1/*
2 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/gfs2_ondisk.h>
16#include <linux/crc32.h>
17#include <linux/delay.h>
18#include <linux/kthread.h>
19#include <linux/freezer.h>
20#include <linux/bio.h>
21#include <linux/blkdev.h>
22#include <linux/writeback.h>
23#include <linux/list_sort.h>
24
25#include "gfs2.h"
26#include "incore.h"
27#include "bmap.h"
28#include "glock.h"
29#include "log.h"
30#include "lops.h"
31#include "meta_io.h"
32#include "util.h"
33#include "dir.h"
34#include "trace_gfs2.h"
35
36/**
37 * gfs2_struct2blk - compute stuff
38 * @sdp: the filesystem
39 * @nstruct: the number of structures
40 * @ssize: the size of the structures
41 *
42 * Compute the number of log descriptor blocks needed to hold a certain number
43 * of structures of a certain size.
44 *
45 * Returns: the number of blocks needed (minimum is always 1)
46 */
47
48unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
49			     unsigned int ssize)
50{
51	unsigned int blks;
52	unsigned int first, second;
53
54	blks = 1;
55	first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
56
57	if (nstruct > first) {
58		second = (sdp->sd_sb.sb_bsize -
59			  sizeof(struct gfs2_meta_header)) / ssize;
60		blks += DIV_ROUND_UP(nstruct - first, second);
61	}
62
63	return blks;
64}
65
66/**
67 * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
68 * @mapping: The associated mapping (maybe NULL)
69 * @bd: The gfs2_bufdata to remove
70 *
71 * The ail lock _must_ be held when calling this function
72 *
73 */
74
75void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
76{
77	bd->bd_tr = NULL;
78	list_del_init(&bd->bd_ail_st_list);
79	list_del_init(&bd->bd_ail_gl_list);
80	atomic_dec(&bd->bd_gl->gl_ail_count);
81	brelse(bd->bd_bh);
82}
83
84/**
85 * gfs2_ail1_start_one - Start I/O on a part of the AIL
86 * @sdp: the filesystem
87 * @wbc: The writeback control structure
88 * @ai: The ail structure
89 *
90 */
91
92static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
93			       struct writeback_control *wbc,
94			       struct gfs2_trans *tr)
95__releases(&sdp->sd_ail_lock)
96__acquires(&sdp->sd_ail_lock)
97{
98	struct gfs2_glock *gl = NULL;
99	struct address_space *mapping;
100	struct gfs2_bufdata *bd, *s;
101	struct buffer_head *bh;
102
103	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
104		bh = bd->bd_bh;
105
106		gfs2_assert(sdp, bd->bd_tr == tr);
107
108		if (!buffer_busy(bh)) {
109			if (!buffer_uptodate(bh))
110				gfs2_io_error_bh(sdp, bh);
111			list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
112			continue;
113		}
114
115		if (!buffer_dirty(bh))
116			continue;
117		if (gl == bd->bd_gl)
118			continue;
119		gl = bd->bd_gl;
120		list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
121		mapping = bh->b_page->mapping;
122		if (!mapping)
123			continue;
124		spin_unlock(&sdp->sd_ail_lock);
125		generic_writepages(mapping, wbc);
126		spin_lock(&sdp->sd_ail_lock);
127		if (wbc->nr_to_write <= 0)
128			break;
129		return 1;
130	}
131
132	return 0;
133}
134
135
136/**
137 * gfs2_ail1_flush - start writeback of some ail1 entries
138 * @sdp: The super block
139 * @wbc: The writeback control structure
140 *
141 * Writes back some ail1 entries, according to the limits in the
142 * writeback control structure
143 */
144
145void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
146{
147	struct list_head *head = &sdp->sd_ail1_list;
148	struct gfs2_trans *tr;
149	struct blk_plug plug;
150
151	trace_gfs2_ail_flush(sdp, wbc, 1);
152	blk_start_plug(&plug);
153	spin_lock(&sdp->sd_ail_lock);
154restart:
155	list_for_each_entry_reverse(tr, head, tr_list) {
156		if (wbc->nr_to_write <= 0)
157			break;
158		if (gfs2_ail1_start_one(sdp, wbc, tr))
159			goto restart;
160	}
161	spin_unlock(&sdp->sd_ail_lock);
162	blk_finish_plug(&plug);
163	trace_gfs2_ail_flush(sdp, wbc, 0);
164}
165
166/**
167 * gfs2_ail1_start - start writeback of all ail1 entries
168 * @sdp: The superblock
169 */
170
171static void gfs2_ail1_start(struct gfs2_sbd *sdp)
172{
173	struct writeback_control wbc = {
174		.sync_mode = WB_SYNC_NONE,
