]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software Limited. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
586e80e6 MP |
8 | #include <linux/device-mapper.h> |
9 | ||
4cc96131 | 10 | #include "dm-rq.h" |
76e33fe4 | 11 | #include "dm-bio-record.h" |
1da177e4 | 12 | #include "dm-path-selector.h" |
b15546f9 | 13 | #include "dm-uevent.h" |
1da177e4 | 14 | |
e5863d9a | 15 | #include <linux/blkdev.h> |
1da177e4 LT |
16 | #include <linux/ctype.h> |
17 | #include <linux/init.h> | |
18 | #include <linux/mempool.h> | |
19 | #include <linux/module.h> | |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/time.h> | |
23 | #include <linux/workqueue.h> | |
35991652 | 24 | #include <linux/delay.h> |
cfae5c9b | 25 | #include <scsi/scsi_dh.h> |
60063497 | 26 | #include <linux/atomic.h> |
78ce23b5 | 27 | #include <linux/blk-mq.h> |
1da177e4 | 28 | |
72d94861 | 29 | #define DM_MSG_PREFIX "multipath" |
4e2d19e4 CS |
30 | #define DM_PG_INIT_DELAY_MSECS 2000 |
31 | #define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1) | |
1da177e4 LT |
32 | |
33 | /* Path properties */ | |
34 | struct pgpath { | |
35 | struct list_head list; | |
36 | ||
37 | struct priority_group *pg; /* Owning PG */ | |
38 | unsigned fail_count; /* Cumulative failure count */ | |
39 | ||
c922d5f7 | 40 | struct dm_path path; |
4e2d19e4 | 41 | struct delayed_work activate_path; |
be7d31cc MS |
42 | |
43 | bool is_active:1; /* Path status */ | |
1da177e4 LT |
44 | }; |
45 | ||
46 | #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) | |
47 | ||
48 | /* | |
49 | * Paths are grouped into Priority Groups and numbered from 1 upwards. | |
50 | * Each has a path selector which controls which path gets used. | |
51 | */ | |
52 | struct priority_group { | |
53 | struct list_head list; | |
54 | ||
55 | struct multipath *m; /* Owning multipath instance */ | |
56 | struct path_selector ps; | |
57 | ||
58 | unsigned pg_num; /* Reference number */ | |
1da177e4 LT |
59 | unsigned nr_pgpaths; /* Number of paths in PG */ |
60 | struct list_head pgpaths; | |
be7d31cc MS |
61 | |
62 | bool bypassed:1; /* Temporarily bypass this PG? */ | |
1da177e4 LT |
63 | }; |
64 | ||
65 | /* Multipath context */ | |
66 | struct multipath { | |
848b8aef | 67 | unsigned long flags; /* Multipath state flags */ |
4e2d19e4 | 68 | |
1fbdd2b3 | 69 | spinlock_t lock; |
848b8aef | 70 | enum dm_queue_mode queue_mode; |
4e2d19e4 | 71 | |
1da177e4 LT |
72 | struct pgpath *current_pgpath; |
73 | struct priority_group *current_pg; | |
74 | struct priority_group *next_pg; /* Switch to this PG if set */ | |
1da177e4 | 75 | |
848b8aef MS |
76 | atomic_t nr_valid_paths; /* Total number of usable paths */ |
77 | unsigned nr_priority_groups; | |
78 | struct list_head priority_groups; | |
1fbdd2b3 | 79 | |
848b8aef MS |
80 | const char *hw_handler_name; |
81 | char *hw_handler_params; | |
82 | wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */ | |
c9e45581 | 83 | unsigned pg_init_retries; /* Number of times to retry pg_init */ |
4e2d19e4 | 84 | unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */ |
91e968aa MS |
85 | atomic_t pg_init_in_progress; /* Only one pg_init allowed at once */ |
86 | atomic_t pg_init_count; /* Number of times pg_init called */ | |
87 | ||
6380f26f | 88 | struct mutex work_mutex; |
20800cb3 | 89 | struct work_struct trigger_event; |
848b8aef | 90 | struct dm_target *ti; |
76e33fe4 MS |
91 | |
92 | struct work_struct process_queued_bios; | |
93 | struct bio_list queued_bios; | |
1da177e4 LT |
94 | }; |
95 | ||
96 | /* | |
76e33fe4 | 97 | * Context information attached to each io we process. |
1da177e4 | 98 | */ |
028867ac | 99 | struct dm_mpath_io { |
1da177e4 | 100 | struct pgpath *pgpath; |
02ab823f | 101 | size_t nr_bytes; |
1da177e4 LT |
102 | }; |
103 | ||
104 | typedef int (*action_fn) (struct pgpath *pgpath); | |
105 | ||
bab7cfc7 | 106 | static struct workqueue_struct *kmultipathd, *kmpath_handlerd; |
c4028958 | 107 | static void trigger_event(struct work_struct *work); |
89bfce76 BVA |
108 | static void activate_or_offline_path(struct pgpath *pgpath); |
109 | static void activate_path_work(struct work_struct *work); | |
76e33fe4 | 110 | static void process_queued_bios(struct work_struct *work); |
1da177e4 | 111 | |
518257b1 MS |
112 | /*----------------------------------------------- |
113 | * Multipath state flags. | |
114 | *-----------------------------------------------*/ | |
115 | ||
116 | #define MPATHF_QUEUE_IO 0 /* Must we queue all I/O? */ | |
117 | #define MPATHF_QUEUE_IF_NO_PATH 1 /* Queue I/O if last path fails? */ | |
118 | #define MPATHF_SAVED_QUEUE_IF_NO_PATH 2 /* Saved state during suspension */ | |
119 | #define MPATHF_RETAIN_ATTACHED_HW_HANDLER 3 /* If there's already a hw_handler present, don't change it. */ | |
120 | #define MPATHF_PG_INIT_DISABLED 4 /* pg_init is not currently allowed */ | |
121 | #define MPATHF_PG_INIT_REQUIRED 5 /* pg_init needs calling? */ | |
122 | #define MPATHF_PG_INIT_DELAY_RETRY 6 /* Delay pg_init retry? */ | |
1da177e4 LT |
123 | |
124 | /*----------------------------------------------- | |
125 | * Allocation routines | |
126 | *-----------------------------------------------*/ | |
127 | ||
128 | static struct pgpath *alloc_pgpath(void) | |
129 | { | |
e69fae56 | 130 | struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL); |
1da177e4 | 131 | |
848b8aef MS |
132 | if (!pgpath) |
133 | return NULL; | |
134 | ||
135 | pgpath->is_active = true; | |
1da177e4 LT |
136 | |
137 | return pgpath; | |
138 | } | |
139 | ||
028867ac | 140 | static void free_pgpath(struct pgpath *pgpath) |
1da177e4 LT |
141 | { |
142 | kfree(pgpath); | |
143 | } | |
144 | ||
145 | static struct priority_group *alloc_priority_group(void) | |
146 | { | |
147 | struct priority_group *pg; | |
148 | ||
e69fae56 | 149 | pg = kzalloc(sizeof(*pg), GFP_KERNEL); |
1da177e4 | 150 | |
e69fae56 MM |
151 | if (pg) |
152 | INIT_LIST_HEAD(&pg->pgpaths); | |
1da177e4 LT |
153 | |
154 | return pg; | |
155 | } | |
156 | ||
157 | static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) | |
158 | { | |
159 | struct pgpath *pgpath, *tmp; | |
160 | ||
161 | list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { | |
162 | list_del(&pgpath->list); | |
163 | dm_put_device(ti, pgpath->path.dev); | |
164 | free_pgpath(pgpath); | |
165 | } | |
166 | } | |
167 | ||
168 | static void free_priority_group(struct priority_group *pg, | |
169 | struct dm_target *ti) | |
170 | { | |
171 | struct path_selector *ps = &pg->ps; | |
172 | ||
173 | if (ps->type) { | |
174 | ps->type->destroy(ps); | |
175 | dm_put_path_selector(ps->type); | |
176 | } | |
177 | ||
178 | free_pgpaths(&pg->pgpaths, ti); | |
179 | kfree(pg); | |
180 | } | |
181 | ||
e83068a5 | 182 | static struct multipath *alloc_multipath(struct dm_target *ti) |
1da177e4 LT |
183 | { |
184 | struct multipath *m; | |
185 | ||
e69fae56 | 186 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
1da177e4 | 187 | if (m) { |
1da177e4 LT |
188 | INIT_LIST_HEAD(&m->priority_groups); |
189 | spin_lock_init(&m->lock); | |
91e968aa | 190 | atomic_set(&m->nr_valid_paths, 0); |
c4028958 | 191 | INIT_WORK(&m->trigger_event, trigger_event); |
6380f26f | 192 | mutex_init(&m->work_mutex); |
8637a6bf | 193 | |
e83068a5 | 194 | m->queue_mode = DM_TYPE_NONE; |
76e33fe4 | 195 | |
28f16c20 MM |
196 | m->ti = ti; |
197 | ti->private = m; | |
1da177e4 LT |
198 | } |
199 | ||
200 | return m; | |
201 | } | |
202 | ||
e83068a5 MS |
203 | static int alloc_multipath_stage2(struct dm_target *ti, struct multipath *m) |
204 | { | |
205 | if (m->queue_mode == DM_TYPE_NONE) { | |
206 | /* | |
207 | * Default to request-based. | |
208 | */ | |
209 | if (dm_use_blk_mq(dm_table_get_md(ti->table))) | |
210 | m->queue_mode = DM_TYPE_MQ_REQUEST_BASED; | |
211 | else | |
212 | m->queue_mode = DM_TYPE_REQUEST_BASED; | |
cd025384 MS |
213 | |
214 | } else if (m->queue_mode == DM_TYPE_BIO_BASED || | |
215 | m->queue_mode == DM_TYPE_NVME_BIO_BASED) { | |
e83068a5 | 216 | INIT_WORK(&m->process_queued_bios, process_queued_bios); |
cd025384 MS |
217 | |
218 | if (m->queue_mode == DM_TYPE_BIO_BASED) { | |
219 | /* | |
220 | * bio-based doesn't support any direct scsi_dh management; | |
221 | * it just discovers if a scsi_dh is attached. | |
222 | */ | |
223 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); | |
224 | } | |
e83068a5 MS |
225 | } |
226 | ||
848b8aef MS |
227 | if (m->queue_mode != DM_TYPE_NVME_BIO_BASED) { |
228 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
229 | atomic_set(&m->pg_init_in_progress, 0); | |
230 | atomic_set(&m->pg_init_count, 0); | |
231 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; | |
232 | init_waitqueue_head(&m->pg_init_wait); | |
233 | } | |
234 | ||
e83068a5 MS |
235 | dm_table_set_type(ti->table, m->queue_mode); |
236 | ||
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 | 339 | |
848b8aef MS |
340 | if (m->queue_mode == DM_TYPE_NVME_BIO_BASED) |
341 | return; | |
342 | ||
1da177e4 | 343 | /* Must we initialise the PG first, and queue I/O till it's ready? */ |
cfae5c9b | 344 | if (m->hw_handler_name) { |
