Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software Limited. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
586e80e6 MP |
8 | #include <linux/device-mapper.h> |
9 | ||
1da177e4 | 10 | #include "dm-path-selector.h" |
b15546f9 | 11 | #include "dm-uevent.h" |
1da177e4 LT |
12 | |
13 | #include <linux/ctype.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/mempool.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/time.h> | |
20 | #include <linux/workqueue.h> | |
35991652 | 21 | #include <linux/delay.h> |
cfae5c9b | 22 | #include <scsi/scsi_dh.h> |
60063497 | 23 | #include <linux/atomic.h> |
1da177e4 | 24 | |
72d94861 | 25 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
26 | #define DM_PG_INIT_DELAY_MSECS 2000 |
27 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
1da177e4 LT |
28 | |
29 | /* Path properties */ | |
30 | struct pgpath { | |
31 | struct list_head list; | |
32 | ||
33 | struct priority_group *pg; /* Owning PG */ | |
6680073d | 34 | unsigned is_active; /* Path status */ |
1da177e4 LT |
35 | unsigned fail_count; /* Cumulative failure count */ |
36 | ||
c922d5f7 | 37 | struct dm_path path; |
4e2d19e4 | 38 | struct delayed_work activate_path; |
1da177e4 LT |
39 | }; |
40 | ||
41 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
42 | ||
43 | /* | |
44 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
45 | * Each has a path selector which controls which path gets used. | |
46 | */ | |
47 | struct priority_group { | |
48 | struct list_head list; | |
49 | ||
50 | struct multipath *m; /* Owning multipath instance */ | |
51 | struct path_selector ps; | |
52 | ||
53 | unsigned pg_num; /* Reference number */ | |
54 | unsigned bypassed; /* Temporarily bypass this PG? */ | |
55 | ||
56 | unsigned nr_pgpaths; /* Number of paths in PG */ | |
57 | struct list_head pgpaths; | |
58 | }; | |
59 | ||
60 | /* Multipath context */ | |
61 | struct multipath { | |
62 | struct list_head list; | |
63 | struct dm_target *ti; | |
64 | ||
cfae5c9b | 65 | const char *hw_handler_name; |
2bfd2e13 | 66 | char *hw_handler_params; |
4e2d19e4 | 67 | |
1fbdd2b3 MS |
68 | spinlock_t lock; |
69 | ||
1da177e4 LT |
70 | unsigned nr_priority_groups; |
71 | struct list_head priority_groups; | |
4e2d19e4 CS |
72 | |
73 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
74 | ||
1da177e4 | 75 | unsigned pg_init_required; /* pg_init needs calling? */ |
c3cd4f6b | 76 | unsigned pg_init_in_progress; /* Only one pg_init allowed at once */ |
4e2d19e4 | 77 | unsigned pg_init_delay_retry; /* Delay pg_init retry? */ |
1da177e4 LT |
78 | |
79 | unsigned nr_valid_paths; /* Total number of usable paths */ | |
80 | struct pgpath *current_pgpath; | |
81 | struct priority_group *current_pg; | |
82 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
83 | unsigned repeat_count; /* I/Os left before calling PS again */ | |
84 | ||
1fbdd2b3 MS |
85 | unsigned queue_io:1; /* Must we queue all I/O? */ |
86 | unsigned queue_if_no_path:1; /* Queue I/O if last path fails? */ | |
87 | unsigned saved_queue_if_no_path:1; /* Saved state during suspension */ | |
a58a935d | 88 | unsigned retain_attached_hw_handler:1; /* If there's already a hw_handler present, don't change it. */ |
1fbdd2b3 | 89 | |
c9e45581 DW |
90 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
91 | unsigned pg_init_count; /* Number of times pg_init called */ | |
4e2d19e4 | 92 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
1da177e4 | 93 | |
1fbdd2b3 | 94 | unsigned queue_size; |
1da177e4 | 95 | struct work_struct process_queued_ios; |
f40c67f0 | 96 | struct list_head 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 | ||
119 | #define MIN_IOS 256 /* Mempool size */ | |
120 | ||
e18b890b | 121 | static struct kmem_cache *_mpio_cache; |
1da177e4 | 122 | |
bab7cfc7 | 123 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 DH |
124 | static void process_queued_ios(struct work_struct *work); |
125 | static void trigger_event(struct work_struct *work); | |
bab7cfc7 | 126 | static void activate_path(struct work_struct *work); |
1da177e4 LT |
127 | |
128 | ||
129 | /*----------------------------------------------- | |
130 | * Allocation routines | |
131 | *-----------------------------------------------*/ | |
132 | ||
133 | static struct pgpath *alloc_pgpath(void) | |
134 | { | |
e69fae56 | 135 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 136 | |
224cb3e9 | 137 | if (pgpath) { |
6680073d | 138 | pgpath->is_active = 1; |
4e2d19e4 | 139 | INIT_DELAYED_WORK(&pgpath->activate_path, activate_path); |
224cb3e9 | 140 | } |
1da177e4 LT |
141 | |
142 | return pgpath; | |
143 | } | |
144 | ||
028867ac | 145 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
146 | { |
147 | kfree(pgpath); | |
148 | } | |
149 | ||
150 | static struct priority_group *alloc_priority_group(void) | |
151 | { | |
152 | struct priority_group *pg; | |
153 | ||
e69fae56 | 154 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 155 | |
e69fae56 MM |
156 | if (pg) |
157 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
158 | |
159 | return pg; | |
160 | } | |
161 | ||
162 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
163 | { | |
164 | struct pgpath *pgpath, *tmp; | |
ae11b1b3 | 165 | struct multipath *m = ti->private; |
1da177e4 LT |
166 | |
167 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
168 | list_del(&pgpath->list); | |
ae11b1b3 HR |
169 | if (m->hw_handler_name) |
170 | scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev)); | |
1da177e4 LT |
171 | dm_put_device(ti, pgpath->path.dev); |
172 | free_pgpath(pgpath); | |
173 | } | |
174 | } | |
175 | ||
176 | static void free_priority_group(struct priority_group *pg, | |
177 | struct dm_target *ti) | |
178 | { | |
179 | struct path_selector *ps = &pg->ps; | |
180 | ||
181 | if (ps->type) { | |
182 | ps->type->destroy(ps); | |
183 | dm_put_path_selector(ps->type); | |
184 | } | |
185 | ||
186 | free_pgpaths(&pg->pgpaths, ti); | |
187 | kfree(pg); | |
188 | } | |
189 | ||
28f16c20 | 190 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
191 | { |
192 | struct multipath *m; | |
193 | ||
e69fae56 | 194 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 195 | if (m) { |
1da177e4 | 196 | INIT_LIST_HEAD(&m->priority_groups); |
f40c67f0 | 197 | INIT_LIST_HEAD(&m->queued_ios); |
1da177e4 LT |
198 | spin_lock_init(&m->lock); |
199 | m->queue_io = 1; | |
4e2d19e4 | 200 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; |
c4028958 DH |
201 | INIT_WORK(&m->process_queued_ios, process_queued_ios); |
202 | INIT_WORK(&m->trigger_event, trigger_event); | |
2bded7bd | 203 | init_waitqueue_head(&m->pg_init_wait); |
6380f26f | 204 | mutex_init(&m->work_mutex); |
93d2341c | 205 | m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache); |
1da177e4 LT |
206 | if (!m->mpio_pool) { |
207 | kfree(m); | |
208 | return NULL; | |
209 | } | |
28f16c20 MM |
210 | m->ti = ti; |
211 | ti->private = m; | |
1da177e4 LT |
212 | } |
213 | ||
214 | return m; | |
215 | } | |
216 | ||
217 | static void free_multipath(struct multipath *m) | |
218 | { | |
219 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
220 | |
221 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
222 | list_del(&pg->list); | |
223 | free_priority_group(pg, m->ti); | |
224 | } | |
225 | ||
cfae5c9b | 226 | kfree(m->hw_handler_name); |
2bfd2e13 | 227 | kfree(m->hw_handler_params); |
1da177e4 LT |
228 | mempool_destroy(m->mpio_pool); |
229 | kfree(m); | |
230 | } | |
231 | ||
466891f9 JN |
232 | static int set_mapinfo(struct multipath *m, union map_info *info) |
233 | { | |
