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> | |
be240ff5 | 23 | #include <linux/timer.h> |
1da177e4 | 24 | #include <linux/workqueue.h> |
35991652 | 25 | #include <linux/delay.h> |
cfae5c9b | 26 | #include <scsi/scsi_dh.h> |
60063497 | 27 | #include <linux/atomic.h> |
78ce23b5 | 28 | #include <linux/blk-mq.h> |
1da177e4 | 29 | |
72d94861 | 30 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
31 | #define DM_PG_INIT_DELAY_MSECS 2000 |
32 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
be240ff5 AP |
33 | #define QUEUE_IF_NO_PATH_TIMEOUT_DEFAULT 0 |
34 | ||
35 | static unsigned long queue_if_no_path_timeout_secs = QUEUE_IF_NO_PATH_TIMEOUT_DEFAULT; | |
1da177e4 LT |
36 | |
37 | /* Path properties */ | |
38 | struct pgpath { | |
39 | struct list_head list; | |
40 | ||
41 | struct priority_group *pg; /* Owning PG */ | |
42 | unsigned fail_count; /* Cumulative failure count */ | |
43 | ||
c922d5f7 | 44 | struct dm_path path; |
4e2d19e4 | 45 | struct delayed_work activate_path; |
be7d31cc MS |
46 | |
47 | bool is_active:1; /* Path status */ | |
1da177e4 LT |
48 | }; |
49 | ||
50 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
51 | ||
52 | /* | |
53 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
54 | * Each has a path selector which controls which path gets used. | |
55 | */ | |
56 | struct priority_group { | |
57 | struct list_head list; | |
58 | ||
59 | struct multipath *m; /* Owning multipath instance */ | |
60 | struct path_selector ps; | |
61 | ||
62 | unsigned pg_num; /* Reference number */ | |
1da177e4 LT |
63 | unsigned nr_pgpaths; /* Number of paths in PG */ |
64 | struct list_head pgpaths; | |
be7d31cc MS |
65 | |
66 | bool bypassed:1; /* Temporarily bypass this PG? */ | |
1da177e4 LT |
67 | }; |
68 | ||
69 | /* Multipath context */ | |
70 | struct multipath { | |
848b8aef | 71 | unsigned long flags; /* Multipath state flags */ |
4e2d19e4 | 72 | |
1fbdd2b3 | 73 | spinlock_t lock; |
848b8aef | 74 | enum dm_queue_mode queue_mode; |
4e2d19e4 | 75 | |
1da177e4 LT |
76 | struct pgpath *current_pgpath; |
77 | struct priority_group *current_pg; | |
78 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
1da177e4 | 79 | |
848b8aef MS |
80 | atomic_t nr_valid_paths; /* Total number of usable paths */ |
81 | unsigned nr_priority_groups; | |
82 | struct list_head priority_groups; | |
1fbdd2b3 | 83 | |
848b8aef MS |
84 | const char *hw_handler_name; |
85 | char *hw_handler_params; | |
86 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
c9e45581 | 87 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
4e2d19e4 | 88 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
91e968aa MS |
89 | atomic_t pg_init_in_progress; /* Only one pg_init allowed at once */ |
90 | atomic_t pg_init_count; /* Number of times pg_init called */ | |
91 | ||
6380f26f | 92 | struct mutex work_mutex; |
20800cb3 | 93 | struct work_struct trigger_event; |
848b8aef | 94 | struct dm_target *ti; |
76e33fe4 MS |
95 | |
96 | struct work_struct process_queued_bios; | |
97 | struct bio_list queued_bios; | |
be240ff5 AP |
98 | |
99 | struct timer_list nopath_timer; /* Timeout for queue_if_no_path */ | |
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); |
be240ff5 | 117 | static void queue_if_no_path_timeout_work(struct timer_list *t); |
1da177e4 | 118 | |
518257b1 MS |
119 | /*----------------------------------------------- |
120 | * Multipath state flags. | |
121 | *-----------------------------------------------*/ | |
122 | ||
123 | #define MPATHF_QUEUE_IO 0 /* Must we queue all I/O? */ | |
124 | #define MPATHF_QUEUE_IF_NO_PATH 1 /* Queue I/O if last path fails? */ | |
125 | #define MPATHF_SAVED_QUEUE_IF_NO_PATH 2 /* Saved state during suspension */ | |
126 | #define MPATHF_RETAIN_ATTACHED_HW_HANDLER 3 /* If there's already a hw_handler present, don't change it. */ | |
127 | #define MPATHF_PG_INIT_DISABLED 4 /* pg_init is not currently allowed */ | |
128 | #define MPATHF_PG_INIT_REQUIRED 5 /* pg_init needs calling? */ | |
129 | #define MPATHF_PG_INIT_DELAY_RETRY 6 /* Delay pg_init retry? */ | |
1da177e4 | 130 | |
374117ad MS |
131 | static bool mpath_double_check_test_bit(int MPATHF_bit, struct multipath *m) |
132 | { | |
133 | bool r = test_bit(MPATHF_bit, &m->flags); | |
134 | ||
135 | if (r) { | |
136 | unsigned long flags; | |
137 | spin_lock_irqsave(&m->lock, flags); | |
138 | r = test_bit(MPATHF_bit, &m->flags); | |
139 | spin_unlock_irqrestore(&m->lock, flags); | |
140 | } | |
141 | ||
142 | return r; | |
143 | } | |
144 | ||
1da177e4 LT |
145 | /*----------------------------------------------- |
146 | * Allocation routines | |
147 | *-----------------------------------------------*/ | |
148 | ||
149 | static struct pgpath *alloc_pgpath(void) | |
150 | { | |
e69fae56 | 151 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 152 | |
848b8aef MS |
153 | if (!pgpath) |
154 | return NULL; | |
155 | ||
156 | pgpath->is_active = true; | |
1da177e4 LT |
157 | |
158 | return pgpath; | |
159 | } | |
160 | ||
028867ac | 161 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
162 | { |
163 | kfree(pgpath); | |
164 | } | |
165 | ||
166 | static struct priority_group *alloc_priority_group(void) | |
167 | { | |
168 | struct priority_group *pg; | |
169 | ||
e69fae56 | 170 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 171 | |
e69fae56 MM |
172 | if (pg) |
173 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
174 | |
175 | return pg; | |
176 | } | |
177 | ||
178 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
179 | { | |
180 | struct pgpath *pgpath, *tmp; | |
181 | ||
182 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
183 | list_del(&pgpath->list); | |
184 | dm_put_device(ti, pgpath->path.dev); | |
185 | free_pgpath(pgpath); | |
186 | } | |
187 | } | |
188 | ||
189 | static void free_priority_group(struct priority_group *pg, | |
190 | struct dm_target *ti) | |
191 | { | |
192 | struct path_selector *ps = &pg->ps; | |
193 | ||
194 | if (ps->type) { | |
195 | ps->type->destroy(ps); | |
196 | dm_put_path_selector(ps->type); | |
197 | } | |
198 | ||
199 | free_pgpaths(&pg->pgpaths, ti); | |
200 | kfree(pg); | |
201 | } | |
202 | ||
e83068a5 | 203 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
204 | { |
205 | struct multipath *m; | |
206 | ||
e69fae56 | 207 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 208 | if (m) { |
1da177e4 LT |
209 | INIT_LIST_HEAD(&m->priority_groups); |
210 | spin_lock_init(&m->lock); | |
91e968aa | 211 | atomic_set(&m->nr_valid_paths, 0); |
c4028958 | 212 | INIT_WORK(&m->trigger_event, trigger_event); |
6380f26f | 213 | mutex_init(&m->work_mutex); |
8637a6bf | 214 | |
e83068a5 | 215 | m->queue_mode = DM_TYPE_NONE; |
76e33fe4 | 216 | |
28f16c20 MM |
217 | m->ti = ti; |
218 | ti->private = m; | |
be240ff5 AP |
219 | |
220 | timer_setup(&m->nopath_timer, queue_if_no_path_timeout_work, 0); | |
1da177e4 LT |
221 | } |
222 | ||
223 | return m; | |
224 | } | |
225 | ||
e83068a5 MS |
226 | static int alloc_multipath_stage2(struct dm_target *ti, struct multipath *m) |
227 | { | |
228 | if (m->queue_mode == DM_TYPE_NONE) { | |
953923c0 | 229 | m->queue_mode = DM_TYPE_REQUEST_BASED; |
8d47e659 | 230 | } else if (m->queue_mode == DM_TYPE_BIO_BASED) { |
e83068a5 | 231 | INIT_WORK(&m->process_queued_bios, process_queued_bios); |
8d47e659 MS |
232 | /* |
233 | * bio-based doesn't support any direct scsi_dh management; | |
234 | * it just discovers if a scsi_dh is attached. | |
235 | */ | |
236 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); | |
e83068a5 MS |
237 | } |
238 | ||
239 | dm_table_set_type(ti->table, m->queue_mode); | |
240 | ||
c3736674 MS |
241 | /* |
242 | * Init fields that are only used when a scsi_dh is attached | |
243 | * - must do this unconditionally (really doesn't hurt non-SCSI uses) | |
244 | */ | |
245 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
246 | atomic_set(&m->pg_init_in_progress, 0); | |
247 | atomic_set(&m->pg_init_count, 0); | |
248 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; | |
249 | init_waitqueue_head(&m->pg_init_wait); | |
250 | ||
e83068a5 MS |
251 | return 0; |
252 | } | |
253 | ||
1da177e4 LT |
254 | static void free_multipath(struct multipath *m) |
255 | { | |
256 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
257 | |
258 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
259 | list_del(&pg->list); | |
260 | free_priority_group(pg, m->ti); | |
261 | } | |
262 | ||
cfae5c9b | 263 | kfree(m->hw_handler_name); |
2bfd2e13 | 264 | kfree(m->hw_handler_params); |
d5ffebdd | 265 | mutex_destroy(&m->work_mutex); |
1da177e4 LT |
266 | kfree(m); |
267 | } | |
268 | ||
2eff1924 MS |
269 | static struct dm_mpath_io *get_mpio(union map_info *info) |
270 | { | |
271 | return info->ptr; | |
272 | } | |
273 | ||
bf661be1 MS |
274 | static size_t multipath_per_bio_data_size(void) |
275 | { | |
276 | return sizeof(struct dm_mpath_io) + sizeof(struct dm_bio_details); | |
277 | } | |
278 | ||
76e33fe4 MS |
279 | static struct dm_mpath_io *get_mpio_from_bio(struct bio *bio) |
280 | { | |
bf661be1 | 281 | return dm_per_bio_data(bio, multipath_per_bio_data_size()); |
76e33fe4 MS |
282 | } |
283 | ||
d07a241d | 284 | static struct dm_bio_details *get_bio_details_from_mpio(struct dm_mpath_io *mpio) |
76e33fe4 | 285 | { |
bf661be1 | 286 | /* dm_bio_details is immediately after the dm_mpath_io in bio's per-bio-data */ |
bf661be1 | 287 | void *bio_details = mpio + 1; |
bf661be1 MS |
288 | return bio_details; |
289 | } | |
290 | ||
63f6e6fd | 291 | static void multipath_init_per_bio_data(struct bio *bio, struct dm_mpath_io **mpio_p) |
bf661be1 MS |
292 | { |
293 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); | |
d07a241d | 294 | struct dm_bio_details *bio_details = get_bio_details_from_mpio(mpio); |
76e33fe4 | 295 | |
d0442f80 MS |
296 | mpio->nr_bytes = bio->bi_iter.bi_size; |
297 | mpio->pgpath = NULL; | |
298 | *mpio_p = mpio; | |
76e33fe4 | 299 | |
d0442f80 | 300 | dm_bio_record(bio_details, bio); |
76e33fe4 MS |
301 | } |
302 | ||
1da177e4 LT |
303 | /*----------------------------------------------- |
304 | * Path selection | |
305 | *-----------------------------------------------*/ | |
306 | ||
3e9f1be1 | 307 | static int __pg_init_all_paths(struct multipath *m) |
