HWPOISON: Add soft page offline support
[linux-2.6/linux-2.6-openrd.git] / drivers / md / dm-log-userspace-base.c
blob7ac2c1450d1004c4de300df3aab50fead2cfe5e0
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/dm-dirty-log.h>
9 #include <linux/device-mapper.h>
10 #include <linux/dm-log-userspace.h>
12 #include "dm-log-userspace-transfer.h"
14 struct flush_entry {
15 int type;
16 region_t region;
17 struct list_head list;
20 struct log_c {
21 struct dm_target *ti;
22 uint32_t region_size;
23 region_t region_count;
24 uint64_t luid;
25 char uuid[DM_UUID_LEN];
27 char *usr_argv_str;
28 uint32_t usr_argc;
31 * in_sync_hint gets set when doing is_remote_recovering. It
32 * represents the first region that needs recovery. IOW, the
33 * first zero bit of sync_bits. This can be useful for to limit
34 * traffic for calls like is_remote_recovering and get_resync_work,
35 * but be take care in its use for anything else.
37 uint64_t in_sync_hint;
39 spinlock_t flush_lock;
40 struct list_head flush_list; /* only for clear and mark requests */
43 static mempool_t *flush_entry_pool;
45 static void *flush_entry_alloc(gfp_t gfp_mask, void *pool_data)
47 return kmalloc(sizeof(struct flush_entry), gfp_mask);
50 static void flush_entry_free(void *element, void *pool_data)
52 kfree(element);
55 static int userspace_do_request(struct log_c *lc, const char *uuid,
56 int request_type, char *data, size_t data_size,
57 char *rdata, size_t *rdata_size)
59 int r;
62 * If the server isn't there, -ESRCH is returned,
63 * and we must keep trying until the server is
64 * restored.
66 retry:
67 r = dm_consult_userspace(uuid, lc->luid, request_type, data,
68 data_size, rdata, rdata_size);
70 if (r != -ESRCH)
71 return r;
73 DMERR(" Userspace log server not found.");
74 while (1) {
75 set_current_state(TASK_INTERRUPTIBLE);
76 schedule_timeout(2*HZ);
77 DMWARN("Attempting to contact userspace log server...");
78 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR,
79 lc->usr_argv_str,
80 strlen(lc->usr_argv_str) + 1,
81 NULL, NULL);
82 if (!r)
83 break;
85 DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
86 r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL,
87 0, NULL, NULL);
88 if (!r)
89 goto retry;
91 DMERR("Error trying to resume userspace log: %d", r);
93 return -ESRCH;
96 static int build_constructor_string(struct dm_target *ti,
97 unsigned argc, char **argv,
98 char **ctr_str)
100 int i, str_size;
101 char *str = NULL;
103 *ctr_str = NULL;
105 for (i = 0, str_size = 0; i < argc; i++)
106 str_size += strlen(argv[i]) + 1; /* +1 for space between args */
108 str_size += 20; /* Max number of chars in a printed u64 number */
110 str = kzalloc(str_size, GFP_KERNEL);
111 if (!str) {
112 DMWARN("Unable to allocate memory for constructor string");
113 return -ENOMEM;
116 str_size = sprintf(str, "%llu", (unsigned long long)ti->len);
117 for (i = 0; i < argc; i++)
118 str_size += sprintf(str + str_size, " %s", argv[i]);
120 *ctr_str = str;
121 return str_size;
125 * userspace_ctr
127 * argv contains:
128 * <UUID> <other args>
129 * Where 'other args' is the userspace implementation specific log
130 * arguments. An example might be:
131 * <UUID> clustered_disk <arg count> <log dev> <region_size> [[no]sync]
133 * So, this module will strip off the <UUID> for identification purposes
134 * when communicating with userspace about a log; but will pass on everything
135 * else.
