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 | 213 | |
8d47e659 | 214 | } else if (m->queue_mode == DM_TYPE_BIO_BASED) { |
e83068a5 | 215 | INIT_WORK(&m->process_queued_bios, process_queued_bios); |
8d47e659 MS |
216 | /* |
217 | * bio-based doesn't support any direct scsi_dh management; | |
218 | * it just discovers if a scsi_dh is attached. | |
219 | */ | |
220 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); | |
e83068a5 MS |
221 | } |
222 | ||
223 | dm_table_set_type(ti->table, m->queue_mode); | |
224 | ||
c3736674 MS |
225 | /* |
226 | * Init fields that are only used when a scsi_dh is attached | |
227 | * - must do this unconditionally (really doesn't hurt non-SCSI uses) | |
228 | */ | |
229 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
230 | atomic_set(&m->pg_init_in_progress, 0); | |
231 | atomic_set(&m->pg_init_count, 0); | |
232 | m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT; | |
233 | init_waitqueue_head(&m->pg_init_wait); | |
234 | ||
e83068a5 MS |
235 | return 0; |
236 | } | |
237 | ||
1da177e4 LT |
238 | static void free_multipath(struct multipath *m) |
239 | { | |
240 | struct priority_group *pg, *tmp; | |
1da177e4 LT |
241 | |
242 | list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { | |
243 | list_del(&pg->list); | |
244 | free_priority_group(pg, m->ti); | |
245 | } | |
246 | ||
cfae5c9b | 247 | kfree(m->hw_handler_name); |
2bfd2e13 | 248 | kfree(m->hw_handler_params); |
d5ffebdd | 249 | mutex_destroy(&m->work_mutex); |
1da177e4 LT |
250 | kfree(m); |
251 | } | |
252 | ||
2eff1924 MS |
253 | static struct dm_mpath_io *get_mpio(union map_info *info) |
254 | { | |
255 | return info->ptr; | |
256 | } | |
257 | ||
bf661be1 MS |
258 | static size_t multipath_per_bio_data_size(void) |
259 | { | |
260 | return sizeof(struct dm_mpath_io) + sizeof(struct dm_bio_details); | |
261 | } | |
262 | ||
76e33fe4 MS |
263 | static struct dm_mpath_io *get_mpio_from_bio(struct bio *bio) |
264 | { | |
bf661be1 | 265 | return dm_per_bio_data(bio, multipath_per_bio_data_size()); |
76e33fe4 MS |
266 | } |
267 | ||
d07a241d | 268 | static struct dm_bio_details *get_bio_details_from_mpio(struct dm_mpath_io *mpio) |
76e33fe4 | 269 | { |
bf661be1 | 270 | /* dm_bio_details is immediately after the dm_mpath_io in bio's per-bio-data */ |
bf661be1 | 271 | void *bio_details = mpio + 1; |
bf661be1 MS |
272 | return bio_details; |
273 | } | |
274 | ||
63f6e6fd | 275 | static void multipath_init_per_bio_data(struct bio *bio, struct dm_mpath_io **mpio_p) |
bf661be1 MS |
276 | { |
277 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); | |
d07a241d | 278 | struct dm_bio_details *bio_details = get_bio_details_from_mpio(mpio); |
76e33fe4 | 279 | |
d0442f80 MS |
280 | mpio->nr_bytes = bio->bi_iter.bi_size; |
281 | mpio->pgpath = NULL; | |
282 | *mpio_p = mpio; | |
76e33fe4 | 283 | |
d0442f80 | 284 | dm_bio_record(bio_details, bio); |
76e33fe4 MS |
285 | } |
286 | ||
1da177e4 LT |
287 | /*----------------------------------------------- |
288 | * Path selection | |
289 | *-----------------------------------------------*/ | |
290 | ||
3e9f1be1 | 291 | static int __pg_init_all_paths(struct multipath *m) |
fb612642 KU |
292 | { |
293 | struct pgpath *pgpath; | |
4e2d19e4 | 294 | unsigned long pg_init_delay = 0; |
fb612642 | 295 | |
b194679f BVA |
296 | lockdep_assert_held(&m->lock); |
297 | ||
91e968aa | 298 | if (atomic_read(&m->pg_init_in_progress) || test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) |
3e9f1be1 | 299 | return 0; |
17f4ff45 | 300 | |
91e968aa | 301 | atomic_inc(&m->pg_init_count); |
518257b1 | 302 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
3e9f1be1 HR |
303 | |
304 | /* Check here to reset pg_init_required */ | |
305 | if (!m->current_pg) | |
306 | return 0; | |
307 | ||
518257b1 | 308 | if (test_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags)) |
4e2d19e4 CS |
309 | pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ? |
310 | m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS); | |
fb612642 KU |
311 | list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) { |
312 | /* Skip failed paths */ | |
313 | if (!pgpath->is_active) | |
314 | continue; | |
4e2d19e4 CS |
315 | if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path, |
316 | pg_init_delay)) | |
91e968aa | 317 | atomic_inc(&m->pg_init_in_progress); |
fb612642 | 318 | } |
91e968aa | 319 | return atomic_read(&m->pg_init_in_progress); |
fb612642 KU |
320 | } |
321 | ||
c1d7ecf7 | 322 | static int pg_init_all_paths(struct multipath *m) |
1da177e4 | 323 | { |
c1d7ecf7 | 324 | int ret; |
2da1610a MS |
325 | unsigned long flags; |
326 | ||
327 | spin_lock_irqsave(&m->lock, flags); | |
c1d7ecf7 | 328 | ret = __pg_init_all_paths(m); |
2da1610a | 329 | spin_unlock_irqrestore(&m->lock, flags); |
c1d7ecf7 BVA |
330 | |
331 | return ret; | |
2da1610a MS |
332 | } |
333 | ||
334 | static void __switch_pg(struct multipath *m, struct priority_group *pg) | |
335 | { | |
336 | m->current_pg = pg; | |
1da177e4 LT |
337 | |
338 | /* Must we initialise the PG first, and queue I/O till it's ready? */ | |
cfae5c9b | 339 | if (m->hw_handler_name) { |
518257b1 MS |
340 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
341 | set_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 342 | } else { |
518257b1 MS |
343 | clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
344 | clear_bit(MPATHF_QUEUE_IO, &m->flags); | |
1da177e4 | 345 | } |
c9e45581 | 346 | |
91e968aa | 347 | atomic_set(&m->pg_init_count, 0); |
1da177e4 LT |
348 | } |
349 | ||
2da1610a MS |
350 | static struct pgpath *choose_path_in_pg(struct multipath *m, |
351 | struct priority_group *pg, | |
352 | size_t nr_bytes) | |
1da177e4 | 353 | { |
2da1610a | 354 | unsigned long flags; |
c922d5f7 | 355 | struct dm_path *path; |
2da1610a | 356 | struct pgpath *pgpath; |
1da177e4 | 357 | |
90a4323c | 358 | path = pg->ps.type->select_path(&pg->ps, nr_bytes); |
1da177e4 | 359 | if (!path) |
2da1610a | 360 | return ERR_PTR(-ENXIO); |
1da177e4 | 361 | |
2da1610a | 362 | pgpath = path_to_pgpath(path); |
1da177e4 | 363 | |
506458ef | 364 | if (unlikely(READ_ONCE(m->current_pg) != pg)) { |
2da1610a MS |
365 | /* Only update current_pgpath if pg changed */ |
366 | spin_lock_irqsave(&m->lock, flags); | |
367 | m->current_pgpath = pgpath; | |
368 | __switch_pg(m, pg); | |
369 | spin_unlock_irqrestore(&m->lock, flags); | |
370 | } | |
1da177e4 | 371 | |
2da1610a | 372 | return pgpath; |
1da177e4 LT |
373 | } |
374 | ||
2da1610a | 375 | static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes) |
1da177e4 | 376 | { |
2da1610a | 377 | unsigned long flags; |
1da177e4 | 378 | struct priority_group *pg; |
2da1610a | 379 | struct pgpath *pgpath; |
d19a55cc | 380 | unsigned bypassed = 1; |
1da177e4 | 381 | |
91e968aa | 382 | if (!atomic_read(&m->nr_valid_paths)) { |
8d47e659 | 383 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
1da177e4 | 384 | goto failed; |
1f271972 | 385 | } |
1da177e4 LT |
386 | |
387 | /* Were we instructed to switch PG? */ | |
506458ef | 388 | if (READ_ONCE(m->next_pg)) { |
2da1610a | 389 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 | 390 | pg = m->next_pg; |
2da1610a MS |
391 | if (!pg) { |
392 | spin_unlock_irqrestore(&m->lock, flags); | |
393 | goto check_current_pg; | |
394 | } | |
1da177e4 | 395 | m->next_pg = NULL; |
2da1610a MS |
396 | spin_unlock_irqrestore(&m->lock, flags); |
397 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
398 | if (!IS_ERR_OR_NULL(pgpath)) | |
399 | return pgpath; | |
1da177e4 LT |
400 | } |
401 | ||
402 | /* Don't change PG until it has no remaining paths */ | |
2da1610a | 403 | check_current_pg: |
506458ef | 404 | pg = READ_ONCE(m->current_pg); |
2da1610a MS |
405 | if (pg) { |
406 | pgpath = choose_path_in_pg(m, pg, nr_bytes); | |
407 | if (!IS_ERR_OR_NULL(pgpath)) | |
408 | return pgpath; | |
409 | } | |
1da177e4 LT |
410 | |
411 | /* | |
412 | * Loop through priority groups until we find a valid path. | |
413 | * First time we skip PGs marked 'bypassed'. | |
f220fd4e MC |
414 | * Second time we only try the ones we skipped, but set |
415 | * pg_init_delay_retry so we do not hammer controllers. | |
1da177e4 LT |
416 | */ |
417 | do { | |
418 | list_for_each_entry(pg, &m->priority_groups, list) { | |
d19a55cc | 419 | if (pg->bypassed == !!bypassed) |
1da177e4 | 420 | continue; |
2da1610a MS |
421 | pgpath = choose_path_in_pg(m, pg, nr_bytes); |
422 | if (!IS_ERR_OR_NULL(pgpath)) { | |
f220fd4e | 423 | if (!bypassed) |
518257b1 | 424 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); |
2da1610a | 425 | return pgpath; |
f220fd4e | 426 | } |
1da177e4 LT |
427 | } |
428 | } while (bypassed--); | |
429 | ||
430 | failed: | |
2da1610a | 431 | spin_lock_irqsave(&m->lock, flags); |
1da177e4 LT |
432 | m->current_pgpath = NULL; |
433 | m->current_pg = NULL; | |
