rf_netbsdkintf.c revision 1.201
1 /* $NetBSD: rf_netbsdkintf.c,v 1.201 2006/02/24 03:20:22 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) 1990, 1993 40 * The Regents of the University of California. All rights reserved. 41 * 42 * This code is derived from software contributed to Berkeley by 43 * the Systems Programming Group of the University of Utah Computer 44 * Science Department. 45 * 46 * Redistribution and use in source and binary forms, with or without 47 * modification, are permitted provided that the following conditions 48 * are met: 49 * 1. Redistributions of source code must retain the above copyright 50 * notice, this list of conditions and the following disclaimer. 51 * 2. Redistributions in binary form must reproduce the above copyright 52 * notice, this list of conditions and the following disclaimer in the 53 * documentation and/or other materials provided with the distribution. 54 * 3. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 71 * 72 * @(#)cd.c 8.2 (Berkeley) 11/16/93 73 */ 74 75 /* 76 * Copyright (c) 1988 University of Utah. 77 * 78 * This code is derived from software contributed to Berkeley by 79 * the Systems Programming Group of the University of Utah Computer 80 * Science Department. 81 * 82 * Redistribution and use in source and binary forms, with or without 83 * modification, are permitted provided that the following conditions 84 * are met: 85 * 1. Redistributions of source code must retain the above copyright 86 * notice, this list of conditions and the following disclaimer. 87 * 2. Redistributions in binary form must reproduce the above copyright 88 * notice, this list of conditions and the following disclaimer in the 89 * documentation and/or other materials provided with the distribution. 90 * 3. All advertising materials mentioning features or use of this software 91 * must display the following acknowledgement: 92 * This product includes software developed by the University of 93 * California, Berkeley and its contributors. 94 * 4. Neither the name of the University nor the names of its contributors 95 * may be used to endorse or promote products derived from this software 96 * without specific prior written permission. 97 * 98 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 99 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 100 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 101 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 102 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 103 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 104 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 105 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 106 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 107 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 108 * SUCH DAMAGE. 109 * 110 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 111 * 112 * @(#)cd.c 8.2 (Berkeley) 11/16/93 113 */ 114 115 /* 116 * Copyright (c) 1995 Carnegie-Mellon University. 117 * All rights reserved. 118 * 119 * Authors: Mark Holland, Jim Zelenka 120 * 121 * Permission to use, copy, modify and distribute this software and 122 * its documentation is hereby granted, provided that both the copyright 123 * notice and this permission notice appear in all copies of the 124 * software, derivative works or modified versions, and any portions 125 * thereof, and that both notices appear in supporting documentation. 126 * 127 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 128 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 129 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 130 * 131 * Carnegie Mellon requests users of this software to return to 132 * 133 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU 134 * School of Computer Science 135 * Carnegie Mellon University 136 * Pittsburgh PA 15213-3890 137 * 138 * any improvements or extensions that they make and grant Carnegie the 139 * rights to redistribute these changes. 140 */ 141 142 /*********************************************************** 143 * 144 * rf_kintf.c -- the kernel interface routines for RAIDframe 145 * 146 ***********************************************************/ 147 148 #include <sys/cdefs.h> 149 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.201 2006/02/24 03:20:22 oster Exp $"); 150 151 #include <sys/param.h> 152 #include <sys/errno.h> 153 #include <sys/pool.h> 154 #include <sys/proc.h> 155 #include <sys/queue.h> 156 #include <sys/disk.h> 157 #include <sys/device.h> 158 #include <sys/stat.h> 159 #include <sys/ioctl.h> 160 #include <sys/fcntl.h> 161 #include <sys/systm.h> 162 #include <sys/namei.h> 163 #include <sys/vnode.h> 164 #include <sys/disklabel.h> 165 #include <sys/conf.h> 166 #include <sys/lock.h> 167 #include <sys/buf.h> 168 #include <sys/bufq.h> 169 #include <sys/user.h> 170 #include <sys/reboot.h> 171 172 #include <dev/raidframe/raidframevar.h> 173 #include <dev/raidframe/raidframeio.h> 174 #include "raid.h" 175 #include "opt_raid_autoconfig.h" 176 #include "rf_raid.h" 177 #include "rf_copyback.h" 178 #include "rf_dag.h" 179 #include "rf_dagflags.h" 180 #include "rf_desc.h" 181 #include "rf_diskqueue.h" 182 #include "rf_etimer.h" 183 #include "rf_general.h" 184 #include "rf_kintf.h" 185 #include "rf_options.h" 186 #include "rf_driver.h" 187 #include "rf_parityscan.h" 188 #include "rf_threadstuff.h" 189 190 #ifdef DEBUG 191 int rf_kdebug_level = 0; 192 #define db1_printf(a) if (rf_kdebug_level > 0) printf a 193 #else /* DEBUG */ 194 #define db1_printf(a) { } 195 #endif /* DEBUG */ 196 197 static RF_Raid_t **raidPtrs; /* global raid device descriptors */ 198 199 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex) 200 201 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a 202 * spare table */ 203 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from 204 * installation process */ 205 206 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures"); 207 208 /* prototypes */ 209 static void KernelWakeupFunc(struct buf *); 210 static void InitBP(struct buf *, struct vnode *, unsigned, 211 dev_t, RF_SectorNum_t, RF_SectorCount_t, caddr_t, void (*) (struct buf *), 212 void *, int, struct proc *); 213 static void raidinit(RF_Raid_t *); 214 215 void raidattach(int); 216 217 dev_type_open(raidopen); 218 dev_type_close(raidclose); 219 dev_type_read(raidread); 220 dev_type_write(raidwrite); 221 dev_type_ioctl(raidioctl); 222 dev_type_strategy(raidstrategy); 223 dev_type_dump(raiddump); 224 dev_type_size(raidsize); 225 226 const struct bdevsw raid_bdevsw = { 227 raidopen, raidclose, raidstrategy, raidioctl, 228 raiddump, raidsize, D_DISK 229 }; 230 231 const struct cdevsw raid_cdevsw = { 232 raidopen, raidclose, raidread, raidwrite, raidioctl, 233 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 234 }; 235 236 /* XXX Not sure if the following should be replacing the raidPtrs above, 237 or if it should be used in conjunction with that... 238 */ 239 240 struct raid_softc { 241 int sc_flags; /* flags */ 242 int sc_cflags; /* configuration flags */ 243 size_t sc_size; /* size of the raid device */ 244 char sc_xname[20]; /* XXX external name */ 245 struct disk sc_dkdev; /* generic disk device info */ 246 struct bufq_state *buf_queue; /* used for the device queue */ 247 }; 248 /* sc_flags */ 249 #define RAIDF_INITED 0x01 /* unit has been initialized */ 250 #define RAIDF_WLABEL 0x02 /* label area is writable */ 251 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 252 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 253 #define RAIDF_LOCKED 0x80 /* unit is locked */ 254 255 #define raidunit(x) DISKUNIT(x) 256 int numraid = 0; 257 258 /* 259 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 260 * Be aware that large numbers can allow the driver to consume a lot of 261 * kernel memory, especially on writes, and in degraded mode reads. 262 * 263 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 264 * a single 64K write will typically require 64K for the old data, 265 * 64K for the old parity, and 64K for the new parity, for a total 266 * of 192K (if the parity buffer is not re-used immediately). 267 * Even it if is used immediately, that's still 128K, which when multiplied 268 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 269 * 270 * Now in degraded mode, for example, a 64K read on the above setup may 271 * require data reconstruction, which will require *all* of the 4 remaining 272 * disks to participate -- 4 * 32K/disk == 128K again. 273 */ 274 275 #ifndef RAIDOUTSTANDING 276 #define RAIDOUTSTANDING 6 277 #endif 278 279 #define RAIDLABELDEV(dev) \ 280 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 281 282 /* declared here, and made public, for the benefit of KVM stuff.. */ 283 struct raid_softc *raid_softc; 284 285 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *, 286 struct disklabel *); 287 static void raidgetdisklabel(dev_t); 288 static void raidmakedisklabel(struct raid_softc *); 289 290 static int raidlock(struct raid_softc *); 291 static void raidunlock(struct raid_softc *); 292 293 static void rf_markalldirty(RF_Raid_t *); 294 295 struct device *raidrootdev; 296 297 void rf_ReconThread(struct rf_recon_req *); 298 void rf_RewriteParityThread(RF_Raid_t *raidPtr); 299 void rf_CopybackThread(RF_Raid_t *raidPtr); 300 void rf_ReconstructInPlaceThread(struct rf_recon_req *); 301 int rf_autoconfig(struct device *self); 302 void rf_buildroothack(RF_ConfigSet_t *); 303 304 RF_AutoConfig_t *rf_find_raid_components(void); 305 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *); 306 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *); 307 static int rf_reasonable_label(RF_ComponentLabel_t *); 308 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *); 309 int rf_set_autoconfig(RF_Raid_t *, int); 310 int rf_set_rootpartition(RF_Raid_t *, int); 311 void rf_release_all_vps(RF_ConfigSet_t *); 312 void rf_cleanup_config_set(RF_ConfigSet_t *); 313 int rf_have_enough_components(RF_ConfigSet_t *); 314 int rf_auto_config_set(RF_ConfigSet_t *, int *); 315 316 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not 317 allow autoconfig to take place. 318 Note that this is overridden by having 319 RAID_AUTOCONFIG as an option in the 320 kernel config file. */ 321 322 struct RF_Pools_s rf_pools; 323 324 void 325 raidattach(int num) 326 { 327 int raidID; 328 int i, rc; 329 330 #ifdef DEBUG 331 printf("raidattach: Asked for %d units\n", num); 332 #endif 333 334 if (num <= 0) { 335 #ifdef DIAGNOSTIC 336 panic("raidattach: count <= 0"); 337 #endif 338 return; 339 } 340 /* This is where all the initialization stuff gets done. */ 341 342 numraid = num; 343 344 /* Make some space for requested number of units... */ 345 346 RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **)); 347 if (raidPtrs == NULL) { 348 panic("raidPtrs is NULL!!"); 349 } 350 351 rf_mutex_init(&rf_sparet_wait_mutex); 352 353 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 354 355 for (i = 0; i < num; i++) 356 raidPtrs[i] = NULL; 357 rc = rf_BootRaidframe(); 358 if (rc == 0) 359 printf("Kernelized RAIDframe activated\n"); 360 else 361 panic("Serious error booting RAID!!"); 362 363 /* put together some datastructures like the CCD device does.. This 364 * lets us lock the device and what-not when it gets opened. */ 365 366 raid_softc = (struct raid_softc *) 367 malloc(num * sizeof(struct raid_softc), 368 M_RAIDFRAME, M_NOWAIT); 369 if (raid_softc == NULL) { 370 printf("WARNING: no memory for RAIDframe driver\n"); 371 return; 372 } 373 374 memset(raid_softc, 0, num * sizeof(struct raid_softc)); 375 376 raidrootdev = (struct device *)malloc(num * sizeof(struct device), 377 M_RAIDFRAME, M_NOWAIT); 378 if (raidrootdev == NULL) { 379 panic("No memory for RAIDframe driver!!?!?!"); 380 } 381 382 for (raidID = 0; raidID < num; raidID++) { 383 bufq_alloc(&raid_softc[raidID].buf_queue, "fcfs", 0); 384 pseudo_disk_init(&raid_softc[raidID].sc_dkdev); 385 386 /* XXXJRT Should use config_attach_pseudo() */ 387 388 raidrootdev[raidID].dv_class = DV_DISK; 389 raidrootdev[raidID].dv_cfdata = NULL; 390 raidrootdev[raidID].dv_unit = raidID; 391 raidrootdev[raidID].dv_parent = NULL; 392 raidrootdev[raidID].dv_flags = 0; 393 snprintf(raidrootdev[raidID].dv_xname, 394 sizeof(raidrootdev[raidID].dv_xname), "raid%d", raidID); 395 396 RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t), 397 (RF_Raid_t *)); 398 if (raidPtrs[raidID] == NULL) { 399 printf("WARNING: raidPtrs[%d] is NULL\n", raidID); 400 numraid = raidID; 401 return; 402 } 403 } 404 405 #ifdef RAID_AUTOCONFIG 406 raidautoconfig = 1; 407 #endif 408 409 /* 410 * Register a finalizer which will be used to auto-config RAID 411 * sets once all real hardware devices have been found. 