xref: /openbmc/linux/fs/gfs2/log.c (revision 021ed9f5)
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 static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
35 
36 /**
37  * gfs2_struct2blk - compute stuff
38  * @sdp: the filesystem
39  * @nstruct: the number of structures
40  *
41  * Compute the number of log descriptor blocks needed to hold a certain number
42  * of structures of a certain size.
43  *
44  * Returns: the number of blocks needed (minimum is always 1)
45  */
46 
47 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
48 {
49 	unsigned int blks;
50 	unsigned int first, second;
51 
52 	blks = 1;
53 	first = sdp->sd_ldptrs;
54 
55 	if (nstruct > first) {
56 		second = sdp->sd_inptrs;
57 		blks += DIV_ROUND_UP(nstruct - first, second);
58 	}
59 
60 	return blks;
61 }
62 
63 /**
64  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
65  * @mapping: The associated mapping (maybe NULL)
66  * @bd: The gfs2_bufdata to remove
67  *
68  * The ail lock _must_ be held when calling this function
69  *
70  */
71 
72 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
73 {
74 	bd->bd_tr = NULL;
75 	list_del_init(&bd->bd_ail_st_list);
76 	list_del_init(&bd->bd_ail_gl_list);
77 	atomic_dec(&bd->bd_gl->gl_ail_count);
78 	brelse(bd->bd_bh);
79 }
80 
81 /**
82  * gfs2_ail1_start_one - Start I/O on a part of the AIL
83  * @sdp: the filesystem
84  * @wbc: The writeback control structure
85  * @ai: The ail structure
86  *
87  */
88 
89 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
90 			       struct writeback_control *wbc,
91 			       struct gfs2_trans *tr,
92 			       bool *withdraw)
93 __releases(&sdp->sd_ail_lock)
94 __acquires(&sdp->sd_ail_lock)
95 {
96 	struct gfs2_glock *gl = NULL;
97 	struct address_space *mapping;
98 	struct gfs2_bufdata *bd, *s;
99 	struct buffer_head *bh;
100 
101 	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
102 		bh = bd->bd_bh;
103 
104 		gfs2_assert(sdp, bd->bd_tr == tr);
105 
106 		if (!buffer_busy(bh)) {
107 			if (!buffer_uptodate(bh) &&
108 			    !test_and_set_bit(SDF_AIL1_IO_ERROR,
109 					      &sdp->sd_flags)) {
110 				gfs2_io_error_bh(sdp, bh);
111 				*withdraw = true;
112 			}
113 			list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
114 			continue;
115 		}
116 
117 		if (!buffer_dirty(bh))
118 			continue;
119 		if (gl == bd->bd_gl)
120 			continue;
121 		gl = bd->bd_gl;
122 		list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
123 		mapping = bh->b_page->mapping;
124 		if (!mapping)
125 			continue;
126 		spin_unlock(&sdp->sd_ail_lock);
127 		generic_writepages(mapping, wbc);
128 		spin_lock(&sdp->sd_ail_lock);
129 		if (wbc->nr_to_write <= 0)
130 			break;
131 		return 1;
132 	}
133 
134 	return 0;
135 }
136 
137 
138 /**
139  * gfs2_ail1_flush - start writeback of some ail1 entries
140  * @sdp: The super block
141  * @wbc: The writeback control structure
142  *
143  * Writes back some ail1 entries, according to the limits in the
144  * writeback control structure
145  */
146 
147 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
148 {
149 	struct list_head *head = &sdp->sd_ail1_list;
150 	struct gfs2_trans *tr;
151 	struct blk_plug plug;
152 	bool withdraw = false;
153 
154 	trace_gfs2_ail_flush(sdp, wbc, 1);
155 	blk_start_plug(&plug);
156 	spin_lock(&sdp->sd_ail_lock);
157 restart:
158 	list_for_each_entry_reverse(tr, head, tr_list) {
159 		if (wbc->nr_to_write <= 0)
160 			break;
161 		if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw) &&
162 		    !gfs2_withdrawn(sdp))
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_committed_revoke > 0)
473 		reserved += gfs2_struct2blk(sdp, sdp->sd_log_committed_revoke);
474 	/* One for the overall header */
475 	if (reserved)
