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 | 13 | |
e5863d9a | 14 | #include <linux/blkdev.h> |
1da177e4 LT |
15 | #include <linux/ctype.h> |
16 | #include <linux/init.h> | |
17 | #include <linux/mempool.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/time.h> | |
22 | #include <linux/workqueue.h> | |
35991652 | 23 | #include <linux/delay.h> |
cfae5c9b | 24 | #include <scsi/scsi_dh.h> |
60063497 | 25 | #include <linux/atomic.h> |
1da177e4 | 26 | |
72d94861 | 27 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
28 | #define DM_PG_INIT_DELAY_MSECS 2000 |
29 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
1da177e4 LT |
30 | |
31 | /* Path properties */ | |
32 | struct pgpath { | |
33 | struct list_head list; | |
34 | ||
35 | struct priority_group *pg; /* Owning PG */ | |
6680073d | 36 | unsigned is_active; /* Path status */ |
1da177e4 LT |
37 | unsigned fail_count; /* Cumulative failure count */ |
38 | ||
c922d5f7 | 39 | struct dm_path path; |
4e2d19e4 | 40 | struct delayed_work activate_path; |
1da177e4 LT |
41 | }; |
42 | ||
43 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
44 | ||
45 | /* | |
46 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
47 | * Each has a path selector which controls which path gets used. | |
48 | */ | |
49 | struct priority_group { | |
50 | struct list_head list; | |
51 | ||
52 | struct multipath *m; /* Owning multipath instance */ | |
53 | struct path_selector ps; | |
54 | ||
55 | unsigned pg_num; /* Reference number */ | |
56 | unsigned bypassed; /* Temporarily bypass this PG? */ | |
57 | ||
58 | unsigned nr_pgpaths; /* Number of paths in PG */ | |
59 | struct list_head pgpaths; | |
60 | }; | |
61 | ||
62 | /* Multipath context */ | |
63 | struct multipath { | |
64 | struct list_head list; | |
65 | struct dm_target *ti; | |
66 | ||
cfae5c9b | 67 | const char *hw_handler_name; |
2bfd2e13 | 68 | char *hw_handler_params; |
4e2d19e4 | 69 | |
1fbdd2b3 MS |
70 | spinlock_t lock; |
71 | ||
1da177e4 LT |
72 | unsigned nr_priority_groups; |
73 | struct list_head priority_groups; | |
4e2d19e4 CS |
74 | |
75 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
76 | ||
1da177e4 | 77 | unsigned pg_init_required; /* pg_init needs calling? */ |
c3cd4f6b | 78 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
4e2d19e4 | 79 | unsigned pg_init_delay_retry; /* Delay pg_init retry? */ |
1da177e4 LT |
80 | |
81 | unsigned nr_valid_paths; /* Total number of usable paths */ | |
82 | struct pgpath *current_pgpath; | |
83 | struct priority_group *current_pg; | |
84 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
85 | unsigned repeat_count; /* I/Os left before calling PS again */ | |
86 | ||
1fbdd2b3 MS |
87 | unsigned queue_io:1; /* Must we queue all I/O? */ |
88 | unsigned queue_if_no_path:1; /* Queue I/O if last path fails? */ | |
89 | unsigned saved_queue_if_no_path:1; /* Saved state during suspension */ | |
a58a935d | 90 | unsigned retain_attached_hw_handler:1; /* If there's already a hw_handler present, don't change it. */ |
954a73d5 | 91 | unsigned pg_init_disabled:1; /* pg_init is not currently allowed */ |
1fbdd2b3 | 92 | |
c9e45581 DW |
93 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
94 | unsigned pg_init_count; /* Number of times pg_init called */ | |
4e2d19e4 | 95 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
1da177e4 | 96 | |
1da177e4 LT |
97 | struct work_struct trigger_event; |
98 | ||
99 | /* | |
028867ac | 100 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
101 | * can resubmit bios on error. |
102 | */ | |
103 | mempool_t *mpio_pool; | |
6380f26f MA |
104 | |
105 | struct mutex work_mutex; | |
1da177e4 LT |
106 | }; |
107 | ||
108 | /* | |
109 | * Context information attached to each bio we process. | |
110 | */ | |
028867ac | 111 | struct dm_mpath_io { |
1da177e4 | 112 | struct pgpath *pgpath; |
02ab823f | 113 | size_t nr_bytes; |
1da177e4 LT |
114 | }; |
115 | ||
116 | typedef int (*action_fn) (struct pgpath *pgpath); | |
117 | ||
e18b890b | 118 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 119 | |
bab7cfc7 | 120 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 | 121 | static void trigger_event(struct work_struct *work); |
bab7cfc7 | 122 | static void activate_path(struct work_struct *work); |
e8099177 | 123 | static int __pgpath_busy(struct pgpath *pgpath); |
1da177e4 LT |
124 | |
125 | ||
126 | /*----------------------------------------------- | |
127 | * Allocation routines | |
128 | *-----------------------------------------------*/ | |
129 | ||
130 | static struct pgpath *alloc_pgpath(void) | |
131 | { | |
e69fae56 | 132 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 133 | |
224cb3e9 | 134 | if (pgpath) { |
6680073d | 135 | pgpath->is_active = 1; |
4e2d19e4 | 136 | INIT_DELAYED_WORK(&pgpath->activate_path, activate_path); |
224cb3e9 | 137 | } |
1da177e4 LT |
138 | |
139 | return pgpath; | |
140 | } | |
141 | ||
028867ac | 142 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
143 | { |
144 | kfree(pgpath); | |
145 | } | |
146 | ||
147 | static struct priority_group *alloc_priority_group(void) | |
148 | { | |
149 | struct priority_group *pg; | |
150 | ||
e69fae56 | 151 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 152 | |
e69fae56 MM |
153 | if (pg) |
154 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
155 | |
156 | return pg; | |
157 | } | |
158 | ||
159 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
160 | { | |
161 | struct pgpath *pgpath, *tmp; | |
162 | ||
163 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
164 | list_del(&pgpath->list); | |
165 | dm_put_device(ti, pgpath->path.dev); | |
166 | free_pgpath(pgpath); | |
167 | } | |
168 | } | |
169 | ||
170 | static void free_priority_group(struct priority_group *pg, | |
171 | struct dm_target *ti) | |
172 | { | |
173 | struct path_selector *ps = &pg->ps; | |
174 | ||
175 | if (ps->type) { | |
176 | ps->type->destroy(ps); | |
177 | dm_put_path_selector(ps->type); | |
178 | } | |
179 | ||
180 | free_pgpaths(&pg->pgpaths, ti); | |
181 | kfree(pg); | |
182 | } | |
183 | ||
28f16c20 | 184 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
185 | { |
186 | struct multipath *m; | |
f4790826 | 187 | unsigned min_ios = dm_get_reserved_rq_based_ios(); |
1da177e4 | 188 | |
e69fae56 | 189 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 190 | if (m) { |
1da177e4 LT |
191 | INIT_LIST_HEAD(&m->priority_groups); |
192 | spin_lock_init(&m->lock); | |
193 | m->queue_io = 1; | |
4e2d19e4 | 194 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; |
c4028958 | 195 | INIT_WORK(&m->trigger_event, trigger_event); |
2bded7bd | 196 | init_waitqueue_head(&m->pg_init_wait); |
6380f26f | 197 | mutex_init(&m->work_mutex); |
f4790826 | 198 | m->mpio_pool = mempool_create_slab_pool(min_ios, _mpio_cache); |
1da177e4 LT |
199 | if (!m->mpio_pool) { |
200 | kfree(m); | |
201 | return NULL; | |
202 | } | |
28f16c20 MM |
203 | m->ti = ti; |
204 | ti->private = m; | |
1da177e4 LT |
205 | } |
206 | ||
207 | return m; | |
208 | } | |
209 | ||
210 | static void free_multipath(struct multipath *m) | |
211 | { | |
212 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
213 | |
214 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
215 | list_del(&pg->list); | |
216 | free_priority_group(pg, m->ti); | |
217 | } | |
218 | ||
cfae5c9b | 219 | kfree(m->hw_handler_name); |
2bfd2e13 | 220 | kfree(m->hw_handler_params); |
1da177e4 LT |
221 | mempool_destroy(m->mpio_pool); |
222 | kfree(m); | |
223 | } | |
224 | ||
