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 int ret; 79 80 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1) 81 return -EINVAL; 82 83 ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), 84 &sc->stripe[stripe].dev); 85 if (ret) 86 return ret; 87 88 sc->stripe[stripe].physical_start = start; 89 90 return 0; 91 } 92 93 /* 94 * Construct a striped mapping. 95 * <number of stripes> <chunk size> [<dev_path> <offset>]+ 96 */ 97 static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv) 98 { 99 struct stripe_c *sc; 100 sector_t width, tmp_len; 101 uint32_t stripes; 102 uint32_t chunk_size; 103 int r; 104 unsigned int i; 105 106 if (argc < 2) { 107 ti->error = "Not enough arguments"; 108 return -EINVAL; 109 } 110 111 if (kstrtouint(argv[0], 10, &stripes) || !stripes) { 112 ti->error = "Invalid stripe count"; 113 return -EINVAL; 114 } 115 116 if (kstrtouint(argv[1], 10, &chunk_size) || !chunk_size) { 117 ti->error = "Invalid chunk_size"; 118 return -EINVAL; 119 } 120 121 width = ti->len; 122 if (sector_div(width, stripes)) { 123 ti->error = "Target length not divisible by " 124 "number of stripes"; 125 return -EINVAL; 126 } 127 128 tmp_len = width; 129 if (sector_div(tmp_len, chunk_size)) { 130 ti->error = "Target length not divisible by " 131 "chunk size"; 132 return -EINVAL; 133 } 134 135 /* 136 * Do we have enough arguments for that many stripes ? 137 */ 138 if (argc != (2 + 2 * stripes)) { 139 ti->error = "Not enough destinations " 140 "specified"; 141 return -EINVAL; 142 } 143 144 sc = alloc_context(stripes); 145 if (!sc) { 146 ti->error = "Memory allocation for striped context " 147 "failed"; 148 return -ENOMEM; 149 } 150 151 INIT_WORK(&sc->trigger_event, trigger_event); 152 153 /* Set pointer to dm target; used in trigger_event */ 154 sc->ti = ti; 155 sc->stripes = stripes; 156 sc->stripe_width = width; 157 158 if (stripes & (stripes - 1)) 159 sc->stripes_shift = -1; 160 else 161 sc->stripes_shift = __ffs(stripes); 162 163 r = dm_set_target_max_io_len(ti, chunk_size); 164 if (r) { 165 kfree(sc); 166 return r; 167 } 168 169 ti->num_flush_bios = stripes; 170 ti->num_discard_bios = stripes; 171 ti->num_write_same_bios = stripes; 172 173 sc->chunk_size = chunk_size; 174 if (chunk_size & (chunk_size - 1)) 175 sc->chunk_size_shift = -1; 176 else 177 sc->chunk_size_shift = __ffs(chunk_size); 178 179 /* 180 * Get the stripe destinations. 181 */ 182 for (i = 0; i < stripes; i++) { 183 argv += 2; 184 185 r = get_stripe(ti, sc, i, argv); 186 if (r < 0) { 187 ti->error = "Couldn't parse stripe destination"; 188 while (i--) 189 dm_put_device(ti, sc->stripe[i].dev); 190 kfree(sc); 191 return r; 192 } 193 atomic_set(&(sc->stripe[i].error_count), 0); 194 } 195 196 ti->private = sc; 197 198 return 0; 199 } 200 201 static void stripe_dtr(struct dm_target *ti) 202 { 203 unsigned int i; 204 struct stripe_c *sc = (struct stripe_c *) ti->private; 205 206 for (i = 0; i < sc->stripes; i++) 207 dm_put_device(ti, sc->stripe[i].dev); 208 209 flush_work(&sc->trigger_event); 210 kfree(sc); 211 } 212 213 static void stripe_map_sector(struct stripe_c *sc, sector_t sector, 214 uint32_t *stripe, sector_t *result) 215 { 216 sector_t chunk = dm_target_offset(sc->ti, sector); 217 sector_t chunk_offset; 218 219 if (sc->chunk_size_shift < 0) 220 chunk_offset = sector_div(chunk, sc->chunk_size); 221 