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 LT |
95 | |
96 | struct work_struct process_queued_ios; | |
1da177e4 LT |
97 | |
98 | struct work_struct trigger_event; | |
99 | ||
100 | /* | |
028867ac | 101 | * We must use a mempool of dm_mpath_io structs so that we |
1da177e4 LT |
102 | * can resubmit bios on error. |
103 | */ | |
104 | mempool_t *mpio_pool; | |
6380f26f MA |
105 | |
106 | struct mutex work_mutex; | |
1da177e4 LT |
107 | }; |
108 | ||
109 | /* | |
110 | * Context information attached to each bio we process. | |
111 | */ | |
028867ac | 112 | struct dm_mpath_io { |
1da177e4 | 113 | struct pgpath *pgpath; |
02ab823f | 114 | size_t nr_bytes; |
1da177e4 LT |
115 | }; |
116 | ||
117 | typedef int (*action_fn) (struct pgpath *pgpath); | |
118 | ||
e18b890b | 119 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 120 | |
bab7cfc7 | 121 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 DH |
122 | static void process_queued_ios(struct work_struct *work); |
123 | static void trigger_event(struct work_struct *work); | |
bab7cfc7 | 124 | static void activate_path(struct work_struct *work); |
e8099177 | 125 | static int __pgpath_busy(struct pgpath *pgpath); |
1da177e4 LT |
126 | |
127 | ||
128 | /*----------------------------------------------- | |
129 | * Allocation routines | |
130 | *-----------------------------------------------*/ | |
131 | ||
132 | static struct pgpath *alloc_pgpath(void) | |
133 | { | |
e69fae56 | 134 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 135 | |
224cb3e9 | 136 | if (pgpath) { |
6680073d | 137 | pgpath->is_active = 1; |
4e2d19e4 | 138 | INIT_DELAYED_WORK(&pgpath->activate_path, activate_path); |
224cb3e9 | 139 | } |
1da177e4 LT |
140 | |
141 | return pgpath; | |
142 | } | |
143 | ||
028867ac | 144 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
145 | { |
146 | kfree(pgpath); | |
147 | } | |
148 | ||
149 | static struct priority_group *alloc_priority_group(void) | |
150 | { | |
151 | struct priority_group *pg; | |
152 | ||
e69fae56 | 153 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 154 | |
e69fae56 MM |
155 | if (pg) |
156 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
157 | |
158 | return pg; | |
159 | } | |
160 | ||
161 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
162 | { | |
163 | struct pgpath *pgpath, *tmp; | |
ae11b1b3 | 164 | struct multipath *m = ti->private; |
1da177e4 LT |
165 | |
166 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
167 | list_del(&pgpath->list); | |
ae11b1b3 HR |
168 | if (m->hw_handler_name) |
169 | scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev)); | |
1da177e4 LT |
170 | dm_put_device(ti, pgpath->path.dev); |
171 | free_pgpath(pgpath); | |
172 | } | |
173 | } | |
174 | ||
175 | static void free_priority_group(struct priority_group *pg, | |
176 | struct dm_target *ti) | |
177 | { | |
178 | struct path_selector *ps = &pg->ps; | |
179 | ||
180 | if (ps->type) { | |
181 | ps->type->destroy(ps); | |
182 | dm_put_path_selector(ps->type); | |
183 | } | |
184 | ||
185 | free_pgpaths(&pg->pgpaths, ti); | |
186 | kfree(pg); | |
187 | } | |
188 | ||
28f16c20 | 189 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
190 | { |
191 | struct multipath *m; | |
f4790826 | 192 | unsigned min_ios = dm_get_reserved_rq_based_ios(); |
1da177e4 | 193 | |
e69fae56 | 194 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 195 | if (m) { |
1da177e4 LT |
196 | INIT_LIST_HEAD(&m->priority_groups); |
197 | spin_lock_init(&m->lock); | |
198 | m->queue_io = 1; | |
4e2d19e4 | 199 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; |
c4028958 DH |
200 | INIT_WORK(&m->process_queued_ios, process_queued_ios); |
201 | INIT_WORK(&m->trigger_event, trigger_event); | |
2bded7bd | 202 | init_waitqueue_head(&m->pg_init_wait); |
6380f26f | 203 | mutex_init(&m->work_mutex); |
f4790826 | 204 | m->mpio_pool = mempool_create_slab_pool(min_ios, _mpio_cache); |
1da177e4 LT |
205 | if (!m->mpio_pool) { |
206 | kfree(m); | |
207 | return NULL; | |
208 | } | |
28f16c20 MM |
209 | m->ti = ti; |
210 | ti->private = m; | |
1da177e4 LT |
211 | } |
212 | ||
213 | return m; | |
214 | } | |
215 | ||
216 | static void free_multipath(struct multipath *m) | |
217 | { | |
218 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
219 | |
220 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
221 | list_del(&pg->list); | |
222 | free_priority_group(pg, m->ti); | |
223 | } | |
224 | ||
cfae5c9b | 225 | kfree(m->hw_handler_name); |
2bfd2e13 | 226 | kfree(m->hw_handler_params); |
1da177e4 LT |
227 | mempool_destroy(m->mpio_pool); |
228 | kfree(m); | |
229 | } | |
230 | ||
466891f9 JN |
231 | static int set_mapinfo(struct multipath *m, union map_info *info) |
232 | { | |
233 | struct dm_mpath_io *mpio; | |
234 | ||
235 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); | |
236 | if (!mpio) | |
237 | return -ENOMEM; | |
238 | ||
239 | memset(mpio, 0, sizeof(*mpio)); | |
240 | info->ptr = mpio; | |
241 | ||
242 | return 0; | |
243 | } | |
244 | ||
245 | static void clear_mapinfo(struct multipath *m, union map_info *info) | |
246 | { | |
247 | struct dm_mpath_io *mpio = info->ptr; | |
248 | ||
249 | info->ptr = NULL; | |
250 | mempool_free(mpio, m->mpio_pool); | |
251 | } | |
1da177e4 LT |
252 | |
253 | /*----------------------------------------------- | |
254 | * Path selection | |
255 | *-----------------------------------------------*/ | |
256 | ||
fb612642 KU |
257 | static void __pg_init_all_paths(struct multipath *m) |
258 | { | |
259 | struct pgpath *pgpath; | |
4e2d19e4 | 260 | unsigned long pg_init_delay = 0; |
fb612642 | 261 | |
17f4ff45 HR |
262 | if (m->pg_init_in_progress || m->pg_init_disabled) |
263 | return; | |
264 | ||
fb612642 KU |
265 | m->pg_init_count++; |
266 | m->pg_init_required = 0; | |
4e2d19e4 CS |
267 | if (m->pg_init_delay_retry) |
268 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? | |
269 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
270 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
271 | /* Skip failed paths */ | |
272 | if (!pgpath->is_active) | |
273 | continue; | |
4e2d19e4 CS |
274 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
275 | pg_init_delay)) | |
fb612642 KU |
276 | m->pg_init_in_progress++; |
277 | } | |
278 | } | |
279 | ||
1da177e4 LT |
280 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) |
281 | { | |
1da177e4 LT |
282 | m->current_pg = pgpath->pg; |
283 | ||
284 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 285 | if (m->hw_handler_name) { |
1da177e4 LT |
286 | m->pg_init_required = 1; |
287 | m->queue_io = 1; | |
288 | } else { | |
289 | m->pg_init_required = 0; | |
290 | m->queue_io = 0; | |
291 | } | |
c9e45581 DW |
292 | |
293 | m->pg_init_count = 0; | |
1da177e4 LT |
294 | } |
295 | ||
02ab823f KU |
296 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
297 | size_t nr_bytes) | |
1da177e4 | 298 | { |
c922d5f7 | 299 | struct dm_path *path; |
1da177e4 | 300 | |
02ab823f | 301 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes); |
1da177e4 LT |
