Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software Limited. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
586e80e6 MP |
8 | #include <linux/device-mapper.h> |
9 | ||
f4790826 | 10 | #include "dm.h" |
1da177e4 | 11 | #include "dm-path-selector.h" |
b15546f9 | 12 | #include "dm-uevent.h" |
1da177e4 LT |
13 | |
14 | #include <linux/ctype.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/mempool.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/pagemap.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/time.h> | |
21 | #include <linux/workqueue.h> | |
35991652 | 22 | #include <linux/delay.h> |
cfae5c9b | 23 | #include <scsi/scsi_dh.h> |
60063497 | 24 | #include <linux/atomic.h> |
1da177e4 | 25 | |
72d94861 | 26 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
27 | #define DM_PG_INIT_DELAY_MSECS 2000 |
28 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
1da177e4 LT |
29 | |
30 | /* Path properties */ | |
31 | struct pgpath { | |
32 | struct list_head list; | |
33 | ||
34 | struct priority_group *pg; /* Owning PG */ | |
6680073d | 35 | unsigned is_active; /* Path status */ |
1da177e4 LT |
36 | unsigned fail_count; /* Cumulative failure count */ |
37 | ||
c922d5f7 | 38 | struct dm_path path; |
4e2d19e4 | 39 | struct delayed_work activate_path; |
1da177e4 LT |
40 | }; |
41 | ||
42 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
43 | ||
44 | /* | |
45 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
46 | * Each has a path selector which controls which path gets used. | |
47 | */ | |
48 | struct priority_group { | |
49 | struct list_head list; | |
50 | ||
51 | struct multipath *m; /* Owning multipath instance */ | |
52 | struct path_selector ps; | |
53 | ||
54 | unsigned pg_num; /* Reference number */ | |
55 | unsigned bypassed; /* Temporarily bypass this PG? */ | |
56 | ||
57 | unsigned nr_pgpaths; /* Number of paths in PG */ | |
58 | struct list_head pgpaths; | |
59 | }; | |
60 | ||
61 | /* Multipath context */ | |
62 | struct multipath { | |
63 | struct list_head list; | |
64 | struct dm_target *ti; | |
65 | ||
cfae5c9b | 66 | const char *hw_handler_name; |
2bfd2e13 | 67 | char *hw_handler_params; |
4e2d19e4 | 68 | |
1fbdd2b3 MS |
69 | spinlock_t lock; |
70 | ||
1da177e4 LT |
71 | unsigned nr_priority_groups; |
72 | struct list_head priority_groups; | |
4e2d19e4 CS |
73 | |
74 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
75 | ||
1da177e4 | 76 | unsigned pg_init_required; /* pg_init needs calling? */ |
c3cd4f6b | 77 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
4e2d19e4 | 78 | unsigned pg_init_delay_retry; /* Delay pg_init retry? */ |
1da177e4 LT |
79 | |
80 | unsigned nr_valid_paths; /* Total number of usable paths */ | |
81 | struct pgpath *current_pgpath; | |
82 | struct priority_group *current_pg; | |
83 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
84 | unsigned repeat_count; /* I/Os left before calling PS again */ | |
85 | ||
1fbdd2b3 MS |
86 | unsigned queue_io:1; /* Must we queue all I/O? */ |
87 | unsigned queue_if_no_path:1; /* Queue I/O if last path fails? */ | |
88 | unsigned saved_queue_if_no_path:1; /* Saved state during suspension */ | |
a58a935d | 89 | unsigned retain_attached_hw_handler:1; /* If there's already a hw_handler present, don't change it. */ |
954a73d5 | 90 | unsigned pg_init_disabled:1; /* pg_init is not currently allowed */ |
1fbdd2b3 | 91 | |
c9e45581 DW |
92 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
93 | unsigned pg_init_count; /* Number of times pg_init called */ | |
4e2d19e4 | 94 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
1da177e4 | 95 | |
1da177e4 LT |
96 | struct work_struct trigger_event; |
97 | ||
98 | /* | |
028867ac | 99 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
100 | * can resubmit bios on error. |
101 | */ | |
102 | mempool_t *mpio_pool; | |
6380f26f MA |
103 | |
104 | struct mutex work_mutex; | |
1da177e4 LT |
105 | }; |
106 | ||
107 | /* | |
108 | * Context information attached to each bio we process. | |
109 | */ | |
028867ac | 110 | struct dm_mpath_io { |
1da177e4 | 111 | struct pgpath *pgpath; |
02ab823f | 112 | size_t nr_bytes; |
1da177e4 LT |
113 | }; |
114 | ||
115 | typedef int (*action_fn) (struct pgpath *pgpath); | |
116 | ||
e18b890b | 117 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 118 | |
bab7cfc7 | 119 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 | 120 | static void trigger_event(struct work_struct *work); |
bab7cfc7 | 121 | static void activate_path(struct work_struct *work); |
e8099177 | 122 | static int __pgpath_busy(struct pgpath *pgpath); |
1da177e4 LT |
123 | |
124 | ||
125 | /*----------------------------------------------- | |
126 | * Allocation routines | |
127 | *-----------------------------------------------*/ | |
128 | ||
129 | static struct pgpath *alloc_pgpath(void) | |
130 | { | |
e69fae56 | 131 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 132 | |
224cb3e9 | 133 | if (pgpath) { |
6680073d | 134 | pgpath->is_active = 1; |
4e2d19e4 | 135 | INIT_DELAYED_WORK(&pgpath->activate_path, activate_path); |
224cb3e9 | 136 | } |
1da177e4 LT |
137 | |
138 | return pgpath; | |
139 | } | |
140 | ||
028867ac | 141 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
142 | { |
143 | kfree(pgpath); | |
144 | } | |
145 | ||
146 | static struct priority_group *alloc_priority_group(void) | |
147 | { | |
148 | struct priority_group *pg; | |
149 | ||
e69fae56 | 150 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 151 | |
e69fae56 MM |
152 | if (pg) |
153 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
154 | |
155 | return pg; | |
156 | } | |
157 | ||
158 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
159 | { | |
160 | struct pgpath *pgpath, *tmp; | |
ae11b1b3 | 161 | struct multipath *m = ti->private; |
1da177e4 LT |
162 | |
163 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
164 | list_del(&pgpath->list); | |
ae11b1b3 HR |
165 | if (m->hw_handler_name) |
166 | scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev)); | |
1da177e4 LT |
167 | dm_put_device(ti, pgpath->path.dev); |
168 | free_pgpath(pgpath); | |
169 | } | |
170 | } | |
171 | ||
172 | static void free_priority_group(struct priority_group *pg, | |
173 | struct dm_target *ti) | |
174 | { | |
175 | struct path_selector *ps = &pg->ps; | |
176 | ||
177 | if (ps->type) { | |
178 | ps->type->destroy(ps); | |
179 | dm_put_path_selector(ps->type); | |
180 | } | |
181 | ||
182 | free_pgpaths(&pg->pgpaths, ti); | |
183 | kfree(pg); | |
184 | } | |
185 | ||
28f16c20 | 186 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
187 | { |
188 | struct multipath *m; | |
f4790826 | 189 | unsigned min_ios = dm_get_reserved_rq_based_ios(); |
1da177e4 | 190 | |
e69fae56 | 191 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 192 | if (m) { |
1da177e4 LT |
193 | INIT_LIST_HEAD(&m->priority_groups); |
194 | spin_lock_init(&m->lock); | |
195 | m->queue_io = 1; | |
4e2d19e4 | 196 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; |
c4028958 | 197 | INIT_WORK(&m->trigger_event, trigger_event); |
2bded7bd | 198 | init_waitqueue_head(&m->pg_init_wait); |
6380f26f | 199 | mutex_init(&m->work_mutex); |
f4790826 | 200 | m->mpio_pool = mempool_create_slab_pool(min_ios, _mpio_cache); |
1da177e4 LT |
201 | if (!m->mpio_pool) { |
202 | kfree(m); | |
203 | return NULL; | |
204 | } | |
