Loading...
1/*
2 * Copyright (C) 2006-2009 Red Hat, Inc.
3 *
4 * This file is released under the LGPL.
5 */
6
7#include <linux/bio.h>
8#include <linux/slab.h>
9#include <linux/dm-dirty-log.h>
10#include <linux/device-mapper.h>
11#include <linux/dm-log-userspace.h>
12#include <linux/module.h>
13#include <linux/workqueue.h>
14
15#include "dm-log-userspace-transfer.h"
16
17#define DM_LOG_USERSPACE_VSN "1.3.0"
18
19struct flush_entry {
20 int type;
21 region_t region;
22 struct list_head list;
23};
24
25/*
26 * This limit on the number of mark and clear request is, to a degree,
27 * arbitrary. However, there is some basis for the choice in the limits
28 * imposed on the size of data payload by dm-log-userspace-transfer.c:
29 * dm_consult_userspace().
30 */
31#define MAX_FLUSH_GROUP_COUNT 32
32
33struct log_c {
34 struct dm_target *ti;
35 struct dm_dev *log_dev;
36 uint32_t region_size;
37 region_t region_count;
38 uint64_t luid;
39 char uuid[DM_UUID_LEN];
40
41 char *usr_argv_str;
42 uint32_t usr_argc;
43
44 /*
45 * in_sync_hint gets set when doing is_remote_recovering. It
46 * represents the first region that needs recovery. IOW, the
47 * first zero bit of sync_bits. This can be useful for to limit
48 * traffic for calls like is_remote_recovering and get_resync_work,
49 * but be take care in its use for anything else.
50 */
51 uint64_t in_sync_hint;
52
53 /*
54 * Mark and clear requests are held until a flush is issued
55 * so that we can group, and thereby limit, the amount of
56 * network traffic between kernel and userspace. The 'flush_lock'
57 * is used to protect these lists.
58 */
59 spinlock_t flush_lock;
60 struct list_head mark_list;
61 struct list_head clear_list;
62
63 /*
64 * Workqueue for flush of clear region requests.
65 */
66 struct workqueue_struct *dmlog_wq;
67 struct delayed_work flush_log_work;
68 atomic_t sched_flush;
69
70 /*
71 * Combine userspace flush and mark requests for efficiency.
72 */
73 uint32_t integrated_flush;
74};
75
76static mempool_t *flush_entry_pool;
77
78static void *flush_entry_alloc(gfp_t gfp_mask, void *pool_data)
79{
80 return kmalloc(sizeof(struct flush_entry), gfp_mask);
81}
82
83static void flush_entry_free(void *element, void *pool_data)
84{
85 kfree(element);
86}
87
88static int userspace_do_request(struct log_c *lc, const char *uuid,
89 int request_type, char *data, size_t data_size,
90 char *rdata, size_t *rdata_size)
91{
92 int r;
93
94 /*
95 * If the server isn't there, -ESRCH is returned,
96 * and we must keep trying until the server is
97 * restored.
98 */
99retry:
100 r = dm_consult_userspace(uuid, lc->luid, request_type, data,
101 data_size, rdata, rdata_size);
102
103 if (r != -ESRCH)
104 return r;
105
106 DMERR(" Userspace log server not found.");
107 while (1) {
108 set_current_state(TASK_INTERRUPTIBLE);
109 schedule_timeout(2*HZ);
110 DMWARN("Attempting to contact userspace log server...");
111 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR,
112 lc->usr_argv_str,
113 strlen(lc->usr_argv_str) + 1,
114 NULL, NULL);
115 if (!r)
116 break;
117 }
118 DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
119 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL,
120 0, NULL, NULL);
121 if (!r)
122 goto retry;
123
124 DMERR("Error trying to resume userspace log: %d", r);
125
126 return -ESRCH;
127}
128
129static int build_constructor_string(struct dm_target *ti,
130 unsigned argc, char **argv,
131 char **ctr_str)
132{
133 int i, str_size;
134 char *str = NULL;
135
136 *ctr_str = NULL;
137
138 /*
139 * Determine overall size of the string.
140 */
141 for (i = 0, str_size = 0; i < argc; i++)
142 str_size += strlen(argv[i]) + 1; /* +1 for space between args */
143
144 str_size += 20; /* Max number of chars in a printed u64 number */
145
146 str = kzalloc(str_size, GFP_KERNEL);
147 if (!str) {
148 DMWARN("Unable to allocate memory for constructor string");
149 return -ENOMEM;
150 }
151
152 str_size = sprintf(str, "%llu", (unsigned long long)ti->len);
153 for (i = 0; i < argc; i++)
154 str_size += sprintf(str + str_size, " %s", argv[i]);
155
156 *ctr_str = str;
157 return str_size;
158}
159
160static void do_flush(struct work_struct *work)
161{
162 int r;
163 struct log_c *lc = container_of(work, struct log_c, flush_log_work.work);
164
165 atomic_set(&lc->sched_flush, 0);
166
167 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, NULL, 0, NULL, NULL);
168
169 if (r)
170 dm_table_event(lc->ti->table);
171}
172
173/*
174 * userspace_ctr
175 *
176 * argv contains:
177 * <UUID> [integrated_flush] <other args>
178 * Where 'other args' are the userspace implementation-specific log
179 * arguments.
180 *
181 * Example:
182 * <UUID> [integrated_flush] clustered-disk <arg count> <log dev>
183 * <region_size> [[no]sync]
184 *
185 * This module strips off the <UUID> and uses it for identification
186 * purposes when communicating with userspace about a log.
187 *
188 * If integrated_flush is defined, the kernel combines flush
189 * and mark requests.
