rf_netbsdkintf.c revision 1.86
1 /* $NetBSD: rf_netbsdkintf.c,v 1.86 2000/05/28 05:23:41 oster Exp $ */ 2 /*- 3 * Copyright (c) 1996, 1997, 1998 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Greg Oster; Jason R. Thorpe. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 38 /* 39 * Copyright (c) 1988 University of Utah. 40 * Copyright (c) 1990, 1993 41 * The Regents of the University of California. All rights reserved. 42 * 43 * This code is derived from software contributed to Berkeley by 44 * the Systems Programming Group of the University of Utah Computer 45 * Science Department. 46 * 47 * Redistribution and use in source and binary forms, with or without 48 * modification, are permitted provided that the following conditions 49 * are met: 50 * 1. Redistributions of source code must retain the above copyright 51 * notice, this list of conditions and the following disclaimer. 52 * 2. Redistributions in binary form must reproduce the above copyright 53 * notice, this list of conditions and the following disclaimer in the 54 * documentation and/or other materials provided with the distribution. 55 * 3. All advertising materials mentioning features or use of this software 56 * must display the following acknowledgement: 57 * This product includes software developed by the University of 58 * California, Berkeley and its contributors. 59 * 4. Neither the name of the University nor the names of its contributors 60 * may be used to endorse or promote products derived from this software 61 * without specific prior written permission. 62 * 63 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 64 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 65 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 66 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 67 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 68 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 69 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 70 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 71 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 72 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 73 * SUCH DAMAGE. 74 * 75 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 76 * 77 * @(#)cd.c 8.2 (Berkeley) 11/16/93 78 */ 79 80 81 82 83 /* 84 * Copyright (c) 1995 Carnegie-Mellon University. 85 * All rights reserved. 86 * 87 * Authors: Mark Holland, Jim Zelenka 88 * 89 * Permission to use, copy, modify and distribute this software and 90 * its documentation is hereby granted, provided that both the copyright 91 * notice and this permission notice appear in all copies of the 92 * software, derivative works or modified versions, and any portions 93 * thereof, and that both notices appear in supporting documentation. 94 * 95 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 96 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 97 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 98 * 99 * Carnegie Mellon requests users of this software to return to 100 * 101 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU 102 * School of Computer Science 103 * Carnegie Mellon University 104 * Pittsburgh PA 15213-3890 105 * 106 * any improvements or extensions that they make and grant Carnegie the 107 * rights to redistribute these changes. 108 */ 109 110 /*********************************************************** 111 * 112 * rf_kintf.c -- the kernel interface routines for RAIDframe 113 * 114 ***********************************************************/ 115 116 #include <sys/errno.h> 117 #include <sys/param.h> 118 #include <sys/pool.h> 119 #include <sys/queue.h> 120 #include <sys/disk.h> 121 #include <sys/device.h> 122 #include <sys/stat.h> 123 #include <sys/ioctl.h> 124 #include <sys/fcntl.h> 125 #include <sys/systm.h> 126 #include <sys/namei.h> 127 #include <sys/vnode.h> 128 #include <sys/param.h> 129 #include <sys/types.h> 130 #include <machine/types.h> 131 #include <sys/disklabel.h> 132 #include <sys/conf.h> 133 #include <sys/lock.h> 134 #include <sys/buf.h> 135 #include <sys/user.h> 136 #include <sys/reboot.h> 137 138 #include "raid.h" 139 #include "opt_raid_autoconfig.h" 140 #include "rf_raid.h" 141 #include "rf_raidframe.h" 142 #include "rf_copyback.h" 143 #include "rf_dag.h" 144 #include "rf_dagflags.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 dev_t dev, RF_SectorNum_t startSect, 179 RF_SectorCount_t numSect, caddr_t buf, 180 void (*cbFunc) (struct buf *), void *cbArg, 181 int logBytesPerSector, struct proc * b_proc); 182 static void raidinit __P((RF_Raid_t *)); 183 184 void raidattach __P((int)); 185 int raidsize __P((dev_t)); 186 int raidopen __P((dev_t, int, int, struct proc *)); 187 int raidclose __P((dev_t, int, int, struct proc *)); 188 int raidioctl __P((dev_t, u_long, caddr_t, int, struct proc *)); 189 int raidwrite __P((dev_t, struct uio *, int)); 190 int raidread __P((dev_t, struct uio *, int)); 191 void raidstrategy __P((struct buf *)); 192 int raiddump __P((dev_t, daddr_t, caddr_t, size_t)); 193 194 /* 195 * Pilfered from ccd.c 196 */ 197 198 struct raidbuf { 199 struct buf rf_buf; /* new I/O buf. MUST BE FIRST!!! */ 200 struct buf *rf_obp; /* ptr. to original I/O buf */ 201 int rf_flags; /* misc. flags */ 202 RF_DiskQueueData_t *req;/* the request that this was part of.. */ 203 }; 204 205 206 #define RAIDGETBUF(rs) pool_get(&(rs)->sc_cbufpool, PR_NOWAIT) 207 #define RAIDPUTBUF(rs, cbp) pool_put(&(rs)->sc_cbufpool, cbp) 208 209 /* XXX Not sure if the following should be replacing the raidPtrs above, 210 or if it should be used in conjunction with that... 211 */ 212 213 struct raid_softc { 214 int sc_flags; /* flags */ 215 int sc_cflags; /* configuration flags */ 216 size_t sc_size; /* size of the raid device */ 217 char sc_xname[20]; /* XXX external name */ 218 struct disk sc_dkdev; /* generic disk device info */ 219 struct pool sc_cbufpool; /* component buffer pool */ 220 struct buf_queue buf_queue; /* used for the device queue */ 221 }; 222 /* sc_flags */ 223 #define RAIDF_INITED 0x01 /* unit has been initialized */ 224 #define RAIDF_WLABEL 0x02 /* label area is writable */ 225 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 226 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 227 #define RAIDF_LOCKED 0x80 /* unit is locked */ 228 229 #define raidunit(x) DISKUNIT(x) 230 int numraid = 0; 231 232 /* 233 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 234 * Be aware that large numbers can allow the driver to consume a lot of 235 * kernel memory, especially on writes, and in degraded mode reads. 236 * 237 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 238 * a single 64K write will typically require 64K for the old data, 239 * 64K for the old parity, and 64K for the new parity, for a total 240 * of 192K (if the parity buffer is not re-used immediately). 241 * Even it if is used immedately, that's still 128K, which when multiplied 242 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 243 * 244 * Now in degraded mode, for example, a 64K read on the above setup may 245 * require data reconstruction, which will require *all* of the 4 remaining 246 * disks to participate -- 4 * 32K/disk == 128K again. 247 */ 248 249 #ifndef RAIDOUTSTANDING 250 #define RAIDOUTSTANDING 6 251 #endif 252 253 #define RAIDLABELDEV(dev) \ 254 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 255 256 /* declared here, and made public, for the benefit of KVM stuff.. */ 257 struct raid_softc *raid_softc; 258 259 static void raidgetdefaultlabel __P((RF_Raid_t *, struct raid_softc *, 260 struct disklabel *)); 261 static void raidgetdisklabel __P((dev_t)); 262 static void raidmakedisklabel __P((struct raid_softc *)); 263 264 static int raidlock __P((struct raid_softc *)); 265 static void raidunlock __P((struct raid_softc *)); 266 267 static void rf_markalldirty __P((RF_Raid_t *)); 268 void rf_mountroot_hook __P((struct device *)); 269 270 struct device *raidrootdev; 271 272 void rf_ReconThread __P((struct rf_recon_req *)); 273 /* XXX what I want is: */ 274 /*void rf_ReconThread __P((RF_Raid_t *raidPtr)); */ 275 void rf_RewriteParityThread __P((RF_Raid_t *raidPtr)); 276 void rf_CopybackThread __P((RF_Raid_t *raidPtr)); 277 void rf_ReconstructInPlaceThread __P((struct rf_recon_req *)); 278 void rf_buildroothack __P((void *)); 279 280 RF_AutoConfig_t *rf_find_raid_components __P((void)); 281 RF_ConfigSet_t *rf_create_auto_sets __P((RF_AutoConfig_t *)); 282 static int rf_does_it_fit __P((RF_ConfigSet_t *,RF_AutoConfig_t *)); 283 static int rf_reasonable_label __P((RF_ComponentLabel_t *)); 284 void rf_create_configuration __P((RF_AutoConfig_t *,RF_Config_t *, 285 RF_Raid_t *)); 286 int rf_set_autoconfig __P((RF_Raid_t *, int)); 287 int rf_set_rootpartition __P((RF_Raid_t *, int)); 288 void rf_release_all_vps __P((RF_ConfigSet_t *)); 289 void rf_cleanup_config_set __P((RF_ConfigSet_t *)); 290 int rf_have_enough_components __P((RF_ConfigSet_t *)); 291 int rf_auto_config_set __P((RF_ConfigSet_t *, int *)); 292 293 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not 294 allow autoconfig to take place. 295 Note that this is overridden by having 296 RAID_AUTOCONFIG as an option in the 297 kernel config file. */ 298 extern struct device *booted_device; 299 300 void 301 raidattach(num) 302 int num; 303 { 304 int raidID; 305 int i, rc; 306 RF_AutoConfig_t *ac_list; /* autoconfig list */ 307 RF_ConfigSet_t *config_sets; 308 309 #ifdef DEBUG 310 printf("raidattach: Asked for %d units\n", num); 311 #endif 312 313 if (num <= 0) { 314 #ifdef DIAGNOSTIC 315 panic("raidattach: count <= 0"); 316 #endif 317 return; 318 } 319 /* This is where all the initialization stuff gets done. */ 320 321 numraid = num; 322 323 /* Make some space for requested number of units... */ 324 325 RF_Calloc(raidPtrs, num, sizeof(RF_Raid_t *), (RF_Raid_t **)); 326 if (raidPtrs == NULL) { 327 panic("raidPtrs is NULL!!\n"); 328 } 329 330 rc = rf_mutex_init(&rf_sparet_wait_mutex); 331 if (rc) { 332 RF_PANIC(); 333 } 334 335 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 336 337 for (i = 0; i < num; i++) 338 raidPtrs[i] = NULL; 339 rc = rf_BootRaidframe(); 340 if (rc == 0) 341 printf("Kernelized RAIDframe activated\n"); 342 else 343 panic("Serious error booting RAID!!\n"); 344 345 /* put together some datastructures like the CCD device does.. This 346 * lets us lock the device and what-not when it gets opened. */ 347 348 raid_softc = (struct raid_softc *) 349 malloc(num * sizeof(struct raid_softc), 350 M_RAIDFRAME, M_NOWAIT); 351 if (raid_softc == NULL) { 352 printf("WARNING: no memory for RAIDframe driver\n"); 353 return; 354 } 355 356 bzero(raid_softc, num * sizeof(struct raid_softc)); 357 358 raidrootdev = (struct device *)malloc(num * sizeof(struct device), 359 M_RAIDFRAME, M_NOWAIT); 360 if (raidrootdev == NULL) { 361 panic("No memory for RAIDframe driver!!?!?!