518257b1 MS |
345 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
346 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 347 | } else { |
518257b1 MS |
348 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
349 | clear_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 350 | } |
c9e45581 | 351 | |
91e968aa | 352 | atomic_set(&m->pg_init_count, 0); |
1da177e4 LT |
353 | } |
354 | ||
2da1610a MS |
355 | static struct pgpath *choose_path_in_pg(struct multipath *m, |
356 | struct priority_group *pg, | |
357 | size_t nr_bytes) | |
1da177e4 | 358 | { |
2da1610a | 359 | unsigned long flags; |
c922d5f7 | 360 | struct dm_path *path; |
2da1610a | 361 | struct pgpath *pgpath; |
1da177e4 | 362 | |
90a4323c | 363 | path = pg->ps.type->select_path(&pg->ps, nr_bytes); |
1da177e4 | 364 | if (!path) |
2da1610a | 365 | return ERR_PTR(-ENXIO); |
1da177e4 | 366 | |
2da1610a | 367 | pgpath = path_to_pgpath(path); |
1da177e4 | 368 | |
506458ef | 369 | if (unlikely(READ_ONCE(m->current_pg) != pg)) { |
2da1610a MS |
370 | /* Only update current_pgpath if pg changed */ |
371 | spin_lock_irqsave(&m->lock, flags); | |
372 | m->current_pgpath = pgpath; | |
373 | __switch_pg(m, pg); | |
374 | spin_unlock_irqrestore(&m->lock, flags); | |
375 | } | |
1da177e4 | 376 | |
2da1610a | 377 | return pgpath; |
1da177e4 LT |
378 | } |
379 | ||
2da1610a | 380 | static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 | 381 | { |
2da1610a | 382 | unsigned long flags; |
1da177e4 | 383 | struct priority_group *pg; |
2da1610a | 384 | struct pgpath *pgpath; |
d19a55cc | 385 | unsigned bypassed = 1; |
1da177e4 | 386 | |
91e968aa | 387 | if (!atomic_read(&m->nr_valid_paths)) { |
848b8aef MS |
388 | if (m->queue_mode != DM_TYPE_NVME_BIO_BASED) |
389 | clear_bit(MPATHF_QUEUE_IO, &m->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)) { | |
f220fd4e | 429 | if (!bypassed) |
518257b1 | 430 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); |
2da1610a | 431 | return pgpath; |
f220fd4e | 432 | } |
1da177e4 LT |
433 | } |
434 | } while (bypassed--); | |
435 | ||
436 | failed: | |
2da1610a | 437 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
438 | m->current_pgpath = NULL; |
439 | m->current_pg = NULL; | |
2da1610a MS |
440 | spin_unlock_irqrestore(&m->lock, flags); |
441 | ||
442 | return NULL; | |
1da177e4 LT |
443 | } |
444 | ||
45e15720 | 445 | /* |
86331f39 BVA |
446 | * dm_report_EIO() is a macro instead of a function to make pr_debug() |
447 | * report the function name and line number of the function from which | |
448 | * it has been invoked. | |
45e15720 | 449 | */ |
86331f39 | 450 | #define dm_report_EIO(m) \ |
18a482f5 | 451 | do { \ |
86331f39 BVA |
452 | struct mapped_device *md = dm_table_get_md((m)->ti->table); \ |
453 | \ | |
454 | pr_debug("%s: returning EIO; QIFNP = %d; SQIFNP = %d; DNFS = %d\n", \ | |
455 | dm_device_name(md), \ | |
456 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
457 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
458 | dm_noflush_suspending((m)->ti)); \ | |
18a482f5 | 459 | } while (0) |
45e15720 | 460 | |
c1fd0abe MS |
461 | /* |
462 | * Check whether bios must be queued in the device-mapper core rather | |
463 | * than here in the target. | |
464 | * | |
465 | * If MPATHF_QUEUE_IF_NO_PATH and MPATHF_SAVED_QUEUE_IF_NO_PATH hold | |
466 | * the same value then we are not between multipath_presuspend() | |
467 | * and multipath_resume() calls and we have no need to check | |
468 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
469 | */ | |
470 | static bool __must_push_back(struct multipath *m, unsigned long flags) | |
471 | { | |
472 | return ((test_bit(MPATHF_QUEUE_IF_NO_PATH, &flags) != | |
473 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &flags)) && | |
474 | dm_noflush_suspending(m->ti)); | |
475 | } | |
476 | ||
477 | /* | |
478 | * Following functions use READ_ONCE to get atomic access to | |
479 | * all m->flags to avoid taking spinlock | |
480 | */ | |
481 | static bool must_push_back_rq(struct multipath *m) | |
482 | { | |
483 | unsigned long flags = READ_ONCE(m->flags); | |
484 | return test_bit(MPATHF_QUEUE_IF_NO_PATH, &flags) || __must_push_back(m, flags); | |
485 | } | |
486 | ||
487 | static bool must_push_back_bio(struct multipath *m) | |
488 | { | |
489 | unsigned long flags = READ_ONCE(m->flags); | |
490 | return __must_push_back(m, flags); | |
491 | } | |
492 | ||
36fcffcc | 493 | /* |
76e33fe4 | 494 | * Map cloned requests (request-based multipath) |
36fcffcc | 495 | */ |
eb8db831 CH |
496 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, |
497 | union map_info *map_context, | |
498 | struct request **__clone) | |
1da177e4 | 499 | { |
7943bd6d | 500 | struct multipath *m = ti->private; |
eb8db831 | 501 | size_t nr_bytes = blk_rq_bytes(rq); |
1da177e4 | 502 | struct pgpath *pgpath; |
f40c67f0 | 503 | struct block_device *bdev; |
eb8db831 | 504 | struct dm_mpath_io *mpio = get_mpio(map_context); |
7083abbb | 505 | struct request_queue *q; |
eb8db831 | 506 | struct request *clone; |
1da177e4 | 507 | |
1da177e4 | 508 | /* Do we need to select a new pgpath? */ |
506458ef | 509 | pgpath = READ_ONCE(m->current_pgpath); |
2da1610a MS |
510 | if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) |
511 | pgpath = choose_pgpath(m, nr_bytes); | |
1da177e4 | 512 | |
9bf59a61 | 513 | if (!pgpath) { |
c1fd0abe | 514 | if (must_push_back_rq(m)) |
b88efd43 | 515 | return DM_MAPIO_DELAY_REQUEUE; |
18a482f5 | 516 | dm_report_EIO(m); /* Failed */ |
f98e0eb6 | 517 | return DM_MAPIO_KILL; |
518257b1 MS |
518 | } else if (test_bit(MPATHF_QUEUE_IO, &m->flags) || |
519 | test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { | |
459b5401 ML |
520 | pg_init_all_paths(m); |
521 | return DM_MAPIO_DELAY_REQUEUE; | |
9bf59a61 | 522 | } |
6afbc01d | 523 | |
2eb6e1e3 KB |
524 | mpio->pgpath = pgpath; |
525 | mpio->nr_bytes = nr_bytes; | |
526 | ||
9bf59a61 | 527 | bdev = pgpath->path.dev->bdev; |
7083abbb BVA |
528 | q = bdev_get_queue(bdev); |
529 | clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, GFP_ATOMIC); | |
eb8db831 CH |
530 | if (IS_ERR(clone)) { |
531 | /* EBUSY, ENODEV or EWOULDBLOCK: requeue */ | |
848b8aef | 532 | if (blk_queue_dying(q)) { |
7083abbb BVA |
533 | atomic_inc(&m->pg_init_in_progress); |
534 | activate_or_offline_path(pgpath); | |
050af08f | 535 | return DM_MAPIO_DELAY_REQUEUE; |
7083abbb | 536 | } |
050af08f ML |
537 | |
538 | /* | |
539 | * blk-mq's SCHED_RESTART can cover this requeue, so we | |
540 | * needn't deal with it by DELAY_REQUEUE. More importantly, | |
541 | * we have to return DM_MAPIO_REQUEUE so that blk-mq can | |
542 | * get the queue busy feedback (via BLK_STS_RESOURCE), | |
543 | * otherwise I/O merging can suffer. | |
544 | */ | |
545 | if (q->mq_ops) | |
546 | return DM_MAPIO_REQUEUE; | |
547 | else | |
548 | return DM_MAPIO_DELAY_REQUEUE; | |
e5863d9a | 549 | } |
eb8db831 CH |
550 | clone->bio = clone->biotail = NULL; |
551 | clone->rq_disk = bdev->bd_disk; | |
552 | clone->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
553 | *__clone = clone; | |
e5863d9a | 554 | |
9bf59a61 MS |
555 | if (pgpath->pg->ps.type->start_io) |
556 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
557 | &pgpath->path, | |
558 | nr_bytes); | |
2eb6e1e3 | 559 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
560 | } |
561 | ||
e5863d9a MS |
562 | static void multipath_release_clone(struct request *clone) |
563 | { | |
eb8db831 | 564 | blk_put_request(clone); |
e5863d9a MS |
565 | } |
566 | ||
76e33fe4 MS |
567 | /* |
568 | * Map cloned bios (bio-based multipath) | |
569 | */ | |
0001ec56 MS |
570 | |
571 | static struct pgpath *__map_bio(struct multipath *m, struct bio *bio) | |
76e33fe4 | 572 | { |
76e33fe4 MS |
573 | struct pgpath *pgpath; |
574 | unsigned long flags; | |
575 | bool queue_io; | |
576 | ||
577 | /* Do we need to select a new pgpath? */ | |
506458ef | 578 | pgpath = READ_ONCE(m->current_pgpath); |
76e33fe4 MS |
579 | queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags); |
580 | if (!pgpath || !queue_io) | |
0001ec56 | 581 | pgpath = choose_pgpath(m, bio->bi_iter.bi_size); |
76e33fe4 | 582 | |
0001ec56 MS |
583 | if ((pgpath && queue_io) || |
584 | (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) { | |
76e33fe4 MS |
585 | /* Queue for the daemon to resubmit */ |
586 | spin_lock_irqsave(&m->lock, flags); | |
587 | bio_list_add(&m->queued_bios, bio); | |
588 | spin_unlock_irqrestore(&m->lock, flags); | |
848b8aef | 589 | |
0001ec56 MS |
590 | /* PG_INIT_REQUIRED cannot be set without QUEUE_IO */ |
591 | if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) | |
592 | pg_init_all_paths(m); | |
593 | else if (!queue_io) | |
594 | queue_work(kmultipathd, &m->process_queued_bios); | |
595 | ||
596 | return ERR_PTR(-EAGAIN); | |
597 | } | |
598 | ||
599 | return pgpath; | |
600 | } | |
601 | ||
602 | static struct pgpath *__map_bio_nvme(struct multipath *m, struct bio *bio) | |
603 | { | |
604 | struct pgpath *pgpath; | |
605 | unsigned long flags; | |
606 | ||
607 | /* Do we need to select a new pgpath? */ | |