302 | if (!path) |
303 | return -ENXIO; | |
304 | ||
305 | m->current_pgpath = path_to_pgpath(path); | |
306 | ||
307 | if (m->current_pg != pg) | |
308 | __switch_pg(m, m->current_pgpath); | |
309 | ||
310 | return 0; | |
311 | } | |
312 | ||
02ab823f | 313 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
314 | { |
315 | struct priority_group *pg; | |
316 | unsigned bypassed = 1; | |
317 | ||
318 | if (!m->nr_valid_paths) | |
319 | goto failed; | |
320 | ||
321 | /* Were we instructed to switch PG? */ | |
322 | if (m->next_pg) { | |
323 | pg = m->next_pg; | |
324 | m->next_pg = NULL; | |
02ab823f | 325 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
326 | return; |
327 | } | |
328 | ||
329 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 330 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
331 | return; |
332 | ||
333 | /* | |
334 | * Loop through priority groups until we find a valid path. | |
335 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
336 | * Second time we only try the ones we skipped, but set |
337 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
338 | */ |
339 | do { | |
340 | list_for_each_entry(pg, &m->priority_groups, list) { | |
341 | if (pg->bypassed == bypassed) | |
342 | continue; | |
f220fd4e MC |
343 | if (!__choose_path_in_pg(m, pg, nr_bytes)) { |
344 | if (!bypassed) | |
345 | m->pg_init_delay_retry = 1; | |
1da177e4 | 346 | return; |
f220fd4e | 347 | } |
1da177e4 LT |
348 | } |
349 | } while (bypassed--); | |
350 | ||
351 | failed: | |
352 | m->current_pgpath = NULL; | |
353 | m->current_pg = NULL; | |
354 | } | |
355 | ||
45e15720 KU |
356 | /* |
357 | * Check whether bios must be queued in the device-mapper core rather | |
358 | * than here in the target. | |
359 | * | |
360 | * m->lock must be held on entry. | |
361 | * | |
362 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
363 | * same value then we are not between multipath_presuspend() | |
364 | * and multipath_resume() calls and we have no need to check | |
365 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
366 | */ | |
367 | static int __must_push_back(struct multipath *m) | |
368 | { | |
e8099177 HR |
369 | return (m->queue_if_no_path || |
370 | (m->queue_if_no_path != m->saved_queue_if_no_path && | |
371 | dm_noflush_suspending(m->ti))); | |
45e15720 KU |
372 | } |
373 | ||
e8099177 HR |
374 | #define pg_ready(m) (!(m)->queue_io && !(m)->pg_init_required) |
375 | ||
f40c67f0 | 376 | static int map_io(struct multipath *m, struct request *clone, |
e8099177 | 377 | union map_info *map_context) |
1da177e4 | 378 | { |
d2a7ad29 | 379 | int r = DM_MAPIO_REMAPPED; |
f40c67f0 | 380 | size_t nr_bytes = blk_rq_bytes(clone); |
1da177e4 LT |
381 | unsigned long flags; |
382 | struct pgpath *pgpath; | |
f40c67f0 | 383 | struct block_device *bdev; |
466891f9 | 384 | struct dm_mpath_io *mpio = map_context->ptr; |
1da177e4 LT |
385 | |
386 | spin_lock_irqsave(&m->lock, flags); | |
387 | ||
388 | /* Do we need to select a new pgpath? */ | |
389 | if (!m->current_pgpath || | |
390 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
02ab823f | 391 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
392 | |
393 | pgpath = m->current_pgpath; | |
394 | ||
e8099177 HR |
395 | if (pgpath) { |
396 | if (pg_ready(m)) { | |
397 | bdev = pgpath->path.dev->bdev; | |
398 | clone->q = bdev_get_queue(bdev); | |
399 | clone->rq_disk = bdev->bd_disk; | |
400 | mpio->pgpath = pgpath; | |
401 | mpio->nr_bytes = nr_bytes; | |
402 | if (pgpath->pg->ps.type->start_io) | |
403 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
404 | &pgpath->path, | |
405 | nr_bytes); | |
406 | } else { | |
407 | __pg_init_all_paths(m); | |
408 | r = DM_MAPIO_REQUEUE; | |
409 | } | |
410 | } else { | |
411 | /* No path */ | |
412 | if (__must_push_back(m)) | |
413 | r = DM_MAPIO_REQUEUE; | |
414 | else | |
415 | r = -EIO; /* Failed */ | |
416 | } | |
1da177e4 LT |
417 | |
418 | spin_unlock_irqrestore(&m->lock, flags); | |
419 | ||
420 | return r; | |
421 | } | |
422 | ||
423 | /* | |
424 | * If we run out of usable paths, should we queue I/O or error it? | |
425 | */ | |
485ef69e AK |
426 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
427 | unsigned save_old_value) | |
1da177e4 LT |
428 | { |
429 | unsigned long flags; | |
430 | ||
431 | spin_lock_irqsave(&m->lock, flags); | |
432 | ||
485ef69e AK |
433 | if (save_old_value) |
434 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
435 | else | |
436 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 437 | m->queue_if_no_path = queue_if_no_path; |
e8099177 | 438 | if (!m->queue_if_no_path) |
c557308e | 439 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 LT |
440 | |
441 | spin_unlock_irqrestore(&m->lock, flags); | |
442 | ||
443 | return 0; | |
444 | } | |
445 | ||
c4028958 | 446 | static void process_queued_ios(struct work_struct *work) |
1da177e4 | 447 | { |
c4028958 DH |
448 | struct multipath *m = |
449 | container_of(work, struct multipath, process_queued_ios); | |
fb612642 | 450 | struct pgpath *pgpath = NULL; |
e54f77dd | 451 | unsigned must_queue = 1; |
1da177e4 LT |
452 | unsigned long flags; |
453 | ||
454 | spin_lock_irqsave(&m->lock, flags); | |
455 | ||
456 | if (!m->current_pgpath) | |
02ab823f | 457 | __choose_pgpath(m, 0); |
1da177e4 LT |
458 | |
459 | pgpath = m->current_pgpath; | |
460 | ||
c3cd4f6b AK |
461 | if ((pgpath && !m->queue_io) || |
462 | (!pgpath && !m->queue_if_no_path)) | |
463 | must_queue = 0; | |
1da177e4 | 464 | |
17f4ff45 | 465 | if (pgpath && m->pg_init_required) |
fb612642 KU |
466 | __pg_init_all_paths(m); |
467 | ||
1da177e4 | 468 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 469 | if (!must_queue) |
e8099177 | 470 | dm_table_run_md_queue_async(m->ti->table); |
1da177e4 LT |
471 | } |
472 | ||
473 | /* | |
474 | * An event is triggered whenever a path is taken out of use. | |
475 | * Includes path failure and PG bypass. | |
476 | */ | |
c4028958 | 477 | static void trigger_event(struct work_struct *work) |
1da177e4 | 478 | { |
c4028958 DH |
479 | struct multipath *m = |
480 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
481 | |
482 | dm_table_event(m->ti->table); | |
483 | } | |
484 | ||
485 | /*----------------------------------------------------------------- | |
486 | * Constructor/argument parsing: | |
487 | * <#multipath feature args> [<arg>]* | |
488 | * <#hw_handler args> [hw_handler [<arg>]*] | |
489 | * <#priority groups> | |
490 | * <initial priority group> | |
491 | * [<selector> <#selector args> [<arg>]* | |
492 | * <#paths> <#per-path selector args> | |
493 | * [<path> [<arg>]* ]+ ]+ | |
494 | *---------------------------------------------------------------*/ | |
498f0103 | 495 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
496 | struct dm_target *ti) |
497 | { | |
498 | int r; | |
499 | struct path_selector_type *pst; | |
500 | unsigned ps_argc; | |
501 | ||
498f0103 | 502 | static struct dm_arg _args[] = { |
72d94861 | 503 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
504 | }; |