fb612642 KU |
308 | { |
309 | struct pgpath *pgpath; | |
4e2d19e4 | 310 | unsigned long pg_init_delay = 0; |
fb612642 | 311 | |
b194679f BVA |
312 | lockdep_assert_held(&m->lock); |
313 | ||
91e968aa | 314 | if (atomic_read(&m->pg_init_in_progress) || test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) |
3e9f1be1 | 315 | return 0; |
17f4ff45 | 316 | |
91e968aa | 317 | atomic_inc(&m->pg_init_count); |
518257b1 | 318 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
3e9f1be1 HR |
319 | |
320 | /* Check here to reset pg_init_required */ | |
321 | if (!m->current_pg) | |
322 | return 0; | |
323 | ||
518257b1 | 324 | if (test_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags)) |
4e2d19e4 CS |
325 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? |
326 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
327 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
328 | /* Skip failed paths */ | |
329 | if (!pgpath->is_active) | |
330 | continue; | |
4e2d19e4 CS |
331 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
332 | pg_init_delay)) | |
91e968aa | 333 | atomic_inc(&m->pg_init_in_progress); |
fb612642 | 334 | } |
91e968aa | 335 | return atomic_read(&m->pg_init_in_progress); |
fb612642 KU |
336 | } |
337 | ||
c1d7ecf7 | 338 | static int pg_init_all_paths(struct multipath *m) |
1da177e4 | 339 | { |
c1d7ecf7 | 340 | int ret; |
2da1610a MS |
341 | unsigned long flags; |
342 | ||
343 | spin_lock_irqsave(&m->lock, flags); | |
c1d7ecf7 | 344 | ret = __pg_init_all_paths(m); |
2da1610a | 345 | spin_unlock_irqrestore(&m->lock, flags); |
c1d7ecf7 BVA |
346 | |
347 | return ret; | |
2da1610a MS |
348 | } |
349 | ||
350 | static void __switch_pg(struct multipath *m, struct priority_group *pg) | |
351 | { | |
69cea0d4 MS |
352 | lockdep_assert_held(&m->lock); |
353 | ||
2da1610a | 354 | m->current_pg = pg; |
1da177e4 LT |
355 | |
356 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 357 | if (m->hw_handler_name) { |
518257b1 MS |
358 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
359 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 360 | } else { |
518257b1 MS |
361 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
362 | clear_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 363 | } |
c9e45581 | 364 | |
91e968aa | 365 | atomic_set(&m->pg_init_count, 0); |
1da177e4 LT |
366 | } |
367 | ||
2da1610a MS |
368 | static struct pgpath *choose_path_in_pg(struct multipath *m, |
369 | struct priority_group *pg, | |
370 | size_t nr_bytes) | |
1da177e4 | 371 | { |
2da1610a | 372 | unsigned long flags; |
c922d5f7 | 373 | struct dm_path *path; |
2da1610a | 374 | struct pgpath *pgpath; |
1da177e4 | 375 | |
90a4323c | 376 | path = pg->ps.type->select_path(&pg->ps, nr_bytes); |
1da177e4 | 377 | if (!path) |
2da1610a | 378 | return ERR_PTR(-ENXIO); |
1da177e4 | 379 | |
2da1610a | 380 | pgpath = path_to_pgpath(path); |
1da177e4 | 381 | |
506458ef | 382 | if (unlikely(READ_ONCE(m->current_pg) != pg)) { |
2da1610a MS |
383 | /* Only update current_pgpath if pg changed */ |
384 | spin_lock_irqsave(&m->lock, flags); | |
385 | m->current_pgpath = pgpath; | |
386 | __switch_pg(m, pg); | |
387 | spin_unlock_irqrestore(&m->lock, flags); | |
388 | } | |
1da177e4 | 389 | |
2da1610a | 390 | return pgpath; |
1da177e4 LT |
391 | } |
392 | ||
2da1610a | 393 | static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 | 394 | { |
2da1610a | 395 | unsigned long flags; |
1da177e4 | 396 | struct priority_group *pg; |
2da1610a | 397 | struct pgpath *pgpath; |
d19a55cc | 398 | unsigned bypassed = 1; |
1da177e4 | 399 | |
91e968aa | 400 | if (!atomic_read(&m->nr_valid_paths)) { |
69cea0d4 | 401 | spin_lock_irqsave(&m->lock, flags); |
8d47e659 | 402 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
69cea0d4 | 403 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 404 | goto failed; |
1f271972 | 405 | } |
1da177e4 LT |
406 | |
407 | /* Were we instructed to switch PG? */ | |
506458ef | 408 | if (READ_ONCE(m->next_pg)) { |
2da1610a | 409 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 | 410 | pg = m->next_pg; |
2da1610a MS |
411 | if (!pg) { |
412 | spin_unlock_irqrestore(&m->lock, flags); | |
413 | goto check_current_pg; | |
414 | } | |
1da177e4 | 415 | m->next_pg = NULL; |
2da1610a MS |
416 | spin_unlock_irqrestore(&m->lock, flags); |
417 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
418 | if (!IS_ERR_OR_NULL(pgpath)) | |
419 | return pgpath; | |
1da177e4 LT |
420 | } |
421 | ||
422 | /* Don't change PG until it has no remaining paths */ | |
2da1610a | 423 | check_current_pg: |
506458ef | 424 | pg = READ_ONCE(m->current_pg); |
2da1610a MS |
425 | if (pg) { |
426 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
427 | if (!IS_ERR_OR_NULL(pgpath)) | |
428 | return pgpath; | |
429 | } | |
1da177e4 LT |
430 | |
431 | /* | |
432 | * Loop through priority groups until we find a valid path. | |
433 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
434 | * Second time we only try the ones we skipped, but set |
435 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
436 | */ |
437 | do { | |
438 | list_for_each_entry(pg, &m->priority_groups, list) { | |
d19a55cc | 439 | if (pg->bypassed == !!bypassed) |
1da177e4 | 440 | continue; |
2da1610a MS |
441 | pgpath = choose_path_in_pg(m, pg, nr_bytes); |
442 | if (!IS_ERR_OR_NULL(pgpath)) { | |
69cea0d4 MS |
443 | if (!bypassed) { |
444 | spin_lock_irqsave(&m->lock, flags); | |
518257b1 | 445 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); |
69cea0d4 MS |
446 | spin_unlock_irqrestore(&m->lock, flags); |
447 | } | |
2da1610a | 448 | return pgpath; |
f220fd4e | 449 | } |
1da177e4 LT |
450 | } |
451 | } while (bypassed--); | |
452 | ||
453 | failed: | |
2da1610a | 454 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
455 | m->current_pgpath = NULL; |
456 | m->current_pg = NULL; | |
2da1610a MS |
457 | spin_unlock_irqrestore(&m->lock, flags); |
458 | ||
459 | return NULL; | |
1da177e4 LT |
460 | } |
461 | ||
45e15720 | 462 | /* |
ac75b09f | 463 | * dm_report_EIO() is a macro instead of a function to make pr_debug_ratelimited() |
86331f39 BVA |
464 | * report the function name and line number of the function from which |
465 | * it has been invoked. | |
45e15720 | 466 | */ |
86331f39 | 467 | #define dm_report_EIO(m) \ |
18a482f5 | 468 | do { \ |
86331f39 BVA |
469 | struct mapped_device *md = dm_table_get_md((m)->ti->table); \ |
470 | \ | |
ac75b09f MS |
471 | DMDEBUG_LIMIT("%s: returning EIO; QIFNP = %d; SQIFNP = %d; DNFS = %d", \ |
472 | dm_device_name(md), \ | |
473 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
474 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
475 | dm_noflush_suspending((m)->ti)); \ | |
18a482f5 | 476 | } while (0) |
45e15720 | 477 | |
c1fd0abe MS |
478 | /* |
479 | * Check whether bios must be queued in the device-mapper core rather | |
480 | * than here in the target. | |
c1fd0abe | 481 | */ |
a862e4e2 | 482 | static bool __must_push_back(struct multipath *m) |
c1fd0abe | 483 | { |
a862e4e2 | 484 | return dm_noflush_suspending(m->ti); |
c1fd0abe MS |
485 | } |
486 | ||
c1fd0abe MS |
487 | static bool must_push_back_rq(struct multipath *m) |
488 | { | |
73265f3f MS |
489 | unsigned long flags; |
490 | bool ret; | |
491 | ||
492 | spin_lock_irqsave(&m->lock, flags); | |
493 | ret = (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) || __must_push_back(m)); | |
494 | spin_unlock_irqrestore(&m->lock, flags); | |
495 | ||
496 | return ret; | |
c1fd0abe MS |
497 | } |
498 | ||
36fcffcc | 499 | /* |
76e33fe4 | 500 | * Map cloned requests (request-based multipath) |
36fcffcc | 501 | */ |
eb8db831 CH |
502 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, |
503 | union map_info *map_context, | |
504 | struct request **__clone) | |
1da177e4 | 505 | { |
7943bd6d | 506 | struct multipath *m = ti->private; |
eb8db831 | 507 | size_t nr_bytes = blk_rq_bytes(rq); |
1da177e4 | 508 | struct pgpath *pgpath; |
f40c67f0 | 509 | struct block_device *bdev; |
eb8db831 | 510 | struct dm_mpath_io *mpio = get_mpio(map_context); |
7083abbb | 511 | struct request_queue *q; |
eb8db831 | 512 | struct request *clone; |
1da177e4 | 513 | |
1da177e4 | 514 | /* Do we need to select a new pgpath? */ |
506458ef | 515 | pgpath = READ_ONCE(m->current_pgpath); |
374117ad | 516 | if (!pgpath || !mpath_double_check_test_bit(MPATHF_QUEUE_IO, m)) |
2da1610a | 517 | pgpath = choose_pgpath(m, nr_bytes); |
1da177e4 | 518 | |
9bf59a61 | 519 | if (!pgpath) { |
c1fd0abe | 520 | if (must_push_back_rq(m)) |
b88efd43 | 521 | return DM_MAPIO_DELAY_REQUEUE; |
18a482f5 | 522 | dm_report_EIO(m); /* Failed */ |
f98e0eb6 | 523 | return DM_MAPIO_KILL; |
374117ad MS |
524 | } else if (mpath_double_check_test_bit(MPATHF_QUEUE_IO, m) || |
525 | mpath_double_check_test_bit(MPATHF_PG_INIT_REQUIRED, m)) { | |
459b5401 ML |
526 | pg_init_all_paths(m); |
527 | return DM_MAPIO_DELAY_REQUEUE; | |
9bf59a61 | 528 | } |
6afbc01d | 529 | |
2eb6e1e3 KB |
530 | mpio->pgpath = pgpath; |
531 | mpio->nr_bytes = nr_bytes; | |
532 | ||
9bf59a61 | 533 | bdev = pgpath->path.dev->bdev; |
7083abbb | 534 | q = bdev_get_queue(bdev); |
ff005a06 CH |
535 | clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, |
536 | BLK_MQ_REQ_NOWAIT); | |
eb8db831 CH |
537 | if (IS_ERR(clone)) { |
538 | /* EBUSY, ENODEV or EWOULDBLOCK: requeue */ | |
848b8aef | 539 | if (blk_queue_dying(q)) { |
7083abbb BVA |
540 | atomic_inc(&m->pg_init_in_progress); |
541 | activate_or_offline_path(pgpath); | |
050af08f | 542 | return DM_MAPIO_DELAY_REQUEUE; |
7083abbb | 543 | } |
050af08f ML |
544 | |
545 | /* | |
546 | * blk-mq's SCHED_RESTART can cover this requeue, so we | |
547 | * needn't deal with it by DELAY_REQUEUE. More importantly, | |
548 | * we have to return DM_MAPIO_REQUEUE so that blk-mq can | |
549 | * get the queue busy feedback (via BLK_STS_RESOURCE), | |
550 | * otherwise I/O merging can suffer. | |
551 | */ | |
6a23e05c | 552 | return DM_MAPIO_REQUEUE; |
e5863d9a | 553 | } |
eb8db831 CH |
554 | clone->bio = clone->biotail = NULL; |
555 | clone->rq_disk = bdev->bd_disk; | |
556 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
557 | *__clone = clone; | |
e5863d9a | 558 | |
9bf59a61 MS |
