rf_netbsdkintf.c revision 1.250.2.1
1 /* $NetBSD: rf_netbsdkintf.c,v 1.250.2.1 2009/01/19 13:19:02 skrll Exp $ */ 2 /*- 3 * Copyright (c) 1996, 1997, 1998, 2008 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 * 18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28 * POSSIBILITY OF SUCH DAMAGE. 29 */ 30 31 /* 32 * Copyright (c) 1990, 1993 33 * The Regents of the University of California. All rights reserved. 34 * 35 * This code is derived from software contributed to Berkeley by 36 * the Systems Programming Group of the University of Utah Computer 37 * Science Department. 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. Neither the name of the University nor the names of its contributors 48 * may be used to endorse or promote products derived from this software 49 * without specific prior written permission. 50 * 51 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 52 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 53 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 54 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 55 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 56 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 57 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 58 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 59 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 60 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 61 * SUCH DAMAGE. 62 * 63 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 64 * 65 * @(#)cd.c 8.2 (Berkeley) 11/16/93 66 */ 67 68 /* 69 * Copyright (c) 1988 University of Utah. 70 * 71 * This code is derived from software contributed to Berkeley by 72 * the Systems Programming Group of the University of Utah Computer 73 * Science Department. 74 * 75 * Redistribution and use in source and binary forms, with or without 76 * modification, are permitted provided that the following conditions 77 * are met: 78 * 1. Redistributions of source code must retain the above copyright 79 * notice, this list of conditions and the following disclaimer. 80 * 2. Redistributions in binary form must reproduce the above copyright 81 * notice, this list of conditions and the following disclaimer in the 82 * documentation and/or other materials provided with the distribution. 83 * 3. All advertising materials mentioning features or use of this software 84 * must display the following acknowledgement: 85 * This product includes software developed by the University of 86 * California, Berkeley and its contributors. 87 * 4. Neither the name of the University nor the names of its contributors 88 * may be used to endorse or promote products derived from this software 89 * without specific prior written permission. 90 * 91 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 92 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 93 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 94 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 95 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 96 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 97 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 98 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 99 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 100 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 101 * SUCH DAMAGE. 102 * 103 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 104 * 105 * @(#)cd.c 8.2 (Berkeley) 11/16/93 106 */ 107 108 /* 109 * Copyright (c) 1995 Carnegie-Mellon University. 110 * All rights reserved. 111 * 112 * Authors: Mark Holland, Jim Zelenka 113 * 114 * Permission to use, copy, modify and distribute this software and 115 * its documentation is hereby granted, provided that both the copyright 116 * notice and this permission notice appear in all copies of the 117 * software, derivative works or modified versions, and any portions 118 * thereof, and that both notices appear in supporting documentation. 119 * 120 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 121 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 122 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 123 * 124 * Carnegie Mellon requests users of this software to return to 125 * 126 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU 127 * School of Computer Science 128 * Carnegie Mellon University 129 * Pittsburgh PA 15213-3890 130 * 131 * any improvements or extensions that they make and grant Carnegie the 132 * rights to redistribute these changes. 133 */ 134 135 /*********************************************************** 136 * 137 * rf_kintf.c -- the kernel interface routines for RAIDframe 138 * 139 ***********************************************************/ 140 141 #include <sys/cdefs.h> 142 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.250.2.1 2009/01/19 13:19:02 skrll Exp $"); 143 144 #ifdef _KERNEL_OPT 145 #include "opt_compat_netbsd.h" 146 #include "opt_raid_autoconfig.h" 147 #include "raid.h" 148 #endif 149 150 #include <sys/param.h> 151 #include <sys/errno.h> 152 #include <sys/pool.h> 153 #include <sys/proc.h> 154 #include <sys/queue.h> 155 #include <sys/disk.h> 156 #include <sys/device.h> 157 #include <sys/stat.h> 158 #include <sys/ioctl.h> 159 #include <sys/fcntl.h> 160 #include <sys/systm.h> 161 #include <sys/vnode.h> 162 #include <sys/disklabel.h> 163 #include <sys/conf.h> 164 #include <sys/buf.h> 165 #include <sys/bufq.h> 166 #include <sys/user.h> 167 #include <sys/reboot.h> 168 #include <sys/kauth.h> 169 170 #include <prop/proplib.h> 171 172 #include <dev/raidframe/raidframevar.h> 173 #include <dev/raidframe/raidframeio.h> 174 175 #include "rf_raid.h" 176 #include "rf_copyback.h" 177 #include "rf_dag.h" 178 #include "rf_dagflags.h" 179 #include "rf_desc.h" 180 #include "rf_diskqueue.h" 181 #include "rf_etimer.h" 182 #include "rf_general.h" 183 #include "rf_kintf.h" 184 #include "rf_options.h" 185 #include "rf_driver.h" 186 #include "rf_parityscan.h" 187 #include "rf_threadstuff.h" 188 189 #ifdef COMPAT_50 190 #include "rf_compat50.h" 191 #endif 192 193 #ifdef DEBUG 194 int rf_kdebug_level = 0; 195 #define db1_printf(a) if (rf_kdebug_level > 0) printf a 196 #else /* DEBUG */ 197 #define db1_printf(a) { } 198 #endif /* DEBUG */ 199 200 static RF_Raid_t **raidPtrs; /* global raid device descriptors */ 201 202 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 203 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex) 204 205 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a 206 * spare table */ 207 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from 208 * installation process */ 209 #endif 210 211 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures"); 212 213 /* prototypes */ 214 static void KernelWakeupFunc(struct buf *); 215 static void InitBP(struct buf *, struct vnode *, unsigned, 216 dev_t, RF_SectorNum_t, RF_SectorCount_t, void *, void (*) (struct buf *), 217 void *, int, struct proc *); 218 static void raidinit(RF_Raid_t *); 219 220 void raidattach(int); 221 static int raid_match(struct device *, struct cfdata *, void *); 222 static void raid_attach(struct device *, struct device *, void *); 223 static int raid_detach(struct device *, int); 224 225 dev_type_open(raidopen); 226 dev_type_close(raidclose); 227 dev_type_read(raidread); 228 dev_type_write(raidwrite); 229 dev_type_ioctl(raidioctl); 230 dev_type_strategy(raidstrategy); 231 dev_type_dump(raiddump); 232 dev_type_size(raidsize); 233 234 const struct bdevsw raid_bdevsw = { 235 raidopen, raidclose, raidstrategy, raidioctl, 236 raiddump, raidsize, D_DISK 237 }; 238 239 const struct cdevsw raid_cdevsw = { 240 raidopen, raidclose, raidread, raidwrite, raidioctl, 241 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 242 }; 243 244 static struct dkdriver rf_dkdriver = { raidstrategy, minphys }; 245 246 /* XXX Not sure if the following should be replacing the raidPtrs above, 247 or if it should be used in conjunction with that... 248 */ 249 250 struct raid_softc { 251 struct device *sc_dev; 252 int sc_flags; /* flags */ 253 int sc_cflags; /* configuration flags */ 254 uint64_t sc_size; /* size of the raid device */ 255 char sc_xname[20]; /* XXX external name */ 256 struct disk sc_dkdev; /* generic disk device info */ 257 struct bufq_state *buf_queue; /* used for the device queue */ 258 }; 259 /* sc_flags */ 260 #define RAIDF_INITED 0x01 /* unit has been initialized */ 261 #define RAIDF_WLABEL 0x02 /* label area is writable */ 262 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 263 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 264 #define RAIDF_LOCKED 0x80 /* unit is locked */ 265 266 #define raidunit(x) DISKUNIT(x) 267 int numraid = 0; 268 269 extern struct cfdriver raid_cd; 270 CFATTACH_DECL_NEW(raid, sizeof(struct raid_softc), 271 raid_match, raid_attach, raid_detach, NULL); 272 273 /* 274 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 275 * Be aware that large numbers can allow the driver to consume a lot of 276 * kernel memory, especially on writes, and in degraded mode reads. 277 * 278 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 279 * a single 64K write will typically require 64K for the old data, 280 * 64K for the old parity, and 64K for the new parity, for a total 281 * of 192K (if the parity buffer is not re-used immediately). 282 * Even it if is used immediately, that's still 128K, which when multiplied 283 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 284 * 285 * Now in degraded mode, for example, a 64K read on the above setup may 286 * require data reconstruction, which will require *all* of the 4 remaining 287 * disks to participate -- 4 * 32K/disk == 128K again. 288 */ 289 290 #ifndef RAIDOUTSTANDING 291 #define RAIDOUTSTANDING 6 292 #endif 293 294 #define RAIDLABELDEV(dev) \ 295 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 296 297 /* declared here, and made public, for the benefit of KVM stuff.. */ 298 struct raid_softc *raid_softc; 299 300 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *, 301 struct disklabel *); 302 static void raidgetdisklabel(dev_t); 303 static void raidmakedisklabel(struct raid_softc *); 304 305 static int raidlock(struct raid_softc *); 306 static void raidunlock(struct raid_softc *); 307 308 static void rf_markalldirty(RF_Raid_t *); 309 static void rf_set_properties(struct raid_softc *, RF_Raid_t *); 310 311 void rf_ReconThread(struct rf_recon_req *); 312 void rf_RewriteParityThread(RF_Raid_t *raidPtr); 313 void rf_CopybackThread(RF_Raid_t *raidPtr); 314 void rf_ReconstructInPlaceThread(struct rf_recon_req *); 315 int rf_autoconfig(struct device *self); 316 void rf_buildroothack(RF_ConfigSet_t *); 317 318 RF_AutoConfig_t *rf_find_raid_components(void); 319 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *); 320 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *); 321 static int rf_reasonable_label(RF_ComponentLabel_t *); 322 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *); 323 int rf_set_autoconfig(RF_Raid_t *, int); 324 int rf_set_rootpartition(RF_Raid_t *, int); 325 void rf_release_all_vps(RF_ConfigSet_t *); 326 void rf_cleanup_config_set(RF_ConfigSet_t *); 327 int rf_have_enough_components(RF_ConfigSet_t *); 328 int rf_auto_config_set(RF_ConfigSet_t *, int *); 329 static int rf_sync_component_caches(RF_Raid_t *raidPtr); 330 331 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not 332 allow autoconfig to take place. 333 Note that this is overridden by having 334 RAID_AUTOCONFIG as an option in the 335 kernel config file. */ 336 337 struct RF_Pools_s rf_pools; 338 339 void 340 raidattach(int num) 341 { 342 int raidID; 343 int i, rc; 344 345 #ifdef DEBUG 346 printf("raidattach: Asked for %d units\n", num); 347 #endif 348 349 if (num <= 0) { 350 #ifdef DIAGNOSTIC 351 panic("raidattach: count <= 0"); 352 #endif 353 return; 354 } 355 /* This is where all the initialization stuff gets done. */ 356 357 numraid = num; 358 359 /* Make some space for requested number of units... */ 360 361 RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **)); 362 if (raidPtrs == NULL) { 363 panic("raidPtrs is NULL!!"); 364 } 365 366 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 367 rf_mutex_init(&rf_sparet_wait_mutex); 368 369 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 370 #endif 371 372 for (i = 0; i < num; i++) 373 raidPtrs[i] = NULL; 374 rc = rf_BootRaidframe(); 375 if (rc == 0) 376 aprint_normal("Kernelized RAIDframe activated\n"); 377 else 378 panic("Serious error booting RAID!!"); 379 380 /* put together some datastructures like the CCD device does.. This 381 * lets us lock the device and what-not when it gets opened. */ 382 383 raid_softc = (struct raid_softc *) 384 malloc(num * sizeof(struct raid_softc), 385 M_RAIDFRAME, M_NOWAIT); 386 if (raid_softc == NULL) { 387 aprint_error("WARNING: no memory for RAIDframe driver\n"); 388 return; 389 } 390 391 memset(raid_softc, 0, num * sizeof(struct raid_softc)); 392 393 for (raidID = 0; raidID < num; raidID++) { 394 bufq_alloc(&raid_softc[raidID].buf_queue, "fcfs", 0); 395 396 RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t), 397 (RF_Raid_t *)); 398 if (raidPtrs[raidID] == NULL) { 399 aprint_error("WARNING: raidPtrs[%d] is NULL\n", raidID); 400 numraid = raidID; 401 return; 402 } 403 } 404 405 if (config_cfattach_attach(raid_cd.cd_name, &raid_ca)) { 406 aprint_error("raidattach: config_cfattach_attach failed?