xref: /linux/fs/gfs2/log.c (revision 63307d015b91e626c97bb82e88054af3d0b74643)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
5  */
6 
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>
22 
23 #include "gfs2.h"
24 #include "incore.h"
25 #include "bmap.h"
26 #include "glock.h"
27 #include "log.h"
28 #include "lops.h"
29 #include "meta_io.h"
30 #include "util.h"
31 #include "dir.h"
32 #include "trace_gfs2.h"
33 
34 /**
35  * gfs2_struct2blk - compute stuff
36  * @sdp: the filesystem
37  * @nstruct: the number of structures
38  * @ssize: the size of the structures
39  *
40  * Compute the number of log descriptor blocks needed to hold a certain number
41  * of structures of a certain size.
42  *
43  * Returns: the number of blocks needed (minimum is always 1)
44  */
45 
46 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
47 			     unsigned int ssize)
48 {
49 	unsigned int blks;
50 	unsigned int first, second;
51 
52 	blks = 1;
53 	first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
54 
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);
59 	}
60 
61 	return blks;
62 }
63 
64 /**
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
68  *
69  * The ail lock _must_ be held when calling this function
70  *
71  */
72 
73 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
74 {
75 	bd->bd_tr = NULL;
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);
79 	brelse(bd->bd_bh);
80 }
81 
82 /**
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
87  *
88  */
89 
90 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
91 			       struct writeback_control *wbc,
92 			       struct gfs2_trans *tr,
93 			       bool *withdraw)
94 __releases(&sdp->sd_ail_lock)
95 __acquires(&sdp->sd_ail_lock)
96 {
97 	struct gfs2_glock *gl = NULL;
98 	struct address_space *mapping;
99 	struct gfs2_bufdata *bd, *s;
100 	struct buffer_head *bh;
101 
102 	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
103 		bh = bd->bd_bh;
104 
105 		gfs2_assert(sdp, bd->bd_tr == tr);
106 
107 		if (!buffer_busy(bh)) {
108 			if (!buffer_uptodate(bh) &&
109 			    !test_and_set_bit(SDF_AIL1_IO_ERROR,
110 					      &sdp->sd_flags)) {
111 				gfs2_io_error_bh(sdp, bh);
112 				*withdraw = true;
113 			}
114 			list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
115 			continue;
116 		}
117 
118 		if (!buffer_dirty(bh))
119 			continue;
120 		if (gl == bd->bd_gl)
121 			continue;
122 		gl = bd->bd_gl;
123 		list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
124 		mapping = bh->b_page->mapping;
125 		if (!mapping)
126 			continue;
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)
131 			break;
132 		return 1;
133 	}
134 
135 	return 0;
136 }
137 
138 
139 /**
140  * gfs2_ail1_flush - start writeback of some ail1 entries
141  * @sdp: The super block
142  * @wbc: The writeback control structure
143  *
144  * Writes back some ail1 entries, according to the limits in the
145  * writeback control structure
146  */
147 
148 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
149 {
150 	struct list_head *head = &sdp->sd_ail1_list;
151 	struct gfs2_trans *tr;
152 	struct blk_plug plug;
153 	bool withdraw = false;
154 
155 	trace_gfs2_ail_flush(sdp, wbc, 1);
156 	blk_start_plug(&plug);
157 	spin_lock(&sdp->sd_ail_lock);
158 restart:
159 	list_for_each_entry_reverse(tr, head, tr_list) {
160 		if (wbc->nr_to_write <= 0)
161 			break;
162 		if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw))
163 			goto restart;
164 	}
165 	spin_unlock(&sdp->sd_ail_lock);
166 	blk_finish_plug(&plug);
167 	if (withdraw)
168 		gfs2_lm_withdraw(sdp, NULL);
169 	trace_gfs2_ail_flush(sdp, wbc, 0);
170 }
171 
172 /**
173  * gfs2_ail1_start - start writeback of all ail1 entries
