rf_netbsdkintf.c revision 1.109.2.1
1 /* $NetBSD: rf_netbsdkintf.c,v 1.109.2.1 2001/09/07 04:45:29 thorpej 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 * Copyright (c) 1995 Carnegie-Mellon University. 82 * All rights reserved. 83 * 84 * Authors: Mark Holland, Jim Zelenka 85 * 86 * Permission to use, copy, modify and distribute this software and 87 * its documentation is hereby granted, provided that both the copyright 88 * notice and this permission notice appear in all copies of the 89 * software, derivative works or modified versions, and any portions 90 * thereof, and that both notices appear in supporting documentation. 91 * 92 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 93 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 94 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 95 * 96 * Carnegie Mellon requests users of this software to return to 97 * 98 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU 99 * School of Computer Science 100 * Carnegie Mellon University 101 * Pittsburgh PA 15213-3890 102 * 103 * any improvements or extensions that they make and grant Carnegie the 104 * rights to redistribute these changes. 105 */ 106 107 /*********************************************************** 108 * 109 * rf_kintf.c -- the kernel interface routines for RAIDframe 110 * 111 ***********************************************************/ 112 113 #include <sys/errno.h> 114 #include <sys/param.h> 115 #include <sys/pool.h> 116 #include <sys/queue.h> 117 #include <sys/disk.h> 118 #include <sys/device.h> 119 #include <sys/stat.h> 120 #include <sys/ioctl.h> 121 #include <sys/fcntl.h> 122 #include <sys/systm.h> 123 #include <sys/namei.h> 124 #include <sys/vnode.h> 125 #include <sys/param.h> 126 #include <sys/types.h> 127 #include <machine/types.h> 128 #include <sys/disklabel.h> 129 #include <sys/conf.h> 130 #include <sys/lock.h> 131 #include <sys/buf.h> 132 #include <sys/user.h> 133 #include <sys/reboot.h> 134 135 #include <miscfs/specfs/specdev.h> 136 137 #include "raid.h" 138 #include "opt_raid_autoconfig.h" 139 #include "rf_raid.h" 140 #include "rf_raidframe.h" 141 #include "rf_copyback.h" 142 #include "rf_dag.h" 143 #include "rf_dagflags.h" 144 #include "rf_desc.h" 145 #include "rf_diskqueue.h" 146 #include "rf_acctrace.h" 147 #include "rf_etimer.h" 148 #include "rf_general.h" 149 #include "rf_debugMem.h" 150 #include "rf_kintf.h" 151 #include "rf_options.h" 152 #include "rf_driver.h" 153 #include "rf_parityscan.h" 154 #include "rf_debugprint.h" 155 #include "rf_threadstuff.h" 156 #include "rf_configure.h" 157 158 int rf_kdebug_level = 0; 159 160 #ifdef DEBUG 161 #define db1_printf(a) if (rf_kdebug_level > 0) printf a 162 #else /* DEBUG */ 163 #define db1_printf(a) { } 164 #endif /* DEBUG */ 165 166 static RF_Raid_t **raidPtrs; /* global raid device descriptors */ 167 168 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex) 169 170 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a 171 * spare table */ 172 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from 173 * installation process */ 174 175 /* prototypes */ 176 static void KernelWakeupFunc(struct buf * bp); 177 static void InitBP(struct buf * bp, struct vnode *, unsigned rw_flag, 178 RF_SectorNum_t startSect, RF_SectorCount_t numSect, 179 caddr_t buf, void (*cbFunc) (struct buf *), void *cbArg, 180 int logBytesPerSector, struct proc * b_proc); 181 static void raidinit(RF_Raid_t *); 182 183 void raidattach(int); 184 185 /* 186 * Pilfered from ccd.c 187 */ 188 189 struct raidbuf { 190 struct buf rf_buf; /* new I/O buf. MUST BE FIRST!!! */ 191 struct buf *rf_obp; /* ptr. to original I/O buf */ 192 int rf_flags; /* misc. flags */ 193 RF_DiskQueueData_t *req;/* the request that this was part of.. */ 194 }; 195 196 197 #define RAIDGETBUF(rs) pool_get(&(rs)->sc_cbufpool, PR_NOWAIT) 198 #define RAIDPUTBUF(rs, cbp) pool_put(&(rs)->sc_cbufpool, cbp) 199 200 /* XXX Not sure if the following should be replacing the raidPtrs above, 201 or if it should be used in conjunction with that... 202 */ 203 204 struct raid_softc { 205 int sc_flags; /* flags */ 206 int sc_cflags; /* configuration flags */ 207 size_t sc_size; /* size of the raid device */ 208 char sc_xname[20]; /* XXX external name */ 209 struct disk sc_dkdev; /* generic disk device info */ 210 struct pool sc_cbufpool; /* component buffer pool */ 211 struct buf_queue buf_queue; /* used for the device queue */ 212 }; 213 /* sc_flags */ 214 #define RAIDF_INITED 0x01 /* unit has been initialized */ 215 #define RAIDF_WLABEL 0x02 /* label area is writable */ 216 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 217 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 218 #define RAIDF_LOCKED 0x80 /* unit is locked */ 219 220 #define raidunit(x) DISKUNIT(x) 221 int numraid = 0; 222 223 /* 224 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 225 * Be aware that large numbers can allow the driver to consume a lot of 226 * kernel memory, especially on writes, and in degraded mode reads. 227 * 228 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 229 * a single 64K write will typically require 64K for the old data, 230 * 64K for the old parity, and 64K for the new parity, for a total 231 * of 192K (if the parity buffer is not re-used immediately). 232 * Even it if is used immedately, that's still 128K, which when multiplied 233 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 234 * 235 * Now in degraded mode, for example, a 64K read on the above setup may 236 * require data reconstruction, which will require *all* of the 4 remaining 237 * disks to participate -- 4 * 32K/disk == 128K again. 238 */ 239 240 #ifndef RAIDOUTSTANDING 241 #define RAIDOUTSTANDING 6 242 #endif 243 244 #define RAIDLABELDEV(dev) \ 245 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 246 247 /* declared here, and made public, for the benefit of KVM stuff.. */ 248 struct raid_softc *raid_softc; 249 250 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *, 251 struct disklabel *); 252 static void raidgetdisklabel(struct vnode *); 253 static void raidmakedisklabel(struct raid_softc *); 254 255 static int raidlock(struct raid_softc *); 256 static void raidunlock(struct raid_softc *); 257 258 static void rf_markalldirty(RF_Raid_t *); 259 void rf_mountroot_hook(struct device *); 260 261 struct device *raidrootdev; 262 263 void rf_ReconThread(struct rf_recon_req *); 264 /* XXX what I want is: */ 265 /*void rf_ReconThread(RF_Raid_t *raidPtr); */ 266 void rf_RewriteParityThread(RF_Raid_t *raidPtr); 267 void rf_CopybackThread(RF_Raid_t *raidPtr); 268 void rf_ReconstructInPlaceThread(struct rf_recon_req *); 269 void rf_buildroothack(void *); 270 271 RF_AutoConfig_t *rf_find_raid_components(void); 272 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *); 273 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *); 274 static int rf_reasonable_label(RF_ComponentLabel_t *); 275 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *); 276 int rf_set_autoconfig(RF_Raid_t *, int); 277 int rf_set_rootpartition(RF_Raid_t *, int); 278 void rf_release_all_vps(RF_ConfigSet_t *); 279 void rf_cleanup_config_set(RF_ConfigSet_t *); 280 int rf_have_enough_components(RF_ConfigSet_t *); 281 int rf_auto_config_set(RF_ConfigSet_t *, int *); 282 283 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not 284 allow autoconfig to take place. 285 Note that this is overridden by having 286 RAID_AUTOCONFIG as an option in the 287 kernel config file. */ 288 289 void 290 raidattach(num) 291 int num; 292 { 293 int raidID; 294 int i, rc; 295 RF_AutoConfig_t *ac_list; /* autoconfig list */ 296 RF_ConfigSet_t *config_sets; 297 298 #ifdef DEBUG 299 printf("raidattach: Asked for %d units\n", num); 300 #endif 301 302 if (num <= 0) { 303 #ifdef DIAGNOSTIC 304 panic("raidattach: count <= 0"); 305 #endif 306 return; 307 } 308 /* This is where all the initialization stuff gets done. */ 309 310 numraid = num; 311 312 /* Make some space for requested number of units... */ 313 314 RF_Calloc(raidPtrs, num, sizeof(RF_Raid_t *), (RF_Raid_t **)); 315 if (raidPtrs == NULL) { 316 panic("raidPtrs is NULL!!\n"); 317 } 318 319 rc = rf_mutex_init(&rf_sparet_wait_mutex); 320 if (rc) { 321 RF_PANIC(); 322 } 323 324 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 325 326 for (i = 0; i < num; i++) 327 raidPtrs[i] = NULL; 328 rc = rf_BootRaidframe(); 329 if (rc == 0) 330 printf("Kernelized RAIDframe activated\n"); 331 else 332 panic("Serious error booting RAID!!\n"); 333 334 /* put together some datastructures like the CCD device does.. This 335 * lets us lock the device and what-not when it gets opened. */ 336 337 raid_softc = (struct raid_softc *) 338 malloc(num * sizeof(struct raid_softc), 339 M_RAIDFRAME, M_NOWAIT); 340 if (raid_softc == NULL) { 341 printf("WARNING: no memory for RAIDframe driver\n"); 342 return; 343 } 344 345 memset(raid_softc, 0, num * sizeof(struct raid_softc)); 346 347 raidrootdev = (struct device *)malloc(num * sizeof(struct device), 348 M_RAIDFRAME, M_NOWAIT); 349 if (raidrootdev == NULL) { 350 panic("No memory for RAIDframe driver!!?!?!\n"); 351 } 352 353 for (raidID = 0; raidID < num; raidID++) { 354 BUFQ_INIT(&raid_softc[raidID].buf_queue); 355 356 raidrootdev[raidID].dv_class = DV_DISK; 357 raidrootdev[raidID].dv_cfdata = NULL; 358 raidrootdev[raidID].dv_unit = raidID; 359 raidrootdev[raidID].dv_parent = NULL; 360 raidrootdev[raidID].dv_flags = 0; 361 sprintf(raidrootdev[raidID].dv_xname,"raid%d",raidID); 362 363 RF_Calloc(raidPtrs[raidID], 1, sizeof(RF_Raid_t), 364 (RF_Raid_t *)); 365 if (raidPtrs[raidID] == NULL) { 366 printf("WARNING: raidPtrs[%d] is NULL\n", raidID); 367 numraid = raidID; 368 return; 369 } 370 } 371 372 #if RAID_AUTOCONFIG 373 raidautoconfig = 1; 374 #endif 375 376 if (raidautoconfig) { 377 /* 1. locate all RAID components on the system */ 378 379 #if DEBUG 380 printf("Searching for raid components...