blob: 0d437c98ab0872a8dd34e0f97b32729965fa7c63 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Mikulas Patocka586e80e2008-10-21 17:44:59 +01008#include <linux/device-mapper.h>
9
Mike Snitzer4cc96132016-05-12 16:28:10 -040010#include "dm-rq.h"
Mike Snitzer76e33fe2016-05-19 16:15:14 -040011#include "dm-bio-record.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include "dm-path-selector.h"
Mike Andersonb15546f2007-10-19 22:48:02 +010013#include "dm-uevent.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070014
Mike Snitzere5863d92014-12-17 21:08:12 -050015#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#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>
Mikulas Patocka35991652012-06-03 00:29:58 +010024#include <linux/delay.h>
Chandra Seetharamancfae5c92008-05-01 14:50:11 -070025#include <scsi/scsi_dh.h>
Arun Sharma600634972011-07-26 16:09:06 -070026#include <linux/atomic.h>
Mike Snitzer78ce23b2016-01-31 17:38:28 -050027#include <linux/blk-mq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
Alasdair G Kergon72d94862006-06-26 00:27:35 -070029#define DM_MSG_PREFIX "multipath"
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000030#define DM_PG_INIT_DELAY_MSECS 2000
31#define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33/* Path properties */
34struct pgpath {
35 struct list_head list;
36
37 struct priority_group *pg; /* Owning PG */
38 unsigned fail_count; /* Cumulative failure count */
39
Josef "Jeff" Sipekc922d5f2006-12-08 02:36:33 -080040 struct dm_path path;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000041 struct delayed_work activate_path;
Mike Snitzerbe7d31c2016-02-10 13:02:21 -050042
43 bool is_active:1; /* Path status */
Linus Torvalds1da177e2005-04-16 15:20:36 -070044};
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 */
52struct 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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 unsigned nr_pgpaths; /* Number of paths in PG */
60 struct list_head pgpaths;
Mike Snitzerbe7d31c2016-02-10 13:02:21 -050061
62 bool bypassed:1; /* Temporarily bypass this PG? */
Linus Torvalds1da177e2005-04-16 15:20:36 -070063};
64
65/* Multipath context */
66struct multipath {
67 struct list_head list;
68 struct dm_target *ti;
69
Chandra Seetharamancfae5c92008-05-01 14:50:11 -070070 const char *hw_handler_name;
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -070071 char *hw_handler_params;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000072
Mike Snitzer1fbdd2b2012-06-03 00:29:43 +010073 spinlock_t lock;
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 unsigned nr_priority_groups;
76 struct list_head priority_groups;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000077
78 wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 struct pgpath *current_pgpath;
81 struct priority_group *current_pg;
82 struct priority_group *next_pg; /* Switch to this PG if set */
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Mike Snitzer518257b2016-03-17 16:32:10 -040084 unsigned long flags; /* Multipath state flags */
Mike Snitzer1fbdd2b2012-06-03 00:29:43 +010085
Dave Wysochanskic9e45582007-10-19 22:47:53 +010086 unsigned pg_init_retries; /* Number of times to retry pg_init */
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000087 unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Mike Snitzer91e968a2016-03-17 17:10:15 -040089 atomic_t nr_valid_paths; /* Total number of usable paths */
90 atomic_t pg_init_in_progress; /* Only one pg_init allowed at once */
91 atomic_t pg_init_count; /* Number of times pg_init called */
92
Mike Snitzere83068a2016-05-24 21:16:51 -040093 unsigned queue_mode;
94
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 /*
Alasdair G Kergon028867a2007-07-12 17:26:32 +010096 * We must use a mempool of dm_mpath_io structs so that we
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * can resubmit bios on error.
98 */
99 mempool_t *mpio_pool;
Mike Anderson6380f262009-12-10 23:52:21 +0000100
101 struct mutex work_mutex;
Mike Snitzer20800cb2016-03-17 17:13:10 -0400102 struct work_struct trigger_event;
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400103
104 struct work_struct process_queued_bios;
105 struct bio_list queued_bios;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106};
107
108/*
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400109 * Context information attached to each io we process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100111struct dm_mpath_io {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 struct pgpath *pgpath;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100113 size_t nr_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114};
115
116typedef int (*action_fn) (struct pgpath *pgpath);
117
Christoph Lametere18b8902006-12-06 20:33:20 -0800118static struct kmem_cache *_mpio_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -0700120static struct workqueue_struct *kmultipathd, *kmpath_handlerd;
David Howellsc4028952006-11-22 14:57:56 +0000121static void trigger_event(struct work_struct *work);
Bart Van Assche042d8db2017-04-27 10:11:14 -0700122static void activate_or_offline_path(struct pgpath *pgpath);
123static void activate_path_work(struct work_struct *work);
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400124static void process_queued_bios(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Mike Snitzer518257b2016-03-17 16:32:10 -0400126/*-----------------------------------------------
127 * Multipath state flags.
128 *-----------------------------------------------*/
129
130#define MPATHF_QUEUE_IO 0 /* Must we queue all I/O? */
131#define MPATHF_QUEUE_IF_NO_PATH 1 /* Queue I/O if last path fails? */
132#define MPATHF_SAVED_QUEUE_IF_NO_PATH 2 /* Saved state during suspension */
133#define MPATHF_RETAIN_ATTACHED_HW_HANDLER 3 /* If there's already a hw_handler present, don't change it. */
134#define MPATHF_PG_INIT_DISABLED 4 /* pg_init is not currently allowed */
135#define MPATHF_PG_INIT_REQUIRED 5 /* pg_init needs calling? */
136#define MPATHF_PG_INIT_DELAY_RETRY 6 /* Delay pg_init retry? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138/*-----------------------------------------------
139 * Allocation routines
140 *-----------------------------------------------*/
141
142static struct pgpath *alloc_pgpath(void)
143{
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700144 struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Mike Anderson224cb3e2008-08-29 09:36:09 +0200146 if (pgpath) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -0500147 pgpath->is_active = true;
Bart Van Assche042d8db2017-04-27 10:11:14 -0700148 INIT_DELAYED_WORK(&pgpath->activate_path, activate_path_work);
Mike Anderson224cb3e2008-08-29 09:36:09 +0200149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
151 return pgpath;
152}
153
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100154static void free_pgpath(struct pgpath *pgpath)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
156 kfree(pgpath);
157}
158
159static struct priority_group *alloc_priority_group(void)
160{
161 struct priority_group *pg;
162
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700163 pg = kzalloc(sizeof(*pg), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700165 if (pg)
166 INIT_LIST_HEAD(&pg->pgpaths);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168 return pg;
169}
170
171static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti)
172{
173 struct pgpath *pgpath, *tmp;
174
175 list_for_each_entry_safe(pgpath, tmp, pgpaths, list) {
176 list_del(&pgpath->list);
177 dm_put_device(ti, pgpath->path.dev);
178 free_pgpath(pgpath);
179 }
180}
181
182static void free_priority_group(struct priority_group *pg,
183 struct dm_target *ti)
184{
185 struct path_selector *ps = &pg->ps;
186
187 if (ps->type) {
188 ps->type->destroy(ps);
189 dm_put_path_selector(ps->type);
190 }
191
192 free_pgpaths(&pg->pgpaths, ti);
193 kfree(pg);
194}
195
Mike Snitzere83068a2016-05-24 21:16:51 -0400196static struct multipath *alloc_multipath(struct dm_target *ti)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
198 struct multipath *m;
199
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700200 m = kzalloc(sizeof(*m), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 if (m) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 INIT_LIST_HEAD(&m->priority_groups);
203 spin_lock_init(&m->lock);
Mike Snitzer518257b2016-03-17 16:32:10 -0400204 set_bit(MPATHF_QUEUE_IO, &m->flags);
Mike Snitzer91e968a2016-03-17 17:10:15 -0400205 atomic_set(&m->nr_valid_paths, 0);
206 atomic_set(&m->pg_init_in_progress, 0);
207 atomic_set(&m->pg_init_count, 0);
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000208 m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT;
David Howellsc4028952006-11-22 14:57:56 +0000209 INIT_WORK(&m->trigger_event, trigger_event);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +0000210 init_waitqueue_head(&m->pg_init_wait);
Mike Anderson6380f262009-12-10 23:52:21 +0000211 mutex_init(&m->work_mutex);
Mike Snitzer8637a6b2016-01-31 12:08:36 -0500212
213 m->mpio_pool = NULL;
Mike Snitzere83068a2016-05-24 21:16:51 -0400214 m->queue_mode = DM_TYPE_NONE;
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400215
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700216 m->ti = ti;
217 ti->private = m;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 }
219
220 return m;
221}
222
Mike Snitzere83068a2016-05-24 21:16:51 -0400223static int alloc_multipath_stage2(struct dm_target *ti, struct multipath *m)
224{
225 if (m->queue_mode == DM_TYPE_NONE) {
226 /*
227 * Default to request-based.
228 */
229 if (dm_use_blk_mq(dm_table_get_md(ti->table)))
230 m->queue_mode = DM_TYPE_MQ_REQUEST_BASED;
231 else
232 m->queue_mode = DM_TYPE_REQUEST_BASED;
233 }
234
235 if (m->queue_mode == DM_TYPE_REQUEST_BASED) {
236 unsigned min_ios = dm_get_reserved_rq_based_ios();
237
238 m->mpio_pool = mempool_create_slab_pool(min_ios, _mpio_cache);
239 if (!m->mpio_pool)
240 return -ENOMEM;
241 }
242 else if (m->queue_mode == DM_TYPE_BIO_BASED) {
243 INIT_WORK(&m->process_queued_bios, process_queued_bios);
244 /*
245 * bio-based doesn't support any direct scsi_dh management;
246 * it just discovers if a scsi_dh is attached.
