sys/dev/disk/dm/device-mapper.c

   1 /*        $NetBSD: device-mapper.c,v 1.22 2010/03/26 15:46:04 jakllsch Exp $ */
   2
   3 /*
   4  * Copyright (c) 2010-2011 Alex Hornung <alex@alexhornung.com>
   5  * Copyright (c) 2010 The NetBSD Foundation, Inc.
   6  * All rights reserved.
   7  *
   8  * This code is derived from software contributed to The NetBSD Foundation
   9  * by Adam Hamsik.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
  22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
  24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  30  * POSSIBILITY OF SUCH DAMAGE.
  31  */
  32
  33 /*
  34  * I want to say thank you to all people who helped me with this project.
  35  */
  36
  37 #include <sys/types.h>
  38 #include <sys/ctype.h>
  39
  40 #include <sys/buf.h>
  41 #include <sys/conf.h>
  42 #include <sys/device.h>
  43 #include <sys/disk.h>
  44 #include <sys/disklabel.h>
  45 #include <sys/dtype.h>
  46 #include <sys/malloc.h>
  47 #include <sys/module.h>
  48 #include <sys/sysctl.h>
  49 #include <dev/disk/dm/dm.h>
  50
  51 #include "netbsd-dm.h"
  52
  53 static  d_ioctl_t       dmioctl;
  54 static  d_open_t        dmopen;
  55 static  d_close_t       dmclose;
  56 static  d_psize_t       dmsize;
  57 static  d_strategy_t    dmstrategy;
  58 static  d_dump_t        dmdump;
  59
  60 /* New module handle and destroy routines */
  61 static int dm_modcmd(module_t mod, int cmd, void *unused);
  62 static int dmdestroy(void);
  63
  64 static void dm_doinit(void);
  65
  66 static int dm_cmd_to_fun(prop_dictionary_t);
  67 static int disk_ioctl_switch(cdev_t, u_long, void *);
  68 static int dm_ioctl_switch(u_long);
  69 #if 0
  70 static void dmminphys(struct buf *);
  71 #endif
  72
  73 /* ***Variable-definitions*** */
  74 struct dev_ops dm_ops = {
  75         { "dm", 0, D_DISK | D_MPSAFE },
  76         .d_open         = dmopen,
  77         .d_close        = dmclose,
  78         .d_read         = physread,
  79         .d_write        = physwrite,
  80         .d_ioctl        = dmioctl,
  81         .d_strategy     = dmstrategy,
  82         .d_psize        = dmsize,
  83         .d_dump         = dmdump,
  84 /* D_DISK */
  85 };
  86
  87 MALLOC_DEFINE(M_DM, "dm", "Device Mapper allocations");
  88
  89 int dm_debug_level = 0;
  90
  91 extern uint64_t dm_dev_counter;
  92
  93 static cdev_t dmcdev;
  94
  95 static moduledata_t dm_mod = {
  96     "dm",
  97     dm_modcmd,
  98     NULL
  99 };
 100 DECLARE_MODULE(dm, dm_mod, SI_SUB_RAID, SI_ORDER_ANY);
 101 MODULE_VERSION(dm, 1);
 102
 103 /*
 104  * This array is used to translate cmd to function pointer.
 105  *
 106  * Interface between libdevmapper and lvm2tools uses different
 107  * names for one IOCTL call because libdevmapper do another thing
 108  * then. When I run "info" or "mknodes" libdevmapper will send same
 109  * ioctl to kernel but will do another things in userspace.
