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 sector_t stripes_mask; 30 31 /* The size of this target / num. stripes */ 32 sector_t stripe_width; 33 34 /* stripe chunk size */ 35 uint32_t chunk_shift; 36 sector_t chunk_mask; 37 38 /* Needed for handling events */ 39 struct dm_target *ti; 40 41 /* Work struct used for triggering events*/ 42 struct work_struct trigger_event; 43 44 struct stripe stripe[0]; 45 }; 46 47 /* 48 * An event is triggered whenever a drive 49 * drops out of a stripe volume. 50 */ 51 static void trigger_event(struct work_struct *work) 52 { 53 struct stripe_c *sc = container_of(work, struct stripe_c, 54 trigger_event); 55 dm_table_event(sc->ti->table); 56 } 57 58 static inline struct stripe_c *alloc_context(unsigned int stripes) 59 { 60 size_t len; 61 62 if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe), 63 stripes)) 64 return NULL; 65 66 len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes); 67 68 return kmalloc(len, GFP_KERNEL); 69 } 70 71 /* 72 * Parse a single <dev> <sector> pair 73 */ 74 static int get_stripe(struct dm_target *ti, struct stripe_c *sc, 75 unsigned int stripe, char **argv) 76 { 77 unsigned long long start; 78 char dummy; 79 80 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1) 81 return -EINVAL; 82 83 if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), 84 &sc->stripe[stripe].dev)) 85 return -ENXIO; 86 87 sc->stripe[stripe].physical_start = start; 88 89 return 0; 90 } 91 92 /* 93 * Construct a striped mapping. 94 * <number of stripes> <chunk size (2^^n)> [<dev_path> <offset>]+ 95 */ 96 static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv) 97 { 98 struct stripe_c *sc; 99 sector_t width; 100 uint32_t stripes; 101 uint32_t chunk_size; 102 char *end; 103 int r; 104 unsigned int i; 105 106 if (argc < 2) { 107 ti->error = "Not enough arguments"; 108 return -EINVAL; 109 } 110 111 stripes = simple_strtoul(argv[0], &end, 10); 112 if (!stripes || *end) { 113 ti->error = "Invalid stripe count"; 114 return -EINVAL; 115 } 116 117 chunk_size = simple_strtoul(argv[1], &end, 10); 118 if (*end) { 119 ti->error = "Invalid chunk_size"; 120 return -EINVAL; 121 } 122 123 /* 124 * chunk_size is a power of two 125 */ 126 if (!is_power_of_2(chunk_size) || 127 (chunk_size < (PAGE_SIZE >> SECTOR_SHIFT))) { 128 ti->error = "Invalid chunk size"; 129 return -EINVAL; 130 } 131 132 if (ti->len & (chunk_size - 1)) { 133 ti->error = "Target length not divisible by " 134 "chunk size"; 135 return -EINVAL; 136 } 137 138 width = ti->len; 139 if (sector_div(width, stripes)) { 140 ti->error = "Target length not divisible by " 141 "number of stripes"; 142 return -EINVAL; 143 } 144 145 /* 146 * Do we have enough arguments for that many stripes ? 147 */ 148 if (argc != (2 + 2 * stripes)) { 149 ti->error = "Not enough destinations " 150 "specified"; 151 return -EINVAL; 152 } 153 154 sc = alloc_context(stripes); 155 if (!sc) { 156 ti->error = "Memory allocation for striped context " 157 "failed"; 158 return -ENOMEM; 159 } 160 161 INIT_WORK(&sc->trigger_event, trigger_event); 162 163 /* Set pointer to dm target; used in trigger_event */ 164 sc->ti = ti; 165 sc->stripes = stripes; 166 sc->stripe_width = width; 167 168 if (stripes & (stripes - 1)) 169 sc->stripes_shift = -1; 170 else { 171 sc->stripes_shift = ffs(stripes) - 1; 172 sc->stripes_mask = ((sector_t) stripes) - 1; 173 } 174 175 ti->split_io = chunk_size; 176 ti->num_flush_requests = stripes; 177 ti->num_discard_requests = stripes; 178 179 sc->chunk_shift = ffs(chunk_size) - 1; 180 sc->chunk_mask = ((sector_t) chunk_size) - 1; 181 182 /* 183 * Get the stripe destinations. 