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/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 r = mempool_init_slab_pool(&lc->flush_entry_pool, FLUSH_ENTRY_POOL_SIZE,
253 _flush_entry_cache);
254 if (r) {
255 DMERR("Failed to create flush_entry_pool");
256 goto out;
257 }
258
259 /*
260 * Send table string and get back any opened device.
261 */
262 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
263 ctr_str, str_size,
264 devices_rdata, &devices_rdata_size);
265
266 if (r < 0) {
267 if (r == -ESRCH)
268 DMERR("Userspace log server not found");
269 else
270 DMERR("Userspace log server failed to create log");
271 goto out;
272 }
273
274 /* Since the region size does not change, get it now */
275 rdata_size = sizeof(rdata);
276 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
277 NULL, 0, (char *)&rdata, &rdata_size);
278
279 if (r) {
280 DMERR("Failed to get region size of dirty log");
281 goto out;
282 }
283
284 lc->region_size = (uint32_t)rdata;
285 lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
286
287 if (devices_rdata_size) {
288 if (devices_rdata[devices_rdata_size - 1] != '\0') {
289 DMERR("DM_ULOG_CTR device return string not properly terminated");
290 r = -EINVAL;
291 goto out;
292 }
293 r = dm_get_device(ti, devices_rdata,
294 dm_table_get_mode(ti->table), &lc->log_dev);
295 if (r)
296 DMERR("Failed to register %s with device-mapper",
297 devices_rdata);
298 }
299
300 if (lc->integrated_flush) {
301 lc->dmlog_wq = alloc_workqueue("dmlogd", WQ_MEM_RECLAIM, 0);
302 if (!lc->dmlog_wq) {
303 DMERR("couldn't start dmlogd");
304 r = -ENOMEM;
305 goto out;
306 }
307
308 INIT_DELAYED_WORK(&lc->flush_log_work, do_flush);
309 atomic_set(&lc->sched_flush, 0);
310 }
311
312out:
313 kfree(devices_rdata);
314 if (r) {
315 mempool_exit(&lc->flush_entry_pool);
316 kfree(lc);
317 kfree(ctr_str);
318 } else {
319 lc->usr_argv_str = ctr_str;
320 log->context = lc;
321 }
322
323 return r;
324}
325
326static void userspace_dtr(struct dm_dirty_log *log)
327{
328 struct log_c *lc = log->context;
329
330 if (lc->integrated_flush) {
331 /* flush workqueue */
332 if (atomic_read(&lc->sched_flush))
333 flush_delayed_work(&lc->flush_log_work);
334
335 destroy_workqueue(lc->dmlog_wq);
336 }
337
338 (void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
339 NULL, 0, NULL, NULL);
340
341 if (lc->log_dev)
342 dm_put_device(lc->ti, lc->log_dev);
343
344 mempool_exit(&lc->flush_entry_pool);
345
346 kfree(lc->usr_argv_str);
347 kfree(lc);
348
349 return;
350}
351
352static int userspace_presuspend(struct dm_dirty_log *log)
353{
354 int r;
355 struct log_c *lc = log->context;
356
357 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
358 NULL, 0, NULL, NULL);
359
360 return r;
361}
362
363static int userspace_postsuspend(struct dm_dirty_log *log)
364{
365 int r;
366 struct log_c *lc = log->context;
367
368 /*
369 * Run planned flush earlier.
370 */
371 if (lc->integrated_flush && atomic_read(&lc->sched_flush))
372 flush_delayed_work(&lc->flush_log_work);
373
374 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
375 NULL, 0, NULL, NULL);
376
377 return r;
378}
379
380static int userspace_resume(struct dm_dirty_log *log)
381{
382 int r;
383 struct log_c *lc = log->context;
384
385 lc->in_sync_hint = 0;
386 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
387 NULL, 0, NULL, NULL);
388
389 return r;
390}
391
392static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
393{
394 struct log_c *lc = log->context;
395
396 return lc->region_size;
397}
398
399/*
400 * userspace_is_clean
401 *
402 * Check whether a region is clean. If there is any sort of
403 * failure when consulting the server, we return not clean.