137 static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
138 unsigned argc, char **argv)
140 int r = 0;
141 int str_size;
142 char *ctr_str = NULL;
143 struct log_c *lc = NULL;
144 uint64_t rdata;
145 size_t rdata_size = sizeof(rdata);
147 if (argc < 3) {
148 DMWARN("Too few arguments to userspace dirty log");
149 return -EINVAL;
152 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
153 if (!lc) {
154 DMWARN("Unable to allocate userspace log context.");
155 return -ENOMEM;
158 /* The ptr value is sufficient for local unique id */
159 lc->luid = (unsigned long)lc;
161 lc->ti = ti;
163 if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
164 DMWARN("UUID argument too long.");
165 kfree(lc);
166 return -EINVAL;
169 strncpy(lc->uuid, argv[0], DM_UUID_LEN);
170 spin_lock_init(&lc->flush_lock);
171 INIT_LIST_HEAD(&lc->flush_list);
173 str_size = build_constructor_string(ti, argc - 1, argv + 1, &ctr_str);
174 if (str_size < 0) {
175 kfree(lc);
176 return str_size;
179 /* Send table string */
180 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
181 ctr_str, str_size, NULL, NULL);
183 if (r == -ESRCH) {
184 DMERR("Userspace log server not found");
185 goto out;
188 /* Since the region size does not change, get it now */
189 rdata_size = sizeof(rdata);
190 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
191 NULL, 0, (char *)&rdata, &rdata_size);
193 if (r) {
194 DMERR("Failed to get region size of dirty log");
195 goto out;
198 lc->region_size = (uint32_t)rdata;
199 lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
201 out:
202 if (r) {
203 kfree(lc);
204 kfree(ctr_str);
205 } else {
206 lc->usr_argv_str = ctr_str;
207 lc->usr_argc = argc;
208 log->context = lc;
211 return r;
214 static void userspace_dtr(struct dm_dirty_log *log)
216 int r;
217 struct log_c *lc = log->context;
219 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
220 NULL, 0,
221 NULL, NULL);
223 kfree(lc->usr_argv_str);
224 kfree(lc);
226 return;
229 static int userspace_presuspend(struct dm_dirty_log *log)
231 int r;
232 struct log_c *lc = log->context;
234 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
235 NULL, 0,
236 NULL, NULL);
238 return r;
241 static int userspace_postsuspend(struct dm_dirty_log *log)
243 int r;
244 struct log_c *lc = log->context;
246 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
247 NULL, 0,
248 NULL, NULL);
250 return r;
253 static int userspace_resume(struct dm_dirty_log *log)
255 int r;
256 struct log_c *lc = log->context;
258 lc->in_sync_hint = 0;
259 r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
260 NULL, 0,
261 NULL, NULL);
263 return r;
266 static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
268 struct log_c *lc = log->context;
270 return lc->region_size;
274 * userspace_is_clean
276 * Check whether a region is clean. If there is any sort of
277 * failure when consulting the server, we return not clean.
279 * Returns: 1 if clean, 0 otherwise
281 static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
283 int r;
284 uint64_t region64 = (uint64_t)region;
285 int64_t is_clean;
286 size_t rdata_size;
287 struct log_c *lc = log->context;
289 rdata_size = sizeof(is_clean);
290 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
291 (char *)&region64, sizeof(region64),
292 (char *)&is_clean, &rdata_size);
294 return (r) ? 0 : (int)is_clean;
298 * userspace_in_sync
300 * Check if the region is in-sync. If there is any sort
301 * of failure when consulting the server, we assume that
302 * the region is not in sync.
304 * If 'can_block' is set, return immediately
306 * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
308 static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
309 int can_block)
311 int r;
312 uint64_t region64 = region;
313 int64_t in_sync;
314 size_t rdata_size;
315 struct log_c *lc = log->context;
318 * We can never respond directly - even if in_sync_hint is
319 * set. This is because another machine could see a device
320 * failure and mark the region out-of-sync. If we don't go
321 * to userspace to ask, we might think the region is in-sync
322 * and allow a read to pick up data that is stale. (This is
323 * very unlikely if a device actually fails; but it is very
324 * likely if a connection to one device from one machine fails.)
326 * There still might be a problem if the mirror caches the region
327 * state as in-sync... but then this call would not be made. So,
328 * that is a mirror problem.
330 if (!can_block)
331 return -EWOULDBLOCK;
333 rdata_size = sizeof(in_sync);
334 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
335 (char *)&region64, sizeof(region64),
336 (char *)&in_sync, &rdata_size);
337 return (r) ? 0 : (int)in_sync;
341 * userspace_flush
343 * This function is ok to block.
344 * The flush happens in two stages. First, it sends all
345 * clear/mark requests that are on the list. Then it
346 * tells the server to commit them. This gives the
347 * server a chance to optimise the commit, instead of
348 * doing it for every request.
350 * Additionally, we could implement another thread that
351 * sends the requests up to the server - reducing the
352 * load on flush. Then the flush would have less in
353 * the list and be responsible for the finishing commit.