190 *
191 * The rest of the line, beginning with 'clustered-disk', is passed
192 * to the userspace ctr function.
193 */
194static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
195 unsigned argc, char **argv)
196{
197 int r = 0;
198 int str_size;
199 char *ctr_str = NULL;
200 struct log_c *lc = NULL;
201 uint64_t rdata;
202 size_t rdata_size = sizeof(rdata);
203 char *devices_rdata = NULL;
204 size_t devices_rdata_size = DM_NAME_LEN;
205
206 if (argc < 3) {
207 DMWARN("Too few arguments to userspace dirty log");
208 return -EINVAL;
209 }
210
211 lc = kzalloc(sizeof(*lc), GFP_KERNEL);
212 if (!lc) {
213 DMWARN("Unable to allocate userspace log context.");
214 return -ENOMEM;
215 }
216
217 /* The ptr value is sufficient for local unique id */
218 lc->luid = (unsigned long)lc;
219
220 lc->ti = ti;
221
222 if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
223 DMWARN("UUID argument too long.");
224 kfree(lc);
225 return -EINVAL;
226 }
227
228 lc->usr_argc = argc;
229
230 strncpy(lc->uuid, argv[0], DM_UUID_LEN);
231 argc--;
232 argv++;
233 spin_lock_init(&lc->flush_lock);
234 INIT_LIST_HEAD(&lc->mark_list);
235 INIT_LIST_HEAD(&lc->clear_list);
236
237 if (!strcasecmp(argv[0], "integrated_flush")) {
238 lc->integrated_flush = 1;
239 argc--;
240 argv++;
241 }
242
243 str_size = build_constructor_string(ti, argc, argv, &ctr_str);
244 if (str_size < 0) {
245 kfree(lc);
246 return str_size;
247 }
248
249 devices_rdata = kzalloc(devices_rdata_size, GFP_KERNEL);
250 if (!devices_rdata) {
251 DMERR("Failed to allocate memory for device information");
252 r = -ENOMEM;
253 goto out;
254 }
255
256 /*
257 * Send table string and get back any opened device.
258 */
259 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
260 ctr_str, str_size,
261 devices_rdata, &devices_rdata_size);
262
263 if (r < 0) {
264 if (r == -ESRCH)
265 DMERR("Userspace log server not found");
266 else
267 DMERR("Userspace log server failed to create log");
268 goto out;
269 }
270
271 /* Since the region size does not change, get it now */
272 rdata_size = sizeof(rdata);
273 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
274 NULL, 0, (char *)&rdata, &rdata_size);
275
276 if (r) {
277 DMERR("Failed to get region size of dirty log");
278 goto out;
279 }
280
281 lc->region_size = (uint32_t)rdata;
282 lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
283
284 if (devices_rdata_size) {
285 if (devices_rdata[devices_rdata_size - 1] != '\0') {
286 DMERR("DM_ULOG_CTR device return string not properly terminated");
287 r = -EINVAL;
288 goto out;
289 }
290 r = dm_get_device(ti, devices_rdata,
291 dm_table_get_mode(ti->table), &lc->log_dev);
292 if (r)
293 DMERR("Failed to register %s with device-mapper",
294 devices_rdata);
295 }
296
297 if (lc->integrated_flush) {
298 lc->dmlog_wq = alloc_workqueue("dmlogd", WQ_MEM_RECLAIM, 0);
299 if (!lc->dmlog_wq) {
300 DMERR("couldn't start dmlogd");
301 r = -ENOMEM;
302 goto out;
303 }
304
305 INIT_DELAYED_WORK(&lc->flush_log_work, do_flush);
306 atomic_set(&lc->sched_flush, 0);
307 }
308
309out:
310 kfree(devices_rdata);
311 if (r) {
312 kfree(lc);
313 kfree(ctr_str);
314 } else {
315 lc->usr_argv_str = ctr_str;
316 log->context = lc;
317 }
318
319 return r;
320}
321
322static void userspace_dtr(struct dm_dirty_log *log)
323{
324 struct log_c *lc = log->context;
325
326 if (lc->integrated_flush) {
327 /* flush workqueue */
328 if (atomic_read(&lc->sched_flush))
329 flush_delayed_work(&lc->flush_log_work);
330
331 destroy_workqueue(lc->dmlog_wq);
332 }
333
334 (void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
335 NULL, 0, NULL, NULL);
336
337 if (lc->log_dev)
338 dm_put_device(lc->ti, lc->log_dev);
339
340 kfree(lc->usr_argv_str);
341 kfree(lc);
342
343 return;
344}
345
346static int userspace_presuspend(struct dm_dirty_log *log)
347{
348 int r;
349 struct log_c *lc = log->context;
350
351 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
352 NULL, 0, NULL, NULL);
353
354 return r;
355}
356
357static int userspace_postsuspend(struct dm_dirty_log *log)
358{
359 int r;
360 struct log_c *lc = log->context;
361
362 /*
363 * Run planned flush earlier.
364 */
365 if (lc->integrated_flush && atomic_read(&lc->sched_flush))
366 flush_delayed_work(&lc->flush_log_work);
367
368 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
369 NULL, 0, NULL, NULL);
370
371 return r;
372}
373
374static int userspace_resume(struct dm_dirty_log *log)
375{
376 int r;
377 struct log_c *lc = log->context;
378
379 lc->in_sync_hint = 0;
380 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
381 NULL, 0, NULL, NULL);
382
383 return r;
384}
385
386static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
387{
388 struct log_c *lc = log->context;
389
390 return lc->region_size;
391}
392
393/*
394 * userspace_is_clean
395 *
396 * Check whether a region is clean. If there is any sort of
397 * failure when consulting the server, we return not clean.
