1 /* 2 * Copyright (C) 2001-2003 Sistina Software (UK) Limited. 3 * 4 * This file is released under the GPL. 5 */ 6 7 #include <linux/device-mapper.h> 8 9 #include <linux/module.h> 10 #include <linux/init.h> 11 #include <linux/blkdev.h> 12 #include <linux/bio.h> 13 #include <linux/slab.h> 14 #include <linux/log2.h> 15 16 #define DM_MSG_PREFIX "striped" 17 #define DM_IO_ERROR_THRESHOLD 15 18 19 struct stripe { 20 struct dm_dev *dev; 21 sector_t physical_start; 22 23 atomic_t error_count; 24 }; 25 26 struct stripe_c { 27 uint32_t stripes; 28 int stripes_shift; 29 30 /* The size of this target / num. stripes */ 31 sector_t stripe_width; 32 33 uint32_t chunk_size; 34 int chunk_size_shift; 35 36 /* Needed for handling events */ 37 struct dm_target *ti; 38 39 /* Work struct used for triggering events*/ 40 struct work_struct trigger_event; 41 42 struct stripe stripe[0]; 43 }; 44 45 /* 46 * An event is triggered whenever a drive 47 * drops out of a stripe volume. 48 */ 49 static void trigger_event(struct work_struct *work) 50 { 51 struct stripe_c *sc = container_of(work, struct stripe_c, 52 trigger_event); 53 dm_table_event(sc->ti->table); 54 } 55 56 static inline struct stripe_c *alloc_context(unsigned int stripes) 57 { 58 size_t len; 59 60 if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe), 61 stripes)) 62 return NULL; 63 64 len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes); 65 66 return kmalloc(len, GFP_KERNEL); 67 } 68 69 /* 70 * Parse a single <dev> <sector> pair 71 */ 72 static int get_stripe(struct dm_target *ti, struct stripe_c *sc, 73 unsigned int stripe, char **argv) 74 { 75 unsigned long long start; 76 char dummy; 77 78 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1) 79 return -EINVAL; 80 81 if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), 82 &sc->stripe[stripe].dev)) 83 return -ENXIO; 84 85 sc->stripe[stripe].physical_start = start; 86 87 return 0; 88 } 89 90 /* 91 * Construct a striped mapping. 92 * <number of stripes> <chunk size> [<dev_path> <offset>]+ 93 */ 94 static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv) 95 { 96 struct stripe_c *sc; 97 sector_t width; 98 uint32_t stripes; 99 uint32_t chunk_size; 100 int r; 101 unsigned int i; 102 103 if (argc < 2) { 104 ti->error = "Not enough arguments"; 105 return -EINVAL; 106 } 107 108 if (kstrtouint(argv[0], 10, &stripes) || !stripes) { 109 ti->error = "Invalid stripe count"; 110 return -EINVAL; 111 } 112 113 if (kstrtouint(argv[1], 10, &chunk_size) || !chunk_size) { 114 ti->error = "Invalid chunk_size"; 115 return -EINVAL; 116 } 117 118 width = ti->len; 119 if (sector_div(width, chunk_size)) { 120 ti->error = "Target length not divisible by " 121 "chunk size"; 122 return -EINVAL; 123 } 124 125 if (sector_div(width, stripes)) { 126 ti->error = "Target length not divisible by " 127 "number of stripes"; 128 return -EINVAL; 129 } 130 131 /* 132 * Do we have enough arguments for that many stripes ? 133 */ 134 if (argc != (2 + 2 * stripes)) { 135 ti->error = "Not enough destinations " 136 "specified"; 137 return -EINVAL; 138 } 139 140 sc = alloc_context(stripes); 141 if (!sc) { 142 ti->error = "Memory allocation for striped context " 143 "failed"; 144 return -ENOMEM; 145 } 146 147 INIT_WORK(&sc->trigger_event, trigger_event); 148 149 /* Set pointer to dm target; used in trigger_event */ 150 sc->ti = ti; 151 sc->stripes = stripes; 152 sc->stripe_width = width; 153 154 if (stripes & (stripes - 1)) 155 sc->stripes_shift = -1; 156 else 157 sc->stripes_shift = __ffs(stripes); 158 159 r = dm_set_target_max_io_len(ti, chunk_size); 160 if (r) 161 return r; 162 163 ti->num_flush_requests = stripes; 164 ti->num_discard_requests = stripes; 165 166 sc->chunk_size = chunk_size; 167 if (chunk_size & (chunk_size - 1)) 168 sc->chunk_size_shift = -1; 169 else 170 sc->chunk_size_shift = __ffs(chunk_size); 171 172 /* 173 * Get the stripe destinations. 