2 * Copyright (C) 2006-2009 Red Hat, Inc.
4 * This file is released under the LGPL.
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>
16 #include "dm-log-userspace-transfer.h"
18 #define DM_LOG_USERSPACE_VSN "1.3.0"
23 struct list_head list;
27 * This limit on the number of mark and clear request is, to a degree,
28 * arbitrary. However, there is some basis for the choice in the limits
29 * imposed on the size of data payload by dm-log-userspace-transfer.c:
30 * dm_consult_userspace().
32 #define MAX_FLUSH_GROUP_COUNT 32
36 struct dm_dev *log_dev;
38 region_t region_count;
40 char uuid[DM_UUID_LEN];
46 * in_sync_hint gets set when doing is_remote_recovering. It
47 * represents the first region that needs recovery. IOW, the
48 * first zero bit of sync_bits. This can be useful for to limit
49 * traffic for calls like is_remote_recovering and get_resync_work,
50 * but be take care in its use for anything else.
52 uint64_t in_sync_hint;
55 * Mark and clear requests are held until a flush is issued
56 * so that we can group, and thereby limit, the amount of
57 * network traffic between kernel and userspace. The 'flush_lock'
58 * is used to protect these lists.
60 spinlock_t flush_lock;
61 struct list_head mark_list;
62 struct list_head clear_list;
65 * Workqueue for flush of clear region requests.
67 struct workqueue_struct *dmlog_wq;
68 struct delayed_work flush_log_work;
72 * Combine userspace flush and mark requests for efficiency.
74 uint32_t integrated_flush;
77 static mempool_t *flush_entry_pool;
79 static void *flush_entry_alloc(gfp_t gfp_mask, void *pool_data)
81 return kmalloc(sizeof(struct flush_entry), gfp_mask);
84 static void flush_entry_free(void *element, void *pool_data)
89 static int userspace_do_request(struct log_c *lc, const char *uuid,
90 int request_type, char *data, size_t data_size,
91 char *rdata, size_t *rdata_size)
96 * If the server isn't there, -ESRCH is returned,
97 * and we must keep trying until the server is
101 r = dm_consult_userspace(uuid, lc->luid, request_type, data,
102 data_size, rdata, rdata_size);
107 DMERR(" Userspace log server not found.");
109 set_current_state(TASK_INTERRUPTIBLE);
110 schedule_timeout(2*HZ);
111 DMWARN("Attempting to contact userspace log server...");
112 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR,
114 strlen(lc->usr_argv_str) + 1,
119 DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
120 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL,
125 DMERR("Error trying to resume userspace log: %d", r);
130 static int build_constructor_string(struct dm_target *ti,
131 unsigned argc, char **argv,
140 * Determine overall size of the string.
142 for (i = 0, str_size = 0; i < argc; i++)
143 str_size += strlen(argv[i]) + 1; /* +1 for space between args */
145 str_size += 20; /* Max number of chars in a printed u64 number */
147 str = kzalloc(str_size, GFP_KERNEL);
149 DMWARN("Unable to allocate memory for constructor string");
153 str_size = sprintf(str, "%llu", (unsigned long long)ti->len);
154 for (i = 0; i < argc; i++)
155 str_size += sprintf(str + str_size, " %s", argv[i]);
161 static void do_flush(struct work_struct *work)
164 struct log_c *lc = container_of(work, struct log_c, flush_log_work.work);
166 atomic_set(&lc->sched_flush, 0);
168 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, NULL, 0, NULL, NULL);
171 dm_table_event(lc->ti->table);
178 * <UUID> [integrated_flush] <other args>
179 * Where 'other args' are the userspace implementation-specific log
183 * <UUID> [integrated_flush] clustered-disk <arg count> <log dev>
184 * <region_size> [[no]sync]
186 * This module strips off the <UUID> and uses it for identification
187 * purposes when communicating with userspace about a log.
