blob: 890c0e8ed13eaae7d78c75f81eb3503314d1dac2 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include "dm-bio-record.h"
Mike Andersonb15546f2007-10-19 22:48:02 +010012#include "dm-uevent.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070013
14#include <linux/ctype.h>
15#include <linux/init.h>
16#include <linux/mempool.h>
17#include <linux/module.h>
18#include <linux/pagemap.h>
19#include <linux/slab.h>
20#include <linux/time.h>
21#include <linux/workqueue.h>
Chandra Seetharamancfae5c92008-05-01 14:50:11 -070022#include <scsi/scsi_dh.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <asm/atomic.h>
24
Alasdair G Kergon72d94862006-06-26 00:27:35 -070025#define DM_MSG_PREFIX "multipath"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#define MESG_STR(x) x, sizeof(x)
27
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;
Mike Anderson224cb3e2008-08-29 09:36:09 +020037 struct work_struct deactivate_path;
Chandra Seetharamane54f77d2009-06-22 10:12:12 +010038 struct work_struct activate_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039};
40
41#define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path)
42
43/*
44 * Paths are grouped into Priority Groups and numbered from 1 upwards.
45 * Each has a path selector which controls which path gets used.
46 */
47struct priority_group {
48 struct list_head list;
49
50 struct multipath *m; /* Owning multipath instance */
51 struct path_selector ps;
52
53 unsigned pg_num; /* Reference number */
54 unsigned bypassed; /* Temporarily bypass this PG? */
55
56 unsigned nr_pgpaths; /* Number of paths in PG */
57 struct list_head pgpaths;
58};
59
60/* Multipath context */
61struct multipath {
62 struct list_head list;
63 struct dm_target *ti;
64
65 spinlock_t lock;
66
Chandra Seetharamancfae5c92008-05-01 14:50:11 -070067 const char *hw_handler_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 unsigned nr_priority_groups;
69 struct list_head priority_groups;
70 unsigned pg_init_required; /* pg_init needs calling? */
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -070071 unsigned pg_init_in_progress; /* Only one pg_init allowed at once */
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73 unsigned nr_valid_paths; /* Total number of usable paths */
74 struct pgpath *current_pgpath;
75 struct priority_group *current_pg;
76 struct priority_group *next_pg; /* Switch to this PG if set */
77 unsigned repeat_count; /* I/Os left before calling PS again */
78
79 unsigned queue_io; /* Must we queue all I/O? */
80 unsigned queue_if_no_path; /* Queue I/O if last path fails? */
Alasdair G Kergon436d4102005-07-12 15:53:03 -070081 unsigned saved_queue_if_no_path;/* Saved state during suspension */
Dave Wysochanskic9e45582007-10-19 22:47:53 +010082 unsigned pg_init_retries; /* Number of times to retry pg_init */
83 unsigned pg_init_count; /* Number of times pg_init called */
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85 struct work_struct process_queued_ios;
86 struct bio_list queued_ios;
87 unsigned queue_size;
88
89 struct work_struct trigger_event;
90
91 /*
Alasdair G Kergon028867a2007-07-12 17:26:32 +010092 * We must use a mempool of dm_mpath_io structs so that we
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 * can resubmit bios on error.
94 */
95 mempool_t *mpio_pool;
96};
97
98/*
99 * Context information attached to each bio we process.
100 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100101struct dm_mpath_io {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 struct pgpath *pgpath;
103 struct dm_bio_details details;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100104 size_t nr_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105};
106
107typedef int (*action_fn) (struct pgpath *pgpath);
108
109#define MIN_IOS 256 /* Mempool size */
110
Christoph Lametere18b8902006-12-06 20:33:20 -0800111static struct kmem_cache *_mpio_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -0700113static struct workqueue_struct *kmultipathd, *kmpath_handlerd;
David Howellsc4028952006-11-22 14:57:56 +0000114static void process_queued_ios(struct work_struct *work);
115static void trigger_event(struct work_struct *work);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -0700116static void activate_path(struct work_struct *work);
Mike Anderson224cb3e2008-08-29 09:36:09 +0200117static void deactivate_path(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
119
120/*-----------------------------------------------
121 * Allocation routines
122 *-----------------------------------------------*/
123
124static struct pgpath *alloc_pgpath(void)
125{
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700126 struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Mike Anderson224cb3e2008-08-29 09:36:09 +0200128 if (pgpath) {
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100129 pgpath->is_active = 1;
Mike Anderson224cb3e2008-08-29 09:36:09 +0200130 INIT_WORK(&pgpath->deactivate_path, deactivate_path);
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100131 INIT_WORK(&pgpath->activate_path, activate_path);
Mike Anderson224cb3e2008-08-29 09:36:09 +0200132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
134 return pgpath;
135}
136
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100137static void free_pgpath(struct pgpath *pgpath)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
139 kfree(pgpath);
140}
141
Mike Anderson224cb3e2008-08-29 09:36:09 +0200142static void deactivate_path(struct work_struct *work)
143{
144 struct pgpath *pgpath =
145 container_of(work, struct pgpath, deactivate_path);
146
147 blk_abort_queue(pgpath->path.dev->bdev->bd_disk->queue);
148}
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150static struct priority_group *alloc_priority_group(void)
151{
152 struct priority_group *pg;
153
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700154 pg = kzalloc(sizeof(*pg), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700156 if (pg)
157 INIT_LIST_HEAD(&pg->pgpaths);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159 return pg;
160}
161
162static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti)
163{
164 struct pgpath *pgpath, *tmp;
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700165 struct multipath *m = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 list_for_each_entry_safe(pgpath, tmp, pgpaths, list) {
168 list_del(&pgpath->list);
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700169 if (m->hw_handler_name)
170 scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 dm_put_device(ti, pgpath->path.dev);
172 free_pgpath(pgpath);
173 }
174}
175
176static void free_priority_group(struct priority_group *pg,
177 struct dm_target *ti)
178{
179 struct path_selector *ps = &pg->ps;
180
181 if (ps->type) {
182 ps->type->destroy(ps);
183 dm_put_path_selector(ps->type);
184 }
185
186 free_pgpaths(&pg->pgpaths, ti);
187 kfree(pg);
188}
189
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700190static struct multipath *alloc_multipath(struct dm_target *ti)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
192 struct multipath *m;
193
Micha³ Miros³awe69fae52006-10-03 01:15:34 -0700194 m = kzalloc(sizeof(*m), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 if (m) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 INIT_LIST_HEAD(&m->priority_groups);
197 spin_lock_init(&m->lock);
198 m->queue_io = 1;
David Howellsc4028952006-11-22 14:57:56 +0000199 INIT_WORK(&m->process_queued_ios, process_queued_ios);
