1 /* $NetBSD: rf_netbsdkintf.c,v 1.151 2002/11/21 15:37:55 oster Exp $ */ 2 /*- 3 * Copyright (c) 1996, 1997, 1998 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; Jason R. Thorpe. 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) 1988 University of Utah. 40 * Copyright (c) 1990, 1993 41 * The Regents of the University of California. All rights reserved. 42 * 43 * This code is derived from software contributed to Berkeley by 44 * the Systems Programming Group of the University of Utah Computer 45 * Science Department. 46 * 47 * Redistribution and use in source and binary forms, with or without 48 * modification, are permitted provided that the following conditions 49 * are met: 50 * 1. Redistributions of source code must retain the above copyright 51 * notice, this list of conditions and the following disclaimer. 52 * 2. Redistributions in binary form must reproduce the above copyright 53 * notice, this list of conditions and the following disclaimer in the 54 * documentation and/or other materials provided with the distribution. 55 * 3. All advertising materials mentioning features or use of this software 56 * must display the following acknowledgement: 57 * This product includes software developed by the University of 58 * California, Berkeley and its contributors. 59 * 4. Neither the name of the University nor the names of its contributors 60 * may be used to endorse or promote products derived from this software 61 * without specific prior written permission. 62 * 63 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 64 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 65 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 66 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 67 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 68 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 69 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 70 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 71 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 72 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 73 * SUCH DAMAGE. 74 * 75 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 76 * 77 * @(#)cd.c 8.2 (Berkeley) 11/16/93 78 */ 79 80 81 82 83 /* 84 * Copyright (c) 1995 Carnegie-Mellon University. 85 * All rights reserved. 86 * 87 * Authors: Mark Holland, Jim Zelenka 88 * 89 * Permission to use, copy, modify and distribute this software and 90 * its documentation is hereby granted, provided that both the copyright 91 * notice and this permission notice appear in all copies of the 92 * software, derivative works or modified versions, and any portions 93 * thereof, and that both notices appear in supporting documentation. 94 * 95 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 96 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 97 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 98 * 99 * Carnegie Mellon requests users of this software to return to 100 * 101 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 102 * School of Computer Science 103 * Carnegie Mellon University 104 * Pittsburgh PA 15213-3890 105 * 106 * any improvements or extensions that they make and grant Carnegie the 107 * rights to redistribute these changes. 108 */ 109 110 /*********************************************************** 111 * 112 * rf_kintf.c -- the kernel interface routines for RAIDframe 113 * 114 ***********************************************************/ 115 116 #include <sys/cdefs.h> 117 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.151 2002/11/21 15:37:55 oster Exp $"); 118 119 #include <sys/param.h> 120 #include <sys/errno.h> 121 #include <sys/pool.h> 122 #include <sys/queue.h> 123 #include <sys/disk.h> 124 #include <sys/device.h> 125 #include <sys/stat.h> 126 #include <sys/ioctl.h> 127 #include <sys/fcntl.h> 128 #include <sys/systm.h> 129 #include <sys/namei.h> 130 #include <sys/vnode.h> 131 #include <sys/disklabel.h> 132 #include <sys/conf.h> 133 #include <sys/lock.h> 134 #include <sys/buf.h> 135 #include <sys/user.h> 136 #include <sys/reboot.h> 137 138 #include <dev/raidframe/raidframevar.h> 139 #include <dev/raidframe/raidframeio.h> 140 #include "raid.h" 141 #include "opt_raid_autoconfig.h" 142 #include "rf_raid.h" 143 #include "rf_copyback.h" 144 #include "rf_dag.h" 145 #include "rf_dagflags.h" 146 #include "rf_desc.h" 147 #include "rf_diskqueue.h" 148 #include "rf_etimer.h" 149 #include "rf_general.h" 150 #include "rf_kintf.h" 151 #include "rf_options.h" 152 #include "rf_driver.h" 153 #include "rf_parityscan.h" 154 #include "rf_threadstuff.h" 155 156 #ifdef DEBUG 157 int rf_kdebug_level = 0; 158 #define db1_printf(a) if (rf_kdebug_level > 0) printf a 159 #else /* DEBUG */ 160 #define db1_printf(a) { } 161 #endif /* DEBUG */ 162 163 static RF_Raid_t **raidPtrs; /* global raid device descriptors */ 164 165 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex) 166 167 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a 168 * spare table */ 169 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from 170 * installation process */ 171 172 /* prototypes */ 173 static void KernelWakeupFunc(struct buf * bp); 174 static void InitBP(struct buf * bp, struct vnode *, unsigned rw_flag, 175 dev_t dev, RF_SectorNum_t startSect, 176 RF_SectorCount_t numSect, caddr_t buf, 177 void (*cbFunc) (struct buf *), void *cbArg, 178 int logBytesPerSector, struct proc * b_proc); 179 static void raidinit(RF_Raid_t *); 180 181 void raidattach(int); 182 183 dev_type_open(raidopen); 184 dev_type_close(raidclose); 185 dev_type_read(raidread); 186 dev_type_write(raidwrite); 187 dev_type_ioctl(raidioctl); 188 dev_type_strategy(raidstrategy); 189 dev_type_dump(raiddump); 190 dev_type_size(raidsize); 191 192 const struct bdevsw raid_bdevsw = { 193 raidopen, raidclose, raidstrategy, raidioctl, 194 raiddump, raidsize, D_DISK 195 }; 196 197 const struct cdevsw raid_cdevsw = { 198 raidopen, raidclose, raidread, raidwrite, raidioctl, 199 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 200 }; 201 202 /* 203 * Pilfered from ccd.c 204 */ 205 206 struct raidbuf { 207 struct buf rf_buf; /* new I/O buf. MUST BE FIRST!!! */ 208 struct buf *rf_obp; /* ptr. to original I/O buf */ 209 RF_DiskQueueData_t *req;/* the request that this was part of.. */ 210 }; 211 212 /* component buffer pool */ 213 struct pool raidframe_cbufpool; 214 215 /* XXX Not sure if the following should be replacing the raidPtrs above, 216 or if it should be used in conjunction with that... 217 */ 218 219 struct raid_softc { 220 int sc_flags; /* flags */ 221 int sc_cflags; /* configuration flags */ 222 size_t sc_size; /* size of the raid device */ 223 char sc_xname[20]; /* XXX external name */ 224 struct disk sc_dkdev; /* generic disk device info */ 225 struct bufq_state buf_queue; /* used for the device queue */ 226 }; 227 /* sc_flags */ 228 #define RAIDF_INITED 0x01 /* unit has been initialized */ 229 #define RAIDF_WLABEL 0x02 /* label area is writable */ 230 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 231 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 232 #define RAIDF_LOCKED 0x80 /* unit is locked */ 233 234 #define raidunit(x) DISKUNIT(x) 235 int numraid = 0; 236 237 /* 238 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 239 * Be aware that large numbers can allow the driver to consume a lot of 240 * kernel memory, especially on writes, and in degraded mode reads. 241 * 242 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 243 * a single 64K write will typically require 64K for the old data, 244 * 64K for the old parity, and 64K for the new parity, for a total 245 * of 192K (if the parity buffer is not re-used immediately). 246 * Even it if is used immediately, that's still 128K, which when multiplied 247 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 248 * 249 * Now in degraded mode, for example, a 64K read on the above setup may 250 * require data reconstruction, which will require *all* of the 4 remaining 251 * disks to participate -- 4 * 32K/disk == 128K again. 252 */ 253 254 #ifndef RAIDOUTSTANDING 255 #define RAIDOUTSTANDING 6 256 #endif 257 258 #define RAIDLABELDEV(dev) \ 259 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 260 261 /* declared here, and made public, for the benefit of KVM stuff.. */ 262 struct raid_softc *raid_softc; 263 264 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *, 265 struct disklabel *); 266 static void raidgetdisklabel(dev_t); 267 static void raidmakedisklabel(struct raid_softc *); 268 269 static int raidlock(struct raid_softc *); 270 static void raidunlock(struct raid_softc *); 271 272 static void rf_markalldirty(RF_Raid_t *); 273 274 struct device *raidrootdev; 275 276 void rf_ReconThread(struct rf_recon_req *); 277 /* XXX what I want is: */ 278 /*void rf_ReconThread(RF_Raid_t *raidPtr); */ 279 void rf_RewriteParityThread(RF_Raid_t *raidPtr); 280 void rf_CopybackThread(RF_Raid_t *raidPtr); 281 void rf_ReconstructInPlaceThread(struct rf_recon_req *); 282 int rf_autoconfig(struct device *self); 283 void rf_buildroothack(RF_ConfigSet_t *); 284 285 RF_AutoConfig_t *rf_find_raid_components(void); 286 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *); 287 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *); 288 static int rf_reasonable_label(RF_ComponentLabel_t *); 289 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *); 290 int rf_set_autoconfig(RF_Raid_t *, int); 291 int rf_set_rootpartition(RF_Raid_t *, int); 292 void rf_release_all_vps(RF_ConfigSet_t *); 293 void rf_cleanup_config_set(RF_ConfigSet_t *); 294 int rf_have_enough_components(RF_ConfigSet_t *); 295 int rf_auto_config_set(RF_ConfigSet_t *, int *); 296 297 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not 298 allow autoconfig to take place. 299 Note that this is overridden by having 300 RAID_AUTOCONFIG as an option in the 301 kernel config file. */ 302 303 void 304 raidattach(num) 305 int num; 306 { 307 int raidID; 308 int i, rc; 309 310 #ifdef DEBUG 311 printf("raidattach: Asked for %d units\n", num); 312 #endif 313 314 if (num <= 0) { 315 #ifdef DIAGNOSTIC 316 panic("raidattach: count <= 0"); 317 #endif 318 return; 319 } 320 /* This is where all the initialization stuff gets done. */ 321 322 numraid = num; 323 324 /* Make some space for requested number of units... */ 325 326 RF_Calloc(raidPtrs, num, sizeof(RF_Raid_t *), (RF_Raid_t **)); 327 if (raidPtrs == NULL) { 328 panic("raidPtrs is NULL!!"); 329 } 330 331 /* Initialize the component buffer pool. */ 332 pool_init(&raidframe_cbufpool, sizeof(struct raidbuf), 0, 333 0, 0, "raidpl", NULL); 334 335 rc = rf_mutex_init(&rf_sparet_wait_mutex); 336 if (rc) { 337 RF_PANIC(); 338 } 339 340 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 341 342 for (i = 0; i < num; i++) 343 raidPtrs[i] = NULL; 344 rc = rf_BootRaidframe(); 345 if (rc == 0) 346 printf("Kernelized RAIDframe activated\n"); 347 else 348 panic("Serious error booting RAID!!"); 349 350 /* put together some datastructures like the CCD device does.. This 351 * lets us lock the device and what-not when it gets opened. */ 352 353 raid_softc = (struct raid_softc *) 354 malloc(num * sizeof(struct raid_softc), 355 M_RAIDFRAME, M_NOWAIT); 356 if (raid_softc == NULL) { 357 printf("WARNING: no memory for RAIDframe driver\n"); 358 return; 359 } 360 361 memset(raid_softc, 0, num * sizeof(struct raid_softc)); 362 363 raidrootdev = (struct device *)malloc(num * sizeof(struct device), 364 M_RAIDFRAME, M_NOWAIT); 365 if (raidrootdev == NULL) { 366 panic("No memory for RAIDframe driver!!?!?!"); 367 } 368 369 for (raidID = 0; raidID < num; raidID++) { 370 bufq_alloc(&raid_softc[raidID].buf_queue, BUFQ_FCFS); 371 372 raidrootdev[raidID].dv_class = DV_DISK; 373 raidrootdev[raidID].dv_cfdata = NULL; 374 raidrootdev[raidID].dv_unit = raidID; 375 raidrootdev[raidID].dv_parent = NULL; 376 raidrootdev[raidID].dv_flags = 0; 377 sprintf(raidrootdev[raidID].dv_xname,"raid%d",raidID); 378 379 RF_Calloc(raidPtrs[raidID], 1, sizeof(RF_Raid_t), 380 (RF_Raid_t *)); 381 if (raidPtrs[raidID] == NULL) { 382 printf("WARNING: raidPtrs[%d] is NULL\n", raidID); 383 numraid = raidID; 384 return; 385 } 386 } 387 388 #ifdef RAID_AUTOCONFIG 389 raidautoconfig = 1; 390 #endif 391 392 /* 393 * Register a finalizer which will be used to auto-config RAID 394 * sets once all real hardware devices have been found. 