]>
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 | ||
f4790826 | 10 | #include "dm.h" |
1da177e4 | 11 | #include "dm-path-selector.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> | |
35991652 | 22 | #include <linux/delay.h> |
cfae5c9b | 23 | #include <scsi/scsi_dh.h> |
60063497 | 24 | #include <linux/atomic.h> |
1da177e4 | 25 | |
72d94861 | 26 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
27 | #define DM_PG_INIT_DELAY_MSECS 2000 |
28 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
1da177e4 LT |
29 | |
30 | /* Path properties */ | |
31 | struct pgpath { | |
32 | struct list_head list; | |
33 | ||
34 | struct priority_group *pg; /* Owning PG */ | |
6680073d | 35 | unsigned is_active; /* Path status */ |
1da177e4 LT |
36 | unsigned fail_count; /* Cumulative failure count */ |
37 | ||
c922d5f7 | 38 | struct dm_path path; |
4e2d19e4 | 39 | struct delayed_work activate_path; |
1da177e4 LT |
40 | }; |
41 | ||
42 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
43 | ||
44 | /* | |
45 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
46 | * Each has a path selector which controls which path gets used. | |
47 | */ | |
48 | struct priority_group { | |
49 | struct list_head list; | |
50 | ||
51 | struct multipath *m; /* Owning multipath instance */ | |
52 | struct path_selector ps; | |
53 | ||
54 | unsigned pg_num; /* Reference number */ | |
55 | unsigned bypassed; /* Temporarily bypass this PG? */ | |
56 | ||
57 | unsigned nr_pgpaths; /* Number of paths in PG */ | |
58 | struct list_head pgpaths; | |
59 | }; | |
60 | ||
61 | /* Multipath context */ | |
62 | struct multipath { | |
63 | struct list_head list; | |
64 | struct dm_target *ti; | |
65 | ||
cfae5c9b | 66 | const char *hw_handler_name; |
2bfd2e13 | 67 | char *hw_handler_params; |
4e2d19e4 | 68 | |
1fbdd2b3 MS |
69 | spinlock_t lock; |
70 | ||
1da177e4 LT |
71 | unsigned nr_priority_groups; |
72 | struct list_head priority_groups; | |
4e2d19e4 CS |
73 | |
74 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
75 | ||
1da177e4 | 76 | unsigned pg_init_required; /* pg_init needs calling? */ |
c3cd4f6b | 77 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
4e2d19e4 | 78 | unsigned pg_init_delay_retry; /* Delay pg_init retry? */ |
1da177e4 LT |
79 | |
80 | unsigned nr_valid_paths; /* Total number of usable paths */ | |
81 | struct pgpath *current_pgpath; | |
82 | struct priority_group *current_pg; | |
83 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
84 | unsigned repeat_count; /* I/Os left before calling PS again */ | |
85 | ||
1fbdd2b3 MS |
86 | unsigned queue_io:1; /* Must we queue all I/O? */ |
87 | unsigned queue_if_no_path:1; /* Queue I/O if last path fails? */ | |
88 | unsigned saved_queue_if_no_path:1; /* Saved state during suspension */ | |
a58a935d | 89 | unsigned retain_attached_hw_handler:1; /* If there's already a hw_handler present, don't change it. */ |
954a73d5 | 90 | unsigned pg_init_disabled:1; /* pg_init is not currently allowed */ |
1fbdd2b3 | 91 | |
c9e45581 DW |
92 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
93 | unsigned pg_init_count; /* Number of times pg_init called */ | |
4e2d19e4 | 94 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
1da177e4 | 95 | |
1fbdd2b3 | 96 | unsigned queue_size; |
1da177e4 | 97 | struct work_struct process_queued_ios; |
f40c67f0 | 98 | struct list_head queued_ios; |
1da177e4 LT |
99 | |
100 | struct work_struct trigger_event; | |
101 | ||
102 | /* | |
028867ac | 103 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
104 | * can resubmit bios on error. |
105 | */ | |
106 | mempool_t *mpio_pool; | |
6380f26f MA |
107 | |
108 | struct mutex work_mutex; | |
1da177e4 LT |
109 | }; |
110 | ||
111 | /* | |
112 | * Context information attached to each bio we process. | |
113 | */ | |
028867ac | 114 | struct dm_mpath_io { |
1da177e4 | 115 | struct pgpath *pgpath; |
02ab823f | 116 | size_t nr_bytes; |
1da177e4 LT |
117 | }; |
118 | ||
119 | typedef int (*action_fn) (struct pgpath *pgpath); | |
120 | ||
e18b890b | 121 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 122 | |
bab7cfc7 | 123 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 DH |
124 | static void process_queued_ios(struct work_struct *work); |
125 | static void trigger_event(struct work_struct *work); | |
bab7cfc7 | 126 | static void activate_path(struct work_struct *work); |
1da177e4 LT |
127 | |
128 | ||
129 | /*----------------------------------------------- | |
130 | * Allocation routines | |
131 | *-----------------------------------------------*/ | |
132 | ||
133 | static struct pgpath *alloc_pgpath(void) | |
134 | { | |
e69fae56 | 135 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 136 | |
224cb3e9 | 137 | if (pgpath) { |
6680073d | 138 | pgpath->is_active = 1; |
4e2d19e4 | 139 | INIT_DELAYED_WORK(&pgpath->activate_path, activate_path); |
224cb3e9 | 140 | } |
1da177e4 LT |
141 | |
142 | return pgpath; | |
143 | } | |
144 | ||
028867ac | 145 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
146 | { |
147 | kfree(pgpath); | |
148 | } | |
149 | ||
150 | static struct priority_group *alloc_priority_group(void) | |
151 | { | |
152 | struct priority_group *pg; | |
153 | ||
e69fae56 | 154 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 155 | |
e69fae56 MM |
156 | if (pg) |
157 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
158 | |
159 | return pg; | |
160 | } | |
161 | ||
162 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
163 | { | |
164 | struct pgpath *pgpath, *tmp; | |
ae11b1b3 | 165 | struct multipath *m = ti->private; |
1da177e4 LT |
166 | |
167 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
168 | list_del(&pgpath->list); | |
ae11b1b3 HR |
169 | if (m->hw_handler_name) |
170 | scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev)); | |
1da177e4 LT |
171 | dm_put_device(ti, pgpath->path.dev); |
172 | free_pgpath(pgpath); | |
173 | } | |
174 | } | |
175 | ||
176 | static void free_priority_group(struct priority_group *pg, | |
177 | struct dm_target *ti) | |
178 | { | |
179 | struct path_selector *ps = &pg->ps; | |
180 | ||
181 | if (ps->type) { | |
182 | ps->type->destroy(ps); | |
183 | dm_put_path_selector(ps->type); | |
184 | } | |
185 | ||
186 | free_pgpaths(&pg->pgpaths, ti); | |
187 | kfree(pg); | |
188 | } | |
189 | ||
28f16c20 | 190 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
191 | { |
192 | struct multipath *m; | |
f4790826 | 193 | unsigned min_ios = dm_get_reserved_rq_based_ios(); |
1da177e4 | 194 | |
e69fae56 | 195 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 196 | if (m) { |
1da177e4 | 197 | INIT_LIST_HEAD(&m->priority_groups); |
f40c67f0 | 198 | INIT_LIST_HEAD(&m->queued_ios); |
1da177e4 LT |
199 | spin_lock_init(&m->lock); |
200 | m->queue_io = 1; | |
4e2d19e4 | 201 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; |
c4028958 DH |
202 | INIT_WORK(&m->process_queued_ios, process_queued_ios); |
203 | INIT_WORK(&m->trigger_event, trigger_event); | |
2bded7bd | 204 | init_waitqueue_head(&m->pg_init_wait); |
6380f26f | 205 | mutex_init(&m->work_mutex); |
f4790826 | 206 | m->mpio_pool = mempool_create_slab_pool(min_ios, _mpio_cache); |
1da177e4 LT |
207 | if (!m->mpio_pool) { |
208 | kfree(m); | |
209 | return NULL; | |
210 | } | |
28f16c20 MM |
211 | m->ti = ti; |
212 | ti->private = m; | |
1da177e4 LT |
213 | } |
214 | ||
215 | return m; | |
216 | } | |
217 | ||
218 | static void free_multipath(struct multipath *m) | |
219 | { | |
220 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
221 | |
222 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
223 | list_del(&pg->list); | |
224 | free_priority_group(pg, m->ti); | |
225 | } | |
226 | ||
cfae5c9b | 227 | kfree(m->hw_handler_name); |
