Commit | Line | Data |
---|---|---|
3bd94003 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
f5db4af4 JB |
2 | /* |
3 | * Copyright (C) 2006-2009 Red Hat, Inc. | |
4 | * | |
5 | * This file is released under the LGPL. | |
6 | */ | |
7 | ||
8 | #include <linux/bio.h> | |
5a0e3ad6 | 9 | #include <linux/slab.h> |
0f30af98 | 10 | #include <linux/jiffies.h> |
f5db4af4 JB |
11 | #include <linux/dm-dirty-log.h> |
12 | #include <linux/device-mapper.h> | |
13 | #include <linux/dm-log-userspace.h> | |
056075c7 | 14 | #include <linux/module.h> |
5066a4df | 15 | #include <linux/workqueue.h> |
f5db4af4 JB |
16 | |
17 | #include "dm-log-userspace-transfer.h" | |
18 | ||
5066a4df | 19 | #define DM_LOG_USERSPACE_VSN "1.3.0" |
86a54a48 | 20 | |
ac1f9ef2 MS |
21 | #define FLUSH_ENTRY_POOL_SIZE 16 |
22 | ||
23 | struct dm_dirty_log_flush_entry { | |
f5db4af4 JB |
24 | int type; |
25 | region_t region; | |
26 | struct list_head list; | |
27 | }; | |
28 | ||
085ae065 JB |
29 | /* |
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(). | |
34 | */ | |
35 | #define MAX_FLUSH_GROUP_COUNT 32 | |
36 | ||
f5db4af4 JB |
37 | struct log_c { |
38 | struct dm_target *ti; | |
5a25f0eb | 39 | struct dm_dev *log_dev; |
f5db4af4 JB |
40 | |
41 | char *usr_argv_str; | |
42 | uint32_t usr_argc; | |
43 | ||
ac1f9ef2 MS |
44 | uint32_t region_size; |
45 | region_t region_count; | |
46 | uint64_t luid; | |
47 | char uuid[DM_UUID_LEN]; | |
f5db4af4 | 48 | |
909cc4fb JB |
49 | /* |
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. | |
54 | */ | |
f5db4af4 | 55 | spinlock_t flush_lock; |
909cc4fb JB |
56 | struct list_head mark_list; |
57 | struct list_head clear_list; | |
5066a4df | 58 | |
ac1f9ef2 MS |
59 | /* |
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. | |
65 | */ | |
66 | uint64_t in_sync_hint; | |
67 | ||
5066a4df DZ |
68 | /* |
69 | * Workqueue for flush of clear region requests. | |
70 | */ | |
71 | struct workqueue_struct *dmlog_wq; | |
72 | struct delayed_work flush_log_work; | |
73 | atomic_t sched_flush; | |
74 | ||
75 | /* | |
76 | * Combine userspace flush and mark requests for efficiency. | |
77 | */ | |
78 | uint32_t integrated_flush; | |
f5db4af4 | 79 | |
6f1c819c | 80 | mempool_t flush_entry_pool; |
ac1f9ef2 | 81 | }; |
f5db4af4 | 82 | |
ac1f9ef2 | 83 | static struct kmem_cache *_flush_entry_cache; |
f5db4af4 JB |
84 | |
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) | |
88 | { | |
89 | int r; | |
90 | ||
91 | /* | |
92 | * If the server isn't there, -ESRCH is returned, | |
93 | * and we must keep trying until the server is | |
94 | * restored. | |
95 | */ | |
96 | retry: | |
7ec23d50 | 97 | r = dm_consult_userspace(uuid, lc->luid, request_type, data, |
f5db4af4 JB |
98 | data_size, rdata, rdata_size); |
99 | ||
100 | if (r != -ESRCH) | |
101 | return r; | |
102 | ||
103 | DMERR(" Userspace log server not found."); | |
104 | while (1) { | |
105 | set_current_state(TASK_INTERRUPTIBLE); | |
106 | schedule_timeout(2*HZ); | |
107 | DMWARN("Attempting to contact userspace log server..."); | |
7ec23d50 JB |
108 | r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR, |
109 | lc->usr_argv_str, | |
f5db4af4 JB |
110 | strlen(lc->usr_argv_str) + 1, |
111 | NULL, NULL); | |
112 | if (!r) | |
113 | break; | |
114 | } | |
115 | DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete"); | |
7ec23d50 | 116 | r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL, |
f5db4af4 JB |
117 | 0, NULL, NULL); |
118 | if (!r) | |
119 | goto retry; | |