175		.nr_to_write = LONG_MAX,
176		.range_start = 0,
177		.range_end = LLONG_MAX,
178	};
179
180	return gfs2_ail1_flush(sdp, &wbc);
181}
182
183/**
184 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
185 * @sdp: the filesystem
186 * @ai: the AIL entry
187 *
188 */
189
190static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
191{
192	struct gfs2_bufdata *bd, *s;
193	struct buffer_head *bh;
194
195	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
196					 bd_ail_st_list) {
197		bh = bd->bd_bh;
198		gfs2_assert(sdp, bd->bd_tr == tr);
199		if (buffer_busy(bh))
200			continue;
201		if (!buffer_uptodate(bh))
202			gfs2_io_error_bh(sdp, bh);
203		list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
204	}
205
206}
207
208/**
209 * gfs2_ail1_empty - Try to empty the ail1 lists
210 * @sdp: The superblock
211 *
212 * Tries to empty the ail1 lists, starting with the oldest first
213 */
214
215static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
216{
217	struct gfs2_trans *tr, *s;
218	int oldest_tr = 1;
219	int ret;
220
221	spin_lock(&sdp->sd_ail_lock);
222	list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
223		gfs2_ail1_empty_one(sdp, tr);
224		if (list_empty(&tr->tr_ail1_list) && oldest_tr)
225			list_move(&tr->tr_list, &sdp->sd_ail2_list);
226		else
227			oldest_tr = 0;
228	}
229	ret = list_empty(&sdp->sd_ail1_list);
230	spin_unlock(&sdp->sd_ail_lock);
231
232	return ret;
233}
234
235static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
236{
237	struct gfs2_trans *tr;
238	struct gfs2_bufdata *bd;
239	struct buffer_head *bh;
240
241	spin_lock(&sdp->sd_ail_lock);
242	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
243		list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
244			bh = bd->bd_bh;
245			if (!buffer_locked(bh))
246				continue;
247			get_bh(bh);
248			spin_unlock(&sdp->sd_ail_lock);
249			wait_on_buffer(bh);
250			brelse(bh);
251			return;
252		}
253	}
254	spin_unlock(&sdp->sd_ail_lock);
255}
256
257/**
258 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
259 * @sdp: the filesystem
260 * @ai: the AIL entry
261 *
262 */
263
264static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
265{
266	struct list_head *head = &tr->tr_ail2_list;
267	struct gfs2_bufdata *bd;
268
269	while (!list_empty(head)) {
270		bd = list_entry(head->prev, struct gfs2_bufdata,
271				bd_ail_st_list);
272		gfs2_assert(sdp, bd->bd_tr == tr);
273		gfs2_remove_from_ail(bd);
274	}
275}
276
277static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
278{
279	struct gfs2_trans *tr, *safe;
280	unsigned int old_tail = sdp->sd_log_tail;
281	int wrap = (new_tail < old_tail);
282	int a, b, rm;
283
284	spin_lock(&sdp->sd_ail_lock);
285
286	list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
287		a = (old_tail <= tr->tr_first);
288		b = (tr->tr_first < new_tail);
289		rm = (wrap) ? (a || b) : (a && b);
290		if (!rm)
291			continue;
292
293		gfs2_ail2_empty_one(sdp, tr);
294		list_del(&tr->tr_list);
295		gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
296		gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
297		kfree(tr);
298	}
299
300	spin_unlock(&sdp->sd_ail_lock);
301}
302
303/**
304 * gfs2_log_release - Release a given number of log blocks
305 * @sdp: The GFS2 superblock
306 * @blks: The number of blocks
307 *
308 */
309
310void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
311{
312
313	atomic_add(blks, &sdp->sd_log_blks_free);
314	trace_gfs2_log_blocks(sdp, blks);
315	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
316				  sdp->sd_jdesc->jd_blocks);
317	up_read(&sdp->sd_log_flush_lock);
318}
319
320/**
321 * gfs2_log_reserve - Make a log reservation
322 * @sdp: The GFS2 superblock
323 * @blks: The number of blocks to reserve
324 *
325 * Note that we never give out the last few blocks of the journal. Thats
326 * due to the fact that there is a small number of header blocks
327 * associated with each log flush. The exact number can't be known until
328 * flush time, so we ensure that we have just enough free blocks at all
329 * times to avoid running out during a log flush.