\n"); 407 } 408 409 #ifdef RAID_AUTOCONFIG 410 raidautoconfig = 1; 411 #endif 412 413 /* 414 * Register a finalizer which will be used to auto-config RAID 415 * sets once all real hardware devices have been found. 416 */ 417 if (config_finalize_register(NULL, rf_autoconfig) != 0) 418 aprint_error("WARNING: unable to register RAIDframe finalizer\n"); 419 } 420 421 int 422 rf_autoconfig(struct device *self) 423 { 424 RF_AutoConfig_t *ac_list; 425 RF_ConfigSet_t *config_sets; 426 427 if (raidautoconfig == 0) 428 return (0); 429 430 /* XXX This code can only be run once. */ 431 raidautoconfig = 0; 432 433 /* 1. locate all RAID components on the system */ 434 #ifdef DEBUG 435 printf("Searching for RAID components...\n"); 436 #endif 437 ac_list = rf_find_raid_components(); 438 439 /* 2. Sort them into their respective sets. */ 440 config_sets = rf_create_auto_sets(ac_list); 441 442 /* 443 * 3. Evaluate each set andconfigure the valid ones. 444 * This gets done in rf_buildroothack(). 445 */ 446 rf_buildroothack(config_sets); 447 448 return 1; 449 } 450 451 void 452 rf_buildroothack(RF_ConfigSet_t *config_sets) 453 { 454 RF_ConfigSet_t *cset; 455 RF_ConfigSet_t *next_cset; 456 int retcode; 457 int raidID; 458 int rootID; 459 int col; 460 int num_root; 461 char *devname; 462 463 rootID = 0; 464 num_root = 0; 465 cset = config_sets; 466 while(cset != NULL ) { 467 next_cset = cset->next; 468 if (rf_have_enough_components(cset) && 469 cset->ac->clabel->autoconfigure==1) { 470 retcode = rf_auto_config_set(cset,&raidID); 471 if (!retcode) { 472 #ifdef DEBUG 473 printf("raid%d: configured ok\n", raidID); 474 #endif 475 if (cset->rootable) { 476 rootID = raidID; 477 num_root++; 478 } 479 } else { 480 /* The autoconfig didn't work :( */ 481 #ifdef DEBUG 482 printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 483 #endif 484 rf_release_all_vps(cset); 485 } 486 } else { 487 /* we're not autoconfiguring this set... 488 release the associated resources */ 489 rf_release_all_vps(cset); 490 } 491 /* cleanup */ 492 rf_cleanup_config_set(cset); 493 cset = next_cset; 494 } 495 496 /* if the user has specified what the root device should be 497 then we don't touch booted_device or boothowto... */ 498 499 if (rootspec != NULL) 500 return; 501 502 /* we found something bootable... */ 503 504 if (num_root == 1) { 505 booted_device = raid_softc[rootID].sc_dev; 506 } else if (num_root > 1) { 507 508 /* 509 * Maybe the MD code can help. If it cannot, then 510 * setroot() will discover that we have no 511 * booted_device and will ask the user if nothing was 512 * hardwired in the kernel config file 513 */ 514 515 if (booted_device == NULL) 516 cpu_rootconf(); 517 if (booted_device == NULL) 518 return; 519 520 num_root = 0; 521 for (raidID = 0; raidID < numraid; raidID++) { 522 if (raidPtrs[raidID]->valid == 0) 523 continue; 524 525 if (raidPtrs[raidID]->root_partition == 0) 526 continue; 527 528 for (col = 0; col < raidPtrs[raidID]->numCol; col++) { 529 devname = raidPtrs[raidID]->Disks[col].devname; 530 devname += sizeof("/dev/") - 1; 531 if (strncmp(devname, device_xname(booted_device), 532 strlen(device_xname(booted_device))) != 0) 533 continue; 534 #ifdef DEBUG 535 printf("raid%d includes boot device %s\n", 536 raidID, devname); 537 #endif 538 num_root++; 539 rootID = raidID; 540 } 541 } 542 543 if (num_root == 1) { 544 booted_device = raid_softc[rootID].sc_dev; 545 } else { 546 /* we can't guess.. require the user to answer... */ 547 boothowto |= RB_ASKNAME; 548 } 549 } 550 } 551 552 553 int 554 raidsize(dev_t dev) 555 { 556 struct raid_softc *rs; 557 struct disklabel *lp; 558 int part, unit, omask, size; 559 560 unit = raidunit(dev); 561 if (unit >= numraid) 562 return (-1); 563 rs = &raid_softc[unit]; 564 565 if ((rs->sc_flags & RAIDF_INITED) == 0) 566 return (-1); 567 568 part = DISKPART(dev); 569 omask = rs->sc_dkdev.dk_openmask & (1 << part); 570 lp = rs->sc_dkdev.dk_label; 571 572 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curlwp)) 573 return (-1); 574 575 if (lp->d_partitions[part].p_fstype != FS_SWAP) 576 size = -1; 577 else 578 size = lp->d_partitions[part].p_size * 579 (lp->d_secsize / DEV_BSIZE); 580 581 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curlwp)) 582 return (-1); 583 584 return (size); 585 586 } 587 588 int 589 raiddump(dev_t dev, daddr_t blkno, void *va, size_t size) 590 { 591 int unit = raidunit(dev); 592 struct raid_softc *rs; 593 const struct bdevsw *bdev; 594 struct disklabel *lp; 595 RF_Raid_t *raidPtr; 596 daddr_t offset; 597 int part, c, sparecol, j, scol, dumpto; 598 int error = 0; 599 600 if (unit >= numraid) 601 return (ENXIO); 602 603 rs = &raid_softc[unit]; 604 raidPtr = raidPtrs[unit]; 605 606 if ((rs->sc_flags & RAIDF_INITED) == 0) 607 return ENXIO; 608 609 /* we only support dumping to RAID 1 sets */ 610 if (raidPtr->Layout.numDataCol != 1 || 611 raidPtr->Layout.numParityCol != 1) 612 return EINVAL; 613 614 615 if ((error = raidlock(rs)) != 0) 616 return error; 617 618 if (size % DEV_BSIZE != 0) { 619 error = EINVAL; 620 goto out; 621 } 622 623 if (blkno + size / DEV_BSIZE > rs->sc_size) { 624 printf("%s: blkno (%" PRIu64 ") + size / DEV_BSIZE (%zu) > " 625 "sc->sc_size (%" PRIu64 ")\n", __func__, blkno, 626 size / DEV_BSIZE, rs->sc_size); 627 error = EINVAL; 628 goto out; 629 } 630 631 part = DISKPART(dev); 632 lp = rs->sc_dkdev.dk_label; 633 offset = lp->d_partitions[part].p_offset + RF_PROTECTED_SECTORS; 634 635 /* figure out what device is alive.. */ 636 637 /* 638 Look for a component to dump to. The preference for the 639 component to dump to is as follows: 640 1) the master 641 2) a used_spare of the master 642 3) the slave 643 4) a used_spare of the slave 644 */ 645 646 dumpto = -1; 647 for (c = 0; c < raidPtr->numCol; c++) { 648 if (raidPtr->Disks[c].status == rf_ds_optimal) { 649 /* this might be the one */ 650 dumpto = c; 651 break; 652 } 653 } 654 655 /* 656 At this point we have possibly selected a live master or a 657 live slave. We now check to see if there is a spared 658 master (or a spared slave), if we didn't find a live master 659 or a live slave. 660 */ 661 662 for (c = 0; c < raidPtr->numSpare; c++) { 663 sparecol = raidPtr->numCol + c; 664 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 665 /* How about this one? */ 666 scol = -1; 667 for(j=0;j<raidPtr->numCol;j++) { 668 if (raidPtr->Disks[j].spareCol == sparecol) { 669 scol = j; 670 break; 671 } 672 } 673 if (scol == 0) { 674 /* 675 We must have found a spared master! 676 We'll take that over anything else 677 found so far. (We couldn't have 678 found a real master before, since 679 this is a used spare, and it's 680 saying that it's replacing the 681 master.) On reboot (with 682 autoconfiguration turned on) 683 sparecol will become the 1st 684 component (component0) of this set. 685 */ 686 dumpto = sparecol; 687 break; 688 } else if (scol != -1) { 689 /* 690 Must be a spared slave. We'll dump 691 to that if we havn't found anything 692 else so far. 693 */ 694 if (dumpto == -1) 695 dumpto = sparecol; 696 } 697 } 698 } 699 700 if (dumpto == -1) { 701 /* we couldn't find any live components to dump to!?!? 702 */ 703 error = EINVAL; 704 goto out; 705 } 706 707 bdev = bdevsw_lookup(raidPtr->Disks[dumpto].dev); 708 709 /* 710 Note that blkno is relative to this particular partition. 711 By adding the offset of this partition in the RAID 712 set, and also adding RF_PROTECTED_SECTORS, we get a 713 value that is relative to the partition used for the 714 underlying component. 715 */ 716 717 error = (*bdev->d_dump)(raidPtr->Disks[dumpto].dev, 718 blkno + offset, va, size); 719 720 out: 721 raidunlock(rs); 722 723 return error; 724 } 725 /* ARGSUSED */ 726 int 727 raidopen(dev_t dev, int flags, int fmt, 728 struct lwp *l) 729 { 730 int unit = raidunit(dev); 731 struct raid_softc *rs; 732 struct disklabel *lp; 733 int part, pmask; 734 int error = 0; 735 736 if (unit >= numraid) 737 return (ENXIO); 738 rs = &raid_softc[unit]; 739 740 if ((error = raidlock(rs)) != 0) 741 return (error); 742 lp = rs->sc_dkdev.dk_label; 743 744 part = DISKPART(dev); 745 746 /* 747 * If there are wedges, and this is not RAW_PART, then we 748 * need to fail. 749 */ 750 if (rs->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) { 751 error = EBUSY; 752 goto bad; 753 } 754 pmask = (1 << part); 755 756 if ((rs->sc_flags & RAIDF_INITED) && 757 (rs->sc_dkdev.dk_openmask == 0)) 758 raidgetdisklabel(dev); 759 760 /* make sure that this partition exists */ 761 762 if (part != RAW_PART) { 763 if (((rs->sc_flags & RAIDF_INITED) == 0) || 764 ((part >= lp->d_npartitions) || 765 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 766 error = ENXIO; 767 goto bad; 768 } 769 } 770 /* Prevent this unit from being unconfigured while open. */ 771 switch (fmt) { 772 case S_IFCHR: 773 rs->sc_dkdev.dk_copenmask |= pmask; 774 break; 775 776 case S_IFBLK: 777 rs->sc_dkdev.dk_bopenmask |= pmask; 778 break; 779 } 780 781 if ((rs->sc_dkdev.dk_openmask == 0) && 782 ((rs->sc_flags & RAIDF_INITED) != 0)) { 783 /* First one... mark things as dirty... Note that we *MUST* 784 have done a configure before this. I DO NOT WANT TO BE 785 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 786 THAT THEY BELONG TOGETHER!!!!! */ 787 /* XXX should check to see if we're only open for reading 788 here... If so, we needn't do this, but then need some 789 other way of keeping track of what's happened.. */ 790 791 rf_markalldirty( raidPtrs[unit] ); 792 } 793 794 795 rs->sc_dkdev.dk_openmask = 796 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 797 798 bad: 799 raidunlock(rs); 800 801 return (error); 802 803 804 } 805 /* ARGSUSED */ 806 int 807 raidclose(dev_t dev, int flags, int fmt, struct lwp *l) 808 { 809 int unit = raidunit(dev); 810 struct cfdata *cf; 811 struct raid_softc *rs; 812 int error = 0; 813 int part; 814 815 if (unit >= numraid) 816 return (ENXIO); 817 rs = &raid_softc[unit]; 818 819 if ((error = raidlock(rs)) != 0) 820 return (error); 821 822 part = DISKPART(dev); 823 824 /* ...that much closer to allowing unconfiguration... */ 825 switch (fmt) { 826 case S_IFCHR: 827 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 828 break; 829 830 case S_IFBLK: 831 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 832 break; 833 } 834 rs->sc_dkdev.dk_openmask = 835 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 836 837 if ((rs->sc_dkdev.dk_openmask == 0) && 838 ((rs->sc_flags & RAIDF_INITED) != 0)) { 839 /* Last one... device is not unconfigured yet. 840 Device shutdown has taken care of setting the 841 clean bits if RAIDF_INITED is not set 842 mark things as clean... */ 843 844 rf_update_component_labels(raidPtrs[unit], 845 RF_FINAL_COMPONENT_UPDATE); 846 if (doing_shutdown) { 847 /* last one, and we're going down, so 848 lights out for this RAID set too. */ 849 error = rf_Shutdown(raidPtrs[unit]); 850 851 /* It's no longer initialized... */ 852 rs->sc_flags &= ~RAIDF_INITED; 853 854 /* detach the device */ 855 856 cf = device_cfdata(rs->sc_dev); 857 error = config_detach(rs->sc_dev, DETACH_QUIET); 858 free(cf, M_RAIDFRAME); 859 860 /* Detach the disk. */ 861 disk_detach(&rs->sc_dkdev); 862 disk_destroy(&rs->sc_dkdev); 863 } 864 } 865 866 raidunlock(rs); 867 return (0); 868 869 } 870 871 void 872 raidstrategy(struct buf *bp) 873 { 874 int s; 875 876 unsigned int raidID = raidunit(bp->b_dev); 877 RF_Raid_t *raidPtr; 878 struct raid_softc *rs = &raid_softc[raidID]; 879 int wlabel; 880 881 if ((rs->sc_flags & RAIDF_INITED) ==0) { 882 bp->b_error = ENXIO; 883 goto done; 884 } 885 if (raidID >= numraid || !raidPtrs[raidID]) { 886 bp->b_error = ENODEV; 887 goto done; 888 } 889 raidPtr = raidPtrs[raidID]; 890 if (!raidPtr->valid) { 891 bp->b_error = ENODEV; 892 goto done; 893 } 894 if (bp->b_bcount == 0) { 895 db1_printf(("b_bcount is zero..