1 /* $NetBSD: rf_disks.c,v 1.58 2005/05/29 22:03:09 christos Exp $ */ 2 /*- 3 * Copyright (c) 1999 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Greg Oster 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 38 /* 39 * Copyright (c) 1995 Carnegie-Mellon University. 40 * All rights reserved. 41 * 42 * Author: Mark Holland 43 * 44 * Permission to use, copy, modify and distribute this software and 45 * its documentation is hereby granted, provided that both the copyright 46 * notice and this permission notice appear in all copies of the 47 * software, derivative works or modified versions, and any portions 48 * thereof, and that both notices appear in supporting documentation. 49 * 50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 53 * 54 * Carnegie Mellon requests users of this software to return to 55 * 56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 57 * School of Computer Science 58 * Carnegie Mellon University 59 * Pittsburgh PA 15213-3890 60 * 61 * any improvements or extensions that they make and grant Carnegie the 62 * rights to redistribute these changes. 63 */ 64 65 /*************************************************************** 66 * rf_disks.c -- code to perform operations on the actual disks 67 ***************************************************************/ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: rf_disks.c,v 1.58 2005/05/29 22:03:09 christos Exp $"); 71 72 #include <dev/raidframe/raidframevar.h> 73 74 #include "rf_raid.h" 75 #include "rf_alloclist.h" 76 #include "rf_utils.h" 77 #include "rf_general.h" 78 #include "rf_options.h" 79 #include "rf_kintf.h" 80 #include "rf_netbsd.h" 81 82 #include <sys/param.h> 83 #include <sys/systm.h> 84 #include <sys/proc.h> 85 #include <sys/ioctl.h> 86 #include <sys/fcntl.h> 87 #include <sys/vnode.h> 88 89 static int rf_AllocDiskStructures(RF_Raid_t *, RF_Config_t *); 90 static void rf_print_label_status( RF_Raid_t *, int, char *, 91 RF_ComponentLabel_t *); 92 static int rf_check_label_vitals( RF_Raid_t *, int, int, char *, 93 RF_ComponentLabel_t *, int, int ); 94 95 #define DPRINTF6(a,b,c,d,e,f) if (rf_diskDebug) printf(a,b,c,d,e,f) 96 #define DPRINTF7(a,b,c,d,e,f,g) if (rf_diskDebug) printf(a,b,c,d,e,f,g) 97 98 /************************************************************************** 99 * 100 * initialize the disks comprising the array 101 * 102 * We want the spare disks to have regular row,col numbers so that we can 103 * easily substitue a spare for a failed disk. But, the driver code assumes 104 * throughout that the array contains numRow by numCol _non-spare_ disks, so 105 * it's not clear how to fit in the spares. This is an unfortunate holdover 106 * from raidSim. The quick and dirty fix is to make row zero bigger than the 107 * rest, and put all the spares in it. This probably needs to get changed 108 * eventually. 109 * 110 **************************************************************************/ 111 112 int 113 rf_ConfigureDisks(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr, 114 RF_Config_t *cfgPtr) 115 { 116 RF_RaidDisk_t *disks; 117 RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL; 118 RF_RowCol_t c; 119 int bs, ret; 120 unsigned i, count, foundone = 0, numFailuresThisRow; 121 int force; 122 123 force = cfgPtr->force; 124 125 ret = rf_AllocDiskStructures(raidPtr, cfgPtr); 126 if (ret) 127 goto fail; 128 129 disks = raidPtr->Disks; 130 131 numFailuresThisRow = 0; 132 for (c = 0; c < raidPtr->numCol; c++) { 133 ret = rf_ConfigureDisk(raidPtr, 134 &cfgPtr->devnames[0][c][0], 135 &disks[c], c); 136 137 if (ret) 138 goto fail; 139 140 if (disks[c].status == rf_ds_optimal) { 141 raidread_component_label( 142 raidPtr->raid_cinfo[c].ci_dev, 143 raidPtr->raid_cinfo[c].ci_vp, 144 &raidPtr->raid_cinfo[c].ci_label); 145 } 146 147 if (disks[c].status != rf_ds_optimal) { 