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 | ||
586e80e6 MP |
8 | #include <linux/device-mapper.h> |
9 | ||
1da177e4 | 10 | #include "dm-path-selector.h" |
b15546f9 | 11 | #include "dm-uevent.h" |
1da177e4 LT |
12 | |
13 | #include <linux/ctype.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/mempool.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/time.h> | |
20 | #include <linux/workqueue.h> | |
cfae5c9b | 21 | #include <scsi/scsi_dh.h> |
1da177e4 LT |
22 | #include <asm/atomic.h> |
23 | ||
72d94861 | 24 | #define DM_MSG_PREFIX "multipath" |
1da177e4 LT |
25 | #define MESG_STR(x) x, sizeof(x) |
26 | ||
27 | /* Path properties */ | |
28 | struct pgpath { | |
29 | struct list_head list; | |
30 | ||
31 | struct priority_group *pg; /* Owning PG */ | |
6680073d | 32 | unsigned is_active; /* Path status */ |
1da177e4 LT |
33 | unsigned fail_count; /* Cumulative failure count */ |
34 | ||
c922d5f7 | 35 | struct dm_path path; |
224cb3e9 | 36 | struct work_struct deactivate_path; |
e54f77dd | 37 | struct work_struct activate_path; |
1da177e4 LT |
38 | }; |
39 | ||
40 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
41 | ||
42 | /* | |
43 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
44 | * Each has a path selector which controls which path gets used. | |
45 | */ | |
46 | struct priority_group { | |
47 | struct list_head list; | |
48 | ||
49 | struct multipath *m; /* Owning multipath instance */ | |
50 | struct path_selector ps; | |
51 | ||
52 | unsigned pg_num; /* Reference number */ | |
53 | unsigned bypassed; /* Temporarily bypass this PG? */ | |
54 | ||
55 | unsigned nr_pgpaths; /* Number of paths in PG */ | |
56 | struct list_head pgpaths; | |
57 | }; | |
58 | ||
59 | /* Multipath context */ | |
60 | struct multipath { | |
61 | struct list_head list; | |
62 | struct dm_target *ti; | |
63 | ||
64 | spinlock_t lock; | |
65 | ||
cfae5c9b | 66 | const char *hw_handler_name; |
2bfd2e13 | 67 | char *hw_handler_params; |
1da177e4 LT |
68 | unsigned nr_priority_groups; |
69 | struct list_head priority_groups; | |
70 | unsigned pg_init_required; /* pg_init needs calling? */ | |
c3cd4f6b | 71 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
1da177e4 LT |
72 | |
73 | unsigned nr_valid_paths; /* Total number of usable paths */ | |
74 | struct pgpath *current_pgpath; | |
75 | struct priority_group *current_pg; | |
76 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
77 | unsigned repeat_count; /* I/Os left before calling PS again */ | |
78 | ||
79 | unsigned queue_io; /* Must we queue all I/O? */ | |
80 | unsigned queue_if_no_path; /* Queue I/O if last path fails? */ | |
436d4108 | 81 | unsigned saved_queue_if_no_path;/* Saved state during suspension */ |
c9e45581 DW |
82 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
83 | unsigned pg_init_count; /* Number of times pg_init called */ | |
1da177e4 LT |
84 | |
85 | struct work_struct process_queued_ios; | |
f40c67f0 | 86 | struct list_head queued_ios; |
1da177e4 LT |
87 | unsigned queue_size; |
88 | ||
89 | struct work_struct trigger_event; | |
90 | ||
91 | /* | |
028867ac | 92 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
93 | * can resubmit bios on error. |
94 | */ | |
95 | mempool_t *mpio_pool; | |
96 | }; | |
97 | ||
98 | /* | |
99 | * Context information attached to each bio we process. | |
100 | */ | |
028867ac | 101 | struct dm_mpath_io { |
1da177e4 | 102 | struct pgpath *pgpath; |
02ab823f | 103 | size_t nr_bytes; |
1da177e4 LT |
104 | }; |
105 | ||
106 | typedef int (*action_fn) (struct pgpath *pgpath); | |
107 | ||
108 | #define MIN_IOS 256 /* Mempool size */ | |
109 | ||
e18b890b | 110 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 111 | |
bab7cfc7 | 112 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 DH |
113 | static void process_queued_ios(struct work_struct *work); |
114 | static void trigger_event(struct work_struct *work); | |
bab7cfc7 | 115 | static void activate_path(struct work_struct *work); |
224cb3e9 | 116 | static void deactivate_path(struct work_struct *work); |
1da177e4 LT |
117 | |
118 | ||
119 | /*----------------------------------------------- | |
120 | * Allocation routines | |
121 | *-----------------------------------------------*/ | |
122 | ||
123 | static struct pgpath *alloc_pgpath(void) | |
124 | { | |
e69fae56 | 125 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 126 | |
224cb3e9 | 127 | if (pgpath) { |
6680073d | 128 | pgpath->is_active = 1; |
224cb3e9 | 129 | INIT_WORK(&pgpath->deactivate_path, deactivate_path); |
e54f77dd | 130 | INIT_WORK(&pgpath->activate_path, activate_path); |
224cb3e9 | 131 | } |
1da177e4 LT |
132 | |
133 | return pgpath; | |
134 | } | |
135 | ||
028867ac | 136 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
137 | { |
138 | kfree(pgpath); | |
139 | } | |
140 | ||
224cb3e9 MA |
141 | static void deactivate_path(struct work_struct *work) |
142 | { | |
143 | struct pgpath *pgpath = | |
144 | container_of(work, struct pgpath, deactivate_path); | |
145 | ||
146 | blk_abort_queue(pgpath->path.dev->bdev->bd_disk->queue); | |
147 | } | |
148 | ||
1da177e4 LT |
149 | static struct priority_group *alloc_priority_group(void) |
150 | { | |
151 | struct priority_group *pg; | |
152 | ||
e69fae56 | 153 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 154 | |
e69fae56 MM |
155 | if (pg) |
156 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
157 | |
158 | return pg; | |
159 | } | |
160 | ||
161 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
162 | { | |
163 | struct pgpath *pgpath, *tmp; | |
ae11b1b3 | 164 | struct multipath *m = ti->private; |
1da177e4 LT |
165 | |
166 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
167 | list_del(&pgpath->list); | |
ae11b1b3 HR |
168 | if (m->hw_handler_name) |
169 | scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev)); | |
1da177e4 LT |
170 | dm_put_device(ti, pgpath->path.dev); |
171 | free_pgpath(pgpath); | |
172 | } | |
173 | } | |
174 | ||
175 | static void free_priority_group(struct priority_group *pg, | |
176 | struct dm_target *ti) | |
177 | { | |
178 | struct path_selector *ps = &pg->ps; | |
179 | ||
180 | if (ps->type) { | |
181 | ps->type->destroy(ps); | |
182 | dm_put_path_selector(ps->type); | |
183 | } | |
184 | ||
185 | free_pgpaths(&pg->pgpaths, ti); | |
186 | kfree(pg); | |
187 | } | |
188 | ||
28f16c20 | 189 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
190 | { |
191 | struct multipath *m; | |
192 | ||
e69fae56 | 193 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 194 | if (m) { |
1da177e4 | 195 | INIT_LIST_HEAD(&m->priority_groups); |
f40c67f0 | 196 | INIT_LIST_HEAD(&m->queued_ios); |
1da177e4 LT |
197 | spin_lock_init(&m->lock); |
198 | m->queue_io = 1; | |
c4028958 DH |
199 | INIT_WORK(&m->process_queued_ios, process_queued_ios); |
200 | INIT_WORK(&m->trigger_event, trigger_event); | |
93d2341c | 201 | m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache); |