\n")); 896 goto done; 897 } 898 899 /* 900 * Do bounds checking and adjust transfer. If there's an 901 * error, the bounds check will flag that for us. 902 */ 903 904 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 905 if (DISKPART(bp->b_dev) == RAW_PART) { 906 uint64_t size; /* device size in DEV_BSIZE unit */ 907 908 if (raidPtr->logBytesPerSector > DEV_BSHIFT) { 909 size = raidPtr->totalSectors << 910 (raidPtr->logBytesPerSector - DEV_BSHIFT); 911 } else { 912 size = raidPtr->totalSectors >> 913 (DEV_BSHIFT - raidPtr->logBytesPerSector); 914 } 915 if (bounds_check_with_mediasize(bp, DEV_BSIZE, size) <= 0) { 916 goto done; 917 } 918 } else { 919 if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) { 920 db1_printf(("Bounds check failed!!:%d %d\n", 921 (int) bp->b_blkno, (int) wlabel)); 922 goto done; 923 } 924 } 925 s = splbio(); 926 927 bp->b_resid = 0; 928 929 /* stuff it onto our queue */ 930 bufq_put(rs->buf_queue, bp); 931 932 /* scheduled the IO to happen at the next convenient time */ 933 wakeup(&(raidPtrs[raidID]->iodone)); 934 935 splx(s); 936 return; 937 938 done: 939 bp->b_resid = bp->b_bcount; 940 biodone(bp); 941 } 942 /* ARGSUSED */ 943 int 944 raidread(dev_t dev, struct uio *uio, int flags) 945 { 946 int unit = raidunit(dev); 947 struct raid_softc *rs; 948 949 if (unit >= numraid) 950 return (ENXIO); 951 rs = &raid_softc[unit]; 952 953 if ((rs->sc_flags & RAIDF_INITED) == 0) 954 return (ENXIO); 955 956 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio)); 957 958 } 959 /* ARGSUSED */ 960 int 961 raidwrite(dev_t dev, struct uio *uio, int flags) 962 { 963 int unit = raidunit(dev); 964 struct raid_softc *rs; 965 966 if (unit >= numraid) 967 return (ENXIO); 968 rs = &raid_softc[unit]; 969 970 if ((rs->sc_flags & RAIDF_INITED) == 0) 971 return (ENXIO); 972 973 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio)); 974 975 } 976 977 int 978 raidioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 979 { 980 int unit = raidunit(dev); 981 int error = 0; 982 int part, pmask; 983 struct cfdata *cf; 984 struct raid_softc *rs; 985 RF_Config_t *k_cfg, *u_cfg; 986 RF_Raid_t *raidPtr; 987 RF_RaidDisk_t *diskPtr; 988 RF_AccTotals_t *totals; 989 RF_DeviceConfig_t *d_cfg, **ucfgp; 990 u_char *specific_buf; 991 int retcode = 0; 992 int column; 993 int raidid; 994 struct rf_recon_req *rrcopy, *rr; 995 RF_ComponentLabel_t *clabel; 996 RF_ComponentLabel_t *ci_label; 997 RF_ComponentLabel_t **clabel_ptr; 998 RF_SingleComponent_t *sparePtr,*componentPtr; 999 RF_SingleComponent_t component; 1000 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 1001 int i, j, d; 1002 #ifdef __HAVE_OLD_DISKLABEL 1003 struct disklabel newlabel; 1004 #endif 1005 struct dkwedge_info *dkw; 1006 1007 if (unit >= numraid) 1008 return (ENXIO); 1009 rs = &raid_softc[unit]; 1010 raidPtr = raidPtrs[unit]; 1011 1012 db1_printf(("raidioctl: %d %d %d %d\n", (int) dev, 1013 (int) DISKPART(dev), (int) unit, (int) cmd)); 1014 1015 /* Must be open for writes for these commands... */ 1016 switch (cmd) { 1017 #ifdef DIOCGSECTORSIZE 1018 case DIOCGSECTORSIZE: 1019 *(u_int *)data = raidPtr->bytesPerSector; 1020 return 0; 1021 case DIOCGMEDIASIZE: 1022 *(off_t *)data = 1023 (off_t)raidPtr->totalSectors * raidPtr->bytesPerSector; 1024 return 0; 1025 #endif 1026 case DIOCSDINFO: 1027 case DIOCWDINFO: 1028 #ifdef __HAVE_OLD_DISKLABEL 1029 case ODIOCWDINFO: 1030 case ODIOCSDINFO: 1031 #endif 1032 case DIOCWLABEL: 1033 case DIOCAWEDGE: 1034 case DIOCDWEDGE: 1035 if ((flag & FWRITE) == 0) 1036 return (EBADF); 1037 } 1038 1039 /* Must be initialized for these... */ 1040 switch (cmd) { 1041 case DIOCGDINFO: 1042 case DIOCSDINFO: 1043 case DIOCWDINFO: 1044 #ifdef __HAVE_OLD_DISKLABEL 1045 case ODIOCGDINFO: 1046 case ODIOCWDINFO: 1047 case ODIOCSDINFO: 1048 case ODIOCGDEFLABEL: 1049 #endif 1050 case DIOCGPART: 1051 case DIOCWLABEL: 1052 case DIOCGDEFLABEL: 1053 case DIOCAWEDGE: 1054 case DIOCDWEDGE: 1055 case DIOCLWEDGES: 1056 case DIOCCACHESYNC: 1057 case RAIDFRAME_SHUTDOWN: 1058 case RAIDFRAME_REWRITEPARITY: 1059 case RAIDFRAME_GET_INFO: 1060 case RAIDFRAME_RESET_ACCTOTALS: 1061 case RAIDFRAME_GET_ACCTOTALS: 1062 case RAIDFRAME_KEEP_ACCTOTALS: 1063 case RAIDFRAME_GET_SIZE: 1064 case RAIDFRAME_FAIL_DISK: 1065 case RAIDFRAME_COPYBACK: 1066 case RAIDFRAME_CHECK_RECON_STATUS: 1067 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1068 case RAIDFRAME_GET_COMPONENT_LABEL: 1069 case RAIDFRAME_SET_COMPONENT_LABEL: 1070 case RAIDFRAME_ADD_HOT_SPARE: 1071 case RAIDFRAME_REMOVE_HOT_SPARE: 1072 case RAIDFRAME_INIT_LABELS: 1073 case RAIDFRAME_REBUILD_IN_PLACE: 1074 case RAIDFRAME_CHECK_PARITY: 1075 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1076 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1077 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1078 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1079 case RAIDFRAME_SET_AUTOCONFIG: 1080 case RAIDFRAME_SET_ROOT: 1081 case RAIDFRAME_DELETE_COMPONENT: 1082 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1083 if ((rs->sc_flags & RAIDF_INITED) == 0) 1084 return (ENXIO); 1085 } 1086 1087 switch (cmd) { 1088 #ifdef COMPAT_50 1089 case RAIDFRAME_GET_INFO50: 1090 return rf_get_info50(raidPtr, data); 1091 1092 case RAIDFRAME_CONFIGURE50: 1093 if ((retcode = rf_config50(raidPtr, unit, data, &k_cfg)) != 0) 1094 return retcode; 1095 goto config; 1096 #endif 1097 /* configure the system */ 1098 case RAIDFRAME_CONFIGURE: 1099 1100 if (raidPtr->valid) { 1101 /* There is a valid RAID set running on this unit! */ 1102 printf("raid%d: Device already configured!\n",unit); 1103 return(EINVAL); 1104 } 1105 1106 /* copy-in the configuration information */ 1107 /* data points to a pointer to the configuration structure */ 1108 1109 u_cfg = *((RF_Config_t **) data); 1110 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *)); 1111 if (k_cfg == NULL) { 1112 return (ENOMEM); 1113 } 1114 retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t)); 1115 if (retcode) { 1116 RF_Free(k_cfg, sizeof(RF_Config_t)); 1117 db1_printf(("rf_ioctl: retcode=%d copyin.1\n", 1118 retcode)); 1119 return (retcode); 1120 } 1121 goto config; 1122 config: 1123 /* allocate a buffer for the layout-specific data, and copy it 1124 * in */ 1125 if (k_cfg->layoutSpecificSize) { 1126 if (k_cfg->layoutSpecificSize > 10000) { 1127 /* sanity check */ 1128 RF_Free(k_cfg, sizeof(RF_Config_t)); 1129 return (EINVAL); 1130 } 1131 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize, 1132 (u_char *)); 1133 if (specific_buf == NULL) { 1134 RF_Free(k_cfg, sizeof(RF_Config_t)); 1135 return (ENOMEM); 1136 } 1137 retcode = copyin(k_cfg->layoutSpecific, specific_buf, 1138 k_cfg->layoutSpecificSize); 1139 if (retcode) { 1140 RF_Free(k_cfg, sizeof(RF_Config_t)); 1141 RF_Free(specific_buf, 1142 k_cfg->layoutSpecificSize); 1143 db1_printf(("rf_ioctl: retcode=%d copyin.2\n", 1144 retcode)); 1145 return (retcode); 1146 } 1147 } else 1148 specific_buf = NULL; 1149 k_cfg->layoutSpecific = specific_buf; 1150 1151 /* should do some kind of sanity check on the configuration. 1152 * Store the sum of all the bytes in the last byte? */ 1153 1154 /* configure the system */ 1155 1156 /* 1157 * Clear the entire RAID descriptor, just to make sure 1158 * there is no stale data left in the case of a 1159 * reconfiguration 1160 */ 1161 memset((char *) raidPtr, 0, sizeof(RF_Raid_t)); 1162 raidPtr->raidid = unit; 1163 1164 retcode = rf_Configure(raidPtr, k_cfg, NULL); 1165 1166 if (retcode == 0) { 1167 1168 /* allow this many simultaneous IO's to 1169 this RAID device */ 1170 raidPtr->openings = RAIDOUTSTANDING; 1171 1172 raidinit(raidPtr); 1173 rf_markalldirty(raidPtr); 1174 } 1175 /* free the buffers. No return code here. */ 1176 if (k_cfg->layoutSpecificSize) { 1177 RF_Free(specific_buf, k_cfg->layoutSpecificSize); 1178 } 1179 RF_Free(k_cfg, sizeof(RF_Config_t)); 1180 1181 return (retcode); 1182 1183 /* shutdown the system */ 1184 case RAIDFRAME_SHUTDOWN: 1185 1186 if ((error = raidlock(rs)) != 0) 1187 return (error); 1188 1189 /* 1190 * If somebody has a partition mounted, we shouldn't 1191 * shutdown. 1192 */ 1193 1194 part = DISKPART(dev); 1195 pmask = (1 << part); 1196 if ((rs->sc_dkdev.dk_openmask & ~pmask) || 1197 ((rs->sc_dkdev.dk_bopenmask & pmask) && 1198 (rs->sc_dkdev.dk_copenmask & pmask))) { 1199 raidunlock(rs); 1200 return (EBUSY); 1201 } 1202 1203 retcode = rf_Shutdown(raidPtr); 1204 1205 /* It's no longer initialized... */ 1206 rs->sc_flags &= ~RAIDF_INITED; 1207 1208 /* free the pseudo device attach bits */ 1209 1210 cf = device_cfdata(rs->sc_dev); 1211 /* XXX this causes us to not return any errors 1212 from the above call to rf_Shutdown() */ 1213 retcode = config_detach(rs->sc_dev, DETACH_QUIET); 1214 free(cf, M_RAIDFRAME); 1215 1216 /* Detach the disk. */ 1217 disk_detach(&rs->sc_dkdev); 1218 disk_destroy(&rs->sc_dkdev); 1219 1220 raidunlock(rs); 1221 1222 return (retcode); 1223 case RAIDFRAME_GET_COMPONENT_LABEL: 1224 clabel_ptr = (RF_ComponentLabel_t **) data; 1225 /* need to read the component label for the disk indicated 1226 by row,column in clabel */ 1227 1228 /* For practice, let's get it directly fromdisk, rather 1229 than from the in-core copy */ 1230 RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ), 1231 (RF_ComponentLabel_t *)); 1232 if (clabel == NULL) 1233 return (ENOMEM); 1234 1235 retcode = copyin( *clabel_ptr, clabel, 1236 sizeof(RF_ComponentLabel_t)); 1237 1238 if (retcode) { 1239 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1240 return(retcode); 1241 } 1242 1243 clabel->row = 0; /* Don't allow looking at anything else.*/ 1244 1245 column = clabel->column; 1246 1247 if ((column < 0) || (column >= raidPtr->numCol + 1248 raidPtr->numSpare)) { 1249 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1250 return(EINVAL); 1251 } 1252 1253 retcode = raidread_component_label(raidPtr->Disks[column].dev, 1254 raidPtr->raid_cinfo[column].ci_vp, 1255 clabel ); 1256 1257 if (retcode == 0) { 1258 retcode = copyout(clabel, *clabel_ptr, 1259 sizeof(RF_ComponentLabel_t)); 1260 } 1261 RF_Free(clabel, sizeof(RF_ComponentLabel_t)); 1262 return (retcode); 1263 1264 case RAIDFRAME_SET_COMPONENT_LABEL: 1265 clabel = (RF_ComponentLabel_t *) data; 1266 1267 /* XXX check the label for valid stuff... */ 1268 /* Note that some things *should not* get modified -- 1269 the user should be re-initing the labels instead of 1270 trying to patch things. 1271 */ 1272 1273 raidid = raidPtr->raidid; 1274 #ifdef DEBUG 1275 printf("raid%d: Got component label:\n", raidid); 1276 printf("raid%d: Version: %d\n", raidid, clabel->version); 1277 printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number); 1278 printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter); 1279 printf("raid%d: Column: %d\n", raidid, clabel->column); 1280 printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns); 1281 printf("raid%d: Clean: %d\n", raidid, clabel->clean); 1282 printf("raid%d: Status: %d\n", raidid, clabel->status); 1283 #endif 1284 clabel->row = 0; 1285 column = clabel->column; 1286 1287 if ((column < 0) || (column >= raidPtr->numCol)) { 1288 return(EINVAL); 1289 } 1290 1291 /* XXX this isn't allowed to do anything for now :-) */ 1292 1293 /* XXX and before it is, we need to fill in the rest 1294 of the fields!?!?!?! */ 1295 #if 0 1296 raidwrite_component_label( 1297 raidPtr->Disks[column].dev, 1298 raidPtr->raid_cinfo[column].ci_vp, 1299 clabel ); 1300 #endif 1301 return (0); 1302 1303 case RAIDFRAME_INIT_LABELS: 1304 clabel = (RF_ComponentLabel_t *) data; 1305 /* 1306 we only want the serial number from 1307 the above. We get all the rest of the information 1308 from the config that was used to create this RAID 1309 set. 1310 */ 1311 1312 raidPtr->serial_number = clabel->serial_number; 1313 1314 RF_Malloc(ci_label, sizeof(RF_ComponentLabel_t), 1315 (RF_ComponentLabel_t *)); 1316 if (ci_label == NULL) 1317 return (ENOMEM); 1318 1319 raid_init_component_label(raidPtr, ci_label); 1320 ci_label->serial_number = clabel->serial_number; 1321 ci_label->row = 0; /* we dont' pretend to support more */ 1322 1323 for(column=0;column<raidPtr->numCol;column++) { 1324 diskPtr = &raidPtr->Disks[column]; 1325 if (!RF_DEAD_DISK(diskPtr->status)) { 1326 ci_label->partitionSize = diskPtr->partitionSize; 1327 ci_label->column = column; 1328 raidwrite_component_label( 1329 raidPtr->Disks[column].dev, 1330 raidPtr->raid_cinfo[column].ci_vp, 1331 ci_label ); 1332 } 1333 } 1334 RF_Free(ci_label, sizeof(RF_ComponentLabel_t)); 1335 1336 return (retcode); 1337 case RAIDFRAME_SET_AUTOCONFIG: 1338 d = rf_set_autoconfig(raidPtr, *(int *) data); 1339 printf("raid%d: New autoconfig value is: %d\n", 1340 raidPtr->raidid, d); 1341 *(int *) data = d; 1342 return (retcode); 1343 1344 case RAIDFRAME_SET_ROOT: 1345 d = rf_set_rootpartition(raidPtr, *(int *) data); 1346 printf("raid%d: New rootpartition value is: %d\n", 1347 raidPtr->raidid, d); 1348 *(int *) data = d; 1349 return (retcode); 1350 1351 /* initialize all parity */ 1352 case RAIDFRAME_REWRITEPARITY: 1353 1354 if (raidPtr->Layout.map->faultsTolerated == 0) { 1355 /* Parity for RAID 0 is trivially correct */ 1356 raidPtr->parity_good = RF_RAID_CLEAN; 1357 return(0); 1358 } 1359 1360 if (raidPtr->parity_rewrite_in_progress == 1) { 1361 /* Re-write is already in progress! */ 1362 return(EINVAL); 1363 } 1364 1365 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1366 rf_RewriteParityThread, 1367 raidPtr,"raid_parity"); 1368 return (retcode); 1369 1370 1371 case RAIDFRAME_ADD_HOT_SPARE: 1372 sparePtr = (RF_SingleComponent_t *) data; 1373 memcpy( &component, sparePtr, sizeof(RF_SingleComponent_t)); 1374 retcode = rf_add_hot_spare(raidPtr, &component); 1375 return(retcode); 1376 1377 case RAIDFRAME_REMOVE_HOT_SPARE: 1378 return(retcode); 1379 1380 case RAIDFRAME_DELETE_COMPONENT: 1381 componentPtr = (RF_SingleComponent_t *)data; 1382 memcpy( &component, componentPtr, 1383 sizeof(RF_SingleComponent_t)); 1384 retcode = rf_delete_component(raidPtr, &component); 1385 return(retcode); 1386 1387 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1388 componentPtr = (RF_SingleComponent_t *)data; 1389 memcpy( &component, componentPtr, 1390 sizeof(RF_SingleComponent_t)); 1391 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1392 return(retcode); 1393 1394 case RAIDFRAME_REBUILD_IN_PLACE: 1395 1396 if (raidPtr->Layout.map->faultsTolerated == 0) { 1397 /* Can't do this on a RAID 0!! */ 1398 return(EINVAL); 1399 } 1400 1401 if (raidPtr->recon_in_progress == 1) { 1402 /* a reconstruct is already in progress! */ 1403 return(EINVAL); 1404 } 1405 1406 componentPtr = (RF_SingleComponent_t *) data; 1407 memcpy( &component, componentPtr, 1408 sizeof(RF_SingleComponent_t)); 1409 component.row = 0; /* we don't support any more */ 1410 column = component.column; 1411 1412 if ((column < 0) || (column >= raidPtr->numCol)) { 1413 return(EINVAL); 1414 } 1415 1416 RF_LOCK_MUTEX(raidPtr->mutex); 1417 if ((raidPtr->Disks[column].status == rf_ds_optimal) && 1418 (raidPtr->numFailures > 0)) { 1419 /* XXX 0 above shouldn't be constant!!! */ 1420 /* some component other than this has failed. 