Merge branch 'akpm'
[linux-2.6/next.git] / drivers / md / dm-log-userspace-base.c
blob196246ef39cbbff10bb2c671dcc4c21e108b02db
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 uint32_t region_size;
35 region_t region_count;
36 uint64_t luid;
37 char uuid[DM_UUID_LEN];
39 char *usr_argv_str;
40 uint32_t usr_argc;
43 * in_sync_hint gets set when doing is_remote_recovering. It
44 * represents the first region that needs recovery. IOW, the
45 * first zero bit of sync_bits. This can be useful for to limit
46 * traffic for calls like is_remote_recovering and get_resync_work,
47 * but be take care in its use for anything else.
49 uint64_t in_sync_hint;
52 * Mark and clear requests are held until a flush is issued
53 * so that we can group, and thereby limit, the amount of
54 * network traffic between kernel and userspace. The 'flush_lock'
55 * is used to protect these lists.
57 spinlock_t flush_lock;
58 struct list_head mark_list;
59 struct list_head clear_list;
62 static mempool_t *flush_entry_pool;
64 static void *flush_entry_alloc(gfp_t gfp_mask, void *pool_data)
66 return kmalloc(sizeof(struct flush_entry), gfp_mask);
69 static void flush_entry_free(void *element, void *pool_data)
71 kfree(element);
74 static int userspace_do_request(struct log_c *lc, const char *uuid,
75 int request_type, char *data, size_t data_size,
76 char *rdata, size_t *rdata_size)
78 int r;
81 * If the server isn't there, -ESRCH is returned,
82 * and we must keep trying until the server is
83 * restored.
85 retry:
86 r = dm_consult_userspace(uuid, lc->luid, request_type, data,
87 data_size, rdata, rdata_size);
89 if (r != -ESRCH)
90 return r;
92 DMERR(" Userspace log server not found.");
93 while (1) {
94 set_current_state(TASK_INTERRUPTIBLE);
95 schedule_timeout(2*HZ);
96 DMWARN("Attempting to contact userspace log server...");
97 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR,
98 lc->usr_argv_str,
99 strlen(lc->usr_argv_str) + 1,
100 NULL, NULL);
101 if (!r)
102 break;
104 DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
105 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL,
106 0, NULL, NULL);
107 if (!r)
108 goto retry;
110 DMERR("Error trying to resume userspace log: %d", r);
112 return -ESRCH;
115 static int build_constructor_string(struct dm_target *ti,
116 unsigned argc, char **argv,
117 char **ctr_str)
119 int i, str_size;
120 char *str = NULL;
122 *ctr_str = NULL;
124 for (i = 0, str_size = 0; i < argc; i++)
125 str_size += strlen(argv[i]) + 1; /* +1 for space between args */
127 str_size += 20; /* Max number of chars in a printed u64 number */
129 str = kzalloc(str_size, GFP_KERNEL);
130 if (!str) {
131 DMWARN("Unable to allocate memory for constructor string");
132 return -ENOMEM;
135 str_size = sprintf(str, "%llu", (unsigned long long)ti->len);
136 for (i = 0; i < argc; i++)
137 str_size += sprintf(str + str_size, " %s", argv[i]);
139 *ctr_str = str;
140 return str_size;
144 * userspace_ctr
146 * argv contains:
147 * <UUID> <other args>
148 * Where 'other args' is the userspace implementation specific log
149 * arguments. An example might be:
150 * <UUID> clustered_disk <arg count> <log dev> <region_size> [[no]sync]
152 * So, this module will strip off the <UUID> for identification purposes
153 * when communicating with userspace about a log; but will pass on everything
154 * else.