200 INIT_WORK(&m->trigger_event, trigger_event);
Matthew Dobson93d23412006-03-26 01:37:50 -0800201 m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 if (!m->mpio_pool) {
203 kfree(m);
204 return NULL;
205 }
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700206 m->ti = ti;
207 ti->private = m;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 }
209
210 return m;
211}
212
213static void free_multipath(struct multipath *m)
214{
215 struct priority_group *pg, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
217 list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) {
218 list_del(&pg->list);
219 free_priority_group(pg, m->ti);
220 }
221
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700222 kfree(m->hw_handler_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 mempool_destroy(m->mpio_pool);
224 kfree(m);
225}
226
227
228/*-----------------------------------------------
229 * Path selection
230 *-----------------------------------------------*/
231
232static void __switch_pg(struct multipath *m, struct pgpath *pgpath)
233{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 m->current_pg = pgpath->pg;
235
236 /* Must we initialise the PG first, and queue I/O till it's ready? */
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700237 if (m->hw_handler_name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 m->pg_init_required = 1;
239 m->queue_io = 1;
240 } else {
241 m->pg_init_required = 0;
242 m->queue_io = 0;
243 }
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100244
245 m->pg_init_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100248static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg,
249 size_t nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Josef "Jeff" Sipekc922d5f2006-12-08 02:36:33 -0800251 struct dm_path *path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100253 path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (!path)
255 return -ENXIO;
256
257 m->current_pgpath = path_to_pgpath(path);
258
259 if (m->current_pg != pg)
260 __switch_pg(m, m->current_pgpath);
261
262 return 0;
263}
264
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100265static void __choose_pgpath(struct multipath *m, size_t nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
267 struct priority_group *pg;
268 unsigned bypassed = 1;
269
270 if (!m->nr_valid_paths)
271 goto failed;
272
273 /* Were we instructed to switch PG? */
274 if (m->next_pg) {
275 pg = m->next_pg;
276 m->next_pg = NULL;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100277 if (!__choose_path_in_pg(m, pg, nr_bytes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 return;
279 }
280
281 /* Don't change PG until it has no remaining paths */
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100282 if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 return;
284
285 /*
286 * Loop through priority groups until we find a valid path.
287 * First time we skip PGs marked 'bypassed'.
288 * Second time we only try the ones we skipped.
289 */
290 do {
291 list_for_each_entry(pg, &m->priority_groups, list) {
292 if (pg->bypassed == bypassed)
293 continue;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100294 if (!__choose_path_in_pg(m, pg, nr_bytes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 return;
296 }
297 } while (bypassed--);
298
299failed:
300 m->current_pgpath = NULL;
301 m->current_pg = NULL;
302}
303
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800304/*
305 * Check whether bios must be queued in the device-mapper core rather
306 * than here in the target.
307 *
308 * m->lock must be held on entry.
309 *
310 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
311 * same value then we are not between multipath_presuspend()
312 * and multipath_resume() calls and we have no need to check
313 * for the DMF_NOFLUSH_SUSPENDING flag.
314 */
315static int __must_push_back(struct multipath *m)
316{
317 return (m->queue_if_no_path != m->saved_queue_if_no_path &&
318 dm_noflush_suspending(m->ti));
319}
320
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100321static int map_io(struct multipath *m, struct bio *bio,
322 struct dm_mpath_io *mpio, unsigned was_queued)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323{
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -0800324 int r = DM_MAPIO_REMAPPED;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100325 size_t nr_bytes = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 unsigned long flags;
327 struct pgpath *pgpath;
328
329 spin_lock_irqsave(&m->lock, flags);
330
331 /* Do we need to select a new pgpath? */
332 if (!m->current_pgpath ||
333 (!m->queue_io && (m->repeat_count && --m->repeat_count == 0)))
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100334 __choose_pgpath(m, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 pgpath = m->current_pgpath;
337
338 if (was_queued)
339 m->queue_size--;
340
341 if ((pgpath && m->queue_io) ||
Alasdair G Kergon436d4102005-07-12 15:53:03 -0700342 (!pgpath && m->queue_if_no_path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 /* Queue for the daemon to resubmit */
344 bio_list_add(&m->queued_ios, bio);
345 m->queue_size++;
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700346 if ((m->pg_init_required && !m->pg_init_in_progress) ||
347 !m->queue_io)
Alasdair G Kergonc5573082005-05-05 16:16:07 -0700348 queue_work(kmultipathd, &m->process_queued_ios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 pgpath = NULL;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -0800350 r = DM_MAPIO_SUBMITTED;
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800351 } else if (pgpath)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 bio->bi_bdev = pgpath->path.dev->bdev;
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800353 else if (__must_push_back(m))
354 r = DM_MAPIO_REQUEUE;
355 else
356 r = -EIO; /* Failed */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358 mpio->pgpath = pgpath;
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100359 mpio->nr_bytes = nr_bytes;
360
361 if (r == DM_MAPIO_REMAPPED && pgpath->pg->ps.type->start_io)
362 pgpath->pg->ps.type->start_io(&pgpath->pg->ps, &pgpath->path,
363 nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 spin_unlock_irqrestore(&m->lock, flags);
366
367 return r;
368}
369
370/*
371 * If we run out of usable paths, should we queue I/O or error it?
372 */
Alasdair G Kergon485ef692005-09-27 21:45:45 -0700373static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path,
374 unsigned save_old_value)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
376 unsigned long flags;
377
378 spin_lock_irqsave(&m->lock, flags);
379
Alasdair G Kergon485ef692005-09-27 21:45:45 -0700380 if (save_old_value)
381 m->saved_queue_if_no_path = m->queue_if_no_path;
382 else
383 m->saved_queue_if_no_path = queue_if_no_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 m->queue_if_no_path = queue_if_no_path;
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700385 if (!m->queue_if_no_path && m->queue_size)
Alasdair G Kergonc5573082005-05-05 16:16:07 -0700386 queue_work(kmultipathd, &m->process_queued_ios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
388 spin_unlock_irqrestore(&m->lock, flags);
389
390 return 0;
391}
392
393/*-----------------------------------------------------------------
394 * The multipath daemon is responsible for resubmitting queued ios.
