1*4882a593Smuzhiyun /*
2*4882a593Smuzhiyun * Copyright (C) 2004-2005 IBM Corp. All Rights Reserved.
3*4882a593Smuzhiyun * Copyright (C) 2006-2009 NEC Corporation.
4*4882a593Smuzhiyun *
5*4882a593Smuzhiyun * dm-queue-length.c
6*4882a593Smuzhiyun *
7*4882a593Smuzhiyun * Module Author: Stefan Bader, IBM
8*4882a593Smuzhiyun * Modified by: Kiyoshi Ueda, NEC
9*4882a593Smuzhiyun *
10*4882a593Smuzhiyun * This file is released under the GPL.
11*4882a593Smuzhiyun *
12*4882a593Smuzhiyun * queue-length path selector - choose a path with the least number of
13*4882a593Smuzhiyun * in-flight I/Os.
14*4882a593Smuzhiyun */
15*4882a593Smuzhiyun
16*4882a593Smuzhiyun #include "dm.h"
17*4882a593Smuzhiyun #include "dm-path-selector.h"
18*4882a593Smuzhiyun
19*4882a593Smuzhiyun #include <linux/slab.h>
20*4882a593Smuzhiyun #include <linux/ctype.h>
21*4882a593Smuzhiyun #include <linux/errno.h>
22*4882a593Smuzhiyun #include <linux/module.h>
23*4882a593Smuzhiyun #include <linux/atomic.h>
24*4882a593Smuzhiyun
25*4882a593Smuzhiyun #define DM_MSG_PREFIX "multipath queue-length"
26*4882a593Smuzhiyun #define QL_MIN_IO 1
27*4882a593Smuzhiyun #define QL_VERSION "0.2.0"
28*4882a593Smuzhiyun
29*4882a593Smuzhiyun struct selector {
30*4882a593Smuzhiyun struct list_head valid_paths;
31*4882a593Smuzhiyun struct list_head failed_paths;
32*4882a593Smuzhiyun spinlock_t lock;
33*4882a593Smuzhiyun };
34*4882a593Smuzhiyun
35*4882a593Smuzhiyun struct path_info {
36*4882a593Smuzhiyun struct list_head list;
37*4882a593Smuzhiyun struct dm_path *path;
38*4882a593Smuzhiyun unsigned repeat_count;
39*4882a593Smuzhiyun atomic_t qlen; /* the number of in-flight I/Os */
40*4882a593Smuzhiyun };
41*4882a593Smuzhiyun
alloc_selector(void)42*4882a593Smuzhiyun static struct selector *alloc_selector(void)
43*4882a593Smuzhiyun {
44*4882a593Smuzhiyun struct selector *s = kmalloc(sizeof(*s), GFP_KERNEL);
45*4882a593Smuzhiyun
46*4882a593Smuzhiyun if (s) {
47*4882a593Smuzhiyun INIT_LIST_HEAD(&s->valid_paths);
48*4882a593Smuzhiyun INIT_LIST_HEAD(&s->failed_paths);
49*4882a593Smuzhiyun spin_lock_init(&s->lock);
50*4882a593Smuzhiyun }
51*4882a593Smuzhiyun
52*4882a593Smuzhiyun return s;
53*4882a593Smuzhiyun }
54*4882a593Smuzhiyun
ql_create(struct path_selector * ps,unsigned argc,char ** argv)55*4882a593Smuzhiyun static int ql_create(struct path_selector *ps, unsigned argc, char **argv)
56*4882a593Smuzhiyun {
57*4882a593Smuzhiyun struct selector *s = alloc_selector();
58*4882a593Smuzhiyun
59*4882a593Smuzhiyun if (!s)
60*4882a593Smuzhiyun return -ENOMEM;
61*4882a593Smuzhiyun
62*4882a593Smuzhiyun ps->context = s;
63*4882a593Smuzhiyun return 0;
64*4882a593Smuzhiyun }
65*4882a593Smuzhiyun
