blob: 5e0090ef4182e71de26afec04d23edb19b3f4e39 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include "dm-path-selector.h"
Mike Andersonb15546f2007-10-19 22:48:02 +010011#include "dm-uevent.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070012
13#include <linux/ctype.h>
14#include <linux/init.h>
15#include <linux/mempool.h>
16#include <linux/module.h>
17#include <linux/pagemap.h>
18#include <linux/slab.h>
19#include <linux/time.h>
20#include <linux/workqueue.h>
Chandra Seetharamancfae5c92008-05-01 14:50:11 -070021#include <scsi/scsi_dh.h>
Arun Sharma60063492011-07-26 16:09:06 -070022#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Alasdair G Kergon72d94862006-06-26 00:27:35 -070024#define DM_MSG_PREFIX "multipath"
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000025#define DM_PG_INIT_DELAY_MSECS 2000
26#define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28/* Path properties */
29struct pgpath {
30 struct list_head list;
31
32 struct priority_group *pg; /* Owning PG */
Kiyoshi Ueda66800732008-10-10 13:36:58 +010033 unsigned is_active; /* Path status */
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 unsigned fail_count; /* Cumulative failure count */
35
Josef "Jeff" Sipekc922d5f2006-12-08 02:36:33 -080036 struct dm_path path;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000037 struct delayed_work activate_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038};
39
40#define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path)
41
42/*
43 * Paths are grouped into Priority Groups and numbered from 1 upwards.
44 * Each has a path selector which controls which path gets used.
45 */
46struct priority_group {
47 struct list_head list;
48
49 struct multipath *m; /* Owning multipath instance */
50 struct path_selector ps;
51
52 unsigned pg_num; /* Reference number */
53 unsigned bypassed; /* Temporarily bypass this PG? */
54
55 unsigned nr_pgpaths; /* Number of paths in PG */
56 struct list_head pgpaths;
57};
58
59/* Multipath context */
60struct multipath {
61 struct list_head list;
62 struct dm_target *ti;
63
64 spinlock_t lock;
65
Chandra Seetharamancfae5c92008-05-01 14:50:11 -070066 const char *hw_handler_name;
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -070067 char *hw_handler_params;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000068
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 unsigned nr_priority_groups;
70 struct list_head priority_groups;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000071
72 wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 unsigned pg_init_required; /* pg_init needs calling? */
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -070075 unsigned pg_init_in_progress; /* Only one pg_init allowed at once */
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000076 unsigned pg_init_delay_retry; /* Delay pg_init retry? */
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78 unsigned nr_valid_paths; /* Total number of usable paths */
79 struct pgpath *current_pgpath;
80 struct priority_group *current_pg;
81 struct priority_group *next_pg; /* Switch to this PG if set */
82 unsigned repeat_count; /* I/Os left before calling PS again */
83
84 unsigned queue_io; /* Must we queue all I/O? */
85 unsigned queue_if_no_path; /* Queue I/O if last path fails? */
Alasdair G Kergon436d4102005-07-12 15:53:03 -070086 unsigned saved_queue_if_no_path;/* Saved state during suspension */
Dave Wysochanskic9e45582007-10-19 22:47:53 +010087 unsigned pg_init_retries; /* Number of times to retry pg_init */
88 unsigned pg_init_count; /* Number of times pg_init called */
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +000089 unsigned pg_init_delay_msecs; /* Number of msecs before pg_init retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
91 struct work_struct process_queued_ios;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +010092 struct list_head queued_ios;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 unsigned queue_size;
94
95 struct work_struct trigger_event;
96
97 /*
Alasdair G Kergon028867a2007-07-12 17:26:32 +010098 * We must use a mempool of dm_mpath_io structs so that we
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 * can resubmit bios on error.
100 */
101 mempool_t *mpio_pool;
Mike Anderson6380f262009-12-10 23:52:21 +0000102
103 struct mutex work_mutex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104};
105
106/*
107 * Context information attached to each bio we process.
108 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100109struct dm_mpath_io {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 struct pgpath *pgpath;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100111 size_t nr_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112};
113
114typedef int (*action_fn) (struct pgpath *pgpath);
115
116#define MIN_IOS 256 /* Mempool size */
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 process_queued_ios(struct work_struct *work);
122static void trigger_event(struct work_struct *work);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -0700123static void activate_path(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
125
126/*-----------------------------------------------
127 * Allocation routines
128 *-----------------------------------------------*/
129
130static struct pgpath *alloc_pgpath(void)
131{
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700132 struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Mike Anderson224cb3e2008-08-29 09:36:09 +0200134 if (pgpath) {
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100135 pgpath->is_active = 1;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000136 INIT_DELAYED_WORK(&pgpath->activate_path, activate_path);
Mike Anderson224cb3e2008-08-29 09:36:09 +0200137 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
139 return pgpath;
140}
141
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100142static void free_pgpath(struct pgpath *pgpath)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
144 kfree(pgpath);
145}
146
147static struct priority_group *alloc_priority_group(void)
148{
149 struct priority_group *pg;
150
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700151 pg = kzalloc(sizeof(*pg), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700153 if (pg)
154 INIT_LIST_HEAD(&pg->pgpaths);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
156 return pg;
157}
158
159static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti)
160{
161 struct pgpath *pgpath, *tmp;
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700162 struct multipath *m = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164 list_for_each_entry_safe(pgpath, tmp, pgpaths, list) {
165 list_del(&pgpath->list);
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700166 if (m->hw_handler_name)
167 scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 dm_put_device(ti, pgpath->path.dev);
169 free_pgpath(pgpath);
170 }
171}
172
173static void free_priority_group(struct priority_group *pg,
174 struct dm_target *ti)
175{
176 struct path_selector *ps = &pg->ps;
177
178 if (ps->type) {
179 ps->type->destroy(ps);
180 dm_put_path_selector(ps->type);
181 }
182
183 free_pgpaths(&pg->pgpaths, ti);
184 kfree(pg);
185}
186
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700187static struct multipath *alloc_multipath(struct dm_target *ti)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
189 struct multipath *m;
190
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700191 m = kzalloc(sizeof(*m), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 if (m) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 INIT_LIST_HEAD(&m->priority_groups);
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100194 INIT_LIST_HEAD(&m->queued_ios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 spin_lock_init(&m->lock);
196 m->queue_io = 1;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000197 m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT;
David Howellsc4028952006-11-22 14:57:56 +0000198 INIT_WORK(&m->process_queued_ios, process_queued_ios);
199 INIT_WORK(&m->trigger_event, trigger_event);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +0000200 init_waitqueue_head(&m->pg_init_wait);
Mike Anderson6380f262009-12-10 23:52:21 +0000201 mutex_init(&m->work_mutex);
Matthew Dobson93d23412006-03-26 01:37:50 -0800202 m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 if (!m->mpio_pool) {
204 kfree(m);
205 return NULL;
206 }
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700207 m->ti = ti;
208 ti->private = m;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 }
210
211 return m;
212}
213
214static void free_multipath(struct multipath *m)
215{
216 struct priority_group *pg, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218 list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) {
219 list_del(&pg->list);
220 free_priority_group(pg, m->ti);
221 }
222
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700223 kfree(m->hw_handler_name);
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700224 kfree(m->hw_handler_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 mempool_destroy(m->mpio_pool);
226 kfree(m);
227}
228
229
230/*-----------------------------------------------
231 * Path selection
232 *-----------------------------------------------*/
233
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000234static void __pg_init_all_paths(struct multipath *m)
235{
236 struct pgpath *pgpath;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000237 unsigned long pg_init_delay = 0;
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000238
239 m->pg_init_count++;
240 m->pg_init_required = 0;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000241 if (m->pg_init_delay_retry)
242 pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ?
