2 * Copyright (C) 2006-2009 Red Hat, Inc.
4 * This file is released under the LGPL.
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>
15 #include "dm-log-userspace-transfer.h"
17 #define DM_LOG_USERSPACE_VSN "1.3.0"
22 struct list_head list
;
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().
31 #define MAX_FLUSH_GROUP_COUNT 32
35 struct dm_dev
*log_dev
;
37 region_t region_count
;
39 char uuid
[DM_UUID_LEN
];
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.
51 uint64_t in_sync_hint
;
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.
59 spinlock_t flush_lock
;
60 struct list_head mark_list
;
61 struct list_head clear_list
;
64 * Workqueue for flush of clear region requests.
66 struct workqueue_struct
*dmlog_wq
;
67 struct delayed_work flush_log_work
;
71 * Combine userspace flush and mark requests for efficiency.
73 uint32_t integrated_flush
;
76 static mempool_t
*flush_entry_pool
;
78 static void *flush_entry_alloc(gfp_t gfp_mask
, void *pool_data
)
80 return kmalloc(sizeof(struct flush_entry
), gfp_mask
);
83 static void flush_entry_free(void *element
, void *pool_data
)
88 static 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
)
95 * If the server isn't there, -ESRCH is returned,
96 * and we must keep trying until the server is
100 r
= dm_consult_userspace(uuid
, lc
->luid
, request_type
, data
,
101 data_size
, rdata
, rdata_size
);
106 DMERR(" Userspace log server not found.");
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
,
113 strlen(lc
->usr_argv_str
) + 1,
118 DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
119 r
= dm_consult_userspace(uuid
, lc
->luid
, DM_ULOG_RESUME
, NULL
,
124 DMERR("Error trying to resume userspace log: %d", r
);
129 static int build_constructor_string(struct dm_target
*ti
,
130 unsigned argc
, char **argv
,
139 * Determine overall size of the string.
141 for (i
= 0, str_size
= 0; i
< argc
; i
++)
142 str_size
+= strlen(argv
[i
]) + 1; /* +1 for space between args */
144 str_size
+= 20; /* Max number of chars in a printed u64 number */
146 str
= kzalloc(str_size
, GFP_KERNEL
);
148 DMWARN("Unable to allocate memory for constructor string");
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
]);
160 static void do_flush(struct work_struct
*work
)
163 struct log_c
*lc
= container_of(work
, struct log_c
, flush_log_work
.work
);
165 atomic_set(&lc
->sched_flush
, 0);
167 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_FLUSH
, NULL
, 0, NULL
, NULL
);
170 dm_table_event(lc
->ti
->table
);
177 * <UUID> [integrated_flush] <other args>
178 * Where 'other args' are the userspace implementation-specific log
182 * <UUID> [integrated_flush] clustered-disk <arg count> <log dev>
183 * <region_size> [[no]sync]
185 * This module strips off the <UUID> and uses it for identification
186 * purposes when communicating with userspace about a log.
188 * If integrated_flush is defined, the kernel combines flush
191 * The rest of the line, beginning with 'clustered-disk', is passed
192 * to the userspace ctr function.
194 static int userspace_ctr(struct dm_dirty_log
*log
, struct dm_target
*ti
,
195 unsigned argc
, char **argv
)
199 char *ctr_str
= NULL
;
200 struct log_c
*lc
= NULL
;
202 size_t rdata_size
= sizeof(rdata
);
203 char *devices_rdata
= NULL
;
204 size_t devices_rdata_size
= DM_NAME_LEN
;
207 DMWARN("Too few arguments to userspace dirty log");
211 lc
= kzalloc(sizeof(*lc
), GFP_KERNEL
);
213 DMWARN("Unable to allocate userspace log context.");
217 /* The ptr value is sufficient for local unique id */
218 lc
->luid
= (unsigned long)lc
;
222 if (strlen(argv
[0]) > (DM_UUID_LEN
- 1)) {
223 DMWARN("UUID argument too long.");
230 strncpy(lc
->uuid
, argv
[0], DM_UUID_LEN
);
233 spin_lock_init(&lc
->flush_lock
);
234 INIT_LIST_HEAD(&lc
->mark_list
);
235 INIT_LIST_HEAD(&lc
->clear_list
);
237 if (!strcasecmp(argv
[0], "integrated_flush")) {
238 lc
->integrated_flush
= 1;
243 str_size
= build_constructor_string(ti
, argc
, argv
, &ctr_str
);
249 devices_rdata
= kzalloc(devices_rdata_size
, GFP_KERNEL
);
250 if (!devices_rdata
) {
251 DMERR("Failed to allocate memory for device information");
257 * Send table string and get back any opened device.
