]>
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 | ||
4cc96131 | 10 | #include "dm-rq.h" |
76e33fe4 | 11 | #include "dm-bio-record.h" |
1da177e4 | 12 | #include "dm-path-selector.h" |
b15546f9 | 13 | #include "dm-uevent.h" |
1da177e4 | 14 | |
e5863d9a | 15 | #include <linux/blkdev.h> |
1da177e4 LT |
16 | #include <linux/ctype.h> |
17 | #include <linux/init.h> | |
18 | #include <linux/mempool.h> | |
19 | #include <linux/module.h> | |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/time.h> | |
23 | #include <linux/workqueue.h> | |
35991652 | 24 | #include <linux/delay.h> |
cfae5c9b | 25 | #include <scsi/scsi_dh.h> |
60063497 | 26 | #include <linux/atomic.h> |
78ce23b5 | 27 | #include <linux/blk-mq.h> |
1da177e4 | 28 | |
72d94861 | 29 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
30 | #define DM_PG_INIT_DELAY_MSECS 2000 |
31 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
1da177e4 LT |
32 | |
33 | /* Path properties */ | |
34 | struct pgpath { | |
35 | struct list_head list; | |
36 | ||
37 | struct priority_group *pg; /* Owning PG */ | |
38 | unsigned fail_count; /* Cumulative failure count */ | |
39 | ||
c922d5f7 | 40 | struct dm_path path; |
4e2d19e4 | 41 | struct delayed_work activate_path; |
be7d31cc MS |
42 | |
43 | bool is_active:1; /* Path status */ | |
1da177e4 LT |
44 | }; |
45 | ||
46 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
47 | ||
48 | /* | |
49 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
50 | * Each has a path selector which controls which path gets used. | |
51 | */ | |
52 | struct priority_group { | |
53 | struct list_head list; | |
54 | ||
55 | struct multipath *m; /* Owning multipath instance */ | |
56 | struct path_selector ps; | |
57 | ||
58 | unsigned pg_num; /* Reference number */ | |
1da177e4 LT |
59 | unsigned nr_pgpaths; /* Number of paths in PG */ |
60 | struct list_head pgpaths; | |
be7d31cc MS |
61 | |
62 | bool bypassed:1; /* Temporarily bypass this PG? */ | |
1da177e4 LT |
63 | }; |
64 | ||
65 | /* Multipath context */ | |
66 | struct multipath { | |
67 | struct list_head list; | |
68 | struct dm_target *ti; | |
69 | ||
cfae5c9b | 70 | const char *hw_handler_name; |
2bfd2e13 | 71 | char *hw_handler_params; |
4e2d19e4 | 72 | |
1fbdd2b3 MS |
73 | spinlock_t lock; |
74 | ||
1da177e4 LT |
75 | unsigned nr_priority_groups; |
76 | struct list_head priority_groups; | |
4e2d19e4 CS |
77 | |
78 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
79 | ||
1da177e4 LT |
80 | struct pgpath *current_pgpath; |
81 | struct priority_group *current_pg; | |
82 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
1da177e4 | 83 | |
518257b1 | 84 | unsigned long flags; /* Multipath state flags */ |
1fbdd2b3 | 85 | |
c9e45581 | 86 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
4e2d19e4 | 87 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
1da177e4 | 88 | |
91e968aa MS |
89 | atomic_t nr_valid_paths; /* Total number of usable paths */ |
90 | atomic_t pg_init_in_progress; /* Only one pg_init allowed at once */ | |
91 | atomic_t pg_init_count; /* Number of times pg_init called */ | |
92 | ||
e83068a5 MS |
93 | unsigned queue_mode; |
94 | ||
6380f26f | 95 | struct mutex work_mutex; |
20800cb3 | 96 | struct work_struct trigger_event; |
76e33fe4 MS |
97 | |
98 | struct work_struct process_queued_bios; | |
99 | struct bio_list queued_bios; | |
1da177e4 LT |
100 | }; |
101 | ||
102 | /* | |
76e33fe4 | 103 | * Context information attached to each io we process. |
1da177e4 | 104 | */ |
028867ac | 105 | struct dm_mpath_io { |
1da177e4 | 106 | struct pgpath *pgpath; |
02ab823f | 107 | size_t nr_bytes; |
1da177e4 LT |
108 | }; |
109 | ||
110 | typedef int (*action_fn) (struct pgpath *pgpath); | |
111 | ||
bab7cfc7 | 112 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 | 113 | static void trigger_event(struct work_struct *work); |
89bfce76 BVA |
114 | static void activate_or_offline_path(struct pgpath *pgpath); |
115 | static void activate_path_work(struct work_struct *work); | |
76e33fe4 | 116 | static void process_queued_bios(struct work_struct *work); |
1da177e4 | 117 | |
518257b1 MS |
118 | /*----------------------------------------------- |
119 | * Multipath state flags. | |
120 | *-----------------------------------------------*/ | |
121 | ||
122 | #define MPATHF_QUEUE_IO 0 /* Must we queue all I/O? */ | |
123 | #define MPATHF_QUEUE_IF_NO_PATH 1 /* Queue I/O if last path fails? */ | |
124 | #define MPATHF_SAVED_QUEUE_IF_NO_PATH 2 /* Saved state during suspension */ | |
125 | #define MPATHF_RETAIN_ATTACHED_HW_HANDLER 3 /* If there's already a hw_handler present, don't change it. */ | |
126 | #define MPATHF_PG_INIT_DISABLED 4 /* pg_init is not currently allowed */ | |
127 | #define MPATHF_PG_INIT_REQUIRED 5 /* pg_init needs calling? */ | |
128 | #define MPATHF_PG_INIT_DELAY_RETRY 6 /* Delay pg_init retry? */ | |
1da177e4 LT |
129 | |
130 | /*----------------------------------------------- | |
131 | * Allocation routines | |
132 | *-----------------------------------------------*/ | |
133 | ||
134 | static struct pgpath *alloc_pgpath(void) | |
135 | { | |
e69fae56 | 136 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 137 | |
224cb3e9 | 138 | if (pgpath) { |
be7d31cc | 139 | pgpath->is_active = true; |
89bfce76 | 140 | INIT_DELAYED_WORK(&pgpath->activate_path, activate_path_work); |
224cb3e9 | 141 | } |
1da177e4 LT |
142 | |
143 | return pgpath; | |
144 | } | |
145 | ||
028867ac | 146 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
147 | { |
148 | kfree(pgpath); | |
149 | } | |
150 | ||
151 | static struct priority_group *alloc_priority_group(void) | |
152 | { | |
153 | struct priority_group *pg; | |
154 | ||
e69fae56 | 155 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 156 | |
e69fae56 MM |
157 | if (pg) |
158 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
159 | |
160 | return pg; | |
161 | } | |
162 | ||
163 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
164 | { | |
165 | struct pgpath *pgpath, *tmp; | |
166 | ||
167 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
168 | list_del(&pgpath->list); | |
169 | dm_put_device(ti, pgpath->path.dev); | |
170 | free_pgpath(pgpath); | |
171 | } | |
172 | } | |
173 | ||
174 | static void free_priority_group(struct priority_group *pg, | |
175 | struct dm_target *ti) | |
176 | { | |
177 | struct path_selector *ps = &pg->ps; | |
178 | ||
179 | if (ps->type) { | |
180 | ps->type->destroy(ps); | |
181 | dm_put_path_selector(ps->type); | |
182 | } | |
183 | ||
184 | free_pgpaths(&pg->pgpaths, ti); | |
185 | kfree(pg); | |
186 | } | |
187 | ||
e83068a5 | 188 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
189 | { |
190 | struct multipath *m; | |
191 | ||
e69fae56 | 192 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 193 | if (m) { |
1da177e4 LT |
194 | INIT_LIST_HEAD(&m->priority_groups); |
195 | spin_lock_init(&m->lock); | |
518257b1 | 196 | set_bit(MPATHF_QUEUE_IO, &m->flags); |
91e968aa MS |
197 | atomic_set(&m->nr_valid_paths, 0); |
198 | atomic_set(&m->pg_init_in_progress, 0); | |
199 | atomic_set(&m->pg_init_count, 0); | |
4e2d19e4 | 200 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; |
c4028958 | 201 | INIT_WORK(&m->trigger_event, trigger_event); |
2bded7bd | 202 | init_waitqueue_head(&m->pg_init_wait); |
6380f26f | 203 | mutex_init(&m->work_mutex); |
8637a6bf | 204 | |
e83068a5 | 205 | m->queue_mode = DM_TYPE_NONE; |
76e33fe4 | 206 | |
28f16c20 MM |
207 | m->ti = ti; |
208 | ti->private = m; | |
1da177e4 LT |
209 | } |
210 | ||
211 | return m; | |
212 | } | |
213 | ||
e83068a5 MS |
214 | static int alloc_multipath_stage2(struct dm_target *ti, struct multipath *m) |
215 | { | |
216 | if (m->queue_mode == DM_TYPE_NONE) { | |
217 | /* | |
218 | * Default to request-based. | |
219 | */ | |
220 | if (dm_use_blk_mq(dm_table_get_md(ti->table))) | |
221 | m->queue_mode = DM_TYPE_MQ_REQUEST_BASED; | |
222 | else | |
223 | m->queue_mode = DM_TYPE_REQUEST_BASED; | |
eb8db831 | 224 | } else if (m->queue_mode == DM_TYPE_BIO_BASED) { |
e83068a5 MS |
225 | INIT_WORK(&m->process_queued_bios, process_queued_bios); |
226 | /* | |
227 | * bio-based doesn't support any direct scsi_dh management; | |
228 | * it just discovers if a scsi_dh is attached. | |
229 | */ | |
230 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); | |
231 | } | |
232 | ||
233 | dm_table_set_type(ti->table, m->queue_mode); | |
234 | ||
235 | return 0; | |
236 | } | |
237 | ||
1da177e4 LT |
238 | static void free_multipath(struct multipath *m) |
239 | { | |
240 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
241 | |
242 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
243 | list_del(&pg->list); | |
244 | free_priority_group(pg, m->ti); | |
245 | } | |
246 | ||
cfae5c9b | 247 | kfree(m->hw_handler_name); |
2bfd2e13 | 248 | kfree(m->hw_handler_params); |
1da177e4 LT |
249 | kfree(m); |
250 | } | |
251 | ||
2eff1924 MS |
252 | static struct dm_mpath_io *get_mpio(union map_info *info) |
253 | { | |
254 | return info->ptr; | |
255 | } | |
256 | ||
bf661be1 MS |
257 | static size_t multipath_per_bio_data_size(void) |
258 | { | |
259 | return sizeof(struct dm_mpath_io) + sizeof(struct dm_bio_details); | |
260 | } | |
261 | ||
76e33fe4 MS |
262 | static struct dm_mpath_io *get_mpio_from_bio(struct bio *bio) |
263 | { | |
bf661be1 | 264 | return dm_per_bio_data(bio, multipath_per_bio_data_size()); |
