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 #include "trans.h"
34 #include "aops.h"
35 
36 static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
37 
38 /**
39  * gfs2_struct2blk - compute stuff
40  * @sdp: the filesystem
41  * @nstruct: the number of structures
42  *
43  * Compute the number of log descriptor blocks needed to hold a certain number
44  * of structures of a certain size.
45  *
46  * Returns: the number of blocks needed (minimum is always 1)
47  */
48 
49 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
50 {
51 	unsigned int blks;
52 	unsigned int first, second;
53 
54 	/* The initial struct gfs2_log_descriptor block */
55 	blks = 1;
56 	first = sdp->sd_ldptrs;
57 
58 	if (nstruct > first) {
59 		/* Subsequent struct gfs2_meta_header blocks */
60 		second = sdp->sd_inptrs;
61 		blks += DIV_ROUND_UP(nstruct - first, second);
62 	}
63 
64 	return blks;
65 }
66 
67 /**
68  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
69  * @bd: The gfs2_bufdata to remove
70  *
71  * The ail lock _must_ be held when calling this function
72  *
73  */
74 
75 void 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 transaction
86  * @sdp: The superblock
87  * @wbc: The writeback control structure
88  * @tr: The transaction to start I/O on
89  * @plug: The block plug currently active
90  */
91 
92 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
93 			       struct writeback_control *wbc,
94 			       struct gfs2_trans *tr, struct blk_plug *plug)
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 	int ret = 0;
103 
104 	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
105 		bh = bd->bd_bh;
106 
107 		gfs2_assert(sdp, bd->bd_tr == tr);
108 
109 		if (!buffer_busy(bh)) {
110 			if (buffer_uptodate(bh)) {
111 				list_move(&bd->bd_ail_st_list,
112 					  &tr->tr_ail2_list);
113 				continue;
114 			}
115 			if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
116 				gfs2_io_error_bh(sdp, bh);
117 				gfs2_withdraw_delayed(sdp);
118 			}
119 		}
120 
121 		if (gfs2_withdrawing_or_withdrawn(sdp)) {
122 			gfs2_remove_from_ail(bd);
123 			continue;
124 		}
125 		if (!buffer_dirty(bh))
126 			continue;
127 		if (gl == bd->bd_gl)
128 			continue;
129 		gl = bd->bd_gl;
130 		list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
131 		mapping = bh->b_folio->mapping;
132 		if (!mapping)
133 			continue;
134 		spin_unlock(&sdp->sd_ail_lock);
135 		BUG_ON(GFS2_SB(mapping->host) != sdp);
136 		if (gfs2_is_jdata(GFS2_I(mapping->host)))
137 			ret = gfs2_jdata_writeback(mapping, wbc);
138 		else
139 			ret = mapping->a_ops->writepages(mapping, wbc);
140 		if (need_resched()) {
141 			blk_finish_plug(plug);
142 			cond_resched();
143 			blk_start_plug(plug);
144 		}
145 		spin_lock(&sdp->sd_ail_lock);
146 		if (ret == -ENODATA) /* if a jdata write into a new hole */
147 			ret = 0; /* ignore it */
148 		mapping_set_error(mapping, ret);
149 		if (ret || wbc->nr_to_write <= 0)
150 			break;
151 		return -EBUSY;
152 	}
153 
154 	return ret;
155 }
156 
157 static void dump_ail_list(struct gfs2_sbd *sdp)
158 {
159 	struct gfs2_trans *tr;
160 	struct gfs2_bufdata *bd;
161 	struct buffer_head *bh;
162 
163 	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
164 		list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
165 					    bd_ail_st_list) {
166 			bh = bd->bd_bh;
167 			fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
168 			       (unsigned long long)bd->bd_blkno, bh);
169 			if (!bh) {
170 				fs_err(sdp, "\n");
171 				continue;
172 			}
173 			fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
174 			       "map:%d new:%d ar:%d aw:%d delay:%d "
175 			       "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
176 			       (unsigned long long)bh->b_blocknr,
177 			       buffer_uptodate(bh), buffer_dirty(bh),
178 			       buffer_locked(bh), buffer_req(bh),
179 			       buffer_mapped(bh), buffer_new(bh),
180 			       buffer_async_read(bh), buffer_async_write(bh),
181 			       buffer_delay(bh), buffer_write_io_error(bh),
182 			       buffer_unwritten(bh),
183 			       buffer_defer_completion(bh),
184 			       buffer_pinned(bh), buffer_escaped(bh));
185 		}
186 	}
187 }
188 
189 /**
190  * gfs2_ail1_flush - start writeback of some ail1 entries
191  * @sdp: The super block
192  * @wbc: The writeback control structure
193  *
194  * Writes back some ail1 entries, according to the limits in the
195  * writeback control structure
196  */
197 
198 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
199 {
200 	struct list_head *head = &sdp->sd_ail1_list;
