rf_netbsdkintf.c revision 1.295.6.2.6.1
1 /* $NetBSD: rf_netbsdkintf.c,v 1.295.6.2.6.1 2014/12/02 22:08:01 snj Exp $ */ 2 3 /*- 4 * Copyright (c) 1996, 1997, 1998, 2008-2011 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Greg Oster; Jason R. Thorpe. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (c) 1988 University of Utah. 34 * Copyright (c) 1990, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * This code is derived from software contributed to Berkeley by 38 * the Systems Programming Group of the University of Utah Computer 39 * Science Department. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 66 * 67 * @(#)cd.c 8.2 (Berkeley) 11/16/93 68 */ 69 70 /* 71 * Copyright (c) 1995 Carnegie-Mellon University. 72 * All rights reserved. 73 * 74 * Authors: Mark Holland, Jim Zelenka 75 * 76 * Permission to use, copy, modify and distribute this software and 77 * its documentation is hereby granted, provided that both the copyright 78 * notice and this permission notice appear in all copies of the 79 * software, derivative works or modified versions, and any portions 80 * thereof, and that both notices appear in supporting documentation. 81 * 82 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 83 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 84 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 85 * 86 * Carnegie Mellon requests users of this software to return to 87 * 88 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU 89 * School of Computer Science 90 * Carnegie Mellon University 91 * Pittsburgh PA 15213-3890 92 * 93 * any improvements or extensions that they make and grant Carnegie the 94 * rights to redistribute these changes. 95 */ 96 97 /*********************************************************** 98 * 99 * rf_kintf.c -- the kernel interface routines for RAIDframe 100 * 101 ***********************************************************/ 102 103 #include <sys/cdefs.h> 104 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.295.6.2.6.1 2014/12/02 22:08:01 snj Exp $"); 105 106 #ifdef _KERNEL_OPT 107 #include "opt_compat_netbsd.h" 108 #include "opt_raid_autoconfig.h" 109 #include "raid.h" 110 #endif 111 112 #include <sys/param.h> 113 #include <sys/errno.h> 114 #include <sys/pool.h> 115 #include <sys/proc.h> 116 #include <sys/queue.h> 117 #include <sys/disk.h> 118 #include <sys/device.h> 119 #include <sys/stat.h> 120 #include <sys/ioctl.h> 121 #include <sys/fcntl.h> 122 #include <sys/systm.h> 123 #include <sys/vnode.h> 124 #include <sys/disklabel.h> 125 #include <sys/conf.h> 126 #include <sys/buf.h> 127 #include <sys/bufq.h> 128 #include <sys/reboot.h> 129 #include <sys/kauth.h> 130 131 #include <prop/proplib.h> 132 133 #include <dev/raidframe/raidframevar.h> 134 #include <dev/raidframe/raidframeio.h> 135 #include <dev/raidframe/rf_paritymap.h> 136 137 #include "rf_raid.h" 138 #include "rf_copyback.h" 139 #include "rf_dag.h" 140 #include "rf_dagflags.h" 141 #include "rf_desc.h" 142 #include "rf_diskqueue.h" 143 #include "rf_etimer.h" 144 #include "rf_general.h" 145 #include "rf_kintf.h" 146 #include "rf_options.h" 147 #include "rf_driver.h" 148 #include "rf_parityscan.h" 149 #include "rf_threadstuff.h" 150 151 #ifdef COMPAT_50 152 #include "rf_compat50.h" 153 #endif 154 155 #ifdef DEBUG 156 int rf_kdebug_level = 0; 157 #define db1_printf(a) if (rf_kdebug_level > 0) printf a 158 #else /* DEBUG */ 159 #define db1_printf(a) { } 160 #endif /* DEBUG */ 161 162 static RF_Raid_t **raidPtrs; /* global raid device descriptors */ 163 164 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 165 static rf_declare_mutex2(rf_sparet_wait_mutex); 166 static rf_declare_cond2(rf_sparet_wait_cv); 167 static rf_declare_cond2(rf_sparet_resp_cv); 168 169 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a 170 * spare table */ 171 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from 172 * installation process */ 173 #endif 174 175 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures"); 176 177 /* prototypes */ 178 static void KernelWakeupFunc(struct buf *); 179 static void InitBP(struct buf *, struct vnode *, unsigned, 180 dev_t, RF_SectorNum_t, RF_SectorCount_t, void *, void (*) (struct buf *), 181 void *, int, struct proc *); 182 static void raidinit(RF_Raid_t *); 183 184 void raidattach(int); 185 static int raid_match(device_t, cfdata_t, void *); 186 static void raid_attach(device_t, device_t, void *); 187 static int raid_detach(device_t, int); 188 189 static int raidread_component_area(dev_t, struct vnode *, void *, size_t, 190 daddr_t, daddr_t); 191 static int raidwrite_component_area(dev_t, struct vnode *, void *, size_t, 192 daddr_t, daddr_t, int); 193 194 static int raidwrite_component_label(unsigned, 195 dev_t, struct vnode *, RF_ComponentLabel_t *); 196 static int raidread_component_label(unsigned, 197 dev_t, struct vnode *, RF_ComponentLabel_t *); 198 199 200 dev_type_open(raidopen); 201 dev_type_close(raidclose); 202 dev_type_read(raidread); 203 dev_type_write(raidwrite); 204 dev_type_ioctl(raidioctl); 205 dev_type_strategy(raidstrategy); 206 dev_type_dump(raiddump); 207 dev_type_size(raidsize); 208 209 const struct bdevsw raid_bdevsw = { 210 raidopen, raidclose, raidstrategy, raidioctl, 211 raiddump, raidsize, D_DISK 212 }; 213 214 const struct cdevsw raid_cdevsw = { 215 raidopen, raidclose, raidread, raidwrite, raidioctl, 216 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 217 }; 218 219 static struct dkdriver rf_dkdriver = { raidstrategy, minphys }; 220 221 /* XXX Not sure if the following should be replacing the raidPtrs above, 222 or if it should be used in conjunction with that... 223 */ 224 225 struct raid_softc { 226 device_t sc_dev; 227 int sc_flags; /* flags */ 228 int sc_cflags; /* configuration flags */ 229 uint64_t sc_size; /* size of the raid device */ 230 char sc_xname[20]; /* XXX external name */ 231 struct disk sc_dkdev; /* generic disk device info */ 232 struct bufq_state *buf_queue; /* used for the device queue */ 233 }; 234 /* sc_flags */ 235 #define RAIDF_INITED 0x01 /* unit has been initialized */ 236 #define RAIDF_WLABEL 0x02 /* label area is writable */ 237 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 238 #define RAIDF_SHUTDOWN 0x08 /* unit is being shutdown */ 239 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 240 #define RAIDF_LOCKED 0x80 /* unit is locked */ 241 242 #define raidunit(x) DISKUNIT(x) 243 int numraid = 0; 244 245 extern struct cfdriver raid_cd; 246 CFATTACH_DECL3_NEW(raid, sizeof(struct raid_softc), 247 raid_match, raid_attach, raid_detach, NULL, NULL, NULL, 248 DVF_DETACH_SHUTDOWN); 249 250 /* 251 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 252 * Be aware that large numbers can allow the driver to consume a lot of 253 * kernel memory, especially on writes, and in degraded mode reads. 254 * 255 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 256 * a single 64K write will typically require 64K for the old data, 257 * 64K for the old parity, and 64K for the new parity, for a total 258 * of 192K (if the parity buffer is not re-used immediately). 259 * Even it if is used immediately, that's still 128K, which when multiplied 260 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 261 * 262 * Now in degraded mode, for example, a 64K read on the above setup may 263 * require data reconstruction, which will require *all* of the 4 remaining 264 * disks to participate -- 4 * 32K/disk == 128K again. 