174  * @sdp: The superblock
175  */
176 
177 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
178 {
179 	struct writeback_control wbc = {
180 		.sync_mode = WB_SYNC_NONE,
181 		.nr_to_write = LONG_MAX,
182 		.range_start = 0,
183 		.range_end = LLONG_MAX,
184 	};
185 
186 	return gfs2_ail1_flush(sdp, &wbc);
187 }
188 
189 /**
190  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
191  * @sdp: the filesystem
192  * @ai: the AIL entry
193  *
194  */
195 
196 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
197 				bool *withdraw)
198 {
199 	struct gfs2_bufdata *bd, *s;
200 	struct buffer_head *bh;
201 
202 	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
203 					 bd_ail_st_list) {
204 		bh = bd->bd_bh;
205 		gfs2_assert(sdp, bd->bd_tr == tr);
206 		if (buffer_busy(bh))
207 			continue;
208 		if (!buffer_uptodate(bh) &&
209 		    !test_and_set_bit(SDF_AIL1_IO_ERROR, &sdp->sd_flags)) {
210 			gfs2_io_error_bh(sdp, bh);
211 			*withdraw = true;
212 		}
213 		list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
214 	}
215 }
216 
217 /**
218  * gfs2_ail1_empty - Try to empty the ail1 lists
219  * @sdp: The superblock
220  *
221  * Tries to empty the ail1 lists, starting with the oldest first
222  */
223 
224 static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
225 {
226 	struct gfs2_trans *tr, *s;
227 	int oldest_tr = 1;
228 	int ret;
229 	bool withdraw = false;
230 
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);
236 		else
237 			oldest_tr = 0;
238 	}
239 	ret = list_empty(&sdp->sd_ail1_list);
240 	spin_unlock(&sdp->sd_ail_lock);
241 
242 	if (withdraw)
243 		gfs2_lm_withdraw(sdp, "fatal: I/O error(s)\n");
244 
245 	return ret;
246 }
247 
248 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
249 {
250 	struct gfs2_trans *tr;
251 	struct gfs2_bufdata *bd;
252 	struct buffer_head *bh;
253 
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) {
257 			bh = bd->bd_bh;
258 			if (!buffer_locked(bh))
259 				continue;
260 			get_bh(bh);
261 			spin_unlock(&sdp->sd_ail_lock);
262 			wait_on_buffer(bh);
263 			brelse(bh);
264 			return;
265 		}
266 	}
267 	spin_unlock(&sdp->sd_ail_lock);
268 }
269 
270 /**
271  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
272  * @sdp: the filesystem
273  * @ai: the AIL entry
274  *
275  */
276 
277 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
278 {
279 	struct list_head *head = &tr->tr_ail2_list;
280 	struct gfs2_bufdata *bd;
281 
282 	while (!list_empty(head)) {
283 		bd = list_entry(head->prev, struct gfs2_bufdata,
284 				bd_ail_st_list);
285 		gfs2_assert(sdp, bd->bd_tr == tr);
286 		gfs2_remove_from_ail(bd);
287 	}
288 }
289 
290 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
291 {
292 	struct gfs2_trans *tr, *safe;
293 	unsigned int old_tail = sdp->sd_log_tail;
294 	int wrap = (new_tail < old_tail);
295 	int a, b, rm;
296 
297 	spin_lock(&sdp->sd_ail_lock);
298 
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);
303 		if (!rm)
304 			continue;
305 
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));
310 		kfree(tr);
311 	}
312 
313 	spin_unlock(&sdp->sd_ail_lock);
314 }
315 
316 /**
317  * gfs2_log_release - Release a given number of log blocks
318  * @sdp: The GFS2 superblock
319  * @blks: The number of blocks
320  *
321  */
322 
323 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
324 {
325 
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);
331 }
332 
333 /**
334  * gfs2_log_reserve - Make a log reservation
335  * @sdp: The GFS2 superblock
336  * @blks: The number of blocks to reserve
337  *
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.
343  *
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.