201 	struct gfs2_trans *tr;
202 	struct blk_plug plug;
203 	int ret;
204 	unsigned long flush_start = jiffies;
205 
206 	trace_gfs2_ail_flush(sdp, wbc, 1);
207 	blk_start_plug(&plug);
208 	spin_lock(&sdp->sd_ail_lock);
209 restart:
210 	ret = 0;
211 	if (time_after(jiffies, flush_start + (HZ * 600))) {
212 		fs_err(sdp, "Error: In %s for ten minutes! t=%d\n",
213 		       __func__, current->journal_info ? 1 : 0);
214 		dump_ail_list(sdp);
215 		goto out;
216 	}
217 	list_for_each_entry_reverse(tr, head, tr_list) {
218 		if (wbc->nr_to_write <= 0)
219 			break;
220 		ret = gfs2_ail1_start_one(sdp, wbc, tr, &plug);
221 		if (ret) {
222 			if (ret == -EBUSY)
223 				goto restart;
224 			break;
225 		}
226 	}
227 out:
228 	spin_unlock(&sdp->sd_ail_lock);
229 	blk_finish_plug(&plug);
230 	if (ret) {
231 		gfs2_lm(sdp, "gfs2_ail1_start_one returned: %d\n", ret);
232 		gfs2_withdraw(sdp);
233 	}
234 	trace_gfs2_ail_flush(sdp, wbc, 0);
235 }
236 
237 /**
238  * gfs2_ail1_start - start writeback of all ail1 entries
239  * @sdp: The superblock
240  */
241 
242 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
243 {
244 	struct writeback_control wbc = {
245 		.sync_mode = WB_SYNC_NONE,
246 		.nr_to_write = LONG_MAX,
247 		.range_start = 0,
248 		.range_end = LLONG_MAX,
249 	};
250 
251 	return gfs2_ail1_flush(sdp, &wbc);
252 }
253 
254 static void gfs2_log_update_flush_tail(struct gfs2_sbd *sdp)
255 {
256 	unsigned int new_flush_tail = sdp->sd_log_head;
257 	struct gfs2_trans *tr;
258 
259 	if (!list_empty(&sdp->sd_ail1_list)) {
260 		tr = list_last_entry(&sdp->sd_ail1_list,
261 				     struct gfs2_trans, tr_list);
262 		new_flush_tail = tr->tr_first;
263 	}
264 	sdp->sd_log_flush_tail = new_flush_tail;
265 }
266 
267 static void gfs2_log_update_head(struct gfs2_sbd *sdp)
268 {
269 	unsigned int new_head = sdp->sd_log_flush_head;
270 
271 	if (sdp->sd_log_flush_tail == sdp->sd_log_head)
272 		sdp->sd_log_flush_tail = new_head;
273 	sdp->sd_log_head = new_head;
274 }
275 
276 /*
277  * gfs2_ail_empty_tr - empty one of the ail lists of a transaction
278  */
279 
280 static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
281 			      struct list_head *head)
282 {
283 	struct gfs2_bufdata *bd;
284 
285 	while (!list_empty(head)) {
286 		bd = list_first_entry(head, struct gfs2_bufdata,
287 				      bd_ail_st_list);
288 		gfs2_assert(sdp, bd->bd_tr == tr);
289 		gfs2_remove_from_ail(bd);
290 	}
291 }
292 
293 /**
294  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
295  * @sdp: the filesystem
296  * @tr: the transaction
297  * @max_revokes: If nonzero, issue revokes for the bd items for written buffers
298  *
299  * returns: the transaction's count of remaining active items
300  */
301 
302 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
303 				int *max_revokes)
304 {
305 	struct gfs2_bufdata *bd, *s;
306 	struct buffer_head *bh;
307 	int active_count = 0;
308 
309 	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
310 					 bd_ail_st_list) {
311 		bh = bd->bd_bh;
312 		gfs2_assert(sdp, bd->bd_tr == tr);
313 		/*
314 		 * If another process flagged an io error, e.g. writing to the
315 		 * journal, error all other bhs and move them off the ail1 to
316 		 * prevent a tight loop when unmount tries to flush ail1,
317 		 * regardless of whether they're still busy. If no outside
318 		 * errors were found and the buffer is busy, move to the next.
319 		 * If the ail buffer is not busy and caught an error, flag it
320 		 * for others.
321 		 */
322 		if (!sdp->sd_log_error && buffer_busy(bh)) {
323 			active_count++;
324 			continue;
325 		}
326 		if (!buffer_uptodate(bh) &&
327 		    !cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
328 			gfs2_io_error_bh(sdp, bh);
329 			gfs2_withdraw_delayed(sdp);
330 		}
331 		/*
332 		 * If we have space for revokes and the bd is no longer on any
333 		 * buf list, we can just add a revoke for it immediately and
334 		 * avoid having to put it on the ail2 list, where it would need
335 		 * to be revoked later.
336 		 */
337 		if (*max_revokes && list_empty(&bd->bd_list)) {
338 			gfs2_add_revoke(sdp, bd);
339 			(*max_revokes)--;
340 			continue;
341 		}
342 		list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
343 	}
344 	return active_count;
345 }
346 
347 /**
348  * gfs2_ail1_empty - Try to empty the ail1 lists
349  * @sdp: The superblock
350  * @max_revokes: If non-zero, add revokes where appropriate
351  *
352  * Tries to empty the ail1 lists, starting with the oldest first.
353  * Returns %true if the ail1 list is now empty.