1da177e4 LT |
202 | if (!m->mpio_pool) { |
203 | kfree(m); | |
204 | return NULL; | |
205 | } | |
28f16c20 MM |
206 | m->ti = ti; |
207 | ti->private = m; | |
1da177e4 LT |
208 | } |
209 | ||
210 | return m; | |
211 | } | |
212 | ||
213 | static void free_multipath(struct multipath *m) | |
214 | { | |
215 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
216 | |
217 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
218 | list_del(&pg->list); | |
219 | free_priority_group(pg, m->ti); | |
220 | } | |
221 | ||
cfae5c9b | 222 | kfree(m->hw_handler_name); |
2bfd2e13 | 223 | kfree(m->hw_handler_params); |
1da177e4 LT |
224 | mempool_destroy(m->mpio_pool); |
225 | kfree(m); | |
226 | } | |
227 | ||
228 | ||
229 | /*----------------------------------------------- | |
230 | * Path selection | |
231 | *-----------------------------------------------*/ | |
232 | ||
233 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) | |
234 | { | |
1da177e4 LT |
235 | m->current_pg = pgpath->pg; |
236 | ||
237 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 238 | if (m->hw_handler_name) { |
1da177e4 LT |
239 | m->pg_init_required = 1; |
240 | m->queue_io = 1; | |
241 | } else { | |
242 | m->pg_init_required = 0; | |
243 | m->queue_io = 0; | |
244 | } | |
c9e45581 DW |
245 | |
246 | m->pg_init_count = 0; | |
1da177e4 LT |
247 | } |
248 | ||
02ab823f KU |
249 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
250 | size_t nr_bytes) | |
1da177e4 | 251 | { |
c922d5f7 | 252 | struct dm_path *path; |
1da177e4 | 253 | |
02ab823f | 254 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes); |
1da177e4 LT |
255 | if (!path) |
256 | return -ENXIO; | |
257 | ||
258 | m->current_pgpath = path_to_pgpath(path); | |
259 | ||
260 | if (m->current_pg != pg) | |
261 | __switch_pg(m, m->current_pgpath); | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
02ab823f | 266 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
267 | { |
268 | struct priority_group *pg; | |
269 | unsigned bypassed = 1; | |
270 | ||
271 | if (!m->nr_valid_paths) | |
272 | goto failed; | |
273 | ||
274 | /* Were we instructed to switch PG? */ | |
275 | if (m->next_pg) { | |
276 | pg = m->next_pg; | |
277 | m->next_pg = NULL; | |
02ab823f | 278 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
279 | return; |
280 | } | |
281 | ||
282 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 283 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
284 | return; |
285 | ||
286 | /* | |
287 | * Loop through priority groups until we find a valid path. | |
288 | * First time we skip PGs marked 'bypassed'. | |
289 | * Second time we only try the ones we skipped. | |
290 | */ | |
291 | do { | |
292 | list_for_each_entry(pg, &m->priority_groups, list) { | |
293 | if (pg->bypassed == bypassed) | |
294 | continue; | |
02ab823f | 295 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
296 | return; |
297 | } | |
298 | } while (bypassed--); | |
299 | ||
300 | failed: | |
301 | m->current_pgpath = NULL; | |
302 | m->current_pg = NULL; | |
303 | } | |
304 | ||
45e15720 KU |
305 | /* |
306 | * Check whether bios must be queued in the device-mapper core rather | |
307 | * than here in the target. | |
308 | * | |
309 | * m->lock must be held on entry. | |
310 | * | |
311 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
312 | * same value then we are not between multipath_presuspend() | |
313 | * and multipath_resume() calls and we have no need to check | |
314 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
315 | */ | |
316 | static int __must_push_back(struct multipath *m) | |
317 | { | |
318 | return (m->queue_if_no_path != m->saved_queue_if_no_path && | |
319 | dm_noflush_suspending(m->ti)); | |
320 | } | |
321 | ||
f40c67f0 | 322 | static int map_io(struct multipath *m, struct request *clone, |
028867ac | 323 | struct dm_mpath_io *mpio, unsigned was_queued) |
1da177e4 | 324 | { |
d2a7ad29 | 325 | int r = DM_MAPIO_REMAPPED; |
f40c67f0 | 326 | size_t nr_bytes = blk_rq_bytes(clone); |
1da177e4 LT |
327 | unsigned long flags; |
328 | struct pgpath *pgpath; | |
f40c67f0 | 329 | struct block_device *bdev; |
1da177e4 LT |
330 | |
331 | spin_lock_irqsave(&m->lock, flags); | |
332 | ||
333 | /* Do we need to select a new pgpath? */ | |
334 | if (!m->current_pgpath || | |
335 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
02ab823f | 336 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
337 | |
338 | pgpath = m->current_pgpath; | |
339 | ||
340 | if (was_queued) | |
341 | m->queue_size--; | |
342 | ||
343 | if ((pgpath && m->queue_io) || | |
436d4108 | 344 | (!pgpath && m->queue_if_no_path)) { |
1da177e4 | 345 | /* Queue for the daemon to resubmit */ |
f40c67f0 | 346 | list_add_tail(&clone->queuelist, &m->queued_ios); |
1da177e4 | 347 | m->queue_size++; |
c3cd4f6b AK |
348 | if ((m->pg_init_required && !m->pg_init_in_progress) || |
349 | !m->queue_io) | |
c557308e | 350 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 351 | pgpath = NULL; |
d2a7ad29 | 352 | r = DM_MAPIO_SUBMITTED; |
f40c67f0 KU |
353 | } else if (pgpath) { |
354 | bdev = pgpath->path.dev->bdev; | |
355 | clone->q = bdev_get_queue(bdev); | |
356 | clone->rq_disk = bdev->bd_disk; | |
357 | } else if (__must_push_back(m)) | |
45e15720 KU |
358 | r = DM_MAPIO_REQUEUE; |
359 | else | |
360 | r = -EIO; /* Failed */ | |
1da177e4 LT |
361 | |
362 | mpio->pgpath = pgpath; | |
02ab823f KU |
363 | mpio->nr_bytes = nr_bytes; |
364 | ||
365 | if (r == DM_MAPIO_REMAPPED && pgpath->pg->ps.type->start_io) | |
366 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, &pgpath->path, | |
367 | nr_bytes); | |
1da177e4 LT |
368 | |
369 | spin_unlock_irqrestore(&m->lock, flags); | |
370 | ||
371 | return r; | |
372 | } | |
373 | ||
374 | /* | |
375 | * If we run out of usable paths, should we queue I/O or error it? | |
376 | */ | |
485ef69e AK |
377 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
378 | unsigned save_old_value) | |
1da177e4 LT |
379 | { |
380 | unsigned long flags; | |
381 | ||
382 | spin_lock_irqsave(&m->lock, flags); | |
383 | ||
485ef69e AK |
384 | if (save_old_value) |
385 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
386 | else | |
387 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 388 | m->queue_if_no_path = queue_if_no_path; |
c3cd4f6b | 389 | if (!m->queue_if_no_path && m->queue_size) |
c557308e | 390 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 LT |
391 | |
392 | spin_unlock_irqrestore(&m->lock, flags); | |
393 | ||
394 | return 0; | |
395 | } | |
396 | ||
397 | /*----------------------------------------------------------------- | |
398 | * The multipath daemon is responsible for resubmitting queued ios. | |
399 | *---------------------------------------------------------------*/ | |
400 | ||
401 | static void dispatch_queued_ios(struct multipath *m) | |
402 | { | |
403 | int r; | |
404 | unsigned long flags; | |