395 *---------------------------------------------------------------*/
396
397static void dispatch_queued_ios(struct multipath *m)
398{
399 int r;
400 unsigned long flags;
401 struct bio *bio = NULL, *next;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100402 struct dm_mpath_io *mpio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 union map_info *info;
404
405 spin_lock_irqsave(&m->lock, flags);
406 bio = bio_list_get(&m->queued_ios);
407 spin_unlock_irqrestore(&m->lock, flags);
408
409 while (bio) {
410 next = bio->bi_next;
411 bio->bi_next = NULL;
412
413 info = dm_get_mapinfo(bio);
414 mpio = info->ptr;
415
416 r = map_io(m, bio, mpio, 1);
417 if (r < 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200418 bio_endio(bio, r);
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -0800419 else if (r == DM_MAPIO_REMAPPED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 generic_make_request(bio);
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800421 else if (r == DM_MAPIO_REQUEUE)
NeilBrown6712ecf2007-09-27 12:47:43 +0200422 bio_endio(bio, -EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
424 bio = next;
425 }
426}
427
David Howellsc4028952006-11-22 14:57:56 +0000428static void process_queued_ios(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
David Howellsc4028952006-11-22 14:57:56 +0000430 struct multipath *m =
431 container_of(work, struct multipath, process_queued_ios);
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100432 struct pgpath *pgpath = NULL, *tmp;
433 unsigned must_queue = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 unsigned long flags;
435
436 spin_lock_irqsave(&m->lock, flags);
437
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700438 if (!m->queue_size)
439 goto out;
440
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 if (!m->current_pgpath)
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +0100442 __choose_pgpath(m, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
444 pgpath = m->current_pgpath;
445
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700446 if ((pgpath && !m->queue_io) ||
447 (!pgpath && !m->queue_if_no_path))
448 must_queue = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Chandra Seetharamanb81aa1c2008-11-13 23:39:00 +0000450 if (m->pg_init_required && !m->pg_init_in_progress && pgpath) {
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100451 m->pg_init_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 m->pg_init_required = 0;
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100453 list_for_each_entry(tmp, &pgpath->pg->pgpaths, list) {
454 if (queue_work(kmpath_handlerd, &tmp->activate_path))
455 m->pg_init_in_progress++;
456 }
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700457 }
Alasdair G Kergonc3cd4f62005-07-12 15:53:04 -0700458out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (!must_queue)
461 dispatch_queued_ios(m);
462}
463
464/*
465 * An event is triggered whenever a path is taken out of use.
466 * Includes path failure and PG bypass.
467 */
David Howellsc4028952006-11-22 14:57:56 +0000468static void trigger_event(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
David Howellsc4028952006-11-22 14:57:56 +0000470 struct multipath *m =
471 container_of(work, struct multipath, trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 dm_table_event(m->ti->table);
474}
475
476/*-----------------------------------------------------------------
477 * Constructor/argument parsing:
478 * <#multipath feature args> [<arg>]*
479 * <#hw_handler args> [hw_handler [<arg>]*]
480 * <#priority groups>
481 * <initial priority group>
482 * [<selector> <#selector args> [<arg>]*
483 * <#paths> <#per-path selector args>
484 * [<path> [<arg>]* ]+ ]+
485 *---------------------------------------------------------------*/
486struct param {
487 unsigned min;
488 unsigned max;
489 char *error;
490};
491
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492static int read_param(struct param *param, char *str, unsigned *v, char **error)
493{
494 if (!str ||
495 (sscanf(str, "%u", v) != 1) ||
496 (*v < param->min) ||
497 (*v > param->max)) {
498 *error = param->error;
499 return -EINVAL;
500 }
501
502 return 0;
503}
504
505struct arg_set {
506 unsigned argc;
507 char **argv;
508};
509
510static char *shift(struct arg_set *as)
511{
512 char *r;
513
514 if (as->argc) {
515 as->argc--;
516 r = *as->argv;
517 as->argv++;
518 return r;
519 }
520
521 return NULL;
522}
523
524static void consume(struct arg_set *as, unsigned n)
525{
526 BUG_ON (as->argc < n);
527 as->argc -= n;
528 as->argv += n;
529}
530
531static int parse_path_selector(struct arg_set *as, struct priority_group *pg,
532 struct dm_target *ti)
533{
534 int r;
535 struct path_selector_type *pst;
536 unsigned ps_argc;
537
538 static struct param _params[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700539 {0, 1024, "invalid number of path selector args"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 };
541
542 pst = dm_get_path_selector(shift(as));
543 if (!pst) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700544 ti->error = "unknown path selector type";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 return -EINVAL;
546 }
547
548 r = read_param(_params, shift(as), &ps_argc, &ti->error);
Mikulas Patocka371b2e32008-07-21 12:00:24 +0100549 if (r) {
550 dm_put_path_selector(pst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 return -EINVAL;
Mikulas Patocka371b2e32008-07-21 12:00:24 +0100552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Mikulas Patocka0e0497c2009-06-22 10:08:02 +0100554 if (ps_argc > as->argc) {
555 dm_put_path_selector(pst);
556 ti->error = "not enough arguments for path selector";
557 return -EINVAL;
558 }
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 r = pst->create(&pg->ps, ps_argc, as->argv);
561 if (r) {
562 dm_put_path_selector(pst);
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700563 ti->error = "path selector constructor failed";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return r;
565 }
566
567 pg->ps.type = pst;
568 consume(as, ps_argc);
569
570 return 0;
571}
572
573static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps,
574 struct dm_target *ti)
575{
576 int r;
577 struct pgpath *p;
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700578 struct multipath *m = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 /* we need at least a path arg */
581 if (as->argc < 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700582 ti->error = "no device given";
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100583 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 }
585
586 p = alloc_pgpath();
587 if (!p)
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100588 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 r = dm_get_device(ti, shift(as), ti->begin, ti->len,
591 dm_table_get_mode(ti->table), &p->path.dev);
592 if (r) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700593 ti->error = "error getting device";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 goto bad;
595 }
596
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700597 if (m->hw_handler_name) {
Hannes Reineckea0cf7ea2009-06-22 10:12:11 +0100598 struct request_queue *q = bdev_get_queue(p->path.dev->bdev);
599
600 r = scsi_dh_attach(q, m->hw_handler_name);
601 if (r == -EBUSY) {
602 /*
603 * Already attached to different hw_handler,
604 * try to reattach with correct one.