476 		reserved++;
477 	return reserved;
478 }
479 
480 static unsigned int current_tail(struct gfs2_sbd *sdp)
481 {
482 	struct gfs2_trans *tr;
483 	unsigned int tail;
484 
485 	spin_lock(&sdp->sd_ail_lock);
486 
487 	if (list_empty(&sdp->sd_ail1_list)) {
488 		tail = sdp->sd_log_head;
489 	} else {
490 		tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
491 				tr_list);
492 		tail = tr->tr_first;
493 	}
494 
495 	spin_unlock(&sdp->sd_ail_lock);
496 
497 	return tail;
498 }
499 
500 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
501 {
502 	unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
503 
504 	ail2_empty(sdp, new_tail);
505 
506 	atomic_add(dist, &sdp->sd_log_blks_free);
507 	trace_gfs2_log_blocks(sdp, dist);
508 	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
509 			     sdp->sd_jdesc->jd_blocks);
510 
511 	sdp->sd_log_tail = new_tail;
512 }
513 
514 
515 static void log_flush_wait(struct gfs2_sbd *sdp)
516 {
517 	DEFINE_WAIT(wait);
518 
519 	if (atomic_read(&sdp->sd_log_in_flight)) {
520 		do {
521 			prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
522 					TASK_UNINTERRUPTIBLE);
523 			if (atomic_read(&sdp->sd_log_in_flight))
524 				io_schedule();
525 		} while(atomic_read(&sdp->sd_log_in_flight));
526 		finish_wait(&sdp->sd_log_flush_wait, &wait);
527 	}
528 }
529 
530 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
531 {
532 	struct gfs2_inode *ipa, *ipb;
533 
534 	ipa = list_entry(a, struct gfs2_inode, i_ordered);
535 	ipb = list_entry(b, struct gfs2_inode, i_ordered);
536 
537 	if (ipa->i_no_addr < ipb->i_no_addr)
538 		return -1;
539 	if (ipa->i_no_addr > ipb->i_no_addr)
540 		return 1;
541 	return 0;
542 }
543 
544 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
545 {
546 	struct gfs2_inode *ip;
547 	LIST_HEAD(written);
548 
549 	spin_lock(&sdp->sd_ordered_lock);
550 	list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
551 	while (!list_empty(&sdp->sd_log_ordered)) {
552 		ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered);
553 		if (ip->i_inode.i_mapping->nrpages == 0) {
554 			test_and_clear_bit(GIF_ORDERED, &ip->i_flags);
555 			list_del(&ip->i_ordered);
556 			continue;
557 		}
558 		list_move(&ip->i_ordered, &written);
559 		spin_unlock(&sdp->sd_ordered_lock);
560 		filemap_fdatawrite(ip->i_inode.i_mapping);
561 		spin_lock(&sdp->sd_ordered_lock);
562 	}
563 	list_splice(&written, &sdp->sd_log_ordered);
564 	spin_unlock(&sdp->sd_ordered_lock);
565 }
566 
567 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
568 {
569 	struct gfs2_inode *ip;
570 
571 	spin_lock(&sdp->sd_ordered_lock);
572 	while (!list_empty(&sdp->sd_log_ordered)) {
573 		ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered);
574 		list_del(&ip->i_ordered);
575 		WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
576 		if (ip->i_inode.i_mapping->nrpages == 0)
577 			continue;
578 		spin_unlock(&sdp->sd_ordered_lock);
579 		filemap_fdatawait(ip->i_inode.i_mapping);
580 		spin_lock(&sdp->sd_ordered_lock);
581 	}
582 	spin_unlock(&sdp->sd_ordered_lock);
583 }
584 
585 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
586 {
587 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
588 
589 	spin_lock(&sdp->sd_ordered_lock);
590 	if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
591 		list_del(&ip->i_ordered);
592 	spin_unlock(&sdp->sd_ordered_lock);
593 }
594 
595 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
596 {
597 	struct buffer_head *bh = bd->bd_bh;
598 	struct gfs2_glock *gl = bd->bd_gl;
599 
600 	bh->b_private = NULL;
601 	bd->bd_blkno = bh->b_blocknr;
602 	gfs2_remove_from_ail(bd); /* drops ref on bh */
603 	bd->bd_bh = NULL;
604 	sdp->sd_log_num_revoke++;
605 	if (atomic_inc_return(&gl->gl_revokes) == 1)