247 */
248 set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags);
249 }
250
251 dm_table_set_type(ti->table, m->queue_mode);
252
253 return 0;
254}
255
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256static void free_multipath(struct multipath *m)
257{
258 struct priority_group *pg, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260 list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) {
261 list_del(&pg->list);
262 free_priority_group(pg, m->ti);
263 }
264
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700265 kfree(m->hw_handler_name);
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700266 kfree(m->hw_handler_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 mempool_destroy(m->mpio_pool);
268 kfree(m);
269}
270
Mike Snitzer2eff1922016-02-03 09:13:14 -0500271static struct dm_mpath_io *get_mpio(union map_info *info)
272{
273 return info->ptr;
274}
275
276static struct dm_mpath_io *set_mpio(struct multipath *m, union map_info *info)
Jun'ichi Nomura466891f2012-03-28 18:41:25 +0100277{
278 struct dm_mpath_io *mpio;
279
Mike Snitzer8637a6b2016-01-31 12:08:36 -0500280 if (!m->mpio_pool) {
281 /* Use blk-mq pdu memory requested via per_io_data_size */
Mike Snitzer2eff1922016-02-03 09:13:14 -0500282 mpio = get_mpio(info);
Mike Snitzer8637a6b2016-01-31 12:08:36 -0500283 memset(mpio, 0, sizeof(*mpio));
284 return mpio;
285 }
286
Jun'ichi Nomura466891f2012-03-28 18:41:25 +0100287 mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC);
288 if (!mpio)
Mike Snitzer2eff1922016-02-03 09:13:14 -0500289 return NULL;
Jun'ichi Nomura466891f2012-03-28 18:41:25 +0100290
291 memset(mpio, 0, sizeof(*mpio));
292 info->ptr = mpio;
293
Mike Snitzer2eff1922016-02-03 09:13:14 -0500294 return mpio;
Jun'ichi Nomura466891f2012-03-28 18:41:25 +0100295}
296
Mike Snitzer2eff1922016-02-03 09:13:14 -0500297static void clear_request_fn_mpio(struct multipath *m, union map_info *info)
Jun'ichi Nomura466891f2012-03-28 18:41:25 +0100298{
Mike Snitzer2eff1922016-02-03 09:13:14 -0500299 /* Only needed for non blk-mq (.request_fn) multipath */
Mike Snitzer8637a6b2016-01-31 12:08:36 -0500300 if (m->mpio_pool) {
301 struct dm_mpath_io *mpio = info->ptr;
Jun'ichi Nomura466891f2012-03-28 18:41:25 +0100302
Mike Snitzer8637a6b2016-01-31 12:08:36 -0500303 info->ptr = NULL;
304 mempool_free(mpio, m->mpio_pool);
305 }
Jun'ichi Nomura466891f2012-03-28 18:41:25 +0100306}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
Mike Snitzerbf661be2016-05-24 15:48:08 -0400308static size_t multipath_per_bio_data_size(void)
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400309{
Mike Snitzerbf661be2016-05-24 15:48:08 -0400310 return sizeof(struct dm_mpath_io) + sizeof(struct dm_bio_details);
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400311}
312
Mike Snitzerbf661be2016-05-24 15:48:08 -0400313static struct dm_mpath_io *get_mpio_from_bio(struct bio *bio)
314{
315 return dm_per_bio_data(bio, multipath_per_bio_data_size());
316}
317
318static struct dm_bio_details *get_bio_details_from_bio(struct bio *bio)
319{
320 /* dm_bio_details is immediately after the dm_mpath_io in bio's per-bio-data */
321 struct dm_mpath_io *mpio = get_mpio_from_bio(bio);
322 void *bio_details = mpio + 1;
323
324 return bio_details;
325}
326
327static void multipath_init_per_bio_data(struct bio *bio, struct dm_mpath_io **mpio_p,
328 struct dm_bio_details **bio_details_p)
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400329{
330 struct dm_mpath_io *mpio = get_mpio_from_bio(bio);
Mike Snitzerbf661be2016-05-24 15:48:08 -0400331 struct dm_bio_details *bio_details = get_bio_details_from_bio(bio);
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400332
333 memset(mpio, 0, sizeof(*mpio));
Mike Snitzerbf661be2016-05-24 15:48:08 -0400334 memset(bio_details, 0, sizeof(*bio_details));
335 dm_bio_record(bio_details, bio);
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400336
Mike Snitzerbf661be2016-05-24 15:48:08 -0400337 if (mpio_p)
338 *mpio_p = mpio;
339 if (bio_details_p)
340 *bio_details_p = bio_details;
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400341}
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343/*-----------------------------------------------
344 * Path selection
345 *-----------------------------------------------*/
346
Hannes Reinecke3e9f1be2014-02-28 15:33:45 +0100347static int __pg_init_all_paths(struct multipath *m)
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000348{
349 struct pgpath *pgpath;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000350 unsigned long pg_init_delay = 0;
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000351
Mike Snitzer91e968a2016-03-17 17:10:15 -0400352 if (atomic_read(&m->pg_init_in_progress) || test_bit(MPATHF_PG_INIT_DISABLED, &m->flags))
Hannes Reinecke3e9f1be2014-02-28 15:33:45 +0100353 return 0;
Hannes Reinecke17f4ff42014-02-28 15:33:42 +0100354
Mike Snitzer91e968a2016-03-17 17:10:15 -0400355 atomic_inc(&m->pg_init_count);
Mike Snitzer518257b2016-03-17 16:32:10 -0400356 clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
Hannes Reinecke3e9f1be2014-02-28 15:33:45 +0100357
358 /* Check here to reset pg_init_required */
359 if (!m->current_pg)
360 return 0;
361
Mike Snitzer518257b2016-03-17 16:32:10 -0400362 if (test_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags))
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000363 pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ?
364 m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS);
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000365 list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) {
366 /* Skip failed paths */
367 if (!pgpath->is_active)
368 continue;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000369 if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path,
370 pg_init_delay))
Mike Snitzer91e968a2016-03-17 17:10:15 -0400371 atomic_inc(&m->pg_init_in_progress);
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000372 }
Mike Snitzer91e968a2016-03-17 17:10:15 -0400373 return atomic_read(&m->pg_init_in_progress);
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000374}
375
Mike Snitzer2da16102016-03-17 18:38:17 -0400376static int pg_init_all_paths(struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
Mike Snitzer2da16102016-03-17 18:38:17 -0400378 int r;
379 unsigned long flags;
380
381 spin_lock_irqsave(&m->lock, flags);
382 r = __pg_init_all_paths(m);
383 spin_unlock_irqrestore(&m->lock, flags);
384
385 return r;
386}
387
388static void __switch_pg(struct multipath *m, struct priority_group *pg)
389{
390 m->current_pg = pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 /* Must we initialise the PG first, and queue I/O till it's ready? */
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700393 if (m->hw_handler_name) {
Mike Snitzer518257b2016-03-17 16:32:10 -0400394 set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
395 set_bit(MPATHF_QUEUE_IO, &m->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 } else {
Mike Snitzer518257b2016-03-17 16:32:10 -0400397 clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
398 clear_bit(MPATHF_QUEUE_IO, &m->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 }
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100400
Mike Snitzer91e968a2016-03-17 17:10:15 -0400401 atomic_set(&m->pg_init_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
Mike Snitzer2da16102016-03-17 18:38:17 -0400404static struct pgpath *choose_path_in_pg(struct multipath *m,
405 struct priority_group *pg,
406 size_t nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
Mike Snitzer2da16102016-03-17 18:38:17 -0400408 unsigned long flags;
Josef "Jeff" Sipekc922d5f2006-12-08 02:36:33 -0800409 struct dm_path *path;
Mike Snitzer2da16102016-03-17 18:38:17 -0400410 struct pgpath *pgpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Mike Snitzer90a43232016-02-17 21:29:17 -0500412 path = pg->ps.type->select_path(&pg->ps, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 if (!path)
Mike Snitzer2da16102016-03-17 18:38:17 -0400414 return ERR_PTR(-ENXIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
Mike Snitzer2da16102016-03-17 18:38:17 -0400416 pgpath = path_to_pgpath(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
Mike Snitzer2da16102016-03-17 18:38:17 -0400418 if (unlikely(lockless_dereference(m->current_pg) != pg)) {
419 /* Only update current_pgpath if pg changed */
420 spin_lock_irqsave(&m->lock, flags);
421 m->current_pgpath = pgpath;
422 __switch_pg(m, pg);
423 spin_unlock_irqrestore(&m->lock, flags);
424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Mike Snitzer2da16102016-03-17 18:38:17 -0400426 return pgpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427}
428
Mike Snitzer2da16102016-03-17 18:38:17 -0400429static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
Mike Snitzer2da16102016-03-17 18:38:17 -0400431 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 struct priority_group *pg;
Mike Snitzer2da16102016-03-17 18:38:17 -0400433 struct pgpath *pgpath;
Mike Snitzer63d32e82017-01-06 15:33:14 -0500434 unsigned bypassed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Mike Snitzer91e968a2016-03-17 17:10:15 -0400436 if (!atomic_read(&m->nr_valid_paths)) {
Mike Snitzer518257b2016-03-17 16:32:10 -0400437 clear_bit(MPATHF_QUEUE_IO, &m->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 goto failed;
Benjamin Marzinski1f271972014-08-13 13:53:42 -0500439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
441 /* Were we instructed to switch PG? */
Mike Snitzer2da16102016-03-17 18:38:17 -0400442 if (lockless_dereference(m->next_pg)) {
443 spin_lock_irqsave(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 pg = m->next_pg;
Mike Snitzer2da16102016-03-17 18:38:17 -0400445 if (!pg) {
446 spin_unlock_irqrestore(&m->lock, flags);
447 goto check_current_pg;
448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 m->next_pg = NULL;
Mike Snitzer2da16102016-03-17 18:38:17 -0400450 spin_unlock_irqrestore(&m->lock, flags);
451 pgpath = choose_path_in_pg(m, pg, nr_bytes);
452 if (!IS_ERR_OR_NULL(pgpath))
453 return pgpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
455
456 /* Don't change PG until it has no remaining paths */
Mike Snitzer2da16102016-03-17 18:38:17 -0400457check_current_pg:
458 pg = lockless_dereference(m->current_pg);
459 if (pg) {
460 pgpath = choose_path_in_pg(m, pg, nr_bytes);
461 if (!IS_ERR_OR_NULL(pgpath))
462 return pgpath;
463 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 /*
466 * Loop through priority groups until we find a valid path.
467 * First time we skip PGs marked 'bypassed'.
Mike Christief220fd42012-06-03 00:29:45 +0100468 * Second time we only try the ones we skipped, but set
469 * pg_init_delay_retry so we do not hammer controllers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 */
471 do {
472 list_for_each_entry(pg, &m->priority_groups, list) {
Mike Snitzer63d32e82017-01-06 15:33:14 -0500473 if (pg->bypassed == !!bypassed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 continue;
Mike Snitzer2da16102016-03-17 18:38:17 -0400475 pgpath = choose_path_in_pg(m, pg, nr_bytes);
476 if (!IS_ERR_OR_NULL(pgpath)) {
Mike Christief220fd42012-06-03 00:29:45 +0100477 if (!bypassed)
Mike Snitzer518257b2016-03-17 16:32:10 -0400478 set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags);
Mike Snitzer2da16102016-03-17 18:38:17 -0400479 return pgpath;
Mike Christief220fd42012-06-03 00:29:45 +0100480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 }
482 } while (bypassed--);
483
484failed:
Mike Snitzer2da16102016-03-17 18:38:17 -0400485 spin_lock_irqsave(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 m->current_pgpath = NULL;
487 m->current_pg = NULL;
Mike Snitzer2da16102016-03-17 18:38:17 -0400488 spin_unlock_irqrestore(&m->lock, flags);
489
490 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491}
492
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800493/*
494 * Check whether bios must be queued in the device-mapper core rather
495 * than here in the target.
496 *
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800497 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
498 * same value then we are not between multipath_presuspend()
499 * and multipath_resume() calls and we have no need to check
500 * for the DMF_NOFLUSH_SUSPENDING flag.