120 | ||
121 | DMERR("Error trying to resume userspace log: %d", r); | |
122 | ||
123 | return -ESRCH; | |
124 | } | |
125 | ||
126 | static int build_constructor_string(struct dm_target *ti, | |
86a3238c | 127 | unsigned int argc, char **argv, |
f5db4af4 JB |
128 | char **ctr_str) |
129 | { | |
130 | int i, str_size; | |
131 | char *str = NULL; | |
132 | ||
133 | *ctr_str = NULL; | |
134 | ||
5066a4df DZ |
135 | /* |
136 | * Determine overall size of the string. | |
137 | */ | |
f5db4af4 JB |
138 | for (i = 0, str_size = 0; i < argc; i++) |
139 | str_size += strlen(argv[i]) + 1; /* +1 for space between args */ | |
140 | ||
141 | str_size += 20; /* Max number of chars in a printed u64 number */ | |
142 | ||
143 | str = kzalloc(str_size, GFP_KERNEL); | |
144 | if (!str) { | |
145 | DMWARN("Unable to allocate memory for constructor string"); | |
146 | return -ENOMEM; | |
147 | } | |
148 | ||
b8313b6d JB |
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]); | |
f5db4af4 JB |
152 | |
153 | *ctr_str = str; | |
154 | return str_size; | |
155 | } | |
156 | ||
5066a4df DZ |
157 | static void do_flush(struct work_struct *work) |
158 | { | |
159 | int r; | |
160 | struct log_c *lc = container_of(work, struct log_c, flush_log_work.work); | |
161 | ||
162 | atomic_set(&lc->sched_flush, 0); | |
163 | ||
164 | r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, NULL, 0, NULL, NULL); | |
165 | ||
166 | if (r) | |
167 | dm_table_event(lc->ti->table); | |
168 | } | |
169 | ||
f5db4af4 JB |
170 | /* |
171 | * userspace_ctr | |
172 | * | |
173 | * argv contains: | |
5066a4df DZ |
174 | * <UUID> [integrated_flush] <other args> |
175 | * Where 'other args' are the userspace implementation-specific log | |
176 | * arguments. | |
177 | * | |
178 | * Example: | |
179 | * <UUID> [integrated_flush] clustered-disk <arg count> <log dev> | |
180 | * <region_size> [[no]sync] | |
181 | * | |
182 | * This module strips off the <UUID> and uses it for identification | |
183 | * purposes when communicating with userspace about a log. | |
f5db4af4 | 184 | * |
5066a4df DZ |
185 | * If integrated_flush is defined, the kernel combines flush |
186 | * and mark requests. | |
187 | * | |
188 | * The rest of the line, beginning with 'clustered-disk', is passed | |
189 | * to the userspace ctr function. | |
f5db4af4 JB |
190 | */ |
191 | static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti, | |
86a3238c | 192 | unsigned int argc, char **argv) |
f5db4af4 JB |
193 | { |
194 | int r = 0; | |
195 | int str_size; | |
196 | char *ctr_str = NULL; | |
197 | struct log_c *lc = NULL; | |
198 | uint64_t rdata; | |
199 | size_t rdata_size = sizeof(rdata); | |
5a25f0eb JB |
200 | char *devices_rdata = NULL; |
201 | size_t devices_rdata_size = DM_NAME_LEN; | |
f5db4af4 JB |
202 | |
203 | if (argc < 3) { | |
204 | DMWARN("Too few arguments to userspace dirty log"); | |
205 | return -EINVAL; | |
206 | } | |
207 | ||
5a25f0eb | 208 | lc = kzalloc(sizeof(*lc), GFP_KERNEL); |
f5db4af4 JB |
209 | if (!lc) { |
210 | DMWARN("Unable to allocate userspace log context."); | |
211 | return -ENOMEM; | |
212 | } | |
213 | ||
7ec23d50 | 214 | /* The ptr value is sufficient for local unique id */ |
bca915aa | 215 | lc->luid = (unsigned long)lc; |
7ec23d50 | 216 | |
f5db4af4 JB |
217 | lc->ti = ti; |
218 | ||
219 | if (strlen(argv[0]) > (DM_UUID_LEN - 1)) { | |
220 | DMWARN("UUID argument too long."); | |
221 | kfree(lc); | |
222 | return -EINVAL; | |
223 | } | |
224 | ||
5066a4df DZ |
