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 <linux/device-mapper.h> 17 18 #define DM_MSG_PREFIX "delay" 19 20 struct delay_c { 21 struct timer_list delay_timer; 22 struct mutex timer_lock; 23 struct work_struct flush_expired_bios; 24 struct list_head delayed_bios; 25 atomic_t may_delay; 26 mempool_t *delayed_pool; 27 28 struct dm_dev *dev_read; 29 sector_t start_read; 30 unsigned read_delay; 31 unsigned reads; 32 33 struct dm_dev *dev_write; 34 sector_t start_write; 35 unsigned write_delay; 36 unsigned writes; 37 }; 38 39 struct dm_delay_info { 40 struct delay_c *context; 41 struct list_head list; 42 struct bio *bio; 43 unsigned long expires; 44 }; 45 46 static DEFINE_MUTEX(delayed_bios_lock); 47 48 static struct workqueue_struct *kdelayd_wq; 49 static struct kmem_cache *delayed_cache; 50 51 static void handle_delayed_timer(unsigned long data) 52 { 53 struct delay_c *dc = (struct delay_c *)data; 54 55 queue_work(kdelayd_wq, &dc->flush_expired_bios); 56 } 57 58 static void queue_timeout(struct delay_c *dc, unsigned long expires) 59 { 60 mutex_lock(&dc->timer_lock); 61 62 if (!timer_pending(&dc->delay_timer) || expires < dc->delay_timer.expires) 63 mod_timer(&dc->delay_timer, expires); 64 65 mutex_unlock(&dc->timer_lock); 66 } 67 68 static void flush_bios(struct bio *bio) 69 { 70 struct bio *n; 71 72 while (bio) { 73 n = bio->bi_next; 74 bio->bi_next = NULL; 75 generic_make_request(bio); 76 bio = n; 77 } 78 } 79 80 static struct bio *flush_delayed_bios(struct delay_c *dc, int flush_all) 81 { 82 struct dm_delay_info *delayed, *next; 83 unsigned long next_expires = 0; 84 int start_timer = 0; 85 struct bio_list flush_bios = { }; 86 87 mutex_lock(&delayed_bios_lock); 88 list_for_each_entry_safe(delayed, next, &dc->delayed_bios, list) { 89 if (flush_all || time_after_eq(jiffies, delayed->expires)) { 90 list_del(&delayed->list); 91 bio_list_add(&flush_bios, delayed->bio); 92 if ((bio_data_dir(delayed->bio) == WRITE)) 93 delayed->context->writes--; 94 else 95 delayed->context->reads--; 96 mempool_free(delayed, dc->delayed_pool); 97 continue; 98 } 99 100 if (!start_timer) { 101 start_timer = 1; 102 next_expires = delayed->expires; 103 } else 104 next_expires = min(next_expires, delayed->expires); 105 } 106 107 mutex_unlock(&delayed_bios_lock); 108 109 if (start_timer) 110 queue_timeout(dc, next_expires); 111 112 return bio_list_get(&flush_bios); 113 } 114 115 static void flush_expired_bios(struct work_struct *work) 116 { 117 struct delay_c *dc; 118 119 dc = container_of(work, struct delay_c, flush_expired_bios); 120 flush_bios(flush_delayed_bios(dc, 0)); 121 } 122 123 /* 124 * Mapping parameters: 125 * <device> <offset> <delay> [<write_device> <write_offset> <write_delay>] 126 * 127 * With separate write parameters, the first set is only used for reads. 128 * Delays are specified in milliseconds. 129 */ 130 static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv) 131 { 132 struct delay_c *dc; 133 unsigned long long tmpll; 134 135 if (argc != 3 && argc != 6) { 136 ti->error = "requires exactly 3 or 6 arguments"; 137 return -EINVAL; 138 } 139 140 dc = kmalloc(sizeof(*dc), GFP_KERNEL); 141 if (!dc) { 142 ti->error = "Cannot allocate context"; 143 return -ENOMEM; 144 } 145 146 dc->reads = dc->writes = 0; 147 148 if (sscanf(argv[1], "%llu", &tmpll) != 1) { 149 ti->error = "Invalid device sector"; 150 goto bad; 151 } 152 dc->start_read = tmpll; 153 154 if (sscanf(argv[2], "%u", &dc->read_delay) != 1) { 155 ti->error = "Invalid delay"; 156 goto bad; 157 } 158 159 if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), 160 &dc->dev_read)) { 161 ti->error = "Device lookup failed"; 162 goto bad; 163 } 164 165 dc->dev_write = NULL; 166 if (argc == 