]>
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 LT |
130 | |
131 | /*----------------------------------------------- | |
132 | * Allocation routines | |
133 | *-----------------------------------------------*/ | |
134 | ||
135 | static struct pgpath *alloc_pgpath(void) | |
136 | { | |
e69fae56 | 137 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 138 | |
848b8aef MS |
139 | if (!pgpath) |
140 | return NULL; | |
141 | ||
142 | pgpath->is_active = true; | |
1da177e4 LT |
143 | |
144 | return pgpath; | |
145 | } | |
146 | ||
028867ac | 147 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
148 | { |
149 | kfree(pgpath); | |
150 | } | |
151 | ||
152 | static struct priority_group *alloc_priority_group(void) | |
153 | { | |
154 | struct priority_group *pg; | |
155 | ||
e69fae56 | 156 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 157 | |
e69fae56 MM |
158 | if (pg) |
159 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
160 | |
161 | return pg; | |
162 | } | |
163 | ||
164 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
165 | { | |
166 | struct pgpath *pgpath, *tmp; | |
167 | ||
168 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
169 | list_del(&pgpath->list); | |
170 | dm_put_device(ti, pgpath->path.dev); | |
171 | free_pgpath(pgpath); | |
172 | } | |
173 | } | |
174 | ||
175 | static void free_priority_group(struct priority_group *pg, | |
176 | struct dm_target *ti) | |
177 | { | |
178 | struct path_selector *ps = &pg->ps; | |
179 | ||
180 | if (ps->type) { | |
181 | ps->type->destroy(ps); | |
182 | dm_put_path_selector(ps->type); | |
183 | } | |
184 | ||
185 | free_pgpaths(&pg->pgpaths, ti); | |
186 | kfree(pg); | |
187 | } | |
188 | ||
e83068a5 | 189 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
190 | { |
191 | struct multipath *m; | |
192 | ||
e69fae56 | 193 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 194 | if (m) { |
1da177e4 LT |
195 | INIT_LIST_HEAD(&m->priority_groups); |
196 | spin_lock_init(&m->lock); | |
91e968aa | 197 | atomic_set(&m->nr_valid_paths, 0); |
c4028958 | 198 | INIT_WORK(&m->trigger_event, trigger_event); |
6380f26f | 199 | mutex_init(&m->work_mutex); |
8637a6bf | 200 | |
e83068a5 | 201 | m->queue_mode = DM_TYPE_NONE; |
76e33fe4 | 202 | |
28f16c20 MM |
203 | m->ti = ti; |
204 | ti->private = m; | |
be240ff5 AP |
205 | |
206 | timer_setup(&m->nopath_timer, queue_if_no_path_timeout_work, 0); | |
1da177e4 LT |
207 | } |
208 | ||
209 | return m; | |
210 | } | |
211 | ||
e83068a5 MS |
212 | static int alloc_multipath_stage2(struct dm_target *ti, struct multipath *m) |
213 | { | |
214 | if (m->queue_mode == DM_TYPE_NONE) { | |
953923c0 | 215 | m->queue_mode = DM_TYPE_REQUEST_BASED; |
8d47e659 | 216 | } else if (m->queue_mode == DM_TYPE_BIO_BASED) { |
e83068a5 | 217 | INIT_WORK(&m->process_queued_bios, process_queued_bios); |
8d47e659 MS |
218 | /* |
219 | * bio-based doesn't support any direct scsi_dh management; | |
220 | * it just discovers if a scsi_dh is attached. | |
221 | */ | |
222 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); | |
e83068a5 MS |
223 | } |
224 | ||
225 | dm_table_set_type(ti->table, m->queue_mode); | |
226 | ||
c3736674 MS |
227 | /* |
228 | * Init fields that are only used when a scsi_dh is attached | |
229 | * - must do this unconditionally (really doesn't hurt non-SCSI uses) | |
230 | */ | |
231 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
232 | atomic_set(&m->pg_init_in_progress, 0); | |
233 | atomic_set(&m->pg_init_count, 0); | |
234 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; | |
235 | init_waitqueue_head(&m->pg_init_wait); | |
236 | ||
e83068a5 MS |
237 | return 0; |
238 | } | |
239 | ||
1da177e4 LT |
240 | static void free_multipath(struct multipath *m) |
241 | { | |
242 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
243 | |
244 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
245 | list_del(&pg->list); | |
246 | free_priority_group(pg, m->ti); | |
247 | } | |
248 | ||
cfae5c9b | 249 | kfree(m->hw_handler_name); |
2bfd2e13 | 250 | kfree(m->hw_handler_params); |
d5ffebdd | 251 | mutex_destroy(&m->work_mutex); |
1da177e4 LT |
252 | kfree(m); |
253 | } | |
254 | ||
2eff1924 MS |
255 | static struct dm_mpath_io *get_mpio(union map_info *info) |
256 | { | |
257 | return info->ptr; | |
258 | } | |
259 | ||
bf661be1 MS |
260 | static size_t multipath_per_bio_data_size(void) |
261 | { | |
262 | return sizeof(struct dm_mpath_io) + sizeof(struct dm_bio_details); | |
263 | } | |
264 | ||
76e33fe4 MS |
265 | static struct dm_mpath_io *get_mpio_from_bio(struct bio *bio) |
266 | { | |
bf661be1 | 267 | return dm_per_bio_data(bio, multipath_per_bio_data_size()); |
76e33fe4 MS |
268 | } |
269 | ||
d07a241d | 270 | static struct dm_bio_details *get_bio_details_from_mpio(struct dm_mpath_io *mpio) |
76e33fe4 | 271 | { |
bf661be1 | 272 | /* dm_bio_details is immediately after the dm_mpath_io in bio's per-bio-data */ |
bf661be1 | 273 | void *bio_details = mpio + 1; |
bf661be1 MS |
274 | return bio_details; |
275 | } | |
276 | ||
63f6e6fd | 277 | static void multipath_init_per_bio_data(struct bio *bio, struct dm_mpath_io **mpio_p) |
bf661be1 MS |
278 | { |
279 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); | |
d07a241d | 280 | struct dm_bio_details *bio_details = get_bio_details_from_mpio(mpio); |
76e33fe4 | 281 | |
d0442f80 MS |
282 | mpio->nr_bytes = bio->bi_iter.bi_size; |
283 | mpio->pgpath = NULL; | |
284 | *mpio_p = mpio; | |
76e33fe4 | 285 | |
d0442f80 | 286 | dm_bio_record(bio_details, bio); |
76e33fe4 MS |
287 | } |
288 | ||
1da177e4 LT |
289 | /*----------------------------------------------- |
290 | * Path selection | |
291 | *-----------------------------------------------*/ | |
292 | ||
3e9f1be1 | 293 | static int __pg_init_all_paths(struct multipath *m) |
fb612642 KU |
294 | { |
295 | struct pgpath *pgpath; | |
4e2d19e4 | 296 | unsigned long pg_init_delay = 0; |
fb612642 | 297 | |
b194679f BVA |
298 | lockdep_assert_held(&m->lock); |
299 | ||
91e968aa | 300 | if (atomic_read(&m->pg_init_in_progress) || test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) |
3e9f1be1 | 301 | return 0; |
17f4ff45 | 302 | |
91e968aa | 303 | atomic_inc(&m->pg_init_count); |
518257b1 | 304 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
3e9f1be1 HR |
305 | |
306 | /* Check here to reset pg_init_required */ | |
307 | if (!m->current_pg) | |
308 | return 0; | |
309 | ||
518257b1 | 310 | if (test_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags)) |
4e2d19e4 CS |
311 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? |
312 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
313 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
314 | /* Skip failed paths */ | |
315 | if (!pgpath->is_active) | |
316 | continue; | |
4e2d19e4 CS |
317 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
318 | pg_init_delay)) | |
91e968aa | 319 | atomic_inc(&m->pg_init_in_progress); |
fb612642 | 320 | } |
91e968aa | 321 | return atomic_read(&m->pg_init_in_progress); |
fb612642 KU |
322 | } |
323 | ||
c1d7ecf7 | 324 | static int pg_init_all_paths(struct multipath *m) |
1da177e4 | 325 | { |
c1d7ecf7 | 326 | int ret; |
2da1610a MS |
327 | unsigned long flags; |
328 | ||
329 | spin_lock_irqsave(&m->lock, flags); | |
c1d7ecf7 | 330 | ret = __pg_init_all_paths(m); |
2da1610a | 331 | spin_unlock_irqrestore(&m->lock, flags); |
c1d7ecf7 BVA |
332 | |
333 | return ret; | |
2da1610a MS |
334 | } |
335 | ||
336 | static void __switch_pg(struct multipath *m, struct priority_group *pg) | |
337 | { | |
69cea0d4 MS |
338 | lockdep_assert_held(&m->lock); |
339 | ||
2da1610a | 340 | m->current_pg = pg; |
1da177e4 LT |
341 | |
342 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 343 | if (m->hw_handler_name) { |
518257b1 MS |
344 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
345 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 346 | } else { |
518257b1 MS |
347 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
348 | clear_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 349 | } |
c9e45581 | 350 | |
91e968aa | 351 | atomic_set(&m->pg_init_count, 0); |
1da177e4 LT |
352 | } |
353 | ||
2da1610a MS |
354 | static struct pgpath *choose_path_in_pg(struct multipath *m, |
355 | struct priority_group *pg, | |
356 | size_t nr_bytes) | |
1da177e4 | 357 | { |
2da1610a | 358 | unsigned long flags; |
c922d5f7 | 359 | struct dm_path *path; |
2da1610a | 360 | struct pgpath *pgpath; |
1da177e4 | 361 | |
90a4323c | 362 | path = pg->ps.type->select_path(&pg->ps, nr_bytes); |
1da177e4 | 363 | if (!path) |
2da1610a | 364 | return ERR_PTR(-ENXIO); |
1da177e4 | 365 | |
2da1610a | 366 | pgpath = path_to_pgpath(path); |
1da177e4 | 367 | |
506458ef | 368 | if (unlikely(READ_ONCE(m->current_pg) != pg)) { |
2da1610a MS |
369 | /* Only update current_pgpath if pg changed */ |
370 | spin_lock_irqsave(&m->lock, flags); | |
371 | m->current_pgpath = pgpath; | |
372 | __switch_pg(m, pg); | |
373 | spin_unlock_irqrestore(&m->lock, flags); | |
374 | } | |
1da177e4 | 375 | |
2da1610a | 376 | return pgpath; |
1da177e4 LT |
377 | } |
378 | ||
2da1610a | 379 | static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 | 380 | { |
2da1610a | 381 | unsigned long flags; |
1da177e4 | 382 | struct priority_group *pg; |
2da1610a | 383 | struct pgpath *pgpath; |
d19a55cc | 384 | unsigned bypassed = 1; |
1da177e4 | 385 | |
91e968aa | 386 | if (!atomic_read(&m->nr_valid_paths)) { |
69cea0d4 | 387 | spin_lock_irqsave(&m->lock, flags); |
8d47e659 | 388 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
69cea0d4 | 389 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 | 390 | goto failed; |
