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 | ||
466891f9 JN |
229 | static int set_mapinfo(struct multipath *m, union map_info *info) |
230 | { | |
231 | struct dm_mpath_io *mpio; | |
232 | ||
233 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); | |
234 | if (!mpio) | |
235 | return -ENOMEM; | |
236 | ||
237 | memset(mpio, 0, sizeof(*mpio)); | |
238 | info->ptr = mpio; | |
239 | ||
240 | return 0; | |
241 | } | |
242 | ||
243 | static void clear_mapinfo(struct multipath *m, union map_info *info) | |
244 | { | |
245 | struct dm_mpath_io *mpio = info->ptr; | |
246 | ||
247 | info->ptr = NULL; | |
248 | mempool_free(mpio, m->mpio_pool); | |
249 | } | |
1da177e4 LT |
250 | |
251 | /*----------------------------------------------- | |
252 | * Path selection | |
253 | *-----------------------------------------------*/ | |
254 | ||
fb612642 KU |
255 | static void __pg_init_all_paths(struct multipath *m) |
256 | { | |
257 | struct pgpath *pgpath; | |
4e2d19e4 | 258 | unsigned long pg_init_delay = 0; |
fb612642 KU |
259 | |
260 | m->pg_init_count++; | |
261 | m->pg_init_required = 0; | |
4e2d19e4 CS |
262 | if (m->pg_init_delay_retry) |
263 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? | |
264 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
265 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
266 | /* Skip failed paths */ | |
267 | if (!pgpath->is_active) | |
268 | continue; | |
4e2d19e4 CS |
269 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
270 | pg_init_delay)) | |
fb612642 KU |
271 | m->pg_init_in_progress++; |
272 | } | |
273 | } | |
274 | ||
1da177e4 LT |
275 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) |
276 | { | |
1da177e4 LT |
277 | m->current_pg = pgpath->pg; |
278 | ||
279 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 280 | if (m->hw_handler_name) { |
1da177e4 LT |
281 | m->pg_init_required = 1; |
282 | m->queue_io = 1; | |
283 | } else { | |
284 | m->pg_init_required = 0; | |
285 | m->queue_io = 0; | |
286 | } | |
c9e45581 DW |
287 | |
288 | m->pg_init_count = 0; | |
1da177e4 LT |
289 | } |
290 | ||
02ab823f KU |
291 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
292 | size_t nr_bytes) | |
1da177e4 | 293 | { |
c922d5f7 | 294 | struct dm_path *path; |
1da177e4 | 295 | |
02ab823f | 296 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes); |
1da177e4 LT |
297 | if (!path) |
298 | return -ENXIO; | |
299 | ||
300 | m->current_pgpath = path_to_pgpath(path); | |
301 | ||
302 | if (m->current_pg != pg) | |
303 | __switch_pg(m, m->current_pgpath); | |
304 | ||
305 | return 0; | |
306 | } | |
307 | ||
02ab823f | 308 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
309 | { |
310 | struct priority_group *pg; | |
311 | unsigned bypassed = 1; | |
312 | ||
313 | if (!m->nr_valid_paths) | |
314 | goto failed; | |
315 | ||
316 | /* Were we instructed to switch PG? */ | |
317 | if (m->next_pg) { | |
318 | pg = m->next_pg; | |
319 | m->next_pg = NULL; | |
02ab823f | 320 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
321 | return; |
322 | } | |
323 | ||
324 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 325 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
326 | return; |
327 | ||
328 | /* | |
329 | * Loop through priority groups until we find a valid path. | |
330 | * First time we skip PGs marked 'bypassed'. | |
331 | * Second time we only try the ones we skipped. | |
332 | */ | |
333 | do { | |
334 | list_for_each_entry(pg, &m->priority_groups, list) { | |
335 | if (pg->bypassed == bypassed) | |
336 | continue; | |
02ab823f | 337 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
338 | return; |
339 | } | |
340 | } while (bypassed--); | |
341 | ||
342 | failed: | |
343 | m->current_pgpath = NULL; | |
344 | m->current_pg = NULL; | |
345 | } | |
346 | ||
45e15720 KU |
347 | /* |
348 | * Check whether bios must be queued in the device-mapper core rather | |
349 | * than here in the target. | |
350 | * | |
351 | * m->lock must be held on entry. | |
352 | * | |
353 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
354 | * same value then we are not between multipath_presuspend() | |
355 | * and multipath_resume() calls and we have no need to check | |
356 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
357 | */ | |
358 | static int __must_push_back(struct multipath *m) | |
359 | { | |
360 | return (m->queue_if_no_path != m->saved_queue_if_no_path && | |
361 | dm_noflush_suspending(m->ti)); | |
362 | } | |
363 | ||
f40c67f0 | 364 | static int map_io(struct multipath *m, struct request *clone, |
466891f9 | 365 | union map_info *map_context, unsigned was_queued) |
1da177e4 | 366 | { |
d2a7ad29 | 367 | int r = DM_MAPIO_REMAPPED; |
f40c67f0 | 368 | size_t nr_bytes = blk_rq_bytes(clone); |
1da177e4 LT |
369 | unsigned long flags; |
370 | struct pgpath *pgpath; | |
f40c67f0 | 371 | struct block_device *bdev; |
466891f9 | 372 | struct dm_mpath_io *mpio = map_context->ptr; |
1da177e4 LT |
373 | |
374 | spin_lock_irqsave(&m->lock, flags); | |
375 | ||
376 | /* Do we need to select a new pgpath? */ | |
377 | if (!m->current_pgpath || | |
378 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
02ab823f | 379 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
380 | |
381 | pgpath = m->current_pgpath; | |
382 | ||
383 | if (was_queued) | |
384 | m->queue_size--; | |
385 | ||
386 | if ((pgpath && m->queue_io) || | |
436d4108 | 387 | (!pgpath && m->queue_if_no_path)) { |
1da177e4 | 388 | /* Queue for the daemon to resubmit */ |
f40c67f0 | 389 | list_add_tail(&clone->queuelist, &m->queued_ios); |
1da177e4 | 390 | m->queue_size++; |
c3cd4f6b AK |
391 | if ((m->pg_init_required && !m->pg_init_in_progress) || |
392 | !m->queue_io) | |
c557308e | 393 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 394 | pgpath = NULL; |
d2a7ad29 | 395 | r = DM_MAPIO_SUBMITTED; |
f40c67f0 KU |
396 | } else if (pgpath) { |
397 | bdev = pgpath->path.dev->bdev; | |
398 | clone->q = bdev_get_queue(bdev); | |
399 | clone->rq_disk = bdev->bd_disk; | |
400 | } else if (__must_push_back(m)) | |
45e15720 KU |
401 | r = DM_MAPIO_REQUEUE; |
402 | else | |
403 | r = -EIO; /* Failed */ | |
1da177e4 LT |
404 | |
405 | mpio->pgpath = pgpath; | |
02ab823f KU |
406 | mpio->nr_bytes = nr_bytes; |
407 | ||
408 | if (r == DM_MAPIO_REMAPPED && pgpath->pg->ps.type->start_io) | |
409 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, &pgpath->path, | |