395 */ 396 if (config_finalize_register(NULL, rf_autoconfig) != 0) 397 printf("WARNING: unable to register RAIDframe finalizer\n"); 398 } 399 400 int 401 rf_autoconfig(struct device *self) 402 { 403 RF_AutoConfig_t *ac_list; 404 RF_ConfigSet_t *config_sets; 405 406 if (raidautoconfig == 0) 407 return (0); 408 409 /* XXX This code can only be run once. */ 410 raidautoconfig = 0; 411 412 /* 1. locate all RAID components on the system */ 413 #ifdef DEBUG 414 printf("Searching for RAID components...\n"); 415 #endif 416 ac_list = rf_find_raid_components(); 417 418 /* 2. Sort them into their respective sets. */ 419 config_sets = rf_create_auto_sets(ac_list); 420 421 /* 422 * 3. Evaluate each set andconfigure the valid ones. 423 * This gets done in rf_buildroothack(). 424 */ 425 rf_buildroothack(config_sets); 426 427 return (1); 428 } 429 430 void 431 rf_buildroothack(RF_ConfigSet_t *config_sets) 432 { 433 RF_ConfigSet_t *cset; 434 RF_ConfigSet_t *next_cset; 435 int retcode; 436 int raidID; 437 int rootID; 438 int num_root; 439 440 rootID = 0; 441 num_root = 0; 442 cset = config_sets; 443 while(cset != NULL ) { 444 next_cset = cset->next; 445 if (rf_have_enough_components(cset) && 446 cset->ac->clabel->autoconfigure==1) { 447 retcode = rf_auto_config_set(cset,&raidID); 448 if (!retcode) { 449 if (cset->rootable) { 450 rootID = raidID; 451 num_root++; 452 } 453 } else { 454 /* The autoconfig didn't work :( */ 455 #if DEBUG 456 printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 457 #endif 458 rf_release_all_vps(cset); 459 } 460 } else { 461 /* we're not autoconfiguring this set... 462 release the associated resources */ 463 rf_release_all_vps(cset); 464 } 465 /* cleanup */ 466 rf_cleanup_config_set(cset); 467 cset = next_cset; 468 } 469 470 /* we found something bootable... */ 471 472 if (num_root == 1) { 473 booted_device = &raidrootdev[rootID]; 474 } else if (num_root > 1) { 475 /* we can't guess.. require the user to answer... */ 476 boothowto |= RB_ASKNAME; 477 } 478 } 479 480 481 int 482 raidsize(dev) 483 dev_t dev; 484 { 485 struct raid_softc *rs; 486 struct disklabel *lp; 487 int part, unit, omask, size; 488 489 unit = raidunit(dev); 490 if (unit >= numraid) 491 return (-1); 492 rs = &raid_softc[unit]; 493 494 if ((rs->sc_flags & RAIDF_INITED) == 0) 495 return (-1); 496 497 part = DISKPART(dev); 498 omask = rs->sc_dkdev.dk_openmask & (1 << part); 499 lp = rs->sc_dkdev.dk_label; 500 501 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc)) 502 return (-1); 503 504 if (lp->d_partitions[part].p_fstype != FS_SWAP) 505 size = -1; 506 else 507 size = lp->d_partitions[part].p_size * 508 (lp->d_secsize / DEV_BSIZE); 509 510 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc)) 511 return (-1); 512 513 return (size); 514 515 } 516 517 int 518 raiddump(dev, blkno, va, size) 519 dev_t dev; 520 daddr_t blkno; 521 caddr_t va; 522 size_t size; 523 { 524 /* Not implemented. */ 525 return ENXIO; 526 } 527 /* ARGSUSED */ 528 int 529 raidopen(dev, flags, fmt, p) 530 dev_t dev; 531 int flags, fmt; 532 struct proc *p; 533 { 534 int unit = raidunit(dev); 535 struct raid_softc *rs; 536 struct disklabel *lp; 537 int part, pmask; 538 int error = 0; 539 540 if (unit >= numraid) 541 return (ENXIO); 542 rs = &raid_softc[unit]; 543 544 if ((error = raidlock(rs)) != 0) 545 return (error); 546 lp = rs->sc_dkdev.dk_label; 547 548 part = DISKPART(dev); 549 pmask = (1 << part); 550 551 if ((rs->sc_flags & RAIDF_INITED) && 552 (rs->sc_dkdev.dk_openmask == 0)) 553 raidgetdisklabel(dev); 554 555 /* make sure that this partition exists */ 556 557 if (part != RAW_PART) { 558 if (((rs->sc_flags & RAIDF_INITED) == 0) || 559 ((part >= lp->d_npartitions) || 560 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 561 error = ENXIO; 562 raidunlock(rs); 563 return (error); 564 } 565 } 566 /* Prevent this unit from being unconfigured while open. */ 567 switch (fmt) { 568 case S_IFCHR: 569 rs->sc_dkdev.dk_copenmask |= pmask; 570 break; 571 572 case S_IFBLK: 573 rs->sc_dkdev.dk_bopenmask |= pmask; 574 break; 575 } 576 577 if ((rs->sc_dkdev.dk_openmask == 0) && 578 ((rs->sc_flags & RAIDF_INITED) != 0)) { 579 /* First one... mark things as dirty... Note that we *MUST* 580 have done a configure before this. I DO NOT WANT TO BE 581 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 582 THAT THEY BELONG TOGETHER!!!!! */ 583 /* XXX should check to see if we're only open for reading 584 here... If so, we needn't do this, but then need some 585 other way of keeping track of what's happened.. */ 586 587 rf_markalldirty( raidPtrs[unit] ); 588 } 589 590 591 rs->sc_dkdev.dk_openmask = 592 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 593 594 raidunlock(rs); 595 596 return (error); 597 598 599 } 600 /* ARGSUSED */ 601 int 602 raidclose(dev, flags, fmt, p) 603 dev_t dev; 604 int flags, fmt; 605 struct proc *p; 606 { 607 int unit = raidunit(dev); 608 struct raid_softc *rs; 609 int error = 0; 610 int part; 611 612 if (unit >= numraid) 613 return (ENXIO); 614 rs = &raid_softc[unit]; 615 616 if ((error = raidlock(rs)) != 0) 617 return (error); 618 619 part = DISKPART(dev); 620 621 /* ...that much closer to allowing unconfiguration... */ 622 switch (fmt) { 623 case S_IFCHR: 624 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 625 break; 626 627 case S_IFBLK: 628 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 629 break; 630 } 631 rs->sc_dkdev.dk_openmask = 632 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 633 634 if ((rs->sc_dkdev.dk_openmask == 0) && 635 ((rs->sc_flags & RAIDF_INITED) != 0)) { 636 /* Last one... device is not unconfigured yet. 637 Device shutdown has taken care of setting the 638 clean bits if RAIDF_INITED is not set 639 mark things as clean... */ 640 641 rf_update_component_labels(raidPtrs[unit], 642 RF_FINAL_COMPONENT_UPDATE); 643 if (doing_shutdown) { 644 /* last one, and we're going down, so 645 lights out for this RAID set too. */ 646 error = rf_Shutdown(raidPtrs[unit]); 647 648 /* It's no longer initialized... */ 649 rs->sc_flags &= ~RAIDF_INITED; 650 651 /* Detach the disk. */ 652 disk_detach(&rs->sc_dkdev); 653 } 654 } 655 656 raidunlock(rs); 657 return (0); 658 659 } 660 661 void 662 raidstrategy(bp) 663 struct buf *bp; 664 { 665 int s; 666 667 unsigned int raidID = raidunit(bp->b_dev); 668 RF_Raid_t *raidPtr; 669 struct raid_softc *rs = &raid_softc[raidID]; 670 struct disklabel *lp; 671 int wlabel; 672 673 if ((rs->sc_flags & RAIDF_INITED) ==0) { 674 bp->b_error = ENXIO; 675 bp->b_flags |= B_ERROR; 676 bp->b_resid = bp->b_bcount; 677 biodone(bp); 678 return; 679 } 680 if (raidID >= numraid || !raidPtrs[raidID]) { 681 bp->b_error = ENODEV; 682 bp->b_flags |= B_ERROR; 683 bp->b_resid = bp->b_bcount; 684 biodone(bp); 685 return; 686 } 687 raidPtr = raidPtrs[raidID]; 688 if (!raidPtr->valid) { 689 bp->b_error = ENODEV; 690 bp->b_flags |= B_ERROR; 691 bp->b_resid = bp->b_bcount; 692 biodone(bp); 693 return; 694 } 695 if (bp->b_bcount == 0) { 696 db1_printf(("b_bcount is zero..\n")); 697 biodone(bp); 698 return; 699 } 700 lp = rs->sc_dkdev.dk_label; 701 702 /* 703 * Do bounds checking and adjust transfer. If there's an 704 * error, the bounds check will flag that for us. 705 */ 706 707 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 708 if (DISKPART(bp->b_dev) != RAW_PART) 709 if (bounds_check_with_label(bp, lp, wlabel) <= 0) { 710 db1_printf(("Bounds check failed!!:%d %d\n", 711 (int) bp->b_blkno, (int) wlabel)); 712 biodone(bp); 713 return; 714 } 715 s = splbio(); 716 717 bp->b_resid = 0; 718 719 /* stuff it onto our queue */ 720 BUFQ_PUT(&rs->buf_queue, bp); 721 722 raidstart(raidPtrs[raidID]); 723 724 splx(s); 725 } 726 /* ARGSUSED */ 727 int 728 raidread(dev, uio, flags) 729 dev_t dev; 730 struct uio *uio; 731 int flags; 732 { 733 int unit = raidunit(dev); 734 struct raid_softc *rs; 735 int part; 736 737 if (unit >= numraid) 738 return (ENXIO); 739 rs = &raid_softc[unit]; 740 741 if ((rs->sc_flags & RAIDF_INITED) == 0) 742 return (ENXIO); 743 part = DISKPART(dev); 744 745 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio)); 746 747 } 748 /* ARGSUSED */ 749 int 750 raidwrite(dev, uio, flags) 751 dev_t dev; 752 struct uio *uio; 753 int flags; 754 { 755 int unit = raidunit(dev); 756 struct raid_softc *rs; 757 758 if (unit >= numraid) 759 return (ENXIO); 760 rs = &raid_softc[unit]; 761 762 if ((rs->sc_flags & RAIDF_INITED) == 0) 763 return (ENXIO); 764 765 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio)); 766 767 } 768 769 int 770 raidioctl(dev, cmd, data, flag, p) 771 dev_t dev; 772 u_long cmd; 773 caddr_t data; 774 int flag; 775 struct proc *p; 776 { 777 int unit = raidunit(dev); 778 int error = 0; 779 int part, pmask; 780 struct raid_softc *rs; 781 RF_Config_t *k_cfg, *u_cfg; 782 RF_Raid_t *raidPtr; 783 RF_RaidDisk_t *diskPtr; 784 RF_AccTotals_t *totals; 785 RF_DeviceConfig_t *d_cfg, **ucfgp; 786 u_char *specific_buf; 787 int retcode = 0; 788 int row; 789 int column; 790 int raidid; 791 struct rf_recon_req *rrcopy, *rr; 792 RF_ComponentLabel_t *clabel; 793 RF_ComponentLabel_t ci_label; 794 RF_ComponentLabel_t **clabel_ptr; 795 RF_SingleComponent_t *sparePtr,*componentPtr; 796 RF_SingleComponent_t hot_spare; 797 RF_SingleComponent_t component; 798 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 799 int i, j, d; 800 #ifdef __HAVE_OLD_DISKLABEL 801 struct disklabel newlabel; 802 #endif 803 804 if (unit >= numraid) 805 return (ENXIO); 806 rs = &raid_softc[unit]; 807 raidPtr = raidPtrs[unit]; 808 809 db1_printf(("raidioctl: %d %d %d %d\n", (int) dev, 810 (int) DISKPART(dev), (int) unit, (int) cmd)); 811 812 /* Must be open for writes for these commands... */ 813 switch (cmd) { 814 case DIOCSDINFO: 815 case DIOCWDINFO: 816 #ifdef __HAVE_OLD_DISKLABEL 817 case ODIOCWDINFO: 818 case ODIOCSDINFO: 819 #endif 820 case DIOCWLABEL: 821 if ((flag & FWRITE) == 0) 822 return (EBADF); 823 } 824 825 /* Must be initialized for these... */ 826 switch (cmd) { 827 case DIOCGDINFO: 828 case DIOCSDINFO: 829 case DIOCWDINFO: 830 #ifdef __HAVE_OLD_DISKLABEL 831 case ODIOCGDINFO: 832 case ODIOCWDINFO: 833 case ODIOCSDINFO: 834 case ODIOCGDEFLABEL: 835 #endif 836 case DIOCGPART: 837 case DIOCWLABEL: 838 case DIOCGDEFLABEL: 839 case RAIDFRAME_SHUTDOWN: 840 case RAIDFRAME_REWRITEPARITY: 841 case RAIDFRAME_GET_INFO: 842 case RAIDFRAME_RESET_ACCTOTALS: 843 case RAIDFRAME_GET_ACCTOTALS: 844 case RAIDFRAME_KEEP_ACCTOTALS: 845 case RAIDFRAME_GET_SIZE: 846 case RAIDFRAME_FAIL_DISK: 847 case RAIDFRAME_COPYBACK: 848 case RAIDFRAME_CHECK_RECON_STATUS: 849 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 850 case RAIDFRAME_GET_COMPONENT_LABEL: 851 case RAIDFRAME_SET_COMPONENT_LABEL: 852 case RAIDFRAME_ADD_HOT_SPARE: 853 case RAIDFRAME_REMOVE_HOT_SPARE: 854 case RAIDFRAME_INIT_LABELS: 855 case RAIDFRAME_REBUILD_IN_PLACE: 856 case RAIDFRAME_CHECK_PARITY: 857 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 858 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 859 case RAIDFRAME_CHECK_COPYBACK_STATUS: 860 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 861 case RAIDFRAME_SET_AUTOCONFIG: 862 case RAIDFRAME_SET_ROOT: 863 case RAIDFRAME_DELETE_COMPONENT: 864 case RAIDFRAME_INCORPORATE_HOT_SPARE: 865 if ((rs->sc_flags & RAIDF_INITED) == 0) 866 return (ENXIO); 867 } 868 869 switch (cmd) { 870 871 /* configure the system */ 872 case RAIDFRAME_CONFIGURE: 873 874 if (raidPtr->valid) { 875 /* There is a valid RAID set running on this unit! */ 876 printf("raid%d: Device already configured!