028867ac | 405 | struct dm_mpath_io *mpio; |
1da177e4 | 406 | union map_info *info; |
f40c67f0 KU |
407 | struct request *clone, *n; |
408 | LIST_HEAD(cl); | |
1da177e4 LT |
409 | |
410 | spin_lock_irqsave(&m->lock, flags); | |
f40c67f0 | 411 | list_splice_init(&m->queued_ios, &cl); |
1da177e4 LT |
412 | spin_unlock_irqrestore(&m->lock, flags); |
413 | ||
f40c67f0 KU |
414 | list_for_each_entry_safe(clone, n, &cl, queuelist) { |
415 | list_del_init(&clone->queuelist); | |
1da177e4 | 416 | |
f40c67f0 | 417 | info = dm_get_rq_mapinfo(clone); |
1da177e4 LT |
418 | mpio = info->ptr; |
419 | ||
f40c67f0 KU |
420 | r = map_io(m, clone, mpio, 1); |
421 | if (r < 0) { | |
422 | mempool_free(mpio, m->mpio_pool); | |
423 | dm_kill_unmapped_request(clone, r); | |
424 | } else if (r == DM_MAPIO_REMAPPED) | |
425 | dm_dispatch_request(clone); | |
426 | else if (r == DM_MAPIO_REQUEUE) { | |
427 | mempool_free(mpio, m->mpio_pool); | |
428 | dm_requeue_unmapped_request(clone); | |
429 | } | |
1da177e4 LT |
430 | } |
431 | } | |
432 | ||
c4028958 | 433 | static void process_queued_ios(struct work_struct *work) |
1da177e4 | 434 | { |
c4028958 DH |
435 | struct multipath *m = |
436 | container_of(work, struct multipath, process_queued_ios); | |
e54f77dd CS |
437 | struct pgpath *pgpath = NULL, *tmp; |
438 | unsigned must_queue = 1; | |
1da177e4 LT |
439 | unsigned long flags; |
440 | ||
441 | spin_lock_irqsave(&m->lock, flags); | |
442 | ||
c3cd4f6b AK |
443 | if (!m->queue_size) |
444 | goto out; | |
445 | ||
1da177e4 | 446 | if (!m->current_pgpath) |
02ab823f | 447 | __choose_pgpath(m, 0); |
1da177e4 LT |
448 | |
449 | pgpath = m->current_pgpath; | |
450 | ||
c3cd4f6b AK |
451 | if ((pgpath && !m->queue_io) || |
452 | (!pgpath && !m->queue_if_no_path)) | |
453 | must_queue = 0; | |
1da177e4 | 454 | |
b81aa1c7 | 455 | if (m->pg_init_required && !m->pg_init_in_progress && pgpath) { |
c9e45581 | 456 | m->pg_init_count++; |
1da177e4 | 457 | m->pg_init_required = 0; |
e54f77dd CS |
458 | list_for_each_entry(tmp, &pgpath->pg->pgpaths, list) { |
459 | if (queue_work(kmpath_handlerd, &tmp->activate_path)) | |
460 | m->pg_init_in_progress++; | |
461 | } | |
c3cd4f6b | 462 | } |
c3cd4f6b | 463 | out: |
1da177e4 | 464 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
465 | if (!must_queue) |
466 | dispatch_queued_ios(m); | |
467 | } | |
468 | ||
469 | /* | |
470 | * An event is triggered whenever a path is taken out of use. | |
471 | * Includes path failure and PG bypass. | |
472 | */ | |
c4028958 | 473 | static void trigger_event(struct work_struct *work) |
1da177e4 | 474 | { |
c4028958 DH |
475 | struct multipath *m = |
476 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
477 | |
478 | dm_table_event(m->ti->table); | |
479 | } | |
480 | ||
481 | /*----------------------------------------------------------------- | |
482 | * Constructor/argument parsing: | |
483 | * <#multipath feature args> [<arg>]* | |
484 | * <#hw_handler args> [hw_handler [<arg>]*] | |
485 | * <#priority groups> | |
486 | * <initial priority group> | |
487 | * [<selector> <#selector args> [<arg>]* | |
488 | * <#paths> <#per-path selector args> | |
489 | * [<path> [<arg>]* ]+ ]+ | |
490 | *---------------------------------------------------------------*/ | |
491 | struct param { | |
492 | unsigned min; | |
493 | unsigned max; | |
494 | char *error; | |
495 | }; | |
496 | ||
1da177e4 LT |
497 | static int read_param(struct param *param, char *str, unsigned *v, char **error) |
498 | { | |
499 | if (!str || | |
500 | (sscanf(str, "%u", v) != 1) || | |
501 | (*v < param->min) || | |
502 | (*v > param->max)) { | |
503 | *error = param->error; | |
504 | return -EINVAL; | |
505 | } | |
506 | ||
507 | return 0; | |
508 | } | |
509 | ||
510 | struct arg_set { | |
511 | unsigned argc; | |
512 | char **argv; | |
513 | }; | |
514 | ||
515 | static char *shift(struct arg_set *as) | |
516 | { | |
517 | char *r; | |
518 | ||
519 | if (as->argc) { | |
520 | as->argc--; | |
521 | r = *as->argv; | |
522 | as->argv++; | |
523 | return r; | |
524 | } | |
525 | ||
526 | return NULL; | |
527 | } | |
528 | ||
529 | static void consume(struct arg_set *as, unsigned n) | |
530 | { | |
531 | BUG_ON (as->argc < n); | |
532 | as->argc -= n; | |
533 | as->argv += n; | |
534 | } | |
535 | ||
536 | static int parse_path_selector(struct arg_set *as, struct priority_group *pg, | |
537 | struct dm_target *ti) | |
538 | { | |
539 | int r; | |
540 | struct path_selector_type *pst; | |
541 | unsigned ps_argc; | |
542 | ||
543 | static struct param _params[] = { | |
72d94861 | 544 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
545 | }; |
546 | ||
547 | pst = dm_get_path_selector(shift(as)); | |
548 | if (!pst) { | |
72d94861 | 549 | ti->error = "unknown path selector type"; |
1da177e4 LT |
550 | return -EINVAL; |
551 | } | |
552 | ||
553 | r = read_param(_params, shift(as), &ps_argc, &ti->error); | |
371b2e34 MP |
554 | if (r) { |
555 | dm_put_path_selector(pst); | |
1da177e4 | 556 | return -EINVAL; |
371b2e34 | 557 | } |
1da177e4 | 558 | |
0e0497c0 MP |
559 | if (ps_argc > as->argc) { |
560 | dm_put_path_selector(pst); | |
561 | ti->error = "not enough arguments for path selector"; | |
562 | return -EINVAL; | |
563 | } | |
564 | ||
1da177e4 LT |
565 | r = pst->create(&pg->ps, ps_argc, as->argv); |
566 | if (r) { | |
567 | dm_put_path_selector(pst); | |
72d94861 | 568 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
569 | return r; |
570 | } | |
571 | ||
572 | pg->ps.type = pst; | |
573 | consume(as, ps_argc); | |
574 | ||
575 | return 0; | |
576 | } | |
577 | ||
578 | static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps, | |
579 | struct dm_target *ti) | |
580 | { | |
581 | int r; | |
582 | struct pgpath *p; | |
ae11b1b3 | 583 | struct multipath *m = ti->private; |
1da177e4 LT |
584 | |
585 | /* we need at least a path arg */ | |
586 | if (as->argc < 1) { | |
72d94861 | 587 | ti->error = "no device given"; |
01460f35 | 588 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
589 | } |
590 | ||
591 | p = alloc_pgpath(); | |
592 | if (!p) | |
01460f35 | 593 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
594 | |
595 | r = dm_get_device(ti, shift(as), ti->begin, ti->len, | |
596 | dm_table_get_mode(ti->table), &p->path.dev); | |
597 | if (r) { | |
72d94861 | 598 | ti->error = "error getting device"; |
1da177e4 LT |
599 | goto bad; |
600 | } | |
601 | ||
ae11b1b3 | 602 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
603 | struct request_queue *q = bdev_get_queue(p->path.dev->bdev); |
604 | ||
605 | r = scsi_dh_attach(q, m->hw_handler_name); | |
606 | if (r == -EBUSY) { | |
607 | /* | |
608 | * Already attached to different hw_handler, | |
609 | * try to reattach with correct one. | |
610 | */ | |
611 | scsi_dh_detach(q); | |
612 | r = scsi_dh_attach(q, m->hw_handler_name); | |
613 | } | |
614 | ||
ae11b1b3 | 615 | if (r < 0) { |