410 | nr_bytes); | |
1da177e4 LT |
411 | |
412 | spin_unlock_irqrestore(&m->lock, flags); | |
413 | ||
414 | return r; | |
415 | } | |
416 | ||
417 | /* | |
418 | * If we run out of usable paths, should we queue I/O or error it? | |
419 | */ | |
485ef69e AK |
420 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
421 | unsigned save_old_value) | |
1da177e4 LT |
422 | { |
423 | unsigned long flags; | |
424 | ||
425 | spin_lock_irqsave(&m->lock, flags); | |
426 | ||
485ef69e AK |
427 | if (save_old_value) |
428 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
429 | else | |
430 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 431 | m->queue_if_no_path = queue_if_no_path; |
c3cd4f6b | 432 | if (!m->queue_if_no_path && m->queue_size) |
c557308e | 433 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 LT |
434 | |
435 | spin_unlock_irqrestore(&m->lock, flags); | |
436 | ||
437 | return 0; | |
438 | } | |
439 | ||
440 | /*----------------------------------------------------------------- | |
441 | * The multipath daemon is responsible for resubmitting queued ios. | |
442 | *---------------------------------------------------------------*/ | |
443 | ||
444 | static void dispatch_queued_ios(struct multipath *m) | |
445 | { | |
446 | int r; | |
447 | unsigned long flags; | |
1da177e4 | 448 | union map_info *info; |
f40c67f0 KU |
449 | struct request *clone, *n; |
450 | LIST_HEAD(cl); | |
1da177e4 LT |
451 | |
452 | spin_lock_irqsave(&m->lock, flags); | |
f40c67f0 | 453 | list_splice_init(&m->queued_ios, &cl); |
1da177e4 LT |
454 | spin_unlock_irqrestore(&m->lock, flags); |
455 | ||
f40c67f0 KU |
456 | list_for_each_entry_safe(clone, n, &cl, queuelist) { |
457 | list_del_init(&clone->queuelist); | |
1da177e4 | 458 | |
f40c67f0 | 459 | info = dm_get_rq_mapinfo(clone); |
1da177e4 | 460 | |
466891f9 | 461 | r = map_io(m, clone, info, 1); |
f40c67f0 | 462 | if (r < 0) { |
466891f9 | 463 | clear_mapinfo(m, info); |
f40c67f0 KU |
464 | dm_kill_unmapped_request(clone, r); |
465 | } else if (r == DM_MAPIO_REMAPPED) | |
466 | dm_dispatch_request(clone); | |
467 | else if (r == DM_MAPIO_REQUEUE) { | |
466891f9 | 468 | clear_mapinfo(m, info); |
f40c67f0 KU |
469 | dm_requeue_unmapped_request(clone); |
470 | } | |
1da177e4 LT |
471 | } |
472 | } | |
473 | ||
c4028958 | 474 | static void process_queued_ios(struct work_struct *work) |
1da177e4 | 475 | { |
c4028958 DH |
476 | struct multipath *m = |
477 | container_of(work, struct multipath, process_queued_ios); | |
fb612642 | 478 | struct pgpath *pgpath = NULL; |
e54f77dd | 479 | unsigned must_queue = 1; |
1da177e4 LT |
480 | unsigned long flags; |
481 | ||
482 | spin_lock_irqsave(&m->lock, flags); | |
483 | ||
c3cd4f6b AK |
484 | if (!m->queue_size) |
485 | goto out; | |
486 | ||
1da177e4 | 487 | if (!m->current_pgpath) |
02ab823f | 488 | __choose_pgpath(m, 0); |
1da177e4 LT |
489 | |
490 | pgpath = m->current_pgpath; | |
491 | ||
c3cd4f6b AK |
492 | if ((pgpath && !m->queue_io) || |
493 | (!pgpath && !m->queue_if_no_path)) | |
494 | must_queue = 0; | |
1da177e4 | 495 | |
fb612642 KU |
496 | if (m->pg_init_required && !m->pg_init_in_progress && pgpath) |
497 | __pg_init_all_paths(m); | |
498 | ||
c3cd4f6b | 499 | out: |
1da177e4 | 500 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
501 | if (!must_queue) |
502 | dispatch_queued_ios(m); | |
503 | } | |
504 | ||
505 | /* | |
506 | * An event is triggered whenever a path is taken out of use. | |
507 | * Includes path failure and PG bypass. | |
508 | */ | |
c4028958 | 509 | static void trigger_event(struct work_struct *work) |
1da177e4 | 510 | { |
c4028958 DH |
511 | struct multipath *m = |
512 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
513 | |
514 | dm_table_event(m->ti->table); | |
515 | } | |
516 | ||
517 | /*----------------------------------------------------------------- | |
518 | * Constructor/argument parsing: | |
519 | * <#multipath feature args> [<arg>]* | |
520 | * <#hw_handler args> [hw_handler [<arg>]*] | |
521 | * <#priority groups> | |
522 | * <initial priority group> | |
523 | * [<selector> <#selector args> [<arg>]* | |
524 | * <#paths> <#per-path selector args> | |
525 | * [<path> [<arg>]* ]+ ]+ | |
526 | *---------------------------------------------------------------*/ | |
498f0103 | 527 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
528 | struct dm_target *ti) |
529 | { | |
530 | int r; | |
531 | struct path_selector_type *pst; | |
532 | unsigned ps_argc; | |
533 | ||
498f0103 | 534 | static struct dm_arg _args[] = { |
72d94861 | 535 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
536 | }; |
537 | ||
498f0103 | 538 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 539 | if (!pst) { |
72d94861 | 540 | ti->error = "unknown path selector type"; |
1da177e4 LT |
541 | return -EINVAL; |
542 | } | |
543 | ||
498f0103 | 544 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
545 | if (r) { |
546 | dm_put_path_selector(pst); | |
1da177e4 | 547 | return -EINVAL; |
371b2e34 | 548 | } |
1da177e4 LT |
549 | |
550 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
551 | if (r) { | |
552 | dm_put_path_selector(pst); | |
72d94861 | 553 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
554 | return r; |
555 | } | |
556 | ||
557 | pg->ps.type = pst; | |
498f0103 | 558 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
559 | |
560 | return 0; | |
561 | } | |
562 | ||
498f0103 | 563 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
564 | struct dm_target *ti) |
565 | { | |
566 | int r; | |
567 | struct pgpath *p; | |
ae11b1b3 | 568 | struct multipath *m = ti->private; |
1da177e4 LT |
569 | |
570 | /* we need at least a path arg */ | |
571 | if (as->argc < 1) { | |
72d94861 | 572 | ti->error = "no device given"; |
01460f35 | 573 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
574 | } |
575 | ||
576 | p = alloc_pgpath(); | |
577 | if (!p) | |
01460f35 | 578 | return ERR_PTR(-ENOMEM); |
1da177e4 | 579 | |
498f0103 | 580 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 581 | &p->path.dev); |
1da177e4 | 582 | if (r) { |
72d94861 | 583 | ti->error = "error getting device"; |
1da177e4 LT |
584 | goto bad; |
585 | } | |
586 | ||
ae11b1b3 | 587 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
588 | struct request_queue *q = bdev_get_queue(p->path.dev->bdev); |
589 | ||
590 | r = scsi_dh_attach(q, m->hw_handler_name); | |
