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 return r; 164 165 ti->num_flush_bios = stripes; 166 ti->num_discard_bios = stripes; 167 ti->num_write_same_bios = stripes; 168 169 sc->chunk_size = chunk_size; 170 if (chunk_size & (chunk_size - 1)) 171 sc->chunk_size_shift = -1; 172 else 173 sc->chunk_size_shift = __ffs(chunk_size); 174 175 /* 176 * Get the stripe destinations. 177 */ 178 for (i = 0; i < stripes; i++) { 179 argv += 2; 180 181 r = get_stripe(ti, sc, i, argv); 182 if (r < 0) { 183 ti->error = "Couldn't parse stripe destination"; 184 while (i--) 185 dm_put_device(ti, sc->stripe[i].dev); 186 kfree(sc); 187 return r; 188 } 189 atomic_set(&(sc->stripe[i].error_count), 0); 190 } 191 192 ti->private = sc; 193 194 return 0; 195 } 196 197 static void stripe_dtr(struct dm_target *ti) 198 { 199 unsigned int i; 200 struct stripe_c *sc = (struct stripe_c *) ti->private; 201 202 for (i = 0; i < sc->stripes; i++) 203 dm_put_device(ti, sc->stripe[i].dev); 204 205 flush_work(&sc->trigger_event); 206 kfree(sc); 207 } 208 209 static void stripe_map_sector(struct stripe_c *sc, sector_t sector, 210 uint32_t *stripe, sector_t *result) 211 { 212 sector_t chunk = dm_target_offset(sc->ti, sector); 213 sector_t chunk_offset; 214 215 if (sc->chunk_size_shift < 0) 216 chunk_offset = sector_div(chunk, sc->chunk_size); 217 else { 218 chunk_offset = chunk & (sc->chunk_size - 1); 219 chunk >>= sc->chunk_size_shift; 220 } 221 222 if (sc->stripes_shift < 0) 223 *stripe = sector_div(chunk, sc->stripes); 224 else { 225 *stripe = chunk & (sc->stripes - 1); 226 chunk >>= sc->stripes_shift; 227 } 228 229 if (sc->chunk_size_shift < 0) 230 chunk *= sc->chunk_size; 231 else 232 chunk <<= sc->chunk_size_shift; 233 234 *result = chunk + chunk_offset; 235 } 236 237 static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector, 238 uint32_t target_stripe, sector_t *result) 239 { 240 uint32_t stripe; 241 242 stripe_map_sector(sc, sector, &stripe, result); 243 if (stripe == target_stripe) 244 return; 245 246 /* round down */ 247 sector = *result; 248 if (sc->chunk_size_shift < 0) 249 *result -= sector_div(sector, sc->chunk_size); 250 else 251 *result = sector & ~(sector_t)(sc->chunk_size - 1); 252 253 if (target_stripe < stripe) 254 *result += sc->chunk_size; /* next chunk */ 255 } 256 257 static int stripe_map_range(struct stripe_c *sc, struct bio *bio, 258 uint32_t target_stripe) 259 { 260 sector_t begin, end; 261 262 stripe_map_range_sector(sc, bio->bi_iter.bi_sector, 263 target_stripe, &begin); 264 stripe_map_range_sector(sc, bio_end_sector(bio), 265 target_stripe, &end); 266 if (begin < end) { 267 bio->bi_bdev = sc->stripe[target_stripe].dev->bdev; 268 bio->bi_iter.bi_sector = begin + 269 sc->stripe[target_stripe].physical_start; 270 bio->bi_iter.bi_size = to_bytes(end - begin); 271 return DM_MAPIO_REMAPPED; 272 } else { 273 /* The range doesn't map to the target stripe */ 274 bio_endio(bio, 0); 275 return DM_MAPIO_SUBMITTED; 276 } 277 } 278 279 static int stripe_map(struct dm_target *ti, struct bio *bio) 280 { 281 struct stripe_c *sc = ti->private; 282 uint32_t stripe; 283 unsigned target_bio_nr; 284 285 if (bio->bi_rw & REQ_FLUSH) { 286 target_bio_nr = dm_bio_get_target_bio_nr(bio); 287 BUG_ON(target_bio_nr >= sc->stripes); 288 bio->bi_bdev = sc->stripe[target_bio_nr].dev->bdev; 289 return DM_MAPIO_REMAPPED; 290 } 291 if (unlikely(bio->bi_rw & REQ_DISCARD) || 292 unlikely(bio->bi_rw & REQ_WRITE_SAME)) { 293 target_bio_nr = dm_bio_get_target_bio_nr(bio); 294 BUG_ON(target_bio_nr >= sc->stripes); 295 return stripe_map_range(sc, bio, target_bio_nr); 296 } 297 298 stripe_map_sector(sc, bio->bi_iter.bi_sector, 299 &stripe, &bio->bi_iter.bi_sector); 300 301 bio->bi_iter.bi_sector += sc->stripe[stripe].physical_start; 302 bio->bi_bdev = sc->stripe[stripe].dev->bdev; 303 304 return DM_MAPIO_REMAPPED; 305 } 306 307 /* 308 * Stripe status: 309 * 310 * INFO 311 * #stripes [stripe_name <stripe_name>] [group word count] 312 * [error count 'A|D' <error count 'A|D'>] 313 * 314 * TABLE 315 * #stripes [stripe chunk size] 316 * [stripe_name physical_start <stripe_name physical_start>] 317 * 318 */ 319 320 static void stripe_status(struct dm_target *ti, status_type_t type, 321 unsigned status_flags, char *result, unsigned maxlen) 322 { 323 struct stripe_c *sc = (struct stripe_c *) ti->private; 324 char buffer[sc->stripes + 1]; 325 unsigned int sz = 0; 326 unsigned int i; 327 328 switch (type) { 329 case STATUSTYPE_INFO: 330 DMEMIT("%d ", sc->stripes); 331 for (i = 0; i < sc->stripes; i++) { 332 DMEMIT("%s ", sc->stripe[i].dev->name); 333 buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ? 334 'D' : 'A'; 335 } 336 buffer[i] = '\0'; 337 DMEMIT("1 %s", buffer); 338 break; 339 340 case STATUSTYPE_TABLE: 341 DMEMIT("%d %llu", sc->stripes, 342 (unsigned long long)sc->chunk_size); 343 for (i = 0; i < sc->stripes; i++) 344 DMEMIT(" %s %llu", sc->stripe[i].dev->name, 345 (unsigned long long)sc->stripe[i].physical_start); 346 break; 347 } 348 } 349 350 static int stripe_end_io(struct dm_target *ti, struct bio *bio, int error) 351 { 352 unsigned i; 353 char major_minor[16]; 354 struct stripe_c *sc = ti->private; 355 356 if (!error) 357 return 0; /* I/O complete */ 358 359 if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD)) 360 return error; 361 362 if (error == -EOPNOTSUPP) 363 return error; 364 365 memset(major_minor, 0, sizeof(major_minor)); 366 sprintf(major_minor, "%d:%d", 367 MAJOR(disk_devt(bio->bi_bdev->bd_disk)), 368 MINOR(disk_devt(bio->bi_bdev->bd_disk))); 369 370 /* 371 * Test to see which stripe drive triggered the event 372 * and increment error count for all stripes on that device. 373 * If the error count for a given device exceeds the threshold 374 * value we will no longer trigger any further events. 375 */ 376 for (i = 0; i < sc->stripes; i++) 377 if (!strcmp(sc->stripe[i].dev->name, major_minor)) { 378 atomic_inc(&(sc->stripe[i].error_count)); 379 if (atomic_read(&(sc->stripe[i].error_count)) < 380 DM_IO_ERROR_THRESHOLD) 381 schedule_work(&sc->trigger_event); 382 } 383 384 return error; 385 } 386 387 static int stripe_iterate_devices(struct dm_target *ti, 388 iterate_devices_callout_fn fn, void *data) 389 { 390 struct stripe_c *sc = ti->private; 391 int ret = 0; 392 unsigned i = 0; 393 394 do { 395 ret = fn(ti, sc->stripe[i].dev, 396 sc->stripe[i].physical_start, 397 sc->stripe_width, data); 398 } while (!ret && ++i < sc->stripes); 399 400 return ret; 401 } 402 403 static void stripe_io_hints(struct dm_target *ti, 404 struct queue_limits *limits) 405 { 406 struct stripe_c *sc = ti->private; 407 unsigned chunk_size = sc->chunk_size << SECTOR_SHIFT; 408 409 blk_limits_io_min(limits, chunk_size); 410 blk_limits_io_opt(limits, chunk_size * sc->stripes); 411 } 412 413 static int stripe_merge(struct dm_target *ti, struct bvec_merge_data *bvm, 414 struct bio_vec *biovec, int max_size) 415 { 416 struct stripe_c *sc = ti->private; 417 sector_t bvm_sector = bvm->bi_sector; 418 uint32_t stripe; 419 struct request_queue *q; 420 421 stripe_map_sector(sc, bvm_sector, &stripe, &bvm_sector); 422 423 q = bdev_get_queue(sc->stripe[stripe].dev->bdev); 424 if (!q->merge_bvec_fn) 425 return max_size; 426 427 bvm->bi_bdev = sc->stripe[stripe].dev->bdev; 428 bvm->bi_sector = sc->stripe[stripe].physical_start + bvm_sector; 429 430 return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); 431 } 432 433 static struct target_type stripe_target = { 434 .name = "striped", 435 .version = {1, 5, 1}, 436 .module = THIS_MODULE, 437 .ctr = stripe_ctr, 438 .dtr = stripe_dtr, 439 .map = stripe_map, 440 .end_io = stripe_end_io, 441 .status = stripe_status, 442 .iterate_devices = stripe_iterate_devices, 443 .io_hints = stripe_io_hints, 444 .merge = stripe_merge, 445 }; 446 447 int __init dm_stripe_init(void) 448 { 449 int r; 450 451 r = dm_register_target(&stripe_target); 452 if (r < 0) { 453 DMWARN("target registration failed"); 454 return r; 455 } 456 457 return r; 458 } 459 460 void dm_stripe_exit(void) 461 { 462 dm_unregister_target(&stripe_target); 463 } 464