189 * If integrated_flush is defined, the kernel combines flush
192 * The rest of the line, beginning with 'clustered-disk', is passed
193 * to the userspace ctr function.
195 static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
196 unsigned argc, char **argv)
200 char *ctr_str = NULL;
201 struct log_c *lc = NULL;
203 size_t rdata_size = sizeof(rdata);
204 char *devices_rdata = NULL;
205 size_t devices_rdata_size = DM_NAME_LEN;
208 DMWARN("Too few arguments to userspace dirty log");
212 lc = kzalloc(sizeof(*lc), GFP_KERNEL);
214 DMWARN("Unable to allocate userspace log context.");
218 /* The ptr value is sufficient for local unique id */
219 lc->luid = (unsigned long)lc;
223 if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
224 DMWARN("UUID argument too long.");
231 strncpy(lc->uuid, argv[0], DM_UUID_LEN);
234 spin_lock_init(&lc->flush_lock);
235 INIT_LIST_HEAD(&lc->mark_list);
236 INIT_LIST_HEAD(&lc->clear_list);
238 if (!strcasecmp(argv[0], "integrated_flush")) {
239 lc->integrated_flush = 1;
244 str_size = build_constructor_string(ti, argc, argv, &ctr_str);
250 devices_rdata = kzalloc(devices_rdata_size, GFP_KERNEL);
251 if (!devices_rdata) {
252 DMERR("Failed to allocate memory for device information");
258 * Send table string and get back any opened device.
260 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
262 devices_rdata, &devices_rdata_size);
266 DMERR("Userspace log server not found");
268 DMERR("Userspace log server failed to create log");
272 /* Since the region size does not change, get it now */
273 rdata_size = sizeof(rdata);
274 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
275 NULL, 0, (char *)&rdata, &rdata_size);
278 DMERR("Failed to get region size of dirty log");
282 lc->region_size = (uint32_t)rdata;
283 lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
285 if (devices_rdata_size) {
286 if (devices_rdata[devices_rdata_size - 1] != '\0') {
287 DMERR("DM_ULOG_CTR device return string not properly terminated");
291 r = dm_get_device(ti, devices_rdata,
292 dm_table_get_mode(ti->table), &lc->log_dev);
294 DMERR("Failed to register %s with device-mapper",
298 if (lc->integrated_flush) {
299 lc->dmlog_wq = alloc_workqueue("dmlogd", WQ_MEM_RECLAIM, 0);
301 DMERR("couldn't start dmlogd");
306 INIT_DELAYED_WORK(&lc->flush_log_work, do_flush);
307 atomic_set(&lc->sched_flush, 0);
311 kfree(devices_rdata);
316 lc->usr_argv_str = ctr_str;
323 static void userspace_dtr(struct dm_dirty_log *log)
325 struct log_c *lc = log->context;
327 if (lc->integrated_flush) {
328 /* flush workqueue */
329 if (atomic_read(&lc->sched_flush))
330 flush_delayed_work(&lc->flush_log_work);
332 destroy_workqueue(lc->dmlog_wq);
335 (void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
336 NULL, 0, NULL, NULL);
339 dm_put_device(lc->ti, lc->log_dev);
341 kfree(lc->usr_argv_str);
347 static int userspace_presuspend(struct dm_dirty_log *log)
350 struct log_c *lc = log->context;
352 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
353 NULL, 0, NULL, NULL);
358 static int userspace_postsuspend(struct dm_dirty_log *log)
361 struct log_c *lc = log->context;
364 * Run planned flush earlier.
366 if (lc->integrated_flush && atomic_read(&lc->sched_flush))
367 flush_delayed_work(&lc->flush_log_work);
369 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
370 NULL, 0, NULL, NULL);
375 static int userspace_resume(struct dm_dirty_log *log)
378 struct log_c *lc = log->context;
380 lc->in_sync_hint = 0;
381 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
382 NULL, 0, NULL, NULL);
387 static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
389 struct log_c *lc = log->context;
391 return lc->region_size;
397 * Check whether a region is clean. If there is any sort of
398 * failure when consulting the server, we return not clean.
