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 1.6 2004/05/19 22:52:48 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 100 return ENODEV; 101 dname += 2; /* point past */ 102 103 /* 104 * Found the device. We can expect one of 105 * two formats for the rest: a unit number, 106 * then either a partition letter for the 107 * compatiblity partition (e.g. h) or a 108 * slice ID and partition (e.g. s2e). 109 * Create a minor number for each of them. 110 */ 111 unit = 0; 112 while ((*dname >= '0') /* unit number */ 113 &&(*dname <= '9')) { 114 unit = unit * 10 + *dname - '0'; 115 dname++; 116 } 117 118 if (*dname == 's') { /* slice */ 119 if (((dname[1] < '1') || (dname[1] > '4')) /* invalid slice */ 120 ||((dname[2] < 'a') || (dname[2] > 'h'))) /* or invalid partition */ 121 return ENODEV; 122 devminor = ((unit & 31) << 3) /* unit */ 123 +(dname[2] - 'a') /* partition */ 124 +((dname[1] - '0' + 1) << 16) /* slice */ 125 +((unit & ~31) << 16); /* high-order unit bits */ 126 } else { /* compatibility partition */ 127 if ((*dname < 'a') || (*dname > 'h')) /* or invalid partition */ 128 return ENODEV; 129 devminor = (*dname - 'a') /* partition */ 130 +((unit & 31) << 3) /* unit */ 131 +((unit & ~31) << 16); /* high-order unit bits */ 132 } 133 134 if ((devminor & 7) == 2) /* partition c */ 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_dev = drive->dev; /* device */ 328 bp->b_blkno = offset / drive->partinfo.disklab->d_secsize; /* block number */ 329 bp->b_saveaddr = bp->b_data; 330 bp->b_data = buf; 331 bp->b_bcount = len; 332 BUF_STRATEGY(bp, 0); /* initiate the transfer */ 333 error = biowait(bp); 334 bp->b_data = bp->b_saveaddr; 335 bp->b_flags |= B_INVAL | B_AGE; 336 bp->b_flags &= ~B_ERROR; 337 brelse(bp); 338 if (error) 339 break; 340 length -= len; /* update pointers */ 341 buf += len; 342 offset += len; 343 } 344 return error; 345 } 346 347 /* 348 * Check a drive for a vinum header. If found, 349 * update the drive information. We come here 350 * with a partially populated drive structure 351 * which includes the device name. 352 * 353 * Return information on what we found. 354 * 355 * This function is called from two places: check_drive, 356 * which wants to find out whether the drive is a 357 * Vinum drive, and config_drive, which asserts that 358 * it is a vinum drive. In the first case, we don't 359 * print error messages (verbose==0), in the second 360 * we do (verbose==1). 361 */ 362 enum drive_label_info 363 read_drive_label(struct drive *drive, int verbose) 364 { 365 int error; 366 int result; /* result of our search */ 367 struct vinum_hdr *vhdr; /* and as header */ 368 369 error = init_drive(drive, 0); /* find the drive */ 370 if (error) /* find the drive */ 371 return DL_CANT_OPEN; /* not ours */ 372 373 vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* allocate buffers */ 374 CHECKALLOC(vhdr, "Can't allocate memory"); 375 376 drive->state = drive_up; /* be optimistic */ 377 error = read_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); 378 if (vhdr->magic == VINUM_MAGIC) { /* ours! */ 379 if (drive->label.name[0] /* we have a name for this drive */ 380 &&(strcmp(drive->label.name, vhdr->label.name))) { /* but it doesn't match the real name */ 381 drive->lasterror = EINVAL; 382 result = DL_WRONG_DRIVE; /* it's the wrong drive */ 383 drive->state = drive_unallocated; /* put it back, it's not ours */ 384 } else 385 