505 | ||
498f0103 | 506 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 507 | if (!pst) { |
72d94861 | 508 | ti->error = "unknown path selector type"; |
1da177e4 LT |
509 | return -EINVAL; |
510 | } | |
511 | ||
498f0103 | 512 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
513 | if (r) { |
514 | dm_put_path_selector(pst); | |
1da177e4 | 515 | return -EINVAL; |
371b2e34 | 516 | } |
1da177e4 LT |
517 | |
518 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
519 | if (r) { | |
520 | dm_put_path_selector(pst); | |
72d94861 | 521 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
522 | return r; |
523 | } | |
524 | ||
525 | pg->ps.type = pst; | |
498f0103 | 526 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
527 | |
528 | return 0; | |
529 | } | |
530 | ||
498f0103 | 531 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
532 | struct dm_target *ti) |
533 | { | |
534 | int r; | |
535 | struct pgpath *p; | |
ae11b1b3 | 536 | struct multipath *m = ti->private; |
a58a935d MS |
537 | struct request_queue *q = NULL; |
538 | const char *attached_handler_name; | |
1da177e4 LT |
539 | |
540 | /* we need at least a path arg */ | |
541 | if (as->argc < 1) { | |
72d94861 | 542 | ti->error = "no device given"; |
01460f35 | 543 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
544 | } |
545 | ||
546 | p = alloc_pgpath(); | |
547 | if (!p) | |
01460f35 | 548 | return ERR_PTR(-ENOMEM); |
1da177e4 | 549 | |
498f0103 | 550 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 551 | &p->path.dev); |
1da177e4 | 552 | if (r) { |
72d94861 | 553 | ti->error = "error getting device"; |
1da177e4 LT |
554 | goto bad; |
555 | } | |
556 | ||
a58a935d MS |
557 | if (m->retain_attached_hw_handler || m->hw_handler_name) |
558 | q = bdev_get_queue(p->path.dev->bdev); | |
559 | ||
560 | if (m->retain_attached_hw_handler) { | |
561 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); | |
562 | if (attached_handler_name) { | |
563 | /* | |
564 | * Reset hw_handler_name to match the attached handler | |
565 | * and clear any hw_handler_params associated with the | |
566 | * ignored handler. | |
567 | * | |
568 | * NB. This modifies the table line to show the actual | |
569 | * handler instead of the original table passed in. | |
570 | */ | |
571 | kfree(m->hw_handler_name); | |
572 | m->hw_handler_name = attached_handler_name; | |
573 | ||
574 | kfree(m->hw_handler_params); | |
575 | m->hw_handler_params = NULL; | |
576 | } | |
577 | } | |
a0cf7ea9 | 578 | |
a58a935d MS |
579 | if (m->hw_handler_name) { |
580 | /* | |
581 | * Increments scsi_dh reference, even when using an | |
582 | * already-attached handler. | |
583 | */ | |
a0cf7ea9 HR |
584 | r = scsi_dh_attach(q, m->hw_handler_name); |
585 | if (r == -EBUSY) { | |
586 | /* | |
a58a935d | 587 | * Already attached to different hw_handler: |
a0cf7ea9 HR |
588 | * try to reattach with correct one. |
589 | */ | |
590 | scsi_dh_detach(q); | |
591 | r = scsi_dh_attach(q, m->hw_handler_name); | |
592 | } | |
593 | ||
ae11b1b3 | 594 | if (r < 0) { |
a0cf7ea9 | 595 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
596 | dm_put_device(ti, p->path.dev); |
597 | goto bad; | |
598 | } | |
2bfd2e13 CS |
599 | |
600 | if (m->hw_handler_params) { | |
601 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
602 | if (r < 0) { | |
603 | ti->error = "unable to set hardware " | |
604 | "handler parameters"; | |
605 | scsi_dh_detach(q); | |
606 | dm_put_device(ti, p->path.dev); | |
607 | goto bad; | |
608 | } | |
609 | } | |
ae11b1b3 HR |
610 | } |
611 | ||
1da177e4 LT |
612 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
613 | if (r) { | |
614 | dm_put_device(ti, p->path.dev); | |
615 | goto bad; | |
616 | } | |
617 | ||
618 | return p; | |
619 | ||
620 | bad: | |
621 | free_pgpath(p); | |
01460f35 | 622 | return ERR_PTR(r); |
1da177e4 LT |
623 | } |
624 | ||
498f0103 | 625 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 626 | struct multipath *m) |
1da177e4 | 627 | { |
498f0103 | 628 | static struct dm_arg _args[] = { |
72d94861 AK |
629 | {1, 1024, "invalid number of paths"}, |
630 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
631 | }; |
632 | ||
633 | int r; | |
498f0103 | 634 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 635 | struct priority_group *pg; |
28f16c20 | 636 | struct dm_target *ti = m->ti; |
1da177e4 LT |
637 | |
638 | if (as->argc < 2) { | |
639 | as->argc = 0; | |
01460f35 BM |
640 | ti->error = "not enough priority group arguments"; |
641 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
642 | } |
643 | ||
644 | pg = alloc_priority_group(); | |
645 | if (!pg) { | |
72d94861 | 646 | ti->error = "couldn't allocate priority group"; |
01460f35 | 647 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
648 | } |
649 | pg->m = m; | |
650 | ||
651 | r = parse_path_selector(as, pg, ti); | |
652 | if (r) | |
653 | goto bad; | |
654 | ||
655 | /* | |
656 | * read the paths | |
657 | */ | |
498f0103 | 658 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
659 | if (r) |
660 | goto bad; | |
661 | ||
498f0103 | 662 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
663 | if (r) |
664 | goto bad; | |
665 | ||
498f0103 | 666 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
667 | for (i = 0; i < pg->nr_pgpaths; i++) { |
668 | struct pgpath *pgpath; | |
498f0103 | 669 | struct dm_arg_set path_args; |
1da177e4 | 670 | |
498f0103 | 671 | if (as->argc < nr_args) { |
148acff6 | 672 | ti->error = "not enough path parameters"; |
6bbf79a1 | 673 | r = -EINVAL; |
1da177e4 | 674 | goto bad; |
148acff6 | 675 | } |
1da177e4 | 676 | |
498f0103 | 677 | path_args.argc = nr_args; |
1da177e4 LT |
678 | path_args.argv = as->argv; |
679 | ||
680 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
681 | if (IS_ERR(pgpath)) { |
682 | r = PTR_ERR(pgpath); | |
1da177e4 | 683 | goto bad; |
01460f35 | 684 | } |
1da177e4 LT |
685 | |
686 | pgpath->pg = pg; | |
687 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 688 | dm_consume_args(as, nr_args); |
1da177e4 LT |
689 | } |
690 | ||
691 | return pg; | |
692 | ||
693 | bad: | |
694 | free_priority_group(pg, ti); | |
01460f35 | 695 | return ERR_PTR(r); |
1da177e4 LT |
696 | } |
697 | ||
498f0103 | 698 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 699 | { |
1da177e4 | 700 | unsigned hw_argc; |
2bfd2e13 | 701 | int ret; |
28f16c20 | 702 | struct dm_target *ti = m->ti; |
1da177e4 | 703 | |
498f0103 | 704 | static struct dm_arg _args[] = { |
72d94861 | 705 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
706 | }; |
707 | ||
498f0103 | 708 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
709 | return -EINVAL; |
710 | ||
711 | if (!hw_argc) | |
712 | return 0; | |
713 | ||
498f0103 | 714 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
510193a2 MS |
715 | if (!try_then_request_module(scsi_dh_handler_exist(m->hw_handler_name), |
716 | "scsi_dh_%s", m->hw_handler_name)) { | |
72d94861 | 717 | ti->error = "unknown hardware handler type"; |
2bfd2e13 CS |
718 | ret = -EINVAL; |
719 | goto fail; | |
