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/ioccom.h>
  48 #include <sys/malloc.h>
  49 #include <sys/module.h>
  50 #include <sys/sysctl.h>
  51 #include <dev/disk/dm/dm.h>
  52
  53 #include "netbsd-dm.h"
  54
  55 static  d_ioctl_t       dmioctl;
  56 static  d_open_t        dmopen;
  57 static  d_close_t       dmclose;
  58 static  d_psize_t       dmsize;
  59 static  d_strategy_t    dmstrategy;
  60 static  d_dump_t        dmdump;
  61
  62 /* attach and detach routines */
  63 void dmattach(int);
  64 static int dm_modcmd(module_t mod, int cmd, void *unused);
  65 static int dmdestroy(void);
  66
  67 static void dm_doinit(void);
  68
  69 static int dm_cmd_to_fun(prop_dictionary_t);
  70 static int disk_ioctl_switch(cdev_t, u_long, void *);
  71 static int dm_ioctl_switch(u_long);
  72 #if 0
  73 static void dmminphys(struct buf *);
  74 #endif
  75
  76 /* ***Variable-definitions*** */
  77 struct dev_ops dm_ops = {
  78         { "dm", 0, D_DISK | D_MPSAFE },
  79         .d_open         = dmopen,
  80         .d_close        = dmclose,
  81         .d_read         = physread,
  82         .d_write        = physwrite,
  83         .d_ioctl        = dmioctl,
  84         .d_strategy     = dmstrategy,
  85         .d_psize        = dmsize,
  86         .d_dump         = dmdump,
  87 /* D_DISK */
  88 };
  89
  90 MALLOC_DEFINE(M_DM, "dm", "Device Mapper allocations");
  91
  92 int dm_debug_level = 0;
  93
  94 extern uint64_t dm_dev_counter;
  95
  96 static cdev_t dmcdev;
  97
  98 static moduledata_t dm_mod = {
  99     "dm",
 100     dm_modcmd,
 101     NULL
 102 };
 103 DECLARE_MODULE(dm, dm_mod, SI_SUB_RAID, SI_ORDER_ANY);
 104 MODULE_VERSION(dm, 1);
 105
 106 /*
 107  * This array is used to translate cmd to function pointer.
 108  *
 109  * Interface between libdevmapper and lvm2tools uses different
 110  * names for one IOCTL call because libdevmapper do another thing
 111  * then. When I run "info" or "mknodes" libdevmapper will send same
 112  * ioctl to kernel but will do another things in userspace.
 113  *
 114  */
 115 static struct cmd_function cmd_fn[] = {
 116                 { .cmd = "version", .fn = dm_get_version_ioctl},
 117                 { .cmd = "targets", .fn = dm_list_versions_ioctl},
 118                 { .cmd = "create",  .fn = dm_dev_create_ioctl},
 119                 { .cmd = "info",    .fn = dm_dev_status_ioctl},
 120                 { .cmd = "mknodes", .fn = dm_dev_status_ioctl},
 121                 { .cmd = "names",   .fn = dm_dev_list_ioctl},
 122                 { .cmd = "suspend", .fn = dm_dev_suspend_ioctl},
 123                 { .cmd = "remove",  .fn = dm_dev_remove_ioctl},
 124                 { .cmd = "remove_all", .fn = dm_dev_remove_all_ioctl},
 125                 { .cmd = "rename",  .fn = dm_dev_rename_ioctl},
 126                 { .cmd = "resume",  .fn = dm_dev_resume_ioctl},
 127                 { .cmd = "clear",   .fn = dm_table_clear_ioctl},
 128                 { .cmd = "deps",    .fn = dm_table_deps_ioctl},
 129                 { .cmd = "reload",  .fn = dm_table_load_ioctl},
 130                 { .cmd = "status",  .fn = dm_table_status_ioctl},
 131                 { .cmd = "table",   .fn = dm_table_status_ioctl},
 132                 { .cmd = "message", .fn = dm_message_ioctl},
 133                 {NULL, NULL}
 134 };
 135
 136 /* New module handle routine */
 137 static int
 138 dm_modcmd(module_t mod, int cmd, void *unused)
 139 {
 140         int error, bmajor, cmajor;
 141
 142         error = 0;
 143         bmajor = -1;
 144         cmajor = -1;
 145
 146         switch (cmd) {
 147         case MOD_LOAD:
 148                 dm_doinit();
 149                 kprintf("Device Mapper version %d.%d.%d loaded\n",
 150                     DM_VERSION_MAJOR, DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL);
 151                 break;
 152
 153         case MOD_UNLOAD:
 154                 /*
 155                  * Disable unloading of dm module if there are any devices
 156                  * defined in driver. This is probably too strong we need
 157                  * to disable auto-unload only if there is mounted dm device
 158                  * present.
