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
14#include "dm-log-userspace-transfer.h"
15
16#define DM_LOG_USERSPACE_VSN "1.1.0"
17
18struct flush_entry {
19 int type;
20 region_t region;
21 struct list_head list;
22};
23
24/*
25 * This limit on the number of mark and clear request is, to a degree,
26 * arbitrary. However, there is some basis for the choice in the limits
27 * imposed on the size of data payload by dm-log-userspace-transfer.c:
28 * dm_consult_userspace().
29 */
30#define MAX_FLUSH_GROUP_COUNT 32
31
32struct log_c {
33 struct dm_target *ti;
34 struct dm_dev *log_dev;
35 uint32_t region_size;
36 region_t region_count;
37 uint64_t luid;
38 char uuid[DM_UUID_LEN];
39
40 char *usr_argv_str;
41 uint32_t usr_argc;
42
43 /*
44 * in_sync_hint gets set when doing is_remote_recovering. It
45 * represents the first region that needs recovery. IOW, the
46 * first zero bit of sync_bits. This can be useful for to limit
47 * traffic for calls like is_remote_recovering and get_resync_work,
48 * but be take care in its use for anything else.
49 */
50 uint64_t in_sync_hint;
51
52 /*
53 * Mark and clear requests are held until a flush is issued
54 * so that we can group, and thereby limit, the amount of
55 * network traffic between kernel and userspace. The 'flush_lock'
56 * is used to protect these lists.
57 */
58 spinlock_t flush_lock;
59 struct list_head mark_list;
60 struct list_head clear_list;
61};
62
63static mempool_t *flush_entry_pool;
64
65static void *flush_entry_alloc(gfp_t gfp_mask, void *pool_data)
66{
67 return kmalloc(sizeof(struct flush_entry), gfp_mask);
68}
69
70static void flush_entry_free(void *element, void *pool_data)
71{
72 kfree(element);
73}
74
75static int userspace_do_request(struct log_c *lc, const char *uuid,
76 int request_type, char *data, size_t data_size,
77 char *rdata, size_t *rdata_size)
78{
79 int r;
80
81 /*
82 * If the server isn't there, -ESRCH is returned,
83 * and we must keep trying until the server is
84 * restored.
85 */
86retry:
87 r = dm_consult_userspace(uuid, lc->luid, request_type, data,
88 data_size, rdata, rdata_size);
89
90 if (r != -ESRCH)
91 return r;
92
93 DMERR(" Userspace log server not found.");
94 while (1) {
95 set_current_state(TASK_INTERRUPTIBLE);
96 schedule_timeout(2*HZ);
97 DMWARN("Attempting to contact userspace log server...");
98 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR,
99 lc->usr_argv_str,
100 strlen(lc->usr_argv_str) + 1,
101 NULL, NULL);
102 if (!r)
103 break;
104 }
105 DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
106 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL,
107 0, NULL, NULL);
108 if (!r)
109 goto retry;
110
111 DMERR("Error trying to resume userspace log: %d", r);
112
113 return -ESRCH;
114}
115
116static int build_constructor_string(struct dm_target *ti,
117 unsigned argc, char **argv,
118 char **ctr_str)
119{
120 int i, str_size;
121 char *str = NULL;
122
123 *ctr_str = NULL;
124
125 for (i = 0, str_size = 0; i < argc; i++)
126 str_size += strlen(argv[i]) + 1; /* +1 for space between args */
127
128 str_size += 20; /* Max number of chars in a printed u64 number */
129
130 str = kzalloc(str_size, GFP_KERNEL);
131 if (!str) {
132 DMWARN("Unable to allocate memory for constructor string");
133 return -ENOMEM;
134 }
135
136 str_size = sprintf(str, "%llu", (unsigned long long)ti->len);
137 for (i = 0; i < argc; i++)
138 str_size += sprintf(str + str_size, " %s", argv[i]);
139
140 *ctr_str = str;
141 return str_size;
142}
143
144/*
145 * userspace_ctr
146 *
147 * argv contains:
148 * <UUID> <other args>
149 * Where 'other args' is the userspace implementation specific log
150 * arguments. An example might be:
151 * <UUID> clustered-disk <arg count> <log dev> <region_size> [[no]sync]
152 *
153 * So, this module will strip off the <UUID> for identification purposes
154 * when communicating with userspace about a log; but will pass on everything
155 * else.