466891f9 JN |
225 | static int set_mapinfo(struct multipath *m, union map_info *info) |
226 | { | |
227 | struct dm_mpath_io *mpio; | |
228 | ||
229 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); | |
230 | if (!mpio) | |
231 | return -ENOMEM; | |
232 | ||
233 | memset(mpio, 0, sizeof(*mpio)); | |
234 | info->ptr = mpio; | |
235 | ||
236 | return 0; | |
237 | } | |
238 | ||
239 | static void clear_mapinfo(struct multipath *m, union map_info *info) | |
240 | { | |
241 | struct dm_mpath_io *mpio = info->ptr; | |
242 | ||
243 | info->ptr = NULL; | |
244 | mempool_free(mpio, m->mpio_pool); | |
245 | } | |
1da177e4 LT |
246 | |
247 | /*----------------------------------------------- | |
248 | * Path selection | |
249 | *-----------------------------------------------*/ | |
250 | ||
3e9f1be1 | 251 | static int __pg_init_all_paths(struct multipath *m) |
fb612642 KU |
252 | { |
253 | struct pgpath *pgpath; | |
4e2d19e4 | 254 | unsigned long pg_init_delay = 0; |
fb612642 | 255 | |
17f4ff45 | 256 | if (m->pg_init_in_progress || m->pg_init_disabled) |
3e9f1be1 | 257 | return 0; |
17f4ff45 | 258 | |
fb612642 KU |
259 | m->pg_init_count++; |
260 | m->pg_init_required = 0; | |
3e9f1be1 HR |
261 | |
262 | /* Check here to reset pg_init_required */ | |
263 | if (!m->current_pg) | |
264 | return 0; | |
265 | ||
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 | } | |
3e9f1be1 | 277 | return m->pg_init_in_progress; |
fb612642 KU |
278 | } |
279 | ||
1da177e4 LT |
280 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) |
281 | { | |
1da177e4 LT |
282 | m->current_pg = pgpath->pg; |
283 | ||
284 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 285 | if (m->hw_handler_name) { |
1da177e4 LT |
286 | m->pg_init_required = 1; |
287 | m->queue_io = 1; | |
288 | } else { | |
289 | m->pg_init_required = 0; | |
290 | m->queue_io = 0; | |
291 | } | |
c9e45581 DW |
292 | |
293 | m->pg_init_count = 0; | |
1da177e4 LT |
294 | } |
295 | ||
02ab823f KU |
296 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
297 | size_t nr_bytes) | |
1da177e4 | 298 | { |
c922d5f7 | 299 | struct dm_path *path; |
1da177e4 | 300 | |
02ab823f | 301 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes); |
1da177e4 LT |
302 | if (!path) |
303 | return -ENXIO; | |
304 | ||
305 | m->current_pgpath = path_to_pgpath(path); | |
306 | ||
307 | if (m->current_pg != pg) | |
308 | __switch_pg(m, m->current_pgpath); | |
309 | ||
310 | return 0; | |
311 | } | |
312 | ||
02ab823f | 313 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
314 | { |
315 | struct priority_group *pg; | |
316 | unsigned bypassed = 1; | |
317 | ||
1f271972 BM |
318 | if (!m->nr_valid_paths) { |
319 | m->queue_io = 0; | |
1da177e4 | 320 | goto failed; |
1f271972 | 321 | } |
1da177e4 LT |
322 | |
323 | /* Were we instructed to switch PG? */ | |
324 | if (m->next_pg) { | |
325 | pg = m->next_pg; | |
326 | m->next_pg = NULL; | |
02ab823f | 327 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
328 | return; |
329 | } | |
330 | ||
331 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 332 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
333 | return; |
334 | ||
335 | /* | |
336 | * Loop through priority groups until we find a valid path. | |
337 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
338 | * Second time we only try the ones we skipped, but set |
339 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
340 | */ |
341 | do { | |
342 | list_for_each_entry(pg, &m->priority_groups, list) { | |
343 | if (pg->bypassed == bypassed) | |
344 | continue; | |
f220fd4e MC |
345 | if (!__choose_path_in_pg(m, pg, nr_bytes)) { |
346 | if (!bypassed) | |
347 | m->pg_init_delay_retry = 1; | |
1da177e4 | 348 | return; |
f220fd4e | 349 | } |
1da177e4 LT |
350 | } |
351 | } while (bypassed--); | |
352 | ||
353 | failed: | |
354 | m->current_pgpath = NULL; | |
355 | m->current_pg = NULL; | |
356 | } | |
357 | ||
45e15720 KU |
358 | /* |
359 | * Check whether bios must be queued in the device-mapper core rather | |
360 | * than here in the target. | |
361 | * | |
362 | * m->lock must be held on entry. | |
363 | * | |
364 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
365 | * same value then we are not between multipath_presuspend() | |
366 | * and multipath_resume() calls and we have no need to check | |
367 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
368 | */ | |
369 | static int __must_push_back(struct multipath *m) | |
370 | { | |
e8099177 HR |
371 | return (m->queue_if_no_path || |
372 | (m->queue_if_no_path != m->saved_queue_if_no_path && | |
373 | dm_noflush_suspending(m->ti))); | |
45e15720 KU |
374 | } |
375 | ||
36fcffcc HR |
376 | /* |
377 | * Map cloned requests | |
378 | */ | |
e5863d9a MS |
379 | static int __multipath_map(struct dm_target *ti, struct request *clone, |
380 | union map_info *map_context, | |
381 | struct request *rq, struct request **__clone) | |
1da177e4 | 382 | { |
36fcffcc | 383 | struct multipath *m = (struct multipath *) ti->private; |
e3bde04f | 384 | int r = DM_MAPIO_REQUEUE; |
e5863d9a | 385 | size_t nr_bytes = clone ? blk_rq_bytes(clone) : blk_rq_bytes(rq); |
1da177e4 | 386 | struct pgpath *pgpath; |
f40c67f0 | 387 | struct block_device *bdev; |
e3bde04f | 388 | struct dm_mpath_io *mpio; |
1da177e4 | 389 | |
2eb6e1e3 | 390 | spin_lock_irq(&m->lock); |
1da177e4 LT |
391 | |
392 | /* Do we need to select a new pgpath? */ | |
393 | if (!m->current_pgpath || | |
394 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
02ab823f | 395 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
396 | |
397 | pgpath = m->current_pgpath; | |
398 | ||
9bf59a61 MS |
399 | if (!pgpath) { |
400 | if (!__must_push_back(m)) | |
401 | r = -EIO; /* Failed */ | |
402 | goto out_unlock; | |
6afbc01d | 403 | } else if (m->queue_io || m->pg_init_required) { |
e3bde04f | 404 | __pg_init_all_paths(m); |
9bf59a61 MS |
405 | goto out_unlock; |
406 | } | |
6afbc01d | 407 | |
9bf59a61 MS |
408 | if (set_mapinfo(m, map_context) < 0) |
409 | /* ENOMEM, requeue */ | |
410 | goto out_unlock; | |
411 | ||
2eb6e1e3 KB |
412 | mpio = map_context->ptr; |
413 | mpio->pgpath = pgpath; | |
414 | mpio->nr_bytes = nr_bytes; | |
415 | ||
9bf59a61 | 416 | bdev = pgpath->path.dev->bdev; |
2eb6e1e3 | 417 | |
2eb6e1e3 KB |
418 | spin_unlock_irq(&m->lock); |
419 | ||
e5863d9a | 420 | if (clone) { |
c5248f79 MS |
421 | /* |
422 | * Old request-based interface: allocated clone is passed in. | |
423 | * Used by: .request_fn stacked on .request_fn path(s). | |
424 | */ | |
e5863d9a MS |
425 | clone->q = bdev_get_queue(bdev); |
426 | clone->rq_disk = bdev->bd_disk; | |
427 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
428 | } else { | |
429 | /* blk-mq request-based interface */ | |
430 | *__clone = blk_get_request(bdev_get_queue(bdev), | |
02233342 | 431 | rq_data_dir(rq), GFP_ATOMIC); |
4c6dd53d | 432 | if (IS_ERR(*__clone)) { |
e5863d9a | 433 | /* ENOMEM, requeue */ |
4c6dd53d | 434 | clear_mapinfo(m, map_context); |
e5863d9a | 435 | return r; |
4c6dd53d | 436 | } |
e5863d9a MS |
437 | (*__clone)->bio = (*__clone)->biotail = NULL; |
438 | (*__clone)->rq_disk = bdev->bd_disk; | |
439 | (*__clone)->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
440 | } | |
441 | ||
9bf59a61 MS |
442 | if (pgpath->pg->ps.type->start_io) |
443 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
444 | &pgpath->path, | |
445 | nr_bytes); | |
2eb6e1e3 | 446 | return DM_MAPIO_REMAPPED; |
1da177e4 | 447 | |
e3bde04f | 448 | out_unlock: |