355 * Returns: 0 on success, < 0 on failure
357 static int userspace_flush(struct dm_dirty_log *log)
359 int r = 0;
360 unsigned long flags;
361 struct log_c *lc = log->context;
362 LIST_HEAD(flush_list);
363 struct flush_entry *fe, *tmp_fe;
365 spin_lock_irqsave(&lc->flush_lock, flags);
366 list_splice_init(&lc->flush_list, &flush_list);
367 spin_unlock_irqrestore(&lc->flush_lock, flags);
369 if (list_empty(&flush_list))
370 return 0;
373 * FIXME: Count up requests, group request types,
374 * allocate memory to stick all requests in and
375 * send to server in one go. Failing the allocation,
376 * do it one by one.
379 list_for_each_entry(fe, &flush_list, list) {
380 r = userspace_do_request(lc, lc->uuid, fe->type,
381 (char *)&fe->region,
382 sizeof(fe->region),
383 NULL, NULL);
384 if (r)
385 goto fail;
388 r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
389 NULL, 0, NULL, NULL);
391 fail:
393 * We can safely remove these entries, even if failure.
394 * Calling code will receive an error and will know that
395 * the log facility has failed.
397 list_for_each_entry_safe(fe, tmp_fe, &flush_list, list) {
398 list_del(&fe->list);
399 mempool_free(fe, flush_entry_pool);
402 if (r)
403 dm_table_event(lc->ti->table);
405 return r;
409 * userspace_mark_region
411 * This function should avoid blocking unless absolutely required.
412 * (Memory allocation is valid for blocking.)
414 static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
416 unsigned long flags;
417 struct log_c *lc = log->context;
418 struct flush_entry *fe;
420 /* Wait for an allocation, but _never_ fail */
421 fe = mempool_alloc(flush_entry_pool, GFP_NOIO);
422 BUG_ON(!fe);
424 spin_lock_irqsave(&lc->flush_lock, flags);
425 fe->type = DM_ULOG_MARK_REGION;
426 fe->region = region;
427 list_add(&fe->list, &lc->flush_list);
428 spin_unlock_irqrestore(&lc->flush_lock, flags);
430 return;
434 * userspace_clear_region
436 * This function must not block.
437 * So, the alloc can't block. In the worst case, it is ok to
438 * fail. It would simply mean we can't clear the region.
439 * Does nothing to current sync context, but does mean
440 * the region will be re-sync'ed on a reload of the mirror
441 * even though it is in-sync.
443 static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
445 unsigned long flags;
446 struct log_c *lc = log->context;
447 struct flush_entry *fe;
450 * If we fail to allocate, we skip the clearing of
451 * the region. This doesn't hurt us in any way, except
452 * to cause the region to be resync'ed when the
453 * device is activated next time.
455 fe = mempool_alloc(flush_entry_pool, GFP_ATOMIC);
456 if (!fe) {
457 DMERR("Failed to allocate memory to clear region.");
458 return;
461 spin_lock_irqsave(&lc->flush_lock, flags);
462 fe->type = DM_ULOG_CLEAR_REGION;
463 fe->region = region;
464 list_add(&fe->list, &lc->flush_list);
465 spin_unlock_irqrestore(&lc->flush_lock, flags);
467 return;
471 * userspace_get_resync_work
473 * Get a region that needs recovery. It is valid to return
474 * an error for this function.
476 * Returns: 1 if region filled, 0 if no work, <0 on error
478 static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
480 int r;
481 size_t rdata_size;
482 struct log_c *lc = log->context;
483 struct {
484 int64_t i; /* 64-bit for mix arch compatibility */
485 region_t r;
486 } pkg;
488 if (lc->in_sync_hint >= lc->region_count)
489 return 0;
491 rdata_size = sizeof(pkg);
492 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
493 NULL, 0,
494 (char *)&pkg, &rdata_size);
496 *region = pkg.r;
497 return (r) ? r : (int)pkg.i;
501 * userspace_set_region_sync
503 * Set the sync status of a given region. This function
504 * must not fail.
506 static void userspace_set_region_sync(struct dm_dirty_log *log,
507 region_t region, int in_sync)
509 int r;
510 struct log_c *lc = log->context;
511 struct {
512 region_t r;
513 int64_t i;
514 } pkg;
516 pkg.r = region;
517 pkg.i = (int64_t)in_sync;
519 r = userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
520 (char *)&pkg, sizeof(pkg),
521 NULL, NULL);
524 * It would be nice to be able to report failures.
525 * However, it is easy emough to detect and resolve.
527 return;
531 * userspace_get_sync_count
533 * If there is any sort of failure when consulting the server,
534 * we assume that the sync count is zero.