a0cf7ea9 | 616 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
617 | dm_put_device(ti, p->path.dev); |
618 | goto bad; | |
619 | } | |
2bfd2e13 CS |
620 | |
621 | if (m->hw_handler_params) { | |
622 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
623 | if (r < 0) { | |
624 | ti->error = "unable to set hardware " | |
625 | "handler parameters"; | |
626 | scsi_dh_detach(q); | |
627 | dm_put_device(ti, p->path.dev); | |
628 | goto bad; | |
629 | } | |
630 | } | |
ae11b1b3 HR |
631 | } |
632 | ||
1da177e4 LT |
633 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
634 | if (r) { | |
635 | dm_put_device(ti, p->path.dev); | |
636 | goto bad; | |
637 | } | |
638 | ||
639 | return p; | |
640 | ||
641 | bad: | |
642 | free_pgpath(p); | |
01460f35 | 643 | return ERR_PTR(r); |
1da177e4 LT |
644 | } |
645 | ||
646 | static struct priority_group *parse_priority_group(struct arg_set *as, | |
28f16c20 | 647 | struct multipath *m) |
1da177e4 LT |
648 | { |
649 | static struct param _params[] = { | |
72d94861 AK |
650 | {1, 1024, "invalid number of paths"}, |
651 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
652 | }; |
653 | ||
654 | int r; | |
655 | unsigned i, nr_selector_args, nr_params; | |
656 | struct priority_group *pg; | |
28f16c20 | 657 | struct dm_target *ti = m->ti; |
1da177e4 LT |
658 | |
659 | if (as->argc < 2) { | |
660 | as->argc = 0; | |
01460f35 BM |
661 | ti->error = "not enough priority group arguments"; |
662 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
663 | } |
664 | ||
665 | pg = alloc_priority_group(); | |
666 | if (!pg) { | |
72d94861 | 667 | ti->error = "couldn't allocate priority group"; |
01460f35 | 668 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
669 | } |
670 | pg->m = m; | |
671 | ||
672 | r = parse_path_selector(as, pg, ti); | |
673 | if (r) | |
674 | goto bad; | |
675 | ||
676 | /* | |
677 | * read the paths | |
678 | */ | |
679 | r = read_param(_params, shift(as), &pg->nr_pgpaths, &ti->error); | |
680 | if (r) | |
681 | goto bad; | |
682 | ||
683 | r = read_param(_params + 1, shift(as), &nr_selector_args, &ti->error); | |
684 | if (r) | |
685 | goto bad; | |
686 | ||
687 | nr_params = 1 + nr_selector_args; | |
688 | for (i = 0; i < pg->nr_pgpaths; i++) { | |
689 | struct pgpath *pgpath; | |
690 | struct arg_set path_args; | |
691 | ||
148acff6 MP |
692 | if (as->argc < nr_params) { |
693 | ti->error = "not enough path parameters"; | |
1da177e4 | 694 | goto bad; |
148acff6 | 695 | } |
1da177e4 LT |
696 | |
697 | path_args.argc = nr_params; | |
698 | path_args.argv = as->argv; | |
699 | ||
700 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
701 | if (IS_ERR(pgpath)) { |
702 | r = PTR_ERR(pgpath); | |
1da177e4 | 703 | goto bad; |
01460f35 | 704 | } |
1da177e4 LT |
705 | |
706 | pgpath->pg = pg; | |
707 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
708 | consume(as, nr_params); | |
709 | } | |
710 | ||
711 | return pg; | |
712 | ||
713 | bad: | |
714 | free_priority_group(pg, ti); | |
01460f35 | 715 | return ERR_PTR(r); |
1da177e4 LT |
716 | } |
717 | ||
28f16c20 | 718 | static int parse_hw_handler(struct arg_set *as, struct multipath *m) |
1da177e4 | 719 | { |
1da177e4 | 720 | unsigned hw_argc; |
2bfd2e13 | 721 | int ret; |
28f16c20 | 722 | struct dm_target *ti = m->ti; |
1da177e4 LT |
723 | |
724 | static struct param _params[] = { | |
72d94861 | 725 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
726 | }; |
727 | ||
cfae5c9b | 728 | if (read_param(_params, shift(as), &hw_argc, &ti->error)) |
1da177e4 LT |
729 | return -EINVAL; |
730 | ||
731 | if (!hw_argc) | |
732 | return 0; | |
733 | ||
e094f4f1 MP |
734 | if (hw_argc > as->argc) { |
735 | ti->error = "not enough arguments for hardware handler"; | |
736 | return -EINVAL; | |
737 | } | |
738 | ||
cfae5c9b CS |
739 | m->hw_handler_name = kstrdup(shift(as), GFP_KERNEL); |
740 | request_module("scsi_dh_%s", m->hw_handler_name); | |
741 | if (scsi_dh_handler_exist(m->hw_handler_name) == 0) { | |
72d94861 | 742 | ti->error = "unknown hardware handler type"; |
2bfd2e13 CS |
743 | ret = -EINVAL; |
744 | goto fail; | |
1da177e4 | 745 | } |
14e98c5c | 746 | |
2bfd2e13 CS |
747 | if (hw_argc > 1) { |
748 | char *p; | |
749 | int i, j, len = 4; | |
750 | ||
751 | for (i = 0; i <= hw_argc - 2; i++) | |
752 | len += strlen(as->argv[i]) + 1; | |
753 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
754 | if (!p) { | |
755 | ti->error = "memory allocation failed"; | |
756 | ret = -ENOMEM; | |
757 | goto fail; | |
758 | } | |
759 | j = sprintf(p, "%d", hw_argc - 1); | |
760 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
761 | j = sprintf(p, "%s", as->argv[i]); | |
762 | } | |
1da177e4 LT |
763 | consume(as, hw_argc - 1); |
764 | ||
765 | return 0; | |
2bfd2e13 CS |
766 | fail: |
767 | kfree(m->hw_handler_name); | |
768 | m->hw_handler_name = NULL; | |
769 | return ret; | |
1da177e4 LT |
770 | } |
771 | ||
28f16c20 | 772 | static int parse_features(struct arg_set *as, struct multipath *m) |
1da177e4 LT |
773 | { |
774 | int r; | |
775 | unsigned argc; | |
28f16c20 | 776 | struct dm_target *ti = m->ti; |
c9e45581 | 777 | const char *param_name; |
1da177e4 LT |
778 | |
779 | static struct param _params[] = { | |
c9e45581 DW |
780 | {0, 3, "invalid number of feature args"}, |
781 | {1, 50, "pg_init_retries must be between 1 and 50"}, | |
1da177e4 LT |
782 | }; |
783 | ||
784 | r = read_param(_params, shift(as), &argc, &ti->error); | |
785 | if (r) | |
786 | return -EINVAL; | |
787 | ||
788 | if (!argc) | |
789 | return 0; | |
790 | ||
c9e45581 DW |
791 | do { |
792 | param_name = shift(as); | |
793 | argc--; | |
794 | ||
795 | if (!strnicmp(param_name, MESG_STR("queue_if_no_path"))) { | |
796 | r = queue_if_no_path(m, 1, 0); | |
797 | continue; | |
798 | } | |
799 | ||
800 | if (!strnicmp(param_name, MESG_STR("pg_init_retries")) && | |
801 | (argc >= 1)) { | |
802 | r = read_param(_params + 1, shift(as), | |
803 | &m->pg_init_retries, &ti->error); | |
804 | argc--; | |
805 | continue; | |
806 | } | |
807 | ||
1da177e4 | 808 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
809 | r = -EINVAL; |
810 | } while (argc && !r); | |
811 | ||
812 | return r; | |
1da177e4 LT |
813 | } |
814 | ||
815 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
816 | char **argv) | |
817 | { | |
818 | /* target parameters */ | |
819 | static struct param _params[] = { | |
72d94861 AK |
820 | {1, 1024, "invalid number of priority groups"}, |
821 | {1, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
822 | }; |
823 | ||
824 | int r; | |
825 | struct multipath *m; | |
826 | struct arg_set as; | |
827 | unsigned pg_count = 0; | |
828 | unsigned next_pg_num; | |
829 | ||
830 | as.argc = argc; | |
831 | as.argv = argv; | |
832 | ||
28f16c20 | 833 | m = alloc_multipath(ti); |
1da177e4 | 834 | if (!m) { |
72d94861 | 835 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
836 | return -EINVAL; |
837 | } | |