76e33fe4 MS |
265 | } |
266 | ||
bf661be1 | 267 | static struct dm_bio_details *get_bio_details_from_bio(struct bio *bio) |
76e33fe4 | 268 | { |
bf661be1 | 269 | /* dm_bio_details is immediately after the dm_mpath_io in bio's per-bio-data */ |
76e33fe4 | 270 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); |
bf661be1 MS |
271 | void *bio_details = mpio + 1; |
272 | ||
273 | return bio_details; | |
274 | } | |
275 | ||
276 | static void multipath_init_per_bio_data(struct bio *bio, struct dm_mpath_io **mpio_p, | |
277 | struct dm_bio_details **bio_details_p) | |
278 | { | |
279 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); | |
280 | struct dm_bio_details *bio_details = get_bio_details_from_bio(bio); | |
76e33fe4 MS |
281 | |
282 | memset(mpio, 0, sizeof(*mpio)); | |
bf661be1 MS |
283 | memset(bio_details, 0, sizeof(*bio_details)); |
284 | dm_bio_record(bio_details, bio); | |
76e33fe4 | 285 | |
bf661be1 MS |
286 | if (mpio_p) |
287 | *mpio_p = mpio; | |
288 | if (bio_details_p) | |
289 | *bio_details_p = bio_details; | |
76e33fe4 MS |
290 | } |
291 | ||
1da177e4 LT |
292 | /*----------------------------------------------- |
293 | * Path selection | |
294 | *-----------------------------------------------*/ | |
295 | ||
3e9f1be1 | 296 | static int __pg_init_all_paths(struct multipath *m) |
fb612642 KU |
297 | { |
298 | struct pgpath *pgpath; | |
4e2d19e4 | 299 | unsigned long pg_init_delay = 0; |
fb612642 | 300 | |
91e968aa | 301 | if (atomic_read(&m->pg_init_in_progress) || test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) |
3e9f1be1 | 302 | return 0; |
17f4ff45 | 303 | |
91e968aa | 304 | atomic_inc(&m->pg_init_count); |
518257b1 | 305 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
3e9f1be1 HR |
306 | |
307 | /* Check here to reset pg_init_required */ | |
308 | if (!m->current_pg) | |
309 | return 0; | |
310 | ||
518257b1 | 311 | if (test_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags)) |
4e2d19e4 CS |
312 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? |
313 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
314 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
315 | /* Skip failed paths */ | |
316 | if (!pgpath->is_active) | |
317 | continue; | |
4e2d19e4 CS |
318 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
319 | pg_init_delay)) | |
91e968aa | 320 | atomic_inc(&m->pg_init_in_progress); |
fb612642 | 321 | } |
91e968aa | 322 | return atomic_read(&m->pg_init_in_progress); |
fb612642 KU |
323 | } |
324 | ||
4813577f | 325 | static void pg_init_all_paths(struct multipath *m) |
1da177e4 | 326 | { |
2da1610a MS |
327 | unsigned long flags; |
328 | ||
329 | spin_lock_irqsave(&m->lock, flags); | |
4813577f | 330 | __pg_init_all_paths(m); |
2da1610a | 331 | spin_unlock_irqrestore(&m->lock, flags); |
2da1610a MS |
332 | } |
333 | ||
334 | static void __switch_pg(struct multipath *m, struct priority_group *pg) | |
335 | { | |
336 | m->current_pg = pg; | |
1da177e4 LT |
337 | |
338 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 339 | if (m->hw_handler_name) { |
518257b1 MS |
340 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
341 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 342 | } else { |
518257b1 MS |
343 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
344 | clear_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 345 | } |
c9e45581 | 346 | |
91e968aa | 347 | atomic_set(&m->pg_init_count, 0); |
1da177e4 LT |
348 | } |
349 | ||
2da1610a MS |
350 | static struct pgpath *choose_path_in_pg(struct multipath *m, |
351 | struct priority_group *pg, | |
352 | size_t nr_bytes) | |
1da177e4 | 353 | { |
2da1610a | 354 | unsigned long flags; |
c922d5f7 | 355 | struct dm_path *path; |
2da1610a | 356 | struct pgpath *pgpath; |
1da177e4 | 357 | |
90a4323c | 358 | path = pg->ps.type->select_path(&pg->ps, nr_bytes); |
1da177e4 | 359 | if (!path) |
2da1610a | 360 | return ERR_PTR(-ENXIO); |
1da177e4 | 361 | |
2da1610a | 362 | pgpath = path_to_pgpath(path); |
1da177e4 | 363 | |
2da1610a MS |
364 | if (unlikely(lockless_dereference(m->current_pg) != pg)) { |
365 | /* Only update current_pgpath if pg changed */ | |
366 | spin_lock_irqsave(&m->lock, flags); | |
367 | m->current_pgpath = pgpath; | |
368 | __switch_pg(m, pg); | |
369 | spin_unlock_irqrestore(&m->lock, flags); | |
370 | } | |
1da177e4 | 371 | |
2da1610a | 372 | return pgpath; |
1da177e4 LT |
373 | } |
374 | ||
2da1610a | 375 | static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 | 376 | { |
2da1610a | 377 | unsigned long flags; |
1da177e4 | 378 | struct priority_group *pg; |
2da1610a | 379 | struct pgpath *pgpath; |
d19a55cc | 380 | unsigned bypassed = 1; |
1da177e4 | 381 | |
91e968aa | 382 | if (!atomic_read(&m->nr_valid_paths)) { |
518257b1 | 383 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
1da177e4 | 384 | goto failed; |
1f271972 | 385 | } |
1da177e4 LT |
386 | |
387 | /* Were we instructed to switch PG? */ | |
2da1610a MS |
388 | if (lockless_dereference(m->next_pg)) { |
389 | spin_lock_irqsave(&m->lock, flags); | |
1da177e4 | 390 | pg = m->next_pg; |
2da1610a MS |
391 | if (!pg) { |
392 | spin_unlock_irqrestore(&m->lock, flags); | |
393 | goto check_current_pg; | |
394 | } | |
1da177e4 | 395 | m->next_pg = NULL; |
2da1610a MS |
396 | spin_unlock_irqrestore(&m->lock, flags); |
397 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
398 | if (!IS_ERR_OR_NULL(pgpath)) | |
399 | return pgpath; | |
1da177e4 LT |
400 | } |
401 | ||
402 | /* Don't change PG until it has no remaining paths */ | |
2da1610a MS |
403 | check_current_pg: |
404 | pg = lockless_dereference(m->current_pg); | |
405 | if (pg) { | |
406 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
407 | if (!IS_ERR_OR_NULL(pgpath)) | |
408 | return pgpath; | |
409 | } | |
1da177e4 LT |
410 | |
411 | /* | |
412 | * Loop through priority groups until we find a valid path. | |
413 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
414 | * Second time we only try the ones we skipped, but set |
415 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
416 | */ |
417 | do { | |
418 | list_for_each_entry(pg, &m->priority_groups, list) { | |
d19a55cc | 419 | if (pg->bypassed == !!bypassed) |
1da177e4 | 420 | continue; |
2da1610a MS |
421 | pgpath = choose_path_in_pg(m, pg, nr_bytes); |
422 | if (!IS_ERR_OR_NULL(pgpath)) { | |
f220fd4e | 423 | if (!bypassed) |
518257b1 | 424 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); |
2da1610a | 425 | return pgpath; |
f220fd4e | 426 | } |
1da177e4 LT |
427 | } |
428 | } while (bypassed--); | |
429 | ||
430 | failed: | |
2da1610a | 431 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
432 | m->current_pgpath = NULL; |
433 | m->current_pg = NULL; | |
2da1610a MS |
434 | spin_unlock_irqrestore(&m->lock, flags); |
435 | ||
436 | return NULL; | |
1da177e4 LT |
437 | } |
438 | ||
45e15720 KU |
439 | /* |
440 | * Check whether bios must be queued in the device-mapper core rather | |
441 | * than here in the target. | |
442 | * | |
45e15720 KU |
443 | * If m->queue_if_no_path and m->saved_queue_if_no_path hold the |
444 | * same value then we are not between multipath_presuspend() | |
445 | * and multipath_resume() calls and we have no need to check | |
446 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
447 | */ | |
76e33fe4 MS |
448 | static bool __must_push_back(struct multipath *m) |
449 | { | |
450 | return ((test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) != | |
451 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)) && | |
452 | dm_noflush_suspending(m->ti)); | |
453 | } | |
454 | ||
455 | static bool must_push_back_rq(struct multipath *m) | |
45e15720 | 456 | { |
1814f2e3 MS |
457 | bool r; |
458 | unsigned long flags; | |
459 | ||
460 | spin_lock_irqsave(&m->lock, flags); | |
461 | r = (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) || | |
462 | __must_push_back(m)); | |
463 | spin_unlock_irqrestore(&m->lock, flags); | |
464 | ||
465 | return r; | |
76e33fe4 MS |
466 | } |
467 | ||
468 | static bool must_push_back_bio(struct multipath *m) | |
469 | { | |
1814f2e3 MS |
470 | bool r; |
471 | unsigned long flags; | |
472 | ||
473 | spin_lock_irqsave(&m->lock, flags); | |
474 | r = __must_push_back(m); | |
475 | spin_unlock_irqrestore(&m->lock, flags); | |
476 | ||
477 | return r; | |
45e15720 KU |
478 | } |
479 | ||
36fcffcc | 480 | /* |
76e33fe4 | 481 | * Map cloned requests (request-based multipath) |
36fcffcc | 482 | */ |
eb8db831 CH |
483 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, |
484 | union map_info *map_context, | |
485 | struct request **__clone) | |
1da177e4 | 486 | { |
7943bd6d | 487 | struct multipath *m = ti->private; |
eb8db831 | 488 | size_t nr_bytes = blk_rq_bytes(rq); |
1da177e4 | 489 | struct pgpath *pgpath; |
f40c67f0 | 490 | struct block_device *bdev; |
eb8db831 | 491 | struct dm_mpath_io *mpio = get_mpio(map_context); |
7083abbb | 492 | struct request_queue *q; |
eb8db831 | 493 | struct request *clone; |
1da177e4 | 494 | |
1da177e4 | 495 | /* Do we need to select a new pgpath? */ |
2da1610a MS |
496 | pgpath = lockless_dereference(m->current_pgpath); |
497 | if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) | |
498 | pgpath = choose_pgpath(m, nr_bytes); | |
1da177e4 | 499 | |
9bf59a61 | 500 | if (!pgpath) { |
b88efd43 MS |
501 | if (must_push_back_rq(m)) |