\n"); 381 #endif 382 ac_list = rf_find_raid_components(); 383 384 /* 2. sort them into their respective sets */ 385 386 config_sets = rf_create_auto_sets(ac_list); 387 388 /* 3. evaluate each set and configure the valid ones 389 This gets done in rf_buildroothack() */ 390 391 /* schedule the creation of the thread to do the 392 "/ on RAID" stuff */ 393 394 kthread_create(rf_buildroothack,config_sets); 395 396 #if 0 397 mountroothook_establish(rf_mountroot_hook, &raidrootdev[0]); 398 #endif 399 } 400 401 } 402 403 void 404 rf_buildroothack(arg) 405 void *arg; 406 { 407 RF_ConfigSet_t *config_sets = arg; 408 RF_ConfigSet_t *cset; 409 RF_ConfigSet_t *next_cset; 410 int retcode; 411 int raidID; 412 int rootID; 413 int num_root; 414 415 rootID = 0; 416 num_root = 0; 417 cset = config_sets; 418 while(cset != NULL ) { 419 next_cset = cset->next; 420 if (rf_have_enough_components(cset) && 421 cset->ac->clabel->autoconfigure==1) { 422 retcode = rf_auto_config_set(cset,&raidID); 423 if (!retcode) { 424 if (cset->rootable) { 425 rootID = raidID; 426 num_root++; 427 } 428 } else { 429 /* The autoconfig didn't work :( */ 430 #if DEBUG 431 printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 432 #endif 433 rf_release_all_vps(cset); 434 } 435 } else { 436 /* we're not autoconfiguring this set... 437 release the associated resources */ 438 rf_release_all_vps(cset); 439 } 440 /* cleanup */ 441 rf_cleanup_config_set(cset); 442 cset = next_cset; 443 } 444 if (boothowto & RB_ASKNAME) { 445 /* We don't auto-config... */ 446 } else { 447 /* They didn't ask, and we found something bootable... */ 448 449 if (num_root == 1) { 450 booted_device = &raidrootdev[rootID]; 451 } else if (num_root > 1) { 452 /* we can't guess.. require the user to answer... */ 453 boothowto |= RB_ASKNAME; 454 } 455 } 456 } 457 458 459 int 460 raidsize(dev) 461 dev_t dev; 462 { 463 #if 1 /* XXXthorpej */ 464 return (-1); 465 #else 466 struct raid_softc *rs; 467 struct disklabel *lp; 468 int part, unit, omask, size; 469 470 unit = raidunit(dev); 471 if (unit >= numraid) 472 return (-1); 473 rs = &raid_softc[unit]; 474 475 if ((rs->sc_flags & RAIDF_INITED) == 0) 476 return (-1); 477 478 part = DISKPART(dev); 479 omask = rs->sc_dkdev.dk_openmask & (1 << part); 480 lp = rs->sc_dkdev.dk_label; 481 482 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc)) 483 return (-1); 484 485 if (lp->d_partitions[part].p_fstype != FS_SWAP) 486 size = -1; 487 else 488 size = lp->d_partitions[part].p_size * 489 (lp->d_secsize / DEV_BSIZE); 490 491 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc)) 492 return (-1); 493 494 return (size); 495 #endif 496 } 497 498 int 499 raiddump(dev, blkno, va, size) 500 dev_t dev; 501 daddr_t blkno; 502 caddr_t va; 503 size_t size; 504 { 505 /* Not implemented. */ 506 return ENXIO; 507 } 508 509 /* ARGSUSED */ 510 int 511 raidopen(devvp, flags, fmt, p) 512 struct vnode *devvp; 513 int flags, fmt; 514 struct proc *p; 515 { 516 int unit = raidunit(devvp->v_rdev); 517 struct raid_softc *rs; 518 struct disklabel *lp; 519 int part, pmask; 520 int error = 0; 521 522 if (unit >= numraid) 523 return (ENXIO); 524 rs = &raid_softc[unit]; 525 devvp->v_devcookie = rs; 526 527 if ((error = raidlock(rs)) != 0) 528 return (error); 529 lp = rs->sc_dkdev.dk_label; 530 531 part = DISKPART(devvp->v_rdev); 532 pmask = (1 << part); 533 534 db1_printf(("Opening raid device number: %d partition: %d\n", 535 unit, part)); 536 537 538 if ((rs->sc_flags & RAIDF_INITED) && 539 (rs->sc_dkdev.dk_openmask == 0)) 540 raidgetdisklabel(devvp); 541 542 /* make sure that this partition exists */ 543 544 if (part != RAW_PART) { 545 db1_printf(("Not a raw partition..\n")); 546 if (((rs->sc_flags & RAIDF_INITED) == 0) || 547 ((part >= lp->d_npartitions) || 548 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 549 error = ENXIO; 550 raidunlock(rs); 551 db1_printf(("Bailing out...\n")); 552 return (error); 553 } 554 } 555 /* Prevent this unit from being unconfigured while open. */ 556 switch (fmt) { 557 case S_IFCHR: 558 rs->sc_dkdev.dk_copenmask |= pmask; 559 break; 560 561 case S_IFBLK: 562 rs->sc_dkdev.dk_bopenmask |= pmask; 563 break; 564 } 565 566 if ((rs->sc_dkdev.dk_openmask == 0) && 567 ((rs->sc_flags & RAIDF_INITED) != 0)) { 568 /* First one... mark things as dirty... Note that we *MUST* 569 have done a configure before this. I DO NOT WANT TO BE 570 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 571 THAT THEY BELONG TOGETHER!!!!! */ 572 /* XXX should check to see if we're only open for reading 573 here... If so, we needn't do this, but then need some 574 other way of keeping track of what's happened.. */ 575 576 rf_markalldirty( raidPtrs[unit] ); 577 } 578 579 580 rs->sc_dkdev.dk_openmask = 581 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 582 583 raidunlock(rs); 584 585 return (error); 586 587 588 } 589 590 /* ARGSUSED */ 591 int 592 raidclose(devvp, flags, fmt, p) 593 struct vnode *devvp; 594 int flags, fmt; 595 struct proc *p; 596 { 597 struct raid_softc *rs = devvp->v_devcookie; 598 int error = 0; 599 int part; 600 601 if ((error = raidlock(rs)) != 0) 602 return (error); 603 604 part = DISKPART(devvp->v_rdev); 605 606 /* ...that much closer to allowing unconfiguration... */ 607 switch (fmt) { 608 case S_IFCHR: 609 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 610 break; 611 612 case S_IFBLK: 613 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 614 break; 615 } 616 rs->sc_dkdev.dk_openmask = 617 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 618 619 if ((rs->sc_dkdev.dk_openmask == 0) && 620 ((rs->sc_flags & RAIDF_INITED) != 0)) { 621 /* Last one... device is not unconfigured yet. 622 Device shutdown has taken care of setting the 623 clean bits if RAIDF_INITED is not set 624 mark things as clean... */ 625 #if 0 626 printf("Last one on raid%d. Updating status.\n", 627 DISKUNIT(devvp->v_rdev)); 628 #endif 629 rf_update_component_labels(raidPtrs[DISKUNIT(devvp->v_rdev)], 630 RF_FINAL_COMPONENT_UPDATE); 631 if (doing_shutdown) { 632 /* last one, and we're going down, so 633 lights out for this RAID set too. */ 634 error = rf_Shutdown(raidPtrs[DISKUNIT(devvp->v_rdev)]); 635 pool_destroy(&rs->sc_cbufpool); 636 637 /* It's no longer initialized... */ 638 rs->sc_flags &= ~RAIDF_INITED; 639 640 /* Detach the disk. */ 641 disk_detach(&rs->sc_dkdev); 642 } 643 } 644 645 raidunlock(rs); 646 return (0); 647 } 648 649 void 650 raidstrategy(bp) 651 struct buf *bp; 652 { 653 int s; 654 655 unsigned int raidID = DISKUNIT(bp->b_devvp->v_rdev); 656 struct raid_softc *rs = bp->b_devvp->v_devcookie; 657 RF_Raid_t *raidPtr; 658 struct disklabel *lp; 659 int wlabel; 660 661 if ((rs->sc_flags & RAIDF_INITED) ==0) { 662 bp->b_error = ENXIO; 663 bp->b_flags |= B_ERROR; 664 bp->b_resid = bp->b_bcount; 665 biodone(bp); 666 return; 667 } 668 raidPtr = raidPtrs[raidID]; 669 if (raidPtr == NULL || raidPtr->valid == 0) { 670 bp->b_error = ENODEV; 671 bp->b_flags |= B_ERROR; 672 bp->b_resid = bp->b_bcount; 673 biodone(bp); 674 return; 675 } 676 if (bp->b_bcount == 0) { 677 db1_printf(("b_bcount is zero..\n")); 678 biodone(bp); 679 return; 680 } 681 lp = rs->sc_dkdev.dk_label; 682 683 /* 684 * Do bounds checking and adjust transfer. If there's an 685 * error, the bounds check will flag that for us. 686 */ 687 688 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 689 if (DISKPART(bp->b_devvp->v_rdev) != RAW_PART && 690 (bp->b_flags & B_DKLABEL) == 0) { 691 if (bounds_check_with_label(bp, lp, wlabel) <= 0) { 692 db1_printf(("Bounds check failed!!:%d %d\n", 693 (int) bp->b_blkno, (int) wlabel)); 694 biodone(bp); 695 return; 696 } 697 } 698 s = splbio(); 699 700 bp->b_resid = 0; 701 702 /* stuff it onto our queue */ 703 BUFQ_INSERT_TAIL(&rs->buf_queue, bp); 704 705 raidstart(raidPtrs[raidID]); 706 707 splx(s); 708 } 709 710 /* ARGSUSED */ 711 int 712 raidread(devvp, uio, flags) 713 struct vnode *devvp; 714 struct uio *uio; 715 int flags; 716 { 717 struct raid_softc *rs = devvp->v_devcookie; 718 719 if ((rs->sc_flags & RAIDF_INITED) == 0) 720 return (ENXIO); 721 722 db1_printf(("raidread: unit: %d partition: %d\n", 723 DISKUNIT(devvp->v_rdev), DISKPART(devvp->v_rdev))); 724 725 return (physio(raidstrategy, NULL, devvp, B_READ, minphys, uio)); 726 } 727 728 /* ARGSUSED */ 729 int 730 raidwrite(devvp, uio, flags) 731 struct vnode *devvp; 732 struct uio *uio; 733 int flags; 734 { 735 struct raid_softc *rs = devvp->v_devcookie; 736 737 if ((rs->sc_flags & RAIDF_INITED) == 0) 738 return (ENXIO); 739 740 db1_printf(("raidwrite\n")); 741 742 return (physio(raidstrategy, NULL, devvp, B_WRITE, minphys, uio)); 743 } 744 745 int 746 raidioctl(devvp, cmd, data, flag, p) 747 struct vnode *devvp; 748 u_long cmd; 749 caddr_t data; 750 int flag; 751 struct proc *p; 752 { 753 struct raid_softc *rs = devvp->v_devcookie; 754 int error = 0; 755 int part, pmask; 756 RF_Config_t *k_cfg, *u_cfg; 757 RF_Raid_t *raidPtr; 758 RF_RaidDisk_t *diskPtr; 759 RF_AccTotals_t *totals; 760 RF_DeviceConfig_t *d_cfg, **ucfgp; 761 u_char *specific_buf; 762 int retcode = 0; 763 int row; 764 int column; 765 struct rf_recon_req *rrcopy, *rr; 766 RF_ComponentLabel_t *clabel; 767 RF_ComponentLabel_t ci_label; 768 RF_ComponentLabel_t **clabel_ptr; 769 RF_SingleComponent_t *sparePtr,*componentPtr; 770 RF_SingleComponent_t hot_spare; 771 RF_SingleComponent_t component; 772 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 773 int i, j, d; 774 #ifdef __HAVE_OLD_DISKLABEL 775 struct disklabel newlabel; 776 #endif 777 778 raidPtr = raidPtrs[DISKUNIT(devvp->v_rdev)]; 779 780 db1_printf(("raidioctl: 0x%x %d %d %ld\n", devvp->v_rdev, 781 DISKPART(devvp->v_rdev), DISKUNIT(devvp->v_rdev), cmd)); 782 783 /* Must be open for writes for these commands... */ 784 switch (cmd) { 785 case DIOCSDINFO: 786 case DIOCWDINFO: 787 #ifdef __HAVE_OLD_DISKLABEL 788 case ODIOCWDINFO: 789 case ODIOCSDINFO: 790 #endif 791 case DIOCWLABEL: 792 if ((flag & FWRITE) == 0) 793 return (EBADF); 794 } 795 796 /* Must be initialized for these... */ 797 switch (cmd) { 798 case DIOCGDINFO: 799 case DIOCSDINFO: 800 case DIOCWDINFO: 801 #ifdef __HAVE_OLD_DISKLABEL 802 case ODIOCGDINFO: 803 case ODIOCWDINFO: 804 case ODIOCSDINFO: 805 case ODIOCGDEFLABEL: 806 #endif 807 case DIOCGPART: 808 case DIOCWLABEL: 809 case DIOCGDEFLABEL: 810 case RAIDFRAME_SHUTDOWN: 811 case RAIDFRAME_REWRITEPARITY: 812 case RAIDFRAME_GET_INFO: 813 case RAIDFRAME_RESET_ACCTOTALS: 814 case RAIDFRAME_GET_ACCTOTALS: 815 case RAIDFRAME_KEEP_ACCTOTALS: 816 case RAIDFRAME_GET_SIZE: 817 case RAIDFRAME_FAIL_DISK: 818 case RAIDFRAME_COPYBACK: 819 case RAIDFRAME_CHECK_RECON_STATUS: 820 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 821 case RAIDFRAME_GET_COMPONENT_LABEL: 822 case RAIDFRAME_SET_COMPONENT_LABEL: 823 case RAIDFRAME_ADD_HOT_SPARE: 824 case RAIDFRAME_REMOVE_HOT_SPARE: 825 case RAIDFRAME_INIT_LABELS: 826 case RAIDFRAME_REBUILD_IN_PLACE: 827 case RAIDFRAME_CHECK_PARITY: 828 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 829 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 830 case RAIDFRAME_CHECK_COPYBACK_STATUS: 831 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 832 case RAIDFRAME_SET_AUTOCONFIG: 833 case RAIDFRAME_SET_ROOT: 834 case RAIDFRAME_DELETE_COMPONENT: 835 case RAIDFRAME_INCORPORATE_HOT_SPARE: 836 if ((rs->sc_flags & RAIDF_INITED) == 0) 837 return (ENXIO); 838 } 839 840 switch (cmd) { 841 842 /* configure the system */ 843 case RAIDFRAME_CONFIGURE: 844 845 if (raidPtr->valid) { 846 /* There is a valid RAID set running on this unit! */ 847 printf("raid%d: Device already configured!\n", 848 DISKUNIT(devvp->v_rdev)); 849 return(EINVAL); 850 } 851 852 /* copy-in the configuration information */ 853 /* data points to a pointer to the configuration structure */ 854 855 u_cfg = *((RF_Config_t **) data); 856 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *)); 857 if (k_cfg == NULL) { 858 return (ENOMEM); 859 } 860 retcode = copyin((caddr_t) u_cfg, (caddr_t) k_cfg, 861 sizeof(RF_Config_t)); 862 if (retcode) { 863 RF_Free(k_cfg, sizeof(RF_Config_t)); 864 db1_printf(("rf_ioctl: retcode=%d copyin.1\n", 865 retcode)); 866 return (retcode); 867 } 868 /* allocate a buffer for the layout-specific data, and copy it 869 * in */ 870 if (k_cfg->layoutSpecificSize) { 871 if (k_cfg->layoutSpecificSize > 10000) { 872 /* sanity check */ 873 RF_Free(k_cfg, sizeof(RF_Config_t)); 874 return (EINVAL); 875 } 876 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize, 877 (u_char *)); 878 if (specific_buf == NULL) { 879 RF_Free(k_cfg, sizeof(RF_Config_t)); 880 return (ENOMEM); 881 } 882 retcode = copyin(k_cfg->layoutSpecific, 883 (caddr_t) specific_buf, 884 k_cfg->layoutSpecificSize); 885 if (retcode) { 886 RF_Free(k_cfg, sizeof(RF_Config_t)); 887 RF_Free(specific_buf, 888 k_cfg->layoutSpecificSize); 889 db1_printf(("rf_ioctl: retcode=%d copyin.2\n", 890 retcode)); 891 return (retcode); 892 } 893 } else 894 specific_buf = NULL; 895 k_cfg->layoutSpecific = specific_buf; 896 897 /* should do some kind of sanity check on the configuration. 