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 The NetBSD Foundation, Inc.
   5  * All rights reserved.
   6  *
   7  * This code is derived from software contributed to The NetBSD Foundation
   8  * by Adam Hamsik.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
  20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
  21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
  23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  29  * POSSIBILITY OF SUCH DAMAGE.
  30  */
  31
  32 /*
  33  * I want to say thank you to all people who helped me with this project.
  34  */
  35
  36 #include <sys/types.h>
  37 #include <sys/param.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/ioccom.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, bmajor, cmajor;
 140
 141         error = 0;
 142         bmajor = -1;
 143         cmajor = -1;
 144
 145         switch (cmd) {
 146         case MOD_LOAD:
 147                 dm_doinit();
 148                 kprintf("Device Mapper version %d.%d.%d loaded\n",
 149                     DM_VERSION_MAJOR, DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL);
 150                 break;
 151
 152         case MOD_UNLOAD:
 153                 /*
 154                  * Disable unloading of dm module if there are any devices
 155                  * defined in driver. This is probably too strong we need
 156                  * to disable auto-unload only if there is mounted dm device
 157                  * present.
 158                  */
 159                 if (dm_dev_counter > 0)
 160                         return EBUSY;
 161
 162                 error = dmdestroy();
 163                 if (error)
 164                         break;
 165                 kprintf("Device Mapper unloaded\n");
 166                 break;
 167
 168         default:
 169                 break;
 170         }
 171
 172         return error;
 173 }
 174
 175 static void
 176 dm_doinit(void)
 177 {
 178         dm_target_init();
 179         dm_dev_init();
 180         dm_pdev_init();
 181         dmcdev = make_dev(&dm_ops, 0, UID_ROOT, GID_OPERATOR, 0640, "mapper/control");
 182 }
 183
 184 /* Destroy routine */
 185 static int
 186 dmdestroy(void)
 187 {
 188         destroy_dev(dmcdev);
 189
 190         dm_dev_uninit();
 191         dm_pdev_uninit();
 192         dm_target_uninit();
 193
 194         return 0;
 195 }
 196
 197 static int
 198 dmopen(struct dev_open_args *ap)
 199 {
 200         cdev_t dev = ap->a_head.a_dev;
 201         dm_dev_t *dmv;
 202
 203         /* Shortcut for the control device */
 204         if (minor(dev) == 0)
 205                 return 0;
 206
 207         if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
 208                 return ENXIO;
 209
 210         dmv->is_open = 1;
 211         dm_dev_unbusy(dmv);
 212
 213         aprint_debug("dm open routine called %" PRIu32 "\n",
 214             minor(ap->a_head.a_dev));
 215         return 0;
 216 }
 217
 218 static int
 219 dmclose(struct dev_close_args *ap)
 220 {
 221         cdev_t dev = ap->a_head.a_dev;
 222         dm_dev_t *dmv;
 223
 224         /* Shortcut for the control device */
 225         if (minor(dev) == 0)
 226                 return 0;
 227
 228         if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
 229                 return ENXIO;
 230
 231         dmv->is_open = 0;
 232         dm_dev_unbusy(dmv);
 233
 234         aprint_debug("dm close routine called %" PRIu32 "\n",
 235             minor(ap->a_head.a_dev));
 236         return 0;
 237 }
 238
 239
 240 static int
 241 dmioctl(struct dev_ioctl_args *ap)
 242 {
 243         cdev_t dev = ap->a_head.a_dev;
 244         u_long cmd = ap->a_cmd;
 245         void *data = ap->a_data;
 246
 247         int r, err;
 248         prop_dictionary_t dm_dict_in;
 249
 250         err = r = 0;
 251
 252         aprint_debug("dmioctl called\n");
 253
 254         KKASSERT(data != NULL);
 255
 256         if (( r = disk_ioctl_switch(dev, cmd, data)) == ENOTTY) {
 257                 struct plistref *pref = (struct plistref *) data;
 258
 259                 /* Check if we were called with NETBSD_DM_IOCTL ioctl
 260                    otherwise quit. */
 261                 if ((r = dm_ioctl_switch(cmd)) != 0)
 262                         return r;
 263
 264                 if((r = prop_dictionary_copyin_ioctl(pref, cmd, &dm_dict_in)) != 0)
 265                         return r;
 266
 267                 if ((r = dm_check_version(dm_dict_in)) != 0)
 268                         goto cleanup_exit;
 269
 270                 /* run ioctl routine */
 271                 if ((err = dm_cmd_to_fun(dm_dict_in)) != 0)
 272                         goto cleanup_exit;
 273
 274 cleanup_exit:
 275                 r = prop_dictionary_copyout_ioctl(pref, cmd, dm_dict_in);
 276                 prop_object_release(dm_dict_in);
 277         }
 278
 279         /*
 280          * Return the error of the actual command if one one has
 281          * happened. Otherwise return 'r' which indicates errors
 282          * that occurred during helper operations.