148 numFailuresThisRow++; 149 } else { 150 if (disks[c].numBlocks < min_numblks) 151 min_numblks = disks[c].numBlocks; 152 DPRINTF6("Disk at col %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", 153 c, disks[c].devname, 154 (long int) disks[c].numBlocks, 155 disks[c].blockSize, 156 (long int) disks[c].numBlocks * 157 disks[c].blockSize / 1024 / 1024); 158 } 159 } 160 /* XXX fix for n-fault tolerant */ 161 /* XXX this should probably check to see how many failures 162 we can handle for this configuration! */ 163 if (numFailuresThisRow > 0) 164 raidPtr->status = rf_rs_degraded; 165 166 /* all disks must be the same size & have the same block size, bs must 167 * be a power of 2 */ 168 bs = 0; 169 foundone = 0; 170 for (c = 0; c < raidPtr->numCol; c++) { 171 if (disks[c].status == rf_ds_optimal) { 172 bs = disks[c].blockSize; 173 foundone = 1; 174 break; 175 } 176 } 177 if (!foundone) { 178 RF_ERRORMSG("RAIDFRAME: Did not find any live disks in the array.\n"); 179 ret = EINVAL; 180 goto fail; 181 } 182 for (count = 0, i = 1; i; i <<= 1) 183 if (bs & i) 184 count++; 185 if (count != 1) { 186 RF_ERRORMSG1("Error: block size on disks (%d) must be a power of 2\n", bs); 187 ret = EINVAL; 188 goto fail; 189 } 190 191 if (rf_CheckLabels( raidPtr, cfgPtr )) { 192 printf("raid%d: There were fatal errors\n", raidPtr->raidid); 193 if (force != 0) { 194 printf("raid%d: Fatal errors being ignored.\n", 195 raidPtr->raidid); 196 } else { 197 ret = EINVAL; 198 goto fail; 199 } 200 } 201 202 for (c = 0; c < raidPtr->numCol; c++) { 203 if (disks[c].status == rf_ds_optimal) { 204 if (disks[c].blockSize != bs) { 205 RF_ERRORMSG1("Error: block size of disk at c %d different from disk at c 0\n", c); 206 ret = EINVAL; 207 goto fail; 208 } 209 if (disks[c].numBlocks != min_numblks) { 210 RF_ERRORMSG2("WARNING: truncating disk at c %d to %d blocks\n", 211 c, (int) min_numblks); 212 disks[c].numBlocks = min_numblks; 213 } 214 } 215 } 216 217 raidPtr->sectorsPerDisk = min_numblks; 218 raidPtr->logBytesPerSector = ffs(bs) - 1; 219 raidPtr->bytesPerSector = bs; 220 raidPtr->sectorMask = bs - 1; 221 return (0); 222 223 fail: 224 225 rf_UnconfigureVnodes( raidPtr ); 226 227 return (ret); 228 } 229 230 231 /**************************************************************************** 232 * set up the data structures describing the spare disks in the array 233 * recall from the above comment that the spare disk descriptors are stored 234 * in row zero, which is specially expanded to hold them. 235 ****************************************************************************/ 236 int 237 rf_ConfigureSpareDisks(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr, 238 RF_Config_t *cfgPtr) 239 { 240 int i, ret; 241 unsigned int bs; 242 RF_RaidDisk_t *disks; 243 int num_spares_done; 244 245 num_spares_done = 0; 246 247 /* The space for the spares should have already been allocated by 248 * ConfigureDisks() */ 249 250 disks = &raidPtr->Disks[raidPtr->numCol]; 251 for (i = 0; i < raidPtr->numSpare; i++) { 252 ret = rf_ConfigureDisk(raidPtr, &cfgPtr->spare_names[i][0], 253 &disks[i], raidPtr->numCol + i); 254 if (ret) 255 goto fail; 256 if (disks[i].status != rf_ds_optimal) { 257 RF_ERRORMSG1("Warning: spare disk %s failed TUR\n", 258 &cfgPtr->spare_names[i][0]); 259 } else { 260 disks[i].status = rf_ds_spare; /* change status to 261 * spare */ 262 DPRINTF6("Spare Disk %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", i, 263 disks[i].devname, 264 (long int) disks[i].numBlocks, disks[i].blockSize, 265 (long int) disks[i].numBlocks * 266 disks[i].blockSize / 1024 / 1024); 267 } 268 num_spares_done++; 269 } 270 271 /* check sizes and block sizes on spare disks */ 272 bs = 1 << raidPtr->logBytesPerSector; 273 for (i = 0; i < raidPtr->numSpare; i++) { 274 if (disks[i].blockSize != bs) { 275 