156 static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
157 unsigned argc, char **argv)
159 int r = 0;
160 int str_size;
161 char *ctr_str = NULL;
162 struct log_c *lc = NULL;
163 uint64_t rdata;
164 size_t rdata_size = sizeof(rdata);
166 if (argc < 3) {
167 DMWARN("Too few arguments to userspace dirty log");
168 return -EINVAL;
171 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
172 if (!lc) {
173 DMWARN("Unable to allocate userspace log context.");
174 return -ENOMEM;
177 /* The ptr value is sufficient for local unique id */
178 lc->luid = (unsigned long)lc;
180 lc->ti = ti;
182 if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
183 DMWARN("UUID argument too long.");
184 kfree(lc);
185 return -EINVAL;
188 strncpy(lc->uuid, argv[0], DM_UUID_LEN);
189 spin_lock_init(&lc->flush_lock);
190 INIT_LIST_HEAD(&lc->mark_list);
191 INIT_LIST_HEAD(&lc->clear_list);
193 str_size = build_constructor_string(ti, argc - 1, argv + 1, &ctr_str);
194 if (str_size < 0) {
195 kfree(lc);
196 return str_size;
199 /* Send table string */
200 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
201 ctr_str, str_size, NULL, NULL);
203 if (r < 0) {
204 if (r == -ESRCH)
205 DMERR("Userspace log server not found");
206 else
207 DMERR("Userspace log server failed to create log");
208 goto out;
211 /* Since the region size does not change, get it now */
212 rdata_size = sizeof(rdata);
213 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
214 NULL, 0, (char *)&rdata, &rdata_size);
216 if (r) {
217 DMERR("Failed to get region size of dirty log");
218 goto out;
221 lc->region_size = (uint32_t)rdata;
222 lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
224 out:
225 if (r) {
226 kfree(lc);
227 kfree(ctr_str);
228 } else {
229 lc->usr_argv_str = ctr_str;
230 lc->usr_argc = argc;
231 log->context = lc;
234 return r;
237 static void userspace_dtr(struct dm_dirty_log *log)
239 struct log_c *lc = log->context;
241 (void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
242 NULL, 0,
243 NULL, NULL);
245 kfree(lc->usr_argv_str);
246 kfree(lc);
248 return;
251 static int userspace_presuspend(struct dm_dirty_log *log)
253 int r;
254 struct log_c *lc = log->context;
256 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
257 NULL, 0,
258 NULL, NULL);
260 return r;
263 static int userspace_postsuspend(struct dm_dirty_log *log)
265 int r;
266 struct log_c *lc = log->context;
268 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
269 NULL, 0,
270 NULL, NULL);
272 return r;
275 static int userspace_resume(struct dm_dirty_log *log)
277 int r;
278 struct log_c *lc = log->context;
280 lc->in_sync_hint = 0;
281 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
282 NULL, 0,
283 NULL, NULL);
285 return r;
288 static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
290 struct log_c *lc = log->context;
292 return lc->region_size;
296 * userspace_is_clean
298 * Check whether a region is clean. If there is any sort of
299 * failure when consulting the server, we return not clean.
301 * Returns: 1 if clean, 0 otherwise
303 static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
305 int r;
306 uint64_t region64 = (uint64_t)region;
307 int64_t is_clean;
308 size_t rdata_size;
309 struct log_c *lc = log->context;
311 rdata_size = sizeof(is_clean);
312 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
313 (char *)&region64, sizeof(region64),
314 (char *)&is_clean, &rdata_size);
316 return (r) ? 0 : (int)is_clean;
320 * userspace_in_sync
322 * Check if the region is in-sync. If there is any sort
323 * of failure when consulting the server, we assume that
324 * the region is not in sync.
326 * If 'can_block' is set, return immediately
328 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
330 static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
331 int can_block)
333 int r;
334 uint64_t region64 = region;
335 int64_t in_sync;
336 size_t rdata_size;
337 struct log_c *lc = log->context;
340 * We can never respond directly - even if in_sync_hint is
341 * set. This is because another machine could see a device
342 * failure and mark the region out-of-sync. If we don't go
343 * to userspace to ask, we might think the region is in-sync
344 * and allow a read to pick up data that is stale. (This is
345 * very unlikely if a device actually fails; but it is very
346 * likely if a connection to one device from one machine fails.)