1f271972 | 391 | } |
1da177e4 LT |
392 | |
393 | /* Were we instructed to switch PG? */ | |
506458ef | 394 | if (READ_ONCE(m->next_pg)) { |
2da1610a | 395 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 | 396 | pg = m->next_pg; |
2da1610a MS |
397 | if (!pg) { |
398 | spin_unlock_irqrestore(&m->lock, flags); | |
399 | goto check_current_pg; | |
400 | } | |
1da177e4 | 401 | m->next_pg = NULL; |
2da1610a MS |
402 | spin_unlock_irqrestore(&m->lock, flags); |
403 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
404 | if (!IS_ERR_OR_NULL(pgpath)) | |
405 | return pgpath; | |
1da177e4 LT |
406 | } |
407 | ||
408 | /* Don't change PG until it has no remaining paths */ | |
2da1610a | 409 | check_current_pg: |
506458ef | 410 | pg = READ_ONCE(m->current_pg); |
2da1610a MS |
411 | if (pg) { |
412 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
413 | if (!IS_ERR_OR_NULL(pgpath)) | |
414 | return pgpath; | |
415 | } | |
1da177e4 LT |
416 | |
417 | /* | |
418 | * Loop through priority groups until we find a valid path. | |
419 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
420 | * Second time we only try the ones we skipped, but set |
421 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
422 | */ |
423 | do { | |
424 | list_for_each_entry(pg, &m->priority_groups, list) { | |
d19a55cc | 425 | if (pg->bypassed == !!bypassed) |
1da177e4 | 426 | continue; |
2da1610a MS |
427 | pgpath = choose_path_in_pg(m, pg, nr_bytes); |
428 | if (!IS_ERR_OR_NULL(pgpath)) { | |
69cea0d4 MS |
429 | if (!bypassed) { |
430 | spin_lock_irqsave(&m->lock, flags); | |
518257b1 | 431 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); |
69cea0d4 MS |
432 | spin_unlock_irqrestore(&m->lock, flags); |
433 | } | |
2da1610a | 434 | return pgpath; |
f220fd4e | 435 | } |
1da177e4 LT |
436 | } |
437 | } while (bypassed--); | |
438 | ||
439 | failed: | |
2da1610a | 440 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
441 | m->current_pgpath = NULL; |
442 | m->current_pg = NULL; | |
2da1610a MS |
443 | spin_unlock_irqrestore(&m->lock, flags); |
444 | ||
445 | return NULL; | |
1da177e4 LT |
446 | } |
447 | ||
45e15720 | 448 | /* |
ac75b09f | 449 | * dm_report_EIO() is a macro instead of a function to make pr_debug_ratelimited() |
86331f39 BVA |
450 | * report the function name and line number of the function from which |
451 | * it has been invoked. | |
45e15720 | 452 | */ |
86331f39 | 453 | #define dm_report_EIO(m) \ |
18a482f5 | 454 | do { \ |
86331f39 BVA |
455 | struct mapped_device *md = dm_table_get_md((m)->ti->table); \ |
456 | \ | |
ac75b09f MS |
457 | DMDEBUG_LIMIT("%s: returning EIO; QIFNP = %d; SQIFNP = %d; DNFS = %d", \ |
458 | dm_device_name(md), \ | |
459 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
460 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
461 | dm_noflush_suspending((m)->ti)); \ | |
18a482f5 | 462 | } while (0) |
45e15720 | 463 | |
c1fd0abe MS |
464 | /* |
465 | * Check whether bios must be queued in the device-mapper core rather | |
466 | * than here in the target. | |
c1fd0abe | 467 | */ |
a862e4e2 | 468 | static bool __must_push_back(struct multipath *m) |
c1fd0abe | 469 | { |
a862e4e2 | 470 | return dm_noflush_suspending(m->ti); |
c1fd0abe MS |
471 | } |
472 | ||
c1fd0abe MS |
473 | static bool must_push_back_rq(struct multipath *m) |
474 | { | |
a862e4e2 | 475 | return test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) || __must_push_back(m); |
c1fd0abe MS |
476 | } |
477 | ||
36fcffcc | 478 | /* |
76e33fe4 | 479 | * Map cloned requests (request-based multipath) |
36fcffcc | 480 | */ |
eb8db831 CH |
481 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, |
482 | union map_info *map_context, | |
483 | struct request **__clone) | |
1da177e4 | 484 | { |
7943bd6d | 485 | struct multipath *m = ti->private; |
eb8db831 | 486 | size_t nr_bytes = blk_rq_bytes(rq); |
1da177e4 | 487 | struct pgpath *pgpath; |
f40c67f0 | 488 | struct block_device *bdev; |
eb8db831 | 489 | struct dm_mpath_io *mpio = get_mpio(map_context); |
7083abbb | 490 | struct request_queue *q; |
eb8db831 | 491 | struct request *clone; |
1da177e4 | 492 | |
1da177e4 | 493 | /* Do we need to select a new pgpath? */ |
506458ef | 494 | pgpath = READ_ONCE(m->current_pgpath); |
2da1610a MS |
495 | if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) |
496 | pgpath = choose_pgpath(m, nr_bytes); | |
1da177e4 | 497 | |
9bf59a61 | 498 | if (!pgpath) { |
c1fd0abe | 499 | if (must_push_back_rq(m)) |
b88efd43 | 500 | return DM_MAPIO_DELAY_REQUEUE; |
18a482f5 | 501 | dm_report_EIO(m); /* Failed */ |
f98e0eb6 | 502 | return DM_MAPIO_KILL; |
518257b1 MS |
503 | } else if (test_bit(MPATHF_QUEUE_IO, &m->flags) || |
504 | test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { | |
459b5401 ML |
505 | pg_init_all_paths(m); |
506 | return DM_MAPIO_DELAY_REQUEUE; | |
9bf59a61 | 507 | } |
6afbc01d | 508 | |
2eb6e1e3 KB |
509 | mpio->pgpath = pgpath; |
510 | mpio->nr_bytes = nr_bytes; | |
511 | ||
9bf59a61 | 512 | bdev = pgpath->path.dev->bdev; |
7083abbb | 513 | q = bdev_get_queue(bdev); |
ff005a06 CH |
514 | clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, |
515 | BLK_MQ_REQ_NOWAIT); | |
eb8db831 CH |
516 | if (IS_ERR(clone)) { |
517 | /* EBUSY, ENODEV or EWOULDBLOCK: requeue */ | |
848b8aef | 518 | if (blk_queue_dying(q)) { |
7083abbb BVA |
519 | atomic_inc(&m->pg_init_in_progress); |
520 | activate_or_offline_path(pgpath); | |
050af08f | 521 | return DM_MAPIO_DELAY_REQUEUE; |
7083abbb | 522 | } |
050af08f ML |
523 | |
524 | /* | |
525 | * blk-mq's SCHED_RESTART can cover this requeue, so we | |
526 | * needn't deal with it by DELAY_REQUEUE. More importantly, | |
527 | * we have to return DM_MAPIO_REQUEUE so that blk-mq can | |
528 | * get the queue busy feedback (via BLK_STS_RESOURCE), | |
529 | * otherwise I/O merging can suffer. | |
530 | */ | |
6a23e05c | 531 | return DM_MAPIO_REQUEUE; |
e5863d9a | 532 | } |
eb8db831 CH |
533 | clone->bio = clone->biotail = NULL; |
534 | clone->rq_disk = bdev->bd_disk; | |
535 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
536 | *__clone = clone; | |
e5863d9a | 537 | |
9bf59a61 MS |
538 | if (pgpath->pg->ps.type->start_io) |
539 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
540 | &pgpath->path, | |
541 | nr_bytes); | |
2eb6e1e3 | 542 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
543 | } |
544 | ||
5de719e3 YY |
545 | static void multipath_release_clone(struct request *clone, |
546 | union map_info *map_context) | |
e5863d9a | 547 | { |
5de719e3 YY |
548 | if (unlikely(map_context)) { |
549 | /* | |
550 | * non-NULL map_context means caller is still map | |
551 | * method; must undo multipath_clone_and_map() | |
552 | */ | |
553 | struct dm_mpath_io *mpio = get_mpio(map_context); | |
554 | struct pgpath *pgpath = mpio->pgpath; | |
555 | ||
556 | if (pgpath && pgpath->pg->ps.type->end_io) | |
557 | pgpath->pg->ps.type->end_io(&pgpath->pg->ps, | |
558 | &pgpath->path, | |
087615bf GKB |
559 | mpio->nr_bytes, |
560 | clone->io_start_time_ns); | |
5de719e3 YY |
561 | } |
562 | ||
eb8db831 | 563 | blk_put_request(clone); |
e5863d9a MS |
564 | } |
565 | ||
76e33fe4 MS |
566 | /* |
567 | * Map cloned bios (bio-based multipath) | |
568 | */ | |
0001ec56 MS |
569 | |
570 | static struct pgpath *__map_bio(struct multipath *m, struct bio *bio) | |
76e33fe4 | 571 | { |
76e33fe4 MS |
572 | struct pgpath *pgpath; |
573 | unsigned long flags; | |
574 | bool queue_io; | |
575 | ||
576 | /* Do we need to select a new pgpath? */ | |
506458ef | 577 | pgpath = READ_ONCE(m->current_pgpath); |
5686dee3 | 578 | if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) |
0001ec56 | 579 | pgpath = choose_pgpath(m, bio->bi_iter.bi_size); |
76e33fe4 | 580 | |
5686dee3 GKB |
581 | /* MPATHF_QUEUE_IO might have been cleared by choose_pgpath. */ |
582 | queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags); | |
583 | ||
76e33fe4 MS |
584 | if ((pgpath && queue_io) || |
585 | (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) { | |
586 | /* Queue for the daemon to resubmit */ | |
587 | spin_lock_irqsave(&m->lock, flags); | |
588 | bio_list_add(&m->queued_bios, bio); | |
589 | spin_unlock_irqrestore(&m->lock, flags); | |
848b8aef | 590 | |
76e33fe4 MS |
591 | /* PG_INIT_REQUIRED cannot be set without QUEUE_IO */ |
592 | if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) | |
593 | pg_init_all_paths(m); | |
594 | else if (!queue_io) | |
595 | queue_work(kmultipathd, &m->process_queued_bios); | |
0001ec56 MS |
596 | |
597 | return ERR_PTR(-EAGAIN); | |
76e33fe4 MS |
598 | } |
599 | ||
0001ec56 MS |
600 | return pgpath; |
601 | } | |
602 | ||
0001ec56 MS |
603 | static int __multipath_map_bio(struct multipath *m, struct bio *bio, |
604 | struct dm_mpath_io *mpio) | |
605 | { | |
dbaf971c | 606 | struct pgpath *pgpath = __map_bio(m, bio); |
848b8aef | 607 | |
0001ec56 | 608 | if (IS_ERR(pgpath)) |
76e33fe4 | 609 | return DM_MAPIO_SUBMITTED; |
76e33fe4 MS |
610 | |
611 | if (!pgpath) { | |
a862e4e2 | 612 | if (__must_push_back(m)) |
ca5beb76 | 613 | return DM_MAPIO_REQUEUE; |
18a482f5 | 614 | dm_report_EIO(m); |
846785e6 | 615 | return DM_MAPIO_KILL; |
76e33fe4 MS |
616 | } |
617 | ||
618 | mpio->pgpath = pgpath; | |
76e33fe4 | 619 | |
4e4cbee9 | 620 | bio->bi_status = 0; |
74d46992 | 621 | bio_set_dev(bio, pgpath->path.dev->bdev); |
1eff9d32 | 622 | bio->bi_opf |= REQ_FAILFAST_TRANSPORT; |
76e33fe4 MS |
623 | |
624 | if (pgpath->pg->ps.type->start_io) | |
625 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
626 | &pgpath->path, | |
d0442f80 | 627 | mpio->nr_bytes); |