502 | return DM_MAPIO_DELAY_REQUEUE; | |
503 | return -EIO; /* Failed */ | |
518257b1 MS |
504 | } else if (test_bit(MPATHF_QUEUE_IO, &m->flags) || |
505 | test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { | |
2da1610a | 506 | pg_init_all_paths(m); |
06eb061f | 507 | return DM_MAPIO_REQUEUE; |
9bf59a61 | 508 | } |
6afbc01d | 509 | |
eb8db831 | 510 | memset(mpio, 0, sizeof(*mpio)); |
2eb6e1e3 KB |
511 | mpio->pgpath = pgpath; |
512 | mpio->nr_bytes = nr_bytes; | |
513 | ||
9bf59a61 | 514 | bdev = pgpath->path.dev->bdev; |
7083abbb BVA |
515 | q = bdev_get_queue(bdev); |
516 | clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, GFP_ATOMIC); | |
eb8db831 CH |
517 | if (IS_ERR(clone)) { |
518 | /* EBUSY, ENODEV or EWOULDBLOCK: requeue */ | |
7083abbb BVA |
519 | bool queue_dying = blk_queue_dying(q); |
520 | DMERR_LIMIT("blk_get_request() returned %ld%s - requeuing", | |
521 | PTR_ERR(clone), queue_dying ? " (path offline)" : ""); | |
522 | if (queue_dying) { | |
523 | atomic_inc(&m->pg_init_in_progress); | |
524 | activate_or_offline_path(pgpath); | |
525 | return DM_MAPIO_REQUEUE; | |
526 | } | |
06eb061f | 527 | return DM_MAPIO_DELAY_REQUEUE; |
e5863d9a | 528 | } |
eb8db831 CH |
529 | clone->bio = clone->biotail = NULL; |
530 | clone->rq_disk = bdev->bd_disk; | |
531 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
532 | *__clone = clone; | |
e5863d9a | 533 | |
9bf59a61 MS |
534 | if (pgpath->pg->ps.type->start_io) |
535 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
536 | &pgpath->path, | |
537 | nr_bytes); | |
2eb6e1e3 | 538 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
539 | } |
540 | ||
e5863d9a MS |
541 | static void multipath_release_clone(struct request *clone) |
542 | { | |
eb8db831 | 543 | blk_put_request(clone); |
e5863d9a MS |
544 | } |
545 | ||
76e33fe4 MS |
546 | /* |
547 | * Map cloned bios (bio-based multipath) | |
548 | */ | |
549 | static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_mpath_io *mpio) | |
550 | { | |
551 | size_t nr_bytes = bio->bi_iter.bi_size; | |
552 | struct pgpath *pgpath; | |
553 | unsigned long flags; | |
554 | bool queue_io; | |
555 | ||
556 | /* Do we need to select a new pgpath? */ | |
557 | pgpath = lockless_dereference(m->current_pgpath); | |
558 | queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags); | |
559 | if (!pgpath || !queue_io) | |
560 | pgpath = choose_pgpath(m, nr_bytes); | |
561 | ||
562 | if ((pgpath && queue_io) || | |
563 | (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) { | |
564 | /* Queue for the daemon to resubmit */ | |
565 | spin_lock_irqsave(&m->lock, flags); | |
566 | bio_list_add(&m->queued_bios, bio); | |
567 | spin_unlock_irqrestore(&m->lock, flags); | |
568 | /* PG_INIT_REQUIRED cannot be set without QUEUE_IO */ | |
569 | if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) | |
570 | pg_init_all_paths(m); | |
571 | else if (!queue_io) | |
572 | queue_work(kmultipathd, &m->process_queued_bios); | |
573 | return DM_MAPIO_SUBMITTED; | |
574 | } | |
575 | ||
576 | if (!pgpath) { | |
577 | if (!must_push_back_bio(m)) | |
578 | return -EIO; | |
579 | return DM_MAPIO_REQUEUE; | |
580 | } | |
581 | ||
582 | mpio->pgpath = pgpath; | |
583 | mpio->nr_bytes = nr_bytes; | |
584 | ||
585 | bio->bi_error = 0; | |
586 | bio->bi_bdev = pgpath->path.dev->bdev; | |
1eff9d32 | 587 | bio->bi_opf |= REQ_FAILFAST_TRANSPORT; |
76e33fe4 MS |
588 | |
589 | if (pgpath->pg->ps.type->start_io) | |
590 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
591 | &pgpath->path, | |
592 | nr_bytes); | |
593 | return DM_MAPIO_REMAPPED; | |
594 | } | |
595 | ||
596 | static int multipath_map_bio(struct dm_target *ti, struct bio *bio) | |
597 | { | |
598 | struct multipath *m = ti->private; | |
bf661be1 MS |
599 | struct dm_mpath_io *mpio = NULL; |
600 | ||
601 | multipath_init_per_bio_data(bio, &mpio, NULL); | |
76e33fe4 MS |
602 | |
603 | return __multipath_map_bio(m, bio, mpio); | |
604 | } | |
605 | ||
7e48c768 | 606 | static void process_queued_io_list(struct multipath *m) |
76e33fe4 | 607 | { |
7e48c768 MS |
608 | if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED) |
609 | dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table)); | |
610 | else if (m->queue_mode == DM_TYPE_BIO_BASED) | |
76e33fe4 MS |
611 | queue_work(kmultipathd, &m->process_queued_bios); |
612 | } | |
613 | ||
614 | static void process_queued_bios(struct work_struct *work) | |
615 | { | |
616 | int r; | |
617 | unsigned long flags; | |
618 | struct bio *bio; | |
619 | struct bio_list bios; | |
620 | struct blk_plug plug; | |
621 | struct multipath *m = | |
622 | container_of(work, struct multipath, process_queued_bios); | |
623 | ||
624 | bio_list_init(&bios); | |
625 | ||
626 | spin_lock_irqsave(&m->lock, flags); | |
627 | ||
628 | if (bio_list_empty(&m->queued_bios)) { | |
629 | spin_unlock_irqrestore(&m->lock, flags); | |
630 | return; | |
631 | } | |
632 | ||
633 | bio_list_merge(&bios, &m->queued_bios); | |
634 | bio_list_init(&m->queued_bios); | |
635 | ||
636 | spin_unlock_irqrestore(&m->lock, flags); | |
637 | ||
638 | blk_start_plug(&plug); | |
639 | while ((bio = bio_list_pop(&bios))) { | |
640 | r = __multipath_map_bio(m, bio, get_mpio_from_bio(bio)); | |
641 | if (r < 0 || r == DM_MAPIO_REQUEUE) { | |
642 | bio->bi_error = r; | |
643 | bio_endio(bio); | |
644 | } else if (r == DM_MAPIO_REMAPPED) | |
645 | generic_make_request(bio); | |
646 | } | |
647 | blk_finish_plug(&plug); | |
648 | } | |
649 | ||
1da177e4 LT |
650 | /* |
651 | * If we run out of usable paths, should we queue I/O or error it? | |
652 | */ | |
be7d31cc MS |
653 | static int queue_if_no_path(struct multipath *m, bool queue_if_no_path, |
654 | bool save_old_value) | |
1da177e4 LT |
655 | { |
656 | unsigned long flags; | |
657 | ||
658 | spin_lock_irqsave(&m->lock, flags); | |
659 | ||
518257b1 MS |
660 | if (save_old_value) { |
661 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) | |
662 | set_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
663 | else | |
664 | clear_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
665 | } else { | |
666 | if (queue_if_no_path) | |
667 | set_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
668 | else | |
669 | clear_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
670 | } | |
671 | if (queue_if_no_path) | |
672 | set_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
485ef69e | 673 | else |
518257b1 MS |
674 | clear_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags); |
675 | ||
1da177e4 LT |
676 | spin_unlock_irqrestore(&m->lock, flags); |
677 | ||
76e33fe4 | 678 | if (!queue_if_no_path) { |
63d832c3 | 679 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 680 | process_queued_io_list(m); |
76e33fe4 | 681 | } |
63d832c3 | 682 | |
1da177e4 LT |
683 | return 0; |
684 | } | |
685 | ||
1da177e4 LT |
686 | /* |
687 | * An event is triggered whenever a path is taken out of use. | |
688 | * Includes path failure and PG bypass. | |
689 | */ | |
c4028958 | 690 | static void trigger_event(struct work_struct *work) |
1da177e4 | 691 | { |
c4028958 DH |
692 | struct multipath *m = |
693 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
694 | |
695 | dm_table_event(m->ti->table); | |
696 | } | |
697 | ||
698 | /*----------------------------------------------------------------- | |
699 | * Constructor/argument parsing: | |
700 | * <#multipath feature args> [<arg>]* | |
701 | * <#hw_handler args> [hw_handler [<arg>]*] | |
702 | * <#priority groups> | |
703 | * <initial priority group> | |
704 | * [<selector> <#selector args> [<arg>]* | |
705 | * <#paths> <#per-path selector args> | |
706 | * [<path> [<arg>]* ]+ ]+ | |
707 | *---------------------------------------------------------------*/ | |
498f0103 | 708 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
709 | struct dm_target *ti) |
710 | { | |
711 | int r; | |
712 | struct path_selector_type *pst; | |
713 | unsigned ps_argc; | |
714 | ||
498f0103 | 715 | static struct dm_arg _args[] = { |
72d94861 | 716 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
717 | }; |
718 | ||
498f0103 | 719 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 720 | if (!pst) { |
72d94861 | 721 | ti->error = "unknown path selector type"; |
1da177e4 LT |
722 | return -EINVAL; |
723 | } | |
724 | ||
498f0103 | 725 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
726 | if (r) { |
727 | dm_put_path_selector(pst); | |
1da177e4 | 728 | return -EINVAL; |
371b2e34 | 729 | } |
1da177e4 LT |
730 | |
731 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
732 | if (r) { | |
733 | dm_put_path_selector(pst); | |
72d94861 | 734 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
735 | return r; |
736 | } | |
737 | ||
738 | pg->ps.type = pst; | |
498f0103 | 739 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
740 | |
741 | return 0; | |
742 | } | |
743 | ||
498f0103 | 744 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
745 | struct dm_target *ti) |
746 | { | |
747 | int r; | |
748 | struct pgpath *p; | |
ae11b1b3 | 749 | struct multipath *m = ti->private; |
a58a935d MS |
750 | struct request_queue *q = NULL; |
751 | const char *attached_handler_name; | |
1da177e4 LT |
752 | |
753 | /* we need at least a path arg */ | |
754 | if (as->argc < 1) { | |
72d94861 | 755 | ti->error = "no device given"; |
01460f35 | 756 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