2da1610a MS |
434 | spin_unlock_irqrestore(&m->lock, flags); |
435 | ||
436 | return NULL; | |
1da177e4 LT |
437 | } |
438 | ||
45e15720 | 439 | /* |
86331f39 BVA |
440 | * dm_report_EIO() is a macro instead of a function to make pr_debug() |
441 | * report the function name and line number of the function from which | |
442 | * it has been invoked. | |
45e15720 | 443 | */ |
86331f39 | 444 | #define dm_report_EIO(m) \ |
18a482f5 | 445 | do { \ |
86331f39 BVA |
446 | struct mapped_device *md = dm_table_get_md((m)->ti->table); \ |
447 | \ | |
448 | pr_debug("%s: returning EIO; QIFNP = %d; SQIFNP = %d; DNFS = %d\n", \ | |
449 | dm_device_name(md), \ | |
450 | test_bit(MPATHF_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
451 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &(m)->flags), \ | |
452 | dm_noflush_suspending((m)->ti)); \ | |
18a482f5 | 453 | } while (0) |
45e15720 | 454 | |
c1fd0abe MS |
455 | /* |
456 | * Check whether bios must be queued in the device-mapper core rather | |
457 | * than here in the target. | |
458 | * | |
459 | * If MPATHF_QUEUE_IF_NO_PATH and MPATHF_SAVED_QUEUE_IF_NO_PATH hold | |
460 | * the same value then we are not between multipath_presuspend() | |
461 | * and multipath_resume() calls and we have no need to check | |
462 | * for the DMF_NOFLUSH_SUSPENDING flag. | |
463 | */ | |
464 | static bool __must_push_back(struct multipath *m, unsigned long flags) | |
465 | { | |
466 | return ((test_bit(MPATHF_QUEUE_IF_NO_PATH, &flags) != | |
467 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &flags)) && | |
468 | dm_noflush_suspending(m->ti)); | |
469 | } | |
470 | ||
471 | /* | |
472 | * Following functions use READ_ONCE to get atomic access to | |
473 | * all m->flags to avoid taking spinlock | |
474 | */ | |
475 | static bool must_push_back_rq(struct multipath *m) | |
476 | { | |
477 | unsigned long flags = READ_ONCE(m->flags); | |
478 | return test_bit(MPATHF_QUEUE_IF_NO_PATH, &flags) || __must_push_back(m, flags); | |
479 | } | |
480 | ||
481 | static bool must_push_back_bio(struct multipath *m) | |
482 | { | |
483 | unsigned long flags = READ_ONCE(m->flags); | |
484 | return __must_push_back(m, flags); | |
485 | } | |
486 | ||
36fcffcc | 487 | /* |
76e33fe4 | 488 | * Map cloned requests (request-based multipath) |
36fcffcc | 489 | */ |
eb8db831 CH |
490 | static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, |
491 | union map_info *map_context, | |
492 | struct request **__clone) | |
1da177e4 | 493 | { |
7943bd6d | 494 | struct multipath *m = ti->private; |
eb8db831 | 495 | size_t nr_bytes = blk_rq_bytes(rq); |
1da177e4 | 496 | struct pgpath *pgpath; |
f40c67f0 | 497 | struct block_device *bdev; |
eb8db831 | 498 | struct dm_mpath_io *mpio = get_mpio(map_context); |
7083abbb | 499 | struct request_queue *q; |
eb8db831 | 500 | struct request *clone; |
1da177e4 | 501 | |
1da177e4 | 502 | /* Do we need to select a new pgpath? */ |
506458ef | 503 | pgpath = READ_ONCE(m->current_pgpath); |
2da1610a MS |
504 | if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags)) |
505 | pgpath = choose_pgpath(m, nr_bytes); | |
1da177e4 | 506 | |
9bf59a61 | 507 | if (!pgpath) { |
c1fd0abe | 508 | if (must_push_back_rq(m)) |
b88efd43 | 509 | return DM_MAPIO_DELAY_REQUEUE; |
18a482f5 | 510 | dm_report_EIO(m); /* Failed */ |
f98e0eb6 | 511 | return DM_MAPIO_KILL; |
518257b1 MS |
512 | } else if (test_bit(MPATHF_QUEUE_IO, &m->flags) || |
513 | test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { | |
459b5401 ML |
514 | pg_init_all_paths(m); |
515 | return DM_MAPIO_DELAY_REQUEUE; | |
9bf59a61 | 516 | } |
6afbc01d | 517 | |
2eb6e1e3 KB |
518 | mpio->pgpath = pgpath; |
519 | mpio->nr_bytes = nr_bytes; | |
520 | ||
9bf59a61 | 521 | bdev = pgpath->path.dev->bdev; |
7083abbb BVA |
522 | q = bdev_get_queue(bdev); |
523 | clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, GFP_ATOMIC); | |
eb8db831 CH |
524 | if (IS_ERR(clone)) { |
525 | /* EBUSY, ENODEV or EWOULDBLOCK: requeue */ | |
848b8aef | 526 | if (blk_queue_dying(q)) { |
7083abbb BVA |
527 | atomic_inc(&m->pg_init_in_progress); |
528 | activate_or_offline_path(pgpath); | |
050af08f | 529 | return DM_MAPIO_DELAY_REQUEUE; |
7083abbb | 530 | } |
050af08f ML |
531 | |
532 | /* | |
533 | * blk-mq's SCHED_RESTART can cover this requeue, so we | |
534 | * needn't deal with it by DELAY_REQUEUE. More importantly, | |
535 | * we have to return DM_MAPIO_REQUEUE so that blk-mq can | |
536 | * get the queue busy feedback (via BLK_STS_RESOURCE), | |
537 | * otherwise I/O merging can suffer. | |
538 | */ | |
539 | if (q->mq_ops) | |
540 | return DM_MAPIO_REQUEUE; | |
541 | else | |
542 | return DM_MAPIO_DELAY_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 | ||
e5863d9a MS |
556 | static void multipath_release_clone(struct request *clone) |
557 | { | |
eb8db831 | 558 | blk_put_request(clone); |
e5863d9a MS |
559 | } |
560 | ||
76e33fe4 MS |
561 | /* |
562 | * Map cloned bios (bio-based multipath) | |
563 | */ | |
0001ec56 MS |
564 | |
565 | static struct pgpath *__map_bio(struct multipath *m, struct bio *bio) | |
76e33fe4 | 566 | { |
76e33fe4 MS |
567 | struct pgpath *pgpath; |
568 | unsigned long flags; | |
569 | bool queue_io; | |
570 | ||
571 | /* Do we need to select a new pgpath? */ | |
506458ef | 572 | pgpath = READ_ONCE(m->current_pgpath); |
76e33fe4 MS |
573 | queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags); |
574 | if (!pgpath || !queue_io) | |
0001ec56 | 575 | pgpath = choose_pgpath(m, bio->bi_iter.bi_size); |
76e33fe4 MS |
576 | |
577 | if ((pgpath && queue_io) || | |
578 | (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) { | |
579 | /* Queue for the daemon to resubmit */ | |
580 | spin_lock_irqsave(&m->lock, flags); | |
581 | bio_list_add(&m->queued_bios, bio); | |
582 | spin_unlock_irqrestore(&m->lock, flags); | |
848b8aef | 583 | |
76e33fe4 MS |
584 | /* PG_INIT_REQUIRED cannot be set without QUEUE_IO */ |
585 | if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) | |
586 | pg_init_all_paths(m); | |
587 | else if (!queue_io) | |
588 | queue_work(kmultipathd, &m->process_queued_bios); | |
0001ec56 MS |
589 | |
590 | return ERR_PTR(-EAGAIN); | |
76e33fe4 MS |
591 | } |
592 | ||
0001ec56 MS |
593 | return pgpath; |
594 | } | |
595 | ||
8d47e659 | 596 | static struct pgpath *__map_bio_fast(struct multipath *m, struct bio *bio) |
0001ec56 MS |
597 | { |
598 | struct pgpath *pgpath; | |
599 | unsigned long flags; | |
600 | ||
601 | /* Do we need to select a new pgpath? */ | |
602 | /* | |
603 | * FIXME: currently only switching path if no path (due to failure, etc) | |
604 | * - which negates the point of using a path selector | |
605 | */ | |
606 | pgpath = READ_ONCE(m->current_pgpath); | |
607 | if (!pgpath) | |
608 | pgpath = choose_pgpath(m, bio->bi_iter.bi_size); | |
609 | ||
610 | if (!pgpath) { | |
611 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { | |
612 | /* Queue for the daemon to resubmit */ | |
613 | spin_lock_irqsave(&m->lock, flags); | |
614 | bio_list_add(&m->queued_bios, bio); | |
615 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 616 | queue_work(kmultipathd, &m->process_queued_bios); |
0001ec56 MS |
617 | |
618 | return ERR_PTR(-EAGAIN); | |
848b8aef | 619 | } |
0001ec56 MS |
620 | return NULL; |
621 | } | |
622 | ||
623 | return pgpath; | |
624 | } | |
625 | ||
626 | static int __multipath_map_bio(struct multipath *m, struct bio *bio, | |
627 | struct dm_mpath_io *mpio) | |
628 | { | |
629 | struct pgpath *pgpath; | |
630 | ||
8d47e659 MS |
631 | if (!m->hw_handler_name) |
632 | pgpath = __map_bio_fast(m, bio); | |
0001ec56 MS |
633 | else |
634 | pgpath = __map_bio(m, bio); | |
848b8aef | 635 | |
0001ec56 | 636 | if (IS_ERR(pgpath)) |
76e33fe4 | 637 | return DM_MAPIO_SUBMITTED; |
76e33fe4 MS |
638 | |
639 | if (!pgpath) { | |
c1fd0abe | 640 | if (must_push_back_bio(m)) |
ca5beb76 | 641 | return DM_MAPIO_REQUEUE; |
18a482f5 | 642 | dm_report_EIO(m); |
846785e6 | 643 | return DM_MAPIO_KILL; |
76e33fe4 MS |
644 | } |
645 | ||
646 | mpio->pgpath = pgpath; | |
76e33fe4 | 647 | |
4e4cbee9 | 648 | bio->bi_status = 0; |
74d46992 | 649 | bio_set_dev(bio, pgpath->path.dev->bdev); |
1eff9d32 | 650 | bio->bi_opf |= REQ_FAILFAST_TRANSPORT; |
76e33fe4 MS |
651 | |
652 | if (pgpath->pg->ps.type->start_io) | |
653 | pgpath->pg->ps.type->start_io(&pgpath->pg->ps, | |
654 | &pgpath->path, | |
d0442f80 | 655 | mpio->nr_bytes); |
76e33fe4 MS |
656 | return DM_MAPIO_REMAPPED; |
657 | } | |
658 | ||
659 | static int multipath_map_bio(struct dm_target *ti, struct bio *bio) | |
660 | { | |
661 | struct multipath *m = ti->private; | |
bf661be1 MS |
662 | struct dm_mpath_io *mpio = NULL; |
663 | ||
63f6e6fd | 664 | multipath_init_per_bio_data(bio, &mpio); |
76e33fe4 MS |
665 | return __multipath_map_bio(m, bio, mpio); |
666 | } | |
667 | ||