1da177e4 | 720 | } |
14e98c5c | 721 | |
2bfd2e13 CS |
722 | if (hw_argc > 1) { |
723 | char *p; | |
724 | int i, j, len = 4; | |
725 | ||
726 | for (i = 0; i <= hw_argc - 2; i++) | |
727 | len += strlen(as->argv[i]) + 1; | |
728 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
729 | if (!p) { | |
730 | ti->error = "memory allocation failed"; | |
731 | ret = -ENOMEM; | |
732 | goto fail; | |
733 | } | |
734 | j = sprintf(p, "%d", hw_argc - 1); | |
735 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
736 | j = sprintf(p, "%s", as->argv[i]); | |
737 | } | |
498f0103 | 738 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
739 | |
740 | return 0; | |
2bfd2e13 CS |
741 | fail: |
742 | kfree(m->hw_handler_name); | |
743 | m->hw_handler_name = NULL; | |
744 | return ret; | |
1da177e4 LT |
745 | } |
746 | ||
498f0103 | 747 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
748 | { |
749 | int r; | |
750 | unsigned argc; | |
28f16c20 | 751 | struct dm_target *ti = m->ti; |
498f0103 | 752 | const char *arg_name; |
1da177e4 | 753 | |
498f0103 | 754 | static struct dm_arg _args[] = { |
a58a935d | 755 | {0, 6, "invalid number of feature args"}, |
c9e45581 | 756 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 757 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
758 | }; |
759 | ||
498f0103 | 760 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
761 | if (r) |
762 | return -EINVAL; | |
763 | ||
764 | if (!argc) | |
765 | return 0; | |
766 | ||
c9e45581 | 767 | do { |
498f0103 | 768 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
769 | argc--; |
770 | ||
498f0103 | 771 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
c9e45581 DW |
772 | r = queue_if_no_path(m, 1, 0); |
773 | continue; | |
774 | } | |
775 | ||
a58a935d MS |
776 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
777 | m->retain_attached_hw_handler = 1; | |
778 | continue; | |
779 | } | |
780 | ||
498f0103 | 781 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 782 | (argc >= 1)) { |
498f0103 | 783 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
784 | argc--; |
785 | continue; | |
786 | } | |
787 | ||
498f0103 | 788 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 789 | (argc >= 1)) { |
498f0103 | 790 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
791 | argc--; |
792 | continue; | |
793 | } | |
794 | ||
1da177e4 | 795 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
796 | r = -EINVAL; |
797 | } while (argc && !r); | |
798 | ||
799 | return r; | |
1da177e4 LT |
800 | } |
801 | ||
802 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
803 | char **argv) | |
804 | { | |
498f0103 MS |
805 | /* target arguments */ |
806 | static struct dm_arg _args[] = { | |
a490a07a MS |
807 | {0, 1024, "invalid number of priority groups"}, |
808 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
809 | }; |
810 | ||
811 | int r; | |
812 | struct multipath *m; | |
498f0103 | 813 | struct dm_arg_set as; |
1da177e4 LT |
814 | unsigned pg_count = 0; |
815 | unsigned next_pg_num; | |
816 | ||
817 | as.argc = argc; | |
818 | as.argv = argv; | |
819 | ||
28f16c20 | 820 | m = alloc_multipath(ti); |
1da177e4 | 821 | if (!m) { |
72d94861 | 822 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
823 | return -EINVAL; |
824 | } | |
825 | ||
28f16c20 | 826 | r = parse_features(&as, m); |
1da177e4 LT |
827 | if (r) |
828 | goto bad; | |
829 | ||
28f16c20 | 830 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
831 | if (r) |
832 | goto bad; | |
833 | ||
498f0103 | 834 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
835 | if (r) |
836 | goto bad; | |
837 | ||
498f0103 | 838 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
839 | if (r) |
840 | goto bad; | |
841 | ||
a490a07a MS |
842 | if ((!m->nr_priority_groups && next_pg_num) || |
843 | (m->nr_priority_groups && !next_pg_num)) { | |
844 | ti->error = "invalid initial priority group"; | |
845 | r = -EINVAL; | |
846 | goto bad; | |
847 | } | |
848 | ||
1da177e4 LT |
849 | /* parse the priority groups */ |
850 | while (as.argc) { | |
851 | struct priority_group *pg; | |
852 | ||
28f16c20 | 853 | pg = parse_priority_group(&as, m); |
01460f35 BM |
854 | if (IS_ERR(pg)) { |
855 | r = PTR_ERR(pg); | |
1da177e4 LT |
856 | goto bad; |
857 | } | |
858 | ||
859 | m->nr_valid_paths += pg->nr_pgpaths; | |
860 | list_add_tail(&pg->list, &m->priority_groups); | |
861 | pg_count++; | |
862 | pg->pg_num = pg_count; | |
863 | if (!--next_pg_num) | |
864 | m->next_pg = pg; | |
865 | } | |
866 | ||
867 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 868 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
869 | r = -EINVAL; |
870 | goto bad; | |
871 | } | |
872 | ||
55a62eef AK |
873 | ti->num_flush_bios = 1; |
874 | ti->num_discard_bios = 1; | |
042bcef8 | 875 | ti->num_write_same_bios = 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 | { |
954a73d5 SKM |
910 | unsigned long flags; |
911 | ||
912 | spin_lock_irqsave(&m->lock, flags); | |
913 | m->pg_init_disabled = 1; | |
914 | spin_unlock_irqrestore(&m->lock, flags); | |
915 | ||
bab7cfc7 | 916 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 917 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 918 | flush_workqueue(kmultipathd); |
43829731 | 919 | flush_work(&m->trigger_event); |
954a73d5 SKM |
920 | |
921 | spin_lock_irqsave(&m->lock, flags); | |
922 | m->pg_init_disabled = 0; | |
923 | spin_unlock_irqrestore(&m->lock, flags); | |
6df400ab KU |
924 | } |
925 | ||
926 | static void multipath_dtr(struct dm_target *ti) | |
927 | { | |
928 | struct multipath *m = ti->private; | |
929 | ||
2bded7bd | 930 | flush_multipath_work(m); |
1da177e4 LT |
931 | free_multipath(m); |
932 | } | |
933 | ||
934 | /* | |
f40c67f0 | 935 | * Map cloned requests |
1da177e4 | 936 | */ |
f40c67f0 | 937 | static int multipath_map(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
938 | union map_info *map_context) |
939 | { | |
940 | int r; | |
1da177e4 LT |
941 | struct multipath *m = (struct multipath *) ti->private; |
942 | ||
466891f9 | 943 | if (set_mapinfo(m, map_context) < 0) |
f40c67f0 KU |
944 | /* ENOMEM, requeue */ |
945 | return DM_MAPIO_REQUEUE; | |
1da177e4 | 946 | |
f40c67f0 | 947 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; |
e8099177 | 948 | r = map_io(m, clone, map_context); |
45e15720 | 949 | if (r < 0 || r == DM_MAPIO_REQUEUE) |
466891f9 | 950 | clear_mapinfo(m, map_context); |
1da177e4 LT |
951 | |
952 | return r; | |
953 | } | |
954 | ||
955 | /* | |
956 | * Take a path out of use. | |
957 | */ | |
958 | static int fail_path(struct pgpath *pgpath) | |
959 | { | |
960 | unsigned long flags; | |
961 | struct multipath *m = pgpath->pg->m; | |
962 | ||
963 | spin_lock_irqsave(&m->lock, flags); | |
964 | ||
6680073d | 965 | if (!pgpath->is_active) |
1da177e4 LT |
966 | goto out; |
967 | ||
72d94861 | 968 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
969 | |
970 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
6680073d | 971 | pgpath->is_active = 0; |
1da177e4 LT |
972 | pgpath->fail_count++; |