234 | struct dm_mpath_io *mpio; | |
235 | ||
236 | mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC); | |
237 | if (!mpio) | |
238 | return -ENOMEM; | |
239 | ||
240 | memset(mpio, 0, sizeof(*mpio)); | |
241 | info->ptr = mpio; | |
242 | ||
243 | return 0; | |
244 | } | |
245 | ||
246 | static void clear_mapinfo(struct multipath *m, union map_info *info) | |
247 | { | |
248 | struct dm_mpath_io *mpio = info->ptr; | |
249 | ||
250 | info->ptr = NULL; | |
251 | mempool_free(mpio, m->mpio_pool); | |
252 | } | |
1da177e4 LT |
253 | |
254 | /*----------------------------------------------- | |
255 | * Path selection | |
256 | *-----------------------------------------------*/ | |
257 | ||
fb612642 KU |
258 | static void __pg_init_all_paths(struct multipath *m) |
259 | { | |
260 | struct pgpath *pgpath; | |
4e2d19e4 | 261 | unsigned long pg_init_delay = 0; |
fb612642 KU |
262 | |
263 | m->pg_init_count++; | |
264 | m->pg_init_required = 0; | |
4e2d19e4 CS |
265 | if (m->pg_init_delay_retry) |
266 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? | |
267 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
268 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
269 | /* Skip failed paths */ | |
270 | if (!pgpath->is_active) | |
271 | continue; | |
4e2d19e4 CS |
272 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
273 | pg_init_delay)) | |
fb612642 KU |
274 | m->pg_init_in_progress++; |
275 | } | |
276 | } | |
277 | ||
1da177e4 LT |
278 | static void __switch_pg(struct multipath *m, struct pgpath *pgpath) |
279 | { | |
1da177e4 LT |
280 | m->current_pg = pgpath->pg; |
281 | ||
282 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 283 | if (m->hw_handler_name) { |
1da177e4 LT |
284 | m->pg_init_required = 1; |
285 | m->queue_io = 1; | |
286 | } else { | |
287 | m->pg_init_required = 0; | |
288 | m->queue_io = 0; | |
289 | } | |
c9e45581 DW |
290 | |
291 | m->pg_init_count = 0; | |
1da177e4 LT |
292 | } |
293 | ||
02ab823f KU |
294 | static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg, |
295 | size_t nr_bytes) | |
1da177e4 | 296 | { |
c922d5f7 | 297 | struct dm_path *path; |
1da177e4 | 298 | |
02ab823f | 299 | path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes); |
1da177e4 LT |
300 | if (!path) |
301 | return -ENXIO; | |
302 | ||
303 | m->current_pgpath = path_to_pgpath(path); | |
304 | ||
305 | if (m->current_pg != pg) | |
306 | __switch_pg(m, m->current_pgpath); | |
307 | ||
308 | return 0; | |
309 | } | |
310 | ||
02ab823f | 311 | static void __choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 LT |
312 | { |
313 | struct priority_group *pg; | |
314 | unsigned bypassed = 1; | |
315 | ||
316 | if (!m->nr_valid_paths) | |
317 | goto failed; | |
318 | ||
319 | /* Were we instructed to switch PG? */ | |
320 | if (m->next_pg) { | |
321 | pg = m->next_pg; | |
322 | m->next_pg = NULL; | |
02ab823f | 323 | if (!__choose_path_in_pg(m, pg, nr_bytes)) |
1da177e4 LT |
324 | return; |
325 | } | |
326 | ||
327 | /* Don't change PG until it has no remaining paths */ | |
02ab823f | 328 | if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes)) |
1da177e4 LT |
329 | return; |
330 | ||
331 | /* | |
332 | * Loop through priority groups until we find a valid path. | |
333 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
334 | * Second time we only try the ones we skipped, but set |
335 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
336 | */ |
337 | do { | |
338 | list_for_each_entry(pg, &m->priority_groups, list) { | |
339 | if (pg->bypassed == bypassed) | |
340 | continue; | |
f220fd4e MC |
341 | if (!__choose_path_in_pg(m, pg, nr_bytes)) { |
342 | if (!bypassed) | |
343 | m->pg_init_delay_retry = 1; | |
1da177e4 | 344 | return; |
f220fd4e | 345 | } |
1da177e4 LT |
346 | } |
347 | } while (bypassed--); | |
348 | ||
349 | failed: | |
350 | m->current_pgpath = NULL; | |
351 | m->current_pg = NULL; | |
352 | } | |
353 | ||
45e15720 KU |
354 | /* |
355 | * Check whether bios must be queued in the device-mapper core rather | |
356 | * than here in the target. | |
357 | * | |
358 | * m->lock must be held on entry. | |
359 | * | |
360 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the | |
361 | * same value then we are not between multipath_presuspend() | |
362 | * and multipath_resume() calls and we have no need to check | |
363 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
364 | */ | |
365 | static int __must_push_back(struct multipath *m) | |
366 | { | |
367 | return (m->queue_if_no_path != m->saved_queue_if_no_path && | |
368 | dm_noflush_suspending(m->ti)); | |
369 | } | |
370 | ||
f40c67f0 | 371 | static int map_io(struct multipath *m, struct request *clone, |
466891f9 | 372 | union map_info *map_context, unsigned was_queued) |
1da177e4 | 373 | { |
d2a7ad29 | 374 | int r = DM_MAPIO_REMAPPED; |
f40c67f0 | 375 | size_t nr_bytes = blk_rq_bytes(clone); |
1da177e4 LT |
376 | unsigned long flags; |
377 | struct pgpath *pgpath; | |
f40c67f0 | 378 | struct block_device *bdev; |
466891f9 | 379 | struct dm_mpath_io *mpio = map_context->ptr; |
1da177e4 LT |
380 | |
381 | spin_lock_irqsave(&m->lock, flags); | |
382 | ||
383 | /* Do we need to select a new pgpath? */ | |
384 | if (!m->current_pgpath || | |
385 | (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) | |
02ab823f | 386 | __choose_pgpath(m, nr_bytes); |
1da177e4 LT |
387 | |
388 | pgpath = m->current_pgpath; | |
389 | ||
390 | if (was_queued) | |
391 | m->queue_size--; | |
392 | ||
393 | if ((pgpath && m->queue_io) || | |
436d4108 | 394 | (!pgpath && m->queue_if_no_path)) { |
1da177e4 | 395 | /* Queue for the daemon to resubmit */ |
f40c67f0 | 396 | list_add_tail(&clone->queuelist, &m->queued_ios); |
1da177e4 | 397 | m->queue_size++; |
c3cd4f6b AK |
398 | if ((m->pg_init_required && !m->pg_init_in_progress) || |
399 | !m->queue_io) | |
c557308e | 400 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 | 401 | pgpath = NULL; |
d2a7ad29 | 402 | r = DM_MAPIO_SUBMITTED; |
f40c67f0 KU |
403 | } else if (pgpath) { |
404 | bdev = pgpath->path.dev->bdev; | |
405 | clone->q = bdev_get_queue(bdev); | |
406 | clone->rq_disk = bdev->bd_disk; | |
407 | } else if (__must_push_back(m)) | |
45e15720 KU |
408 | r = DM_MAPIO_REQUEUE; |
409 | else | |
410 | r = -EIO; /* Failed */ | |
1da177e4 LT |
411 | |
412 | mpio->pgpath = pgpath; | |
02ab823f KU |
413 | mpio->nr_bytes = nr_bytes; |
414 | ||
415 | if (r == DM_MAPIO_REMAPPED && pgpath->pg->ps.type->start_io) | |
416 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, &pgpath->path, | |
417 | nr_bytes); | |
1da177e4 LT |
418 | |
419 | spin_unlock_irqrestore(&m->lock, flags); | |
420 | ||
421 | return r; | |
422 | } | |
423 | ||
424 | /* | |
425 | * If we run out of usable paths, should we queue I/O or error it? | |
426 | */ | |
485ef69e AK |
427 | static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path, |
428 | unsigned save_old_value) | |
1da177e4 LT |
429 | { |
430 | unsigned long flags; | |
431 | ||
432 | spin_lock_irqsave(&m->lock, flags); | |
433 | ||
485ef69e AK |
434 | if (save_old_value) |
435 | m->saved_queue_if_no_path = m->queue_if_no_path; | |
436 | else | |
437 | m->saved_queue_if_no_path = queue_if_no_path; | |
1da177e4 | 438 | m->queue_if_no_path = queue_if_no_path; |
c3cd4f6b | 439 | if (!m->queue_if_no_path && m->queue_size) |
c557308e | 440 | queue_work(kmultipathd, &m->process_queued_ios); |
1da177e4 LT |
441 | |
442 | spin_unlock_irqrestore(&m->lock, flags); | |
443 | ||
444 | return 0; | |
445 | } | |
446 | ||
447 | /*----------------------------------------------------------------- | |