898 * Store the sum of all the bytes in the last byte? */ 899 900 /* configure the system */ 901 902 /* 903 * Clear the entire RAID descriptor, just to make sure 904 * there is no stale data left in the case of a 905 * reconfiguration 906 */ 907 memset((char *) raidPtr, 0, sizeof(RF_Raid_t)); 908 raidPtr->raidid = DISKUNIT(devvp->v_rdev); 909 910 retcode = rf_Configure(raidPtr, k_cfg, NULL); 911 912 if (retcode == 0) { 913 914 /* allow this many simultaneous IO's to 915 this RAID device */ 916 raidPtr->openings = RAIDOUTSTANDING; 917 918 raidinit(raidPtr); 919 rf_markalldirty(raidPtr); 920 } 921 /* free the buffers. No return code here. */ 922 if (k_cfg->layoutSpecificSize) { 923 RF_Free(specific_buf, k_cfg->layoutSpecificSize); 924 } 925 RF_Free(k_cfg, sizeof(RF_Config_t)); 926 927 return (retcode); 928 929 /* shutdown the system */ 930 case RAIDFRAME_SHUTDOWN: 931 932 if ((error = raidlock(rs)) != 0) 933 return (error); 934 935 /* 936 * If somebody has a partition mounted, we shouldn't 937 * shutdown. 938 */ 939 940 part = DISKPART(devvp->v_rdev); 941 pmask = (1 << part); 942 if ((rs->sc_dkdev.dk_openmask & ~pmask) || 943 ((rs->sc_dkdev.dk_bopenmask & pmask) && 944 (rs->sc_dkdev.dk_copenmask & pmask))) { 945 raidunlock(rs); 946 return (EBUSY); 947 } 948 949 retcode = rf_Shutdown(raidPtr); 950 951 pool_destroy(&rs->sc_cbufpool); 952 953 /* It's no longer initialized... */ 954 rs->sc_flags &= ~RAIDF_INITED; 955 956 /* Detach the disk. */ 957 disk_detach(&rs->sc_dkdev); 958 959 raidunlock(rs); 960 961 return (retcode); 962 case RAIDFRAME_GET_COMPONENT_LABEL: 963 clabel_ptr = (RF_ComponentLabel_t **) data; 964 /* need to read the component label for the disk indicated 965 by row,column in clabel */ 966 967 /* For practice, let's get it directly fromdisk, rather 968 than from the in-core copy */ 969 RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ), 970 (RF_ComponentLabel_t *)); 971 if (clabel == NULL) 972 return (ENOMEM); 973 974 memset((char *) clabel, 0, sizeof(RF_ComponentLabel_t)); 975 976 retcode = copyin( *clabel_ptr, clabel, 977 sizeof(RF_ComponentLabel_t)); 978 979 if (retcode) { 980 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 981 return(retcode); 982 } 983 984 row = clabel->row; 985 column = clabel->column; 986 987 if ((row < 0) || (row >= raidPtr->numRow) || 988 (column < 0) || (column >= raidPtr->numCol + 989 raidPtr->numSpare)) { 990 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 991 return(EINVAL); 992 } 993 994 raidread_component_label(raidPtr->raid_cinfo[row][column].ci_vp, 995 clabel); 996 997 retcode = copyout((caddr_t) clabel, 998 (caddr_t) *clabel_ptr, 999 sizeof(RF_ComponentLabel_t)); 1000 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1001 return (retcode); 1002 1003 case RAIDFRAME_SET_COMPONENT_LABEL: 1004 clabel = (RF_ComponentLabel_t *) data; 1005 1006 /* XXX check the label for valid stuff... */ 1007 /* Note that some things *should not* get modified -- 1008 the user should be re-initing the labels instead of 1009 trying to patch things. 1010 */ 1011 1012 printf("Got component label:\n"); 1013 printf("Version: %d\n",clabel->version); 1014 printf("Serial Number: %d\n",clabel->serial_number); 1015 printf("Mod counter: %d\n",clabel->mod_counter); 1016 printf("Row: %d\n", clabel->row); 1017 printf("Column: %d\n", clabel->column); 1018 printf("Num Rows: %d\n", clabel->num_rows); 1019 printf("Num Columns: %d\n", clabel->num_columns); 1020 printf("Clean: %d\n", clabel->clean); 1021 printf("Status: %d\n", clabel->status); 1022 1023 row = clabel->row; 1024 column = clabel->column; 1025 1026 if ((row < 0) || (row >= raidPtr->numRow) || 1027 (column < 0) || (column >= raidPtr->numCol)) { 1028 return(EINVAL); 1029 } 1030 1031 /* XXX this isn't allowed to do anything for now :-) */ 1032 1033 /* XXX and before it is, we need to fill in the rest 1034 of the fields!?!?!?! */ 1035 #if 0 1036 raidwrite_component_label( 1037 raidPtr->raid_cinfo[row][column].ci_vp, clabel); 1038 #endif 1039 return (0); 1040 1041 case RAIDFRAME_INIT_LABELS: 1042 clabel = (RF_ComponentLabel_t *) data; 1043 /* 1044 we only want the serial number from 1045 the above. We get all the rest of the information 1046 from the config that was used to create this RAID 1047 set. 1048 */ 1049 1050 raidPtr->serial_number = clabel->serial_number; 1051 1052 raid_init_component_label(raidPtr, &ci_label); 1053 ci_label.serial_number = clabel->serial_number; 1054 1055 for(row=0;row<raidPtr->numRow;row++) { 1056 ci_label.row = row; 1057 for(column=0;column<raidPtr->numCol;column++) { 1058 diskPtr = &raidPtr->Disks[row][column]; 1059 if (!RF_DEAD_DISK(diskPtr->status)) { 1060 ci_label.partitionSize = diskPtr->partitionSize; 1061 ci_label.column = column; 1062 raidwrite_component_label( 1063 raidPtr->raid_cinfo[row][column].ci_vp, 1064 &ci_label ); 1065 } 1066 } 1067 } 1068 1069 return (retcode); 1070 case RAIDFRAME_SET_AUTOCONFIG: 1071 d = rf_set_autoconfig(raidPtr, *(int *) data); 1072 printf("New autoconfig value is: %d\n", d); 1073 *(int *) data = d; 1074 return (retcode); 1075 1076 case RAIDFRAME_SET_ROOT: 1077 d = rf_set_rootpartition(raidPtr, *(int *) data); 1078 printf("New rootpartition value is: %d\n", d); 1079 *(int *) data = d; 1080 return (retcode); 1081 1082 /* initialize all parity */ 1083 case RAIDFRAME_REWRITEPARITY: 1084 1085 if (raidPtr->Layout.map->faultsTolerated == 0) { 1086 /* Parity for RAID 0 is trivially correct */ 1087 raidPtr->parity_good = RF_RAID_CLEAN; 1088 return(0); 1089 } 1090 1091 if (raidPtr->parity_rewrite_in_progress == 1) { 1092 /* Re-write is already in progress! */ 1093 return(EINVAL); 1094 } 1095 1096 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1097 rf_RewriteParityThread, 1098 raidPtr,"raid_parity"); 1099 return (retcode); 1100 1101 1102 case RAIDFRAME_ADD_HOT_SPARE: 1103 sparePtr = (RF_SingleComponent_t *) data; 1104 memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t)); 1105 retcode = rf_add_hot_spare(raidPtr, &hot_spare); 1106 return(retcode); 1107 1108 case RAIDFRAME_REMOVE_HOT_SPARE: 1109 return(retcode); 1110 1111 case RAIDFRAME_DELETE_COMPONENT: 1112 componentPtr = (RF_SingleComponent_t *)data; 1113 memcpy( &component, componentPtr, 1114 sizeof(RF_SingleComponent_t)); 1115 retcode = rf_delete_component(raidPtr, &component); 1116 return(retcode); 1117 1118 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1119 componentPtr = (RF_SingleComponent_t *)data; 1120 memcpy( &component, componentPtr, 1121 sizeof(RF_SingleComponent_t)); 1122 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1123 return(retcode); 1124 1125 case RAIDFRAME_REBUILD_IN_PLACE: 1126 1127 if (raidPtr->Layout.map->faultsTolerated == 0) { 1128 /* Can't do this on a RAID 0!! */ 1129 return(EINVAL); 1130 } 1131 1132 if (raidPtr->recon_in_progress == 1) { 1133 /* a reconstruct is already in progress! */ 1134 return(EINVAL); 1135 } 1136 1137 componentPtr = (RF_SingleComponent_t *) data; 1138 memcpy( &component, componentPtr, 1139 sizeof(RF_SingleComponent_t)); 1140 row = component.row; 1141 column = component.column; 1142 printf("Rebuild: %d %d\n",row, column); 1143 if ((row < 0) || (row >= raidPtr->numRow) || 1144 (column < 0) || (column >= raidPtr->numCol)) { 1145 return(EINVAL); 1146 } 1147 1148 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1149 if (rrcopy == NULL) 1150 return(ENOMEM); 1151 1152 rrcopy->raidPtr = (void *) raidPtr; 1153 rrcopy->row = row; 1154 rrcopy->col = column; 1155 1156 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1157 rf_ReconstructInPlaceThread, 1158 rrcopy,"raid_reconip"); 1159 return(retcode); 1160 1161 case RAIDFRAME_GET_INFO: 1162 if (!raidPtr->valid) 1163 return (ENODEV); 1164 ucfgp = (RF_DeviceConfig_t **) data; 1165 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1166 (RF_DeviceConfig_t *)); 1167 if (d_cfg == NULL) 1168 return (ENOMEM); 1169 memset((char *) d_cfg, 0, sizeof(RF_DeviceConfig_t)); 1170 d_cfg->rows = raidPtr->numRow; 1171 d_cfg->cols = raidPtr->numCol; 1172 d_cfg->ndevs = raidPtr->numRow * raidPtr->numCol; 1173 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1174 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1175 return (ENOMEM); 1176 } 1177 d_cfg->nspares = raidPtr->numSpare; 1178 if (d_cfg->nspares >= RF_MAX_DISKS) { 1179 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1180 return (ENOMEM); 1181 } 1182 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1183 d = 0; 1184 for (i = 0; i < d_cfg->rows; i++) { 1185 for (j = 0; j < d_cfg->cols; j++) { 1186 d_cfg->devs[d] = raidPtr->Disks[i][j]; 1187 d++; 1188 } 1189 } 1190 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1191 d_cfg->spares[i] = raidPtr->Disks[0][j]; 1192 } 1193 retcode = copyout((caddr_t) d_cfg, (caddr_t) * ucfgp, 1194 sizeof(RF_DeviceConfig_t)); 1195 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1196 1197 return (retcode); 1198 1199 case RAIDFRAME_CHECK_PARITY: 1200 *(int *) data = raidPtr->parity_good; 1201 return (0); 1202 1203 case RAIDFRAME_RESET_ACCTOTALS: 1204 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1205 return (0); 1206 1207 case RAIDFRAME_GET_ACCTOTALS: 1208 totals = (RF_AccTotals_t *) data; 1209 *totals = raidPtr->acc_totals; 1210 return (0); 1211 1212 case RAIDFRAME_KEEP_ACCTOTALS: 1213 raidPtr->keep_acc_totals = *(int *)data; 1214 return (0); 1215 1216 case RAIDFRAME_GET_SIZE: 1217 *(int *) data = raidPtr->totalSectors; 1218 return (0); 1219 1220 /* fail a disk & optionally start reconstruction */ 1221 case RAIDFRAME_FAIL_DISK: 1222 1223 if (raidPtr->Layout.map->faultsTolerated == 0) { 1224 /* Can't do this on a RAID 0!! */ 1225 return(EINVAL); 1226 } 1227 1228 rr = (struct rf_recon_req *) data; 1229 1230 if (rr->row < 0 || rr->row >= raidPtr->numRow 1231 || rr->col < 0 || rr->col >= raidPtr->numCol) 1232 return (EINVAL); 1233 1234 printf("raid%d: Failing the disk: row: %d col: %d\n", 1235 DISKUNIT(devvp->v_rdev), rr->row, rr->col); 1236 1237 /* make a copy of the recon request so that we don't rely on 1238 * the user's buffer */ 1239 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1240 if (rrcopy == NULL) 1241 return(ENOMEM); 1242 bcopy(rr, rrcopy, sizeof(*rr)); 1243 rrcopy->raidPtr = (void *) raidPtr; 1244 1245 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1246 rf_ReconThread, 1247 rrcopy,"raid_recon"); 1248 return (0); 1249 1250 /* invoke a copyback operation after recon on whatever disk 1251 * needs it, if any */ 1252 case RAIDFRAME_COPYBACK: 1253 1254 if (raidPtr->Layout.map->faultsTolerated == 0) { 1255 /* This makes no sense on a RAID 0!! */ 1256 return(EINVAL); 1257 } 1258 1259 if (raidPtr->copyback_in_progress == 1) { 1260 /* Copyback is already in progress! */ 1261 return(EINVAL); 1262 } 1263 1264 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1265 rf_CopybackThread, 1266 raidPtr,"raid_copyback"); 1267 return (retcode); 1268 1269 /* return the percentage completion of reconstruction */ 1270 case RAIDFRAME_CHECK_RECON_STATUS: 1271 if (raidPtr->Layout.map->faultsTolerated == 0) { 1272 /* This makes no sense on a RAID 0, so tell the 1273 user it's done. */ 1274 *(int *) data = 100; 1275 return(0); 1276 } 1277 row = 0; /* XXX we only consider a single row... */ 1278 if (raidPtr->status[row] != rf_rs_reconstructing) 1279 *(int *) data = 100; 1280 else 1281 *(int *) data = raidPtr->reconControl[row]->percentComplete; 1282 return (0); 1283 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1284 progressInfoPtr = (RF_ProgressInfo_t **) data; 1285 row = 0; /* XXX we only consider a single row... */ 1286 if (raidPtr->status[row] != rf_rs_reconstructing) { 1287 progressInfo.remaining = 0; 1288 progressInfo.completed = 100; 1289 progressInfo.total = 100; 1290 } else { 1291 progressInfo.total = 1292 raidPtr->reconControl[row]->numRUsTotal; 1293 progressInfo.completed = 1294 raidPtr->reconControl[row]->numRUsComplete; 1295 progressInfo.remaining = progressInfo.total - 1296 progressInfo.completed; 1297 } 1298 retcode = copyout((caddr_t) &progressInfo, 1299 (caddr_t) *progressInfoPtr, 1300 sizeof(RF_ProgressInfo_t)); 1301 return (retcode); 1302 1303 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1304 if (raidPtr->Layout.map->faultsTolerated == 0) { 1305 /* This makes no sense on a RAID 0, so tell the 1306 user it's done. */ 1307 *(int *) data = 100; 1308 return(0); 1309 } 1310 if (raidPtr->parity_rewrite_in_progress == 1) { 1311 *(int *) data = 100 * 1312 raidPtr->parity_rewrite_stripes_done / 1313 raidPtr->Layout.numStripe; 1314 } else { 1315 *(int *) data = 100; 1316 } 1317 return (0); 1318 1319 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1320 progressInfoPtr = (RF_ProgressInfo_t **) data; 1321 if (raidPtr->parity_rewrite_in_progress == 1) { 1322 progressInfo.total = raidPtr->Layout.numStripe; 1323 progressInfo.completed = 1324 raidPtr->parity_rewrite_stripes_done; 1325 progressInfo.remaining = progressInfo.total - 1326 progressInfo.completed; 1327 } else { 1328 progressInfo.remaining = 0; 1329 progressInfo.completed = 100; 1330 progressInfo.total = 100; 1331 } 1332 retcode = copyout((caddr_t) &progressInfo, 1333 (caddr_t) *progressInfoPtr, 1334 sizeof(RF_ProgressInfo_t)); 1335 return (retcode); 1336 1337 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1338 if (raidPtr->Layout.map->faultsTolerated == 0) { 1339 /* This makes no sense on a RAID 0 */ 1340 *(int *) data = 100; 1341 return(0); 1342 } 1343 if (raidPtr->copyback_in_progress == 1) { 1344 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1345 raidPtr->Layout.numStripe; 1346 } else { 1347 *(int *) data = 100; 1348 } 1349 return (0); 1350 1351 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1352 progressInfoPtr = (RF_ProgressInfo_t **) data; 1353 if (raidPtr->copyback_in_progress == 1) { 1354 progressInfo.total = raidPtr->Layout.numStripe; 1355 progressInfo.completed = 1356 raidPtr->copyback_stripes_done; 1357 progressInfo.remaining = progressInfo.total - 1358 progressInfo.completed; 1359 } else { 1360 progressInfo.remaining = 0; 1361 progressInfo.completed = 100; 1362 progressInfo.total = 100; 1363 } 1364 retcode = copyout((caddr_t) &progressInfo, 1365 (caddr_t) *progressInfoPtr, 1366 sizeof(RF_ProgressInfo_t)); 1367 return (retcode); 1368 1369 /* the sparetable daemon calls this to wait for the kernel to 1370 * need a spare table. this ioctl does not return until a 1371 * spare table is needed. XXX -- calling mpsleep here in the 1372 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1373 * -- I should either compute the spare table in the kernel, 1374 * or have a different -- XXX XXX -- interface (a different 1375 * character device) for delivering the table -- XXX */ 1376 #if 0 1377 case RAIDFRAME_SPARET_WAIT: 1378 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1379 while (!rf_sparet_wait_queue) 1380 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE); 1381 waitreq = rf_sparet_wait_queue; 1382 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1383 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1384 1385 /* structure assignment */ 1386 *((RF_SparetWait_t *) data) = *waitreq; 1387 1388 RF_Free(waitreq, sizeof(*waitreq)); 1389 return (0); 1390 1391 /* wakes up a process waiting on SPARET_WAIT and puts an error 1392 * code in it that will cause the dameon to exit */ 1393 case RAIDFRAME_ABORT_SPARET_WAIT: 1394 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1395 waitreq->fcol = -1; 1396 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1397 waitreq->next = rf_sparet_wait_queue; 1398 rf_sparet_wait_queue = waitreq; 1399 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1400 wakeup(&rf_sparet_wait_queue); 1401 return (0); 1402 1403 /* used by the spare table daemon to deliver a spare table 1404 * into the kernel */ 1405 case RAIDFRAME_SEND_SPARET: 1406 1407 /* install the spare table */ 1408 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1409 1410 /* respond to the requestor. the return status of the spare 1411 * table installation is passed in the "fcol" field */ 1412 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1413 waitreq->fcol = retcode; 1414 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1415 waitreq->next = rf_sparet_resp_queue; 1416 rf_sparet_resp_queue = waitreq; 1417 wakeup(&rf_sparet_resp_queue); 1418 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1419 1420 return (retcode); 1421 #endif 1422 1423 default: 1424 break; /* fall through to the os-specific code below */ 1425 1426 } 1427 1428 if (!raidPtr->valid) 1429 return (EINVAL); 1430 1431 /* 1432 * Add support for "regular" device ioctls here. 1433 */ 1434 1435 switch (cmd) { 1436 case DIOCGDINFO: 1437 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1438 break; 1439 #ifdef __HAVE_OLD_DISKLABEL 1440 case ODIOCGDINFO: 1441 newlabel = *(rs->sc_dkdev.dk_label); 1442 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1443 return ENOTTY; 1444 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1445 break; 1446 #endif 1447 1448 case DIOCGPART: 1449 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1450 ((struct partinfo *) data)->part = 1451 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(devvp->v_rdev)]; 1452 break; 1453 1454 case DIOCWDINFO: 1455 case DIOCSDINFO: 1456 #ifdef __HAVE_OLD_DISKLABEL 1457 case ODIOCWDINFO: 1458 case ODIOCSDINFO: 1459 #endif 1460 { 1461 struct disklabel *lp; 1462 #ifdef __HAVE_OLD_DISKLABEL 1463 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1464 memset(&newlabel, 0, sizeof newlabel); 1465 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1466 lp = &newlabel; 1467 } else 1468 #endif 1469 lp = (struct disklabel *)data; 1470 1471 if ((error = raidlock(rs)) != 0) 1472 return (error); 1473 1474 rs->sc_flags |= RAIDF_LABELLING; 1475 1476 error = setdisklabel(rs->sc_dkdev.dk_label, 1477 lp, 0, rs->sc_dkdev.dk_cpulabel); 1478 if (error == 0) { 1479 if (cmd == DIOCWDINFO 1480 #ifdef __HAVE_OLD_DISKLABEL 1481 || cmd == ODIOCWDINFO 1482 #endif 1483 ) 1484 error = writedisklabel(devvp, raidstrategy, 1485 rs->sc_dkdev.dk_label, 1486 rs->sc_dkdev.dk_cpulabel); 1487 } 1488 rs->sc_flags &= ~RAIDF_LABELLING; 1489 1490 raidunlock(rs); 1491 1492 if (error) 1493 return (error); 1494 break; 1495 } 1496 1497 case DIOCWLABEL: 1498 if (*(int *) data != 0) 1499 rs->sc_flags |= RAIDF_WLABEL; 1500 else 1501 rs->sc_flags &= ~RAIDF_WLABEL; 1502 break; 1503 1504 case DIOCGDEFLABEL: 1505 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1506 break; 1507 1508 #ifdef __HAVE_OLD_DISKLABEL 1509 case ODIOCGDEFLABEL: 1510 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1511 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1512 return ENOTTY; 1513 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1514 break; 1515 #endif 1516 1517 default: 1518 retcode = ENOTTY; 1519 } 1520 return (retcode); 1521 } 1522 1523 1524 /* raidinit -- complete the rest of the initialization for the 1525 RAIDframe device. */ 1526 1527 1528 static void 1529 raidinit(raidPtr) 1530 RF_Raid_t *raidPtr; 1531 { 1532 struct raid_softc *rs; 1533 int unit; 1534 1535 unit = raidPtr->raidid; 1536 1537 rs = &raid_softc[unit]; 1538 pool_init(&rs->sc_cbufpool, sizeof(struct raidbuf), 0, 1539 0, 0, "raidpl", 0, NULL, NULL, M_RAIDFRAME); 1540 1541 1542 /* XXX should check return code first... */ 1543 rs->sc_flags |= RAIDF_INITED; 1544 1545 sprintf(rs->sc_xname, "raid%d", unit); /* XXX doesn't check bounds. */ 1546 1547 rs->sc_dkdev.dk_name = rs->sc_xname; 1548 1549 /* disk_attach actually creates space for the CPU disklabel, among 1550 * other things, so it's critical to call this *BEFORE* we try putzing 1551 * with disklabels. */ 1552 1553 disk_attach(&rs->sc_dkdev); 1554 1555 /* XXX There may be a weird interaction here between this, and 1556 * protectedSectors, as used in RAIDframe. */ 1557 1558 rs->sc_size = raidPtr->totalSectors; 1559 1560 } 1561 1562 /* wake up the daemon & tell it to get us a spare table 1563 * XXX 1564 * the entries in the queues should be tagged with the raidPtr 1565 * so that in the extremely rare case that two recons happen at once, 1566 * we know for which device were requesting a spare table 1567 * XXX 1568 * 1569 * XXX This code is not currently used. GO 1570 */ 1571 int 1572 rf_GetSpareTableFromDaemon(req) 1573 RF_SparetWait_t *req; 1574 { 1575 int retcode; 1576 1577 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1578 req->next = rf_sparet_wait_queue; 1579 rf_sparet_wait_queue = req; 1580 wakeup(&rf_sparet_wait_queue); 1581 1582 /* mpsleep unlocks the mutex */ 1583 while (!rf_sparet_resp_queue) { 1584 tsleep(&rf_sparet_resp_queue, PRIBIO, 1585 "raidframe getsparetable", 0); 1586 } 1587 req = rf_sparet_resp_queue; 1588 rf_sparet_resp_queue = req->next; 1589 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1590 1591 retcode = req->fcol; 1592 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1593 * alloc'd */ 1594 return (retcode); 1595 } 1596 1597 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1598 * bp & passes it down. 1599 * any calls originating in the kernel must use non-blocking I/O 1600 * do some extra sanity checking to return "appropriate" error values for 1601 * certain conditions (to make some standard utilities work) 1602 * 1603 * Formerly known as: rf_DoAccessKernel 1604 */ 1605 void 1606 raidstart(raidPtr) 1607 RF_Raid_t *raidPtr; 1608 { 1609 RF_SectorCount_t num_blocks, pb, sum; 1610 RF_RaidAddr_t raid_addr; 1611 int retcode; 1612 struct partition *pp; 1613 daddr_t blocknum; 1614 int unit; 1615 struct raid_softc *rs; 1616 int do_async; 1617 struct buf *bp; 1618 1619 unit = raidPtr->raidid; 1620 rs = &raid_softc[unit]; 1621 1622 /* quick check to see if anything has died recently */ 1623 RF_LOCK_MUTEX(raidPtr->mutex); 1624 if (raidPtr->numNewFailures > 0) { 1625 rf_update_component_labels(raidPtr, 1626 RF_NORMAL_COMPONENT_UPDATE); 1627 raidPtr->numNewFailures--; 1628 } 1629 RF_UNLOCK_MUTEX(raidPtr->mutex); 1630 1631 /* Check to see if we're at the limit... */ 1632 RF_LOCK_MUTEX(raidPtr->mutex); 1633 while (raidPtr->openings > 0) { 1634 RF_UNLOCK_MUTEX(raidPtr->mutex); 1635 1636 /* get the next item, if any, from the queue */ 1637 if ((bp = BUFQ_FIRST(&rs->buf_queue)) == NULL) { 1638 /* nothing more to do */ 1639 return; 1640 } 1641 BUFQ_REMOVE(&rs->buf_queue, bp); 1642 1643 /* Ok, for the bp we have here, bp->b_blkno is relative to the 1644 * partition.. Need to make it absolute to the underlying 1645 * device.. */ 1646 1647 blocknum = bp->b_blkno; 1648 if (DISKPART(bp->b_devvp->v_rdev) != RAW_PART && 1649 (bp->b_flags & B_DKLABEL) == 0) { 1650 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_devvp->v_rdev)]; 1651 blocknum += pp->p_offset; 1652 } 1653 1654 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 1655 (int) blocknum)); 1656 1657 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 1658 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 1659 1660 /* *THIS* is where we adjust what block we're going to... 1661 * but DO NOT TOUCH bp->b_blkno!!! */ 1662 raid_addr = blocknum; 1663 1664 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 1665 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 1666 sum = raid_addr + num_blocks + pb; 1667 if (1 || rf_debugKernelAccess) { 1668 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 1669 (int) raid_addr, (int) sum, (int) num_blocks, 1670 (int) pb, (int) bp->b_resid)); 1671 } 1672 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 1673 || (sum < num_blocks) || (sum < pb)) { 1674 bp->b_error = ENOSPC; 1675 bp->b_flags |= B_ERROR; 1676 bp->b_resid = bp->b_bcount; 1677 biodone(bp); 1678 RF_LOCK_MUTEX(raidPtr->mutex); 1679 continue; 1680 } 1681 /* 1682 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 1683 */ 1684 1685 if (bp->b_bcount & raidPtr->sectorMask) { 1686 bp->b_error = EINVAL; 1687 bp->b_flags |= B_ERROR; 1688 bp->b_resid = bp->b_bcount; 1689 biodone(bp); 1690 RF_LOCK_MUTEX(raidPtr->mutex); 1691 continue; 1692 1693 } 1694 db1_printf(("Calling DoAccess..