1 /*- 2 * Copyright (c) 1997, 1998 3 * Nan Yang Computer Services Limited. All rights reserved. 4 * 5 * This software is distributed under the so-called ``Berkeley 6 * License'': 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by Nan Yang Computer 19 * Services Limited. 20 * 4. Neither the name of the Company nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * This software is provided ``as is'', and any express or implied 25 * warranties, including, but not limited to, the implied warranties of 26 * merchantability and fitness for a particular purpose are disclaimed. 27 * In no event shall the company or contributors be liable for any 28 * direct, indirect, incidental, special, exemplary, or consequential 29 * damages (including, but not limited to, procurement of substitute 30 * goods or services; loss of use, data, or profits; or business 31 * interruption) however caused and on any theory of liability, whether 32 * in contract, strict liability, or tort (including negligence or 33 * otherwise) arising in any way out of the use of this software, even if 34 * advised of the possibility of such damage. 35 * 36 * $Id: vinumio.c,v 1.30 2000/05/10 23:23:30 grog Exp grog $ 37 * $FreeBSD: src/sys/dev/vinum/vinumio.c,v 1.52.2.6 2002/05/02 08:43:44 grog Exp $ 38 * $DragonFly: src/sys/dev/raid/vinum/vinumio.c,v 1.10 2006/02/17 19:18:06 dillon Exp $ 39 */ 40 41 #include "vinumhdr.h" 42 #include "request.h" 43 #include <vm/vm_zone.h> 44 45 static char *sappend(char *txt, char *s); 46 static int drivecmp(const void *va, const void *vb); 47 48 /* 49 * Open the device associated with the drive, and set drive's vp. 50 * Return an error number 51 */ 52 int 53 open_drive(struct drive *drive, struct proc *p, int verbose) 54 { 55 int devmajor; /* major devs for disk device */ 56 int devminor; /* minor devs for disk device */ 57 int unit; 58 char *dname; 59 60 if (bcmp(drive->devicename, "/dev/", 5)) /* device name doesn't start with /dev */ 61 return ENOENT; /* give up */ 62 if (drive->flags & VF_OPEN) /* open already, */ 63 return EBUSY; /* don't do it again */ 64 65 /* 66 * Yes, Bruce, I know this is horrible, but we 67 * don't have a root file system when we first 68 * try to do this. If you can come up with a 69 * better solution, I'd really like it. I'm 70 * just putting it in now to add ammuntion to 71 * moving the system to devfs. 72 */ 73 dname = &drive->devicename[5]; 74 drive->dev = NULL; /* no device yet */ 75 76 /* Find the device */ 77 if (bcmp(dname, "ad", 2) == 0) /* IDE disk */ 78 devmajor = 116; 79 else if (bcmp(dname, "wd", 2) == 0) /* IDE disk */ 80 devmajor = 3; 81 else if (bcmp(dname, "da", 2) == 0) 82 devmajor = 13; 83 else if (bcmp(dname, "vn", 2) == 0) 84 devmajor = 43; 85 else if (bcmp(dname, "md", 2) == 0) 86 devmajor = 95; 87 else if (bcmp(dname, "amrd", 4) == 0) { 88 devmajor = 133; 89 dname += 2; 90 } else if (bcmp(dname, "mlxd", 4) == 0) { 91 devmajor = 131; 92 dname += 2; 93 } else if (bcmp(dname, "idad", 4) == 0) { 94 devmajor = 109; 95 dname += 2; 96 } else if (bcmp(dname, "twed", 4) == 0) { /* 3ware raid */ 97 devmajor = 147; 98 dname += 2; 99 } else if (bcmp(dname, "ar", 2) == 0) { 100 devmajor = 157; 101 } else 102 return ENODEV; 103 dname += 2; /* point past */ 104 105 /* 106 * Found the device. We can expect one of 107 * two formats for the rest: a unit number, 108 * then either a partition letter for the 109 * compatiblity partition (e.g. h) or a 110 * slice ID and partition (e.g. s2e). 111 * Create a minor number for each of them. 112 */ 113 unit = 0; 114 while ((*dname >= '0') /* unit number */ 115 &&(*dname <= '9')) { 116 unit = unit * 10 + *dname - '0'; 117 dname++; 118 } 119 120 if (*dname == 's') { /* slice */ 121 if (((dname[1] < '1') || (dname[1] > '4')) /* invalid slice */ 122 ||((dname[2] < 'a') || (dname[2] > 'p'))) /* or invalid partition */ 123 return ENODEV; 124 devminor = dkmakeminor(unit, dname[1] - '0' + 1, (dname[2] - 'a')); 125 } else { /* compatibility partition */ 126 if ((*dname < 'a') || (*dname > 'p')) /* or invalid partition */ 127 return ENODEV; 128 devminor = dkmakeminor(unit, 0, (dname[0] - 'a')); 129 } 130 131 /* 132 * Disallow partition c 133 */ 134 if ((((devminor >> 17) & 0x08) | (devminor & 7)) == 2) 135 return ENOTTY; /* not buying that */ 136 137 drive->dev = udev2dev(makeudev(devmajor, devminor), 