 283          */
 284         return (err != 0)?err:r;
 285 }
 286
 287 /*
 288  * Translate command sent from libdevmapper to func.
 289  */
 290 static int
 291 dm_cmd_to_fun(prop_dictionary_t dm_dict){
 292         int i, r;
 293         prop_string_t command;
 294
 295         r = 0;
 296
 297         if ((command = prop_dictionary_get(dm_dict, DM_IOCTL_COMMAND)) == NULL)
 298                 return EINVAL;
 299
 300         for(i = 0; cmd_fn[i].cmd != NULL; i++)
 301                 if (prop_string_equals_cstring(command, cmd_fn[i].cmd))
 302                         break;
 303
 304         if (cmd_fn[i].cmd == NULL)
 305                 return EINVAL;
 306
 307         aprint_debug("ioctl %s called\n", cmd_fn[i].cmd);
 308         r = cmd_fn[i].fn(dm_dict);
 309
 310         return r;
 311 }
 312
 313 /* Call apropriate ioctl handler function. */
 314 static int
 315 dm_ioctl_switch(u_long cmd)
 316 {
 317
 318         switch(cmd) {
 319
 320         case NETBSD_DM_IOCTL:
 321                 aprint_debug("dm NetBSD_DM_IOCTL called\n");
 322                 break;
 323         default:
 324                  aprint_debug("dm unknown ioctl called\n");
 325                  return ENOTTY;
 326                  break; /* NOT REACHED */
 327         }
 328
 329          return 0;
 330 }
 331
 332  /*
 333   * Check for disk specific ioctls.
 334   */
 335
 336 static int
 337 disk_ioctl_switch(cdev_t dev, u_long cmd, void *data)
 338 {
 339         dm_dev_t *dmv;
 340
 341         /* disk ioctls make sense only on block devices */
 342         if (minor(dev) == 0)
 343                 return ENOTTY;
 344
 345         switch(cmd) {
 346         case DIOCGPART:
 347         {
 348                 struct partinfo *dpart;
 349                 u_int64_t size;
 350                 dpart = (void *)data;
 351                 bzero(dpart, sizeof(*dpart));
 352
 353                 if ((dmv = dev->si_drv1) == NULL)
 354                         return ENODEV;
 355                 if (dmv->diskp->d_info.d_media_blksize == 0) {
 356                         return ENOTSUP;
 357                 } else {
 358                         size = dm_table_size(&dmv->table_head);
 359                         dpart->media_offset  = 0;
 360                         dpart->media_size    = size * DEV_BSIZE;
 361                         dpart->media_blocks  = size;
 362                         dpart->media_blksize = DEV_BSIZE;
 363                         dpart->fstype = FS_BSDFFS;
 364                 }
 365                 break;
 366         }
 367
 368         default:
 369                 aprint_debug("unknown disk_ioctl called\n");
 370                 return ENOTTY;
 371                 break; /* NOT REACHED */
 372         }
 373
 374         return 0;
 375 }
 376
 377 /*
 378  * Do all IO operations on dm logical devices.