1421 Let's not make things worse than they already 1422 are... */ 1423 printf("raid%d: Unable to reconstruct to disk at:\n", 1424 raidPtr->raidid); 1425 printf("raid%d: Col: %d Too many failures.\n", 1426 raidPtr->raidid, column); 1427 RF_UNLOCK_MUTEX(raidPtr->mutex); 1428 return (EINVAL); 1429 } 1430 if (raidPtr->Disks[column].status == 1431 rf_ds_reconstructing) { 1432 printf("raid%d: Unable to reconstruct to disk at:\n", 1433 raidPtr->raidid); 1434 printf("raid%d: Col: %d Reconstruction already occuring!\n", raidPtr->raidid, column); 1435 1436 RF_UNLOCK_MUTEX(raidPtr->mutex); 1437 return (EINVAL); 1438 } 1439 if (raidPtr->Disks[column].status == rf_ds_spared) { 1440 RF_UNLOCK_MUTEX(raidPtr->mutex); 1441 return (EINVAL); 1442 } 1443 RF_UNLOCK_MUTEX(raidPtr->mutex); 1444 1445 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1446 if (rrcopy == NULL) 1447 return(ENOMEM); 1448 1449 rrcopy->raidPtr = (void *) raidPtr; 1450 rrcopy->col = column; 1451 1452 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1453 rf_ReconstructInPlaceThread, 1454 rrcopy,"raid_reconip"); 1455 return(retcode); 1456 1457 case RAIDFRAME_GET_INFO: 1458 if (!raidPtr->valid) 1459 return (ENODEV); 1460 ucfgp = (RF_DeviceConfig_t **) data; 1461 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1462 (RF_DeviceConfig_t *)); 1463 if (d_cfg == NULL) 1464 return (ENOMEM); 1465 d_cfg->rows = 1; /* there is only 1 row now */ 1466 d_cfg->cols = raidPtr->numCol; 1467 d_cfg->ndevs = raidPtr->numCol; 1468 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1469 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1470 return (ENOMEM); 1471 } 1472 d_cfg->nspares = raidPtr->numSpare; 1473 if (d_cfg->nspares >= RF_MAX_DISKS) { 1474 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1475 return (ENOMEM); 1476 } 1477 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1478 d = 0; 1479 for (j = 0; j < d_cfg->cols; j++) { 1480 d_cfg->devs[d] = raidPtr->Disks[j]; 1481 d++; 1482 } 1483 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1484 d_cfg->spares[i] = raidPtr->Disks[j]; 1485 } 1486 retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t)); 1487 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1488 1489 return (retcode); 1490 1491 case RAIDFRAME_CHECK_PARITY: 1492 *(int *) data = raidPtr->parity_good; 1493 return (0); 1494 1495 case RAIDFRAME_RESET_ACCTOTALS: 1496 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1497 return (0); 1498 1499 case RAIDFRAME_GET_ACCTOTALS: 1500 totals = (RF_AccTotals_t *) data; 1501 *totals = raidPtr->acc_totals; 1502 return (0); 1503 1504 case RAIDFRAME_KEEP_ACCTOTALS: 1505 raidPtr->keep_acc_totals = *(int *)data; 1506 return (0); 1507 1508 case RAIDFRAME_GET_SIZE: 1509 *(int *) data = raidPtr->totalSectors; 1510 return (0); 1511 1512 /* fail a disk & optionally start reconstruction */ 1513 case RAIDFRAME_FAIL_DISK: 1514 1515 if (raidPtr->Layout.map->faultsTolerated == 0) { 1516 /* Can't do this on a RAID 0!! */ 1517 return(EINVAL); 1518 } 1519 1520 rr = (struct rf_recon_req *) data; 1521 rr->row = 0; 1522 if (rr->col < 0 || rr->col >= raidPtr->numCol) 1523 return (EINVAL); 1524 1525 1526 RF_LOCK_MUTEX(raidPtr->mutex); 1527 if (raidPtr->status == rf_rs_reconstructing) { 1528 /* you can't fail a disk while we're reconstructing! */ 1529 /* XXX wrong for RAID6 */ 1530 RF_UNLOCK_MUTEX(raidPtr->mutex); 1531 return (EINVAL); 1532 } 1533 if ((raidPtr->Disks[rr->col].status == 1534 rf_ds_optimal) && (raidPtr->numFailures > 0)) { 1535 /* some other component has failed. Let's not make 1536 things worse. XXX wrong for RAID6 */ 1537 RF_UNLOCK_MUTEX(raidPtr->mutex); 1538 return (EINVAL); 1539 } 1540 if (raidPtr->Disks[rr->col].status == rf_ds_spared) { 1541 /* Can't fail a spared disk! */ 1542 RF_UNLOCK_MUTEX(raidPtr->mutex); 1543 return (EINVAL); 1544 } 1545 RF_UNLOCK_MUTEX(raidPtr->mutex); 1546 1547 /* make a copy of the recon request so that we don't rely on 1548 * the user's buffer */ 1549 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1550 if (rrcopy == NULL) 1551 return(ENOMEM); 1552 memcpy(rrcopy, rr, sizeof(*rr)); 1553 rrcopy->raidPtr = (void *) raidPtr; 1554 1555 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1556 rf_ReconThread, 1557 rrcopy,"raid_recon"); 1558 return (0); 1559 1560 /* invoke a copyback operation after recon on whatever disk 1561 * needs it, if any */ 1562 case RAIDFRAME_COPYBACK: 1563 1564 if (raidPtr->Layout.map->faultsTolerated == 0) { 1565 /* This makes no sense on a RAID 0!! */ 1566 return(EINVAL); 1567 } 1568 1569 if (raidPtr->copyback_in_progress == 1) { 1570 /* Copyback is already in progress! */ 1571 return(EINVAL); 1572 } 1573 1574 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1575 rf_CopybackThread, 1576 raidPtr,"raid_copyback"); 1577 return (retcode); 1578 1579 /* return the percentage completion of reconstruction */ 1580 case RAIDFRAME_CHECK_RECON_STATUS: 1581 if (raidPtr->Layout.map->faultsTolerated == 0) { 1582 /* This makes no sense on a RAID 0, so tell the 1583 user it's done. */ 1584 *(int *) data = 100; 1585 return(0); 1586 } 1587 if (raidPtr->status != rf_rs_reconstructing) 1588 *(int *) data = 100; 1589 else { 1590 if (raidPtr->reconControl->numRUsTotal > 0) { 1591 *(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal); 1592 } else { 1593 *(int *) data = 0; 1594 } 1595 } 1596 return (0); 1597 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1598 progressInfoPtr = (RF_ProgressInfo_t **) data; 1599 if (raidPtr->status != rf_rs_reconstructing) { 1600 progressInfo.remaining = 0; 1601 progressInfo.completed = 100; 1602 progressInfo.total = 100; 1603 } else { 1604 progressInfo.total = 1605 raidPtr->reconControl->numRUsTotal; 1606 progressInfo.completed = 1607 raidPtr->reconControl->numRUsComplete; 1608 progressInfo.remaining = progressInfo.total - 1609 progressInfo.completed; 1610 } 1611 retcode = copyout(&progressInfo, *progressInfoPtr, 1612 sizeof(RF_ProgressInfo_t)); 1613 return (retcode); 1614 1615 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1616 if (raidPtr->Layout.map->faultsTolerated == 0) { 1617 /* This makes no sense on a RAID 0, so tell the 1618 user it's done. */ 1619 *(int *) data = 100; 1620 return(0); 1621 } 1622 if (raidPtr->parity_rewrite_in_progress == 1) { 1623 *(int *) data = 100 * 1624 raidPtr->parity_rewrite_stripes_done / 1625 raidPtr->Layout.numStripe; 1626 } else { 1627 *(int *) data = 100; 1628 } 1629 return (0); 1630 1631 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1632 progressInfoPtr = (RF_ProgressInfo_t **) data; 1633 if (raidPtr->parity_rewrite_in_progress == 1) { 1634 progressInfo.total = raidPtr->Layout.numStripe; 1635 progressInfo.completed = 1636 raidPtr->parity_rewrite_stripes_done; 1637 progressInfo.remaining = progressInfo.total - 1638 progressInfo.completed; 1639 } else { 1640 progressInfo.remaining = 0; 1641 progressInfo.completed = 100; 1642 progressInfo.total = 100; 1643 } 1644 retcode = copyout(&progressInfo, *progressInfoPtr, 1645 sizeof(RF_ProgressInfo_t)); 1646 return (retcode); 1647 1648 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1649 if (raidPtr->Layout.map->faultsTolerated == 0) { 1650 /* This makes no sense on a RAID 0 */ 1651 *(int *) data = 100; 1652 return(0); 1653 } 1654 if (raidPtr->copyback_in_progress == 1) { 1655 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1656 raidPtr->Layout.numStripe; 1657 } else { 1658 *(int *) data = 100; 1659 } 1660 return (0); 1661 1662 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1663 progressInfoPtr = (RF_ProgressInfo_t **) data; 1664 if (raidPtr->copyback_in_progress == 1) { 1665 progressInfo.total = raidPtr->Layout.numStripe; 1666 progressInfo.completed = 1667 raidPtr->copyback_stripes_done; 1668 progressInfo.remaining = progressInfo.total - 1669 progressInfo.completed; 1670 } else { 1671 progressInfo.remaining = 0; 1672 progressInfo.completed = 100; 1673 progressInfo.total = 100; 1674 } 1675 retcode = copyout(&progressInfo, *progressInfoPtr, 1676 sizeof(RF_ProgressInfo_t)); 1677 return (retcode); 1678 1679 /* the sparetable daemon calls this to wait for the kernel to 1680 * need a spare table. this ioctl does not return until a 1681 * spare table is needed. XXX -- calling mpsleep here in the 1682 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1683 * -- I should either compute the spare table in the kernel, 1684 * or have a different -- XXX XXX -- interface (a different 1685 * character device) for delivering the table -- XXX */ 1686 #if 0 1687 case RAIDFRAME_SPARET_WAIT: 1688 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1689 while (!rf_sparet_wait_queue) 1690 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE); 1691 waitreq = rf_sparet_wait_queue; 1692 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1693 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1694 1695 /* structure assignment */ 1696 *((RF_SparetWait_t *) data) = *waitreq; 1697 1698 RF_Free(waitreq, sizeof(*waitreq)); 1699 return (0); 1700 1701 /* wakes up a process waiting on SPARET_WAIT and puts an error 1702 * code in it that will cause the dameon to exit */ 1703 case RAIDFRAME_ABORT_SPARET_WAIT: 1704 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1705 waitreq->fcol = -1; 1706 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1707 waitreq->next = rf_sparet_wait_queue; 1708 rf_sparet_wait_queue = waitreq; 1709 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1710 wakeup(&rf_sparet_wait_queue); 1711 return (0); 1712 1713 /* used by the spare table daemon to deliver a spare table 1714 * into the kernel */ 1715 case RAIDFRAME_SEND_SPARET: 1716 1717 /* install the spare table */ 1718 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1719 1720 /* respond to the requestor. the return status of the spare 1721 * table installation is passed in the "fcol" field */ 1722 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1723 waitreq->fcol = retcode; 1724 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1725 waitreq->next = rf_sparet_resp_queue; 1726 rf_sparet_resp_queue = waitreq; 1727 wakeup(&rf_sparet_resp_queue); 1728 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1729 1730 return (retcode); 1731 #endif 1732 1733 default: 1734 break; /* fall through to the os-specific code below */ 1735 1736 } 1737 1738 if (!raidPtr->valid) 1739 return (EINVAL); 1740 1741 /* 1742 * Add support for "regular" device ioctls here. 1743 */ 1744 1745 switch (cmd) { 1746 case DIOCGDINFO: 1747 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1748 break; 1749 #ifdef __HAVE_OLD_DISKLABEL 1750 case ODIOCGDINFO: 1751 newlabel = *(rs->sc_dkdev.dk_label); 1752 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1753 return ENOTTY; 1754 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1755 break; 1756 #endif 1757 1758 case DIOCGPART: 1759 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1760 ((struct partinfo *) data)->part = 1761 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1762 break; 1763 1764 case DIOCWDINFO: 1765 case DIOCSDINFO: 1766 #ifdef __HAVE_OLD_DISKLABEL 1767 case ODIOCWDINFO: 1768 case ODIOCSDINFO: 1769 #endif 1770 { 1771 struct disklabel *lp; 1772 #ifdef __HAVE_OLD_DISKLABEL 1773 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1774 memset(&newlabel, 0, sizeof newlabel); 1775 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1776 lp = &newlabel; 1777 } else 1778 #endif 1779 lp = (struct disklabel *)data; 1780 1781 if ((error = raidlock(rs)) != 0) 1782 return (error); 1783 1784 rs->sc_flags |= RAIDF_LABELLING; 1785 1786 error = setdisklabel(rs->sc_dkdev.dk_label, 1787 lp, 0, rs->sc_dkdev.dk_cpulabel); 1788 if (error == 0) { 1789 if (cmd == DIOCWDINFO 1790 #ifdef __HAVE_OLD_DISKLABEL 1791 || cmd == ODIOCWDINFO 1792 #endif 1793 ) 1794 error = writedisklabel(RAIDLABELDEV(dev), 1795 raidstrategy, rs->sc_dkdev.dk_label, 1796 rs->sc_dkdev.dk_cpulabel); 1797 } 1798 rs->sc_flags &= ~RAIDF_LABELLING; 1799 1800 raidunlock(rs); 1801 1802 if (error) 1803 return (error); 1804 break; 1805 } 1806 1807 case DIOCWLABEL: 1808 if (*(int *) data != 0) 1809 rs->sc_flags |= RAIDF_WLABEL; 1810 else 1811 rs->sc_flags &= ~RAIDF_WLABEL; 1812 break; 1813 1814 case DIOCGDEFLABEL: 1815 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1816 break; 1817 1818 #ifdef __HAVE_OLD_DISKLABEL 1819 case ODIOCGDEFLABEL: 1820 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1821 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1822 return ENOTTY; 1823 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1824 break; 1825 #endif 1826 1827 case DIOCAWEDGE: 1828 case DIOCDWEDGE: 1829 dkw = (void *)data; 1830 1831 /* If the ioctl