591 | if (r == -EBUSY) { | |
592 | /* | |
593 | * Already attached to different hw_handler, | |
594 | * try to reattach with correct one. | |
595 | */ | |
596 | scsi_dh_detach(q); | |
597 | r = scsi_dh_attach(q, m->hw_handler_name); | |
598 | } | |
599 | ||
ae11b1b3 | 600 | if (r < 0) { |
a0cf7ea9 | 601 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
602 | dm_put_device(ti, p->path.dev); |
603 | goto bad; | |
604 | } | |
2bfd2e13 CS |
605 | |
606 | if (m->hw_handler_params) { | |
607 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
608 | if (r < 0) { | |
609 | ti->error = "unable to set hardware " | |
610 | "handler parameters"; | |
611 | scsi_dh_detach(q); | |
612 | dm_put_device(ti, p->path.dev); | |
613 | goto bad; | |
614 | } | |
615 | } | |
ae11b1b3 HR |
616 | } |
617 | ||
1da177e4 LT |
618 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
619 | if (r) { | |
620 | dm_put_device(ti, p->path.dev); | |
621 | goto bad; | |
622 | } | |
623 | ||
624 | return p; | |
625 | ||
626 | bad: | |
627 | free_pgpath(p); | |
01460f35 | 628 | return ERR_PTR(r); |
1da177e4 LT |
629 | } |
630 | ||
498f0103 | 631 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 632 | struct multipath *m) |
1da177e4 | 633 | { |
498f0103 | 634 | static struct dm_arg _args[] = { |
72d94861 AK |
635 | {1, 1024, "invalid number of paths"}, |
636 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
637 | }; |
638 | ||
639 | int r; | |
498f0103 | 640 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 641 | struct priority_group *pg; |
28f16c20 | 642 | struct dm_target *ti = m->ti; |
1da177e4 LT |
643 | |
644 | if (as->argc < 2) { | |
645 | as->argc = 0; | |
01460f35 BM |
646 | ti->error = "not enough priority group arguments"; |
647 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
648 | } |
649 | ||
650 | pg = alloc_priority_group(); | |
651 | if (!pg) { | |
72d94861 | 652 | ti->error = "couldn't allocate priority group"; |
01460f35 | 653 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
654 | } |
655 | pg->m = m; | |
656 | ||
657 | r = parse_path_selector(as, pg, ti); | |
658 | if (r) | |
659 | goto bad; | |
660 | ||
661 | /* | |
662 | * read the paths | |
663 | */ | |
498f0103 | 664 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
665 | if (r) |
666 | goto bad; | |
667 | ||
498f0103 | 668 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
669 | if (r) |
670 | goto bad; | |
671 | ||
498f0103 | 672 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
673 | for (i = 0; i < pg->nr_pgpaths; i++) { |
674 | struct pgpath *pgpath; | |
498f0103 | 675 | struct dm_arg_set path_args; |
1da177e4 | 676 | |
498f0103 | 677 | if (as->argc < nr_args) { |
148acff6 | 678 | ti->error = "not enough path parameters"; |
6bbf79a1 | 679 | r = -EINVAL; |
1da177e4 | 680 | goto bad; |
148acff6 | 681 | } |
1da177e4 | 682 | |
498f0103 | 683 | path_args.argc = nr_args; |
1da177e4 LT |
684 | path_args.argv = as->argv; |
685 | ||
686 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
687 | if (IS_ERR(pgpath)) { |
688 | r = PTR_ERR(pgpath); | |
1da177e4 | 689 | goto bad; |
01460f35 | 690 | } |
1da177e4 LT |
691 | |
692 | pgpath->pg = pg; | |
693 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 694 | dm_consume_args(as, nr_args); |
1da177e4 LT |
695 | } |
696 | ||
697 | return pg; | |
698 | ||
699 | bad: | |
700 | free_priority_group(pg, ti); | |
01460f35 | 701 | return ERR_PTR(r); |
1da177e4 LT |
702 | } |
703 | ||
498f0103 | 704 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 705 | { |
1da177e4 | 706 | unsigned hw_argc; |
2bfd2e13 | 707 | int ret; |
28f16c20 | 708 | struct dm_target *ti = m->ti; |
1da177e4 | 709 | |
498f0103 | 710 | static struct dm_arg _args[] = { |
72d94861 | 711 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
712 | }; |
713 | ||
498f0103 | 714 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
715 | return -EINVAL; |
716 | ||
717 | if (!hw_argc) | |
718 | return 0; | |
719 | ||
498f0103 | 720 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
cfae5c9b CS |
721 | request_module("scsi_dh_%s", m->hw_handler_name); |
722 | if (scsi_dh_handler_exist(m->hw_handler_name) == 0) { | |
72d94861 | 723 | ti->error = "unknown hardware handler type"; |
2bfd2e13 CS |
724 | ret = -EINVAL; |
725 | goto fail; | |
1da177e4 | 726 | } |
14e98c5c | 727 | |
2bfd2e13 CS |
728 | if (hw_argc > 1) { |
729 | char *p; | |
730 | int i, j, len = 4; | |
731 | ||
732 | for (i = 0; i <= hw_argc - 2; i++) | |
733 | len += strlen(as->argv[i]) + 1; | |
734 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
735 | if (!p) { | |
736 | ti->error = "memory allocation failed"; | |
737 | ret = -ENOMEM; | |
738 | goto fail; | |
739 | } | |
740 | j = sprintf(p, "%d", hw_argc - 1); | |
741 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
742 | j = sprintf(p, "%s", as->argv[i]); | |
743 | } | |
498f0103 | 744 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
745 | |
746 | return 0; | |
2bfd2e13 CS |
747 | fail: |
748 | kfree(m->hw_handler_name); | |
749 | m->hw_handler_name = NULL; | |
750 | return ret; | |
1da177e4 LT |
751 | } |
752 | ||
498f0103 | 753 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
754 | { |
755 | int r; | |
756 | unsigned argc; | |
28f16c20 | 757 | struct dm_target *ti = m->ti; |
498f0103 | 758 | const char *arg_name; |
1da177e4 | 759 | |
498f0103 | 760 | static struct dm_arg _args[] = { |
4e2d19e4 | 761 | {0, 5, "invalid number of feature args"}, |
c9e45581 | 762 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 763 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
764 | }; |
765 | ||
498f0103 | 766 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
767 | if (r) |
768 | return -EINVAL; | |
769 | ||
770 | if (!argc) | |
771 | return 0; | |
772 | ||
c9e45581 | 773 | do { |
498f0103 | 774 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
775 | argc--; |
776 | ||
498f0103 | 777 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
c9e45581 DW |
778 | r = queue_if_no_path(m, 1, 0); |
779 | continue; | |
780 | } | |
781 | ||
498f0103 | 782 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 783 | (argc >= 1)) { |
498f0103 | 784 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
785 | argc--; |
786 | continue; | |
787 | } | |
788 | ||
498f0103 | 789 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 790 | (argc >= 1)) { |