559 | if (pgpath->pg->ps.type->start_io) |
560 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
561 | &pgpath->path, | |
562 | nr_bytes); | |
2eb6e1e3 | 563 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
564 | } |
565 | ||
5de719e3 YY |
566 | static void multipath_release_clone(struct request *clone, |
567 | union map_info *map_context) | |
e5863d9a | 568 | { |
5de719e3 YY |
569 | if (unlikely(map_context)) { |
570 | /* | |
571 | * non-NULL map_context means caller is still map | |
572 | * method; must undo multipath_clone_and_map() | |
573 | */ | |
574 | struct dm_mpath_io *mpio = get_mpio(map_context); | |
575 | struct pgpath *pgpath = mpio->pgpath; | |
576 | ||
577 | if (pgpath && pgpath->pg->ps.type->end_io) | |
578 | pgpath->pg->ps.type->end_io(&pgpath->pg->ps, | |
579 | &pgpath->path, | |
087615bf GKB |
580 | mpio->nr_bytes, |
581 | clone->io_start_time_ns); | |
5de719e3 YY |
582 | } |
583 | ||
eb8db831 | 584 | blk_put_request(clone); |
e5863d9a MS |
585 | } |
586 | ||
76e33fe4 MS |
587 | /* |
588 | * Map cloned bios (bio-based multipath) | |
589 | */ | |
0001ec56 | 590 | |
17213ec1 | 591 | static void __multipath_queue_bio(struct multipath *m, struct bio *bio) |
f45f1186 | 592 | { |
f45f1186 | 593 | /* Queue for the daemon to resubmit */ |
f45f1186 MS |
594 | bio_list_add(&m->queued_bios, bio); |
595 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) | |
596 | queue_work(kmultipathd, &m->process_queued_bios); | |
17213ec1 MS |
597 | } |
598 | ||
599 | static void multipath_queue_bio(struct multipath *m, struct bio *bio) | |
600 | { | |
601 | unsigned long flags; | |
602 | ||
603 | spin_lock_irqsave(&m->lock, flags); | |
604 | __multipath_queue_bio(m, bio); | |
f45f1186 MS |
605 | spin_unlock_irqrestore(&m->lock, flags); |
606 | } | |
607 | ||
0001ec56 | 608 | static struct pgpath *__map_bio(struct multipath *m, struct bio *bio) |
76e33fe4 | 609 | { |
76e33fe4 MS |
610 | struct pgpath *pgpath; |
611 | unsigned long flags; | |
76e33fe4 MS |
612 | |
613 | /* Do we need to select a new pgpath? */ | |
506458ef | 614 | pgpath = READ_ONCE(m->current_pgpath); |
374117ad | 615 | if (!pgpath || !mpath_double_check_test_bit(MPATHF_QUEUE_IO, m)) |
0001ec56 | 616 | pgpath = choose_pgpath(m, bio->bi_iter.bi_size); |
76e33fe4 | 617 | |
17213ec1 | 618 | if (!pgpath) { |
76e33fe4 | 619 | spin_lock_irqsave(&m->lock, flags); |
17213ec1 MS |
620 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { |
621 | __multipath_queue_bio(m, bio); | |
622 | pgpath = ERR_PTR(-EAGAIN); | |
623 | } | |
76e33fe4 | 624 | spin_unlock_irqrestore(&m->lock, flags); |
848b8aef | 625 | |
374117ad MS |
626 | } else if (mpath_double_check_test_bit(MPATHF_QUEUE_IO, m) || |
627 | mpath_double_check_test_bit(MPATHF_PG_INIT_REQUIRED, m)) { | |
f45f1186 | 628 | multipath_queue_bio(m, bio); |
17213ec1 | 629 | pg_init_all_paths(m); |
0001ec56 | 630 | return ERR_PTR(-EAGAIN); |
76e33fe4 MS |
631 | } |
632 | ||
0001ec56 MS |
633 | return pgpath; |
634 | } | |
635 | ||
0001ec56 MS |
636 | static int __multipath_map_bio(struct multipath *m, struct bio *bio, |
637 | struct dm_mpath_io *mpio) | |
638 | { | |
dbaf971c | 639 | struct pgpath *pgpath = __map_bio(m, bio); |
848b8aef | 640 | |
0001ec56 | 641 | if (IS_ERR(pgpath)) |
76e33fe4 | 642 | return DM_MAPIO_SUBMITTED; |
76e33fe4 MS |
643 | |
644 | if (!pgpath) { | |
a862e4e2 | 645 | if (__must_push_back(m)) |
ca5beb76 | 646 | return DM_MAPIO_REQUEUE; |
18a482f5 | 647 | dm_report_EIO(m); |
846785e6 | 648 | return DM_MAPIO_KILL; |
76e33fe4 MS |
649 | } |
650 | ||
651 | mpio->pgpath = pgpath; | |
76e33fe4 | 652 | |
4e4cbee9 | 653 | bio->bi_status = 0; |
74d46992 | 654 | bio_set_dev(bio, pgpath->path.dev->bdev); |
1eff9d32 | 655 | bio->bi_opf |= REQ_FAILFAST_TRANSPORT; |
76e33fe4 MS |
656 | |
657 | if (pgpath->pg->ps.type->start_io) | |
658 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
659 | &pgpath->path, | |
d0442f80 | 660 | mpio->nr_bytes); |
76e33fe4 MS |
661 | return DM_MAPIO_REMAPPED; |
662 | } | |
663 | ||
664 | static int multipath_map_bio(struct dm_target *ti, struct bio *bio) | |
665 | { | |
666 | struct multipath *m = ti->private; | |
bf661be1 MS |
667 | struct dm_mpath_io *mpio = NULL; |
668 | ||
63f6e6fd | 669 | multipath_init_per_bio_data(bio, &mpio); |
76e33fe4 MS |
670 | return __multipath_map_bio(m, bio, mpio); |
671 | } | |
672 | ||
7e48c768 | 673 | static void process_queued_io_list(struct multipath *m) |
76e33fe4 | 674 | { |
953923c0 | 675 | if (m->queue_mode == DM_TYPE_REQUEST_BASED) |
7e48c768 | 676 | dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table)); |
8d47e659 | 677 | else if (m->queue_mode == DM_TYPE_BIO_BASED) |
76e33fe4 MS |
678 | queue_work(kmultipathd, &m->process_queued_bios); |
679 | } | |
680 | ||
681 | static void process_queued_bios(struct work_struct *work) | |
682 | { | |
683 | int r; | |
684 | unsigned long flags; | |
685 | struct bio *bio; | |
686 | struct bio_list bios; | |
687 | struct blk_plug plug; | |
688 | struct multipath *m = | |
689 | container_of(work, struct multipath, process_queued_bios); | |
690 | ||
691 | bio_list_init(&bios); | |
692 | ||
693 | spin_lock_irqsave(&m->lock, flags); | |
694 | ||
695 | if (bio_list_empty(&m->queued_bios)) { | |
696 | spin_unlock_irqrestore(&m->lock, flags); | |
697 | return; | |
698 | } | |
699 | ||
700 | bio_list_merge(&bios, &m->queued_bios); | |
701 | bio_list_init(&m->queued_bios); | |
702 | ||
703 | spin_unlock_irqrestore(&m->lock, flags); | |
704 | ||
705 | blk_start_plug(&plug); | |
706 | while ((bio = bio_list_pop(&bios))) { | |
1836df08 MS |
707 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); |
708 | dm_bio_restore(get_bio_details_from_mpio(mpio), bio); | |
709 | r = __multipath_map_bio(m, bio, mpio); | |
846785e6 CH |
710 | switch (r) { |
711 | case DM_MAPIO_KILL: | |
4e4cbee9 CH |
712 | bio->bi_status = BLK_STS_IOERR; |
713 | bio_endio(bio); | |
047385b3 | 714 | break; |
846785e6 | 715 | case DM_MAPIO_REQUEUE: |
4e4cbee9 | 716 | bio->bi_status = BLK_STS_DM_REQUEUE; |
76e33fe4 | 717 | bio_endio(bio); |
846785e6 CH |
718 | break; |
719 | case DM_MAPIO_REMAPPED: | |
ed00aabd | 720 | submit_bio_noacct(bio); |
846785e6 | 721 | break; |
8192a0cd | 722 | case DM_MAPIO_SUBMITTED: |
9157c8d3 BVA |
723 | break; |
724 | default: | |
725 | WARN_ONCE(true, "__multipath_map_bio() returned %d\n", r); | |
846785e6 | 726 | } |
76e33fe4 MS |
727 | } |
728 | blk_finish_plug(&plug); | |
729 | } | |
730 | ||
1da177e4 LT |
731 | /* |
732 | * If we run out of usable paths, should we queue I/O or error it? | |
733 | */ | |
be7d31cc | 734 | static int queue_if_no_path(struct multipath *m, bool queue_if_no_path, |
4c3f4838 | 735 | bool save_old_value, const char *caller) |
1da177e4 LT |
736 | { |
737 | unsigned long flags; | |
553ec94c | 738 | bool queue_if_no_path_bit, saved_queue_if_no_path_bit; |
4c3f4838 MS |
739 | const char *dm_dev_name = dm_device_name(dm_table_get_md(m->ti->table)); |
740 | ||
741 | DMDEBUG("%s: %s caller=%s queue_if_no_path=%d save_old_value=%d", | |
742 | dm_dev_name, __func__, caller, queue_if_no_path, save_old_value); | |
1da177e4 LT |
743 | |
744 | spin_lock_irqsave(&m->lock, flags); | |
553ec94c MS |
745 | |
746 | queue_if_no_path_bit = test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
747 | saved_queue_if_no_path_bit = test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
748 | ||
749 | if (save_old_value) { | |
750 | if (unlikely(!queue_if_no_path_bit && saved_queue_if_no_path_bit)) { | |
751 | DMERR("%s: QIFNP disabled but saved as enabled, saving again loses state, not saving!", | |
4c3f4838 | 752 | dm_dev_name); |
553ec94c MS |
753 | } else |
754 | assign_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags, queue_if_no_path_bit); | |
755 | } else if (!queue_if_no_path && saved_queue_if_no_path_bit) { | |
756 | /* due to "fail_if_no_path" message, need to honor it. */ | |
757 | clear_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
758 | } | |
c1fd0abe | 759 | assign_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags, queue_if_no_path); |
553ec94c | 760 | |
4c3f4838 MS |
761 | DMDEBUG("%s: after %s changes; QIFNP = %d; SQIFNP = %d; DNFS = %d", |
762 | dm_dev_name, __func__, | |
763 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags), | |
764 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags), | |
765 | dm_noflush_suspending(m->ti)); | |
766 | ||
1da177e4 LT |
767 | spin_unlock_irqrestore(&m->lock, flags); |
768 | ||
76e33fe4 | 769 | if (!queue_if_no_path) { |
63d832c3 | 770 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 771 | process_queued_io_list(m); |
76e33fe4 | 772 | } |
63d832c3 | 773 | |
1da177e4 LT |
774 | return 0; |
775 | } | |
776 | ||
be240ff5 AP |
777 | /* |
778 | * If the queue_if_no_path timeout fires, turn off queue_if_no_path and | |
779 | * process any queued I/O. | |
780 | */ | |
781 | static void queue_if_no_path_timeout_work(struct timer_list *t) | |
782 | { | |
783 | struct multipath *m = from_timer(m, t, nopath_timer); | |
784 | struct mapped_device *md = dm_table_get_md(m->ti->table); | |
785 | ||
786 | DMWARN("queue_if_no_path timeout on %s, failing queued IO", dm_device_name(md)); | |
4c3f4838 | 787 | queue_if_no_path(m, false, false, __func__); |
be240ff5 AP |
788 | } |
789 | ||
790 | /* | |
791 | * Enable the queue_if_no_path timeout if necessary. | |
792 | * Called with m->lock held. | |
793 | */ | |
794 | static void enable_nopath_timeout(struct multipath *m) | |
795 | { | |
796 | unsigned long queue_if_no_path_timeout = | |
797 | READ_ONCE(queue_if_no_path_timeout_secs) * HZ; | |
798 | ||
799 | lockdep_assert_held(&m->lock); | |
800 | ||
801 | if (queue_if_no_path_timeout > 0 && | |
802 | atomic_read(&m->nr_valid_paths) == 0 && | |
803 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
804 | mod_timer(&m->nopath_timer, | |
805 | jiffies + queue_if_no_path_timeout); | |
806 | } | |
807 | } | |
808 | ||
809 | static void disable_nopath_timeout(struct multipath *m) | |
810 | { | |
811 | del_timer_sync(&m->nopath_timer); | |
812 | } | |
813 | ||
1da177e4 LT |
814 | /* |
815 | * An event is triggered whenever a path is taken out of use. | |
816 | * Includes path failure and PG bypass. | |
817 | */ | |
