]>
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> |
78ce23b5 | 26 | #include <linux/blk-mq.h> |
1da177e4 | 27 | |
72d94861 | 28 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
29 | #define DM_PG_INIT_DELAY_MSECS 2000 |
30 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
1da177e4 LT |
31 | |
32 | /* Path properties */ | |
33 | struct pgpath { | |
34 | struct list_head list; | |
35 | ||
36 | struct priority_group *pg; /* Owning PG */ | |
37 | unsigned fail_count; /* Cumulative failure count */ | |
38 | ||
c922d5f7 | 39 | struct dm_path path; |
4e2d19e4 | 40 | struct delayed_work activate_path; |
be7d31cc MS |
41 | |
42 | bool is_active:1; /* Path status */ | |
1da177e4 LT |
43 | }; |
44 | ||
45 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
46 | ||
47 | /* | |
48 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
49 | * Each has a path selector which controls which path gets used. | |
50 | */ | |
51 | struct priority_group { | |
52 | struct list_head list; | |
53 | ||
54 | struct multipath *m; /* Owning multipath instance */ | |
55 | struct path_selector ps; | |
56 | ||
57 | unsigned pg_num; /* Reference number */ | |
1da177e4 LT |
58 | unsigned nr_pgpaths; /* Number of paths in PG */ |
59 | struct list_head pgpaths; | |
be7d31cc MS |
60 | |
61 | bool bypassed:1; /* Temporarily bypass this PG? */ | |
1da177e4 LT |
62 | }; |
63 | ||
64 | /* Multipath context */ | |
65 | struct multipath { | |
66 | struct list_head list; | |
67 | struct dm_target *ti; | |
68 | ||
cfae5c9b | 69 | const char *hw_handler_name; |
2bfd2e13 | 70 | char *hw_handler_params; |
4e2d19e4 | 71 | |
1fbdd2b3 MS |
72 | spinlock_t lock; |
73 | ||
1da177e4 LT |
74 | unsigned nr_priority_groups; |
75 | struct list_head priority_groups; | |
4e2d19e4 CS |
76 | |
77 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
78 | ||
c3cd4f6b | 79 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
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 */ | |
1da177e4 | 85 | |
be7d31cc MS |
86 | bool queue_io:1; /* Must we queue all I/O? */ |
87 | bool queue_if_no_path:1; /* Queue I/O if last path fails? */ | |
88 | bool saved_queue_if_no_path:1; /* Saved state during suspension */ | |
89 | bool retain_attached_hw_handler:1; /* If there's already a hw_handler present, don't change it. */ | |
90 | bool pg_init_disabled:1; /* pg_init is not currently allowed */ | |
91 | bool pg_init_required:1; /* pg_init needs calling? */ | |
92 | bool pg_init_delay_retry:1; /* Delay pg_init retry? */ | |
1fbdd2b3 | 93 | |
c9e45581 DW |
94 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
95 | unsigned pg_init_count; /* Number of times pg_init called */ | |
4e2d19e4 | 96 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
1da177e4 | 97 | |
1da177e4 LT |
98 | struct work_struct trigger_event; |
99 | ||
100 | /* | |
028867ac | 101 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
102 | * can resubmit bios on error. |
103 | */ | |
104 | mempool_t *mpio_pool; | |
6380f26f MA |
105 | |
106 | struct mutex work_mutex; | |
1da177e4 LT |
107 | }; |
108 | ||
109 | /* | |
110 | * Context information attached to each bio we process. | |
111 | */ | |
028867ac | 112 | struct dm_mpath_io { |
1da177e4 | 113 | struct pgpath *pgpath; |
02ab823f | 114 | size_t nr_bytes; |
1da177e4 LT |
115 | }; |
116 | ||
117 | typedef int (*action_fn) (struct pgpath *pgpath); | |
118 | ||
e18b890b | 119 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 120 | |
bab7cfc7 | 121 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 | 122 | static void trigger_event(struct work_struct *work); |
bab7cfc7 | 123 | static void activate_path(struct work_struct *work); |
e8099177 | 124 | static int __pgpath_busy(struct pgpath *pgpath); |
1da177e4 LT |
125 | |
126 | ||
127 | /*----------------------------------------------- | |
128 | * Allocation routines | |
129 | *-----------------------------------------------*/ | |
130 | ||
131 | static struct pgpath *alloc_pgpath(void) | |
132 | { | |
e69fae56 | 133 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 134 | |
224cb3e9 | 135 | if (pgpath) { |
be7d31cc | 136 | pgpath->is_active = true; |
4e2d19e4 | 137 | INIT_DELAYED_WORK(&pgpath->activate_path, activate_path); |
224cb3e9 | 138 | } |
1da177e4 LT |
139 | |
140 | return pgpath; | |
141 | } | |
142 | ||
028867ac | 143 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
144 | { |
145 | kfree(pgpath); | |
146 | } | |
147 | ||
148 | static struct priority_group *alloc_priority_group(void) | |
149 | { | |
150 | struct priority_group *pg; | |
151 | ||
e69fae56 | 152 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 153 | |
e69fae56 MM |
154 | if (pg) |
155 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
156 | |
157 | return pg; | |
158 | } | |
159 | ||
160 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
161 | { | |
162 | struct pgpath *pgpath, *tmp; | |
163 | ||
164 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
165 | list_del(&pgpath->list); | |
166 | dm_put_device(ti, pgpath->path.dev); | |
167 | free_pgpath(pgpath); | |
168 | } | |
169 | } | |
170 | ||
171 | static void free_priority_group(struct priority_group *pg, | |
172 | struct dm_target *ti) | |
173 | { | |
174 | struct path_selector *ps = &pg->ps; | |
175 | ||
176 | if (ps->type) { | |
177 | ps->type->destroy(ps); | |
178 | dm_put_path_selector(ps->type); | |
179 | } | |
180 | ||
181 | free_pgpaths(&pg->pgpaths, ti); | |
182 | kfree(pg); | |
183 | } | |
184 | ||
8637a6bf | 185 | static struct multipath *alloc_multipath(struct dm_target *ti, bool use_blk_mq) |
1da177e4 LT |
186 | { |
187 | struct multipath *m; | |
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); | |
be7d31cc | 193 | m->queue_io = true; |
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); |
8637a6bf MS |
198 | |
199 | m->mpio_pool = NULL; | |
200 | if (!use_blk_mq) { | |
201 | unsigned min_ios = dm_get_reserved_rq_based_ios(); | |
202 | ||
203 | m->mpio_pool = mempool_create_slab_pool(min_ios, _mpio_cache); | |
204 | if (!m->mpio_pool) { | |
205 | kfree(m); | |
206 | return NULL; | |
207 | } | |
1da177e4 | 208 | } |
8637a6bf | 209 | |
28f16c20 MM |
210 | m->ti = ti; |
211 | ti->private = m; | |
1da177e4 LT |
212 | } |
213 | ||
214 | return m; | |
215 | } | |
216 | ||
217 | static void free_multipath(struct multipath *m) | |
218 | { | |
219 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
220 | |
221 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
222 | list_del(&pg->list); | |
223 | free_priority_group(pg, m->ti); | |
224 | } | |
225 | ||
cfae5c9b | 226 | kfree(m->hw_handler_name); |
2bfd2e13 | 227 | kfree(m->hw_handler_params); |
1da177e4 LT |
228 | mempool_destroy(m->mpio_pool); |
229 | kfree(m); | |
230 | } | |
231 | ||
2eff1924 MS |
232 | static struct dm_mpath_io *get_mpio(union map_info *info) |
233 | { | |
234 | return info->ptr; | |
235 | } | |
236 | ||
237 | static struct dm_mpath_io *set_mpio(struct multipath *m, union map_info *info) | |
466891f9 JN |
238 | { |
239 | struct dm_mpath_io *mpio; | |
240 | ||
8637a6bf MS |
241 | if (!m->mpio_pool) { |
242 | /* Use blk-mq pdu memory requested via per_io_data_size */ | |
2eff1924 | 243 | mpio = get_mpio(info); |
8637a6bf MS |
244 | memset(mpio, 0, sizeof(*mpio)); |
245 | return mpio; | |
246 | } | |
247 | ||
466891f9 JN |
248 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); |
249 | if (!mpio) | |
2eff1924 | 250 | return NULL; |
466891f9 JN |
251 | |
252 | memset(mpio, 0, sizeof(*mpio)); | |
253 | info->ptr = mpio; | |
254 | ||
2eff1924 | 255 | return mpio; |
466891f9 JN |
256 | } |
257 | ||
2eff1924 | 258 | static void clear_request_fn_mpio(struct multipath *m, union map_info *info) |
