xref: /linux/drivers/md/dm-log-userspace-base.c (revision 98f21c54f99519329c18e2625b0ea6db14524d09)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2006-2009 Red Hat, Inc.
4  *
5  * This file is released under the LGPL.
6  */
7 
8 #include <linux/bio.h>
9 #include <linux/slab.h>
10 #include <linux/jiffies.h>
11 #include <linux/dm-dirty-log.h>
12 #include <linux/device-mapper.h>
13 #include <linux/dm-log-userspace.h>
14 #include <linux/module.h>
15 #include <linux/workqueue.h>
16 
17 #include "dm-log-userspace-transfer.h"
18 
19 #define DM_LOG_USERSPACE_VSN "1.3.0"
20 
21 #define FLUSH_ENTRY_POOL_SIZE 16
22 
23 struct dm_dirty_log_flush_entry {
24 	int type;
25 	region_t region;
26 	struct list_head list;
27 };
28 
29 /*
30  * This limit on the number of mark and clear request is, to a degree,
31  * arbitrary.  However, there is some basis for the choice in the limits
32  * imposed on the size of data payload by dm-log-userspace-transfer.c:
33  * dm_consult_userspace().
34  */
35 #define MAX_FLUSH_GROUP_COUNT 32
36 
37 struct log_c {
38 	struct dm_target *ti;
39 	struct dm_dev *log_dev;
40 
41 	char *usr_argv_str;
42 	uint32_t usr_argc;
43 
44 	uint32_t region_size;
45 	region_t region_count;
46 	uint64_t luid;
47 	char uuid[DM_UUID_LEN];
48 
49 	/*
50 	 * Mark and clear requests are held until a flush is issued
51 	 * so that we can group, and thereby limit, the amount of
52 	 * network traffic between kernel and userspace.  The 'flush_lock'
53 	 * is used to protect these lists.
54 	 */
55 	spinlock_t flush_lock;
56 	struct list_head mark_list;
57 	struct list_head clear_list;
58 
59 	/*
60 	 * in_sync_hint gets set when doing is_remote_recovering.  It
61 	 * represents the first region that needs recovery.  IOW, the
62 	 * first zero bit of sync_bits.  This can be useful for to limit
63 	 * traffic for calls like is_remote_recovering and get_resync_work,
64 	 * but be take care in its use for anything else.
65 	 */
66 	uint64_t in_sync_hint;
67 
68 	/*
69 	 * Workqueue for flush of clear region requests.
70 	 */
71 	struct workqueue_struct *dmlog_wq;
72 	struct delayed_work flush_log_work;
73 	atomic_t sched_flush;
74 
75 	/*
76 	 * Combine userspace flush and mark requests for efficiency.
77 	 */
78 	uint32_t integrated_flush;
79 
80 	mempool_t flush_entry_pool;
81 };
82 
83 static struct kmem_cache *_flush_entry_cache;
84 
85 static int userspace_do_request(struct log_c *lc, const char *uuid,
86 				int request_type, char *data, size_t data_size,
87 				char *rdata, size_t *rdata_size)
88 {
89 	int r;
90 
91 	/*
92 	 * If the server isn't there, -ESRCH is returned,
93 	 * and we must keep trying until the server is
94 	 * restored.
95 	 */
96 retry:
97 	r = dm_consult_userspace(uuid, lc->luid, request_type, data,
98 				 data_size, rdata, rdata_size);
99 
100 	if (r != -ESRCH)
101 		return r;
102 
103 	DMERR(" Userspace log server not found.");
104 	while (1) {
105 		set_current_state(TASK_INTERRUPTIBLE);
106 		schedule_timeout(2*HZ);
107 		DMWARN("Attempting to contact userspace log server...");
108 		r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR,
109 					 lc->usr_argv_str,
110 					 strlen(lc->usr_argv_str) + 1,
111 					 NULL, NULL);
112 		if (!r)
113 			break;
114 	}
115 	DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
116 	r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL,
117 				 0, NULL, NULL);
118 	if (!r)
119 		goto retry;
120 
121 	DMERR("Error trying to resume userspace log: %d", r);
122 
123 	return -ESRCH;
124 }
125 
126 static int build_constructor_string(struct dm_target *ti,
127 				    unsigned int argc, char **argv,
128 				    char **ctr_str)
129 {
130 	int i, str_size;
131 	char *str = NULL;
132 
133 	*ctr_str = NULL;
134 
135 	/*
136 	 * Determine overall size of the string.