\n")); 1695 1696 1697 RF_LOCK_MUTEX(raidPtr->mutex); 1698 raidPtr->openings--; 1699 RF_UNLOCK_MUTEX(raidPtr->mutex); 1700 1701 /* 1702 * Everything is async. 1703 */ 1704 do_async = 1; 1705 1706 disk_busy(&rs->sc_dkdev); 1707 1708 /* XXX we're still at splbio() here... do we *really* 1709 need to be? */ 1710 1711 /* don't ever condition on bp->b_flags & B_WRITE. 1712 * always condition on B_READ instead */ 1713 1714 retcode = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 1715 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 1716 do_async, raid_addr, num_blocks, 1717 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 1718 1719 RF_LOCK_MUTEX(raidPtr->mutex); 1720 } 1721 RF_UNLOCK_MUTEX(raidPtr->mutex); 1722 } 1723 1724 /* 1725 * invoke an I/O from kernel mode. Disk queue should be 1726 * locked upon entry 1727 */ 1728 int 1729 rf_DispatchKernelIO(queue, req) 1730 RF_DiskQueue_t *queue; 1731 RF_DiskQueueData_t *req; 1732 { 1733 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 1734 struct buf *bp; 1735 struct raidbuf *raidbp = NULL; 1736 struct raid_softc *rs; 1737 int unit; 1738 int s; 1739 1740 s=0; 1741 /* s = splbio();*/ /* want to test this */ 1742 /* XXX along with the vnode, we also need the softc associated with 1743 * this device.. */ 1744 1745 req->queue = queue; 1746 1747 unit = queue->raidPtr->raidid; 1748 1749 db1_printf(("DispatchKernelIO unit: %d\n", unit)); 1750 1751 if (unit >= numraid) { 1752 printf("Invalid unit number: %d %d\n", unit, numraid); 1753 panic("Invalid Unit number in rf_DispatchKernelIO\n"); 1754 } 1755 rs = &raid_softc[unit]; 1756 1757 bp = req->bp; 1758 #if 1 1759 /* XXX when there is a physical disk failure, someone is passing us a 1760 * buffer that contains old stuff!! Attempt to deal with this problem 1761 * without taking a performance hit... (not sure where the real bug 1762 * is. It's buried in RAIDframe somewhere) :-( GO ) */ 1763 1764 if (bp->b_flags & B_ERROR) { 1765 bp->b_flags &= ~B_ERROR; 1766 } 1767 if (bp->b_error != 0) { 1768 bp->b_error = 0; 1769 } 1770 #endif 1771 raidbp = RAIDGETBUF(rs); 1772 1773 raidbp->rf_flags = 0; /* XXX not really used anywhere... */ 1774 1775 /* 1776 * context for raidiodone 1777 */ 1778 raidbp->rf_obp = bp; 1779 raidbp->req = req; 1780 1781 LIST_INIT(&raidbp->rf_buf.b_dep); 1782 1783 switch (req->type) { 1784 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 1785 /* XXX need to do something extra here.. */ 1786 /* I'm leaving this in, as I've never actually seen it used, 1787 * and I'd like folks to report it... GO */ 1788 printf(("WAKEUP CALLED\n")); 1789 queue->numOutstanding++; 1790 1791 /* XXX need to glue the original buffer into this?? */ 1792 1793 KernelWakeupFunc(&raidbp->rf_buf); 1794 break; 1795 1796 case RF_IO_TYPE_READ: 1797 case RF_IO_TYPE_WRITE: 1798 1799 if (req->tracerec) { 1800 RF_ETIMER_START(req->tracerec->timer); 1801 } 1802 InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp, 1803 op | bp->b_flags, req->sectorOffset, req->numSector, 1804 req->buf, KernelWakeupFunc, (void *) req, 1805 queue->raidPtr->logBytesPerSector, req->b_proc); 1806 1807 if (rf_debugKernelAccess) { 1808 db1_printf(("dispatch: bp->b_blkno = %ld\n", 1809 (long) bp->b_blkno)); 1810 } 1811 queue->numOutstanding++; 1812 queue->last_deq_sector = req->sectorOffset; 1813 /* acc wouldn't have been let in if there were any pending 1814 * reqs at any other priority */ 1815 queue->curPriority = req->priority; 1816 1817 db1_printf(("Going for %c to unit %d row %d col %d\n", 1818 req->type, unit, queue->row, queue->col)); 1819 db1_printf(("sector %d count %d (%d bytes) %d\n", 1820 (int) req->sectorOffset, (int) req->numSector, 1821 (int) (req->numSector << 1822 queue->raidPtr->logBytesPerSector), 1823 (int) queue->raidPtr->logBytesPerSector)); 1824 if ((raidbp->rf_buf.b_flags & B_READ) == 0) { 1825 raidbp->rf_buf.b_vp->v_numoutput++; 1826 } 1827 VOP_STRATEGY(&raidbp->rf_buf); 1828 1829 break; 1830 1831 default: 1832 panic("bad req->type in rf_DispatchKernelIO"); 1833 } 1834 db1_printf(("Exiting from DispatchKernelIO\n")); 1835 /* splx(s); */ /* want to test this */ 1836 return (0); 1837 } 1838 /* this is the callback function associated with a I/O invoked from 1839 kernel code. 1840 */ 1841 static void 1842 KernelWakeupFunc(vbp) 1843 struct buf *vbp; 1844 { 1845 RF_DiskQueueData_t *req = NULL; 1846 RF_DiskQueue_t *queue; 1847 struct raidbuf *raidbp = (struct raidbuf *) vbp; 1848 struct buf *bp; 1849 struct raid_softc *rs; 1850 int unit; 1851 int s; 1852 1853 s = splbio(); 1854 db1_printf(("recovering the request queue:\n")); 1855 req = raidbp->req; 1856 1857 bp = raidbp->rf_obp; 1858 1859 queue = (RF_DiskQueue_t *) req->queue; 1860 1861 if (raidbp->rf_buf.b_flags & B_ERROR) { 1862 bp->b_flags |= B_ERROR; 1863 bp->b_error = raidbp->rf_buf.b_error ? 1864 raidbp->rf_buf.b_error : EIO; 1865 } 1866 1867 /* XXX methinks this could be wrong... */ 1868 #if 1 1869 bp->b_resid = raidbp->rf_buf.b_resid; 1870 #endif 1871 1872 if (req->tracerec) { 1873 RF_ETIMER_STOP(req->tracerec->timer); 1874 RF_ETIMER_EVAL(req->tracerec->timer); 1875 RF_LOCK_MUTEX(rf_tracing_mutex); 1876 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 1877 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 1878 req->tracerec->num_phys_ios++; 1879 RF_UNLOCK_MUTEX(rf_tracing_mutex); 1880 } 1881 bp->b_bcount = raidbp->rf_buf.b_bcount; /* XXXX ?? */ 1882 1883 unit = queue->raidPtr->raidid; /* *Much* simpler :-> */ 1884 1885 1886 /* XXX Ok, let's get aggressive... If B_ERROR is set, let's go 1887 * ballistic, and mark the component as hosed... */ 1888 1889 if (bp->b_flags & B_ERROR) { 1890 /* Mark the disk as dead */ 1891 /* but only mark it once... */ 1892 if (queue->raidPtr->Disks[queue->row][queue->col].status == 1893 rf_ds_optimal) { 1894 printf("raid%d: IO Error. Marking %s as failed.\n", 1895 unit, queue->raidPtr->Disks[queue->row][queue->col].devname); 1896 queue->raidPtr->Disks[queue->row][queue->col].status = 1897 rf_ds_failed; 1898 queue->raidPtr->status[queue->row] = rf_rs_degraded; 1899 queue->raidPtr->numFailures++; 1900 queue->raidPtr->numNewFailures++; 1901 } else { /* Disk is already dead... */ 1902 /* printf("Disk already marked as dead!\n"); */ 1903 } 1904 1905 } 1906 1907 rs = &raid_softc[unit]; 1908 RAIDPUTBUF(rs, raidbp); 1909 1910 rf_DiskIOComplete(queue, req, (bp->b_flags & B_ERROR) ? 1 : 0); 1911 (req->CompleteFunc) (req->argument, (bp->b_flags & B_ERROR) ? 1 : 0); 1912 1913 splx(s); 1914 } 1915 1916 /* 1917 * initialize a buf structure for doing an I/O in the kernel. 1918 */ 1919 static void 1920 InitBP(bp, b_vp, rw_flag, startSect, numSect, buf, cbFunc, cbArg, 1921 logBytesPerSector, b_proc) 1922 struct buf *bp; 1923 struct vnode *b_vp; 1924 unsigned rw_flag; 1925 RF_SectorNum_t startSect; 1926 RF_SectorCount_t numSect; 1927 caddr_t buf; 1928 void (*cbFunc) (struct buf *); 1929 void *cbArg; 1930 int logBytesPerSector; 1931 struct proc *b_proc; 1932 { 1933 /* bp->b_flags = B_PHYS | rw_flag; */ 1934 bp->b_flags = B_CALL | rw_flag; /* XXX need B_PHYS here too??? */ 1935 bp->b_bcount = numSect << logBytesPerSector; 1936 bp->b_bufsize = bp->b_bcount; 1937 bp->b_error = 0; 1938 bp->b_devvp = b_vp; 1939 bp->b_data = buf; 1940 bp->b_blkno = startSect; 1941 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 1942 if (bp->b_bcount == 0) { 1943 panic("bp->b_bcount is zero in InitBP!!\n"); 1944 } 1945 bp->b_proc = b_proc; 1946 bp->b_iodone = cbFunc; 1947 bp->b_vp = b_vp; 1948 1949 } 1950 1951 static void 1952 raidgetdefaultlabel(raidPtr, rs, lp) 1953 RF_Raid_t *raidPtr; 1954 struct raid_softc *rs; 1955 struct disklabel *lp; 1956 { 1957 db1_printf(("Building a default label...\n")); 1958 memset(lp, 0, sizeof(*lp)); 1959 1960 /* fabricate a label... */ 1961 lp->d_secperunit = raidPtr->totalSectors; 1962 lp->d_secsize = raidPtr->bytesPerSector; 1963 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 1964 lp->d_ntracks = 4 * raidPtr->numCol; 1965 lp->d_ncylinders = raidPtr->totalSectors / 1966 (lp->d_nsectors * lp->d_ntracks); 1967 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1968 1969 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 1970 lp->d_type = DTYPE_RAID; 1971 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1972 lp->d_rpm = 3600; 1973 lp->d_interleave = 1; 1974 lp->d_flags = 0; 1975 1976 lp->d_partitions[RAW_PART].p_offset = 0; 1977 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 1978 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 1979 lp->d_npartitions = RAW_PART + 1; 1980 1981 lp->d_magic = DISKMAGIC; 1982 lp->d_magic2 = DISKMAGIC; 1983 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 1984 1985 } 1986 /* 1987 * Read the disklabel from the raid device. If one is not present, fake one 1988 * up. 1989 */ 1990 static void 1991 raidgetdisklabel(devvp) 1992 struct vnode *devvp; 1993 { 1994 struct raid_softc *rs = devvp->v_devcookie; 1995 char *errstring; 1996 struct disklabel *lp = rs->sc_dkdev.dk_label; 1997 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 1998 RF_Raid_t *raidPtr; 1999 2000 db1_printf(("Getting the disklabel...\n")); 2001 2002 memset(clp, 0, sizeof(*clp)); 2003 2004 raidPtr = raidPtrs[DISKUNIT(devvp->v_rdev)]; 2005 2006 raidgetdefaultlabel(raidPtr, rs, lp); 2007 2008 /* 2009 * Call the generic disklabel extraction routine. 2010 */ 2011 errstring = readdisklabel(devvp, raidstrategy, 2012 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2013 if (errstring) 2014 raidmakedisklabel(rs); 2015 else { 2016 int i; 2017 struct partition *pp; 2018 2019 /* 2020 * Sanity check whether the found disklabel is valid. 2021 * 2022 * This is necessary since total size of the raid device 2023 * may vary when an interleave is changed even though exactly 2024 * same componets are used, and old disklabel may used 2025 * if that is found. 2026 */ 2027 if (lp->d_secperunit != rs->sc_size) 2028 printf("WARNING: %s: " 2029 "total sector size in disklabel (%d) != " 2030 "the size of raid (%ld)\n", rs->sc_xname, 2031 lp->d_secperunit, (long) rs->sc_size); 2032 for (i = 0; i < lp->d_npartitions; i++) { 2033 pp = &lp->d_partitions[i]; 2034 if (pp->p_offset + pp->p_size > rs->sc_size) 2035 printf("WARNING: %s: end of partition `%c' " 2036 "exceeds the size of raid (%ld)\n", 2037 rs->sc_xname, 'a' + i, (long) rs->sc_size); 2038 } 2039 } 2040 } 2041 /* 2042 * Take care of things one might want to take care of in the event 2043 * that a disklabel isn't present. 