RF_ERRORMSG3("Block size of %d on spare disk %s is not the same as on other disks (%d)\n", disks[i].blockSize, disks[i].devname, bs); 276 ret = EINVAL; 277 goto fail; 278 } 279 if (disks[i].numBlocks < raidPtr->sectorsPerDisk) { 280 RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %ld blocks)\n", 281 disks[i].devname, disks[i].blockSize, 282 (long int) raidPtr->sectorsPerDisk); 283 ret = EINVAL; 284 goto fail; 285 } else 286 if (disks[i].numBlocks > raidPtr->sectorsPerDisk) { 287 RF_ERRORMSG3("Warning: truncating spare disk %s to %ld blocks (from %ld)\n", 288 disks[i].devname, 289 (long int) raidPtr->sectorsPerDisk, 290 (long int) disks[i].numBlocks); 291 292 disks[i].numBlocks = raidPtr->sectorsPerDisk; 293 } 294 } 295 296 return (0); 297 298 fail: 299 300 /* Release the hold on the main components. We've failed to allocate 301 * a spare, and since we're failing, we need to free things.. 302 303 XXX failing to allocate a spare is *not* that big of a deal... 304 We *can* survive without it, if need be, esp. if we get hot 305 adding working. 306 307 If we don't fail out here, then we need a way to remove this spare... 308 that should be easier to do here than if we are "live"... 309 310 */ 311 312 rf_UnconfigureVnodes( raidPtr ); 313 314 return (ret); 315 } 316 317 static int 318 rf_AllocDiskStructures(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr) 319 { 320 int ret; 321 322 /* We allocate RF_MAXSPARE on the first row so that we 323 have room to do hot-swapping of spares */ 324 RF_MallocAndAdd(raidPtr->Disks, (raidPtr->numCol + RF_MAXSPARE) * 325 sizeof(RF_RaidDisk_t), (RF_RaidDisk_t *), 326 raidPtr->cleanupList); 327 if (raidPtr->Disks == NULL) { 328 ret = ENOMEM; 329 goto fail; 330 } 331 332 /* get space for device specific stuff.. */ 333 RF_MallocAndAdd(raidPtr->raid_cinfo, 334 (raidPtr->numCol + RF_MAXSPARE) * 335 sizeof(struct raidcinfo), (struct raidcinfo *), 336 raidPtr->cleanupList); 337 338 if (raidPtr->raid_cinfo == NULL) { 339 ret = ENOMEM; 340 goto fail; 341 } 342 343 return(0); 344 fail: 345 rf_UnconfigureVnodes( raidPtr ); 346 347 return(ret); 348 } 349 350 351 /* configure a single disk during auto-configuration at boot */ 352 int 353 rf_AutoConfigureDisks(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr, 354 RF_AutoConfig_t *auto_config) 355 { 356 RF_RaidDisk_t *disks; 357 RF_RaidDisk_t *diskPtr; 358 RF_RowCol_t c; 359 RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL; 360 int bs, ret; 361 int numFailuresThisRow; 362 RF_AutoConfig_t *ac; 363 int parity_good; 364 int mod_counter; 365 int mod_counter_found; 366 367 #if DEBUG 368 printf("Starting autoconfiguration of RAID set...\n"); 369 #endif 370 371 ret = rf_AllocDiskStructures(raidPtr, cfgPtr); 372 if (ret) 373 goto fail; 374 375 disks = raidPtr->Disks; 376 377 /* assume the parity will be fine.. */ 378 parity_good = RF_RAID_CLEAN; 379 380 /* Check for mod_counters that are too low */ 381 mod_counter_found = 0; 382 mod_counter = 0; 383 ac = auto_config; 384 while(ac!=NULL) { 385 if (mod_counter_found==0) { 386 mod_counter = ac->clabel->mod_counter; 387 mod_counter_found = 1; 388 } else { 389 if (ac->clabel->mod_counter > mod_counter) { 390 mod_counter = ac->clabel->mod_counter; 391 } 392 } 393 ac->flag = 0; /* clear the general purpose flag */ 394 ac = ac->next; 395 } 396 397 bs = 0; 398 399 numFailuresThisRow = 0; 400 for (c = 0; c < raidPtr->numCol; c++) { 401 diskPtr = &disks[c]; 402 403 /* find this row/col in the autoconfig */ 404 #if DEBUG 405 printf("Looking for %d in autoconfig\n",c); 406 #endif 407 ac = auto_config; 408 while(ac!