2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
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.
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>
35 #include "trace_gfs2.h"
38 * gfs2_struct2blk - compute stuff
39 * @sdp: the filesystem
40 * @nstruct: the number of structures
41 * @ssize: the size of the structures
43 * Compute the number of log descriptor blocks needed to hold a certain number
44 * of structures of a certain size.
46 * Returns: the number of blocks needed (minimum is always 1)
49 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
53 unsigned int first, second;
56 first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
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);
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
72 * The ail lock _must_ be held when calling this function
76 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
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);
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
93 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
94 struct writeback_control *wbc,
95 struct gfs2_trans *tr,
97 __releases(&sdp->sd_ail_lock)
98 __acquires(&sdp->sd_ail_lock)
100 struct gfs2_glock *gl = NULL;
101 struct address_space *mapping;
102 struct gfs2_bufdata *bd, *s;
103 struct buffer_head *bh;
105 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
108 gfs2_assert(sdp, bd->bd_tr == tr);
110 if (!buffer_busy(bh)) {
111 if (!buffer_uptodate(bh) &&
112 !test_and_set_bit(SDF_AIL1_IO_ERROR,
114 gfs2_io_error_bh(sdp, bh);
117 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
121 if (!buffer_dirty(bh))
126 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
127 mapping = bh->b_page->mapping;
130 spin_unlock(&sdp->sd_ail_lock);
131 generic_writepages(mapping, wbc);
132 spin_lock(&sdp->sd_ail_lock);
133 if (wbc->nr_to_write <= 0)
143 * gfs2_ail1_flush - start writeback of some ail1 entries
144 * @sdp: The super block
145 * @wbc: The writeback control structure
147 * Writes back some ail1 entries, according to the limits in the
148 * writeback control structure
151 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
153 struct list_head *head = &sdp->sd_ail1_list;
154 struct gfs2_trans *tr;
155 struct blk_plug plug;
156 bool withdraw = false;
158 trace_gfs2_ail_flush(sdp, wbc, 1);
159 blk_start_plug(&plug);
160 spin_lock(&sdp->sd_ail_lock);
162 list_for_each_entry_reverse(tr, head, tr_list) {
163 if (wbc->nr_to_write <= 0)
165 if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw))
168 spin_unlock(&sdp->sd_ail_lock);
169 blk_finish_plug(&plug);
171 gfs2_lm_withdraw(sdp, NULL);
172 trace_gfs2_ail_flush(sdp, wbc, 0);
176 * gfs2_ail1_start - start writeback of all ail1 entries
177 * @sdp: The superblock
180 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
182 struct writeback_control wbc = {
183 .sync_mode = WB_SYNC_NONE,
184 .nr_to_write = LONG_MAX,
186 .range_end = LLONG_MAX,
189 return gfs2_ail1_flush(sdp, &wbc);
193 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
194 * @sdp: the filesystem
199 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
202 struct gfs2_bufdata *bd, *s;
203 struct buffer_head *bh;
205 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
208 gfs2_assert(sdp, bd->bd_tr == tr);
211 if (!buffer_uptodate(bh) &&
212 !test_and_set_bit(SDF_AIL1_IO_ERROR, &sdp->sd_flags)) {
213 gfs2_io_error_bh(sdp, bh);
216 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
221 * gfs2_ail1_empty - Try to empty the ail1 lists
222 * @sdp: The superblock
224 * Tries to empty the ail1 lists, starting with the oldest first
227 static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
229 struct gfs2_trans *tr, *s;
232 bool withdraw = false;
234 spin_lock(&sdp->sd_ail_lock);
235 list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
236 gfs2_ail1_empty_one(sdp, tr, &withdraw);
237 if (list_empty(&tr->tr_ail1_list) && oldest_tr)
