1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2006-2009 Red Hat, Inc.
5 * This file is released under the LGPL.
9 #include <linux/slab.h>
10 #include <linux/jiffies.h>
11 #include <linux/dm-dirty-log.h>
12 #include <linux/device-mapper.h>
13 #include <linux/dm-log-userspace.h>
14 #include <linux/module.h>
15 #include <linux/workqueue.h>
17 #include "dm-log-userspace-transfer.h"
19 #define DM_LOG_USERSPACE_VSN "1.3.0"
21 #define FLUSH_ENTRY_POOL_SIZE 16
23 struct dm_dirty_log_flush_entry
{
26 struct list_head list
;
30 * This limit on the number of mark and clear request is, to a degree,
31 * arbitrary. However, there is some basis for the choice in the limits
32 * imposed on the size of data payload by dm-log-userspace-transfer.c:
33 * dm_consult_userspace().
35 #define MAX_FLUSH_GROUP_COUNT 32
39 struct dm_dev
*log_dev
;
45 region_t region_count
;
47 char uuid
[DM_UUID_LEN
];
50 * Mark and clear requests are held until a flush is issued
51 * so that we can group, and thereby limit, the amount of
52 * network traffic between kernel and userspace. The 'flush_lock'
53 * is used to protect these lists.
55 spinlock_t flush_lock
;
56 struct list_head mark_list
;
57 struct list_head clear_list
;
60 * in_sync_hint gets set when doing is_remote_recovering. It
61 * represents the first region that needs recovery. IOW, the
62 * first zero bit of sync_bits. This can be useful for to limit
63 * traffic for calls like is_remote_recovering and get_resync_work,
64 * but be take care in its use for anything else.
66 uint64_t in_sync_hint
;
69 * Workqueue for flush of clear region requests.
71 struct workqueue_struct
*dmlog_wq
;
72 struct delayed_work flush_log_work
;
76 * Combine userspace flush and mark requests for efficiency.
78 uint32_t integrated_flush
;
80 mempool_t flush_entry_pool
;
83 static struct kmem_cache
*_flush_entry_cache
;
85 static int userspace_do_request(struct log_c
*lc
, const char *uuid
,
86 int request_type
, char *data
, size_t data_size
,
87 char *rdata
, size_t *rdata_size
)
92 * If the server isn't there, -ESRCH is returned,
93 * and we must keep trying until the server is
97 r
= dm_consult_userspace(uuid
, lc
->luid
, request_type
, data
,
98 data_size
, rdata
, rdata_size
);
103 DMERR(" Userspace log server not found.");
105 set_current_state(TASK_INTERRUPTIBLE
);
106 schedule_timeout(2*HZ
);
107 DMWARN("Attempting to contact userspace log server...");
108 r
= dm_consult_userspace(uuid
, lc
->luid
, DM_ULOG_CTR
,
110 strlen(lc
->usr_argv_str
) + 1,
115 DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
116 r
= dm_consult_userspace(uuid
, lc
->luid
, DM_ULOG_RESUME
, NULL
,
121 DMERR("Error trying to resume userspace log: %d", r
);
126 static int build_constructor_string(struct dm_target
*ti
,
127 unsigned int argc
, char **argv
,
136 * Determine overall size of the string.
138 for (i
= 0, str_size
= 0; i
< argc
; i
++)
139 str_size
+= strlen(argv
[i
]) + 1; /* +1 for space between args */
141 str_size
+= 20; /* Max number of chars in a printed u64 number */
143 str
= kzalloc(str_size
, GFP_KERNEL
);
145 DMWARN("Unable to allocate memory for constructor string");
149 str_size
= sprintf(str
, "%llu", (unsigned long long)ti
->len
);
150 for (i
= 0; i
< argc
; i
++)
151 str_size
+= sprintf(str
+ str_size
, " %s", argv
[i
]);
157 static void do_flush(struct work_struct
*work
)
160 struct log_c
*lc
= container_of(work
, struct log_c
, flush_log_work
.work
);
162 atomic_set(&lc
->sched_flush
, 0);
164 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_FLUSH
, NULL
, 0, NULL
, NULL
);
167 dm_table_event(lc
->ti
->table
);
174 * <UUID> [integrated_flush] <other args>
175 * Where 'other args' are the userspace implementation-specific log
179 * <UUID> [integrated_flush] clustered-disk <arg count> <log dev>
180 * <region_size> [[no]sync]
182 * This module strips off the <UUID> and uses it for identification
183 * purposes when communicating with userspace about a log.