757 | } |
758 | ||
759 | p = alloc_pgpath(); | |
760 | if (!p) | |
01460f35 | 761 | return ERR_PTR(-ENOMEM); |
1da177e4 | 762 | |
498f0103 | 763 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 764 | &p->path.dev); |
1da177e4 | 765 | if (r) { |
72d94861 | 766 | ti->error = "error getting device"; |
1da177e4 LT |
767 | goto bad; |
768 | } | |
769 | ||
518257b1 | 770 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) || m->hw_handler_name) |
a58a935d MS |
771 | q = bdev_get_queue(p->path.dev->bdev); |
772 | ||
518257b1 | 773 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) { |
1bab0de0 | 774 | retain: |
a58a935d MS |
775 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); |
776 | if (attached_handler_name) { | |
54cd640d | 777 | /* |
778 | * Clear any hw_handler_params associated with a | |
779 | * handler that isn't already attached. | |
780 | */ | |
781 | if (m->hw_handler_name && strcmp(attached_handler_name, m->hw_handler_name)) { | |
782 | kfree(m->hw_handler_params); | |
783 | m->hw_handler_params = NULL; | |
784 | } | |
785 | ||
a58a935d MS |
786 | /* |
787 | * Reset hw_handler_name to match the attached handler | |
a58a935d MS |
788 | * |
789 | * NB. This modifies the table line to show the actual | |
790 | * handler instead of the original table passed in. | |
791 | */ | |
792 | kfree(m->hw_handler_name); | |
793 | m->hw_handler_name = attached_handler_name; | |
a58a935d MS |
794 | } |
795 | } | |
a0cf7ea9 | 796 | |
a58a935d | 797 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
798 | r = scsi_dh_attach(q, m->hw_handler_name); |
799 | if (r == -EBUSY) { | |
1bab0de0 | 800 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 801 | |
1bab0de0 CH |
802 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
803 | bdevname(p->path.dev->bdev, b)); | |
804 | goto retain; | |
805 | } | |
ae11b1b3 | 806 | if (r < 0) { |
a0cf7ea9 | 807 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
808 | dm_put_device(ti, p->path.dev); |
809 | goto bad; | |
810 | } | |
2bfd2e13 CS |
811 | |
812 | if (m->hw_handler_params) { | |
813 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
814 | if (r < 0) { | |
815 | ti->error = "unable to set hardware " | |
816 | "handler parameters"; | |
2bfd2e13 CS |
817 | dm_put_device(ti, p->path.dev); |
818 | goto bad; | |
819 | } | |
820 | } | |
ae11b1b3 HR |
821 | } |
822 | ||
1da177e4 LT |
823 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
824 | if (r) { | |
825 | dm_put_device(ti, p->path.dev); | |
826 | goto bad; | |
827 | } | |
828 | ||
829 | return p; | |
830 | ||
831 | bad: | |
832 | free_pgpath(p); | |
01460f35 | 833 | return ERR_PTR(r); |
1da177e4 LT |
834 | } |
835 | ||
498f0103 | 836 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 837 | struct multipath *m) |
1da177e4 | 838 | { |
498f0103 | 839 | static struct dm_arg _args[] = { |
72d94861 AK |
840 | {1, 1024, "invalid number of paths"}, |
841 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
842 | }; |
843 | ||
844 | int r; | |
498f0103 | 845 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 846 | struct priority_group *pg; |
28f16c20 | 847 | struct dm_target *ti = m->ti; |
1da177e4 LT |
848 | |
849 | if (as->argc < 2) { | |
850 | as->argc = 0; | |
01460f35 BM |
851 | ti->error = "not enough priority group arguments"; |
852 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
853 | } |
854 | ||
855 | pg = alloc_priority_group(); | |
856 | if (!pg) { | |
72d94861 | 857 | ti->error = "couldn't allocate priority group"; |
01460f35 | 858 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
859 | } |
860 | pg->m = m; | |
861 | ||
862 | r = parse_path_selector(as, pg, ti); | |
863 | if (r) | |
864 | goto bad; | |
865 | ||
866 | /* | |
867 | * read the paths | |
868 | */ | |
498f0103 | 869 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
870 | if (r) |
871 | goto bad; | |
872 | ||
498f0103 | 873 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
874 | if (r) |
875 | goto bad; | |
876 | ||
498f0103 | 877 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
878 | for (i = 0; i < pg->nr_pgpaths; i++) { |
879 | struct pgpath *pgpath; | |
498f0103 | 880 | struct dm_arg_set path_args; |
1da177e4 | 881 | |
498f0103 | 882 | if (as->argc < nr_args) { |
148acff6 | 883 | ti->error = "not enough path parameters"; |
6bbf79a1 | 884 | r = -EINVAL; |
1da177e4 | 885 | goto bad; |
148acff6 | 886 | } |
1da177e4 | 887 | |
498f0103 | 888 | path_args.argc = nr_args; |
1da177e4 LT |
889 | path_args.argv = as->argv; |
890 | ||
891 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
892 | if (IS_ERR(pgpath)) { |
893 | r = PTR_ERR(pgpath); | |
1da177e4 | 894 | goto bad; |
01460f35 | 895 | } |
1da177e4 LT |
896 | |
897 | pgpath->pg = pg; | |
898 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 899 | dm_consume_args(as, nr_args); |
1da177e4 LT |
900 | } |
901 | ||
902 | return pg; | |
903 | ||
904 | bad: | |
905 | free_priority_group(pg, ti); | |
01460f35 | 906 | return ERR_PTR(r); |
1da177e4 LT |
907 | } |
908 | ||
498f0103 | 909 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 910 | { |
1da177e4 | 911 | unsigned hw_argc; |
2bfd2e13 | 912 | int ret; |
28f16c20 | 913 | struct dm_target *ti = m->ti; |
1da177e4 | 914 | |
498f0103 | 915 | static struct dm_arg _args[] = { |
72d94861 | 916 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
917 | }; |
918 | ||
498f0103 | 919 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
920 | return -EINVAL; |
921 | ||
922 | if (!hw_argc) | |
923 | return 0; | |
924 | ||
e83068a5 | 925 | if (m->queue_mode == DM_TYPE_BIO_BASED) { |
76e33fe4 MS |
926 | dm_consume_args(as, hw_argc); |
927 | DMERR("bio-based multipath doesn't allow hardware handler args"); | |
928 | return 0; | |
929 | } | |
930 | ||
498f0103 | 931 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
f97dc421 | 932 | if (!m->hw_handler_name) |
933 | return -EINVAL; | |
14e98c5c | 934 | |
2bfd2e13 CS |
935 | if (hw_argc > 1) { |
936 | char *p; | |
937 | int i, j, len = 4; | |
938 | ||
939 | for (i = 0; i <= hw_argc - 2; i++) | |
940 | len += strlen(as->argv[i]) + 1; | |
941 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
942 | if (!p) { | |
943 | ti->error = "memory allocation failed"; | |
944 | ret = -ENOMEM; | |
945 | goto fail; | |
946 | } | |
947 | j = sprintf(p, "%d", hw_argc - 1); | |
948 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
949 | j = sprintf(p, "%s", as->argv[i]); | |
950 | } | |
498f0103 | 951 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
952 | |
953 | return 0; | |
2bfd2e13 CS |
954 | fail: |
955 | kfree(m->hw_handler_name); | |
956 | m->hw_handler_name = NULL; | |
957 | return ret; | |
1da177e4 LT |
958 | } |
959 | ||
498f0103 | 960 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
961 | { |
962 | int r; | |
963 | unsigned argc; | |
28f16c20 | 964 | struct dm_target *ti = m->ti; |
498f0103 | 965 | const char *arg_name; |
1da177e4 | 966 | |
498f0103 | 967 | static struct dm_arg _args[] = { |
e83068a5 | 968 | {0, 8, "invalid number of feature args"}, |
c9e45581 | 969 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 970 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
971 | }; |
972 | ||
498f0103 | 973 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
974 | if (r) |
975 | return -EINVAL; | |
976 | ||
977 | if (!argc) | |
978 | return 0; | |
979 | ||
c9e45581 | 980 | do { |
498f0103 | 981 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
982 | argc--; |
983 | ||
498f0103 | 984 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
be7d31cc | 985 | r = queue_if_no_path(m, true, false); |
c9e45581 DW |
986 | continue; |
987 | } | |
988 | ||
a58a935d | 989 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
518257b1 | 990 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); |
a58a935d MS |
991 | continue; |
992 | } | |
993 | ||
498f0103 | 994 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 995 | (argc >= 1)) { |
498f0103 | 996 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
997 | argc--; |
998 | continue; | |
999 | } | |
1000 | ||
498f0103 | 1001 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 1002 | (argc >= 1)) { |
498f0103 | 1003 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
1004 | argc--; |
1005 | continue; | |
1006 | } | |
1007 | ||
e83068a5 MS |
1008 | if (!strcasecmp(arg_name, "queue_mode") && |
1009 | (argc >= 1)) { | |
1010 | const char *queue_mode_name = dm_shift_arg(as); | |
1011 | ||
1012 | if (!strcasecmp(queue_mode_name, "bio")) | |
1013 | m->queue_mode = DM_TYPE_BIO_BASED; | |
1014 | else if (!strcasecmp(queue_mode_name, "rq")) | |
1015 | m->queue_mode = DM_TYPE_REQUEST_BASED; | |
1016 | else if (!strcasecmp(queue_mode_name, "mq")) | |
1017 | m->queue_mode = DM_TYPE_MQ_REQUEST_BASED; | |
1018 | else { | |
1019 | ti->error = "Unknown 'queue_mode' requested"; | |
1020 | r = -EINVAL; | |
1021 | } | |
1022 | argc--; | |
1023 | continue; | |
1024 | } | |
1025 | ||
1da177e4 | 1026 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
1027 | r = -EINVAL; |
1028 | } while (argc && !r); | |
1029 | ||
1030 | return r; | |
1da177e4 LT |
1031 | } |
1032 | ||
e83068a5 | 1033 | static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv) |
1da177e4 | 1034 | { |
498f0103 MS |
1035 | /* target arguments */ |
1036 | static struct dm_arg _args[] = { | |
a490a07a MS |