174 */ 175 for (i = 0; i < stripes; i++) { 176 argv += 2; 177 178 r = get_stripe(ti, sc, i, argv); 179 if (r < 0) { 180 ti->error = "Couldn't parse stripe destination"; 181 while (i--) 182 dm_put_device(ti, sc->stripe[i].dev); 183 kfree(sc); 184 return r; 185 } 186 atomic_set(&(sc->stripe[i].error_count), 0); 187 } 188 189 ti->private = sc; 190 191 return 0; 192 } 193 194 static void stripe_dtr(struct dm_target *ti) 195 { 196 unsigned int i; 197 struct stripe_c *sc = (struct stripe_c *) ti->private; 198 199 for (i = 0; i < sc->stripes; i++) 200 dm_put_device(ti, sc->stripe[i].dev); 201 202 flush_work_sync(&sc->trigger_event); 203 kfree(sc); 204 } 205 206 static void stripe_map_sector(struct stripe_c *sc, sector_t sector, 207 uint32_t *stripe, sector_t *result) 208 { 209 sector_t chunk = dm_target_offset(sc->ti, sector); 210 sector_t chunk_offset; 211 212 if (sc->chunk_size_shift < 0) 213 chunk_offset = sector_div(chunk, sc->chunk_size); 214 else { 215 chunk_offset = chunk & (sc->chunk_size - 1); 216 chunk >>= sc->chunk_size_shift; 217 } 218 219 if (sc->stripes_shift < 0) 220 *stripe = sector_div(chunk, sc->stripes); 221 else { 222 *stripe = chunk & (sc->stripes - 1); 223 chunk >>= sc->stripes_shift; 224 } 225 226 if (sc->chunk_size_shift < 0) 227 chunk *= sc->chunk_size; 228 else 229 chunk <<= sc->chunk_size_shift; 230 231 *result = chunk + chunk_offset; 232 } 233 234 static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector, 235 uint32_t target_stripe, sector_t *result) 236 { 237 uint32_t stripe; 238 239 stripe_map_sector(sc, sector, &stripe, result); 240 if (stripe == target_stripe) 241 return; 242 243 /* round down */ 244 sector = *result; 245 if (sc->chunk_size_shift < 0) 246 *result -= sector_div(sector, sc->chunk_size); 247 else 248 *result = sector & ~(sector_t)(sc->chunk_size - 1); 249 250 if (target_stripe < stripe) 251 *result += sc->chunk_size; /* next chunk */ 252 } 253 254 static int stripe_map_discard(struct stripe_c *sc, struct bio *bio, 255 uint32_t target_stripe) 256 { 257 sector_t begin, end; 258 259 stripe_map_range_sector(sc, bio->bi_sector, target_stripe, &begin); 260 stripe_map_range_sector(sc, bio->bi_sector + bio_sectors(bio), 261 target_stripe, &end); 262 if (begin < end) { 263 bio->bi_bdev = sc->stripe[target_stripe].dev->bdev; 264 bio->bi_sector = begin + sc->stripe[target_stripe].physical_start; 265 bio->bi_size = to_bytes(end - begin); 266 return DM_MAPIO_REMAPPED; 267 } else { 268 /* The range doesn't map to the target stripe */ 269 bio_endio(bio, 0); 270 return DM_MAPIO_SUBMITTED; 271 } 272 } 273 274 static int stripe_map(struct dm_target *ti, struct bio *bio, 275 union map_info *map_context) 276 { 277 struct stripe_c *sc = ti->private; 278 uint32_t stripe; 279 unsigned target_request_nr; 280 281 if (bio->bi_rw & REQ_FLUSH) { 282 target_request_nr = map_context->target_request_nr; 283 BUG_ON(target_request_nr >= sc->stripes); 284 bio->bi_bdev = sc->stripe[target_request_nr].dev->bdev; 285 return DM_MAPIO_REMAPPED; 286 } 287 if (unlikely(bio->bi_rw & REQ_DISCARD)) { 288 target_request_nr = map_context->target_request_nr; 289 BUG_ON(target_request_nr >= sc->stripes); 290 return stripe_map_discard(sc, bio, target_request_nr); 291 } 292 293 stripe_map_sector(sc, bio->bi_sector, &stripe, &bio->bi_sector); 294 295 bio->bi_sector += sc->stripe[stripe].physical_start; 296 bio->bi_bdev = sc->stripe[stripe].dev->bdev; 297 298 return DM_MAPIO_REMAPPED; 299 } 300 301 /* 302 * Stripe status: 303 * 304 * INFO 305 * #stripes [stripe_name <stripe_name>] [group word count] 306 * [error count 'A|D' <error count 'A|D'>] 307 * 308 * TABLE 309 * #stripes [stripe chunk size] 310 * [stripe_name physical_start <stripe_name physical_start>] 311 * 312 */ 313 314 static int stripe_status(struct dm_target *ti, status_type_t type, 315 unsigned status_flags, char *result, unsigned maxlen) 316 { 317 struct stripe_c *sc = (struct stripe_c *) ti->private; 318 char buffer[sc->stripes + 1]; 319 unsigned int sz = 0; 320 unsigned int i; 321 322 switch (type) { 323 case STATUSTYPE_INFO: 324 DMEMIT("%d ", sc->stripes); 325 for (i = 0; i < sc->stripes; i++) { 326 DMEMIT("%s ", sc->stripe[i].dev->name); 327 buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ? 