185 * If integrated_flush is defined, the kernel combines flush
188 * The rest of the line, beginning with 'clustered-disk', is passed
189 * to the userspace ctr function.
191 static int userspace_ctr(struct dm_dirty_log
*log
, struct dm_target
*ti
,
192 unsigned int argc
, char **argv
)
196 char *ctr_str
= NULL
;
197 struct log_c
*lc
= NULL
;
199 size_t rdata_size
= sizeof(rdata
);
200 char *devices_rdata
= NULL
;
201 size_t devices_rdata_size
= DM_NAME_LEN
;
204 DMWARN("Too few arguments to userspace dirty log");
208 lc
= kzalloc(sizeof(*lc
), GFP_KERNEL
);
210 DMWARN("Unable to allocate userspace log context.");
214 /* The ptr value is sufficient for local unique id */
215 lc
->luid
= (unsigned long)lc
;
219 if (strlen(argv
[0]) > (DM_UUID_LEN
- 1)) {
220 DMWARN("UUID argument too long.");
227 strscpy(lc
->uuid
, argv
[0], sizeof(lc
->uuid
));
230 spin_lock_init(&lc
->flush_lock
);
231 INIT_LIST_HEAD(&lc
->mark_list
);
232 INIT_LIST_HEAD(&lc
->clear_list
);
234 if (!strcasecmp(argv
[0], "integrated_flush")) {
235 lc
->integrated_flush
= 1;
240 str_size
= build_constructor_string(ti
, argc
, argv
, &ctr_str
);
246 devices_rdata
= kzalloc(devices_rdata_size
, GFP_KERNEL
);
247 if (!devices_rdata
) {
248 DMERR("Failed to allocate memory for device information");
253 r
= mempool_init_slab_pool(&lc
->flush_entry_pool
, FLUSH_ENTRY_POOL_SIZE
,
256 DMERR("Failed to create flush_entry_pool");
261 * Send table string and get back any opened device.
263 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_CTR
,
265 devices_rdata
, &devices_rdata_size
);
269 DMERR("Userspace log server not found");
271 DMERR("Userspace log server failed to create log");
275 /* Since the region size does not change, get it now */
276 rdata_size
= sizeof(rdata
);
277 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_GET_REGION_SIZE
,
278 NULL
, 0, (char *)&rdata
, &rdata_size
);
281 DMERR("Failed to get region size of dirty log");
285 lc
->region_size
= (uint32_t)rdata
;
286 lc
->region_count
= dm_sector_div_up(ti
->len
, lc
->region_size
);
288 if (devices_rdata_size
) {
289 if (devices_rdata
[devices_rdata_size
- 1] != '\0') {
290 DMERR("DM_ULOG_CTR device return string not properly terminated");
294 r
= dm_get_device(ti
, devices_rdata
,
295 dm_table_get_mode(ti
->table
), &lc
->log_dev
);
297 DMERR("Failed to register %s with device-mapper",
301 if (lc
->integrated_flush
) {
302 lc
->dmlog_wq
= alloc_workqueue("dmlogd", WQ_MEM_RECLAIM
, 0);
304 DMERR("couldn't start dmlogd");
309 INIT_DELAYED_WORK(&lc
->flush_log_work
, do_flush
);
310 atomic_set(&lc
->sched_flush
, 0);
314 kfree(devices_rdata
);
316 mempool_exit(&lc
->flush_entry_pool
);
320 lc
->usr_argv_str
= ctr_str
;
327 static void userspace_dtr(struct dm_dirty_log
*log
)
329 struct log_c
*lc
= log
->context
;
331 if (lc
->integrated_flush
) {
332 /* flush workqueue */
333 if (atomic_read(&lc
->sched_flush
))
334 flush_delayed_work(&lc
->flush_log_work
);
336 destroy_workqueue(lc
->dmlog_wq
);
339 (void) dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_DTR
,
340 NULL
, 0, NULL
, NULL
);
343 dm_put_device(lc
->ti
, lc
->log_dev
);
345 mempool_exit(&lc
->flush_entry_pool
);
347 kfree(lc
->usr_argv_str
);
351 static int userspace_presuspend(struct dm_dirty_log
*log
)
354 struct log_c
*lc
= log
->context
;
356 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_PRESUSPEND
,
357 NULL
, 0, NULL
, NULL
);
362 static int userspace_postsuspend(struct dm_dirty_log
*log
)
365 struct log_c
*lc
= log
->context
;
368 * Run planned flush earlier.
