rf_netbsdkintf.c revision 1.250.4.13.2.1
1 /* $NetBSD: rf_netbsdkintf.c,v 1.250.4.13.2.1 2014/11/20 12:25:10 sborrill 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 1.250.4.13.2.1 2014/11/20 12:25:10 sborrill Exp $"); 143 144 #include <sys/param.h> 145 #include <sys/errno.h> 146 #include <sys/pool.h> 147 #include <sys/proc.h> 148 #include <sys/queue.h> 149 #include <sys/disk.h> 150 #include <sys/device.h> 151 #include <sys/stat.h> 152 #include <sys/ioctl.h> 153 #include <sys/fcntl.h> 154 #include <sys/systm.h> 155 #include <sys/vnode.h> 156 #include <sys/disklabel.h> 157 #include <sys/conf.h> 158 #include <sys/buf.h> 159 #include <sys/bufq.h> 160 #include <sys/user.h> 161 #include <sys/reboot.h> 162 #include <sys/kauth.h> 163 164 #include <prop/proplib.h> 165 166 #include <dev/raidframe/raidframevar.h> 167 #include <dev/raidframe/raidframeio.h> 168 #include <dev/raidframe/rf_paritymap.h> 169 #include "raid.h" 170 #include "opt_raid_autoconfig.h" 171 #include "rf_raid.h" 172 #include "rf_copyback.h" 173 #include "rf_dag.h" 174 #include "rf_dagflags.h" 175 #include "rf_desc.h" 176 #include "rf_diskqueue.h" 177 #include "rf_etimer.h" 178 #include "rf_general.h" 179 #include "rf_kintf.h" 180 #include "rf_options.h" 181 #include "rf_driver.h" 182 #include "rf_parityscan.h" 183 #include "rf_threadstuff.h" 184 185 #ifdef DEBUG 186 int rf_kdebug_level = 0; 187 #define db1_printf(a) if (rf_kdebug_level > 0) printf a 188 #else /* DEBUG */ 189 #define db1_printf(a) { } 190 #endif /* DEBUG */ 191 192 static RF_Raid_t **raidPtrs; /* global raid device descriptors */ 193 194 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 195 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex) 196 197 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a 198 * spare table */ 199 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from 200 * installation process */ 201 #endif 202 203 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures"); 204 205 /* prototypes */ 206 static void KernelWakeupFunc(struct buf *); 207 static void InitBP(struct buf *, struct vnode *, unsigned, 208 dev_t, RF_SectorNum_t, RF_SectorCount_t, void *, void (*) (struct buf *), 209 void *, int, struct proc *); 210 static void raidinit(RF_Raid_t *); 211 212 void raidattach(int); 213 static int raid_match(struct device *, struct cfdata *, void *); 214 static void raid_attach(struct device *, struct device *, void *); 215 static int raid_detach(struct device *, int); 216 217 static int raidread_component_area(dev_t, struct vnode *, void *, size_t, 218 daddr_t, daddr_t); 219 static int raidwrite_component_area(dev_t, struct vnode *, void *, size_t, 220 daddr_t, daddr_t, int); 221 222 static int raidwrite_component_label(unsigned, 223 dev_t, struct vnode *, RF_ComponentLabel_t *); 224 static int raidread_component_label(unsigned, 225 dev_t, struct vnode *, RF_ComponentLabel_t *); 226 227 228 dev_type_open(raidopen); 229 dev_type_close(raidclose); 230 dev_type_read(raidread); 231 dev_type_write(raidwrite); 232 dev_type_ioctl(raidioctl); 233 dev_type_strategy(raidstrategy); 234 dev_type_dump(raiddump); 235 dev_type_size(raidsize); 236 237 const struct bdevsw raid_bdevsw = { 238 raidopen, raidclose, raidstrategy, raidioctl, 239 raiddump, raidsize, D_DISK 240 }; 241 242 const struct cdevsw raid_cdevsw = { 243 raidopen, raidclose, raidread, raidwrite, raidioctl, 244 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 245 }; 246 247 static struct dkdriver rf_dkdriver = { raidstrategy, minphys }; 248 249 /* XXX Not sure if the following should be replacing the raidPtrs above, 250 or if it should be used in conjunction with that... 251 */ 252 253 struct raid_softc { 254 struct device *sc_dev; 255 int sc_flags; /* flags */ 256 int sc_cflags; /* configuration flags */ 257 uint64_t sc_size; /* size of the raid device */ 258 char sc_xname[20]; /* XXX external name */ 259 struct disk sc_dkdev; /* generic disk device info */ 260 struct bufq_state *buf_queue; /* used for the device queue */ 261 }; 262 /* sc_flags */ 263 #define RAIDF_INITED 0x01 /* unit has been initialized */ 264 #define RAIDF_WLABEL 0x02 /* label area is writable */ 265 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 266 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 267 #define RAIDF_LOCKED 0x80 /* unit is locked */ 268 269 #define raidunit(x) DISKUNIT(x) 270 int numraid = 0; 271 272 extern struct cfdriver raid_cd; 273 CFATTACH_DECL_NEW(raid, sizeof(struct raid_softc), 274 raid_match, raid_attach, raid_detach, NULL); 275 276 /* 277 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 278 * Be aware that large numbers can allow the driver to consume a lot of 279 * kernel memory, especially on writes, and in degraded mode reads. 280 * 281 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 282 * a single 64K write will typically require 64K for the old data, 283 * 64K for the old parity, and 64K for the new parity, for a total 284 * of 192K (if the parity buffer is not re-used immediately). 285 * Even it if is used immediately, that's still 128K, which when multiplied 286 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 287 * 288 * Now in degraded mode, for example, a 64K read on the above setup may 289 * require data reconstruction, which will require *all* of the 4 remaining 290 * disks to participate -- 4 * 32K/disk == 128K again. 291 */ 292 293 #ifndef RAIDOUTSTANDING 294 #define RAIDOUTSTANDING 6 295 #endif 296 297 #define RAIDLABELDEV(dev) \ 298 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 299 300 /* declared here, and made public, for the benefit of KVM stuff.. */ 301 struct raid_softc *raid_softc; 302 303 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *, 304 struct disklabel *); 305 static void raidgetdisklabel(dev_t); 306 static void raidmakedisklabel(struct raid_softc *); 307 308 static int raidlock(struct raid_softc *); 309 static void raidunlock(struct raid_softc *); 310 311 static void rf_markalldirty(RF_Raid_t *); 312 static void rf_set_properties(struct raid_softc *, RF_Raid_t *); 313 314 void rf_ReconThread(struct rf_recon_req *); 315 void rf_RewriteParityThread(RF_Raid_t *raidPtr); 316 void rf_CopybackThread(RF_Raid_t *raidPtr); 317 void rf_ReconstructInPlaceThread(struct rf_recon_req *); 318 int rf_autoconfig(struct device *self); 319 void rf_buildroothack(RF_ConfigSet_t *); 320 321 RF_AutoConfig_t *rf_find_raid_components(void); 322 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *); 323 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *); 324 static int rf_reasonable_label(RF_ComponentLabel_t *, uint64_t); 325 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *); 326 int rf_set_autoconfig(RF_Raid_t *, int); 327 int rf_set_rootpartition(RF_Raid_t *, int); 328 void rf_release_all_vps(RF_ConfigSet_t *); 329 void rf_cleanup_config_set(RF_ConfigSet_t *); 330 int rf_have_enough_components(RF_ConfigSet_t *); 331 int rf_auto_config_set(RF_ConfigSet_t *, int *); 332 static void rf_fix_old_label_size(RF_ComponentLabel_t *, uint64_t); 333 334 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not 335 allow autoconfig to take place. 336 Note that this is overridden by having 337 RAID_AUTOCONFIG as an option in the 338 kernel config file. */ 339 340 struct RF_Pools_s rf_pools; 341 342 void 343 raidattach(int num) 344 { 345 int raidID; 346 int i, rc; 347 348 aprint_debug("raidattach: Asked for %d units\n", num); 349 350 if (num <= 0) { 351 #ifdef DIAGNOSTIC 352 panic("raidattach: count <= 0"); 353 #endif 354 return; 355 } 356 /* This is where all the initialization stuff gets done. */ 357 358 numraid = num; 359 360 /* Make some space for requested number of units... */ 361 362 RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **)); 363 if (raidPtrs == NULL) { 364 panic("raidPtrs is NULL!!"); 365 } 366 367 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 368 rf_mutex_init(&rf_sparet_wait_mutex); 369 370 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 371 #endif 372 373 for (i = 0; i < num; i++) 374 raidPtrs[i] = NULL; 375 rc = rf_BootRaidframe(); 376 if (rc == 0) 377 aprint_normal("Kernelized RAIDframe activated\n"); 378 else 379 panic("Serious error booting RAID!!"); 380 381 /* put together some datastructures like the CCD device does.. This 382 * lets us lock the device and what-not when it gets opened. */ 383 384 raid_softc = (struct raid_softc *) 385 malloc(num * sizeof(struct raid_softc), 386 M_RAIDFRAME, M_NOWAIT); 387 if (raid_softc == NULL) { 388 aprint_error("WARNING: no memory for RAIDframe driver\n"); 389 return; 390 } 391 392 memset(raid_softc, 0, num * sizeof(struct raid_softc)); 393 394 for (raidID = 0; raidID < num; raidID++) { 395 bufq_alloc(&raid_softc[raidID].buf_queue, "fcfs", 0); 396 397 RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t), 398 (RF_Raid_t *)); 399 if (raidPtrs[raidID] == NULL) { 400 aprint_error("WARNING: raidPtrs[%d] is NULL\n", raidID); 401 numraid = raidID; 402 return; 403 } 404 } 405 406 if (config_cfattach_attach(raid_cd.cd_name, &raid_ca)) { 407 aprint_error("raidattach: config_cfattach_attach failed?\n"); 408 } 409 410 #ifdef RAID_AUTOCONFIG 411 raidautoconfig = 1; 412 #endif 413 414 /* 415 * Register a finalizer which will be used to auto-config RAID 416 * sets once all real hardware devices have been found. 417 */ 418 if (config_finalize_register(NULL, rf_autoconfig) != 0) 419 aprint_error("WARNING: unable to register RAIDframe finalizer\n"); 420 } 421 422 int 423 rf_autoconfig(struct device *self) 424 { 425 RF_AutoConfig_t *ac_list; 426 RF_ConfigSet_t *config_sets; 427 428 if (raidautoconfig == 0) 429 return (0); 430 431 /* XXX This code can only be run once. */ 432 raidautoconfig = 0; 433 434 /* 1. locate all RAID components on the system */ 435 aprint_debug("Searching for RAID components...\n"); 436 ac_list = rf_find_raid_components(); 437 438 /* 2. Sort them into their respective sets. */ 439 config_sets = rf_create_auto_sets(ac_list); 440 441 /* 442 * 3. Evaluate each set andconfigure the valid ones. 443 * This gets done in rf_buildroothack(). 444 */ 445 rf_buildroothack(config_sets); 446 447 return 1; 448 } 449 450 void 451 rf_buildroothack(RF_ConfigSet_t *config_sets) 452 { 453 RF_ConfigSet_t *cset; 454 RF_ConfigSet_t *next_cset; 455 int retcode; 456 int raidID; 457 int rootID; 458 int col; 459 int num_root; 460 char *devname; 461 462 rootID = 0; 463 num_root = 0; 464 cset = config_sets; 465 while(cset != NULL ) { 466 next_cset = cset->next; 467 if (rf_have_enough_components(cset) && 468 cset->ac->clabel->autoconfigure==1) { 469 retcode = rf_auto_config_set(cset,&raidID); 470 if (!retcode) { 471 aprint_debug("raid%d: configured ok\n", raidID); 472 if (cset->rootable) { 473 rootID = raidID; 474 num_root++; 475 } 476 } else { 477 /* The autoconfig didn't work :( */ 478 aprint_debug("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 479 rf_release_all_vps(cset); 480 } 481 } else { 482 /* we're not autoconfiguring this set... 483 release the associated resources */ 484 rf_release_all_vps(cset); 485 } 486 /* cleanup */ 487 rf_cleanup_config_set(cset); 488 cset = next_cset; 489 } 490 491 /* if the user has specified what the root device should be 492 then we don't touch booted_device or boothowto... */ 493 494 if (rootspec != NULL) 495 return; 496 497 /* we found something bootable... */ 498 499 if (num_root == 1) { 500 booted_device = raid_softc[rootID].sc_dev; 501 } else if (num_root > 1) { 502 503 /* 504 * Maybe the MD code can help. If it cannot, then 505 * setroot() will discover that we have no 506 * booted_device and will ask the user if nothing was 507 * hardwired in the kernel config file 508 */ 509 510 if (booted_device == NULL) 511 cpu_rootconf(); 512 if (booted_device == NULL) 513 return; 514 515 num_root = 0; 516 for (raidID = 0; raidID < numraid; raidID++) { 517 if (raidPtrs[raidID]->valid == 0) 518 continue; 519 520 if (raidPtrs[raidID]->root_partition == 0) 521 continue; 522 523 for (col = 0; col < raidPtrs[raidID]->numCol; col++) { 524 devname = raidPtrs[raidID]->Disks[col].devname; 525 devname += sizeof("/dev/") - 1; 526 if (strncmp(devname, device_xname(booted_device), 527 strlen(device_xname(booted_device))) != 0) 528 continue; 529 aprint_debug("raid%d includes boot device %s\n", 530 raidID, devname); 531 num_root++; 532 rootID = raidID; 533 } 534 } 535 536 if (num_root == 1) { 537 booted_device = raid_softc[rootID].sc_dev; 538 } else { 539 /* we can't guess.. require the user to answer... */ 540 boothowto |= RB_ASKNAME; 541 } 542 } 543 } 544 545 546 int 547 raidsize(dev_t dev) 548 { 549 struct raid_softc *rs; 550 struct disklabel *lp; 551 int part, unit, omask, size; 552 553 unit = raidunit(dev); 554 if (unit >= numraid) 555 return (-1); 556 rs = &raid_softc[unit]; 557 558 if ((rs->sc_flags & RAIDF_INITED) == 0) 559 return (-1); 560 561 part = DISKPART(dev); 562 omask = rs->sc_dkdev.dk_openmask & (1 << part); 563 lp = rs->sc_dkdev.dk_label; 564 565 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curlwp)) 566 return (-1); 567 568 if (lp->d_partitions[part].p_fstype != FS_SWAP) 569 size = -1; 570 else 571 size = lp->d_partitions[part].p_size * 572 (lp->d_secsize / DEV_BSIZE); 573 574 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curlwp)) 575 return (-1); 576 577 return (size); 578 579 } 580 581 int 582 raiddump(dev_t dev, daddr_t blkno, void *va, size_t size) 583 { 584 int unit = raidunit(dev); 585 struct raid_softc *rs; 586 const struct bdevsw *bdev; 587 struct disklabel *lp; 588 RF_Raid_t *raidPtr; 589 daddr_t offset; 590 int part, c, sparecol, j, scol, dumpto; 591 int error = 0; 592 593 if (unit >= numraid) 594 return (ENXIO); 595 596 rs = &raid_softc[unit]; 597 raidPtr = raidPtrs[unit]; 598 599 if ((rs->sc_flags & RAIDF_INITED) == 0) 600 return ENXIO; 601 602 /* we only support dumping to RAID 1 sets */ 603 if (raidPtr->Layout.numDataCol != 1 || 604 raidPtr->Layout.numParityCol != 1) 605 return EINVAL; 606 607 608 if ((error = raidlock(rs)) != 0) 609 return error; 610 611 if (size % DEV_BSIZE != 0) { 612 error = EINVAL; 613 goto out; 614 } 615 616 if (blkno + size / DEV_BSIZE > rs->sc_size) { 617 printf("%s: blkno (%" PRIu64 ") + size / DEV_BSIZE (%zu) > " 618 "sc->sc_size (%" PRIu64 ")\n", __func__, blkno, 619 size / DEV_BSIZE, rs->sc_size); 620 error = EINVAL; 621 goto out; 622 } 623 624 part = DISKPART(dev); 625 lp = rs->sc_dkdev.dk_label; 626 offset = lp->d_partitions[part].p_offset + RF_PROTECTED_SECTORS; 627 628 /* figure out what device is alive.. */ 629 630 /* 631 Look for a component to dump to. The preference for the 632 component to dump to is as follows: 633 1) the master 634 2) a used_spare of the master 635 3) the slave 636 4) a used_spare of the slave 637 */ 638 639 dumpto = -1; 640 for (c = 0; c < raidPtr->numCol; c++) { 641 if (raidPtr->Disks[c].status == rf_ds_optimal) { 642 /* this might be the one */ 643 dumpto = c; 644 break; 645 } 646 } 647 648 /* 649 At this point we have possibly selected a live master or a 650 live slave. We now check to see if there is a spared 651 master (or a spared slave), if we didn't find a live master 652 or a live slave. 