349  *
350  * Returns: errno
351  */
352 
353 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
354 {
355 	int ret = 0;
356 	unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
357 	unsigned wanted = blks + reserved_blks;
358 	DEFINE_WAIT(wait);
359 	int did_wait = 0;
360 	unsigned int free_blocks;
361 
362 	if (gfs2_assert_warn(sdp, blks) ||
363 	    gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
364 		return -EINVAL;
365 	atomic_add(blks, &sdp->sd_log_blks_needed);
366 retry:
367 	free_blocks = atomic_read(&sdp->sd_log_blks_free);
368 	if (unlikely(free_blocks <= wanted)) {
369 		do {
370 			prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
371 					TASK_UNINTERRUPTIBLE);
372 			wake_up(&sdp->sd_logd_waitq);
373 			did_wait = 1;
374 			if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
375 				io_schedule();
376 			free_blocks = atomic_read(&sdp->sd_log_blks_free);
377 		} while(free_blocks <= wanted);
378 		finish_wait(&sdp->sd_log_waitq, &wait);
379 	}
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);
385 		goto retry;
386 	}
387 	atomic_sub(blks, &sdp->sd_log_blks_needed);
388 	trace_gfs2_log_blocks(sdp, -blks);
389 
390 	/*
391 	 * If we waited, then so might others, wake them up _after_ we get
392 	 * our share of the log.
393 	 */
394 	if (unlikely(did_wait))
395 		wake_up(&sdp->sd_log_waitq);
396 
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);
400 		ret = -EROFS;
401 	}
402 	if (atomic_dec_and_test(&sdp->sd_reserving_log))
403 		wake_up(&sdp->sd_reserving_log_wait);
404 	return ret;
405 }
406 
407 /**
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
412  *
413  *   Compute the distance (in the journal direction) between two
414  *   blocks in the journal
415  *
416  * Returns: the distance in blocks
417  */
418 
419 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
420 					unsigned int older)
421 {
422 	int dist;
423 
424 	dist = newer - older;
425 	if (dist < 0)
426 		dist += sdp->sd_jdesc->jd_blocks;
427 
428 	return dist;
429 }
430 
431 /**
432  * calc_reserved - Calculate the number of blocks to reserve when
433  *                 refunding a transaction's unused buffers.
434  * @sdp: The GFS2 superblock
435  *
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.
442  *
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.
450  *
451  * Also, we need to reserve blocks for revoke journal entries and one for an
452  * overall header for the lot.
453  *
454  * Returns: the number of blocks reserved
455  */
456 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
457 {
458 	unsigned int reserved = 0;
459 	unsigned int mbuf;
460 	unsigned int dbuf;
461 	struct gfs2_trans *tr = sdp->sd_log_tr;
462 
463 	if (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));
470 	}
471 
472 	if (sdp->sd_log_commited_revoke > 0)
473 		reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
474 					  sizeof(u64));
475 	/* One for the overall header */
476 	if (reserved)
477 		reserved++;
478 	return reserved;
479 }
480 
481 static unsigned int current_tail(struct gfs2_sbd *sdp)
482 {
483 	struct gfs2_trans *tr;
484 	unsigned int tail;
485 
486 	spin_lock(&sdp->sd_ail_lock);
487 
488 	if (list_empty(&sdp->sd_ail1_list)) {
489 		tail = sdp->sd_log_head;
490 	} else {
491 		tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
492 				tr_list);
493 		tail = tr->tr_first;
494 	}
495 
496 	spin_unlock(&sdp->sd_ail_lock);
497 
498 	return tail;
499 }
500 
501 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
502 {
503 	unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
504 
505 	ail2_empty(sdp, new_tail);
506 
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);
511 
512 	sdp->sd_log_tail = new_tail;
513 }
514 
515 
516 static void log_flush_wait(struct gfs2_sbd *sdp)
517 {
518 	DEFINE_WAIT(wait);
519 
520 	if (atomic_read(&sdp->sd_log_in_flight)) {
521 		do {
522 			prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
523 					TASK_UNINTERRUPTIBLE);
524 			if (atomic_read(&sdp->sd_log_in_flight))
525 				io_schedule();
526 		} while(atomic_read(&sdp->sd_log_in_flight));
527 		finish_wait(&sdp->sd_log_flush_wait, &wait);
528 	}
529 }