404 *
405 * Returns: 1 if clean, 0 otherwise
406 */
407static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
408{
409 int r;
410 uint64_t region64 = (uint64_t)region;
411 int64_t is_clean;
412 size_t rdata_size;
413 struct log_c *lc = log->context;
414
415 rdata_size = sizeof(is_clean);
416 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
417 (char *)®ion64, sizeof(region64),
418 (char *)&is_clean, &rdata_size);
419
420 return (r) ? 0 : (int)is_clean;
421}
422
423/*
424 * userspace_in_sync
425 *
426 * Check if the region is in-sync. If there is any sort
427 * of failure when consulting the server, we assume that
428 * the region is not in sync.
429 *
430 * If 'can_block' is set, return immediately
431 *
432 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
433 */
434static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
435 int can_block)
436{
437 int r;
438 uint64_t region64 = region;
439 int64_t in_sync;
440 size_t rdata_size;
441 struct log_c *lc = log->context;
442
443 /*
444 * We can never respond directly - even if in_sync_hint is
445 * set. This is because another machine could see a device
446 * failure and mark the region out-of-sync. If we don't go
447 * to userspace to ask, we might think the region is in-sync
448 * and allow a read to pick up data that is stale. (This is
449 * very unlikely if a device actually fails; but it is very
450 * likely if a connection to one device from one machine fails.)
451 *
452 * There still might be a problem if the mirror caches the region
453 * state as in-sync... but then this call would not be made. So,
454 * that is a mirror problem.
455 */
456 if (!can_block)
457 return -EWOULDBLOCK;
458
459 rdata_size = sizeof(in_sync);
460 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
461 (char *)®ion64, sizeof(region64),
462 (char *)&in_sync, &rdata_size);
463 return (r) ? 0 : (int)in_sync;
464}
465
466static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
467{
468 int r = 0;
469 struct dm_dirty_log_flush_entry *fe;
470
471 list_for_each_entry(fe, flush_list, list) {
472 r = userspace_do_request(lc, lc->uuid, fe->type,
473 (char *)&fe->region,
474 sizeof(fe->region),
475 NULL, NULL);
476 if (r)
477 break;
478 }
479
480 return r;
481}
482
483static int flush_by_group(struct log_c *lc, struct list_head *flush_list,
484 int flush_with_payload)
485{
486 int r = 0;
487 int count;
488 uint32_t type = 0;
489 struct dm_dirty_log_flush_entry *fe, *tmp_fe;
490 LIST_HEAD(tmp_list);
491 uint64_t group[MAX_FLUSH_GROUP_COUNT];
492
493 /*
494 * Group process the requests
495 */
496 while (!list_empty(flush_list)) {
497 count = 0;
498
499 list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
500 group[count] = fe->region;
501 count++;
502
503 list_move(&fe->list, &tmp_list);
504
505 type = fe->type;
506 if (count >= MAX_FLUSH_GROUP_COUNT)
507 break;
508 }
509
510 if (flush_with_payload) {
511 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
512 (char *)(group),
513 count * sizeof(uint64_t),
514 NULL, NULL);
515 /*
516 * Integrated flush failed.
517 */
518 if (r)
519 break;
520 } else {
521 r = userspace_do_request(lc, lc->uuid, type,
522 (char *)(group),
523 count * sizeof(uint64_t),
524 NULL, NULL);
525 if (r) {
526 /*
527 * Group send failed. Attempt one-by-one.
528 */
529 list_splice_init(&tmp_list, flush_list);
530 r = flush_one_by_one(lc, flush_list);
531 break;
532 }
533 }
534 }
535
536 /*
537 * Must collect flush_entrys that were successfully processed
538 * as a group so that they will be free'd by the caller.
539 */
540 list_splice_init(&tmp_list, flush_list);
541
542 return r;
543}
544
545/*
546 * userspace_flush
547 *
548 * This function is ok to block.