330 *
331 * We no longer flush the log here, instead we wake up logd to do that
332 * for us. To avoid the thundering herd and to ensure that we deal fairly
333 * with queued waiters, we use an exclusive wait. This means that when we
334 * get woken with enough journal space to get our reservation, we need to
335 * wake the next waiter on the list.
336 *
337 * Returns: errno
338 */
339
340int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
341{
342	int ret = 0;
343	unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
344	unsigned wanted = blks + reserved_blks;
345	DEFINE_WAIT(wait);
346	int did_wait = 0;
347	unsigned int free_blocks;
348
349	if (gfs2_assert_warn(sdp, blks) ||
350	    gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
351		return -EINVAL;
352retry:
353	free_blocks = atomic_read(&sdp->sd_log_blks_free);
354	if (unlikely(free_blocks <= wanted)) {
355		do {
356			prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
357					TASK_UNINTERRUPTIBLE);
358			wake_up(&sdp->sd_logd_waitq);
359			did_wait = 1;
360			if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
361				io_schedule();
362			free_blocks = atomic_read(&sdp->sd_log_blks_free);
363		} while(free_blocks <= wanted);
364		finish_wait(&sdp->sd_log_waitq, &wait);
365	}
366	atomic_inc(&sdp->sd_reserving_log);
367	if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
368				free_blocks - blks) != free_blocks) {
369		if (atomic_dec_and_test(&sdp->sd_reserving_log))
370			wake_up(&sdp->sd_reserving_log_wait);
371		goto retry;
372	}
373	trace_gfs2_log_blocks(sdp, -blks);
374
375	/*
376	 * If we waited, then so might others, wake them up _after_ we get
377	 * our share of the log.
378	 */
379	if (unlikely(did_wait))
380		wake_up(&sdp->sd_log_waitq);
381
382	down_read(&sdp->sd_log_flush_lock);
383	if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
384		gfs2_log_release(sdp, blks);
385		ret = -EROFS;
386	}
387	if (atomic_dec_and_test(&sdp->sd_reserving_log))
388		wake_up(&sdp->sd_reserving_log_wait);
389	return ret;
390}
391
392/**
393 * log_distance - Compute distance between two journal blocks
394 * @sdp: The GFS2 superblock
395 * @newer: The most recent journal block of the pair
396 * @older: The older journal block of the pair
397 *
398 *   Compute the distance (in the journal direction) between two
399 *   blocks in the journal
400 *
401 * Returns: the distance in blocks
402 */
403
404static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
405					unsigned int older)
406{
407	int dist;
408
409	dist = newer - older;
410	if (dist < 0)
411		dist += sdp->sd_jdesc->jd_blocks;
412
413	return dist;
414}
415
416/**
417 * calc_reserved - Calculate the number of blocks to reserve when
418 *                 refunding a transaction's unused buffers.
419 * @sdp: The GFS2 superblock
420 *
421 * This is complex.  We need to reserve room for all our currently used
422 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
423 * all our journaled data buffers for journaled files (e.g. files in the
424 * meta_fs like rindex, or files for which chattr +j was done.)
425 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
426 * will count it as free space (sd_log_blks_free) and corruption will follow.
427 *
428 * We can have metadata bufs and jdata bufs in the same journal.  So each
429 * type gets its own log header, for which we need to reserve a block.
430 * In fact, each type has the potential for needing more than one header
431 * in cases where we have more buffers than will fit on a journal page.
