]>
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 | { | |
338 | m->current_pg = pg; | |
1da177e4 LT |
339 | |
340 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 341 | if (m->hw_handler_name) { |
518257b1 MS |
342 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
343 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 344 | } else { |
518257b1 MS |
345 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
346 | clear_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 347 | } |
c9e45581 | 348 | |
91e968aa | 349 | atomic_set(&m->pg_init_count, 0); |
1da177e4 LT |
350 | } |
351 | ||
2da1610a MS |
352 | static struct pgpath *choose_path_in_pg(struct multipath *m, |
353 | struct priority_group *pg, | |
354 | size_t nr_bytes) | |
1da177e4 | 355 | { |
2da1610a | 356 | unsigned long flags; |
c922d5f7 | 357 | struct dm_path *path; |
2da1610a | 358 | struct pgpath *pgpath; |
1da177e4 | 359 | |
90a4323c | 360 | path = pg->ps.type->select_path(&pg->ps, nr_bytes); |
1da177e4 | 361 | if (!path) |
2da1610a | 362 | return ERR_PTR(-ENXIO); |
1da177e4 | 363 | |
2da1610a | 364 | pgpath = path_to_pgpath(path); |
1da177e4 | 365 | |
506458ef | 366 | if (unlikely(READ_ONCE(m->current_pg) != pg)) { |
2da1610a MS |
367 | /* Only update current_pgpath if pg changed */ |
368 | spin_lock_irqsave(&m->lock, flags); | |
369 | m->current_pgpath = pgpath; | |
370 | __switch_pg(m, pg); | |
371 | spin_unlock_irqrestore(&m->lock, flags); | |
372 | } | |
1da177e4 | 373 | |
2da1610a | 374 | return pgpath; |
1da177e4 LT |
375 | } |
376 | ||
2da1610a | 377 | static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 | 378 | { |
2da1610a | 379 | unsigned long flags; |
1da177e4 | 380 | struct priority_group *pg; |
2da1610a | 381 | struct pgpath *pgpath; |
d19a55cc | 382 | unsigned bypassed = 1; |
1da177e4 | 383 | |
91e968aa | 384 | if (!atomic_read(&m->nr_valid_paths)) { |
8d47e659 | 385 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
1da177e4 | 386 | goto failed; |
1f271972 | 387 | } |
1da177e4 LT |
388 | |
389 | /* Were we instructed to switch PG? */ | |
506458ef | 390 | if (READ_ONCE(m->next_pg)) { |
2da1610a | 391 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 | 392 | pg = m->next_pg; |
2da1610a MS |
393 | if (!pg) { |
394 | spin_unlock_irqrestore(&m->lock, flags); | |
395 | goto check_current_pg; | |
396 | } | |
1da177e4 | 397 | m->next_pg = NULL; |
2da1610a MS |
398 | spin_unlock_irqrestore(&m->lock, flags); |
399 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
400 | if (!IS_ERR_OR_NULL(pgpath)) | |
401 | return pgpath; | |
1da177e4 LT |
402 | } |
403 | ||
404 | /* Don't change PG until it has no remaining paths */ | |
2da1610a | 405 | check_current_pg: |
506458ef | 406 | pg = READ_ONCE(m->current_pg); |
2da1610a MS |
407 | if (pg) { |
408 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
409 | if (!IS_ERR_OR_NULL(pgpath)) | |
410 | return pgpath; | |
411 | } | |
1da177e4 LT |
412 | |
413 | /* | |
414 | * Loop through priority groups until we find a valid path. | |
415 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
416 | * Second time we only try the ones we skipped, but set |
417 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
418 | */ |
419 | do { | |
420 | list_for_each_entry(pg, &m->priority_groups, list) { | |
d19a55cc | 421 | if (pg->bypassed == !!bypassed) |
1da177e4 | 422 | continue; |
2da1610a MS |
423 | pgpath = choose_path_in_pg(m, pg, nr_bytes); |
424 | if (!IS_ERR_OR_NULL(pgpath)) { | |
f220fd4e | 425 | if (!bypassed) |
518257b1 | 426 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); |
2da1610a | 427 | return pgpath; |
f220fd4e | 428 | } |
1da177e4 LT |
429 | } |
430 | } while (bypassed--); | |
431 | ||
432 | failed: | |
2da1610a | 433 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
434 | m->current_pgpath = NULL; |
435 | m->current_pg = NULL; | |
2da1610a MS |
436 | spin_unlock_irqrestore(&m->lock, flags); |
437 | ||
438 | return NULL; | |
1da177e4 LT |
439 | } |
440 | ||
45e15720 | 441 | /* |
86331f39 BVA |
442 | * dm_report_EIO() is a macro instead of a function to make pr_debug() |
443 | * report the function name and line number of the function from which | |
444 | * it has been invoked. | |
45e15720 | 445 | */ |
86331f39 | 446 | #define dm_report_EIO(m) \ |
18a482f5 | 447 | do { \ |
86331f39 BVA |
448 | struct mapped_device *md = dm_table_get_md((m)->ti->table); \ |
449 | \ | |
450 | pr_debug("%s: returning EIO; QIFNP = %d; SQIFNP = %d; DNFS = %d\n", \ | |
451 | dm_device_name(md), \ | |
452 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
453 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
454 | dm_noflush_suspending((m)->ti)); \ | |
18a482f5 | 455 | } while (0) |
45e15720 | 456 | |
c1fd0abe MS |
457 | /* |
458 | * Check whether bios must be queued in the device-mapper core rather | |
459 | * than here in the target. | |
460 | * | |
461 | * If MPATHF_QUEUE_IF_NO_PATH and MPATHF_SAVED_QUEUE_IF_NO_PATH hold | |
462 | * the same value then we are not between multipath_presuspend() | |
463 | * and multipath_resume() calls and we have no need to check | |
464 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
465 | */ | |
466 | static bool __must_push_back(struct multipath *m, unsigned long flags) | |
467 | { | |
468 | return ((test_bit(MPATHF_QUEUE_IF_NO_PATH, &flags) != | |
469 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &flags)) && | |
470 | dm_noflush_suspending(m->ti)); | |
471 | } | |
472 | ||
473 | /* | |
474 | * Following functions use READ_ONCE to get atomic access to | |
475 | * all m->flags to avoid taking spinlock | |
476 | */ | |
477 | static bool must_push_back_rq(struct multipath *m) | |
478 | { | |
479 | unsigned long flags = READ_ONCE(m->flags); | |
480 | return test_bit(MPATHF_QUEUE_IF_NO_PATH, &flags) || __must_push_back(m, flags); | |
481 | } | |
482 | ||
483 | static bool must_push_back_bio(struct multipath *m) | |
484 | { | |
485 | unsigned long flags = READ_ONCE(m->flags); | |
486 | return __must_push_back(m, flags); | |
487 | } | |
488 | ||
36fcffcc | 489 | /* |
76e33fe4 | 490 | * Map cloned requests (request-based multipath) |
36fcffcc | 491 | */ |
eb8db831 CH |
492 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, |
493 | union map_info *map_context, | |
494 | struct request **__clone) | |
1da177e4 | 495 | { |
7943bd6d | 496 | struct multipath *m = ti->private; |
eb8db831 | 497 | size_t nr_bytes = blk_rq_bytes(rq); |
1da177e4 | 498 | struct pgpath *pgpath; |
f40c67f0 | 499 | struct block_device *bdev; |
eb8db831 | 500 | struct dm_mpath_io *mpio = get_mpio(map_context); |
7083abbb | 501 | struct request_queue *q; |
eb8db831 | 502 | struct request *clone; |
1da177e4 | 503 | |
1da177e4 | 504 | /* Do we need to select a new pgpath? */ |
506458ef | 505 | pgpath = READ_ONCE(m->current_pgpath); |
2da1610a MS |
506 | if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) |
507 | pgpath = choose_pgpath(m, nr_bytes); | |
1da177e4 | 508 | |
9bf59a61 | 509 | if (!pgpath) { |
c1fd0abe | 510 | if (must_push_back_rq(m)) |
b88efd43 | 511 | return DM_MAPIO_DELAY_REQUEUE; |
18a482f5 | 512 | dm_report_EIO(m); /* Failed */ |
f98e0eb6 | 513 | return DM_MAPIO_KILL; |
518257b1 MS |
514 | } else if (test_bit(MPATHF_QUEUE_IO, &m->flags) || |
515 | test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { | |
459b5401 ML |
516 | pg_init_all_paths(m); |
517 | return DM_MAPIO_DELAY_REQUEUE; | |
9bf59a61 | 518 | } |
6afbc01d | 519 | |
2eb6e1e3 KB |
520 | mpio->pgpath = pgpath; |
521 | mpio->nr_bytes = nr_bytes; | |
522 | ||
9bf59a61 | 523 | bdev = pgpath->path.dev->bdev; |
7083abbb | 524 | q = bdev_get_queue(bdev); |
ff005a06 CH |
525 | clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, |
526 | BLK_MQ_REQ_NOWAIT); | |
eb8db831 CH |
527 | if (IS_ERR(clone)) { |
528 | /* EBUSY, ENODEV or EWOULDBLOCK: requeue */ | |
848b8aef | 529 | if (blk_queue_dying(q)) { |
7083abbb BVA |
530 | atomic_inc(&m->pg_init_in_progress); |
531 | activate_or_offline_path(pgpath); | |
050af08f | 532 | return DM_MAPIO_DELAY_REQUEUE; |
7083abbb | 533 | } |
050af08f ML |
534 | |
535 | /* | |
536 | * blk-mq's SCHED_RESTART can cover this requeue, so we | |
537 | * needn't deal with it by DELAY_REQUEUE. More importantly, | |
538 | * we have to return DM_MAPIO_REQUEUE so that blk-mq can | |
539 | * get the queue busy feedback (via BLK_STS_RESOURCE), | |