549 * The flush happens in two stages. First, it sends all
550 * clear/mark requests that are on the list. Then it
551 * tells the server to commit them. This gives the
552 * server a chance to optimise the commit, instead of
553 * doing it for every request.
554 *
555 * Additionally, we could implement another thread that
556 * sends the requests up to the server - reducing the
557 * load on flush. Then the flush would have less in
558 * the list and be responsible for the finishing commit.
559 *
560 * Returns: 0 on success, < 0 on failure
561 */
562static int userspace_flush(struct dm_dirty_log *log)
563{
564 int r = 0;
565 unsigned long flags;
566 struct log_c *lc = log->context;
567 LIST_HEAD(mark_list);
568 LIST_HEAD(clear_list);
569 int mark_list_is_empty;
570 int clear_list_is_empty;
571 struct dm_dirty_log_flush_entry *fe, *tmp_fe;
572 mempool_t *flush_entry_pool = &lc->flush_entry_pool;
573
574 spin_lock_irqsave(&lc->flush_lock, flags);
575 list_splice_init(&lc->mark_list, &mark_list);
576 list_splice_init(&lc->clear_list, &clear_list);
577 spin_unlock_irqrestore(&lc->flush_lock, flags);
578
579 mark_list_is_empty = list_empty(&mark_list);
580 clear_list_is_empty = list_empty(&clear_list);
581
582 if (mark_list_is_empty && clear_list_is_empty)
583 return 0;
584
585 r = flush_by_group(lc, &clear_list, 0);
586 if (r)
587 goto out;
588
589 if (!lc->integrated_flush) {
590 r = flush_by_group(lc, &mark_list, 0);
591 if (r)
592 goto out;
593 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
594 NULL, 0, NULL, NULL);
595 goto out;
596 }
597
598 /*
599 * Send integrated flush request with mark_list as payload.
600 */
601 r = flush_by_group(lc, &mark_list, 1);
602 if (r)
603 goto out;
604
605 if (mark_list_is_empty && !atomic_read(&lc->sched_flush)) {
606 /*
607 * When there are only clear region requests,
608 * we schedule a flush in the future.
609 */
610 queue_delayed_work(lc->dmlog_wq, &lc->flush_log_work, 3 * HZ);
611 atomic_set(&lc->sched_flush, 1);
612 } else {
613 /*
614 * Cancel pending flush because we
615 * have already flushed in mark_region.
616 */
617 cancel_delayed_work(&lc->flush_log_work);
618 atomic_set(&lc->sched_flush, 0);
619 }
620
621out:
622 /*
623 * We can safely remove these entries, even after failure.
624 * Calling code will receive an error and will know that
625 * the log facility has failed.
626 */
627 list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) {
628 list_del(&fe->list);
629 mempool_free(fe, flush_entry_pool);
630 }
631 list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) {
632 list_del(&fe->list);
633 mempool_free(fe, flush_entry_pool);
634 }
635
636 if (r)
637 dm_table_event(lc->ti->table);
638
639 return r;
640}
641
642/*
643 * userspace_mark_region
644 *
645 * This function should avoid blocking unless absolutely required.
646 * (Memory allocation is valid for blocking.)
647 */
648static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
649{
650 unsigned long flags;
651 struct log_c *lc = log->context;
652 struct dm_dirty_log_flush_entry *fe;
653
654 /* Wait for an allocation, but _never_ fail */
655 fe = mempool_alloc(&lc->flush_entry_pool, GFP_NOIO);
656 BUG_ON(!fe);
657
658 spin_lock_irqsave(&lc->flush_lock, flags);
659 fe->type = DM_ULOG_MARK_REGION;
660 fe->region = region;
661 list_add(&fe->list, &lc->mark_list);
662 spin_unlock_irqrestore(&lc->flush_lock, flags);
663
664 return;
665}
666
667/*
668 * userspace_clear_region
669 *
670 * This function must not block.