137 	 */
138 	for (i = 0, str_size = 0; i < argc; i++)
139 		str_size += strlen(argv[i]) + 1; /* +1 for space between args */
140 
141 	str_size += 20; /* Max number of chars in a printed u64 number */
142 	str_size++; /* For NUL-terminator */
143 
144 	str = kzalloc(str_size, GFP_KERNEL);
145 	if (!str) {
146 		DMWARN("Unable to allocate memory for constructor string");
147 		return -ENOMEM;
148 	}
149 
150 	str_size = sprintf(str, "%llu", (unsigned long long)ti->len);
151 	for (i = 0; i < argc; i++)
152 		str_size += sprintf(str + str_size, " %s", argv[i]);
153 
154 	*ctr_str = str;
155 	return str_size;
156 }
157 
158 static void do_flush(struct work_struct *work)
159 {
160 	int r;
161 	struct log_c *lc = container_of(work, struct log_c, flush_log_work.work);
162 
163 	atomic_set(&lc->sched_flush, 0);
164 
165 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, NULL, 0, NULL, NULL);
166 
167 	if (r)
168 		dm_table_event(lc->ti->table);
169 }
170 
171 /*
172  * userspace_ctr
173  *
174  * argv contains:
175  *	<UUID> [integrated_flush] <other args>
176  * Where 'other args' are the userspace implementation-specific log
177  * arguments.
178  *
179  * Example:
180  *	<UUID> [integrated_flush] clustered-disk <arg count> <log dev>
181  *	<region_size> [[no]sync]
182  *
183  * This module strips off the <UUID> and uses it for identification
184  * purposes when communicating with userspace about a log.
185  *
186  * If integrated_flush is defined, the kernel combines flush
187  * and mark requests.
188  *
189  * The rest of the line, beginning with 'clustered-disk', is passed
190  * to the userspace ctr function.
191  */
192 static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
193 			 unsigned int argc, char **argv)
194 {
195 	int r = 0;
196 	int str_size;
197 	char *ctr_str = NULL;
198 	struct log_c *lc = NULL;
199 	uint64_t rdata;
200 	size_t rdata_size = sizeof(rdata);
201 	char *devices_rdata = NULL;
202 	size_t devices_rdata_size = DM_NAME_LEN;
203 
204 	if (argc < 3) {
205 		DMWARN("Too few arguments to userspace dirty log");
206 		return -EINVAL;
207 	}
208 
209 	lc = kzalloc_obj(*lc);
210 	if (!lc) {
211 		DMWARN("Unable to allocate userspace log context.");
212 		return -ENOMEM;
213 	}
214 
215 	/* The ptr value is sufficient for local unique id */
216 	lc->luid = (unsigned long)lc;
217 
218 	lc->ti = ti;
219 
220 	if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
221 		DMWARN("UUID argument too long.");
222 		kfree(lc);
223 		return -EINVAL;
224 	}
225 
226 	lc->usr_argc = argc;
227 
228 	strscpy(lc->uuid, argv[0], sizeof(lc->uuid));
229 	argc--;
230 	argv++;
231 	spin_lock_init(&lc->flush_lock);
232 	INIT_LIST_HEAD(&lc->mark_list);
233 	INIT_LIST_HEAD(&lc->clear_list);
234 
235 	if (!strcasecmp(argv[0], "integrated_flush")) {
236 		lc->integrated_flush = 1;
237 		argc--;
238 		argv++;
239 	}
240 
241 	str_size = build_constructor_string(ti, argc, argv, &ctr_str);
242 	if (str_size < 0) {
243 		kfree(lc);
244 		return str_size;
245 	}
246 
247 	devices_rdata = kzalloc(devices_rdata_size, GFP_KERNEL);
248 	if (!devices_rdata) {
249 		DMERR("Failed to allocate memory for device information");
250 		r = -ENOMEM;
251 		goto out;
252 	}
253 
254 	r = mempool_init_slab_pool(&lc->flush_entry_pool, FLUSH_ENTRY_POOL_SIZE,
255 				   _flush_entry_cache);
256 	if (r) {
257 		DMERR("Failed to create flush_entry_pool");
258 		goto out;
259 	}
260 
261 	/*
262 	 * Send table string and get back any opened device.
