Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software Limited. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
9 | #include "dm-path-selector.h" | |
1da177e4 LT |
10 | #include "dm-bio-list.h" |
11 | #include "dm-bio-record.h" | |
b15546f9 | 12 | #include "dm-uevent.h" |
1da177e4 LT |
13 | |
14 | #include <linux/ctype.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mempool.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/pagemap.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/time.h> | |
21 | #include <linux/workqueue.h> | |
cfae5c9b | 22 | #include <scsi/scsi_dh.h> |
1da177e4 LT |
23 | #include <asm/atomic.h> |
24 | ||
72d94861 | 25 | #define DM_MSG_PREFIX "multipath" |
1da177e4 LT |
26 | #define MESG_STR(x) x, sizeof(x) |
27 | ||
28 | /* Path properties */ | |
29 | struct pgpath { | |
30 | struct list_head list; | |
31 | ||
32 | struct priority_group *pg; /* Owning PG */ | |
33 | unsigned fail_count; /* Cumulative failure count */ | |
34 | ||
c922d5f7 | 35 | struct dm_path path; |
1da177e4 LT |
36 | }; |
37 | ||
38 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
39 | ||
40 | /* | |
41 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
42 | * Each has a path selector which controls which path gets used. | |
43 | */ | |
44 | struct priority_group { | |
45 | struct list_head list; | |
46 | ||
47 | struct multipath *m; /* Owning multipath instance */ | |
48 | struct path_selector ps; | |
49 | ||
50 | unsigned pg_num; /* Reference number */ | |
51 | unsigned bypassed; /* Temporarily bypass this PG? */ | |
52 | ||
53 | unsigned nr_pgpaths; /* Number of paths in PG */ | |
54 | struct list_head pgpaths; | |
55 | }; | |
56 | ||
57 | /* Multipath context */ | |
58 | struct multipath { | |
59 | struct list_head list; | |
60 | struct dm_target *ti; | |
61 | ||
62 | spinlock_t lock; | |
63 | ||
cfae5c9b | 64 | const char *hw_handler_name; |
bab7cfc7 | 65 | struct work_struct activate_path; |
1da177e4 LT |
66 | unsigned nr_priority_groups; |
67 | struct list_head priority_groups; | |
68 | unsigned pg_init_required; /* pg_init needs calling? */ | |
c3cd4f6b | 69 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
1da177e4 LT |
70 | |
71 | unsigned nr_valid_paths; /* Total number of usable paths */ | |
72 | struct pgpath *current_pgpath; | |
73 | struct priority_group *current_pg; | |
74 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
75 | unsigned repeat_count; /* I/Os left before calling PS again */ | |
76 | ||
77 | unsigned queue_io; /* Must we queue all I/O? */ | |
78 | unsigned queue_if_no_path; /* Queue I/O if last path fails? */ | |
436d4108 | 79 | unsigned saved_queue_if_no_path;/* Saved state during suspension */ |
c9e45581 DW |
80 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
81 | unsigned pg_init_count; /* Number of times pg_init called */ | |
1da177e4 LT |
82 | |
83 | struct work_struct process_queued_ios; | |
84 | struct bio_list queued_ios; | |
85 | unsigned queue_size; | |
86 | ||
87 | struct work_struct trigger_event; | |
88 | ||
89 | /* | |
028867ac | 90 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
91 | * can resubmit bios on error. |
92 | */ | |
93 | mempool_t *mpio_pool; | |
94 | }; | |
95 | ||
96 | /* | |
97 | * Context information attached to each bio we process. | |
98 | */ | |
028867ac | 99 | struct dm_mpath_io { |
1da177e4 LT |
100 | struct pgpath *pgpath; |
101 | struct dm_bio_details details; | |
102 | }; | |
103 | ||
104 | typedef int (*action_fn) (struct pgpath *pgpath); | |
105 | ||
106 | #define MIN_IOS 256 /* Mempool size */ | |
107 | ||
e18b890b | 108 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 109 | |
bab7cfc7 | 110 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 DH |
111 | static void process_queued_ios(struct work_struct *work); |
112 | static void trigger_event(struct work_struct *work); | |
bab7cfc7 | 113 | static void activate_path(struct work_struct *work); |
1da177e4 LT |
114 | |
115 | ||
116 | /*----------------------------------------------- | |
117 | * Allocation routines | |
118 | *-----------------------------------------------*/ | |
119 | ||
120 | static struct pgpath *alloc_pgpath(void) | |
121 | { | |
e69fae56 | 122 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 123 | |
e69fae56 | 124 | if (pgpath) |
1da177e4 | 125 | pgpath->path.is_active = 1; |
1da177e4 LT |
126 | |
127 | return pgpath; | |
128 | } | |
129 | ||
028867ac | 130 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
131 | { |
132 | kfree(pgpath); | |
133 | } | |
134 | ||
135 | static struct priority_group *alloc_priority_group(void) | |
136 | { | |
137 | struct priority_group *pg; | |
138 | ||
e69fae56 | 139 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 140 | |
e69fae56 MM |
141 | if (pg) |
142 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
143 | |
144 | return pg; | |
145 | } | |
146 | ||
147 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
148 | { | |
149 | struct pgpath *pgpath, *tmp; | |
ae11b1b3 | 150 | struct multipath *m = ti->private; |
1da177e4 LT |
151 | |
152 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
153 | list_del(&pgpath->list); | |
ae11b1b3 HR |
154 | if (m->hw_handler_name) |
155 | scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev)); | |
1da177e4 LT |
156 | dm_put_device(ti, pgpath->path.dev); |
157 | free_pgpath(pgpath); | |
158 | } | |
159 | } | |
160 | ||
161 | static void free_priority_group(struct priority_group *pg, | |
162 | struct dm_target *ti) | |
163 | { | |
164 | struct path_selector *ps = &pg->ps; | |
165 | ||
166 | if (ps->type) { | |
167 | ps->type->destroy(ps); | |
168 | dm_put_path_selector(ps->type); | |
169 | } | |
170 | ||
171 | free_pgpaths(&pg->pgpaths, ti); | |
172 | kfree(pg); | |
173 | } | |
174 | ||
28f16c20 | 175 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
176 | { |
177 | struct multipath *m; | |
178 | ||
e69fae56 | 179 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 180 | if (m) { |
1da177e4 LT |
181 | INIT_LIST_HEAD(&m->priority_groups); |
182 | spin_lock_init(&m->lock); | |