7e48c768 | 668 | static void process_queued_io_list(struct multipath *m) |
76e33fe4 | 669 | { |
7e48c768 MS |
670 | if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED) |
671 | dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table)); | |
8d47e659 | 672 | else if (m->queue_mode == DM_TYPE_BIO_BASED) |
76e33fe4 MS |
673 | queue_work(kmultipathd, &m->process_queued_bios); |
674 | } | |
675 | ||
676 | static void process_queued_bios(struct work_struct *work) | |
677 | { | |
678 | int r; | |
679 | unsigned long flags; | |
680 | struct bio *bio; | |
681 | struct bio_list bios; | |
682 | struct blk_plug plug; | |
683 | struct multipath *m = | |
684 | container_of(work, struct multipath, process_queued_bios); | |
685 | ||
686 | bio_list_init(&bios); | |
687 | ||
688 | spin_lock_irqsave(&m->lock, flags); | |
689 | ||
690 | if (bio_list_empty(&m->queued_bios)) { | |
691 | spin_unlock_irqrestore(&m->lock, flags); | |
692 | return; | |
693 | } | |
694 | ||
695 | bio_list_merge(&bios, &m->queued_bios); | |
696 | bio_list_init(&m->queued_bios); | |
697 | ||
698 | spin_unlock_irqrestore(&m->lock, flags); | |
699 | ||
700 | blk_start_plug(&plug); | |
701 | while ((bio = bio_list_pop(&bios))) { | |
1836df08 MS |
702 | struct dm_mpath_io *mpio = get_mpio_from_bio(bio); |
703 | dm_bio_restore(get_bio_details_from_mpio(mpio), bio); | |
704 | r = __multipath_map_bio(m, bio, mpio); | |
846785e6 CH |
705 | switch (r) { |
706 | case DM_MAPIO_KILL: | |
4e4cbee9 CH |
707 | bio->bi_status = BLK_STS_IOERR; |
708 | bio_endio(bio); | |
047385b3 | 709 | break; |
846785e6 | 710 | case DM_MAPIO_REQUEUE: |
4e4cbee9 | 711 | bio->bi_status = BLK_STS_DM_REQUEUE; |
76e33fe4 | 712 | bio_endio(bio); |
846785e6 CH |
713 | break; |
714 | case DM_MAPIO_REMAPPED: | |
76e33fe4 | 715 | generic_make_request(bio); |
846785e6 | 716 | break; |
8192a0cd | 717 | case DM_MAPIO_SUBMITTED: |
9157c8d3 BVA |
718 | break; |
719 | default: | |
720 | WARN_ONCE(true, "__multipath_map_bio() returned %d\n", r); | |
846785e6 | 721 | } |
76e33fe4 MS |
722 | } |
723 | blk_finish_plug(&plug); | |
724 | } | |
725 | ||
1da177e4 LT |
726 | /* |
727 | * If we run out of usable paths, should we queue I/O or error it? | |
728 | */ | |
be7d31cc MS |
729 | static int queue_if_no_path(struct multipath *m, bool queue_if_no_path, |
730 | bool save_old_value) | |
1da177e4 LT |
731 | { |
732 | unsigned long flags; | |
733 | ||
734 | spin_lock_irqsave(&m->lock, flags); | |
5307e2ad LW |
735 | assign_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags, |
736 | (save_old_value && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) || | |
737 | (!save_old_value && queue_if_no_path)); | |
c1fd0abe | 738 | assign_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags, queue_if_no_path); |
1da177e4 LT |
739 | spin_unlock_irqrestore(&m->lock, flags); |
740 | ||
76e33fe4 | 741 | if (!queue_if_no_path) { |
63d832c3 | 742 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 743 | process_queued_io_list(m); |
76e33fe4 | 744 | } |
63d832c3 | 745 | |
1da177e4 LT |
746 | return 0; |
747 | } | |
748 | ||
1da177e4 LT |
749 | /* |
750 | * An event is triggered whenever a path is taken out of use. | |
751 | * Includes path failure and PG bypass. | |
752 | */ | |
c4028958 | 753 | static void trigger_event(struct work_struct *work) |
1da177e4 | 754 | { |
c4028958 DH |
755 | struct multipath *m = |
756 | container_of(work, struct multipath, trigger_event); | |
1da177e4 LT |
757 | |
758 | dm_table_event(m->ti->table); | |
759 | } | |
760 | ||
761 | /*----------------------------------------------------------------- | |
762 | * Constructor/argument parsing: | |
763 | * <#multipath feature args> [<arg>]* | |
764 | * <#hw_handler args> [hw_handler [<arg>]*] | |
765 | * <#priority groups> | |
766 | * <initial priority group> | |
767 | * [<selector> <#selector args> [<arg>]* | |
768 | * <#paths> <#per-path selector args> | |
769 | * [<path> [<arg>]* ]+ ]+ | |
770 | *---------------------------------------------------------------*/ | |
498f0103 | 771 | static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg, |
1da177e4 LT |
772 | struct dm_target *ti) |
773 | { | |
774 | int r; | |
775 | struct path_selector_type *pst; | |
776 | unsigned ps_argc; | |
777 | ||
5916a22b | 778 | static const struct dm_arg _args[] = { |
72d94861 | 779 | {0, 1024, "invalid number of path selector args"}, |
1da177e4 LT |
780 | }; |
781 | ||
498f0103 | 782 | pst = dm_get_path_selector(dm_shift_arg(as)); |
1da177e4 | 783 | if (!pst) { |
72d94861 | 784 | ti->error = "unknown path selector type"; |
1da177e4 LT |
785 | return -EINVAL; |
786 | } | |
787 | ||
498f0103 | 788 | r = dm_read_arg_group(_args, as, &ps_argc, &ti->error); |
371b2e34 MP |
789 | if (r) { |
790 | dm_put_path_selector(pst); | |
1da177e4 | 791 | return -EINVAL; |
371b2e34 | 792 | } |
1da177e4 LT |
793 | |
794 | r = pst->create(&pg->ps, ps_argc, as->argv); | |
795 | if (r) { | |
796 | dm_put_path_selector(pst); | |
72d94861 | 797 | ti->error = "path selector constructor failed"; |
1da177e4 LT |
798 | return r; |
799 | } | |
800 | ||
801 | pg->ps.type = pst; | |
498f0103 | 802 | dm_consume_args(as, ps_argc); |
1da177e4 LT |
803 | |
804 | return 0; | |
805 | } | |
806 | ||
e8f74a0f MS |
807 | static int setup_scsi_dh(struct block_device *bdev, struct multipath *m, |
808 | const char *attached_handler_name, char **error) | |
1da177e4 | 809 | { |
848b8aef | 810 | struct request_queue *q = bdev_get_queue(bdev); |
848b8aef | 811 | int r; |
a58a935d | 812 | |
518257b1 | 813 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) { |
1bab0de0 | 814 | retain: |
a58a935d | 815 | if (attached_handler_name) { |
54cd640d | 816 | /* |
817 | * Clear any hw_handler_params associated with a | |
818 | * handler that isn't already attached. | |
819 | */ | |
820 | if (m->hw_handler_name && strcmp(attached_handler_name, m->hw_handler_name)) { | |
821 | kfree(m->hw_handler_params); | |
822 | m->hw_handler_params = NULL; | |
823 | } | |
824 | ||
a58a935d MS |
825 | /* |
826 | * Reset hw_handler_name to match the attached handler | |
a58a935d MS |
827 | * |
828 | * NB. This modifies the table line to show the actual | |
829 | * handler instead of the original table passed in. | |
830 | */ | |
831 | kfree(m->hw_handler_name); | |
832 | m->hw_handler_name = attached_handler_name; | |
a58a935d MS |
833 | } |
834 | } | |
a0cf7ea9 | 835 | |
a58a935d | 836 | if (m->hw_handler_name) { |
a0cf7ea9 HR |
837 | r = scsi_dh_attach(q, m->hw_handler_name); |
838 | if (r == -EBUSY) { | |
1bab0de0 | 839 | char b[BDEVNAME_SIZE]; |
a0cf7ea9 | 840 | |
1bab0de0 | 841 | printk(KERN_INFO "dm-mpath: retaining handler on device %s\n", |
848b8aef | 842 | bdevname(bdev, b)); |
1bab0de0 CH |
843 | goto retain; |
844 | } | |
ae11b1b3 | 845 | if (r < 0) { |
848b8aef MS |
846 | *error = "error attaching hardware handler"; |
847 | return r; | |
ae11b1b3 | 848 | } |
2bfd2e13 CS |
849 | |
850 | if (m->hw_handler_params) { | |
851 | r = scsi_dh_set_params(q, m->hw_handler_params); | |
852 | if (r < 0) { | |
848b8aef MS |
853 | *error = "unable to set hardware handler parameters"; |
854 | return r; | |
2bfd2e13 CS |
855 | } |
856 | } | |
ae11b1b3 HR |
857 | } |
858 | ||
848b8aef MS |
859 | return 0; |
860 | } | |
861 | ||
862 | static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps, | |
863 | struct dm_target *ti) | |
864 | { | |
865 | int r; | |
866 | struct pgpath *p; | |
867 | struct multipath *m = ti->private; | |
e8f74a0f MS |
868 | struct request_queue *q; |
869 | const char *attached_handler_name; | |
848b8aef MS |
870 | |
871 | /* we need at least a path arg */ | |
872 | if (as->argc < 1) { | |
873 | ti->error = "no device given"; | |
874 | return ERR_PTR(-EINVAL); | |
875 | } | |
876 | ||
877 | p = alloc_pgpath(); | |
878 | if (!p) | |
879 | return ERR_PTR(-ENOMEM); | |
880 | ||
881 | r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table), | |
882 | &p->path.dev); | |
883 | if (r) { | |
884 | ti->error = "error getting device"; | |
885 | goto bad; | |
886 | } | |
887 | ||
e8f74a0f MS |
888 | q = bdev_get_queue(p->path.dev->bdev); |
889 | attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL); | |
e457edf0 | 890 | if (attached_handler_name || m->hw_handler_name) { |
848b8aef | 891 | INIT_DELAYED_WORK(&p->activate_path, activate_path_work); |
e8f74a0f | 892 | r = setup_scsi_dh(p->path.dev->bdev, m, attached_handler_name, &ti->error); |
848b8aef MS |
893 | if (r) { |
894 | dm_put_device(ti, p->path.dev); | |
895 | goto bad; | |
896 | } | |
897 | } | |
898 | ||
1da177e4 LT |
899 | r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); |
900 | if (r) { | |
901 | dm_put_device(ti, p->path.dev); | |
902 | goto bad; | |
903 | } | |
904 | ||
905 | return p; | |
1da177e4 LT |
906 | bad: |
907 | free_pgpath(p); | |
01460f35 | 908 | return ERR_PTR(r); |
1da177e4 LT |
909 | } |
910 | ||