263 	 */
264 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
265 				 ctr_str, str_size,
266 				 devices_rdata, &devices_rdata_size);
267 
268 	if (r < 0) {
269 		if (r == -ESRCH)
270 			DMERR("Userspace log server not found");
271 		else
272 			DMERR("Userspace log server failed to create log");
273 		goto out;
274 	}
275 
276 	/* Since the region size does not change, get it now */
277 	rdata_size = sizeof(rdata);
278 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
279 				 NULL, 0, (char *)&rdata, &rdata_size);
280 
281 	if (r) {
282 		DMERR("Failed to get region size of dirty log");
283 		goto out;
284 	}
285 
286 	lc->region_size = (uint32_t)rdata;
287 	lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
288 
289 	if (devices_rdata_size) {
290 		if (devices_rdata[devices_rdata_size - 1] != '\0') {
291 			DMERR("DM_ULOG_CTR device return string not properly terminated");
292 			r = -EINVAL;
293 			goto out;
294 		}
295 		r = dm_get_device(ti, devices_rdata,
296 				  dm_table_get_mode(ti->table), &lc->log_dev);
297 		if (r)
298 			DMERR("Failed to register %s with device-mapper",
299 			      devices_rdata);
300 	}
301 
302 	if (lc->integrated_flush) {
303 		lc->dmlog_wq = alloc_workqueue("dmlogd",
304 					       WQ_MEM_RECLAIM | WQ_PERCPU, 0);
305 		if (!lc->dmlog_wq) {
306 			DMERR("couldn't start dmlogd");
307 			r = -ENOMEM;
308 			goto out;
309 		}
310 
311 		INIT_DELAYED_WORK(&lc->flush_log_work, do_flush);
312 		atomic_set(&lc->sched_flush, 0);
313 	}
314 
315 out:
316 	kfree(devices_rdata);
317 	if (r) {
318 		mempool_exit(&lc->flush_entry_pool);
319 		kfree(lc);
320 		kfree(ctr_str);
321 	} else {
322 		lc->usr_argv_str = ctr_str;
323 		log->context = lc;
324 	}
325 
326 	return r;
327 }
328 
329 static void userspace_dtr(struct dm_dirty_log *log)
330 {
331 	struct log_c *lc = log->context;
332 
333 	if (lc->integrated_flush) {
334 		/* flush workqueue */
335 		if (atomic_read(&lc->sched_flush))
336 			flush_delayed_work(&lc->flush_log_work);
337 
338 		destroy_workqueue(lc->dmlog_wq);
339 	}
340 
341 	(void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
342 				    NULL, 0, NULL, NULL);
343 
344 	if (lc->log_dev)
345 		dm_put_device(lc->ti, lc->log_dev);
346 
347 	mempool_exit(&lc->flush_entry_pool);
348 
349 	kfree(lc->usr_argv_str);
350 	kfree(lc);
351 }
352 
353 static int userspace_presuspend(struct dm_dirty_log *log)
354 {
355 	int r;
356 	struct log_c *lc = log->context;
357 
358 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
359 				 NULL, 0, NULL, NULL);
360 
361 	return r;
362 }
363 
364 static int userspace_postsuspend(struct dm_dirty_log *log)
365 {
366 	int r;
367 	struct log_c *lc = log->context;
368 
369 	/*
370 	 * Run planned flush earlier.
371 	 */
372 	if (lc->integrated_flush && atomic_read(&lc->sched_flush))
373 		flush_delayed_work(&lc->flush_log_work);
374 
375 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
376 				 NULL, 0, NULL, NULL);
377 
378 	return r;
379 }
380 
381 static int userspace_resume(struct dm_dirty_log *log)
382 {
383 	int r;
384 	struct log_c *lc = log->context;
385 
386 	lc->in_sync_hint = 0;
387 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
388 				 NULL, 0, NULL, NULL);
389 
390 	return r;
391 }
392 
393 static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
394 {
395 	struct log_c *lc = log->context;
396 
397 	return lc->region_size;
398 }
399 
400 /*
401  * userspace_is_clean
402  *
403  * Check whether a region is clean.  If there is any sort of
404  * failure when consulting the server, we return not clean.