466891f9 | 259 | { |
2eff1924 | 260 | /* Only needed for non blk-mq (.request_fn) multipath */ |
8637a6bf MS |
261 | if (m->mpio_pool) { |
262 | struct dm_mpath_io *mpio = info->ptr; | |
466891f9 | 263 | |
8637a6bf MS |
264 | info->ptr = NULL; |
265 | mempool_free(mpio, m->mpio_pool); | |
266 | } | |
466891f9 | 267 | } |
1da177e4 LT |
268 | |
269 | /*----------------------------------------------- | |
270 | * Path selection | |
271 | *-----------------------------------------------*/ | |
272 | ||
3e9f1be1 | 273 | static int __pg_init_all_paths(struct multipath *m) |
fb612642 KU |
274 | { |
275 | struct pgpath *pgpath; | |
4e2d19e4 | 276 | unsigned long pg_init_delay = 0; |
fb612642 | 277 | |
17f4ff45 | 278 | if (m->pg_init_in_progress || m->pg_init_disabled) |
3e9f1be1 | 279 | return 0; |
17f4ff45 | 280 | |
fb612642 | 281 | m->pg_init_count++; |
be7d31cc | 282 | m->pg_init_required = false; |
3e9f1be1 HR |
283 | |
284 | /* Check here to reset pg_init_required */ | |
285 | if (!m->current_pg) | |
286 | return 0; | |
287 | ||
4e2d19e4 CS |
288 | if (m->pg_init_delay_retry) |
289 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? | |
290 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
291 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
292 | /* Skip failed paths */ | |
293 | if (!pgpath->is_active) | |
294 | continue; | |
4e2d19e4 CS |
295 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
296 | pg_init_delay)) | |
fb612642 KU |
297 | m->pg_init_in_progress++; |
298 | } | |
3e9f1be1 | 299 | return m->pg_init_in_progress; |
fb612642 KU |
300 | } |
301 | ||
1da177e4 LT |
302 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) |
303 | { | |
1da177e4 LT |
304 | m->current_pg = pgpath->pg; |
305 | ||
306 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 307 | if (m->hw_handler_name) { |
be7d31cc MS |
308 | m->pg_init_required = true; |
309 | m->queue_io = true; | |
1da177e4 | 310 | } else { |
be7d31cc MS |
311 | m->pg_init_required = false; |
312 | m->queue_io = false; | |
1da177e4 | 313 | } |
c9e45581 DW |
314 | |
315 | m->pg_init_count = 0; | |
1da177e4 LT |
316 | } |
317 | ||
02ab823f KU |
318 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
319 | size_t nr_bytes) | |
1da177e4 | 320 | { |
c922d5f7 | 321 | struct dm_path *path; |
1da177e4 | 322 | |
90a4323c | 323 | path = pg->ps.type->select_path(&pg->ps, nr_bytes); |
1da177e4 LT |
324 | if (!path) |
325 | return -ENXIO; | |
326 | ||
327 | m->current_pgpath = path_to_pgpath(path); | |
328 | ||
329 | if (m->current_pg != pg) | |
330 | __switch_pg(m, m->current_pgpath); | |
331 | ||
332 | return 0; | |
333 | } | |
334 | ||
02ab823f | 335 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
336 | { |
337 | struct priority_group *pg; | |
be7d31cc | 338 | bool bypassed = true; |
1da177e4 | 339 | |
1f271972 | 340 | if (!m->nr_valid_paths) { |
be7d31cc | 341 | m->queue_io = false; |
1da177e4 | 342 | goto failed; |
1f271972 | 343 | } |
1da177e4 LT |
344 | |
345 | /* Were we instructed to switch PG? */ | |
346 | if (m->next_pg) { | |
347 | pg = m->next_pg; | |
348 | m->next_pg = NULL; | |
02ab823f | 349 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
350 | return; |
351 | } | |
352 | ||
353 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 354 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
355 | return; |
356 | ||
357 | /* | |
358 | * Loop through priority groups until we find a valid path. | |
359 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
360 | * Second time we only try the ones we skipped, but set |
361 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
362 | */ |
363 | do { | |
364 | list_for_each_entry(pg, &m->priority_groups, list) { | |
365 | if (pg->bypassed == bypassed) | |
366 | continue; | |
f220fd4e MC |
367 | if (!__choose_path_in_pg(m, pg, nr_bytes)) { |
368 | if (!bypassed) | |
be7d31cc | 369 | m->pg_init_delay_retry = true; |
1da177e4 | 370 | return; |
f220fd4e | 371 | } |
1da177e4 LT |
372 | } |
373 | } while (bypassed--); | |
374 | ||
375 | failed: | |
376 | m->current_pgpath = NULL; | |
377 | m->current_pg = NULL; | |
378 | } | |
379 | ||
45e15720 KU |
380 | /* |
381 | * Check whether bios must be queued in the device-mapper core rather | |
382 | * than here in the target. | |
383 | * | |
384 | * m->lock must be held on entry. | |
385 | * | |
386 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
387 | * same value then we are not between multipath_presuspend() | |
388 | * and multipath_resume() calls and we have no need to check | |
389 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
390 | */ | |
391 | static int __must_push_back(struct multipath *m) | |
392 | { | |
e8099177 HR |
393 | return (m->queue_if_no_path || |
394 | (m->queue_if_no_path != m->saved_queue_if_no_path && | |
395 | dm_noflush_suspending(m->ti))); | |
45e15720 KU |
396 | } |
397 | ||
36fcffcc HR |
398 | /* |
399 | * Map cloned requests | |
400 | */ | |
e5863d9a MS |
401 | static int __multipath_map(struct dm_target *ti, struct request *clone, |
402 | union map_info *map_context, | |
403 | struct request *rq, struct request **__clone) | |
1da177e4 | 404 | { |
7943bd6d | 405 | struct multipath *m = ti->private; |
e3bde04f | 406 | int r = DM_MAPIO_REQUEUE; |
e5863d9a | 407 | size_t nr_bytes = clone ? blk_rq_bytes(clone) : blk_rq_bytes(rq); |
1da177e4 | 408 | struct pgpath *pgpath; |
f40c67f0 | 409 | struct block_device *bdev; |
e3bde04f | 410 | struct dm_mpath_io *mpio; |
1da177e4 | 411 | |
2eb6e1e3 | 412 | spin_lock_irq(&m->lock); |
1da177e4 LT |
413 | |
414 | /* Do we need to select a new pgpath? */ | |
21136f89 | 415 | if (!m->current_pgpath || !m->queue_io) |
02ab823f | 416 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
417 | |
418 | pgpath = m->current_pgpath; | |
419 | ||
9bf59a61 MS |
420 | if (!pgpath) { |
421 | if (!__must_push_back(m)) | |
422 | r = -EIO; /* Failed */ | |
423 | goto out_unlock; | |
6afbc01d | 424 | } else if (m->queue_io || m->pg_init_required) { |
e3bde04f | 425 | __pg_init_all_paths(m); |
9bf59a61 MS |
426 | goto out_unlock; |
427 | } | |
6afbc01d | 428 | |
2eff1924 MS |
429 | mpio = set_mpio(m, map_context); |
430 | if (!mpio) | |
9bf59a61 MS |
431 | /* ENOMEM, requeue */ |
432 | goto out_unlock; | |
433 | ||
2eb6e1e3 KB |
434 | mpio->pgpath = pgpath; |
435 | mpio->nr_bytes = nr_bytes; | |
436 | ||
9bf59a61 | 437 | bdev = pgpath->path.dev->bdev; |
2eb6e1e3 | 438 | |
2eb6e1e3 KB |
439 | spin_unlock_irq(&m->lock); |
440 | ||
e5863d9a | 441 | if (clone) { |
c5248f79 MS |
442 | /* |
443 | * Old request-based interface: allocated clone is passed in. | |
444 | * Used by: .request_fn stacked on .request_fn path(s). | |
445 | */ | |
e5863d9a MS |
446 | clone->q = bdev_get_queue(bdev); |
447 | clone->rq_disk = bdev->bd_disk; | |
448 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
449 | } else { | |
eca7ee6d MS |
450 | /* |
451 | * blk-mq request-based interface; used by both: | |
452 | * .request_fn stacked on blk-mq path(s) and | |
453 | * blk-mq stacked on blk-mq path(s). | |
454 | */ | |
78ce23b5 MS |
455 | *__clone = blk_mq_alloc_request(bdev_get_queue(bdev), |
456 | rq_data_dir(rq), BLK_MQ_REQ_NOWAIT); | |
4c6dd53d | 457 | if (IS_ERR(*__clone)) { |
e5863d9a | 458 | /* ENOMEM, requeue */ |
2eff1924 | 459 | clear_request_fn_mpio(m, map_context); |
e5863d9a | 460 | return r; |
4c6dd53d | 461 | } |
e5863d9a MS |
462 | (*__clone)->bio = (*__clone)->biotail = NULL; |
463 | (*__clone)->rq_disk = bdev->bd_disk; | |
464 | (*__clone)->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
465 | } | |
466 | ||
9bf59a61 MS |
467 | if (pgpath->pg->ps.type->start_io) |
468 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