156 */
157static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
158 unsigned argc, char **argv)
159{
160 int r = 0;
161 int str_size;
162 char *ctr_str = NULL;
163 struct log_c *lc = NULL;
164 uint64_t rdata;
165 size_t rdata_size = sizeof(rdata);
166 char *devices_rdata = NULL;
167 size_t devices_rdata_size = DM_NAME_LEN;
168
169 if (argc < 3) {
170 DMWARN("Too few arguments to userspace dirty log");
171 return -EINVAL;
172 }
173
174 lc = kzalloc(sizeof(*lc), GFP_KERNEL);
175 if (!lc) {
176 DMWARN("Unable to allocate userspace log context.");
177 return -ENOMEM;
178 }
179
180 /* The ptr value is sufficient for local unique id */
181 lc->luid = (unsigned long)lc;
182
183 lc->ti = ti;
184
185 if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
186 DMWARN("UUID argument too long.");
187 kfree(lc);
188 return -EINVAL;
189 }
190
191 strncpy(lc->uuid, argv[0], DM_UUID_LEN);
192 spin_lock_init(&lc->flush_lock);
193 INIT_LIST_HEAD(&lc->mark_list);
194 INIT_LIST_HEAD(&lc->clear_list);
195
196 str_size = build_constructor_string(ti, argc - 1, argv + 1, &ctr_str);
197 if (str_size < 0) {
198 kfree(lc);
199 return str_size;
200 }
201
202 devices_rdata = kzalloc(devices_rdata_size, GFP_KERNEL);
203 if (!devices_rdata) {
204 DMERR("Failed to allocate memory for device information");
205 r = -ENOMEM;
206 goto out;
207 }
208
209 /*
210 * Send table string and get back any opened device.
211 */
212 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
213 ctr_str, str_size,
214 devices_rdata, &devices_rdata_size);
215
216 if (r < 0) {
217 if (r == -ESRCH)
218 DMERR("Userspace log server not found");
219 else
220 DMERR("Userspace log server failed to create log");
221 goto out;
222 }
223
224 /* Since the region size does not change, get it now */
225 rdata_size = sizeof(rdata);
226 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
227 NULL, 0, (char *)&rdata, &rdata_size);
228
229 if (r) {
230 DMERR("Failed to get region size of dirty log");
231 goto out;
232 }
233
234 lc->region_size = (uint32_t)rdata;
235 lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
236
237 if (devices_rdata_size) {
238 if (devices_rdata[devices_rdata_size - 1] != '\0') {
239 DMERR("DM_ULOG_CTR device return string not properly terminated");
240 r = -EINVAL;
241 goto out;
242 }
243 r = dm_get_device(ti, devices_rdata,
244 dm_table_get_mode(ti->table), &lc->log_dev);
245 if (r)
246 DMERR("Failed to register %s with device-mapper",
247 devices_rdata);
248 }
249out:
250 kfree(devices_rdata);
251 if (r) {
252 kfree(lc);
253 kfree(ctr_str);
254 } else {
255 lc->usr_argv_str = ctr_str;
256 lc->usr_argc = argc;
257 log->context = lc;
258 }
259
260 return r;
261}
262
263static void userspace_dtr(struct dm_dirty_log *log)
264{
265 struct log_c *lc = log->context;
266
267 (void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
268 NULL, 0,
269 NULL, NULL);
270
271 if (lc->log_dev)
272 dm_put_device(lc->ti, lc->log_dev);
273
274 kfree(lc->usr_argv_str);
275 kfree(lc);
276
277 return;
278}
279
280static int userspace_presuspend(struct dm_dirty_log *log)
281{
282 int r;
283 struct log_c *lc = log->context;
284
285 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
286 NULL, 0,
287 NULL, NULL);
288
289 return r;
290}
291
292static int userspace_postsuspend(struct dm_dirty_log *log)
293{
294 int r;
295 struct log_c *lc = log->context;
296
297 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
298 NULL, 0,
299 NULL, NULL);
300
301 return r;
302}
303
304static int userspace_resume(struct dm_dirty_log *log)
305{
306 int r;
307 struct log_c *lc = log->context;
308
309 lc->in_sync_hint = 0;
310 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
311 NULL, 0,
312 NULL, NULL);
313
314 return r;
315}
316
317static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
318{
319 struct log_c *lc = log->context;
320
321 return lc->region_size;
322}
323
324/*
325 * userspace_is_clean
326 *
327 * Check whether a region is clean. If there is any sort of
328 * failure when consulting the server, we return not clean.