671 * So, the alloc can't block. In the worst case, it is ok to
672 * fail. It would simply mean we can't clear the region.
673 * Does nothing to current sync context, but does mean
674 * the region will be re-sync'ed on a reload of the mirror
675 * even though it is in-sync.
676 */
677static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
678{
679 unsigned long flags;
680 struct log_c *lc = log->context;
681 struct dm_dirty_log_flush_entry *fe;
682
683 /*
684 * If we fail to allocate, we skip the clearing of
685 * the region. This doesn't hurt us in any way, except
686 * to cause the region to be resync'ed when the
687 * device is activated next time.
688 */
689 fe = mempool_alloc(&lc->flush_entry_pool, GFP_ATOMIC);
690 if (!fe) {
691 DMERR("Failed to allocate memory to clear region.");
692 return;
693 }
694
695 spin_lock_irqsave(&lc->flush_lock, flags);
696 fe->type = DM_ULOG_CLEAR_REGION;
697 fe->region = region;
698 list_add(&fe->list, &lc->clear_list);
699 spin_unlock_irqrestore(&lc->flush_lock, flags);
700
701 return;
702}
703
704/*
705 * userspace_get_resync_work
706 *
707 * Get a region that needs recovery. It is valid to return
708 * an error for this function.
709 *
710 * Returns: 1 if region filled, 0 if no work, <0 on error
711 */
712static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
713{
714 int r;
715 size_t rdata_size;
716 struct log_c *lc = log->context;
717 struct {
718 int64_t i; /* 64-bit for mix arch compatibility */
719 region_t r;
720 } pkg;
721
722 if (lc->in_sync_hint >= lc->region_count)
723 return 0;
724
725 rdata_size = sizeof(pkg);
726 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
727 NULL, 0, (char *)&pkg, &rdata_size);
728
729 *region = pkg.r;
730 return (r) ? r : (int)pkg.i;
731}
732
733/*
734 * userspace_set_region_sync
735 *
736 * Set the sync status of a given region. This function
737 * must not fail.
738 */
739static void userspace_set_region_sync(struct dm_dirty_log *log,
740 region_t region, int in_sync)
741{
742 struct log_c *lc = log->context;
743 struct {
744 region_t r;
745 int64_t i;
746 } pkg;
747
748 pkg.r = region;
749 pkg.i = (int64_t)in_sync;
750
751 (void) userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
752 (char *)&pkg, sizeof(pkg), NULL, NULL);
753
754 /*
755 * It would be nice to be able to report failures.
756 * However, it is easy enough to detect and resolve.
757 */
758 return;
759}
760
761/*
762 * userspace_get_sync_count
763 *
764 * If there is any sort of failure when consulting the server,
765 * we assume that the sync count is zero.
766 *
767 * Returns: sync count on success, 0 on failure
768 */
769static region_t userspace_get_sync_count(struct dm_dirty_log *log)
770{
771 int r;
772 size_t rdata_size;
773 uint64_t sync_count;
774 struct log_c *lc = log->context;
775
776 rdata_size = sizeof(sync_count);
777 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
778 NULL, 0, (char *)&sync_count, &rdata_size);
779
780 if (r)
781 return 0;
782
783 if (sync_count >= lc->region_count)
784 lc->in_sync_hint = lc->region_count;
785
786 return (region_t)sync_count;
787}
788
789/*
790 * userspace_status
791 *
792 * Returns: amount of space consumed
793 */
794static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
795 char *result, unsigned maxlen)
796{
797 int r = 0;
798 char *table_args;
799 size_t sz = (size_t)maxlen;
800 struct log_c *lc = log->context;
801
802 switch (status_type) {
803 case STATUSTYPE_INFO:
804 r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
805 NULL, 0, result, &sz);
806
807 if (r) {
808 sz = 0;
809 DMEMIT("%s 1 COM_FAILURE", log->type->name);
810 }