606 		gfs2_glock_hold(gl);
607 	set_bit(GLF_LFLUSH, &gl->gl_flags);
608 	list_add(&bd->bd_list, &sdp->sd_log_revokes);
609 }
610 
611 void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
612 {
613 	if (atomic_dec_return(&gl->gl_revokes) == 0) {
614 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
615 		gfs2_glock_queue_put(gl);
616 	}
617 }
618 
619 void gfs2_write_revokes(struct gfs2_sbd *sdp)
620 {
621 	struct gfs2_trans *tr;
622 	struct gfs2_bufdata *bd, *tmp;
623 	int have_revokes = 0;
624 	int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
625 
626 	gfs2_ail1_empty(sdp);
627 	spin_lock(&sdp->sd_ail_lock);
628 	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
629 		list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
630 			if (list_empty(&bd->bd_list)) {
631 				have_revokes = 1;
632 				goto done;
633 			}
634 		}
635 	}
636 done:
637 	spin_unlock(&sdp->sd_ail_lock);
638 	if (have_revokes == 0)
639 		return;
640 	while (sdp->sd_log_num_revoke > max_revokes)
641 		max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
642 	max_revokes -= sdp->sd_log_num_revoke;
643 	if (!sdp->sd_log_num_revoke) {
644 		atomic_dec(&sdp->sd_log_blks_free);
645 		/* If no blocks have been reserved, we need to also
646 		 * reserve a block for the header */
647 		if (!sdp->sd_log_blks_reserved)
648 			atomic_dec(&sdp->sd_log_blks_free);
649 	}
650 	gfs2_log_lock(sdp);
651 	spin_lock(&sdp->sd_ail_lock);
652 	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
653 		list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
654 			if (max_revokes == 0)
655 				goto out_of_blocks;
656 			if (!list_empty(&bd->bd_list))
657 				continue;
658 			gfs2_add_revoke(sdp, bd);
659 			max_revokes--;
660 		}
661 	}
662 out_of_blocks:
663 	spin_unlock(&sdp->sd_ail_lock);
664 	gfs2_log_unlock(sdp);
665 
666 	if (!sdp->sd_log_num_revoke) {
667 		atomic_inc(&sdp->sd_log_blks_free);
668 		if (!sdp->sd_log_blks_reserved)
669 			atomic_inc(&sdp->sd_log_blks_free);
670 	}
671 }
672 
673 /**
674  * gfs2_write_log_header - Write a journal log header buffer at lblock
675  * @sdp: The GFS2 superblock
676  * @jd: journal descriptor of the journal to which we are writing
677  * @seq: sequence number
678  * @tail: tail of the log
679  * @lblock: value for lh_blkno (block number relative to start of journal)
680  * @flags: log header flags GFS2_LOG_HEAD_*
681  * @op_flags: flags to pass to the bio
682  *
683  * Returns: the initialized log buffer descriptor
684  */
685 
686 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
687 			   u64 seq, u32 tail, u32 lblock, u32 flags,
688 			   int op_flags)
689 {
690 	struct gfs2_log_header *lh;
691 	u32 hash, crc;
692 	struct page *page;
693 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
694 	struct timespec64 tv;
695 	struct super_block *sb = sdp->sd_vfs;
696 	u64 dblock;
697 
698 	if (gfs2_withdrawn(sdp))
699 		goto out;
700 
701 	page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
702 	lh = page_address(page);
703 	clear_page(lh);
704 
705 	lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
706 	lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
707 	lh->lh_header.__pad0 = cpu_to_be64(0);
708 	lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
709 	lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
710 	lh->lh_sequence = cpu_to_be64(seq);
711 	lh->lh_flags = cpu_to_be32(flags);
712 	lh->lh_tail = cpu_to_be32(tail);
713 	lh->lh_blkno = cpu_to_be32(lblock);
714 	hash = ~crc32(~0, lh, LH_V1_SIZE);
715 	lh->lh_hash = cpu_to_be32(hash);
716 
717 	ktime_get_coarse_real_ts64(&tv);
718 	lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
719 	lh->lh_sec = cpu_to_be64(tv.tv_sec);
720 	if (!list_empty(&jd->extent_list))