432 * Metadata journal entries take up half the space of journaled buffer entries.
433 * Thus, metadata entries have buf_limit (502) and journaled buffers have
434 * databuf_limit (251) before they cause a wrap around.
435 *
436 * Also, we need to reserve blocks for revoke journal entries and one for an
437 * overall header for the lot.
438 *
439 * Returns: the number of blocks reserved
440 */
441static unsigned int calc_reserved(struct gfs2_sbd *sdp)
442{
443	unsigned int reserved = 0;
444	unsigned int mbuf;
445	unsigned int dbuf;
446	struct gfs2_trans *tr = sdp->sd_log_tr;
447
448	if (tr) {
449		mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
450		dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
451		reserved = mbuf + dbuf;
452		/* Account for header blocks */
453		reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
454		reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
455	}
456
457	if (sdp->sd_log_commited_revoke > 0)
458		reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
459					  sizeof(u64));
460	/* One for the overall header */
461	if (reserved)
462		reserved++;
463	return reserved;
464}
465
466static unsigned int current_tail(struct gfs2_sbd *sdp)
467{
468	struct gfs2_trans *tr;
469	unsigned int tail;
470
471	spin_lock(&sdp->sd_ail_lock);
472
473	if (list_empty(&sdp->sd_ail1_list)) {
474		tail = sdp->sd_log_head;
475	} else {
476		tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
477				tr_list);
478		tail = tr->tr_first;
479	}
480
481	spin_unlock(&sdp->sd_ail_lock);
482
483	return tail;
484}
485
486static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
487{
488	unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
489
490	ail2_empty(sdp, new_tail);
491
492	atomic_add(dist, &sdp->sd_log_blks_free);
493	trace_gfs2_log_blocks(sdp, dist);
494	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
495			     sdp->sd_jdesc->jd_blocks);
496
497	sdp->sd_log_tail = new_tail;
498}
499
500
501static void log_flush_wait(struct gfs2_sbd *sdp)
502{
503	DEFINE_WAIT(wait);
504
505	if (atomic_read(&sdp->sd_log_in_flight)) {
506		do {
507			prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
508					TASK_UNINTERRUPTIBLE);
509			if (atomic_read(&sdp->sd_log_in_flight))
510				io_schedule();
511		} while(atomic_read(&sdp->sd_log_in_flight));
512		finish_wait(&sdp->sd_log_flush_wait, &wait);
513	}
514}
515
516static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
517{
518	struct gfs2_inode *ipa, *ipb;
519
520	ipa = list_entry(a, struct gfs2_inode, i_ordered);
521	ipb = list_entry(b, struct gfs2_inode, i_ordered);
522
523	if (ipa->i_no_addr < ipb->i_no_addr)
524		return -1;
525	if (ipa->i_no_addr > ipb->i_no_addr)
526		return 1;
527	return 0;
528}
529
530static void gfs2_ordered_write(struct gfs2_sbd *sdp)
531{
532	struct gfs2_inode *ip;
533	LIST_HEAD(written);
534
535	spin_lock(&sdp->sd_ordered_lock);
536	list_sort(NULL, &sdp->sd_log_le_ordered, &ip_cmp);
537	while (!list_empty(&sdp->sd_log_le_ordered)) {
538		ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
539		list_move(&ip->i_ordered, &written);
540		if (ip->i_inode.i_mapping->nrpages == 0)
541			continue;
542		spin_unlock(&sdp->sd_ordered_lock);
543		filemap_fdatawrite(ip->i_inode.i_mapping);
544		spin_lock(&sdp->sd_ordered_lock);
545	}
546	list_splice(&written, &sdp->sd_log_le_ordered);
547	spin_unlock(&sdp->sd_ordered_lock);
548}
549
550static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
551{
552	struct gfs2_inode *ip;
553
554	spin_lock(&sdp->sd_ordered_lock);
555	while (!list_empty(&sdp->sd_log_le_ordered)) {
556		ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
557		list_del(&ip->i_ordered);
558		WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
559		if (ip->i_inode.i_mapping->nrpages == 0)
560			continue;
561		spin_unlock(&sdp->sd_ordered_lock);
562		filemap_fdatawait(ip->i_inode.i_mapping);
