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/param.h>
  39 #include <sys/ctype.h>
  40
  41 #include <sys/buf.h>
  42 #include <sys/conf.h>
  43 #include <sys/device.h>
  44 #include <sys/disk.h>
  45 #include <sys/disklabel.h>
  46 #include <sys/dtype.h>
  47 #include <sys/malloc.h>
  48 #include <sys/module.h>
  49 #include <sys/sysctl.h>
  50 #include <dev/disk/dm/dm.h>
  51
  52 #include "netbsd-dm.h"
  53
  54 static  d_ioctl_t       dmioctl;
  55 static  d_open_t        dmopen;
  56 static  d_close_t       dmclose;
  57 static  d_psize_t       dmsize;
  58 static  d_strategy_t    dmstrategy;
  59 static  d_dump_t        dmdump;
  60
  61 /* attach and detach routines */
  62 void dmattach(int);
  63 static int dm_modcmd(module_t mod, int cmd, void *unused);
  64 static int dmdestroy(void);
  65
  66 static void dm_doinit(void);
  67
  68 static int dm_cmd_to_fun(prop_dictionary_t);
  69 static int disk_ioctl_switch(cdev_t, u_long, void *);
  70 static int dm_ioctl_switch(u_long);
  71 #if 0
  72 static void dmminphys(struct buf *);
  73 #endif
  74
  75 /* ***Variable-definitions*** */
  76 struct dev_ops dm_ops = {
  77         { "dm", 0, D_DISK | D_MPSAFE },
  78         .d_open         = dmopen,
  79         .d_close        = dmclose,
  80         .d_read         = physread,
  81         .d_write        = physwrite,
  82         .d_ioctl        = dmioctl,
  83         .d_strategy     = dmstrategy,
  84         .d_psize        = dmsize,
  85         .d_dump         = dmdump,
  86 /* D_DISK */
  87 };
  88
  89 MALLOC_DEFINE(M_DM, "dm", "Device Mapper allocations");
  90
  91 int dm_debug_level = 0;
  92
  93 extern uint64_t dm_dev_counter;
  94
  95 static cdev_t dmcdev;
  96
  97 static moduledata_t dm_mod = {
  98     "dm",
  99     dm_modcmd,
 100     NULL
 101 };
 102 DECLARE_MODULE(dm, dm_mod, SI_SUB_RAID, SI_ORDER_ANY);
 103 MODULE_VERSION(dm, 1);
 104
 105 /*
 106  * This array is used to translate cmd to function pointer.
 107  *
 108  * Interface between libdevmapper and lvm2tools uses different
 109  * names for one IOCTL call because libdevmapper do another thing
 110  * then. When I run "info" or "mknodes" libdevmapper will send same
 111  * ioctl to kernel but will do another things in userspace.
 112  *
 113  */
 114 static struct cmd_function cmd_fn[] = {
 115                 { .cmd = "version", .fn = dm_get_version_ioctl},
 116                 { .cmd = "targets", .fn = dm_list_versions_ioctl},
 117                 { .cmd = "create",  .fn = dm_dev_create_ioctl},
 118                 { .cmd = "info",    .fn = dm_dev_status_ioctl},
 119                 { .cmd = "mknodes", .fn = dm_dev_status_ioctl},
 120                 { .cmd = "names",   .fn = dm_dev_list_ioctl},
 121                 { .cmd = "suspend", .fn = dm_dev_suspend_ioctl},
 122                 { .cmd = "remove",  .fn = dm_dev_remove_ioctl},
 123                 { .cmd = "remove_all", .fn = dm_dev_remove_all_ioctl},
 124                 { .cmd = "rename",  .fn = dm_dev_rename_ioctl},
 125                 { .cmd = "resume",  .fn = dm_dev_resume_ioctl},
 126                 { .cmd = "clear",   .fn = dm_table_clear_ioctl},
 127                 { .cmd = "deps",    .fn = dm_table_deps_ioctl},
 128                 { .cmd = "reload",  .fn = dm_table_load_ioctl},
 129                 { .cmd = "status",  .fn = dm_table_status_ioctl},
 130                 { .cmd = "table",   .fn = dm_table_status_ioctl},
 131                 { .cmd = "message", .fn = dm_message_ioctl},
 132                 {NULL, NULL}
 133 };
 134
 135 /* New module handle routine */
 136 static int
 137 dm_modcmd(module_t mod, int cmd, void *unused)
 138 {
 139         int error;
 140
 141         error = 0;
 142
 143         switch (cmd) {
 144         case MOD_LOAD:
 145                 dm_doinit();
 146                 kprintf("Device Mapper version %d.%d.%d loaded\n",
 147                     DM_VERSION_MAJOR, DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL);
 148                 break;
 149
 150         case MOD_UNLOAD:
 151                 /*
 152                  * Disable unloading of dm module if there are any devices
 153                  * defined in driver. This is probably too strong we need
 154                  * to disable auto-unload only if there is mounted dm device
 155                  * present.
