| 1 | /* |
| 2 | * multipath.c : Multiple Devices driver for Linux |
| 3 | * |
| 4 | * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat |
| 5 | * |
| 6 | * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman |
| 7 | * |
| 8 | * MULTIPATH management functions. |
| 9 | * |
| 10 | * derived from raid1.c. |
| 11 | * |
| 12 | * This program is free software; you can redistribute it and/or modify |
| 13 | * it under the terms of the GNU General Public License as published by |
| 14 | * the Free Software Foundation; either version 2, or (at your option) |
| 15 | * any later version. |
| 16 | * |
| 17 | * You should have received a copy of the GNU General Public License |
| 18 | * (for example /usr/src/linux/COPYING); if not, write to the Free |
| 19 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
| 20 | */ |
| 21 | |
| 22 | #include <linux/blkdev.h> |
| 23 | #include <linux/module.h> |
| 24 | #include <linux/raid/md_u.h> |
| 25 | #include <linux/seq_file.h> |
| 26 | #include <linux/slab.h> |
| 27 | #include "md.h" |
| 28 | #include "multipath.h" |
| 29 | |
| 30 | #define MAX_WORK_PER_DISK 128 |
| 31 | |
| 32 | #define NR_RESERVED_BUFS 32 |
| 33 | |
| 34 | static int multipath_map (struct mpconf *conf) |
| 35 | { |
| 36 | int i, disks = conf->raid_disks; |
| 37 | |
| 38 | /* |
| 39 | * Later we do read balancing on the read side |
| 40 | * now we use the first available disk. |
| 41 | */ |
| 42 | |
| 43 | rcu_read_lock(); |
| 44 | for (i = 0; i < disks; i++) { |
| 45 | struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev); |
| 46 | if (rdev && test_bit(In_sync, &rdev->flags) && |
| 47 | !test_bit(Faulty, &rdev->flags)) { |
| 48 | atomic_inc(&rdev->nr_pending); |
| 49 | rcu_read_unlock(); |
| 50 | return i; |
| 51 | } |
| 52 | } |
| 53 | rcu_read_unlock(); |
| 54 | |
| 55 | pr_crit_ratelimited("multipath_map(): no more operational IO paths?\n"); |
| 56 | return (-1); |
| 57 | } |
| 58 | |
| 59 | static void multipath_reschedule_retry (struct multipath_bh *mp_bh) |
| 60 | { |
| 61 | unsigned long flags; |
| 62 | struct mddev *mddev = mp_bh->mddev; |
| 63 | struct mpconf *conf = mddev->private; |
| 64 | |
| 65 | spin_lock_irqsave(&conf->device_lock, flags); |
| 66 | list_add(&mp_bh->retry_list, &conf->retry_list); |
| 67 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 68 | md_wakeup_thread(mddev->thread); |
| 69 | } |
| 70 | |
| 71 | /* |
| 72 | * multipath_end_bh_io() is called when we have finished servicing a multipathed |
| 73 | * operation and are ready to return a success/failure code to the buffer |
| 74 | * cache layer. |
| 75 | */ |
| 76 | static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err) |
| 77 | { |
| 78 | struct bio *bio = mp_bh->master_bio; |
| 79 | struct mpconf *conf = mp_bh->mddev->private; |
| 80 | |
| 81 | bio->bi_error = err; |
| 82 | bio_endio(bio); |
| 83 | mempool_free(mp_bh, conf->pool); |
| 84 | } |
| 85 | |
| 86 | static void multipath_end_request(struct bio *bio) |
| 87 | { |
| 88 | struct multipath_bh *mp_bh = bio->bi_private; |
| 89 | struct mpconf *conf = mp_bh->mddev->private; |
| 90 | struct md_rdev *rdev = conf->multipaths[mp_bh->path].rdev; |
| 91 | |
| 92 | if (!bio->bi_error) |
| 93 | multipath_end_bh_io(mp_bh, 0); |
| 94 | else if (!(bio->bi_opf & REQ_RAHEAD)) { |
| 95 | /* |
| 96 | * oops, IO error: |
| 97 | */ |
| 98 | char b[BDEVNAME_SIZE]; |
| 99 | md_error (mp_bh->mddev, rdev); |
| 100 | pr_info("multipath: %s: rescheduling sector %llu\n", |