605 */
606 scsi_dh_detach(q);
607 r = scsi_dh_attach(q, m->hw_handler_name);
608 }
609
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700610 if (r < 0) {
Hannes Reineckea0cf7ea2009-06-22 10:12:11 +0100611 ti->error = "error attaching hardware handler";
Hannes Reineckeae11b1b2008-07-17 17:49:02 -0700612 dm_put_device(ti, p->path.dev);
613 goto bad;
614 }
615 }
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error);
618 if (r) {
619 dm_put_device(ti, p->path.dev);
620 goto bad;
621 }
622
623 return p;
624
625 bad:
626 free_pgpath(p);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100627 return ERR_PTR(r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629
630static struct priority_group *parse_priority_group(struct arg_set *as,
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700631 struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 static struct param _params[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700634 {1, 1024, "invalid number of paths"},
635 {0, 1024, "invalid number of selector args"}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 };
637
638 int r;
639 unsigned i, nr_selector_args, nr_params;
640 struct priority_group *pg;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700641 struct dm_target *ti = m->ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643 if (as->argc < 2) {
644 as->argc = 0;
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100645 ti->error = "not enough priority group arguments";
646 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 }
648
649 pg = alloc_priority_group();
650 if (!pg) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700651 ti->error = "couldn't allocate priority group";
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100652 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
654 pg->m = m;
655
656 r = parse_path_selector(as, pg, ti);
657 if (r)
658 goto bad;
659
660 /*
661 * read the paths
662 */
663 r = read_param(_params, shift(as), &pg->nr_pgpaths, &ti->error);
664 if (r)
665 goto bad;
666
667 r = read_param(_params + 1, shift(as), &nr_selector_args, &ti->error);
668 if (r)
669 goto bad;
670
671 nr_params = 1 + nr_selector_args;
672 for (i = 0; i < pg->nr_pgpaths; i++) {
673 struct pgpath *pgpath;
674 struct arg_set path_args;
675
Mikulas Patocka148acff2008-07-21 12:00:30 +0100676 if (as->argc < nr_params) {
677 ti->error = "not enough path parameters";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 goto bad;
Mikulas Patocka148acff2008-07-21 12:00:30 +0100679 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
681 path_args.argc = nr_params;
682 path_args.argv = as->argv;
683
684 pgpath = parse_path(&path_args, &pg->ps, ti);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100685 if (IS_ERR(pgpath)) {
686 r = PTR_ERR(pgpath);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 goto bad;
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
690 pgpath->pg = pg;
691 list_add_tail(&pgpath->list, &pg->pgpaths);
692 consume(as, nr_params);
693 }
694
695 return pg;
696
697 bad:
698 free_priority_group(pg, ti);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100699 return ERR_PTR(r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700}
701
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700702static int parse_hw_handler(struct arg_set *as, struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 unsigned hw_argc;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700705 struct dm_target *ti = m->ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 static struct param _params[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700708 {0, 1024, "invalid number of hardware handler args"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 };
710
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700711 if (read_param(_params, shift(as), &hw_argc, &ti->error))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 return -EINVAL;
713
714 if (!hw_argc)
715 return 0;
716
Mikulas Patockae094f4f2009-06-22 10:12:10 +0100717 if (hw_argc > as->argc) {
718 ti->error = "not enough arguments for hardware handler";
719 return -EINVAL;
720 }
721
Chandra Seetharamancfae5c92008-05-01 14:50:11 -0700722 m->hw_handler_name = kstrdup(shift(as), GFP_KERNEL);
723 request_module("scsi_dh_%s", m->hw_handler_name);
724 if (scsi_dh_handler_exist(m->hw_handler_name) == 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700725 ti->error = "unknown hardware handler type";
Chandra Seetharamanfe9233f2008-05-23 18:16:40 -0700726 kfree(m->hw_handler_name);
727 m->hw_handler_name = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 return -EINVAL;
729 }
Chandra Seetharaman14e98c52008-11-13 23:39:06 +0000730
731 if (hw_argc > 1)
732 DMWARN("Ignoring user-specified arguments for "
733 "hardware handler \"%s\"", m->hw_handler_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 consume(as, hw_argc - 1);
735
736 return 0;
737}
738
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700739static int parse_features(struct arg_set *as, struct multipath *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
741 int r;
742 unsigned argc;
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700743 struct dm_target *ti = m->ti;
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100744 const char *param_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
746 static struct param _params[] = {
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100747 {0, 3, "invalid number of feature args"},
748 {1, 50, "pg_init_retries must be between 1 and 50"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 };
750
751 r = read_param(_params, shift(as), &argc, &ti->error);
752 if (r)
753 return -EINVAL;
754
755 if (!argc)
756 return 0;
757
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100758 do {
759 param_name = shift(as);
760 argc--;
761
762 if (!strnicmp(param_name, MESG_STR("queue_if_no_path"))) {
763 r = queue_if_no_path(m, 1, 0);
764 continue;
765 }
766
767 if (!strnicmp(param_name, MESG_STR("pg_init_retries")) &&
768 (argc >= 1)) {
769 r = read_param(_params + 1, shift(as),
770 &m->pg_init_retries, &ti->error);
771 argc--;
772 continue;
773 }
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 ti->error = "Unrecognised multipath feature request";
Dave Wysochanskic9e45582007-10-19 22:47:53 +0100776 r = -EINVAL;
777 } while (argc && !r);
778
779 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780}
781
782static int multipath_ctr(struct dm_target *ti, unsigned int argc,
783 char **argv)
784{
785 /* target parameters */
786 static struct param _params[] = {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700787 {1, 1024, "invalid number of priority groups"},
788 {1, 1024, "invalid initial priority group number"},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 };
790
791 int r;
792 struct multipath *m;
793 struct arg_set as;
794 unsigned pg_count = 0;
795 unsigned next_pg_num;
796
797 as.argc = argc;
798 as.argv = argv;
799
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700800 m = alloc_multipath(ti);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 if (!m) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700802 ti->error = "can't allocate multipath";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 return -EINVAL;
804 }
805
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700806 r = parse_features(&as, m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 if (r)
808 goto bad;
809