259 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_CTR
,
261 devices_rdata
, &devices_rdata_size
);
265 DMERR("Userspace log server not found");
267 DMERR("Userspace log server failed to create log");
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
);
277 DMERR("Failed to get region size of dirty log");
281 lc
->region_size
= (uint32_t)rdata
;
282 lc
->region_count
= dm_sector_div_up(ti
->len
, lc
->region_size
);
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");
290 r
= dm_get_device(ti
, devices_rdata
,
291 dm_table_get_mode(ti
->table
), &lc
->log_dev
);
293 DMERR("Failed to register %s with device-mapper",
297 if (lc
->integrated_flush
) {
298 lc
->dmlog_wq
= alloc_workqueue("dmlogd", WQ_MEM_RECLAIM
, 0);
300 DMERR("couldn't start dmlogd");
305 INIT_DELAYED_WORK(&lc
->flush_log_work
, do_flush
);
306 atomic_set(&lc
->sched_flush
, 0);
310 kfree(devices_rdata
);
315 lc
->usr_argv_str
= ctr_str
;
322 static void userspace_dtr(struct dm_dirty_log
*log
)
324 struct log_c
*lc
= log
->context
;
326 if (lc
->integrated_flush
) {
327 /* flush workqueue */
328 if (atomic_read(&lc
->sched_flush
))
329 flush_delayed_work(&lc
->flush_log_work
);
331 destroy_workqueue(lc
->dmlog_wq
);
334 (void) dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_DTR
,
335 NULL
, 0, NULL
, NULL
);
338 dm_put_device(lc
->ti
, lc
->log_dev
);
340 kfree(lc
->usr_argv_str
);
346 static int userspace_presuspend(struct dm_dirty_log
*log
)
349 struct log_c
*lc
= log
->context
;
351 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_PRESUSPEND
,
352 NULL
, 0, NULL
, NULL
);
357 static int userspace_postsuspend(struct dm_dirty_log
*log
)
360 struct log_c
*lc
= log
->context
;
363 * Run planned flush earlier.
365 if (lc
->integrated_flush
&& atomic_read(&lc
->sched_flush
))
366 flush_delayed_work(&lc
->flush_log_work
);
368 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_POSTSUSPEND
,
369 NULL
, 0, NULL
, NULL
);
374 static int userspace_resume(struct dm_dirty_log
*log
)
377 struct log_c
*lc
= log
->context
;
379 lc
->in_sync_hint
= 0;
380 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_RESUME
,
381 NULL
, 0, NULL
, NULL
);
386 static uint32_t userspace_get_region_size(struct dm_dirty_log
*log
)
388 struct log_c
*lc
= log
->context
;
390 return lc
->region_size
;
396 * Check whether a region is clean. If there is any sort of
397 * failure when consulting the server, we return not clean.
399 * Returns: 1 if clean, 0 otherwise
401 static int userspace_is_clean(struct dm_dirty_log
*log
, region_t region
)
404 uint64_t region64
= (uint64_t)region
;
407 struct log_c
*lc
= log
->context
;
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
);
414 return (r
) ? 0 : (int)is_clean
;
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.
424 * If 'can_block' is set, return immediately
426 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
428 static int userspace_in_sync(struct dm_dirty_log
*log
, region_t region
,
432 uint64_t region64
= region
;
435 struct log_c
*lc
= log
->context
;
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.)
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.
