]>
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 | ||
7e0d574f | 93 | enum dm_queue_mode queue_mode; |
e83068a5 | 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 | |
b194679f BVA |
301 | lockdep_assert_held(&m->lock); |
302 | ||
91e968aa | 303 | if (atomic_read(&m->pg_init_in_progress) || test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) |
3e9f1be1 | 304 | return 0; |
17f4ff45 | 305 | |
91e968aa | 306 | atomic_inc(&m->pg_init_count); |
518257b1 | 307 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
3e9f1be1 HR |
308 | |
309 | /* Check here to reset pg_init_required */ | |
310 | if (!m->current_pg) | |
311 | return 0; | |
312 | ||
518257b1 | 313 | if (test_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags)) |
4e2d19e4 CS |
314 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? |
315 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
316 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
317 | /* Skip failed paths */ | |
318 | if (!pgpath->is_active) | |
319 | continue; | |
4e2d19e4 CS |
320 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
321 | pg_init_delay)) | |
91e968aa | 322 | atomic_inc(&m->pg_init_in_progress); |
fb612642 | 323 | } |
91e968aa | 324 | return atomic_read(&m->pg_init_in_progress); |
fb612642 KU |
325 | } |
326 | ||
c1d7ecf7 | 327 | static int pg_init_all_paths(struct multipath *m) |
1da177e4 | 328 | { |
c1d7ecf7 | 329 | int ret; |
2da1610a MS |
330 | unsigned long flags; |
331 | ||
332 | spin_lock_irqsave(&m->lock, flags); | |
c1d7ecf7 | 333 | ret = __pg_init_all_paths(m); |
2da1610a | 334 | spin_unlock_irqrestore(&m->lock, flags); |
c1d7ecf7 BVA |
335 | |
336 | return ret; | |
2da1610a MS |
337 | } |
338 | ||
339 | static void __switch_pg(struct multipath *m, struct priority_group *pg) | |
340 | { | |
341 | m->current_pg = pg; | |
1da177e4 LT |
342 | |
343 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 344 | if (m->hw_handler_name) { |
518257b1 MS |
345 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
346 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 347 | } else { |
518257b1 MS |
348 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
349 | clear_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 350 | } |
c9e45581 | 351 | |
91e968aa | 352 | atomic_set(&m->pg_init_count, 0); |
1da177e4 LT |
353 | } |
354 | ||
2da1610a MS |
355 | static struct pgpath *choose_path_in_pg(struct multipath *m, |
356 | struct priority_group *pg, | |
357 | size_t nr_bytes) | |
1da177e4 | 358 | { |
2da1610a | 359 | unsigned long flags; |
c922d5f7 | 360 | struct dm_path *path; |
2da1610a | 361 | struct pgpath *pgpath; |
1da177e4 | 362 | |
90a4323c | 363 | path = pg->ps.type->select_path(&pg->ps, nr_bytes); |
1da177e4 | 364 | if (!path) |
2da1610a | 365 | return ERR_PTR(-ENXIO); |
1da177e4 | 366 | |
2da1610a | 367 | pgpath = path_to_pgpath(path); |
1da177e4 | 368 | |
2da1610a MS |
369 | if (unlikely(lockless_dereference(m->current_pg) != pg)) { |
370 | /* Only update current_pgpath if pg changed */ | |
371 | spin_lock_irqsave(&m->lock, flags); | |
372 | m->current_pgpath = pgpath; | |
373 | __switch_pg(m, pg); | |
374 | spin_unlock_irqrestore(&m->lock, flags); | |
375 | } | |
1da177e4 | 376 | |
2da1610a | 377 | return pgpath; |
1da177e4 LT |
378 | } |
379 | ||
2da1610a | 380 | static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 | 381 | { |
2da1610a | 382 | unsigned long flags; |
1da177e4 | 383 | struct priority_group *pg; |
2da1610a | 384 | struct pgpath *pgpath; |
d19a55cc | 385 | unsigned bypassed = 1; |
1da177e4 | 386 | |
91e968aa | 387 | if (!atomic_read(&m->nr_valid_paths)) { |
518257b1 | 388 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
1da177e4 | 389 | goto failed; |
1f271972 | 390 | } |
1da177e4 LT |
391 | |
392 | /* Were we instructed to switch PG? */ | |
2da1610a MS |
393 | if (lockless_dereference(m->next_pg)) { |
394 | spin_lock_irqsave(&m->lock, flags); | |
1da177e4 | 395 | pg = m->next_pg; |
2da1610a MS |
396 | if (!pg) { |
397 | spin_unlock_irqrestore(&m->lock, flags); | |
398 | goto check_current_pg; | |
399 | } | |
1da177e4 | 400 | m->next_pg = NULL; |
2da1610a MS |
401 | spin_unlock_irqrestore(&m->lock, flags); |
402 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
403 | if (!IS_ERR_OR_NULL(pgpath)) | |
404 | return pgpath; | |
1da177e4 LT |
405 | } |
406 | ||
407 | /* Don't change PG until it has no remaining paths */ | |
2da1610a MS |
408 | check_current_pg: |
409 | pg = lockless_dereference(m->current_pg); | |
410 | if (pg) { | |
411 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
412 | if (!IS_ERR_OR_NULL(pgpath)) | |
413 | return pgpath; | |
414 | } | |
1da177e4 LT |
415 | |
416 | /* | |
417 | * Loop through priority groups until we find a valid path. | |
418 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
419 | * Second time we only try the ones we skipped, but set |
420 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
421 | */ |
422 | do { | |
423 | list_for_each_entry(pg, &m->priority_groups, list) { | |
d19a55cc | 424 | if (pg->bypassed == !!bypassed) |
1da177e4 | 425 | continue; |
2da1610a MS |
426 | pgpath = choose_path_in_pg(m, pg, nr_bytes); |
427 | if (!IS_ERR_OR_NULL(pgpath)) { | |
f220fd4e | 428 | if (!bypassed) |
518257b1 | 429 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); |
2da1610a | 430 | return pgpath; |
f220fd4e | 431 | } |
1da177e4 LT |
432 | } |
433 | } while (bypassed--); | |
434 | ||
435 | failed: | |
2da1610a | 436 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
437 | m->current_pgpath = NULL; |
438 | m->current_pg = NULL; | |
2da1610a MS |
439 | spin_unlock_irqrestore(&m->lock, flags); |
440 | ||
441 | return NULL; | |
1da177e4 LT |
442 | } |
443 | ||
45e15720 | 444 | /* |
86331f39 BVA |
445 | * dm_report_EIO() is a macro instead of a function to make pr_debug() |
446 | * report the function name and line number of the function from which | |
447 | * it has been invoked. | |
45e15720 | 448 | */ |
86331f39 | 449 | #define dm_report_EIO(m) \ |
18a482f5 | 450 | do { \ |
86331f39 BVA |
451 | struct mapped_device *md = dm_table_get_md((m)->ti->table); \ |
452 | \ | |
453 | pr_debug("%s: returning EIO; QIFNP = %d; SQIFNP = %d; DNFS = %d\n", \ | |
454 | dm_device_name(md), \ | |
455 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
456 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
457 | dm_noflush_suspending((m)->ti)); \ | |
18a482f5 | 458 | } while (0) |
45e15720 | 459 | |
36fcffcc | 460 | /* |
76e33fe4 | 461 | * Map cloned requests (request-based multipath) |
36fcffcc | 462 | */ |
eb8db831 CH |
463 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, |
464 | union map_info *map_context, | |
465 | struct request **__clone) | |
1da177e4 | 466 | { |
7943bd6d | 467 | struct multipath *m = ti->private; |
eb8db831 | 468 | size_t nr_bytes = blk_rq_bytes(rq); |
1da177e4 | 469 | struct pgpath *pgpath; |
f40c67f0 | 470 | struct block_device *bdev; |
eb8db831 | 471 | struct dm_mpath_io *mpio = get_mpio(map_context); |
7083abbb | 472 | struct request_queue *q; |
eb8db831 | 473 | struct request *clone; |
1da177e4 | 474 | |
1da177e4 | 475 | /* Do we need to select a new pgpath? */ |
2da1610a MS |
476 | pgpath = lockless_dereference(m->current_pgpath); |
477 | if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) | |
478 | pgpath = choose_pgpath(m, nr_bytes); | |
1da177e4 | 479 | |
9bf59a61 | 480 | if (!pgpath) { |
ca5beb76 | 481 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
b88efd43 | 482 | return DM_MAPIO_DELAY_REQUEUE; |
18a482f5 | 483 | dm_report_EIO(m); /* Failed */ |
f98e0eb6 | 484 | return DM_MAPIO_KILL; |
518257b1 MS |
485 | } else if (test_bit(MPATHF_QUEUE_IO, &m->flags) || |
486 | test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { | |
c1d7ecf7 BVA |
487 | if (pg_init_all_paths(m)) |
488 | return DM_MAPIO_DELAY_REQUEUE; | |
06eb061f | 489 | return DM_MAPIO_REQUEUE; |