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700810 r = parse_hw_handler(&as, m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 if (r)
812 goto bad;
813
814 r = read_param(_params, shift(&as), &m->nr_priority_groups, &ti->error);
815 if (r)
816 goto bad;
817
818 r = read_param(_params + 1, shift(&as), &next_pg_num, &ti->error);
819 if (r)
820 goto bad;
821
822 /* parse the priority groups */
823 while (as.argc) {
824 struct priority_group *pg;
825
Micha³ Miros³aw28f16c22006-10-03 01:15:33 -0700826 pg = parse_priority_group(&as, m);
Benjamin Marzinski01460f32008-10-10 13:36:57 +0100827 if (IS_ERR(pg)) {
828 r = PTR_ERR(pg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 goto bad;
830 }
831
832 m->nr_valid_paths += pg->nr_pgpaths;
833 list_add_tail(&pg->list, &m->priority_groups);
834 pg_count++;
835 pg->pg_num = pg_count;
836 if (!--next_pg_num)
837 m->next_pg = pg;
838 }
839
840 if (pg_count != m->nr_priority_groups) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700841 ti->error = "priority group count mismatch";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 r = -EINVAL;
843 goto bad;
844 }
845
Mikulas Patocka86279212009-06-22 10:12:24 +0100846 ti->num_flush_requests = 1;
847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 return 0;
849
850 bad:
851 free_multipath(m);
852 return r;
853}
854
855static void multipath_dtr(struct dm_target *ti)
856{
857 struct multipath *m = (struct multipath *) ti->private;
Alasdair G Kergona044d012005-07-12 15:53:02 -0700858
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -0700859 flush_workqueue(kmpath_handlerd);
Alasdair G Kergona044d012005-07-12 15:53:02 -0700860 flush_workqueue(kmultipathd);
Mikulas Patocka53b351f2009-06-22 10:12:13 +0100861 flush_scheduled_work();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 free_multipath(m);
863}
864
865/*
866 * Map bios, recording original fields for later in case we have to resubmit
867 */
868static int multipath_map(struct dm_target *ti, struct bio *bio,
869 union map_info *map_context)
870{
871 int r;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100872 struct dm_mpath_io *mpio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 struct multipath *m = (struct multipath *) ti->private;
874
875 mpio = mempool_alloc(m->mpio_pool, GFP_NOIO);
876 dm_bio_record(&mpio->details, bio);
877
878 map_context->ptr = mpio;
Mike Christie6000a362008-08-19 18:45:30 -0500879 bio->bi_rw |= (1 << BIO_RW_FAILFAST_TRANSPORT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 r = map_io(m, bio, mpio, 0);
Kiyoshi Ueda45e15722006-12-08 02:41:10 -0800881 if (r < 0 || r == DM_MAPIO_REQUEUE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 mempool_free(mpio, m->mpio_pool);
883
884 return r;
885}
886
887/*
888 * Take a path out of use.
889 */
890static int fail_path(struct pgpath *pgpath)
891{
892 unsigned long flags;
893 struct multipath *m = pgpath->pg->m;
894
895 spin_lock_irqsave(&m->lock, flags);
896
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100897 if (!pgpath->is_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 goto out;
899
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700900 DMWARN("Failing path %s.", pgpath->path.dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path);
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100903 pgpath->is_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 pgpath->fail_count++;
905
906 m->nr_valid_paths--;
907
908 if (pgpath == m->current_pgpath)
909 m->current_pgpath = NULL;
910
Mike Andersonb15546f2007-10-19 22:48:02 +0100911 dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti,
912 pgpath->path.dev->name, m->nr_valid_paths);
913
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +0000914 schedule_work(&m->trigger_event);
Mike Anderson224cb3e2008-08-29 09:36:09 +0200915 queue_work(kmultipathd, &pgpath->deactivate_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917out:
918 spin_unlock_irqrestore(&m->lock, flags);
919
920 return 0;
921}
922
923/*
924 * Reinstate a previously-failed path
925 */
926static int reinstate_path(struct pgpath *pgpath)
927{
928 int r = 0;
929 unsigned long flags;
930 struct multipath *m = pgpath->pg->m;
931
932 spin_lock_irqsave(&m->lock, flags);
933
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100934 if (pgpath->is_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 goto out;
936
Alasdair G Kergondef052d2008-07-21 12:00:31 +0100937 if (!pgpath->pg->ps.type->reinstate_path) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 DMWARN("Reinstate path not supported by path selector %s",
939 pgpath->pg->ps.type->name);
940 r = -EINVAL;
941 goto out;
942 }
943
944 r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path);
945 if (r)
946 goto out;
947
Kiyoshi Ueda66800732008-10-10 13:36:58 +0100948 pgpath->is_active = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100950 if (!m->nr_valid_paths++ && m->queue_size) {
951 m->current_pgpath = NULL;
Alasdair G Kergonc5573082005-05-05 16:16:07 -0700952 queue_work(kmultipathd, &m->process_queued_ios);
Chandra Seetharamane54f77d2009-06-22 10:12:12 +0100953 } else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) {
954 if (queue_work(kmpath_handlerd, &pgpath->activate_path))
955 m->pg_init_in_progress++;
956 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Mike Andersonb15546f2007-10-19 22:48:02 +0100958 dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti,
959 pgpath->path.dev->name, m->nr_valid_paths);
960
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +0000961 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963out:
964 spin_unlock_irqrestore(&m->lock, flags);
965
966 return r;
967}
968
969/*
970 * Fail or reinstate all paths that match the provided struct dm_dev.
971 */
972static int action_dev(struct multipath *m, struct dm_dev *dev,
973 action_fn action)
974{
975 int r = 0;
976 struct pgpath *pgpath;
977 struct priority_group *pg;
978
979 list_for_each_entry(pg, &m->priority_groups, list) {
980 list_for_each_entry(pgpath, &pg->pgpaths, list) {
981 if (pgpath->path.dev == dev)
982 r = action(pgpath);
983 }
984 }
985
986 return r;
987}
988
989/*
990 * Temporarily try to avoid having to use the specified PG
991 */
992static void bypass_pg(struct multipath *m, struct priority_group *pg,
993 int bypassed)
994{
995 unsigned long flags;
996
997 spin_lock_irqsave(&m->lock, flags);
998
999 pg->bypassed = bypassed;
1000 m->current_pgpath = NULL;
1001 m->current_pg = NULL;
1002
1003 spin_unlock_irqrestore(&m->lock, flags);
1004
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001005 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006}
1007
1008/*
1009 * Switch to using the specified PG from the next I/O that gets mapped
1010 */
1011static int switch_pg_num(struct multipath *m, const char *pgstr)
1012{
1013 struct priority_group *pg;
1014 unsigned pgnum;
1015 unsigned long flags;
1016
1017 if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum ||
1018 (pgnum > m->nr_priority_groups)) {
1019 DMWARN("invalid PG number supplied to switch_pg_num");
1020 return -EINVAL;
1021 }
1022
1023 spin_lock_irqsave(&m->lock, flags);
1024 list_for_each_entry(pg, &m->priority_groups, list) {
1025 pg->bypassed = 0;
1026 if (--pgnum)
1027 continue;
1028
1029 m->current_pgpath = NULL;
1030 m->current_pg = NULL;
1031 m->next_pg = pg;
1032 }
1033 spin_unlock_irqrestore(&m->lock, flags);
1034
Alasdair G Kergonfe9cf302009-01-06 03:05:13 +00001035 schedule_work(&m->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 return 0;
1037}
1038
1039/*
1040 * Set/clear bypassed status of a PG.
