Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software Limited. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
586e80e6 MP |
8 | #include <linux/device-mapper.h> |
9 | ||
1da177e4 | 10 | #include "dm-path-selector.h" |
b15546f9 | 11 | #include "dm-uevent.h" |
1da177e4 LT |
12 | |
13 | #include <linux/ctype.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/mempool.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/time.h> | |
20 | #include <linux/workqueue.h> | |
cfae5c9b | 21 | #include <scsi/scsi_dh.h> |
60063497 | 22 | #include <linux/atomic.h> |
1da177e4 | 23 | |
72d94861 | 24 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
25 | #define DM_PG_INIT_DELAY_MSECS 2000 |
26 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
1da177e4 LT |
27 | |
28 | /* Path properties */ | |
29 | struct pgpath { | |
30 | struct list_head list; | |
31 | ||
32 | struct priority_group *pg; /* Owning PG */ | |
6680073d | 33 | unsigned is_active; /* Path status */ |
1da177e4 LT |
34 | unsigned fail_count; /* Cumulative failure count */ |
35 | ||
c922d5f7 | 36 | struct dm_path path; |
4e2d19e4 | 37 | struct delayed_work activate_path; |
1da177e4 LT |
38 | }; |
39 | ||
40 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
41 | ||
42 | /* | |
43 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
44 | * Each has a path selector which controls which path gets used. | |
45 | */ | |
46 | struct priority_group { | |
47 | struct list_head list; | |
48 | ||
49 | struct multipath *m; /* Owning multipath instance */ | |
50 | struct path_selector ps; | |
51 | ||
52 | unsigned pg_num; /* Reference number */ | |
53 | unsigned bypassed; /* Temporarily bypass this PG? */ | |
54 | ||
55 | unsigned nr_pgpaths; /* Number of paths in PG */ | |
56 | struct list_head pgpaths; | |
57 | }; | |
58 | ||
59 | /* Multipath context */ | |
60 | struct multipath { | |
61 | struct list_head list; | |
62 | struct dm_target *ti; | |
63 | ||
64 | spinlock_t lock; | |
65 | ||
cfae5c9b | 66 | const char *hw_handler_name; |
2bfd2e13 | 67 | char *hw_handler_params; |
4e2d19e4 | 68 | |
1da177e4 LT |
69 | unsigned nr_priority_groups; |
70 | struct list_head priority_groups; | |
4e2d19e4 CS |
71 | |
72 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
73 | ||
1da177e4 | 74 | unsigned pg_init_required; /* pg_init needs calling? */ |
c3cd4f6b | 75 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
4e2d19e4 | 76 | unsigned pg_init_delay_retry; /* Delay pg_init retry? */ |
1da177e4 LT |
77 | |
78 | unsigned nr_valid_paths; /* Total number of usable paths */ | |
79 | struct pgpath *current_pgpath; | |
80 | struct priority_group *current_pg; | |
81 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
82 | unsigned repeat_count; /* I/Os left before calling PS again */ | |
83 | ||
84 | unsigned queue_io; /* Must we queue all I/O? */ | |
85 | unsigned queue_if_no_path; /* Queue I/O if last path fails? */ | |
436d4108 | 86 | unsigned saved_queue_if_no_path;/* Saved state during suspension */ |
c9e45581 DW |
87 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
88 | unsigned pg_init_count; /* Number of times pg_init called */ | |
4e2d19e4 | 89 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
1da177e4 LT |
90 | |
91 | struct work_struct process_queued_ios; | |
f40c67f0 | 92 | struct list_head queued_ios; |
1da177e4 LT |
93 | unsigned queue_size; |
94 | ||
95 | struct work_struct trigger_event; | |
96 | ||
97 | /* | |
028867ac | 98 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
99 | * can resubmit bios on error. |
100 | */ | |
101 | mempool_t *mpio_pool; | |
6380f26f MA |
102 | |
103 | struct mutex work_mutex; | |
1da177e4 LT |
104 | }; |
105 | ||
106 | /* | |
107 | * Context information attached to each bio we process. | |
108 | */ | |
028867ac | 109 | struct dm_mpath_io { |
1da177e4 | 110 | struct pgpath *pgpath; |
02ab823f | 111 | size_t nr_bytes; |
1da177e4 LT |
112 | }; |
113 | ||
114 | typedef int (*action_fn) (struct pgpath *pgpath); | |
115 | ||
116 | #define MIN_IOS 256 /* Mempool size */ | |
117 | ||
e18b890b | 118 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 119 | |
bab7cfc7 | 120 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 DH |
121 | static void process_queued_ios(struct work_struct *work); |
122 | static void trigger_event(struct work_struct *work); | |
bab7cfc7 | 123 | static void activate_path(struct work_struct *work); |
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; | |
ae11b1b3 | 162 | struct multipath *m = ti->private; |
1da177e4 LT |
163 | |
164 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
165 | list_del(&pgpath->list); | |
ae11b1b3 HR |
166 | if (m->hw_handler_name) |
167 | scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev)); | |
1da177e4 LT |
168 | dm_put_device(ti, pgpath->path.dev); |
169 | free_pgpath(pgpath); | |
170 | } | |
171 | } | |
172 | ||
173 | static void free_priority_group(struct priority_group *pg, | |
174 | struct dm_target *ti) | |
175 | { | |
176 | struct path_selector *ps = &pg->ps; | |
177 | ||
178 | if (ps->type) { | |
179 | ps->type->destroy(ps); | |
180 | dm_put_path_selector(ps->type); | |
181 | } | |
182 | ||
183 | free_pgpaths(&pg->pgpaths, ti); | |
184 | kfree(pg); | |
185 | } | |
186 | ||
28f16c20 | 187 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
188 | { |
189 | struct multipath *m; | |
190 | ||
e69fae56 | 191 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 192 | if (m) { |
1da177e4 | 193 | INIT_LIST_HEAD(&m->priority_groups); |
f40c67f0 | 194 | INIT_LIST_HEAD(&m->queued_ios); |
1da177e4 LT |
195 | spin_lock_init(&m->lock); |
196 | m->queue_io = 1; | |
4e2d19e4 | 197 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; |
c4028958 DH |
198 | INIT_WORK(&m->process_queued_ios, process_queued_ios); |
199 | INIT_WORK(&m->trigger_event, trigger_event); | |
2bded7bd | 200 | init_waitqueue_head(&m->pg_init_wait); |
6380f26f | 201 | mutex_init(&m->work_mutex); |
93d2341c | 202 | m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache); |
1da177e4 LT |
203 | if (!m->mpio_pool) { |
204 | kfree(m); | |
205 | return NULL; | |
206 | } | |
28f16c20 MM |
207 | m->ti = ti; |
208 | ti->private = m; | |
1da177e4 LT |
209 | } |
210 | ||
211 | return m; | |
212 | } | |
213 | ||
214 | static void free_multipath(struct multipath *m) | |
215 | { | |
216 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
217 | |
218 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
219 | list_del(&pg->list); | |
220 | free_priority_group(pg, m->ti); | |
221 | } | |
222 | ||
cfae5c9b | 223 | kfree(m->hw_handler_name); |
2bfd2e13 | 224 | kfree(m->hw_handler_params); |
1da177e4 LT |
225 | mempool_destroy(m->mpio_pool); |
226 | kfree(m); | |
227 | } | |
228 | ||
229 | ||
230 | /*----------------------------------------------- | |
231 | * Path selection | |
232 | *-----------------------------------------------*/ | |
233 | ||
fb612642 KU |
234 | static void __pg_init_all_paths(struct multipath *m) |
235 | { | |
236 | struct pgpath *pgpath; | |
4e2d19e4 | 237 | unsigned long pg_init_delay = 0; |
fb612642 KU |
238 | |
239 | m->pg_init_count++; | |