398 *
399 * Returns: 1 if clean, 0 otherwise
400 */
401static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
402{
403 int r;
404 uint64_t region64 = (uint64_t)region;
405 int64_t is_clean;
406 size_t rdata_size;
407 struct log_c *lc = log->context;
408
409 rdata_size = sizeof(is_clean);
410 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
411 (char *)®ion64, sizeof(region64),
412 (char *)&is_clean, &rdata_size);
413
414 return (r) ? 0 : (int)is_clean;
415}
416
417/*
418 * userspace_in_sync
419 *
420 * Check if the region is in-sync. If there is any sort
421 * of failure when consulting the server, we assume that
422 * the region is not in sync.
423 *
424 * If 'can_block' is set, return immediately
425 *
426 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
427 */
428static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
429 int can_block)
430{
431 int r;
432 uint64_t region64 = region;
433 int64_t in_sync;
434 size_t rdata_size;
435 struct log_c *lc = log->context;
436
437 /*
438 * We can never respond directly - even if in_sync_hint is
439 * set. This is because another machine could see a device
440 * failure and mark the region out-of-sync. If we don't go
441 * to userspace to ask, we might think the region is in-sync
442 * and allow a read to pick up data that is stale. (This is
443 * very unlikely if a device actually fails; but it is very
444 * likely if a connection to one device from one machine fails.)
445 *
446 * There still might be a problem if the mirror caches the region
447 * state as in-sync... but then this call would not be made. So,
448 * that is a mirror problem.
449 */
450 if (!can_block)
451 return -EWOULDBLOCK;
452
453 rdata_size = sizeof(in_sync);
454 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
455 (char *)®ion64, sizeof(region64),
456 (char *)&in_sync, &rdata_size);
457 return (r) ? 0 : (int)in_sync;
458}
459
460static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
461{
462 int r = 0;
463 struct flush_entry *fe;
464
465 list_for_each_entry(fe, flush_list, list) {
466 r = userspace_do_request(lc, lc->uuid, fe->type,
467 (char *)&fe->region,
468 sizeof(fe->region),
469 NULL, NULL);
470 if (r)
471 break;
472 }
473
474 return r;
475}
476
477static int flush_by_group(struct log_c *lc, struct list_head *flush_list,
478 int flush_with_payload)
479{
480 int r = 0;
481 int count;
482 uint32_t type = 0;
483 struct flush_entry *fe, *tmp_fe;
484 LIST_HEAD(tmp_list);
485 uint64_t group[MAX_FLUSH_GROUP_COUNT];
486
487 /*
488 * Group process the requests
489 */
490 while (!list_empty(flush_list)) {
491 count = 0;
492
493 list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
494 group[count] = fe->region;
495 count++;
496
497 list_move(&fe->list, &tmp_list);
498
499 type = fe->type;
500 if (count >= MAX_FLUSH_GROUP_COUNT)
501 break;
502 }
503
504 if (flush_with_payload) {
505 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
506 (char *)(group),
507 count * sizeof(uint64_t),
508 NULL, NULL);
509 /*
510 * Integrated flush failed.
511 */
512 if (r)
513 break;
514 } else {
515 r = userspace_do_request(lc, lc->uuid, type,
516 (char *)(group),
517 count * sizeof(uint64_t),
518 NULL, NULL);
519 if (r) {
520 /*
521 * Group send failed. Attempt one-by-one.
522 */
523 list_splice_init(&tmp_list, flush_list);
524 r = flush_one_by_one(lc, flush_list);
525 break;
526 }
527 }
528 }
529
530 /*
531 * Must collect flush_entrys that were successfully processed
532 * as a group so that they will be free'd by the caller.
533 */
534 list_splice_init(&tmp_list, flush_list);
535
536 return r;
537}
538
539/*
540 * userspace_flush
541 *
542 * This function is ok to block.
543 * The flush happens in two stages. First, it sends all
544 * clear/mark requests that are on the list. Then it
545 * tells the server to commit them. This gives the
546 * server a chance to optimise the commit, instead of
547 * doing it for every request.
548 *
549 * Additionally, we could implement another thread that
550 * sends the requests up to the server - reducing the
551 * load on flush. Then the flush would have less in
552 * the list and be responsible for the finishing commit.
553 *
554 * Returns: 0 on success, < 0 on failure
555 */
556static int userspace_flush(struct dm_dirty_log *log)
557{
558 int r = 0;
559 unsigned long flags;
560 struct log_c *lc = log->context;
561 LIST_HEAD(mark_list);
562 LIST_HEAD(clear_list);
563 int mark_list_is_empty;
564 int clear_list_is_empty;
565 struct flush_entry *fe, *tmp_fe;
566
567 spin_lock_irqsave(&lc->flush_lock, flags);
568 list_splice_init(&lc->mark_list, &mark_list);
569 list_splice_init(&lc->clear_list, &clear_list);
570 spin_unlock_irqrestore(&lc->flush_lock, flags);
571
572 mark_list_is_empty = list_empty(&mark_list);
573 clear_list_is_empty = list_empty(&clear_list);
574
575 if (mark_list_is_empty && clear_list_is_empty)
576 return 0;
577
578 r = flush_by_group(lc, &clear_list, 0);
579 if (r)
580 goto out;
581
582 if (!lc->integrated_flush) {
583 r = flush_by_group(lc, &mark_list, 0);
584 if (r)
585 goto out;
586 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
587 NULL, 0, NULL, NULL);
588 goto out;
589 }
590
591 /*
592 * Send integrated flush request with mark_list as payload.
593 */
594 r = flush_by_group(lc, &mark_list, 1);
595 if (r)
596 goto out;
597
598 if (mark_list_is_empty && !atomic_read(&lc->sched_flush)) {
599 /*
600 * When there are only clear region requests,
601 * we schedule a flush in the future.