result = DL_OURS; 386 /* 387 * We copy the drive anyway so that we have 388 * the correct name in the drive info. This 389 * may not be the name specified 390 */ 391 drive->label = vhdr->label; /* put in the label information */ 392 } else if (vhdr->magic == VINUM_NOMAGIC) /* was ours, but we gave it away */ 393 result = DL_DELETED_LABEL; /* and return the info */ 394 else 395 result = DL_NOT_OURS; /* we could have it, but we don't yet */ 396 Free(vhdr); /* that's all. */ 397 return result; 398 } 399 400 /* 401 * Check a drive for a vinum header. If found, 402 * read configuration information from the drive and 403 * incorporate the data into the configuration. 404 * 405 * Return drive number. 406 */ 407 struct drive * 408 check_drive(char *devicename) 409 { 410 int driveno; 411 int i; 412 struct drive *drive; 413 414 driveno = find_drive_by_dev(devicename, 1); /* if entry doesn't exist, create it */ 415 drive = &vinum_conf.drive[driveno]; /* and get a pointer */ 416 417 if (read_drive_label(drive, 0) == DL_OURS) { /* one of ours */ 418 for (i = 0; i < vinum_conf.drives_allocated; i++) { /* see if the name already exists */ 419 if ((i != driveno) /* not this drive */ 420 &&(DRIVE[i].state != drive_unallocated) /* and it's allocated */ 421 &&(strcmp(DRIVE[i].label.name, 422 DRIVE[driveno].label.name) == 0)) { /* and it has the same name */ 423 struct drive *mydrive = &DRIVE[i]; 424 425 if (mydrive->devicename[0] == '/') { /* we know a device name for it */ 426 /* 427 * set an error, but don't take the 428 * drive down: that would cause unneeded 429 * error messages. 430 */ 431 drive->lasterror = EEXIST; 432 break; 433 } else { /* it's just a place holder, */ 434 int sdno; 435 436 for (sdno = 0; sdno < vinum_conf.subdisks_allocated; sdno++) { /* look at each subdisk */ 437 if ((SD[sdno].driveno == i) /* it's pointing to this one, */ 438 &&(SD[sdno].state != sd_unallocated)) { /* and it's a real subdisk */ 439 SD[sdno].driveno = drive->driveno; /* point to the one we found */ 440 update_sd_state(sdno); /* and update its state */ 441 } 442 } 443 bzero(mydrive, sizeof(struct drive)); /* don't deallocate it, just remove it */ 444 } 445 } 446 } 447 } else { 448 if (drive->lasterror == 0) 449 drive->lasterror = ENODEV; 450 close_drive(drive); 451 drive->state = drive_down; 452 } 453 return drive; 454 } 455 456 static char * 457 sappend(char *txt, char *s) 458 { 459 while ((*s++ = *txt++) != 0); 460 return s - 1; 461 } 462 463 void 464 format_config(char *config, int len) 465 { 466 int i; 467 int j; 468 char *s = config; 469 char *configend = &config[len]; 470 471 bzero(config, len); 472 473 /* First write the volume configuration */ 474 for (i = 0; i < vinum_conf.volumes_allocated; i++) { 475 struct volume *vol; 476 477 vol = &vinum_conf.volume[i]; 478 if ((vol->state > volume_uninit) 479 && (vol->name[0] != '\0')) { /* paranoia */ 480 snprintf(s, 481 configend - s, 482 "volume %s state %s", 483 vol->name, 484 volume_state(vol->state)); 485 while (*s) 486 s++; /* find the end */ 487 if (vol->preferred_plex >= 0) /* preferences, */ 488 snprintf(s, 489 configend - s, 490 " readpol prefer %s", 491 vinum_conf.plex[vol->preferred_plex].name); 492 while (*s) 493 s++; /* find the end */ 494 s = sappend("\n", s); 495 } 496 } 497 498 /* Then the plex configuration */ 499 for (i = 0; i < vinum_conf.plexes_allocated; i++) { 500 struct plex *plex; 501 502 plex = &vinum_conf.plex[i]; 