\n",unit); 877 return(EINVAL); 878 } 879 880 /* copy-in the configuration information */ 881 /* data points to a pointer to the configuration structure */ 882 883 u_cfg = *((RF_Config_t **) data); 884 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *)); 885 if (k_cfg == NULL) { 886 return (ENOMEM); 887 } 888 retcode = copyin((caddr_t) u_cfg, (caddr_t) k_cfg, 889 sizeof(RF_Config_t)); 890 if (retcode) { 891 RF_Free(k_cfg, sizeof(RF_Config_t)); 892 db1_printf(("rf_ioctl: retcode=%d copyin.1\n", 893 retcode)); 894 return (retcode); 895 } 896 /* allocate a buffer for the layout-specific data, and copy it 897 * in */ 898 if (k_cfg->layoutSpecificSize) { 899 if (k_cfg->layoutSpecificSize > 10000) { 900 /* sanity check */ 901 RF_Free(k_cfg, sizeof(RF_Config_t)); 902 return (EINVAL); 903 } 904 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize, 905 (u_char *)); 906 if (specific_buf == NULL) { 907 RF_Free(k_cfg, sizeof(RF_Config_t)); 908 return (ENOMEM); 909 } 910 retcode = copyin(k_cfg->layoutSpecific, 911 (caddr_t) specific_buf, 912 k_cfg->layoutSpecificSize); 913 if (retcode) { 914 RF_Free(k_cfg, sizeof(RF_Config_t)); 915 RF_Free(specific_buf, 916 k_cfg->layoutSpecificSize); 917 db1_printf(("rf_ioctl: retcode=%d copyin.2\n", 918 retcode)); 919 return (retcode); 920 } 921 } else 922 specific_buf = NULL; 923 k_cfg->layoutSpecific = specific_buf; 924 925 /* should do some kind of sanity check on the configuration. 926 * Store the sum of all the bytes in the last byte? */ 927 928 /* configure the system */ 929 930 /* 931 * Clear the entire RAID descriptor, just to make sure 932 * there is no stale data left in the case of a 933 * reconfiguration 934 */ 935 memset((char *) raidPtr, 0, sizeof(RF_Raid_t)); 936 raidPtr->raidid = unit; 937 938 retcode = rf_Configure(raidPtr, k_cfg, NULL); 939 940 if (retcode == 0) { 941 942 /* allow this many simultaneous IO's to 943 this RAID device */ 944 raidPtr->openings = RAIDOUTSTANDING; 945 946 raidinit(raidPtr); 947 rf_markalldirty(raidPtr); 948 } 949 /* free the buffers. No return code here. */ 950 if (k_cfg->layoutSpecificSize) { 951 RF_Free(specific_buf, k_cfg->layoutSpecificSize); 952 } 953 RF_Free(k_cfg, sizeof(RF_Config_t)); 954 955 return (retcode); 956 957 /* shutdown the system */ 958 case RAIDFRAME_SHUTDOWN: 959 960 if ((error = raidlock(rs)) != 0) 961 return (error); 962 963 /* 964 * If somebody has a partition mounted, we shouldn't 965 * shutdown. 966 */ 967 968 part = DISKPART(dev); 969 pmask = (1 << part); 970 if ((rs->sc_dkdev.dk_openmask & ~pmask) || 971 ((rs->sc_dkdev.dk_bopenmask & pmask) && 972 (rs->sc_dkdev.dk_copenmask & pmask))) { 973 raidunlock(rs); 974 return (EBUSY); 975 } 976 977 retcode = rf_Shutdown(raidPtr); 978 979 /* It's no longer initialized... */ 980 rs->sc_flags &= ~RAIDF_INITED; 981 982 /* Detach the disk. */ 983 disk_detach(&rs->sc_dkdev); 984 985 raidunlock(rs); 986 987 return (retcode); 988 case RAIDFRAME_GET_COMPONENT_LABEL: 989 clabel_ptr = (RF_ComponentLabel_t **) data; 990 /* need to read the component label for the disk indicated 991 by row,column in clabel */ 992 993 /* For practice, let's get it directly fromdisk, rather 994 than from the in-core copy */ 995 RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ), 996 (RF_ComponentLabel_t *)); 997 if (clabel == NULL) 998 return (ENOMEM); 999 1000 memset((char *) clabel, 0, sizeof(RF_ComponentLabel_t)); 1001 1002 retcode = copyin( *clabel_ptr, clabel, 1003 sizeof(RF_ComponentLabel_t)); 1004 1005 if (retcode) { 1006 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1007 return(retcode); 1008 } 1009 1010 row = clabel->row; 1011 column = clabel->column; 1012 1013 if ((row < 0) || (row >= raidPtr->numRow) || 1014 (column < 0) || (column >= raidPtr->numCol + 1015 raidPtr->numSpare)) { 1016 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1017 return(EINVAL); 1018 } 1019 1020 raidread_component_label(raidPtr->Disks[row][column].dev, 1021 raidPtr->raid_cinfo[row][column].ci_vp, 1022 clabel ); 1023 1024 retcode = copyout((caddr_t) clabel, 1025 (caddr_t) *clabel_ptr, 1026 sizeof(RF_ComponentLabel_t)); 1027 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1028 return (retcode); 1029 1030 case RAIDFRAME_SET_COMPONENT_LABEL: 1031 clabel = (RF_ComponentLabel_t *) data; 1032 1033 /* XXX check the label for valid stuff... */ 1034 /* Note that some things *should not* get modified -- 1035 the user should be re-initing the labels instead of 1036 trying to patch things. 1037 */ 1038 1039 raidid = raidPtr->raidid; 1040 printf("raid%d: Got component label:\n", raidid); 1041 printf("raid%d: Version: %d\n", raidid, clabel->version); 1042 printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number); 1043 printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter); 1044 printf("raid%d: Row: %d\n", raidid, clabel->row); 1045 printf("raid%d: Column: %d\n", raidid, clabel->column); 1046 printf("raid%d: Num Rows: %d\n", raidid, clabel->num_rows); 1047 printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns); 1048 printf("raid%d: Clean: %d\n", raidid, clabel->clean); 1049 printf("raid%d: Status: %d\n", raidid, clabel->status); 1050 1051 row = clabel->row; 1052 column = clabel->column; 1053 1054 if ((row < 0) || (row >= raidPtr->numRow) || 1055 (column < 0) || (column >= raidPtr->numCol)) { 1056 return(EINVAL); 1057 } 1058 1059 /* XXX this isn't allowed to do anything for now :-) */ 1060 1061 /* XXX and before it is, we need to fill in the rest 1062 of the fields!?!?!?! */ 1063 #if 0 1064 raidwrite_component_label( 1065 raidPtr->Disks[row][column].dev, 1066 raidPtr->raid_cinfo[row][column].ci_vp, 1067 clabel ); 1068 #endif 1069 return (0); 1070 1071 case RAIDFRAME_INIT_LABELS: 1072 clabel = (RF_ComponentLabel_t *) data; 1073 /* 1074 we only want the serial number from 1075 the above. We get all the rest of the information 1076 from the config that was used to create this RAID 1077 set. 1078 */ 1079 1080 raidPtr->serial_number = clabel->serial_number; 1081 1082 raid_init_component_label(raidPtr, &ci_label); 1083 ci_label.serial_number = clabel->serial_number; 1084 1085 for(row=0;row<raidPtr->numRow;row++) { 1086 ci_label.row = row; 1087 for(column=0;column<raidPtr->numCol;column++) { 1088 diskPtr = &raidPtr->Disks[row][column]; 1089 if (!RF_DEAD_DISK(diskPtr->status)) { 1090 ci_label.partitionSize = diskPtr->partitionSize; 1091 ci_label.column = column; 1092 raidwrite_component_label( 1093 raidPtr->Disks[row][column].dev, 1094 raidPtr->raid_cinfo[row][column].ci_vp, 1095 &ci_label ); 1096 } 1097 } 1098 } 1099 1100 return (retcode); 1101 case RAIDFRAME_SET_AUTOCONFIG: 1102 d = rf_set_autoconfig(raidPtr, *(int *) data); 1103 printf("raid%d: New autoconfig value is: %d\n", 1104 raidPtr->raidid, d); 1105 *(int *) data = d; 1106 return (retcode); 1107 1108 case RAIDFRAME_SET_ROOT: 1109 d = rf_set_rootpartition(raidPtr, *(int *) data); 1110 printf("raid%d: New rootpartition value is: %d\n", 1111 raidPtr->raidid, d); 1112 *(int *) data = d; 1113 return (retcode); 1114 1115 /* initialize all parity */ 1116 case RAIDFRAME_REWRITEPARITY: 1117 1118 if (raidPtr->Layout.map->faultsTolerated == 0) { 1119 /* Parity for RAID 0 is trivially correct */ 1120 raidPtr->parity_good = RF_RAID_CLEAN; 1121 return(0); 1122 } 1123 1124 if (raidPtr->parity_rewrite_in_progress == 1) { 1125 /* Re-write is already in progress! */ 1126 return(EINVAL); 1127 } 1128 1129 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1130 rf_RewriteParityThread, 1131 raidPtr,"raid_parity"); 1132 return (retcode); 1133 1134 1135 case RAIDFRAME_ADD_HOT_SPARE: 1136 sparePtr = (RF_SingleComponent_t *) data; 1137 memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t)); 1138 retcode = rf_add_hot_spare(raidPtr, &hot_spare); 1139 return(retcode); 1140 1141 case RAIDFRAME_REMOVE_HOT_SPARE: 1142 return(retcode); 1143 1144 case RAIDFRAME_DELETE_COMPONENT: 1145 componentPtr = (RF_SingleComponent_t *)data; 1146 memcpy( &component, componentPtr, 1147 sizeof(RF_SingleComponent_t)); 1148 retcode = rf_delete_component(raidPtr, &component); 1149 return(retcode); 1150 1151 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1152 componentPtr = (RF_SingleComponent_t *)data; 1153 memcpy( &component, componentPtr, 1154 sizeof(RF_SingleComponent_t)); 1155 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1156 return(retcode); 1157 1158 case RAIDFRAME_REBUILD_IN_PLACE: 1159 1160 if (raidPtr->Layout.map->faultsTolerated == 0) { 1161 /* Can't do this on a RAID 0!! */ 1162 return(EINVAL); 1163 } 1164 1165 if (raidPtr->recon_in_progress == 1) { 1166 /* a reconstruct is already in progress! */ 1167 return(EINVAL); 1168 } 1169 1170 componentPtr = (RF_SingleComponent_t *) data; 1171 memcpy( &component, componentPtr, 1172 sizeof(RF_SingleComponent_t)); 1173 row = component.row; 1174 column = component.column; 1175 1176 if ((row < 0) || (row >= raidPtr->numRow) || 1177 (column < 0) || (column >= raidPtr->numCol)) { 1178 return(EINVAL); 1179 } 1180 1181 RF_LOCK_MUTEX(raidPtr->mutex); 1182 if ((raidPtr->Disks[row][column].status == rf_ds_optimal) && 1183 (raidPtr->numFailures > 0)) { 1184 /* XXX 0 above shouldn't be constant!!! */ 1185 /* some component other than this has failed. 1186 Let's not make things worse than they already 1187 are... */ 1188 printf("raid%d: Unable to reconstruct to disk at:\n", 1189 raidPtr->raidid); 1190 printf("raid%d: Row: %d Col: %d Too many failures.\n", 1191 raidPtr->raidid, row, column); 1192 RF_UNLOCK_MUTEX(raidPtr->mutex); 1193 return (EINVAL); 1194 } 1195 if (raidPtr->Disks[row][column].status == 1196 rf_ds_reconstructing) { 1197 printf("raid%d: Unable to reconstruct to disk at:\n", 1198 raidPtr->raidid); 1199 printf("raid%d: Row: %d Col: %d Reconstruction already occuring!\n", raidPtr->raidid, row, column); 1200 1201 RF_UNLOCK_MUTEX(raidPtr->mutex); 1202 return (EINVAL); 1203 } 1204 if (raidPtr->Disks[row][column].status == rf_ds_spared) { 1205 RF_UNLOCK_MUTEX(raidPtr->mutex); 1206 return (EINVAL); 1207 } 1208 RF_UNLOCK_MUTEX(raidPtr->mutex); 1209 1210 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1211 if (rrcopy == NULL) 1212 return(ENOMEM); 1213 1214 rrcopy->raidPtr = (void *) raidPtr; 1215 rrcopy->row = row; 1216 rrcopy->col = column; 1217 1218 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1219 rf_ReconstructInPlaceThread, 1220 rrcopy,"raid_reconip"); 1221 return(retcode); 1222 1223 case RAIDFRAME_GET_INFO: 1224 if (!raidPtr->valid) 1225 return (ENODEV); 1226 ucfgp = (RF_DeviceConfig_t **) data; 1227 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1228 (RF_DeviceConfig_t *)); 1229 if (d_cfg == NULL) 1230 return (ENOMEM); 1231 memset((char *) d_cfg, 0, sizeof(RF_DeviceConfig_t)); 1232 d_cfg->rows = raidPtr->numRow; 1233 d_cfg->cols = raidPtr->numCol; 1234 d_cfg->ndevs = raidPtr->numRow * raidPtr->numCol; 1235 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1236 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1237 return (ENOMEM); 1238 } 1239 d_cfg->nspares = raidPtr->numSpare; 1240 if (d_cfg->nspares >= RF_MAX_DISKS) { 1241 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1242 return (ENOMEM); 1243 } 1244 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1245 d = 0; 1246 for (i = 0; i < d_cfg->rows; i++) { 1247 for (j = 0; j < d_cfg->cols; j++) { 1248 d_cfg->devs[d] = raidPtr->Disks[i][j]; 1249 d++; 1250 } 1251 } 1252 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1253 d_cfg->spares[i] = raidPtr->Disks[0][j]; 1254 } 1255 retcode = copyout((caddr_t) d_cfg, (caddr_t) * ucfgp, 1256 sizeof(RF_DeviceConfig_t)); 1257 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1258 1259 return (retcode); 1260 1261 case RAIDFRAME_CHECK_PARITY: 1262 *(int *) data = raidPtr->parity_good; 1263 return (0); 1264 1265 case RAIDFRAME_RESET_ACCTOTALS: 1266 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1267 return (0); 1268 1269 case RAIDFRAME_GET_ACCTOTALS: 1270 totals = (RF_AccTotals_t *) data; 1271 *totals = raidPtr->acc_totals; 1272 return (0); 1273 1274 case RAIDFRAME_KEEP_ACCTOTALS: 1275 raidPtr->keep_acc_totals = *(int *)data; 1276 return (0); 1277 1278 case RAIDFRAME_GET_SIZE: 1279 *(int *) data = raidPtr->totalSectors; 1280 return (0); 1281 1282 /* fail a disk & optionally start reconstruction */ 1283 case RAIDFRAME_FAIL_DISK: 1284 1285 if (raidPtr->Layout.map->faultsTolerated == 0) { 1286 /* Can't do this on a RAID 0!! */ 1287 return(EINVAL); 1288 } 1289 1290 rr = (struct rf_recon_req *) data; 1291 1292 if (rr->row < 0 || rr->row >= raidPtr->numRow 1293 || rr->col < 0 || rr->col >= raidPtr->numCol) 1294 return (EINVAL); 1295 1296 1297 RF_LOCK_MUTEX(raidPtr->mutex); 1298 if ((raidPtr->Disks[rr->row][rr->col].status == 1299 rf_ds_optimal) && (raidPtr->numFailures > 0)) { 1300 /* some other component has failed. Let's not make 1301 things worse. XXX wrong for RAID6 */ 1302 RF_UNLOCK_MUTEX(raidPtr->mutex); 1303 return (EINVAL); 1304 } 1305 if (raidPtr->Disks[rr->row][rr->col].status == rf_ds_spared) { 1306 /* Can't fail a spared disk! */ 1307 RF_UNLOCK_MUTEX(raidPtr->mutex); 1308 return (EINVAL); 1309 } 1310 RF_UNLOCK_MUTEX(raidPtr->mutex); 1311 1312 /* make a copy of the recon request so that we don't rely on 1313 * the user's buffer */ 1314 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1315 if (rrcopy == NULL) 1316 return(ENOMEM); 1317 memcpy(rrcopy, rr, sizeof(*rr)); 1318 rrcopy->raidPtr = (void *) raidPtr; 1319 1320 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1321 rf_ReconThread, 1322 rrcopy,"raid_recon"); 1323 return (0); 1324 1325 /* invoke a copyback operation after recon on whatever disk 1326 * needs it, if any */ 1327 case RAIDFRAME_COPYBACK: 1328 1329 if (raidPtr->Layout.map->faultsTolerated == 0) { 1330 /* This makes no sense on a RAID 0!! */ 1331 return(EINVAL); 1332 } 1333 1334 if (raidPtr->copyback_in_progress == 1) { 1335 /* Copyback is already in progress! */ 1336 return(EINVAL); 1337 } 1338 1339 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1340 rf_CopybackThread, 1341 raidPtr,"raid_copyback"); 1342 return (retcode); 1343 1344 /* return the percentage completion of reconstruction */ 1345 case RAIDFRAME_CHECK_RECON_STATUS: 1346 if (raidPtr->Layout.map->faultsTolerated == 0) { 1347 /* This makes no sense on a RAID 0, so tell the 1348 user it's done. */ 1349 *(int *) data = 100; 1350 return(0); 1351 } 1352 row = 0; /* XXX we only consider a single row... */ 1353 if (raidPtr->status[row] != rf_rs_reconstructing) 1354 *(int *) data = 100; 1355 else 1356 *(int *) data = raidPtr->reconControl[row]->percentComplete; 1357 return (0); 1358 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1359 progressInfoPtr = (RF_ProgressInfo_t **) data; 1360 row = 0; /* XXX we only consider a single row... */ 1361 if (raidPtr->status[row] != rf_rs_reconstructing) { 1362 progressInfo.remaining = 0; 1363 progressInfo.completed = 100; 1364 progressInfo.total = 100; 1365 } else { 1366 progressInfo.total = 1367 raidPtr->reconControl[row]->numRUsTotal; 1368 progressInfo.completed = 1369 raidPtr->reconControl[row]->numRUsComplete; 1370 progressInfo.remaining = progressInfo.total - 1371 progressInfo.completed; 1372 } 1373 retcode = copyout((caddr_t) &progressInfo, 1374 (caddr_t) *progressInfoPtr, 1375 sizeof(RF_ProgressInfo_t)); 1376 return (retcode); 1377 1378 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1379 if (raidPtr->Layout.map->faultsTolerated == 0) { 1380 /* This makes no sense on a RAID 0, so tell the 1381 user it's done. */ 1382 *(int *) data = 100; 1383 return(0); 1384 } 1385 if (raidPtr->parity_rewrite_in_progress == 1) { 1386 *(int *) data = 100 * 1387 raidPtr->parity_rewrite_stripes_done / 1388 raidPtr->Layout.numStripe; 1389 } else { 1390 *(int *) data = 100; 1391 } 1392 return (0); 1393 1394 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1395 progressInfoPtr = (RF_ProgressInfo_t **) data; 1396 if (raidPtr->parity_rewrite_in_progress == 1) { 1397 progressInfo.total = raidPtr->Layout.numStripe; 1398 progressInfo.completed = 1399 raidPtr->parity_rewrite_stripes_done; 1400 progressInfo.remaining = progressInfo.total - 1401 progressInfo.completed; 1402 } else { 1403 progressInfo.remaining = 0; 1404 progressInfo.completed = 100; 1405 progressInfo.total = 100; 1406 } 1407 retcode = copyout((caddr_t) &progressInfo, 1408 (caddr_t) *progressInfoPtr, 1409 sizeof(RF_ProgressInfo_t)); 1410 return (retcode); 1411 1412 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1413 if (raidPtr->Layout.map->faultsTolerated == 0) { 1414 /* This makes no sense on a RAID 0 */ 1415 *(int *) data = 100; 1416 return(0); 1417 } 1418 if (raidPtr->copyback_in_progress == 1) { 1419 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1420 raidPtr->Layout.numStripe; 1421 } else { 1422 *(int *) data = 100; 1423 } 1424 return (0); 1425 1426 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1427 progressInfoPtr = (RF_ProgressInfo_t **) data; 1428 if (raidPtr->copyback_in_progress == 1) { 1429 progressInfo.total = raidPtr->Layout.numStripe; 1430 progressInfo.completed = 1431 raidPtr->copyback_stripes_done; 1432 progressInfo.remaining = progressInfo.total - 1433 progressInfo.completed; 1434 } else { 1435 progressInfo.remaining = 0; 1436 progressInfo.completed = 100; 1437 progressInfo.total = 100; 1438 } 1439 retcode = copyout((caddr_t) &progressInfo, 1440 (caddr_t) *progressInfoPtr, 1441 sizeof(RF_ProgressInfo_t)); 1442 return (retcode); 1443 1444 /* the sparetable daemon calls this to wait for the kernel to 1445 * need a spare table. this ioctl does not return until a 1446 * spare table is needed. XXX -- calling mpsleep here in the 1447 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1448 * -- I should either compute the spare table in the kernel, 1449 * or have a different -- XXX XXX -- interface (a different 1450 * character device) for delivering the table -- XXX */ 1451 #if 0 1452 case RAIDFRAME_SPARET_WAIT: 1453 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1454 while (!rf_sparet_wait_queue) 1455 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE); 1456 waitreq = rf_sparet_wait_queue; 1457 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1458 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1459 1460 /* structure assignment */ 1461 *((RF_SparetWait_t *) data) = *waitreq; 1462 1463 RF_Free(waitreq, sizeof(*waitreq)); 1464 return (0); 1465 1466 /* wakes up a process waiting on SPARET_WAIT and puts an error 1467 * code in it that will cause the dameon to exit */ 1468 case RAIDFRAME_ABORT_SPARET_WAIT: 1469 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1470 waitreq->fcol = -1; 1471 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1472 waitreq->next = rf_sparet_wait_queue; 1473 rf_sparet_wait_queue = waitreq; 1474 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1475 wakeup(&rf_sparet_wait_queue); 1476 return (0); 1477 1478 /* used by the spare table daemon to deliver a spare table 1479 * into the kernel */ 1480 case RAIDFRAME_SEND_SPARET: 1481 1482 /* install the spare table */ 1483 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1484 1485 /* respond to the requestor. the return status of the spare 1486 * table installation is passed in the "fcol" field */ 1487 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1488 waitreq->fcol = retcode; 1489 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1490 waitreq->next = rf_sparet_resp_queue; 1491 rf_sparet_resp_queue = waitreq; 1492 wakeup(&rf_sparet_resp_queue); 1493 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1494 1495 return (retcode); 1496 #endif 1497 1498 default: 1499 break; /* fall through to the os-specific code below */ 1500 1501 } 1502 1503 if (!raidPtr->valid) 1504 return (EINVAL); 1505 1506 /* 1507 * Add support for "regular" device ioctls here. 1508 */ 1509 1510 switch (cmd) { 1511 case DIOCGDINFO: 1512 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1513 break; 1514 #ifdef __HAVE_OLD_DISKLABEL 1515 case ODIOCGDINFO: 1516 newlabel = *(rs->sc_dkdev.dk_label); 1517 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1518 return ENOTTY; 1519 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1520 break; 1521 #endif 1522 1523 case DIOCGPART: 1524 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1525 ((struct partinfo *) data)->part = 1526 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1527 break; 1528 1529 case DIOCWDINFO: 1530 case DIOCSDINFO: 1531 #ifdef __HAVE_OLD_DISKLABEL 1532 case ODIOCWDINFO: 1533 case ODIOCSDINFO: 1534 #endif 1535 { 1536 struct disklabel *lp; 1537 #ifdef __HAVE_OLD_DISKLABEL 1538 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1539 memset(&newlabel, 0, sizeof newlabel); 1540 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1541 lp = &newlabel; 1542 } else 1543 #endif 1544 lp = (struct disklabel *)data; 1545 1546 if ((error = raidlock(rs)) != 0) 1547 return (error); 1548 1549 rs->sc_flags |= RAIDF_LABELLING; 1550 1551 error = setdisklabel(rs->sc_dkdev.dk_label, 1552 lp, 0, rs->sc_dkdev.dk_cpulabel); 1553 if (error == 0) { 1554 if (cmd == DIOCWDINFO 1555 #ifdef __HAVE_OLD_DISKLABEL 1556 || cmd == ODIOCWDINFO 1557 #endif 1558 ) 1559 error = writedisklabel(RAIDLABELDEV(dev), 1560 raidstrategy, rs->sc_dkdev.dk_label, 1561 rs->sc_dkdev.dk_cpulabel); 1562 } 1563 rs->sc_flags &= ~RAIDF_LABELLING; 1564 1565 raidunlock(rs); 1566 1567 if (error) 1568 return (error); 1569 break; 1570 } 1571 1572 case DIOCWLABEL: 1573 if (*(int *) data != 0) 1574 rs->sc_flags |= RAIDF_WLABEL; 1575 else 1576 rs->sc_flags &= ~RAIDF_WLABEL; 1577 break; 1578 1579 case DIOCGDEFLABEL: 1580 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1581 break; 1582 1583 #ifdef __HAVE_OLD_DISKLABEL 1584 case ODIOCGDEFLABEL: 1585 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1586 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1587 return ENOTTY; 1588 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1589 break; 1590 #endif 1591 1592 default: 1593 retcode = ENOTTY; 1594 } 1595 return (retcode); 1596 1597 } 1598 1599 1600 /* raidinit -- complete the rest of the initialization for the 1601 RAIDframe device. */ 1602 1603 1604 static void 1605 raidinit(raidPtr) 1606 RF_Raid_t *raidPtr; 1607 { 1608 struct raid_softc *rs; 1609 int unit; 1610 1611 unit = raidPtr->raidid; 1612 1613 rs = &raid_softc[unit]; 1614 1615 /* XXX should check return code first... */ 1616 rs->sc_flags |= RAIDF_INITED; 1617 1618 sprintf(rs->sc_xname, "raid%d", unit); /* XXX doesn't check bounds. */ 1619 1620 rs->sc_dkdev.dk_name = rs->sc_xname; 1621 1622 /* disk_attach actually creates space for the CPU disklabel, among 1623 * other things, so it's critical to call this *BEFORE* we try putzing 1624 * with disklabels. */ 1625 1626 disk_attach(&rs->sc_dkdev); 1627 1628 /* XXX There may be a weird interaction here between this, and 1629 * protectedSectors, as used in RAIDframe. */ 1630 1631 rs->sc_size = raidPtr->totalSectors; 1632 1633 } 1634 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 1635 /* wake up the daemon & tell it to get us a spare table 1636 * XXX 1637 * the entries in the queues should be tagged with the raidPtr 1638 * so that in the extremely rare case that two recons happen at once, 1639 * we know for which device were requesting a spare table 1640 * XXX 1641 * 1642 * XXX This code is not currently used. GO 1643 */ 1644 int 1645 rf_GetSpareTableFromDaemon(req) 1646 RF_SparetWait_t *req; 1647 { 1648 int retcode; 1649 1650 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1651 req->next = rf_sparet_wait_queue; 1652 rf_sparet_wait_queue = req; 1653 wakeup(&rf_sparet_wait_queue); 1654 1655 /* mpsleep unlocks the mutex */ 1656 while (!rf_sparet_resp_queue) { 1657 tsleep(&rf_sparet_resp_queue, PRIBIO, 1658 "raidframe getsparetable", 0); 1659 } 1660 req = rf_sparet_resp_queue; 1661 rf_sparet_resp_queue = req->next; 1662 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1663 1664 retcode = req->fcol; 1665 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1666 * alloc'd */ 1667 return (retcode); 1668 } 1669 #endif 1670 1671 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1672 * bp & passes it down. 1673 * any calls originating in the kernel must