973 | ||
974 | m->nr_valid_paths--; | |
975 | ||
976 | if (pgpath == m->current_pgpath) | |
977 | m->current_pgpath = NULL; | |
978 | ||
b15546f9 MA |
979 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
980 | pgpath->path.dev->name, m->nr_valid_paths); | |
981 | ||
fe9cf30e | 982 | schedule_work(&m->trigger_event); |
1da177e4 LT |
983 | |
984 | out: | |
985 | spin_unlock_irqrestore(&m->lock, flags); | |
986 | ||
987 | return 0; | |
988 | } | |
989 | ||
990 | /* | |
991 | * Reinstate a previously-failed path | |
992 | */ | |
993 | static int reinstate_path(struct pgpath *pgpath) | |
994 | { | |
995 | int r = 0; | |
996 | unsigned long flags; | |
997 | struct multipath *m = pgpath->pg->m; | |
998 | ||
999 | spin_lock_irqsave(&m->lock, flags); | |
1000 | ||
6680073d | 1001 | if (pgpath->is_active) |
1da177e4 LT |
1002 | goto out; |
1003 | ||
def052d2 | 1004 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
1005 | DMWARN("Reinstate path not supported by path selector %s", |
1006 | pgpath->pg->ps.type->name); | |
1007 | r = -EINVAL; | |
1008 | goto out; | |
1009 | } | |
1010 | ||
1011 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1012 | if (r) | |
1013 | goto out; | |
1014 | ||
6680073d | 1015 | pgpath->is_active = 1; |
1da177e4 | 1016 | |
e8099177 | 1017 | if (!m->nr_valid_paths++) { |
e54f77dd | 1018 | m->current_pgpath = NULL; |
c557308e | 1019 | queue_work(kmultipathd, &m->process_queued_ios); |
e54f77dd | 1020 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1021 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
e54f77dd CS |
1022 | m->pg_init_in_progress++; |
1023 | } | |
1da177e4 | 1024 | |
b15546f9 MA |
1025 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
1026 | pgpath->path.dev->name, m->nr_valid_paths); | |
1027 | ||
fe9cf30e | 1028 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1029 | |
1030 | out: | |
1031 | spin_unlock_irqrestore(&m->lock, flags); | |
1032 | ||
1033 | return r; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1038 | */ | |
1039 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1040 | action_fn action) | |
1041 | { | |
19040c0b | 1042 | int r = -EINVAL; |
1da177e4 LT |
1043 | struct pgpath *pgpath; |
1044 | struct priority_group *pg; | |
1045 | ||
1046 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1047 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1048 | if (pgpath->path.dev == dev) | |
1049 | r = action(pgpath); | |
1050 | } | |
1051 | } | |
1052 | ||
1053 | return r; | |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * Temporarily try to avoid having to use the specified PG | |
1058 | */ | |
1059 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
1060 | int bypassed) | |
1061 | { | |
1062 | unsigned long flags; | |
1063 | ||
1064 | spin_lock_irqsave(&m->lock, flags); | |
1065 | ||
1066 | pg->bypassed = bypassed; | |
1067 | m->current_pgpath = NULL; | |
1068 | m->current_pg = NULL; | |
1069 | ||
1070 | spin_unlock_irqrestore(&m->lock, flags); | |
1071 | ||
fe9cf30e | 1072 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1073 | } |
1074 | ||
1075 | /* | |
1076 | * Switch to using the specified PG from the next I/O that gets mapped | |
1077 | */ | |
1078 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1079 | { | |
1080 | struct priority_group *pg; | |
1081 | unsigned pgnum; | |
1082 | unsigned long flags; | |
31998ef1 | 1083 | char dummy; |
1da177e4 | 1084 | |
31998ef1 | 1085 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1086 | (pgnum > m->nr_priority_groups)) { |
1087 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1088 | return -EINVAL; | |
1089 | } | |
1090 | ||
1091 | spin_lock_irqsave(&m->lock, flags); | |
1092 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1093 | pg->bypassed = 0; | |
1094 | if (--pgnum) | |
1095 | continue; | |
1096 | ||
1097 | m->current_pgpath = NULL; | |
1098 | m->current_pg = NULL; | |
1099 | m->next_pg = pg; | |
1100 | } | |
1101 | spin_unlock_irqrestore(&m->lock, flags); | |
1102 | ||
fe9cf30e | 1103 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1104 | return 0; |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Set/clear bypassed status of a PG. | |
1109 | * PGs are numbered upwards from 1 in the order they were declared. | |
1110 | */ | |
1111 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
1112 | { | |
1113 | struct priority_group *pg; | |
1114 | unsigned pgnum; | |
31998ef1 | 1115 | char dummy; |
1da177e4 | 1116 | |
31998ef1 | 1117 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1118 | (pgnum > m->nr_priority_groups)) { |
1119 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1120 | return -EINVAL; | |
1121 | } | |
1122 | ||
1123 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1124 | if (!--pgnum) | |
1125 | break; | |
1126 | } | |
1127 | ||
1128 | bypass_pg(m, pg, bypassed); | |
1129 | return 0; | |
1130 | } | |
1131 | ||
c9e45581 DW |
1132 | /* |
1133 | * Should we retry pg_init immediately? | |
1134 | */ | |
1135 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1136 | { | |
1137 | unsigned long flags; | |
1138 | int limit_reached = 0; | |
1139 | ||
1140 | spin_lock_irqsave(&m->lock, flags); | |
1141 | ||
954a73d5 | 1142 | if (m->pg_init_count <= m->pg_init_retries && !m->pg_init_disabled) |
c9e45581 DW |
1143 | m->pg_init_required = 1; |
1144 | else | |
1145 | limit_reached = 1; | |
1146 | ||
1147 | spin_unlock_irqrestore(&m->lock, flags); | |
1148 | ||
1149 | return limit_reached; | |
1150 | } | |
1151 | ||
3ae31f6a | 1152 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1153 | { |
83c0d5d5 | 1154 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1155 | struct priority_group *pg = pgpath->pg; |
1156 | struct multipath *m = pg->m; | |
1157 | unsigned long flags; | |
4e2d19e4 | 1158 | unsigned delay_retry = 0; |
cfae5c9b CS |
1159 | |
1160 | /* device or driver problems */ | |
1161 | switch (errors) { | |
1162 | case SCSI_DH_OK: | |
1163 | break; | |
1164 | case SCSI_DH_NOSYS: | |
1165 | if (!m->hw_handler_name) { | |
1166 | errors = 0; | |
1167 | break; | |
1168 | } | |
f7b934c8 MB |
1169 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1170 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1171 | /* |
1172 | * Fail path for now, so we do not ping pong | |
1173 | */ | |
1174 | fail_path(pgpath); | |
1175 | break; | |
1176 | case SCSI_DH_DEV_TEMP_BUSY: | |
1177 | /* | |
1178 | * Probably doing something like FW upgrade on the | |
1179 | * controller so try the other pg. | |
1180 | */ | |
1181 | bypass_pg(m, pg, 1); | |
1182 | break; | |
cfae5c9b | 1183 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1184 | /* Wait before retrying. */ |
1185 | delay_retry = 1; | |
cfae5c9b CS |
1186 | case SCSI_DH_IMM_RETRY: |
1187 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1188 | if (pg_init_limit_reached(m, pgpath)) | |
1189 | fail_path(pgpath); | |
1190 | errors = 0; | |
1191 | break; | |
1192 | default: | |
1193 | /* | |
1194 | * We probably do not want to fail the path for a device | |