608 | /* | |
609 | * FIXME: currently only switching path if no path (due to failure, etc) | |
610 | * - which negates the point of using a path selector | |
611 | */ | |
612 | pgpath = READ_ONCE(m->current_pgpath); | |
613 | if (!pgpath) | |
614 | pgpath = choose_pgpath(m, bio->bi_iter.bi_size); | |
615 | ||
616 | if (!pgpath) { | |
617 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
618 | /* Queue for the daemon to resubmit */ | |
619 | spin_lock_irqsave(&m->lock, flags); | |
620 | bio_list_add(&m->queued_bios, bio); | |
621 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 622 | queue_work(kmultipathd, &m->process_queued_bios); |
0001ec56 MS |
623 | |
624 | return ERR_PTR(-EAGAIN); | |
848b8aef | 625 | } |
0001ec56 MS |
626 | return NULL; |
627 | } | |
628 | ||
629 | return pgpath; | |
630 | } | |
631 | ||
632 | static int __multipath_map_bio(struct multipath *m, struct bio *bio, | |
633 | struct dm_mpath_io *mpio) | |
634 | { | |
635 | struct pgpath *pgpath; | |
636 | ||
637 | if (m->queue_mode == DM_TYPE_NVME_BIO_BASED) | |
638 | pgpath = __map_bio_nvme(m, bio); | |
639 | else | |
640 | pgpath = __map_bio(m, bio); | |
848b8aef | 641 | |
0001ec56 | 642 | if (IS_ERR(pgpath)) |
76e33fe4 | 643 | return DM_MAPIO_SUBMITTED; |
76e33fe4 MS |
644 | |
645 | if (!pgpath) { | |
c1fd0abe | 646 | if (must_push_back_bio(m)) |
ca5beb76 | 647 | return DM_MAPIO_REQUEUE; |
18a482f5 | 648 | dm_report_EIO(m); |
846785e6 | 649 | return DM_MAPIO_KILL; |
76e33fe4 MS |
650 | } |
651 | ||
652 | mpio->pgpath = pgpath; | |
76e33fe4 | 653 | |
4e4cbee9 | 654 | bio->bi_status = 0; |
74d46992 | 655 | bio_set_dev(bio, pgpath->path.dev->bdev); |
1eff9d32 | 656 | bio->bi_opf |= REQ_FAILFAST_TRANSPORT; |
76e33fe4 MS |
657 | |
658 | if (pgpath->pg->ps.type->start_io) | |
659 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
660 | &pgpath->path, | |
d0442f80 | 661 | mpio->nr_bytes); |
76e33fe4 MS |
662 | return DM_MAPIO_REMAPPED; |
663 | } | |
664 | ||
665 | static int multipath_map_bio(struct dm_target *ti, struct bio *bio) | |
666 | { | |
667 | struct multipath *m = ti->private; | |
bf661be1 MS |
668 | struct dm_mpath_io *mpio = NULL; |
669 | ||
63f6e6fd | 670 | multipath_init_per_bio_data(bio, &mpio); |
76e33fe4 MS |
671 | return __multipath_map_bio(m, bio, mpio); |
672 | } | |
673 | ||
7e48c768 | 674 | static void process_queued_io_list(struct multipath *m) |
76e33fe4 | 675 | { |
7e48c768 MS |
676 | if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED) |
677 | dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table)); | |
cd025384 MS |
678 | else if (m->queue_mode == DM_TYPE_BIO_BASED || |
679 | m->queue_mode == DM_TYPE_NVME_BIO_BASED) | |
76e33fe4 MS |
680 | queue_work(kmultipathd, &m->process_queued_bios); |
681 | } | |
682 | ||
683 | static void process_queued_bios(struct work_struct *work) | |
684 | { | |
685 | int r; | |
686 | unsigned long flags; | |
687 | struct bio *bio; | |
688 | struct bio_list bios; | |
689 | struct blk_plug plug; | |
690 | struct multipath *m = | |
691 | container_of(work, struct multipath, process_queued_bios); | |
692 | ||
693 | bio_list_init(&bios); | |
694 | ||
695 | spin_lock_irqsave(&m->lock, flags); | |
696 | ||
697 | if (bio_list_empty(&m->queued_bios)) { | |
698 | spin_unlock_irqrestore(&m->lock, flags); | |
699 | return; | |
700 | } | |
701 | ||
702 | bio_list_merge(&bios, &m->queued_bios); | |
703 | bio_list_init(&m->queued_bios); | |
704 | ||
705 | spin_unlock_irqrestore(&m->lock, flags); | |
706 | ||
707 | blk_start_plug(&plug); | |
708 | while ((bio = bio_list_pop(&bios))) { | |
1836df08 MS |
709 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); |
710 | dm_bio_restore(get_bio_details_from_mpio(mpio), bio); | |
711 | r = __multipath_map_bio(m, bio, mpio); | |
846785e6 CH |
712 | switch (r) { |
713 | case DM_MAPIO_KILL: | |
4e4cbee9 CH |
714 | bio->bi_status = BLK_STS_IOERR; |
715 | bio_endio(bio); | |
047385b3 | 716 | break; |
846785e6 | 717 | case DM_MAPIO_REQUEUE: |
4e4cbee9 | 718 | bio->bi_status = BLK_STS_DM_REQUEUE; |
76e33fe4 | 719 | bio_endio(bio); |
846785e6 CH |
720 | break; |
721 | case DM_MAPIO_REMAPPED: | |
76e33fe4 | 722 | generic_make_request(bio); |
846785e6 | 723 | break; |
9157c8d3 BVA |
724 | case 0: |
725 | break; | |
726 | default: | |
727 | WARN_ONCE(true, "__multipath_map_bio() returned %d\n", r); | |
846785e6 | 728 | } |
76e33fe4 MS |
729 | } |
730 | blk_finish_plug(&plug); | |
731 | } | |
732 | ||
1da177e4 LT |
733 | /* |
734 | * If we run out of usable paths, should we queue I/O or error it? | |
735 | */ | |
be7d31cc MS |
736 | static int queue_if_no_path(struct multipath *m, bool queue_if_no_path, |
737 | bool save_old_value) | |
1da177e4 LT |
738 | { |
739 | unsigned long flags; | |
740 | ||
741 | spin_lock_irqsave(&m->lock, flags); | |
5307e2ad LW |
742 | assign_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags, |
743 | (save_old_value && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) || | |
744 | (!save_old_value && queue_if_no_path)); | |
c1fd0abe | 745 | assign_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags, queue_if_no_path); |
1da177e4 LT |
746 | spin_unlock_irqrestore(&m->lock, flags); |
747 | ||
76e33fe4 | 748 | if (!queue_if_no_path) { |
63d832c3 | 749 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 750 | process_queued_io_list(m); |
76e33fe4 | 751 | } |
63d832c3 | 752 | |
1da177e4 LT |
753 | return 0; |
754 | } | |
755 | ||
1da177e4 LT |
756 | /* |
757 | * An event is triggered whenever a path is taken out of use. | |
758 | * Includes path failure and PG bypass. | |
759 | */ | |
c4028958 | 760 | static void trigger_event(struct work_struct *work) |
1da177e4 | 761 | { |
c4028958 DH |
762 | struct multipath *m = |
763 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
764 | |
765 | dm_table_event(m->ti->table); | |
766 | } | |
767 | ||
768 | /*----------------------------------------------------------------- | |
769 | * Constructor/argument parsing: | |
770 | * <#multipath feature args> [<arg>]* | |
771 | * <#hw_handler args> [hw_handler [<arg>]*] | |
772 | * <#priority groups> | |
773 | * <initial priority group> | |
774 | * [<selector> <#selector args> [<arg>]* | |
775 | * <#paths> <#per-path selector args> | |
776 | * [<path> [<arg>]* ]+ ]+ | |
777 | *---------------------------------------------------------------*/ | |
498f0103 | 778 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
779 | struct dm_target *ti) |
780 | { | |
781 | int r; | |
782 | struct path_selector_type *pst; | |
783 | unsigned ps_argc; | |
784 | ||
5916a22b | 785 | static const struct dm_arg _args[] = { |
72d94861 | 786 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
787 | }; |
788 | ||
498f0103 | 789 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 790 | if (!pst) { |
72d94861 | 791 | ti->error = "unknown path selector type"; |
1da177e4 LT |
792 | return -EINVAL; |
793 | } | |
794 | ||
498f0103 | 795 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
796 | if (r) { |
797 | dm_put_path_selector(pst); | |
1da177e4 | 798 | return -EINVAL; |
371b2e34 | 799 | } |
1da177e4 LT |
800 | |
801 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
802 | if (r) { | |
803 | dm_put_path_selector(pst); | |
72d94861 | 804 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
805 | return r; |
806 | } | |
807 | ||
808 | pg->ps.type = pst; | |
498f0103 | 809 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
810 | |
811 | return 0; | |
812 | } | |
813 | ||
848b8aef | 814 | static int setup_scsi_dh(struct block_device *bdev, struct multipath *m, char **error) |
1da177e4 | 815 | { |
848b8aef | 816 | struct request_queue *q = bdev_get_queue(bdev); |
a58a935d | 817 | const char *attached_handler_name; |
848b8aef | 818 | int r; |
a58a935d | 819 | |
518257b1 | 820 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) { |
1bab0de0 | 821 | retain: |
a58a935d MS |
822 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); |
823 | if (attached_handler_name) { | |
54cd640d | 824 | /* |
825 | * Clear any hw_handler_params associated with a | |
826 | * handler that isn't already attached. | |
827 | */ | |
828 | if (m->hw_handler_name && strcmp(attached_handler_name, m->hw_handler_name)) { | |
829 | kfree(m->hw_handler_params); | |
830 | m->hw_handler_params = NULL; | |
831 | } | |
832 | ||
a58a935d MS |
833 | /* |
834 | * Reset hw_handler_name to match the attached handler | |
a58a935d MS |
835 | * |
836 | * NB. This modifies the table line to show the actual | |
837 | * handler instead of the original table passed in. | |
838 | */ | |
839 | kfree(m->hw_handler_name); | |
840 | m->hw_handler_name = attached_handler_name; | |
a58a935d MS |
841 | } |
842 | } | |
a0cf7ea9 | 843 | |
a58a935d | 844 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
845 | r = scsi_dh_attach(q, m->hw_handler_name); |
846 | if (r == -EBUSY) { | |
1bab0de0 | 847 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 848 | |
1bab0de0 | 849 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
848b8aef | 850 | bdevname(bdev, b)); |
1bab0de0 CH |
851 | goto retain; |