354  */
355 
356 static bool gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
357 {
358 	struct gfs2_trans *tr, *s;
359 	int oldest_tr = 1;
360 	bool empty;
361 
362 	spin_lock(&sdp->sd_ail_lock);
363 	list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
364 		if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr)
365 			list_move(&tr->tr_list, &sdp->sd_ail2_list);
366 		else
367 			oldest_tr = 0;
368 	}
369 	gfs2_log_update_flush_tail(sdp);
370 	empty = list_empty(&sdp->sd_ail1_list);
371 	spin_unlock(&sdp->sd_ail_lock);
372 
373 	return empty;
374 }
375 
376 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
377 {
378 	struct gfs2_trans *tr;
379 	struct gfs2_bufdata *bd;
380 	struct buffer_head *bh;
381 
382 	spin_lock(&sdp->sd_ail_lock);
383 	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
384 		list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
385 			bh = bd->bd_bh;
386 			if (!buffer_locked(bh))
387 				continue;
388 			get_bh(bh);
389 			spin_unlock(&sdp->sd_ail_lock);
390 			wait_on_buffer(bh);
391 			brelse(bh);
392 			return;
393 		}
394 	}
395 	spin_unlock(&sdp->sd_ail_lock);
396 }
397 
398 static void __ail2_empty(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
399 {
400 	gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
401 	list_del(&tr->tr_list);
402 	gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
403 	gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
404 	gfs2_trans_free(sdp, tr);
405 }
406 
407 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
408 {
409 	struct list_head *ail2_list = &sdp->sd_ail2_list;
410 	unsigned int old_tail = sdp->sd_log_tail;
411 	struct gfs2_trans *tr, *safe;
412 
413 	spin_lock(&sdp->sd_ail_lock);
414 	if (old_tail <= new_tail) {
415 		list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
416 			if (old_tail <= tr->tr_first && tr->tr_first < new_tail)
417 				__ail2_empty(sdp, tr);
418 		}
419 	} else {
420 		list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
421 			if (old_tail <= tr->tr_first || tr->tr_first < new_tail)
422 				__ail2_empty(sdp, tr);
423 		}
424 	}
425 	spin_unlock(&sdp->sd_ail_lock);
426 }
427 
428 /**
429  * gfs2_log_is_empty - Check if the log is empty
430  * @sdp: The GFS2 superblock
431  */
432 
433 bool gfs2_log_is_empty(struct gfs2_sbd *sdp) {
434 	return atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks;
435 }
436 
437 static bool __gfs2_log_try_reserve_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
438 {
439 	unsigned int available;
440 
441 	available = atomic_read(&sdp->sd_log_revokes_available);
442 	while (available >= revokes) {
443 		if (atomic_try_cmpxchg(&sdp->sd_log_revokes_available,
444 				       &available, available - revokes))
445 			return true;
446 	}
447 	return false;
448 }
449 
450 /**
451  * gfs2_log_release_revokes - Release a given number of revokes
452  * @sdp: The GFS2 superblock
453  * @revokes: The number of revokes to release
454  *
455  * sdp->sd_log_flush_lock must be held.
456  */
457 void gfs2_log_release_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
458 {
459 	if (revokes)
460 		atomic_add(revokes, &sdp->sd_log_revokes_available);
461 }
462 
463 /**
464  * gfs2_log_release - Release a given number of log blocks
465  * @sdp: The GFS2 superblock
466  * @blks: The number of blocks
467  *
468  */
469 
470 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
471 {
472 	atomic_add(blks, &sdp->sd_log_blks_free);
473 	trace_gfs2_log_blocks(sdp, blks);
474 	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
475 				  sdp->sd_jdesc->jd_blocks);
476 	if (atomic_read(&sdp->sd_log_blks_needed))
477 		wake_up(&sdp->sd_log_waitq);
478 }
479 
480 /**
481  * __gfs2_log_try_reserve - Try to make a log reservation
482  * @sdp: The GFS2 superblock
483  * @blks: The number of blocks to reserve
484  * @taboo_blks: The number of blocks to leave free
485  *
486  * Try to do the same as __gfs2_log_reserve(), but fail if no more log
487  * space is immediately available.
488  */
489 static bool __gfs2_log_try_reserve(struct gfs2_sbd *sdp, unsigned int blks,
490 				   unsigned int taboo_blks)
491 {
492 	unsigned wanted = blks + taboo_blks;
493 	unsigned int free_blocks;
494 
495 	free_blocks = atomic_read(&sdp->sd_log_blks_free);
496 	while (free_blocks >= wanted) {
497 		if (atomic_try_cmpxchg(&sdp->sd_log_blks_free, &free_blocks,
498 				       free_blocks - blks)) {
499 			trace_gfs2_log_blocks(sdp, -blks);
500 			return true;
501 		}
502 	}
503 	return false;
504 }
505 
506 /**
507  * __gfs2_log_reserve - Make a log reservation
508  * @sdp: The GFS2 superblock
509  * @blks: The number of blocks to reserve
510  * @taboo_blks: The number of blocks to leave free
511  *
512  * @taboo_blks is set to 0 for logd, and to GFS2_LOG_FLUSH_MIN_BLOCKS
513  * for all other processes.  This ensures that when the log is almost full,
514  * logd will still be able to call gfs2_log_flush one more time  without
515  * blocking, which will advance the tail and make some more log space
516  * available.
517  *
518  * We no longer flush the log here, instead we wake up logd to do that
519  * for us. To avoid the thundering herd and to ensure that we deal fairly
520  * with queued waiters, we use an exclusive wait. This means that when we
521  * get woken with enough journal space to get our reservation, we need to
522  * wake the next waiter on the list.
523  */
524 
525 static void __gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks,
526 			       unsigned int taboo_blks)
527 {
528 	unsigned wanted = blks + taboo_blks;
529 	unsigned int free_blocks;
530 
531 	atomic_add(blks, &sdp->sd_log_blks_needed);
532 	for (;;) {
533 		if (current != sdp->sd_logd_process)
534 			wake_up(&sdp->sd_logd_waitq);
535 		io_wait_event(sdp->sd_log_waitq,
536 			(free_blocks = atomic_read(&sdp->sd_log_blks_free),
537 			 free_blocks >= wanted));
538 		do {
539 			if (atomic_try_cmpxchg(&sdp->sd_log_blks_free,
540 					       &free_blocks,
541 					       free_blocks - blks))
542 				goto reserved;
543 		} while (free_blocks >= wanted);
544 	}
545 
546 reserved:
547 	trace_gfs2_log_blocks(sdp, -blks);
548 	if (atomic_sub_return(blks, &sdp->sd_log_blks_needed))
549 		wake_up(&sdp->sd_log_waitq);
550 }
551 
552 /**
553  * gfs2_log_try_reserve - Try to make a log reservation
554  * @sdp: The GFS2 superblock
555  * @tr: The transaction
556  * @extra_revokes: The number of additional revokes reserved (output)
557  *
558  * This is similar to gfs2_log_reserve, but sdp->sd_log_flush_lock must be
559  * held for correct revoke accounting.