540 | * otherwise I/O merging can suffer. | |
541 | */ | |
6a23e05c | 542 | return DM_MAPIO_REQUEUE; |
e5863d9a | 543 | } |
eb8db831 CH |
544 | clone->bio = clone->biotail = NULL; |
545 | clone->rq_disk = bdev->bd_disk; | |
546 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
547 | *__clone = clone; | |
e5863d9a | 548 | |
9bf59a61 MS |
549 | if (pgpath->pg->ps.type->start_io) |
550 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
551 | &pgpath->path, | |
552 | nr_bytes); | |
2eb6e1e3 | 553 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
554 | } |
555 | ||
5de719e3 YY |
556 | static void multipath_release_clone(struct request *clone, |
557 | union map_info *map_context) | |
e5863d9a | 558 | { |
5de719e3 YY |
559 | if (unlikely(map_context)) { |
560 | /* | |
561 | * non-NULL map_context means caller is still map | |
562 | * method; must undo multipath_clone_and_map() | |
563 | */ | |
564 | struct dm_mpath_io *mpio = get_mpio(map_context); | |
565 | struct pgpath *pgpath = mpio->pgpath; | |
566 | ||
567 | if (pgpath && pgpath->pg->ps.type->end_io) | |
568 | pgpath->pg->ps.type->end_io(&pgpath->pg->ps, | |
569 | &pgpath->path, | |
570 | mpio->nr_bytes); | |
571 | } | |
572 | ||
eb8db831 | 573 | blk_put_request(clone); |
e5863d9a MS |
574 | } |
575 | ||
76e33fe4 MS |
576 | /* |
577 | * Map cloned bios (bio-based multipath) | |
578 | */ | |
0001ec56 MS |
579 | |
580 | static struct pgpath *__map_bio(struct multipath *m, struct bio *bio) | |
76e33fe4 | 581 | { |
76e33fe4 MS |
582 | struct pgpath *pgpath; |
583 | unsigned long flags; | |
584 | bool queue_io; | |
585 | ||
586 | /* Do we need to select a new pgpath? */ | |
506458ef | 587 | pgpath = READ_ONCE(m->current_pgpath); |
5686dee3 | 588 | if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) |
0001ec56 | 589 | pgpath = choose_pgpath(m, bio->bi_iter.bi_size); |
76e33fe4 | 590 | |
5686dee3 GKB |
591 | /* MPATHF_QUEUE_IO might have been cleared by choose_pgpath. */ |
592 | queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags); | |
593 | ||
76e33fe4 MS |
594 | if ((pgpath && queue_io) || |
595 | (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) { | |
596 | /* Queue for the daemon to resubmit */ | |
597 | spin_lock_irqsave(&m->lock, flags); | |
598 | bio_list_add(&m->queued_bios, bio); | |
599 | spin_unlock_irqrestore(&m->lock, flags); | |
848b8aef | 600 | |
76e33fe4 MS |
601 | /* PG_INIT_REQUIRED cannot be set without QUEUE_IO */ |
602 | if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) | |
603 | pg_init_all_paths(m); | |
604 | else if (!queue_io) | |
605 | queue_work(kmultipathd, &m->process_queued_bios); | |
0001ec56 MS |
606 | |
607 | return ERR_PTR(-EAGAIN); | |
76e33fe4 MS |
608 | } |
609 | ||
0001ec56 MS |
610 | return pgpath; |
611 | } | |
612 | ||
0001ec56 MS |
613 | static int __multipath_map_bio(struct multipath *m, struct bio *bio, |
614 | struct dm_mpath_io *mpio) | |
615 | { | |
dbaf971c | 616 | struct pgpath *pgpath = __map_bio(m, bio); |
848b8aef | 617 | |
0001ec56 | 618 | if (IS_ERR(pgpath)) |
76e33fe4 | 619 | return DM_MAPIO_SUBMITTED; |
76e33fe4 MS |
620 | |
621 | if (!pgpath) { | |
c1fd0abe | 622 | if (must_push_back_bio(m)) |
ca5beb76 | 623 | return DM_MAPIO_REQUEUE; |
18a482f5 | 624 | dm_report_EIO(m); |
846785e6 | 625 | return DM_MAPIO_KILL; |
76e33fe4 MS |
626 | } |
627 | ||
628 | mpio->pgpath = pgpath; | |
76e33fe4 | 629 | |
4e4cbee9 | 630 | bio->bi_status = 0; |
74d46992 | 631 | bio_set_dev(bio, pgpath->path.dev->bdev); |
1eff9d32 | 632 | bio->bi_opf |= REQ_FAILFAST_TRANSPORT; |
76e33fe4 MS |
633 | |
634 | if (pgpath->pg->ps.type->start_io) | |
635 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
636 | &pgpath->path, | |
d0442f80 | 637 | mpio->nr_bytes); |
76e33fe4 MS |
638 | return DM_MAPIO_REMAPPED; |
639 | } | |
640 | ||
641 | static int multipath_map_bio(struct dm_target *ti, struct bio *bio) | |
642 | { | |
643 | struct multipath *m = ti->private; | |
bf661be1 MS |
644 | struct dm_mpath_io *mpio = NULL; |
645 | ||
63f6e6fd | 646 | multipath_init_per_bio_data(bio, &mpio); |
76e33fe4 MS |
647 | return __multipath_map_bio(m, bio, mpio); |
648 | } | |
649 | ||
7e48c768 | 650 | static void process_queued_io_list(struct multipath *m) |
76e33fe4 | 651 | { |
953923c0 | 652 | if (m->queue_mode == DM_TYPE_REQUEST_BASED) |
7e48c768 | 653 | dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table)); |
8d47e659 | 654 | else if (m->queue_mode == DM_TYPE_BIO_BASED) |
76e33fe4 MS |
655 | queue_work(kmultipathd, &m->process_queued_bios); |
656 | } | |
657 | ||
658 | static void process_queued_bios(struct work_struct *work) | |
659 | { | |
660 | int r; | |
661 | unsigned long flags; | |
662 | struct bio *bio; | |
663 | struct bio_list bios; | |
664 | struct blk_plug plug; | |
665 | struct multipath *m = | |
666 | container_of(work, struct multipath, process_queued_bios); | |
667 | ||
668 | bio_list_init(&bios); | |
669 | ||
670 | spin_lock_irqsave(&m->lock, flags); | |
671 | ||
672 | if (bio_list_empty(&m->queued_bios)) { | |
673 | spin_unlock_irqrestore(&m->lock, flags); | |
674 | return; | |
675 | } | |
676 | ||
677 | bio_list_merge(&bios, &m->queued_bios); | |
678 | bio_list_init(&m->queued_bios); | |
679 | ||
680 | spin_unlock_irqrestore(&m->lock, flags); | |
681 | ||
682 | blk_start_plug(&plug); | |
683 | while ((bio = bio_list_pop(&bios))) { | |
1836df08 MS |
684 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); |
685 | dm_bio_restore(get_bio_details_from_mpio(mpio), bio); | |
686 | r = __multipath_map_bio(m, bio, mpio); | |
846785e6 CH |
687 | switch (r) { |
688 | case DM_MAPIO_KILL: | |
4e4cbee9 CH |
689 | bio->bi_status = BLK_STS_IOERR; |
690 | bio_endio(bio); | |
047385b3 | 691 | break; |
846785e6 | 692 | case DM_MAPIO_REQUEUE: |
4e4cbee9 | 693 | bio->bi_status = BLK_STS_DM_REQUEUE; |
76e33fe4 | 694 | bio_endio(bio); |
846785e6 CH |
695 | break; |
696 | case DM_MAPIO_REMAPPED: | |
76e33fe4 | 697 | generic_make_request(bio); |
846785e6 | 698 | break; |
8192a0cd | 699 | case DM_MAPIO_SUBMITTED: |
9157c8d3 BVA |
700 | break; |
701 | default: | |
702 | WARN_ONCE(true, "__multipath_map_bio() returned %d\n", r); | |
846785e6 | 703 | } |
76e33fe4 MS |
704 | } |
705 | blk_finish_plug(&plug); | |
706 | } | |
707 | ||
1da177e4 LT |
708 | /* |
709 | * If we run out of usable paths, should we queue I/O or error it? | |
710 | */ | |
be7d31cc MS |
711 | static int queue_if_no_path(struct multipath *m, bool queue_if_no_path, |
712 | bool save_old_value) | |
1da177e4 LT |
713 | { |
714 | unsigned long flags; | |
715 | ||
716 | spin_lock_irqsave(&m->lock, flags); | |
5307e2ad LW |
717 | assign_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags, |
718 | (save_old_value && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) || | |
719 | (!save_old_value && queue_if_no_path)); | |
c1fd0abe | 720 | assign_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags, queue_if_no_path); |
1da177e4 LT |
721 | spin_unlock_irqrestore(&m->lock, flags); |
722 | ||
76e33fe4 | 723 | if (!queue_if_no_path) { |
63d832c3 | 724 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 725 | process_queued_io_list(m); |
76e33fe4 | 726 | } |
63d832c3 | 727 | |
1da177e4 LT |
728 | return 0; |
729 | } | |
730 | ||
be240ff5 AP |
731 | /* |
732 | * If the queue_if_no_path timeout fires, turn off queue_if_no_path and | |
733 | * process any queued I/O. | |
734 | */ | |
735 | static void queue_if_no_path_timeout_work(struct timer_list *t) | |
736 | { | |
737 | struct multipath *m = from_timer(m, t, nopath_timer); | |
738 | struct mapped_device *md = dm_table_get_md(m->ti->table); | |
739 | ||
740 | DMWARN("queue_if_no_path timeout on %s, failing queued IO", dm_device_name(md)); | |
741 | queue_if_no_path(m, false, false); | |
742 | } | |
743 | ||
744 | /* | |
745 | * Enable the queue_if_no_path timeout if necessary. | |
746 | * Called with m->lock held. | |
747 | */ | |
748 | static void enable_nopath_timeout(struct multipath *m) | |
749 | { | |
750 | unsigned long queue_if_no_path_timeout = | |
751 | READ_ONCE(queue_if_no_path_timeout_secs) * HZ; | |
752 | ||
753 | lockdep_assert_held(&m->lock); | |
754 | ||
755 | if (queue_if_no_path_timeout > 0 && | |
756 | atomic_read(&m->nr_valid_paths) == 0 && | |
757 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
758 | mod_timer(&m->nopath_timer, | |