653 */ 654 655 for (c = 0; c < raidPtr->numSpare; c++) { 656 sparecol = raidPtr->numCol + c; 657 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 658 /* How about this one? */ 659 scol = -1; 660 for(j=0;j<raidPtr->numCol;j++) { 661 if (raidPtr->Disks[j].spareCol == sparecol) { 662 scol = j; 663 break; 664 } 665 } 666 if (scol == 0) { 667 /* 668 We must have found a spared master! 669 We'll take that over anything else 670 found so far. (We couldn't have 671 found a real master before, since 672 this is a used spare, and it's 673 saying that it's replacing the 674 master.) On reboot (with 675 autoconfiguration turned on) 676 sparecol will become the 1st 677 component (component0) of this set. 678 */ 679 dumpto = sparecol; 680 break; 681 } else if (scol != -1) { 682 /* 683 Must be a spared slave. We'll dump 684 to that if we havn't found anything 685 else so far. 686 */ 687 if (dumpto == -1) 688 dumpto = sparecol; 689 } 690 } 691 } 692 693 if (dumpto == -1) { 694 /* we couldn't find any live components to dump to!?!? 695 */ 696 error = EINVAL; 697 goto out; 698 } 699 700 bdev = bdevsw_lookup(raidPtr->Disks[dumpto].dev); 701 702 /* 703 Note that blkno is relative to this particular partition. 704 By adding the offset of this partition in the RAID 705 set, and also adding RF_PROTECTED_SECTORS, we get a 706 value that is relative to the partition used for the 707 underlying component. 708 */ 709 710 error = (*bdev->d_dump)(raidPtr->Disks[dumpto].dev, 711 blkno + offset, va, size); 712 713 out: 714 raidunlock(rs); 715 716 return error; 717 } 718 /* ARGSUSED */ 719 int 720 raidopen(dev_t dev, int flags, int fmt, 721 struct lwp *l) 722 { 723 int unit = raidunit(dev); 724 struct raid_softc *rs; 725 struct disklabel *lp; 726 int part, pmask; 727 int error = 0; 728 729 if (unit >= numraid) 730 return (ENXIO); 731 rs = &raid_softc[unit]; 732 733 if ((error = raidlock(rs)) != 0) 734 return (error); 735 lp = rs->sc_dkdev.dk_label; 736 737 part = DISKPART(dev); 738 739 /* 740 * If there are wedges, and this is not RAW_PART, then we 741 * need to fail. 742 */ 743 if (rs->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) { 744 error = EBUSY; 745 goto bad; 746 } 747 pmask = (1 << part); 748 749 if ((rs->sc_flags & RAIDF_INITED) && 750 (rs->sc_dkdev.dk_openmask == 0)) 751 raidgetdisklabel(dev); 752 753 /* make sure that this partition exists */ 754 755 if (part != RAW_PART) { 756 if (((rs->sc_flags & RAIDF_INITED) == 0) || 757 ((part >= lp->d_npartitions) || 758 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 759 error = ENXIO; 760 goto bad; 761 } 762 } 763 /* Prevent this unit from being unconfigured while open. */ 764 switch (fmt) { 765 case S_IFCHR: 766 rs->sc_dkdev.dk_copenmask |= pmask; 767 break; 768 769 case S_IFBLK: 770 rs->sc_dkdev.dk_bopenmask |= pmask; 771 break; 772 } 773 774 if ((rs->sc_dkdev.dk_openmask == 0) && 775 ((rs->sc_flags & RAIDF_INITED) != 0)) { 776 /* First one... mark things as dirty... Note that we *MUST* 777 have done a configure before this. I DO NOT WANT TO BE 778 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 779 THAT THEY BELONG TOGETHER!!!!! */ 780 /* XXX should check to see if we're only open for reading 781 here... If so, we needn't do this, but then need some 782 other way of keeping track of what's happened.. */ 783 784 rf_markalldirty( raidPtrs[unit] ); 785 } 786 787 788 rs->sc_dkdev.dk_openmask = 789 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 790 791 bad: 792 raidunlock(rs); 793 794 return (error); 795 796 797 } 798 /* ARGSUSED */ 799 int 800 raidclose(dev_t dev, int flags, int fmt, struct lwp *l) 801 { 802 int unit = raidunit(dev); 803 struct cfdata *cf; 804 struct raid_softc *rs; 805 int error = 0; 806 int part; 807 808 if (unit >= numraid) 809 return (ENXIO); 810 rs = &raid_softc[unit]; 811 812 if ((error = raidlock(rs)) != 0) 813 return (error); 814 815 part = DISKPART(dev); 816 817 /* ...that much closer to allowing unconfiguration... */ 818 switch (fmt) { 819 case S_IFCHR: 820 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 821 break; 822 823 case S_IFBLK: 824 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 825 break; 826 } 827 rs->sc_dkdev.dk_openmask = 828 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 829 830 if ((rs->sc_dkdev.dk_openmask == 0) && 831 ((rs->sc_flags & RAIDF_INITED) != 0)) { 832 /* Last one... device is not unconfigured yet. 833 Device shutdown has taken care of setting the 834 clean bits if RAIDF_INITED is not set 835 mark things as clean... */ 836 837 rf_update_component_labels(raidPtrs[unit], 838 RF_FINAL_COMPONENT_UPDATE); 839 if (doing_shutdown) { 840 /* last one, and we're going down, so 841 lights out for this RAID set too. */ 842 error = rf_Shutdown(raidPtrs[unit]); 843 844 /* It's no longer initialized... */ 845 rs->sc_flags &= ~RAIDF_INITED; 846 847 /* detach the device */ 848 849 cf = device_cfdata(rs->sc_dev); 850 error = config_detach(rs->sc_dev, DETACH_QUIET); 851 free(cf, M_RAIDFRAME); 852 853 /* Detach the disk. */ 854 dkwedge_delall(&rs->sc_dkdev); 855 disk_detach(&rs->sc_dkdev); 856 disk_destroy(&rs->sc_dkdev); 857 } 858 } 859 860 raidunlock(rs); 861 return (0); 862 863 } 864 865 void 866 raidstrategy(struct buf *bp) 867 { 868 int s; 869 870 unsigned int raidID = raidunit(bp->b_dev); 871 RF_Raid_t *raidPtr; 872 struct raid_softc *rs = &raid_softc[raidID]; 873 int wlabel; 874 875 if ((rs->sc_flags & RAIDF_INITED) ==0) { 876 bp->b_error = ENXIO; 877 goto done; 878 } 879 if (raidID >= numraid || !raidPtrs[raidID]) { 880 bp->b_error = ENODEV; 881 goto done; 882 } 883 raidPtr = raidPtrs[raidID]; 884 if (!raidPtr->valid) { 885 bp->b_error = ENODEV; 886 goto done; 887 } 888 if (bp->b_bcount == 0) { 889 db1_printf(("b_bcount is zero..\n")); 890 goto done; 891 } 892 893 /* 894 * Do bounds checking and adjust transfer. If there's an 895 * error, the bounds check will flag that for us. 896 */ 897 898 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 899 if (DISKPART(bp->b_dev) == RAW_PART) { 900 uint64_t size; /* device size in DEV_BSIZE unit */ 901 902 if (raidPtr->logBytesPerSector > DEV_BSHIFT) { 903 size = raidPtr->totalSectors << 904 (raidPtr->logBytesPerSector - DEV_BSHIFT); 905 } else { 906 size = raidPtr->totalSectors >> 907 (DEV_BSHIFT - raidPtr->logBytesPerSector); 908 } 909 if (bounds_check_with_mediasize(bp, DEV_BSIZE, size) <= 0) { 910 goto done; 911 } 912 } else { 913 if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) { 914 db1_printf(("Bounds check failed!!:%d %d\n", 915 (int) bp->b_blkno, (int) wlabel)); 916 goto done; 917 } 918 } 919 s = splbio(); 920 921 bp->b_resid = 0; 922 923 /* stuff it onto our queue */ 924 BUFQ_PUT(rs->buf_queue, bp); 925 926 /* scheduled the IO to happen at the next convenient time */ 927 wakeup(&(raidPtrs[raidID]->iodone)); 928 929 splx(s); 930 return; 931 932 done: 933 bp->b_resid = bp->b_bcount; 934 biodone(bp); 935 } 936 /* ARGSUSED */ 937 int 938 raidread(dev_t dev, struct uio *uio, int flags) 939 { 940 int unit = raidunit(dev); 941 struct raid_softc *rs; 942 943 if (unit >= numraid) 944 return (ENXIO); 945 rs = &raid_softc[unit]; 946 947 if ((rs->sc_flags & RAIDF_INITED) == 0) 948 return (ENXIO); 949 950 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio)); 951 952 } 953 /* ARGSUSED */ 954 int 955 raidwrite(dev_t dev, struct uio *uio, int flags) 956 { 957 int unit = raidunit(dev); 958 struct raid_softc *rs; 959 960 if (unit >= numraid) 961 return (ENXIO); 962 rs = &raid_softc[unit]; 963 964 if ((rs->sc_flags & RAIDF_INITED) == 0) 965 return (ENXIO); 966 967 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio)); 968 969 } 970 971 int 972 raidioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 973 { 974 int unit = raidunit(dev); 975 int error = 0; 976 int part, pmask, s; 977 struct cfdata *cf; 978 struct raid_softc *rs; 979 RF_Config_t *k_cfg, *u_cfg; 980 RF_Raid_t *raidPtr; 981 RF_RaidDisk_t *diskPtr; 982 RF_AccTotals_t *totals; 983 RF_DeviceConfig_t *d_cfg, **ucfgp; 984 u_char *specific_buf; 985 int retcode = 0; 986 int column; 987 /* int raidid; */ 988 struct rf_recon_req *rrcopy, *rr; 989 RF_ComponentLabel_t *clabel; 990 RF_ComponentLabel_t *ci_label; 991 RF_ComponentLabel_t **clabel_ptr; 992 RF_SingleComponent_t *sparePtr,*componentPtr; 993 RF_SingleComponent_t component; 994 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 995 int i, j, d; 996 #ifdef __HAVE_OLD_DISKLABEL 997 struct disklabel newlabel; 998 #endif 999 struct dkwedge_info *dkw; 1000 1001 if (unit >= numraid) 1002 return (ENXIO); 1003 rs = &raid_softc[unit]; 1004 raidPtr = raidPtrs[unit]; 1005 1006 db1_printf(("raidioctl: %d %d %d %lu\n", (int) dev, 1007 (int) DISKPART(dev), (int) unit, cmd)); 1008 1009 /* Must be open for writes for these commands... */ 1010 switch (cmd) { 1011 #ifdef DIOCGSECTORSIZE 1012 case DIOCGSECTORSIZE: 1013 *(u_int *)data = raidPtr->bytesPerSector; 1014 return 0; 1015 case DIOCGMEDIASIZE: 1016 *(off_t *)data = 1017 (off_t)raidPtr->totalSectors * raidPtr->bytesPerSector; 1018 return 0; 1019 #endif 1020 case DIOCSDINFO: 1021 case DIOCWDINFO: 1022 #ifdef __HAVE_OLD_DISKLABEL 1023 case ODIOCWDINFO: 1024 case ODIOCSDINFO: 1025 #endif 1026 case DIOCWLABEL: 1027 case DIOCAWEDGE: 1028 case DIOCDWEDGE: 1029 case DIOCSSTRATEGY: 1030 if ((flag & FWRITE) == 0) 1031 return (EBADF); 1032 } 1033 1034 /* Must be initialized for these... */ 1035 switch (cmd) { 1036 case DIOCGDINFO: 1037 case DIOCSDINFO: 1038 case DIOCWDINFO: 1039 #ifdef __HAVE_OLD_DISKLABEL 1040 case ODIOCGDINFO: 1041 case ODIOCWDINFO: 1042 case ODIOCSDINFO: 1043 case ODIOCGDEFLABEL: 1044 #endif 1045 case DIOCGPART: 1046 case DIOCWLABEL: 1047 case DIOCGDEFLABEL: 1048 case DIOCAWEDGE: 1049 case DIOCDWEDGE: 1050 case DIOCLWEDGES: 1051 case DIOCCACHESYNC: 1052 case RAIDFRAME_SHUTDOWN: 1053 case RAIDFRAME_REWRITEPARITY: 1054 case RAIDFRAME_GET_INFO: 1055 case RAIDFRAME_RESET_ACCTOTALS: 1056 case RAIDFRAME_GET_ACCTOTALS: 1057 case RAIDFRAME_KEEP_ACCTOTALS: 1058 case RAIDFRAME_GET_SIZE: 1059 case RAIDFRAME_FAIL_DISK: 1060 case RAIDFRAME_COPYBACK: 1061 case RAIDFRAME_CHECK_RECON_STATUS: 1062 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1063 case RAIDFRAME_GET_COMPONENT_LABEL: 1064 case RAIDFRAME_SET_COMPONENT_LABEL: 1065 case RAIDFRAME_ADD_HOT_SPARE: 1066 case RAIDFRAME_REMOVE_HOT_SPARE: 1067 case RAIDFRAME_INIT_LABELS: 1068 case RAIDFRAME_REBUILD_IN_PLACE: 1069 case RAIDFRAME_CHECK_PARITY: 1070 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1071 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1072 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1073 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1074 case RAIDFRAME_SET_AUTOCONFIG: 1075 case RAIDFRAME_SET_ROOT: 1076 case RAIDFRAME_DELETE_COMPONENT: 1077 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1078 case RAIDFRAME_PARITYMAP_STATUS: 1079 case RAIDFRAME_PARITYMAP_GET_DISABLE: 1080 case RAIDFRAME_PARITYMAP_SET_DISABLE: 1081 case RAIDFRAME_PARITYMAP_SET_PARAMS: 1082 case DIOCGSTRATEGY: 1083 case DIOCSSTRATEGY: 1084 if ((rs->sc_flags & RAIDF_INITED) == 0) 1085 return (ENXIO); 1086 } 1087 1088 switch (cmd) { 1089 1090 /* configure the system */ 1091 case RAIDFRAME_CONFIGURE: 1092 1093 if (raidPtr->valid) { 1094 /* There is a valid RAID set running on this unit! */ 1095 printf("raid%d: Device already configured!