0); 138 139 drive->dev->si_iosize_max = DFLTPHYS; 140 if (dev_is_good(drive->dev)) 141 drive->lasterror = dev_dopen(drive->dev, FWRITE, 0, NULL); 142 else 143 drive->lasterror = ENOENT; 144 145 if (drive->lasterror != 0) { /* failed */ 146 drive->state = drive_down; /* just force it down */ 147 if (verbose) 148 log(LOG_WARNING, 149 "vinum open_drive %s: failed with error %d\n", 150 drive->devicename, drive->lasterror); 151 } else 152 drive->flags |= VF_OPEN; /* we're open now */ 153 154 return drive->lasterror; 155 } 156 157 /* 158 * Set some variables in the drive struct 159 * in more convenient form. Return error indication 160 */ 161 int 162 set_drive_parms(struct drive *drive) 163 { 164 drive->blocksize = BLKDEV_IOSIZE; /* do we need this? */ 165 drive->secsperblock = drive->blocksize /* number of sectors per block */ 166 / drive->partinfo.disklab->d_secsize; 167 168 /* Now update the label part */ 169 bcopy(hostname, drive->label.sysname, VINUMHOSTNAMELEN); /* put in host name */ 170 getmicrotime(&drive->label.date_of_birth); /* and current time */ 171 drive->label.drive_size = ((u_int64_t) drive->partinfo.part->p_size) /* size of the drive in bytes */ 172 *((u_int64_t) drive->partinfo.disklab->d_secsize); 173 #if VINUMDEBUG 174 if (debug & DEBUG_BIGDRIVE) /* pretend we're 100 times as big */ 175 drive->label.drive_size *= 100; 176 #endif 177 178 /* number of sectors available for subdisks */ 179 drive->sectors_available = drive->label.drive_size / DEV_BSIZE - DATASTART; 180 181 /* 182 * Bug in 3.0 as of January 1998: you can open 183 * non-existent slices. They have a length of 0. 184 */ 185 if (drive->label.drive_size < MINVINUMSLICE) { /* too small to worry about */ 186 set_drive_state(drive->driveno, drive_down, setstate_force); 187 drive->lasterror = ENOSPC; 188 return ENOSPC; 189 } 190 drive->freelist_size = INITIAL_DRIVE_FREELIST; /* initial number of entries */ 191 drive->freelist = (struct drive_freelist *) 192 Malloc(INITIAL_DRIVE_FREELIST * sizeof(struct drive_freelist)); 193 if (drive->freelist == NULL) /* can't malloc, dammit */ 194 return ENOSPC; 195 drive->freelist_entries = 1; /* just (almost) the complete drive */ 196 drive->freelist[0].offset = DATASTART; /* starts here */ 197 drive->freelist[0].sectors = (drive->label.drive_size >> DEV_BSHIFT) - DATASTART; /* and it's this long */ 198 if (drive->label.name[0] != '\0') /* got a name */ 199 set_drive_state(drive->driveno, drive_up, setstate_force); /* our drive is accessible */ 200 else /* we know about it, but that's all */ 201 drive->state = drive_referenced; 202 return 0; 203 } 204 205 /* 206 * Initialize a drive: open the device and add device 207 * information 208 */ 209 int 210 init_drive(struct drive *drive, int verbose) 211 { 212 if (drive->devicename[0] != '/') { 213 drive->lasterror = EINVAL; 214 log(LOG_ERR, "vinum: Can't open drive without drive name\n"); 215 return EINVAL; 216 } 217 drive->lasterror = open_drive(drive, curproc, verbose); /* open the drive */ 218 if (drive->lasterror) 219 return drive->lasterror; 220 221 drive->lasterror = dev_dioctl( 222 drive->dev, 223 DIOCGPART, 224 (caddr_t) & drive->partinfo, 225 FREAD, 226 curthread); 227 if (drive->lasterror) { 228 if (verbose) 229 log(LOG_WARNING, 230 "vinum open_drive %s: Can't get partition information, drive->lasterror %d\n", 231 drive->devicename, 232 drive->lasterror); 233 close_drive(drive); 234 return drive->lasterror; 235 } 236 if (drive->partinfo.part->p_fstype != FS_VINUM) { /* not Vinum */ 237 drive->lasterror = EFTYPE; 238 if (verbose) 239 log(LOG_WARNING, 240 "vinum open_drive %s: Wrong partition type for vinum\n", 241 drive->devicename); 242 close_drive(drive); 243 return EFTYPE; 244 } 245 return set_drive_parms(drive); /* set various odds and ends */ 246 } 247 248 /* Close a drive if it's open. */ 249 void 250 close_drive(struct drive *drive) 251 { 252 LOCKDRIVE(drive); /* keep the daemon out */ 253 if (drive->flags & VF_OPEN) 254 close_locked_drive(drive); /* and close it */ 255 if (drive->state > drive_down) /* if it's up */ 256 drive->state = drive_down; /* make sure it's down */ 257 unlockdrive(drive); 258 } 259 260 /* 261 * Real drive close code, called with drive already locked. 262 * We have also checked that the drive is open. No errors. 