469 | &pgpath->path, | |
470 | nr_bytes); | |
2eb6e1e3 | 471 | return DM_MAPIO_REMAPPED; |
1da177e4 | 472 | |
e3bde04f | 473 | out_unlock: |
2eb6e1e3 | 474 | spin_unlock_irq(&m->lock); |
1da177e4 LT |
475 | |
476 | return r; | |
477 | } | |
478 | ||
e5863d9a MS |
479 | static int multipath_map(struct dm_target *ti, struct request *clone, |
480 | union map_info *map_context) | |
481 | { | |
482 | return __multipath_map(ti, clone, map_context, NULL, NULL); | |
483 | } | |
484 | ||
485 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, | |
486 | union map_info *map_context, | |
487 | struct request **clone) | |
488 | { | |
489 | return __multipath_map(ti, NULL, map_context, rq, clone); | |
490 | } | |
491 | ||
492 | static void multipath_release_clone(struct request *clone) | |
493 | { | |
78ce23b5 | 494 | blk_mq_free_request(clone); |
e5863d9a MS |
495 | } |
496 | ||
1da177e4 LT |
497 | /* |
498 | * If we run out of usable paths, should we queue I/O or error it? | |
499 | */ | |
be7d31cc MS |
500 | static int queue_if_no_path(struct multipath *m, bool queue_if_no_path, |
501 | bool save_old_value) | |
1da177e4 LT |
502 | { |
503 | unsigned long flags; | |
504 | ||
505 | spin_lock_irqsave(&m->lock, flags); | |
506 | ||
485ef69e AK |
507 | if (save_old_value) |
508 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
509 | else | |
510 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 511 | m->queue_if_no_path = queue_if_no_path; |
1da177e4 LT |
512 | spin_unlock_irqrestore(&m->lock, flags); |
513 | ||
63d832c3 HR |
514 | if (!queue_if_no_path) |
515 | dm_table_run_md_queue_async(m->ti->table); | |
516 | ||
1da177e4 LT |
517 | return 0; |
518 | } | |
519 | ||
1da177e4 LT |
520 | /* |
521 | * An event is triggered whenever a path is taken out of use. | |
522 | * Includes path failure and PG bypass. | |
523 | */ | |
c4028958 | 524 | static void trigger_event(struct work_struct *work) |
1da177e4 | 525 | { |
c4028958 DH |
526 | struct multipath *m = |
527 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
528 | |
529 | dm_table_event(m->ti->table); | |
530 | } | |
531 | ||
532 | /*----------------------------------------------------------------- | |
533 | * Constructor/argument parsing: | |
534 | * <#multipath feature args> [<arg>]* | |
535 | * <#hw_handler args> [hw_handler [<arg>]*] | |
536 | * <#priority groups> | |
537 | * <initial priority group> | |
538 | * [<selector> <#selector args> [<arg>]* | |
539 | * <#paths> <#per-path selector args> | |
540 | * [<path> [<arg>]* ]+ ]+ | |
541 | *---------------------------------------------------------------*/ | |
498f0103 | 542 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
543 | struct dm_target *ti) |
544 | { | |
545 | int r; | |
546 | struct path_selector_type *pst; | |
547 | unsigned ps_argc; | |
548 | ||
498f0103 | 549 | static struct dm_arg _args[] = { |
72d94861 | 550 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
551 | }; |
552 | ||
498f0103 | 553 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 554 | if (!pst) { |
72d94861 | 555 | ti->error = "unknown path selector type"; |
1da177e4 LT |
556 | return -EINVAL; |
557 | } | |
558 | ||
498f0103 | 559 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
560 | if (r) { |
561 | dm_put_path_selector(pst); | |
1da177e4 | 562 | return -EINVAL; |
371b2e34 | 563 | } |
1da177e4 LT |
564 | |
565 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
566 | if (r) { | |
567 | dm_put_path_selector(pst); | |
72d94861 | 568 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
569 | return r; |
570 | } | |
571 | ||
572 | pg->ps.type = pst; | |
498f0103 | 573 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
574 | |
575 | return 0; | |
576 | } | |
577 | ||
498f0103 | 578 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
579 | struct dm_target *ti) |
580 | { | |
581 | int r; | |
582 | struct pgpath *p; | |
ae11b1b3 | 583 | struct multipath *m = ti->private; |
a58a935d MS |
584 | struct request_queue *q = NULL; |
585 | const char *attached_handler_name; | |
1da177e4 LT |
586 | |
587 | /* we need at least a path arg */ | |
588 | if (as->argc < 1) { | |
72d94861 | 589 | ti->error = "no device given"; |
01460f35 | 590 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
591 | } |
592 | ||
593 | p = alloc_pgpath(); | |
594 | if (!p) | |
01460f35 | 595 | return ERR_PTR(-ENOMEM); |
1da177e4 | 596 | |
498f0103 | 597 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 598 | &p->path.dev); |
1da177e4 | 599 | if (r) { |
72d94861 | 600 | ti->error = "error getting device"; |
1da177e4 LT |
601 | goto bad; |
602 | } | |
603 | ||
a58a935d MS |
604 | if (m->retain_attached_hw_handler || m->hw_handler_name) |
605 | q = bdev_get_queue(p->path.dev->bdev); | |
606 | ||
607 | if (m->retain_attached_hw_handler) { | |
1bab0de0 | 608 | retain: |
a58a935d MS |
609 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); |
610 | if (attached_handler_name) { | |
611 | /* | |
612 | * Reset hw_handler_name to match the attached handler | |
613 | * and clear any hw_handler_params associated with the | |
614 | * ignored handler. | |
615 | * | |
616 | * NB. This modifies the table line to show the actual | |
617 | * handler instead of the original table passed in. | |
618 | */ | |
619 | kfree(m->hw_handler_name); | |
620 | m->hw_handler_name = attached_handler_name; | |
621 | ||
622 | kfree(m->hw_handler_params); | |
623 | m->hw_handler_params = NULL; | |
624 | } | |
625 | } | |
a0cf7ea9 | 626 | |
a58a935d | 627 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
628 | r = scsi_dh_attach(q, m->hw_handler_name); |
629 | if (r == -EBUSY) { | |
1bab0de0 | 630 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 631 | |
1bab0de0 CH |
632 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
633 | bdevname(p->path.dev->bdev, b)); | |
634 | goto retain; | |
635 | } | |
ae11b1b3 | 636 | if (r < 0) { |
a0cf7ea9 | 637 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
638 | dm_put_device(ti, p->path.dev); |
639 | goto bad; | |
640 | } | |
2bfd2e13 CS |
641 | |
642 | if (m->hw_handler_params) { | |
643 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
644 | if (r < 0) { | |
645 | ti->error = "unable to set hardware " | |
646 | "handler parameters"; | |
2bfd2e13 CS |
647 | dm_put_device(ti, p->path.dev); |
648 | goto bad; | |
649 | } | |
650 | } | |
ae11b1b3 HR |
651 | } |
652 | ||
1da177e4 LT |
653 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
654 | if (r) { | |
655 | dm_put_device(ti, p->path.dev); | |
656 | goto bad; | |
657 | } | |
658 | ||
659 | return p; | |
660 | ||
661 | bad: | |
662 | free_pgpath(p); | |
01460f35 | 663 | return ERR_PTR(r); |
1da177e4 LT |
664 | } |
665 | ||
498f0103 | 666 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 667 | struct multipath *m) |
1da177e4 | 668 | { |
498f0103 | 669 | static struct dm_arg _args[] = { |
72d94861 AK |
670 | {1, 1024, "invalid number of paths"}, |
671 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
672 | }; |
673 | ||
674 | int r; | |
498f0103 | 675 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 676 | struct priority_group *pg; |
28f16c20 | 677 | struct dm_target *ti = m->ti; |
1da177e4 LT |
678 | |
679 | if (as->argc < 2) { | |
680 | as->argc = 0; | |
01460f35 BM |
681 | ti->error = "not enough priority group arguments"; |
682 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
683 | } |
684 | ||
685 | pg = alloc_priority_group(); | |
686 | if (!pg) { | |
72d94861 | 687 | ti->error = "couldn't allocate priority group"; |
01460f35 | 688 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
689 | } |
690 | pg->m = m; | |
691 | ||