happens here, the parent is us. */ 1832 (void)strcpy(dkw->dkw_parent, rs->sc_xname); 1833 return cmd == DIOCAWEDGE ? dkwedge_add(dkw) : dkwedge_del(dkw); 1834 1835 case DIOCLWEDGES: 1836 return dkwedge_list(&rs->sc_dkdev, 1837 (struct dkwedge_list *)data, l); 1838 case DIOCCACHESYNC: 1839 return rf_sync_component_caches(raidPtr); 1840 default: 1841 retcode = ENOTTY; 1842 } 1843 return (retcode); 1844 1845 } 1846 1847 1848 /* raidinit -- complete the rest of the initialization for the 1849 RAIDframe device. */ 1850 1851 1852 static void 1853 raidinit(RF_Raid_t *raidPtr) 1854 { 1855 struct cfdata *cf; 1856 struct raid_softc *rs; 1857 int unit; 1858 1859 unit = raidPtr->raidid; 1860 1861 rs = &raid_softc[unit]; 1862 1863 /* XXX should check return code first... */ 1864 rs->sc_flags |= RAIDF_INITED; 1865 1866 /* XXX doesn't check bounds. */ 1867 snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit); 1868 1869 /* attach the pseudo device */ 1870 cf = malloc(sizeof(*cf), M_RAIDFRAME, M_WAITOK); 1871 cf->cf_name = raid_cd.cd_name; 1872 cf->cf_atname = raid_cd.cd_name; 1873 cf->cf_unit = unit; 1874 cf->cf_fstate = FSTATE_STAR; 1875 1876 rs->sc_dev = config_attach_pseudo(cf); 1877 1878 if (rs->sc_dev==NULL) { 1879 printf("raid%d: config_attach_pseudo failed\n", 1880 raidPtr->raidid); 1881 } 1882 1883 /* disk_attach actually creates space for the CPU disklabel, among 1884 * other things, so it's critical to call this *BEFORE* we try putzing 1885 * with disklabels. */ 1886 1887 disk_init(&rs->sc_dkdev, rs->sc_xname, &rf_dkdriver); 1888 disk_attach(&rs->sc_dkdev); 1889 1890 /* XXX There may be a weird interaction here between this, and 1891 * protectedSectors, as used in RAIDframe. */ 1892 1893 rs->sc_size = raidPtr->totalSectors; 1894 1895 dkwedge_discover(&rs->sc_dkdev); 1896 1897 rf_set_properties(rs, raidPtr); 1898 1899 } 1900 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 1901 /* wake up the daemon & tell it to get us a spare table 1902 * XXX 1903 * the entries in the queues should be tagged with the raidPtr 1904 * so that in the extremely rare case that two recons happen at once, 1905 * we know for which device were requesting a spare table 1906 * XXX 1907 * 1908 * XXX This code is not currently used. GO 1909 */ 1910 int 1911 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req) 1912 { 1913 int retcode; 1914 1915 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1916 req->next = rf_sparet_wait_queue; 1917 rf_sparet_wait_queue = req; 1918 wakeup(&rf_sparet_wait_queue); 1919 1920 /* mpsleep unlocks the mutex */ 1921 while (!rf_sparet_resp_queue) { 1922 tsleep(&rf_sparet_resp_queue, PRIBIO, 1923 "raidframe getsparetable", 0); 1924 } 1925 req = rf_sparet_resp_queue; 1926 rf_sparet_resp_queue = req->next; 1927 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1928 1929 retcode = req->fcol; 1930 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1931 * alloc'd */ 1932 return (retcode); 1933 } 1934 #endif 1935 1936 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1937 * bp & passes it down. 1938 * any calls originating in the kernel must use non-blocking I/O 1939 * do some extra sanity checking to return "appropriate" error values for 1940 * certain conditions (to make some standard utilities work) 1941 * 1942 * Formerly known as: rf_DoAccessKernel 1943 */ 1944 void 1945 raidstart(RF_Raid_t *raidPtr) 1946 { 1947 RF_SectorCount_t num_blocks, pb, sum; 1948 RF_RaidAddr_t raid_addr; 1949 struct partition *pp; 1950 daddr_t blocknum; 1951 int unit; 1952 struct raid_softc *rs; 1953 int do_async; 1954 struct buf *bp; 1955 int rc; 1956 1957 unit = raidPtr->raidid; 1958 rs = &raid_softc[unit]; 1959 1960 /* quick check to see if anything has died recently */ 1961 RF_LOCK_MUTEX(raidPtr->mutex); 1962 if (raidPtr->numNewFailures > 0) { 1963 RF_UNLOCK_MUTEX(raidPtr->mutex); 1964 rf_update_component_labels(raidPtr, 1965 RF_NORMAL_COMPONENT_UPDATE); 1966 RF_LOCK_MUTEX(raidPtr->mutex); 1967 raidPtr->numNewFailures--; 1968 } 1969 1970 /* Check to see if we're at the limit... */ 1971 while (raidPtr->openings > 0) { 1972 RF_UNLOCK_MUTEX(raidPtr->mutex); 1973 1974 /* get the next item, if any, from the queue */ 1975 if ((bp = bufq_get(rs->buf_queue)) == NULL) { 1976 /* nothing more to do */ 1977 return; 1978 } 1979 1980 /* Ok, for the bp we have here, bp->b_blkno is relative to the 1981 * partition.. Need to make it absolute to the underlying 1982 * device.. */ 1983 1984 blocknum = bp->b_blkno; 1985 if (DISKPART(bp->b_dev) != RAW_PART) { 1986 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 1987 blocknum += pp->p_offset; 1988 } 1989 1990 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 1991 (int) blocknum)); 1992 1993 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 1994 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 1995 1996 /* *THIS* is where we adjust what block we're going to... 1997 * but DO NOT TOUCH bp->b_blkno!!! */ 1998 raid_addr = blocknum; 1999 2000 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 2001 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 2002 sum = raid_addr + num_blocks + pb; 2003 if (1 || rf_debugKernelAccess) { 2004 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 2005 (int) raid_addr, (int) sum, (int) num_blocks, 2006 (int) pb, (int) bp->b_resid)); 2007 } 2008 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 2009 || (sum < num_blocks) || (sum < pb)) { 2010 bp->b_error = ENOSPC; 2011 bp->b_resid = bp->b_bcount; 2012 biodone(bp); 2013 RF_LOCK_MUTEX(raidPtr->mutex); 2014 continue; 2015 } 2016 /* 2017 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 2018 */ 2019 2020 if (bp->b_bcount & raidPtr->sectorMask) { 2021 bp->b_error = EINVAL; 2022 bp->b_resid = bp->b_bcount; 2023 biodone(bp); 2024 RF_LOCK_MUTEX(raidPtr->mutex); 2025 continue; 2026 2027 } 2028 db1_printf(("Calling DoAccess..\n")); 2029 2030 2031 RF_LOCK_MUTEX(raidPtr->mutex); 2032 raidPtr->openings--; 2033 RF_UNLOCK_MUTEX(raidPtr->mutex); 2034 2035 /* 2036 * Everything is async. 2037 */ 2038 do_async = 1; 2039 2040 disk_busy(&rs->sc_dkdev); 2041 2042 /* XXX we're still at splbio() here... do we *really* 2043 need to be? */ 2044 2045 /* don't ever condition on bp->b_flags & B_WRITE. 2046 * always condition on B_READ instead */ 2047 2048 rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 2049 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 2050 do_async, raid_addr, num_blocks, 2051 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 2052 2053 if (rc) { 2054 bp->b_error = rc; 2055 bp->b_resid = bp->b_bcount; 2056 biodone(bp); 2057 /* continue loop */ 2058 } 2059 2060 RF_LOCK_MUTEX(raidPtr->mutex); 2061 } 2062 RF_UNLOCK_MUTEX(raidPtr->mutex); 2063 } 2064 2065 2066 2067 2068 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 2069 2070 int 2071 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req) 2072 { 2073 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 2074 struct buf *bp; 2075 2076 req->queue = queue; 2077 2078 #if DIAGNOSTIC 2079 if (queue->raidPtr->raidid >= numraid) { 2080 printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid, 2081 numraid); 2082 panic("Invalid Unit number in rf_DispatchKernelIO"); 2083 } 2084 #endif 2085 2086 bp = req->bp; 2087 2088 switch (req->type) { 2089 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 2090 /* XXX need to do something extra here.. */ 2091 /* I'm leaving this in, as I've never actually seen it used, 2092 * and I'd like folks to report it... GO */ 2093 printf(("WAKEUP CALLED\n")); 2094 queue->numOutstanding++; 2095 2096 bp->b_flags = 0; 2097 bp->b_private = req; 2098 2099 KernelWakeupFunc(bp); 2100 break; 2101 2102 case RF_IO_TYPE_READ: 2103 case RF_IO_TYPE_WRITE: 2104 #if RF_ACC_TRACE > 0 2105 if (req->tracerec) { 2106 RF_ETIMER_START(req->tracerec->timer); 2107 } 2108 #endif 2109 InitBP(bp, queue->rf_cinfo->ci_vp, 2110 op, queue->rf_cinfo->ci_dev, 2111 req->sectorOffset, req->numSector, 2112 req->buf, KernelWakeupFunc, (void *) req, 2113 queue->raidPtr->logBytesPerSector, req->b_proc); 2114 2115 if (rf_debugKernelAccess) { 2116 db1_printf(("dispatch: bp->b_blkno = %ld\n", 2117 (long) bp->b_blkno)); 2118 } 2119 queue->numOutstanding++; 2120 queue->last_deq_sector = req->sectorOffset; 2121 /* acc wouldn't have been let in if there were any pending 2122 * reqs at any other priority */ 2123 queue->curPriority = req->priority; 2124 2125 db1_printf(("Going for %c to unit %d col %d\n", 2126 req->type, queue->raidPtr->raidid, 2127 queue->col)); 2128 db1_printf(("sector %d count %d (%d bytes) %d\n", 2129 (int) req->sectorOffset, (int) req->numSector, 2130 (int) (req->numSector << 2131 queue->raidPtr->logBytesPerSector), 2132 (int) queue->raidPtr->logBytesPerSector)); 2133 bdev_strategy(bp); 2134 2135 break; 2136 2137 default: 2138 panic("bad req->type in rf_DispatchKernelIO"); 2139 } 2140 db1_printf(("Exiting from DispatchKernelIO\n")); 2141 2142 return (0); 2143 } 2144 /* this is the callback function associated with a I/O invoked from 2145 kernel code. 2146 */ 2147 static void 2148 KernelWakeupFunc(struct buf *bp) 2149 { 2150 RF_DiskQueueData_t *req = NULL; 2151 RF_DiskQueue_t *queue; 2152 int s; 2153 2154 s = splbio(); 2155 db1_printf(("recovering the request queue:\n")); 2156 req = bp->b_private; 2157 2158 queue = (RF_DiskQueue_t *) req->queue; 2159 2160 #if RF_ACC_TRACE > 0 2161 if (req->tracerec) { 2162 RF_ETIMER_STOP(req->tracerec->timer); 2163 RF_ETIMER_EVAL(req->tracerec->timer); 2164 RF_LOCK_MUTEX(rf_tracing_mutex); 2165 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2166 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2167 req->tracerec->num_phys_ios++; 2168 RF_UNLOCK_MUTEX(rf_tracing_mutex); 2169 } 2170 #endif 2171 2172 /* XXX Ok, let's get aggressive... If b_error is set, let's go 2173 * ballistic, and mark the component as hosed... */ 2174 2175 if (bp->b_error != 0) { 2176 /* Mark the disk as dead */ 2177 /* but only mark it once... */ 2178 /* and only if it wouldn't leave this RAID set 2179 completely broken */ 2180 if (((queue->raidPtr->Disks[queue->col].status == 2181 rf_ds_optimal) || 2182 (queue->raidPtr->Disks[queue->col].status == 2183 rf_ds_used_spare)) && 2184 (queue->raidPtr->numFailures < 2185 queue->raidPtr->Layout.map->faultsTolerated)) { 2186 printf("raid%d: IO Error. Marking %s as failed.\n", 2187 queue->raidPtr->raidid, 2188 queue->raidPtr->Disks[queue->col].devname); 2189 queue->raidPtr->Disks[queue->col].status = 2190 rf_ds_failed; 2191 queue->raidPtr->status = rf_rs_degraded; 2192 queue->raidPtr->numFailures++; 2193 queue->raidPtr->numNewFailures++; 2194 } else { /* Disk is already dead... */ 2195 /* printf("Disk already marked as dead!\n"); */ 2196 } 2197 2198 } 2199 2200 /* Fill in the error value */ 2201 2202 req->error = bp->b_error; 2203 2204 simple_lock(&queue->raidPtr->iodone_lock); 2205 2206 /* Drop this one on the "finished" queue... */ 2207 TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries); 2208 2209 /* Let the raidio thread know there is work to be done. */ 2210 wakeup(&(queue->raidPtr->iodone)); 2211 2212 simple_unlock(&queue->raidPtr->iodone_lock); 2213 2214 splx(s); 2215 } 2216 2217 2218 2219 /* 2220 * initialize a buf structure for doing an I/O in the kernel. 2221 */ 2222 static void 2223 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev, 2224 RF_SectorNum_t startSect, RF_SectorCount_t numSect, void *bf, 2225 void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector, 2226 struct proc *b_proc) 2227 { 2228 /* bp->b_flags = B_PHYS | rw_flag; */ 2229 bp->b_flags = rw_flag; /* XXX need B_PHYS here too??? */ 2230 bp->b_oflags = 0; 2231 bp->b_cflags = 0; 2232 bp->b_bcount = numSect << logBytesPerSector; 2233 bp->b_bufsize = bp->b_bcount; 2234 bp->b_error = 0; 2235 bp->b_dev = dev; 2236 bp->b_data = bf; 2237 bp->b_blkno = startSect; 2238 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 2239 if (bp->b_bcount == 0) { 2240 panic("bp->b_bcount is zero in InitBP!!"); 2241 } 2242 bp->b_proc = b_proc; 2243 bp->b_iodone = cbFunc; 2244 bp->b_private = cbArg; 2245 } 2246 2247 static void 2248 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs, 2249 struct disklabel *lp) 2250 { 2251 memset(lp, 0, sizeof(*lp)); 2252 2253 /* fabricate a label... */ 2254 lp->d_secperunit = raidPtr->totalSectors; 2255 lp->d_secsize = raidPtr->bytesPerSector; 2256 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 2257 lp->d_ntracks = 4 * raidPtr->numCol; 2258 lp->d_ncylinders = raidPtr->totalSectors / 2259 (lp->d_nsectors * lp->d_ntracks); 2260 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 2261 2262 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 2263 lp->d_type = DTYPE_RAID; 2264 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 2265 lp->d_rpm = 3600; 2266 lp->d_interleave = 1; 2267 lp->d_flags = 0; 2268 2269 lp->d_partitions[RAW_PART].p_offset = 0; 2270 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 2271 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 2272 lp->d_npartitions = RAW_PART + 1; 2273 2274 lp->d_magic = DISKMAGIC; 2275 lp->d_magic2 = DISKMAGIC; 2276 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 2277 2278 } 2279 /* 2280 * Read the disklabel from the raid device. If one is not present, fake one 2281 * up. 2282 */ 2283 static void 2284 raidgetdisklabel(dev_t dev) 2285 { 2286 int unit = raidunit(dev); 2287 struct raid_softc *rs = &raid_softc[unit]; 2288 const char *errstring; 2289 struct disklabel *lp = rs->sc_dkdev.dk_label; 2290 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 2291 RF_Raid_t *raidPtr; 2292 2293 db1_printf(("Getting the disklabel...