]>
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; |
7253a334 | 66 | struct pgpath *pgpath_to_activate; |
1da177e4 LT |
67 | unsigned nr_priority_groups; |
68 | struct list_head priority_groups; | |
69 | unsigned pg_init_required; /* pg_init needs calling? */ | |
c3cd4f6b | 70 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
1da177e4 LT |
71 | |
72 | unsigned nr_valid_paths; /* Total number of usable paths */ | |
73 | struct pgpath *current_pgpath; | |
74 | struct priority_group *current_pg; | |
75 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
76 | unsigned repeat_count; /* I/Os left before calling PS again */ | |
77 | ||
78 | unsigned queue_io; /* Must we queue all I/O? */ | |
79 | unsigned queue_if_no_path; /* Queue I/O if last path fails? */ | |
436d4108 | 80 | unsigned saved_queue_if_no_path;/* Saved state during suspension */ |
c9e45581 DW |
81 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
82 | unsigned pg_init_count; /* Number of times pg_init called */ | |
1da177e4 LT |
83 | |
84 | struct work_struct process_queued_ios; | |
85 | struct bio_list queued_ios; | |
86 | unsigned queue_size; | |
87 | ||
88 | struct work_struct trigger_event; | |
89 | ||
90 | /* | |
028867ac | 91 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
92 | * can resubmit bios on error. |
93 | */ | |
94 | mempool_t *mpio_pool; | |
95 | }; | |
96 | ||
97 | /* | |
98 | * Context information attached to each bio we process. | |
99 | */ | |
028867ac | 100 | struct dm_mpath_io { |
1da177e4 LT |
101 | struct pgpath *pgpath; |
102 | struct dm_bio_details details; | |
103 | }; | |
104 | ||
105 | typedef int (*action_fn) (struct pgpath *pgpath); | |
106 | ||
107 | #define MIN_IOS 256 /* Mempool size */ | |
108 | ||
e18b890b | 109 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 110 | |
bab7cfc7 | 111 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 DH |
112 | static void process_queued_ios(struct work_struct *work); |
113 | static void trigger_event(struct work_struct *work); | |
bab7cfc7 | 114 | static void activate_path(struct work_struct *work); |
1da177e4 LT |
115 | |
116 | ||
117 | /*----------------------------------------------- | |
118 | * Allocation routines | |
119 | *-----------------------------------------------*/ | |
120 | ||
121 | static struct pgpath *alloc_pgpath(void) | |
122 | { | |
e69fae56 | 123 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 124 | |
e69fae56 | 125 | if (pgpath) |
1da177e4 | 126 | pgpath->path.is_active = 1; |
1da177e4 LT |
127 | |
128 | return pgpath; | |
129 | } | |
130 | ||
028867ac | 131 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
132 | { |
133 | kfree(pgpath); | |
134 | } | |
135 | ||
136 | static struct priority_group *alloc_priority_group(void) | |
137 | { | |
138 | struct priority_group *pg; | |
139 | ||
e69fae56 | 140 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 141 | |
e69fae56 MM |
142 | if (pg) |
143 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
144 | |
145 | return pg; | |
146 | } | |
147 | ||
148 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
149 | { | |
7253a334 | 150 | unsigned long flags; |
1da177e4 | 151 | struct pgpath *pgpath, *tmp; |
ae11b1b3 | 152 | struct multipath *m = ti->private; |
1da177e4 LT |
153 | |
154 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
155 | list_del(&pgpath->list); | |
ae11b1b3 HR |
156 | if (m->hw_handler_name) |
157 | scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev)); | |
1da177e4 | 158 | dm_put_device(ti, pgpath->path.dev); |
7253a334 CS |
159 | spin_lock_irqsave(&m->lock, flags); |
160 | if (m->pgpath_to_activate == pgpath) | |
161 | m->pgpath_to_activate = NULL; | |
162 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
163 | free_pgpath(pgpath); |
164 | } | |
165 | } | |
166 | ||
167 | static void free_priority_group(struct priority_group *pg, | |
168 | struct dm_target *ti) | |
169 | { | |
170 | struct path_selector *ps = &pg->ps; | |
171 | ||
172 | if (ps->type) { | |
173 | ps->type->destroy(ps); | |
174 | dm_put_path_selector(ps->type); | |
175 | } | |
176 | ||
177 | free_pgpaths(&pg->pgpaths, ti); | |
178 | kfree(pg); | |
179 | } | |
180 | ||
28f16c20 | 181 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
182 | { |
183 | struct multipath *m; | |
184 | ||
e69fae56 | 185 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 186 | if (m) { |
1da177e4 LT |
187 | INIT_LIST_HEAD(&m->priority_groups); |
188 | spin_lock_init(&m->lock); | |
189 | m->queue_io = 1; | |
c4028958 DH |
190 | INIT_WORK(&m->process_queued_ios, process_queued_ios); |
191 | INIT_WORK(&m->trigger_event, trigger_event); | |
bab7cfc7 | 192 | INIT_WORK(&m->activate_path, activate_path); |
93d2341c | 193 | m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache); |
1da177e4 LT |
194 | if (!m->mpio_pool) { |
195 | kfree(m); | |
196 | return NULL; | |
197 | } | |
28f16c20 MM |
198 | m->ti = ti; |
199 | ti->private = m; | |
1da177e4 LT |
200 | } |
201 | ||
202 | return m; | |
203 | } | |
204 | ||
205 | static void free_multipath(struct multipath *m) | |
206 | { | |
207 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
208 | |
209 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
210 | list_del(&pg->list); | |
211 | free_priority_group(pg, m->ti); | |
212 | } | |
213 | ||
cfae5c9b | 214 | kfree(m->hw_handler_name); |
1da177e4 LT |
215 | mempool_destroy(m->mpio_pool); |
216 | kfree(m); | |
217 | } | |
218 | ||
219 | ||
220 | /*----------------------------------------------- | |
221 | * Path selection | |
222 | *-----------------------------------------------*/ | |
223 | ||
224 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) | |
225 | { | |
1da177e4 LT |
226 | m->current_pg = pgpath->pg; |
227 | ||
228 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 229 | if (m->hw_handler_name) { |
1da177e4 LT |
230 | m->pg_init_required = 1; |
231 | m->queue_io = 1; | |
232 | } else { | |
233 | m->pg_init_required = 0; | |