c4028958 | 818 | static void trigger_event(struct work_struct *work) |
1da177e4 | 819 | { |
c4028958 DH |
820 | struct multipath *m = |
821 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
822 | |
823 | dm_table_event(m->ti->table); | |
824 | } | |
825 | ||
826 | /*----------------------------------------------------------------- | |
827 | * Constructor/argument parsing: | |
828 | * <#multipath feature args> [<arg>]* | |
829 | * <#hw_handler args> [hw_handler [<arg>]*] | |
830 | * <#priority groups> | |
831 | * <initial priority group> | |
832 | * [<selector> <#selector args> [<arg>]* | |
833 | * <#paths> <#per-path selector args> | |
834 | * [<path> [<arg>]* ]+ ]+ | |
835 | *---------------------------------------------------------------*/ | |
498f0103 | 836 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
837 | struct dm_target *ti) |
838 | { | |
839 | int r; | |
840 | struct path_selector_type *pst; | |
841 | unsigned ps_argc; | |
842 | ||
5916a22b | 843 | static const struct dm_arg _args[] = { |
72d94861 | 844 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
845 | }; |
846 | ||
498f0103 | 847 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 848 | if (!pst) { |
72d94861 | 849 | ti->error = "unknown path selector type"; |
1da177e4 LT |
850 | return -EINVAL; |
851 | } | |
852 | ||
498f0103 | 853 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
854 | if (r) { |
855 | dm_put_path_selector(pst); | |
1da177e4 | 856 | return -EINVAL; |
371b2e34 | 857 | } |
1da177e4 LT |
858 | |
859 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
860 | if (r) { | |
861 | dm_put_path_selector(pst); | |
72d94861 | 862 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
863 | return r; |
864 | } | |
865 | ||
866 | pg->ps.type = pst; | |
498f0103 | 867 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
868 | |
869 | return 0; | |
870 | } | |
871 | ||
e8f74a0f | 872 | static int setup_scsi_dh(struct block_device *bdev, struct multipath *m, |
b592211c | 873 | const char **attached_handler_name, char **error) |
1da177e4 | 874 | { |
848b8aef | 875 | struct request_queue *q = bdev_get_queue(bdev); |
848b8aef | 876 | int r; |
a58a935d | 877 | |
374117ad | 878 | if (mpath_double_check_test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, m)) { |
1bab0de0 | 879 | retain: |
b592211c | 880 | if (*attached_handler_name) { |
54cd640d | 881 | /* |
882 | * Clear any hw_handler_params associated with a | |
883 | * handler that isn't already attached. | |
884 | */ | |
b592211c | 885 | if (m->hw_handler_name && strcmp(*attached_handler_name, m->hw_handler_name)) { |
54cd640d | 886 | kfree(m->hw_handler_params); |
887 | m->hw_handler_params = NULL; | |
888 | } | |
889 | ||
a58a935d MS |
890 | /* |
891 | * Reset hw_handler_name to match the attached handler | |
a58a935d MS |
892 | * |
893 | * NB. This modifies the table line to show the actual | |
894 | * handler instead of the original table passed in. | |
895 | */ | |
896 | kfree(m->hw_handler_name); | |
b592211c MS |
897 | m->hw_handler_name = *attached_handler_name; |
898 | *attached_handler_name = NULL; | |
a58a935d MS |
899 | } |
900 | } | |
a0cf7ea9 | 901 | |
a58a935d | 902 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
903 | r = scsi_dh_attach(q, m->hw_handler_name); |
904 | if (r == -EBUSY) { | |
1bab0de0 | 905 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 906 | |
1bab0de0 | 907 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
848b8aef | 908 | bdevname(bdev, b)); |
1bab0de0 CH |
909 | goto retain; |
910 | } | |
ae11b1b3 | 911 | if (r < 0) { |
848b8aef MS |
912 | *error = "error attaching hardware handler"; |
913 | return r; | |
ae11b1b3 | 914 | } |
2bfd2e13 CS |
915 | |
916 | if (m->hw_handler_params) { | |
917 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
918 | if (r < 0) { | |
848b8aef MS |
919 | *error = "unable to set hardware handler parameters"; |
920 | return r; | |
2bfd2e13 CS |
921 | } |
922 | } | |
ae11b1b3 HR |
923 | } |
924 | ||
848b8aef MS |
925 | return 0; |
926 | } | |
927 | ||
928 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, | |
929 | struct dm_target *ti) | |
930 | { | |
931 | int r; | |
932 | struct pgpath *p; | |
933 | struct multipath *m = ti->private; | |
e8f74a0f | 934 | struct request_queue *q; |
b592211c | 935 | const char *attached_handler_name = NULL; |
848b8aef MS |
936 | |
937 | /* we need at least a path arg */ | |
938 | if (as->argc < 1) { | |
939 | ti->error = "no device given"; | |
940 | return ERR_PTR(-EINVAL); | |
941 | } | |
942 | ||
943 | p = alloc_pgpath(); | |
944 | if (!p) | |
945 | return ERR_PTR(-ENOMEM); | |
946 | ||
947 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), | |
948 | &p->path.dev); | |
949 | if (r) { | |
950 | ti->error = "error getting device"; | |
951 | goto bad; | |
952 | } | |
953 | ||
e8f74a0f MS |
954 | q = bdev_get_queue(p->path.dev->bdev); |
955 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); | |
e457edf0 | 956 | if (attached_handler_name || m->hw_handler_name) { |
848b8aef | 957 | INIT_DELAYED_WORK(&p->activate_path, activate_path_work); |
b592211c | 958 | r = setup_scsi_dh(p->path.dev->bdev, m, &attached_handler_name, &ti->error); |
940bc471 | 959 | kfree(attached_handler_name); |
848b8aef MS |
960 | if (r) { |
961 | dm_put_device(ti, p->path.dev); | |
962 | goto bad; | |
963 | } | |
964 | } | |
965 | ||
1da177e4 LT |
966 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
967 | if (r) { | |
968 | dm_put_device(ti, p->path.dev); | |
969 | goto bad; | |
970 | } | |
971 | ||
972 | return p; | |
1da177e4 LT |
973 | bad: |
974 | free_pgpath(p); | |
01460f35 | 975 | return ERR_PTR(r); |
1da177e4 LT |
976 | } |
977 | ||
498f0103 | 978 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 979 | struct multipath *m) |
1da177e4 | 980 | { |
5916a22b | 981 | static const struct dm_arg _args[] = { |
72d94861 AK |
982 | {1, 1024, "invalid number of paths"}, |
983 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
984 | }; |
985 | ||
986 | int r; | |
498f0103 | 987 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 988 | struct priority_group *pg; |
28f16c20 | 989 | struct dm_target *ti = m->ti; |
1da177e4 LT |
990 | |
991 | if (as->argc < 2) { | |
992 | as->argc = 0; | |
01460f35 BM |
993 | ti->error = "not enough priority group arguments"; |
994 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
995 | } |
996 | ||
997 | pg = alloc_priority_group(); | |
998 | if (!pg) { | |
72d94861 | 999 | ti->error = "couldn't allocate priority group"; |
01460f35 | 1000 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1001 | } |
1002 | pg->m = m; | |
1003 | ||
1004 | r = parse_path_selector(as, pg, ti); | |
1005 | if (r) | |
1006 | goto bad; | |
1007 | ||
1008 | /* | |
1009 | * read the paths | |
1010 | */ | |
498f0103 | 1011 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
1012 | if (r) |
1013 | goto bad; | |
1014 | ||
498f0103 | 1015 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
1016 | if (r) |
1017 | goto bad; | |
1018 | ||
498f0103 | 1019 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
1020 | for (i = 0; i < pg->nr_pgpaths; i++) { |
1021 | struct pgpath *pgpath; | |
498f0103 | 1022 | struct dm_arg_set path_args; |
1da177e4 | 1023 | |
498f0103 | 1024 | if (as->argc < nr_args) { |
148acff6 | 1025 | ti->error = "not enough path parameters"; |
6bbf79a1 | 1026 | r = -EINVAL; |
1da177e4 | 1027 | goto bad; |
148acff6 | 1028 | } |
1da177e4 | 1029 | |
498f0103 | 1030 | path_args.argc = nr_args; |
1da177e4 LT |
1031 | path_args.argv = as->argv; |
1032 | ||
1033 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
1034 | if (IS_ERR(pgpath)) { |
1035 | r = PTR_ERR(pgpath); | |
1da177e4 | 1036 | goto bad; |
01460f35 | 1037 | } |
1da177e4 LT |
1038 | |
1039 | pgpath->pg = pg; | |
1040 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 1041 | dm_consume_args(as, nr_args); |
1da177e4 LT |
1042 | } |
1043 | ||
1044 | return pg; | |
1045 | ||
1046 | bad: | |
1047 | free_priority_group(pg, ti); | |
01460f35 | 1048 | return ERR_PTR(r); |
1da177e4 LT |
1049 | } |
1050 | ||
498f0103 | 1051 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 1052 | { |
1da177e4 | 1053 | unsigned hw_argc; |
2bfd2e13 | 1054 | int ret; |
28f16c20 | 1055 | struct dm_target *ti = m->ti; |
1da177e4 | 1056 | |
5916a22b | 1057 | static const struct dm_arg _args[] = { |
72d94861 | 1058 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
1059 | }; |
1060 | ||
498f0103 | 1061 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
1062 | return -EINVAL; |
1063 | ||
1064 | if (!hw_argc) | |
1065 | return 0; | |
1066 | ||
8d47e659 | 1067 | if (m->queue_mode == DM_TYPE_BIO_BASED) { |
76e33fe4 MS |
1068 | dm_consume_args(as, hw_argc); |
1069 | DMERR("bio-based multipath doesn't allow hardware handler args"); | |
1070 | return 0; | |
1071 | } | |
1072 | ||
498f0103 | 1073 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
f97dc421 | 1074 | if (!m->hw_handler_name) |
1075 | return -EINVAL; | |
14e98c5c | 1076 | |
2bfd2e13 CS |
1077 | if (hw_argc > 1) { |
1078 | char *p; | |
1079 | int i, j, len = 4; | |
1080 | ||
1081 | for (i = 0; i <= hw_argc - 2; i++) | |
1082 | len += strlen(as->argv[i]) + 1; | |
1083 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
1084 | if (!p) { | |
1085 | ti->error = "memory allocation failed"; | |
1086 | ret = -ENOMEM; | |
1087 | goto fail; | |
1088 | } | |
1089 | j = sprintf(p, "%d", hw_argc - 1); | |
1090 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
1091 | j = sprintf(p, "%s", as->argv[i]); | |
1092 | } | |
498f0103 | 1093 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
1094 | |
1095 | return 0; | |
2bfd2e13 CS |
1096 | fail: |
1097 | kfree(m->hw_handler_name); | |
1098 | m->hw_handler_name = NULL; | |
1099 | return ret; | |
1da177e4 LT |
1100 | } |
1101 | ||
498f0103 | 1102 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
1103 | { |
1104 | int r; | |
1105 | unsigned argc; | |
28f16c20 | 1106 | struct dm_target *ti = m->ti; |
498f0103 | 1107 | const char *arg_name; |
1da177e4 | 1108 | |
5916a22b | 1109 | static const struct dm_arg _args[] = { |
e83068a5 | 1110 | {0, 8, "invalid number of feature args"}, |
c9e45581 | 1111 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 1112 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