602 */
603 queue_delayed_work(lc->dmlog_wq, &lc->flush_log_work, 3 * HZ);
604 atomic_set(&lc->sched_flush, 1);
605 } else {
606 /*
607 * Cancel pending flush because we
608 * have already flushed in mark_region.
609 */
610 cancel_delayed_work(&lc->flush_log_work);
611 atomic_set(&lc->sched_flush, 0);
612 }
613
614out:
615 /*
616 * We can safely remove these entries, even after failure.
617 * Calling code will receive an error and will know that
618 * the log facility has failed.
619 */
620 list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) {
621 list_del(&fe->list);
622 mempool_free(fe, flush_entry_pool);
623 }
624 list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) {
625 list_del(&fe->list);
626 mempool_free(fe, flush_entry_pool);
627 }
628
629 if (r)
630 dm_table_event(lc->ti->table);
631
632 return r;
633}
634
635/*
636 * userspace_mark_region
637 *
638 * This function should avoid blocking unless absolutely required.
639 * (Memory allocation is valid for blocking.)
640 */
641static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
642{
643 unsigned long flags;
644 struct log_c *lc = log->context;
645 struct flush_entry *fe;
646
647 /* Wait for an allocation, but _never_ fail */
648 fe = mempool_alloc(flush_entry_pool, GFP_NOIO);
649 BUG_ON(!fe);
650
651 spin_lock_irqsave(&lc->flush_lock, flags);
652 fe->type = DM_ULOG_MARK_REGION;
653 fe->region = region;
654 list_add(&fe->list, &lc->mark_list);
655 spin_unlock_irqrestore(&lc->flush_lock, flags);
656
657 return;
658}
659
660/*
661 * userspace_clear_region
662 *
663 * This function must not block.
664 * So, the alloc can't block. In the worst case, it is ok to
665 * fail. It would simply mean we can't clear the region.
666 * Does nothing to current sync context, but does mean
667 * the region will be re-sync'ed on a reload of the mirror
668 * even though it is in-sync.
669 */
670static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
671{
672 unsigned long flags;
673 struct log_c *lc = log->context;
674 struct flush_entry *fe;
675
676 /*
677 * If we fail to allocate, we skip the clearing of
678 * the region. This doesn't hurt us in any way, except
679 * to cause the region to be resync'ed when the
680 * device is activated next time.
681 */
682 fe = mempool_alloc(flush_entry_pool, GFP_ATOMIC);
683 if (!fe) {
684 DMERR("Failed to allocate memory to clear region.");
685 return;
686 }
687
688 spin_lock_irqsave(&lc->flush_lock, flags);
689 fe->type = DM_ULOG_CLEAR_REGION;
690 fe->region = region;
691 list_add(&fe->list, &lc->clear_list);
692 spin_unlock_irqrestore(&lc->flush_lock, flags);
693
694 return;
695}
696
697/*
698 * userspace_get_resync_work
699 *
700 * Get a region that needs recovery. It is valid to return
701 * an error for this function.
702 *
703 * Returns: 1 if region filled, 0 if no work, <0 on error
704 */
705static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
706{
707 int r;
708 size_t rdata_size;
709 struct log_c *lc = log->context;
710 struct {
711 int64_t i; /* 64-bit for mix arch compatibility */
712 region_t r;
713 } pkg;
714
715 if (lc->in_sync_hint >= lc->region_count)
716 return 0;
717
718 rdata_size = sizeof(pkg);
719 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
720 NULL, 0, (char *)&pkg, &rdata_size);
721
722 *region = pkg.r;
723 return (r) ? r : (int)pkg.i;
724}
725
726/*
727 * userspace_set_region_sync
728 *
729 * Set the sync status of a given region. This function
730 * must not fail.
731 */
732static void userspace_set_region_sync(struct dm_dirty_log *log,
733 region_t region, int in_sync)
734{
735 int r;
736 struct log_c *lc = log->context;
737 struct {
738 region_t r;
739 int64_t i;
740 } pkg;
741
742 pkg.r = region;
743 pkg.i = (int64_t)in_sync;
744
745 r = userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
746 (char *)&pkg, sizeof(pkg), NULL, NULL);
747
748 /*
749 * It would be nice to be able to report failures.
750 * However, it is easy emough to detect and resolve.
751 */
752 return;
753}
754
755/*
756 * userspace_get_sync_count
757 *
758 * If there is any sort of failure when consulting the server,
759 * we assume that the sync count is zero.