498f0103 | 791 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
792 | argc--; |
793 | continue; | |
794 | } | |
795 | ||
1da177e4 | 796 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
797 | r = -EINVAL; |
798 | } while (argc && !r); | |
799 | ||
800 | return r; | |
1da177e4 LT |
801 | } |
802 | ||
803 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
804 | char **argv) | |
805 | { | |
498f0103 MS |
806 | /* target arguments */ |
807 | static struct dm_arg _args[] = { | |
a490a07a MS |
808 | {0, 1024, "invalid number of priority groups"}, |
809 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
810 | }; |
811 | ||
812 | int r; | |
813 | struct multipath *m; | |
498f0103 | 814 | struct dm_arg_set as; |
1da177e4 LT |
815 | unsigned pg_count = 0; |
816 | unsigned next_pg_num; | |
817 | ||
818 | as.argc = argc; | |
819 | as.argv = argv; | |
820 | ||
28f16c20 | 821 | m = alloc_multipath(ti); |
1da177e4 | 822 | if (!m) { |
72d94861 | 823 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
824 | return -EINVAL; |
825 | } | |
826 | ||
28f16c20 | 827 | r = parse_features(&as, m); |
1da177e4 LT |
828 | if (r) |
829 | goto bad; | |
830 | ||
28f16c20 | 831 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
832 | if (r) |
833 | goto bad; | |
834 | ||
498f0103 | 835 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
836 | if (r) |
837 | goto bad; | |
838 | ||
498f0103 | 839 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
840 | if (r) |
841 | goto bad; | |
842 | ||
a490a07a MS |
843 | if ((!m->nr_priority_groups && next_pg_num) || |
844 | (m->nr_priority_groups && !next_pg_num)) { | |
845 | ti->error = "invalid initial priority group"; | |
846 | r = -EINVAL; | |
847 | goto bad; | |
848 | } | |
849 | ||
1da177e4 LT |
850 | /* parse the priority groups */ |
851 | while (as.argc) { | |
852 | struct priority_group *pg; | |
853 | ||
28f16c20 | 854 | pg = parse_priority_group(&as, m); |
01460f35 BM |
855 | if (IS_ERR(pg)) { |
856 | r = PTR_ERR(pg); | |
1da177e4 LT |
857 | goto bad; |
858 | } | |
859 | ||
860 | m->nr_valid_paths += pg->nr_pgpaths; | |
861 | list_add_tail(&pg->list, &m->priority_groups); | |
862 | pg_count++; | |
863 | pg->pg_num = pg_count; | |
864 | if (!--next_pg_num) | |
865 | m->next_pg = pg; | |
866 | } | |
867 | ||
868 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 869 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
870 | r = -EINVAL; |
871 | goto bad; | |
872 | } | |
873 | ||
8627921f | 874 | ti->num_flush_requests = 1; |
959eb4e5 | 875 | ti->num_discard_requests = 1; |
8627921f | 876 | |
1da177e4 LT |
877 | return 0; |
878 | ||
879 | bad: | |
880 | free_multipath(m); | |
881 | return r; | |
882 | } | |
883 | ||
2bded7bd KU |
884 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
885 | { | |
886 | DECLARE_WAITQUEUE(wait, current); | |
887 | unsigned long flags; | |
888 | ||
889 | add_wait_queue(&m->pg_init_wait, &wait); | |
890 | ||
891 | while (1) { | |
892 | set_current_state(TASK_UNINTERRUPTIBLE); | |
893 | ||
894 | spin_lock_irqsave(&m->lock, flags); | |
895 | if (!m->pg_init_in_progress) { | |
896 | spin_unlock_irqrestore(&m->lock, flags); | |
897 | break; | |
898 | } | |
899 | spin_unlock_irqrestore(&m->lock, flags); | |
900 | ||
901 | io_schedule(); | |
902 | } | |
903 | set_current_state(TASK_RUNNING); | |
904 | ||
905 | remove_wait_queue(&m->pg_init_wait, &wait); | |
906 | } | |
907 | ||
908 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 909 | { |
bab7cfc7 | 910 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 911 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 912 | flush_workqueue(kmultipathd); |
d5ffa387 | 913 | flush_work_sync(&m->trigger_event); |
6df400ab KU |
914 | } |
915 | ||
916 | static void multipath_dtr(struct dm_target *ti) | |
917 | { | |
918 | struct multipath *m = ti->private; | |
919 | ||
2bded7bd | 920 | flush_multipath_work(m); |
1da177e4 LT |
921 | free_multipath(m); |
922 | } | |
923 | ||
924 | /* | |
f40c67f0 | 925 | * Map cloned requests |
1da177e4 | 926 | */ |
f40c67f0 | 927 | static int multipath_map(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
928 | union map_info *map_context) |
929 | { | |
930 | int r; | |
1da177e4 LT |
931 | struct multipath *m = (struct multipath *) ti->private; |
932 | ||
466891f9 | 933 | if (set_mapinfo(m, map_context) < 0) |
f40c67f0 KU |
934 | /* ENOMEM, requeue */ |
935 | return DM_MAPIO_REQUEUE; | |
1da177e4 | 936 | |
f40c67f0 | 937 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; |
466891f9 | 938 | r = map_io(m, clone, map_context, 0); |
45e15720 | 939 | if (r < 0 || r == DM_MAPIO_REQUEUE) |
466891f9 | 940 | clear_mapinfo(m, map_context); |
1da177e4 LT |
941 | |
942 | return r; | |
943 | } | |
944 | ||
945 | /* | |
946 | * Take a path out of use. | |
947 | */ | |
948 | static int fail_path(struct pgpath *pgpath) | |
949 | { | |
950 | unsigned long flags; | |
951 | struct multipath *m = pgpath->pg->m; | |
952 | ||
953 | spin_lock_irqsave(&m->lock, flags); | |
954 | ||
6680073d | 955 | if (!pgpath->is_active) |
1da177e4 LT |
956 | goto out; |
957 | ||
72d94861 | 958 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
959 | |
960 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
6680073d | 961 | pgpath->is_active = 0; |
1da177e4 LT |
962 | pgpath->fail_count++; |
963 | ||
964 | m->nr_valid_paths--; | |
965 | ||
966 | if (pgpath == m->current_pgpath) | |
967 | m->current_pgpath = NULL; | |
968 | ||
b15546f9 MA |
969 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
970 | pgpath->path.dev->name, m->nr_valid_paths); | |
971 | ||
fe9cf30e | 972 | schedule_work(&m->trigger_event); |
1da177e4 LT |
973 | |
974 | out: | |
975 | spin_unlock_irqrestore(&m->lock, flags); | |
976 | ||
977 | return 0; | |
978 | } | |
979 | ||
980 | /* | |
981 | * Reinstate a previously-failed path | |
982 | */ | |
983 | static int reinstate_path(struct pgpath *pgpath) | |
984 | { | |
985 | int r = 0; | |
986 | unsigned long flags; | |
987 | struct multipath *m = pgpath->pg->m; | |
988 | ||
989 | spin_lock_irqsave(&m->lock, flags); | |
990 | ||
6680073d | 991 | if (pgpath->is_active) |
1da177e4 LT |
992 | goto out; |
993 | ||
def052d2 | 994 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
995 | DMWARN("Reinstate path not supported by path selector %s", |
996 | pgpath->pg->ps.type->name); | |
997 | r = -EINVAL; | |
998 | goto out; | |
999 | } | |