1113 | }; |
1114 | ||
498f0103 | 1115 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
1116 | if (r) |
1117 | return -EINVAL; | |
1118 | ||
1119 | if (!argc) | |
1120 | return 0; | |
1121 | ||
c9e45581 | 1122 | do { |
498f0103 | 1123 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
1124 | argc--; |
1125 | ||
498f0103 | 1126 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
4c3f4838 | 1127 | r = queue_if_no_path(m, true, false, __func__); |
c9e45581 DW |
1128 | continue; |
1129 | } | |
1130 | ||
a58a935d | 1131 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
518257b1 | 1132 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); |
a58a935d MS |
1133 | continue; |
1134 | } | |
1135 | ||
498f0103 | 1136 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 1137 | (argc >= 1)) { |
498f0103 | 1138 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
1139 | argc--; |
1140 | continue; | |
1141 | } | |
1142 | ||
498f0103 | 1143 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 1144 | (argc >= 1)) { |
498f0103 | 1145 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
1146 | argc--; |
1147 | continue; | |
1148 | } | |
1149 | ||
e83068a5 MS |
1150 | if (!strcasecmp(arg_name, "queue_mode") && |
1151 | (argc >= 1)) { | |
1152 | const char *queue_mode_name = dm_shift_arg(as); | |
1153 | ||
1154 | if (!strcasecmp(queue_mode_name, "bio")) | |
1155 | m->queue_mode = DM_TYPE_BIO_BASED; | |
953923c0 MS |
1156 | else if (!strcasecmp(queue_mode_name, "rq") || |
1157 | !strcasecmp(queue_mode_name, "mq")) | |
e83068a5 | 1158 | m->queue_mode = DM_TYPE_REQUEST_BASED; |
e83068a5 MS |
1159 | else { |
1160 | ti->error = "Unknown 'queue_mode' requested"; | |
1161 | r = -EINVAL; | |
1162 | } | |
1163 | argc--; | |
1164 | continue; | |
1165 | } | |
1166 | ||
1da177e4 | 1167 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
1168 | r = -EINVAL; |
1169 | } while (argc && !r); | |
1170 | ||
1171 | return r; | |
1da177e4 LT |
1172 | } |
1173 | ||
e83068a5 | 1174 | static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv) |
1da177e4 | 1175 | { |
498f0103 | 1176 | /* target arguments */ |
5916a22b | 1177 | static const struct dm_arg _args[] = { |
a490a07a MS |
1178 | {0, 1024, "invalid number of priority groups"}, |
1179 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
1180 | }; |
1181 | ||
1182 | int r; | |
1183 | struct multipath *m; | |
498f0103 | 1184 | struct dm_arg_set as; |
1da177e4 LT |
1185 | unsigned pg_count = 0; |
1186 | unsigned next_pg_num; | |
be240ff5 | 1187 | unsigned long flags; |
1da177e4 LT |
1188 | |
1189 | as.argc = argc; | |
1190 | as.argv = argv; | |
1191 | ||
e83068a5 | 1192 | m = alloc_multipath(ti); |
1da177e4 | 1193 | if (!m) { |
72d94861 | 1194 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
1195 | return -EINVAL; |
1196 | } | |
1197 | ||
28f16c20 | 1198 | r = parse_features(&as, m); |
1da177e4 LT |
1199 | if (r) |
1200 | goto bad; | |
1201 | ||
e83068a5 MS |
1202 | r = alloc_multipath_stage2(ti, m); |
1203 | if (r) | |
1204 | goto bad; | |
1205 | ||
28f16c20 | 1206 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
1207 | if (r) |
1208 | goto bad; | |
1209 | ||
498f0103 | 1210 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
1211 | if (r) |
1212 | goto bad; | |
1213 | ||
498f0103 | 1214 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
1215 | if (r) |
1216 | goto bad; | |
1217 | ||
a490a07a MS |
1218 | if ((!m->nr_priority_groups && next_pg_num) || |
1219 | (m->nr_priority_groups && !next_pg_num)) { | |
1220 | ti->error = "invalid initial priority group"; | |
1221 | r = -EINVAL; | |
1222 | goto bad; | |
1223 | } | |
1224 | ||
1da177e4 LT |
1225 | /* parse the priority groups */ |
1226 | while (as.argc) { | |
1227 | struct priority_group *pg; | |
91e968aa | 1228 | unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths); |
1da177e4 | 1229 | |
28f16c20 | 1230 | pg = parse_priority_group(&as, m); |
01460f35 BM |
1231 | if (IS_ERR(pg)) { |
1232 | r = PTR_ERR(pg); | |
1da177e4 LT |
1233 | goto bad; |
1234 | } | |
1235 | ||
91e968aa MS |
1236 | nr_valid_paths += pg->nr_pgpaths; |
1237 | atomic_set(&m->nr_valid_paths, nr_valid_paths); | |
1238 | ||
1da177e4 LT |
1239 | list_add_tail(&pg->list, &m->priority_groups); |
1240 | pg_count++; | |
1241 | pg->pg_num = pg_count; | |
1242 | if (!--next_pg_num) | |
1243 | m->next_pg = pg; | |
1244 | } | |
1245 | ||
1246 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 1247 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
1248 | r = -EINVAL; |
1249 | goto bad; | |
1250 | } | |
1251 | ||
be240ff5 AP |
1252 | spin_lock_irqsave(&m->lock, flags); |
1253 | enable_nopath_timeout(m); | |
1254 | spin_unlock_irqrestore(&m->lock, flags); | |
1255 | ||
55a62eef AK |
1256 | ti->num_flush_bios = 1; |
1257 | ti->num_discard_bios = 1; | |
042bcef8 | 1258 | ti->num_write_same_bios = 1; |
ac62d620 | 1259 | ti->num_write_zeroes_bios = 1; |
8d47e659 | 1260 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
bf661be1 | 1261 | ti->per_io_data_size = multipath_per_bio_data_size(); |
eb8db831 | 1262 | else |
8637a6bf | 1263 | ti->per_io_data_size = sizeof(struct dm_mpath_io); |
8627921f | 1264 | |
1da177e4 LT |
1265 | return 0; |
1266 | ||
1267 | bad: | |
1268 | free_multipath(m); | |
1269 | return r; | |
1270 | } | |
1271 | ||
2bded7bd KU |
1272 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
1273 | { | |
9f4c3f87 | 1274 | DEFINE_WAIT(wait); |
2bded7bd KU |
1275 | |
1276 | while (1) { | |
9f4c3f87 | 1277 | prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE); |
2bded7bd | 1278 | |
91e968aa | 1279 | if (!atomic_read(&m->pg_init_in_progress)) |
2bded7bd | 1280 | break; |
2bded7bd KU |
1281 | |
1282 | io_schedule(); | |
1283 | } | |
9f4c3f87 | 1284 | finish_wait(&m->pg_init_wait, &wait); |
2bded7bd KU |
1285 | } |
1286 | ||
1287 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 1288 | { |
848b8aef MS |
1289 | if (m->hw_handler_name) { |
1290 | set_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1291 | smp_mb__after_atomic(); | |
1292 | ||
935fcc56 | 1293 | if (atomic_read(&m->pg_init_in_progress)) |
1294 | flush_workqueue(kmpath_handlerd); | |
848b8aef MS |
1295 | multipath_wait_for_pg_init_completion(m); |
1296 | ||
1297 | clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1298 | smp_mb__after_atomic(); | |
1299 | } | |
954a73d5 | 1300 | |
935fcc56 | 1301 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
1302 | flush_work(&m->process_queued_bios); | |
43829731 | 1303 | flush_work(&m->trigger_event); |
6df400ab KU |
1304 | } |
1305 | ||
1306 | static void multipath_dtr(struct dm_target *ti) | |
1307 | { | |
1308 | struct multipath *m = ti->private; | |
1309 | ||
be240ff5 | 1310 | disable_nopath_timeout(m); |
2bded7bd | 1311 | flush_multipath_work(m); |
1da177e4 LT |
1312 | free_multipath(m); |
1313 | } | |
1314 | ||
1da177e4 LT |
1315 | /* |
1316 | * Take a path out of use. | |
1317 | */ | |
1318 | static int fail_path(struct pgpath *pgpath) | |
1319 | { | |
1320 | unsigned long flags; | |
1321 | struct multipath *m = pgpath->pg->m; | |
1322 | ||
1323 | spin_lock_irqsave(&m->lock, flags); | |
1324 | ||
6680073d | 1325 | if (!pgpath->is_active) |
1da177e4 LT |
1326 | goto out; |
1327 | ||
04867370 MS |
1328 | DMWARN("%s: Failing path %s.", |
1329 | dm_device_name(dm_table_get_md(m->ti->table)), | |
1330 | pgpath->path.dev->name); | |
1da177e4 LT |
1331 | |
1332 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
be7d31cc | 1333 | pgpath->is_active = false; |
1da177e4 LT |
1334 | pgpath->fail_count++; |
1335 | ||
91e968aa | 1336 | atomic_dec(&m->nr_valid_paths); |
1da177e4 LT |
1337 | |
1338 | if (pgpath == m->current_pgpath) | |
1339 | m->current_pgpath = NULL; | |
1340 | ||
b15546f9 | 1341 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
91e968aa | 1342 | pgpath->path.dev->name, atomic_read(&m->nr_valid_paths)); |
b15546f9 | 1343 | |
fe9cf30e | 1344 | schedule_work(&m->trigger_event); |
1da177e4 | 1345 | |
be240ff5 AP |
1346 | enable_nopath_timeout(m); |
1347 | ||
1da177e4 LT |
1348 | out: |
1349 | spin_unlock_irqrestore(&m->lock, flags); | |
1350 | ||
1351 | return 0; | |
1352 | } | |
1353 | ||
1354 | /* | |
1355 | * Reinstate a previously-failed path | |
1356 | */ | |
1357 | static int reinstate_path(struct pgpath *pgpath) | |
1358 | { | |
63d832c3 | 1359 | int r = 0, run_queue = 0; |
1da177e4 LT |
1360 | unsigned long flags; |
1361 | struct multipath *m = pgpath->pg->m; | |
91e968aa | 1362 | unsigned nr_valid_paths; |
1da177e4 LT |
1363 | |
1364 | spin_lock_irqsave(&m->lock, flags); | |
1365 | ||
6680073d | 1366 | if (pgpath->is_active) |
1da177e4 LT |
1367 | goto out; |
1368 | ||
04867370 MS |
1369 | DMWARN("%s: Reinstating path %s.", |
1370 | dm_device_name(dm_table_get_md(m->ti->table)), | |
1371 | pgpath->path.dev->name); | |
1da177e4 LT |
1372 | |
1373 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1374 | if (r) | |
1375 | goto out; | |
1376 | ||
be7d31cc | 1377 | pgpath->is_active = true; |
1da177e4 | 1378 | |
91e968aa MS |
1379 | nr_valid_paths = atomic_inc_return(&m->nr_valid_paths); |
1380 | if (nr_valid_paths == 1) { | |
e54f77dd | 1381 | m->current_pgpath = NULL; |
63d832c3 | 1382 | run_queue = 1; |
e54f77dd | 1383 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1384 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
91e968aa | 1385 | atomic_inc(&m->pg_init_in_progress); |
e54f77dd | 1386 | } |
1da177e4 | 1387 | |
b15546f9 | 1388 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
91e968aa | 1389 | pgpath->path.dev->name, nr_valid_paths); |
b15546f9 | 1390 | |
fe9cf30e | 1391 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1392 | |
1393 | out: | |
1394 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 1395 | if (run_queue) { |
63d832c3 | 1396 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1397 | process_queued_io_list(m); |
76e33fe4 | 1398 | } |
1da177e4 | 1399 | |
be240ff5 AP |
1400 | if (pgpath->is_active) |
1401 | disable_nopath_timeout(m); | |
1402 | ||
1da177e4 LT |
1403 | return r; |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1408 | */ | |
1409 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1410 | action_fn action) | |
1411 | { | |