263 */ 264 void 265 close_locked_drive(struct drive *drive) 266 { 267 /* 268 * If we can't access the drive, we can't flush 269 * the queues, which spec_close() will try to 270 * do. Get rid of them here first. 271 */ 272 drive->lasterror = dev_dclose(drive->dev, 0, 0, NULL); 273 drive->flags &= ~VF_OPEN; /* no longer open */ 274 } 275 276 /* 277 * Remove drive from the configuration. 278 * Caller must ensure that it isn't active. 279 */ 280 void 281 remove_drive(int driveno) 282 { 283 struct drive *drive = &vinum_conf.drive[driveno]; 284 struct vinum_hdr *vhdr; /* buffer for header */ 285 int error; 286 287 if (drive->state > drive_referenced) { /* real drive */ 288 if (drive->state == drive_up) { 289 vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* allocate buffer */ 290 CHECKALLOC(vhdr, "Can't allocate memory"); 291 error = read_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); 292 if (error) 293 drive->lasterror = error; 294 else { 295 vhdr->magic = VINUM_NOMAGIC; /* obliterate the magic, but leave the rest */ 296 write_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); 297 } 298 Free(vhdr); 299 } 300 free_drive(drive); /* close it and free resources */ 301 save_config(); /* and save the updated configuration */ 302 } 303 } 304 305 /* 306 * Transfer drive data. Usually called from one of these defines; 307 * #define read_drive(a, b, c, d) driveio (a, b, c, d, B_READ) 308 * #define write_drive(a, b, c, d) driveio (a, b, c, d, B_WRITE) 309 * 310 * length and offset are in bytes, but must be multiples of sector 311 * size. The function *does not check* for this condition, and 312 * truncates ruthlessly. 313 * Return error number 314 */ 315 int 316 driveio(struct drive *drive, char *buf, size_t length, off_t offset, int flag) 317 { 318 int error; 319 struct buf *bp; 320 321 error = 0; /* to keep the compiler happy */ 322 while (length) { /* divide into small enough blocks */ 323 int len = min(length, MAXBSIZE); /* maximum block device transfer is MAXBSIZE */ 324 325 bp = geteblk(len); /* get a buffer header */ 326 bp->b_flags = flag; 327 bp->b_bio1.bio_blkno = offset / drive->partinfo.disklab->d_secsize; /* block number */ 328 bp->b_saveaddr = bp->b_data; 329 bp->b_data = buf; 330 bp->b_bcount = len; 331 dev_dstrategy(drive->dev, &bp->b_bio1); 332 error = biowait(bp); 333 bp->b_data = bp->b_saveaddr; 334 bp->b_flags |= B_INVAL | B_AGE; 335 bp->b_flags &= ~B_ERROR; 336 brelse(bp); 337 if (error) 338 break; 339 length -= len; /* update pointers */ 340 buf += len; 341 offset += len; 342 } 343 return error; 344 } 345 346 /* 347 * Check a drive for a vinum header. If found, 348 * update the drive information. We come here 349 * with a partially populated drive structure 350 * which includes the device name. 351 * 352 * Return information on what we found. 353 * 354 * This function is called from two places: check_drive, 355 * which wants to find out whether the drive is a 356 * Vinum drive, and config_drive, which asserts that 357 * it is a vinum drive. In the first case, we don't 358 * print error messages (verbose==0), in the second 359 * we do (verbose==1). 360 */ 361 enum drive_label_info 362 read_drive_label(struct drive *drive, int verbose) 363 { 364 int error; 365 int result; /* result of our search */ 366 struct vinum_hdr *vhdr; /* and as header */ 367 368 error = init_drive(drive, 0); /* find the drive */ 369 if (error) /* find the drive */ 370 return DL_CANT_OPEN; /* not ours */ 371 372 vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* allocate buffers */ 373 CHECKALLOC(vhdr, "Can't allocate memory"); 374 375 drive->state = drive_up; /* be optimistic */ 376 error = read_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); 377 if (vhdr->magic == VINUM_MAGIC) { /* ours! */ 378 if (drive->label.name[0] /* we have a name for this drive */ 379 &&(strcmp(drive->label.name, vhdr->label.name))) { /* but it doesn't match the real name */ 380 drive->lasterror = EINVAL; 381 result = DL_WRONG_DRIVE; /* it's the wrong drive */ 382 drive->state = drive_unallocated; /* put it back, it's not ours */ 383 } else 384 result = DL_OURS; 385 /* 386 * We copy the drive anyway so that we have 387 * the correct name in the drive info. This 388 * may not be the name specified 389 */ 390 drive->label = vhdr->label; /* put in the label information */ 391 } else if (vhdr->magic == VINUM_NOMAGIC) /* was ours, but we gave it away */ 392 result = DL_DELETED_LABEL; /* and return the info */ 393 else 394 result = DL_NOT_OURS; /* we could have it, but we don't yet */ 395 Free(vhdr); /* that's all. */ 396 return result; 397 } 398 399 /* 400 * Check a drive for a vinum header. If found, 401 * read configuration information from the drive and 402 * incorporate the data into the configuration. 