183 | m->queue_io = 1; | |
c4028958 DH |
184 | INIT_WORK(&m->process_queued_ios, process_queued_ios); |
185 | INIT_WORK(&m->trigger_event, trigger_event); | |
bab7cfc7 | 186 | INIT_WORK(&m->activate_path, activate_path); |
93d2341c | 187 | m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache); |
1da177e4 LT |
188 | if (!m->mpio_pool) { |
189 | kfree(m); | |
190 | return NULL; | |
191 | } | |
28f16c20 MM |
192 | m->ti = ti; |
193 | ti->private = m; | |
1da177e4 LT |
194 | } |
195 | ||
196 | return m; | |
197 | } | |
198 | ||
199 | static void free_multipath(struct multipath *m) | |
200 | { | |
201 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
202 | |
203 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
204 | list_del(&pg->list); | |
205 | free_priority_group(pg, m->ti); | |
206 | } | |
207 | ||
cfae5c9b | 208 | kfree(m->hw_handler_name); |
1da177e4 LT |
209 | mempool_destroy(m->mpio_pool); |
210 | kfree(m); | |
211 | } | |
212 | ||
213 | ||
214 | /*----------------------------------------------- | |
215 | * Path selection | |
216 | *-----------------------------------------------*/ | |
217 | ||
218 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) | |
219 | { | |
1da177e4 LT |
220 | m->current_pg = pgpath->pg; |
221 | ||
222 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 223 | if (m->hw_handler_name) { |
1da177e4 LT |
224 | m->pg_init_required = 1; |
225 | m->queue_io = 1; | |
226 | } else { | |
227 | m->pg_init_required = 0; | |
228 | m->queue_io = 0; | |
229 | } | |
c9e45581 DW |
230 | |
231 | m->pg_init_count = 0; | |
1da177e4 LT |
232 | } |
233 | ||
234 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg) | |
235 | { | |
c922d5f7 | 236 | struct dm_path *path; |
1da177e4 LT |
237 | |
238 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count); | |
239 | if (!path) | |
240 | return -ENXIO; | |
241 | ||
242 | m->current_pgpath = path_to_pgpath(path); | |
243 | ||
244 | if (m->current_pg != pg) | |
245 | __switch_pg(m, m->current_pgpath); | |
246 | ||
247 | return 0; | |
248 | } | |
249 | ||
250 | static void __choose_pgpath(struct multipath *m) | |
251 | { | |
252 | struct priority_group *pg; | |
253 | unsigned bypassed = 1; | |
254 | ||
255 | if (!m->nr_valid_paths) | |
256 | goto failed; | |
257 | ||
258 | /* Were we instructed to switch PG? */ | |
259 | if (m->next_pg) { | |
260 | pg = m->next_pg; | |
261 | m->next_pg = NULL; | |
262 | if (!__choose_path_in_pg(m, pg)) | |
263 | return; | |
264 | } | |
265 | ||
266 | /* Don't change PG until it has no remaining paths */ | |
267 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg)) | |
268 | return; | |
269 | ||
270 | /* | |
271 | * Loop through priority groups until we find a valid path. | |
272 | * First time we skip PGs marked 'bypassed'. | |
273 | * Second time we only try the ones we skipped. | |
274 | */ | |
275 | do { | |
276 | list_for_each_entry(pg, &m->priority_groups, list) { | |
277 | if (pg->bypassed == bypassed) | |
278 | continue; | |
279 | if (!__choose_path_in_pg(m, pg)) | |
280 | return; | |
281 | } | |
282 | } while (bypassed--); | |
283 | ||
284 | failed: | |
285 | m->current_pgpath = NULL; | |
286 | m->current_pg = NULL; | |
287 | } | |
288 | ||
45e15720 KU |
289 | /* |
290 | * Check whether bios must be queued in the device-mapper core rather | |
291 | * than here in the target. | |
292 | * | |
293 | * m->lock must be held on entry. | |
294 | * | |
295 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
296 | * same value then we are not between multipath_presuspend() | |
297 | * and multipath_resume() calls and we have no need to check | |
298 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
299 | */ | |
300 | static int __must_push_back(struct multipath *m) | |
301 | { | |
302 | return (m->queue_if_no_path != m->saved_queue_if_no_path && | |
303 | dm_noflush_suspending(m->ti)); | |
304 | } | |
305 | ||
028867ac AK |
306 | static int map_io(struct multipath *m, struct bio *bio, |
307 | struct dm_mpath_io *mpio, unsigned was_queued) | |
1da177e4 | 308 | { |
d2a7ad29 | 309 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
310 | unsigned long flags; |
311 | struct pgpath *pgpath; | |
312 | ||
313 | spin_lock_irqsave(&m->lock, flags); | |
314 | ||
315 | /* Do we need to select a new pgpath? */ | |
316 | if (!m->current_pgpath || | |
317 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
318 | __choose_pgpath(m); | |
319 | ||
320 | pgpath = m->current_pgpath; | |
321 | ||
322 | if (was_queued) | |
323 | m->queue_size--; | |
324 | ||
325 | if ((pgpath && m->queue_io) || | |
436d4108 | 326 | (!pgpath && m->queue_if_no_path)) { |
1da177e4 LT |
327 | /* Queue for the daemon to resubmit */ |
328 | bio_list_add(&m->queued_ios, bio); | |
329 | m->queue_size++; | |
c3cd4f6b AK |
330 | if ((m->pg_init_required && !m->pg_init_in_progress) || |
331 | !m->queue_io) | |
c557308e | 332 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 333 | pgpath = NULL; |
d2a7ad29 | 334 | r = DM_MAPIO_SUBMITTED; |
45e15720 | 335 | } else if (pgpath) |
1da177e4 | 336 | bio->bi_bdev = pgpath->path.dev->bdev; |
45e15720 KU |
337 | else if (__must_push_back(m)) |
338 | r = DM_MAPIO_REQUEUE; | |
339 | else | |
340 | r = -EIO; /* Failed */ | |
1da177e4 LT |
341 | |
342 | mpio->pgpath = pgpath; | |
343 | ||
344 | spin_unlock_irqrestore(&m->lock, flags); | |
345 | ||
346 | return r; | |
347 | } | |
348 | ||
349 | /* | |
350 | * If we run out of usable paths, should we queue I/O or error it? | |
351 | */ | |
485ef69e AK |
352 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
353 | unsigned save_old_value) | |
1da177e4 LT |
354 | { |
355 | unsigned long flags; | |
356 | ||
357 | spin_lock_irqsave(&m->lock, flags); | |
358 | ||
485ef69e AK |
359 | if (save_old_value) |
360 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
361 | else | |
362 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 363 | m->queue_if_no_path = queue_if_no_path; |
c3cd4f6b | 364 | if (!m->queue_if_no_path && m->queue_size) |
c557308e | 365 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 LT |