234 | m->queue_io = 0; | |
235 | } | |
c9e45581 DW |
236 | |
237 | m->pg_init_count = 0; | |
1da177e4 LT |
238 | } |
239 | ||
240 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg) | |
241 | { | |
c922d5f7 | 242 | struct dm_path *path; |
1da177e4 LT |
243 | |
244 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count); | |
245 | if (!path) | |
246 | return -ENXIO; | |
247 | ||
248 | m->current_pgpath = path_to_pgpath(path); | |
249 | ||
250 | if (m->current_pg != pg) | |
251 | __switch_pg(m, m->current_pgpath); | |
252 | ||
253 | return 0; | |
254 | } | |
255 | ||
256 | static void __choose_pgpath(struct multipath *m) | |
257 | { | |
258 | struct priority_group *pg; | |
259 | unsigned bypassed = 1; | |
260 | ||
261 | if (!m->nr_valid_paths) | |
262 | goto failed; | |
263 | ||
264 | /* Were we instructed to switch PG? */ | |
265 | if (m->next_pg) { | |
266 | pg = m->next_pg; | |
267 | m->next_pg = NULL; | |
268 | if (!__choose_path_in_pg(m, pg)) | |
269 | return; | |
270 | } | |
271 | ||
272 | /* Don't change PG until it has no remaining paths */ | |
273 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg)) | |
274 | return; | |
275 | ||
276 | /* | |
277 | * Loop through priority groups until we find a valid path. | |
278 | * First time we skip PGs marked 'bypassed'. | |
279 | * Second time we only try the ones we skipped. | |
280 | */ | |
281 | do { | |
282 | list_for_each_entry(pg, &m->priority_groups, list) { | |
283 | if (pg->bypassed == bypassed) | |
284 | continue; | |
285 | if (!__choose_path_in_pg(m, pg)) | |
286 | return; | |
287 | } | |
288 | } while (bypassed--); | |
289 | ||
290 | failed: | |
291 | m->current_pgpath = NULL; | |
292 | m->current_pg = NULL; | |
293 | } | |
294 | ||
45e15720 KU |
295 | /* |
296 | * Check whether bios must be queued in the device-mapper core rather | |
297 | * than here in the target. | |
298 | * | |
299 | * m->lock must be held on entry. | |
300 | * | |
301 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
302 | * same value then we are not between multipath_presuspend() | |
303 | * and multipath_resume() calls and we have no need to check | |
304 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
305 | */ | |
306 | static int __must_push_back(struct multipath *m) | |
307 | { | |
308 | return (m->queue_if_no_path != m->saved_queue_if_no_path && | |
309 | dm_noflush_suspending(m->ti)); | |
310 | } | |
311 | ||
028867ac AK |
312 | static int map_io(struct multipath *m, struct bio *bio, |
313 | struct dm_mpath_io *mpio, unsigned was_queued) | |
1da177e4 | 314 | { |
d2a7ad29 | 315 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
316 | unsigned long flags; |
317 | struct pgpath *pgpath; | |
318 | ||
319 | spin_lock_irqsave(&m->lock, flags); | |
320 | ||
321 | /* Do we need to select a new pgpath? */ | |
322 | if (!m->current_pgpath || | |
323 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
324 | __choose_pgpath(m); | |
325 | ||
326 | pgpath = m->current_pgpath; | |
327 | ||
328 | if (was_queued) | |
329 | m->queue_size--; | |
330 | ||
331 | if ((pgpath && m->queue_io) || | |
436d4108 | 332 | (!pgpath && m->queue_if_no_path)) { |
1da177e4 LT |
333 | /* Queue for the daemon to resubmit */ |
334 | bio_list_add(&m->queued_ios, bio); | |
335 | m->queue_size++; | |
c3cd4f6b AK |
336 | if ((m->pg_init_required && !m->pg_init_in_progress) || |
337 | !m->queue_io) | |
c557308e | 338 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 339 | pgpath = NULL; |
d2a7ad29 | 340 | r = DM_MAPIO_SUBMITTED; |
45e15720 | 341 | } else if (pgpath) |
1da177e4 | 342 | bio->bi_bdev = pgpath->path.dev->bdev; |
45e15720 KU |
343 | else if (__must_push_back(m)) |
344 | r = DM_MAPIO_REQUEUE; | |
345 | else | |
346 | r = -EIO; /* Failed */ | |
1da177e4 LT |
347 | |
348 | mpio->pgpath = pgpath; | |
349 | ||
350 | spin_unlock_irqrestore(&m->lock, flags); | |
351 | ||
352 | return r; | |
353 | } | |
354 | ||
355 | /* | |
356 | * If we run out of usable paths, should we queue I/O or error it? | |
357 | */ | |
485ef69e AK |
358 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
359 | unsigned save_old_value) | |
1da177e4 LT |
360 | { |
361 | unsigned long flags; | |
362 | ||
363 | spin_lock_irqsave(&m->lock, flags); | |
364 | ||
485ef69e AK |
365 | if (save_old_value) |
366 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
367 | else | |
368 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 369 | m->queue_if_no_path = queue_if_no_path; |
c3cd4f6b | 370 | if (!m->queue_if_no_path && m->queue_size) |
c557308e | 371 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 LT |
372 | |
373 | spin_unlock_irqrestore(&m->lock, flags); | |
374 | ||
375 | return 0; | |
376 | } | |
377 | ||
378 | /*----------------------------------------------------------------- | |
379 | * The multipath daemon is responsible for resubmitting queued ios. | |
380 | *---------------------------------------------------------------*/ | |
381 | ||
382 | static void dispatch_queued_ios(struct multipath *m) | |
383 | { | |
384 | int r; | |
385 | unsigned long flags; | |
386 | struct bio *bio = NULL, *next; | |
028867ac | 387 | struct dm_mpath_io *mpio; |
1da177e4 LT |
388 | union map_info *info; |
389 | ||
390 | spin_lock_irqsave(&m->lock, flags); | |
391 | bio = bio_list_get(&m->queued_ios); | |
392 | spin_unlock_irqrestore(&m->lock, flags); | |
393 | ||
394 | while (bio) { | |
395 | next = bio->bi_next; | |
396 | bio->bi_next = NULL; | |
397 | ||
398 | info = dm_get_mapinfo(bio); | |
399 | mpio = info->ptr; | |
400 | ||
401 | r = map_io(m, bio, mpio, 1); | |
402 | if (r < 0) | |
6712ecf8 | 403 | bio_endio(bio, r); |
d2a7ad29 | 404 | else if (r == DM_MAPIO_REMAPPED) |
1da177e4 | 405 | generic_make_request(bio); |
45e15720 | 406 | else if (r == DM_MAPIO_REQUEUE) |
6712ecf8 | 407 | bio_endio(bio, -EIO); |
1da177e4 LT |
408 | |
409 | bio = next; | |
410 | } | |
411 | } | |
412 | ||
c4028958 | 413 | static void process_queued_ios(struct work_struct *work) |