838 | ||
28f16c20 | 839 | r = parse_features(&as, m); |
1da177e4 LT |
840 | if (r) |
841 | goto bad; | |
842 | ||
28f16c20 | 843 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
844 | if (r) |
845 | goto bad; | |
846 | ||
847 | r = read_param(_params, shift(&as), &m->nr_priority_groups, &ti->error); | |
848 | if (r) | |
849 | goto bad; | |
850 | ||
851 | r = read_param(_params + 1, shift(&as), &next_pg_num, &ti->error); | |
852 | if (r) | |
853 | goto bad; | |
854 | ||
855 | /* parse the priority groups */ | |
856 | while (as.argc) { | |
857 | struct priority_group *pg; | |
858 | ||
28f16c20 | 859 | pg = parse_priority_group(&as, m); |
01460f35 BM |
860 | if (IS_ERR(pg)) { |
861 | r = PTR_ERR(pg); | |
1da177e4 LT |
862 | goto bad; |
863 | } | |
864 | ||
865 | m->nr_valid_paths += pg->nr_pgpaths; | |
866 | list_add_tail(&pg->list, &m->priority_groups); | |
867 | pg_count++; | |
868 | pg->pg_num = pg_count; | |
869 | if (!--next_pg_num) | |
870 | m->next_pg = pg; | |
871 | } | |
872 | ||
873 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 874 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
875 | r = -EINVAL; |
876 | goto bad; | |
877 | } | |
878 | ||
8627921f MP |
879 | ti->num_flush_requests = 1; |
880 | ||
1da177e4 LT |
881 | return 0; |
882 | ||
883 | bad: | |
884 | free_multipath(m); | |
885 | return r; | |
886 | } | |
887 | ||
888 | static void multipath_dtr(struct dm_target *ti) | |
889 | { | |
890 | struct multipath *m = (struct multipath *) ti->private; | |
a044d016 | 891 | |
bab7cfc7 | 892 | flush_workqueue(kmpath_handlerd); |
a044d016 | 893 | flush_workqueue(kmultipathd); |
53b351f9 | 894 | flush_scheduled_work(); |
1da177e4 LT |
895 | free_multipath(m); |
896 | } | |
897 | ||
898 | /* | |
f40c67f0 | 899 | * Map cloned requests |
1da177e4 | 900 | */ |
f40c67f0 | 901 | static int multipath_map(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
902 | union map_info *map_context) |
903 | { | |
904 | int r; | |
028867ac | 905 | struct dm_mpath_io *mpio; |
1da177e4 LT |
906 | struct multipath *m = (struct multipath *) ti->private; |
907 | ||
f40c67f0 KU |
908 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); |
909 | if (!mpio) | |
910 | /* ENOMEM, requeue */ | |
911 | return DM_MAPIO_REQUEUE; | |
912 | memset(mpio, 0, sizeof(*mpio)); | |
1da177e4 LT |
913 | |
914 | map_context->ptr = mpio; | |
f40c67f0 KU |
915 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; |
916 | r = map_io(m, clone, mpio, 0); | |
45e15720 | 917 | if (r < 0 || r == DM_MAPIO_REQUEUE) |
1da177e4 LT |
918 | mempool_free(mpio, m->mpio_pool); |
919 | ||
920 | return r; | |
921 | } | |
922 | ||
923 | /* | |
924 | * Take a path out of use. | |
925 | */ | |
926 | static int fail_path(struct pgpath *pgpath) | |
927 | { | |
928 | unsigned long flags; | |
929 | struct multipath *m = pgpath->pg->m; | |
930 | ||
931 | spin_lock_irqsave(&m->lock, flags); | |
932 | ||
6680073d | 933 | if (!pgpath->is_active) |
1da177e4 LT |
934 | goto out; |
935 | ||
72d94861 | 936 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
937 | |
938 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
6680073d | 939 | pgpath->is_active = 0; |
1da177e4 LT |
940 | pgpath->fail_count++; |
941 | ||
942 | m->nr_valid_paths--; | |
943 | ||
944 | if (pgpath == m->current_pgpath) | |
945 | m->current_pgpath = NULL; | |
946 | ||
b15546f9 MA |
947 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
948 | pgpath->path.dev->name, m->nr_valid_paths); | |
949 | ||
fe9cf30e | 950 | schedule_work(&m->trigger_event); |
224cb3e9 | 951 | queue_work(kmultipathd, &pgpath->deactivate_path); |
1da177e4 LT |
952 | |
953 | out: | |
954 | spin_unlock_irqrestore(&m->lock, flags); | |
955 | ||
956 | return 0; | |
957 | } | |
958 | ||
959 | /* | |
960 | * Reinstate a previously-failed path | |
961 | */ | |
962 | static int reinstate_path(struct pgpath *pgpath) | |
963 | { | |
964 | int r = 0; | |
965 | unsigned long flags; | |
966 | struct multipath *m = pgpath->pg->m; | |
967 | ||
968 | spin_lock_irqsave(&m->lock, flags); | |
969 | ||
6680073d | 970 | if (pgpath->is_active) |
1da177e4 LT |
971 | goto out; |
972 | ||
def052d2 | 973 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
974 | DMWARN("Reinstate path not supported by path selector %s", |
975 | pgpath->pg->ps.type->name); | |
976 | r = -EINVAL; | |
977 | goto out; | |
978 | } | |
979 | ||
980 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
981 | if (r) | |
982 | goto out; | |
983 | ||
6680073d | 984 | pgpath->is_active = 1; |
1da177e4 | 985 | |
e54f77dd CS |
986 | if (!m->nr_valid_paths++ && m->queue_size) { |
987 | m->current_pgpath = NULL; | |
c557308e | 988 | queue_work(kmultipathd, &m->process_queued_ios); |
e54f77dd CS |
989 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
990 | if (queue_work(kmpath_handlerd, &pgpath->activate_path)) | |
991 | m->pg_init_in_progress++; | |
992 | } | |
1da177e4 | 993 | |
b15546f9 MA |
994 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
995 | pgpath->path.dev->name, m->nr_valid_paths); | |
996 | ||
fe9cf30e | 997 | schedule_work(&m->trigger_event); |
1da177e4 LT |
998 | |
999 | out: | |
1000 | spin_unlock_irqrestore(&m->lock, flags); | |
1001 | ||
1002 | return r; | |
1003 | } | |
1004 | ||
1005 | /* | |
1006 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1007 | */ | |
1008 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1009 | action_fn action) | |
1010 | { | |
1011 | int r = 0; | |
1012 | struct pgpath *pgpath; | |
1013 | struct priority_group *pg; | |
1014 | ||
1015 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1016 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1017 | if (pgpath->path.dev == dev) | |
1018 | r = action(pgpath); | |
1019 | } | |
1020 | } | |
1021 | ||
1022 | return r; | |
1023 | } | |
1024 | ||
1025 | /* | |
1026 | * Temporarily try to avoid having to use the specified PG | |
1027 | */ | |
1028 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
1029 | int bypassed) | |
1030 | { | |
1031 | unsigned long flags; | |
1032 | ||
1033 | spin_lock_irqsave(&m->lock, flags); | |
1034 | ||
1035 | pg->bypassed = bypassed; | |
1036 | m->current_pgpath = NULL; | |
1037 | m->current_pg = NULL; | |
1038 | ||
1039 | spin_unlock_irqrestore(&m->lock, flags); | |
1040 | ||
fe9cf30e | 1041 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1042 | } |
1043 | ||
1044 | /* | |
1045 | * Switch to using the specified PG from the next I/O that gets mapped | |
1046 | */ | |
1047 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1048 | { | |
1049 | struct priority_group *pg; | |
1050 | unsigned pgnum; | |
1051 | unsigned long flags; | |
1052 | ||
1053 | if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || | |
1054 | (pgnum > m->nr_priority_groups)) { | |
1055 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1056 | return -EINVAL; | |
1057 | } | |
1058 | ||
1059 | spin_lock_irqsave(&m->lock, flags); | |