403 * 404 * Return drive number. 405 */ 406 struct drive * 407 check_drive(char *devicename) 408 { 409 int driveno; 410 int i; 411 struct drive *drive; 412 413 driveno = find_drive_by_dev(devicename, 1); /* if entry doesn't exist, create it */ 414 drive = &vinum_conf.drive[driveno]; /* and get a pointer */ 415 416 if (read_drive_label(drive, 0) == DL_OURS) { /* one of ours */ 417 for (i = 0; i < vinum_conf.drives_allocated; i++) { /* see if the name already exists */ 418 if ((i != driveno) /* not this drive */ 419 &&(DRIVE[i].state != drive_unallocated) /* and it's allocated */ 420 &&(strcmp(DRIVE[i].label.name, 421 DRIVE[driveno].label.name) == 0)) { /* and it has the same name */ 422 struct drive *mydrive = &DRIVE[i]; 423 424 if (mydrive->devicename[0] == '/') { /* we know a device name for it */ 425 /* 426 * set an error, but don't take the 427 * drive down: that would cause unneeded 428 * error messages. 429 */ 430 drive->lasterror = EEXIST; 431 break; 432 } else { /* it's just a place holder, */ 433 int sdno; 434 435 for (sdno = 0; sdno < vinum_conf.subdisks_allocated; sdno++) { /* look at each subdisk */ 436 if ((SD[sdno].driveno == i) /* it's pointing to this one, */ 437 &&(SD[sdno].state != sd_unallocated)) { /* and it's a real subdisk */ 438 SD[sdno].driveno = drive->driveno; /* point to the one we found */ 439 update_sd_state(sdno); /* and update its state */ 440 } 441 } 442 bzero(mydrive, sizeof(struct drive)); /* don't deallocate it, just remove it */ 443 } 444 } 445 } 446 } else { 447 if (drive->lasterror == 0) 448 drive->lasterror = ENODEV; 449 close_drive(drive); 450 drive->state = drive_down; 451 } 452 return drive; 453 } 454 455 static char * 456 sappend(char *txt, char *s) 457 { 458 while ((*s++ = *txt++) != 0); 459 return s - 1; 460 } 461 462 void 463 format_config(char *config, int len) 464 { 465 int i; 466 int j; 467 char *s = config; 468 char *configend = &config[len]; 469 470 bzero(config, len); 471 472 /* First write the volume configuration */ 473 for (i = 0; i < vinum_conf.volumes_allocated; i++) { 474 struct volume *vol; 475 476 vol = &vinum_conf.volume[i]; 477 if ((vol->state > volume_uninit) 478 && (vol->name[0] != '\0')) { /* paranoia */ 479 snprintf(s, 480 configend - s, 481 "volume %s state %s", 482 vol->name, 483 volume_state(vol->state)); 484 while (*s) 485 s++; /* find the end */ 486 if (vol->preferred_plex >= 0) /* preferences, */ 487 snprintf(s, 488 configend - s, 489 " readpol prefer %s", 490 vinum_conf.plex[vol->preferred_plex].name); 491 while (*s) 492 s++; /* find the end */ 493 s = sappend("\n", s); 494 } 495 } 496 497 /* Then the plex configuration */ 498 for (i = 0; i < vinum_conf.plexes_allocated; i++) { 499 struct plex *plex; 500 501 plex = &vinum_conf.plex[i]; 502 if ((plex->state > plex_referenced) 503 && (plex->name[0] != '\0')) { /* paranoia */ 504 snprintf(s, 505 configend - s, 506 "plex name %s state %s org %s ", 507 plex->name, 508 plex_state(plex->state), 509 plex_org(plex->organization)); 510 while (*s) 511 s++; /* find the end */ 512 if (isstriped(plex)) { 513 snprintf(s, 514 configend - s, 515 "%ds ", 516 (int) plex->stripesize); 517 while (*s) 518 s++; /* find the end */ 519 } 520 if (plex->volno >= 0) /* we have a volume */ 521 snprintf(s, 522 configend - s, 523 "vol %s ", 524 vinum_conf.volume[plex->volno].name); 525 while (*s) 526 s++; /* find the end */ 527 for (j = 0; j < plex->subdisks; j++) { 528 snprintf(s, 529 configend - s, 530 " sd %s", 531 vinum_conf.sd[plex->sdnos[j]].name); 532 } 533 s = sappend("\n", s); 534 } 535 } 536 537 /* And finally the subdisk configuration */ 538 for (i = 0; i < vinum_conf.subdisks_allocated; i++) { 539 struct sd *sd; 540 char *drivename; 541 542 sd = &SD[i]; 543 if ((sd->state != sd_referenced) 544 && (sd->state != sd_unallocated) 545 && (sd->name[0] != '\0')) { /* paranoia */ 546 drivename = vinum_conf.drive[sd->driveno].label.name; 547 /* 548 * XXX We've seen cases of dead subdisks 549 * which don't have a drive. If we let them 550 * through here, the drive name is null, so 551 * they get the drive named 'plex'. 