412 */ 413 if (config_finalize_register(NULL, rf_autoconfig) != 0) 414 printf("WARNING: unable to register RAIDframe finalizer\n"); 415 } 416 417 int 418 rf_autoconfig(struct device *self) 419 { 420 RF_AutoConfig_t *ac_list; 421 RF_ConfigSet_t *config_sets; 422 423 if (raidautoconfig == 0) 424 return (0); 425 426 /* XXX This code can only be run once. */ 427 raidautoconfig = 0; 428 429 /* 1. locate all RAID components on the system */ 430 #ifdef DEBUG 431 printf("Searching for RAID components...\n"); 432 #endif 433 ac_list = rf_find_raid_components(); 434 435 /* 2. Sort them into their respective sets. */ 436 config_sets = rf_create_auto_sets(ac_list); 437 438 /* 439 * 3. Evaluate each set andconfigure the valid ones. 440 * This gets done in rf_buildroothack(). 441 */ 442 rf_buildroothack(config_sets); 443 444 return (1); 445 } 446 447 void 448 rf_buildroothack(RF_ConfigSet_t *config_sets) 449 { 450 RF_ConfigSet_t *cset; 451 RF_ConfigSet_t *next_cset; 452 int retcode; 453 int raidID; 454 int rootID; 455 int num_root; 456 457 rootID = 0; 458 num_root = 0; 459 cset = config_sets; 460 while(cset != NULL ) { 461 next_cset = cset->next; 462 if (rf_have_enough_components(cset) && 463 cset->ac->clabel->autoconfigure==1) { 464 retcode = rf_auto_config_set(cset,&raidID); 465 if (!retcode) { 466 if (cset->rootable) { 467 rootID = raidID; 468 num_root++; 469 } 470 } else { 471 /* The autoconfig didn't work :( */ 472 #if DEBUG 473 printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 474 #endif 475 rf_release_all_vps(cset); 476 } 477 } else { 478 /* we're not autoconfiguring this set... 479 release the associated resources */ 480 rf_release_all_vps(cset); 481 } 482 /* cleanup */ 483 rf_cleanup_config_set(cset); 484 cset = next_cset; 485 } 486 487 /* we found something bootable... */ 488 489 if (num_root == 1) { 490 booted_device = &raidrootdev[rootID]; 491 } else if (num_root > 1) { 492 /* we can't guess.. require the user to answer... */ 493 boothowto |= RB_ASKNAME; 494 } 495 } 496 497 498 int 499 raidsize(dev_t dev) 500 { 501 struct raid_softc *rs; 502 struct disklabel *lp; 503 int part, unit, omask, size; 504 505 unit = raidunit(dev); 506 if (unit >= numraid) 507 return (-1); 508 rs = &raid_softc[unit]; 509 510 if ((rs->sc_flags & RAIDF_INITED) == 0) 511 return (-1); 512 513 part = DISKPART(dev); 514 omask = rs->sc_dkdev.dk_openmask & (1 << part); 515 lp = rs->sc_dkdev.dk_label; 516 517 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curlwp)) 518 return (-1); 519 520 if (lp->d_partitions[part].p_fstype != FS_SWAP) 521 size = -1; 522 else 523 size = lp->d_partitions[part].p_size * 524 (lp->d_secsize / DEV_BSIZE); 525 526 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curlwp)) 527 return (-1); 528 529 return (size); 530 531 } 532 533 int 534 raiddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size) 535 { 536 /* Not implemented. */ 537 return ENXIO; 538 } 539 /* ARGSUSED */ 540 int 541 raidopen(dev_t dev, int flags, int fmt, struct lwp *l) 542 { 543 int unit = raidunit(dev); 544 struct raid_softc *rs; 545 struct disklabel *lp; 546 int part, pmask; 547 int error = 0; 548 549 if (unit >= numraid) 550 return (ENXIO); 551 rs = &raid_softc[unit]; 552 553 if ((error = raidlock(rs)) != 0) 554 return (error); 555 lp = rs->sc_dkdev.dk_label; 556 557 part = DISKPART(dev); 558 pmask = (1 << part); 559 560 if ((rs->sc_flags & RAIDF_INITED) && 561 (rs->sc_dkdev.dk_openmask == 0)) 562 raidgetdisklabel(dev); 563 564 /* make sure that this partition exists */ 565 566 if (part != RAW_PART) { 567 if (((rs->sc_flags & RAIDF_INITED) == 0) || 568 ((part >= lp->d_npartitions) || 569 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 570 error = ENXIO; 571 raidunlock(rs); 572 return (error); 573 } 574 } 575 /* Prevent this unit from being unconfigured while open. */ 576 switch (fmt) { 577 case S_IFCHR: 578 rs->sc_dkdev.dk_copenmask |= pmask; 579 break; 580 581 case S_IFBLK: 582 rs->sc_dkdev.dk_bopenmask |= pmask; 583 break; 584 } 585 586 if ((rs->sc_dkdev.dk_openmask == 0) && 587 ((rs->sc_flags & RAIDF_INITED) != 0)) { 588 /* First one... mark things as dirty... Note that we *MUST* 589 have done a configure before this. I DO NOT WANT TO BE 590 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 591 THAT THEY BELONG TOGETHER!!!!! */ 592 /* XXX should check to see if we're only open for reading 593 here... If so, we needn't do this, but then need some 594 other way of keeping track of what's happened.. */ 595 596 rf_markalldirty( raidPtrs[unit] ); 597 } 598 599 600 rs->sc_dkdev.dk_openmask = 601 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 602 603 raidunlock(rs); 604 605 return (error); 606 607 608 } 609 /* ARGSUSED */ 610 int 611 raidclose(dev_t dev, int flags, int fmt, struct lwp *l) 612 { 613 int unit = raidunit(dev); 614 struct raid_softc *rs; 615 int error = 0; 616 int part; 617 618 if (unit >= numraid) 619 return (ENXIO); 620 rs = &raid_softc[unit]; 621 622 if ((error = raidlock(rs)) != 0) 623 return (error); 624 625 part = DISKPART(dev); 626 627 /* ...that much closer to allowing unconfiguration... */ 628 switch (fmt) { 629 case S_IFCHR: 630 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 631 break; 632 633 case S_IFBLK: 634 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 635 break; 636 } 637 rs->sc_dkdev.dk_openmask = 638 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 639 640 if ((rs->sc_dkdev.dk_openmask == 0) && 641 ((rs->sc_flags & RAIDF_INITED) != 0)) { 642 /* Last one... device is not unconfigured yet. 643 Device shutdown has taken care of setting the 644 clean bits if RAIDF_INITED is not set 645 mark things as clean... */ 646 647 rf_update_component_labels(raidPtrs[unit], 648 RF_FINAL_COMPONENT_UPDATE); 649 if (doing_shutdown) { 650 /* last one, and we're going down, so 651 lights out for this RAID set too. */ 652 error = rf_Shutdown(raidPtrs[unit]); 653 654 /* It's no longer initialized... */ 655 rs->sc_flags &= ~RAIDF_INITED; 656 657 /* Detach the disk. */ 658 pseudo_disk_detach(&rs->sc_dkdev); 659 } 660 } 661 662 raidunlock(rs); 663 return (0); 664 665 } 666 667 void 668 raidstrategy(struct buf *bp) 669 { 670 int s; 671 672 unsigned int raidID = raidunit(bp->b_dev); 673 RF_Raid_t *raidPtr; 674 struct raid_softc *rs = &raid_softc[raidID]; 675 int wlabel; 676 677 if ((rs->sc_flags & RAIDF_INITED) ==0) { 678 bp->b_error = ENXIO; 679 bp->b_flags |= B_ERROR; 680 goto done; 681 } 682 if (raidID >= numraid || !raidPtrs[raidID]) { 683 bp->b_error = ENODEV; 684 bp->b_flags |= B_ERROR; 685 goto done; 686 } 687 raidPtr = raidPtrs[raidID]; 688 if (!raidPtr->valid) { 689 bp->b_error = ENODEV; 690 bp->b_flags |= B_ERROR; 691 goto done; 692 } 693 if (bp->b_bcount == 0) { 694 db1_printf(("b_bcount is zero..\n")); 695 goto done; 696 } 697 698 /* 699 * Do bounds checking and adjust transfer. If there's an 700 * error, the bounds check will flag that for us. 701 */ 702 703 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 704 if (DISKPART(bp->b_dev) == RAW_PART) { 705 uint64_t size; /* device size in DEV_BSIZE unit */ 706 707 if (raidPtr->logBytesPerSector > DEV_BSHIFT) { 708 size = raidPtr->totalSectors << 709 (raidPtr->logBytesPerSector - DEV_BSHIFT); 710 } else { 711 size = raidPtr->totalSectors >> 712 (DEV_BSHIFT - raidPtr->logBytesPerSector); 713 } 714 if (bounds_check_with_mediasize(bp, DEV_BSIZE, size) <= 0) { 715 goto done; 716 } 717 } else { 718 if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) { 719 db1_printf(("Bounds check failed!!:%d %d\n", 720 (int) bp->b_blkno, (int) wlabel)); 721 goto done; 722 } 723 } 724 s = splbio(); 725 726 bp->b_resid = 0; 727 728 /* stuff it onto our queue */ 729 BUFQ_PUT(rs->buf_queue, bp); 730 731 /* scheduled the IO to happen at the next convenient time */ 732 wakeup(&(raidPtrs[raidID]->iodone)); 733 734 splx(s); 735 return; 736 737 done: 738 bp->b_resid = bp->b_bcount; 739 biodone(bp); 740 } 741 /* ARGSUSED */ 742 int 743 raidread(dev_t dev, struct uio *uio, int flags) 744 { 745 int unit = raidunit(dev); 746 struct raid_softc *rs; 747 748 if (unit >= numraid) 749 return (ENXIO); 750 rs = &raid_softc[unit]; 751 752 if ((rs->sc_flags & RAIDF_INITED) == 0) 753 return (ENXIO); 754 755 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio)); 756 757 } 758 /* ARGSUSED */ 759 int 760 raidwrite(dev_t dev, struct uio *uio, int flags) 761 { 762 int unit = raidunit(dev); 763 struct raid_softc *rs; 764 765 if (unit >= numraid) 766 return (ENXIO); 767 rs = &raid_softc[unit]; 768 769 if ((rs->sc_flags & RAIDF_INITED) == 0) 770 return (ENXIO); 771 772 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio)); 773 774 } 775 776 int 777 raidioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l) 778 { 779 int unit = raidunit(dev); 780 int error = 0; 781 int part, pmask; 782 struct raid_softc *rs; 783 RF_Config_t *k_cfg, *u_cfg; 784 RF_Raid_t *raidPtr; 785 RF_RaidDisk_t *diskPtr; 786 RF_AccTotals_t *totals; 787 RF_DeviceConfig_t *d_cfg, **ucfgp; 788 u_char *specific_buf; 789 int retcode = 0; 790 int column; 791 int raidid; 792 struct rf_recon_req *rrcopy, *rr; 793 RF_ComponentLabel_t *clabel; 794 RF_ComponentLabel_t ci_label; 795 RF_ComponentLabel_t **clabel_ptr; 796 RF_SingleComponent_t *sparePtr,*componentPtr; 797 RF_SingleComponent_t hot_spare; 798 RF_SingleComponent_t component; 799 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 800 int i, j, d; 801 #ifdef __HAVE_OLD_DISKLABEL 802 struct disklabel newlabel; 803 #endif 804 805 if (unit >= numraid) 806 return (ENXIO); 807 rs = &raid_softc[unit]; 808 raidPtr = raidPtrs[unit]; 809 810 db1_printf(("raidioctl: %d %d %d %d\n", (int) dev, 811 (int) DISKPART(dev), (int) unit, (int) cmd)); 812 813 /* Must be open for writes for these commands... */ 814 switch (cmd) { 815 case DIOCSDINFO: 816 case DIOCWDINFO: 817 #ifdef __HAVE_OLD_DISKLABEL 818 case ODIOCWDINFO: 819 case ODIOCSDINFO: 820 #endif 821 case DIOCWLABEL: 822 if ((flag & FWRITE) == 0) 823 return (EBADF); 824 } 825 826 /* Must be initialized for these... */ 827 switch (cmd) { 828 case DIOCGDINFO: 829 case DIOCSDINFO: 830 case DIOCWDINFO: 831 #ifdef __HAVE_OLD_DISKLABEL 832 case ODIOCGDINFO: 833 case ODIOCWDINFO: 834 case ODIOCSDINFO: 835 case ODIOCGDEFLABEL: 836 #endif 837 case DIOCGPART: 838 case DIOCWLABEL: 839 case DIOCGDEFLABEL: 840 case RAIDFRAME_SHUTDOWN: 841 case RAIDFRAME_REWRITEPARITY: 842 case RAIDFRAME_GET_INFO: 843 case RAIDFRAME_RESET_ACCTOTALS: 844 case RAIDFRAME_GET_ACCTOTALS: 845 case RAIDFRAME_KEEP_ACCTOTALS: 846 case RAIDFRAME_GET_SIZE: 847 case RAIDFRAME_FAIL_DISK: 848 case RAIDFRAME_COPYBACK: 849 case RAIDFRAME_CHECK_RECON_STATUS: 850 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 851 case RAIDFRAME_GET_COMPONENT_LABEL: 852 case RAIDFRAME_SET_COMPONENT_LABEL: 853 case RAIDFRAME_ADD_HOT_SPARE: 854 case RAIDFRAME_REMOVE_HOT_SPARE: 855 case RAIDFRAME_INIT_LABELS: 856 case RAIDFRAME_REBUILD_IN_PLACE: 857 case RAIDFRAME_CHECK_PARITY: 858 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 859 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 860 case RAIDFRAME_CHECK_COPYBACK_STATUS: 861 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 862 case RAIDFRAME_SET_AUTOCONFIG: 863 case RAIDFRAME_SET_ROOT: 864 case RAIDFRAME_DELETE_COMPONENT: 865 case RAIDFRAME_INCORPORATE_HOT_SPARE: 866 if ((rs->sc_flags & RAIDF_INITED) == 0) 867 return (ENXIO); 868 } 869 870 switch (cmd) { 871 872 /* configure the system */ 873 case RAIDFRAME_CONFIGURE: 874 875 if (raidPtr->valid) { 876 /* There is a valid RAID set running on this unit! */ 877 printf("raid%d: Device already configured!