240 | m->pg_init_required = 0; | |
4e2d19e4 CS |
241 | if (m->pg_init_delay_retry) |
242 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? | |
243 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
244 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
245 | /* Skip failed paths */ | |
246 | if (!pgpath->is_active) | |
247 | continue; | |
4e2d19e4 CS |
248 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
249 | pg_init_delay)) | |
fb612642 KU |
250 | m->pg_init_in_progress++; |
251 | } | |
252 | } | |
253 | ||
1da177e4 LT |
254 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) |
255 | { | |
1da177e4 LT |
256 | m->current_pg = pgpath->pg; |
257 | ||
258 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 259 | if (m->hw_handler_name) { |
1da177e4 LT |
260 | m->pg_init_required = 1; |
261 | m->queue_io = 1; | |
262 | } else { | |
263 | m->pg_init_required = 0; | |
264 | m->queue_io = 0; | |
265 | } | |
c9e45581 DW |
266 | |
267 | m->pg_init_count = 0; | |
1da177e4 LT |
268 | } |
269 | ||
02ab823f KU |
270 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
271 | size_t nr_bytes) | |
1da177e4 | 272 | { |
c922d5f7 | 273 | struct dm_path *path; |
1da177e4 | 274 | |
02ab823f | 275 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes); |
1da177e4 LT |
276 | if (!path) |
277 | return -ENXIO; | |
278 | ||
279 | m->current_pgpath = path_to_pgpath(path); | |
280 | ||
281 | if (m->current_pg != pg) | |
282 | __switch_pg(m, m->current_pgpath); | |
283 | ||
284 | return 0; | |
285 | } | |
286 | ||
02ab823f | 287 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
288 | { |
289 | struct priority_group *pg; | |
290 | unsigned bypassed = 1; | |
291 | ||
292 | if (!m->nr_valid_paths) | |
293 | goto failed; | |
294 | ||
295 | /* Were we instructed to switch PG? */ | |
296 | if (m->next_pg) { | |
297 | pg = m->next_pg; | |
298 | m->next_pg = NULL; | |
02ab823f | 299 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
300 | return; |
301 | } | |
302 | ||
303 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 304 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
305 | return; |
306 | ||
307 | /* | |
308 | * Loop through priority groups until we find a valid path. | |
309 | * First time we skip PGs marked 'bypassed'. | |
310 | * Second time we only try the ones we skipped. | |
311 | */ | |
312 | do { | |
313 | list_for_each_entry(pg, &m->priority_groups, list) { | |
314 | if (pg->bypassed == bypassed) | |
315 | continue; | |
02ab823f | 316 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
317 | return; |
318 | } | |
319 | } while (bypassed--); | |
320 | ||
321 | failed: | |
322 | m->current_pgpath = NULL; | |
323 | m->current_pg = NULL; | |
324 | } | |
325 | ||
45e15720 KU |
326 | /* |
327 | * Check whether bios must be queued in the device-mapper core rather | |
328 | * than here in the target. | |
329 | * | |
330 | * m->lock must be held on entry. | |
331 | * | |
332 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
333 | * same value then we are not between multipath_presuspend() | |
334 | * and multipath_resume() calls and we have no need to check | |
335 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
336 | */ | |
337 | static int __must_push_back(struct multipath *m) | |
338 | { | |
339 | return (m->queue_if_no_path != m->saved_queue_if_no_path && | |
340 | dm_noflush_suspending(m->ti)); | |
341 | } | |
342 | ||
f40c67f0 | 343 | static int map_io(struct multipath *m, struct request *clone, |
028867ac | 344 | struct dm_mpath_io *mpio, unsigned was_queued) |
1da177e4 | 345 | { |
d2a7ad29 | 346 | int r = DM_MAPIO_REMAPPED; |
f40c67f0 | 347 | size_t nr_bytes = blk_rq_bytes(clone); |
1da177e4 LT |
348 | unsigned long flags; |
349 | struct pgpath *pgpath; | |
f40c67f0 | 350 | struct block_device *bdev; |
1da177e4 LT |
351 | |
352 | spin_lock_irqsave(&m->lock, flags); | |
353 | ||
354 | /* Do we need to select a new pgpath? */ | |
355 | if (!m->current_pgpath || | |
356 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
02ab823f | 357 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
358 | |
359 | pgpath = m->current_pgpath; | |
360 | ||
361 | if (was_queued) | |
362 | m->queue_size--; | |
363 | ||
364 | if ((pgpath && m->queue_io) || | |
436d4108 | 365 | (!pgpath && m->queue_if_no_path)) { |
1da177e4 | 366 | /* Queue for the daemon to resubmit */ |
f40c67f0 | 367 | list_add_tail(&clone->queuelist, &m->queued_ios); |
1da177e4 | 368 | m->queue_size++; |
c3cd4f6b AK |
369 | if ((m->pg_init_required && !m->pg_init_in_progress) || |
370 | !m->queue_io) | |
c557308e | 371 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 372 | pgpath = NULL; |
d2a7ad29 | 373 | r = DM_MAPIO_SUBMITTED; |
f40c67f0 KU |
374 | } else if (pgpath) { |
375 | bdev = pgpath->path.dev->bdev; | |
376 | clone->q = bdev_get_queue(bdev); | |
377 | clone->rq_disk = bdev->bd_disk; | |
378 | } else if (__must_push_back(m)) | |
45e15720 KU |
379 | r = DM_MAPIO_REQUEUE; |
380 | else | |
381 | r = -EIO; /* Failed */ | |
1da177e4 LT |
382 | |
383 | mpio->pgpath = pgpath; | |
02ab823f KU |
384 | mpio->nr_bytes = nr_bytes; |
385 | ||
386 | if (r == DM_MAPIO_REMAPPED && pgpath->pg->ps.type->start_io) | |
387 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, &pgpath->path, | |
388 | nr_bytes); | |
1da177e4 LT |
389 | |
390 | spin_unlock_irqrestore(&m->lock, flags); | |
391 | ||
392 | return r; | |
393 | } | |
394 | ||
395 | /* | |
396 | * If we run out of usable paths, should we queue I/O or error it? | |
397 | */ | |
485ef69e AK |
398 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
399 | unsigned save_old_value) | |
1da177e4 LT |
400 | { |
401 | unsigned long flags; | |
402 | ||
403 | spin_lock_irqsave(&m->lock, flags); | |
404 | ||
485ef69e AK |
405 | if (save_old_value) |
406 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
407 | else | |
408 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 409 | m->queue_if_no_path = queue_if_no_path; |
c3cd4f6b | 410 | if (!m->queue_if_no_path && m->queue_size) |
c557308e | 411 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 LT |
412 | |
413 | spin_unlock_irqrestore(&m->lock, flags); | |
414 | ||
415 | return 0; | |
416 | } | |
417 | ||
418 | /*----------------------------------------------------------------- | |
419 | * The multipath daemon is responsible for resubmitting queued ios. | |
420 | *---------------------------------------------------------------*/ | |
421 | ||
422 | static void dispatch_queued_ios(struct multipath *m) | |
423 | { | |
424 | int r; | |
425 | unsigned long flags; | |
028867ac | 426 | struct dm_mpath_io *mpio; |
1da177e4 | 427 | union map_info *info; |
f40c67f0 KU |
428 | struct request *clone, *n; |
429 | LIST_HEAD(cl); | |
1da177e4 LT |
430 | |
431 | spin_lock_irqsave(&m->lock, flags); | |
f40c67f0 | 432 | list_splice_init(&m->queued_ios, &cl); |
1da177e4 LT |
433 | spin_unlock_irqrestore(&m->lock, flags); |
434 | ||
f40c67f0 KU |
435 | list_for_each_entry_safe(clone, n, &cl, queuelist) { |
436 | list_del_init(&clone->queuelist); | |
1da177e4 | 437 | |
f40c67f0 | 438 | info = dm_get_rq_mapinfo(clone); |
1da177e4 LT |
439 | mpio = info->ptr; |
