Heinz Mauelshagen | 3bd9400 | 2023-01-25 21:00:44 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 2 | /* |
| 3 | * Copyright (C) 2004-2005 IBM Corp. All Rights Reserved. |
| 4 | * Copyright (C) 2006-2009 NEC Corporation. |
| 5 | * |
| 6 | * dm-queue-length.c |
| 7 | * |
| 8 | * Module Author: Stefan Bader, IBM |
| 9 | * Modified by: Kiyoshi Ueda, NEC |
| 10 | * |
| 11 | * This file is released under the GPL. |
| 12 | * |
| 13 | * queue-length path selector - choose a path with the least number of |
| 14 | * in-flight I/Os. |
| 15 | */ |
| 16 | |
| 17 | #include "dm.h" |
| 18 | #include "dm-path-selector.h" |
| 19 | |
| 20 | #include <linux/slab.h> |
| 21 | #include <linux/ctype.h> |
| 22 | #include <linux/errno.h> |
| 23 | #include <linux/module.h> |
Arun Sharma | 60063497 | 2011-07-26 16:09:06 -0700 | [diff] [blame] | 24 | #include <linux/atomic.h> |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 25 | |
| 26 | #define DM_MSG_PREFIX "multipath queue-length" |
Mike Snitzer | 21136f8 | 2016-02-10 11:58:45 -0500 | [diff] [blame] | 27 | #define QL_MIN_IO 1 |
| 28 | #define QL_VERSION "0.2.0" |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 29 | |
| 30 | struct selector { |
| 31 | struct list_head valid_paths; |
| 32 | struct list_head failed_paths; |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 33 | spinlock_t lock; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 34 | }; |
| 35 | |
| 36 | struct path_info { |
| 37 | struct list_head list; |
| 38 | struct dm_path *path; |
Heinz Mauelshagen | 86a3238 | 2023-01-25 21:14:58 +0100 | [diff] [blame] | 39 | unsigned int repeat_count; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 40 | atomic_t qlen; /* the number of in-flight I/Os */ |
| 41 | }; |
| 42 | |
| 43 | static struct selector *alloc_selector(void) |
| 44 | { |
| 45 | struct selector *s = kmalloc(sizeof(*s), GFP_KERNEL); |
| 46 | |
| 47 | if (s) { |
| 48 | INIT_LIST_HEAD(&s->valid_paths); |
| 49 | INIT_LIST_HEAD(&s->failed_paths); |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 50 | spin_lock_init(&s->lock); |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 51 | } |
| 52 | |
| 53 | return s; |
| 54 | } |
| 55 | |
Heinz Mauelshagen | 86a3238 | 2023-01-25 21:14:58 +0100 | [diff] [blame] | 56 | static int ql_create(struct path_selector *ps, unsigned int argc, char **argv) |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 57 | { |
| 58 | struct selector *s = alloc_selector(); |
| 59 | |
| 60 | if (!s) |
| 61 | return -ENOMEM; |
| 62 | |
| 63 | ps->context = s; |
| 64 | return 0; |
| 65 | } |
| 66 | |
| 67 | static void ql_free_paths(struct list_head *paths) |
| 68 | { |
| 69 | struct path_info *pi, *next; |
| 70 | |
| 71 | list_for_each_entry_safe(pi, next, paths, list) { |
| 72 | list_del(&pi->list); |
| 73 | kfree(pi); |
| 74 | } |
| 75 | } |
| 76 | |
| 77 | static void ql_destroy(struct path_selector *ps) |
| 78 | { |
| 79 | struct selector *s = ps->context; |
| 80 | |
| 81 | ql_free_paths(&s->valid_paths); |
| 82 | ql_free_paths(&s->failed_paths); |
| 83 | kfree(s); |
| 84 | ps->context = NULL; |
| 85 | } |
| 86 | |
| 87 | static int ql_status(struct path_selector *ps, struct dm_path *path, |
Heinz Mauelshagen | 86a3238 | 2023-01-25 21:14:58 +0100 | [diff] [blame] | 88 | status_type_t type, char *result, unsigned int maxlen) |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 89 | { |
Heinz Mauelshagen | 86a3238 | 2023-01-25 21:14:58 +0100 | [diff] [blame] | 90 | unsigned int sz = 0; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 91 | struct path_info *pi; |
| 92 | |
| 93 | /* When called with NULL path, return selector status/args. */ |
| 94 | if (!path) |
| 95 | DMEMIT("0 "); |
| 96 | else { |
| 97 | pi = path->pscontext; |
| 98 | |
| 99 | switch (type) { |
| 100 | case STATUSTYPE_INFO: |
| 101 | DMEMIT("%d ", atomic_read(&pi->qlen)); |
| 102 | break; |
| 103 | case STATUSTYPE_TABLE: |
| 104 | DMEMIT("%u ", pi->repeat_count); |
| 105 | break; |