370 if (lc
->integrated_flush
&& atomic_read(&lc
->sched_flush
))
371 flush_delayed_work(&lc
->flush_log_work
);
373 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_POSTSUSPEND
,
374 NULL
, 0, NULL
, NULL
);
379 static int userspace_resume(struct dm_dirty_log
*log
)
382 struct log_c
*lc
= log
->context
;
384 lc
->in_sync_hint
= 0;
385 r
= dm_consult_userspace(lc
->uuid
, lc
->luid
, DM_ULOG_RESUME
,
386 NULL
, 0, NULL
, NULL
);
391 static uint32_t userspace_get_region_size(struct dm_dirty_log
*log
)
393 struct log_c
*lc
= log
->context
;
395 return lc
->region_size
;
401 * Check whether a region is clean. If there is any sort of
402 * failure when consulting the server, we return not clean.
404 * Returns: 1 if clean, 0 otherwise
406 static int userspace_is_clean(struct dm_dirty_log
*log
, region_t region
)
409 uint64_t region64
= (uint64_t)region
;
412 struct log_c
*lc
= log
->context
;
414 rdata_size
= sizeof(is_clean
);
415 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_IS_CLEAN
,
416 (char *)®ion64
, sizeof(region64
),
417 (char *)&is_clean
, &rdata_size
);
419 return (r
) ? 0 : (int)is_clean
;
425 * Check if the region is in-sync. If there is any sort
426 * of failure when consulting the server, we assume that
427 * the region is not in sync.
429 * If 'can_block' is set, return immediately
431 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
433 static int userspace_in_sync(struct dm_dirty_log
*log
, region_t region
,
437 uint64_t region64
= region
;
440 struct log_c
*lc
= log
->context
;
443 * We can never respond directly - even if in_sync_hint is
444 * set. This is because another machine could see a device
445 * failure and mark the region out-of-sync. If we don't go
446 * to userspace to ask, we might think the region is in-sync
447 * and allow a read to pick up data that is stale. (This is
448 * very unlikely if a device actually fails; but it is very
449 * likely if a connection to one device from one machine fails.)
451 * There still might be a problem if the mirror caches the region
452 * state as in-sync... but then this call would not be made. So,
453 * that is a mirror problem.
458 rdata_size
= sizeof(in_sync
);
459 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_IN_SYNC
,
460 (char *)®ion64
, sizeof(region64
),
461 (char *)&in_sync
, &rdata_size
);
462 return (r
) ? 0 : (int)in_sync
;
465 static int flush_one_by_one(struct log_c
*lc
, struct list_head
*flush_list
)
468 struct dm_dirty_log_flush_entry
*fe
;
470 list_for_each_entry(fe
, flush_list
, list
) {
471 r
= userspace_do_request(lc
, lc
->uuid
, fe
->type
,
482 static int flush_by_group(struct log_c
*lc
, struct list_head
*flush_list
,
483 int flush_with_payload
)
488 struct dm_dirty_log_flush_entry
*fe
, *tmp_fe
;
490 uint64_t group
[MAX_FLUSH_GROUP_COUNT
];
493 * Group process the requests
495 while (!list_empty(flush_list
)) {
498 list_for_each_entry_safe(fe
, tmp_fe
, flush_list
, list
) {
499 group
[count
] = fe
->region
;
502 list_move(&fe
->list
, &tmp_list
);
505 if (count
>= MAX_FLUSH_GROUP_COUNT
)
509 if (flush_with_payload
) {
510 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_FLUSH
,
512 count
* sizeof(uint64_t),
515 * Integrated flush failed.
520 r
= userspace_do_request(lc
, lc
->uuid
, type
,
522 count
* sizeof(uint64_t),
526 * Group send failed. Attempt one-by-one.
528 list_splice_init(&tmp_list
, flush_list
);
529 r
= flush_one_by_one(lc
, flush_list
);
536 * Must collect flush_entrys that were successfully processed
537 * as a group so that they will be free'd by the caller.
539 list_splice_init(&tmp_list
, flush_list
);
547 * This function is ok to block.