265 */ 266 267 #ifndef RAIDOUTSTANDING 268 #define RAIDOUTSTANDING 6 269 #endif 270 271 #define RAIDLABELDEV(dev) \ 272 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 273 274 /* declared here, and made public, for the benefit of KVM stuff.. */ 275 struct raid_softc *raid_softc; 276 277 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *, 278 struct disklabel *); 279 static void raidgetdisklabel(dev_t); 280 static void raidmakedisklabel(struct raid_softc *); 281 282 static int raidlock(struct raid_softc *); 283 static void raidunlock(struct raid_softc *); 284 285 static int raid_detach_unlocked(struct raid_softc *); 286 287 static void rf_markalldirty(RF_Raid_t *); 288 static void rf_set_properties(struct raid_softc *, RF_Raid_t *); 289 290 void rf_ReconThread(struct rf_recon_req *); 291 void rf_RewriteParityThread(RF_Raid_t *raidPtr); 292 void rf_CopybackThread(RF_Raid_t *raidPtr); 293 void rf_ReconstructInPlaceThread(struct rf_recon_req *); 294 int rf_autoconfig(device_t); 295 void rf_buildroothack(RF_ConfigSet_t *); 296 297 RF_AutoConfig_t *rf_find_raid_components(void); 298 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *); 299 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *); 300 int rf_reasonable_label(RF_ComponentLabel_t *, uint64_t); 301 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *); 302 int rf_set_autoconfig(RF_Raid_t *, int); 303 int rf_set_rootpartition(RF_Raid_t *, int); 304 void rf_release_all_vps(RF_ConfigSet_t *); 305 void rf_cleanup_config_set(RF_ConfigSet_t *); 306 int rf_have_enough_components(RF_ConfigSet_t *); 307 int rf_auto_config_set(RF_ConfigSet_t *, int *); 308 static void rf_fix_old_label_size(RF_ComponentLabel_t *, uint64_t); 309 310 /* 311 * Debugging, mostly. Set to 0 to not allow autoconfig to take place. 312 * Note that this is overridden by having RAID_AUTOCONFIG as an option 313 * in the kernel config file. 314 */ 315 #ifdef RAID_AUTOCONFIG 316 int raidautoconfig = 1; 317 #else 318 int raidautoconfig = 0; 319 #endif 320 static bool raidautoconfigdone = false; 321 322 struct RF_Pools_s rf_pools; 323 324 void 325 raidattach(int num) 326 { 327 int raidID; 328 int i, rc; 329 330 aprint_debug("raidattach: Asked for %d units\n", num); 331 332 if (num <= 0) { 333 #ifdef DIAGNOSTIC 334 panic("raidattach: count <= 0"); 335 #endif 336 return; 337 } 338 /* This is where all the initialization stuff gets done. */ 339 340 numraid = num; 341 342 /* Make some space for requested number of units... */ 343 344 RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **)); 345 if (raidPtrs == NULL) { 346 panic("raidPtrs is NULL!!"); 347 } 348 349 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 350 rf_init_mutex2(rf_sparet_wait_mutex, IPL_VM); 351 rf_init_cond2(rf_sparet_wait_cv, "sparetw"); 352 rf_init_cond2(rf_sparet_resp_cv, "rfgst"); 353 354 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 355 #endif 356 357 for (i = 0; i < num; i++) 358 raidPtrs[i] = NULL; 359 rc = rf_BootRaidframe(); 360 if (rc == 0) 361 aprint_verbose("Kernelized RAIDframe activated\n"); 362 else 363 panic("Serious error booting RAID!!"); 364 365 /* put together some datastructures like the CCD device does.. This 366 * lets us lock the device and what-not when it gets opened. */ 367 368 raid_softc = (struct raid_softc *) 369 malloc(num * sizeof(struct raid_softc), 370 M_RAIDFRAME, M_NOWAIT); 371 if (raid_softc == NULL) { 372 aprint_error("WARNING: no memory for RAIDframe driver\n"); 373 return; 374 } 375 376 memset(raid_softc, 0, num * sizeof(struct raid_softc)); 377 378 for (raidID = 0; raidID < num; raidID++) { 379 bufq_alloc(&raid_softc[raidID].buf_queue, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK); 380 381 RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t), 382 (RF_Raid_t *)); 383 if (raidPtrs[raidID] == NULL) { 384 aprint_error("WARNING: raidPtrs[%d] is NULL\n", raidID); 385 numraid = raidID; 386 return; 387 } 388 } 389 390 if (config_cfattach_attach(raid_cd.cd_name, &raid_ca)) { 391 aprint_error("raidattach: config_cfattach_attach failed?\n"); 392 } 393 394 raidautoconfigdone = false; 395 396 /* 397 * Register a finalizer which will be used to auto-config RAID 398 * sets once all real hardware devices have been found. 399 */ 400 if (config_finalize_register(NULL, rf_autoconfig) != 0) 401 aprint_error("WARNING: unable to register RAIDframe finalizer\n"); 402 } 403 404 int 405 rf_autoconfig(device_t self) 406 { 407 RF_AutoConfig_t *ac_list; 408 RF_ConfigSet_t *config_sets; 409 410 if (!raidautoconfig || raidautoconfigdone == true) 411 return (0); 412 413 /* XXX This code can only be run once. */ 414 raidautoconfigdone = true; 415 416 /* 1. locate all RAID components on the system */ 417 aprint_debug("Searching for RAID components...\n"); 418 ac_list = rf_find_raid_components(); 419 420 /* 2. Sort them into their respective sets. */ 421 config_sets = rf_create_auto_sets(ac_list); 422 423 /* 424 * 3. Evaluate each set andconfigure the valid ones. 425 * This gets done in rf_buildroothack(). 426 */ 427 rf_buildroothack(config_sets); 428 429 return 1; 430 } 431 432 void 433 rf_buildroothack(RF_ConfigSet_t *config_sets) 434 { 435 RF_ConfigSet_t *cset; 436 RF_ConfigSet_t *next_cset; 437 int retcode; 438 int raidID; 439 int rootID; 440 int col; 441 int num_root; 442 char *devname; 443 444 rootID = 0; 445 num_root = 0; 446 cset = config_sets; 447 while (cset != NULL) { 448 next_cset = cset->next; 449 if (rf_have_enough_components(cset) && 450 cset->ac->clabel->autoconfigure==1) { 451 retcode = rf_auto_config_set(cset,&raidID); 452 if (!retcode) { 453 aprint_debug("raid%d: configured ok\n", raidID); 454 if (cset->rootable) { 455 rootID = raidID; 456 num_root++; 457 } 458 } else { 459 /* The autoconfig didn't work :( */ 460 aprint_debug("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 461 rf_release_all_vps(cset); 462 } 463 } else { 464 /* we're not autoconfiguring this set... 465 release the associated resources */ 466 rf_release_all_vps(cset); 467 } 468 /* cleanup */ 469 rf_cleanup_config_set(cset); 470 cset = next_cset; 471 } 472 473 /* if the user has specified what the root device should be 474 then we don't touch booted_device or boothowto... */ 475 476 if (rootspec != NULL) 477 return; 478 479 /* we found something bootable... */ 480 481 if (num_root == 1) { 482 booted_device = raid_softc[rootID].sc_dev; 483 } else if (num_root > 1) { 484 485 /* 486 * Maybe the MD code can help. If it cannot, then 487 * setroot() will discover that we have no 488 * booted_device and will ask the user if nothing was 489 * hardwired in the kernel config file 490 */ 491 492 if (booted_device == NULL) 493 cpu_rootconf(); 494 if (booted_device == NULL) 495 return; 496 497 num_root = 0; 498 for (raidID = 0; raidID < numraid; raidID++) { 499 if (raidPtrs[raidID]->valid == 0) 500 continue; 501 502 if (raidPtrs[raidID]->root_partition == 0) 503 continue; 504 505 for (col = 0; col < raidPtrs[raidID]->numCol; col++) { 506 devname = raidPtrs[raidID]->Disks[col].devname; 507 devname += sizeof("/dev/") - 1; 508 if (strncmp(devname, device_xname(booted_device), 509 strlen(device_xname(booted_device))) != 0) 510 continue; 511 aprint_debug("raid%d includes boot device %s\n", 512 raidID, devname); 513 num_root++; 514 rootID = raidID; 515 } 516 } 517 518 if (num_root == 1) { 519 booted_device = raid_softc[rootID].sc_dev; 520 } else { 521 /* we can't guess.. require the user to answer... */ 522 boothowto |= RB_ASKNAME; 523 } 524 } 525 } 526 527 528 int 529 raidsize(dev_t dev) 530 { 531 struct raid_softc *rs; 532 struct disklabel *lp; 533 int part, unit, omask, size; 534 535 unit = raidunit(dev); 536 if (unit >= numraid) 537 return (-1); 538 rs = &raid_softc[unit]; 539 540 if ((rs->sc_flags & RAIDF_INITED) == 0) 541 return (-1); 542 543 part = DISKPART(dev); 544 omask = rs->sc_dkdev.dk_openmask & (1 << part); 545 lp = rs->sc_dkdev.dk_label; 546 547 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curlwp)) 548 return (-1); 549 550 if (lp->d_partitions[part].p_fstype != FS_SWAP) 551 size = -1; 552 else 553 size = lp->d_partitions[part].p_size * 554 (lp->d_secsize / DEV_BSIZE); 555 556 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curlwp)) 557 return (-1); 558 559 return (size); 560 561 } 562 563 int 564 raiddump(dev_t dev, daddr_t blkno, void *va, size_t size) 565 { 566 int unit = raidunit(dev); 567 struct raid_softc *rs; 568 const struct bdevsw *bdev; 569 struct disklabel *lp; 570 RF_Raid_t *raidPtr; 571 daddr_t offset; 572 int part, c, sparecol, j, scol, dumpto; 573 int error = 0; 574 575 if (unit >= numraid) 576 return (ENXIO); 577 578 rs = &raid_softc[unit]; 579 raidPtr = raidPtrs[unit]; 580 581 if ((rs->sc_flags & RAIDF_INITED) == 0) 582 return ENXIO; 583 584 /* we only support dumping to RAID 1 sets */ 585 if (raidPtr->Layout.numDataCol != 1 || 586 raidPtr->Layout.numParityCol != 1) 587 return EINVAL; 588 589 590 if ((error = raidlock(rs)) != 0) 591 return error; 592 593 if (size % DEV_BSIZE != 0) { 594 error = EINVAL; 595 goto out; 596 } 597 598 if (blkno + size / DEV_BSIZE > rs->sc_size) { 599 printf("%s: blkno (%" PRIu64 ") + size / DEV_BSIZE (%zu) > " 600 "sc->sc_size (%" PRIu64 ")\n", __func__, blkno, 601 size / DEV_BSIZE, rs->sc_size); 602 error = EINVAL; 603 goto out; 604 } 605 606 part = DISKPART(dev); 607 lp = rs->sc_dkdev.dk_label; 608 offset = lp->d_partitions[part].p_offset + RF_PROTECTED_SECTORS; 609 610 /* figure out what device is alive.. */ 611 612 /* 613 Look for a component to dump to. The preference for the 614 component to dump to is as follows: 615 1) the master 616 2) a used_spare of the master 617 3) the slave 618 4) a used_spare of the slave 619 */ 620 621 dumpto = -1; 622 for (c = 0; c < raidPtr->numCol; c++) { 623 if (raidPtr->Disks[c].status == rf_ds_optimal) { 624 /* this might be the one */ 625 dumpto = c; 626 break; 627 } 628 } 629 630 /* 631 At this point we have possibly selected a live master or a 632 live slave. We now check to see if there is a spared 633 master (or a spared slave), if we didn't find a live master 634 or a live slave. 635 */ 636 637 for (c = 0; c < raidPtr->numSpare; c++) { 638 sparecol = raidPtr->numCol + c; 639 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 640 /* How about this one? */ 641 scol = -1; 642 for(j=0;j<raidPtr->numCol;j++) { 643 if (raidPtr->Disks[j].spareCol == sparecol) { 644 scol = j; 645 break; 646 } 647 } 648 if (scol == 0) { 649 /* 650 We must have found a spared master! 651 We'll take that over anything else 652 found so far. (We couldn't have 653 found a real master before, since 654 this is a used spare, and it's 655 saying that it's replacing the 656 master.) On reboot (with 657 autoconfiguration turned on) 658 sparecol will become the 1st 659 component (component0) of this set. 660 */ 661 dumpto = sparecol; 662 break; 663 } else if (scol != -1) { 664 /* 665 Must be a spared slave. We'll dump 666 to that if we havn't found anything 667 else so far. 668 */ 669 if (dumpto == -1) 670 dumpto = sparecol; 671 } 672 } 673 } 674 675 if (dumpto == -1) { 676 /* we couldn't find any live components to dump to!?!? 677 */ 678 error = EINVAL; 679 goto out; 680 } 681 682 bdev = bdevsw_lookup(raidPtr->Disks[dumpto].dev); 683 684 /* 685 Note that blkno is relative to this particular partition. 686 By adding the offset of this partition in the RAID 687 set, and also adding RF_PROTECTED_SECTORS, we get a 688 value that is relative to the partition used for the 689 underlying component. 690 */ 691 692 error = (*bdev->d_dump)(raidPtr->Disks[dumpto].dev, 693 blkno + offset, va, size); 694 695 out: 696 raidunlock(rs); 697 698 return error; 699 } 700 /* ARGSUSED */ 701 int 702 raidopen(dev_t dev, int flags, int fmt, 703 struct lwp *l) 704 { 705 int unit = raidunit(dev); 706 struct raid_softc *rs; 707 struct disklabel *lp; 708 int part, pmask; 709 int error = 0; 710 711 if (unit >= numraid) 712 return (ENXIO); 713 rs = &raid_softc[unit]; 714 715 if ((error = raidlock(rs)) != 0) 716 return (error); 717 718 if ((rs->sc_flags & RAIDF_SHUTDOWN) != 0) { 719 error = EBUSY; 720 goto bad; 721 } 722 723 lp = rs->sc_dkdev.dk_label; 724 725 part = DISKPART(dev); 726 727 /* 728 * If there are wedges, and this is not RAW_PART, then we 729 * need to fail. 730 */ 731 if (rs->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) { 732 error = EBUSY; 733 goto bad; 734 } 735 pmask = (1 << part); 736 737 if ((rs->sc_flags & RAIDF_INITED) && 738 (rs->sc_dkdev.dk_openmask == 0)) 739 raidgetdisklabel(dev); 740 741 /* make sure that this partition exists */ 742 743 if (part != RAW_PART) { 744 if (((rs->sc_flags & RAIDF_INITED) == 0) || 745 ((part >= lp->d_npartitions) || 746 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 747 error = ENXIO; 748 goto bad; 749 } 750 } 751 /* Prevent this unit from being unconfigured while open. */ 752 switch (fmt) { 753 case S_IFCHR: 754 rs->sc_dkdev.dk_copenmask |= pmask; 755 break; 756 757 case S_IFBLK: 758 rs->sc_dkdev.dk_bopenmask |= pmask; 759 break; 760 } 761 762 if ((rs->sc_dkdev.dk_openmask == 0) && 763 ((rs->sc_flags & RAIDF_INITED) != 0)) { 764 /* First one... mark things as dirty... Note that we *MUST* 765 have done a configure before this. I DO NOT WANT TO BE 766 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 767 THAT THEY BELONG TOGETHER!!!!! */ 768 /* XXX should check to see if we're only open for reading 769 here... If so, we needn't do this, but then need some 770 other way of keeping track of what's happened.. */ 771 772 rf_markalldirty(raidPtrs[unit]); 773 } 774 775 776 rs->sc_dkdev.dk_openmask = 777 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 778 779 bad: 780 raidunlock(rs); 781 782 return (error); 783 784 785 } 786 /* ARGSUSED */ 787 int 788 raidclose(dev_t dev, int flags, int fmt, struct lwp *l) 789 { 790 int unit = raidunit(dev); 791 struct raid_softc *rs; 792 int error = 0; 793 int part; 794 795 if (unit >= numraid) 796 return (ENXIO); 797 rs = &raid_softc[unit]; 798 799 if ((error = raidlock(rs)) != 0) 800 return (error); 801 802 part = DISKPART(dev); 803 804 /* ...that much closer to allowing unconfiguration... */ 805 switch (fmt) { 806 case S_IFCHR: 807 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 808 break; 809 810 case S_IFBLK: 811 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 812 break; 813 } 814 rs->sc_dkdev.dk_openmask = 815 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 816 817 if ((rs->sc_dkdev.dk_openmask == 0) && 818 ((rs->sc_flags & RAIDF_INITED) != 0)) { 819 /* Last one... device is not unconfigured yet. 820 Device shutdown has taken care of setting the 821 clean bits if RAIDF_INITED is not set 822 mark things as clean... */ 823 824 rf_update_component_labels(raidPtrs[unit], 825 RF_FINAL_COMPONENT_UPDATE); 826 827 /* If the kernel is shutting down, it will detach 828 * this RAID set soon enough. 