405  *
406  * Returns: 1 if clean, 0 otherwise
407  */
408 static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
409 {
410 	int r;
411 	uint64_t region64 = (uint64_t)region;
412 	int64_t is_clean;
413 	size_t rdata_size;
414 	struct log_c *lc = log->context;
415 
416 	rdata_size = sizeof(is_clean);
417 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
418 				 (char *)&region64, sizeof(region64),
419 				 (char *)&is_clean, &rdata_size);
420 
421 	return (r) ? 0 : (int)is_clean;
422 }
423 
424 /*
425  * userspace_in_sync
426  *
427  * Check if the region is in-sync.  If there is any sort
428  * of failure when consulting the server, we assume that
429  * the region is not in sync.
430  *
431  * If 'can_block' is set, return immediately
432  *
433  * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
434  */
435 static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
436 			     int can_block)
437 {
438 	int r;
439 	uint64_t region64 = region;
440 	int64_t in_sync;
441 	size_t rdata_size;
442 	struct log_c *lc = log->context;
443 
444 	/*
445 	 * We can never respond directly - even if in_sync_hint is
446 	 * set.  This is because another machine could see a device
447 	 * failure and mark the region out-of-sync.  If we don't go
448 	 * to userspace to ask, we might think the region is in-sync
449 	 * and allow a read to pick up data that is stale.  (This is
450 	 * very unlikely if a device actually fails; but it is very
451 	 * likely if a connection to one device from one machine fails.)
452 	 *
453 	 * There still might be a problem if the mirror caches the region
454 	 * state as in-sync... but then this call would not be made.  So,
455 	 * that is a mirror problem.
456 	 */
457 	if (!can_block)
458 		return -EWOULDBLOCK;
459 
460 	rdata_size = sizeof(in_sync);
461 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
462 				 (char *)&region64, sizeof(region64),
463 				 (char *)&in_sync, &rdata_size);
464 	return (r) ? 0 : (int)in_sync;
465 }
466 
467 static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
468 {
469 	int r = 0;
470 	struct dm_dirty_log_flush_entry *fe;
471 
472 	list_for_each_entry(fe, flush_list, list) {
473 		r = userspace_do_request(lc, lc->uuid, fe->type,
474 					 (char *)&fe->region,
475 					 sizeof(fe->region),
476 					 NULL, NULL);
477 		if (r)
478 			break;
479 	}
480 
481 	return r;
482 }
483 
484 static int flush_by_group(struct log_c *lc, struct list_head *flush_list,
485 			  int flush_with_payload)
486 {
487 	int r = 0;
488 	int count;
489 	uint32_t type = 0;
490 	struct dm_dirty_log_flush_entry *fe, *tmp_fe;
491 	LIST_HEAD(tmp_list);
492 	uint64_t group[MAX_FLUSH_GROUP_COUNT];
493 
494 	/*
495 	 * Group process the requests
496 	 */
497 	while (!list_empty(flush_list)) {
498 		count = 0;
499 
500 		list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
501 			group[count] = fe->region;
502 			count++;
503 
504 			list_move(&fe->list, &tmp_list);
505 
506 			type = fe->type;
507 			if (count >= MAX_FLUSH_GROUP_COUNT)
508 				break;
509 		}
510 
511 		if (flush_with_payload) {
512 			r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
513 						 (char *)(group),
514 						 count * sizeof(uint64_t),
515 						 NULL, NULL);
516 			/*
517 			 * Integrated flush failed.
518 			 */
519 			if (r)
520 				break;
521 		} else {
522 			r = userspace_do_request(lc, lc->uuid, type,
523 						 (char *)(group),
524 						 count * sizeof(uint64_t),
525 						 NULL, NULL);
526 			if (r) {
527 				/*
528 				 * Group send failed.  Attempt one-by-one.
529 				 */
530 				list_splice_init(&tmp_list, flush_list);
531 				r = flush_one_by_one(lc, flush_list);
532 				break;
533 			}
534 		}
535 	}
536 
537 	/*
538 	 * Must collect flush_entrys that were successfully processed
539 	 * as a group so that they will be free'd by the caller.
540 	 */
541 	list_splice_init(&tmp_list, flush_list);
542 
543 	return r;
544 }
545 
546 /*
547  * userspace_flush
548  *
549  * This function is ok to block.