2bfd2e13 | 228 | kfree(m->hw_handler_params); |
1da177e4 LT |
229 | mempool_destroy(m->mpio_pool); |
230 | kfree(m); | |
231 | } | |
232 | ||
466891f9 JN |
233 | static int set_mapinfo(struct multipath *m, union map_info *info) |
234 | { | |
235 | struct dm_mpath_io *mpio; | |
236 | ||
237 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); | |
238 | if (!mpio) | |
239 | return -ENOMEM; | |
240 | ||
241 | memset(mpio, 0, sizeof(*mpio)); | |
242 | info->ptr = mpio; | |
243 | ||
244 | return 0; | |
245 | } | |
246 | ||
247 | static void clear_mapinfo(struct multipath *m, union map_info *info) | |
248 | { | |
249 | struct dm_mpath_io *mpio = info->ptr; | |
250 | ||
251 | info->ptr = NULL; | |
252 | mempool_free(mpio, m->mpio_pool); | |
253 | } | |
1da177e4 LT |
254 | |
255 | /*----------------------------------------------- | |
256 | * Path selection | |
257 | *-----------------------------------------------*/ | |
258 | ||
fb612642 KU |
259 | static void __pg_init_all_paths(struct multipath *m) |
260 | { | |
261 | struct pgpath *pgpath; | |
4e2d19e4 | 262 | unsigned long pg_init_delay = 0; |
fb612642 KU |
263 | |
264 | m->pg_init_count++; | |
265 | m->pg_init_required = 0; | |
4e2d19e4 CS |
266 | if (m->pg_init_delay_retry) |
267 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? | |
268 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
269 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
270 | /* Skip failed paths */ | |
271 | if (!pgpath->is_active) | |
272 | continue; | |
4e2d19e4 CS |
273 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
274 | pg_init_delay)) | |
fb612642 KU |
275 | m->pg_init_in_progress++; |
276 | } | |
277 | } | |
278 | ||
1da177e4 LT |
279 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) |
280 | { | |
1da177e4 LT |
281 | m->current_pg = pgpath->pg; |
282 | ||
283 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 284 | if (m->hw_handler_name) { |
1da177e4 LT |
285 | m->pg_init_required = 1; |
286 | m->queue_io = 1; | |
287 | } else { | |
288 | m->pg_init_required = 0; | |
289 | m->queue_io = 0; | |
290 | } | |
c9e45581 DW |
291 | |
292 | m->pg_init_count = 0; | |
1da177e4 LT |
293 | } |
294 | ||
02ab823f KU |
295 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
296 | size_t nr_bytes) | |
1da177e4 | 297 | { |
c922d5f7 | 298 | struct dm_path *path; |
1da177e4 | 299 | |
02ab823f | 300 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes); |
1da177e4 LT |
301 | if (!path) |
302 | return -ENXIO; | |
303 | ||
304 | m->current_pgpath = path_to_pgpath(path); | |
305 | ||
306 | if (m->current_pg != pg) | |
307 | __switch_pg(m, m->current_pgpath); | |
308 | ||
309 | return 0; | |
310 | } | |
311 | ||
02ab823f | 312 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
313 | { |
314 | struct priority_group *pg; | |
315 | unsigned bypassed = 1; | |
316 | ||
317 | if (!m->nr_valid_paths) | |
318 | goto failed; | |
319 | ||
320 | /* Were we instructed to switch PG? */ | |
321 | if (m->next_pg) { | |
322 | pg = m->next_pg; | |
323 | m->next_pg = NULL; | |
02ab823f | 324 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
325 | return; |
326 | } | |
327 | ||
328 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 329 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
330 | return; |
331 | ||
332 | /* | |
333 | * Loop through priority groups until we find a valid path. | |
334 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
335 | * Second time we only try the ones we skipped, but set |
336 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
337 | */ |
338 | do { | |
339 | list_for_each_entry(pg, &m->priority_groups, list) { | |
340 | if (pg->bypassed == bypassed) | |
341 | continue; | |
f220fd4e MC |
342 | if (!__choose_path_in_pg(m, pg, nr_bytes)) { |
343 | if (!bypassed) | |
344 | m->pg_init_delay_retry = 1; | |
1da177e4 | 345 | return; |
f220fd4e | 346 | } |
1da177e4 LT |
347 | } |
348 | } while (bypassed--); | |
349 | ||
350 | failed: | |
351 | m->current_pgpath = NULL; | |
352 | m->current_pg = NULL; | |
353 | } | |
354 | ||
45e15720 KU |
355 | /* |
356 | * Check whether bios must be queued in the device-mapper core rather | |
357 | * than here in the target. | |
358 | * | |
359 | * m->lock must be held on entry. | |
360 | * | |
361 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
362 | * same value then we are not between multipath_presuspend() | |
363 | * and multipath_resume() calls and we have no need to check | |
364 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
365 | */ | |
366 | static int __must_push_back(struct multipath *m) | |
367 | { | |
368 | return (m->queue_if_no_path != m->saved_queue_if_no_path && | |
369 | dm_noflush_suspending(m->ti)); | |
370 | } | |
371 | ||
f40c67f0 | 372 | static int map_io(struct multipath *m, struct request *clone, |
466891f9 | 373 | union map_info *map_context, unsigned was_queued) |
1da177e4 | 374 | { |
d2a7ad29 | 375 | int r = DM_MAPIO_REMAPPED; |
f40c67f0 | 376 | size_t nr_bytes = blk_rq_bytes(clone); |
1da177e4 LT |
377 | unsigned long flags; |
378 | struct pgpath *pgpath; | |
f40c67f0 | 379 | struct block_device *bdev; |
466891f9 | 380 | struct dm_mpath_io *mpio = map_context->ptr; |
1da177e4 LT |
381 | |
382 | spin_lock_irqsave(&m->lock, flags); | |
383 | ||
384 | /* Do we need to select a new pgpath? */ | |
385 | if (!m->current_pgpath || | |
386 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
02ab823f | 387 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
388 | |
389 | pgpath = m->current_pgpath; | |
390 | ||
391 | if (was_queued) | |
392 | m->queue_size--; | |
393 | ||
394 | if ((pgpath && m->queue_io) || | |
436d4108 | 395 | (!pgpath && m->queue_if_no_path)) { |
1da177e4 | 396 | /* Queue for the daemon to resubmit */ |
f40c67f0 | 397 | list_add_tail(&clone->queuelist, &m->queued_ios); |
1da177e4 | 398 | m->queue_size++; |
c3cd4f6b AK |
399 | if ((m->pg_init_required && !m->pg_init_in_progress) || |
400 | !m->queue_io) | |
c557308e | 401 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 402 | pgpath = NULL; |
d2a7ad29 | 403 | r = DM_MAPIO_SUBMITTED; |
f40c67f0 KU |
404 | } else if (pgpath) { |
405 | bdev = pgpath->path.dev->bdev; | |
406 | clone->q = bdev_get_queue(bdev); | |
407 | clone->rq_disk = bdev->bd_disk; | |
408 | } else if (__must_push_back(m)) | |
45e15720 KU |
409 | r = DM_MAPIO_REQUEUE; |
410 | else | |
411 | r = -EIO; /* Failed */ | |
1da177e4 LT |
412 | |
413 | mpio->pgpath = pgpath; | |
02ab823f KU |
414 | mpio->nr_bytes = nr_bytes; |
415 | ||
416 | if (r == DM_MAPIO_REMAPPED && pgpath->pg->ps.type->start_io) | |
417 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, &pgpath->path, | |
418 | nr_bytes); | |
1da177e4 LT |
419 | |
420 | spin_unlock_irqrestore(&m->lock, flags); | |
421 | ||
422 | return r; | |
423 | } | |
424 | ||
425 | /* | |
426 | * If we run out of usable paths, should we queue I/O or error it? | |
427 | */ | |
485ef69e AK |
428 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
429 | unsigned save_old_value) | |
1da177e4 LT |
430 | { |
431 | unsigned long flags; | |
432 | ||
433 | spin_lock_irqsave(&m->lock, flags); | |
434 | ||
485ef69e AK |
435 | if (save_old_value) |
436 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
437 | else | |
438 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 439 | m->queue_if_no_path = queue_if_no_path; |
c3cd4f6b | 440 | if (!m->queue_if_no_path && m->queue_size) |
c557308e | 441 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 LT |
442 | |
443 | spin_unlock_irqrestore(&m->lock, flags); | |
444 | ||
445 | return 0; | |
446 | } | |
447 | ||
448 | /*----------------------------------------------------------------- | |
449 | * The multipath daemon is responsible for resubmitting queued ios. | |