 110  *
 111  */
 112 static struct cmd_function cmd_fn[] = {
 113                 { .cmd = "version", .fn = NULL},
 114                 { .cmd = "targets", .fn = dm_list_versions_ioctl},
 115                 { .cmd = "create",  .fn = dm_dev_create_ioctl},
 116                 { .cmd = "info",    .fn = dm_dev_status_ioctl},
 117                 { .cmd = "mknodes", .fn = dm_dev_status_ioctl},
 118                 { .cmd = "names",   .fn = dm_dev_list_ioctl},
 119                 { .cmd = "suspend", .fn = dm_dev_suspend_ioctl},
 120                 { .cmd = "remove",  .fn = dm_dev_remove_ioctl},
 121                 { .cmd = "remove_all", .fn = dm_dev_remove_all_ioctl},
 122                 { .cmd = "rename",  .fn = dm_dev_rename_ioctl},
 123                 { .cmd = "resume",  .fn = dm_dev_resume_ioctl},
 124                 { .cmd = "clear",   .fn = dm_table_clear_ioctl},
 125                 { .cmd = "deps",    .fn = dm_table_deps_ioctl},
 126                 { .cmd = "reload",  .fn = dm_table_load_ioctl},
 127                 { .cmd = "status",  .fn = dm_table_status_ioctl},
 128                 { .cmd = "table",   .fn = dm_table_status_ioctl},
 129                 { .cmd = "message", .fn = dm_message_ioctl},
 130 };
 131
 132 /* New module handle routine */
 133 static int
 134 dm_modcmd(module_t mod, int cmd, void *unused)
 135 {
 136         int error;
 137
 138         error = 0;
 139
 140         switch (cmd) {
 141         case MOD_LOAD:
 142                 dm_doinit();
 143                 kprintf("Device Mapper version %d.%d.%d loaded\n",
 144                     DM_VERSION_MAJOR, DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL);
 145                 break;
 146
 147         case MOD_UNLOAD:
 148                 /*
 149                  * Disable unloading of dm module if there are any devices
 150                  * defined in driver. This is probably too strong we need
 151                  * to disable auto-unload only if there is mounted dm device
 152                  * present.
 153                  */
 154                 if (dm_dev_counter > 0)
 155                         return EBUSY;
 156
 157                 error = dmdestroy();
 158                 if (error)
 159                         break;
 160                 kprintf("Device Mapper unloaded\n");
 161                 break;
 162
 163         default:
 164                 break;
 165         }
 166
 167         return error;
 168 }
 169
 170 static void
 171 dm_doinit(void)
 172 {
 173         dm_target_init();
 174         dm_dev_init();
 175         dm_pdev_init();
 176         dmcdev = make_dev(&dm_ops, 0, UID_ROOT, GID_OPERATOR, 0640, "mapper/control");
 177 }
 178
 179 /* Destroy routine */
 180 static int
 181 dmdestroy(void)
 182 {
 183         destroy_dev(dmcdev);
 184
 185         dm_dev_uninit();
 186         dm_pdev_uninit();
 187         dm_target_uninit();
 188
 189         return 0;
 190 }
 191
 192 static int
 193 dmopen(struct dev_open_args *ap)
 194 {
 195         cdev_t dev = ap->a_head.a_dev;
 196         dm_dev_t *dmv;
 197
 198         /* Shortcut for the control device */
 199         if (minor(dev) == 0)
 200                 return 0;
 201
 202         if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
 203                 return ENXIO;
 204
 205         dmv->is_open = 1;
 206         dm_dev_unbusy(dmv);
 207
 208         aprint_debug("dm open routine called %" PRIu32 "\n",
 209             minor(ap->a_head.a_dev));
 210         return 0;
 211 }
 212
 213 static int
 214 dmclose(struct dev_close_args *ap)
 215 {
 216         cdev_t dev = ap->a_head.a_dev;
 217         dm_dev_t *dmv;
 218
 219         /* Shortcut for the control device */
 220         if (minor(dev) == 0)
 221                 return 0;
 222
 223         if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
 224                 return ENXIO;
 225
 226         dmv->is_open = 0;
 227         dm_dev_unbusy(dmv);
 228
 229         aprint_debug("dm close routine called %" PRIu32 "\n",
 230             minor(ap->a_head.a_dev));
 231         return 0;
 232 }
 233
 234
 235 static int
 236 dmioctl(struct dev_ioctl_args *ap)
 237 {
 238         cdev_t dev = ap->a_head.a_dev;
 239         u_long cmd = ap->a_cmd;
 240         void *data = ap->a_data;
 241         struct plistref *pref = (struct plistref *)data;
 242
 243         int r, err;
 244         prop_dictionary_t dm_dict_in;
 245
 246         err = r = 0;
 247
 248         aprint_debug("dmioctl called\n");
 249         KKASSERT(data != NULL);
 250
 251         if ((r = disk_ioctl_switch(dev, cmd, data)) != ENOTTY)
 252                 return r;  /* Handled disk ioctl */
 253
 254         if ((r = dm_ioctl_switch(cmd)) != 0)
 255                 return r;  /* Not NETBSD_DM_IOCTL */
 256
 257         if ((r = prop_dictionary_copyin_ioctl(pref, cmd, &dm_dict_in)) != 0)
 258                 return r;
 259
 260         if ((r = dm_check_version(dm_dict_in)) == 0)
 261                 err = dm_cmd_to_fun(dm_dict_in);
 262
 263         r = prop_dictionary_copyout_ioctl(pref, cmd, dm_dict_in);
 264         prop_object_release(dm_dict_in);
 265
 266         /* Return the dm ioctl error if any. */
 267         if (err)
 268                 return err;
 269         return r;
 270 }
 271
 272 /*
 273  * Translate command sent from libdevmapper to func.