\n"); 362 } 363 364 for (raidID = 0; raidID < num; raidID++) { 365 BUFQ_INIT(&raid_softc[raidID].buf_queue); 366 367 raidrootdev[raidID].dv_class = DV_DISK; 368 raidrootdev[raidID].dv_cfdata = NULL; 369 raidrootdev[raidID].dv_unit = raidID; 370 raidrootdev[raidID].dv_parent = NULL; 371 raidrootdev[raidID].dv_flags = 0; 372 sprintf(raidrootdev[raidID].dv_xname,"raid%d",raidID); 373 374 RF_Calloc(raidPtrs[raidID], 1, sizeof(RF_Raid_t), 375 (RF_Raid_t *)); 376 if (raidPtrs[raidID] == NULL) { 377 printf("WARNING: raidPtrs[%d] is NULL\n", raidID); 378 numraid = raidID; 379 return; 380 } 381 } 382 383 #if RAID_AUTOCONFIG 384 raidautoconfig = 1; 385 #endif 386 387 if (raidautoconfig) { 388 /* 1. locate all RAID components on the system */ 389 390 #if DEBUG 391 printf("Searching for raid components...\n"); 392 #endif 393 ac_list = rf_find_raid_components(); 394 395 /* 2. sort them into their respective sets */ 396 397 config_sets = rf_create_auto_sets(ac_list); 398 399 /* 3. evaluate each set and configure the valid ones 400 This gets done in rf_buildroothack() */ 401 402 /* schedule the creation of the thread to do the 403 "/ on RAID" stuff */ 404 405 kthread_create(rf_buildroothack,config_sets); 406 407 #if 0 408 mountroothook_establish(rf_mountroot_hook, &raidrootdev[0]); 409 #endif 410 } 411 412 } 413 414 void 415 rf_buildroothack(arg) 416 void *arg; 417 { 418 RF_ConfigSet_t *config_sets = arg; 419 RF_ConfigSet_t *cset; 420 RF_ConfigSet_t *next_cset; 421 int retcode; 422 int raidID; 423 int rootID; 424 int num_root; 425 426 num_root = 0; 427 cset = config_sets; 428 while(cset != NULL ) { 429 next_cset = cset->next; 430 if (rf_have_enough_components(cset) && 431 cset->ac->clabel->autoconfigure==1) { 432 retcode = rf_auto_config_set(cset,&raidID); 433 if (!retcode) { 434 if (cset->rootable) { 435 rootID = raidID; 436 num_root++; 437 } 438 } else { 439 /* The autoconfig didn't work :( */ 440 #if DEBUG 441 printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 442 #endif 443 rf_release_all_vps(cset); 444 } 445 } else { 446 /* we're not autoconfiguring this set... 447 release the associated resources */ 448 rf_release_all_vps(cset); 449 } 450 /* cleanup */ 451 rf_cleanup_config_set(cset); 452 cset = next_cset; 453 } 454 if (boothowto & RB_ASKNAME) { 455 /* We don't auto-config... */ 456 } else { 457 /* They didn't ask, and we found something bootable... */ 458 459 if (num_root == 1) { 460 booted_device = &raidrootdev[rootID]; 461 } else if (num_root > 1) { 462 /* we can't guess.. require the user to answer... */ 463 boothowto |= RB_ASKNAME; 464 } 465 } 466 } 467 468 469 int 470 raidsize(dev) 471 dev_t dev; 472 { 473 struct raid_softc *rs; 474 struct disklabel *lp; 475 int part, unit, omask, size; 476 477 unit = raidunit(dev); 478 if (unit >= numraid) 479 return (-1); 480 rs = &raid_softc[unit]; 481 482 if ((rs->sc_flags & RAIDF_INITED) == 0) 483 return (-1); 484 485 part = DISKPART(dev); 486 omask = rs->sc_dkdev.dk_openmask & (1 << part); 487 lp = rs->sc_dkdev.dk_label; 488 489 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc)) 490 return (-1); 491 492 if (lp->d_partitions[part].p_fstype != FS_SWAP) 493 size = -1; 494 else 495 size = lp->d_partitions[part].p_size * 496 (lp->d_secsize / DEV_BSIZE); 497 498 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc)) 499 return (-1); 500 501 return (size); 502 503 } 504 505 int 506 raiddump(dev, blkno, va, size) 507 dev_t dev; 508 daddr_t blkno; 509 caddr_t va; 510 size_t size; 511 { 512 /* Not implemented. */ 513 return ENXIO; 514 } 515 /* ARGSUSED */ 516 int 517 raidopen(dev, flags, fmt, p) 518 dev_t dev; 519 int flags, fmt; 520 struct proc *p; 521 { 522 int unit = raidunit(dev); 523 struct raid_softc *rs; 524 struct disklabel *lp; 525 int part, pmask; 526 int error = 0; 527 528 if (unit >= numraid) 529 return (ENXIO); 530 rs = &raid_softc[unit]; 531 532 if ((error = raidlock(rs)) != 0) 533 return (error); 534 lp = rs->sc_dkdev.dk_label; 535 536 part = DISKPART(dev); 537 pmask = (1 << part); 538 539 db1_printf(("Opening raid device number: %d partition: %d\n", 540 unit, part)); 541 542 543 if ((rs->sc_flags & RAIDF_INITED) && 544 (rs->sc_dkdev.dk_openmask == 0)) 545 raidgetdisklabel(dev); 546 547 /* make sure that this partition exists */ 548 549 if (part != RAW_PART) { 550 db1_printf(("Not a raw partition..\n")); 551 if (((rs->sc_flags & RAIDF_INITED) == 0) || 552 ((part >= lp->d_npartitions) || 553 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 554 error = ENXIO; 555 raidunlock(rs); 556 db1_printf(("Bailing out...\n")); 557 return (error); 558 } 559 } 560 /* Prevent this unit from being unconfigured while open. */ 561 switch (fmt) { 562 case S_IFCHR: 563 rs->sc_dkdev.dk_copenmask |= pmask; 564 break; 565 566 case S_IFBLK: 567 rs->sc_dkdev.dk_bopenmask |= pmask; 568 break; 569 } 570 571 if ((rs->sc_dkdev.dk_openmask == 0) && 572 ((rs->sc_flags & RAIDF_INITED) != 0)) { 573 /* First one... mark things as dirty... Note that we *MUST* 574 have done a configure before this. I DO NOT WANT TO BE 575 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 576 THAT THEY BELONG TOGETHER!!!!! */ 577 /* XXX should check to see if we're only open for reading 578 here... If so, we needn't do this, but then need some 579 other way of keeping track of what's happened.. */ 580 581 rf_markalldirty( raidPtrs[unit] ); 582 } 583 584 585 rs->sc_dkdev.dk_openmask = 586 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 587 588 raidunlock(rs); 589 590 return (error); 591 592 593 } 594 /* ARGSUSED */ 595 int 596 raidclose(dev, flags, fmt, p) 597 dev_t dev; 598 int flags, fmt; 599 struct proc *p; 600 { 601 int unit = raidunit(dev); 602 struct raid_softc *rs; 603 int error = 0; 604 int part; 605 606 if (unit >= numraid) 607 return (ENXIO); 608 rs = &raid_softc[unit]; 609 610 if ((error = raidlock(rs)) != 0) 611 return (error); 612 613 part = DISKPART(dev); 614 615 /* ...that much closer to allowing unconfiguration... */ 616 switch (fmt) { 617 case S_IFCHR: 618 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 619 break; 620 621 case S_IFBLK: 622 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 623 break; 624 } 625 rs->sc_dkdev.dk_openmask = 626 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 627 628 if ((rs->sc_dkdev.dk_openmask == 0) && 629 ((rs->sc_flags & RAIDF_INITED) != 0)) { 630 /* Last one... device is not unconfigured yet. 631 Device shutdown has taken care of setting the 632 clean bits if RAIDF_INITED is not set 633 mark things as clean... */ 634 #if 0 635 printf("Last one on raid%d. Updating status.\n",unit); 636 #endif 637 rf_final_update_component_labels( raidPtrs[unit] ); 638 } 639 640 raidunlock(rs); 641 return (0); 642 643 } 644 645 void 646 raidstrategy(bp) 647 struct buf *bp; 648 { 649 int s; 650 651 unsigned int raidID = raidunit(bp->b_dev); 652 RF_Raid_t *raidPtr; 653 struct raid_softc *rs = &raid_softc[raidID]; 654 struct disklabel *lp; 655 int wlabel; 656 657 if ((rs->sc_flags & RAIDF_INITED) ==0) { 658 bp->b_error = ENXIO; 659 bp->b_flags = B_ERROR; 660 bp->b_resid = bp->b_bcount; 661 biodone(bp); 662 return; 663 } 664 if (raidID >= numraid || !raidPtrs[raidID]) { 665 bp->b_error = ENODEV; 666 bp->b_flags |= B_ERROR; 667 bp->b_resid = bp->b_bcount; 668 biodone(bp); 669 return; 670 } 671 raidPtr = raidPtrs[raidID]; 672 if (!raidPtr->valid) { 673 bp->b_error = ENODEV; 674 bp->b_flags |= B_ERROR; 675 bp->b_resid = bp->b_bcount; 676 biodone(bp); 677 return; 678 } 679 if (bp->b_bcount == 0) { 680 db1_printf(("b_bcount is zero..\n")); 681 biodone(bp); 682 return; 683 } 684 lp = rs->sc_dkdev.dk_label; 685 686 /* 687 * Do bounds checking and adjust transfer. If there's an 688 * error, the bounds check will flag that for us. 689 */ 690 691 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 692 if (DISKPART(bp->b_dev) != RAW_PART) 693 if (bounds_check_with_label(bp, lp, wlabel) <= 0) { 694 db1_printf(("Bounds check failed!!:%d %d\n", 695 (int) bp->b_blkno, (int) wlabel)); 696 biodone(bp); 697 return; 698 } 699 s = splbio(); 700 701 bp->b_resid = 0; 702 703 /* stuff it onto our queue */ 704 BUFQ_INSERT_TAIL(&rs->buf_queue, bp); 705 706 raidstart(raidPtrs[raidID]); 707 708 splx(s); 709 } 710 /* ARGSUSED */ 711 int 712 raidread(dev, uio, flags) 713 dev_t dev; 714 struct uio *uio; 715 int flags; 716 { 717 int unit = raidunit(dev); 718 struct raid_softc *rs; 719 int part; 720 721 if (unit >= numraid) 722 return (ENXIO); 723 rs = &raid_softc[unit]; 724 725 if ((rs->sc_flags & RAIDF_INITED) == 0) 726 return (ENXIO); 727 part = DISKPART(dev); 728 729 db1_printf(("raidread: unit: %d partition: %d\n", unit, part)); 730 731 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio)); 732 733 } 734 /* ARGSUSED */ 735 int 736 raidwrite(dev, uio, flags) 737 dev_t dev; 738 struct uio *uio; 739 int flags; 740 { 741 int unit = raidunit(dev); 742 struct raid_softc *rs; 743 744 if (unit >= numraid) 745 return (ENXIO); 746 rs = &raid_softc[unit]; 747 748 if ((rs->sc_flags & RAIDF_INITED) == 0) 749 return (ENXIO); 750 db1_printf(("raidwrite\n")); 751 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio)); 752 753 } 754 755 int 756 raidioctl(dev, cmd, data, flag, p) 757 dev_t dev; 758 u_long cmd; 759 caddr_t data; 760 int flag; 761 struct proc *p; 762 { 763 int unit = raidunit(dev); 764 int error = 0; 765 int part, pmask; 766 struct raid_softc *rs; 767 RF_Config_t *k_cfg, *u_cfg; 768 RF_Raid_t *raidPtr; 769 RF_RaidDisk_t *diskPtr; 770 RF_AccTotals_t *totals; 771 RF_DeviceConfig_t *d_cfg, **ucfgp; 772 u_char *specific_buf; 773 int retcode = 0; 774 int row; 775 int column; 776 struct rf_recon_req *rrcopy, *rr; 777 RF_ComponentLabel_t *clabel; 778 RF_ComponentLabel_t ci_label; 779 RF_ComponentLabel_t **clabel_ptr; 780 RF_SingleComponent_t *sparePtr,*componentPtr; 781 RF_SingleComponent_t hot_spare; 782 RF_SingleComponent_t component; 783 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 784 int i, j, d; 785 786 if (unit >= numraid) 787 return (ENXIO); 788 rs = &raid_softc[unit]; 789 raidPtr = raidPtrs[unit]; 790 791 db1_printf(("raidioctl: %d %d %d %d\n", (int) dev, 792 (int) DISKPART(dev), (int) unit, (int) cmd)); 793 794 /* Must be open for writes for these commands... */ 795 switch (cmd) { 796 case DIOCSDINFO: 797 case DIOCWDINFO: 798 case DIOCWLABEL: 799 if ((flag & FWRITE) == 0) 800 return (EBADF); 801 } 802 803 /* Must be initialized for these... */ 804 switch (cmd) { 805 case DIOCGDINFO: 806 case DIOCSDINFO: 807 case DIOCWDINFO: 808 case DIOCGPART: 809 case DIOCWLABEL: 810 case DIOCGDEFLABEL: 811 case RAIDFRAME_SHUTDOWN: 812 case RAIDFRAME_REWRITEPARITY: 813 case RAIDFRAME_GET_INFO: 814 case RAIDFRAME_RESET_ACCTOTALS: 815 case RAIDFRAME_GET_ACCTOTALS: 816 case RAIDFRAME_KEEP_ACCTOTALS: 817 case RAIDFRAME_GET_SIZE: 818 case RAIDFRAME_FAIL_DISK: 819 case RAIDFRAME_COPYBACK: 820 case RAIDFRAME_CHECK_RECON_STATUS: 821 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 822 case RAIDFRAME_GET_COMPONENT_LABEL: 823 case RAIDFRAME_SET_COMPONENT_LABEL: 824 case RAIDFRAME_ADD_HOT_SPARE: 825 case RAIDFRAME_REMOVE_HOT_SPARE: 826 case RAIDFRAME_INIT_LABELS: 827 case RAIDFRAME_REBUILD_IN_PLACE: 828 case RAIDFRAME_CHECK_PARITY: 829 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 830 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 831 case RAIDFRAME_CHECK_COPYBACK_STATUS: 832 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 833 case RAIDFRAME_SET_AUTOCONFIG: 834 case RAIDFRAME_SET_ROOT: 835 case RAIDFRAME_DELETE_COMPONENT: 836 case RAIDFRAME_INCORPORATE_HOT_SPARE: 837 if ((rs->sc_flags & RAIDF_INITED) == 0) 838 return (ENXIO); 839 } 840 841 switch (cmd) { 842 843 /* configure the system */ 844 case RAIDFRAME_CONFIGURE: 845 846 if (raidPtr->valid) { 847 /* There is a valid RAID set running on this unit! */ 848 printf("raid%d: Device already configured!