1195 | * error, but this is what the old dm did. In future | |
1196 | * patches we can do more advanced handling. | |
1197 | */ | |
1198 | fail_path(pgpath); | |
1199 | } | |
1200 | ||
1201 | spin_lock_irqsave(&m->lock, flags); | |
1202 | if (errors) { | |
e54f77dd CS |
1203 | if (pgpath == m->current_pgpath) { |
1204 | DMERR("Could not failover device. Error %d.", errors); | |
1205 | m->current_pgpath = NULL; | |
1206 | m->current_pg = NULL; | |
1207 | } | |
d0259bf0 | 1208 | } else if (!m->pg_init_required) |
cfae5c9b | 1209 | pg->bypassed = 0; |
cfae5c9b | 1210 | |
d0259bf0 KU |
1211 | if (--m->pg_init_in_progress) |
1212 | /* Activations of other paths are still on going */ | |
1213 | goto out; | |
1214 | ||
1215 | if (!m->pg_init_required) | |
1216 | m->queue_io = 0; | |
1217 | ||
4e2d19e4 | 1218 | m->pg_init_delay_retry = delay_retry; |
d0259bf0 KU |
1219 | queue_work(kmultipathd, &m->process_queued_ios); |
1220 | ||
2bded7bd KU |
1221 | /* |
1222 | * Wake up any thread waiting to suspend. | |
1223 | */ | |
1224 | wake_up(&m->pg_init_wait); | |
1225 | ||
d0259bf0 | 1226 | out: |
cfae5c9b CS |
1227 | spin_unlock_irqrestore(&m->lock, flags); |
1228 | } | |
1229 | ||
bab7cfc7 CS |
1230 | static void activate_path(struct work_struct *work) |
1231 | { | |
e54f77dd | 1232 | struct pgpath *pgpath = |
4e2d19e4 | 1233 | container_of(work, struct pgpath, activate_path.work); |
bab7cfc7 | 1234 | |
3ae31f6a | 1235 | scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev), |
83c0d5d5 | 1236 | pg_init_done, pgpath); |
bab7cfc7 CS |
1237 | } |
1238 | ||
7e782af5 HR |
1239 | static int noretry_error(int error) |
1240 | { | |
1241 | switch (error) { | |
1242 | case -EOPNOTSUPP: | |
1243 | case -EREMOTEIO: | |
1244 | case -EILSEQ: | |
1245 | case -ENODATA: | |
cc9d3c38 | 1246 | case -ENOSPC: |
7e782af5 HR |
1247 | return 1; |
1248 | } | |
1249 | ||
1250 | /* Anything else could be a path failure, so should be retried */ | |
1251 | return 0; | |
1252 | } | |
1253 | ||
1da177e4 LT |
1254 | /* |
1255 | * end_io handling | |
1256 | */ | |
f40c67f0 | 1257 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1258 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1259 | { |
f40c67f0 KU |
1260 | /* |
1261 | * We don't queue any clone request inside the multipath target | |
1262 | * during end I/O handling, since those clone requests don't have | |
1263 | * bio clones. If we queue them inside the multipath target, | |
1264 | * we need to make bio clones, that requires memory allocation. | |
1265 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1266 | * don't have bio clones.) | |
1267 | * Instead of queueing the clone request here, we queue the original | |
1268 | * request into dm core, which will remake a clone request and | |
1269 | * clone bios for it and resubmit it later. | |
1270 | */ | |
1271 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1272 | unsigned long flags; |
1da177e4 | 1273 | |
f40c67f0 | 1274 | if (!error && !clone->errors) |
1da177e4 LT |
1275 | return 0; /* I/O complete */ |
1276 | ||
f84cb8a4 MS |
1277 | if (noretry_error(error)) { |
1278 | if ((clone->cmd_flags & REQ_WRITE_SAME) && | |
1279 | !clone->q->limits.max_write_same_sectors) { | |
1280 | struct queue_limits *limits; | |
1281 | ||
1282 | /* device doesn't really support WRITE SAME, disable it */ | |
1283 | limits = dm_get_queue_limits(dm_table_get_md(m->ti->table)); | |
1284 | limits->max_write_same_sectors = 0; | |
1285 | } | |
959eb4e5 | 1286 | return error; |
f84cb8a4 | 1287 | } |
959eb4e5 | 1288 | |
cfae5c9b CS |
1289 | if (mpio->pgpath) |
1290 | fail_path(mpio->pgpath); | |
1da177e4 | 1291 | |
640eb3b0 | 1292 | spin_lock_irqsave(&m->lock, flags); |
751b2a7d HR |
1293 | if (!m->nr_valid_paths) { |
1294 | if (!m->queue_if_no_path) { | |
1295 | if (!__must_push_back(m)) | |
1296 | r = -EIO; | |
1297 | } else { | |
1298 | if (error == -EBADE) | |
1299 | r = error; | |
1300 | } | |
1301 | } | |
640eb3b0 | 1302 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1303 | |
f40c67f0 | 1304 | return r; |
1da177e4 LT |
1305 | } |
1306 | ||
f40c67f0 | 1307 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1308 | int error, union map_info *map_context) |
1309 | { | |
028867ac AK |
1310 | struct multipath *m = ti->private; |
1311 | struct dm_mpath_io *mpio = map_context->ptr; | |
a71a261f | 1312 | struct pgpath *pgpath; |
1da177e4 LT |
1313 | struct path_selector *ps; |
1314 | int r; | |
1315 | ||
466891f9 JN |
1316 | BUG_ON(!mpio); |
1317 | ||
f40c67f0 | 1318 | r = do_end_io(m, clone, error, mpio); |
a71a261f | 1319 | pgpath = mpio->pgpath; |
1da177e4 LT |
1320 | if (pgpath) { |
1321 | ps = &pgpath->pg->ps; | |
1322 | if (ps->type->end_io) | |
02ab823f | 1323 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1324 | } |
466891f9 | 1325 | clear_mapinfo(m, map_context); |
1da177e4 LT |
1326 | |
1327 | return r; | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1332 | * the last path fails we must error any remaining I/O. |
1333 | * Note that if the freeze_bdev fails while suspending, the | |
1334 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1335 | */ |
1336 | static void multipath_presuspend(struct dm_target *ti) | |
1337 | { | |
1338 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1339 | |
485ef69e | 1340 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1341 | } |
1342 | ||
6df400ab KU |
1343 | static void multipath_postsuspend(struct dm_target *ti) |
1344 | { | |
6380f26f MA |
1345 | struct multipath *m = ti->private; |
1346 | ||
1347 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1348 | flush_multipath_work(m); |
6380f26f | 1349 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1350 | } |
1351 | ||
436d4108 AK |
1352 | /* |
1353 | * Restore the queue_if_no_path setting. | |
1354 | */ | |
1da177e4 LT |
1355 | static void multipath_resume(struct dm_target *ti) |
1356 | { | |
1357 | struct multipath *m = (struct multipath *) ti->private; | |
1358 | unsigned long flags; | |
1359 | ||
1360 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1361 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1362 | spin_unlock_irqrestore(&m->lock, flags); |
1363 | } | |
1364 | ||
1365 | /* | |
1366 | * Info output has the following format: | |
1367 | * num_multipath_feature_args [multipath_feature_args]* | |
1368 | * num_handler_status_args [handler_status_args]* | |
1369 | * num_groups init_group_number | |
1370 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1371 | * num_paths num_selector_args | |
1372 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1373 | * | |
1374 | * Table output has the following format (identical to the constructor string): | |
1375 | * num_feature_args [features_args]* | |
1376 | * num_handler_args hw_handler [hw_handler_args]* | |
1377 | * num_groups init_group_number | |
1378 | * [priority selector-name num_ps_args [ps_args]* | |
1379 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1380 | */ | |
fd7c092e MP |
1381 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1382 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1383 | { |