829 */ 830 } 831 832 raidunlock(rs); 833 return (0); 834 835 } 836 837 void 838 raidstrategy(struct buf *bp) 839 { 840 unsigned int raidID = raidunit(bp->b_dev); 841 RF_Raid_t *raidPtr; 842 struct raid_softc *rs = &raid_softc[raidID]; 843 int wlabel; 844 845 if ((rs->sc_flags & RAIDF_INITED) ==0) { 846 bp->b_error = ENXIO; 847 goto done; 848 } 849 if (raidID >= numraid || !raidPtrs[raidID]) { 850 bp->b_error = ENODEV; 851 goto done; 852 } 853 raidPtr = raidPtrs[raidID]; 854 if (!raidPtr->valid) { 855 bp->b_error = ENODEV; 856 goto done; 857 } 858 if (bp->b_bcount == 0) { 859 db1_printf(("b_bcount is zero..\n")); 860 goto done; 861 } 862 863 /* 864 * Do bounds checking and adjust transfer. If there's an 865 * error, the bounds check will flag that for us. 866 */ 867 868 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 869 if (DISKPART(bp->b_dev) == RAW_PART) { 870 uint64_t size; /* device size in DEV_BSIZE unit */ 871 872 if (raidPtr->logBytesPerSector > DEV_BSHIFT) { 873 size = raidPtr->totalSectors << 874 (raidPtr->logBytesPerSector - DEV_BSHIFT); 875 } else { 876 size = raidPtr->totalSectors >> 877 (DEV_BSHIFT - raidPtr->logBytesPerSector); 878 } 879 if (bounds_check_with_mediasize(bp, DEV_BSIZE, size) <= 0) { 880 goto done; 881 } 882 } else { 883 if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) { 884 db1_printf(("Bounds check failed!!:%d %d\n", 885 (int) bp->b_blkno, (int) wlabel)); 886 goto done; 887 } 888 } 889 890 rf_lock_mutex2(raidPtr->iodone_lock); 891 892 bp->b_resid = 0; 893 894 /* stuff it onto our queue */ 895 bufq_put(rs->buf_queue, bp); 896 897 /* scheduled the IO to happen at the next convenient time */ 898 rf_signal_cond2(raidPtr->iodone_cv); 899 rf_unlock_mutex2(raidPtr->iodone_lock); 900 901 return; 902 903 done: 904 bp->b_resid = bp->b_bcount; 905 biodone(bp); 906 } 907 /* ARGSUSED */ 908 int 909 raidread(dev_t dev, struct uio *uio, int flags) 910 { 911 int unit = raidunit(dev); 912 struct raid_softc *rs; 913 914 if (unit >= numraid) 915 return (ENXIO); 916 rs = &raid_softc[unit]; 917 918 if ((rs->sc_flags & RAIDF_INITED) == 0) 919 return (ENXIO); 920 921 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio)); 922 923 } 924 /* ARGSUSED */ 925 int 926 raidwrite(dev_t dev, struct uio *uio, int flags) 927 { 928 int unit = raidunit(dev); 929 struct raid_softc *rs; 930 931 if (unit >= numraid) 932 return (ENXIO); 933 rs = &raid_softc[unit]; 934 935 if ((rs->sc_flags & RAIDF_INITED) == 0) 936 return (ENXIO); 937 938 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio)); 939 940 } 941 942 static int 943 raid_detach_unlocked(struct raid_softc *rs) 944 { 945 int error; 946 RF_Raid_t *raidPtr; 947 948 raidPtr = raidPtrs[device_unit(rs->sc_dev)]; 949 950 /* 951 * If somebody has a partition mounted, we shouldn't 952 * shutdown. 953 */ 954 if (rs->sc_dkdev.dk_openmask != 0) 955 return EBUSY; 956 957 if ((rs->sc_flags & RAIDF_INITED) == 0) 958 ; /* not initialized: nothing to do */ 959 else if ((error = rf_Shutdown(raidPtr)) != 0) 960 return error; 961 else 962 rs->sc_flags &= ~(RAIDF_INITED|RAIDF_SHUTDOWN); 963 964 /* Detach the disk. */ 965 dkwedge_delall(&rs->sc_dkdev); 966 disk_detach(&rs->sc_dkdev); 967 disk_destroy(&rs->sc_dkdev); 968 969 aprint_normal_dev(rs->sc_dev, "detached\n"); 970 971 return 0; 972 } 973 974 int 975 raidioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 976 { 977 int unit = raidunit(dev); 978 int error = 0; 979 int part, pmask, s; 980 cfdata_t cf; 981 struct raid_softc *rs; 982 RF_Config_t *k_cfg, *u_cfg; 983 RF_Raid_t *raidPtr; 984 RF_RaidDisk_t *diskPtr; 985 RF_AccTotals_t *totals; 986 RF_DeviceConfig_t *d_cfg, **ucfgp; 987 u_char *specific_buf; 988 int retcode = 0; 989 int column; 990 /* int raidid; */ 991 struct rf_recon_req *rrcopy, *rr; 992 RF_ComponentLabel_t *clabel; 993 RF_ComponentLabel_t *ci_label; 994 RF_ComponentLabel_t **clabel_ptr; 995 RF_SingleComponent_t *sparePtr,*componentPtr; 996 RF_SingleComponent_t component; 997 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 998 int i, j, d; 999 #ifdef __HAVE_OLD_DISKLABEL 1000 struct disklabel newlabel; 1001 #endif 1002 struct dkwedge_info *dkw; 1003 1004 if (unit >= numraid) 1005 return (ENXIO); 1006 rs = &raid_softc[unit]; 1007 raidPtr = raidPtrs[unit]; 1008 1009 db1_printf(("raidioctl: %d %d %d %lu\n", (int) dev, 1010 (int) DISKPART(dev), (int) unit, cmd)); 1011 1012 /* Must be open for writes for these commands... */ 1013 switch (cmd) { 1014 #ifdef DIOCGSECTORSIZE 1015 case DIOCGSECTORSIZE: 1016 *(u_int *)data = raidPtr->bytesPerSector; 1017 return 0; 1018 case DIOCGMEDIASIZE: 1019 *(off_t *)data = 1020 (off_t)raidPtr->totalSectors * raidPtr->bytesPerSector; 1021 return 0; 1022 #endif 1023 case DIOCSDINFO: 1024 case DIOCWDINFO: 1025 #ifdef __HAVE_OLD_DISKLABEL 1026 case ODIOCWDINFO: 1027 case ODIOCSDINFO: 1028 #endif 1029 case DIOCWLABEL: 1030 case DIOCAWEDGE: 1031 case DIOCDWEDGE: 1032 case DIOCSSTRATEGY: 1033 if ((flag & FWRITE) == 0) 1034 return (EBADF); 1035 } 1036 1037 /* Must be initialized for these... */ 1038 switch (cmd) { 1039 case DIOCGDINFO: 1040 case DIOCSDINFO: 1041 case DIOCWDINFO: 1042 #ifdef __HAVE_OLD_DISKLABEL 1043 case ODIOCGDINFO: 1044 case ODIOCWDINFO: 1045 case ODIOCSDINFO: 1046 case ODIOCGDEFLABEL: 1047 #endif 1048 case DIOCGPART: 1049 case DIOCWLABEL: 1050 case DIOCGDEFLABEL: 1051 case DIOCAWEDGE: 1052 case DIOCDWEDGE: 1053 case DIOCLWEDGES: 1054 case DIOCCACHESYNC: 1055 case RAIDFRAME_SHUTDOWN: 1056 case RAIDFRAME_REWRITEPARITY: 1057 case RAIDFRAME_GET_INFO: 1058 case RAIDFRAME_RESET_ACCTOTALS: 1059 case RAIDFRAME_GET_ACCTOTALS: 1060 case RAIDFRAME_KEEP_ACCTOTALS: 1061 case RAIDFRAME_GET_SIZE: 1062 case RAIDFRAME_FAIL_DISK: 1063 case RAIDFRAME_COPYBACK: 1064 case RAIDFRAME_CHECK_RECON_STATUS: 1065 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1066 case RAIDFRAME_GET_COMPONENT_LABEL: 1067 case RAIDFRAME_SET_COMPONENT_LABEL: 1068 case RAIDFRAME_ADD_HOT_SPARE: 1069 case RAIDFRAME_REMOVE_HOT_SPARE: 1070 case RAIDFRAME_INIT_LABELS: 1071 case RAIDFRAME_REBUILD_IN_PLACE: 1072 case RAIDFRAME_CHECK_PARITY: 1073 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1074 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1075 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1076 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1077 case RAIDFRAME_SET_AUTOCONFIG: 1078 case RAIDFRAME_SET_ROOT: 1079 case RAIDFRAME_DELETE_COMPONENT: 1080 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1081 case RAIDFRAME_PARITYMAP_STATUS: 1082 case RAIDFRAME_PARITYMAP_GET_DISABLE: 1083 case RAIDFRAME_PARITYMAP_SET_DISABLE: 1084 case RAIDFRAME_PARITYMAP_SET_PARAMS: 1085 case DIOCGSTRATEGY: 1086 case DIOCSSTRATEGY: 1087 if ((rs->sc_flags & RAIDF_INITED) == 0) 1088 return (ENXIO); 1089 } 1090 1091 switch (cmd) { 1092 #ifdef COMPAT_50 1093 case RAIDFRAME_GET_INFO50: 1094 return rf_get_info50(raidPtr, data); 1095 1096 case RAIDFRAME_CONFIGURE50: 1097 if ((retcode = rf_config50(raidPtr, unit, data, &k_cfg)) != 0) 1098 return retcode; 1099 goto config; 1100 #endif 1101 /* configure the system */ 1102 case RAIDFRAME_CONFIGURE: 1103 1104 if (raidPtr->valid) { 1105 /* There is a valid RAID set running on this unit! */ 1106 printf("raid%d: Device already configured!