use non-blocking I/O 1674 * do some extra sanity checking to return "appropriate" error values for 1675 * certain conditions (to make some standard utilities work) 1676 * 1677 * Formerly known as: rf_DoAccessKernel 1678 */ 1679 void 1680 raidstart(raidPtr) 1681 RF_Raid_t *raidPtr; 1682 { 1683 RF_SectorCount_t num_blocks, pb, sum; 1684 RF_RaidAddr_t raid_addr; 1685 struct partition *pp; 1686 daddr_t blocknum; 1687 int unit; 1688 struct raid_softc *rs; 1689 int do_async; 1690 struct buf *bp; 1691 1692 unit = raidPtr->raidid; 1693 rs = &raid_softc[unit]; 1694 1695 /* quick check to see if anything has died recently */ 1696 RF_LOCK_MUTEX(raidPtr->mutex); 1697 if (raidPtr->numNewFailures > 0) { 1698 RF_UNLOCK_MUTEX(raidPtr->mutex); 1699 rf_update_component_labels(raidPtr, 1700 RF_NORMAL_COMPONENT_UPDATE); 1701 RF_LOCK_MUTEX(raidPtr->mutex); 1702 raidPtr->numNewFailures--; 1703 } 1704 1705 /* Check to see if we're at the limit... */ 1706 while (raidPtr->openings > 0) { 1707 RF_UNLOCK_MUTEX(raidPtr->mutex); 1708 1709 /* get the next item, if any, from the queue */ 1710 if ((bp = BUFQ_GET(&rs->buf_queue)) == NULL) { 1711 /* nothing more to do */ 1712 return; 1713 } 1714 1715 /* Ok, for the bp we have here, bp->b_blkno is relative to the 1716 * partition.. Need to make it absolute to the underlying 1717 * device.. */ 1718 1719 blocknum = bp->b_blkno; 1720 if (DISKPART(bp->b_dev) != RAW_PART) { 1721 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 1722 blocknum += pp->p_offset; 1723 } 1724 1725 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 1726 (int) blocknum)); 1727 1728 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 1729 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 1730 1731 /* *THIS* is where we adjust what block we're going to... 1732 * but DO NOT TOUCH bp->b_blkno!!! */ 1733 raid_addr = blocknum; 1734 1735 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 1736 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 1737 sum = raid_addr + num_blocks + pb; 1738 if (1 || rf_debugKernelAccess) { 1739 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 1740 (int) raid_addr, (int) sum, (int) num_blocks, 1741 (int) pb, (int) bp->b_resid)); 1742 } 1743 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 1744 || (sum < num_blocks) || (sum < pb)) { 1745 bp->b_error = ENOSPC; 1746 bp->b_flags |= B_ERROR; 1747 bp->b_resid = bp->b_bcount; 1748 biodone(bp); 1749 RF_LOCK_MUTEX(raidPtr->mutex); 1750 continue; 1751 } 1752 /* 1753 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 1754 */ 1755 1756 if (bp->b_bcount & raidPtr->sectorMask) { 1757 bp->b_error = EINVAL; 1758 bp->b_flags |= B_ERROR; 1759 bp->b_resid = bp->b_bcount; 1760 biodone(bp); 1761 RF_LOCK_MUTEX(raidPtr->mutex); 1762 continue; 1763 1764 } 1765 db1_printf(("Calling DoAccess..\n")); 1766 1767 1768 RF_LOCK_MUTEX(raidPtr->mutex); 1769 raidPtr->openings--; 1770 RF_UNLOCK_MUTEX(raidPtr->mutex); 1771 1772 /* 1773 * Everything is async. 1774 */ 1775 do_async = 1; 1776 1777 disk_busy(&rs->sc_dkdev); 1778 1779 /* XXX we're still at splbio() here... do we *really* 1780 need to be? */ 1781 1782 /* don't ever condition on bp->b_flags & B_WRITE. 1783 * always condition on B_READ instead */ 1784 1785 bp->b_error = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 1786 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 1787 do_async, raid_addr, num_blocks, 1788 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 1789 1790 if (bp->b_error) { 1791 bp->b_flags |= B_ERROR; 1792 } 1793 1794 RF_LOCK_MUTEX(raidPtr->mutex); 1795 } 1796 RF_UNLOCK_MUTEX(raidPtr->mutex); 1797 } 1798 1799 1800 1801 1802 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 1803 1804 int 1805 rf_DispatchKernelIO(queue, req) 1806 RF_DiskQueue_t *queue; 1807 RF_DiskQueueData_t *req; 1808 { 1809 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 1810 struct buf *bp; 1811 struct raidbuf *raidbp = NULL; 1812 1813 req->queue = queue; 1814 1815 #if DIAGNOSTIC 1816 if (queue->raidPtr->raidid >= numraid) { 1817 printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid, 1818 numraid); 1819 panic("Invalid Unit number in rf_DispatchKernelIO"); 1820 } 1821 #endif 1822 1823 bp = req->bp; 1824 #if 1 1825 /* XXX when there is a physical disk failure, someone is passing us a 1826 * buffer that contains old stuff!! Attempt to deal with this problem 1827 * without taking a performance hit... (not sure where the real bug 1828 * is. It's buried in RAIDframe somewhere) :-( GO ) */ 1829 1830 if (bp->b_flags & B_ERROR) { 1831 bp->b_flags &= ~B_ERROR; 1832 } 1833 if (bp->b_error != 0) { 1834 bp->b_error = 0; 1835 } 1836 #endif 1837 raidbp = pool_get(&raidframe_cbufpool, PR_NOWAIT); 1838 1839 /* 1840 * context for raidiodone 1841 */ 1842 raidbp->rf_obp = bp; 1843 raidbp->req = req; 1844 1845 LIST_INIT(&raidbp->rf_buf.b_dep); 1846 1847 switch (req->type) { 1848 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 1849 /* XXX need to do something extra here.. */ 1850 /* I'm leaving this in, as I've never actually seen it used, 1851 * and I'd like folks to report it... GO */ 1852 printf(("WAKEUP CALLED\n")); 1853 queue->numOutstanding++; 1854 1855 /* XXX need to glue the original buffer into this?? */ 1856 1857 KernelWakeupFunc(&raidbp->rf_buf); 1858 break; 1859 1860 case RF_IO_TYPE_READ: 1861 case RF_IO_TYPE_WRITE: 1862 1863 if (req->tracerec) { 1864 RF_ETIMER_START(req->tracerec->timer); 1865 } 1866 InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp, 1867 op | bp->b_flags, queue->rf_cinfo->ci_dev, 1868 req->sectorOffset, req->numSector, 1869 req->buf, KernelWakeupFunc, (void *) req, 1870 queue->raidPtr->logBytesPerSector, req->b_proc); 1871 1872 if (rf_debugKernelAccess) { 1873 db1_printf(("dispatch: bp->b_blkno = %ld\n", 1874 (long) bp->b_blkno)); 1875 } 1876 queue->numOutstanding++; 1877 queue->last_deq_sector = req->sectorOffset; 1878 /* acc wouldn't have been let in if there were any pending 1879 * reqs at any other priority */ 1880 queue->curPriority = req->priority; 1881 1882 db1_printf(("Going for %c to unit %d row %d col %d\n", 1883 req->type, queue->raidPtr->raidid, 1884 queue->row, queue->col)); 1885 db1_printf(("sector %d count %d (%d bytes) %d\n", 1886 (int) req->sectorOffset, (int) req->numSector, 1887 (int) (req->numSector << 1888 queue->raidPtr->logBytesPerSector), 1889 (int) queue->raidPtr->logBytesPerSector)); 1890 if ((raidbp->rf_buf.b_flags & B_READ) == 0) { 1891 raidbp->rf_buf.b_vp->v_numoutput++; 1892 } 1893 VOP_STRATEGY(&raidbp->rf_buf); 1894 1895 break; 1896 1897 default: 1898 panic("bad req->type in rf_DispatchKernelIO"); 1899 } 1900 db1_printf(("Exiting from DispatchKernelIO\n")); 1901 1902 return (0); 1903 } 1904 /* this is the callback function associated with a I/O invoked from 1905 kernel code. 1906 */ 1907 static void 1908 KernelWakeupFunc(vbp) 1909 struct buf *vbp; 1910 { 1911 RF_DiskQueueData_t *req = NULL; 1912 RF_DiskQueue_t *queue; 1913 struct raidbuf *raidbp = (struct raidbuf *) vbp; 1914 struct buf *bp; 1915 int s; 1916 1917 s = splbio(); 1918 db1_printf(("recovering the request queue:\n")); 1919 req = raidbp->req; 1920 1921 bp = raidbp->rf_obp; 1922 1923 queue = (RF_DiskQueue_t *) req->queue; 1924 1925 if (raidbp->rf_buf.b_flags & B_ERROR) { 1926 bp->b_flags |= B_ERROR; 1927 bp->b_error = raidbp->rf_buf.b_error ? 1928 raidbp->rf_buf.b_error : EIO; 1929 } 1930 1931 /* XXX methinks this could be wrong... */ 1932 #if 1 1933 bp->b_resid = raidbp->rf_buf.b_resid; 1934 #endif 1935 1936 if (req->tracerec) { 1937 RF_ETIMER_STOP(req->tracerec->timer); 1938 RF_ETIMER_EVAL(req->tracerec->timer); 1939 RF_LOCK_MUTEX(rf_tracing_mutex); 1940 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 1941 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 1942 req->tracerec->num_phys_ios++; 1943 RF_UNLOCK_MUTEX(rf_tracing_mutex); 1944 } 1945 bp->b_bcount = raidbp->rf_buf.b_bcount; /* XXXX ?? */ 1946 1947 /* XXX Ok, let's get aggressive... If B_ERROR is set, let's go 1948 * ballistic, and mark the component as hosed... */ 1949 1950 if (bp->b_flags & B_ERROR) { 1951 /* Mark the disk as dead */ 1952 /* but only mark it once... */ 1953 if (queue->raidPtr->Disks[queue->row][queue->col].status == 1954 rf_ds_optimal) { 1955 printf("raid%d: IO Error. Marking %s as failed.\n", 1956 queue->raidPtr->raidid, 1957 queue->raidPtr->Disks[queue->row][queue->col].devname); 1958 queue->raidPtr->Disks[queue->row][queue->col].status = 1959 rf_ds_failed; 1960 queue->raidPtr->status[queue->row] = rf_rs_degraded; 1961 queue->raidPtr->numFailures++; 1962 queue->raidPtr->numNewFailures++; 1963 } else { /* Disk is already dead... */ 1964 /* printf("Disk already marked as dead!\n"); */ 1965 } 1966 1967 } 1968 1969 pool_put(&raidframe_cbufpool, raidbp); 1970 1971 /* Fill in the error value */ 1972 1973 req->error = (bp->b_flags & B_ERROR) ? bp->b_error : 0; 1974 1975 simple_lock(&queue->raidPtr->iodone_lock); 1976 1977 /* Drop this one on the "finished" queue... */ 1978 TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries); 1979 1980 /* Let the raidio thread know there is work to be done. */ 1981 wakeup(&(queue->raidPtr->iodone)); 1982 1983 simple_unlock(&queue->raidPtr->iodone_lock); 1984 1985 splx(s); 1986 } 1987 1988 1989 1990 /* 1991 * initialize a buf structure for doing an I/O in the kernel. 1992 */ 1993 static void 1994 InitBP(bp, b_vp, rw_flag, dev, startSect, numSect, buf, cbFunc, cbArg, 1995 logBytesPerSector, b_proc) 1996 struct buf *bp; 1997 struct vnode *b_vp; 1998 unsigned rw_flag; 1999 dev_t dev; 2000 RF_SectorNum_t startSect; 2001 RF_SectorCount_t numSect; 2002 caddr_t buf; 2003 void (*cbFunc) (struct buf *); 2004 void *cbArg; 2005 int logBytesPerSector; 2006 struct proc *b_proc; 2007 { 2008 /* bp->b_flags = B_PHYS | rw_flag; */ 2009 bp->b_flags = B_CALL | rw_flag; /* XXX need B_PHYS here too??? */ 2010 bp->b_bcount = numSect << logBytesPerSector; 2011 bp->b_bufsize = bp->b_bcount; 2012 bp->b_error = 0; 2013 bp->b_dev = dev; 2014 bp->b_data = buf; 2015 bp->b_blkno = startSect; 2016 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 2017 if (bp->b_bcount == 0) { 2018 panic("bp->b_bcount is zero in InitBP!!"); 2019 } 2020 bp->b_proc = b_proc; 2021 bp->b_iodone = cbFunc; 2022 bp->b_vp = b_vp; 2023 2024 } 2025 2026 static void 2027 raidgetdefaultlabel(raidPtr, rs, lp) 2028 RF_Raid_t *raidPtr; 2029 struct raid_softc *rs; 2030 struct disklabel *lp; 2031 { 2032 memset(lp, 0, sizeof(*lp)); 2033 2034 /* fabricate a label... */ 2035 lp->d_secperunit = raidPtr->totalSectors; 2036 lp->d_secsize = raidPtr->bytesPerSector; 2037 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 2038 lp->d_ntracks = 4 * raidPtr->numCol; 2039 lp->d_ncylinders = raidPtr->totalSectors / 2040 (lp->d_nsectors * lp->d_ntracks); 2041 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 2042 2043 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 2044 lp->d_type = DTYPE_RAID; 2045 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 2046 lp->d_rpm = 3600; 2047 lp->d_interleave = 1; 2048 lp->d_flags = 0; 2049 2050 lp->d_partitions[RAW_PART].p_offset = 0; 2051 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 2052 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 2053 lp->d_npartitions = RAW_PART + 1; 2054 2055 lp->d_magic = DISKMAGIC; 2056 lp->d_magic2 = DISKMAGIC; 2057 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 2058 2059 } 2060 /* 2061 * Read the disklabel from the raid device. If one is not present, fake one 2062 * up. 2063 */ 2064 static void 2065 raidgetdisklabel(dev) 2066 dev_t dev; 2067 { 2068 int unit = raidunit(dev); 2069 struct raid_softc *rs = &raid_softc[unit]; 2070 char *errstring; 2071 struct disklabel *lp = rs->sc_dkdev.dk_label; 2072 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 2073 RF_Raid_t *raidPtr; 2074 2075 db1_printf(("Getting the disklabel...\n")); 2076 2077 memset(clp, 0, sizeof(*clp)); 2078 2079 raidPtr = raidPtrs[unit]; 2080 2081 raidgetdefaultlabel(raidPtr, rs, lp); 2082 2083 /* 2084 * Call the generic disklabel extraction routine. 