9bf59a61 | 490 | } |
6afbc01d | 491 | |
eb8db831 | 492 | memset(mpio, 0, sizeof(*mpio)); |
2eb6e1e3 KB |
493 | mpio->pgpath = pgpath; |
494 | mpio->nr_bytes = nr_bytes; | |
495 | ||
9bf59a61 | 496 | bdev = pgpath->path.dev->bdev; |
7083abbb BVA |
497 | q = bdev_get_queue(bdev); |
498 | clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, GFP_ATOMIC); | |
eb8db831 CH |
499 | if (IS_ERR(clone)) { |
500 | /* EBUSY, ENODEV or EWOULDBLOCK: requeue */ | |
7083abbb | 501 | bool queue_dying = blk_queue_dying(q); |
7083abbb BVA |
502 | if (queue_dying) { |
503 | atomic_inc(&m->pg_init_in_progress); | |
504 | activate_or_offline_path(pgpath); | |
7083abbb | 505 | } |
06eb061f | 506 | return DM_MAPIO_DELAY_REQUEUE; |
e5863d9a | 507 | } |
eb8db831 CH |
508 | clone->bio = clone->biotail = NULL; |
509 | clone->rq_disk = bdev->bd_disk; | |
510 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
511 | *__clone = clone; | |
e5863d9a | 512 | |
9bf59a61 MS |
513 | if (pgpath->pg->ps.type->start_io) |
514 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
515 | &pgpath->path, | |
516 | nr_bytes); | |
2eb6e1e3 | 517 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
518 | } |
519 | ||
e5863d9a MS |
520 | static void multipath_release_clone(struct request *clone) |
521 | { | |
eb8db831 | 522 | blk_put_request(clone); |
e5863d9a MS |
523 | } |
524 | ||
76e33fe4 MS |
525 | /* |
526 | * Map cloned bios (bio-based multipath) | |
527 | */ | |
528 | static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_mpath_io *mpio) | |
529 | { | |
530 | size_t nr_bytes = bio->bi_iter.bi_size; | |
531 | struct pgpath *pgpath; | |
532 | unsigned long flags; | |
533 | bool queue_io; | |
534 | ||
535 | /* Do we need to select a new pgpath? */ | |
536 | pgpath = lockless_dereference(m->current_pgpath); | |
537 | queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags); | |
538 | if (!pgpath || !queue_io) | |
539 | pgpath = choose_pgpath(m, nr_bytes); | |
540 | ||
541 | if ((pgpath && queue_io) || | |
542 | (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) { | |
543 | /* Queue for the daemon to resubmit */ | |
544 | spin_lock_irqsave(&m->lock, flags); | |
545 | bio_list_add(&m->queued_bios, bio); | |
546 | spin_unlock_irqrestore(&m->lock, flags); | |
547 | /* PG_INIT_REQUIRED cannot be set without QUEUE_IO */ | |
548 | if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) | |
549 | pg_init_all_paths(m); | |
550 | else if (!queue_io) | |
551 | queue_work(kmultipathd, &m->process_queued_bios); | |
552 | return DM_MAPIO_SUBMITTED; | |
553 | } | |
554 | ||
555 | if (!pgpath) { | |
ca5beb76 BVA |
556 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
557 | return DM_MAPIO_REQUEUE; | |
18a482f5 | 558 | dm_report_EIO(m); |
846785e6 | 559 | return DM_MAPIO_KILL; |
76e33fe4 MS |
560 | } |
561 | ||
562 | mpio->pgpath = pgpath; | |
563 | mpio->nr_bytes = nr_bytes; | |
564 | ||
4e4cbee9 | 565 | bio->bi_status = 0; |
74d46992 | 566 | bio_set_dev(bio, pgpath->path.dev->bdev); |
1eff9d32 | 567 | bio->bi_opf |= REQ_FAILFAST_TRANSPORT; |
76e33fe4 MS |
568 | |
569 | if (pgpath->pg->ps.type->start_io) | |
570 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
571 | &pgpath->path, | |
572 | nr_bytes); | |
573 | return DM_MAPIO_REMAPPED; | |
574 | } | |
575 | ||
576 | static int multipath_map_bio(struct dm_target *ti, struct bio *bio) | |
577 | { | |
578 | struct multipath *m = ti->private; | |
bf661be1 MS |
579 | struct dm_mpath_io *mpio = NULL; |
580 | ||
581 | multipath_init_per_bio_data(bio, &mpio, NULL); | |
76e33fe4 MS |
582 | |
583 | return __multipath_map_bio(m, bio, mpio); | |
584 | } | |
585 | ||
7e48c768 | 586 | static void process_queued_io_list(struct multipath *m) |
76e33fe4 | 587 | { |
7e48c768 MS |
588 | if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED) |
589 | dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table)); | |
590 | else if (m->queue_mode == DM_TYPE_BIO_BASED) | |
76e33fe4 MS |
591 | queue_work(kmultipathd, &m->process_queued_bios); |
592 | } | |
593 | ||
594 | static void process_queued_bios(struct work_struct *work) | |
595 | { | |
596 | int r; | |
597 | unsigned long flags; | |
598 | struct bio *bio; | |
599 | struct bio_list bios; | |
600 | struct blk_plug plug; | |
601 | struct multipath *m = | |
602 | container_of(work, struct multipath, process_queued_bios); | |
603 | ||
604 | bio_list_init(&bios); | |
605 | ||
606 | spin_lock_irqsave(&m->lock, flags); | |
607 | ||
608 | if (bio_list_empty(&m->queued_bios)) { | |
609 | spin_unlock_irqrestore(&m->lock, flags); | |
610 | return; | |
611 | } | |
612 | ||
613 | bio_list_merge(&bios, &m->queued_bios); | |
614 | bio_list_init(&m->queued_bios); | |
615 | ||
616 | spin_unlock_irqrestore(&m->lock, flags); | |
617 | ||
618 | blk_start_plug(&plug); | |
619 | while ((bio = bio_list_pop(&bios))) { | |
620 | r = __multipath_map_bio(m, bio, get_mpio_from_bio(bio)); | |
846785e6 CH |
621 | switch (r) { |
622 | case DM_MAPIO_KILL: | |
4e4cbee9 CH |
623 | bio->bi_status = BLK_STS_IOERR; |
624 | bio_endio(bio); | |
047385b3 | 625 | break; |
846785e6 | 626 | case DM_MAPIO_REQUEUE: |
4e4cbee9 | 627 | bio->bi_status = BLK_STS_DM_REQUEUE; |
76e33fe4 | 628 | bio_endio(bio); |
846785e6 CH |
629 | break; |
630 | case DM_MAPIO_REMAPPED: | |
76e33fe4 | 631 | generic_make_request(bio); |
846785e6 | 632 | break; |
9157c8d3 BVA |
633 | case 0: |
634 | break; | |
635 | default: | |
636 | WARN_ONCE(true, "__multipath_map_bio() returned %d\n", r); | |
846785e6 | 637 | } |
76e33fe4 MS |
638 | } |
639 | blk_finish_plug(&plug); | |
640 | } | |
641 | ||
9a8ac3ae BVA |
642 | static void assign_bit(bool value, long nr, unsigned long *addr) |
643 | { | |
644 | if (value) | |
645 | set_bit(nr, addr); | |
646 | else | |
647 | clear_bit(nr, addr); | |
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); | |
9a8ac3ae BVA |
659 | assign_bit((save_old_value && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) || |
660 | (!save_old_value && queue_if_no_path), | |
661 | MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
662 | assign_bit(queue_if_no_path || dm_noflush_suspending(m->ti), | |
663 | MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
1da177e4 LT |
664 | spin_unlock_irqrestore(&m->lock, flags); |
665 | ||
76e33fe4 | 666 | if (!queue_if_no_path) { |
63d832c3 | 667 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 668 | process_queued_io_list(m); |
76e33fe4 | 669 | } |
63d832c3 | 670 | |
1da177e4 LT |
671 | return 0; |
672 | } | |
673 | ||
1da177e4 LT |
674 | /* |
675 | * An event is triggered whenever a path is taken out of use. | |
676 | * Includes path failure and PG bypass. | |
677 | */ | |
c4028958 | 678 | static void trigger_event(struct work_struct *work) |
1da177e4 | 679 | { |
c4028958 DH |
680 | struct multipath *m = |
681 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
682 | |
683 | dm_table_event(m->ti->table); | |
684 | } | |
685 | ||
686 | /*----------------------------------------------------------------- | |
687 | * Constructor/argument parsing: | |
688 | * <#multipath feature args> [<arg>]* | |
689 | * <#hw_handler args> [hw_handler [<arg>]*] | |
690 | * <#priority groups> | |
691 | * <initial priority group> | |
692 | * [<selector> <#selector args> [<arg>]* | |
693 | * <#paths> <#per-path selector args> | |
694 | * [<path> [<arg>]* ]+ ]+ | |
695 | *---------------------------------------------------------------*/ | |
498f0103 | 696 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
697 | struct dm_target *ti) |
698 | { | |
699 | int r; | |
700 | struct path_selector_type *pst; | |
701 | unsigned ps_argc; | |
702 | ||
5916a22b | 703 | static const struct dm_arg _args[] = { |
72d94861 | 704 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
705 | }; |
706 | ||
498f0103 | 707 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 708 | if (!pst) { |