450 | *---------------------------------------------------------------*/ | |
451 | ||
452 | static void dispatch_queued_ios(struct multipath *m) | |
453 | { | |
454 | int r; | |
455 | unsigned long flags; | |
1da177e4 | 456 | union map_info *info; |
f40c67f0 KU |
457 | struct request *clone, *n; |
458 | LIST_HEAD(cl); | |
1da177e4 LT |
459 | |
460 | spin_lock_irqsave(&m->lock, flags); | |
f40c67f0 | 461 | list_splice_init(&m->queued_ios, &cl); |
1da177e4 LT |
462 | spin_unlock_irqrestore(&m->lock, flags); |
463 | ||
f40c67f0 KU |
464 | list_for_each_entry_safe(clone, n, &cl, queuelist) { |
465 | list_del_init(&clone->queuelist); | |
1da177e4 | 466 | |
f40c67f0 | 467 | info = dm_get_rq_mapinfo(clone); |
1da177e4 | 468 | |
466891f9 | 469 | r = map_io(m, clone, info, 1); |
f40c67f0 | 470 | if (r < 0) { |
466891f9 | 471 | clear_mapinfo(m, info); |
f40c67f0 KU |
472 | dm_kill_unmapped_request(clone, r); |
473 | } else if (r == DM_MAPIO_REMAPPED) | |
474 | dm_dispatch_request(clone); | |
475 | else if (r == DM_MAPIO_REQUEUE) { | |
466891f9 | 476 | clear_mapinfo(m, info); |
f40c67f0 KU |
477 | dm_requeue_unmapped_request(clone); |
478 | } | |
1da177e4 LT |
479 | } |
480 | } | |
481 | ||
c4028958 | 482 | static void process_queued_ios(struct work_struct *work) |
1da177e4 | 483 | { |
c4028958 DH |
484 | struct multipath *m = |
485 | container_of(work, struct multipath, process_queued_ios); | |
fb612642 | 486 | struct pgpath *pgpath = NULL; |
e54f77dd | 487 | unsigned must_queue = 1; |
1da177e4 LT |
488 | unsigned long flags; |
489 | ||
490 | spin_lock_irqsave(&m->lock, flags); | |
491 | ||
492 | if (!m->current_pgpath) | |
02ab823f | 493 | __choose_pgpath(m, 0); |
1da177e4 LT |
494 | |
495 | pgpath = m->current_pgpath; | |
496 | ||
c3cd4f6b AK |
497 | if ((pgpath && !m->queue_io) || |
498 | (!pgpath && !m->queue_if_no_path)) | |
499 | must_queue = 0; | |
1da177e4 | 500 | |
954a73d5 SKM |
501 | if (m->pg_init_required && !m->pg_init_in_progress && pgpath && |
502 | !m->pg_init_disabled) | |
fb612642 KU |
503 | __pg_init_all_paths(m); |
504 | ||
1da177e4 | 505 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
506 | if (!must_queue) |
507 | dispatch_queued_ios(m); | |
508 | } | |
509 | ||
510 | /* | |
511 | * An event is triggered whenever a path is taken out of use. | |
512 | * Includes path failure and PG bypass. | |
513 | */ | |
c4028958 | 514 | static void trigger_event(struct work_struct *work) |
1da177e4 | 515 | { |
c4028958 DH |
516 | struct multipath *m = |
517 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
518 | |
519 | dm_table_event(m->ti->table); | |
520 | } | |
521 | ||
522 | /*----------------------------------------------------------------- | |
523 | * Constructor/argument parsing: | |
524 | * <#multipath feature args> [<arg>]* | |
525 | * <#hw_handler args> [hw_handler [<arg>]*] | |
526 | * <#priority groups> | |
527 | * <initial priority group> | |
528 | * [<selector> <#selector args> [<arg>]* | |
529 | * <#paths> <#per-path selector args> | |
530 | * [<path> [<arg>]* ]+ ]+ | |
531 | *---------------------------------------------------------------*/ | |
498f0103 | 532 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
533 | struct dm_target *ti) |
534 | { | |
535 | int r; | |
536 | struct path_selector_type *pst; | |
537 | unsigned ps_argc; | |
538 | ||
498f0103 | 539 | static struct dm_arg _args[] = { |
72d94861 | 540 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
541 | }; |
542 | ||
498f0103 | 543 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 544 | if (!pst) { |
72d94861 | 545 | ti->error = "unknown path selector type"; |
1da177e4 LT |
546 | return -EINVAL; |
547 | } | |
548 | ||
498f0103 | 549 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
550 | if (r) { |
551 | dm_put_path_selector(pst); | |
1da177e4 | 552 | return -EINVAL; |
371b2e34 | 553 | } |
1da177e4 LT |
554 | |
555 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
556 | if (r) { | |
557 | dm_put_path_selector(pst); | |
72d94861 | 558 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
559 | return r; |
560 | } | |
561 | ||
562 | pg->ps.type = pst; | |
498f0103 | 563 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
564 | |
565 | return 0; | |
566 | } | |
567 | ||
498f0103 | 568 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
569 | struct dm_target *ti) |
570 | { | |
571 | int r; | |
572 | struct pgpath *p; | |
ae11b1b3 | 573 | struct multipath *m = ti->private; |
a58a935d MS |
574 | struct request_queue *q = NULL; |
575 | const char *attached_handler_name; | |
1da177e4 LT |
576 | |
577 | /* we need at least a path arg */ | |
578 | if (as->argc < 1) { | |
72d94861 | 579 | ti->error = "no device given"; |
01460f35 | 580 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
581 | } |
582 | ||
583 | p = alloc_pgpath(); | |
584 | if (!p) | |
01460f35 | 585 | return ERR_PTR(-ENOMEM); |
1da177e4 | 586 | |
498f0103 | 587 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 588 | &p->path.dev); |
1da177e4 | 589 | if (r) { |
72d94861 | 590 | ti->error = "error getting device"; |
1da177e4 LT |
591 | goto bad; |
592 | } | |
593 | ||
a58a935d MS |
594 | if (m->retain_attached_hw_handler || m->hw_handler_name) |
595 | q = bdev_get_queue(p->path.dev->bdev); | |
596 | ||
597 | if (m->retain_attached_hw_handler) { | |
598 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); | |
599 | if (attached_handler_name) { | |
600 | /* | |
601 | * Reset hw_handler_name to match the attached handler | |
602 | * and clear any hw_handler_params associated with the | |
603 | * ignored handler. | |
604 | * | |
605 | * NB. This modifies the table line to show the actual | |
606 | * handler instead of the original table passed in. | |
607 | */ | |
608 | kfree(m->hw_handler_name); | |
609 | m->hw_handler_name = attached_handler_name; | |
610 | ||
611 | kfree(m->hw_handler_params); | |
612 | m->hw_handler_params = NULL; | |
613 | } | |
614 | } | |
a0cf7ea9 | 615 | |
a58a935d MS |
616 | if (m->hw_handler_name) { |
617 | /* | |
618 | * Increments scsi_dh reference, even when using an | |
619 | * already-attached handler. | |
620 | */ | |
a0cf7ea9 HR |
621 | r = scsi_dh_attach(q, m->hw_handler_name); |
622 | if (r == -EBUSY) { | |
623 | /* | |
a58a935d | 624 | * Already attached to different hw_handler: |
a0cf7ea9 HR |
625 | * try to reattach with correct one. |
626 | */ | |
627 | scsi_dh_detach(q); | |
628 | r = scsi_dh_attach(q, m->hw_handler_name); | |
629 | } | |
630 | ||
ae11b1b3 | 631 | if (r < 0) { |
a0cf7ea9 | 632 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
633 | dm_put_device(ti, p->path.dev); |
634 | goto bad; | |
635 | } | |
2bfd2e13 CS |
636 | |
637 | if (m->hw_handler_params) { | |
638 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
639 | if (r < 0) { | |
640 | ti->error = "unable to set hardware " | |
641 | "handler parameters"; | |
642 | scsi_dh_detach(q); | |
643 | dm_put_device(ti, p->path.dev); | |
644 | goto bad; | |
645 | } | |
646 | } | |
ae11b1b3 HR |
647 | } |
648 | ||
1da177e4 LT |
649 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
650 | if (r) { | |
651 | dm_put_device(ti, p->path.dev); | |
652 | goto bad; | |
653 | } | |
654 | ||
655 | return p; | |
656 | ||
657 | bad: | |
658 | free_pgpath(p); | |
01460f35 | 659 | return ERR_PTR(r); |
1da177e4 LT |
660 | } |
661 | ||
498f0103 | 662 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 663 | struct multipath *m) |
1da177e4 | 664 | { |
498f0103 | 665 | static struct dm_arg _args[] = { |
72d94861 AK |
666 | {1, 1024, "invalid number of paths"}, |
667 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
668 | }; |
669 | ||
670 | int r; | |
498f0103 | 671 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 672 | struct priority_group *pg; |