1041 * PGs are numbered upwards from 1 in the order they were declared.
1042 */
1043static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed)
1044{
1045 struct priority_group *pg;
1046 unsigned pgnum;
1047
1048 if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum ||
1049 (pgnum > m->nr_priority_groups)) {
1050 DMWARN("invalid PG number supplied to bypass_pg");
1051 return -EINVAL;
1052 }
1053
1054 list_for_each_entry(pg, &m->priority_groups, list) {
1055 if (!--pgnum)
1056 break;
1057 }
1058
1059 bypass_pg(m, pg, bypassed);
1060 return 0;
1061}
1062
1063/*
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001064 * Should we retry pg_init immediately?
1065 */
1066static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
1067{
1068 unsigned long flags;
1069 int limit_reached = 0;
1070
1071 spin_lock_irqsave(&m->lock, flags);
1072
1073 if (m->pg_init_count <= m->pg_init_retries)
1074 m->pg_init_required = 1;
1075 else
1076 limit_reached = 1;
1077
1078 spin_unlock_irqrestore(&m->lock, flags);
1079
1080 return limit_reached;
1081}
1082
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001083static void pg_init_done(struct dm_path *path, int errors)
1084{
1085 struct pgpath *pgpath = path_to_pgpath(path);
1086 struct priority_group *pg = pgpath->pg;
1087 struct multipath *m = pg->m;
1088 unsigned long flags;
1089
1090 /* device or driver problems */
1091 switch (errors) {
1092 case SCSI_DH_OK:
1093 break;
1094 case SCSI_DH_NOSYS:
1095 if (!m->hw_handler_name) {
1096 errors = 0;
1097 break;
1098 }
1099 DMERR("Cannot failover device because scsi_dh_%s was not "
1100 "loaded.", m->hw_handler_name);
1101 /*
1102 * Fail path for now, so we do not ping pong
1103 */
1104 fail_path(pgpath);
1105 break;
1106 case SCSI_DH_DEV_TEMP_BUSY:
1107 /*
1108 * Probably doing something like FW upgrade on the
1109 * controller so try the other pg.
1110 */
1111 bypass_pg(m, pg, 1);
1112 break;
1113 /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */
1114 case SCSI_DH_RETRY:
1115 case SCSI_DH_IMM_RETRY:
1116 case SCSI_DH_RES_TEMP_UNAVAIL:
1117 if (pg_init_limit_reached(m, pgpath))
1118 fail_path(pgpath);
1119 errors = 0;
1120 break;
1121 default:
1122 /*
1123 * We probably do not want to fail the path for a device
1124 * error, but this is what the old dm did. In future
1125 * patches we can do more advanced handling.
1126 */
1127 fail_path(pgpath);
1128 }
1129
1130 spin_lock_irqsave(&m->lock, flags);
1131 if (errors) {
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001132 if (pgpath == m->current_pgpath) {
1133 DMERR("Could not failover device. Error %d.", errors);
1134 m->current_pgpath = NULL;
1135 m->current_pg = NULL;
1136 }
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001137 } else if (!m->pg_init_required) {
1138 m->queue_io = 0;
1139 pg->bypassed = 0;
1140 }
1141
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001142 m->pg_init_in_progress--;
1143 if (!m->pg_init_in_progress)
1144 queue_work(kmultipathd, &m->process_queued_ios);
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001145 spin_unlock_irqrestore(&m->lock, flags);
1146}
1147
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001148static void activate_path(struct work_struct *work)
1149{
1150 int ret;
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001151 struct pgpath *pgpath =
1152 container_of(work, struct pgpath, activate_path);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001153
Chandra Seetharamane54f77d2009-06-22 10:12:12 +01001154 ret = scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev));
1155 pg_init_done(&pgpath->path, ret);
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001156}
1157
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158/*
1159 * end_io handling
1160 */
1161static int do_end_io(struct multipath *m, struct bio *bio,
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001162 int error, struct dm_mpath_io *mpio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163{
Stefan Bader640eb3b2005-11-21 21:32:35 -08001164 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
1166 if (!error)
1167 return 0; /* I/O complete */
1168
Lars Marowsky-Bree4f588022005-06-08 15:50:31 -07001169 if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
1170 return error;
1171
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -07001172 if (error == -EOPNOTSUPP)
1173 return error;
1174
Stefan Bader640eb3b2005-11-21 21:32:35 -08001175 spin_lock_irqsave(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 if (!m->nr_valid_paths) {
Kiyoshi Ueda45e15722006-12-08 02:41:10 -08001177 if (__must_push_back(m)) {
1178 spin_unlock_irqrestore(&m->lock, flags);
1179 return DM_ENDIO_REQUEUE;
1180 } else if (!m->queue_if_no_path) {
Stefan Bader640eb3b2005-11-21 21:32:35 -08001181 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 return -EIO;
1183 } else {
Stefan Bader640eb3b2005-11-21 21:32:35 -08001184 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 goto requeue;
1186 }
1187 }
Stefan Bader640eb3b2005-11-21 21:32:35 -08001188 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001190 if (mpio->pgpath)
1191 fail_path(mpio->pgpath);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
1193 requeue:
1194 dm_bio_restore(&mpio->details, bio);
1195
1196 /* queue for the daemon to resubmit or fail */
Stefan Bader640eb3b2005-11-21 21:32:35 -08001197 spin_lock_irqsave(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 bio_list_add(&m->queued_ios, bio);
1199 m->queue_size++;
1200 if (!m->queue_io)
Alasdair G Kergonc5573082005-05-05 16:16:07 -07001201 queue_work(kmultipathd, &m->process_queued_ios);
Stefan Bader640eb3b2005-11-21 21:32:35 -08001202 spin_unlock_irqrestore(&m->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001204 return DM_ENDIO_INCOMPLETE; /* io not complete */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
1207static int multipath_end_io(struct dm_target *ti, struct bio *bio,
1208 int error, union map_info *map_context)
1209{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001210 struct multipath *m = ti->private;
1211 struct dm_mpath_io *mpio = map_context->ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 struct pgpath *pgpath = mpio->pgpath;
1213 struct path_selector *ps;
1214 int r;
1215
1216 r = do_end_io(m, bio, error, mpio);
1217 if (pgpath) {
1218 ps = &pgpath->pg->ps;
1219 if (ps->type->end_io)
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +01001220 ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001222 if (r != DM_ENDIO_INCOMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 mempool_free(mpio, m->mpio_pool);
1224
1225 return r;
1226}
1227
1228/*
1229 * Suspend can't complete until all the I/O is processed so if
Alasdair G Kergon436d4102005-07-12 15:53:03 -07001230 * the last path fails we must error any remaining I/O.