536 * Returns: sync count on success, 0 on failure
538 static region_t userspace_get_sync_count(struct dm_dirty_log *log)
540 int r;
541 size_t rdata_size;
542 uint64_t sync_count;
543 struct log_c *lc = log->context;
545 rdata_size = sizeof(sync_count);
546 r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
547 NULL, 0,
548 (char *)&sync_count, &rdata_size);
550 if (r)
551 return 0;
553 if (sync_count >= lc->region_count)
554 lc->in_sync_hint = lc->region_count;
556 return (region_t)sync_count;
560 * userspace_status
562 * Returns: amount of space consumed
564 static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
565 char *result, unsigned maxlen)
567 int r = 0;
568 char *table_args;
569 size_t sz = (size_t)maxlen;
570 struct log_c *lc = log->context;
572 switch (status_type) {
573 case STATUSTYPE_INFO:
574 r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
575 NULL, 0,
576 result, &sz);
578 if (r) {
579 sz = 0;
580 DMEMIT("%s 1 COM_FAILURE", log->type->name);
582 break;
583 case STATUSTYPE_TABLE:
584 sz = 0;
585 table_args = strchr(lc->usr_argv_str, ' ');
586 BUG_ON(!table_args); /* There will always be a ' ' */
587 table_args++;
589 DMEMIT("%s %u %s %s ", log->type->name, lc->usr_argc,
590 lc->uuid, table_args);
591 break;
593 return (r) ? 0 : (int)sz;
597 * userspace_is_remote_recovering
599 * Returns: 1 if region recovering, 0 otherwise
601 static int userspace_is_remote_recovering(struct dm_dirty_log *log,
602 region_t region)
604 int r;
605 uint64_t region64 = region;
606 struct log_c *lc = log->context;
607 static unsigned long long limit;
608 struct {
609 int64_t is_recovering;
610 uint64_t in_sync_hint;
611 } pkg;
612 size_t rdata_size = sizeof(pkg);
615 * Once the mirror has been reported to be in-sync,
616 * it will never again ask for recovery work. So,
617 * we can safely say there is not a remote machine
618 * recovering if the device is in-sync. (in_sync_hint
619 * must be reset at resume time.)
621 if (region < lc->in_sync_hint)
622 return 0;
623 else if (jiffies < limit)
624 return 1;
626 limit = jiffies + (HZ / 4);
627 r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
628 (char *)&region64, sizeof(region64),
629 (char *)&pkg, &rdata_size);
630 if (r)
631 return 1;
633 lc->in_sync_hint = pkg.in_sync_hint;
635 return (int)pkg.is_recovering;
638 static struct dm_dirty_log_type _userspace_type = {
639 .name = "userspace",
640 .module = THIS_MODULE,
641 .ctr = userspace_ctr,
642 .dtr = userspace_dtr,
643 .presuspend = userspace_presuspend,
644 .postsuspend = userspace_postsuspend,
645 .resume = userspace_resume,
646 .get_region_size = userspace_get_region_size,
647 .is_clean = userspace_is_clean,
648 .in_sync = userspace_in_sync,
649 .flush = userspace_flush,
650 .mark_region = userspace_mark_region,
651 .clear_region = userspace_clear_region,
652 .get_resync_work = userspace_get_resync_work,
653 .set_region_sync = userspace_set_region_sync,
654 .get_sync_count = userspace_get_sync_count,
655 .status = userspace_status,
656 .is_remote_recovering = userspace_is_remote_recovering,
659 static int __init userspace_dirty_log_init(void)
661 int r = 0;
663 flush_entry_pool = mempool_create(100, flush_entry_alloc,
664 flush_entry_free, NULL);
666 if (!flush_entry_pool) {
667 DMWARN("Unable to create flush_entry_pool: No memory.");
668 return -ENOMEM;
671 r = dm_ulog_tfr_init();
672 if (r) {
673 DMWARN("Unable to initialize userspace log communications");
674 mempool_destroy(flush_entry_pool);
675 return r;
678 r = dm_dirty_log_type_register(&_userspace_type);
679 if (r) {
680 DMWARN("Couldn't register userspace dirty log type");
681 dm_ulog_tfr_exit();
682 mempool_destroy(flush_entry_pool);
683 return r;
686 DMINFO("version 1.0.0 loaded");
687 return 0;
690 static void __exit userspace_dirty_log_exit(void)
692 dm_dirty_log_type_unregister(&_userspace_type);
693 dm_ulog_tfr_exit();
694 mempool_destroy(flush_entry_pool);
696 DMINFO("version 1.0.0 unloaded");
697 return;
700 module_init(userspace_dirty_log_init);
701 module_exit(userspace_dirty_log_exit);
703 MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
704 MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>");
705 MODULE_LICENSE("GPL");