spi-topcliff-pch: add recovery processing in case wait-event timeout
[zen-stable.git] / drivers / md / dm-log-userspace-base.c
blob9429159d9ee335c78c57f19e023b68d602683acc
1 /*
2 * Copyright (C) 2006-2009 Red Hat, Inc.
4 * This file is released under the LGPL.
5 */
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>
14 #include "dm-log-userspace-transfer.h"
16 #define DM_LOG_USERSPACE_VSN "1.1.0"
18 struct flush_entry {
19 int type;
20 region_t region;
21 struct list_head list;
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().
30 #define MAX_FLUSH_GROUP_COUNT 32
32 struct 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];
40 char *usr_argv_str;
41 uint32_t usr_argc;
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.
50 uint64_t in_sync_hint;
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.
58 spinlock_t flush_lock;
59 struct list_head mark_list;
60 struct list_head clear_list;
63 static mempool_t *flush_entry_pool;
65 static void *flush_entry_alloc(gfp_t gfp_mask, void *pool_data)
67 return kmalloc(sizeof(struct flush_entry), gfp_mask);
70 static void flush_entry_free(void *element, void *pool_data)
72 kfree(element);
75 static 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)
79 int r;
82 * If the server isn't there, -ESRCH is returned,
83 * and we must keep trying until the server is
84 * restored.
86 retry:
87 r = dm_consult_userspace(uuid, lc->luid, request_type, data,
88 data_size, rdata, rdata_size);
90 if (r != -ESRCH)
91 return r;
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;
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;
111 DMERR("Error trying to resume userspace log: %d", r);
113 return -ESRCH;
116 static int build_constructor_string(struct dm_target *ti,
117 unsigned argc, char **argv,
118 char **ctr_str)
120 int i, str_size;
121 char *str = NULL;
123 *ctr_str = NULL;
125 for (i = 0, str_size = 0; i < argc; i++)
126 str_size += strlen(argv[i]) + 1; /* +1 for space between args */
128 str_size += 20; /* Max number of chars in a printed u64 number */
130 str = kzalloc(str_size, GFP_KERNEL);
131 if (!str) {
132 DMWARN("Unable to allocate memory for constructor string");
133 return -ENOMEM;
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]);
140 *ctr_str = str;
141 return str_size;
145 * userspace_ctr
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]
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.
157 static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
158 unsigned argc, char **argv)
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;
169 if (argc < 3) {
170 DMWARN("Too few arguments to userspace dirty log");
171 return -EINVAL;
174 lc = kzalloc(sizeof(*lc), GFP_KERNEL);
175 if (!lc) {
176 DMWARN("Unable to allocate userspace log context.");
177 return -ENOMEM;
180 /* The ptr value is sufficient for local unique id */
181 lc->luid = (unsigned long)lc;
183 lc->ti = ti;
185 if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
186 DMWARN("UUID argument too long.");
187 kfree(lc);
188 return -EINVAL;
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);
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;
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;
210 * Send table string and get back any opened device.
212 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
213 ctr_str, str_size,
214 devices_rdata, &devices_rdata_size);
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;
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);
229 if (r) {
230 DMERR("Failed to get region size of dirty log");
231 goto out;
234 lc->region_size = (uint32_t)rdata;
235 lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
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;
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);
249 out:
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;
260 return r;
263 static void userspace_dtr(struct dm_dirty_log *log)
265 struct log_c *lc = log->context;
267 (void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
268 NULL, 0,
269 NULL, NULL);
271 if (lc->log_dev)
272 dm_put_device(lc->ti, lc->log_dev);
274 kfree(lc->usr_argv_str);
275 kfree(lc);
277 return;
280 static int userspace_presuspend(struct dm_dirty_log *log)
282 int r;
283 struct log_c *lc = log->context;
285 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
286 NULL, 0,
287 NULL, NULL);
289 return r;
292 static int userspace_postsuspend(struct dm_dirty_log *log)
294 int r;
295 struct log_c *lc = log->context;
297 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
298 NULL, 0,
299 NULL, NULL);
301 return r;
304 static int userspace_resume(struct dm_dirty_log *log)
306 int r;
307 struct log_c *lc = log->context;
309 lc->in_sync_hint = 0;
310 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
311 NULL, 0,
312 NULL, NULL);
314 return r;
317 static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
319 struct log_c *lc = log->context;
321 return lc->region_size;
325 * userspace_is_clean
327 * Check whether a region is clean. If there is any sort of
328 * failure when consulting the server, we return not clean.
330 * Returns: 1 if clean, 0 otherwise
332 static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
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;
340 rdata_size = sizeof(is_clean);
341 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
342 (char *)&region64, sizeof(region64),
343 (char *)&is_clean, &rdata_size);
345 return (r) ? 0 : (int)is_clean;
349 * userspace_in_sync
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.
355 * If 'can_block' is set, return immediately
357 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
359 static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
360 int can_block)
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;
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.)
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.
381 if (!can_block)
382 return -EWOULDBLOCK;
384 rdata_size = sizeof(in_sync);
385 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
386 (char *)&region64, sizeof(region64),
387 (char *)&in_sync, &rdata_size);
388 return (r) ? 0 : (int)in_sync;
391 static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
393 int r = 0;
394 struct flush_entry *fe;
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;
405 return r;
408 static int flush_by_group(struct log_c *lc, struct list_head *flush_list)
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];
418 * Group process the requests
420 while (!list_empty(flush_list)) {
421 count = 0;
423 list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
424 group[count] = fe->region;
425 count++;
427 list_move(&fe->list, &tmp_list);
429 type = fe->type;
430 if (count >= MAX_FLUSH_GROUP_COUNT)
431 break;
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;
447 * Must collect flush_entrys that were successfully processed
448 * as a group so that they will be free'd by the caller.