\n",unit); 1096 return(EINVAL); 1097 } 1098 1099 /* copy-in the configuration information */ 1100 /* data points to a pointer to the configuration structure */ 1101 1102 u_cfg = *((RF_Config_t **) data); 1103 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *)); 1104 if (k_cfg == NULL) { 1105 return (ENOMEM); 1106 } 1107 retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t)); 1108 if (retcode) { 1109 RF_Free(k_cfg, sizeof(RF_Config_t)); 1110 db1_printf(("rf_ioctl: retcode=%d copyin.1\n", 1111 retcode)); 1112 return (retcode); 1113 } 1114 /* allocate a buffer for the layout-specific data, and copy it 1115 * in */ 1116 if (k_cfg->layoutSpecificSize) { 1117 if (k_cfg->layoutSpecificSize > 10000) { 1118 /* sanity check */ 1119 RF_Free(k_cfg, sizeof(RF_Config_t)); 1120 return (EINVAL); 1121 } 1122 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize, 1123 (u_char *)); 1124 if (specific_buf == NULL) { 1125 RF_Free(k_cfg, sizeof(RF_Config_t)); 1126 return (ENOMEM); 1127 } 1128 retcode = copyin(k_cfg->layoutSpecific, specific_buf, 1129 k_cfg->layoutSpecificSize); 1130 if (retcode) { 1131 RF_Free(k_cfg, sizeof(RF_Config_t)); 1132 RF_Free(specific_buf, 1133 k_cfg->layoutSpecificSize); 1134 db1_printf(("rf_ioctl: retcode=%d copyin.2\n", 1135 retcode)); 1136 return (retcode); 1137 } 1138 } else 1139 specific_buf = NULL; 1140 k_cfg->layoutSpecific = specific_buf; 1141 1142 /* should do some kind of sanity check on the configuration. 1143 * Store the sum of all the bytes in the last byte? */ 1144 1145 /* configure the system */ 1146 1147 /* 1148 * Clear the entire RAID descriptor, just to make sure 1149 * there is no stale data left in the case of a 1150 * reconfiguration 1151 */ 1152 memset((char *) raidPtr, 0, sizeof(RF_Raid_t)); 1153 raidPtr->raidid = unit; 1154 1155 retcode = rf_Configure(raidPtr, k_cfg, NULL); 1156 1157 if (retcode == 0) { 1158 1159 /* allow this many simultaneous IO's to 1160 this RAID device */ 1161 raidPtr->openings = RAIDOUTSTANDING; 1162 1163 raidinit(raidPtr); 1164 rf_markalldirty(raidPtr); 1165 } 1166 /* free the buffers. No return code here. */ 1167 if (k_cfg->layoutSpecificSize) { 1168 RF_Free(specific_buf, k_cfg->layoutSpecificSize); 1169 } 1170 RF_Free(k_cfg, sizeof(RF_Config_t)); 1171 1172 return (retcode); 1173 1174 /* shutdown the system */ 1175 case RAIDFRAME_SHUTDOWN: 1176 1177 if ((error = raidlock(rs)) != 0) 1178 return (error); 1179 1180 /* 1181 * If somebody has a partition mounted, we shouldn't 1182 * shutdown. 1183 */ 1184 1185 part = DISKPART(dev); 1186 pmask = (1 << part); 1187 if ((rs->sc_dkdev.dk_openmask & ~pmask) || 1188 ((rs->sc_dkdev.dk_bopenmask & pmask) && 1189 (rs->sc_dkdev.dk_copenmask & pmask))) { 1190 raidunlock(rs); 1191 return (EBUSY); 1192 } 1193 1194 retcode = rf_Shutdown(raidPtr); 1195 1196 /* It's no longer initialized... */ 1197 rs->sc_flags &= ~RAIDF_INITED; 1198 1199 /* free the pseudo device attach bits */ 1200 1201 cf = device_cfdata(rs->sc_dev); 1202 /* XXX this causes us to not return any errors 1203 from the above call to rf_Shutdown() */ 1204 retcode = config_detach(rs->sc_dev, DETACH_QUIET); 1205 free(cf, M_RAIDFRAME); 1206 1207 /* Detach the disk. */ 1208 dkwedge_delall(&rs->sc_dkdev); 1209 disk_detach(&rs->sc_dkdev); 1210 disk_destroy(&rs->sc_dkdev); 1211 1212 raidunlock(rs); 1213 1214 return (retcode); 1215 case RAIDFRAME_GET_COMPONENT_LABEL: 1216 clabel_ptr = (RF_ComponentLabel_t **) data; 1217 /* need to read the component label for the disk indicated 1218 by row,column in clabel */ 1219 1220 /* 1221 * Perhaps there should be an option to skip the in-core 1222 * copy and hit the disk, as with disklabel(8). 1223 */ 1224 RF_Malloc(clabel, sizeof(*clabel), (RF_ComponentLabel_t *)); 1225 1226 retcode = copyin( *clabel_ptr, clabel, 1227 sizeof(RF_ComponentLabel_t)); 1228 1229 if (retcode) { 1230 return(retcode); 1231 } 1232 1233 clabel->row = 0; /* Don't allow looking at anything else.*/ 1234 1235 column = clabel->column; 1236 1237 if ((column < 0) || (column >= raidPtr->numCol + 1238 raidPtr->numSpare)) { 1239 return(EINVAL); 1240 } 1241 1242 RF_Free(clabel, sizeof(*clabel)); 1243 1244 clabel = raidget_component_label(raidPtr, column); 1245 1246 if (retcode == 0) { 1247 retcode = copyout(clabel, *clabel_ptr, 1248 sizeof(RF_ComponentLabel_t)); 1249 } 1250 return (retcode); 1251 1252 #if 0 1253 case RAIDFRAME_SET_COMPONENT_LABEL: 1254 clabel = (RF_ComponentLabel_t *) data; 1255 1256 /* XXX check the label for valid stuff... */ 1257 /* Note that some things *should not* get modified -- 1258 the user should be re-initing the labels instead of 1259 trying to patch things. 1260 */ 1261 1262 raidid = raidPtr->raidid; 1263 #ifdef DEBUG 1264 printf("raid%d: Got component label:\n", raidid); 1265 printf("raid%d: Version: %d\n", raidid, clabel->version); 1266 printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number); 1267 printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter); 1268 printf("raid%d: Column: %d\n", raidid, clabel->column); 1269 printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns); 1270 printf("raid%d: Clean: %d\n", raidid, clabel->clean); 1271 printf("raid%d: Status: %d\n", raidid, clabel->status); 1272 #endif 1273 clabel->row = 0; 1274 column = clabel->column; 1275 1276 if ((column < 0) || (column >= raidPtr->numCol)) { 1277 return(EINVAL); 1278 } 1279 1280 /* XXX this isn't allowed to do anything for now :-) */ 1281 1282 /* XXX and before it is, we need to fill in the rest 1283 of the fields!?!?!?! */ 1284 memcpy(raidget_component_label(raidPtr, column), 1285 clabel, sizeof(*clabel)); 1286 raidflush_component_label(raidPtr, column); 1287 return (0); 1288 #endif 1289 1290 case RAIDFRAME_INIT_LABELS: 1291 clabel = (RF_ComponentLabel_t *) data; 1292 /* 1293 we only want the serial number from 1294 the above. We get all the rest of the information 1295 from the config that was used to create this RAID 1296 set. 1297 */ 1298 1299 raidPtr->serial_number = clabel->serial_number; 1300 1301 for(column=0;column<raidPtr->numCol;column++) { 1302 diskPtr = &raidPtr->Disks[column]; 1303 if (!RF_DEAD_DISK(diskPtr->status)) { 1304 ci_label = raidget_component_label(raidPtr, 1305 column); 1306 /* Zeroing this is important. */ 1307 memset(ci_label, 0, sizeof(*ci_label)); 1308 raid_init_component_label(raidPtr, ci_label); 1309 ci_label->serial_number = 1310 raidPtr->serial_number; 1311 ci_label->row = 0; /* we dont' pretend to support more */ 1312 rf_component_label_set_partitionsize(ci_label, 1313 diskPtr->partitionSize); 1314 ci_label->column = column; 1315 raidflush_component_label(raidPtr, column); 1316 } 1317 /* XXXjld what about the spares? */ 1318 } 1319 1320 return (retcode); 1321 case RAIDFRAME_SET_AUTOCONFIG: 1322 d = rf_set_autoconfig(raidPtr, *(int *) data); 1323 printf("raid%d: New autoconfig value is: %d\n", 1324 raidPtr->raidid, d); 1325 *(int *) data = d; 1326 return (retcode); 1327 1328 case RAIDFRAME_SET_ROOT: 1329 d = rf_set_rootpartition(raidPtr, *(int *) data); 1330 printf("raid%d: New rootpartition value is: %d\n", 1331 raidPtr->raidid, d); 1332 *(int *) data = d; 1333 return (retcode); 1334 1335 /* initialize all parity */ 1336 case RAIDFRAME_REWRITEPARITY: 1337 1338 if (raidPtr->Layout.map->faultsTolerated == 0) { 1339 /* Parity for RAID 0 is trivially correct */ 1340 raidPtr->parity_good = RF_RAID_CLEAN; 1341 return(0); 1342 } 1343 1344 if (raidPtr->parity_rewrite_in_progress == 1) { 1345 /* Re-write is already in progress! */ 1346 return(EINVAL); 1347 } 1348 1349 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1350 rf_RewriteParityThread, 1351 raidPtr,"raid_parity"); 1352 return (retcode); 1353 1354 1355 case RAIDFRAME_ADD_HOT_SPARE: 1356 sparePtr = (RF_SingleComponent_t *) data; 1357 memcpy( &component, sparePtr, sizeof(RF_SingleComponent_t)); 1358 retcode = rf_add_hot_spare(raidPtr, &component); 1359 return(retcode); 1360 1361 case RAIDFRAME_REMOVE_HOT_SPARE: 1362 return(retcode); 1363 1364 case RAIDFRAME_DELETE_COMPONENT: 1365 componentPtr = (RF_SingleComponent_t *)data; 1366 memcpy( &component, componentPtr, 1367 sizeof(RF_SingleComponent_t)); 1368 retcode = rf_delete_component(raidPtr, &component); 1369 return(retcode); 1370 1371 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1372 componentPtr = (RF_SingleComponent_t *)data; 1373 memcpy( &component, componentPtr, 1374 sizeof(RF_SingleComponent_t)); 1375 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1376 return(retcode); 1377 1378 case RAIDFRAME_REBUILD_IN_PLACE: 1379 1380 if (raidPtr->Layout.map->faultsTolerated == 0) { 1381 /* Can't do this on a RAID 0!! */ 1382 return(EINVAL); 1383 } 1384 1385 if (raidPtr->recon_in_progress == 1) { 1386 /* a reconstruct is already in progress! */ 1387 return(EINVAL); 1388 } 1389 1390 componentPtr = (RF_SingleComponent_t *) data; 1391 memcpy( &component, componentPtr, 1392 sizeof(RF_SingleComponent_t)); 1393 component.row = 0; /* we don't support any more */ 1394 column = component.column; 1395 1396 if ((column < 0) || (column >= raidPtr->numCol)) { 1397 return(EINVAL); 1398 } 1399 1400 RF_LOCK_MUTEX(raidPtr->mutex); 1401 if ((raidPtr->Disks[column].status == rf_ds_optimal) && 1402 (raidPtr->numFailures > 0)) { 1403 /* XXX 0 above shouldn't be constant!!! */ 1404 /* some component other than this has failed. 1405 Let's not make things worse than they already 1406 are... */ 1407 printf("raid%d: Unable to reconstruct to disk at:\n", 1408 raidPtr->raidid); 1409 printf("raid%d: Col: %d Too many failures.\n", 1410 raidPtr->raidid, column); 1411 RF_UNLOCK_MUTEX(raidPtr->mutex); 1412 return (EINVAL); 1413 } 1414 if (raidPtr->Disks[column].status == 1415 rf_ds_reconstructing) { 1416 printf("raid%d: Unable to reconstruct to disk at:\n", 1417 raidPtr->raidid); 1418 printf("raid%d: Col: %d Reconstruction already occuring!\n", raidPtr->raidid, column); 1419 1420 RF_UNLOCK_MUTEX(raidPtr->mutex); 1421 return (EINVAL); 1422 } 1423 if (raidPtr->Disks[column].status == rf_ds_spared) { 1424 RF_UNLOCK_MUTEX(raidPtr->mutex); 1425 return (EINVAL); 1426 } 1427 RF_UNLOCK_MUTEX(raidPtr->mutex); 1428 1429 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1430 if (rrcopy == NULL) 1431 return(ENOMEM); 1432 1433 rrcopy->raidPtr = (void *) raidPtr; 1434 rrcopy->col = column; 1435 1436 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1437 rf_ReconstructInPlaceThread, 1438 rrcopy,"raid_reconip"); 1439 return(retcode); 1440 1441 case RAIDFRAME_GET_INFO: 1442 if (!raidPtr->valid) 1443 return (ENODEV); 1444 ucfgp = (RF_DeviceConfig_t **) data; 1445 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1446 (RF_DeviceConfig_t *)); 1447 if (d_cfg == NULL) 1448 return (ENOMEM); 1449 d_cfg->rows = 1; /* there is only 1 row now */ 1450 d_cfg->cols = raidPtr->numCol; 1451 d_cfg->ndevs = raidPtr->numCol; 1452 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1453 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1454 return (ENOMEM); 1455 } 1456 d_cfg->nspares = raidPtr->numSpare; 1457 if (d_cfg->nspares >= RF_MAX_DISKS) { 1458 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1459 return (ENOMEM); 1460 } 1461 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1462 d = 0; 1463 for (j = 0; j < d_cfg->cols; j++) { 1464 d_cfg->devs[d] = raidPtr->Disks[j]; 1465 d++; 1466 } 1467 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1468 d_cfg->spares[i] = raidPtr->Disks[j]; 1469 if (d_cfg->spares[i].status == rf_ds_rebuilding_spare) { 1470 /* XXX: raidctl(8) expects to see this as a used spare */ 1471 d_cfg->spares[i].status = rf_ds_used_spare; 1472 } 1473 } 1474 retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t)); 1475 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1476 1477 return (retcode); 1478 1479 case RAIDFRAME_CHECK_PARITY: 1480 *(int *) data = raidPtr->parity_good; 1481 return (0); 1482 1483 case RAIDFRAME_PARITYMAP_STATUS: 1484 rf_paritymap_status(raidPtr->parity_map, 1485 (struct rf_pmstat *)data); 1486 return 0; 1487 1488 case RAIDFRAME_PARITYMAP_SET_PARAMS: 1489 if (raidPtr->parity_map == NULL) 1490 return ENOENT; /* ??? */ 1491 if (0 != rf_paritymap_set_params(raidPtr->parity_map, 1492 (struct rf_pmparams *)data, 1)) 1493 return EINVAL; 1494 return 0; 1495 1496 case RAIDFRAME_PARITYMAP_GET_DISABLE: 1497 *(int *) data = rf_paritymap_get_disable(raidPtr); 1498 return 0; 1499 1500 case RAIDFRAME_PARITYMAP_SET_DISABLE: 1501 rf_paritymap_set_disable(raidPtr, *(int *)data); 1502 /* XXX should errors be passed up? */ 1503 return 0; 1504 1505 case RAIDFRAME_RESET_ACCTOTALS: 1506 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1507 return (0); 1508 1509 case RAIDFRAME_GET_ACCTOTALS: 1510 totals = (RF_AccTotals_t *) data; 1511 *totals = raidPtr->acc_totals; 1512 return (0); 1513 1514 case RAIDFRAME_KEEP_ACCTOTALS: 1515 raidPtr->keep_acc_totals = *(int *)data; 1516 return (0); 1517 1518 case RAIDFRAME_GET_SIZE: 1519 *(int *) data = raidPtr->totalSectors; 1520 return (0); 1521 1522 /* fail a disk & optionally start reconstruction */ 1523 case RAIDFRAME_FAIL_DISK: 1524 1525 if (raidPtr->Layout.map->faultsTolerated == 0) { 1526 /* Can't do this on a RAID 0!! */ 1527 return(EINVAL); 1528 } 1529 1530 rr = (struct rf_recon_req *) data; 1531 rr->row = 0; 1532 if (rr->col < 0 || rr->col >= raidPtr->numCol) 1533 return (EINVAL); 1534 1535 1536 RF_LOCK_MUTEX(raidPtr->mutex); 1537 if (raidPtr->status == rf_rs_reconstructing) { 1538 /* you can't fail a disk while we're reconstructing! */ 1539 /* XXX wrong for RAID6 */ 1540 RF_UNLOCK_MUTEX(raidPtr->mutex); 1541 return (EINVAL); 1542 } 1543 if ((raidPtr->Disks[rr->col].status == 1544 rf_ds_optimal) && (raidPtr->numFailures > 0)) { 1545 /* some other component has failed. Let's not make 1546 things worse. XXX wrong for RAID6 */ 1547 RF_UNLOCK_MUTEX(raidPtr->mutex); 1548 return (EINVAL); 1549 } 1550 if (raidPtr->Disks[rr->col].status == rf_ds_spared) { 1551 /* Can't fail a spared disk! */ 1552 RF_UNLOCK_MUTEX(raidPtr->mutex); 1553 return (EINVAL); 1554 } 1555 RF_UNLOCK_MUTEX(raidPtr->mutex); 1556 1557 /* make a copy of the