225 | lc->usr_argc = argc; |
226 | ||
18ac5233 | 227 | strscpy(lc->uuid, argv[0], sizeof(lc->uuid)); |
5066a4df DZ |
228 | argc--; |
229 | argv++; | |
f5db4af4 | 230 | spin_lock_init(&lc->flush_lock); |
909cc4fb JB |
231 | INIT_LIST_HEAD(&lc->mark_list); |
232 | INIT_LIST_HEAD(&lc->clear_list); | |
f5db4af4 | 233 | |
5066a4df DZ |
234 | if (!strcasecmp(argv[0], "integrated_flush")) { |
235 | lc->integrated_flush = 1; | |
236 | argc--; | |
237 | argv++; | |
238 | } | |
239 | ||
240 | str_size = build_constructor_string(ti, argc, argv, &ctr_str); | |
f5db4af4 JB |
241 | if (str_size < 0) { |
242 | kfree(lc); | |
243 | return str_size; | |
244 | } | |
245 | ||
5a25f0eb JB |
246 | devices_rdata = kzalloc(devices_rdata_size, GFP_KERNEL); |
247 | if (!devices_rdata) { | |
248 | DMERR("Failed to allocate memory for device information"); | |
249 | r = -ENOMEM; | |
250 | goto out; | |
251 | } | |
252 | ||
6f1c819c KO |
253 | r = mempool_init_slab_pool(&lc->flush_entry_pool, FLUSH_ENTRY_POOL_SIZE, |
254 | _flush_entry_cache); | |
255 | if (r) { | |
ac1f9ef2 | 256 | DMERR("Failed to create flush_entry_pool"); |
ac1f9ef2 MS |
257 | goto out; |
258 | } | |
259 | ||
5a25f0eb JB |
260 | /* |
261 | * Send table string and get back any opened device. | |
262 | */ | |
7ec23d50 | 263 | r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR, |
5a25f0eb JB |
264 | ctr_str, str_size, |
265 | devices_rdata, &devices_rdata_size); | |
f5db4af4 | 266 | |
4a038677 JB |
267 | if (r < 0) { |
268 | if (r == -ESRCH) | |
269 | DMERR("Userspace log server not found"); | |
270 | else | |
271 | DMERR("Userspace log server failed to create log"); | |
f5db4af4 JB |
272 | goto out; |
273 | } | |
274 | ||
275 | /* Since the region size does not change, get it now */ | |
276 | rdata_size = sizeof(rdata); | |
7ec23d50 | 277 | r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE, |
f5db4af4 JB |
278 | NULL, 0, (char *)&rdata, &rdata_size); |
279 | ||
280 | if (r) { | |
281 | DMERR("Failed to get region size of dirty log"); | |
282 | goto out; | |
283 | } | |
284 | ||
285 | lc->region_size = (uint32_t)rdata; | |
286 | lc->region_count = dm_sector_div_up(ti->len, lc->region_size); | |
287 | ||
5a25f0eb JB |
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"); | |
291 | r = -EINVAL; | |
292 | goto out; | |
293 | } | |
294 | r = dm_get_device(ti, devices_rdata, | |
295 | dm_table_get_mode(ti->table), &lc->log_dev); | |
296 | if (r) | |
297 | DMERR("Failed to register %s with device-mapper", | |
298 | devices_rdata); | |
299 | } | |
5066a4df DZ |
300 | |
301 | if (lc->integrated_flush) { | |
302 | lc->dmlog_wq = alloc_workqueue("dmlogd", WQ_MEM_RECLAIM, 0); | |
303 | if (!lc->dmlog_wq) { | |
304 | DMERR("couldn't start dmlogd"); | |
305 | r = -ENOMEM; | |
306 | goto out; | |
307 | } | |
308 | ||
309 | INIT_DELAYED_WORK(&lc->flush_log_work, do_flush); | |
310 | atomic_set(&lc->sched_flush, 0); | |
311 | } | |
312 | ||
f5db4af4 | 313 | out: |
5a25f0eb | 314 | kfree(devices_rdata); |
f5db4af4 | 315 | if (r) { |
6f1c819c | 316 | mempool_exit(&lc->flush_entry_pool); |
f5db4af4 JB |
317 | kfree(lc); |
318 | kfree(ctr_str); | |
319 | } else { | |
320 | lc->usr_argv_str = ctr_str; | |
f5db4af4 JB |
321 | log->context = lc; |
322 | } | |
323 | ||
324 | return r; | |
325 | } | |
326 | ||
327 | static void userspace_dtr(struct dm_dirty_log *log) | |
328 | { | |
f5db4af4 JB |
329 | struct log_c *lc = log->context; |
330 | ||
5066a4df DZ |
331 | if (lc->integrated_flush) { |
332 | /* flush workqueue */ | |