 159                  */
 160                 if (dm_dev_counter > 0)
 161                         return EBUSY;
 162
 163                 error = dmdestroy();
 164                 if (error)
 165                         break;
 166                 kprintf("Device Mapper unloaded\n");
 167                 break;
 168
 169         default:
 170                 break;
 171         }
 172
 173         return error;
 174 }
 175
 176 static void
 177 dm_doinit(void)
 178 {
 179         dm_target_init();
 180         dm_dev_init();
 181         dm_pdev_init();
 182         dmcdev = make_dev(&dm_ops, 0, UID_ROOT, GID_OPERATOR, 0640, "mapper/control");
 183 }
 184
 185 /* Destroy routine */
 186 static int
 187 dmdestroy(void)
 188 {
 189         destroy_dev(dmcdev);
 190
 191         dm_dev_uninit();
 192         dm_pdev_uninit();
 193         dm_target_uninit();
 194
 195         return 0;
 196 }
 197
 198 static int
 199 dmopen(struct dev_open_args *ap)
 200 {
 201         cdev_t dev = ap->a_head.a_dev;
 202         dm_dev_t *dmv;
 203
 204         /* Shortcut for the control device */
 205         if (minor(dev) == 0)
 206                 return 0;
 207
 208         if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
 209                 return ENXIO;
 210
 211         dmv->is_open = 1;
 212         dm_dev_unbusy(dmv);
 213
 214         aprint_debug("dm open routine called %" PRIu32 "\n",
 215             minor(ap->a_head.a_dev));
 216         return 0;
 217 }
 218
 219 static int
 220 dmclose(struct dev_close_args *ap)
 221 {
 222         cdev_t dev = ap->a_head.a_dev;
 223         dm_dev_t *dmv;
 224
 225         /* Shortcut for the control device */
 226         if (minor(dev) == 0)
 227                 return 0;
 228
 229         if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
 230                 return ENXIO;
 231
 232         dmv->is_open = 0;
 233         dm_dev_unbusy(dmv);
 234
 235         aprint_debug("dm close routine called %" PRIu32 "\n",
 236             minor(ap->a_head.a_dev));
 237         return 0;
 238 }
 239
 240
 241 static int
 242 dmioctl(struct dev_ioctl_args *ap)
 243 {
 244         cdev_t dev = ap->a_head.a_dev;
 245         u_long cmd = ap->a_cmd;
 246         void *data = ap->a_data;
 247
 248         int r, err;
 249         prop_dictionary_t dm_dict_in;
 250
 251         err = r = 0;
 252
 253         aprint_debug("dmioctl called\n");
 254
 255         KKASSERT(data != NULL);
 256
 257         if (( r = disk_ioctl_switch(dev, cmd, data)) == ENOTTY) {
 258                 struct plistref *pref = (struct plistref *) data;
 259
 260                 /* Check if we were called with NETBSD_DM_IOCTL ioctl
 261                    otherwise quit. */
 262                 if ((r = dm_ioctl_switch(cmd)) != 0)
 263                         return r;
 264
 265                 if((r = prop_dictionary_copyin_ioctl(pref, cmd, &dm_dict_in)) != 0)
 266                         return r;
 267
 268                 if ((r = dm_check_version(dm_dict_in)) != 0)
 269                         goto cleanup_exit;
 270
 271                 /* run ioctl routine */
 272                 if ((err = dm_cmd_to_fun(dm_dict_in)) != 0)
 273                         goto cleanup_exit;
 274
 275 cleanup_exit:
 276                 r = prop_dictionary_copyout_ioctl(pref, cmd, dm_dict_in);
 277                 prop_object_release(dm_dict_in);
 278         }
 279
 280         /*
 281          * Return the error of the actual command if one one has
 282          * happened. Otherwise return 'r' which indicates errors
 283          * that occurred during helper operations.