550  * The flush happens in two stages.  First, it sends all
551  * clear/mark requests that are on the list.  Then it
552  * tells the server to commit them.  This gives the
553  * server a chance to optimise the commit, instead of
554  * doing it for every request.
555  *
556  * Additionally, we could implement another thread that
557  * sends the requests up to the server - reducing the
558  * load on flush.  Then the flush would have less in
559  * the list and be responsible for the finishing commit.
560  *
561  * Returns: 0 on success, < 0 on failure
562  */
563 static int userspace_flush(struct dm_dirty_log *log)
564 {
565 	int r = 0;
566 	unsigned long flags;
567 	struct log_c *lc = log->context;
568 	LIST_HEAD(mark_list);
569 	LIST_HEAD(clear_list);
570 	int mark_list_is_empty;
571 	int clear_list_is_empty;
572 	struct dm_dirty_log_flush_entry *fe, *tmp_fe;
573 	mempool_t *flush_entry_pool = &lc->flush_entry_pool;
574 
575 	spin_lock_irqsave(&lc->flush_lock, flags);
576 	list_splice_init(&lc->mark_list, &mark_list);
577 	list_splice_init(&lc->clear_list, &clear_list);
578 	spin_unlock_irqrestore(&lc->flush_lock, flags);
579 
580 	mark_list_is_empty = list_empty(&mark_list);
581 	clear_list_is_empty = list_empty(&clear_list);
582 
583 	if (mark_list_is_empty && clear_list_is_empty)
584 		return 0;
585 
586 	r = flush_by_group(lc, &clear_list, 0);
587 	if (r)
588 		goto out;
589 
590 	if (!lc->integrated_flush) {
591 		r = flush_by_group(lc, &mark_list, 0);
592 		if (r)
593 			goto out;
594 		r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
595 					 NULL, 0, NULL, NULL);
596 		goto out;
597 	}
598 
599 	/*
600 	 * Send integrated flush request with mark_list as payload.
601 	 */
602 	r = flush_by_group(lc, &mark_list, 1);
603 	if (r)
604 		goto out;
605 
606 	if (mark_list_is_empty && !atomic_read(&lc->sched_flush)) {
607 		/*
608 		 * When there are only clear region requests,
609 		 * we schedule a flush in the future.
610 		 */
611 		queue_delayed_work(lc->dmlog_wq, &lc->flush_log_work, 3 * HZ);
612 		atomic_set(&lc->sched_flush, 1);
613 	} else {
614 		/*
615 		 * Cancel pending flush because we
616 		 * have already flushed in mark_region.
617 		 */
618 		cancel_delayed_work(&lc->flush_log_work);
619 		atomic_set(&lc->sched_flush, 0);
620 	}
621 
622 out:
623 	/*
624 	 * We can safely remove these entries, even after failure.
625 	 * Calling code will receive an error and will know that
626 	 * the log facility has failed.
627 	 */
628 	list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) {
629 		list_del(&fe->list);
630 		mempool_free(fe, flush_entry_pool);
631 	}
632 	list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) {
633 		list_del(&fe->list);
634 		mempool_free(fe, flush_entry_pool);
635 	}
636 
637 	if (r)
638 		dm_table_event(lc->ti->table);
639 
640 	return r;
641 }
642 
643 /*
644  * userspace_mark_region
645  *
646  * This function should avoid blocking unless absolutely required.
647  * (Memory allocation is valid for blocking.)
648  */
649 static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
650 {
651 	unsigned long flags;
652 	struct log_c *lc = log->context;
653 	struct dm_dirty_log_flush_entry *fe;
654 
655 	/* Wait for an allocation, but _never_ fail */
656 	fe = mempool_alloc(&lc->flush_entry_pool, GFP_NOIO);
657 	BUG_ON(!fe);
658 
659 	spin_lock_irqsave(&lc->flush_lock, flags);
660 	fe->type = DM_ULOG_MARK_REGION;
661 	fe->region = region;
662 	list_add(&fe->list, &lc->mark_list);
663 	spin_unlock_irqrestore(&lc->flush_lock, flags);
664 }
665 
666 /*
667  * userspace_clear_region
668  *
669  * This function must not block.
670  * So, the alloc can't block.  In the worst case, it is ok to
671  * fail.  It would simply mean we can't clear the region.