1060 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1061 | pg->bypassed = 0; | |
1062 | if (--pgnum) | |
1063 | continue; | |
1064 | ||
1065 | m->current_pgpath = NULL; | |
1066 | m->current_pg = NULL; | |
1067 | m->next_pg = pg; | |
1068 | } | |
1069 | spin_unlock_irqrestore(&m->lock, flags); | |
1070 | ||
fe9cf30e | 1071 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1072 | return 0; |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * Set/clear bypassed status of a PG. | |
1077 | * PGs are numbered upwards from 1 in the order they were declared. | |
1078 | */ | |
1079 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
1080 | { | |
1081 | struct priority_group *pg; | |
1082 | unsigned pgnum; | |
1083 | ||
1084 | if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || | |
1085 | (pgnum > m->nr_priority_groups)) { | |
1086 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1087 | return -EINVAL; | |
1088 | } | |
1089 | ||
1090 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1091 | if (!--pgnum) | |
1092 | break; | |
1093 | } | |
1094 | ||
1095 | bypass_pg(m, pg, bypassed); | |
1096 | return 0; | |
1097 | } | |
1098 | ||
c9e45581 DW |
1099 | /* |
1100 | * Should we retry pg_init immediately? | |
1101 | */ | |
1102 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1103 | { | |
1104 | unsigned long flags; | |
1105 | int limit_reached = 0; | |
1106 | ||
1107 | spin_lock_irqsave(&m->lock, flags); | |
1108 | ||
1109 | if (m->pg_init_count <= m->pg_init_retries) | |
1110 | m->pg_init_required = 1; | |
1111 | else | |
1112 | limit_reached = 1; | |
1113 | ||
1114 | spin_unlock_irqrestore(&m->lock, flags); | |
1115 | ||
1116 | return limit_reached; | |
1117 | } | |
1118 | ||
cfae5c9b CS |
1119 | static void pg_init_done(struct dm_path *path, int errors) |
1120 | { | |
1121 | struct pgpath *pgpath = path_to_pgpath(path); | |
1122 | struct priority_group *pg = pgpath->pg; | |
1123 | struct multipath *m = pg->m; | |
1124 | unsigned long flags; | |
1125 | ||
1126 | /* device or driver problems */ | |
1127 | switch (errors) { | |
1128 | case SCSI_DH_OK: | |
1129 | break; | |
1130 | case SCSI_DH_NOSYS: | |
1131 | if (!m->hw_handler_name) { | |
1132 | errors = 0; | |
1133 | break; | |
1134 | } | |
1135 | DMERR("Cannot failover device because scsi_dh_%s was not " | |
1136 | "loaded.", m->hw_handler_name); | |
1137 | /* | |
1138 | * Fail path for now, so we do not ping pong | |
1139 | */ | |
1140 | fail_path(pgpath); | |
1141 | break; | |
1142 | case SCSI_DH_DEV_TEMP_BUSY: | |
1143 | /* | |
1144 | * Probably doing something like FW upgrade on the | |
1145 | * controller so try the other pg. | |
1146 | */ | |
1147 | bypass_pg(m, pg, 1); | |
1148 | break; | |
1149 | /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */ | |
1150 | case SCSI_DH_RETRY: | |
1151 | case SCSI_DH_IMM_RETRY: | |
1152 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1153 | if (pg_init_limit_reached(m, pgpath)) | |
1154 | fail_path(pgpath); | |
1155 | errors = 0; | |
1156 | break; | |
1157 | default: | |
1158 | /* | |
1159 | * We probably do not want to fail the path for a device | |
1160 | * error, but this is what the old dm did. In future | |
1161 | * patches we can do more advanced handling. | |
1162 | */ | |
1163 | fail_path(pgpath); | |
1164 | } | |
1165 | ||
1166 | spin_lock_irqsave(&m->lock, flags); | |
1167 | if (errors) { | |
e54f77dd CS |
1168 | if (pgpath == m->current_pgpath) { |
1169 | DMERR("Could not failover device. Error %d.", errors); | |
1170 | m->current_pgpath = NULL; | |
1171 | m->current_pg = NULL; | |
1172 | } | |
cfae5c9b CS |
1173 | } else if (!m->pg_init_required) { |
1174 | m->queue_io = 0; | |
1175 | pg->bypassed = 0; | |
1176 | } | |
1177 | ||
e54f77dd CS |
1178 | m->pg_init_in_progress--; |
1179 | if (!m->pg_init_in_progress) | |
1180 | queue_work(kmultipathd, &m->process_queued_ios); | |
cfae5c9b CS |
1181 | spin_unlock_irqrestore(&m->lock, flags); |
1182 | } | |
1183 | ||
bab7cfc7 CS |
1184 | static void activate_path(struct work_struct *work) |
1185 | { | |
1186 | int ret; | |
e54f77dd CS |
1187 | struct pgpath *pgpath = |
1188 | container_of(work, struct pgpath, activate_path); | |
bab7cfc7 | 1189 | |
e54f77dd CS |
1190 | ret = scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev)); |
1191 | pg_init_done(&pgpath->path, ret); | |
bab7cfc7 CS |
1192 | } |
1193 | ||
1da177e4 LT |
1194 | /* |
1195 | * end_io handling | |
1196 | */ | |
f40c67f0 | 1197 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1198 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1199 | { |
f40c67f0 KU |
1200 | /* |
1201 | * We don't queue any clone request inside the multipath target | |
1202 | * during end I/O handling, since those clone requests don't have | |
1203 | * bio clones. If we queue them inside the multipath target, | |
1204 | * we need to make bio clones, that requires memory allocation. | |
1205 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1206 | * don't have bio clones.) | |
1207 | * Instead of queueing the clone request here, we queue the original | |
1208 | * request into dm core, which will remake a clone request and | |
1209 | * clone bios for it and resubmit it later. | |
1210 | */ | |
1211 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1212 | unsigned long flags; |
1da177e4 | 1213 | |
f40c67f0 | 1214 | if (!error && !clone->errors) |
1da177e4 LT |
1215 | return 0; /* I/O complete */ |
1216 | ||
f6a80ea8 AK |
1217 | if (error == -EOPNOTSUPP) |
1218 | return error; | |
1219 | ||
cfae5c9b CS |
1220 | if (mpio->pgpath) |
1221 | fail_path(mpio->pgpath); | |
1da177e4 | 1222 | |
640eb3b0 | 1223 | spin_lock_irqsave(&m->lock, flags); |
f40c67f0 KU |
1224 | if (!m->nr_valid_paths && !m->queue_if_no_path && !__must_push_back(m)) |
1225 | r = -EIO; | |
640eb3b0 | 1226 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1227 | |
f40c67f0 | 1228 | return r; |
1da177e4 LT |
1229 | } |
1230 | ||
f40c67f0 | 1231 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1232 | int error, union map_info *map_context) |
1233 | { | |
028867ac AK |
1234 | struct multipath *m = ti->private; |
1235 | struct dm_mpath_io *mpio = map_context->ptr; | |
1da177e4 LT |
1236 | struct pgpath *pgpath = mpio->pgpath; |
1237 | struct path_selector *ps; | |
1238 | int r; | |
1239 | ||
f40c67f0 | 1240 | r = do_end_io(m, clone, error, mpio); |
1da177e4 LT |
1241 | if (pgpath) { |
1242 | ps = &pgpath->pg->ps; | |
1243 | if (ps->type->end_io) | |
02ab823f | 1244 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1245 | } |
f40c67f0 | 1246 | mempool_free(mpio, m->mpio_pool); |
1da177e4 LT |
1247 | |
1248 | return r; | |
1249 | } | |
1250 | ||
1251 | /* | |
1252 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1253 | * the last path fails we must error any remaining I/O. |
1254 | * Note that if the freeze_bdev fails while suspending, the | |
1255 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1256 | */ |