72d94861 | 709 | ti->error = "unknown path selector type"; |
1da177e4 LT |
710 | return -EINVAL; |
711 | } | |
712 | ||
498f0103 | 713 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
714 | if (r) { |
715 | dm_put_path_selector(pst); | |
1da177e4 | 716 | return -EINVAL; |
371b2e34 | 717 | } |
1da177e4 LT |
718 | |
719 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
720 | if (r) { | |
721 | dm_put_path_selector(pst); | |
72d94861 | 722 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
723 | return r; |
724 | } | |
725 | ||
726 | pg->ps.type = pst; | |
498f0103 | 727 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
728 | |
729 | return 0; | |
730 | } | |
731 | ||
498f0103 | 732 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, |
1da177e4 LT |
733 | struct dm_target *ti) |
734 | { | |
735 | int r; | |
736 | struct pgpath *p; | |
ae11b1b3 | 737 | struct multipath *m = ti->private; |
a58a935d MS |
738 | struct request_queue *q = NULL; |
739 | const char *attached_handler_name; | |
1da177e4 LT |
740 | |
741 | /* we need at least a path arg */ | |
742 | if (as->argc < 1) { | |
72d94861 | 743 | ti->error = "no device given"; |
01460f35 | 744 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
745 | } |
746 | ||
747 | p = alloc_pgpath(); | |
748 | if (!p) | |
01460f35 | 749 | return ERR_PTR(-ENOMEM); |
1da177e4 | 750 | |
498f0103 | 751 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), |
8215d6ec | 752 | &p->path.dev); |
1da177e4 | 753 | if (r) { |
72d94861 | 754 | ti->error = "error getting device"; |
1da177e4 LT |
755 | goto bad; |
756 | } | |
757 | ||
518257b1 | 758 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) || m->hw_handler_name) |
a58a935d MS |
759 | q = bdev_get_queue(p->path.dev->bdev); |
760 | ||
518257b1 | 761 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) { |
1bab0de0 | 762 | retain: |
a58a935d MS |
763 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); |
764 | if (attached_handler_name) { | |
54cd640d | 765 | /* |
766 | * Clear any hw_handler_params associated with a | |
767 | * handler that isn't already attached. | |
768 | */ | |
769 | if (m->hw_handler_name && strcmp(attached_handler_name, m->hw_handler_name)) { | |
770 | kfree(m->hw_handler_params); | |
771 | m->hw_handler_params = NULL; | |
772 | } | |
773 | ||
a58a935d MS |
774 | /* |
775 | * Reset hw_handler_name to match the attached handler | |
a58a935d MS |
776 | * |
777 | * NB. This modifies the table line to show the actual | |
778 | * handler instead of the original table passed in. | |
779 | */ | |
780 | kfree(m->hw_handler_name); | |
781 | m->hw_handler_name = attached_handler_name; | |
a58a935d MS |
782 | } |
783 | } | |
a0cf7ea9 | 784 | |
a58a935d | 785 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
786 | r = scsi_dh_attach(q, m->hw_handler_name); |
787 | if (r == -EBUSY) { | |
1bab0de0 | 788 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 789 | |
1bab0de0 CH |
790 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
791 | bdevname(p->path.dev->bdev, b)); | |
792 | goto retain; | |
793 | } | |
ae11b1b3 | 794 | if (r < 0) { |
a0cf7ea9 | 795 | ti->error = "error attaching hardware handler"; |
ae11b1b3 HR |
796 | dm_put_device(ti, p->path.dev); |
797 | goto bad; | |
798 | } | |
2bfd2e13 CS |
799 | |
800 | if (m->hw_handler_params) { | |
801 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
802 | if (r < 0) { | |
803 | ti->error = "unable to set hardware " | |
804 | "handler parameters"; | |
2bfd2e13 CS |
805 | dm_put_device(ti, p->path.dev); |
806 | goto bad; | |
807 | } | |
808 | } | |
ae11b1b3 HR |
809 | } |
810 | ||
1da177e4 LT |
811 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
812 | if (r) { | |
813 | dm_put_device(ti, p->path.dev); | |
814 | goto bad; | |
815 | } | |
816 | ||
817 | return p; | |
818 | ||
819 | bad: | |
820 | free_pgpath(p); | |
01460f35 | 821 | return ERR_PTR(r); |
1da177e4 LT |
822 | } |
823 | ||
498f0103 | 824 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 825 | struct multipath *m) |
1da177e4 | 826 | { |
5916a22b | 827 | static const struct dm_arg _args[] = { |
72d94861 AK |
828 | {1, 1024, "invalid number of paths"}, |
829 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
830 | }; |
831 | ||
832 | int r; | |
498f0103 | 833 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 834 | struct priority_group *pg; |
28f16c20 | 835 | struct dm_target *ti = m->ti; |
1da177e4 LT |
836 | |
837 | if (as->argc < 2) { | |
838 | as->argc = 0; | |
01460f35 BM |
839 | ti->error = "not enough priority group arguments"; |
840 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
841 | } |
842 | ||
843 | pg = alloc_priority_group(); | |
844 | if (!pg) { | |
72d94861 | 845 | ti->error = "couldn't allocate priority group"; |
01460f35 | 846 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
847 | } |
848 | pg->m = m; | |
849 | ||
850 | r = parse_path_selector(as, pg, ti); | |
851 | if (r) | |
852 | goto bad; | |
853 | ||
854 | /* | |
855 | * read the paths | |
856 | */ | |
498f0103 | 857 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
858 | if (r) |
859 | goto bad; | |
860 | ||
498f0103 | 861 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
862 | if (r) |
863 | goto bad; | |
864 | ||
498f0103 | 865 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
866 | for (i = 0; i < pg->nr_pgpaths; i++) { |
867 | struct pgpath *pgpath; | |
498f0103 | 868 | struct dm_arg_set path_args; |
1da177e4 | 869 | |
498f0103 | 870 | if (as->argc < nr_args) { |
148acff6 | 871 | ti->error = "not enough path parameters"; |
6bbf79a1 | 872 | r = -EINVAL; |
1da177e4 | 873 | goto bad; |
148acff6 | 874 | } |
1da177e4 | 875 | |
498f0103 | 876 | path_args.argc = nr_args; |
1da177e4 LT |
877 | path_args.argv = as->argv; |
878 | ||
879 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
880 | if (IS_ERR(pgpath)) { |
881 | r = PTR_ERR(pgpath); | |
1da177e4 | 882 | goto bad; |
01460f35 | 883 | } |
1da177e4 LT |
884 | |
885 | pgpath->pg = pg; | |
886 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 887 | dm_consume_args(as, nr_args); |
1da177e4 LT |
888 | } |
889 | ||
890 | return pg; | |
891 | ||
892 | bad: | |
893 | free_priority_group(pg, ti); | |
01460f35 | 894 | return ERR_PTR(r); |
1da177e4 LT |
895 | } |
896 | ||
498f0103 | 897 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 898 | { |
1da177e4 | 899 | unsigned hw_argc; |
2bfd2e13 | 900 | int ret; |
28f16c20 | 901 | struct dm_target *ti = m->ti; |
1da177e4 | 902 | |
5916a22b | 903 | static const struct dm_arg _args[] = { |
72d94861 | 904 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
905 | }; |
906 | ||
498f0103 | 907 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
908 | return -EINVAL; |
909 | ||
910 | if (!hw_argc) | |
911 | return 0; | |
912 | ||
e83068a5 | 913 | if (m->queue_mode == DM_TYPE_BIO_BASED) { |
76e33fe4 MS |
914 | dm_consume_args(as, hw_argc); |
915 | DMERR("bio-based multipath doesn't allow hardware handler args"); | |
916 | return 0; | |
917 | } | |
918 | ||
498f0103 | 919 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
f97dc421 | 920 | if (!m->hw_handler_name) |
921 | return -EINVAL; | |
14e98c5c | 922 | |
2bfd2e13 CS |
923 | if (hw_argc > 1) { |
924 | char *p; | |
925 | int i, j, len = 4; | |
926 | ||
927 | for (i = 0; i <= hw_argc - 2; i++) | |
928 | len += strlen(as->argv[i]) + 1; | |
929 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
930 | if (!p) { | |
931 | ti->error = "memory allocation failed"; | |
932 | ret = -ENOMEM; | |
933 | goto fail; | |
934 | } | |
935 | j = sprintf(p, "%d", hw_argc - 1); | |
936 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
937 | j = sprintf(p, "%s", as->argv[i]); | |
938 | } | |
498f0103 | 939 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
940 | |
941 | return 0; | |
2bfd2e13 CS |
942 | fail: |
943 | kfree(m->hw_handler_name); | |