498f0103 | 911 | static struct priority_group *parse_priority_group(struct dm_arg_set *as, |
28f16c20 | 912 | struct multipath *m) |
1da177e4 | 913 | { |
5916a22b | 914 | static const struct dm_arg _args[] = { |
72d94861 AK |
915 | {1, 1024, "invalid number of paths"}, |
916 | {0, 1024, "invalid number of selector args"} | |
1da177e4 LT |
917 | }; |
918 | ||
919 | int r; | |
498f0103 | 920 | unsigned i, nr_selector_args, nr_args; |
1da177e4 | 921 | struct priority_group *pg; |
28f16c20 | 922 | struct dm_target *ti = m->ti; |
1da177e4 LT |
923 | |
924 | if (as->argc < 2) { | |
925 | as->argc = 0; | |
01460f35 BM |
926 | ti->error = "not enough priority group arguments"; |
927 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
928 | } |
929 | ||
930 | pg = alloc_priority_group(); | |
931 | if (!pg) { | |
72d94861 | 932 | ti->error = "couldn't allocate priority group"; |
01460f35 | 933 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
934 | } |
935 | pg->m = m; | |
936 | ||
937 | r = parse_path_selector(as, pg, ti); | |
938 | if (r) | |
939 | goto bad; | |
940 | ||
941 | /* | |
942 | * read the paths | |
943 | */ | |
498f0103 | 944 | r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error); |
1da177e4 LT |
945 | if (r) |
946 | goto bad; | |
947 | ||
498f0103 | 948 | r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error); |
1da177e4 LT |
949 | if (r) |
950 | goto bad; | |
951 | ||
498f0103 | 952 | nr_args = 1 + nr_selector_args; |
1da177e4 LT |
953 | for (i = 0; i < pg->nr_pgpaths; i++) { |
954 | struct pgpath *pgpath; | |
498f0103 | 955 | struct dm_arg_set path_args; |
1da177e4 | 956 | |
498f0103 | 957 | if (as->argc < nr_args) { |
148acff6 | 958 | ti->error = "not enough path parameters"; |
6bbf79a1 | 959 | r = -EINVAL; |
1da177e4 | 960 | goto bad; |
148acff6 | 961 | } |
1da177e4 | 962 | |
498f0103 | 963 | path_args.argc = nr_args; |
1da177e4 LT |
964 | path_args.argv = as->argv; |
965 | ||
966 | pgpath = parse_path(&path_args, &pg->ps, ti); | |
01460f35 BM |
967 | if (IS_ERR(pgpath)) { |
968 | r = PTR_ERR(pgpath); | |
1da177e4 | 969 | goto bad; |
01460f35 | 970 | } |
1da177e4 LT |
971 | |
972 | pgpath->pg = pg; | |
973 | list_add_tail(&pgpath->list, &pg->pgpaths); | |
498f0103 | 974 | dm_consume_args(as, nr_args); |
1da177e4 LT |
975 | } |
976 | ||
977 | return pg; | |
978 | ||
979 | bad: | |
980 | free_priority_group(pg, ti); | |
01460f35 | 981 | return ERR_PTR(r); |
1da177e4 LT |
982 | } |
983 | ||
498f0103 | 984 | static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m) |
1da177e4 | 985 | { |
1da177e4 | 986 | unsigned hw_argc; |
2bfd2e13 | 987 | int ret; |
28f16c20 | 988 | struct dm_target *ti = m->ti; |
1da177e4 | 989 | |
5916a22b | 990 | static const struct dm_arg _args[] = { |
72d94861 | 991 | {0, 1024, "invalid number of hardware handler args"}, |
1da177e4 LT |
992 | }; |
993 | ||
498f0103 | 994 | if (dm_read_arg_group(_args, as, &hw_argc, &ti->error)) |
1da177e4 LT |
995 | return -EINVAL; |
996 | ||
997 | if (!hw_argc) | |
998 | return 0; | |
999 | ||
8d47e659 | 1000 | if (m->queue_mode == DM_TYPE_BIO_BASED) { |
76e33fe4 MS |
1001 | dm_consume_args(as, hw_argc); |
1002 | DMERR("bio-based multipath doesn't allow hardware handler args"); | |
1003 | return 0; | |
1004 | } | |
1005 | ||
498f0103 | 1006 | m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL); |
f97dc421 | 1007 | if (!m->hw_handler_name) |
1008 | return -EINVAL; | |
14e98c5c | 1009 | |
2bfd2e13 CS |
1010 | if (hw_argc > 1) { |
1011 | char *p; | |
1012 | int i, j, len = 4; | |
1013 | ||
1014 | for (i = 0; i <= hw_argc - 2; i++) | |
1015 | len += strlen(as->argv[i]) + 1; | |
1016 | p = m->hw_handler_params = kzalloc(len, GFP_KERNEL); | |
1017 | if (!p) { | |
1018 | ti->error = "memory allocation failed"; | |
1019 | ret = -ENOMEM; | |
1020 | goto fail; | |
1021 | } | |
1022 | j = sprintf(p, "%d", hw_argc - 1); | |
1023 | for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1) | |
1024 | j = sprintf(p, "%s", as->argv[i]); | |
1025 | } | |
498f0103 | 1026 | dm_consume_args(as, hw_argc - 1); |
1da177e4 LT |
1027 | |
1028 | return 0; | |
2bfd2e13 CS |
1029 | fail: |
1030 | kfree(m->hw_handler_name); | |
1031 | m->hw_handler_name = NULL; | |
1032 | return ret; | |
1da177e4 LT |
1033 | } |
1034 | ||
498f0103 | 1035 | static int parse_features(struct dm_arg_set *as, struct multipath *m) |
1da177e4 LT |
1036 | { |
1037 | int r; | |
1038 | unsigned argc; | |
28f16c20 | 1039 | struct dm_target *ti = m->ti; |
498f0103 | 1040 | const char *arg_name; |
1da177e4 | 1041 | |
5916a22b | 1042 | static const struct dm_arg _args[] = { |
e83068a5 | 1043 | {0, 8, "invalid number of feature args"}, |
c9e45581 | 1044 | {1, 50, "pg_init_retries must be between 1 and 50"}, |
4e2d19e4 | 1045 | {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"}, |
1da177e4 LT |
1046 | }; |
1047 | ||
498f0103 | 1048 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
1da177e4 LT |
1049 | if (r) |
1050 | return -EINVAL; | |
1051 | ||
1052 | if (!argc) | |
1053 | return 0; | |
1054 | ||
c9e45581 | 1055 | do { |
498f0103 | 1056 | arg_name = dm_shift_arg(as); |
c9e45581 DW |
1057 | argc--; |
1058 | ||
498f0103 | 1059 | if (!strcasecmp(arg_name, "queue_if_no_path")) { |
be7d31cc | 1060 | r = queue_if_no_path(m, true, false); |
c9e45581 DW |
1061 | continue; |
1062 | } | |
1063 | ||
a58a935d | 1064 | if (!strcasecmp(arg_name, "retain_attached_hw_handler")) { |
518257b1 | 1065 | set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags); |
a58a935d MS |
1066 | continue; |
1067 | } | |
1068 | ||
498f0103 | 1069 | if (!strcasecmp(arg_name, "pg_init_retries") && |
c9e45581 | 1070 | (argc >= 1)) { |
498f0103 | 1071 | r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error); |
c9e45581 DW |
1072 | argc--; |
1073 | continue; | |
1074 | } | |
1075 | ||
498f0103 | 1076 | if (!strcasecmp(arg_name, "pg_init_delay_msecs") && |
4e2d19e4 | 1077 | (argc >= 1)) { |
498f0103 | 1078 | r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error); |
4e2d19e4 CS |
1079 | argc--; |
1080 | continue; | |
1081 | } | |
1082 | ||
e83068a5 MS |
1083 | if (!strcasecmp(arg_name, "queue_mode") && |
1084 | (argc >= 1)) { | |
1085 | const char *queue_mode_name = dm_shift_arg(as); | |
1086 | ||
1087 | if (!strcasecmp(queue_mode_name, "bio")) | |
1088 | m->queue_mode = DM_TYPE_BIO_BASED; | |
1089 | else if (!strcasecmp(queue_mode_name, "rq")) | |
1090 | m->queue_mode = DM_TYPE_REQUEST_BASED; | |
1091 | else if (!strcasecmp(queue_mode_name, "mq")) | |
1092 | m->queue_mode = DM_TYPE_MQ_REQUEST_BASED; | |
1093 | else { | |
1094 | ti->error = "Unknown 'queue_mode' requested"; | |
1095 | r = -EINVAL; | |
1096 | } | |
1097 | argc--; | |
1098 | continue; | |
1099 | } | |
1100 | ||
1da177e4 | 1101 | ti->error = "Unrecognised multipath feature request"; |
c9e45581 DW |
1102 | r = -EINVAL; |
1103 | } while (argc && !r); | |
1104 | ||
1105 | return r; | |
1da177e4 LT |
1106 | } |
1107 | ||
e83068a5 | 1108 | static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv) |
1da177e4 | 1109 | { |
498f0103 | 1110 | /* target arguments */ |
5916a22b | 1111 | static const struct dm_arg _args[] = { |
a490a07a MS |
1112 | {0, 1024, "invalid number of priority groups"}, |
1113 | {0, 1024, "invalid initial priority group number"}, | |
1da177e4 LT |
1114 | }; |
1115 | ||
1116 | int r; | |
1117 | struct multipath *m; | |
498f0103 | 1118 | struct dm_arg_set as; |
1da177e4 LT |
1119 | unsigned pg_count = 0; |
1120 | unsigned next_pg_num; | |
1121 | ||
1122 | as.argc = argc; | |
1123 | as.argv = argv; | |
1124 | ||
e83068a5 | 1125 | m = alloc_multipath(ti); |
1da177e4 | 1126 | if (!m) { |
72d94861 | 1127 | ti->error = "can't allocate multipath"; |
1da177e4 LT |
1128 | return -EINVAL; |
1129 | } | |
1130 | ||
28f16c20 | 1131 | r = parse_features(&as, m); |
1da177e4 LT |
1132 | if (r) |
1133 | goto bad; | |
1134 | ||
e83068a5 MS |
1135 | r = alloc_multipath_stage2(ti, m); |
1136 | if (r) | |
1137 | goto bad; | |
1138 | ||
28f16c20 | 1139 | r = parse_hw_handler(&as, m); |
1da177e4 LT |
1140 | if (r) |
1141 | goto bad; | |
1142 | ||
498f0103 | 1143 | r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error); |
1da177e4 LT |
1144 | if (r) |
1145 | goto bad; | |
1146 | ||
498f0103 | 1147 | r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error); |
1da177e4 LT |
1148 | if (r) |
1149 | goto bad; | |
1150 | ||
a490a07a MS |
1151 | if ((!m->nr_priority_groups && next_pg_num) || |
1152 | (m->nr_priority_groups && !next_pg_num)) { | |
1153 | ti->error = "invalid initial priority group"; | |
1154 | r = -EINVAL; | |
1155 | goto bad; | |
1156 | } | |
1157 | ||
1da177e4 LT |
1158 | /* parse the priority groups */ |
1159 | while (as.argc) { | |
1160 | struct priority_group *pg; | |
91e968aa | 1161 | unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths); |
1da177e4 | 1162 | |
28f16c20 | 1163 | pg = parse_priority_group(&as, m); |