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
;
460 static int flush_one_by_one(struct log_c
*lc
, struct list_head
*flush_list
)
463 struct flush_entry
*fe
;
465 list_for_each_entry(fe
, flush_list
, list
) {
466 r
= userspace_do_request(lc
, lc
->uuid
, fe
->type
,
477 static int flush_by_group(struct log_c
*lc
, struct list_head
*flush_list
,
478 int flush_with_payload
)
483 struct flush_entry
*fe
, *tmp_fe
;
485 uint64_t group
[MAX_FLUSH_GROUP_COUNT
];
488 * Group process the requests
490 while (!list_empty(flush_list
)) {
493 list_for_each_entry_safe(fe
, tmp_fe
, flush_list
, list
) {
494 group
[count
] = fe
->region
;
497 list_move(&fe
->list
, &tmp_list
);
500 if (count
>= MAX_FLUSH_GROUP_COUNT
)
504 if (flush_with_payload
) {
505 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_FLUSH
,
507 count
* sizeof(uint64_t),
510 * Integrated flush failed.
515 r
= userspace_do_request(lc
, lc
->uuid
, type
,
517 count
* sizeof(uint64_t),
521 * Group send failed. Attempt one-by-one.
523 list_splice_init(&tmp_list
, flush_list
);
524 r
= flush_one_by_one(lc
, flush_list
);
531 * Must collect flush_entrys that were successfully processed
532 * as a group so that they will be free'd by the caller.
534 list_splice_init(&tmp_list
, flush_list
);
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.
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.
554 * Returns: 0 on success, < 0 on failure
556 static int userspace_flush(struct dm_dirty_log
*log
)
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
;
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
);
572 mark_list_is_empty
= list_empty(&mark_list
);
573 clear_list_is_empty
= list_empty(&clear_list
);
575 if (mark_list_is_empty
&& clear_list_is_empty
)
578 r
= flush_by_group(lc
, &clear_list
, 0);
582 if (!lc
->integrated_flush
) {
583 r
= flush_by_group(lc
, &mark_list
, 0);
586 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_FLUSH
,
587 NULL
, 0, NULL
, NULL
);
592 * Send integrated flush request with mark_list as payload.
594 r
= flush_by_group(lc
, &mark_list
, 1);
598 if (mark_list_is_empty
&& !atomic_read(&lc
->sched_flush
)) {
600 * When there are only clear region requests,
601 * we schedule a flush in the future.
603 queue_delayed_work(lc
->dmlog_wq
, &lc
->flush_log_work
, 3 * HZ
);
604 atomic_set(&lc
->sched_flush
, 1);
607 * Cancel pending flush because we
608 * have already flushed in mark_region.
610 cancel_delayed_work(&lc
->flush_log_work
);
611 atomic_set(&lc
->sched_flush
, 0);
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.
620 list_for_each_entry_safe(fe
, tmp_fe
, &mark_list
, list
) {
622 mempool_free(fe
, flush_entry_pool
);
624 list_for_each_entry_safe(fe
, tmp_fe
, &clear_list
, list
) {
626 mempool_free(fe
, flush_entry_pool
);
630 dm_table_event(lc
->ti
->table
);
636 * userspace_mark_region
638 * This function should avoid blocking unless absolutely required.
639 * (Memory allocation is valid for blocking.)
641 static void userspace_mark_region(struct dm_dirty_log
*log
, region_t region
)
644 struct log_c
*lc
= log
->context
;
645 struct flush_entry
*fe
;
647 /* Wait for an allocation, but _never_ fail */
648 fe
= mempool_alloc(flush_entry_pool
, GFP_NOIO
);
651 spin_lock_irqsave(&lc
->flush_lock
, flags
);
652 fe
->type
= DM_ULOG_MARK_REGION
;
654 list_add(&fe
->list
, &lc
->mark_list
);
655 spin_unlock_irqrestore(&lc
->flush_lock
, flags
);
661 * userspace_clear_region
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.
670 static void userspace_clear_region(struct dm_dirty_log
*log
, region_t region
)
673 struct log_c
*lc
= log
->context
;
674 struct flush_entry
*fe
;
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.
682 fe
= mempool_alloc(flush_entry_pool
, GFP_ATOMIC
);
684 DMERR("Failed to allocate memory to clear region.");
688 spin_lock_irqsave(&lc
->flush_lock
, flags
);
689 fe
->type
= DM_ULOG_CLEAR_REGION
;
691 list_add(&fe
->list
, &lc
->clear_list
);
692 spin_unlock_irqrestore(&lc
->flush_lock
, flags
);
698 * userspace_get_resync_work
700 * Get a region that needs recovery. It is valid to return
701 * an error for this function.