366 | |
367 | spin_unlock_irqrestore(&m->lock, flags); | |
368 | ||
369 | return 0; | |
370 | } | |
371 | ||
372 | /*----------------------------------------------------------------- | |
373 | * The multipath daemon is responsible for resubmitting queued ios. | |
374 | *---------------------------------------------------------------*/ | |
375 | ||
376 | static void dispatch_queued_ios(struct multipath *m) | |
377 | { | |
378 | int r; | |
379 | unsigned long flags; | |
380 | struct bio *bio = NULL, *next; | |
028867ac | 381 | struct dm_mpath_io *mpio; |
1da177e4 LT |
382 | union map_info *info; |
383 | ||
384 | spin_lock_irqsave(&m->lock, flags); | |
385 | bio = bio_list_get(&m->queued_ios); | |
386 | spin_unlock_irqrestore(&m->lock, flags); | |
387 | ||
388 | while (bio) { | |
389 | next = bio->bi_next; | |
390 | bio->bi_next = NULL; | |
391 | ||
392 | info = dm_get_mapinfo(bio); | |
393 | mpio = info->ptr; | |
394 | ||
395 | r = map_io(m, bio, mpio, 1); | |
396 | if (r < 0) | |
6712ecf8 | 397 | bio_endio(bio, r); |
d2a7ad29 | 398 | else if (r == DM_MAPIO_REMAPPED) |
1da177e4 | 399 | generic_make_request(bio); |
45e15720 | 400 | else if (r == DM_MAPIO_REQUEUE) |
6712ecf8 | 401 | bio_endio(bio, -EIO); |
1da177e4 LT |
402 | |
403 | bio = next; | |
404 | } | |
405 | } | |
406 | ||
c4028958 | 407 | static void process_queued_ios(struct work_struct *work) |
1da177e4 | 408 | { |
c4028958 DH |
409 | struct multipath *m = |
410 | container_of(work, struct multipath, process_queued_ios); | |
c3cd4f6b AK |
411 | struct pgpath *pgpath = NULL; |
412 | unsigned init_required = 0, must_queue = 1; | |
1da177e4 LT |
413 | unsigned long flags; |
414 | ||
415 | spin_lock_irqsave(&m->lock, flags); | |
416 | ||
c3cd4f6b AK |
417 | if (!m->queue_size) |
418 | goto out; | |
419 | ||
1da177e4 LT |
420 | if (!m->current_pgpath) |
421 | __choose_pgpath(m); | |
422 | ||
423 | pgpath = m->current_pgpath; | |
424 | ||
c3cd4f6b AK |
425 | if ((pgpath && !m->queue_io) || |
426 | (!pgpath && !m->queue_if_no_path)) | |
427 | must_queue = 0; | |
1da177e4 | 428 | |
c3cd4f6b | 429 | if (m->pg_init_required && !m->pg_init_in_progress) { |
c9e45581 | 430 | m->pg_init_count++; |
1da177e4 | 431 | m->pg_init_required = 0; |
c3cd4f6b AK |
432 | m->pg_init_in_progress = 1; |
433 | init_required = 1; | |
434 | } | |
1da177e4 | 435 | |
c3cd4f6b | 436 | out: |
1da177e4 LT |
437 | spin_unlock_irqrestore(&m->lock, flags); |
438 | ||
bab7cfc7 CS |
439 | if (init_required) |
440 | queue_work(kmpath_handlerd, &m->activate_path); | |
1da177e4 LT |
441 | |
442 | if (!must_queue) | |
443 | dispatch_queued_ios(m); | |
444 | } | |
445 | ||
446 | /* | |
447 | * An event is triggered whenever a path is taken out of use. | |
448 | * Includes path failure and PG bypass. | |
449 | */ | |
c4028958 | 450 | static void trigger_event(struct work_struct *work) |
1da177e4 | 451 | { |
c4028958 DH |
452 | struct multipath *m = |
453 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
454 | |
455 | dm_table_event(m->ti->table); | |
456 | } | |
457 | ||
458 | /*----------------------------------------------------------------- | |
459 | * Constructor/argument parsing: | |
460 | * <#multipath feature args> [<arg>]* | |
461 | * <#hw_handler args> [hw_handler [<arg>]*] | |
462 | * <#priority groups> | |
463 | * <initial priority group> | |
464 | * [<selector> <#selector args> [<arg>]* | |
465 | * <#paths> <#per-path selector args> | |
466 | * [<path> [<arg>]* ]+ ]+ | |
467 | *---------------------------------------------------------------*/ | |
468 | struct param { | |
469 | unsigned min; | |
470 | unsigned max; | |
471 | char *error; | |
472 | }; | |
473 | ||
1da177e4 LT |
474 | static int read_param(struct param *param, char *str, unsigned *v, char **error) |
475 | { | |
476 | if (!str || | |
477 | (sscanf(str, "%u", v) != 1) || | |
478 | (*v < param->min) || | |
479 | (*v > param->max)) { | |
480 | *error = param->error; | |
481 | return -EINVAL; | |
482 | } | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
487 | struct arg_set { | |
488 | unsigned argc; | |
489 | char **argv; | |
490 | }; | |
491 | ||
492 | static char *shift(struct arg_set *as) | |
493 | { | |
494 | char *r; | |
495 | ||
496 | if (as->argc) { | |
497 | as->argc--; | |
498 | r = *as->argv; | |
499 | as->argv++; | |
500 | return r; | |
501 | } | |
502 | ||
503 | return NULL; | |
504 | } | |
505 | ||
506 | static void consume(struct arg_set *as, unsigned n) | |
507 | { | |
508 | BUG_ON (as->argc < n); | |
509 | as->argc -= n; | |
510 | as->argv += n; | |
511 | } | |
512 | ||
513 | static int parse_path_selector(struct arg_set *as, struct priority_group *pg, | |
514 | struct dm_target *ti) | |
515 | { | |
516 | int r; | |
517 | struct path_selector_type *pst; | |
518 | unsigned ps_argc; | |
519 | ||
520 | static struct param _params[] = { | |
72d94861 | 521 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
522 | }; |
523 | ||
524 | pst = dm_get_path_selector(shift(as)); | |
525 | if (!pst) { | |
72d94861 | 526 | ti->error = "unknown path selector type"; |
1da177e4 LT |
527 | return -EINVAL; |
528 | } | |
529 | ||
530 | r = read_param(_params, shift(as), &ps_argc, &ti->error); | |
371b2e34 MP |
531 | if (r) { |
532 | dm_put_path_selector(pst); | |
1da177e4 | 533 | return -EINVAL; |
371b2e34 | 534 | } |
1da177e4 LT |
535 | |
536 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
537 | if (r) { | |
538 | dm_put_path_selector(pst); | |
72d94861 | 539 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
540 | return r; |
541 | } | |
542 | ||
543 | pg->ps.type = pst; | |
544 | consume(as, ps_argc); | |
545 | ||
546 | return 0; | |
547 | } | |
548 | ||
549 | static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps, | |
550 | struct dm_target *ti) | |
551 | { | |
552 | int r; | |
553 | struct pgpath *p; | |
ae11b1b3 | 554 | struct multipath *m = ti->private; |
1da177e4 LT |
555 | |
556 | /* we need at least a path arg */ | |
557 | if (as->argc < 1) { | |
72d94861 | 558 | ti->error = "no device given"; |
1da177e4 LT |
559 | return NULL; |
560 | } | |