1da177e4 | 414 | { |
c4028958 DH |
415 | struct multipath *m = |
416 | container_of(work, struct multipath, process_queued_ios); | |
c3cd4f6b AK |
417 | struct pgpath *pgpath = NULL; |
418 | unsigned init_required = 0, must_queue = 1; | |
1da177e4 LT |
419 | unsigned long flags; |
420 | ||
421 | spin_lock_irqsave(&m->lock, flags); | |
422 | ||
c3cd4f6b AK |
423 | if (!m->queue_size) |
424 | goto out; | |
425 | ||
1da177e4 LT |
426 | if (!m->current_pgpath) |
427 | __choose_pgpath(m); | |
428 | ||
429 | pgpath = m->current_pgpath; | |
7253a334 | 430 | m->pgpath_to_activate = m->current_pgpath; |
1da177e4 | 431 | |
c3cd4f6b AK |
432 | if ((pgpath && !m->queue_io) || |
433 | (!pgpath && !m->queue_if_no_path)) | |
434 | must_queue = 0; | |
1da177e4 | 435 | |
c3cd4f6b | 436 | if (m->pg_init_required && !m->pg_init_in_progress) { |
c9e45581 | 437 | m->pg_init_count++; |
1da177e4 | 438 | m->pg_init_required = 0; |
c3cd4f6b AK |
439 | m->pg_init_in_progress = 1; |
440 | init_required = 1; | |
441 | } | |
1da177e4 | 442 | |
c3cd4f6b | 443 | out: |
1da177e4 LT |
444 | spin_unlock_irqrestore(&m->lock, flags); |
445 | ||
bab7cfc7 CS |
446 | if (init_required) |
447 | queue_work(kmpath_handlerd, &m->activate_path); | |
1da177e4 LT |
448 | |
449 | if (!must_queue) | |
450 | dispatch_queued_ios(m); | |
451 | } | |
452 | ||
453 | /* | |
454 | * An event is triggered whenever a path is taken out of use. | |
455 | * Includes path failure and PG bypass. | |
456 | */ | |
c4028958 | 457 | static void trigger_event(struct work_struct *work) |
1da177e4 | 458 | { |
c4028958 DH |
459 | struct multipath *m = |
460 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
461 | |
462 | dm_table_event(m->ti->table); | |
463 | } | |
464 | ||
465 | /*----------------------------------------------------------------- | |
466 | * Constructor/argument parsing: | |
467 | * <#multipath feature args> [<arg>]* | |
468 | * <#hw_handler args> [hw_handler [<arg>]*] | |
469 | * <#priority groups> | |
470 | * <initial priority group> | |
471 | * [<selector> <#selector args> [<arg>]* | |
472 | * <#paths> <#per-path selector args> | |
473 | * [<path> [<arg>]* ]+ ]+ | |
474 | *---------------------------------------------------------------*/ | |
475 | struct param { | |
476 | unsigned min; | |
477 | unsigned max; | |
478 | char *error; | |
479 | }; | |
480 | ||
1da177e4 LT |
481 | static int read_param(struct param *param, char *str, unsigned *v, char **error) |
482 | { | |
483 | if (!str || | |
484 | (sscanf(str, "%u", v) != 1) || | |
485 | (*v < param->min) || | |
486 | (*v > param->max)) { | |
487 | *error = param->error; | |
488 | return -EINVAL; | |
489 | } | |
490 | ||
491 | return 0; | |
492 | } | |
493 | ||
494 | struct arg_set { | |
495 | unsigned argc; | |
496 | char **argv; | |
497 | }; | |
498 | ||
499 | static char *shift(struct arg_set *as) | |
500 | { | |
501 | char *r; | |
502 | ||
503 | if (as->argc) { | |
504 | as->argc--; | |
505 | r = *as->argv; | |
506 | as->argv++; | |
507 | return r; | |
508 | } | |
509 | ||
510 | return NULL; | |
511 | } | |
512 | ||
513 | static void consume(struct arg_set *as, unsigned n) | |
514 | { | |
515 | BUG_ON (as->argc < n); | |
516 | as->argc -= n; | |
517 | as->argv += n; | |
518 | } | |
519 | ||
520 | static int parse_path_selector(struct arg_set *as, struct priority_group *pg, | |
521 | struct dm_target *ti) | |
522 | { | |
523 | int r; | |
524 | struct path_selector_type *pst; | |
525 | unsigned ps_argc; | |
526 | ||
527 | static struct param _params[] = { | |
72d94861 | 528 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
529 | }; |
530 | ||
531 | pst = dm_get_path_selector(shift(as)); | |
532 | if (!pst) { | |
72d94861 | 533 | ti->error = "unknown path selector type"; |
1da177e4 LT |
534 | return -EINVAL; |
535 | } | |
536 | ||
537 | r = read_param(_params, shift(as), &ps_argc, &ti->error); | |
371b2e34 MP |
538 | if (r) { |
539 | dm_put_path_selector(pst); | |
1da177e4 | 540 | return -EINVAL; |
371b2e34 | 541 | } |
1da177e4 LT |
542 | |
543 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
544 | if (r) { | |
545 | dm_put_path_selector(pst); | |
72d94861 | 546 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
547 | return r; |
548 | } | |
549 | ||
550 | pg->ps.type = pst; | |
551 | consume(as, ps_argc); | |
552 | ||
553 | return 0; | |
554 | } | |
555 | ||
556 | static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps, | |
557 | struct dm_target *ti) | |
558 | { | |
559 | int r; | |
560 | struct pgpath *p; | |
ae11b1b3 | 561 | struct multipath *m = ti->private; |
1da177e4 LT |
562 | |
563 | /* we need at least a path arg */ | |
564 | if (as->argc < 1) { | |
72d94861 | 565 | ti->error = "no device given"; |
01460f35 | 566 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
567 | } |
568 | ||
569 | p = alloc_pgpath(); | |
570 | if (!p) | |
01460f35 | 571 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
572 | |
573 | r = dm_get_device(ti, shift(as), ti->begin, ti->len, | |
574 | dm_table_get_mode(ti->table), &p->path.dev); | |
575 | if (r) { | |
72d94861 | 576 | ti->error = "error getting device"; |
1da177e4 LT |
577 | goto bad; |
578 | } | |
579 | ||
ae11b1b3 HR |
580 | if (m->hw_handler_name) { |
581 | r = scsi_dh_attach(bdev_get_queue(p->path.dev->bdev), | |
582 | m->hw_handler_name); | |
583 | if (r < 0) { | |
584 | dm_put_device(ti, p->path.dev); | |
585 | goto bad; | |
586 | } | |
587 | } | |
588 | ||
1da177e4 LT |
589 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
590 | if (r) { | |
591 | dm_put_device(ti, p->path.dev); | |
592 | goto bad; | |
593 | } | |
594 | ||
595 | return p; | |
596 | ||
597 | bad: | |
598 | free_pgpath(p); | |
01460f35 | 599 | return ERR_PTR(r); |
1da177e4 LT |
600 | } |
601 | ||
602 | static struct priority_group *parse_priority_group(struct arg_set *as, | |
28f16c20 | 603 | struct multipath *m) |
1da177e4 LT |
604 | { |
605 | static struct param _params[] = { | |