19040c0b | 1412 | int r = -EINVAL; |
1da177e4 LT |
1413 | struct pgpath *pgpath; |
1414 | struct priority_group *pg; | |
1415 | ||
1416 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1417 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1418 | if (pgpath->path.dev == dev) | |
1419 | r = action(pgpath); | |
1420 | } | |
1421 | } | |
1422 | ||
1423 | return r; | |
1424 | } | |
1425 | ||
1426 | /* | |
1427 | * Temporarily try to avoid having to use the specified PG | |
1428 | */ | |
1429 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
be7d31cc | 1430 | bool bypassed) |
1da177e4 LT |
1431 | { |
1432 | unsigned long flags; | |
1433 | ||
1434 | spin_lock_irqsave(&m->lock, flags); | |
1435 | ||
1436 | pg->bypassed = bypassed; | |
1437 | m->current_pgpath = NULL; | |
1438 | m->current_pg = NULL; | |
1439 | ||
1440 | spin_unlock_irqrestore(&m->lock, flags); | |
1441 | ||
fe9cf30e | 1442 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1443 | } |
1444 | ||
1445 | /* | |
1446 | * Switch to using the specified PG from the next I/O that gets mapped | |
1447 | */ | |
1448 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1449 | { | |
1450 | struct priority_group *pg; | |
1451 | unsigned pgnum; | |
1452 | unsigned long flags; | |
31998ef1 | 1453 | char dummy; |
1da177e4 | 1454 | |
31998ef1 | 1455 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1456 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1457 | DMWARN("invalid PG number supplied to switch_pg_num"); |
1458 | return -EINVAL; | |
1459 | } | |
1460 | ||
1461 | spin_lock_irqsave(&m->lock, flags); | |
1462 | list_for_each_entry(pg, &m->priority_groups, list) { | |
be7d31cc | 1463 | pg->bypassed = false; |
1da177e4 LT |
1464 | if (--pgnum) |
1465 | continue; | |
1466 | ||
1467 | m->current_pgpath = NULL; | |
1468 | m->current_pg = NULL; | |
1469 | m->next_pg = pg; | |
1470 | } | |
1471 | spin_unlock_irqrestore(&m->lock, flags); | |
1472 | ||
fe9cf30e | 1473 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1474 | return 0; |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * Set/clear bypassed status of a PG. | |
1479 | * PGs are numbered upwards from 1 in the order they were declared. | |
1480 | */ | |
be7d31cc | 1481 | static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed) |
1da177e4 LT |
1482 | { |
1483 | struct priority_group *pg; | |
1484 | unsigned pgnum; | |
31998ef1 | 1485 | char dummy; |
1da177e4 | 1486 | |
31998ef1 | 1487 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1488 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1489 | DMWARN("invalid PG number supplied to bypass_pg"); |
1490 | return -EINVAL; | |
1491 | } | |
1492 | ||
1493 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1494 | if (!--pgnum) | |
1495 | break; | |
1496 | } | |
1497 | ||
1498 | bypass_pg(m, pg, bypassed); | |
1499 | return 0; | |
1500 | } | |
1501 | ||
c9e45581 DW |
1502 | /* |
1503 | * Should we retry pg_init immediately? | |
1504 | */ | |
be7d31cc | 1505 | static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) |
c9e45581 DW |
1506 | { |
1507 | unsigned long flags; | |
be7d31cc | 1508 | bool limit_reached = false; |
c9e45581 DW |
1509 | |
1510 | spin_lock_irqsave(&m->lock, flags); | |
1511 | ||
91e968aa MS |
1512 | if (atomic_read(&m->pg_init_count) <= m->pg_init_retries && |
1513 | !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) | |
518257b1 | 1514 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
c9e45581 | 1515 | else |
be7d31cc | 1516 | limit_reached = true; |
c9e45581 DW |
1517 | |
1518 | spin_unlock_irqrestore(&m->lock, flags); | |
1519 | ||
1520 | return limit_reached; | |
1521 | } | |
1522 | ||
3ae31f6a | 1523 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1524 | { |
83c0d5d5 | 1525 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1526 | struct priority_group *pg = pgpath->pg; |
1527 | struct multipath *m = pg->m; | |
1528 | unsigned long flags; | |
be7d31cc | 1529 | bool delay_retry = false; |
cfae5c9b CS |
1530 | |
1531 | /* device or driver problems */ | |
1532 | switch (errors) { | |
1533 | case SCSI_DH_OK: | |
1534 | break; | |
1535 | case SCSI_DH_NOSYS: | |
1536 | if (!m->hw_handler_name) { | |
1537 | errors = 0; | |
1538 | break; | |
1539 | } | |
f7b934c8 MB |
1540 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1541 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1542 | /* |
1543 | * Fail path for now, so we do not ping pong | |
1544 | */ | |
1545 | fail_path(pgpath); | |
1546 | break; | |
1547 | case SCSI_DH_DEV_TEMP_BUSY: | |
1548 | /* | |
1549 | * Probably doing something like FW upgrade on the | |
1550 | * controller so try the other pg. | |
1551 | */ | |
be7d31cc | 1552 | bypass_pg(m, pg, true); |
cfae5c9b | 1553 | break; |
cfae5c9b | 1554 | case SCSI_DH_RETRY: |
4e2d19e4 | 1555 | /* Wait before retrying. */ |
4ecc5081 | 1556 | delay_retry = true; |
df561f66 | 1557 | fallthrough; |
cfae5c9b CS |
1558 | case SCSI_DH_IMM_RETRY: |
1559 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1560 | if (pg_init_limit_reached(m, pgpath)) | |
1561 | fail_path(pgpath); | |
1562 | errors = 0; | |
1563 | break; | |
ec31f3f7 | 1564 | case SCSI_DH_DEV_OFFLINED: |
cfae5c9b CS |
1565 | default: |
1566 | /* | |
1567 | * We probably do not want to fail the path for a device | |
1568 | * error, but this is what the old dm did. In future | |
1569 | * patches we can do more advanced handling. | |
1570 | */ | |
1571 | fail_path(pgpath); | |
1572 | } | |
1573 | ||
1574 | spin_lock_irqsave(&m->lock, flags); | |
1575 | if (errors) { | |
e54f77dd CS |
1576 | if (pgpath == m->current_pgpath) { |
1577 | DMERR("Could not failover device. Error %d.", errors); | |
1578 | m->current_pgpath = NULL; | |
1579 | m->current_pg = NULL; | |
1580 | } | |
518257b1 | 1581 | } else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
be7d31cc | 1582 | pg->bypassed = false; |
cfae5c9b | 1583 | |
91e968aa | 1584 | if (atomic_dec_return(&m->pg_init_in_progress) > 0) |
d0259bf0 KU |
1585 | /* Activations of other paths are still on going */ |
1586 | goto out; | |
1587 | ||
518257b1 MS |
1588 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { |
1589 | if (delay_retry) | |
1590 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1591 | else | |
1592 | clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1593 | ||
3e9f1be1 HR |
1594 | if (__pg_init_all_paths(m)) |
1595 | goto out; | |
1596 | } | |
518257b1 | 1597 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
d0259bf0 | 1598 | |
7e48c768 | 1599 | process_queued_io_list(m); |
76e33fe4 | 1600 | |
2bded7bd KU |
1601 | /* |
1602 | * Wake up any thread waiting to suspend. | |
1603 | */ | |
1604 | wake_up(&m->pg_init_wait); | |
1605 | ||
d0259bf0 | 1606 | out: |
cfae5c9b CS |
1607 | spin_unlock_irqrestore(&m->lock, flags); |
1608 | } | |
1609 | ||
89bfce76 | 1610 | static void activate_or_offline_path(struct pgpath *pgpath) |
bab7cfc7 | 1611 | { |
f10e06b7 | 1612 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); |
bab7cfc7 | 1613 | |
f10e06b7 MS |
1614 | if (pgpath->is_active && !blk_queue_dying(q)) |
1615 | scsi_dh_activate(q, pg_init_done, pgpath); | |
3a017509 HR |
1616 | else |
1617 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1618 | } |
1619 | ||
89bfce76 BVA |
1620 | static void activate_path_work(struct work_struct *work) |
1621 | { | |
1622 | struct pgpath *pgpath = | |
1623 | container_of(work, struct pgpath, activate_path.work); | |
1624 | ||
1625 | activate_or_offline_path(pgpath); | |
1626 | } | |
1627 | ||
b79f10ee | 1628 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
2a842aca | 1629 | blk_status_t error, union map_info *map_context) |
1da177e4 | 1630 | { |
b79f10ee CH |
1631 | struct dm_mpath_io *mpio = get_mpio(map_context); |
1632 | struct pgpath *pgpath = mpio->pgpath; | |
7ed8578a | 1633 | int r = DM_ENDIO_DONE; |
b79f10ee | 1634 | |
f40c67f0 KU |
1635 | /* |
1636 | * We don't queue any clone request inside the multipath target | |
1637 | * during end I/O handling, since those clone requests don't have | |
1638 | * bio clones. If we queue them inside the multipath target, | |
1639 | * we need to make bio clones, that requires memory allocation. | |
4cc96131 | 1640 | * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests |
f40c67f0 KU |
1641 | * don't have bio clones.) |
1642 | * Instead of queueing the clone request here, we queue the original | |
1643 | * request into dm core, which will remake a clone request and | |
1644 | * clone bios for it and resubmit it later. | |
1645 | */ | |
a1275677 | 1646 | if (error && blk_path_error(error)) { |
b79f10ee | 1647 | struct multipath *m = ti->private; |
1da177e4 | 1648 | |
ac514ffc MS |
1649 | if (error == BLK_STS_RESOURCE) |
1650 | r = DM_ENDIO_DELAY_REQUEUE; | |
1651 | else | |
1652 | r = DM_ENDIO_REQUEUE; | |
1da177e4 | 1653 | |
b79f10ee CH |
1654 | if (pgpath) |
1655 | fail_path(pgpath); | |
1da177e4 | 1656 | |
73265f3f | 1657 | if (!atomic_read(&m->nr_valid_paths) && |
c1fd0abe | 1658 | !must_push_back_rq(m)) { |
2a842aca | 1659 | if (error == BLK_STS_IOERR) |
18a482f5 | 1660 | dm_report_EIO(m); |
7ed8578a CH |
1661 | /* complete with the original error */ |
1662 | r = DM_ENDIO_DONE; | |
1663 | } | |
b79f10ee | 1664 | } |
466891f9 | 1665 | |
1da177e4 | 1666 | if (pgpath) { |
b79f10ee CH |
1667 | struct path_selector *ps = &pgpath->pg->ps; |
1668 | ||
1da177e4 | 1669 | if (ps->type->end_io) |
087615bf GKB |
1670 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes, |
1671 | clone->io_start_time_ns); | |
1da177e4 | 1672 | } |
1da177e4 | 1673 | |
7ed8578a | 1674 | return r; |
1da177e4 LT |
1675 | } |
1676 | ||
4e4cbee9 | 1677 | static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, |
848b8aef | 1678 | blk_status_t *error) |
76e33fe4 | 1679 | { |
14ef1e48 CH |
1680 | struct multipath *m = ti->private; |
1681 | struct dm_mpath_io *mpio = get_mpio_from_bio(clone); | |
1682 | struct pgpath *pgpath = mpio->pgpath; | |
76e33fe4 | 1683 | unsigned long flags; |
1be56909 | 1684 | int r = DM_ENDIO_DONE; |
76e33fe4 | 1685 | |
a1275677 | 1686 | if (!*error || !blk_path_error(*error)) |
14ef1e48 | 1687 | goto done; |
76e33fe4 | 1688 | |
14ef1e48 CH |
1689 | if (pgpath) |
1690 | fail_path(pgpath); | |