759 | jiffies + queue_if_no_path_timeout); | |
760 | } | |
761 | } | |
762 | ||
763 | static void disable_nopath_timeout(struct multipath *m) | |
764 | { | |
765 | del_timer_sync(&m->nopath_timer); | |
766 | } | |
767 | ||
1da177e4 LT |
768 | /* |
769 | * An event is triggered whenever a path is taken out of use. | |
770 | * Includes path failure and PG bypass. | |
771 | */ | |
c4028958 | 772 | static void trigger_event(struct work_struct *work) |
1da177e4 | 773 | { |
c4028958 DH |
774 | struct multipath *m = |
775 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
776 | |
777 | dm_table_event(m->ti->table); | |
778 | } | |
779 | ||
780 | /*----------------------------------------------------------------- | |
781 | * Constructor/argument parsing: | |
782 | * <#multipath feature args> [<arg>]* | |
783 | * <#hw_handler args> [hw_handler [<arg>]*] | |
784 | * <#priority groups> | |
785 | * <initial priority group> | |
786 | * [<selector> <#selector args> [<arg>]* | |
787 | * <#paths> <#per-path selector args> | |
788 | * [<path> [<arg>]* ]+ ]+ | |
789 | *---------------------------------------------------------------*/ | |
498f0103 | 790 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
791 | struct dm_target *ti) |
792 | { | |
793 | int r; | |
794 | struct path_selector_type *pst; | |
795 | unsigned ps_argc; | |
796 | ||
5916a22b | 797 | static const struct dm_arg _args[] = { |
72d94861 | 798 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
799 | }; |
800 | ||
498f0103 | 801 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 802 | if (!pst) { |
72d94861 | 803 | ti->error = "unknown path selector type"; |
1da177e4 LT |
804 | return -EINVAL; |
805 | } | |
806 | ||
498f0103 | 807 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
808 | if (r) { |
809 | dm_put_path_selector(pst); | |
1da177e4 | 810 | return -EINVAL; |
371b2e34 | 811 | } |
1da177e4 LT |
812 | |
813 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
814 | if (r) { | |
815 | dm_put_path_selector(pst); | |
72d94861 | 816 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
817 | return r; |
818 | } | |
819 | ||
820 | pg->ps.type = pst; | |
498f0103 | 821 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
822 | |
823 | return 0; | |
824 | } | |
825 | ||
e8f74a0f | 826 | static int setup_scsi_dh(struct block_device *bdev, struct multipath *m, |
b592211c | 827 | const char **attached_handler_name, char **error) |
1da177e4 | 828 | { |
848b8aef | 829 | struct request_queue *q = bdev_get_queue(bdev); |
848b8aef | 830 | int r; |
a58a935d | 831 | |
518257b1 | 832 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) { |
1bab0de0 | 833 | retain: |
b592211c | 834 | if (*attached_handler_name) { |
54cd640d | 835 | /* |
836 | * Clear any hw_handler_params associated with a | |
837 | * handler that isn't already attached. | |
838 | */ | |
b592211c | 839 | if (m->hw_handler_name && strcmp(*attached_handler_name, m->hw_handler_name)) { |
54cd640d | 840 | kfree(m->hw_handler_params); |
841 | m->hw_handler_params = NULL; | |
842 | } | |
843 | ||
a58a935d MS |
844 | /* |
845 | * Reset hw_handler_name to match the attached handler | |
a58a935d MS |
846 | * |
847 | * NB. This modifies the table line to show the actual | |
848 | * handler instead of the original table passed in. | |
849 | */ | |
850 | kfree(m->hw_handler_name); | |
b592211c MS |
851 | m->hw_handler_name = *attached_handler_name; |
852 | *attached_handler_name = NULL; | |
a58a935d MS |
853 | } |
854 | } | |
a0cf7ea9 | 855 | |
a58a935d | 856 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
857 | r = scsi_dh_attach(q, m->hw_handler_name); |
858 | if (r == -EBUSY) { | |
1bab0de0 | 859 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 860 | |
1bab0de0 | 861 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
848b8aef | 862 | bdevname(bdev, b)); |
1bab0de0 CH |
863 | goto retain; |
864 | } | |
ae11b1b3 | 865 | if (r < 0) { |
848b8aef MS |
866 | *error = "error attaching hardware handler"; |
867 | return r; | |
ae11b1b3 | 868 | } |
2bfd2e13 CS |
869 | |
870 | if (m->hw_handler_params) { | |
871 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
872 | if (r < 0) { | |
848b8aef MS |
873 | *error = "unable to set hardware handler parameters"; |
874 | return r; | |
2bfd2e13 CS |
875 | } |
876 | } | |
ae11b1b3 HR |
877 | } |
878 | ||
848b8aef MS |
879 | return 0; |
880 | } | |
881 | ||
882 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, | |
883 | struct dm_target *ti) | |
884 | { | |
885 | int r; | |
886 | struct pgpath *p; | |
887 | struct multipath *m = ti->private; | |
e8f74a0f | 888 | struct request_queue *q; |
b592211c | 889 | const char *attached_handler_name = NULL; |
848b8aef MS |
890 | |
891 | /* we need at least a path arg */ | |
892 | if (as->argc < 1) { | |
893 | ti->error = "no device given"; | |
894 | return ERR_PTR(-EINVAL); | |
895 | } | |
896 | ||
897 | p = alloc_pgpath(); | |
898 | if (!p) | |
899 | return ERR_PTR(-ENOMEM); | |
900 | ||
901 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), | |
902 | &p->path.dev); | |
903 | if (r) { | |
904 | ti->error = "error getting device"; | |
905 | goto bad; | |
906 | } | |
907 | ||
e8f74a0f MS |
908 | q = bdev_get_queue(p->path.dev->bdev); |
909 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); | |
e457edf0 | 910 | if (attached_handler_name || m->hw_handler_name) { |
848b8aef | 911 | INIT_DELAYED_WORK(&p->activate_path, activate_path_work); |
b592211c | 912 | r = setup_scsi_dh(p->path.dev->bdev, m, &attached_handler_name, &ti->error); |
940bc471 | 913 | kfree(attached_handler_name); |
848b8aef MS |
914 | if (r) { |
915 | dm_put_device(ti, p->path.dev); | |
916 | goto bad; | |
917 | } | |
918 | } | |
919 | ||
1da177e4 LT |
920 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
921 | if (r) { | |
922 | dm_put_device(ti, p->path.dev); | |
923 | goto bad; | |
924 | } | |
925 | ||
926 | return p; | |
1da177e4 LT |
927 | bad: |
928 | free_pgpath(p); | |
01460f35 | 929 | return ERR_PTR(r); |
1da177e4 LT |
930 | } |
931 | ||
498f0103 | 932 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 933 | struct multipath *m) |
1da177e4 | 934 | { |
5916a22b | 935 | static const struct dm_arg _args[] = { |
72d94861 AK |
936 | {1, 1024, "invalid number of paths"}, |
937 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
938 | }; |
939 | ||
940 | int r; | |
498f0103 | 941 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 942 | struct priority_group *pg; |
28f16c20 | 943 | struct dm_target *ti = m->ti; |
1da177e4 LT |
944 | |
945 | if (as->argc < 2) { | |
946 | as->argc = 0; | |
01460f35 BM |
947 | ti->error = "not enough priority group arguments"; |
948 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
949 | } |
950 | ||
951 | pg = alloc_priority_group(); | |
952 | if (!pg) { | |
72d94861 | 953 | ti->error = "couldn't allocate priority group"; |
01460f35 | 954 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
955 | } |
956 | pg->m = m; | |
957 | ||
958 | r = parse_path_selector(as, pg, ti); | |
959 | if (r) | |
960 | goto bad; | |
961 | ||
962 | /* | |
963 | * read the paths | |
964 | */ | |
498f0103 | 965 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
966 | if (r) |
967 | goto bad; | |
968 | ||
498f0103 | 969 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
970 | if (r) |
971 | goto bad; | |
972 | ||
498f0103 | 973 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
974 | for (i = 0; i < pg->nr_pgpaths; i++) { |
975 | struct pgpath *pgpath; | |
498f0103 | 976 | struct dm_arg_set path_args; |
1da177e4 | 977 | |
498f0103 | 978 | if (as->argc < nr_args) { |
148acff6 | 979 | ti->error = "not enough path parameters"; |
6bbf79a1 | 980 | r = -EINVAL; |
1da177e4 | 981 | goto bad; |
148acff6 | 982 | } |
1da177e4 | 983 | |
498f0103 | 984 | path_args.argc = nr_args; |
1da177e4 LT |
985 | path_args.argv = as->argv; |
986 | ||
987 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
988 | if (IS_ERR(pgpath)) { |
989 | r = PTR_ERR(pgpath); | |
1da177e4 | 990 | goto bad; |
01460f35 | 991 | } |
1da177e4 LT |
992 | |
993 | pgpath->pg = pg; | |
994 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 995 | dm_consume_args(as, nr_args); |
1da177e4 LT |
996 | } |
997 | ||
998 | return pg; | |
999 | ||
1000 | bad: | |
1001 | free_priority_group(pg, ti); | |
01460f35 | 1002 | return ERR_PTR(r); |
