drivers/md/multipath.c

   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                         atomic_inc(&rdev->nr_pending);
  48                         rcu_read_unlock();
  49                         return i;
  50                 }
  51         }
  52         rcu_read_unlock();
  53
  54         printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
  55         return (-1);
  56 }
  57
  58 static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
  59 {
  60         unsigned long flags;
  61         struct mddev *mddev = mp_bh->mddev;
  62         struct mpconf *conf = mddev->private;
  63
  64         spin_lock_irqsave(&conf->device_lock, flags);
  65         list_add(&mp_bh->retry_list, &conf->retry_list);
  66         spin_unlock_irqrestore(&conf->device_lock, flags);
  67         md_wakeup_thread(mddev->thread);
  68 }
  69
  70 /*
  71  * multipath_end_bh_io() is called when we have finished servicing a multipathed
  72  * operation and are ready to return a success/failure code to the buffer
  73  * cache layer.
  74  */
  75 static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
  76 {
  77         struct bio *bio = mp_bh->master_bio;
  78         struct mpconf *conf = mp_bh->mddev->private;
  79
  80         bio->bi_error = err;
  81         bio_endio(bio);
  82         mempool_free(mp_bh, conf->pool);
  83 }
  84
  85 static void multipath_end_request(struct bio *bio)
  86 {
  87         struct multipath_bh *mp_bh = bio->bi_private;
  88         struct mpconf *conf = mp_bh->mddev->private;
  89         struct md_rdev *rdev = conf->multipaths[mp_bh->path].rdev;
  90
  91         if (!bio->bi_error)
  92                 multipath_end_bh_io(mp_bh, 0);
  93         else if (!(bio->bi_rw & REQ_RAHEAD)) {
  94                 /*
  95                  * oops, IO error:
  96                  */
  97                 char b[BDEVNAME_SIZE];
  98                 md_error (mp_bh->mddev, rdev);
  99                 printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
 100                        bdevname(rdev->bdev,b),
 101                        (unsigned long long)bio->bi_iter.bi_sector);
 102                 multipath_reschedule_retry(mp_bh);
 103         } else
 104                 multipath_end_bh_io(mp_bh, bio->bi_error);
 105         rdev_dec_pending(rdev, conf->mddev);
 106 }
 107
 108 static void multipath_make_request(struct mddev *mddev, struct bio * bio)
 109 {
 110         struct mpconf *conf = mddev->private;
 111         struct multipath_bh * mp_bh;
 112         struct multipath_info *multipath;
 113
 114         if (unlikely(bio->bi_rw & REQ_FLUSH)) {
 115                 md_flush_request(mddev, bio);
 116                 return;
 117         }
 118
 119         mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
 120
 121         mp_bh->master_bio = bio;
 122         mp_bh->mddev = mddev;
 123
 124         mp_bh->path = multipath_map(conf);
 125         if (mp_bh->path < 0) {
 126                 bio_io_error(bio);
 127                 mempool_free(mp_bh, conf->pool);
 128                 return;
 129         }
 130         multipath = conf->multipaths + mp_bh->path;
 131
 132         bio_init(&mp_bh->bio);
 133         __bio_clone_fast(&mp_bh->bio, bio);
 134
 135         mp_bh->bio.bi_iter.bi_sector += multipath->rdev->data_offset;
 136         mp_bh->bio.bi_bdev = multipath->rdev->bdev;
 137         mp_bh->bio.bi_rw |= REQ_FAILFAST_TRANSPORT;
 138         mp_bh->bio.bi_end_io = multipath_end_request;
 139         mp_bh->bio.bi_private = mp_bh;
 140         generic_make_request(&mp_bh->bio);
 141         return;
 142 }
 143
 144 static void multipath_status (struct seq_file *seq, struct mddev *mddev)
 145 {
 146         struct mpconf *conf = mddev->private;
 147         int i;
 148
 149         seq_printf (seq, " [%d/%d] [", conf->raid_disks,
 150                     conf->raid_disks - mddev->degraded);
 151         for (i = 0; i < conf->raid_disks; i++)
 152                 seq_printf (seq, "%s",
 153                                conf->multipaths[i].rdev &&
 154                                test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
 155         seq_printf (seq, "]");
 156 }
 157
 158 static int multipath_congested(struct mddev *mddev, int bits)
 159 {
 160         struct mpconf *conf = mddev->private;
 161         int i, ret = 0;
 162
 163         rcu_read_lock();
 164         for (i = 0; i < mddev->raid_disks ; i++) {
 165                 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
 166                 if (rdev && !test_bit(Faulty, &rdev->flags)) {
 167                         struct request_queue *q = bdev_get_queue(rdev->bdev);
 168
 169                         ret |= bdi_congested(&q->backing_dev_info, bits);
 170                         /* Just like multipath_map, we just check the
 171                          * first available device
 172                          */
 173                         break;
 174                 }
 175         }
 176         rcu_read_unlock();
 177         return ret;
 178 }
 179
 180 /*
 181  * Careful, this can execute in IRQ contexts as well!