 284          */
 285         return (err != 0)?err:r;
 286 }
 287
 288 /*
 289  * Translate command sent from libdevmapper to func.
 290  */
 291 static int
 292 dm_cmd_to_fun(prop_dictionary_t dm_dict){
 293         int i, r;
 294         prop_string_t command;
 295
 296         r = 0;
 297
 298         if ((command = prop_dictionary_get(dm_dict, DM_IOCTL_COMMAND)) == NULL)
 299                 return EINVAL;
 300
 301         for(i = 0; cmd_fn[i].cmd != NULL; i++)
 302                 if (prop_string_equals_cstring(command, cmd_fn[i].cmd))
 303                         break;
 304
 305         if (cmd_fn[i].cmd == NULL)
 306                 return EINVAL;
 307
 308         aprint_debug("ioctl %s called\n", cmd_fn[i].cmd);
 309         r = cmd_fn[i].fn(dm_dict);
 310
 311         return r;
 312 }
 313
 314 /* Call apropriate ioctl handler function. */
 315 static int
 316 dm_ioctl_switch(u_long cmd)
 317 {
 318
 319         switch(cmd) {
 320
 321         case NETBSD_DM_IOCTL:
 322                 aprint_debug("dm NetBSD_DM_IOCTL called\n");
 323                 break;
 324         default:
 325                  aprint_debug("dm unknown ioctl called\n");
 326                  return ENOTTY;
 327                  break; /* NOT REACHED */
 328         }
 329
 330          return 0;
 331 }
 332
 333  /*
 334   * Check for disk specific ioctls.
 335   */
 336
 337 static int
 338 disk_ioctl_switch(cdev_t dev, u_long cmd, void *data)
 339 {
 340         dm_dev_t *dmv;
 341
 342         /* disk ioctls make sense only on block devices */
 343         if (minor(dev) == 0)
 344                 return ENOTTY;
 345
 346         switch(cmd) {
 347         case DIOCGPART:
 348         {
 349                 struct partinfo *dpart;
 350                 u_int64_t size;
 351                 dpart = (void *)data;
 352                 bzero(dpart, sizeof(*dpart));
 353
 354                 if ((dmv = dev->si_drv1) == NULL)
 355                         return ENODEV;
 356                 if (dmv->diskp->d_info.d_media_blksize == 0) {
 357                         return ENOTSUP;
 358                 } else {
 359                         size = dm_table_size(&dmv->table_head);
 360                         dpart->media_offset  = 0;
 361                         dpart->media_size    = size * DEV_BSIZE;
 362                         dpart->media_blocks  = size;
 363                         dpart->media_blksize = DEV_BSIZE;
 364                         dpart->fstype = FS_BSDFFS;
 365                 }
 366                 break;
 367         }
 368
 369         default:
 370                 aprint_debug("unknown disk_ioctl called\n");
 371                 return ENOTTY;
 372                 break; /* NOT REACHED */
 373         }
 374
 375         return 0;
 376 }
 377
 378 /*
 379  * Do all IO operations on dm logical devices.