440 | ||
f40c67f0 KU |
441 | r = map_io(m, clone, mpio, 1); |
442 | if (r < 0) { | |
443 | mempool_free(mpio, m->mpio_pool); | |
444 | dm_kill_unmapped_request(clone, r); | |
445 | } else if (r == DM_MAPIO_REMAPPED) | |
446 | dm_dispatch_request(clone); | |
447 | else if (r == DM_MAPIO_REQUEUE) { | |
448 | mempool_free(mpio, m->mpio_pool); | |
449 | dm_requeue_unmapped_request(clone); | |
450 | } | |
1da177e4 LT |
451 | } |
452 | } | |
453 | ||
c4028958 | 454 | static void process_queued_ios(struct work_struct *work) |
1da177e4 | 455 | { |
c4028958 DH |
456 | struct multipath *m = |
457 | container_of(work, struct multipath, process_queued_ios); | |
fb612642 | 458 | struct pgpath *pgpath = NULL; |
e54f77dd | 459 | unsigned must_queue = 1; |
1da177e4 LT |
460 | unsigned long flags; |
461 | ||
462 | spin_lock_irqsave(&m->lock, flags); | |
463 | ||
c3cd4f6b AK |
464 | if (!m->queue_size) |
465 | goto out; | |
466 | ||
1da177e4 | 467 | if (!m->current_pgpath) |
02ab823f | 468 | __choose_pgpath(m, 0); |
1da177e4 LT |
469 | |
470 | pgpath = m->current_pgpath; | |
471 | ||
c3cd4f6b AK |
472 | if ((pgpath && !m->queue_io) || |
473 | (!pgpath && !m->queue_if_no_path)) | |
474 | must_queue = 0; | |
1da177e4 | 475 | |
fb612642 KU |
476 | if (m->pg_init_required && !m->pg_init_in_progress && pgpath) |
477 | __pg_init_all_paths(m); | |
478 | ||
c3cd4f6b | 479 | out: |
1da177e4 | 480 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
481 | if (!must_queue) |
482 | dispatch_queued_ios(m); | |
483 | } | |
484 | ||
485 | /* | |
486 | * An event is triggered whenever a path is taken out of use. | |
487 | * Includes path failure and PG bypass. | |
488 | */ | |
c4028958 | 489 | static void trigger_event(struct work_struct *work) |
1da177e4 | 490 | { |
c4028958 DH |
491 | struct multipath *m = |
492 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
493 | |
494 | dm_table_event(m->ti->table); | |
495 | } | |
496 | ||
497 | /*----------------------------------------------------------------- | |
498 | * Constructor/argument parsing: | |
499 | * <#multipath feature args> [<arg>]* | |
500 | * <#hw_handler args> [hw_handler [<arg>]*] | |
501 | * <#priority groups> | |
502 | * <initial priority group> | |
503 | * [<selector> <#selector args> [<arg>]* | |
504 | * <#paths> <#per-path selector args> | |
505 | * [<path> [<arg>]* ]+ ]+ | |
506 | *---------------------------------------------------------------*/ | |
498f0103 | 507 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
508 | struct dm_target *ti) |
509 | { | |
510 | int r; | |
511 | struct path_selector_type *pst; | |
512 | unsigned ps_argc; | |
513 | ||
498f0103 | 514 | static struct dm_arg _args[] = { |
72d94861 | 515 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
516 | }; |
517 | ||
498f0103 | 518 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 519 | if (!pst) { |
72d94861 | 520 | ti->error = "unknown path selector type"; |
1da177e4 LT |
521 | return -EINVAL; |
522 | } | |
523 | ||
498f0103 | 524 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
525 | if (r) { |
526 | dm_put_path_selector(pst); | |
1da177e4 | 527 | return -EINVAL; |
371b2e34 | 528 | } |
1da177e4 LT |
529 | |
530 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
531 | if (r) { | |
532 | dm_put_path_selector(pst); | |
72d94861 | 533 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
534 | return r; |
535 | } | |
536 | ||
537 | pg->ps.type = pst; | |
498f0103 | 538 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
539 | |
540 | return 0; | |
541 | } | |
542 | ||
498f0103 | 543 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
544 | struct dm_target *ti) |
545 | { | |
546 | int r; | |
547 | struct pgpath *p; | |
ae11b1b3 | 548 | struct multipath *m = ti->private; |
1da177e4 LT |
549 | |
550 | /* we need at least a path arg */ | |
551 | if (as->argc < 1) { | |
72d94861 | 552 | ti->error = "no device given"; |
01460f35 | 553 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
554 | } |
555 | ||
556 | p = alloc_pgpath(); | |
557 | if (!p) | |
01460f35 | 558 | return ERR_PTR(-ENOMEM); |
1da177e4 | 559 | |
498f0103 | 560 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 561 | &p->path.dev); |
1da177e4 | 562 | if (r) { |
72d94861 | 563 | ti->error = "error getting device"; |
1da177e4 LT |
564 | goto bad; |
565 | } | |
566 | ||
ae11b1b3 | 567 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
568 | struct request_queue *q = bdev_get_queue(p->path.dev->bdev); |
569 | ||
570 | r = scsi_dh_attach(q, m->hw_handler_name); | |
571 | if (r == -EBUSY) { | |
572 | /* | |
573 | * Already attached to different hw_handler, | |
574 | * try to reattach with correct one. | |
575 | */ | |
576 | scsi_dh_detach(q); | |
577 | r = scsi_dh_attach(q, m->hw_handler_name); | |
578 | } | |
579 | ||
ae11b1b3 | 580 | if (r < 0) { |
a0cf7ea9 | 581 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
582 | dm_put_device(ti, p->path.dev); |
583 | goto bad; | |
584 | } | |
2bfd2e13 CS |
585 | |
586 | if (m->hw_handler_params) { | |
587 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
588 | if (r < 0) { | |
589 | ti->error = "unable to set hardware " | |
590 | "handler parameters"; | |
591 | scsi_dh_detach(q); | |
592 | dm_put_device(ti, p->path.dev); | |
593 | goto bad; | |
594 | } | |
595 | } | |
ae11b1b3 HR |
596 | } |
597 | ||
1da177e4 LT |
598 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
599 | if (r) { | |
600 | dm_put_device(ti, p->path.dev); | |
601 | goto bad; | |
602 | } | |
603 | ||
604 | return p; | |
605 | ||
606 | bad: | |
607 | free_pgpath(p); | |
01460f35 | 608 | return ERR_PTR(r); |
1da177e4 LT |
609 | } |
610 | ||
498f0103 | 611 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 612 | struct multipath *m) |
1da177e4 | 613 | { |
498f0103 | 614 | static struct dm_arg _args[] = { |
72d94861 AK |
615 | {1, 1024, "invalid number of paths"}, |
616 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
617 | }; |
618 | ||
619 | int r; | |
498f0103 | 620 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 621 | struct priority_group *pg; |
28f16c20 | 622 | struct dm_target *ti = m->ti; |
1da177e4 LT |
623 | |
624 | if (as->argc < 2) { | |
625 | as->argc = 0; | |
01460f35 BM |
626 | ti->error = "not enough priority group arguments"; |
627 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
628 | } |
629 | ||
630 | pg = alloc_priority_group(); | |
631 | if (!pg) { | |
72d94861 | 632 | ti->error = "couldn't allocate priority group"; |
01460f35 | 633 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
634 | } |
635 | pg->m = m; | |
636 | ||
637 | r = parse_path_selector(as, pg, ti); | |
638 | if (r) | |
639 | goto bad; | |
640 | ||
641 | /* | |
642 | * read the paths | |
643 | */ | |
498f0103 | 644 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
645 | if (r) |
646 | goto bad; | |
647 | ||
498f0103 | 648 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
649 | if (r) |
650 | goto bad; | |
651 | ||
498f0103 | 652 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
653 | for (i = 0; i < pg->nr_pgpaths; i++) { |
654 | struct pgpath *pgpath; | |