552 * 553 * This is a breakage limiter, not a fix. 554 */ 555 if (drivename[0] == '\0') 556 drivename = "*invalid*"; 557 snprintf(s, 558 configend - s, 559 "sd name %s drive %s plex %s len %llus driveoffset %llus state %s", 560 sd->name, 561 drivename, 562 vinum_conf.plex[sd->plexno].name, 563 (unsigned long long) sd->sectors, 564 (unsigned long long) sd->driveoffset, 565 sd_state(sd->state)); 566 while (*s) 567 s++; /* find the end */ 568 if (sd->plexno >= 0) 569 snprintf(s, 570 configend - s, 571 " plexoffset %llds", 572 (long long) sd->plexoffset); 573 else 574 snprintf(s, configend - s, " detached"); 575 while (*s) 576 s++; /* find the end */ 577 if (sd->flags & VF_RETRYERRORS) { 578 snprintf(s, configend - s, " retryerrors"); 579 while (*s) 580 s++; /* find the end */ 581 } 582 snprintf(s, configend - s, " \n"); 583 while (*s) 584 s++; /* find the end */ 585 } 586 } 587 if (s > &config[len - 2]) 588 panic("vinum: configuration data overflow"); 589 } 590 591 /* 592 * issue a save config request to the d�mon. The actual work 593 * is done in process context by daemon_save_config 594 */ 595 void 596 save_config(void) 597 { 598 queue_daemon_request(daemonrq_saveconfig, (union daemoninfo) NULL); 599 } 600 601 /* 602 * Write the configuration to all vinum slices. This 603 * is performed by the d�mon only 604 */ 605 void 606 daemon_save_config(void) 607 { 608 int error; 609 int written_config; /* set when we first write the config to disk */ 610 int driveno; 611 struct drive *drive; /* point to current drive info */ 612 struct vinum_hdr *vhdr; /* and as header */ 613 char *config; /* point to config data */ 614 int wlabel_on; /* to set writing label on/off */ 615 616 /* don't save the configuration while we're still working on it */ 617 if (vinum_conf.flags & VF_CONFIGURING) 618 return; 619 written_config = 0; /* no config written yet */ 620 /* Build a volume header */ 621 vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* get space for the config data */ 622 CHECKALLOC(vhdr, "Can't allocate config data"); 623 vhdr->magic = VINUM_MAGIC; /* magic number */ 624 vhdr->config_length = MAXCONFIG; /* length of following config info */ 625 626 config = Malloc(MAXCONFIG); /* get space for the config data */ 627 CHECKALLOC(config, "Can't allocate config data"); 628 629 format_config(config, MAXCONFIG); 630 error = 0; /* no errors yet */ 631 for (driveno = 0; driveno < vinum_conf.drives_allocated; driveno++) { 632 drive = &vinum_conf.drive[driveno]; /* point to drive */ 633 if (drive->state > drive_referenced) { 634 LOCKDRIVE(drive); /* don't let it change */ 635 636 /* 637 * First, do some drive consistency checks. Some 638 * of these are kludges, others require a process 639 * context and couldn't be done before 640 */ 641 if ((drive->devicename[0] == '\0') 642 || (drive->label.name[0] == '\0')) { 643 unlockdrive(drive); 644 free_drive(drive); /* get rid of it */ 645 break; 646 } 647 if (((drive->flags & VF_OPEN) == 0) /* drive not open */ 648 &&(drive->state > drive_down)) { /* and it thinks it's not down */ 649 unlockdrive(drive); 650 set_drive_state(driveno, drive_down, setstate_force); /* tell it what's what */ 651 continue; 652 } 653 if ((drive->state == drive_down) /* it's down */ 654 &&(drive->flags & VF_OPEN)) { /* but open, */ 655 unlockdrive(drive); 656 close_drive(drive); /* close it */ 657 } else if (drive->state > drive_down) { 658 getmicrotime(&drive->label.last_update); /* time of last update is now */ 659 bcopy((char *) &drive->label, /* and the label info from the drive structure */ 660 (char *) &vhdr->label, 661 sizeof(vhdr->label)); 662 if ((drive->state != drive_unallocated) 663 && (drive->state != drive_referenced)) { /* and it's a real drive */ 664 wlabel_on = 1; /* enable writing the label */ 665 error = dev_dioctl(drive->dev, /* make the label writeable */ 666 DIOCWLABEL, 667 (caddr_t) & wlabel_on, 668 FWRITE, 669 curthread); 670 if (error == 0) 671 error = write_drive(drive, (char *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); 672 if (error == 0) 673 error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET); /* first config copy */ 674 if (error == 0) 675 