 156                  */
 157                 if (dm_dev_counter > 0)
 158                         return EBUSY;
 159
 160                 error = dmdestroy();
 161                 if (error)
 162                         break;
 163                 kprintf("Device Mapper unloaded\n");
 164                 break;
 165
 166         default:
 167                 break;
 168         }
 169
 170         return error;
 171 }
 172
 173 static void
 174 dm_doinit(void)
 175 {
 176         dm_target_init();
 177         dm_dev_init();
 178         dm_pdev_init();
 179         dmcdev = make_dev(&dm_ops, 0, UID_ROOT, GID_OPERATOR, 0640, "mapper/control");
 180 }
 181
 182 /* Destroy routine */
 183 static int
 184 dmdestroy(void)
 185 {
 186         destroy_dev(dmcdev);
 187
 188         dm_dev_uninit();
 189         dm_pdev_uninit();
 190         dm_target_uninit();
 191
 192         return 0;
 193 }
 194
 195 static int
 196 dmopen(struct dev_open_args *ap)
 197 {
 198         cdev_t dev = ap->a_head.a_dev;
 199         dm_dev_t *dmv;
 200
 201         /* Shortcut for the control device */
 202         if (minor(dev) == 0)
 203                 return 0;
 204
 205         if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
 206                 return ENXIO;
 207
 208         dmv->is_open = 1;
 209         dm_dev_unbusy(dmv);
 210
 211         aprint_debug("dm open routine called %" PRIu32 "\n",
 212             minor(ap->a_head.a_dev));
 213         return 0;
 214 }
 215
 216 static int
 217 dmclose(struct dev_close_args *ap)
 218 {
 219         cdev_t dev = ap->a_head.a_dev;
 220         dm_dev_t *dmv;
 221
 222         /* Shortcut for the control device */
 223         if (minor(dev) == 0)
 224                 return 0;
 225
 226         if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
 227                 return ENXIO;
 228
 229         dmv->is_open = 0;
 230         dm_dev_unbusy(dmv);
 231
 232         aprint_debug("dm close routine called %" PRIu32 "\n",
 233             minor(ap->a_head.a_dev));
 234         return 0;
 235 }
 236
 237
 238 static int
 239 dmioctl(struct dev_ioctl_args *ap)
 240 {
 241         cdev_t dev = ap->a_head.a_dev;
 242         u_long cmd = ap->a_cmd;
 243         void *data = ap->a_data;
 244
 245         int r, err;
 246         prop_dictionary_t dm_dict_in;
 247
 248         err = r = 0;
 249
 250         aprint_debug("dmioctl called\n");
 251
 252         KKASSERT(data != NULL);
 253
 254         if (( r = disk_ioctl_switch(dev, cmd, data)) == ENOTTY) {
 255                 struct plistref *pref = (struct plistref *) data;
 256
 257                 /* Check if we were called with NETBSD_DM_IOCTL ioctl
 258                    otherwise quit. */
 259                 if ((r = dm_ioctl_switch(cmd)) != 0)
 260                         return r;
 261
 262                 if((r = prop_dictionary_copyin_ioctl(pref, cmd, &dm_dict_in)) != 0)
 263                         return r;
 264
 265                 if ((r = dm_check_version(dm_dict_in)) != 0)
 266                         goto cleanup_exit;
 267
 268                 /* run ioctl routine */
 269                 if ((err = dm_cmd_to_fun(dm_dict_in)) != 0)
 270                         goto cleanup_exit;
 271
 272 cleanup_exit:
 273                 r = prop_dictionary_copyout_ioctl(pref, cmd, dm_dict_in);
 274                 prop_object_release(dm_dict_in);
 275         }
 276
 277         /*
 278          * Return the error of the actual command if one one has
 279          * happened. Otherwise return 'r' which indicates errors
 280          * that occurred during helper operations.