944 | m->hw_handler_name = NULL; | |
945 | return ret; | |
1da177e4 LT |
946 | } |
947 | ||
498f0103 | 948 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
949 | { |
950 | int r; | |
951 | unsigned argc; | |
28f16c20 | 952 | struct dm_target *ti = m->ti; |
498f0103 | 953 | const char *arg_name; |
1da177e4 | 954 | |
5916a22b | 955 | static const struct dm_arg _args[] = { |
e83068a5 | 956 | {0, 8, "invalid number of feature args"}, |
c9e45581 | 957 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 958 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
959 | }; |
960 | ||
498f0103 | 961 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
962 | if (r) |
963 | return -EINVAL; | |
964 | ||
965 | if (!argc) | |
966 | return 0; | |
967 | ||
c9e45581 | 968 | do { |
498f0103 | 969 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
970 | argc--; |
971 | ||
498f0103 | 972 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
be7d31cc | 973 | r = queue_if_no_path(m, true, false); |
c9e45581 DW |
974 | continue; |
975 | } | |
976 | ||
a58a935d | 977 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
518257b1 | 978 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); |
a58a935d MS |
979 | continue; |
980 | } | |
981 | ||
498f0103 | 982 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 983 | (argc >= 1)) { |
498f0103 | 984 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
985 | argc--; |
986 | continue; | |
987 | } | |
988 | ||
498f0103 | 989 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 990 | (argc >= 1)) { |
498f0103 | 991 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
992 | argc--; |
993 | continue; | |
994 | } | |
995 | ||
e83068a5 MS |
996 | if (!strcasecmp(arg_name, "queue_mode") && |
997 | (argc >= 1)) { | |
998 | const char *queue_mode_name = dm_shift_arg(as); | |
999 | ||
1000 | if (!strcasecmp(queue_mode_name, "bio")) | |
1001 | m->queue_mode = DM_TYPE_BIO_BASED; | |
1002 | else if (!strcasecmp(queue_mode_name, "rq")) | |
1003 | m->queue_mode = DM_TYPE_REQUEST_BASED; | |
1004 | else if (!strcasecmp(queue_mode_name, "mq")) | |
1005 | m->queue_mode = DM_TYPE_MQ_REQUEST_BASED; | |
1006 | else { | |
1007 | ti->error = "Unknown 'queue_mode' requested"; | |
1008 | r = -EINVAL; | |
1009 | } | |
1010 | argc--; | |
1011 | continue; | |
1012 | } | |
1013 | ||
1da177e4 | 1014 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
1015 | r = -EINVAL; |
1016 | } while (argc && !r); | |
1017 | ||
1018 | return r; | |
1da177e4 LT |
1019 | } |
1020 | ||
e83068a5 | 1021 | static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv) |
1da177e4 | 1022 | { |
498f0103 | 1023 | /* target arguments */ |
5916a22b | 1024 | static const struct dm_arg _args[] = { |
a490a07a MS |
1025 | {0, 1024, "invalid number of priority groups"}, |
1026 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
1027 | }; |
1028 | ||
1029 | int r; | |
1030 | struct multipath *m; | |
498f0103 | 1031 | struct dm_arg_set as; |
1da177e4 LT |
1032 | unsigned pg_count = 0; |
1033 | unsigned next_pg_num; | |
1034 | ||
1035 | as.argc = argc; | |
1036 | as.argv = argv; | |
1037 | ||
e83068a5 | 1038 | m = alloc_multipath(ti); |
1da177e4 | 1039 | if (!m) { |
72d94861 | 1040 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
1041 | return -EINVAL; |
1042 | } | |
1043 | ||
28f16c20 | 1044 | r = parse_features(&as, m); |
1da177e4 LT |
1045 | if (r) |
1046 | goto bad; | |
1047 | ||
e83068a5 MS |
1048 | r = alloc_multipath_stage2(ti, m); |
1049 | if (r) | |
1050 | goto bad; | |
1051 | ||
28f16c20 | 1052 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
1053 | if (r) |
1054 | goto bad; | |
1055 | ||
498f0103 | 1056 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
1057 | if (r) |
1058 | goto bad; | |
1059 | ||
498f0103 | 1060 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
1061 | if (r) |
1062 | goto bad; | |
1063 | ||
a490a07a MS |
1064 | if ((!m->nr_priority_groups && next_pg_num) || |
1065 | (m->nr_priority_groups && !next_pg_num)) { | |
1066 | ti->error = "invalid initial priority group"; | |
1067 | r = -EINVAL; | |
1068 | goto bad; | |
1069 | } | |
1070 | ||
1da177e4 LT |
1071 | /* parse the priority groups */ |
1072 | while (as.argc) { | |
1073 | struct priority_group *pg; | |
91e968aa | 1074 | unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths); |
1da177e4 | 1075 | |
28f16c20 | 1076 | pg = parse_priority_group(&as, m); |
01460f35 BM |
1077 | if (IS_ERR(pg)) { |
1078 | r = PTR_ERR(pg); | |
1da177e4 LT |
1079 | goto bad; |
1080 | } | |
1081 | ||
91e968aa MS |
1082 | nr_valid_paths += pg->nr_pgpaths; |
1083 | atomic_set(&m->nr_valid_paths, nr_valid_paths); | |
1084 | ||
1da177e4 LT |
1085 | list_add_tail(&pg->list, &m->priority_groups); |
1086 | pg_count++; | |
1087 | pg->pg_num = pg_count; | |
1088 | if (!--next_pg_num) | |
1089 | m->next_pg = pg; | |
1090 | } | |
1091 | ||
1092 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 1093 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
1094 | r = -EINVAL; |
1095 | goto bad; | |
1096 | } | |
1097 | ||
55a62eef AK |
1098 | ti->num_flush_bios = 1; |
1099 | ti->num_discard_bios = 1; | |
042bcef8 | 1100 | ti->num_write_same_bios = 1; |
ac62d620 | 1101 | ti->num_write_zeroes_bios = 1; |
e83068a5 | 1102 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
bf661be1 | 1103 | ti->per_io_data_size = multipath_per_bio_data_size(); |
eb8db831 | 1104 | else |
8637a6bf | 1105 | ti->per_io_data_size = sizeof(struct dm_mpath_io); |
8627921f | 1106 | |
1da177e4 LT |
1107 | return 0; |
1108 | ||
1109 | bad: | |
1110 | free_multipath(m); | |
1111 | return r; | |
1112 | } | |
1113 | ||
2bded7bd KU |
1114 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
1115 | { | |
9f4c3f87 | 1116 | DEFINE_WAIT(wait); |
2bded7bd KU |
1117 | |
1118 | while (1) { | |
9f4c3f87 | 1119 | prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE); |
2bded7bd | 1120 | |
91e968aa | 1121 | if (!atomic_read(&m->pg_init_in_progress)) |
2bded7bd | 1122 | break; |
2bded7bd KU |
1123 | |
1124 | io_schedule(); | |
1125 | } | |
9f4c3f87 | 1126 | finish_wait(&m->pg_init_wait, &wait); |
2bded7bd KU |
1127 | } |
1128 | ||
1129 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 1130 | { |
518257b1 MS |
1131 | set_bit(MPATHF_PG_INIT_DISABLED, &m->flags); |
1132 | smp_mb__after_atomic(); | |
954a73d5 | 1133 | |
bab7cfc7 | 1134 | flush_workqueue(kmpath_handlerd); |
2bded7bd | 1135 | multipath_wait_for_pg_init_completion(m); |
a044d016 | 1136 | flush_workqueue(kmultipathd); |
43829731 | 1137 | flush_work(&m->trigger_event); |
954a73d5 | 1138 | |
518257b1 MS |
1139 | clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags); |
1140 | smp_mb__after_atomic(); | |
6df400ab KU |
1141 | } |
1142 | ||
1143 | static void multipath_dtr(struct dm_target *ti) | |
1144 | { | |
1145 | struct multipath *m = ti->private; | |
1146 | ||
2bded7bd | 1147 | flush_multipath_work(m); |
1da177e4 LT |
1148 | free_multipath(m); |
1149 | } | |
1150 | ||
1da177e4 LT |
1151 | /* |
1152 | * Take a path out of use. | |
1153 | */ | |
1154 | static int fail_path(struct pgpath *pgpath) | |
1155 | { | |
1156 | unsigned long flags; | |
1157 | struct multipath *m = pgpath->pg->m; | |
1158 | ||
1159 | spin_lock_irqsave(&m->lock, flags); | |
1160 | ||
6680073d | 1161 | if (!pgpath->is_active) |
1da177e4 LT |
1162 | goto out; |
1163 | ||
72d94861 | 1164 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1165 | |
1166 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
be7d31cc | 1167 | pgpath->is_active = false; |
1da177e4 LT |
1168 | pgpath->fail_count++; |
1169 | ||
91e968aa | 1170 | atomic_dec(&m->nr_valid_paths); |
1da177e4 LT |
1171 | |
1172 | if (pgpath == m->current_pgpath) | |
1173 | m->current_pgpath = NULL; | |
1174 | ||
b15546f9 | 1175 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
91e968aa | 1176 | pgpath->path.dev->name, atomic_read(&m->nr_valid_paths)); |