760 *
761 * Returns: sync count on success, 0 on failure
762 */
763static region_t userspace_get_sync_count(struct dm_dirty_log *log)
764{
765 int r;
766 size_t rdata_size;
767 uint64_t sync_count;
768 struct log_c *lc = log->context;
769
770 rdata_size = sizeof(sync_count);
771 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
772 NULL, 0, (char *)&sync_count, &rdata_size);
773
774 if (r)
775 return 0;
776
777 if (sync_count >= lc->region_count)
778 lc->in_sync_hint = lc->region_count;
779
780 return (region_t)sync_count;
781}
782
783/*
784 * userspace_status
785 *
786 * Returns: amount of space consumed
787 */
788static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
789 char *result, unsigned maxlen)
790{
791 int r = 0;
792 char *table_args;
793 size_t sz = (size_t)maxlen;
794 struct log_c *lc = log->context;
795
796 switch (status_type) {
797 case STATUSTYPE_INFO:
798 r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
799 NULL, 0, result, &sz);
800
801 if (r) {
802 sz = 0;
803 DMEMIT("%s 1 COM_FAILURE", log->type->name);
804 }
805 break;
806 case STATUSTYPE_TABLE:
807 sz = 0;
808 table_args = strchr(lc->usr_argv_str, ' ');
809 BUG_ON(!table_args); /* There will always be a ' ' */
810 table_args++;
811
812 DMEMIT("%s %u %s ", log->type->name, lc->usr_argc, lc->uuid);
813 if (lc->integrated_flush)
814 DMEMIT("integrated_flush ");
815 DMEMIT("%s ", table_args);
816 break;
817 }
818 return (r) ? 0 : (int)sz;
819}
820
821/*
822 * userspace_is_remote_recovering
823 *
824 * Returns: 1 if region recovering, 0 otherwise
825 */
826static int userspace_is_remote_recovering(struct dm_dirty_log *log,
827 region_t region)
828{
829 int r;
830 uint64_t region64 = region;
831 struct log_c *lc = log->context;
832 static unsigned long long limit;
833 struct {
834 int64_t is_recovering;
835 uint64_t in_sync_hint;
836 } pkg;
837 size_t rdata_size = sizeof(pkg);
838
839 /*
840 * Once the mirror has been reported to be in-sync,
841 * it will never again ask for recovery work. So,
842 * we can safely say there is not a remote machine
843 * recovering if the device is in-sync. (in_sync_hint
844 * must be reset at resume time.)
845 */
846 if (region < lc->in_sync_hint)
847 return 0;
848 else if (jiffies < limit)
849 return 1;
850
851 limit = jiffies + (HZ / 4);
852 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
853 (char *)®ion64, sizeof(region64),
854 (char *)&pkg, &rdata_size);
855 if (r)
856 return 1;
857
858 lc->in_sync_hint = pkg.in_sync_hint;
859
860 return (int)pkg.is_recovering;
861}
862
863static struct dm_dirty_log_type _userspace_type = {
864 .name = "userspace",
865 .module = THIS_MODULE,
866 .ctr = userspace_ctr,
867 .dtr = userspace_dtr,
868 .presuspend = userspace_presuspend,
869 .postsuspend = userspace_postsuspend,
870 .resume = userspace_resume,
871 .get_region_size = userspace_get_region_size,
872 .is_clean = userspace_is_clean,
873 .in_sync = userspace_in_sync,
874 .flush = userspace_flush,
875 .mark_region = userspace_mark_region,
876 .clear_region = userspace_clear_region,
877 .get_resync_work = userspace_get_resync_work,
878 .set_region_sync = userspace_set_region_sync,
879 .get_sync_count = userspace_get_sync_count,
880 .status = userspace_status,
881 .is_remote_recovering = userspace_is_remote_recovering,
882};
883
884static int __init userspace_dirty_log_init(void)
885{
886 int r = 0;
887
888 flush_entry_pool = mempool_create(100, flush_entry_alloc,
889 flush_entry_free, NULL);
890
891 if (!flush_entry_pool) {
892 DMWARN("Unable to create flush_entry_pool: No memory.");
893 return -ENOMEM;
894 }
895
896 r = dm_ulog_tfr_init();
897 if (r) {
898 DMWARN("Unable to initialize userspace log communications");
899 mempool_destroy(flush_entry_pool);
900 return r;
901 }
902
903 r = dm_dirty_log_type_register(&_userspace_type);
904 if (r) {
905 DMWARN("Couldn't register userspace dirty log type");
906 dm_ulog_tfr_exit();
907 mempool_destroy(flush_entry_pool);
908 return r;
909 }
910
911 DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
912 return 0;
913}
914
915static void __exit userspace_dirty_log_exit(void)
916{
917 dm_dirty_log_type_unregister(&_userspace_type);
918 dm_ulog_tfr_exit();
919 mempool_destroy(flush_entry_pool);
920
921 DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
922 return;
923}
924
925module_init(userspace_dirty_log_init);
926module_exit(userspace_dirty_log_exit);
927
928MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
929MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>");
930MODULE_LICENSE("GPL");
1/*
2 * Copyright (C) 2006-2009 Red Hat, Inc.
3 *
4 * This file is released under the LGPL.
5 */
6
7#include <linux/bio.h>
8#include <linux/slab.h>
9#include <linux/jiffies.h>
10#include <linux/dm-dirty-log.h>
11#include <linux/device-mapper.h>
12#include <linux/dm-log-userspace.h>
13#include <linux/module.h>
14#include <linux/workqueue.h>
15
16#include "dm-log-userspace-transfer.h"
17
18#define DM_LOG_USERSPACE_VSN "1.3.0"
19
20#define FLUSH_ENTRY_POOL_SIZE 16
21
22struct dm_dirty_log_flush_entry {
23 int type;
24 region_t region;
25 struct list_head list;
26};
27
28/*
29 * This limit on the number of mark and clear request is, to a degree,
30 * arbitrary. However, there is some basis for the choice in the limits
31 * imposed on the size of data payload by dm-log-userspace-transfer.c:
32 * dm_consult_userspace().
33 */
34#define MAX_FLUSH_GROUP_COUNT 32
35
36struct log_c {
37 struct dm_target *ti;
38 struct dm_dev *log_dev;
39
40 char *usr_argv_str;
41 uint32_t usr_argc;
42
43 uint32_t region_size;
44 region_t region_count;
45 uint64_t luid;
46 char uuid[DM_UUID_LEN];
47
48 /*
49 * Mark and clear requests are held until a flush is issued
50 * so that we can group, and thereby limit, the amount of
51 * network traffic between kernel and userspace. The 'flush_lock'
52 * is used to protect these lists.