error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET + MAXCONFIG); /* second copy */ 676 wlabel_on = 0; /* enable writing the label */ 677 if (error == 0) 678 error = dev_dioctl(drive->dev, /* make the label non-writeable again */ 679 DIOCWLABEL, 680 (caddr_t) & wlabel_on, 681 FWRITE, 682 curthread); 683 unlockdrive(drive); 684 if (error) { 685 log(LOG_ERR, 686 "vinum: Can't write config to %s, error %d\n", 687 drive->devicename, 688 error); 689 set_drive_state(drive->driveno, drive_down, setstate_force); 690 } else 691 written_config = 1; /* we've written it on at least one drive */ 692 } 693 } else /* not worth looking at, */ 694 unlockdrive(drive); /* just unlock it again */ 695 } 696 } 697 Free(vhdr); 698 Free(config); 699 } 700 701 /* 702 * Disk labels are a mess. The correct way to 703 * access them is with the DIOC[GSW]DINFO ioctls, 704 * but some programs, such as newfs, access the 705 * disk directly, so we have to write things 706 * there. We do this only on request. If a user 707 * request tries to read it directly, we fake up 708 * one on the fly. 709 */ 710 711 /* 712 * get_volume_label returns a label structure to lp, which 713 * is allocated by the caller 714 */ 715 void 716 get_volume_label(char *name, int plexes, u_int64_t size, struct disklabel *lp) 717 { 718 bzero(lp, sizeof(struct disklabel)); 719 720 strncpy(lp->d_typename, "vinum", sizeof(lp->d_typename)); 721 lp->d_type = DTYPE_VINUM; 722 strncpy(lp->d_packname, name, min(sizeof(lp->d_packname), sizeof(name))); 723 lp->d_rpm = 14400 * plexes; /* to keep them guessing */ 724 lp->d_interleave = 1; 725 lp->d_flags = 0; 726 727 /* 728 * A Vinum volume has a single track with all 729 * its sectors. 730 */ 731 lp->d_secsize = DEV_BSIZE; /* bytes per sector */ 732 lp->d_nsectors = size; /* data sectors per track */ 733 lp->d_ntracks = 1; /* tracks per cylinder */ 734 lp->d_ncylinders = 1; /* data cylinders per unit */ 735 lp->d_secpercyl = size; /* data sectors per cylinder */ 736 lp->d_secperunit = size; /* data sectors per unit */ 737 738 lp->d_bbsize = BBSIZE; 739 lp->d_sbsize = SBSIZE; 740 741 lp->d_magic = DISKMAGIC; 742 lp->d_magic2 = DISKMAGIC; 743 744 /* 745 * Set up partitions a, b and c to be identical 746 * and the size of the volume. a is UFS, b is 747 * swap, c is nothing. 748 */ 749 lp->d_partitions[0].p_size = size; 750 lp->d_partitions[0].p_fsize = 1024; 751 lp->d_partitions[0].p_fstype = FS_BSDFFS; /* FreeBSD File System :-) */ 752 lp->d_partitions[0].p_fsize = 1024; /* FS fragment size */ 753 lp->d_partitions[0].p_frag = 8; /* and fragments per block */ 754 lp->d_partitions[SWAP_PART].p_size = size; 755 lp->d_partitions[SWAP_PART].p_fstype = FS_SWAP; /* swap partition */ 756 lp->d_partitions[LABEL_PART].p_size = size; 757 lp->d_npartitions = LABEL_PART + 1; 758 strncpy(lp->d_packname, name, min(sizeof(lp->d_packname), sizeof(name))); 759 lp->d_checksum = dkcksum(lp); 760 } 761 762 /* Write a volume label. This implements the VINUM_LABEL ioctl. */ 763 int 764 write_volume_label(int volno) 765 { 766 struct disklabel *lp; 767 struct buf *bp; 768 struct disklabel *dlp; 769 struct volume *vol; 770 int error; 771 dev_t dev; 772 773 lp = (struct disklabel *) Malloc((sizeof(struct disklabel) + (DEV_BSIZE - 1)) & (DEV_BSIZE - 1)); 774 if (lp == 0) 775 return ENOMEM; 776 777 if ((unsigned) (volno) >= (unsigned) vinum_conf.volumes_allocated) /* invalid volume */ 778 return ENOENT; 779 780 vol = &VOL[volno]; /* volume in question */ 781 if (vol->state <= volume_uninit) /* nothing there */ 782 return ENXIO; 783 else if (vol->state < volume_up) /* not accessible */ 784 return EIO; /* I/O error */ 785 786 get_volume_label(vol->name, vol->plexes, vol->size, lp); /* get the label */ 787 788 /* 789 * Now write to disk. This code is derived from the 790 * system writedisklabel (), which does silly things 791 * like reading the label and refusing to write 792 * unless it's already there. 793 */ 794 bp = geteblk((int) lp->d_secsize); /* get a buffer */ 795 dev = make_adhoc_dev(&vinum_cdevsw, vol->volno); 796 bp->b_bio1.bio_blkno = LABELSECTOR * ((int) lp->d_secsize / DEV_BSIZE); 797 bp->b_bcount = lp->d_secsize; 798 bzero(bp->b_data, lp->d_secsize); 799 dlp = (struct disklabel *) bp->b_data; 800 *dlp = *lp; 801 bp->b_flags &= ~B_INVAL; 802 bp->b_flags |= B_WRITE; 803 804 /* 805 * This should read: 806 * 807 * vinumstrategy (bp); 808 * 809 * Negotiate with phk to get it fixed. 