1 /* 2 * Copyright (C) 2005-2007 Red Hat GmbH 3 * 4 * A target that delays reads and/or writes and can send 5 * them to different devices. 6 * 7 * This file is released under the GPL. 8 */ 9 10 #include <linux/module.h> 11 #include <linux/init.h> 12 #include <linux/blkdev.h> 13 #include <linux/bio.h> 14 #include <linux/slab.h> 15 16 #include "dm.h" 17 #include "dm-bio-list.h" 18 19 #define DM_MSG_PREFIX "delay" 20 21 struct delay_c { 22 struct timer_list delay_timer; 23 struct mutex timer_lock; 24 struct work_struct flush_expired_bios; 25 struct list_head delayed_bios; 26 atomic_t may_delay; 27 mempool_t *delayed_pool; 28 29 struct dm_dev *dev_read; 30 sector_t start_read; 31 unsigned read_delay; 32 unsigned reads; 33 34 struct dm_dev *dev_write; 35 sector_t start_write; 36 unsigned write_delay; 37 unsigned writes; 38 }; 39 40 struct dm_delay_info { 41 struct delay_c *context; 42 struct list_head list; 43 struct bio *bio; 44 unsigned long expires; 45 }; 46 47 static DEFINE_MUTEX(delayed_bios_lock); 48 49 static struct workqueue_struct *kdelayd_wq; 50 static struct kmem_cache *delayed_cache; 51 52 static void handle_delayed_timer(unsigned long data) 53 { 54 struct delay_c *dc = (struct delay_c *)data; 55 56 queue_work(kdelayd_wq, &dc->flush_expired_bios); 57 } 58 59 static void queue_timeout(struct delay_c *dc, unsigned long expires) 60 { 61 mutex_lock(&dc->timer_lock); 62 63 if (!timer_pending(&dc->delay_timer) || expires < dc->delay_timer.expires) 64 mod_timer(&dc->delay_timer, expires); 65 66 mutex_unlock(&dc->timer_lock); 67 } 68 69 static void flush_bios(struct bio *bio) 70 { 71 struct bio *n; 72 73 while (bio) { 74 n = bio->bi_next; 75 bio->bi_next = NULL; 76 generic_make_request(bio); 77 bio = n; 78 } 79 } 80 81 static struct bio *flush_delayed_bios(struct delay_c *dc, int flush_all) 82 { 83 struct dm_delay_info *delayed, *next; 84 unsigned long next_expires = 0; 85 int start_timer = 0; 86 struct bio_list flush_bios = { }; 87 88 mutex_lock(&delayed_bios_lock); 89 list_for_each_entry_safe(delayed, next, &dc->delayed_bios, list) { 90 if (flush_all || time_after_eq(jiffies, delayed->expires)) { 91 list_del(&delayed->list); 92 bio_list_add(&flush_bios, delayed->bio); 93 if ((bio_data_dir(delayed->bio) == WRITE)) 94 delayed->context->writes--; 95 else 96 delayed->context->reads--; 97 mempool_free(delayed, dc->delayed_pool); 98 continue; 99 } 100 101 if (!start_timer) { 102 start_timer = 1; 103 next_expires = delayed->expires; 104 } else 105 next_expires = min(next_expires, delayed->expires); 106 } 107 108 mutex_unlock(&delayed_bios_lock); 109 110 if (start_timer) 111 queue_timeout(dc, next_expires); 112 113 return bio_list_get(&flush_bios); 114 } 115 116 static void flush_expired_bios(struct work_struct *work) 117 { 118 struct delay_c *dc; 119 120 dc = container_of(work, struct delay_c, flush_expired_bios); 121 flush_bios(flush_delayed_bios(dc, 0)); 122 } 123 124 /* 125 * Mapping parameters: 126 * <device> <offset> <delay> [<write_device> <write_offset> <write_delay>] 127 * 128 * With separate write parameters, the first set is only used for reads. 