\n",unit); 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 bzero((char *) raidPtr, sizeof(RF_Raid_t)); 908 raidPtr->raidid = unit; 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(dev); 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 bzero((char *) clabel, 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 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 990 return(EINVAL); 991 } 992 993 raidread_component_label(raidPtr->Disks[row][column].dev, 994 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->Disks[row][column].dev, 1038 raidPtr->raid_cinfo[row][column].ci_vp, 1039 clabel ); 1040 #endif 1041 return (0); 1042 1043 case RAIDFRAME_INIT_LABELS: 1044 clabel = (RF_ComponentLabel_t *) data; 1045 /* 1046 we only want the serial number from 1047 the above. We get all the rest of the information 1048 from the config that was used to create this RAID 1049 set. 1050 */ 1051 1052 raidPtr->serial_number = clabel->serial_number; 1053 1054 raid_init_component_label(raidPtr, &ci_label); 1055 ci_label.serial_number = clabel->serial_number; 1056 1057 for(row=0;row<raidPtr->numRow;row++) { 1058 ci_label.row = row; 1059 for(column=0;column<raidPtr->numCol;column++) { 1060 diskPtr = &raidPtr->Disks[row][column]; 1061 ci_label.partitionSize = diskPtr->partitionSize; 1062 ci_label.column = column; 1063 raidwrite_component_label( 1064 raidPtr->Disks[row][column].dev, 1065 raidPtr->raid_cinfo[row][column].ci_vp, 1066 &ci_label ); 1067 } 1068 } 1069 1070 return (retcode); 1071 case RAIDFRAME_SET_AUTOCONFIG: 1072 d = rf_set_autoconfig(raidPtr, *(int *) data); 1073 printf("New autoconfig value is: %d\n", d); 1074 *(int *) data = d; 1075 return (retcode); 1076 1077 case RAIDFRAME_SET_ROOT: 1078 d = rf_set_rootpartition(raidPtr, *(int *) data); 1079 printf("New rootpartition value is: %d\n", d); 1080 *(int *) data = d; 1081 return (retcode); 1082 1083 /* initialize all parity */ 1084 case RAIDFRAME_REWRITEPARITY: 1085 1086 if (raidPtr->Layout.map->faultsTolerated == 0) { 1087 /* Parity for RAID 0 is trivially correct */ 1088 raidPtr->parity_good = RF_RAID_CLEAN; 1089 return(0); 1090 } 1091 1092 if (raidPtr->parity_rewrite_in_progress == 1) { 1093 /* Re-write is already in progress! */ 1094 return(EINVAL); 1095 } 1096 1097 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1098 rf_RewriteParityThread, 1099 raidPtr,"raid_parity"); 1100 return (retcode); 1101 1102 1103 case RAIDFRAME_ADD_HOT_SPARE: 1104 sparePtr = (RF_SingleComponent_t *) data; 1105 memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t)); 1106 retcode = rf_add_hot_spare(raidPtr, &hot_spare); 1107 return(retcode); 1108 1109 case RAIDFRAME_REMOVE_HOT_SPARE: 1110 return(retcode); 1111 1112 case RAIDFRAME_DELETE_COMPONENT: 1113 componentPtr = (RF_SingleComponent_t *)data; 1114 memcpy( &component, componentPtr, 1115 sizeof(RF_SingleComponent_t)); 1116 retcode = rf_delete_component(raidPtr, &component); 1117 return(retcode); 1118 1119 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1120 componentPtr = (RF_SingleComponent_t *)data; 1121 memcpy( &component, componentPtr, 1122 sizeof(RF_SingleComponent_t)); 1123 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1124 return(retcode); 1125 1126 case RAIDFRAME_REBUILD_IN_PLACE: 1127 1128 if (raidPtr->Layout.map->faultsTolerated == 0) { 1129 /* Can't do this on a RAID 0!! */ 1130 return(EINVAL); 1131 } 1132 1133 if (raidPtr->recon_in_progress == 1) { 1134 /* a reconstruct is already in progress! */ 1135 return(EINVAL); 1136 } 1137 1138 componentPtr = (RF_SingleComponent_t *) data; 1139 memcpy( &component, componentPtr, 1140 sizeof(RF_SingleComponent_t)); 1141 row = component.row; 1142 column = component.column; 1143 printf("Rebuild: %d %d\n",row, column); 1144 if ((row < 0) || (row >= raidPtr->numRow) || 1145 (column < 0) || (column >= raidPtr->numCol)) { 1146 return(EINVAL); 1147 } 1148 1149 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1150 if (rrcopy == NULL) 1151 return(ENOMEM); 1152 1153 rrcopy->raidPtr = (void *) raidPtr; 1154 rrcopy->row = row; 1155 rrcopy->col = column; 1156 1157 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1158 rf_ReconstructInPlaceThread, 1159 rrcopy,"raid_reconip"); 1160 return(retcode); 1161 1162 case RAIDFRAME_GET_INFO: 1163 if (!raidPtr->valid) 1164 return (ENODEV); 1165 ucfgp = (RF_DeviceConfig_t **) data; 1166 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1167 (RF_DeviceConfig_t *)); 1168 if (d_cfg == NULL) 1169 return (ENOMEM); 1170 bzero((char *) d_cfg, sizeof(RF_DeviceConfig_t)); 1171 d_cfg->rows = raidPtr->numRow; 1172 d_cfg->cols = raidPtr->numCol; 1173 d_cfg->ndevs = raidPtr->numRow * raidPtr->numCol; 1174 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1175 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1176 return (ENOMEM); 1177 } 1178 d_cfg->nspares = raidPtr->numSpare; 1179 if (d_cfg->nspares >= RF_MAX_DISKS) { 1180 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1181 return (ENOMEM); 1182 } 1183 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1184 d = 0; 1185 for (i = 0; i < d_cfg->rows; i++) { 1186 for (j = 0; j < d_cfg->cols; j++) { 1187 d_cfg->devs[d] = raidPtr->Disks[i][j]; 1188 d++; 1189 } 1190 } 1191 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1192 d_cfg->spares[i] = raidPtr->Disks[0][j]; 1193 } 1194 retcode = copyout((caddr_t) d_cfg, (caddr_t) * ucfgp, 1195 sizeof(RF_DeviceConfig_t)); 1196 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1197 1198 return (retcode); 1199 1200 case RAIDFRAME_CHECK_PARITY: 1201 *(int *) data = raidPtr->parity_good; 1202 return (0); 1203 1204 case RAIDFRAME_RESET_ACCTOTALS: 1205 bzero(&raidPtr->acc_totals, sizeof(raidPtr->acc_totals)); 1206 return (0); 1207 1208 case RAIDFRAME_GET_ACCTOTALS: 1209 totals = (RF_AccTotals_t *) data; 1210 *totals = raidPtr->acc_totals; 1211 return (0); 1212 1213 case RAIDFRAME_KEEP_ACCTOTALS: 1214 raidPtr->keep_acc_totals = *(int *)data; 1215 return (0); 1216 1217 case RAIDFRAME_GET_SIZE: 1218 *(int *) data = raidPtr->totalSectors; 1219 return (0); 1220 1221 /* fail a disk & optionally start reconstruction */ 1222 case RAIDFRAME_FAIL_DISK: 1223 1224 if (raidPtr->Layout.map->faultsTolerated == 0) { 1225 /* Can't do this on a RAID 0!! */ 1226 return(EINVAL); 1227 } 1228 1229 rr = (struct rf_recon_req *) data; 1230 1231 if (rr->row < 0 || rr->row >= raidPtr->numRow 1232 || rr->col < 0 || rr->col >= raidPtr->numCol) 1233 return (EINVAL); 1234 1235 printf("raid%d: Failing the disk: row: %d col: %d\n", 1236 unit, rr->row, rr->col); 1237 1238 /* make a copy of the recon request so that we don't rely on 1239 * the user's buffer */ 1240 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1241 if (rrcopy == NULL) 1242 return(ENOMEM); 1243 bcopy(rr, rrcopy, sizeof(*rr)); 1244 rrcopy->raidPtr = (void *) raidPtr; 1245 1246 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1247 rf_ReconThread, 1248 rrcopy,"raid_recon"); 1249 return (0); 1250 1251 /* invoke a copyback operation after recon on whatever disk 1252 * needs it, if any */ 1253 case RAIDFRAME_COPYBACK: 1254 1255 if (raidPtr->Layout.map->faultsTolerated == 0) { 1256 /* This makes no sense on a RAID 0!! */ 1257 return(EINVAL); 1258 } 1259 1260 if (raidPtr->copyback_in_progress == 1) { 1261 /* Copyback is already in progress! */ 1262 return(EINVAL); 1263 } 1264 1265 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1266 rf_CopybackThread, 1267 raidPtr,"raid_copyback"); 1268 return (retcode); 1269 1270 /* return the percentage completion of reconstruction */ 1271 case RAIDFRAME_CHECK_RECON_STATUS: 1272 if (raidPtr->Layout.map->faultsTolerated == 0) { 1273 /* This makes no sense on a RAID 0, so tell the 1274 user it's done. */ 1275 *(int *) data = 100; 1276 return(0); 1277 } 1278 row = 0; /* XXX we only consider a single row... */ 1279 if (raidPtr->status[row] != rf_rs_reconstructing) 1280 *(int *) data = 100; 1281 else 1282 *(int *) data = raidPtr->reconControl[row]->percentComplete; 1283 return (0); 1284 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1285 progressInfoPtr = (RF_ProgressInfo_t **) data; 1286 row = 0; /* XXX we only consider a single row... */ 1287 if (raidPtr->status[row] != rf_rs_reconstructing) { 1288 progressInfo.remaining = 0; 1289 progressInfo.completed = 100; 1290 progressInfo.total = 100; 1291 } else { 1292 progressInfo.total = 1293 raidPtr->reconControl[row]->numRUsTotal; 1294 progressInfo.completed = 1295 raidPtr->reconControl[row]->numRUsComplete; 1296 progressInfo.remaining = progressInfo.total - 1297 progressInfo.completed; 1298 } 1299 retcode = copyout((caddr_t) &progressInfo, 1300 (caddr_t) *progressInfoPtr, 1301 sizeof(RF_ProgressInfo_t)); 1302 return (retcode); 1303 1304 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1305 if (raidPtr->Layout.map->faultsTolerated == 0) { 1306 /* This makes no sense on a RAID 0, so tell the 1307 user it's done. */ 1308 *(int *) data = 100; 1309 return(0); 1310 } 1311 if (raidPtr->parity_rewrite_in_progress == 1) { 1312 *(int *) data = 100 * 1313 raidPtr->parity_rewrite_stripes_done / 1314 raidPtr->Layout.numStripe; 1315 } else { 1316 *(int *) data = 100; 1317 } 1318 return (0); 1319 1320 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1321 progressInfoPtr = (RF_ProgressInfo_t **) data; 1322 if (raidPtr->parity_rewrite_in_progress == 1) { 1323 progressInfo.total = raidPtr->Layout.numStripe; 1324 progressInfo.completed = 1325 raidPtr->parity_rewrite_stripes_done; 1326 progressInfo.remaining = progressInfo.total - 1327 progressInfo.completed; 1328 } else { 1329 progressInfo.remaining = 0; 1330 progressInfo.completed = 100; 1331 progressInfo.total = 100; 1332 } 1333 retcode = copyout((caddr_t) &progressInfo, 1334 (caddr_t) *progressInfoPtr, 1335 sizeof(RF_ProgressInfo_t)); 1336 return (retcode); 1337 1338 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1339 if (raidPtr->Layout.map->faultsTolerated == 0) { 1340 /* This makes no sense on a RAID 0 */ 1341 *(int *) data = 100; 1342 return(0); 1343 } 1344 if (raidPtr->copyback_in_progress == 1) { 1345 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1346 raidPtr->Layout.numStripe; 1347 } else { 1348 *(int *) data = 100; 1349 } 1350 return (0); 1351 1352 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1353 if (raidPtr->copyback_in_progress == 1) { 1354 progressInfo.total = raidPtr->Layout.numStripe; 1355 progressInfo.completed = 1356 raidPtr->parity_rewrite_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 1440 case DIOCGPART: 1441 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1442 ((struct partinfo *) data)->part = 1443 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1444 break; 1445 1446 case DIOCWDINFO: 1447 case DIOCSDINFO: 1448 if ((error = raidlock(rs)) != 0) 1449 return (error); 1450 1451 rs->sc_flags |= RAIDF_LABELLING; 1452 1453 error = setdisklabel(rs->sc_dkdev.dk_label, 1454 (struct disklabel *) data, 0, rs->sc_dkdev.dk_cpulabel); 1455 if (error == 0) { 1456 if (cmd == DIOCWDINFO) 1457 error = writedisklabel(RAIDLABELDEV(dev), 1458 raidstrategy, rs->sc_dkdev.dk_label, 1459 rs->sc_dkdev.dk_cpulabel); 1460 } 1461 rs->sc_flags &= ~RAIDF_LABELLING; 1462 1463 raidunlock(rs); 1464 1465 if (error) 1466 return (error); 1467 break; 1468 1469 case DIOCWLABEL: 1470 if (*(int *) data != 0) 1471 rs->sc_flags |= RAIDF_WLABEL; 1472 else 1473 rs->sc_flags &= ~RAIDF_WLABEL; 1474 break; 1475 1476 case DIOCGDEFLABEL: 1477 raidgetdefaultlabel(raidPtr, rs, 1478 (struct disklabel *) data); 1479 break; 1480 1481 default: 1482 retcode = ENOTTY; 1483 } 1484 return (retcode); 1485 1486 } 1487 1488 1489 /* raidinit -- complete the rest of the initialization for the 1490 RAIDframe device. */ 1491 1492 1493 static void 1494 raidinit(raidPtr) 1495 RF_Raid_t *raidPtr; 1496 { 1497 struct raid_softc *rs; 1498 int unit; 1499 1500 unit = raidPtr->raidid; 1501 1502 rs = &raid_softc[unit]; 1503 pool_init(&rs->sc_cbufpool, sizeof(struct raidbuf), 0, 1504 0, 0, "raidpl", 0, NULL, NULL, M_RAIDFRAME); 1505 1506 1507 /* XXX should check return code first... */ 1508 rs->sc_flags |= RAIDF_INITED; 1509 1510 sprintf(rs->sc_xname, "raid%d", unit); /* XXX doesn't check bounds. */ 1511 1512 rs->sc_dkdev.dk_name = rs->sc_xname; 1513 1514 /* disk_attach actually creates space for the CPU disklabel, among 1515 * other things, so it's critical to call this *BEFORE* we try putzing 1516 * with disklabels. */ 1517 1518 disk_attach(&rs->sc_dkdev); 1519 1520 /* XXX There may be a weird interaction here between this, and 1521 * protectedSectors, as used in RAIDframe. */ 1522 1523 rs->sc_size = raidPtr->totalSectors; 1524 1525 } 1526 1527 /* wake up the daemon & tell it to get us a spare table 1528 * XXX 1529 * the entries in the queues should be tagged with the raidPtr 1530 * so that in the extremely rare case that two recons happen at once, 1531 * we know for which device were requesting a spare table 1532 * XXX 1533 * 1534 * XXX This code is not currently used. GO 1535 */ 1536 int 1537 rf_GetSpareTableFromDaemon(req) 1538 RF_SparetWait_t *req; 1539 { 1540 int retcode; 1541 1542 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1543 req->next = rf_sparet_wait_queue; 1544 rf_sparet_wait_queue = req; 1545 wakeup(&rf_sparet_wait_queue); 1546 1547 /* mpsleep unlocks the mutex */ 1548 while (!rf_sparet_resp_queue) { 1549 tsleep(&rf_sparet_resp_queue, PRIBIO, 1550 "raidframe getsparetable", 0); 1551 } 1552 req = rf_sparet_resp_queue; 1553 rf_sparet_resp_queue = req->next; 1554 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1555 1556 retcode = req->fcol; 1557 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1558 * alloc'd */ 1559 return (retcode); 1560 } 1561 1562 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1563 * bp & passes it down. 1564 * any calls originating in the kernel must use non-blocking I/O 1565 * do some extra sanity checking to return "appropriate" error values for 1566 * certain conditions (to make some standard utilities work) 1567 * 1568 * Formerly known as: rf_DoAccessKernel 1569 */ 1570 void 1571 raidstart(raidPtr) 1572 RF_Raid_t *raidPtr; 1573 { 1574 RF_SectorCount_t num_blocks, pb, sum; 1575 RF_RaidAddr_t raid_addr; 1576 int retcode; 1577 struct partition *pp; 1578 daddr_t blocknum; 1579 int unit; 1580 struct raid_softc *rs; 1581 int do_async; 1582 struct buf *bp; 1583 1584 unit = raidPtr->raidid; 1585 rs = &raid_softc[unit]; 1586 1587 /* quick check to see if anything has died recently */ 1588 RF_LOCK_MUTEX(raidPtr->mutex); 1589 if (raidPtr->numNewFailures > 0) { 1590 rf_update_component_labels(raidPtr); 1591 raidPtr->numNewFailures--; 1592 } 1593 RF_UNLOCK_MUTEX(raidPtr->mutex); 1594 1595 /* Check to see if we're at the limit... */ 1596 RF_LOCK_MUTEX(raidPtr->mutex); 1597 while (raidPtr->openings > 0) { 1598 RF_UNLOCK_MUTEX(raidPtr->mutex); 1599 1600 /* get the next item, if any, from the queue */ 1601 if ((bp = BUFQ_FIRST(&rs->buf_queue)) == NULL) { 1602 /* nothing more to do */ 1603 return; 1604 } 1605 BUFQ_REMOVE(&rs->buf_queue, bp); 1606 1607 /* Ok, for the bp we have here, bp->b_blkno is relative to the 1608 * partition.. Need to make it absolute to the underlying 1609 * device.. */ 1610 1611 blocknum = bp->b_blkno; 1612 if (DISKPART(bp->b_dev) != RAW_PART) { 1613 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 1614 blocknum += pp->p_offset; 1615 } 1616 1617 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 1618 (int) blocknum)); 1619 1620 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 1621 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 1622 1623 /* *THIS* is where we adjust what block we're going to... 1624 * but DO NOT TOUCH bp->b_blkno!!! */ 1625 raid_addr = blocknum; 1626 1627 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 1628 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 1629 sum = raid_addr + num_blocks + pb; 1630 if (1 || rf_debugKernelAccess) { 1631 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 1632 (int) raid_addr, (int) sum, (int) num_blocks, 1633 (int) pb, (int) bp->b_resid)); 1634 } 1635 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 1636 || (sum < num_blocks) || (sum < pb)) { 1637 bp->b_error = ENOSPC; 1638 bp->b_flags |= B_ERROR; 1639 bp->b_resid = bp->b_bcount; 1640 biodone(bp); 1641 RF_LOCK_MUTEX(raidPtr->mutex); 1642 continue; 1643 } 1644 /* 1645 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 1646 */ 1647 1648 if (bp->b_bcount & raidPtr->sectorMask) { 1649 bp->b_error = EINVAL; 1650 bp->b_flags |= B_ERROR; 1651 bp->b_resid = bp->b_bcount; 1652 biodone(bp); 1653 RF_LOCK_MUTEX(raidPtr->mutex); 1654 continue; 1655 1656 } 1657 db1_printf(("Calling DoAccess..\n")); 1658 1659 1660 RF_LOCK_MUTEX(raidPtr->mutex); 1661 raidPtr->openings--; 1662 RF_UNLOCK_MUTEX(raidPtr->mutex); 1663 1664 /* 1665 * Everything is async. 1666 */ 1667 do_async = 1; 1668 1669 /* don't ever condition on bp->b_flags & B_WRITE. 1670 * always condition on B_READ instead */ 1671 1672 /* XXX we're still at splbio() here... do we *really* 1673 need to be? */ 1674 1675 1676 retcode = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 1677 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 1678 do_async, raid_addr, num_blocks, 1679 bp->b_data, bp, NULL, NULL, 1680 RF_DAG_NONBLOCKING_IO, NULL, NULL, NULL); 1681 1682 1683 RF_LOCK_MUTEX(raidPtr->mutex); 1684 } 1685 RF_UNLOCK_MUTEX(raidPtr->mutex); 1686 } 1687 1688 1689 1690 1691 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 1692 1693 int 1694 rf_DispatchKernelIO(queue, req) 1695 RF_DiskQueue_t *queue; 1696 RF_DiskQueueData_t *req; 1697 { 1698 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 1699 struct buf *bp; 1700 struct raidbuf *raidbp = NULL; 1701 struct raid_softc *rs; 1702 int unit; 1703 int s; 1704 1705 s=0; 1706 /* s = splbio();*/ /* want to test this */ 1707 /* XXX along with the vnode, we also need the softc associated with 1708 * this device.. */ 1709 1710 req->queue = queue; 1711 1712 unit = queue->raidPtr->raidid; 1713 1714 db1_printf(("DispatchKernelIO unit: %d\n", unit)); 1715 1716 if (unit >= numraid) { 1717 printf("Invalid unit number: %d %d\n", unit, numraid); 1718 panic("Invalid Unit number in rf_DispatchKernelIO\n"); 1719 } 1720 rs = &raid_softc[unit]; 1721 1722 /* XXX is this the right place? */ 1723 disk_busy(&rs->sc_dkdev); 1724 1725 bp = req->bp; 1726 #if 1 1727 /* XXX when there is a physical disk failure, someone is passing us a 1728 * buffer that contains old stuff!! Attempt to deal with this problem 1729 * without taking a performance hit... (not sure where the real bug 1730 * is. It's buried in RAIDframe somewhere) :-( GO ) */ 1731 1732 if (bp->b_flags & B_ERROR) { 1733 bp->b_flags &= ~B_ERROR; 1734 } 1735 if (bp->b_error != 0) { 1736 bp->b_error = 0; 1737 } 1738 #endif 1739 raidbp = RAIDGETBUF(rs); 1740 1741 raidbp->rf_flags = 0; /* XXX not really used anywhere... */ 1742 1743 /* 1744 * context for raidiodone 1745 */ 1746 raidbp->rf_obp = bp; 1747 raidbp->req = req; 1748 1749 LIST_INIT(&raidbp->rf_buf.b_dep); 1750 1751 switch (req->type) { 1752 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 1753 /* XXX need to do something extra here.. */ 1754 /* I'm leaving this in, as I've never actually seen it used, 1755 * and I'd like folks to report it... GO */ 1756 printf(("WAKEUP CALLED\n")); 1757 queue->numOutstanding++; 1758 1759 /* XXX need to glue the original buffer into this?? */ 1760 1761 KernelWakeupFunc(&raidbp->rf_buf); 1762 break; 1763 1764 case RF_IO_TYPE_READ: 1765 case RF_IO_TYPE_WRITE: 1766 1767 if (req->tracerec) { 1768 RF_ETIMER_START(req->tracerec->timer); 1769 } 1770 InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp, 1771 op | bp->b_flags, queue->rf_cinfo->ci_dev, 1772 req->sectorOffset, req->numSector, 1773 req->buf, KernelWakeupFunc, (void *) req, 1774 queue->raidPtr->logBytesPerSector, req->b_proc); 1775 1776 if (rf_debugKernelAccess) { 1777 db1_printf(("dispatch: bp->b_blkno = %ld\n", 1778 (long) bp->b_blkno)); 1779 } 1780 queue->numOutstanding++; 1781 queue->last_deq_sector = req->sectorOffset; 1782 /* acc wouldn't have been let in if there were any pending 1783 * reqs at any other priority */ 1784 queue->curPriority = req->priority; 1785 1786 db1_printf(("Going for %c to unit %d row %d col %d\n", 1787 req->type, unit, queue->row, queue->col)); 1788 db1_printf(("sector %d count %d (%d bytes) %d\n", 1789 (int) req->sectorOffset, (int) req->numSector, 1790 (int) (req->numSector << 1791 queue->raidPtr->logBytesPerSector), 1792 (int) queue->raidPtr->logBytesPerSector)); 1793 if ((raidbp->rf_buf.b_flags & B_READ) == 0) { 1794 raidbp->rf_buf.b_vp->v_numoutput++; 1795 } 1796 VOP_STRATEGY(&raidbp->rf_buf); 1797 1798 break; 1799 1800 default: 1801 panic("bad req->type in rf_DispatchKernelIO"); 1802 } 1803 db1_printf(("Exiting from DispatchKernelIO\n")); 1804 /* splx(s); */ /* want to test this */ 1805 return (0); 1806 } 1807 /* this is the callback function associated with a I/O invoked from 1808 kernel code. 1809 */ 1810 static void 1811 KernelWakeupFunc(vbp) 1812 struct buf *vbp; 1813 { 1814 RF_DiskQueueData_t *req = NULL; 1815 RF_DiskQueue_t *queue; 1816 struct raidbuf *raidbp = (struct raidbuf *) vbp; 1817 struct buf *bp; 1818 struct raid_softc *rs; 1819 int unit; 1820 int s; 1821 1822 s = splbio(); 1823 db1_printf(("recovering the request queue:\n")); 1824 req = raidbp->req; 1825 1826 bp = raidbp->rf_obp; 1827 1828 queue = (RF_DiskQueue_t *) req->queue; 1829 1830 if (raidbp->rf_buf.b_flags & B_ERROR) { 1831 bp->b_flags |= B_ERROR; 1832 bp->b_error = raidbp->rf_buf.b_error ? 