328 'D' : 'A'; 329 } 330 buffer[i] = '\0'; 331 DMEMIT("1 %s", buffer); 332 break; 333 334 case STATUSTYPE_TABLE: 335 DMEMIT("%d %llu", sc->stripes, 336 (unsigned long long)sc->chunk_size); 337 for (i = 0; i < sc->stripes; i++) 338 DMEMIT(" %s %llu", sc->stripe[i].dev->name, 339 (unsigned long long)sc->stripe[i].physical_start); 340 break; 341 } 342 return 0; 343 } 344 345 static int stripe_end_io(struct dm_target *ti, struct bio *bio, 346 int error, union map_info *map_context) 347 { 348 unsigned i; 349 char major_minor[16]; 350 struct stripe_c *sc = ti->private; 351 352 if (!error) 353 return 0; /* I/O complete */ 354 355 if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD)) 356 return error; 357 358 if (error == -EOPNOTSUPP) 359 return error; 360 361 memset(major_minor, 0, sizeof(major_minor)); 362 sprintf(major_minor, "%d:%d", 363 MAJOR(disk_devt(bio->bi_bdev->bd_disk)), 364 MINOR(disk_devt(bio->bi_bdev->bd_disk))); 365 366 /* 367 * Test to see which stripe drive triggered the event 368 * and increment error count for all stripes on that device. 369 * If the error count for a given device exceeds the threshold 370 * value we will no longer trigger any further events. 371 */ 372 for (i = 0; i < sc->stripes; i++) 373 if (!strcmp(sc->stripe[i].dev->name, major_minor)) { 374 atomic_inc(&(sc->stripe[i].error_count)); 375 if (atomic_read(&(sc->stripe[i].error_count)) < 376 DM_IO_ERROR_THRESHOLD) 377 schedule_work(&sc->trigger_event); 378 } 379 380 return error; 381 } 382 383 static int stripe_iterate_devices(struct dm_target *ti, 384 iterate_devices_callout_fn fn, void *data) 385 { 386 struct stripe_c *sc = ti->private; 387 int ret = 0; 388 unsigned i = 0; 389 390 do { 391 ret = fn(ti, sc->stripe[i].dev, 392 sc->stripe[i].physical_start, 393 sc->stripe_width, data); 394 } while (!ret && ++i < sc->stripes); 395 396 return ret; 397 } 398 399 static void stripe_io_hints(struct dm_target *ti, 400 struct queue_limits *limits) 401 { 402 struct stripe_c *sc = ti->private; 403 unsigned chunk_size = sc->chunk_size << SECTOR_SHIFT; 404 405 blk_limits_io_min(limits, chunk_size); 406 blk_limits_io_opt(limits, chunk_size * sc->stripes); 407 } 408 409 static int stripe_merge(struct dm_target *ti, struct bvec_merge_data *bvm, 410 struct bio_vec *biovec, int max_size) 411 { 412 struct stripe_c *sc = ti->private; 413 sector_t bvm_sector = bvm->bi_sector; 414 uint32_t stripe; 415 struct request_queue *q; 416 417 stripe_map_sector(sc, bvm_sector, &stripe, &bvm_sector); 418 419 q = bdev_get_queue(sc->stripe[stripe].dev->bdev); 420 if (!q->merge_bvec_fn) 421 return max_size; 422 423 bvm->bi_bdev = sc->stripe[stripe].dev->bdev; 424 bvm->bi_sector = sc->stripe[stripe].physical_start + bvm_sector; 425 426 return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); 427 } 428 429 static struct target_type stripe_target = { 430 .name = "striped", 431 .version = {1, 5, 0}, 432 .module = THIS_MODULE, 433 .ctr = stripe_ctr, 434 .dtr = stripe_dtr, 435 .map = stripe_map, 436 .end_io = stripe_end_io, 437 .status = stripe_status, 438 .iterate_devices = stripe_iterate_devices, 439 .io_hints = stripe_io_hints, 440 .merge = stripe_merge, 441 }; 442 443 int __init dm_stripe_init(void) 444 { 445 int r; 446 447 r = dm_register_target(&stripe_target); 448 if (r < 0) { 449 DMWARN("target registration failed"); 450 return r; 451 } 452 453 return r; 454 } 455 456 void dm_stripe_exit(void) 457 { 458 dm_unregister_target(&stripe_target); 459 } 460