72d94861 AK |
606 | {1, 1024, "invalid number of paths"}, |
607 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
608 | }; |
609 | ||
610 | int r; | |
611 | unsigned i, nr_selector_args, nr_params; | |
612 | struct priority_group *pg; | |
28f16c20 | 613 | struct dm_target *ti = m->ti; |
1da177e4 LT |
614 | |
615 | if (as->argc < 2) { | |
616 | as->argc = 0; | |
01460f35 BM |
617 | ti->error = "not enough priority group arguments"; |
618 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
619 | } |
620 | ||
621 | pg = alloc_priority_group(); | |
622 | if (!pg) { | |
72d94861 | 623 | ti->error = "couldn't allocate priority group"; |
01460f35 | 624 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
625 | } |
626 | pg->m = m; | |
627 | ||
628 | r = parse_path_selector(as, pg, ti); | |
629 | if (r) | |
630 | goto bad; | |
631 | ||
632 | /* | |
633 | * read the paths | |
634 | */ | |
635 | r = read_param(_params, shift(as), &pg->nr_pgpaths, &ti->error); | |
636 | if (r) | |
637 | goto bad; | |
638 | ||
639 | r = read_param(_params + 1, shift(as), &nr_selector_args, &ti->error); | |
640 | if (r) | |
641 | goto bad; | |
642 | ||
643 | nr_params = 1 + nr_selector_args; | |
644 | for (i = 0; i < pg->nr_pgpaths; i++) { | |
645 | struct pgpath *pgpath; | |
646 | struct arg_set path_args; | |
647 | ||
148acff6 MP |
648 | if (as->argc < nr_params) { |
649 | ti->error = "not enough path parameters"; | |
1da177e4 | 650 | goto bad; |
148acff6 | 651 | } |
1da177e4 LT |
652 | |
653 | path_args.argc = nr_params; | |
654 | path_args.argv = as->argv; | |
655 | ||
656 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
657 | if (IS_ERR(pgpath)) { |
658 | r = PTR_ERR(pgpath); | |
1da177e4 | 659 | goto bad; |
01460f35 | 660 | } |
1da177e4 LT |
661 | |
662 | pgpath->pg = pg; | |
663 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
664 | consume(as, nr_params); | |
665 | } | |
666 | ||
667 | return pg; | |
668 | ||
669 | bad: | |
670 | free_priority_group(pg, ti); | |
01460f35 | 671 | return ERR_PTR(r); |
1da177e4 LT |
672 | } |
673 | ||
28f16c20 | 674 | static int parse_hw_handler(struct arg_set *as, struct multipath *m) |
1da177e4 | 675 | { |
1da177e4 | 676 | unsigned hw_argc; |
28f16c20 | 677 | struct dm_target *ti = m->ti; |
1da177e4 LT |
678 | |
679 | static struct param _params[] = { | |
72d94861 | 680 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
681 | }; |
682 | ||
cfae5c9b | 683 | if (read_param(_params, shift(as), &hw_argc, &ti->error)) |
1da177e4 LT |
684 | return -EINVAL; |
685 | ||
686 | if (!hw_argc) | |
687 | return 0; | |
688 | ||
cfae5c9b CS |
689 | m->hw_handler_name = kstrdup(shift(as), GFP_KERNEL); |
690 | request_module("scsi_dh_%s", m->hw_handler_name); | |
691 | if (scsi_dh_handler_exist(m->hw_handler_name) == 0) { | |
72d94861 | 692 | ti->error = "unknown hardware handler type"; |
fe9233fb CS |
693 | kfree(m->hw_handler_name); |
694 | m->hw_handler_name = NULL; | |
1da177e4 LT |
695 | return -EINVAL; |
696 | } | |
1da177e4 LT |
697 | consume(as, hw_argc - 1); |
698 | ||
699 | return 0; | |
700 | } | |
701 | ||
28f16c20 | 702 | static int parse_features(struct arg_set *as, struct multipath *m) |
1da177e4 LT |
703 | { |
704 | int r; | |
705 | unsigned argc; | |
28f16c20 | 706 | struct dm_target *ti = m->ti; |
c9e45581 | 707 | const char *param_name; |
1da177e4 LT |
708 | |
709 | static struct param _params[] = { | |
c9e45581 DW |
710 | {0, 3, "invalid number of feature args"}, |
711 | {1, 50, "pg_init_retries must be between 1 and 50"}, | |
1da177e4 LT |
712 | }; |
713 | ||
714 | r = read_param(_params, shift(as), &argc, &ti->error); | |
715 | if (r) | |
716 | return -EINVAL; | |
717 | ||
718 | if (!argc) | |
719 | return 0; | |
720 | ||
c9e45581 DW |
721 | do { |
722 | param_name = shift(as); | |
723 | argc--; | |
724 | ||
725 | if (!strnicmp(param_name, MESG_STR("queue_if_no_path"))) { | |
726 | r = queue_if_no_path(m, 1, 0); | |
727 | continue; | |
728 | } | |
729 | ||
730 | if (!strnicmp(param_name, MESG_STR("pg_init_retries")) && | |
731 | (argc >= 1)) { | |
732 | r = read_param(_params + 1, shift(as), | |
733 | &m->pg_init_retries, &ti->error); | |
734 | argc--; | |
735 | continue; | |
736 | } | |
737 | ||
1da177e4 | 738 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
739 | r = -EINVAL; |
740 | } while (argc && !r); | |
741 | ||
742 | return r; | |
1da177e4 LT |
743 | } |
744 | ||
745 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
746 | char **argv) | |
747 | { | |
748 | /* target parameters */ | |
749 | static struct param _params[] = { | |
72d94861 AK |
750 | {1, 1024, "invalid number of priority groups"}, |
751 | {1, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
752 | }; |
753 | ||
754 | int r; | |
755 | struct multipath *m; | |
756 | struct arg_set as; | |
757 | unsigned pg_count = 0; | |
758 | unsigned next_pg_num; | |
759 | ||
760 | as.argc = argc; | |
761 | as.argv = argv; | |
762 | ||
28f16c20 | 763 | m = alloc_multipath(ti); |
1da177e4 | 764 | if (!m) { |
72d94861 | 765 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
766 | return -EINVAL; |
767 | } | |
768 | ||
28f16c20 | 769 | r = parse_features(&as, m); |
1da177e4 LT |
770 | if (r) |
771 | goto bad; | |
772 | ||
28f16c20 | 773 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
774 | if (r) |
775 | goto bad; | |
776 | ||
777 | r = read_param(_params, shift(&as), &m->nr_priority_groups, &ti->error); | |
778 | if (r) | |
779 | goto bad; | |
780 | ||
781 | r = read_param(_params + 1, shift(&as), &next_pg_num, &ti->error); | |
782 | if (r) | |
783 | goto bad; | |
784 | ||
785 | /* parse the priority groups */ | |
786 | while (as.argc) { | |
787 | struct priority_group *pg; | |
788 | ||
28f16c20 | 789 | pg = parse_priority_group(&as, m); |
01460f35 BM |
790 | if (IS_ERR(pg)) { |
791 | r = PTR_ERR(pg); | |
1da177e4 LT |
792 | goto bad; |
793 | } | |
794 | ||
795 | m->nr_valid_paths += pg->nr_pgpaths; | |