129 * Delays are specified in milliseconds. 130 */ 131 static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv) 132 { 133 struct delay_c *dc; 134 unsigned long long tmpll; 135 136 if (argc != 3 && argc != 6) { 137 ti->error = "requires exactly 3 or 6 arguments"; 138 return -EINVAL; 139 } 140 141 dc = kmalloc(sizeof(*dc), GFP_KERNEL); 142 if (!dc) { 143 ti->error = "Cannot allocate context"; 144 return -ENOMEM; 145 } 146 147 dc->reads = dc->writes = 0; 148 149 if (sscanf(argv[1], "%llu", &tmpll) != 1) { 150 ti->error = "Invalid device sector"; 151 goto bad; 152 } 153 dc->start_read = tmpll; 154 155 if (sscanf(argv[2], "%u", &dc->read_delay) != 1) { 156 ti->error = "Invalid delay"; 157 goto bad; 158 } 159 160 if (dm_get_device(ti, argv[0], dc->start_read, ti->len, 161 dm_table_get_mode(ti->table), &dc->dev_read)) { 162 ti->error = "Device lookup failed"; 163 goto bad; 164 } 165 166 dc->dev_write = NULL; 167 if (argc == 3) 168 goto out; 169 170 if (sscanf(argv[4], "%llu", &tmpll) != 1) { 171 ti->error = "Invalid write device sector"; 172 goto bad_dev_read; 173 } 174 dc->start_write = tmpll; 175 176 if (sscanf(argv[5], "%u", &dc->write_delay) != 1) { 177 ti->error = "Invalid write delay"; 178 goto bad_dev_read; 179 } 180 181 if (dm_get_device(ti, argv[3], dc->start_write, ti->len, 182 dm_table_get_mode(ti->table), &dc->dev_write)) { 183 ti->error = "Write device lookup failed"; 184 goto bad_dev_read; 185 } 186 187 out: 188 dc->delayed_pool = mempool_create_slab_pool(128, delayed_cache); 189 if (!dc->delayed_pool) { 190 DMERR("Couldn't create delayed bio pool."); 191 goto bad_dev_write; 192 } 193 194 setup_timer(&dc->delay_timer, handle_delayed_timer, (unsigned long)dc); 195 196 INIT_WORK(&dc->flush_expired_bios, flush_expired_bios); 197 INIT_LIST_HEAD(&dc->delayed_bios); 198 mutex_init(&dc->timer_lock); 199 atomic_set(&dc->may_delay, 1); 200 201 ti->private = dc; 202 return 0; 203 204 bad_dev_write: 205 if (dc->dev_write) 206 dm_put_device(ti, dc->dev_write); 207 bad_dev_read: 208 dm_put_device(ti, dc->dev_read); 209 bad: 210 kfree(dc); 211 return -EINVAL; 212 } 213 214 static void delay_dtr(struct dm_target *ti) 215 { 216 struct delay_c *dc = ti->private; 217 218 flush_workqueue(kdelayd_wq); 219 220 dm_put_device(ti, dc->dev_read); 221 222 if (dc->dev_write) 223 dm_put_device(ti, dc->dev_write); 224 225 mempool_destroy(dc->delayed_pool); 226 kfree(dc); 227 } 228 229 static int delay_bio(struct delay_c *dc, int delay, struct bio *bio) 230 { 231 struct dm_delay_info *delayed; 232 unsigned long expires = 0; 233 234 if (!delay || !atomic_read(&dc->may_delay)) 235 return 1; 236 237 delayed = mempool_alloc(dc->delayed_pool, GFP_NOIO); 238 239 delayed->context = dc; 240 delayed->bio = bio; 241 delayed->expires = expires = jiffies + (delay * HZ / 1000); 242 243 mutex_lock(&delayed_bios_lock); 244 245 if (bio_data_dir(bio) == WRITE) 246 dc->writes++; 247 else 248 dc->reads++; 249 250 list_add_tail(&delayed->list, &dc->delayed_bios); 