2eb6e1e3 | 449 | spin_unlock_irq(&m->lock); |
1da177e4 LT |
450 | |
451 | return r; | |
452 | } | |
453 | ||
e5863d9a MS |
454 | static int multipath_map(struct dm_target *ti, struct request *clone, |
455 | union map_info *map_context) | |
456 | { | |
457 | return __multipath_map(ti, clone, map_context, NULL, NULL); | |
458 | } | |
459 | ||
460 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, | |
461 | union map_info *map_context, | |
462 | struct request **clone) | |
463 | { | |
464 | return __multipath_map(ti, NULL, map_context, rq, clone); | |
465 | } | |
466 | ||
467 | static void multipath_release_clone(struct request *clone) | |
468 | { | |
469 | blk_put_request(clone); | |
470 | } | |
471 | ||
1da177e4 LT |
472 | /* |
473 | * If we run out of usable paths, should we queue I/O or error it? | |
474 | */ | |
485ef69e AK |
475 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
476 | unsigned save_old_value) | |
1da177e4 LT |
477 | { |
478 | unsigned long flags; | |
479 | ||
480 | spin_lock_irqsave(&m->lock, flags); | |
481 | ||
485ef69e AK |
482 | if (save_old_value) |
483 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
484 | else | |
485 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 486 | m->queue_if_no_path = queue_if_no_path; |
1da177e4 LT |
487 | spin_unlock_irqrestore(&m->lock, flags); |
488 | ||
63d832c3 HR |
489 | if (!queue_if_no_path) |
490 | dm_table_run_md_queue_async(m->ti->table); | |
491 | ||
1da177e4 LT |
492 | return 0; |
493 | } | |
494 | ||
1da177e4 LT |
495 | /* |
496 | * An event is triggered whenever a path is taken out of use. | |
497 | * Includes path failure and PG bypass. | |
498 | */ | |
c4028958 | 499 | static void trigger_event(struct work_struct *work) |
1da177e4 | 500 | { |
c4028958 DH |
501 | struct multipath *m = |
502 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
503 | |
504 | dm_table_event(m->ti->table); | |
505 | } | |
506 | ||
507 | /*----------------------------------------------------------------- | |
508 | * Constructor/argument parsing: | |
509 | * <#multipath feature args> [<arg>]* | |
510 | * <#hw_handler args> [hw_handler [<arg>]*] | |
511 | * <#priority groups> | |
512 | * <initial priority group> | |
513 | * [<selector> <#selector args> [<arg>]* | |
514 | * <#paths> <#per-path selector args> | |
515 | * [<path> [<arg>]* ]+ ]+ | |
516 | *---------------------------------------------------------------*/ | |
498f0103 | 517 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
518 | struct dm_target *ti) |
519 | { | |
520 | int r; | |
521 | struct path_selector_type *pst; | |
522 | unsigned ps_argc; | |
523 | ||
498f0103 | 524 | static struct dm_arg _args[] = { |
72d94861 | 525 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
526 | }; |
527 | ||
498f0103 | 528 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 529 | if (!pst) { |
72d94861 | 530 | ti->error = "unknown path selector type"; |
1da177e4 LT |
531 | return -EINVAL; |
532 | } | |
533 | ||
498f0103 | 534 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
535 | if (r) { |
536 | dm_put_path_selector(pst); | |
1da177e4 | 537 | return -EINVAL; |
371b2e34 | 538 | } |
1da177e4 LT |
539 | |
540 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
541 | if (r) { | |
542 | dm_put_path_selector(pst); | |
72d94861 | 543 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
544 | return r; |
545 | } | |
546 | ||
547 | pg->ps.type = pst; | |
498f0103 | 548 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
549 | |
550 | return 0; | |
551 | } | |
552 | ||
498f0103 | 553 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
554 | struct dm_target *ti) |
555 | { | |
556 | int r; | |
557 | struct pgpath *p; | |
ae11b1b3 | 558 | struct multipath *m = ti->private; |
a58a935d MS |
559 | struct request_queue *q = NULL; |
560 | const char *attached_handler_name; | |
1da177e4 LT |
561 | |
562 | /* we need at least a path arg */ | |
563 | if (as->argc < 1) { | |
72d94861 | 564 | ti->error = "no device given"; |
01460f35 | 565 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
566 | } |
567 | ||
568 | p = alloc_pgpath(); | |
569 | if (!p) | |
01460f35 | 570 | return ERR_PTR(-ENOMEM); |
1da177e4 | 571 | |
498f0103 | 572 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 573 | &p->path.dev); |
1da177e4 | 574 | if (r) { |
72d94861 | 575 | ti->error = "error getting device"; |
1da177e4 LT |
576 | goto bad; |
577 | } | |
578 | ||
a58a935d MS |
579 | if (m->retain_attached_hw_handler || m->hw_handler_name) |
580 | q = bdev_get_queue(p->path.dev->bdev); | |
581 | ||
582 | if (m->retain_attached_hw_handler) { | |
1bab0de0 | 583 | retain: |
a58a935d MS |
584 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); |
585 | if (attached_handler_name) { | |
586 | /* | |
587 | * Reset hw_handler_name to match the attached handler | |
588 | * and clear any hw_handler_params associated with the | |
589 | * ignored handler. | |
590 | * | |
591 | * NB. This modifies the table line to show the actual | |
592 | * handler instead of the original table passed in. | |
593 | */ | |
594 | kfree(m->hw_handler_name); | |
595 | m->hw_handler_name = attached_handler_name; | |
596 | ||
597 | kfree(m->hw_handler_params); | |
598 | m->hw_handler_params = NULL; | |
599 | } | |
600 | } | |
a0cf7ea9 | 601 | |
a58a935d | 602 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
603 | r = scsi_dh_attach(q, m->hw_handler_name); |
604 | if (r == -EBUSY) { | |
1bab0de0 | 605 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 606 | |
1bab0de0 CH |
607 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
608 | bdevname(p->path.dev->bdev, b)); | |
609 | goto retain; | |
610 | } | |
ae11b1b3 | 611 | if (r < 0) { |
a0cf7ea9 | 612 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
613 | dm_put_device(ti, p->path.dev); |
614 | goto bad; | |
615 | } | |
2bfd2e13 CS |
616 | |
617 | if (m->hw_handler_params) { | |
618 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
619 | if (r < 0) { | |
620 | ti->error = "unable to set hardware " | |
621 | "handler parameters"; | |
2bfd2e13 CS |
622 | dm_put_device(ti, p->path.dev); |
623 | goto bad; | |
624 | } | |
625 | } | |
ae11b1b3 HR |
626 | } |
627 | ||
1da177e4 LT |
628 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
629 | if (r) { | |
630 | dm_put_device(ti, p->path.dev); | |
631 | goto bad; | |
632 | } | |
633 | ||
634 | return p; | |
635 | ||
636 | bad: | |
637 | free_pgpath(p); | |
01460f35 | 638 | return ERR_PTR(r); |
1da177e4 LT |
639 | } |
640 | ||
498f0103 | 641 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 642 | struct multipath *m) |
1da177e4 | 643 | { |
498f0103 | 644 | static struct dm_arg _args[] = { |
72d94861 AK |
645 | {1, 1024, "invalid number of paths"}, |
646 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
647 | }; |
648 | ||
649 | int r; | |
498f0103 | 650 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 651 | struct priority_group *pg; |
28f16c20 | 652 | struct dm_target *ti = m->ti; |
1da177e4 LT |
653 | |
654 | if (as->argc < 2) { | |
655 | as->argc = 0; | |
01460f35 BM |
656 | ti->error = "not enough priority group arguments"; |
657 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
658 | } |
659 | ||
660 | pg = alloc_priority_group(); | |
661 | if (!pg) { | |
72d94861 | 662 | ti->error = "couldn't allocate priority group"; |
01460f35 | 663 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
664 | } |
665 | pg->m = m; | |
666 | ||
667 | r = parse_path_selector(as, pg, ti); | |
668 | if (r) | |