548 * The flush happens in two stages. First, it sends all
549 * clear/mark requests that are on the list. Then it
550 * tells the server to commit them. This gives the
551 * server a chance to optimise the commit, instead of
552 * doing it for every request.
554 * Additionally, we could implement another thread that
555 * sends the requests up to the server - reducing the
556 * load on flush. Then the flush would have less in
557 * the list and be responsible for the finishing commit.
559 * Returns: 0 on success, < 0 on failure
561 static int userspace_flush(struct dm_dirty_log
*log
)
565 struct log_c
*lc
= log
->context
;
566 LIST_HEAD(mark_list
);
567 LIST_HEAD(clear_list
);
568 int mark_list_is_empty
;
569 int clear_list_is_empty
;
570 struct dm_dirty_log_flush_entry
*fe
, *tmp_fe
;
571 mempool_t
*flush_entry_pool
= &lc
->flush_entry_pool
;
573 spin_lock_irqsave(&lc
->flush_lock
, flags
);
574 list_splice_init(&lc
->mark_list
, &mark_list
);
575 list_splice_init(&lc
->clear_list
, &clear_list
);
576 spin_unlock_irqrestore(&lc
->flush_lock
, flags
);
578 mark_list_is_empty
= list_empty(&mark_list
);
579 clear_list_is_empty
= list_empty(&clear_list
);
581 if (mark_list_is_empty
&& clear_list_is_empty
)
584 r
= flush_by_group(lc
, &clear_list
, 0);
588 if (!lc
->integrated_flush
) {
589 r
= flush_by_group(lc
, &mark_list
, 0);
592 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_FLUSH
,
593 NULL
, 0, NULL
, NULL
);
598 * Send integrated flush request with mark_list as payload.
600 r
= flush_by_group(lc
, &mark_list
, 1);
604 if (mark_list_is_empty
&& !atomic_read(&lc
->sched_flush
)) {
606 * When there are only clear region requests,
607 * we schedule a flush in the future.
609 queue_delayed_work(lc
->dmlog_wq
, &lc
->flush_log_work
, 3 * HZ
);
610 atomic_set(&lc
->sched_flush
, 1);
613 * Cancel pending flush because we
614 * have already flushed in mark_region.
616 cancel_delayed_work(&lc
->flush_log_work
);
617 atomic_set(&lc
->sched_flush
, 0);
622 * We can safely remove these entries, even after failure.
623 * Calling code will receive an error and will know that
624 * the log facility has failed.
626 list_for_each_entry_safe(fe
, tmp_fe
, &mark_list
, list
) {
628 mempool_free(fe
, flush_entry_pool
);
630 list_for_each_entry_safe(fe
, tmp_fe
, &clear_list
, list
) {
632 mempool_free(fe
, flush_entry_pool
);
636 dm_table_event(lc
->ti
->table
);
642 * userspace_mark_region
644 * This function should avoid blocking unless absolutely required.
645 * (Memory allocation is valid for blocking.)
647 static void userspace_mark_region(struct dm_dirty_log
*log
, region_t region
)
650 struct log_c
*lc
= log
->context
;
651 struct dm_dirty_log_flush_entry
*fe
;
653 /* Wait for an allocation, but _never_ fail */
654 fe
= mempool_alloc(&lc
->flush_entry_pool
, GFP_NOIO
);
657 spin_lock_irqsave(&lc
->flush_lock
, flags
);
658 fe
->type
= DM_ULOG_MARK_REGION
;
660 list_add(&fe
->list
, &lc
->mark_list
);
661 spin_unlock_irqrestore(&lc
->flush_lock
, flags
);
665 * userspace_clear_region
667 * This function must not block.
668 * So, the alloc can't block. In the worst case, it is ok to
669 * fail. It would simply mean we can't clear the region.
670 * Does nothing to current sync context, but does mean
671 * the region will be re-sync'ed on a reload of the mirror
672 * even though it is in-sync.
674 static void userspace_clear_region(struct dm_dirty_log
*log
, region_t region
)
677 struct log_c
*lc
= log
->context
;
678 struct dm_dirty_log_flush_entry
*fe
;
681 * If we fail to allocate, we skip the clearing of
682 * the region. This doesn't hurt us in any way, except
683 * to cause the region to be resync'ed when the
684 * device is activated next time.