Tushar Sugandhi | 8ec4566 | 2021-07-12 17:49:03 -0700 | [diff] [blame] | 106 | case STATUSTYPE_IMA: |
| 107 | *result = '\0'; |
| 108 | break; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 109 | } |
| 110 | } |
| 111 | |
| 112 | return sz; |
| 113 | } |
| 114 | |
| 115 | static int ql_add_path(struct path_selector *ps, struct dm_path *path, |
| 116 | int argc, char **argv, char **error) |
| 117 | { |
| 118 | struct selector *s = ps->context; |
| 119 | struct path_info *pi; |
Heinz Mauelshagen | 86a3238 | 2023-01-25 21:14:58 +0100 | [diff] [blame] | 120 | unsigned int repeat_count = QL_MIN_IO; |
Mikulas Patocka | 31998ef | 2012-03-28 18:41:26 +0100 | [diff] [blame] | 121 | char dummy; |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 122 | unsigned long flags; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 123 | |
| 124 | /* |
| 125 | * Arguments: [<repeat_count>] |
Heinz Mauelshagen | 8ca817c | 2023-02-01 22:31:43 +0100 | [diff] [blame] | 126 | * <repeat_count>: The number of I/Os before switching path. |
| 127 | * If not given, default (QL_MIN_IO) is used. |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 128 | */ |
| 129 | if (argc > 1) { |
| 130 | *error = "queue-length ps: incorrect number of arguments"; |
| 131 | return -EINVAL; |
| 132 | } |
| 133 | |
Mikulas Patocka | 31998ef | 2012-03-28 18:41:26 +0100 | [diff] [blame] | 134 | if ((argc == 1) && (sscanf(argv[0], "%u%c", &repeat_count, &dummy) != 1)) { |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 135 | *error = "queue-length ps: invalid repeat count"; |
| 136 | return -EINVAL; |
| 137 | } |
| 138 | |
Mike Snitzer | 21136f8 | 2016-02-10 11:58:45 -0500 | [diff] [blame] | 139 | if (repeat_count > 1) { |
| 140 | DMWARN_LIMIT("repeat_count > 1 is deprecated, using 1 instead"); |
| 141 | repeat_count = 1; |
| 142 | } |
| 143 | |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 144 | /* Allocate the path information structure */ |
| 145 | pi = kmalloc(sizeof(*pi), GFP_KERNEL); |
| 146 | if (!pi) { |
| 147 | *error = "queue-length ps: Error allocating path information"; |
| 148 | return -ENOMEM; |
| 149 | } |
| 150 | |
| 151 | pi->path = path; |
| 152 | pi->repeat_count = repeat_count; |
| 153 | atomic_set(&pi->qlen, 0); |
| 154 | |
| 155 | path->pscontext = pi; |
| 156 | |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 157 | spin_lock_irqsave(&s->lock, flags); |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 158 | list_add_tail(&pi->list, &s->valid_paths); |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 159 | spin_unlock_irqrestore(&s->lock, flags); |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 160 | |
| 161 | return 0; |
| 162 | } |
| 163 | |
| 164 | static void ql_fail_path(struct path_selector *ps, struct dm_path *path) |
| 165 | { |
| 166 | struct selector *s = ps->context; |
| 167 | struct path_info *pi = path->pscontext; |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 168 | unsigned long flags; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 169 | |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 170 | spin_lock_irqsave(&s->lock, flags); |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 171 | list_move(&pi->list, &s->failed_paths); |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 172 | spin_unlock_irqrestore(&s->lock, flags); |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 173 | } |
| 174 | |
| 175 | static int ql_reinstate_path(struct path_selector *ps, struct dm_path *path) |
| 176 | { |
| 177 | struct selector *s = ps->context; |
| 178 | struct path_info *pi = path->pscontext; |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 179 | unsigned long flags; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 180 | |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 181 | spin_lock_irqsave(&s->lock, flags); |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 182 | list_move_tail(&pi->list, &s->valid_paths); |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 183 | spin_unlock_irqrestore(&s->lock, flags); |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 184 | |