 182  */
 183 static void multipath_error (struct mddev *mddev, struct md_rdev *rdev)
 184 {
 185         struct mpconf *conf = mddev->private;
 186         char b[BDEVNAME_SIZE];
 187
 188         if (conf->raid_disks - mddev->degraded <= 1) {
 189                 /*
 190                  * Uh oh, we can do nothing if this is our last path, but
 191                  * first check if this is a queued request for a device
 192                  * which has just failed.
 193                  */
 194                 printk(KERN_ALERT
 195                        "multipath: only one IO path left and IO error.\n");
 196                 /* leave it active... it's all we have */
 197                 return;
 198         }
 199         /*
 200          * Mark disk as unusable
 201          */
 202         if (test_and_clear_bit(In_sync, &rdev->flags)) {
 203                 unsigned long flags;
 204                 spin_lock_irqsave(&conf->device_lock, flags);
 205                 mddev->degraded++;
 206                 spin_unlock_irqrestore(&conf->device_lock, flags);
 207         }
 208         set_bit(Faulty, &rdev->flags);
 209         set_bit(MD_CHANGE_DEVS, &mddev->flags);
 210         printk(KERN_ALERT "multipath: IO failure on %s,"
 211                " disabling IO path.\n"
 212                "multipath: Operation continuing"
 213                " on %d IO paths.\n",
 214                bdevname(rdev->bdev, b),
 215                conf->raid_disks - mddev->degraded);
 216 }
 217
 218 static void print_multipath_conf (struct mpconf *conf)
 219 {
 220         int i;
 221         struct multipath_info *tmp;
 222
 223         printk("MULTIPATH conf printout:\n");
 224         if (!conf) {
 225                 printk("(conf==NULL)\n");
 226                 return;
 227         }
 228         printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
 229                          conf->raid_disks);
 230
 231         for (i = 0; i < conf->raid_disks; i++) {
 232                 char b[BDEVNAME_SIZE];
 233                 tmp = conf->multipaths + i;
 234                 if (tmp->rdev)
 235                         printk(" disk%d, o:%d, dev:%s\n",
 236                                 i,!test_bit(Faulty, &tmp->rdev->flags),
 237                                bdevname(tmp->rdev->bdev,b));
 238         }
 239 }
 240
 241 static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
 242 {
 243         struct mpconf *conf = mddev->private;
 244         struct request_queue *q;
 245         int err = -EEXIST;
 246         int path;
 247         struct multipath_info *p;
 248         int first = 0;
 249         int last = mddev->raid_disks - 1;
 250
 251         if (rdev->raid_disk >= 0)
 252                 first = last = rdev->raid_disk;
 253
 254         print_multipath_conf(conf);
 255
 256         for (path = first; path <= last; path++)
 257                 if ((p=conf->multipaths+path)->rdev == NULL) {
 258                         q = rdev->bdev->bd_disk->queue;
 259                         disk_stack_limits(mddev->gendisk, rdev->bdev,
 260                                           rdev->data_offset << 9);
 261
 262                         err = md_integrity_add_rdev(rdev, mddev);
 263                         if (err)
 264                                 break;
 265                         spin_lock_irq(&conf->device_lock);
 266                         mddev->degraded--;
 267                         rdev->raid_disk = path;
 268                         set_bit(In_sync, &rdev->flags);
 269                         spin_unlock_irq(&conf->device_lock);
 270                         rcu_assign_pointer(p->rdev, rdev);
 271                         err = 0;
 272                         break;
 273                 }
 274
 275         print_multipath_conf(conf);
 276
 277         return err;
 278 }
 279
 280 static int multipath_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