498f0103 | 655 | struct dm_arg_set path_args; |
1da177e4 | 656 | |
498f0103 | 657 | if (as->argc < nr_args) { |
148acff6 | 658 | ti->error = "not enough path parameters"; |
6bbf79a1 | 659 | r = -EINVAL; |
1da177e4 | 660 | goto bad; |
148acff6 | 661 | } |
1da177e4 | 662 | |
498f0103 | 663 | path_args.argc = nr_args; |
1da177e4 LT |
664 | path_args.argv = as->argv; |
665 | ||
666 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
667 | if (IS_ERR(pgpath)) { |
668 | r = PTR_ERR(pgpath); | |
1da177e4 | 669 | goto bad; |
01460f35 | 670 | } |
1da177e4 LT |
671 | |
672 | pgpath->pg = pg; | |
673 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 674 | dm_consume_args(as, nr_args); |
1da177e4 LT |
675 | } |
676 | ||
677 | return pg; | |
678 | ||
679 | bad: | |
680 | free_priority_group(pg, ti); | |
01460f35 | 681 | return ERR_PTR(r); |
1da177e4 LT |
682 | } |
683 | ||
498f0103 | 684 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 685 | { |
1da177e4 | 686 | unsigned hw_argc; |
2bfd2e13 | 687 | int ret; |
28f16c20 | 688 | struct dm_target *ti = m->ti; |
1da177e4 | 689 | |
498f0103 | 690 | static struct dm_arg _args[] = { |
72d94861 | 691 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
692 | }; |
693 | ||
498f0103 | 694 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
695 | return -EINVAL; |
696 | ||
697 | if (!hw_argc) | |
698 | return 0; | |
699 | ||
498f0103 | 700 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
cfae5c9b CS |
701 | request_module("scsi_dh_%s", m->hw_handler_name); |
702 | if (scsi_dh_handler_exist(m->hw_handler_name) == 0) { | |
72d94861 | 703 | ti->error = "unknown hardware handler type"; |
2bfd2e13 CS |
704 | ret = -EINVAL; |
705 | goto fail; | |
1da177e4 | 706 | } |
14e98c5c | 707 | |
2bfd2e13 CS |
708 | if (hw_argc > 1) { |
709 | char *p; | |
710 | int i, j, len = 4; | |
711 | ||
712 | for (i = 0; i <= hw_argc - 2; i++) | |
713 | len += strlen(as->argv[i]) + 1; | |
714 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
715 | if (!p) { | |
716 | ti->error = "memory allocation failed"; | |
717 | ret = -ENOMEM; | |
718 | goto fail; | |
719 | } | |
720 | j = sprintf(p, "%d", hw_argc - 1); | |
721 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
722 | j = sprintf(p, "%s", as->argv[i]); | |
723 | } | |
498f0103 | 724 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
725 | |
726 | return 0; | |
2bfd2e13 CS |
727 | fail: |
728 | kfree(m->hw_handler_name); | |
729 | m->hw_handler_name = NULL; | |
730 | return ret; | |
1da177e4 LT |
731 | } |
732 | ||
498f0103 | 733 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
734 | { |
735 | int r; | |
736 | unsigned argc; | |
28f16c20 | 737 | struct dm_target *ti = m->ti; |
498f0103 | 738 | const char *arg_name; |
1da177e4 | 739 | |
498f0103 | 740 | static struct dm_arg _args[] = { |
4e2d19e4 | 741 | {0, 5, "invalid number of feature args"}, |
c9e45581 | 742 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 743 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
744 | }; |
745 | ||
498f0103 | 746 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
747 | if (r) |
748 | return -EINVAL; | |
749 | ||
750 | if (!argc) | |
751 | return 0; | |
752 | ||
c9e45581 | 753 | do { |
498f0103 | 754 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
755 | argc--; |
756 | ||
498f0103 | 757 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
c9e45581 DW |
758 | r = queue_if_no_path(m, 1, 0); |
759 | continue; | |
760 | } | |
761 | ||
498f0103 | 762 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 763 | (argc >= 1)) { |
498f0103 | 764 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
765 | argc--; |
766 | continue; | |
767 | } | |
768 | ||
498f0103 | 769 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 770 | (argc >= 1)) { |
498f0103 | 771 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
772 | argc--; |
773 | continue; | |
774 | } | |
775 | ||
1da177e4 | 776 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
777 | r = -EINVAL; |
778 | } while (argc && !r); | |
779 | ||
780 | return r; | |
1da177e4 LT |
781 | } |
782 | ||
783 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
784 | char **argv) | |
785 | { | |
498f0103 MS |
786 | /* target arguments */ |
787 | static struct dm_arg _args[] = { | |
a490a07a MS |
788 | {0, 1024, "invalid number of priority groups"}, |
789 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
790 | }; |
791 | ||
792 | int r; | |
793 | struct multipath *m; | |
498f0103 | 794 | struct dm_arg_set as; |
1da177e4 LT |
795 | unsigned pg_count = 0; |
796 | unsigned next_pg_num; | |
797 | ||
798 | as.argc = argc; | |
799 | as.argv = argv; | |
800 | ||
28f16c20 | 801 | m = alloc_multipath(ti); |
1da177e4 | 802 | if (!m) { |
72d94861 | 803 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
804 | return -EINVAL; |
805 | } | |
806 | ||
28f16c20 | 807 | r = parse_features(&as, m); |
1da177e4 LT |
808 | if (r) |
809 | goto bad; | |
810 | ||
28f16c20 | 811 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
812 | if (r) |
813 | goto bad; | |
814 | ||
498f0103 | 815 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
816 | if (r) |
817 | goto bad; | |
818 | ||
498f0103 | 819 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
820 | if (r) |
821 | goto bad; | |
822 | ||
a490a07a MS |
823 | if ((!m->nr_priority_groups && next_pg_num) || |
824 | (m->nr_priority_groups && !next_pg_num)) { | |
825 | ti->error = "invalid initial priority group"; | |
826 | r = -EINVAL; | |
827 | goto bad; | |
828 | } | |
829 | ||
1da177e4 LT |
830 | /* parse the priority groups */ |
831 | while (as.argc) { | |
832 | struct priority_group *pg; | |
833 | ||
28f16c20 | 834 | pg = parse_priority_group(&as, m); |
01460f35 BM |
835 | if (IS_ERR(pg)) { |
836 | r = PTR_ERR(pg); | |
1da177e4 LT |
837 | goto bad; |
838 | } | |
839 | ||
840 | m->nr_valid_paths += pg->nr_pgpaths; | |
841 | list_add_tail(&pg->list, &m->priority_groups); | |
842 | pg_count++; | |
843 | pg->pg_num = pg_count; | |
844 | if (!--next_pg_num) | |
845 | m->next_pg = pg; | |
846 | } | |
847 | ||
848 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 849 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
850 | r = -EINVAL; |
851 | goto bad; | |
852 | } | |
853 | ||
8627921f | 854 | ti->num_flush_requests = 1; |
959eb4e5 | 855 | ti->num_discard_requests = 1; |
8627921f | 856 | |
1da177e4 LT |
857 | return 0; |
858 | ||
859 | bad: | |
860 | free_multipath(m); | |
861 | return r; | |
862 | } | |
863 | ||
2bded7bd KU |
864 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
865 | { | |
866 | DECLARE_WAITQUEUE(wait, current); | |
867 | unsigned long flags; | |
868 | ||
869 | add_wait_queue(&m->pg_init_wait, &wait); | |
870 | ||
871 | while (1) { | |
872 | set_current_state(TASK_UNINTERRUPTIBLE); | |
873 | ||
874 | spin_lock_irqsave(&m->lock, flags); | |
875 | if (!m->pg_init_in_progress) { | |
876 | spin_unlock_irqrestore(&m->lock, flags); | |
877 | break; | |
878 | } | |
879 | spin_unlock_irqrestore(&m->lock, flags); | |
880 | ||
881 | io_schedule(); | |
882 | } | |
883 | set_current_state(TASK_RUNNING); | |
884 | ||
885 | remove_wait_queue(&m->pg_init_wait, &wait); | |