 380  */
 381 static int
 382 dmstrategy(struct dev_strategy_args *ap)
 383 {
 384         cdev_t dev = ap->a_head.a_dev;
 385         struct bio *bio = ap->a_bio;
 386         struct buf *bp = bio->bio_buf;
 387         int bypass;
 388
 389         dm_dev_t *dmv;
 390         dm_table_t  *tbl;
 391         dm_table_entry_t *table_en;
 392         struct buf *nestbuf;
 393
 394         uint32_t dev_type;
 395
 396         uint64_t buf_start, buf_len, issued_len;
 397         uint64_t table_start, table_end;
 398         uint64_t start, end;
 399
 400         buf_start = bio->bio_offset;
 401         buf_len = bp->b_bcount;
 402
 403         tbl = NULL;
 404
 405         table_end = 0;
 406         dev_type = 0;
 407         issued_len = 0;
 408
 409         dmv = dev->si_drv1;
 410
 411         switch(bp->b_cmd) {
 412         case BUF_CMD_READ:
 413         case BUF_CMD_WRITE:
 414         case BUF_CMD_FREEBLKS:
 415                 bypass = 0;
 416                 break;
 417         case BUF_CMD_FLUSH:
 418                 bypass = 1;
 419                 KKASSERT(buf_len == 0);
 420                 break;
 421         default:
 422                 bp->b_error = EIO;
 423                 bp->b_resid = bp->b_bcount;
 424                 biodone(bio);
 425                 return 0;
 426         }
 427
 428         if (bypass == 0 &&
 429             bounds_check_with_mediasize(bio, DEV_BSIZE,
 430                                         dm_table_size(&dmv->table_head)) <= 0) {
 431                 bp->b_resid = bp->b_bcount;
 432                 biodone(bio);
 433                 return 0;
 434         }
 435
 436         /* Select active table */
 437         tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
 438
 439         nestiobuf_init(bio);
 440         devstat_start_transaction(&dmv->stats);
 441
 442         /*
 443          * Find out what tables I want to select.
 444          */
 445         SLIST_FOREACH(table_en, tbl, next) {
 446                 /*
 447                  * I need need number of bytes not blocks.
 448                  */
 449                 table_start = table_en->start * DEV_BSIZE;
 450                 table_end = table_start + (table_en->length) * DEV_BSIZE;
 451
 452                 /*
 453                  * Calculate the start and end
 454                  */
 455                 start = MAX(table_start, buf_start);
 456                 end = MIN(table_end, buf_start + buf_len);
 457
 458                 aprint_debug("----------------------------------------\n");
 459                 aprint_debug("table_start %010" PRIu64", table_end %010"
 460                     PRIu64 "\n", table_start, table_end);
 461                 aprint_debug("buf_start %010" PRIu64", buf_len %010"
 462                     PRIu64"\n", buf_start, buf_len);
 463                 aprint_debug("start-buf_start %010"PRIu64", end %010"
 464                     PRIu64"\n", start - buf_start, end);
 465                 aprint_debug("start %010" PRIu64" , end %010"
 466                     PRIu64"\n", start, end);
 467                 aprint_debug("\n----------------------------------------\n");
 468
 469                 if (bypass) {
 470                         nestbuf = getpbuf(NULL);
 471                         nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
 472
 473                         nestiobuf_add(bio, nestbuf, 0, 0, &dmv->stats);
 474                         nestbuf->b_bio1.bio_offset = 0;
 475                         table_en->target->strategy(table_en, nestbuf);
 476                 } else if (start < end) {
 477                         nestbuf = getpbuf(NULL);
 478                         nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
 479
 480                         nestiobuf_add(bio, nestbuf,
 481                                       start - buf_start, (end - start),
 482                                       &dmv->stats);
 483                         issued_len += end - start;
 484
 485                         nestbuf->b_bio1.bio_offset = (start - table_start);
 486                         table_en->target->strategy(table_en, nestbuf);
 487                 }
 488         }
 489
 490         if (issued_len < buf_len)
 491                 nestiobuf_error(bio, EINVAL);
 492         nestiobuf_start(bio);
 493         dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
 494
 495         return 0;
 496 }
 497
 498 static int
 499 dmdump(struct dev_dump_args *ap)
 500 {
 501         cdev_t dev = ap->a_head.a_dev;
 502         dm_dev_t *dmv;
 503         dm_table_t  *tbl;
 504         dm_table_entry_t *table_en;
 505         uint32_t dev_type;
 506         uint64_t buf_start, buf_len, issued_len;
 507         uint64_t table_start, table_end;
 508         uint64_t start, end, data_offset;
 509         off_t offset;
 510         size_t length;
 511         int error = 0;
 512
 513         buf_start = ap->a_offset;
 514         buf_len = ap->a_length;
 515
 516         tbl = NULL;
 517
 518         table_end = 0;
 519         dev_type = 0;
 520         issued_len = 0;
 521
 522         dmv = dev->si_drv1;
 523
 524         /* Select active table */
 525         tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
 526
 527
 528         /*
 529          * Find out what tables I want to select.
 530          */
 531         SLIST_FOREACH(table_en, tbl, next) {
 532                 /*
 533                  * I need need number of bytes not blocks.