53 */
54 spinlock_t flush_lock;
55 struct list_head mark_list;
56 struct list_head clear_list;
57
58 /*
59 * in_sync_hint gets set when doing is_remote_recovering. It
60 * represents the first region that needs recovery. IOW, the
61 * first zero bit of sync_bits. This can be useful for to limit
62 * traffic for calls like is_remote_recovering and get_resync_work,
63 * but be take care in its use for anything else.
64 */
65 uint64_t in_sync_hint;
66
67 /*
68 * Workqueue for flush of clear region requests.
69 */
70 struct workqueue_struct *dmlog_wq;
71 struct delayed_work flush_log_work;
72 atomic_t sched_flush;
73
74 /*
75 * Combine userspace flush and mark requests for efficiency.
76 */
77 uint32_t integrated_flush;
78
79 mempool_t *flush_entry_pool;
80};
81
82static struct kmem_cache *_flush_entry_cache;
83
84static int userspace_do_request(struct log_c *lc, const char *uuid,
85 int request_type, char *data, size_t data_size,
86 char *rdata, size_t *rdata_size)
87{
88 int r;
89
90 /*
91 * If the server isn't there, -ESRCH is returned,
92 * and we must keep trying until the server is
93 * restored.
94 */
95retry:
96 r = dm_consult_userspace(uuid, lc->luid, request_type, data,
97 data_size, rdata, rdata_size);
98
99 if (r != -ESRCH)
100 return r;
101
102 DMERR(" Userspace log server not found.");
103 while (1) {
104 set_current_state(TASK_INTERRUPTIBLE);
105 schedule_timeout(2*HZ);
106 DMWARN("Attempting to contact userspace log server...");
107 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR,
108 lc->usr_argv_str,
109 strlen(lc->usr_argv_str) + 1,
110 NULL, NULL);
111 if (!r)
112 break;
113 }
114 DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
115 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL,
116 0, NULL, NULL);
117 if (!r)
118 goto retry;
119
120 DMERR("Error trying to resume userspace log: %d", r);
121
122 return -ESRCH;
123}
124
125static int build_constructor_string(struct dm_target *ti,
126 unsigned argc, char **argv,
127 char **ctr_str)
128{
129 int i, str_size;
130 char *str = NULL;
131
132 *ctr_str = NULL;
133
134 /*
135 * Determine overall size of the string.
136 */
137 for (i = 0, str_size = 0; i < argc; i++)
138 str_size += strlen(argv[i]) + 1; /* +1 for space between args */
139
140 str_size += 20; /* Max number of chars in a printed u64 number */
141
142 str = kzalloc(str_size, GFP_KERNEL);
143 if (!str) {
144 DMWARN("Unable to allocate memory for constructor string");
145 return -ENOMEM;
146 }
147
148 str_size = sprintf(str, "%llu", (unsigned long long)ti->len);
149 for (i = 0; i < argc; i++)
150 str_size += sprintf(str + str_size, " %s", argv[i]);
151
152 *ctr_str = str;
153 return str_size;
154}
155
156static void do_flush(struct work_struct *work)
157{
158 int r;
159 struct log_c *lc = container_of(work, struct log_c, flush_log_work.work);
160
161 atomic_set(&lc->sched_flush, 0);
162
163 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, NULL, 0, NULL, NULL);
164
165 if (r)
166 dm_table_event(lc->ti->table);
167}
168
169/*
170 * userspace_ctr
171 *
172 * argv contains:
173 * <UUID> [integrated_flush] <other args>
174 * Where 'other args' are the userspace implementation-specific log
175 * arguments.
176 *
177 * Example:
178 * <UUID> [integrated_flush] clustered-disk <arg count> <log dev>
179 * <region_size> [[no]sync]
180 *
181 * This module strips off the <UUID> and uses it for identification
182 * purposes when communicating with userspace about a log.
183 *
184 * If integrated_flush is defined, the kernel combines flush
185 * and mark requests.
186 *
187 * The rest of the line, beginning with 'clustered-disk', is passed
188 * to the userspace ctr function.
189 */
190static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
191 unsigned argc, char **argv)
192{
193 int r = 0;
194 int str_size;
195 char *ctr_str = NULL;
196 struct log_c *lc = NULL;
197 uint64_t rdata;
198 size_t rdata_size = sizeof(rdata);
199 char *devices_rdata = NULL;
200 size_t devices_rdata_size = DM_NAME_LEN;
201
202 if (argc < 3) {
203 DMWARN("Too few arguments to userspace dirty log");
204 return -EINVAL;
205 }
206
207 lc = kzalloc(sizeof(*lc), GFP_KERNEL);
208 if (!lc) {
209 DMWARN("Unable to allocate userspace log context.");
210 return -ENOMEM;
211 }
212
213 /* The ptr value is sufficient for local unique id */
214 lc->luid = (unsigned long)lc;
215
216 lc->ti = ti;
217
218 if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
219 DMWARN("UUID argument too long.");
220 kfree(lc);
221 return -EINVAL;
222 }
223
224 lc->usr_argc = argc;
225
226 strncpy(lc->uuid, argv[0], DM_UUID_LEN);
227 argc--;
228 argv++;
229 spin_lock_init(&lc->flush_lock);
230 INIT_LIST_HEAD(&lc->mark_list);
231 INIT_LIST_HEAD(&lc->clear_list);
232
233 if (!strcasecmp(argv[0], "integrated_flush")) {
234 lc->integrated_flush = 1;
235 argc--;
236 argv++;
237 }
238
239 str_size = build_constructor_string(ti, argc, argv, &ctr_str);
240 if (str_size < 0) {
241 kfree(lc);
242 return str_size;
243 }
244
245 devices_rdata = kzalloc(devices_rdata_size, GFP_KERNEL);
246 if (!devices_rdata) {
247 DMERR("Failed to allocate memory for device information");
248 r = -ENOMEM;
249 goto out;
250 }
251
252 lc->flush_entry_pool = mempool_create_slab_pool(FLUSH_ENTRY_POOL_SIZE,
253 _flush_entry_cache);
254 if (!lc->flush_entry_pool) {
255 DMERR("Failed to create flush_entry_pool");
256 r = -ENOMEM;
257 goto out;
258 }
259
260 /*
261 * Send table string and get back any opened device.