28f16c20 | 673 | struct dm_target *ti = m->ti; |
1da177e4 LT |
674 | |
675 | if (as->argc < 2) { | |
676 | as->argc = 0; | |
01460f35 BM |
677 | ti->error = "not enough priority group arguments"; |
678 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
679 | } |
680 | ||
681 | pg = alloc_priority_group(); | |
682 | if (!pg) { | |
72d94861 | 683 | ti->error = "couldn't allocate priority group"; |
01460f35 | 684 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
685 | } |
686 | pg->m = m; | |
687 | ||
688 | r = parse_path_selector(as, pg, ti); | |
689 | if (r) | |
690 | goto bad; | |
691 | ||
692 | /* | |
693 | * read the paths | |
694 | */ | |
498f0103 | 695 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
696 | if (r) |
697 | goto bad; | |
698 | ||
498f0103 | 699 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
700 | if (r) |
701 | goto bad; | |
702 | ||
498f0103 | 703 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
704 | for (i = 0; i < pg->nr_pgpaths; i++) { |
705 | struct pgpath *pgpath; | |
498f0103 | 706 | struct dm_arg_set path_args; |
1da177e4 | 707 | |
498f0103 | 708 | if (as->argc < nr_args) { |
148acff6 | 709 | ti->error = "not enough path parameters"; |
6bbf79a1 | 710 | r = -EINVAL; |
1da177e4 | 711 | goto bad; |
148acff6 | 712 | } |
1da177e4 | 713 | |
498f0103 | 714 | path_args.argc = nr_args; |
1da177e4 LT |
715 | path_args.argv = as->argv; |
716 | ||
717 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
718 | if (IS_ERR(pgpath)) { |
719 | r = PTR_ERR(pgpath); | |
1da177e4 | 720 | goto bad; |
01460f35 | 721 | } |
1da177e4 LT |
722 | |
723 | pgpath->pg = pg; | |
724 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 725 | dm_consume_args(as, nr_args); |
1da177e4 LT |
726 | } |
727 | ||
728 | return pg; | |
729 | ||
730 | bad: | |
731 | free_priority_group(pg, ti); | |
01460f35 | 732 | return ERR_PTR(r); |
1da177e4 LT |
733 | } |
734 | ||
498f0103 | 735 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 736 | { |
1da177e4 | 737 | unsigned hw_argc; |
2bfd2e13 | 738 | int ret; |
28f16c20 | 739 | struct dm_target *ti = m->ti; |
1da177e4 | 740 | |
498f0103 | 741 | static struct dm_arg _args[] = { |
72d94861 | 742 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
743 | }; |
744 | ||
498f0103 | 745 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
746 | return -EINVAL; |
747 | ||
748 | if (!hw_argc) | |
749 | return 0; | |
750 | ||
498f0103 | 751 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
510193a2 MS |
752 | if (!try_then_request_module(scsi_dh_handler_exist(m->hw_handler_name), |
753 | "scsi_dh_%s", m->hw_handler_name)) { | |
72d94861 | 754 | ti->error = "unknown hardware handler type"; |
2bfd2e13 CS |
755 | ret = -EINVAL; |
756 | goto fail; | |
1da177e4 | 757 | } |
14e98c5c | 758 | |
2bfd2e13 CS |
759 | if (hw_argc > 1) { |
760 | char *p; | |
761 | int i, j, len = 4; | |
762 | ||
763 | for (i = 0; i <= hw_argc - 2; i++) | |
764 | len += strlen(as->argv[i]) + 1; | |
765 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
766 | if (!p) { | |
767 | ti->error = "memory allocation failed"; | |
768 | ret = -ENOMEM; | |
769 | goto fail; | |
770 | } | |
771 | j = sprintf(p, "%d", hw_argc - 1); | |
772 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
773 | j = sprintf(p, "%s", as->argv[i]); | |
774 | } | |
498f0103 | 775 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
776 | |
777 | return 0; | |
2bfd2e13 CS |
778 | fail: |
779 | kfree(m->hw_handler_name); | |
780 | m->hw_handler_name = NULL; | |
781 | return ret; | |
1da177e4 LT |
782 | } |
783 | ||
498f0103 | 784 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
785 | { |
786 | int r; | |
787 | unsigned argc; | |
28f16c20 | 788 | struct dm_target *ti = m->ti; |
498f0103 | 789 | const char *arg_name; |
1da177e4 | 790 | |
498f0103 | 791 | static struct dm_arg _args[] = { |
a58a935d | 792 | {0, 6, "invalid number of feature args"}, |
c9e45581 | 793 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 794 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
795 | }; |
796 | ||
498f0103 | 797 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
798 | if (r) |
799 | return -EINVAL; | |
800 | ||
801 | if (!argc) | |
802 | return 0; | |
803 | ||
c9e45581 | 804 | do { |
498f0103 | 805 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
806 | argc--; |
807 | ||
498f0103 | 808 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
c9e45581 DW |
809 | r = queue_if_no_path(m, 1, 0); |
810 | continue; | |
811 | } | |
812 | ||
a58a935d MS |
813 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
814 | m->retain_attached_hw_handler = 1; | |
815 | continue; | |
816 | } | |
817 | ||
498f0103 | 818 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 819 | (argc >= 1)) { |
498f0103 | 820 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
821 | argc--; |
822 | continue; | |
823 | } | |
824 | ||
498f0103 | 825 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 826 | (argc >= 1)) { |
498f0103 | 827 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
828 | argc--; |
829 | continue; | |
830 | } | |
831 | ||
1da177e4 | 832 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
833 | r = -EINVAL; |
834 | } while (argc && !r); | |
835 | ||
836 | return r; | |
1da177e4 LT |
837 | } |
838 | ||
839 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
840 | char **argv) | |
841 | { | |
498f0103 MS |
842 | /* target arguments */ |
843 | static struct dm_arg _args[] = { | |
a490a07a MS |
844 | {0, 1024, "invalid number of priority groups"}, |
845 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
846 | }; |
847 | ||
848 | int r; | |
849 | struct multipath *m; | |
498f0103 | 850 | struct dm_arg_set as; |
1da177e4 LT |
851 | unsigned pg_count = 0; |
852 | unsigned next_pg_num; | |
853 | ||
854 | as.argc = argc; | |
855 | as.argv = argv; | |
856 | ||
28f16c20 | 857 | m = alloc_multipath(ti); |
1da177e4 | 858 | if (!m) { |
72d94861 | 859 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
860 | return -EINVAL; |
861 | } | |
862 | ||
28f16c20 | 863 | r = parse_features(&as, m); |
1da177e4 LT |
864 | if (r) |
865 | goto bad; | |
866 | ||
28f16c20 | 867 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
868 | if (r) |
869 | goto bad; | |
870 | ||
498f0103 | 871 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
872 | if (r) |
873 | goto bad; | |
874 | ||
498f0103 | 875 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
876 | if (r) |
877 | goto bad; | |
878 | ||
a490a07a MS |
879 | if ((!m->nr_priority_groups && next_pg_num) || |
880 | (m->nr_priority_groups && !next_pg_num)) { | |
881 | ti->error = "invalid initial priority group"; | |
882 | r = -EINVAL; | |
883 | goto bad; | |
884 | } | |
885 | ||
1da177e4 LT |
886 | /* parse the priority groups */ |
887 | while (as.argc) { | |
888 | struct priority_group *pg; | |
889 | ||
28f16c20 | 890 | pg = parse_priority_group(&as, m); |
01460f35 BM |
891 | if (IS_ERR(pg)) { |
892 | r = PTR_ERR(pg); | |
1da177e4 LT |
893 | goto bad; |
894 | } | |
895 | ||
896 | m->nr_valid_paths += pg->nr_pgpaths; | |
897 | list_add_tail(&pg->list, &m->priority_groups); | |
898 | pg_count++; | |
899 | pg->pg_num = pg_count; | |
900 | if (!--next_pg_num) | |
901 | m->next_pg = pg; | |
902 | } | |
903 | ||
904 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 905 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
906 | r = -EINVAL; |
907 | goto bad; | |
908 | } | |
909 | ||
55a62eef AK |
910 | ti->num_flush_bios = 1; |
911 | ti->num_discard_bios = 1; | |
042bcef8 | 912 | ti->num_write_same_bios = 1; |
8627921f | 913 | |
1da177e4 LT |
914 | return 0; |
915 | ||
916 | bad: | |
917 | free_multipath(m); | |
918 | return r; | |
919 | } | |
920 | ||
2bded7bd KU |
