1 /* 2 * Copyright (C) 2001-2003 Sistina Software (UK) Limited. 3 * 4 * This file is released under the GPL. 5 */ 6 7 #include "dm.h" 8 #include <linux/device-mapper.h> 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 #include <linux/log2.h> 16 17 #define DM_MSG_PREFIX "striped" 18 #define DM_IO_ERROR_THRESHOLD 15 19 20 struct stripe { 21 struct dm_dev *dev; 22 sector_t physical_start; 23 24 atomic_t error_count; 25 }; 26 27 struct stripe_c { 28 uint32_t stripes; 29 int stripes_shift; 30 31 /* The size of this target / num. stripes */ 32 sector_t stripe_width; 33 34 uint32_t chunk_size; 35 int chunk_size_shift; 36 37 /* Needed for handling events */ 38 struct dm_target *ti; 39 40 /* Work struct used for triggering events*/ 41 struct work_struct trigger_event; 42 43 struct stripe stripe[0]; 44 }; 45 46 /* 47 * An event is triggered whenever a drive 48 * drops out of a stripe volume. 49 */ 50 static void trigger_event(struct work_struct *work) 51 { 52 struct stripe_c *sc = container_of(work, struct stripe_c, 53 trigger_event); 54 dm_table_event(sc->ti->table); 55 } 56 57 static inline struct stripe_c *alloc_context(unsigned int stripes) 58 { 59 size_t len; 60 61 if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe), 62 stripes)) 63 return NULL; 64 65 len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes); 66 67 return kmalloc(len, GFP_KERNEL); 68 } 69 70 /* 71 * Parse a single <dev> <sector> pair 72 */ 73 static int get_stripe(struct dm_target *ti, struct stripe_c *sc, 74 unsigned int stripe, char **argv) 75 { 76 unsigned long long start; 77 char dummy; 78 79 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1) 80 return -EINVAL; 81 82 if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), 83 &sc->stripe[stripe].dev)) 84 return -ENXIO; 85 86 sc->stripe[stripe].physical_start = start; 87 88 return 0; 89 } 90 91 /* 92 * Construct a striped mapping. 93 * <number of stripes> <chunk size> [<dev_path> <offset>]+ 94 */ 95 static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv) 96 { 97 struct stripe_c *sc; 98 sector_t width, tmp_len; 99 uint32_t stripes; 100 uint32_t chunk_size; 101 int r; 102 unsigned int i; 103 104 if (argc < 2) { 105 ti->error = "Not enough arguments"; 106 return -EINVAL; 107 } 108 109 if (kstrtouint(argv[0], 10, &stripes) || !stripes) { 110 ti->error = "Invalid stripe count"; 111 return -EINVAL; 112 } 113 114 if (kstrtouint(argv[1], 10, &chunk_size) || !chunk_size) { 115 ti->error = "Invalid chunk_size"; 116 return -EINVAL; 117 } 118 119 width = ti->len; 120 if (sector_div(width, stripes)) { 121 ti->error = "Target length not divisible by " 122 "number of stripes"; 123 return -EINVAL; 124 } 125 126 tmp_len = width; 127 if (sector_div(tmp_len, chunk_size)) { 128 ti->error = "Target length not divisible by " 129 "chunk size"; 130 return -EINVAL; 131 } 132 133 /* 134 * Do we have enough arguments for that many stripes ? 135 */ 136 if (argc != (2 + 2 * stripes)) { 137 ti->error = "Not enough destinations " 138 "specified"; 139 return -EINVAL; 140 } 141 142 sc = alloc_context(stripes); 143 if (!sc) { 144 ti->error = "Memory allocation for striped context " 145 "failed"; 146 return -ENOMEM; 147 } 148 149 INIT_WORK(&sc->trigger_event, trigger_event); 150 151 /* Set pointer to dm target; used in trigger_event */ 152 sc->ti = ti; 153 sc->stripes = stripes; 154 sc->stripe_width = width; 155 156 if (stripes & (stripes - 1)) 157 sc->stripes_shift = -1; 158 else 159 sc->stripes_shift = __ffs(stripes); 160 161 r = dm_set_target_max_io_len(ti, chunk_size); 162 if (r) { 163 kfree(sc); 164 return r; 165 } 166 167 ti->num_flush_bios = stripes; 168 ti->num_discard_bios = stripes; 169 ti->num_write_same_bios = stripes; 170 171 sc->chunk_size = chunk_size; 172 if (chunk_size & (chunk_size - 1)) 173 sc->chunk_size_shift = -1; 174 else 175 sc->chunk_size_shift = __ffs(chunk_size); 176 177 /* 178 * Get the stripe destinations. 