243 m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS);
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000244 list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) {
245 /* Skip failed paths */
246 if (!pgpath->is_active)
247 continue;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000248 if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path,
249 pg_init_delay))
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000250 m->pg_init_in_progress++;
251 }
252}
253
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254static void __switch_pg(struct multipath *m, struct pgpath *pgpath)
255{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 m->current_pg = pgpath->pg;
257
258 /* Must we initialise the PG first, and queue I/O till it's ready? */
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700259 if (m->hw_handler_name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 m->pg_init_required = 1;
261 m->queue_io = 1;
262 } else {
263 m->pg_init_required = 0;
264 m->queue_io = 0;
265 }
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100266
267 m->pg_init_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268}
269
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100270static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg,
271 size_t nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Josef "Jeff" Sipekc922d5f2006-12-08 02:36:33 -0800273 struct dm_path *path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100275 path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 if (!path)
277 return -ENXIO;
278
279 m->current_pgpath = path_to_pgpath(path);
280
281 if (m->current_pg != pg)
282 __switch_pg(m, m->current_pgpath);
283
284 return 0;
285}
286
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100287static void __choose_pgpath(struct multipath *m, size_t nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
289 struct priority_group *pg;
290 unsigned bypassed = 1;
291
292 if (!m->nr_valid_paths)
293 goto failed;
294
295 /* Were we instructed to switch PG? */
296 if (m->next_pg) {
297 pg = m->next_pg;
298 m->next_pg = NULL;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100299 if (!__choose_path_in_pg(m, pg, nr_bytes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 return;
301 }
302
303 /* Don't change PG until it has no remaining paths */
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100304 if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 return;
306
307 /*
308 * Loop through priority groups until we find a valid path.
309 * First time we skip PGs marked 'bypassed'.
310 * Second time we only try the ones we skipped.
311 */
312 do {
313 list_for_each_entry(pg, &m->priority_groups, list) {
314 if (pg->bypassed == bypassed)
315 continue;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100316 if (!__choose_path_in_pg(m, pg, nr_bytes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 return;
318 }
319 } while (bypassed--);
320
321failed:
322 m->current_pgpath = NULL;
323 m->current_pg = NULL;
324}
325
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800326/*
327 * Check whether bios must be queued in the device-mapper core rather
328 * than here in the target.
329 *
330 * m->lock must be held on entry.
331 *
332 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
333 * same value then we are not between multipath_presuspend()
334 * and multipath_resume() calls and we have no need to check
335 * for the DMF_NOFLUSH_SUSPENDING flag.
336 */
337static int __must_push_back(struct multipath *m)
338{
339 return (m->queue_if_no_path != m->saved_queue_if_no_path &&
340 dm_noflush_suspending(m->ti));
341}
342
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100343static int map_io(struct multipath *m, struct request *clone,
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100344 struct dm_mpath_io *mpio, unsigned was_queued)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -0800346 int r = DM_MAPIO_REMAPPED;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100347 size_t nr_bytes = blk_rq_bytes(clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 unsigned long flags;
349 struct pgpath *pgpath;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100350 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
352 spin_lock_irqsave(&m->lock, flags);
353
354 /* Do we need to select a new pgpath? */
355 if (!m->current_pgpath ||
356 (!m->queue_io && (m->repeat_count && --m->repeat_count == 0)))
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100357 __choose_pgpath(m, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 pgpath = m->current_pgpath;
360
361 if (was_queued)
362 m->queue_size--;
363
364 if ((pgpath && m->queue_io) ||
Alasdair G Kergon436d4102005-07-12 15:53:03 -0700365 (!pgpath && m->queue_if_no_path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 /* Queue for the daemon to resubmit */
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100367 list_add_tail(&clone->queuelist, &m->queued_ios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 m->queue_size++;
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700369 if ((m->pg_init_required && !m->pg_init_in_progress) ||
370 !m->queue_io)
Alasdair G Kergonc5573082005-05-05 16:16:07 -0700371 queue_work(kmultipathd, &m->process_queued_ios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 pgpath = NULL;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -0800373 r = DM_MAPIO_SUBMITTED;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100374 } else if (pgpath) {
375 bdev = pgpath->path.dev->bdev;
376 clone->q = bdev_get_queue(bdev);
377 clone->rq_disk = bdev->bd_disk;
378 } else if (__must_push_back(m))
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800379 r = DM_MAPIO_REQUEUE;
380 else
381 r = -EIO; /* Failed */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
383 mpio->pgpath = pgpath;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100384 mpio->nr_bytes = nr_bytes;
385
386 if (r == DM_MAPIO_REMAPPED && pgpath->pg->ps.type->start_io)
387 pgpath->pg->ps.type->start_io(&pgpath->pg->ps, &pgpath->path,
388 nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390 spin_unlock_irqrestore(&m->lock, flags);
391
392 return r;
393}
394
395/*
396 * If we run out of usable paths, should we queue I/O or error it?
397 */
Alasdair G Kergon485ef692005-09-27 21:45:45 -0700398static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path,
399 unsigned save_old_value)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400{
401 unsigned long flags;
402
403 spin_lock_irqsave(&m->lock, flags);
404
Alasdair G Kergon485ef692005-09-27 21:45:45 -0700405 if (save_old_value)
406 m->saved_queue_if_no_path = m->queue_if_no_path;
407 else
408 m->saved_queue_if_no_path = queue_if_no_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 m->queue_if_no_path = queue_if_no_path;
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700410 if (!m->queue_if_no_path && m->queue_size)
Alasdair G Kergonc5573082005-05-05 16:16:07 -0700411 queue_work(kmultipathd, &m->process_queued_ios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413 spin_unlock_irqrestore(&m->lock, flags);
414
415 return 0;
416}
417
418/*-----------------------------------------------------------------
419 * The multipath daemon is responsible for resubmitting queued ios.
420 *---------------------------------------------------------------*/
421
422static void dispatch_queued_ios(struct multipath *m)
423{
424 int r;
425 unsigned long flags;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100426 struct dm_mpath_io *mpio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 union map_info *info;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100428 struct request *clone, *n;
429 LIST_HEAD(cl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431 spin_lock_irqsave(&m->lock, flags);
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100432 list_splice_init(&m->queued_ios, &cl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 spin_unlock_irqrestore(&m->lock, flags);
434
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100435 list_for_each_entry_safe(clone, n, &cl, queuelist) {
436 list_del_init(&clone->queuelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100438 info = dm_get_rq_mapinfo(clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 mpio = info->ptr;
440
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100441 r = map_io(m, clone, mpio, 1);
442 if (r < 0) {
443 mempool_free(mpio, m->mpio_pool);
444 dm_kill_unmapped_request(clone, r);
445 } else if (r == DM_MAPIO_REMAPPED)
446 dm_dispatch_request(clone);
447 else if (r == DM_MAPIO_REQUEUE) {
448 mempool_free(mpio, m->mpio_pool);
449 dm_requeue_unmapped_request(clone);
450 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 }
452}
453
David Howellsc4028952006-11-22 14:57:56 +0000454static void process_queued_ios(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
David Howellsc4028952006-11-22 14:57:56 +0000456 struct multipath *m =
457 container_of(work, struct multipath, process_queued_ios);
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000458 struct pgpath *pgpath = NULL;
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100459 unsigned must_queue = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 unsigned long flags;
461
462 spin_lock_irqsave(&m->lock, flags);
463
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700464 if (!m->queue_size)
465 goto out;
466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 if (!m->current_pgpath)
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100468 __choose_pgpath(m, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 pgpath = m->current_pgpath;
471
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700472 if ((pgpath && !m->queue_io) ||
473 (!pgpath && !m->queue_if_no_path))
474 must_queue = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Kiyoshi Uedafb612642010-03-06 02:32:18 +0000476 if (m->pg_init_required && !m->pg_init_in_progress && pgpath)
477 __pg_init_all_paths(m);
478
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700479out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 if (!must_queue)
482 dispatch_queued_ios(m);
483}
484
485/*
486 * An event is triggered whenever a path is taken out of use.