| 185 | return 0; |
| 186 | } |
| 187 | |
| 188 | /* |
| 189 | * Select a path having the minimum number of in-flight I/Os |
| 190 | */ |
Mike Snitzer | 90a4323 | 2016-02-17 21:29:17 -0500 | [diff] [blame] | 191 | static struct dm_path *ql_select_path(struct path_selector *ps, size_t nr_bytes) |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 192 | { |
| 193 | struct selector *s = ps->context; |
| 194 | struct path_info *pi = NULL, *best = NULL; |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 195 | struct dm_path *ret = NULL; |
| 196 | unsigned long flags; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 197 | |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 198 | spin_lock_irqsave(&s->lock, flags); |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 199 | if (list_empty(&s->valid_paths)) |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 200 | goto out; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 201 | |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 202 | list_for_each_entry(pi, &s->valid_paths, list) { |
| 203 | if (!best || |
| 204 | (atomic_read(&pi->qlen) < atomic_read(&best->qlen))) |
| 205 | best = pi; |
| 206 | |
| 207 | if (!atomic_read(&best->qlen)) |
| 208 | break; |
| 209 | } |
| 210 | |
| 211 | if (!best) |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 212 | goto out; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 213 | |
Khazhismel Kumykov | f204260 | 2018-01-19 15:07:37 -0800 | [diff] [blame] | 214 | /* Move most recently used to least preferred to evenly balance. */ |
| 215 | list_move_tail(&best->list, &s->valid_paths); |
| 216 | |
Mike Snitzer | 9659f81 | 2016-02-15 14:25:00 -0500 | [diff] [blame] | 217 | ret = best->path; |
| 218 | out: |
| 219 | spin_unlock_irqrestore(&s->lock, flags); |
| 220 | return ret; |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 221 | } |
| 222 | |
| 223 | static int ql_start_io(struct path_selector *ps, struct dm_path *path, |
| 224 | size_t nr_bytes) |
| 225 | { |
| 226 | struct path_info *pi = path->pscontext; |
| 227 | |
| 228 | atomic_inc(&pi->qlen); |
| 229 | |
| 230 | return 0; |
| 231 | } |
| 232 | |
| 233 | static int ql_end_io(struct path_selector *ps, struct dm_path *path, |
Gabriel Krisman Bertazi | 087615bf | 2020-04-30 16:48:29 -0400 | [diff] [blame] | 234 | size_t nr_bytes, u64 start_time) |
Kiyoshi Ueda | fd5e033 | 2009-06-22 10:12:27 +0100 | [diff] [blame] | 235 | { |
| 236 | struct path_info *pi = path->pscontext; |
| 237 | |
| 238 | atomic_dec(&pi->qlen); |
| 239 | |
| 240 | return 0; |
| 241 | } |
| 242 | |
| 243 | static struct path_selector_type ql_ps = { |
| 244 | .name = "queue-length", |
| 245 | .module = THIS_MODULE, |
| 246 | .table_args = 1, |
| 247 | .info_args = 1, |
| 248 | .create = ql_create, |
| 249 | .destroy = ql_destroy, |
| 250 | .status = ql_status, |
| 251 | .add_path = ql_add_path, |
| 252 | .fail_path = ql_fail_path, |
| 253 | .reinstate_path = ql_reinstate_path, |
| 254 | .select_path = ql_select_path, |
| 255 | .start_io = ql_start_io, |
| 256 | .end_io = ql_end_io, |
| 257 | }; |
| 258 | |
| 259 | static int __init dm_ql_init(void) |
| 260 | { |
| 261 | int r = dm_register_path_selector(&ql_ps); |
| 262 | |
| 263 | if (r < 0) |
| 264 | DMERR("register failed %d", r); |
| 265 | |
| 266 | DMINFO("version " QL_VERSION " loaded"); |
| 267 | |
| 268 | return r; |
| 269 | } |
| 270 | |
| 271 | static void __exit dm_ql_exit(void) |
| 272 | { |
| 273 | int r = dm_unregister_path_selector(&ql_ps); |
| 274 | |
| 275 | if (r < 0) |
| 276 | DMERR("unregister failed %d", r); |
| 277 | } |
| 278 | |
| 279 | module_init(dm_ql_init); |
| 280 | module_exit(dm_ql_exit); |
| 281 | |
| 282 | MODULE_AUTHOR("Stefan Bader <Stefan.Bader at de.ibm.com>"); |
| 283 | MODULE_DESCRIPTION( |
| 284 | "(C) Copyright IBM Corp. 2004,2005 All Rights Reserved.\n" |
| 285 | DM_NAME " path selector to balance the number of in-flight I/Os" |
| 286 | ); |
| 287 | MODULE_LICENSE("GPL"); |