669 | goto bad; | |
670 | ||
671 | /* | |
672 | * read the paths | |
673 | */ | |
498f0103 | 674 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
675 | if (r) |
676 | goto bad; | |
677 | ||
498f0103 | 678 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
679 | if (r) |
680 | goto bad; | |
681 | ||
498f0103 | 682 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
683 | for (i = 0; i < pg->nr_pgpaths; i++) { |
684 | struct pgpath *pgpath; | |
498f0103 | 685 | struct dm_arg_set path_args; |
1da177e4 | 686 | |
498f0103 | 687 | if (as->argc < nr_args) { |
148acff6 | 688 | ti->error = "not enough path parameters"; |
6bbf79a1 | 689 | r = -EINVAL; |
1da177e4 | 690 | goto bad; |
148acff6 | 691 | } |
1da177e4 | 692 | |
498f0103 | 693 | path_args.argc = nr_args; |
1da177e4 LT |
694 | path_args.argv = as->argv; |
695 | ||
696 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
697 | if (IS_ERR(pgpath)) { |
698 | r = PTR_ERR(pgpath); | |
1da177e4 | 699 | goto bad; |
01460f35 | 700 | } |
1da177e4 LT |
701 | |
702 | pgpath->pg = pg; | |
703 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 704 | dm_consume_args(as, nr_args); |
1da177e4 LT |
705 | } |
706 | ||
707 | return pg; | |
708 | ||
709 | bad: | |
710 | free_priority_group(pg, ti); | |
01460f35 | 711 | return ERR_PTR(r); |
1da177e4 LT |
712 | } |
713 | ||
498f0103 | 714 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 715 | { |
1da177e4 | 716 | unsigned hw_argc; |
2bfd2e13 | 717 | int ret; |
28f16c20 | 718 | struct dm_target *ti = m->ti; |
1da177e4 | 719 | |
498f0103 | 720 | static struct dm_arg _args[] = { |
72d94861 | 721 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
722 | }; |
723 | ||
498f0103 | 724 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
725 | return -EINVAL; |
726 | ||
727 | if (!hw_argc) | |
728 | return 0; | |
729 | ||
498f0103 | 730 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
14e98c5c | 731 | |
2bfd2e13 CS |
732 | if (hw_argc > 1) { |
733 | char *p; | |
734 | int i, j, len = 4; | |
735 | ||
736 | for (i = 0; i <= hw_argc - 2; i++) | |
737 | len += strlen(as->argv[i]) + 1; | |
738 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
739 | if (!p) { | |
740 | ti->error = "memory allocation failed"; | |
741 | ret = -ENOMEM; | |
742 | goto fail; | |
743 | } | |
744 | j = sprintf(p, "%d", hw_argc - 1); | |
745 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
746 | j = sprintf(p, "%s", as->argv[i]); | |
747 | } | |
498f0103 | 748 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
749 | |
750 | return 0; | |
2bfd2e13 CS |
751 | fail: |
752 | kfree(m->hw_handler_name); | |
753 | m->hw_handler_name = NULL; | |
754 | return ret; | |
1da177e4 LT |
755 | } |
756 | ||
498f0103 | 757 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
758 | { |
759 | int r; | |
760 | unsigned argc; | |
28f16c20 | 761 | struct dm_target *ti = m->ti; |
498f0103 | 762 | const char *arg_name; |
1da177e4 | 763 | |
498f0103 | 764 | static struct dm_arg _args[] = { |
a58a935d | 765 | {0, 6, "invalid number of feature args"}, |
c9e45581 | 766 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 767 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
768 | }; |
769 | ||
498f0103 | 770 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
771 | if (r) |
772 | return -EINVAL; | |
773 | ||
774 | if (!argc) | |
775 | return 0; | |
776 | ||
c9e45581 | 777 | do { |
498f0103 | 778 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
779 | argc--; |
780 | ||
498f0103 | 781 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
c9e45581 DW |
782 | r = queue_if_no_path(m, 1, 0); |
783 | continue; | |
784 | } | |
785 | ||
a58a935d MS |
786 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
787 | m->retain_attached_hw_handler = 1; | |
788 | continue; | |
789 | } | |
790 | ||
498f0103 | 791 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 792 | (argc >= 1)) { |
498f0103 | 793 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
794 | argc--; |
795 | continue; | |
796 | } | |
797 | ||
498f0103 | 798 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 799 | (argc >= 1)) { |
498f0103 | 800 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
801 | argc--; |
802 | continue; | |
803 | } | |
804 | ||
1da177e4 | 805 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
806 | r = -EINVAL; |
807 | } while (argc && !r); | |
808 | ||
809 | return r; | |
1da177e4 LT |
810 | } |
811 | ||
812 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
813 | char **argv) | |
814 | { | |
498f0103 MS |
815 | /* target arguments */ |
816 | static struct dm_arg _args[] = { | |
a490a07a MS |
817 | {0, 1024, "invalid number of priority groups"}, |
818 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
819 | }; |
820 | ||
821 | int r; | |
822 | struct multipath *m; | |
498f0103 | 823 | struct dm_arg_set as; |
1da177e4 LT |
824 | unsigned pg_count = 0; |
825 | unsigned next_pg_num; | |
826 | ||
827 | as.argc = argc; | |
828 | as.argv = argv; | |
829 | ||
28f16c20 | 830 | m = alloc_multipath(ti); |
1da177e4 | 831 | if (!m) { |
72d94861 | 832 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
833 | return -EINVAL; |
834 | } | |
835 | ||
28f16c20 | 836 | r = parse_features(&as, m); |
1da177e4 LT |
837 | if (r) |
838 | goto bad; | |
839 | ||
28f16c20 | 840 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
841 | if (r) |
842 | goto bad; | |
843 | ||
498f0103 | 844 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
845 | if (r) |
846 | goto bad; | |
847 | ||
498f0103 | 848 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
849 | if (r) |
850 | goto bad; | |
851 | ||
a490a07a MS |
852 | if ((!m->nr_priority_groups && next_pg_num) || |
853 | (m->nr_priority_groups && !next_pg_num)) { | |
854 | ti->error = "invalid initial priority group"; | |
855 | r = -EINVAL; | |
856 | goto bad; | |
857 | } | |
858 | ||
1da177e4 LT |
859 | /* parse the priority groups */ |
860 | while (as.argc) { | |
861 | struct priority_group *pg; | |
862 | ||
28f16c20 | 863 | pg = parse_priority_group(&as, m); |
01460f35 BM |
864 | if (IS_ERR(pg)) { |
865 | r = PTR_ERR(pg); | |
1da177e4 LT |
866 | goto bad; |
867 | } | |
868 | ||
869 | m->nr_valid_paths += pg->nr_pgpaths; | |
870 | list_add_tail(&pg->list, &m->priority_groups); | |
871 | pg_count++; | |
872 | pg->pg_num = pg_count; | |
873 | if (!--next_pg_num) | |
874 | m->next_pg = pg; | |
875 | } | |
876 | ||
877 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 878 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
879 | r = -EINVAL; |
880 | goto bad; | |
881 | } | |
882 | ||
55a62eef AK |
883 | ti->num_flush_bios = 1; |
884 | ti->num_discard_bios = 1; | |
042bcef8 | 885 | ti->num_write_same_bios = 1; |
8627921f | 886 | |
1da177e4 LT |
887 | return 0; |
888 | ||
889 | bad: | |
890 | free_multipath(m); | |
891 | return r; | |
892 | } | |
893 | ||
2bded7bd KU |
894 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
895 | { | |
896 | DECLARE_WAITQUEUE(wait, current); | |
897 | unsigned long flags; | |
898 | ||
899 | add_wait_queue(&m->pg_init_wait, &wait); | |
900 | ||
901 | while (1) { | |
902 | set_current_state(TASK_UNINTERRUPTIBLE); | |
903 | ||
904 | spin_lock_irqsave(&m->lock, flags); | |
905 | if (!m->pg_init_in_progress) { | |
906 | spin_unlock_irqrestore(&m->lock, flags); | |
907 | break; | |
908 | } | |
909 | spin_unlock_irqrestore(&m->lock, flags); | |
910 | ||
911 | io_schedule(); | |
912 | } | |
913 | set_current_state(TASK_RUNNING); | |
914 | ||