721 		dblock = gfs2_log_bmap(jd, lblock);
722 	else {
723 		int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
724 		if (gfs2_assert_withdraw(sdp, ret == 0))
725 			return;
726 	}
727 	lh->lh_addr = cpu_to_be64(dblock);
728 	lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
729 
730 	/* We may only write local statfs, quota, etc., when writing to our
731 	   own journal. The values are left 0 when recovering a journal
732 	   different from our own. */
733 	if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
734 		lh->lh_statfs_addr =
735 			cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
736 		lh->lh_quota_addr =
737 			cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
738 
739 		spin_lock(&sdp->sd_statfs_spin);
740 		lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
741 		lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
742 		lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
743 		spin_unlock(&sdp->sd_statfs_spin);
744 	}
745 
746 	BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
747 
748 	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
749 		     sb->s_blocksize - LH_V1_SIZE - 4);
750 	lh->lh_crc = cpu_to_be32(crc);
751 
752 	gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
753 	gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
754 out:
755 	log_flush_wait(sdp);
756 }
757 
758 /**
759  * log_write_header - Get and initialize a journal header buffer
760  * @sdp: The GFS2 superblock
761  * @flags: The log header flags, including log header origin
762  *
763  * Returns: the initialized log buffer descriptor
764  */
765 
766 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
767 {
768 	unsigned int tail;
769 	int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
770 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
771 
772 	gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
773 	tail = current_tail(sdp);
774 
775 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
776 		gfs2_ordered_wait(sdp);
777 		log_flush_wait(sdp);
778 		op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
779 	}
780 	sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
781 	gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
782 			      sdp->sd_log_flush_head, flags, op_flags);
783 	gfs2_log_incr_head(sdp);
784 
785 	if (sdp->sd_log_tail != tail)
786 		log_pull_tail(sdp, tail);
787 }
788 
789 /**
790  * gfs2_log_flush - flush incore transaction(s)
791  * @sdp: the filesystem
792  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
793  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
794  *
795  */
796 
797 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
798 {
799 	struct gfs2_trans *tr;
800 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
801 
802 	down_write(&sdp->sd_log_flush_lock);
803 
804 	/* Log might have been flushed while we waited for the flush lock */
805 	if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
806 		up_write(&sdp->sd_log_flush_lock);
807 		return;
808 	}
809 	trace_gfs2_log_flush(sdp, 1, flags);
810 
811 	if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
812 		clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
813 
814 	sdp->sd_log_flush_head = sdp->sd_log_head;
815 	tr = sdp->sd_log_tr;
816 	if (tr) {
817 		sdp->sd_log_tr = NULL;
818 		INIT_LIST_HEAD(&tr->tr_ail1_list);
819 		INIT_LIST_HEAD(&tr->tr_ail2_list);
820 		tr->tr_first = sdp->sd_log_flush_head;
821 		if (unlikely (state == SFS_FROZEN))
822 			gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
823 	}
824 
825 	if (unlikely(state == SFS_FROZEN))
826 		gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
827 	gfs2_assert_withdraw(sdp,
828 			sdp->sd_log_num_revoke == sdp->sd_log_committed_revoke);