\n",unit); 878 return(EINVAL); 879 } 880 881 /* copy-in the configuration information */ 882 /* data points to a pointer to the configuration structure */ 883 884 u_cfg = *((RF_Config_t **) data); 885 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *)); 886 if (k_cfg == NULL) { 887 return (ENOMEM); 888 } 889 retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t)); 890 if (retcode) { 891 RF_Free(k_cfg, sizeof(RF_Config_t)); 892 db1_printf(("rf_ioctl: retcode=%d copyin.1\n", 893 retcode)); 894 return (retcode); 895 } 896 /* allocate a buffer for the layout-specific data, and copy it 897 * in */ 898 if (k_cfg->layoutSpecificSize) { 899 if (k_cfg->layoutSpecificSize > 10000) { 900 /* sanity check */ 901 RF_Free(k_cfg, sizeof(RF_Config_t)); 902 return (EINVAL); 903 } 904 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize, 905 (u_char *)); 906 if (specific_buf == NULL) { 907 RF_Free(k_cfg, sizeof(RF_Config_t)); 908 return (ENOMEM); 909 } 910 retcode = copyin(k_cfg->layoutSpecific, specific_buf, 911 k_cfg->layoutSpecificSize); 912 if (retcode) { 913 RF_Free(k_cfg, sizeof(RF_Config_t)); 914 RF_Free(specific_buf, 915 k_cfg->layoutSpecificSize); 916 db1_printf(("rf_ioctl: retcode=%d copyin.2\n", 917 retcode)); 918 return (retcode); 919 } 920 } else 921 specific_buf = NULL; 922 k_cfg->layoutSpecific = specific_buf; 923 924 /* should do some kind of sanity check on the configuration. 925 * Store the sum of all the bytes in the last byte? */ 926 927 /* configure the system */ 928 929 /* 930 * Clear the entire RAID descriptor, just to make sure 931 * there is no stale data left in the case of a 932 * reconfiguration 933 */ 934 memset((char *) raidPtr, 0, sizeof(RF_Raid_t)); 935 raidPtr->raidid = unit; 936 937 retcode = rf_Configure(raidPtr, k_cfg, NULL); 938 939 if (retcode == 0) { 940 941 /* allow this many simultaneous IO's to 942 this RAID device */ 943 raidPtr->openings = RAIDOUTSTANDING; 944 945 raidinit(raidPtr); 946 rf_markalldirty(raidPtr); 947 } 948 /* free the buffers. No return code here. */ 949 if (k_cfg->layoutSpecificSize) { 950 RF_Free(specific_buf, k_cfg->layoutSpecificSize); 951 } 952 RF_Free(k_cfg, sizeof(RF_Config_t)); 953 954 return (retcode); 955 956 /* shutdown the system */ 957 case RAIDFRAME_SHUTDOWN: 958 959 if ((error = raidlock(rs)) != 0) 960 return (error); 961 962 /* 963 * If somebody has a partition mounted, we shouldn't 964 * shutdown. 965 */ 966 967 part = DISKPART(dev); 968 pmask = (1 << part); 969 if ((rs->sc_dkdev.dk_openmask & ~pmask) || 970 ((rs->sc_dkdev.dk_bopenmask & pmask) && 971 (rs->sc_dkdev.dk_copenmask & pmask))) { 972 raidunlock(rs); 973 return (EBUSY); 974 } 975 976 retcode = rf_Shutdown(raidPtr); 977 978 /* It's no longer initialized... */ 979 rs->sc_flags &= ~RAIDF_INITED; 980 981 /* Detach the disk. */ 982 pseudo_disk_detach(&rs->sc_dkdev); 983 984 raidunlock(rs); 985 986 return (retcode); 987 case RAIDFRAME_GET_COMPONENT_LABEL: 988 clabel_ptr = (RF_ComponentLabel_t **) data; 989 /* need to read the component label for the disk indicated 990 by row,column in clabel */ 991 992 /* For practice, let's get it directly fromdisk, rather 993 than from the in-core copy */ 994 RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ), 995 (RF_ComponentLabel_t *)); 996 if (clabel == NULL) 997 return (ENOMEM); 998 999 retcode = copyin( *clabel_ptr, clabel, 1000 sizeof(RF_ComponentLabel_t)); 1001 1002 if (retcode) { 1003 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1004 return(retcode); 1005 } 1006 1007 clabel->row = 0; /* Don't allow looking at anything else.*/ 1008 1009 column = clabel->column; 1010 1011 if ((column < 0) || (column >= raidPtr->numCol + 1012 raidPtr->numSpare)) { 1013 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1014 return(EINVAL); 1015 } 1016 1017 retcode = raidread_component_label(raidPtr->Disks[column].dev, 1018 raidPtr->raid_cinfo[column].ci_vp, 1019 clabel ); 1020 1021 if (retcode == 0) { 1022 retcode = copyout(clabel, *clabel_ptr, 1023 sizeof(RF_ComponentLabel_t)); 1024 } 1025 RF_Free(clabel, sizeof(RF_ComponentLabel_t)); 1026 return (retcode); 1027 1028 case RAIDFRAME_SET_COMPONENT_LABEL: 1029 clabel = (RF_ComponentLabel_t *) data; 1030 1031 /* XXX check the label for valid stuff... */ 1032 /* Note that some things *should not* get modified -- 1033 the user should be re-initing the labels instead of 1034 trying to patch things. 1035 */ 1036 1037 raidid = raidPtr->raidid; 1038 #if DEBUG 1039 printf("raid%d: Got component label:\n", raidid); 1040 printf("raid%d: Version: %d\n", raidid, clabel->version); 1041 printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number); 1042 printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter); 1043 printf("raid%d: Column: %d\n", raidid, clabel->column); 1044 printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns); 1045 printf("raid%d: Clean: %d\n", raidid, clabel->clean); 1046 printf("raid%d: Status: %d\n", raidid, clabel->status); 1047 #endif 1048 clabel->row = 0; 1049 column = clabel->column; 1050 1051 if ((column < 0) || (column >= raidPtr->numCol)) { 1052 return(EINVAL); 1053 } 1054 1055 /* XXX this isn't allowed to do anything for now :-) */ 1056 1057 /* XXX and before it is, we need to fill in the rest 1058 of the fields!?!?!?! */ 1059 #if 0 1060 raidwrite_component_label( 1061 raidPtr->Disks[column].dev, 1062 raidPtr->raid_cinfo[column].ci_vp, 1063 clabel ); 1064 #endif 1065 return (0); 1066 1067 case RAIDFRAME_INIT_LABELS: 1068 clabel = (RF_ComponentLabel_t *) data; 1069 /* 1070 we only want the serial number from 1071 the above. We get all the rest of the information 1072 from the config that was used to create this RAID 1073 set. 1074 */ 1075 1076 raidPtr->serial_number = clabel->serial_number; 1077 1078 raid_init_component_label(raidPtr, &ci_label); 1079 ci_label.serial_number = clabel->serial_number; 1080 ci_label.row = 0; /* we dont' pretend to support more */ 1081 1082 for(column=0;column<raidPtr->numCol;column++) { 1083 diskPtr = &raidPtr->Disks[column]; 1084 if (!RF_DEAD_DISK(diskPtr->status)) { 1085 ci_label.partitionSize = diskPtr->partitionSize; 1086 ci_label.column = column; 1087 raidwrite_component_label( 1088 raidPtr->Disks[column].dev, 1089 raidPtr->raid_cinfo[column].ci_vp, 1090 &ci_label ); 1091 } 1092 } 1093 1094 return (retcode); 1095 case RAIDFRAME_SET_AUTOCONFIG: 1096 d = rf_set_autoconfig(raidPtr, *(int *) data); 1097 printf("raid%d: New autoconfig value is: %d\n", 1098 raidPtr->raidid, d); 1099 *(int *) data = d; 1100 return (retcode); 1101 1102 case RAIDFRAME_SET_ROOT: 1103 d = rf_set_rootpartition(raidPtr, *(int *) data); 1104 printf("raid%d: New rootpartition value is: %d\n", 1105 raidPtr->raidid, d); 1106 *(int *) data = d; 1107 return (retcode); 1108 1109 /* initialize all parity */ 1110 case RAIDFRAME_REWRITEPARITY: 1111 1112 if (raidPtr->Layout.map->faultsTolerated == 0) { 1113 /* Parity for RAID 0 is trivially correct */ 1114 raidPtr->parity_good = RF_RAID_CLEAN; 1115 return(0); 1116 } 1117 1118 if (raidPtr->parity_rewrite_in_progress == 1) { 1119 /* Re-write is already in progress! */ 1120 return(EINVAL); 1121 } 1122 1123 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1124 rf_RewriteParityThread, 1125 raidPtr,"raid_parity"); 1126 return (retcode); 1127 1128 1129 case RAIDFRAME_ADD_HOT_SPARE: 1130 sparePtr = (RF_SingleComponent_t *) data; 1131 memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t)); 1132 retcode = rf_add_hot_spare(raidPtr, &hot_spare); 1133 return(retcode); 1134 1135 case RAIDFRAME_REMOVE_HOT_SPARE: 1136 return(retcode); 1137 1138 case RAIDFRAME_DELETE_COMPONENT: 1139 componentPtr = (RF_SingleComponent_t *)data; 1140 memcpy( &component, componentPtr, 1141 sizeof(RF_SingleComponent_t)); 1142 retcode = rf_delete_component(raidPtr, &component); 1143 return(retcode); 1144 1145 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1146 componentPtr = (RF_SingleComponent_t *)data; 1147 memcpy( &component, componentPtr, 1148 sizeof(RF_SingleComponent_t)); 1149 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1150 return(retcode); 1151 1152 case RAIDFRAME_REBUILD_IN_PLACE: 1153 1154 if (raidPtr->Layout.map->faultsTolerated == 0) { 1155 /* Can't do this on a RAID 0!! */ 1156 return(EINVAL); 1157 } 1158 1159 if (raidPtr->recon_in_progress == 1) { 1160 /* a reconstruct is already in progress! */ 1161 return(EINVAL); 1162 } 1163 1164 componentPtr = (RF_SingleComponent_t *) data; 1165 memcpy( &component, componentPtr, 1166 sizeof(RF_SingleComponent_t)); 1167 component.row = 0; /* we don't support any more */ 1168 column = component.column; 1169 1170 if ((column < 0) || (column >= raidPtr->numCol)) { 1171 return(EINVAL); 1172 } 1173 1174 RF_LOCK_MUTEX(raidPtr->mutex); 1175 if ((raidPtr->Disks[column].status == rf_ds_optimal) && 1176 (raidPtr->numFailures > 0)) { 1177 /* XXX 0 above shouldn't be constant!!! */ 1178 /* some component other than this has failed. 1179 Let's not make things worse than they already 1180 are... */ 1181 printf("raid%d: Unable to reconstruct to disk at:\n", 1182 raidPtr->raidid); 1183 printf("raid%d: Col: %d Too many failures.\n", 1184 raidPtr->raidid, column); 1185 RF_UNLOCK_MUTEX(raidPtr->mutex); 1186 return (EINVAL); 1187 } 1188 if (raidPtr->Disks[column].status == 1189 rf_ds_reconstructing) { 1190 printf("raid%d: Unable to reconstruct to disk at:\n", 1191 raidPtr->raidid); 1192 printf("raid%d: Col: %d Reconstruction already occuring!