b15546f9 | 1177 | |
fe9cf30e | 1178 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1179 | |
1180 | out: | |
1181 | spin_unlock_irqrestore(&m->lock, flags); | |
1182 | ||
1183 | return 0; | |
1184 | } | |
1185 | ||
1186 | /* | |
1187 | * Reinstate a previously-failed path | |
1188 | */ | |
1189 | static int reinstate_path(struct pgpath *pgpath) | |
1190 | { | |
63d832c3 | 1191 | int r = 0, run_queue = 0; |
1da177e4 LT |
1192 | unsigned long flags; |
1193 | struct multipath *m = pgpath->pg->m; | |
91e968aa | 1194 | unsigned nr_valid_paths; |
1da177e4 LT |
1195 | |
1196 | spin_lock_irqsave(&m->lock, flags); | |
1197 | ||
6680073d | 1198 | if (pgpath->is_active) |
1da177e4 LT |
1199 | goto out; |
1200 | ||
ec31f3f7 | 1201 | DMWARN("Reinstating path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1202 | |
1203 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1204 | if (r) | |
1205 | goto out; | |
1206 | ||
be7d31cc | 1207 | pgpath->is_active = true; |
1da177e4 | 1208 | |
91e968aa MS |
1209 | nr_valid_paths = atomic_inc_return(&m->nr_valid_paths); |
1210 | if (nr_valid_paths == 1) { | |
e54f77dd | 1211 | m->current_pgpath = NULL; |
63d832c3 | 1212 | run_queue = 1; |
e54f77dd | 1213 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1214 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
91e968aa | 1215 | atomic_inc(&m->pg_init_in_progress); |
e54f77dd | 1216 | } |
1da177e4 | 1217 | |
b15546f9 | 1218 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
91e968aa | 1219 | pgpath->path.dev->name, nr_valid_paths); |
b15546f9 | 1220 | |
fe9cf30e | 1221 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1222 | |
1223 | out: | |
1224 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 1225 | if (run_queue) { |
63d832c3 | 1226 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1227 | process_queued_io_list(m); |
76e33fe4 | 1228 | } |
1da177e4 LT |
1229 | |
1230 | return r; | |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1235 | */ | |
1236 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1237 | action_fn action) | |
1238 | { | |
19040c0b | 1239 | int r = -EINVAL; |
1da177e4 LT |
1240 | struct pgpath *pgpath; |
1241 | struct priority_group *pg; | |
1242 | ||
1243 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1244 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1245 | if (pgpath->path.dev == dev) | |
1246 | r = action(pgpath); | |
1247 | } | |
1248 | } | |
1249 | ||
1250 | return r; | |
1251 | } | |
1252 | ||
1253 | /* | |
1254 | * Temporarily try to avoid having to use the specified PG | |
1255 | */ | |
1256 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
be7d31cc | 1257 | bool bypassed) |
1da177e4 LT |
1258 | { |
1259 | unsigned long flags; | |
1260 | ||
1261 | spin_lock_irqsave(&m->lock, flags); | |
1262 | ||
1263 | pg->bypassed = bypassed; | |
1264 | m->current_pgpath = NULL; | |
1265 | m->current_pg = NULL; | |
1266 | ||
1267 | spin_unlock_irqrestore(&m->lock, flags); | |
1268 | ||
fe9cf30e | 1269 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1270 | } |
1271 | ||
1272 | /* | |
1273 | * Switch to using the specified PG from the next I/O that gets mapped | |
1274 | */ | |
1275 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1276 | { | |
1277 | struct priority_group *pg; | |
1278 | unsigned pgnum; | |
1279 | unsigned long flags; | |
31998ef1 | 1280 | char dummy; |
1da177e4 | 1281 | |
31998ef1 | 1282 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1283 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1284 | DMWARN("invalid PG number supplied to switch_pg_num"); |
1285 | return -EINVAL; | |
1286 | } | |
1287 | ||
1288 | spin_lock_irqsave(&m->lock, flags); | |
1289 | list_for_each_entry(pg, &m->priority_groups, list) { | |
be7d31cc | 1290 | pg->bypassed = false; |
1da177e4 LT |
1291 | if (--pgnum) |
1292 | continue; | |
1293 | ||
1294 | m->current_pgpath = NULL; | |
1295 | m->current_pg = NULL; | |
1296 | m->next_pg = pg; | |
1297 | } | |
1298 | spin_unlock_irqrestore(&m->lock, flags); | |
1299 | ||
fe9cf30e | 1300 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1301 | return 0; |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Set/clear bypassed status of a PG. | |
1306 | * PGs are numbered upwards from 1 in the order they were declared. | |
1307 | */ | |
be7d31cc | 1308 | static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed) |
1da177e4 LT |
1309 | { |
1310 | struct priority_group *pg; | |
1311 | unsigned pgnum; | |
31998ef1 | 1312 | char dummy; |
1da177e4 | 1313 | |
31998ef1 | 1314 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1315 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1316 | DMWARN("invalid PG number supplied to bypass_pg"); |
1317 | return -EINVAL; | |
1318 | } | |
1319 | ||
1320 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1321 | if (!--pgnum) | |
1322 | break; | |
1323 | } | |
1324 | ||
1325 | bypass_pg(m, pg, bypassed); | |
1326 | return 0; | |
1327 | } | |
1328 | ||
c9e45581 DW |
1329 | /* |
1330 | * Should we retry pg_init immediately? | |
1331 | */ | |
be7d31cc | 1332 | static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) |
c9e45581 DW |
1333 | { |
1334 | unsigned long flags; | |
be7d31cc | 1335 | bool limit_reached = false; |
c9e45581 DW |
1336 | |
1337 | spin_lock_irqsave(&m->lock, flags); | |
1338 | ||
91e968aa MS |
1339 | if (atomic_read(&m->pg_init_count) <= m->pg_init_retries && |
1340 | !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) | |
518257b1 | 1341 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
c9e45581 | 1342 | else |
be7d31cc | 1343 | limit_reached = true; |
c9e45581 DW |
1344 | |
1345 | spin_unlock_irqrestore(&m->lock, flags); | |
1346 | ||
1347 | return limit_reached; | |
1348 | } | |
1349 | ||
3ae31f6a | 1350 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1351 | { |
83c0d5d5 | 1352 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1353 | struct priority_group *pg = pgpath->pg; |
1354 | struct multipath *m = pg->m; | |
1355 | unsigned long flags; | |
be7d31cc | 1356 | bool delay_retry = false; |
cfae5c9b CS |
1357 | |
1358 | /* device or driver problems */ | |
1359 | switch (errors) { | |
1360 | case SCSI_DH_OK: | |
1361 | break; | |
1362 | case SCSI_DH_NOSYS: | |
1363 | if (!m->hw_handler_name) { | |
1364 | errors = 0; | |
1365 | break; | |
1366 | } | |
f7b934c8 MB |
1367 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1368 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1369 | /* |
1370 | * Fail path for now, so we do not ping pong | |
1371 | */ | |
1372 | fail_path(pgpath); | |
1373 | break; | |
1374 | case SCSI_DH_DEV_TEMP_BUSY: | |
1375 | /* | |
1376 | * Probably doing something like FW upgrade on the | |
1377 | * controller so try the other pg. | |
1378 | */ | |
be7d31cc | 1379 | bypass_pg(m, pg, true); |
cfae5c9b | 1380 | break; |
cfae5c9b | 1381 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1382 | /* Wait before retrying. */ |
1383 | delay_retry = 1; | |
7b06e09a | 1384 | /* fall through */ |
cfae5c9b CS |
1385 | case SCSI_DH_IMM_RETRY: |
1386 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1387 | if (pg_init_limit_reached(m, pgpath)) | |
1388 | fail_path(pgpath); | |
1389 | errors = 0; | |
1390 | break; | |
ec31f3f7 | 1391 | case SCSI_DH_DEV_OFFLINED: |
cfae5c9b CS |
1392 | default: |
1393 | /* | |
1394 | * We probably do not want to fail the path for a device | |
1395 | * error, but this is what the old dm did. In future | |
1396 | * patches we can do more advanced handling. | |
1397 | */ | |
1398 | fail_path(pgpath); | |
1399 | } | |
1400 | ||
1401 | spin_lock_irqsave(&m->lock, flags); | |
1402 | if (errors) { | |
e54f77dd CS |
1403 | if (pgpath == m->current_pgpath) { |
1404 | DMERR("Could not failover device. Error %d.", errors); | |
1405 | m->current_pgpath = NULL; | |
1406 | m->current_pg = NULL; | |
1407 | } | |