76e33fe4 MS |
628 | return DM_MAPIO_REMAPPED; |
629 | } | |
630 | ||
631 | static int multipath_map_bio(struct dm_target *ti, struct bio *bio) | |
632 | { | |
633 | struct multipath *m = ti->private; | |
bf661be1 MS |
634 | struct dm_mpath_io *mpio = NULL; |
635 | ||
63f6e6fd | 636 | multipath_init_per_bio_data(bio, &mpio); |
76e33fe4 MS |
637 | return __multipath_map_bio(m, bio, mpio); |
638 | } | |
639 | ||
7e48c768 | 640 | static void process_queued_io_list(struct multipath *m) |
76e33fe4 | 641 | { |
953923c0 | 642 | if (m->queue_mode == DM_TYPE_REQUEST_BASED) |
7e48c768 | 643 | dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table)); |
8d47e659 | 644 | else if (m->queue_mode == DM_TYPE_BIO_BASED) |
76e33fe4 MS |
645 | queue_work(kmultipathd, &m->process_queued_bios); |
646 | } | |
647 | ||
648 | static void process_queued_bios(struct work_struct *work) | |
649 | { | |
650 | int r; | |
651 | unsigned long flags; | |
652 | struct bio *bio; | |
653 | struct bio_list bios; | |
654 | struct blk_plug plug; | |
655 | struct multipath *m = | |
656 | container_of(work, struct multipath, process_queued_bios); | |
657 | ||
658 | bio_list_init(&bios); | |
659 | ||
660 | spin_lock_irqsave(&m->lock, flags); | |
661 | ||
662 | if (bio_list_empty(&m->queued_bios)) { | |
663 | spin_unlock_irqrestore(&m->lock, flags); | |
664 | return; | |
665 | } | |
666 | ||
667 | bio_list_merge(&bios, &m->queued_bios); | |
668 | bio_list_init(&m->queued_bios); | |
669 | ||
670 | spin_unlock_irqrestore(&m->lock, flags); | |
671 | ||
672 | blk_start_plug(&plug); | |
673 | while ((bio = bio_list_pop(&bios))) { | |
1836df08 MS |
674 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); |
675 | dm_bio_restore(get_bio_details_from_mpio(mpio), bio); | |
676 | r = __multipath_map_bio(m, bio, mpio); | |
846785e6 CH |
677 | switch (r) { |
678 | case DM_MAPIO_KILL: | |
4e4cbee9 CH |
679 | bio->bi_status = BLK_STS_IOERR; |
680 | bio_endio(bio); | |
047385b3 | 681 | break; |
846785e6 | 682 | case DM_MAPIO_REQUEUE: |
4e4cbee9 | 683 | bio->bi_status = BLK_STS_DM_REQUEUE; |
76e33fe4 | 684 | bio_endio(bio); |
846785e6 CH |
685 | break; |
686 | case DM_MAPIO_REMAPPED: | |
76e33fe4 | 687 | generic_make_request(bio); |
846785e6 | 688 | break; |
8192a0cd | 689 | case DM_MAPIO_SUBMITTED: |
9157c8d3 BVA |
690 | break; |
691 | default: | |
692 | WARN_ONCE(true, "__multipath_map_bio() returned %d\n", r); | |
846785e6 | 693 | } |
76e33fe4 MS |
694 | } |
695 | blk_finish_plug(&plug); | |
696 | } | |
697 | ||
1da177e4 LT |
698 | /* |
699 | * If we run out of usable paths, should we queue I/O or error it? | |
700 | */ | |
be7d31cc | 701 | static int queue_if_no_path(struct multipath *m, bool queue_if_no_path, |
4c3f4838 | 702 | bool save_old_value, const char *caller) |
1da177e4 LT |
703 | { |
704 | unsigned long flags; | |
553ec94c | 705 | bool queue_if_no_path_bit, saved_queue_if_no_path_bit; |
4c3f4838 MS |
706 | const char *dm_dev_name = dm_device_name(dm_table_get_md(m->ti->table)); |
707 | ||
708 | DMDEBUG("%s: %s caller=%s queue_if_no_path=%d save_old_value=%d", | |
709 | dm_dev_name, __func__, caller, queue_if_no_path, save_old_value); | |
1da177e4 LT |
710 | |
711 | spin_lock_irqsave(&m->lock, flags); | |
553ec94c MS |
712 | |
713 | queue_if_no_path_bit = test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
714 | saved_queue_if_no_path_bit = test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
715 | ||
716 | if (save_old_value) { | |
717 | if (unlikely(!queue_if_no_path_bit && saved_queue_if_no_path_bit)) { | |
718 | DMERR("%s: QIFNP disabled but saved as enabled, saving again loses state, not saving!", | |
4c3f4838 | 719 | dm_dev_name); |
553ec94c MS |
720 | } else |
721 | assign_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags, queue_if_no_path_bit); | |
722 | } else if (!queue_if_no_path && saved_queue_if_no_path_bit) { | |
723 | /* due to "fail_if_no_path" message, need to honor it. */ | |
724 | clear_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
725 | } | |
c1fd0abe | 726 | assign_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags, queue_if_no_path); |
553ec94c | 727 | |
4c3f4838 MS |
728 | DMDEBUG("%s: after %s changes; QIFNP = %d; SQIFNP = %d; DNFS = %d", |
729 | dm_dev_name, __func__, | |
730 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags), | |
731 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags), | |
732 | dm_noflush_suspending(m->ti)); | |
733 | ||
1da177e4 LT |
734 | spin_unlock_irqrestore(&m->lock, flags); |
735 | ||
76e33fe4 | 736 | if (!queue_if_no_path) { |
63d832c3 | 737 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 738 | process_queued_io_list(m); |
76e33fe4 | 739 | } |
63d832c3 | 740 | |
1da177e4 LT |
741 | return 0; |
742 | } | |
743 | ||
be240ff5 AP |
744 | /* |
745 | * If the queue_if_no_path timeout fires, turn off queue_if_no_path and | |
746 | * process any queued I/O. | |
747 | */ | |
748 | static void queue_if_no_path_timeout_work(struct timer_list *t) | |
749 | { | |
750 | struct multipath *m = from_timer(m, t, nopath_timer); | |
751 | struct mapped_device *md = dm_table_get_md(m->ti->table); | |
752 | ||
753 | DMWARN("queue_if_no_path timeout on %s, failing queued IO", dm_device_name(md)); | |
4c3f4838 | 754 | queue_if_no_path(m, false, false, __func__); |
be240ff5 AP |
755 | } |
756 | ||
757 | /* | |
758 | * Enable the queue_if_no_path timeout if necessary. | |
759 | * Called with m->lock held. | |
760 | */ | |
761 | static void enable_nopath_timeout(struct multipath *m) | |
762 | { | |
763 | unsigned long queue_if_no_path_timeout = | |
764 | READ_ONCE(queue_if_no_path_timeout_secs) * HZ; | |
765 | ||
766 | lockdep_assert_held(&m->lock); | |
767 | ||
768 | if (queue_if_no_path_timeout > 0 && | |
769 | atomic_read(&m->nr_valid_paths) == 0 && | |
770 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
771 | mod_timer(&m->nopath_timer, | |
772 | jiffies + queue_if_no_path_timeout); | |
773 | } | |
774 | } | |
775 | ||
776 | static void disable_nopath_timeout(struct multipath *m) | |
777 | { | |
778 | del_timer_sync(&m->nopath_timer); | |
779 | } | |
780 | ||
1da177e4 LT |
781 | /* |
782 | * An event is triggered whenever a path is taken out of use. | |
783 | * Includes path failure and PG bypass. | |
784 | */ | |
c4028958 | 785 | static void trigger_event(struct work_struct *work) |
1da177e4 | 786 | { |
c4028958 DH |
787 | struct multipath *m = |
788 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
789 | |
790 | dm_table_event(m->ti->table); | |
791 | } | |
792 | ||
793 | /*----------------------------------------------------------------- | |
794 | * Constructor/argument parsing: | |
795 | * <#multipath feature args> [<arg>]* | |
796 | * <#hw_handler args> [hw_handler [<arg>]*] | |
797 | * <#priority groups> | |
798 | * <initial priority group> | |
799 | * [<selector> <#selector args> [<arg>]* | |
800 | * <#paths> <#per-path selector args> | |
801 | * [<path> [<arg>]* ]+ ]+ | |
802 | *---------------------------------------------------------------*/ | |
498f0103 | 803 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
804 | struct dm_target *ti) |
805 | { | |
806 | int r; | |
807 | struct path_selector_type *pst; | |
808 | unsigned ps_argc; | |
809 | ||
5916a22b | 810 | static const struct dm_arg _args[] = { |
72d94861 | 811 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
812 | }; |
813 | ||
498f0103 | 814 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 815 | if (!pst) { |
72d94861 | 816 | ti->error = "unknown path selector type"; |
1da177e4 LT |
817 | return -EINVAL; |
818 | } | |
819 | ||
498f0103 | 820 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
821 | if (r) { |
822 | dm_put_path_selector(pst); | |
1da177e4 | 823 | return -EINVAL; |
371b2e34 | 824 | } |
1da177e4 LT |
825 | |
826 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
827 | if (r) { | |
828 | dm_put_path_selector(pst); | |
72d94861 | 829 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
830 | return r; |
831 | } | |
832 | ||
833 | pg->ps.type = pst; | |
498f0103 | 834 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
835 | |
836 | return 0; | |
837 | } | |
838 | ||
e8f74a0f | 839 | static int setup_scsi_dh(struct block_device *bdev, struct multipath *m, |
b592211c | 840 | const char **attached_handler_name, char **error) |
1da177e4 | 841 | { |
848b8aef | 842 | struct request_queue *q = bdev_get_queue(bdev); |
848b8aef | 843 | int r; |
a58a935d | 844 | |
518257b1 | 845 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) { |
1bab0de0 | 846 | retain: |
b592211c | 847 | if (*attached_handler_name) { |
54cd640d | 848 | /* |
849 | * Clear any hw_handler_params associated with a | |
850 | * handler that isn't already attached. | |
851 | */ | |
b592211c | 852 | if (m->hw_handler_name && strcmp(*attached_handler_name, m->hw_handler_name)) { |
54cd640d | 853 | kfree(m->hw_handler_params); |
854 | m->hw_handler_params = NULL; | |
855 | } | |
856 | ||
a58a935d MS |
857 | /* |
858 | * Reset hw_handler_name to match the attached handler | |
a58a935d MS |
859 | * |
860 | * NB. This modifies the table line to show the actual | |
861 | * handler instead of the original table passed in. | |
862 | */ | |
863 | kfree(m->hw_handler_name); | |
b592211c MS |
864 | m->hw_handler_name = *attached_handler_name; |
865 | *attached_handler_name = NULL; | |
a58a935d MS |
866 | } |
867 | } | |
a0cf7ea9 | 868 | |
a58a935d | 869 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
870 | r = scsi_dh_attach(q, m->hw_handler_name); |
871 | if (r == -EBUSY) { | |
1bab0de0 | 872 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 873 | |