 281          */
 282         return (err != 0)?err:r;
 283 }
 284
 285 /*
 286  * Translate command sent from libdevmapper to func.
 287  */
 288 static int
 289 dm_cmd_to_fun(prop_dictionary_t dm_dict){
 290         int i, r;
 291         prop_string_t command;
 292
 293         r = 0;
 294
 295         if ((command = prop_dictionary_get(dm_dict, DM_IOCTL_COMMAND)) == NULL)
 296                 return EINVAL;
 297
 298         for(i = 0; cmd_fn[i].cmd != NULL; i++)
 299                 if (prop_string_equals_cstring(command, cmd_fn[i].cmd))
 300                         break;
 301
 302         if (cmd_fn[i].cmd == NULL)
 303                 return EINVAL;
 304
 305         aprint_debug("ioctl %s called\n", cmd_fn[i].cmd);
 306         r = cmd_fn[i].fn(dm_dict);
 307
 308         return r;
 309 }
 310
 311 /* Call apropriate ioctl handler function. */
 312 static int
 313 dm_ioctl_switch(u_long cmd)
 314 {
 315
 316         switch(cmd) {
 317
 318         case NETBSD_DM_IOCTL:
 319                 aprint_debug("dm NetBSD_DM_IOCTL called\n");
 320                 break;
 321         default:
 322                  aprint_debug("dm unknown ioctl called\n");
 323                  return ENOTTY;
 324                  break; /* NOT REACHED */
 325         }
 326
 327          return 0;
 328 }
 329
 330  /*
 331   * Check for disk specific ioctls.
 332   */
 333
 334 static int
 335 disk_ioctl_switch(cdev_t dev, u_long cmd, void *data)
 336 {
 337         dm_dev_t *dmv;
 338
 339         /* disk ioctls make sense only on block devices */
 340         if (minor(dev) == 0)
 341                 return ENOTTY;
 342
 343         switch(cmd) {
 344         case DIOCGPART:
 345         {
 346                 struct partinfo *dpart;
 347                 u_int64_t size;
 348                 dpart = data;
 349                 bzero(dpart, sizeof(*dpart));
 350
 351                 if ((dmv = dev->si_drv1) == NULL)
 352                         return ENODEV;
 353                 if (dmv->diskp->d_info.d_media_blksize == 0) {
 354                         return ENOTSUP;
 355                 } else {
 356                         size = dm_table_size(&dmv->table_head);
 357                         dpart->media_offset  = 0;
 358                         dpart->media_size    = size * DEV_BSIZE;
 359                         dpart->media_blocks  = size;
 360                         dpart->media_blksize = DEV_BSIZE;
 361                         dpart->fstype = FS_BSDFFS;
 362                 }
 363                 break;
 364         }
 365
 366         default:
 367                 aprint_debug("unknown disk_ioctl called\n");
 368                 return ENOTTY;
 369                 break; /* NOT REACHED */
 370         }
 371
 372         return 0;
 373 }
 374
 375 /*
 376  * Do all IO operations on dm logical devices.