238 list_move(&tr->tr_list, &sdp->sd_ail2_list);
242 ret = list_empty(&sdp->sd_ail1_list);
243 spin_unlock(&sdp->sd_ail_lock);
246 gfs2_lm_withdraw(sdp, "fatal: I/O error(s)\n");
251 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
253 struct gfs2_trans *tr;
254 struct gfs2_bufdata *bd;
255 struct buffer_head *bh;
257 spin_lock(&sdp->sd_ail_lock);
258 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
259 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
261 if (!buffer_locked(bh))
264 spin_unlock(&sdp->sd_ail_lock);
270 spin_unlock(&sdp->sd_ail_lock);
274 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
275 * @sdp: the filesystem
280 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
282 struct list_head *head = &tr->tr_ail2_list;
283 struct gfs2_bufdata *bd;
285 while (!list_empty(head)) {
286 bd = list_entry(head->prev, struct gfs2_bufdata,
288 gfs2_assert(sdp, bd->bd_tr == tr);
289 gfs2_remove_from_ail(bd);
293 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
295 struct gfs2_trans *tr, *safe;
296 unsigned int old_tail = sdp->sd_log_tail;
297 int wrap = (new_tail < old_tail);
300 spin_lock(&sdp->sd_ail_lock);
302 list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
303 a = (old_tail <= tr->tr_first);
304 b = (tr->tr_first < new_tail);
305 rm = (wrap) ? (a || b) : (a && b);
309 gfs2_ail2_empty_one(sdp, tr);
310 list_del(&tr->tr_list);
311 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
312 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
316 spin_unlock(&sdp->sd_ail_lock);
320 * gfs2_log_release - Release a given number of log blocks
321 * @sdp: The GFS2 superblock
322 * @blks: The number of blocks
326 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
329 atomic_add(blks, &sdp->sd_log_blks_free);
330 trace_gfs2_log_blocks(sdp, blks);
331 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
332 sdp->sd_jdesc->jd_blocks);
333 up_read(&sdp->sd_log_flush_lock);
337 * gfs2_log_reserve - Make a log reservation
338 * @sdp: The GFS2 superblock
339 * @blks: The number of blocks to reserve
341 * Note that we never give out the last few blocks of the journal. Thats
342 * due to the fact that there is a small number of header blocks
343 * associated with each log flush. The exact number can't be known until
344 * flush time, so we ensure that we have just enough free blocks at all
345 * times to avoid running out during a log flush.
347 * We no longer flush the log here, instead we wake up logd to do that
348 * for us. To avoid the thundering herd and to ensure that we deal fairly
349 * with queued waiters, we use an exclusive wait. This means that when we
350 * get woken with enough journal space to get our reservation, we need to
351 * wake the next waiter on the list.
356 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
359 unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
360 unsigned wanted = blks + reserved_blks;
363 unsigned int free_blocks;
365 if (gfs2_assert_warn(sdp, blks) ||
366 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
368 atomic_add(blks, &sdp->sd_log_blks_needed);
370 free_blocks = atomic_read(&sdp->sd_log_blks_free);
371 if (unlikely(free_blocks <= wanted)) {
373 prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
374 TASK_UNINTERRUPTIBLE);
375 wake_up(&sdp->sd_logd_waitq);
377 if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
379 free_blocks = atomic_read(&sdp->sd_log_blks_free);
380 } while(free_blocks <= wanted);
381 finish_wait(&sdp->sd_log_waitq, &wait);
383 atomic_inc(&sdp->sd_reserving_log);
384 if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
385 free_blocks - blks) != free_blocks) {
386 if (atomic_dec_and_test(&sdp->sd_reserving_log))
387 wake_up(&sdp->sd_reserving_log_wait);
390 atomic_sub(blks, &sdp->sd_log_blks_needed);
391 trace_gfs2_log_blocks(sdp, -blks);
394 * If we waited, then so might others, wake them up _after_ we get
395 * our share of the log.