\n")); 2294 2295 memset(clp, 0, sizeof(*clp)); 2296 2297 raidPtr = raidPtrs[unit]; 2298 2299 raidgetdefaultlabel(raidPtr, rs, lp); 2300 2301 /* 2302 * Call the generic disklabel extraction routine. 2303 */ 2304 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 2305 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2306 if (errstring) 2307 raidmakedisklabel(rs); 2308 else { 2309 int i; 2310 struct partition *pp; 2311 2312 /* 2313 * Sanity check whether the found disklabel is valid. 2314 * 2315 * This is necessary since total size of the raid device 2316 * may vary when an interleave is changed even though exactly 2317 * same components are used, and old disklabel may used 2318 * if that is found. 2319 */ 2320 if (lp->d_secperunit != rs->sc_size) 2321 printf("raid%d: WARNING: %s: " 2322 "total sector size in disklabel (%d) != " 2323 "the size of raid (%ld)\n", unit, rs->sc_xname, 2324 lp->d_secperunit, (long) rs->sc_size); 2325 for (i = 0; i < lp->d_npartitions; i++) { 2326 pp = &lp->d_partitions[i]; 2327 if (pp->p_offset + pp->p_size > rs->sc_size) 2328 printf("raid%d: WARNING: %s: end of partition `%c' " 2329 "exceeds the size of raid (%ld)\n", 2330 unit, rs->sc_xname, 'a' + i, (long) rs->sc_size); 2331 } 2332 } 2333 2334 } 2335 /* 2336 * Take care of things one might want to take care of in the event 2337 * that a disklabel isn't present. 2338 */ 2339 static void 2340 raidmakedisklabel(struct raid_softc *rs) 2341 { 2342 struct disklabel *lp = rs->sc_dkdev.dk_label; 2343 db1_printf(("Making a label..\n")); 2344 2345 /* 2346 * For historical reasons, if there's no disklabel present 2347 * the raw partition must be marked FS_BSDFFS. 2348 */ 2349 2350 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2351 2352 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2353 2354 lp->d_checksum = dkcksum(lp); 2355 } 2356 /* 2357 * Wait interruptibly for an exclusive lock. 2358 * 2359 * XXX 2360 * Several drivers do this; it should be abstracted and made MP-safe. 2361 * (Hmm... where have we seen this warning before :-> GO ) 2362 */ 2363 static int 2364 raidlock(struct raid_softc *rs) 2365 { 2366 int error; 2367 2368 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2369 rs->sc_flags |= RAIDF_WANTED; 2370 if ((error = 2371 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2372 return (error); 2373 } 2374 rs->sc_flags |= RAIDF_LOCKED; 2375 return (0); 2376 } 2377 /* 2378 * Unlock and wake up any waiters. 2379 */ 2380 static void 2381 raidunlock(struct raid_softc *rs) 2382 { 2383 2384 rs->sc_flags &= ~RAIDF_LOCKED; 2385 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2386 rs->sc_flags &= ~RAIDF_WANTED; 2387 wakeup(rs); 2388 } 2389 } 2390 2391 2392 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2393 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2394 2395 int 2396 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter) 2397 { 2398 RF_ComponentLabel_t clabel; 2399 raidread_component_label(dev, b_vp, &clabel); 2400 clabel.mod_counter = mod_counter; 2401 clabel.clean = RF_RAID_CLEAN; 2402 raidwrite_component_label(dev, b_vp, &clabel); 2403 return(0); 2404 } 2405 2406 2407 int 2408 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter) 2409 { 2410 RF_ComponentLabel_t clabel; 2411 raidread_component_label(dev, b_vp, &clabel); 2412 clabel.mod_counter = mod_counter; 2413 clabel.clean = RF_RAID_DIRTY; 2414 raidwrite_component_label(dev, b_vp, &clabel); 2415 return(0); 2416 } 2417 2418 /* ARGSUSED */ 2419 int 2420 raidread_component_label(dev_t dev, struct vnode *b_vp, 2421 RF_ComponentLabel_t *clabel) 2422 { 2423 struct buf *bp; 2424 const struct bdevsw *bdev; 2425 int error; 2426 2427 /* XXX should probably ensure that we don't try to do this if 2428 someone has changed rf_protected_sectors. */ 2429 2430 if (b_vp == NULL) { 2431 /* For whatever reason, this component is not valid. 2432 Don't try to read a component label from it. */ 2433 return(EINVAL); 2434 } 2435 2436 /* get a block of the appropriate size... */ 2437 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2438 bp->b_dev = dev; 2439 2440 /* get our ducks in a row for the read */ 2441 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2442 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2443 bp->b_flags |= B_READ; 2444 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2445 2446 bdev = bdevsw_lookup(bp->b_dev); 2447 if (bdev == NULL) 2448 return (ENXIO); 2449 (*bdev->d_strategy)(bp); 2450 2451 error = biowait(bp); 2452 2453 if (!error) { 2454 memcpy(clabel, bp->b_data, 2455 sizeof(RF_ComponentLabel_t)); 2456 } 2457 2458 brelse(bp, 0); 2459 return(error); 2460 } 2461 /* ARGSUSED */ 2462 int 2463 raidwrite_component_label(dev_t dev, struct vnode *b_vp, 2464 RF_ComponentLabel_t *clabel) 2465 { 2466 struct buf *bp; 2467 const struct bdevsw *bdev; 2468 int error; 2469 2470 /* get a block of the appropriate size... */ 2471 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2472 bp->b_dev = dev; 2473 2474 /* get our ducks in a row for the write */ 2475 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2476 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2477 bp->b_flags |= B_WRITE; 2478 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2479 2480 memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE ); 2481 2482 memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t)); 2483 2484 bdev = bdevsw_lookup(bp->b_dev); 2485 if (bdev == NULL) 2486 return (ENXIO); 2487 (*bdev->d_strategy)(bp); 2488 error = biowait(bp); 2489 brelse(bp, 0); 2490 if (error) { 2491 #if 1 2492 printf("Failed to write RAID component info!\n"); 2493 #endif 2494 } 2495 2496 return(error); 2497 } 2498 2499 void 2500 rf_markalldirty(RF_Raid_t *raidPtr) 2501 { 2502 RF_ComponentLabel_t clabel; 2503 int sparecol; 2504 int c; 2505 int j; 2506 int scol = -1; 2507 2508 raidPtr->mod_counter++; 2509 for (c = 0; c < raidPtr->numCol; c++) { 2510 /* we don't want to touch (at all) a disk that has 2511 failed */ 2512 if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) { 2513 raidread_component_label( 2514 raidPtr->Disks[c].dev, 2515 raidPtr->raid_cinfo[c].ci_vp, 2516 &clabel); 2517 if (clabel.status == rf_ds_spared) { 2518 /* XXX do something special... 2519 but whatever you do, don't 2520 try to access it!! */ 2521 } else { 2522 raidmarkdirty( 2523 raidPtr->Disks[c].dev, 2524 raidPtr->raid_cinfo[c].ci_vp, 2525 raidPtr->mod_counter); 2526 } 2527 } 2528 } 2529 2530 for( c = 0; c < raidPtr->numSpare ; c++) { 2531 sparecol = raidPtr->numCol + c; 2532 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2533 /* 2534 2535 we claim this disk is "optimal" if it's 2536 rf_ds_used_spare, as that means it should be 2537 directly substitutable for the disk it replaced. 2538 We note that too... 2539 2540 */ 2541 2542 for(j=0;j<raidPtr->numCol;j++) { 2543 if (raidPtr->Disks[j].spareCol == sparecol) { 2544 scol = j; 2545 break; 2546 } 2547 } 2548 2549 raidread_component_label( 2550 raidPtr->Disks[sparecol].dev, 2551 raidPtr->raid_cinfo[sparecol].ci_vp, 2552 &clabel); 2553 /* make sure status is noted */ 2554 2555 raid_init_component_label(raidPtr, &clabel); 2556 2557 clabel.row = 0; 2558 clabel.column = scol; 2559 /* Note: we *don't* change status from rf_ds_used_spare 2560 to rf_ds_optimal */ 2561 /* clabel.status = rf_ds_optimal; */ 2562 2563 raidmarkdirty(raidPtr->Disks[sparecol].dev, 2564 raidPtr->raid_cinfo[sparecol].ci_vp, 2565 raidPtr->mod_counter); 2566 } 2567 } 2568 } 2569 2570 2571 void 2572 rf_update_component_labels(RF_Raid_t *raidPtr, int final) 2573 { 2574 RF_ComponentLabel_t clabel; 2575 int sparecol; 2576 int c; 2577 int j; 2578 int scol; 2579 2580 scol = -1; 2581 2582 /* XXX should do extra checks to make sure things really are clean, 2583 rather than blindly setting the clean bit... */ 2584 2585 raidPtr->mod_counter++; 2586 2587 for (c = 0; c < raidPtr->numCol; c++) { 2588 if (raidPtr->Disks[c].status == rf_ds_optimal) { 2589 raidread_component_label( 2590 raidPtr->Disks[c].dev, 2591 raidPtr->raid_cinfo[c].ci_vp, 2592 &clabel); 2593 /* make sure status is noted */ 2594 clabel.status = rf_ds_optimal; 2595 2596 /* bump the counter */ 2597 clabel.mod_counter = raidPtr->mod_counter; 2598 2599 /* note what unit we are configured as */ 2600 clabel.last_unit = raidPtr->raidid; 2601 2602 raidwrite_component_label( 2603 raidPtr->Disks[c].dev, 2604 raidPtr->raid_cinfo[c].ci_vp, 2605 &clabel); 2606 if (final == RF_FINAL_COMPONENT_UPDATE) { 2607 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2608 raidmarkclean( 2609 raidPtr->Disks[c].dev, 2610 raidPtr->raid_cinfo[c].ci_vp, 2611 raidPtr->mod_counter); 2612 } 2613 } 2614 } 2615 /* else we don't touch it.. */ 2616 } 2617 2618 for( c = 0; c < raidPtr->numSpare ; c++) { 2619 sparecol = raidPtr->numCol + c; 2620 /* Need to ensure that the reconstruct actually completed! */ 2621 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2622 /* 2623 2624 we claim this disk is "optimal" if it's 2625 rf_ds_used_spare, as that means it should be 2626 directly substitutable for the disk it replaced. 2627 We note that too... 2628 2629 */ 2630 2631 for(j=0;j<raidPtr->numCol;j++) { 2632 if (raidPtr->Disks[j].spareCol == sparecol) { 2633 scol = j; 2634 break; 2635 } 2636 } 2637 2638 /* XXX shouldn't *really* need this... */ 2639 raidread_component_label( 2640 raidPtr->Disks[sparecol].dev, 2641 raidPtr->raid_cinfo[sparecol].ci_vp, 2642 &clabel); 2643 /* make sure status is noted */ 2644 2645 raid_init_component_label(raidPtr, &clabel); 2646 2647 clabel.mod_counter = raidPtr->mod_counter; 2648 clabel.column = scol; 2649 clabel.status = rf_ds_optimal; 2650 clabel.last_unit = raidPtr->raidid; 2651 2652 raidwrite_component_label( 2653 raidPtr->Disks[sparecol].dev, 2654 raidPtr->raid_cinfo[sparecol].ci_vp, 2655 &clabel); 2656 if (final == RF_FINAL_COMPONENT_UPDATE) { 2657 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2658 raidmarkclean( raidPtr->Disks[sparecol].dev, 2659 raidPtr->raid_cinfo[sparecol].ci_vp, 2660 raidPtr->mod_counter); 2661 } 2662 } 2663 } 2664 } 2665 } 2666 2667 void 2668 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured) 2669 { 2670 2671 if (vp != NULL) { 2672 if (auto_configured == 1) { 2673 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2674 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2675 vput(vp); 2676 2677 } else { 2678 (void) vn_close(vp, FREAD | FWRITE, curlwp->l_cred); 2679 } 2680 } 2681 } 2682 2683 2684 void 2685 rf_UnconfigureVnodes(RF_Raid_t *raidPtr) 2686 { 2687 int r,c; 2688 struct vnode *vp; 2689 int acd; 2690 2691 2692 /* We take this opportunity to close the vnodes like we should.. */ 2693 2694 for (c = 0; c < raidPtr->numCol; c++) { 2695 vp = raidPtr->raid_cinfo[c].ci_vp; 2696 acd = raidPtr->Disks[c].auto_configured; 2697 rf_close_component(raidPtr, vp, acd); 2698 raidPtr->raid_cinfo[c].ci_vp = NULL; 2699 raidPtr->Disks[c].auto_configured = 0; 2700 } 2701 2702 for (r = 0; r < raidPtr->numSpare; r++) { 2703 vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp; 2704 acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured; 2705 rf_close_component(raidPtr, vp, acd); 2706 raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL; 2707 raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0; 2708 } 2709 } 2710 2711 2712 void 2713 rf_ReconThread(struct rf_recon_req *req) 2714 { 2715 int s; 2716 RF_Raid_t *raidPtr; 2717 2718 s = splbio(); 2719 raidPtr = (RF_Raid_t *) req->raidPtr; 2720 raidPtr->recon_in_progress = 1; 2721 2722 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col, 2723 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2724 2725 RF_Free(req, sizeof(*req)); 2726 2727 raidPtr->recon_in_progress = 0; 2728 splx(s); 2729 2730 /* That's all... */ 2731 kthread_exit(0); /* does not return */ 2732 } 2733 2734 void 2735 rf_RewriteParityThread(RF_Raid_t *raidPtr) 2736 { 2737 int retcode; 2738 int s; 2739 2740 raidPtr->parity_rewrite_stripes_done = 0; 2741 raidPtr->parity_rewrite_in_progress = 1; 2742 s = splbio(); 2743 retcode = rf_RewriteParity(raidPtr); 2744 splx(s); 2745 if (retcode) { 2746 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2747 } else { 2748 /* set the clean bit! If we shutdown correctly, 2749 the clean bit on each component label will get 2750 set */ 2751 raidPtr->parity_good = RF_RAID_CLEAN; 2752 } 2753 raidPtr->parity_rewrite_in_progress = 0; 2754 2755 /* Anyone waiting for us to stop? If so, inform them... */ 2756 if (raidPtr->waitShutdown) { 2757 wakeup(&raidPtr->parity_rewrite_in_progress); 2758 } 2759 2760 /* That's all... */ 2761 kthread_exit(0); /* does not return */ 2762 } 2763 2764 2765 void 2766 rf_CopybackThread(RF_Raid_t *raidPtr) 2767 { 2768 int s; 2769 2770 raidPtr->copyback_in_progress = 1; 2771 s = splbio(); 2772 rf_CopybackReconstructedData(raidPtr); 2773 splx(s); 2774 raidPtr->copyback_in_progress = 0; 2775 2776 /* That's all... */ 2777 kthread_exit(0); /* does not return */ 2778 } 2779 2780 2781 void 2782 rf_ReconstructInPlaceThread(struct rf_recon_req *req) 2783 { 2784 int s; 2785 RF_Raid_t *raidPtr; 2786 2787 s = splbio(); 2788 raidPtr = req->raidPtr; 2789 raidPtr->recon_in_progress = 1; 2790 rf_ReconstructInPlace(raidPtr, req->col); 2791 RF_Free(req, sizeof(*req)); 2792 raidPtr->recon_in_progress = 0; 2793 splx(s); 2794 2795 /* That's all... */ 2796 kthread_exit(0); /* does not return */ 2797 } 2798 2799 static RF_AutoConfig_t * 2800 rf_get_component(RF_AutoConfig_t *ac_list, dev_t dev, struct vnode *vp, 2801 const char *cname, RF_SectorCount_t size) 2802 { 2803 int good_one = 0; 2804 RF_ComponentLabel_t *clabel; 2805 RF_AutoConfig_t *ac; 2806 2807 clabel = malloc(sizeof(RF_ComponentLabel_t), M_RAIDFRAME, M_NOWAIT); 2808 if (clabel == NULL) { 2809 oomem: 2810 while(ac_list) { 2811 ac = ac_list; 2812 if (ac->clabel) 2813 free(ac->clabel, M_RAIDFRAME); 2814 ac_list = ac_list->next; 2815 free(ac, M_RAIDFRAME); 2816 } 2817 printf("RAID auto config: out of memory!