1037 | {0, 1024, "invalid number of priority groups"}, |
1038 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
1039 | }; |
1040 | ||
1041 | int r; | |
1042 | struct multipath *m; | |
498f0103 | 1043 | struct dm_arg_set as; |
1da177e4 LT |
1044 | unsigned pg_count = 0; |
1045 | unsigned next_pg_num; | |
1046 | ||
1047 | as.argc = argc; | |
1048 | as.argv = argv; | |
1049 | ||
e83068a5 | 1050 | m = alloc_multipath(ti); |
1da177e4 | 1051 | if (!m) { |
72d94861 | 1052 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
1053 | return -EINVAL; |
1054 | } | |
1055 | ||
28f16c20 | 1056 | r = parse_features(&as, m); |
1da177e4 LT |
1057 | if (r) |
1058 | goto bad; | |
1059 | ||
e83068a5 MS |
1060 | r = alloc_multipath_stage2(ti, m); |
1061 | if (r) | |
1062 | goto bad; | |
1063 | ||
28f16c20 | 1064 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
1065 | if (r) |
1066 | goto bad; | |
1067 | ||
498f0103 | 1068 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
1069 | if (r) |
1070 | goto bad; | |
1071 | ||
498f0103 | 1072 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
1073 | if (r) |
1074 | goto bad; | |
1075 | ||
a490a07a MS |
1076 | if ((!m->nr_priority_groups && next_pg_num) || |
1077 | (m->nr_priority_groups && !next_pg_num)) { | |
1078 | ti->error = "invalid initial priority group"; | |
1079 | r = -EINVAL; | |
1080 | goto bad; | |
1081 | } | |
1082 | ||
1da177e4 LT |
1083 | /* parse the priority groups */ |
1084 | while (as.argc) { | |
1085 | struct priority_group *pg; | |
91e968aa | 1086 | unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths); |
1da177e4 | 1087 | |
28f16c20 | 1088 | pg = parse_priority_group(&as, m); |
01460f35 BM |
1089 | if (IS_ERR(pg)) { |
1090 | r = PTR_ERR(pg); | |
1da177e4 LT |
1091 | goto bad; |
1092 | } | |
1093 | ||
91e968aa MS |
1094 | nr_valid_paths += pg->nr_pgpaths; |
1095 | atomic_set(&m->nr_valid_paths, nr_valid_paths); | |
1096 | ||
1da177e4 LT |
1097 | list_add_tail(&pg->list, &m->priority_groups); |
1098 | pg_count++; | |
1099 | pg->pg_num = pg_count; | |
1100 | if (!--next_pg_num) | |
1101 | m->next_pg = pg; | |
1102 | } | |
1103 | ||
1104 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 1105 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
1106 | r = -EINVAL; |
1107 | goto bad; | |
1108 | } | |
1109 | ||
55a62eef AK |
1110 | ti->num_flush_bios = 1; |
1111 | ti->num_discard_bios = 1; | |
042bcef8 | 1112 | ti->num_write_same_bios = 1; |
e83068a5 | 1113 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
bf661be1 | 1114 | ti->per_io_data_size = multipath_per_bio_data_size(); |
eb8db831 | 1115 | else |
8637a6bf | 1116 | ti->per_io_data_size = sizeof(struct dm_mpath_io); |
8627921f | 1117 | |
1da177e4 LT |
1118 | return 0; |
1119 | ||
1120 | bad: | |
1121 | free_multipath(m); | |
1122 | return r; | |
1123 | } | |
1124 | ||
2bded7bd KU |
1125 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
1126 | { | |
9f4c3f87 | 1127 | DEFINE_WAIT(wait); |
2bded7bd KU |
1128 | |
1129 | while (1) { | |
9f4c3f87 | 1130 | prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE); |
2bded7bd | 1131 | |
91e968aa | 1132 | if (!atomic_read(&m->pg_init_in_progress)) |
2bded7bd | 1133 | break; |
2bded7bd KU |
1134 | |
1135 | io_schedule(); | |
1136 | } | |
9f4c3f87 | 1137 | finish_wait(&m->pg_init_wait, &wait); |
2bded7bd KU |
1138 | } |
1139 | ||
1140 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 1141 | { |
518257b1 MS |
1142 | set_bit(MPATHF_PG_INIT_DISABLED, &m->flags); |
1143 | smp_mb__after_atomic(); | |
954a73d5 | 1144 | |
bab7cfc7 | 1145 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 1146 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 1147 | flush_workqueue(kmultipathd); |
43829731 | 1148 | flush_work(&m->trigger_event); |
954a73d5 | 1149 | |
518257b1 MS |
1150 | clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags); |
1151 | smp_mb__after_atomic(); | |
6df400ab KU |
1152 | } |
1153 | ||
1154 | static void multipath_dtr(struct dm_target *ti) | |
1155 | { | |
1156 | struct multipath *m = ti->private; | |
1157 | ||
2bded7bd | 1158 | flush_multipath_work(m); |
1da177e4 LT |
1159 | free_multipath(m); |
1160 | } | |
1161 | ||
1da177e4 LT |
1162 | /* |
1163 | * Take a path out of use. | |
1164 | */ | |
1165 | static int fail_path(struct pgpath *pgpath) | |
1166 | { | |
1167 | unsigned long flags; | |
1168 | struct multipath *m = pgpath->pg->m; | |
1169 | ||
1170 | spin_lock_irqsave(&m->lock, flags); | |
1171 | ||
6680073d | 1172 | if (!pgpath->is_active) |
1da177e4 LT |
1173 | goto out; |
1174 | ||
72d94861 | 1175 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1176 | |
1177 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
be7d31cc | 1178 | pgpath->is_active = false; |
1da177e4 LT |
1179 | pgpath->fail_count++; |
1180 | ||
91e968aa | 1181 | atomic_dec(&m->nr_valid_paths); |
1da177e4 LT |
1182 | |
1183 | if (pgpath == m->current_pgpath) | |
1184 | m->current_pgpath = NULL; | |
1185 | ||
b15546f9 | 1186 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
91e968aa | 1187 | pgpath->path.dev->name, atomic_read(&m->nr_valid_paths)); |
b15546f9 | 1188 | |
fe9cf30e | 1189 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1190 | |
1191 | out: | |
1192 | spin_unlock_irqrestore(&m->lock, flags); | |
1193 | ||
1194 | return 0; | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * Reinstate a previously-failed path | |
1199 | */ | |
1200 | static int reinstate_path(struct pgpath *pgpath) | |
1201 | { | |
63d832c3 | 1202 | int r = 0, run_queue = 0; |
1da177e4 LT |
1203 | unsigned long flags; |
1204 | struct multipath *m = pgpath->pg->m; | |
91e968aa | 1205 | unsigned nr_valid_paths; |
1da177e4 LT |
1206 | |
1207 | spin_lock_irqsave(&m->lock, flags); | |
1208 | ||
6680073d | 1209 | if (pgpath->is_active) |
1da177e4 LT |
1210 | goto out; |
1211 | ||
ec31f3f7 | 1212 | DMWARN("Reinstating path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1213 | |
1214 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1215 | if (r) | |
1216 | goto out; | |
1217 | ||
be7d31cc | 1218 | pgpath->is_active = true; |
1da177e4 | 1219 | |
91e968aa MS |
1220 | nr_valid_paths = atomic_inc_return(&m->nr_valid_paths); |
1221 | if (nr_valid_paths == 1) { | |
e54f77dd | 1222 | m->current_pgpath = NULL; |
63d832c3 | 1223 | run_queue = 1; |
e54f77dd | 1224 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1225 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
91e968aa | 1226 | atomic_inc(&m->pg_init_in_progress); |
e54f77dd | 1227 | } |
1da177e4 | 1228 | |
b15546f9 | 1229 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
91e968aa | 1230 | pgpath->path.dev->name, nr_valid_paths); |
b15546f9 | 1231 | |
fe9cf30e | 1232 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1233 | |
1234 | out: | |
1235 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 1236 | if (run_queue) { |
63d832c3 | 1237 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1238 | process_queued_io_list(m); |
76e33fe4 | 1239 | } |
1da177e4 LT |
1240 | |
1241 | return r; | |
1242 | } | |
1243 | ||
1244 | /* | |
1245 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1246 | */ | |
1247 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1248 | action_fn action) | |
1249 | { | |
19040c0b | 1250 | int r = -EINVAL; |
1da177e4 LT |
1251 | struct pgpath *pgpath; |
1252 | struct priority_group *pg; | |
1253 | ||
1254 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1255 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1256 | if (pgpath->path.dev == dev) | |
1257 | r = action(pgpath); | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | return r; | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * Temporarily try to avoid having to use the specified PG | |
1266 | */ | |
1267 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
be7d31cc | 1268 | bool bypassed) |
1da177e4 LT |
1269 | { |
1270 | unsigned long flags; | |
1271 | ||
1272 | spin_lock_irqsave(&m->lock, flags); | |
1273 | ||
1274 | pg->bypassed = bypassed; | |
1275 | m->current_pgpath = NULL; | |
1276 | m->current_pg = NULL; | |
1277 | ||
1278 | spin_unlock_irqrestore(&m->lock, flags); | |
1279 | ||
fe9cf30e | 1280 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1281 | } |
1282 | ||
1283 | /* | |
1284 | * Switch to using the specified PG from the next I/O that gets mapped | |
1285 | */ | |
1286 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1287 | { | |
1288 | struct priority_group *pg; | |
1289 | unsigned pgnum; | |
1290 | unsigned long flags; | |
31998ef1 | 1291 | char dummy; |
1da177e4 | 1292 | |
31998ef1 | 1293 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1294 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1295 | DMWARN("invalid PG number supplied to switch_pg_num"); |
1296 | return -EINVAL; | |
1297 | } | |
1298 | ||
1299 | spin_lock_irqsave(&m->lock, flags); | |
1300 | list_for_each_entry(pg, &m->priority_groups, list) { | |
be7d31cc | 1301 | pg->bypassed = false; |
1da177e4 LT |
1302 | if (--pgnum) |
1303 | continue; | |
1304 | ||
1305 | m->current_pgpath = NULL; | |
1306 | m->current_pg = NULL; | |
1307 | m->next_pg = pg; | |
1308 | } | |
1309 | spin_unlock_irqrestore(&m->lock, flags); | |
1310 | ||
fe9cf30e | 1311 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1312 | return 0; |
1313 | } | |
1314 | ||
1315 | /* | |
1316 | * Set/clear bypassed status of a PG. | |