501 */
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400502static bool __must_push_back(struct multipath *m)
503{
504 return ((test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) !=
505 test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)) &&
506 dm_noflush_suspending(m->ti));
507}
508
509static bool must_push_back_rq(struct multipath *m)
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800510{
Mike Snitzer1814f2e2016-07-25 21:08:51 -0400511 bool r;
512 unsigned long flags;
513
514 spin_lock_irqsave(&m->lock, flags);
515 r = (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) ||
516 __must_push_back(m));
517 spin_unlock_irqrestore(&m->lock, flags);
518
519 return r;
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400520}
521
522static bool must_push_back_bio(struct multipath *m)
523{
Mike Snitzer1814f2e2016-07-25 21:08:51 -0400524 bool r;
525 unsigned long flags;
526
527 spin_lock_irqsave(&m->lock, flags);
528 r = __must_push_back(m);
529 spin_unlock_irqrestore(&m->lock, flags);
530
531 return r;
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800532}
533
Hannes Reinecke36fcffc2014-02-28 15:33:47 +0100534/*
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400535 * Map cloned requests (request-based multipath)
Hannes Reinecke36fcffc2014-02-28 15:33:47 +0100536 */
Mike Snitzere5863d92014-12-17 21:08:12 -0500537static int __multipath_map(struct dm_target *ti, struct request *clone,
538 union map_info *map_context,
539 struct request *rq, struct request **__clone)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
Mike Snitzer7943bd62016-02-02 21:53:15 -0500541 struct multipath *m = ti->private;
Hannes Reineckee3bde042014-02-28 15:33:46 +0100542 int r = DM_MAPIO_REQUEUE;
Mike Snitzere5863d92014-12-17 21:08:12 -0500543 size_t nr_bytes = clone ? blk_rq_bytes(clone) : blk_rq_bytes(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 struct pgpath *pgpath;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100545 struct block_device *bdev;
Hannes Reineckee3bde042014-02-28 15:33:46 +0100546 struct dm_mpath_io *mpio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 /* Do we need to select a new pgpath? */
Mike Snitzer2da16102016-03-17 18:38:17 -0400549 pgpath = lockless_dereference(m->current_pgpath);
550 if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags))
551 pgpath = choose_pgpath(m, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Mike Snitzer9bf59a62014-02-28 15:33:48 +0100553 if (!pgpath) {
Mike Snitzerb88efd42016-09-09 19:26:19 -0400554 if (must_push_back_rq(m))
555 return DM_MAPIO_DELAY_REQUEUE;
556 return -EIO; /* Failed */
Mike Snitzer518257b2016-03-17 16:32:10 -0400557 } else if (test_bit(MPATHF_QUEUE_IO, &m->flags) ||
558 test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) {
Mike Snitzer2da16102016-03-17 18:38:17 -0400559 pg_init_all_paths(m);
560 return r;
Mike Snitzer9bf59a62014-02-28 15:33:48 +0100561 }
Mike Snitzer6afbc012014-07-08 11:55:09 -0400562
Mike Snitzer2eff1922016-02-03 09:13:14 -0500563 mpio = set_mpio(m, map_context);
564 if (!mpio)
Mike Snitzer9bf59a62014-02-28 15:33:48 +0100565 /* ENOMEM, requeue */
Mike Snitzer2da16102016-03-17 18:38:17 -0400566 return r;
Mike Snitzer9bf59a62014-02-28 15:33:48 +0100567
Mike Snitzer9bf59a62014-02-28 15:33:48 +0100568 mpio->pgpath = pgpath;
569 mpio->nr_bytes = nr_bytes;
Keith Busch2eb6e1e2014-10-17 17:46:36 -0600570
571 bdev = pgpath->path.dev->bdev;
572
Mike Snitzere5863d92014-12-17 21:08:12 -0500573 if (clone) {
Mike Snitzerc5248f72016-02-20 14:02:49 -0500574 /*
575 * Old request-based interface: allocated clone is passed in.
576 * Used by: .request_fn stacked on .request_fn path(s).
577 */
Mike Snitzere5863d92014-12-17 21:08:12 -0500578 clone->q = bdev_get_queue(bdev);
579 clone->rq_disk = bdev->bd_disk;
580 clone->cmd_flags |= REQ_FAILFAST_TRANSPORT;
581 } else {
Mike Snitzereca7ee62016-02-20 13:45:38 -0500582 /*
583 * blk-mq request-based interface; used by both:
584 * .request_fn stacked on blk-mq path(s) and
585 * blk-mq stacked on blk-mq path(s).
586 */
Mike Snitzer78ce23b2016-01-31 17:38:28 -0500587 *__clone = blk_mq_alloc_request(bdev_get_queue(bdev),
588 rq_data_dir(rq), BLK_MQ_REQ_NOWAIT);
Mike Snitzer4c6dd532015-05-27 15:23:56 -0400589 if (IS_ERR(*__clone)) {
Mike Snitzere5863d92014-12-17 21:08:12 -0500590 /* ENOMEM, requeue */
Mike Snitzer2eff1922016-02-03 09:13:14 -0500591 clear_request_fn_mpio(m, map_context);
Mike Snitzere5863d92014-12-17 21:08:12 -0500592 return r;
Mike Snitzer4c6dd532015-05-27 15:23:56 -0400593 }
Mike Snitzere5863d92014-12-17 21:08:12 -0500594 (*__clone)->bio = (*__clone)->biotail = NULL;
595 (*__clone)->rq_disk = bdev->bd_disk;
596 (*__clone)->cmd_flags |= REQ_FAILFAST_TRANSPORT;
597 }
598
Mike Snitzer9bf59a62014-02-28 15:33:48 +0100599 if (pgpath->pg->ps.type->start_io)
600 pgpath->pg->ps.type->start_io(&pgpath->pg->ps,
601 &pgpath->path,
602 nr_bytes);
Keith Busch2eb6e1e2014-10-17 17:46:36 -0600603 return DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
Mike Snitzere5863d92014-12-17 21:08:12 -0500606static int multipath_map(struct dm_target *ti, struct request *clone,
607 union map_info *map_context)
608{
609 return __multipath_map(ti, clone, map_context, NULL, NULL);
610}
611
612static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
613 union map_info *map_context,
614 struct request **clone)
615{
616 return __multipath_map(ti, NULL, map_context, rq, clone);
617}
618
619static void multipath_release_clone(struct request *clone)
620{
Mike Snitzer78ce23b2016-01-31 17:38:28 -0500621 blk_mq_free_request(clone);
Mike Snitzere5863d92014-12-17 21:08:12 -0500622}
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624/*
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400625 * Map cloned bios (bio-based multipath)
626 */
627static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_mpath_io *mpio)
628{
629 size_t nr_bytes = bio->bi_iter.bi_size;
630 struct pgpath *pgpath;
631 unsigned long flags;
632 bool queue_io;
633
634 /* Do we need to select a new pgpath? */
635 pgpath = lockless_dereference(m->current_pgpath);
636 queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags);
637 if (!pgpath || !queue_io)
638 pgpath = choose_pgpath(m, nr_bytes);
639
640 if ((pgpath && queue_io) ||
641 (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) {
642 /* Queue for the daemon to resubmit */
643 spin_lock_irqsave(&m->lock, flags);
644 bio_list_add(&m->queued_bios, bio);
645 spin_unlock_irqrestore(&m->lock, flags);
646 /* PG_INIT_REQUIRED cannot be set without QUEUE_IO */
647 if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags))
648 pg_init_all_paths(m);
649 else if (!queue_io)
650 queue_work(kmultipathd, &m->process_queued_bios);
651 return DM_MAPIO_SUBMITTED;
652 }
653
654 if (!pgpath) {
655 if (!must_push_back_bio(m))
656 return -EIO;
657 return DM_MAPIO_REQUEUE;
658 }
659
660 mpio->pgpath = pgpath;
661 mpio->nr_bytes = nr_bytes;
662
663 bio->bi_error = 0;
664 bio->bi_bdev = pgpath->path.dev->bdev;
Jens Axboe1eff9d32016-08-05 15:35:16 -0600665 bio->bi_opf |= REQ_FAILFAST_TRANSPORT;
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400666
667 if (pgpath->pg->ps.type->start_io)
668 pgpath->pg->ps.type->start_io(&pgpath->pg->ps,
669 &pgpath->path,
670 nr_bytes);
671 return DM_MAPIO_REMAPPED;
672}
673
674static int multipath_map_bio(struct dm_target *ti, struct bio *bio)
675{
676 struct multipath *m = ti->private;
Mike Snitzerbf661be2016-05-24 15:48:08 -0400677 struct dm_mpath_io *mpio = NULL;
678
679 multipath_init_per_bio_data(bio, &mpio, NULL);
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400680
681 return __multipath_map_bio(m, bio, mpio);
682}
683
Mike Snitzer7e48c762016-09-14 10:47:03 -0400684static void process_queued_io_list(struct multipath *m)
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400685{
Mike Snitzer7e48c762016-09-14 10:47:03 -0400686 if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED)
687 dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table));
688 else if (m->queue_mode == DM_TYPE_BIO_BASED)
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400689 queue_work(kmultipathd, &m->process_queued_bios);
690}
691
692static void process_queued_bios(struct work_struct *work)
693{
694 int r;
695 unsigned long flags;
696 struct bio *bio;
697 struct bio_list bios;
698 struct blk_plug plug;
699 struct multipath *m =
700 container_of(work, struct multipath, process_queued_bios);
701
702 bio_list_init(&bios);
703
704 spin_lock_irqsave(&m->lock, flags);
705
706 if (bio_list_empty(&m->queued_bios)) {
707 spin_unlock_irqrestore(&m->lock, flags);
708 return;
709 }
710
711 bio_list_merge(&bios, &m->queued_bios);
712 bio_list_init(&m->queued_bios);
713
714 spin_unlock_irqrestore(&m->lock, flags);
715
716 blk_start_plug(&plug);
717 while ((bio = bio_list_pop(&bios))) {
718 r = __multipath_map_bio(m, bio, get_mpio_from_bio(bio));
719 if (r < 0 || r == DM_MAPIO_REQUEUE) {
720 bio->bi_error = r;
721 bio_endio(bio);
722 } else if (r == DM_MAPIO_REMAPPED)
723 generic_make_request(bio);
724 }
725 blk_finish_plug(&plug);
726}
727
728/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 * If we run out of usable paths, should we queue I/O or error it?
730 */
Mike Snitzerbe7d31c2016-02-10 13:02:21 -0500731static int queue_if_no_path(struct multipath *m, bool queue_if_no_path,
732 bool save_old_value)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
734 unsigned long flags;
735
736 spin_lock_irqsave(&m->lock, flags);
737
Mike Snitzer518257b2016-03-17 16:32:10 -0400738 if (save_old_value) {
739 if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
740 set_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags);
741 else
742 clear_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags);
743 } else {
744 if (queue_if_no_path)
745 set_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags);
746 else
747 clear_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags);
748 }
749 if (queue_if_no_path)
750 set_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags);
Alasdair G Kergon485ef692005-09-27 21:45:45 -0700751 else
Mike Snitzer518257b2016-03-17 16:32:10 -0400752 clear_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags);
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 spin_unlock_irqrestore(&m->lock, flags);
755
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400756 if (!queue_if_no_path) {
Hannes Reinecke63d832c2014-05-26 14:45:39 +0200757 dm_table_run_md_queue_async(m->ti->table);
Mike Snitzer7e48c762016-09-14 10:47:03 -0400758 process_queued_io_list(m);
Mike Snitzer76e33fe2016-05-19 16:15:14 -0400759 }
Hannes Reinecke63d832c2014-05-26 14:45:39 +0200760
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 return 0;
762}
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764/*
765 * An event is triggered whenever a path is taken out of use.
766 * Includes path failure and PG bypass.