503 if ((plex->state > plex_referenced) 504 && (plex->name[0] != '\0')) { /* paranoia */ 505 snprintf(s, 506 configend - s, 507 "plex name %s state %s org %s ", 508 plex->name, 509 plex_state(plex->state), 510 plex_org(plex->organization)); 511 while (*s) 512 s++; /* find the end */ 513 if (isstriped(plex)) { 514 snprintf(s, 515 configend - s, 516 "%ds ", 517 (int) plex->stripesize); 518 while (*s) 519 s++; /* find the end */ 520 } 521 if (plex->volno >= 0) /* we have a volume */ 522 snprintf(s, 523 configend - s, 524 "vol %s ", 525 vinum_conf.volume[plex->volno].name); 526 while (*s) 527 s++; /* find the end */ 528 for (j = 0; j < plex->subdisks; j++) { 529 snprintf(s, 530 configend - s, 531 " sd %s", 532 vinum_conf.sd[plex->sdnos[j]].name); 533 } 534 s = sappend("\n", s); 535 } 536 } 537 538 /* And finally the subdisk configuration */ 539 for (i = 0; i < vinum_conf.subdisks_allocated; i++) { 540 struct sd *sd; 541 char *drivename; 542 543 sd = &SD[i]; 544 if ((sd->state != sd_referenced) 545 && (sd->state != sd_unallocated) 546 && (sd->name[0] != '\0')) { /* paranoia */ 547 drivename = vinum_conf.drive[sd->driveno].label.name; 548 /* 549 * XXX We've seen cases of dead subdisks 550 * which don't have a drive. If we let them 551 * through here, the drive name is null, so 552 * they get the drive named 'plex'. 553 * 554 * This is a breakage limiter, not a fix. 555 */ 556 if (drivename[0] == '\0') 557 drivename = "*invalid*"; 558 snprintf(s, 559 configend - s, 560 "sd name %s drive %s plex %s len %llus driveoffset %llus state %s", 561 sd->name, 562 drivename, 563 vinum_conf.plex[sd->plexno].name, 564 (unsigned long long) sd->sectors, 565 (unsigned long long) sd->driveoffset, 566 sd_state(sd->state)); 567 while (*s) 568 s++; /* find the end */ 569 if (sd->plexno >= 0) 570 snprintf(s, 571 configend - s, 572 " plexoffset %llds", 573 (long long) sd->plexoffset); 574 else 575 snprintf(s, configend - s, " detached"); 576 while (*s) 577 s++; /* find the end */ 578 if (sd->flags & VF_RETRYERRORS) { 579 snprintf(s, configend - s, " retryerrors"); 580 while (*s) 581 s++; /* find the end */ 582 } 583 snprintf(s, configend - s, " \n"); 584 while (*s) 585 s++; /* find the end */ 586 } 587 } 588 if (s > &config[len - 2]) 589 panic("vinum: configuration data overflow"); 590 } 591 592 /* 593 * issue a save config request to the d�mon. The actual work 594 * is done in process context by daemon_save_config 595 */ 596 void 597 save_config(void) 598 { 599 queue_daemon_request(daemonrq_saveconfig, (union daemoninfo) NULL); 600 } 601 602 /* 603 * Write the configuration to all vinum slices. This 604 * is performed by the d�mon only 605 */ 606 void 607 daemon_save_config(void) 608 { 609 int error; 610 int written_config; /* set when we first write the config to disk */ 611 int driveno; 612 struct drive *drive; /* point to current drive info */ 613 struct vinum_hdr *vhdr; /* and as header */ 614 char *config; /* point to config data */ 615 int wlabel_on; /* to set writing label on/off */ 616 617 /* don't save the configuration while we're still working on it */ 618 if (vinum_conf.flags & VF_CONFIGURING) 619 return; 620 written_config = 0; /* no config written yet */ 621 /* Build a volume header */ 622 vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* get space for the config data */ 623 CHECKALLOC(vhdr, "Can't allocate config data"); 624 vhdr->magic = VINUM_MAGIC; /* magic