530 
531 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
532 {
533 	struct gfs2_inode *ipa, *ipb;
534 
535 	ipa = list_entry(a, struct gfs2_inode, i_ordered);
536 	ipb = list_entry(b, struct gfs2_inode, i_ordered);
537 
538 	if (ipa->i_no_addr < ipb->i_no_addr)
539 		return -1;
540 	if (ipa->i_no_addr > ipb->i_no_addr)
541 		return 1;
542 	return 0;
543 }
544 
545 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
546 {
547 	struct gfs2_inode *ip;
548 	LIST_HEAD(written);
549 
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);
557 			continue;
558 		}
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);
563 	}
564 	list_splice(&written, &sdp->sd_log_ordered);
565 	spin_unlock(&sdp->sd_ordered_lock);
566 }
567 
568 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
569 {
570 	struct gfs2_inode *ip;
571 
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)
578 			continue;
579 		spin_unlock(&sdp->sd_ordered_lock);
580 		filemap_fdatawait(ip->i_inode.i_mapping);
581 		spin_lock(&sdp->sd_ordered_lock);
582 	}
583 	spin_unlock(&sdp->sd_ordered_lock);
584 }
585 
586 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
587 {
588 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
589 
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);
594 }
595 
596 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
597 {
598 	struct buffer_head *bh = bd->bd_bh;
599 	struct gfs2_glock *gl = bd->bd_gl;
600 
601 	bh->b_private = NULL;
602 	bd->bd_blkno = bh->b_blocknr;
603 	gfs2_remove_from_ail(bd); /* drops ref on bh */
604 	bd->bd_bh = NULL;
605 	sdp->sd_log_num_revoke++;
606 	if (atomic_inc_return(&gl->gl_revokes) == 1)
607 		gfs2_glock_hold(gl);
608 	set_bit(GLF_LFLUSH, &gl->gl_flags);
609 	list_add(&bd->bd_list, &sdp->sd_log_revokes);
610 }
611 
612 void gfs2_write_revokes(struct gfs2_sbd *sdp)
613 {
614 	struct gfs2_trans *tr;
615 	struct gfs2_bufdata *bd, *tmp;
616 	int have_revokes = 0;
617 	int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
618 
619 	gfs2_ail1_empty(sdp);
620 	spin_lock(&sdp->sd_ail_lock);
621 	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
622 		list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
623 			if (list_empty(&bd->bd_list)) {
624 				have_revokes = 1;
625 				goto done;
626 			}
627 		}
628 	}
629 done:
630 	spin_unlock(&sdp->sd_ail_lock);
631 	if (have_revokes == 0)
632 		return;
633 	while (sdp->sd_log_num_revoke > max_revokes)
634 		max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
635 	max_revokes -= sdp->sd_log_num_revoke;
636 	if (!sdp->sd_log_num_revoke) {
637 		atomic_dec(&sdp->sd_log_blks_free);
638 		/* If no blocks have been reserved, we need to also
639 		 * reserve a block for the header */
640 		if (!sdp->sd_log_blks_reserved)
641 			atomic_dec(&sdp->sd_log_blks_free);
642 	}
643 	gfs2_log_lock(sdp);
644 	spin_lock(&sdp->sd_ail_lock);
645 	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
646 		list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
647 			if (max_revokes == 0)
648 				goto out_of_blocks;
649 			if (!list_empty(&bd->bd_list))
650 				continue;
651 			gfs2_add_revoke(sdp, bd);
652 			max_revokes--;
653 		}
654 	}
655 out_of_blocks:
656 	spin_unlock(&sdp->sd_ail_lock);
657 	gfs2_log_unlock(sdp);
658 
659 	if (!sdp->sd_log_num_revoke) {
660 		atomic_inc(&sdp->sd_log_blks_free);
661 		if (!sdp->sd_log_blks_reserved)
662 			atomic_inc(&sdp->sd_log_blks_free);
663 	}
664 }
665 
666 /**
667  * gfs2_write_log_header - Write a journal log header buffer at lblock
668  * @sdp: The GFS2 superblock
669  * @jd: journal descriptor of the journal to which we are writing
670  * @seq: sequence number
671  * @tail: tail of the log
672  * @lblock: value for lh_blkno (block number relative to start of journal)
673  * @flags: log header flags GFS2_LOG_HEAD_*
674  * @op_flags: flags to pass to the bio
675  *
676  * Returns: the initialized log buffer descriptor
677  */
678 
679 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
680 			   u64 seq, u32 tail, u32 lblock, u32 flags,
681 			   int op_flags)
682 {
683 	struct gfs2_log_header *lh;
684 	u32 hash, crc;
685 	struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