1317 | * PGs are numbered upwards from 1 in the order they were declared. | |
1318 | */ | |
be7d31cc | 1319 | static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed) |
1da177e4 LT |
1320 | { |
1321 | struct priority_group *pg; | |
1322 | unsigned pgnum; | |
31998ef1 | 1323 | char dummy; |
1da177e4 | 1324 | |
31998ef1 | 1325 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1326 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1327 | DMWARN("invalid PG number supplied to bypass_pg"); |
1328 | return -EINVAL; | |
1329 | } | |
1330 | ||
1331 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1332 | if (!--pgnum) | |
1333 | break; | |
1334 | } | |
1335 | ||
1336 | bypass_pg(m, pg, bypassed); | |
1337 | return 0; | |
1338 | } | |
1339 | ||
c9e45581 DW |
1340 | /* |
1341 | * Should we retry pg_init immediately? | |
1342 | */ | |
be7d31cc | 1343 | static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) |
c9e45581 DW |
1344 | { |
1345 | unsigned long flags; | |
be7d31cc | 1346 | bool limit_reached = false; |
c9e45581 DW |
1347 | |
1348 | spin_lock_irqsave(&m->lock, flags); | |
1349 | ||
91e968aa MS |
1350 | if (atomic_read(&m->pg_init_count) <= m->pg_init_retries && |
1351 | !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) | |
518257b1 | 1352 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
c9e45581 | 1353 | else |
be7d31cc | 1354 | limit_reached = true; |
c9e45581 DW |
1355 | |
1356 | spin_unlock_irqrestore(&m->lock, flags); | |
1357 | ||
1358 | return limit_reached; | |
1359 | } | |
1360 | ||
3ae31f6a | 1361 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1362 | { |
83c0d5d5 | 1363 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1364 | struct priority_group *pg = pgpath->pg; |
1365 | struct multipath *m = pg->m; | |
1366 | unsigned long flags; | |
be7d31cc | 1367 | bool delay_retry = false; |
cfae5c9b CS |
1368 | |
1369 | /* device or driver problems */ | |
1370 | switch (errors) { | |
1371 | case SCSI_DH_OK: | |
1372 | break; | |
1373 | case SCSI_DH_NOSYS: | |
1374 | if (!m->hw_handler_name) { | |
1375 | errors = 0; | |
1376 | break; | |
1377 | } | |
f7b934c8 MB |
1378 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1379 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1380 | /* |
1381 | * Fail path for now, so we do not ping pong | |
1382 | */ | |
1383 | fail_path(pgpath); | |
1384 | break; | |
1385 | case SCSI_DH_DEV_TEMP_BUSY: | |
1386 | /* | |
1387 | * Probably doing something like FW upgrade on the | |
1388 | * controller so try the other pg. | |
1389 | */ | |
be7d31cc | 1390 | bypass_pg(m, pg, true); |
cfae5c9b | 1391 | break; |
cfae5c9b | 1392 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1393 | /* Wait before retrying. */ |
1394 | delay_retry = 1; | |
cfae5c9b CS |
1395 | case SCSI_DH_IMM_RETRY: |
1396 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1397 | if (pg_init_limit_reached(m, pgpath)) | |
1398 | fail_path(pgpath); | |
1399 | errors = 0; | |
1400 | break; | |
ec31f3f7 | 1401 | case SCSI_DH_DEV_OFFLINED: |
cfae5c9b CS |
1402 | default: |
1403 | /* | |
1404 | * We probably do not want to fail the path for a device | |
1405 | * error, but this is what the old dm did. In future | |
1406 | * patches we can do more advanced handling. | |
1407 | */ | |
1408 | fail_path(pgpath); | |
1409 | } | |
1410 | ||
1411 | spin_lock_irqsave(&m->lock, flags); | |
1412 | if (errors) { | |
e54f77dd CS |
1413 | if (pgpath == m->current_pgpath) { |
1414 | DMERR("Could not failover device. Error %d.", errors); | |
1415 | m->current_pgpath = NULL; | |
1416 | m->current_pg = NULL; | |
1417 | } | |
518257b1 | 1418 | } else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
be7d31cc | 1419 | pg->bypassed = false; |
cfae5c9b | 1420 | |
91e968aa | 1421 | if (atomic_dec_return(&m->pg_init_in_progress) > 0) |
d0259bf0 KU |
1422 | /* Activations of other paths are still on going */ |
1423 | goto out; | |
1424 | ||
518257b1 MS |
1425 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { |
1426 | if (delay_retry) | |
1427 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1428 | else | |
1429 | clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1430 | ||
3e9f1be1 HR |
1431 | if (__pg_init_all_paths(m)) |
1432 | goto out; | |
1433 | } | |
518257b1 | 1434 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
d0259bf0 | 1435 | |
7e48c768 | 1436 | process_queued_io_list(m); |
76e33fe4 | 1437 | |
2bded7bd KU |
1438 | /* |
1439 | * Wake up any thread waiting to suspend. | |
1440 | */ | |
1441 | wake_up(&m->pg_init_wait); | |
1442 | ||
d0259bf0 | 1443 | out: |
cfae5c9b CS |
1444 | spin_unlock_irqrestore(&m->lock, flags); |
1445 | } | |
1446 | ||
89bfce76 | 1447 | static void activate_or_offline_path(struct pgpath *pgpath) |
bab7cfc7 | 1448 | { |
f10e06b7 | 1449 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); |
bab7cfc7 | 1450 | |
f10e06b7 MS |
1451 | if (pgpath->is_active && !blk_queue_dying(q)) |
1452 | scsi_dh_activate(q, pg_init_done, pgpath); | |
3a017509 HR |
1453 | else |
1454 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1455 | } |
1456 | ||
89bfce76 BVA |
1457 | static void activate_path_work(struct work_struct *work) |
1458 | { | |
1459 | struct pgpath *pgpath = | |
1460 | container_of(work, struct pgpath, activate_path.work); | |
1461 | ||
1462 | activate_or_offline_path(pgpath); | |
1463 | } | |
1464 | ||
7e782af5 HR |
1465 | static int noretry_error(int error) |
1466 | { | |
1467 | switch (error) { | |
8ff232c1 HR |
1468 | case -EBADE: |
1469 | /* | |
1470 | * EBADE signals an reservation conflict. | |
1471 | * We shouldn't fail the path here as we can communicate with | |
1472 | * the target. We should failover to the next path, but in | |
1473 | * doing so we might be causing a ping-pong between paths. | |
1474 | * So just return the reservation conflict error. | |
1475 | */ | |
7e782af5 HR |
1476 | case -EOPNOTSUPP: |
1477 | case -EREMOTEIO: | |
1478 | case -EILSEQ: | |
1479 | case -ENODATA: | |
cc9d3c38 | 1480 | case -ENOSPC: |
7e782af5 HR |
1481 | return 1; |
1482 | } | |
1483 | ||
1484 | /* Anything else could be a path failure, so should be retried */ | |
1485 | return 0; | |
1486 | } | |
1487 | ||
1da177e4 LT |
1488 | /* |
1489 | * end_io handling | |
1490 | */ | |
f40c67f0 | 1491 | static int do_end_io(struct multipath *m, struct request *clone, |
028867ac | 1492 | int error, struct dm_mpath_io *mpio) |
1da177e4 | 1493 | { |
f40c67f0 KU |
1494 | /* |
1495 | * We don't queue any clone request inside the multipath target | |
1496 | * during end I/O handling, since those clone requests don't have | |
1497 | * bio clones. If we queue them inside the multipath target, | |
1498 | * we need to make bio clones, that requires memory allocation. | |
4cc96131 | 1499 | * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests |
f40c67f0 KU |
1500 | * don't have bio clones.) |
1501 | * Instead of queueing the clone request here, we queue the original | |
1502 | * request into dm core, which will remake a clone request and | |
1503 | * clone bios for it and resubmit it later. | |
1504 | */ | |
1505 | int r = DM_ENDIO_REQUEUE; | |
1da177e4 | 1506 | |
f40c67f0 | 1507 | if (!error && !clone->errors) |
1da177e4 LT |
1508 | return 0; /* I/O complete */ |
1509 | ||
7eee4ae2 | 1510 | if (noretry_error(error)) |
959eb4e5 MS |
1511 | return error; |
1512 | ||
cfae5c9b CS |
1513 | if (mpio->pgpath) |
1514 | fail_path(mpio->pgpath); | |
1da177e4 | 1515 | |
91e968aa | 1516 | if (!atomic_read(&m->nr_valid_paths)) { |
518257b1 | 1517 | if (!test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { |
76e33fe4 | 1518 | if (!must_push_back_rq(m)) |
751b2a7d | 1519 | r = -EIO; |
751b2a7d HR |
1520 | } |
1521 | } | |
1da177e4 | 1522 | |
f40c67f0 | 1523 | return r; |
1da177e4 LT |
1524 | } |
1525 | ||
f40c67f0 | 1526 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
1da177e4 LT |
1527 | int error, union map_info *map_context) |
1528 | { | |
028867ac | 1529 | struct multipath *m = ti->private; |
2eff1924 | 1530 | struct dm_mpath_io *mpio = get_mpio(map_context); |
a71a261f | 1531 | struct pgpath *pgpath; |
1da177e4 LT |
1532 | struct path_selector *ps; |
1533 | int r; | |
1534 | ||
466891f9 JN |
1535 | BUG_ON(!mpio); |
1536 | ||
2eff1924 | 1537 | r = do_end_io(m, clone, error, mpio); |
a71a261f | 1538 | pgpath = mpio->pgpath; |
1da177e4 LT |
1539 | if (pgpath) { |
1540 | ps = &pgpath->pg->ps; | |
1541 | if (ps->type->end_io) | |
02ab823f | 1542 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1543 | } |
1da177e4 LT |
1544 | |
1545 | return r; | |
1546 | } | |
1547 | ||
76e33fe4 MS |
1548 | static int do_end_io_bio(struct multipath *m, struct bio *clone, |
1549 | int error, struct dm_mpath_io *mpio) | |
1550 | { | |
1551 | unsigned long flags; | |
1552 | ||
1553 | if (!error) | |
1554 | return 0; /* I/O complete */ | |
1555 | ||
1556 | if (noretry_error(error)) | |
1557 | return error; | |
1558 | ||
1559 | if (mpio->pgpath) | |
1560 | fail_path(mpio->pgpath); | |
1561 | ||
1562 | if (!atomic_read(&m->nr_valid_paths)) { | |
1563 | if (!test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
1564 | if (!must_push_back_bio(m)) | |
1565 | return -EIO; | |
1566 | return DM_ENDIO_REQUEUE; | |
76e33fe4 MS |
1567 | } |
1568 | } | |
1569 | ||
1570 | /* Queue for the daemon to resubmit */ | |
bf661be1 | 1571 | dm_bio_restore(get_bio_details_from_bio(clone), clone); |