28f16c20 MM |
205 | m->ti = ti; |
206 | ti->private = m; | |
1da177e4 LT |
207 | } |
208 | ||
209 | return m; | |
210 | } | |
211 | ||
212 | static void free_multipath(struct multipath *m) | |
213 | { | |
214 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
215 | |
216 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
217 | list_del(&pg->list); | |
218 | free_priority_group(pg, m->ti); | |
219 | } | |
220 | ||
cfae5c9b | 221 | kfree(m->hw_handler_name); |
2bfd2e13 | 222 | kfree(m->hw_handler_params); |
1da177e4 LT |
223 | mempool_destroy(m->mpio_pool); |
224 | kfree(m); | |
225 | } | |
226 | ||
466891f9 JN |
227 | static int set_mapinfo(struct multipath *m, union map_info *info) |
228 | { | |
229 | struct dm_mpath_io *mpio; | |
230 | ||
231 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); | |
232 | if (!mpio) | |
233 | return -ENOMEM; | |
234 | ||
235 | memset(mpio, 0, sizeof(*mpio)); | |
236 | info->ptr = mpio; | |
237 | ||
238 | return 0; | |
239 | } | |
240 | ||
241 | static void clear_mapinfo(struct multipath *m, union map_info *info) | |
242 | { | |
243 | struct dm_mpath_io *mpio = info->ptr; | |
244 | ||
245 | info->ptr = NULL; | |
246 | mempool_free(mpio, m->mpio_pool); | |
247 | } | |
1da177e4 LT |
248 | |
249 | /*----------------------------------------------- | |
250 | * Path selection | |
251 | *-----------------------------------------------*/ | |
252 | ||
3e9f1be1 | 253 | static int __pg_init_all_paths(struct multipath *m) |
fb612642 KU |
254 | { |
255 | struct pgpath *pgpath; | |
4e2d19e4 | 256 | unsigned long pg_init_delay = 0; |
fb612642 | 257 | |
17f4ff45 | 258 | if (m->pg_init_in_progress || m->pg_init_disabled) |
3e9f1be1 | 259 | return 0; |
17f4ff45 | 260 | |
fb612642 KU |
261 | m->pg_init_count++; |
262 | m->pg_init_required = 0; | |
3e9f1be1 HR |
263 | |
264 | /* Check here to reset pg_init_required */ | |
265 | if (!m->current_pg) | |
266 | return 0; | |
267 | ||
4e2d19e4 CS |
268 | if (m->pg_init_delay_retry) |
269 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? | |
270 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
271 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
272 | /* Skip failed paths */ | |
273 | if (!pgpath->is_active) | |
274 | continue; | |
4e2d19e4 CS |
275 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
276 | pg_init_delay)) | |
fb612642 KU |
277 | m->pg_init_in_progress++; |
278 | } | |
3e9f1be1 | 279 | return m->pg_init_in_progress; |
fb612642 KU |
280 | } |
281 | ||
1da177e4 LT |
282 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) |
283 | { | |
1da177e4 LT |
284 | m->current_pg = pgpath->pg; |
285 | ||
286 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 287 | if (m->hw_handler_name) { |
1da177e4 LT |
288 | m->pg_init_required = 1; |
289 | m->queue_io = 1; | |
290 | } else { | |
291 | m->pg_init_required = 0; | |
292 | m->queue_io = 0; | |
293 | } | |
c9e45581 DW |
294 | |
295 | m->pg_init_count = 0; | |
1da177e4 LT |
296 | } |
297 | ||
02ab823f KU |
298 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
299 | size_t nr_bytes) | |
1da177e4 | 300 | { |
c922d5f7 | 301 | struct dm_path *path; |
1da177e4 | 302 | |
02ab823f | 303 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes); |
1da177e4 LT |
304 | if (!path) |
305 | return -ENXIO; | |
306 | ||
307 | m->current_pgpath = path_to_pgpath(path); | |
308 | ||
309 | if (m->current_pg != pg) | |
310 | __switch_pg(m, m->current_pgpath); | |
311 | ||
312 | return 0; | |
313 | } | |
314 | ||
02ab823f | 315 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
316 | { |
317 | struct priority_group *pg; | |
318 | unsigned bypassed = 1; | |
319 | ||
320 | if (!m->nr_valid_paths) | |
321 | goto failed; | |
322 | ||
323 | /* Were we instructed to switch PG? */ | |
324 | if (m->next_pg) { | |
325 | pg = m->next_pg; | |
326 | m->next_pg = NULL; | |
02ab823f | 327 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
328 | return; |
329 | } | |
330 | ||
331 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 332 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
333 | return; |
334 | ||
335 | /* | |
336 | * Loop through priority groups until we find a valid path. | |
337 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
338 | * Second time we only try the ones we skipped, but set |
339 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
340 | */ |
341 | do { | |
342 | list_for_each_entry(pg, &m->priority_groups, list) { | |
343 | if (pg->bypassed == bypassed) | |
344 | continue; | |
f220fd4e MC |
345 | if (!__choose_path_in_pg(m, pg, nr_bytes)) { |
346 | if (!bypassed) | |
347 | m->pg_init_delay_retry = 1; | |
1da177e4 | 348 | return; |
f220fd4e | 349 | } |
1da177e4 LT |
350 | } |
351 | } while (bypassed--); | |
352 | ||
353 | failed: | |
354 | m->current_pgpath = NULL; | |
355 | m->current_pg = NULL; | |
356 | } | |
357 | ||
45e15720 KU |
358 | /* |
359 | * Check whether bios must be queued in the device-mapper core rather | |
360 | * than here in the target. | |
361 | * | |
362 | * m->lock must be held on entry. | |
363 | * | |
364 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
365 | * same value then we are not between multipath_presuspend() | |
366 | * and multipath_resume() calls and we have no need to check | |
367 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
368 | */ | |
369 | static int __must_push_back(struct multipath *m) | |
370 | { | |
e8099177 HR |
371 | return (m->queue_if_no_path || |
372 | (m->queue_if_no_path != m->saved_queue_if_no_path && | |
373 | dm_noflush_suspending(m->ti))); | |
45e15720 KU |
374 | } |
375 | ||
e8099177 HR |
376 | #define pg_ready(m) (!(m)->queue_io && !(m)->pg_init_required) |
377 | ||
36fcffcc HR |
378 | /* |
379 | * Map cloned requests | |
380 | */ | |
381 | static int multipath_map(struct dm_target *ti, struct request *clone, | |
382 | union map_info *map_context) | |
1da177e4 | 383 | { |
36fcffcc | 384 | struct multipath *m = (struct multipath *) ti->private; |
e3bde04f | 385 | int r = DM_MAPIO_REQUEUE; |
f40c67f0 | 386 | size_t nr_bytes = blk_rq_bytes(clone); |
1da177e4 LT |
387 | unsigned long flags; |
388 | struct pgpath *pgpath; | |
f40c67f0 | 389 | struct block_device *bdev; |
e3bde04f | 390 | struct dm_mpath_io *mpio; |
1da177e4 LT |
391 | |
392 | spin_lock_irqsave(&m->lock, flags); | |
393 | ||
394 | /* Do we need to select a new pgpath? */ | |
395 | if (!m->current_pgpath || | |
396 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
02ab823f | 397 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
398 | |
399 | pgpath = m->current_pgpath; | |
400 | ||
9bf59a61 MS |
401 | if (!pgpath) { |
402 | if (!__must_push_back(m)) | |
403 | r = -EIO; /* Failed */ | |
404 | goto out_unlock; | |
405 | } | |
406 | if (!pg_ready(m)) { | |
e3bde04f | 407 | __pg_init_all_paths(m); |
9bf59a61 MS |
408 | goto out_unlock; |
409 | } | |
410 | if (set_mapinfo(m, map_context) < 0) | |
411 | /* ENOMEM, requeue */ | |
412 | goto out_unlock; | |
413 | ||
414 | bdev = pgpath->path.dev->bdev; | |
415 | clone->q = bdev_get_queue(bdev); | |
416 | clone->rq_disk = bdev->bd_disk; | |
417 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
418 | mpio = map_context->ptr; | |
419 | mpio->pgpath = pgpath; | |
420 | mpio->nr_bytes = nr_bytes; | |
421 | if (pgpath->pg->ps.type->start_io) | |
422 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