1000 | ||
1001 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1002 | if (r) | |
1003 | goto out; | |
1004 | ||
6680073d | 1005 | pgpath->is_active = 1; |
1da177e4 | 1006 | |
e54f77dd CS |
1007 | if (!m->nr_valid_paths++ && m->queue_size) { |
1008 | m->current_pgpath = NULL; | |
c557308e | 1009 | queue_work(kmultipathd, &m->process_queued_ios); |
e54f77dd | 1010 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1011 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
e54f77dd CS |
1012 | m->pg_init_in_progress++; |
1013 | } | |
1da177e4 | 1014 | |
b15546f9 MA |
1015 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
1016 | pgpath->path.dev->name, m->nr_valid_paths); | |
1017 | ||
fe9cf30e | 1018 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1019 | |
1020 | out: | |
1021 | spin_unlock_irqrestore(&m->lock, flags); | |
1022 | ||
1023 | return r; | |
1024 | } | |
1025 | ||
1026 | /* | |
1027 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1028 | */ | |
1029 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1030 | action_fn action) | |
1031 | { | |
19040c0b | 1032 | int r = -EINVAL; |
1da177e4 LT |
1033 | struct pgpath *pgpath; |
1034 | struct priority_group *pg; | |
1035 | ||
1036 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1037 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1038 | if (pgpath->path.dev == dev) | |
1039 | r = action(pgpath); | |
1040 | } | |
1041 | } | |
1042 | ||
1043 | return r; | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * Temporarily try to avoid having to use the specified PG | |
1048 | */ | |
1049 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
1050 | int bypassed) | |
1051 | { | |
1052 | unsigned long flags; | |
1053 | ||
1054 | spin_lock_irqsave(&m->lock, flags); | |
1055 | ||
1056 | pg->bypassed = bypassed; | |
1057 | m->current_pgpath = NULL; | |
1058 | m->current_pg = NULL; | |
1059 | ||
1060 | spin_unlock_irqrestore(&m->lock, flags); | |
1061 | ||
fe9cf30e | 1062 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1063 | } |
1064 | ||
1065 | /* | |
1066 | * Switch to using the specified PG from the next I/O that gets mapped | |
1067 | */ | |
1068 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1069 | { | |
1070 | struct priority_group *pg; | |
1071 | unsigned pgnum; | |
1072 | unsigned long flags; | |
31998ef1 | 1073 | char dummy; |
1da177e4 | 1074 | |
31998ef1 | 1075 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1076 | (pgnum > m->nr_priority_groups)) { |
1077 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1078 | return -EINVAL; | |
1079 | } | |
1080 | ||
1081 | spin_lock_irqsave(&m->lock, flags); | |
1082 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1083 | pg->bypassed = 0; | |
1084 | if (--pgnum) | |
1085 | continue; | |
1086 | ||
1087 | m->current_pgpath = NULL; | |
1088 | m->current_pg = NULL; | |
1089 | m->next_pg = pg; | |
1090 | } | |
1091 | spin_unlock_irqrestore(&m->lock, flags); | |
1092 | ||
fe9cf30e | 1093 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1094 | return 0; |
1095 | } | |
1096 | ||
1097 | /* | |
1098 | * Set/clear bypassed status of a PG. | |
1099 | * PGs are numbered upwards from 1 in the order they were declared. | |
1100 | */ | |
1101 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
1102 | { | |
1103 | struct priority_group *pg; | |
1104 | unsigned pgnum; | |
31998ef1 | 1105 | char dummy; |
1da177e4 | 1106 | |
31998ef1 | 1107 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1108 | (pgnum > m->nr_priority_groups)) { |
1109 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1110 | return -EINVAL; | |
1111 | } | |
1112 | ||
1113 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1114 | if (!--pgnum) | |
1115 | break; | |
1116 | } | |
1117 | ||
1118 | bypass_pg(m, pg, bypassed); | |
1119 | return 0; | |
1120 | } | |
1121 | ||
c9e45581 DW |
1122 | /* |
1123 | * Should we retry pg_init immediately? | |
1124 | */ | |
1125 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1126 | { | |
1127 | unsigned long flags; | |
1128 | int limit_reached = 0; | |
1129 | ||
1130 | spin_lock_irqsave(&m->lock, flags); | |
1131 | ||
1132 | if (m->pg_init_count <= m->pg_init_retries) | |
1133 | m->pg_init_required = 1; | |
1134 | else | |
1135 | limit_reached = 1; | |
1136 | ||
1137 | spin_unlock_irqrestore(&m->lock, flags); | |
1138 | ||
1139 | return limit_reached; | |
1140 | } | |
1141 | ||
3ae31f6a | 1142 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1143 | { |
83c0d5d5 | 1144 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1145 | struct priority_group *pg = pgpath->pg; |
1146 | struct multipath *m = pg->m; | |
1147 | unsigned long flags; | |
4e2d19e4 | 1148 | unsigned delay_retry = 0; |
cfae5c9b CS |
1149 | |
1150 | /* device or driver problems */ | |
1151 | switch (errors) { | |
1152 | case SCSI_DH_OK: | |
1153 | break; | |
1154 | case SCSI_DH_NOSYS: | |
1155 | if (!m->hw_handler_name) { | |
1156 | errors = 0; | |
1157 | break; | |
1158 | } | |
f7b934c8 MB |
1159 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1160 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1161 | /* |
1162 | * Fail path for now, so we do not ping pong | |
1163 | */ | |
1164 | fail_path(pgpath); | |
1165 | break; | |
1166 | case SCSI_DH_DEV_TEMP_BUSY: | |
1167 | /* | |
1168 | * Probably doing something like FW upgrade on the | |
1169 | * controller so try the other pg. | |
1170 | */ | |
1171 | bypass_pg(m, pg, 1); | |
1172 | break; | |
cfae5c9b | 1173 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1174 | /* Wait before retrying. */ |
1175 | delay_retry = 1; | |
cfae5c9b CS |
1176 | case SCSI_DH_IMM_RETRY: |
1177 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1178 | if (pg_init_limit_reached(m, pgpath)) | |
1179 | fail_path(pgpath); | |
1180 | errors = 0; | |
1181 | break; | |
1182 | default: | |
1183 | /* | |
1184 | * We probably do not want to fail the path for a device | |
1185 | * error, but this is what the old dm did. In future | |
1186 | * patches we can do more advanced handling. | |
1187 | */ | |
1188 | fail_path(pgpath); | |
1189 | } | |
1190 | ||
1191 | spin_lock_irqsave(&m->lock, flags); | |
1192 | if (errors) { | |
e54f77dd CS |
1193 | if (pgpath == m->current_pgpath) { |
1194 | DMERR("Could not failover device. Error %d.", errors); | |
1195 | m->current_pgpath = NULL; | |
1196 | m->current_pg = NULL; | |