1257 | static void multipath_presuspend(struct dm_target *ti) | |
1258 | { | |
1259 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1260 | |
485ef69e | 1261 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1262 | } |
1263 | ||
436d4108 AK |
1264 | /* |
1265 | * Restore the queue_if_no_path setting. | |
1266 | */ | |
1da177e4 LT |
1267 | static void multipath_resume(struct dm_target *ti) |
1268 | { | |
1269 | struct multipath *m = (struct multipath *) ti->private; | |
1270 | unsigned long flags; | |
1271 | ||
1272 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1273 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1274 | spin_unlock_irqrestore(&m->lock, flags); |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Info output has the following format: | |
1279 | * num_multipath_feature_args [multipath_feature_args]* | |
1280 | * num_handler_status_args [handler_status_args]* | |
1281 | * num_groups init_group_number | |
1282 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1283 | * num_paths num_selector_args | |
1284 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1285 | * | |
1286 | * Table output has the following format (identical to the constructor string): | |
1287 | * num_feature_args [features_args]* | |
1288 | * num_handler_args hw_handler [hw_handler_args]* | |
1289 | * num_groups init_group_number | |
1290 | * [priority selector-name num_ps_args [ps_args]* | |
1291 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1292 | */ | |
1293 | static int multipath_status(struct dm_target *ti, status_type_t type, | |
1294 | char *result, unsigned int maxlen) | |
1295 | { | |
1296 | int sz = 0; | |
1297 | unsigned long flags; | |
1298 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1299 | struct priority_group *pg; |
1300 | struct pgpath *p; | |
1301 | unsigned pg_num; | |
1302 | char state; | |
1303 | ||
1304 | spin_lock_irqsave(&m->lock, flags); | |
1305 | ||
1306 | /* Features */ | |
1307 | if (type == STATUSTYPE_INFO) | |
c9e45581 DW |
1308 | DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count); |
1309 | else { | |
1310 | DMEMIT("%u ", m->queue_if_no_path + | |
1311 | (m->pg_init_retries > 0) * 2); | |
1312 | if (m->queue_if_no_path) | |
1313 | DMEMIT("queue_if_no_path "); | |
1314 | if (m->pg_init_retries) | |
1315 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
1316 | } | |
1da177e4 | 1317 | |
cfae5c9b | 1318 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1319 | DMEMIT("0 "); |
1320 | else | |
cfae5c9b | 1321 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1322 | |
1323 | DMEMIT("%u ", m->nr_priority_groups); | |
1324 | ||
1325 | if (m->next_pg) | |
1326 | pg_num = m->next_pg->pg_num; | |
1327 | else if (m->current_pg) | |
1328 | pg_num = m->current_pg->pg_num; | |
1329 | else | |
1330 | pg_num = 1; | |
1331 | ||
1332 | DMEMIT("%u ", pg_num); | |
1333 | ||
1334 | switch (type) { | |
1335 | case STATUSTYPE_INFO: | |
1336 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1337 | if (pg->bypassed) | |
1338 | state = 'D'; /* Disabled */ | |
1339 | else if (pg == m->current_pg) | |
1340 | state = 'A'; /* Currently Active */ | |
1341 | else | |
1342 | state = 'E'; /* Enabled */ | |
1343 | ||
1344 | DMEMIT("%c ", state); | |
1345 | ||
1346 | if (pg->ps.type->status) | |
1347 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1348 | result + sz, | |
1349 | maxlen - sz); | |
1350 | else | |
1351 | DMEMIT("0 "); | |
1352 | ||
1353 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1354 | pg->ps.type->info_args); | |
1355 | ||
1356 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1357 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1358 | p->is_active ? "A" : "F", |
1da177e4 LT |
1359 | p->fail_count); |
1360 | if (pg->ps.type->status) | |
1361 | sz += pg->ps.type->status(&pg->ps, | |
1362 | &p->path, type, result + sz, | |
1363 | maxlen - sz); | |
1364 | } | |
1365 | } | |
1366 | break; | |
1367 | ||
1368 | case STATUSTYPE_TABLE: | |
1369 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1370 | DMEMIT("%s ", pg->ps.type->name); | |
1371 | ||
1372 | if (pg->ps.type->status) | |
1373 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1374 | result + sz, | |
1375 | maxlen - sz); | |
1376 | else | |
1377 | DMEMIT("0 "); | |
1378 | ||
1379 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1380 | pg->ps.type->table_args); | |
1381 | ||
1382 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1383 | DMEMIT("%s ", p->path.dev->name); | |
1384 | if (pg->ps.type->status) | |
1385 | sz += pg->ps.type->status(&pg->ps, | |
1386 | &p->path, type, result + sz, | |
1387 | maxlen - sz); | |
1388 | } | |
1389 | } | |
1390 | break; | |
1391 | } | |
1392 | ||
1393 | spin_unlock_irqrestore(&m->lock, flags); | |
1394 | ||
1395 | return 0; | |
1396 | } | |
1397 | ||
1398 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1399 | { | |
1400 | int r; | |
1401 | struct dm_dev *dev; | |
1402 | struct multipath *m = (struct multipath *) ti->private; | |
1403 | action_fn action; | |
1404 | ||
1405 | if (argc == 1) { | |
1406 | if (!strnicmp(argv[0], MESG_STR("queue_if_no_path"))) | |
485ef69e | 1407 | return queue_if_no_path(m, 1, 0); |
1da177e4 | 1408 | else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path"))) |
485ef69e | 1409 | return queue_if_no_path(m, 0, 0); |
1da177e4 LT |
1410 | } |
1411 | ||
1412 | if (argc != 2) | |
1413 | goto error; | |
1414 | ||
1415 | if (!strnicmp(argv[0], MESG_STR("disable_group"))) | |
1416 | return bypass_pg_num(m, argv[1], 1); | |
1417 | else if (!strnicmp(argv[0], MESG_STR("enable_group"))) | |
1418 | return bypass_pg_num(m, argv[1], 0); | |
1419 | else if (!strnicmp(argv[0], MESG_STR("switch_group"))) | |
1420 | return switch_pg_num(m, argv[1]); | |
1421 | else if (!strnicmp(argv[0], MESG_STR("reinstate_path"))) | |
1422 | action = reinstate_path; | |
1423 | else if (!strnicmp(argv[0], MESG_STR("fail_path"))) | |
1424 | action = fail_path; | |
1425 | else | |
1426 | goto error; | |
1427 | ||
1428 | r = dm_get_device(ti, argv[1], ti->begin, ti->len, | |
1429 | dm_table_get_mode(ti->table), &dev); | |
1430 | if (r) { | |
72d94861 | 1431 | DMWARN("message: error getting device %s", |
1da177e4 LT |
1432 | argv[1]); |
1433 | return -EINVAL; | |
1434 | } | |
1435 | ||
1436 | r = action_dev(m, dev, action); | |
1437 | ||
1438 | dm_put_device(ti, dev); | |
1439 | ||
1440 | return r; | |
1441 | ||
1442 | error: | |
1443 | DMWARN("Unrecognised multipath message received."); | |
1444 | return -EINVAL; | |
1445 | } | |
1446 | ||
647b3d00 | 1447 | static int multipath_ioctl(struct dm_target *ti, unsigned int cmd, |
9af4aa30 MB |
1448 | unsigned long arg) |
1449 | { | |
1450 | struct multipath *m = (struct multipath *) ti->private; | |
1451 | struct block_device *bdev = NULL; | |
633a08b8 | 1452 | fmode_t mode = 0; |
9af4aa30 MB |
1453 | unsigned long flags; |
1454 | int r = 0; | |
1455 | ||