179 */ 180 for (i = 0; i < stripes; i++) { 181 argv += 2; 182 183 r = get_stripe(ti, sc, i, argv); 184 if (r < 0) { 185 ti->error = "Couldn't parse stripe destination"; 186 while (i--) 187 dm_put_device(ti, sc->stripe[i].dev); 188 kfree(sc); 189 return r; 190 } 191 atomic_set(&(sc->stripe[i].error_count), 0); 192 } 193 194 ti->private = sc; 195 196 return 0; 197 } 198 199 static void stripe_dtr(struct dm_target *ti) 200 { 201 unsigned int i; 202 struct stripe_c *sc = (struct stripe_c *) ti->private; 203 204 for (i = 0; i < sc->stripes; i++) 205 dm_put_device(ti, sc->stripe[i].dev); 206 207 flush_work(&sc->trigger_event); 208 kfree(sc); 209 } 210 211 static void stripe_map_sector(struct stripe_c *sc, sector_t sector, 212 uint32_t *stripe, sector_t *result) 213 { 214 sector_t chunk = dm_target_offset(sc->ti, sector); 215 sector_t chunk_offset; 216 217 if (sc->chunk_size_shift < 0) 218 chunk_offset = sector_div(chunk, sc->chunk_size); 219 else { 220 chunk_offset = chunk & (sc->chunk_size - 1); 221 chunk >>= sc->chunk_size_shift; 222 } 223 224 if (sc->stripes_shift < 0) 225 *stripe = sector_div(chunk, sc->stripes); 226 else { 227 *stripe = chunk & (sc->stripes - 1); 228 chunk >>= sc->stripes_shift; 229 } 230 231 if (sc->chunk_size_shift < 0) 232 chunk *= sc->chunk_size; 233 else 234 chunk <<= sc->chunk_size_shift; 235 236 *result = chunk + chunk_offset; 237 } 238 239 static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector, 240 uint32_t target_stripe, sector_t *result) 241 { 242 uint32_t stripe; 243 244 stripe_map_sector(sc, sector, &stripe, result); 245 if (stripe == target_stripe) 246 return; 247 248 /* round down */ 249 sector = *result; 250 if (sc->chunk_size_shift < 0) 251 *result -= sector_div(sector, sc->chunk_size); 252 else 253 *result = sector & ~(sector_t)(sc->chunk_size - 1); 254 255 if (target_stripe < stripe) 256 *result += sc->chunk_size; /* next chunk */ 257 } 258 259 static int stripe_map_range(struct stripe_c *sc, struct bio *bio, 260 uint32_t target_stripe) 261 { 262 sector_t begin, end; 263 264 stripe_map_range_sector(sc, bio->bi_iter.bi_sector, 265 target_stripe, &begin); 266 stripe_map_range_sector(sc, bio_end_sector(bio), 267 target_stripe, &end); 268 if (begin < end) { 269 bio->bi_bdev = sc->stripe[target_stripe].dev->bdev; 270 bio->bi_iter.bi_sector = begin + 271 sc->stripe[target_stripe].physical_start; 272 bio->bi_iter.bi_size = to_bytes(end - begin); 273 return DM_MAPIO_REMAPPED; 274 } else { 275 /* The range doesn't map to the target stripe */ 276 bio_endio(bio, 0); 277 return DM_MAPIO_SUBMITTED; 278 } 279 } 280 281 static int stripe_map(struct dm_target *ti, struct bio *bio) 282 { 283 struct stripe_c *sc = ti->private; 284 uint32_t stripe; 285 unsigned target_bio_nr; 286 287 if (bio->bi_rw & REQ_FLUSH) { 288 target_bio_nr = dm_bio_get_target_bio_nr(bio); 289 BUG_ON(target_bio_nr >= sc->stripes); 290 bio->bi_bdev = sc->stripe[target_bio_nr].dev->bdev; 291 return DM_MAPIO_REMAPPED; 292 } 293 if (unlikely(bio->bi_rw & REQ_DISCARD) || 294 unlikely(bio->bi_rw & REQ_WRITE_SAME)) { 295 target_bio_nr = dm_bio_get_target_bio_nr(bio); 296 BUG_ON(target_bio_nr >= sc->stripes); 297 return stripe_map_range(sc, bio, target_bio_nr); 298 } 299 300 stripe_map_sector(sc, bio->bi_iter.bi_sector, 301 &stripe, &bio->bi_iter.bi_sector); 302 303 bio->bi_iter.bi_sector += sc->stripe[stripe].physical_start; 304 bio->bi_bdev = sc->stripe[stripe].dev->bdev; 305 306 return DM_MAPIO_REMAPPED; 307 } 308 309 /* 310 * Stripe status: 311 * 312 * INFO 313 * #stripes [stripe_name <stripe_name>] [group word count] 314 * [error count 'A|D' <error count 'A|D'>] 315 * 316 * TABLE 317 * #stripes [stripe chunk size] 318 * [stripe_name physical_start <stripe_name physical_start>] 319 * 320 */ 321 322 static void stripe_status(struct dm_target *ti, status_type_t type, 323 unsigned status_flags, char *result, unsigned maxlen) 324 { 325 struct stripe_c *sc = (struct stripe_c *) ti->private; 326 char buffer[sc->stripes + 1]; 327 unsigned int sz = 0; 328 unsigned int i; 329 330 switch (type) { 331 case STATUSTYPE_INFO: 332 DMEMIT("%d ", sc->stripes); 333 for (i = 0; i < sc->stripes; i++) { 334 DMEMIT("%s ", sc->stripe[i].dev->name); 335 buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ? 