| 101 | bdevname(rdev->bdev,b), |
| 102 | (unsigned long long)bio->bi_iter.bi_sector); |
| 103 | multipath_reschedule_retry(mp_bh); |
| 104 | } else |
| 105 | multipath_end_bh_io(mp_bh, bio->bi_error); |
| 106 | rdev_dec_pending(rdev, conf->mddev); |
| 107 | } |
| 108 | |
| 109 | static void multipath_make_request(struct mddev *mddev, struct bio * bio) |
| 110 | { |
| 111 | struct mpconf *conf = mddev->private; |
| 112 | struct multipath_bh * mp_bh; |
| 113 | struct multipath_info *multipath; |
| 114 | |
| 115 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
| 116 | md_flush_request(mddev, bio); |
| 117 | return; |
| 118 | } |
| 119 | |
| 120 | mp_bh = mempool_alloc(conf->pool, GFP_NOIO); |
| 121 | |
| 122 | mp_bh->master_bio = bio; |
| 123 | mp_bh->mddev = mddev; |
| 124 | |
| 125 | mp_bh->path = multipath_map(conf); |
| 126 | if (mp_bh->path < 0) { |
| 127 | bio_io_error(bio); |
| 128 | mempool_free(mp_bh, conf->pool); |
| 129 | return; |
| 130 | } |
| 131 | multipath = conf->multipaths + mp_bh->path; |
| 132 | |
| 133 | bio_init(&mp_bh->bio, NULL, 0); |
| 134 | __bio_clone_fast(&mp_bh->bio, bio); |
| 135 | |
| 136 | mp_bh->bio.bi_iter.bi_sector += multipath->rdev->data_offset; |
| 137 | mp_bh->bio.bi_bdev = multipath->rdev->bdev; |
| 138 | mp_bh->bio.bi_opf |= REQ_FAILFAST_TRANSPORT; |
| 139 | mp_bh->bio.bi_end_io = multipath_end_request; |
| 140 | mp_bh->bio.bi_private = mp_bh; |
| 141 | mddev_check_writesame(mddev, &mp_bh->bio); |
| 142 | generic_make_request(&mp_bh->bio); |
| 143 | return; |
| 144 | } |
| 145 | |
| 146 | static void multipath_status(struct seq_file *seq, struct mddev *mddev) |
| 147 | { |
| 148 | struct mpconf *conf = mddev->private; |
| 149 | int i; |
| 150 | |
| 151 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, |
| 152 | conf->raid_disks - mddev->degraded); |
| 153 | rcu_read_lock(); |
| 154 | for (i = 0; i < conf->raid_disks; i++) { |
| 155 | struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev); |
| 156 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
| 157 | } |
| 158 | rcu_read_unlock(); |
| 159 | seq_printf (seq, "]"); |
| 160 | } |
| 161 | |
| 162 | static int multipath_congested(struct mddev *mddev, int bits) |
| 163 | { |
| 164 | struct mpconf *conf = mddev->private; |
| 165 | int i, ret = 0; |
| 166 | |
| 167 | rcu_read_lock(); |
| 168 | for (i = 0; i < mddev->raid_disks ; i++) { |
| 169 | struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev); |
| 170 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
| 171 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
| 172 | |
| 173 | ret |= bdi_congested(q->backing_dev_info, bits); |
| 174 | /* Just like multipath_map, we just check the |
| 175 | * first available device |
| 176 | */ |
| 177 | break; |
| 178 | } |
| 179 | } |
| 180 | rcu_read_unlock(); |
| 181 | return ret; |
| 182 | } |
| 183 | |
| 184 | /* |
| 185 | * Careful, this can execute in IRQ contexts as well! |
| 186 | */ |
| 187 | static void multipath_error (struct mddev *mddev, struct md_rdev *rdev) |
| 188 | { |
| 189 | struct mpconf *conf = mddev->private; |
| 190 | char b[BDEVNAME_SIZE]; |
| 191 | |
| 192 | if (conf->raid_disks - mddev->degraded <= 1) { |
| 193 | /* |
| 194 | * Uh oh, we can do nothing if this is our last path, but |
| 195 | * first check if this is a queued request for a device |
| 196 | * which has just failed. |
| 197 | */ |
| 198 | pr_warn("multipath: only one IO path left and IO error.\n"); |