921 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
922 | { | |
923 | DECLARE_WAITQUEUE(wait, current); | |
924 | unsigned long flags; | |
925 | ||
926 | add_wait_queue(&m->pg_init_wait, &wait); | |
927 | ||
928 | while (1) { | |
929 | set_current_state(TASK_UNINTERRUPTIBLE); | |
930 | ||
931 | spin_lock_irqsave(&m->lock, flags); | |
932 | if (!m->pg_init_in_progress) { | |
933 | spin_unlock_irqrestore(&m->lock, flags); | |
934 | break; | |
935 | } | |
936 | spin_unlock_irqrestore(&m->lock, flags); | |
937 | ||
938 | io_schedule(); | |
939 | } | |
940 | set_current_state(TASK_RUNNING); | |
941 | ||
942 | remove_wait_queue(&m->pg_init_wait, &wait); | |
943 | } | |
944 | ||
945 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 946 | { |
954a73d5 SKM |
947 | unsigned long flags; |
948 | ||
949 | spin_lock_irqsave(&m->lock, flags); | |
950 | m->pg_init_disabled = 1; | |
951 | spin_unlock_irqrestore(&m->lock, flags); | |
952 | ||
bab7cfc7 | 953 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 954 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 955 | flush_workqueue(kmultipathd); |
43829731 | 956 | flush_work(&m->trigger_event); |
954a73d5 SKM |
957 | |
958 | spin_lock_irqsave(&m->lock, flags); | |
959 | m->pg_init_disabled = 0; | |
960 | spin_unlock_irqrestore(&m->lock, flags); | |
6df400ab KU |
961 | } |
962 | ||
963 | static void multipath_dtr(struct dm_target *ti) | |
964 | { | |
965 | struct multipath *m = ti->private; | |
966 | ||
2bded7bd | 967 | flush_multipath_work(m); |
1da177e4 LT |
968 | free_multipath(m); |
969 | } | |
970 | ||
971 | /* | |
f40c67f0 | 972 | * Map cloned requests |
1da177e4 | 973 | */ |
f40c67f0 | 974 | static int multipath_map(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
975 | union map_info *map_context) |
976 | { | |
977 | int r; | |
1da177e4 LT |
978 | struct multipath *m = (struct multipath *) ti->private; |
979 | ||
466891f9 | 980 | if (set_mapinfo(m, map_context) < 0) |
f40c67f0 KU |
981 | /* ENOMEM, requeue */ |
982 | return DM_MAPIO_REQUEUE; | |
1da177e4 | 983 | |
f40c67f0 | 984 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; |
466891f9 | 985 | r = map_io(m, clone, map_context, 0); |
45e15720 | 986 | if (r < 0 || r == DM_MAPIO_REQUEUE) |
466891f9 | 987 | clear_mapinfo(m, map_context); |
1da177e4 LT |
988 | |
989 | return r; | |
990 | } | |
991 | ||
992 | /* | |
993 | * Take a path out of use. | |
994 | */ | |
995 | static int fail_path(struct pgpath *pgpath) | |
996 | { | |
997 | unsigned long flags; | |
998 | struct multipath *m = pgpath->pg->m; | |
999 | ||
1000 | spin_lock_irqsave(&m->lock, flags); | |
1001 | ||
6680073d | 1002 | if (!pgpath->is_active) |
1da177e4 LT |
1003 | goto out; |
1004 | ||
72d94861 | 1005 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1006 | |
1007 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
6680073d | 1008 | pgpath->is_active = 0; |
1da177e4 LT |
1009 | pgpath->fail_count++; |
1010 | ||
1011 | m->nr_valid_paths--; | |
1012 | ||
1013 | if (pgpath == m->current_pgpath) | |
1014 | m->current_pgpath = NULL; | |
1015 | ||
b15546f9 MA |
1016 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
1017 | pgpath->path.dev->name, m->nr_valid_paths); | |
1018 | ||
fe9cf30e | 1019 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1020 | |
1021 | out: | |
1022 | spin_unlock_irqrestore(&m->lock, flags); | |
1023 | ||
1024 | return 0; | |
1025 | } | |
1026 | ||
1027 | /* | |
1028 | * Reinstate a previously-failed path | |
1029 | */ | |
1030 | static int reinstate_path(struct pgpath *pgpath) | |
1031 | { | |
1032 | int r = 0; | |
1033 | unsigned long flags; | |
1034 | struct multipath *m = pgpath->pg->m; | |
1035 | ||
1036 | spin_lock_irqsave(&m->lock, flags); | |
1037 | ||
6680073d | 1038 | if (pgpath->is_active) |
1da177e4 LT |
1039 | goto out; |
1040 | ||
def052d2 | 1041 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
1042 | DMWARN("Reinstate path not supported by path selector %s", |
1043 | pgpath->pg->ps.type->name); | |
1044 | r = -EINVAL; | |
1045 | goto out; | |
1046 | } | |
1047 | ||
1048 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1049 | if (r) | |
1050 | goto out; | |
1051 | ||
6680073d | 1052 | pgpath->is_active = 1; |
1da177e4 | 1053 | |
e54f77dd CS |
1054 | if (!m->nr_valid_paths++ && m->queue_size) { |
1055 | m->current_pgpath = NULL; | |
c557308e | 1056 | queue_work(kmultipathd, &m->process_queued_ios); |
e54f77dd | 1057 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1058 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
e54f77dd CS |
1059 | m->pg_init_in_progress++; |
1060 | } | |
1da177e4 | 1061 | |
b15546f9 MA |
1062 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
1063 | pgpath->path.dev->name, m->nr_valid_paths); | |
1064 | ||
fe9cf30e | 1065 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1066 | |
1067 | out: | |
1068 | spin_unlock_irqrestore(&m->lock, flags); | |
1069 | ||
1070 | return r; | |
1071 | } | |
1072 | ||
1073 | /* | |
1074 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1075 | */ | |
1076 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1077 | action_fn action) | |
1078 | { | |
19040c0b | 1079 | int r = -EINVAL; |
1da177e4 LT |
1080 | struct pgpath *pgpath; |
1081 | struct priority_group *pg; | |
1082 | ||
1083 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1084 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1085 | if (pgpath->path.dev == dev) | |
1086 | r = action(pgpath); | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | return r; | |
1091 | } | |
1092 | ||
1093 | /* | |
1094 | * Temporarily try to avoid having to use the specified PG | |
1095 | */ | |
1096 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
1097 | int bypassed) | |
1098 | { | |
1099 | unsigned long flags; | |
1100 | ||
1101 | spin_lock_irqsave(&m->lock, flags); | |
1102 | ||
1103 | pg->bypassed = bypassed; | |
1104 | m->current_pgpath = NULL; | |
1105 | m->current_pg = NULL; | |
1106 | ||
1107 | spin_unlock_irqrestore(&m->lock, flags); | |
1108 | ||
fe9cf30e | 1109 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1110 | } |
1111 | ||
1112 | /* | |
1113 | * Switch to using the specified PG from the next I/O that gets mapped | |
1114 | */ | |
1115 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1116 | { | |
1117 | struct priority_group *pg; | |
1118 | unsigned pgnum; | |
1119 | unsigned long flags; | |
31998ef1 | 1120 | char dummy; |
1da177e4 | 1121 | |
31998ef1 | 1122 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1123 | (pgnum > m->nr_priority_groups)) { |
1124 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1125 | return -EINVAL; | |
1126 | } | |
1127 | ||
1128 | spin_lock_irqsave(&m->lock, flags); | |
1129 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1130 | pg->bypassed = 0; | |
1131 | if (--pgnum) | |
1132 | continue; | |
1133 | ||
1134 | m->current_pgpath = NULL; | |
1135 | m->current_pg = NULL; | |
1136 | m->next_pg = pg; | |
1137 | } | |
1138 | spin_unlock_irqrestore(&m->lock, flags); | |
1139 | ||
fe9cf30e | 1140 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1141 | return 0; |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * Set/clear bypassed status of a PG. | |
1146 | * PGs are numbered upwards from 1 in the order they were declared. | |
1147 | */ | |
1148 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
1149 | { | |
1150 | struct priority_group *pg; | |
1151 | unsigned pgnum; | |
31998ef1 | 1152 | char dummy; |
1da177e4 | 1153 | |
31998ef1 | 1154 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1155 | (pgnum > m->nr_priority_groups)) { |
1156 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1157 | return -EINVAL; | |
1158 | } | |
1159 | ||
1160 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1161 | if (!--pgnum) | |