recon request so that we don't rely on 1558 * the user's buffer */ 1559 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1560 if (rrcopy == NULL) 1561 return(ENOMEM); 1562 memcpy(rrcopy, rr, sizeof(*rr)); 1563 rrcopy->raidPtr = (void *) raidPtr; 1564 1565 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1566 rf_ReconThread, 1567 rrcopy,"raid_recon"); 1568 return (0); 1569 1570 /* invoke a copyback operation after recon on whatever disk 1571 * needs it, if any */ 1572 case RAIDFRAME_COPYBACK: 1573 1574 if (raidPtr->Layout.map->faultsTolerated == 0) { 1575 /* This makes no sense on a RAID 0!! */ 1576 return(EINVAL); 1577 } 1578 1579 if (raidPtr->copyback_in_progress == 1) { 1580 /* Copyback is already in progress! */ 1581 return(EINVAL); 1582 } 1583 1584 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1585 rf_CopybackThread, 1586 raidPtr,"raid_copyback"); 1587 return (retcode); 1588 1589 /* return the percentage completion of reconstruction */ 1590 case RAIDFRAME_CHECK_RECON_STATUS: 1591 if (raidPtr->Layout.map->faultsTolerated == 0) { 1592 /* This makes no sense on a RAID 0, so tell the 1593 user it's done. */ 1594 *(int *) data = 100; 1595 return(0); 1596 } 1597 if (raidPtr->status != rf_rs_reconstructing) 1598 *(int *) data = 100; 1599 else { 1600 if (raidPtr->reconControl->numRUsTotal > 0) { 1601 *(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal); 1602 } else { 1603 *(int *) data = 0; 1604 } 1605 } 1606 return (0); 1607 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1608 progressInfoPtr = (RF_ProgressInfo_t **) data; 1609 if (raidPtr->status != rf_rs_reconstructing) { 1610 progressInfo.remaining = 0; 1611 progressInfo.completed = 100; 1612 progressInfo.total = 100; 1613 } else { 1614 progressInfo.total = 1615 raidPtr->reconControl->numRUsTotal; 1616 progressInfo.completed = 1617 raidPtr->reconControl->numRUsComplete; 1618 progressInfo.remaining = progressInfo.total - 1619 progressInfo.completed; 1620 } 1621 retcode = copyout(&progressInfo, *progressInfoPtr, 1622 sizeof(RF_ProgressInfo_t)); 1623 return (retcode); 1624 1625 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1626 if (raidPtr->Layout.map->faultsTolerated == 0) { 1627 /* This makes no sense on a RAID 0, so tell the 1628 user it's done. */ 1629 *(int *) data = 100; 1630 return(0); 1631 } 1632 if (raidPtr->parity_rewrite_in_progress == 1) { 1633 *(int *) data = 100 * 1634 raidPtr->parity_rewrite_stripes_done / 1635 raidPtr->Layout.numStripe; 1636 } else { 1637 *(int *) data = 100; 1638 } 1639 return (0); 1640 1641 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1642 progressInfoPtr = (RF_ProgressInfo_t **) data; 1643 if (raidPtr->parity_rewrite_in_progress == 1) { 1644 progressInfo.total = raidPtr->Layout.numStripe; 1645 progressInfo.completed = 1646 raidPtr->parity_rewrite_stripes_done; 1647 progressInfo.remaining = progressInfo.total - 1648 progressInfo.completed; 1649 } else { 1650 progressInfo.remaining = 0; 1651 progressInfo.completed = 100; 1652 progressInfo.total = 100; 1653 } 1654 retcode = copyout(&progressInfo, *progressInfoPtr, 1655 sizeof(RF_ProgressInfo_t)); 1656 return (retcode); 1657 1658 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1659 if (raidPtr->Layout.map->faultsTolerated == 0) { 1660 /* This makes no sense on a RAID 0 */ 1661 *(int *) data = 100; 1662 return(0); 1663 } 1664 if (raidPtr->copyback_in_progress == 1) { 1665 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1666 raidPtr->Layout.numStripe; 1667 } else { 1668 *(int *) data = 100; 1669 } 1670 return (0); 1671 1672 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1673 progressInfoPtr = (RF_ProgressInfo_t **) data; 1674 if (raidPtr->copyback_in_progress == 1) { 1675 progressInfo.total = raidPtr->Layout.numStripe; 1676 progressInfo.completed = 1677 raidPtr->copyback_stripes_done; 1678 progressInfo.remaining = progressInfo.total - 1679 progressInfo.completed; 1680 } else { 1681 progressInfo.remaining = 0; 1682 progressInfo.completed = 100; 1683 progressInfo.total = 100; 1684 } 1685 retcode = copyout(&progressInfo, *progressInfoPtr, 1686 sizeof(RF_ProgressInfo_t)); 1687 return (retcode); 1688 1689 /* the sparetable daemon calls this to wait for the kernel to 1690 * need a spare table. this ioctl does not return until a 1691 * spare table is needed. XXX -- calling mpsleep here in the 1692 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1693 * -- I should either compute the spare table in the kernel, 1694 * or have a different -- XXX XXX -- interface (a different 1695 * character device) for delivering the table -- XXX */ 1696 #if 0 1697 case RAIDFRAME_SPARET_WAIT: 1698 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1699 while (!rf_sparet_wait_queue) 1700 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE); 1701 waitreq = rf_sparet_wait_queue; 1702 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1703 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1704 1705 /* structure assignment */ 1706 *((RF_SparetWait_t *) data) = *waitreq; 1707 1708 RF_Free(waitreq, sizeof(*waitreq)); 1709 return (0); 1710 1711 /* wakes up a process waiting on SPARET_WAIT and puts an error 1712 * code in it that will cause the dameon to exit */ 1713 case RAIDFRAME_ABORT_SPARET_WAIT: 1714 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1715 waitreq->fcol = -1; 1716 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1717 waitreq->next = rf_sparet_wait_queue; 1718 rf_sparet_wait_queue = waitreq; 1719 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1720 wakeup(&rf_sparet_wait_queue); 1721 return (0); 1722 1723 /* used by the spare table daemon to deliver a spare table 1724 * into the kernel */ 1725 case RAIDFRAME_SEND_SPARET: 1726 1727 /* install the spare table */ 1728 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1729 1730 /* respond to the requestor. the return status of the spare 1731 * table installation is passed in the "fcol" field */ 1732 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1733 waitreq->fcol = retcode; 1734 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1735 waitreq->next = rf_sparet_resp_queue; 1736 rf_sparet_resp_queue = waitreq; 1737 wakeup(&rf_sparet_resp_queue); 1738 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1739 1740 return (retcode); 1741 #endif 1742 1743 default: 1744 break; /* fall through to the os-specific code below */ 1745 1746 } 1747 1748 if (!raidPtr->valid) 1749 return (EINVAL); 1750 1751 /* 1752 * Add support for "regular" device ioctls here. 1753 */ 1754 1755 switch (cmd) { 1756 case DIOCGDINFO: 1757 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1758 break; 1759 #ifdef __HAVE_OLD_DISKLABEL 1760 case ODIOCGDINFO: 1761 newlabel = *(rs->sc_dkdev.dk_label); 1762 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1763 return ENOTTY; 1764 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1765 break; 1766 #endif 1767 1768 case DIOCGPART: 1769 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1770 ((struct partinfo *) data)->part = 1771 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1772 break; 1773 1774 case DIOCWDINFO: 1775 case DIOCSDINFO: 1776 #ifdef __HAVE_OLD_DISKLABEL 1777 case ODIOCWDINFO: 1778 case ODIOCSDINFO: 1779 #endif 1780 { 1781 struct disklabel *lp; 1782 #ifdef __HAVE_OLD_DISKLABEL 1783 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1784 memset(&newlabel, 0, sizeof newlabel); 1785 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1786 lp = &newlabel; 1787 } else 1788 #endif 1789 lp = (struct disklabel *)data; 1790 1791 if ((error = raidlock(rs)) != 0) 1792 return (error); 1793 1794 rs->sc_flags |= RAIDF_LABELLING; 1795 1796 error = setdisklabel(rs->sc_dkdev.dk_label, 1797 lp, 0, rs->sc_dkdev.dk_cpulabel); 1798 if (error == 0) { 1799 if (cmd == DIOCWDINFO 1800 #ifdef __HAVE_OLD_DISKLABEL 1801 || cmd == ODIOCWDINFO 1802 #endif 1803 ) 1804 error = writedisklabel(RAIDLABELDEV(dev), 1805 raidstrategy, rs->sc_dkdev.dk_label, 1806 rs->sc_dkdev.dk_cpulabel); 1807 } 1808 rs->sc_flags &= ~RAIDF_LABELLING; 1809 1810 raidunlock(rs); 1811 1812 if (error) 1813 return (error); 1814 break; 1815 } 1816 1817 case DIOCWLABEL: 1818 if (*(int *) data != 0) 1819 rs->sc_flags |= RAIDF_WLABEL; 1820 else 1821 rs->sc_flags &= ~RAIDF_WLABEL; 1822 break; 1823 1824 case DIOCGDEFLABEL: 1825 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1826 break; 1827 1828 #ifdef __HAVE_OLD_DISKLABEL 1829 case ODIOCGDEFLABEL: 1830 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1831 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1832 return ENOTTY; 1833 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1834 break; 1835 #endif 1836 1837 case DIOCAWEDGE: 1838 case DIOCDWEDGE: 1839 dkw = (void *)data; 1840 1841 /* If the ioctl happens here, the parent is us. */ 1842 (void)strcpy(dkw->dkw_parent, rs->sc_xname); 1843 return cmd == DIOCAWEDGE ? dkwedge_add(dkw) : dkwedge_del(dkw); 1844 1845 case DIOCLWEDGES: 1846 return dkwedge_list(&rs->sc_dkdev, 1847 (struct dkwedge_list *)data, l); 1848 case DIOCCACHESYNC: 1849 return rf_sync_component_caches(raidPtr); 1850 1851 case DIOCGSTRATEGY: 1852 { 1853 struct disk_strategy *dks = (void *)data; 1854 1855 s = splbio(); 1856 strlcpy(dks->dks_name, bufq_getstrategyname(rs->buf_queue), 1857 sizeof(dks->dks_name)); 1858 splx(s); 1859 dks->dks_paramlen = 0; 1860 1861 return 0; 1862 } 1863 1864 case DIOCSSTRATEGY: 1865 { 1866 struct disk_strategy *dks = (void *)data; 1867 struct bufq_state *new; 1868 struct bufq_state *old; 1869 1870 if (dks->dks_param != NULL) { 1871 return EINVAL; 1872 } 1873 dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */ 1874 error = bufq_alloc(&new, dks->dks_name, 1875 BUFQ_EXACT|BUFQ_SORT_RAWBLOCK); 1876 if (error) { 1877 return error; 1878 } 1879 s = splbio(); 1880 old = rs->buf_queue; 1881 bufq_move(new, old); 1882 rs->buf_queue = new; 1883 splx(s); 1884 bufq_free(old); 1885 1886 return 0; 1887 } 1888 1889 default: 1890 retcode = ENOTTY; 1891 } 1892 return (retcode); 1893 1894 } 1895 1896 1897 /* raidinit -- complete the rest of the initialization for the 1898 RAIDframe device. */ 1899 1900 1901 static void 1902 raidinit(RF_Raid_t *raidPtr) 1903 { 1904 struct cfdata *cf; 1905 struct raid_softc *rs; 1906 int unit; 1907 1908 unit = raidPtr->raidid; 1909 1910 rs = &raid_softc[unit]; 1911 1912 /* XXX should check return code first... */ 1913 rs->sc_flags |= RAIDF_INITED; 1914 1915 /* XXX doesn't check bounds. */ 1916 snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit); 1917 1918 /* attach the pseudo device */ 1919 cf = malloc(sizeof(*cf), M_RAIDFRAME, M_WAITOK); 1920 cf->cf_name = raid_cd.cd_name; 1921 cf->cf_atname = raid_cd.cd_name; 1922 cf->cf_unit = unit; 1923 cf->cf_fstate = FSTATE_STAR; 1924 1925 rs->sc_dev = config_attach_pseudo(cf); 1926 1927 if (rs->sc_dev==NULL) { 1928 printf("raid%d: config_attach_pseudo failed\n", 1929 raidPtr->raidid); 1930 } 1931 1932 /* disk_attach actually creates space for the CPU disklabel, among 1933 * other things, so it's critical to call this *BEFORE* we try putzing 1934 * with disklabels. */ 1935 1936 disk_init(&rs->sc_dkdev, rs->sc_xname, &rf_dkdriver); 1937 disk_attach(&rs->sc_dkdev); 1938 disk_blocksize(&rs->sc_dkdev, raidPtr->bytesPerSector); 1939 1940 /* XXX There may be a weird interaction here between this, and 1941 * protectedSectors, as used in RAIDframe. */ 1942 1943 rs->sc_size = raidPtr->totalSectors; 1944 1945 dkwedge_discover(&rs->sc_dkdev); 1946 1947 rf_set_properties(rs, raidPtr); 1948 1949 } 1950 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 1951 /* wake up the daemon & tell it to get us a spare table 1952 * XXX 1953 * the entries in the queues should be tagged with the raidPtr 1954 * so that in the extremely rare case that two recons happen at once, 1955 * we know for which device were requesting a spare table 1956 * XXX 1957 * 1958 * XXX This code is not currently used. GO 1959 */ 1960 int 1961 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req) 1962 { 1963 int retcode; 1964 1965 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1966 req->next = rf_sparet_wait_queue; 1967 rf_sparet_wait_queue = req; 1968 wakeup(&rf_sparet_wait_queue); 1969 1970 /* mpsleep unlocks the mutex */ 1971 while (!rf_sparet_resp_queue) { 1972 tsleep(&rf_sparet_resp_queue, PRIBIO, 1973 "raidframe getsparetable", 0); 1974 } 1975 req = rf_sparet_resp_queue; 1976 rf_sparet_resp_queue = req->next; 1977 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1978 1979 retcode = req->fcol; 1980 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1981 * alloc'd */ 1982 return (retcode); 1983 } 1984 #endif 1985 1986 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1987 * bp & passes it down. 1988 * any calls originating in the kernel must use non-blocking I/O 1989 * do some extra sanity checking to return "appropriate" error values for 1990 * certain conditions (to make some standard utilities work) 1991 * 1992 * Formerly known as: rf_DoAccessKernel 1993 */ 1994 void 1995 raidstart(RF_Raid_t *raidPtr) 1996 { 1997 RF_SectorCount_t num_blocks, pb, sum; 1998 RF_RaidAddr_t raid_addr; 1999 struct partition *pp; 2000 daddr_t blocknum; 2001 int unit; 2002 struct raid_softc *rs; 2003 int do_async; 2004 struct buf *bp; 2005 int rc; 2006 2007 unit = raidPtr->raidid; 2008 rs = &raid_softc[unit]; 2009 2010 /* quick check to see if anything has died recently */ 2011 RF_LOCK_MUTEX(raidPtr->mutex); 2012 if (raidPtr->numNewFailures > 0) { 2013 RF_UNLOCK_MUTEX(raidPtr->mutex); 2014 