400 * Returns: 1 if clean, 0 otherwise
402 static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
405 uint64_t region64 = (uint64_t)region;
408 struct log_c *lc = log->context;
410 rdata_size = sizeof(is_clean);
411 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
412 (char *)®ion64, sizeof(region64),
413 (char *)&is_clean, &rdata_size);
415 return (r) ? 0 : (int)is_clean;
421 * Check if the region is in-sync. If there is any sort
422 * of failure when consulting the server, we assume that
423 * the region is not in sync.
425 * If 'can_block' is set, return immediately
427 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
429 static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
433 uint64_t region64 = region;
436 struct log_c *lc = log->context;
439 * We can never respond directly - even if in_sync_hint is
440 * set. This is because another machine could see a device
441 * failure and mark the region out-of-sync. If we don't go
442 * to userspace to ask, we might think the region is in-sync
443 * and allow a read to pick up data that is stale. (This is
444 * very unlikely if a device actually fails; but it is very
445 * likely if a connection to one device from one machine fails.)
447 * There still might be a problem if the mirror caches the region
448 * state as in-sync... but then this call would not be made. So,
449 * that is a mirror problem.
454 rdata_size = sizeof(in_sync);
455 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
456 (char *)®ion64, sizeof(region64),
457 (char *)&in_sync, &rdata_size);
458 return (r) ? 0 : (int)in_sync;
461 static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
464 struct flush_entry *fe;
466 list_for_each_entry(fe, flush_list, list) {
467 r = userspace_do_request(lc, lc->uuid, fe->type,
478 static int flush_by_group(struct log_c *lc, struct list_head *flush_list,
479 int flush_with_payload)
484 struct flush_entry *fe, *tmp_fe;
486 uint64_t group[MAX_FLUSH_GROUP_COUNT];
489 * Group process the requests
491 while (!list_empty(flush_list)) {
494 list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
495 group[count] = fe->region;
498 list_move(&fe->list, &tmp_list);
501 if (count >= MAX_FLUSH_GROUP_COUNT)
505 if (flush_with_payload) {
506 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
508 count * sizeof(uint64_t),
511 * Integrated flush failed.
516 r = userspace_do_request(lc, lc->uuid, type,
518 count * sizeof(uint64_t),
522 * Group send failed. Attempt one-by-one.
524 list_splice_init(&tmp_list, flush_list);
525 r = flush_one_by_one(lc, flush_list);
532 * Must collect flush_entrys that were successfully processed
533 * as a group so that they will be free'd by the caller.
535 list_splice_init(&tmp_list, flush_list);
543 * This function is ok to block.
544 * The flush happens in two stages. First, it sends all
545 * clear/mark requests that are on the list. Then it
546 * tells the server to commit them. This gives the
547 * server a chance to optimise the commit, instead of
548 * doing it for every request.
550 * Additionally, we could implement another thread that
551 * sends the requests up to the server - reducing the
552 * load on flush. Then the flush would have less in
553 * the list and be responsible for the finishing commit.
555 * Returns: 0 on success, < 0 on failure
557 static int userspace_flush(struct dm_dirty_log *log)
561 struct log_c *lc = log->context;
562 LIST_HEAD(mark_list);
563 LIST_HEAD(clear_list);
564 int mark_list_is_empty;
565 int clear_list_is_empty;
566 struct flush_entry *fe, *tmp_fe;
568 spin_lock_irqsave(&lc->flush_lock, flags);
569 list_splice_init(&lc->mark_list, &mark_list);
570 list_splice_init(&lc->clear_list, &clear_list);
571 spin_unlock_irqrestore(&lc->flush_lock, flags);
573 mark_list_is_empty = list_empty(&mark_list);
574 clear_list_is_empty = list_empty(&clear_list);
576 if (mark_list_is_empty && clear_list_is_empty)
579 r = flush_by_group(lc, &clear_list, 0);
583 if (!lc->integrated_flush) {
584 r = flush_by_group(lc, &mark_list, 0);
587 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
588 NULL, 0, NULL, NULL);
593 * Send integrated flush request with mark_list as payload.