1833 raidbp->rf_buf.b_error : EIO; 1834 } 1835 1836 /* XXX methinks this could be wrong... */ 1837 #if 1 1838 bp->b_resid = raidbp->rf_buf.b_resid; 1839 #endif 1840 1841 if (req->tracerec) { 1842 RF_ETIMER_STOP(req->tracerec->timer); 1843 RF_ETIMER_EVAL(req->tracerec->timer); 1844 RF_LOCK_MUTEX(rf_tracing_mutex); 1845 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 1846 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 1847 req->tracerec->num_phys_ios++; 1848 RF_UNLOCK_MUTEX(rf_tracing_mutex); 1849 } 1850 bp->b_bcount = raidbp->rf_buf.b_bcount; /* XXXX ?? */ 1851 1852 unit = queue->raidPtr->raidid; /* *Much* simpler :-> */ 1853 1854 1855 /* XXX Ok, let's get aggressive... If B_ERROR is set, let's go 1856 * ballistic, and mark the component as hosed... */ 1857 1858 if (bp->b_flags & B_ERROR) { 1859 /* Mark the disk as dead */ 1860 /* but only mark it once... */ 1861 if (queue->raidPtr->Disks[queue->row][queue->col].status == 1862 rf_ds_optimal) { 1863 printf("raid%d: IO Error. Marking %s as failed.\n", 1864 unit, queue->raidPtr->Disks[queue->row][queue->col].devname); 1865 queue->raidPtr->Disks[queue->row][queue->col].status = 1866 rf_ds_failed; 1867 queue->raidPtr->status[queue->row] = rf_rs_degraded; 1868 queue->raidPtr->numFailures++; 1869 queue->raidPtr->numNewFailures++; 1870 /* XXX here we should bump the version number for each component, and write that data out */ 1871 } else { /* Disk is already dead... */ 1872 /* printf("Disk already marked as dead!\n"); */ 1873 } 1874 1875 } 1876 1877 rs = &raid_softc[unit]; 1878 RAIDPUTBUF(rs, raidbp); 1879 1880 1881 if (bp->b_resid == 0) { 1882 /* XXX is this the right place for a disk_unbusy()??!??!?!? */ 1883 disk_unbusy(&rs->sc_dkdev, (bp->b_bcount - bp->b_resid)); 1884 } 1885 1886 rf_DiskIOComplete(queue, req, (bp->b_flags & B_ERROR) ? 1 : 0); 1887 (req->CompleteFunc) (req->argument, (bp->b_flags & B_ERROR) ? 1 : 0); 1888 1889 splx(s); 1890 } 1891 1892 1893 1894 /* 1895 * initialize a buf structure for doing an I/O in the kernel. 1896 */ 1897 static void 1898 InitBP(bp, b_vp, rw_flag, dev, startSect, numSect, buf, cbFunc, cbArg, 1899 logBytesPerSector, b_proc) 1900 struct buf *bp; 1901 struct vnode *b_vp; 1902 unsigned rw_flag; 1903 dev_t dev; 1904 RF_SectorNum_t startSect; 1905 RF_SectorCount_t numSect; 1906 caddr_t buf; 1907 void (*cbFunc) (struct buf *); 1908 void *cbArg; 1909 int logBytesPerSector; 1910 struct proc *b_proc; 1911 { 1912 /* bp->b_flags = B_PHYS | rw_flag; */ 1913 bp->b_flags = B_CALL | rw_flag; /* XXX need B_PHYS here too??? */ 1914 bp->b_bcount = numSect << logBytesPerSector; 1915 bp->b_bufsize = bp->b_bcount; 1916 bp->b_error = 0; 1917 bp->b_dev = dev; 1918 bp->b_data = buf; 1919 bp->b_blkno = startSect; 1920 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 1921 if (bp->b_bcount == 0) { 1922 panic("bp->b_bcount is zero in InitBP!!\n"); 1923 } 1924 bp->b_proc = b_proc; 1925 bp->b_iodone = cbFunc; 1926 bp->b_vp = b_vp; 1927 1928 } 1929 1930 static void 1931 raidgetdefaultlabel(raidPtr, rs, lp) 1932 RF_Raid_t *raidPtr; 1933 struct raid_softc *rs; 1934 struct disklabel *lp; 1935 { 1936 db1_printf(("Building a default label...\n")); 1937 bzero(lp, sizeof(*lp)); 1938 1939 /* fabricate a label... */ 1940 lp->d_secperunit = raidPtr->totalSectors; 1941 lp->d_secsize = raidPtr->bytesPerSector; 1942 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 1943 lp->d_ntracks = 1; 1944 lp->d_ncylinders = raidPtr->totalSectors / 1945 (lp->d_nsectors * lp->d_ntracks); 1946 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1947 1948 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 1949 lp->d_type = DTYPE_RAID; 1950 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1951 lp->d_rpm = 3600; 1952 lp->d_interleave = 1; 1953 lp->d_flags = 0; 1954 1955 lp->d_partitions[RAW_PART].p_offset = 0; 1956 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 1957 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 1958 lp->d_npartitions = RAW_PART + 1; 1959 1960 lp->d_magic = DISKMAGIC; 1961 lp->d_magic2 = DISKMAGIC; 1962 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 1963 1964 } 1965 /* 1966 * Read the disklabel from the raid device. If one is not present, fake one 1967 * up. 1968 */ 1969 static void 1970 raidgetdisklabel(dev) 1971 dev_t dev; 1972 { 1973 int unit = raidunit(dev); 1974 struct raid_softc *rs = &raid_softc[unit]; 1975 char *errstring; 1976 struct disklabel *lp = rs->sc_dkdev.dk_label; 1977 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 1978 RF_Raid_t *raidPtr; 1979 1980 db1_printf(("Getting the disklabel...\n")); 1981 1982 bzero(clp, sizeof(*clp)); 1983 1984 raidPtr = raidPtrs[unit]; 1985 1986 raidgetdefaultlabel(raidPtr, rs, lp); 1987 1988 /* 1989 * Call the generic disklabel extraction routine. 1990 */ 1991 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 1992 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 1993 if (errstring) 1994 raidmakedisklabel(rs); 1995 else { 1996 int i; 1997 struct partition *pp; 1998 1999 /* 2000 * Sanity check whether the found disklabel is valid. 2001 * 2002 * This is necessary since total size of the raid device 2003 * may vary when an interleave is changed even though exactly 2004 * same componets are used, and old disklabel may used 2005 * if that is found. 2006 */ 2007 if (lp->d_secperunit != rs->sc_size) 2008 printf("WARNING: %s: " 2009 "total sector size in disklabel (%d) != " 2010 "the size of raid (%ld)\n", rs->sc_xname, 2011 lp->d_secperunit, (long) rs->sc_size); 2012 for (i = 0; i < lp->d_npartitions; i++) { 2013 pp = &lp->d_partitions[i]; 2014 if (pp->p_offset + pp->p_size > rs->sc_size) 2015 printf("WARNING: %s: end of partition `%c' " 2016 "exceeds the size of raid (%ld)\n", 2017 rs->sc_xname, 'a' + i, (long) rs->sc_size); 2018 } 2019 } 2020 2021 } 2022 /* 2023 * Take care of things one might want to take care of in the event 2024 * that a disklabel isn't present. 2025 */ 2026 static void 2027 raidmakedisklabel(rs) 2028 struct raid_softc *rs; 2029 { 2030 struct disklabel *lp = rs->sc_dkdev.dk_label; 2031 db1_printf(("Making a label..\n")); 2032 2033 /* 2034 * For historical reasons, if there's no disklabel present 2035 * the raw partition must be marked FS_BSDFFS. 2036 */ 2037 2038 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2039 2040 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2041 2042 lp->d_checksum = dkcksum(lp); 2043 } 2044 /* 2045 * Lookup the provided name in the filesystem. If the file exists, 2046 * is a valid block device, and isn't being used by anyone else, 2047 * set *vpp to the file's vnode. 2048 * You'll find the original of this in ccd.c 2049 */ 2050 int 2051 raidlookup(path, p, vpp) 2052 char *path; 2053 struct proc *p; 2054 struct vnode **vpp; /* result */ 2055 { 2056 struct nameidata nd; 2057 struct vnode *vp; 2058 struct vattr va; 2059 int error; 2060 2061 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p); 2062 if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) { 2063 #ifdef DEBUG 2064 printf("RAIDframe: vn_open returned %d\n", error); 2065 #endif 2066 return (error); 2067 } 2068 vp = nd.ni_vp; 2069 if (vp->v_usecount > 1) { 2070 VOP_UNLOCK(vp, 0); 2071 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2072 return (EBUSY); 2073 } 2074 if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) { 2075 VOP_UNLOCK(vp, 0); 2076 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2077 return (error); 2078 } 2079 /* XXX: eventually we should handle VREG, too. */ 2080 if (va.va_type != VBLK) { 2081 VOP_UNLOCK(vp, 0); 2082 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2083 return (ENOTBLK); 2084 } 2085 VOP_UNLOCK(vp, 0); 2086 *vpp = vp; 2087 return (0); 2088 } 2089 /* 2090 * Wait interruptibly for an exclusive lock. 2091 * 2092 * XXX 2093 * Several drivers do this; it should be abstracted and made MP-safe. 2094 * (Hmm... where have we seen this warning before :-> GO ) 2095 */ 2096 static int 2097 raidlock(rs) 2098 struct raid_softc *rs; 2099 { 2100 int error; 2101 2102 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2103 rs->sc_flags |= RAIDF_WANTED; 2104 if ((error = 2105 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2106 return (error); 2107 } 2108 rs->sc_flags |= RAIDF_LOCKED; 2109 return (0); 2110 } 2111 /* 2112 * Unlock and wake up any waiters. 