560  */
561 
562 bool gfs2_log_try_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
563 			  unsigned int *extra_revokes)
564 {
565 	unsigned int blks = tr->tr_reserved;
566 	unsigned int revokes = tr->tr_revokes;
567 	unsigned int revoke_blks = 0;
568 
569 	*extra_revokes = 0;
570 	if (revokes && !__gfs2_log_try_reserve_revokes(sdp, revokes)) {
571 		revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
572 		*extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
573 		blks += revoke_blks;
574 	}
575 	if (!blks)
576 		return true;
577 	if (__gfs2_log_try_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS))
578 		return true;
579 	if (!revoke_blks)
580 		gfs2_log_release_revokes(sdp, revokes);
581 	return false;
582 }
583 
584 /**
585  * gfs2_log_reserve - Make a log reservation
586  * @sdp: The GFS2 superblock
587  * @tr: The transaction
588  * @extra_revokes: The number of additional revokes reserved (output)
589  *
590  * sdp->sd_log_flush_lock must not be held.
591  */
592 
593 void gfs2_log_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
594 		      unsigned int *extra_revokes)
595 {
596 	unsigned int blks = tr->tr_reserved;
597 	unsigned int revokes = tr->tr_revokes;
598 	unsigned int revoke_blks;
599 
600 	*extra_revokes = 0;
601 	if (revokes) {
602 		revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
603 		*extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
604 		blks += revoke_blks;
605 	}
606 	__gfs2_log_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS);
607 }
608 
609 /**
610  * log_distance - Compute distance between two journal blocks
611  * @sdp: The GFS2 superblock
612  * @newer: The most recent journal block of the pair
613  * @older: The older journal block of the pair
614  *
615  *   Compute the distance (in the journal direction) between two
616  *   blocks in the journal
617  *
618  * Returns: the distance in blocks
619  */
620 
621 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
622 					unsigned int older)
623 {
624 	int dist;
625 
626 	dist = newer - older;
627 	if (dist < 0)
628 		dist += sdp->sd_jdesc->jd_blocks;
629 
630 	return dist;
631 }
632 
633 /**
634  * calc_reserved - Calculate the number of blocks to keep reserved
635  * @sdp: The GFS2 superblock
636  *
637  * This is complex.  We need to reserve room for all our currently used
638  * metadata blocks (e.g. normal file I/O rewriting file time stamps) and
639  * all our journaled data blocks for journaled files (e.g. files in the
640  * meta_fs like rindex, or files for which chattr +j was done.)
641  * If we don't reserve enough space, corruption will follow.
642  *
643  * We can have metadata blocks and jdata blocks in the same journal.  Each
644  * type gets its own log descriptor, for which we need to reserve a block.
645  * In fact, each type has the potential for needing more than one log descriptor
646  * in cases where we have more blocks than will fit in a log descriptor.
647  * Metadata journal entries take up half the space of journaled buffer entries.
648  *
649  * Also, we need to reserve blocks for revoke journal entries and one for an
650  * overall header for the lot.
651  *
652  * Returns: the number of blocks reserved
653  */
654 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
655 {
656 	unsigned int reserved = GFS2_LOG_FLUSH_MIN_BLOCKS;
657 	unsigned int blocks;
658 	struct gfs2_trans *tr = sdp->sd_log_tr;
659 
660 	if (tr) {
661 		blocks = tr->tr_num_buf_new - tr->tr_num_buf_rm;
662 		reserved += blocks + DIV_ROUND_UP(blocks, buf_limit(sdp));
663 		blocks = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
664 		reserved += blocks + DIV_ROUND_UP(blocks, databuf_limit(sdp));
665 	}
666 	return reserved;
667 }
668 
669 static void log_pull_tail(struct gfs2_sbd *sdp)
670 {
671 	unsigned int new_tail = sdp->sd_log_flush_tail;
672 	unsigned int dist;
673 
674 	if (new_tail == sdp->sd_log_tail)
675 		return;
676 	dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
677 	ail2_empty(sdp, new_tail);
678 	gfs2_log_release(sdp, dist);
679 	sdp->sd_log_tail = new_tail;
680 }
681 
682 
683 void log_flush_wait(struct gfs2_sbd *sdp)
684 {
685 	DEFINE_WAIT(wait);
686 
687 	if (atomic_read(&sdp->sd_log_in_flight)) {
688 		do {
689 			prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
690 					TASK_UNINTERRUPTIBLE);
691 			if (atomic_read(&sdp->sd_log_in_flight))
692 				io_schedule();
693 		} while(atomic_read(&sdp->sd_log_in_flight));
694 		finish_wait(&sdp->sd_log_flush_wait, &wait);
695 	}
696 }
697 
698 static int ip_cmp(void *priv, const struct list_head *a, const struct list_head *b)
699 {
700 	struct gfs2_inode *ipa, *ipb;
701 
702 	ipa = list_entry(a, struct gfs2_inode, i_ordered);
703 	ipb = list_entry(b, struct gfs2_inode, i_ordered);
704 
705 	if (ipa->i_no_addr < ipb->i_no_addr)
706 		return -1;
707 	if (ipa->i_no_addr > ipb->i_no_addr)
708 		return 1;
709 	return 0;
710 }
711 
712 static void __ordered_del_inode(struct gfs2_inode *ip)
713 {
714 	if (!list_empty(&ip->i_ordered))
715 		list_del_init(&ip->i_ordered);
716 }
717 
718 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
719 {
720 	struct gfs2_inode *ip;
721 	LIST_HEAD(written);
722 