=NULL) { 409 if (ac->clabel==NULL) { 410 /* big-time bad news. */ 411 goto fail; 412 } 413 if ((ac->clabel->column == c) && 414 (ac->clabel->mod_counter == mod_counter)) { 415 /* it's this one... */ 416 /* flag it as 'used', so we don't 417 free it later. */ 418 ac->flag = 1; 419 #if DEBUG 420 printf("Found: %s at %d\n", 421 ac->devname,c); 422 #endif 423 424 break; 425 } 426 ac=ac->next; 427 } 428 429 if (ac==NULL) { 430 /* we didn't find an exact match with a 431 correct mod_counter above... can we find 432 one with an incorrect mod_counter to use 433 instead? (this one, if we find it, will be 434 marked as failed once the set configures) 435 */ 436 437 ac = auto_config; 438 while(ac!=NULL) { 439 if (ac->clabel==NULL) { 440 /* big-time bad news. */ 441 goto fail; 442 } 443 if (ac->clabel->column == c) { 444 /* it's this one... 445 flag it as 'used', so we 446 don't free it later. */ 447 ac->flag = 1; 448 #if DEBUG 449 printf("Found(low mod_counter): %s at %d\n", 450 ac->devname,c); 451 #endif 452 453 break; 454 } 455 ac=ac->next; 456 } 457 } 458 459 460 461 if (ac!=NULL) { 462 /* Found it. Configure it.. */ 463 diskPtr->blockSize = ac->clabel->blockSize; 464 diskPtr->numBlocks = ac->clabel->numBlocks; 465 /* Note: rf_protectedSectors is already 466 factored into numBlocks here */ 467 raidPtr->raid_cinfo[c].ci_vp = ac->vp; 468 raidPtr->raid_cinfo[c].ci_dev = ac->dev; 469 470 memcpy(&raidPtr->raid_cinfo[c].ci_label, 471 ac->clabel, sizeof(*ac->clabel)); 472 snprintf(diskPtr->devname, sizeof(diskPtr->devname), 473 "/dev/%s", ac->devname); 474 475 /* note the fact that this component was 476 autoconfigured. You'll need this info 477 later. Trust me :) */ 478 diskPtr->auto_configured = 1; 479 diskPtr->dev = ac->dev; 480 481 /* 482 * we allow the user to specify that 483 * only a fraction of the disks should 484 * be used this is just for debug: it 485 * speeds up the parity scan 486 */ 487 488 diskPtr->numBlocks = diskPtr->numBlocks * 489 rf_sizePercentage / 100; 490 491 /* XXX these will get set multiple times, 492 but since we're autoconfiguring, they'd 493 better be always the same each time! 494 If not, this is the least of your worries */ 495 496 bs = diskPtr->blockSize; 497 min_numblks = diskPtr->numBlocks; 498 499 /* this gets done multiple times, but that's 500 fine -- the serial number will be the same 501 for all components, guaranteed */ 502 raidPtr->serial_number = ac->clabel->serial_number; 503 /* check the last time the label was modified */ 504 505 if (ac->clabel->mod_counter != mod_counter) { 506 /* Even though we've filled in all of 507 the above, we don't trust this 508 component since it's modification 509 counter is not in sync with the 510 rest, and we really consider it to 511 be failed. */ 512 disks[c].status = rf_ds_failed; 513 numFailuresThisRow++; 514 } else { 515 if (ac->clabel->clean != RF_RAID_CLEAN) { 516 parity_good = RF_RAID_DIRTY; 517 } 518 } 519 } else { 520 /* Didn't find it at all!! Component must 521 really be dead */ 522 disks[c].status = rf_ds_failed; 523 snprintf(disks[c].devname, sizeof(disks[c].devname), 524 "component%d", c); 525 numFailuresThisRow++; 526 } 527 } 528 /* XXX fix for n-fault tolerant */ 529 /* XXX this should probably check to see how many failures 530 we can handle for this configuration! */ 531 if (numFailuresThisRow > 0) { 532 raidPtr->status = rf_rs_degraded; 533 raidPtr->numFailures = numFailuresThisRow; 534 } 535 536 /* close the device for the ones that didn't get used */ 537 538 ac = auto_config; 539 while(ac!=NULL) { 540 if (ac->flag == 0) { 541 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 542 VOP_CLOSE(ac->vp, FREAD | FWRITE, NOCRED, 0); 543 vput(ac->vp); 544 ac->vp = NULL; 545 #if DEBUG 546 printf("Released %s from auto-config set.