829 
830 	gfs2_ordered_write(sdp);
831 	lops_before_commit(sdp, tr);
832 	gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
833 
834 	if (sdp->sd_log_head != sdp->sd_log_flush_head) {
835 		log_flush_wait(sdp);
836 		log_write_header(sdp, flags);
837 	} else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
838 		atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
839 		trace_gfs2_log_blocks(sdp, -1);
840 		log_write_header(sdp, flags);
841 	}
842 	lops_after_commit(sdp, tr);
843 
844 	gfs2_log_lock(sdp);
845 	sdp->sd_log_head = sdp->sd_log_flush_head;
846 	sdp->sd_log_blks_reserved = 0;
847 	sdp->sd_log_committed_revoke = 0;
848 
849 	spin_lock(&sdp->sd_ail_lock);
850 	if (tr && !list_empty(&tr->tr_ail1_list)) {
851 		list_add(&tr->tr_list, &sdp->sd_ail1_list);
852 		tr = NULL;
853 	}
854 	spin_unlock(&sdp->sd_ail_lock);
855 	gfs2_log_unlock(sdp);
856 
857 	if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
858 		if (!sdp->sd_log_idle) {
859 			for (;;) {
860 				gfs2_ail1_start(sdp);
861 				gfs2_ail1_wait(sdp);
862 				if (gfs2_ail1_empty(sdp))
863 					break;
864 			}
865 			atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
866 			trace_gfs2_log_blocks(sdp, -1);
867 			log_write_header(sdp, flags);
868 			sdp->sd_log_head = sdp->sd_log_flush_head;
869 		}
870 		if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
871 			     GFS2_LOG_HEAD_FLUSH_FREEZE))
872 			gfs2_log_shutdown(sdp);
873 		if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
874 			atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
875 	}
876 
877 	trace_gfs2_log_flush(sdp, 0, flags);
878 	up_write(&sdp->sd_log_flush_lock);
879 
880 	kfree(tr);
881 }
882 
883 /**
884  * gfs2_merge_trans - Merge a new transaction into a cached transaction
885  * @old: Original transaction to be expanded
886  * @new: New transaction to be merged
887  */
888 
889 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
890 {
891 	WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
892 
893 	old->tr_num_buf_new	+= new->tr_num_buf_new;
894 	old->tr_num_databuf_new	+= new->tr_num_databuf_new;
895 	old->tr_num_buf_rm	+= new->tr_num_buf_rm;
896 	old->tr_num_databuf_rm	+= new->tr_num_databuf_rm;
897 	old->tr_num_revoke	+= new->tr_num_revoke;
898 	old->tr_num_revoke_rm	+= new->tr_num_revoke_rm;
899 
900 	list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
901 	list_splice_tail_init(&new->tr_buf, &old->tr_buf);
902 }
903 
904 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
905 {
906 	unsigned int reserved;
907 	unsigned int unused;
908 	unsigned int maxres;
909 
910 	gfs2_log_lock(sdp);
911 
912 	if (sdp->sd_log_tr) {
913 		gfs2_merge_trans(sdp->sd_log_tr, tr);
914 	} else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
915 		gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
916 		sdp->sd_log_tr = tr;
917 		set_bit(TR_ATTACHED, &tr->tr_flags);
918 	}
919 
920 	sdp->sd_log_committed_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
921 	reserved = calc_reserved(sdp);
922 	maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
923 	gfs2_assert_withdraw(sdp, maxres >= reserved);
924 	unused = maxres - reserved;
925 	atomic_add(unused, &sdp->sd_log_blks_free);
926 	trace_gfs2_log_blocks(sdp, unused);
927 	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
928 			     sdp->sd_jdesc->jd_blocks);
929 	sdp->sd_log_blks_reserved = reserved;
930 
931 	gfs2_log_unlock(sdp);
932 }
933 
934 /**
935  * gfs2_log_commit - Commit a transaction to the log
936  * @sdp: the filesystem
937  * @tr: the transaction
938  *
939  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
940  * or the total number of used blocks (pinned blocks plus AIL blocks)
941  * is greater than thresh2.