| 199 | /* leave it active... it's all we have */ |
| 200 | return; |
| 201 | } |
| 202 | /* |
| 203 | * Mark disk as unusable |
| 204 | */ |
| 205 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
| 206 | unsigned long flags; |
| 207 | spin_lock_irqsave(&conf->device_lock, flags); |
| 208 | mddev->degraded++; |
| 209 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 210 | } |
| 211 | set_bit(Faulty, &rdev->flags); |
| 212 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
| 213 | pr_err("multipath: IO failure on %s, disabling IO path.\n" |
| 214 | "multipath: Operation continuing on %d IO paths.\n", |
| 215 | bdevname(rdev->bdev, b), |
| 216 | conf->raid_disks - mddev->degraded); |
| 217 | } |
| 218 | |
| 219 | static void print_multipath_conf (struct mpconf *conf) |
| 220 | { |
| 221 | int i; |
| 222 | struct multipath_info *tmp; |
| 223 | |
| 224 | pr_debug("MULTIPATH conf printout:\n"); |
| 225 | if (!conf) { |
| 226 | pr_debug("(conf==NULL)\n"); |
| 227 | return; |
| 228 | } |
| 229 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
| 230 | conf->raid_disks); |
| 231 | |
| 232 | for (i = 0; i < conf->raid_disks; i++) { |
| 233 | char b[BDEVNAME_SIZE]; |
| 234 | tmp = conf->multipaths + i; |
| 235 | if (tmp->rdev) |
| 236 | pr_debug(" disk%d, o:%d, dev:%s\n", |
| 237 | i,!test_bit(Faulty, &tmp->rdev->flags), |
| 238 | bdevname(tmp->rdev->bdev,b)); |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
| 243 | { |
| 244 | struct mpconf *conf = mddev->private; |
| 245 | struct request_queue *q; |
| 246 | int err = -EEXIST; |
| 247 | int path; |
| 248 | struct multipath_info *p; |
| 249 | int first = 0; |
| 250 | int last = mddev->raid_disks - 1; |
| 251 | |
| 252 | if (rdev->raid_disk >= 0) |
| 253 | first = last = rdev->raid_disk; |
| 254 | |
| 255 | print_multipath_conf(conf); |
| 256 | |
| 257 | for (path = first; path <= last; path++) |
| 258 | if ((p=conf->multipaths+path)->rdev == NULL) { |
| 259 | q = rdev->bdev->bd_disk->queue; |
| 260 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 261 | rdev->data_offset << 9); |
| 262 | |
| 263 | err = md_integrity_add_rdev(rdev, mddev); |
| 264 | if (err) |
| 265 | break; |
| 266 | spin_lock_irq(&conf->device_lock); |
| 267 | mddev->degraded--; |
| 268 | rdev->raid_disk = path; |
| 269 | set_bit(In_sync, &rdev->flags); |
| 270 | spin_unlock_irq(&conf->device_lock); |
| 271 | rcu_assign_pointer(p->rdev, rdev); |
| 272 | err = 0; |
| 273 | break; |
| 274 | } |
| 275 | |
| 276 | print_multipath_conf(conf); |
| 277 | |
| 278 | return err; |
| 279 | } |
| 280 | |
| 281 | static int multipath_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
| 282 | { |
| 283 | struct mpconf *conf = mddev->private; |
| 284 | int err = 0; |
| 285 | int number = rdev->raid_disk; |
| 286 | struct multipath_info *p = conf->multipaths + number; |
| 287 | |
| 288 | print_multipath_conf(conf); |
| 289 | |
| 290 | if (rdev == p->rdev) { |
| 291 | if (test_bit(In_sync, &rdev->flags) || |
| 292 | atomic_read(&rdev->nr_pending)) { |
| 293 | pr_warn("hot-remove-disk, slot %d is identified but is still operational!\n", number); |
| 294 | err = -EBUSY; |
| 295 | goto abort; |
| 296 | } |
| 297 | p->rdev = NULL; |
| 298 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
| 299 | synchronize_rcu(); |
| 300 | if (atomic_read(&rdev->nr_pending)) { |
| 301 | /* lost the race, try later */ |
| 302 | err = -EBUSY; |
| 303 | p->rdev = rdev; |