333 | if (atomic_read(&lc->sched_flush)) | |
334 | flush_delayed_work(&lc->flush_log_work); | |
335 | ||
336 | destroy_workqueue(lc->dmlog_wq); | |
337 | } | |
338 | ||
4a038677 | 339 | (void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR, |
5066a4df | 340 | NULL, 0, NULL, NULL); |
f5db4af4 | 341 | |
5a25f0eb JB |
342 | if (lc->log_dev) |
343 | dm_put_device(lc->ti, lc->log_dev); | |
344 | ||
6f1c819c | 345 | mempool_exit(&lc->flush_entry_pool); |
ac1f9ef2 | 346 | |
f5db4af4 JB |
347 | kfree(lc->usr_argv_str); |
348 | kfree(lc); | |
f5db4af4 JB |
349 | } |
350 | ||
351 | static int userspace_presuspend(struct dm_dirty_log *log) | |
352 | { | |
353 | int r; | |
354 | struct log_c *lc = log->context; | |
355 | ||
7ec23d50 | 356 | r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND, |
5066a4df | 357 | NULL, 0, NULL, NULL); |
f5db4af4 JB |
358 | |
359 | return r; | |
360 | } | |
361 | ||
362 | static int userspace_postsuspend(struct dm_dirty_log *log) | |
363 | { | |
364 | int r; | |
365 | struct log_c *lc = log->context; | |
366 | ||
5066a4df DZ |
367 | /* |
368 | * Run planned flush earlier. | |
369 | */ | |
370 | if (lc->integrated_flush && atomic_read(&lc->sched_flush)) | |
371 | flush_delayed_work(&lc->flush_log_work); | |
372 | ||
7ec23d50 | 373 | r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND, |
5066a4df | 374 | NULL, 0, NULL, NULL); |
f5db4af4 JB |
375 | |
376 | return r; | |
377 | } | |
378 | ||
379 | static int userspace_resume(struct dm_dirty_log *log) | |
380 | { | |
381 | int r; | |
382 | struct log_c *lc = log->context; | |
383 | ||
384 | lc->in_sync_hint = 0; | |
7ec23d50 | 385 | r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME, |
5066a4df | 386 | NULL, 0, NULL, NULL); |
f5db4af4 JB |
387 | |
388 | return r; | |
389 | } | |
390 | ||
391 | static uint32_t userspace_get_region_size(struct dm_dirty_log *log) | |
392 | { | |
393 | struct log_c *lc = log->context; | |
394 | ||
395 | return lc->region_size; | |
396 | } | |
397 | ||
398 | /* | |
399 | * userspace_is_clean | |
400 | * | |
401 | * Check whether a region is clean. If there is any sort of | |
402 | * failure when consulting the server, we return not clean. | |
403 | * | |
404 | * Returns: 1 if clean, 0 otherwise | |
405 | */ | |
406 | static int userspace_is_clean(struct dm_dirty_log *log, region_t region) | |
407 | { | |
408 | int r; | |
409 | uint64_t region64 = (uint64_t)region; | |
410 | int64_t is_clean; | |
411 | size_t rdata_size; | |
412 | struct log_c *lc = log->context; | |
413 | ||
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); | |
418 | ||
419 | return (r) ? 0 : (int)is_clean; | |
420 | } | |
421 | ||
422 | /* | |
423 | * userspace_in_sync | |
424 | * | |
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. | |
428 | * | |
429 | * If 'can_block' is set, return immediately | |
430 | * | |
431 | * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK | |
432 | */ | |
433 | static int userspace_in_sync(struct dm_dirty_log *log, region_t region, | |
434 | int can_block) | |
435 | { | |
436 | int r; | |
437 | uint64_t region64 = region; | |
438 | int64_t in_sync; | |
439 | size_t rdata_size; | |
440 | struct log_c *lc = log->context; | |
441 | ||
442 | /* | |
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.) | |
450 | * | |
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. | |
454 | */ | |
455 | if (!can_block) | |
456 | return -EWOULDBLOCK; | |
457 | ||
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; | |
463 | } | |