1162 | break; | |
1163 | } | |
1164 | ||
1165 | bypass_pg(m, pg, bypassed); | |
1166 | return 0; | |
1167 | } | |
1168 | ||
c9e45581 DW |
1169 | /* |
1170 | * Should we retry pg_init immediately? | |
1171 | */ | |
1172 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1173 | { | |
1174 | unsigned long flags; | |
1175 | int limit_reached = 0; | |
1176 | ||
1177 | spin_lock_irqsave(&m->lock, flags); | |
1178 | ||
954a73d5 | 1179 | if (m->pg_init_count <= m->pg_init_retries && !m->pg_init_disabled) |
c9e45581 DW |
1180 | m->pg_init_required = 1; |
1181 | else | |
1182 | limit_reached = 1; | |
1183 | ||
1184 | spin_unlock_irqrestore(&m->lock, flags); | |
1185 | ||
1186 | return limit_reached; | |
1187 | } | |
1188 | ||
3ae31f6a | 1189 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1190 | { |
83c0d5d5 | 1191 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1192 | struct priority_group *pg = pgpath->pg; |
1193 | struct multipath *m = pg->m; | |
1194 | unsigned long flags; | |
4e2d19e4 | 1195 | unsigned delay_retry = 0; |
cfae5c9b CS |
1196 | |
1197 | /* device or driver problems */ | |
1198 | switch (errors) { | |
1199 | case SCSI_DH_OK: | |
1200 | break; | |
1201 | case SCSI_DH_NOSYS: | |
1202 | if (!m->hw_handler_name) { | |
1203 | errors = 0; | |
1204 | break; | |
1205 | } | |
f7b934c8 MB |
1206 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1207 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1208 | /* |
1209 | * Fail path for now, so we do not ping pong | |
1210 | */ | |
1211 | fail_path(pgpath); | |
1212 | break; | |
1213 | case SCSI_DH_DEV_TEMP_BUSY: | |
1214 | /* | |
1215 | * Probably doing something like FW upgrade on the | |
1216 | * controller so try the other pg. | |
1217 | */ | |
1218 | bypass_pg(m, pg, 1); | |
1219 | break; | |
cfae5c9b | 1220 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1221 | /* Wait before retrying. */ |
1222 | delay_retry = 1; | |
cfae5c9b CS |
1223 | case SCSI_DH_IMM_RETRY: |
1224 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1225 | if (pg_init_limit_reached(m, pgpath)) | |
1226 | fail_path(pgpath); | |
1227 | errors = 0; | |
1228 | break; | |
1229 | default: | |
1230 | /* | |
1231 | * We probably do not want to fail the path for a device | |
1232 | * error, but this is what the old dm did. In future | |
1233 | * patches we can do more advanced handling. | |
1234 | */ | |
1235 | fail_path(pgpath); | |
1236 | } | |
1237 | ||
1238 | spin_lock_irqsave(&m->lock, flags); | |
1239 | if (errors) { | |
e54f77dd CS |
1240 | if (pgpath == m->current_pgpath) { |
1241 | DMERR("Could not failover device. Error %d.", errors); | |
1242 | m->current_pgpath = NULL; | |
1243 | m->current_pg = NULL; | |
1244 | } | |
d0259bf0 | 1245 | } else if (!m->pg_init_required) |
cfae5c9b | 1246 | pg->bypassed = 0; |
cfae5c9b | 1247 | |
d0259bf0 KU |
1248 | if (--m->pg_init_in_progress) |
1249 | /* Activations of other paths are still on going */ | |
1250 | goto out; | |
1251 | ||
1252 | if (!m->pg_init_required) | |
1253 | m->queue_io = 0; | |
1254 | ||
4e2d19e4 | 1255 | m->pg_init_delay_retry = delay_retry; |
d0259bf0 KU |
1256 | queue_work(kmultipathd, &m->process_queued_ios); |
1257 | ||
2bded7bd KU |
1258 | /* |
1259 | * Wake up any thread waiting to suspend. | |
1260 | */ | |
1261 | wake_up(&m->pg_init_wait); | |
1262 | ||
d0259bf0 | 1263 | out: |
cfae5c9b CS |
1264 | spin_unlock_irqrestore(&m->lock, flags); |
1265 | } | |
1266 | ||
bab7cfc7 CS |
1267 | static void activate_path(struct work_struct *work) |
1268 | { | |
e54f77dd | 1269 | struct pgpath *pgpath = |
4e2d19e4 | 1270 | container_of(work, struct pgpath, activate_path.work); |
bab7cfc7 | 1271 | |
3ae31f6a | 1272 | scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev), |
83c0d5d5 | 1273 | pg_init_done, pgpath); |
bab7cfc7 CS |
1274 | } |
1275 | ||
7e782af5 HR |
1276 | static int noretry_error(int error) |
1277 | { | |
1278 | switch (error) { | |
1279 | case -EOPNOTSUPP: | |
1280 | case -EREMOTEIO: | |
1281 | case -EILSEQ: | |
1282 | case -ENODATA: | |
cc9d3c38 | 1283 | case -ENOSPC: |
7e782af5 HR |
1284 | return 1; |
1285 | } | |
1286 | ||
1287 | /* Anything else could be a path failure, so should be retried */ | |
1288 | return 0; | |
1289 | } | |
1290 | ||
1da177e4 LT |
1291 | /* |
1292 | * end_io handling | |
1293 | */ | |
f40c67f0 | 1294 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1295 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1296 | { |
f40c67f0 KU |
1297 | /* |
1298 | * We don't queue any clone request inside the multipath target | |
1299 | * during end I/O handling, since those clone requests don't have | |
1300 | * bio clones. If we queue them inside the multipath target, | |
1301 | * we need to make bio clones, that requires memory allocation. | |
1302 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1303 | * don't have bio clones.) | |
1304 | * Instead of queueing the clone request here, we queue the original | |
1305 | * request into dm core, which will remake a clone request and | |
1306 | * clone bios for it and resubmit it later. | |
1307 | */ | |
1308 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1309 | unsigned long flags; |
1da177e4 | 1310 | |
f40c67f0 | 1311 | if (!error && !clone->errors) |
1da177e4 LT |
1312 | return 0; /* I/O complete */ |
1313 | ||
f84cb8a4 MS |
1314 | if (noretry_error(error)) { |
1315 | if ((clone->cmd_flags & REQ_WRITE_SAME) && | |
1316 | !clone->q->limits.max_write_same_sectors) { | |
1317 | struct queue_limits *limits; | |
1318 | ||
1319 | /* device doesn't really support WRITE SAME, disable it */ | |
1320 | limits = dm_get_queue_limits(dm_table_get_md(m->ti->table)); | |
1321 | limits->max_write_same_sectors = 0; | |
1322 | } | |
959eb4e5 | 1323 | return error; |
f84cb8a4 | 1324 | } |
959eb4e5 | 1325 | |
cfae5c9b CS |
1326 | if (mpio->pgpath) |
1327 | fail_path(mpio->pgpath); | |
1da177e4 | 1328 | |
640eb3b0 | 1329 | spin_lock_irqsave(&m->lock, flags); |
751b2a7d HR |
1330 | if (!m->nr_valid_paths) { |
1331 | if (!m->queue_if_no_path) { | |
1332 | if (!__must_push_back(m)) | |
1333 | r = -EIO; | |
1334 | } else { | |
1335 | if (error == -EBADE) | |
1336 | r = error; | |
1337 | } | |
1338 | } | |
640eb3b0 | 1339 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1340 | |
f40c67f0 | 1341 | return r; |
1da177e4 LT |
1342 | } |
1343 | ||
f40c67f0 | 1344 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1345 | int error, union map_info *map_context) |
1346 | { | |
028867ac AK |
1347 | struct multipath *m = ti->private; |
1348 | struct dm_mpath_io *mpio = map_context->ptr; | |
a71a261f | 1349 | struct pgpath *pgpath; |
1da177e4 LT |
1350 | struct path_selector *ps; |
1351 | int r; | |
1352 | ||
466891f9 JN |
1353 | BUG_ON(!mpio); |
1354 | ||
f40c67f0 | 1355 | r = do_end_io(m, clone, error, mpio); |
a71a261f | 1356 | pgpath = mpio->pgpath; |
1da177e4 LT |
1357 | if (pgpath) { |
1358 | ps = &pgpath->pg->ps; | |
1359 | if (ps->type->end_io) | |
02ab823f | 1360 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1361 | } |
466891f9 | 1362 | clear_mapinfo(m, map_context); |
1da177e4 LT |
1363 | |
1364 | return r; | |
1365 | } | |
1366 | ||
1367 | /* | |
1368 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1369 | * the last path fails we must error any remaining I/O. |
1370 | * Note that if the freeze_bdev fails while suspending, the | |
1371 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1372 | */ |
1373 | static void multipath_presuspend(struct dm_target *ti) | |
1374 | { | |
1375 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1376 | |
485ef69e | 1377 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1378 | } |
1379 | ||
6df400ab KU |
1380 | static void multipath_postsuspend(struct dm_target *ti) |