796 | list_add_tail(&pg->list, &m->priority_groups); | |
797 | pg_count++; | |
798 | pg->pg_num = pg_count; | |
799 | if (!--next_pg_num) | |
800 | m->next_pg = pg; | |
801 | } | |
802 | ||
803 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 804 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
805 | r = -EINVAL; |
806 | goto bad; | |
807 | } | |
808 | ||
1da177e4 LT |
809 | return 0; |
810 | ||
811 | bad: | |
812 | free_multipath(m); | |
813 | return r; | |
814 | } | |
815 | ||
816 | static void multipath_dtr(struct dm_target *ti) | |
817 | { | |
818 | struct multipath *m = (struct multipath *) ti->private; | |
a044d016 | 819 | |
bab7cfc7 | 820 | flush_workqueue(kmpath_handlerd); |
a044d016 | 821 | flush_workqueue(kmultipathd); |
1da177e4 LT |
822 | free_multipath(m); |
823 | } | |
824 | ||
825 | /* | |
826 | * Map bios, recording original fields for later in case we have to resubmit | |
827 | */ | |
828 | static int multipath_map(struct dm_target *ti, struct bio *bio, | |
829 | union map_info *map_context) | |
830 | { | |
831 | int r; | |
028867ac | 832 | struct dm_mpath_io *mpio; |
1da177e4 LT |
833 | struct multipath *m = (struct multipath *) ti->private; |
834 | ||
835 | mpio = mempool_alloc(m->mpio_pool, GFP_NOIO); | |
836 | dm_bio_record(&mpio->details, bio); | |
837 | ||
838 | map_context->ptr = mpio; | |
839 | bio->bi_rw |= (1 << BIO_RW_FAILFAST); | |
840 | r = map_io(m, bio, mpio, 0); | |
45e15720 | 841 | if (r < 0 || r == DM_MAPIO_REQUEUE) |
1da177e4 LT |
842 | mempool_free(mpio, m->mpio_pool); |
843 | ||
844 | return r; | |
845 | } | |
846 | ||
847 | /* | |
848 | * Take a path out of use. | |
849 | */ | |
850 | static int fail_path(struct pgpath *pgpath) | |
851 | { | |
852 | unsigned long flags; | |
853 | struct multipath *m = pgpath->pg->m; | |
854 | ||
855 | spin_lock_irqsave(&m->lock, flags); | |
856 | ||
857 | if (!pgpath->path.is_active) | |
858 | goto out; | |
859 | ||
72d94861 | 860 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
861 | |
862 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
863 | pgpath->path.is_active = 0; | |
864 | pgpath->fail_count++; | |
865 | ||
866 | m->nr_valid_paths--; | |
867 | ||
868 | if (pgpath == m->current_pgpath) | |
869 | m->current_pgpath = NULL; | |
870 | ||
b15546f9 MA |
871 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
872 | pgpath->path.dev->name, m->nr_valid_paths); | |
873 | ||
c557308e | 874 | queue_work(kmultipathd, &m->trigger_event); |
1da177e4 LT |
875 | |
876 | out: | |
877 | spin_unlock_irqrestore(&m->lock, flags); | |
878 | ||
879 | return 0; | |
880 | } | |
881 | ||
882 | /* | |
883 | * Reinstate a previously-failed path | |
884 | */ | |
885 | static int reinstate_path(struct pgpath *pgpath) | |
886 | { | |
887 | int r = 0; | |
888 | unsigned long flags; | |
889 | struct multipath *m = pgpath->pg->m; | |
890 | ||
891 | spin_lock_irqsave(&m->lock, flags); | |
892 | ||
893 | if (pgpath->path.is_active) | |
894 | goto out; | |
895 | ||
def052d2 | 896 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
897 | DMWARN("Reinstate path not supported by path selector %s", |
898 | pgpath->pg->ps.type->name); | |
899 | r = -EINVAL; | |
900 | goto out; | |
901 | } | |
902 | ||
903 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
904 | if (r) | |
905 | goto out; | |
906 | ||
907 | pgpath->path.is_active = 1; | |
908 | ||
909 | m->current_pgpath = NULL; | |
c3cd4f6b | 910 | if (!m->nr_valid_paths++ && m->queue_size) |
c557308e | 911 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 912 | |
b15546f9 MA |
913 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
914 | pgpath->path.dev->name, m->nr_valid_paths); | |
915 | ||
c557308e | 916 | queue_work(kmultipathd, &m->trigger_event); |
1da177e4 LT |
917 | |
918 | out: | |
919 | spin_unlock_irqrestore(&m->lock, flags); | |
920 | ||
921 | return r; | |
922 | } | |
923 | ||
924 | /* | |
925 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
926 | */ | |
927 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
928 | action_fn action) | |
929 | { | |
930 | int r = 0; | |
931 | struct pgpath *pgpath; | |
932 | struct priority_group *pg; | |
933 | ||
934 | list_for_each_entry(pg, &m->priority_groups, list) { | |
935 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
936 | if (pgpath->path.dev == dev) | |
937 | r = action(pgpath); | |
938 | } | |
939 | } | |
940 | ||
941 | return r; | |
942 | } | |
943 | ||
944 | /* | |
945 | * Temporarily try to avoid having to use the specified PG | |
946 | */ | |
947 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
948 | int bypassed) | |
949 | { | |
950 | unsigned long flags; | |
951 | ||
952 | spin_lock_irqsave(&m->lock, flags); | |
953 | ||
954 | pg->bypassed = bypassed; | |
955 | m->current_pgpath = NULL; | |
956 | m->current_pg = NULL; | |
957 | ||
958 | spin_unlock_irqrestore(&m->lock, flags); | |
959 | ||
c557308e | 960 | queue_work(kmultipathd, &m->trigger_event); |
1da177e4 LT |
961 | } |
962 | ||
963 | /* | |
964 | * Switch to using the specified PG from the next I/O that gets mapped | |
965 | */ | |
966 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
967 | { | |
968 | struct priority_group *pg; | |
969 | unsigned pgnum; | |
970 | unsigned long flags; | |
971 | ||
972 | if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || | |
973 | (pgnum > m->nr_priority_groups)) { | |
974 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
975 | return -EINVAL; | |
976 | } | |
977 | ||
978 | spin_lock_irqsave(&m->lock, flags); | |
979 | list_for_each_entry(pg, &m->priority_groups, list) { | |
980 | pg->bypassed = 0; | |
981 | if (--pgnum) | |
982 | continue; | |
983 | ||
984 | m->current_pgpath = NULL; | |
985 | m->current_pg = NULL; | |
986 | m->next_pg = pg; | |
987 | } | |
988 | spin_unlock_irqrestore(&m->lock, flags); | |
989 | ||
c557308e | 990 | queue_work(kmultipathd, &m->trigger_event); |
1da177e4 LT |
991 | return 0; |
992 | } | |
993 | ||
994 | /* | |