 274  */
 275 static int
 276 dm_cmd_to_fun(prop_dictionary_t dm_dict)
 277 {
 278         int i, size;
 279         prop_string_t command;
 280         struct cmd_function *p = NULL;
 281
 282         size = sizeof(cmd_fn) / sizeof(cmd_fn[0]);
 283
 284         if ((command = prop_dictionary_get(dm_dict, DM_IOCTL_COMMAND)) == NULL)
 285                 return EINVAL;
 286
 287         for (i = 0; i < size; i++) {
 288                 p = &cmd_fn[i];
 289                 if (prop_string_equals_cstring(command, p->cmd))
 290                         break;
 291         }
 292
 293         KKASSERT(p);
 294         if (i == size) {
 295                 aprint_debug("Unknown ioctl\n");
 296                 return EINVAL;
 297         }
 298
 299         aprint_debug("ioctl %s called %p\n", p->cmd, p->fn);
 300         if (p->fn == NULL)
 301                 return 0;  /* No handler required */
 302
 303         return p->fn(dm_dict);
 304 }
 305
 306 /* Call apropriate ioctl handler function. */
 307 static int
 308 dm_ioctl_switch(u_long cmd)
 309 {
 310
 311         switch(cmd) {
 312         case NETBSD_DM_IOCTL:
 313                 aprint_debug("dm NETBSD_DM_IOCTL called\n");
 314                 break;
 315         default:
 316                 aprint_debug("dm unknown ioctl called\n");
 317                 return ENOTTY;
 318                 break; /* NOT REACHED */
 319         }
 320
 321         return 0;
 322 }
 323
 324  /*
 325   * Check for disk specific ioctls.
 326   */
 327
 328 static int
 329 disk_ioctl_switch(cdev_t dev, u_long cmd, void *data)
 330 {
 331         dm_dev_t *dmv;
 332
 333         /* disk ioctls make sense only on block devices */
 334         if (minor(dev) == 0)
 335                 return ENOTTY;
 336
 337         switch(cmd) {
 338         case DIOCGPART:
 339                 if ((dmv = dev->si_drv1) == NULL)
 340                         return ENODEV;
 341                 if (dmv->diskp->d_info.d_media_blksize == 0) {
 342                         return ENOTSUP;
 343                 } else {
 344                         u_int64_t size;
 345                         struct partinfo *dpart = data;
 346                         bzero(dpart, sizeof(*dpart));
 347
 348                         size = dm_table_size(&dmv->table_head);
 349                         dpart->media_offset  = 0;
 350                         dpart->media_size    = size * DEV_BSIZE;
 351                         dpart->media_blocks  = size;
 352                         dpart->media_blksize = DEV_BSIZE;
 353                         dpart->fstype = FS_BSDFFS;
 354                 }
 355                 break;
 356
 357         default:
 358                 aprint_debug("unknown disk_ioctl called\n");
 359                 return ENOTTY;
 360                 break; /* NOT REACHED */
 361         }
 362
 363         return 0;
 364 }
 365
 366 /*
 367  * Do all IO operations on dm logical devices.