1da177e4 LT |
1003 | } |
1004 | ||
498f0103 | 1005 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 1006 | { |
1da177e4 | 1007 | unsigned hw_argc; |
2bfd2e13 | 1008 | int ret; |
28f16c20 | 1009 | struct dm_target *ti = m->ti; |
1da177e4 | 1010 | |
5916a22b | 1011 | static const struct dm_arg _args[] = { |
72d94861 | 1012 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
1013 | }; |
1014 | ||
498f0103 | 1015 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
1016 | return -EINVAL; |
1017 | ||
1018 | if (!hw_argc) | |
1019 | return 0; | |
1020 | ||
8d47e659 | 1021 | if (m->queue_mode == DM_TYPE_BIO_BASED) { |
76e33fe4 MS |
1022 | dm_consume_args(as, hw_argc); |
1023 | DMERR("bio-based multipath doesn't allow hardware handler args"); | |
1024 | return 0; | |
1025 | } | |
1026 | ||
498f0103 | 1027 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
f97dc421 | 1028 | if (!m->hw_handler_name) |
1029 | return -EINVAL; | |
14e98c5c | 1030 | |
2bfd2e13 CS |
1031 | if (hw_argc > 1) { |
1032 | char *p; | |
1033 | int i, j, len = 4; | |
1034 | ||
1035 | for (i = 0; i <= hw_argc - 2; i++) | |
1036 | len += strlen(as->argv[i]) + 1; | |
1037 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
1038 | if (!p) { | |
1039 | ti->error = "memory allocation failed"; | |
1040 | ret = -ENOMEM; | |
1041 | goto fail; | |
1042 | } | |
1043 | j = sprintf(p, "%d", hw_argc - 1); | |
1044 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
1045 | j = sprintf(p, "%s", as->argv[i]); | |
1046 | } | |
498f0103 | 1047 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
1048 | |
1049 | return 0; | |
2bfd2e13 CS |
1050 | fail: |
1051 | kfree(m->hw_handler_name); | |
1052 | m->hw_handler_name = NULL; | |
1053 | return ret; | |
1da177e4 LT |
1054 | } |
1055 | ||
498f0103 | 1056 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
1057 | { |
1058 | int r; | |
1059 | unsigned argc; | |
28f16c20 | 1060 | struct dm_target *ti = m->ti; |
498f0103 | 1061 | const char *arg_name; |
1da177e4 | 1062 | |
5916a22b | 1063 | static const struct dm_arg _args[] = { |
e83068a5 | 1064 | {0, 8, "invalid number of feature args"}, |
c9e45581 | 1065 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 1066 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
1067 | }; |
1068 | ||
498f0103 | 1069 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
1070 | if (r) |
1071 | return -EINVAL; | |
1072 | ||
1073 | if (!argc) | |
1074 | return 0; | |
1075 | ||
c9e45581 | 1076 | do { |
498f0103 | 1077 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
1078 | argc--; |
1079 | ||
498f0103 | 1080 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
be7d31cc | 1081 | r = queue_if_no_path(m, true, false); |
c9e45581 DW |
1082 | continue; |
1083 | } | |
1084 | ||
a58a935d | 1085 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
518257b1 | 1086 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); |
a58a935d MS |
1087 | continue; |
1088 | } | |
1089 | ||
498f0103 | 1090 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 1091 | (argc >= 1)) { |
498f0103 | 1092 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
1093 | argc--; |
1094 | continue; | |
1095 | } | |
1096 | ||
498f0103 | 1097 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 1098 | (argc >= 1)) { |
498f0103 | 1099 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
1100 | argc--; |
1101 | continue; | |
1102 | } | |
1103 | ||
e83068a5 MS |
1104 | if (!strcasecmp(arg_name, "queue_mode") && |
1105 | (argc >= 1)) { | |
1106 | const char *queue_mode_name = dm_shift_arg(as); | |
1107 | ||
1108 | if (!strcasecmp(queue_mode_name, "bio")) | |
1109 | m->queue_mode = DM_TYPE_BIO_BASED; | |
953923c0 MS |
1110 | else if (!strcasecmp(queue_mode_name, "rq") || |
1111 | !strcasecmp(queue_mode_name, "mq")) | |
e83068a5 | 1112 | m->queue_mode = DM_TYPE_REQUEST_BASED; |
e83068a5 MS |
1113 | else { |
1114 | ti->error = "Unknown 'queue_mode' requested"; | |
1115 | r = -EINVAL; | |
1116 | } | |
1117 | argc--; | |
1118 | continue; | |
1119 | } | |
1120 | ||
1da177e4 | 1121 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
1122 | r = -EINVAL; |
1123 | } while (argc && !r); | |
1124 | ||
1125 | return r; | |
1da177e4 LT |
1126 | } |
1127 | ||
e83068a5 | 1128 | static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv) |
1da177e4 | 1129 | { |
498f0103 | 1130 | /* target arguments */ |
5916a22b | 1131 | static const struct dm_arg _args[] = { |
a490a07a MS |
1132 | {0, 1024, "invalid number of priority groups"}, |
1133 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
1134 | }; |
1135 | ||
1136 | int r; | |
1137 | struct multipath *m; | |
498f0103 | 1138 | struct dm_arg_set as; |
1da177e4 LT |
1139 | unsigned pg_count = 0; |
1140 | unsigned next_pg_num; | |
be240ff5 | 1141 | unsigned long flags; |
1da177e4 LT |
1142 | |
1143 | as.argc = argc; | |
1144 | as.argv = argv; | |
1145 | ||
e83068a5 | 1146 | m = alloc_multipath(ti); |
1da177e4 | 1147 | if (!m) { |
72d94861 | 1148 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
1149 | return -EINVAL; |
1150 | } | |
1151 | ||
28f16c20 | 1152 | r = parse_features(&as, m); |
1da177e4 LT |
1153 | if (r) |
1154 | goto bad; | |
1155 | ||
e83068a5 MS |
1156 | r = alloc_multipath_stage2(ti, m); |
1157 | if (r) | |
1158 | goto bad; | |
1159 | ||
28f16c20 | 1160 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
1161 | if (r) |
1162 | goto bad; | |
1163 | ||
498f0103 | 1164 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
1165 | if (r) |
1166 | goto bad; | |
1167 | ||
498f0103 | 1168 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
1169 | if (r) |
1170 | goto bad; | |
1171 | ||
a490a07a MS |
1172 | if ((!m->nr_priority_groups && next_pg_num) || |
1173 | (m->nr_priority_groups && !next_pg_num)) { | |
1174 | ti->error = "invalid initial priority group"; | |
1175 | r = -EINVAL; | |
1176 | goto bad; | |
1177 | } | |
1178 | ||
1da177e4 LT |
1179 | /* parse the priority groups */ |
1180 | while (as.argc) { | |
1181 | struct priority_group *pg; | |
91e968aa | 1182 | unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths); |
1da177e4 | 1183 | |
28f16c20 | 1184 | pg = parse_priority_group(&as, m); |
01460f35 BM |
1185 | if (IS_ERR(pg)) { |
1186 | r = PTR_ERR(pg); | |
1da177e4 LT |
1187 | goto bad; |
1188 | } | |
1189 | ||
91e968aa MS |
1190 | nr_valid_paths += pg->nr_pgpaths; |
1191 | atomic_set(&m->nr_valid_paths, nr_valid_paths); | |
1192 | ||
1da177e4 LT |
1193 | list_add_tail(&pg->list, &m->priority_groups); |
1194 | pg_count++; | |
1195 | pg->pg_num = pg_count; | |
1196 | if (!--next_pg_num) | |
1197 | m->next_pg = pg; | |
1198 | } | |
1199 | ||
1200 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 1201 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
1202 | r = -EINVAL; |
1203 | goto bad; | |
1204 | } | |
1205 | ||
be240ff5 AP |
1206 | spin_lock_irqsave(&m->lock, flags); |
1207 | enable_nopath_timeout(m); | |
1208 | spin_unlock_irqrestore(&m->lock, flags); | |
1209 | ||
55a62eef AK |
1210 | ti->num_flush_bios = 1; |
1211 | ti->num_discard_bios = 1; | |
042bcef8 | 1212 | ti->num_write_same_bios = 1; |
ac62d620 | 1213 | ti->num_write_zeroes_bios = 1; |
8d47e659 | 1214 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
bf661be1 | 1215 | ti->per_io_data_size = multipath_per_bio_data_size(); |
eb8db831 | 1216 | else |
8637a6bf | 1217 | ti->per_io_data_size = sizeof(struct dm_mpath_io); |
8627921f | 1218 | |
1da177e4 LT |
1219 | return 0; |
1220 | ||
1221 | bad: | |
1222 | free_multipath(m); | |
1223 | return r; | |
1224 | } | |
1225 | ||
2bded7bd KU |
1226 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
1227 | { | |
9f4c3f87 | 1228 | DEFINE_WAIT(wait); |
2bded7bd KU |
1229 | |
1230 | while (1) { | |
9f4c3f87 | 1231 | prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE); |
2bded7bd | 1232 | |
91e968aa | 1233 | if (!atomic_read(&m->pg_init_in_progress)) |
2bded7bd | 1234 | break; |
2bded7bd KU |
1235 | |
1236 | io_schedule(); | |
1237 | } | |
9f4c3f87 | 1238 | finish_wait(&m->pg_init_wait, &wait); |
2bded7bd KU |
1239 | } |
1240 | ||
1241 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 1242 | { |
848b8aef MS |
1243 | if (m->hw_handler_name) { |
1244 | set_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1245 | smp_mb__after_atomic(); | |
1246 | ||
935fcc56 | 1247 | if (atomic_read(&m->pg_init_in_progress)) |