| 304 | goto abort; |
| 305 | } |
| 306 | } |
| 307 | err = md_integrity_register(mddev); |
| 308 | } |
| 309 | abort: |
| 310 | |
| 311 | print_multipath_conf(conf); |
| 312 | return err; |
| 313 | } |
| 314 | |
| 315 | /* |
| 316 | * This is a kernel thread which: |
| 317 | * |
| 318 | * 1. Retries failed read operations on working multipaths. |
| 319 | * 2. Updates the raid superblock when problems encounter. |
| 320 | * 3. Performs writes following reads for array syncronising. |
| 321 | */ |
| 322 | |
| 323 | static void multipathd(struct md_thread *thread) |
| 324 | { |
| 325 | struct mddev *mddev = thread->mddev; |
| 326 | struct multipath_bh *mp_bh; |
| 327 | struct bio *bio; |
| 328 | unsigned long flags; |
| 329 | struct mpconf *conf = mddev->private; |
| 330 | struct list_head *head = &conf->retry_list; |
| 331 | |
| 332 | md_check_recovery(mddev); |
| 333 | for (;;) { |
| 334 | char b[BDEVNAME_SIZE]; |
| 335 | spin_lock_irqsave(&conf->device_lock, flags); |
| 336 | if (list_empty(head)) |
| 337 | break; |
| 338 | mp_bh = list_entry(head->prev, struct multipath_bh, retry_list); |
| 339 | list_del(head->prev); |
| 340 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 341 | |
| 342 | bio = &mp_bh->bio; |
| 343 | bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector; |
| 344 | |
| 345 | if ((mp_bh->path = multipath_map (conf))<0) { |
| 346 | pr_err("multipath: %s: unrecoverable IO read error for block %llu\n", |
| 347 | bdevname(bio->bi_bdev,b), |
| 348 | (unsigned long long)bio->bi_iter.bi_sector); |
| 349 | multipath_end_bh_io(mp_bh, -EIO); |
| 350 | } else { |
| 351 | pr_err("multipath: %s: redirecting sector %llu to another IO path\n", |
| 352 | bdevname(bio->bi_bdev,b), |
| 353 | (unsigned long long)bio->bi_iter.bi_sector); |
| 354 | *bio = *(mp_bh->master_bio); |
| 355 | bio->bi_iter.bi_sector += |
| 356 | conf->multipaths[mp_bh->path].rdev->data_offset; |
| 357 | bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev; |
| 358 | bio->bi_opf |= REQ_FAILFAST_TRANSPORT; |
| 359 | bio->bi_end_io = multipath_end_request; |
| 360 | bio->bi_private = mp_bh; |
| 361 | generic_make_request(bio); |
| 362 | } |
| 363 | } |
| 364 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 365 | } |
| 366 | |
| 367 | static sector_t multipath_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
| 368 | { |
| 369 | WARN_ONCE(sectors || raid_disks, |
| 370 | "%s does not support generic reshape\n", __func__); |
| 371 | |
| 372 | return mddev->dev_sectors; |
| 373 | } |
| 374 | |
| 375 | static int multipath_run (struct mddev *mddev) |
| 376 | { |
| 377 | struct mpconf *conf; |
| 378 | int disk_idx; |
| 379 | struct multipath_info *disk; |
| 380 | struct md_rdev *rdev; |
| 381 | int working_disks; |
| 382 | |
| 383 | if (md_check_no_bitmap(mddev)) |
| 384 | return -EINVAL; |
| 385 | |
| 386 | if (mddev->level != LEVEL_MULTIPATH) { |
| 387 | pr_warn("multipath: %s: raid level not set to multipath IO (%d)\n", |
| 388 | mdname(mddev), mddev->level); |
| 389 | goto out; |
| 390 | } |
| 391 | /* |
| 392 | * copy the already verified devices into our private MULTIPATH |
| 393 | * bookkeeping area. [whatever we allocate in multipath_run(), |
| 394 | * should be freed in multipath_free()] |
| 395 | */ |
| 396 | |
| 397 | conf = kzalloc(sizeof(struct mpconf), GFP_KERNEL); |
| 398 | mddev->private = conf; |
| 399 | if (!conf) |
| 400 | goto out; |
| 401 | |