448 | * The multipath daemon is responsible for resubmitting queued ios. | |
449 | *---------------------------------------------------------------*/ | |
450 | ||
451 | static void dispatch_queued_ios(struct multipath *m) | |
452 | { | |
453 | int r; | |
454 | unsigned long flags; | |
1da177e4 | 455 | union map_info *info; |
f40c67f0 KU |
456 | struct request *clone, *n; |
457 | LIST_HEAD(cl); | |
1da177e4 LT |
458 | |
459 | spin_lock_irqsave(&m->lock, flags); | |
f40c67f0 | 460 | list_splice_init(&m->queued_ios, &cl); |
1da177e4 LT |
461 | spin_unlock_irqrestore(&m->lock, flags); |
462 | ||
f40c67f0 KU |
463 | list_for_each_entry_safe(clone, n, &cl, queuelist) { |
464 | list_del_init(&clone->queuelist); | |
1da177e4 | 465 | |
f40c67f0 | 466 | info = dm_get_rq_mapinfo(clone); |
1da177e4 | 467 | |
466891f9 | 468 | r = map_io(m, clone, info, 1); |
f40c67f0 | 469 | if (r < 0) { |
466891f9 | 470 | clear_mapinfo(m, info); |
f40c67f0 KU |
471 | dm_kill_unmapped_request(clone, r); |
472 | } else if (r == DM_MAPIO_REMAPPED) | |
473 | dm_dispatch_request(clone); | |
474 | else if (r == DM_MAPIO_REQUEUE) { | |
466891f9 | 475 | clear_mapinfo(m, info); |
f40c67f0 KU |
476 | dm_requeue_unmapped_request(clone); |
477 | } | |
1da177e4 LT |
478 | } |
479 | } | |
480 | ||
c4028958 | 481 | static void process_queued_ios(struct work_struct *work) |
1da177e4 | 482 | { |
c4028958 DH |
483 | struct multipath *m = |
484 | container_of(work, struct multipath, process_queued_ios); | |
fb612642 | 485 | struct pgpath *pgpath = NULL; |
e54f77dd | 486 | unsigned must_queue = 1; |
1da177e4 LT |
487 | unsigned long flags; |
488 | ||
489 | spin_lock_irqsave(&m->lock, flags); | |
490 | ||
491 | if (!m->current_pgpath) | |
02ab823f | 492 | __choose_pgpath(m, 0); |
1da177e4 LT |
493 | |
494 | pgpath = m->current_pgpath; | |
495 | ||
c3cd4f6b AK |
496 | if ((pgpath && !m->queue_io) || |
497 | (!pgpath && !m->queue_if_no_path)) | |
498 | must_queue = 0; | |
1da177e4 | 499 | |
fb612642 KU |
500 | if (m->pg_init_required && !m->pg_init_in_progress && pgpath) |
501 | __pg_init_all_paths(m); | |
502 | ||
1da177e4 | 503 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
504 | if (!must_queue) |
505 | dispatch_queued_ios(m); | |
506 | } | |
507 | ||
508 | /* | |
509 | * An event is triggered whenever a path is taken out of use. | |
510 | * Includes path failure and PG bypass. | |
511 | */ | |
c4028958 | 512 | static void trigger_event(struct work_struct *work) |
1da177e4 | 513 | { |
c4028958 DH |
514 | struct multipath *m = |
515 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
516 | |
517 | dm_table_event(m->ti->table); | |
518 | } | |
519 | ||
520 | /*----------------------------------------------------------------- | |
521 | * Constructor/argument parsing: | |
522 | * <#multipath feature args> [<arg>]* | |
523 | * <#hw_handler args> [hw_handler [<arg>]*] | |
524 | * <#priority groups> | |
525 | * <initial priority group> | |
526 | * [<selector> <#selector args> [<arg>]* | |
527 | * <#paths> <#per-path selector args> | |
528 | * [<path> [<arg>]* ]+ ]+ | |
529 | *---------------------------------------------------------------*/ | |
498f0103 | 530 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
531 | struct dm_target *ti) |
532 | { | |
533 | int r; | |
534 | struct path_selector_type *pst; | |
535 | unsigned ps_argc; | |
536 | ||
498f0103 | 537 | static struct dm_arg _args[] = { |
72d94861 | 538 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
539 | }; |
540 | ||
498f0103 | 541 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 542 | if (!pst) { |
72d94861 | 543 | ti->error = "unknown path selector type"; |
1da177e4 LT |
544 | return -EINVAL; |
545 | } | |
546 | ||
498f0103 | 547 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
548 | if (r) { |
549 | dm_put_path_selector(pst); | |
1da177e4 | 550 | return -EINVAL; |
371b2e34 | 551 | } |
1da177e4 LT |
552 | |
553 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
554 | if (r) { | |
555 | dm_put_path_selector(pst); | |
72d94861 | 556 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
557 | return r; |
558 | } | |
559 | ||
560 | pg->ps.type = pst; | |
498f0103 | 561 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
562 | |
563 | return 0; | |
564 | } | |
565 | ||
498f0103 | 566 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
567 | struct dm_target *ti) |
568 | { | |
569 | int r; | |
570 | struct pgpath *p; | |
ae11b1b3 | 571 | struct multipath *m = ti->private; |
a58a935d MS |
572 | struct request_queue *q = NULL; |
573 | const char *attached_handler_name; | |
1da177e4 LT |
574 | |
575 | /* we need at least a path arg */ | |
576 | if (as->argc < 1) { | |
72d94861 | 577 | ti->error = "no device given"; |
01460f35 | 578 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
579 | } |
580 | ||
581 | p = alloc_pgpath(); | |
582 | if (!p) | |
01460f35 | 583 | return ERR_PTR(-ENOMEM); |
1da177e4 | 584 | |
498f0103 | 585 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 586 | &p->path.dev); |
1da177e4 | 587 | if (r) { |
72d94861 | 588 | ti->error = "error getting device"; |
1da177e4 LT |
589 | goto bad; |
590 | } | |
591 | ||
a58a935d MS |
592 | if (m->retain_attached_hw_handler || m->hw_handler_name) |
593 | q = bdev_get_queue(p->path.dev->bdev); | |
594 | ||
595 | if (m->retain_attached_hw_handler) { | |
596 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); | |
597 | if (attached_handler_name) { | |
598 | /* | |
599 | * Reset hw_handler_name to match the attached handler | |
600 | * and clear any hw_handler_params associated with the | |
601 | * ignored handler. | |
602 | * | |
603 | * NB. This modifies the table line to show the actual | |
604 | * handler instead of the original table passed in. | |
605 | */ | |
606 | kfree(m->hw_handler_name); | |
607 | m->hw_handler_name = attached_handler_name; | |
608 | ||
609 | kfree(m->hw_handler_params); | |
610 | m->hw_handler_params = NULL; | |
611 | } | |
612 | } | |
a0cf7ea9 | 613 | |
a58a935d MS |
614 | if (m->hw_handler_name) { |
615 | /* | |
616 | * Increments scsi_dh reference, even when using an | |
617 | * already-attached handler. | |
618 | */ | |
a0cf7ea9 HR |
619 | r = scsi_dh_attach(q, m->hw_handler_name); |
620 | if (r == -EBUSY) { | |
621 | /* | |
a58a935d | 622 | * Already attached to different hw_handler: |
a0cf7ea9 HR |
623 | * try to reattach with correct one. |
624 | */ | |
625 | scsi_dh_detach(q); | |
626 | r = scsi_dh_attach(q, m->hw_handler_name); | |
627 | } | |
628 | ||
ae11b1b3 | 629 | if (r < 0) { |
a0cf7ea9 | 630 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
631 | dm_put_device(ti, p->path.dev); |
632 | goto bad; | |
633 | } | |
2bfd2e13 CS |
634 | |
635 | if (m->hw_handler_params) { | |
636 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
637 | if (r < 0) { | |
638 | ti->error = "unable to set hardware " | |
639 | "handler parameters"; | |
640 | scsi_dh_detach(q); | |
641 | dm_put_device(ti, p->path.dev); | |
642 | goto bad; | |
643 | } | |
644 | } | |
ae11b1b3 HR |
645 | } |
646 | ||
1da177e4 LT |
647 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
648 | if (r) { | |
649 | dm_put_device(ti, p->path.dev); | |
650 | goto bad; | |
651 | } | |
652 | ||
653 | return p; | |
654 | ||
655 | bad: | |
656 | free_pgpath(p); | |
01460f35 | 657 | return ERR_PTR(r); |
1da177e4 LT |
658 | } |
659 | ||
498f0103 | 660 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 661 | struct multipath *m) |
1da177e4 | 662 | { |
498f0103 | 663 | static struct dm_arg _args[] = { |
72d94861 AK |
664 | {1, 1024, "invalid number of paths"}, |