487 * Includes path failure and PG bypass.
488 */
David Howellsc4028952006-11-22 14:57:56 +0000489static void trigger_event(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
David Howellsc4028952006-11-22 14:57:56 +0000491 struct multipath *m =
492 container_of(work, struct multipath, trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 dm_table_event(m->ti->table);
495}
496
497/*-----------------------------------------------------------------
498 * Constructor/argument parsing:
499 * <#multipath feature args> [<arg>]*
500 * <#hw_handler args> [hw_handler [<arg>]*]
501 * <#priority groups>
502 * <initial priority group>
503 * [<selector> <#selector args> [<arg>]*
504 * <#paths> <#per-path selector args>
505 * [<path> [<arg>]* ]+ ]+
506 *---------------------------------------------------------------*/
Mike Snitzer498f0102011-08-02 12:32:04 +0100507static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 struct dm_target *ti)
509{
510 int r;
511 struct path_selector_type *pst;
512 unsigned ps_argc;
513
Mike Snitzer498f0102011-08-02 12:32:04 +0100514 static struct dm_arg _args[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700515 {0, 1024, "invalid number of path selector args"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 };
517
Mike Snitzer498f0102011-08-02 12:32:04 +0100518 pst = dm_get_path_selector(dm_shift_arg(as));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 if (!pst) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700520 ti->error = "unknown path selector type";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 return -EINVAL;
522 }
523
Mike Snitzer498f0102011-08-02 12:32:04 +0100524 r = dm_read_arg_group(_args, as, &ps_argc, &ti->error);
Mikulas Patocka371b2e32008-07-21 12:00:24 +0100525 if (r) {
526 dm_put_path_selector(pst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 return -EINVAL;
Mikulas Patocka371b2e32008-07-21 12:00:24 +0100528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
530 r = pst->create(&pg->ps, ps_argc, as->argv);
531 if (r) {
532 dm_put_path_selector(pst);
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700533 ti->error = "path selector constructor failed";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 return r;
535 }
536
537 pg->ps.type = pst;
Mike Snitzer498f0102011-08-02 12:32:04 +0100538 dm_consume_args(as, ps_argc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540 return 0;
541}
542
Mike Snitzer498f0102011-08-02 12:32:04 +0100543static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 struct dm_target *ti)
545{
546 int r;
547 struct pgpath *p;
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700548 struct multipath *m = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 /* we need at least a path arg */
551 if (as->argc < 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700552 ti->error = "no device given";
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100553 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 }
555
556 p = alloc_pgpath();
557 if (!p)
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100558 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Mike Snitzer498f0102011-08-02 12:32:04 +0100560 r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table),
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +0000561 &p->path.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 if (r) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700563 ti->error = "error getting device";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 goto bad;
565 }
566
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700567 if (m->hw_handler_name) {
Hannes Reineckea0cf7ea2009-06-22 10:12:11 +0100568 struct request_queue *q = bdev_get_queue(p->path.dev->bdev);
569
570 r = scsi_dh_attach(q, m->hw_handler_name);
571 if (r == -EBUSY) {
572 /*
573 * Already attached to different hw_handler,
574 * try to reattach with correct one.
575 */
576 scsi_dh_detach(q);
577 r = scsi_dh_attach(q, m->hw_handler_name);
578 }
579
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700580 if (r < 0) {
Hannes Reineckea0cf7ea2009-06-22 10:12:11 +0100581 ti->error = "error attaching hardware handler";
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700582 dm_put_device(ti, p->path.dev);
583 goto bad;
584 }
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700585
586 if (m->hw_handler_params) {
587 r = scsi_dh_set_params(q, m->hw_handler_params);
588 if (r < 0) {
589 ti->error = "unable to set hardware "
590 "handler parameters";
591 scsi_dh_detach(q);
592 dm_put_device(ti, p->path.dev);
593 goto bad;
594 }
595 }
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700596 }
597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error);
599 if (r) {
600 dm_put_device(ti, p->path.dev);
601 goto bad;
602 }
603
604 return p;
605
606 bad:
607 free_pgpath(p);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100608 return ERR_PTR(r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
Mike Snitzer498f0102011-08-02 12:32:04 +0100611static struct priority_group *parse_priority_group(struct dm_arg_set *as,
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700612 struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
Mike Snitzer498f0102011-08-02 12:32:04 +0100614 static struct dm_arg _args[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700615 {1, 1024, "invalid number of paths"},
616 {0, 1024, "invalid number of selector args"}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 };
618
619 int r;
Mike Snitzer498f0102011-08-02 12:32:04 +0100620 unsigned i, nr_selector_args, nr_args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 struct priority_group *pg;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700622 struct dm_target *ti = m->ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624 if (as->argc < 2) {
625 as->argc = 0;
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100626 ti->error = "not enough priority group arguments";
627 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
629
630 pg = alloc_priority_group();
631 if (!pg) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700632 ti->error = "couldn't allocate priority group";
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100633 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635 pg->m = m;
636
637 r = parse_path_selector(as, pg, ti);
638 if (r)
639 goto bad;
640
641 /*
642 * read the paths
643 */
Mike Snitzer498f0102011-08-02 12:32:04 +0100644 r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 if (r)
646 goto bad;
647
Mike Snitzer498f0102011-08-02 12:32:04 +0100648 r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 if (r)
650 goto bad;
651
Mike Snitzer498f0102011-08-02 12:32:04 +0100652 nr_args = 1 + nr_selector_args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 for (i = 0; i < pg->nr_pgpaths; i++) {
654 struct pgpath *pgpath;
Mike Snitzer498f0102011-08-02 12:32:04 +0100655 struct dm_arg_set path_args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Mike Snitzer498f0102011-08-02 12:32:04 +0100657 if (as->argc < nr_args) {
Mikulas Patocka148acff2008-07-21 12:00:30 +0100658 ti->error = "not enough path parameters";
Alasdair G Kergon6bbf79a2010-08-12 04:13:49 +0100659 r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 goto bad;
Mikulas Patocka148acff2008-07-21 12:00:30 +0100661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Mike Snitzer498f0102011-08-02 12:32:04 +0100663 path_args.argc = nr_args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 path_args.argv = as->argv;
665
666 pgpath = parse_path(&path_args, &pg->ps, ti);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100667 if (IS_ERR(pgpath)) {
668 r = PTR_ERR(pgpath);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 goto bad;
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
672 pgpath->pg = pg;
673 list_add_tail(&pgpath->list, &pg->pgpaths);
Mike Snitzer498f0102011-08-02 12:32:04 +0100674 dm_consume_args(as, nr_args);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 }
676
677 return pg;
678
679 bad:
680 free_priority_group(pg, ti);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100681 return ERR_PTR(r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682}
683
Mike Snitzer498f0102011-08-02 12:32:04 +0100684static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 unsigned hw_argc;
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700687 int ret;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700688 struct dm_target *ti = m->ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Mike Snitzer498f0102011-08-02 12:32:04 +0100690 static struct dm_arg _args[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700691 {0, 1024, "invalid number of hardware handler args"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 };
693
Mike Snitzer498f0102011-08-02 12:32:04 +0100694 if (dm_read_arg_group(_args, as, &hw_argc, &ti->error))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 return -EINVAL;
696
697 if (!hw_argc)
698 return 0;
699
Mike Snitzer498f0102011-08-02 12:32:04 +0100700 m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL);
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700701 request_module("scsi_dh_%s", m->hw_handler_name);