423 | &pgpath->path, | |
424 | nr_bytes); | |
425 | r = DM_MAPIO_REMAPPED; | |
1da177e4 | 426 | |
e3bde04f | 427 | out_unlock: |
1da177e4 LT |
428 | spin_unlock_irqrestore(&m->lock, flags); |
429 | ||
430 | return r; | |
431 | } | |
432 | ||
433 | /* | |
434 | * If we run out of usable paths, should we queue I/O or error it? | |
435 | */ | |
485ef69e AK |
436 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
437 | unsigned save_old_value) | |
1da177e4 LT |
438 | { |
439 | unsigned long flags; | |
440 | ||
441 | spin_lock_irqsave(&m->lock, flags); | |
442 | ||
485ef69e AK |
443 | if (save_old_value) |
444 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
445 | else | |
446 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 447 | m->queue_if_no_path = queue_if_no_path; |
e8099177 | 448 | if (!m->queue_if_no_path) |
3e9f1be1 | 449 | dm_table_run_md_queue_async(m->ti->table); |
1da177e4 LT |
450 | |
451 | spin_unlock_irqrestore(&m->lock, flags); | |
452 | ||
453 | return 0; | |
454 | } | |
455 | ||
1da177e4 LT |
456 | /* |
457 | * An event is triggered whenever a path is taken out of use. | |
458 | * Includes path failure and PG bypass. | |
459 | */ | |
c4028958 | 460 | static void trigger_event(struct work_struct *work) |
1da177e4 | 461 | { |
c4028958 DH |
462 | struct multipath *m = |
463 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
464 | |
465 | dm_table_event(m->ti->table); | |
466 | } | |
467 | ||
468 | /*----------------------------------------------------------------- | |
469 | * Constructor/argument parsing: | |
470 | * <#multipath feature args> [<arg>]* | |
471 | * <#hw_handler args> [hw_handler [<arg>]*] | |
472 | * <#priority groups> | |
473 | * <initial priority group> | |
474 | * [<selector> <#selector args> [<arg>]* | |
475 | * <#paths> <#per-path selector args> | |
476 | * [<path> [<arg>]* ]+ ]+ | |
477 | *---------------------------------------------------------------*/ | |
498f0103 | 478 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
479 | struct dm_target *ti) |
480 | { | |
481 | int r; | |
482 | struct path_selector_type *pst; | |
483 | unsigned ps_argc; | |
484 | ||
498f0103 | 485 | static struct dm_arg _args[] = { |
72d94861 | 486 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
487 | }; |
488 | ||
498f0103 | 489 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 490 | if (!pst) { |
72d94861 | 491 | ti->error = "unknown path selector type"; |
1da177e4 LT |
492 | return -EINVAL; |
493 | } | |
494 | ||
498f0103 | 495 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
496 | if (r) { |
497 | dm_put_path_selector(pst); | |
1da177e4 | 498 | return -EINVAL; |
371b2e34 | 499 | } |
1da177e4 LT |
500 | |
501 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
502 | if (r) { | |
503 | dm_put_path_selector(pst); | |
72d94861 | 504 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
505 | return r; |
506 | } | |
507 | ||
508 | pg->ps.type = pst; | |
498f0103 | 509 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
510 | |
511 | return 0; | |
512 | } | |
513 | ||
498f0103 | 514 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
515 | struct dm_target *ti) |
516 | { | |
517 | int r; | |
518 | struct pgpath *p; | |
ae11b1b3 | 519 | struct multipath *m = ti->private; |
a58a935d MS |
520 | struct request_queue *q = NULL; |
521 | const char *attached_handler_name; | |
1da177e4 LT |
522 | |
523 | /* we need at least a path arg */ | |
524 | if (as->argc < 1) { | |
72d94861 | 525 | ti->error = "no device given"; |
01460f35 | 526 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
527 | } |
528 | ||
529 | p = alloc_pgpath(); | |
530 | if (!p) | |
01460f35 | 531 | return ERR_PTR(-ENOMEM); |
1da177e4 | 532 | |
498f0103 | 533 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 534 | &p->path.dev); |
1da177e4 | 535 | if (r) { |
72d94861 | 536 | ti->error = "error getting device"; |
1da177e4 LT |
537 | goto bad; |
538 | } | |
539 | ||
a58a935d MS |
540 | if (m->retain_attached_hw_handler || m->hw_handler_name) |
541 | q = bdev_get_queue(p->path.dev->bdev); | |
542 | ||
543 | if (m->retain_attached_hw_handler) { | |
544 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); | |
545 | if (attached_handler_name) { | |
546 | /* | |
547 | * Reset hw_handler_name to match the attached handler | |
548 | * and clear any hw_handler_params associated with the | |
549 | * ignored handler. | |
550 | * | |
551 | * NB. This modifies the table line to show the actual | |
552 | * handler instead of the original table passed in. | |
553 | */ | |
554 | kfree(m->hw_handler_name); | |
555 | m->hw_handler_name = attached_handler_name; | |
556 | ||
557 | kfree(m->hw_handler_params); | |
558 | m->hw_handler_params = NULL; | |
559 | } | |
560 | } | |
a0cf7ea9 | 561 | |
a58a935d MS |
562 | if (m->hw_handler_name) { |
563 | /* | |
564 | * Increments scsi_dh reference, even when using an | |
565 | * already-attached handler. | |
566 | */ | |
a0cf7ea9 HR |
567 | r = scsi_dh_attach(q, m->hw_handler_name); |
568 | if (r == -EBUSY) { | |
569 | /* | |
a58a935d | 570 | * Already attached to different hw_handler: |
a0cf7ea9 HR |
571 | * try to reattach with correct one. |
572 | */ | |
573 | scsi_dh_detach(q); | |
574 | r = scsi_dh_attach(q, m->hw_handler_name); | |
575 | } | |
576 | ||
ae11b1b3 | 577 | if (r < 0) { |
a0cf7ea9 | 578 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
579 | dm_put_device(ti, p->path.dev); |
580 | goto bad; | |
581 | } | |
2bfd2e13 CS |
582 | |
583 | if (m->hw_handler_params) { | |
584 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
585 | if (r < 0) { | |
586 | ti->error = "unable to set hardware " | |
587 | "handler parameters"; | |
588 | scsi_dh_detach(q); | |
589 | dm_put_device(ti, p->path.dev); | |
590 | goto bad; | |
591 | } | |
592 | } | |
ae11b1b3 HR |
593 | } |
594 | ||
1da177e4 LT |
595 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
596 | if (r) { | |
597 | dm_put_device(ti, p->path.dev); | |
598 | goto bad; | |
599 | } | |
600 | ||
601 | return p; | |
602 | ||
603 | bad: | |
604 | free_pgpath(p); | |
01460f35 | 605 | return ERR_PTR(r); |
1da177e4 LT |
606 | } |
607 | ||
498f0103 | 608 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 609 | struct multipath *m) |
1da177e4 | 610 | { |
498f0103 | 611 | static struct dm_arg _args[] = { |
72d94861 AK |
612 | {1, 1024, "invalid number of paths"}, |
613 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
614 | }; |
615 | ||
616 | int r; | |
498f0103 | 617 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 618 | struct priority_group *pg; |
28f16c20 | 619 | struct dm_target *ti = m->ti; |
1da177e4 LT |
620 | |
621 | if (as->argc < 2) { | |
622 | as->argc = 0; | |
01460f35 BM |
623 | ti->error = "not enough priority group arguments"; |
624 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
625 | } |
626 | ||
627 | pg = alloc_priority_group(); | |
628 | if (!pg) { | |
72d94861 | 629 | ti->error = "couldn't allocate priority group"; |
01460f35 | 630 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
631 | } |
632 | pg->m = m; | |
633 | ||
634 | r = parse_path_selector(as, pg, ti); | |
635 | if (r) | |
636 | goto bad; | |
637 | ||
638 | /* | |
639 | * read the paths | |
640 | */ | |
498f0103 | 641 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
642 | if (r) |
643 | goto bad; | |
644 | ||