1248 | flush_workqueue(kmpath_handlerd); | |
848b8aef MS |
1249 | multipath_wait_for_pg_init_completion(m); |
1250 | ||
1251 | clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1252 | smp_mb__after_atomic(); | |
1253 | } | |
954a73d5 | 1254 | |
935fcc56 | 1255 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
1256 | flush_work(&m->process_queued_bios); | |
43829731 | 1257 | flush_work(&m->trigger_event); |
6df400ab KU |
1258 | } |
1259 | ||
1260 | static void multipath_dtr(struct dm_target *ti) | |
1261 | { | |
1262 | struct multipath *m = ti->private; | |
1263 | ||
be240ff5 | 1264 | disable_nopath_timeout(m); |
2bded7bd | 1265 | flush_multipath_work(m); |
1da177e4 LT |
1266 | free_multipath(m); |
1267 | } | |
1268 | ||
1da177e4 LT |
1269 | /* |
1270 | * Take a path out of use. | |
1271 | */ | |
1272 | static int fail_path(struct pgpath *pgpath) | |
1273 | { | |
1274 | unsigned long flags; | |
1275 | struct multipath *m = pgpath->pg->m; | |
1276 | ||
1277 | spin_lock_irqsave(&m->lock, flags); | |
1278 | ||
6680073d | 1279 | if (!pgpath->is_active) |
1da177e4 LT |
1280 | goto out; |
1281 | ||
72d94861 | 1282 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1283 | |
1284 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
be7d31cc | 1285 | pgpath->is_active = false; |
1da177e4 LT |
1286 | pgpath->fail_count++; |
1287 | ||
91e968aa | 1288 | atomic_dec(&m->nr_valid_paths); |
1da177e4 LT |
1289 | |
1290 | if (pgpath == m->current_pgpath) | |
1291 | m->current_pgpath = NULL; | |
1292 | ||
b15546f9 | 1293 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
91e968aa | 1294 | pgpath->path.dev->name, atomic_read(&m->nr_valid_paths)); |
b15546f9 | 1295 | |
fe9cf30e | 1296 | schedule_work(&m->trigger_event); |
1da177e4 | 1297 | |
be240ff5 AP |
1298 | enable_nopath_timeout(m); |
1299 | ||
1da177e4 LT |
1300 | out: |
1301 | spin_unlock_irqrestore(&m->lock, flags); | |
1302 | ||
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * Reinstate a previously-failed path | |
1308 | */ | |
1309 | static int reinstate_path(struct pgpath *pgpath) | |
1310 | { | |
63d832c3 | 1311 | int r = 0, run_queue = 0; |
1da177e4 LT |
1312 | unsigned long flags; |
1313 | struct multipath *m = pgpath->pg->m; | |
91e968aa | 1314 | unsigned nr_valid_paths; |
1da177e4 LT |
1315 | |
1316 | spin_lock_irqsave(&m->lock, flags); | |
1317 | ||
6680073d | 1318 | if (pgpath->is_active) |
1da177e4 LT |
1319 | goto out; |
1320 | ||
ec31f3f7 | 1321 | DMWARN("Reinstating path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1322 | |
1323 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1324 | if (r) | |
1325 | goto out; | |
1326 | ||
be7d31cc | 1327 | pgpath->is_active = true; |
1da177e4 | 1328 | |
91e968aa MS |
1329 | nr_valid_paths = atomic_inc_return(&m->nr_valid_paths); |
1330 | if (nr_valid_paths == 1) { | |
e54f77dd | 1331 | m->current_pgpath = NULL; |
63d832c3 | 1332 | run_queue = 1; |
e54f77dd | 1333 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1334 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
91e968aa | 1335 | atomic_inc(&m->pg_init_in_progress); |
e54f77dd | 1336 | } |
1da177e4 | 1337 | |
b15546f9 | 1338 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
91e968aa | 1339 | pgpath->path.dev->name, nr_valid_paths); |
b15546f9 | 1340 | |
fe9cf30e | 1341 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1342 | |
1343 | out: | |
1344 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 1345 | if (run_queue) { |
63d832c3 | 1346 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1347 | process_queued_io_list(m); |
76e33fe4 | 1348 | } |
1da177e4 | 1349 | |
be240ff5 AP |
1350 | if (pgpath->is_active) |
1351 | disable_nopath_timeout(m); | |
1352 | ||
1da177e4 LT |
1353 | return r; |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1358 | */ | |
1359 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1360 | action_fn action) | |
1361 | { | |
19040c0b | 1362 | int r = -EINVAL; |
1da177e4 LT |
1363 | struct pgpath *pgpath; |
1364 | struct priority_group *pg; | |
1365 | ||
1366 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1367 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1368 | if (pgpath->path.dev == dev) | |
1369 | r = action(pgpath); | |
1370 | } | |
1371 | } | |
1372 | ||
1373 | return r; | |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Temporarily try to avoid having to use the specified PG | |
1378 | */ | |
1379 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
be7d31cc | 1380 | bool bypassed) |
1da177e4 LT |
1381 | { |
1382 | unsigned long flags; | |
1383 | ||
1384 | spin_lock_irqsave(&m->lock, flags); | |
1385 | ||
1386 | pg->bypassed = bypassed; | |
1387 | m->current_pgpath = NULL; | |
1388 | m->current_pg = NULL; | |
1389 | ||
1390 | spin_unlock_irqrestore(&m->lock, flags); | |
1391 | ||
fe9cf30e | 1392 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1393 | } |
1394 | ||
1395 | /* | |
1396 | * Switch to using the specified PG from the next I/O that gets mapped | |
1397 | */ | |
1398 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1399 | { | |
1400 | struct priority_group *pg; | |
1401 | unsigned pgnum; | |
1402 | unsigned long flags; | |
31998ef1 | 1403 | char dummy; |
1da177e4 | 1404 | |
31998ef1 | 1405 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1406 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1407 | DMWARN("invalid PG number supplied to switch_pg_num"); |
1408 | return -EINVAL; | |
1409 | } | |
1410 | ||
1411 | spin_lock_irqsave(&m->lock, flags); | |
1412 | list_for_each_entry(pg, &m->priority_groups, list) { | |
be7d31cc | 1413 | pg->bypassed = false; |
1da177e4 LT |
1414 | if (--pgnum) |
1415 | continue; | |
1416 | ||
1417 | m->current_pgpath = NULL; | |
1418 | m->current_pg = NULL; | |
1419 | m->next_pg = pg; | |
1420 | } | |
1421 | spin_unlock_irqrestore(&m->lock, flags); | |
1422 | ||
fe9cf30e | 1423 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1424 | return 0; |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * Set/clear bypassed status of a PG. | |
1429 | * PGs are numbered upwards from 1 in the order they were declared. | |
1430 | */ | |
be7d31cc | 1431 | static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed) |
1da177e4 LT |
1432 | { |
1433 | struct priority_group *pg; | |
1434 | unsigned pgnum; | |
31998ef1 | 1435 | char dummy; |
1da177e4 | 1436 | |
31998ef1 | 1437 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1438 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1439 | DMWARN("invalid PG number supplied to bypass_pg"); |
1440 | return -EINVAL; | |
1441 | } | |
1442 | ||
1443 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1444 | if (!--pgnum) | |
1445 | break; | |
1446 | } | |
1447 | ||
1448 | bypass_pg(m, pg, bypassed); | |
1449 | return 0; | |
1450 | } | |
1451 | ||
c9e45581 DW |
1452 | /* |
1453 | * Should we retry pg_init immediately? | |
1454 | */ | |
be7d31cc | 1455 | static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) |
c9e45581 DW |
1456 | { |
1457 | unsigned long flags; | |
be7d31cc | 1458 | bool limit_reached = false; |
c9e45581 DW |
1459 | |
1460 | spin_lock_irqsave(&m->lock, flags); | |
1461 | ||
91e968aa MS |
1462 | if (atomic_read(&m->pg_init_count) <= m->pg_init_retries && |
1463 | !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) | |
518257b1 | 1464 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
c9e45581 | 1465 | else |
be7d31cc | 1466 | limit_reached = true; |
c9e45581 DW |
1467 | |
1468 | spin_unlock_irqrestore(&m->lock, flags); | |
1469 | ||
1470 | return limit_reached; | |
1471 | } | |
1472 | ||
3ae31f6a | 1473 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1474 | { |
83c0d5d5 | 1475 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1476 | struct priority_group *pg = pgpath->pg; |
1477 | struct multipath *m = pg->m; | |
1478 | unsigned long flags; | |
be7d31cc | 1479 | bool delay_retry = false; |
cfae5c9b CS |
1480 | |
1481 | /* device or driver problems */ | |
1482 | switch (errors) { | |
1483 | case SCSI_DH_OK: | |
1484 | break; | |
1485 | case SCSI_DH_NOSYS: | |
1486 | if (!m->hw_handler_name) { | |
1487 | errors = 0; | |
1488 | break; | |
1489 | } | |
f7b934c8 MB |
1490 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1491 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1492 | /* |