665 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
666 | }; |
667 | ||
668 | int r; | |
498f0103 | 669 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 670 | struct priority_group *pg; |
28f16c20 | 671 | struct dm_target *ti = m->ti; |
1da177e4 LT |
672 | |
673 | if (as->argc < 2) { | |
674 | as->argc = 0; | |
01460f35 BM |
675 | ti->error = "not enough priority group arguments"; |
676 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
677 | } |
678 | ||
679 | pg = alloc_priority_group(); | |
680 | if (!pg) { | |
72d94861 | 681 | ti->error = "couldn't allocate priority group"; |
01460f35 | 682 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
683 | } |
684 | pg->m = m; | |
685 | ||
686 | r = parse_path_selector(as, pg, ti); | |
687 | if (r) | |
688 | goto bad; | |
689 | ||
690 | /* | |
691 | * read the paths | |
692 | */ | |
498f0103 | 693 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
694 | if (r) |
695 | goto bad; | |
696 | ||
498f0103 | 697 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
698 | if (r) |
699 | goto bad; | |
700 | ||
498f0103 | 701 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
702 | for (i = 0; i < pg->nr_pgpaths; i++) { |
703 | struct pgpath *pgpath; | |
498f0103 | 704 | struct dm_arg_set path_args; |
1da177e4 | 705 | |
498f0103 | 706 | if (as->argc < nr_args) { |
148acff6 | 707 | ti->error = "not enough path parameters"; |
6bbf79a1 | 708 | r = -EINVAL; |
1da177e4 | 709 | goto bad; |
148acff6 | 710 | } |
1da177e4 | 711 | |
498f0103 | 712 | path_args.argc = nr_args; |
1da177e4 LT |
713 | path_args.argv = as->argv; |
714 | ||
715 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
716 | if (IS_ERR(pgpath)) { |
717 | r = PTR_ERR(pgpath); | |
1da177e4 | 718 | goto bad; |
01460f35 | 719 | } |
1da177e4 LT |
720 | |
721 | pgpath->pg = pg; | |
722 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 723 | dm_consume_args(as, nr_args); |
1da177e4 LT |
724 | } |
725 | ||
726 | return pg; | |
727 | ||
728 | bad: | |
729 | free_priority_group(pg, ti); | |
01460f35 | 730 | return ERR_PTR(r); |
1da177e4 LT |
731 | } |
732 | ||
498f0103 | 733 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 734 | { |
1da177e4 | 735 | unsigned hw_argc; |
2bfd2e13 | 736 | int ret; |
28f16c20 | 737 | struct dm_target *ti = m->ti; |
1da177e4 | 738 | |
498f0103 | 739 | static struct dm_arg _args[] = { |
72d94861 | 740 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
741 | }; |
742 | ||
498f0103 | 743 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
744 | return -EINVAL; |
745 | ||
746 | if (!hw_argc) | |
747 | return 0; | |
748 | ||
498f0103 | 749 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
510193a2 MS |
750 | if (!try_then_request_module(scsi_dh_handler_exist(m->hw_handler_name), |
751 | "scsi_dh_%s", m->hw_handler_name)) { | |
72d94861 | 752 | ti->error = "unknown hardware handler type"; |
2bfd2e13 CS |
753 | ret = -EINVAL; |
754 | goto fail; | |
1da177e4 | 755 | } |
14e98c5c | 756 | |
2bfd2e13 CS |
757 | if (hw_argc > 1) { |
758 | char *p; | |
759 | int i, j, len = 4; | |
760 | ||
761 | for (i = 0; i <= hw_argc - 2; i++) | |
762 | len += strlen(as->argv[i]) + 1; | |
763 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
764 | if (!p) { | |
765 | ti->error = "memory allocation failed"; | |
766 | ret = -ENOMEM; | |
767 | goto fail; | |
768 | } | |
769 | j = sprintf(p, "%d", hw_argc - 1); | |
770 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
771 | j = sprintf(p, "%s", as->argv[i]); | |
772 | } | |
498f0103 | 773 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
774 | |
775 | return 0; | |
2bfd2e13 CS |
776 | fail: |
777 | kfree(m->hw_handler_name); | |
778 | m->hw_handler_name = NULL; | |
779 | return ret; | |
1da177e4 LT |
780 | } |
781 | ||
498f0103 | 782 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
783 | { |
784 | int r; | |
785 | unsigned argc; | |
28f16c20 | 786 | struct dm_target *ti = m->ti; |
498f0103 | 787 | const char *arg_name; |
1da177e4 | 788 | |
498f0103 | 789 | static struct dm_arg _args[] = { |
a58a935d | 790 | {0, 6, "invalid number of feature args"}, |
c9e45581 | 791 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 792 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
793 | }; |
794 | ||
498f0103 | 795 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
796 | if (r) |
797 | return -EINVAL; | |
798 | ||
799 | if (!argc) | |
800 | return 0; | |
801 | ||
c9e45581 | 802 | do { |
498f0103 | 803 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
804 | argc--; |
805 | ||
498f0103 | 806 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
c9e45581 DW |
807 | r = queue_if_no_path(m, 1, 0); |
808 | continue; | |
809 | } | |
810 | ||
a58a935d MS |
811 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
812 | m->retain_attached_hw_handler = 1; | |
813 | continue; | |
814 | } | |
815 | ||
498f0103 | 816 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 817 | (argc >= 1)) { |
498f0103 | 818 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
819 | argc--; |
820 | continue; | |
821 | } | |
822 | ||
498f0103 | 823 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 824 | (argc >= 1)) { |
498f0103 | 825 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
826 | argc--; |
827 | continue; | |
828 | } | |
829 | ||
1da177e4 | 830 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
831 | r = -EINVAL; |
832 | } while (argc && !r); | |
833 | ||
834 | return r; | |
1da177e4 LT |
835 | } |
836 | ||
837 | static int multipath_ctr(struct dm_target *ti, unsigned int argc, | |
838 | char **argv) | |
839 | { | |
498f0103 MS |
840 | /* target arguments */ |
841 | static struct dm_arg _args[] = { | |
a490a07a MS |
842 | {0, 1024, "invalid number of priority groups"}, |
843 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
844 | }; |
845 | ||
846 | int r; | |
847 | struct multipath *m; | |
498f0103 | 848 | struct dm_arg_set as; |
1da177e4 LT |
849 | unsigned pg_count = 0; |
850 | unsigned next_pg_num; | |
851 | ||
852 | as.argc = argc; | |
853 | as.argv = argv; | |
854 | ||
28f16c20 | 855 | m = alloc_multipath(ti); |
1da177e4 | 856 | if (!m) { |
72d94861 | 857 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
858 | return -EINVAL; |
859 | } | |
860 | ||
28f16c20 | 861 | r = parse_features(&as, m); |
1da177e4 LT |
862 | if (r) |
863 | goto bad; | |
864 | ||
28f16c20 | 865 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
866 | if (r) |
867 | goto bad; | |
868 | ||
498f0103 | 869 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
870 | if (r) |
871 | goto bad; | |
872 | ||
498f0103 | 873 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
874 | if (r) |
875 | goto bad; | |
876 | ||
a490a07a MS |
877 | if ((!m->nr_priority_groups && next_pg_num) || |
878 | (m->nr_priority_groups && !next_pg_num)) { | |
879 | ti->error = "invalid initial priority group"; | |
880 | r = -EINVAL; | |
881 | goto bad; | |
882 | } | |
883 | ||
1da177e4 LT |
884 | /* parse the priority groups */ |
885 | while (as.argc) { | |
886 | struct priority_group *pg; | |
887 | ||
28f16c20 | 888 | pg = parse_priority_group(&as, m); |
01460f35 BM |
889 | if (IS_ERR(pg)) { |
890 | r = PTR_ERR(pg); | |
1da177e4 LT |
891 | goto bad; |
892 | } | |
893 | ||
894 | m->nr_valid_paths += pg->nr_pgpaths; | |
895 | list_add_tail(&pg->list, &m->priority_groups); | |
896 | pg_count++; | |
897 | pg->pg_num = pg_count; | |
898 | if (!--next_pg_num) | |
899 | m->next_pg = pg; | |
900 | } | |
901 | ||
902 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 903 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