76e33fe4 | 1691 | |
a271a89c MS |
1692 | if (!atomic_read(&m->nr_valid_paths)) { |
1693 | spin_lock_irqsave(&m->lock, flags); | |
1694 | if (!test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
1695 | if (__must_push_back(m)) { | |
1696 | r = DM_ENDIO_REQUEUE; | |
1697 | } else { | |
1698 | dm_report_EIO(m); | |
1699 | *error = BLK_STS_IOERR; | |
1700 | } | |
1701 | spin_unlock_irqrestore(&m->lock, flags); | |
1702 | goto done; | |
c1fd0abe | 1703 | } |
a271a89c | 1704 | spin_unlock_irqrestore(&m->lock, flags); |
18a482f5 | 1705 | } |
76e33fe4 | 1706 | |
f45f1186 | 1707 | multipath_queue_bio(m, clone); |
1be56909 | 1708 | r = DM_ENDIO_INCOMPLETE; |
14ef1e48 | 1709 | done: |
76e33fe4 | 1710 | if (pgpath) { |
14ef1e48 CH |
1711 | struct path_selector *ps = &pgpath->pg->ps; |
1712 | ||
76e33fe4 | 1713 | if (ps->type->end_io) |
087615bf GKB |
1714 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes, |
1715 | dm_start_time_ns_from_clone(clone)); | |
76e33fe4 MS |
1716 | } |
1717 | ||
1be56909 | 1718 | return r; |
76e33fe4 MS |
1719 | } |
1720 | ||
1da177e4 | 1721 | /* |
553ec94c MS |
1722 | * Suspend with flush can't complete until all the I/O is processed |
1723 | * so if the last path fails we must error any remaining I/O. | |
1724 | * - Note that if the freeze_bdev fails while suspending, the | |
1725 | * queue_if_no_path state is lost - userspace should reset it. | |
1726 | * Otherwise, during noflush suspend, queue_if_no_path will not change. | |
1da177e4 LT |
1727 | */ |
1728 | static void multipath_presuspend(struct dm_target *ti) | |
1729 | { | |
7943bd6d | 1730 | struct multipath *m = ti->private; |
1da177e4 | 1731 | |
553ec94c MS |
1732 | /* FIXME: bio-based shouldn't need to always disable queue_if_no_path */ |
1733 | if (m->queue_mode == DM_TYPE_BIO_BASED || !dm_noflush_suspending(m->ti)) | |
4c3f4838 | 1734 | queue_if_no_path(m, false, true, __func__); |
1da177e4 LT |
1735 | } |
1736 | ||
6df400ab KU |
1737 | static void multipath_postsuspend(struct dm_target *ti) |
1738 | { | |
6380f26f MA |
1739 | struct multipath *m = ti->private; |
1740 | ||
1741 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1742 | flush_multipath_work(m); |
6380f26f | 1743 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1744 | } |
1745 | ||
436d4108 AK |
1746 | /* |
1747 | * Restore the queue_if_no_path setting. | |
1748 | */ | |
1da177e4 LT |
1749 | static void multipath_resume(struct dm_target *ti) |
1750 | { | |
7943bd6d | 1751 | struct multipath *m = ti->private; |
1814f2e3 | 1752 | unsigned long flags; |
1da177e4 | 1753 | |
1814f2e3 | 1754 | spin_lock_irqsave(&m->lock, flags); |
553ec94c MS |
1755 | if (test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)) { |
1756 | set_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
1757 | clear_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
1758 | } | |
4c3f4838 MS |
1759 | |
1760 | DMDEBUG("%s: %s finished; QIFNP = %d; SQIFNP = %d", | |
1761 | dm_device_name(dm_table_get_md(m->ti->table)), __func__, | |
1762 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags), | |
1763 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)); | |
1764 | ||
1814f2e3 | 1765 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1766 | } |
1767 | ||
1768 | /* | |
1769 | * Info output has the following format: | |
1770 | * num_multipath_feature_args [multipath_feature_args]* | |
1771 | * num_handler_status_args [handler_status_args]* | |
1772 | * num_groups init_group_number | |
1773 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1774 | * num_paths num_selector_args | |
1775 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1776 | * | |
1777 | * Table output has the following format (identical to the constructor string): | |
1778 | * num_feature_args [features_args]* | |
1779 | * num_handler_args hw_handler [hw_handler_args]* | |
1780 | * num_groups init_group_number | |
1781 | * [priority selector-name num_ps_args [ps_args]* | |
1782 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1783 | */ | |
fd7c092e MP |
1784 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1785 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1786 | { |
1787 | int sz = 0; | |
1788 | unsigned long flags; | |
7943bd6d | 1789 | struct multipath *m = ti->private; |
1da177e4 LT |
1790 | struct priority_group *pg; |
1791 | struct pgpath *p; | |
1792 | unsigned pg_num; | |
1793 | char state; | |
1794 | ||
1795 | spin_lock_irqsave(&m->lock, flags); | |
1796 | ||
1797 | /* Features */ | |
1798 | if (type == STATUSTYPE_INFO) | |
91e968aa MS |
1799 | DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags), |
1800 | atomic_read(&m->pg_init_count)); | |
c9e45581 | 1801 | else { |
518257b1 | 1802 | DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) + |
4e2d19e4 | 1803 | (m->pg_init_retries > 0) * 2 + |
a58a935d | 1804 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
e83068a5 MS |
1805 | test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) + |
1806 | (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2); | |
1807 | ||
518257b1 | 1808 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
c9e45581 DW |
1809 | DMEMIT("queue_if_no_path "); |
1810 | if (m->pg_init_retries) | |
1811 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1812 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1813 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
518257b1 | 1814 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) |
a58a935d | 1815 | DMEMIT("retain_attached_hw_handler "); |
e83068a5 MS |
1816 | if (m->queue_mode != DM_TYPE_REQUEST_BASED) { |
1817 | switch(m->queue_mode) { | |
1818 | case DM_TYPE_BIO_BASED: | |
1819 | DMEMIT("queue_mode bio "); | |
1820 | break; | |
7e0d574f BVA |
1821 | default: |
1822 | WARN_ON_ONCE(true); | |
1823 | break; | |
e83068a5 MS |
1824 | } |
1825 | } | |
c9e45581 | 1826 | } |
1da177e4 | 1827 | |
cfae5c9b | 1828 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1829 | DMEMIT("0 "); |
1830 | else | |
cfae5c9b | 1831 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1832 | |
1833 | DMEMIT("%u ", m->nr_priority_groups); | |
1834 | ||
1835 | if (m->next_pg) | |
1836 | pg_num = m->next_pg->pg_num; | |
1837 | else if (m->current_pg) | |
1838 | pg_num = m->current_pg->pg_num; | |
1839 | else | |
a490a07a | 1840 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1841 | |
1842 | DMEMIT("%u ", pg_num); | |
1843 | ||
1844 | switch (type) { | |
1845 | case STATUSTYPE_INFO: | |
1846 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1847 | if (pg->bypassed) | |
1848 | state = 'D'; /* Disabled */ | |
1849 | else if (pg == m->current_pg) | |
1850 | state = 'A'; /* Currently Active */ | |
1851 | else | |
1852 | state = 'E'; /* Enabled */ | |
1853 | ||
1854 | DMEMIT("%c ", state); | |
1855 | ||
1856 | if (pg->ps.type->status) | |
1857 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1858 | result + sz, | |
1859 | maxlen - sz); | |
1860 | else | |
1861 | DMEMIT("0 "); | |
1862 | ||
1863 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1864 | pg->ps.type->info_args); | |
1865 | ||
1866 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1867 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1868 | p->is_active ? "A" : "F", |
1da177e4 LT |
1869 | p->fail_count); |
1870 | if (pg->ps.type->status) | |
1871 | sz += pg->ps.type->status(&pg->ps, | |
1872 | &p->path, type, result + sz, | |
1873 | maxlen - sz); | |
1874 | } | |
1875 | } | |
1876 | break; | |
1877 | ||
1878 | case STATUSTYPE_TABLE: | |
1879 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1880 | DMEMIT("%s ", pg->ps.type->name); | |
1881 | ||
1882 | if (pg->ps.type->status) | |
1883 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1884 | result + sz, | |
1885 | maxlen - sz); | |
1886 | else | |
1887 | DMEMIT("0 "); | |
1888 | ||
1889 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1890 | pg->ps.type->table_args); | |
1891 | ||
1892 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1893 | DMEMIT("%s ", p->path.dev->name); | |
1894 | if (pg->ps.type->status) | |
1895 | sz += pg->ps.type->status(&pg->ps, | |
1896 | &p->path, type, result + sz, | |
1897 | maxlen - sz); | |
1898 | } | |
1899 | } | |
1900 | break; | |
1901 | } | |
1902 | ||
1903 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1904 | } |
1905 | ||
1eb5fa84 MS |
1906 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv, |
1907 | char *result, unsigned maxlen) | |
1da177e4 | 1908 | { |
6380f26f | 1909 | int r = -EINVAL; |
1da177e4 | 1910 | struct dm_dev *dev; |
7943bd6d | 1911 | struct multipath *m = ti->private; |
1da177e4 | 1912 | action_fn action; |
be240ff5 | 1913 | unsigned long flags; |
1da177e4 | 1914 | |
6380f26f MA |
1915 | mutex_lock(&m->work_mutex); |
1916 | ||
c2f3d24b KU |
1917 | if (dm_suspended(ti)) { |
1918 | r = -EBUSY; | |
1919 | goto out; | |
1920 | } | |
1921 | ||
1da177e4 | 1922 | if (argc == 1) { |
498f0103 | 1923 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
4c3f4838 | 1924 | r = queue_if_no_path(m, true, false, __func__); |
be240ff5 AP |
1925 | spin_lock_irqsave(&m->lock, flags); |
1926 | enable_nopath_timeout(m); | |
1927 | spin_unlock_irqrestore(&m->lock, flags); | |
6380f26f | 1928 | goto out; |
498f0103 | 1929 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
4c3f4838 | 1930 | r = queue_if_no_path(m, false, false, __func__); |
be240ff5 | 1931 | disable_nopath_timeout(m); |
6380f26f MA |
1932 | goto out; |
1933 | } | |
1da177e4 LT |
1934 | } |
1935 | ||
6380f26f | 1936 | if (argc != 2) { |
a356e426 | 1937 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1938 | goto out; |
1939 | } | |
1da177e4 | 1940 | |
498f0103 | 1941 | if (!strcasecmp(argv[0], "disable_group")) { |
be7d31cc | 1942 | r = bypass_pg_num(m, argv[1], true); |
6380f26f | 1943 | goto out; |
498f0103 | 1944 | } else if (!strcasecmp(argv[0], "enable_group")) { |
be7d31cc | 1945 | r = bypass_pg_num(m, argv[1], false); |
6380f26f | 1946 | goto out; |
498f0103 | 1947 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1948 | r = switch_pg_num(m, argv[1]); |
1949 | goto out; | |
498f0103 | 1950 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1951 | action = reinstate_path; |
498f0103 | 1952 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1953 | action = fail_path; |