397 if (unlikely(did_wait))
398 wake_up(&sdp->sd_log_waitq);
400 down_read(&sdp->sd_log_flush_lock);
401 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
402 gfs2_log_release(sdp, blks);
405 if (atomic_dec_and_test(&sdp->sd_reserving_log))
406 wake_up(&sdp->sd_reserving_log_wait);
411 * log_distance - Compute distance between two journal blocks
412 * @sdp: The GFS2 superblock
413 * @newer: The most recent journal block of the pair
414 * @older: The older journal block of the pair
416 * Compute the distance (in the journal direction) between two
417 * blocks in the journal
419 * Returns: the distance in blocks
422 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
427 dist = newer - older;
429 dist += sdp->sd_jdesc->jd_blocks;
435 * calc_reserved - Calculate the number of blocks to reserve when
436 * refunding a transaction's unused buffers.
437 * @sdp: The GFS2 superblock
439 * This is complex. We need to reserve room for all our currently used
440 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
441 * all our journaled data buffers for journaled files (e.g. files in the
442 * meta_fs like rindex, or files for which chattr +j was done.)
443 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
444 * will count it as free space (sd_log_blks_free) and corruption will follow.
446 * We can have metadata bufs and jdata bufs in the same journal. So each
447 * type gets its own log header, for which we need to reserve a block.
448 * In fact, each type has the potential for needing more than one header
449 * in cases where we have more buffers than will fit on a journal page.
450 * Metadata journal entries take up half the space of journaled buffer entries.
451 * Thus, metadata entries have buf_limit (502) and journaled buffers have
452 * databuf_limit (251) before they cause a wrap around.
454 * Also, we need to reserve blocks for revoke journal entries and one for an
455 * overall header for the lot.
457 * Returns: the number of blocks reserved
459 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
461 unsigned int reserved = 0;
464 struct gfs2_trans *tr = sdp->sd_log_tr;
467 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
468 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
469 reserved = mbuf + dbuf;
470 /* Account for header blocks */
471 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
472 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
475 if (sdp->sd_log_commited_revoke > 0)
476 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
478 /* One for the overall header */
484 static unsigned int current_tail(struct gfs2_sbd *sdp)
486 struct gfs2_trans *tr;
489 spin_lock(&sdp->sd_ail_lock);
491 if (list_empty(&sdp->sd_ail1_list)) {
492 tail = sdp->sd_log_head;
494 tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
499 spin_unlock(&sdp->sd_ail_lock);
504 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
506 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
508 ail2_empty(sdp, new_tail);
510 atomic_add(dist, &sdp->sd_log_blks_free);
511 trace_gfs2_log_blocks(sdp, dist);
512 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
513 sdp->sd_jdesc->jd_blocks);
515 sdp->sd_log_tail = new_tail;
519 static void log_flush_wait(struct gfs2_sbd *sdp)
523 if (atomic_read(&sdp->sd_log_in_flight)) {
525 prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
526 TASK_UNINTERRUPTIBLE);
527 if (atomic_read(&sdp->sd_log_in_flight))
529 } while(atomic_read(&sdp->sd_log_in_flight));
530 finish_wait(&sdp->sd_log_flush_wait, &wait);
534 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
536 struct gfs2_inode *ipa, *ipb;
538 ipa = list_entry(a, struct gfs2_inode, i_ordered);
539 ipb = list_entry(b, struct gfs2_inode, i_ordered);
541 if (ipa->i_no_addr < ipb->i_no_addr)
543 if (ipa->i_no_addr > ipb->i_no_addr)
548 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
550 struct gfs2_inode *ip;
553 spin_lock(&sdp->sd_ordered_lock);
554 list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
555 while (!list_empty(&sdp->sd_log_ordered)) {
556 ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered);
557 if (ip->i_inode.i_mapping->nrpages == 0) {
558 test_and_clear_bit(GIF_ORDERED, &ip->i_flags);
559 list_del(&ip->i_ordered);
562 list_move(&ip->i_ordered, &written);
563 spin_unlock(&sdp->sd_ordered_lock);
564 filemap_fdatawrite(ip->i_inode.i_mapping);
565 spin_lock(&sdp->sd_ordered_lock);
567 list_splice(&written, &sdp->sd_log_ordered);