329 *
330 * Returns: 1 if clean, 0 otherwise
331 */
332static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
333{
334 int r;
335 uint64_t region64 = (uint64_t)region;
336 int64_t is_clean;
337 size_t rdata_size;
338 struct log_c *lc = log->context;
339
340 rdata_size = sizeof(is_clean);
341 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
342 (char *)®ion64, sizeof(region64),
343 (char *)&is_clean, &rdata_size);
344
345 return (r) ? 0 : (int)is_clean;
346}
347
348/*
349 * userspace_in_sync
350 *
351 * Check if the region is in-sync. If there is any sort
352 * of failure when consulting the server, we assume that
353 * the region is not in sync.
354 *
355 * If 'can_block' is set, return immediately
356 *
357 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
358 */
359static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
360 int can_block)
361{
362 int r;
363 uint64_t region64 = region;
364 int64_t in_sync;
365 size_t rdata_size;
366 struct log_c *lc = log->context;
367
368 /*
369 * We can never respond directly - even if in_sync_hint is
370 * set. This is because another machine could see a device
371 * failure and mark the region out-of-sync. If we don't go
372 * to userspace to ask, we might think the region is in-sync
373 * and allow a read to pick up data that is stale. (This is
374 * very unlikely if a device actually fails; but it is very
375 * likely if a connection to one device from one machine fails.)
376 *
377 * There still might be a problem if the mirror caches the region
378 * state as in-sync... but then this call would not be made. So,
379 * that is a mirror problem.
380 */
381 if (!can_block)
382 return -EWOULDBLOCK;
383
384 rdata_size = sizeof(in_sync);
385 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
386 (char *)®ion64, sizeof(region64),
387 (char *)&in_sync, &rdata_size);
388 return (r) ? 0 : (int)in_sync;
389}
390
391static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
392{
393 int r = 0;
394 struct flush_entry *fe;
395
396 list_for_each_entry(fe, flush_list, list) {
397 r = userspace_do_request(lc, lc->uuid, fe->type,
398 (char *)&fe->region,
399 sizeof(fe->region),
400 NULL, NULL);
401 if (r)
402 break;
403 }
404
405 return r;
406}
407
408static int flush_by_group(struct log_c *lc, struct list_head *flush_list)
409{
410 int r = 0;
411 int count;
412 uint32_t type = 0;
413 struct flush_entry *fe, *tmp_fe;
414 LIST_HEAD(tmp_list);
415 uint64_t group[MAX_FLUSH_GROUP_COUNT];
416
417 /*
418 * Group process the requests
419 */
420 while (!list_empty(flush_list)) {
421 count = 0;
422
423 list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
424 group[count] = fe->region;
425 count++;
426
427 list_move(&fe->list, &tmp_list);
428
429 type = fe->type;
430 if (count >= MAX_FLUSH_GROUP_COUNT)
431 break;
432 }
433
434 r = userspace_do_request(lc, lc->uuid, type,
435 (char *)(group),
436 count * sizeof(uint64_t),
437 NULL, NULL);
438 if (r) {
439 /* Group send failed. Attempt one-by-one. */
440 list_splice_init(&tmp_list, flush_list);
441 r = flush_one_by_one(lc, flush_list);
442 break;
443 }
444 }
445
446 /*
447 * Must collect flush_entrys that were successfully processed
448 * as a group so that they will be free'd by the caller.
449 */
450 list_splice_init(&tmp_list, flush_list);
451
452 return r;
453}
454
455/*
456 * userspace_flush
457 *
458 * This function is ok to block.
459 * The flush happens in two stages. First, it sends all
460 * clear/mark requests that are on the list. Then it
461 * tells the server to commit them. This gives the
462 * server a chance to optimise the commit, instead of
463 * doing it for every request.
464 *
465 * Additionally, we could implement another thread that
466 * sends the requests up to the server - reducing the
467 * load on flush. Then the flush would have less in
468 * the list and be responsible for the finishing commit.
469 *
470 * Returns: 0 on success, < 0 on failure
471 */
472static int userspace_flush(struct dm_dirty_log *log)
473{
474 int r = 0;
475 unsigned long flags;
476 struct log_c *lc = log->context;
477 LIST_HEAD(mark_list);
478 LIST_HEAD(clear_list);
479 struct flush_entry *fe, *tmp_fe;
480
481 spin_lock_irqsave(&lc->flush_lock, flags);
482 list_splice_init(&lc->mark_list, &mark_list);
483 list_splice_init(&lc->clear_list, &clear_list);
484 spin_unlock_irqrestore(&lc->flush_lock, flags);
485
486 if (list_empty(&mark_list) && list_empty(&clear_list))
487 return 0;
488
489 r = flush_by_group(lc, &mark_list);
490 if (r)
491 goto fail;
492
493 r = flush_by_group(lc, &clear_list);
494 if (r)
495 goto fail;
496
497 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
498 NULL, 0, NULL, NULL);
499
500fail:
501 /*
502 * We can safely remove these entries, even if failure.