\n",unit); 1107 return(EINVAL); 1108 } 1109 1110 /* copy-in the configuration information */ 1111 /* data points to a pointer to the configuration structure */ 1112 1113 u_cfg = *((RF_Config_t **) data); 1114 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *)); 1115 if (k_cfg == NULL) { 1116 return (ENOMEM); 1117 } 1118 retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t)); 1119 if (retcode) { 1120 RF_Free(k_cfg, sizeof(RF_Config_t)); 1121 db1_printf(("rf_ioctl: retcode=%d copyin.1\n", 1122 retcode)); 1123 return (retcode); 1124 } 1125 goto config; 1126 config: 1127 /* allocate a buffer for the layout-specific data, and copy it 1128 * in */ 1129 if (k_cfg->layoutSpecificSize) { 1130 if (k_cfg->layoutSpecificSize > 10000) { 1131 /* sanity check */ 1132 RF_Free(k_cfg, sizeof(RF_Config_t)); 1133 return (EINVAL); 1134 } 1135 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize, 1136 (u_char *)); 1137 if (specific_buf == NULL) { 1138 RF_Free(k_cfg, sizeof(RF_Config_t)); 1139 return (ENOMEM); 1140 } 1141 retcode = copyin(k_cfg->layoutSpecific, specific_buf, 1142 k_cfg->layoutSpecificSize); 1143 if (retcode) { 1144 RF_Free(k_cfg, sizeof(RF_Config_t)); 1145 RF_Free(specific_buf, 1146 k_cfg->layoutSpecificSize); 1147 db1_printf(("rf_ioctl: retcode=%d copyin.2\n", 1148 retcode)); 1149 return (retcode); 1150 } 1151 } else 1152 specific_buf = NULL; 1153 k_cfg->layoutSpecific = specific_buf; 1154 1155 /* should do some kind of sanity check on the configuration. 1156 * Store the sum of all the bytes in the last byte? */ 1157 1158 /* configure the system */ 1159 1160 /* 1161 * Clear the entire RAID descriptor, just to make sure 1162 * there is no stale data left in the case of a 1163 * reconfiguration 1164 */ 1165 memset(raidPtr, 0, sizeof(*raidPtr)); 1166 raidPtr->raidid = unit; 1167 1168 retcode = rf_Configure(raidPtr, k_cfg, NULL); 1169 1170 if (retcode == 0) { 1171 1172 /* allow this many simultaneous IO's to 1173 this RAID device */ 1174 raidPtr->openings = RAIDOUTSTANDING; 1175 1176 raidinit(raidPtr); 1177 rf_markalldirty(raidPtr); 1178 } 1179 /* free the buffers. No return code here. */ 1180 if (k_cfg->layoutSpecificSize) { 1181 RF_Free(specific_buf, k_cfg->layoutSpecificSize); 1182 } 1183 RF_Free(k_cfg, sizeof(RF_Config_t)); 1184 1185 return (retcode); 1186 1187 /* shutdown the system */ 1188 case RAIDFRAME_SHUTDOWN: 1189 1190 part = DISKPART(dev); 1191 pmask = (1 << part); 1192 1193 if ((error = raidlock(rs)) != 0) 1194 return (error); 1195 1196 if ((rs->sc_dkdev.dk_openmask & ~pmask) || 1197 ((rs->sc_dkdev.dk_bopenmask & pmask) && 1198 (rs->sc_dkdev.dk_copenmask & pmask))) 1199 retcode = EBUSY; 1200 else { 1201 rs->sc_flags |= RAIDF_SHUTDOWN; 1202 rs->sc_dkdev.dk_copenmask &= ~pmask; 1203 rs->sc_dkdev.dk_bopenmask &= ~pmask; 1204 rs->sc_dkdev.dk_openmask &= ~pmask; 1205 retcode = 0; 1206 } 1207 1208 raidunlock(rs); 1209 1210 if (retcode != 0) 1211 return retcode; 1212 1213 /* free the pseudo device attach bits */ 1214 1215 cf = device_cfdata(rs->sc_dev); 1216 if ((retcode = config_detach(rs->sc_dev, DETACH_QUIET)) == 0) 1217 free(cf, M_RAIDFRAME); 1218 1219 return (retcode); 1220 case RAIDFRAME_GET_COMPONENT_LABEL: 1221 clabel_ptr = (RF_ComponentLabel_t **) data; 1222 /* need to read the component label for the disk indicated 1223 by row,column in clabel */ 1224 1225 /* 1226 * Perhaps there should be an option to skip the in-core 1227 * copy and hit the disk, as with disklabel(8). 1228 */ 1229 RF_Malloc(clabel, sizeof(*clabel), (RF_ComponentLabel_t *)); 1230 1231 retcode = copyin(*clabel_ptr, clabel, sizeof(*clabel)); 1232 1233 if (retcode) { 1234 RF_Free(clabel, sizeof(*clabel)); 1235 return retcode; 1236 } 1237 1238 clabel->row = 0; /* Don't allow looking at anything else.*/ 1239 1240 column = clabel->column; 1241 1242 if ((column < 0) || (column >= raidPtr->numCol + 1243 raidPtr->numSpare)) { 1244 RF_Free(clabel, sizeof(*clabel)); 1245 return EINVAL; 1246 } 1247 1248 RF_Free(clabel, sizeof(*clabel)); 1249 1250 clabel = raidget_component_label(raidPtr, column); 1251 1252 return copyout(clabel, *clabel_ptr, sizeof(**clabel_ptr)); 1253 1254 #if 0 1255 case RAIDFRAME_SET_COMPONENT_LABEL: 1256 clabel = (RF_ComponentLabel_t *) data; 1257 1258 /* XXX check the label for valid stuff... */ 1259 /* Note that some things *should not* get modified -- 1260 the user should be re-initing the labels instead of 1261 trying to patch things. 1262 */ 1263 1264 raidid = raidPtr->raidid; 1265 #ifdef DEBUG 1266 printf("raid%d: Got component label:\n", raidid); 1267 printf("raid%d: Version: %d\n", raidid, clabel->version); 1268 printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number); 1269 printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter); 1270 printf("raid%d: Column: %d\n", raidid, clabel->column); 1271 printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns); 1272 printf("raid%d: Clean: %d\n", raidid, clabel->clean); 1273 printf("raid%d: Status: %d\n", raidid, clabel->status); 1274 #endif 1275 clabel->row = 0; 1276 column = clabel->column; 1277 1278 if ((column < 0) || (column >= raidPtr->numCol)) { 1279 return(EINVAL); 1280 } 1281 1282 /* XXX this isn't allowed to do anything for now :-) */ 1283 1284 /* XXX and before it is, we need to fill in the rest 1285 of the fields!?!?!?! */ 1286 memcpy(raidget_component_label(raidPtr, column), 1287 clabel, sizeof(*clabel)); 1288 raidflush_component_label(raidPtr, column); 1289 return (0); 1290 #endif 1291 1292 case RAIDFRAME_INIT_LABELS: 1293 clabel = (RF_ComponentLabel_t *) data; 1294 /* 1295 we only want the serial number from 1296 the above. We get all the rest of the information 1297 from the config that was used to create this RAID 1298 set. 1299 */ 1300 1301 raidPtr->serial_number = clabel->serial_number; 1302 1303 for(column=0;column<raidPtr->numCol;column++) { 1304 diskPtr = &raidPtr->Disks[column]; 1305 if (!RF_DEAD_DISK(diskPtr->status)) { 1306 ci_label = raidget_component_label(raidPtr, 1307 column); 1308 /* Zeroing this is important. */ 1309 memset(ci_label, 0, sizeof(*ci_label)); 1310 raid_init_component_label(raidPtr, ci_label); 1311 ci_label->serial_number = 1312 raidPtr->serial_number; 1313 ci_label->row = 0; /* we dont' pretend to support more */ 1314 rf_component_label_set_partitionsize(ci_label, 1315 diskPtr->partitionSize); 1316 ci_label->column = column; 1317 raidflush_component_label(raidPtr, column); 1318 } 1319 /* XXXjld what about the spares? */ 1320 } 1321 1322 return (retcode); 1323 case RAIDFRAME_SET_AUTOCONFIG: 1324 d = rf_set_autoconfig(raidPtr, *(int *) data); 1325 printf("raid%d: New autoconfig value is: %d\n", 1326 raidPtr->raidid, d); 1327 *(int *) data = d; 1328 return (retcode); 1329 1330 case RAIDFRAME_SET_ROOT: 1331 d = rf_set_rootpartition(raidPtr, *(int *) data); 1332 printf("raid%d: New rootpartition value is: %d\n", 1333 raidPtr->raidid, d); 1334 *(int *) data = d; 1335 return (retcode); 1336 1337 /* initialize all parity */ 1338 case RAIDFRAME_REWRITEPARITY: 1339 1340 if (raidPtr->Layout.map->faultsTolerated == 0) { 1341 /* Parity for RAID 0 is trivially correct */ 1342 raidPtr->parity_good = RF_RAID_CLEAN; 1343 return(0); 1344 } 1345 1346 if (raidPtr->parity_rewrite_in_progress == 1) { 1347 /* Re-write is already in progress! */ 1348 return(EINVAL); 1349 } 1350 1351 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1352 rf_RewriteParityThread, 1353 raidPtr,"raid_parity"); 1354 return (retcode); 1355 1356 1357 case RAIDFRAME_ADD_HOT_SPARE: 1358 sparePtr = (RF_SingleComponent_t *) data; 1359 memcpy( &component, sparePtr, sizeof(RF_SingleComponent_t)); 1360 retcode = rf_add_hot_spare(raidPtr, &component); 1361 return(retcode); 1362 1363 case RAIDFRAME_REMOVE_HOT_SPARE: 1364 return(retcode); 1365 1366 case RAIDFRAME_DELETE_COMPONENT: 1367 componentPtr = (RF_SingleComponent_t *)data; 1368 memcpy( &component, componentPtr, 1369 sizeof(RF_SingleComponent_t)); 1370 retcode = rf_delete_component(raidPtr, &component); 1371 