686 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
687 	struct timespec64 tv;
688 	struct super_block *sb = sdp->sd_vfs;
689 	u64 dblock;
690 
691 	lh = page_address(page);
692 	clear_page(lh);
693 
694 	lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
695 	lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
696 	lh->lh_header.__pad0 = cpu_to_be64(0);
697 	lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
698 	lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
699 	lh->lh_sequence = cpu_to_be64(seq);
700 	lh->lh_flags = cpu_to_be32(flags);
701 	lh->lh_tail = cpu_to_be32(tail);
702 	lh->lh_blkno = cpu_to_be32(lblock);
703 	hash = ~crc32(~0, lh, LH_V1_SIZE);
704 	lh->lh_hash = cpu_to_be32(hash);
705 
706 	ktime_get_coarse_real_ts64(&tv);
707 	lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
708 	lh->lh_sec = cpu_to_be64(tv.tv_sec);
709 	if (!list_empty(&jd->extent_list))
710 		dblock = gfs2_log_bmap(sdp);
711 	else {
712 		int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
713 		if (gfs2_assert_withdraw(sdp, ret == 0))
714 			return;
715 	}
716 	lh->lh_addr = cpu_to_be64(dblock);
717 	lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
718 
719 	/* We may only write local statfs, quota, etc., when writing to our
720 	   own journal. The values are left 0 when recovering a journal
721 	   different from our own. */
722 	if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
723 		lh->lh_statfs_addr =
724 			cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
725 		lh->lh_quota_addr =
726 			cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
727 
728 		spin_lock(&sdp->sd_statfs_spin);
729 		lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
730 		lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
731 		lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
732 		spin_unlock(&sdp->sd_statfs_spin);
733 	}
734 
735 	BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
736 
737 	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
738 		     sb->s_blocksize - LH_V1_SIZE - 4);
739 	lh->lh_crc = cpu_to_be32(crc);
740 
741 	gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
742 	gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
743 	log_flush_wait(sdp);
744 }
745 
746 /**
747  * log_write_header - Get and initialize a journal header buffer
748  * @sdp: The GFS2 superblock
749  * @flags: The log header flags, including log header origin
750  *
751  * Returns: the initialized log buffer descriptor
752  */
753 
754 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
755 {
756 	unsigned int tail;
757 	int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
758 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
759 
760 	gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
761 	tail = current_tail(sdp);
762 
763 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
764 		gfs2_ordered_wait(sdp);
765 		log_flush_wait(sdp);
766 		op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
767 	}
768 	sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
769 	gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
770 			      sdp->sd_log_flush_head, flags, op_flags);
771 
772 	if (sdp->sd_log_tail != tail)
773 		log_pull_tail(sdp, tail);
774 }
775 
776 /**
777  * gfs2_log_flush - flush incore transaction(s)
778  * @sdp: the filesystem
779  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
780  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
781  *
782  */
783 
784 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
785 {
786 	struct gfs2_trans *tr;
787 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
788 
789 	down_write(&sdp->sd_log_flush_lock);
790 
791 	/* Log might have been flushed while we waited for the flush lock */
792 	if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
793 		up_write(&sdp->sd_log_flush_lock);
794 		return;
795 	}
796 	trace_gfs2_log_flush(sdp, 1, flags);
797 
798 	if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
799 		clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
800 
801 	sdp->sd_log_flush_head = sdp->sd_log_head;
802 	tr = sdp->sd_log_tr;
803 	if (tr) {