1384 | int sz = 0; | |
1385 | unsigned long flags; | |
1386 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1387 | struct priority_group *pg; |
1388 | struct pgpath *p; | |
1389 | unsigned pg_num; | |
1390 | char state; | |
1391 | ||
1392 | spin_lock_irqsave(&m->lock, flags); | |
1393 | ||
1394 | /* Features */ | |
1395 | if (type == STATUSTYPE_INFO) | |
e8099177 | 1396 | DMEMIT("2 %u %u ", m->queue_io, m->pg_init_count); |
c9e45581 DW |
1397 | else { |
1398 | DMEMIT("%u ", m->queue_if_no_path + | |
4e2d19e4 | 1399 | (m->pg_init_retries > 0) * 2 + |
a58a935d MS |
1400 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
1401 | m->retain_attached_hw_handler); | |
c9e45581 DW |
1402 | if (m->queue_if_no_path) |
1403 | DMEMIT("queue_if_no_path "); | |
1404 | if (m->pg_init_retries) | |
1405 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1406 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1407 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
a58a935d MS |
1408 | if (m->retain_attached_hw_handler) |
1409 | DMEMIT("retain_attached_hw_handler "); | |
c9e45581 | 1410 | } |
1da177e4 | 1411 | |
cfae5c9b | 1412 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1413 | DMEMIT("0 "); |
1414 | else | |
cfae5c9b | 1415 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1416 | |
1417 | DMEMIT("%u ", m->nr_priority_groups); | |
1418 | ||
1419 | if (m->next_pg) | |
1420 | pg_num = m->next_pg->pg_num; | |
1421 | else if (m->current_pg) | |
1422 | pg_num = m->current_pg->pg_num; | |
1423 | else | |
a490a07a | 1424 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1425 | |
1426 | DMEMIT("%u ", pg_num); | |
1427 | ||
1428 | switch (type) { | |
1429 | case STATUSTYPE_INFO: | |
1430 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1431 | if (pg->bypassed) | |
1432 | state = 'D'; /* Disabled */ | |
1433 | else if (pg == m->current_pg) | |
1434 | state = 'A'; /* Currently Active */ | |
1435 | else | |
1436 | state = 'E'; /* Enabled */ | |
1437 | ||
1438 | DMEMIT("%c ", state); | |
1439 | ||
1440 | if (pg->ps.type->status) | |
1441 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1442 | result + sz, | |
1443 | maxlen - sz); | |
1444 | else | |
1445 | DMEMIT("0 "); | |
1446 | ||
1447 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1448 | pg->ps.type->info_args); | |
1449 | ||
1450 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1451 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1452 | p->is_active ? "A" : "F", |
1da177e4 LT |
1453 | p->fail_count); |
1454 | if (pg->ps.type->status) | |
1455 | sz += pg->ps.type->status(&pg->ps, | |
1456 | &p->path, type, result + sz, | |
1457 | maxlen - sz); | |
1458 | } | |
1459 | } | |
1460 | break; | |
1461 | ||
1462 | case STATUSTYPE_TABLE: | |
1463 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1464 | DMEMIT("%s ", pg->ps.type->name); | |
1465 | ||
1466 | if (pg->ps.type->status) | |
1467 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1468 | result + sz, | |
1469 | maxlen - sz); | |
1470 | else | |
1471 | DMEMIT("0 "); | |
1472 | ||
1473 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1474 | pg->ps.type->table_args); | |
1475 | ||
1476 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1477 | DMEMIT("%s ", p->path.dev->name); | |
1478 | if (pg->ps.type->status) | |
1479 | sz += pg->ps.type->status(&pg->ps, | |
1480 | &p->path, type, result + sz, | |
1481 | maxlen - sz); | |
1482 | } | |
1483 | } | |
1484 | break; | |
1485 | } | |
1486 | ||
1487 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1488 | } |
1489 | ||
1490 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1491 | { | |
6380f26f | 1492 | int r = -EINVAL; |
1da177e4 LT |
1493 | struct dm_dev *dev; |
1494 | struct multipath *m = (struct multipath *) ti->private; | |
1495 | action_fn action; | |
1496 | ||
6380f26f MA |
1497 | mutex_lock(&m->work_mutex); |
1498 | ||
c2f3d24b KU |
1499 | if (dm_suspended(ti)) { |
1500 | r = -EBUSY; | |
1501 | goto out; | |
1502 | } | |
1503 | ||
1da177e4 | 1504 | if (argc == 1) { |
498f0103 | 1505 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
6380f26f MA |
1506 | r = queue_if_no_path(m, 1, 0); |
1507 | goto out; | |
498f0103 | 1508 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
6380f26f MA |
1509 | r = queue_if_no_path(m, 0, 0); |
1510 | goto out; | |
1511 | } | |
1da177e4 LT |
1512 | } |
1513 | ||
6380f26f MA |
1514 | if (argc != 2) { |
1515 | DMWARN("Unrecognised multipath message received."); | |
1516 | goto out; | |
1517 | } | |
1da177e4 | 1518 | |
498f0103 | 1519 | if (!strcasecmp(argv[0], "disable_group")) { |
6380f26f MA |
1520 | r = bypass_pg_num(m, argv[1], 1); |
1521 | goto out; | |
498f0103 | 1522 | } else if (!strcasecmp(argv[0], "enable_group")) { |
6380f26f MA |
1523 | r = bypass_pg_num(m, argv[1], 0); |
1524 | goto out; | |
498f0103 | 1525 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1526 | r = switch_pg_num(m, argv[1]); |
1527 | goto out; | |
498f0103 | 1528 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1529 | action = reinstate_path; |
498f0103 | 1530 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1531 | action = fail_path; |
6380f26f MA |
1532 | else { |
1533 | DMWARN("Unrecognised multipath message received."); | |
1534 | goto out; | |
1535 | } | |
1da177e4 | 1536 | |
8215d6ec | 1537 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1538 | if (r) { |
72d94861 | 1539 | DMWARN("message: error getting device %s", |
1da177e4 | 1540 | argv[1]); |
6380f26f | 1541 | goto out; |
1da177e4 LT |
1542 | } |
1543 | ||
1544 | r = action_dev(m, dev, action); | |
1545 | ||
1546 | dm_put_device(ti, dev); | |
1547 | ||
6380f26f MA |
1548 | out: |
1549 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1550 | return r; |
1da177e4 LT |
1551 | } |
1552 | ||
647b3d00 | 1553 | static int multipath_ioctl(struct dm_target *ti, unsigned int cmd, |
9af4aa30 MB |
1554 | unsigned long arg) |
1555 | { | |
35991652 | 1556 | struct multipath *m = ti->private; |
7ba10aa6 | 1557 | struct pgpath *pgpath; |
35991652 MP |
1558 | struct block_device *bdev; |
1559 | fmode_t mode; | |
9af4aa30 | 1560 | unsigned long flags; |
35991652 MP |
1561 | int r; |
1562 | ||
35991652 MP |
1563 | bdev = NULL; |
1564 | mode = 0; | |
1565 | r = 0; | |
9af4aa30 MB |
1566 | |
1567 | spin_lock_irqsave(&m->lock, flags); | |
1568 | ||
1569 | if (!m->current_pgpath) | |
02ab823f | 1570 | __choose_pgpath(m, 0); |
9af4aa30 | 1571 | |
7ba10aa6 MS |
1572 | pgpath = m->current_pgpath; |
1573 | ||
1574 | if (pgpath) { | |
1575 | bdev = pgpath->path.dev->bdev; | |
1576 | mode = pgpath->path.dev->mode; | |
e90dae1f | 1577 | } |
9af4aa30 | 1578 | |
7ba10aa6 | 1579 | if ((pgpath && m->queue_io) || (!pgpath && m->queue_if_no_path)) |
6c182cd8 | 1580 | r = -ENOTCONN; |
9af4aa30 MB |
1581 | else if (!bdev) |
1582 | r = -EIO; | |
1583 | ||
1584 | spin_unlock_irqrestore(&m->lock, flags); | |
1585 | ||
ec8013be PB |
1586 | /* |
1587 | * Only pass ioctls through if the device sizes match exactly. | |
1588 | */ | |
a1989b33 HR |