518257b1 | 1408 | } else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
be7d31cc | 1409 | pg->bypassed = false; |
cfae5c9b | 1410 | |
91e968aa | 1411 | if (atomic_dec_return(&m->pg_init_in_progress) > 0) |
d0259bf0 KU |
1412 | /* Activations of other paths are still on going */ |
1413 | goto out; | |
1414 | ||
518257b1 MS |
1415 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { |
1416 | if (delay_retry) | |
1417 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1418 | else | |
1419 | clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1420 | ||
3e9f1be1 HR |
1421 | if (__pg_init_all_paths(m)) |
1422 | goto out; | |
1423 | } | |
518257b1 | 1424 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
d0259bf0 | 1425 | |
7e48c768 | 1426 | process_queued_io_list(m); |
76e33fe4 | 1427 | |
2bded7bd KU |
1428 | /* |
1429 | * Wake up any thread waiting to suspend. | |
1430 | */ | |
1431 | wake_up(&m->pg_init_wait); | |
1432 | ||
d0259bf0 | 1433 | out: |
cfae5c9b CS |
1434 | spin_unlock_irqrestore(&m->lock, flags); |
1435 | } | |
1436 | ||
89bfce76 | 1437 | static void activate_or_offline_path(struct pgpath *pgpath) |
bab7cfc7 | 1438 | { |
f10e06b7 | 1439 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); |
bab7cfc7 | 1440 | |
f10e06b7 MS |
1441 | if (pgpath->is_active && !blk_queue_dying(q)) |
1442 | scsi_dh_activate(q, pg_init_done, pgpath); | |
3a017509 HR |
1443 | else |
1444 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1445 | } |
1446 | ||
89bfce76 BVA |
1447 | static void activate_path_work(struct work_struct *work) |
1448 | { | |
1449 | struct pgpath *pgpath = | |
1450 | container_of(work, struct pgpath, activate_path.work); | |
1451 | ||
1452 | activate_or_offline_path(pgpath); | |
1453 | } | |
1454 | ||
2a842aca | 1455 | static int noretry_error(blk_status_t error) |
7e782af5 HR |
1456 | { |
1457 | switch (error) { | |
2a842aca CH |
1458 | case BLK_STS_NOTSUPP: |
1459 | case BLK_STS_NOSPC: | |
1460 | case BLK_STS_TARGET: | |
1461 | case BLK_STS_NEXUS: | |
1462 | case BLK_STS_MEDIUM: | |
7e782af5 HR |
1463 | return 1; |
1464 | } | |
1465 | ||
1466 | /* Anything else could be a path failure, so should be retried */ | |
1467 | return 0; | |
1468 | } | |
1469 | ||
b79f10ee | 1470 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
2a842aca | 1471 | blk_status_t error, union map_info *map_context) |
1da177e4 | 1472 | { |
b79f10ee CH |
1473 | struct dm_mpath_io *mpio = get_mpio(map_context); |
1474 | struct pgpath *pgpath = mpio->pgpath; | |
7ed8578a | 1475 | int r = DM_ENDIO_DONE; |
b79f10ee | 1476 | |
f40c67f0 KU |
1477 | /* |
1478 | * We don't queue any clone request inside the multipath target | |
1479 | * during end I/O handling, since those clone requests don't have | |
1480 | * bio clones. If we queue them inside the multipath target, | |
1481 | * we need to make bio clones, that requires memory allocation. | |
4cc96131 | 1482 | * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests |
f40c67f0 KU |
1483 | * don't have bio clones.) |
1484 | * Instead of queueing the clone request here, we queue the original | |
1485 | * request into dm core, which will remake a clone request and | |
1486 | * clone bios for it and resubmit it later. | |
1487 | */ | |
b79f10ee CH |
1488 | if (error && !noretry_error(error)) { |
1489 | struct multipath *m = ti->private; | |
1da177e4 | 1490 | |
7ed8578a | 1491 | r = DM_ENDIO_REQUEUE; |
1da177e4 | 1492 | |
b79f10ee CH |
1493 | if (pgpath) |
1494 | fail_path(pgpath); | |
1da177e4 | 1495 | |
b79f10ee | 1496 | if (atomic_read(&m->nr_valid_paths) == 0 && |
7ed8578a | 1497 | !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { |
2a842aca | 1498 | if (error == BLK_STS_IOERR) |
18a482f5 | 1499 | dm_report_EIO(m); |
7ed8578a CH |
1500 | /* complete with the original error */ |
1501 | r = DM_ENDIO_DONE; | |
1502 | } | |
b79f10ee | 1503 | } |
466891f9 | 1504 | |
1da177e4 | 1505 | if (pgpath) { |
b79f10ee CH |
1506 | struct path_selector *ps = &pgpath->pg->ps; |
1507 | ||
1da177e4 | 1508 | if (ps->type->end_io) |
02ab823f | 1509 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1510 | } |
1da177e4 | 1511 | |
7ed8578a | 1512 | return r; |
1da177e4 LT |
1513 | } |
1514 | ||
4e4cbee9 CH |
1515 | static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, |
1516 | blk_status_t *error) | |
76e33fe4 | 1517 | { |
14ef1e48 CH |
1518 | struct multipath *m = ti->private; |
1519 | struct dm_mpath_io *mpio = get_mpio_from_bio(clone); | |
1520 | struct pgpath *pgpath = mpio->pgpath; | |
76e33fe4 | 1521 | unsigned long flags; |
1be56909 | 1522 | int r = DM_ENDIO_DONE; |
76e33fe4 | 1523 | |
4e4cbee9 | 1524 | if (!*error || noretry_error(*error)) |
14ef1e48 | 1525 | goto done; |
76e33fe4 | 1526 | |
14ef1e48 CH |
1527 | if (pgpath) |
1528 | fail_path(pgpath); | |
76e33fe4 | 1529 | |
ca5beb76 | 1530 | if (atomic_read(&m->nr_valid_paths) == 0 && |
18a482f5 CH |
1531 | !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { |
1532 | dm_report_EIO(m); | |
4e4cbee9 | 1533 | *error = BLK_STS_IOERR; |
14ef1e48 | 1534 | goto done; |
18a482f5 | 1535 | } |
76e33fe4 MS |
1536 | |
1537 | /* Queue for the daemon to resubmit */ | |
bf661be1 | 1538 | dm_bio_restore(get_bio_details_from_bio(clone), clone); |
76e33fe4 MS |
1539 | |
1540 | spin_lock_irqsave(&m->lock, flags); | |
1541 | bio_list_add(&m->queued_bios, clone); | |
1542 | spin_unlock_irqrestore(&m->lock, flags); | |
1543 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) | |
1544 | queue_work(kmultipathd, &m->process_queued_bios); | |
1545 | ||
1be56909 | 1546 | r = DM_ENDIO_INCOMPLETE; |
14ef1e48 | 1547 | done: |
76e33fe4 | 1548 | if (pgpath) { |
14ef1e48 CH |
1549 | struct path_selector *ps = &pgpath->pg->ps; |
1550 | ||
76e33fe4 MS |
1551 | if (ps->type->end_io) |
1552 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); | |
1553 | } | |
1554 | ||
1be56909 | 1555 | return r; |
76e33fe4 MS |
1556 | } |
1557 | ||
1da177e4 LT |
1558 | /* |
1559 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1560 | * the last path fails we must error any remaining I/O. |
1561 | * Note that if the freeze_bdev fails while suspending, the | |
1562 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1563 | */ |
1564 | static void multipath_presuspend(struct dm_target *ti) | |
1565 | { | |
7943bd6d | 1566 | struct multipath *m = ti->private; |
1da177e4 | 1567 | |
be7d31cc | 1568 | queue_if_no_path(m, false, true); |
1da177e4 LT |
1569 | } |
1570 | ||
6df400ab KU |
1571 | static void multipath_postsuspend(struct dm_target *ti) |
1572 | { | |
6380f26f MA |
1573 | struct multipath *m = ti->private; |
1574 | ||
1575 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1576 | flush_multipath_work(m); |
6380f26f | 1577 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1578 | } |
1579 | ||
436d4108 AK |
1580 | /* |
1581 | * Restore the queue_if_no_path setting. | |
1582 | */ | |
1da177e4 LT |
1583 | static void multipath_resume(struct dm_target *ti) |
1584 | { | |
7943bd6d | 1585 | struct multipath *m = ti->private; |
1814f2e3 | 1586 | unsigned long flags; |
1da177e4 | 1587 | |
1814f2e3 | 1588 | spin_lock_irqsave(&m->lock, flags); |
9a8ac3ae BVA |
1589 | assign_bit(test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags), |
1590 | MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
1814f2e3 | 1591 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1592 | } |
1593 | ||
1594 | /* | |
1595 | * Info output has the following format: | |
1596 | * num_multipath_feature_args [multipath_feature_args]* | |
1597 | * num_handler_status_args [handler_status_args]* | |
1598 | * num_groups init_group_number | |
1599 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1600 | * num_paths num_selector_args | |
1601 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1602 | * | |
1603 | * Table output has the following format (identical to the constructor string): | |