 377  */
 378 static int
 379 dmstrategy(struct dev_strategy_args *ap)
 380 {
 381         cdev_t dev = ap->a_head.a_dev;
 382         struct bio *bio = ap->a_bio;
 383         struct buf *bp = bio->bio_buf;
 384         int bypass;
 385
 386         dm_dev_t *dmv;
 387         dm_table_t  *tbl;
 388         dm_table_entry_t *table_en;
 389         struct buf *nestbuf;
 390
 391         uint64_t buf_start, buf_len, issued_len;
 392         uint64_t table_start, table_end;
 393         uint64_t start, end;
 394
 395         buf_start = bio->bio_offset;
 396         buf_len = bp->b_bcount;
 397
 398         tbl = NULL;
 399
 400         table_end = 0;
 401         issued_len = 0;
 402
 403         dmv = dev->si_drv1;
 404
 405         switch(bp->b_cmd) {
 406         case BUF_CMD_READ:
 407         case BUF_CMD_WRITE:
 408         case BUF_CMD_FREEBLKS:
 409                 bypass = 0;
 410                 break;
 411         case BUF_CMD_FLUSH:
 412                 bypass = 1;
 413                 KKASSERT(buf_len == 0);
 414                 break;
 415         default:
 416                 bp->b_error = EIO;
 417                 bp->b_resid = bp->b_bcount;
 418                 biodone(bio);
 419                 return 0;
 420         }
 421
 422         if (bypass == 0 &&
 423             bounds_check_with_mediasize(bio, DEV_BSIZE,
 424                                         dm_table_size(&dmv->table_head)) <= 0) {
 425                 bp->b_resid = bp->b_bcount;
 426                 biodone(bio);
 427                 return 0;
 428         }
 429
 430         /* Select active table */
 431         tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
 432
 433         nestiobuf_init(bio);
 434         devstat_start_transaction(&dmv->stats);
 435
 436         /*
 437          * Find out what tables I want to select.
 438          */
 439         SLIST_FOREACH(table_en, tbl, next) {
 440                 /*
 441                  * I need need number of bytes not blocks.
 442                  */
 443                 table_start = table_en->start * DEV_BSIZE;
 444                 table_end = table_start + (table_en->length) * DEV_BSIZE;
 445
 446                 /*
 447                  * Calculate the start and end
 448                  */
 449                 start = MAX(table_start, buf_start);
 450                 end = MIN(table_end, buf_start + buf_len);
 451
 452                 aprint_debug("----------------------------------------\n");
 453                 aprint_debug("table_start %010" PRIu64", table_end %010"
 454                     PRIu64 "\n", table_start, table_end);
 455                 aprint_debug("buf_start %010" PRIu64", buf_len %010"
 456                     PRIu64"\n", buf_start, buf_len);
 457                 aprint_debug("start-buf_start %010"PRIu64", end %010"
 458                     PRIu64"\n", start - buf_start, end);
 459                 aprint_debug("start %010" PRIu64" , end %010"
 460                     PRIu64"\n", start, end);
 461                 aprint_debug("\n----------------------------------------\n");
 462
 463                 if (bypass) {
 464                         nestbuf = getpbuf(NULL);
 465                         nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
 466
 467                         nestiobuf_add(bio, nestbuf, 0, 0, &dmv->stats);
 468                         nestbuf->b_bio1.bio_offset = 0;
 469                         table_en->target->strategy(table_en, nestbuf);
 470                 } else if (start < end) {
 471                         nestbuf = getpbuf(NULL);
 472                         nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
 473
 474                         nestiobuf_add(bio, nestbuf,
 475                                       start - buf_start, (end - start),
 476                                       &dmv->stats);
 477                         issued_len += end - start;
 478
 479                         nestbuf->b_bio1.bio_offset = (start - table_start);
 480                         table_en->target->strategy(table_en, nestbuf);
 481                 }
 482         }
 483
 484         if (issued_len < buf_len)
 485                 nestiobuf_error(bio, EINVAL);
 486         nestiobuf_start(bio);
 487         dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
 488
 489         return 0;
 490 }
 491
 492 static int
 493 dmdump(struct dev_dump_args *ap)
 494 {
 495         cdev_t dev = ap->a_head.a_dev;
 496         dm_dev_t *dmv;
 497         dm_table_t  *tbl;
 498         dm_table_entry_t *table_en;
 499         uint64_t buf_start, buf_len, issued_len;
 500         uint64_t table_start, table_end;
 501         uint64_t start, end, data_offset;
 502         off_t offset;
 503         size_t length;
 504         int error = 0;
 505
 506         buf_start = ap->a_offset;
 507         buf_len = ap->a_length;
 508
 509         tbl = NULL;
 510
 511         table_end = 0;
 512         issued_len = 0;
 513
 514         dmv = dev->si_drv1;
 515
 516         /* Select active table */
 517         tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
 518
 519
 520         /*
 521          * Find out what tables I want to select.
 522          */
 523         SLIST_FOREACH(table_en, tbl, next) {
 524                 /*
 525                  * I need need number of bytes not blocks.