692 | r = parse_path_selector(as, pg, ti); | |
693 | if (r) | |
694 | goto bad; | |
695 | ||
696 | /* | |
697 | * read the paths | |
698 | */ | |
498f0103 | 699 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
700 | if (r) |
701 | goto bad; | |
702 | ||
498f0103 | 703 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
704 | if (r) |
705 | goto bad; | |
706 | ||
498f0103 | 707 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
708 | for (i = 0; i < pg->nr_pgpaths; i++) { |
709 | struct pgpath *pgpath; | |
498f0103 | 710 | struct dm_arg_set path_args; |
1da177e4 | 711 | |
498f0103 | 712 | if (as->argc < nr_args) { |
148acff6 | 713 | ti->error = "not enough path parameters"; |
6bbf79a1 | 714 | r = -EINVAL; |
1da177e4 | 715 | goto bad; |
148acff6 | 716 | } |
1da177e4 | 717 | |
498f0103 | 718 | path_args.argc = nr_args; |
1da177e4 LT |
719 | path_args.argv = as->argv; |
720 | ||
721 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
722 | if (IS_ERR(pgpath)) { |
723 | r = PTR_ERR(pgpath); | |
1da177e4 | 724 | goto bad; |
01460f35 | 725 | } |
1da177e4 LT |
726 | |
727 | pgpath->pg = pg; | |
728 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 729 | dm_consume_args(as, nr_args); |
1da177e4 LT |
730 | } |
731 | ||
732 | return pg; | |
733 | ||
734 | bad: | |
735 | free_priority_group(pg, ti); | |
01460f35 | 736 | return ERR_PTR(r); |
1da177e4 LT |
737 | } |
738 | ||
498f0103 | 739 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 740 | { |
1da177e4 | 741 | unsigned hw_argc; |
2bfd2e13 | 742 | int ret; |
28f16c20 | 743 | struct dm_target *ti = m->ti; |
1da177e4 | 744 | |
498f0103 | 745 | static struct dm_arg _args[] = { |
72d94861 | 746 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
747 | }; |
748 | ||
498f0103 | 749 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
750 | return -EINVAL; |
751 | ||
752 | if (!hw_argc) | |
753 | return 0; | |
754 | ||
498f0103 | 755 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
14e98c5c | 756 | |
2bfd2e13 CS |
757 | if (hw_argc > 1) { |
758 | char *p; | |
759 | int i, j, len = 4; | |
760 | ||
761 | for (i = 0; i <= hw_argc - 2; i++) | |
762 | len += strlen(as->argv[i]) + 1; | |
763 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
764 | if (!p) { | |
765 | ti->error = "memory allocation failed"; | |
766 | ret = -ENOMEM; | |
767 | goto fail; | |
768 | } | |
769 | j = sprintf(p, "%d", hw_argc - 1); | |
770 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
771 | j = sprintf(p, "%s", as->argv[i]); | |
772 | } | |
498f0103 | 773 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
774 | |
775 | return 0; | |
2bfd2e13 CS |
776 | fail: |
777 | kfree(m->hw_handler_name); | |
778 | m->hw_handler_name = NULL; | |
779 | return ret; | |
1da177e4 LT |
780 | } |
781 | ||
498f0103 | 782 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
783 | { |
784 | int r; | |
785 | unsigned argc; | |
28f16c20 | 786 | struct dm_target *ti = m->ti; |
498f0103 | 787 | const char *arg_name; |
1da177e4 | 788 | |
498f0103 | 789 | static struct dm_arg _args[] = { |
a58a935d | 790 | {0, 6, "invalid number of feature args"}, |
c9e45581 | 791 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 792 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
793 | }; |
794 | ||
498f0103 | 795 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
796 | if (r) |
797 | return -EINVAL; | |
798 | ||
799 | if (!argc) | |
800 | return 0; | |
801 | ||
c9e45581 | 802 | do { |
498f0103 | 803 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
804 | argc--; |
805 | ||
498f0103 | 806 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
be7d31cc | 807 | r = queue_if_no_path(m, true, false); |
c9e45581 DW |
808 | continue; |
809 | } | |
810 | ||
a58a935d | 811 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
be7d31cc | 812 | m->retain_attached_hw_handler = true; |
a58a935d MS |
813 | continue; |
814 | } | |
815 | ||
498f0103 | 816 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 817 | (argc >= 1)) { |
498f0103 | 818 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
819 | argc--; |
820 | continue; | |
821 | } | |
822 | ||
498f0103 | 823 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 824 | (argc >= 1)) { |
498f0103 | 825 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
826 | argc--; |
827 | continue; | |
828 | } | |
829 | ||
1da177e4 | 830 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
831 | r = -EINVAL; |
832 | } while (argc && !r); | |
833 | ||
834 | return r; | |
1da177e4 LT |
835 | } |
836 | ||
837 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
838 | char **argv) | |
839 | { | |
498f0103 MS |
840 | /* target arguments */ |
841 | static struct dm_arg _args[] = { | |
a490a07a MS |
842 | {0, 1024, "invalid number of priority groups"}, |
843 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
844 | }; |
845 | ||
846 | int r; | |
847 | struct multipath *m; | |
498f0103 | 848 | struct dm_arg_set as; |
1da177e4 LT |
849 | unsigned pg_count = 0; |
850 | unsigned next_pg_num; | |
8637a6bf | 851 | bool use_blk_mq = dm_use_blk_mq(dm_table_get_md(ti->table)); |
1da177e4 LT |
852 | |
853 | as.argc = argc; | |
854 | as.argv = argv; | |
855 | ||
8637a6bf | 856 | m = alloc_multipath(ti, use_blk_mq); |
1da177e4 | 857 | if (!m) { |
72d94861 | 858 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
859 | return -EINVAL; |
860 | } | |
861 | ||
28f16c20 | 862 | r = parse_features(&as, m); |
1da177e4 LT |
863 | if (r) |
864 | goto bad; | |
865 | ||
28f16c20 | 866 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
867 | if (r) |
868 | goto bad; | |
869 | ||
498f0103 | 870 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
871 | if (r) |
872 | goto bad; | |
873 | ||
498f0103 | 874 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
875 | if (r) |
876 | goto bad; | |
877 | ||
a490a07a MS |
878 | if ((!m->nr_priority_groups && next_pg_num) || |
879 | (m->nr_priority_groups && !next_pg_num)) { | |
880 | ti->error = "invalid initial priority group"; | |
881 | r = -EINVAL; | |
882 | goto bad; | |
883 | } | |
884 | ||
1da177e4 LT |
885 | /* parse the priority groups */ |
886 | while (as.argc) { | |
887 | struct priority_group *pg; | |
888 | ||
28f16c20 | 889 | pg = parse_priority_group(&as, m); |
01460f35 BM |
890 | if (IS_ERR(pg)) { |
891 | r = PTR_ERR(pg); | |
1da177e4 LT |
892 | goto bad; |
893 | } | |
894 | ||
895 | m->nr_valid_paths += pg->nr_pgpaths; | |
896 | list_add_tail(&pg->list, &m->priority_groups); | |
897 | pg_count++; | |
898 | pg->pg_num = pg_count; | |
899 | if (!--next_pg_num) | |
900 | m->next_pg = pg; | |
901 | } | |
902 | ||
903 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 904 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
905 | r = -EINVAL; |
906 | goto bad; | |
907 | } | |
908 | ||
55a62eef AK |
909 | ti->num_flush_bios = 1; |
910 | ti->num_discard_bios = 1; | |
042bcef8 | 911 | ti->num_write_same_bios = 1; |
8637a6bf MS |
912 | if (use_blk_mq) |
913 | ti->per_io_data_size = sizeof(struct dm_mpath_io); | |
8627921f | 914 | |
1da177e4 LT |
915 | return 0; |
916 | ||
917 | bad: | |
918 | free_multipath(m); | |
919 | return r; | |
920 | } | |
921 | ||
2bded7bd KU |
922 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
923 | { | |
924 | DECLARE_WAITQUEUE(wait, current); | |
925 | unsigned long flags; | |
926 | ||
927 | add_wait_queue(&m->pg_init_wait, &wait); | |
928 | ||
929 | while (1) { | |
930 | set_current_state(TASK_UNINTERRUPTIBLE); | |
931 | ||
932 | spin_lock_irqsave(&m->lock, flags); | |
933 | if (!m->pg_init_in_progress) { | |
934 | spin_unlock_irqrestore(&m->lock, flags); | |
935 | break; | |
936 | } | |
937 | spin_unlock_irqrestore(&m->lock, flags); | |
938 | ||