503 * Calling code will receive an error and will know that
504 * the log facility has failed.
505 */
506 list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) {
507 list_del(&fe->list);
508 mempool_free(fe, flush_entry_pool);
509 }
510 list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) {
511 list_del(&fe->list);
512 mempool_free(fe, flush_entry_pool);
513 }
514
515 if (r)
516 dm_table_event(lc->ti->table);
517
518 return r;
519}
520
521/*
522 * userspace_mark_region
523 *
524 * This function should avoid blocking unless absolutely required.
525 * (Memory allocation is valid for blocking.)
526 */
527static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
528{
529 unsigned long flags;
530 struct log_c *lc = log->context;
531 struct flush_entry *fe;
532
533 /* Wait for an allocation, but _never_ fail */
534 fe = mempool_alloc(flush_entry_pool, GFP_NOIO);
535 BUG_ON(!fe);
536
537 spin_lock_irqsave(&lc->flush_lock, flags);
538 fe->type = DM_ULOG_MARK_REGION;
539 fe->region = region;
540 list_add(&fe->list, &lc->mark_list);
541 spin_unlock_irqrestore(&lc->flush_lock, flags);
542
543 return;
544}
545
546/*
547 * userspace_clear_region
548 *
549 * This function must not block.
550 * So, the alloc can't block. In the worst case, it is ok to
551 * fail. It would simply mean we can't clear the region.
552 * Does nothing to current sync context, but does mean
553 * the region will be re-sync'ed on a reload of the mirror
554 * even though it is in-sync.
555 */
556static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
557{
558 unsigned long flags;
559 struct log_c *lc = log->context;
560 struct flush_entry *fe;
561
562 /*
563 * If we fail to allocate, we skip the clearing of
564 * the region. This doesn't hurt us in any way, except
565 * to cause the region to be resync'ed when the
566 * device is activated next time.
567 */
568 fe = mempool_alloc(flush_entry_pool, GFP_ATOMIC);
569 if (!fe) {
570 DMERR("Failed to allocate memory to clear region.");
571 return;
572 }
573
574 spin_lock_irqsave(&lc->flush_lock, flags);
575 fe->type = DM_ULOG_CLEAR_REGION;
576 fe->region = region;
577 list_add(&fe->list, &lc->clear_list);
578 spin_unlock_irqrestore(&lc->flush_lock, flags);
579
580 return;
581}
582
583/*
584 * userspace_get_resync_work
585 *
586 * Get a region that needs recovery. It is valid to return
587 * an error for this function.
588 *
589 * Returns: 1 if region filled, 0 if no work, <0 on error
590 */
591static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
592{
593 int r;
594 size_t rdata_size;
595 struct log_c *lc = log->context;
596 struct {
597 int64_t i; /* 64-bit for mix arch compatibility */
598 region_t r;
599 } pkg;
600
601 if (lc->in_sync_hint >= lc->region_count)
602 return 0;
603
604 rdata_size = sizeof(pkg);
605 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
606 NULL, 0,
607 (char *)&pkg, &rdata_size);
608
609 *region = pkg.r;
610 return (r) ? r : (int)pkg.i;
611}
612
613/*
614 * userspace_set_region_sync
615 *
616 * Set the sync status of a given region. This function
617 * must not fail.
618 */
619static void userspace_set_region_sync(struct dm_dirty_log *log,
620 region_t region, int in_sync)
621{
622 int r;
623 struct log_c *lc = log->context;
624 struct {
625 region_t r;
626 int64_t i;
627 } pkg;
628
629 pkg.r = region;
630 pkg.i = (int64_t)in_sync;
631
632 r = userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
633 (char *)&pkg, sizeof(pkg),
634 NULL, NULL);
635
636 /*
637 * It would be nice to be able to report failures.
638 * However, it is easy emough to detect and resolve.
639 */
640 return;
641}
642
643/*
644 * userspace_get_sync_count
645 *
646 * If there is any sort of failure when consulting the server,
647 * we assume that the sync count is zero.