1bab0de0 | 874 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
848b8aef | 875 | bdevname(bdev, b)); |
1bab0de0 CH |
876 | goto retain; |
877 | } | |
ae11b1b3 | 878 | if (r < 0) { |
848b8aef MS |
879 | *error = "error attaching hardware handler"; |
880 | return r; | |
ae11b1b3 | 881 | } |
2bfd2e13 CS |
882 | |
883 | if (m->hw_handler_params) { | |
884 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
885 | if (r < 0) { | |
848b8aef MS |
886 | *error = "unable to set hardware handler parameters"; |
887 | return r; | |
2bfd2e13 CS |
888 | } |
889 | } | |
ae11b1b3 HR |
890 | } |
891 | ||
848b8aef MS |
892 | return 0; |
893 | } | |
894 | ||
895 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, | |
896 | struct dm_target *ti) | |
897 | { | |
898 | int r; | |
899 | struct pgpath *p; | |
900 | struct multipath *m = ti->private; | |
e8f74a0f | 901 | struct request_queue *q; |
b592211c | 902 | const char *attached_handler_name = NULL; |
848b8aef MS |
903 | |
904 | /* we need at least a path arg */ | |
905 | if (as->argc < 1) { | |
906 | ti->error = "no device given"; | |
907 | return ERR_PTR(-EINVAL); | |
908 | } | |
909 | ||
910 | p = alloc_pgpath(); | |
911 | if (!p) | |
912 | return ERR_PTR(-ENOMEM); | |
913 | ||
914 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), | |
915 | &p->path.dev); | |
916 | if (r) { | |
917 | ti->error = "error getting device"; | |
918 | goto bad; | |
919 | } | |
920 | ||
e8f74a0f MS |
921 | q = bdev_get_queue(p->path.dev->bdev); |
922 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); | |
e457edf0 | 923 | if (attached_handler_name || m->hw_handler_name) { |
848b8aef | 924 | INIT_DELAYED_WORK(&p->activate_path, activate_path_work); |
b592211c | 925 | r = setup_scsi_dh(p->path.dev->bdev, m, &attached_handler_name, &ti->error); |
940bc471 | 926 | kfree(attached_handler_name); |
848b8aef MS |
927 | if (r) { |
928 | dm_put_device(ti, p->path.dev); | |
929 | goto bad; | |
930 | } | |
931 | } | |
932 | ||
1da177e4 LT |
933 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
934 | if (r) { | |
935 | dm_put_device(ti, p->path.dev); | |
936 | goto bad; | |
937 | } | |
938 | ||
939 | return p; | |
1da177e4 LT |
940 | bad: |
941 | free_pgpath(p); | |
01460f35 | 942 | return ERR_PTR(r); |
1da177e4 LT |
943 | } |
944 | ||
498f0103 | 945 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 946 | struct multipath *m) |
1da177e4 | 947 | { |
5916a22b | 948 | static const struct dm_arg _args[] = { |
72d94861 AK |
949 | {1, 1024, "invalid number of paths"}, |
950 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
951 | }; |
952 | ||
953 | int r; | |
498f0103 | 954 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 955 | struct priority_group *pg; |
28f16c20 | 956 | struct dm_target *ti = m->ti; |
1da177e4 LT |
957 | |
958 | if (as->argc < 2) { | |
959 | as->argc = 0; | |
01460f35 BM |
960 | ti->error = "not enough priority group arguments"; |
961 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
962 | } |
963 | ||
964 | pg = alloc_priority_group(); | |
965 | if (!pg) { | |
72d94861 | 966 | ti->error = "couldn't allocate priority group"; |
01460f35 | 967 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
968 | } |
969 | pg->m = m; | |
970 | ||
971 | r = parse_path_selector(as, pg, ti); | |
972 | if (r) | |
973 | goto bad; | |
974 | ||
975 | /* | |
976 | * read the paths | |
977 | */ | |
498f0103 | 978 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
979 | if (r) |
980 | goto bad; | |
981 | ||
498f0103 | 982 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
983 | if (r) |
984 | goto bad; | |
985 | ||
498f0103 | 986 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
987 | for (i = 0; i < pg->nr_pgpaths; i++) { |
988 | struct pgpath *pgpath; | |
498f0103 | 989 | struct dm_arg_set path_args; |
1da177e4 | 990 | |
498f0103 | 991 | if (as->argc < nr_args) { |
148acff6 | 992 | ti->error = "not enough path parameters"; |
6bbf79a1 | 993 | r = -EINVAL; |
1da177e4 | 994 | goto bad; |
148acff6 | 995 | } |
1da177e4 | 996 | |
498f0103 | 997 | path_args.argc = nr_args; |
1da177e4 LT |
998 | path_args.argv = as->argv; |
999 | ||
1000 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
1001 | if (IS_ERR(pgpath)) { |
1002 | r = PTR_ERR(pgpath); | |
1da177e4 | 1003 | goto bad; |
01460f35 | 1004 | } |
1da177e4 LT |
1005 | |
1006 | pgpath->pg = pg; | |
1007 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 1008 | dm_consume_args(as, nr_args); |
1da177e4 LT |
1009 | } |
1010 | ||
1011 | return pg; | |
1012 | ||
1013 | bad: | |
1014 | free_priority_group(pg, ti); | |
01460f35 | 1015 | return ERR_PTR(r); |
1da177e4 LT |
1016 | } |
1017 | ||
498f0103 | 1018 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 1019 | { |
1da177e4 | 1020 | unsigned hw_argc; |
2bfd2e13 | 1021 | int ret; |
28f16c20 | 1022 | struct dm_target *ti = m->ti; |
1da177e4 | 1023 | |
5916a22b | 1024 | static const struct dm_arg _args[] = { |
72d94861 | 1025 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
1026 | }; |
1027 | ||
498f0103 | 1028 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
1029 | return -EINVAL; |
1030 | ||
1031 | if (!hw_argc) | |
1032 | return 0; | |
1033 | ||
8d47e659 | 1034 | if (m->queue_mode == DM_TYPE_BIO_BASED) { |
76e33fe4 MS |
1035 | dm_consume_args(as, hw_argc); |
1036 | DMERR("bio-based multipath doesn't allow hardware handler args"); | |
1037 | return 0; | |
1038 | } | |
1039 | ||
498f0103 | 1040 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
f97dc421 | 1041 | if (!m->hw_handler_name) |
1042 | return -EINVAL; | |
14e98c5c | 1043 | |
2bfd2e13 CS |
1044 | if (hw_argc > 1) { |
1045 | char *p; | |
1046 | int i, j, len = 4; | |
1047 | ||
1048 | for (i = 0; i <= hw_argc - 2; i++) | |
1049 | len += strlen(as->argv[i]) + 1; | |
1050 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
1051 | if (!p) { | |
1052 | ti->error = "memory allocation failed"; | |
1053 | ret = -ENOMEM; | |
1054 | goto fail; | |
1055 | } | |
1056 | j = sprintf(p, "%d", hw_argc - 1); | |
1057 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
1058 | j = sprintf(p, "%s", as->argv[i]); | |
1059 | } | |
498f0103 | 1060 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
1061 | |
1062 | return 0; | |
2bfd2e13 CS |
1063 | fail: |
1064 | kfree(m->hw_handler_name); | |
1065 | m->hw_handler_name = NULL; | |
1066 | return ret; | |
1da177e4 LT |
1067 | } |
1068 | ||
498f0103 | 1069 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
1070 | { |
1071 | int r; | |
1072 | unsigned argc; | |
28f16c20 | 1073 | struct dm_target *ti = m->ti; |
498f0103 | 1074 | const char *arg_name; |
1da177e4 | 1075 | |
5916a22b | 1076 | static const struct dm_arg _args[] = { |
e83068a5 | 1077 | {0, 8, "invalid number of feature args"}, |
c9e45581 | 1078 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 1079 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
1080 | }; |
1081 | ||
498f0103 | 1082 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
1083 | if (r) |
1084 | return -EINVAL; | |
1085 | ||
1086 | if (!argc) | |
1087 | return 0; | |
1088 | ||
c9e45581 | 1089 | do { |
498f0103 | 1090 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
1091 | argc--; |
1092 | ||
498f0103 | 1093 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
4c3f4838 | 1094 | r = queue_if_no_path(m, true, false, __func__); |
c9e45581 DW |
1095 | continue; |
1096 | } | |
1097 | ||
a58a935d | 1098 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
518257b1 | 1099 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); |
a58a935d MS |
1100 | continue; |
1101 | } | |
1102 | ||
498f0103 | 1103 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 1104 | (argc >= 1)) { |
498f0103 | 1105 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
1106 | argc--; |
1107 | continue; | |
1108 | } | |
1109 | ||
498f0103 | 1110 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 1111 | (argc >= 1)) { |
498f0103 | 1112 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
1113 | argc--; |
1114 | continue; | |
1115 | } | |
1116 | ||
e83068a5 MS |
1117 | if (!strcasecmp(arg_name, "queue_mode") && |
1118 | (argc >= 1)) { | |
1119 | const char *queue_mode_name = dm_shift_arg(as); | |
1120 | ||
1121 | if (!strcasecmp(queue_mode_name, "bio")) | |
1122 | m->queue_mode = DM_TYPE_BIO_BASED; | |
953923c0 MS |
1123 | else if (!strcasecmp(queue_mode_name, "rq") || |
1124 | !strcasecmp(queue_mode_name, "mq")) | |
e83068a5 | 1125 | m->queue_mode = DM_TYPE_REQUEST_BASED; |
e83068a5 MS |
1126 | else { |
1127 | ti->error = "Unknown 'queue_mode' requested"; | |
1128 | r = -EINVAL; | |
1129 | } | |
1130 | argc--; | |
1131 | continue; | |
1132 | } | |
1133 | ||
1da177e4 | 1134 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
1135 | r = -EINVAL; |
1136 | } while (argc && !r); | |
1137 | ||
1138 | return r; | |
1da177e4 LT |
1139 | } |
1140 | ||
e83068a5 | 1141 | static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv) |
1da177e4 | 1142 | { |
498f0103 | 1143 | /* target arguments */ |
5916a22b | 1144 | static const struct dm_arg _args[] = { |
a490a07a MS |
1145 | {0, 1024, "invalid number of priority groups"}, |
1146 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
1147 | }; |
1148 | ||
1149 | int r; | |
1150 | struct multipath *m; | |
498f0103 | 1151 | struct dm_arg_set as; |
1da177e4 LT |
1152 | unsigned pg_count = 0; |
1153 | unsigned next_pg_num; | |
be240ff5 | 1154 | unsigned long flags; |
1da177e4 LT |
1155 | |
1156 | as.argc = argc; | |
1157 | as.argv = argv; | |
1158 | ||
e83068a5 | 1159 | m = alloc_multipath(ti); |
1da177e4 | 1160 | if (!m) { |