767 */
David Howellsc4028952006-11-22 14:57:56 +0000768static void trigger_event(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
David Howellsc4028952006-11-22 14:57:56 +0000770 struct multipath *m =
771 container_of(work, struct multipath, trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
773 dm_table_event(m->ti->table);
774}
775
776/*-----------------------------------------------------------------
777 * Constructor/argument parsing:
778 * <#multipath feature args> [<arg>]*
779 * <#hw_handler args> [hw_handler [<arg>]*]
780 * <#priority groups>
781 * <initial priority group>
782 * [<selector> <#selector args> [<arg>]*
783 * <#paths> <#per-path selector args>
784 * [<path> [<arg>]* ]+ ]+
785 *---------------------------------------------------------------*/
Mike Snitzer498f0102011-08-02 12:32:04 +0100786static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 struct dm_target *ti)
788{
789 int r;
790 struct path_selector_type *pst;
791 unsigned ps_argc;
792
Mike Snitzer498f0102011-08-02 12:32:04 +0100793 static struct dm_arg _args[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700794 {0, 1024, "invalid number of path selector args"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 };
796
Mike Snitzer498f0102011-08-02 12:32:04 +0100797 pst = dm_get_path_selector(dm_shift_arg(as));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 if (!pst) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700799 ti->error = "unknown path selector type";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 return -EINVAL;
801 }
802
Mike Snitzer498f0102011-08-02 12:32:04 +0100803 r = dm_read_arg_group(_args, as, &ps_argc, &ti->error);
Mikulas Patocka371b2e32008-07-21 12:00:24 +0100804 if (r) {
805 dm_put_path_selector(pst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return -EINVAL;
Mikulas Patocka371b2e32008-07-21 12:00:24 +0100807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
809 r = pst->create(&pg->ps, ps_argc, as->argv);
810 if (r) {
811 dm_put_path_selector(pst);
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700812 ti->error = "path selector constructor failed";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return r;
814 }
815
816 pg->ps.type = pst;
Mike Snitzer498f0102011-08-02 12:32:04 +0100817 dm_consume_args(as, ps_argc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
819 return 0;
820}
821
Mike Snitzer498f0102011-08-02 12:32:04 +0100822static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 struct dm_target *ti)
824{
825 int r;
826 struct pgpath *p;
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700827 struct multipath *m = ti->private;
Mike Snitzera58a9352012-07-27 15:08:04 +0100828 struct request_queue *q = NULL;
829 const char *attached_handler_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831 /* we need at least a path arg */
832 if (as->argc < 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700833 ti->error = "no device given";
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100834 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 }
836
837 p = alloc_pgpath();
838 if (!p)
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100839 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Mike Snitzer498f0102011-08-02 12:32:04 +0100841 r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table),
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +0000842 &p->path.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 if (r) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700844 ti->error = "error getting device";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 goto bad;
846 }
847
Mike Snitzer518257b2016-03-17 16:32:10 -0400848 if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) || m->hw_handler_name)
Mike Snitzera58a9352012-07-27 15:08:04 +0100849 q = bdev_get_queue(p->path.dev->bdev);
Hannes Reineckea0cf7ea2009-06-22 10:12:11 +0100850
Mike Snitzer518257b2016-03-17 16:32:10 -0400851 if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) {
Christoph Hellwig1bab0de2015-08-27 14:16:54 +0200852retain:
Mike Snitzera58a9352012-07-27 15:08:04 +0100853 attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL);
854 if (attached_handler_name) {
855 /*
856 * Reset hw_handler_name to match the attached handler
857 * and clear any hw_handler_params associated with the
858 * ignored handler.
859 *
860 * NB. This modifies the table line to show the actual
861 * handler instead of the original table passed in.
862 */
863 kfree(m->hw_handler_name);
864 m->hw_handler_name = attached_handler_name;
865
866 kfree(m->hw_handler_params);
867 m->hw_handler_params = NULL;
868 }
869 }
870
871 if (m->hw_handler_name) {
Hannes Reineckea0cf7ea2009-06-22 10:12:11 +0100872 r = scsi_dh_attach(q, m->hw_handler_name);
873 if (r == -EBUSY) {
Christoph Hellwig1bab0de2015-08-27 14:16:54 +0200874 char b[BDEVNAME_SIZE];
Hannes Reineckea0cf7ea2009-06-22 10:12:11 +0100875
Christoph Hellwig1bab0de2015-08-27 14:16:54 +0200876 printk(KERN_INFO "dm-mpath: retaining handler on device %s\n",
877 bdevname(p->path.dev->bdev, b));
878 goto retain;
879 }
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700880 if (r < 0) {
Hannes Reineckea0cf7ea2009-06-22 10:12:11 +0100881 ti->error = "error attaching hardware handler";
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700882 dm_put_device(ti, p->path.dev);
883 goto bad;
884 }
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700885
886 if (m->hw_handler_params) {
887 r = scsi_dh_set_params(q, m->hw_handler_params);
888 if (r < 0) {
889 ti->error = "unable to set hardware "
890 "handler parameters";
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700891 dm_put_device(ti, p->path.dev);
892 goto bad;
893 }
894 }
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700895 }
896
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error);
898 if (r) {
899 dm_put_device(ti, p->path.dev);
900 goto bad;
901 }
902
903 return p;
904
905 bad:
906 free_pgpath(p);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100907 return ERR_PTR(r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908}
909
Mike Snitzer498f0102011-08-02 12:32:04 +0100910static struct priority_group *parse_priority_group(struct dm_arg_set *as,
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700911 struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912{
Mike Snitzer498f0102011-08-02 12:32:04 +0100913 static struct dm_arg _args[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700914 {1, 1024, "invalid number of paths"},
915 {0, 1024, "invalid number of selector args"}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 };
917
918 int r;
Mike Snitzer498f0102011-08-02 12:32:04 +0100919 unsigned i, nr_selector_args, nr_args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 struct priority_group *pg;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700921 struct dm_target *ti = m->ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 if (as->argc < 2) {
924 as->argc = 0;
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100925 ti->error = "not enough priority group arguments";
926 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 }
928
929 pg = alloc_priority_group();
930 if (!pg) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700931 ti->error = "couldn't allocate priority group";
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100932 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 }
934 pg->m = m;
935
936 r = parse_path_selector(as, pg, ti);
937 if (r)
938 goto bad;
939
940 /*
941 * read the paths
942 */
Mike Snitzer498f0102011-08-02 12:32:04 +0100943 r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (r)
945 goto bad;
946
Mike Snitzer498f0102011-08-02 12:32:04 +0100947 r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (r)
949 goto bad;
950
Mike Snitzer498f0102011-08-02 12:32:04 +0100951 nr_args = 1 + nr_selector_args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 for (i = 0; i < pg->nr_pgpaths; i++) {
953 struct pgpath *pgpath;
Mike Snitzer498f0102011-08-02 12:32:04 +0100954 struct dm_arg_set path_args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Mike Snitzer498f0102011-08-02 12:32:04 +0100956 if (as->argc < nr_args) {
Mikulas Patocka148acff2008-07-21 12:00:30 +0100957 ti->error = "not enough path parameters";
Alasdair G Kergon6bbf79a2010-08-12 04:13:49 +0100958 r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 goto bad;
Mikulas Patocka148acff2008-07-21 12:00:30 +0100960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Mike Snitzer498f0102011-08-02 12:32:04 +0100962 path_args.argc = nr_args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 path_args.argv = as->argv;
964
965 pgpath = parse_path(&path_args, &pg->ps, ti);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100966 if (IS_ERR(pgpath)) {
967 r = PTR_ERR(pgpath);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 goto bad;
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100969 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 pgpath->pg = pg;
972 list_add_tail(&pgpath->list, &pg->pgpaths);
Mike Snitzer498f0102011-08-02 12:32:04 +0100973 dm_consume_args(as, nr_args);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 }
975
976 return pg;
977
978 bad:
979 free_priority_group(pg, ti);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100980 return ERR_PTR(r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981}
982
Mike Snitzer498f0102011-08-02 12:32:04 +0100983static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 unsigned hw_argc;
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700986 int ret;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700987 struct dm_target *ti = m->ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Mike Snitzer498f0102011-08-02 12:32:04 +0100989 static struct dm_arg _args[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700990 {0, 1024, "invalid number of hardware handler args"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 };
992
Mike Snitzer498f0102011-08-02 12:32:04 +0100993 if (dm_read_arg_group(_args, as, &hw_argc, &ti->error))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 return -EINVAL;
995
996 if (!hw_argc)
997 return 0;
998
Mike Snitzere83068a2016-05-24 21:16:51 -0400999 if (m->queue_mode == DM_TYPE_BIO_BASED) {
Mike Snitzer76e33fe2016-05-19 16:15:14 -04001000 dm_consume_args(as, hw_argc);
1001 DMERR("bio-based multipath doesn't allow hardware handler args");
1002 return 0;
1003 }
1004
Mike Snitzer498f0102011-08-02 12:32:04 +01001005 m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL);
Chandra Seetharaman14e98c52008-11-13 23:39:06 +00001006
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -07001007 if (hw_argc > 1) {
1008 char *p;
1009 int i, j, len = 4;
1010
1011 for (i = 0; i <= hw_argc - 2; i++)
1012 len += strlen(as->argv[i]) + 1;
1013 p = m->hw_handler_params = kzalloc(len, GFP_KERNEL);
1014 if (!p) {
1015 ti->error = "memory allocation failed";
1016 ret = -ENOMEM;
1017 goto fail;
1018 }
1019 j = sprintf(p, "%d", hw_argc - 1);
1020 for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1)
1021 j = sprintf(p, "%s", as->argv[i]);
1022 }
Mike Snitzer498f0102011-08-02 12:32:04 +01001023 dm_consume_args(as, hw_argc - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025 return 0;
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -07001026fail:
1027 kfree(m->hw_handler_name);
1028 m->hw_handler_name = NULL;
1029 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030}
1031
Mike Snitzer498f0102011-08-02 12:32:04 +01001032static int parse_features(struct dm_arg_set *as, struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
1034 int r;
1035 unsigned argc;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -07001036 struct dm_target *ti = m->ti;
Mike Snitzer498f0102011-08-02 12:32:04 +01001037 const char *arg_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Mike Snitzer498f0102011-08-02 12:32:04 +01001039 static struct dm_arg _args[] = {
Mike Snitzere83068a2016-05-24 21:16:51 -04001040 {0, 8, "invalid number of feature args"},
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001041 {1, 50, "pg_init_retries must be between 1 and 50"},
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001042 {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 };
1044
Mike Snitzer498f0102011-08-02 12:32:04 +01001045 r = dm_read_arg_group(_args, as, &argc, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 if (r)
1047 return -EINVAL;
1048
1049 if (!argc)
1050 return 0;
1051
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001052 do {
Mike Snitzer498f0102011-08-02 12:32:04 +01001053 arg_name = dm_shift_arg(as);
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001054 argc--;
1055
Mike Snitzer498f0102011-08-02 12:32:04 +01001056 if (!strcasecmp(arg_name, "queue_if_no_path")) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001057 r = queue_if_no_path(m, true, false);
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001058 continue;
1059 }
1060
Mike Snitzera58a9352012-07-27 15:08:04 +01001061 if (!strcasecmp(arg_name, "retain_attached_hw_handler")) {
Mike Snitzer518257b2016-03-17 16:32:10 -04001062 set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags);
Mike Snitzera58a9352012-07-27 15:08:04 +01001063 continue;
1064 }
1065
Mike Snitzer498f0102011-08-02 12:32:04 +01001066 if (!strcasecmp(arg_name, "pg_init_retries") &&
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001067 (argc >= 1)) {