498f0103 | 645 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
646 | if (r) |
647 | goto bad; | |
648 | ||
498f0103 | 649 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
650 | for (i = 0; i < pg->nr_pgpaths; i++) { |
651 | struct pgpath *pgpath; | |
498f0103 | 652 | struct dm_arg_set path_args; |
1da177e4 | 653 | |
498f0103 | 654 | if (as->argc < nr_args) { |
148acff6 | 655 | ti->error = "not enough path parameters"; |
6bbf79a1 | 656 | r = -EINVAL; |
1da177e4 | 657 | goto bad; |
148acff6 | 658 | } |
1da177e4 | 659 | |
498f0103 | 660 | path_args.argc = nr_args; |
1da177e4 LT |
661 | path_args.argv = as->argv; |
662 | ||
663 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
664 | if (IS_ERR(pgpath)) { |
665 | r = PTR_ERR(pgpath); | |
1da177e4 | 666 | goto bad; |
01460f35 | 667 | } |
1da177e4 LT |
668 | |
669 | pgpath->pg = pg; | |
670 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 671 | dm_consume_args(as, nr_args); |
1da177e4 LT |
672 | } |
673 | ||
674 | return pg; | |
675 | ||
676 | bad: | |
677 | free_priority_group(pg, ti); | |
01460f35 | 678 | return ERR_PTR(r); |
1da177e4 LT |
679 | } |
680 | ||
498f0103 | 681 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 682 | { |
1da177e4 | 683 | unsigned hw_argc; |
2bfd2e13 | 684 | int ret; |
28f16c20 | 685 | struct dm_target *ti = m->ti; |
1da177e4 | 686 | |
498f0103 | 687 | static struct dm_arg _args[] = { |
72d94861 | 688 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
689 | }; |
690 | ||
498f0103 | 691 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
692 | return -EINVAL; |
693 | ||
694 | if (!hw_argc) | |
695 | return 0; | |
696 | ||
498f0103 | 697 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
510193a2 MS |
698 | if (!try_then_request_module(scsi_dh_handler_exist(m->hw_handler_name), |
699 | "scsi_dh_%s", m->hw_handler_name)) { | |
72d94861 | 700 | ti->error = "unknown hardware handler type"; |
2bfd2e13 CS |
701 | ret = -EINVAL; |
702 | goto fail; | |
1da177e4 | 703 | } |
14e98c5c | 704 | |
2bfd2e13 CS |
705 | if (hw_argc > 1) { |
706 | char *p; | |
707 | int i, j, len = 4; | |
708 | ||
709 | for (i = 0; i <= hw_argc - 2; i++) | |
710 | len += strlen(as->argv[i]) + 1; | |
711 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
712 | if (!p) { | |
713 | ti->error = "memory allocation failed"; | |
714 | ret = -ENOMEM; | |
715 | goto fail; | |
716 | } | |
717 | j = sprintf(p, "%d", hw_argc - 1); | |
718 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
719 | j = sprintf(p, "%s", as->argv[i]); | |
720 | } | |
498f0103 | 721 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
722 | |
723 | return 0; | |
2bfd2e13 CS |
724 | fail: |
725 | kfree(m->hw_handler_name); | |
726 | m->hw_handler_name = NULL; | |
727 | return ret; | |
1da177e4 LT |
728 | } |
729 | ||
498f0103 | 730 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
731 | { |
732 | int r; | |
733 | unsigned argc; | |
28f16c20 | 734 | struct dm_target *ti = m->ti; |
498f0103 | 735 | const char *arg_name; |
1da177e4 | 736 | |
498f0103 | 737 | static struct dm_arg _args[] = { |
a58a935d | 738 | {0, 6, "invalid number of feature args"}, |
c9e45581 | 739 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 740 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
741 | }; |
742 | ||
498f0103 | 743 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
744 | if (r) |
745 | return -EINVAL; | |
746 | ||
747 | if (!argc) | |
748 | return 0; | |
749 | ||
c9e45581 | 750 | do { |
498f0103 | 751 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
752 | argc--; |
753 | ||
498f0103 | 754 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
c9e45581 DW |
755 | r = queue_if_no_path(m, 1, 0); |
756 | continue; | |
757 | } | |
758 | ||
a58a935d MS |
759 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
760 | m->retain_attached_hw_handler = 1; | |
761 | continue; | |
762 | } | |
763 | ||
498f0103 | 764 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 765 | (argc >= 1)) { |
498f0103 | 766 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
767 | argc--; |
768 | continue; | |
769 | } | |
770 | ||
498f0103 | 771 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 772 | (argc >= 1)) { |
498f0103 | 773 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
774 | argc--; |
775 | continue; | |
776 | } | |
777 | ||
1da177e4 | 778 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
779 | r = -EINVAL; |
780 | } while (argc && !r); | |
781 | ||
782 | return r; | |
1da177e4 LT |
783 | } |
784 | ||
785 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
786 | char **argv) | |
787 | { | |
498f0103 MS |
788 | /* target arguments */ |
789 | static struct dm_arg _args[] = { | |
a490a07a MS |
790 | {0, 1024, "invalid number of priority groups"}, |
791 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
792 | }; |
793 | ||
794 | int r; | |
795 | struct multipath *m; | |
498f0103 | 796 | struct dm_arg_set as; |
1da177e4 LT |
797 | unsigned pg_count = 0; |
798 | unsigned next_pg_num; | |
799 | ||
800 | as.argc = argc; | |
801 | as.argv = argv; | |
802 | ||
28f16c20 | 803 | m = alloc_multipath(ti); |
1da177e4 | 804 | if (!m) { |
72d94861 | 805 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
806 | return -EINVAL; |
807 | } | |
808 | ||
28f16c20 | 809 | r = parse_features(&as, m); |
1da177e4 LT |
810 | if (r) |
811 | goto bad; | |
812 | ||
28f16c20 | 813 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
814 | if (r) |
815 | goto bad; | |
816 | ||
498f0103 | 817 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
818 | if (r) |
819 | goto bad; | |
820 | ||
498f0103 | 821 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
822 | if (r) |
823 | goto bad; | |
824 | ||
a490a07a MS |
825 | if ((!m->nr_priority_groups && next_pg_num) || |
826 | (m->nr_priority_groups && !next_pg_num)) { | |
827 | ti->error = "invalid initial priority group"; | |
828 | r = -EINVAL; | |
829 | goto bad; | |
830 | } | |
831 | ||
1da177e4 LT |
832 | /* parse the priority groups */ |
833 | while (as.argc) { | |
834 | struct priority_group *pg; | |
835 | ||
28f16c20 | 836 | pg = parse_priority_group(&as, m); |
01460f35 BM |
837 | if (IS_ERR(pg)) { |
838 | r = PTR_ERR(pg); | |
1da177e4 LT |
839 | goto bad; |
840 | } | |
841 | ||
842 | m->nr_valid_paths += pg->nr_pgpaths; | |
843 | list_add_tail(&pg->list, &m->priority_groups); | |
844 | pg_count++; | |
845 | pg->pg_num = pg_count; | |
846 | if (!--next_pg_num) | |
847 | m->next_pg = pg; | |
848 | } | |
849 | ||
850 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 851 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
852 | r = -EINVAL; |
853 | goto bad; | |
854 | } | |
855 | ||
55a62eef AK |
856 | ti->num_flush_bios = 1; |
857 | ti->num_discard_bios = 1; | |
042bcef8 | 858 | ti->num_write_same_bios = 1; |
8627921f | 859 | |
1da177e4 LT |
860 | return 0; |
861 | ||
862 | bad: | |
863 | free_multipath(m); | |
864 | return r; | |
865 | } | |
866 | ||
2bded7bd KU |
867 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
868 | { | |
869 | DECLARE_WAITQUEUE(wait, current); | |
870 | unsigned long flags; | |
871 | ||
872 | add_wait_queue(&m->pg_init_wait, &wait); | |
873 | ||
874 | while (1) { | |
875 | set_current_state(TASK_UNINTERRUPTIBLE); | |
876 | ||
877 | spin_lock_irqsave(&m->lock, flags); | |
878 | if (!m->pg_init_in_progress) { | |