464 | ||
085ae065 JB |
465 | static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list) |
466 | { | |
467 | int r = 0; | |
ac1f9ef2 | 468 | struct dm_dirty_log_flush_entry *fe; |
085ae065 JB |
469 | |
470 | list_for_each_entry(fe, flush_list, list) { | |
471 | r = userspace_do_request(lc, lc->uuid, fe->type, | |
472 | (char *)&fe->region, | |
473 | sizeof(fe->region), | |
474 | NULL, NULL); | |
475 | if (r) | |
476 | break; | |
477 | } | |
478 | ||
479 | return r; | |
480 | } | |
481 | ||
5066a4df DZ |
482 | static int flush_by_group(struct log_c *lc, struct list_head *flush_list, |
483 | int flush_with_payload) | |
085ae065 JB |
484 | { |
485 | int r = 0; | |
486 | int count; | |
487 | uint32_t type = 0; | |
ac1f9ef2 | 488 | struct dm_dirty_log_flush_entry *fe, *tmp_fe; |
085ae065 JB |
489 | LIST_HEAD(tmp_list); |
490 | uint64_t group[MAX_FLUSH_GROUP_COUNT]; | |
491 | ||
492 | /* | |
493 | * Group process the requests | |
494 | */ | |
495 | while (!list_empty(flush_list)) { | |
496 | count = 0; | |
497 | ||
498 | list_for_each_entry_safe(fe, tmp_fe, flush_list, list) { | |
499 | group[count] = fe->region; | |
500 | count++; | |
501 | ||
6c9b27ab | 502 | list_move(&fe->list, &tmp_list); |
085ae065 JB |
503 | |
504 | type = fe->type; | |
505 | if (count >= MAX_FLUSH_GROUP_COUNT) | |
506 | break; | |
507 | } | |
508 | ||
5066a4df DZ |
509 | if (flush_with_payload) { |
510 | r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, | |
511 | (char *)(group), | |
512 | count * sizeof(uint64_t), | |
513 | NULL, NULL); | |
514 | /* | |
515 | * Integrated flush failed. | |
516 | */ | |
517 | if (r) | |
518 | break; | |
519 | } else { | |
520 | r = userspace_do_request(lc, lc->uuid, type, | |
521 | (char *)(group), | |
522 | count * sizeof(uint64_t), | |
523 | NULL, NULL); | |
524 | if (r) { | |
525 | /* | |
526 | * Group send failed. Attempt one-by-one. | |
527 | */ | |
528 | list_splice_init(&tmp_list, flush_list); | |
529 | r = flush_one_by_one(lc, flush_list); | |
530 | break; | |
531 | } | |
085ae065 JB |
532 | } |
533 | } | |
534 | ||
535 | /* | |
536 | * Must collect flush_entrys that were successfully processed | |
537 | * as a group so that they will be free'd by the caller. | |
538 | */ | |
539 | list_splice_init(&tmp_list, flush_list); | |
540 | ||
541 | return r; | |
542 | } | |
543 | ||
f5db4af4 JB |
544 | /* |
545 | * userspace_flush | |
546 | * | |
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. | |
553 | * | |
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. | |
558 | * | |
559 | * Returns: 0 on success, < 0 on failure | |
560 | */ | |
561 | static int userspace_flush(struct dm_dirty_log *log) | |
562 | { | |
563 | int r = 0; | |
564 | unsigned long flags; | |
565 | struct log_c *lc = log->context; | |
909cc4fb JB |
566 | LIST_HEAD(mark_list); |
567 | LIST_HEAD(clear_list); | |
5066a4df DZ |
568 | int mark_list_is_empty; |
569 | int clear_list_is_empty; | |
ac1f9ef2 | 570 | struct dm_dirty_log_flush_entry *fe, *tmp_fe; |
6f1c819c | 571 | mempool_t *flush_entry_pool = &lc->flush_entry_pool; |
f5db4af4 JB |
572 | |
573 | spin_lock_irqsave(&lc->flush_lock, flags); | |
909cc4fb JB |
574 | list_splice_init(&lc->mark_list, &mark_list); |
575 | list_splice_init(&lc->clear_list, &clear_list); | |
f5db4af4 JB |
576 | spin_unlock_irqrestore(&lc->flush_lock, flags); |
577 | ||
5066a4df DZ |
578 | mark_list_is_empty = list_empty(&mark_list); |
579 | clear_list_is_empty = list_empty(&clear_list); | |
580 | ||