Mike Snitzer498f0102011-08-02 12:32:04 +01001068 r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error);
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001069 argc--;
1070 continue;
1071 }
1072
Mike Snitzer498f0102011-08-02 12:32:04 +01001073 if (!strcasecmp(arg_name, "pg_init_delay_msecs") &&
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001074 (argc >= 1)) {
Mike Snitzer498f0102011-08-02 12:32:04 +01001075 r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error);
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001076 argc--;
1077 continue;
1078 }
1079
Mike Snitzere83068a2016-05-24 21:16:51 -04001080 if (!strcasecmp(arg_name, "queue_mode") &&
1081 (argc >= 1)) {
1082 const char *queue_mode_name = dm_shift_arg(as);
1083
1084 if (!strcasecmp(queue_mode_name, "bio"))
1085 m->queue_mode = DM_TYPE_BIO_BASED;
1086 else if (!strcasecmp(queue_mode_name, "rq"))
1087 m->queue_mode = DM_TYPE_REQUEST_BASED;
1088 else if (!strcasecmp(queue_mode_name, "mq"))
1089 m->queue_mode = DM_TYPE_MQ_REQUEST_BASED;
1090 else {
1091 ti->error = "Unknown 'queue_mode' requested";
1092 r = -EINVAL;
1093 }
1094 argc--;
1095 continue;
1096 }
1097
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 ti->error = "Unrecognised multipath feature request";
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001099 r = -EINVAL;
1100 } while (argc && !r);
1101
1102 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103}
1104
Mike Snitzere83068a2016-05-24 21:16:51 -04001105static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
Mike Snitzer498f0102011-08-02 12:32:04 +01001107 /* target arguments */
1108 static struct dm_arg _args[] = {
Mike Snitzera490a072011-03-24 13:54:33 +00001109 {0, 1024, "invalid number of priority groups"},
1110 {0, 1024, "invalid initial priority group number"},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 };
1112
1113 int r;
1114 struct multipath *m;
Mike Snitzer498f0102011-08-02 12:32:04 +01001115 struct dm_arg_set as;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 unsigned pg_count = 0;
1117 unsigned next_pg_num;
1118
1119 as.argc = argc;
1120 as.argv = argv;
1121
Mike Snitzere83068a2016-05-24 21:16:51 -04001122 m = alloc_multipath(ti);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 if (!m) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001124 ti->error = "can't allocate multipath";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return -EINVAL;
1126 }
1127
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -07001128 r = parse_features(&as, m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 if (r)
1130 goto bad;
1131
Mike Snitzere83068a2016-05-24 21:16:51 -04001132 r = alloc_multipath_stage2(ti, m);
1133 if (r)
1134 goto bad;
1135
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -07001136 r = parse_hw_handler(&as, m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 if (r)
1138 goto bad;
1139
Mike Snitzer498f0102011-08-02 12:32:04 +01001140 r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 if (r)
1142 goto bad;
1143
Mike Snitzer498f0102011-08-02 12:32:04 +01001144 r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 if (r)
1146 goto bad;
1147
Mike Snitzera490a072011-03-24 13:54:33 +00001148 if ((!m->nr_priority_groups && next_pg_num) ||
1149 (m->nr_priority_groups && !next_pg_num)) {
1150 ti->error = "invalid initial priority group";
1151 r = -EINVAL;
1152 goto bad;
1153 }
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 /* parse the priority groups */
1156 while (as.argc) {
1157 struct priority_group *pg;
Mike Snitzer91e968a2016-03-17 17:10:15 -04001158 unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -07001160 pg = parse_priority_group(&as, m);
Benjamin Marzinski01460f32008-10-10 13:36:57 +01001161 if (IS_ERR(pg)) {
1162 r = PTR_ERR(pg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 goto bad;
1164 }
1165
Mike Snitzer91e968a2016-03-17 17:10:15 -04001166 nr_valid_paths += pg->nr_pgpaths;
1167 atomic_set(&m->nr_valid_paths, nr_valid_paths);
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 list_add_tail(&pg->list, &m->priority_groups);
1170 pg_count++;
1171 pg->pg_num = pg_count;
1172 if (!--next_pg_num)
1173 m->next_pg = pg;
1174 }
1175
1176 if (pg_count != m->nr_priority_groups) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001177 ti->error = "priority group count mismatch";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 r = -EINVAL;
1179 goto bad;
1180 }
1181
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001182 ti->num_flush_bios = 1;
1183 ti->num_discard_bios = 1;
Mike Snitzer042bcef2013-05-10 14:37:16 +01001184 ti->num_write_same_bios = 1;
Mike Snitzere83068a2016-05-24 21:16:51 -04001185 if (m->queue_mode == DM_TYPE_BIO_BASED)
Mike Snitzerbf661be2016-05-24 15:48:08 -04001186 ti->per_io_data_size = multipath_per_bio_data_size();
Mike Snitzere83068a2016-05-24 21:16:51 -04001187 else if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED)
Mike Snitzer8637a6b2016-01-31 12:08:36 -05001188 ti->per_io_data_size = sizeof(struct dm_mpath_io);
Mikulas Patocka86279212009-06-22 10:12:24 +01001189
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 return 0;
1191
1192 bad:
1193 free_multipath(m);
1194 return r;
1195}
1196
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001197static void multipath_wait_for_pg_init_completion(struct multipath *m)
1198{
Bart Van Assche9f4c3f82016-08-31 15:16:43 -07001199 DEFINE_WAIT(wait);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001200
1201 while (1) {
Bart Van Assche9f4c3f82016-08-31 15:16:43 -07001202 prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001203
Mike Snitzer91e968a2016-03-17 17:10:15 -04001204 if (!atomic_read(&m->pg_init_in_progress))
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001205 break;
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001206
1207 io_schedule();
1208 }
Bart Van Assche9f4c3f82016-08-31 15:16:43 -07001209 finish_wait(&m->pg_init_wait, &wait);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001210}
1211
1212static void flush_multipath_work(struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
Mike Snitzer518257b2016-03-17 16:32:10 -04001214 set_bit(MPATHF_PG_INIT_DISABLED, &m->flags);
1215 smp_mb__after_atomic();
Shiva Krishna Merla954a73d2013-10-30 03:26:38 +00001216
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001217 flush_workqueue(kmpath_handlerd);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001218 multipath_wait_for_pg_init_completion(m);
Alasdair G Kergona044d012005-07-12 15:53:02 -07001219 flush_workqueue(kmultipathd);
Tejun Heo43829732012-08-20 14:51:24 -07001220 flush_work(&m->trigger_event);
Shiva Krishna Merla954a73d2013-10-30 03:26:38 +00001221
Mike Snitzer518257b2016-03-17 16:32:10 -04001222 clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags);
1223 smp_mb__after_atomic();
Kiyoshi Ueda6df400a2009-12-10 23:52:19 +00001224}
1225
1226static void multipath_dtr(struct dm_target *ti)
1227{
1228 struct multipath *m = ti->private;
1229
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001230 flush_multipath_work(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 free_multipath(m);
1232}
1233
1234/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 * Take a path out of use.
1236 */
1237static int fail_path(struct pgpath *pgpath)
1238{
1239 unsigned long flags;
1240 struct multipath *m = pgpath->pg->m;
1241
1242 spin_lock_irqsave(&m->lock, flags);
1243
Kiyoshi Ueda66800732008-10-10 13:36:58 +01001244 if (!pgpath->is_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 goto out;
1246
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001247 DMWARN("Failing path %s.", pgpath->path.dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249 pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path);
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001250 pgpath->is_active = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 pgpath->fail_count++;
1252
Mike Snitzer91e968a2016-03-17 17:10:15 -04001253 atomic_dec(&m->nr_valid_paths);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
1255 if (pgpath == m->current_pgpath)
1256 m->current_pgpath = NULL;
1257
Mike Andersonb15546f2007-10-19 22:48:02 +01001258 dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti,
Mike Snitzer91e968a2016-03-17 17:10:15 -04001259 pgpath->path.dev->name, atomic_read(&m->nr_valid_paths));
Mike Andersonb15546f2007-10-19 22:48:02 +01001260
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001261 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
1263out:
1264 spin_unlock_irqrestore(&m->lock, flags);
1265
1266 return 0;
1267}
1268
1269/*
1270 * Reinstate a previously-failed path
1271 */
1272static int reinstate_path(struct pgpath *pgpath)
1273{
Hannes Reinecke63d832c2014-05-26 14:45:39 +02001274 int r = 0, run_queue = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 unsigned long flags;
1276 struct multipath *m = pgpath->pg->m;
Mike Snitzer91e968a2016-03-17 17:10:15 -04001277 unsigned nr_valid_paths;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
1279 spin_lock_irqsave(&m->lock, flags);
1280
Kiyoshi Ueda66800732008-10-10 13:36:58 +01001281 if (pgpath->is_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 goto out;
1283
Mike Snitzerec31f3f2016-02-20 12:49:43 -05001284 DMWARN("Reinstating path %s.", pgpath->path.dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
1286 r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path);
1287 if (r)
1288 goto out;
1289
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001290 pgpath->is_active = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
Mike Snitzer91e968a2016-03-17 17:10:15 -04001292 nr_valid_paths = atomic_inc_return(&m->nr_valid_paths);
1293 if (nr_valid_paths == 1) {
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001294 m->current_pgpath = NULL;
Hannes Reinecke63d832c2014-05-26 14:45:39 +02001295 run_queue = 1;
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001296 } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) {
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001297 if (queue_work(kmpath_handlerd, &pgpath->activate_path.work))
Mike Snitzer91e968a2016-03-17 17:10:15 -04001298 atomic_inc(&m->pg_init_in_progress);
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Mike Andersonb15546f2007-10-19 22:48:02 +01001301 dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti,
Mike Snitzer91e968a2016-03-17 17:10:15 -04001302 pgpath->path.dev->name, nr_valid_paths);
Mike Andersonb15546f2007-10-19 22:48:02 +01001303
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001304 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
1306out:
1307 spin_unlock_irqrestore(&m->lock, flags);
Mike Snitzer76e33fe2016-05-19 16:15:14 -04001308 if (run_queue) {
Hannes Reinecke63d832c2014-05-26 14:45:39 +02001309 dm_table_run_md_queue_async(m->ti->table);
Mike Snitzer7e48c762016-09-14 10:47:03 -04001310 process_queued_io_list(m);
Mike Snitzer76e33fe2016-05-19 16:15:14 -04001311 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
1313 return r;
1314}
1315
1316/*
1317 * Fail or reinstate all paths that match the provided struct dm_dev.
1318 */
1319static int action_dev(struct multipath *m, struct dm_dev *dev,
1320 action_fn action)
1321{
Mike Snitzer19040c02011-03-24 13:54:31 +00001322 int r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 struct pgpath *pgpath;
1324 struct priority_group *pg;
1325
1326 list_for_each_entry(pg, &m->priority_groups, list) {
1327 list_for_each_entry(pgpath, &pg->pgpaths, list) {
1328 if (pgpath->path.dev == dev)
1329 r = action(pgpath);
1330 }
1331 }
1332
1333 return r;
1334}
1335
1336/*
1337 * Temporarily try to avoid having to use the specified PG
1338 */
1339static void bypass_pg(struct multipath *m, struct priority_group *pg,
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001340 bool bypassed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
1342 unsigned long flags;
1343
1344 spin_lock_irqsave(&m->lock, flags);
1345
1346 pg->bypassed = bypassed;
1347 m->current_pgpath = NULL;
1348 m->current_pg = NULL;
1349
1350 spin_unlock_irqrestore(&m->lock, flags);
1351
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001352 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353}
1354
1355/*
1356 * Switch to using the specified PG from the next I/O that gets mapped
1357 */
1358static int switch_pg_num(struct multipath *m, const char *pgstr)
1359{
1360 struct priority_group *pg;
1361 unsigned pgnum;
1362 unsigned long flags;
Mikulas Patocka31998ef2012-03-28 18:41:26 +01001363 char dummy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Mikulas Patocka31998ef2012-03-28 18:41:26 +01001365 if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 (pgnum > m->nr_priority_groups)) {
1367 DMWARN("invalid PG number supplied to switch_pg_num");
1368 return -EINVAL;
1369 }
1370
1371 spin_lock_irqsave(&m->lock, flags);
1372 list_for_each_entry(pg, &m->priority_groups, list) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001373 pg->bypassed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (--pgnum)
1375 continue;
1376
1377 m->current_pgpath = NULL;
1378 m->current_pg = NULL;
1379 m->next_pg = pg;
1380 }
1381 spin_unlock_irqrestore(&m->lock, flags);
1382
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001383 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 return 0;
1385}
1386
1387/*
1388 * Set/clear bypassed status of a PG.