852 | } | |
ae11b1b3 | 853 | if (r < 0) { |
848b8aef MS |
854 | *error = "error attaching hardware handler"; |
855 | return r; | |
ae11b1b3 | 856 | } |
2bfd2e13 CS |
857 | |
858 | if (m->hw_handler_params) { | |
859 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
860 | if (r < 0) { | |
848b8aef MS |
861 | *error = "unable to set hardware handler parameters"; |
862 | return r; | |
2bfd2e13 CS |
863 | } |
864 | } | |
ae11b1b3 HR |
865 | } |
866 | ||
848b8aef MS |
867 | return 0; |
868 | } | |
869 | ||
870 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, | |
871 | struct dm_target *ti) | |
872 | { | |
873 | int r; | |
874 | struct pgpath *p; | |
875 | struct multipath *m = ti->private; | |
876 | ||
877 | /* we need at least a path arg */ | |
878 | if (as->argc < 1) { | |
879 | ti->error = "no device given"; | |
880 | return ERR_PTR(-EINVAL); | |
881 | } | |
882 | ||
883 | p = alloc_pgpath(); | |
884 | if (!p) | |
885 | return ERR_PTR(-ENOMEM); | |
886 | ||
887 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), | |
888 | &p->path.dev); | |
889 | if (r) { | |
890 | ti->error = "error getting device"; | |
891 | goto bad; | |
892 | } | |
893 | ||
894 | if (m->queue_mode != DM_TYPE_NVME_BIO_BASED) { | |
895 | INIT_DELAYED_WORK(&p->activate_path, activate_path_work); | |
896 | r = setup_scsi_dh(p->path.dev->bdev, m, &ti->error); | |
897 | if (r) { | |
898 | dm_put_device(ti, p->path.dev); | |
899 | goto bad; | |
900 | } | |
901 | } | |
902 | ||
1da177e4 LT |
903 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
904 | if (r) { | |
905 | dm_put_device(ti, p->path.dev); | |
906 | goto bad; | |
907 | } | |
908 | ||
909 | return p; | |
1da177e4 LT |
910 | bad: |
911 | free_pgpath(p); | |
01460f35 | 912 | return ERR_PTR(r); |
1da177e4 LT |
913 | } |
914 | ||
498f0103 | 915 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 916 | struct multipath *m) |
1da177e4 | 917 | { |
5916a22b | 918 | static const struct dm_arg _args[] = { |
72d94861 AK |
919 | {1, 1024, "invalid number of paths"}, |
920 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
921 | }; |
922 | ||
923 | int r; | |
498f0103 | 924 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 925 | struct priority_group *pg; |
28f16c20 | 926 | struct dm_target *ti = m->ti; |
1da177e4 LT |
927 | |
928 | if (as->argc < 2) { | |
929 | as->argc = 0; | |
01460f35 BM |
930 | ti->error = "not enough priority group arguments"; |
931 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
932 | } |
933 | ||
934 | pg = alloc_priority_group(); | |
935 | if (!pg) { | |
72d94861 | 936 | ti->error = "couldn't allocate priority group"; |
01460f35 | 937 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
938 | } |
939 | pg->m = m; | |
940 | ||
941 | r = parse_path_selector(as, pg, ti); | |
942 | if (r) | |
943 | goto bad; | |
944 | ||
945 | /* | |
946 | * read the paths | |
947 | */ | |
498f0103 | 948 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
949 | if (r) |
950 | goto bad; | |
951 | ||
498f0103 | 952 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
953 | if (r) |
954 | goto bad; | |
955 | ||
498f0103 | 956 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
957 | for (i = 0; i < pg->nr_pgpaths; i++) { |
958 | struct pgpath *pgpath; | |
498f0103 | 959 | struct dm_arg_set path_args; |
1da177e4 | 960 | |
498f0103 | 961 | if (as->argc < nr_args) { |
148acff6 | 962 | ti->error = "not enough path parameters"; |
6bbf79a1 | 963 | r = -EINVAL; |
1da177e4 | 964 | goto bad; |
148acff6 | 965 | } |
1da177e4 | 966 | |
498f0103 | 967 | path_args.argc = nr_args; |
1da177e4 LT |
968 | path_args.argv = as->argv; |
969 | ||
970 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
971 | if (IS_ERR(pgpath)) { |
972 | r = PTR_ERR(pgpath); | |
1da177e4 | 973 | goto bad; |
01460f35 | 974 | } |
1da177e4 LT |
975 | |
976 | pgpath->pg = pg; | |
977 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 978 | dm_consume_args(as, nr_args); |
1da177e4 LT |
979 | } |
980 | ||
981 | return pg; | |
982 | ||
983 | bad: | |
984 | free_priority_group(pg, ti); | |
01460f35 | 985 | return ERR_PTR(r); |
1da177e4 LT |
986 | } |
987 | ||
498f0103 | 988 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 989 | { |
1da177e4 | 990 | unsigned hw_argc; |
2bfd2e13 | 991 | int ret; |
28f16c20 | 992 | struct dm_target *ti = m->ti; |
1da177e4 | 993 | |
5916a22b | 994 | static const struct dm_arg _args[] = { |
72d94861 | 995 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
996 | }; |
997 | ||
498f0103 | 998 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
999 | return -EINVAL; |
1000 | ||
1001 | if (!hw_argc) | |
1002 | return 0; | |
1003 | ||
cd025384 MS |
1004 | if (m->queue_mode == DM_TYPE_BIO_BASED || |
1005 | m->queue_mode == DM_TYPE_NVME_BIO_BASED) { | |
76e33fe4 MS |
1006 | dm_consume_args(as, hw_argc); |
1007 | DMERR("bio-based multipath doesn't allow hardware handler args"); | |
1008 | return 0; | |
1009 | } | |
1010 | ||
498f0103 | 1011 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
f97dc421 | 1012 | if (!m->hw_handler_name) |
1013 | return -EINVAL; | |
14e98c5c | 1014 | |
2bfd2e13 CS |
1015 | if (hw_argc > 1) { |
1016 | char *p; | |
1017 | int i, j, len = 4; | |
1018 | ||
1019 | for (i = 0; i <= hw_argc - 2; i++) | |
1020 | len += strlen(as->argv[i]) + 1; | |
1021 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
1022 | if (!p) { | |
1023 | ti->error = "memory allocation failed"; | |
1024 | ret = -ENOMEM; | |
1025 | goto fail; | |
1026 | } | |
1027 | j = sprintf(p, "%d", hw_argc - 1); | |
1028 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
1029 | j = sprintf(p, "%s", as->argv[i]); | |
1030 | } | |
498f0103 | 1031 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
1032 | |
1033 | return 0; | |
2bfd2e13 CS |
1034 | fail: |
1035 | kfree(m->hw_handler_name); | |
1036 | m->hw_handler_name = NULL; | |
1037 | return ret; | |
1da177e4 LT |
1038 | } |
1039 | ||
498f0103 | 1040 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
1041 | { |
1042 | int r; | |
1043 | unsigned argc; | |
28f16c20 | 1044 | struct dm_target *ti = m->ti; |
498f0103 | 1045 | const char *arg_name; |
1da177e4 | 1046 | |
5916a22b | 1047 | static const struct dm_arg _args[] = { |
e83068a5 | 1048 | {0, 8, "invalid number of feature args"}, |
c9e45581 | 1049 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 1050 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
1051 | }; |
1052 | ||
498f0103 | 1053 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
1054 | if (r) |
1055 | return -EINVAL; | |
1056 | ||
1057 | if (!argc) | |
1058 | return 0; | |
1059 | ||
c9e45581 | 1060 | do { |
498f0103 | 1061 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
1062 | argc--; |
1063 | ||
498f0103 | 1064 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
be7d31cc | 1065 | r = queue_if_no_path(m, true, false); |
c9e45581 DW |
1066 | continue; |
1067 | } | |
1068 | ||
a58a935d | 1069 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
518257b1 | 1070 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); |
a58a935d MS |
1071 | continue; |
1072 | } | |
1073 | ||
498f0103 | 1074 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 1075 | (argc >= 1)) { |
498f0103 | 1076 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
1077 | argc--; |
1078 | continue; | |
1079 | } | |
1080 | ||
498f0103 | 1081 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 1082 | (argc >= 1)) { |
498f0103 | 1083 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
1084 | argc--; |
1085 | continue; | |
1086 | } | |
1087 | ||
e83068a5 MS |
1088 | if (!strcasecmp(arg_name, "queue_mode") && |
1089 | (argc >= 1)) { | |
1090 | const char *queue_mode_name = dm_shift_arg(as); | |
1091 | ||
1092 | if (!strcasecmp(queue_mode_name, "bio")) | |
1093 | m->queue_mode = DM_TYPE_BIO_BASED; | |
cd025384 MS |
1094 | else if (!strcasecmp(queue_mode_name, "nvme")) |
1095 | m->queue_mode = DM_TYPE_NVME_BIO_BASED; | |
e83068a5 MS |
1096 | else if (!strcasecmp(queue_mode_name, "rq")) |
1097 | m->queue_mode = DM_TYPE_REQUEST_BASED; | |
1098 | else if (!strcasecmp(queue_mode_name, "mq")) | |
1099 | m->queue_mode = DM_TYPE_MQ_REQUEST_BASED; | |
1100 | else { | |
1101 | ti->error = "Unknown 'queue_mode' requested"; | |
1102 | r = -EINVAL; | |
1103 | } | |
1104 | argc--; | |
1105 | continue; | |
1106 | } | |
1107 | ||
1da177e4 | 1108 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
1109 | r = -EINVAL; |
1110 | } while (argc && !r); | |
1111 | ||
1112 | return r; | |
1da177e4 LT |
1113 | } |
1114 | ||
e83068a5 | 1115 | static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv) |
1da177e4 | 1116 | { |
498f0103 | 1117 | /* target arguments */ |
5916a22b | 1118 | static const struct dm_arg _args[] = { |
a490a07a MS |
1119 | {0, 1024, "invalid number of priority groups"}, |
1120 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
1121 | }; |
1122 | ||
1123 | int r; | |
1124 | struct multipath *m; | |
498f0103 | 1125 | struct dm_arg_set as; |
1da177e4 LT |
1126 | unsigned pg_count = 0; |
1127 | unsigned next_pg_num; | |
1128 | ||
1129 | as.argc = argc; | |