else { 222 chunk_offset = chunk & (sc->chunk_size - 1); 223 chunk >>= sc->chunk_size_shift; 224 } 225 226 if (sc->stripes_shift < 0) 227 *stripe = sector_div(chunk, sc->stripes); 228 else { 229 *stripe = chunk & (sc->stripes - 1); 230 chunk >>= sc->stripes_shift; 231 } 232 233 if (sc->chunk_size_shift < 0) 234 chunk *= sc->chunk_size; 235 else 236 chunk <<= sc->chunk_size_shift; 237 238 *result = chunk + chunk_offset; 239 } 240 241 static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector, 242 uint32_t target_stripe, sector_t *result) 243 { 244 uint32_t stripe; 245 246 stripe_map_sector(sc, sector, &stripe, result); 247 if (stripe == target_stripe) 248 return; 249 250 /* round down */ 251 sector = *result; 252 if (sc->chunk_size_shift < 0) 253 *result -= sector_div(sector, sc->chunk_size); 254 else 255 *result = sector & ~(sector_t)(sc->chunk_size - 1); 256 257 if (target_stripe < stripe) 258 *result += sc->chunk_size; /* next chunk */ 259 } 260 261 static int stripe_map_range(struct stripe_c *sc, struct bio *bio, 262 uint32_t target_stripe) 263 { 264 sector_t begin, end; 265 266 stripe_map_range_sector(sc, bio->bi_iter.bi_sector, 267 target_stripe, &begin); 268 stripe_map_range_sector(sc, bio_end_sector(bio), 269 target_stripe, &end); 270 if (begin < end) { 271 bio->bi_bdev = sc->stripe[target_stripe].dev->bdev; 272 bio->bi_iter.bi_sector = begin + 273 sc->stripe[target_stripe].physical_start; 274 bio->bi_iter.bi_size = to_bytes(end - begin); 275 return DM_MAPIO_REMAPPED; 276 } else { 277 /* The range doesn't map to the target stripe */ 278 bio_endio(bio); 279 return DM_MAPIO_SUBMITTED; 280 } 281 } 282 283 static int stripe_map(struct dm_target *ti, struct bio *bio) 284 { 285 struct stripe_c *sc = ti->private; 286 uint32_t stripe; 287 unsigned target_bio_nr; 288 289 if (bio->bi_opf & REQ_PREFLUSH) { 290 target_bio_nr = dm_bio_get_target_bio_nr(bio); 291 BUG_ON(target_bio_nr >= sc->stripes); 292 bio->bi_bdev = sc->stripe[target_bio_nr].dev->bdev; 293 return DM_MAPIO_REMAPPED; 294 } 295 if (unlikely(bio_op(bio) == REQ_OP_DISCARD) || 296 unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) { 297 target_bio_nr = dm_bio_get_target_bio_nr(bio); 298 BUG_ON(target_bio_nr >= sc->stripes); 299 return stripe_map_range(sc, bio, target_bio_nr); 300 } 301 302 stripe_map_sector(sc, bio->bi_iter.bi_sector, 303 &stripe, &bio->bi_iter.bi_sector); 304 305 bio->bi_iter.bi_sector += sc->stripe[stripe].physical_start; 306 bio->bi_bdev = sc->stripe[stripe].dev->bdev; 307 308 return DM_MAPIO_REMAPPED; 309 } 310 311 static long stripe_direct_access(struct dm_target *ti, sector_t sector, 312 void **kaddr, pfn_t *pfn, long size) 313 { 314 struct stripe_c *sc = ti->private; 315 uint32_t stripe; 316 struct block_device *bdev; 317 struct blk_dax_ctl dax = { 318 .size = size, 319 }; 320 long ret; 321 322 stripe_map_sector(sc, sector, &stripe, &dax.sector); 323 324 dax.sector += sc->stripe[stripe].physical_start; 325 bdev = sc->stripe[stripe].dev->bdev; 326 327 ret = bdev_direct_access(bdev, &dax); 328 *kaddr = dax.addr; 329 *pfn = dax.pfn; 330 331 return ret; 332 } 333 334 /* 335 * Stripe status: 336 * 337 * INFO 338 * #stripes [stripe_name <stripe_name>] [group word count] 339 * [error count 'A|D' <error count 'A|D'>] 340 * 341 * TABLE 342 * #stripes [stripe chunk size] 343 * [stripe_name