1389 * PGs are numbered upwards from 1 in the order they were declared.
1390 */
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001391static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392{
1393 struct priority_group *pg;
1394 unsigned pgnum;
Mikulas Patocka31998ef2012-03-28 18:41:26 +01001395 char dummy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
Mikulas Patocka31998ef2012-03-28 18:41:26 +01001397 if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 (pgnum > m->nr_priority_groups)) {
1399 DMWARN("invalid PG number supplied to bypass_pg");
1400 return -EINVAL;
1401 }
1402
1403 list_for_each_entry(pg, &m->priority_groups, list) {
1404 if (!--pgnum)
1405 break;
1406 }
1407
1408 bypass_pg(m, pg, bypassed);
1409 return 0;
1410}
1411
1412/*
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001413 * Should we retry pg_init immediately?
1414 */
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001415static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001416{
1417 unsigned long flags;
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001418 bool limit_reached = false;
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001419
1420 spin_lock_irqsave(&m->lock, flags);
1421
Mike Snitzer91e968a2016-03-17 17:10:15 -04001422 if (atomic_read(&m->pg_init_count) <= m->pg_init_retries &&
1423 !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags))
Mike Snitzer518257b2016-03-17 16:32:10 -04001424 set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001425 else
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001426 limit_reached = true;
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001427
1428 spin_unlock_irqrestore(&m->lock, flags);
1429
1430 return limit_reached;
1431}
1432
Chandra Seetharaman3ae31f62009-10-21 09:22:46 -07001433static void pg_init_done(void *data, int errors)
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001434{
Moger, Babu83c0d5d2010-03-06 02:29:45 +00001435 struct pgpath *pgpath = data;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001436 struct priority_group *pg = pgpath->pg;
1437 struct multipath *m = pg->m;
1438 unsigned long flags;
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001439 bool delay_retry = false;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001440
1441 /* device or driver problems */
1442 switch (errors) {
1443 case SCSI_DH_OK:
1444 break;
1445 case SCSI_DH_NOSYS:
1446 if (!m->hw_handler_name) {
1447 errors = 0;
1448 break;
1449 }
Moger, Babuf7b934c2010-03-06 02:29:49 +00001450 DMERR("Could not failover the device: Handler scsi_dh_%s "
1451 "Error %d.", m->hw_handler_name, errors);
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001452 /*
1453 * Fail path for now, so we do not ping pong
1454 */
1455 fail_path(pgpath);
1456 break;
1457 case SCSI_DH_DEV_TEMP_BUSY:
1458 /*
1459 * Probably doing something like FW upgrade on the
1460 * controller so try the other pg.
1461 */
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001462 bypass_pg(m, pg, true);
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001463 break;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001464 case SCSI_DH_RETRY:
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001465 /* Wait before retrying. */
1466 delay_retry = 1;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001467 case SCSI_DH_IMM_RETRY:
1468 case SCSI_DH_RES_TEMP_UNAVAIL:
1469 if (pg_init_limit_reached(m, pgpath))
1470 fail_path(pgpath);
1471 errors = 0;
1472 break;
Mike Snitzerec31f3f2016-02-20 12:49:43 -05001473 case SCSI_DH_DEV_OFFLINED:
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001474 default:
1475 /*
1476 * We probably do not want to fail the path for a device
1477 * error, but this is what the old dm did. In future
1478 * patches we can do more advanced handling.
1479 */
1480 fail_path(pgpath);
1481 }
1482
1483 spin_lock_irqsave(&m->lock, flags);
1484 if (errors) {
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001485 if (pgpath == m->current_pgpath) {
1486 DMERR("Could not failover device. Error %d.", errors);
1487 m->current_pgpath = NULL;
1488 m->current_pg = NULL;
1489 }
Mike Snitzer518257b2016-03-17 16:32:10 -04001490 } else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags))
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001491 pg->bypassed = false;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001492
Mike Snitzer91e968a2016-03-17 17:10:15 -04001493 if (atomic_dec_return(&m->pg_init_in_progress) > 0)
Kiyoshi Uedad0259bf2010-03-06 02:30:02 +00001494 /* Activations of other paths are still on going */
1495 goto out;
1496
Mike Snitzer518257b2016-03-17 16:32:10 -04001497 if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) {
1498 if (delay_retry)
1499 set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags);
1500 else
1501 clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags);
1502
Hannes Reinecke3e9f1be2014-02-28 15:33:45 +01001503 if (__pg_init_all_paths(m))
1504 goto out;
1505 }
Mike Snitzer518257b2016-03-17 16:32:10 -04001506 clear_bit(MPATHF_QUEUE_IO, &m->flags);
Kiyoshi Uedad0259bf2010-03-06 02:30:02 +00001507
Mike Snitzer7e48c762016-09-14 10:47:03 -04001508 process_queued_io_list(m);
Mike Snitzer76e33fe2016-05-19 16:15:14 -04001509
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001510 /*
1511 * Wake up any thread waiting to suspend.
1512 */
1513 wake_up(&m->pg_init_wait);
1514
Kiyoshi Uedad0259bf2010-03-06 02:30:02 +00001515out:
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001516 spin_unlock_irqrestore(&m->lock, flags);
1517}
1518
Bart Van Assche042d8db2017-04-27 10:11:14 -07001519static void activate_or_offline_path(struct pgpath *pgpath)
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001520{
Mike Snitzerf10e06b2016-09-01 12:06:37 -04001521 struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001522
Mike Snitzerf10e06b2016-09-01 12:06:37 -04001523 if (pgpath->is_active && !blk_queue_dying(q))
1524 scsi_dh_activate(q, pg_init_done, pgpath);
Hannes Reinecke3a017502014-02-28 15:33:49 +01001525 else
1526 pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001527}
1528
Bart Van Assche042d8db2017-04-27 10:11:14 -07001529static void activate_path_work(struct work_struct *work)
1530{
1531 struct pgpath *pgpath =
1532 container_of(work, struct pgpath, activate_path.work);
1533
1534 activate_or_offline_path(pgpath);
1535}
1536
Hannes Reinecke7e782af2013-07-01 15:16:26 +02001537static int noretry_error(int error)
1538{
1539 switch (error) {
Hannes Reinecke8ff232c2015-07-15 13:23:24 +02001540 case -EBADE:
1541 /*
1542 * EBADE signals an reservation conflict.
1543 * We shouldn't fail the path here as we can communicate with
1544 * the target. We should failover to the next path, but in
1545 * doing so we might be causing a ping-pong between paths.
1546 * So just return the reservation conflict error.
1547 */
Hannes Reinecke7e782af2013-07-01 15:16:26 +02001548 case -EOPNOTSUPP:
1549 case -EREMOTEIO:
1550 case -EILSEQ:
1551 case -ENODATA:
Jun'ichi Nomuracc9d3c32013-09-13 14:54:30 +09001552 case -ENOSPC:
Hannes Reinecke7e782af2013-07-01 15:16:26 +02001553 return 1;
1554 }
1555
1556 /* Anything else could be a path failure, so should be retried */
1557 return 0;
1558}
1559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560/*
1561 * end_io handling
1562 */
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001563static int do_end_io(struct multipath *m, struct request *clone,
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001564 int error, struct dm_mpath_io *mpio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565{
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001566 /*
1567 * We don't queue any clone request inside the multipath target
1568 * during end I/O handling, since those clone requests don't have
1569 * bio clones. If we queue them inside the multipath target,
1570 * we need to make bio clones, that requires memory allocation.
Mike Snitzer4cc96132016-05-12 16:28:10 -04001571 * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001572 * don't have bio clones.)
1573 * Instead of queueing the clone request here, we queue the original
1574 * request into dm core, which will remake a clone request and
1575 * clone bios for it and resubmit it later.
1576 */
1577 int r = DM_ENDIO_REQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001579 if (!error && !clone->errors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 return 0; /* I/O complete */
1581
Mike Snitzer7eee4ae2014-06-02 15:50:06 -04001582 if (noretry_error(error))
Mike Snitzer959eb4e2010-08-12 04:14:32 +01001583 return error;
1584
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001585 if (mpio->pgpath)
1586 fail_path(mpio->pgpath);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Mike Snitzer91e968a2016-03-17 17:10:15 -04001588 if (!atomic_read(&m->nr_valid_paths)) {
Mike Snitzer518257b2016-03-17 16:32:10 -04001589 if (!test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
Mike Snitzer76e33fe2016-05-19 16:15:14 -04001590 if (!must_push_back_rq(m))
Hannes Reinecke751b2a72011-01-18 10:13:12 +01001591 r = -EIO;
Hannes Reinecke751b2a72011-01-18 10:13:12 +01001592 }
1593 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001595 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596}
1597
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001598static int multipath_end_io(struct dm_target *ti, struct request *clone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 int error, union map_info *map_context)
1600{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001601 struct multipath *m = ti->private;
Mike Snitzer2eff1922016-02-03 09:13:14 -05001602 struct dm_mpath_io *mpio = get_mpio(map_context);
Wei Yongjuna71a2612012-10-12 16:59:42 +01001603 struct pgpath *pgpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 struct path_selector *ps;
1605 int r;
1606
Jun'ichi Nomura466891f2012-03-28 18:41:25 +01001607 BUG_ON(!mpio);
1608
Mike Snitzer2eff1922016-02-03 09:13:14 -05001609 r = do_end_io(m, clone, error, mpio);
Wei Yongjuna71a2612012-10-12 16:59:42 +01001610 pgpath = mpio->pgpath;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 if (pgpath) {
1612 ps = &pgpath->pg->ps;
1613 if (ps->type->end_io)
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +01001614 ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
Mike Snitzer2eff1922016-02-03 09:13:14 -05001616 clear_request_fn_mpio(m, map_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617
1618 return r;
1619}
1620
Mike Snitzer76e33fe2016-05-19 16:15:14 -04001621static int do_end_io_bio(struct multipath *m, struct bio *clone,
1622 int error, struct dm_mpath_io *mpio)
1623{
1624 unsigned long flags;
1625
1626 if (!error)
1627 return 0; /* I/O complete */
1628
1629 if (noretry_error(error))
1630 return error;
1631
1632 if (mpio->pgpath)
1633 fail_path(mpio->pgpath);
1634
1635 if (!atomic_read(&m->nr_valid_paths)) {
1636 if (!test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
1637 if (!must_push_back_bio(m))
1638 return -EIO;
1639 return DM_ENDIO_REQUEUE;
Mike Snitzer76e33fe2016-05-19 16:15:14 -04001640 }
1641 }
1642
1643 /* Queue for the daemon to resubmit */
Mike Snitzerbf661be2016-05-24 15:48:08 -04001644 dm_bio_restore(get_bio_details_from_bio(clone), clone);
Mike Snitzer76e33fe2016-05-19 16:15:14 -04001645
1646 spin_lock_irqsave(&m->lock, flags);
1647 bio_list_add(&m->queued_bios, clone);
1648 spin_unlock_irqrestore(&m->lock, flags);
1649 if (!test_bit(MPATHF_QUEUE_IO, &m->flags))
1650 queue_work(kmultipathd, &m->process_queued_bios);
1651
1652 return DM_ENDIO_INCOMPLETE;
1653}
1654
1655static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, int error)
1656{
1657 struct multipath *m = ti->private;
1658 struct dm_mpath_io *mpio = get_mpio_from_bio(clone);
1659 struct pgpath *pgpath;
1660 struct path_selector *ps;
1661 int r;
1662
1663 BUG_ON(!mpio);
1664
1665 r = do_end_io_bio(m, clone, error, mpio);
1666 pgpath = mpio->pgpath;
1667 if (pgpath) {
1668 ps = &pgpath->pg->ps;
1669 if (ps->type->end_io)
1670 ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes);
1671 }
1672
1673 return r;
1674}
1675
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676/*
1677 * Suspend can't complete until all the I/O is processed so if
Alasdair G Kergon436d4102005-07-12 15:53:03 -07001678 * the last path fails we must error any remaining I/O.