262 */
263 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
264 ctr_str, str_size,
265 devices_rdata, &devices_rdata_size);
266
267 if (r < 0) {
268 if (r == -ESRCH)
269 DMERR("Userspace log server not found");
270 else
271 DMERR("Userspace log server failed to create log");
272 goto out;
273 }
274
275 /* Since the region size does not change, get it now */
276 rdata_size = sizeof(rdata);
277 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
278 NULL, 0, (char *)&rdata, &rdata_size);
279
280 if (r) {
281 DMERR("Failed to get region size of dirty log");
282 goto out;
283 }
284
285 lc->region_size = (uint32_t)rdata;
286 lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
287
288 if (devices_rdata_size) {
289 if (devices_rdata[devices_rdata_size - 1] != '\0') {
290 DMERR("DM_ULOG_CTR device return string not properly terminated");
291 r = -EINVAL;
292 goto out;
293 }
294 r = dm_get_device(ti, devices_rdata,
295 dm_table_get_mode(ti->table), &lc->log_dev);
296 if (r)
297 DMERR("Failed to register %s with device-mapper",
298 devices_rdata);
299 }
300
301 if (lc->integrated_flush) {
302 lc->dmlog_wq = alloc_workqueue("dmlogd", WQ_MEM_RECLAIM, 0);
303 if (!lc->dmlog_wq) {
304 DMERR("couldn't start dmlogd");
305 r = -ENOMEM;
306 goto out;
307 }
308
309 INIT_DELAYED_WORK(&lc->flush_log_work, do_flush);
310 atomic_set(&lc->sched_flush, 0);
311 }
312
313out:
314 kfree(devices_rdata);
315 if (r) {
316 mempool_destroy(lc->flush_entry_pool);
317 kfree(lc);
318 kfree(ctr_str);
319 } else {
320 lc->usr_argv_str = ctr_str;
321 log->context = lc;
322 }
323
324 return r;
325}
326
327static void userspace_dtr(struct dm_dirty_log *log)
328{
329 struct log_c *lc = log->context;
330
331 if (lc->integrated_flush) {
332 /* flush workqueue */
333 if (atomic_read(&lc->sched_flush))
334 flush_delayed_work(&lc->flush_log_work);
335
336 destroy_workqueue(lc->dmlog_wq);
337 }
338
339 (void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
340 NULL, 0, NULL, NULL);
341
342 if (lc->log_dev)
343 dm_put_device(lc->ti, lc->log_dev);
344
345 mempool_destroy(lc->flush_entry_pool);
346
347 kfree(lc->usr_argv_str);
348 kfree(lc);
349
350 return;
351}
352
353static 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
364static 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
381static 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
393static 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 */
408static 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 *)®ion64, 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 */
435static 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 *)®ion64, sizeof(region64),
463 (char *)&in_sync, &rdata_size);
464 return (r) ? 0 : (int)in_sync;
465}
466
467static 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
484static 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 */
563static 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
622out:
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 */
649static 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 return;
666}
667
668/*
669 * userspace_clear_region
670 *
671 * This function must not block.
672 * So, the alloc can't block. In the worst case, it is ok to
673 * fail. It would simply mean we can't clear the region.
674 * Does nothing to current sync context, but does mean
675 * the region will be re-sync'ed on a reload of the mirror
676 * even though it is in-sync.
677 */
678static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
679{
680 unsigned long flags;
681 struct log_c *lc = log->context;
682 struct dm_dirty_log_flush_entry *fe;
683
684 /*
685 * If we fail to allocate, we skip the clearing of
686 * the region. This doesn't hurt us in any way, except
687 * to cause the region to be resync'ed when the
688 * device is activated next time.
689 */
690 fe = mempool_alloc(lc->flush_entry_pool, GFP_ATOMIC);
691 if (!fe) {
692 DMERR("Failed to allocate memory to clear region.");
693 return;
694 }
695
696 spin_lock_irqsave(&lc->flush_lock, flags);
697 fe->type = DM_ULOG_CLEAR_REGION;
698 fe->region = region;
699 list_add(&fe->list, &lc->clear_list);
700 spin_unlock_irqrestore(&lc->flush_lock, flags);
701
702 return;
703}
704
705/*
706 * userspace_get_resync_work
707 *
708 * Get a region that needs recovery. It is valid to return
709 * an error for this function.
710 *
711 * Returns: 1 if region filled, 0 if no work, <0 on error
712 */
713static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
714{
715 int r;
716 size_t rdata_size;
717 struct log_c *lc = log->context;
718 struct {
719 int64_t i; /* 64-bit for mix arch compatibility */
720 region_t r;
721 } pkg;
722
723 if (lc->in_sync_hint >= lc->region_count)
724 return 0;
725
726 rdata_size = sizeof(pkg);
727 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
728 NULL, 0, (char *)&pkg, &rdata_size);
729
730 *region = pkg.r;
731 return (r) ? r : (int)pkg.i;
732}
733
734/*
735 * userspace_set_region_sync
736 *
737 * Set the sync status of a given region. This function
738 * must not fail.