1493 | * Fail path for now, so we do not ping pong | |
1494 | */ | |
1495 | fail_path(pgpath); | |
1496 | break; | |
1497 | case SCSI_DH_DEV_TEMP_BUSY: | |
1498 | /* | |
1499 | * Probably doing something like FW upgrade on the | |
1500 | * controller so try the other pg. | |
1501 | */ | |
be7d31cc | 1502 | bypass_pg(m, pg, true); |
cfae5c9b | 1503 | break; |
cfae5c9b | 1504 | case SCSI_DH_RETRY: |
4e2d19e4 | 1505 | /* Wait before retrying. */ |
4ecc5081 | 1506 | delay_retry = true; |
7b06e09a | 1507 | /* fall through */ |
cfae5c9b CS |
1508 | case SCSI_DH_IMM_RETRY: |
1509 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1510 | if (pg_init_limit_reached(m, pgpath)) | |
1511 | fail_path(pgpath); | |
1512 | errors = 0; | |
1513 | break; | |
ec31f3f7 | 1514 | case SCSI_DH_DEV_OFFLINED: |
cfae5c9b CS |
1515 | default: |
1516 | /* | |
1517 | * We probably do not want to fail the path for a device | |
1518 | * error, but this is what the old dm did. In future | |
1519 | * patches we can do more advanced handling. | |
1520 | */ | |
1521 | fail_path(pgpath); | |
1522 | } | |
1523 | ||
1524 | spin_lock_irqsave(&m->lock, flags); | |
1525 | if (errors) { | |
e54f77dd CS |
1526 | if (pgpath == m->current_pgpath) { |
1527 | DMERR("Could not failover device. Error %d.", errors); | |
1528 | m->current_pgpath = NULL; | |
1529 | m->current_pg = NULL; | |
1530 | } | |
518257b1 | 1531 | } else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
be7d31cc | 1532 | pg->bypassed = false; |
cfae5c9b | 1533 | |
91e968aa | 1534 | if (atomic_dec_return(&m->pg_init_in_progress) > 0) |
d0259bf0 KU |
1535 | /* Activations of other paths are still on going */ |
1536 | goto out; | |
1537 | ||
518257b1 MS |
1538 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { |
1539 | if (delay_retry) | |
1540 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1541 | else | |
1542 | clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1543 | ||
3e9f1be1 HR |
1544 | if (__pg_init_all_paths(m)) |
1545 | goto out; | |
1546 | } | |
518257b1 | 1547 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
d0259bf0 | 1548 | |
7e48c768 | 1549 | process_queued_io_list(m); |
76e33fe4 | 1550 | |
2bded7bd KU |
1551 | /* |
1552 | * Wake up any thread waiting to suspend. | |
1553 | */ | |
1554 | wake_up(&m->pg_init_wait); | |
1555 | ||
d0259bf0 | 1556 | out: |
cfae5c9b CS |
1557 | spin_unlock_irqrestore(&m->lock, flags); |
1558 | } | |
1559 | ||
89bfce76 | 1560 | static void activate_or_offline_path(struct pgpath *pgpath) |
bab7cfc7 | 1561 | { |
f10e06b7 | 1562 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); |
bab7cfc7 | 1563 | |
f10e06b7 MS |
1564 | if (pgpath->is_active && !blk_queue_dying(q)) |
1565 | scsi_dh_activate(q, pg_init_done, pgpath); | |
3a017509 HR |
1566 | else |
1567 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1568 | } |
1569 | ||
89bfce76 BVA |
1570 | static void activate_path_work(struct work_struct *work) |
1571 | { | |
1572 | struct pgpath *pgpath = | |
1573 | container_of(work, struct pgpath, activate_path.work); | |
1574 | ||
1575 | activate_or_offline_path(pgpath); | |
1576 | } | |
1577 | ||
b79f10ee | 1578 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
2a842aca | 1579 | blk_status_t error, union map_info *map_context) |
1da177e4 | 1580 | { |
b79f10ee CH |
1581 | struct dm_mpath_io *mpio = get_mpio(map_context); |
1582 | struct pgpath *pgpath = mpio->pgpath; | |
7ed8578a | 1583 | int r = DM_ENDIO_DONE; |
b79f10ee | 1584 | |
f40c67f0 KU |
1585 | /* |
1586 | * We don't queue any clone request inside the multipath target | |
1587 | * during end I/O handling, since those clone requests don't have | |
1588 | * bio clones. If we queue them inside the multipath target, | |
1589 | * we need to make bio clones, that requires memory allocation. | |
4cc96131 | 1590 | * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests |
f40c67f0 KU |
1591 | * don't have bio clones.) |
1592 | * Instead of queueing the clone request here, we queue the original | |
1593 | * request into dm core, which will remake a clone request and | |
1594 | * clone bios for it and resubmit it later. | |
1595 | */ | |
a1275677 | 1596 | if (error && blk_path_error(error)) { |
b79f10ee | 1597 | struct multipath *m = ti->private; |
1da177e4 | 1598 | |
ac514ffc MS |
1599 | if (error == BLK_STS_RESOURCE) |
1600 | r = DM_ENDIO_DELAY_REQUEUE; | |
1601 | else | |
1602 | r = DM_ENDIO_REQUEUE; | |
1da177e4 | 1603 | |
b79f10ee CH |
1604 | if (pgpath) |
1605 | fail_path(pgpath); | |
1da177e4 | 1606 | |
b79f10ee | 1607 | if (atomic_read(&m->nr_valid_paths) == 0 && |
c1fd0abe | 1608 | !must_push_back_rq(m)) { |
2a842aca | 1609 | if (error == BLK_STS_IOERR) |
18a482f5 | 1610 | dm_report_EIO(m); |
7ed8578a CH |
1611 | /* complete with the original error */ |
1612 | r = DM_ENDIO_DONE; | |
1613 | } | |
b79f10ee | 1614 | } |
466891f9 | 1615 | |
1da177e4 | 1616 | if (pgpath) { |
b79f10ee CH |
1617 | struct path_selector *ps = &pgpath->pg->ps; |
1618 | ||
1da177e4 | 1619 | if (ps->type->end_io) |
02ab823f | 1620 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1621 | } |
1da177e4 | 1622 | |
7ed8578a | 1623 | return r; |
1da177e4 LT |
1624 | } |
1625 | ||
4e4cbee9 | 1626 | static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, |
848b8aef | 1627 | blk_status_t *error) |
76e33fe4 | 1628 | { |
14ef1e48 CH |
1629 | struct multipath *m = ti->private; |
1630 | struct dm_mpath_io *mpio = get_mpio_from_bio(clone); | |
1631 | struct pgpath *pgpath = mpio->pgpath; | |
76e33fe4 | 1632 | unsigned long flags; |
1be56909 | 1633 | int r = DM_ENDIO_DONE; |
76e33fe4 | 1634 | |
a1275677 | 1635 | if (!*error || !blk_path_error(*error)) |
14ef1e48 | 1636 | goto done; |
76e33fe4 | 1637 | |
14ef1e48 CH |
1638 | if (pgpath) |
1639 | fail_path(pgpath); | |
76e33fe4 | 1640 | |
ca5beb76 | 1641 | if (atomic_read(&m->nr_valid_paths) == 0 && |
18a482f5 | 1642 | !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { |
c1fd0abe MS |
1643 | if (must_push_back_bio(m)) { |
1644 | r = DM_ENDIO_REQUEUE; | |
1645 | } else { | |
1646 | dm_report_EIO(m); | |
1647 | *error = BLK_STS_IOERR; | |
1648 | } | |
14ef1e48 | 1649 | goto done; |
18a482f5 | 1650 | } |
76e33fe4 | 1651 | |
76e33fe4 MS |
1652 | spin_lock_irqsave(&m->lock, flags); |
1653 | bio_list_add(&m->queued_bios, clone); | |
1654 | spin_unlock_irqrestore(&m->lock, flags); | |
1655 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) | |
1656 | queue_work(kmultipathd, &m->process_queued_bios); | |
1657 | ||
1be56909 | 1658 | r = DM_ENDIO_INCOMPLETE; |
14ef1e48 | 1659 | done: |
76e33fe4 | 1660 | if (pgpath) { |
14ef1e48 CH |
1661 | struct path_selector *ps = &pgpath->pg->ps; |
1662 | ||
76e33fe4 MS |
1663 | if (ps->type->end_io) |
1664 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); | |
1665 | } | |
1666 | ||
1be56909 | 1667 | return r; |
76e33fe4 MS |
1668 | } |
1669 | ||
1da177e4 LT |
1670 | /* |
1671 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1672 | * the last path fails we must error any remaining I/O. |
1673 | * Note that if the freeze_bdev fails while suspending, the | |
1674 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1675 | */ |
1676 | static void multipath_presuspend(struct dm_target *ti) | |
1677 | { | |
7943bd6d | 1678 | struct multipath *m = ti->private; |
1da177e4 | 1679 | |
be7d31cc | 1680 | queue_if_no_path(m, false, true); |
1da177e4 LT |
1681 | } |
1682 | ||
6df400ab KU |
1683 | static void multipath_postsuspend(struct dm_target *ti) |
1684 | { | |
6380f26f MA |
1685 | struct multipath *m = ti->private; |
1686 | ||
1687 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1688 | flush_multipath_work(m); |
6380f26f | 1689 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1690 | } |
1691 | ||
436d4108 AK |
1692 | /* |
1693 | * Restore the queue_if_no_path setting. | |
1694 | */ | |
1da177e4 LT |
1695 | static void multipath_resume(struct dm_target *ti) |
1696 | { | |
7943bd6d | 1697 | struct multipath *m = ti->private; |
1814f2e3 | 1698 | unsigned long flags; |
1da177e4 | 1699 | |
1814f2e3 | 1700 | spin_lock_irqsave(&m->lock, flags); |
5307e2ad LW |
1701 | assign_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags, |
1702 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)); | |
1814f2e3 | 1703 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1704 | } |
1705 | ||
1706 | /* | |