1604 | * num_feature_args [features_args]* | |
1605 | * num_handler_args hw_handler [hw_handler_args]* | |
1606 | * num_groups init_group_number | |
1607 | * [priority selector-name num_ps_args [ps_args]* | |
1608 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1609 | */ | |
fd7c092e MP |
1610 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1611 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1612 | { |
1613 | int sz = 0; | |
1614 | unsigned long flags; | |
7943bd6d | 1615 | struct multipath *m = ti->private; |
1da177e4 LT |
1616 | struct priority_group *pg; |
1617 | struct pgpath *p; | |
1618 | unsigned pg_num; | |
1619 | char state; | |
1620 | ||
1621 | spin_lock_irqsave(&m->lock, flags); | |
1622 | ||
1623 | /* Features */ | |
1624 | if (type == STATUSTYPE_INFO) | |
91e968aa MS |
1625 | DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags), |
1626 | atomic_read(&m->pg_init_count)); | |
c9e45581 | 1627 | else { |
518257b1 | 1628 | DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) + |
4e2d19e4 | 1629 | (m->pg_init_retries > 0) * 2 + |
a58a935d | 1630 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
e83068a5 MS |
1631 | test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) + |
1632 | (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2); | |
1633 | ||
518257b1 | 1634 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
c9e45581 DW |
1635 | DMEMIT("queue_if_no_path "); |
1636 | if (m->pg_init_retries) | |
1637 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1638 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1639 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
518257b1 | 1640 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) |
a58a935d | 1641 | DMEMIT("retain_attached_hw_handler "); |
e83068a5 MS |
1642 | if (m->queue_mode != DM_TYPE_REQUEST_BASED) { |
1643 | switch(m->queue_mode) { | |
1644 | case DM_TYPE_BIO_BASED: | |
1645 | DMEMIT("queue_mode bio "); | |
1646 | break; | |
1647 | case DM_TYPE_MQ_REQUEST_BASED: | |
1648 | DMEMIT("queue_mode mq "); | |
1649 | break; | |
7e0d574f BVA |
1650 | default: |
1651 | WARN_ON_ONCE(true); | |
1652 | break; | |
e83068a5 MS |
1653 | } |
1654 | } | |
c9e45581 | 1655 | } |
1da177e4 | 1656 | |
cfae5c9b | 1657 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1658 | DMEMIT("0 "); |
1659 | else | |
cfae5c9b | 1660 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1661 | |
1662 | DMEMIT("%u ", m->nr_priority_groups); | |
1663 | ||
1664 | if (m->next_pg) | |
1665 | pg_num = m->next_pg->pg_num; | |
1666 | else if (m->current_pg) | |
1667 | pg_num = m->current_pg->pg_num; | |
1668 | else | |
a490a07a | 1669 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1670 | |
1671 | DMEMIT("%u ", pg_num); | |
1672 | ||
1673 | switch (type) { | |
1674 | case STATUSTYPE_INFO: | |
1675 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1676 | if (pg->bypassed) | |
1677 | state = 'D'; /* Disabled */ | |
1678 | else if (pg == m->current_pg) | |
1679 | state = 'A'; /* Currently Active */ | |
1680 | else | |
1681 | state = 'E'; /* Enabled */ | |
1682 | ||
1683 | DMEMIT("%c ", state); | |
1684 | ||
1685 | if (pg->ps.type->status) | |
1686 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1687 | result + sz, | |
1688 | maxlen - sz); | |
1689 | else | |
1690 | DMEMIT("0 "); | |
1691 | ||
1692 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1693 | pg->ps.type->info_args); | |
1694 | ||
1695 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1696 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1697 | p->is_active ? "A" : "F", |
1da177e4 LT |
1698 | p->fail_count); |
1699 | if (pg->ps.type->status) | |
1700 | sz += pg->ps.type->status(&pg->ps, | |
1701 | &p->path, type, result + sz, | |
1702 | maxlen - sz); | |
1703 | } | |
1704 | } | |
1705 | break; | |
1706 | ||
1707 | case STATUSTYPE_TABLE: | |
1708 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1709 | DMEMIT("%s ", pg->ps.type->name); | |
1710 | ||
1711 | if (pg->ps.type->status) | |
1712 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1713 | result + sz, | |
1714 | maxlen - sz); | |
1715 | else | |
1716 | DMEMIT("0 "); | |
1717 | ||
1718 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1719 | pg->ps.type->table_args); | |
1720 | ||
1721 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1722 | DMEMIT("%s ", p->path.dev->name); | |
1723 | if (pg->ps.type->status) | |
1724 | sz += pg->ps.type->status(&pg->ps, | |
1725 | &p->path, type, result + sz, | |
1726 | maxlen - sz); | |
1727 | } | |
1728 | } | |
1729 | break; | |
1730 | } | |
1731 | ||
1732 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1733 | } |
1734 | ||
1735 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1736 | { | |
6380f26f | 1737 | int r = -EINVAL; |
1da177e4 | 1738 | struct dm_dev *dev; |
7943bd6d | 1739 | struct multipath *m = ti->private; |
1da177e4 LT |
1740 | action_fn action; |
1741 | ||
6380f26f MA |
1742 | mutex_lock(&m->work_mutex); |
1743 | ||
c2f3d24b KU |
1744 | if (dm_suspended(ti)) { |
1745 | r = -EBUSY; | |
1746 | goto out; | |
1747 | } | |
1748 | ||
1da177e4 | 1749 | if (argc == 1) { |
498f0103 | 1750 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
be7d31cc | 1751 | r = queue_if_no_path(m, true, false); |
6380f26f | 1752 | goto out; |
498f0103 | 1753 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
be7d31cc | 1754 | r = queue_if_no_path(m, false, false); |
6380f26f MA |
1755 | goto out; |
1756 | } | |
1da177e4 LT |
1757 | } |
1758 | ||
6380f26f | 1759 | if (argc != 2) { |
a356e426 | 1760 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1761 | goto out; |
1762 | } | |
1da177e4 | 1763 | |
498f0103 | 1764 | if (!strcasecmp(argv[0], "disable_group")) { |
be7d31cc | 1765 | r = bypass_pg_num(m, argv[1], true); |
6380f26f | 1766 | goto out; |
498f0103 | 1767 | } else if (!strcasecmp(argv[0], "enable_group")) { |
be7d31cc | 1768 | r = bypass_pg_num(m, argv[1], false); |
6380f26f | 1769 | goto out; |
498f0103 | 1770 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1771 | r = switch_pg_num(m, argv[1]); |
1772 | goto out; | |
498f0103 | 1773 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1774 | action = reinstate_path; |
498f0103 | 1775 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1776 | action = fail_path; |
6380f26f | 1777 | else { |
a356e426 | 1778 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1779 | goto out; |
1780 | } | |
1da177e4 | 1781 | |
8215d6ec | 1782 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1783 | if (r) { |
72d94861 | 1784 | DMWARN("message: error getting device %s", |
1da177e4 | 1785 | argv[1]); |
6380f26f | 1786 | goto out; |
1da177e4 LT |
1787 | } |
1788 | ||
1789 | r = action_dev(m, dev, action); | |
1790 | ||
1791 | dm_put_device(ti, dev); | |
1792 | ||
6380f26f MA |
1793 | out: |
1794 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1795 | return r; |
1da177e4 LT |
1796 | } |
1797 | ||
e56f81e0 CH |
1798 | static int multipath_prepare_ioctl(struct dm_target *ti, |
1799 | struct block_device **bdev, fmode_t *mode) | |
9af4aa30 | 1800 | { |
35991652 | 1801 | struct multipath *m = ti->private; |
2da1610a | 1802 | struct pgpath *current_pgpath; |
35991652 MP |
1803 | int r; |
1804 | ||
2da1610a MS |
1805 | current_pgpath = lockless_dereference(m->current_pgpath); |
1806 | if (!current_pgpath) | |
1807 | current_pgpath = choose_pgpath(m, 0); | |
9af4aa30 | 1808 | |
2da1610a | 1809 | if (current_pgpath) { |
518257b1 | 1810 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) { |
2da1610a MS |
1811 | *bdev = current_pgpath->path.dev->bdev; |
1812 | *mode = current_pgpath->path.dev->mode; | |
43e43c9e JN |
1813 | r = 0; |
1814 | } else { | |
1815 | /* pg_init has not started or completed */ | |
1816 | r = -ENOTCONN; | |
1817 | } | |
1818 | } else { | |
1819 | /* No path is available */ | |
518257b1 | 1820 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