561 | ||
562 | p = alloc_pgpath(); | |
563 | if (!p) | |
564 | return NULL; | |
565 | ||
566 | r = dm_get_device(ti, shift(as), ti->begin, ti->len, | |
567 | dm_table_get_mode(ti->table), &p->path.dev); | |
568 | if (r) { | |
72d94861 | 569 | ti->error = "error getting device"; |
1da177e4 LT |
570 | goto bad; |
571 | } | |
572 | ||
ae11b1b3 HR |
573 | if (m->hw_handler_name) { |
574 | r = scsi_dh_attach(bdev_get_queue(p->path.dev->bdev), | |
575 | m->hw_handler_name); | |
576 | if (r < 0) { | |
577 | dm_put_device(ti, p->path.dev); | |
578 | goto bad; | |
579 | } | |
580 | } | |
581 | ||
1da177e4 LT |
582 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
583 | if (r) { | |
584 | dm_put_device(ti, p->path.dev); | |
585 | goto bad; | |
586 | } | |
587 | ||
588 | return p; | |
589 | ||
590 | bad: | |
591 | free_pgpath(p); | |
592 | return NULL; | |
593 | } | |
594 | ||
595 | static struct priority_group *parse_priority_group(struct arg_set *as, | |
28f16c20 | 596 | struct multipath *m) |
1da177e4 LT |
597 | { |
598 | static struct param _params[] = { | |
72d94861 AK |
599 | {1, 1024, "invalid number of paths"}, |
600 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
601 | }; |
602 | ||
603 | int r; | |
604 | unsigned i, nr_selector_args, nr_params; | |
605 | struct priority_group *pg; | |
28f16c20 | 606 | struct dm_target *ti = m->ti; |
1da177e4 LT |
607 | |
608 | if (as->argc < 2) { | |
609 | as->argc = 0; | |
72d94861 | 610 | ti->error = "not enough priority group aruments"; |
1da177e4 LT |
611 | return NULL; |
612 | } | |
613 | ||
614 | pg = alloc_priority_group(); | |
615 | if (!pg) { | |
72d94861 | 616 | ti->error = "couldn't allocate priority group"; |
1da177e4 LT |
617 | return NULL; |
618 | } | |
619 | pg->m = m; | |
620 | ||
621 | r = parse_path_selector(as, pg, ti); | |
622 | if (r) | |
623 | goto bad; | |
624 | ||
625 | /* | |
626 | * read the paths | |
627 | */ | |
628 | r = read_param(_params, shift(as), &pg->nr_pgpaths, &ti->error); | |
629 | if (r) | |
630 | goto bad; | |
631 | ||
632 | r = read_param(_params + 1, shift(as), &nr_selector_args, &ti->error); | |
633 | if (r) | |
634 | goto bad; | |
635 | ||
636 | nr_params = 1 + nr_selector_args; | |
637 | for (i = 0; i < pg->nr_pgpaths; i++) { | |
638 | struct pgpath *pgpath; | |
639 | struct arg_set path_args; | |
640 | ||
148acff6 MP |
641 | if (as->argc < nr_params) { |
642 | ti->error = "not enough path parameters"; | |
1da177e4 | 643 | goto bad; |
148acff6 | 644 | } |
1da177e4 LT |
645 | |
646 | path_args.argc = nr_params; | |
647 | path_args.argv = as->argv; | |
648 | ||
649 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
650 | if (!pgpath) | |
651 | goto bad; | |
652 | ||
653 | pgpath->pg = pg; | |
654 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
655 | consume(as, nr_params); | |
656 | } | |
657 | ||
658 | return pg; | |
659 | ||
660 | bad: | |
661 | free_priority_group(pg, ti); | |
662 | return NULL; | |
663 | } | |
664 | ||
28f16c20 | 665 | static int parse_hw_handler(struct arg_set *as, struct multipath *m) |
1da177e4 | 666 | { |
1da177e4 | 667 | unsigned hw_argc; |
28f16c20 | 668 | struct dm_target *ti = m->ti; |
1da177e4 LT |
669 | |
670 | static struct param _params[] = { | |
72d94861 | 671 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
672 | }; |
673 | ||
cfae5c9b | 674 | if (read_param(_params, shift(as), &hw_argc, &ti->error)) |
1da177e4 LT |
675 | return -EINVAL; |
676 | ||
677 | if (!hw_argc) | |
678 | return 0; | |
679 | ||
cfae5c9b CS |
680 | m->hw_handler_name = kstrdup(shift(as), GFP_KERNEL); |
681 | request_module("scsi_dh_%s", m->hw_handler_name); | |
682 | if (scsi_dh_handler_exist(m->hw_handler_name) == 0) { | |
72d94861 | 683 | ti->error = "unknown hardware handler type"; |
fe9233fb CS |
684 | kfree(m->hw_handler_name); |
685 | m->hw_handler_name = NULL; | |
1da177e4 LT |
686 | return -EINVAL; |
687 | } | |
1da177e4 LT |
688 | consume(as, hw_argc - 1); |
689 | ||
690 | return 0; | |
691 | } | |
692 | ||
28f16c20 | 693 | static int parse_features(struct arg_set *as, struct multipath *m) |
1da177e4 LT |
694 | { |
695 | int r; | |
696 | unsigned argc; | |
28f16c20 | 697 | struct dm_target *ti = m->ti; |
c9e45581 | 698 | const char *param_name; |
1da177e4 LT |
699 | |
700 | static struct param _params[] = { | |
c9e45581 DW |
701 | {0, 3, "invalid number of feature args"}, |
702 | {1, 50, "pg_init_retries must be between 1 and 50"}, | |
1da177e4 LT |
703 | }; |
704 | ||
705 | r = read_param(_params, shift(as), &argc, &ti->error); | |
706 | if (r) | |
707 | return -EINVAL; | |
708 | ||
709 | if (!argc) | |
710 | return 0; | |
711 | ||
c9e45581 DW |
712 | do { |
713 | param_name = shift(as); | |
714 | argc--; | |
715 | ||
716 | if (!strnicmp(param_name, MESG_STR("queue_if_no_path"))) { | |
717 | r = queue_if_no_path(m, 1, 0); | |
718 | continue; | |
719 | } | |
720 | ||
721 | if (!strnicmp(param_name, MESG_STR("pg_init_retries")) && | |
722 | (argc >= 1)) { | |
723 | r = read_param(_params + 1, shift(as), | |
724 | &m->pg_init_retries, &ti->error); | |
725 | argc--; | |
726 | continue; | |
727 | } | |
728 | ||
1da177e4 | 729 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
730 | r = -EINVAL; |
731 | } while (argc && !r); | |
732 | ||
733 | return r; | |
1da177e4 LT |
734 | } |
735 | ||
736 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
737 | char **argv) | |
738 | { | |
739 | /* target parameters */ | |
740 | static struct param _params[] = { | |
72d94861 AK |
741 | {1, 1024, "invalid number of priority groups"}, |
742 | {1, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
743 | }; |
744 | ||
745 | int r; | |
746 | struct multipath *m; | |
747 | struct arg_set as; | |
748 | unsigned pg_count = 0; | |
749 | unsigned next_pg_num; | |
750 | ||
751 | as.argc = argc; | |
752 | as.argv = argv; | |
753 | ||
28f16c20 | 754 | m = alloc_multipath(ti); |
1da177e4 | 755 | if (!m) { |
72d94861 | 756 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
757 | return -EINVAL; |
758 | } | |
759 | ||
28f16c20 | 760 | r = parse_features(&as, m); |