2044 */ 2045 static void 2046 raidmakedisklabel(rs) 2047 struct raid_softc *rs; 2048 { 2049 struct disklabel *lp = rs->sc_dkdev.dk_label; 2050 db1_printf(("Making a label..\n")); 2051 2052 /* 2053 * For historical reasons, if there's no disklabel present 2054 * the raw partition must be marked FS_BSDFFS. 2055 */ 2056 2057 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2058 2059 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2060 2061 lp->d_checksum = dkcksum(lp); 2062 } 2063 /* 2064 * Lookup the provided name in the filesystem. If the file exists, 2065 * is a valid block device, and isn't being used by anyone else, 2066 * set *vpp to the file's vnode. 2067 * You'll find the original of this in ccd.c 2068 */ 2069 int 2070 raidlookup(path, p, vpp) 2071 char *path; 2072 struct proc *p; 2073 struct vnode **vpp; /* result */ 2074 { 2075 struct nameidata nd; 2076 struct vnode *vp; 2077 struct vattr va; 2078 int error; 2079 2080 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p); 2081 if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) { 2082 #ifdef DEBUG 2083 printf("RAIDframe: vn_open returned %d\n", error); 2084 #endif 2085 return (error); 2086 } 2087 vp = nd.ni_vp; 2088 if (vp->v_usecount > 1) { 2089 VOP_UNLOCK(vp, 0); 2090 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2091 return (EBUSY); 2092 } 2093 if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) { 2094 VOP_UNLOCK(vp, 0); 2095 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2096 return (error); 2097 } 2098 /* XXX: eventually we should handle VREG, too. */ 2099 if (va.va_type != VBLK) { 2100 VOP_UNLOCK(vp, 0); 2101 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2102 return (ENOTBLK); 2103 } 2104 VOP_UNLOCK(vp, 0); 2105 *vpp = vp; 2106 return (0); 2107 } 2108 /* 2109 * Wait interruptibly for an exclusive lock. 2110 * 2111 * XXX 2112 * Several drivers do this; it should be abstracted and made MP-safe. 2113 * (Hmm... where have we seen this warning before :-> GO ) 2114 */ 2115 static int 2116 raidlock(rs) 2117 struct raid_softc *rs; 2118 { 2119 int error; 2120 2121 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2122 rs->sc_flags |= RAIDF_WANTED; 2123 if ((error = 2124 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2125 return (error); 2126 } 2127 rs->sc_flags |= RAIDF_LOCKED; 2128 return (0); 2129 } 2130 /* 2131 * Unlock and wake up any waiters. 2132 */ 2133 static void 2134 raidunlock(rs) 2135 struct raid_softc *rs; 2136 { 2137 2138 rs->sc_flags &= ~RAIDF_LOCKED; 2139 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2140 rs->sc_flags &= ~RAIDF_WANTED; 2141 wakeup(rs); 2142 } 2143 } 2144 2145 2146 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2147 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2148 2149 int 2150 raidmarkclean(struct vnode *b_vp, int mod_counter) 2151 { 2152 RF_ComponentLabel_t clabel; 2153 raidread_component_label(b_vp, &clabel); 2154 clabel.mod_counter = mod_counter; 2155 clabel.clean = RF_RAID_CLEAN; 2156 raidwrite_component_label(b_vp, &clabel); 2157 return(0); 2158 } 2159 2160 2161 int 2162 raidmarkdirty(struct vnode *b_vp, int mod_counter) 2163 { 2164 RF_ComponentLabel_t clabel; 2165 raidread_component_label(b_vp, &clabel); 2166 clabel.mod_counter = mod_counter; 2167 clabel.clean = RF_RAID_DIRTY; 2168 raidwrite_component_label(b_vp, &clabel); 2169 return(0); 2170 } 2171 2172 /* ARGSUSED */ 2173 int 2174 raidread_component_label(b_vp, clabel) 2175 struct vnode *b_vp; 2176 RF_ComponentLabel_t *clabel; 2177 { 2178 struct buf *bp; 2179 int error; 2180 2181 /* XXX should probably ensure that we don't try to do this if 2182 someone has changed rf_protected_sectors. */ 2183 2184 if (b_vp == NULL) { 2185 /* For whatever reason, this component is not valid. 2186 Don't try to read a component label from it. */ 2187 return(EINVAL); 2188 } 2189 2190 /* get a block of the appropriate size... */ 2191 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2192 bp->b_devvp = b_vp; 2193 2194 /* get our ducks in a row for the read */ 2195 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2196 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2197 bp->b_flags |= B_READ; 2198 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2199 2200 (*bdevsw[major(bp->b_devvp->v_rdev)].d_strategy)(bp); 2201 2202 error = biowait(bp); 2203 2204 if (!error) { 2205 memcpy(clabel, bp->b_data, 2206 sizeof(RF_ComponentLabel_t)); 2207 #if 0 2208 rf_print_component_label( clabel ); 2209 #endif 2210 } else { 2211 #if 0 2212 printf("Failed to read RAID component label!\n"); 2213 #endif 2214 } 2215 2216 brelse(bp); 2217 return(error); 2218 } 2219 2220 /* ARGSUSED */ 2221 int 2222 raidwrite_component_label(b_vp, clabel) 2223 struct vnode *b_vp; 2224 RF_ComponentLabel_t *clabel; 2225 { 2226 struct buf *bp; 2227 int error; 2228 2229 /* get a block of the appropriate size... */ 2230 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2231 bgetdevvp(b_vp, bp); 2232 2233 /* get our ducks in a row for the write */ 2234 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2235 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2236 bp->b_flags |= B_WRITE; 2237 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2238 2239 memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE ); 2240 2241 memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t)); 2242 2243 (*bdevsw[major(bp->b_devvp->v_rdev)].d_strategy)(bp); 2244 error = biowait(bp); 2245 bp->b_flags |= B_INVAL; 2246 brelse(bp); 2247 if (error) { 2248 #if 1 2249 printf("Failed to write RAID component info!\n"); 2250 #endif 2251 } 2252 2253 return(error); 2254 } 2255 2256 void 2257 rf_markalldirty(raidPtr) 2258 RF_Raid_t *raidPtr; 2259 { 2260 RF_ComponentLabel_t clabel; 2261 int r,c; 2262 2263 raidPtr->mod_counter++; 2264 for (r = 0; r < raidPtr->numRow; r++) { 2265 for (c = 0; c < raidPtr->numCol; c++) { 2266 /* we don't want to touch (at all) a disk that has 2267 failed */ 2268 if (!RF_DEAD_DISK(raidPtr->Disks[r][c].status)) { 2269 raidread_component_label( 2270 raidPtr->raid_cinfo[r][c].ci_vp, 2271 &clabel); 2272 if (clabel.status == rf_ds_spared) { 2273 /* XXX do something special... 2274 but whatever you do, don't 2275 try to access it!! */ 2276 } else { 2277 #if 0 2278 clabel.status = 2279 raidPtr->Disks[r][c].status; 2280 raidwrite_component_label( 2281 raidPtr->raid_cinfo[r][c].ci_vp, 2282 &clabel); 2283 #endif 2284 raidmarkdirty( 2285 raidPtr->raid_cinfo[r][c].ci_vp, 2286 raidPtr->mod_counter); 2287 } 2288 } 2289 } 2290 } 2291 /* printf("Component labels marked dirty.\n"); */ 2292 #if 0 2293 for( c = 0; c < raidPtr->numSpare ; c++) { 2294 sparecol = raidPtr->numCol + c; 2295 if (raidPtr->Disks[r][sparecol].status == rf_ds_used_spare) { 2296 /* 2297 2298 XXX this is where we get fancy and map this spare 2299 into it's correct spot in the array. 2300 2301 */ 2302 /* 2303 2304 we claim this disk is "optimal" if it's 2305 rf_ds_used_spare, as that means it should be 2306 directly substitutable for the disk it replaced. 2307 We note that too... 2308 2309 */ 2310 2311 for(i=0;i<raidPtr->numRow;i++) { 2312 for(j=0;j<raidPtr->numCol;j++) { 2313 if ((raidPtr->Disks[i][j].spareRow == 2314 r) && 2315 (raidPtr->Disks[i][j].spareCol == 2316 sparecol)) { 2317 srow = r; 2318 scol = sparecol; 2319 break; 2320 } 2321 } 2322 } 2323 2324 raidread_component_label( 2325 raidPtr->raid_cinfo[r][sparecol].ci_vp, 2326 &clabel); 2327 /* make sure status is noted */ 2328 clabel.version = RF_COMPONENT_LABEL_VERSION; 2329 clabel.mod_counter = raidPtr->mod_counter; 2330 clabel.serial_number = raidPtr->serial_number; 2331 clabel.row = srow; 2332 clabel.column = scol; 2333 clabel.num_rows = raidPtr->numRow; 2334 clabel.num_columns = raidPtr->numCol; 2335 clabel.clean = RF_RAID_DIRTY; /* changed in a bit*/ 2336 clabel.status = rf_ds_optimal; 2337 raidwrite_component_label( 2338 raidPtr->raid_cinfo[r][sparecol].ci_vp, 2339 &clabel); 2340 raidmarkclean(raidPtr->raid_cinfo[r][sparecol].ci_vp); 2341 } 2342 } 2343 #endif 2344 } 2345 2346 2347 void 2348 rf_update_component_labels(raidPtr, final) 2349 RF_Raid_t *raidPtr; 2350 int final; 2351 { 2352 RF_ComponentLabel_t clabel; 2353 int sparecol; 2354 int r,c; 2355 int i,j; 2356 int srow, scol; 2357 2358 srow = -1; 2359 scol = -1; 2360 2361 /* XXX should do extra checks to make sure things really are clean, 2362 rather than blindly setting the clean bit... */ 2363 2364 raidPtr->mod_counter++; 2365 2366 for (r = 0; r < raidPtr->numRow; r++) { 2367 for (c = 0; c < raidPtr->numCol; c++) { 2368 if (raidPtr->Disks[r][c].status == rf_ds_optimal) { 2369 raidread_component_label( 2370 raidPtr->raid_cinfo[r][c].ci_vp, 2371 &clabel); 2372 /* make sure status is noted */ 2373 clabel.status = rf_ds_optimal; 2374 /* bump the counter */ 2375 clabel.mod_counter = raidPtr->mod_counter; 2376 2377 raidwrite_component_label( 2378 raidPtr->raid_cinfo[r][c].ci_vp, 2379 &clabel); 2380 if (final == RF_FINAL_COMPONENT_UPDATE) { 2381 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2382 raidmarkclean( 2383 raidPtr->raid_cinfo[r][c].ci_vp, 2384 raidPtr->mod_counter); 2385 } 2386 } 2387 } 2388 /* else we don't touch it.. */ 2389 } 2390 } 2391 2392 for( c = 0; c < raidPtr->numSpare ; c++) { 2393 sparecol = raidPtr->numCol + c; 2394 if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) { 2395 /* 2396 2397 we claim this disk is "optimal" if it's 2398 rf_ds_used_spare, as that means it should be 2399 directly substitutable for the disk it replaced. 2400 We note that too... 2401 2402 */ 2403 2404 for(i=0;i<raidPtr->numRow;i++) { 2405 for(j=0;j<raidPtr->numCol;j++) { 2406 if ((raidPtr->Disks[i][j].spareRow == 2407 0) && 2408 (raidPtr->Disks[i][j].spareCol == 2409 sparecol)) { 2410 srow = i; 2411 scol = j; 2412 break; 2413 } 2414 } 2415 } 2416 2417 /* XXX shouldn't *really* need this... */ 2418 raidread_component_label( 2419 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2420 &clabel); 2421 /* make sure status is noted */ 2422 2423 raid_init_component_label(raidPtr, &clabel); 2424 2425 clabel.mod_counter = raidPtr->mod_counter; 2426 clabel.row = srow; 2427 clabel.column = scol; 2428 clabel.status = rf_ds_optimal; 2429 2430 raidwrite_component_label( 2431 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2432 &clabel); 2433 if (final == RF_FINAL_COMPONENT_UPDATE) { 2434 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2435 raidmarkclean( 2436 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2437 raidPtr->mod_counter); 2438 } 2439 } 2440 } 2441 } 2442 /* printf("Component labels updated\n"); */ 2443 } 2444 2445 void 2446 rf_close_component(raidPtr, vp, auto_configured) 2447 RF_Raid_t *raidPtr; 2448 struct vnode *vp; 2449 int auto_configured; 2450 { 2451 struct proc *p; 2452 2453 p = raidPtr->engine_thread; 2454 2455 if (vp != NULL) { 2456 if (auto_configured == 1) { 2457 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2458 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2459 vput(vp); 2460 2461 } else { 2462 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2463 } 2464 } else { 2465 printf("vnode was NULL\n"); 2466 } 2467 } 2468 2469 2470 void 2471 rf_UnconfigureVnodes(raidPtr) 2472 RF_Raid_t *raidPtr; 2473 { 2474 int r,c; 