72d94861 | 1161 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
1162 | return -EINVAL; |
1163 | } | |
1164 | ||
28f16c20 | 1165 | r = parse_features(&as, m); |
1da177e4 LT |
1166 | if (r) |
1167 | goto bad; | |
1168 | ||
e83068a5 MS |
1169 | r = alloc_multipath_stage2(ti, m); |
1170 | if (r) | |
1171 | goto bad; | |
1172 | ||
28f16c20 | 1173 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
1174 | if (r) |
1175 | goto bad; | |
1176 | ||
498f0103 | 1177 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
1178 | if (r) |
1179 | goto bad; | |
1180 | ||
498f0103 | 1181 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
1182 | if (r) |
1183 | goto bad; | |
1184 | ||
a490a07a MS |
1185 | if ((!m->nr_priority_groups && next_pg_num) || |
1186 | (m->nr_priority_groups && !next_pg_num)) { | |
1187 | ti->error = "invalid initial priority group"; | |
1188 | r = -EINVAL; | |
1189 | goto bad; | |
1190 | } | |
1191 | ||
1da177e4 LT |
1192 | /* parse the priority groups */ |
1193 | while (as.argc) { | |
1194 | struct priority_group *pg; | |
91e968aa | 1195 | unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths); |
1da177e4 | 1196 | |
28f16c20 | 1197 | pg = parse_priority_group(&as, m); |
01460f35 BM |
1198 | if (IS_ERR(pg)) { |
1199 | r = PTR_ERR(pg); | |
1da177e4 LT |
1200 | goto bad; |
1201 | } | |
1202 | ||
91e968aa MS |
1203 | nr_valid_paths += pg->nr_pgpaths; |
1204 | atomic_set(&m->nr_valid_paths, nr_valid_paths); | |
1205 | ||
1da177e4 LT |
1206 | list_add_tail(&pg->list, &m->priority_groups); |
1207 | pg_count++; | |
1208 | pg->pg_num = pg_count; | |
1209 | if (!--next_pg_num) | |
1210 | m->next_pg = pg; | |
1211 | } | |
1212 | ||
1213 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 1214 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
1215 | r = -EINVAL; |
1216 | goto bad; | |
1217 | } | |
1218 | ||
be240ff5 AP |
1219 | spin_lock_irqsave(&m->lock, flags); |
1220 | enable_nopath_timeout(m); | |
1221 | spin_unlock_irqrestore(&m->lock, flags); | |
1222 | ||
55a62eef AK |
1223 | ti->num_flush_bios = 1; |
1224 | ti->num_discard_bios = 1; | |
042bcef8 | 1225 | ti->num_write_same_bios = 1; |
ac62d620 | 1226 | ti->num_write_zeroes_bios = 1; |
8d47e659 | 1227 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
bf661be1 | 1228 | ti->per_io_data_size = multipath_per_bio_data_size(); |
eb8db831 | 1229 | else |
8637a6bf | 1230 | ti->per_io_data_size = sizeof(struct dm_mpath_io); |
8627921f | 1231 | |
1da177e4 LT |
1232 | return 0; |
1233 | ||
1234 | bad: | |
1235 | free_multipath(m); | |
1236 | return r; | |
1237 | } | |
1238 | ||
2bded7bd KU |
1239 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
1240 | { | |
9f4c3f87 | 1241 | DEFINE_WAIT(wait); |
2bded7bd KU |
1242 | |
1243 | while (1) { | |
9f4c3f87 | 1244 | prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE); |
2bded7bd | 1245 | |
91e968aa | 1246 | if (!atomic_read(&m->pg_init_in_progress)) |
2bded7bd | 1247 | break; |
2bded7bd KU |
1248 | |
1249 | io_schedule(); | |
1250 | } | |
9f4c3f87 | 1251 | finish_wait(&m->pg_init_wait, &wait); |
2bded7bd KU |
1252 | } |
1253 | ||
1254 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 1255 | { |
848b8aef MS |
1256 | if (m->hw_handler_name) { |
1257 | set_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1258 | smp_mb__after_atomic(); | |
1259 | ||
935fcc56 | 1260 | if (atomic_read(&m->pg_init_in_progress)) |
1261 | flush_workqueue(kmpath_handlerd); | |
848b8aef MS |
1262 | multipath_wait_for_pg_init_completion(m); |
1263 | ||
1264 | clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1265 | smp_mb__after_atomic(); | |
1266 | } | |
954a73d5 | 1267 | |
935fcc56 | 1268 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
1269 | flush_work(&m->process_queued_bios); | |
43829731 | 1270 | flush_work(&m->trigger_event); |
6df400ab KU |
1271 | } |
1272 | ||
1273 | static void multipath_dtr(struct dm_target *ti) | |
1274 | { | |
1275 | struct multipath *m = ti->private; | |
1276 | ||
be240ff5 | 1277 | disable_nopath_timeout(m); |
2bded7bd | 1278 | flush_multipath_work(m); |
1da177e4 LT |
1279 | free_multipath(m); |
1280 | } | |
1281 | ||
1da177e4 LT |
1282 | /* |
1283 | * Take a path out of use. | |
1284 | */ | |
1285 | static int fail_path(struct pgpath *pgpath) | |
1286 | { | |
1287 | unsigned long flags; | |
1288 | struct multipath *m = pgpath->pg->m; | |
1289 | ||
1290 | spin_lock_irqsave(&m->lock, flags); | |
1291 | ||
6680073d | 1292 | if (!pgpath->is_active) |
1da177e4 LT |
1293 | goto out; |
1294 | ||
04867370 MS |
1295 | DMWARN("%s: Failing path %s.", |
1296 | dm_device_name(dm_table_get_md(m->ti->table)), | |
1297 | pgpath->path.dev->name); | |
1da177e4 LT |
1298 | |
1299 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
be7d31cc | 1300 | pgpath->is_active = false; |
1da177e4 LT |
1301 | pgpath->fail_count++; |
1302 | ||
91e968aa | 1303 | atomic_dec(&m->nr_valid_paths); |
1da177e4 LT |
1304 | |
1305 | if (pgpath == m->current_pgpath) | |
1306 | m->current_pgpath = NULL; | |
1307 | ||
b15546f9 | 1308 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
91e968aa | 1309 | pgpath->path.dev->name, atomic_read(&m->nr_valid_paths)); |
b15546f9 | 1310 | |
fe9cf30e | 1311 | schedule_work(&m->trigger_event); |
1da177e4 | 1312 | |
be240ff5 AP |
1313 | enable_nopath_timeout(m); |
1314 | ||
1da177e4 LT |
1315 | out: |
1316 | spin_unlock_irqrestore(&m->lock, flags); | |
1317 | ||
1318 | return 0; | |
1319 | } | |
1320 | ||
1321 | /* | |
1322 | * Reinstate a previously-failed path | |
1323 | */ | |
1324 | static int reinstate_path(struct pgpath *pgpath) | |
1325 | { | |
63d832c3 | 1326 | int r = 0, run_queue = 0; |
1da177e4 LT |
1327 | unsigned long flags; |
1328 | struct multipath *m = pgpath->pg->m; | |
91e968aa | 1329 | unsigned nr_valid_paths; |
1da177e4 LT |
1330 | |
1331 | spin_lock_irqsave(&m->lock, flags); | |
1332 | ||
6680073d | 1333 | if (pgpath->is_active) |
1da177e4 LT |
1334 | goto out; |
1335 | ||
04867370 MS |
1336 | DMWARN("%s: Reinstating path %s.", |
1337 | dm_device_name(dm_table_get_md(m->ti->table)), | |
1338 | pgpath->path.dev->name); | |
1da177e4 LT |
1339 | |
1340 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1341 | if (r) | |
1342 | goto out; | |
1343 | ||
be7d31cc | 1344 | pgpath->is_active = true; |
1da177e4 | 1345 | |
91e968aa MS |
1346 | nr_valid_paths = atomic_inc_return(&m->nr_valid_paths); |
1347 | if (nr_valid_paths == 1) { | |
e54f77dd | 1348 | m->current_pgpath = NULL; |
63d832c3 | 1349 | run_queue = 1; |
e54f77dd | 1350 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1351 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
91e968aa | 1352 | atomic_inc(&m->pg_init_in_progress); |
e54f77dd | 1353 | } |
1da177e4 | 1354 | |
b15546f9 | 1355 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
91e968aa | 1356 | pgpath->path.dev->name, nr_valid_paths); |
b15546f9 | 1357 | |
fe9cf30e | 1358 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1359 | |
1360 | out: | |
1361 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 1362 | if (run_queue) { |
63d832c3 | 1363 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1364 | process_queued_io_list(m); |
76e33fe4 | 1365 | } |
1da177e4 | 1366 | |
be240ff5 AP |
1367 | if (pgpath->is_active) |
1368 | disable_nopath_timeout(m); | |
1369 | ||
1da177e4 LT |
1370 | return r; |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1375 | */ | |
1376 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1377 | action_fn action) | |
1378 | { | |
19040c0b | 1379 | int r = -EINVAL; |
1da177e4 LT |
1380 | struct pgpath *pgpath; |
1381 | struct priority_group *pg; | |
1382 | ||
1383 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1384 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1385 | if (pgpath->path.dev == dev) | |
1386 | r = action(pgpath); | |
1387 | } | |
1388 | } | |
1389 | ||
1390 | return r; | |
1391 | } | |
1392 | ||
1393 | /* | |
1394 | * Temporarily try to avoid having to use the specified PG | |
1395 | */ | |
1396 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
be7d31cc | 1397 | bool bypassed) |
1da177e4 LT |
1398 | { |
1399 | unsigned long flags; | |
1400 | ||
1401 | spin_lock_irqsave(&m->lock, flags); | |
1402 | ||
1403 | pg->bypassed = bypassed; | |
1404 | m->current_pgpath = NULL; | |
1405 | m->current_pg = NULL; | |
1406 | ||
1407 | spin_unlock_irqrestore(&m->lock, flags); | |
1408 | ||
fe9cf30e | 1409 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1410 | } |
1411 | ||
1412 | /* | |
1413 | * Switch to using the specified PG from the next I/O that gets mapped | |
1414 | */ | |
1415 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1416 | { | |
1417 | struct priority_group *pg; | |
1418 | unsigned pgnum; | |
1419 | unsigned long flags; | |
31998ef1 | 1420 | char dummy; |
1da177e4 | 1421 | |
31998ef1 | 1422 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1423 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1424 | DMWARN("invalid PG number supplied to switch_pg_num"); |
1425 | return -EINVAL; | |
1426 | } | |
1427 | ||
1428 | spin_lock_irqsave(&m->lock, flags); | |
1429 | list_for_each_entry(pg, &m->priority_groups, list) { | |
be7d31cc | 1430 | pg->bypassed = false; |
1da177e4 LT |
1431 | if (--pgnum) |
1432 | continue; | |
1433 | ||
1434 | m->current_pgpath = NULL; | |
1435 | m->current_pg = NULL; | |
1436 | m->next_pg = pg; | |
1437 | } | |
1438 | spin_unlock_irqrestore(&m->lock, flags); | |
1439 | ||
fe9cf30e | 1440 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1441 | return 0; |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * Set/clear bypassed status of a PG. | |