886 | } | |
887 | ||
888 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 889 | { |
bab7cfc7 | 890 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 891 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 892 | flush_workqueue(kmultipathd); |
d5ffa387 | 893 | flush_work_sync(&m->trigger_event); |
6df400ab KU |
894 | } |
895 | ||
896 | static void multipath_dtr(struct dm_target *ti) | |
897 | { | |
898 | struct multipath *m = ti->private; | |
899 | ||
2bded7bd | 900 | flush_multipath_work(m); |
1da177e4 LT |
901 | free_multipath(m); |
902 | } | |
903 | ||
904 | /* | |
f40c67f0 | 905 | * Map cloned requests |
1da177e4 | 906 | */ |
f40c67f0 | 907 | static int multipath_map(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
908 | union map_info *map_context) |
909 | { | |
910 | int r; | |
028867ac | 911 | struct dm_mpath_io *mpio; |
1da177e4 LT |
912 | struct multipath *m = (struct multipath *) ti->private; |
913 | ||
f40c67f0 KU |
914 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); |
915 | if (!mpio) | |
916 | /* ENOMEM, requeue */ | |
917 | return DM_MAPIO_REQUEUE; | |
918 | memset(mpio, 0, sizeof(*mpio)); | |
1da177e4 LT |
919 | |
920 | map_context->ptr = mpio; | |
f40c67f0 KU |
921 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; |
922 | r = map_io(m, clone, mpio, 0); | |
45e15720 | 923 | if (r < 0 || r == DM_MAPIO_REQUEUE) |
1da177e4 LT |
924 | mempool_free(mpio, m->mpio_pool); |
925 | ||
926 | return r; | |
927 | } | |
928 | ||
929 | /* | |
930 | * Take a path out of use. | |
931 | */ | |
932 | static int fail_path(struct pgpath *pgpath) | |
933 | { | |
934 | unsigned long flags; | |
935 | struct multipath *m = pgpath->pg->m; | |
936 | ||
937 | spin_lock_irqsave(&m->lock, flags); | |
938 | ||
6680073d | 939 | if (!pgpath->is_active) |
1da177e4 LT |
940 | goto out; |
941 | ||
72d94861 | 942 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
943 | |
944 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
6680073d | 945 | pgpath->is_active = 0; |
1da177e4 LT |
946 | pgpath->fail_count++; |
947 | ||
948 | m->nr_valid_paths--; | |
949 | ||
950 | if (pgpath == m->current_pgpath) | |
951 | m->current_pgpath = NULL; | |
952 | ||
b15546f9 MA |
953 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
954 | pgpath->path.dev->name, m->nr_valid_paths); | |
955 | ||
fe9cf30e | 956 | schedule_work(&m->trigger_event); |
1da177e4 LT |
957 | |
958 | out: | |
959 | spin_unlock_irqrestore(&m->lock, flags); | |
960 | ||
961 | return 0; | |
962 | } | |
963 | ||
964 | /* | |
965 | * Reinstate a previously-failed path | |
966 | */ | |
967 | static int reinstate_path(struct pgpath *pgpath) | |
968 | { | |
969 | int r = 0; | |
970 | unsigned long flags; | |
971 | struct multipath *m = pgpath->pg->m; | |
972 | ||
973 | spin_lock_irqsave(&m->lock, flags); | |
974 | ||
6680073d | 975 | if (pgpath->is_active) |
1da177e4 LT |
976 | goto out; |
977 | ||
def052d2 | 978 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
979 | DMWARN("Reinstate path not supported by path selector %s", |
980 | pgpath->pg->ps.type->name); | |
981 | r = -EINVAL; | |
982 | goto out; | |
983 | } | |
984 | ||
985 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
986 | if (r) | |
987 | goto out; | |
988 | ||
6680073d | 989 | pgpath->is_active = 1; |
1da177e4 | 990 | |
e54f77dd CS |
991 | if (!m->nr_valid_paths++ && m->queue_size) { |
992 | m->current_pgpath = NULL; | |
c557308e | 993 | queue_work(kmultipathd, &m->process_queued_ios); |
e54f77dd | 994 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 995 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
e54f77dd CS |
996 | m->pg_init_in_progress++; |
997 | } | |
1da177e4 | 998 | |
b15546f9 MA |
999 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
1000 | pgpath->path.dev->name, m->nr_valid_paths); | |
1001 | ||
fe9cf30e | 1002 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1003 | |
1004 | out: | |
1005 | spin_unlock_irqrestore(&m->lock, flags); | |
1006 | ||
1007 | return r; | |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1012 | */ | |
1013 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1014 | action_fn action) | |
1015 | { | |
19040c0b | 1016 | int r = -EINVAL; |
1da177e4 LT |
1017 | struct pgpath *pgpath; |
1018 | struct priority_group *pg; | |
1019 | ||
1020 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1021 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1022 | if (pgpath->path.dev == dev) | |
1023 | r = action(pgpath); | |
1024 | } | |
1025 | } | |
1026 | ||
1027 | return r; | |
1028 | } | |
1029 | ||
1030 | /* | |
1031 | * Temporarily try to avoid having to use the specified PG | |
1032 | */ | |
1033 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
1034 | int bypassed) | |
1035 | { | |
1036 | unsigned long flags; | |
1037 | ||
1038 | spin_lock_irqsave(&m->lock, flags); | |
1039 | ||
1040 | pg->bypassed = bypassed; | |
1041 | m->current_pgpath = NULL; | |
1042 | m->current_pg = NULL; | |
1043 | ||
1044 | spin_unlock_irqrestore(&m->lock, flags); | |
1045 | ||
fe9cf30e | 1046 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1047 | } |
1048 | ||
1049 | /* | |
1050 | * Switch to using the specified PG from the next I/O that gets mapped | |
1051 | */ | |
1052 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1053 | { | |
1054 | struct priority_group *pg; | |
1055 | unsigned pgnum; | |
1056 | unsigned long flags; | |
1057 | ||
1058 | if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || | |
1059 | (pgnum > m->nr_priority_groups)) { | |
1060 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1061 | return -EINVAL; | |
1062 | } | |
1063 | ||
1064 | spin_lock_irqsave(&m->lock, flags); | |
1065 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1066 | pg->bypassed = 0; | |
1067 | if (--pgnum) | |
1068 | continue; | |
1069 | ||
1070 | m->current_pgpath = NULL; | |
1071 | m->current_pg = NULL; | |
1072 | m->next_pg = pg; | |
1073 | } | |
1074 | spin_unlock_irqrestore(&m->lock, flags); | |
1075 | ||
fe9cf30e | 1076 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1077 | return 0; |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * Set/clear bypassed status of a PG. | |
1082 | * PGs are numbered upwards from 1 in the order they were declared. | |
1083 | */ | |
1084 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
1085 | { | |
1086 | struct priority_group *pg; | |
1087 | unsigned pgnum; | |
1088 | ||
1089 | if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || | |
1090 | (pgnum > m->nr_priority_groups)) { | |
1091 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1092 | return -EINVAL; | |
1093 | } | |
1094 | ||
1095 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1096 | if (!--pgnum) | |
1097 | break; | |
1098 | } | |
1099 | ||
1100 | bypass_pg(m, pg, bypassed); | |
1101 | return 0; | |
1102 | } | |
1103 | ||
c9e45581 DW |
1104 | /* |
1105 | * Should we retry pg_init immediately? | |
1106 | */ | |
1107 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1108 | { | |
1109 | unsigned long flags; | |
1110 | int limit_reached = 0; | |
1111 | ||
1112 | spin_lock_irqsave(&m->lock, flags); | |