595 r = flush_by_group(lc, &mark_list, 1);
599 if (mark_list_is_empty && !atomic_read(&lc->sched_flush)) {
601 * When there are only clear region requests,
602 * we schedule a flush in the future.
604 queue_delayed_work(lc->dmlog_wq, &lc->flush_log_work, 3 * HZ);
605 atomic_set(&lc->sched_flush, 1);
608 * Cancel pending flush because we
609 * have already flushed in mark_region.
611 cancel_delayed_work(&lc->flush_log_work);
612 atomic_set(&lc->sched_flush, 0);
617 * We can safely remove these entries, even after failure.
618 * Calling code will receive an error and will know that
619 * the log facility has failed.
621 list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) {
623 mempool_free(fe, flush_entry_pool);
625 list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) {
627 mempool_free(fe, flush_entry_pool);
631 dm_table_event(lc->ti->table);
637 * userspace_mark_region
639 * This function should avoid blocking unless absolutely required.
640 * (Memory allocation is valid for blocking.)
642 static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
645 struct log_c *lc = log->context;
646 struct flush_entry *fe;
648 /* Wait for an allocation, but _never_ fail */
649 fe = mempool_alloc(flush_entry_pool, GFP_NOIO);
652 spin_lock_irqsave(&lc->flush_lock, flags);
653 fe->type = DM_ULOG_MARK_REGION;
655 list_add(&fe->list, &lc->mark_list);
656 spin_unlock_irqrestore(&lc->flush_lock, flags);
662 * userspace_clear_region
664 * This function must not block.
665 * So, the alloc can't block. In the worst case, it is ok to
666 * fail. It would simply mean we can't clear the region.
667 * Does nothing to current sync context, but does mean
668 * the region will be re-sync'ed on a reload of the mirror
669 * even though it is in-sync.
671 static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
674 struct log_c *lc = log->context;
675 struct flush_entry *fe;
678 * If we fail to allocate, we skip the clearing of
679 * the region. This doesn't hurt us in any way, except
680 * to cause the region to be resync'ed when the
681 * device is activated next time.
683 fe = mempool_alloc(flush_entry_pool, GFP_ATOMIC);
685 DMERR("Failed to allocate memory to clear region.");
689 spin_lock_irqsave(&lc->flush_lock, flags);
690 fe->type = DM_ULOG_CLEAR_REGION;
692 list_add(&fe->list, &lc->clear_list);
693 spin_unlock_irqrestore(&lc->flush_lock, flags);
699 * userspace_get_resync_work
701 * Get a region that needs recovery. It is valid to return
702 * an error for this function.
704 * Returns: 1 if region filled, 0 if no work, <0 on error
706 static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
710 struct log_c *lc = log->context;
712 int64_t i; /* 64-bit for mix arch compatibility */
716 if (lc->in_sync_hint >= lc->region_count)
719 rdata_size = sizeof(pkg);
720 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
721 NULL, 0, (char *)&pkg, &rdata_size);
724 return (r) ? r : (int)pkg.i;
728 * userspace_set_region_sync
730 * Set the sync status of a given region. This function
733 static void userspace_set_region_sync(struct dm_dirty_log *log,
734 region_t region, int in_sync)
737 struct log_c *lc = log->context;
744 pkg.i = (int64_t)in_sync;
746 r = userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
747 (char *)&pkg, sizeof(pkg), NULL, NULL);
750 * It would be nice to be able to report failures.
751 * However, it is easy emough to detect and resolve.
757 * userspace_get_sync_count
759 * If there is any sort of failure when consulting the server,
760 * we assume that the sync count is zero.