995 | * Set/clear bypassed status of a PG. | |
996 | * PGs are numbered upwards from 1 in the order they were declared. | |
997 | */ | |
998 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
999 | { | |
1000 | struct priority_group *pg; | |
1001 | unsigned pgnum; | |
1002 | ||
1003 | if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || | |
1004 | (pgnum > m->nr_priority_groups)) { | |
1005 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1006 | return -EINVAL; | |
1007 | } | |
1008 | ||
1009 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1010 | if (!--pgnum) | |
1011 | break; | |
1012 | } | |
1013 | ||
1014 | bypass_pg(m, pg, bypassed); | |
1015 | return 0; | |
1016 | } | |
1017 | ||
c9e45581 DW |
1018 | /* |
1019 | * Should we retry pg_init immediately? | |
1020 | */ | |
1021 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1022 | { | |
1023 | unsigned long flags; | |
1024 | int limit_reached = 0; | |
1025 | ||
1026 | spin_lock_irqsave(&m->lock, flags); | |
1027 | ||
1028 | if (m->pg_init_count <= m->pg_init_retries) | |
1029 | m->pg_init_required = 1; | |
1030 | else | |
1031 | limit_reached = 1; | |
1032 | ||
1033 | spin_unlock_irqrestore(&m->lock, flags); | |
1034 | ||
1035 | return limit_reached; | |
1036 | } | |
1037 | ||
cfae5c9b CS |
1038 | static void pg_init_done(struct dm_path *path, int errors) |
1039 | { | |
1040 | struct pgpath *pgpath = path_to_pgpath(path); | |
1041 | struct priority_group *pg = pgpath->pg; | |
1042 | struct multipath *m = pg->m; | |
1043 | unsigned long flags; | |
1044 | ||
1045 | /* device or driver problems */ | |
1046 | switch (errors) { | |
1047 | case SCSI_DH_OK: | |
1048 | break; | |
1049 | case SCSI_DH_NOSYS: | |
1050 | if (!m->hw_handler_name) { | |
1051 | errors = 0; | |
1052 | break; | |
1053 | } | |
1054 | DMERR("Cannot failover device because scsi_dh_%s was not " | |
1055 | "loaded.", m->hw_handler_name); | |
1056 | /* | |
1057 | * Fail path for now, so we do not ping pong | |
1058 | */ | |
1059 | fail_path(pgpath); | |
1060 | break; | |
1061 | case SCSI_DH_DEV_TEMP_BUSY: | |
1062 | /* | |
1063 | * Probably doing something like FW upgrade on the | |
1064 | * controller so try the other pg. | |
1065 | */ | |
1066 | bypass_pg(m, pg, 1); | |
1067 | break; | |
1068 | /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */ | |
1069 | case SCSI_DH_RETRY: | |
1070 | case SCSI_DH_IMM_RETRY: | |
1071 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1072 | if (pg_init_limit_reached(m, pgpath)) | |
1073 | fail_path(pgpath); | |
1074 | errors = 0; | |
1075 | break; | |
1076 | default: | |
1077 | /* | |
1078 | * We probably do not want to fail the path for a device | |
1079 | * error, but this is what the old dm did. In future | |
1080 | * patches we can do more advanced handling. | |
1081 | */ | |
1082 | fail_path(pgpath); | |
1083 | } | |
1084 | ||
1085 | spin_lock_irqsave(&m->lock, flags); | |
1086 | if (errors) { | |
1087 | DMERR("Could not failover device. Error %d.", errors); | |
1088 | m->current_pgpath = NULL; | |
1089 | m->current_pg = NULL; | |
1090 | } else if (!m->pg_init_required) { | |
1091 | m->queue_io = 0; | |
1092 | pg->bypassed = 0; | |
1093 | } | |
1094 | ||
1095 | m->pg_init_in_progress = 0; | |
1096 | queue_work(kmultipathd, &m->process_queued_ios); | |
1097 | spin_unlock_irqrestore(&m->lock, flags); | |
1098 | } | |
1099 | ||
bab7cfc7 CS |
1100 | static void activate_path(struct work_struct *work) |
1101 | { | |
1102 | int ret; | |
1103 | struct multipath *m = | |
1104 | container_of(work, struct multipath, activate_path); | |
7253a334 CS |
1105 | struct dm_path *path; |
1106 | unsigned long flags; | |
bab7cfc7 | 1107 | |
7253a334 CS |
1108 | spin_lock_irqsave(&m->lock, flags); |
1109 | path = &m->pgpath_to_activate->path; | |
1110 | m->pgpath_to_activate = NULL; | |
1111 | spin_unlock_irqrestore(&m->lock, flags); | |
1112 | if (!path) | |
1113 | return; | |
bab7cfc7 CS |
1114 | ret = scsi_dh_activate(bdev_get_queue(path->dev->bdev)); |
1115 | pg_init_done(path, ret); | |
1116 | } | |
1117 | ||
1da177e4 LT |
1118 | /* |
1119 | * end_io handling | |
1120 | */ | |
1121 | static int do_end_io(struct multipath *m, struct bio *bio, | |
028867ac | 1122 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1123 | { |
640eb3b0 | 1124 | unsigned long flags; |
1da177e4 LT |
1125 | |
1126 | if (!error) | |
1127 | return 0; /* I/O complete */ | |
1128 | ||
4f58802f LMB |
1129 | if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio)) |
1130 | return error; | |
1131 | ||
f6a80ea8 AK |
1132 | if (error == -EOPNOTSUPP) |
1133 | return error; | |
1134 | ||
640eb3b0 | 1135 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 | 1136 | if (!m->nr_valid_paths) { |
45e15720 KU |
1137 | if (__must_push_back(m)) { |
1138 | spin_unlock_irqrestore(&m->lock, flags); | |
1139 | return DM_ENDIO_REQUEUE; | |
1140 | } else if (!m->queue_if_no_path) { | |
640eb3b0 | 1141 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1142 | return -EIO; |
1143 | } else { | |
640eb3b0 | 1144 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1145 | goto requeue; |
1146 | } | |
1147 | } | |
640eb3b0 | 1148 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1149 | |
cfae5c9b CS |
1150 | if (mpio->pgpath) |
1151 | fail_path(mpio->pgpath); | |
1da177e4 LT |
1152 | |
1153 | requeue: | |
1154 | dm_bio_restore(&mpio->details, bio); | |
1155 | ||
1156 | /* queue for the daemon to resubmit or fail */ | |
640eb3b0 | 1157 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
1158 | bio_list_add(&m->queued_ios, bio); |
1159 | m->queue_size++; | |
1160 | if (!m->queue_io) | |
c557308e | 1161 | queue_work(kmultipathd, &m->process_queued_ios); |
640eb3b0 | 1162 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1163 | |
d2a7ad29 | 1164 | return DM_ENDIO_INCOMPLETE; /* io not complete */ |
1da177e4 LT |
1165 | } |
1166 | ||
1167 | static int multipath_end_io(struct dm_target *ti, struct bio *bio, | |