915 | remove_wait_queue(&m->pg_init_wait, &wait); | |
916 | } | |
917 | ||
918 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 919 | { |
954a73d5 SKM |
920 | unsigned long flags; |
921 | ||
922 | spin_lock_irqsave(&m->lock, flags); | |
923 | m->pg_init_disabled = 1; | |
924 | spin_unlock_irqrestore(&m->lock, flags); | |
925 | ||
bab7cfc7 | 926 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 927 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 928 | flush_workqueue(kmultipathd); |
43829731 | 929 | flush_work(&m->trigger_event); |
954a73d5 SKM |
930 | |
931 | spin_lock_irqsave(&m->lock, flags); | |
932 | m->pg_init_disabled = 0; | |
933 | spin_unlock_irqrestore(&m->lock, flags); | |
6df400ab KU |
934 | } |
935 | ||
936 | static void multipath_dtr(struct dm_target *ti) | |
937 | { | |
938 | struct multipath *m = ti->private; | |
939 | ||
2bded7bd | 940 | flush_multipath_work(m); |
1da177e4 LT |
941 | free_multipath(m); |
942 | } | |
943 | ||
1da177e4 LT |
944 | /* |
945 | * Take a path out of use. | |
946 | */ | |
947 | static int fail_path(struct pgpath *pgpath) | |
948 | { | |
949 | unsigned long flags; | |
950 | struct multipath *m = pgpath->pg->m; | |
951 | ||
952 | spin_lock_irqsave(&m->lock, flags); | |
953 | ||
6680073d | 954 | if (!pgpath->is_active) |
1da177e4 LT |
955 | goto out; |
956 | ||
72d94861 | 957 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
958 | |
959 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
6680073d | 960 | pgpath->is_active = 0; |
1da177e4 LT |
961 | pgpath->fail_count++; |
962 | ||
963 | m->nr_valid_paths--; | |
964 | ||
965 | if (pgpath == m->current_pgpath) | |
966 | m->current_pgpath = NULL; | |
967 | ||
b15546f9 MA |
968 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
969 | pgpath->path.dev->name, m->nr_valid_paths); | |
970 | ||
fe9cf30e | 971 | schedule_work(&m->trigger_event); |
1da177e4 LT |
972 | |
973 | out: | |
974 | spin_unlock_irqrestore(&m->lock, flags); | |
975 | ||
976 | return 0; | |
977 | } | |
978 | ||
979 | /* | |
980 | * Reinstate a previously-failed path | |
981 | */ | |
982 | static int reinstate_path(struct pgpath *pgpath) | |
983 | { | |
63d832c3 | 984 | int r = 0, run_queue = 0; |
1da177e4 LT |
985 | unsigned long flags; |
986 | struct multipath *m = pgpath->pg->m; | |
987 | ||
988 | spin_lock_irqsave(&m->lock, flags); | |
989 | ||
6680073d | 990 | if (pgpath->is_active) |
1da177e4 LT |
991 | goto out; |
992 | ||
def052d2 | 993 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
994 | DMWARN("Reinstate path not supported by path selector %s", |
995 | pgpath->pg->ps.type->name); | |
996 | r = -EINVAL; | |
997 | goto out; | |
998 | } | |
999 | ||
1000 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1001 | if (r) | |
1002 | goto out; | |
1003 | ||
6680073d | 1004 | pgpath->is_active = 1; |
1da177e4 | 1005 | |
e8099177 | 1006 | if (!m->nr_valid_paths++) { |
e54f77dd | 1007 | m->current_pgpath = NULL; |
63d832c3 | 1008 | run_queue = 1; |
e54f77dd | 1009 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1010 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
e54f77dd CS |
1011 | m->pg_init_in_progress++; |
1012 | } | |
1da177e4 | 1013 | |
b15546f9 MA |
1014 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
1015 | pgpath->path.dev->name, m->nr_valid_paths); | |
1016 | ||
fe9cf30e | 1017 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1018 | |
1019 | out: | |
1020 | spin_unlock_irqrestore(&m->lock, flags); | |
63d832c3 HR |
1021 | if (run_queue) |
1022 | dm_table_run_md_queue_async(m->ti->table); | |
1da177e4 LT |
1023 | |
1024 | return r; | |
1025 | } | |
1026 | ||
1027 | /* | |
1028 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1029 | */ | |
1030 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1031 | action_fn action) | |
1032 | { | |
19040c0b | 1033 | int r = -EINVAL; |
1da177e4 LT |
1034 | struct pgpath *pgpath; |
1035 | struct priority_group *pg; | |
1036 | ||
1037 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1038 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1039 | if (pgpath->path.dev == dev) | |
1040 | r = action(pgpath); | |
1041 | } | |
1042 | } | |
1043 | ||
1044 | return r; | |
1045 | } | |
1046 | ||
1047 | /* | |
1048 | * Temporarily try to avoid having to use the specified PG | |
1049 | */ | |
1050 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
1051 | int bypassed) | |
1052 | { | |
1053 | unsigned long flags; | |
1054 | ||
1055 | spin_lock_irqsave(&m->lock, flags); | |
1056 | ||
1057 | pg->bypassed = bypassed; | |
1058 | m->current_pgpath = NULL; | |
1059 | m->current_pg = NULL; | |
1060 | ||
1061 | spin_unlock_irqrestore(&m->lock, flags); | |
1062 | ||
fe9cf30e | 1063 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1064 | } |
1065 | ||
1066 | /* | |
1067 | * Switch to using the specified PG from the next I/O that gets mapped | |
1068 | */ | |
1069 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1070 | { | |
1071 | struct priority_group *pg; | |
1072 | unsigned pgnum; | |
1073 | unsigned long flags; | |
31998ef1 | 1074 | char dummy; |
1da177e4 | 1075 | |
31998ef1 | 1076 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1077 | (pgnum > m->nr_priority_groups)) { |
1078 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1079 | return -EINVAL; | |
1080 | } | |
1081 | ||
1082 | spin_lock_irqsave(&m->lock, flags); | |
1083 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1084 | pg->bypassed = 0; | |
1085 | if (--pgnum) | |
1086 | continue; | |
1087 | ||
1088 | m->current_pgpath = NULL; | |
1089 | m->current_pg = NULL; | |
1090 | m->next_pg = pg; | |
1091 | } | |
1092 | spin_unlock_irqrestore(&m->lock, flags); | |
1093 | ||
fe9cf30e | 1094 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1095 | return 0; |
1096 | } | |
1097 | ||
1098 | /* | |
1099 | * Set/clear bypassed status of a PG. | |
1100 | * PGs are numbered upwards from 1 in the order they were declared. | |
1101 | */ | |
1102 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
1103 | { | |
1104 | struct priority_group *pg; | |
1105 | unsigned pgnum; | |
31998ef1 | 1106 | char dummy; |
1da177e4 | 1107 | |
31998ef1 | 1108 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1109 | (pgnum > m->nr_priority_groups)) { |
1110 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1111 | return -EINVAL; | |
1112 | } | |
1113 | ||
1114 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1115 | if (!--pgnum) | |
1116 | break; | |
1117 | } | |
1118 | ||
1119 | bypass_pg(m, pg, bypassed); | |
1120 | return 0; | |
1121 | } | |
1122 | ||
c9e45581 DW |
1123 | /* |
1124 | * Should we retry pg_init immediately? | |
1125 | */ | |
1126 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1127 | { | |
1128 | unsigned long flags; | |
1129 | int limit_reached = 0; | |
1130 | ||
1131 | spin_lock_irqsave(&m->lock, flags); | |
1132 | ||
954a73d5 | 1133 | if (m->pg_init_count <= m->pg_init_retries && !m->pg_init_disabled) |
c9e45581 DW |
1134 | m->pg_init_required = 1; |
1135 | else | |
1136 | limit_reached = 1; | |
1137 | ||
1138 | spin_unlock_irqrestore(&m->lock, flags); | |
1139 | ||
1140 | return limit_reached; | |
1141 | } | |
1142 | ||
3ae31f6a | 1143 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1144 | { |
83c0d5d5 | 1145 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1146 | struct priority_group *pg = pgpath->pg; |
1147 | struct multipath *m = pg->m; | |
1148 | unsigned long flags; | |
4e2d19e4 | 1149 | unsigned delay_retry = 0; |
cfae5c9b CS |
1150 | |
1151 | /* device or driver problems */ | |