 368  */
 369 static int
 370 dmstrategy(struct dev_strategy_args *ap)
 371 {
 372         cdev_t dev = ap->a_head.a_dev;
 373         struct bio *bio = ap->a_bio;
 374         struct buf *bp = bio->bio_buf;
 375         int bypass;
 376
 377         dm_dev_t *dmv;
 378         dm_table_t  *tbl;
 379         dm_table_entry_t *table_en;
 380         struct buf *nestbuf;
 381
 382         uint64_t buf_start, buf_len, issued_len;
 383         uint64_t table_start, table_end;
 384         uint64_t start, end;
 385
 386         buf_start = bio->bio_offset;
 387         buf_len = bp->b_bcount;
 388
 389         tbl = NULL;
 390
 391         table_end = 0;
 392         issued_len = 0;
 393
 394         dmv = dev->si_drv1;
 395
 396         switch(bp->b_cmd) {
 397         case BUF_CMD_READ:
 398         case BUF_CMD_WRITE:
 399         case BUF_CMD_FREEBLKS:
 400                 bypass = 0;
 401                 break;
 402         case BUF_CMD_FLUSH:
 403                 bypass = 1;
 404                 KKASSERT(buf_len == 0);
 405                 break;
 406         default:
 407                 bp->b_error = EIO;
 408                 bp->b_resid = bp->b_bcount;
 409                 biodone(bio);
 410                 return 0;
 411         }
 412
 413         if (bypass == 0 &&
 414             bounds_check_with_mediasize(bio, DEV_BSIZE,
 415                                         dm_table_size(&dmv->table_head)) <= 0) {
 416                 bp->b_resid = bp->b_bcount;
 417                 biodone(bio);
 418                 return 0;
 419         }
 420
 421         /* Select active table */
 422         tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
 423
 424         nestiobuf_init(bio);
 425         devstat_start_transaction(&dmv->stats);
 426
 427         /*
 428          * Find out what tables I want to select.
 429          */
 430         SLIST_FOREACH(table_en, tbl, next) {
 431                 /*
 432                  * I need need number of bytes not blocks.
 433                  */
 434                 table_start = table_en->start * DEV_BSIZE;
 435                 table_end = table_start + (table_en->length) * DEV_BSIZE;
 436
 437                 /*
 438                  * Calculate the start and end
 439                  */
 440                 start = MAX(table_start, buf_start);
 441                 end = MIN(table_end, buf_start + buf_len);
 442
 443                 if (dm_debug_level) {
 444                         aprint_normal("----------------------------------------\n");
 445                         aprint_normal("table_start %010" PRIu64", table_end %010"
 446                             PRIu64 "\n", table_start, table_end);
 447                         aprint_normal("buf_start %010" PRIu64", buf_len %010"
 448                             PRIu64"\n", buf_start, buf_len);
 449                         aprint_normal("start-buf_start %010"PRIu64", end %010"
 450                             PRIu64"\n", start - buf_start, end);
 451                         aprint_normal("start %010" PRIu64", end %010"
 452                             PRIu64"\n", start, end);
 453                         aprint_normal("----------------------------------------\n");
 454                 }
 455
 456                 if (bypass) {
 457                         nestbuf = getpbuf(NULL);
 458                         nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
 459
 460                         nestiobuf_add(bio, nestbuf, 0, 0, &dmv->stats);
 461                         nestbuf->b_bio1.bio_offset = 0;
 462                         table_en->target->strategy(table_en, nestbuf);
 463                 } else if (start < end) {
 464                         nestbuf = getpbuf(NULL);
 465                         nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
 466
 467                         nestiobuf_add(bio, nestbuf,
 468                                       start - buf_start, (end - start),
 469                                       &dmv->stats);
 470                         issued_len += end - start;
 471
 472                         nestbuf->b_bio1.bio_offset = (start - table_start);
 473                         table_en->target->strategy(table_en, nestbuf);
 474                 }
 475         }
 476
 477         if (issued_len < buf_len)
 478                 nestiobuf_error(bio, EINVAL);
 479         nestiobuf_start(bio);
 480         dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
 481
 482         return 0;
 483 }
 484
 485 static int
 486 dmdump(struct dev_dump_args *ap)
 487 {
 488         cdev_t dev = ap->a_head.a_dev;
 489         dm_dev_t *dmv;
 490         dm_table_t  *tbl;
 491         dm_table_entry_t *table_en;
 492         uint64_t buf_start, buf_len, issued_len;
 493         uint64_t table_start, table_end;
 494         uint64_t start, end, data_offset;
 495         off_t offset;
 496         size_t length;
 497         int error = 0;
 498
 499         buf_start = ap->a_offset;
 500         buf_len = ap->a_length;
 501
 502         tbl = NULL;
 503
 504         table_end = 0;
 505         issued_len = 0;
 506
 507         dmv = dev->si_drv1;
 508
 509         /* Select active table */
 510         tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
 511
 512
 513         /*
 514          * Find out what tables I want to select.