1231 * Note that if the freeze_bdev fails while suspending, the
1232 * queue_if_no_path state is lost - userspace should reset it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 */
1234static void multipath_presuspend(struct dm_target *ti)
1235{
1236 struct multipath *m = (struct multipath *) ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Alasdair G Kergon485ef692005-09-27 21:45:45 -07001238 queue_if_no_path(m, 0, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239}
1240
Alasdair G Kergon436d4102005-07-12 15:53:03 -07001241/*
1242 * Restore the queue_if_no_path setting.
1243 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244static void multipath_resume(struct dm_target *ti)
1245{
1246 struct multipath *m = (struct multipath *) ti->private;
1247 unsigned long flags;
1248
1249 spin_lock_irqsave(&m->lock, flags);
Alasdair G Kergon436d4102005-07-12 15:53:03 -07001250 m->queue_if_no_path = m->saved_queue_if_no_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 spin_unlock_irqrestore(&m->lock, flags);
1252}
1253
1254/*
1255 * Info output has the following format:
1256 * num_multipath_feature_args [multipath_feature_args]*
1257 * num_handler_status_args [handler_status_args]*
1258 * num_groups init_group_number
1259 * [A|D|E num_ps_status_args [ps_status_args]*
1260 * num_paths num_selector_args
1261 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1262 *
1263 * Table output has the following format (identical to the constructor string):
1264 * num_feature_args [features_args]*
1265 * num_handler_args hw_handler [hw_handler_args]*
1266 * num_groups init_group_number
1267 * [priority selector-name num_ps_args [ps_args]*
1268 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1269 */
1270static int multipath_status(struct dm_target *ti, status_type_t type,
1271 char *result, unsigned int maxlen)
1272{
1273 int sz = 0;
1274 unsigned long flags;
1275 struct multipath *m = (struct multipath *) ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 struct priority_group *pg;
1277 struct pgpath *p;
1278 unsigned pg_num;
1279 char state;
1280
1281 spin_lock_irqsave(&m->lock, flags);
1282
1283 /* Features */
1284 if (type == STATUSTYPE_INFO)
Dave Wysochanskic9e45582007-10-19 22:47:53 +01001285 DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count);
1286 else {
1287 DMEMIT("%u ", m->queue_if_no_path +
1288 (m->pg_init_retries > 0) * 2);
1289 if (m->queue_if_no_path)
1290 DMEMIT("queue_if_no_path ");
1291 if (m->pg_init_retries)
1292 DMEMIT("pg_init_retries %u ", m->pg_init_retries);
1293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001295 if (!m->hw_handler_name || type == STATUSTYPE_INFO)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 DMEMIT("0 ");
1297 else
Chandra Seetharamancfae5c92008-05-01 14:50:11 -07001298 DMEMIT("1 %s ", m->hw_handler_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
1300 DMEMIT("%u ", m->nr_priority_groups);
1301
1302 if (m->next_pg)
1303 pg_num = m->next_pg->pg_num;
1304 else if (m->current_pg)
1305 pg_num = m->current_pg->pg_num;
1306 else
1307 pg_num = 1;
1308
1309 DMEMIT("%u ", pg_num);
1310
1311 switch (type) {
1312 case STATUSTYPE_INFO:
1313 list_for_each_entry(pg, &m->priority_groups, list) {
1314 if (pg->bypassed)
1315 state = 'D'; /* Disabled */
1316 else if (pg == m->current_pg)
1317 state = 'A'; /* Currently Active */
1318 else
1319 state = 'E'; /* Enabled */
1320
1321 DMEMIT("%c ", state);
1322
1323 if (pg->ps.type->status)
1324 sz += pg->ps.type->status(&pg->ps, NULL, type,
1325 result + sz,
1326 maxlen - sz);
1327 else
1328 DMEMIT("0 ");
1329
1330 DMEMIT("%u %u ", pg->nr_pgpaths,
1331 pg->ps.type->info_args);
1332
1333 list_for_each_entry(p, &pg->pgpaths, list) {
1334 DMEMIT("%s %s %u ", p->path.dev->name,
Kiyoshi Ueda66800732008-10-10 13:36:58 +01001335 p->is_active ? "A" : "F",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 p->fail_count);
1337 if (pg->ps.type->status)
1338 sz += pg->ps.type->status(&pg->ps,
1339 &p->path, type, result + sz,
1340 maxlen - sz);
1341 }
1342 }
1343 break;
1344
1345 case STATUSTYPE_TABLE:
1346 list_for_each_entry(pg, &m->priority_groups, list) {
1347 DMEMIT("%s ", pg->ps.type->name);
1348
1349 if (pg->ps.type->status)
1350 sz += pg->ps.type->status(&pg->ps, NULL, type,
1351 result + sz,
1352 maxlen - sz);
1353 else
1354 DMEMIT("0 ");
1355
1356 DMEMIT("%u %u ", pg->nr_pgpaths,
1357 pg->ps.type->table_args);
1358
1359 list_for_each_entry(p, &pg->pgpaths, list) {
1360 DMEMIT("%s ", p->path.dev->name);
1361 if (pg->ps.type->status)
1362 sz += pg->ps.type->status(&pg->ps,
1363 &p->path, type, result + sz,
1364 maxlen - sz);
1365 }
1366 }
1367 break;
1368 }
1369
1370 spin_unlock_irqrestore(&m->lock, flags);
1371
1372 return 0;
1373}
1374
1375static int multipath_message(struct dm_target *ti, unsigned argc, char **argv)
1376{
1377 int r;
1378 struct dm_dev *dev;
1379 struct multipath *m = (struct multipath *) ti->private;
1380 action_fn action;
1381
1382 if (argc == 1) {
1383 if (!strnicmp(argv[0], MESG_STR("queue_if_no_path")))
Alasdair G Kergon485ef692005-09-27 21:45:45 -07001384 return queue_if_no_path(m, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path")))
Alasdair G Kergon485ef692005-09-27 21:45:45 -07001386 return queue_if_no_path(m, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 }
1388
1389 if (argc != 2)
1390 goto error;
1391
1392 if (!strnicmp(argv[0], MESG_STR("disable_group")))
1393 return bypass_pg_num(m, argv[1], 1);
1394 else if (!strnicmp(argv[0], MESG_STR("enable_group")))