return(retcode); 1372 1373 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1374 componentPtr = (RF_SingleComponent_t *)data; 1375 memcpy( &component, componentPtr, 1376 sizeof(RF_SingleComponent_t)); 1377 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1378 return(retcode); 1379 1380 case RAIDFRAME_REBUILD_IN_PLACE: 1381 1382 if (raidPtr->Layout.map->faultsTolerated == 0) { 1383 /* Can't do this on a RAID 0!! */ 1384 return(EINVAL); 1385 } 1386 1387 if (raidPtr->recon_in_progress == 1) { 1388 /* a reconstruct is already in progress! */ 1389 return(EINVAL); 1390 } 1391 1392 componentPtr = (RF_SingleComponent_t *) data; 1393 memcpy( &component, componentPtr, 1394 sizeof(RF_SingleComponent_t)); 1395 component.row = 0; /* we don't support any more */ 1396 column = component.column; 1397 1398 if ((column < 0) || (column >= raidPtr->numCol)) { 1399 return(EINVAL); 1400 } 1401 1402 rf_lock_mutex2(raidPtr->mutex); 1403 if ((raidPtr->Disks[column].status == rf_ds_optimal) && 1404 (raidPtr->numFailures > 0)) { 1405 /* XXX 0 above shouldn't be constant!!! */ 1406 /* some component other than this has failed. 1407 Let's not make things worse than they already 1408 are... */ 1409 printf("raid%d: Unable to reconstruct to disk at:\n", 1410 raidPtr->raidid); 1411 printf("raid%d: Col: %d Too many failures.\n", 1412 raidPtr->raidid, column); 1413 rf_unlock_mutex2(raidPtr->mutex); 1414 return (EINVAL); 1415 } 1416 if (raidPtr->Disks[column].status == 1417 rf_ds_reconstructing) { 1418 printf("raid%d: Unable to reconstruct to disk at:\n", 1419 raidPtr->raidid); 1420 printf("raid%d: Col: %d Reconstruction already occuring!\n", raidPtr->raidid, column); 1421 1422 rf_unlock_mutex2(raidPtr->mutex); 1423 return (EINVAL); 1424 } 1425 if (raidPtr->Disks[column].status == rf_ds_spared) { 1426 rf_unlock_mutex2(raidPtr->mutex); 1427 return (EINVAL); 1428 } 1429 rf_unlock_mutex2(raidPtr->mutex); 1430 1431 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1432 if (rrcopy == NULL) 1433 return(ENOMEM); 1434 1435 rrcopy->raidPtr = (void *) raidPtr; 1436 rrcopy->col = column; 1437 1438 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1439 rf_ReconstructInPlaceThread, 1440 rrcopy,"raid_reconip"); 1441 return(retcode); 1442 1443 case RAIDFRAME_GET_INFO: 1444 if (!raidPtr->valid) 1445 return (ENODEV); 1446 ucfgp = (RF_DeviceConfig_t **) data; 1447 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1448 (RF_DeviceConfig_t *)); 1449 if (d_cfg == NULL) 1450 return (ENOMEM); 1451 d_cfg->rows = 1; /* there is only 1 row now */ 1452 d_cfg->cols = raidPtr->numCol; 1453 d_cfg->ndevs = raidPtr->numCol; 1454 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1455 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1456 return (ENOMEM); 1457 } 1458 d_cfg->nspares = raidPtr->numSpare; 1459 if (d_cfg->nspares >= RF_MAX_DISKS) { 1460 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1461 return (ENOMEM); 1462 } 1463 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1464 d = 0; 1465 for (j = 0; j < d_cfg->cols; j++) { 1466 d_cfg->devs[d] = raidPtr->Disks[j]; 1467 d++; 1468 } 1469 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1470 d_cfg->spares[i] = raidPtr->Disks[j]; 1471 if (d_cfg->spares[i].status == rf_ds_rebuilding_spare) { 1472 /* XXX: raidctl(8) expects to see this as a used spare */ 1473 d_cfg->spares[i].status = rf_ds_used_spare; 1474 } 1475 } 1476 retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t)); 1477 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1478 1479 return (retcode); 1480 1481 case RAIDFRAME_CHECK_PARITY: 1482 *(int *) data = raidPtr->parity_good; 1483 return (0); 1484 1485 case RAIDFRAME_PARITYMAP_STATUS: 1486 if (rf_paritymap_ineligible(raidPtr)) 1487 return EINVAL; 1488 rf_paritymap_status(raidPtr->parity_map, 1489 (struct rf_pmstat *)data); 1490 return 0; 1491 1492 case RAIDFRAME_PARITYMAP_SET_PARAMS: 1493 if (rf_paritymap_ineligible(raidPtr)) 1494 return EINVAL; 1495 if (raidPtr->parity_map == NULL) 1496 return ENOENT; /* ??? */ 1497 if (0 != rf_paritymap_set_params(raidPtr->parity_map, 1498 (struct rf_pmparams *)data, 1)) 1499 return EINVAL; 1500 return 0; 1501 1502 case RAIDFRAME_PARITYMAP_GET_DISABLE: 1503 if (rf_paritymap_ineligible(raidPtr)) 1504 return EINVAL; 1505 *(int *) data = rf_paritymap_get_disable(raidPtr); 1506 return 0; 1507 1508 case RAIDFRAME_PARITYMAP_SET_DISABLE: 1509 if (rf_paritymap_ineligible(raidPtr)) 1510 return EINVAL; 1511 rf_paritymap_set_disable(raidPtr, *(int *)data); 1512 /* XXX should errors be passed up? */ 1513 return 0; 1514 1515 case RAIDFRAME_RESET_ACCTOTALS: 1516 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1517 return (0); 1518 1519 case RAIDFRAME_GET_ACCTOTALS: 1520 totals = (RF_AccTotals_t *) data; 1521 *totals = raidPtr->acc_totals; 1522 return (0); 1523 1524 case RAIDFRAME_KEEP_ACCTOTALS: 1525 raidPtr->keep_acc_totals = *(int *)data; 1526 return (0); 1527 1528 case RAIDFRAME_GET_SIZE: 1529 *(int *) data = raidPtr->totalSectors; 1530 return (0); 1531 1532 /* fail a disk & optionally start reconstruction */ 1533 case RAIDFRAME_FAIL_DISK: 1534 1535 if (raidPtr->Layout.map->faultsTolerated == 0) { 1536 /* Can't do this on a RAID 0!! */ 1537 return(EINVAL); 1538 } 1539 1540 rr = (struct rf_recon_req *) data; 1541 rr->row = 0; 1542 if (rr->col < 0 || rr->col >= raidPtr->numCol) 1543 return (EINVAL); 1544 1545 1546 rf_lock_mutex2(raidPtr->mutex); 1547 if (raidPtr->status == rf_rs_reconstructing) { 1548 /* you can't fail a disk while we're reconstructing! */ 1549 /* XXX wrong for RAID6 */ 1550 rf_unlock_mutex2(raidPtr->mutex); 1551 return (EINVAL); 1552 } 1553 if ((raidPtr->Disks[rr->col].status == 1554 rf_ds_optimal) && (raidPtr->numFailures > 0)) { 1555 /* some other component has failed. Let's not make 1556 things worse. XXX wrong for RAID6 */ 1557 rf_unlock_mutex2(raidPtr->mutex); 1558 return (EINVAL); 1559 } 1560 if (raidPtr->Disks[rr->col].status == rf_ds_spared) { 1561 /* Can't fail a spared disk! */ 1562 rf_unlock_mutex2(raidPtr->mutex); 1563 return (EINVAL); 1564 } 1565 rf_unlock_mutex2(raidPtr->mutex); 1566 1567 /* make a copy of the recon request so that we don't rely on 1568 * the user's buffer */ 1569 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1570 if (rrcopy == NULL) 1571 return(ENOMEM); 1572 memcpy(rrcopy, rr, sizeof(*rr)); 1573 rrcopy->raidPtr = (void *) raidPtr; 1574 1575 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1576 rf_ReconThread, 1577 rrcopy,"raid_recon"); 1578 return (0); 1579 1580 /* invoke a copyback operation after recon on whatever disk 1581 * needs it, if any */ 1582 case RAIDFRAME_COPYBACK: 1583 1584 if (raidPtr->Layout.map->faultsTolerated == 0) { 1585 /* This makes no sense on a RAID 0!! */ 1586 return(EINVAL); 1587 } 1588 1589 if (raidPtr->copyback_in_progress == 1) { 1590 /* Copyback is already in progress! */ 1591 return(EINVAL); 1592 } 1593 1594 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1595 rf_CopybackThread, 1596 raidPtr,"raid_copyback"); 1597 return (retcode); 1598 1599 /* return the percentage completion of reconstruction */ 1600 case RAIDFRAME_CHECK_RECON_STATUS: 1601 if (raidPtr->Layout.map->faultsTolerated == 0) { 1602 /* This makes no sense on a RAID 0, so tell the 1603 user it's done. */ 1604 *(int *) data = 100; 1605 return(0); 1606 } 1607 if (raidPtr->status != rf_rs_reconstructing) 1608 *(int *) data = 100; 1609 else { 1610 if (raidPtr->reconControl->numRUsTotal > 0) { 1611 *(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal); 