1707 | * Info output has the following format: | |
1708 | * num_multipath_feature_args [multipath_feature_args]* | |
1709 | * num_handler_status_args [handler_status_args]* | |
1710 | * num_groups init_group_number | |
1711 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1712 | * num_paths num_selector_args | |
1713 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1714 | * | |
1715 | * Table output has the following format (identical to the constructor string): | |
1716 | * num_feature_args [features_args]* | |
1717 | * num_handler_args hw_handler [hw_handler_args]* | |
1718 | * num_groups init_group_number | |
1719 | * [priority selector-name num_ps_args [ps_args]* | |
1720 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1721 | */ | |
fd7c092e MP |
1722 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1723 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1724 | { |
1725 | int sz = 0; | |
1726 | unsigned long flags; | |
7943bd6d | 1727 | struct multipath *m = ti->private; |
1da177e4 LT |
1728 | struct priority_group *pg; |
1729 | struct pgpath *p; | |
1730 | unsigned pg_num; | |
1731 | char state; | |
1732 | ||
1733 | spin_lock_irqsave(&m->lock, flags); | |
1734 | ||
1735 | /* Features */ | |
1736 | if (type == STATUSTYPE_INFO) | |
91e968aa MS |
1737 | DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags), |
1738 | atomic_read(&m->pg_init_count)); | |
c9e45581 | 1739 | else { |
518257b1 | 1740 | DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) + |
4e2d19e4 | 1741 | (m->pg_init_retries > 0) * 2 + |
a58a935d | 1742 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
e83068a5 MS |
1743 | test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) + |
1744 | (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2); | |
1745 | ||
518257b1 | 1746 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
c9e45581 DW |
1747 | DMEMIT("queue_if_no_path "); |
1748 | if (m->pg_init_retries) | |
1749 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1750 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1751 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
518257b1 | 1752 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) |
a58a935d | 1753 | DMEMIT("retain_attached_hw_handler "); |
e83068a5 MS |
1754 | if (m->queue_mode != DM_TYPE_REQUEST_BASED) { |
1755 | switch(m->queue_mode) { | |
1756 | case DM_TYPE_BIO_BASED: | |
1757 | DMEMIT("queue_mode bio "); | |
1758 | break; | |
7e0d574f BVA |
1759 | default: |
1760 | WARN_ON_ONCE(true); | |
1761 | break; | |
e83068a5 MS |
1762 | } |
1763 | } | |
c9e45581 | 1764 | } |
1da177e4 | 1765 | |
cfae5c9b | 1766 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1767 | DMEMIT("0 "); |
1768 | else | |
cfae5c9b | 1769 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1770 | |
1771 | DMEMIT("%u ", m->nr_priority_groups); | |
1772 | ||
1773 | if (m->next_pg) | |
1774 | pg_num = m->next_pg->pg_num; | |
1775 | else if (m->current_pg) | |
1776 | pg_num = m->current_pg->pg_num; | |
1777 | else | |
a490a07a | 1778 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1779 | |
1780 | DMEMIT("%u ", pg_num); | |
1781 | ||
1782 | switch (type) { | |
1783 | case STATUSTYPE_INFO: | |
1784 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1785 | if (pg->bypassed) | |
1786 | state = 'D'; /* Disabled */ | |
1787 | else if (pg == m->current_pg) | |
1788 | state = 'A'; /* Currently Active */ | |
1789 | else | |
1790 | state = 'E'; /* Enabled */ | |
1791 | ||
1792 | DMEMIT("%c ", state); | |
1793 | ||
1794 | if (pg->ps.type->status) | |
1795 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1796 | result + sz, | |
1797 | maxlen - sz); | |
1798 | else | |
1799 | DMEMIT("0 "); | |
1800 | ||
1801 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1802 | pg->ps.type->info_args); | |
1803 | ||
1804 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1805 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1806 | p->is_active ? "A" : "F", |
1da177e4 LT |
1807 | p->fail_count); |
1808 | if (pg->ps.type->status) | |
1809 | sz += pg->ps.type->status(&pg->ps, | |
1810 | &p->path, type, result + sz, | |
1811 | maxlen - sz); | |
1812 | } | |
1813 | } | |
1814 | break; | |
1815 | ||
1816 | case STATUSTYPE_TABLE: | |
1817 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1818 | DMEMIT("%s ", pg->ps.type->name); | |
1819 | ||
1820 | if (pg->ps.type->status) | |
1821 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1822 | result + sz, | |
1823 | maxlen - sz); | |
1824 | else | |
1825 | DMEMIT("0 "); | |
1826 | ||
1827 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1828 | pg->ps.type->table_args); | |
1829 | ||
1830 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1831 | DMEMIT("%s ", p->path.dev->name); | |
1832 | if (pg->ps.type->status) | |
1833 | sz += pg->ps.type->status(&pg->ps, | |
1834 | &p->path, type, result + sz, | |
1835 | maxlen - sz); | |
1836 | } | |
1837 | } | |
1838 | break; | |
1839 | } | |
1840 | ||
1841 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1842 | } |
1843 | ||
1eb5fa84 MS |
1844 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv, |
1845 | char *result, unsigned maxlen) | |
1da177e4 | 1846 | { |
6380f26f | 1847 | int r = -EINVAL; |
1da177e4 | 1848 | struct dm_dev *dev; |
7943bd6d | 1849 | struct multipath *m = ti->private; |
1da177e4 | 1850 | action_fn action; |
be240ff5 | 1851 | unsigned long flags; |
1da177e4 | 1852 | |
6380f26f MA |
1853 | mutex_lock(&m->work_mutex); |
1854 | ||
c2f3d24b KU |
1855 | if (dm_suspended(ti)) { |
1856 | r = -EBUSY; | |
1857 | goto out; | |
1858 | } | |
1859 | ||
1da177e4 | 1860 | if (argc == 1) { |
498f0103 | 1861 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
be7d31cc | 1862 | r = queue_if_no_path(m, true, false); |
be240ff5 AP |
1863 | spin_lock_irqsave(&m->lock, flags); |
1864 | enable_nopath_timeout(m); | |
1865 | spin_unlock_irqrestore(&m->lock, flags); | |
6380f26f | 1866 | goto out; |
498f0103 | 1867 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
be7d31cc | 1868 | r = queue_if_no_path(m, false, false); |
be240ff5 | 1869 | disable_nopath_timeout(m); |
6380f26f MA |
1870 | goto out; |
1871 | } | |
1da177e4 LT |
1872 | } |
1873 | ||
6380f26f | 1874 | if (argc != 2) { |
a356e426 | 1875 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1876 | goto out; |
1877 | } | |
1da177e4 | 1878 | |
498f0103 | 1879 | if (!strcasecmp(argv[0], "disable_group")) { |
be7d31cc | 1880 | r = bypass_pg_num(m, argv[1], true); |
6380f26f | 1881 | goto out; |
498f0103 | 1882 | } else if (!strcasecmp(argv[0], "enable_group")) { |
be7d31cc | 1883 | r = bypass_pg_num(m, argv[1], false); |
6380f26f | 1884 | goto out; |
498f0103 | 1885 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1886 | r = switch_pg_num(m, argv[1]); |
1887 | goto out; | |
498f0103 | 1888 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1889 | action = reinstate_path; |
498f0103 | 1890 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1891 | action = fail_path; |
6380f26f | 1892 | else { |
a356e426 | 1893 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1894 | goto out; |
1895 | } | |
1da177e4 | 1896 | |
8215d6ec | 1897 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1898 | if (r) { |
72d94861 | 1899 | DMWARN("message: error getting device %s", |
1da177e4 | 1900 | argv[1]); |
6380f26f | 1901 | goto out; |
1da177e4 LT |
1902 | } |
1903 | ||
1904 | r = action_dev(m, dev, action); | |
1905 | ||
1906 | dm_put_device(ti, dev); | |
1907 | ||
6380f26f MA |
1908 | out: |
1909 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1910 | return r; |
1da177e4 LT |
1911 | } |
1912 | ||
e56f81e0 | 1913 | static int multipath_prepare_ioctl(struct dm_target *ti, |
5bd5e8d8 | 1914 | struct block_device **bdev) |
9af4aa30 | 1915 | { |
35991652 | 1916 | struct multipath *m = ti->private; |
2da1610a | 1917 | struct pgpath *current_pgpath; |
35991652 MP |
1918 | int r; |
1919 | ||
506458ef | 1920 | current_pgpath = READ_ONCE(m->current_pgpath); |
2da1610a MS |
1921 | if (!current_pgpath) |
1922 | current_pgpath = choose_pgpath(m, 0); | |
9af4aa30 | 1923 | |
2da1610a | 1924 | if (current_pgpath) { |
518257b1 | 1925 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) { |
2da1610a | 1926 | *bdev = current_pgpath->path.dev->bdev; |
43e43c9e JN |
1927 | r = 0; |
1928 | } else { | |