1679 * Note that if the freeze_bdev fails while suspending, the
1680 * queue_if_no_path state is lost - userspace should reset it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 */
1682static void multipath_presuspend(struct dm_target *ti)
1683{
Mike Snitzer7943bd62016-02-02 21:53:15 -05001684 struct multipath *m = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001686 queue_if_no_path(m, false, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687}
1688
Kiyoshi Ueda6df400a2009-12-10 23:52:19 +00001689static void multipath_postsuspend(struct dm_target *ti)
1690{
Mike Anderson6380f262009-12-10 23:52:21 +00001691 struct multipath *m = ti->private;
1692
1693 mutex_lock(&m->work_mutex);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001694 flush_multipath_work(m);
Mike Anderson6380f262009-12-10 23:52:21 +00001695 mutex_unlock(&m->work_mutex);
Kiyoshi Ueda6df400a2009-12-10 23:52:19 +00001696}
1697
Alasdair G Kergon436d4102005-07-12 15:53:03 -07001698/*
1699 * Restore the queue_if_no_path setting.
1700 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701static void multipath_resume(struct dm_target *ti)
1702{
Mike Snitzer7943bd62016-02-02 21:53:15 -05001703 struct multipath *m = ti->private;
Mike Snitzer1814f2e2016-07-25 21:08:51 -04001704 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
Mike Snitzer1814f2e2016-07-25 21:08:51 -04001706 spin_lock_irqsave(&m->lock, flags);
Mike Snitzer518257b2016-03-17 16:32:10 -04001707 if (test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags))
1708 set_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags);
1709 else
1710 clear_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags);
Mike Snitzer1814f2e2016-07-25 21:08:51 -04001711 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712}
1713
1714/*
1715 * Info output has the following format:
1716 * num_multipath_feature_args [multipath_feature_args]*
1717 * num_handler_status_args [handler_status_args]*
1718 * num_groups init_group_number
1719 * [A|D|E num_ps_status_args [ps_status_args]*
1720 * num_paths num_selector_args
1721 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1722 *
1723 * Table output has the following format (identical to the constructor string):
1724 * num_feature_args [features_args]*
1725 * num_handler_args hw_handler [hw_handler_args]*
1726 * num_groups init_group_number
1727 * [priority selector-name num_ps_args [ps_args]*
1728 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1729 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00001730static void multipath_status(struct dm_target *ti, status_type_t type,
1731 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732{
1733 int sz = 0;
1734 unsigned long flags;
Mike Snitzer7943bd62016-02-02 21:53:15 -05001735 struct multipath *m = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 struct priority_group *pg;
1737 struct pgpath *p;
1738 unsigned pg_num;
1739 char state;
1740
1741 spin_lock_irqsave(&m->lock, flags);
1742
1743 /* Features */
1744 if (type == STATUSTYPE_INFO)
Mike Snitzer91e968a2016-03-17 17:10:15 -04001745 DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags),
1746 atomic_read(&m->pg_init_count));
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001747 else {
Mike Snitzer518257b2016-03-17 16:32:10 -04001748 DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) +
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001749 (m->pg_init_retries > 0) * 2 +
Mike Snitzera58a9352012-07-27 15:08:04 +01001750 (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 +
Mike Snitzere83068a2016-05-24 21:16:51 -04001751 test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) +
1752 (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2);
1753
Mike Snitzer518257b2016-03-17 16:32:10 -04001754 if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001755 DMEMIT("queue_if_no_path ");
1756 if (m->pg_init_retries)
1757 DMEMIT("pg_init_retries %u ", m->pg_init_retries);
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001758 if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT)
1759 DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs);
Mike Snitzer518257b2016-03-17 16:32:10 -04001760 if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags))
Mike Snitzera58a9352012-07-27 15:08:04 +01001761 DMEMIT("retain_attached_hw_handler ");
Mike Snitzere83068a2016-05-24 21:16:51 -04001762 if (m->queue_mode != DM_TYPE_REQUEST_BASED) {
1763 switch(m->queue_mode) {
1764 case DM_TYPE_BIO_BASED:
1765 DMEMIT("queue_mode bio ");
1766 break;
1767 case DM_TYPE_MQ_REQUEST_BASED:
1768 DMEMIT("queue_mode mq ");
1769 break;
1770 }
1771 }
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001774 if (!m->hw_handler_name || type == STATUSTYPE_INFO)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 DMEMIT("0 ");
1776 else
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001777 DMEMIT("1 %s ", m->hw_handler_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
1779 DMEMIT("%u ", m->nr_priority_groups);
1780
1781 if (m->next_pg)
1782 pg_num = m->next_pg->pg_num;
1783 else if (m->current_pg)
1784 pg_num = m->current_pg->pg_num;
1785 else
Mike Snitzera490a072011-03-24 13:54:33 +00001786 pg_num = (m->nr_priority_groups ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
1788 DMEMIT("%u ", pg_num);
1789
1790 switch (type) {
1791 case STATUSTYPE_INFO:
1792 list_for_each_entry(pg, &m->priority_groups, list) {
1793 if (pg->bypassed)
1794 state = 'D'; /* Disabled */
1795 else if (pg == m->current_pg)
1796 state = 'A'; /* Currently Active */
1797 else
1798 state = 'E'; /* Enabled */
1799
1800 DMEMIT("%c ", state);
1801
1802 if (pg->ps.type->status)
1803 sz += pg->ps.type->status(&pg->ps, NULL, type,
1804 result + sz,
1805 maxlen - sz);
1806 else
1807 DMEMIT("0 ");
1808
1809 DMEMIT("%u %u ", pg->nr_pgpaths,
1810 pg->ps.type->info_args);
1811
1812 list_for_each_entry(p, &pg->pgpaths, list) {
1813 DMEMIT("%s %s %u ", p->path.dev->name,
Kiyoshi Ueda66800732008-10-10 13:36:58 +01001814 p->is_active ? "A" : "F",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 p->fail_count);
1816 if (pg->ps.type->status)
1817 sz += pg->ps.type->status(&pg->ps,
1818 &p->path, type, result + sz,
1819 maxlen - sz);
1820 }
1821 }
1822 break;
1823
1824 case STATUSTYPE_TABLE:
1825 list_for_each_entry(pg, &m->priority_groups, list) {
1826 DMEMIT("%s ", pg->ps.type->name);
1827
1828 if (pg->ps.type->status)
1829 sz += pg->ps.type->status(&pg->ps, NULL, type,
1830 result + sz,
1831 maxlen - sz);
1832 else
1833 DMEMIT("0 ");
1834
1835 DMEMIT("%u %u ", pg->nr_pgpaths,
1836 pg->ps.type->table_args);
1837
1838 list_for_each_entry(p, &pg->pgpaths, list) {
1839 DMEMIT("%s ", p->path.dev->name);
1840 if (pg->ps.type->status)
1841 sz += pg->ps.type->status(&pg->ps,
1842 &p->path, type, result + sz,
1843 maxlen - sz);
1844 }
1845 }
1846 break;
1847 }
1848
1849 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850}
1851
1852static int multipath_message(struct dm_target *ti, unsigned argc, char **argv)
1853{
Mike Anderson6380f262009-12-10 23:52:21 +00001854 int r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 struct dm_dev *dev;
Mike Snitzer7943bd62016-02-02 21:53:15 -05001856 struct multipath *m = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 action_fn action;
1858
Mike Anderson6380f262009-12-10 23:52:21 +00001859 mutex_lock(&m->work_mutex);
1860
Kiyoshi Uedac2f3d242009-12-10 23:52:27 +00001861 if (dm_suspended(ti)) {
1862 r = -EBUSY;
1863 goto out;
1864 }
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 if (argc == 1) {
Mike Snitzer498f0102011-08-02 12:32:04 +01001867 if (!strcasecmp(argv[0], "queue_if_no_path")) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001868 r = queue_if_no_path(m, true, false);
Mike Anderson6380f262009-12-10 23:52:21 +00001869 goto out;
Mike Snitzer498f0102011-08-02 12:32:04 +01001870 } else if (!strcasecmp(argv[0], "fail_if_no_path")) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001871 r = queue_if_no_path(m, false, false);
Mike Anderson6380f262009-12-10 23:52:21 +00001872 goto out;
1873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 }
1875
Mike Anderson6380f262009-12-10 23:52:21 +00001876 if (argc != 2) {
Jose Castilloa356e422014-01-29 17:52:45 +01001877 DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc);
Mike Anderson6380f262009-12-10 23:52:21 +00001878 goto out;
1879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Mike Snitzer498f0102011-08-02 12:32:04 +01001881 if (!strcasecmp(argv[0], "disable_group")) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001882 r = bypass_pg_num(m, argv[1], true);
Mike Anderson6380f262009-12-10 23:52:21 +00001883 goto out;
Mike Snitzer498f0102011-08-02 12:32:04 +01001884 } else if (!strcasecmp(argv[0], "enable_group")) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001885 r = bypass_pg_num(m, argv[1], false);
Mike Anderson6380f262009-12-10 23:52:21 +00001886 goto out;
Mike Snitzer498f0102011-08-02 12:32:04 +01001887 } else if (!strcasecmp(argv[0], "switch_group")) {
Mike Anderson6380f262009-12-10 23:52:21 +00001888 r = switch_pg_num(m, argv[1]);
1889 goto out;
Mike Snitzer498f0102011-08-02 12:32:04 +01001890 } else if (!strcasecmp(argv[0], "reinstate_path"))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 action = reinstate_path;
Mike Snitzer498f0102011-08-02 12:32:04 +01001892 else if (!strcasecmp(argv[0], "fail_path"))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 action = fail_path;
Mike Anderson6380f262009-12-10 23:52:21 +00001894 else {
Jose Castilloa356e422014-01-29 17:52:45 +01001895 DMWARN("Unrecognised multipath message received: %s", argv[0]);
Mike Anderson6380f262009-12-10 23:52:21 +00001896 goto out;
1897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +00001899 r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 if (r) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001901 DMWARN("message: error getting device %s",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 argv[1]);
Mike Anderson6380f262009-12-10 23:52:21 +00001903 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 }
1905
1906 r = action_dev(m, dev, action);
1907
1908 dm_put_device(ti, dev);
1909
Mike Anderson6380f262009-12-10 23:52:21 +00001910out:
1911 mutex_unlock(&m->work_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913}
1914
Christoph Hellwige56f81e2015-10-15 14:10:50 +02001915static int multipath_prepare_ioctl(struct dm_target *ti,
1916 struct block_device **bdev, fmode_t *mode)
Milan Broz9af4aa32006-10-03 01:15:20 -07001917{
Mikulas Patocka35991652012-06-03 00:29:58 +01001918 struct multipath *m = ti->private;
Mike Snitzer2da16102016-03-17 18:38:17 -04001919 struct pgpath *current_pgpath;
Mikulas Patocka35991652012-06-03 00:29:58 +01001920 int r;
1921
Mike Snitzer2da16102016-03-17 18:38:17 -04001922 current_pgpath = lockless_dereference(m->current_pgpath);
1923 if (!current_pgpath)
1924 current_pgpath = choose_pgpath(m, 0);
Milan Broz9af4aa32006-10-03 01:15:20 -07001925
Mike Snitzer2da16102016-03-17 18:38:17 -04001926 if (current_pgpath) {
Mike Snitzer518257b2016-03-17 16:32:10 -04001927 if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) {
Mike Snitzer2da16102016-03-17 18:38:17 -04001928 *bdev = current_pgpath->path.dev->bdev;
1929 *mode = current_pgpath->path.dev->mode;
Junichi Nomura43e43c92015-11-17 09:36:56 +00001930 r = 0;
1931 } else {
1932 /* pg_init has not started or completed */
1933 r = -ENOTCONN;
1934 }
1935 } else {
1936 /* No path is available */
Mike Snitzer518257b2016-03-17 16:32:10 -04001937 if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
Junichi Nomura43e43c92015-11-17 09:36:56 +00001938 r = -ENOTCONN;
1939 else
1940 r = -EIO;
Milan Broze90dae12006-10-03 01:15:22 -07001941 }
Milan Broz9af4aa32006-10-03 01:15:20 -07001942
Junichi Nomura5bbbfdf2015-11-17 09:39:26 +00001943 if (r == -ENOTCONN) {
Mike Snitzer2da16102016-03-17 18:38:17 -04001944 if (!lockless_dereference(m->current_pg)) {
Hannes Reinecke3e9f1be2014-02-28 15:33:45 +01001945 /* Path status changed, redo selection */
Mike Snitzer2da16102016-03-17 18:38:17 -04001946 (void) choose_pgpath(m, 0);
Hannes Reinecke3e9f1be2014-02-28 15:33:45 +01001947 }
Mike Snitzer518257b2016-03-17 16:32:10 -04001948 if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags))
Mike Snitzer2da16102016-03-17 18:38:17 -04001949 pg_init_all_paths(m);
Hannes Reinecke63d832c2014-05-26 14:45:39 +02001950 dm_table_run_md_queue_async(m->ti->table);
Mike Snitzer7e48c762016-09-14 10:47:03 -04001951 process_queued_io_list(m);
Hannes Reinecke3e9f1be2014-02-28 15:33:45 +01001952 }
Mikulas Patocka35991652012-06-03 00:29:58 +01001953
Christoph Hellwige56f81e2015-10-15 14:10:50 +02001954 /*
1955 * Only pass ioctls through if the device sizes match exactly.