01460f35 BM |
1164 | if (IS_ERR(pg)) { |
1165 | r = PTR_ERR(pg); | |
1da177e4 LT |
1166 | goto bad; |
1167 | } | |
1168 | ||
91e968aa MS |
1169 | nr_valid_paths += pg->nr_pgpaths; |
1170 | atomic_set(&m->nr_valid_paths, nr_valid_paths); | |
1171 | ||
1da177e4 LT |
1172 | list_add_tail(&pg->list, &m->priority_groups); |
1173 | pg_count++; | |
1174 | pg->pg_num = pg_count; | |
1175 | if (!--next_pg_num) | |
1176 | m->next_pg = pg; | |
1177 | } | |
1178 | ||
1179 | if (pg_count != m->nr_priority_groups) { | |
72d94861 | 1180 | ti->error = "priority group count mismatch"; |
1da177e4 LT |
1181 | r = -EINVAL; |
1182 | goto bad; | |
1183 | } | |
1184 | ||
55a62eef AK |
1185 | ti->num_flush_bios = 1; |
1186 | ti->num_discard_bios = 1; | |
042bcef8 | 1187 | ti->num_write_same_bios = 1; |
ac62d620 | 1188 | ti->num_write_zeroes_bios = 1; |
8d47e659 | 1189 | if (m->queue_mode == DM_TYPE_BIO_BASED) |
bf661be1 | 1190 | ti->per_io_data_size = multipath_per_bio_data_size(); |
eb8db831 | 1191 | else |
8637a6bf | 1192 | ti->per_io_data_size = sizeof(struct dm_mpath_io); |
8627921f | 1193 | |
1da177e4 LT |
1194 | return 0; |
1195 | ||
1196 | bad: | |
1197 | free_multipath(m); | |
1198 | return r; | |
1199 | } | |
1200 | ||
2bded7bd KU |
1201 | static void multipath_wait_for_pg_init_completion(struct multipath *m) |
1202 | { | |
9f4c3f87 | 1203 | DEFINE_WAIT(wait); |
2bded7bd KU |
1204 | |
1205 | while (1) { | |
9f4c3f87 | 1206 | prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE); |
2bded7bd | 1207 | |
91e968aa | 1208 | if (!atomic_read(&m->pg_init_in_progress)) |
2bded7bd | 1209 | break; |
2bded7bd KU |
1210 | |
1211 | io_schedule(); | |
1212 | } | |
9f4c3f87 | 1213 | finish_wait(&m->pg_init_wait, &wait); |
2bded7bd KU |
1214 | } |
1215 | ||
1216 | static void flush_multipath_work(struct multipath *m) | |
1da177e4 | 1217 | { |
848b8aef MS |
1218 | if (m->hw_handler_name) { |
1219 | set_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1220 | smp_mb__after_atomic(); | |
1221 | ||
1222 | flush_workqueue(kmpath_handlerd); | |
1223 | multipath_wait_for_pg_init_completion(m); | |
1224 | ||
1225 | clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags); | |
1226 | smp_mb__after_atomic(); | |
1227 | } | |
954a73d5 | 1228 | |
a044d016 | 1229 | flush_workqueue(kmultipathd); |
43829731 | 1230 | flush_work(&m->trigger_event); |
6df400ab KU |
1231 | } |
1232 | ||
1233 | static void multipath_dtr(struct dm_target *ti) | |
1234 | { | |
1235 | struct multipath *m = ti->private; | |
1236 | ||
2bded7bd | 1237 | flush_multipath_work(m); |
1da177e4 LT |
1238 | free_multipath(m); |
1239 | } | |
1240 | ||
1da177e4 LT |
1241 | /* |
1242 | * Take a path out of use. | |
1243 | */ | |
1244 | static int fail_path(struct pgpath *pgpath) | |
1245 | { | |
1246 | unsigned long flags; | |
1247 | struct multipath *m = pgpath->pg->m; | |
1248 | ||
1249 | spin_lock_irqsave(&m->lock, flags); | |
1250 | ||
6680073d | 1251 | if (!pgpath->is_active) |
1da177e4 LT |
1252 | goto out; |
1253 | ||
72d94861 | 1254 | DMWARN("Failing path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1255 | |
1256 | pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); | |
be7d31cc | 1257 | pgpath->is_active = false; |
1da177e4 LT |
1258 | pgpath->fail_count++; |
1259 | ||
91e968aa | 1260 | atomic_dec(&m->nr_valid_paths); |
1da177e4 LT |
1261 | |
1262 | if (pgpath == m->current_pgpath) | |
1263 | m->current_pgpath = NULL; | |
1264 | ||
b15546f9 | 1265 | dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti, |
91e968aa | 1266 | pgpath->path.dev->name, atomic_read(&m->nr_valid_paths)); |
b15546f9 | 1267 | |
fe9cf30e | 1268 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1269 | |
1270 | out: | |
1271 | spin_unlock_irqrestore(&m->lock, flags); | |
1272 | ||
1273 | return 0; | |
1274 | } | |
1275 | ||
1276 | /* | |
1277 | * Reinstate a previously-failed path | |
1278 | */ | |
1279 | static int reinstate_path(struct pgpath *pgpath) | |
1280 | { | |
63d832c3 | 1281 | int r = 0, run_queue = 0; |
1da177e4 LT |
1282 | unsigned long flags; |
1283 | struct multipath *m = pgpath->pg->m; | |
91e968aa | 1284 | unsigned nr_valid_paths; |
1da177e4 LT |
1285 | |
1286 | spin_lock_irqsave(&m->lock, flags); | |
1287 | ||
6680073d | 1288 | if (pgpath->is_active) |
1da177e4 LT |
1289 | goto out; |
1290 | ||
ec31f3f7 | 1291 | DMWARN("Reinstating path %s.", pgpath->path.dev->name); |
1da177e4 LT |
1292 | |
1293 | r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); | |
1294 | if (r) | |
1295 | goto out; | |
1296 | ||
be7d31cc | 1297 | pgpath->is_active = true; |
1da177e4 | 1298 | |
91e968aa MS |
1299 | nr_valid_paths = atomic_inc_return(&m->nr_valid_paths); |
1300 | if (nr_valid_paths == 1) { | |
e54f77dd | 1301 | m->current_pgpath = NULL; |
63d832c3 | 1302 | run_queue = 1; |
e54f77dd | 1303 | } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) { |
4e2d19e4 | 1304 | if (queue_work(kmpath_handlerd, &pgpath->activate_path.work)) |
91e968aa | 1305 | atomic_inc(&m->pg_init_in_progress); |
e54f77dd | 1306 | } |
1da177e4 | 1307 | |
b15546f9 | 1308 | dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti, |
91e968aa | 1309 | pgpath->path.dev->name, nr_valid_paths); |
b15546f9 | 1310 | |
fe9cf30e | 1311 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1312 | |
1313 | out: | |
1314 | spin_unlock_irqrestore(&m->lock, flags); | |
76e33fe4 | 1315 | if (run_queue) { |
63d832c3 | 1316 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1317 | process_queued_io_list(m); |
76e33fe4 | 1318 | } |
1da177e4 LT |
1319 | |
1320 | return r; | |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * Fail or reinstate all paths that match the provided struct dm_dev. | |
1325 | */ | |
1326 | static int action_dev(struct multipath *m, struct dm_dev *dev, | |
1327 | action_fn action) | |
1328 | { | |
19040c0b | 1329 | int r = -EINVAL; |
1da177e4 LT |
1330 | struct pgpath *pgpath; |
1331 | struct priority_group *pg; | |
1332 | ||
1333 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1334 | list_for_each_entry(pgpath, &pg->pgpaths, list) { | |
1335 | if (pgpath->path.dev == dev) | |
1336 | r = action(pgpath); | |
1337 | } | |
1338 | } | |
1339 | ||
1340 | return r; | |
1341 | } | |
1342 | ||
1343 | /* | |
1344 | * Temporarily try to avoid having to use the specified PG | |
1345 | */ | |
1346 | static void bypass_pg(struct multipath *m, struct priority_group *pg, | |
be7d31cc | 1347 | bool bypassed) |
1da177e4 LT |
1348 | { |
1349 | unsigned long flags; | |
1350 | ||
1351 | spin_lock_irqsave(&m->lock, flags); | |
1352 | ||
1353 | pg->bypassed = bypassed; | |
1354 | m->current_pgpath = NULL; | |
1355 | m->current_pg = NULL; | |
1356 | ||
1357 | spin_unlock_irqrestore(&m->lock, flags); | |
1358 | ||
fe9cf30e | 1359 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1360 | } |
1361 | ||
1362 | /* | |
1363 | * Switch to using the specified PG from the next I/O that gets mapped | |
1364 | */ | |
1365 | static int switch_pg_num(struct multipath *m, const char *pgstr) | |
1366 | { | |
1367 | struct priority_group *pg; | |
1368 | unsigned pgnum; | |
1369 | unsigned long flags; | |
31998ef1 | 1370 | char dummy; |
1da177e4 | 1371 | |
31998ef1 | 1372 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1373 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1374 | DMWARN("invalid PG number supplied to switch_pg_num"); |
1375 | return -EINVAL; | |
1376 | } | |
1377 | ||
1378 | spin_lock_irqsave(&m->lock, flags); | |
1379 | list_for_each_entry(pg, &m->priority_groups, list) { | |
be7d31cc | 1380 | pg->bypassed = false; |
1da177e4 LT |
1381 | if (--pgnum) |
1382 | continue; | |
1383 | ||
1384 | m->current_pgpath = NULL; | |
1385 | m->current_pg = NULL; | |
1386 | m->next_pg = pg; | |
1387 | } | |
1388 | spin_unlock_irqrestore(&m->lock, flags); | |
1389 | ||
fe9cf30e | 1390 | schedule_work(&m->trigger_event); |
1da177e4 LT |
1391 | return 0; |
1392 | } | |
1393 | ||
1394 | /* | |
1395 | * Set/clear bypassed status of a PG. | |
1396 | * PGs are numbered upwards from 1 in the order they were declared. | |
1397 | */ | |
be7d31cc | 1398 | static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed) |
1da177e4 LT |
1399 | { |
1400 | struct priority_group *pg; | |
1401 | unsigned pgnum; | |
31998ef1 | 1402 | char dummy; |
1da177e4 | 1403 | |
31998ef1 | 1404 | if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum || |
cc5bd925 | 1405 | !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) { |
1da177e4 LT |
1406 | DMWARN("invalid PG number supplied to bypass_pg"); |
1407 | return -EINVAL; | |
1408 | } | |
1409 | ||
1410 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1411 | if (!--pgnum) | |
1412 | break; | |
1413 | } | |
1414 | ||
1415 | bypass_pg(m, pg, bypassed); | |
1416 | return 0; | |
1417 | } | |
1418 | ||
c9e45581 DW |
1419 | /* |
1420 | * Should we retry pg_init immediately? | |