348 * There still might be a problem if the mirror caches the region
349 * state as in-sync... but then this call would not be made. So,
350 * that is a mirror problem.
352 if (!can_block)
353 return -EWOULDBLOCK;
355 rdata_size = sizeof(in_sync);
356 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
357 (char *)&region64, sizeof(region64),
358 (char *)&in_sync, &rdata_size);
359 return (r) ? 0 : (int)in_sync;
362 static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
364 int r = 0;
365 struct flush_entry *fe;
367 list_for_each_entry(fe, flush_list, list) {
368 r = userspace_do_request(lc, lc->uuid, fe->type,
369 (char *)&fe->region,
370 sizeof(fe->region),
371 NULL, NULL);
372 if (r)
373 break;
376 return r;
379 static int flush_by_group(struct log_c *lc, struct list_head *flush_list)
381 int r = 0;
382 int count;
383 uint32_t type = 0;
384 struct flush_entry *fe, *tmp_fe;
385 LIST_HEAD(tmp_list);
386 uint64_t group[MAX_FLUSH_GROUP_COUNT];
389 * Group process the requests
391 while (!list_empty(flush_list)) {
392 count = 0;
394 list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
395 group[count] = fe->region;
396 count++;
398 list_move(&fe->list, &tmp_list);
400 type = fe->type;
401 if (count >= MAX_FLUSH_GROUP_COUNT)
402 break;
405 r = userspace_do_request(lc, lc->uuid, type,
406 (char *)(group),
407 count * sizeof(uint64_t),
408 NULL, NULL);
409 if (r) {
410 /* Group send failed. Attempt one-by-one. */
411 list_splice_init(&tmp_list, flush_list);
412 r = flush_one_by_one(lc, flush_list);
413 break;
418 * Must collect flush_entrys that were successfully processed
419 * as a group so that they will be free'd by the caller.
421 list_splice_init(&tmp_list, flush_list);
423 return r;
427 * userspace_flush
429 * This function is ok to block.
430 * The flush happens in two stages. First, it sends all
431 * clear/mark requests that are on the list. Then it
432 * tells the server to commit them. This gives the
433 * server a chance to optimise the commit, instead of
434 * doing it for every request.
436 * Additionally, we could implement another thread that
437 * sends the requests up to the server - reducing the
438 * load on flush. Then the flush would have less in
439 * the list and be responsible for the finishing commit.
441 * Returns: 0 on success, < 0 on failure
443 static int userspace_flush(struct dm_dirty_log *log)
445 int r = 0;
446 unsigned long flags;
447 struct log_c *lc = log->context;
448 LIST_HEAD(mark_list);
449 LIST_HEAD(clear_list);
450 struct flush_entry *fe, *tmp_fe;
452 spin_lock_irqsave(&lc->flush_lock, flags);
453 list_splice_init(&lc->mark_list, &mark_list);
454 list_splice_init(&lc->clear_list, &clear_list);
455 spin_unlock_irqrestore(&lc->flush_lock, flags);
457 if (list_empty(&mark_list) && list_empty(&clear_list))
458 return 0;
460 r = flush_by_group(lc, &mark_list);
461 if (r)
462 goto fail;
464 r = flush_by_group(lc, &clear_list);
465 if (r)
466 goto fail;
468 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
469 NULL, 0, NULL, NULL);
471 fail:
473 * We can safely remove these entries, even if failure.
474 * Calling code will receive an error and will know that
475 * the log facility has failed.
477 list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) {
478 list_del(&fe->list);
479 mempool_free(fe, flush_entry_pool);
481 list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) {
482 list_del(&fe->list);
483 mempool_free(fe, flush_entry_pool);
486 if (r)
487 dm_table_event(lc->ti->table);
489 return r;
493 * userspace_mark_region
495 * This function should avoid blocking unless absolutely required.