568 spin_unlock(&sdp->sd_ordered_lock);
571 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
573 struct gfs2_inode *ip;
575 spin_lock(&sdp->sd_ordered_lock);
576 while (!list_empty(&sdp->sd_log_ordered)) {
577 ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered);
578 list_del(&ip->i_ordered);
579 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
580 if (ip->i_inode.i_mapping->nrpages == 0)
582 spin_unlock(&sdp->sd_ordered_lock);
583 filemap_fdatawait(ip->i_inode.i_mapping);
584 spin_lock(&sdp->sd_ordered_lock);
586 spin_unlock(&sdp->sd_ordered_lock);
589 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
591 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
593 spin_lock(&sdp->sd_ordered_lock);
594 if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
595 list_del(&ip->i_ordered);
596 spin_unlock(&sdp->sd_ordered_lock);
599 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
601 struct buffer_head *bh = bd->bd_bh;
602 struct gfs2_glock *gl = bd->bd_gl;
604 bh->b_private = NULL;
605 bd->bd_blkno = bh->b_blocknr;
606 gfs2_remove_from_ail(bd); /* drops ref on bh */
608 sdp->sd_log_num_revoke++;
609 if (atomic_inc_return(&gl->gl_revokes) == 1)
611 set_bit(GLF_LFLUSH, &gl->gl_flags);
612 list_add(&bd->bd_list, &sdp->sd_log_revokes);
615 void gfs2_write_revokes(struct gfs2_sbd *sdp)
617 struct gfs2_trans *tr;
618 struct gfs2_bufdata *bd, *tmp;
619 int have_revokes = 0;
620 int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
622 gfs2_ail1_empty(sdp);
623 spin_lock(&sdp->sd_ail_lock);
624 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
625 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
626 if (list_empty(&bd->bd_list)) {
633 spin_unlock(&sdp->sd_ail_lock);
634 if (have_revokes == 0)
636 while (sdp->sd_log_num_revoke > max_revokes)
637 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
638 max_revokes -= sdp->sd_log_num_revoke;
639 if (!sdp->sd_log_num_revoke) {
640 atomic_dec(&sdp->sd_log_blks_free);
641 /* If no blocks have been reserved, we need to also
642 * reserve a block for the header */
643 if (!sdp->sd_log_blks_reserved)
644 atomic_dec(&sdp->sd_log_blks_free);
647 spin_lock(&sdp->sd_ail_lock);
648 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
649 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
650 if (max_revokes == 0)
652 if (!list_empty(&bd->bd_list))
654 gfs2_add_revoke(sdp, bd);
659 spin_unlock(&sdp->sd_ail_lock);
660 gfs2_log_unlock(sdp);
662 if (!sdp->sd_log_num_revoke) {
663 atomic_inc(&sdp->sd_log_blks_free);
664 if (!sdp->sd_log_blks_reserved)
665 atomic_inc(&sdp->sd_log_blks_free);
670 * gfs2_write_log_header - Write a journal log header buffer at lblock
671 * @sdp: The GFS2 superblock
672 * @jd: journal descriptor of the journal to which we are writing
673 * @seq: sequence number
674 * @tail: tail of the log
675 * @lblock: value for lh_blkno (block number relative to start of journal)
676 * @flags: log header flags GFS2_LOG_HEAD_*
677 * @op_flags: flags to pass to the bio
679 * Returns: the initialized log buffer descriptor
682 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
683 u64 seq, u32 tail, u32 lblock, u32 flags,
686 struct gfs2_log_header *lh;
688 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
689 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
690 struct timespec64 tv;
691 struct super_block *sb = sdp->sd_vfs;
694 lh = page_address(page);
697 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
698 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
699 lh->lh_header.__pad0 = cpu_to_be64(0);
700 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
701 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
702 lh->lh_sequence = cpu_to_be64(seq);
703 lh->lh_flags = cpu_to_be32(flags);
704 lh->lh_tail = cpu_to_be32(tail);
705 lh->lh_blkno = cpu_to_be32(lblock);
706 hash = ~crc32(~0, lh, LH_V1_SIZE);
707 lh->lh_hash = cpu_to_be32(hash);
709 ktime_get_coarse_real_ts64(&tv);
710 lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
711 lh->lh_sec = cpu_to_be64(tv.tv_sec);
712 if (!list_empty(&jd->extent_list))
713 dblock = gfs2_log_bmap(sdp);
715 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
716 if (gfs2_assert_withdraw(sdp, ret == 0))
719 lh->lh_addr = cpu_to_be64(dblock);
720 lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
722 /* We may only write local statfs, quota, etc., when writing to our
723 own journal. The values are left 0 when recovering a journal