ql_free_paths(struct list_head * paths)66*4882a593Smuzhiyun static void ql_free_paths(struct list_head *paths)
67*4882a593Smuzhiyun {
68*4882a593Smuzhiyun struct path_info *pi, *next;
69*4882a593Smuzhiyun
70*4882a593Smuzhiyun list_for_each_entry_safe(pi, next, paths, list) {
71*4882a593Smuzhiyun list_del(&pi->list);
72*4882a593Smuzhiyun kfree(pi);
73*4882a593Smuzhiyun }
74*4882a593Smuzhiyun }
75*4882a593Smuzhiyun
ql_destroy(struct path_selector * ps)76*4882a593Smuzhiyun static void ql_destroy(struct path_selector *ps)
77*4882a593Smuzhiyun {
78*4882a593Smuzhiyun struct selector *s = ps->context;
79*4882a593Smuzhiyun
80*4882a593Smuzhiyun ql_free_paths(&s->valid_paths);
81*4882a593Smuzhiyun ql_free_paths(&s->failed_paths);
82*4882a593Smuzhiyun kfree(s);
83*4882a593Smuzhiyun ps->context = NULL;
84*4882a593Smuzhiyun }
85*4882a593Smuzhiyun
ql_status(struct path_selector * ps,struct dm_path * path,status_type_t type,char * result,unsigned maxlen)86*4882a593Smuzhiyun static int ql_status(struct path_selector *ps, struct dm_path *path,
87*4882a593Smuzhiyun status_type_t type, char *result, unsigned maxlen)
88*4882a593Smuzhiyun {
89*4882a593Smuzhiyun unsigned sz = 0;
90*4882a593Smuzhiyun struct path_info *pi;
91*4882a593Smuzhiyun
92*4882a593Smuzhiyun /* When called with NULL path, return selector status/args. */
93*4882a593Smuzhiyun if (!path)
94*4882a593Smuzhiyun DMEMIT("0 ");
95*4882a593Smuzhiyun else {
96*4882a593Smuzhiyun pi = path->pscontext;
97*4882a593Smuzhiyun
98*4882a593Smuzhiyun switch (type) {
99*4882a593Smuzhiyun case STATUSTYPE_INFO:
100*4882a593Smuzhiyun DMEMIT("%d ", atomic_read(&pi->qlen));
101*4882a593Smuzhiyun break;
102*4882a593Smuzhiyun case STATUSTYPE_TABLE:
103*4882a593Smuzhiyun DMEMIT("%u ", pi->repeat_count);
104*4882a593Smuzhiyun break;
105*4882a593Smuzhiyun }
106*4882a593Smuzhiyun }
107*4882a593Smuzhiyun
108*4882a593Smuzhiyun return sz;
109*4882a593Smuzhiyun }
110*4882a593Smuzhiyun
ql_add_path(struct path_selector * ps,struct dm_path * path,int argc,char ** argv,char ** error)111*4882a593Smuzhiyun static int ql_add_path(struct path_selector *ps, struct dm_path *path,
112*4882a593Smuzhiyun int argc, char **argv, char **error)
113*4882a593Smuzhiyun {
114*4882a593Smuzhiyun struct selector *s = ps->context;
115*4882a593Smuzhiyun struct path_info *pi;
116*4882a593Smuzhiyun unsigned repeat_count = QL_MIN_IO;
117*4882a593Smuzhiyun char dummy;
118*4882a593Smuzhiyun unsigned long flags;
119*4882a593Smuzhiyun
120*4882a593Smuzhiyun /*
121*4882a593Smuzhiyun * Arguments: [<repeat_count>]
122*4882a593Smuzhiyun * <repeat_count>: The number of I/Os before switching path.
123*4882a593Smuzhiyun * If not given, default (QL_MIN_IO) is used.