1da177e4 LT |
761 | if (r) |
762 | goto bad; | |
763 | ||
28f16c20 | 764 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
765 | if (r) |
766 | goto bad; | |
767 | ||
768 | r = read_param(_params, shift(&as), &m->nr_priority_groups, &ti->error); | |
769 | if (r) | |
770 | goto bad; | |
771 | ||
772 | r = read_param(_params + 1, shift(&as), &next_pg_num, &ti->error); | |
773 | if (r) | |
774 | goto bad; | |
775 | ||
776 | /* parse the priority groups */ | |
777 | while (as.argc) { | |
778 | struct priority_group *pg; | |
779 | ||
28f16c20 | 780 | pg = parse_priority_group(&as, m); |
1da177e4 LT |
781 | if (!pg) { |
782 | r = -EINVAL; | |
783 | goto bad; | |
784 | } | |
785 | ||
786 | m->nr_valid_paths += pg->nr_pgpaths; | |
787 | list_add_tail(&pg->list, &m->priority_groups); | |
788 | pg_count++; | |
789 | pg->pg_num = pg_count; | |
790 | if (!--next_pg_num) | |
791 | m->next_pg = pg; | |
792 | } | |
793 | ||
794 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 795 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
796 | r = -EINVAL; |
797 | goto bad; | |
798 | } | |
799 | ||
1da177e4 LT |
800 | return 0; |
801 | ||
802 | bad: | |
803 | free_multipath(m); | |
804 | return r; | |
805 | } | |
806 | ||
807 | static void multipath_dtr(struct dm_target *ti) | |
808 | { | |
809 | struct multipath *m = (struct multipath *) ti->private; | |
a044d016 | 810 | |
bab7cfc7 | 811 | flush_workqueue(kmpath_handlerd); |
a044d016 | 812 | flush_workqueue(kmultipathd); |
1da177e4 LT |
813 | free_multipath(m); |
814 | } | |
815 | ||
816 | /* | |
817 | * Map bios, recording original fields for later in case we have to resubmit | |
818 | */ | |
819 | static int multipath_map(struct dm_target *ti, struct bio *bio, | |
820 | union map_info *map_context) | |
821 | { | |
822 | int r; | |
028867ac | 823 | struct dm_mpath_io *mpio; |
1da177e4 LT |
824 | struct multipath *m = (struct multipath *) ti->private; |
825 | ||
826 | mpio = mempool_alloc(m->mpio_pool, GFP_NOIO); | |
827 | dm_bio_record(&mpio->details, bio); | |
828 | ||
829 | map_context->ptr = mpio; | |
830 | bio->bi_rw |= (1 << BIO_RW_FAILFAST); | |
831 | r = map_io(m, bio, mpio, 0); | |
45e15720 | 832 | if (r < 0 || r == DM_MAPIO_REQUEUE) |
1da177e4 LT |
833 | mempool_free(mpio, m->mpio_pool); |
834 | ||
835 | return r; | |
836 | } | |
837 | ||
838 | /* | |
839 | * Take a path out of use. | |
840 | */ | |
841 | static int fail_path(struct pgpath *pgpath) | |
842 | { | |
843 | unsigned long flags; | |
844 | struct multipath *m = pgpath->pg->m; | |
845 | ||
846 | spin_lock_irqsave(&m->lock, flags); | |
847 | ||
848 | if (!pgpath->path.is_active) | |
849 | goto out; | |
850 | ||
72d94861 | 851 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
852 | |
853 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
854 | pgpath->path.is_active = 0; | |
855 | pgpath->fail_count++; | |
856 | ||
857 | m->nr_valid_paths--; | |
858 | ||
859 | if (pgpath == m->current_pgpath) | |
860 | m->current_pgpath = NULL; | |
861 | ||
b15546f9 MA |
862 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
863 | pgpath->path.dev->name, m->nr_valid_paths); | |
864 | ||
c557308e | 865 | queue_work(kmultipathd, &m->trigger_event); |
1da177e4 LT |
866 | |
867 | out: | |
868 | spin_unlock_irqrestore(&m->lock, flags); | |
869 | ||
870 | return 0; | |
871 | } | |
872 | ||
873 | /* | |
874 | * Reinstate a previously-failed path | |
875 | */ | |
876 | static int reinstate_path(struct pgpath *pgpath) | |
877 | { | |
878 | int r = 0; | |
879 | unsigned long flags; | |
880 | struct multipath *m = pgpath->pg->m; | |
881 | ||
882 | spin_lock_irqsave(&m->lock, flags); | |
883 | ||
884 | if (pgpath->path.is_active) | |
885 | goto out; | |
886 | ||
def052d2 | 887 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
888 | DMWARN("Reinstate path not supported by path selector %s", |
889 | pgpath->pg->ps.type->name); | |
890 | r = -EINVAL; | |
891 | goto out; | |
892 | } | |
893 | ||
894 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
895 | if (r) | |
896 | goto out; | |
897 | ||
898 | pgpath->path.is_active = 1; | |
899 | ||
900 | m->current_pgpath = NULL; | |
c3cd4f6b | 901 | if (!m->nr_valid_paths++ && m->queue_size) |
c557308e | 902 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 903 | |
b15546f9 MA |
904 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
905 | pgpath->path.dev->name, m->nr_valid_paths); | |
906 | ||
c557308e | 907 | queue_work(kmultipathd, &m->trigger_event); |
1da177e4 LT |
908 | |
909 | out: | |
910 | spin_unlock_irqrestore(&m->lock, flags); | |
911 | ||
912 | return r; | |
913 | } | |
914 | ||
915 | /* | |
916 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
917 | */ | |
918 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
919 | action_fn action) | |
920 | { | |
921 | int r = 0; | |
922 | struct pgpath *pgpath; | |
923 | struct priority_group *pg; | |
924 | ||
925 | list_for_each_entry(pg, &m->priority_groups, list) { | |
926 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
927 | if (pgpath->path.dev == dev) | |
928 | r = action(pgpath); | |
929 | } | |
930 | } | |
931 | ||
932 | return r; | |
933 | } | |
934 | ||
935 | /* | |
936 | * Temporarily try to avoid having to use the specified PG | |
937 | */ | |
938 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
939 | int bypassed) | |
940 | { | |
941 | unsigned long flags; | |
942 | ||
943 | spin_lock_irqsave(&m->lock, flags); | |
944 | ||
945 | pg->bypassed = bypassed; | |
946 | m->current_pgpath = NULL; | |
947 | m->current_pg = NULL; | |
948 | ||
949 | spin_unlock_irqrestore(&m->lock, flags); | |
950 | ||
c557308e | 951 | queue_work(kmultipathd, &m->trigger_event); |
1da177e4 LT |
952 | } |
953 | ||
954 | /* | |
955 | * Switch to using the specified PG from the next I/O that gets mapped | |
956 | */ | |
957 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
958 | { | |
959 | struct priority_group *pg; | |
960 | unsigned pgnum; | |
961 | unsigned long flags; | |
962 | ||
963 | if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || | |
964 | (pgnum > m->nr_priority_groups)) { | |
965 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
966 | return -EINVAL; | |
967 | } | |
968 | ||
969 | spin_lock_irqsave(&m->lock, flags); | |
970 | list_for_each_entry(pg, &m->priority_groups, list) { | |
971 | pg->bypassed = 0; | |
972 | if (--pgnum) | |
973 | continue; | |
974 | ||
975 | m->current_pgpath = NULL; | |
976 | m->current_pg = NULL; | |
977 | m->next_pg = pg; | |
978 | } | |
979 | spin_unlock_irqrestore(&m->lock, flags); | |
980 | ||
c557308e | 981 | queue_work(kmultipathd, &m->trigger_event); |
1da177e4 LT |
982 | return 0; |
983 | } | |
984 | ||
985 | /* | |
986 | * Set/clear bypassed status of a PG. | |
987 | * PGs are numbered upwards from 1 in the order they were declared. | |
988 | */ | |
989 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
990 | { | |
991 | struct priority_group *pg; | |
992 | unsigned pgnum; | |
993 | ||
994 | if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || | |
995 | (pgnum > m->nr_priority_groups)) { | |
996 | DMWARN("invalid PG number supplied to bypass_pg"); | |
997 | return -EINVAL; | |
998 | } | |
999 | ||
1000 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1001 | if (!--pgnum) | |
1002 | break; | |
1003 | } | |
1004 | ||
1005 | bypass_pg(m, pg, bypassed); | |
1006 | return 0; | |
1007 | } | |
1008 | ||
c9e45581 DW |
1009 | /* |
1010 | * Should we retry pg_init immediately? | |
1011 | */ | |
1012 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1013 | { | |
1014 | unsigned long flags; | |
1015 | int limit_reached = 0; | |
1016 | ||
1017 | spin_lock_irqsave(&m->lock, flags); | |
1018 | ||
1019 | if (m->pg_init_count <= m->pg_init_retries) | |
1020 | m->pg_init_required = 1; | |
1021 | else | |
1022 | limit_reached = 1; | |
1023 | ||
1024 | spin_unlock_irqrestore(&m->lock, flags); | |
1025 | ||
1026 | return limit_reached; | |
1027 | } | |
1028 | ||
cfae5c9b CS |
1029 | static void pg_init_done(struct dm_path *path, int errors) |
1030 | { | |
1031 | struct pgpath *pgpath = path_to_pgpath(path); | |
1032 | struct priority_group *pg = pgpath->pg; | |
1033 | struct multipath *m = pg->m; | |
1034 | unsigned long flags; | |
1035 | ||
1036 | /* device or driver problems */ | |
1037 | switch (errors) { | |
1038 | case SCSI_DH_OK: | |
1039 | break; | |
1040 | case SCSI_DH_NOSYS: | |
1041 | if (!m->hw_handler_name) { | |
1042 | errors = 0; | |
1043 | break; | |
1044 | } | |
1045 | DMERR("Cannot failover device because scsi_dh_%s was not " | |
1046 | "loaded.", m->hw_handler_name); | |
1047 | /* | |
1048 | * Fail path for now, so we do not ping pong | |
1049 | */ | |
1050 | fail_path(pgpath); | |
1051 | break; | |
1052 | case SCSI_DH_DEV_TEMP_BUSY: | |
1053 | /* | |
1054 | * Probably doing something like FW upgrade on the | |
1055 | * controller so try the other pg. | |
1056 | */ | |
1057 | bypass_pg(m, pg, 1); | |
1058 | break; | |
1059 | /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */ | |
1060 | case SCSI_DH_RETRY: | |
1061 | case SCSI_DH_IMM_RETRY: | |
1062 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1063 | if (pg_init_limit_reached(m, pgpath)) | |
1064 | fail_path(pgpath); | |
1065 | errors = 0; | |
1066 | break; | |
1067 | default: | |
1068 | /* | |
1069 | * We probably do not want to fail the path for a device | |
1070 | * error, but this is what the old dm did. In future | |
1071 | * patches we can do more advanced handling. | |
1072 | */ | |
1073 | fail_path(pgpath); | |
1074 | } | |
1075 | ||
1076 | spin_lock_irqsave(&m->lock, flags); | |
1077 | if (errors) { | |
1078 | DMERR("Could not failover device. Error %d.", errors); | |
1079 | m->current_pgpath = NULL; | |
1080 | m->current_pg = NULL; | |
1081 | } else if (!m->pg_init_required) { | |
1082 | m->queue_io = 0; | |
1083 | pg->bypassed = 0; | |
1084 | } | |
1085 | ||
1086 | m->pg_init_in_progress = 0; | |
1087 | queue_work(kmultipathd, &m->process_queued_ios); | |
1088 | spin_unlock_irqrestore(&m->lock, flags); | |
1089 | } | |
1090 | ||
bab7cfc7 CS |
1091 | static void activate_path(struct work_struct *work) |
1092 | { | |
1093 | int ret; | |
1094 | struct multipath *m = | |
1095 | container_of(work, struct multipath, activate_path); | |
1096 | struct dm_path *path = &m->current_pgpath->path; | |
1097 | ||
1098 | ret = scsi_dh_activate(bdev_get_queue(path->dev->bdev)); | |
1099 | pg_init_done(path, ret); | |
1100 | } | |
1101 | ||
1da177e4 LT |
1102 | /* |
1103 | * end_io handling | |
1104 | */ | |
1105 | static int do_end_io(struct multipath *m, struct bio *bio, | |
028867ac | 1106 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1107 | { |
640eb3b0 | 1108 | unsigned long flags; |
1da177e4 LT |
1109 | |
1110 | if (!error) | |
1111 | return 0; /* I/O complete */ | |
1112 | ||
4f58802f LMB |
1113 | if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio)) |
1114 | return error; | |
1115 | ||
f6a80ea8 AK |
1116 | if (error == -EOPNOTSUPP) |
1117 | return error; | |
1118 | ||
640eb3b0 | 1119 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 | 1120 | if (!m->nr_valid_paths) { |
45e15720 KU |
1121 | if (__must_push_back(m)) { |
1122 | spin_unlock_irqrestore(&m->lock, flags); | |
1123 | return DM_ENDIO_REQUEUE; | |
1124 | } else if (!m->queue_if_no_path) { | |
640eb3b0 | 1125 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1126 | return -EIO; |
1127 | } else { | |
640eb3b0 | 1128 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1129 | goto requeue; |
1130 | } | |
1131 | } | |
640eb3b0 | 1132 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1133 | |
cfae5c9b CS |
1134 | if (mpio->pgpath) |
1135 | fail_path(mpio->pgpath); | |
1da177e4 LT |
1136 | |
1137 | requeue: | |
1138 | dm_bio_restore(&mpio->details, bio); | |
1139 | ||
1140 | /* queue for the daemon to resubmit or fail */ | |
640eb3b0 | 1141 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
1142 | bio_list_add(&m->queued_ios, bio); |
1143 | m->queue_size++; | |
1144 | if (!m->queue_io) | |
c557308e | 1145 | queue_work(kmultipathd, &m->process_queued_ios); |
640eb3b0 | 1146 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1147 | |