810 */ 811 dev_dstrategy(dev, &bp->b_bio1); 812 error = biowait(bp); 813 bp->b_flags |= B_INVAL | B_AGE; 814 bp->b_flags &= ~B_ERROR; 815 brelse(bp); 816 return error; 817 } 818 819 /* Look at all disks on the system for vinum slices */ 820 int 821 vinum_scandisk(char *devicename[], int drives) 822 { 823 struct drive *volatile drive; 824 volatile int driveno; 825 int firstdrive; /* first drive in this list */ 826 volatile int gooddrives; /* number of usable drives found */ 827 int firsttime; /* set if we have never configured before */ 828 int error; 829 char *config_text; /* read the config info from disk into here */ 830 char *volatile cptr; /* pointer into config information */ 831 char *eptr; /* end pointer into config information */ 832 char *config_line; /* copy the config line to */ 833 volatile int status; 834 int *volatile drivelist; /* list of drive indices */ 835 #define DRIVENAMELEN 64 836 #define DRIVEPARTS 35 /* max partitions per drive, excluding c */ 837 char partname[DRIVENAMELEN]; /* for creating partition names */ 838 839 status = 0; /* success indication */ 840 vinum_conf.flags |= VF_READING_CONFIG; /* reading config from disk */ 841 842 gooddrives = 0; /* number of usable drives found */ 843 firstdrive = vinum_conf.drives_used; /* the first drive */ 844 firsttime = vinum_conf.drives_used == 0; /* are we a virgin? */ 845 846 /* allocate a drive pointer list */ 847 drivelist = (int *) Malloc(drives * DRIVEPARTS * sizeof(int)); 848 CHECKALLOC(drivelist, "Can't allocate memory"); 849 850 /* Open all drives and find which was modified most recently */ 851 for (driveno = 0; driveno < drives; driveno++) { 852 char part; /* UNIX partition */ 853 int slice; 854 int founddrive; /* flag when we find a vinum drive */ 855 int has_slice = 0; 856 int has_part = 0; 857 char *tmp; 858 859 founddrive = 0; /* no vinum drive found yet on this spindle */ 860 861 /* 862 * If the device path contains a slice we do not try to tack on 863 * another slice. If the device path has a partition we only check 864 * that partition. 865 */ 866 if ((tmp = rindex(devicename[driveno], '/')) == NULL) 867 tmp = devicename[driveno]; 868 while (*tmp && (*tmp < '0' || *tmp > '9')) 869 ++tmp; 870 while (*tmp && *tmp >= '0' && *tmp <= '9') 871 ++tmp; 872 if (*tmp == 's') 873 has_slice = strtol(tmp + 1, &tmp, 0); 874 if (*tmp >= 'a' && *tmp <= 'p') 875 has_part = *tmp; 876 877 /* 878 * Scan slices if no slice was specified, only if no partition was 879 * specified. 880 */ 881 if (has_slice == 0 && has_part == 0) 882 for (slice = 1; slice < 5; slice++) { 883 if (has_slice && slice != has_slice) 884 continue; 885 886 for (part = 'a'; part <= 'p'; part++) { 887 if (has_part && part != has_part) 888 continue; 889 if (part == 'c') 890 continue; 891 snprintf(partname, DRIVENAMELEN, 892 "%ss%d%c", devicename[driveno], slice, part); 893 drive = check_drive(partname); /* try to open it */ 894 if ((drive->lasterror != 0) /* didn't work, */ 895 ||(drive->state != drive_up)) 896 free_drive(drive); /* get rid of it */ 897 else if (drive->flags & VF_CONFIGURED) /* already read this config, */ 898 log(LOG_WARNING, 899 "vinum: already read config from %s\n", /* say so */ 900 drive->label.name); 901 else { 902 drivelist[gooddrives] = drive->driveno; /* keep the drive index */ 903 drive->flags &= ~VF_NEWBORN; /* which is no longer newly born */ 904 gooddrives++; 905 founddrive++; 906 } 907 } 908 } 909 if (founddrive == 0 && has_slice == 0) { /* didn't find anything, */ 910 for (part = 'a'; part <= 'p'; part++) { /* try the compatibility partition */ 911 if (has_part && has_part != part) 912 continue; 913 if (part == 'c') 914 continue; 915 if (has_part) { 916 snprintf(partname, DRIVENAMELEN, 917 "%s", devicename[driveno]); 918 } else { 919 snprintf(partname, DRIVENAMELEN, 920 "%s%c", devicename[driveno], part); 921 } 922 drive = check_drive(partname); /* try to open it */ 923 if ((drive->lasterror != 0) /* didn't work, */ 924 ||(drive->state != drive_up)) 925 free_drive(drive); /* get rid of it */ 926 else if (drive->flags & VF_CONFIGURED) /* already read this config, */ 927 log(LOG_WARNING, 928 "vinum: already read config from %s\n", /* say so */ 929 drive->label.name); 930 else { 931 drivelist[gooddrives] = drive->driveno; /* keep the drive index */ 932 drive->flags &= ~VF_NEWBORN; /* which is no longer newly born */ 933 gooddrives++; 934 } 935 } 936 } 937 } 938 939 if (gooddrives == 0) { 940 if (firsttime) 941 log(LOG_WARNING, "vinum: no drives found\n"); 942 else 943 log(LOG_INFO, "vinum: no additional drives found\n"); 944 return ENOENT; 945 } 946 /* 947 * We now have at least one drive 948 * open. Sort them in order of config time 949 * and merge the config info with what we 950 * have already. 