563		spin_lock(&sdp->sd_ordered_lock);
564	}
565	spin_unlock(&sdp->sd_ordered_lock);
566}
567
568void gfs2_ordered_del_inode(struct gfs2_inode *ip)
569{
570	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
571
572	spin_lock(&sdp->sd_ordered_lock);
573	if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
574		list_del(&ip->i_ordered);
575	spin_unlock(&sdp->sd_ordered_lock);
576}
577
578void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
579{
580	struct buffer_head *bh = bd->bd_bh;
581	struct gfs2_glock *gl = bd->bd_gl;
582
583	bh->b_private = NULL;
584	bd->bd_blkno = bh->b_blocknr;
585	gfs2_remove_from_ail(bd); /* drops ref on bh */
586	bd->bd_bh = NULL;
587	bd->bd_ops = &gfs2_revoke_lops;
588	sdp->sd_log_num_revoke++;
589	atomic_inc(&gl->gl_revokes);
590	set_bit(GLF_LFLUSH, &gl->gl_flags);
591	list_add(&bd->bd_list, &sdp->sd_log_le_revoke);
592}
593
594void gfs2_write_revokes(struct gfs2_sbd *sdp)
595{
596	struct gfs2_trans *tr;
597	struct gfs2_bufdata *bd, *tmp;
598	int have_revokes = 0;
599	int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
600
601	gfs2_ail1_empty(sdp);
602	spin_lock(&sdp->sd_ail_lock);
603	list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
604		list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
605			if (list_empty(&bd->bd_list)) {
606				have_revokes = 1;
607				goto done;
608			}
609		}
610	}
611done:
612	spin_unlock(&sdp->sd_ail_lock);
613	if (have_revokes == 0)
614		return;
615	while (sdp->sd_log_num_revoke > max_revokes)
616		max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
617	max_revokes -= sdp->sd_log_num_revoke;
618	if (!sdp->sd_log_num_revoke) {
619		atomic_dec(&sdp->sd_log_blks_free);
620		/* If no blocks have been reserved, we need to also
621		 * reserve a block for the header */
622		if (!sdp->sd_log_blks_reserved)
623			atomic_dec(&sdp->sd_log_blks_free);
624	}
625	gfs2_log_lock(sdp);
626	spin_lock(&sdp->sd_ail_lock);
627	list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
628		list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
629			if (max_revokes == 0)
630				goto out_of_blocks;
631			if (!list_empty(&bd->bd_list))
632				continue;
633			gfs2_add_revoke(sdp, bd);
634			max_revokes--;
635		}
636	}
637out_of_blocks:
638	spin_unlock(&sdp->sd_ail_lock);
639	gfs2_log_unlock(sdp);
640
641	if (!sdp->sd_log_num_revoke) {
642		atomic_inc(&sdp->sd_log_blks_free);
643		if (!sdp->sd_log_blks_reserved)
644			atomic_inc(&sdp->sd_log_blks_free);
645	}
646}
647
648/**
649 * log_write_header - Get and initialize a journal header buffer
650 * @sdp: The GFS2 superblock
651 *
652 * Returns: the initialized log buffer descriptor
653 */
654
655static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
656{
657	struct gfs2_log_header *lh;
658	unsigned int tail;
659	u32 hash;
660	int rw = WRITE_FLUSH_FUA | REQ_META;
661	struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
662	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
663	lh = page_address(page);
664	clear_page(lh);
665
666	gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
667
668	tail = current_tail(sdp);
669
670	lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
671	lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
672	lh->lh_header.__pad0 = cpu_to_be64(0);
673	lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
674	lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
675	lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
676	lh->lh_flags = cpu_to_be32(flags);
677	lh->lh_tail = cpu_to_be32(tail);
678	lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
679	hash = gfs2_disk_hash(page_address(page), sizeof(struct gfs2_log_header));
680	lh->lh_hash = cpu_to_be32(hash);
681
682	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
683		gfs2_ordered_wait(sdp);
684		log_flush_wait(sdp);
685		rw = WRITE_SYNC | REQ_META | REQ_PRIO;
686	}
687