rf_update_component_labels(raidPtr, 2015 RF_NORMAL_COMPONENT_UPDATE); 2016 RF_LOCK_MUTEX(raidPtr->mutex); 2017 raidPtr->numNewFailures--; 2018 } 2019 2020 /* Check to see if we're at the limit... */ 2021 while (raidPtr->openings > 0) { 2022 RF_UNLOCK_MUTEX(raidPtr->mutex); 2023 2024 /* get the next item, if any, from the queue */ 2025 if ((bp = BUFQ_GET(rs->buf_queue)) == NULL) { 2026 /* nothing more to do */ 2027 return; 2028 } 2029 2030 /* Ok, for the bp we have here, bp->b_blkno is relative to the 2031 * partition.. Need to make it absolute to the underlying 2032 * device.. */ 2033 2034 blocknum = bp->b_blkno << DEV_BSHIFT >> raidPtr->logBytesPerSector; 2035 if (DISKPART(bp->b_dev) != RAW_PART) { 2036 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 2037 blocknum += pp->p_offset; 2038 } 2039 2040 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 2041 (int) blocknum)); 2042 2043 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 2044 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 2045 2046 /* *THIS* is where we adjust what block we're going to... 2047 * but DO NOT TOUCH bp->b_blkno!!! */ 2048 raid_addr = blocknum; 2049 2050 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 2051 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 2052 sum = raid_addr + num_blocks + pb; 2053 if (1 || rf_debugKernelAccess) { 2054 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 2055 (int) raid_addr, (int) sum, (int) num_blocks, 2056 (int) pb, (int) bp->b_resid)); 2057 } 2058 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 2059 || (sum < num_blocks) || (sum < pb)) { 2060 bp->b_error = ENOSPC; 2061 bp->b_resid = bp->b_bcount; 2062 biodone(bp); 2063 RF_LOCK_MUTEX(raidPtr->mutex); 2064 continue; 2065 } 2066 /* 2067 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 2068 */ 2069 2070 if (bp->b_bcount & raidPtr->sectorMask) { 2071 bp->b_error = EINVAL; 2072 bp->b_resid = bp->b_bcount; 2073 biodone(bp); 2074 RF_LOCK_MUTEX(raidPtr->mutex); 2075 continue; 2076 2077 } 2078 db1_printf(("Calling DoAccess..\n")); 2079 2080 2081 RF_LOCK_MUTEX(raidPtr->mutex); 2082 raidPtr->openings--; 2083 RF_UNLOCK_MUTEX(raidPtr->mutex); 2084 2085 /* 2086 * Everything is async. 2087 */ 2088 do_async = 1; 2089 2090 disk_busy(&rs->sc_dkdev); 2091 2092 /* XXX we're still at splbio() here... do we *really* 2093 need to be? */ 2094 2095 /* don't ever condition on bp->b_flags & B_WRITE. 2096 * always condition on B_READ instead */ 2097 2098 rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 2099 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 2100 do_async, raid_addr, num_blocks, 2101 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 2102 2103 if (rc) { 2104 bp->b_error = rc; 2105 bp->b_resid = bp->b_bcount; 2106 biodone(bp); 2107 /* continue loop */ 2108 } 2109 2110 RF_LOCK_MUTEX(raidPtr->mutex); 2111 } 2112 RF_UNLOCK_MUTEX(raidPtr->mutex); 2113 } 2114 2115 2116 2117 2118 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 2119 2120 int 2121 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req) 2122 { 2123 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 2124 struct buf *bp; 2125 2126 req->queue = queue; 2127 2128 #if DIAGNOSTIC 2129 if (queue->raidPtr->raidid >= numraid) { 2130 printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid, 2131 numraid); 2132 panic("Invalid Unit number in rf_DispatchKernelIO"); 2133 } 2134 #endif 2135 2136 bp = req->bp; 2137 2138 switch (req->type) { 2139 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 2140 /* XXX need to do something extra here.. */ 2141 /* I'm leaving this in, as I've never actually seen it used, 2142 * and I'd like folks to report it... GO */ 2143 printf(("WAKEUP CALLED\n")); 2144 queue->numOutstanding++; 2145 2146 bp->b_flags = 0; 2147 bp->b_private = req; 2148 2149 KernelWakeupFunc(bp); 2150 break; 2151 2152 case RF_IO_TYPE_READ: 2153 case RF_IO_TYPE_WRITE: 2154 #if RF_ACC_TRACE > 0 2155 if (req->tracerec) { 2156 RF_ETIMER_START(req->tracerec->timer); 2157 } 2158 #endif 2159 InitBP(bp, queue->rf_cinfo->ci_vp, 2160 op, queue->rf_cinfo->ci_dev, 2161 req->sectorOffset, req->numSector, 2162 req->buf, KernelWakeupFunc, (void *) req, 2163 queue->raidPtr->logBytesPerSector, req->b_proc); 2164 2165 if (rf_debugKernelAccess) { 2166 db1_printf(("dispatch: bp->b_blkno = %ld\n", 2167 (long) bp->b_blkno)); 2168 } 2169 queue->numOutstanding++; 2170 queue->last_deq_sector = req->sectorOffset; 2171 /* acc wouldn't have been let in if there were any pending 2172 * reqs at any other priority */ 2173 queue->curPriority = req->priority; 2174 2175 db1_printf(("Going for %c to unit %d col %d\n", 2176 req->type, queue->raidPtr->raidid, 2177 queue->col)); 2178 db1_printf(("sector %d count %d (%d bytes) %d\n", 2179 (int) req->sectorOffset, (int) req->numSector, 2180 (int) (req->numSector << 2181 queue->raidPtr->logBytesPerSector), 2182 (int) queue->raidPtr->logBytesPerSector)); 2183 2184 /* 2185 * XXX: drop lock here since this can block at 2186 * least with backing SCSI devices. Retake it 2187 * to minimize fuss with calling interfaces. 2188 */ 2189 2190 RF_UNLOCK_QUEUE_MUTEX(queue, "unusedparam"); 2191 bdev_strategy(bp); 2192 RF_LOCK_QUEUE_MUTEX(queue, "unusedparam"); 2193 break; 2194 2195 default: 2196 panic("bad req->type in rf_DispatchKernelIO"); 2197 } 2198 db1_printf(("Exiting from DispatchKernelIO\n")); 2199 2200 return (0); 2201 } 2202 /* this is the callback function associated with a I/O invoked from 2203 kernel code. 2204 */ 2205 static void 2206 KernelWakeupFunc(struct buf *bp) 2207 { 2208 RF_DiskQueueData_t *req = NULL; 2209 RF_DiskQueue_t *queue; 2210 int s; 2211 2212 s = splbio(); 2213 db1_printf(("recovering the request queue:\n")); 2214 req = bp->b_private; 2215 2216 queue = (RF_DiskQueue_t *) req->queue; 2217 2218 #if RF_ACC_TRACE > 0 2219 if (req->tracerec) { 2220 RF_ETIMER_STOP(req->tracerec->timer); 2221 RF_ETIMER_EVAL(req->tracerec->timer); 2222 RF_LOCK_MUTEX(rf_tracing_mutex); 2223 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2224 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2225 req->tracerec->num_phys_ios++; 2226 RF_UNLOCK_MUTEX(rf_tracing_mutex); 2227 } 2228 #endif 2229 2230 /* XXX Ok, let's get aggressive... If b_error is set, let's go 2231 * ballistic, and mark the component as hosed... */ 2232 2233 if (bp->b_error != 0) { 2234 /* Mark the disk as dead */ 2235 /* but only mark it once... */ 2236 /* and only if it wouldn't leave this RAID set 2237 completely broken */ 2238 if (((queue->raidPtr->Disks[queue->col].status == 2239 rf_ds_optimal) || 2240 (queue->raidPtr->Disks[queue->col].status == 2241 rf_ds_used_spare)) && 2242 (queue->raidPtr->numFailures < 2243 queue->raidPtr->Layout.map->faultsTolerated)) { 2244 printf("raid%d: IO Error. Marking %s as failed.\n", 2245 queue->raidPtr->raidid, 2246 queue->raidPtr->Disks[queue->col].devname); 2247 queue->raidPtr->Disks[queue->col].status = 2248 rf_ds_failed; 2249 queue->raidPtr->status = rf_rs_degraded; 2250 queue->raidPtr->numFailures++; 2251 queue->raidPtr->numNewFailures++; 2252 } else { /* Disk is already dead... */ 2253 /* printf("Disk already marked as dead!\n"); */ 2254 } 2255 2256 } 2257 2258 /* Fill in the error value */ 2259 2260 req->error = bp->b_error; 2261 2262 simple_lock(&queue->raidPtr->iodone_lock); 2263 2264 /* Drop this one on the "finished" queue... */ 2265 TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries); 2266 2267 /* Let the raidio thread know there is work to be done. */ 2268 wakeup(&(queue->raidPtr->iodone)); 2269 2270 simple_unlock(&queue->raidPtr->iodone_lock); 2271 2272 splx(s); 2273 } 2274 2275 2276 2277 /* 2278 * initialize a buf structure for doing an I/O in the kernel. 2279 */ 2280 static void 2281 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev, 2282 RF_SectorNum_t startSect, RF_SectorCount_t numSect, void *bf, 2283 void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector, 2284 struct proc *b_proc) 2285 { 2286 /* bp->b_flags = B_PHYS | rw_flag; */ 2287 bp->b_flags = rw_flag; /* XXX need B_PHYS here too??? */ 2288 bp->b_oflags = 0; 2289 bp->b_cflags = 0; 2290 bp->b_bcount = numSect << logBytesPerSector; 2291 bp->b_bufsize = bp->b_bcount; 2292 bp->b_error = 0; 2293 bp->b_dev = dev; 2294 bp->b_data = bf; 2295 bp->b_blkno = startSect << logBytesPerSector >> DEV_BSHIFT; 2296 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 2297 if (bp->b_bcount == 0) { 2298 panic("bp->b_bcount is zero in InitBP!!"); 2299 } 2300 bp->b_proc = b_proc; 2301 bp->b_iodone = cbFunc; 2302 bp->b_private = cbArg; 2303 } 2304 2305 static void 2306 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs, 2307 struct disklabel *lp) 2308 { 2309 memset(lp, 0, sizeof(*lp)); 2310 2311 /* fabricate a label... */ 2312 lp->d_secperunit = raidPtr->totalSectors; 2313 lp->d_secsize = raidPtr->bytesPerSector; 2314 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 2315 lp->d_ntracks = 4 * raidPtr->numCol; 2316 lp->d_ncylinders = raidPtr->totalSectors / 2317 (lp->d_nsectors * lp->d_ntracks); 2318 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 2319 2320 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 2321 lp->d_type = DTYPE_RAID; 2322 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 2323 lp->d_rpm = 3600; 2324 lp->d_interleave = 1; 2325 lp->d_flags = 0; 2326 2327 lp->d_partitions[RAW_PART].p_offset = 0; 2328 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 2329 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 2330 lp->d_npartitions = RAW_PART + 1; 2331 2332 lp->d_magic = DISKMAGIC; 2333 lp->d_magic2 = DISKMAGIC; 2334 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 2335 2336 } 2337 /* 2338 * Read the disklabel from the raid device. If one is not present, fake one 2339 * up. 2340 */ 2341 static void 2342 raidgetdisklabel(dev_t dev) 2343 { 2344 int unit = raidunit(dev); 2345 struct raid_softc *rs = &raid_softc[unit]; 2346 const char *errstring; 2347 struct disklabel *lp = rs->sc_dkdev.dk_label; 2348 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 2349 RF_Raid_t *raidPtr; 2350 2351 db1_printf(("Getting the disklabel...\n")); 2352 2353 memset(clp, 0, sizeof(*clp)); 2354 2355 raidPtr = raidPtrs[unit]; 2356 2357 raidgetdefaultlabel(raidPtr, rs, lp); 2358 2359 /* 2360 * Call the generic disklabel extraction routine. 2361 */ 2362 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 2363 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2364 if (errstring) 2365 raidmakedisklabel(rs); 2366 else { 2367 int i; 2368 struct partition *pp; 2369 2370 /* 2371 * Sanity check whether the found disklabel is valid. 2372 * 2373 * This is necessary since total size of the raid device 2374 * may vary when an interleave is changed even though exactly 2375 * same components are used, and old disklabel may used 2376 * if that is found. 2377 */ 2378 if (lp->d_secperunit != rs->sc_size) 2379 printf("raid%d: WARNING: %s: " 2380 "total sector size in disklabel (%" PRIu32 ") != " 2381 "the size of raid (%" PRIu64 ")\n", unit, rs->sc_xname, 2382 lp->d_secperunit, rs->sc_size); 2383 for (i = 0; i < lp->d_npartitions; i++) { 2384 pp = &lp->d_partitions[i]; 2385 if (pp->p_offset + pp->p_size > rs->sc_size) 2386 printf("raid%d: WARNING: %s: end of partition `%c' " 2387 "exceeds the size of raid (%" PRIu64 ")\n", 2388 unit, rs->sc_xname, 'a' + i, rs->sc_size); 2389 } 2390 } 2391 2392 } 2393 /* 2394 * Take care of things one might want to take care of in the event 2395 * that a disklabel isn't present. 2396 */ 2397 static void 2398 raidmakedisklabel(struct raid_softc *rs) 2399 { 2400 struct disklabel *lp = rs->sc_dkdev.dk_label; 2401 db1_printf(("Making a label..\n")); 2402 2403 /* 2404 * For historical reasons, if there's no disklabel present 2405 * the raw partition must be marked FS_BSDFFS. 2406 */ 2407 2408 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2409 2410 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2411 2412 lp->d_checksum = dkcksum(lp); 2413 } 2414 /* 2415 * Wait interruptibly for an exclusive lock. 2416 * 2417 * XXX 2418 * Several drivers do this; it should be abstracted and made MP-safe. 2419 * (Hmm... where have we seen this warning before :-> GO ) 2420 */ 2421 static int 2422 raidlock(struct raid_softc *rs) 2423 { 2424 int error; 2425 2426 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2427 rs->sc_flags |= RAIDF_WANTED; 2428 if ((error = 2429 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2430 return (error); 2431 } 2432 rs->sc_flags |= RAIDF_LOCKED; 2433 return (0); 2434 } 2435 /* 2436 * Unlock and wake up any waiters. 