\n"); 2818 return NULL; /* XXX probably should panic? */ 2819 } 2820 2821 if (!raidread_component_label(dev, vp, clabel)) { 2822 /* Got the label. Does it look reasonable? */ 2823 if (rf_reasonable_label(clabel) && 2824 (clabel->partitionSize <= size)) { 2825 #ifdef DEBUG 2826 printf("Component on: %s: %llu\n", 2827 cname, (unsigned long long)size); 2828 rf_print_component_label(clabel); 2829 #endif 2830 /* if it's reasonable, add it, else ignore it. */ 2831 ac = malloc(sizeof(RF_AutoConfig_t), M_RAIDFRAME, 2832 M_NOWAIT); 2833 if (ac == NULL) { 2834 free(clabel, M_RAIDFRAME); 2835 goto oomem; 2836 } 2837 strlcpy(ac->devname, cname, sizeof(ac->devname)); 2838 ac->dev = dev; 2839 ac->vp = vp; 2840 ac->clabel = clabel; 2841 ac->next = ac_list; 2842 ac_list = ac; 2843 good_one = 1; 2844 } 2845 } 2846 if (!good_one) { 2847 /* cleanup */ 2848 free(clabel, M_RAIDFRAME); 2849 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2850 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2851 vput(vp); 2852 } 2853 return ac_list; 2854 } 2855 2856 RF_AutoConfig_t * 2857 rf_find_raid_components() 2858 { 2859 struct vnode *vp; 2860 struct disklabel label; 2861 struct device *dv; 2862 dev_t dev; 2863 int bmajor, bminor, wedge; 2864 int error; 2865 int i; 2866 RF_AutoConfig_t *ac_list; 2867 2868 2869 /* initialize the AutoConfig list */ 2870 ac_list = NULL; 2871 2872 /* we begin by trolling through *all* the devices on the system */ 2873 2874 for (dv = alldevs.tqh_first; dv != NULL; 2875 dv = dv->dv_list.tqe_next) { 2876 2877 /* we are only interested in disks... */ 2878 if (device_class(dv) != DV_DISK) 2879 continue; 2880 2881 /* we don't care about floppies... */ 2882 if (device_is_a(dv, "fd")) { 2883 continue; 2884 } 2885 2886 /* we don't care about CD's... */ 2887 if (device_is_a(dv, "cd")) { 2888 continue; 2889 } 2890 2891 /* we don't care about md's... */ 2892 if (device_is_a(dv, "md")) { 2893 continue; 2894 } 2895 2896 /* hdfd is the Atari/Hades floppy driver */ 2897 if (device_is_a(dv, "hdfd")) { 2898 continue; 2899 } 2900 2901 /* fdisa is the Atari/Milan floppy driver */ 2902 if (device_is_a(dv, "fdisa")) { 2903 continue; 2904 } 2905 2906 /* need to find the device_name_to_block_device_major stuff */ 2907 bmajor = devsw_name2blk(device_xname(dv), NULL, 0); 2908 2909 /* get a vnode for the raw partition of this disk */ 2910 2911 wedge = device_is_a(dv, "dk"); 2912 bminor = minor(device_unit(dv)); 2913 dev = wedge ? makedev(bmajor, bminor) : 2914 MAKEDISKDEV(bmajor, bminor, RAW_PART); 2915 if (bdevvp(dev, &vp)) 2916 panic("RAID can't alloc vnode"); 2917 2918 error = VOP_OPEN(vp, FREAD, NOCRED); 2919 2920 if (error) { 2921 /* "Who cares." Continue looking 2922 for something that exists*/ 2923 vput(vp); 2924 continue; 2925 } 2926 2927 if (wedge) { 2928 struct dkwedge_info dkw; 2929 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, 2930 NOCRED); 2931 if (error) { 2932 printf("RAIDframe: can't get wedge info for " 2933 "dev %s (%d)\n", device_xname(dv), error); 2934 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2935 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2936 vput(vp); 2937 continue; 2938 } 2939 2940 if (strcmp(dkw.dkw_ptype, DKW_PTYPE_RAIDFRAME) != 0) { 2941 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2942 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2943 vput(vp); 2944 continue; 2945 } 2946 2947 ac_list = rf_get_component(ac_list, dev, vp, 2948 device_xname(dv), dkw.dkw_size); 2949 continue; 2950 } 2951 2952 /* Ok, the disk exists. Go get the disklabel. */ 2953 error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED); 2954 if (error) { 2955 /* 2956 * XXX can't happen - open() would 2957 * have errored out (or faked up one) 2958 */ 2959 if (error != ENOTTY) 2960 printf("RAIDframe: can't get label for dev " 2961 "%s (%d)\n", device_xname(dv), error); 2962 } 2963 2964 /* don't need this any more. We'll allocate it again 2965 a little later if we really do... */ 2966 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2967 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2968 vput(vp); 2969 2970 if (error) 2971 continue; 2972 2973 for (i = 0; i < label.d_npartitions; i++) { 2974 char cname[sizeof(ac_list->devname)]; 2975 2976 /* We only support partitions marked as RAID */ 2977 if (label.d_partitions[i].p_fstype != FS_RAID) 2978 continue; 2979 2980 dev = MAKEDISKDEV(bmajor, device_unit(dv), i); 2981 if (bdevvp(dev, &vp)) 2982 panic("RAID can't alloc vnode"); 2983 2984 error = VOP_OPEN(vp, FREAD, NOCRED); 2985 if (error) { 2986 /* Whatever... */ 2987 vput(vp); 2988 continue; 2989 } 2990 snprintf(cname, sizeof(cname), "%s%c", 2991 device_xname(dv), 'a' + i); 2992 ac_list = rf_get_component(ac_list, dev, vp, cname, 2993 label.d_partitions[i].p_size); 2994 } 2995 } 2996 return ac_list; 2997 } 2998 2999 3000 static int 3001 rf_reasonable_label(RF_ComponentLabel_t *clabel) 3002 { 3003 3004 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 3005 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 3006 ((clabel->clean == RF_RAID_CLEAN) || 3007 (clabel->clean == RF_RAID_DIRTY)) && 3008 clabel->row >=0 && 3009 clabel->column >= 0 && 3010 clabel->num_rows > 0 && 3011 clabel->num_columns > 0 && 3012 clabel->row < clabel->num_rows && 3013 clabel->column < clabel->num_columns && 3014 clabel->blockSize > 0 && 3015 clabel->numBlocks > 0) { 3016 /* label looks reasonable enough... */ 3017 return(1); 3018 } 3019 return(0); 3020 } 3021 3022 3023 #ifdef DEBUG 3024 void 3025 rf_print_component_label(RF_ComponentLabel_t *clabel) 3026 { 3027 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 3028 clabel->row, clabel->column, 3029 clabel->num_rows, clabel->num_columns); 3030 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 3031 clabel->version, clabel->serial_number, 3032 clabel->mod_counter); 3033 printf(" Clean: %s Status: %d\n", 3034 clabel->clean ? "Yes" : "No", clabel->status ); 3035 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 3036 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 3037 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n", 3038 (char) clabel->parityConfig, clabel->blockSize, 3039 clabel->numBlocks); 3040 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" ); 3041 printf(" Contains root partition: %s\n", 3042 clabel->root_partition ? "Yes" : "No" ); 3043 printf(" Last configured as: raid%d\n", clabel->last_unit ); 3044 #if 0 3045 printf(" Config order: %d\n", clabel->config_order); 3046 #endif 3047 3048 } 3049 #endif 3050 3051 RF_ConfigSet_t * 3052 rf_create_auto_sets(RF_AutoConfig_t *ac_list) 3053 { 3054 RF_AutoConfig_t *ac; 3055 RF_ConfigSet_t *config_sets; 3056 RF_ConfigSet_t *cset; 3057 RF_AutoConfig_t *ac_next; 3058 3059 3060 config_sets = NULL; 3061 3062 /* Go through the AutoConfig list, and figure out which components 3063 belong to what sets. */ 3064 ac = ac_list; 3065 while(ac!=NULL) { 3066 /* we're going to putz with ac->next, so save it here 3067 for use at the end of the loop */ 3068 ac_next = ac->next; 3069 3070 if (config_sets == NULL) { 3071 /* will need at least this one... */ 3072 config_sets = (RF_ConfigSet_t *) 3073 malloc(sizeof(RF_ConfigSet_t), 3074 M_RAIDFRAME, M_NOWAIT); 3075 if (config_sets == NULL) { 3076 panic("rf_create_auto_sets: No memory!"); 3077 } 3078 /* this one is easy :) */ 3079 config_sets->ac = ac; 3080 config_sets->next = NULL; 3081 config_sets->rootable = 0; 3082 ac->next = NULL; 3083 } else { 3084 /* which set does this component fit into? */ 3085 cset = config_sets; 3086 while(cset!=NULL) { 3087 if (rf_does_it_fit(cset, ac)) { 3088 /* looks like it matches... */ 3089 ac->next = cset->ac; 3090 cset->ac = ac; 3091 break; 3092 } 3093 cset = cset->next; 3094 } 3095 if (cset==NULL) { 3096 /* didn't find a match above... new set..*/ 3097 cset = (RF_ConfigSet_t *) 3098 malloc(sizeof(RF_ConfigSet_t), 3099 M_RAIDFRAME, M_NOWAIT); 3100 if (cset == NULL) { 3101 panic("rf_create_auto_sets: No memory!"); 3102 } 3103 cset->ac = ac; 3104 ac->next = NULL; 3105 cset->next = config_sets; 3106 cset->rootable = 0; 3107 config_sets = cset; 3108 } 3109 } 3110 ac = ac_next; 3111 } 3112 3113 3114 return(config_sets); 3115 } 3116 3117 static int 3118 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac) 3119 { 3120 RF_ComponentLabel_t *clabel1, *clabel2; 3121 3122 /* If this one matches the *first* one in the set, that's good 3123 enough, since the other members of the set would have been 3124 through here too... */ 3125 /* note that we are not checking partitionSize here.. 3126 3127 Note that we are also not checking the mod_counters here. 3128 If everything else matches execpt the mod_counter, that's 3129 good enough for this test. We will deal with the mod_counters 3130 a little later in the autoconfiguration process. 3131 3132 (clabel1->mod_counter == clabel2->mod_counter) && 3133 3134 The reason we don't check for this is that failed disks 3135 will have lower modification counts. If those disks are 3136 not added to the set they used to belong to, then they will 3137 form their own set, which may result in 2 different sets, 3138 for example, competing to be configured at raid0, and 3139 perhaps competing to be the root filesystem set. If the 3140 wrong ones get configured, or both attempt to become /, 3141 weird behaviour and or serious lossage will occur. Thus we 3142 need to bring them into the fold here, and kick them out at 3143 a later point. 3144 3145 */ 3146 3147 clabel1 = cset->ac->clabel; 3148 clabel2 = ac->clabel; 3149 if ((clabel1->version == clabel2->version) && 3150 (clabel1->serial_number == clabel2->serial_number) && 3151 (clabel1->num_rows == clabel2->num_rows) && 3152 (clabel1->num_columns == clabel2->num_columns) && 3153 (clabel1->sectPerSU == clabel2->sectPerSU) && 3154 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 3155 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 3156 (clabel1->parityConfig == clabel2->parityConfig) && 3157 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 3158 (clabel1->blockSize == clabel2->blockSize) && 3159 (clabel1->numBlocks == clabel2->numBlocks) && 3160 (clabel1->autoconfigure == clabel2->autoconfigure) && 3161 (clabel1->root_partition == clabel2->root_partition) && 3162 (clabel1->last_unit == clabel2->last_unit) && 3163 (clabel1->config_order == clabel2->config_order)) { 3164 /* if it get's here, it almost *has* to be a match */ 3165 } else { 3166 /* it's not consistent with somebody in the set.. 3167 punt */ 3168 return(0); 3169 } 3170 /* all was fine.. it must fit... */ 3171 return(1); 3172 } 3173 3174 int 3175 rf_have_enough_components(RF_ConfigSet_t *cset) 3176 { 3177 RF_AutoConfig_t *ac; 3178 RF_AutoConfig_t *auto_config; 3179 RF_ComponentLabel_t *clabel; 3180 int c; 3181 int num_cols; 3182 int num_missing; 3183 int mod_counter; 3184 int mod_counter_found; 3185 int even_pair_failed; 3186 char parity_type; 3187 3188 3189 /* check to see that we have enough 'live' components 3190 of this set. If so, we can configure it if necessary */ 3191 3192 num_cols = cset->ac->clabel->num_columns; 3193 parity_type = cset->ac->clabel->parityConfig; 3194 3195 /* XXX Check for duplicate components!?!?!? */ 3196 3197 /* Determine what the mod_counter is supposed to be for this set. */ 3198 3199 mod_counter_found = 0; 3200 mod_counter = 0; 3201 ac = cset->ac; 3202 while(ac!=NULL) { 3203 if (mod_counter_found==0) { 3204 mod_counter = ac->clabel->mod_counter; 3205 mod_counter_found = 1; 3206 } else { 3207 if (ac->clabel->mod_counter > mod_counter) { 3208 mod_counter = ac->clabel->mod_counter; 3209 } 3210 } 3211 ac = ac->next; 3212 } 3213 3214 num_missing = 0; 3215 auto_config = cset->ac; 3216 3217 even_pair_failed = 0; 3218 for(c=0; c<num_cols; c++) { 3219 ac = auto_config; 3220 while(ac!