physical_start <stripe_name physical_start>] 344 * 345 */ 346 347 static void stripe_status(struct dm_target *ti, status_type_t type, 348 unsigned status_flags, char *result, unsigned maxlen) 349 { 350 struct stripe_c *sc = (struct stripe_c *) ti->private; 351 char buffer[sc->stripes + 1]; 352 unsigned int sz = 0; 353 unsigned int i; 354 355 switch (type) { 356 case STATUSTYPE_INFO: 357 DMEMIT("%d ", sc->stripes); 358 for (i = 0; i < sc->stripes; i++) { 359 DMEMIT("%s ", sc->stripe[i].dev->name); 360 buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ? 361 'D' : 'A'; 362 } 363 buffer[i] = '\0'; 364 DMEMIT("1 %s", buffer); 365 break; 366 367 case STATUSTYPE_TABLE: 368 DMEMIT("%d %llu", sc->stripes, 369 (unsigned long long)sc->chunk_size); 370 for (i = 0; i < sc->stripes; i++) 371 DMEMIT(" %s %llu", sc->stripe[i].dev->name, 372 (unsigned long long)sc->stripe[i].physical_start); 373 break; 374 } 375 } 376 377 static int stripe_end_io(struct dm_target *ti, struct bio *bio, int error) 378 { 379 unsigned i; 380 char major_minor[16]; 381 struct stripe_c *sc = ti->private; 382 383 if (!error) 384 return 0; /* I/O complete */ 385 386 if ((error == -EWOULDBLOCK) && (bio->bi_opf & REQ_RAHEAD)) 387 return error; 388 389 if (error == -EOPNOTSUPP) 390 return error; 391 392 memset(major_minor, 0, sizeof(major_minor)); 393 sprintf(major_minor, "%d:%d", 394 MAJOR(disk_devt(bio->bi_bdev->bd_disk)), 395 MINOR(disk_devt(bio->bi_bdev->bd_disk))); 396 397 /* 398 * Test to see which stripe drive triggered the event 399 * and increment error count for all stripes on that device. 400 * If the error count for a given device exceeds the threshold 401 * value we will no longer trigger any further events. 402 */ 403 for (i = 0; i < sc->stripes; i++) 404 if (!strcmp(sc->stripe[i].dev->name, major_minor)) { 405 atomic_inc(&(sc->stripe[i].error_count)); 406 if (atomic_read(&(sc->stripe[i].error_count)) < 407 DM_IO_ERROR_THRESHOLD) 408 schedule_work(&sc->trigger_event); 409 } 410 411 return error; 412 } 413 414 static int stripe_iterate_devices(struct dm_target *ti, 415 iterate_devices_callout_fn fn, void *data) 416 { 417 struct stripe_c *sc = ti->private; 418 int ret = 0; 419 unsigned i = 0; 420 421 do { 422 ret = fn(ti, sc->stripe[i].dev, 423 sc->stripe[i].physical_start, 424 sc->stripe_width, data); 425 } while (!ret && ++i < sc->stripes); 426 427 return ret; 428 } 429 430 static void stripe_io_hints(struct dm_target *ti, 431 struct queue_limits *limits) 432 { 433 struct stripe_c *sc = ti->private; 434 unsigned chunk_size = sc->chunk_size << SECTOR_SHIFT; 435 436 blk_limits_io_min(limits, chunk_size); 437 blk_limits_io_opt(limits, chunk_size * sc->stripes); 438 } 439 440 static struct target_type stripe_target = { 441 .name = "striped", 442 .version = {1, 6, 0}, 443 .module = THIS_MODULE, 444 .ctr = stripe_ctr, 445 .dtr = stripe_dtr, 446 .map = stripe_map, 447 .end_io = stripe_end_io, 448 .status = stripe_status, 449 .iterate_devices = stripe_iterate_devices, 450 .io_hints = stripe_io_hints, 451 .direct_access = stripe_direct_access, 452 }; 453 454 int __init dm_stripe_init(void) 455 { 456 int r; 457 458 r = dm_register_target(&stripe_target); 459 if (r < 0) 460 DMWARN("target registration failed"); 461 462 return r; 463 } 464 465 void dm_stripe_exit(void) 466 { 467 dm_unregister_target(&stripe_target); 468 } 469