3) 167 goto out; 168 169 if (sscanf(argv[4], "%llu", &tmpll) != 1) { 170 ti->error = "Invalid write device sector"; 171 goto bad_dev_read; 172 } 173 dc->start_write = tmpll; 174 175 if (sscanf(argv[5], "%u", &dc->write_delay) != 1) { 176 ti->error = "Invalid write delay"; 177 goto bad_dev_read; 178 } 179 180 if (dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), 181 &dc->dev_write)) { 182 ti->error = "Write device lookup failed"; 183 goto bad_dev_read; 184 } 185 186 out: 187 dc->delayed_pool = mempool_create_slab_pool(128, delayed_cache); 188 if (!dc->delayed_pool) { 189 DMERR("Couldn't create delayed bio pool."); 190 goto bad_dev_write; 191 } 192 193 setup_timer(&dc->delay_timer, handle_delayed_timer, (unsigned long)dc); 194 195 INIT_WORK(&dc->flush_expired_bios, flush_expired_bios); 196 INIT_LIST_HEAD(&dc->delayed_bios); 197 mutex_init(&dc->timer_lock); 198 atomic_set(&dc->may_delay, 1); 199 200 ti->num_flush_requests = 1; 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 if (bio_sectors(bio)) 283 bio->bi_sector = dc->start_write + 284 (bio->bi_sector - ti->begin); 285 286 return delay_bio(dc, dc->write_delay, bio); 287 } 288 289 bio->bi_bdev = dc->dev_read->bdev; 290 bio->bi_sector = dc->start_read + 291 (bio->bi_sector - ti->begin); 292 293 return delay_bio(dc, dc->read_delay, bio); 294 } 295 296 static int delay_status(struct dm_target *ti, status_type_t type, 297 char *result, unsigned maxlen) 298 { 299 struct delay_c *dc = ti->private; 300 int sz = 0; 301 302 switch (type) { 303 case STATUSTYPE_INFO: 304 DMEMIT("%u %u", dc->reads, dc->writes); 305 break; 306 307 case STATUSTYPE_TABLE: 308 DMEMIT("%s %llu %u", dc->dev_read->name, 309 (unsigned long long) dc->start_read, 310 dc->read_delay); 311 if (dc->dev_write) 312 DMEMIT(" %s %llu %u", dc->dev_write->name, 313 (unsigned long long) dc->start_write, 314 dc->write_delay); 315 break; 316 } 317 318 return 0; 319 } 320 321 static int delay_iterate_devices(struct dm_target *ti, 322 iterate_devices_callout_fn fn, void *data) 323 { 324 struct delay_c *dc = ti->private; 325 int ret = 0; 326 327 ret = fn(ti, dc->dev_read, dc->start_read, ti->len, data); 328 if (ret) 329 goto out; 330 331 if (dc->dev_write) 332 ret = fn(ti, dc->dev_write, dc->start_write, ti->len, data); 333 334 out: 335 return ret; 336 } 337 338 static struct target_type delay_target = { 339 .name = "delay", 340 .version = {1, 1, 0}, 341 .module = THIS_MODULE, 342 .ctr = delay_ctr, 343 .dtr = delay_dtr, 344 .map = delay_map, 345 .presuspend = delay_presuspend, 346 .resume = delay_resume, 347 .status = delay_status, 348 .iterate_devices = delay_iterate_devices, 349 }; 350 351 static int __init dm_delay_init(void) 352 { 353 int r = -ENOMEM; 354 355 kdelayd_wq = create_workqueue("kdelayd"); 356 if (!kdelayd_wq) { 357 DMERR("Couldn't start kdelayd"); 358 goto bad_queue; 359 } 360 361 delayed_cache = KMEM_CACHE(dm_delay_info, 0); 362 if (!delayed_cache) { 363 DMERR("Couldn't create delayed bio cache."); 364 goto bad_memcache; 365 } 366 367 r = dm_register_target(&delay_target); 368 if (r < 0) { 369 DMERR("register failed %d", r); 370 goto bad_register; 371 } 372 373 return 0; 374 375 bad_register: 376 kmem_cache_destroy(delayed_cache); 377 bad_memcache: 378 destroy_workqueue(kdelayd_wq); 379 bad_queue: 380 return r; 381 } 382 383 static void __exit dm_delay_exit(void) 384 { 385 dm_unregister_target(&delay_target); 386 kmem_cache_destroy(delayed_cache); 387 destroy_workqueue(kdelayd_wq); 388 } 389 390 /* Module hooks */ 391 module_init(dm_delay_init); 392 module_exit(dm_delay_exit); 393 394 MODULE_DESCRIPTION(DM_NAME " delay target"); 395 MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>"); 396 MODULE_LICENSE("GPL"); 397