686 fe
= mempool_alloc(&lc
->flush_entry_pool
, GFP_ATOMIC
);
688 DMERR("Failed to allocate memory to clear region.");
692 spin_lock_irqsave(&lc
->flush_lock
, flags
);
693 fe
->type
= DM_ULOG_CLEAR_REGION
;
695 list_add(&fe
->list
, &lc
->clear_list
);
696 spin_unlock_irqrestore(&lc
->flush_lock
, flags
);
700 * userspace_get_resync_work
702 * Get a region that needs recovery. It is valid to return
703 * an error for this function.
705 * Returns: 1 if region filled, 0 if no work, <0 on error
707 static int userspace_get_resync_work(struct dm_dirty_log
*log
, region_t
*region
)
711 struct log_c
*lc
= log
->context
;
713 int64_t i
; /* 64-bit for mix arch compatibility */
717 if (lc
->in_sync_hint
>= lc
->region_count
)
720 rdata_size
= sizeof(pkg
);
721 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_GET_RESYNC_WORK
,
722 NULL
, 0, (char *)&pkg
, &rdata_size
);
725 return (r
) ? r
: (int)pkg
.i
;
729 * userspace_set_region_sync
731 * Set the sync status of a given region. This function
734 static void userspace_set_region_sync(struct dm_dirty_log
*log
,
735 region_t region
, int in_sync
)
737 struct log_c
*lc
= log
->context
;
744 pkg
.i
= (int64_t)in_sync
;
746 (void) 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 enough 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 int 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
);
821 return (r
) ? 0 : (int)sz
;
825 * userspace_is_remote_recovering
827 * Returns: 1 if region recovering, 0 otherwise
829 static int userspace_is_remote_recovering(struct dm_dirty_log
*log
,
833 uint64_t region64
= region
;
834 struct log_c
*lc
= log
->context
;
835 static unsigned long limit
;
837 int64_t is_recovering
;
838 uint64_t in_sync_hint
;
840 size_t rdata_size
= sizeof(pkg
);
843 * Once the mirror has been reported to be in-sync,
844 * it will never again ask for recovery work. So,
845 * we can safely say there is not a remote machine
846 * recovering if the device is in-sync. (in_sync_hint
847 * must be reset at resume time.)
849 if (region
< lc
->in_sync_hint
)
851 else if (time_after(limit
, jiffies
))
854 limit
= jiffies
+ (HZ
/ 4);
855 r
= userspace_do_request(lc
, lc
->uuid
, DM_ULOG_IS_REMOTE_RECOVERING
,
856 (char *)®ion64
, sizeof(region64
),
857 (char *)&pkg
, &rdata_size
);
861 lc
->in_sync_hint
= pkg
.in_sync_hint
;
863 return (int)pkg
.is_recovering
;
866 static struct dm_dirty_log_type _userspace_type
= {
868 .module
= THIS_MODULE
,
869 .ctr
= userspace_ctr
,
870 .dtr
= userspace_dtr
,
871 .presuspend
= userspace_presuspend
,
872 .postsuspend
= userspace_postsuspend
,
873 .resume
= userspace_resume
,
874 .get_region_size
= userspace_get_region_size
,
875 .is_clean
= userspace_is_clean
,
876 .in_sync
= userspace_in_sync
,
877 .flush
= userspace_flush
,
878 .mark_region
= userspace_mark_region
,
879 .clear_region
= userspace_clear_region
,
880 .get_resync_work
= userspace_get_resync_work
,
881 .set_region_sync
= userspace_set_region_sync
,
882 .get_sync_count
= userspace_get_sync_count
,
883 .status
= userspace_status
,
884 .is_remote_recovering
= userspace_is_remote_recovering
,
887 static int __init
userspace_dirty_log_init(void)
891 _flush_entry_cache
= KMEM_CACHE(dm_dirty_log_flush_entry
, 0);
892 if (!_flush_entry_cache
) {
893 DMWARN("Unable to create flush_entry_cache: No memory.");
897 r
= dm_ulog_tfr_init();
899 DMWARN("Unable to initialize userspace log communications");
900 kmem_cache_destroy(_flush_entry_cache
);
904 r
= dm_dirty_log_type_register(&_userspace_type
);
906 DMWARN("Couldn't register userspace dirty log type");
908 kmem_cache_destroy(_flush_entry_cache
);
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 kmem_cache_destroy(_flush_entry_cache
);
922 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@lists.linux.dev>");
930 MODULE_LICENSE("GPL");