648 *
649 * Returns: sync count on success, 0 on failure
650 */
651static region_t userspace_get_sync_count(struct dm_dirty_log *log)
652{
653 int r;
654 size_t rdata_size;
655 uint64_t sync_count;
656 struct log_c *lc = log->context;
657
658 rdata_size = sizeof(sync_count);
659 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
660 NULL, 0,
661 (char *)&sync_count, &rdata_size);
662
663 if (r)
664 return 0;
665
666 if (sync_count >= lc->region_count)
667 lc->in_sync_hint = lc->region_count;
668
669 return (region_t)sync_count;
670}
671
672/*
673 * userspace_status
674 *
675 * Returns: amount of space consumed
676 */
677static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
678 char *result, unsigned maxlen)
679{
680 int r = 0;
681 char *table_args;
682 size_t sz = (size_t)maxlen;
683 struct log_c *lc = log->context;
684
685 switch (status_type) {
686 case STATUSTYPE_INFO:
687 r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
688 NULL, 0,
689 result, &sz);
690
691 if (r) {
692 sz = 0;
693 DMEMIT("%s 1 COM_FAILURE", log->type->name);
694 }
695 break;
696 case STATUSTYPE_TABLE:
697 sz = 0;
698 table_args = strchr(lc->usr_argv_str, ' ');
699 BUG_ON(!table_args); /* There will always be a ' ' */
700 table_args++;
701
702 DMEMIT("%s %u %s %s ", log->type->name, lc->usr_argc,
703 lc->uuid, table_args);
704 break;
705 }
706 return (r) ? 0 : (int)sz;
707}
708
709/*
710 * userspace_is_remote_recovering
711 *
712 * Returns: 1 if region recovering, 0 otherwise
713 */
714static int userspace_is_remote_recovering(struct dm_dirty_log *log,
715 region_t region)
716{
717 int r;
718 uint64_t region64 = region;
719 struct log_c *lc = log->context;
720 static unsigned long long limit;
721 struct {
722 int64_t is_recovering;
723 uint64_t in_sync_hint;
724 } pkg;
725 size_t rdata_size = sizeof(pkg);
726
727 /*
728 * Once the mirror has been reported to be in-sync,
729 * it will never again ask for recovery work. So,
730 * we can safely say there is not a remote machine
731 * recovering if the device is in-sync. (in_sync_hint
732 * must be reset at resume time.)
733 */
734 if (region < lc->in_sync_hint)
735 return 0;
736 else if (jiffies < limit)
737 return 1;
738
739 limit = jiffies + (HZ / 4);
740 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
741 (char *)®ion64, sizeof(region64),
742 (char *)&pkg, &rdata_size);
743 if (r)
744 return 1;
745
746 lc->in_sync_hint = pkg.in_sync_hint;
747
748 return (int)pkg.is_recovering;
749}
750
751static struct dm_dirty_log_type _userspace_type = {
752 .name = "userspace",
753 .module = THIS_MODULE,
754 .ctr = userspace_ctr,
755 .dtr = userspace_dtr,
756 .presuspend = userspace_presuspend,
757 .postsuspend = userspace_postsuspend,
758 .resume = userspace_resume,
759 .get_region_size = userspace_get_region_size,
760 .is_clean = userspace_is_clean,
761 .in_sync = userspace_in_sync,
762 .flush = userspace_flush,
763 .mark_region = userspace_mark_region,
764 .clear_region = userspace_clear_region,
765 .get_resync_work = userspace_get_resync_work,
766 .set_region_sync = userspace_set_region_sync,
767 .get_sync_count = userspace_get_sync_count,
768 .status = userspace_status,
769 .is_remote_recovering = userspace_is_remote_recovering,
770};
771
772static int __init userspace_dirty_log_init(void)
773{
774 int r = 0;
775
776 flush_entry_pool = mempool_create(100, flush_entry_alloc,
777 flush_entry_free, NULL);
778
779 if (!flush_entry_pool) {
780 DMWARN("Unable to create flush_entry_pool: No memory.");
781 return -ENOMEM;
782 }
783
784 r = dm_ulog_tfr_init();
785 if (r) {
786 DMWARN("Unable to initialize userspace log communications");
787 mempool_destroy(flush_entry_pool);
788 return r;
789 }
790
791 r = dm_dirty_log_type_register(&_userspace_type);
792 if (r) {
793 DMWARN("Couldn't register userspace dirty log type");
794 dm_ulog_tfr_exit();
795 mempool_destroy(flush_entry_pool);
796 return r;
797 }
798
799 DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
800 return 0;
801}
802
803static void __exit userspace_dirty_log_exit(void)
804{
805 dm_dirty_log_type_unregister(&_userspace_type);
806 dm_ulog_tfr_exit();
807 mempool_destroy(flush_entry_pool);
808
809 DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
810 return;
811}
812
813module_init(userspace_dirty_log_init);
814module_exit(userspace_dirty_log_exit);
815
816MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
817MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>");
818MODULE_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 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");