1612 } else { 1613 *(int *) data = 0; 1614 } 1615 } 1616 return (0); 1617 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1618 progressInfoPtr = (RF_ProgressInfo_t **) data; 1619 if (raidPtr->status != rf_rs_reconstructing) { 1620 progressInfo.remaining = 0; 1621 progressInfo.completed = 100; 1622 progressInfo.total = 100; 1623 } else { 1624 progressInfo.total = 1625 raidPtr->reconControl->numRUsTotal; 1626 progressInfo.completed = 1627 raidPtr->reconControl->numRUsComplete; 1628 progressInfo.remaining = progressInfo.total - 1629 progressInfo.completed; 1630 } 1631 retcode = copyout(&progressInfo, *progressInfoPtr, 1632 sizeof(RF_ProgressInfo_t)); 1633 return (retcode); 1634 1635 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1636 if (raidPtr->Layout.map->faultsTolerated == 0) { 1637 /* This makes no sense on a RAID 0, so tell the 1638 user it's done. */ 1639 *(int *) data = 100; 1640 return(0); 1641 } 1642 if (raidPtr->parity_rewrite_in_progress == 1) { 1643 *(int *) data = 100 * 1644 raidPtr->parity_rewrite_stripes_done / 1645 raidPtr->Layout.numStripe; 1646 } else { 1647 *(int *) data = 100; 1648 } 1649 return (0); 1650 1651 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1652 progressInfoPtr = (RF_ProgressInfo_t **) data; 1653 if (raidPtr->parity_rewrite_in_progress == 1) { 1654 progressInfo.total = raidPtr->Layout.numStripe; 1655 progressInfo.completed = 1656 raidPtr->parity_rewrite_stripes_done; 1657 progressInfo.remaining = progressInfo.total - 1658 progressInfo.completed; 1659 } else { 1660 progressInfo.remaining = 0; 1661 progressInfo.completed = 100; 1662 progressInfo.total = 100; 1663 } 1664 retcode = copyout(&progressInfo, *progressInfoPtr, 1665 sizeof(RF_ProgressInfo_t)); 1666 return (retcode); 1667 1668 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1669 if (raidPtr->Layout.map->faultsTolerated == 0) { 1670 /* This makes no sense on a RAID 0 */ 1671 *(int *) data = 100; 1672 return(0); 1673 } 1674 if (raidPtr->copyback_in_progress == 1) { 1675 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1676 raidPtr->Layout.numStripe; 1677 } else { 1678 *(int *) data = 100; 1679 } 1680 return (0); 1681 1682 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1683 progressInfoPtr = (RF_ProgressInfo_t **) data; 1684 if (raidPtr->copyback_in_progress == 1) { 1685 progressInfo.total = raidPtr->Layout.numStripe; 1686 progressInfo.completed = 1687 raidPtr->copyback_stripes_done; 1688 progressInfo.remaining = progressInfo.total - 1689 progressInfo.completed; 1690 } else { 1691 progressInfo.remaining = 0; 1692 progressInfo.completed = 100; 1693 progressInfo.total = 100; 1694 } 1695 retcode = copyout(&progressInfo, *progressInfoPtr, 1696 sizeof(RF_ProgressInfo_t)); 1697 return (retcode); 1698 1699 /* the sparetable daemon calls this to wait for the kernel to 1700 * need a spare table. this ioctl does not return until a 1701 * spare table is needed. XXX -- calling mpsleep here in the 1702 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1703 * -- I should either compute the spare table in the kernel, 1704 * or have a different -- XXX XXX -- interface (a different 1705 * character device) for delivering the table -- XXX */ 1706 #if 0 1707 case RAIDFRAME_SPARET_WAIT: 1708 rf_lock_mutex2(rf_sparet_wait_mutex); 1709 while (!rf_sparet_wait_queue) 1710 rf_wait_cond2(rf_sparet_wait_cv, rf_sparet_wait_mutex); 1711 waitreq = rf_sparet_wait_queue; 1712 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1713 rf_unlock_mutex2(rf_sparet_wait_mutex); 1714 1715 /* structure assignment */ 1716 *((RF_SparetWait_t *) data) = *waitreq; 1717 1718 RF_Free(waitreq, sizeof(*waitreq)); 1719 return (0); 1720 1721 /* wakes up a process waiting on SPARET_WAIT and puts an error 1722 * code in it that will cause the dameon to exit */ 1723 case RAIDFRAME_ABORT_SPARET_WAIT: 1724 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1725 waitreq->fcol = -1; 1726 rf_lock_mutex2(rf_sparet_wait_mutex); 1727 waitreq->next = rf_sparet_wait_queue; 1728 rf_sparet_wait_queue = waitreq; 1729 rf_broadcast_conf2(rf_sparet_wait_cv); 1730 rf_unlock_mutex2(rf_sparet_wait_mutex); 1731 return (0); 1732 1733 /* used by the spare table daemon to deliver a spare table 1734 * into the kernel */ 1735 case RAIDFRAME_SEND_SPARET: 1736 1737 /* install the spare table */ 1738 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1739 1740 /* respond to the requestor. the return status of the spare 1741 * table installation is passed in the "fcol" field */ 1742 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1743 waitreq->fcol = retcode; 1744 rf_lock_mutex2(rf_sparet_wait_mutex); 1745 waitreq->next = rf_sparet_resp_queue; 1746 rf_sparet_resp_queue = waitreq; 1747 rf_broadcast_cond2(rf_sparet_resp_cv); 1748 rf_unlock_mutex2(rf_sparet_wait_mutex); 1749 1750 return (retcode); 1751 #endif 1752 1753 default: 1754 break; /* fall through to the os-specific code below */ 1755 1756 } 1757 1758 if (!raidPtr->valid) 1759 return (EINVAL); 1760 1761 /* 1762 * Add support for "regular" device ioctls here. 1763 */ 1764 1765 error = disk_ioctl(&rs->sc_dkdev, cmd, data, flag, l); 1766 if (error != EPASSTHROUGH) 1767 return (error); 1768 1769 switch (cmd) { 1770 case DIOCGDINFO: 1771 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1772 break; 1773 #ifdef __HAVE_OLD_DISKLABEL 1774 case ODIOCGDINFO: 1775 newlabel = *(rs->sc_dkdev.dk_label); 1776 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1777 return ENOTTY; 1778 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1779 break; 1780 #endif 1781 1782 case DIOCGPART: 1783 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1784 ((struct partinfo *) data)->part = 1785 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1786 break; 1787 1788 case DIOCWDINFO: 1789 case DIOCSDINFO: 1790 #ifdef __HAVE_OLD_DISKLABEL 1791 case ODIOCWDINFO: 1792 case ODIOCSDINFO: 1793 #endif 1794 { 1795 struct disklabel *lp; 1796 #ifdef __HAVE_OLD_DISKLABEL 1797 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1798 memset(&newlabel, 0, sizeof newlabel); 1799 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1800 lp = &newlabel; 1801 } else 1802 #endif 1803 lp = (struct disklabel *)data; 1804 1805 if ((error = raidlock(rs)) != 0) 1806 return (error); 1807 1808 rs->sc_flags |= RAIDF_LABELLING; 1809 1810 error = setdisklabel(rs->sc_dkdev.dk_label, 1811 lp, 0, rs->sc_dkdev.dk_cpulabel); 1812 if (error == 0) { 1813 if (cmd == DIOCWDINFO 1814 #ifdef __HAVE_OLD_DISKLABEL 1815 || cmd == ODIOCWDINFO 1816 #endif 1817 ) 1818 error = writedisklabel(RAIDLABELDEV(dev), 1819 raidstrategy, rs->sc_dkdev.dk_label, 1820 rs->sc_dkdev.dk_cpulabel); 1821 } 1822 rs->sc_flags &= ~RAIDF_LABELLING; 1823 1824 raidunlock(rs); 1825 1826 if (error) 1827 return (error); 1828 break; 1829 } 1830 1831 case DIOCWLABEL: 1832 if (*(int *) data != 0) 1833 rs->sc_flags |= RAIDF_WLABEL; 1834 else 1835 rs->sc_flags &= ~RAIDF_WLABEL; 1836 break; 1837 1838 case DIOCGDEFLABEL: 1839 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1840 break; 1841 1842 #ifdef __HAVE_OLD_DISKLABEL 1843 case ODIOCGDEFLABEL: 1844 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1845 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1846 return ENOTTY; 1847 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1848 break; 1849 #endif 1850 1851 case DIOCAWEDGE: 1852 case DIOCDWEDGE: 1853 dkw = (void *)data; 1854 1855 /* If the ioctl happens here, the parent is us. */ 1856 (void)strcpy(dkw->dkw_parent, rs->sc_xname); 