1421 | */ | |
be7d31cc | 1422 | static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath) |
c9e45581 DW |
1423 | { |
1424 | unsigned long flags; | |
be7d31cc | 1425 | bool limit_reached = false; |
c9e45581 DW |
1426 | |
1427 | spin_lock_irqsave(&m->lock, flags); | |
1428 | ||
91e968aa MS |
1429 | if (atomic_read(&m->pg_init_count) <= m->pg_init_retries && |
1430 | !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags)) | |
518257b1 | 1431 | set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags); |
c9e45581 | 1432 | else |
be7d31cc | 1433 | limit_reached = true; |
c9e45581 DW |
1434 | |
1435 | spin_unlock_irqrestore(&m->lock, flags); | |
1436 | ||
1437 | return limit_reached; | |
1438 | } | |
1439 | ||
3ae31f6a | 1440 | static void pg_init_done(void *data, int errors) |
cfae5c9b | 1441 | { |
83c0d5d5 | 1442 | struct pgpath *pgpath = data; |
cfae5c9b CS |
1443 | struct priority_group *pg = pgpath->pg; |
1444 | struct multipath *m = pg->m; | |
1445 | unsigned long flags; | |
be7d31cc | 1446 | bool delay_retry = false; |
cfae5c9b CS |
1447 | |
1448 | /* device or driver problems */ | |
1449 | switch (errors) { | |
1450 | case SCSI_DH_OK: | |
1451 | break; | |
1452 | case SCSI_DH_NOSYS: | |
1453 | if (!m->hw_handler_name) { | |
1454 | errors = 0; | |
1455 | break; | |
1456 | } | |
f7b934c8 MB |
1457 | DMERR("Could not failover the device: Handler scsi_dh_%s " |
1458 | "Error %d.", m->hw_handler_name, errors); | |
cfae5c9b CS |
1459 | /* |
1460 | * Fail path for now, so we do not ping pong | |
1461 | */ | |
1462 | fail_path(pgpath); | |
1463 | break; | |
1464 | case SCSI_DH_DEV_TEMP_BUSY: | |
1465 | /* | |
1466 | * Probably doing something like FW upgrade on the | |
1467 | * controller so try the other pg. | |
1468 | */ | |
be7d31cc | 1469 | bypass_pg(m, pg, true); |
cfae5c9b | 1470 | break; |
cfae5c9b | 1471 | case SCSI_DH_RETRY: |
4e2d19e4 CS |
1472 | /* Wait before retrying. */ |
1473 | delay_retry = 1; | |
7b06e09a | 1474 | /* fall through */ |
cfae5c9b CS |
1475 | case SCSI_DH_IMM_RETRY: |
1476 | case SCSI_DH_RES_TEMP_UNAVAIL: | |
1477 | if (pg_init_limit_reached(m, pgpath)) | |
1478 | fail_path(pgpath); | |
1479 | errors = 0; | |
1480 | break; | |
ec31f3f7 | 1481 | case SCSI_DH_DEV_OFFLINED: |
cfae5c9b CS |
1482 | default: |
1483 | /* | |
1484 | * We probably do not want to fail the path for a device | |
1485 | * error, but this is what the old dm did. In future | |
1486 | * patches we can do more advanced handling. | |
1487 | */ | |
1488 | fail_path(pgpath); | |
1489 | } | |
1490 | ||
1491 | spin_lock_irqsave(&m->lock, flags); | |
1492 | if (errors) { | |
e54f77dd CS |
1493 | if (pgpath == m->current_pgpath) { |
1494 | DMERR("Could not failover device. Error %d.", errors); | |
1495 | m->current_pgpath = NULL; | |
1496 | m->current_pg = NULL; | |
1497 | } | |
518257b1 | 1498 | } else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
be7d31cc | 1499 | pg->bypassed = false; |
cfae5c9b | 1500 | |
91e968aa | 1501 | if (atomic_dec_return(&m->pg_init_in_progress) > 0) |
d0259bf0 KU |
1502 | /* Activations of other paths are still on going */ |
1503 | goto out; | |
1504 | ||
518257b1 MS |
1505 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) { |
1506 | if (delay_retry) | |
1507 | set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1508 | else | |
1509 | clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags); | |
1510 | ||
3e9f1be1 HR |
1511 | if (__pg_init_all_paths(m)) |
1512 | goto out; | |
1513 | } | |
518257b1 | 1514 | clear_bit(MPATHF_QUEUE_IO, &m->flags); |
d0259bf0 | 1515 | |
7e48c768 | 1516 | process_queued_io_list(m); |
76e33fe4 | 1517 | |
2bded7bd KU |
1518 | /* |
1519 | * Wake up any thread waiting to suspend. | |
1520 | */ | |
1521 | wake_up(&m->pg_init_wait); | |
1522 | ||
d0259bf0 | 1523 | out: |
cfae5c9b CS |
1524 | spin_unlock_irqrestore(&m->lock, flags); |
1525 | } | |
1526 | ||
89bfce76 | 1527 | static void activate_or_offline_path(struct pgpath *pgpath) |
bab7cfc7 | 1528 | { |
f10e06b7 | 1529 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); |
bab7cfc7 | 1530 | |
f10e06b7 MS |
1531 | if (pgpath->is_active && !blk_queue_dying(q)) |
1532 | scsi_dh_activate(q, pg_init_done, pgpath); | |
3a017509 HR |
1533 | else |
1534 | pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED); | |
bab7cfc7 CS |
1535 | } |
1536 | ||
89bfce76 BVA |
1537 | static void activate_path_work(struct work_struct *work) |
1538 | { | |
1539 | struct pgpath *pgpath = | |
1540 | container_of(work, struct pgpath, activate_path.work); | |
1541 | ||
1542 | activate_or_offline_path(pgpath); | |
1543 | } | |
1544 | ||
b79f10ee | 1545 | static int multipath_end_io(struct dm_target *ti, struct request *clone, |
2a842aca | 1546 | blk_status_t error, union map_info *map_context) |
1da177e4 | 1547 | { |
b79f10ee CH |
1548 | struct dm_mpath_io *mpio = get_mpio(map_context); |
1549 | struct pgpath *pgpath = mpio->pgpath; | |
7ed8578a | 1550 | int r = DM_ENDIO_DONE; |
b79f10ee | 1551 | |
f40c67f0 KU |
1552 | /* |
1553 | * We don't queue any clone request inside the multipath target | |
1554 | * during end I/O handling, since those clone requests don't have | |
1555 | * bio clones. If we queue them inside the multipath target, | |
1556 | * we need to make bio clones, that requires memory allocation. | |
4cc96131 | 1557 | * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests |
f40c67f0 KU |
1558 | * don't have bio clones.) |
1559 | * Instead of queueing the clone request here, we queue the original | |
1560 | * request into dm core, which will remake a clone request and | |
1561 | * clone bios for it and resubmit it later. | |
1562 | */ | |
a1275677 | 1563 | if (error && blk_path_error(error)) { |
b79f10ee | 1564 | struct multipath *m = ti->private; |
1da177e4 | 1565 | |
ac514ffc MS |
1566 | if (error == BLK_STS_RESOURCE) |
1567 | r = DM_ENDIO_DELAY_REQUEUE; | |
1568 | else | |
1569 | r = DM_ENDIO_REQUEUE; | |
1da177e4 | 1570 | |
b79f10ee CH |
1571 | if (pgpath) |
1572 | fail_path(pgpath); | |
1da177e4 | 1573 | |
b79f10ee | 1574 | if (atomic_read(&m->nr_valid_paths) == 0 && |
c1fd0abe | 1575 | !must_push_back_rq(m)) { |
2a842aca | 1576 | if (error == BLK_STS_IOERR) |
18a482f5 | 1577 | dm_report_EIO(m); |
7ed8578a CH |
1578 | /* complete with the original error */ |
1579 | r = DM_ENDIO_DONE; | |
1580 | } | |
b79f10ee | 1581 | } |
466891f9 | 1582 | |
1da177e4 | 1583 | if (pgpath) { |
b79f10ee CH |
1584 | struct path_selector *ps = &pgpath->pg->ps; |
1585 | ||
1da177e4 | 1586 | if (ps->type->end_io) |
02ab823f | 1587 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); |
1da177e4 | 1588 | } |
1da177e4 | 1589 | |
7ed8578a | 1590 | return r; |
1da177e4 LT |
1591 | } |
1592 | ||
4e4cbee9 | 1593 | static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, |
848b8aef | 1594 | blk_status_t *error) |
76e33fe4 | 1595 | { |
14ef1e48 CH |
1596 | struct multipath *m = ti->private; |
1597 | struct dm_mpath_io *mpio = get_mpio_from_bio(clone); | |
1598 | struct pgpath *pgpath = mpio->pgpath; | |
76e33fe4 | 1599 | unsigned long flags; |
1be56909 | 1600 | int r = DM_ENDIO_DONE; |
76e33fe4 | 1601 | |
a1275677 | 1602 | if (!*error || !blk_path_error(*error)) |
14ef1e48 | 1603 | goto done; |
76e33fe4 | 1604 | |
14ef1e48 CH |
1605 | if (pgpath) |
1606 | fail_path(pgpath); | |
76e33fe4 | 1607 | |
ca5beb76 | 1608 | if (atomic_read(&m->nr_valid_paths) == 0 && |
18a482f5 | 1609 | !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) { |
c1fd0abe MS |
1610 | if (must_push_back_bio(m)) { |
1611 | r = DM_ENDIO_REQUEUE; | |
1612 | } else { | |
1613 | dm_report_EIO(m); | |
1614 | *error = BLK_STS_IOERR; | |
1615 | } | |
14ef1e48 | 1616 | goto done; |
18a482f5 | 1617 | } |
76e33fe4 | 1618 | |
76e33fe4 MS |
1619 | spin_lock_irqsave(&m->lock, flags); |
1620 | bio_list_add(&m->queued_bios, clone); | |
1621 | spin_unlock_irqrestore(&m->lock, flags); | |
1622 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) | |
1623 | queue_work(kmultipathd, &m->process_queued_bios); | |
1624 | ||
1be56909 | 1625 | r = DM_ENDIO_INCOMPLETE; |
14ef1e48 | 1626 | done: |
76e33fe4 | 1627 | if (pgpath) { |
14ef1e48 CH |
1628 | struct path_selector *ps = &pgpath->pg->ps; |
1629 | ||
76e33fe4 MS |
1630 | if (ps->type->end_io) |
1631 | ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes); | |
1632 | } | |
1633 | ||
1be56909 | 1634 | return r; |
76e33fe4 MS |
1635 | } |
1636 | ||
1da177e4 LT |
1637 | /* |
1638 | * Suspend can't complete until all the I/O is processed so if | |
436d4108 AK |
1639 | * the last path fails we must error any remaining I/O. |
1640 | * Note that if the freeze_bdev fails while suspending, the | |
1641 | * queue_if_no_path state is lost - userspace should reset it. | |
1da177e4 LT |
1642 | */ |
1643 | static void multipath_presuspend(struct dm_target *ti) | |
1644 | { | |
7943bd6d | 1645 | struct multipath *m = ti->private; |