d2a7ad29 | 1148 | return DM_ENDIO_INCOMPLETE; /* io not complete */ |
1da177e4 LT |
1149 | } |
1150 | ||
1151 | static int multipath_end_io(struct dm_target *ti, struct bio *bio, | |
1152 | int error, union map_info *map_context) | |
1153 | { | |
028867ac AK |
1154 | struct multipath *m = ti->private; |
1155 | struct dm_mpath_io *mpio = map_context->ptr; | |
1da177e4 LT |
1156 | struct pgpath *pgpath = mpio->pgpath; |
1157 | struct path_selector *ps; | |
1158 | int r; | |
1159 | ||
1160 | r = do_end_io(m, bio, error, mpio); | |
1161 | if (pgpath) { | |
1162 | ps = &pgpath->pg->ps; | |
1163 | if (ps->type->end_io) | |
1164 | ps->type->end_io(ps, &pgpath->path); | |
1165 | } | |
d2a7ad29 | 1166 | if (r != DM_ENDIO_INCOMPLETE) |
1da177e4 LT |
1167 | mempool_free(mpio, m->mpio_pool); |
1168 | ||
1169 | return r; | |
1170 | } | |
1171 | ||
1172 | /* | |
1173 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1174 | * the last path fails we must error any remaining I/O. |
1175 | * Note that if the freeze_bdev fails while suspending, the | |
1176 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1177 | */ |
1178 | static void multipath_presuspend(struct dm_target *ti) | |
1179 | { | |
1180 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1181 | |
485ef69e | 1182 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1183 | } |
1184 | ||
436d4108 AK |
1185 | /* |
1186 | * Restore the queue_if_no_path setting. | |
1187 | */ | |
1da177e4 LT |
1188 | static void multipath_resume(struct dm_target *ti) |
1189 | { | |
1190 | struct multipath *m = (struct multipath *) ti->private; | |
1191 | unsigned long flags; | |
1192 | ||
1193 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1194 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1195 | spin_unlock_irqrestore(&m->lock, flags); |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * Info output has the following format: | |
1200 | * num_multipath_feature_args [multipath_feature_args]* | |
1201 | * num_handler_status_args [handler_status_args]* | |
1202 | * num_groups init_group_number | |
1203 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1204 | * num_paths num_selector_args | |
1205 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1206 | * | |
1207 | * Table output has the following format (identical to the constructor string): | |
1208 | * num_feature_args [features_args]* | |
1209 | * num_handler_args hw_handler [hw_handler_args]* | |
1210 | * num_groups init_group_number | |
1211 | * [priority selector-name num_ps_args [ps_args]* | |
1212 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1213 | */ | |
1214 | static int multipath_status(struct dm_target *ti, status_type_t type, | |
1215 | char *result, unsigned int maxlen) | |
1216 | { | |
1217 | int sz = 0; | |
1218 | unsigned long flags; | |
1219 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1220 | struct priority_group *pg; |
1221 | struct pgpath *p; | |
1222 | unsigned pg_num; | |
1223 | char state; | |
1224 | ||
1225 | spin_lock_irqsave(&m->lock, flags); | |
1226 | ||
1227 | /* Features */ | |
1228 | if (type == STATUSTYPE_INFO) | |
c9e45581 DW |
1229 | DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count); |
1230 | else { | |
1231 | DMEMIT("%u ", m->queue_if_no_path + | |
1232 | (m->pg_init_retries > 0) * 2); | |
1233 | if (m->queue_if_no_path) | |
1234 | DMEMIT("queue_if_no_path "); | |
1235 | if (m->pg_init_retries) | |
1236 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
1237 | } | |
1da177e4 | 1238 | |
cfae5c9b | 1239 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1240 | DMEMIT("0 "); |
1241 | else | |
cfae5c9b | 1242 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1243 | |
1244 | DMEMIT("%u ", m->nr_priority_groups); | |
1245 | ||
1246 | if (m->next_pg) | |
1247 | pg_num = m->next_pg->pg_num; | |
1248 | else if (m->current_pg) | |
1249 | pg_num = m->current_pg->pg_num; | |
1250 | else | |
1251 | pg_num = 1; | |
1252 | ||
1253 | DMEMIT("%u ", pg_num); | |
1254 | ||
1255 | switch (type) { | |
1256 | case STATUSTYPE_INFO: | |
1257 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1258 | if (pg->bypassed) | |
1259 | state = 'D'; /* Disabled */ | |
1260 | else if (pg == m->current_pg) | |
1261 | state = 'A'; /* Currently Active */ | |
1262 | else | |
1263 | state = 'E'; /* Enabled */ | |
1264 | ||
1265 | DMEMIT("%c ", state); | |
1266 | ||
1267 | if (pg->ps.type->status) | |
1268 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1269 | result + sz, | |
1270 | maxlen - sz); | |
1271 | else | |
1272 | DMEMIT("0 "); | |
1273 | ||
1274 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1275 | pg->ps.type->info_args); | |
1276 | ||
1277 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1278 | DMEMIT("%s %s %u ", p->path.dev->name, | |
1279 | p->path.is_active ? "A" : "F", | |
1280 | p->fail_count); | |
1281 | if (pg->ps.type->status) | |
1282 | sz += pg->ps.type->status(&pg->ps, | |
1283 | &p->path, type, result + sz, | |
1284 | maxlen - sz); | |
1285 | } | |
1286 | } | |
1287 | break; | |
1288 | ||
1289 | case STATUSTYPE_TABLE: | |
1290 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1291 | DMEMIT("%s ", pg->ps.type->name); | |
1292 | ||
1293 | if (pg->ps.type->status) | |
1294 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1295 | result + sz, | |
1296 | maxlen - sz); | |
1297 | else | |
1298 | DMEMIT("0 "); | |
1299 | ||
1300 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1301 | pg->ps.type->table_args); | |
1302 | ||
1303 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1304 | DMEMIT("%s ", p->path.dev->name); | |
1305 | if (pg->ps.type->status) | |
1306 | sz += pg->ps.type->status(&pg->ps, | |
1307 | &p->path, type, result + sz, | |
1308 | maxlen - sz); | |
1309 | } | |
1310 | } | |
1311 | break; | |
1312 | } | |
1313 | ||
1314 | spin_unlock_irqrestore(&m->lock, flags); | |
1315 | ||
1316 | return 0; | |
1317 | } | |
1318 | ||
1319 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1320 | { | |
1321 | int r; | |
1322 | struct dm_dev *dev; | |