688	sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
689	gfs2_log_write_page(sdp, page);
690	gfs2_log_flush_bio(sdp, rw);
691	log_flush_wait(sdp);
692
693	if (sdp->sd_log_tail != tail)
694		log_pull_tail(sdp, tail);
695}
696
697/**
698 * gfs2_log_flush - flush incore transaction(s)
699 * @sdp: the filesystem
700 * @gl: The glock structure to flush.  If NULL, flush the whole incore log
701 *
702 */
703
704void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl,
705		    enum gfs2_flush_type type)
706{
707	struct gfs2_trans *tr;
708	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
709
710	down_write(&sdp->sd_log_flush_lock);
711
712	/* Log might have been flushed while we waited for the flush lock */
713	if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
714		up_write(&sdp->sd_log_flush_lock);
715		return;
716	}
717	trace_gfs2_log_flush(sdp, 1);
718
719	sdp->sd_log_flush_head = sdp->sd_log_head;
720	sdp->sd_log_flush_wrapped = 0;
721	tr = sdp->sd_log_tr;
722	if (tr) {
723		sdp->sd_log_tr = NULL;
724		INIT_LIST_HEAD(&tr->tr_ail1_list);
725		INIT_LIST_HEAD(&tr->tr_ail2_list);
726		tr->tr_first = sdp->sd_log_flush_head;
727		if (unlikely (state == SFS_FROZEN))
728			gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
729	}
730
731	if (unlikely(state == SFS_FROZEN))
732		gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
733	gfs2_assert_withdraw(sdp,
734			sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
735
736	gfs2_ordered_write(sdp);
737	lops_before_commit(sdp, tr);
738	gfs2_log_flush_bio(sdp, WRITE);
739
740	if (sdp->sd_log_head != sdp->sd_log_flush_head) {
741		log_flush_wait(sdp);
742		log_write_header(sdp, 0);
743	} else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
744		atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
745		trace_gfs2_log_blocks(sdp, -1);
746		log_write_header(sdp, 0);
747	}
748	lops_after_commit(sdp, tr);
749
750	gfs2_log_lock(sdp);
751	sdp->sd_log_head = sdp->sd_log_flush_head;
752	sdp->sd_log_blks_reserved = 0;
753	sdp->sd_log_commited_revoke = 0;
754
755	spin_lock(&sdp->sd_ail_lock);
756	if (tr && !list_empty(&tr->tr_ail1_list)) {
757		list_add(&tr->tr_list, &sdp->sd_ail1_list);
758		tr = NULL;
759	}
760	spin_unlock(&sdp->sd_ail_lock);
761	gfs2_log_unlock(sdp);
762
763	if (type != NORMAL_FLUSH) {
764		if (!sdp->sd_log_idle) {
765			for (;;) {
766				gfs2_ail1_start(sdp);
767				gfs2_ail1_wait(sdp);
768				if (gfs2_ail1_empty(sdp))
769					break;
770			}
771			atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
772			trace_gfs2_log_blocks(sdp, -1);
773			sdp->sd_log_flush_wrapped = 0;
774			log_write_header(sdp, 0);
775			sdp->sd_log_head = sdp->sd_log_flush_head;
776		}
777		if (type == SHUTDOWN_FLUSH || type == FREEZE_FLUSH)
778			gfs2_log_shutdown(sdp);
779		if (type == FREEZE_FLUSH)
780			atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
781	}
782
783	trace_gfs2_log_flush(sdp, 0);
784	up_write(&sdp->sd_log_flush_lock);
785
786	kfree(tr);
787}
788
789/**
790 * gfs2_merge_trans - Merge a new transaction into a cached transaction
791 * @old: Original transaction to be expanded
792 * @new: New transaction to be merged
793 */
794
795static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
796{
797	WARN_ON_ONCE(old->tr_attached != 1);
798
799	old->tr_num_buf_new	+= new->tr_num_buf_new;
800	old->tr_num_databuf_new	+= new->tr_num_databuf_new;
801	old->tr_num_buf_rm	+= new->tr_num_buf_rm;
802	old->tr_num_databuf_rm	+= new->tr_num_databuf_rm;
803	old->tr_num_revoke	+= new->tr_num_revoke;
804	old->tr_num_revoke_rm	+= new->tr_num_revoke_rm;
805
806	list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
807	list_splice_tail_init(&new->tr_buf, &old->tr_buf);
808}
809
810static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
811{
812	unsigned int reserved;
813	unsigned int unused;
814	unsigned int maxres;
815
816	gfs2_log_lock(sdp);