 526                  */
 527                 table_start = table_en->start * DEV_BSIZE;
 528                 table_end = table_start + (table_en->length) * DEV_BSIZE;
 529
 530                 /*
 531                  * Calculate the start and end
 532                  */
 533                 start = MAX(table_start, buf_start);
 534                 end = MIN(table_end, buf_start + buf_len);
 535
 536                 if (ap->a_length == 0) {
 537                         if (table_en->target->dump == NULL) {
 538                                 error = ENXIO;
 539                                 goto out;
 540                         }
 541
 542                         table_en->target->dump(table_en, NULL, 0, 0);
 543                 } else if (start < end) {
 544                         data_offset = start - buf_start;
 545                         offset = start - table_start;
 546                         length = end - start;
 547
 548                         if (table_en->target->dump == NULL) {
 549                                 error = ENXIO;
 550                                 goto out;
 551                         }
 552
 553                         table_en->target->dump(table_en,
 554                             (char *)ap->a_virtual + data_offset,
 555                             length, offset);
 556
 557                         issued_len += end - start;
 558                 }
 559         }
 560
 561         if (issued_len < buf_len)
 562                 error = EINVAL;
 563
 564 out:
 565         dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
 566
 567         return error;
 568 }
 569
 570 static int
 571 dmsize(struct dev_psize_args *ap)
 572 {
 573         cdev_t dev = ap->a_head.a_dev;
 574         dm_dev_t *dmv;
 575         uint64_t size;
 576
 577         size = 0;
 578
 579         if ((dmv = dev->si_drv1) == NULL)
 580                 return ENXIO;
 581
 582         size = dm_table_size(&dmv->table_head);
 583         ap->a_result = (int64_t)size;
 584
 585         return 0;
 586 }
 587
 588 #if 0
 589 static void
 590 dmminphys(struct buf *bp)
 591 {
 592
 593         bp->b_bcount = MIN(bp->b_bcount, MAXPHYS);
 594 }
 595 #endif
 596
 597 void
 598 dmsetdiskinfo(struct disk *disk, dm_table_head_t *head)
 599 {
 600         struct disk_info info;
 601         uint64_t dmp_size;
 602
 603         dmp_size = dm_table_size(head);
 604
 605         bzero(&info, sizeof(struct disk_info));
 606         info.d_media_blksize = DEV_BSIZE;
 607         info.d_media_blocks = dmp_size;
 608 #if 0
 609         /* this is set by disk_setdiskinfo */
 610         info.d_media_size = dmp_size * DEV_BSIZE;
 611 #endif
 612         info.d_dsflags = DSO_MBRQUIET | DSO_DEVICEMAPPER | DSO_RAWPSIZE;
 613
 614         info.d_secpertrack = 32;
 615         info.d_nheads = 64;
 616         info.d_secpercyl = info.d_secpertrack * info.d_nheads;
 617         info.d_ncylinders = dmp_size / info.d_secpercyl;
 618
 619         disk_setdiskinfo(disk, &info);
 620 }
 621
 622 /*
 623  * Transform char s to uint64_t offset number.
 624  */
 625 uint64_t
 626 atoi64(const char *s)
 627 {
 628         uint64_t n;
 629         n = 0;
 630
 631         while (*s != '\0') {
 632                 if (!isdigit(*s))
 633                         break;
 634
 635                 n = (10 * n) + (*s - '0');
 636                 s++;
 637         }
 638
 639         return n;
 640 }
 641
 642 void
 643 dm_builtin_init(void *arg)
 644 {
 645         modeventhand_t evh = (modeventhand_t)arg;
 646
 647         KKASSERT(evh != NULL);
 648         evh(NULL, MOD_LOAD, NULL);
 649 }
 650
 651 void
 652 dm_builtin_uninit(void *arg)
 653 {
 654         modeventhand_t evh = (modeventhand_t)arg;
 655
 656         KKASSERT(evh != NULL);
 657         evh(NULL, MOD_UNLOAD, NULL);
 658 }
 659
 660 TUNABLE_INT("debug.dm_debug", &dm_debug_level);
 661 SYSCTL_INT(_debug, OID_AUTO, dm_debug, CTLFLAG_RW, &dm_debug_level,
 662                0, "Eanble device mapper debugging");
 663