939 | io_schedule(); | |
940 | } | |
941 | set_current_state(TASK_RUNNING); | |
942 | ||
943 | remove_wait_queue(&m->pg_init_wait, &wait); | |
944 | } | |
945 | ||
946 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 947 | { |
954a73d5 SKM |
948 | unsigned long flags; |
949 | ||
950 | spin_lock_irqsave(&m->lock, flags); | |
be7d31cc | 951 | m->pg_init_disabled = true; |
954a73d5 SKM |
952 | spin_unlock_irqrestore(&m->lock, flags); |
953 | ||
bab7cfc7 | 954 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 955 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 956 | flush_workqueue(kmultipathd); |
43829731 | 957 | flush_work(&m->trigger_event); |
954a73d5 SKM |
958 | |
959 | spin_lock_irqsave(&m->lock, flags); | |
be7d31cc | 960 | m->pg_init_disabled = false; |
954a73d5 | 961 | spin_unlock_irqrestore(&m->lock, flags); |
6df400ab KU |
962 | } |
963 | ||
964 | static void multipath_dtr(struct dm_target *ti) | |
965 | { | |
966 | struct multipath *m = ti->private; | |
967 | ||
2bded7bd | 968 | flush_multipath_work(m); |
1da177e4 LT |
969 | free_multipath(m); |
970 | } | |
971 | ||
1da177e4 LT |
972 | /* |
973 | * Take a path out of use. | |
974 | */ | |
975 | static int fail_path(struct pgpath *pgpath) | |
976 | { | |
977 | unsigned long flags; | |
978 | struct multipath *m = pgpath->pg->m; | |
979 | ||
980 | spin_lock_irqsave(&m->lock, flags); | |
981 | ||
6680073d | 982 | if (!pgpath->is_active) |
1da177e4 LT |
983 | goto out; |
984 | ||
72d94861 | 985 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
986 | |
987 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
be7d31cc | 988 | pgpath->is_active = false; |
1da177e4 LT |
989 | pgpath->fail_count++; |
990 | ||
991 | m->nr_valid_paths--; | |
992 | ||
993 | if (pgpath == m->current_pgpath) | |
994 | m->current_pgpath = NULL; | |
995 | ||
b15546f9 MA |
996 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
997 | pgpath->path.dev->name, m->nr_valid_paths); | |
998 | ||
fe9cf30e | 999 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1000 | |
1001 | out: | |
1002 | spin_unlock_irqrestore(&m->lock, flags); | |
1003 | ||
1004 | return 0; | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * Reinstate a previously-failed path | |
1009 | */ | |
1010 | static int reinstate_path(struct pgpath *pgpath) | |
1011 | { | |
63d832c3 | 1012 | int r = 0, run_queue = 0; |
1da177e4 LT |
1013 | unsigned long flags; |
1014 | struct multipath *m = pgpath->pg->m; | |
1015 | ||
1016 | spin_lock_irqsave(&m->lock, flags); | |
1017 | ||
6680073d | 1018 | if (pgpath->is_active) |
1da177e4 LT |
1019 | goto out; |
1020 | ||
def052d2 | 1021 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
1022 | DMWARN("Reinstate path not supported by path selector %s", |
1023 | pgpath->pg->ps.type->name); | |
1024 | r = -EINVAL; | |
1025 | goto out; | |
1026 | } | |
1027 | ||
1028 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1029 | if (r) | |
1030 | goto out; | |
1031 | ||
be7d31cc | 1032 | pgpath->is_active = true; |
1da177e4 | 1033 | |
e8099177 | 1034 | if (!m->nr_valid_paths++) { |
e54f77dd | 1035 | m->current_pgpath = NULL; |
63d832c3 | 1036 | run_queue = 1; |
e54f77dd | 1037 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1038 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
e54f77dd CS |
1039 | m->pg_init_in_progress++; |
1040 | } | |
1da177e4 | 1041 | |
b15546f9 MA |
1042 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
1043 | pgpath->path.dev->name, m->nr_valid_paths); | |
1044 | ||
fe9cf30e | 1045 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1046 | |
1047 | out: | |
1048 | spin_unlock_irqrestore(&m->lock, flags); | |
63d832c3 HR |
1049 | if (run_queue) |
1050 | dm_table_run_md_queue_async(m->ti->table); | |
1da177e4 LT |
1051 | |
1052 | return r; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1057 | */ | |
1058 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1059 | action_fn action) | |
1060 | { | |
19040c0b | 1061 | int r = -EINVAL; |
1da177e4 LT |
1062 | struct pgpath *pgpath; |
1063 | struct priority_group *pg; | |
1064 | ||
1065 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1066 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1067 | if (pgpath->path.dev == dev) | |
1068 | r = action(pgpath); | |
1069 | } | |
1070 | } | |
1071 | ||
1072 | return r; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * Temporarily try to avoid having to use the specified PG | |
1077 | */ | |
1078 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
be7d31cc | 1079 | bool bypassed) |
1da177e4 LT |
1080 | { |
1081 | unsigned long flags; | |
1082 | ||
1083 | spin_lock_irqsave(&m->lock, flags); | |
1084 | ||
1085 | pg->bypassed = bypassed; | |
1086 | m->current_pgpath = NULL; | |
1087 | m->current_pg = NULL; | |
1088 | ||
1089 | spin_unlock_irqrestore(&m->lock, flags); | |
1090 | ||
fe9cf30e | 1091 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1092 | } |
1093 | ||
1094 | /* | |
1095 | * Switch to using the specified PG from the next I/O that gets mapped | |
1096 | */ | |
1097 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1098 | { | |
1099 | struct priority_group *pg; | |
1100 | unsigned pgnum; | |
1101 | unsigned long flags; | |
31998ef1 | 1102 | char dummy; |
1da177e4 | 1103 | |
31998ef1 | 1104 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1105 | (pgnum > m->nr_priority_groups)) { |
1106 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1107 | return -EINVAL; | |
1108 | } | |
1109 | ||
1110 | spin_lock_irqsave(&m->lock, flags); | |
1111 | list_for_each_entry(pg, &m->priority_groups, list) { | |
be7d31cc | 1112 | pg->bypassed = false; |
1da177e4 LT |
1113 | if (--pgnum) |
1114 | continue; | |
1115 | ||
1116 | m->current_pgpath = NULL; | |
1117 | m->current_pg = NULL; | |
1118 | m->next_pg = pg; | |
1119 | } | |
1120 | spin_unlock_irqrestore(&m->lock, flags); | |
1121 | ||
fe9cf30e | 1122 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1123 | return 0; |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * Set/clear bypassed status of a PG. | |
1128 | * PGs are numbered upwards from 1 in the order they were declared. | |
1129 | */ | |
be7d31cc | 1130 | static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed) |
1da177e4 LT |
1131 | { |
1132 | struct priority_group *pg; | |
1133 | unsigned pgnum; | |
31998ef1 | 1134 | char dummy; |
1da177e4 | 1135 | |
31998ef1 | 1136 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1137 | (pgnum > m->nr_priority_groups)) { |
1138 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1139 | return -EINVAL; | |
1140 | } | |
1141 | ||
1142 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1143 | if (!--pgnum) | |
1144 | break; | |
1145 | } | |
1146 | ||
1147 | bypass_pg(m, pg, bypassed); | |
1148 | return 0; | |
1149 | } | |
1150 | ||
c9e45581 DW |
1151 | /* |
1152 | * Should we retry pg_init immediately? | |
1153 | */ | |
be7d31cc | 1154 | static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) |
c9e45581 DW |
1155 | { |
1156 | unsigned long flags; | |
be7d31cc | 1157 | bool limit_reached = false; |
c9e45581 DW |
1158 | |
1159 | spin_lock_irqsave(&m->lock, flags); | |
1160 | ||
954a73d5 | 1161 | if (m->pg_init_count <= m->pg_init_retries && !m->pg_init_disabled) |
be7d31cc | 1162 | m->pg_init_required = true; |
c9e45581 | 1163 | else |
be7d31cc | 1164 | limit_reached = true; |
c9e45581 DW |
1165 | |
1166 | spin_unlock_irqrestore(&m->lock, flags); | |
1167 | ||
1168 | return limit_reached; | |
1169 | } | |
1170 | ||
3ae31f6a | 1171 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1172 | { |
83c0d5d5 | 1173 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1174 | struct priority_group *pg = pgpath->pg; |