 379  */
 380 static int
 381 dmstrategy(struct dev_strategy_args *ap)
 382 {
 383         cdev_t dev = ap->a_head.a_dev;
 384         struct bio *bio = ap->a_bio;
 385         struct buf *bp = bio->bio_buf;
 386         int bypass;
 387
 388         dm_dev_t *dmv;
 389         dm_table_t  *tbl;
 390         dm_table_entry_t *table_en;
 391         struct buf *nestbuf;
 392
 393         uint32_t dev_type;
 394
 395         uint64_t buf_start, buf_len, issued_len;
 396         uint64_t table_start, table_end;
 397         uint64_t start, end;
 398
 399         buf_start = bio->bio_offset;
 400         buf_len = bp->b_bcount;
 401
 402         tbl = NULL;
 403
 404         table_end = 0;
 405         dev_type = 0;
 406         issued_len = 0;
 407
 408         dmv = dev->si_drv1;
 409
 410         switch(bp->b_cmd) {
 411         case BUF_CMD_READ:
 412         case BUF_CMD_WRITE:
 413         case BUF_CMD_FREEBLKS:
 414                 bypass = 0;
 415                 break;
 416         case BUF_CMD_FLUSH:
 417                 bypass = 1;
 418                 KKASSERT(buf_len == 0);
 419                 break;
 420         default:
 421                 bp->b_error = EIO;
 422                 bp->b_resid = bp->b_bcount;
 423                 biodone(bio);
 424                 return 0;
 425         }
 426
 427         if (bypass == 0 &&
 428             bounds_check_with_mediasize(bio, DEV_BSIZE,
 429                                         dm_table_size(&dmv->table_head)) <= 0) {
 430                 bp->b_resid = bp->b_bcount;
 431                 biodone(bio);
 432                 return 0;
 433         }
 434
 435         /* Select active table */
 436         tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
 437
 438         nestiobuf_init(bio);
 439         devstat_start_transaction(&dmv->stats);
 440
 441         /*
 442          * Find out what tables I want to select.
 443          */
 444         SLIST_FOREACH(table_en, tbl, next) {
 445                 /*
 446                  * I need need number of bytes not blocks.
 447                  */
 448                 table_start = table_en->start * DEV_BSIZE;
 449                 table_end = table_start + (table_en->length) * DEV_BSIZE;
 450
 451                 /*
 452                  * Calculate the start and end
 453                  */
 454                 start = MAX(table_start, buf_start);
 455                 end = MIN(table_end, buf_start + buf_len);
 456
 457                 aprint_debug("----------------------------------------\n");
 458                 aprint_debug("table_start %010" PRIu64", table_end %010"
 459                     PRIu64 "\n", table_start, table_end);
 460                 aprint_debug("buf_start %010" PRIu64", buf_len %010"
 461                     PRIu64"\n", buf_start, buf_len);
 462                 aprint_debug("start-buf_start %010"PRIu64", end %010"
 463                     PRIu64"\n", start - buf_start, end);
 464                 aprint_debug("start %010" PRIu64" , end %010"
 465                     PRIu64"\n", start, end);
 466                 aprint_debug("\n----------------------------------------\n");
 467
 468                 if (bypass) {
 469                         nestbuf = getpbuf(NULL);
 470                         nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
 471
 472                         nestiobuf_add(bio, nestbuf, 0, 0, &dmv->stats);
 473                         nestbuf->b_bio1.bio_offset = 0;
 474                         table_en->target->strategy(table_en, nestbuf);
 475                 } else if (start < end) {
 476                         nestbuf = getpbuf(NULL);
 477                         nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
 478
 479                         nestiobuf_add(bio, nestbuf,
 480                                       start - buf_start, (end - start),
 481                                       &dmv->stats);
 482                         issued_len += end - start;
 483
 484                         nestbuf->b_bio1.bio_offset = (start - table_start);
 485                         table_en->target->strategy(table_en, nestbuf);
 486                 }
 487         }
 488
 489         if (issued_len < buf_len)
 490                 nestiobuf_error(bio, EINVAL);
 491         nestiobuf_start(bio);
 492         dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
 493
 494         return 0;
 495 }
 496
 497 static int
 498 dmdump(struct dev_dump_args *ap)
 499 {
 500         cdev_t dev = ap->a_head.a_dev;
 501         dm_dev_t *dmv;
 502         dm_table_t  *tbl;
 503         dm_table_entry_t *table_en;
 504         uint32_t dev_type;
 505         uint64_t buf_start, buf_len, issued_len;
 506         uint64_t table_start, table_end;
 507         uint64_t start, end, data_offset;
 508         off_t offset;
 509         size_t length;
 510         int error = 0;
 511
 512         buf_start = ap->a_offset;
 513         buf_len = ap->a_length;
 514
 515         tbl = NULL;
 516
 517         table_end = 0;
 518         dev_type = 0;
 519         issued_len = 0;
 520
 521         dmv = dev->si_drv1;
 522
 523         /* Select active table */
 524         tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
 525
 526
 527         /*
 528          * Find out what tables I want to select.