904 | r = -EINVAL; |
905 | goto bad; | |
906 | } | |
907 | ||
55a62eef AK |
908 | ti->num_flush_bios = 1; |
909 | ti->num_discard_bios = 1; | |
042bcef8 | 910 | ti->num_write_same_bios = 1; |
8627921f | 911 | |
1da177e4 LT |
912 | return 0; |
913 | ||
914 | bad: | |
915 | free_multipath(m); | |
916 | return r; | |
917 | } | |
918 | ||
2bded7bd KU |
919 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
920 | { | |
921 | DECLARE_WAITQUEUE(wait, current); | |
922 | unsigned long flags; | |
923 | ||
924 | add_wait_queue(&m->pg_init_wait, &wait); | |
925 | ||
926 | while (1) { | |
927 | set_current_state(TASK_UNINTERRUPTIBLE); | |
928 | ||
929 | spin_lock_irqsave(&m->lock, flags); | |
930 | if (!m->pg_init_in_progress) { | |
931 | spin_unlock_irqrestore(&m->lock, flags); | |
932 | break; | |
933 | } | |
934 | spin_unlock_irqrestore(&m->lock, flags); | |
935 | ||
936 | io_schedule(); | |
937 | } | |
938 | set_current_state(TASK_RUNNING); | |
939 | ||
940 | remove_wait_queue(&m->pg_init_wait, &wait); | |
941 | } | |
942 | ||
943 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 944 | { |
bab7cfc7 | 945 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 946 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 947 | flush_workqueue(kmultipathd); |
43829731 | 948 | flush_work(&m->trigger_event); |
6df400ab KU |
949 | } |
950 | ||
951 | static void multipath_dtr(struct dm_target *ti) | |
952 | { | |
953 | struct multipath *m = ti->private; | |
954 | ||
2bded7bd | 955 | flush_multipath_work(m); |
1da177e4 LT |
956 | free_multipath(m); |
957 | } | |
958 | ||
959 | /* | |
f40c67f0 | 960 | * Map cloned requests |
1da177e4 | 961 | */ |
f40c67f0 | 962 | static int multipath_map(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
963 | union map_info *map_context) |
964 | { | |
965 | int r; | |
1da177e4 LT |
966 | struct multipath *m = (struct multipath *) ti->private; |
967 | ||
466891f9 | 968 | if (set_mapinfo(m, map_context) < 0) |
f40c67f0 KU |
969 | /* ENOMEM, requeue */ |
970 | return DM_MAPIO_REQUEUE; | |
1da177e4 | 971 | |
f40c67f0 | 972 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; |
466891f9 | 973 | r = map_io(m, clone, map_context, 0); |
45e15720 | 974 | if (r < 0 || r == DM_MAPIO_REQUEUE) |
466891f9 | 975 | clear_mapinfo(m, map_context); |
1da177e4 LT |
976 | |
977 | return r; | |
978 | } | |
979 | ||
980 | /* | |
981 | * Take a path out of use. | |
982 | */ | |
983 | static int fail_path(struct pgpath *pgpath) | |
984 | { | |
985 | unsigned long flags; | |
986 | struct multipath *m = pgpath->pg->m; | |
987 | ||
988 | spin_lock_irqsave(&m->lock, flags); | |
989 | ||
6680073d | 990 | if (!pgpath->is_active) |
1da177e4 LT |
991 | goto out; |
992 | ||
72d94861 | 993 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
994 | |
995 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
6680073d | 996 | pgpath->is_active = 0; |
1da177e4 LT |
997 | pgpath->fail_count++; |
998 | ||
999 | m->nr_valid_paths--; | |
1000 | ||
1001 | if (pgpath == m->current_pgpath) | |
1002 | m->current_pgpath = NULL; | |
1003 | ||
b15546f9 MA |
1004 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
1005 | pgpath->path.dev->name, m->nr_valid_paths); | |
1006 | ||
fe9cf30e | 1007 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1008 | |
1009 | out: | |
1010 | spin_unlock_irqrestore(&m->lock, flags); | |
1011 | ||
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * Reinstate a previously-failed path | |
1017 | */ | |
1018 | static int reinstate_path(struct pgpath *pgpath) | |
1019 | { | |
1020 | int r = 0; | |
1021 | unsigned long flags; | |
1022 | struct multipath *m = pgpath->pg->m; | |
1023 | ||
1024 | spin_lock_irqsave(&m->lock, flags); | |
1025 | ||
6680073d | 1026 | if (pgpath->is_active) |
1da177e4 LT |
1027 | goto out; |
1028 | ||
def052d2 | 1029 | if (!pgpath->pg->ps.type->reinstate_path) { |
1da177e4 LT |
1030 | DMWARN("Reinstate path not supported by path selector %s", |
1031 | pgpath->pg->ps.type->name); | |
1032 | r = -EINVAL; | |
1033 | goto out; | |
1034 | } | |
1035 | ||
1036 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1037 | if (r) | |
1038 | goto out; | |
1039 | ||
6680073d | 1040 | pgpath->is_active = 1; |
1da177e4 | 1041 | |
e54f77dd CS |
1042 | if (!m->nr_valid_paths++ && m->queue_size) { |
1043 | m->current_pgpath = NULL; | |
c557308e | 1044 | queue_work(kmultipathd, &m->process_queued_ios); |
e54f77dd | 1045 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1046 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
e54f77dd CS |
1047 | m->pg_init_in_progress++; |
1048 | } | |
1da177e4 | 1049 | |
b15546f9 MA |
1050 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
1051 | pgpath->path.dev->name, m->nr_valid_paths); | |
1052 | ||
fe9cf30e | 1053 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1054 | |
1055 | out: | |
1056 | spin_unlock_irqrestore(&m->lock, flags); | |
1057 | ||
1058 | return r; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1063 | */ | |
1064 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1065 | action_fn action) | |
1066 | { | |
19040c0b | 1067 | int r = -EINVAL; |
1da177e4 LT |
1068 | struct pgpath *pgpath; |
1069 | struct priority_group *pg; | |
1070 | ||
1071 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1072 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1073 | if (pgpath->path.dev == dev) | |
1074 | r = action(pgpath); | |
1075 | } | |
1076 | } | |
1077 | ||
1078 | return r; | |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * Temporarily try to avoid having to use the specified PG | |
1083 | */ | |
1084 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
1085 | int bypassed) | |
1086 | { | |
1087 | unsigned long flags; | |
1088 | ||
1089 | spin_lock_irqsave(&m->lock, flags); | |
1090 | ||
1091 | pg->bypassed = bypassed; | |
1092 | m->current_pgpath = NULL; | |
1093 | m->current_pg = NULL; | |
1094 | ||
1095 | spin_unlock_irqrestore(&m->lock, flags); | |
1096 | ||
fe9cf30e | 1097 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1098 | } |
1099 | ||
1100 | /* | |
1101 | * Switch to using the specified PG from the next I/O that gets mapped | |
1102 | */ | |
1103 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1104 | { | |
1105 | struct priority_group *pg; | |
1106 | unsigned pgnum; | |
1107 | unsigned long flags; | |
31998ef1 | 1108 | char dummy; |
1da177e4 | 1109 | |
31998ef1 | 1110 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1111 | (pgnum > m->nr_priority_groups)) { |
1112 | DMWARN("invalid PG number supplied to switch_pg_num"); | |
1113 | return -EINVAL; | |
1114 | } | |
1115 | ||
1116 | spin_lock_irqsave(&m->lock, flags); | |
1117 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1118 | pg->bypassed = 0; | |
1119 | if (--pgnum) | |
1120 | continue; | |
1121 | ||
1122 | m->current_pgpath = NULL; | |
1123 | m->current_pg = NULL; | |
1124 | m->next_pg = pg; | |
1125 | } | |
1126 | spin_unlock_irqrestore(&m->lock, flags); | |
1127 | ||
fe9cf30e | 1128 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1129 | return 0; |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * Set/clear bypassed status of a PG. | |
1134 | * PGs are numbered upwards from 1 in the order they were declared. | |
1135 | */ | |
1136 | static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) | |
1137 | { | |
1138 | struct priority_group *pg; | |
1139 | unsigned pgnum; | |
31998ef1 | 1140 | char dummy; |
1da177e4 | 1141 | |
31998ef1 | 1142 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
1da177e4 LT |