1175 | struct multipath *m = pg->m; | |
1176 | unsigned long flags; | |
be7d31cc | 1177 | bool delay_retry = false; |
cfae5c9b CS |
1178 | |
1179 | /* device or driver problems */ | |
1180 | switch (errors) { | |
1181 | case SCSI_DH_OK: | |
1182 | break; | |
1183 | case SCSI_DH_NOSYS: | |
1184 | if (!m->hw_handler_name) { | |
1185 | errors = 0; | |
1186 | break; | |
1187 | } | |
f7b934c8 MB |
1188 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1189 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1190 | /* |
1191 | * Fail path for now, so we do not ping pong | |
1192 | */ | |
1193 | fail_path(pgpath); | |
1194 | break; | |
1195 | case SCSI_DH_DEV_TEMP_BUSY: | |
1196 | /* | |
1197 | * Probably doing something like FW upgrade on the | |
1198 | * controller so try the other pg. | |
1199 | */ | |
be7d31cc | 1200 | bypass_pg(m, pg, true); |
cfae5c9b | 1201 | break; |
cfae5c9b | 1202 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1203 | /* Wait before retrying. */ |
1204 | delay_retry = 1; | |
cfae5c9b CS |
1205 | case SCSI_DH_IMM_RETRY: |
1206 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1207 | if (pg_init_limit_reached(m, pgpath)) | |
1208 | fail_path(pgpath); | |
1209 | errors = 0; | |
1210 | break; | |
1211 | default: | |
1212 | /* | |
1213 | * We probably do not want to fail the path for a device | |
1214 | * error, but this is what the old dm did. In future | |
1215 | * patches we can do more advanced handling. | |
1216 | */ | |
1217 | fail_path(pgpath); | |
1218 | } | |
1219 | ||
1220 | spin_lock_irqsave(&m->lock, flags); | |
1221 | if (errors) { | |
e54f77dd CS |
1222 | if (pgpath == m->current_pgpath) { |
1223 | DMERR("Could not failover device. Error %d.", errors); | |
1224 | m->current_pgpath = NULL; | |
1225 | m->current_pg = NULL; | |
1226 | } | |
d0259bf0 | 1227 | } else if (!m->pg_init_required) |
be7d31cc | 1228 | pg->bypassed = false; |
cfae5c9b | 1229 | |
d0259bf0 KU |
1230 | if (--m->pg_init_in_progress) |
1231 | /* Activations of other paths are still on going */ | |
1232 | goto out; | |
1233 | ||
3e9f1be1 HR |
1234 | if (m->pg_init_required) { |
1235 | m->pg_init_delay_retry = delay_retry; | |
1236 | if (__pg_init_all_paths(m)) | |
1237 | goto out; | |
1238 | } | |
be7d31cc | 1239 | m->queue_io = false; |
d0259bf0 | 1240 | |
2bded7bd KU |
1241 | /* |
1242 | * Wake up any thread waiting to suspend. | |
1243 | */ | |
1244 | wake_up(&m->pg_init_wait); | |
1245 | ||
d0259bf0 | 1246 | out: |
cfae5c9b CS |
1247 | spin_unlock_irqrestore(&m->lock, flags); |
1248 | } | |
1249 | ||
bab7cfc7 CS |
1250 | static void activate_path(struct work_struct *work) |
1251 | { | |
e54f77dd | 1252 | struct pgpath *pgpath = |
4e2d19e4 | 1253 | container_of(work, struct pgpath, activate_path.work); |
bab7cfc7 | 1254 | |
3a017509 HR |
1255 | if (pgpath->is_active) |
1256 | scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev), | |
1257 | pg_init_done, pgpath); | |
1258 | else | |
1259 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1260 | } |
1261 | ||
7e782af5 HR |
1262 | static int noretry_error(int error) |
1263 | { | |
1264 | switch (error) { | |
1265 | case -EOPNOTSUPP: | |
1266 | case -EREMOTEIO: | |
1267 | case -EILSEQ: | |
1268 | case -ENODATA: | |
cc9d3c38 | 1269 | case -ENOSPC: |
7e782af5 HR |
1270 | return 1; |
1271 | } | |
1272 | ||
1273 | /* Anything else could be a path failure, so should be retried */ | |
1274 | return 0; | |
1275 | } | |
1276 | ||
1da177e4 LT |
1277 | /* |
1278 | * end_io handling | |
1279 | */ | |
f40c67f0 | 1280 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1281 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1282 | { |
f40c67f0 KU |
1283 | /* |
1284 | * We don't queue any clone request inside the multipath target | |
1285 | * during end I/O handling, since those clone requests don't have | |
1286 | * bio clones. If we queue them inside the multipath target, | |
1287 | * we need to make bio clones, that requires memory allocation. | |
1288 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1289 | * don't have bio clones.) | |
1290 | * Instead of queueing the clone request here, we queue the original | |
1291 | * request into dm core, which will remake a clone request and | |
1292 | * clone bios for it and resubmit it later. | |
1293 | */ | |
1294 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1295 | unsigned long flags; |
1da177e4 | 1296 | |
f40c67f0 | 1297 | if (!error && !clone->errors) |
1da177e4 LT |
1298 | return 0; /* I/O complete */ |
1299 | ||
7eee4ae2 | 1300 | if (noretry_error(error)) |
959eb4e5 MS |
1301 | return error; |
1302 | ||
cfae5c9b CS |
1303 | if (mpio->pgpath) |
1304 | fail_path(mpio->pgpath); | |
1da177e4 | 1305 | |
640eb3b0 | 1306 | spin_lock_irqsave(&m->lock, flags); |
751b2a7d HR |
1307 | if (!m->nr_valid_paths) { |
1308 | if (!m->queue_if_no_path) { | |
1309 | if (!__must_push_back(m)) | |
1310 | r = -EIO; | |
1311 | } else { | |
1312 | if (error == -EBADE) | |
1313 | r = error; | |
1314 | } | |
1315 | } | |
640eb3b0 | 1316 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1317 | |
f40c67f0 | 1318 | return r; |
1da177e4 LT |
1319 | } |
1320 | ||
f40c67f0 | 1321 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1322 | int error, union map_info *map_context) |
1323 | { | |
028867ac | 1324 | struct multipath *m = ti->private; |
2eff1924 | 1325 | struct dm_mpath_io *mpio = get_mpio(map_context); |
a71a261f | 1326 | struct pgpath *pgpath; |
1da177e4 LT |
1327 | struct path_selector *ps; |
1328 | int r; | |
1329 | ||
466891f9 JN |
1330 | BUG_ON(!mpio); |
1331 | ||
2eff1924 | 1332 | r = do_end_io(m, clone, error, mpio); |
a71a261f | 1333 | pgpath = mpio->pgpath; |
1da177e4 LT |
1334 | if (pgpath) { |
1335 | ps = &pgpath->pg->ps; | |
1336 | if (ps->type->end_io) | |
02ab823f | 1337 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1338 | } |
2eff1924 | 1339 | clear_request_fn_mpio(m, map_context); |
1da177e4 LT |
1340 | |
1341 | return r; | |
1342 | } | |
1343 | ||
1344 | /* | |
1345 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1346 | * the last path fails we must error any remaining I/O. |
1347 | * Note that if the freeze_bdev fails while suspending, the | |
1348 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1349 | */ |
1350 | static void multipath_presuspend(struct dm_target *ti) | |
1351 | { | |
7943bd6d | 1352 | struct multipath *m = ti->private; |
1da177e4 | 1353 | |
be7d31cc | 1354 | queue_if_no_path(m, false, true); |
1da177e4 LT |
1355 | } |
1356 | ||
6df400ab KU |
1357 | static void multipath_postsuspend(struct dm_target *ti) |
1358 | { | |
6380f26f MA |
1359 | struct multipath *m = ti->private; |
1360 | ||
1361 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1362 | flush_multipath_work(m); |
6380f26f | 1363 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1364 | } |
1365 | ||
436d4108 AK |
1366 | /* |
1367 | * Restore the queue_if_no_path setting. | |
1368 | */ | |
1da177e4 LT |
1369 | static void multipath_resume(struct dm_target *ti) |
1370 | { | |
7943bd6d | 1371 | struct multipath *m = ti->private; |
1da177e4 LT |
1372 | unsigned long flags; |
1373 | ||
1374 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1375 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1376 | spin_unlock_irqrestore(&m->lock, flags); |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * Info output has the following format: | |
1381 | * num_multipath_feature_args [multipath_feature_args]* | |
1382 | * num_handler_status_args [handler_status_args]* | |
1383 | * num_groups init_group_number | |
1384 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1385 | * num_paths num_selector_args | |