 529          */
 530         SLIST_FOREACH(table_en, tbl, next) {
 531                 /*
 532                  * I need need number of bytes not blocks.
 533                  */
 534                 table_start = table_en->start * DEV_BSIZE;
 535                 table_end = table_start + (table_en->length) * DEV_BSIZE;
 536
 537                 /*
 538                  * Calculate the start and end
 539                  */
 540                 start = MAX(table_start, buf_start);
 541                 end = MIN(table_end, buf_start + buf_len);
 542
 543                 if (ap->a_length == 0) {
 544                         if (table_en->target->dump == NULL) {
 545                                 error = ENXIO;
 546                                 goto out;
 547                         }
 548
 549                         table_en->target->dump(table_en, NULL, 0, 0);
 550                 } else if (start < end) {
 551                         data_offset = start - buf_start;
 552                         offset = start - table_start;
 553                         length = end - start;
 554
 555                         if (table_en->target->dump == NULL) {
 556                                 error = ENXIO;
 557                                 goto out;
 558                         }
 559
 560                         table_en->target->dump(table_en,
 561                             (char *)ap->a_virtual + data_offset,
 562                             length, offset);
 563
 564                         issued_len += end - start;
 565                 }
 566         }
 567
 568         if (issued_len < buf_len)
 569                 error = EINVAL;
 570
 571 out:
 572         dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
 573
 574         return error;
 575 }
 576
 577 static int
 578 dmsize(struct dev_psize_args *ap)
 579 {
 580         cdev_t dev = ap->a_head.a_dev;
 581         dm_dev_t *dmv;
 582         uint64_t size;
 583
 584         size = 0;
 585
 586         if ((dmv = dev->si_drv1) == NULL)
 587                 return ENXIO;
 588
 589         size = dm_table_size(&dmv->table_head);
 590         ap->a_result = (int64_t)size;
 591
 592         return 0;
 593 }
 594
 595 #if 0
 596 static void
 597 dmminphys(struct buf *bp)
 598 {
 599
 600         bp->b_bcount = MIN(bp->b_bcount, MAXPHYS);
 601 }
 602 #endif
 603
 604 void
 605 dmsetdiskinfo(struct disk *disk, dm_table_head_t *head)
 606 {
 607         struct disk_info info;
 608         uint64_t dmp_size;
 609
 610         dmp_size = dm_table_size(head);
 611
 612         bzero(&info, sizeof(struct disk_info));
 613         info.d_media_blksize = DEV_BSIZE;
 614         info.d_media_blocks = dmp_size;
 615 #if 0
 616         /* this is set by disk_setdiskinfo */
 617         info.d_media_size = dmp_size * DEV_BSIZE;
 618 #endif
 619         info.d_dsflags = DSO_MBRQUIET | DSO_DEVICEMAPPER;
 620
 621         info.d_secpertrack = 32;
 622         info.d_nheads = 64;
 623         info.d_secpercyl = info.d_secpertrack * info.d_nheads;
 624         info.d_ncylinders = dmp_size / info.d_secpercyl;
 625
 626         disk_setdiskinfo(disk, &info);
 627 }
 628
 629 /*
 630  * Transform char s to uint64_t offset number.
 631  */
 632 uint64_t
 633 atoi64(const char *s)
 634 {
 635         uint64_t n;
 636         n = 0;
 637
 638         while (*s != '\0') {
 639                 if (!isdigit(*s))
 640                         break;
 641
 642                 n = (10 * n) + (*s - '0');
 643                 s++;
 644         }
 645
 646         return n;
 647 }
 648
 649 void
 650 dm_builtin_init(void *arg)
 651 {
 652         modeventhand_t evh = (modeventhand_t)arg;
 653
 654         KKASSERT(evh != NULL);
 655         evh(NULL, MOD_LOAD, NULL);
 656 }
 657
 658 void
 659 dm_builtin_uninit(void *arg)
 660 {
 661         modeventhand_t evh = (modeventhand_t)arg;
 662
 663         KKASSERT(evh != NULL);
 664         evh(NULL, MOD_UNLOAD, NULL);
 665 }
 666
 667 TUNABLE_INT("debug.dm_debug", &dm_debug_level);
 668 SYSCTL_INT(_debug, OID_AUTO, dm_debug, CTLFLAG_RW, &dm_debug_level,
 669                0, "Eanble device mapper debugging");
 670