\n", raidPtr->raidid, column); 1193 1194 RF_UNLOCK_MUTEX(raidPtr->mutex); 1195 return (EINVAL); 1196 } 1197 if (raidPtr->Disks[column].status == rf_ds_spared) { 1198 RF_UNLOCK_MUTEX(raidPtr->mutex); 1199 return (EINVAL); 1200 } 1201 RF_UNLOCK_MUTEX(raidPtr->mutex); 1202 1203 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1204 if (rrcopy == NULL) 1205 return(ENOMEM); 1206 1207 rrcopy->raidPtr = (void *) raidPtr; 1208 rrcopy->col = column; 1209 1210 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1211 rf_ReconstructInPlaceThread, 1212 rrcopy,"raid_reconip"); 1213 return(retcode); 1214 1215 case RAIDFRAME_GET_INFO: 1216 if (!raidPtr->valid) 1217 return (ENODEV); 1218 ucfgp = (RF_DeviceConfig_t **) data; 1219 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1220 (RF_DeviceConfig_t *)); 1221 if (d_cfg == NULL) 1222 return (ENOMEM); 1223 d_cfg->rows = 1; /* there is only 1 row now */ 1224 d_cfg->cols = raidPtr->numCol; 1225 d_cfg->ndevs = raidPtr->numCol; 1226 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1227 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1228 return (ENOMEM); 1229 } 1230 d_cfg->nspares = raidPtr->numSpare; 1231 if (d_cfg->nspares >= RF_MAX_DISKS) { 1232 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1233 return (ENOMEM); 1234 } 1235 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1236 d = 0; 1237 for (j = 0; j < d_cfg->cols; j++) { 1238 d_cfg->devs[d] = raidPtr->Disks[j]; 1239 d++; 1240 } 1241 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1242 d_cfg->spares[i] = raidPtr->Disks[j]; 1243 } 1244 retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t)); 1245 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1246 1247 return (retcode); 1248 1249 case RAIDFRAME_CHECK_PARITY: 1250 *(int *) data = raidPtr->parity_good; 1251 return (0); 1252 1253 case RAIDFRAME_RESET_ACCTOTALS: 1254 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1255 return (0); 1256 1257 case RAIDFRAME_GET_ACCTOTALS: 1258 totals = (RF_AccTotals_t *) data; 1259 *totals = raidPtr->acc_totals; 1260 return (0); 1261 1262 case RAIDFRAME_KEEP_ACCTOTALS: 1263 raidPtr->keep_acc_totals = *(int *)data; 1264 return (0); 1265 1266 case RAIDFRAME_GET_SIZE: 1267 *(int *) data = raidPtr->totalSectors; 1268 return (0); 1269 1270 /* fail a disk & optionally start reconstruction */ 1271 case RAIDFRAME_FAIL_DISK: 1272 1273 if (raidPtr->Layout.map->faultsTolerated == 0) { 1274 /* Can't do this on a RAID 0!! */ 1275 return(EINVAL); 1276 } 1277 1278 rr = (struct rf_recon_req *) data; 1279 rr->row = 0; 1280 if (rr->col < 0 || rr->col >= raidPtr->numCol) 1281 return (EINVAL); 1282 1283 1284 RF_LOCK_MUTEX(raidPtr->mutex); 1285 if (raidPtr->status == rf_rs_reconstructing) { 1286 /* you can't fail a disk while we're reconstructing! */ 1287 /* XXX wrong for RAID6 */ 1288 RF_UNLOCK_MUTEX(raidPtr->mutex); 1289 return (EINVAL); 1290 } 1291 if ((raidPtr->Disks[rr->col].status == 1292 rf_ds_optimal) && (raidPtr->numFailures > 0)) { 1293 /* some other component has failed. Let's not make 1294 things worse. XXX wrong for RAID6 */ 1295 RF_UNLOCK_MUTEX(raidPtr->mutex); 1296 return (EINVAL); 1297 } 1298 if (raidPtr->Disks[rr->col].status == rf_ds_spared) { 1299 /* Can't fail a spared disk! */ 1300 RF_UNLOCK_MUTEX(raidPtr->mutex); 1301 return (EINVAL); 1302 } 1303 RF_UNLOCK_MUTEX(raidPtr->mutex); 1304 1305 /* make a copy of the recon request so that we don't rely on 1306 * the user's buffer */ 1307 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1308 if (rrcopy == NULL) 1309 return(ENOMEM); 1310 memcpy(rrcopy, rr, sizeof(*rr)); 1311 rrcopy->raidPtr = (void *) raidPtr; 1312 1313 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1314 rf_ReconThread, 1315 rrcopy,"raid_recon"); 1316 return (0); 1317 1318 /* invoke a copyback operation after recon on whatever disk 1319 * needs it, if any */ 1320 case RAIDFRAME_COPYBACK: 1321 1322 if (raidPtr->Layout.map->faultsTolerated == 0) { 1323 /* This makes no sense on a RAID 0!! */ 1324 return(EINVAL); 1325 } 1326 1327 if (raidPtr->copyback_in_progress == 1) { 1328 /* Copyback is already in progress! */ 1329 return(EINVAL); 1330 } 1331 1332 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1333 rf_CopybackThread, 1334 raidPtr,"raid_copyback"); 1335 return (retcode); 1336 1337 /* return the percentage completion of reconstruction */ 1338 case RAIDFRAME_CHECK_RECON_STATUS: 1339 if (raidPtr->Layout.map->faultsTolerated == 0) { 1340 /* This makes no sense on a RAID 0, so tell the 1341 user it's done. */ 1342 *(int *) data = 100; 1343 return(0); 1344 } 1345 if (raidPtr->status != rf_rs_reconstructing) 1346 *(int *) data = 100; 1347 else { 1348 if (raidPtr->reconControl->numRUsTotal > 0) { 1349 *(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal); 1350 } else { 1351 *(int *) data = 0; 1352 } 1353 } 1354 return (0); 1355 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1356 progressInfoPtr = (RF_ProgressInfo_t **) data; 1357 if (raidPtr->status != rf_rs_reconstructing) { 1358 progressInfo.remaining = 0; 1359 progressInfo.completed = 100; 1360 progressInfo.total = 100; 1361 } else { 1362 progressInfo.total = 1363 raidPtr->reconControl->numRUsTotal; 1364 progressInfo.completed = 1365 raidPtr->reconControl->numRUsComplete; 1366 progressInfo.remaining = progressInfo.total - 1367 progressInfo.completed; 1368 } 1369 retcode = copyout(&progressInfo, *progressInfoPtr, 1370 sizeof(RF_ProgressInfo_t)); 1371 return (retcode); 1372 1373 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1374 if (raidPtr->Layout.map->faultsTolerated == 0) { 1375 /* This makes no sense on a RAID 0, so tell the 1376 user it's done. */ 1377 *(int *) data = 100; 1378 return(0); 1379 } 1380 if (raidPtr->parity_rewrite_in_progress == 1) { 1381 *(int *) data = 100 * 1382 raidPtr->parity_rewrite_stripes_done / 1383 raidPtr->Layout.numStripe; 1384 } else { 1385 *(int *) data = 100; 1386 } 1387 return (0); 1388 1389 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1390 progressInfoPtr = (RF_ProgressInfo_t **) data; 1391 if (raidPtr->parity_rewrite_in_progress == 1) { 1392 progressInfo.total = raidPtr->Layout.numStripe; 1393 progressInfo.completed = 1394 raidPtr->parity_rewrite_stripes_done; 1395 progressInfo.remaining = progressInfo.total - 1396 progressInfo.completed; 1397 } else { 1398 progressInfo.remaining = 0; 1399 progressInfo.completed = 100; 1400 progressInfo.total = 100; 1401 } 1402 retcode = copyout(&progressInfo, *progressInfoPtr, 1403 sizeof(RF_ProgressInfo_t)); 1404 return (retcode); 1405 1406 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1407 if (raidPtr->Layout.map->faultsTolerated == 0) { 1408 /* This makes no sense on a RAID 0 */ 1409 *(int *) data = 100; 1410 return(0); 1411 } 1412 if (raidPtr->copyback_in_progress == 1) { 1413 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1414 raidPtr->Layout.numStripe; 1415 } else { 1416 *(int *) data = 100; 1417 } 1418 return (0); 1419 1420 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1421 progressInfoPtr = (RF_ProgressInfo_t **) data; 1422 if (raidPtr->copyback_in_progress == 1) { 1423 progressInfo.total = raidPtr->Layout.numStripe; 1424 progressInfo.completed = 1425 raidPtr->copyback_stripes_done; 1426 progressInfo.remaining = progressInfo.total - 1427 progressInfo.completed; 1428 } else { 1429 progressInfo.remaining = 0; 1430 progressInfo.completed = 100; 1431 progressInfo.total = 100; 1432 } 1433 retcode = copyout(&progressInfo, *progressInfoPtr, 1434 sizeof(RF_ProgressInfo_t)); 1435 return (retcode); 1436 1437 /* the sparetable daemon calls this to wait for the kernel to 1438 * need a spare table. this ioctl does not return until a 1439 * spare table is needed. XXX -- calling mpsleep here in the 1440 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1441 * -- I should either compute the spare table in the kernel, 1442 * or have a different -- XXX XXX -- interface (a different 1443 * character device) for delivering the table -- XXX */ 1444 #if 0 1445 case RAIDFRAME_SPARET_WAIT: 1446 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1447 while (!rf_sparet_wait_queue) 1448 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE); 1449 waitreq = rf_sparet_wait_queue; 1450 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1451 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1452 1453 /* structure assignment */ 1454 *((RF_SparetWait_t *) data) = *waitreq; 1455 1456 RF_Free(waitreq, sizeof(*waitreq)); 1457 return (0); 1458 1459 /* wakes up a process waiting on SPARET_WAIT and puts an error 1460 * code in it that will cause the dameon to exit */ 1461 case RAIDFRAME_ABORT_SPARET_WAIT: 1462 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1463 waitreq->fcol = -1; 1464 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1465 waitreq->next = rf_sparet_wait_queue; 1466 rf_sparet_wait_queue = waitreq; 1467 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1468 wakeup(&rf_sparet_wait_queue); 1469 return (0); 1470 1471 /* used by the spare table daemon to deliver a spare table 1472 * into the kernel */ 1473 case RAIDFRAME_SEND_SPARET: 1474 1475 /* install the spare table */ 1476 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1477 1478 /* respond to the requestor. the return status of the spare 1479 * table installation is passed in the "fcol" field */ 1480 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1481 waitreq->fcol = retcode; 1482 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1483 waitreq->next = rf_sparet_resp_queue; 1484 rf_sparet_resp_queue = waitreq; 1485 wakeup(&rf_sparet_resp_queue); 1486 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1487 1488 return (retcode); 1489 #endif 1490 1491 default: 1492 break; /* fall through to the os-specific code below */ 1493 1494 } 1495 1496 if (!raidPtr->valid) 1497 return (EINVAL); 1498 1499 /* 1500 * Add support for "regular" device ioctls here. 1501 */ 1502 1503 switch (cmd) { 1504 case DIOCGDINFO: 1505 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1506 break; 1507 #ifdef __HAVE_OLD_DISKLABEL 1508 case ODIOCGDINFO: 1509 newlabel = *(rs->sc_dkdev.dk_label); 1510 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1511 return ENOTTY; 1512 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1513 break; 1514 #endif 1515 1516 case DIOCGPART: 1517 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1518 ((struct partinfo *) data)->part = 1519 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1520 break; 1521 1522 case DIOCWDINFO: 1523 case DIOCSDINFO: 1524 #ifdef __HAVE_OLD_DISKLABEL 1525 case ODIOCWDINFO: 1526 case ODIOCSDINFO: 1527 #endif 1528 { 1529 struct disklabel *lp; 1530 #ifdef __HAVE_OLD_DISKLABEL 1531 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1532 memset(&newlabel, 0, sizeof newlabel); 1533 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1534 lp = &newlabel; 1535 } else 1536 #endif 1537 lp = (struct disklabel *)data; 1538 1539 if ((error = raidlock(rs)) != 0) 1540 return (error); 1541 1542 rs->sc_flags |= RAIDF_LABELLING; 1543 1544 error = setdisklabel(rs->sc_dkdev.dk_label, 1545 lp, 0, rs->sc_dkdev.dk_cpulabel); 1546 if (error == 0) { 1547 if (cmd == DIOCWDINFO 1548 #ifdef __HAVE_OLD_DISKLABEL 1549 || cmd == ODIOCWDINFO 1550 #endif 1551 ) 1552 error = writedisklabel(RAIDLABELDEV(dev), 1553 raidstrategy, rs->sc_dkdev.dk_label, 1554 rs->sc_dkdev.dk_cpulabel); 1555 } 1556 rs->sc_flags &= ~RAIDF_LABELLING; 1557 1558 raidunlock(rs); 1559 1560 if (error) 1561 return (error); 1562 break; 1563 } 1564 1565 case DIOCWLABEL: 1566 if (*(int *) data != 0) 1567 rs->sc_flags |= RAIDF_WLABEL; 1568 else 1569 rs->sc_flags &= ~RAIDF_WLABEL; 1570 break; 1571 1572 case DIOCGDEFLABEL: 1573 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1574 break; 1575 1576 #ifdef __HAVE_OLD_DISKLABEL 1577 case ODIOCGDEFLABEL: 1578 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1579 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1580 return ENOTTY; 1581 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1582 break; 1583 #endif 1584 1585 default: 1586 retcode = ENOTTY; 1587 } 1588 return (retcode); 1589 1590 } 1591 1592 1593 /* raidinit -- complete the rest of the initialization for the 1594 RAIDframe device. */ 1595 1596 1597 static void 1598 raidinit(RF_Raid_t *raidPtr) 1599 { 1600 struct raid_softc *rs; 1601 int unit; 1602 1603 unit = raidPtr->raidid; 1604 1605 rs = &raid_softc[unit]; 1606 1607 /* XXX should check return code first... */ 1608 rs->sc_flags |= RAIDF_INITED; 1609 1610 /* XXX doesn't check bounds. */ 1611 snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit); 1612 1613 rs->sc_dkdev.dk_name = rs->sc_xname; 1614 1615 /* disk_attach actually creates space for the CPU disklabel, among 1616 * other things, so it's critical to call this *BEFORE* we try putzing 1617 * with disklabels. */ 1618 1619 pseudo_disk_attach(&rs->sc_dkdev); 1620 1621 /* XXX There may be a weird interaction here between this, and 1622 * protectedSectors, as used in RAIDframe. */ 1623 1624 rs->sc_size = raidPtr->totalSectors; 1625 } 1626 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 1627 /* wake up the daemon & tell it to get us a spare table 1628 * XXX 1629 * the entries in the queues should be tagged with the raidPtr 1630 * so that in the extremely rare case that two recons happen at once, 1631 * we know for which device were requesting a spare table 1632 * XXX 1633 * 1634 * XXX This code is not currently used. GO 1635 */ 1636 int 1637 