496 * (Memory allocation is valid for blocking.)
498 static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
500 unsigned long flags;
501 struct log_c *lc = log->context;
502 struct flush_entry *fe;
504 /* Wait for an allocation, but _never_ fail */
505 fe = mempool_alloc(flush_entry_pool, GFP_NOIO);
506 BUG_ON(!fe);
508 spin_lock_irqsave(&lc->flush_lock, flags);
509 fe->type = DM_ULOG_MARK_REGION;
510 fe->region = region;
511 list_add(&fe->list, &lc->mark_list);
512 spin_unlock_irqrestore(&lc->flush_lock, flags);
514 return;
518 * userspace_clear_region
520 * This function must not block.
521 * So, the alloc can't block. In the worst case, it is ok to
522 * fail. It would simply mean we can't clear the region.
523 * Does nothing to current sync context, but does mean
524 * the region will be re-sync'ed on a reload of the mirror
525 * even though it is in-sync.
527 static void userspace_clear_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;
534 * If we fail to allocate, we skip the clearing of
535 * the region. This doesn't hurt us in any way, except
536 * to cause the region to be resync'ed when the
537 * device is activated next time.
539 fe = mempool_alloc(flush_entry_pool, GFP_ATOMIC);
540 if (!fe) {
541 DMERR("Failed to allocate memory to clear region.");
542 return;
545 spin_lock_irqsave(&lc->flush_lock, flags);
546 fe->type = DM_ULOG_CLEAR_REGION;
547 fe->region = region;
548 list_add(&fe->list, &lc->clear_list);
549 spin_unlock_irqrestore(&lc->flush_lock, flags);
551 return;
555 * userspace_get_resync_work
557 * Get a region that needs recovery. It is valid to return
558 * an error for this function.
560 * Returns: 1 if region filled, 0 if no work, <0 on error
562 static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
564 int r;
565 size_t rdata_size;
566 struct log_c *lc = log->context;
567 struct {
568 int64_t i; /* 64-bit for mix arch compatibility */
569 region_t r;
570 } pkg;
572 if (lc->in_sync_hint >= lc->region_count)
573 return 0;
575 rdata_size = sizeof(pkg);
576 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
577 NULL, 0,
578 (char *)&pkg, &rdata_size);
580 *region = pkg.r;
581 return (r) ? r : (int)pkg.i;
585 * userspace_set_region_sync
587 * Set the sync status of a given region. This function
588 * must not fail.
590 static void userspace_set_region_sync(struct dm_dirty_log *log,
591 region_t region, int in_sync)
593 int r;
594 struct log_c *lc = log->context;
595 struct {
596 region_t r;
597 int64_t i;
598 } pkg;
600 pkg.r = region;
601 pkg.i = (int64_t)in_sync;
603 r = userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
604 (char *)&pkg, sizeof(pkg),
605 NULL, NULL);
608 * It would be nice to be able to report failures.
609 * However, it is easy emough to detect and resolve.
611 return;
615 * userspace_get_sync_count
617 * If there is any sort of failure when consulting the server,
618 * we assume that the sync count is zero.