number */ 625 vhdr->config_length = MAXCONFIG; /* length of following config info */ 626 627 config = Malloc(MAXCONFIG); /* get space for the config data */ 628 CHECKALLOC(config, "Can't allocate config data"); 629 630 format_config(config, MAXCONFIG); 631 error = 0; /* no errors yet */ 632 for (driveno = 0; driveno < vinum_conf.drives_allocated; driveno++) { 633 drive = &vinum_conf.drive[driveno]; /* point to drive */ 634 if (drive->state > drive_referenced) { 635 LOCKDRIVE(drive); /* don't let it change */ 636 637 /* 638 * First, do some drive consistency checks. Some 639 * of these are kludges, others require a process 640 * context and couldn't be done before 641 */ 642 if ((drive->devicename[0] == '\0') 643 || (drive->label.name[0] == '\0')) { 644 unlockdrive(drive); 645 free_drive(drive); /* get rid of it */ 646 break; 647 } 648 if (((drive->flags & VF_OPEN) == 0) /* drive not open */ 649 &&(drive->state > drive_down)) { /* and it thinks it's not down */ 650 unlockdrive(drive); 651 set_drive_state(driveno, drive_down, setstate_force); /* tell it what's what */ 652 continue; 653 } 654 if ((drive->state == drive_down) /* it's down */ 655 &&(drive->flags & VF_OPEN)) { /* but open, */ 656 unlockdrive(drive); 657 close_drive(drive); /* close it */ 658 } else if (drive->state > drive_down) { 659 getmicrotime(&drive->label.last_update); /* time of last update is now */ 660 bcopy((char *) &drive->label, /* and the label info from the drive structure */ 661 (char *) &vhdr->label, 662 sizeof(vhdr->label)); 663 if ((drive->state != drive_unallocated) 664 && (drive->state != drive_referenced)) { /* and it's a real drive */ 665 wlabel_on = 1; /* enable writing the label */ 666 error = dev_dioctl(drive->dev, /* make the label writeable */ 667 DIOCWLABEL, 668 (caddr_t) & wlabel_on, 669 FWRITE, 670 curthread); 671 if (error == 0) 672 error = write_drive(drive, (char *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); 673 if (error == 0) 674 error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET); /* first config copy */ 675 if (error == 0) 676 error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET + MAXCONFIG); /* second copy */ 677 wlabel_on = 0; /* enable writing the label */ 678 if (error == 0) 679 error = dev_dioctl(drive->dev, /* make the label non-writeable again */ 680 DIOCWLABEL, 681 (caddr_t) & wlabel_on, 682 FWRITE, 683 curthread); 684 unlockdrive(drive); 685 if (error) { 686 log(LOG_ERR, 687 "vinum: Can't write config to %s, error %d\n", 688 drive->devicename, 689 error); 690 set_drive_state(drive->driveno, drive_down, setstate_force); 691 } else 692 written_config = 1; /* we've written it on at least one drive */ 693 } 694 } else /* not worth looking at, */ 695 unlockdrive(drive); /* just unlock it again */ 696 } 697 } 698 Free(vhdr); 699 Free(config); 700 } 701 702 /* 703 * Disk labels are a mess. The correct way to 704 * access them is with the DIOC[GSW]DINFO ioctls, 705 * but some programs, such as newfs, access the 706 * disk directly, so we have to write things 707 * there. We do this only on request. If a user 708 * request tries to read it directly, we fake up 709 * one on the fly. 710 */ 711 712 /* 713 * get_volume_label returns a label structure to lp, which 714 * is allocated by the caller 715 */ 716 void 717 get_volume_label(char *name, int plexes, u_int64_t size, struct disklabel *lp) 718 { 719 bzero(lp, sizeof(struct disklabel)); 720 721 strncpy(lp->d_typename, "vinum", sizeof(lp->d_typename)); 722 lp->d_type = DTYPE_VINUM; 723 strncpy(lp->d_packname, name, min(sizeof(lp->d_packname), sizeof(name))); 724 lp->d_rpm = 14400 * plexes; /* to keep them guessing */ 725 lp->d_interleave = 1; 726 lp->d_flags = 0; 727 728 /* 729 * A Vinum volume has a single track with all 730 * its sectors. 