336 'D' : 'A'; 337 } 338 buffer[i] = '\0'; 339 DMEMIT("1 %s", buffer); 340 break; 341 342 case STATUSTYPE_TABLE: 343 DMEMIT("%d %llu", sc->stripes, 344 (unsigned long long)sc->chunk_size); 345 for (i = 0; i < sc->stripes; i++) 346 DMEMIT(" %s %llu", sc->stripe[i].dev->name, 347 (unsigned long long)sc->stripe[i].physical_start); 348 break; 349 } 350 } 351 352 static int stripe_end_io(struct dm_target *ti, struct bio *bio, int error) 353 { 354 unsigned i; 355 char major_minor[16]; 356 struct stripe_c *sc = ti->private; 357 358 if (!error) 359 return 0; /* I/O complete */ 360 361 if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD)) 362 return error; 363 364 if (error == -EOPNOTSUPP) 365 return error; 366 367 memset(major_minor, 0, sizeof(major_minor)); 368 sprintf(major_minor, "%d:%d", 369 MAJOR(disk_devt(bio->bi_bdev->bd_disk)), 370 MINOR(disk_devt(bio->bi_bdev->bd_disk))); 371 372 /* 373 * Test to see which stripe drive triggered the event 374 * and increment error count for all stripes on that device. 375 * If the error count for a given device exceeds the threshold 376 * value we will no longer trigger any further events. 377 */ 378 for (i = 0; i < sc->stripes; i++) 379 if (!strcmp(sc->stripe[i].dev->name, major_minor)) { 380 atomic_inc(&(sc->stripe[i].error_count)); 381 if (atomic_read(&(sc->stripe[i].error_count)) < 382 DM_IO_ERROR_THRESHOLD) 383 schedule_work(&sc->trigger_event); 384 } 385 386 return error; 387 } 388 389 static int stripe_iterate_devices(struct dm_target *ti, 390 iterate_devices_callout_fn fn, void *data) 391 { 392 struct stripe_c *sc = ti->private; 393 int ret = 0; 394 unsigned i = 0; 395 396 do { 397 ret = fn(ti, sc->stripe[i].dev, 398 sc->stripe[i].physical_start, 399 sc->stripe_width, data); 400 } while (!ret && ++i < sc->stripes); 401 402 return ret; 403 } 404 405 static void stripe_io_hints(struct dm_target *ti, 406 struct queue_limits *limits) 407 { 408 struct stripe_c *sc = ti->private; 409 unsigned chunk_size = sc->chunk_size << SECTOR_SHIFT; 410 411 blk_limits_io_min(limits, chunk_size); 412 blk_limits_io_opt(limits, chunk_size * sc->stripes); 413 } 414 415 static int stripe_merge(struct dm_target *ti, struct bvec_merge_data *bvm, 416 struct bio_vec *biovec, int max_size) 417 { 418 struct stripe_c *sc = ti->private; 419 sector_t bvm_sector = bvm->bi_sector; 420 uint32_t stripe; 421 struct request_queue *q; 422 423 stripe_map_sector(sc, bvm_sector, &stripe, &bvm_sector); 424 425 q = bdev_get_queue(sc->stripe[stripe].dev->bdev); 426 if (!q->merge_bvec_fn) 427 return max_size; 428 429 bvm->bi_bdev = sc->stripe[stripe].dev->bdev; 430 bvm->bi_sector = sc->stripe[stripe].physical_start + bvm_sector; 431 432 return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); 433 } 434 435 static struct target_type stripe_target = { 436 .name = "striped", 437 .version = {1, 5, 1}, 438 .module = THIS_MODULE, 439 .ctr = stripe_ctr, 440 .dtr = stripe_dtr, 441 .map = stripe_map, 442 .end_io = stripe_end_io, 443 .status = stripe_status, 444 .iterate_devices = stripe_iterate_devices, 445 .io_hints = stripe_io_hints, 446 .merge = stripe_merge, 447 }; 448 449 int __init dm_stripe_init(void) 450 { 451 int r; 452 453 r = dm_register_target(&stripe_target); 454 if (r < 0) 455 DMWARN("target registration failed"); 456 457 return r; 458 } 459 460 void dm_stripe_exit(void) 461 { 462 dm_unregister_target(&stripe_target); 463 } 464