 281 {
 282         struct mpconf *conf = mddev->private;
 283         int err = 0;
 284         int number = rdev->raid_disk;
 285         struct multipath_info *p = conf->multipaths + number;
 286
 287         print_multipath_conf(conf);
 288
 289         if (rdev == p->rdev) {
 290                 if (test_bit(In_sync, &rdev->flags) ||
 291                     atomic_read(&rdev->nr_pending)) {
 292                         printk(KERN_ERR "hot-remove-disk, slot %d is identified"
 293                                " but is still operational!\n", number);
 294                         err = -EBUSY;
 295                         goto abort;
 296                 }
 297                 p->rdev = NULL;
 298                 synchronize_rcu();
 299                 if (atomic_read(&rdev->nr_pending)) {
 300                         /* lost the race, try later */
 301                         err = -EBUSY;
 302                         p->rdev = rdev;
 303                         goto abort;
 304                 }
 305                 err = md_integrity_register(mddev);
 306         }
 307 abort:
 308
 309         print_multipath_conf(conf);
 310         return err;
 311 }
 312
 313 /*
 314  * This is a kernel thread which:
 315  *
 316  *      1.      Retries failed read operations on working multipaths.
 317  *      2.      Updates the raid superblock when problems encounter.
 318  *      3.      Performs writes following reads for array syncronising.
 319  */
 320
 321 static void multipathd(struct md_thread *thread)
 322 {
 323         struct mddev *mddev = thread->mddev;
 324         struct multipath_bh *mp_bh;
 325         struct bio *bio;
 326         unsigned long flags;
 327         struct mpconf *conf = mddev->private;
 328         struct list_head *head = &conf->retry_list;
 329
 330         md_check_recovery(mddev);
 331         for (;;) {
 332                 char b[BDEVNAME_SIZE];
 333                 spin_lock_irqsave(&conf->device_lock, flags);
 334                 if (list_empty(head))
 335                         break;
 336                 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
 337                 list_del(head->prev);
 338                 spin_unlock_irqrestore(&conf->device_lock, flags);
 339
 340                 bio = &mp_bh->bio;
 341                 bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector;
 342
 343                 if ((mp_bh->path = multipath_map (conf))<0) {
 344                         printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
 345                                 " error for block %llu\n",
 346                                 bdevname(bio->bi_bdev,b),
 347                                 (unsigned long long)bio->bi_iter.bi_sector);
 348                         multipath_end_bh_io(mp_bh, -EIO);
 349                 } else {
 350                         printk(KERN_ERR "multipath: %s: redirecting sector %llu"
 351                                 " 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_rw |= 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                 printk("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                 printk(KERN_ERR
 401                         "multipath: couldn't allocate memory for %s\n",
 402                         mdname(mddev));
 403                 goto out;
 404         }
 405
 406         conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
 407                                    GFP_KERNEL);
 408         if (!conf->multipaths) {
 409                 printk(KERN_ERR
 410                         "multipath: couldn't allocate memory for %s\n",
 411                         mdname(mddev));
 412                 goto out_free_conf;
 413         }
 414
 415         working_disks = 0;
 416         rdev_for_each(rdev, mddev) {
 417                 disk_idx = rdev->raid_disk;
 418                 if (disk_idx < 0 ||