\n", 547 ac->devname); 548 #endif 549 } 550 ac = ac->next; 551 } 552 553 raidPtr->mod_counter = mod_counter; 554 555 /* note the state of the parity, if any */ 556 raidPtr->parity_good = parity_good; 557 raidPtr->sectorsPerDisk = min_numblks; 558 raidPtr->logBytesPerSector = ffs(bs) - 1; 559 raidPtr->bytesPerSector = bs; 560 raidPtr->sectorMask = bs - 1; 561 return (0); 562 563 fail: 564 565 rf_UnconfigureVnodes( raidPtr ); 566 567 return (ret); 568 569 } 570 571 /* configure a single disk in the array */ 572 int 573 rf_ConfigureDisk(RF_Raid_t *raidPtr, char *bf, RF_RaidDisk_t *diskPtr, 574 RF_RowCol_t col) 575 { 576 char *p; 577 struct partinfo dpart; 578 struct vnode *vp; 579 struct vattr va; 580 struct proc *proc; 581 int error; 582 583 p = rf_find_non_white(bf); 584 if (p[strlen(p) - 1] == '\n') { 585 /* strip off the newline */ 586 p[strlen(p) - 1] = '\0'; 587 } 588 (void) strcpy(diskPtr->devname, p); 589 590 proc = raidPtr->engine_thread; 591 592 /* Let's start by claiming the component is fine and well... */ 593 diskPtr->status = rf_ds_optimal; 594 595 raidPtr->raid_cinfo[col].ci_vp = NULL; 596 raidPtr->raid_cinfo[col].ci_dev = 0; 597 598 if (!strcmp("absent", diskPtr->devname)) { 599 printf("Ignoring missing component at column %d\n", col); 600 sprintf(diskPtr->devname, "component%d", col); 601 diskPtr->status = rf_ds_failed; 602 return (0); 603 } 604 605 error = raidlookup(diskPtr->devname, proc, &vp); 606 if (error) { 607 printf("raidlookup on device: %s failed!\n", diskPtr->devname); 608 if (error == ENXIO) { 609 /* the component isn't there... must be dead :-( */ 610 diskPtr->status = rf_ds_failed; 611 } else { 612 return (error); 613 } 614 } 615 if (diskPtr->status == rf_ds_optimal) { 616 617 if ((error = VOP_GETATTR(vp, &va, proc->p_ucred, proc)) != 0) { 618 return (error); 619 } 620 error = VOP_IOCTL(vp, DIOCGPART, &dpart, 621 FREAD, proc->p_ucred, proc); 622 if (error) { 623 return (error); 624 } 625 626 diskPtr->blockSize = dpart.disklab->d_secsize; 627 628 diskPtr->numBlocks = dpart.part->p_size - rf_protectedSectors; 629 diskPtr->partitionSize = dpart.part->p_size; 630 631 raidPtr->raid_cinfo[col].ci_vp = vp; 632 raidPtr->raid_cinfo[col].ci_dev = va.va_rdev; 633 634 /* This component was not automatically configured */ 635 diskPtr->auto_configured = 0; 636 diskPtr->dev = va.va_rdev; 637 638 /* we allow the user to specify that only a fraction of the 639 * disks should be used this is just for debug: it speeds up 640 * the parity scan */ 641 diskPtr->numBlocks = diskPtr->numBlocks * 642 rf_sizePercentage / 100; 643 } 644 return (0); 645 } 646 647 static void 648 rf_print_label_status(RF_Raid_t *raidPtr, int column, char *dev_name, 649 RF_ComponentLabel_t *ci_label) 650 { 651 652 printf("raid%d: Component %s being configured at col: %d\n", 653 raidPtr->raidid, dev_name, column ); 654 printf(" Column: %d Num Columns: %d\n", 655 ci_label->column, 656 ci_label->num_columns); 657 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 658 ci_label->version, ci_label->serial_number, 659 ci_label->mod_counter); 660 printf(" Clean: %s Status: %d\n", 661 ci_label->clean ? "Yes" : "No", ci_label->status ); 662 } 663 664 static int rf_check_label_vitals(RF_Raid_t *raidPtr, int row, int column, 665 char *dev_name, RF_ComponentLabel_t *ci_label, 666 int serial_number, int mod_counter) 667 { 668 int fatal_error = 0; 669 670 if (serial_number != ci_label->serial_number) { 671 printf("%s has a different serial number: %d %d\n", 672 dev_name, serial_number, ci_label->serial_number); 673 fatal_error = 1; 674 } 675 if (mod_counter != ci_label->mod_counter) { 676 printf("%s has a different modfication count: %d %d\n", 677 dev_name, mod_counter, ci_label->mod_counter); 678 } 679 680 if (row != ci_label->row) { 681 printf("Row out of alignment for: %s\n", dev_name); 682 fatal_error = 1; 683 } 684 if (column != ci_label->column) { 685 printf("Column out of alignment for: %s\n", dev_name); 686 fatal_error = 1; 687 } 688 if (raidPtr->numCol != ci_label->num_columns) { 689 printf("Number of columns do not match for: %s\n", dev_name); 690 fatal_error = 1; 691 } 692 if (ci_label->clean == 0) { 693 /* it's not clean, but that's not fatal */ 694 printf("%s is not clean!