1da177e4 | 1646 | |
be7d31cc | 1647 | queue_if_no_path(m, false, true); |
1da177e4 LT |
1648 | } |
1649 | ||
6df400ab KU |
1650 | static void multipath_postsuspend(struct dm_target *ti) |
1651 | { | |
6380f26f MA |
1652 | struct multipath *m = ti->private; |
1653 | ||
1654 | mutex_lock(&m->work_mutex); | |
2bded7bd | 1655 | flush_multipath_work(m); |
6380f26f | 1656 | mutex_unlock(&m->work_mutex); |
6df400ab KU |
1657 | } |
1658 | ||
436d4108 AK |
1659 | /* |
1660 | * Restore the queue_if_no_path setting. | |
1661 | */ | |
1da177e4 LT |
1662 | static void multipath_resume(struct dm_target *ti) |
1663 | { | |
7943bd6d | 1664 | struct multipath *m = ti->private; |
1814f2e3 | 1665 | unsigned long flags; |
1da177e4 | 1666 | |
1814f2e3 | 1667 | spin_lock_irqsave(&m->lock, flags); |
5307e2ad LW |
1668 | assign_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags, |
1669 | test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)); | |
1814f2e3 | 1670 | spin_unlock_irqrestore(&m->lock, flags); |
1da177e4 LT |
1671 | } |
1672 | ||
1673 | /* | |
1674 | * Info output has the following format: | |
1675 | * num_multipath_feature_args [multipath_feature_args]* | |
1676 | * num_handler_status_args [handler_status_args]* | |
1677 | * num_groups init_group_number | |
1678 | * [A|D|E num_ps_status_args [ps_status_args]* | |
1679 | * num_paths num_selector_args | |
1680 | * [path_dev A|F fail_count [selector_args]* ]+ ]+ | |
1681 | * | |
1682 | * Table output has the following format (identical to the constructor string): | |
1683 | * num_feature_args [features_args]* | |
1684 | * num_handler_args hw_handler [hw_handler_args]* | |
1685 | * num_groups init_group_number | |
1686 | * [priority selector-name num_ps_args [ps_args]* | |
1687 | * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ | |
1688 | */ | |
fd7c092e MP |
1689 | static void multipath_status(struct dm_target *ti, status_type_t type, |
1690 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
1691 | { |
1692 | int sz = 0; | |
1693 | unsigned long flags; | |
7943bd6d | 1694 | struct multipath *m = ti->private; |
1da177e4 LT |
1695 | struct priority_group *pg; |
1696 | struct pgpath *p; | |
1697 | unsigned pg_num; | |
1698 | char state; | |
1699 | ||
1700 | spin_lock_irqsave(&m->lock, flags); | |
1701 | ||
1702 | /* Features */ | |
1703 | if (type == STATUSTYPE_INFO) | |
91e968aa MS |
1704 | DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags), |
1705 | atomic_read(&m->pg_init_count)); | |
c9e45581 | 1706 | else { |
518257b1 | 1707 | DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) + |
4e2d19e4 | 1708 | (m->pg_init_retries > 0) * 2 + |
a58a935d | 1709 | (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 + |
e83068a5 MS |
1710 | test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) + |
1711 | (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2); | |
1712 | ||
518257b1 | 1713 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
c9e45581 DW |
1714 | DMEMIT("queue_if_no_path "); |
1715 | if (m->pg_init_retries) | |
1716 | DMEMIT("pg_init_retries %u ", m->pg_init_retries); | |
4e2d19e4 CS |
1717 | if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) |
1718 | DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs); | |
518257b1 | 1719 | if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) |
a58a935d | 1720 | DMEMIT("retain_attached_hw_handler "); |
e83068a5 MS |
1721 | if (m->queue_mode != DM_TYPE_REQUEST_BASED) { |
1722 | switch(m->queue_mode) { | |
1723 | case DM_TYPE_BIO_BASED: | |
1724 | DMEMIT("queue_mode bio "); | |
1725 | break; | |
1726 | case DM_TYPE_MQ_REQUEST_BASED: | |
1727 | DMEMIT("queue_mode mq "); | |
1728 | break; | |
7e0d574f BVA |
1729 | default: |
1730 | WARN_ON_ONCE(true); | |
1731 | break; | |
e83068a5 MS |
1732 | } |
1733 | } | |
c9e45581 | 1734 | } |
1da177e4 | 1735 | |
cfae5c9b | 1736 | if (!m->hw_handler_name || type == STATUSTYPE_INFO) |
1da177e4 LT |
1737 | DMEMIT("0 "); |
1738 | else | |
cfae5c9b | 1739 | DMEMIT("1 %s ", m->hw_handler_name); |
1da177e4 LT |
1740 | |
1741 | DMEMIT("%u ", m->nr_priority_groups); | |
1742 | ||
1743 | if (m->next_pg) | |
1744 | pg_num = m->next_pg->pg_num; | |
1745 | else if (m->current_pg) | |
1746 | pg_num = m->current_pg->pg_num; | |
1747 | else | |
a490a07a | 1748 | pg_num = (m->nr_priority_groups ? 1 : 0); |
1da177e4 LT |
1749 | |
1750 | DMEMIT("%u ", pg_num); | |
1751 | ||
1752 | switch (type) { | |
1753 | case STATUSTYPE_INFO: | |
1754 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1755 | if (pg->bypassed) | |
1756 | state = 'D'; /* Disabled */ | |
1757 | else if (pg == m->current_pg) | |
1758 | state = 'A'; /* Currently Active */ | |
1759 | else | |
1760 | state = 'E'; /* Enabled */ | |
1761 | ||
1762 | DMEMIT("%c ", state); | |
1763 | ||
1764 | if (pg->ps.type->status) | |
1765 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1766 | result + sz, | |
1767 | maxlen - sz); | |
1768 | else | |
1769 | DMEMIT("0 "); | |
1770 | ||
1771 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1772 | pg->ps.type->info_args); | |
1773 | ||
1774 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1775 | DMEMIT("%s %s %u ", p->path.dev->name, | |
6680073d | 1776 | p->is_active ? "A" : "F", |
1da177e4 LT |
1777 | p->fail_count); |
1778 | if (pg->ps.type->status) | |
1779 | sz += pg->ps.type->status(&pg->ps, | |
1780 | &p->path, type, result + sz, | |
1781 | maxlen - sz); | |
1782 | } | |
1783 | } | |
1784 | break; | |
1785 | ||
1786 | case STATUSTYPE_TABLE: | |
1787 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1788 | DMEMIT("%s ", pg->ps.type->name); | |
1789 | ||
1790 | if (pg->ps.type->status) | |
1791 | sz += pg->ps.type->status(&pg->ps, NULL, type, | |
1792 | result + sz, | |
1793 | maxlen - sz); | |
1794 | else | |
1795 | DMEMIT("0 "); | |
1796 | ||
1797 | DMEMIT("%u %u ", pg->nr_pgpaths, | |
1798 | pg->ps.type->table_args); | |
1799 | ||
1800 | list_for_each_entry(p, &pg->pgpaths, list) { | |
1801 | DMEMIT("%s ", p->path.dev->name); | |
1802 | if (pg->ps.type->status) | |
1803 | sz += pg->ps.type->status(&pg->ps, | |
1804 | &p->path, type, result + sz, | |
1805 | maxlen - sz); | |
1806 | } | |
1807 | } | |
1808 | break; | |
1809 | } | |
1810 | ||
1811 | spin_unlock_irqrestore(&m->lock, flags); | |
1da177e4 LT |
1812 | } |
1813 | ||
1eb5fa84 MS |
1814 | static int multipath_message(struct dm_target *ti, unsigned argc, char **argv, |
1815 | char *result, unsigned maxlen) | |
1da177e4 | 1816 | { |
6380f26f | 1817 | int r = -EINVAL; |
1da177e4 | 1818 | struct dm_dev *dev; |
7943bd6d | 1819 | struct multipath *m = ti->private; |
1da177e4 LT |
1820 | action_fn action; |
1821 | ||
6380f26f MA |
1822 | mutex_lock(&m->work_mutex); |
1823 | ||
c2f3d24b KU |
1824 | if (dm_suspended(ti)) { |
1825 | r = -EBUSY; | |
1826 | goto out; | |
1827 | } | |
1828 | ||
1da177e4 | 1829 | if (argc == 1) { |
498f0103 | 1830 | if (!strcasecmp(argv[0], "queue_if_no_path")) { |
be7d31cc | 1831 | r = queue_if_no_path(m, true, false); |
6380f26f | 1832 | goto out; |
498f0103 | 1833 | } else if (!strcasecmp(argv[0], "fail_if_no_path")) { |
be7d31cc | 1834 | r = queue_if_no_path(m, false, false); |
6380f26f MA |
1835 | goto out; |
1836 | } | |
1da177e4 LT |
1837 | } |
1838 | ||
6380f26f | 1839 | if (argc != 2) { |
a356e426 | 1840 | DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc); |
6380f26f MA |
1841 | goto out; |
1842 | } | |
1da177e4 | 1843 | |
498f0103 | 1844 | if (!strcasecmp(argv[0], "disable_group")) { |
be7d31cc | 1845 | r = bypass_pg_num(m, argv[1], true); |
6380f26f | 1846 | goto out; |
498f0103 | 1847 | } else if (!strcasecmp(argv[0], "enable_group")) { |
be7d31cc | 1848 | r = bypass_pg_num(m, argv[1], false); |
6380f26f | 1849 | goto out; |
498f0103 | 1850 | } else if (!strcasecmp(argv[0], "switch_group")) { |
6380f26f MA |
1851 | r = switch_pg_num(m, argv[1]); |
1852 | goto out; | |
498f0103 | 1853 | } else if (!strcasecmp(argv[0], "reinstate_path")) |
1da177e4 | 1854 | action = reinstate_path; |
498f0103 | 1855 | else if (!strcasecmp(argv[0], "fail_path")) |
1da177e4 | 1856 | action = fail_path; |
6380f26f | 1857 | else { |
a356e426 | 1858 | DMWARN("Unrecognised multipath message received: %s", argv[0]); |
6380f26f MA |
1859 | goto out; |
1860 | } | |
1da177e4 | 1861 | |
8215d6ec | 1862 | r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev); |
1da177e4 | 1863 | if (r) { |
72d94861 | 1864 | DMWARN("message: error getting device %s", |
1da177e4 | 1865 | argv[1]); |
6380f26f | 1866 | goto out; |
1da177e4 LT |
1867 | } |
1868 | ||
1869 | r = action_dev(m, dev, action); | |
1870 | ||
1871 | dm_put_device(ti, dev); | |
1872 | ||
6380f26f MA |
1873 | out: |
1874 | mutex_unlock(&m->work_mutex); | |
1da177e4 | 1875 | return r; |
1da177e4 LT |
1876 | } |
1877 | ||
e56f81e0 | 1878 | static int multipath_prepare_ioctl(struct dm_target *ti, |