2437 */ 2438 static void 2439 raidunlock(struct raid_softc *rs) 2440 { 2441 2442 rs->sc_flags &= ~RAIDF_LOCKED; 2443 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2444 rs->sc_flags &= ~RAIDF_WANTED; 2445 wakeup(rs); 2446 } 2447 } 2448 2449 2450 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2451 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2452 #define RF_PARITY_MAP_SIZE RF_PARITYMAP_NBYTE 2453 2454 static daddr_t 2455 rf_component_info_offset(void) 2456 { 2457 2458 return RF_COMPONENT_INFO_OFFSET; 2459 } 2460 2461 static daddr_t 2462 rf_component_info_size(unsigned secsize) 2463 { 2464 daddr_t info_size; 2465 2466 KASSERT(secsize); 2467 if (secsize > RF_COMPONENT_INFO_SIZE) 2468 info_size = secsize; 2469 else 2470 info_size = RF_COMPONENT_INFO_SIZE; 2471 2472 return info_size; 2473 } 2474 2475 static daddr_t 2476 rf_parity_map_offset(RF_Raid_t *raidPtr) 2477 { 2478 daddr_t map_offset; 2479 2480 KASSERT(raidPtr->bytesPerSector); 2481 if (raidPtr->bytesPerSector > RF_COMPONENT_INFO_SIZE) 2482 map_offset = raidPtr->bytesPerSector; 2483 else 2484 map_offset = RF_COMPONENT_INFO_SIZE; 2485 map_offset += rf_component_info_offset(); 2486 2487 return map_offset; 2488 } 2489 2490 static daddr_t 2491 rf_parity_map_size(RF_Raid_t *raidPtr) 2492 { 2493 daddr_t map_size; 2494 2495 if (raidPtr->bytesPerSector > RF_PARITY_MAP_SIZE) 2496 map_size = raidPtr->bytesPerSector; 2497 else 2498 map_size = RF_PARITY_MAP_SIZE; 2499 2500 return map_size; 2501 } 2502 2503 int 2504 raidmarkclean(RF_Raid_t *raidPtr, RF_RowCol_t col) 2505 { 2506 RF_ComponentLabel_t *clabel; 2507 2508 clabel = raidget_component_label(raidPtr, col); 2509 clabel->clean = RF_RAID_CLEAN; 2510 raidflush_component_label(raidPtr, col); 2511 return(0); 2512 } 2513 2514 2515 int 2516 raidmarkdirty(RF_Raid_t *raidPtr, RF_RowCol_t col) 2517 { 2518 RF_ComponentLabel_t *clabel; 2519 2520 clabel = raidget_component_label(raidPtr, col); 2521 clabel->clean = RF_RAID_DIRTY; 2522 raidflush_component_label(raidPtr, col); 2523 return(0); 2524 } 2525 2526 int 2527 raidfetch_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2528 { 2529 KASSERT(raidPtr->bytesPerSector); 2530 return raidread_component_label(raidPtr->bytesPerSector, 2531 raidPtr->Disks[col].dev, 2532 raidPtr->raid_cinfo[col].ci_vp, 2533 &raidPtr->raid_cinfo[col].ci_label); 2534 } 2535 2536 RF_ComponentLabel_t * 2537 raidget_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2538 { 2539 return &raidPtr->raid_cinfo[col].ci_label; 2540 } 2541 2542 int 2543 raidflush_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2544 { 2545 RF_ComponentLabel_t *label; 2546 2547 label = &raidPtr->raid_cinfo[col].ci_label; 2548 label->mod_counter = raidPtr->mod_counter; 2549 #ifndef RF_NO_PARITY_MAP 2550 label->parity_map_modcount = label->mod_counter; 2551 #endif 2552 return raidwrite_component_label(raidPtr->bytesPerSector, 2553 raidPtr->Disks[col].dev, 2554 raidPtr->raid_cinfo[col].ci_vp, label); 2555 } 2556 2557 2558 static int 2559 raidread_component_label(unsigned secsize, dev_t dev, struct vnode *b_vp, 2560 RF_ComponentLabel_t *clabel) 2561 { 2562 return raidread_component_area(dev, b_vp, clabel, 2563 sizeof(RF_ComponentLabel_t), 2564 rf_component_info_offset(), 2565 rf_component_info_size(secsize)); 2566 } 2567 2568 /* ARGSUSED */ 2569 static int 2570 raidread_component_area(dev_t dev, struct vnode *b_vp, void *data, 2571 size_t msize, daddr_t offset, daddr_t dsize) 2572 { 2573 struct buf *bp; 2574 const struct bdevsw *bdev; 2575 int error; 2576 2577 /* XXX should probably ensure that we don't try to do this if 2578 someone has changed rf_protected_sectors. */ 2579 2580 if (b_vp == NULL) { 2581 /* For whatever reason, this component is not valid. 2582 Don't try to read a component label from it. */ 2583 return(EINVAL); 2584 } 2585 2586 /* get a block of the appropriate size... */ 2587 bp = geteblk((int)dsize); 2588 bp->b_dev = dev; 2589 2590 /* get our ducks in a row for the read */ 2591 bp->b_blkno = offset / DEV_BSIZE; 2592 bp->b_bcount = dsize; 2593 bp->b_flags |= B_READ; 2594 bp->b_resid = dsize; 2595 2596 bdev = bdevsw_lookup(bp->b_dev); 2597 if (bdev == NULL) 2598 return (ENXIO); 2599 (*bdev->d_strategy)(bp); 2600 2601 error = biowait(bp); 2602 2603 if (!error) { 2604 memcpy(data, bp->b_data, msize); 2605 } 2606 2607 brelse(bp, 0); 2608 return(error); 2609 } 2610 2611 2612 static int 2613 raidwrite_component_label(unsigned secsize, dev_t dev, struct vnode *b_vp, 2614 RF_ComponentLabel_t *clabel) 2615 { 2616 return raidwrite_component_area(dev, b_vp, clabel, 2617 sizeof(RF_ComponentLabel_t), 2618 rf_component_info_offset(), 2619 rf_component_info_size(secsize), 0); 2620 } 2621 2622 /* ARGSUSED */ 2623 static int 2624 raidwrite_component_area(dev_t dev, struct vnode *b_vp, void *data, 2625 size_t msize, daddr_t offset, daddr_t dsize, int asyncp) 2626 { 2627 struct buf *bp; 2628 const struct bdevsw *bdev; 2629 int error; 2630 2631 /* get a block of the appropriate size... */ 2632 bp = geteblk((int)dsize); 2633 bp->b_dev = dev; 2634 2635 /* get our ducks in a row for the write */ 2636 bp->b_blkno = offset / DEV_BSIZE; 2637 bp->b_bcount = dsize; 2638 bp->b_flags |= B_WRITE | (asyncp ? B_ASYNC : 0); 2639 bp->b_resid = dsize; 2640 2641 memset(bp->b_data, 0, dsize); 2642 memcpy(bp->b_data, data, msize); 2643 2644 bdev = bdevsw_lookup(bp->b_dev); 2645 if (bdev == NULL) 2646 return (ENXIO); 2647 (*bdev->d_strategy)(bp); 2648 if (asyncp) 2649 return 0; 2650 error = biowait(bp); 2651 brelse(bp, 0); 2652 if (error) { 2653 #if 1 2654 printf("Failed to write RAID component info!\n"); 2655 #endif 2656 } 2657 2658 return(error); 2659 } 2660 2661 void 2662 rf_paritymap_kern_write(RF_Raid_t *raidPtr, struct rf_paritymap_ondisk *map) 2663 { 2664 int c; 2665 2666 for (c = 0; c < raidPtr->numCol; c++) { 2667 /* Skip dead disks. */ 2668 if (RF_DEAD_DISK(raidPtr->Disks[c].status)) 2669 continue; 2670 /* XXXjld: what if an error occurs here? */ 2671 raidwrite_component_area(raidPtr->Disks[c].dev, 2672 raidPtr->raid_cinfo[c].ci_vp, map, 2673 RF_PARITYMAP_NBYTE, 2674 rf_parity_map_offset(raidPtr), 2675 rf_parity_map_size(raidPtr), 0); 2676 } 2677 } 2678 2679 void 2680 rf_paritymap_kern_read(RF_Raid_t *raidPtr, struct rf_paritymap_ondisk *map) 2681 { 2682 struct rf_paritymap_ondisk tmp; 2683 int c,first; 2684 2685 first=1; 2686 for (c = 0; c < raidPtr->numCol; c++) { 2687 /* Skip dead disks. */ 2688 if (RF_DEAD_DISK(raidPtr->Disks[c].status)) 2689 continue; 2690 raidread_component_area(raidPtr->Disks[c].dev, 2691 raidPtr->raid_cinfo[c].ci_vp, &tmp, 2692 RF_PARITYMAP_NBYTE, 2693 rf_parity_map_offset(raidPtr), 2694 rf_parity_map_size(raidPtr)); 2695 if (first) { 2696 memcpy(map, &tmp, sizeof(*map)); 2697 first = 0; 2698 } else { 2699 rf_paritymap_merge(map, &tmp); 2700 } 2701 } 2702 } 2703 2704 void 2705 rf_markalldirty(RF_Raid_t *raidPtr) 2706 { 2707 RF_ComponentLabel_t *clabel; 2708 int sparecol; 2709 int c; 2710 int j; 2711 int scol = -1; 2712 2713 raidPtr->mod_counter++; 2714 for (c = 0; c < raidPtr->numCol; c++) { 2715 /* we don't want to touch (at all) a disk that has 2716 failed */ 2717 if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) { 2718 clabel = raidget_component_label(raidPtr, c); 2719 if (clabel->status == rf_ds_spared) { 2720 /* XXX do something special... 2721 but whatever you do, don't 2722 try to access it!! */ 2723 } else { 2724 raidmarkdirty(raidPtr, c); 2725 } 2726 } 2727 } 2728 2729 for( c = 0; c < raidPtr->numSpare ; c++) { 2730 sparecol = raidPtr->numCol + c; 2731 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2732 /* 2733 2734 we claim this disk is "optimal" if it's 2735 rf_ds_used_spare, as that means it should be 2736 directly substitutable for the disk it replaced. 2737 We note that too... 2738 2739 */ 2740 2741 for(j=0;j<raidPtr->numCol;j++) { 2742 if (raidPtr->Disks[j].spareCol == sparecol) { 2743 scol = j; 2744 break; 2745 } 2746 } 2747 2748 clabel = raidget_component_label(raidPtr, sparecol); 2749 /* make sure status is noted */ 2750 2751 raid_init_component_label(raidPtr, clabel); 2752 2753 clabel->row = 0; 2754 clabel->column = scol; 2755 /* Note: we *don't* change status from rf_ds_used_spare 2756 to rf_ds_optimal */ 2757 /* clabel.status = rf_ds_optimal; */ 2758 2759 raidmarkdirty(raidPtr, sparecol); 2760 } 2761 } 2762 } 2763 2764 2765 void 2766 rf_update_component_labels(RF_Raid_t *raidPtr, int final) 2767 { 2768 RF_ComponentLabel_t *clabel; 2769 int sparecol; 2770 int c; 2771 int j; 2772 int scol; 2773 2774 scol = -1; 2775 2776 /* XXX should do extra checks to make sure things really are clean, 2777 rather than blindly setting the clean bit... */ 2778 2779 raidPtr->mod_counter++; 2780 2781 for (c = 0; c < raidPtr->numCol; c++) { 2782 if (raidPtr->Disks[c].status == rf_ds_optimal) { 2783 clabel = raidget_component_label(raidPtr, c); 2784 /* make sure status is noted */ 2785 clabel->status = rf_ds_optimal; 2786 2787 /* note what unit we are configured as */ 2788 clabel->last_unit = raidPtr->raidid; 2789 2790 raidflush_component_label(raidPtr, c); 2791 if (final == RF_FINAL_COMPONENT_UPDATE) { 2792 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2793 raidmarkclean(raidPtr, c); 2794 } 2795 } 2796 } 2797 /* else we don't touch it.. */ 2798 } 2799 2800 for( c = 0; c < raidPtr->numSpare ; c++) { 2801 sparecol = raidPtr->numCol + c; 2802 /* Need to ensure that the reconstruct actually completed! */ 2803 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2804 /* 2805 2806 we claim this disk is "optimal" if it's 2807 rf_ds_used_spare, as that means it should be 2808 directly substitutable for the disk it replaced. 2809 We note that too... 2810 2811 */ 2812 2813 for(j=0;j<raidPtr->numCol;j++) { 2814 if (raidPtr->Disks[j].spareCol == sparecol) { 2815 scol = j; 2816 break; 2817 } 2818 } 2819 2820 /* XXX shouldn't *really* need this... */ 2821 clabel = raidget_component_label(raidPtr, sparecol); 2822 /* make sure status is noted */ 2823 2824 raid_init_component_label(raidPtr, clabel); 2825 2826 clabel->column = scol; 2827 clabel->status = rf_ds_optimal; 2828 clabel->last_unit = raidPtr->raidid; 2829 2830 raidflush_component_label(raidPtr, sparecol); 2831 if (final == RF_FINAL_COMPONENT_UPDATE) { 2832 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2833 raidmarkclean(raidPtr, sparecol); 2834 } 2835 } 2836 } 2837 } 2838 } 2839 2840 void 2841 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured) 2842 { 2843 2844 if (vp != NULL) { 2845 if (auto_configured == 1) { 2846 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2847 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2848 vput(vp); 2849 2850 } else { 2851 (void) vn_close(vp, FREAD | FWRITE, curlwp->l_cred); 2852 } 2853 } 2854 } 2855 2856 2857 void 2858 rf_UnconfigureVnodes(RF_Raid_t *raidPtr) 2859 { 2860 int r,c; 2861 struct vnode *vp; 2862 int acd; 2863 2864 2865 /* We take this opportunity to close the vnodes like we should.. */ 2866 2867 for (c = 0; c < raidPtr->numCol; c++) { 2868 vp = raidPtr->raid_cinfo[c].ci_vp; 2869 acd = raidPtr->Disks[c].auto_configured; 2870 rf_close_component(raidPtr, vp, acd); 2871 raidPtr->raid_cinfo[c].ci_vp = NULL; 2872 raidPtr->Disks[c].auto_configured = 0; 2873 } 2874 2875 for (r = 0; r < raidPtr->numSpare; r++) { 2876 vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp; 2877 acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured; 2878 rf_close_component(raidPtr, vp, acd); 2879 raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL; 2880 raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0; 2881 } 2882 } 2883 2884 2885 void 2886 rf_ReconThread(struct rf_recon_req *req) 2887 { 2888 int s; 2889 RF_Raid_t *raidPtr; 2890 2891 s = splbio(); 2892 raidPtr = (RF_Raid_t *) req->raidPtr; 2893 raidPtr->recon_in_progress = 1; 2894 2895 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col, 2896 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2897 2898 RF_Free(req, sizeof(*req)); 2899 2900 raidPtr->recon_in_progress = 0; 2901 splx(s); 2902 2903 /* That's all... */ 2904 kthread_exit(0); /* does not return */ 2905 } 2906 2907 void 2908 rf_RewriteParityThread(RF_Raid_t *raidPtr) 2909 { 2910 int retcode; 2911 int s; 2912 2913 raidPtr->parity_rewrite_stripes_done = 0; 2914 raidPtr->parity_rewrite_in_progress = 1; 2915 s = splbio(); 2916 retcode = rf_RewriteParity(raidPtr); 2917 splx(s); 2918 if (retcode) { 2919 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2920 } else { 2921 /* set the clean bit! If we shutdown correctly, 2922 the clean bit on each component label will get 2923 set */ 2924 raidPtr->parity_good = RF_RAID_CLEAN; 2925 } 2926 raidPtr->parity_rewrite_in_progress = 0; 2927 2928 /* Anyone waiting for us to stop? If so, inform them... */ 2929 if (raidPtr->waitShutdown) { 2930 wakeup(&raidPtr->parity_rewrite_in_progress); 2931 } 2932 2933 /* That's all... */ 2934 kthread_exit(0); /* does not return */ 2935 } 2936 2937 2938 void 2939 rf_CopybackThread(RF_Raid_t *raidPtr) 2940 { 2941 int s; 2942 2943 raidPtr->copyback_in_progress = 1; 2944 s = splbio(); 2945 rf_CopybackReconstructedData(raidPtr); 2946 splx(s); 2947 raidPtr->copyback_in_progress = 0; 2948 2949 /* That's all... */ 2950 kthread_exit(0); /* does not return */ 2951 } 2952 2953 2954 void 2955 rf_ReconstructInPlaceThread(struct rf_recon_req *req) 2956 { 2957 int s; 2958 RF_Raid_t *raidPtr; 2959 2960 s = splbio(); 2961 raidPtr = req->raidPtr; 2962 raidPtr->recon_in_progress = 1; 2963 rf_ReconstructInPlace(raidPtr, req->col); 2964 RF_Free(req, sizeof(*req)); 2965 raidPtr->recon_in_progress = 0; 2966 splx(s); 2967 2968 /* That's all... */ 2969 kthread_exit(0); /* does not return */ 2970 } 2971 2972 static RF_AutoConfig_t * 2973 rf_get_component(RF_AutoConfig_t *ac_list, dev_t dev, struct vnode *vp, 2974 const char *cname, RF_SectorCount_t size, uint64_t numsecs, 2975 unsigned secsize) 2976 { 2977 int good_one = 0; 2978 RF_ComponentLabel_t *clabel; 2979 RF_AutoConfig_t *ac; 2980 2981 clabel = malloc(sizeof(RF_ComponentLabel_t), M_RAIDFRAME, M_NOWAIT); 2982 if (clabel == NULL) { 2983 oomem: 2984 while(ac_list) { 2985 ac = ac_list; 2986 if (ac->clabel) 2987 free(ac->clabel, M_RAIDFRAME); 2988 ac_list = ac_list->next; 2989 free(ac, M_RAIDFRAME); 2990 } 2991 printf("RAID auto config: out of memory!