251 252 mutex_unlock(&delayed_bios_lock); 253 254 queue_timeout(dc, expires); 255 256 return 0; 257 } 258 259 static void delay_presuspend(struct dm_target *ti) 260 { 261 struct delay_c *dc = ti->private; 262 263 atomic_set(&dc->may_delay, 0); 264 del_timer_sync(&dc->delay_timer); 265 flush_bios(flush_delayed_bios(dc, 1)); 266 } 267 268 static void delay_resume(struct dm_target *ti) 269 { 270 struct delay_c *dc = ti->private; 271 272 atomic_set(&dc->may_delay, 1); 273 } 274 275 static int delay_map(struct dm_target *ti, struct bio *bio, 276 union map_info *map_context) 277 { 278 struct delay_c *dc = ti->private; 279 280 if ((bio_data_dir(bio) == WRITE) && (dc->dev_write)) { 281 bio->bi_bdev = dc->dev_write->bdev; 282 bio->bi_sector = dc->start_write + 283 (bio->bi_sector - ti->begin); 284 285 return delay_bio(dc, dc->write_delay, bio); 286 } 287 288 bio->bi_bdev = dc->dev_read->bdev; 289 bio->bi_sector = dc->start_read + 290 (bio->bi_sector - ti->begin); 291 292 return delay_bio(dc, dc->read_delay, bio); 293 } 294 295 static int delay_status(struct dm_target *ti, status_type_t type, 296 char *result, unsigned maxlen) 297 { 298 struct delay_c *dc = ti->private; 299 int sz = 0; 300 301 switch (type) { 302 case STATUSTYPE_INFO: 303 DMEMIT("%u %u", dc->reads, dc->writes); 304 break; 305 306 case STATUSTYPE_TABLE: 307 DMEMIT("%s %llu %u", dc->dev_read->name, 308 (unsigned long long) dc->start_read, 309 dc->read_delay); 310 if (dc->dev_write) 311 DMEMIT(" %s %llu %u", dc->dev_write->name, 312 (unsigned long long) dc->start_write, 313 dc->write_delay); 314 break; 315 } 316 317 return 0; 318 } 319 320 static struct target_type delay_target = { 321 .name = "delay", 322 .version = {1, 0, 2}, 323 .module = THIS_MODULE, 324 .ctr = delay_ctr, 325 .dtr = delay_dtr, 326 .map = delay_map, 327 .presuspend = delay_presuspend, 328 .resume = delay_resume, 329 .status = delay_status, 330 }; 331 332 static int __init dm_delay_init(void) 333 { 334 int r = -ENOMEM; 335 336 kdelayd_wq = create_workqueue("kdelayd"); 337 if (!kdelayd_wq) { 338 DMERR("Couldn't start kdelayd"); 339 goto bad_queue; 340 } 341 342 delayed_cache = KMEM_CACHE(dm_delay_info, 0); 343 if (!delayed_cache) { 344 DMERR("Couldn't create delayed bio cache."); 345 goto bad_memcache; 346 } 347 348 r = dm_register_target(&delay_target); 349 if (r < 0) { 350 DMERR("register failed %d", r); 351 goto bad_register; 352 } 353 354 return 0; 355 356 bad_register: 357 kmem_cache_destroy(delayed_cache); 358 bad_memcache: 359 destroy_workqueue(kdelayd_wq); 360 bad_queue: 361 return r; 362 } 363 364 static void __exit dm_delay_exit(void) 365 { 366 int r = dm_unregister_target(&delay_target); 367 368 if (r < 0) 369 DMERR("unregister failed %d", r); 370 371 kmem_cache_destroy(delayed_cache); 372 destroy_workqueue(kdelayd_wq); 373 } 374 375 /* Module hooks */ 376 module_init(dm_delay_init); 377 module_exit(dm_delay_exit); 378 379 MODULE_DESCRIPTION(DM_NAME " delay target"); 380 MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>"); 381 MODULE_LICENSE("GPL"); 382