672  * Does nothing to current sync context, but does mean
673  * the region will be re-sync'ed on a reload of the mirror
674  * even though it is in-sync.
675  */
676 static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
677 {
678 	unsigned long flags;
679 	struct log_c *lc = log->context;
680 	struct dm_dirty_log_flush_entry *fe;
681 
682 	/*
683 	 * If we fail to allocate, we skip the clearing of
684 	 * the region.  This doesn't hurt us in any way, except
685 	 * to cause the region to be resync'ed when the
686 	 * device is activated next time.
687 	 */
688 	fe = mempool_alloc(&lc->flush_entry_pool, GFP_ATOMIC);
689 	if (!fe) {
690 		DMERR("Failed to allocate memory to clear region.");
691 		return;
692 	}
693 
694 	spin_lock_irqsave(&lc->flush_lock, flags);
695 	fe->type = DM_ULOG_CLEAR_REGION;
696 	fe->region = region;
697 	list_add(&fe->list, &lc->clear_list);
698 	spin_unlock_irqrestore(&lc->flush_lock, flags);
699 }
700 
701 /*
702  * userspace_get_resync_work
703  *
704  * Get a region that needs recovery.  It is valid to return
705  * an error for this function.
706  *
707  * Returns: 1 if region filled, 0 if no work, <0 on error
708  */
709 static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
710 {
711 	int r;
712 	size_t rdata_size;
713 	struct log_c *lc = log->context;
714 	struct {
715 		int64_t i; /* 64-bit for mix arch compatibility */
716 		region_t r;
717 	} pkg;
718 
719 	if (lc->in_sync_hint >= lc->region_count)
720 		return 0;
721 
722 	rdata_size = sizeof(pkg);
723 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
724 				 NULL, 0, (char *)&pkg, &rdata_size);
725 
726 	*region = pkg.r;
727 	return (r) ? r : (int)pkg.i;
728 }
729 
730 /*
731  * userspace_set_region_sync
732  *
733  * Set the sync status of a given region.  This function
734  * must not fail.
735  */
736 static void userspace_set_region_sync(struct dm_dirty_log *log,
737 				      region_t region, int in_sync)
738 {
739 	struct log_c *lc = log->context;
740 	struct {
741 		region_t r;
742 		int64_t i;
743 	} pkg;
744 
745 	pkg.r = region;
746 	pkg.i = (int64_t)in_sync;
747 
748 	(void) userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
749 				    (char *)&pkg, sizeof(pkg), NULL, NULL);
750 
751 	/*
752 	 * It would be nice to be able to report failures.
753 	 * However, it is easy enough to detect and resolve.
754 	 */
755 }
756 
757 /*
758  * userspace_get_sync_count
759  *
760  * If there is any sort of failure when consulting the server,
761  * we assume that the sync count is zero.
762  *
763  * Returns: sync count on success, 0 on failure
764  */
765 static region_t userspace_get_sync_count(struct dm_dirty_log *log)
766 {
767 	int r;
768 	size_t rdata_size;
769 	uint64_t sync_count;
770 	struct log_c *lc = log->context;
771 
772 	rdata_size = sizeof(sync_count);
773 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
774 				 NULL, 0, (char *)&sync_count, &rdata_size);
775 
776 	if (r)
777 		return 0;
778 
779 	if (sync_count >= lc->region_count)
780 		lc->in_sync_hint = lc->region_count;
781 
782 	return (region_t)sync_count;
783 }
784 
785 /*
786  * userspace_status
787  *
788  * Returns: amount of space consumed
789  */
790 static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
791 			    char *result, unsigned int maxlen)
792 {
793 	int r = 0;
794 	char *table_args;
795 	size_t sz = (size_t)maxlen;
796 	struct log_c *lc = log->context;
797 
798 	switch (status_type) {
799 	case STATUSTYPE_INFO:
800 		r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
801 					 NULL, 0, result, &sz);
802 
803 		if (r) {
804 			sz = 0;
805 			DMEMIT("%s 1 COM_FAILURE", log->type->name);
806 		}
807 		break;