1929 | /* pg_init has not started or completed */ | |
1930 | r = -ENOTCONN; | |
1931 | } | |
1932 | } else { | |
1933 | /* No path is available */ | |
518257b1 | 1934 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
43e43c9e JN |
1935 | r = -ENOTCONN; |
1936 | else | |
1937 | r = -EIO; | |
e90dae1f | 1938 | } |
9af4aa30 | 1939 | |
5bbbfdf6 | 1940 | if (r == -ENOTCONN) { |
506458ef | 1941 | if (!READ_ONCE(m->current_pg)) { |
3e9f1be1 | 1942 | /* Path status changed, redo selection */ |
2da1610a | 1943 | (void) choose_pgpath(m, 0); |
3e9f1be1 | 1944 | } |
518257b1 | 1945 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
2da1610a | 1946 | pg_init_all_paths(m); |
63d832c3 | 1947 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1948 | process_queued_io_list(m); |
3e9f1be1 | 1949 | } |
35991652 | 1950 | |
e56f81e0 CH |
1951 | /* |
1952 | * Only pass ioctls through if the device sizes match exactly. | |
1953 | */ | |
1954 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
1955 | return 1; | |
1956 | return r; | |
9af4aa30 MB |
1957 | } |
1958 | ||
af4874e0 MS |
1959 | static int multipath_iterate_devices(struct dm_target *ti, |
1960 | iterate_devices_callout_fn fn, void *data) | |
1961 | { | |
1962 | struct multipath *m = ti->private; | |
1963 | struct priority_group *pg; | |
1964 | struct pgpath *p; | |
1965 | int ret = 0; | |
1966 | ||
1967 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1968 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1969 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1970 | if (ret) |
1971 | goto out; | |
1972 | } | |
1973 | } | |
1974 | ||
1975 | out: | |
1976 | return ret; | |
1977 | } | |
1978 | ||
9f54cec5 | 1979 | static int pgpath_busy(struct pgpath *pgpath) |
f40c67f0 KU |
1980 | { |
1981 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1982 | ||
52b09914 | 1983 | return blk_lld_busy(q); |
f40c67f0 KU |
1984 | } |
1985 | ||
1986 | /* | |
1987 | * We return "busy", only when we can map I/Os but underlying devices | |
1988 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1989 | * the underlying queue). | |
1990 | * In other words, if we want to kill I/Os or queue them inside us | |
1991 | * due to map unavailability, we don't return "busy". Otherwise, | |
1992 | * dm core won't give us the I/Os and we can't do what we want. | |
1993 | */ | |
1994 | static int multipath_busy(struct dm_target *ti) | |
1995 | { | |
be7d31cc | 1996 | bool busy = false, has_active = false; |
f40c67f0 | 1997 | struct multipath *m = ti->private; |
2da1610a | 1998 | struct priority_group *pg, *next_pg; |
f40c67f0 | 1999 | struct pgpath *pgpath; |
f40c67f0 | 2000 | |
b88efd43 MS |
2001 | /* pg_init in progress */ |
2002 | if (atomic_read(&m->pg_init_in_progress)) | |
2da1610a MS |
2003 | return true; |
2004 | ||
b88efd43 MS |
2005 | /* no paths available, for blk-mq: rely on IO mapping to delay requeue */ |
2006 | if (!atomic_read(&m->nr_valid_paths) && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) | |
953923c0 | 2007 | return (m->queue_mode != DM_TYPE_REQUEST_BASED); |
b88efd43 | 2008 | |
f40c67f0 | 2009 | /* Guess which priority_group will be used at next mapping time */ |
506458ef WD |
2010 | pg = READ_ONCE(m->current_pg); |
2011 | next_pg = READ_ONCE(m->next_pg); | |
2012 | if (unlikely(!READ_ONCE(m->current_pgpath) && next_pg)) | |
2da1610a MS |
2013 | pg = next_pg; |
2014 | ||
2015 | if (!pg) { | |
f40c67f0 KU |
2016 | /* |
2017 | * We don't know which pg will be used at next mapping time. | |
2da1610a | 2018 | * We don't call choose_pgpath() here to avoid to trigger |
f40c67f0 KU |
2019 | * pg_init just by busy checking. |
2020 | * So we don't know whether underlying devices we will be using | |
2021 | * at next mapping time are busy or not. Just try mapping. | |
2022 | */ | |
2da1610a MS |
2023 | return busy; |
2024 | } | |
f40c67f0 KU |
2025 | |
2026 | /* | |
2027 | * If there is one non-busy active path at least, the path selector | |
2028 | * will be able to select it. So we consider such a pg as not busy. | |
2029 | */ | |
be7d31cc | 2030 | busy = true; |
2da1610a | 2031 | list_for_each_entry(pgpath, &pg->pgpaths, list) { |
f40c67f0 | 2032 | if (pgpath->is_active) { |
be7d31cc | 2033 | has_active = true; |
9f54cec5 | 2034 | if (!pgpath_busy(pgpath)) { |
be7d31cc | 2035 | busy = false; |
f40c67f0 KU |
2036 | break; |
2037 | } | |
2038 | } | |
2da1610a | 2039 | } |
f40c67f0 | 2040 | |
2da1610a | 2041 | if (!has_active) { |
f40c67f0 KU |
2042 | /* |
2043 | * No active path in this pg, so this pg won't be used and | |
2044 | * the current_pg will be changed at next mapping time. | |
2045 | * We need to try mapping to determine it. | |
2046 | */ | |
be7d31cc | 2047 | busy = false; |
2da1610a | 2048 | } |
f40c67f0 KU |
2049 | |
2050 | return busy; | |
2051 | } | |
2052 | ||
1da177e4 LT |
2053 | /*----------------------------------------------------------------- |
2054 | * Module setup | |
2055 | *---------------------------------------------------------------*/ | |
2056 | static struct target_type multipath_target = { | |
2057 | .name = "multipath", | |
636be424 | 2058 | .version = {1, 14, 0}, |
8c5c1473 SM |
2059 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE | |
2060 | DM_TARGET_PASSES_INTEGRITY, | |
1da177e4 LT |
2061 | .module = THIS_MODULE, |
2062 | .ctr = multipath_ctr, | |
2063 | .dtr = multipath_dtr, | |
e5863d9a MS |
2064 | .clone_and_map_rq = multipath_clone_and_map, |
2065 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 2066 | .rq_end_io = multipath_end_io, |
76e33fe4 MS |
2067 | .map = multipath_map_bio, |
2068 | .end_io = multipath_end_io_bio, | |
2069 | .presuspend = multipath_presuspend, | |
2070 | .postsuspend = multipath_postsuspend, | |
2071 | .resume = multipath_resume, | |
2072 | .status = multipath_status, | |
2073 | .message = multipath_message, | |
2074 | .prepare_ioctl = multipath_prepare_ioctl, | |
2075 | .iterate_devices = multipath_iterate_devices, | |
2076 | .busy = multipath_busy, | |
2077 | }; | |
2078 | ||
1da177e4 LT |
2079 | static int __init dm_multipath_init(void) |
2080 | { | |
2081 | int r; | |
2082 | ||
4d4d66ab | 2083 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 2084 | if (!kmultipathd) { |
0cd33124 | 2085 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
2086 | r = -ENOMEM; |
2087 | goto bad_alloc_kmultipathd; | |
c557308e AK |
2088 | } |
2089 | ||
bab7cfc7 CS |
2090 | /* |
2091 | * A separate workqueue is used to handle the device handlers | |
2092 | * to avoid overloading existing workqueue. Overloading the | |
2093 | * old workqueue would also create a bottleneck in the | |
2094 | * path of the storage hardware device activation. | |
2095 | */ | |
4d4d66ab TH |
2096 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
2097 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
2098 | if (!kmpath_handlerd) { |
2099 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
2100 | r = -ENOMEM; |
2101 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
2102 | } |
2103 | ||
7e6358d2 | 2104 | r = dm_register_target(&multipath_target); |
2105 | if (r < 0) { | |
2106 | DMERR("request-based register failed %d", r); | |
2107 | r = -EINVAL; | |
2108 | goto bad_register_target; | |
2109 | } | |
2110 | ||
ff658e9c JT |
2111 | return 0; |
2112 | ||
7e6358d2 | 2113 | bad_register_target: |
2114 | destroy_workqueue(kmpath_handlerd); | |
ff658e9c JT |
2115 | bad_alloc_kmpath_handlerd: |
2116 | destroy_workqueue(kmultipathd); | |
2117 | bad_alloc_kmultipathd: | |
1da177e4 LT |
2118 | return r; |
2119 | } | |
2120 | ||
2121 | static void __exit dm_multipath_exit(void) | |
2122 | { | |
bab7cfc7 | 2123 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
2124 | destroy_workqueue(kmultipathd); |
2125 | ||
10d3bd09 | 2126 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
2127 | } |
2128 | ||
1da177e4 LT |
2129 | module_init(dm_multipath_init); |
2130 | module_exit(dm_multipath_exit); | |
2131 | ||
be240ff5 AP |
2132 | module_param_named(queue_if_no_path_timeout_secs, |
2133 | queue_if_no_path_timeout_secs, ulong, S_IRUGO | S_IWUSR); | |
2134 | MODULE_PARM_DESC(queue_if_no_path_timeout_secs, "No available paths queue IO timeout in seconds"); | |
2135 | ||
1da177e4 LT |
2136 | MODULE_DESCRIPTION(DM_NAME " multipath target"); |
2137 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
2138 | MODULE_LICENSE("GPL"); |