1168 | int error, union map_info *map_context) | |
1169 | { | |
028867ac AK |
1170 | struct multipath *m = ti->private; |
1171 | struct dm_mpath_io *mpio = map_context->ptr; | |
1da177e4 LT |
1172 | struct pgpath *pgpath = mpio->pgpath; |
1173 | struct path_selector *ps; | |
1174 | int r; | |
1175 | ||
1176 | r = do_end_io(m, bio, error, mpio); | |
1177 | if (pgpath) { | |
1178 | ps = &pgpath->pg->ps; | |
1179 | if (ps->type->end_io) | |
1180 | ps->type->end_io(ps, &pgpath->path); | |
1181 | } | |
d2a7ad29 | 1182 | if (r != DM_ENDIO_INCOMPLETE) |
1da177e4 LT |
1183 | mempool_free(mpio, m->mpio_pool); |
1184 | ||
1185 | return r; | |
1186 | } | |
1187 | ||
1188 | /* | |
1189 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1190 | * the last path fails we must error any remaining I/O. |
1191 | * Note that if the freeze_bdev fails while suspending, the | |
1192 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1193 | */ |
1194 | static void multipath_presuspend(struct dm_target *ti) | |
1195 | { | |
1196 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1197 | |
485ef69e | 1198 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1199 | } |
1200 | ||
436d4108 AK |
1201 | /* |
1202 | * Restore the queue_if_no_path setting. | |
1203 | */ | |
1da177e4 LT |
1204 | static void multipath_resume(struct dm_target *ti) |
1205 | { | |
1206 | struct multipath *m = (struct multipath *) ti->private; | |
1207 | unsigned long flags; | |
1208 | ||
1209 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1210 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1211 | spin_unlock_irqrestore(&m->lock, flags); |
1212 | } | |
1213 | ||
1214 | /* | |
1215 | * Info output has the following format: | |
1216 | * num_multipath_feature_args [multipath_feature_args]* | |
1217 | * num_handler_status_args [handler_status_args]* | |
1218 | * num_groups init_group_number | |
1219 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1220 | * num_paths num_selector_args | |
1221 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1222 | * | |
1223 | * Table output has the following format (identical to the constructor string): | |
1224 | * num_feature_args [features_args]* | |
1225 | * num_handler_args hw_handler [hw_handler_args]* | |
1226 | * num_groups init_group_number | |
1227 | * [priority selector-name num_ps_args [ps_args]* | |
1228 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1229 | */ | |
1230 | static int multipath_status(struct dm_target *ti, status_type_t type, | |
1231 | char *result, unsigned int maxlen) | |
1232 | { | |
1233 | int sz = 0; | |
1234 | unsigned long flags; | |
1235 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1236 | struct priority_group *pg; |
1237 | struct pgpath *p; | |
1238 | unsigned pg_num; | |
1239 | char state; | |
1240 | ||
1241 | spin_lock_irqsave(&m->lock, flags); | |
1242 | ||
1243 | /* Features */ | |
1244 | if (type == STATUSTYPE_INFO) | |
c9e45581 DW |
1245 | DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count); |
1246 | else { | |
1247 | DMEMIT("%u ", m->queue_if_no_path + | |
1248 | (m->pg_init_retries > 0) * 2); | |
1249 | if (m->queue_if_no_path) | |
1250 | DMEMIT("queue_if_no_path "); | |
1251 | if (m->pg_init_retries) | |
1252 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
1253 | } | |
1da177e4 | 1254 | |
cfae5c9b | 1255 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1256 | DMEMIT("0 "); |
1257 | else | |
cfae5c9b | 1258 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1259 | |
1260 | DMEMIT("%u ", m->nr_priority_groups); | |
1261 | ||
1262 | if (m->next_pg) | |
1263 | pg_num = m->next_pg->pg_num; | |
1264 | else if (m->current_pg) | |
1265 | pg_num = m->current_pg->pg_num; | |
1266 | else | |
1267 | pg_num = 1; | |
1268 | ||
1269 | DMEMIT("%u ", pg_num); | |
1270 | ||
1271 | switch (type) { | |
1272 | case STATUSTYPE_INFO: | |
1273 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1274 | if (pg->bypassed) | |
1275 | state = 'D'; /* Disabled */ | |
1276 | else if (pg == m->current_pg) | |
1277 | state = 'A'; /* Currently Active */ | |
1278 | else | |
1279 | state = 'E'; /* Enabled */ | |
1280 | ||
1281 | DMEMIT("%c ", state); | |
1282 | ||
1283 | if (pg->ps.type->status) | |
1284 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1285 | result + sz, | |
1286 | maxlen - sz); | |
1287 | else | |
1288 | DMEMIT("0 "); | |
1289 | ||
1290 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1291 | pg->ps.type->info_args); | |
1292 | ||
1293 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1294 | DMEMIT("%s %s %u ", p->path.dev->name, | |
1295 | p->path.is_active ? "A" : "F", | |
1296 | p->fail_count); | |
1297 | if (pg->ps.type->status) | |
1298 | sz += pg->ps.type->status(&pg->ps, | |
1299 | &p->path, type, result + sz, | |
1300 | maxlen - sz); | |
1301 | } | |
1302 | } | |
1303 | break; | |
1304 | ||
1305 | case STATUSTYPE_TABLE: | |
1306 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1307 | DMEMIT("%s ", pg->ps.type->name); | |
1308 | ||
1309 | if (pg->ps.type->status) | |
1310 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1311 | result + sz, | |
1312 | maxlen - sz); | |
1313 | else | |
1314 | DMEMIT("0 "); | |
1315 | ||
1316 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1317 | pg->ps.type->table_args); | |
1318 | ||
1319 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1320 | DMEMIT("%s ", p->path.dev->name); | |
1321 | if (pg->ps.type->status) | |
1322 | sz += pg->ps.type->status(&pg->ps, | |
1323 | &p->path, type, result + sz, | |
1324 | maxlen - sz); | |
1325 | } | |
1326 | } | |
1327 | break; | |
1328 | } | |
1329 | ||
1330 | spin_unlock_irqrestore(&m->lock, flags); | |
1331 | ||
1332 | return 0; | |
1333 | } | |
1334 | ||
1335 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1336 | { | |
1337 | int r; | |
1338 | struct dm_dev *dev; | |
1339 | struct multipath *m = (struct multipath *) ti->private; | |
1340 | action_fn action; | |
1341 | ||