1130 | as.argv = argv; | |
1131 | ||
e83068a5 | 1132 | m = alloc_multipath(ti); |
1da177e4 | 1133 | if (!m) { |
72d94861 | 1134 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
1135 | return -EINVAL; |
1136 | } | |
1137 | ||
28f16c20 | 1138 | r = parse_features(&as, m); |
1da177e4 LT |
1139 | if (r) |
1140 | goto bad; | |
1141 | ||
e83068a5 MS |
1142 | r = alloc_multipath_stage2(ti, m); |
1143 | if (r) | |
1144 | goto bad; | |
1145 | ||
28f16c20 | 1146 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
1147 | if (r) |
1148 | goto bad; | |
1149 | ||
498f0103 | 1150 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
1151 | if (r) |
1152 | goto bad; | |
1153 | ||
498f0103 | 1154 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
1155 | if (r) |
1156 | goto bad; | |
1157 | ||
a490a07a MS |
1158 | if ((!m->nr_priority_groups && next_pg_num) || |
1159 | (m->nr_priority_groups && !next_pg_num)) { | |
1160 | ti->error = "invalid initial priority group"; | |
1161 | r = -EINVAL; | |
1162 | goto bad; | |
1163 | } | |
1164 | ||
1da177e4 LT |
1165 | /* parse the priority groups */ |
1166 | while (as.argc) { | |
1167 | struct priority_group *pg; | |
91e968aa | 1168 | unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths); |
1da177e4 | 1169 | |
28f16c20 | 1170 | pg = parse_priority_group(&as, m); |
01460f35 BM |
1171 | if (IS_ERR(pg)) { |
1172 | r = PTR_ERR(pg); | |
1da177e4 LT |
1173 | goto bad; |
1174 | } | |
1175 | ||
91e968aa MS |
1176 | nr_valid_paths += pg->nr_pgpaths; |
1177 | atomic_set(&m->nr_valid_paths, nr_valid_paths); | |
1178 | ||
1da177e4 LT |
1179 | list_add_tail(&pg->list, &m->priority_groups); |
1180 | pg_count++; | |
1181 | pg->pg_num = pg_count; | |
1182 | if (!--next_pg_num) | |
1183 | m->next_pg = pg; | |
1184 | } | |
1185 | ||
1186 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 1187 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
1188 | r = -EINVAL; |
1189 | goto bad; | |
1190 | } | |
1191 | ||
55a62eef AK |
1192 | ti->num_flush_bios = 1; |
1193 | ti->num_discard_bios = 1; | |
042bcef8 | 1194 | ti->num_write_same_bios = 1; |
ac62d620 | 1195 | ti->num_write_zeroes_bios = 1; |
cd025384 | 1196 | if (m->queue_mode == DM_TYPE_BIO_BASED || m->queue_mode == DM_TYPE_NVME_BIO_BASED) |
bf661be1 | 1197 | ti->per_io_data_size = multipath_per_bio_data_size(); |
eb8db831 | 1198 | else |
8637a6bf | 1199 | ti->per_io_data_size = sizeof(struct dm_mpath_io); |
8627921f | 1200 | |
1da177e4 LT |
1201 | return 0; |
1202 | ||
1203 | bad: | |
1204 | free_multipath(m); | |
1205 | return r; | |
1206 | } | |
1207 | ||
2bded7bd KU |
1208 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
1209 | { | |
9f4c3f87 | 1210 | DEFINE_WAIT(wait); |
2bded7bd KU |
1211 | |
1212 | while (1) { | |
9f4c3f87 | 1213 | prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE); |
2bded7bd | 1214 | |
91e968aa | 1215 | if (!atomic_read(&m->pg_init_in_progress)) |
2bded7bd | 1216 | break; |
2bded7bd KU |
1217 | |
1218 | io_schedule(); | |
1219 | } | |
9f4c3f87 | 1220 | finish_wait(&m->pg_init_wait, &wait); |
2bded7bd KU |
1221 | } |
1222 | ||
1223 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 1224 | { |
848b8aef MS |
1225 | if (m->hw_handler_name) { |
1226 | set_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1227 | smp_mb__after_atomic(); | |
1228 | ||
1229 | flush_workqueue(kmpath_handlerd); | |
1230 | multipath_wait_for_pg_init_completion(m); | |
1231 | ||
1232 | clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1233 | smp_mb__after_atomic(); | |
1234 | } | |
954a73d5 | 1235 | |
a044d016 | 1236 | flush_workqueue(kmultipathd); |
43829731 | 1237 | flush_work(&m->trigger_event); |
6df400ab KU |
1238 | } |
1239 | ||
1240 | static void multipath_dtr(struct dm_target *ti) | |
1241 | { | |
1242 | struct multipath *m = ti->private; | |
1243 | ||
2bded7bd | 1244 | flush_multipath_work(m); |
1da177e4 LT |
1245 | free_multipath(m); |
1246 | } | |
1247 | ||
1da177e4 LT |
1248 | /* |
1249 | * Take a path out of use. | |
1250 | */ | |
1251 | static int fail_path(struct pgpath *pgpath) | |
1252 | { | |
1253 | unsigned long flags; | |
1254 | struct multipath *m = pgpath->pg->m; | |
1255 | ||
1256 | spin_lock_irqsave(&m->lock, flags); | |
1257 | ||
6680073d | 1258 | if (!pgpath->is_active) |
1da177e4 LT |
1259 | goto out; |
1260 | ||
72d94861 | 1261 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1262 | |
1263 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
be7d31cc | 1264 | pgpath->is_active = false; |
1da177e4 LT |
1265 | pgpath->fail_count++; |
1266 | ||
91e968aa | 1267 | atomic_dec(&m->nr_valid_paths); |
1da177e4 LT |
1268 | |
1269 | if (pgpath == m->current_pgpath) | |
1270 | m->current_pgpath = NULL; | |
1271 | ||
b15546f9 | 1272 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
91e968aa | 1273 | pgpath->path.dev->name, atomic_read(&m->nr_valid_paths)); |
b15546f9 | 1274 | |
fe9cf30e | 1275 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1276 | |
1277 | out: | |
1278 | spin_unlock_irqrestore(&m->lock, flags); | |
1279 | ||
1280 | return 0; | |
1281 | } | |
1282 | ||
1283 | /* | |
1284 | * Reinstate a previously-failed path | |
1285 | */ | |
1286 | static int reinstate_path(struct pgpath *pgpath) | |
1287 | { | |
63d832c3 | 1288 | int r = 0, run_queue = 0; |
1da177e4 LT |
1289 | unsigned long flags; |
1290 | struct multipath *m = pgpath->pg->m; | |
91e968aa | 1291 | unsigned nr_valid_paths; |
1da177e4 LT |
1292 | |
1293 | spin_lock_irqsave(&m->lock, flags); | |
1294 | ||
6680073d | 1295 | if (pgpath->is_active) |
1da177e4 LT |
1296 | goto out; |
1297 | ||
ec31f3f7 | 1298 | DMWARN("Reinstating path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1299 | |
1300 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1301 | if (r) | |
1302 | goto out; | |
1303 | ||
be7d31cc | 1304 | pgpath->is_active = true; |
1da177e4 | 1305 | |
91e968aa MS |
1306 | nr_valid_paths = atomic_inc_return(&m->nr_valid_paths); |
1307 | if (nr_valid_paths == 1) { | |
e54f77dd | 1308 | m->current_pgpath = NULL; |
63d832c3 | 1309 | run_queue = 1; |
e54f77dd | 1310 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1311 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
91e968aa | 1312 | atomic_inc(&m->pg_init_in_progress); |
e54f77dd | 1313 | } |
1da177e4 | 1314 | |
b15546f9 | 1315 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
91e968aa | 1316 | pgpath->path.dev->name, nr_valid_paths); |
b15546f9 | 1317 | |
fe9cf30e | 1318 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1319 | |
1320 | out: | |
1321 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 1322 | if (run_queue) { |
63d832c3 | 1323 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1324 | process_queued_io_list(m); |
76e33fe4 | 1325 | } |
1da177e4 LT |
1326 | |
1327 | return r; | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1332 | */ | |
1333 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1334 | action_fn action) | |
1335 | { | |
19040c0b | 1336 | int r = -EINVAL; |
1da177e4 LT |
1337 | struct pgpath *pgpath; |
1338 | struct priority_group *pg; | |
1339 | ||
1340 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1341 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1342 | if (pgpath->path.dev == dev) | |
1343 | r = action(pgpath); | |
1344 | } | |
1345 | } | |
1346 | ||
1347 | return r; | |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * Temporarily try to avoid having to use the specified PG | |
1352 | */ | |
1353 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
be7d31cc | 1354 | bool bypassed) |
1da177e4 LT |
1355 | { |
1356 | unsigned long flags; | |
1357 | ||
1358 | spin_lock_irqsave(&m->lock, flags); | |
1359 | ||
1360 | pg->bypassed = bypassed; | |
1361 | m->current_pgpath = NULL; | |
1362 | m->current_pg = NULL; | |
1363 | ||
1364 | spin_unlock_irqrestore(&m->lock, flags); | |
1365 | ||
fe9cf30e | 1366 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1367 | } |
1368 | ||
1369 | /* | |
1370 | * Switch to using the specified PG from the next I/O that gets mapped | |
1371 | */ | |
1372 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1373 | { | |
1374 | struct priority_group *pg; | |
1375 | unsigned pgnum; | |
1376 | unsigned long flags; | |
31998ef1 | 1377 | char dummy; |
1da177e4 | 1378 | |
31998ef1 | 1379 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1380 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1381 | DMWARN("invalid PG number supplied to switch_pg_num"); |
1382 | return -EINVAL; | |
1383 | } | |
1384 | ||
1385 | spin_lock_irqsave(&m->lock, flags); | |
1386 | list_for_each_entry(pg, &m->priority_groups, list) { | |
be7d31cc | 1387 | pg->bypassed = false; |
1da177e4 LT |
1388 | if (--pgnum) |
1389 | continue; | |
1390 | ||
1391 | m->current_pgpath = NULL; | |
1392 | m->current_pg = NULL; | |
1393 | m->next_pg = pg; | |
1394 | } | |
1395 | spin_unlock_irqrestore(&m->lock, flags); | |
1396 | ||
fe9cf30e | 1397 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1398 | return 0; |
1399 | } | |
1400 | ||
1401 | /* | |
1402 | * Set/clear bypassed status of a PG. | |