1152 | switch (errors) { | |
1153 | case SCSI_DH_OK: | |
1154 | break; | |
1155 | case SCSI_DH_NOSYS: | |
1156 | if (!m->hw_handler_name) { | |
1157 | errors = 0; | |
1158 | break; | |
1159 | } | |
f7b934c8 MB |
1160 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1161 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1162 | /* |
1163 | * Fail path for now, so we do not ping pong | |
1164 | */ | |
1165 | fail_path(pgpath); | |
1166 | break; | |
1167 | case SCSI_DH_DEV_TEMP_BUSY: | |
1168 | /* | |
1169 | * Probably doing something like FW upgrade on the | |
1170 | * controller so try the other pg. | |
1171 | */ | |
1172 | bypass_pg(m, pg, 1); | |
1173 | break; | |
cfae5c9b | 1174 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1175 | /* Wait before retrying. */ |
1176 | delay_retry = 1; | |
cfae5c9b CS |
1177 | case SCSI_DH_IMM_RETRY: |
1178 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1179 | if (pg_init_limit_reached(m, pgpath)) | |
1180 | fail_path(pgpath); | |
1181 | errors = 0; | |
1182 | break; | |
1183 | default: | |
1184 | /* | |
1185 | * We probably do not want to fail the path for a device | |
1186 | * error, but this is what the old dm did. In future | |
1187 | * patches we can do more advanced handling. | |
1188 | */ | |
1189 | fail_path(pgpath); | |
1190 | } | |
1191 | ||
1192 | spin_lock_irqsave(&m->lock, flags); | |
1193 | if (errors) { | |
e54f77dd CS |
1194 | if (pgpath == m->current_pgpath) { |
1195 | DMERR("Could not failover device. Error %d.", errors); | |
1196 | m->current_pgpath = NULL; | |
1197 | m->current_pg = NULL; | |
1198 | } | |
d0259bf0 | 1199 | } else if (!m->pg_init_required) |
cfae5c9b | 1200 | pg->bypassed = 0; |
cfae5c9b | 1201 | |
d0259bf0 KU |
1202 | if (--m->pg_init_in_progress) |
1203 | /* Activations of other paths are still on going */ | |
1204 | goto out; | |
1205 | ||
3e9f1be1 HR |
1206 | if (m->pg_init_required) { |
1207 | m->pg_init_delay_retry = delay_retry; | |
1208 | if (__pg_init_all_paths(m)) | |
1209 | goto out; | |
1210 | } | |
1211 | m->queue_io = 0; | |
d0259bf0 | 1212 | |
2bded7bd KU |
1213 | /* |
1214 | * Wake up any thread waiting to suspend. | |
1215 | */ | |
1216 | wake_up(&m->pg_init_wait); | |
1217 | ||
d0259bf0 | 1218 | out: |
cfae5c9b CS |
1219 | spin_unlock_irqrestore(&m->lock, flags); |
1220 | } | |
1221 | ||
bab7cfc7 CS |
1222 | static void activate_path(struct work_struct *work) |
1223 | { | |
e54f77dd | 1224 | struct pgpath *pgpath = |
4e2d19e4 | 1225 | container_of(work, struct pgpath, activate_path.work); |
bab7cfc7 | 1226 | |
3a017509 HR |
1227 | if (pgpath->is_active) |
1228 | scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev), | |
1229 | pg_init_done, pgpath); | |
1230 | else | |
1231 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1232 | } |
1233 | ||
7e782af5 HR |
1234 | static int noretry_error(int error) |
1235 | { | |
1236 | switch (error) { | |
1237 | case -EOPNOTSUPP: | |
1238 | case -EREMOTEIO: | |
1239 | case -EILSEQ: | |
1240 | case -ENODATA: | |
cc9d3c38 | 1241 | case -ENOSPC: |
7e782af5 HR |
1242 | return 1; |
1243 | } | |
1244 | ||
1245 | /* Anything else could be a path failure, so should be retried */ | |
1246 | return 0; | |
1247 | } | |
1248 | ||
1da177e4 LT |
1249 | /* |
1250 | * end_io handling | |
1251 | */ | |
f40c67f0 | 1252 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1253 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1254 | { |
f40c67f0 KU |
1255 | /* |
1256 | * We don't queue any clone request inside the multipath target | |
1257 | * during end I/O handling, since those clone requests don't have | |
1258 | * bio clones. If we queue them inside the multipath target, | |
1259 | * we need to make bio clones, that requires memory allocation. | |
1260 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1261 | * don't have bio clones.) | |
1262 | * Instead of queueing the clone request here, we queue the original | |
1263 | * request into dm core, which will remake a clone request and | |
1264 | * clone bios for it and resubmit it later. | |
1265 | */ | |
1266 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1267 | unsigned long flags; |
1da177e4 | 1268 | |
f40c67f0 | 1269 | if (!error && !clone->errors) |
1da177e4 LT |
1270 | return 0; /* I/O complete */ |
1271 | ||
7eee4ae2 | 1272 | if (noretry_error(error)) |
959eb4e5 MS |
1273 | return error; |
1274 | ||
cfae5c9b CS |
1275 | if (mpio->pgpath) |
1276 | fail_path(mpio->pgpath); | |
1da177e4 | 1277 | |
640eb3b0 | 1278 | spin_lock_irqsave(&m->lock, flags); |
751b2a7d HR |
1279 | if (!m->nr_valid_paths) { |
1280 | if (!m->queue_if_no_path) { | |
1281 | if (!__must_push_back(m)) | |
1282 | r = -EIO; | |
1283 | } else { | |
1284 | if (error == -EBADE) | |
1285 | r = error; | |
1286 | } | |
1287 | } | |
640eb3b0 | 1288 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1289 | |
f40c67f0 | 1290 | return r; |
1da177e4 LT |
1291 | } |
1292 | ||
f40c67f0 | 1293 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1294 | int error, union map_info *map_context) |
1295 | { | |
028867ac AK |
1296 | struct multipath *m = ti->private; |
1297 | struct dm_mpath_io *mpio = map_context->ptr; | |
a71a261f | 1298 | struct pgpath *pgpath; |
1da177e4 LT |
1299 | struct path_selector *ps; |
1300 | int r; | |
1301 | ||
466891f9 JN |
1302 | BUG_ON(!mpio); |
1303 | ||
f40c67f0 | 1304 | r = do_end_io(m, clone, error, mpio); |
a71a261f | 1305 | pgpath = mpio->pgpath; |
1da177e4 LT |
1306 | if (pgpath) { |
1307 | ps = &pgpath->pg->ps; | |
1308 | if (ps->type->end_io) | |
02ab823f | 1309 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1310 | } |
466891f9 | 1311 | clear_mapinfo(m, map_context); |
1da177e4 LT |
1312 | |
1313 | return r; | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1318 | * the last path fails we must error any remaining I/O. |
1319 | * Note that if the freeze_bdev fails while suspending, the | |
1320 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1321 | */ |
1322 | static void multipath_presuspend(struct dm_target *ti) | |
1323 | { | |
1324 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1325 | |
485ef69e | 1326 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1327 | } |
1328 | ||
6df400ab KU |
1329 | static void multipath_postsuspend(struct dm_target *ti) |
1330 | { | |
6380f26f MA |
1331 | struct multipath *m = ti->private; |
1332 | ||
1333 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1334 | flush_multipath_work(m); |
6380f26f | 1335 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1336 | } |
1337 | ||
436d4108 AK |
1338 | /* |
1339 | * Restore the queue_if_no_path setting. | |
1340 | */ | |
1da177e4 LT |
1341 | static void multipath_resume(struct dm_target *ti) |
1342 | { | |
1343 | struct multipath *m = (struct multipath *) ti->private; | |
1344 | unsigned long flags; | |
1345 | ||
1346 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1347 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1348 | spin_unlock_irqrestore(&m->lock, flags); |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * Info output has the following format: | |
1353 | * num_multipath_feature_args [multipath_feature_args]* | |
1354 | * num_handler_status_args [handler_status_args]* | |
1355 | * num_groups init_group_number | |
1356 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1357 | * num_paths num_selector_args | |
1358 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1359 | * | |
1360 | * Table output has the following format (identical to the constructor string): | |