 534                  */
 535                 table_start = table_en->start * DEV_BSIZE;
 536                 table_end = table_start + (table_en->length) * DEV_BSIZE;
 537
 538                 /*
 539                  * Calculate the start and end
 540                  */
 541                 start = MAX(table_start, buf_start);
 542                 end = MIN(table_end, buf_start + buf_len);
 543
 544                 if (ap->a_length == 0) {
 545                         if (table_en->target->dump == NULL) {
 546                                 error = ENXIO;
 547                                 goto out;
 548                         }
 549
 550                         table_en->target->dump(table_en, NULL, 0, 0);
 551                 } else if (start < end) {
 552                         data_offset = start - buf_start;
 553                         offset = start - table_start;
 554                         length = end - start;
 555
 556                         if (table_en->target->dump == NULL) {
 557                                 error = ENXIO;
 558                                 goto out;
 559                         }
 560
 561                         table_en->target->dump(table_en,
 562                             (char *)ap->a_virtual + data_offset,
 563                             length, offset);
 564
 565                         issued_len += end - start;
 566                 }
 567         }
 568
 569         if (issued_len < buf_len)
 570                 error = EINVAL;
 571
 572 out:
 573         dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
 574
 575         return error;
 576 }
 577
 578 static int
 579 dmsize(struct dev_psize_args *ap)
 580 {
 581         cdev_t dev = ap->a_head.a_dev;
 582         dm_dev_t *dmv;
 583         uint64_t size;
 584
 585         size = 0;
 586
 587         if ((dmv = dev->si_drv1) == NULL)
 588                 return ENXIO;
 589
 590         size = dm_table_size(&dmv->table_head);
 591         ap->a_result = (int64_t)size;
 592
 593         return 0;
 594 }
 595
 596 #if 0
 597 static void
 598 dmminphys(struct buf *bp)
 599 {
 600
 601         bp->b_bcount = MIN(bp->b_bcount, MAXPHYS);
 602 }
 603 #endif
 604
 605 void
 606 dmsetdiskinfo(struct disk *disk, dm_table_head_t *head)
 607 {
 608         struct disk_info info;
 609         uint64_t dmp_size;
 610
 611         dmp_size = dm_table_size(head);
 612
 613         bzero(&info, sizeof(struct disk_info));
 614         info.d_media_blksize = DEV_BSIZE;
 615         info.d_media_blocks = dmp_size;
 616 #if 0
 617         /* this is set by disk_setdiskinfo */
 618         info.d_media_size = dmp_size * DEV_BSIZE;
 619 #endif
 620         info.d_dsflags = DSO_MBRQUIET | DSO_DEVICEMAPPER;
 621
 622         info.d_secpertrack = 32;
 623         info.d_nheads = 64;
 624         info.d_secpercyl = info.d_secpertrack * info.d_nheads;
 625         info.d_ncylinders = dmp_size / info.d_secpercyl;
 626
 627         disk_setdiskinfo(disk, &info);
 628 }
 629
 630 /*
 631  * Transform char s to uint64_t offset number.
 632  */
 633 uint64_t
 634 atoi64(const char *s)
 635 {
 636         uint64_t n;
 637         n = 0;
 638
 639         while (*s != '\0') {
 640                 if (!isdigit(*s))
 641                         break;
 642
 643                 n = (10 * n) + (*s - '0');
 644                 s++;
 645         }
 646
 647         return n;
 648 }
 649
 650 void
 651 dm_builtin_init(void *arg)
 652 {
 653         modeventhand_t evh = (modeventhand_t)arg;
 654
 655         KKASSERT(evh != NULL);
 656         evh(NULL, MOD_LOAD, NULL);
 657 }
 658
 659 void
 660 dm_builtin_uninit(void *arg)
 661 {
 662         modeventhand_t evh = (modeventhand_t)arg;
 663
 664         KKASSERT(evh != NULL);
 665         evh(NULL, MOD_UNLOAD, NULL);
 666 }
 667
 668 TUNABLE_INT("debug.dm_debug", &dm_debug_level);
 669 SYSCTL_INT(_debug, OID_AUTO, dm_debug, CTLFLAG_RW, &dm_debug_level,
 670                0, "Eanble device mapper debugging");
 671