1403 | * PGs are numbered upwards from 1 in the order they were declared. | |
1404 | */ | |
be7d31cc | 1405 | static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed) |
1da177e4 LT |
1406 | { |
1407 | struct priority_group *pg; | |
1408 | unsigned pgnum; | |
31998ef1 | 1409 | char dummy; |
1da177e4 | 1410 | |
31998ef1 | 1411 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1412 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1413 | DMWARN("invalid PG number supplied to bypass_pg"); |
1414 | return -EINVAL; | |
1415 | } | |
1416 | ||
1417 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1418 | if (!--pgnum) | |
1419 | break; | |
1420 | } | |
1421 | ||
1422 | bypass_pg(m, pg, bypassed); | |
1423 | return 0; | |
1424 | } | |
1425 | ||
c9e45581 DW |
1426 | /* |
1427 | * Should we retry pg_init immediately? | |
1428 | */ | |
be7d31cc | 1429 | static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) |
c9e45581 DW |
1430 | { |
1431 | unsigned long flags; | |
be7d31cc | 1432 | bool limit_reached = false; |
c9e45581 DW |
1433 | |
1434 | spin_lock_irqsave(&m->lock, flags); | |
1435 | ||
91e968aa MS |
1436 | if (atomic_read(&m->pg_init_count) <= m->pg_init_retries && |
1437 | !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) | |
518257b1 | 1438 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
c9e45581 | 1439 | else |
be7d31cc | 1440 | limit_reached = true; |
c9e45581 DW |
1441 | |
1442 | spin_unlock_irqrestore(&m->lock, flags); | |
1443 | ||
1444 | return limit_reached; | |
1445 | } | |
1446 | ||
3ae31f6a | 1447 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1448 | { |
83c0d5d5 | 1449 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1450 | struct priority_group *pg = pgpath->pg; |
1451 | struct multipath *m = pg->m; | |
1452 | unsigned long flags; | |
be7d31cc | 1453 | bool delay_retry = false; |
cfae5c9b CS |
1454 | |
1455 | /* device or driver problems */ | |
1456 | switch (errors) { | |
1457 | case SCSI_DH_OK: | |
1458 | break; | |
1459 | case SCSI_DH_NOSYS: | |
1460 | if (!m->hw_handler_name) { | |
1461 | errors = 0; | |
1462 | break; | |
1463 | } | |
f7b934c8 MB |
1464 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1465 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1466 | /* |
1467 | * Fail path for now, so we do not ping pong | |
1468 | */ | |
1469 | fail_path(pgpath); | |
1470 | break; | |
1471 | case SCSI_DH_DEV_TEMP_BUSY: | |
1472 | /* | |
1473 | * Probably doing something like FW upgrade on the | |
1474 | * controller so try the other pg. | |
1475 | */ | |
be7d31cc | 1476 | bypass_pg(m, pg, true); |
cfae5c9b | 1477 | break; |
cfae5c9b | 1478 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1479 | /* Wait before retrying. */ |
1480 | delay_retry = 1; | |
7b06e09a | 1481 | /* fall through */ |
cfae5c9b CS |
1482 | case SCSI_DH_IMM_RETRY: |
1483 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1484 | if (pg_init_limit_reached(m, pgpath)) | |
1485 | fail_path(pgpath); | |
1486 | errors = 0; | |
1487 | break; | |
ec31f3f7 | 1488 | case SCSI_DH_DEV_OFFLINED: |
cfae5c9b CS |
1489 | default: |
1490 | /* | |
1491 | * We probably do not want to fail the path for a device | |
1492 | * error, but this is what the old dm did. In future | |
1493 | * patches we can do more advanced handling. | |
1494 | */ | |
1495 | fail_path(pgpath); | |
1496 | } | |
1497 | ||
1498 | spin_lock_irqsave(&m->lock, flags); | |
1499 | if (errors) { | |
e54f77dd CS |
1500 | if (pgpath == m->current_pgpath) { |
1501 | DMERR("Could not failover device. Error %d.", errors); | |
1502 | m->current_pgpath = NULL; | |
1503 | m->current_pg = NULL; | |
1504 | } | |
518257b1 | 1505 | } else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
be7d31cc | 1506 | pg->bypassed = false; |
cfae5c9b | 1507 | |
91e968aa | 1508 | if (atomic_dec_return(&m->pg_init_in_progress) > 0) |
d0259bf0 KU |
1509 | /* Activations of other paths are still on going */ |
1510 | goto out; | |
1511 | ||
518257b1 MS |
1512 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { |
1513 | if (delay_retry) | |
1514 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1515 | else | |
1516 | clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1517 | ||
3e9f1be1 HR |
1518 | if (__pg_init_all_paths(m)) |
1519 | goto out; | |
1520 | } | |
518257b1 | 1521 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
d0259bf0 | 1522 | |
7e48c768 | 1523 | process_queued_io_list(m); |
76e33fe4 | 1524 | |
2bded7bd KU |
1525 | /* |
1526 | * Wake up any thread waiting to suspend. | |
1527 | */ | |
1528 | wake_up(&m->pg_init_wait); | |
1529 | ||
d0259bf0 | 1530 | out: |
cfae5c9b CS |
1531 | spin_unlock_irqrestore(&m->lock, flags); |
1532 | } | |
1533 | ||
89bfce76 | 1534 | static void activate_or_offline_path(struct pgpath *pgpath) |
bab7cfc7 | 1535 | { |
f10e06b7 | 1536 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); |
bab7cfc7 | 1537 | |
f10e06b7 MS |
1538 | if (pgpath->is_active && !blk_queue_dying(q)) |
1539 | scsi_dh_activate(q, pg_init_done, pgpath); | |
3a017509 HR |
1540 | else |
1541 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1542 | } |
1543 | ||
89bfce76 BVA |
1544 | static void activate_path_work(struct work_struct *work) |
1545 | { | |
1546 | struct pgpath *pgpath = | |
1547 | container_of(work, struct pgpath, activate_path.work); | |
1548 | ||
1549 | activate_or_offline_path(pgpath); | |
1550 | } | |
1551 | ||
2a842aca | 1552 | static int noretry_error(blk_status_t error) |
7e782af5 HR |
1553 | { |
1554 | switch (error) { | |
2a842aca CH |
1555 | case BLK_STS_NOTSUPP: |
1556 | case BLK_STS_NOSPC: | |
1557 | case BLK_STS_TARGET: | |
1558 | case BLK_STS_NEXUS: | |
1559 | case BLK_STS_MEDIUM: | |
7e782af5 HR |
1560 | return 1; |
1561 | } | |
1562 | ||
1563 | /* Anything else could be a path failure, so should be retried */ | |
1564 | return 0; | |
1565 | } | |
1566 | ||
b79f10ee | 1567 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
2a842aca | 1568 | blk_status_t error, union map_info *map_context) |
1da177e4 | 1569 | { |
b79f10ee CH |
1570 | struct dm_mpath_io *mpio = get_mpio(map_context); |
1571 | struct pgpath *pgpath = mpio->pgpath; | |
7ed8578a | 1572 | int r = DM_ENDIO_DONE; |
b79f10ee | 1573 | |
f40c67f0 KU |
1574 | /* |
1575 | * We don't queue any clone request inside the multipath target | |
1576 | * during end I/O handling, since those clone requests don't have | |
1577 | * bio clones. If we queue them inside the multipath target, | |
1578 | * we need to make bio clones, that requires memory allocation. | |
4cc96131 | 1579 | * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests |
f40c67f0 KU |
1580 | * don't have bio clones.) |
1581 | * Instead of queueing the clone request here, we queue the original | |
1582 | * request into dm core, which will remake a clone request and | |
1583 | * clone bios for it and resubmit it later. | |
1584 | */ | |
b79f10ee CH |
1585 | if (error && !noretry_error(error)) { |
1586 | struct multipath *m = ti->private; | |
1da177e4 | 1587 | |
7ed8578a | 1588 | r = DM_ENDIO_REQUEUE; |
1da177e4 | 1589 | |
b79f10ee CH |
1590 | if (pgpath) |
1591 | fail_path(pgpath); | |
1da177e4 | 1592 | |
b79f10ee | 1593 | if (atomic_read(&m->nr_valid_paths) == 0 && |
c1fd0abe | 1594 | !must_push_back_rq(m)) { |
2a842aca | 1595 | if (error == BLK_STS_IOERR) |
18a482f5 | 1596 | dm_report_EIO(m); |
7ed8578a CH |
1597 | /* complete with the original error */ |
1598 | r = DM_ENDIO_DONE; | |
1599 | } | |
b79f10ee | 1600 | } |
466891f9 | 1601 | |
1da177e4 | 1602 | if (pgpath) { |
b79f10ee CH |
1603 | struct path_selector *ps = &pgpath->pg->ps; |
1604 | ||
1da177e4 | 1605 | if (ps->type->end_io) |
02ab823f | 1606 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1607 | } |
1da177e4 | 1608 | |
7ed8578a | 1609 | return r; |
1da177e4 LT |
1610 | } |
1611 | ||
4e4cbee9 | 1612 | static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, |
848b8aef | 1613 | blk_status_t *error) |
76e33fe4 | 1614 | { |
14ef1e48 CH |
1615 | struct multipath *m = ti->private; |
1616 | struct dm_mpath_io *mpio = get_mpio_from_bio(clone); | |
1617 | struct pgpath *pgpath = mpio->pgpath; | |
76e33fe4 | 1618 | unsigned long flags; |
1be56909 | 1619 | int r = DM_ENDIO_DONE; |
76e33fe4 | 1620 | |
4e4cbee9 | 1621 | if (!*error || noretry_error(*error)) |
14ef1e48 | 1622 | goto done; |
76e33fe4 | 1623 | |
14ef1e48 CH |
1624 | if (pgpath) |
1625 | fail_path(pgpath); | |
76e33fe4 | 1626 | |
ca5beb76 | 1627 | if (atomic_read(&m->nr_valid_paths) == 0 && |
18a482f5 | 1628 | !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { |
c1fd0abe MS |
1629 | if (must_push_back_bio(m)) { |
1630 | r = DM_ENDIO_REQUEUE; | |
1631 | } else { | |
1632 | dm_report_EIO(m); | |
1633 | *error = BLK_STS_IOERR; | |
1634 | } | |
14ef1e48 | 1635 | goto done; |
18a482f5 | 1636 | } |
76e33fe4 | 1637 | |
76e33fe4 MS |
1638 | spin_lock_irqsave(&m->lock, flags); |
1639 | bio_list_add(&m->queued_bios, clone); | |
1640 | spin_unlock_irqrestore(&m->lock, flags); | |
1641 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) | |
1642 | queue_work(kmultipathd, &m->process_queued_bios); | |
1643 | ||