1456 | spin_lock_irqsave(&m->lock, flags); | |
1457 | ||
1458 | if (!m->current_pgpath) | |
02ab823f | 1459 | __choose_pgpath(m, 0); |
9af4aa30 | 1460 | |
e90dae1f | 1461 | if (m->current_pgpath) { |
9af4aa30 | 1462 | bdev = m->current_pgpath->path.dev->bdev; |
633a08b8 | 1463 | mode = m->current_pgpath->path.dev->mode; |
e90dae1f | 1464 | } |
9af4aa30 MB |
1465 | |
1466 | if (m->queue_io) | |
1467 | r = -EAGAIN; | |
1468 | else if (!bdev) | |
1469 | r = -EIO; | |
1470 | ||
1471 | spin_unlock_irqrestore(&m->lock, flags); | |
1472 | ||
633a08b8 | 1473 | return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
9af4aa30 MB |
1474 | } |
1475 | ||
af4874e0 MS |
1476 | static int multipath_iterate_devices(struct dm_target *ti, |
1477 | iterate_devices_callout_fn fn, void *data) | |
1478 | { | |
1479 | struct multipath *m = ti->private; | |
1480 | struct priority_group *pg; | |
1481 | struct pgpath *p; | |
1482 | int ret = 0; | |
1483 | ||
1484 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1485 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1486 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1487 | if (ret) |
1488 | goto out; | |
1489 | } | |
1490 | } | |
1491 | ||
1492 | out: | |
1493 | return ret; | |
1494 | } | |
1495 | ||
f40c67f0 KU |
1496 | static int __pgpath_busy(struct pgpath *pgpath) |
1497 | { | |
1498 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1499 | ||
1500 | return dm_underlying_device_busy(q); | |
1501 | } | |
1502 | ||
1503 | /* | |
1504 | * We return "busy", only when we can map I/Os but underlying devices | |
1505 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1506 | * the underlying queue). | |
1507 | * In other words, if we want to kill I/Os or queue them inside us | |
1508 | * due to map unavailability, we don't return "busy". Otherwise, | |
1509 | * dm core won't give us the I/Os and we can't do what we want. | |
1510 | */ | |
1511 | static int multipath_busy(struct dm_target *ti) | |
1512 | { | |
1513 | int busy = 0, has_active = 0; | |
1514 | struct multipath *m = ti->private; | |
1515 | struct priority_group *pg; | |
1516 | struct pgpath *pgpath; | |
1517 | unsigned long flags; | |
1518 | ||
1519 | spin_lock_irqsave(&m->lock, flags); | |
1520 | ||
1521 | /* Guess which priority_group will be used at next mapping time */ | |
1522 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1523 | pg = m->next_pg; | |
1524 | else if (likely(m->current_pg)) | |
1525 | pg = m->current_pg; | |
1526 | else | |
1527 | /* | |
1528 | * We don't know which pg will be used at next mapping time. | |
1529 | * We don't call __choose_pgpath() here to avoid to trigger | |
1530 | * pg_init just by busy checking. | |
1531 | * So we don't know whether underlying devices we will be using | |
1532 | * at next mapping time are busy or not. Just try mapping. | |
1533 | */ | |
1534 | goto out; | |
1535 | ||
1536 | /* | |
1537 | * If there is one non-busy active path at least, the path selector | |
1538 | * will be able to select it. So we consider such a pg as not busy. | |
1539 | */ | |
1540 | busy = 1; | |
1541 | list_for_each_entry(pgpath, &pg->pgpaths, list) | |
1542 | if (pgpath->is_active) { | |
1543 | has_active = 1; | |
1544 | ||
1545 | if (!__pgpath_busy(pgpath)) { | |
1546 | busy = 0; | |
1547 | break; | |
1548 | } | |
1549 | } | |
1550 | ||
1551 | if (!has_active) | |
1552 | /* | |
1553 | * No active path in this pg, so this pg won't be used and | |
1554 | * the current_pg will be changed at next mapping time. | |
1555 | * We need to try mapping to determine it. | |
1556 | */ | |
1557 | busy = 0; | |
1558 | ||
1559 | out: | |
1560 | spin_unlock_irqrestore(&m->lock, flags); | |
1561 | ||
1562 | return busy; | |
1563 | } | |
1564 | ||
1da177e4 LT |
1565 | /*----------------------------------------------------------------- |
1566 | * Module setup | |
1567 | *---------------------------------------------------------------*/ | |
1568 | static struct target_type multipath_target = { | |
1569 | .name = "multipath", | |
af4874e0 | 1570 | .version = {1, 1, 0}, |
1da177e4 LT |
1571 | .module = THIS_MODULE, |
1572 | .ctr = multipath_ctr, | |
1573 | .dtr = multipath_dtr, | |
f40c67f0 KU |
1574 | .map_rq = multipath_map, |
1575 | .rq_end_io = multipath_end_io, | |
1da177e4 LT |
1576 | .presuspend = multipath_presuspend, |
1577 | .resume = multipath_resume, | |
1578 | .status = multipath_status, | |
1579 | .message = multipath_message, | |
9af4aa30 | 1580 | .ioctl = multipath_ioctl, |
af4874e0 | 1581 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1582 | .busy = multipath_busy, |
1da177e4 LT |
1583 | }; |
1584 | ||
1585 | static int __init dm_multipath_init(void) | |
1586 | { | |
1587 | int r; | |
1588 | ||
1589 | /* allocate a slab for the dm_ios */ | |
028867ac | 1590 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1591 | if (!_mpio_cache) |
1592 | return -ENOMEM; | |
1593 | ||
1594 | r = dm_register_target(&multipath_target); | |
1595 | if (r < 0) { | |
0cd33124 | 1596 | DMERR("register failed %d", r); |
1da177e4 LT |
1597 | kmem_cache_destroy(_mpio_cache); |
1598 | return -EINVAL; | |
1599 | } | |
1600 | ||
c557308e AK |
1601 | kmultipathd = create_workqueue("kmpathd"); |
1602 | if (!kmultipathd) { | |
0cd33124 | 1603 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1604 | dm_unregister_target(&multipath_target); |
1605 | kmem_cache_destroy(_mpio_cache); | |
1606 | return -ENOMEM; | |
1607 | } | |
1608 | ||
bab7cfc7 CS |
1609 | /* |
1610 | * A separate workqueue is used to handle the device handlers | |
1611 | * to avoid overloading existing workqueue. Overloading the | |
1612 | * old workqueue would also create a bottleneck in the | |
1613 | * path of the storage hardware device activation. | |
1614 | */ | |
1615 | kmpath_handlerd = create_singlethread_workqueue("kmpath_handlerd"); | |
1616 | if (!kmpath_handlerd) { | |
1617 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1618 | destroy_workqueue(kmultipathd); | |
1619 | dm_unregister_target(&multipath_target); | |
1620 | kmem_cache_destroy(_mpio_cache); | |
1621 | return -ENOMEM; | |
1622 | } | |
1623 | ||
72d94861 | 1624 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1625 | multipath_target.version[0], multipath_target.version[1], |
1626 | multipath_target.version[2]); | |
1627 | ||
1628 | return r; | |
1629 | } | |
1630 | ||
1631 | static void __exit dm_multipath_exit(void) | |
1632 | { | |
bab7cfc7 | 1633 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1634 | destroy_workqueue(kmultipathd); |
1635 | ||
10d3bd09 | 1636 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1637 | kmem_cache_destroy(_mpio_cache); |
1638 | } | |
1639 | ||
1da177e4 LT |
1640 | module_init(dm_multipath_init); |
1641 | module_exit(dm_multipath_exit); | |
1642 | ||
1643 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1644 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1645 | MODULE_LICENSE("GPL"); |