1956 */
1957 if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT)
1958 return 1;
1959 return r;
Milan Broz9af4aa32006-10-03 01:15:20 -07001960}
1961
Mike Snitzeraf4874e2009-06-22 10:12:33 +01001962static int multipath_iterate_devices(struct dm_target *ti,
1963 iterate_devices_callout_fn fn, void *data)
1964{
1965 struct multipath *m = ti->private;
1966 struct priority_group *pg;
1967 struct pgpath *p;
1968 int ret = 0;
1969
1970 list_for_each_entry(pg, &m->priority_groups, list) {
1971 list_for_each_entry(p, &pg->pgpaths, list) {
Mike Snitzer5dea2712009-07-23 20:30:42 +01001972 ret = fn(ti, p->path.dev, ti->begin, ti->len, data);
Mike Snitzeraf4874e2009-06-22 10:12:33 +01001973 if (ret)
1974 goto out;
1975 }
1976 }
1977
1978out:
1979 return ret;
1980}
1981
Mike Snitzer9f54cec2016-02-11 21:42:28 -05001982static int pgpath_busy(struct pgpath *pgpath)
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001983{
1984 struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev);
1985
Mike Snitzer52b09912015-02-23 16:36:41 -05001986 return blk_lld_busy(q);
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001987}
1988
1989/*
1990 * We return "busy", only when we can map I/Os but underlying devices
1991 * are busy (so even if we map I/Os now, the I/Os will wait on
1992 * the underlying queue).
1993 * In other words, if we want to kill I/Os or queue them inside us
1994 * due to map unavailability, we don't return "busy". Otherwise,
1995 * dm core won't give us the I/Os and we can't do what we want.
1996 */
1997static int multipath_busy(struct dm_target *ti)
1998{
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05001999 bool busy = false, has_active = false;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002000 struct multipath *m = ti->private;
Mike Snitzer2da16102016-03-17 18:38:17 -04002001 struct priority_group *pg, *next_pg;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002002 struct pgpath *pgpath;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002003
Mike Snitzerb88efd42016-09-09 19:26:19 -04002004 /* pg_init in progress */
2005 if (atomic_read(&m->pg_init_in_progress))
Mike Snitzer2da16102016-03-17 18:38:17 -04002006 return true;
2007
Mike Snitzerb88efd42016-09-09 19:26:19 -04002008 /* no paths available, for blk-mq: rely on IO mapping to delay requeue */
2009 if (!atomic_read(&m->nr_valid_paths) && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
2010 return (m->queue_mode != DM_TYPE_MQ_REQUEST_BASED);
2011
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002012 /* Guess which priority_group will be used at next mapping time */
Mike Snitzer2da16102016-03-17 18:38:17 -04002013 pg = lockless_dereference(m->current_pg);
2014 next_pg = lockless_dereference(m->next_pg);
2015 if (unlikely(!lockless_dereference(m->current_pgpath) && next_pg))
2016 pg = next_pg;
2017
2018 if (!pg) {
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002019 /*
2020 * We don't know which pg will be used at next mapping time.
Mike Snitzer2da16102016-03-17 18:38:17 -04002021 * We don't call choose_pgpath() here to avoid to trigger
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002022 * pg_init just by busy checking.
2023 * So we don't know whether underlying devices we will be using
2024 * at next mapping time are busy or not. Just try mapping.
2025 */
Mike Snitzer2da16102016-03-17 18:38:17 -04002026 return busy;
2027 }
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002028
2029 /*
2030 * If there is one non-busy active path at least, the path selector
2031 * will be able to select it. So we consider such a pg as not busy.
2032 */
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05002033 busy = true;
Mike Snitzer2da16102016-03-17 18:38:17 -04002034 list_for_each_entry(pgpath, &pg->pgpaths, list) {
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002035 if (pgpath->is_active) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05002036 has_active = true;
Mike Snitzer9f54cec2016-02-11 21:42:28 -05002037 if (!pgpath_busy(pgpath)) {
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05002038 busy = false;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002039 break;
2040 }
2041 }
Mike Snitzer2da16102016-03-17 18:38:17 -04002042 }
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002043
Mike Snitzer2da16102016-03-17 18:38:17 -04002044 if (!has_active) {
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002045 /*
2046 * No active path in this pg, so this pg won't be used and
2047 * the current_pg will be changed at next mapping time.
2048 * We need to try mapping to determine it.
2049 */
Mike Snitzerbe7d31c2016-02-10 13:02:21 -05002050 busy = false;
Mike Snitzer2da16102016-03-17 18:38:17 -04002051 }
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002052
2053 return busy;
2054}
2055
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056/*-----------------------------------------------------------------
2057 * Module setup
2058 *---------------------------------------------------------------*/
2059static struct target_type multipath_target = {
2060 .name = "multipath",
Mike Snitzere83068a2016-05-24 21:16:51 -04002061 .version = {1, 12, 0},
Mike Snitzer16f12262016-01-31 17:22:27 -05002062 .features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 .module = THIS_MODULE,
2064 .ctr = multipath_ctr,
2065 .dtr = multipath_dtr,
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002066 .map_rq = multipath_map,
Mike Snitzere5863d92014-12-17 21:08:12 -05002067 .clone_and_map_rq = multipath_clone_and_map,
2068 .release_clone_rq = multipath_release_clone,
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01002069 .rq_end_io = multipath_end_io,
Mike Snitzer76e33fe2016-05-19 16:15:14 -04002070 .map = multipath_map_bio,
2071 .end_io = multipath_end_io_bio,
2072 .presuspend = multipath_presuspend,
2073 .postsuspend = multipath_postsuspend,
2074 .resume = multipath_resume,
2075 .status = multipath_status,
2076 .message = multipath_message,
2077 .prepare_ioctl = multipath_prepare_ioctl,
2078 .iterate_devices = multipath_iterate_devices,
2079 .busy = multipath_busy,
2080};
2081
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082static int __init dm_multipath_init(void)
2083{
2084 int r;
2085
Mike Snitzer76e33fe2016-05-19 16:15:14 -04002086 /* allocate a slab for the dm_mpath_ios */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002087 _mpio_cache = KMEM_CACHE(dm_mpath_io, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 if (!_mpio_cache)
2089 return -ENOMEM;
2090
2091 r = dm_register_target(&multipath_target);
2092 if (r < 0) {
Mike Snitzer76e33fe2016-05-19 16:15:14 -04002093 DMERR("request-based register failed %d", r);
Johannes Thumshirnff658e92015-01-11 12:45:23 +01002094 r = -EINVAL;
2095 goto bad_register_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 }
2097
Tejun Heo4d4d66a2011-01-13 19:59:57 +00002098 kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0);
Alasdair G Kergonc5573082005-05-05 16:16:07 -07002099 if (!kmultipathd) {
Alasdair G Kergon0cd33122007-07-12 17:27:01 +01002100 DMERR("failed to create workqueue kmpathd");
Johannes Thumshirnff658e92015-01-11 12:45:23 +01002101 r = -ENOMEM;
2102 goto bad_alloc_kmultipathd;
Alasdair G Kergonc5573082005-05-05 16:16:07 -07002103 }
2104
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07002105 /*
2106 * A separate workqueue is used to handle the device handlers
2107 * to avoid overloading existing workqueue. Overloading the
2108 * old workqueue would also create a bottleneck in the
2109 * path of the storage hardware device activation.
2110 */
Tejun Heo4d4d66a2011-01-13 19:59:57 +00002111 kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd",
2112 WQ_MEM_RECLAIM);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07002113 if (!kmpath_handlerd) {
2114 DMERR("failed to create workqueue kmpath_handlerd");
Johannes Thumshirnff658e92015-01-11 12:45:23 +01002115 r = -ENOMEM;
2116 goto bad_alloc_kmpath_handlerd;
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07002117 }
2118
Johannes Thumshirnff658e92015-01-11 12:45:23 +01002119 return 0;
2120
2121bad_alloc_kmpath_handlerd:
2122 destroy_workqueue(kmultipathd);
2123bad_alloc_kmultipathd:
2124 dm_unregister_target(&multipath_target);
2125bad_register_target:
2126 kmem_cache_destroy(_mpio_cache);
2127
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 return r;
2129}
2130
2131static void __exit dm_multipath_exit(void)
2132{
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07002133 destroy_workqueue(kmpath_handlerd);
Alasdair G Kergonc5573082005-05-05 16:16:07 -07002134 destroy_workqueue(kmultipathd);
2135
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00002136 dm_unregister_target(&multipath_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 kmem_cache_destroy(_mpio_cache);
2138}
2139
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140module_init(dm_multipath_init);
2141module_exit(dm_multipath_exit);
2142
2143MODULE_DESCRIPTION(DM_NAME " multipath target");
2144MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
2145MODULE_LICENSE("GPL");