124*4882a593Smuzhiyun */
125*4882a593Smuzhiyun if (argc > 1) {
126*4882a593Smuzhiyun *error = "queue-length ps: incorrect number of arguments";
127*4882a593Smuzhiyun return -EINVAL;
128*4882a593Smuzhiyun }
129*4882a593Smuzhiyun
130*4882a593Smuzhiyun if ((argc == 1) && (sscanf(argv[0], "%u%c", &repeat_count, &dummy) != 1)) {
131*4882a593Smuzhiyun *error = "queue-length ps: invalid repeat count";
132*4882a593Smuzhiyun return -EINVAL;
133*4882a593Smuzhiyun }
134*4882a593Smuzhiyun
135*4882a593Smuzhiyun if (repeat_count > 1) {
136*4882a593Smuzhiyun DMWARN_LIMIT("repeat_count > 1 is deprecated, using 1 instead");
137*4882a593Smuzhiyun repeat_count = 1;
138*4882a593Smuzhiyun }
139*4882a593Smuzhiyun
140*4882a593Smuzhiyun /* Allocate the path information structure */
141*4882a593Smuzhiyun pi = kmalloc(sizeof(*pi), GFP_KERNEL);
142*4882a593Smuzhiyun if (!pi) {
143*4882a593Smuzhiyun *error = "queue-length ps: Error allocating path information";
144*4882a593Smuzhiyun return -ENOMEM;
145*4882a593Smuzhiyun }
146*4882a593Smuzhiyun
147*4882a593Smuzhiyun pi->path = path;
148*4882a593Smuzhiyun pi->repeat_count = repeat_count;
149*4882a593Smuzhiyun atomic_set(&pi->qlen, 0);
150*4882a593Smuzhiyun
151*4882a593Smuzhiyun path->pscontext = pi;
152*4882a593Smuzhiyun
153*4882a593Smuzhiyun spin_lock_irqsave(&s->lock, flags);
154*4882a593Smuzhiyun list_add_tail(&pi->list, &s->valid_paths);
155*4882a593Smuzhiyun spin_unlock_irqrestore(&s->lock, flags);
156*4882a593Smuzhiyun
157*4882a593Smuzhiyun return 0;
158*4882a593Smuzhiyun }
159*4882a593Smuzhiyun
ql_fail_path(struct path_selector * ps,struct dm_path * path)160*4882a593Smuzhiyun static void ql_fail_path(struct path_selector *ps, struct dm_path *path)
161*4882a593Smuzhiyun {
162*4882a593Smuzhiyun struct selector *s = ps->context;
163*4882a593Smuzhiyun struct path_info *pi = path->pscontext;
164*4882a593Smuzhiyun unsigned long flags;
165*4882a593Smuzhiyun
166*4882a593Smuzhiyun spin_lock_irqsave(&s->lock, flags);
167*4882a593Smuzhiyun list_move(&pi->list, &s->failed_paths);
168*4882a593Smuzhiyun spin_unlock_irqrestore(&s->lock, flags);
169*4882a593Smuzhiyun }
170*4882a593Smuzhiyun
ql_reinstate_path(struct path_selector * ps,struct dm_path * path)171*4882a593Smuzhiyun static int ql_reinstate_path(struct path_selector *ps, struct dm_path *path)
172*4882a593Smuzhiyun {
173*4882a593Smuzhiyun struct selector *s = ps->context;
174*4882a593Smuzhiyun struct path_info *pi = path->pscontext;
175*4882a593Smuzhiyun unsigned long flags;
176*4882a593Smuzhiyun
177*4882a593Smuzhiyun spin_lock_irqsave(&s->lock, flags);
178*4882a593Smuzhiyun list_move_tail(&pi->list, &s->valid_paths);
179*4882a593Smuzhiyun spin_unlock_irqrestore(&s->lock, flags);
180*4882a593Smuzhiyun
181*4882a593Smuzhiyun return 0;
182*4882a593Smuzhiyun }
183*4882a593Smuzhiyun
184*4882a593Smuzhiyun /*
185*4882a593Smuzhiyun * Select a path having the minimum number of in-flight I/Os
186*4882a593Smuzhiyun */
ql_select_path(struct path_selector * ps,size_t nr_bytes)187*4882a593Smuzhiyun static struct dm_path *ql_select_path(struct path_selector *ps, size_t nr_bytes)
188*4882a593Smuzhiyun {
189*4882a593Smuzhiyun struct selector *s = ps->context;
190*4882a593Smuzhiyun struct path_info *pi = NULL, *best = NULL;
191*4882a593Smuzhiyun struct dm_path *ret = NULL;
192*4882a593Smuzhiyun unsigned long flags;
193*4882a593Smuzhiyun
194*4882a593Smuzhiyun spin_lock_irqsave(&s->lock, flags);
195*4882a593Smuzhiyun if (list_empty(&s->valid_paths))
196*4882a593Smuzhiyun goto out;
197*4882a593Smuzhiyun
198*4882a593Smuzhiyun list_for_each_entry(pi, &s->valid_paths, list) {
199*4882a593Smuzhiyun if (!best ||