731 */ 732 lp->d_secsize = DEV_BSIZE; /* bytes per sector */ 733 lp->d_nsectors = size; /* data sectors per track */ 734 lp->d_ntracks = 1; /* tracks per cylinder */ 735 lp->d_ncylinders = 1; /* data cylinders per unit */ 736 lp->d_secpercyl = size; /* data sectors per cylinder */ 737 lp->d_secperunit = size; /* data sectors per unit */ 738 739 lp->d_bbsize = BBSIZE; 740 lp->d_sbsize = SBSIZE; 741 742 lp->d_magic = DISKMAGIC; 743 lp->d_magic2 = DISKMAGIC; 744 745 /* 746 * Set up partitions a, b and c to be identical 747 * and the size of the volume. a is UFS, b is 748 * swap, c is nothing. 749 */ 750 lp->d_partitions[0].p_size = size; 751 lp->d_partitions[0].p_fsize = 1024; 752 lp->d_partitions[0].p_fstype = FS_BSDFFS; /* FreeBSD File System :-) */ 753 lp->d_partitions[0].p_fsize = 1024; /* FS fragment size */ 754 lp->d_partitions[0].p_frag = 8; /* and fragments per block */ 755 lp->d_partitions[SWAP_PART].p_size = size; 756 lp->d_partitions[SWAP_PART].p_fstype = FS_SWAP; /* swap partition */ 757 lp->d_partitions[LABEL_PART].p_size = size; 758 lp->d_npartitions = LABEL_PART + 1; 759 strncpy(lp->d_packname, name, min(sizeof(lp->d_packname), sizeof(name))); 760 lp->d_checksum = dkcksum(lp); 761 } 762 763 /* Write a volume label. This implements the VINUM_LABEL ioctl. */ 764 int 765 write_volume_label(int volno) 766 { 767 struct disklabel *lp; 768 struct buf *bp; 769 struct disklabel *dlp; 770 struct volume *vol; 771 int error; 772 dev_t dev; 773 774 lp = (struct disklabel *) Malloc((sizeof(struct disklabel) + (DEV_BSIZE - 1)) & (DEV_BSIZE - 1)); 775 if (lp == 0) 776 return ENOMEM; 777 778 if ((unsigned) (volno) >= (unsigned) vinum_conf.volumes_allocated) /* invalid volume */ 779 return ENOENT; 780 781 vol = &VOL[volno]; /* volume in question */ 782 if (vol->state <= volume_uninit) /* nothing there */ 783 return ENXIO; 784 else if (vol->state < volume_up) /* not accessible */ 785 return EIO; /* I/O error */ 786 787 get_volume_label(vol->name, vol->plexes, vol->size, lp); /* get the label */ 788 789 /* 790 * Now write to disk. This code is derived from the 791 * system writedisklabel (), which does silly things 792 * like reading the label and refusing to write 793 * unless it's already there. 794 */ 795 bp = geteblk((int) lp->d_secsize); /* get a buffer */ 796 dev = make_adhoc_dev(&vinum_cdevsw, vol->volno); 797 bp->b_dev = dev; 798 bp->b_blkno = LABELSECTOR * ((int) lp->d_secsize / DEV_BSIZE); 799 bp->b_bcount = lp->d_secsize; 800 bzero(bp->b_data, lp->d_secsize); 801 dlp = (struct disklabel *) bp->b_data; 802 *dlp = *lp; 803 bp->b_flags &= ~B_INVAL; 804 bp->b_flags |= B_WRITE; 805 806 /* 807 * This should read: 808 * 809 * vinumstrategy (bp); 810 * 811 * Negotiate with phk to get it fixed. 