1143 | (pgnum > m->nr_priority_groups)) { |
1144 | DMWARN("invalid PG number supplied to bypass_pg"); | |
1145 | return -EINVAL; | |
1146 | } | |
1147 | ||
1148 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1149 | if (!--pgnum) | |
1150 | break; | |
1151 | } | |
1152 | ||
1153 | bypass_pg(m, pg, bypassed); | |
1154 | return 0; | |
1155 | } | |
1156 | ||
c9e45581 DW |
1157 | /* |
1158 | * Should we retry pg_init immediately? | |
1159 | */ | |
1160 | static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) | |
1161 | { | |
1162 | unsigned long flags; | |
1163 | int limit_reached = 0; | |
1164 | ||
1165 | spin_lock_irqsave(&m->lock, flags); | |
1166 | ||
1167 | if (m->pg_init_count <= m->pg_init_retries) | |
1168 | m->pg_init_required = 1; | |
1169 | else | |
1170 | limit_reached = 1; | |
1171 | ||
1172 | spin_unlock_irqrestore(&m->lock, flags); | |
1173 | ||
1174 | return limit_reached; | |
1175 | } | |
1176 | ||
3ae31f6a | 1177 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1178 | { |
83c0d5d5 | 1179 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1180 | struct priority_group *pg = pgpath->pg; |
1181 | struct multipath *m = pg->m; | |
1182 | unsigned long flags; | |
4e2d19e4 | 1183 | unsigned delay_retry = 0; |
cfae5c9b CS |
1184 | |
1185 | /* device or driver problems */ | |
1186 | switch (errors) { | |
1187 | case SCSI_DH_OK: | |
1188 | break; | |
1189 | case SCSI_DH_NOSYS: | |
1190 | if (!m->hw_handler_name) { | |
1191 | errors = 0; | |
1192 | break; | |
1193 | } | |
f7b934c8 MB |
1194 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1195 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1196 | /* |
1197 | * Fail path for now, so we do not ping pong | |
1198 | */ | |
1199 | fail_path(pgpath); | |
1200 | break; | |
1201 | case SCSI_DH_DEV_TEMP_BUSY: | |
1202 | /* | |
1203 | * Probably doing something like FW upgrade on the | |
1204 | * controller so try the other pg. | |
1205 | */ | |
1206 | bypass_pg(m, pg, 1); | |
1207 | break; | |
cfae5c9b | 1208 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1209 | /* Wait before retrying. */ |
1210 | delay_retry = 1; | |
cfae5c9b CS |
1211 | case SCSI_DH_IMM_RETRY: |
1212 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1213 | if (pg_init_limit_reached(m, pgpath)) | |
1214 | fail_path(pgpath); | |
1215 | errors = 0; | |
1216 | break; | |
1217 | default: | |
1218 | /* | |
1219 | * We probably do not want to fail the path for a device | |
1220 | * error, but this is what the old dm did. In future | |
1221 | * patches we can do more advanced handling. | |
1222 | */ | |
1223 | fail_path(pgpath); | |
1224 | } | |
1225 | ||
1226 | spin_lock_irqsave(&m->lock, flags); | |
1227 | if (errors) { | |
e54f77dd CS |
1228 | if (pgpath == m->current_pgpath) { |
1229 | DMERR("Could not failover device. Error %d.", errors); | |
1230 | m->current_pgpath = NULL; | |
1231 | m->current_pg = NULL; | |
1232 | } | |
d0259bf0 | 1233 | } else if (!m->pg_init_required) |
cfae5c9b | 1234 | pg->bypassed = 0; |
cfae5c9b | 1235 | |
d0259bf0 KU |
1236 | if (--m->pg_init_in_progress) |
1237 | /* Activations of other paths are still on going */ | |
1238 | goto out; | |
1239 | ||
1240 | if (!m->pg_init_required) | |
1241 | m->queue_io = 0; | |
1242 | ||
4e2d19e4 | 1243 | m->pg_init_delay_retry = delay_retry; |
d0259bf0 KU |
1244 | queue_work(kmultipathd, &m->process_queued_ios); |
1245 | ||
2bded7bd KU |
1246 | /* |
1247 | * Wake up any thread waiting to suspend. | |
1248 | */ | |
1249 | wake_up(&m->pg_init_wait); | |
1250 | ||
d0259bf0 | 1251 | out: |
cfae5c9b CS |
1252 | spin_unlock_irqrestore(&m->lock, flags); |
1253 | } | |
1254 | ||
bab7cfc7 CS |
1255 | static void activate_path(struct work_struct *work) |
1256 | { | |
e54f77dd | 1257 | struct pgpath *pgpath = |
4e2d19e4 | 1258 | container_of(work, struct pgpath, activate_path.work); |
bab7cfc7 | 1259 | |
3ae31f6a | 1260 | scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev), |
83c0d5d5 | 1261 | pg_init_done, pgpath); |
bab7cfc7 CS |
1262 | } |
1263 | ||
1da177e4 LT |
1264 | /* |
1265 | * end_io handling | |
1266 | */ | |
f40c67f0 | 1267 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1268 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1269 | { |
f40c67f0 KU |
1270 | /* |
1271 | * We don't queue any clone request inside the multipath target | |
1272 | * during end I/O handling, since those clone requests don't have | |
1273 | * bio clones. If we queue them inside the multipath target, | |
1274 | * we need to make bio clones, that requires memory allocation. | |
1275 | * (See drivers/md/dm.c:end_clone_bio() about why the clone requests | |
1276 | * don't have bio clones.) | |
1277 | * Instead of queueing the clone request here, we queue the original | |
1278 | * request into dm core, which will remake a clone request and | |
1279 | * clone bios for it and resubmit it later. | |
1280 | */ | |
1281 | int r = DM_ENDIO_REQUEUE; | |
640eb3b0 | 1282 | unsigned long flags; |
1da177e4 | 1283 | |
f40c67f0 | 1284 | if (!error && !clone->errors) |
1da177e4 LT |
1285 | return 0; /* I/O complete */ |
1286 | ||
6f13f6fb | 1287 | if (error == -EOPNOTSUPP || error == -EREMOTEIO || error == -EILSEQ) |
959eb4e5 MS |
1288 | return error; |
1289 | ||
cfae5c9b CS |
1290 | if (mpio->pgpath) |
1291 | fail_path(mpio->pgpath); | |
1da177e4 | 1292 | |
640eb3b0 | 1293 | spin_lock_irqsave(&m->lock, flags); |
751b2a7d HR |
1294 | if (!m->nr_valid_paths) { |
1295 | if (!m->queue_if_no_path) { | |
1296 | if (!__must_push_back(m)) | |
1297 | r = -EIO; | |
1298 | } else { | |
1299 | if (error == -EBADE) | |
1300 | r = error; | |
1301 | } | |
1302 | } | |
640eb3b0 | 1303 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 1304 | |
f40c67f0 | 1305 | return r; |
1da177e4 LT |
1306 | } |
1307 | ||
f40c67f0 | 1308 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1309 | int error, union map_info *map_context) |
1310 | { | |
028867ac AK |
1311 | struct multipath *m = ti->private; |
1312 | struct dm_mpath_io *mpio = map_context->ptr; | |
a71a261f | 1313 | struct pgpath *pgpath; |
1da177e4 LT |
1314 | struct path_selector *ps; |
1315 | int r; | |
1316 | ||
466891f9 JN |
1317 | BUG_ON(!mpio); |
1318 | ||
f40c67f0 | 1319 | r = do_end_io(m, clone, error, mpio); |
a71a261f | 1320 | pgpath = mpio->pgpath; |
1da177e4 LT |
1321 | if (pgpath) { |
1322 | ps = &pgpath->pg->ps; | |
1323 | if (ps->type->end_io) | |
02ab823f | 1324 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1325 | } |
466891f9 | 1326 | clear_mapinfo(m, map_context); |
1da177e4 LT |
1327 | |
1328 | return r; | |
1329 | } | |
1330 | ||
1331 | /* | |
1332 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1333 | * the last path fails we must error any remaining I/O. |
1334 | * Note that if the freeze_bdev fails while suspending, the | |
1335 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1336 | */ |
1337 | static void multipath_presuspend(struct dm_target *ti) | |
1338 | { | |
1339 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 | 1340 | |
485ef69e | 1341 | queue_if_no_path(m, 0, 1); |
1da177e4 LT |
1342 | } |
1343 | ||
6df400ab KU |
1344 | static void multipath_postsuspend(struct dm_target *ti) |
1345 | { | |
6380f26f MA |
1346 | struct multipath *m = ti->private; |
1347 | ||
1348 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1349 | flush_multipath_work(m); |
6380f26f | 1350 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1351 | } |
1352 | ||
436d4108 AK |
1353 | /* |
1354 | * Restore the queue_if_no_path setting. | |
1355 | */ | |
1da177e4 LT |
1356 | static void multipath_resume(struct dm_target *ti) |