1381 | { | |
6380f26f MA |
1382 | struct multipath *m = ti->private; |
1383 | ||
1384 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1385 | flush_multipath_work(m); |
6380f26f | 1386 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1387 | } |
1388 | ||
436d4108 AK |
1389 | /* |
1390 | * Restore the queue_if_no_path setting. | |
1391 | */ | |
1da177e4 LT |
1392 | static void multipath_resume(struct dm_target *ti) |
1393 | { | |
1394 | struct multipath *m = (struct multipath *) ti->private; | |
1395 | unsigned long flags; | |
1396 | ||
1397 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1398 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1399 | spin_unlock_irqrestore(&m->lock, flags); |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * Info output has the following format: | |
1404 | * num_multipath_feature_args [multipath_feature_args]* | |
1405 | * num_handler_status_args [handler_status_args]* | |
1406 | * num_groups init_group_number | |
1407 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1408 | * num_paths num_selector_args | |
1409 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1410 | * | |
1411 | * Table output has the following format (identical to the constructor string): | |
1412 | * num_feature_args [features_args]* | |
1413 | * num_handler_args hw_handler [hw_handler_args]* | |
1414 | * num_groups init_group_number | |
1415 | * [priority selector-name num_ps_args [ps_args]* | |
1416 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1417 | */ | |
fd7c092e MP |
1418 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1419 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1420 | { |
1421 | int sz = 0; | |
1422 | unsigned long flags; | |
1423 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1424 | struct priority_group *pg; |
1425 | struct pgpath *p; | |
1426 | unsigned pg_num; | |
1427 | char state; | |
1428 | ||
1429 | spin_lock_irqsave(&m->lock, flags); | |
1430 | ||
1431 | /* Features */ | |
1432 | if (type == STATUSTYPE_INFO) | |
c9e45581 DW |
1433 | DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count); |
1434 | else { | |
1435 | DMEMIT("%u ", m->queue_if_no_path + | |
4e2d19e4 | 1436 | (m->pg_init_retries > 0) * 2 + |
a58a935d MS |
1437 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
1438 | m->retain_attached_hw_handler); | |
c9e45581 DW |
1439 | if (m->queue_if_no_path) |
1440 | DMEMIT("queue_if_no_path "); | |
1441 | if (m->pg_init_retries) | |
1442 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1443 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1444 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
a58a935d MS |
1445 | if (m->retain_attached_hw_handler) |
1446 | DMEMIT("retain_attached_hw_handler "); | |
c9e45581 | 1447 | } |
1da177e4 | 1448 | |
cfae5c9b | 1449 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1450 | DMEMIT("0 "); |
1451 | else | |
cfae5c9b | 1452 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1453 | |
1454 | DMEMIT("%u ", m->nr_priority_groups); | |
1455 | ||
1456 | if (m->next_pg) | |
1457 | pg_num = m->next_pg->pg_num; | |
1458 | else if (m->current_pg) | |
1459 | pg_num = m->current_pg->pg_num; | |
1460 | else | |
a490a07a | 1461 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1462 | |
1463 | DMEMIT("%u ", pg_num); | |
1464 | ||
1465 | switch (type) { | |
1466 | case STATUSTYPE_INFO: | |
1467 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1468 | if (pg->bypassed) | |
1469 | state = 'D'; /* Disabled */ | |
1470 | else if (pg == m->current_pg) | |
1471 | state = 'A'; /* Currently Active */ | |
1472 | else | |
1473 | state = 'E'; /* Enabled */ | |
1474 | ||
1475 | DMEMIT("%c ", state); | |
1476 | ||
1477 | if (pg->ps.type->status) | |
1478 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1479 | result + sz, | |
1480 | maxlen - sz); | |
1481 | else | |
1482 | DMEMIT("0 "); | |
1483 | ||
1484 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1485 | pg->ps.type->info_args); | |
1486 | ||
1487 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1488 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1489 | p->is_active ? "A" : "F", |
1da177e4 LT |
1490 | p->fail_count); |
1491 | if (pg->ps.type->status) | |
1492 | sz += pg->ps.type->status(&pg->ps, | |
1493 | &p->path, type, result + sz, | |
1494 | maxlen - sz); | |
1495 | } | |
1496 | } | |
1497 | break; | |
1498 | ||
1499 | case STATUSTYPE_TABLE: | |
1500 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1501 | DMEMIT("%s ", pg->ps.type->name); | |
1502 | ||
1503 | if (pg->ps.type->status) | |
1504 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1505 | result + sz, | |
1506 | maxlen - sz); | |
1507 | else | |
1508 | DMEMIT("0 "); | |
1509 | ||
1510 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1511 | pg->ps.type->table_args); | |
1512 | ||
1513 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1514 | DMEMIT("%s ", p->path.dev->name); | |
1515 | if (pg->ps.type->status) | |
1516 | sz += pg->ps.type->status(&pg->ps, | |
1517 | &p->path, type, result + sz, | |
1518 | maxlen - sz); | |
1519 | } | |
1520 | } | |
1521 | break; | |
1522 | } | |
1523 | ||
1524 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1525 | } |
1526 | ||
1527 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1528 | { | |
6380f26f | 1529 | int r = -EINVAL; |
1da177e4 LT |
1530 | struct dm_dev *dev; |
1531 | struct multipath *m = (struct multipath *) ti->private; | |
1532 | action_fn action; | |
1533 | ||
6380f26f MA |
1534 | mutex_lock(&m->work_mutex); |
1535 | ||
c2f3d24b KU |
1536 | if (dm_suspended(ti)) { |
1537 | r = -EBUSY; | |
1538 | goto out; | |
1539 | } | |
1540 | ||
1da177e4 | 1541 | if (argc == 1) { |
498f0103 | 1542 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
6380f26f MA |
1543 | r = queue_if_no_path(m, 1, 0); |
1544 | goto out; | |
498f0103 | 1545 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
6380f26f MA |
1546 | r = queue_if_no_path(m, 0, 0); |
1547 | goto out; | |
1548 | } | |
1da177e4 LT |
1549 | } |
1550 | ||
6380f26f MA |
1551 | if (argc != 2) { |
1552 | DMWARN("Unrecognised multipath message received."); | |
1553 | goto out; | |
1554 | } | |
1da177e4 | 1555 | |
498f0103 | 1556 | if (!strcasecmp(argv[0], "disable_group")) { |
6380f26f MA |
1557 | r = bypass_pg_num(m, argv[1], 1); |
1558 | goto out; | |
498f0103 | 1559 | } else if (!strcasecmp(argv[0], "enable_group")) { |
6380f26f MA |
1560 | r = bypass_pg_num(m, argv[1], 0); |
1561 | goto out; | |
498f0103 | 1562 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1563 | r = switch_pg_num(m, argv[1]); |
1564 | goto out; | |
498f0103 | 1565 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1566 | action = reinstate_path; |
498f0103 | 1567 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1568 | action = fail_path; |
6380f26f MA |
1569 | else { |
1570 | DMWARN("Unrecognised multipath message received."); | |
1571 | goto out; | |
1572 | } | |
1da177e4 | 1573 | |
8215d6ec | 1574 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1575 | if (r) { |
72d94861 | 1576 | DMWARN("message: error getting device %s", |
1da177e4 | 1577 | argv[1]); |
6380f26f | 1578 | goto out; |
1da177e4 LT |
1579 | } |
1580 | ||
1581 | r = action_dev(m, dev, action); | |
1582 | ||
1583 | dm_put_device(ti, dev); | |
1584 | ||
6380f26f MA |
1585 | out: |
1586 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1587 | return r; |
1da177e4 LT |
1588 | } |
1589 | ||
647b3d00 | 1590 | static int multipath_ioctl(struct dm_target *ti, unsigned int cmd, |
9af4aa30 MB |
1591 | unsigned long arg) |
1592 | { | |
35991652 | 1593 | struct multipath *m = ti->private; |
7ba10aa6 | 1594 | struct pgpath *pgpath; |
35991652 MP |
1595 | struct block_device *bdev; |