581 | if (mark_list_is_empty && clear_list_is_empty) | |
f5db4af4 JB |
582 | return 0; |
583 | ||
5066a4df | 584 | r = flush_by_group(lc, &clear_list, 0); |
085ae065 | 585 | if (r) |
5066a4df DZ |
586 | goto out; |
587 | ||
588 | if (!lc->integrated_flush) { | |
589 | r = flush_by_group(lc, &mark_list, 0); | |
590 | if (r) | |
591 | goto out; | |
592 | r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, | |
593 | NULL, 0, NULL, NULL); | |
594 | goto out; | |
595 | } | |
909cc4fb | 596 | |
5066a4df DZ |
597 | /* |
598 | * Send integrated flush request with mark_list as payload. | |
599 | */ | |
600 | r = flush_by_group(lc, &mark_list, 1); | |
085ae065 | 601 | if (r) |
5066a4df | 602 | goto out; |
f5db4af4 | 603 | |
5066a4df DZ |
604 | if (mark_list_is_empty && !atomic_read(&lc->sched_flush)) { |
605 | /* | |
606 | * When there are only clear region requests, | |
607 | * we schedule a flush in the future. | |
608 | */ | |
609 | queue_delayed_work(lc->dmlog_wq, &lc->flush_log_work, 3 * HZ); | |
610 | atomic_set(&lc->sched_flush, 1); | |
611 | } else { | |
612 | /* | |
613 | * Cancel pending flush because we | |
614 | * have already flushed in mark_region. | |
615 | */ | |
616 | cancel_delayed_work(&lc->flush_log_work); | |
617 | atomic_set(&lc->sched_flush, 0); | |
618 | } | |
f5db4af4 | 619 | |
5066a4df | 620 | out: |
f5db4af4 | 621 | /* |
5066a4df | 622 | * We can safely remove these entries, even after failure. |
f5db4af4 JB |
623 | * Calling code will receive an error and will know that |
624 | * the log facility has failed. | |
625 | */ | |
909cc4fb JB |
626 | list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) { |
627 | list_del(&fe->list); | |
628 | mempool_free(fe, flush_entry_pool); | |
629 | } | |
630 | list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) { | |
f5db4af4 JB |
631 | list_del(&fe->list); |
632 | mempool_free(fe, flush_entry_pool); | |
633 | } | |
634 | ||
635 | if (r) | |
636 | dm_table_event(lc->ti->table); | |
637 | ||
638 | return r; | |
639 | } | |
640 | ||
641 | /* | |
642 | * userspace_mark_region | |
643 | * | |
644 | * This function should avoid blocking unless absolutely required. | |
645 | * (Memory allocation is valid for blocking.) | |
646 | */ | |
647 | static void userspace_mark_region(struct dm_dirty_log *log, region_t region) | |
648 | { | |
649 | unsigned long flags; | |
650 | struct log_c *lc = log->context; | |
ac1f9ef2 | 651 | struct dm_dirty_log_flush_entry *fe; |
f5db4af4 JB |
652 | |
653 | /* Wait for an allocation, but _never_ fail */ | |
6f1c819c | 654 | fe = mempool_alloc(&lc->flush_entry_pool, GFP_NOIO); |
f5db4af4 JB |
655 | BUG_ON(!fe); |
656 | ||
657 | spin_lock_irqsave(&lc->flush_lock, flags); | |
658 | fe->type = DM_ULOG_MARK_REGION; | |
659 | fe->region = region; | |
909cc4fb | 660 | list_add(&fe->list, &lc->mark_list); |
f5db4af4 | 661 | spin_unlock_irqrestore(&lc->flush_lock, flags); |
f5db4af4 JB |
662 | } |
663 | ||
664 | /* | |
665 | * userspace_clear_region | |
666 | * | |
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. | |
673 | */ | |
674 | static void userspace_clear_region(struct dm_dirty_log *log, region_t region) | |
675 | { | |
676 | unsigned long flags; | |
677 | struct log_c *lc = log->context; | |
ac1f9ef2 | 678 | struct dm_dirty_log_flush_entry *fe; |
f5db4af4 JB |
679 | |
680 | /* | |
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. | |
685 | */ | |
6f1c819c | 686 | fe = mempool_alloc(&lc->flush_entry_pool, GFP_ATOMIC); |
f5db4af4 JB |
687 | if (!fe) { |
