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