739 */
740static void userspace_set_region_sync(struct dm_dirty_log *log,
741 region_t region, int in_sync)
742{
743 struct log_c *lc = log->context;
744 struct {
745 region_t r;
746 int64_t i;
747 } pkg;
748
749 pkg.r = region;
750 pkg.i = (int64_t)in_sync;
751
752 (void) userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
753 (char *)&pkg, sizeof(pkg), NULL, NULL);
754
755 /*
756 * It would be nice to be able to report failures.
757 * However, it is easy enough to detect and resolve.
758 */
759 return;
760}
761
762/*
763 * userspace_get_sync_count
764 *
765 * If there is any sort of failure when consulting the server,
766 * we assume that the sync count is zero.
767 *
768 * Returns: sync count on success, 0 on failure
769 */
770static region_t userspace_get_sync_count(struct dm_dirty_log *log)
771{
772 int r;
773 size_t rdata_size;
774 uint64_t sync_count;
775 struct log_c *lc = log->context;
776
777 rdata_size = sizeof(sync_count);
778 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
779 NULL, 0, (char *)&sync_count, &rdata_size);
780
781 if (r)
782 return 0;
783
784 if (sync_count >= lc->region_count)
785 lc->in_sync_hint = lc->region_count;
786
787 return (region_t)sync_count;
788}
789
790/*
791 * userspace_status
792 *
793 * Returns: amount of space consumed
794 */
795static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
796 char *result, unsigned maxlen)
797{
798 int r = 0;
799 char *table_args;
800 size_t sz = (size_t)maxlen;
801 struct log_c *lc = log->context;
802
803 switch (status_type) {
804 case STATUSTYPE_INFO:
805 r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
806 NULL, 0, result, &sz);
807
808 if (r) {
809 sz = 0;
810 DMEMIT("%s 1 COM_FAILURE", log->type->name);
811 }
812 break;
813 case STATUSTYPE_TABLE:
814 sz = 0;
815 table_args = strchr(lc->usr_argv_str, ' ');
816 BUG_ON(!table_args); /* There will always be a ' ' */
817 table_args++;
818
819 DMEMIT("%s %u %s ", log->type->name, lc->usr_argc, lc->uuid);
820 if (lc->integrated_flush)
821 DMEMIT("integrated_flush ");
822 DMEMIT("%s ", table_args);
823 break;
824 }
825 return (r) ? 0 : (int)sz;
826}
827
828/*
829 * userspace_is_remote_recovering
830 *
831 * Returns: 1 if region recovering, 0 otherwise
832 */
833static int userspace_is_remote_recovering(struct dm_dirty_log *log,
834 region_t region)
835{
836 int r;
837 uint64_t region64 = region;
838 struct log_c *lc = log->context;
839 static unsigned long limit;
840 struct {
841 int64_t is_recovering;
842 uint64_t in_sync_hint;
843 } pkg;
844 size_t rdata_size = sizeof(pkg);
845
846 /*
847 * Once the mirror has been reported to be in-sync,
848 * it will never again ask for recovery work. So,
849 * we can safely say there is not a remote machine
850 * recovering if the device is in-sync. (in_sync_hint
851 * must be reset at resume time.)
852 */
853 if (region < lc->in_sync_hint)
854 return 0;
855 else if (time_after(limit, jiffies))
856 return 1;
857
858 limit = jiffies + (HZ / 4);
859 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
860 (char *)®ion64, sizeof(region64),
861 (char *)&pkg, &rdata_size);
862 if (r)
863 return 1;
864
865 lc->in_sync_hint = pkg.in_sync_hint;
866
867 return (int)pkg.is_recovering;
868}
869
870static struct dm_dirty_log_type _userspace_type = {
871 .name = "userspace",
872 .module = THIS_MODULE,
873 .ctr = userspace_ctr,
874 .dtr = userspace_dtr,
875 .presuspend = userspace_presuspend,
876 .postsuspend = userspace_postsuspend,
877 .resume = userspace_resume,
878 .get_region_size = userspace_get_region_size,
879 .is_clean = userspace_is_clean,
880 .in_sync = userspace_in_sync,
881 .flush = userspace_flush,
882 .mark_region = userspace_mark_region,
883 .clear_region = userspace_clear_region,
884 .get_resync_work = userspace_get_resync_work,
885 .set_region_sync = userspace_set_region_sync,
886 .get_sync_count = userspace_get_sync_count,
887 .status = userspace_status,
888 .is_remote_recovering = userspace_is_remote_recovering,
889};
890
891static int __init userspace_dirty_log_init(void)
892{
893 int r = 0;
894
895 _flush_entry_cache = KMEM_CACHE(dm_dirty_log_flush_entry, 0);
896 if (!_flush_entry_cache) {
897 DMWARN("Unable to create flush_entry_cache: No memory.");
898 return -ENOMEM;
899 }
900
901 r = dm_ulog_tfr_init();
902 if (r) {
903 DMWARN("Unable to initialize userspace log communications");
904 kmem_cache_destroy(_flush_entry_cache);
905 return r;
906 }
907
908 r = dm_dirty_log_type_register(&_userspace_type);
909 if (r) {
910 DMWARN("Couldn't register userspace dirty log type");
911 dm_ulog_tfr_exit();
912 kmem_cache_destroy(_flush_entry_cache);
913 return r;
914 }
915
916 DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
917 return 0;
918}
919
920static void __exit userspace_dirty_log_exit(void)
921{
922 dm_dirty_log_type_unregister(&_userspace_type);
923 dm_ulog_tfr_exit();
924 kmem_cache_destroy(_flush_entry_cache);
925
926 DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
927 return;
928}
929
930module_init(userspace_dirty_log_init);
931module_exit(userspace_dirty_log_exit);
932
933MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
934MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>");
935MODULE_LICENSE("GPL");