1113 | ||
1114 | if (m->pg_init_count <= m->pg_init_retries) | |
1115 | m->pg_init_required = 1; | |
1116 | else | |
1117 | limit_reached = 1; | |
1118 | ||
1119 | spin_unlock_irqrestore(&m->lock, flags); | |
1120 | ||
1121 | return limit_reached; | |
1122 | } | |
1123 | ||
3ae31f6a | 1124 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1125 | { |
83c0d5d5 | 1126 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1127 | struct priority_group *pg = pgpath->pg; |
1128 | struct multipath *m = pg->m; | |
1129 | unsigned long flags; | |
4e2d19e4 | 1130 | unsigned delay_retry = 0; |
cfae5c9b CS |
1131 | |
1132 | /* device or driver problems */ | |
1133 | switch (errors) { | |
1134 | case SCSI_DH_OK: | |
1135 | break; | |
1136 | case SCSI_DH_NOSYS: | |
1137 | if (!m->hw_handler_name) { | |
1138 | errors = 0; | |
1139 | break; | |
1140 | } | |
f7b934c8 MB |
1141 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1142 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1143 | /* |
1144 | * Fail path for now, so we do not ping pong | |
1145 | */ | |
1146 | fail_path(pgpath); | |
1147 | break; | |
1148 | case SCSI_DH_DEV_TEMP_BUSY: | |
1149 | /* | |
1150 | * Probably doing something like FW upgrade on the | |
1151 | * controller so try the other pg. | |
1152 | */ | |
1153 | bypass_pg(m, pg, 1); | |
1154 | break; | |
cfae5c9b | 1155 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1156 | /* Wait before retrying. */ |
1157 | delay_retry = 1; | |
cfae5c9b CS |
1158 | case SCSI_DH_IMM_RETRY: |
1159 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1160 | if (pg_init_limit_reached(m, pgpath)) | |
1161 | fail_path(pgpath); | |
1162 | errors = 0; | |
1163 | break; | |
1164 | default: | |
1165 | /* | |
1166 | * We probably do not want to fail the path for a device | |
1167 | * error, but this is what the old dm did. In future | |
1168 | * patches we can do more advanced handling. | |
1169 | */ | |
1170 | fail_path(pgpath); | |
1171 | } | |
1172 | ||
1173 | spin_lock_irqsave(&m->lock, flags); | |
1174 | if (errors) { | |
e54f77dd CS |
1175 | if (pgpath == m->current_pgpath) { |
1176 | DMERR("Could not failover device. Error %d.", errors); | |
1177 | m->current_pgpath = NULL; | |
1178 | m->current_pg = NULL; | |
1179 | } | |
d0259bf0 | 1180 | } else if (!m->pg_init_required) |
cfae5c9b | 1181 | pg->bypassed = 0; |
cfae5c9b | 1182 | |
d0259bf0 KU |
1183 | if (--m->pg_init_in_progress) |
1184 | /* Activations of other paths are still on going */ | |
1185 | goto out; | |
1186 | ||
1187 | if (!m->pg_init_required) | |
1188 | m->queue_io = 0; | |
1189 | ||
4e2d19e4 | 1190 | m->pg_init_delay_retry = delay_retry; |
d0259bf0 KU |
1191 | queue_work(kmultipathd, &m->process_queued_ios); |
1192 | ||
2bded7bd KU |
1193 | /* |
1194 | * Wake up any thread waiting to suspend. | |
1195 | */ | |
1196 | wake_up(&m->pg_init_wait); | |
1197 | ||
d0259bf0 | 1198 | out: |
cfae5c9b CS |
1199 | spin_unlock_irqrestore(&m->lock, flags); |
1200 | } | |
1201 | ||
bab7cfc7 CS |
1202 | static void activate_path(struct work_struct *work) |
1203 | { | |
e54f77dd | 1204 | struct pgpath *pgpath = |
4e2d19e4 | 1205 | container_of(work, struct pgpath, activate_path.work); |
bab7cfc7 | 1206 | |
3ae31f6a | 1207 | scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev), |
83c0d5d5 | 1208 | pg_init_done, pgpath); |
bab7cfc7 CS |
1209 | } |
1210 | ||
1da177e4 LT |
1211 | /* |
1212 | * end_io handling | |
1213 | */ | |
f40c67f0 | 1214 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1215 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1216 | { |
f40c67f0 KU |
1217 | /* |
1218 | * We don't queue any clone request inside the multipath target | |
1219 | * during end I/O handling, since those clone requests don't have | |
1220 | * bio clones. If we queue them inside the multipath target, | |
1221 | * we need to make bio clones, that requires memory allocation. | |
1222 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1223 | * don't have bio clones.) | |
1224 | * Instead of queueing the clone request here, we queue the original | |
1225 | * request into dm core, which will remake a clone request and | |
1226 | * clone bios for it and resubmit it later. | |
1227 | */ | |
1228 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1229 | unsigned long flags; |
1da177e4 | 1230 | |
f40c67f0 | 1231 | if (!error && !clone->errors) |
1da177e4 LT |
1232 | return 0; /* I/O complete */ |
1233 | ||
6f13f6fb | 1234 | if (error == -EOPNOTSUPP || error == -EREMOTEIO || error == -EILSEQ) |
959eb4e5 MS |
1235 | return error; |
1236 | ||
cfae5c9b CS |
1237 | if (mpio->pgpath) |
1238 | fail_path(mpio->pgpath); | |
1da177e4 | 1239 | |
640eb3b0 | 1240 | spin_lock_irqsave(&m->lock, flags); |
751b2a7d HR |
1241 | if (!m->nr_valid_paths) { |
1242 | if (!m->queue_if_no_path) { | |
1243 | if (!__must_push_back(m)) | |
1244 | r = -EIO; | |
1245 | } else { | |
1246 | if (error == -EBADE) | |
1247 | r = error; | |
1248 | } | |
1249 | } | |
640eb3b0 | 1250 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1251 | |
f40c67f0 | 1252 | return r; |
1da177e4 LT |
1253 | } |
1254 | ||
f40c67f0 | 1255 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1256 | int error, union map_info *map_context) |
1257 | { | |
028867ac AK |
1258 | struct multipath *m = ti->private; |
1259 | struct dm_mpath_io *mpio = map_context->ptr; | |
1da177e4 LT |
1260 | struct pgpath *pgpath = mpio->pgpath; |
1261 | struct path_selector *ps; | |
1262 | int r; | |
1263 | ||
f40c67f0 | 1264 | r = do_end_io(m, clone, error, mpio); |
1da177e4 LT |
1265 | if (pgpath) { |
1266 | ps = &pgpath->pg->ps; | |
1267 | if (ps->type->end_io) | |
02ab823f | 1268 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1269 | } |
f40c67f0 | 1270 | mempool_free(mpio, m->mpio_pool); |
1da177e4 LT |
1271 | |
1272 | return r; | |
1273 | } | |
1274 | ||
1275 | /* | |
1276 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1277 | * the last path fails we must error any remaining I/O. |
1278 | * Note that if the freeze_bdev fails while suspending, the | |
1279 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1280 | */ |
1281 | static void multipath_presuspend(struct dm_target *ti) | |
1282 | { | |
1283 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1284 | |
485ef69e | 1285 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1286 | } |
1287 | ||
6df400ab KU |
1288 | static void multipath_postsuspend(struct dm_target *ti) |
1289 | { | |
6380f26f MA |
1290 | struct multipath *m = ti->private; |
1291 | ||
1292 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1293 | flush_multipath_work(m); |
6380f26f | 1294 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1295 | } |
1296 | ||
436d4108 AK |
1297 | /* |
1298 | * Restore the queue_if_no_path setting. | |
1299 | */ | |
1da177e4 LT |
1300 | static void multipath_resume(struct dm_target *ti) |
1301 | { | |
1302 | struct multipath *m = (struct multipath *) ti->private; | |
1303 | unsigned long flags; | |
1304 | ||
1305 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1306 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1307 | spin_unlock_irqrestore(&m->lock, flags); |