879 | spin_unlock_irqrestore(&m->lock, flags); | |
880 | break; | |
881 | } | |
882 | spin_unlock_irqrestore(&m->lock, flags); | |
883 | ||
884 | io_schedule(); | |
885 | } | |
886 | set_current_state(TASK_RUNNING); | |
887 | ||
888 | remove_wait_queue(&m->pg_init_wait, &wait); | |
889 | } | |
890 | ||
891 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 892 | { |
954a73d5 SKM |
893 | unsigned long flags; |
894 | ||
895 | spin_lock_irqsave(&m->lock, flags); | |
896 | m->pg_init_disabled = 1; | |
897 | spin_unlock_irqrestore(&m->lock, flags); | |
898 | ||
bab7cfc7 | 899 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 900 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 901 | flush_workqueue(kmultipathd); |
43829731 | 902 | flush_work(&m->trigger_event); |
954a73d5 SKM |
903 | |
904 | spin_lock_irqsave(&m->lock, flags); | |
905 | m->pg_init_disabled = 0; | |
906 | spin_unlock_irqrestore(&m->lock, flags); | |
6df400ab KU |
907 | } |
908 | ||
909 | static void multipath_dtr(struct dm_target *ti) | |
910 | { | |
911 | struct multipath *m = ti->private; | |
912 | ||
2bded7bd | 913 | flush_multipath_work(m); |
1da177e4 LT |
914 | free_multipath(m); |
915 | } | |
916 | ||
1da177e4 LT |
917 | /* |
918 | * Take a path out of use. | |
919 | */ | |
920 | static int fail_path(struct pgpath *pgpath) | |
921 | { | |
922 | unsigned long flags; | |
923 | struct multipath *m = pgpath->pg->m; | |
924 | ||
925 | spin_lock_irqsave(&m->lock, flags); | |
926 | ||
6680073d | 927 | if (!pgpath->is_active) |
1da177e4 LT |
928 | goto out; |
929 | ||
72d94861 | 930 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
931 | |
932 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
6680073d | 933 | pgpath->is_active = 0; |
1da177e4 LT |
934 | pgpath->fail_count++; |
935 | ||
936 | m->nr_valid_paths--; | |
937 | ||
938 | if (pgpath == m->current_pgpath) | |
939 | m->current_pgpath = NULL; | |
940 | ||
b15546f9 MA |
941 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
942 | pgpath->path.dev->name, m->nr_valid_paths); | |
943 | ||
fe9cf30e | 944 | schedule_work(&m->trigger_event); |
1da177e4 LT |
945 | |
946 | out: | |
947 | spin_unlock_irqrestore(&m->lock, flags); | |
948 | ||
949 | return 0; | |
950 | } | |
951 | ||
952 | /* | |
953 | * Reinstate a previously-failed path | |
954 | */ | |
955 | static int reinstate_path(struct pgpath *pgpath) | |
956 | { | |
957 | int r = 0; | |
958 | unsigned long flags; | |
959 | struct multipath *m = pgpath->pg->m; | |
960 | ||
961 | spin_lock_irqsave(&m->lock, flags); | |
962 | ||
6680073d | 963 | if (pgpath->is_active) |
1da177e4 LT |
964 | goto out; |
965 | ||
def052d2 | 966 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
967 | DMWARN("Reinstate path not supported by path selector %s", |
968 | pgpath->pg->ps.type->name); | |
969 | r = -EINVAL; | |
970 | goto out; | |
971 | } | |
972 | ||
973 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
974 | if (r) | |
975 | goto out; | |
976 | ||
6680073d | 977 | pgpath->is_active = 1; |
1da177e4 | 978 | |
e8099177 | 979 | if (!m->nr_valid_paths++) { |
e54f77dd | 980 | m->current_pgpath = NULL; |
3e9f1be1 | 981 | dm_table_run_md_queue_async(m->ti->table); |
e54f77dd | 982 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 983 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
e54f77dd CS |
984 | m->pg_init_in_progress++; |
985 | } | |
1da177e4 | 986 | |
b15546f9 MA |
987 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
988 | pgpath->path.dev->name, m->nr_valid_paths); | |
989 | ||
fe9cf30e | 990 | schedule_work(&m->trigger_event); |
1da177e4 LT |
991 | |
992 | out: | |
993 | spin_unlock_irqrestore(&m->lock, flags); | |
994 | ||
995 | return r; | |
996 | } | |
997 | ||
998 | /* | |
999 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1000 | */ | |
1001 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1002 | action_fn action) | |
1003 | { | |
19040c0b | 1004 | int r = -EINVAL; |
1da177e4 LT |
1005 | struct pgpath *pgpath; |
1006 | struct priority_group *pg; | |
1007 | ||
1008 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1009 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1010 | if (pgpath->path.dev == dev) | |
1011 | r = action(pgpath); | |
1012 | } | |
1013 | } | |
1014 | ||
1015 | return r; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * Temporarily try to avoid having to use the specified PG | |
1020 | */ | |
1021 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
1022 | int bypassed) | |
1023 | { | |
1024 | unsigned long flags; | |
1025 | ||
1026 | spin_lock_irqsave(&m->lock, flags); | |
1027 | ||
1028 | pg->bypassed = bypassed; | |
1029 | m->current_pgpath = NULL; | |
1030 | m->current_pg = NULL; | |
1031 | ||
1032 | spin_unlock_irqrestore(&m->lock, flags); | |
1033 | ||
fe9cf30e | 1034 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1035 | } |
1036 | ||
1037 | /* | |
1038 | * Switch to using the specified PG from the next I/O that gets mapped | |
1039 | */ | |
1040 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1041 | { | |
1042 | struct priority_group *pg; | |
1043 | unsigned pgnum; | |
1044 | unsigned long flags; | |
31998ef1 | 1045 | char dummy; |
1da177e4 | 1046 | |
31998ef1 | 1047 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1048 | (pgnum > m->nr_priority_groups)) { |
1049 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1050 | return -EINVAL; | |
1051 | } | |
1052 | ||
1053 | spin_lock_irqsave(&m->lock, flags); | |
1054 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1055 | pg->bypassed = 0; | |
1056 | if (--pgnum) | |
1057 | continue; | |
1058 | ||
1059 | m->current_pgpath = NULL; | |
1060 | m->current_pg = NULL; | |
1061 | m->next_pg = pg; | |
1062 | } | |
1063 | spin_unlock_irqrestore(&m->lock, flags); | |
1064 | ||
fe9cf30e | 1065 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1066 | return 0; |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * Set/clear bypassed status of a PG. | |
1071 | * PGs are numbered upwards from 1 in the order they were declared. | |
1072 | */ | |
1073 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
1074 | { | |
1075 | struct priority_group *pg; | |
1076 | unsigned pgnum; | |
31998ef1 | 1077 | char dummy; |
1da177e4 | 1078 | |
31998ef1 | 1079 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1080 | (pgnum > m->nr_priority_groups)) { |
1081 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1082 | return -EINVAL; | |
1083 | } | |
1084 | ||
1085 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1086 | if (!--pgnum) | |
1087 | break; | |
1088 | } | |
1089 | ||
1090 | bypass_pg(m, pg, bypassed); | |
1091 | return 0; | |
1092 | } | |
1093 | ||
c9e45581 DW |
1094 | /* |
1095 | * Should we retry pg_init immediately? | |
1096 | */ | |
1097 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1098 | { | |
1099 | unsigned long flags; | |
1100 | int limit_reached = 0; | |
1101 | ||
1102 | spin_lock_irqsave(&m->lock, flags); | |
1103 | ||
954a73d5 | 1104 | if (m->pg_init_count <= m->pg_init_retries && !m->pg_init_disabled) |
c9e45581 DW |
1105 | m->pg_init_required = 1; |
1106 | else | |
1107 | limit_reached = 1; | |
1108 | ||
1109 | spin_unlock_irqrestore(&m->lock, flags); | |
1110 | ||
1111 | return limit_reached; | |
1112 | } | |
1113 | ||
3ae31f6a | 1114 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1115 | { |