804 		sdp->sd_log_tr = NULL;
805 		INIT_LIST_HEAD(&tr->tr_ail1_list);
806 		INIT_LIST_HEAD(&tr->tr_ail2_list);
807 		tr->tr_first = sdp->sd_log_flush_head;
808 		if (unlikely (state == SFS_FROZEN))
809 			gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
810 	}
811 
812 	if (unlikely(state == SFS_FROZEN))
813 		gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
814 	gfs2_assert_withdraw(sdp,
815 			sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
816 
817 	gfs2_ordered_write(sdp);
818 	lops_before_commit(sdp, tr);
819 	gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
820 
821 	if (sdp->sd_log_head != sdp->sd_log_flush_head) {
822 		log_flush_wait(sdp);
823 		log_write_header(sdp, flags);
824 	} else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
825 		atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
826 		trace_gfs2_log_blocks(sdp, -1);
827 		log_write_header(sdp, flags);
828 	}
829 	lops_after_commit(sdp, tr);
830 
831 	gfs2_log_lock(sdp);
832 	sdp->sd_log_head = sdp->sd_log_flush_head;
833 	sdp->sd_log_blks_reserved = 0;
834 	sdp->sd_log_commited_revoke = 0;
835 
836 	spin_lock(&sdp->sd_ail_lock);
837 	if (tr && !list_empty(&tr->tr_ail1_list)) {
838 		list_add(&tr->tr_list, &sdp->sd_ail1_list);
839 		tr = NULL;
840 	}
841 	spin_unlock(&sdp->sd_ail_lock);
842 	gfs2_log_unlock(sdp);
843 
844 	if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
845 		if (!sdp->sd_log_idle) {
846 			for (;;) {
847 				gfs2_ail1_start(sdp);
848 				gfs2_ail1_wait(sdp);
849 				if (gfs2_ail1_empty(sdp))
850 					break;
851 			}
852 			atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
853 			trace_gfs2_log_blocks(sdp, -1);
854 			log_write_header(sdp, flags);
855 			sdp->sd_log_head = sdp->sd_log_flush_head;
856 		}
857 		if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
858 			     GFS2_LOG_HEAD_FLUSH_FREEZE))
859 			gfs2_log_shutdown(sdp);
860 		if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
861 			atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
862 	}
863 
864 	trace_gfs2_log_flush(sdp, 0, flags);
865 	up_write(&sdp->sd_log_flush_lock);
866 
867 	kfree(tr);
868 }
869 
870 /**
871  * gfs2_merge_trans - Merge a new transaction into a cached transaction
872  * @old: Original transaction to be expanded
873  * @new: New transaction to be merged
874  */
875 
876 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
877 {
878 	WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
879 
880 	old->tr_num_buf_new	+= new->tr_num_buf_new;
881 	old->tr_num_databuf_new	+= new->tr_num_databuf_new;
882 	old->tr_num_buf_rm	+= new->tr_num_buf_rm;
883 	old->tr_num_databuf_rm	+= new->tr_num_databuf_rm;
884 	old->tr_num_revoke	+= new->tr_num_revoke;
885 	old->tr_num_revoke_rm	+= new->tr_num_revoke_rm;
886 
887 	list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
888 	list_splice_tail_init(&new->tr_buf, &old->tr_buf);
889 }
890 
891 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
892 {
893 	unsigned int reserved;
894 	unsigned int unused;
895 	unsigned int maxres;
896 
897 	gfs2_log_lock(sdp);
898 
899 	if (sdp->sd_log_tr) {
900 		gfs2_merge_trans(sdp->sd_log_tr, tr);
901 	} else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
902 		gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
903 		sdp->sd_log_tr = tr;
904 		set_bit(TR_ATTACHED, &tr->tr_flags);
905 	}
906 
907 	sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
908 	reserved = calc_reserved(sdp);
909 	maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
910 	gfs2_assert_withdraw(sdp, maxres >= reserved);
911 	unused = maxres - reserved;
912 	atomic_add(unused, &sdp->sd_log_blks_free);
913 	trace_gfs2_log_blocks(sdp, unused);
914 	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
915 			     sdp->sd_jdesc->jd_blocks);
916 	sdp->sd_log_blks_reserved = reserved;
917 
918 	gfs2_log_unlock(sdp);
919 }
920 
921 /**
922  * gfs2_log_commit - Commit a transaction to the log
923  * @sdp: the filesystem
924  * @tr: the transaction
925  *
926  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
927  * or the total number of used blocks (pinned blocks plus AIL blocks)
928  * is greater than thresh2.