184 */ 185 for (i = 0; i < stripes; i++) { 186 argv += 2; 187 188 r = get_stripe(ti, sc, i, argv); 189 if (r < 0) { 190 ti->error = "Couldn't parse stripe destination"; 191 while (i--) 192 dm_put_device(ti, sc->stripe[i].dev); 193 kfree(sc); 194 return r; 195 } 196 atomic_set(&(sc->stripe[i].error_count), 0); 197 } 198 199 ti->private = sc; 200 201 return 0; 202 } 203 204 static void stripe_dtr(struct dm_target *ti) 205 { 206 unsigned int i; 207 struct stripe_c *sc = (struct stripe_c *) ti->private; 208 209 for (i = 0; i < sc->stripes; i++) 210 dm_put_device(ti, sc->stripe[i].dev); 211 212 flush_work_sync(&sc->trigger_event); 213 kfree(sc); 214 } 215 216 static void stripe_map_sector(struct stripe_c *sc, sector_t sector, 217 uint32_t *stripe, sector_t *result) 218 { 219 sector_t offset = dm_target_offset(sc->ti, sector); 220 sector_t chunk = offset >> sc->chunk_shift; 221 222 if (sc->stripes_shift < 0) 223 *stripe = sector_div(chunk, sc->stripes); 224 else { 225 *stripe = chunk & sc->stripes_mask; 226 chunk >>= sc->stripes_shift; 227 } 228 229 *result = (chunk << sc->chunk_shift) | (offset & sc->chunk_mask); 230 } 231 232 static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector, 233 uint32_t target_stripe, sector_t *result) 234 { 235 uint32_t stripe; 236 237 stripe_map_sector(sc, sector, &stripe, result); 238 if (stripe == target_stripe) 239 return; 240 *result &= ~sc->chunk_mask; /* round down */ 241 if (target_stripe < stripe) 242 *result += sc->chunk_mask + 1; /* next chunk */ 243 } 244 245 static int stripe_map_discard(struct stripe_c *sc, struct bio *bio, 246 uint32_t target_stripe) 247 { 248 sector_t begin, end; 249 250 stripe_map_range_sector(sc, bio->bi_sector, target_stripe, &begin); 251 stripe_map_range_sector(sc, bio->bi_sector + bio_sectors(bio), 252 target_stripe, &end); 253 if (begin < end) { 254 bio->bi_bdev = sc->stripe[target_stripe].dev->bdev; 255 bio->bi_sector = begin + sc->stripe[target_stripe].physical_start; 256 bio->bi_size = to_bytes(end - begin); 257 return DM_MAPIO_REMAPPED; 258 } else { 259 /* The range doesn't map to the target stripe */ 260 bio_endio(bio, 0); 261 return DM_MAPIO_SUBMITTED; 262 } 263 } 264 265 static int stripe_map(struct dm_target *ti, struct bio *bio, 266 union map_info *map_context) 267 { 268 struct stripe_c *sc = ti->private; 269 uint32_t stripe; 270 unsigned target_request_nr; 271 272 if (bio->bi_rw & REQ_FLUSH) { 273 target_request_nr = map_context->target_request_nr; 274 BUG_ON(target_request_nr >= sc->stripes); 275 bio->bi_bdev = sc->stripe[target_request_nr].dev->bdev; 276 return DM_MAPIO_REMAPPED; 277 } 278 if (unlikely(bio->bi_rw & REQ_DISCARD)) { 279 target_request_nr = map_context->target_request_nr; 280 BUG_ON(target_request_nr >= sc->stripes); 281 return stripe_map_discard(sc, bio, target_request_nr); 282 } 283 284 stripe_map_sector(sc, bio->bi_sector, &stripe, &bio->bi_sector); 285 286 bio->bi_sector += sc->stripe[stripe].physical_start; 287 bio->bi_bdev = sc->stripe[stripe].dev->bdev; 288 289 return DM_MAPIO_REMAPPED; 290 } 291 292 /* 293 * Stripe status: 294 * 295 * INFO 296 * #stripes [stripe_name <stripe_name>] [group word count] 297 * [error count 'A|D' <error count 'A|D'>] 298 * 299 * TABLE 300 * #stripes [stripe chunk size] 301 * [stripe_name physical_start <stripe_name physical_start>] 302 * 303 */ 304 305 static int stripe_status(struct dm_target *ti, 306 status_type_t type, char *result, unsigned int maxlen) 307 { 308 struct stripe_c *sc = (struct stripe_c *) ti->private; 309 char buffer[sc->stripes + 1]; 310 unsigned int sz = 0; 311 unsigned int i; 312 313 switch (type) { 314 case STATUSTYPE_INFO: 315 DMEMIT("%d ", sc->stripes); 316 for (i = 0; i < sc->stripes; i++) { 317 DMEMIT("%s ", sc->stripe[i].dev->name); 318 buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ? 