| 402 | conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks, |
| 403 | GFP_KERNEL); |
| 404 | if (!conf->multipaths) |
| 405 | goto out_free_conf; |
| 406 | |
| 407 | working_disks = 0; |
| 408 | rdev_for_each(rdev, mddev) { |
| 409 | disk_idx = rdev->raid_disk; |
| 410 | if (disk_idx < 0 || |
| 411 | disk_idx >= mddev->raid_disks) |
| 412 | continue; |
| 413 | |
| 414 | disk = conf->multipaths + disk_idx; |
| 415 | disk->rdev = rdev; |
| 416 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 417 | rdev->data_offset << 9); |
| 418 | |
| 419 | if (!test_bit(Faulty, &rdev->flags)) |
| 420 | working_disks++; |
| 421 | } |
| 422 | |
| 423 | conf->raid_disks = mddev->raid_disks; |
| 424 | conf->mddev = mddev; |
| 425 | spin_lock_init(&conf->device_lock); |
| 426 | INIT_LIST_HEAD(&conf->retry_list); |
| 427 | |
| 428 | if (!working_disks) { |
| 429 | pr_warn("multipath: no operational IO paths for %s\n", |
| 430 | mdname(mddev)); |
| 431 | goto out_free_conf; |
| 432 | } |
| 433 | mddev->degraded = conf->raid_disks - working_disks; |
| 434 | |
| 435 | conf->pool = mempool_create_kmalloc_pool(NR_RESERVED_BUFS, |
| 436 | sizeof(struct multipath_bh)); |
| 437 | if (conf->pool == NULL) |
| 438 | goto out_free_conf; |
| 439 | |
| 440 | mddev->thread = md_register_thread(multipathd, mddev, |
| 441 | "multipath"); |
| 442 | if (!mddev->thread) |
| 443 | goto out_free_conf; |
| 444 | |
| 445 | pr_info("multipath: array %s active with %d out of %d IO paths\n", |
| 446 | mdname(mddev), conf->raid_disks - mddev->degraded, |
| 447 | mddev->raid_disks); |
| 448 | /* |
| 449 | * Ok, everything is just fine now |
| 450 | */ |
| 451 | md_set_array_sectors(mddev, multipath_size(mddev, 0, 0)); |
| 452 | |
| 453 | if (md_integrity_register(mddev)) |
| 454 | goto out_free_conf; |
| 455 | |
| 456 | return 0; |
| 457 | |
| 458 | out_free_conf: |
| 459 | mempool_destroy(conf->pool); |
| 460 | kfree(conf->multipaths); |
| 461 | kfree(conf); |
| 462 | mddev->private = NULL; |
| 463 | out: |
| 464 | return -EIO; |
| 465 | } |
| 466 | |
| 467 | static void multipath_free(struct mddev *mddev, void *priv) |
| 468 | { |
| 469 | struct mpconf *conf = priv; |
| 470 | |
| 471 | mempool_destroy(conf->pool); |
| 472 | kfree(conf->multipaths); |
| 473 | kfree(conf); |
| 474 | } |
| 475 | |
| 476 | static struct md_personality multipath_personality = |
| 477 | { |
| 478 | .name = "multipath", |
| 479 | .level = LEVEL_MULTIPATH, |
| 480 | .owner = THIS_MODULE, |
| 481 | .make_request = multipath_make_request, |
| 482 | .run = multipath_run, |
| 483 | .free = multipath_free, |
| 484 | .status = multipath_status, |
| 485 | .error_handler = multipath_error, |
| 486 | .hot_add_disk = multipath_add_disk, |
| 487 | .hot_remove_disk= multipath_remove_disk, |
| 488 | .size = multipath_size, |
| 489 | .congested = multipath_congested, |
| 490 | }; |
| 491 | |
| 492 | static int __init multipath_init (void) |
| 493 | { |
| 494 | return register_md_personality (&multipath_personality); |
| 495 | } |
| 496 | |
| 497 | static void __exit multipath_exit (void) |
| 498 | { |
| 499 | unregister_md_personality (&multipath_personality); |
| 500 | } |
| 501 | |
| 502 | module_init(multipath_init); |
| 503 | module_exit(multipath_exit); |
| 504 | MODULE_LICENSE("GPL"); |
| 505 | MODULE_DESCRIPTION("simple multi-path personality for MD"); |
| 506 | MODULE_ALIAS("md-personality-7"); /* MULTIPATH */ |
| 507 | MODULE_ALIAS("md-multipath"); |
| 508 | MODULE_ALIAS("md-level--4"); |