929  *
930  * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
931  * journal size.
932  *
933  * Returns: errno
934  */
935 
936 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
937 {
938 	log_refund(sdp, tr);
939 
940 	if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
941 	    ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
942 	    atomic_read(&sdp->sd_log_thresh2)))
943 		wake_up(&sdp->sd_logd_waitq);
944 }
945 
946 /**
947  * gfs2_log_shutdown - write a shutdown header into a journal
948  * @sdp: the filesystem
949  *
950  */
951 
952 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
953 {
954 	gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
955 	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
956 	gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
957 
958 	sdp->sd_log_flush_head = sdp->sd_log_head;
959 
960 	log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
961 
962 	gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
963 	gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
964 
965 	sdp->sd_log_head = sdp->sd_log_flush_head;
966 	sdp->sd_log_tail = sdp->sd_log_head;
967 }
968 
969 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
970 {
971 	return (atomic_read(&sdp->sd_log_pinned) +
972 		atomic_read(&sdp->sd_log_blks_needed) >=
973 		atomic_read(&sdp->sd_log_thresh1));
974 }
975 
976 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
977 {
978 	unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
979 
980 	if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
981 		return 1;
982 
983 	return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
984 		atomic_read(&sdp->sd_log_thresh2);
985 }
986 
987 /**
988  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
989  * @sdp: Pointer to GFS2 superblock
990  *
991  * Also, periodically check to make sure that we're using the most recent
992  * journal index.
993  */
994 
995 int gfs2_logd(void *data)
996 {
997 	struct gfs2_sbd *sdp = data;
998 	unsigned long t = 1;
999 	DEFINE_WAIT(wait);
1000 	bool did_flush;
1001 
1002 	while (!kthread_should_stop()) {
1003 
1004 		/* Check for errors writing to the journal */
1005 		if (sdp->sd_log_error) {
1006 			gfs2_lm_withdraw(sdp,
1007 					 "GFS2: fsid=%s: error %d: "
1008 					 "withdrawing the file system to "
1009 					 "prevent further damage.\n",
1010 					 sdp->sd_fsname, sdp->sd_log_error);
1011 		}
1012 
1013 		did_flush = false;
1014 		if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1015 			gfs2_ail1_empty(sdp);
1016 			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1017 				       GFS2_LFC_LOGD_JFLUSH_REQD);
1018 			did_flush = true;
1019 		}
1020 
1021 		if (gfs2_ail_flush_reqd(sdp)) {
1022 			gfs2_ail1_start(sdp);
1023 			gfs2_ail1_wait(sdp);
1024 			gfs2_ail1_empty(sdp);
1025 			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1026 				       GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1027 			did_flush = true;
1028 		}
1029 
1030 		if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1031 			wake_up(&sdp->sd_log_waitq);
1032 
1033 		t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1034 
1035 		try_to_freeze();
1036 
1037 		do {
1038 			prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1039 					TASK_INTERRUPTIBLE);
1040 			if (!gfs2_ail_flush_reqd(sdp) &&
1041 			    !gfs2_jrnl_flush_reqd(sdp) &&
1042 			    !kthread_should_stop())
1043 				t = schedule_timeout(t);
1044 		} while(t && !gfs2_ail_flush_reqd(sdp) &&
1045 			!gfs2_jrnl_flush_reqd(sdp) &&
1046 			!kthread_should_stop());
1047 		finish_wait(&sdp->sd_logd_waitq, &wait);
1048 	}
1049 
1050 	return 0;
1051 }
1052 
1053