2085 */ 2086 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 2087 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2088 if (errstring) 2089 raidmakedisklabel(rs); 2090 else { 2091 int i; 2092 struct partition *pp; 2093 2094 /* 2095 * Sanity check whether the found disklabel is valid. 2096 * 2097 * This is necessary since total size of the raid device 2098 * may vary when an interleave is changed even though exactly 2099 * same componets are used, and old disklabel may used 2100 * if that is found. 2101 */ 2102 if (lp->d_secperunit != rs->sc_size) 2103 printf("raid%d: WARNING: %s: " 2104 "total sector size in disklabel (%d) != " 2105 "the size of raid (%ld)\n", unit, rs->sc_xname, 2106 lp->d_secperunit, (long) rs->sc_size); 2107 for (i = 0; i < lp->d_npartitions; i++) { 2108 pp = &lp->d_partitions[i]; 2109 if (pp->p_offset + pp->p_size > rs->sc_size) 2110 printf("raid%d: WARNING: %s: end of partition `%c' " 2111 "exceeds the size of raid (%ld)\n", 2112 unit, rs->sc_xname, 'a' + i, (long) rs->sc_size); 2113 } 2114 } 2115 2116 } 2117 /* 2118 * Take care of things one might want to take care of in the event 2119 * that a disklabel isn't present. 2120 */ 2121 static void 2122 raidmakedisklabel(rs) 2123 struct raid_softc *rs; 2124 { 2125 struct disklabel *lp = rs->sc_dkdev.dk_label; 2126 db1_printf(("Making a label..\n")); 2127 2128 /* 2129 * For historical reasons, if there's no disklabel present 2130 * the raw partition must be marked FS_BSDFFS. 2131 */ 2132 2133 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2134 2135 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2136 2137 lp->d_checksum = dkcksum(lp); 2138 } 2139 /* 2140 * Lookup the provided name in the filesystem. If the file exists, 2141 * is a valid block device, and isn't being used by anyone else, 2142 * set *vpp to the file's vnode. 2143 * You'll find the original of this in ccd.c 2144 */ 2145 int 2146 raidlookup(path, p, vpp) 2147 char *path; 2148 struct proc *p; 2149 struct vnode **vpp; /* result */ 2150 { 2151 struct nameidata nd; 2152 struct vnode *vp; 2153 struct vattr va; 2154 int error; 2155 2156 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p); 2157 if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) { 2158 return (error); 2159 } 2160 vp = nd.ni_vp; 2161 if (vp->v_usecount > 1) { 2162 VOP_UNLOCK(vp, 0); 2163 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2164 return (EBUSY); 2165 } 2166 if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) { 2167 VOP_UNLOCK(vp, 0); 2168 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2169 return (error); 2170 } 2171 /* XXX: eventually we should handle VREG, too. */ 2172 if (va.va_type != VBLK) { 2173 VOP_UNLOCK(vp, 0); 2174 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2175 return (ENOTBLK); 2176 } 2177 VOP_UNLOCK(vp, 0); 2178 *vpp = vp; 2179 return (0); 2180 } 2181 /* 2182 * Wait interruptibly for an exclusive lock. 2183 * 2184 * XXX 2185 * Several drivers do this; it should be abstracted and made MP-safe. 2186 * (Hmm... where have we seen this warning before :-> GO ) 2187 */ 2188 static int 2189 raidlock(rs) 2190 struct raid_softc *rs; 2191 { 2192 int error; 2193 2194 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2195 rs->sc_flags |= RAIDF_WANTED; 2196 if ((error = 2197 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2198 return (error); 2199 } 2200 rs->sc_flags |= RAIDF_LOCKED; 2201 return (0); 2202 } 2203 /* 2204 * Unlock and wake up any waiters. 2205 */ 2206 static void 2207 raidunlock(rs) 2208 struct raid_softc *rs; 2209 { 2210 2211 rs->sc_flags &= ~RAIDF_LOCKED; 2212 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2213 rs->sc_flags &= ~RAIDF_WANTED; 2214 wakeup(rs); 2215 } 2216 } 2217 2218 2219 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2220 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2221 2222 int 2223 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter) 2224 { 2225 RF_ComponentLabel_t clabel; 2226 raidread_component_label(dev, b_vp, &clabel); 2227 clabel.mod_counter = mod_counter; 2228 clabel.clean = RF_RAID_CLEAN; 2229 raidwrite_component_label(dev, b_vp, &clabel); 2230 return(0); 2231 } 2232 2233 2234 int 2235 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter) 2236 { 2237 RF_ComponentLabel_t clabel; 2238 raidread_component_label(dev, b_vp, &clabel); 2239 clabel.mod_counter = mod_counter; 2240 clabel.clean = RF_RAID_DIRTY; 2241 raidwrite_component_label(dev, b_vp, &clabel); 2242 return(0); 2243 } 2244 2245 /* ARGSUSED */ 2246 int 2247 raidread_component_label(dev, b_vp, clabel) 2248 dev_t dev; 2249 struct vnode *b_vp; 2250 RF_ComponentLabel_t *clabel; 2251 { 2252 struct buf *bp; 2253 const struct bdevsw *bdev; 2254 int error; 2255 2256 /* XXX should probably ensure that we don't try to do this if 2257 someone has changed rf_protected_sectors. */ 2258 2259 if (b_vp == NULL) { 2260 /* For whatever reason, this component is not valid. 2261 Don't try to read a component label from it. */ 2262 return(EINVAL); 2263 } 2264 2265 /* get a block of the appropriate size... */ 2266 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2267 bp->b_dev = dev; 2268 2269 /* get our ducks in a row for the read */ 2270 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2271 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2272 bp->b_flags |= B_READ; 2273 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2274 2275 bdev = bdevsw_lookup(bp->b_dev); 2276 if (bdev == NULL) 2277 return (ENXIO); 2278 (*bdev->d_strategy)(bp); 2279 2280 error = biowait(bp); 2281 2282 if (!error) { 2283 memcpy(clabel, bp->b_data, 2284 sizeof(RF_ComponentLabel_t)); 2285 } 2286 2287 brelse(bp); 2288 return(error); 2289 } 2290 /* ARGSUSED */ 2291 int 2292 raidwrite_component_label(dev, b_vp, clabel) 2293 dev_t dev; 2294 struct vnode *b_vp; 2295 RF_ComponentLabel_t *clabel; 2296 { 2297 struct buf *bp; 2298 const struct bdevsw *bdev; 2299 int error; 2300 2301 /* get a block of the appropriate size... */ 2302 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2303 bp->b_dev = dev; 2304 2305 /* get our ducks in a row for the write */ 2306 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2307 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2308 bp->b_flags |= B_WRITE; 2309 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2310 2311 memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE ); 2312 2313 memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t)); 2314 2315 bdev = bdevsw_lookup(bp->b_dev); 2316 if (bdev == NULL) 2317 return (ENXIO); 2318 (*bdev->d_strategy)(bp); 2319 error = biowait(bp); 2320 brelse(bp); 2321 if (error) { 2322 #if 1 2323 printf("Failed to write RAID component info!\n"); 2324 #endif 2325 } 2326 2327 return(error); 2328 } 2329 2330 void 2331 rf_markalldirty(raidPtr) 2332 RF_Raid_t *raidPtr; 2333 { 2334 RF_ComponentLabel_t clabel; 2335 int sparecol; 2336 int r,c; 2337 int i,j; 2338 int srow, scol; 2339 2340 raidPtr->mod_counter++; 2341 for (r = 0; r < raidPtr->numRow; r++) { 2342 for (c = 0; c < raidPtr->numCol; c++) { 2343 /* we don't want to touch (at all) a disk that has 2344 failed */ 2345 if (!RF_DEAD_DISK(raidPtr->Disks[r][c].status)) { 2346 raidread_component_label( 2347 raidPtr->Disks[r][c].dev, 2348 raidPtr->raid_cinfo[r][c].ci_vp, 2349 &clabel); 2350 if (clabel.status == rf_ds_spared) { 2351 /* XXX do something special... 2352 but whatever you do, don't 2353 try to access it!! */ 2354 } else { 2355 raidmarkdirty( 2356 raidPtr->Disks[r][c].dev, 2357 raidPtr->raid_cinfo[r][c].ci_vp, 2358 raidPtr->mod_counter); 2359 } 2360 } 2361 } 2362 } 2363 2364 for( c = 0; c < raidPtr->numSpare ; c++) { 2365 sparecol = raidPtr->numCol + c; 2366 if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) { 2367 /* 2368 2369 we claim this disk is "optimal" if it's 2370 rf_ds_used_spare, as that means it should be 2371 directly substitutable for the disk it replaced. 2372 We note that too... 2373 2374 */ 2375 2376 for(i=0;i<raidPtr->numRow;i++) { 2377 for(j=0;j<raidPtr->numCol;j++) { 2378 if ((raidPtr->Disks[i][j].spareRow == 2379 0) && 2380 (raidPtr->Disks[i][j].spareCol == 2381 sparecol)) { 2382 srow = i; 2383 scol = j; 2384 break; 2385 } 2386 } 2387 } 2388 2389 raidread_component_label( 2390 raidPtr->Disks[0][sparecol].dev, 2391 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2392 &clabel); 2393 /* make sure status is noted */ 2394 2395 raid_init_component_label(raidPtr, &clabel); 2396 2397 clabel.row = srow; 2398 clabel.column = scol; 2399 /* Note: we *don't* change status from rf_ds_used_spare 2400 to rf_ds_optimal */ 2401 /* clabel.status = rf_ds_optimal; */ 2402 2403 raidmarkdirty(raidPtr->Disks[0][sparecol].dev, 2404 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2405 raidPtr->mod_counter); 2406 } 2407 } 2408 } 2409 2410 2411 void 2412 rf_update_component_labels(raidPtr, final) 2413 RF_Raid_t *raidPtr; 2414 int final; 2415 { 2416 RF_ComponentLabel_t clabel; 2417 int sparecol; 2418 int r,c; 2419 int i,j; 2420 int srow, scol; 2421 2422 srow = -1; 2423 scol = -1; 2424 2425 /* XXX should do extra checks to make sure things really are clean, 2426 rather than blindly setting the clean bit... */ 2427 2428 raidPtr->mod_counter++; 2429 2430 for (r = 0; r < raidPtr->numRow; r++) { 2431 for (c = 0; c < raidPtr->numCol; c++) { 2432 if (raidPtr->Disks[r][c].status == rf_ds_optimal) { 2433 raidread_component_label( 2434 raidPtr->Disks[r][c].dev, 2435 raidPtr->raid_cinfo[r][c].ci_vp, 2436 &clabel); 2437 /* make sure status is noted */ 2438 clabel.status = rf_ds_optimal; 2439 /* bump the counter */ 2440 clabel.mod_counter = raidPtr->mod_counter; 2441 2442 raidwrite_component_label( 2443 raidPtr->Disks[r][c].dev, 2444 raidPtr->raid_cinfo[r][c].ci_vp, 2445 &clabel); 2446 if (final == RF_FINAL_COMPONENT_UPDATE) { 2447 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2448 raidmarkclean( 2449 raidPtr->Disks[r][c].dev, 2450 raidPtr->raid_cinfo[r][c].ci_vp, 2451 raidPtr->mod_counter); 2452 } 2453 } 2454 } 2455 /* else we don't touch it.. */ 2456 } 2457 } 2458 2459 for( c = 0; c < raidPtr->numSpare ; c++) { 2460 sparecol = raidPtr->numCol + c; 2461 /* Need to ensure that the reconstruct actually completed! */ 2462 if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) { 2463 /* 2464 2465 we claim this disk is "optimal" if it's 2466 rf_ds_used_spare, as that means it should be 2467 directly substitutable for the disk it replaced. 