1386 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1387 | * | |
1388 | * Table output has the following format (identical to the constructor string): | |
1389 | * num_feature_args [features_args]* | |
1390 | * num_handler_args hw_handler [hw_handler_args]* | |
1391 | * num_groups init_group_number | |
1392 | * [priority selector-name num_ps_args [ps_args]* | |
1393 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1394 | */ | |
fd7c092e MP |
1395 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1396 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1397 | { |
1398 | int sz = 0; | |
1399 | unsigned long flags; | |
7943bd6d | 1400 | struct multipath *m = ti->private; |
1da177e4 LT |
1401 | struct priority_group *pg; |
1402 | struct pgpath *p; | |
1403 | unsigned pg_num; | |
1404 | char state; | |
1405 | ||
1406 | spin_lock_irqsave(&m->lock, flags); | |
1407 | ||
1408 | /* Features */ | |
1409 | if (type == STATUSTYPE_INFO) | |
e8099177 | 1410 | DMEMIT("2 %u %u ", m->queue_io, m->pg_init_count); |
c9e45581 DW |
1411 | else { |
1412 | DMEMIT("%u ", m->queue_if_no_path + | |
4e2d19e4 | 1413 | (m->pg_init_retries > 0) * 2 + |
a58a935d MS |
1414 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
1415 | m->retain_attached_hw_handler); | |
c9e45581 DW |
1416 | if (m->queue_if_no_path) |
1417 | DMEMIT("queue_if_no_path "); | |
1418 | if (m->pg_init_retries) | |
1419 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1420 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1421 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
a58a935d MS |
1422 | if (m->retain_attached_hw_handler) |
1423 | DMEMIT("retain_attached_hw_handler "); | |
c9e45581 | 1424 | } |
1da177e4 | 1425 | |
cfae5c9b | 1426 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1427 | DMEMIT("0 "); |
1428 | else | |
cfae5c9b | 1429 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1430 | |
1431 | DMEMIT("%u ", m->nr_priority_groups); | |
1432 | ||
1433 | if (m->next_pg) | |
1434 | pg_num = m->next_pg->pg_num; | |
1435 | else if (m->current_pg) | |
1436 | pg_num = m->current_pg->pg_num; | |
1437 | else | |
a490a07a | 1438 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1439 | |
1440 | DMEMIT("%u ", pg_num); | |
1441 | ||
1442 | switch (type) { | |
1443 | case STATUSTYPE_INFO: | |
1444 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1445 | if (pg->bypassed) | |
1446 | state = 'D'; /* Disabled */ | |
1447 | else if (pg == m->current_pg) | |
1448 | state = 'A'; /* Currently Active */ | |
1449 | else | |
1450 | state = 'E'; /* Enabled */ | |
1451 | ||
1452 | DMEMIT("%c ", state); | |
1453 | ||
1454 | if (pg->ps.type->status) | |
1455 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1456 | result + sz, | |
1457 | maxlen - sz); | |
1458 | else | |
1459 | DMEMIT("0 "); | |
1460 | ||
1461 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1462 | pg->ps.type->info_args); | |
1463 | ||
1464 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1465 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1466 | p->is_active ? "A" : "F", |
1da177e4 LT |
1467 | p->fail_count); |
1468 | if (pg->ps.type->status) | |
1469 | sz += pg->ps.type->status(&pg->ps, | |
1470 | &p->path, type, result + sz, | |
1471 | maxlen - sz); | |
1472 | } | |
1473 | } | |
1474 | break; | |
1475 | ||
1476 | case STATUSTYPE_TABLE: | |
1477 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1478 | DMEMIT("%s ", pg->ps.type->name); | |
1479 | ||
1480 | if (pg->ps.type->status) | |
1481 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1482 | result + sz, | |
1483 | maxlen - sz); | |
1484 | else | |
1485 | DMEMIT("0 "); | |
1486 | ||
1487 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1488 | pg->ps.type->table_args); | |
1489 | ||
1490 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1491 | DMEMIT("%s ", p->path.dev->name); | |
1492 | if (pg->ps.type->status) | |
1493 | sz += pg->ps.type->status(&pg->ps, | |
1494 | &p->path, type, result + sz, | |
1495 | maxlen - sz); | |
1496 | } | |
1497 | } | |
1498 | break; | |
1499 | } | |
1500 | ||
1501 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1502 | } |
1503 | ||
1504 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1505 | { | |
6380f26f | 1506 | int r = -EINVAL; |
1da177e4 | 1507 | struct dm_dev *dev; |
7943bd6d | 1508 | struct multipath *m = ti->private; |
1da177e4 LT |
1509 | action_fn action; |
1510 | ||
6380f26f MA |
1511 | mutex_lock(&m->work_mutex); |
1512 | ||
c2f3d24b KU |
1513 | if (dm_suspended(ti)) { |
1514 | r = -EBUSY; | |
1515 | goto out; | |
1516 | } | |
1517 | ||
1da177e4 | 1518 | if (argc == 1) { |
498f0103 | 1519 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
be7d31cc | 1520 | r = queue_if_no_path(m, true, false); |
6380f26f | 1521 | goto out; |
498f0103 | 1522 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
be7d31cc | 1523 | r = queue_if_no_path(m, false, false); |
6380f26f MA |
1524 | goto out; |
1525 | } | |
1da177e4 LT |
1526 | } |
1527 | ||
6380f26f | 1528 | if (argc != 2) { |
a356e426 | 1529 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1530 | goto out; |
1531 | } | |
1da177e4 | 1532 | |
498f0103 | 1533 | if (!strcasecmp(argv[0], "disable_group")) { |
be7d31cc | 1534 | r = bypass_pg_num(m, argv[1], true); |
6380f26f | 1535 | goto out; |
498f0103 | 1536 | } else if (!strcasecmp(argv[0], "enable_group")) { |
be7d31cc | 1537 | r = bypass_pg_num(m, argv[1], false); |
6380f26f | 1538 | goto out; |
498f0103 | 1539 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1540 | r = switch_pg_num(m, argv[1]); |
1541 | goto out; | |
498f0103 | 1542 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1543 | action = reinstate_path; |
498f0103 | 1544 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1545 | action = fail_path; |
6380f26f | 1546 | else { |
a356e426 | 1547 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1548 | goto out; |
1549 | } | |
1da177e4 | 1550 | |
8215d6ec | 1551 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1552 | if (r) { |
72d94861 | 1553 | DMWARN("message: error getting device %s", |
1da177e4 | 1554 | argv[1]); |
6380f26f | 1555 | goto out; |
1da177e4 LT |
1556 | } |
1557 | ||
1558 | r = action_dev(m, dev, action); | |
1559 | ||
1560 | dm_put_device(ti, dev); | |
1561 | ||
6380f26f MA |
1562 | out: |
1563 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1564 | return r; |
1da177e4 LT |
1565 | } |
1566 | ||
e56f81e0 CH |
1567 | static int multipath_prepare_ioctl(struct dm_target *ti, |
1568 | struct block_device **bdev, fmode_t *mode) | |
9af4aa30 | 1569 | { |
35991652 | 1570 | struct multipath *m = ti->private; |
9af4aa30 | 1571 | unsigned long flags; |
35991652 MP |
1572 | int r; |
1573 | ||
9af4aa30 MB |
1574 | spin_lock_irqsave(&m->lock, flags); |
1575 | ||
1576 | if (!m->current_pgpath) | |
02ab823f | 1577 | __choose_pgpath(m, 0); |
9af4aa30 | 1578 | |
43e43c9e JN |
1579 | if (m->current_pgpath) { |
1580 | if (!m->queue_io) { | |
1581 | *bdev = m->current_pgpath->path.dev->bdev; | |
1582 | *mode = m->current_pgpath->path.dev->mode; | |
1583 | r = 0; | |
1584 | } else { | |
1585 | /* pg_init has not started or completed */ | |
1586 | r = -ENOTCONN; | |
1587 | } | |
1588 | } else { | |
1589 | /* No path is available */ | |
1590 | if (m->queue_if_no_path) | |
1591 | r = -ENOTCONN; | |
1592 | else | |
1593 | r = -EIO; | |
e90dae1f | 1594 | } |
9af4aa30 | 1595 | |
9af4aa30 MB |
1596 | spin_unlock_irqrestore(&m->lock, flags); |
1597 | ||
5bbbfdf6 | 1598 | if (r == -ENOTCONN) { |
3e9f1be1 HR |
1599 | spin_lock_irqsave(&m->lock, flags); |
1600 | if (!m->current_pg) { | |
1601 | /* Path status changed, redo selection */ | |
1602 | __choose_pgpath(m, 0); | |