620 * Returns: sync count on success, 0 on failure
622 static region_t userspace_get_sync_count(struct dm_dirty_log *log)
624 int r;
625 size_t rdata_size;
626 uint64_t sync_count;
627 struct log_c *lc = log->context;
629 rdata_size = sizeof(sync_count);
630 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
631 NULL, 0,
632 (char *)&sync_count, &rdata_size);
634 if (r)
635 return 0;
637 if (sync_count >= lc->region_count)
638 lc->in_sync_hint = lc->region_count;
640 return (region_t)sync_count;
644 * userspace_status
646 * Returns: amount of space consumed
648 static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
649 char *result, unsigned maxlen)
651 int r = 0;
652 char *table_args;
653 size_t sz = (size_t)maxlen;
654 struct log_c *lc = log->context;
656 switch (status_type) {
657 case STATUSTYPE_INFO:
658 r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
659 NULL, 0,
660 result, &sz);
662 if (r) {
663 sz = 0;
664 DMEMIT("%s 1 COM_FAILURE", log->type->name);
666 break;
667 case STATUSTYPE_TABLE:
668 sz = 0;
669 table_args = strchr(lc->usr_argv_str, ' ');
670 BUG_ON(!table_args); /* There will always be a ' ' */
671 table_args++;
673 DMEMIT("%s %u %s %s ", log->type->name, lc->usr_argc,
674 lc->uuid, table_args);
675 break;
677 return (r) ? 0 : (int)sz;
681 * userspace_is_remote_recovering
683 * Returns: 1 if region recovering, 0 otherwise
685 static int userspace_is_remote_recovering(struct dm_dirty_log *log,
686 region_t region)
688 int r;
689 uint64_t region64 = region;
690 struct log_c *lc = log->context;
691 static unsigned long long limit;
692 struct {
693 int64_t is_recovering;
694 uint64_t in_sync_hint;
695 } pkg;
696 size_t rdata_size = sizeof(pkg);
699 * Once the mirror has been reported to be in-sync,
700 * it will never again ask for recovery work. So,
701 * we can safely say there is not a remote machine
702 * recovering if the device is in-sync. (in_sync_hint
703 * must be reset at resume time.)
705 if (region < lc->in_sync_hint)
706 return 0;
707 else if (jiffies < limit)
708 return 1;
710 limit = jiffies + (HZ / 4);
711 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
712 (char *)&region64, sizeof(region64),
713 (char *)&pkg, &rdata_size);
714 if (r)
715 return 1;
717 lc->in_sync_hint = pkg.in_sync_hint;
719 return (int)pkg.is_recovering;
722 static struct dm_dirty_log_type _userspace_type = {
723 .name = "userspace",
724 .module = THIS_MODULE,
725 .ctr = userspace_ctr,
726 .dtr = userspace_dtr,
727 .presuspend = userspace_presuspend,
728 .postsuspend = userspace_postsuspend,
729 .resume = userspace_resume,
730 .get_region_size = userspace_get_region_size,
731 .is_clean = userspace_is_clean,
732 .in_sync = userspace_in_sync,
733 .flush = userspace_flush,
734 .mark_region = userspace_mark_region,
735 .clear_region = userspace_clear_region,
736 .get_resync_work = userspace_get_resync_work,
737 .set_region_sync = userspace_set_region_sync,
738 .get_sync_count = userspace_get_sync_count,
739 .status = userspace_status,
740 .is_remote_recovering = userspace_is_remote_recovering,
743 static int __init userspace_dirty_log_init(void)
745 int r = 0;
747 flush_entry_pool = mempool_create(100, flush_entry_alloc,
748 flush_entry_free, NULL);
750 if (!flush_entry_pool) {
751 DMWARN("Unable to create flush_entry_pool: No memory.");
752 return -ENOMEM;
755 r = dm_ulog_tfr_init();
756 if (r) {
757 DMWARN("Unable to initialize userspace log communications");
758 mempool_destroy(flush_entry_pool);
759 return r;
762 r = dm_dirty_log_type_register(&_userspace_type);
763 if (r) {
764 DMWARN("Couldn't register userspace dirty log type");
765 dm_ulog_tfr_exit();
766 mempool_destroy(flush_entry_pool);
767 return r;
770 DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
771 return 0;
774 static void __exit userspace_dirty_log_exit(void)
776 dm_dirty_log_type_unregister(&_userspace_type);
777 dm_ulog_tfr_exit();
778 mempool_destroy(flush_entry_pool);
780 DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
781 return;
784 module_init(userspace_dirty_log_init);
785 module_exit(userspace_dirty_log_exit);
787 MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
788 MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>");
789 MODULE_LICENSE("GPL");