2468 We note that too... 2469 2470 */ 2471 2472 for(i=0;i<raidPtr->numRow;i++) { 2473 for(j=0;j<raidPtr->numCol;j++) { 2474 if ((raidPtr->Disks[i][j].spareRow == 2475 0) && 2476 (raidPtr->Disks[i][j].spareCol == 2477 sparecol)) { 2478 srow = i; 2479 scol = j; 2480 break; 2481 } 2482 } 2483 } 2484 2485 /* XXX shouldn't *really* need this... */ 2486 raidread_component_label( 2487 raidPtr->Disks[0][sparecol].dev, 2488 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2489 &clabel); 2490 /* make sure status is noted */ 2491 2492 raid_init_component_label(raidPtr, &clabel); 2493 2494 clabel.mod_counter = raidPtr->mod_counter; 2495 clabel.row = srow; 2496 clabel.column = scol; 2497 clabel.status = rf_ds_optimal; 2498 2499 raidwrite_component_label( 2500 raidPtr->Disks[0][sparecol].dev, 2501 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2502 &clabel); 2503 if (final == RF_FINAL_COMPONENT_UPDATE) { 2504 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2505 raidmarkclean( raidPtr->Disks[0][sparecol].dev, 2506 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2507 raidPtr->mod_counter); 2508 } 2509 } 2510 } 2511 } 2512 } 2513 2514 void 2515 rf_close_component(raidPtr, vp, auto_configured) 2516 RF_Raid_t *raidPtr; 2517 struct vnode *vp; 2518 int auto_configured; 2519 { 2520 struct proc *p; 2521 2522 p = raidPtr->engine_thread; 2523 2524 if (vp != NULL) { 2525 if (auto_configured == 1) { 2526 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2527 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2528 vput(vp); 2529 2530 } else { 2531 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2532 } 2533 } 2534 } 2535 2536 2537 void 2538 rf_UnconfigureVnodes(raidPtr) 2539 RF_Raid_t *raidPtr; 2540 { 2541 int r,c; 2542 struct vnode *vp; 2543 int acd; 2544 2545 2546 /* We take this opportunity to close the vnodes like we should.. */ 2547 2548 for (r = 0; r < raidPtr->numRow; r++) { 2549 for (c = 0; c < raidPtr->numCol; c++) { 2550 vp = raidPtr->raid_cinfo[r][c].ci_vp; 2551 acd = raidPtr->Disks[r][c].auto_configured; 2552 rf_close_component(raidPtr, vp, acd); 2553 raidPtr->raid_cinfo[r][c].ci_vp = NULL; 2554 raidPtr->Disks[r][c].auto_configured = 0; 2555 } 2556 } 2557 for (r = 0; r < raidPtr->numSpare; r++) { 2558 vp = raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp; 2559 acd = raidPtr->Disks[0][raidPtr->numCol + r].auto_configured; 2560 rf_close_component(raidPtr, vp, acd); 2561 raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp = NULL; 2562 raidPtr->Disks[0][raidPtr->numCol + r].auto_configured = 0; 2563 } 2564 } 2565 2566 2567 void 2568 rf_ReconThread(req) 2569 struct rf_recon_req *req; 2570 { 2571 int s; 2572 RF_Raid_t *raidPtr; 2573 2574 s = splbio(); 2575 raidPtr = (RF_Raid_t *) req->raidPtr; 2576 raidPtr->recon_in_progress = 1; 2577 2578 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->row, req->col, 2579 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2580 2581 /* XXX get rid of this! we don't need it at all.. */ 2582 RF_Free(req, sizeof(*req)); 2583 2584 raidPtr->recon_in_progress = 0; 2585 splx(s); 2586 2587 /* That's all... */ 2588 kthread_exit(0); /* does not return */ 2589 } 2590 2591 void 2592 rf_RewriteParityThread(raidPtr) 2593 RF_Raid_t *raidPtr; 2594 { 2595 int retcode; 2596 int s; 2597 2598 raidPtr->parity_rewrite_in_progress = 1; 2599 s = splbio(); 2600 retcode = rf_RewriteParity(raidPtr); 2601 splx(s); 2602 if (retcode) { 2603 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2604 } else { 2605 /* set the clean bit! If we shutdown correctly, 2606 the clean bit on each component label will get 2607 set */ 2608 raidPtr->parity_good = RF_RAID_CLEAN; 2609 } 2610 raidPtr->parity_rewrite_in_progress = 0; 2611 2612 /* Anyone waiting for us to stop? If so, inform them... */ 2613 if (raidPtr->waitShutdown) { 2614 wakeup(&raidPtr->parity_rewrite_in_progress); 2615 } 2616 2617 /* That's all... */ 2618 kthread_exit(0); /* does not return */ 2619 } 2620 2621 2622 void 2623 rf_CopybackThread(raidPtr) 2624 RF_Raid_t *raidPtr; 2625 { 2626 int s; 2627 2628 raidPtr->copyback_in_progress = 1; 2629 s = splbio(); 2630 rf_CopybackReconstructedData(raidPtr); 2631 splx(s); 2632 raidPtr->copyback_in_progress = 0; 2633 2634 /* That's all... */ 2635 kthread_exit(0); /* does not return */ 2636 } 2637 2638 2639 void 2640 rf_ReconstructInPlaceThread(req) 2641 struct rf_recon_req *req; 2642 { 2643 int retcode; 2644 int s; 2645 RF_Raid_t *raidPtr; 2646 2647 s = splbio(); 2648 raidPtr = req->raidPtr; 2649 raidPtr->recon_in_progress = 1; 2650 retcode = rf_ReconstructInPlace(raidPtr, req->row, req->col); 2651 RF_Free(req, sizeof(*req)); 2652 raidPtr->recon_in_progress = 0; 2653 splx(s); 2654 2655 /* That's all... */ 2656 kthread_exit(0); /* does not return */ 2657 } 2658 2659 RF_AutoConfig_t * 2660 rf_find_raid_components() 2661 { 2662 struct vnode *vp; 2663 struct disklabel label; 2664 struct device *dv; 2665 dev_t dev; 2666 int bmajor; 2667 int error; 2668 int i; 2669 int good_one; 2670 RF_ComponentLabel_t *clabel; 2671 RF_AutoConfig_t *ac_list; 2672 RF_AutoConfig_t *ac; 2673 2674 2675 /* initialize the AutoConfig list */ 2676 ac_list = NULL; 2677 2678 /* we begin by trolling through *all* the devices on the system */ 2679 2680 for (dv = alldevs.tqh_first; dv != NULL; 2681 dv = dv->dv_list.tqe_next) { 2682 2683 /* we are only interested in disks... */ 2684 if (dv->dv_class != DV_DISK) 2685 continue; 2686 2687 /* we don't care about floppies... */ 2688 if (!strcmp(dv->dv_cfdata->cf_name,"fd")) { 2689 continue; 2690 } 2691 2692 /* we don't care about CD's... */ 2693 if (!strcmp(dv->dv_cfdata->cf_name,"cd")) { 2694 continue; 2695 } 2696 2697 /* hdfd is the Atari/Hades floppy driver */ 2698 if (!strcmp(dv->dv_cfdata->cf_name,"hdfd")) { 2699 continue; 2700 } 2701 /* fdisa is the Atari/Milan floppy driver */ 2702 if (!strcmp(dv->dv_cfdata->cf_name,"fdisa")) { 2703 continue; 2704 } 2705 2706 /* need to find the device_name_to_block_device_major stuff */ 2707 bmajor = devsw_name2blk(dv->dv_xname, NULL, 0); 2708 2709 /* get a vnode for the raw partition of this disk */ 2710 2711 dev = MAKEDISKDEV(bmajor, dv->dv_unit, RAW_PART); 2712 if (bdevvp(dev, &vp)) 2713 panic("RAID can't alloc vnode"); 2714 2715 error = VOP_OPEN(vp, FREAD, NOCRED, 0); 2716 2717 if (error) { 2718 /* "Who cares." Continue looking 2719 for something that exists*/ 2720 vput(vp); 2721 continue; 2722 } 2723 2724 /* Ok, the disk exists. Go get the disklabel. */ 2725 error = VOP_IOCTL(vp, DIOCGDINFO, (caddr_t)&label, 2726 FREAD, NOCRED, 0); 2727 if (error) { 2728 /* 2729 * XXX can't happen - open() would 2730 * have errored out (or faked up one) 2731 */ 2732 printf("can't get label for dev %s%c (%d)!?!?\n", 2733 dv->dv_xname, 'a' + RAW_PART, error); 2734 } 2735 2736 /* don't need this any more. We'll allocate it again 2737 a little later if we really do... */ 2738 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2739 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2740 vput(vp); 2741 2742 for (i=0; i < label.d_npartitions; i++) { 2743 /* We only support partitions marked as RAID */ 2744 if (label.d_partitions[i].p_fstype != FS_RAID) 2745 continue; 2746 2747 dev = MAKEDISKDEV(bmajor, dv->dv_unit, i); 2748 if (bdevvp(dev, &vp)) 2749 panic("RAID can't alloc vnode"); 2750 2751 error = VOP_OPEN(vp, FREAD, NOCRED, 0); 2752 if (error) { 2753 /* Whatever... */ 2754 vput(vp); 2755 continue; 2756 } 2757 2758 good_one = 0; 2759 2760 clabel = (RF_ComponentLabel_t *) 2761 malloc(sizeof(RF_ComponentLabel_t), 2762 M_RAIDFRAME, M_NOWAIT); 2763 if (clabel == NULL) { 2764 /* XXX CLEANUP HERE */ 2765 printf("RAID auto config: out of memory!\n"); 2766 return(NULL); /* XXX probably should panic? */ 2767 } 2768 2769 if (!raidread_component_label(dev, vp, clabel)) { 2770 /* Got the label. Does it look reasonable? */ 2771 if (rf_reasonable_label(clabel) && 2772 (clabel->partitionSize <= 2773 label.d_partitions[i].p_size)) { 2774 #if DEBUG 2775 printf("Component on: %s%c: %d\n", 2776 dv->dv_xname, 'a'+i, 2777 label.d_partitions[i].p_size); 2778 rf_print_component_label(clabel); 2779 #endif 2780 /* if it's reasonable, add it, 2781 else ignore it. */ 2782 ac = (RF_AutoConfig_t *) 2783 malloc(sizeof(RF_AutoConfig_t), 2784 M_RAIDFRAME, 2785 M_NOWAIT); 2786 if (ac == NULL) { 2787 /* XXX should panic?? */ 2788 return(NULL); 2789 } 2790 2791 sprintf(ac->devname, "%s%c", 2792 dv->dv_xname, 'a'+i); 2793 ac->dev = dev; 2794 ac->vp = vp; 2795 ac->clabel = clabel; 2796 ac->next = ac_list; 2797 ac_list = ac; 2798 good_one = 1; 2799 } 2800 } 2801 if (!good_one) { 2802 /* cleanup */ 2803 free(clabel, M_RAIDFRAME); 2804 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2805 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2806 vput(vp); 2807 } 2808 } 2809 } 2810 return(ac_list); 2811 } 2812 2813 static int 2814 rf_reasonable_label(clabel) 2815 RF_ComponentLabel_t *clabel; 2816 { 2817 2818 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 2819 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 2820 ((clabel->clean == RF_RAID_CLEAN) || 2821 (clabel->clean == RF_RAID_DIRTY)) && 2822 clabel->row >=0 && 2823 clabel->column >= 0 && 2824 clabel->num_rows > 0 && 2825 clabel->num_columns > 0 && 2826 clabel->row < clabel->num_rows && 2827 clabel->column < clabel->num_columns && 2828 clabel->blockSize > 0 && 2829 clabel->numBlocks > 0) { 2830 /* label looks reasonable enough... */ 2831 return(1); 2832 } 2833 return(0); 2834 } 2835 2836 2837 #if DEBUG 2838 void 2839 rf_print_component_label(clabel) 2840 RF_ComponentLabel_t *clabel; 2841 { 2842 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 2843 clabel->row, clabel->column, 2844 clabel->num_rows, clabel->num_columns); 2845 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 2846 clabel->version, clabel->serial_number, 2847 clabel->mod_counter); 2848 printf(" Clean: %s Status: %d\n", 2849 clabel->clean ? "Yes" : "No", clabel->status ); 2850 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 2851 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 2852 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n", 2853 (char) clabel->parityConfig, clabel->blockSize, 2854 clabel->numBlocks); 2855 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" ); 2856 printf(" Contains root partition: %s\n", 2857 clabel->root_partition ? "Yes" : "No" ); 2858 printf(" Last configured as: raid%d\n", clabel->last_unit ); 2859 #if 0 2860 printf(" Config order: %d\n", clabel->config_order); 2861 #endif 2862 2863 } 2864 #endif 2865 2866 RF_ConfigSet_t * 2867 rf_create_auto_sets(ac_list) 2868 RF_AutoConfig_t *ac_list; 2869 { 2870 RF_AutoConfig_t *ac; 2871 RF_ConfigSet_t *config_sets; 2872 RF_ConfigSet_t *cset; 2873 RF_AutoConfig_t *ac_next; 2874 2875 2876 config_sets = NULL; 2877 2878 /* Go through the AutoConfig list, and figure out which components 2879 belong to what sets. */ 2880 ac = ac_list; 2881 while(ac!=NULL) { 2882 /* we're going to putz with ac->next, so save it here 2883 for use at the end of the loop */ 2884 ac_next = ac->next; 2885 2886 if (config_sets == NULL) { 2887 /* will need at least this one... */ 2888 config_sets = (RF_ConfigSet_t *) 2889 malloc(sizeof(RF_ConfigSet_t), 2890 M_RAIDFRAME, M_NOWAIT); 2891 if (config_sets == NULL) { 2892 panic("rf_create_auto_sets: No memory!"); 2893 } 2894 /* this one is easy :) */ 2895 config_sets->ac = ac; 2896 config_sets->next = NULL; 2897 config_sets->rootable = 0; 2898 ac->next = NULL; 2899 } else { 2900 /* which set does this component fit into? */ 2901 cset = config_sets; 2902 while(cset!=NULL) { 2903 if (rf_does_it_fit(cset, ac)) { 2904 /* looks like it matches... */ 2905 ac->next = cset->ac; 2906 cset->ac = ac; 2907 break; 2908 } 2909 cset = cset->next; 2910 } 2911 if (cset==NULL) { 2912 /* didn't find a match above... new set..*/ 2913 cset = (RF_ConfigSet_t *) 2914 malloc(sizeof(RF_ConfigSet_t), 2915 M_RAIDFRAME, M_NOWAIT); 2916 if (cset == NULL) { 2917 panic("rf_create_auto_sets: No memory!"); 2918 } 2919 cset->ac = ac; 2920 ac->next = NULL; 2921 cset->next = config_sets; 2922 cset->rootable = 0; 2923 config_sets = cset; 2924 } 2925 } 2926 ac = ac_next; 2927 } 2928 2929 2930 return(config_sets); 2931 } 2932 2933 static int 2934 rf_does_it_fit(cset, ac) 2935 RF_ConfigSet_t *cset; 2936 RF_AutoConfig_t *ac; 2937 { 2938 RF_ComponentLabel_t *clabel1, *clabel2; 2939 2940 /* If this one matches the *first* one in the set, that's good 2941 enough, since the other members of the set would have been 2942 through here too... */ 2943 /* note that we are not checking partitionSize here.. 2944 2945 Note that we are also not checking the mod_counters here. 2946 If everything else matches execpt the mod_counter, that's 2947 good enough for this test. We will deal with the mod_counters 2948 a little later in the autoconfiguration process. 2949 2950 (clabel1->mod_counter == clabel2->mod_counter) && 2951 2952 The reason we don't check for this is that failed disks 2953 will have lower modification counts. If those disks are 2954 not added to the set they used to belong to, then they will 2955 form their own set, which may result in 2 different sets, 2956 for example, competing to be configured at raid0, and 2957 perhaps competing to be the root filesystem set. If the 2958 wrong ones get configured, or both attempt to become /, 2959 weird behaviour and or serious lossage will occur. Thus we 2960 need to bring them into the fold here, and kick them out at 2961 a later point. 