\n", dev_name); 695 } 696 return(fatal_error); 697 } 698 699 700 /* 701 702 rf_CheckLabels() - check all the component labels for consistency. 703 Return an error if there is anything major amiss. 704 705 */ 706 707 int 708 rf_CheckLabels(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr) 709 { 710 int c; 711 char *dev_name; 712 RF_ComponentLabel_t *ci_label; 713 int serial_number = 0; 714 int mod_number = 0; 715 int fatal_error = 0; 716 int mod_values[4]; 717 int mod_count[4]; 718 int ser_values[4]; 719 int ser_count[4]; 720 int num_ser; 721 int num_mod; 722 int i; 723 int found; 724 int hosed_column; 725 int too_fatal; 726 int parity_good; 727 int force; 728 729 hosed_column = -1; 730 too_fatal = 0; 731 force = cfgPtr->force; 732 733 /* 734 We're going to try to be a little intelligent here. If one 735 component's label is bogus, and we can identify that it's the 736 *only* one that's gone, we'll mark it as "failed" and allow 737 the configuration to proceed. This will be the *only* case 738 that we'll proceed if there would be (otherwise) fatal errors. 739 740 Basically we simply keep a count of how many components had 741 what serial number. If all but one agree, we simply mark 742 the disagreeing component as being failed, and allow 743 things to come up "normally". 744 745 We do this first for serial numbers, and then for "mod_counter". 746 747 */ 748 749 num_ser = 0; 750 num_mod = 0; 751 752 for (c = 0; c < raidPtr->numCol; c++) { 753 ci_label = &raidPtr->raid_cinfo[c].ci_label; 754 found=0; 755 for(i=0;i<num_ser;i++) { 756 if (ser_values[i] == ci_label->serial_number) { 757 ser_count[i]++; 758 found=1; 759 break; 760 } 761 } 762 if (!found) { 763 ser_values[num_ser] = ci_label->serial_number; 764 ser_count[num_ser] = 1; 765 num_ser++; 766 if (num_ser>2) { 767 fatal_error = 1; 768 break; 769 } 770 } 771 found=0; 772 for(i=0;i<num_mod;i++) { 773 if (mod_values[i] == ci_label->mod_counter) { 774 mod_count[i]++; 775 found=1; 776 break; 777 } 778 } 779 if (!found) { 780 mod_values[num_mod] = ci_label->mod_counter; 781 mod_count[num_mod] = 1; 782 num_mod++; 783 if (num_mod>2) { 784 fatal_error = 1; 785 break; 786 } 787 } 788 } 789 #if DEBUG 790 printf("raid%d: Summary of serial numbers:\n", raidPtr->raidid); 791 for(i=0;i<num_ser;i++) { 792 printf("%d %d\n", ser_values[i], ser_count[i]); 793 } 794 printf("raid%d: Summary of mod counters:\n", raidPtr->raidid); 795 for(i=0;i<num_mod;i++) { 796 printf("%d %d\n", mod_values[i], mod_count[i]); 797 } 798 #endif 799 serial_number = ser_values[0]; 800 if (num_ser == 2) { 801 if ((ser_count[0] == 1) || (ser_count[1] == 1)) { 802 /* Locate the maverick component */ 803 if (ser_count[1] > ser_count[0]) { 804 serial_number = ser_values[1]; 805 } 806 807 for (c = 0; c < raidPtr->numCol; c++) { 808 ci_label = &raidPtr->raid_cinfo[c].ci_label; 809 if (serial_number != ci_label->serial_number) { 810 hosed_column = c; 811 break; 812 } 813 } 814 printf("Hosed component: %s\n", 815 &cfgPtr->devnames[0][hosed_column][0]); 816 if (!force) { 817 /* we'll fail this component, as if there are 818 other major errors, we arn't forcing things 819 and we'll abort the config anyways */ 820 raidPtr->Disks[hosed_column].status 821 = rf_ds_failed; 822 raidPtr->numFailures++; 823 raidPtr->status = rf_rs_degraded; 824 } 825 } else { 826 too_fatal = 1; 827 } 828 if (cfgPtr->parityConfig == '0') { 829 /* We've identified two different serial numbers. 