 515          */
 516         SLIST_FOREACH(table_en, tbl, next) {
 517                 /*
 518                  * I need need number of bytes not blocks.
 519                  */
 520                 table_start = table_en->start * DEV_BSIZE;
 521                 table_end = table_start + (table_en->length) * DEV_BSIZE;
 522
 523                 /*
 524                  * Calculate the start and end
 525                  */
 526                 start = MAX(table_start, buf_start);
 527                 end = MIN(table_end, buf_start + buf_len);
 528
 529                 if (ap->a_length == 0) {
 530                         if (table_en->target->dump == NULL) {
 531                                 error = ENXIO;
 532                                 goto out;
 533                         }
 534
 535                         table_en->target->dump(table_en, NULL, 0, 0);
 536                 } else if (start < end) {
 537                         data_offset = start - buf_start;
 538                         offset = start - table_start;
 539                         length = end - start;
 540
 541                         if (table_en->target->dump == NULL) {
 542                                 error = ENXIO;
 543                                 goto out;
 544                         }
 545
 546                         table_en->target->dump(table_en,
 547                             (char *)ap->a_virtual + data_offset,
 548                             length, offset);
 549
 550                         issued_len += end - start;
 551                 }
 552         }
 553
 554         if (issued_len < buf_len)
 555                 error = EINVAL;
 556
 557 out:
 558         dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
 559
 560         return error;
 561 }
 562
 563 static int
 564 dmsize(struct dev_psize_args *ap)
 565 {
 566         cdev_t dev = ap->a_head.a_dev;
 567         dm_dev_t *dmv;
 568         uint64_t size;
 569
 570         size = 0;
 571
 572         if ((dmv = dev->si_drv1) == NULL)
 573                 return ENXIO;
 574
 575         size = dm_table_size(&dmv->table_head);
 576         ap->a_result = (int64_t)size;
 577
 578         return 0;
 579 }
 580
 581 #if 0
 582 static void
 583 dmminphys(struct buf *bp)
 584 {
 585
 586         bp->b_bcount = MIN(bp->b_bcount, MAXPHYS);
 587 }
 588 #endif
 589
 590 void
 591 dmsetdiskinfo(struct disk *disk, dm_table_head_t *head)
 592 {
 593         struct disk_info info;
 594         uint64_t dmp_size;
 595
 596         dmp_size = dm_table_size(head);
 597
 598         bzero(&info, sizeof(struct disk_info));
 599         info.d_media_blksize = DEV_BSIZE;
 600         info.d_media_blocks = dmp_size;
 601 #if 0
 602         /* this is set by disk_setdiskinfo */
 603         info.d_media_size = dmp_size * DEV_BSIZE;
 604 #endif
 605         info.d_dsflags = DSO_MBRQUIET | DSO_DEVICEMAPPER | DSO_RAWPSIZE;
 606
 607         info.d_secpertrack = 32;
 608         info.d_nheads = 64;
 609         info.d_secpercyl = info.d_secpertrack * info.d_nheads;
 610         info.d_ncylinders = dmp_size / info.d_secpercyl;
 611
 612         disk_setdiskinfo(disk, &info);
 613 }
 614
 615 /*
 616  * Transform char s to uint64_t offset number.
 617  */
 618 uint64_t
 619 atoi64(const char *s)
 620 {
 621         uint64_t n;
 622         n = 0;
 623
 624         while (*s != '\0') {
 625                 if (!isdigit(*s))
 626                         break;
 627
 628                 n = (10 * n) + (*s - '0');
 629                 s++;
 630         }
 631
 632         return n;
 633 }
 634
 635 void
 636 dm_builtin_init(void *arg)
 637 {
 638         modeventhand_t evh = (modeventhand_t)arg;
 639
 640         KKASSERT(evh != NULL);
 641         evh(NULL, MOD_LOAD, NULL);
 642 }
 643
 644 void
 645 dm_builtin_uninit(void *arg)
 646 {
 647         modeventhand_t evh = (modeventhand_t)arg;
 648
 649         KKASSERT(evh != NULL);
 650         evh(NULL, MOD_UNLOAD, NULL);
 651 }
 652
 653 TUNABLE_INT("debug.dm_debug", &dm_debug_level);
 654 SYSCTL_INT(_debug, OID_AUTO, dm_debug, CTLFLAG_RW, &dm_debug_level,
 655                0, "Enable device mapper debugging");
 656