\n"); 2992 return NULL; /* XXX probably should panic? */ 2993 } 2994 2995 if (!raidread_component_label(secsize, dev, vp, clabel)) { 2996 /* Got the label. Does it look reasonable? */ 2997 if (rf_reasonable_label(clabel, numsecs) && 2998 (rf_component_label_partitionsize(clabel) <= size)) { 2999 #ifdef DEBUG 3000 printf("Component on: %s: %llu\n", 3001 cname, (unsigned long long)size); 3002 rf_print_component_label(clabel); 3003 #endif 3004 /* if it's reasonable, add it, else ignore it. */ 3005 ac = malloc(sizeof(RF_AutoConfig_t), M_RAIDFRAME, 3006 M_NOWAIT); 3007 if (ac == NULL) { 3008 free(clabel, M_RAIDFRAME); 3009 goto oomem; 3010 } 3011 strlcpy(ac->devname, cname, sizeof(ac->devname)); 3012 ac->dev = dev; 3013 ac->vp = vp; 3014 ac->clabel = clabel; 3015 ac->next = ac_list; 3016 ac_list = ac; 3017 good_one = 1; 3018 } 3019 } 3020 if (!good_one) { 3021 /* cleanup */ 3022 free(clabel, M_RAIDFRAME); 3023 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3024 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3025 vput(vp); 3026 } 3027 return ac_list; 3028 } 3029 3030 RF_AutoConfig_t * 3031 rf_find_raid_components() 3032 { 3033 struct vnode *vp; 3034 struct disklabel label; 3035 struct device *dv; 3036 dev_t dev; 3037 int bmajor, bminor, wedge; 3038 int error; 3039 int i; 3040 RF_AutoConfig_t *ac_list; 3041 uint64_t numsecs; 3042 unsigned secsize; 3043 3044 RF_ASSERT(raidPtr->bytesPerSector < rf_component_info_offset()); 3045 3046 /* initialize the AutoConfig list */ 3047 ac_list = NULL; 3048 3049 /* we begin by trolling through *all* the devices on the system */ 3050 3051 for (dv = alldevs.tqh_first; dv != NULL; 3052 dv = dv->dv_list.tqe_next) { 3053 3054 /* we are only interested in disks... */ 3055 if (device_class(dv) != DV_DISK) 3056 continue; 3057 3058 /* we don't care about floppies... */ 3059 if (device_is_a(dv, "fd")) { 3060 continue; 3061 } 3062 3063 /* we don't care about CD's... */ 3064 if (device_is_a(dv, "cd")) { 3065 continue; 3066 } 3067 3068 /* we don't care about md's... */ 3069 if (device_is_a(dv, "md")) { 3070 continue; 3071 } 3072 3073 /* hdfd is the Atari/Hades floppy driver */ 3074 if (device_is_a(dv, "hdfd")) { 3075 continue; 3076 } 3077 3078 /* fdisa is the Atari/Milan floppy driver */ 3079 if (device_is_a(dv, "fdisa")) { 3080 continue; 3081 } 3082 3083 /* need to find the device_name_to_block_device_major stuff */ 3084 bmajor = devsw_name2blk(device_xname(dv), NULL, 0); 3085 3086 /* get a vnode for the raw partition of this disk */ 3087 3088 wedge = device_is_a(dv, "dk"); 3089 bminor = minor(device_unit(dv)); 3090 dev = wedge ? makedev(bmajor, bminor) : 3091 MAKEDISKDEV(bmajor, bminor, RAW_PART); 3092 if (bdevvp(dev, &vp)) 3093 panic("RAID can't alloc vnode"); 3094 3095 error = VOP_OPEN(vp, FREAD, NOCRED); 3096 3097 if (error) { 3098 /* "Who cares." Continue looking 3099 for something that exists*/ 3100 vput(vp); 3101 continue; 3102 } 3103 3104 error = getdisksize(vp, &numsecs, &secsize); 3105 if (error) { 3106 vput(vp); 3107 continue; 3108 } 3109 if (wedge) { 3110 struct dkwedge_info dkw; 3111 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, 3112 NOCRED); 3113 if (error) { 3114 printf("RAIDframe: can't get wedge info for " 3115 "dev %s (%d)\n", device_xname(dv), error); 3116 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3117 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3118 vput(vp); 3119 continue; 3120 } 3121 3122 if (strcmp(dkw.dkw_ptype, DKW_PTYPE_RAIDFRAME) != 0) { 3123 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3124 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3125 vput(vp); 3126 continue; 3127 } 3128 3129 ac_list = rf_get_component(ac_list, dev, vp, 3130 device_xname(dv), dkw.dkw_size, numsecs, secsize); 3131 continue; 3132 } 3133 3134 /* Ok, the disk exists. Go get the disklabel. */ 3135 error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED); 3136 if (error) { 3137 /* 3138 * XXX can't happen - open() would 3139 * have errored out (or faked up one) 3140 */ 3141 if (error != ENOTTY) 3142 printf("RAIDframe: can't get label for dev " 3143 "%s (%d)\n", device_xname(dv), error); 3144 } 3145 3146 /* don't need this any more. We'll allocate it again 3147 a little later if we really do... */ 3148 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3149 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3150 vput(vp); 3151 3152 if (error) 3153 continue; 3154 3155 for (i = 0; i < label.d_npartitions; i++) { 3156 char cname[sizeof(ac_list->devname)]; 3157 3158 /* We only support partitions marked as RAID */ 3159 if (label.d_partitions[i].p_fstype != FS_RAID) 3160 continue; 3161 3162 dev = MAKEDISKDEV(bmajor, device_unit(dv), i); 3163 if (bdevvp(dev, &vp)) 3164 panic("RAID can't alloc vnode"); 3165 3166 error = VOP_OPEN(vp, FREAD, NOCRED); 3167 if (error) { 3168 /* Whatever... */ 3169 vput(vp); 3170 continue; 3171 } 3172 snprintf(cname, sizeof(cname), "%s%c", 3173 device_xname(dv), 'a' + i); 3174 ac_list = rf_get_component(ac_list, dev, vp, cname, 3175 label.d_partitions[i].p_size, numsecs, secsize); 3176 } 3177 } 3178 return ac_list; 3179 } 3180 3181 3182 static int 3183 rf_reasonable_label(RF_ComponentLabel_t *clabel, uint64_t numsecs) 3184 { 3185 3186 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 3187 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 3188 ((clabel->clean == RF_RAID_CLEAN) || 3189 (clabel->clean == RF_RAID_DIRTY)) && 3190 clabel->row >=0 && 3191 clabel->column >= 0 && 3192 clabel->num_rows > 0 && 3193 clabel->num_columns > 0 && 3194 clabel->row < clabel->num_rows && 3195 clabel->column < clabel->num_columns && 3196 clabel->blockSize > 0 && 3197 /* 3198 * numBlocksHi may contain garbage, but it is ok since 3199 * the type is unsigned. If it is really garbage, 3200 * rf_fix_old_label_size() will fix it. 3201 */ 3202 rf_component_label_numblocks(clabel) > 0) { 3203 /* 3204 * label looks reasonable enough... 3205 * let's make sure it has no old garbage. 3206 */ 3207 rf_fix_old_label_size(clabel, numsecs); 3208 return(1); 3209 } 3210 return(0); 3211 } 3212 3213 3214 /* 3215 * For reasons yet unknown, some old component labels have garbage in 3216 * the newer numBlocksHi region, and this causes lossage. Since those 3217 * disks will also have numsecs set to less than 32 bits of sectors, 3218 * we can determine when this corruption has occured, and fix it. 3219 * 3220 * The exact same problem, with the same unknown reason, happens to 3221 * the partitionSizeHi member as well. 3222 */ 3223 static void 3224 rf_fix_old_label_size(RF_ComponentLabel_t *clabel, uint64_t numsecs) 3225 { 3226 3227 if (numsecs < ((uint64_t)1 << 32)) { 3228 if (clabel->numBlocksHi) { 3229 printf("WARNING: total sectors < 32 bits, yet " 3230 "numBlocksHi set\n" 3231 "WARNING: resetting numBlocksHi to zero.\n"); 3232 clabel->numBlocksHi = 0; 3233 } 3234 3235 if (clabel->partitionSizeHi) { 3236 printf("WARNING: total sectors < 32 bits, yet " 3237 "partitionSizeHi set\n" 3238 "WARNING: resetting partitionSizeHi to zero.\n"); 3239 clabel->partitionSizeHi = 0; 3240 } 3241 } 3242 } 3243 3244 3245 #ifdef DEBUG 3246 void 3247 rf_print_component_label(RF_ComponentLabel_t *clabel) 3248 { 3249 uint64_t numBlocks; 3250 3251 numBlocks = rf_component_label_numblocks(clabel); 3252 3253 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 3254 clabel->row, clabel->column, 3255 clabel->num_rows, clabel->num_columns); 3256 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 3257 clabel->version, clabel->serial_number, 3258 clabel->mod_counter); 3259 printf(" Clean: %s Status: %d\n", 3260 clabel->clean ? "Yes" : "No", clabel->status ); 3261 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 3262 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 3263 printf(" RAID Level: %c blocksize: %d numBlocks: %"PRIu64"\n", 3264 (char) clabel->parityConfig, clabel->blockSize, numBlocks); 3265 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" ); 3266 printf(" Contains root partition: %s\n", 3267 clabel->root_partition ? "Yes" : "No" ); 3268 printf(" Last configured as: raid%d\n", clabel->last_unit ); 3269 #if 0 3270 printf(" Config order: %d\n", clabel->config_order); 3271 #endif 3272 3273 } 3274 #endif 3275 3276 RF_ConfigSet_t * 3277 rf_create_auto_sets(RF_AutoConfig_t *ac_list) 3278 { 3279 RF_AutoConfig_t *ac; 3280 RF_ConfigSet_t *config_sets; 3281 RF_ConfigSet_t *cset; 3282 RF_AutoConfig_t *ac_next; 3283 3284 3285 config_sets = NULL; 3286 3287 /* Go through the AutoConfig list, and figure out which components 3288 belong to what sets. */ 3289 ac = ac_list; 3290 while(ac!=NULL) { 3291 /* we're going to putz with ac->next, so save it here 3292 for use at the end of the loop */ 3293 ac_next = ac->next; 3294 3295 if (config_sets == NULL) { 3296 /* will need at least this one... */ 3297 config_sets = (RF_ConfigSet_t *) 3298 malloc(sizeof(RF_ConfigSet_t), 3299 M_RAIDFRAME, M_NOWAIT); 3300 if (config_sets == NULL) { 3301 panic("rf_create_auto_sets: No memory!"); 3302 } 3303 /* this one is easy :) */ 3304 config_sets->ac = ac; 3305 config_sets->next = NULL; 3306 config_sets->rootable = 0; 3307 ac->next = NULL; 3308 } else { 3309 /* which set does this component fit into? */ 3310 cset = config_sets; 3311 while(cset!=NULL) { 3312 if (rf_does_it_fit(cset, ac)) { 3313 /* looks like it matches... */ 3314 ac->next = cset->ac; 3315 cset->ac = ac; 3316 break; 3317 } 3318 cset = cset->next; 3319 } 3320 if (cset==NULL) { 3321 /* didn't find a match above... new set..*/ 3322 cset = (RF_ConfigSet_t *) 3323 malloc(sizeof(RF_ConfigSet_t), 3324 M_RAIDFRAME, M_NOWAIT); 3325 if (cset == NULL) { 3326 panic("rf_create_auto_sets: No memory!"); 3327 } 3328 cset->ac = ac; 3329 ac->next = NULL; 3330 cset->next = config_sets; 3331 cset->rootable = 0; 3332 config_sets = cset; 3333 } 3334 } 3335 ac = ac_next; 3336 } 3337 3338 3339 return(config_sets); 3340 } 3341 3342 static int 3343 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac) 3344 { 3345 RF_ComponentLabel_t *clabel1, *clabel2; 3346 3347 /* If this one matches the *first* one in the set, that's good 3348 enough, since the other members of the set would have been 3349 through here too... */ 3350 /* note that we are not checking partitionSize here.. 3351 3352 Note that we are also not checking the mod_counters here. 3353 If everything else matches execpt the mod_counter, that's 3354 good enough for this test. We will deal with the mod_counters 3355 a little later in the autoconfiguration process. 3356 3357 (clabel1->mod_counter == clabel2->mod_counter) && 3358 3359 The reason we don't check for this is that failed disks 3360 will have lower modification counts. If those disks are 3361 not added to the set they used to belong to, then they will 3362 form their own set, which may result in 2 different sets, 3363 for example, competing to be configured at raid0, and 3364 perhaps competing to be the root filesystem set. If the 3365 wrong ones get configured, or both attempt to become /, 3366 weird behaviour and or serious lossage will occur. Thus we 3367 need to bring them into the fold here, and kick them out at 3368 a later point. 3369 3370 */ 3371 3372 clabel1 = cset->ac->clabel; 3373 clabel2 = ac->clabel; 3374 if ((clabel1->version == clabel2->version) && 3375 (clabel1->serial_number == clabel2->serial_number) && 3376 (clabel1->num_rows == clabel2->num_rows) && 3377 (clabel1->num_columns == clabel2->num_columns) && 3378 (clabel1->sectPerSU == clabel2->sectPerSU) && 3379 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 3380 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 3381 (clabel1->parityConfig == clabel2->parityConfig) && 3382 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 3383 (clabel1->blockSize == clabel2->blockSize) && 3384 rf_component_label_numblocks(clabel1) == 3385 rf_component_label_numblocks(clabel2) && 3386 (clabel1->autoconfigure == clabel2->autoconfigure) && 3387 (clabel1->root_partition == clabel2->root_partition) && 3388 (clabel1->last_unit == clabel2->last_unit) && 3389 (clabel1->config_order == clabel2->config_order)) { 3390 /* if it get's here, it almost *has* to be a match */ 3391 } else { 3392 /* it's not consistent with somebody in the set.. 3393 punt */ 3394 return(0); 3395 } 3396 /* all was fine.. it must fit... */ 3397 return(1); 3398 } 3399 3400 int 3401 rf_have_enough_components(RF_ConfigSet_t *cset) 3402 { 3403 RF_AutoConfig_t *ac; 3404 RF_AutoConfig_t *auto_config; 3405 RF_ComponentLabel_t *clabel; 3406 int c; 3407 int num_cols; 3408 int num_missing; 3409 int mod_counter; 3410 int mod_counter_found; 3411 int even_pair_failed; 3412 char parity_type; 3413 3414 3415 /* check to see that we have enough 'live' components 3416 of this set. If so, we can configure it if necessary */ 3417 3418 num_cols = cset->ac->clabel->num_columns; 3419 parity_type = cset->ac->clabel->parityConfig; 3420 3421 /* XXX Check for duplicate components!?!?!? */ 3422 3423 /* Determine what the mod_counter is supposed to be for this set. */ 3424 3425 mod_counter_found = 0; 3426 mod_counter = 0; 3427 ac = cset->ac; 3428 while(ac!=NULL) { 3429 if (mod_counter_found==0) { 3430 mod_counter = ac->clabel->mod_counter; 3431 mod_counter_found = 1; 3432 } else { 3433 if (ac->clabel->mod_counter > mod_counter) { 3434 mod_counter = ac->clabel->mod_counter; 3435 } 3436 } 3437 ac = ac->next; 3438 } 3439 3440 num_missing = 0; 3441 auto_config = cset->ac; 3442 3443 even_pair_failed = 0; 3444 for(c=0; c<num_cols; c++) { 3445 ac = auto_config; 3446 while(ac!