1197 | } | |
d0259bf0 | 1198 | } else if (!m->pg_init_required) |
cfae5c9b | 1199 | pg->bypassed = 0; |
cfae5c9b | 1200 | |
d0259bf0 KU |
1201 | if (--m->pg_init_in_progress) |
1202 | /* Activations of other paths are still on going */ | |
1203 | goto out; | |
1204 | ||
1205 | if (!m->pg_init_required) | |
1206 | m->queue_io = 0; | |
1207 | ||
4e2d19e4 | 1208 | m->pg_init_delay_retry = delay_retry; |
d0259bf0 KU |
1209 | queue_work(kmultipathd, &m->process_queued_ios); |
1210 | ||
2bded7bd KU |
1211 | /* |
1212 | * Wake up any thread waiting to suspend. | |
1213 | */ | |
1214 | wake_up(&m->pg_init_wait); | |
1215 | ||
d0259bf0 | 1216 | out: |
cfae5c9b CS |
1217 | spin_unlock_irqrestore(&m->lock, flags); |
1218 | } | |
1219 | ||
bab7cfc7 CS |
1220 | static void activate_path(struct work_struct *work) |
1221 | { | |
e54f77dd | 1222 | struct pgpath *pgpath = |
4e2d19e4 | 1223 | container_of(work, struct pgpath, activate_path.work); |
bab7cfc7 | 1224 | |
3ae31f6a | 1225 | scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev), |
83c0d5d5 | 1226 | pg_init_done, pgpath); |
bab7cfc7 CS |
1227 | } |
1228 | ||
1da177e4 LT |
1229 | /* |
1230 | * end_io handling | |
1231 | */ | |
f40c67f0 | 1232 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1233 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1234 | { |
f40c67f0 KU |
1235 | /* |
1236 | * We don't queue any clone request inside the multipath target | |
1237 | * during end I/O handling, since those clone requests don't have | |
1238 | * bio clones. If we queue them inside the multipath target, | |
1239 | * we need to make bio clones, that requires memory allocation. | |
1240 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1241 | * don't have bio clones.) | |
1242 | * Instead of queueing the clone request here, we queue the original | |
1243 | * request into dm core, which will remake a clone request and | |
1244 | * clone bios for it and resubmit it later. | |
1245 | */ | |
1246 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1247 | unsigned long flags; |
1da177e4 | 1248 | |
f40c67f0 | 1249 | if (!error && !clone->errors) |
1da177e4 LT |
1250 | return 0; /* I/O complete */ |
1251 | ||
6f13f6fb | 1252 | if (error == -EOPNOTSUPP || error == -EREMOTEIO || error == -EILSEQ) |
959eb4e5 MS |
1253 | return error; |
1254 | ||
cfae5c9b CS |
1255 | if (mpio->pgpath) |
1256 | fail_path(mpio->pgpath); | |
1da177e4 | 1257 | |
640eb3b0 | 1258 | spin_lock_irqsave(&m->lock, flags); |
751b2a7d HR |
1259 | if (!m->nr_valid_paths) { |
1260 | if (!m->queue_if_no_path) { | |
1261 | if (!__must_push_back(m)) | |
1262 | r = -EIO; | |
1263 | } else { | |
1264 | if (error == -EBADE) | |
1265 | r = error; | |
1266 | } | |
1267 | } | |
640eb3b0 | 1268 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1269 | |
f40c67f0 | 1270 | return r; |
1da177e4 LT |
1271 | } |
1272 | ||
f40c67f0 | 1273 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1274 | int error, union map_info *map_context) |
1275 | { | |
028867ac AK |
1276 | struct multipath *m = ti->private; |
1277 | struct dm_mpath_io *mpio = map_context->ptr; | |
1da177e4 LT |
1278 | struct pgpath *pgpath = mpio->pgpath; |
1279 | struct path_selector *ps; | |
1280 | int r; | |
1281 | ||
466891f9 JN |
1282 | BUG_ON(!mpio); |
1283 | ||
f40c67f0 | 1284 | r = do_end_io(m, clone, error, mpio); |
1da177e4 LT |
1285 | if (pgpath) { |
1286 | ps = &pgpath->pg->ps; | |
1287 | if (ps->type->end_io) | |
02ab823f | 1288 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1289 | } |
466891f9 | 1290 | clear_mapinfo(m, map_context); |
1da177e4 LT |
1291 | |
1292 | return r; | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1297 | * the last path fails we must error any remaining I/O. |
1298 | * Note that if the freeze_bdev fails while suspending, the | |
1299 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1300 | */ |
1301 | static void multipath_presuspend(struct dm_target *ti) | |
1302 | { | |
1303 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1304 | |
485ef69e | 1305 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1306 | } |
1307 | ||
6df400ab KU |
1308 | static void multipath_postsuspend(struct dm_target *ti) |
1309 | { | |
6380f26f MA |
1310 | struct multipath *m = ti->private; |
1311 | ||
1312 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1313 | flush_multipath_work(m); |
6380f26f | 1314 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1315 | } |
1316 | ||
436d4108 AK |
1317 | /* |
1318 | * Restore the queue_if_no_path setting. | |
1319 | */ | |
1da177e4 LT |
1320 | static void multipath_resume(struct dm_target *ti) |
1321 | { | |
1322 | struct multipath *m = (struct multipath *) ti->private; | |
1323 | unsigned long flags; | |
1324 | ||
1325 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1326 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1327 | spin_unlock_irqrestore(&m->lock, flags); |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Info output has the following format: | |
1332 | * num_multipath_feature_args [multipath_feature_args]* | |
1333 | * num_handler_status_args [handler_status_args]* | |
1334 | * num_groups init_group_number | |
1335 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1336 | * num_paths num_selector_args | |
1337 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1338 | * | |
1339 | * Table output has the following format (identical to the constructor string): | |
1340 | * num_feature_args [features_args]* | |
1341 | * num_handler_args hw_handler [hw_handler_args]* | |
1342 | * num_groups init_group_number | |
1343 | * [priority selector-name num_ps_args [ps_args]* | |
1344 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1345 | */ | |
1346 | static int multipath_status(struct dm_target *ti, status_type_t type, | |
1347 | char *result, unsigned int maxlen) | |
1348 | { | |
1349 | int sz = 0; | |
1350 | unsigned long flags; | |
1351 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1352 | struct priority_group *pg; |
1353 | struct pgpath *p; | |
1354 | unsigned pg_num; | |
1355 | char state; | |
1356 | ||
1357 | spin_lock_irqsave(&m->lock, flags); | |
1358 | ||
1359 | /* Features */ | |
1360 | if (type == STATUSTYPE_INFO) | |
c9e45581 DW |
1361 | DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count); |
1362 | else { | |
1363 | DMEMIT("%u ", m->queue_if_no_path + | |
4e2d19e4 CS |
1364 | (m->pg_init_retries > 0) * 2 + |