83c0d5d5 | 1116 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1117 | struct priority_group *pg = pgpath->pg; |
1118 | struct multipath *m = pg->m; | |
1119 | unsigned long flags; | |
4e2d19e4 | 1120 | unsigned delay_retry = 0; |
cfae5c9b CS |
1121 | |
1122 | /* device or driver problems */ | |
1123 | switch (errors) { | |
1124 | case SCSI_DH_OK: | |
1125 | break; | |
1126 | case SCSI_DH_NOSYS: | |
1127 | if (!m->hw_handler_name) { | |
1128 | errors = 0; | |
1129 | break; | |
1130 | } | |
f7b934c8 MB |
1131 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1132 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1133 | /* |
1134 | * Fail path for now, so we do not ping pong | |
1135 | */ | |
1136 | fail_path(pgpath); | |
1137 | break; | |
1138 | case SCSI_DH_DEV_TEMP_BUSY: | |
1139 | /* | |
1140 | * Probably doing something like FW upgrade on the | |
1141 | * controller so try the other pg. | |
1142 | */ | |
1143 | bypass_pg(m, pg, 1); | |
1144 | break; | |
cfae5c9b | 1145 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1146 | /* Wait before retrying. */ |
1147 | delay_retry = 1; | |
cfae5c9b CS |
1148 | case SCSI_DH_IMM_RETRY: |
1149 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1150 | if (pg_init_limit_reached(m, pgpath)) | |
1151 | fail_path(pgpath); | |
1152 | errors = 0; | |
1153 | break; | |
1154 | default: | |
1155 | /* | |
1156 | * We probably do not want to fail the path for a device | |
1157 | * error, but this is what the old dm did. In future | |
1158 | * patches we can do more advanced handling. | |
1159 | */ | |
1160 | fail_path(pgpath); | |
1161 | } | |
1162 | ||
1163 | spin_lock_irqsave(&m->lock, flags); | |
1164 | if (errors) { | |
e54f77dd CS |
1165 | if (pgpath == m->current_pgpath) { |
1166 | DMERR("Could not failover device. Error %d.", errors); | |
1167 | m->current_pgpath = NULL; | |
1168 | m->current_pg = NULL; | |
1169 | } | |
d0259bf0 | 1170 | } else if (!m->pg_init_required) |
cfae5c9b | 1171 | pg->bypassed = 0; |
cfae5c9b | 1172 | |
d0259bf0 KU |
1173 | if (--m->pg_init_in_progress) |
1174 | /* Activations of other paths are still on going */ | |
1175 | goto out; | |
1176 | ||
3e9f1be1 HR |
1177 | if (m->pg_init_required) { |
1178 | m->pg_init_delay_retry = delay_retry; | |
1179 | if (__pg_init_all_paths(m)) | |
1180 | goto out; | |
1181 | } | |
1182 | m->queue_io = 0; | |
d0259bf0 | 1183 | |
2bded7bd KU |
1184 | /* |
1185 | * Wake up any thread waiting to suspend. | |
1186 | */ | |
1187 | wake_up(&m->pg_init_wait); | |
1188 | ||
d0259bf0 | 1189 | out: |
cfae5c9b CS |
1190 | spin_unlock_irqrestore(&m->lock, flags); |
1191 | } | |
1192 | ||
bab7cfc7 CS |
1193 | static void activate_path(struct work_struct *work) |
1194 | { | |
e54f77dd | 1195 | struct pgpath *pgpath = |
4e2d19e4 | 1196 | container_of(work, struct pgpath, activate_path.work); |
bab7cfc7 | 1197 | |
3a017509 HR |
1198 | if (pgpath->is_active) |
1199 | scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev), | |
1200 | pg_init_done, pgpath); | |
1201 | else | |
1202 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1203 | } |
1204 | ||
7e782af5 HR |
1205 | static int noretry_error(int error) |
1206 | { | |
1207 | switch (error) { | |
1208 | case -EOPNOTSUPP: | |
1209 | case -EREMOTEIO: | |
1210 | case -EILSEQ: | |
1211 | case -ENODATA: | |
cc9d3c38 | 1212 | case -ENOSPC: |
7e782af5 HR |
1213 | return 1; |
1214 | } | |
1215 | ||
1216 | /* Anything else could be a path failure, so should be retried */ | |
1217 | return 0; | |
1218 | } | |
1219 | ||
1da177e4 LT |
1220 | /* |
1221 | * end_io handling | |
1222 | */ | |
f40c67f0 | 1223 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1224 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1225 | { |
f40c67f0 KU |
1226 | /* |
1227 | * We don't queue any clone request inside the multipath target | |
1228 | * during end I/O handling, since those clone requests don't have | |
1229 | * bio clones. If we queue them inside the multipath target, | |
1230 | * we need to make bio clones, that requires memory allocation. | |
1231 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1232 | * don't have bio clones.) | |
1233 | * Instead of queueing the clone request here, we queue the original | |
1234 | * request into dm core, which will remake a clone request and | |
1235 | * clone bios for it and resubmit it later. | |
1236 | */ | |
1237 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1238 | unsigned long flags; |
1da177e4 | 1239 | |
f40c67f0 | 1240 | if (!error && !clone->errors) |
1da177e4 LT |
1241 | return 0; /* I/O complete */ |
1242 | ||
f84cb8a4 MS |
1243 | if (noretry_error(error)) { |
1244 | if ((clone->cmd_flags & REQ_WRITE_SAME) && | |
1245 | !clone->q->limits.max_write_same_sectors) { | |
1246 | struct queue_limits *limits; | |
1247 | ||
1248 | /* device doesn't really support WRITE SAME, disable it */ | |
1249 | limits = dm_get_queue_limits(dm_table_get_md(m->ti->table)); | |
1250 | limits->max_write_same_sectors = 0; | |
1251 | } | |
959eb4e5 | 1252 | return error; |
f84cb8a4 | 1253 | } |
959eb4e5 | 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; | |
a71a261f | 1278 | struct pgpath *pgpath; |
1da177e4 LT |
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); |
a71a261f | 1285 | pgpath = mpio->pgpath; |
1da177e4 LT |
1286 | if (pgpath) { |
1287 | ps = &pgpath->pg->ps; | |
1288 | if (ps->type->end_io) | |
02ab823f | 1289 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1290 | } |
466891f9 | 1291 | clear_mapinfo(m, map_context); |
1da177e4 LT |
1292 | |
1293 | return r; | |
1294 | } | |
1295 | ||
1296 | /* | |
1297 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1298 | * the last path fails we must error any remaining I/O. |
1299 | * Note that if the freeze_bdev fails while suspending, the | |
1300 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1301 | */ |
1302 | static void multipath_presuspend(struct dm_target *ti) | |
1303 | { | |
1304 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1305 | |
485ef69e | 1306 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1307 | } |
1308 | ||
6df400ab KU |
1309 | static void multipath_postsuspend(struct dm_target *ti) |
1310 | { | |
6380f26f MA |
1311 | struct multipath *m = ti->private; |
1312 | ||
1313 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1314 | flush_multipath_work(m); |
6380f26f | 1315 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1316 | } |
1317 | ||
436d4108 AK |
1318 | /* |
1319 | * Restore the queue_if_no_path setting. | |
1320 | */ | |
1da177e4 LT |
1321 | static void multipath_resume(struct dm_target *ti) |
1322 | { | |
1323 | struct multipath *m = (struct multipath *) ti->private; | |
1324 | unsigned long flags; | |
1325 | ||
1326 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1327 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1328 | spin_unlock_irqrestore(&m->lock, flags); |
1329 | } | |
1330 | ||
1331 | /* | |
1332 | * Info output has the following format: | |
1333 | * num_multipath_feature_args [multipath_feature_args]* | |
1334 | * num_handler_status_args [handler_status_args]* | |
1335 | * num_groups init_group_number | |
1336 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1337 | * num_paths num_selector_args | |
1338 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1339 | * | |
1340 | * Table output has the following format (identical to the constructor string): | |