 419                     disk_idx >= mddev->raid_disks)
 420                         continue;
 421
 422                 disk = conf->multipaths + disk_idx;
 423                 disk->rdev = rdev;
 424                 disk_stack_limits(mddev->gendisk, rdev->bdev,
 425                                   rdev->data_offset << 9);
 426
 427                 if (!test_bit(Faulty, &rdev->flags))
 428                         working_disks++;
 429         }
 430
 431         conf->raid_disks = mddev->raid_disks;
 432         conf->mddev = mddev;
 433         spin_lock_init(&conf->device_lock);
 434         INIT_LIST_HEAD(&conf->retry_list);
 435
 436         if (!working_disks) {
 437                 printk(KERN_ERR "multipath: no operational IO paths for %s\n",
 438                         mdname(mddev));
 439                 goto out_free_conf;
 440         }
 441         mddev->degraded = conf->raid_disks - working_disks;
 442
 443         conf->pool = mempool_create_kmalloc_pool(NR_RESERVED_BUFS,
 444                                                  sizeof(struct multipath_bh));
 445         if (conf->pool == NULL) {
 446                 printk(KERN_ERR
 447                         "multipath: couldn't allocate memory for %s\n",
 448                         mdname(mddev));
 449                 goto out_free_conf;
 450         }
 451
 452         {
 453                 mddev->thread = md_register_thread(multipathd, mddev,
 454                                                    "multipath");
 455                 if (!mddev->thread) {
 456                         printk(KERN_ERR "multipath: couldn't allocate thread"
 457                                 " for %s\n", mdname(mddev));
 458                         goto out_free_conf;
 459                 }
 460         }
 461
 462         printk(KERN_INFO
 463                 "multipath: array %s active with %d out of %d IO paths\n",
 464                 mdname(mddev), conf->raid_disks - mddev->degraded,
 465                mddev->raid_disks);
 466         /*
 467          * Ok, everything is just fine now
 468          */
 469         md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
 470
 471         if (md_integrity_register(mddev))
 472                 goto out_free_conf;
 473
 474         return 0;
 475
 476 out_free_conf:
 477         mempool_destroy(conf->pool);
 478         kfree(conf->multipaths);
 479         kfree(conf);
 480         mddev->private = NULL;
 481 out:
 482         return -EIO;
 483 }
 484
 485 static void multipath_free(struct mddev *mddev, void *priv)
 486 {
 487         struct mpconf *conf = priv;
 488
 489         mempool_destroy(conf->pool);
 490         kfree(conf->multipaths);
 491         kfree(conf);
 492 }
 493
 494 static struct md_personality multipath_personality =
 495 {
 496         .name           = "multipath",
 497         .level          = LEVEL_MULTIPATH,
 498         .owner          = THIS_MODULE,
 499         .make_request   = multipath_make_request,
 500         .run            = multipath_run,
 501         .free           = multipath_free,
 502         .status         = multipath_status,
 503         .error_handler  = multipath_error,
 504         .hot_add_disk   = multipath_add_disk,
 505         .hot_remove_disk= multipath_remove_disk,
 506         .size           = multipath_size,
 507         .congested      = multipath_congested,
 508 };
 509
 510 static int __init multipath_init (void)
 511 {
 512         return register_md_personality (&multipath_personality);
 513 }
 514
 515 static void __exit multipath_exit (void)
 516 {
 517         unregister_md_personality (&multipath_personality);
 518 }
 519
 520 module_init(multipath_init);
 521 module_exit(multipath_exit);
 522 MODULE_LICENSE("GPL");
 523 MODULE_DESCRIPTION("simple multi-path personality for MD");
 524 MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
 525 MODULE_ALIAS("md-multipath");
 526 MODULE_ALIAS("md-level--4");