812 */ 813 BUF_STRATEGY(bp, 0); 814 error = biowait(bp); 815 bp->b_flags |= B_INVAL | B_AGE; 816 bp->b_flags &= ~B_ERROR; 817 brelse(bp); 818 return error; 819 } 820 821 /* Look at all disks on the system for vinum slices */ 822 int 823 vinum_scandisk(char *devicename[], int drives) 824 { 825 struct drive *volatile drive; 826 volatile int driveno; 827 int firstdrive; /* first drive in this list */ 828 volatile int gooddrives; /* number of usable drives found */ 829 int firsttime; /* set if we have never configured before */ 830 int error; 831 char *config_text; /* read the config info from disk into here */ 832 char *volatile cptr; /* pointer into config information */ 833 char *eptr; /* end pointer into config information */ 834 char *config_line; /* copy the config line to */ 835 volatile int status; 836 int *volatile drivelist; /* list of drive indices */ 837 #define DRIVENAMELEN 64 838 #define DRIVEPARTS 35 /* max partitions per drive, excluding c */ 839 char partname[DRIVENAMELEN]; /* for creating partition names */ 840 841 status = 0; /* success indication */ 842 vinum_conf.flags |= VF_READING_CONFIG; /* reading config from disk */ 843 844 gooddrives = 0; /* number of usable drives found */ 845 firstdrive = vinum_conf.drives_used; /* the first drive */ 846 firsttime = vinum_conf.drives_used == 0; /* are we a virgin? */ 847 848 /* allocate a drive pointer list */ 849 drivelist = (int *) Malloc(drives * DRIVEPARTS * sizeof(int)); 850 CHECKALLOC(drivelist, "Can't allocate memory"); 851 852 /* Open all drives and find which was modified most recently */ 853 for (driveno = 0; driveno < drives; driveno++) { 854 char part; /* UNIX partition */ 855 int slice; 856 int founddrive; /* flag when we find a vinum drive */ 857 858 founddrive = 0; /* no vinum drive found yet on this spindle */ 859 /* first try the partition table */ 860 for (slice = 1; slice < 5; slice++) 861 for (part = 'a'; part < 'i'; part++) { 862 if (part != 'c') { /* don't do the c partition */ 863 snprintf(partname, 864 DRIVENAMELEN, 865 "%ss%d%c", 866 devicename[driveno], 867 slice, 868 part); 869 drive = check_drive(partname); /* try to open it */ 870 if ((drive->lasterror != 0) /* didn't work, */ 871 ||(drive->state != drive_up)) 872 free_drive(drive); /* get rid of it */ 873 else if (drive->flags & VF_CONFIGURED) /* already read this config, */ 874 log(LOG_WARNING, 875 "vinum: already read config from %s\n", /* say so */ 876 drive->label.name); 877 else { 878 drivelist[gooddrives] = drive->driveno; /* keep the drive index */ 879 drive->flags &= ~VF_NEWBORN; /* which is no longer newly born */ 880 gooddrives++; 881 founddrive++; 882 } 883 } 884 } 885 if (founddrive == 0) { /* didn't find anything, */ 886 for (part = 'a'; part < 'i'; part++) /* try the compatibility partition */ 887 if (part != 'c') { /* don't do the c partition */ 888 snprintf(partname, /* /dev/sd0a */ 889 DRIVENAMELEN, 890 "%s%c", 891 devicename[driveno], 892 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 } 906 } 907 } 908 } 909 910 if (gooddrives == 0) { 911 if (firsttime) 912 log(LOG_WARNING, "vinum: no drives found\n"); 913 else 914 log(LOG_INFO, "vinum: no additional drives found\n"); 915 return ENOENT; 916 } 917 /* 918 * We now have at least one drive 919 * open. Sort them in order of config time 920 * and merge the config info with what we 921 * have already. 