1308 | } | |
1309 | ||
1310 | /* | |
1311 | * Info output has the following format: | |
1312 | * num_multipath_feature_args [multipath_feature_args]* | |
1313 | * num_handler_status_args [handler_status_args]* | |
1314 | * num_groups init_group_number | |
1315 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1316 | * num_paths num_selector_args | |
1317 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1318 | * | |
1319 | * Table output has the following format (identical to the constructor string): | |
1320 | * num_feature_args [features_args]* | |
1321 | * num_handler_args hw_handler [hw_handler_args]* | |
1322 | * num_groups init_group_number | |
1323 | * [priority selector-name num_ps_args [ps_args]* | |
1324 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1325 | */ | |
1326 | static int multipath_status(struct dm_target *ti, status_type_t type, | |
1327 | char *result, unsigned int maxlen) | |
1328 | { | |
1329 | int sz = 0; | |
1330 | unsigned long flags; | |
1331 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1332 | struct priority_group *pg; |
1333 | struct pgpath *p; | |
1334 | unsigned pg_num; | |
1335 | char state; | |
1336 | ||
1337 | spin_lock_irqsave(&m->lock, flags); | |
1338 | ||
1339 | /* Features */ | |
1340 | if (type == STATUSTYPE_INFO) | |
c9e45581 DW |
1341 | DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count); |
1342 | else { | |
1343 | DMEMIT("%u ", m->queue_if_no_path + | |
4e2d19e4 CS |
1344 | (m->pg_init_retries > 0) * 2 + |
1345 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2); | |
c9e45581 DW |
1346 | if (m->queue_if_no_path) |
1347 | DMEMIT("queue_if_no_path "); | |
1348 | if (m->pg_init_retries) | |
1349 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1350 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1351 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
c9e45581 | 1352 | } |
1da177e4 | 1353 | |
cfae5c9b | 1354 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1355 | DMEMIT("0 "); |
1356 | else | |
cfae5c9b | 1357 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1358 | |
1359 | DMEMIT("%u ", m->nr_priority_groups); | |
1360 | ||
1361 | if (m->next_pg) | |
1362 | pg_num = m->next_pg->pg_num; | |
1363 | else if (m->current_pg) | |
1364 | pg_num = m->current_pg->pg_num; | |
1365 | else | |
a490a07a | 1366 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1367 | |
1368 | DMEMIT("%u ", pg_num); | |
1369 | ||
1370 | switch (type) { | |
1371 | case STATUSTYPE_INFO: | |
1372 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1373 | if (pg->bypassed) | |
1374 | state = 'D'; /* Disabled */ | |
1375 | else if (pg == m->current_pg) | |
1376 | state = 'A'; /* Currently Active */ | |
1377 | else | |
1378 | state = 'E'; /* Enabled */ | |
1379 | ||
1380 | DMEMIT("%c ", state); | |
1381 | ||
1382 | if (pg->ps.type->status) | |
1383 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1384 | result + sz, | |
1385 | maxlen - sz); | |
1386 | else | |
1387 | DMEMIT("0 "); | |
1388 | ||
1389 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1390 | pg->ps.type->info_args); | |
1391 | ||
1392 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1393 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1394 | p->is_active ? "A" : "F", |
1da177e4 LT |
1395 | p->fail_count); |
1396 | if (pg->ps.type->status) | |
1397 | sz += pg->ps.type->status(&pg->ps, | |
1398 | &p->path, type, result + sz, | |
1399 | maxlen - sz); | |
1400 | } | |
1401 | } | |
1402 | break; | |
1403 | ||
1404 | case STATUSTYPE_TABLE: | |
1405 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1406 | DMEMIT("%s ", pg->ps.type->name); | |
1407 | ||
1408 | if (pg->ps.type->status) | |
1409 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1410 | result + sz, | |
1411 | maxlen - sz); | |
1412 | else | |
1413 | DMEMIT("0 "); | |
1414 | ||
1415 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1416 | pg->ps.type->table_args); | |
1417 | ||
1418 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1419 | DMEMIT("%s ", p->path.dev->name); | |
1420 | if (pg->ps.type->status) | |
1421 | sz += pg->ps.type->status(&pg->ps, | |
1422 | &p->path, type, result + sz, | |
1423 | maxlen - sz); | |
1424 | } | |
1425 | } | |
1426 | break; | |
1427 | } | |
1428 | ||
1429 | spin_unlock_irqrestore(&m->lock, flags); | |
1430 | ||
1431 | return 0; | |
1432 | } | |
1433 | ||
1434 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1435 | { | |
6380f26f | 1436 | int r = -EINVAL; |
1da177e4 LT |
1437 | struct dm_dev *dev; |
1438 | struct multipath *m = (struct multipath *) ti->private; | |
1439 | action_fn action; | |
1440 | ||
6380f26f MA |
1441 | mutex_lock(&m->work_mutex); |
1442 | ||
c2f3d24b KU |
1443 | if (dm_suspended(ti)) { |
1444 | r = -EBUSY; | |
1445 | goto out; | |
1446 | } | |
1447 | ||
1da177e4 | 1448 | if (argc == 1) { |
498f0103 | 1449 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
6380f26f MA |
1450 | r = queue_if_no_path(m, 1, 0); |
1451 | goto out; | |
498f0103 | 1452 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
6380f26f MA |
1453 | r = queue_if_no_path(m, 0, 0); |
1454 | goto out; | |
1455 | } | |
1da177e4 LT |
1456 | } |
1457 | ||
6380f26f MA |
1458 | if (argc != 2) { |
1459 | DMWARN("Unrecognised multipath message received."); | |
1460 | goto out; | |
1461 | } | |
1da177e4 | 1462 | |
498f0103 | 1463 | if (!strcasecmp(argv[0], "disable_group")) { |
6380f26f MA |
1464 | r = bypass_pg_num(m, argv[1], 1); |
1465 | goto out; | |
498f0103 | 1466 | } else if (!strcasecmp(argv[0], "enable_group")) { |
6380f26f MA |
1467 | r = bypass_pg_num(m, argv[1], 0); |
1468 | goto out; | |
498f0103 | 1469 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1470 | r = switch_pg_num(m, argv[1]); |
1471 | goto out; | |
498f0103 | 1472 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1473 | action = reinstate_path; |
498f0103 | 1474 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1475 | action = fail_path; |
6380f26f MA |
1476 | else { |
1477 | DMWARN("Unrecognised multipath message received."); | |
1478 | goto out; | |
1479 | } | |
1da177e4 | 1480 | |
8215d6ec | 1481 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1482 | if (r) { |
72d94861 | 1483 | DMWARN("message: error getting device %s", |
1da177e4 | 1484 | argv[1]); |
6380f26f | 1485 | goto out; |
1da177e4 LT |
1486 | } |
1487 | ||
1488 | r = action_dev(m, dev, action); | |
1489 | ||
1490 | dm_put_device(ti, dev); | |
1491 | ||
6380f26f MA |
1492 | out: |
1493 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1494 | return r; |
1da177e4 LT |
1495 | } |
1496 | ||
647b3d00 | 1497 | static int multipath_ioctl(struct dm_target *ti, unsigned int cmd, |
9af4aa30 MB |
1498 | unsigned long arg) |
1499 | { | |
1500 | struct multipath *m = (struct multipath *) ti->private; | |
1501 | struct block_device *bdev = NULL; | |
633a08b8 | 1502 | fmode_t mode = 0; |
9af4aa30 MB |
1503 | unsigned long flags; |
1504 | int r = 0; | |
1505 | ||
1506 | spin_lock_irqsave(&m->lock, flags); | |
1507 | ||
1508 | if (!m->current_pgpath) | |
02ab823f | 1509 | __choose_pgpath(m, 0); |
9af4aa30 | 1510 | |