817
818	if (sdp->sd_log_tr) {
819		gfs2_merge_trans(sdp->sd_log_tr, tr);
820	} else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
821		gfs2_assert_withdraw(sdp, tr->tr_alloced);
822		sdp->sd_log_tr = tr;
823		tr->tr_attached = 1;
824	}
825
826	sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
827	reserved = calc_reserved(sdp);
828	maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
829	gfs2_assert_withdraw(sdp, maxres >= reserved);
830	unused = maxres - reserved;
831	atomic_add(unused, &sdp->sd_log_blks_free);
832	trace_gfs2_log_blocks(sdp, unused);
833	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
834			     sdp->sd_jdesc->jd_blocks);
835	sdp->sd_log_blks_reserved = reserved;
836
837	gfs2_log_unlock(sdp);
838}
839
840/**
841 * gfs2_log_commit - Commit a transaction to the log
842 * @sdp: the filesystem
843 * @tr: the transaction
844 *
845 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
846 * or the total number of used blocks (pinned blocks plus AIL blocks)
847 * is greater than thresh2.
848 *
849 * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
850 * journal size.
851 *
852 * Returns: errno
853 */
854
855void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
856{
857	log_refund(sdp, tr);
858
859	if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
860	    ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
861	    atomic_read(&sdp->sd_log_thresh2)))
862		wake_up(&sdp->sd_logd_waitq);
863}
864
865/**
866 * gfs2_log_shutdown - write a shutdown header into a journal
867 * @sdp: the filesystem
868 *
869 */
870
871void gfs2_log_shutdown(struct gfs2_sbd *sdp)
872{
873	gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
874	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
875	gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
876
877	sdp->sd_log_flush_head = sdp->sd_log_head;
878	sdp->sd_log_flush_wrapped = 0;
879
880	log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT);
881
882	gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
883	gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
884
885	sdp->sd_log_head = sdp->sd_log_flush_head;
886	sdp->sd_log_tail = sdp->sd_log_head;
887}
888
889static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
890{
891	return (atomic_read(&sdp->sd_log_pinned) >= atomic_read(&sdp->sd_log_thresh1));
892}
893
894static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
895{
896	unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
897	return used_blocks >= atomic_read(&sdp->sd_log_thresh2);
898}
899
900/**
901 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
902 * @sdp: Pointer to GFS2 superblock
903 *
904 * Also, periodically check to make sure that we're using the most recent
905 * journal index.
906 */
907
908int gfs2_logd(void *data)
909{
910	struct gfs2_sbd *sdp = data;
911	unsigned long t = 1;
912	DEFINE_WAIT(wait);
913
914	while (!kthread_should_stop()) {
915
916		if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
917			gfs2_ail1_empty(sdp);
918			gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
919		}
920
921		if (gfs2_ail_flush_reqd(sdp)) {
922			gfs2_ail1_start(sdp);
923			gfs2_ail1_wait(sdp);
924			gfs2_ail1_empty(sdp);
925			gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
926		}
927
928		if (!gfs2_ail_flush_reqd(sdp))
929			wake_up(&sdp->sd_log_waitq);
930
931		t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
932
933		try_to_freeze();
934
935		do {
936			prepare_to_wait(&sdp->sd_logd_waitq, &wait,
937					TASK_INTERRUPTIBLE);
938			if (!gfs2_ail_flush_reqd(sdp) &&
939			    !gfs2_jrnl_flush_reqd(sdp) &&
940			    !kthread_should_stop())
941				t = schedule_timeout(t);
942		} while(t && !gfs2_ail_flush_reqd(sdp) &&
943			!gfs2_jrnl_flush_reqd(sdp) &&
944			!kthread_should_stop());
945		finish_wait(&sdp->sd_logd_waitq, &wait);
946	}
947
948	return 0;
949}
950
951