319 'D' : 'A'; 320 } 321 buffer[i] = '\0'; 322 DMEMIT("1 %s", buffer); 323 break; 324 325 case STATUSTYPE_TABLE: 326 DMEMIT("%d %llu", sc->stripes, 327 (unsigned long long)sc->chunk_mask + 1); 328 for (i = 0; i < sc->stripes; i++) 329 DMEMIT(" %s %llu", sc->stripe[i].dev->name, 330 (unsigned long long)sc->stripe[i].physical_start); 331 break; 332 } 333 return 0; 334 } 335 336 static int stripe_end_io(struct dm_target *ti, struct bio *bio, 337 int error, union map_info *map_context) 338 { 339 unsigned i; 340 char major_minor[16]; 341 struct stripe_c *sc = ti->private; 342 343 if (!error) 344 return 0; /* I/O complete */ 345 346 if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD)) 347 return error; 348 349 if (error == -EOPNOTSUPP) 350 return error; 351 352 memset(major_minor, 0, sizeof(major_minor)); 353 sprintf(major_minor, "%d:%d", 354 MAJOR(disk_devt(bio->bi_bdev->bd_disk)), 355 MINOR(disk_devt(bio->bi_bdev->bd_disk))); 356 357 /* 358 * Test to see which stripe drive triggered the event 359 * and increment error count for all stripes on that device. 360 * If the error count for a given device exceeds the threshold 361 * value we will no longer trigger any further events. 362 */ 363 for (i = 0; i < sc->stripes; i++) 364 if (!strcmp(sc->stripe[i].dev->name, major_minor)) { 365 atomic_inc(&(sc->stripe[i].error_count)); 366 if (atomic_read(&(sc->stripe[i].error_count)) < 367 DM_IO_ERROR_THRESHOLD) 368 schedule_work(&sc->trigger_event); 369 } 370 371 return error; 372 } 373 374 static int stripe_iterate_devices(struct dm_target *ti, 375 iterate_devices_callout_fn fn, void *data) 376 { 377 struct stripe_c *sc = ti->private; 378 int ret = 0; 379 unsigned i = 0; 380 381 do { 382 ret = fn(ti, sc->stripe[i].dev, 383 sc->stripe[i].physical_start, 384 sc->stripe_width, data); 385 } while (!ret && ++i < sc->stripes); 386 387 return ret; 388 } 389 390 static void stripe_io_hints(struct dm_target *ti, 391 struct queue_limits *limits) 392 { 393 struct stripe_c *sc = ti->private; 394 unsigned chunk_size = (sc->chunk_mask + 1) << 9; 395 396 blk_limits_io_min(limits, chunk_size); 397 blk_limits_io_opt(limits, chunk_size * sc->stripes); 398 } 399 400 static int stripe_merge(struct dm_target *ti, struct bvec_merge_data *bvm, 401 struct bio_vec *biovec, int max_size) 402 { 403 struct stripe_c *sc = ti->private; 404 sector_t bvm_sector = bvm->bi_sector; 405 uint32_t stripe; 406 struct request_queue *q; 407 408 stripe_map_sector(sc, bvm_sector, &stripe, &bvm_sector); 409 410 q = bdev_get_queue(sc->stripe[stripe].dev->bdev); 411 if (!q->merge_bvec_fn) 412 return max_size; 413 414 bvm->bi_bdev = sc->stripe[stripe].dev->bdev; 415 bvm->bi_sector = sc->stripe[stripe].physical_start + bvm_sector; 416 417 return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); 418 } 419 420 static struct target_type stripe_target = { 421 .name = "striped", 422 .version = {1, 4, 0}, 423 .module = THIS_MODULE, 424 .ctr = stripe_ctr, 425 .dtr = stripe_dtr, 426 .map = stripe_map, 427 .end_io = stripe_end_io, 428 .status = stripe_status, 429 .iterate_devices = stripe_iterate_devices, 430 .io_hints = stripe_io_hints, 431 .merge = stripe_merge, 432 }; 433 434 int __init dm_stripe_init(void) 435 { 436 int r; 437 438 r = dm_register_target(&stripe_target); 439 if (r < 0) { 440 DMWARN("target registration failed"); 441 return r; 442 } 443 444 return r; 445 } 446 447 void dm_stripe_exit(void) 448 { 449 dm_unregister_target(&stripe_target); 450 } 451