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req) 1638 { 1639 int retcode; 1640 1641 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1642 req->next = rf_sparet_wait_queue; 1643 rf_sparet_wait_queue = req; 1644 wakeup(&rf_sparet_wait_queue); 1645 1646 /* mpsleep unlocks the mutex */ 1647 while (!rf_sparet_resp_queue) { 1648 tsleep(&rf_sparet_resp_queue, PRIBIO, 1649 "raidframe getsparetable", 0); 1650 } 1651 req = rf_sparet_resp_queue; 1652 rf_sparet_resp_queue = req->next; 1653 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1654 1655 retcode = req->fcol; 1656 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1657 * alloc'd */ 1658 return (retcode); 1659 } 1660 #endif 1661 1662 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1663 * bp & passes it down. 1664 * any calls originating in the kernel must use non-blocking I/O 1665 * do some extra sanity checking to return "appropriate" error values for 1666 * certain conditions (to make some standard utilities work) 1667 * 1668 * Formerly known as: rf_DoAccessKernel 1669 */ 1670 void 1671 raidstart(RF_Raid_t *raidPtr) 1672 { 1673 RF_SectorCount_t num_blocks, pb, sum; 1674 RF_RaidAddr_t raid_addr; 1675 struct partition *pp; 1676 daddr_t blocknum; 1677 int unit; 1678 struct raid_softc *rs; 1679 int do_async; 1680 struct buf *bp; 1681 int rc; 1682 1683 unit = raidPtr->raidid; 1684 rs = &raid_softc[unit]; 1685 1686 /* quick check to see if anything has died recently */ 1687 RF_LOCK_MUTEX(raidPtr->mutex); 1688 if (raidPtr->numNewFailures > 0) { 1689 RF_UNLOCK_MUTEX(raidPtr->mutex); 1690 rf_update_component_labels(raidPtr, 1691 RF_NORMAL_COMPONENT_UPDATE); 1692 RF_LOCK_MUTEX(raidPtr->mutex); 1693 raidPtr->numNewFailures--; 1694 } 1695 1696 /* Check to see if we're at the limit... */ 1697 while (raidPtr->openings > 0) { 1698 RF_UNLOCK_MUTEX(raidPtr->mutex); 1699 1700 /* get the next item, if any, from the queue */ 1701 if ((bp = BUFQ_GET(rs->buf_queue)) == NULL) { 1702 /* nothing more to do */ 1703 return; 1704 } 1705 1706 /* Ok, for the bp we have here, bp->b_blkno is relative to the 1707 * partition.. Need to make it absolute to the underlying 1708 * device.. */ 1709 1710 blocknum = bp->b_blkno; 1711 if (DISKPART(bp->b_dev) != RAW_PART) { 1712 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 1713 blocknum += pp->p_offset; 1714 } 1715 1716 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 1717 (int) blocknum)); 1718 1719 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 1720 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 1721 1722 /* *THIS* is where we adjust what block we're going to... 1723 * but DO NOT TOUCH bp->b_blkno!!! */ 1724 raid_addr = blocknum; 1725 1726 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 1727 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 1728 sum = raid_addr + num_blocks + pb; 1729 if (1 || rf_debugKernelAccess) { 1730 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 1731 (int) raid_addr, (int) sum, (int) num_blocks, 1732 (int) pb, (int) bp->b_resid)); 1733 } 1734 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 1735 || (sum < num_blocks) || (sum < pb)) { 1736 bp->b_error = ENOSPC; 1737 bp->b_flags |= B_ERROR; 1738 bp->b_resid = bp->b_bcount; 1739 biodone(bp); 1740 RF_LOCK_MUTEX(raidPtr->mutex); 1741 continue; 1742 } 1743 /* 1744 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 1745 */ 1746 1747 if (bp->b_bcount & raidPtr->sectorMask) { 1748 bp->b_error = EINVAL; 1749 bp->b_flags |= B_ERROR; 1750 bp->b_resid = bp->b_bcount; 1751 biodone(bp); 1752 RF_LOCK_MUTEX(raidPtr->mutex); 1753 continue; 1754 1755 } 1756 db1_printf(("Calling DoAccess..\n")); 1757 1758 1759 RF_LOCK_MUTEX(raidPtr->mutex); 1760 raidPtr->openings--; 1761 RF_UNLOCK_MUTEX(raidPtr->mutex); 1762 1763 /* 1764 * Everything is async. 1765 */ 1766 do_async = 1; 1767 1768 disk_busy(&rs->sc_dkdev); 1769 1770 /* XXX we're still at splbio() here... do we *really* 1771 need to be? */ 1772 1773 /* don't ever condition on bp->b_flags & B_WRITE. 1774 * always condition on B_READ instead */ 1775 1776 rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 1777 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 1778 do_async, raid_addr, num_blocks, 1779 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 1780 1781 if (rc) { 1782 bp->b_error = rc; 1783 bp->b_flags |= B_ERROR; 1784 bp->b_resid = bp->b_bcount; 1785 biodone(bp); 1786 /* continue loop */ 1787 } 1788 1789 RF_LOCK_MUTEX(raidPtr->mutex); 1790 } 1791 RF_UNLOCK_MUTEX(raidPtr->mutex); 1792 } 1793 1794 1795 1796 1797 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 1798 1799 int 1800 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req) 1801 { 1802 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 1803 struct buf *bp; 1804 1805 req->queue = queue; 1806 1807 #if DIAGNOSTIC 1808 if (queue->raidPtr->raidid >= numraid) { 1809 printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid, 1810 numraid); 1811 panic("Invalid Unit number in rf_DispatchKernelIO"); 1812 } 1813 #endif 1814 1815 bp = req->bp; 1816 1817 switch (req->type) { 1818 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 1819 /* XXX need to do something extra here.. */ 1820 /* I'm leaving this in, as I've never actually seen it used, 1821 * and I'd like folks to report it... GO */ 1822 printf(("WAKEUP CALLED\n")); 1823 queue->numOutstanding++; 1824 1825 bp->b_flags = 0; 1826 bp->b_fspriv.bf_private = req; 1827 1828 KernelWakeupFunc(bp); 1829 break; 1830 1831 case RF_IO_TYPE_READ: 1832 case RF_IO_TYPE_WRITE: 1833 #if RF_ACC_TRACE > 0 1834 if (req->tracerec) { 1835 RF_ETIMER_START(req->tracerec->timer); 1836 } 1837 #endif 1838 InitBP(bp, queue->rf_cinfo->ci_vp, 1839 op, queue->rf_cinfo->ci_dev, 1840 req->sectorOffset, req->numSector, 1841 req->buf, KernelWakeupFunc, (void *) req, 1842 queue->raidPtr->logBytesPerSector, req->b_proc); 1843 1844 if (rf_debugKernelAccess) { 1845 db1_printf(("dispatch: bp->b_blkno = %ld\n", 1846 (long) bp->b_blkno)); 1847 } 1848 queue->numOutstanding++; 1849 queue->last_deq_sector = req->sectorOffset; 1850 /* acc wouldn't have been let in if there were any pending 1851 * reqs at any other priority */ 1852 queue->curPriority = req->priority; 1853 1854 db1_printf(("Going for %c to unit %d col %d\n", 1855 req->type, queue->raidPtr->raidid, 1856 queue->col)); 1857 db1_printf(("sector %d count %d (%d bytes) %d\n", 1858 (int) req->sectorOffset, (int) req->numSector, 1859 (int) (req->numSector << 1860 queue->raidPtr->logBytesPerSector), 1861 (int) queue->raidPtr->logBytesPerSector)); 1862 VOP_STRATEGY(bp->b_vp, bp); 1863 1864 break; 1865 1866 default: 1867 panic("bad req->type in rf_DispatchKernelIO"); 1868 } 1869 db1_printf(("Exiting from DispatchKernelIO\n")); 1870 1871 return (0); 1872 } 1873 /* this is the callback function associated with a I/O invoked from 1874 kernel code. 1875 */ 1876 static void 1877 KernelWakeupFunc(struct buf *bp) 1878 { 1879 RF_DiskQueueData_t *req = NULL; 1880 RF_DiskQueue_t *queue; 1881 int s; 1882 1883 s = splbio(); 1884 db1_printf(("recovering the request queue:\n")); 1885 req = bp->b_fspriv.bf_private; 1886 1887 queue = (RF_DiskQueue_t *) req->queue; 1888 1889 #if RF_ACC_TRACE > 0 1890 if (req->tracerec) { 1891 RF_ETIMER_STOP(req->tracerec->timer); 1892 RF_ETIMER_EVAL(req->tracerec->timer); 1893 RF_LOCK_MUTEX(rf_tracing_mutex); 1894 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 1895 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 1896 req->tracerec->num_phys_ios++; 1897 RF_UNLOCK_MUTEX(rf_tracing_mutex); 1898 } 1899 #endif 1900 1901 /* XXX Ok, let's get aggressive... If B_ERROR is set, let's go 1902 * ballistic, and mark the component as hosed... */ 1903 1904 if (bp->b_flags & B_ERROR) { 1905 /* Mark the disk as dead */ 1906 /* but only mark it once... */ 1907 /* and only if it wouldn't leave this RAID set 1908 completely broken */ 1909 if (((queue->raidPtr->Disks[queue->col].status == 1910 rf_ds_optimal) || 1911 (queue->raidPtr->Disks[queue->col].status == 1912 rf_ds_used_spare)) && 1913 (queue->raidPtr->numFailures < 1914 queue->raidPtr->Layout.map->faultsTolerated)) { 1915 printf("raid%d: IO Error. Marking %s as failed.\n", 1916 queue->raidPtr->raidid, 1917 queue->raidPtr->Disks[queue->col].devname); 1918 queue->raidPtr->Disks[queue->col].status = 1919 rf_ds_failed; 1920 queue->raidPtr->status = rf_rs_degraded; 1921 queue->raidPtr->numFailures++; 1922 queue->raidPtr->numNewFailures++; 1923 } else { /* Disk is already dead... */ 1924 /* printf("Disk already marked as dead!\n"); */ 1925 } 1926 1927 } 1928 1929 /* Fill in the error value */ 1930 1931 req->error = (bp->b_flags & B_ERROR) ? bp->b_error : 0; 1932 1933 simple_lock(&queue->raidPtr->iodone_lock); 1934 1935 /* Drop this one on the "finished" queue... */ 1936 TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries); 1937 1938 /* Let the raidio thread know there is work to be done. */ 1939 wakeup(&(queue->raidPtr->iodone)); 1940 1941 simple_unlock(&queue->raidPtr->iodone_lock); 1942 1943 splx(s); 1944 } 1945 1946 1947 1948 /* 1949 * initialize a buf structure for doing an I/O in the kernel. 1950 */ 1951 static void 1952 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev, 1953 RF_SectorNum_t startSect, RF_SectorCount_t numSect, caddr_t bf, 1954 void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector, 1955 struct proc *b_proc) 1956 { 1957 /* bp->b_flags = B_PHYS | rw_flag; */ 1958 bp->b_flags = B_CALL | rw_flag; /* XXX need B_PHYS here too??? */ 1959 bp->b_bcount = numSect << logBytesPerSector; 1960 bp->b_bufsize = bp->b_bcount; 1961 bp->b_error = 0; 1962 bp->b_dev = dev; 1963 bp->b_data = bf; 1964 bp->b_blkno = startSect; 1965 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 1966 if (bp->b_bcount == 0) { 1967 panic("bp->b_bcount is zero in InitBP!!"); 1968 } 1969 bp->b_proc = b_proc; 1970 bp->b_iodone = cbFunc; 1971 bp->b_fspriv.bf_private = cbArg; 1972 bp->b_vp = b_vp; 1973 if ((bp->b_flags & B_READ) == 0) { 1974 bp->b_vp->v_numoutput++; 1975 } 1976 1977 } 1978 1979 static void 1980 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs, 1981 struct disklabel *lp) 1982 { 1983 memset(lp, 0, sizeof(*lp)); 1984 1985 /* fabricate a label... */ 1986 lp->d_secperunit = raidPtr->totalSectors; 1987 lp->d_secsize = raidPtr->bytesPerSector; 1988 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 1989 lp->d_ntracks = 4 * raidPtr->numCol; 1990 lp->d_ncylinders = raidPtr->totalSectors / 1991 (lp->d_nsectors * lp->d_ntracks); 1992 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1993 1994 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 1995 lp->d_type = DTYPE_RAID; 1996 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1997 lp->d_rpm = 3600; 1998 lp->d_interleave = 1; 1999 lp->d_flags = 0; 2000 2001 lp->d_partitions[RAW_PART].p_offset = 0; 2002 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 2003 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 2004 lp->d_npartitions = RAW_PART + 1; 2005 2006 lp->d_magic = DISKMAGIC; 2007 lp->d_magic2 = DISKMAGIC; 2008 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 2009 2010 } 2011 /* 2012 * Read the disklabel from the raid device. If one is not present, fake one 2013 * up. 2014 */ 2015 static void 2016 raidgetdisklabel(dev_t dev) 2017 { 2018 int unit = raidunit(dev); 2019 struct raid_softc *rs = &raid_softc[unit]; 2020 const char *errstring; 2021 struct disklabel *lp = rs->sc_dkdev.dk_label; 2022 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 2023 RF_Raid_t *raidPtr; 2024 2025 db1_printf(("Getting the disklabel...\n")); 2026 2027 memset(clp, 0, sizeof(*clp)); 2028 2029 raidPtr = raidPtrs[unit]; 2030 2031 raidgetdefaultlabel(raidPtr, rs, lp); 2032 2033 /* 2034 * Call the generic disklabel extraction routine. 