1be56909 | 1644 | r = DM_ENDIO_INCOMPLETE; |
14ef1e48 | 1645 | done: |
76e33fe4 | 1646 | if (pgpath) { |
14ef1e48 CH |
1647 | struct path_selector *ps = &pgpath->pg->ps; |
1648 | ||
76e33fe4 MS |
1649 | if (ps->type->end_io) |
1650 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); | |
1651 | } | |
1652 | ||
1be56909 | 1653 | return r; |
76e33fe4 MS |
1654 | } |
1655 | ||
1da177e4 LT |
1656 | /* |
1657 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1658 | * the last path fails we must error any remaining I/O. |
1659 | * Note that if the freeze_bdev fails while suspending, the | |
1660 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1661 | */ |
1662 | static void multipath_presuspend(struct dm_target *ti) | |
1663 | { | |
7943bd6d | 1664 | struct multipath *m = ti->private; |
1da177e4 | 1665 | |
be7d31cc | 1666 | queue_if_no_path(m, false, true); |
1da177e4 LT |
1667 | } |
1668 | ||
6df400ab KU |
1669 | static void multipath_postsuspend(struct dm_target *ti) |
1670 | { | |
6380f26f MA |
1671 | struct multipath *m = ti->private; |
1672 | ||
1673 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1674 | flush_multipath_work(m); |
6380f26f | 1675 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1676 | } |
1677 | ||
436d4108 AK |
1678 | /* |
1679 | * Restore the queue_if_no_path setting. | |
1680 | */ | |
1da177e4 LT |
1681 | static void multipath_resume(struct dm_target *ti) |
1682 | { | |
7943bd6d | 1683 | struct multipath *m = ti->private; |
1814f2e3 | 1684 | unsigned long flags; |
1da177e4 | 1685 | |
1814f2e3 | 1686 | spin_lock_irqsave(&m->lock, flags); |
5307e2ad LW |
1687 | assign_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags, |
1688 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)); | |
1814f2e3 | 1689 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1690 | } |
1691 | ||
1692 | /* | |
1693 | * Info output has the following format: | |
1694 | * num_multipath_feature_args [multipath_feature_args]* | |
1695 | * num_handler_status_args [handler_status_args]* | |
1696 | * num_groups init_group_number | |
1697 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1698 | * num_paths num_selector_args | |
1699 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1700 | * | |
1701 | * Table output has the following format (identical to the constructor string): | |
1702 | * num_feature_args [features_args]* | |
1703 | * num_handler_args hw_handler [hw_handler_args]* | |
1704 | * num_groups init_group_number | |
1705 | * [priority selector-name num_ps_args [ps_args]* | |
1706 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1707 | */ | |
fd7c092e MP |
1708 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1709 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1710 | { |
1711 | int sz = 0; | |
1712 | unsigned long flags; | |
7943bd6d | 1713 | struct multipath *m = ti->private; |
1da177e4 LT |
1714 | struct priority_group *pg; |
1715 | struct pgpath *p; | |
1716 | unsigned pg_num; | |
1717 | char state; | |
1718 | ||
1719 | spin_lock_irqsave(&m->lock, flags); | |
1720 | ||
1721 | /* Features */ | |
1722 | if (type == STATUSTYPE_INFO) | |
91e968aa MS |
1723 | DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags), |
1724 | atomic_read(&m->pg_init_count)); | |
c9e45581 | 1725 | else { |
518257b1 | 1726 | DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) + |
4e2d19e4 | 1727 | (m->pg_init_retries > 0) * 2 + |
a58a935d | 1728 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
e83068a5 MS |
1729 | test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) + |
1730 | (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2); | |
1731 | ||
518257b1 | 1732 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
c9e45581 DW |
1733 | DMEMIT("queue_if_no_path "); |
1734 | if (m->pg_init_retries) | |
1735 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1736 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1737 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
518257b1 | 1738 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) |
a58a935d | 1739 | DMEMIT("retain_attached_hw_handler "); |
e83068a5 MS |
1740 | if (m->queue_mode != DM_TYPE_REQUEST_BASED) { |
1741 | switch(m->queue_mode) { | |
1742 | case DM_TYPE_BIO_BASED: | |
1743 | DMEMIT("queue_mode bio "); | |
1744 | break; | |
cd025384 MS |
1745 | case DM_TYPE_NVME_BIO_BASED: |
1746 | DMEMIT("queue_mode nvme "); | |
1747 | break; | |
e83068a5 MS |
1748 | case DM_TYPE_MQ_REQUEST_BASED: |
1749 | DMEMIT("queue_mode mq "); | |
1750 | break; | |
7e0d574f BVA |
1751 | default: |
1752 | WARN_ON_ONCE(true); | |
1753 | break; | |
e83068a5 MS |
1754 | } |
1755 | } | |
c9e45581 | 1756 | } |
1da177e4 | 1757 | |
cfae5c9b | 1758 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1759 | DMEMIT("0 "); |
1760 | else | |
cfae5c9b | 1761 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1762 | |
1763 | DMEMIT("%u ", m->nr_priority_groups); | |
1764 | ||
1765 | if (m->next_pg) | |
1766 | pg_num = m->next_pg->pg_num; | |
1767 | else if (m->current_pg) | |
1768 | pg_num = m->current_pg->pg_num; | |
1769 | else | |
a490a07a | 1770 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1771 | |
1772 | DMEMIT("%u ", pg_num); | |
1773 | ||
1774 | switch (type) { | |
1775 | case STATUSTYPE_INFO: | |
1776 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1777 | if (pg->bypassed) | |
1778 | state = 'D'; /* Disabled */ | |
1779 | else if (pg == m->current_pg) | |
1780 | state = 'A'; /* Currently Active */ | |
1781 | else | |
1782 | state = 'E'; /* Enabled */ | |
1783 | ||
1784 | DMEMIT("%c ", state); | |
1785 | ||
1786 | if (pg->ps.type->status) | |
1787 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1788 | result + sz, | |
1789 | maxlen - sz); | |
1790 | else | |
1791 | DMEMIT("0 "); | |
1792 | ||
1793 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1794 | pg->ps.type->info_args); | |
1795 | ||
1796 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1797 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1798 | p->is_active ? "A" : "F", |
1da177e4 LT |
1799 | p->fail_count); |
1800 | if (pg->ps.type->status) | |
1801 | sz += pg->ps.type->status(&pg->ps, | |
1802 | &p->path, type, result + sz, | |
1803 | maxlen - sz); | |
1804 | } | |
1805 | } | |
1806 | break; | |
1807 | ||
1808 | case STATUSTYPE_TABLE: | |
1809 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1810 | DMEMIT("%s ", pg->ps.type->name); | |
1811 | ||
1812 | if (pg->ps.type->status) | |
1813 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1814 | result + sz, | |
1815 | maxlen - sz); | |
1816 | else | |
1817 | DMEMIT("0 "); | |
1818 | ||
1819 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1820 | pg->ps.type->table_args); | |
1821 | ||
1822 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1823 | DMEMIT("%s ", p->path.dev->name); | |
1824 | if (pg->ps.type->status) | |
1825 | sz += pg->ps.type->status(&pg->ps, | |
1826 | &p->path, type, result + sz, | |
1827 | maxlen - sz); | |
1828 | } | |
1829 | } | |
1830 | break; | |
1831 | } | |
1832 | ||
1833 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1834 | } |
1835 | ||
1836 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) | |
1837 | { | |
6380f26f | 1838 | int r = -EINVAL; |
1da177e4 | 1839 | struct dm_dev *dev; |
7943bd6d | 1840 | struct multipath *m = ti->private; |
1da177e4 LT |
1841 | action_fn action; |
1842 | ||
6380f26f MA |
1843 | mutex_lock(&m->work_mutex); |
1844 | ||
c2f3d24b KU |
1845 | if (dm_suspended(ti)) { |
1846 | r = -EBUSY; | |
1847 | goto out; | |
1848 | } | |
1849 | ||
1da177e4 | 1850 | if (argc == 1) { |
498f0103 | 1851 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
be7d31cc | 1852 | r = queue_if_no_path(m, true, false); |
6380f26f | 1853 | goto out; |
498f0103 | 1854 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
be7d31cc | 1855 | r = queue_if_no_path(m, false, false); |
6380f26f MA |
1856 | goto out; |
1857 | } | |
1da177e4 LT |
1858 | } |
1859 | ||
6380f26f | 1860 | if (argc != 2) { |
a356e426 | 1861 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1862 | goto out; |
1863 | } | |
1da177e4 | 1864 | |
498f0103 | 1865 | if (!strcasecmp(argv[0], "disable_group")) { |
be7d31cc | 1866 | r = bypass_pg_num(m, argv[1], true); |
6380f26f | 1867 | goto out; |
498f0103 | 1868 | } else if (!strcasecmp(argv[0], "enable_group")) { |
be7d31cc | 1869 | r = bypass_pg_num(m, argv[1], false); |
6380f26f | 1870 | goto out; |
498f0103 | 1871 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1872 | r = switch_pg_num(m, argv[1]); |
1873 | goto out; | |
498f0103 | 1874 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1875 | action = reinstate_path; |
498f0103 | 1876 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1877 | action = fail_path; |
6380f26f | 1878 | else { |
a356e426 | 1879 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1880 | goto out; |
1881 | } | |
1da177e4 | 1882 | |
8215d6ec | 1883 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1884 | if (r) { |
72d94861 | 1885 | DMWARN("message: error getting device %s", |
1da177e4 | 1886 | argv[1]); |