1342 | if (argc == 1) { | |
1343 | if (!strnicmp(argv[0], MESG_STR("queue_if_no_path"))) | |
485ef69e | 1344 | return queue_if_no_path(m, 1, 0); |
1da177e4 | 1345 | else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path"))) |
485ef69e | 1346 | return queue_if_no_path(m, 0, 0); |
1da177e4 LT |
1347 | } |
1348 | ||
1349 | if (argc != 2) | |
1350 | goto error; | |
1351 | ||
1352 | if (!strnicmp(argv[0], MESG_STR("disable_group"))) | |
1353 | return bypass_pg_num(m, argv[1], 1); | |
1354 | else if (!strnicmp(argv[0], MESG_STR("enable_group"))) | |
1355 | return bypass_pg_num(m, argv[1], 0); | |
1356 | else if (!strnicmp(argv[0], MESG_STR("switch_group"))) | |
1357 | return switch_pg_num(m, argv[1]); | |
1358 | else if (!strnicmp(argv[0], MESG_STR("reinstate_path"))) | |
1359 | action = reinstate_path; | |
1360 | else if (!strnicmp(argv[0], MESG_STR("fail_path"))) | |
1361 | action = fail_path; | |
1362 | else | |
1363 | goto error; | |
1364 | ||
1365 | r = dm_get_device(ti, argv[1], ti->begin, ti->len, | |
1366 | dm_table_get_mode(ti->table), &dev); | |
1367 | if (r) { | |
72d94861 | 1368 | DMWARN("message: error getting device %s", |
1da177e4 LT |
1369 | argv[1]); |
1370 | return -EINVAL; | |
1371 | } | |
1372 | ||
1373 | r = action_dev(m, dev, action); | |
1374 | ||
1375 | dm_put_device(ti, dev); | |
1376 | ||
1377 | return r; | |
1378 | ||
1379 | error: | |
1380 | DMWARN("Unrecognised multipath message received."); | |
1381 | return -EINVAL; | |
1382 | } | |
1383 | ||
9af4aa30 MB |
1384 | static int multipath_ioctl(struct dm_target *ti, struct inode *inode, |
1385 | struct file *filp, unsigned int cmd, | |
1386 | unsigned long arg) | |
1387 | { | |
1388 | struct multipath *m = (struct multipath *) ti->private; | |
1389 | struct block_device *bdev = NULL; | |
1390 | unsigned long flags; | |
e90dae1f MB |
1391 | struct file fake_file = {}; |
1392 | struct dentry fake_dentry = {}; | |
9af4aa30 MB |
1393 | int r = 0; |
1394 | ||
c649bb9c | 1395 | fake_file.f_path.dentry = &fake_dentry; |
e90dae1f | 1396 | |
9af4aa30 MB |
1397 | spin_lock_irqsave(&m->lock, flags); |
1398 | ||
1399 | if (!m->current_pgpath) | |
1400 | __choose_pgpath(m); | |
1401 | ||
e90dae1f | 1402 | if (m->current_pgpath) { |
9af4aa30 | 1403 | bdev = m->current_pgpath->path.dev->bdev; |
e90dae1f MB |
1404 | fake_dentry.d_inode = bdev->bd_inode; |
1405 | fake_file.f_mode = m->current_pgpath->path.dev->mode; | |
1406 | } | |
9af4aa30 MB |
1407 | |
1408 | if (m->queue_io) | |
1409 | r = -EAGAIN; | |
1410 | else if (!bdev) | |
1411 | r = -EIO; | |
1412 | ||
1413 | spin_unlock_irqrestore(&m->lock, flags); | |
1414 | ||
e90dae1f MB |
1415 | return r ? : blkdev_driver_ioctl(bdev->bd_inode, &fake_file, |
1416 | bdev->bd_disk, cmd, arg); | |
9af4aa30 MB |
1417 | } |
1418 | ||
1da177e4 LT |
1419 | /*----------------------------------------------------------------- |
1420 | * Module setup | |
1421 | *---------------------------------------------------------------*/ | |
1422 | static struct target_type multipath_target = { | |
1423 | .name = "multipath", | |
9af4aa30 | 1424 | .version = {1, 0, 5}, |
1da177e4 LT |
1425 | .module = THIS_MODULE, |
1426 | .ctr = multipath_ctr, | |
1427 | .dtr = multipath_dtr, | |
1428 | .map = multipath_map, | |
1429 | .end_io = multipath_end_io, | |
1430 | .presuspend = multipath_presuspend, | |
1431 | .resume = multipath_resume, | |
1432 | .status = multipath_status, | |
1433 | .message = multipath_message, | |
9af4aa30 | 1434 | .ioctl = multipath_ioctl, |
1da177e4 LT |
1435 | }; |
1436 | ||
1437 | static int __init dm_multipath_init(void) | |
1438 | { | |
1439 | int r; | |
1440 | ||
1441 | /* allocate a slab for the dm_ios */ | |
028867ac | 1442 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1443 | if (!_mpio_cache) |
1444 | return -ENOMEM; | |
1445 | ||
1446 | r = dm_register_target(&multipath_target); | |
1447 | if (r < 0) { | |
0cd33124 | 1448 | DMERR("register failed %d", r); |
1da177e4 LT |
1449 | kmem_cache_destroy(_mpio_cache); |
1450 | return -EINVAL; | |
1451 | } | |
1452 | ||
c557308e AK |
1453 | kmultipathd = create_workqueue("kmpathd"); |
1454 | if (!kmultipathd) { | |
0cd33124 | 1455 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1456 | dm_unregister_target(&multipath_target); |
1457 | kmem_cache_destroy(_mpio_cache); | |
1458 | return -ENOMEM; | |
1459 | } | |
1460 | ||
bab7cfc7 CS |
1461 | /* |
1462 | * A separate workqueue is used to handle the device handlers | |
1463 | * to avoid overloading existing workqueue. Overloading the | |
1464 | * old workqueue would also create a bottleneck in the | |
1465 | * path of the storage hardware device activation. | |
1466 | */ | |
1467 | kmpath_handlerd = create_singlethread_workqueue("kmpath_handlerd"); | |
1468 | if (!kmpath_handlerd) { | |
1469 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1470 | destroy_workqueue(kmultipathd); | |
1471 | dm_unregister_target(&multipath_target); | |
1472 | kmem_cache_destroy(_mpio_cache); | |
1473 | return -ENOMEM; | |
1474 | } | |
1475 | ||
72d94861 | 1476 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1477 | multipath_target.version[0], multipath_target.version[1], |
1478 | multipath_target.version[2]); | |
1479 | ||
1480 | return r; | |
1481 | } | |
1482 | ||
1483 | static void __exit dm_multipath_exit(void) | |
1484 | { | |
1485 | int r; | |
1486 | ||
bab7cfc7 | 1487 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1488 | destroy_workqueue(kmultipathd); |
1489 | ||
1da177e4 LT |
1490 | r = dm_unregister_target(&multipath_target); |
1491 | if (r < 0) | |
0cd33124 | 1492 | DMERR("target unregister failed %d", r); |
1da177e4 LT |
1493 | kmem_cache_destroy(_mpio_cache); |
1494 | } | |
1495 | ||
1da177e4 LT |
1496 | module_init(dm_multipath_init); |
1497 | module_exit(dm_multipath_exit); | |
1498 | ||
1499 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1500 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1501 | MODULE_LICENSE("GPL"); |