2113 */ 2114 static void 2115 raidunlock(rs) 2116 struct raid_softc *rs; 2117 { 2118 2119 rs->sc_flags &= ~RAIDF_LOCKED; 2120 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2121 rs->sc_flags &= ~RAIDF_WANTED; 2122 wakeup(rs); 2123 } 2124 } 2125 2126 2127 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2128 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2129 2130 int 2131 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter) 2132 { 2133 RF_ComponentLabel_t clabel; 2134 raidread_component_label(dev, b_vp, &clabel); 2135 clabel.mod_counter = mod_counter; 2136 clabel.clean = RF_RAID_CLEAN; 2137 raidwrite_component_label(dev, b_vp, &clabel); 2138 return(0); 2139 } 2140 2141 2142 int 2143 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter) 2144 { 2145 RF_ComponentLabel_t clabel; 2146 raidread_component_label(dev, b_vp, &clabel); 2147 clabel.mod_counter = mod_counter; 2148 clabel.clean = RF_RAID_DIRTY; 2149 raidwrite_component_label(dev, b_vp, &clabel); 2150 return(0); 2151 } 2152 2153 /* ARGSUSED */ 2154 int 2155 raidread_component_label(dev, b_vp, clabel) 2156 dev_t dev; 2157 struct vnode *b_vp; 2158 RF_ComponentLabel_t *clabel; 2159 { 2160 struct buf *bp; 2161 int error; 2162 2163 /* XXX should probably ensure that we don't try to do this if 2164 someone has changed rf_protected_sectors. */ 2165 2166 /* get a block of the appropriate size... */ 2167 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2168 bp->b_dev = dev; 2169 2170 /* get our ducks in a row for the read */ 2171 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2172 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2173 bp->b_flags = B_BUSY | B_READ; 2174 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2175 2176 (*bdevsw[major(bp->b_dev)].d_strategy)(bp); 2177 2178 error = biowait(bp); 2179 2180 if (!error) { 2181 memcpy(clabel, bp->b_data, 2182 sizeof(RF_ComponentLabel_t)); 2183 #if 0 2184 rf_print_component_label( clabel ); 2185 #endif 2186 } else { 2187 #if 0 2188 printf("Failed to read RAID component label!\n"); 2189 #endif 2190 } 2191 2192 bp->b_flags = B_INVAL | B_AGE; 2193 brelse(bp); 2194 return(error); 2195 } 2196 /* ARGSUSED */ 2197 int 2198 raidwrite_component_label(dev, b_vp, clabel) 2199 dev_t dev; 2200 struct vnode *b_vp; 2201 RF_ComponentLabel_t *clabel; 2202 { 2203 struct buf *bp; 2204 int error; 2205 2206 /* get a block of the appropriate size... */ 2207 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2208 bp->b_dev = dev; 2209 2210 /* get our ducks in a row for the write */ 2211 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2212 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2213 bp->b_flags = B_BUSY | B_WRITE; 2214 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2215 2216 memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE ); 2217 2218 memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t)); 2219 2220 (*bdevsw[major(bp->b_dev)].d_strategy)(bp); 2221 error = biowait(bp); 2222 bp->b_flags = B_INVAL | B_AGE; 2223 brelse(bp); 2224 if (error) { 2225 #if 1 2226 printf("Failed to write RAID component info!\n"); 2227 #endif 2228 } 2229 2230 return(error); 2231 } 2232 2233 void 2234 rf_markalldirty(raidPtr) 2235 RF_Raid_t *raidPtr; 2236 { 2237 RF_ComponentLabel_t clabel; 2238 int r,c; 2239 2240 raidPtr->mod_counter++; 2241 for (r = 0; r < raidPtr->numRow; r++) { 2242 for (c = 0; c < raidPtr->numCol; c++) { 2243 if (raidPtr->Disks[r][c].status != rf_ds_failed) { 2244 raidread_component_label( 2245 raidPtr->Disks[r][c].dev, 2246 raidPtr->raid_cinfo[r][c].ci_vp, 2247 &clabel); 2248 if (clabel.status == rf_ds_spared) { 2249 /* XXX do something special... 2250 but whatever you do, don't 2251 try to access it!! */ 2252 } else { 2253 #if 0 2254 clabel.status = 2255 raidPtr->Disks[r][c].status; 2256 raidwrite_component_label( 2257 raidPtr->Disks[r][c].dev, 2258 raidPtr->raid_cinfo[r][c].ci_vp, 2259 &clabel); 2260 #endif 2261 raidmarkdirty( 2262 raidPtr->Disks[r][c].dev, 2263 raidPtr->raid_cinfo[r][c].ci_vp, 2264 raidPtr->mod_counter); 2265 } 2266 } 2267 } 2268 } 2269 /* printf("Component labels marked dirty.\n"); */ 2270 #if 0 2271 for( c = 0; c < raidPtr->numSpare ; c++) { 2272 sparecol = raidPtr->numCol + c; 2273 if (raidPtr->Disks[r][sparecol].status == rf_ds_used_spare) { 2274 /* 2275 2276 XXX this is where we get fancy and map this spare 2277 into it's correct spot in the array. 2278 2279 */ 2280 /* 2281 2282 we claim this disk is "optimal" if it's 2283 rf_ds_used_spare, as that means it should be 2284 directly substitutable for the disk it replaced. 2285 We note that too... 2286 2287 */ 2288 2289 for(i=0;i<raidPtr->numRow;i++) { 2290 for(j=0;j<raidPtr->numCol;j++) { 2291 if ((raidPtr->Disks[i][j].spareRow == 2292 r) && 2293 (raidPtr->Disks[i][j].spareCol == 2294 sparecol)) { 2295 srow = r; 2296 scol = sparecol; 2297 break; 2298 } 2299 } 2300 } 2301 2302 raidread_component_label( 2303 raidPtr->Disks[r][sparecol].dev, 2304 raidPtr->raid_cinfo[r][sparecol].ci_vp, 2305 &clabel); 2306 /* make sure status is noted */ 2307 clabel.version = RF_COMPONENT_LABEL_VERSION; 2308 clabel.mod_counter = raidPtr->mod_counter; 2309 clabel.serial_number = raidPtr->serial_number; 2310 clabel.row = srow; 2311 clabel.column = scol; 2312 clabel.num_rows = raidPtr->numRow; 2313 clabel.num_columns = raidPtr->numCol; 2314 clabel.clean = RF_RAID_DIRTY; /* changed in a bit*/ 2315 clabel.status = rf_ds_optimal; 2316 raidwrite_component_label( 2317 raidPtr->Disks[r][sparecol].dev, 2318 raidPtr->raid_cinfo[r][sparecol].ci_vp, 2319 &clabel); 2320 raidmarkclean( raidPtr->Disks[r][sparecol].dev, 2321 raidPtr->raid_cinfo[r][sparecol].ci_vp); 2322 } 2323 } 2324 2325 #endif 2326 } 2327 2328 2329 void 2330 rf_update_component_labels(raidPtr) 2331 RF_Raid_t *raidPtr; 2332 { 2333 RF_ComponentLabel_t clabel; 2334 int sparecol; 2335 int r,c; 2336 int i,j; 2337 int srow, scol; 2338 2339 srow = -1; 2340 scol = -1; 2341 2342 /* XXX should do extra checks to make sure things really are clean, 2343 rather than blindly setting the clean bit... */ 2344 2345 raidPtr->mod_counter++; 2346 2347 for (r = 0; r < raidPtr->numRow; r++) { 2348 for (c = 0; c < raidPtr->numCol; c++) { 2349 if (raidPtr->Disks[r][c].status == rf_ds_optimal) { 2350 raidread_component_label( 2351 raidPtr->Disks[r][c].dev, 2352 raidPtr->raid_cinfo[r][c].ci_vp, 2353 &clabel); 2354 /* make sure status is noted */ 2355 clabel.status = rf_ds_optimal; 2356 /* bump the counter */ 2357 clabel.mod_counter = raidPtr->mod_counter; 2358 2359 raidwrite_component_label( 2360 raidPtr->Disks[r][c].dev, 2361 raidPtr->raid_cinfo[r][c].ci_vp, 2362 &clabel); 2363 } 2364 /* else we don't touch it.. */ 2365 } 2366 } 2367 2368 for( c = 0; c < raidPtr->numSpare ; c++) { 2369 sparecol = raidPtr->numCol + c; 2370 if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) { 2371 /* 2372 2373 we claim this disk is "optimal" if it's 2374 rf_ds_used_spare, as that means it should be 2375 directly substitutable for the disk it replaced. 2376 We note that too... 2377 2378 */ 2379 2380 for(i=0;i<raidPtr->numRow;i++) { 2381 for(j=0;j<raidPtr->numCol;j++) { 2382 if ((raidPtr->Disks[i][j].spareRow == 2383 0) && 2384 (raidPtr->Disks[i][j].spareCol == 2385 sparecol)) { 2386 srow = i; 2387 scol = j; 2388 break; 2389 } 2390 } 2391 } 2392 2393 /* XXX shouldn't *really* need this... */ 2394 raidread_component_label( 2395 raidPtr->Disks[0][sparecol].dev, 2396 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2397 &clabel); 2398 /* make sure status is noted */ 2399 2400 raid_init_component_label(raidPtr, &clabel); 2401 2402 clabel.mod_counter = raidPtr->mod_counter; 2403 clabel.row = srow; 2404 clabel.column = scol; 2405 clabel.status = rf_ds_optimal; 2406 2407 raidwrite_component_label( 2408 raidPtr->Disks[0][sparecol].dev, 2409 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2410 &clabel); 2411 } 2412 } 2413 /* printf("Component labels updated\n"); */ 2414 } 2415 2416 2417 void 2418 rf_final_update_component_labels(raidPtr) 2419 RF_Raid_t *raidPtr; 2420 { 2421 RF_ComponentLabel_t clabel; 2422 int sparecol; 2423 int r,c; 2424 int i,j; 2425 int srow, scol; 2426 2427 srow = -1; 2428 scol = -1; 2429 2430 /* XXX should do extra checks to make sure things really are clean, 2431 rather than blindly setting the clean bit... */ 2432 2433 raidPtr->mod_counter++; 2434 2435 for (r = 0; r < raidPtr->numRow; r++) { 2436 for (c = 0; c < raidPtr->numCol; c++) { 2437 if (raidPtr->Disks[r][c].status == rf_ds_optimal) { 2438 raidread_component_label( 2439 raidPtr->Disks[r][c].dev, 2440 raidPtr->raid_cinfo[r][c].ci_vp, 2441 &clabel); 2442 /* make sure status is noted */ 2443 clabel.status = rf_ds_optimal; 2444 /* bump the counter */ 2445 clabel.mod_counter = raidPtr->mod_counter; 2446 2447 raidwrite_component_label( 2448 raidPtr->Disks[r][c].dev, 2449 raidPtr->raid_cinfo[r][c].ci_vp, 2450 &clabel); 2451 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2452 raidmarkclean( 2453 raidPtr->Disks[r][c].dev, 2454 raidPtr->raid_cinfo[r][c].ci_vp, 2455 raidPtr->mod_counter); 2456 } 2457 } 2458 /* else we don't touch it.. */ 2459 } 2460 } 2461 2462 for( c = 0; c < raidPtr->numSpare ; c++) { 2463 sparecol = raidPtr->numCol + c; 2464 if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) { 2465 /* 2466 2467 we claim this disk is "optimal" if it's 2468 rf_ds_used_spare, as that means it should be 2469 directly substitutable for the disk it replaced. 2470 We note that too... 2471 2472 */ 2473 2474 for(i=0;i<raidPtr->numRow;i++) { 2475 for(j=0;j<raidPtr->numCol;j++) { 2476 if ((raidPtr->Disks[i][j].spareRow == 2477 0) && 2478 (raidPtr->Disks[i][j].spareCol == 2479 sparecol)) { 2480 srow = i; 2481 scol = j; 2482 break; 2483 } 2484 } 2485 } 2486 2487 /* XXX shouldn't *really* need this... */ 2488 raidread_component_label( 2489 raidPtr->Disks[0][sparecol].dev, 2490 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2491 &clabel); 2492 /* make sure status is noted */ 2493 2494 raid_init_component_label(raidPtr, &clabel); 2495 2496 clabel.mod_counter = raidPtr->mod_counter; 2497 clabel.row = srow; 2498 clabel.column = scol; 2499 clabel.status = rf_ds_optimal; 2500 2501 raidwrite_component_label( 2502 raidPtr->Disks[0][sparecol].dev, 2503 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2504 &clabel); 2505 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2506 raidmarkclean( raidPtr->Disks[0][sparecol].dev, 2507 raidPtr->raid_cinfo[0][sparecol].ci_vp, 2508 raidPtr->mod_counter); 2509 } 2510 } 2511 } 2512 /* printf("Component labels updated\n"); */ 2513 } 2514 2515 void 2516 rf_close_component(raidPtr, vp, auto_configured) 2517 RF_Raid_t *raidPtr; 2518 struct vnode *vp; 2519 int auto_configured; 2520 { 2521 struct proc *p; 2522 2523 p = raidPtr->engine_thread; 2524 2525 if (vp != NULL) { 2526 if (auto_configured == 1) { 2527 VOP_CLOSE(vp, FREAD, NOCRED, 0); 2528 vput(vp); 2529 2530 } else { 2531 VOP_UNLOCK(vp, 0); 2532 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p); 2533 } 2534 } else { 2535 printf("vnode was NULL\n"); 2536 } 2537 } 2538 2539 2540 void 2541 rf_UnconfigureVnodes(raidPtr) 2542 RF_Raid_t *raidPtr; 2543 { 2544 int r,c; 2545 struct proc *p; 2546 struct vnode *vp; 2547 int acd; 2548 2549 2550 /* We take this opportunity to close the vnodes like we should.. */ 2551 2552 p = raidPtr->engine_thread; 2553 2554 for (r = 0; r < raidPtr->numRow; r++) { 2555 for (c = 0; c < raidPtr->numCol; c++) { 2556 printf("Closing vnode for row: %d col: %d\n", r, c); 2557 vp = raidPtr->raid_cinfo[r][c].ci_vp; 2558 acd = raidPtr->Disks[r][c].auto_configured; 2559 rf_close_component(raidPtr, vp, acd); 2560 raidPtr->raid_cinfo[r][c].ci_vp = NULL; 2561 raidPtr->Disks[r][c].auto_configured = 0; 2562 } 2563 } 2564 for (r = 0; r < raidPtr->numSpare; r++) { 2565 printf("Closing vnode for spare: %d\n", r); 2566 vp = raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp; 2567 acd = raidPtr->Disks[0][raidPtr->numCol + r].auto_configured; 2568 rf_close_component(raidPtr, vp, acd); 2569 raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp = NULL; 2570 raidPtr->Disks[0][raidPtr->numCol + r].auto_configured = 0; 2571 } 2572 } 2573 2574 2575 void 2576 rf_ReconThread(req) 2577 struct rf_recon_req *req; 2578 { 2579 int s; 2580 RF_Raid_t *raidPtr; 2581 2582 s = splbio(); 2583 raidPtr = (RF_Raid_t *) req->raidPtr; 2584 raidPtr->recon_in_progress = 1; 2585 2586 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->row, req->col, 2587 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2588 2589 /* XXX get rid of this! we don't need it at all.. */ 2590 RF_Free(req, sizeof(*req)); 2591 2592 raidPtr->recon_in_progress = 0; 2593 splx(s); 2594 2595 /* That's all... */ 2596 kthread_exit(0); /* does not return */ 2597 } 2598 2599 void 2600 rf_RewriteParityThread(raidPtr) 2601 RF_Raid_t *raidPtr; 2602 { 2603 int retcode; 2604 int s; 2605 2606 raidPtr->parity_rewrite_in_progress = 1; 2607 s = splbio(); 2608 retcode = rf_RewriteParity(raidPtr); 2609 splx(s); 2610 if (retcode) { 2611 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2612 } else { 2613 /* set the clean bit! If we shutdown correctly, 2614 the clean bit on each component label will get 2615 set */ 2616 raidPtr->parity_good = RF_RAID_CLEAN; 2617 } 2618 raidPtr->parity_rewrite_in_progress = 0; 2619 2620 /* Anyone waiting for us to stop? If so, inform them... */ 2621 if (raidPtr->waitShutdown) { 2622 wakeup(&raidPtr->parity_rewrite_in_progress); 2623 } 2624 2625 /* That's all... */ 2626 kthread_exit(0); /* does not return */ 2627 } 2628 2629 2630 void 2631 rf_CopybackThread(raidPtr) 2632 RF_Raid_t *raidPtr; 2633 { 2634 int s; 2635 2636 raidPtr->copyback_in_progress = 1; 2637 s = splbio(); 2638 rf_CopybackReconstructedData(raidPtr); 2639 splx(s); 2640 raidPtr->copyback_in_progress = 0; 2641 2642 /* That's all... */ 2643 kthread_exit(0); /* does not return */ 2644 } 2645 2646 2647 void 2648 rf_ReconstructInPlaceThread(req) 2649 struct rf_recon_req *req; 2650 { 2651 int retcode; 2652 int s; 2653 RF_Raid_t *raidPtr; 2654 2655 s = splbio(); 2656 raidPtr = req->raidPtr; 2657 raidPtr->recon_in_progress = 1; 2658 retcode = rf_ReconstructInPlace(raidPtr, req->row, req->col); 2659 RF_Free(req, sizeof(*req)); 2660 raidPtr->recon_in_progress = 0; 2661 splx(s); 2662 2663 /* That's all... */ 2664 kthread_exit(0); /* does not return */ 2665 } 2666 2667 void 2668 rf_mountroot_hook(dev) 2669 struct device *dev; 2670 { 2671 2672 } 2673 2674 2675 RF_AutoConfig_t * 2676 rf_find_raid_components() 2677 { 2678 struct devnametobdevmaj *dtobdm; 2679 struct vnode *vp; 2680 struct disklabel label; 2681 struct device *dv; 2682 char *cd_name; 2683 dev_t dev; 2684 int error; 2685 int i; 2686 int good_one; 2687 RF_ComponentLabel_t *clabel; 2688 RF_AutoConfig_t *ac_list; 2689 RF_AutoConfig_t *ac; 2690 2691 2692 /* initialize the AutoConfig list */ 2693 ac_list = NULL; 2694 2695 if (raidautoconfig) { 2696 2697 /* we begin by trolling through *all* the devices on the system */ 2698 2699 for (dv = alldevs.tqh_first; dv != NULL; 2700 dv = dv->dv_list.tqe_next) { 2701 2702 /* we are only interested in disks... */ 2703 if (dv->dv_class != DV_DISK) 2704 continue; 2705 2706 /* we don't care about floppies... */ 2707 if (!strcmp(dv->dv_cfdata->cf_driver->cd_name,"fd")) { 2708 continue; 2709 } 2710 2711 /* need to find the device_name_to_block_device_major stuff */ 2712 cd_name = dv->dv_cfdata->cf_driver->cd_name; 2713 dtobdm = dev_name2blk; 2714 while (dtobdm->d_name && strcmp(dtobdm->d_name, cd_name)) { 2715 dtobdm++; 2716 } 2717 2718 /* get a vnode for the raw partition of this disk */ 2719 2720 dev = MAKEDISKDEV(dtobdm->d_maj, dv->dv_unit, RAW_PART); 2721 if (bdevvp(dev, &vp)) 2722 panic("RAID can't alloc vnode"); 2723 2724 error = VOP_OPEN(vp, FREAD, NOCRED, 0); 2725 2726 if (error) { 2727 /* "Who cares." Continue looking 2728 for something that exists*/ 2729 vput(vp); 2730 continue; 2731 } 2732 2733 /* Ok, the disk exists. Go get the disklabel. */ 2734 error = VOP_IOCTL(vp, DIOCGDINFO, (caddr_t)&label, 2735 FREAD, NOCRED, 0); 2736 if (error) { 2737 /* 2738 * XXX can't happen - open() would 2739 * have errored out (or faked up one) 2740 */ 2741 printf("can't get label for dev %s%c (%d)!?!?\n", 2742 dv->dv_xname, 'a' + RAW_PART, error); 2743 } 2744 2745 /* don't need this any more. We'll allocate it again 2746 a little later if we really do... */ 2747 VOP_CLOSE(vp, FREAD, NOCRED, 0); 2748 vput(vp); 2749 2750 for (i=0; i < label.d_npartitions; i++) { 2751 /* We only support partitions marked as RAID */ 2752 if (label.d_partitions[i].p_fstype != FS_RAID) 2753 continue; 2754 2755 dev = MAKEDISKDEV(dtobdm->d_maj, dv->dv_unit, i); 2756 if (bdevvp(dev, &vp)) 2757 panic("RAID can't alloc vnode"); 2758 2759 error = VOP_OPEN(vp, FREAD, NOCRED, 0); 2760 if (error) { 2761 /* Whatever... */ 2762 vput(vp); 2763 continue; 2764 } 2765 2766 good_one = 0; 2767 2768 clabel = (RF_ComponentLabel_t *) 2769 malloc(sizeof(RF_ComponentLabel_t), 2770 M_RAIDFRAME, M_NOWAIT); 2771 if (clabel == NULL) { 2772 /* XXX CLEANUP HERE */ 2773 printf("RAID auto config: out of memory!\n"); 2774 return(NULL); /* XXX probably should panic? */ 2775 } 2776 2777 if (!raidread_component_label(dev, vp, clabel)) { 2778 /* Got the label. Does it look reasonable? */ 2779 if (rf_reasonable_label(clabel) && 2780 (clabel->partitionSize <= 2781 label.d_partitions[i].p_size)) { 2782 #if DEBUG 2783 printf("Component on: %s%c: %d\n", 2784 dv->dv_xname, 'a'+i, 2785 label.d_partitions[i].p_size); 2786 rf_print_component_label(clabel); 2787 #endif 2788 /* if it's reasonable, add it, 2789 else ignore it. */ 2790 ac = (RF_AutoConfig_t *) 2791 malloc(sizeof(RF_AutoConfig_t), 2792 M_RAIDFRAME, 2793 M_NOWAIT); 2794 if (ac == NULL) { 2795 /* XXX should panic?? */ 2796 return(NULL); 2797 } 2798 2799 sprintf(ac->devname, "%s%c", 2800 dv->dv_xname, 'a'+i); 2801 ac->dev = dev; 2802 ac->vp = vp; 2803 ac->clabel = clabel; 2804 ac->next = ac_list; 2805 ac_list = ac; 2806 good_one = 1; 2807 } 2808 } 2809 if (!good_one) { 2810 /* cleanup */ 2811 free(clabel, M_RAIDFRAME); 2812 VOP_CLOSE(vp, FREAD, NOCRED, 0); 2813 vput(vp); 2814 } 2815 } 2816 } 2817 } 2818 return(ac_list); 2819 } 2820 2821 static int 2822 rf_reasonable_label(clabel) 2823 RF_ComponentLabel_t *clabel; 2824 { 2825 2826 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 2827 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 2828 ((clabel->clean == RF_RAID_CLEAN) || 2829 (clabel->clean == RF_RAID_DIRTY)) && 2830 clabel->row >=0 && 2831 clabel->column >= 0 && 2832 clabel->num_rows > 0 && 2833 clabel->num_columns > 0 && 2834 clabel->row < clabel->num_rows && 2835 clabel->column < clabel->num_columns && 2836 clabel->blockSize > 0 && 2837 clabel->numBlocks > 0) { 2838 /* label looks reasonable enough... */ 2839 return(1); 2840 } 2841 return(0); 2842 } 2843 2844 2845 void 2846 rf_print_component_label(clabel) 2847 RF_ComponentLabel_t *clabel; 2848 { 2849 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 2850 clabel->row, clabel->column, 2851 clabel->num_rows, clabel->num_columns); 2852 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 2853 clabel->version, clabel->serial_number, 2854 clabel->mod_counter); 2855 printf(" Clean: %s Status: %d\n", 2856 clabel->clean ? "Yes" : "No", clabel->status ); 2857 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 2858 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 2859 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n", 2860 (char) clabel->parityConfig, clabel->blockSize, 2861 clabel->numBlocks); 2862 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" ); 2863 printf(" Contains root partition: %s\n", 2864 clabel->root_partition ? "Yes" : "No" ); 2865 printf(" Last configured as: raid%d\n", clabel->last_unit ); 2866 #if 0 2867 printf(" Config order: %d\n", clabel->config_order); 2868 #endif 2869 2870 } 2871 2872 RF_ConfigSet_t * 2873 rf_create_auto_sets(ac_list) 2874 RF_AutoConfig_t *ac_list; 2875 { 2876 RF_AutoConfig_t *ac; 2877 RF_ConfigSet_t *config_sets; 2878 RF_ConfigSet_t *cset; 2879 RF_AutoConfig_t *ac_next; 2880 2881 2882 config_sets = NULL; 2883 2884 /* Go through the AutoConfig list, and figure out which components 2885 belong to what sets. */ 2886 ac = ac_list; 2887 while(ac!=NULL) { 2888 /* we're going to putz with ac->next, so save it here 2889 for use at the end of the loop */ 2890 ac_next = ac->next; 2891 2892 if (config_sets == NULL) { 2893 /* will need at least this one... */ 2894 config_sets = (RF_ConfigSet_t *) 2895 malloc(sizeof(RF_ConfigSet_t), 2896 M_RAIDFRAME, M_NOWAIT); 2897 if (config_sets == NULL) { 2898 panic("rf_create_auto_sets: No memory!\n"); 2899 } 2900 /* this one is easy :) */ 2901 config_sets->ac = ac; 2902 config_sets->next = NULL; 2903 config_sets->rootable = 0; 2904 ac->next = NULL; 2905 } else { 2906 /* which set does this component fit into? */ 2907 cset = config_sets; 2908 while(cset!=NULL) { 2909 if (rf_does_it_fit(cset, ac)) { 2910 /* looks like it matches... */ 2911 ac->next = cset->ac; 2912 cset->ac = ac; 2913 break; 2914 } 2915 cset = cset->next; 2916 } 2917 if (cset==NULL) { 2918 /* didn't find a match above... new set..*/ 2919 cset = (RF_ConfigSet_t *) 2920 malloc(sizeof(RF_ConfigSet_t), 2921 M_RAIDFRAME, M_NOWAIT); 2922 if (cset == NULL) { 2923 panic("rf_create_auto_sets: No memory!\n"); 2924 } 2925 cset->ac = ac; 2926 ac->next = NULL; 2927 cset->next = config_sets; 2928 cset->rootable = 0; 2929 config_sets = cset; 2930 } 2931 } 2932 ac = ac_next; 2933 } 2934 2935 2936 return(config_sets); 2937 } 2938 2939 static int 2940 rf_does_it_fit(cset, ac) 2941 RF_ConfigSet_t *cset; 2942 RF_AutoConfig_t *ac; 2943 { 2944 RF_ComponentLabel_t *clabel1, *clabel2; 2945 2946 /* If this one matches the *first* one in the set, that's good 2947 enough, since the other members of the set would have been 2948 through here too... */ 2949 /* note that we are not checking partitionSize here.. 2950 2951 Note that we are also not checking the mod_counters here. 2952 If everything else matches execpt the mod_counter, that's 2953 good enough for this test. We will deal with the mod_counters 2954 a little later in the autoconfiguration process. 