830 RAID 0 can't cope with that, so we'll punt */ 831 too_fatal = 1; 832 } 833 834 } 835 836 /* record the serial number for later. If we bail later, setting 837 this doesn't matter, otherwise we've got the best guess at the 838 correct serial number */ 839 raidPtr->serial_number = serial_number; 840 841 mod_number = mod_values[0]; 842 if (num_mod == 2) { 843 if ((mod_count[0] == 1) || (mod_count[1] == 1)) { 844 /* Locate the maverick component */ 845 if (mod_count[1] > mod_count[0]) { 846 mod_number = mod_values[1]; 847 } else if (mod_count[1] < mod_count[0]) { 848 mod_number = mod_values[0]; 849 } else { 850 /* counts of different modification values 851 are the same. Assume greater value is 852 the correct one, all other things 853 considered */ 854 if (mod_values[0] > mod_values[1]) { 855 mod_number = mod_values[0]; 856 } else { 857 mod_number = mod_values[1]; 858 } 859 860 } 861 862 for (c = 0; c < raidPtr->numCol; c++) { 863 ci_label = &raidPtr->raid_cinfo[c].ci_label; 864 if (mod_number != ci_label->mod_counter) { 865 if (hosed_column == c) { 866 /* same one. Can 867 deal with it. */ 868 } else { 869 hosed_column = c; 870 if (num_ser != 1) { 871 too_fatal = 1; 872 break; 873 } 874 } 875 } 876 } 877 printf("Hosed component: %s\n", 878 &cfgPtr->devnames[0][hosed_column][0]); 879 if (!force) { 880 /* we'll fail this component, as if there are 881 other major errors, we arn't forcing things 882 and we'll abort the config anyways */ 883 if (raidPtr->Disks[hosed_column].status != rf_ds_failed) { 884 raidPtr->Disks[hosed_column].status 885 = rf_ds_failed; 886 raidPtr->numFailures++; 887 raidPtr->status = rf_rs_degraded; 888 } 889 } 890 } else { 891 too_fatal = 1; 892 } 893 if (cfgPtr->parityConfig == '0') { 894 /* We've identified two different mod counters. 895 RAID 0 can't cope with that, so we'll punt */ 896 too_fatal = 1; 897 } 898 } 899 900 raidPtr->mod_counter = mod_number; 901 902 if (too_fatal) { 903 /* we've had both a serial number mismatch, and a mod_counter 904 mismatch -- and they involved two different components!! 905 Bail -- make things fail so that the user must force 906 the issue... */ 907 hosed_column = -1; 908 fatal_error = 1; 909 } 910 911 if (num_ser > 2) { 912 printf("raid%d: Too many different serial numbers!\n", 913 raidPtr->raidid); 914 fatal_error = 1; 915 } 916 917 if (num_mod > 2) { 918 printf("raid%d: Too many different mod counters!\n", 919 raidPtr->raidid); 920 fatal_error = 1; 921 } 922 923 /* we start by assuming the parity will be good, and flee from 924 that notion at the slightest sign of trouble */ 925 926 parity_good = RF_RAID_CLEAN; 927 928 for (c = 0; c < raidPtr->numCol; c++) { 929 dev_name = &cfgPtr->devnames[0][c][0]; 930 ci_label = &raidPtr->raid_cinfo[c].ci_label; 931 932 if (c == hosed_column) { 933 printf("raid%d: Ignoring %s\n", 934 raidPtr->raidid, dev_name); 935 } else { 936 rf_print_label_status( raidPtr, c, dev_name, ci_label); 937 if (rf_check_label_vitals( raidPtr, 0, c, 938 dev_name, ci_label, 939 serial_number, 940 mod_number )) { 941 fatal_error = 1; 942 } 943 if (ci_label->clean != RF_RAID_CLEAN) { 944 parity_good = RF_RAID_DIRTY; 945 } 946 } 947 } 948 949 if (fatal_error) { 950 parity_good = RF_RAID_DIRTY; 951 } 952 953 /* we note the state of the parity */ 954 raidPtr->parity_good = parity_good; 955 956 return(fatal_error); 957 } 958 959 int 960 rf_add_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr) 961 { 962 RF_RaidDisk_t *disks; 963 RF_DiskQueue_t *spareQueues; 964 int ret; 965 unsigned int bs; 966 int spare_number; 967 968 ret=0; 969 970 if (raidPtr->numSpare >= RF_MAXSPARE) { 971 RF_ERRORMSG1("Too many spares: %d\n", raidPtr->numSpare); 972 return(EINVAL); 973 } 974 975 RF_LOCK_MUTEX(raidPtr->mutex); 976 while (raidPtr->adding_hot_spare==1) { 977 ltsleep(&(raidPtr->adding_hot_spare), PRIBIO, "raidhs", 0, 978 &(raidPtr->mutex)); 979 } 980 raidPtr->adding_hot_spare=1; 981 RF_UNLOCK_MUTEX(raidPtr->mutex); 982 983 /* the beginning of the spares... */ 984 disks = &raidPtr->Disks[raidPtr->numCol]; 985 986 spare_number = raidPtr->numSpare; 987 988 ret = rf_ConfigureDisk(raidPtr, sparePtr->component_name, 989 &disks[spare_number], 990 raidPtr->numCol + spare_number); 991 992 if (ret) 993 goto fail; 994 if (disks[spare_number].status != rf_ds_optimal) { 995 RF_ERRORMSG1("Warning: spare disk %s failed TUR\n", 996 sparePtr->component_name); 997 rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0); 998 ret=EINVAL; 999 goto fail; 1000 } else { 1001 disks[spare_number].status = rf_ds_spare; 1002 DPRINTF6("Spare Disk %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", spare_number, 1003 disks[spare_number].devname, 1004 (long int) disks[spare_number].numBlocks, 1005 disks[spare_number].blockSize, 1006 (long int) disks[spare_number].numBlocks * 1007 disks[spare_number].blockSize / 1024 / 1024); 1008 } 1009 1010 1011 /* check sizes and block sizes on the spare disk */ 1012 bs = 1 << raidPtr->logBytesPerSector; 1013 if (disks[spare_number].blockSize != bs) { 1014 RF_ERRORMSG3("Block size of %d on spare disk %s is not the same as on other disks (%d)\n", disks[spare_number].blockSize, disks[spare_number].devname, bs); 1015 rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0); 1016 ret = EINVAL; 1017 goto fail; 1018 } 1019 if (disks[spare_number].numBlocks < raidPtr->sectorsPerDisk) { 1020 RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %ld blocks)\n", 1021 disks[spare_number].devname, 1022 disks[spare_number].blockSize, 1023 (long int) raidPtr->sectorsPerDisk); 1024 rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0); 1025 ret = EINVAL; 1026 goto fail; 1027 } else { 1028 if (disks[spare_number].numBlocks > 1029 raidPtr->sectorsPerDisk) { 1030 RF_ERRORMSG3("Warning: truncating spare disk %s to %ld blocks (from %ld)\n", 1031 disks[spare_number].devname, 1032 (long int) raidPtr->sectorsPerDisk, 1033 (long int) disks[spare_number].numBlocks); 1034 1035 disks[spare_number].numBlocks = raidPtr->sectorsPerDisk; 1036 } 1037 } 1038 1039 spareQueues = &raidPtr->Queues[raidPtr->numCol]; 1040 ret = rf_ConfigureDiskQueue( raidPtr, &spareQueues[spare_number], 1041 raidPtr->numCol + spare_number, 1042 raidPtr->qType, 1043 raidPtr->sectorsPerDisk, 1044 raidPtr->Disks[raidPtr->numCol + 1045 spare_number].dev, 1046 raidPtr->maxOutstanding, 1047 &raidPtr->shutdownList, 1048 raidPtr->cleanupList); 1049 1050 RF_LOCK_MUTEX(raidPtr->mutex); 1051 raidPtr->numSpare++; 1052 RF_UNLOCK_MUTEX(raidPtr->mutex); 1053 1054 fail: 1055 RF_LOCK_MUTEX(raidPtr->mutex); 1056 raidPtr->adding_hot_spare=0; 1057 wakeup(&(raidPtr->adding_hot_spare)); 1058 RF_UNLOCK_MUTEX(raidPtr->mutex); 1059 1060 return(ret); 1061 } 1062 1063 int 1064 rf_remove_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr) 1065 { 1066 int spare_number; 1067 1068 1069 if (raidPtr->numSpare==0) { 1070 printf("No spares to remove!\n"); 1071 return(EINVAL); 1072 } 1073 1074 spare_number = sparePtr->column; 1075 1076 return(EINVAL); /* XXX not implemented yet */ 1077 #if 0 1078 if (spare_number < 0 || spare_number > raidPtr->numSpare) { 1079 return(EINVAL); 1080 } 1081 1082 /* verify that this spare isn't in use... */ 1083 1084 1085 1086 1087 /* it's gone.. */ 1088 1089 raidPtr->numSpare--; 1090 1091 return(0); 1092 #endif 1093 } 1094 1095 1096 int 1097 rf_delete_component(RF_Raid_t *raidPtr, RF_SingleComponent_t *component) 1098 { 1099 RF_RaidDisk_t *disks; 1100 1101 if ((component->column < 0) || 1102 (component->column >= raidPtr->numCol)) { 1103 return(EINVAL); 1104 } 1105 1106 disks = &raidPtr->Disks[component->column]; 1107 1108 /* 1. This component must be marked as 'failed' */ 1109 1110 return(EINVAL); /* Not implemented yet. */ 1111 } 1112 1113 int 1114 rf_incorporate_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *component) 1115 { 1116 1117 /* Issues here include how to 'move' this in if there is IO 1118 taking place (e.g. component queues and such) */ 1119 1120 return(EINVAL); /* Not implemented yet. */ 1121 } 1122