951 */ 952 qsort(drivelist, gooddrives, sizeof(int), drivecmp); 953 config_text = (char *) Malloc(MAXCONFIG * 2); /* allocate buffers */ 954 CHECKALLOC(config_text, "Can't allocate memory"); 955 config_line = (char *) Malloc(MAXCONFIGLINE * 2); /* allocate buffers */ 956 CHECKALLOC(config_line, "Can't allocate memory"); 957 for (driveno = 0; driveno < gooddrives; driveno++) { /* now include the config */ 958 drive = &DRIVE[drivelist[driveno]]; /* point to the drive */ 959 960 if (firsttime && (driveno == 0)) /* we've never configured before, */ 961 log(LOG_INFO, "vinum: reading configuration from %s\n", drive->devicename); 962 else 963 log(LOG_INFO, "vinum: updating configuration from %s\n", drive->devicename); 964 965 if (drive->state == drive_up) 966 /* Read in both copies of the configuration information */ 967 error = read_drive(drive, config_text, MAXCONFIG * 2, VINUM_CONFIG_OFFSET); 968 else { 969 error = EIO; 970 printf("vinum_scandisk: %s is %s\n", drive->devicename, drive_state(drive->state)); 971 } 972 973 if (error != 0) { 974 log(LOG_ERR, "vinum: Can't read device %s, error %d\n", drive->devicename, error); 975 free_drive(drive); /* give it back */ 976 status = error; 977 } 978 /* 979 * At this point, check that the two copies 980 * are the same, and do something useful if 981 * not. In particular, consider which is 982 * newer, and what this means for the 983 * integrity of the data on the drive. 984 */ 985 else { 986 vinum_conf.drives_used++; /* another drive in use */ 987 /* Parse the configuration, and add it to the global configuration */ 988 for (cptr = config_text; *cptr != '\0';) { /* love this style(9) */ 989 volatile int parse_status; /* return value from parse_config */ 990 991 for (eptr = config_line; (*cptr != '\n') && (*cptr != '\0');) /* until the end of the line */ 992 *eptr++ = *cptr++; 993 *eptr = '\0'; /* and delimit */ 994 if (setjmp(command_fail) == 0) { /* come back here on error and continue */ 995 parse_status = parse_config(config_line, &keyword_set, 1); /* parse the config line */ 996 if (parse_status < 0) { /* error in config */ 997 /* 998 * This config should have been parsed in user 999 * space. If we run into problems here, something 1000 * serious is afoot. Complain and let the user 1001 * snarf the config to see what's wrong. 1002 */ 1003 log(LOG_ERR, 1004 "vinum: Config error on %s, aborting integration\n", 1005 drive->devicename); 1006 free_drive(drive); /* give it back */ 1007 status = EINVAL; 1008 } 1009 } 1010 while (*cptr == '\n') 1011 cptr++; /* skip to next line */ 1012 } 1013 } 1014 drive->flags |= VF_CONFIGURED; /* read this drive's configuration */ 1015 } 1016 1017 Free(config_line); 1018 Free(config_text); 1019 Free(drivelist); 1020 vinum_conf.flags &= ~VF_READING_CONFIG; /* no longer reading from disk */ 1021 if (status != 0) 1022 printf("vinum: couldn't read configuration"); 1023 else 1024 updateconfig(VF_READING_CONFIG); /* update from disk config */ 1025 return status; 1026 } 1027 1028 /* 1029 * Compare the modification dates of the drives, for qsort. 1030 * Return 1 if a < b, 0 if a == b, 01 if a > b: in other 1031 * words, sort backwards. 1032 */ 1033 int 1034 drivecmp(const void *va, const void *vb) 1035 { 1036 const struct drive *a = &DRIVE[*(const int *) va]; 1037 const struct drive *b = &DRIVE[*(const int *) vb]; 1038 1039 if ((a->label.last_update.tv_sec == b->label.last_update.tv_sec) 1040 && (a->label.last_update.tv_usec == b->label.last_update.tv_usec)) 1041 return 0; 1042 else if ((a->label.last_update.tv_sec > b->label.last_update.tv_sec) 1043 || ((a->label.last_update.tv_sec == b->label.last_update.tv_sec) 1044 && (a->label.last_update.tv_usec > b->label.last_update.tv_usec))) 1045 return -1; 1046 else 1047 return 1; 1048 } 1049 /* Local Variables: */ 1050 /* fill-column: 50 */ 1051 /* End: */ 1052