1446 | * PGs are numbered upwards from 1 in the order they were declared. | |
1447 | */ | |
be7d31cc | 1448 | static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed) |
1da177e4 LT |
1449 | { |
1450 | struct priority_group *pg; | |
1451 | unsigned pgnum; | |
31998ef1 | 1452 | char dummy; |
1da177e4 | 1453 | |
31998ef1 | 1454 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1455 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1456 | DMWARN("invalid PG number supplied to bypass_pg"); |
1457 | return -EINVAL; | |
1458 | } | |
1459 | ||
1460 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1461 | if (!--pgnum) | |
1462 | break; | |
1463 | } | |
1464 | ||
1465 | bypass_pg(m, pg, bypassed); | |
1466 | return 0; | |
1467 | } | |
1468 | ||
c9e45581 DW |
1469 | /* |
1470 | * Should we retry pg_init immediately? | |
1471 | */ | |
be7d31cc | 1472 | static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) |
c9e45581 DW |
1473 | { |
1474 | unsigned long flags; | |
be7d31cc | 1475 | bool limit_reached = false; |
c9e45581 DW |
1476 | |
1477 | spin_lock_irqsave(&m->lock, flags); | |
1478 | ||
91e968aa MS |
1479 | if (atomic_read(&m->pg_init_count) <= m->pg_init_retries && |
1480 | !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) | |
518257b1 | 1481 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
c9e45581 | 1482 | else |
be7d31cc | 1483 | limit_reached = true; |
c9e45581 DW |
1484 | |
1485 | spin_unlock_irqrestore(&m->lock, flags); | |
1486 | ||
1487 | return limit_reached; | |
1488 | } | |
1489 | ||
3ae31f6a | 1490 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1491 | { |
83c0d5d5 | 1492 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1493 | struct priority_group *pg = pgpath->pg; |
1494 | struct multipath *m = pg->m; | |
1495 | unsigned long flags; | |
be7d31cc | 1496 | bool delay_retry = false; |
cfae5c9b CS |
1497 | |
1498 | /* device or driver problems */ | |
1499 | switch (errors) { | |
1500 | case SCSI_DH_OK: | |
1501 | break; | |
1502 | case SCSI_DH_NOSYS: | |
1503 | if (!m->hw_handler_name) { | |
1504 | errors = 0; | |
1505 | break; | |
1506 | } | |
f7b934c8 MB |
1507 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1508 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1509 | /* |
1510 | * Fail path for now, so we do not ping pong | |
1511 | */ | |
1512 | fail_path(pgpath); | |
1513 | break; | |
1514 | case SCSI_DH_DEV_TEMP_BUSY: | |
1515 | /* | |
1516 | * Probably doing something like FW upgrade on the | |
1517 | * controller so try the other pg. | |
1518 | */ | |
be7d31cc | 1519 | bypass_pg(m, pg, true); |
cfae5c9b | 1520 | break; |
cfae5c9b | 1521 | case SCSI_DH_RETRY: |
4e2d19e4 | 1522 | /* Wait before retrying. */ |
4ecc5081 | 1523 | delay_retry = true; |
7b06e09a | 1524 | /* fall through */ |
cfae5c9b CS |
1525 | case SCSI_DH_IMM_RETRY: |
1526 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1527 | if (pg_init_limit_reached(m, pgpath)) | |
1528 | fail_path(pgpath); | |
1529 | errors = 0; | |
1530 | break; | |
ec31f3f7 | 1531 | case SCSI_DH_DEV_OFFLINED: |
cfae5c9b CS |
1532 | default: |
1533 | /* | |
1534 | * We probably do not want to fail the path for a device | |
1535 | * error, but this is what the old dm did. In future | |
1536 | * patches we can do more advanced handling. | |
1537 | */ | |
1538 | fail_path(pgpath); | |
1539 | } | |
1540 | ||
1541 | spin_lock_irqsave(&m->lock, flags); | |
1542 | if (errors) { | |
e54f77dd CS |
1543 | if (pgpath == m->current_pgpath) { |
1544 | DMERR("Could not failover device. Error %d.", errors); | |
1545 | m->current_pgpath = NULL; | |
1546 | m->current_pg = NULL; | |
1547 | } | |
518257b1 | 1548 | } else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
be7d31cc | 1549 | pg->bypassed = false; |
cfae5c9b | 1550 | |
91e968aa | 1551 | if (atomic_dec_return(&m->pg_init_in_progress) > 0) |
d0259bf0 KU |
1552 | /* Activations of other paths are still on going */ |
1553 | goto out; | |
1554 | ||
518257b1 MS |
1555 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { |
1556 | if (delay_retry) | |
1557 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1558 | else | |
1559 | clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1560 | ||
3e9f1be1 HR |
1561 | if (__pg_init_all_paths(m)) |
1562 | goto out; | |
1563 | } | |
518257b1 | 1564 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
d0259bf0 | 1565 | |
7e48c768 | 1566 | process_queued_io_list(m); |
76e33fe4 | 1567 | |
2bded7bd KU |
1568 | /* |
1569 | * Wake up any thread waiting to suspend. | |
1570 | */ | |
1571 | wake_up(&m->pg_init_wait); | |
1572 | ||
d0259bf0 | 1573 | out: |
cfae5c9b CS |
1574 | spin_unlock_irqrestore(&m->lock, flags); |
1575 | } | |
1576 | ||
89bfce76 | 1577 | static void activate_or_offline_path(struct pgpath *pgpath) |
bab7cfc7 | 1578 | { |
f10e06b7 | 1579 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); |
bab7cfc7 | 1580 | |
f10e06b7 MS |
1581 | if (pgpath->is_active && !blk_queue_dying(q)) |
1582 | scsi_dh_activate(q, pg_init_done, pgpath); | |
3a017509 HR |
1583 | else |
1584 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1585 | } |
1586 | ||
89bfce76 BVA |
1587 | static void activate_path_work(struct work_struct *work) |
1588 | { | |
1589 | struct pgpath *pgpath = | |
1590 | container_of(work, struct pgpath, activate_path.work); | |
1591 | ||
1592 | activate_or_offline_path(pgpath); | |
1593 | } | |
1594 | ||
b79f10ee | 1595 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
2a842aca | 1596 | blk_status_t error, union map_info *map_context) |
1da177e4 | 1597 | { |
b79f10ee CH |
1598 | struct dm_mpath_io *mpio = get_mpio(map_context); |
1599 | struct pgpath *pgpath = mpio->pgpath; | |
7ed8578a | 1600 | int r = DM_ENDIO_DONE; |
b79f10ee | 1601 | |
f40c67f0 KU |
1602 | /* |
1603 | * We don't queue any clone request inside the multipath target | |
1604 | * during end I/O handling, since those clone requests don't have | |
1605 | * bio clones. If we queue them inside the multipath target, | |
1606 | * we need to make bio clones, that requires memory allocation. | |
4cc96131 | 1607 | * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests |
f40c67f0 KU |
1608 | * don't have bio clones.) |
1609 | * Instead of queueing the clone request here, we queue the original | |
1610 | * request into dm core, which will remake a clone request and | |
1611 | * clone bios for it and resubmit it later. | |
1612 | */ | |
a1275677 | 1613 | if (error && blk_path_error(error)) { |
b79f10ee | 1614 | struct multipath *m = ti->private; |
1da177e4 | 1615 | |
ac514ffc MS |
1616 | if (error == BLK_STS_RESOURCE) |
1617 | r = DM_ENDIO_DELAY_REQUEUE; | |
1618 | else | |
1619 | r = DM_ENDIO_REQUEUE; | |
1da177e4 | 1620 | |
b79f10ee CH |
1621 | if (pgpath) |
1622 | fail_path(pgpath); | |
1da177e4 | 1623 | |
a271a89c MS |
1624 | if (!atomic_read(&m->nr_valid_paths)) { |
1625 | unsigned long flags; | |
1626 | spin_lock_irqsave(&m->lock, flags); | |
1627 | if (!must_push_back_rq(m)) { | |
1628 | if (error == BLK_STS_IOERR) | |
1629 | dm_report_EIO(m); | |
1630 | /* complete with the original error */ | |
1631 | r = DM_ENDIO_DONE; | |
1632 | } | |
1633 | spin_unlock_irqrestore(&m->lock, flags); | |
7ed8578a | 1634 | } |
b79f10ee | 1635 | } |
466891f9 | 1636 | |
1da177e4 | 1637 | if (pgpath) { |
b79f10ee CH |
1638 | struct path_selector *ps = &pgpath->pg->ps; |
1639 | ||
1da177e4 | 1640 | if (ps->type->end_io) |
087615bf GKB |
1641 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes, |
1642 | clone->io_start_time_ns); | |
1da177e4 | 1643 | } |
1da177e4 | 1644 | |
7ed8578a | 1645 | return r; |
1da177e4 LT |
1646 | } |
1647 | ||
4e4cbee9 | 1648 | static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, |
848b8aef | 1649 | blk_status_t *error) |
76e33fe4 | 1650 | { |
14ef1e48 CH |
1651 | struct multipath *m = ti->private; |
1652 | struct dm_mpath_io *mpio = get_mpio_from_bio(clone); | |
1653 | struct pgpath *pgpath = mpio->pgpath; | |
76e33fe4 | 1654 | unsigned long flags; |
1be56909 | 1655 | int r = DM_ENDIO_DONE; |
76e33fe4 | 1656 | |
a1275677 | 1657 | if (!*error || !blk_path_error(*error)) |
14ef1e48 | 1658 | goto done; |
76e33fe4 | 1659 | |
14ef1e48 CH |
1660 | if (pgpath) |
1661 | fail_path(pgpath); | |
76e33fe4 | 1662 | |
a271a89c MS |
1663 | if (!atomic_read(&m->nr_valid_paths)) { |
1664 | spin_lock_irqsave(&m->lock, flags); | |
1665 | if (!test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
1666 | if (__must_push_back(m)) { | |
1667 | r = DM_ENDIO_REQUEUE; | |
1668 | } else { | |
1669 | dm_report_EIO(m); | |
1670 | *error = BLK_STS_IOERR; | |
1671 | } | |
1672 | spin_unlock_irqrestore(&m->lock, flags); | |
1673 | goto done; | |
c1fd0abe | 1674 | } |
a271a89c | 1675 | spin_unlock_irqrestore(&m->lock, flags); |
18a482f5 | 1676 | } |
76e33fe4 | 1677 | |
76e33fe4 MS |
1678 | spin_lock_irqsave(&m->lock, flags); |
1679 | bio_list_add(&m->queued_bios, clone); | |
76e33fe4 MS |
1680 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) |
1681 | queue_work(kmultipathd, &m->process_queued_bios); | |
69cea0d4 | 1682 | spin_unlock_irqrestore(&m->lock, flags); |
76e33fe4 | 1683 | |
1be56909 | 1684 | r = DM_ENDIO_INCOMPLETE; |
14ef1e48 | 1685 | done: |
76e33fe4 | 1686 | if (pgpath) { |
14ef1e48 CH |
1687 | struct path_selector *ps = &pgpath->pg->ps; |
1688 | ||
76e33fe4 | 1689 | if (ps->type->end_io) |
087615bf GKB |
1690 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes, |
1691 | dm_start_time_ns_from_clone(clone)); | |
76e33fe4 MS |
1692 | } |
1693 | ||
1be56909 | 1694 | return r; |
76e33fe4 MS |
1695 | } |
1696 | ||
1da177e4 | 1697 | /* |
553ec94c MS |