1323 | struct multipath *m = (struct multipath *) ti->private; | |
1324 | action_fn action; | |
1325 | ||
1326 | if (argc == 1) { | |
1327 | if (!strnicmp(argv[0], MESG_STR("queue_if_no_path"))) | |
485ef69e | 1328 | return queue_if_no_path(m, 1, 0); |
1da177e4 | 1329 | else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path"))) |
485ef69e | 1330 | return queue_if_no_path(m, 0, 0); |
1da177e4 LT |
1331 | } |
1332 | ||
1333 | if (argc != 2) | |
1334 | goto error; | |
1335 | ||
1336 | if (!strnicmp(argv[0], MESG_STR("disable_group"))) | |
1337 | return bypass_pg_num(m, argv[1], 1); | |
1338 | else if (!strnicmp(argv[0], MESG_STR("enable_group"))) | |
1339 | return bypass_pg_num(m, argv[1], 0); | |
1340 | else if (!strnicmp(argv[0], MESG_STR("switch_group"))) | |
1341 | return switch_pg_num(m, argv[1]); | |
1342 | else if (!strnicmp(argv[0], MESG_STR("reinstate_path"))) | |
1343 | action = reinstate_path; | |
1344 | else if (!strnicmp(argv[0], MESG_STR("fail_path"))) | |
1345 | action = fail_path; | |
1346 | else | |
1347 | goto error; | |
1348 | ||
1349 | r = dm_get_device(ti, argv[1], ti->begin, ti->len, | |
1350 | dm_table_get_mode(ti->table), &dev); | |
1351 | if (r) { | |
72d94861 | 1352 | DMWARN("message: error getting device %s", |
1da177e4 LT |
1353 | argv[1]); |
1354 | return -EINVAL; | |
1355 | } | |
1356 | ||
1357 | r = action_dev(m, dev, action); | |
1358 | ||
1359 | dm_put_device(ti, dev); | |
1360 | ||
1361 | return r; | |
1362 | ||
1363 | error: | |
1364 | DMWARN("Unrecognised multipath message received."); | |
1365 | return -EINVAL; | |
1366 | } | |
1367 | ||
9af4aa30 MB |
1368 | static int multipath_ioctl(struct dm_target *ti, struct inode *inode, |
1369 | struct file *filp, unsigned int cmd, | |
1370 | unsigned long arg) | |
1371 | { | |
1372 | struct multipath *m = (struct multipath *) ti->private; | |
1373 | struct block_device *bdev = NULL; | |
1374 | unsigned long flags; | |
e90dae1f MB |
1375 | struct file fake_file = {}; |
1376 | struct dentry fake_dentry = {}; | |
9af4aa30 MB |
1377 | int r = 0; |
1378 | ||
c649bb9c | 1379 | fake_file.f_path.dentry = &fake_dentry; |
e90dae1f | 1380 | |
9af4aa30 MB |
1381 | spin_lock_irqsave(&m->lock, flags); |
1382 | ||
1383 | if (!m->current_pgpath) | |
1384 | __choose_pgpath(m); | |
1385 | ||
e90dae1f | 1386 | if (m->current_pgpath) { |
9af4aa30 | 1387 | bdev = m->current_pgpath->path.dev->bdev; |
e90dae1f MB |
1388 | fake_dentry.d_inode = bdev->bd_inode; |
1389 | fake_file.f_mode = m->current_pgpath->path.dev->mode; | |
1390 | } | |
9af4aa30 MB |
1391 | |
1392 | if (m->queue_io) | |
1393 | r = -EAGAIN; | |
1394 | else if (!bdev) | |
1395 | r = -EIO; | |
1396 | ||
1397 | spin_unlock_irqrestore(&m->lock, flags); | |
1398 | ||
e90dae1f MB |
1399 | return r ? : blkdev_driver_ioctl(bdev->bd_inode, &fake_file, |
1400 | bdev->bd_disk, cmd, arg); | |
9af4aa30 MB |
1401 | } |
1402 | ||
1da177e4 LT |
1403 | /*----------------------------------------------------------------- |
1404 | * Module setup | |
1405 | *---------------------------------------------------------------*/ | |
1406 | static struct target_type multipath_target = { | |
1407 | .name = "multipath", | |
9af4aa30 | 1408 | .version = {1, 0, 5}, |
1da177e4 LT |
1409 | .module = THIS_MODULE, |
1410 | .ctr = multipath_ctr, | |
1411 | .dtr = multipath_dtr, | |
1412 | .map = multipath_map, | |
1413 | .end_io = multipath_end_io, | |
1414 | .presuspend = multipath_presuspend, | |
1415 | .resume = multipath_resume, | |
1416 | .status = multipath_status, | |
1417 | .message = multipath_message, | |
9af4aa30 | 1418 | .ioctl = multipath_ioctl, |
1da177e4 LT |
1419 | }; |
1420 | ||
1421 | static int __init dm_multipath_init(void) | |
1422 | { | |
1423 | int r; | |
1424 | ||
1425 | /* allocate a slab for the dm_ios */ | |
028867ac | 1426 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1427 | if (!_mpio_cache) |
1428 | return -ENOMEM; | |
1429 | ||
1430 | r = dm_register_target(&multipath_target); | |
1431 | if (r < 0) { | |
0cd33124 | 1432 | DMERR("register failed %d", r); |
1da177e4 LT |
1433 | kmem_cache_destroy(_mpio_cache); |
1434 | return -EINVAL; | |
1435 | } | |
1436 | ||
c557308e AK |
1437 | kmultipathd = create_workqueue("kmpathd"); |
1438 | if (!kmultipathd) { | |
0cd33124 | 1439 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1440 | dm_unregister_target(&multipath_target); |
1441 | kmem_cache_destroy(_mpio_cache); | |
1442 | return -ENOMEM; | |
1443 | } | |
1444 | ||
bab7cfc7 CS |
1445 | /* |
1446 | * A separate workqueue is used to handle the device handlers | |
1447 | * to avoid overloading existing workqueue. Overloading the | |
1448 | * old workqueue would also create a bottleneck in the | |
1449 | * path of the storage hardware device activation. | |
1450 | */ | |
1451 | kmpath_handlerd = create_singlethread_workqueue("kmpath_handlerd"); | |
1452 | if (!kmpath_handlerd) { | |
1453 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1454 | destroy_workqueue(kmultipathd); | |
1455 | dm_unregister_target(&multipath_target); | |
1456 | kmem_cache_destroy(_mpio_cache); | |
1457 | return -ENOMEM; | |
1458 | } | |
1459 | ||
72d94861 | 1460 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1461 | multipath_target.version[0], multipath_target.version[1], |
1462 | multipath_target.version[2]); | |
1463 | ||
1464 | return r; | |
1465 | } | |
1466 | ||
1467 | static void __exit dm_multipath_exit(void) | |
1468 | { | |
1469 | int r; | |
1470 | ||
bab7cfc7 | 1471 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1472 | destroy_workqueue(kmultipathd); |
1473 | ||
1da177e4 LT |
1474 | r = dm_unregister_target(&multipath_target); |
1475 | if (r < 0) | |
0cd33124 | 1476 | DMERR("target unregister failed %d", r); |
1da177e4 LT |
1477 | kmem_cache_destroy(_mpio_cache); |
1478 | } | |
1479 | ||
1da177e4 LT |
1480 | module_init(dm_multipath_init); |
1481 | module_exit(dm_multipath_exit); | |
1482 | ||
1483 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1484 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1485 | MODULE_LICENSE("GPL"); |