6380f26f | 1887 | goto out; |
1da177e4 LT |
1888 | } |
1889 | ||
1890 | r = action_dev(m, dev, action); | |
1891 | ||
1892 | dm_put_device(ti, dev); | |
1893 | ||
6380f26f MA |
1894 | out: |
1895 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1896 | return r; |
1da177e4 LT |
1897 | } |
1898 | ||
e56f81e0 CH |
1899 | static int multipath_prepare_ioctl(struct dm_target *ti, |
1900 | struct block_device **bdev, fmode_t *mode) | |
9af4aa30 | 1901 | { |
35991652 | 1902 | struct multipath *m = ti->private; |
2da1610a | 1903 | struct pgpath *current_pgpath; |
35991652 MP |
1904 | int r; |
1905 | ||
506458ef | 1906 | current_pgpath = READ_ONCE(m->current_pgpath); |
2da1610a MS |
1907 | if (!current_pgpath) |
1908 | current_pgpath = choose_pgpath(m, 0); | |
9af4aa30 | 1909 | |
2da1610a | 1910 | if (current_pgpath) { |
518257b1 | 1911 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) { |
2da1610a MS |
1912 | *bdev = current_pgpath->path.dev->bdev; |
1913 | *mode = current_pgpath->path.dev->mode; | |
43e43c9e JN |
1914 | r = 0; |
1915 | } else { | |
1916 | /* pg_init has not started or completed */ | |
1917 | r = -ENOTCONN; | |
1918 | } | |
1919 | } else { | |
1920 | /* No path is available */ | |
518257b1 | 1921 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
43e43c9e JN |
1922 | r = -ENOTCONN; |
1923 | else | |
1924 | r = -EIO; | |
e90dae1f | 1925 | } |
9af4aa30 | 1926 | |
5bbbfdf6 | 1927 | if (r == -ENOTCONN) { |
506458ef | 1928 | if (!READ_ONCE(m->current_pg)) { |
3e9f1be1 | 1929 | /* Path status changed, redo selection */ |
2da1610a | 1930 | (void) choose_pgpath(m, 0); |
3e9f1be1 | 1931 | } |
518257b1 | 1932 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
2da1610a | 1933 | pg_init_all_paths(m); |
63d832c3 | 1934 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1935 | process_queued_io_list(m); |
3e9f1be1 | 1936 | } |
35991652 | 1937 | |
e56f81e0 CH |
1938 | /* |
1939 | * Only pass ioctls through if the device sizes match exactly. | |
1940 | */ | |
1941 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
1942 | return 1; | |
1943 | return r; | |
9af4aa30 MB |
1944 | } |
1945 | ||
af4874e0 MS |
1946 | static int multipath_iterate_devices(struct dm_target *ti, |
1947 | iterate_devices_callout_fn fn, void *data) | |
1948 | { | |
1949 | struct multipath *m = ti->private; | |
1950 | struct priority_group *pg; | |
1951 | struct pgpath *p; | |
1952 | int ret = 0; | |
1953 | ||
1954 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1955 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1956 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1957 | if (ret) |
1958 | goto out; | |
1959 | } | |
1960 | } | |
1961 | ||
1962 | out: | |
1963 | return ret; | |
1964 | } | |
1965 | ||
9f54cec5 | 1966 | static int pgpath_busy(struct pgpath *pgpath) |
f40c67f0 KU |
1967 | { |
1968 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1969 | ||
52b09914 | 1970 | return blk_lld_busy(q); |
f40c67f0 KU |
1971 | } |
1972 | ||
1973 | /* | |
1974 | * We return "busy", only when we can map I/Os but underlying devices | |
1975 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1976 | * the underlying queue). | |
1977 | * In other words, if we want to kill I/Os or queue them inside us | |
1978 | * due to map unavailability, we don't return "busy". Otherwise, | |
1979 | * dm core won't give us the I/Os and we can't do what we want. | |
1980 | */ | |
1981 | static int multipath_busy(struct dm_target *ti) | |
1982 | { | |
be7d31cc | 1983 | bool busy = false, has_active = false; |
f40c67f0 | 1984 | struct multipath *m = ti->private; |
2da1610a | 1985 | struct priority_group *pg, *next_pg; |
f40c67f0 | 1986 | struct pgpath *pgpath; |
f40c67f0 | 1987 | |
b88efd43 MS |
1988 | /* pg_init in progress */ |
1989 | if (atomic_read(&m->pg_init_in_progress)) | |
2da1610a MS |
1990 | return true; |
1991 | ||
b88efd43 MS |
1992 | /* no paths available, for blk-mq: rely on IO mapping to delay requeue */ |
1993 | if (!atomic_read(&m->nr_valid_paths) && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) | |
1994 | return (m->queue_mode != DM_TYPE_MQ_REQUEST_BASED); | |
1995 | ||
f40c67f0 | 1996 | /* Guess which priority_group will be used at next mapping time */ |
506458ef WD |
1997 | pg = READ_ONCE(m->current_pg); |
1998 | next_pg = READ_ONCE(m->next_pg); | |
1999 | if (unlikely(!READ_ONCE(m->current_pgpath) && next_pg)) | |
2da1610a MS |
2000 | pg = next_pg; |
2001 | ||
2002 | if (!pg) { | |
f40c67f0 KU |
2003 | /* |
2004 | * We don't know which pg will be used at next mapping time. | |
2da1610a | 2005 | * We don't call choose_pgpath() here to avoid to trigger |
f40c67f0 KU |
2006 | * pg_init just by busy checking. |
2007 | * So we don't know whether underlying devices we will be using | |
2008 | * at next mapping time are busy or not. Just try mapping. | |
2009 | */ | |
2da1610a MS |
2010 | return busy; |
2011 | } | |
f40c67f0 KU |
2012 | |
2013 | /* | |
2014 | * If there is one non-busy active path at least, the path selector | |
2015 | * will be able to select it. So we consider such a pg as not busy. | |
2016 | */ | |
be7d31cc | 2017 | busy = true; |
2da1610a | 2018 | list_for_each_entry(pgpath, &pg->pgpaths, list) { |
f40c67f0 | 2019 | if (pgpath->is_active) { |
be7d31cc | 2020 | has_active = true; |
9f54cec5 | 2021 | if (!pgpath_busy(pgpath)) { |
be7d31cc | 2022 | busy = false; |
f40c67f0 KU |
2023 | break; |
2024 | } | |
2025 | } | |
2da1610a | 2026 | } |
f40c67f0 | 2027 | |
2da1610a | 2028 | if (!has_active) { |
f40c67f0 KU |
2029 | /* |
2030 | * No active path in this pg, so this pg won't be used and | |
2031 | * the current_pg will be changed at next mapping time. | |
2032 | * We need to try mapping to determine it. | |
2033 | */ | |
be7d31cc | 2034 | busy = false; |
2da1610a | 2035 | } |
f40c67f0 KU |
2036 | |
2037 | return busy; | |
2038 | } | |
2039 | ||
1da177e4 LT |
2040 | /*----------------------------------------------------------------- |
2041 | * Module setup | |
2042 | *---------------------------------------------------------------*/ | |
2043 | static struct target_type multipath_target = { | |
2044 | .name = "multipath", | |
e83068a5 | 2045 | .version = {1, 12, 0}, |
16f12266 | 2046 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE, |
1da177e4 LT |
2047 | .module = THIS_MODULE, |
2048 | .ctr = multipath_ctr, | |
2049 | .dtr = multipath_dtr, | |
e5863d9a MS |
2050 | .clone_and_map_rq = multipath_clone_and_map, |
2051 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 2052 | .rq_end_io = multipath_end_io, |
76e33fe4 MS |
2053 | .map = multipath_map_bio, |
2054 | .end_io = multipath_end_io_bio, | |
2055 | .presuspend = multipath_presuspend, | |
2056 | .postsuspend = multipath_postsuspend, | |
2057 | .resume = multipath_resume, | |
2058 | .status = multipath_status, | |
2059 | .message = multipath_message, | |
2060 | .prepare_ioctl = multipath_prepare_ioctl, | |
2061 | .iterate_devices = multipath_iterate_devices, | |
2062 | .busy = multipath_busy, | |
2063 | }; | |
2064 | ||
1da177e4 LT |
2065 | static int __init dm_multipath_init(void) |
2066 | { | |
2067 | int r; | |
2068 | ||
4d4d66ab | 2069 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 2070 | if (!kmultipathd) { |
0cd33124 | 2071 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
2072 | r = -ENOMEM; |
2073 | goto bad_alloc_kmultipathd; | |
c557308e AK |
2074 | } |
2075 | ||
bab7cfc7 CS |
2076 | /* |
2077 | * A separate workqueue is used to handle the device handlers | |
2078 | * to avoid overloading existing workqueue. Overloading the | |
2079 | * old workqueue would also create a bottleneck in the | |
2080 | * path of the storage hardware device activation. | |
2081 | */ | |
4d4d66ab TH |
2082 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
2083 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
2084 | if (!kmpath_handlerd) { |
2085 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
2086 | r = -ENOMEM; |
2087 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
2088 | } |
2089 | ||
7e6358d2 | 2090 | r = dm_register_target(&multipath_target); |
2091 | if (r < 0) { | |
2092 | DMERR("request-based register failed %d", r); | |
2093 | r = -EINVAL; | |
2094 | goto bad_register_target; | |
2095 | } | |
2096 | ||
ff658e9c JT |
2097 | return 0; |
2098 | ||
7e6358d2 | 2099 | bad_register_target: |
2100 | destroy_workqueue(kmpath_handlerd); | |
ff658e9c JT |
2101 | bad_alloc_kmpath_handlerd: |
2102 | destroy_workqueue(kmultipathd); | |
2103 | bad_alloc_kmultipathd: | |
1da177e4 LT |
2104 | return r; |
2105 | } | |
2106 | ||
2107 | static void __exit dm_multipath_exit(void) | |
2108 | { | |
bab7cfc7 | 2109 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
2110 | destroy_workqueue(kmultipathd); |
2111 | ||
10d3bd09 | 2112 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
2113 | } |
2114 | ||
1da177e4 LT |
2115 | module_init(dm_multipath_init); |
2116 | module_exit(dm_multipath_exit); | |
2117 | ||
2118 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
2119 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
2120 | MODULE_LICENSE("GPL"); |