922 */ 923 qsort(drivelist, gooddrives, sizeof(int), drivecmp); 924 config_text = (char *) Malloc(MAXCONFIG * 2); /* allocate buffers */ 925 CHECKALLOC(config_text, "Can't allocate memory"); 926 config_line = (char *) Malloc(MAXCONFIGLINE * 2); /* allocate buffers */ 927 CHECKALLOC(config_line, "Can't allocate memory"); 928 for (driveno = 0; driveno < gooddrives; driveno++) { /* now include the config */ 929 drive = &DRIVE[drivelist[driveno]]; /* point to the drive */ 930 931 if (firsttime && (driveno == 0)) /* we've never configured before, */ 932 log(LOG_INFO, "vinum: reading configuration from %s\n", drive->devicename); 933 else 934 log(LOG_INFO, "vinum: updating configuration from %s\n", drive->devicename); 935 936 if (drive->state == drive_up) 937 /* Read in both copies of the configuration information */ 938 error = read_drive(drive, config_text, MAXCONFIG * 2, VINUM_CONFIG_OFFSET); 939 else { 940 error = EIO; 941 printf("vinum_scandisk: %s is %s\n", drive->devicename, drive_state(drive->state)); 942 } 943 944 if (error != 0) { 945 log(LOG_ERR, "vinum: Can't read device %s, error %d\n", drive->devicename, error); 946 free_drive(drive); /* give it back */ 947 status = error; 948 } 949 /* 950 * At this point, check that the two copies 951 * are the same, and do something useful if 952 * not. In particular, consider which is 953 * newer, and what this means for the 954 * integrity of the data on the drive. 955 */ 956 else { 957 vinum_conf.drives_used++; /* another drive in use */ 958 /* Parse the configuration, and add it to the global configuration */ 959 for (cptr = config_text; *cptr != '\0';) { /* love this style(9) */ 960 volatile int parse_status; /* return value from parse_config */ 961 962 for (eptr = config_line; (*cptr != '\n') && (*cptr != '\0');) /* until the end of the line */ 963 *eptr++ = *cptr++; 964 *eptr = '\0'; /* and delimit */ 965 if (setjmp(command_fail) == 0) { /* come back here on error and continue */ 966 parse_status = parse_config(config_line, &keyword_set, 1); /* parse the config line */ 967 if (parse_status < 0) { /* error in config */ 968 /* 969 * This config should have been parsed in user 970 * space. If we run into problems here, something 971 * serious is afoot. Complain and let the user 972 * snarf the config to see what's wrong. 973 */ 974 log(LOG_ERR, 975 "vinum: Config error on %s, aborting integration\n", 976 drive->devicename); 977 free_drive(drive); /* give it back */ 978 status = EINVAL; 979 } 980 } 981 while (*cptr == '\n') 982 cptr++; /* skip to next line */ 983 } 984 } 985 drive->flags |= VF_CONFIGURED; /* read this drive's configuration */ 986 } 987 988 Free(config_text); 989 Free(drivelist); 990 vinum_conf.flags &= ~VF_READING_CONFIG; /* no longer reading from disk */ 991 if (status != 0) 992 printf("vinum: couldn't read configuration"); 993 else 994 updateconfig(VF_READING_CONFIG); /* update from disk config */ 995 return status; 996 } 997 998 /* 999 * Compare the modification dates of the drives, for qsort. 1000 * Return 1 if a < b, 0 if a == b, 01 if a > b: in other 1001 * words, sort backwards. 1002 */ 1003 int 1004 drivecmp(const void *va, const void *vb) 1005 { 1006 const struct drive *a = &DRIVE[*(const int *) va]; 1007 const struct drive *b = &DRIVE[*(const int *) vb]; 1008 1009 if ((a->label.last_update.tv_sec == b->label.last_update.tv_sec) 1010 && (a->label.last_update.tv_usec == b->label.last_update.tv_usec)) 1011 return 0; 1012 else if ((a->label.last_update.tv_sec > b->label.last_update.tv_sec) 1013 || ((a->label.last_update.tv_sec == b->label.last_update.tv_sec) 1014 && (a->label.last_update.tv_usec > b->label.last_update.tv_usec))) 1015 return -1; 1016 else 1017 return 1; 1018 } 1019 /* Local Variables: */ 1020 /* fill-column: 50 */ 1021 /* End: */ 1022