942  *
943  * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
944  * journal size.
945  *
946  * Returns: errno
947  */
948 
949 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
950 {
951 	log_refund(sdp, tr);
952 
953 	if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
954 	    ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
955 	    atomic_read(&sdp->sd_log_thresh2)))
956 		wake_up(&sdp->sd_logd_waitq);
957 }
958 
959 /**
960  * gfs2_log_shutdown - write a shutdown header into a journal
961  * @sdp: the filesystem
962  *
963  */
964 
965 static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
966 {
967 	gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
968 	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
969 	gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
970 
971 	sdp->sd_log_flush_head = sdp->sd_log_head;
972 
973 	log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
974 
975 	gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
976 	gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
977 
978 	sdp->sd_log_head = sdp->sd_log_flush_head;
979 	sdp->sd_log_tail = sdp->sd_log_head;
980 }
981 
982 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
983 {
984 	return (atomic_read(&sdp->sd_log_pinned) +
985 		atomic_read(&sdp->sd_log_blks_needed) >=
986 		atomic_read(&sdp->sd_log_thresh1));
987 }
988 
989 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
990 {
991 	unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
992 
993 	if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
994 		return 1;
995 
996 	return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
997 		atomic_read(&sdp->sd_log_thresh2);
998 }
999 
1000 /**
1001  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1002  * @sdp: Pointer to GFS2 superblock
1003  *
1004  * Also, periodically check to make sure that we're using the most recent
1005  * journal index.
1006  */
1007 
1008 int gfs2_logd(void *data)
1009 {
1010 	struct gfs2_sbd *sdp = data;
1011 	unsigned long t = 1;
1012 	DEFINE_WAIT(wait);
1013 	bool did_flush;
1014 
1015 	while (!kthread_should_stop()) {
1016 
1017 		/* Check for errors writing to the journal */
1018 		if (sdp->sd_log_error) {
1019 			gfs2_lm_withdraw(sdp,
1020 					 "GFS2: fsid=%s: error %d: "
1021 					 "withdrawing the file system to "
1022 					 "prevent further damage.\n",
1023 					 sdp->sd_fsname, sdp->sd_log_error);
1024 		}
1025 
1026 		did_flush = false;
1027 		if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1028 			gfs2_ail1_empty(sdp);
1029 			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1030 				       GFS2_LFC_LOGD_JFLUSH_REQD);
1031 			did_flush = true;
1032 		}
1033 
1034 		if (gfs2_ail_flush_reqd(sdp)) {
1035 			gfs2_ail1_start(sdp);
1036 			gfs2_ail1_wait(sdp);
1037 			gfs2_ail1_empty(sdp);
1038 			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1039 				       GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1040 			did_flush = true;
1041 		}
1042 
1043 		if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1044 			wake_up(&sdp->sd_log_waitq);
1045 
1046 		t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1047 
1048 		try_to_freeze();
1049 
1050 		do {
1051 			prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1052 					TASK_INTERRUPTIBLE);
1053 			if (!gfs2_ail_flush_reqd(sdp) &&
1054 			    !gfs2_jrnl_flush_reqd(sdp) &&
1055 			    !kthread_should_stop())
1056 				t = schedule_timeout(t);
1057 		} while(t && !gfs2_ail_flush_reqd(sdp) &&
1058 			!gfs2_jrnl_flush_reqd(sdp) &&
1059 			!kthread_should_stop());
1060 		finish_wait(&sdp->sd_logd_waitq, &wait);
1061 	}
1062 
1063 	return 0;
1064 }
1065 
1066