1603 | } | |
1604 | if (m->pg_init_required) | |
1605 | __pg_init_all_paths(m); | |
4cdd2ad7 | 1606 | spin_unlock_irqrestore(&m->lock, flags); |
63d832c3 | 1607 | dm_table_run_md_queue_async(m->ti->table); |
3e9f1be1 | 1608 | } |
35991652 | 1609 | |
e56f81e0 CH |
1610 | /* |
1611 | * Only pass ioctls through if the device sizes match exactly. | |
1612 | */ | |
1613 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
1614 | return 1; | |
1615 | return r; | |
9af4aa30 MB |
1616 | } |
1617 | ||
af4874e0 MS |
1618 | static int multipath_iterate_devices(struct dm_target *ti, |
1619 | iterate_devices_callout_fn fn, void *data) | |
1620 | { | |
1621 | struct multipath *m = ti->private; | |
1622 | struct priority_group *pg; | |
1623 | struct pgpath *p; | |
1624 | int ret = 0; | |
1625 | ||
1626 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1627 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1628 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1629 | if (ret) |
1630 | goto out; | |
1631 | } | |
1632 | } | |
1633 | ||
1634 | out: | |
1635 | return ret; | |
1636 | } | |
1637 | ||
f40c67f0 KU |
1638 | static int __pgpath_busy(struct pgpath *pgpath) |
1639 | { | |
1640 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1641 | ||
52b09914 | 1642 | return blk_lld_busy(q); |
f40c67f0 KU |
1643 | } |
1644 | ||
1645 | /* | |
1646 | * We return "busy", only when we can map I/Os but underlying devices | |
1647 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1648 | * the underlying queue). | |
1649 | * In other words, if we want to kill I/Os or queue them inside us | |
1650 | * due to map unavailability, we don't return "busy". Otherwise, | |
1651 | * dm core won't give us the I/Os and we can't do what we want. | |
1652 | */ | |
1653 | static int multipath_busy(struct dm_target *ti) | |
1654 | { | |
be7d31cc | 1655 | bool busy = false, has_active = false; |
f40c67f0 KU |
1656 | struct multipath *m = ti->private; |
1657 | struct priority_group *pg; | |
1658 | struct pgpath *pgpath; | |
1659 | unsigned long flags; | |
1660 | ||
1661 | spin_lock_irqsave(&m->lock, flags); | |
1662 | ||
7a7a3b45 JN |
1663 | /* pg_init in progress or no paths available */ |
1664 | if (m->pg_init_in_progress || | |
1665 | (!m->nr_valid_paths && m->queue_if_no_path)) { | |
be7d31cc | 1666 | busy = true; |
b63349a7 HR |
1667 | goto out; |
1668 | } | |
f40c67f0 KU |
1669 | /* Guess which priority_group will be used at next mapping time */ |
1670 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1671 | pg = m->next_pg; | |
1672 | else if (likely(m->current_pg)) | |
1673 | pg = m->current_pg; | |
1674 | else | |
1675 | /* | |
1676 | * We don't know which pg will be used at next mapping time. | |
1677 | * We don't call __choose_pgpath() here to avoid to trigger | |
1678 | * pg_init just by busy checking. | |
1679 | * So we don't know whether underlying devices we will be using | |
1680 | * at next mapping time are busy or not. Just try mapping. | |
1681 | */ | |
1682 | goto out; | |
1683 | ||
1684 | /* | |
1685 | * If there is one non-busy active path at least, the path selector | |
1686 | * will be able to select it. So we consider such a pg as not busy. | |
1687 | */ | |
be7d31cc | 1688 | busy = true; |
f40c67f0 KU |
1689 | list_for_each_entry(pgpath, &pg->pgpaths, list) |
1690 | if (pgpath->is_active) { | |
be7d31cc | 1691 | has_active = true; |
f40c67f0 KU |
1692 | |
1693 | if (!__pgpath_busy(pgpath)) { | |
be7d31cc | 1694 | busy = false; |
f40c67f0 KU |
1695 | break; |
1696 | } | |
1697 | } | |
1698 | ||
1699 | if (!has_active) | |
1700 | /* | |
1701 | * No active path in this pg, so this pg won't be used and | |
1702 | * the current_pg will be changed at next mapping time. | |
1703 | * We need to try mapping to determine it. | |
1704 | */ | |
be7d31cc | 1705 | busy = false; |
f40c67f0 KU |
1706 | |
1707 | out: | |
1708 | spin_unlock_irqrestore(&m->lock, flags); | |
1709 | ||
1710 | return busy; | |
1711 | } | |
1712 | ||
1da177e4 LT |
1713 | /*----------------------------------------------------------------- |
1714 | * Module setup | |
1715 | *---------------------------------------------------------------*/ | |
1716 | static struct target_type multipath_target = { | |
1717 | .name = "multipath", | |
16f12266 MS |
1718 | .version = {1, 11, 0}, |
1719 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE, | |
1da177e4 LT |
1720 | .module = THIS_MODULE, |
1721 | .ctr = multipath_ctr, | |
1722 | .dtr = multipath_dtr, | |
f40c67f0 | 1723 | .map_rq = multipath_map, |
e5863d9a MS |
1724 | .clone_and_map_rq = multipath_clone_and_map, |
1725 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 1726 | .rq_end_io = multipath_end_io, |
1da177e4 | 1727 | .presuspend = multipath_presuspend, |
6df400ab | 1728 | .postsuspend = multipath_postsuspend, |
1da177e4 LT |
1729 | .resume = multipath_resume, |
1730 | .status = multipath_status, | |
1731 | .message = multipath_message, | |
e56f81e0 | 1732 | .prepare_ioctl = multipath_prepare_ioctl, |
af4874e0 | 1733 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1734 | .busy = multipath_busy, |
1da177e4 LT |
1735 | }; |
1736 | ||
1737 | static int __init dm_multipath_init(void) | |
1738 | { | |
1739 | int r; | |
1740 | ||
1741 | /* allocate a slab for the dm_ios */ | |
028867ac | 1742 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1743 | if (!_mpio_cache) |
1744 | return -ENOMEM; | |
1745 | ||
1746 | r = dm_register_target(&multipath_target); | |
1747 | if (r < 0) { | |
0cd33124 | 1748 | DMERR("register failed %d", r); |
ff658e9c JT |
1749 | r = -EINVAL; |
1750 | goto bad_register_target; | |
1da177e4 LT |
1751 | } |
1752 | ||
4d4d66ab | 1753 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1754 | if (!kmultipathd) { |
0cd33124 | 1755 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
1756 | r = -ENOMEM; |
1757 | goto bad_alloc_kmultipathd; | |
c557308e AK |
1758 | } |
1759 | ||
bab7cfc7 CS |
1760 | /* |
1761 | * A separate workqueue is used to handle the device handlers | |
1762 | * to avoid overloading existing workqueue. Overloading the | |
1763 | * old workqueue would also create a bottleneck in the | |
1764 | * path of the storage hardware device activation. | |
1765 | */ | |
4d4d66ab TH |
1766 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1767 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1768 | if (!kmpath_handlerd) { |
1769 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
1770 | r = -ENOMEM; |
1771 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
1772 | } |
1773 | ||
72d94861 | 1774 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1775 | multipath_target.version[0], multipath_target.version[1], |
1776 | multipath_target.version[2]); | |
1777 | ||
ff658e9c JT |
1778 | return 0; |
1779 | ||
1780 | bad_alloc_kmpath_handlerd: | |
1781 | destroy_workqueue(kmultipathd); | |
1782 | bad_alloc_kmultipathd: | |
1783 | dm_unregister_target(&multipath_target); | |
1784 | bad_register_target: | |
1785 | kmem_cache_destroy(_mpio_cache); | |
1786 | ||
1da177e4 LT |
1787 | return r; |
1788 | } | |
1789 | ||
1790 | static void __exit dm_multipath_exit(void) | |
1791 | { | |
bab7cfc7 | 1792 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1793 | destroy_workqueue(kmultipathd); |
1794 | ||
10d3bd09 | 1795 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1796 | kmem_cache_destroy(_mpio_cache); |
1797 | } | |
1798 | ||
1da177e4 LT |
1799 | module_init(dm_multipath_init); |
1800 | module_exit(dm_multipath_exit); | |
1801 | ||
1802 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1803 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1804 | MODULE_LICENSE("GPL"); |