2955 2956 (clabel1->mod_counter == clabel2->mod_counter) && 2957 2958 The reason we don't check for this is that failed disks 2959 will have lower modification counts. If those disks are 2960 not added to the set they used to belong to, then they will 2961 form their own set, which may result in 2 different sets, 2962 for example, competing to be configured at raid0, and 2963 perhaps competing to be the root filesystem set. If the 2964 wrong ones get configured, or both attempt to become /, 2965 weird behaviour and or serious lossage will occur. Thus we 2966 need to bring them into the fold here, and kick them out at 2967 a later point. 2968 2969 */ 2970 2971 clabel1 = cset->ac->clabel; 2972 clabel2 = ac->clabel; 2973 if ((clabel1->version == clabel2->version) && 2974 (clabel1->serial_number == clabel2->serial_number) && 2975 (clabel1->num_rows == clabel2->num_rows) && 2976 (clabel1->num_columns == clabel2->num_columns) && 2977 (clabel1->sectPerSU == clabel2->sectPerSU) && 2978 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 2979 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 2980 (clabel1->parityConfig == clabel2->parityConfig) && 2981 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 2982 (clabel1->blockSize == clabel2->blockSize) && 2983 (clabel1->numBlocks == clabel2->numBlocks) && 2984 (clabel1->autoconfigure == clabel2->autoconfigure) && 2985 (clabel1->root_partition == clabel2->root_partition) && 2986 (clabel1->last_unit == clabel2->last_unit) && 2987 (clabel1->config_order == clabel2->config_order)) { 2988 /* if it get's here, it almost *has* to be a match */ 2989 } else { 2990 /* it's not consistent with somebody in the set.. 2991 punt */ 2992 return(0); 2993 } 2994 /* all was fine.. it must fit... */ 2995 return(1); 2996 } 2997 2998 int 2999 rf_have_enough_components(cset) 3000 RF_ConfigSet_t *cset; 3001 { 3002 RF_AutoConfig_t *ac; 3003 RF_AutoConfig_t *auto_config; 3004 RF_ComponentLabel_t *clabel; 3005 int r,c; 3006 int num_rows; 3007 int num_cols; 3008 int num_missing; 3009 int mod_counter; 3010 3011 /* check to see that we have enough 'live' components 3012 of this set. If so, we can configure it if necessary */ 3013 3014 num_rows = cset->ac->clabel->num_rows; 3015 num_cols = cset->ac->clabel->num_columns; 3016 3017 /* XXX Check for duplicate components!?!?!? */ 3018 3019 /* Determine what the mod_counter is supposed to be for this set. */ 3020 3021 mod_counter = -1; 3022 ac = cset->ac; 3023 while(ac!=NULL) { 3024 if (ac->clabel->mod_counter > mod_counter) { 3025 mod_counter = ac->clabel->mod_counter; 3026 } 3027 ac = ac->next; 3028 } 3029 3030 num_missing = 0; 3031 auto_config = cset->ac; 3032 3033 for(r=0; r<num_rows; r++) { 3034 for(c=0; c<num_cols; c++) { 3035 ac = auto_config; 3036 while(ac!=NULL) { 3037 if (ac->clabel==NULL) { 3038 /* big-time bad news. */ 3039 goto fail; 3040 } 3041 if ((ac->clabel->row == r) && 3042 (ac->clabel->column == c) && 3043 (ac->clabel->mod_counter == mod_counter)) { 3044 /* it's this one... */ 3045 #if DEBUG 3046 printf("Found: %s at %d,%d\n", 3047 ac->devname,r,c); 3048 #endif 3049 break; 3050 } 3051 ac=ac->next; 3052 } 3053 if (ac==NULL) { 3054 /* Didn't find one here! */ 3055 num_missing++; 3056 } 3057 } 3058 } 3059 3060 clabel = cset->ac->clabel; 3061 3062 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3063 ((clabel->parityConfig == '1') && (num_missing > 1)) || 3064 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3065 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3066 /* XXX this needs to be made *much* more general */ 3067 /* Too many failures */ 3068 return(0); 3069 } 3070 /* otherwise, all is well, and we've got enough to take a kick 3071 at autoconfiguring this set */ 3072 return(1); 3073 fail: 3074 return(0); 3075 3076 } 3077 3078 void 3079 rf_create_configuration(ac,config,raidPtr) 3080 RF_AutoConfig_t *ac; 3081 RF_Config_t *config; 3082 RF_Raid_t *raidPtr; 3083 { 3084 RF_ComponentLabel_t *clabel; 3085 int i; 3086 3087 clabel = ac->clabel; 3088 3089 /* 1. Fill in the common stuff */ 3090 config->numRow = clabel->num_rows; 3091 config->numCol = clabel->num_columns; 3092 config->numSpare = 0; /* XXX should this be set here? */ 3093 config->sectPerSU = clabel->sectPerSU; 3094 config->SUsPerPU = clabel->SUsPerPU; 3095 config->SUsPerRU = clabel->SUsPerRU; 3096 config->parityConfig = clabel->parityConfig; 3097 /* XXX... */ 3098 strcpy(config->diskQueueType,"fifo"); 3099 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3100 config->layoutSpecificSize = 0; /* XXX ?? */ 3101 3102 while(ac!=NULL) { 3103 /* row/col values will be in range due to the checks 3104 in reasonable_label() */ 3105 strcpy(config->devnames[ac->clabel->row][ac->clabel->column], 3106 ac->devname); 3107 ac = ac->next; 3108 } 3109 3110 for(i=0;i<RF_MAXDBGV;i++) { 3111 config->debugVars[i][0] = NULL; 3112 } 3113 } 3114 3115 int 3116 rf_set_autoconfig(raidPtr, new_value) 3117 RF_Raid_t *raidPtr; 3118 int new_value; 3119 { 3120 RF_ComponentLabel_t clabel; 3121 struct vnode *vp; 3122 dev_t dev; 3123 int row, column; 3124 3125 raidPtr->autoconfigure = new_value; 3126 for(row=0; row<raidPtr->numRow; row++) { 3127 for(column=0; column<raidPtr->numCol; column++) { 3128 if (raidPtr->Disks[row][column].status == 3129 rf_ds_optimal) { 3130 dev = raidPtr->Disks[row][column].dev; 3131 vp = raidPtr->raid_cinfo[row][column].ci_vp; 3132 raidread_component_label(dev, vp, &clabel); 3133 clabel.autoconfigure = new_value; 3134 raidwrite_component_label(dev, vp, &clabel); 3135 } 3136 } 3137 } 3138 return(new_value); 3139 } 3140 3141 int 3142 rf_set_rootpartition(raidPtr, new_value) 3143 RF_Raid_t *raidPtr; 3144 int new_value; 3145 { 3146 RF_ComponentLabel_t clabel; 3147 struct vnode *vp; 3148 dev_t dev; 3149 int row, column; 3150 3151 raidPtr->root_partition = new_value; 3152 for(row=0; row<raidPtr->numRow; row++) { 3153 for(column=0; column<raidPtr->numCol; column++) { 3154 if (raidPtr->Disks[row][column].status == 3155 rf_ds_optimal) { 3156 dev = raidPtr->Disks[row][column].dev; 3157 vp = raidPtr->raid_cinfo[row][column].ci_vp; 3158 raidread_component_label(dev, vp, &clabel); 3159 clabel.root_partition = new_value; 3160 raidwrite_component_label(dev, vp, &clabel); 3161 } 3162 } 3163 } 3164 return(new_value); 3165 } 3166 3167 void 3168 rf_release_all_vps(cset) 3169 RF_ConfigSet_t *cset; 3170 { 3171 RF_AutoConfig_t *ac; 3172 3173 ac = cset->ac; 3174 while(ac!=NULL) { 3175 /* Close the vp, and give it back */ 3176 if (ac->vp) { 3177 VOP_CLOSE(ac->vp, FREAD, NOCRED, 0); 3178 vput(ac->vp); 3179 ac->vp = NULL; 3180 } 3181 ac = ac->next; 3182 } 3183 } 3184 3185 3186 void 3187 rf_cleanup_config_set(cset) 3188 RF_ConfigSet_t *cset; 3189 { 3190 RF_AutoConfig_t *ac; 3191 RF_AutoConfig_t *next_ac; 3192 3193 ac = cset->ac; 3194 while(ac!=NULL) { 3195 next_ac = ac->next; 3196 /* nuke the label */ 3197 free(ac->clabel, M_RAIDFRAME); 3198 /* cleanup the config structure */ 3199 free(ac, M_RAIDFRAME); 3200 /* "next.." */ 3201 ac = next_ac; 3202 } 3203 /* and, finally, nuke the config set */ 3204 free(cset, M_RAIDFRAME); 3205 } 3206 3207 3208 void 3209 raid_init_component_label(raidPtr, clabel) 3210 RF_Raid_t *raidPtr; 3211 RF_ComponentLabel_t *clabel; 3212 { 3213 /* current version number */ 3214 clabel->version = RF_COMPONENT_LABEL_VERSION; 3215 clabel->serial_number = raidPtr->serial_number; 3216 clabel->mod_counter = raidPtr->mod_counter; 3217 clabel->num_rows = raidPtr->numRow; 3218 clabel->num_columns = raidPtr->numCol; 3219 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3220 clabel->status = rf_ds_optimal; /* "It's good!" */ 3221 3222 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3223 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3224 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3225 3226 clabel->blockSize = raidPtr->bytesPerSector; 3227 clabel->numBlocks = raidPtr->sectorsPerDisk; 3228 3229 /* XXX not portable */ 3230 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3231 clabel->maxOutstanding = raidPtr->maxOutstanding; 3232 clabel->autoconfigure = raidPtr->autoconfigure; 3233 clabel->root_partition = raidPtr->root_partition; 3234 clabel->last_unit = raidPtr->raidid; 3235 clabel->config_order = raidPtr->config_order; 3236 } 3237 3238 int 3239 rf_auto_config_set(cset,unit) 3240 RF_ConfigSet_t *cset; 3241 int *unit; 3242 { 3243 RF_Raid_t *raidPtr; 3244 RF_Config_t *config; 3245 int raidID; 3246 int retcode; 3247 3248 printf("RAID autoconfigure\n"); 3249 3250 retcode = 0; 3251 *unit = -1; 3252 3253 /* 1. Create a config structure */ 3254 3255 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3256 M_RAIDFRAME, 3257 M_NOWAIT); 3258 if (config==NULL) { 3259 printf("Out of mem!?!?\n"); 3260 /* XXX do something more intelligent here. */ 3261 return(1); 3262 } 3263 3264 memset(config, 0, sizeof(RF_Config_t)); 3265 3266 /* XXX raidID needs to be set correctly.. */ 3267 3268 /* 3269 2. Figure out what RAID ID this one is supposed to live at 3270 See if we can get the same RAID dev that it was configured 3271 on last time.. 3272 */ 3273 3274 raidID = cset->ac->clabel->last_unit; 3275 if ((raidID < 0) || (raidID >= numraid)) { 3276 /* let's not wander off into lala land. */ 3277 raidID = numraid - 1; 3278 } 3279 if (raidPtrs[raidID]->valid != 0) { 3280 3281 /* 3282 Nope... Go looking for an alternative... 3283 Start high so we don't immediately use raid0 if that's 3284 not taken. 3285 */ 3286 3287 for(raidID = numraid; raidID >= 0; raidID--) { 3288 if (raidPtrs[raidID]->valid == 0) { 3289 /* can use this one! */ 3290 break; 3291 } 3292 } 3293 } 3294 3295 if (raidID < 0) { 3296 /* punt... */ 3297 printf("Unable to auto configure this set!\n"); 3298 printf("(Out of RAID devs!)\n"); 3299 return(1); 3300 } 3301 printf("Configuring raid%d:\n",raidID); 3302 raidPtr = raidPtrs[raidID]; 3303 3304 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3305 raidPtr->raidid = raidID; 3306 raidPtr->openings = RAIDOUTSTANDING; 3307 3308 /* 3. Build the configuration structure */ 3309 rf_create_configuration(cset->ac, config, raidPtr); 3310 3311 /* 4. Do the configuration */ 3312 retcode = rf_Configure(raidPtr, config, cset->ac); 3313 3314 if (retcode == 0) { 3315 3316 raidinit(raidPtrs[raidID]); 3317 3318 rf_markalldirty(raidPtrs[raidID]); 3319 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3320 if (cset->ac->clabel->root_partition==1) { 3321 /* everything configured just fine. Make a note 3322 that this set is eligible to be root. */ 3323 cset->rootable = 1; 3324 /* XXX do this here? */ 3325 raidPtrs[raidID]->root_partition = 1; 3326 } 3327 } 3328 3329 /* 5. Cleanup */ 3330 free(config, M_RAIDFRAME); 3331 3332 *unit = raidID; 3333 return(retcode); 3334 } 3335
Indexes created Tue Sep 23 23:09:58 GMT 2025