1341 | * num_feature_args [features_args]* | |
1342 | * num_handler_args hw_handler [hw_handler_args]* | |
1343 | * num_groups init_group_number | |
1344 | * [priority selector-name num_ps_args [ps_args]* | |
1345 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1346 | */ | |
fd7c092e MP |
1347 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1348 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1349 | { |
1350 | int sz = 0; | |
1351 | unsigned long flags; | |
1352 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1353 | struct priority_group *pg; |
1354 | struct pgpath *p; | |
1355 | unsigned pg_num; | |
1356 | char state; | |
1357 | ||
1358 | spin_lock_irqsave(&m->lock, flags); | |
1359 | ||
1360 | /* Features */ | |
1361 | if (type == STATUSTYPE_INFO) | |
e8099177 | 1362 | DMEMIT("2 %u %u ", m->queue_io, m->pg_init_count); |
c9e45581 DW |
1363 | else { |
1364 | DMEMIT("%u ", m->queue_if_no_path + | |
4e2d19e4 | 1365 | (m->pg_init_retries > 0) * 2 + |
a58a935d MS |
1366 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
1367 | m->retain_attached_hw_handler); | |
c9e45581 DW |
1368 | if (m->queue_if_no_path) |
1369 | DMEMIT("queue_if_no_path "); | |
1370 | if (m->pg_init_retries) | |
1371 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1372 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1373 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
a58a935d MS |
1374 | if (m->retain_attached_hw_handler) |
1375 | DMEMIT("retain_attached_hw_handler "); | |
c9e45581 | 1376 | } |
1da177e4 | 1377 | |
cfae5c9b | 1378 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1379 | DMEMIT("0 "); |
1380 | else | |
cfae5c9b | 1381 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1382 | |
1383 | DMEMIT("%u ", m->nr_priority_groups); | |
1384 | ||
1385 | if (m->next_pg) | |
1386 | pg_num = m->next_pg->pg_num; | |
1387 | else if (m->current_pg) | |
1388 | pg_num = m->current_pg->pg_num; | |
1389 | else | |
a490a07a | 1390 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1391 | |
1392 | DMEMIT("%u ", pg_num); | |
1393 | ||
1394 | switch (type) { | |
1395 | case STATUSTYPE_INFO: | |
1396 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1397 | if (pg->bypassed) | |
1398 | state = 'D'; /* Disabled */ | |
1399 | else if (pg == m->current_pg) | |
1400 | state = 'A'; /* Currently Active */ | |
1401 | else | |
1402 | state = 'E'; /* Enabled */ | |
1403 | ||
1404 | DMEMIT("%c ", state); | |
1405 | ||
1406 | if (pg->ps.type->status) | |
1407 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1408 | result + sz, | |
1409 | maxlen - sz); | |
1410 | else | |
1411 | DMEMIT("0 "); | |
1412 | ||
1413 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1414 | pg->ps.type->info_args); | |
1415 | ||
1416 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1417 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1418 | p->is_active ? "A" : "F", |
1da177e4 LT |
1419 | p->fail_count); |
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 | case STATUSTYPE_TABLE: | |
1429 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1430 | DMEMIT("%s ", pg->ps.type->name); | |
1431 | ||
1432 | if (pg->ps.type->status) | |
1433 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1434 | result + sz, | |
1435 | maxlen - sz); | |
1436 | else | |
1437 | DMEMIT("0 "); | |
1438 | ||
1439 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1440 | pg->ps.type->table_args); | |
1441 | ||
1442 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1443 | DMEMIT("%s ", p->path.dev->name); | |
1444 | if (pg->ps.type->status) | |
1445 | sz += pg->ps.type->status(&pg->ps, | |
1446 | &p->path, type, result + sz, | |
1447 | maxlen - sz); | |
1448 | } | |
1449 | } | |
1450 | break; | |
1451 | } | |
1452 | ||
1453 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1454 | } |
1455 | ||
1456 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1457 | { | |
6380f26f | 1458 | int r = -EINVAL; |
1da177e4 LT |
1459 | struct dm_dev *dev; |
1460 | struct multipath *m = (struct multipath *) ti->private; | |
1461 | action_fn action; | |
1462 | ||
6380f26f MA |
1463 | mutex_lock(&m->work_mutex); |
1464 | ||
c2f3d24b KU |
1465 | if (dm_suspended(ti)) { |
1466 | r = -EBUSY; | |
1467 | goto out; | |
1468 | } | |
1469 | ||
1da177e4 | 1470 | if (argc == 1) { |
498f0103 | 1471 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
6380f26f MA |
1472 | r = queue_if_no_path(m, 1, 0); |
1473 | goto out; | |
498f0103 | 1474 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
6380f26f MA |
1475 | r = queue_if_no_path(m, 0, 0); |
1476 | goto out; | |
1477 | } | |
1da177e4 LT |
1478 | } |
1479 | ||
6380f26f | 1480 | if (argc != 2) { |
a356e426 | 1481 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1482 | goto out; |
1483 | } | |
1da177e4 | 1484 | |
498f0103 | 1485 | if (!strcasecmp(argv[0], "disable_group")) { |
6380f26f MA |
1486 | r = bypass_pg_num(m, argv[1], 1); |
1487 | goto out; | |
498f0103 | 1488 | } else if (!strcasecmp(argv[0], "enable_group")) { |
6380f26f MA |
1489 | r = bypass_pg_num(m, argv[1], 0); |
1490 | goto out; | |
498f0103 | 1491 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1492 | r = switch_pg_num(m, argv[1]); |
1493 | goto out; | |
498f0103 | 1494 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1495 | action = reinstate_path; |
498f0103 | 1496 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1497 | action = fail_path; |
6380f26f | 1498 | else { |
a356e426 | 1499 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1500 | goto out; |
1501 | } | |
1da177e4 | 1502 | |
8215d6ec | 1503 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1504 | if (r) { |
72d94861 | 1505 | DMWARN("message: error getting device %s", |
1da177e4 | 1506 | argv[1]); |
6380f26f | 1507 | goto out; |
1da177e4 LT |
1508 | } |
1509 | ||
1510 | r = action_dev(m, dev, action); | |
1511 | ||
1512 | dm_put_device(ti, dev); | |
1513 | ||
6380f26f MA |
1514 | out: |
1515 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1516 | return r; |
1da177e4 LT |
1517 | } |
1518 | ||
647b3d00 | 1519 | static int multipath_ioctl(struct dm_target *ti, unsigned int cmd, |
9af4aa30 MB |
1520 | unsigned long arg) |
1521 | { | |
35991652 | 1522 | struct multipath *m = ti->private; |
7ba10aa6 | 1523 | struct pgpath *pgpath; |
35991652 MP |
1524 | struct block_device *bdev; |
1525 | fmode_t mode; | |
9af4aa30 | 1526 | unsigned long flags; |
35991652 MP |
1527 | int r; |
1528 | ||
35991652 MP |
1529 | bdev = NULL; |
1530 | mode = 0; | |
1531 | r = 0; | |
9af4aa30 MB |
1532 | |
1533 | spin_lock_irqsave(&m->lock, flags); | |
1534 | ||
1535 | if (!m->current_pgpath) | |
02ab823f | 1536 | __choose_pgpath(m, 0); |
9af4aa30 | 1537 | |
7ba10aa6 MS |
1538 | pgpath = m->current_pgpath; |
1539 | ||
1540 | if (pgpath) { | |
1541 | bdev = pgpath->path.dev->bdev; | |
1542 | mode = pgpath->path.dev->mode; | |
e90dae1f | 1543 | } |
9af4aa30 | 1544 | |
7ba10aa6 | 1545 | if ((pgpath && m->queue_io) || (!pgpath && m->queue_if_no_path)) |
6c182cd8 | 1546 | r = -ENOTCONN; |
9af4aa30 MB |
1547 | else if (!bdev) |
1548 | r = -EIO; | |
1549 | ||
1550 | spin_unlock_irqrestore(&m->lock, flags); | |
1551 | ||
ec8013be PB |
1552 | /* |
1553 | * Only pass ioctls through if the device sizes match exactly. | |
1554 | */ | |
a1989b33 HR |