2475 struct proc *p; 2476 struct vnode *vp; 2477 int acd; 2478 2479 2480 /* We take this opportunity to close the vnodes like we should.. */ 2481 2482 p = raidPtr->engine_thread; 2483 2484 for (r = 0; r < raidPtr->numRow; r++) { 2485 for (c = 0; c < raidPtr->numCol; c++) { 2486 printf("Closing vnode for row: %d col: %d\n", r, c); 2487 vp = raidPtr->raid_cinfo[r][c].ci_vp; 2488 acd = raidPtr->Disks[r][c].auto_configured; 2489 rf_close_component(raidPtr, vp, acd); 2490 raidPtr->raid_cinfo[r][c].ci_vp = NULL; 2491 raidPtr->Disks[r][c].auto_configured = 0; 2492 } 2493 } 2494 for (r = 0; r < raidPtr->numSpare; r++) { 2495 printf("Closing vnode for spare: %d\n", r); 2496 vp = raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp; 2497 acd = raidPtr->Disks[0][raidPtr->numCol + r].auto_configured; 2498 rf_close_component(raidPtr, vp, acd); 2499 raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp = NULL; 2500 raidPtr->Disks[0][raidPtr->numCol + r].auto_configured = 0; 2501 } 2502 } 2503 2504 2505 void 2506 rf_ReconThread(req) 2507 struct rf_recon_req *req; 2508 { 2509 int s; 2510 RF_Raid_t *raidPtr; 2511 2512 s = splbio(); 2513 raidPtr = (RF_Raid_t *) req->raidPtr; 2514 raidPtr->recon_in_progress = 1; 2515 2516 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->row, req->col, 2517 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2518 2519 /* XXX get rid of this! we don't need it at all.. */ 2520 RF_Free(req, sizeof(*req)); 2521 2522 raidPtr->recon_in_progress = 0; 2523 splx(s); 2524 2525 /* That's all... */ 2526 kthread_exit(0); /* does not return */ 2527 } 2528 2529 void 2530 rf_RewriteParityThread(raidPtr) 2531 RF_Raid_t *raidPtr; 2532 { 2533 int retcode; 2534 int s; 2535 2536 raidPtr->parity_rewrite_in_progress = 1; 2537 s = splbio(); 2538 retcode = rf_RewriteParity(raidPtr); 2539 splx(s); 2540 if (retcode) { 2541 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2542 } else { 2543 /* set the clean bit! If we shutdown correctly, 2544 the clean bit on each component label will get 2545 set */ 2546 raidPtr->parity_good = RF_RAID_CLEAN; 2547 } 2548 raidPtr->parity_rewrite_in_progress = 0; 2549 2550 /* Anyone waiting for us to stop? If so, inform them... */ 2551 if (raidPtr->waitShutdown) { 2552 wakeup(&raidPtr->parity_rewrite_in_progress); 2553 } 2554 2555 /* That's all... */ 2556 kthread_exit(0); /* does not return */ 2557 } 2558 2559 2560 void 2561 rf_CopybackThread(raidPtr) 2562 RF_Raid_t *raidPtr; 2563 { 2564 int s; 2565 2566 raidPtr->copyback_in_progress = 1; 2567 s = splbio(); 2568 rf_CopybackReconstructedData(raidPtr); 2569 splx(s); 2570 raidPtr->copyback_in_progress = 0; 2571 2572 /* That's all... */ 2573 kthread_exit(0); /* does not return */ 2574 } 2575 2576 2577 void 2578 rf_ReconstructInPlaceThread(req) 2579 struct rf_recon_req *req; 2580 { 2581 int retcode; 2582 int s; 2583 RF_Raid_t *raidPtr; 2584 2585 s = splbio(); 2586 raidPtr = req->raidPtr; 2587 raidPtr->recon_in_progress = 1; 2588 retcode = rf_ReconstructInPlace(raidPtr, req->row, req->col); 2589 RF_Free(req, sizeof(*req)); 2590 raidPtr->recon_in_progress = 0; 2591 splx(s); 2592 2593 /* That's all... */ 2594 kthread_exit(0); /* does not return */ 2595 } 2596 2597 void 2598 rf_mountroot_hook(dev) 2599 struct device *dev; 2600 { 2601 2602 } 2603 2604 2605 RF_AutoConfig_t * 2606 rf_find_raid_components() 2607 { 2608 struct devnametobdevmaj *dtobdm; 2609 struct vnode *vp; 2610 struct disklabel label; 2611 struct device *dv; 2612 char *cd_name; 2613 dev_t dev; 2614 int error; 2615 int i; 2616 int good_one; 2617 RF_ComponentLabel_t *clabel; 2618 RF_AutoConfig_t *ac_list; 2619 RF_AutoConfig_t *ac; 2620 2621 2622 /* initialize the AutoConfig list */ 2623 ac_list = NULL; 2624 2625 /* we begin by trolling through *all* the devices on the system */ 2626 2627 for (dv = alldevs.tqh_first; dv != NULL; 2628 dv = dv->dv_list.tqe_next) { 2629 2630 /* we are only interested in disks... */ 2631 if (dv->dv_class != DV_DISK) 2632 continue; 2633 2634 /* we don't care about floppies... */ 2635 if (!strcmp(dv->dv_cfdata->cf_driver->cd_name,"fd")) { 2636 continue; 2637 } 2638 2639 /* need to find the device_name_to_block_device_major stuff */ 2640 cd_name = dv->dv_cfdata->cf_driver->cd_name; 2641 dtobdm = dev_name2blk; 2642 while (dtobdm->d_name && strcmp(dtobdm->d_name, cd_name)) { 2643 dtobdm++; 2644 } 2645 2646 /* get a vnode for the raw partition of this disk */ 2647 2648 dev = MAKEDISKDEV(dtobdm->d_maj, dv->dv_unit, RAW_PART); 2649 if (bdevvp(dev, &vp)) 2650 panic("RAID can't alloc vnode"); 2651 2652 error = VOP_OPEN(vp, FREAD, NOCRED, 0); 2653 2654 if (error) { 2655 /* "Who cares." Continue looking 2656 for something that exists*/ 2657 vput(vp); 2658 continue; 2659 } 2660 2661 /* Ok, the disk exists. Go get the disklabel. */ 2662 error = VOP_IOCTL(vp, DIOCGDINFO, (caddr_t)&label, 2663 FREAD, NOCRED, 0); 2664 if (error) { 2665 /* 2666 * XXX can't happen - open() would 2667 * have errored out (or faked up one) 2668 */ 2669 printf("can't get label for dev %s%c (%d)!?!?\n", 2670 dv->dv_xname, 'a' + RAW_PART, error); 2671 } 2672 2673 /* don't need this any more. We'll allocate it again 2674 a little later if we really do... */ 2675 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2676 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2677 vput(vp); 2678 2679 for (i=0; i < label.d_npartitions; i++) { 2680 /* We only support partitions marked as RAID */ 2681 if (label.d_partitions[i].p_fstype != FS_RAID) 2682 continue; 2683 2684 dev = MAKEDISKDEV(dtobdm->d_maj, dv->dv_unit, i); 2685 if (bdevvp(dev, &vp)) 2686 panic("RAID can't alloc vnode"); 2687 2688 error = VOP_OPEN(vp, FREAD, NOCRED, 0); 2689 if (error) { 2690 /* Whatever... */ 2691 vput(vp); 2692 continue; 2693 } 2694 2695 good_one = 0; 2696 2697 clabel = (RF_ComponentLabel_t *) 2698 malloc(sizeof(RF_ComponentLabel_t), 2699 M_RAIDFRAME, M_NOWAIT); 2700 if (clabel == NULL) { 2701 /* XXX CLEANUP HERE */ 2702 printf("RAID auto config: out of memory!\n"); 2703 return(NULL); /* XXX probably should panic? */ 2704 } 2705 2706 if (!raidread_component_label(vp, clabel)) { 2707 /* Got the label. Does it look reasonable? */ 2708 if (rf_reasonable_label(clabel) && 2709 (clabel->partitionSize <= 2710 label.d_partitions[i].p_size)) { 2711 #if DEBUG 2712 printf("Component on: %s%c: %d\n", 2713 dv->dv_xname, 'a'+i, 2714 label.d_partitions[i].p_size); 2715 rf_print_component_label(clabel); 2716 #endif 2717 /* if it's reasonable, add it, 2718 else ignore it. */ 2719 ac = (RF_AutoConfig_t *) 2720 malloc(sizeof(RF_AutoConfig_t), 2721 M_RAIDFRAME, 2722 M_NOWAIT); 2723 if (ac == NULL) { 2724 /* XXX should panic?? */ 2725 return(NULL); 2726 } 2727 2728 sprintf(ac->devname, "%s%c", 2729 dv->dv_xname, 'a'+i); 2730 ac->vp = vp; 2731 ac->clabel = clabel; 2732 ac->next = ac_list; 2733 ac_list = ac; 2734 good_one = 1; 2735 } 2736 } 2737 if (!good_one) { 2738 /* cleanup */ 2739 free(clabel, M_RAIDFRAME); 2740 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2741 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2742 vput(vp); 2743 } 2744 } 2745 } 2746 return(ac_list); 2747 } 2748 2749 static int 2750 rf_reasonable_label(clabel) 2751 RF_ComponentLabel_t *clabel; 2752 { 2753 2754 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 2755 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 2756 ((clabel->clean == RF_RAID_CLEAN) || 2757 (clabel->clean == RF_RAID_DIRTY)) && 2758 clabel->row >=0 && 2759 clabel->column >= 0 && 2760 clabel->num_rows > 0 && 2761 clabel->num_columns > 0 && 2762 clabel->row < clabel->num_rows && 2763 clabel->column < clabel->num_columns && 2764 clabel->blockSize > 0 && 2765 clabel->numBlocks > 0) { 2766 /* label looks reasonable enough... */ 2767 return(1); 2768 } 2769 return(0); 2770 } 2771 2772 2773 void 2774 rf_print_component_label(clabel) 2775 RF_ComponentLabel_t *clabel; 2776 { 2777 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 2778 clabel->row, clabel->column, 2779 clabel->num_rows, clabel->num_columns); 2780 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 2781 clabel->version, clabel->serial_number, 2782 clabel->mod_counter); 2783 printf(" Clean: %s Status: %d\n", 2784 clabel->clean ? "Yes" : "No", clabel->status ); 2785 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 2786 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 2787 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n", 2788 (char) clabel->parityConfig, clabel->blockSize, 2789 clabel->numBlocks); 2790 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" ); 2791 printf(" Contains root partition: %s\n", 2792 clabel->root_partition ? "Yes" : "No" ); 2793 printf(" Last configured as: raid%d\n", clabel->last_unit ); 2794 #if 0 2795 printf(" Config order: %d\n", clabel->config_order); 2796 #endif 2797 2798 } 2799 2800 RF_ConfigSet_t * 2801 rf_create_auto_sets(ac_list) 2802 RF_AutoConfig_t *ac_list; 2803 { 2804 RF_AutoConfig_t *ac; 2805 RF_ConfigSet_t *config_sets; 2806 RF_ConfigSet_t *cset; 2807 RF_AutoConfig_t *ac_next; 2808 2809 2810 config_sets = NULL; 2811 2812 /* Go through the AutoConfig list, and figure out which components 2813 belong to what sets. */ 2814 ac = ac_list; 2815 while(ac!=NULL) { 2816 /* we're going to putz with ac->next, so save it here 2817 for use at the end of the loop */ 2818 ac_next = ac->next; 2819 2820 if (config_sets == NULL) { 2821 /* will need at least this one... */ 2822 config_sets = (RF_ConfigSet_t *) 2823 malloc(sizeof(RF_ConfigSet_t), 2824 M_RAIDFRAME, M_NOWAIT); 2825 if (config_sets == NULL) { 2826 panic("rf_create_auto_sets: No memory!\n"); 2827 } 2828 /* this one is easy :) */ 2829 config_sets->ac = ac; 2830 config_sets->next = NULL; 2831 config_sets->rootable = 0; 2832 ac->next = NULL; 2833 } else { 2834 /* which set does this component fit into? */ 2835 cset = config_sets; 2836 while(cset!=NULL) { 2837 if (rf_does_it_fit(cset, ac)) { 2838 /* looks like it matches... */ 2839 ac->next = cset->ac; 2840 cset->ac = ac; 2841 break; 2842 } 2843 cset = cset->next; 2844 } 2845 if (cset==NULL) { 2846 /* didn't find a match above... new set..*/ 2847 cset = (RF_ConfigSet_t *) 2848 malloc(sizeof(RF_ConfigSet_t), 2849 M_RAIDFRAME, M_NOWAIT); 2850 if (cset == NULL) { 2851 panic("rf_create_auto_sets: No memory!\n"); 2852 } 2853 cset->ac = ac; 2854 ac->next = NULL; 2855 cset->next = config_sets; 2856 cset->rootable = 0; 2857 config_sets = cset; 2858 } 2859 } 2860 ac = ac_next; 2861 } 2862 2863 2864 return(config_sets); 2865 } 2866 2867 static int 2868 rf_does_it_fit(cset, ac) 2869 RF_ConfigSet_t *cset; 2870 RF_AutoConfig_t *ac; 2871 { 2872 RF_ComponentLabel_t *clabel1, *clabel2; 2873 2874 /* If this one matches the *first* one in the set, that's good 2875 enough, since the other members of the set would have been 2876 through here too... */ 2877 /* note that we are not checking partitionSize here.. 2878 2879 Note that we are also not checking the mod_counters here. 2880 If everything else matches execpt the mod_counter, that's 2881 good enough for this test. We will deal with the mod_counters 2882 a little later in the autoconfiguration process. 2883 2884 (clabel1->mod_counter == clabel2->mod_counter) && 2885 2886 The reason we don't check for this is that failed disks 2887 will have lower modification counts. If those disks are 2888 not added to the set they used to belong to, then they will 2889 form their own set, which may result in 2 different sets, 2890 for example, competing to be configured at raid0, and 2891 perhaps competing to be the root filesystem set. If the 2892 wrong ones get configured, or both attempt to become /, 2893 weird behaviour and or serious lossage will occur. Thus we 2894 need to bring them into the fold here, and kick them out at 2895 a later point. 