702 if (scsi_dh_handler_exist(m->hw_handler_name) == 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700703 ti->error = "unknown hardware handler type";
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700704 ret = -EINVAL;
705 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
Chandra Seetharaman14e98c52008-11-13 23:39:06 +0000707
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700708 if (hw_argc > 1) {
709 char *p;
710 int i, j, len = 4;
711
712 for (i = 0; i <= hw_argc - 2; i++)
713 len += strlen(as->argv[i]) + 1;
714 p = m->hw_handler_params = kzalloc(len, GFP_KERNEL);
715 if (!p) {
716 ti->error = "memory allocation failed";
717 ret = -ENOMEM;
718 goto fail;
719 }
720 j = sprintf(p, "%d", hw_argc - 1);
721 for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1)
722 j = sprintf(p, "%s", as->argv[i]);
723 }
Mike Snitzer498f0102011-08-02 12:32:04 +0100724 dm_consume_args(as, hw_argc - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726 return 0;
Chandra Seetharaman2bfd2e12009-08-03 12:42:45 -0700727fail:
728 kfree(m->hw_handler_name);
729 m->hw_handler_name = NULL;
730 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
732
Mike Snitzer498f0102011-08-02 12:32:04 +0100733static int parse_features(struct dm_arg_set *as, struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
735 int r;
736 unsigned argc;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700737 struct dm_target *ti = m->ti;
Mike Snitzer498f0102011-08-02 12:32:04 +0100738 const char *arg_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Mike Snitzer498f0102011-08-02 12:32:04 +0100740 static struct dm_arg _args[] = {
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000741 {0, 5, "invalid number of feature args"},
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100742 {1, 50, "pg_init_retries must be between 1 and 50"},
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000743 {0, 60000, "pg_init_delay_msecs must be between 0 and 60000"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 };
745
Mike Snitzer498f0102011-08-02 12:32:04 +0100746 r = dm_read_arg_group(_args, as, &argc, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 if (r)
748 return -EINVAL;
749
750 if (!argc)
751 return 0;
752
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100753 do {
Mike Snitzer498f0102011-08-02 12:32:04 +0100754 arg_name = dm_shift_arg(as);
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100755 argc--;
756
Mike Snitzer498f0102011-08-02 12:32:04 +0100757 if (!strcasecmp(arg_name, "queue_if_no_path")) {
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100758 r = queue_if_no_path(m, 1, 0);
759 continue;
760 }
761
Mike Snitzer498f0102011-08-02 12:32:04 +0100762 if (!strcasecmp(arg_name, "pg_init_retries") &&
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100763 (argc >= 1)) {
Mike Snitzer498f0102011-08-02 12:32:04 +0100764 r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error);
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100765 argc--;
766 continue;
767 }
768
Mike Snitzer498f0102011-08-02 12:32:04 +0100769 if (!strcasecmp(arg_name, "pg_init_delay_msecs") &&
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000770 (argc >= 1)) {
Mike Snitzer498f0102011-08-02 12:32:04 +0100771 r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error);
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000772 argc--;
773 continue;
774 }
775
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 ti->error = "Unrecognised multipath feature request";
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100777 r = -EINVAL;
778 } while (argc && !r);
779
780 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
783static int multipath_ctr(struct dm_target *ti, unsigned int argc,
784 char **argv)
785{
Mike Snitzer498f0102011-08-02 12:32:04 +0100786 /* target arguments */
787 static struct dm_arg _args[] = {
Mike Snitzera490a072011-03-24 13:54:33 +0000788 {0, 1024, "invalid number of priority groups"},
789 {0, 1024, "invalid initial priority group number"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 };
791
792 int r;
793 struct multipath *m;
Mike Snitzer498f0102011-08-02 12:32:04 +0100794 struct dm_arg_set as;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 unsigned pg_count = 0;
796 unsigned next_pg_num;
797
798 as.argc = argc;
799 as.argv = argv;
800
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700801 m = alloc_multipath(ti);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 if (!m) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700803 ti->error = "can't allocate multipath";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 return -EINVAL;
805 }
806
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700807 r = parse_features(&as, m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (r)
809 goto bad;
810
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700811 r = parse_hw_handler(&as, m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 if (r)
813 goto bad;
814
Mike Snitzer498f0102011-08-02 12:32:04 +0100815 r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 if (r)
817 goto bad;
818
Mike Snitzer498f0102011-08-02 12:32:04 +0100819 r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 if (r)
821 goto bad;
822
Mike Snitzera490a072011-03-24 13:54:33 +0000823 if ((!m->nr_priority_groups && next_pg_num) ||
824 (m->nr_priority_groups && !next_pg_num)) {
825 ti->error = "invalid initial priority group";
826 r = -EINVAL;
827 goto bad;
828 }
829
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 /* parse the priority groups */
831 while (as.argc) {
832 struct priority_group *pg;
833
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700834 pg = parse_priority_group(&as, m);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100835 if (IS_ERR(pg)) {
836 r = PTR_ERR(pg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 goto bad;
838 }
839
840 m->nr_valid_paths += pg->nr_pgpaths;
841 list_add_tail(&pg->list, &m->priority_groups);
842 pg_count++;
843 pg->pg_num = pg_count;
844 if (!--next_pg_num)
845 m->next_pg = pg;
846 }
847
848 if (pg_count != m->nr_priority_groups) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700849 ti->error = "priority group count mismatch";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 r = -EINVAL;
851 goto bad;
852 }
853
Mikulas Patocka86279212009-06-22 10:12:24 +0100854 ti->num_flush_requests = 1;
Mike Snitzer959eb4e2010-08-12 04:14:32 +0100855 ti->num_discard_requests = 1;
Mikulas Patocka86279212009-06-22 10:12:24 +0100856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 return 0;
858
859 bad:
860 free_multipath(m);
861 return r;
862}
863
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +0000864static void multipath_wait_for_pg_init_completion(struct multipath *m)
865{
866 DECLARE_WAITQUEUE(wait, current);
867 unsigned long flags;
868
869 add_wait_queue(&m->pg_init_wait, &wait);
870
871 while (1) {
872 set_current_state(TASK_UNINTERRUPTIBLE);
873
874 spin_lock_irqsave(&m->lock, flags);
875 if (!m->pg_init_in_progress) {
876 spin_unlock_irqrestore(&m->lock, flags);
877 break;
878 }
879 spin_unlock_irqrestore(&m->lock, flags);
880
881 io_schedule();
882 }
883 set_current_state(TASK_RUNNING);
884
885 remove_wait_queue(&m->pg_init_wait, &wait);
886}
887
888static void flush_multipath_work(struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -0700890 flush_workqueue(kmpath_handlerd);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +0000891 multipath_wait_for_pg_init_completion(m);
Alasdair G Kergona044d012005-07-12 15:53:02 -0700892 flush_workqueue(kmultipathd);
Tejun Heod5ffa382011-01-13 19:59:56 +0000893 flush_work_sync(&m->trigger_event);
Kiyoshi Ueda6df400a2009-12-10 23:52:19 +0000894}
895
896static void multipath_dtr(struct dm_target *ti)
897{
898 struct multipath *m = ti->private;
899
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +0000900 flush_multipath_work(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 free_multipath(m);
902}
903
904/*
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100905 * Map cloned requests
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 */
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100907static int multipath_map(struct dm_target *ti, struct request *clone,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 union map_info *map_context)
909{
910 int r;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100911 struct dm_mpath_io *mpio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 struct multipath *m = (struct multipath *) ti->private;
913
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100914 mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC);
915 if (!mpio)
916 /* ENOMEM, requeue */
917 return DM_MAPIO_REQUEUE;
918 memset(mpio, 0, sizeof(*mpio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 map_context->ptr = mpio;
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +0100921 clone->cmd_flags |= REQ_FAILFAST_TRANSPORT;
922 r = map_io(m, clone, mpio, 0);
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800923 if (r < 0 || r == DM_MAPIO_REQUEUE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 mempool_free(mpio, m->mpio_pool);
925
926 return r;
927}
928
929/*
930 * Take a path out of use.