688 | DMERR("Failed to allocate memory to clear region."); | |
689 | return; | |
690 | } | |
691 | ||
692 | spin_lock_irqsave(&lc->flush_lock, flags); | |
693 | fe->type = DM_ULOG_CLEAR_REGION; | |
694 | fe->region = region; | |
909cc4fb | 695 | list_add(&fe->list, &lc->clear_list); |
f5db4af4 | 696 | spin_unlock_irqrestore(&lc->flush_lock, flags); |
f5db4af4 JB |
697 | } |
698 | ||
699 | /* | |
700 | * userspace_get_resync_work | |
701 | * | |
702 | * Get a region that needs recovery. It is valid to return | |
703 | * an error for this function. | |
704 | * | |
705 | * Returns: 1 if region filled, 0 if no work, <0 on error | |
706 | */ | |
707 | static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region) | |
708 | { | |
709 | int r; | |
710 | size_t rdata_size; | |
711 | struct log_c *lc = log->context; | |
712 | struct { | |
713 | int64_t i; /* 64-bit for mix arch compatibility */ | |
714 | region_t r; | |
715 | } pkg; | |
716 | ||
717 | if (lc->in_sync_hint >= lc->region_count) | |
718 | return 0; | |
719 | ||
720 | rdata_size = sizeof(pkg); | |
721 | r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK, | |
5066a4df | 722 | NULL, 0, (char *)&pkg, &rdata_size); |
f5db4af4 JB |
723 | |
724 | *region = pkg.r; | |
725 | return (r) ? r : (int)pkg.i; | |
726 | } | |
727 | ||
728 | /* | |
729 | * userspace_set_region_sync | |
730 | * | |
731 | * Set the sync status of a given region. This function | |
732 | * must not fail. | |
733 | */ | |
734 | static void userspace_set_region_sync(struct dm_dirty_log *log, | |
735 | region_t region, int in_sync) | |
736 | { | |
f5db4af4 JB |
737 | struct log_c *lc = log->context; |
738 | struct { | |
739 | region_t r; | |
740 | int64_t i; | |
741 | } pkg; | |
742 | ||
743 | pkg.r = region; | |
744 | pkg.i = (int64_t)in_sync; | |
745 | ||
18cc980a NMG |
746 | (void) userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC, |
747 | (char *)&pkg, sizeof(pkg), NULL, NULL); | |
f5db4af4 JB |
748 | |
749 | /* | |
750 | * It would be nice to be able to report failures. | |
18cc980a | 751 | * However, it is easy enough to detect and resolve. |
f5db4af4 | 752 | */ |
f5db4af4 JB |
753 | } |
754 | ||
755 | /* | |
756 | * userspace_get_sync_count | |
757 | * | |
758 | * If there is any sort of failure when consulting the server, | |
759 | * we assume that the sync count is zero. | |
760 | * | |
761 | * Returns: sync count on success, 0 on failure | |
762 | */ | |
763 | static region_t userspace_get_sync_count(struct dm_dirty_log *log) | |
764 | { | |
765 | int r; | |
766 | size_t rdata_size; | |
767 | uint64_t sync_count; | |
768 | struct log_c *lc = log->context; | |
769 | ||
770 | rdata_size = sizeof(sync_count); | |
771 | r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT, | |
5066a4df | 772 | NULL, 0, (char *)&sync_count, &rdata_size); |
f5db4af4 JB |
773 | |
774 | if (r) | |
775 | return 0; | |
776 | ||
777 | if (sync_count >= lc->region_count) | |
778 | lc->in_sync_hint = lc->region_count; | |
779 | ||
780 | return (region_t)sync_count; | |
781 | } | |
782 | ||
783 | /* | |
784 | * userspace_status | |
785 | * | |
786 | * Returns: amount of space consumed | |
787 | */ | |
788 | static int userspace_status(struct dm_dirty_log *log, status_type_t status_type, | |
86a3238c | 789 | char *result, unsigned int maxlen) |
f5db4af4 JB |
790 | { |
791 | int r = 0; | |
b8313b6d | 792 | char *table_args; |
f5db4af4 JB |
793 | size_t sz = (size_t)maxlen; |
794 | struct log_c *lc = log->context; | |
795 | ||