811 break;
812 case STATUSTYPE_TABLE:
813 sz = 0;
814 table_args = strchr(lc->usr_argv_str, ' ');
815 BUG_ON(!table_args); /* There will always be a ' ' */
816 table_args++;
817
818 DMEMIT("%s %u %s ", log->type->name, lc->usr_argc, lc->uuid);
819 if (lc->integrated_flush)
820 DMEMIT("integrated_flush ");
821 DMEMIT("%s ", table_args);
822 break;
823 case STATUSTYPE_IMA:
824 *result = '\0';
825 break;
826 }
827 return (r) ? 0 : (int)sz;
828}
829
830/*
831 * userspace_is_remote_recovering
832 *
833 * Returns: 1 if region recovering, 0 otherwise
834 */
835static int userspace_is_remote_recovering(struct dm_dirty_log *log,
836 region_t region)
837{
838 int r;
839 uint64_t region64 = region;
840 struct log_c *lc = log->context;
841 static unsigned long limit;
842 struct {
843 int64_t is_recovering;
844 uint64_t in_sync_hint;
845 } pkg;
846 size_t rdata_size = sizeof(pkg);
847
848 /*
849 * Once the mirror has been reported to be in-sync,
850 * it will never again ask for recovery work. So,
851 * we can safely say there is not a remote machine
852 * recovering if the device is in-sync. (in_sync_hint
853 * must be reset at resume time.)
854 */
855 if (region < lc->in_sync_hint)
856 return 0;
857 else if (time_after(limit, jiffies))
858 return 1;
859
860 limit = jiffies + (HZ / 4);
861 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
862 (char *)®ion64, sizeof(region64),
863 (char *)&pkg, &rdata_size);
864 if (r)
865 return 1;
866
867 lc->in_sync_hint = pkg.in_sync_hint;
868
869 return (int)pkg.is_recovering;
870}
871
872static struct dm_dirty_log_type _userspace_type = {
873 .name = "userspace",
874 .module = THIS_MODULE,
875 .ctr = userspace_ctr,
876 .dtr = userspace_dtr,
877 .presuspend = userspace_presuspend,
878 .postsuspend = userspace_postsuspend,
879 .resume = userspace_resume,
880 .get_region_size = userspace_get_region_size,
881 .is_clean = userspace_is_clean,
882 .in_sync = userspace_in_sync,
883 .flush = userspace_flush,
884 .mark_region = userspace_mark_region,
885 .clear_region = userspace_clear_region,
886 .get_resync_work = userspace_get_resync_work,
887 .set_region_sync = userspace_set_region_sync,
888 .get_sync_count = userspace_get_sync_count,
889 .status = userspace_status,
890 .is_remote_recovering = userspace_is_remote_recovering,
891};
892
893static int __init userspace_dirty_log_init(void)
894{
895 int r = 0;
896
897 _flush_entry_cache = KMEM_CACHE(dm_dirty_log_flush_entry, 0);
898 if (!_flush_entry_cache) {
899 DMWARN("Unable to create flush_entry_cache: No memory.");
900 return -ENOMEM;
901 }
902
903 r = dm_ulog_tfr_init();
904 if (r) {
905 DMWARN("Unable to initialize userspace log communications");
906 kmem_cache_destroy(_flush_entry_cache);
907 return r;
908 }
909
910 r = dm_dirty_log_type_register(&_userspace_type);
911 if (r) {
912 DMWARN("Couldn't register userspace dirty log type");
913 dm_ulog_tfr_exit();
914 kmem_cache_destroy(_flush_entry_cache);
915 return r;
916 }
917
918 DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
919 return 0;
920}
921
922static void __exit userspace_dirty_log_exit(void)
923{
924 dm_dirty_log_type_unregister(&_userspace_type);
925 dm_ulog_tfr_exit();
926 kmem_cache_destroy(_flush_entry_cache);
927
928 DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
929 return;
930}
931
932module_init(userspace_dirty_log_init);
933module_exit(userspace_dirty_log_exit);
934
935MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
936MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>");
937MODULE_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/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");