200*4882a593Smuzhiyun (atomic_read(&pi->qlen) < atomic_read(&best->qlen)))
201*4882a593Smuzhiyun best = pi;
202*4882a593Smuzhiyun
203*4882a593Smuzhiyun if (!atomic_read(&best->qlen))
204*4882a593Smuzhiyun break;
205*4882a593Smuzhiyun }
206*4882a593Smuzhiyun
207*4882a593Smuzhiyun if (!best)
208*4882a593Smuzhiyun goto out;
209*4882a593Smuzhiyun
210*4882a593Smuzhiyun /* Move most recently used to least preferred to evenly balance. */
211*4882a593Smuzhiyun list_move_tail(&best->list, &s->valid_paths);
212*4882a593Smuzhiyun
213*4882a593Smuzhiyun ret = best->path;
214*4882a593Smuzhiyun out:
215*4882a593Smuzhiyun spin_unlock_irqrestore(&s->lock, flags);
216*4882a593Smuzhiyun return ret;
217*4882a593Smuzhiyun }
218*4882a593Smuzhiyun
ql_start_io(struct path_selector * ps,struct dm_path * path,size_t nr_bytes)219*4882a593Smuzhiyun static int ql_start_io(struct path_selector *ps, struct dm_path *path,
220*4882a593Smuzhiyun size_t nr_bytes)
221*4882a593Smuzhiyun {
222*4882a593Smuzhiyun struct path_info *pi = path->pscontext;
223*4882a593Smuzhiyun
224*4882a593Smuzhiyun atomic_inc(&pi->qlen);
225*4882a593Smuzhiyun
226*4882a593Smuzhiyun return 0;
227*4882a593Smuzhiyun }
228*4882a593Smuzhiyun
ql_end_io(struct path_selector * ps,struct dm_path * path,size_t nr_bytes,u64 start_time)229*4882a593Smuzhiyun static int ql_end_io(struct path_selector *ps, struct dm_path *path,
230*4882a593Smuzhiyun size_t nr_bytes, u64 start_time)
231*4882a593Smuzhiyun {
232*4882a593Smuzhiyun struct path_info *pi = path->pscontext;
233*4882a593Smuzhiyun
234*4882a593Smuzhiyun atomic_dec(&pi->qlen);
235*4882a593Smuzhiyun
236*4882a593Smuzhiyun return 0;
237*4882a593Smuzhiyun }
238*4882a593Smuzhiyun
239*4882a593Smuzhiyun static struct path_selector_type ql_ps = {
240*4882a593Smuzhiyun .name = "queue-length",
241*4882a593Smuzhiyun .module = THIS_MODULE,
242*4882a593Smuzhiyun .table_args = 1,
243*4882a593Smuzhiyun .info_args = 1,
244*4882a593Smuzhiyun .create = ql_create,
245*4882a593Smuzhiyun .destroy = ql_destroy,
246*4882a593Smuzhiyun .status = ql_status,
247*4882a593Smuzhiyun .add_path = ql_add_path,
248*4882a593Smuzhiyun .fail_path = ql_fail_path,
249*4882a593Smuzhiyun .reinstate_path = ql_reinstate_path,
250*4882a593Smuzhiyun .select_path = ql_select_path,
251*4882a593Smuzhiyun .start_io = ql_start_io,
252*4882a593Smuzhiyun .end_io = ql_end_io,
253*4882a593Smuzhiyun };
254*4882a593Smuzhiyun
dm_ql_init(void)255*4882a593Smuzhiyun static int __init dm_ql_init(void)
256*4882a593Smuzhiyun {
257*4882a593Smuzhiyun int r = dm_register_path_selector(&ql_ps);
258*4882a593Smuzhiyun
259*4882a593Smuzhiyun if (r < 0)
260*4882a593Smuzhiyun DMERR("register failed %d", r);
261*4882a593Smuzhiyun
262*4882a593Smuzhiyun DMINFO("version " QL_VERSION " loaded");
263*4882a593Smuzhiyun
264*4882a593Smuzhiyun return r;
265*4882a593Smuzhiyun }
266*4882a593Smuzhiyun
dm_ql_exit(void)267*4882a593Smuzhiyun static void __exit dm_ql_exit(void)
268*4882a593Smuzhiyun {
269*4882a593Smuzhiyun int r = dm_unregister_path_selector(&ql_ps);
270*4882a593Smuzhiyun
271*4882a593Smuzhiyun if (r < 0)
272*4882a593Smuzhiyun DMERR("unregister failed %d", r);
273*4882a593Smuzhiyun }
274*4882a593Smuzhiyun
275*4882a593Smuzhiyun module_init(dm_ql_init);
276*4882a593Smuzhiyun module_exit(dm_ql_exit);
277*4882a593Smuzhiyun
278*4882a593Smuzhiyun MODULE_AUTHOR("Stefan Bader <Stefan.Bader at de.ibm.com>");
279*4882a593Smuzhiyun MODULE_DESCRIPTION(
280*4882a593Smuzhiyun "(C) Copyright IBM Corp. 2004,2005 All Rights Reserved.\n"
281*4882a593Smuzhiyun DM_NAME " path selector to balance the number of in-flight I/Os"
282*4882a593Smuzhiyun );
283*4882a593Smuzhiyun MODULE_LICENSE("GPL");
284