1698 | * Suspend with flush can't complete until all the I/O is processed |
1699 | * so if the last path fails we must error any remaining I/O. | |
1700 | * - Note that if the freeze_bdev fails while suspending, the | |
1701 | * queue_if_no_path state is lost - userspace should reset it. | |
1702 | * Otherwise, during noflush suspend, queue_if_no_path will not change. | |
1da177e4 LT |
1703 | */ |
1704 | static void multipath_presuspend(struct dm_target *ti) | |
1705 | { | |
7943bd6d | 1706 | struct multipath *m = ti->private; |
1da177e4 | 1707 | |
553ec94c MS |
1708 | /* FIXME: bio-based shouldn't need to always disable queue_if_no_path */ |
1709 | if (m->queue_mode == DM_TYPE_BIO_BASED || !dm_noflush_suspending(m->ti)) | |
4c3f4838 | 1710 | queue_if_no_path(m, false, true, __func__); |
1da177e4 LT |
1711 | } |
1712 | ||
6df400ab KU |
1713 | static void multipath_postsuspend(struct dm_target *ti) |
1714 | { | |
6380f26f MA |
1715 | struct multipath *m = ti->private; |
1716 | ||
1717 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1718 | flush_multipath_work(m); |
6380f26f | 1719 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1720 | } |
1721 | ||
436d4108 AK |
1722 | /* |
1723 | * Restore the queue_if_no_path setting. | |
1724 | */ | |
1da177e4 LT |
1725 | static void multipath_resume(struct dm_target *ti) |
1726 | { | |
7943bd6d | 1727 | struct multipath *m = ti->private; |
1814f2e3 | 1728 | unsigned long flags; |
1da177e4 | 1729 | |
1814f2e3 | 1730 | spin_lock_irqsave(&m->lock, flags); |
553ec94c MS |
1731 | if (test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)) { |
1732 | set_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags); | |
1733 | clear_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags); | |
1734 | } | |
4c3f4838 MS |
1735 | |
1736 | DMDEBUG("%s: %s finished; QIFNP = %d; SQIFNP = %d", | |
1737 | dm_device_name(dm_table_get_md(m->ti->table)), __func__, | |
1738 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags), | |
1739 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)); | |
1740 | ||
1814f2e3 | 1741 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1742 | } |
1743 | ||
1744 | /* | |
1745 | * Info output has the following format: | |
1746 | * num_multipath_feature_args [multipath_feature_args]* | |
1747 | * num_handler_status_args [handler_status_args]* | |
1748 | * num_groups init_group_number | |
1749 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1750 | * num_paths num_selector_args | |
1751 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1752 | * | |
1753 | * Table output has the following format (identical to the constructor string): | |
1754 | * num_feature_args [features_args]* | |
1755 | * num_handler_args hw_handler [hw_handler_args]* | |
1756 | * num_groups init_group_number | |
1757 | * [priority selector-name num_ps_args [ps_args]* | |
1758 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1759 | */ | |
fd7c092e MP |
1760 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1761 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1762 | { |
1763 | int sz = 0; | |
1764 | unsigned long flags; | |
7943bd6d | 1765 | struct multipath *m = ti->private; |
1da177e4 LT |
1766 | struct priority_group *pg; |
1767 | struct pgpath *p; | |
1768 | unsigned pg_num; | |
1769 | char state; | |
1770 | ||
1771 | spin_lock_irqsave(&m->lock, flags); | |
1772 | ||
1773 | /* Features */ | |
1774 | if (type == STATUSTYPE_INFO) | |
91e968aa MS |
1775 | DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags), |
1776 | atomic_read(&m->pg_init_count)); | |
c9e45581 | 1777 | else { |
518257b1 | 1778 | DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) + |
4e2d19e4 | 1779 | (m->pg_init_retries > 0) * 2 + |
a58a935d | 1780 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
e83068a5 MS |
1781 | test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) + |
1782 | (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2); | |
1783 | ||
518257b1 | 1784 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
c9e45581 DW |
1785 | DMEMIT("queue_if_no_path "); |
1786 | if (m->pg_init_retries) | |
1787 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1788 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1789 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
518257b1 | 1790 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) |
a58a935d | 1791 | DMEMIT("retain_attached_hw_handler "); |
e83068a5 MS |
1792 | if (m->queue_mode != DM_TYPE_REQUEST_BASED) { |
1793 | switch(m->queue_mode) { | |
1794 | case DM_TYPE_BIO_BASED: | |
1795 | DMEMIT("queue_mode bio "); | |
1796 | break; | |
7e0d574f BVA |
1797 | default: |
1798 | WARN_ON_ONCE(true); | |
1799 | break; | |
e83068a5 MS |
1800 | } |
1801 | } | |
c9e45581 | 1802 | } |
1da177e4 | 1803 | |
cfae5c9b | 1804 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1805 | DMEMIT("0 "); |
1806 | else | |
cfae5c9b | 1807 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1808 | |
1809 | DMEMIT("%u ", m->nr_priority_groups); | |
1810 | ||
1811 | if (m->next_pg) | |
1812 | pg_num = m->next_pg->pg_num; | |
1813 | else if (m->current_pg) | |
1814 | pg_num = m->current_pg->pg_num; | |
1815 | else | |
a490a07a | 1816 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1817 | |
1818 | DMEMIT("%u ", pg_num); | |
1819 | ||
1820 | switch (type) { | |
1821 | case STATUSTYPE_INFO: | |
1822 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1823 | if (pg->bypassed) | |
1824 | state = 'D'; /* Disabled */ | |
1825 | else if (pg == m->current_pg) | |
1826 | state = 'A'; /* Currently Active */ | |
1827 | else | |
1828 | state = 'E'; /* Enabled */ | |
1829 | ||
1830 | DMEMIT("%c ", state); | |
1831 | ||
1832 | if (pg->ps.type->status) | |
1833 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1834 | result + sz, | |
1835 | maxlen - sz); | |
1836 | else | |
1837 | DMEMIT("0 "); | |
1838 | ||
1839 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1840 | pg->ps.type->info_args); | |
1841 | ||
1842 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1843 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1844 | p->is_active ? "A" : "F", |
1da177e4 LT |
1845 | p->fail_count); |
1846 | if (pg->ps.type->status) | |
1847 | sz += pg->ps.type->status(&pg->ps, | |
1848 | &p->path, type, result + sz, | |
1849 | maxlen - sz); | |
1850 | } | |
1851 | } | |
1852 | break; | |
1853 | ||
1854 | case STATUSTYPE_TABLE: | |
1855 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1856 | DMEMIT("%s ", pg->ps.type->name); | |
1857 | ||
1858 | if (pg->ps.type->status) | |
1859 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1860 | result + sz, | |
1861 | maxlen - sz); | |
1862 | else | |
1863 | DMEMIT("0 "); | |
1864 | ||
1865 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1866 | pg->ps.type->table_args); | |
1867 | ||
1868 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1869 | DMEMIT("%s ", p->path.dev->name); | |
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 | ||
1879 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1880 | } |
1881 | ||
1eb5fa84 MS |
1882 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv, |
1883 | char *result, unsigned maxlen) | |
1da177e4 | 1884 | { |
6380f26f | 1885 | int r = -EINVAL; |
1da177e4 | 1886 | struct dm_dev *dev; |
7943bd6d | 1887 | struct multipath *m = ti->private; |
1da177e4 | 1888 | action_fn action; |
be240ff5 | 1889 | unsigned long flags; |
1da177e4 | 1890 | |
6380f26f MA |
1891 | mutex_lock(&m->work_mutex); |
1892 | ||
c2f3d24b KU |
1893 | if (dm_suspended(ti)) { |
1894 | r = -EBUSY; | |
1895 | goto out; | |
1896 | } | |
1897 | ||
1da177e4 | 1898 | if (argc == 1) { |
498f0103 | 1899 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
4c3f4838 | 1900 | r = queue_if_no_path(m, true, false, __func__); |
be240ff5 AP |
1901 | spin_lock_irqsave(&m->lock, flags); |
1902 | enable_nopath_timeout(m); | |
1903 | spin_unlock_irqrestore(&m->lock, flags); | |
6380f26f | 1904 | goto out; |
498f0103 | 1905 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
4c3f4838 | 1906 | r = queue_if_no_path(m, false, false, __func__); |
be240ff5 | 1907 | disable_nopath_timeout(m); |
6380f26f MA |
1908 | goto out; |
1909 | } | |
1da177e4 LT |
1910 | } |
1911 | ||
6380f26f | 1912 | if (argc != 2) { |
a356e426 | 1913 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1914 | goto out; |
1915 | } | |
1da177e4 | 1916 | |
498f0103 | 1917 | if (!strcasecmp(argv[0], "disable_group")) { |
be7d31cc | 1918 | r = bypass_pg_num(m, argv[1], true); |
6380f26f | 1919 | goto out; |
498f0103 | 1920 | } else if (!strcasecmp(argv[0], "enable_group")) { |
be7d31cc | 1921 | r = bypass_pg_num(m, argv[1], false); |
6380f26f | 1922 | goto out; |
498f0103 | 1923 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1924 | r = switch_pg_num(m, argv[1]); |
1925 | goto out; | |
498f0103 | 1926 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1927 | action = reinstate_path; |
498f0103 | 1928 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1929 | action = fail_path; |
6380f26f | 1930 | else { |
a356e426 | 1931 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1932 | goto out; |
1933 | } | |
1da177e4 | 1934 | |
8215d6ec | 1935 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1936 | if (r) { |
72d94861 | 1937 | DMWARN("message: error getting device %s", |
1da177e4 | 1938 | argv[1]); |
6380f26f | 1939 | goto out; |
1da177e4 LT |
1940 | } |
1941 | ||
1942 | r = action_dev(m, dev, action); | |
1943 | ||
1944 | dm_put_device(ti, dev); | |
1945 | ||
6380f26f MA |
1946 | out: |
1947 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1948 | return r; |
1da177e4 LT |
1949 | } |