796 | switch (status_type) { | |
797 | case STATUSTYPE_INFO: | |
798 | r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO, | |
5066a4df | 799 | NULL, 0, result, &sz); |
f5db4af4 JB |
800 | |
801 | if (r) { | |
802 | sz = 0; | |
803 | DMEMIT("%s 1 COM_FAILURE", log->type->name); | |
804 | } | |
805 | break; | |
806 | case STATUSTYPE_TABLE: | |
807 | sz = 0; | |
0d03d59d | 808 | table_args = strchr(lc->usr_argv_str, ' '); |
b8313b6d JB |
809 | BUG_ON(!table_args); /* There will always be a ' ' */ |
810 | table_args++; | |
811 | ||
5066a4df DZ |
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); | |
f5db4af4 | 816 | break; |
8ec45662 TS |
817 | case STATUSTYPE_IMA: |
818 | *result = '\0'; | |
819 | break; | |
f5db4af4 JB |
820 | } |
821 | return (r) ? 0 : (int)sz; | |
822 | } | |
823 | ||
824 | /* | |
825 | * userspace_is_remote_recovering | |
826 | * | |
827 | * Returns: 1 if region recovering, 0 otherwise | |
828 | */ | |
829 | static int userspace_is_remote_recovering(struct dm_dirty_log *log, | |
830 | region_t region) | |
831 | { | |
832 | int r; | |
833 | uint64_t region64 = region; | |
834 | struct log_c *lc = log->context; | |
0f30af98 | 835 | static unsigned long limit; |
f5db4af4 JB |
836 | struct { |
837 | int64_t is_recovering; | |
838 | uint64_t in_sync_hint; | |
839 | } pkg; | |
840 | size_t rdata_size = sizeof(pkg); | |
841 | ||
842 | /* | |
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.) | |
848 | */ | |
849 | if (region < lc->in_sync_hint) | |
850 | return 0; | |
0f30af98 | 851 | else if (time_after(limit, jiffies)) |
f5db4af4 JB |
852 | return 1; |
853 | ||
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); | |
858 | if (r) | |
859 | return 1; | |
860 | ||
861 | lc->in_sync_hint = pkg.in_sync_hint; | |
862 | ||
863 | return (int)pkg.is_recovering; | |
864 | } | |
865 | ||
866 | static struct dm_dirty_log_type _userspace_type = { | |
867 | .name = "userspace", | |
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, | |
885 | }; | |
886 | ||
887 | static int __init userspace_dirty_log_init(void) | |
888 | { | |
889 | int r = 0; | |
890 | ||
ac1f9ef2 MS |
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."); | |
f5db4af4 JB |
894 | return -ENOMEM; |
895 | } | |
896 | ||
897 | r = dm_ulog_tfr_init(); | |
898 | if (r) { | |
899 | DMWARN("Unable to initialize userspace log communications"); | |
ac1f9ef2 | 900 | kmem_cache_destroy(_flush_entry_cache); |
f5db4af4 JB |
901 | return r; |
902 | } | |
903 | ||
904 | r = dm_dirty_log_type_register(&_userspace_type); | |
905 | if (r) { | |
906 | DMWARN("Couldn't register userspace dirty log type"); | |
907 | dm_ulog_tfr_exit(); | |
ac1f9ef2 | 908 | kmem_cache_destroy(_flush_entry_cache); |
f5db4af4 JB |
909 | return r; |
910 | } | |
911 | ||
86a54a48 | 912 | DMINFO("version " DM_LOG_USERSPACE_VSN " loaded"); |
f5db4af4 JB |
913 | return 0; |
914 | } | |
915 | ||
916 | static void __exit userspace_dirty_log_exit(void) | |
917 | { | |
918 | dm_dirty_log_type_unregister(&_userspace_type); | |
919 | dm_ulog_tfr_exit(); | |
ac1f9ef2 | 920 | kmem_cache_destroy(_flush_entry_cache); |
f5db4af4 | 921 | |
86a54a48 | 922 | DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded"); |
f5db4af4 JB |
923 | } |
924 | ||
925 | module_init(userspace_dirty_log_init); | |
926 | module_exit(userspace_dirty_log_exit); | |
927 | ||
928 | MODULE_DESCRIPTION(DM_NAME " userspace dirty log link"); | |
929 | MODULE_AUTHOR("Jonathan Brassow <dm-devel@redhat.com>"); | |
930 | MODULE_LICENSE("GPL"); |