2035 */ 2036 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 2037 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2038 if (errstring) 2039 raidmakedisklabel(rs); 2040 else { 2041 int i; 2042 struct partition *pp; 2043 2044 /* 2045 * Sanity check whether the found disklabel is valid. 2046 * 2047 * This is necessary since total size of the raid device 2048 * may vary when an interleave is changed even though exactly 2049 * same componets are used, and old disklabel may used 2050 * if that is found. 2051 */ 2052 if (lp->d_secperunit != rs->sc_size) 2053 printf("raid%d: WARNING: %s: " 2054 "total sector size in disklabel (%d) != " 2055 "the size of raid (%ld)\n", unit, rs->sc_xname, 2056 lp->d_secperunit, (long) rs->sc_size); 2057 for (i = 0; i < lp->d_npartitions; i++) { 2058 pp = &lp->d_partitions[i]; 2059 if (pp->p_offset + pp->p_size > rs->sc_size) 2060 printf("raid%d: WARNING: %s: end of partition `%c' " 2061 "exceeds the size of raid (%ld)\n", 2062 unit, rs->sc_xname, 'a' + i, (long) rs->sc_size); 2063 } 2064 } 2065 2066 } 2067 /* 2068 * Take care of things one might want to take care of in the event 2069 * that a disklabel isn't present. 2070 */ 2071 static void 2072 raidmakedisklabel(struct raid_softc *rs) 2073 { 2074 struct disklabel *lp = rs->sc_dkdev.dk_label; 2075 db1_printf(("Making a label..\n")); 2076 2077 /* 2078 * For historical reasons, if there's no disklabel present 2079 * the raw partition must be marked FS_BSDFFS. 2080 */ 2081 2082 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2083 2084 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2085 2086 lp->d_checksum = dkcksum(lp); 2087 } 2088 /* 2089 * Lookup the provided name in the filesystem. If the file exists, 2090 * is a valid block device, and isn't being used by anyone else, 2091 * set *vpp to the file's vnode. 2092 * You'll find the original of this in ccd.c 2093 */ 2094 int 2095 raidlookup(char *path, struct lwp *l, struct vnode **vpp) 2096 { 2097 struct nameidata nd; 2098 struct vnode *vp; 2099 struct proc *p; 2100 struct vattr va; 2101 int error; 2102 2103 p = l ? l->l_proc : NULL; 2104 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, l); 2105 if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) { 2106 return (error); 2107 } 2108 vp = nd.ni_vp; 2109 if (vp->v_usecount > 1) { 2110 VOP_UNLOCK(vp, 0); 2111 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l); 2112 return (EBUSY); 2113 } 2114 if ((error = VOP_GETATTR(vp, &va, p->p_ucred, l)) != 0) { 2115 VOP_UNLOCK(vp, 0); 2116 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l); 2117 return (error); 2118 } 2119 /* XXX: eventually we should handle VREG, too. */ 2120 if (va.va_type != VBLK) { 2121 VOP_UNLOCK(vp, 0); 2122 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l); 2123 return (ENOTBLK); 2124 } 2125 VOP_UNLOCK(vp, 0); 2126 *vpp = vp; 2127 return (0); 2128 } 2129 /* 2130 * Wait interruptibly for an exclusive lock. 2131 * 2132 * XXX 2133 * Several drivers do this; it should be abstracted and made MP-safe. 2134 * (Hmm... where have we seen this warning before :-> GO ) 2135 */ 2136 static int 2137 raidlock(struct raid_softc *rs) 2138 { 2139 int error; 2140 2141 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2142 rs->sc_flags |= RAIDF_WANTED; 2143 if ((error = 2144 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2145 return (error); 2146 } 2147 rs->sc_flags |= RAIDF_LOCKED; 2148 return (0); 2149 } 2150 /* 2151 * Unlock and wake up any waiters. 2152 */ 2153 static void 2154 raidunlock(struct raid_softc *rs) 2155 { 2156 2157 rs->sc_flags &= ~RAIDF_LOCKED; 2158 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2159 rs->sc_flags &= ~RAIDF_WANTED; 2160 wakeup(rs); 2161 } 2162 } 2163 2164 2165 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2166 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2167 2168 int 2169 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter) 2170 { 2171 RF_ComponentLabel_t clabel; 2172 raidread_component_label(dev, b_vp, &clabel); 2173 clabel.mod_counter = mod_counter; 2174 clabel.clean = RF_RAID_CLEAN; 2175 raidwrite_component_label(dev, b_vp, &clabel); 2176 return(0); 2177 } 2178 2179 2180 int 2181 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter) 2182 { 2183 RF_ComponentLabel_t clabel; 2184 raidread_component_label(dev, b_vp, &clabel); 2185 clabel.mod_counter = mod_counter; 2186 clabel.clean = RF_RAID_DIRTY; 2187 raidwrite_component_label(dev, b_vp, &clabel); 2188 return(0); 2189 } 2190 2191 /* ARGSUSED */ 2192 int 2193 raidread_component_label(dev_t dev, struct vnode *b_vp, 2194 RF_ComponentLabel_t *clabel) 2195 { 2196 struct buf *bp; 2197 const struct bdevsw *bdev; 2198 int error; 2199 2200 /* XXX should probably ensure that we don't try to do this if 2201 someone has changed rf_protected_sectors. */ 2202 2203 if (b_vp == NULL) { 2204 /* For whatever reason, this component is not valid. 2205 Don't try to read a component label from it. */ 2206 return(EINVAL); 2207 } 2208 2209 /* get a block of the appropriate size... */ 2210 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2211 bp->b_dev = dev; 2212 2213 /* get our ducks in a row for the read */ 2214 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2215 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2216 bp->b_flags |= B_READ; 2217 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2218 2219 bdev = bdevsw_lookup(bp->b_dev); 2220 if (bdev == NULL) 2221 return (ENXIO); 2222 (*bdev->d_strategy)(bp); 2223 2224 error = biowait(bp); 2225 2226 if (!error) { 2227 memcpy(clabel, bp->b_data, 2228 sizeof(RF_ComponentLabel_t)); 2229 } 2230 2231 brelse(bp); 2232 return(error); 2233 } 2234 /* ARGSUSED */ 2235 int 2236 raidwrite_component_label(dev_t dev, struct vnode *b_vp, 2237 RF_ComponentLabel_t *clabel) 2238 { 2239 struct buf *bp; 2240 const struct bdevsw *bdev; 2241 int error; 2242 2243 /* get a block of the appropriate size... */ 2244 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2245 bp->b_dev = dev; 2246 2247 /* get our ducks in a row for the write */ 2248 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2249 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2250 bp->b_flags |= B_WRITE; 2251 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2252 2253 memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE ); 2254 2255 memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t)); 2256 2257 bdev = bdevsw_lookup(bp->b_dev); 2258 if (bdev == NULL) 2259 return (ENXIO); 2260 (*bdev->d_strategy)(bp); 2261 error = biowait(bp); 2262 brelse(bp); 2263 if (error) { 2264 #if 1 2265 printf("Failed to write RAID component info!\n"); 2266 #endif 2267 } 2268 2269 return(error); 2270 } 2271 2272 void 2273 rf_markalldirty(RF_Raid_t *raidPtr) 2274 { 2275 RF_ComponentLabel_t clabel; 2276 int sparecol; 2277 int c; 2278 int j; 2279 int scol = -1; 2280 2281 raidPtr->mod_counter++; 2282 for (c = 0; c < raidPtr->numCol; c++) { 2283 /* we don't want to touch (at all) a disk that has 2284 failed */ 2285 if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) { 2286 raidread_component_label( 2287 raidPtr->Disks[c].dev, 2288 raidPtr->raid_cinfo[c].ci_vp, 2289 &clabel); 2290 if (clabel.status == rf_ds_spared) { 2291 /* XXX do something special... 2292 but whatever you do, don't 2293 try to access it!! */ 2294 } else { 2295 raidmarkdirty( 2296 raidPtr->Disks[c].dev, 2297 raidPtr->raid_cinfo[c].ci_vp, 2298 raidPtr->mod_counter); 2299 } 2300 } 2301 } 2302 2303 for( c = 0; c < raidPtr->numSpare ; c++) { 2304 sparecol = raidPtr->numCol + c; 2305 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2306 /* 2307 2308 we claim this disk is "optimal" if it's 2309 rf_ds_used_spare, as that means it should be 2310 directly substitutable for the disk it replaced. 2311 We note that too... 2312 2313 */ 2314 2315 for(j=0;j<raidPtr->numCol;j++) { 2316 if (raidPtr->Disks[j].spareCol == sparecol) { 2317 scol = j; 2318 break; 2319 } 2320 } 2321 2322 raidread_component_label( 2323 raidPtr->Disks[sparecol].dev, 2324 raidPtr->raid_cinfo[sparecol].ci_vp, 2325 &clabel); 2326 /* make sure status is noted */ 2327 2328 raid_init_component_label(raidPtr, &clabel); 2329 2330 clabel.row = 0; 2331 clabel.column = scol; 2332 /* Note: we *don't* change status from rf_ds_used_spare 2333 to rf_ds_optimal */ 2334 /* clabel.status = rf_ds_optimal; */ 2335 2336 raidmarkdirty(raidPtr->Disks[sparecol].dev, 2337 raidPtr->raid_cinfo[sparecol].ci_vp, 2338 raidPtr->mod_counter); 2339 } 2340 } 2341 } 2342 2343 2344 void 2345 rf_update_component_labels(RF_Raid_t *raidPtr, int final) 2346 { 2347 RF_ComponentLabel_t clabel; 2348 int sparecol; 2349 int c; 2350 int j; 2351 int scol; 2352 2353 scol = -1; 2354 2355 /* XXX should do extra checks to make sure things really are clean, 2356 rather than blindly setting the clean bit... */ 2357 2358 raidPtr->mod_counter++; 2359 2360 for (c = 0; c < raidPtr->numCol; c++) { 2361 if (raidPtr->Disks[c].status == rf_ds_optimal) { 2362 raidread_component_label( 2363 raidPtr->Disks[c].dev, 2364 raidPtr->raid_cinfo[c].ci_vp, 2365 &clabel); 2366 /* make sure status is noted */ 2367 clabel.status = rf_ds_optimal; 2368 2369 /* bump the counter */ 2370 clabel.mod_counter = raidPtr->mod_counter; 2371 2372 raidwrite_component_label( 2373 raidPtr->Disks[c].dev, 2374 raidPtr->raid_cinfo[c].ci_vp, 2375 &clabel); 2376 if (final == RF_FINAL_COMPONENT_UPDATE) { 2377 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2378 raidmarkclean( 2379 raidPtr->Disks[c].dev, 2380 raidPtr->raid_cinfo[c].ci_vp, 2381 raidPtr->mod_counter); 2382 } 2383 } 2384 } 2385 /* else we don't touch it.. */ 2386 } 2387 2388 for( c = 0; c < raidPtr->numSpare ; c++) { 2389 sparecol = raidPtr->numCol + c; 2390 /* Need to ensure that the reconstruct actually completed! */ 2391 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2392 /* 2393 2394 we claim this disk is "optimal" if it's 2395 rf_ds_used_spare, as that means it should be 2396 directly substitutable for the disk it replaced. 2397 We note that too... 2398 2399 */ 2400 2401 for(j=0;j<raidPtr->numCol;j++) { 2402 if (raidPtr->Disks[j].spareCol == sparecol) { 2403 scol = j; 2404 break; 2405 } 2406 } 2407 2408 /* XXX shouldn't *really* need this... */ 2409 raidread_component_label( 2410 raidPtr->Disks[sparecol].dev, 2411 raidPtr->raid_cinfo[sparecol].ci_vp, 2412 &clabel); 2413 /* make sure status is noted */ 2414 2415 raid_init_component_label(raidPtr, &clabel); 2416 2417 clabel.mod_counter = raidPtr->mod_counter; 2418 clabel.column = scol; 2419 clabel.status = rf_ds_optimal; 2420 2421 raidwrite_component_label( 2422 raidPtr->Disks[sparecol].dev, 2423 raidPtr->raid_cinfo[sparecol].ci_vp, 2424 &clabel); 2425 if (final == RF_FINAL_COMPONENT_UPDATE) { 2426 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2427 raidmarkclean( raidPtr->Disks[sparecol].dev, 2428 raidPtr->raid_cinfo[sparecol].ci_vp, 2429 raidPtr->mod_counter); 2430 } 2431 } 2432 } 2433 } 2434 } 2435 2436 void 2437 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured) 2438 { 2439 struct proc *p; 2440 struct lwp *l; 2441 2442 p = raidPtr->engine_thread; 2443 l = LIST_FIRST(&p->p_lwps); 2444 2445 if (vp != NULL) { 2446 if (auto_configured == 1) { 2447 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2448 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2449 vput(vp); 2450 2451 } else { 2452 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l); 2453 } 2454 } 2455 } 2456 2457 2458 void 2459 rf_UnconfigureVnodes(RF_Raid_t *raidPtr) 2460 { 2461 int r,c; 2462 struct vnode *vp; 2463 int acd; 2464 2465 2466 /* We take this opportunity to close the vnodes like we should.. */ 2467 2468 for (c = 0; c < raidPtr->numCol; c++) { 2469 vp = raidPtr->raid_cinfo[c].ci_vp; 2470 acd = raidPtr->Disks[c].auto_configured; 2471 rf_close_component(raidPtr, vp, acd); 2472 raidPtr->raid_cinfo[c].ci_vp = NULL; 2473 