808 	case STATUSTYPE_TABLE:
809 		sz = 0;
810 		table_args = strchr(lc->usr_argv_str, ' ');
811 		BUG_ON(!table_args); /* There will always be a ' ' */
812 		table_args++;
813 
814 		DMEMIT("%s %u %s ", log->type->name, lc->usr_argc, lc->uuid);
815 		if (lc->integrated_flush)
816 			DMEMIT("integrated_flush ");
817 		DMEMIT("%s ", table_args);
818 		break;
819 	case STATUSTYPE_IMA:
820 		*result = '\0';
821 		break;
822 	}
823 	return (r) ? 0 : (int)sz;
824 }
825 
826 /*
827  * userspace_is_remote_recovering
828  *
829  * Returns: 1 if region recovering, 0 otherwise
830  */
831 static int userspace_is_remote_recovering(struct dm_dirty_log *log,
832 					  region_t region)
833 {
834 	int r;
835 	uint64_t region64 = region;
836 	struct log_c *lc = log->context;
837 	static unsigned long limit;
838 	struct {
839 		int64_t is_recovering;
840 		uint64_t in_sync_hint;
841 	} pkg;
842 	size_t rdata_size = sizeof(pkg);
843 
844 	/*
845 	 * Once the mirror has been reported to be in-sync,
846 	 * it will never again ask for recovery work.  So,
847 	 * we can safely say there is not a remote machine
848 	 * recovering if the device is in-sync.  (in_sync_hint
849 	 * must be reset at resume time.)
850 	 */
851 	if (region < lc->in_sync_hint)
852 		return 0;
853 	else if (time_after(limit, jiffies))
854 		return 1;
855 
856 	limit = jiffies + (HZ / 4);
857 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
858 				 (char *)&region64, sizeof(region64),
859 				 (char *)&pkg, &rdata_size);
860 	if (r)
861 		return 1;
862 
863 	lc->in_sync_hint = pkg.in_sync_hint;
864 
865 	return (int)pkg.is_recovering;
866 }
867 
868 static struct dm_dirty_log_type _userspace_type = {
869 	.name = "userspace",
870 	.module = THIS_MODULE,
871 	.ctr = userspace_ctr,
872 	.dtr = userspace_dtr,
873 	.presuspend = userspace_presuspend,
874 	.postsuspend = userspace_postsuspend,
875 	.resume = userspace_resume,
876 	.get_region_size = userspace_get_region_size,
877 	.is_clean = userspace_is_clean,
878 	.in_sync = userspace_in_sync,
879 	.flush = userspace_flush,
880 	.mark_region = userspace_mark_region,
881 	.clear_region = userspace_clear_region,
882 	.get_resync_work = userspace_get_resync_work,
883 	.set_region_sync = userspace_set_region_sync,
884 	.get_sync_count = userspace_get_sync_count,
885 	.status = userspace_status,
886 	.is_remote_recovering = userspace_is_remote_recovering,
887 };
888 
889 static int __init userspace_dirty_log_init(void)
890 {
891 	int r = 0;
892 
893 	_flush_entry_cache = KMEM_CACHE(dm_dirty_log_flush_entry, 0);
894 	if (!_flush_entry_cache) {
895 		DMWARN("Unable to create flush_entry_cache: No memory.");
896 		return -ENOMEM;
897 	}
898 
899 	r = dm_ulog_tfr_init();
900 	if (r) {
901 		DMWARN("Unable to initialize userspace log communications");
902 		kmem_cache_destroy(_flush_entry_cache);
903 		return r;
904 	}
905 
906 	r = dm_dirty_log_type_register(&_userspace_type);
907 	if (r) {
908 		DMWARN("Couldn't register userspace dirty log type");
909 		dm_ulog_tfr_exit();
910 		kmem_cache_destroy(_flush_entry_cache);
911 		return r;
912 	}
913 
914 	DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
915 	return 0;
916 }
917 
918 static void __exit userspace_dirty_log_exit(void)
919 {
920 	dm_dirty_log_type_unregister(&_userspace_type);
921 	dm_ulog_tfr_exit();
922 	kmem_cache_destroy(_flush_entry_cache);
923 
924 	DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
925 }
926 
927 module_init(userspace_dirty_log_init);
928 module_exit(userspace_dirty_log_exit);
929 
930 MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
931 MODULE_AUTHOR("Jonathan Brassow <dm-devel@lists.linux.dev>");
932 MODULE_LICENSE("GPL");
933