450 list_splice_init(&tmp_list, flush_list);
452 return r;
456 * userspace_flush
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.
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.
470 * Returns: 0 on success, < 0 on failure
472 static int userspace_flush(struct dm_dirty_log *log)
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;
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);
486 if (list_empty(&mark_list) && list_empty(&clear_list))
487 return 0;
489 r = flush_by_group(lc, &mark_list);
490 if (r)
491 goto fail;
493 r = flush_by_group(lc, &clear_list);
494 if (r)
495 goto fail;
497 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
498 NULL, 0, NULL, NULL);
500 fail:
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.
506 list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) {
507 list_del(&fe->list);
508 mempool_free(fe, flush_entry_pool);
510 list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) {
511 list_del(&fe->list);
512 mempool_free(fe, flush_entry_pool);
515 if (r)
516 dm_table_event(lc->ti->table);
518 return r;
522 * userspace_mark_region
524 * This function should avoid blocking unless absolutely required.
525 * (Memory allocation is valid for blocking.)
527 static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
529 unsigned long flags;
530 struct log_c *lc = log->context;
531 struct flush_entry *fe;
533 /* Wait for an allocation, but _never_ fail */
534 fe = mempool_alloc(flush_entry_pool, GFP_NOIO);
535 BUG_ON(!fe);
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);
543 return;
547 * userspace_clear_region
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.
556 static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
558 unsigned long flags;
559 struct log_c *lc = log->context;
560 struct flush_entry *fe;
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.
568 fe = mempool_alloc(flush_entry_pool, GFP_ATOMIC);
569 if (!fe) {
570 DMERR("Failed to allocate memory to clear region.");
571 return;
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);
580 return;
584 * userspace_get_resync_work
586 * Get a region that needs recovery. It is valid to return
587 * an error for this function.
589 * Returns: 1 if region filled, 0 if no work, <0 on error
591 static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
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;
601 if (lc->in_sync_hint >= lc->region_count)
602 return 0;
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);
609 *region = pkg.r;
610 return (r) ? r : (int)pkg.i;
614 * userspace_set_region_sync
616 * Set the sync status of a given region. This function
617 * must not fail.
619 static void userspace_set_region_sync(struct dm_dirty_log *log,
620 region_t region, int in_sync)
622 int r;
623 struct log_c *lc = log->context;
624 struct {
625 region_t r;
626 int64_t i;
627 } pkg;
629 pkg.r = region;
630 pkg.i = (int64_t)in_sync;
632 r = userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
633 (char *)&pkg, sizeof(pkg),
634 NULL, NULL);
637 * It would be nice to be able to report failures.
638 * However, it is easy emough to detect and resolve.
640 return;
644 * userspace_get_sync_count
646 * If there is any sort of failure when consulting the server,
647 * we assume that the sync count is zero.
649 * Returns: sync count on success, 0 on failure
651 static region_t userspace_get_sync_count(struct dm_dirty_log *log)
653 int r;
654 size_t rdata_size;
655 uint64_t sync_count;
656 struct log_c *lc = log->context;
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);
663 if (r)
664 return 0;
666 if (sync_count >= lc->region_count)
667 lc->in_sync_hint = lc->region_count;
669 return (region_t)sync_count;
673 * userspace_status
675 * Returns: amount of space consumed
677 static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
678 char *result, unsigned maxlen)
680 int r = 0;
681 char *table_args;
682 size_t sz = (size_t)maxlen;
683 struct log_c *lc = log->context;
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);
691 if (r) {
692 sz = 0;
693 DMEMIT("%s 1 COM_FAILURE", log->type->name);
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++;
702 DMEMIT("%s %u %s %s ", log->type->name, lc->usr_argc,
703 lc->uuid, table_args);
704 break;
706 return (r) ? 0 : (int)sz;
710 * userspace_is_remote_recovering
712 * Returns: 1 if region recovering, 0 otherwise
714 static int userspace_is_remote_recovering(struct dm_dirty_log *log,
715 region_t region)
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);
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.)
734 if (region < lc->in_sync_hint)
735 return 0;
736 else if (jiffies < limit)
737 return 1;
739 limit = jiffies + (HZ / 4);
740 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
741 (char *)&region64, sizeof(region64),
742 (char *)&pkg, &rdata_size);
743 if (r)
744 return 1;
746 lc->in_sync_hint = pkg.in_sync_hint;
748 return (int)pkg.is_recovering;
751 static 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,
772 static int __init userspace_dirty_log_init(void)
774 int r = 0;
776 flush_entry_pool = mempool_create(100, flush_entry_alloc,
777 flush_entry_free, NULL);
779 if (!flush_entry_pool) {
780 DMWARN("Unable to create flush_entry_pool: No memory.");
781 return -ENOMEM;
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;
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;
799 DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
800 return 0;
803 static void __exit userspace_dirty_log_exit(void)
805 dm_dirty_log_type_unregister(&_userspace_type);
806 dm_ulog_tfr_exit();
807 mempool_destroy(flush_entry_pool);
809 DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
810 return;
813 module_init(userspace_dirty_log_init);
814 module_exit(userspace_dirty_log_exit);
816 MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
817 MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>");
818 MODULE_LICENSE("GPL");