1596 | fmode_t mode; | |
9af4aa30 | 1597 | unsigned long flags; |
35991652 MP |
1598 | int r; |
1599 | ||
35991652 MP |
1600 | bdev = NULL; |
1601 | mode = 0; | |
1602 | r = 0; | |
9af4aa30 MB |
1603 | |
1604 | spin_lock_irqsave(&m->lock, flags); | |
1605 | ||
1606 | if (!m->current_pgpath) | |
02ab823f | 1607 | __choose_pgpath(m, 0); |
9af4aa30 | 1608 | |
7ba10aa6 MS |
1609 | pgpath = m->current_pgpath; |
1610 | ||
1611 | if (pgpath) { | |
1612 | bdev = pgpath->path.dev->bdev; | |
1613 | mode = pgpath->path.dev->mode; | |
e90dae1f | 1614 | } |
9af4aa30 | 1615 | |
7ba10aa6 | 1616 | if ((pgpath && m->queue_io) || (!pgpath && m->queue_if_no_path)) |
6c182cd8 | 1617 | r = -ENOTCONN; |
9af4aa30 MB |
1618 | else if (!bdev) |
1619 | r = -EIO; | |
1620 | ||
1621 | spin_unlock_irqrestore(&m->lock, flags); | |
1622 | ||
ec8013be PB |
1623 | /* |
1624 | * Only pass ioctls through if the device sizes match exactly. | |
1625 | */ | |
1626 | if (!r && ti->len != i_size_read(bdev->bd_inode) >> SECTOR_SHIFT) | |
1627 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
1628 | ||
6c182cd8 | 1629 | if (r == -ENOTCONN && !fatal_signal_pending(current)) |
35991652 | 1630 | queue_work(kmultipathd, &m->process_queued_ios); |
35991652 | 1631 | |
633a08b8 | 1632 | return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
9af4aa30 MB |
1633 | } |
1634 | ||
af4874e0 MS |
1635 | static int multipath_iterate_devices(struct dm_target *ti, |
1636 | iterate_devices_callout_fn fn, void *data) | |
1637 | { | |
1638 | struct multipath *m = ti->private; | |
1639 | struct priority_group *pg; | |
1640 | struct pgpath *p; | |
1641 | int ret = 0; | |
1642 | ||
1643 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1644 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1645 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1646 | if (ret) |
1647 | goto out; | |
1648 | } | |
1649 | } | |
1650 | ||
1651 | out: | |
1652 | return ret; | |
1653 | } | |
1654 | ||
f40c67f0 KU |
1655 | static int __pgpath_busy(struct pgpath *pgpath) |
1656 | { | |
1657 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1658 | ||
1659 | return dm_underlying_device_busy(q); | |
1660 | } | |
1661 | ||
1662 | /* | |
1663 | * We return "busy", only when we can map I/Os but underlying devices | |
1664 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1665 | * the underlying queue). | |
1666 | * In other words, if we want to kill I/Os or queue them inside us | |
1667 | * due to map unavailability, we don't return "busy". Otherwise, | |
1668 | * dm core won't give us the I/Os and we can't do what we want. | |
1669 | */ | |
1670 | static int multipath_busy(struct dm_target *ti) | |
1671 | { | |
1672 | int busy = 0, has_active = 0; | |
1673 | struct multipath *m = ti->private; | |
1674 | struct priority_group *pg; | |
1675 | struct pgpath *pgpath; | |
1676 | unsigned long flags; | |
1677 | ||
1678 | spin_lock_irqsave(&m->lock, flags); | |
1679 | ||
1680 | /* Guess which priority_group will be used at next mapping time */ | |
1681 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1682 | pg = m->next_pg; | |
1683 | else if (likely(m->current_pg)) | |
1684 | pg = m->current_pg; | |
1685 | else | |
1686 | /* | |
1687 | * We don't know which pg will be used at next mapping time. | |
1688 | * We don't call __choose_pgpath() here to avoid to trigger | |
1689 | * pg_init just by busy checking. | |
1690 | * So we don't know whether underlying devices we will be using | |
1691 | * at next mapping time are busy or not. Just try mapping. | |
1692 | */ | |
1693 | goto out; | |
1694 | ||
1695 | /* | |
1696 | * If there is one non-busy active path at least, the path selector | |
1697 | * will be able to select it. So we consider such a pg as not busy. | |
1698 | */ | |
1699 | busy = 1; | |
1700 | list_for_each_entry(pgpath, &pg->pgpaths, list) | |
1701 | if (pgpath->is_active) { | |
1702 | has_active = 1; | |
1703 | ||
1704 | if (!__pgpath_busy(pgpath)) { | |
1705 | busy = 0; | |
1706 | break; | |
1707 | } | |
1708 | } | |
1709 | ||
1710 | if (!has_active) | |
1711 | /* | |
1712 | * No active path in this pg, so this pg won't be used and | |
1713 | * the current_pg will be changed at next mapping time. | |
1714 | * We need to try mapping to determine it. | |
1715 | */ | |
1716 | busy = 0; | |
1717 | ||
1718 | out: | |
1719 | spin_unlock_irqrestore(&m->lock, flags); | |
1720 | ||
1721 | return busy; | |
1722 | } | |
1723 | ||
1da177e4 LT |
1724 | /*----------------------------------------------------------------- |
1725 | * Module setup | |
1726 | *---------------------------------------------------------------*/ | |
1727 | static struct target_type multipath_target = { | |
1728 | .name = "multipath", | |
954a73d5 | 1729 | .version = {1, 6, 0}, |
1da177e4 LT |
1730 | .module = THIS_MODULE, |
1731 | .ctr = multipath_ctr, | |
1732 | .dtr = multipath_dtr, | |
f40c67f0 KU |
1733 | .map_rq = multipath_map, |
1734 | .rq_end_io = multipath_end_io, | |
1da177e4 | 1735 | .presuspend = multipath_presuspend, |
6df400ab | 1736 | .postsuspend = multipath_postsuspend, |
1da177e4 LT |
1737 | .resume = multipath_resume, |
1738 | .status = multipath_status, | |
1739 | .message = multipath_message, | |
9af4aa30 | 1740 | .ioctl = multipath_ioctl, |
af4874e0 | 1741 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1742 | .busy = multipath_busy, |
1da177e4 LT |
1743 | }; |
1744 | ||
1745 | static int __init dm_multipath_init(void) | |
1746 | { | |
1747 | int r; | |
1748 | ||
1749 | /* allocate a slab for the dm_ios */ | |
028867ac | 1750 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1751 | if (!_mpio_cache) |
1752 | return -ENOMEM; | |
1753 | ||
1754 | r = dm_register_target(&multipath_target); | |
1755 | if (r < 0) { | |
0cd33124 | 1756 | DMERR("register failed %d", r); |
1da177e4 LT |
1757 | kmem_cache_destroy(_mpio_cache); |
1758 | return -EINVAL; | |
1759 | } | |
1760 | ||
4d4d66ab | 1761 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1762 | if (!kmultipathd) { |
0cd33124 | 1763 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1764 | dm_unregister_target(&multipath_target); |
1765 | kmem_cache_destroy(_mpio_cache); | |
1766 | return -ENOMEM; | |
1767 | } | |
1768 | ||
bab7cfc7 CS |
1769 | /* |
1770 | * A separate workqueue is used to handle the device handlers | |
1771 | * to avoid overloading existing workqueue. Overloading the | |
1772 | * old workqueue would also create a bottleneck in the | |
1773 | * path of the storage hardware device activation. | |
1774 | */ | |
4d4d66ab TH |
1775 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1776 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1777 | if (!kmpath_handlerd) { |
1778 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1779 | destroy_workqueue(kmultipathd); | |
1780 | dm_unregister_target(&multipath_target); | |
1781 | kmem_cache_destroy(_mpio_cache); | |
1782 | return -ENOMEM; | |
1783 | } | |
1784 | ||
72d94861 | 1785 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1786 | multipath_target.version[0], multipath_target.version[1], |
1787 | multipath_target.version[2]); | |
1788 | ||
1789 | return r; | |
1790 | } | |
1791 | ||
1792 | static void __exit dm_multipath_exit(void) | |
1793 | { | |
bab7cfc7 | 1794 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1795 | destroy_workqueue(kmultipathd); |
1796 | ||
10d3bd09 | 1797 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1798 | kmem_cache_destroy(_mpio_cache); |
1799 | } | |
1800 | ||
1da177e4 LT |
1801 | module_init(dm_multipath_init); |
1802 | module_exit(dm_multipath_exit); | |
1803 | ||
1804 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1805 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1806 | MODULE_LICENSE("GPL"); |