723 	spin_lock(&sdp->sd_ordered_lock);
724 	list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
725 	while (!list_empty(&sdp->sd_log_ordered)) {
726 		ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
727 		if (ip->i_inode.i_mapping->nrpages == 0) {
728 			__ordered_del_inode(ip);
729 			continue;
730 		}
731 		list_move(&ip->i_ordered, &written);
732 		spin_unlock(&sdp->sd_ordered_lock);
733 		filemap_fdatawrite(ip->i_inode.i_mapping);
734 		spin_lock(&sdp->sd_ordered_lock);
735 	}
736 	list_splice(&written, &sdp->sd_log_ordered);
737 	spin_unlock(&sdp->sd_ordered_lock);
738 }
739 
740 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
741 {
742 	struct gfs2_inode *ip;
743 
744 	spin_lock(&sdp->sd_ordered_lock);
745 	while (!list_empty(&sdp->sd_log_ordered)) {
746 		ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
747 		__ordered_del_inode(ip);
748 		if (ip->i_inode.i_mapping->nrpages == 0)
749 			continue;
750 		spin_unlock(&sdp->sd_ordered_lock);
751 		filemap_fdatawait(ip->i_inode.i_mapping);
752 		spin_lock(&sdp->sd_ordered_lock);
753 	}
754 	spin_unlock(&sdp->sd_ordered_lock);
755 }
756 
757 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
758 {
759 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
760 
761 	spin_lock(&sdp->sd_ordered_lock);
762 	__ordered_del_inode(ip);
763 	spin_unlock(&sdp->sd_ordered_lock);
764 }
765 
766 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
767 {
768 	struct buffer_head *bh = bd->bd_bh;
769 	struct gfs2_glock *gl = bd->bd_gl;
770 
771 	sdp->sd_log_num_revoke++;
772 	if (atomic_inc_return(&gl->gl_revokes) == 1)
773 		gfs2_glock_hold(gl);
774 	bh->b_private = NULL;
775 	bd->bd_blkno = bh->b_blocknr;
776 	gfs2_remove_from_ail(bd); /* drops ref on bh */
777 	bd->bd_bh = NULL;
778 	set_bit(GLF_LFLUSH, &gl->gl_flags);
779 	list_add(&bd->bd_list, &sdp->sd_log_revokes);
780 }
781 
782 void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
783 {
784 	if (atomic_dec_return(&gl->gl_revokes) == 0) {
785 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
786 		gfs2_glock_put_async(gl);
787 	}
788 }
789 
790 /**
791  * gfs2_flush_revokes - Add as many revokes to the system transaction as we can
792  * @sdp: The GFS2 superblock
793  *
794  * Our usual strategy is to defer writing revokes as much as we can in the hope
795  * that we'll eventually overwrite the journal, which will make those revokes
796  * go away.  This changes when we flush the log: at that point, there will
797  * likely be some left-over space in the last revoke block of that transaction.
798  * We can fill that space with additional revokes for blocks that have already
799  * been written back.  This will basically come at no cost now, and will save
800  * us from having to keep track of those blocks on the AIL2 list later.
801  */
802 void gfs2_flush_revokes(struct gfs2_sbd *sdp)
803 {
804 	/* number of revokes we still have room for */
805 	unsigned int max_revokes = atomic_read(&sdp->sd_log_revokes_available);
806 
807 	gfs2_log_lock(sdp);
808 	gfs2_ail1_empty(sdp, max_revokes);
809 	gfs2_log_unlock(sdp);
810 
811 	if (gfs2_withdrawing(sdp))
812 		gfs2_withdraw(sdp);
813 }
814 
815 /**
816  * gfs2_write_log_header - Write a journal log header buffer at lblock
817  * @sdp: The GFS2 superblock
818  * @jd: journal descriptor of the journal to which we are writing
819  * @seq: sequence number
820  * @tail: tail of the log
821  * @lblock: value for lh_blkno (block number relative to start of journal)
822  * @flags: log header flags GFS2_LOG_HEAD_*
823  * @op_flags: flags to pass to the bio
824  *
825  * Returns: the initialized log buffer descriptor
826  */
827 
828 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
829 			   u64 seq, u32 tail, u32 lblock, u32 flags,
830 			   blk_opf_t op_flags)
831 {
832 	struct gfs2_log_header *lh;
833 	u32 hash, crc;
834 	struct page *page;
835 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
836 	struct timespec64 tv;
837 	struct super_block *sb = sdp->sd_vfs;
838 	u64 dblock;
839 
840 	if (gfs2_withdrawing_or_withdrawn(sdp))
841 		return;
842 
843 	page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
844 	lh = page_address(page);
845 	clear_page(lh);
846 
847 	lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
848 	lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
849 	lh->lh_header.__pad0 = cpu_to_be64(0);
850 	lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
851 	lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
852 	lh->lh_sequence = cpu_to_be64(seq);
853 	lh->lh_flags = cpu_to_be32(flags);
854 	lh->lh_tail = cpu_to_be32(tail);
855 	lh->lh_blkno = cpu_to_be32(lblock);
856 	hash = ~crc32(~0, lh, LH_V1_SIZE);
857 	lh->lh_hash = cpu_to_be32(hash);
858 
859 	ktime_get_coarse_real_ts64(&tv);
860 	lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
861 	lh->lh_sec = cpu_to_be64(tv.tv_sec);
862 	if (!list_empty(&jd->extent_list))
863 		dblock = gfs2_log_bmap(jd, lblock);
864 	else {
865 		unsigned int extlen;
866 		int ret;
867 
868 		extlen = 1;
869 		ret = gfs2_get_extent(jd->jd_inode, lblock, &dblock, &extlen);
870 		if (gfs2_assert_withdraw(sdp, ret == 0))
871 			return;
872 	}
873 	lh->lh_addr = cpu_to_be64(dblock);