2962 2963 */ 2964 2965 clabel1 = cset->ac->clabel; 2966 clabel2 = ac->clabel; 2967 if ((clabel1->version == clabel2->version) && 2968 (clabel1->serial_number == clabel2->serial_number) && 2969 (clabel1->num_rows == clabel2->num_rows) && 2970 (clabel1->num_columns == clabel2->num_columns) && 2971 (clabel1->sectPerSU == clabel2->sectPerSU) && 2972 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 2973 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 2974 (clabel1->parityConfig == clabel2->parityConfig) && 2975 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 2976 (clabel1->blockSize == clabel2->blockSize) && 2977 (clabel1->numBlocks == clabel2->numBlocks) && 2978 (clabel1->autoconfigure == clabel2->autoconfigure) && 2979 (clabel1->root_partition == clabel2->root_partition) && 2980 (clabel1->last_unit == clabel2->last_unit) && 2981 (clabel1->config_order == clabel2->config_order)) { 2982 /* if it get's here, it almost *has* to be a match */ 2983 } else { 2984 /* it's not consistent with somebody in the set.. 2985 punt */ 2986 return(0); 2987 } 2988 /* all was fine.. it must fit... */ 2989 return(1); 2990 } 2991 2992 int 2993 rf_have_enough_components(cset) 2994 RF_ConfigSet_t *cset; 2995 { 2996 RF_AutoConfig_t *ac; 2997 RF_AutoConfig_t *auto_config; 2998 RF_ComponentLabel_t *clabel; 2999 int r,c; 3000 int num_rows; 3001 int num_cols; 3002 int num_missing; 3003 int mod_counter; 3004 int mod_counter_found; 3005 int even_pair_failed; 3006 char parity_type; 3007 3008 3009 /* check to see that we have enough 'live' components 3010 of this set. If so, we can configure it if necessary */ 3011 3012 num_rows = cset->ac->clabel->num_rows; 3013 num_cols = cset->ac->clabel->num_columns; 3014 parity_type = cset->ac->clabel->parityConfig; 3015 3016 /* XXX Check for duplicate components!?!?!? */ 3017 3018 /* Determine what the mod_counter is supposed to be for this set. */ 3019 3020 mod_counter_found = 0; 3021 mod_counter = 0; 3022 ac = cset->ac; 3023 while(ac!=NULL) { 3024 if (mod_counter_found==0) { 3025 mod_counter = ac->clabel->mod_counter; 3026 mod_counter_found = 1; 3027 } else { 3028 if (ac->clabel->mod_counter > mod_counter) { 3029 mod_counter = ac->clabel->mod_counter; 3030 } 3031 } 3032 ac = ac->next; 3033 } 3034 3035 num_missing = 0; 3036 auto_config = cset->ac; 3037 3038 for(r=0; r<num_rows; r++) { 3039 even_pair_failed = 0; 3040 for(c=0; c<num_cols; c++) { 3041 ac = auto_config; 3042 while(ac!=NULL) { 3043 if ((ac->clabel->row == r) && 3044 (ac->clabel->column == c) && 3045 (ac->clabel->mod_counter == mod_counter)) { 3046 /* it's this one... */ 3047 #if DEBUG 3048 printf("Found: %s at %d,%d\n", 3049 ac->devname,r,c); 3050 #endif 3051 break; 3052 } 3053 ac=ac->next; 3054 } 3055 if (ac==NULL) { 3056 /* Didn't find one here! */ 3057 /* special case for RAID 1, especially 3058 where there are more than 2 3059 components (where RAIDframe treats 3060 things a little differently :( ) */ 3061 if (parity_type == '1') { 3062 if (c%2 == 0) { /* even component */ 3063 even_pair_failed = 1; 3064 } else { /* odd component. If 3065 we're failed, and 3066 so is the even 3067 component, it's 3068 "Good Night, Charlie" */ 3069 if (even_pair_failed == 1) { 3070 return(0); 3071 } 3072 } 3073 } else { 3074 /* normal accounting */ 3075 num_missing++; 3076 } 3077 } 3078 if ((parity_type == '1') && (c%2 == 1)) { 3079 /* Just did an even component, and we didn't 3080 bail.. reset the even_pair_failed flag, 3081 and go on to the next component.... */ 3082 even_pair_failed = 0; 3083 } 3084 } 3085 } 3086 3087 clabel = cset->ac->clabel; 3088 3089 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3090 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3091 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3092 /* XXX this needs to be made *much* more general */ 3093 /* Too many failures */ 3094 return(0); 3095 } 3096 /* otherwise, all is well, and we've got enough to take a kick 3097 at autoconfiguring this set */ 3098 return(1); 3099 } 3100 3101 void 3102 rf_create_configuration(ac,config,raidPtr) 3103 RF_AutoConfig_t *ac; 3104 RF_Config_t *config; 3105 RF_Raid_t *raidPtr; 3106 { 3107 RF_ComponentLabel_t *clabel; 3108 int i; 3109 3110 clabel = ac->clabel; 3111 3112 /* 1. Fill in the common stuff */ 3113 config->numRow = clabel->num_rows; 3114 config->numCol = clabel->num_columns; 3115 config->numSpare = 0; /* XXX should this be set here? */ 3116 config->sectPerSU = clabel->sectPerSU; 3117 config->SUsPerPU = clabel->SUsPerPU; 3118 config->SUsPerRU = clabel->SUsPerRU; 3119 config->parityConfig = clabel->parityConfig; 3120 /* XXX... */ 3121 strcpy(config->diskQueueType,"fifo"); 3122 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3123 config->layoutSpecificSize = 0; /* XXX ?? */ 3124 3125 while(ac!=NULL) { 3126 /* row/col values will be in range due to the checks 3127 in reasonable_label() */ 3128 strcpy(config->devnames[ac->clabel->row][ac->clabel->column], 3129 ac->devname); 3130 ac = ac->next; 3131 } 3132 3133 for(i=0;i<RF_MAXDBGV;i++) { 3134 config->debugVars[i][0] = NULL; 3135 } 3136 } 3137 3138 int 3139 rf_set_autoconfig(raidPtr, new_value) 3140 RF_Raid_t *raidPtr; 3141 int new_value; 3142 { 3143 RF_ComponentLabel_t clabel; 3144 struct vnode *vp; 3145 dev_t dev; 3146 int row, column; 3147 int sparecol; 3148 3149 raidPtr->autoconfigure = new_value; 3150 for(row=0; row<raidPtr->numRow; row++) { 3151 for(column=0; column<raidPtr->numCol; column++) { 3152 if (raidPtr->Disks[row][column].status == 3153 rf_ds_optimal) { 3154 dev = raidPtr->Disks[row][column].dev; 3155 vp = raidPtr->raid_cinfo[row][column].ci_vp; 3156 raidread_component_label(dev, vp, &clabel); 3157 clabel.autoconfigure = new_value; 3158 raidwrite_component_label(dev, vp, &clabel); 3159 } 3160 } 3161 } 3162 for(column = 0; column < raidPtr->numSpare ; column++) { 3163 sparecol = raidPtr->numCol + column; 3164 if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) { 3165 dev = raidPtr->Disks[0][sparecol].dev; 3166 vp = raidPtr->raid_cinfo[0][sparecol].ci_vp; 3167 raidread_component_label(dev, vp, &clabel); 3168 clabel.autoconfigure = new_value; 3169 raidwrite_component_label(dev, vp, &clabel); 3170 } 3171 } 3172 return(new_value); 3173 } 3174 3175 int 3176 rf_set_rootpartition(raidPtr, new_value) 3177 RF_Raid_t *raidPtr; 3178 int new_value; 3179 { 3180 RF_ComponentLabel_t clabel; 3181 struct vnode *vp; 3182 dev_t dev; 3183 int row, column; 3184 int sparecol; 3185 3186 raidPtr->root_partition = new_value; 3187 for(row=0; row<raidPtr->numRow; row++) { 3188 for(column=0; column<raidPtr->numCol; column++) { 3189 if (raidPtr->Disks[row][column].status == 3190 rf_ds_optimal) { 3191 dev = raidPtr->Disks[row][column].dev; 3192 vp = raidPtr->raid_cinfo[row][column].ci_vp; 3193 raidread_component_label(dev, vp, &clabel); 3194 clabel.root_partition = new_value; 3195 raidwrite_component_label(dev, vp, &clabel); 3196 } 3197 } 3198 } 3199 for(column = 0; column < raidPtr->numSpare ; column++) { 3200 sparecol = raidPtr->numCol + column; 3201 if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) { 3202 dev = raidPtr->Disks[0][sparecol].dev; 3203 vp = raidPtr->raid_cinfo[0][sparecol].ci_vp; 3204 raidread_component_label(dev, vp, &clabel); 3205 clabel.root_partition = new_value; 3206 raidwrite_component_label(dev, vp, &clabel); 3207 } 3208 } 3209 return(new_value); 3210 } 3211 3212 void 3213 rf_release_all_vps(cset) 3214 RF_ConfigSet_t *cset; 3215 { 3216 RF_AutoConfig_t *ac; 3217 3218 ac = cset->ac; 3219 while(ac!=NULL) { 3220 /* Close the vp, and give it back */ 3221 if (ac->vp) { 3222 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3223 VOP_CLOSE(ac->vp, FREAD, NOCRED, 0); 3224 vput(ac->vp); 3225 ac->vp = NULL; 3226 } 3227 ac = ac->next; 3228 } 3229 } 3230 3231 3232 void 3233 rf_cleanup_config_set(cset) 3234 RF_ConfigSet_t *cset; 3235 { 3236 RF_AutoConfig_t *ac; 3237 RF_AutoConfig_t *next_ac; 3238 3239 ac = cset->ac; 3240 while(ac!=NULL) { 3241 next_ac = ac->next; 3242 /* nuke the label */ 3243 free(ac->clabel, M_RAIDFRAME); 3244 /* cleanup the config structure */ 3245 free(ac, M_RAIDFRAME); 3246 /* "next.." */ 3247 ac = next_ac; 3248 } 3249 /* and, finally, nuke the config set */ 3250 free(cset, M_RAIDFRAME); 3251 } 3252 3253 3254 void 3255 raid_init_component_label(raidPtr, clabel) 3256 RF_Raid_t *raidPtr; 3257 RF_ComponentLabel_t *clabel; 3258 { 3259 /* current version number */ 3260 clabel->version = RF_COMPONENT_LABEL_VERSION; 3261 clabel->serial_number = raidPtr->serial_number; 3262 clabel->mod_counter = raidPtr->mod_counter; 3263 clabel->num_rows = raidPtr->numRow; 3264 clabel->num_columns = raidPtr->numCol; 3265 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3266 clabel->status = rf_ds_optimal; /* "It's good!" */ 3267 3268 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3269 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3270 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3271 3272 clabel->blockSize = raidPtr->bytesPerSector; 3273 clabel->numBlocks = raidPtr->sectorsPerDisk; 3274 3275 /* XXX not portable */ 3276 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3277 clabel->maxOutstanding = raidPtr->maxOutstanding; 3278 clabel->autoconfigure = raidPtr->autoconfigure; 3279 clabel->root_partition = raidPtr->root_partition; 3280 clabel->last_unit = raidPtr->raidid; 3281 clabel->config_order = raidPtr->config_order; 3282 } 3283 3284 int 3285 rf_auto_config_set(cset,unit) 3286 RF_ConfigSet_t *cset; 3287 int *unit; 3288 { 3289 RF_Raid_t *raidPtr; 3290 RF_Config_t *config; 3291 int raidID; 3292 int retcode; 3293 3294 #if DEBUG 3295 printf("RAID autoconfigure\n"); 3296 #endif 3297 3298 retcode = 0; 3299 *unit = -1; 3300 3301 /* 1. Create a config structure */ 3302 3303 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3304 M_RAIDFRAME, 3305 M_NOWAIT); 3306 if (config==NULL) { 3307 printf("Out of mem!?!?\n"); 3308 /* XXX do something more intelligent here. */ 3309 return(1); 3310 } 3311 3312 memset(config, 0, sizeof(RF_Config_t)); 3313 3314 /* 3315 2. Figure out what RAID ID this one is supposed to live at 3316 See if we can get the same RAID dev that it was configured 3317 on last time.. 3318 */ 3319 3320 raidID = cset->ac->clabel->last_unit; 3321 if ((raidID < 0) || (raidID >= numraid)) { 3322 /* let's not wander off into lala land. */ 3323 raidID = numraid - 1; 3324 } 3325 if (raidPtrs[raidID]->valid != 0) { 3326 3327 /* 3328 Nope... Go looking for an alternative... 3329 Start high so we don't immediately use raid0 if that's 3330 not taken. 3331 */ 3332 3333 for(raidID = numraid - 1; raidID >= 0; raidID--) { 3334 if (raidPtrs[raidID]->valid == 0) { 3335 /* can use this one! */ 3336 break; 3337 } 3338 } 3339 } 3340 3341 if (raidID < 0) { 3342 /* punt... */ 3343 printf("Unable to auto configure this set!\n"); 3344 printf("(Out of RAID devs!)\n"); 3345 return(1); 3346 } 3347 3348 #if DEBUG 3349 printf("Configuring raid%d:\n",raidID); 3350 #endif 3351 3352 raidPtr = raidPtrs[raidID]; 3353 3354 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3355 raidPtr->raidid = raidID; 3356 raidPtr->openings = RAIDOUTSTANDING; 3357 3358 /* 3. Build the configuration structure */ 3359 rf_create_configuration(cset->ac, config, raidPtr); 3360 3361 /* 4. Do the configuration */ 3362 retcode = rf_Configure(raidPtr, config, cset->ac); 3363 3364 if (retcode == 0) { 3365 3366 raidinit(raidPtrs[raidID]); 3367 3368 rf_markalldirty(raidPtrs[raidID]); 3369 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3370 if (cset->ac->clabel->root_partition==1) { 3371 /* everything configured just fine. Make a note 3372 that this set is eligible to be root. */ 3373 cset->rootable = 1; 3374 /* XXX do this here? */ 3375 raidPtrs[raidID]->root_partition = 1; 3376 } 3377 } 3378 3379 /* 5. Cleanup */ 3380 free(config, M_RAIDFRAME); 3381 3382 *unit = raidID; 3383 return(retcode); 3384 } 3385 3386 void 3387 rf_disk_unbusy(desc) 3388 RF_RaidAccessDesc_t *desc; 3389 { 3390 struct buf *bp; 3391 3392 bp = (struct buf *)desc->bp; 3393 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3394 (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ)); 3395 } 3396