931 */
932static int fail_path(struct pgpath *pgpath)
933{
934 unsigned long flags;
935 struct multipath *m = pgpath->pg->m;
936
937 spin_lock_irqsave(&m->lock, flags);
938
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100939 if (!pgpath->is_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 goto out;
941
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700942 DMWARN("Failing path %s.", pgpath->path.dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path);
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100945 pgpath->is_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 pgpath->fail_count++;
947
948 m->nr_valid_paths--;
949
950 if (pgpath == m->current_pgpath)
951 m->current_pgpath = NULL;
952
Mike Andersonb15546f2007-10-19 22:48:02 +0100953 dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti,
954 pgpath->path.dev->name, m->nr_valid_paths);
955
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +0000956 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958out:
959 spin_unlock_irqrestore(&m->lock, flags);
960
961 return 0;
962}
963
964/*
965 * Reinstate a previously-failed path
966 */
967static int reinstate_path(struct pgpath *pgpath)
968{
969 int r = 0;
970 unsigned long flags;
971 struct multipath *m = pgpath->pg->m;
972
973 spin_lock_irqsave(&m->lock, flags);
974
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100975 if (pgpath->is_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 goto out;
977
Alasdair G Kergondef052d2008-07-21 12:00:31 +0100978 if (!pgpath->pg->ps.type->reinstate_path) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 DMWARN("Reinstate path not supported by path selector %s",
980 pgpath->pg->ps.type->name);
981 r = -EINVAL;
982 goto out;
983 }
984
985 r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path);
986 if (r)
987 goto out;
988
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100989 pgpath->is_active = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100991 if (!m->nr_valid_paths++ && m->queue_size) {
992 m->current_pgpath = NULL;
Alasdair G Kergonc5573082005-05-05 16:16:07 -0700993 queue_work(kmultipathd, &m->process_queued_ios);
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100994 } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) {
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +0000995 if (queue_work(kmpath_handlerd, &pgpath->activate_path.work))
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100996 m->pg_init_in_progress++;
997 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Mike Andersonb15546f2007-10-19 22:48:02 +0100999 dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti,
1000 pgpath->path.dev->name, m->nr_valid_paths);
1001
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001002 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004out:
1005 spin_unlock_irqrestore(&m->lock, flags);
1006
1007 return r;
1008}
1009
1010/*
1011 * Fail or reinstate all paths that match the provided struct dm_dev.
1012 */
1013static int action_dev(struct multipath *m, struct dm_dev *dev,
1014 action_fn action)
1015{
Mike Snitzer19040c02011-03-24 13:54:31 +00001016 int r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 struct pgpath *pgpath;
1018 struct priority_group *pg;
1019
1020 list_for_each_entry(pg, &m->priority_groups, list) {
1021 list_for_each_entry(pgpath, &pg->pgpaths, list) {
1022 if (pgpath->path.dev == dev)
1023 r = action(pgpath);
1024 }
1025 }
1026
1027 return r;
1028}
1029
1030/*
1031 * Temporarily try to avoid having to use the specified PG
1032 */
1033static void bypass_pg(struct multipath *m, struct priority_group *pg,
1034 int bypassed)
1035{
1036 unsigned long flags;
1037
1038 spin_lock_irqsave(&m->lock, flags);
1039
1040 pg->bypassed = bypassed;
1041 m->current_pgpath = NULL;
1042 m->current_pg = NULL;
1043
1044 spin_unlock_irqrestore(&m->lock, flags);
1045
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001046 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047}
1048
1049/*
1050 * Switch to using the specified PG from the next I/O that gets mapped
1051 */
1052static int switch_pg_num(struct multipath *m, const char *pgstr)
1053{
1054 struct priority_group *pg;
1055 unsigned pgnum;
1056 unsigned long flags;
1057
1058 if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum ||
1059 (pgnum > m->nr_priority_groups)) {
1060 DMWARN("invalid PG number supplied to switch_pg_num");
1061 return -EINVAL;
1062 }
1063
1064 spin_lock_irqsave(&m->lock, flags);
1065 list_for_each_entry(pg, &m->priority_groups, list) {
1066 pg->bypassed = 0;
1067 if (--pgnum)
1068 continue;
1069
1070 m->current_pgpath = NULL;
1071 m->current_pg = NULL;
1072 m->next_pg = pg;
1073 }
1074 spin_unlock_irqrestore(&m->lock, flags);
1075
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001076 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 return 0;
1078}
1079
1080/*
1081 * Set/clear bypassed status of a PG.
1082 * PGs are numbered upwards from 1 in the order they were declared.
1083 */
1084static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed)
1085{
1086 struct priority_group *pg;
1087 unsigned pgnum;
1088
1089 if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum ||
1090 (pgnum > m->nr_priority_groups)) {
1091 DMWARN("invalid PG number supplied to bypass_pg");
1092 return -EINVAL;
1093 }
1094
1095 list_for_each_entry(pg, &m->priority_groups, list) {
1096 if (!--pgnum)
1097 break;
1098 }
1099
1100 bypass_pg(m, pg, bypassed);
1101 return 0;
1102}
1103
1104/*
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001105 * Should we retry pg_init immediately?
1106 */
1107static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
1108{
1109 unsigned long flags;
1110 int limit_reached = 0;
1111
1112 spin_lock_irqsave(&m->lock, flags);
1113
1114 if (m->pg_init_count <= m->pg_init_retries)
1115 m->pg_init_required = 1;
1116 else
1117 limit_reached = 1;
1118
1119 spin_unlock_irqrestore(&m->lock, flags);
1120
1121 return limit_reached;
1122}
1123
Chandra Seetharaman3ae31f62009-10-21 09:22:46 -07001124static void pg_init_done(void *data, int errors)
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001125{
Moger, Babu83c0d5d2010-03-06 02:29:45 +00001126 struct pgpath *pgpath = data;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001127 struct priority_group *pg = pgpath->pg;
1128 struct multipath *m = pg->m;
1129 unsigned long flags;
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001130 unsigned delay_retry = 0;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001131
1132 /* device or driver problems */
1133 switch (errors) {
1134 case SCSI_DH_OK:
1135 break;
1136 case SCSI_DH_NOSYS:
1137 if (!m->hw_handler_name) {
1138 errors = 0;
1139 break;
1140 }
Moger, Babuf7b934c2010-03-06 02:29:49 +00001141 DMERR("Could not failover the device: Handler scsi_dh_%s "
1142 "Error %d.", m->hw_handler_name, errors);
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001143 /*
1144 * Fail path for now, so we do not ping pong
1145 */
1146 fail_path(pgpath);
1147 break;
1148 case SCSI_DH_DEV_TEMP_BUSY:
1149 /*
1150 * Probably doing something like FW upgrade on the
1151 * controller so try the other pg.