1361 | * num_feature_args [features_args]* | |
1362 | * num_handler_args hw_handler [hw_handler_args]* | |
1363 | * num_groups init_group_number | |
1364 | * [priority selector-name num_ps_args [ps_args]* | |
1365 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1366 | */ | |
fd7c092e MP |
1367 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1368 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1369 | { |
1370 | int sz = 0; | |
1371 | unsigned long flags; | |
1372 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1373 | struct priority_group *pg; |
1374 | struct pgpath *p; | |
1375 | unsigned pg_num; | |
1376 | char state; | |
1377 | ||
1378 | spin_lock_irqsave(&m->lock, flags); | |
1379 | ||
1380 | /* Features */ | |
1381 | if (type == STATUSTYPE_INFO) | |
e8099177 | 1382 | DMEMIT("2 %u %u ", m->queue_io, m->pg_init_count); |
c9e45581 DW |
1383 | else { |
1384 | DMEMIT("%u ", m->queue_if_no_path + | |
4e2d19e4 | 1385 | (m->pg_init_retries > 0) * 2 + |
a58a935d MS |
1386 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
1387 | m->retain_attached_hw_handler); | |
c9e45581 DW |
1388 | if (m->queue_if_no_path) |
1389 | DMEMIT("queue_if_no_path "); | |
1390 | if (m->pg_init_retries) | |
1391 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1392 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1393 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
a58a935d MS |
1394 | if (m->retain_attached_hw_handler) |
1395 | DMEMIT("retain_attached_hw_handler "); | |
c9e45581 | 1396 | } |
1da177e4 | 1397 | |
cfae5c9b | 1398 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1399 | DMEMIT("0 "); |
1400 | else | |
cfae5c9b | 1401 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1402 | |
1403 | DMEMIT("%u ", m->nr_priority_groups); | |
1404 | ||
1405 | if (m->next_pg) | |
1406 | pg_num = m->next_pg->pg_num; | |
1407 | else if (m->current_pg) | |
1408 | pg_num = m->current_pg->pg_num; | |
1409 | else | |
a490a07a | 1410 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1411 | |
1412 | DMEMIT("%u ", pg_num); | |
1413 | ||
1414 | switch (type) { | |
1415 | case STATUSTYPE_INFO: | |
1416 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1417 | if (pg->bypassed) | |
1418 | state = 'D'; /* Disabled */ | |
1419 | else if (pg == m->current_pg) | |
1420 | state = 'A'; /* Currently Active */ | |
1421 | else | |
1422 | state = 'E'; /* Enabled */ | |
1423 | ||
1424 | DMEMIT("%c ", state); | |
1425 | ||
1426 | if (pg->ps.type->status) | |
1427 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1428 | result + sz, | |
1429 | maxlen - sz); | |
1430 | else | |
1431 | DMEMIT("0 "); | |
1432 | ||
1433 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1434 | pg->ps.type->info_args); | |
1435 | ||
1436 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1437 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1438 | p->is_active ? "A" : "F", |
1da177e4 LT |
1439 | p->fail_count); |
1440 | if (pg->ps.type->status) | |
1441 | sz += pg->ps.type->status(&pg->ps, | |
1442 | &p->path, type, result + sz, | |
1443 | maxlen - sz); | |
1444 | } | |
1445 | } | |
1446 | break; | |
1447 | ||
1448 | case STATUSTYPE_TABLE: | |
1449 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1450 | DMEMIT("%s ", pg->ps.type->name); | |
1451 | ||
1452 | if (pg->ps.type->status) | |
1453 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1454 | result + sz, | |
1455 | maxlen - sz); | |
1456 | else | |
1457 | DMEMIT("0 "); | |
1458 | ||
1459 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1460 | pg->ps.type->table_args); | |
1461 | ||
1462 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1463 | DMEMIT("%s ", p->path.dev->name); | |
1464 | if (pg->ps.type->status) | |
1465 | sz += pg->ps.type->status(&pg->ps, | |
1466 | &p->path, type, result + sz, | |
1467 | maxlen - sz); | |
1468 | } | |
1469 | } | |
1470 | break; | |
1471 | } | |
1472 | ||
1473 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1474 | } |
1475 | ||
1476 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1477 | { | |
6380f26f | 1478 | int r = -EINVAL; |
1da177e4 LT |
1479 | struct dm_dev *dev; |
1480 | struct multipath *m = (struct multipath *) ti->private; | |
1481 | action_fn action; | |
1482 | ||
6380f26f MA |
1483 | mutex_lock(&m->work_mutex); |
1484 | ||
c2f3d24b KU |
1485 | if (dm_suspended(ti)) { |
1486 | r = -EBUSY; | |
1487 | goto out; | |
1488 | } | |
1489 | ||
1da177e4 | 1490 | if (argc == 1) { |
498f0103 | 1491 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
6380f26f MA |
1492 | r = queue_if_no_path(m, 1, 0); |
1493 | goto out; | |
498f0103 | 1494 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
6380f26f MA |
1495 | r = queue_if_no_path(m, 0, 0); |
1496 | goto out; | |
1497 | } | |
1da177e4 LT |
1498 | } |
1499 | ||
6380f26f | 1500 | if (argc != 2) { |
a356e426 | 1501 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1502 | goto out; |
1503 | } | |
1da177e4 | 1504 | |
498f0103 | 1505 | if (!strcasecmp(argv[0], "disable_group")) { |
6380f26f MA |
1506 | r = bypass_pg_num(m, argv[1], 1); |
1507 | goto out; | |
498f0103 | 1508 | } else if (!strcasecmp(argv[0], "enable_group")) { |
6380f26f MA |
1509 | r = bypass_pg_num(m, argv[1], 0); |
1510 | goto out; | |
498f0103 | 1511 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1512 | r = switch_pg_num(m, argv[1]); |
1513 | goto out; | |
498f0103 | 1514 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1515 | action = reinstate_path; |
498f0103 | 1516 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1517 | action = fail_path; |
6380f26f | 1518 | else { |
a356e426 | 1519 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1520 | goto out; |
1521 | } | |
1da177e4 | 1522 | |
8215d6ec | 1523 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1524 | if (r) { |
72d94861 | 1525 | DMWARN("message: error getting device %s", |
1da177e4 | 1526 | argv[1]); |
6380f26f | 1527 | goto out; |
1da177e4 LT |
1528 | } |
1529 | ||
1530 | r = action_dev(m, dev, action); | |
1531 | ||
1532 | dm_put_device(ti, dev); | |
1533 | ||
6380f26f MA |
1534 | out: |
1535 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1536 | return r; |
1da177e4 LT |
1537 | } |
1538 | ||
e56f81e0 CH |
1539 | static int multipath_prepare_ioctl(struct dm_target *ti, |
1540 | struct block_device **bdev, fmode_t *mode) | |
9af4aa30 | 1541 | { |
35991652 | 1542 | struct multipath *m = ti->private; |
9af4aa30 | 1543 | unsigned long flags; |
35991652 MP |
1544 | int r; |
1545 | ||
9af4aa30 MB |
1546 | spin_lock_irqsave(&m->lock, flags); |
1547 | ||
1548 | if (!m->current_pgpath) | |
02ab823f | 1549 | __choose_pgpath(m, 0); |
9af4aa30 | 1550 | |
43e43c9e JN |
1551 | if (m->current_pgpath) { |
1552 | if (!m->queue_io) { | |
1553 | *bdev = m->current_pgpath->path.dev->bdev; | |
1554 | *mode = m->current_pgpath->path.dev->mode; | |
1555 | r = 0; | |
1556 | } else { | |
1557 | /* pg_init has not started or completed */ | |
1558 | r = -ENOTCONN; | |
1559 | } | |
1560 | } else { | |
1561 | /* No path is available */ | |
1562 | if (m->queue_if_no_path) | |
1563 | r = -ENOTCONN; | |
1564 | else | |
1565 | r = -EIO; | |
e90dae1f | 1566 | } |
9af4aa30 | 1567 | |
9af4aa30 MB |
1568 | spin_unlock_irqrestore(&m->lock, flags); |
1569 | ||
5bbbfdf6 | 1570 | if (r == -ENOTCONN) { |
3e9f1be1 HR |
1571 | spin_lock_irqsave(&m->lock, flags); |
1572 | if (!m->current_pg) { | |
1573 | /* Path status changed, redo selection */ | |
1574 | __choose_pgpath(m, 0); | |
1575 | } | |