=NULL) { 3447 if ((ac->clabel->column == c) && 3448 (ac->clabel->mod_counter == mod_counter)) { 3449 /* it's this one... */ 3450 #ifdef DEBUG 3451 printf("Found: %s at %d\n", 3452 ac->devname,c); 3453 #endif 3454 break; 3455 } 3456 ac=ac->next; 3457 } 3458 if (ac==NULL) { 3459 /* Didn't find one here! */ 3460 /* special case for RAID 1, especially 3461 where there are more than 2 3462 components (where RAIDframe treats 3463 things a little differently :( ) */ 3464 if (parity_type == '1') { 3465 if (c%2 == 0) { /* even component */ 3466 even_pair_failed = 1; 3467 } else { /* odd component. If 3468 we're failed, and 3469 so is the even 3470 component, it's 3471 "Good Night, Charlie" */ 3472 if (even_pair_failed == 1) { 3473 return(0); 3474 } 3475 } 3476 } else { 3477 /* normal accounting */ 3478 num_missing++; 3479 } 3480 } 3481 if ((parity_type == '1') && (c%2 == 1)) { 3482 /* Just did an even component, and we didn't 3483 bail.. reset the even_pair_failed flag, 3484 and go on to the next component.... */ 3485 even_pair_failed = 0; 3486 } 3487 } 3488 3489 clabel = cset->ac->clabel; 3490 3491 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3492 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3493 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3494 /* XXX this needs to be made *much* more general */ 3495 /* Too many failures */ 3496 return(0); 3497 } 3498 /* otherwise, all is well, and we've got enough to take a kick 3499 at autoconfiguring this set */ 3500 return(1); 3501 } 3502 3503 void 3504 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config, 3505 RF_Raid_t *raidPtr) 3506 { 3507 RF_ComponentLabel_t *clabel; 3508 int i; 3509 3510 clabel = ac->clabel; 3511 3512 /* 1. Fill in the common stuff */ 3513 config->numRow = clabel->num_rows = 1; 3514 config->numCol = clabel->num_columns; 3515 config->numSpare = 0; /* XXX should this be set here? */ 3516 config->sectPerSU = clabel->sectPerSU; 3517 config->SUsPerPU = clabel->SUsPerPU; 3518 config->SUsPerRU = clabel->SUsPerRU; 3519 config->parityConfig = clabel->parityConfig; 3520 /* XXX... */ 3521 strcpy(config->diskQueueType,"fifo"); 3522 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3523 config->layoutSpecificSize = 0; /* XXX ?? */ 3524 3525 while(ac!=NULL) { 3526 /* row/col values will be in range due to the checks 3527 in reasonable_label() */ 3528 strcpy(config->devnames[0][ac->clabel->column], 3529 ac->devname); 3530 ac = ac->next; 3531 } 3532 3533 for(i=0;i<RF_MAXDBGV;i++) { 3534 config->debugVars[i][0] = 0; 3535 } 3536 } 3537 3538 int 3539 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value) 3540 { 3541 RF_ComponentLabel_t *clabel; 3542 int column; 3543 int sparecol; 3544 3545 raidPtr->autoconfigure = new_value; 3546 3547 for(column=0; column<raidPtr->numCol; column++) { 3548 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3549 clabel = raidget_component_label(raidPtr, column); 3550 clabel->autoconfigure = new_value; 3551 raidflush_component_label(raidPtr, column); 3552 } 3553 } 3554 for(column = 0; column < raidPtr->numSpare ; column++) { 3555 sparecol = raidPtr->numCol + column; 3556 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3557 clabel = raidget_component_label(raidPtr, sparecol); 3558 clabel->autoconfigure = new_value; 3559 raidflush_component_label(raidPtr, sparecol); 3560 } 3561 } 3562 return(new_value); 3563 } 3564 3565 int 3566 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value) 3567 { 3568 RF_ComponentLabel_t *clabel; 3569 int column; 3570 int sparecol; 3571 3572 raidPtr->root_partition = new_value; 3573 for(column=0; column<raidPtr->numCol; column++) { 3574 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3575 clabel = raidget_component_label(raidPtr, column); 3576 clabel->root_partition = new_value; 3577 raidflush_component_label(raidPtr, column); 3578 } 3579 } 3580 for(column = 0; column < raidPtr->numSpare ; column++) { 3581 sparecol = raidPtr->numCol + column; 3582 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3583 clabel = raidget_component_label(raidPtr, sparecol); 3584 clabel->root_partition = new_value; 3585 raidflush_component_label(raidPtr, sparecol); 3586 } 3587 } 3588 return(new_value); 3589 } 3590 3591 void 3592 rf_release_all_vps(RF_ConfigSet_t *cset) 3593 { 3594 RF_AutoConfig_t *ac; 3595 3596 ac = cset->ac; 3597 while(ac!=NULL) { 3598 /* Close the vp, and give it back */ 3599 if (ac->vp) { 3600 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3601 VOP_CLOSE(ac->vp, FREAD, NOCRED); 3602 vput(ac->vp); 3603 ac->vp = NULL; 3604 } 3605 ac = ac->next; 3606 } 3607 } 3608 3609 3610 void 3611 rf_cleanup_config_set(RF_ConfigSet_t *cset) 3612 { 3613 RF_AutoConfig_t *ac; 3614 RF_AutoConfig_t *next_ac; 3615 3616 ac = cset->ac; 3617 while(ac!=NULL) { 3618 next_ac = ac->next; 3619 /* nuke the label */ 3620 free(ac->clabel, M_RAIDFRAME); 3621 /* cleanup the config structure */ 3622 free(ac, M_RAIDFRAME); 3623 /* "next.." */ 3624 ac = next_ac; 3625 } 3626 /* and, finally, nuke the config set */ 3627 free(cset, M_RAIDFRAME); 3628 } 3629 3630 3631 void 3632 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel) 3633 { 3634 /* current version number */ 3635 clabel->version = RF_COMPONENT_LABEL_VERSION; 3636 clabel->serial_number = raidPtr->serial_number; 3637 clabel->mod_counter = raidPtr->mod_counter; 3638 3639 clabel->num_rows = 1; 3640 clabel->num_columns = raidPtr->numCol; 3641 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3642 clabel->status = rf_ds_optimal; /* "It's good!" */ 3643 3644 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3645 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3646 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3647 3648 clabel->blockSize = raidPtr->bytesPerSector; 3649 rf_component_label_set_numblocks(clabel, raidPtr->sectorsPerDisk); 3650 3651 /* XXX not portable */ 3652 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3653 clabel->maxOutstanding = raidPtr->maxOutstanding; 3654 clabel->autoconfigure = raidPtr->autoconfigure; 3655 clabel->root_partition = raidPtr->root_partition; 3656 clabel->last_unit = raidPtr->raidid; 3657 clabel->config_order = raidPtr->config_order; 3658 3659 #ifndef RF_NO_PARITY_MAP 3660 rf_paritymap_init_label(raidPtr->parity_map, clabel); 3661 #endif 3662 } 3663 3664 int 3665 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit) 3666 { 3667 RF_Raid_t *raidPtr; 3668 RF_Config_t *config; 3669 int raidID; 3670 int retcode; 3671 3672 #ifdef DEBUG 3673 printf("RAID autoconfigure\n"); 3674 #endif 3675 3676 retcode = 0; 3677 *unit = -1; 3678 3679 /* 1. Create a config structure */ 3680 3681 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3682 M_RAIDFRAME, 3683 M_NOWAIT); 3684 if (config==NULL) { 3685 printf("Out of mem!?!?\n"); 3686 /* XXX do something more intelligent here. */ 3687 return(1); 3688 } 3689 3690 memset(config, 0, sizeof(RF_Config_t)); 3691 3692 /* 3693 2. Figure out what RAID ID this one is supposed to live at 3694 See if we can get the same RAID dev that it was configured 3695 on last time.. 3696 */ 3697 3698 raidID = cset->ac->clabel->last_unit; 3699 if ((raidID < 0) || (raidID >= numraid)) { 3700 /* let's not wander off into lala land. */ 3701 raidID = numraid - 1; 3702 } 3703 if (raidPtrs[raidID]->valid != 0) { 3704 3705 /* 3706 Nope... Go looking for an alternative... 3707 Start high so we don't immediately use raid0 if that's 3708 not taken. 3709 */ 3710 3711 for(raidID = numraid - 1; raidID >= 0; raidID--) { 3712 if (raidPtrs[raidID]->valid == 0) { 3713 /* can use this one! */ 3714 break; 3715 } 3716 } 3717 } 3718 3719 if (raidID < 0) { 3720 /* punt... */ 3721 printf("Unable to auto configure this set!\n"); 3722 printf("(Out of RAID devs!)\n"); 3723 free(config, M_RAIDFRAME); 3724 return(1); 3725 } 3726 3727 #ifdef DEBUG 3728 printf("Configuring raid%d:\n",raidID); 3729 #endif 3730 3731 raidPtr = raidPtrs[raidID]; 3732 3733 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3734 raidPtr->raidid = raidID; 3735 raidPtr->openings = RAIDOUTSTANDING; 3736 3737 /* 3. Build the configuration structure */ 3738 rf_create_configuration(cset->ac, config, raidPtr); 3739 3740 /* 4. Do the configuration */ 3741 retcode = rf_Configure(raidPtr, config, cset->ac); 3742 3743 if (retcode == 0) { 3744 3745 raidinit(raidPtrs[raidID]); 3746 3747 rf_markalldirty(raidPtrs[raidID]); 3748 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3749 if (cset->ac->clabel->root_partition==1) { 3750 /* everything configured just fine. Make a note 3751 that this set is eligible to be root. */ 3752 cset->rootable = 1; 3753 /* XXX do this here? */ 3754 raidPtrs[raidID]->root_partition = 1; 3755 } 3756 } 3757 3758 /* 5. Cleanup */ 3759 free(config, M_RAIDFRAME); 3760 3761 *unit = raidID; 3762 return(retcode); 3763 } 3764 3765 void 3766 rf_disk_unbusy(RF_RaidAccessDesc_t *desc) 3767 { 3768 struct buf *bp; 3769 3770 bp = (struct buf *)desc->bp; 3771 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3772 (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ)); 3773 } 3774 3775 void 3776 rf_pool_init(struct pool *p, size_t size, const char *w_chan, 3777 size_t xmin, size_t xmax) 3778 { 3779 pool_init(p, size, 0, 0, 0, w_chan, NULL, IPL_BIO); 3780 pool_sethiwat(p, xmax); 3781 pool_prime(p, xmin); 3782 pool_setlowat(p, xmin); 3783 } 3784 3785 /* 3786 * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see 3787 * if there is IO pending and if that IO could possibly be done for a 3788 * given RAID set. Returns 0 if IO is waiting and can be done, 1 3789 * otherwise. 3790 * 3791 */ 3792 3793 int 3794 rf_buf_queue_check(int raidid) 3795 { 3796 if ((BUFQ_PEEK(raid_softc[raidid].buf_queue) != NULL) && 3797 raidPtrs[raidid]->openings > 0) { 3798 /* there is work to do */ 3799 return 0; 3800 } 3801 /* default is nothing to do */ 3802 return 1; 3803 } 3804 3805 int 3806 rf_getdisksize(struct vnode *vp, struct lwp *l, RF_RaidDisk_t *diskPtr) 3807 { 3808 struct partinfo dpart; 3809 struct dkwedge_info dkw; 3810 int error; 3811 3812 error = VOP_IOCTL(vp, DIOCGPART, &dpart, FREAD, l->l_cred); 3813 if (error == 0) { 3814 diskPtr->blockSize = dpart.disklab->d_secsize; 3815 diskPtr->numBlocks = dpart.part->p_size - rf_protectedSectors; 3816 diskPtr->partitionSize = dpart.part->p_size; 3817 return 0; 3818 } 3819 3820 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, l->l_cred); 3821 if (error == 0) { 3822 struct disk *pdk; 3823 3824 if ((pdk = disk_find(dkw.dkw_parent)) != NULL) 3825 diskPtr->blockSize = DEV_BSIZE << pdk->dk_blkshift; 3826 else 3827 diskPtr->blockSize = 512; /* XXX */ 3828 diskPtr->numBlocks = dkw.dkw_size - rf_protectedSectors; 3829 diskPtr->partitionSize = dkw.dkw_size; 3830 return 0; 3831 } 3832 return error; 3833 } 3834 3835 static int 3836 raid_match(struct device *self, struct cfdata *cfdata, 3837 void *aux) 3838 { 3839 return 1; 3840 } 3841 3842 static void 3843 raid_attach(struct device *parent, struct device *self, 3844 void *aux) 3845 { 3846 3847 } 3848 3849 3850 static int 3851 raid_detach(struct device *self, int flags) 3852 { 3853 struct raid_softc *rs = (struct raid_softc *)self; 3854 3855 if (rs->sc_flags & RAIDF_INITED) 3856 return EBUSY; 3857 3858 return 0; 3859 } 3860 3861 static void 3862 rf_set_properties(struct raid_softc *rs, RF_Raid_t *raidPtr) 3863 { 3864 prop_dictionary_t disk_info, odisk_info, geom; 3865 disk_info = prop_dictionary_create(); 3866 geom = prop_dictionary_create(); 3867 prop_dictionary_set_uint64(geom, "sectors-per-unit", 3868 raidPtr->totalSectors); 3869 prop_dictionary_set_uint32(geom, "sector-size", 3870 raidPtr->bytesPerSector); 3871 3872 prop_dictionary_set_uint16(geom, "sectors-per-track", 3873 raidPtr->Layout.dataSectorsPerStripe); 3874 prop_dictionary_set_uint16(geom, "tracks-per-cylinder", 3875 4 * raidPtr->numCol); 3876 3877 prop_dictionary_set_uint64(geom, "cylinders-per-unit", 3878 raidPtr->totalSectors / (raidPtr->Layout.dataSectorsPerStripe * 3879 (4 * raidPtr->numCol))); 3880 3881 prop_dictionary_set(disk_info, "geometry", geom); 3882 prop_object_release(geom); 3883 prop_dictionary_set(device_properties(rs->sc_dev), 3884 "disk-info", disk_info); 3885 odisk_info = rs->sc_dkdev.dk_info; 3886 rs->sc_dkdev.dk_info = disk_info; 3887 if (odisk_info) 3888 prop_object_release(odisk_info); 3889 } 3890 3891 /* 3892 * Implement forwarding of the DIOCCACHESYNC ioctl to each of the components. 3893 * We end up returning whatever error was returned by the first cache flush 3894 * that fails. 3895 */ 3896 3897 int 3898 rf_sync_component_caches(RF_Raid_t *raidPtr) 3899 { 3900 int c, sparecol; 3901 int e,error; 3902 int force = 1; 3903 3904 error = 0; 3905 for (c = 0; c < raidPtr->numCol; c++) { 3906 if (raidPtr->Disks[c].status == rf_ds_optimal) { 3907 e = VOP_IOCTL(raidPtr->raid_cinfo[c].ci_vp, DIOCCACHESYNC, 3908 &force, FWRITE, NOCRED); 3909 if (e) { 3910 if (e != ENODEV) 3911 printf("raid%d: cache flush to component %s failed.\n", 3912 raidPtr->raidid, raidPtr->Disks[c].devname); 3913 if (error == 0) { 3914 error = e; 3915 } 3916 } 3917 } 3918 } 3919 3920 for( c = 0; c < raidPtr->numSpare ; c++) { 3921 sparecol = raidPtr->numCol + c; 3922 /* Need to ensure that the reconstruct actually completed! */ 3923 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3924 e = VOP_IOCTL(raidPtr->raid_cinfo[sparecol].ci_vp, 3925 DIOCCACHESYNC, &force, FWRITE, NOCRED); 3926 if (e) { 3927 if (e != ENODEV) 3928 printf("raid%d: cache flush to component %s failed.\n", 3929 raidPtr->raidid, raidPtr->Disks[sparecol].devname); 3930 if (error == 0) { 3931 error = e; 3932 } 3933 } 3934 } 3935 } 3936 return error; 3937 } 3938
Indexes created Tue Sep 30 17:09:57 GMT 2025