762 * Returns: sync count on success, 0 on failure
764 static region_t userspace_get_sync_count(struct dm_dirty_log *log)
769 struct log_c *lc = log->context;
771 rdata_size = sizeof(sync_count);
772 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
773 NULL, 0, (char *)&sync_count, &rdata_size);
778 if (sync_count >= lc->region_count)
779 lc->in_sync_hint = lc->region_count;
781 return (region_t)sync_count;
787 * Returns: amount of space consumed
789 static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
790 char *result, unsigned maxlen)
794 size_t sz = (size_t)maxlen;
795 struct log_c *lc = log->context;
797 switch (status_type) {
798 case STATUSTYPE_INFO:
799 r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
800 NULL, 0, result, &sz);
804 DMEMIT("%s 1 COM_FAILURE", log->type->name);
807 case STATUSTYPE_TABLE:
809 table_args = strchr(lc->usr_argv_str, ' ');
810 BUG_ON(!table_args); /* There will always be a ' ' */
813 DMEMIT("%s %u %s ", log->type->name, lc->usr_argc, lc->uuid);
814 if (lc->integrated_flush)
815 DMEMIT("integrated_flush ");
816 DMEMIT("%s ", table_args);
819 return (r) ? 0 : (int)sz;
823 * userspace_is_remote_recovering
825 * Returns: 1 if region recovering, 0 otherwise
827 static int userspace_is_remote_recovering(struct dm_dirty_log *log,
831 uint64_t region64 = region;
832 struct log_c *lc = log->context;
833 static unsigned long limit;
835 int64_t is_recovering;
836 uint64_t in_sync_hint;
838 size_t rdata_size = sizeof(pkg);
841 * Once the mirror has been reported to be in-sync,
842 * it will never again ask for recovery work. So,
843 * we can safely say there is not a remote machine
844 * recovering if the device is in-sync. (in_sync_hint
845 * must be reset at resume time.)
847 if (region < lc->in_sync_hint)
849 else if (time_after(limit, jiffies))
852 limit = jiffies + (HZ / 4);
853 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
854 (char *)®ion64, sizeof(region64),
855 (char *)&pkg, &rdata_size);
859 lc->in_sync_hint = pkg.in_sync_hint;
861 return (int)pkg.is_recovering;
864 static struct dm_dirty_log_type _userspace_type = {
866 .module = THIS_MODULE,
867 .ctr = userspace_ctr,
868 .dtr = userspace_dtr,
869 .presuspend = userspace_presuspend,
870 .postsuspend = userspace_postsuspend,
871 .resume = userspace_resume,
872 .get_region_size = userspace_get_region_size,
873 .is_clean = userspace_is_clean,
874 .in_sync = userspace_in_sync,
875 .flush = userspace_flush,
876 .mark_region = userspace_mark_region,
877 .clear_region = userspace_clear_region,
878 .get_resync_work = userspace_get_resync_work,
879 .set_region_sync = userspace_set_region_sync,
880 .get_sync_count = userspace_get_sync_count,
881 .status = userspace_status,
882 .is_remote_recovering = userspace_is_remote_recovering,
885 static int __init userspace_dirty_log_init(void)
889 flush_entry_pool = mempool_create(100, flush_entry_alloc,
890 flush_entry_free, NULL);
892 if (!flush_entry_pool) {
893 DMWARN("Unable to create flush_entry_pool: No memory.");
897 r = dm_ulog_tfr_init();
899 DMWARN("Unable to initialize userspace log communications");
900 mempool_destroy(flush_entry_pool);
904 r = dm_dirty_log_type_register(&_userspace_type);
906 DMWARN("Couldn't register userspace dirty log type");
908 mempool_destroy(flush_entry_pool);
912 DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
916 static void __exit userspace_dirty_log_exit(void)
918 dm_dirty_log_type_unregister(&_userspace_type);
920 mempool_destroy(flush_entry_pool);
922 DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
926 module_init(userspace_dirty_log_init);
927 module_exit(userspace_dirty_log_exit);
929 MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
931 MODULE_LICENSE("GPL");