43e43c9e JN |
1821 | r = -ENOTCONN; |
1822 | else | |
1823 | r = -EIO; | |
e90dae1f | 1824 | } |
9af4aa30 | 1825 | |
5bbbfdf6 | 1826 | if (r == -ENOTCONN) { |
2da1610a | 1827 | if (!lockless_dereference(m->current_pg)) { |
3e9f1be1 | 1828 | /* Path status changed, redo selection */ |
2da1610a | 1829 | (void) choose_pgpath(m, 0); |
3e9f1be1 | 1830 | } |
518257b1 | 1831 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
2da1610a | 1832 | pg_init_all_paths(m); |
63d832c3 | 1833 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1834 | process_queued_io_list(m); |
3e9f1be1 | 1835 | } |
35991652 | 1836 | |
e56f81e0 CH |
1837 | /* |
1838 | * Only pass ioctls through if the device sizes match exactly. | |
1839 | */ | |
1840 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
1841 | return 1; | |
1842 | return r; | |
9af4aa30 MB |
1843 | } |
1844 | ||
af4874e0 MS |
1845 | static int multipath_iterate_devices(struct dm_target *ti, |
1846 | iterate_devices_callout_fn fn, void *data) | |
1847 | { | |
1848 | struct multipath *m = ti->private; | |
1849 | struct priority_group *pg; | |
1850 | struct pgpath *p; | |
1851 | int ret = 0; | |
1852 | ||
1853 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1854 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1855 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1856 | if (ret) |
1857 | goto out; | |
1858 | } | |
1859 | } | |
1860 | ||
1861 | out: | |
1862 | return ret; | |
1863 | } | |
1864 | ||
9f54cec5 | 1865 | static int pgpath_busy(struct pgpath *pgpath) |
f40c67f0 KU |
1866 | { |
1867 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1868 | ||
52b09914 | 1869 | return blk_lld_busy(q); |
f40c67f0 KU |
1870 | } |
1871 | ||
1872 | /* | |
1873 | * We return "busy", only when we can map I/Os but underlying devices | |
1874 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1875 | * the underlying queue). | |
1876 | * In other words, if we want to kill I/Os or queue them inside us | |
1877 | * due to map unavailability, we don't return "busy". Otherwise, | |
1878 | * dm core won't give us the I/Os and we can't do what we want. | |
1879 | */ | |
1880 | static int multipath_busy(struct dm_target *ti) | |
1881 | { | |
be7d31cc | 1882 | bool busy = false, has_active = false; |
f40c67f0 | 1883 | struct multipath *m = ti->private; |
2da1610a | 1884 | struct priority_group *pg, *next_pg; |
f40c67f0 | 1885 | struct pgpath *pgpath; |
f40c67f0 | 1886 | |
b88efd43 MS |
1887 | /* pg_init in progress */ |
1888 | if (atomic_read(&m->pg_init_in_progress)) | |
2da1610a MS |
1889 | return true; |
1890 | ||
b88efd43 MS |
1891 | /* no paths available, for blk-mq: rely on IO mapping to delay requeue */ |
1892 | if (!atomic_read(&m->nr_valid_paths) && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) | |
1893 | return (m->queue_mode != DM_TYPE_MQ_REQUEST_BASED); | |
1894 | ||
f40c67f0 | 1895 | /* Guess which priority_group will be used at next mapping time */ |
2da1610a MS |
1896 | pg = lockless_dereference(m->current_pg); |
1897 | next_pg = lockless_dereference(m->next_pg); | |
1898 | if (unlikely(!lockless_dereference(m->current_pgpath) && next_pg)) | |
1899 | pg = next_pg; | |
1900 | ||
1901 | if (!pg) { | |
f40c67f0 KU |
1902 | /* |
1903 | * We don't know which pg will be used at next mapping time. | |
2da1610a | 1904 | * We don't call choose_pgpath() here to avoid to trigger |
f40c67f0 KU |
1905 | * pg_init just by busy checking. |
1906 | * So we don't know whether underlying devices we will be using | |
1907 | * at next mapping time are busy or not. Just try mapping. | |
1908 | */ | |
2da1610a MS |
1909 | return busy; |
1910 | } | |
f40c67f0 KU |
1911 | |
1912 | /* | |
1913 | * If there is one non-busy active path at least, the path selector | |
1914 | * will be able to select it. So we consider such a pg as not busy. | |
1915 | */ | |
be7d31cc | 1916 | busy = true; |
2da1610a | 1917 | list_for_each_entry(pgpath, &pg->pgpaths, list) { |
f40c67f0 | 1918 | if (pgpath->is_active) { |
be7d31cc | 1919 | has_active = true; |
9f54cec5 | 1920 | if (!pgpath_busy(pgpath)) { |
be7d31cc | 1921 | busy = false; |
f40c67f0 KU |
1922 | break; |
1923 | } | |
1924 | } | |
2da1610a | 1925 | } |
f40c67f0 | 1926 | |
2da1610a | 1927 | if (!has_active) { |
f40c67f0 KU |
1928 | /* |
1929 | * No active path in this pg, so this pg won't be used and | |
1930 | * the current_pg will be changed at next mapping time. | |
1931 | * We need to try mapping to determine it. | |
1932 | */ | |
be7d31cc | 1933 | busy = false; |
2da1610a | 1934 | } |
f40c67f0 KU |
1935 | |
1936 | return busy; | |
1937 | } | |
1938 | ||
1da177e4 LT |
1939 | /*----------------------------------------------------------------- |
1940 | * Module setup | |
1941 | *---------------------------------------------------------------*/ | |
1942 | static struct target_type multipath_target = { | |
1943 | .name = "multipath", | |
e83068a5 | 1944 | .version = {1, 12, 0}, |
16f12266 | 1945 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE, |
1da177e4 LT |
1946 | .module = THIS_MODULE, |
1947 | .ctr = multipath_ctr, | |
1948 | .dtr = multipath_dtr, | |
e5863d9a MS |
1949 | .clone_and_map_rq = multipath_clone_and_map, |
1950 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 1951 | .rq_end_io = multipath_end_io, |
76e33fe4 MS |
1952 | .map = multipath_map_bio, |
1953 | .end_io = multipath_end_io_bio, | |
1954 | .presuspend = multipath_presuspend, | |
1955 | .postsuspend = multipath_postsuspend, | |
1956 | .resume = multipath_resume, | |
1957 | .status = multipath_status, | |
1958 | .message = multipath_message, | |
1959 | .prepare_ioctl = multipath_prepare_ioctl, | |
1960 | .iterate_devices = multipath_iterate_devices, | |
1961 | .busy = multipath_busy, | |
1962 | }; | |
1963 | ||
1da177e4 LT |
1964 | static int __init dm_multipath_init(void) |
1965 | { | |
1966 | int r; | |
1967 | ||
1da177e4 LT |
1968 | r = dm_register_target(&multipath_target); |
1969 | if (r < 0) { | |
76e33fe4 | 1970 | DMERR("request-based register failed %d", r); |
ff658e9c JT |
1971 | r = -EINVAL; |
1972 | goto bad_register_target; | |
1da177e4 LT |
1973 | } |
1974 | ||
4d4d66ab | 1975 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 1976 | if (!kmultipathd) { |
0cd33124 | 1977 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
1978 | r = -ENOMEM; |
1979 | goto bad_alloc_kmultipathd; | |
c557308e AK |
1980 | } |
1981 | ||
bab7cfc7 CS |
1982 | /* |
1983 | * A separate workqueue is used to handle the device handlers | |
1984 | * to avoid overloading existing workqueue. Overloading the | |
1985 | * old workqueue would also create a bottleneck in the | |
1986 | * path of the storage hardware device activation. | |
1987 | */ | |
4d4d66ab TH |
1988 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
1989 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
1990 | if (!kmpath_handlerd) { |
1991 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
1992 | r = -ENOMEM; |
1993 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
1994 | } |
1995 | ||
ff658e9c JT |
1996 | return 0; |
1997 | ||
1998 | bad_alloc_kmpath_handlerd: | |
1999 | destroy_workqueue(kmultipathd); | |
2000 | bad_alloc_kmultipathd: | |
2001 | dm_unregister_target(&multipath_target); | |
2002 | bad_register_target: | |
1da177e4 LT |
2003 | return r; |
2004 | } | |
2005 | ||
2006 | static void __exit dm_multipath_exit(void) | |
2007 | { | |
bab7cfc7 | 2008 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
2009 | destroy_workqueue(kmultipathd); |
2010 | ||
10d3bd09 | 2011 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
2012 | } |
2013 | ||
1da177e4 LT |
2014 | module_init(dm_multipath_init); |
2015 | module_exit(dm_multipath_exit); | |
2016 | ||
2017 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
2018 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
2019 | MODULE_LICENSE("GPL"); |