1357 | { | |
1358 | struct multipath *m = (struct multipath *) ti->private; | |
1359 | unsigned long flags; | |
1360 | ||
1361 | spin_lock_irqsave(&m->lock, flags); | |
436d4108 | 1362 | m->queue_if_no_path = m->saved_queue_if_no_path; |
1da177e4 LT |
1363 | spin_unlock_irqrestore(&m->lock, flags); |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Info output has the following format: | |
1368 | * num_multipath_feature_args [multipath_feature_args]* | |
1369 | * num_handler_status_args [handler_status_args]* | |
1370 | * num_groups init_group_number | |
1371 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1372 | * num_paths num_selector_args | |
1373 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1374 | * | |
1375 | * Table output has the following format (identical to the constructor string): | |
1376 | * num_feature_args [features_args]* | |
1377 | * num_handler_args hw_handler [hw_handler_args]* | |
1378 | * num_groups init_group_number | |
1379 | * [priority selector-name num_ps_args [ps_args]* | |
1380 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1381 | */ | |
fd7c092e MP |
1382 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1383 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1384 | { |
1385 | int sz = 0; | |
1386 | unsigned long flags; | |
1387 | struct multipath *m = (struct multipath *) ti->private; | |
1da177e4 LT |
1388 | struct priority_group *pg; |
1389 | struct pgpath *p; | |
1390 | unsigned pg_num; | |
1391 | char state; | |
1392 | ||
1393 | spin_lock_irqsave(&m->lock, flags); | |
1394 | ||
1395 | /* Features */ | |
1396 | if (type == STATUSTYPE_INFO) | |
c9e45581 DW |
1397 | DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count); |
1398 | else { | |
1399 | DMEMIT("%u ", m->queue_if_no_path + | |
4e2d19e4 | 1400 | (m->pg_init_retries > 0) * 2 + |
a58a935d MS |
1401 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
1402 | m->retain_attached_hw_handler); | |
c9e45581 DW |
1403 | if (m->queue_if_no_path) |
1404 | DMEMIT("queue_if_no_path "); | |
1405 | if (m->pg_init_retries) | |
1406 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1407 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1408 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
a58a935d MS |
1409 | if (m->retain_attached_hw_handler) |
1410 | DMEMIT("retain_attached_hw_handler "); | |
c9e45581 | 1411 | } |
1da177e4 | 1412 | |
cfae5c9b | 1413 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1414 | DMEMIT("0 "); |
1415 | else | |
cfae5c9b | 1416 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1417 | |
1418 | DMEMIT("%u ", m->nr_priority_groups); | |
1419 | ||
1420 | if (m->next_pg) | |
1421 | pg_num = m->next_pg->pg_num; | |
1422 | else if (m->current_pg) | |
1423 | pg_num = m->current_pg->pg_num; | |
1424 | else | |
a490a07a | 1425 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1426 | |
1427 | DMEMIT("%u ", pg_num); | |
1428 | ||
1429 | switch (type) { | |
1430 | case STATUSTYPE_INFO: | |
1431 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1432 | if (pg->bypassed) | |
1433 | state = 'D'; /* Disabled */ | |
1434 | else if (pg == m->current_pg) | |
1435 | state = 'A'; /* Currently Active */ | |
1436 | else | |
1437 | state = 'E'; /* Enabled */ | |
1438 | ||
1439 | DMEMIT("%c ", state); | |
1440 | ||
1441 | if (pg->ps.type->status) | |
1442 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1443 | result + sz, | |
1444 | maxlen - sz); | |
1445 | else | |
1446 | DMEMIT("0 "); | |
1447 | ||
1448 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1449 | pg->ps.type->info_args); | |
1450 | ||
1451 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1452 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1453 | p->is_active ? "A" : "F", |
1da177e4 LT |
1454 | p->fail_count); |
1455 | if (pg->ps.type->status) | |
1456 | sz += pg->ps.type->status(&pg->ps, | |
1457 | &p->path, type, result + sz, | |
1458 | maxlen - sz); | |
1459 | } | |
1460 | } | |
1461 | break; | |
1462 | ||
1463 | case STATUSTYPE_TABLE: | |
1464 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1465 | DMEMIT("%s ", pg->ps.type->name); | |
1466 | ||
1467 | if (pg->ps.type->status) | |
1468 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1469 | result + sz, | |
1470 | maxlen - sz); | |
1471 | else | |
1472 | DMEMIT("0 "); | |
1473 | ||
1474 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1475 | pg->ps.type->table_args); | |
1476 | ||
1477 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1478 | DMEMIT("%s ", p->path.dev->name); | |
1479 | if (pg->ps.type->status) | |
1480 | sz += pg->ps.type->status(&pg->ps, | |
1481 | &p->path, type, result + sz, | |
1482 | maxlen - sz); | |
1483 | } | |
1484 | } | |
1485 | break; | |
1486 | } | |
1487 | ||
1488 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1489 | } |
1490 | ||
1491 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1492 | { | |
6380f26f | 1493 | int r = -EINVAL; |
1da177e4 LT |
1494 | struct dm_dev *dev; |
1495 | struct multipath *m = (struct multipath *) ti->private; | |
1496 | action_fn action; | |
1497 | ||
6380f26f MA |
1498 | mutex_lock(&m->work_mutex); |
1499 | ||
c2f3d24b KU |
1500 | if (dm_suspended(ti)) { |
1501 | r = -EBUSY; | |
1502 | goto out; | |
1503 | } | |
1504 | ||
1da177e4 | 1505 | if (argc == 1) { |
498f0103 | 1506 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
6380f26f MA |
1507 | r = queue_if_no_path(m, 1, 0); |
1508 | goto out; | |
498f0103 | 1509 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
6380f26f MA |
1510 | r = queue_if_no_path(m, 0, 0); |
1511 | goto out; | |
1512 | } | |
1da177e4 LT |
1513 | } |
1514 | ||
6380f26f MA |
1515 | if (argc != 2) { |
1516 | DMWARN("Unrecognised multipath message received."); | |
1517 | goto out; | |
1518 | } | |
1da177e4 | 1519 | |
498f0103 | 1520 | if (!strcasecmp(argv[0], "disable_group")) { |
6380f26f MA |
1521 | r = bypass_pg_num(m, argv[1], 1); |
1522 | goto out; | |
498f0103 | 1523 | } else if (!strcasecmp(argv[0], "enable_group")) { |
6380f26f MA |
1524 | r = bypass_pg_num(m, argv[1], 0); |
1525 | goto out; | |
498f0103 | 1526 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1527 | r = switch_pg_num(m, argv[1]); |
1528 | goto out; | |
498f0103 | 1529 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1530 | action = reinstate_path; |
498f0103 | 1531 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1532 | action = fail_path; |
6380f26f MA |
1533 | else { |
1534 | DMWARN("Unrecognised multipath message received."); | |
1535 | goto out; | |
1536 | } | |
1da177e4 | 1537 | |
8215d6ec | 1538 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1539 | if (r) { |
72d94861 | 1540 | DMWARN("message: error getting device %s", |
1da177e4 | 1541 | argv[1]); |
6380f26f | 1542 | goto out; |
1da177e4 LT |
1543 | } |
1544 | ||
1545 | r = action_dev(m, dev, action); | |
1546 | ||
1547 | dm_put_device(ti, dev); | |
1548 | ||
6380f26f MA |
1549 | out: |
1550 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1551 | return r; |
1da177e4 LT |
1552 | } |
1553 | ||
647b3d00 | 1554 | static int multipath_ioctl(struct dm_target *ti, unsigned int cmd, |
9af4aa30 MB |
1555 | unsigned long arg) |
1556 | { | |
35991652 | 1557 | struct multipath *m = ti->private; |
7ba10aa6 | 1558 | struct pgpath *pgpath; |
35991652 MP |
1559 | struct block_device *bdev; |
1560 | fmode_t mode; | |
9af4aa30 | 1561 | unsigned long flags; |
35991652 MP |
1562 | int r; |
1563 | ||
1564 | again: | |
1565 | bdev = NULL; | |
1566 | mode = 0; | |
1567 | r = 0; | |
9af4aa30 MB |
1568 | |