e90dae1f | 1511 | if (m->current_pgpath) { |
9af4aa30 | 1512 | bdev = m->current_pgpath->path.dev->bdev; |
633a08b8 | 1513 | mode = m->current_pgpath->path.dev->mode; |
e90dae1f | 1514 | } |
9af4aa30 MB |
1515 | |
1516 | if (m->queue_io) | |
1517 | r = -EAGAIN; | |
1518 | else if (!bdev) | |
1519 | r = -EIO; | |
1520 | ||
1521 | spin_unlock_irqrestore(&m->lock, flags); | |
1522 | ||
ec8013be PB |
1523 | /* |
1524 | * Only pass ioctls through if the device sizes match exactly. | |
1525 | */ | |
1526 | if (!r && ti->len != i_size_read(bdev->bd_inode) >> SECTOR_SHIFT) | |
1527 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
1528 | ||
633a08b8 | 1529 | return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
9af4aa30 MB |
1530 | } |
1531 | ||
af4874e0 MS |
1532 | static int multipath_iterate_devices(struct dm_target *ti, |
1533 | iterate_devices_callout_fn fn, void *data) | |
1534 | { | |
1535 | struct multipath *m = ti->private; | |
1536 | struct priority_group *pg; | |
1537 | struct pgpath *p; | |
1538 | int ret = 0; | |
1539 | ||
1540 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1541 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1542 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1543 | if (ret) |
1544 | goto out; | |
1545 | } | |
1546 | } | |
1547 | ||
1548 | out: | |
1549 | return ret; | |
1550 | } | |
1551 | ||
f40c67f0 KU |
1552 | static int __pgpath_busy(struct pgpath *pgpath) |
1553 | { | |
1554 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1555 | ||
1556 | return dm_underlying_device_busy(q); | |
1557 | } | |
1558 | ||
1559 | /* | |
1560 | * We return "busy", only when we can map I/Os but underlying devices | |
1561 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1562 | * the underlying queue). | |
1563 | * In other words, if we want to kill I/Os or queue them inside us | |
1564 | * due to map unavailability, we don't return "busy". Otherwise, | |
1565 | * dm core won't give us the I/Os and we can't do what we want. | |
1566 | */ | |
1567 | static int multipath_busy(struct dm_target *ti) | |
1568 | { | |
1569 | int busy = 0, has_active = 0; | |
1570 | struct multipath *m = ti->private; | |
1571 | struct priority_group *pg; | |
1572 | struct pgpath *pgpath; | |
1573 | unsigned long flags; | |
1574 | ||
1575 | spin_lock_irqsave(&m->lock, flags); | |
1576 | ||
1577 | /* Guess which priority_group will be used at next mapping time */ | |
1578 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1579 | pg = m->next_pg; | |
1580 | else if (likely(m->current_pg)) | |
1581 | pg = m->current_pg; | |
1582 | else | |
1583 | /* | |
1584 | * We don't know which pg will be used at next mapping time. | |
1585 | * We don't call __choose_pgpath() here to avoid to trigger | |
1586 | * pg_init just by busy checking. | |
1587 | * So we don't know whether underlying devices we will be using | |
1588 | * at next mapping time are busy or not. Just try mapping. | |
1589 | */ | |
1590 | goto out; | |
1591 | ||
1592 | /* | |
1593 | * If there is one non-busy active path at least, the path selector | |
1594 | * will be able to select it. So we consider such a pg as not busy. | |
1595 | */ | |
1596 | busy = 1; | |
1597 | list_for_each_entry(pgpath, &pg->pgpaths, list) | |
1598 | if (pgpath->is_active) { | |
1599 | has_active = 1; | |
1600 | ||
1601 | if (!__pgpath_busy(pgpath)) { | |
1602 | busy = 0; | |
1603 | break; | |
1604 | } | |
1605 | } | |
1606 | ||
1607 | if (!has_active) | |
1608 | /* | |
1609 | * No active path in this pg, so this pg won't be used and | |
1610 | * the current_pg will be changed at next mapping time. | |
1611 | * We need to try mapping to determine it. | |
1612 | */ | |
1613 | busy = 0; | |
1614 | ||
1615 | out: | |
1616 | spin_unlock_irqrestore(&m->lock, flags); | |
1617 | ||
1618 | return busy; | |
1619 | } | |
1620 | ||
1da177e4 LT |
1621 | /*----------------------------------------------------------------- |
1622 | * Module setup | |
1623 | *---------------------------------------------------------------*/ | |
1624 | static struct target_type multipath_target = { | |
1625 | .name = "multipath", | |
19040c0b | 1626 | .version = {1, 3, 0}, |
1da177e4 LT |
1627 | .module = THIS_MODULE, |
1628 | .ctr = multipath_ctr, | |
1629 | .dtr = multipath_dtr, | |
f40c67f0 KU |
1630 | .map_rq = multipath_map, |
1631 | .rq_end_io = multipath_end_io, | |
1da177e4 | 1632 | .presuspend = multipath_presuspend, |
6df400ab | 1633 | .postsuspend = multipath_postsuspend, |
1da177e4 LT |
1634 | .resume = multipath_resume, |
1635 | .status = multipath_status, | |
1636 | .message = multipath_message, | |
9af4aa30 | 1637 | .ioctl = multipath_ioctl, |
af4874e0 | 1638 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1639 | .busy = multipath_busy, |
1da177e4 LT |
1640 | }; |
1641 | ||
1642 | static int __init dm_multipath_init(void) | |
1643 | { | |
1644 | int r; | |
1645 | ||
1646 | /* allocate a slab for the dm_ios */ | |
028867ac | 1647 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1648 | if (!_mpio_cache) |
1649 | return -ENOMEM; | |
1650 | ||
1651 | r = dm_register_target(&multipath_target); | |
1652 | if (r < 0) { | |
0cd33124 | 1653 | DMERR("register failed %d", r); |
1da177e4 LT |
1654 | kmem_cache_destroy(_mpio_cache); |
1655 | return -EINVAL; | |
1656 | } | |
1657 | ||
4d4d66ab | 1658 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1659 | if (!kmultipathd) { |
0cd33124 | 1660 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1661 | dm_unregister_target(&multipath_target); |
1662 | kmem_cache_destroy(_mpio_cache); | |
1663 | return -ENOMEM; | |
1664 | } | |
1665 | ||
bab7cfc7 CS |
1666 | /* |
1667 | * A separate workqueue is used to handle the device handlers | |
1668 | * to avoid overloading existing workqueue. Overloading the | |
1669 | * old workqueue would also create a bottleneck in the | |
1670 | * path of the storage hardware device activation. | |
1671 | */ | |
4d4d66ab TH |
1672 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1673 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1674 | if (!kmpath_handlerd) { |
1675 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1676 | destroy_workqueue(kmultipathd); | |
1677 | dm_unregister_target(&multipath_target); | |
1678 | kmem_cache_destroy(_mpio_cache); | |
1679 | return -ENOMEM; | |
1680 | } | |
1681 | ||
72d94861 | 1682 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1683 | multipath_target.version[0], multipath_target.version[1], |
1684 | multipath_target.version[2]); | |
1685 | ||
1686 | return r; | |
1687 | } | |
1688 | ||
1689 | static void __exit dm_multipath_exit(void) | |
1690 | { | |
bab7cfc7 | 1691 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1692 | destroy_workqueue(kmultipathd); |
1693 | ||
10d3bd09 | 1694 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1695 | kmem_cache_destroy(_mpio_cache); |
1696 | } | |
1697 | ||
1da177e4 LT |
1698 | module_init(dm_multipath_init); |
1699 | module_exit(dm_multipath_exit); | |
1700 | ||
1701 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1702 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1703 | MODULE_LICENSE("GPL"); |