1365 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2); | |
c9e45581 DW |
1366 | if (m->queue_if_no_path) |
1367 | DMEMIT("queue_if_no_path "); | |
1368 | if (m->pg_init_retries) | |
1369 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1370 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1371 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
c9e45581 | 1372 | } |
1da177e4 | 1373 | |
cfae5c9b | 1374 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1375 | DMEMIT("0 "); |
1376 | else | |
cfae5c9b | 1377 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1378 | |
1379 | DMEMIT("%u ", m->nr_priority_groups); | |
1380 | ||
1381 | if (m->next_pg) | |
1382 | pg_num = m->next_pg->pg_num; | |
1383 | else if (m->current_pg) | |
1384 | pg_num = m->current_pg->pg_num; | |
1385 | else | |
a490a07a | 1386 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1387 | |
1388 | DMEMIT("%u ", pg_num); | |
1389 | ||
1390 | switch (type) { | |
1391 | case STATUSTYPE_INFO: | |
1392 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1393 | if (pg->bypassed) | |
1394 | state = 'D'; /* Disabled */ | |
1395 | else if (pg == m->current_pg) | |
1396 | state = 'A'; /* Currently Active */ | |
1397 | else | |
1398 | state = 'E'; /* Enabled */ | |
1399 | ||
1400 | DMEMIT("%c ", state); | |
1401 | ||
1402 | if (pg->ps.type->status) | |
1403 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1404 | result + sz, | |
1405 | maxlen - sz); | |
1406 | else | |
1407 | DMEMIT("0 "); | |
1408 | ||
1409 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1410 | pg->ps.type->info_args); | |
1411 | ||
1412 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1413 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1414 | p->is_active ? "A" : "F", |
1da177e4 LT |
1415 | p->fail_count); |
1416 | if (pg->ps.type->status) | |
1417 | sz += pg->ps.type->status(&pg->ps, | |
1418 | &p->path, type, result + sz, | |
1419 | maxlen - sz); | |
1420 | } | |
1421 | } | |
1422 | break; | |
1423 | ||
1424 | case STATUSTYPE_TABLE: | |
1425 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1426 | DMEMIT("%s ", pg->ps.type->name); | |
1427 | ||
1428 | if (pg->ps.type->status) | |
1429 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1430 | result + sz, | |
1431 | maxlen - sz); | |
1432 | else | |
1433 | DMEMIT("0 "); | |
1434 | ||
1435 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1436 | pg->ps.type->table_args); | |
1437 | ||
1438 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1439 | DMEMIT("%s ", p->path.dev->name); | |
1440 | if (pg->ps.type->status) | |
1441 | sz += pg->ps.type->status(&pg->ps, | |
1442 | &p->path, type, result + sz, | |
1443 | maxlen - sz); | |
1444 | } | |
1445 | } | |
1446 | break; | |
1447 | } | |
1448 | ||
1449 | spin_unlock_irqrestore(&m->lock, flags); | |
1450 | ||
1451 | return 0; | |
1452 | } | |
1453 | ||
1454 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1455 | { | |
6380f26f | 1456 | int r = -EINVAL; |
1da177e4 LT |
1457 | struct dm_dev *dev; |
1458 | struct multipath *m = (struct multipath *) ti->private; | |
1459 | action_fn action; | |
1460 | ||
6380f26f MA |
1461 | mutex_lock(&m->work_mutex); |
1462 | ||
c2f3d24b KU |
1463 | if (dm_suspended(ti)) { |
1464 | r = -EBUSY; | |
1465 | goto out; | |
1466 | } | |
1467 | ||
1da177e4 | 1468 | if (argc == 1) { |
498f0103 | 1469 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
6380f26f MA |
1470 | r = queue_if_no_path(m, 1, 0); |
1471 | goto out; | |
498f0103 | 1472 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
6380f26f MA |
1473 | r = queue_if_no_path(m, 0, 0); |
1474 | goto out; | |
1475 | } | |
1da177e4 LT |
1476 | } |
1477 | ||
6380f26f MA |
1478 | if (argc != 2) { |
1479 | DMWARN("Unrecognised multipath message received."); | |
1480 | goto out; | |
1481 | } | |
1da177e4 | 1482 | |
498f0103 | 1483 | if (!strcasecmp(argv[0], "disable_group")) { |
6380f26f MA |
1484 | r = bypass_pg_num(m, argv[1], 1); |
1485 | goto out; | |
498f0103 | 1486 | } else if (!strcasecmp(argv[0], "enable_group")) { |
6380f26f MA |
1487 | r = bypass_pg_num(m, argv[1], 0); |
1488 | goto out; | |
498f0103 | 1489 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1490 | r = switch_pg_num(m, argv[1]); |
1491 | goto out; | |
498f0103 | 1492 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1493 | action = reinstate_path; |
498f0103 | 1494 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1495 | action = fail_path; |
6380f26f MA |
1496 | else { |
1497 | DMWARN("Unrecognised multipath message received."); | |
1498 | goto out; | |
1499 | } | |
1da177e4 | 1500 | |
8215d6ec | 1501 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1502 | if (r) { |
72d94861 | 1503 | DMWARN("message: error getting device %s", |
1da177e4 | 1504 | argv[1]); |
6380f26f | 1505 | goto out; |
1da177e4 LT |
1506 | } |
1507 | ||
1508 | r = action_dev(m, dev, action); | |
1509 | ||
1510 | dm_put_device(ti, dev); | |
1511 | ||
6380f26f MA |
1512 | out: |
1513 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1514 | return r; |
1da177e4 LT |
1515 | } |
1516 | ||
647b3d00 | 1517 | static int multipath_ioctl(struct dm_target *ti, unsigned int cmd, |
9af4aa30 MB |
1518 | unsigned long arg) |
1519 | { | |
1520 | struct multipath *m = (struct multipath *) ti->private; | |
1521 | struct block_device *bdev = NULL; | |
633a08b8 | 1522 | fmode_t mode = 0; |
9af4aa30 MB |
1523 | unsigned long flags; |
1524 | int r = 0; | |
1525 | ||
1526 | spin_lock_irqsave(&m->lock, flags); | |
1527 | ||
1528 | if (!m->current_pgpath) | |
02ab823f | 1529 | __choose_pgpath(m, 0); |
9af4aa30 | 1530 | |
e90dae1f | 1531 | if (m->current_pgpath) { |
9af4aa30 | 1532 | bdev = m->current_pgpath->path.dev->bdev; |
633a08b8 | 1533 | mode = m->current_pgpath->path.dev->mode; |
e90dae1f | 1534 | } |
9af4aa30 MB |
1535 | |
1536 | if (m->queue_io) | |
1537 | r = -EAGAIN; | |
1538 | else if (!bdev) | |
1539 | r = -EIO; | |
1540 | ||
1541 | spin_unlock_irqrestore(&m->lock, flags); | |
1542 | ||
ec8013be PB |
1543 | /* |
1544 | * Only pass ioctls through if the device sizes match exactly. | |
1545 | */ | |
1546 | if (!r && ti->len != i_size_read(bdev->bd_inode) >> SECTOR_SHIFT) | |
1547 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
1548 | ||
633a08b8 | 1549 | return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
9af4aa30 MB |
1550 | } |
1551 | ||