874 	lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
875 
876 	/* We may only write local statfs, quota, etc., when writing to our
877 	   own journal. The values are left 0 when recovering a journal
878 	   different from our own. */
879 	if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
880 		lh->lh_statfs_addr =
881 			cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
882 		lh->lh_quota_addr =
883 			cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
884 
885 		spin_lock(&sdp->sd_statfs_spin);
886 		lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
887 		lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
888 		lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
889 		spin_unlock(&sdp->sd_statfs_spin);
890 	}
891 
892 	BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
893 
894 	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
895 		     sb->s_blocksize - LH_V1_SIZE - 4);
896 	lh->lh_crc = cpu_to_be32(crc);
897 
898 	gfs2_log_write(sdp, jd, page, sb->s_blocksize, 0, dblock);
899 	gfs2_log_submit_bio(&jd->jd_log_bio, REQ_OP_WRITE | op_flags);
900 }
901 
902 /**
903  * log_write_header - Get and initialize a journal header buffer
904  * @sdp: The GFS2 superblock
905  * @flags: The log header flags, including log header origin
906  *
907  * Returns: the initialized log buffer descriptor
908  */
909 
910 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
911 {
912 	blk_opf_t op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
913 	struct super_block *sb = sdp->sd_vfs;
914 
915 	gfs2_assert_withdraw(sdp, sb->s_writers.frozen != SB_FREEZE_COMPLETE);
916 
917 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
918 		gfs2_ordered_wait(sdp);
919 		log_flush_wait(sdp);
920 		op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
921 	}
922 	sdp->sd_log_idle = (sdp->sd_log_flush_tail == sdp->sd_log_flush_head);
923 	gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++,
924 			      sdp->sd_log_flush_tail, sdp->sd_log_flush_head,
925 			      flags, op_flags);
926 	gfs2_log_incr_head(sdp);
927 	log_flush_wait(sdp);
928 	log_pull_tail(sdp);
929 	gfs2_log_update_head(sdp);
930 }
931 
932 /**
933  * gfs2_ail_drain - drain the ail lists after a withdraw
934  * @sdp: Pointer to GFS2 superblock
935  */
936 void gfs2_ail_drain(struct gfs2_sbd *sdp)
937 {
938 	struct gfs2_trans *tr;
939 
940 	spin_lock(&sdp->sd_ail_lock);
941 	/*
942 	 * For transactions on the sd_ail1_list we need to drain both the
943 	 * ail1 and ail2 lists. That's because function gfs2_ail1_start_one
944 	 * (temporarily) moves items from its tr_ail1 list to tr_ail2 list
945 	 * before revokes are sent for that block. Items on the sd_ail2_list
946 	 * should have already gotten beyond that point, so no need.
947 	 */
948 	while (!list_empty(&sdp->sd_ail1_list)) {
949 		tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
950 				      tr_list);
951 		gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
952 		gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
953 		list_del(&tr->tr_list);
954 		gfs2_trans_free(sdp, tr);
955 	}
956 	while (!list_empty(&sdp->sd_ail2_list)) {
957 		tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
958 				      tr_list);
959 		gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
960 		list_del(&tr->tr_list);
961 		gfs2_trans_free(sdp, tr);
962 	}
963 	gfs2_drain_revokes(sdp);
964 	spin_unlock(&sdp->sd_ail_lock);
965 }
966 
967 /**
968  * empty_ail1_list - try to start IO and empty the ail1 list
969  * @sdp: Pointer to GFS2 superblock
970  */
971 static void empty_ail1_list(struct gfs2_sbd *sdp)
972 {
973 	unsigned long start = jiffies;
974 	bool empty = false;
975 
976 	while (!empty) {
977 		if (time_after(jiffies, start + (HZ * 600))) {
978 			fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n",
979 			       __func__, current->journal_info ? 1 : 0);
980 			dump_ail_list(sdp);
981 			return;
982 		}
983 		gfs2_ail1_start(sdp);
984 		gfs2_ail1_wait(sdp);
985 		empty = gfs2_ail1_empty(sdp, 0);
986 
987 		if (gfs2_withdrawing_or_withdrawn(sdp))
988 			break;
989 	}
990 
991 	if (gfs2_withdrawing(sdp))
992 		gfs2_withdraw(sdp);
993 }
994 
995 /**
996  * trans_drain - drain the buf and databuf queue for a failed transaction
997  * @tr: the transaction to drain
998  *
999  * When this is called, we're taking an error exit for a log write that failed
1000  * but since we bypassed the after_commit functions, we need to remove the
1001  * items from the buf and databuf queue.
1002  */
1003 static void trans_drain(struct gfs2_trans *tr)
1004 {
1005 	struct gfs2_bufdata *bd;
1006 	struct list_head *head;
1007 
1008 	if (!tr)
1009 		return;
1010 
1011 	head = &tr->tr_buf;
1012 	while (!list_empty(head)) {
1013 		bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1014 		list_del_init(&bd->bd_list);
1015 		if (!list_empty(&bd->bd_ail_st_list))
1016 			gfs2_remove_from_ail(bd);
1017 		kmem_cache_free(gfs2_bufdata_cachep, bd);
1018 	}
1019 	head = &tr->tr_databuf;
1020 	while (!list_empty(head)) {
1021 		bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1022 		list_del_init(&bd->bd_list);
1023 		if (!list_empty(&bd->bd_ail_st_list))