1152 */
1153 bypass_pg(m, pg, 1);
1154 break;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001155 case SCSI_DH_RETRY:
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001156 /* Wait before retrying. */
1157 delay_retry = 1;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001158 case SCSI_DH_IMM_RETRY:
1159 case SCSI_DH_RES_TEMP_UNAVAIL:
1160 if (pg_init_limit_reached(m, pgpath))
1161 fail_path(pgpath);
1162 errors = 0;
1163 break;
1164 default:
1165 /*
1166 * We probably do not want to fail the path for a device
1167 * error, but this is what the old dm did. In future
1168 * patches we can do more advanced handling.
1169 */
1170 fail_path(pgpath);
1171 }
1172
1173 spin_lock_irqsave(&m->lock, flags);
1174 if (errors) {
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001175 if (pgpath == m->current_pgpath) {
1176 DMERR("Could not failover device. Error %d.", errors);
1177 m->current_pgpath = NULL;
1178 m->current_pg = NULL;
1179 }
Kiyoshi Uedad0259bf2010-03-06 02:30:02 +00001180 } else if (!m->pg_init_required)
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001181 pg->bypassed = 0;
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001182
Kiyoshi Uedad0259bf2010-03-06 02:30:02 +00001183 if (--m->pg_init_in_progress)
1184 /* Activations of other paths are still on going */
1185 goto out;
1186
1187 if (!m->pg_init_required)
1188 m->queue_io = 0;
1189
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001190 m->pg_init_delay_retry = delay_retry;
Kiyoshi Uedad0259bf2010-03-06 02:30:02 +00001191 queue_work(kmultipathd, &m->process_queued_ios);
1192
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001193 /*
1194 * Wake up any thread waiting to suspend.
1195 */
1196 wake_up(&m->pg_init_wait);
1197
Kiyoshi Uedad0259bf2010-03-06 02:30:02 +00001198out:
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001199 spin_unlock_irqrestore(&m->lock, flags);
1200}
1201
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001202static void activate_path(struct work_struct *work)
1203{
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001204 struct pgpath *pgpath =
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001205 container_of(work, struct pgpath, activate_path.work);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001206
Chandra Seetharaman3ae31f62009-10-21 09:22:46 -07001207 scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev),
Moger, Babu83c0d5d2010-03-06 02:29:45 +00001208 pg_init_done, pgpath);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001209}
1210
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211/*
1212 * end_io handling
1213 */
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001214static int do_end_io(struct multipath *m, struct request *clone,
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001215 int error, struct dm_mpath_io *mpio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216{
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001217 /*
1218 * We don't queue any clone request inside the multipath target
1219 * during end I/O handling, since those clone requests don't have
1220 * bio clones. If we queue them inside the multipath target,
1221 * we need to make bio clones, that requires memory allocation.
1222 * (See drivers/md/dm.c:end_clone_bio() about why the clone requests
1223 * don't have bio clones.)
1224 * Instead of queueing the clone request here, we queue the original
1225 * request into dm core, which will remake a clone request and
1226 * clone bios for it and resubmit it later.
1227 */
1228 int r = DM_ENDIO_REQUEUE;
Stefan Bader640eb3b2005-11-21 21:32:35 -08001229 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001231 if (!error && !clone->errors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 return 0; /* I/O complete */
1233
Martin K. Petersen6f13f6f2011-05-29 13:02:55 +01001234 if (error == -EOPNOTSUPP || error == -EREMOTEIO || error == -EILSEQ)
Mike Snitzer959eb4e2010-08-12 04:14:32 +01001235 return error;
1236
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001237 if (mpio->pgpath)
1238 fail_path(mpio->pgpath);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Stefan Bader640eb3b2005-11-21 21:32:35 -08001240 spin_lock_irqsave(&m->lock, flags);
Hannes Reinecke751b2a72011-01-18 10:13:12 +01001241 if (!m->nr_valid_paths) {
1242 if (!m->queue_if_no_path) {
1243 if (!__must_push_back(m))
1244 r = -EIO;
1245 } else {
1246 if (error == -EBADE)
1247 r = error;
1248 }
1249 }
Stefan Bader640eb3b2005-11-21 21:32:35 -08001250 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001252 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253}
1254
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001255static int multipath_end_io(struct dm_target *ti, struct request *clone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 int error, union map_info *map_context)
1257{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001258 struct multipath *m = ti->private;
1259 struct dm_mpath_io *mpio = map_context->ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 struct pgpath *pgpath = mpio->pgpath;
1261 struct path_selector *ps;
1262 int r;
1263
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001264 r = do_end_io(m, clone, error, mpio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 if (pgpath) {
1266 ps = &pgpath->pg->ps;
1267 if (ps->type->end_io)
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +01001268 ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 }
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001270 mempool_free(mpio, m->mpio_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
1272 return r;
1273}
1274
1275/*
1276 * Suspend can't complete until all the I/O is processed so if
Alasdair G Kergon436d4102005-07-12 15:53:03 -07001277 * the last path fails we must error any remaining I/O.
1278 * Note that if the freeze_bdev fails while suspending, the
1279 * queue_if_no_path state is lost - userspace should reset it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 */
1281static void multipath_presuspend(struct dm_target *ti)
1282{
1283 struct multipath *m = (struct multipath *) ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
Alasdair G Kergon485ef692005-09-27 21:45:45 -07001285 queue_if_no_path(m, 0, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286}
1287
Kiyoshi Ueda6df400a2009-12-10 23:52:19 +00001288static void multipath_postsuspend(struct dm_target *ti)
1289{
Mike Anderson6380f262009-12-10 23:52:21 +00001290 struct multipath *m = ti->private;
1291
1292 mutex_lock(&m->work_mutex);
Kiyoshi Ueda2bded7b2010-03-06 02:32:13 +00001293 flush_multipath_work(m);
Mike Anderson6380f262009-12-10 23:52:21 +00001294 mutex_unlock(&m->work_mutex);
Kiyoshi Ueda6df400a2009-12-10 23:52:19 +00001295}
1296
Alasdair G Kergon436d4102005-07-12 15:53:03 -07001297/*
1298 * Restore the queue_if_no_path setting.