1569 | spin_lock_irqsave(&m->lock, flags); | |
1570 | ||
1571 | if (!m->current_pgpath) | |
02ab823f | 1572 | __choose_pgpath(m, 0); |
9af4aa30 | 1573 | |
7ba10aa6 MS |
1574 | pgpath = m->current_pgpath; |
1575 | ||
1576 | if (pgpath) { | |
1577 | bdev = pgpath->path.dev->bdev; | |
1578 | mode = pgpath->path.dev->mode; | |
e90dae1f | 1579 | } |
9af4aa30 | 1580 | |
7ba10aa6 | 1581 | if ((pgpath && m->queue_io) || (!pgpath && m->queue_if_no_path)) |
9af4aa30 MB |
1582 | r = -EAGAIN; |
1583 | else if (!bdev) | |
1584 | r = -EIO; | |
1585 | ||
1586 | spin_unlock_irqrestore(&m->lock, flags); | |
1587 | ||
ec8013be PB |
1588 | /* |
1589 | * Only pass ioctls through if the device sizes match exactly. | |
1590 | */ | |
1591 | if (!r && ti->len != i_size_read(bdev->bd_inode) >> SECTOR_SHIFT) | |
1592 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
1593 | ||
35991652 MP |
1594 | if (r == -EAGAIN && !fatal_signal_pending(current)) { |
1595 | queue_work(kmultipathd, &m->process_queued_ios); | |
1596 | msleep(10); | |
1597 | goto again; | |
1598 | } | |
1599 | ||
633a08b8 | 1600 | return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
9af4aa30 MB |
1601 | } |
1602 | ||
af4874e0 MS |
1603 | static int multipath_iterate_devices(struct dm_target *ti, |
1604 | iterate_devices_callout_fn fn, void *data) | |
1605 | { | |
1606 | struct multipath *m = ti->private; | |
1607 | struct priority_group *pg; | |
1608 | struct pgpath *p; | |
1609 | int ret = 0; | |
1610 | ||
1611 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1612 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1613 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1614 | if (ret) |
1615 | goto out; | |
1616 | } | |
1617 | } | |
1618 | ||
1619 | out: | |
1620 | return ret; | |
1621 | } | |
1622 | ||
f40c67f0 KU |
1623 | static int __pgpath_busy(struct pgpath *pgpath) |
1624 | { | |
1625 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1626 | ||
1627 | return dm_underlying_device_busy(q); | |
1628 | } | |
1629 | ||
1630 | /* | |
1631 | * We return "busy", only when we can map I/Os but underlying devices | |
1632 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1633 | * the underlying queue). | |
1634 | * In other words, if we want to kill I/Os or queue them inside us | |
1635 | * due to map unavailability, we don't return "busy". Otherwise, | |
1636 | * dm core won't give us the I/Os and we can't do what we want. | |
1637 | */ | |
1638 | static int multipath_busy(struct dm_target *ti) | |
1639 | { | |
1640 | int busy = 0, has_active = 0; | |
1641 | struct multipath *m = ti->private; | |
1642 | struct priority_group *pg; | |
1643 | struct pgpath *pgpath; | |
1644 | unsigned long flags; | |
1645 | ||
1646 | spin_lock_irqsave(&m->lock, flags); | |
1647 | ||
1648 | /* Guess which priority_group will be used at next mapping time */ | |
1649 | if (unlikely(!m->current_pgpath && m->next_pg)) | |
1650 | pg = m->next_pg; | |
1651 | else if (likely(m->current_pg)) | |
1652 | pg = m->current_pg; | |
1653 | else | |
1654 | /* | |
1655 | * We don't know which pg will be used at next mapping time. | |
1656 | * We don't call __choose_pgpath() here to avoid to trigger | |
1657 | * pg_init just by busy checking. | |
1658 | * So we don't know whether underlying devices we will be using | |
1659 | * at next mapping time are busy or not. Just try mapping. | |
1660 | */ | |
1661 | goto out; | |
1662 | ||
1663 | /* | |
1664 | * If there is one non-busy active path at least, the path selector | |
1665 | * will be able to select it. So we consider such a pg as not busy. | |
1666 | */ | |
1667 | busy = 1; | |
1668 | list_for_each_entry(pgpath, &pg->pgpaths, list) | |
1669 | if (pgpath->is_active) { | |
1670 | has_active = 1; | |
1671 | ||
1672 | if (!__pgpath_busy(pgpath)) { | |
1673 | busy = 0; | |
1674 | break; | |
1675 | } | |
1676 | } | |
1677 | ||
1678 | if (!has_active) | |
1679 | /* | |
1680 | * No active path in this pg, so this pg won't be used and | |
1681 | * the current_pg will be changed at next mapping time. | |
1682 | * We need to try mapping to determine it. | |
1683 | */ | |
1684 | busy = 0; | |
1685 | ||
1686 | out: | |
1687 | spin_unlock_irqrestore(&m->lock, flags); | |
1688 | ||
1689 | return busy; | |
1690 | } | |
1691 | ||
1da177e4 LT |
1692 | /*----------------------------------------------------------------- |
1693 | * Module setup | |
1694 | *---------------------------------------------------------------*/ | |
1695 | static struct target_type multipath_target = { | |
1696 | .name = "multipath", | |
fd7c092e | 1697 | .version = {1, 5, 1}, |
1da177e4 LT |
1698 | .module = THIS_MODULE, |
1699 | .ctr = multipath_ctr, | |
1700 | .dtr = multipath_dtr, | |
f40c67f0 KU |
1701 | .map_rq = multipath_map, |
1702 | .rq_end_io = multipath_end_io, | |
1da177e4 | 1703 | .presuspend = multipath_presuspend, |
6df400ab | 1704 | .postsuspend = multipath_postsuspend, |
1da177e4 LT |
1705 | .resume = multipath_resume, |
1706 | .status = multipath_status, | |
1707 | .message = multipath_message, | |
9af4aa30 | 1708 | .ioctl = multipath_ioctl, |
af4874e0 | 1709 | .iterate_devices = multipath_iterate_devices, |
f40c67f0 | 1710 | .busy = multipath_busy, |
1da177e4 LT |
1711 | }; |
1712 | ||
1713 | static int __init dm_multipath_init(void) | |
1714 | { | |
1715 | int r; | |
1716 | ||
1717 | /* allocate a slab for the dm_ios */ | |
028867ac | 1718 | _mpio_cache = KMEM_CACHE(dm_mpath_io, 0); |
1da177e4 LT |
1719 | if (!_mpio_cache) |
1720 | return -ENOMEM; | |
1721 | ||
1722 | r = dm_register_target(&multipath_target); | |
1723 | if (r < 0) { | |
0cd33124 | 1724 | DMERR("register failed %d", r); |
1da177e4 LT |
1725 | kmem_cache_destroy(_mpio_cache); |
1726 | return -EINVAL; | |
1727 | } | |
1728 | ||
4d4d66ab | 1729 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1730 | if (!kmultipathd) { |
0cd33124 | 1731 | DMERR("failed to create workqueue kmpathd"); |
c557308e AK |
1732 | dm_unregister_target(&multipath_target); |
1733 | kmem_cache_destroy(_mpio_cache); | |
1734 | return -ENOMEM; | |
1735 | } | |
1736 | ||
bab7cfc7 CS |
1737 | /* |
1738 | * A separate workqueue is used to handle the device handlers | |
1739 | * to avoid overloading existing workqueue. Overloading the | |
1740 | * old workqueue would also create a bottleneck in the | |
1741 | * path of the storage hardware device activation. | |
1742 | */ | |
4d4d66ab TH |
1743 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1744 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1745 | if (!kmpath_handlerd) { |
1746 | DMERR("failed to create workqueue kmpath_handlerd"); | |
1747 | destroy_workqueue(kmultipathd); | |
1748 | dm_unregister_target(&multipath_target); | |
1749 | kmem_cache_destroy(_mpio_cache); | |
1750 | return -ENOMEM; | |
1751 | } | |
1752 | ||
72d94861 | 1753 | DMINFO("version %u.%u.%u loaded", |
1da177e4 LT |
1754 | multipath_target.version[0], multipath_target.version[1], |
1755 | multipath_target.version[2]); | |
1756 | ||
1757 | return r; | |
1758 | } | |
1759 | ||
1760 | static void __exit dm_multipath_exit(void) | |
1761 | { | |
bab7cfc7 | 1762 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
1763 | destroy_workqueue(kmultipathd); |
1764 | ||
10d3bd09 | 1765 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
1766 | kmem_cache_destroy(_mpio_cache); |
1767 | } | |
1768 | ||
1da177e4 LT |
1769 | module_init(dm_multipath_init); |
1770 | module_exit(dm_multipath_exit); | |
1771 | ||
1772 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
1773 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
1774 | MODULE_LICENSE("GPL"); |