1024 			gfs2_remove_from_ail(bd);
1025 		kmem_cache_free(gfs2_bufdata_cachep, bd);
1026 	}
1027 }
1028 
1029 /**
1030  * gfs2_log_flush - flush incore transaction(s)
1031  * @sdp: The filesystem
1032  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
1033  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
1034  *
1035  */
1036 
1037 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
1038 {
1039 	struct gfs2_trans *tr = NULL;
1040 	unsigned int reserved_blocks = 0, used_blocks = 0;
1041 	bool frozen = test_bit(SDF_FROZEN, &sdp->sd_flags);
1042 	unsigned int first_log_head;
1043 	unsigned int reserved_revokes = 0;
1044 
1045 	down_write(&sdp->sd_log_flush_lock);
1046 	trace_gfs2_log_flush(sdp, 1, flags);
1047 
1048 repeat:
1049 	/*
1050 	 * Do this check while holding the log_flush_lock to prevent new
1051 	 * buffers from being added to the ail via gfs2_pin()
1052 	 */
1053 	if (gfs2_withdrawing_or_withdrawn(sdp) ||
1054 	    !test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
1055 		goto out;
1056 
1057 	/* Log might have been flushed while we waited for the flush lock */
1058 	if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags))
1059 		goto out;
1060 
1061 	first_log_head = sdp->sd_log_head;
1062 	sdp->sd_log_flush_head = first_log_head;
1063 
1064 	tr = sdp->sd_log_tr;
1065 	if (tr || sdp->sd_log_num_revoke) {
1066 		if (reserved_blocks)
1067 			gfs2_log_release(sdp, reserved_blocks);
1068 		reserved_blocks = sdp->sd_log_blks_reserved;
1069 		reserved_revokes = sdp->sd_log_num_revoke;
1070 		if (tr) {
1071 			sdp->sd_log_tr = NULL;
1072 			tr->tr_first = first_log_head;
1073 			if (unlikely(frozen)) {
1074 				if (gfs2_assert_withdraw_delayed(sdp,
1075 				       !tr->tr_num_buf_new && !tr->tr_num_databuf_new))
1076 					goto out_withdraw;
1077 			}
1078 		}
1079 	} else if (!reserved_blocks) {
1080 		unsigned int taboo_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
1081 
1082 		reserved_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
1083 		if (current == sdp->sd_logd_process)
1084 			taboo_blocks = 0;
1085 
1086 		if (!__gfs2_log_try_reserve(sdp, reserved_blocks, taboo_blocks)) {
1087 			up_write(&sdp->sd_log_flush_lock);
1088 			__gfs2_log_reserve(sdp, reserved_blocks, taboo_blocks);
1089 			down_write(&sdp->sd_log_flush_lock);
1090 			goto repeat;
1091 		}
1092 		BUG_ON(sdp->sd_log_num_revoke);
1093 	}
1094 
1095 	if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
1096 		clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
1097 
1098 	if (unlikely(frozen))
1099 		if (gfs2_assert_withdraw_delayed(sdp, !reserved_revokes))
1100 			goto out_withdraw;
1101 
1102 	gfs2_ordered_write(sdp);
1103 	if (gfs2_withdrawing_or_withdrawn(sdp))
1104 		goto out_withdraw;
1105 	lops_before_commit(sdp, tr);
1106 	if (gfs2_withdrawing_or_withdrawn(sdp))
1107 		goto out_withdraw;
1108 	if (sdp->sd_jdesc)
1109 		gfs2_log_submit_bio(&sdp->sd_jdesc->jd_log_bio, REQ_OP_WRITE);
1110 	if (gfs2_withdrawing_or_withdrawn(sdp))
1111 		goto out_withdraw;
1112 
1113 	if (sdp->sd_log_head != sdp->sd_log_flush_head) {
1114 		log_write_header(sdp, flags);
1115 	} else if (sdp->sd_log_tail != sdp->sd_log_flush_tail && !sdp->sd_log_idle) {
1116 		log_write_header(sdp, flags);
1117 	}
1118 	if (gfs2_withdrawing_or_withdrawn(sdp))
1119 		goto out_withdraw;
1120 	lops_after_commit(sdp, tr);
1121 
1122 	gfs2_log_lock(sdp);
1123 	sdp->sd_log_blks_reserved = 0;
1124 
1125 	spin_lock(&sdp->sd_ail_lock);
1126 	if (tr && !list_empty(&tr->tr_ail1_list)) {
1127 		list_add(&tr->tr_list, &sdp->sd_ail1_list);
1128 		tr = NULL;
1129 	}
1130 	spin_unlock(&sdp->sd_ail_lock);
1131 	gfs2_log_unlock(sdp);
1132 
1133 	if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
1134 		if (!sdp->sd_log_idle) {
1135 			empty_ail1_list(sdp);
1136 			if (gfs2_withdrawing_or_withdrawn(sdp))
1137 				goto out_withdraw;
1138 			log_write_header(sdp, flags);
1139 		}
1140 		if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
1141 			     GFS2_LOG_HEAD_FLUSH_FREEZE))
1142 			gfs2_log_shutdown(sdp);
1143 	}
1144 
1145 out_end:
1146 	used_blocks = log_distance(sdp, sdp->sd_log_flush_head, first_log_head);
1147 	reserved_revokes += atomic_read(&sdp->sd_log_revokes_available);
1148 	atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
1149 	gfs2_assert_withdraw(sdp, reserved_revokes % sdp->sd_inptrs == sdp->sd_ldptrs);
1150 	if (reserved_revokes > sdp->sd_ldptrs)
1151 		reserved_blocks += (reserved_revokes - sdp->sd_ldptrs) / sdp->sd_inptrs;
1152 out:
1153 	if (used_blocks != reserved_blocks) {
1154 		gfs2_assert_withdraw_delayed(sdp, used_blocks < reserved_blocks);
1155 		gfs2_log_release(sdp, reserved_blocks - used_blocks);
1156 	}
1157 	up_write(&sdp->sd_log_flush_lock);
1158 	gfs2_trans_free(sdp, tr);
1159 	if (gfs2_withdrawing(sdp))
1160 		gfs2_withdraw(sdp);
1161 	trace_gfs2_log_flush(sdp, 0, flags);
1162 	return;
1163 
1164 out_withdraw:
1165 	trans_drain(tr);
1166 	/**
1167 	 * If the tr_list is empty, we're withdrawing during a log
1168 	 * flush that targets a transaction, but the transaction was
1169 	 * never queued onto any of the ail lists. Here we add it to
1170 	 * ail1 just so that ail_drain() will find and free it.