703 * Returns: 1 if region filled, 0 if no work, <0 on error
705 static int userspace_get_resync_work(struct dm_dirty_log
*log
, region_t
*region
)
709 struct log_c
*lc
= log
->context
;
711 int64_t i
; /* 64-bit for mix arch compatibility */
715 if (lc
->in_sync_hint
>= lc
->region_count
)
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
);
723 return (r
) ? r
: (int)pkg
.i
;
727 * userspace_set_region_sync
729 * Set the sync status of a given region. This function
732 static void userspace_set_region_sync(struct dm_dirty_log
*log
,
733 region_t region
, int in_sync
)
736 struct log_c
*lc
= log
->context
;
743 pkg
.i
= (int64_t)in_sync
;
745 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_SET_REGION_SYNC
,
746 (char *)&pkg
, sizeof(pkg
), NULL
, NULL
);
749 * It would be nice to be able to report failures.
750 * However, it is easy emough to detect and resolve.
756 * userspace_get_sync_count
758 * If there is any sort of failure when consulting the server,
759 * we assume that the sync count is zero.
761 * Returns: sync count on success, 0 on failure
763 static region_t
userspace_get_sync_count(struct dm_dirty_log
*log
)
768 struct log_c
*lc
= log
->context
;
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
);
777 if (sync_count
>= lc
->region_count
)
778 lc
->in_sync_hint
= lc
->region_count
;
780 return (region_t
)sync_count
;
786 * Returns: amount of space consumed
788 static int userspace_status(struct dm_dirty_log
*log
, status_type_t status_type
,
789 char *result
, unsigned maxlen
)
793 size_t sz
= (size_t)maxlen
;
794 struct log_c
*lc
= log
->context
;
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
);
803 DMEMIT("%s 1 COM_FAILURE", log
->type
->name
);
806 case STATUSTYPE_TABLE
:
808 table_args
= strchr(lc
->usr_argv_str
, ' ');
809 BUG_ON(!table_args
); /* There will always be a ' ' */
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
);
818 return (r
) ? 0 : (int)sz
;
822 * userspace_is_remote_recovering
824 * Returns: 1 if region recovering, 0 otherwise
826 static int userspace_is_remote_recovering(struct dm_dirty_log
*log
,
830 uint64_t region64
= region
;
831 struct log_c
*lc
= log
->context
;
832 static unsigned long long limit
;
834 int64_t is_recovering
;
835 uint64_t in_sync_hint
;
837 size_t rdata_size
= sizeof(pkg
);
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.)
846 if (region
< lc
->in_sync_hint
)
848 else if (jiffies
< limit
)
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
);
858 lc
->in_sync_hint
= pkg
.in_sync_hint
;
860 return (int)pkg
.is_recovering
;
863 static struct dm_dirty_log_type _userspace_type
= {
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
,
884 static int __init
userspace_dirty_log_init(void)
888 flush_entry_pool
= mempool_create(100, flush_entry_alloc
,
889 flush_entry_free
, NULL
);
891 if (!flush_entry_pool
) {
892 DMWARN("Unable to create flush_entry_pool: No memory.");
896 r
= dm_ulog_tfr_init();
898 DMWARN("Unable to initialize userspace log communications");
899 mempool_destroy(flush_entry_pool
);
903 r
= dm_dirty_log_type_register(&_userspace_type
);
905 DMWARN("Couldn't register userspace dirty log type");
907 mempool_destroy(flush_entry_pool
);
911 DMINFO("version " DM_LOG_USERSPACE_VSN
" loaded");
915 static void __exit
userspace_dirty_log_exit(void)
917 dm_dirty_log_type_unregister(&_userspace_type
);
919 mempool_destroy(flush_entry_pool
);
921 DMINFO("version " DM_LOG_USERSPACE_VSN
" unloaded");
925 module_init(userspace_dirty_log_init
);
926 module_exit(userspace_dirty_log_exit
);
928 MODULE_DESCRIPTION(DM_NAME
" userspace dirty log link");
929 MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>");
930 MODULE_LICENSE("GPL");