76e33fe4 MS |
1572 | |
1573 | spin_lock_irqsave(&m->lock, flags); | |
1574 | bio_list_add(&m->queued_bios, clone); | |
1575 | spin_unlock_irqrestore(&m->lock, flags); | |
1576 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) | |
1577 | queue_work(kmultipathd, &m->process_queued_bios); | |
1578 | ||
1579 | return DM_ENDIO_INCOMPLETE; | |
1580 | } | |
1581 | ||
1582 | static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, int error) | |
1583 | { | |
1584 | struct multipath *m = ti->private; | |
1585 | struct dm_mpath_io *mpio = get_mpio_from_bio(clone); | |
1586 | struct pgpath *pgpath; | |
1587 | struct path_selector *ps; | |
1588 | int r; | |
1589 | ||
1590 | BUG_ON(!mpio); | |
1591 | ||
1592 | r = do_end_io_bio(m, clone, error, mpio); | |
1593 | pgpath = mpio->pgpath; | |
1594 | if (pgpath) { | |
1595 | ps = &pgpath->pg->ps; | |
1596 | if (ps->type->end_io) | |
1597 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); | |
1598 | } | |
1599 | ||
1600 | return r; | |
1601 | } | |
1602 | ||
1da177e4 LT |
1603 | /* |
1604 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1605 | * the last path fails we must error any remaining I/O. |
1606 | * Note that if the freeze_bdev fails while suspending, the | |
1607 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1608 | */ |
1609 | static void multipath_presuspend(struct dm_target *ti) | |
1610 | { | |
7943bd6d | 1611 | struct multipath *m = ti->private; |
1da177e4 | 1612 | |
be7d31cc | 1613 | queue_if_no_path(m, false, true); |
1da177e4 LT |
1614 | } |
1615 | ||
6df400ab KU |
1616 | static void multipath_postsuspend(struct dm_target *ti) |
1617 | { | |
6380f26f MA |
1618 | struct multipath *m = ti->private; |
1619 | ||
1620 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1621 | flush_multipath_work(m); |
6380f26f | 1622 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1623 | } |
1624 | ||
436d4108 AK |
1625 | /* |
1626 | * Restore the queue_if_no_path setting. | |
1627 | */ | |
1da177e4 LT |
1628 | static void multipath_resume(struct dm_target *ti) |
1629 | { | |
7943bd6d | 1630 | struct multipath *m = ti->private; |
1814f2e3 | 1631 | unsigned long flags; |
1da177e4 | 1632 | |
1814f2e3 | 1633 | spin_lock_irqsave(&m->lock, flags); |
518257b1 MS |
1634 | if (test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)) |
1635 | set_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
1636 | else | |
1637 | clear_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
1814f2e3 | 1638 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1639 | } |
1640 | ||
1641 | /* | |
1642 | * Info output has the following format: | |
1643 | * num_multipath_feature_args [multipath_feature_args]* | |
1644 | * num_handler_status_args [handler_status_args]* | |
1645 | * num_groups init_group_number | |
1646 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1647 | * num_paths num_selector_args | |
1648 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1649 | * | |
1650 | * Table output has the following format (identical to the constructor string): | |
1651 | * num_feature_args [features_args]* | |
1652 | * num_handler_args hw_handler [hw_handler_args]* | |
1653 | * num_groups init_group_number | |
1654 | * [priority selector-name num_ps_args [ps_args]* | |
1655 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1656 | */ | |
fd7c092e MP |
1657 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1658 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1659 | { |
1660 | int sz = 0; | |
1661 | unsigned long flags; | |
7943bd6d | 1662 | struct multipath *m = ti->private; |
1da177e4 LT |
1663 | struct priority_group *pg; |
1664 | struct pgpath *p; | |
1665 | unsigned pg_num; | |
1666 | char state; | |
1667 | ||
1668 | spin_lock_irqsave(&m->lock, flags); | |
1669 | ||
1670 | /* Features */ | |
1671 | if (type == STATUSTYPE_INFO) | |
91e968aa MS |
1672 | DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags), |
1673 | atomic_read(&m->pg_init_count)); | |
c9e45581 | 1674 | else { |
518257b1 | 1675 | DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) + |
4e2d19e4 | 1676 | (m->pg_init_retries > 0) * 2 + |
a58a935d | 1677 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
e83068a5 MS |
1678 | test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) + |
1679 | (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2); | |
1680 | ||
518257b1 | 1681 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
c9e45581 DW |
1682 | DMEMIT("queue_if_no_path "); |
1683 | if (m->pg_init_retries) | |
1684 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1685 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1686 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
518257b1 | 1687 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) |
a58a935d | 1688 | DMEMIT("retain_attached_hw_handler "); |
e83068a5 MS |
1689 | if (m->queue_mode != DM_TYPE_REQUEST_BASED) { |
1690 | switch(m->queue_mode) { | |
1691 | case DM_TYPE_BIO_BASED: | |
1692 | DMEMIT("queue_mode bio "); | |
1693 | break; | |
1694 | case DM_TYPE_MQ_REQUEST_BASED: | |
1695 | DMEMIT("queue_mode mq "); | |
1696 | break; | |
1697 | } | |
1698 | } | |
c9e45581 | 1699 | } |
1da177e4 | 1700 | |
cfae5c9b | 1701 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1702 | DMEMIT("0 "); |
1703 | else | |
cfae5c9b | 1704 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1705 | |
1706 | DMEMIT("%u ", m->nr_priority_groups); | |
1707 | ||
1708 | if (m->next_pg) | |
1709 | pg_num = m->next_pg->pg_num; | |
1710 | else if (m->current_pg) | |
1711 | pg_num = m->current_pg->pg_num; | |
1712 | else | |
a490a07a | 1713 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1714 | |
1715 | DMEMIT("%u ", pg_num); | |
1716 | ||
1717 | switch (type) { | |
1718 | case STATUSTYPE_INFO: | |
1719 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1720 | if (pg->bypassed) | |
1721 | state = 'D'; /* Disabled */ | |
1722 | else if (pg == m->current_pg) | |
1723 | state = 'A'; /* Currently Active */ | |
1724 | else | |
1725 | state = 'E'; /* Enabled */ | |
1726 | ||
1727 | DMEMIT("%c ", state); | |
1728 | ||
1729 | if (pg->ps.type->status) | |
1730 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1731 | result + sz, | |
1732 | maxlen - sz); | |
1733 | else | |
1734 | DMEMIT("0 "); | |
1735 | ||
1736 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1737 | pg->ps.type->info_args); | |
1738 | ||
1739 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1740 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1741 | p->is_active ? "A" : "F", |
1da177e4 LT |
1742 | p->fail_count); |
1743 | if (pg->ps.type->status) | |
1744 | sz += pg->ps.type->status(&pg->ps, | |
1745 | &p->path, type, result + sz, | |
1746 | maxlen - sz); | |
1747 | } | |
1748 | } | |
1749 | break; | |
1750 | ||
1751 | case STATUSTYPE_TABLE: | |
1752 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1753 | DMEMIT("%s ", pg->ps.type->name); | |
1754 | ||
1755 | if (pg->ps.type->status) | |
1756 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1757 | result + sz, | |
1758 | maxlen - sz); | |
1759 | else | |
1760 | DMEMIT("0 "); | |
1761 | ||
1762 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1763 | pg->ps.type->table_args); | |
1764 | ||
1765 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1766 | DMEMIT("%s ", p->path.dev->name); | |
1767 | if (pg->ps.type->status) | |
1768 | sz += pg->ps.type->status(&pg->ps, | |
1769 | &p->path, type, result + sz, | |
1770 | maxlen - sz); | |
1771 | } | |
1772 | } | |
1773 | break; | |
1774 | } | |
1775 | ||
1776 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1777 | } |
1778 | ||
1779 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1780 | { | |
6380f26f | 1781 | int r = -EINVAL; |
1da177e4 | 1782 | struct dm_dev *dev; |
7943bd6d | 1783 | struct multipath *m = ti->private; |
1da177e4 LT |
1784 | action_fn action; |
1785 | ||
6380f26f MA |
1786 | mutex_lock(&m->work_mutex); |
1787 | ||
c2f3d24b KU |
1788 | if (dm_suspended(ti)) { |
1789 | r = -EBUSY; | |
1790 | goto out; | |
1791 | } | |
1792 | ||
1da177e4 | 1793 | if (argc == 1) { |
498f0103 | 1794 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
be7d31cc | 1795 | r = queue_if_no_path(m, true, false); |
6380f26f | 1796 | goto out; |
498f0103 | 1797 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
be7d31cc | 1798 | r = queue_if_no_path(m, false, false); |
6380f26f MA |
1799 | goto out; |
1800 | } | |
1da177e4 LT |
1801 | } |
1802 | ||
6380f26f | 1803 | if (argc != 2) { |
a356e426 | 1804 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1805 | goto out; |
1806 | } | |
1da177e4 | 1807 | |
498f0103 | 1808 | if (!strcasecmp(argv[0], "disable_group")) { |
be7d31cc | 1809 | r = bypass_pg_num(m, argv[1], true); |
6380f26f | 1810 | goto out; |
498f0103 | 1811 | } else if (!strcasecmp(argv[0], "enable_group")) { |
be7d31cc | 1812 | r = bypass_pg_num(m, argv[1], false); |
6380f26f | 1813 | goto out; |
498f0103 | 1814 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1815 | r = switch_pg_num(m, argv[1]); |
1816 | goto out; | |
498f0103 | 1817 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1818 | action = reinstate_path; |
498f0103 | 1819 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1820 | action = fail_path; |
6380f26f | 1821 | else { |
a356e426 | 1822 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1823 | goto out; |