724 different from our own. */
725 if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
727 cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
729 cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
731 spin_lock(&sdp->sd_statfs_spin);
732 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
733 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
734 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
735 spin_unlock(&sdp->sd_statfs_spin);
738 BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
740 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
741 sb->s_blocksize - LH_V1_SIZE - 4);
742 lh->lh_crc = cpu_to_be32(crc);
744 gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
745 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
750 * log_write_header - Get and initialize a journal header buffer
751 * @sdp: The GFS2 superblock
752 * @flags: The log header flags, including log header origin
754 * Returns: the initialized log buffer descriptor
757 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
760 int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
761 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
763 gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
764 tail = current_tail(sdp);
766 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
767 gfs2_ordered_wait(sdp);
769 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
771 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
772 gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
773 sdp->sd_log_flush_head, flags, op_flags);
775 if (sdp->sd_log_tail != tail)
776 log_pull_tail(sdp, tail);
780 * gfs2_log_flush - flush incore transaction(s)
781 * @sdp: the filesystem
782 * @gl: The glock structure to flush. If NULL, flush the whole incore log
783 * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
787 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
789 struct gfs2_trans *tr;
790 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
792 down_write(&sdp->sd_log_flush_lock);
794 /* Log might have been flushed while we waited for the flush lock */
795 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
796 up_write(&sdp->sd_log_flush_lock);
799 trace_gfs2_log_flush(sdp, 1, flags);
801 if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
802 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
804 sdp->sd_log_flush_head = sdp->sd_log_head;
807 sdp->sd_log_tr = NULL;
808 INIT_LIST_HEAD(&tr->tr_ail1_list);
809 INIT_LIST_HEAD(&tr->tr_ail2_list);
810 tr->tr_first = sdp->sd_log_flush_head;
811 if (unlikely (state == SFS_FROZEN))
812 gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
815 if (unlikely(state == SFS_FROZEN))
816 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
817 gfs2_assert_withdraw(sdp,
818 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
820 gfs2_ordered_write(sdp);
821 lops_before_commit(sdp, tr);
822 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
824 if (sdp->sd_log_head != sdp->sd_log_flush_head) {
826 log_write_header(sdp, flags);
827 } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
828 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
829 trace_gfs2_log_blocks(sdp, -1);
830 log_write_header(sdp, flags);
832 lops_after_commit(sdp, tr);
835 sdp->sd_log_head = sdp->sd_log_flush_head;
836 sdp->sd_log_blks_reserved = 0;
837 sdp->sd_log_commited_revoke = 0;
839 spin_lock(&sdp->sd_ail_lock);
840 if (tr && !list_empty(&tr->tr_ail1_list)) {
841 list_add(&tr->tr_list, &sdp->sd_ail1_list);
844 spin_unlock(&sdp->sd_ail_lock);
845 gfs2_log_unlock(sdp);
847 if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
848 if (!sdp->sd_log_idle) {
850 gfs2_ail1_start(sdp);
852 if (gfs2_ail1_empty(sdp))
855 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
856 trace_gfs2_log_blocks(sdp, -1);
857 log_write_header(sdp, flags);
858 sdp->sd_log_head = sdp->sd_log_flush_head;
860 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
861 GFS2_LOG_HEAD_FLUSH_FREEZE))
862 gfs2_log_shutdown(sdp);
863 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
864 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
867 trace_gfs2_log_flush(sdp, 0, flags);
868 up_write(&sdp->sd_log_flush_lock);
874 * gfs2_merge_trans - Merge a new transaction into a cached transaction
875 * @old: Original transaction to be expanded
876 * @new: New transaction to be merged
879 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