1299 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300static void multipath_resume(struct dm_target *ti)
1301{
1302 struct multipath *m = (struct multipath *) ti->private;
1303 unsigned long flags;
1304
1305 spin_lock_irqsave(&m->lock, flags);
Alasdair G Kergon436d4102005-07-12 15:53:03 -07001306 m->queue_if_no_path = m->saved_queue_if_no_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 spin_unlock_irqrestore(&m->lock, flags);
1308}
1309
1310/*
1311 * Info output has the following format:
1312 * num_multipath_feature_args [multipath_feature_args]*
1313 * num_handler_status_args [handler_status_args]*
1314 * num_groups init_group_number
1315 * [A|D|E num_ps_status_args [ps_status_args]*
1316 * num_paths num_selector_args
1317 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1318 *
1319 * Table output has the following format (identical to the constructor string):
1320 * num_feature_args [features_args]*
1321 * num_handler_args hw_handler [hw_handler_args]*
1322 * num_groups init_group_number
1323 * [priority selector-name num_ps_args [ps_args]*
1324 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1325 */
1326static int multipath_status(struct dm_target *ti, status_type_t type,
1327 char *result, unsigned int maxlen)
1328{
1329 int sz = 0;
1330 unsigned long flags;
1331 struct multipath *m = (struct multipath *) ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 struct priority_group *pg;
1333 struct pgpath *p;
1334 unsigned pg_num;
1335 char state;
1336
1337 spin_lock_irqsave(&m->lock, flags);
1338
1339 /* Features */
1340 if (type == STATUSTYPE_INFO)
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001341 DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count);
1342 else {
1343 DMEMIT("%u ", m->queue_if_no_path +
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001344 (m->pg_init_retries > 0) * 2 +
1345 (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2);
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001346 if (m->queue_if_no_path)
1347 DMEMIT("queue_if_no_path ");
1348 if (m->pg_init_retries)
1349 DMEMIT("pg_init_retries %u ", m->pg_init_retries);
Chandra Seetharaman4e2d19e2011-01-13 20:00:01 +00001350 if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT)
1351 DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs);
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001354 if (!m->hw_handler_name || type == STATUSTYPE_INFO)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 DMEMIT("0 ");
1356 else
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001357 DMEMIT("1 %s ", m->hw_handler_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 DMEMIT("%u ", m->nr_priority_groups);
1360
1361 if (m->next_pg)
1362 pg_num = m->next_pg->pg_num;
1363 else if (m->current_pg)
1364 pg_num = m->current_pg->pg_num;
1365 else
Mike Snitzera490a072011-03-24 13:54:33 +00001366 pg_num = (m->nr_priority_groups ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
1368 DMEMIT("%u ", pg_num);
1369
1370 switch (type) {
1371 case STATUSTYPE_INFO:
1372 list_for_each_entry(pg, &m->priority_groups, list) {
1373 if (pg->bypassed)
1374 state = 'D'; /* Disabled */
1375 else if (pg == m->current_pg)
1376 state = 'A'; /* Currently Active */
1377 else
1378 state = 'E'; /* Enabled */
1379
1380 DMEMIT("%c ", state);
1381
1382 if (pg->ps.type->status)
1383 sz += pg->ps.type->status(&pg->ps, NULL, type,
1384 result + sz,
1385 maxlen - sz);
1386 else
1387 DMEMIT("0 ");
1388
1389 DMEMIT("%u %u ", pg->nr_pgpaths,
1390 pg->ps.type->info_args);
1391
1392 list_for_each_entry(p, &pg->pgpaths, list) {
1393 DMEMIT("%s %s %u ", p->path.dev->name,
Kiyoshi Ueda66800732008-10-10 13:36:58 +01001394 p->is_active ? "A" : "F",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 p->fail_count);
1396 if (pg->ps.type->status)
1397 sz += pg->ps.type->status(&pg->ps,
1398 &p->path, type, result + sz,
1399 maxlen - sz);
1400 }
1401 }
1402 break;
1403
1404 case STATUSTYPE_TABLE:
1405 list_for_each_entry(pg, &m->priority_groups, list) {
1406 DMEMIT("%s ", pg->ps.type->name);
1407
1408 if (pg->ps.type->status)
1409 sz += pg->ps.type->status(&pg->ps, NULL, type,
1410 result + sz,
1411 maxlen - sz);
1412 else
1413 DMEMIT("0 ");
1414
1415 DMEMIT("%u %u ", pg->nr_pgpaths,
1416 pg->ps.type->table_args);
1417
1418 list_for_each_entry(p, &pg->pgpaths, list) {
1419 DMEMIT("%s ", p->path.dev->name);
1420 if (pg->ps.type->status)
1421 sz += pg->ps.type->status(&pg->ps,
1422 &p->path, type, result + sz,
1423 maxlen - sz);
1424 }
1425 }
1426 break;
1427 }
1428
1429 spin_unlock_irqrestore(&m->lock, flags);
1430
1431 return 0;
1432}
1433
1434static int multipath_message(struct dm_target *ti, unsigned argc, char **argv)
1435{
Mike Anderson6380f262009-12-10 23:52:21 +00001436 int r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 struct dm_dev *dev;
1438 struct multipath *m = (struct multipath *) ti->private;
1439 action_fn action;
1440
Mike Anderson6380f262009-12-10 23:52:21 +00001441 mutex_lock(&m->work_mutex);
1442
Kiyoshi Uedac2f3d242009-12-10 23:52:27 +00001443 if (dm_suspended(ti)) {
1444 r = -EBUSY;
1445 goto out;
1446 }
1447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 if (argc == 1) {
Mike Snitzer498f0102011-08-02 12:32:04 +01001449 if (!strcasecmp(argv[0], "queue_if_no_path")) {
Mike Anderson6380f262009-12-10 23:52:21 +00001450 r = queue_if_no_path(m, 1, 0);
1451 goto out;
Mike Snitzer498f0102011-08-02 12:32:04 +01001452 } else if (!strcasecmp(argv[0], "fail_if_no_path")) {
Mike Anderson6380f262009-12-10 23:52:21 +00001453 r = queue_if_no_path(m, 0, 0);
1454 goto out;
1455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 }
1457
Mike Anderson6380f262009-12-10 23:52:21 +00001458 if (argc != 2) {
1459 DMWARN("Unrecognised multipath message received.");
1460 goto out;
1461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Mike Snitzer498f0102011-08-02 12:32:04 +01001463 if (!strcasecmp(argv[0], "disable_group")) {
Mike Anderson6380f262009-12-10 23:52:21 +00001464 r = bypass_pg_num(m, argv[1], 1);
1465 goto out;
Mike Snitzer498f0102011-08-02 12:32:04 +01001466 } else if (!strcasecmp(argv[0], "enable_group")) {
Mike Anderson6380f262009-12-10 23:52:21 +00001467 r = bypass_pg_num(m, argv[1], 0);
1468 goto out;
Mike Snitzer498f0102011-08-02 12:32:04 +01001469 } else if (!strcasecmp(argv[0], "switch_group")) {
Mike Anderson6380f262009-12-10 23:52:21 +00001470 r = switch_pg_num(m, argv[1]);
1471 goto out;
Mike Snitzer498f0102011-08-02 12:32:04 +01001472 } else if (!strcasecmp(argv[0], "reinstate_path"))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 action = reinstate_path;
Mike Snitzer498f0102011-08-02 12:32:04 +01001474 else if (!strcasecmp(argv[0], "fail_path"))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 action = fail_path;
Mike Anderson6380f262009-12-10 23:52:21 +00001476 else {
1477 DMWARN("Unrecognised multipath message received.");
1478 goto out;
1479 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +00001481 r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 if (r) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001483 DMWARN("message: error getting device %s",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 argv[1]);
Mike Anderson6380f262009-12-10 23:52:21 +00001485 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 }
1487
1488 r = action_dev(m, dev, action);
1489
1490 dm_put_device(ti, dev);
1491
Mike Anderson6380f262009-12-10 23:52:21 +00001492out:
1493 mutex_unlock(&m->work_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495}
1496
Al Viro647b3d02007-08-28 22:15:59 -04001497static int multipath_ioctl(struct dm_target *ti, unsigned int cmd,
Milan Broz9af4aa32006-10-03 01:15:20 -07001498 unsigned long arg)
1499{
1500 struct multipath *m = (struct multipath *) ti->private;
1501 struct block_device *bdev = NULL;
Al Viro633a08b2007-08-29 20:34:12 -04001502 fmode_t mode = 0;
Milan Broz9af4aa32006-10-03 01:15:20 -07001503 unsigned long flags;
1504 int r = 0;
1505
1506 spin_lock_irqsave(&m->lock, flags);
1507
1508 if (!m->current_pgpath)
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +01001509 __choose_pgpath(m, 0);
Milan Broz9af4aa32006-10-03 01:15:20 -07001510
Milan Broze90dae12006-10-03 01:15:22 -07001511 if (m->current_pgpath) {
Milan Broz9af4aa32006-10-03 01:15:20 -07001512 bdev = m->current_pgpath->path.dev->bdev;
Al Viro633a08b2007-08-29 20:34:12 -04001513 mode = m->current_pgpath->path.dev->mode;
Milan Broze90dae12006-10-03 01:15:22 -07001514 }
Milan Broz9af4aa32006-10-03 01:15:20 -07001515
1516 if (m->queue_io)
1517 r = -EAGAIN;
1518 else if (!bdev)
1519 r = -EIO;
1520
1521 spin_unlock_irqrestore(&m->lock, flags);
1522
Al Viro633a08b2007-08-29 20:34:12 -04001523 return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg);
Milan Broz9af4aa32006-10-03 01:15:20 -07001524}
1525
Mike Snitzeraf4874e2009-06-22 10:12:33 +01001526static int multipath_iterate_devices(struct dm_target *ti,
1527 iterate_devices_callout_fn fn, void *data)
1528{
1529 struct multipath *m = ti->private;
1530 struct priority_group *pg;
1531 struct pgpath *p;
1532 int ret = 0;
1533
1534 list_for_each_entry(pg, &m->priority_groups, list) {
1535 list_for_each_entry(p, &pg->pgpaths, list) {
Mike Snitzer5dea2712009-07-23 20:30:42 +01001536 ret = fn(ti, p->path.dev, ti->begin, ti->len, data);
Mike Snitzeraf4874e2009-06-22 10:12:33 +01001537 if (ret)
1538 goto out;
1539 }
1540 }
1541
1542out:
1543 return ret;
1544}
1545
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001546static int __pgpath_busy(struct pgpath *pgpath)
1547{
1548 struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev);
1549
1550 return dm_underlying_device_busy(q);
1551}
1552
1553/*
1554 * We return "busy", only when we can map I/Os but underlying devices
1555 * are busy (so even if we map I/Os now, the I/Os will wait on
1556 * the underlying queue).