1950 | ||
e56f81e0 | 1951 | static int multipath_prepare_ioctl(struct dm_target *ti, |
5bd5e8d8 | 1952 | struct block_device **bdev) |
9af4aa30 | 1953 | { |
35991652 | 1954 | struct multipath *m = ti->private; |
2da1610a | 1955 | struct pgpath *current_pgpath; |
69cea0d4 | 1956 | unsigned long flags; |
35991652 MP |
1957 | int r; |
1958 | ||
506458ef | 1959 | current_pgpath = READ_ONCE(m->current_pgpath); |
2361ae59 | 1960 | if (!current_pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) |
2da1610a | 1961 | current_pgpath = choose_pgpath(m, 0); |
9af4aa30 | 1962 | |
2da1610a | 1963 | if (current_pgpath) { |
518257b1 | 1964 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) { |
2da1610a | 1965 | *bdev = current_pgpath->path.dev->bdev; |
43e43c9e JN |
1966 | r = 0; |
1967 | } else { | |
1968 | /* pg_init has not started or completed */ | |
1969 | r = -ENOTCONN; | |
1970 | } | |
1971 | } else { | |
1972 | /* No path is available */ | |
a271a89c MS |
1973 | r = -EIO; |
1974 | spin_lock_irqsave(&m->lock, flags); | |
518257b1 | 1975 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
43e43c9e | 1976 | r = -ENOTCONN; |
a271a89c | 1977 | spin_unlock_irqrestore(&m->lock, flags); |
e90dae1f | 1978 | } |
9af4aa30 | 1979 | |
5bbbfdf6 | 1980 | if (r == -ENOTCONN) { |
506458ef | 1981 | if (!READ_ONCE(m->current_pg)) { |
3e9f1be1 | 1982 | /* Path status changed, redo selection */ |
2da1610a | 1983 | (void) choose_pgpath(m, 0); |
3e9f1be1 | 1984 | } |
69cea0d4 | 1985 | spin_lock_irqsave(&m->lock, flags); |
518257b1 | 1986 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
69cea0d4 MS |
1987 | (void) __pg_init_all_paths(m); |
1988 | spin_unlock_irqrestore(&m->lock, flags); | |
63d832c3 | 1989 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1990 | process_queued_io_list(m); |
3e9f1be1 | 1991 | } |
35991652 | 1992 | |
e56f81e0 CH |
1993 | /* |
1994 | * Only pass ioctls through if the device sizes match exactly. | |
1995 | */ | |
1996 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
1997 | return 1; | |
1998 | return r; | |
9af4aa30 MB |
1999 | } |
2000 | ||
af4874e0 MS |
2001 | static int multipath_iterate_devices(struct dm_target *ti, |
2002 | iterate_devices_callout_fn fn, void *data) | |
2003 | { | |
2004 | struct multipath *m = ti->private; | |
2005 | struct priority_group *pg; | |
2006 | struct pgpath *p; | |
2007 | int ret = 0; | |
2008 | ||
2009 | list_for_each_entry(pg, &m->priority_groups, list) { | |
2010 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 2011 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
2012 | if (ret) |
2013 | goto out; | |
2014 | } | |
2015 | } | |
2016 | ||
2017 | out: | |
2018 | return ret; | |
2019 | } | |
2020 | ||
9f54cec5 | 2021 | static int pgpath_busy(struct pgpath *pgpath) |
f40c67f0 KU |
2022 | { |
2023 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
2024 | ||
52b09914 | 2025 | return blk_lld_busy(q); |
f40c67f0 KU |
2026 | } |
2027 | ||
2028 | /* | |
2029 | * We return "busy", only when we can map I/Os but underlying devices | |
2030 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
2031 | * the underlying queue). | |
2032 | * In other words, if we want to kill I/Os or queue them inside us | |
2033 | * due to map unavailability, we don't return "busy". Otherwise, | |
2034 | * dm core won't give us the I/Os and we can't do what we want. | |
2035 | */ | |
2036 | static int multipath_busy(struct dm_target *ti) | |
2037 | { | |
be7d31cc | 2038 | bool busy = false, has_active = false; |
f40c67f0 | 2039 | struct multipath *m = ti->private; |
2da1610a | 2040 | struct priority_group *pg, *next_pg; |
f40c67f0 | 2041 | struct pgpath *pgpath; |
f40c67f0 | 2042 | |
b88efd43 MS |
2043 | /* pg_init in progress */ |
2044 | if (atomic_read(&m->pg_init_in_progress)) | |
2da1610a MS |
2045 | return true; |
2046 | ||
b88efd43 | 2047 | /* no paths available, for blk-mq: rely on IO mapping to delay requeue */ |
a271a89c MS |
2048 | if (!atomic_read(&m->nr_valid_paths)) { |
2049 | unsigned long flags; | |
2050 | spin_lock_irqsave(&m->lock, flags); | |
2051 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
2052 | spin_unlock_irqrestore(&m->lock, flags); | |
2053 | return (m->queue_mode != DM_TYPE_REQUEST_BASED); | |
2054 | } | |
2055 | spin_unlock_irqrestore(&m->lock, flags); | |
2056 | } | |
b88efd43 | 2057 | |
f40c67f0 | 2058 | /* Guess which priority_group will be used at next mapping time */ |
506458ef WD |
2059 | pg = READ_ONCE(m->current_pg); |
2060 | next_pg = READ_ONCE(m->next_pg); | |
2061 | if (unlikely(!READ_ONCE(m->current_pgpath) && next_pg)) | |
2da1610a MS |
2062 | pg = next_pg; |
2063 | ||
2064 | if (!pg) { | |
f40c67f0 KU |
2065 | /* |
2066 | * We don't know which pg will be used at next mapping time. | |
2da1610a | 2067 | * We don't call choose_pgpath() here to avoid to trigger |
f40c67f0 KU |
2068 | * pg_init just by busy checking. |
2069 | * So we don't know whether underlying devices we will be using | |
2070 | * at next mapping time are busy or not. Just try mapping. | |
2071 | */ | |
2da1610a MS |
2072 | return busy; |
2073 | } | |
f40c67f0 KU |
2074 | |
2075 | /* | |
2076 | * If there is one non-busy active path at least, the path selector | |
2077 | * will be able to select it. So we consider such a pg as not busy. | |
2078 | */ | |
be7d31cc | 2079 | busy = true; |
2da1610a | 2080 | list_for_each_entry(pgpath, &pg->pgpaths, list) { |
f40c67f0 | 2081 | if (pgpath->is_active) { |
be7d31cc | 2082 | has_active = true; |
9f54cec5 | 2083 | if (!pgpath_busy(pgpath)) { |
be7d31cc | 2084 | busy = false; |
f40c67f0 KU |
2085 | break; |
2086 | } | |
2087 | } | |
2da1610a | 2088 | } |
f40c67f0 | 2089 | |
2da1610a | 2090 | if (!has_active) { |
f40c67f0 KU |
2091 | /* |
2092 | * No active path in this pg, so this pg won't be used and | |
2093 | * the current_pg will be changed at next mapping time. | |
2094 | * We need to try mapping to determine it. | |
2095 | */ | |
be7d31cc | 2096 | busy = false; |
2da1610a | 2097 | } |
f40c67f0 KU |
2098 | |
2099 | return busy; | |
2100 | } | |
2101 | ||
1da177e4 LT |
2102 | /*----------------------------------------------------------------- |
2103 | * Module setup | |
2104 | *---------------------------------------------------------------*/ | |
2105 | static struct target_type multipath_target = { | |
2106 | .name = "multipath", | |
636be424 | 2107 | .version = {1, 14, 0}, |
8c5c1473 SM |
2108 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE | |
2109 | DM_TARGET_PASSES_INTEGRITY, | |
1da177e4 LT |
2110 | .module = THIS_MODULE, |
2111 | .ctr = multipath_ctr, | |
2112 | .dtr = multipath_dtr, | |
e5863d9a MS |
2113 | .clone_and_map_rq = multipath_clone_and_map, |
2114 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 2115 | .rq_end_io = multipath_end_io, |
76e33fe4 MS |
2116 | .map = multipath_map_bio, |
2117 | .end_io = multipath_end_io_bio, | |
2118 | .presuspend = multipath_presuspend, | |
2119 | .postsuspend = multipath_postsuspend, | |
2120 | .resume = multipath_resume, | |
2121 | .status = multipath_status, | |
2122 | .message = multipath_message, | |
2123 | .prepare_ioctl = multipath_prepare_ioctl, | |
2124 | .iterate_devices = multipath_iterate_devices, | |
2125 | .busy = multipath_busy, | |
2126 | }; | |
2127 | ||
1da177e4 LT |
2128 | static int __init dm_multipath_init(void) |
2129 | { | |
2130 | int r; | |
2131 | ||
4d4d66ab | 2132 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 2133 | if (!kmultipathd) { |
0cd33124 | 2134 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
2135 | r = -ENOMEM; |
2136 | goto bad_alloc_kmultipathd; | |
c557308e AK |
2137 | } |
2138 | ||
bab7cfc7 CS |
2139 | /* |
2140 | * A separate workqueue is used to handle the device handlers | |
2141 | * to avoid overloading existing workqueue. Overloading the | |
2142 | * old workqueue would also create a bottleneck in the | |
2143 | * path of the storage hardware device activation. | |
2144 | */ | |
4d4d66ab TH |
2145 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
2146 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
2147 | if (!kmpath_handlerd) { |
2148 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
2149 | r = -ENOMEM; |
2150 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
2151 | } |
2152 | ||
7e6358d2 | 2153 | r = dm_register_target(&multipath_target); |
2154 | if (r < 0) { | |
2155 | DMERR("request-based register failed %d", r); | |
2156 | r = -EINVAL; | |
2157 | goto bad_register_target; | |
2158 | } | |
2159 | ||
ff658e9c JT |
2160 | return 0; |
2161 | ||
7e6358d2 | 2162 | bad_register_target: |
2163 | destroy_workqueue(kmpath_handlerd); | |
ff658e9c JT |
2164 | bad_alloc_kmpath_handlerd: |
2165 | destroy_workqueue(kmultipathd); | |
2166 | bad_alloc_kmultipathd: | |
1da177e4 LT |
2167 | return r; |
2168 | } | |
2169 | ||
2170 | static void __exit dm_multipath_exit(void) | |
2171 | { | |
bab7cfc7 | 2172 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
2173 | destroy_workqueue(kmultipathd); |
2174 | ||
10d3bd09 | 2175 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
2176 | } |
2177 | ||
1da177e4 LT |
2178 | module_init(dm_multipath_init); |
2179 | module_exit(dm_multipath_exit); | |
2180 | ||
be240ff5 AP |
2181 | module_param_named(queue_if_no_path_timeout_secs, |
2182 | queue_if_no_path_timeout_secs, ulong, S_IRUGO | S_IWUSR); | |
2183 | MODULE_PARM_DESC(queue_if_no_path_timeout_secs, "No available paths queue IO timeout in seconds"); | |
2184 | ||
1da177e4 LT |
2185 | MODULE_DESCRIPTION(DM_NAME " multipath target"); |
2186 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
2187 | MODULE_LICENSE("GPL"); |