=NULL) { 3221 if ((ac->clabel->column == c) && 3222 (ac->clabel->mod_counter == mod_counter)) { 3223 /* it's this one... */ 3224 #ifdef DEBUG 3225 printf("Found: %s at %d\n", 3226 ac->devname,c); 3227 #endif 3228 break; 3229 } 3230 ac=ac->next; 3231 } 3232 if (ac==NULL) { 3233 /* Didn't find one here! */ 3234 /* special case for RAID 1, especially 3235 where there are more than 2 3236 components (where RAIDframe treats 3237 things a little differently :( ) */ 3238 if (parity_type == '1') { 3239 if (c%2 == 0) { /* even component */ 3240 even_pair_failed = 1; 3241 } else { /* odd component. If 3242 we're failed, and 3243 so is the even 3244 component, it's 3245 "Good Night, Charlie" */ 3246 if (even_pair_failed == 1) { 3247 return(0); 3248 } 3249 } 3250 } else { 3251 /* normal accounting */ 3252 num_missing++; 3253 } 3254 } 3255 if ((parity_type == '1') && (c%2 == 1)) { 3256 /* Just did an even component, and we didn't 3257 bail.. reset the even_pair_failed flag, 3258 and go on to the next component.... */ 3259 even_pair_failed = 0; 3260 } 3261 } 3262 3263 clabel = cset->ac->clabel; 3264 3265 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3266 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3267 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3268 /* XXX this needs to be made *much* more general */ 3269 /* Too many failures */ 3270 return(0); 3271 } 3272 /* otherwise, all is well, and we've got enough to take a kick 3273 at autoconfiguring this set */ 3274 return(1); 3275 } 3276 3277 void 3278 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config, 3279 RF_Raid_t *raidPtr) 3280 { 3281 RF_ComponentLabel_t *clabel; 3282 int i; 3283 3284 clabel = ac->clabel; 3285 3286 /* 1. Fill in the common stuff */ 3287 config->numRow = clabel->num_rows = 1; 3288 config->numCol = clabel->num_columns; 3289 config->numSpare = 0; /* XXX should this be set here? */ 3290 config->sectPerSU = clabel->sectPerSU; 3291 config->SUsPerPU = clabel->SUsPerPU; 3292 config->SUsPerRU = clabel->SUsPerRU; 3293 config->parityConfig = clabel->parityConfig; 3294 /* XXX... */ 3295 strcpy(config->diskQueueType,"fifo"); 3296 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3297 config->layoutSpecificSize = 0; /* XXX ?? */ 3298 3299 while(ac!=NULL) { 3300 /* row/col values will be in range due to the checks 3301 in reasonable_label() */ 3302 strcpy(config->devnames[0][ac->clabel->column], 3303 ac->devname); 3304 ac = ac->next; 3305 } 3306 3307 for(i=0;i<RF_MAXDBGV;i++) { 3308 config->debugVars[i][0] = 0; 3309 } 3310 } 3311 3312 int 3313 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value) 3314 { 3315 RF_ComponentLabel_t clabel; 3316 struct vnode *vp; 3317 dev_t dev; 3318 int column; 3319 int sparecol; 3320 3321 raidPtr->autoconfigure = new_value; 3322 3323 for(column=0; column<raidPtr->numCol; column++) { 3324 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3325 dev = raidPtr->Disks[column].dev; 3326 vp = raidPtr->raid_cinfo[column].ci_vp; 3327 raidread_component_label(dev, vp, &clabel); 3328 clabel.autoconfigure = new_value; 3329 raidwrite_component_label(dev, vp, &clabel); 3330 } 3331 } 3332 for(column = 0; column < raidPtr->numSpare ; column++) { 3333 sparecol = raidPtr->numCol + column; 3334 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3335 dev = raidPtr->Disks[sparecol].dev; 3336 vp = raidPtr->raid_cinfo[sparecol].ci_vp; 3337 raidread_component_label(dev, vp, &clabel); 3338 clabel.autoconfigure = new_value; 3339 raidwrite_component_label(dev, vp, &clabel); 3340 } 3341 } 3342 return(new_value); 3343 } 3344 3345 int 3346 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value) 3347 { 3348 RF_ComponentLabel_t clabel; 3349 struct vnode *vp; 3350 dev_t dev; 3351 int column; 3352 int sparecol; 3353 3354 raidPtr->root_partition = new_value; 3355 for(column=0; column<raidPtr->numCol; column++) { 3356 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3357 dev = raidPtr->Disks[column].dev; 3358 vp = raidPtr->raid_cinfo[column].ci_vp; 3359 raidread_component_label(dev, vp, &clabel); 3360 clabel.root_partition = new_value; 3361 raidwrite_component_label(dev, vp, &clabel); 3362 } 3363 } 3364 for(column = 0; column < raidPtr->numSpare ; column++) { 3365 sparecol = raidPtr->numCol + column; 3366 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3367 dev = raidPtr->Disks[sparecol].dev; 3368 vp = raidPtr->raid_cinfo[sparecol].ci_vp; 3369 raidread_component_label(dev, vp, &clabel); 3370 clabel.root_partition = new_value; 3371 raidwrite_component_label(dev, vp, &clabel); 3372 } 3373 } 3374 return(new_value); 3375 } 3376 3377 void 3378 rf_release_all_vps(RF_ConfigSet_t *cset) 3379 { 3380 RF_AutoConfig_t *ac; 3381 3382 ac = cset->ac; 3383 while(ac!=NULL) { 3384 /* Close the vp, and give it back */ 3385 if (ac->vp) { 3386 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3387 VOP_CLOSE(ac->vp, FREAD, NOCRED); 3388 vput(ac->vp); 3389 ac->vp = NULL; 3390 } 3391 ac = ac->next; 3392 } 3393 } 3394 3395 3396 void 3397 rf_cleanup_config_set(RF_ConfigSet_t *cset) 3398 { 3399 RF_AutoConfig_t *ac; 3400 RF_AutoConfig_t *next_ac; 3401 3402 ac = cset->ac; 3403 while(ac!=NULL) { 3404 next_ac = ac->next; 3405 /* nuke the label */ 3406 free(ac->clabel, M_RAIDFRAME); 3407 /* cleanup the config structure */ 3408 free(ac, M_RAIDFRAME); 3409 /* "next.." */ 3410 ac = next_ac; 3411 } 3412 /* and, finally, nuke the config set */ 3413 free(cset, M_RAIDFRAME); 3414 } 3415 3416 3417 void 3418 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel) 3419 { 3420 /* current version number */ 3421 clabel->version = RF_COMPONENT_LABEL_VERSION; 3422 clabel->serial_number = raidPtr->serial_number; 3423 clabel->mod_counter = raidPtr->mod_counter; 3424 clabel->num_rows = 1; 3425 clabel->num_columns = raidPtr->numCol; 3426 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3427 clabel->status = rf_ds_optimal; /* "It's good!" */ 3428 3429 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3430 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3431 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3432 3433 clabel->blockSize = raidPtr->bytesPerSector; 3434 clabel->numBlocks = raidPtr->sectorsPerDisk; 3435 3436 /* XXX not portable */ 3437 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3438 clabel->maxOutstanding = raidPtr->maxOutstanding; 3439 clabel->autoconfigure = raidPtr->autoconfigure; 3440 clabel->root_partition = raidPtr->root_partition; 3441 clabel->last_unit = raidPtr->raidid; 3442 clabel->config_order = raidPtr->config_order; 3443 } 3444 3445 int 3446 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit) 3447 { 3448 RF_Raid_t *raidPtr; 3449 RF_Config_t *config; 3450 int raidID; 3451 int retcode; 3452 3453 #ifdef DEBUG 3454 printf("RAID autoconfigure\n"); 3455 #endif 3456 3457 retcode = 0; 3458 *unit = -1; 3459 3460 /* 1. Create a config structure */ 3461 3462 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3463 M_RAIDFRAME, 3464 M_NOWAIT); 3465 if (config==NULL) { 3466 printf("Out of mem!?!?\n"); 3467 /* XXX do something more intelligent here. */ 3468 return(1); 3469 } 3470 3471 memset(config, 0, sizeof(RF_Config_t)); 3472 3473 /* 3474 2. Figure out what RAID ID this one is supposed to live at 3475 See if we can get the same RAID dev that it was configured 3476 on last time.. 3477 */ 3478 3479 raidID = cset->ac->clabel->last_unit; 3480 if ((raidID < 0) || (raidID >= numraid)) { 3481 /* let's not wander off into lala land. */ 3482 raidID = numraid - 1; 3483 } 3484 if (raidPtrs[raidID]->valid != 0) { 3485 3486 /* 3487 Nope... Go looking for an alternative... 3488 Start high so we don't immediately use raid0 if that's 3489 not taken. 3490 */ 3491 3492 for(raidID = numraid - 1; raidID >= 0; raidID--) { 3493 if (raidPtrs[raidID]->valid == 0) { 3494 /* can use this one! */ 3495 break; 3496 } 3497 } 3498 } 3499 3500 if (raidID < 0) { 3501 /* punt... */ 3502 printf("Unable to auto configure this set!\n"); 3503 printf("(Out of RAID devs!)\n"); 3504 free(config, M_RAIDFRAME); 3505 return(1); 3506 } 3507 3508 #ifdef DEBUG 3509 printf("Configuring raid%d:\n",raidID); 3510 #endif 3511 3512 raidPtr = raidPtrs[raidID]; 3513 3514 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3515 raidPtr->raidid = raidID; 3516 raidPtr->openings = RAIDOUTSTANDING; 3517 3518 /* 3. Build the configuration structure */ 3519 rf_create_configuration(cset->ac, config, raidPtr); 3520 3521 /* 4. Do the configuration */ 3522 retcode = rf_Configure(raidPtr, config, cset->ac); 3523 3524 if (retcode == 0) { 3525 3526 raidinit(raidPtrs[raidID]); 3527 3528 rf_markalldirty(raidPtrs[raidID]); 3529 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3530 if (cset->ac->clabel->root_partition==1) { 3531 /* everything configured just fine. Make a note 3532 that this set is eligible to be root. */ 3533 cset->rootable = 1; 3534 /* XXX do this here? */ 3535 raidPtrs[raidID]->root_partition = 1; 3536 } 3537 } 3538 3539 /* 5. Cleanup */ 3540 free(config, M_RAIDFRAME); 3541 3542 *unit = raidID; 3543 return(retcode); 3544 } 3545 3546 void 3547 rf_disk_unbusy(RF_RaidAccessDesc_t *desc) 3548 { 3549 struct buf *bp; 3550 3551 bp = (struct buf *)desc->bp; 3552 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3553 (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ)); 3554 } 3555 3556 void 3557 rf_pool_init(struct pool *p, size_t size, const char *w_chan, 3558 size_t xmin, size_t xmax) 3559 { 3560 pool_init(p, size, 0, 0, 0, w_chan, NULL, IPL_BIO); 3561 pool_sethiwat(p, xmax); 3562 pool_prime(p, xmin); 3563 pool_setlowat(p, xmin); 3564 } 3565 3566 /* 3567 * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see 3568 * if there is IO pending and if that IO could possibly be done for a 3569 * given RAID set. Returns 0 if IO is waiting and can be done, 1 3570 * otherwise. 3571 * 3572 */ 3573 3574 int 3575 rf_buf_queue_check(int raidid) 3576 { 3577 if ((bufq_peek(raid_softc[raidid].buf_queue) != NULL) && 3578 raidPtrs[raidid]->openings > 0) { 3579 /* there is work to do */ 3580 return 0; 3581 } 3582 /* default is nothing to do */ 3583 return 1; 3584 } 3585 3586 int 3587 rf_getdisksize(struct vnode *vp, struct lwp *l, RF_RaidDisk_t *diskPtr) 3588 { 3589 struct partinfo dpart; 3590 struct dkwedge_info dkw; 3591 int error; 3592 3593 error = VOP_IOCTL(vp, DIOCGPART, &dpart, FREAD, l->l_cred); 3594 if (error == 0) { 3595 diskPtr->blockSize = dpart.disklab->d_secsize; 3596 diskPtr->numBlocks = dpart.part->p_size - rf_protectedSectors; 3597 diskPtr->partitionSize = dpart.part->p_size; 3598 return 0; 3599 } 3600 3601 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, l->l_cred); 3602 if (error == 0) { 3603 diskPtr->blockSize = 512; /* XXX */ 3604 diskPtr->numBlocks = dkw.dkw_size - rf_protectedSectors; 3605 diskPtr->partitionSize = dkw.dkw_size; 3606 return 0; 3607 } 3608 return error; 3609 } 3610 3611 static int 3612 raid_match(struct device *self, struct cfdata *cfdata, 3613 void *aux) 3614 { 3615 return 1; 3616 } 3617 3618 static void 3619 raid_attach(struct device *parent, struct device *self, 3620 void *aux) 3621 { 3622 3623 } 3624 3625 3626 static int 3627 raid_detach(struct device *self, int flags) 3628 { 3629 struct raid_softc *rs = (struct raid_softc *)self; 3630 3631 if (rs->sc_flags & RAIDF_INITED) 3632 return EBUSY; 3633 3634 return 0; 3635 } 3636 3637 static void 3638 rf_set_properties(struct raid_softc *rs, RF_Raid_t *raidPtr) 3639 { 3640 prop_dictionary_t disk_info, odisk_info, geom; 3641 disk_info = prop_dictionary_create(); 3642 geom = prop_dictionary_create(); 3643 prop_dictionary_set_uint64(geom, "sectors-per-unit", 3644 raidPtr->totalSectors); 3645 prop_dictionary_set_uint32(geom, "sector-size", 3646 raidPtr->bytesPerSector); 3647 3648 prop_dictionary_set_uint16(geom, "sectors-per-track", 3649 raidPtr->Layout.dataSectorsPerStripe); 3650 prop_dictionary_set_uint16(geom, "tracks-per-cylinder", 3651 4 * raidPtr->numCol); 3652 3653 prop_dictionary_set_uint64(geom, "cylinders-per-unit", 3654 raidPtr->totalSectors / (raidPtr->Layout.dataSectorsPerStripe * 3655 (4 * raidPtr->numCol))); 3656 3657 prop_dictionary_set(disk_info, "geometry", geom); 3658 prop_object_release(geom); 3659 prop_dictionary_set(device_properties(rs->sc_dev), 3660 "disk-info", disk_info); 3661 odisk_info = rs->sc_dkdev.dk_info; 3662 rs->sc_dkdev.dk_info = disk_info; 3663 if (odisk_info) 3664 prop_object_release(odisk_info); 3665 } 3666 3667 /* 3668 * Implement forwarding of the DIOCCACHESYNC ioctl to each of the components. 3669 * We end up returning whatever error was returned by the first cache flush 3670 * that fails. 3671 */ 3672 3673 static int 3674 rf_sync_component_caches(RF_Raid_t *raidPtr) 3675 { 3676 int c, sparecol; 3677 int e,error; 3678 int force = 1; 3679 3680 error = 0; 3681 for (c = 0; c < raidPtr->numCol; c++) { 3682 if (raidPtr->Disks[c].status == rf_ds_optimal) { 3683 e = VOP_IOCTL(raidPtr->raid_cinfo[c].ci_vp, DIOCCACHESYNC, 3684 &force, FWRITE, NOCRED); 3685 if (e) { 3686 printf("raid%d: cache flush to component %s failed.\n", 3687 raidPtr->raidid, raidPtr->Disks[c].devname); 3688 if (error == 0) { 3689 error = e; 3690 } 3691 } 3692 } 3693 } 3694 3695 for( c = 0; c < raidPtr->numSpare ; c++) { 3696 sparecol = raidPtr->numCol + c; 3697 /* Need to ensure that the reconstruct actually completed! */ 3698 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3699 e = VOP_IOCTL(raidPtr->raid_cinfo[sparecol].ci_vp, 3700 DIOCCACHESYNC, &force, FWRITE, NOCRED); 3701 if (e) { 3702 printf("raid%d: cache flush to component %s failed.\n", 3703 raidPtr->raidid, raidPtr->Disks[sparecol].devname); 3704 if (error == 0) { 3705 error = e; 3706 } 3707 } 3708 } 3709 } 3710 return error; 3711 } 3712
Indexes created Tue Sep 30 17:09:57 GMT 2025