1557 * In other words, if we want to kill I/Os or queue them inside us
1558 * due to map unavailability, we don't return "busy". Otherwise,
1559 * dm core won't give us the I/Os and we can't do what we want.
1560 */
1561static int multipath_busy(struct dm_target *ti)
1562{
1563 int busy = 0, has_active = 0;
1564 struct multipath *m = ti->private;
1565 struct priority_group *pg;
1566 struct pgpath *pgpath;
1567 unsigned long flags;
1568
1569 spin_lock_irqsave(&m->lock, flags);
1570
1571 /* Guess which priority_group will be used at next mapping time */
1572 if (unlikely(!m->current_pgpath && m->next_pg))
1573 pg = m->next_pg;
1574 else if (likely(m->current_pg))
1575 pg = m->current_pg;
1576 else
1577 /*
1578 * We don't know which pg will be used at next mapping time.
1579 * We don't call __choose_pgpath() here to avoid to trigger
1580 * pg_init just by busy checking.
1581 * So we don't know whether underlying devices we will be using
1582 * at next mapping time are busy or not. Just try mapping.
1583 */
1584 goto out;
1585
1586 /*
1587 * If there is one non-busy active path at least, the path selector
1588 * will be able to select it. So we consider such a pg as not busy.
1589 */
1590 busy = 1;
1591 list_for_each_entry(pgpath, &pg->pgpaths, list)
1592 if (pgpath->is_active) {
1593 has_active = 1;
1594
1595 if (!__pgpath_busy(pgpath)) {
1596 busy = 0;
1597 break;
1598 }
1599 }
1600
1601 if (!has_active)
1602 /*
1603 * No active path in this pg, so this pg won't be used and
1604 * the current_pg will be changed at next mapping time.
1605 * We need to try mapping to determine it.
1606 */
1607 busy = 0;
1608
1609out:
1610 spin_unlock_irqrestore(&m->lock, flags);
1611
1612 return busy;
1613}
1614
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615/*-----------------------------------------------------------------
1616 * Module setup
1617 *---------------------------------------------------------------*/
1618static struct target_type multipath_target = {
1619 .name = "multipath",
Mike Snitzer19040c02011-03-24 13:54:31 +00001620 .version = {1, 3, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 .module = THIS_MODULE,
1622 .ctr = multipath_ctr,
1623 .dtr = multipath_dtr,
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001624 .map_rq = multipath_map,
1625 .rq_end_io = multipath_end_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 .presuspend = multipath_presuspend,
Kiyoshi Ueda6df400a2009-12-10 23:52:19 +00001627 .postsuspend = multipath_postsuspend,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 .resume = multipath_resume,
1629 .status = multipath_status,
1630 .message = multipath_message,
Milan Broz9af4aa32006-10-03 01:15:20 -07001631 .ioctl = multipath_ioctl,
Mike Snitzeraf4874e2009-06-22 10:12:33 +01001632 .iterate_devices = multipath_iterate_devices,
Kiyoshi Uedaf40c67f2009-06-22 10:12:37 +01001633 .busy = multipath_busy,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634};
1635
1636static int __init dm_multipath_init(void)
1637{
1638 int r;
1639
1640 /* allocate a slab for the dm_ios */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001641 _mpio_cache = KMEM_CACHE(dm_mpath_io, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 if (!_mpio_cache)
1643 return -ENOMEM;
1644
1645 r = dm_register_target(&multipath_target);
1646 if (r < 0) {
Alasdair G Kergon0cd33122007-07-12 17:27:01 +01001647 DMERR("register failed %d", r);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 kmem_cache_destroy(_mpio_cache);
1649 return -EINVAL;
1650 }
1651
Tejun Heo4d4d66a2011-01-13 19:59:57 +00001652 kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0);
Alasdair G Kergonc5573082005-05-05 16:16:07 -07001653 if (!kmultipathd) {
Alasdair G Kergon0cd33122007-07-12 17:27:01 +01001654 DMERR("failed to create workqueue kmpathd");
Alasdair G Kergonc5573082005-05-05 16:16:07 -07001655 dm_unregister_target(&multipath_target);
1656 kmem_cache_destroy(_mpio_cache);
1657 return -ENOMEM;
1658 }
1659
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001660 /*
1661 * A separate workqueue is used to handle the device handlers
1662 * to avoid overloading existing workqueue. Overloading the
1663 * old workqueue would also create a bottleneck in the
1664 * path of the storage hardware device activation.
1665 */
Tejun Heo4d4d66a2011-01-13 19:59:57 +00001666 kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd",
1667 WQ_MEM_RECLAIM);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001668 if (!kmpath_handlerd) {
1669 DMERR("failed to create workqueue kmpath_handlerd");
1670 destroy_workqueue(kmultipathd);
1671 dm_unregister_target(&multipath_target);
1672 kmem_cache_destroy(_mpio_cache);
1673 return -ENOMEM;
1674 }
1675
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001676 DMINFO("version %u.%u.%u loaded",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 multipath_target.version[0], multipath_target.version[1],
1678 multipath_target.version[2]);
1679
1680 return r;
1681}
1682
1683static void __exit dm_multipath_exit(void)
1684{
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001685 destroy_workqueue(kmpath_handlerd);
Alasdair G Kergonc5573082005-05-05 16:16:07 -07001686 destroy_workqueue(kmultipathd);
1687
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00001688 dm_unregister_target(&multipath_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 kmem_cache_destroy(_mpio_cache);
1690}
1691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692module_init(dm_multipath_init);
1693module_exit(dm_multipath_exit);
1694
1695MODULE_DESCRIPTION(DM_NAME " multipath target");
1696MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
1697MODULE_LICENSE("GPL");