af4874e0 MS |
1552 | static int multipath_iterate_devices(struct dm_target *ti, |
1553 | iterate_devices_callout_fn fn, void *data) | |
1554 | { | |
1555 | struct multipath *m = ti->private; | |
1556 | struct priority_group *pg; | |
1557 | struct pgpath *p; | |
1558 | int ret = 0; | |
1559 | ||
1560 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1561 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1562 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1563 | if (ret) |
1564 | goto out; | |
1565 | } | |
1566 | } | |
1567 | ||
1568 | out: | |
1569 | return ret; | |
1570 | } | |
1571 | ||
f40c67f0 KU |
1572 | static int __pgpath_busy(struct pgpath *pgpath) |
1573 | { | |
1574 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1575 | ||
1576 | return dm_underlying_device_busy(q); | |
1577 | } | |
1578 | ||
1579 | /* | |
1580 | * We return "busy", only when we can map I/Os but underlying devices | |
1581 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1582 | * the underlying queue). | |
1583 | * In other words, if we want to kill I/Os or queue them inside us | |
1584 | * due to map unavailability, we don't return "busy". Otherwise, | |
1585 | * dm core won't give us the I/Os and we can't do what we want. | |
1586 | */ | |
1587 | static int multipath_busy(struct dm_target *ti) | |
1588 | { | |
1589 | int busy = 0, has_active = 0; | |
1590 | struct multipath *m = ti->private; | |
1591 | struct priority_group *pg; | |
1592 | struct pgpath *pgpath; | |
1593 | unsigned long flags; | |
1594 | ||
1595 | spin_lock_irqsave(&m->lock, flags); | |
1596 | ||
1597 | /* Guess which priority_group will be used at next mapping time */ | |
1598 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1599 | pg = m->next_pg; | |
1600 | else if (likely(m->current_pg)) | |
1601 | pg = m->current_pg; | |
1602 | else | |
1603 | /* | |
1604 | * We don't know which pg will be used at next mapping time. | |
1605 | * We don't call __choose_pgpath() here to avoid to trigger | |
1606 | * pg_init just by busy checking. | |
1607 | * So we don't know whether underlying devices we will be using | |
1608 | * at next mapping time are busy or not. Just try mapping. | |
1609 | */ | |
1610 | goto out; | |
1611 | ||
1612 | /* | |
1613 | * If there is one non-busy active path at least, the path selector | |
1614 | * will be able to select it. So we consider such a pg as not busy. | |
1615 | */ | |
1616 | busy = 1; | |
1617 | list_for_each_entry(pgpath, &pg->pgpaths, list) | |
1618 | if (pgpath->is_active) { | |
1619 | has_active = 1; | |
1620 | ||
1621 | if (!__pgpath_busy(pgpath)) { | |
1622 | busy = 0; | |
1623 | break; | |
1624 | } | |
1625 | } | |
1626 | ||
1627 | if (!has_active) | |
1628 | /* | |
1629 | * No active path in this pg, so this pg won't be used and | |
1630 | * the current_pg will be changed at next mapping time. | |
1631 | * We need to try mapping to determine it. | |
1632 | */ | |
1633 | busy = 0; | |
1634 | ||
1635 | out: | |
1636 | spin_unlock_irqrestore(&m->lock, flags); | |
1637 | ||
1638 | return busy; | |
1639 | } | |
1640 | ||
1da177e4 LT |
1641 | /*----------------------------------------------------------------- |
1642 | * Module setup | |
1643 | *---------------------------------------------------------------*/ | |
1644 | static struct target_type multipath_target = { | |
1645 | .name = "multipath", | |
19040c0b | 1646 | .version = {1, 3, 0}, |
1da177e4 LT |
1647 | .module = THIS_MODULE, |
1648 | .ctr = multipath_ctr, | |
1649 | .dtr = multipath_dtr, | |
f40c67f0 KU |
1650 | .map_rq = multipath_map, |
1651 | .rq_end_io = multipath_end_io, | |
1da177e4 | 1652 | .presuspend = multipath_presuspend, |
6df400ab | 1653 | .postsuspend = multipath_postsuspend, |
1da177e4 LT |
1654 | .resume = multipath_resume, |
1655 | .status = multipath_status, | |
1656 | .message = multipath_message, | |
9af4aa30 | 1657 | .ioctl = multipath_ioctl, |
af4874e0 | 1658 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1659 | .busy = multipath_busy, |
1da177e4 LT |
1660 | }; |
1661 | ||
1662 | static int __init dm_multipath_init(void) | |
1663 | { | |
1664 | int r; | |
1665 | ||
1666 | /* allocate a slab for the dm_ios */ | |
028867ac | 1667 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1668 | if (!_mpio_cache) |
1669 | return -ENOMEM; | |
1670 | ||
1671 | r = dm_register_target(&multipath_target); | |
1672 | if (r < 0) { | |
0cd33124 | 1673 | DMERR("register failed %d", r); |
1da177e4 LT |
1674 | kmem_cache_destroy(_mpio_cache); |
1675 | return -EINVAL; | |
1676 | } | |
1677 | ||
4d4d66ab | 1678 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1679 | if (!kmultipathd) { |
0cd33124 | 1680 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1681 | dm_unregister_target(&multipath_target); |
1682 | kmem_cache_destroy(_mpio_cache); | |
1683 | return -ENOMEM; | |
1684 | } | |
1685 | ||
bab7cfc7 CS |
1686 | /* |
1687 | * A separate workqueue is used to handle the device handlers | |
1688 | * to avoid overloading existing workqueue. Overloading the | |
1689 | * old workqueue would also create a bottleneck in the | |
1690 | * path of the storage hardware device activation. | |
1691 | */ | |
4d4d66ab TH |
1692 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1693 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1694 | if (!kmpath_handlerd) { |
1695 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1696 | destroy_workqueue(kmultipathd); | |
1697 | dm_unregister_target(&multipath_target); | |
1698 | kmem_cache_destroy(_mpio_cache); | |
1699 | return -ENOMEM; | |
1700 | } | |
1701 | ||
72d94861 | 1702 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1703 | multipath_target.version[0], multipath_target.version[1], |
1704 | multipath_target.version[2]); | |
1705 | ||
1706 | return r; | |
1707 | } | |
1708 | ||
1709 | static void __exit dm_multipath_exit(void) | |
1710 | { | |
bab7cfc7 | 1711 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1712 | destroy_workqueue(kmultipathd); |
1713 | ||
10d3bd09 | 1714 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1715 | kmem_cache_destroy(_mpio_cache); |
1716 | } | |
1717 | ||
1da177e4 LT |
1718 | module_init(dm_multipath_init); |
1719 | module_exit(dm_multipath_exit); | |
1720 | ||
1721 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1722 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1723 | MODULE_LICENSE("GPL"); |