1589 | if (!bdev || ti->len != i_size_read(bdev->bd_inode) >> SECTOR_SHIFT) { |
1590 | int err = scsi_verify_blk_ioctl(NULL, cmd); | |
1591 | if (err) | |
1592 | r = err; | |
1593 | } | |
ec8013be | 1594 | |
6c182cd8 | 1595 | if (r == -ENOTCONN && !fatal_signal_pending(current)) |
35991652 | 1596 | queue_work(kmultipathd, &m->process_queued_ios); |
35991652 | 1597 | |
633a08b8 | 1598 | return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
9af4aa30 MB |
1599 | } |
1600 | ||
af4874e0 MS |
1601 | static int multipath_iterate_devices(struct dm_target *ti, |
1602 | iterate_devices_callout_fn fn, void *data) | |
1603 | { | |
1604 | struct multipath *m = ti->private; | |
1605 | struct priority_group *pg; | |
1606 | struct pgpath *p; | |
1607 | int ret = 0; | |
1608 | ||
1609 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1610 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1611 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1612 | if (ret) |
1613 | goto out; | |
1614 | } | |
1615 | } | |
1616 | ||
1617 | out: | |
1618 | return ret; | |
1619 | } | |
1620 | ||
f40c67f0 KU |
1621 | static int __pgpath_busy(struct pgpath *pgpath) |
1622 | { | |
1623 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1624 | ||
1625 | return dm_underlying_device_busy(q); | |
1626 | } | |
1627 | ||
1628 | /* | |
1629 | * We return "busy", only when we can map I/Os but underlying devices | |
1630 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1631 | * the underlying queue). | |
1632 | * In other words, if we want to kill I/Os or queue them inside us | |
1633 | * due to map unavailability, we don't return "busy". Otherwise, | |
1634 | * dm core won't give us the I/Os and we can't do what we want. | |
1635 | */ | |
1636 | static int multipath_busy(struct dm_target *ti) | |
1637 | { | |
1638 | int busy = 0, has_active = 0; | |
1639 | struct multipath *m = ti->private; | |
1640 | struct priority_group *pg; | |
1641 | struct pgpath *pgpath; | |
1642 | unsigned long flags; | |
1643 | ||
1644 | spin_lock_irqsave(&m->lock, flags); | |
1645 | ||
b63349a7 | 1646 | /* pg_init in progress, requeue until done */ |
e8099177 | 1647 | if (!pg_ready(m)) { |
b63349a7 HR |
1648 | busy = 1; |
1649 | goto out; | |
1650 | } | |
f40c67f0 KU |
1651 | /* Guess which priority_group will be used at next mapping time */ |
1652 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1653 | pg = m->next_pg; | |
1654 | else if (likely(m->current_pg)) | |
1655 | pg = m->current_pg; | |
1656 | else | |
1657 | /* | |
1658 | * We don't know which pg will be used at next mapping time. | |
1659 | * We don't call __choose_pgpath() here to avoid to trigger | |
1660 | * pg_init just by busy checking. | |
1661 | * So we don't know whether underlying devices we will be using | |
1662 | * at next mapping time are busy or not. Just try mapping. | |
1663 | */ | |
1664 | goto out; | |
1665 | ||
1666 | /* | |
1667 | * If there is one non-busy active path at least, the path selector | |
1668 | * will be able to select it. So we consider such a pg as not busy. | |
1669 | */ | |
1670 | busy = 1; | |
1671 | list_for_each_entry(pgpath, &pg->pgpaths, list) | |
1672 | if (pgpath->is_active) { | |
1673 | has_active = 1; | |
1674 | ||
1675 | if (!__pgpath_busy(pgpath)) { | |
1676 | busy = 0; | |
1677 | break; | |
1678 | } | |
1679 | } | |
1680 | ||
1681 | if (!has_active) | |
1682 | /* | |
1683 | * No active path in this pg, so this pg won't be used and | |
1684 | * the current_pg will be changed at next mapping time. | |
1685 | * We need to try mapping to determine it. | |
1686 | */ | |
1687 | busy = 0; | |
1688 | ||
1689 | out: | |
1690 | spin_unlock_irqrestore(&m->lock, flags); | |
1691 | ||
1692 | return busy; | |
1693 | } | |
1694 | ||
1da177e4 LT |
1695 | /*----------------------------------------------------------------- |
1696 | * Module setup | |
1697 | *---------------------------------------------------------------*/ | |
1698 | static struct target_type multipath_target = { | |
1699 | .name = "multipath", | |
e8099177 | 1700 | .version = {1, 7, 0}, |
1da177e4 LT |
1701 | .module = THIS_MODULE, |
1702 | .ctr = multipath_ctr, | |
1703 | .dtr = multipath_dtr, | |
f40c67f0 KU |
1704 | .map_rq = multipath_map, |
1705 | .rq_end_io = multipath_end_io, | |
1da177e4 | 1706 | .presuspend = multipath_presuspend, |
6df400ab | 1707 | .postsuspend = multipath_postsuspend, |
1da177e4 LT |
1708 | .resume = multipath_resume, |
1709 | .status = multipath_status, | |
1710 | .message = multipath_message, | |
9af4aa30 | 1711 | .ioctl = multipath_ioctl, |
af4874e0 | 1712 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1713 | .busy = multipath_busy, |
1da177e4 LT |
1714 | }; |
1715 | ||
1716 | static int __init dm_multipath_init(void) | |
1717 | { | |
1718 | int r; | |
1719 | ||
1720 | /* allocate a slab for the dm_ios */ | |
028867ac | 1721 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1722 | if (!_mpio_cache) |
1723 | return -ENOMEM; | |
1724 | ||
1725 | r = dm_register_target(&multipath_target); | |
1726 | if (r < 0) { | |
0cd33124 | 1727 | DMERR("register failed %d", r); |
1da177e4 LT |
1728 | kmem_cache_destroy(_mpio_cache); |
1729 | return -EINVAL; | |
1730 | } | |
1731 | ||
4d4d66ab | 1732 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1733 | if (!kmultipathd) { |
0cd33124 | 1734 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1735 | dm_unregister_target(&multipath_target); |
1736 | kmem_cache_destroy(_mpio_cache); | |
1737 | return -ENOMEM; | |
1738 | } | |
1739 | ||
bab7cfc7 CS |
1740 | /* |
1741 | * A separate workqueue is used to handle the device handlers | |
1742 | * to avoid overloading existing workqueue. Overloading the | |
1743 | * old workqueue would also create a bottleneck in the | |
1744 | * path of the storage hardware device activation. | |
1745 | */ | |
4d4d66ab TH |
1746 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1747 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1748 | if (!kmpath_handlerd) { |
1749 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1750 | destroy_workqueue(kmultipathd); | |
1751 | dm_unregister_target(&multipath_target); | |
1752 | kmem_cache_destroy(_mpio_cache); | |
1753 | return -ENOMEM; | |
1754 | } | |
1755 | ||
72d94861 | 1756 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1757 | multipath_target.version[0], multipath_target.version[1], |
1758 | multipath_target.version[2]); | |
1759 | ||
1760 | return r; | |
1761 | } | |
1762 | ||
1763 | static void __exit dm_multipath_exit(void) | |
1764 | { | |
bab7cfc7 | 1765 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1766 | destroy_workqueue(kmultipathd); |
1767 | ||
10d3bd09 | 1768 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1769 | kmem_cache_destroy(_mpio_cache); |
1770 | } | |
1771 | ||
1da177e4 LT |
1772 | module_init(dm_multipath_init); |
1773 | module_exit(dm_multipath_exit); | |
1774 | ||
1775 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1776 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1777 | MODULE_LICENSE("GPL"); |