1576 | if (m->pg_init_required) | |
1577 | __pg_init_all_paths(m); | |
4cdd2ad7 | 1578 | spin_unlock_irqrestore(&m->lock, flags); |
63d832c3 | 1579 | dm_table_run_md_queue_async(m->ti->table); |
3e9f1be1 | 1580 | } |
35991652 | 1581 | |
e56f81e0 CH |
1582 | /* |
1583 | * Only pass ioctls through if the device sizes match exactly. | |
1584 | */ | |
1585 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
1586 | return 1; | |
1587 | return r; | |
9af4aa30 MB |
1588 | } |
1589 | ||
af4874e0 MS |
1590 | static int multipath_iterate_devices(struct dm_target *ti, |
1591 | iterate_devices_callout_fn fn, void *data) | |
1592 | { | |
1593 | struct multipath *m = ti->private; | |
1594 | struct priority_group *pg; | |
1595 | struct pgpath *p; | |
1596 | int ret = 0; | |
1597 | ||
1598 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1599 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1600 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1601 | if (ret) |
1602 | goto out; | |
1603 | } | |
1604 | } | |
1605 | ||
1606 | out: | |
1607 | return ret; | |
1608 | } | |
1609 | ||
f40c67f0 KU |
1610 | static int __pgpath_busy(struct pgpath *pgpath) |
1611 | { | |
1612 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1613 | ||
52b09914 | 1614 | return blk_lld_busy(q); |
f40c67f0 KU |
1615 | } |
1616 | ||
1617 | /* | |
1618 | * We return "busy", only when we can map I/Os but underlying devices | |
1619 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1620 | * the underlying queue). | |
1621 | * In other words, if we want to kill I/Os or queue them inside us | |
1622 | * due to map unavailability, we don't return "busy". Otherwise, | |
1623 | * dm core won't give us the I/Os and we can't do what we want. | |
1624 | */ | |
1625 | static int multipath_busy(struct dm_target *ti) | |
1626 | { | |
1627 | int busy = 0, has_active = 0; | |
1628 | struct multipath *m = ti->private; | |
1629 | struct priority_group *pg; | |
1630 | struct pgpath *pgpath; | |
1631 | unsigned long flags; | |
1632 | ||
1633 | spin_lock_irqsave(&m->lock, flags); | |
1634 | ||
7a7a3b45 JN |
1635 | /* pg_init in progress or no paths available */ |
1636 | if (m->pg_init_in_progress || | |
1637 | (!m->nr_valid_paths && m->queue_if_no_path)) { | |
b63349a7 HR |
1638 | busy = 1; |
1639 | goto out; | |
1640 | } | |
f40c67f0 KU |
1641 | /* Guess which priority_group will be used at next mapping time */ |
1642 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1643 | pg = m->next_pg; | |
1644 | else if (likely(m->current_pg)) | |
1645 | pg = m->current_pg; | |
1646 | else | |
1647 | /* | |
1648 | * We don't know which pg will be used at next mapping time. | |
1649 | * We don't call __choose_pgpath() here to avoid to trigger | |
1650 | * pg_init just by busy checking. | |
1651 | * So we don't know whether underlying devices we will be using | |
1652 | * at next mapping time are busy or not. Just try mapping. | |
1653 | */ | |
1654 | goto out; | |
1655 | ||
1656 | /* | |
1657 | * If there is one non-busy active path at least, the path selector | |
1658 | * will be able to select it. So we consider such a pg as not busy. | |
1659 | */ | |
1660 | busy = 1; | |
1661 | list_for_each_entry(pgpath, &pg->pgpaths, list) | |
1662 | if (pgpath->is_active) { | |
1663 | has_active = 1; | |
1664 | ||
1665 | if (!__pgpath_busy(pgpath)) { | |
1666 | busy = 0; | |
1667 | break; | |
1668 | } | |
1669 | } | |
1670 | ||
1671 | if (!has_active) | |
1672 | /* | |
1673 | * No active path in this pg, so this pg won't be used and | |
1674 | * the current_pg will be changed at next mapping time. | |
1675 | * We need to try mapping to determine it. | |
1676 | */ | |
1677 | busy = 0; | |
1678 | ||
1679 | out: | |
1680 | spin_unlock_irqrestore(&m->lock, flags); | |
1681 | ||
1682 | return busy; | |
1683 | } | |
1684 | ||
1da177e4 LT |
1685 | /*----------------------------------------------------------------- |
1686 | * Module setup | |
1687 | *---------------------------------------------------------------*/ | |
1688 | static struct target_type multipath_target = { | |
1689 | .name = "multipath", | |
16f12266 MS |
1690 | .version = {1, 11, 0}, |
1691 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE, | |
1da177e4 LT |
1692 | .module = THIS_MODULE, |
1693 | .ctr = multipath_ctr, | |
1694 | .dtr = multipath_dtr, | |
f40c67f0 | 1695 | .map_rq = multipath_map, |
e5863d9a MS |
1696 | .clone_and_map_rq = multipath_clone_and_map, |
1697 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 1698 | .rq_end_io = multipath_end_io, |
1da177e4 | 1699 | .presuspend = multipath_presuspend, |
6df400ab | 1700 | .postsuspend = multipath_postsuspend, |
1da177e4 LT |
1701 | .resume = multipath_resume, |
1702 | .status = multipath_status, | |
1703 | .message = multipath_message, | |
e56f81e0 | 1704 | .prepare_ioctl = multipath_prepare_ioctl, |
af4874e0 | 1705 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1706 | .busy = multipath_busy, |
1da177e4 LT |
1707 | }; |
1708 | ||
1709 | static int __init dm_multipath_init(void) | |
1710 | { | |
1711 | int r; | |
1712 | ||
1713 | /* allocate a slab for the dm_ios */ | |
028867ac | 1714 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1715 | if (!_mpio_cache) |
1716 | return -ENOMEM; | |
1717 | ||
1718 | r = dm_register_target(&multipath_target); | |
1719 | if (r < 0) { | |
0cd33124 | 1720 | DMERR("register failed %d", r); |
ff658e9c JT |
1721 | r = -EINVAL; |
1722 | goto bad_register_target; | |
1da177e4 LT |
1723 | } |
1724 | ||
4d4d66ab | 1725 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1726 | if (!kmultipathd) { |
0cd33124 | 1727 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
1728 | r = -ENOMEM; |
1729 | goto bad_alloc_kmultipathd; | |
c557308e AK |
1730 | } |
1731 | ||
bab7cfc7 CS |
1732 | /* |
1733 | * A separate workqueue is used to handle the device handlers | |
1734 | * to avoid overloading existing workqueue. Overloading the | |
1735 | * old workqueue would also create a bottleneck in the | |
1736 | * path of the storage hardware device activation. | |
1737 | */ | |
4d4d66ab TH |
1738 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1739 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1740 | if (!kmpath_handlerd) { |
1741 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
1742 | r = -ENOMEM; |
1743 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
1744 | } |
1745 | ||
72d94861 | 1746 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1747 | multipath_target.version[0], multipath_target.version[1], |
1748 | multipath_target.version[2]); | |
1749 | ||
ff658e9c JT |
1750 | return 0; |
1751 | ||
1752 | bad_alloc_kmpath_handlerd: | |
1753 | destroy_workqueue(kmultipathd); | |
1754 | bad_alloc_kmultipathd: | |
1755 | dm_unregister_target(&multipath_target); | |
1756 | bad_register_target: | |
1757 | kmem_cache_destroy(_mpio_cache); | |
1758 | ||
1da177e4 LT |
1759 | return r; |
1760 | } | |
1761 | ||
1762 | static void __exit dm_multipath_exit(void) | |
1763 | { | |
bab7cfc7 | 1764 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1765 | destroy_workqueue(kmultipathd); |
1766 | ||
10d3bd09 | 1767 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1768 | kmem_cache_destroy(_mpio_cache); |
1769 | } | |
1770 | ||
1da177e4 LT |
1771 | module_init(dm_multipath_init); |
1772 | module_exit(dm_multipath_exit); | |
1773 | ||
1774 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1775 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1776 | MODULE_LICENSE("GPL"); |