1395 return bypass_pg_num(m, argv[1], 0);
1396 else if (!strnicmp(argv[0], MESG_STR("switch_group")))
1397 return switch_pg_num(m, argv[1]);
1398 else if (!strnicmp(argv[0], MESG_STR("reinstate_path")))
1399 action = reinstate_path;
1400 else if (!strnicmp(argv[0], MESG_STR("fail_path")))
1401 action = fail_path;
1402 else
1403 goto error;
1404
1405 r = dm_get_device(ti, argv[1], ti->begin, ti->len,
1406 dm_table_get_mode(ti->table), &dev);
1407 if (r) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001408 DMWARN("message: error getting device %s",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 argv[1]);
1410 return -EINVAL;
1411 }
1412
1413 r = action_dev(m, dev, action);
1414
1415 dm_put_device(ti, dev);
1416
1417 return r;
1418
1419error:
1420 DMWARN("Unrecognised multipath message received.");
1421 return -EINVAL;
1422}
1423
Al Viro647b3d02007-08-28 22:15:59 -04001424static int multipath_ioctl(struct dm_target *ti, unsigned int cmd,
Milan Broz9af4aa32006-10-03 01:15:20 -07001425 unsigned long arg)
1426{
1427 struct multipath *m = (struct multipath *) ti->private;
1428 struct block_device *bdev = NULL;
Al Viro633a08b2007-08-29 20:34:12 -04001429 fmode_t mode = 0;
Milan Broz9af4aa32006-10-03 01:15:20 -07001430 unsigned long flags;
1431 int r = 0;
1432
1433 spin_lock_irqsave(&m->lock, flags);
1434
1435 if (!m->current_pgpath)
Kiyoshi Ueda02ab8232009-06-22 10:12:27 +01001436 __choose_pgpath(m, 0);
Milan Broz9af4aa32006-10-03 01:15:20 -07001437
Milan Broze90dae12006-10-03 01:15:22 -07001438 if (m->current_pgpath) {
Milan Broz9af4aa32006-10-03 01:15:20 -07001439 bdev = m->current_pgpath->path.dev->bdev;
Al Viro633a08b2007-08-29 20:34:12 -04001440 mode = m->current_pgpath->path.dev->mode;
Milan Broze90dae12006-10-03 01:15:22 -07001441 }
Milan Broz9af4aa32006-10-03 01:15:20 -07001442
1443 if (m->queue_io)
1444 r = -EAGAIN;
1445 else if (!bdev)
1446 r = -EIO;
1447
1448 spin_unlock_irqrestore(&m->lock, flags);
1449
Al Viro633a08b2007-08-29 20:34:12 -04001450 return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg);
Milan Broz9af4aa32006-10-03 01:15:20 -07001451}
1452
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453/*-----------------------------------------------------------------
1454 * Module setup
1455 *---------------------------------------------------------------*/
1456static struct target_type multipath_target = {
1457 .name = "multipath",
Milan Broz9af4aa32006-10-03 01:15:20 -07001458 .version = {1, 0, 5},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 .module = THIS_MODULE,
1460 .ctr = multipath_ctr,
1461 .dtr = multipath_dtr,
1462 .map = multipath_map,
1463 .end_io = multipath_end_io,
1464 .presuspend = multipath_presuspend,
1465 .resume = multipath_resume,
1466 .status = multipath_status,
1467 .message = multipath_message,
Milan Broz9af4aa32006-10-03 01:15:20 -07001468 .ioctl = multipath_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469};
1470
1471static int __init dm_multipath_init(void)
1472{
1473 int r;
1474
1475 /* allocate a slab for the dm_ios */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001476 _mpio_cache = KMEM_CACHE(dm_mpath_io, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (!_mpio_cache)
1478 return -ENOMEM;
1479
1480 r = dm_register_target(&multipath_target);
1481 if (r < 0) {
Alasdair G Kergon0cd33122007-07-12 17:27:01 +01001482 DMERR("register failed %d", r);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 kmem_cache_destroy(_mpio_cache);
1484 return -EINVAL;
1485 }
1486
Alasdair G Kergonc5573082005-05-05 16:16:07 -07001487 kmultipathd = create_workqueue("kmpathd");
1488 if (!kmultipathd) {
Alasdair G Kergon0cd33122007-07-12 17:27:01 +01001489 DMERR("failed to create workqueue kmpathd");
Alasdair G Kergonc5573082005-05-05 16:16:07 -07001490 dm_unregister_target(&multipath_target);
1491 kmem_cache_destroy(_mpio_cache);
1492 return -ENOMEM;
1493 }
1494
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001495 /*
1496 * A separate workqueue is used to handle the device handlers
1497 * to avoid overloading existing workqueue. Overloading the
1498 * old workqueue would also create a bottleneck in the
1499 * path of the storage hardware device activation.
1500 */
1501 kmpath_handlerd = create_singlethread_workqueue("kmpath_handlerd");
1502 if (!kmpath_handlerd) {
1503 DMERR("failed to create workqueue kmpath_handlerd");
1504 destroy_workqueue(kmultipathd);
1505 dm_unregister_target(&multipath_target);
1506 kmem_cache_destroy(_mpio_cache);
1507 return -ENOMEM;
1508 }
1509
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001510 DMINFO("version %u.%u.%u loaded",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 multipath_target.version[0], multipath_target.version[1],
1512 multipath_target.version[2]);
1513
1514 return r;
1515}
1516
1517static void __exit dm_multipath_exit(void)
1518{
Chandra Seetharamanbab7cfc2008-05-01 14:50:22 -07001519 destroy_workqueue(kmpath_handlerd);
Alasdair G Kergonc5573082005-05-05 16:16:07 -07001520 destroy_workqueue(kmultipathd);
1521
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00001522 dm_unregister_target(&multipath_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 kmem_cache_destroy(_mpio_cache);
1524}
1525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526module_init(dm_multipath_init);
1527module_exit(dm_multipath_exit);
1528
1529MODULE_DESCRIPTION(DM_NAME " multipath target");
1530MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
1531MODULE_LICENSE("GPL");