1171 	 */
1172 	spin_lock(&sdp->sd_ail_lock);
1173 	if (tr && list_empty(&tr->tr_list))
1174 		list_add(&tr->tr_list, &sdp->sd_ail1_list);
1175 	spin_unlock(&sdp->sd_ail_lock);
1176 	tr = NULL;
1177 	goto out_end;
1178 }
1179 
1180 /**
1181  * gfs2_merge_trans - Merge a new transaction into a cached transaction
1182  * @sdp: the filesystem
1183  * @new: New transaction to be merged
1184  */
1185 
1186 static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
1187 {
1188 	struct gfs2_trans *old = sdp->sd_log_tr;
1189 
1190 	WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
1191 
1192 	old->tr_num_buf_new	+= new->tr_num_buf_new;
1193 	old->tr_num_databuf_new	+= new->tr_num_databuf_new;
1194 	old->tr_num_buf_rm	+= new->tr_num_buf_rm;
1195 	old->tr_num_databuf_rm	+= new->tr_num_databuf_rm;
1196 	old->tr_revokes		+= new->tr_revokes;
1197 	old->tr_num_revoke	+= new->tr_num_revoke;
1198 
1199 	list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
1200 	list_splice_tail_init(&new->tr_buf, &old->tr_buf);
1201 
1202 	spin_lock(&sdp->sd_ail_lock);
1203 	list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
1204 	list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
1205 	spin_unlock(&sdp->sd_ail_lock);
1206 }
1207 
1208 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1209 {
1210 	unsigned int reserved;
1211 	unsigned int unused;
1212 	unsigned int maxres;
1213 
1214 	gfs2_log_lock(sdp);
1215 
1216 	if (sdp->sd_log_tr) {
1217 		gfs2_merge_trans(sdp, tr);
1218 	} else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
1219 		gfs2_assert_withdraw(sdp, !test_bit(TR_ONSTACK, &tr->tr_flags));
1220 		sdp->sd_log_tr = tr;
1221 		set_bit(TR_ATTACHED, &tr->tr_flags);
1222 	}
1223 
1224 	reserved = calc_reserved(sdp);
1225 	maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
1226 	gfs2_assert_withdraw(sdp, maxres >= reserved);
1227 	unused = maxres - reserved;
1228 	if (unused)
1229 		gfs2_log_release(sdp, unused);
1230 	sdp->sd_log_blks_reserved = reserved;
1231 
1232 	gfs2_log_unlock(sdp);
1233 }
1234 
1235 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
1236 {
1237 	return atomic_read(&sdp->sd_log_pinned) +
1238 	       atomic_read(&sdp->sd_log_blks_needed) >=
1239 	       atomic_read(&sdp->sd_log_thresh1);
1240 }
1241 
1242 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
1243 {
1244 	return sdp->sd_jdesc->jd_blocks -
1245 	       atomic_read(&sdp->sd_log_blks_free) +
1246 	       atomic_read(&sdp->sd_log_blks_needed) >=
1247 	       atomic_read(&sdp->sd_log_thresh2);
1248 }
1249 
1250 /**
1251  * gfs2_log_commit - Commit a transaction to the log
1252  * @sdp: the filesystem
1253  * @tr: the transaction
1254  *
1255  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
1256  * or the total number of used blocks (pinned blocks plus AIL blocks)
1257  * is greater than thresh2.
1258  *
1259  * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of
1260  * journal size.
1261  *
1262  * Returns: errno
1263  */
1264 
1265 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1266 {
1267 	log_refund(sdp, tr);
1268 
1269 	if (gfs2_ail_flush_reqd(sdp) || gfs2_jrnl_flush_reqd(sdp))
1270 		wake_up(&sdp->sd_logd_waitq);
1271 }
1272 
1273 /**
1274  * gfs2_log_shutdown - write a shutdown header into a journal
1275  * @sdp: the filesystem
1276  *
1277  */
1278 
1279 static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
1280 {
1281 	gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
1282 	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
1283 	gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
1284 
1285 	log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
1286 	log_pull_tail(sdp);
1287 
1288 	gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
1289 	gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
1290 }
1291 
1292 /**
1293  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1294  * @data: Pointer to GFS2 superblock
1295  *
1296  * Also, periodically check to make sure that we're using the most recent
1297  * journal index.
1298  */
1299 
1300 int gfs2_logd(void *data)
1301 {
1302 	struct gfs2_sbd *sdp = data;
1303 	unsigned long t = 1;
1304 
1305 	set_freezable();
1306 	while (!kthread_should_stop()) {
1307 		if (gfs2_withdrawing_or_withdrawn(sdp))
1308 			break;
1309 
1310 		/* Check for errors writing to the journal */
1311 		if (sdp->sd_log_error) {
1312 			gfs2_lm(sdp,
1313 				"GFS2: fsid=%s: error %d: "
1314 				"withdrawing the file system to "
1315 				"prevent further damage.\n",
1316 				sdp->sd_fsname, sdp->sd_log_error);
1317 			gfs2_withdraw(sdp);
1318 			break;
1319 		}
1320 
1321 		if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1322 			gfs2_ail1_empty(sdp, 0);
1323 			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1324 						  GFS2_LFC_LOGD_JFLUSH_REQD);
1325 		}
1326 
1327 		if (test_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags) ||
1328 		    gfs2_ail_flush_reqd(sdp)) {
1329 			clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
1330 			gfs2_ail1_start(sdp);
1331 			gfs2_ail1_wait(sdp);
1332 			gfs2_ail1_empty(sdp, 0);
1333 			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1334 						  GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1335 		}
1336 
1337 		t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1338 
1339 		t = wait_event_freezable_timeout(sdp->sd_logd_waitq,
1340 				test_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags) ||
1341 				gfs2_ail_flush_reqd(sdp) ||
1342 				gfs2_jrnl_flush_reqd(sdp) ||
1343 				sdp->sd_log_error ||
1344 				gfs2_withdrawing_or_withdrawn(sdp) ||
1345 				kthread_should_stop(),
1346 				t);
1347 	}
1348 
1349 	if (gfs2_withdrawing(sdp))
1350 		gfs2_withdraw(sdp);
1351 
1352 	return 0;
1353 }
1354 
1355