1555 | if (!bdev || ti->len != i_size_read(bdev->bd_inode) >> SECTOR_SHIFT) { |
1556 | int err = scsi_verify_blk_ioctl(NULL, cmd); | |
1557 | if (err) | |
1558 | r = err; | |
1559 | } | |
ec8013be | 1560 | |
3e9f1be1 HR |
1561 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
1562 | spin_lock_irqsave(&m->lock, flags); | |
1563 | if (!m->current_pg) { | |
1564 | /* Path status changed, redo selection */ | |
1565 | __choose_pgpath(m, 0); | |
1566 | } | |
1567 | if (m->pg_init_required) | |
1568 | __pg_init_all_paths(m); | |
3e9f1be1 | 1569 | dm_table_run_md_queue_async(m->ti->table); |
4cdd2ad7 | 1570 | spin_unlock_irqrestore(&m->lock, flags); |
3e9f1be1 | 1571 | } |
35991652 | 1572 | |
633a08b8 | 1573 | return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
9af4aa30 MB |
1574 | } |
1575 | ||
af4874e0 MS |
1576 | static int multipath_iterate_devices(struct dm_target *ti, |
1577 | iterate_devices_callout_fn fn, void *data) | |
1578 | { | |
1579 | struct multipath *m = ti->private; | |
1580 | struct priority_group *pg; | |
1581 | struct pgpath *p; | |
1582 | int ret = 0; | |
1583 | ||
1584 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1585 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1586 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1587 | if (ret) |
1588 | goto out; | |
1589 | } | |
1590 | } | |
1591 | ||
1592 | out: | |
1593 | return ret; | |
1594 | } | |
1595 | ||
f40c67f0 KU |
1596 | static int __pgpath_busy(struct pgpath *pgpath) |
1597 | { | |
1598 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1599 | ||
1600 | return dm_underlying_device_busy(q); | |
1601 | } | |
1602 | ||
1603 | /* | |
1604 | * We return "busy", only when we can map I/Os but underlying devices | |
1605 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1606 | * the underlying queue). | |
1607 | * In other words, if we want to kill I/Os or queue them inside us | |
1608 | * due to map unavailability, we don't return "busy". Otherwise, | |
1609 | * dm core won't give us the I/Os and we can't do what we want. | |
1610 | */ | |
1611 | static int multipath_busy(struct dm_target *ti) | |
1612 | { | |
1613 | int busy = 0, has_active = 0; | |
1614 | struct multipath *m = ti->private; | |
1615 | struct priority_group *pg; | |
1616 | struct pgpath *pgpath; | |
1617 | unsigned long flags; | |
1618 | ||
1619 | spin_lock_irqsave(&m->lock, flags); | |
1620 | ||
b63349a7 | 1621 | /* pg_init in progress, requeue until done */ |
e8099177 | 1622 | if (!pg_ready(m)) { |
b63349a7 HR |
1623 | busy = 1; |
1624 | goto out; | |
1625 | } | |
f40c67f0 KU |
1626 | /* Guess which priority_group will be used at next mapping time */ |
1627 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1628 | pg = m->next_pg; | |
1629 | else if (likely(m->current_pg)) | |
1630 | pg = m->current_pg; | |
1631 | else | |
1632 | /* | |
1633 | * We don't know which pg will be used at next mapping time. | |
1634 | * We don't call __choose_pgpath() here to avoid to trigger | |
1635 | * pg_init just by busy checking. | |
1636 | * So we don't know whether underlying devices we will be using | |
1637 | * at next mapping time are busy or not. Just try mapping. | |
1638 | */ | |
1639 | goto out; | |
1640 | ||
1641 | /* | |
1642 | * If there is one non-busy active path at least, the path selector | |
1643 | * will be able to select it. So we consider such a pg as not busy. | |
1644 | */ | |
1645 | busy = 1; | |
1646 | list_for_each_entry(pgpath, &pg->pgpaths, list) | |
1647 | if (pgpath->is_active) { | |
1648 | has_active = 1; | |
1649 | ||
1650 | if (!__pgpath_busy(pgpath)) { | |
1651 | busy = 0; | |
1652 | break; | |
1653 | } | |
1654 | } | |
1655 | ||
1656 | if (!has_active) | |
1657 | /* | |
1658 | * No active path in this pg, so this pg won't be used and | |
1659 | * the current_pg will be changed at next mapping time. | |
1660 | * We need to try mapping to determine it. | |
1661 | */ | |
1662 | busy = 0; | |
1663 | ||
1664 | out: | |
1665 | spin_unlock_irqrestore(&m->lock, flags); | |
1666 | ||
1667 | return busy; | |
1668 | } | |
1669 | ||
1da177e4 LT |
1670 | /*----------------------------------------------------------------- |
1671 | * Module setup | |
1672 | *---------------------------------------------------------------*/ | |
1673 | static struct target_type multipath_target = { | |
1674 | .name = "multipath", | |
e8099177 | 1675 | .version = {1, 7, 0}, |
1da177e4 LT |
1676 | .module = THIS_MODULE, |
1677 | .ctr = multipath_ctr, | |
1678 | .dtr = multipath_dtr, | |
f40c67f0 KU |
1679 | .map_rq = multipath_map, |
1680 | .rq_end_io = multipath_end_io, | |
1da177e4 | 1681 | .presuspend = multipath_presuspend, |
6df400ab | 1682 | .postsuspend = multipath_postsuspend, |
1da177e4 LT |
1683 | .resume = multipath_resume, |
1684 | .status = multipath_status, | |
1685 | .message = multipath_message, | |
9af4aa30 | 1686 | .ioctl = multipath_ioctl, |
af4874e0 | 1687 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1688 | .busy = multipath_busy, |
1da177e4 LT |
1689 | }; |
1690 | ||
1691 | static int __init dm_multipath_init(void) | |
1692 | { | |
1693 | int r; | |
1694 | ||
1695 | /* allocate a slab for the dm_ios */ | |
028867ac | 1696 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1697 | if (!_mpio_cache) |
1698 | return -ENOMEM; | |
1699 | ||
1700 | r = dm_register_target(&multipath_target); | |
1701 | if (r < 0) { | |
0cd33124 | 1702 | DMERR("register failed %d", r); |
1da177e4 LT |
1703 | kmem_cache_destroy(_mpio_cache); |
1704 | return -EINVAL; | |
1705 | } | |
1706 | ||
4d4d66ab | 1707 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1708 | if (!kmultipathd) { |
0cd33124 | 1709 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1710 | dm_unregister_target(&multipath_target); |
1711 | kmem_cache_destroy(_mpio_cache); | |
1712 | return -ENOMEM; | |
1713 | } | |
1714 | ||
bab7cfc7 CS |
1715 | /* |
1716 | * A separate workqueue is used to handle the device handlers | |
1717 | * to avoid overloading existing workqueue. Overloading the | |
1718 | * old workqueue would also create a bottleneck in the | |
1719 | * path of the storage hardware device activation. | |
1720 | */ | |
4d4d66ab TH |
1721 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1722 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1723 | if (!kmpath_handlerd) { |
1724 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1725 | destroy_workqueue(kmultipathd); | |
1726 | dm_unregister_target(&multipath_target); | |
1727 | kmem_cache_destroy(_mpio_cache); | |
1728 | return -ENOMEM; | |
1729 | } | |
1730 | ||
72d94861 | 1731 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1732 | multipath_target.version[0], multipath_target.version[1], |
1733 | multipath_target.version[2]); | |
1734 | ||
1735 | return r; | |
1736 | } | |
1737 | ||
1738 | static void __exit dm_multipath_exit(void) | |
1739 | { | |
bab7cfc7 | 1740 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1741 | destroy_workqueue(kmultipathd); |
1742 | ||
10d3bd09 | 1743 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1744 | kmem_cache_destroy(_mpio_cache); |
1745 | } | |
1746 | ||
1da177e4 LT |
1747 | module_init(dm_multipath_init); |
1748 | module_exit(dm_multipath_exit); | |
1749 | ||
1750 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1751 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1752 | MODULE_LICENSE("GPL"); |