2896 2897 */ 2898 2899 clabel1 = cset->ac->clabel; 2900 clabel2 = ac->clabel; 2901 if ((clabel1->version == clabel2->version) && 2902 (clabel1->serial_number == clabel2->serial_number) && 2903 (clabel1->num_rows == clabel2->num_rows) && 2904 (clabel1->num_columns == clabel2->num_columns) && 2905 (clabel1->sectPerSU == clabel2->sectPerSU) && 2906 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 2907 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 2908 (clabel1->parityConfig == clabel2->parityConfig) && 2909 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 2910 (clabel1->blockSize == clabel2->blockSize) && 2911 (clabel1->numBlocks == clabel2->numBlocks) && 2912 (clabel1->autoconfigure == clabel2->autoconfigure) && 2913 (clabel1->root_partition == clabel2->root_partition) && 2914 (clabel1->last_unit == clabel2->last_unit) && 2915 (clabel1->config_order == clabel2->config_order)) { 2916 /* if it get's here, it almost *has* to be a match */ 2917 } else { 2918 /* it's not consistent with somebody in the set.. 2919 punt */ 2920 return(0); 2921 } 2922 /* all was fine.. it must fit... */ 2923 return(1); 2924 } 2925 2926 int 2927 rf_have_enough_components(cset) 2928 RF_ConfigSet_t *cset; 2929 { 2930 RF_AutoConfig_t *ac; 2931 RF_AutoConfig_t *auto_config; 2932 RF_ComponentLabel_t *clabel; 2933 int r,c; 2934 int num_rows; 2935 int num_cols; 2936 int num_missing; 2937 int mod_counter; 2938 int mod_counter_found; 2939 int even_pair_failed; 2940 char parity_type; 2941 2942 2943 /* check to see that we have enough 'live' components 2944 of this set. If so, we can configure it if necessary */ 2945 2946 num_rows = cset->ac->clabel->num_rows; 2947 num_cols = cset->ac->clabel->num_columns; 2948 parity_type = cset->ac->clabel->parityConfig; 2949 2950 /* XXX Check for duplicate components!?!?!? */ 2951 2952 /* Determine what the mod_counter is supposed to be for this set. */ 2953 2954 mod_counter_found = 0; 2955 mod_counter = 0; 2956 ac = cset->ac; 2957 while(ac!=NULL) { 2958 if (mod_counter_found==0) { 2959 mod_counter = ac->clabel->mod_counter; 2960 mod_counter_found = 1; 2961 } else { 2962 if (ac->clabel->mod_counter > mod_counter) { 2963 mod_counter = ac->clabel->mod_counter; 2964 } 2965 } 2966 ac = ac->next; 2967 } 2968 2969 num_missing = 0; 2970 auto_config = cset->ac; 2971 2972 for(r=0; r<num_rows; r++) { 2973 even_pair_failed = 0; 2974 for(c=0; c<num_cols; c++) { 2975 ac = auto_config; 2976 while(ac!=NULL) { 2977 if ((ac->clabel->row == r) && 2978 (ac->clabel->column == c) && 2979 (ac->clabel->mod_counter == mod_counter)) { 2980 /* it's this one... */ 2981 #if DEBUG 2982 printf("Found: %s at %d,%d\n", 2983 ac->devname,r,c); 2984 #endif 2985 break; 2986 } 2987 ac=ac->next; 2988 } 2989 if (ac==NULL) { 2990 /* Didn't find one here! */ 2991 /* special case for RAID 1, especially 2992 where there are more than 2 2993 components (where RAIDframe treats 2994 things a little differently :( ) */ 2995 if (parity_type == '1') { 2996 if (c%2 == 0) { /* even component */ 2997 even_pair_failed = 1; 2998 } else { /* odd component. If 2999 we're failed, and 3000 so is the even 3001 component, it's 3002 "Good Night, Charlie" */ 3003 if (even_pair_failed == 1) { 3004 return(0); 3005 } 3006 } 3007 } else { 3008 /* normal accounting */ 3009 num_missing++; 3010 } 3011 } 3012 if ((parity_type == '1') && (c%2 == 1)) { 3013 /* Just did an even component, and we didn't 3014 bail.. reset the even_pair_failed flag, 3015 and go on to the next component.... */ 3016 even_pair_failed = 0; 3017 } 3018 } 3019 } 3020 3021 clabel = cset->ac->clabel; 3022 3023 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3024 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3025 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3026 /* XXX this needs to be made *much* more general */ 3027 /* Too many failures */ 3028 return(0); 3029 } 3030 /* otherwise, all is well, and we've got enough to take a kick 3031 at autoconfiguring this set */ 3032 return(1); 3033 } 3034 3035 void 3036 rf_create_configuration(ac,config,raidPtr) 3037 RF_AutoConfig_t *ac; 3038 RF_Config_t *config; 3039 RF_Raid_t *raidPtr; 3040 { 3041 RF_ComponentLabel_t *clabel; 3042 int i; 3043 3044 clabel = ac->clabel; 3045 3046 /* 1. Fill in the common stuff */ 3047 config->numRow = clabel->num_rows; 3048 config->numCol = clabel->num_columns; 3049 config->numSpare = 0; /* XXX should this be set here? */ 3050 config->sectPerSU = clabel->sectPerSU; 3051 config->SUsPerPU = clabel->SUsPerPU; 3052 config->SUsPerRU = clabel->SUsPerRU; 3053 config->parityConfig = clabel->parityConfig; 3054 /* XXX... */ 3055 strcpy(config->diskQueueType,"fifo"); 3056 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3057 config->layoutSpecificSize = 0; /* XXX ?? */ 3058 3059 while(ac!=NULL) { 3060 /* row/col values will be in range due to the checks 3061 in reasonable_label() */ 3062 strcpy(config->devnames[ac->clabel->row][ac->clabel->column], 3063 ac->devname); 3064 ac = ac->next; 3065 } 3066 3067 for(i=0;i<RF_MAXDBGV;i++) { 3068 config->debugVars[i][0] = NULL; 3069 } 3070 } 3071 3072 int 3073 rf_set_autoconfig(raidPtr, new_value) 3074 RF_Raid_t *raidPtr; 3075 int new_value; 3076 { 3077 RF_ComponentLabel_t clabel; 3078 struct vnode *vp; 3079 int row, column; 3080 3081 raidPtr->autoconfigure = new_value; 3082 for(row=0; row<raidPtr->numRow; row++) { 3083 for(column=0; column<raidPtr->numCol; column++) { 3084 if (raidPtr->Disks[row][column].status == 3085 rf_ds_optimal) { 3086 vp = raidPtr->raid_cinfo[row][column].ci_vp; 3087 raidread_component_label(vp, &clabel); 3088 clabel.autoconfigure = new_value; 3089 raidwrite_component_label(vp, &clabel); 3090 } 3091 } 3092 } 3093 return(new_value); 3094 } 3095 3096 int 3097 rf_set_rootpartition(raidPtr, new_value) 3098 RF_Raid_t *raidPtr; 3099 int new_value; 3100 { 3101 RF_ComponentLabel_t clabel; 3102 struct vnode *vp; 3103 int row, column; 3104 3105 raidPtr->root_partition = new_value; 3106 for(row=0; row<raidPtr->numRow; row++) { 3107 for(column=0; column<raidPtr->numCol; column++) { 3108 if (raidPtr->Disks[row][column].status == 3109 rf_ds_optimal) { 3110 vp = raidPtr->raid_cinfo[row][column].ci_vp; 3111 raidread_component_label(vp, &clabel); 3112 clabel.root_partition = new_value; 3113 raidwrite_component_label(vp, &clabel); 3114 } 3115 } 3116 } 3117 return(new_value); 3118 } 3119 3120 void 3121 rf_release_all_vps(cset) 3122 RF_ConfigSet_t *cset; 3123 { 3124 RF_AutoConfig_t *ac; 3125 3126 ac = cset->ac; 3127 while(ac!=NULL) { 3128 /* Close the vp, and give it back */ 3129 if (ac->vp) { 3130 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3131 VOP_CLOSE(ac->vp, FREAD, NOCRED, 0); 3132 vput(ac->vp); 3133 ac->vp = NULL; 3134 } 3135 ac = ac->next; 3136 } 3137 } 3138 3139 3140 void 3141 rf_cleanup_config_set(cset) 3142 RF_ConfigSet_t *cset; 3143 { 3144 RF_AutoConfig_t *ac; 3145 RF_AutoConfig_t *next_ac; 3146 3147 ac = cset->ac; 3148 while(ac!=NULL) { 3149 next_ac = ac->next; 3150 /* nuke the label */ 3151 free(ac->clabel, M_RAIDFRAME); 3152 /* cleanup the config structure */ 3153 free(ac, M_RAIDFRAME); 3154 /* "next.." */ 3155 ac = next_ac; 3156 } 3157 /* and, finally, nuke the config set */ 3158 free(cset, M_RAIDFRAME); 3159 } 3160 3161 3162 void 3163 raid_init_component_label(raidPtr, clabel) 3164 RF_Raid_t *raidPtr; 3165 RF_ComponentLabel_t *clabel; 3166 { 3167 /* current version number */ 3168 clabel->version = RF_COMPONENT_LABEL_VERSION; 3169 clabel->serial_number = raidPtr->serial_number; 3170 clabel->mod_counter = raidPtr->mod_counter; 3171 clabel->num_rows = raidPtr->numRow; 3172 clabel->num_columns = raidPtr->numCol; 3173 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3174 clabel->status = rf_ds_optimal; /* "It's good!" */ 3175 3176 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3177 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3178 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3179 3180 clabel->blockSize = raidPtr->bytesPerSector; 3181 clabel->numBlocks = raidPtr->sectorsPerDisk; 3182 3183 /* XXX not portable */ 3184 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3185 clabel->maxOutstanding = raidPtr->maxOutstanding; 3186 clabel->autoconfigure = raidPtr->autoconfigure; 3187 clabel->root_partition = raidPtr->root_partition; 3188 clabel->last_unit = raidPtr->raidid; 3189 clabel->config_order = raidPtr->config_order; 3190 } 3191 3192 int 3193 rf_auto_config_set(cset,unit) 3194 RF_ConfigSet_t *cset; 3195 int *unit; 3196 { 3197 RF_Raid_t *raidPtr; 3198 RF_Config_t *config; 3199 int raidID; 3200 int retcode; 3201 3202 printf("RAID autoconfigure\n"); 3203 3204 retcode = 0; 3205 *unit = -1; 3206 3207 /* 1. Create a config structure */ 3208 3209 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3210 M_RAIDFRAME, 3211 M_NOWAIT); 3212 if (config==NULL) { 3213 printf("Out of mem!?!?\n"); 3214 /* XXX do something more intelligent here. */ 3215 return(1); 3216 } 3217 3218 memset(config, 0, sizeof(RF_Config_t)); 3219 3220 /* XXX raidID needs to be set correctly.. */ 3221 3222 /* 3223 2. Figure out what RAID ID this one is supposed to live at 3224 See if we can get the same RAID dev that it was configured 3225 on last time.. 3226 */ 3227 3228 raidID = cset->ac->clabel->last_unit; 3229 if ((raidID < 0) || (raidID >= numraid)) { 3230 /* let's not wander off into lala land. */ 3231 raidID = numraid - 1; 3232 } 3233 if (raidPtrs[raidID]->valid != 0) { 3234 3235 /* 3236 Nope... Go looking for an alternative... 3237 Start high so we don't immediately use raid0 if that's 3238 not taken. 3239 */ 3240 3241 for(raidID = numraid; raidID >= 0; raidID--) { 3242 if (raidPtrs[raidID]->valid == 0) { 3243 /* can use this one! */ 3244 break; 3245 } 3246 } 3247 } 3248 3249 if (raidID < 0) { 3250 /* punt... */ 3251 printf("Unable to auto configure this set!\n"); 3252 printf("(Out of RAID devs!)\n"); 3253 return(1); 3254 } 3255 printf("Configuring raid%d:\n",raidID); 3256 raidPtr = raidPtrs[raidID]; 3257 3258 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3259 raidPtr->raidid = raidID; 3260 raidPtr->openings = RAIDOUTSTANDING; 3261 3262 /* 3. Build the configuration structure */ 3263 rf_create_configuration(cset->ac, config, raidPtr); 3264 3265 /* 4. Do the configuration */ 3266 retcode = rf_Configure(raidPtr, config, cset->ac); 3267 3268 if (retcode == 0) { 3269 3270 raidinit(raidPtrs[raidID]); 3271 3272 rf_markalldirty(raidPtrs[raidID]); 3273 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3274 if (cset->ac->clabel->root_partition==1) { 3275 /* everything configured just fine. Make a note 3276 that this set is eligible to be root. */ 3277 cset->rootable = 1; 3278 /* XXX do this here? */ 3279 raidPtrs[raidID]->root_partition = 1; 3280 } 3281 } 3282 3283 /* 5. Cleanup */ 3284 free(config, M_RAIDFRAME); 3285 3286 *unit = raidID; 3287 return(retcode); 3288 } 3289 3290 void 3291 rf_disk_unbusy(desc) 3292 RF_RaidAccessDesc_t *desc; 3293 { 3294 struct buf *bp; 3295 3296 bp = (struct buf *)desc->bp; 3297 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3298 (bp->b_bcount - bp->b_resid)); 3299 } 3300
Indexes created Tue Sep 30 20:09:53 GMT 2025