1824 | } | |
1da177e4 | 1825 | |
8215d6ec | 1826 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1827 | if (r) { |
72d94861 | 1828 | DMWARN("message: error getting device %s", |
1da177e4 | 1829 | argv[1]); |
6380f26f | 1830 | goto out; |
1da177e4 LT |
1831 | } |
1832 | ||
1833 | r = action_dev(m, dev, action); | |
1834 | ||
1835 | dm_put_device(ti, dev); | |
1836 | ||
6380f26f MA |
1837 | out: |
1838 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1839 | return r; |
1da177e4 LT |
1840 | } |
1841 | ||
e56f81e0 CH |
1842 | static int multipath_prepare_ioctl(struct dm_target *ti, |
1843 | struct block_device **bdev, fmode_t *mode) | |
9af4aa30 | 1844 | { |
35991652 | 1845 | struct multipath *m = ti->private; |
2da1610a | 1846 | struct pgpath *current_pgpath; |
35991652 MP |
1847 | int r; |
1848 | ||
2da1610a MS |
1849 | current_pgpath = lockless_dereference(m->current_pgpath); |
1850 | if (!current_pgpath) | |
1851 | current_pgpath = choose_pgpath(m, 0); | |
9af4aa30 | 1852 | |
2da1610a | 1853 | if (current_pgpath) { |
518257b1 | 1854 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) { |
2da1610a MS |
1855 | *bdev = current_pgpath->path.dev->bdev; |
1856 | *mode = current_pgpath->path.dev->mode; | |
43e43c9e JN |
1857 | r = 0; |
1858 | } else { | |
1859 | /* pg_init has not started or completed */ | |
1860 | r = -ENOTCONN; | |
1861 | } | |
1862 | } else { | |
1863 | /* No path is available */ | |
518257b1 | 1864 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
43e43c9e JN |
1865 | r = -ENOTCONN; |
1866 | else | |
1867 | r = -EIO; | |
e90dae1f | 1868 | } |
9af4aa30 | 1869 | |
5bbbfdf6 | 1870 | if (r == -ENOTCONN) { |
2da1610a | 1871 | if (!lockless_dereference(m->current_pg)) { |
3e9f1be1 | 1872 | /* Path status changed, redo selection */ |
2da1610a | 1873 | (void) choose_pgpath(m, 0); |
3e9f1be1 | 1874 | } |
518257b1 | 1875 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
2da1610a | 1876 | pg_init_all_paths(m); |
63d832c3 | 1877 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1878 | process_queued_io_list(m); |
3e9f1be1 | 1879 | } |
35991652 | 1880 | |
e56f81e0 CH |
1881 | /* |
1882 | * Only pass ioctls through if the device sizes match exactly. | |
1883 | */ | |
1884 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
1885 | return 1; | |
1886 | return r; | |
9af4aa30 MB |
1887 | } |
1888 | ||
af4874e0 MS |
1889 | static int multipath_iterate_devices(struct dm_target *ti, |
1890 | iterate_devices_callout_fn fn, void *data) | |
1891 | { | |
1892 | struct multipath *m = ti->private; | |
1893 | struct priority_group *pg; | |
1894 | struct pgpath *p; | |
1895 | int ret = 0; | |
1896 | ||
1897 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1898 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1899 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1900 | if (ret) |
1901 | goto out; | |
1902 | } | |
1903 | } | |
1904 | ||
1905 | out: | |
1906 | return ret; | |
1907 | } | |
1908 | ||
9f54cec5 | 1909 | static int pgpath_busy(struct pgpath *pgpath) |
f40c67f0 KU |
1910 | { |
1911 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1912 | ||
52b09914 | 1913 | return blk_lld_busy(q); |
f40c67f0 KU |
1914 | } |
1915 | ||
1916 | /* | |
1917 | * We return "busy", only when we can map I/Os but underlying devices | |
1918 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1919 | * the underlying queue). | |
1920 | * In other words, if we want to kill I/Os or queue them inside us | |
1921 | * due to map unavailability, we don't return "busy". Otherwise, | |
1922 | * dm core won't give us the I/Os and we can't do what we want. | |
1923 | */ | |
1924 | static int multipath_busy(struct dm_target *ti) | |
1925 | { | |
be7d31cc | 1926 | bool busy = false, has_active = false; |
f40c67f0 | 1927 | struct multipath *m = ti->private; |
2da1610a | 1928 | struct priority_group *pg, *next_pg; |
f40c67f0 | 1929 | struct pgpath *pgpath; |
f40c67f0 | 1930 | |
b88efd43 MS |
1931 | /* pg_init in progress */ |
1932 | if (atomic_read(&m->pg_init_in_progress)) | |
2da1610a MS |
1933 | return true; |
1934 | ||
b88efd43 MS |
1935 | /* no paths available, for blk-mq: rely on IO mapping to delay requeue */ |
1936 | if (!atomic_read(&m->nr_valid_paths) && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) | |
1937 | return (m->queue_mode != DM_TYPE_MQ_REQUEST_BASED); | |
1938 | ||
f40c67f0 | 1939 | /* Guess which priority_group will be used at next mapping time */ |
2da1610a MS |
1940 | pg = lockless_dereference(m->current_pg); |
1941 | next_pg = lockless_dereference(m->next_pg); | |
1942 | if (unlikely(!lockless_dereference(m->current_pgpath) && next_pg)) | |
1943 | pg = next_pg; | |
1944 | ||
1945 | if (!pg) { | |
f40c67f0 KU |
1946 | /* |
1947 | * We don't know which pg will be used at next mapping time. | |
2da1610a | 1948 | * We don't call choose_pgpath() here to avoid to trigger |
f40c67f0 KU |
1949 | * pg_init just by busy checking. |
1950 | * So we don't know whether underlying devices we will be using | |
1951 | * at next mapping time are busy or not. Just try mapping. | |
1952 | */ | |
2da1610a MS |
1953 | return busy; |
1954 | } | |
f40c67f0 KU |
1955 | |
1956 | /* | |
1957 | * If there is one non-busy active path at least, the path selector | |
1958 | * will be able to select it. So we consider such a pg as not busy. | |
1959 | */ | |
be7d31cc | 1960 | busy = true; |
2da1610a | 1961 | list_for_each_entry(pgpath, &pg->pgpaths, list) { |
f40c67f0 | 1962 | if (pgpath->is_active) { |
be7d31cc | 1963 | has_active = true; |
9f54cec5 | 1964 | if (!pgpath_busy(pgpath)) { |
be7d31cc | 1965 | busy = false; |
f40c67f0 KU |
1966 | break; |
1967 | } | |
1968 | } | |
2da1610a | 1969 | } |
f40c67f0 | 1970 | |
2da1610a | 1971 | if (!has_active) { |
f40c67f0 KU |
1972 | /* |
1973 | * No active path in this pg, so this pg won't be used and | |
1974 | * the current_pg will be changed at next mapping time. | |
1975 | * We need to try mapping to determine it. | |
1976 | */ | |
be7d31cc | 1977 | busy = false; |
2da1610a | 1978 | } |
f40c67f0 KU |
1979 | |
1980 | return busy; | |
1981 | } | |
1982 | ||
1da177e4 LT |
1983 | /*----------------------------------------------------------------- |
1984 | * Module setup | |
1985 | *---------------------------------------------------------------*/ | |
1986 | static struct target_type multipath_target = { | |
1987 | .name = "multipath", | |
e83068a5 | 1988 | .version = {1, 12, 0}, |
16f12266 | 1989 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE, |
1da177e4 LT |
1990 | .module = THIS_MODULE, |
1991 | .ctr = multipath_ctr, | |
1992 | .dtr = multipath_dtr, | |
e5863d9a MS |
1993 | .clone_and_map_rq = multipath_clone_and_map, |
1994 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 1995 | .rq_end_io = multipath_end_io, |
76e33fe4 MS |
1996 | .map = multipath_map_bio, |
1997 | .end_io = multipath_end_io_bio, | |
1998 | .presuspend = multipath_presuspend, | |
1999 | .postsuspend = multipath_postsuspend, | |
2000 | .resume = multipath_resume, | |
2001 | .status = multipath_status, | |
2002 | .message = multipath_message, | |
2003 | .prepare_ioctl = multipath_prepare_ioctl, | |
2004 | .iterate_devices = multipath_iterate_devices, | |
2005 | .busy = multipath_busy, | |
2006 | }; | |
2007 | ||
1da177e4 LT |
2008 | static int __init dm_multipath_init(void) |
2009 | { | |
2010 | int r; | |
2011 | ||
1da177e4 LT |
2012 | r = dm_register_target(&multipath_target); |
2013 | if (r < 0) { | |
76e33fe4 | 2014 | DMERR("request-based register failed %d", r); |
ff658e9c JT |
2015 | r = -EINVAL; |
2016 | goto bad_register_target; | |
1da177e4 LT |
2017 | } |
2018 | ||
4d4d66ab | 2019 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 2020 | if (!kmultipathd) { |
0cd33124 | 2021 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
2022 | r = -ENOMEM; |
2023 | goto bad_alloc_kmultipathd; | |
c557308e AK |
2024 | } |
2025 | ||
bab7cfc7 CS |
2026 | /* |
2027 | * A separate workqueue is used to handle the device handlers | |
2028 | * to avoid overloading existing workqueue. Overloading the | |
2029 | * old workqueue would also create a bottleneck in the | |
2030 | * path of the storage hardware device activation. | |
2031 | */ | |
4d4d66ab TH |
2032 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
2033 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
2034 | if (!kmpath_handlerd) { |
2035 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
2036 | r = -ENOMEM; |
2037 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
2038 | } |
2039 | ||
ff658e9c JT |
2040 | return 0; |
2041 | ||
2042 | bad_alloc_kmpath_handlerd: | |
2043 | destroy_workqueue(kmultipathd); | |
2044 | bad_alloc_kmultipathd: | |
2045 | dm_unregister_target(&multipath_target); | |
2046 | bad_register_target: | |
1da177e4 LT |
2047 | return r; |
2048 | } | |
2049 | ||
2050 | static void __exit dm_multipath_exit(void) | |
2051 | { | |
bab7cfc7 | 2052 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
2053 | destroy_workqueue(kmultipathd); |
2054 | ||
10d3bd09 | 2055 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
2056 | } |
2057 | ||
1da177e4 LT |
2058 | module_init(dm_multipath_init); |
2059 | module_exit(dm_multipath_exit); | |
2060 | ||
2061 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
2062 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
2063 | MODULE_LICENSE("GPL"); |