881 WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
883 old->tr_num_buf_new += new->tr_num_buf_new;
884 old->tr_num_databuf_new += new->tr_num_databuf_new;
885 old->tr_num_buf_rm += new->tr_num_buf_rm;
886 old->tr_num_databuf_rm += new->tr_num_databuf_rm;
887 old->tr_num_revoke += new->tr_num_revoke;
888 old->tr_num_revoke_rm += new->tr_num_revoke_rm;
890 list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
891 list_splice_tail_init(&new->tr_buf, &old->tr_buf);
894 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
896 unsigned int reserved;
902 if (sdp->sd_log_tr) {
903 gfs2_merge_trans(sdp->sd_log_tr, tr);
904 } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
905 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
907 set_bit(TR_ATTACHED, &tr->tr_flags);
910 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
911 reserved = calc_reserved(sdp);
912 maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
913 gfs2_assert_withdraw(sdp, maxres >= reserved);
914 unused = maxres - reserved;
915 atomic_add(unused, &sdp->sd_log_blks_free);
916 trace_gfs2_log_blocks(sdp, unused);
917 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
918 sdp->sd_jdesc->jd_blocks);
919 sdp->sd_log_blks_reserved = reserved;
921 gfs2_log_unlock(sdp);
925 * gfs2_log_commit - Commit a transaction to the log
926 * @sdp: the filesystem
927 * @tr: the transaction
929 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
930 * or the total number of used blocks (pinned blocks plus AIL blocks)
931 * is greater than thresh2.
933 * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
939 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
943 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
944 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
945 atomic_read(&sdp->sd_log_thresh2)))
946 wake_up(&sdp->sd_logd_waitq);
950 * gfs2_log_shutdown - write a shutdown header into a journal
951 * @sdp: the filesystem
955 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
957 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
958 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
959 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
961 sdp->sd_log_flush_head = sdp->sd_log_head;
963 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
965 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
966 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
968 sdp->sd_log_head = sdp->sd_log_flush_head;
969 sdp->sd_log_tail = sdp->sd_log_head;
972 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
974 return (atomic_read(&sdp->sd_log_pinned) +
975 atomic_read(&sdp->sd_log_blks_needed) >=
976 atomic_read(&sdp->sd_log_thresh1));
979 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
981 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
983 if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
986 return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
987 atomic_read(&sdp->sd_log_thresh2);
991 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
992 * @sdp: Pointer to GFS2 superblock
994 * Also, periodically check to make sure that we're using the most recent
998 int gfs2_logd(void *data)
1000 struct gfs2_sbd *sdp = data;
1001 unsigned long t = 1;
1005 while (!kthread_should_stop()) {
1007 /* Check for errors writing to the journal */
1008 if (sdp->sd_log_error) {
1009 gfs2_lm_withdraw(sdp,
1010 "GFS2: fsid=%s: error %d: "
1011 "withdrawing the file system to "
1012 "prevent further damage.\n",
1013 sdp->sd_fsname, sdp->sd_log_error);
1017 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1018 gfs2_ail1_empty(sdp);
1019 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1020 GFS2_LFC_LOGD_JFLUSH_REQD);
1024 if (gfs2_ail_flush_reqd(sdp)) {
1025 gfs2_ail1_start(sdp);
1026 gfs2_ail1_wait(sdp);
1027 gfs2_ail1_empty(sdp);
1028 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1029 GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1033 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1034 wake_up(&sdp->sd_log_waitq);
1036 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1041 prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1042 TASK_INTERRUPTIBLE);
1043 if (!gfs2_ail_flush_reqd(sdp) &&
1044 !gfs2_jrnl_flush_reqd(sdp) &&
1045 !kthread_should_stop())
1046 t = schedule_timeout(t);
1047 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1048 !gfs2_jrnl_flush_reqd(sdp) &&
1049 !kthread_should_stop());
1050 finish_wait(&sdp->sd_logd_waitq, &wait);