raidPtr->Disks[c].auto_configured = 0; 2474 } 2475 2476 for (r = 0; r < raidPtr->numSpare; r++) { 2477 vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp; 2478 acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured; 2479 rf_close_component(raidPtr, vp, acd); 2480 raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL; 2481 raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0; 2482 } 2483 } 2484 2485 2486 void 2487 rf_ReconThread(struct rf_recon_req *req) 2488 { 2489 int s; 2490 RF_Raid_t *raidPtr; 2491 2492 s = splbio(); 2493 raidPtr = (RF_Raid_t *) req->raidPtr; 2494 raidPtr->recon_in_progress = 1; 2495 2496 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col, 2497 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2498 2499 RF_Free(req, sizeof(*req)); 2500 2501 raidPtr->recon_in_progress = 0; 2502 splx(s); 2503 2504 /* That's all... */ 2505 kthread_exit(0); /* does not return */ 2506 } 2507 2508 void 2509 rf_RewriteParityThread(RF_Raid_t *raidPtr) 2510 { 2511 int retcode; 2512 int s; 2513 2514 raidPtr->parity_rewrite_stripes_done = 0; 2515 raidPtr->parity_rewrite_in_progress = 1; 2516 s = splbio(); 2517 retcode = rf_RewriteParity(raidPtr); 2518 splx(s); 2519 if (retcode) { 2520 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2521 } else { 2522 /* set the clean bit! If we shutdown correctly, 2523 the clean bit on each component label will get 2524 set */ 2525 raidPtr->parity_good = RF_RAID_CLEAN; 2526 } 2527 raidPtr->parity_rewrite_in_progress = 0; 2528 2529 /* Anyone waiting for us to stop? If so, inform them... */ 2530 if (raidPtr->waitShutdown) { 2531 wakeup(&raidPtr->parity_rewrite_in_progress); 2532 } 2533 2534 /* That's all... */ 2535 kthread_exit(0); /* does not return */ 2536 } 2537 2538 2539 void 2540 rf_CopybackThread(RF_Raid_t *raidPtr) 2541 { 2542 int s; 2543 2544 raidPtr->copyback_in_progress = 1; 2545 s = splbio(); 2546 rf_CopybackReconstructedData(raidPtr); 2547 splx(s); 2548 raidPtr->copyback_in_progress = 0; 2549 2550 /* That's all... */ 2551 kthread_exit(0); /* does not return */ 2552 } 2553 2554 2555 void 2556 rf_ReconstructInPlaceThread(struct rf_recon_req *req) 2557 { 2558 int s; 2559 RF_Raid_t *raidPtr; 2560 2561 s = splbio(); 2562 raidPtr = req->raidPtr; 2563 raidPtr->recon_in_progress = 1; 2564 rf_ReconstructInPlace(raidPtr, req->col); 2565 RF_Free(req, sizeof(*req)); 2566 raidPtr->recon_in_progress = 0; 2567 splx(s); 2568 2569 /* That's all... */ 2570 kthread_exit(0); /* does not return */ 2571 } 2572 2573 RF_AutoConfig_t * 2574 rf_find_raid_components() 2575 { 2576 struct vnode *vp; 2577 struct disklabel label; 2578 struct device *dv; 2579 dev_t dev; 2580 int bmajor; 2581 int error; 2582 int i; 2583 int good_one; 2584 RF_ComponentLabel_t *clabel; 2585 RF_AutoConfig_t *ac_list; 2586 RF_AutoConfig_t *ac; 2587 2588 2589 /* initialize the AutoConfig list */ 2590 ac_list = NULL; 2591 2592 /* we begin by trolling through *all* the devices on the system */ 2593 2594 for (dv = alldevs.tqh_first; dv != NULL; 2595 dv = dv->dv_list.tqe_next) { 2596 2597 /* we are only interested in disks... */ 2598 if (device_class(dv) != DV_DISK) 2599 continue; 2600 2601 /* we don't care about floppies... */ 2602 if (!strcmp(dv->dv_cfdata->cf_name,"fd")) { 2603 continue; 2604 } 2605 2606 /* we don't care about CD's... */ 2607 if (!strcmp(dv->dv_cfdata->cf_name,"cd")) { 2608 continue; 2609 } 2610 2611 /* hdfd is the Atari/Hades floppy driver */ 2612 if (!strcmp(dv->dv_cfdata->cf_name,"hdfd")) { 2613 continue; 2614 } 2615 /* fdisa is the Atari/Milan floppy driver */ 2616 if (!strcmp(dv->dv_cfdata->cf_name,"fdisa")) { 2617 continue; 2618 } 2619 2620 /* need to find the device_name_to_block_device_major stuff */ 2621 bmajor = devsw_name2blk(dv->dv_xname, NULL, 0); 2622 2623 /* get a vnode for the raw partition of this disk */ 2624 2625 dev = MAKEDISKDEV(bmajor, dv->dv_unit, RAW_PART); 2626 if (bdevvp(dev, &vp)) 2627 panic("RAID can't alloc vnode"); 2628 2629 error = VOP_OPEN(vp, FREAD, NOCRED, 0); 2630 2631 if (error) { 2632 /* "Who cares." Continue looking 2633 for something that exists*/ 2634 vput(vp); 2635 continue; 2636 } 2637 2638 /* Ok, the disk exists. Go get the disklabel. */ 2639 error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED, 0); 2640 if (error) { 2641 /* 2642 * XXX can't happen - open() would 2643 * have errored out (or faked up one) 2644 */ 2645 if (error != ENOTTY) 2646 printf("RAIDframe: can't get label for dev " 2647 "%s (%d)\n", dv->dv_xname, error); 2648 } 2649 2650 /* don't need this any more. We'll allocate it again 2651 a little later if we really do... */ 2652 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2653 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2654 vput(vp); 2655 2656 if (error) 2657 continue; 2658 2659 for (i=0; i < label.d_npartitions; i++) { 2660 /* We only support partitions marked as RAID */ 2661 if (label.d_partitions[i].p_fstype != FS_RAID) 2662 continue; 2663 2664 dev = MAKEDISKDEV(bmajor, dv->dv_unit, i); 2665 if (bdevvp(dev, &vp)) 2666 panic("RAID can't alloc vnode"); 2667 2668 error = VOP_OPEN(vp, FREAD, NOCRED, 0); 2669 if (error) { 2670 /* Whatever... */ 2671 vput(vp); 2672 continue; 2673 } 2674 2675 good_one = 0; 2676 2677 clabel = (RF_ComponentLabel_t *) 2678 malloc(sizeof(RF_ComponentLabel_t), 2679 M_RAIDFRAME, M_NOWAIT); 2680 if (clabel == NULL) { 2681 /* XXX CLEANUP HERE */ 2682 printf("RAID auto config: out of memory!\n"); 2683 return(NULL); /* XXX probably should panic? */ 2684 } 2685 2686 if (!raidread_component_label(dev, vp, clabel)) { 2687 /* Got the label. Does it look reasonable? */ 2688 if (rf_reasonable_label(clabel) && 2689 (clabel->partitionSize <= 2690 label.d_partitions[i].p_size)) { 2691 #if DEBUG 2692 printf("Component on: %s%c: %d\n", 2693 dv->dv_xname, 'a'+i, 2694 label.d_partitions[i].p_size); 2695 rf_print_component_label(clabel); 2696 #endif 2697 /* if it's reasonable, add it, 2698 else ignore it. */ 2699 ac = (RF_AutoConfig_t *) 2700 malloc(sizeof(RF_AutoConfig_t), 2701 M_RAIDFRAME, 2702 M_NOWAIT); 2703 if (ac == NULL) { 2704 /* XXX should panic?? */ 2705 return(NULL); 2706 } 2707 2708 snprintf(ac->devname, 2709 sizeof(ac->devname), "%s%c", 2710 dv->dv_xname, 'a'+i); 2711 ac->dev = dev; 2712 ac->vp = vp; 2713 ac->clabel = clabel; 2714 ac->next = ac_list; 2715 ac_list = ac; 2716 good_one = 1; 2717 } 2718 } 2719 if (!good_one) { 2720 /* cleanup */ 2721 free(clabel, M_RAIDFRAME); 2722 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2723 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0); 2724 vput(vp); 2725 } 2726 } 2727 } 2728 return(ac_list); 2729 } 2730 2731 static int 2732 rf_reasonable_label(RF_ComponentLabel_t *clabel) 2733 { 2734 2735 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 2736 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 2737 ((clabel->clean == RF_RAID_CLEAN) || 2738 (clabel->clean == RF_RAID_DIRTY)) && 2739 clabel->row >=0 && 2740 clabel->column >= 0 && 2741 clabel->num_rows > 0 && 2742 clabel->num_columns > 0 && 2743 clabel->row < clabel->num_rows && 2744 clabel->column < clabel->num_columns && 2745 clabel->blockSize > 0 && 2746 clabel->numBlocks > 0) { 2747 /* label looks reasonable enough... */ 2748 return(1); 2749 } 2750 return(0); 2751 } 2752 2753 2754 #if DEBUG 2755 void 2756 rf_print_component_label(RF_ComponentLabel_t *clabel) 2757 { 2758 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 2759 clabel->row, clabel->column, 2760 clabel->num_rows, clabel->num_columns); 2761 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 2762 clabel->version, clabel->serial_number, 2763 clabel->mod_counter); 2764 printf(" Clean: %s Status: %d\n", 2765 clabel->clean ? "Yes" : "No", clabel->status ); 2766 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 2767 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 2768 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n", 2769 (char) clabel->parityConfig, clabel->blockSize, 2770 clabel->numBlocks); 2771 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" ); 2772 printf(" Contains root partition: %s\n", 2773 clabel->root_partition ? "Yes" : "No" ); 2774 printf(" Last configured as: raid%d\n", clabel->last_unit ); 2775 #if 0 2776 printf(" Config order: %d\n", clabel->config_order); 2777 #endif 2778 2779 } 2780 #endif 2781 2782 RF_ConfigSet_t * 2783 rf_create_auto_sets(RF_AutoConfig_t *ac_list) 2784 { 2785 RF_AutoConfig_t *ac; 2786 RF_ConfigSet_t *config_sets; 2787 RF_ConfigSet_t *cset; 2788 RF_AutoConfig_t *ac_next; 2789 2790 2791 config_sets = NULL; 2792 2793 /* Go through the AutoConfig list, and figure out which components 2794 belong to what sets. */ 2795 ac = ac_list; 2796 while(ac!=NULL) { 2797 /* we're going to putz with ac->next, so save it here 2798 for use at the end of the loop */ 2799 ac_next = ac->next; 2800 2801 if (config_sets == NULL) { 2802 /* will need at least this one... */ 2803 config_sets = (RF_ConfigSet_t *) 2804 malloc(sizeof(RF_ConfigSet_t), 2805 M_RAIDFRAME, M_NOWAIT); 2806 if (config_sets == NULL) { 2807 panic("rf_create_auto_sets: No memory!"); 2808 } 2809 /* this one is easy :) */ 2810 config_sets->ac = ac; 2811 config_sets->next = NULL; 2812 config_sets->rootable = 0; 2813 ac->next = NULL; 2814 } else { 2815 /* which set does this component fit into? */ 2816 cset = config_sets; 2817 while(cset!=NULL) { 2818 if (rf_does_it_fit(cset, ac)) { 2819 /* looks like it matches... */ 2820 ac->next = cset->ac; 2821 cset->ac = ac; 2822 break; 2823 } 2824 cset = cset->next; 2825 } 2826 if (cset==NULL) { 2827 /* didn't find a match above... new set..*/ 2828 cset = (RF_ConfigSet_t *) 2829 malloc(sizeof(RF_ConfigSet_t), 2830 M_RAIDFRAME, M_NOWAIT); 2831 if (cset == NULL) { 2832 panic("rf_create_auto_sets: No memory!"); 2833 } 2834 cset->ac = ac; 2835 ac->next = NULL; 2836 cset->next = config_sets; 2837 cset->rootable = 0; 2838 config_sets = cset; 2839 } 2840 } 2841 ac = ac_next; 2842 } 2843 2844 2845 return(config_sets); 2846 } 2847 2848 static int 2849 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac) 2850 { 2851 RF_ComponentLabel_t *clabel1, *clabel2; 2852 2853 /* If this one matches the *first* one in the set, that's good 2854 enough, since the other members of the set would have been 2855 through here too... */ 2856 /* note that we are not checking partitionSize here.. 2857 2858 Note that we are also not checking the mod_counters here. 2859 If everything else matches execpt the mod_counter, that's 2860 good enough for this test. We will deal with the mod_counters 2861 a little later in the autoconfiguration process. 2862 2863 (clabel1->mod_counter == clabel2->mod_counter) && 2864 2865 The reason we don't check for this is that failed disks 2866 will have lower modification counts. If those disks are 2867 not added to the set they used to belong to, then they will 2868 form their own set, which may result in 2 different sets, 2869 for example, competing to be configured at raid0, and 2870 perhaps competing to be the root filesystem set. If the 2871 wrong ones get configured, or both attempt to become /, 2872 weird behaviour and or serious lossage will occur. Thus we 2873 need to bring them into the fold here, and kick them out at 2874 a later point. 