6380f26f | 1954 | else { |
a356e426 | 1955 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1956 | goto out; |
1957 | } | |
1da177e4 | 1958 | |
8215d6ec | 1959 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1960 | if (r) { |
72d94861 | 1961 | DMWARN("message: error getting device %s", |
1da177e4 | 1962 | argv[1]); |
6380f26f | 1963 | goto out; |
1da177e4 LT |
1964 | } |
1965 | ||
1966 | r = action_dev(m, dev, action); | |
1967 | ||
1968 | dm_put_device(ti, dev); | |
1969 | ||
6380f26f MA |
1970 | out: |
1971 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1972 | return r; |
1da177e4 LT |
1973 | } |
1974 | ||
e56f81e0 | 1975 | static int multipath_prepare_ioctl(struct dm_target *ti, |
5bd5e8d8 | 1976 | struct block_device **bdev) |
9af4aa30 | 1977 | { |
35991652 | 1978 | struct multipath *m = ti->private; |
564dbb13 | 1979 | struct pgpath *pgpath; |
69cea0d4 | 1980 | unsigned long flags; |
35991652 MP |
1981 | int r; |
1982 | ||
564dbb13 | 1983 | pgpath = READ_ONCE(m->current_pgpath); |
374117ad | 1984 | if (!pgpath || !mpath_double_check_test_bit(MPATHF_QUEUE_IO, m)) |
564dbb13 | 1985 | pgpath = choose_pgpath(m, 0); |
9af4aa30 | 1986 | |
564dbb13 | 1987 | if (pgpath) { |
374117ad | 1988 | if (!mpath_double_check_test_bit(MPATHF_QUEUE_IO, m)) { |
564dbb13 | 1989 | *bdev = pgpath->path.dev->bdev; |
43e43c9e JN |
1990 | r = 0; |
1991 | } else { | |
1992 | /* pg_init has not started or completed */ | |
1993 | r = -ENOTCONN; | |
1994 | } | |
1995 | } else { | |
1996 | /* No path is available */ | |
a271a89c MS |
1997 | r = -EIO; |
1998 | spin_lock_irqsave(&m->lock, flags); | |
518257b1 | 1999 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
43e43c9e | 2000 | r = -ENOTCONN; |
a271a89c | 2001 | spin_unlock_irqrestore(&m->lock, flags); |
e90dae1f | 2002 | } |
9af4aa30 | 2003 | |
5bbbfdf6 | 2004 | if (r == -ENOTCONN) { |
506458ef | 2005 | if (!READ_ONCE(m->current_pg)) { |
3e9f1be1 | 2006 | /* Path status changed, redo selection */ |
2da1610a | 2007 | (void) choose_pgpath(m, 0); |
3e9f1be1 | 2008 | } |
69cea0d4 | 2009 | spin_lock_irqsave(&m->lock, flags); |
518257b1 | 2010 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
69cea0d4 MS |
2011 | (void) __pg_init_all_paths(m); |
2012 | spin_unlock_irqrestore(&m->lock, flags); | |
63d832c3 | 2013 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 2014 | process_queued_io_list(m); |
3e9f1be1 | 2015 | } |
35991652 | 2016 | |
e56f81e0 CH |
2017 | /* |
2018 | * Only pass ioctls through if the device sizes match exactly. | |
2019 | */ | |
2020 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
2021 | return 1; | |
2022 | return r; | |
9af4aa30 MB |
2023 | } |
2024 | ||
af4874e0 MS |
2025 | static int multipath_iterate_devices(struct dm_target *ti, |
2026 | iterate_devices_callout_fn fn, void *data) | |
2027 | { | |
2028 | struct multipath *m = ti->private; | |
2029 | struct priority_group *pg; | |
2030 | struct pgpath *p; | |
2031 | int ret = 0; | |
2032 | ||
2033 | list_for_each_entry(pg, &m->priority_groups, list) { | |
2034 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 2035 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
2036 | if (ret) |
2037 | goto out; | |
2038 | } | |
2039 | } | |
2040 | ||
2041 | out: | |
2042 | return ret; | |
2043 | } | |
2044 | ||
9f54cec5 | 2045 | static int pgpath_busy(struct pgpath *pgpath) |
f40c67f0 KU |
2046 | { |
2047 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
2048 | ||
52b09914 | 2049 | return blk_lld_busy(q); |
f40c67f0 KU |
2050 | } |
2051 | ||
2052 | /* | |
2053 | * We return "busy", only when we can map I/Os but underlying devices | |
2054 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
2055 | * the underlying queue). | |
2056 | * In other words, if we want to kill I/Os or queue them inside us | |
2057 | * due to map unavailability, we don't return "busy". Otherwise, | |
2058 | * dm core won't give us the I/Os and we can't do what we want. | |
2059 | */ | |
2060 | static int multipath_busy(struct dm_target *ti) | |
2061 | { | |
be7d31cc | 2062 | bool busy = false, has_active = false; |
f40c67f0 | 2063 | struct multipath *m = ti->private; |
2da1610a | 2064 | struct priority_group *pg, *next_pg; |
f40c67f0 | 2065 | struct pgpath *pgpath; |
f40c67f0 | 2066 | |
b88efd43 MS |
2067 | /* pg_init in progress */ |
2068 | if (atomic_read(&m->pg_init_in_progress)) | |
2da1610a MS |
2069 | return true; |
2070 | ||
b88efd43 | 2071 | /* no paths available, for blk-mq: rely on IO mapping to delay requeue */ |
a271a89c MS |
2072 | if (!atomic_read(&m->nr_valid_paths)) { |
2073 | unsigned long flags; | |
2074 | spin_lock_irqsave(&m->lock, flags); | |
2075 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
2076 | spin_unlock_irqrestore(&m->lock, flags); | |
2077 | return (m->queue_mode != DM_TYPE_REQUEST_BASED); | |
2078 | } | |
2079 | spin_unlock_irqrestore(&m->lock, flags); | |
2080 | } | |
b88efd43 | 2081 | |
f40c67f0 | 2082 | /* Guess which priority_group will be used at next mapping time */ |
506458ef WD |
2083 | pg = READ_ONCE(m->current_pg); |
2084 | next_pg = READ_ONCE(m->next_pg); | |
2085 | if (unlikely(!READ_ONCE(m->current_pgpath) && next_pg)) | |
2da1610a MS |
2086 | pg = next_pg; |
2087 | ||
2088 | if (!pg) { | |
f40c67f0 KU |
2089 | /* |
2090 | * We don't know which pg will be used at next mapping time. | |
2da1610a | 2091 | * We don't call choose_pgpath() here to avoid to trigger |
f40c67f0 KU |
2092 | * pg_init just by busy checking. |
2093 | * So we don't know whether underlying devices we will be using | |
2094 | * at next mapping time are busy or not. Just try mapping. | |
2095 | */ | |
2da1610a MS |
2096 | return busy; |
2097 | } | |
f40c67f0 KU |
2098 | |
2099 | /* | |
2100 | * If there is one non-busy active path at least, the path selector | |
2101 | * will be able to select it. So we consider such a pg as not busy. | |
2102 | */ | |
be7d31cc | 2103 | busy = true; |
2da1610a | 2104 | list_for_each_entry(pgpath, &pg->pgpaths, list) { |
f40c67f0 | 2105 | if (pgpath->is_active) { |
be7d31cc | 2106 | has_active = true; |
9f54cec5 | 2107 | if (!pgpath_busy(pgpath)) { |
be7d31cc | 2108 | busy = false; |
f40c67f0 KU |
2109 | break; |
2110 | } | |
2111 | } | |
2da1610a | 2112 | } |
f40c67f0 | 2113 | |
2da1610a | 2114 | if (!has_active) { |
f40c67f0 KU |
2115 | /* |
2116 | * No active path in this pg, so this pg won't be used and | |
2117 | * the current_pg will be changed at next mapping time. | |
2118 | * We need to try mapping to determine it. | |
2119 | */ | |
be7d31cc | 2120 | busy = false; |
2da1610a | 2121 | } |
f40c67f0 KU |
2122 | |
2123 | return busy; | |
2124 | } | |
2125 | ||
1da177e4 LT |
2126 | /*----------------------------------------------------------------- |
2127 | * Module setup | |
2128 | *---------------------------------------------------------------*/ | |
2129 | static struct target_type multipath_target = { | |
2130 | .name = "multipath", | |
636be424 | 2131 | .version = {1, 14, 0}, |
8c5c1473 SM |
2132 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE | |
2133 | DM_TARGET_PASSES_INTEGRITY, | |
1da177e4 LT |
2134 | .module = THIS_MODULE, |
2135 | .ctr = multipath_ctr, | |
2136 | .dtr = multipath_dtr, | |
e5863d9a MS |
2137 | .clone_and_map_rq = multipath_clone_and_map, |
2138 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 2139 | .rq_end_io = multipath_end_io, |
76e33fe4 MS |
2140 | .map = multipath_map_bio, |
2141 | .end_io = multipath_end_io_bio, | |
2142 | .presuspend = multipath_presuspend, | |
2143 | .postsuspend = multipath_postsuspend, | |
2144 | .resume = multipath_resume, | |
2145 | .status = multipath_status, | |
2146 | .message = multipath_message, | |
2147 | .prepare_ioctl = multipath_prepare_ioctl, | |
2148 | .iterate_devices = multipath_iterate_devices, | |
2149 | .busy = multipath_busy, | |
2150 | }; | |
2151 | ||
1da177e4 LT |
2152 | static int __init dm_multipath_init(void) |
2153 | { | |
2154 | int r; | |
2155 | ||
4d4d66ab | 2156 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 2157 | if (!kmultipathd) { |
0cd33124 | 2158 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
2159 | r = -ENOMEM; |
2160 | goto bad_alloc_kmultipathd; | |
c557308e AK |
2161 | } |
2162 | ||
bab7cfc7 CS |
2163 | /* |
2164 | * A separate workqueue is used to handle the device handlers | |
2165 | * to avoid overloading existing workqueue. Overloading the | |
2166 | * old workqueue would also create a bottleneck in the | |
2167 | * path of the storage hardware device activation. | |
2168 | */ | |
4d4d66ab TH |
2169 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
2170 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
2171 | if (!kmpath_handlerd) { |
2172 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
2173 | r = -ENOMEM; |
2174 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
2175 | } |
2176 | ||
7e6358d2 | 2177 | r = dm_register_target(&multipath_target); |
2178 | if (r < 0) { | |
2179 | DMERR("request-based register failed %d", r); | |
2180 | r = -EINVAL; | |
2181 | goto bad_register_target; | |
2182 | } | |
2183 | ||
ff658e9c JT |
2184 | return 0; |
2185 | ||
7e6358d2 | 2186 | bad_register_target: |
2187 | destroy_workqueue(kmpath_handlerd); | |
ff658e9c JT |
2188 | bad_alloc_kmpath_handlerd: |
2189 | destroy_workqueue(kmultipathd); | |
2190 | bad_alloc_kmultipathd: | |
1da177e4 LT |
2191 | return r; |
2192 | } | |
2193 | ||
2194 | static void __exit dm_multipath_exit(void) | |
2195 | { | |
bab7cfc7 | 2196 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
2197 | destroy_workqueue(kmultipathd); |
2198 | ||
10d3bd09 | 2199 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
2200 | } |
2201 | ||
1da177e4 LT |
2202 | module_init(dm_multipath_init); |
2203 | module_exit(dm_multipath_exit); | |
2204 | ||
be240ff5 AP |
2205 | module_param_named(queue_if_no_path_timeout_secs, |
2206 | queue_if_no_path_timeout_secs, ulong, S_IRUGO | S_IWUSR); | |
2207 | MODULE_PARM_DESC(queue_if_no_path_timeout_secs, "No available paths queue IO timeout in seconds"); | |
2208 | ||
1da177e4 LT |
2209 | MODULE_DESCRIPTION(DM_NAME " multipath target"); |
2210 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
2211 | MODULE_LICENSE("GPL"); |