5bd5e8d8 | 1879 | struct block_device **bdev) |
9af4aa30 | 1880 | { |
35991652 | 1881 | struct multipath *m = ti->private; |
2da1610a | 1882 | struct pgpath *current_pgpath; |
35991652 MP |
1883 | int r; |
1884 | ||
506458ef | 1885 | current_pgpath = READ_ONCE(m->current_pgpath); |
2da1610a MS |
1886 | if (!current_pgpath) |
1887 | current_pgpath = choose_pgpath(m, 0); | |
9af4aa30 | 1888 | |
2da1610a | 1889 | if (current_pgpath) { |
518257b1 | 1890 | if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) { |
2da1610a | 1891 | *bdev = current_pgpath->path.dev->bdev; |
43e43c9e JN |
1892 | r = 0; |
1893 | } else { | |
1894 | /* pg_init has not started or completed */ | |
1895 | r = -ENOTCONN; | |
1896 | } | |
1897 | } else { | |
1898 | /* No path is available */ | |
518257b1 | 1899 | if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) |
43e43c9e JN |
1900 | r = -ENOTCONN; |
1901 | else | |
1902 | r = -EIO; | |
e90dae1f | 1903 | } |
9af4aa30 | 1904 | |
5bbbfdf6 | 1905 | if (r == -ENOTCONN) { |
506458ef | 1906 | if (!READ_ONCE(m->current_pg)) { |
3e9f1be1 | 1907 | /* Path status changed, redo selection */ |
2da1610a | 1908 | (void) choose_pgpath(m, 0); |
3e9f1be1 | 1909 | } |
518257b1 | 1910 | if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) |
2da1610a | 1911 | pg_init_all_paths(m); |
63d832c3 | 1912 | dm_table_run_md_queue_async(m->ti->table); |
7e48c768 | 1913 | process_queued_io_list(m); |
3e9f1be1 | 1914 | } |
35991652 | 1915 | |
e56f81e0 CH |
1916 | /* |
1917 | * Only pass ioctls through if the device sizes match exactly. | |
1918 | */ | |
1919 | if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT) | |
1920 | return 1; | |
1921 | return r; | |
9af4aa30 MB |
1922 | } |
1923 | ||
af4874e0 MS |
1924 | static int multipath_iterate_devices(struct dm_target *ti, |
1925 | iterate_devices_callout_fn fn, void *data) | |
1926 | { | |
1927 | struct multipath *m = ti->private; | |
1928 | struct priority_group *pg; | |
1929 | struct pgpath *p; | |
1930 | int ret = 0; | |
1931 | ||
1932 | list_for_each_entry(pg, &m->priority_groups, list) { | |
1933 | list_for_each_entry(p, &pg->pgpaths, list) { | |
5dea271b | 1934 | ret = fn(ti, p->path.dev, ti->begin, ti->len, data); |
af4874e0 MS |
1935 | if (ret) |
1936 | goto out; | |
1937 | } | |
1938 | } | |
1939 | ||
1940 | out: | |
1941 | return ret; | |
1942 | } | |
1943 | ||
9f54cec5 | 1944 | static int pgpath_busy(struct pgpath *pgpath) |
f40c67f0 KU |
1945 | { |
1946 | struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev); | |
1947 | ||
52b09914 | 1948 | return blk_lld_busy(q); |
f40c67f0 KU |
1949 | } |
1950 | ||
1951 | /* | |
1952 | * We return "busy", only when we can map I/Os but underlying devices | |
1953 | * are busy (so even if we map I/Os now, the I/Os will wait on | |
1954 | * the underlying queue). | |
1955 | * In other words, if we want to kill I/Os or queue them inside us | |
1956 | * due to map unavailability, we don't return "busy". Otherwise, | |
1957 | * dm core won't give us the I/Os and we can't do what we want. | |
1958 | */ | |
1959 | static int multipath_busy(struct dm_target *ti) | |
1960 | { | |
be7d31cc | 1961 | bool busy = false, has_active = false; |
f40c67f0 | 1962 | struct multipath *m = ti->private; |
2da1610a | 1963 | struct priority_group *pg, *next_pg; |
f40c67f0 | 1964 | struct pgpath *pgpath; |
f40c67f0 | 1965 | |
b88efd43 MS |
1966 | /* pg_init in progress */ |
1967 | if (atomic_read(&m->pg_init_in_progress)) | |
2da1610a MS |
1968 | return true; |
1969 | ||
b88efd43 MS |
1970 | /* no paths available, for blk-mq: rely on IO mapping to delay requeue */ |
1971 | if (!atomic_read(&m->nr_valid_paths) && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) | |
1972 | return (m->queue_mode != DM_TYPE_MQ_REQUEST_BASED); | |
1973 | ||
f40c67f0 | 1974 | /* Guess which priority_group will be used at next mapping time */ |
506458ef WD |
1975 | pg = READ_ONCE(m->current_pg); |
1976 | next_pg = READ_ONCE(m->next_pg); | |
1977 | if (unlikely(!READ_ONCE(m->current_pgpath) && next_pg)) | |
2da1610a MS |
1978 | pg = next_pg; |
1979 | ||
1980 | if (!pg) { | |
f40c67f0 KU |
1981 | /* |
1982 | * We don't know which pg will be used at next mapping time. | |
2da1610a | 1983 | * We don't call choose_pgpath() here to avoid to trigger |
f40c67f0 KU |
1984 | * pg_init just by busy checking. |
1985 | * So we don't know whether underlying devices we will be using | |
1986 | * at next mapping time are busy or not. Just try mapping. | |
1987 | */ | |
2da1610a MS |
1988 | return busy; |
1989 | } | |
f40c67f0 KU |
1990 | |
1991 | /* | |
1992 | * If there is one non-busy active path at least, the path selector | |
1993 | * will be able to select it. So we consider such a pg as not busy. | |
1994 | */ | |
be7d31cc | 1995 | busy = true; |
2da1610a | 1996 | list_for_each_entry(pgpath, &pg->pgpaths, list) { |
f40c67f0 | 1997 | if (pgpath->is_active) { |
be7d31cc | 1998 | has_active = true; |
9f54cec5 | 1999 | if (!pgpath_busy(pgpath)) { |
be7d31cc | 2000 | busy = false; |
f40c67f0 KU |
2001 | break; |
2002 | } | |
2003 | } | |
2da1610a | 2004 | } |
f40c67f0 | 2005 | |
2da1610a | 2006 | if (!has_active) { |
f40c67f0 KU |
2007 | /* |
2008 | * No active path in this pg, so this pg won't be used and | |
2009 | * the current_pg will be changed at next mapping time. | |
2010 | * We need to try mapping to determine it. | |
2011 | */ | |
be7d31cc | 2012 | busy = false; |
2da1610a | 2013 | } |
f40c67f0 KU |
2014 | |
2015 | return busy; | |
2016 | } | |
2017 | ||
1da177e4 LT |
2018 | /*----------------------------------------------------------------- |
2019 | * Module setup | |
2020 | *---------------------------------------------------------------*/ | |
2021 | static struct target_type multipath_target = { | |
2022 | .name = "multipath", | |
8c5c1473 SM |
2023 | .version = {1, 13, 0}, |
2024 | .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE | | |
2025 | DM_TARGET_PASSES_INTEGRITY, | |
1da177e4 LT |
2026 | .module = THIS_MODULE, |
2027 | .ctr = multipath_ctr, | |
2028 | .dtr = multipath_dtr, | |
e5863d9a MS |
2029 | .clone_and_map_rq = multipath_clone_and_map, |
2030 | .release_clone_rq = multipath_release_clone, | |
f40c67f0 | 2031 | .rq_end_io = multipath_end_io, |
76e33fe4 MS |
2032 | .map = multipath_map_bio, |
2033 | .end_io = multipath_end_io_bio, | |
2034 | .presuspend = multipath_presuspend, | |
2035 | .postsuspend = multipath_postsuspend, | |
2036 | .resume = multipath_resume, | |
2037 | .status = multipath_status, | |
2038 | .message = multipath_message, | |
2039 | .prepare_ioctl = multipath_prepare_ioctl, | |
2040 | .iterate_devices = multipath_iterate_devices, | |
2041 | .busy = multipath_busy, | |
2042 | }; | |
2043 | ||
1da177e4 LT |
2044 | static int __init dm_multipath_init(void) |
2045 | { | |
2046 | int r; | |
2047 | ||
4d4d66ab | 2048 | kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0); |
c557308e | 2049 | if (!kmultipathd) { |
0cd33124 | 2050 | DMERR("failed to create workqueue kmpathd"); |
ff658e9c JT |
2051 | r = -ENOMEM; |
2052 | goto bad_alloc_kmultipathd; | |
c557308e AK |
2053 | } |
2054 | ||
bab7cfc7 CS |
2055 | /* |
2056 | * A separate workqueue is used to handle the device handlers | |
2057 | * to avoid overloading existing workqueue. Overloading the | |
2058 | * old workqueue would also create a bottleneck in the | |
2059 | * path of the storage hardware device activation. | |
2060 | */ | |
4d4d66ab TH |
2061 | kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd", |
2062 | WQ_MEM_RECLAIM); | |
bab7cfc7 CS |
2063 | if (!kmpath_handlerd) { |
2064 | DMERR("failed to create workqueue kmpath_handlerd"); | |
ff658e9c JT |
2065 | r = -ENOMEM; |
2066 | goto bad_alloc_kmpath_handlerd; | |
bab7cfc7 CS |
2067 | } |
2068 | ||
7e6358d2 | 2069 | r = dm_register_target(&multipath_target); |
2070 | if (r < 0) { | |
2071 | DMERR("request-based register failed %d", r); | |
2072 | r = -EINVAL; | |
2073 | goto bad_register_target; | |
2074 | } | |
2075 | ||
ff658e9c JT |
2076 | return 0; |
2077 | ||
7e6358d2 | 2078 | bad_register_target: |
2079 | destroy_workqueue(kmpath_handlerd); | |
ff658e9c JT |
2080 | bad_alloc_kmpath_handlerd: |
2081 | destroy_workqueue(kmultipathd); | |
2082 | bad_alloc_kmultipathd: | |
1da177e4 LT |
2083 | return r; |
2084 | } | |
2085 | ||
2086 | static void __exit dm_multipath_exit(void) | |
2087 | { | |
bab7cfc7 | 2088 | destroy_workqueue(kmpath_handlerd); |
c557308e AK |
2089 | destroy_workqueue(kmultipathd); |
2090 | ||
10d3bd09 | 2091 | dm_unregister_target(&multipath_target); |
1da177e4 LT |
2092 | } |
2093 | ||
1da177e4 LT |
2094 | module_init(dm_multipath_init); |
2095 | module_exit(dm_multipath_exit); | |
2096 | ||
2097 | MODULE_DESCRIPTION(DM_NAME " multipath target"); | |
2098 | MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); | |
2099 | MODULE_LICENSE("GPL"); |