2875 2876 */ 2877 2878 clabel1 = cset->ac->clabel; 2879 clabel2 = ac->clabel; 2880 if ((clabel1->version == clabel2->version) && 2881 (clabel1->serial_number == clabel2->serial_number) && 2882 (clabel1->num_rows == clabel2->num_rows) && 2883 (clabel1->num_columns == clabel2->num_columns) && 2884 (clabel1->sectPerSU == clabel2->sectPerSU) && 2885 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 2886 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 2887 (clabel1->parityConfig == clabel2->parityConfig) && 2888 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 2889 (clabel1->blockSize == clabel2->blockSize) && 2890 (clabel1->numBlocks == clabel2->numBlocks) && 2891 (clabel1->autoconfigure == clabel2->autoconfigure) && 2892 (clabel1->root_partition == clabel2->root_partition) && 2893 (clabel1->last_unit == clabel2->last_unit) && 2894 (clabel1->config_order == clabel2->config_order)) { 2895 /* if it get's here, it almost *has* to be a match */ 2896 } else { 2897 /* it's not consistent with somebody in the set.. 2898 punt */ 2899 return(0); 2900 } 2901 /* all was fine.. it must fit... */ 2902 return(1); 2903 } 2904 2905 int 2906 rf_have_enough_components(RF_ConfigSet_t *cset) 2907 { 2908 RF_AutoConfig_t *ac; 2909 RF_AutoConfig_t *auto_config; 2910 RF_ComponentLabel_t *clabel; 2911 int c; 2912 int num_cols; 2913 int num_missing; 2914 int mod_counter; 2915 int mod_counter_found; 2916 int even_pair_failed; 2917 char parity_type; 2918 2919 2920 /* check to see that we have enough 'live' components 2921 of this set. If so, we can configure it if necessary */ 2922 2923 num_cols = cset->ac->clabel->num_columns; 2924 parity_type = cset->ac->clabel->parityConfig; 2925 2926 /* XXX Check for duplicate components!?!?!? */ 2927 2928 /* Determine what the mod_counter is supposed to be for this set. */ 2929 2930 mod_counter_found = 0; 2931 mod_counter = 0; 2932 ac = cset->ac; 2933 while(ac!=NULL) { 2934 if (mod_counter_found==0) { 2935 mod_counter = ac->clabel->mod_counter; 2936 mod_counter_found = 1; 2937 } else { 2938 if (ac->clabel->mod_counter > mod_counter) { 2939 mod_counter = ac->clabel->mod_counter; 2940 } 2941 } 2942 ac = ac->next; 2943 } 2944 2945 num_missing = 0; 2946 auto_config = cset->ac; 2947 2948 even_pair_failed = 0; 2949 for(c=0; c<num_cols; c++) { 2950 ac = auto_config; 2951 while(ac!=NULL) { 2952 if ((ac->clabel->column == c) && 2953 (ac->clabel->mod_counter == mod_counter)) { 2954 /* it's this one... */ 2955 #if DEBUG 2956 printf("Found: %s at %d\n", 2957 ac->devname,c); 2958 #endif 2959 break; 2960 } 2961 ac=ac->next; 2962 } 2963 if (ac==NULL) { 2964 /* Didn't find one here! */ 2965 /* special case for RAID 1, especially 2966 where there are more than 2 2967 components (where RAIDframe treats 2968 things a little differently :( ) */ 2969 if (parity_type == '1') { 2970 if (c%2 == 0) { /* even component */ 2971 even_pair_failed = 1; 2972 } else { /* odd component. If 2973 we're failed, and 2974 so is the even 2975 component, it's 2976 "Good Night, Charlie" */ 2977 if (even_pair_failed == 1) { 2978 return(0); 2979 } 2980 } 2981 } else { 2982 /* normal accounting */ 2983 num_missing++; 2984 } 2985 } 2986 if ((parity_type == '1') && (c%2 == 1)) { 2987 /* Just did an even component, and we didn't 2988 bail.. reset the even_pair_failed flag, 2989 and go on to the next component.... */ 2990 even_pair_failed = 0; 2991 } 2992 } 2993 2994 clabel = cset->ac->clabel; 2995 2996 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 2997 ((clabel->parityConfig == '4') && (num_missing > 1)) || 2998 ((clabel->parityConfig == '5') && (num_missing > 1))) { 2999 /* XXX this needs to be made *much* more general */ 3000 /* Too many failures */ 3001 return(0); 3002 } 3003 /* otherwise, all is well, and we've got enough to take a kick 3004 at autoconfiguring this set */ 3005 return(1); 3006 } 3007 3008 void 3009 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config, 3010 RF_Raid_t *raidPtr) 3011 { 3012 RF_ComponentLabel_t *clabel; 3013 int i; 3014 3015 clabel = ac->clabel; 3016 3017 /* 1. Fill in the common stuff */ 3018 config->numRow = clabel->num_rows = 1; 3019 config->numCol = clabel->num_columns; 3020 config->numSpare = 0; /* XXX should this be set here? */ 3021 config->sectPerSU = clabel->sectPerSU; 3022 config->SUsPerPU = clabel->SUsPerPU; 3023 config->SUsPerRU = clabel->SUsPerRU; 3024 config->parityConfig = clabel->parityConfig; 3025 /* XXX... */ 3026 strcpy(config->diskQueueType,"fifo"); 3027 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3028 config->layoutSpecificSize = 0; /* XXX ?? */ 3029 3030 while(ac!=NULL) { 3031 /* row/col values will be in range due to the checks 3032 in reasonable_label() */ 3033 strcpy(config->devnames[0][ac->clabel->column], 3034 ac->devname); 3035 ac = ac->next; 3036 } 3037 3038 for(i=0;i<RF_MAXDBGV;i++) { 3039 config->debugVars[i][0] = 0; 3040 } 3041 } 3042 3043 int 3044 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value) 3045 { 3046 RF_ComponentLabel_t clabel; 3047 struct vnode *vp; 3048 dev_t dev; 3049 int column; 3050 int sparecol; 3051 3052 raidPtr->autoconfigure = new_value; 3053 3054 for(column=0; column<raidPtr->numCol; column++) { 3055 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3056 dev = raidPtr->Disks[column].dev; 3057 vp = raidPtr->raid_cinfo[column].ci_vp; 3058 raidread_component_label(dev, vp, &clabel); 3059 clabel.autoconfigure = new_value; 3060 raidwrite_component_label(dev, vp, &clabel); 3061 } 3062 } 3063 for(column = 0; column < raidPtr->numSpare ; column++) { 3064 sparecol = raidPtr->numCol + column; 3065 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3066 dev = raidPtr->Disks[sparecol].dev; 3067 vp = raidPtr->raid_cinfo[sparecol].ci_vp; 3068 raidread_component_label(dev, vp, &clabel); 3069 clabel.autoconfigure = new_value; 3070 raidwrite_component_label(dev, vp, &clabel); 3071 } 3072 } 3073 return(new_value); 3074 } 3075 3076 int 3077 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value) 3078 { 3079 RF_ComponentLabel_t clabel; 3080 struct vnode *vp; 3081 dev_t dev; 3082 int column; 3083 int sparecol; 3084 3085 raidPtr->root_partition = new_value; 3086 for(column=0; column<raidPtr->numCol; column++) { 3087 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3088 dev = raidPtr->Disks[column].dev; 3089 vp = raidPtr->raid_cinfo[column].ci_vp; 3090 raidread_component_label(dev, vp, &clabel); 3091 clabel.root_partition = new_value; 3092 raidwrite_component_label(dev, vp, &clabel); 3093 } 3094 } 3095 for(column = 0; column < raidPtr->numSpare ; column++) { 3096 sparecol = raidPtr->numCol + column; 3097 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3098 dev = raidPtr->Disks[sparecol].dev; 3099 vp = raidPtr->raid_cinfo[sparecol].ci_vp; 3100 raidread_component_label(dev, vp, &clabel); 3101 clabel.root_partition = new_value; 3102 raidwrite_component_label(dev, vp, &clabel); 3103 } 3104 } 3105 return(new_value); 3106 } 3107 3108 void 3109 rf_release_all_vps(RF_ConfigSet_t *cset) 3110 { 3111 RF_AutoConfig_t *ac; 3112 3113 ac = cset->ac; 3114 while(ac!=NULL) { 3115 /* Close the vp, and give it back */ 3116 if (ac->vp) { 3117 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3118 VOP_CLOSE(ac->vp, FREAD, NOCRED, 0); 3119 vput(ac->vp); 3120 ac->vp = NULL; 3121 } 3122 ac = ac->next; 3123 } 3124 } 3125 3126 3127 void 3128 rf_cleanup_config_set(RF_ConfigSet_t *cset) 3129 { 3130 RF_AutoConfig_t *ac; 3131 RF_AutoConfig_t *next_ac; 3132 3133 ac = cset->ac; 3134 while(ac!=NULL) { 3135 next_ac = ac->next; 3136 /* nuke the label */ 3137 free(ac->clabel, M_RAIDFRAME); 3138 /* cleanup the config structure */ 3139 free(ac, M_RAIDFRAME); 3140 /* "next.." */ 3141 ac = next_ac; 3142 } 3143 /* and, finally, nuke the config set */ 3144 free(cset, M_RAIDFRAME); 3145 } 3146 3147 3148 void 3149 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel) 3150 { 3151 /* current version number */ 3152 clabel->version = RF_COMPONENT_LABEL_VERSION; 3153 clabel->serial_number = raidPtr->serial_number; 3154 clabel->mod_counter = raidPtr->mod_counter; 3155 clabel->num_rows = 1; 3156 clabel->num_columns = raidPtr->numCol; 3157 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3158 clabel->status = rf_ds_optimal; /* "It's good!" */ 3159 3160 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3161 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3162 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3163 3164 clabel->blockSize = raidPtr->bytesPerSector; 3165 clabel->numBlocks = raidPtr->sectorsPerDisk; 3166 3167 /* XXX not portable */ 3168 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3169 clabel->maxOutstanding = raidPtr->maxOutstanding; 3170 clabel->autoconfigure = raidPtr->autoconfigure; 3171 clabel->root_partition = raidPtr->root_partition; 3172 clabel->last_unit = raidPtr->raidid; 3173 clabel->config_order = raidPtr->config_order; 3174 } 3175 3176 int 3177 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit) 3178 { 3179 RF_Raid_t *raidPtr; 3180 RF_Config_t *config; 3181 int raidID; 3182 int retcode; 3183 3184 #if DEBUG 3185 printf("RAID autoconfigure\n"); 3186 #endif 3187 3188 retcode = 0; 3189 *unit = -1; 3190 3191 /* 1. Create a config structure */ 3192 3193 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3194 M_RAIDFRAME, 3195 M_NOWAIT); 3196 if (config==NULL) { 3197 printf("Out of mem!?!?\n"); 3198 /* XXX do something more intelligent here. */ 3199 return(1); 3200 } 3201 3202 memset(config, 0, sizeof(RF_Config_t)); 3203 3204 /* 3205 2. Figure out what RAID ID this one is supposed to live at 3206 See if we can get the same RAID dev that it was configured 3207 on last time.. 3208 */ 3209 3210 raidID = cset->ac->clabel->last_unit; 3211 if ((raidID < 0) || (raidID >= numraid)) { 3212 /* let's not wander off into lala land. */ 3213 raidID = numraid - 1; 3214 } 3215 if (raidPtrs[raidID]->valid != 0) { 3216 3217 /* 3218 Nope... Go looking for an alternative... 3219 Start high so we don't immediately use raid0 if that's 3220 not taken. 3221 */ 3222 3223 for(raidID = numraid - 1; raidID >= 0; raidID--) { 3224 if (raidPtrs[raidID]->valid == 0) { 3225 /* can use this one! */ 3226 break; 3227 } 3228 } 3229 } 3230 3231 if (raidID < 0) { 3232 /* punt... */ 3233 printf("Unable to auto configure this set!\n"); 3234 printf("(Out of RAID devs!)\n"); 3235 return(1); 3236 } 3237 3238 #if DEBUG 3239 printf("Configuring raid%d:\n",raidID); 3240 #endif 3241 3242 raidPtr = raidPtrs[raidID]; 3243 3244 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3245 raidPtr->raidid = raidID; 3246 raidPtr->openings = RAIDOUTSTANDING; 3247 3248 /* 3. Build the configuration structure */ 3249 rf_create_configuration(cset->ac, config, raidPtr); 3250 3251 /* 4. Do the configuration */ 3252 retcode = rf_Configure(raidPtr, config, cset->ac); 3253 3254 if (retcode == 0) { 3255 3256 raidinit(raidPtrs[raidID]); 3257 3258 rf_markalldirty(raidPtrs[raidID]); 3259 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3260 if (cset->ac->clabel->root_partition==1) { 3261 /* everything configured just fine. Make a note 3262 that this set is eligible to be root. */ 3263 cset->rootable = 1; 3264 /* XXX do this here? */ 3265 raidPtrs[raidID]->root_partition = 1; 3266 } 3267 } 3268 3269 /* 5. Cleanup */ 3270 free(config, M_RAIDFRAME); 3271 3272 *unit = raidID; 3273 return(retcode); 3274 } 3275 3276 void 3277 rf_disk_unbusy(RF_RaidAccessDesc_t *desc) 3278 { 3279 struct buf *bp; 3280 3281 bp = (struct buf *)desc->bp; 3282 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3283 (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ)); 3284 } 3285 3286 void 3287 rf_pool_init(struct pool *p, size_t size, const char *w_chan, 3288 size_t xmin, size_t xmax) 3289 { 3290 pool_init(p, size, 0, 0, 0, w_chan, NULL); 3291 pool_sethiwat(p, xmax); 3292 pool_prime(p, xmin); 3293 pool_setlowat(p, xmin); 3294 } 3295 3296 /* 3297 * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see 3298 * if there is IO pending and if that IO could possibly be done for a 3299 * given RAID set. Returns 0 if IO is waiting and can be done, 1 3300 * otherwise. 3301 * 3302 */ 3303 3304 int 3305 rf_buf_queue_check(int raidid) 3306 { 3307 if ((BUFQ_PEEK(raid_softc[raidid].buf_queue) != NULL) && 3308 raidPtrs[raidid]->openings > 0) { 3309 /* there is work to do */ 3310 return 0; 3311 } 3312 /* default is nothing to do */ 3313 return 1; 3314 } 3315
Indexes created Fri Sep 26 08:10:20 GMT 2025