1857 return cmd == DIOCAWEDGE ? dkwedge_add(dkw) : dkwedge_del(dkw); 1858 1859 case DIOCLWEDGES: 1860 return dkwedge_list(&rs->sc_dkdev, 1861 (struct dkwedge_list *)data, l); 1862 case DIOCCACHESYNC: 1863 return rf_sync_component_caches(raidPtr); 1864 1865 case DIOCGSTRATEGY: 1866 { 1867 struct disk_strategy *dks = (void *)data; 1868 1869 s = splbio(); 1870 strlcpy(dks->dks_name, bufq_getstrategyname(rs->buf_queue), 1871 sizeof(dks->dks_name)); 1872 splx(s); 1873 dks->dks_paramlen = 0; 1874 1875 return 0; 1876 } 1877 1878 case DIOCSSTRATEGY: 1879 { 1880 struct disk_strategy *dks = (void *)data; 1881 struct bufq_state *new; 1882 struct bufq_state *old; 1883 1884 if (dks->dks_param != NULL) { 1885 return EINVAL; 1886 } 1887 dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */ 1888 error = bufq_alloc(&new, dks->dks_name, 1889 BUFQ_EXACT|BUFQ_SORT_RAWBLOCK); 1890 if (error) { 1891 return error; 1892 } 1893 s = splbio(); 1894 old = rs->buf_queue; 1895 bufq_move(new, old); 1896 rs->buf_queue = new; 1897 splx(s); 1898 bufq_free(old); 1899 1900 return 0; 1901 } 1902 1903 default: 1904 retcode = ENOTTY; 1905 } 1906 return (retcode); 1907 1908 } 1909 1910 1911 /* raidinit -- complete the rest of the initialization for the 1912 RAIDframe device. */ 1913 1914 1915 static void 1916 raidinit(RF_Raid_t *raidPtr) 1917 { 1918 cfdata_t cf; 1919 struct raid_softc *rs; 1920 int unit; 1921 1922 unit = raidPtr->raidid; 1923 1924 rs = &raid_softc[unit]; 1925 1926 /* XXX should check return code first... */ 1927 rs->sc_flags |= RAIDF_INITED; 1928 1929 /* XXX doesn't check bounds. */ 1930 snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit); 1931 1932 /* attach the pseudo device */ 1933 cf = malloc(sizeof(*cf), M_RAIDFRAME, M_WAITOK); 1934 cf->cf_name = raid_cd.cd_name; 1935 cf->cf_atname = raid_cd.cd_name; 1936 cf->cf_unit = unit; 1937 cf->cf_fstate = FSTATE_STAR; 1938 1939 rs->sc_dev = config_attach_pseudo(cf); 1940 1941 if (rs->sc_dev == NULL) { 1942 printf("raid%d: config_attach_pseudo failed\n", 1943 raidPtr->raidid); 1944 rs->sc_flags &= ~RAIDF_INITED; 1945 free(cf, M_RAIDFRAME); 1946 return; 1947 } 1948 1949 /* disk_attach actually creates space for the CPU disklabel, among 1950 * other things, so it's critical to call this *BEFORE* we try putzing 1951 * with disklabels. */ 1952 1953 disk_init(&rs->sc_dkdev, rs->sc_xname, &rf_dkdriver); 1954 disk_attach(&rs->sc_dkdev); 1955 disk_blocksize(&rs->sc_dkdev, raidPtr->bytesPerSector); 1956 1957 /* XXX There may be a weird interaction here between this, and 1958 * protectedSectors, as used in RAIDframe. */ 1959 1960 rs->sc_size = raidPtr->totalSectors; 1961 1962 dkwedge_discover(&rs->sc_dkdev); 1963 1964 rf_set_properties(rs, raidPtr); 1965 1966 } 1967 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 1968 /* wake up the daemon & tell it to get us a spare table 1969 * XXX 1970 * the entries in the queues should be tagged with the raidPtr 1971 * so that in the extremely rare case that two recons happen at once, 1972 * we know for which device were requesting a spare table 1973 * XXX 1974 * 1975 * XXX This code is not currently used. GO 1976 */ 1977 int 1978 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req) 1979 { 1980 int retcode; 1981 1982 rf_lock_mutex2(rf_sparet_wait_mutex); 1983 req->next = rf_sparet_wait_queue; 1984 rf_sparet_wait_queue = req; 1985 rf_broadcast_cond2(rf_sparet_wait_cv); 1986 1987 /* mpsleep unlocks the mutex */ 1988 while (!rf_sparet_resp_queue) { 1989 rf_wait_cond2(rf_sparet_resp_cv, rf_sparet_wait_mutex); 1990 } 1991 req = rf_sparet_resp_queue; 1992 rf_sparet_resp_queue = req->next; 1993 rf_unlock_mutex2(rf_sparet_wait_mutex); 1994 1995 retcode = req->fcol; 1996 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1997 * alloc'd */ 1998 return (retcode); 1999 } 2000 #endif 2001 2002 /* a wrapper around rf_DoAccess that extracts appropriate info from the 2003 * bp & passes it down. 2004 * any calls originating in the kernel must use non-blocking I/O 2005 * do some extra sanity checking to return "appropriate" error values for 2006 * certain conditions (to make some standard utilities work) 2007 * 2008 * Formerly known as: rf_DoAccessKernel 2009 */ 2010 void 2011 raidstart(RF_Raid_t *raidPtr) 2012 { 2013 RF_SectorCount_t num_blocks, pb, sum; 2014 RF_RaidAddr_t raid_addr; 2015 struct partition *pp; 2016 daddr_t blocknum; 2017 int unit; 2018 struct raid_softc *rs; 2019 int do_async; 2020 struct buf *bp; 2021 int rc; 2022 2023 unit = raidPtr->raidid; 2024 rs = &raid_softc[unit]; 2025 2026 /* quick check to see if anything has died recently */ 2027 rf_lock_mutex2(raidPtr->mutex); 2028 if (raidPtr->numNewFailures > 0) { 2029 rf_unlock_mutex2(raidPtr->mutex); 2030 rf_update_component_labels(raidPtr, 2031 RF_NORMAL_COMPONENT_UPDATE); 2032 rf_lock_mutex2(raidPtr->mutex); 2033 raidPtr->numNewFailures--; 2034 } 2035 2036 /* Check to see if we're at the limit... */ 2037 while (raidPtr->openings > 0) { 2038 rf_unlock_mutex2(raidPtr->mutex); 2039 2040 /* get the next item, if any, from the queue */ 2041 if ((bp = bufq_get(rs->buf_queue)) == NULL) { 2042 /* nothing more to do */ 2043 return; 2044 } 2045 2046 /* Ok, for the bp we have here, bp->b_blkno is relative to the 2047 * partition.. Need to make it absolute to the underlying 2048 * device.. */ 2049 2050 blocknum = bp->b_blkno << DEV_BSHIFT >> raidPtr->logBytesPerSector; 2051 if (DISKPART(bp->b_dev) != RAW_PART) { 2052 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 2053 blocknum += pp->p_offset; 2054 } 2055 2056 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 2057 (int) blocknum)); 2058 2059 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 2060 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 2061 2062 /* *THIS* is where we adjust what block we're going to... 2063 * but DO NOT TOUCH bp->b_blkno!!! */ 2064 raid_addr = blocknum; 2065 2066 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 2067 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 2068 sum = raid_addr + num_blocks + pb; 2069 if (1 || rf_debugKernelAccess) { 2070 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 2071 (int) raid_addr, (int) sum, (int) num_blocks, 2072 (int) pb, (int) bp->b_resid)); 2073 } 2074 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 2075 || (sum < num_blocks) || (sum < pb)) { 2076 bp->b_error = ENOSPC; 2077 bp->b_resid = bp->b_bcount; 2078 biodone(bp); 2079 rf_lock_mutex2(raidPtr->mutex); 2080 continue; 2081 } 2082 /* 2083 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 2084 */ 2085 2086 if (bp->b_bcount & raidPtr->sectorMask) { 2087 bp->b_error = EINVAL; 2088 bp->b_resid = bp->b_bcount; 2089 biodone(bp); 2090 rf_lock_mutex2(raidPtr->mutex); 2091 continue; 2092 2093 } 2094 db1_printf(("Calling DoAccess..\n")); 2095 2096 2097 rf_lock_mutex2(raidPtr->mutex); 2098 raidPtr->openings--; 2099 rf_unlock_mutex2(raidPtr->mutex); 2100 2101 /* 2102 * Everything is async. 2103 */ 2104 do_async = 1; 2105 2106 disk_busy(&rs->sc_dkdev); 2107 2108 /* XXX we're still at splbio() here... do we *really* 2109 need to be? */ 2110 2111 /* don't ever condition on bp->b_flags & B_WRITE. 2112 * always condition on B_READ instead */ 2113 2114 rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 2115 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 2116 do_async, raid_addr, num_blocks, 2117 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 2118 2119 if (rc) { 2120 bp->b_error = rc; 2121 bp->b_resid = bp->b_bcount; 2122 biodone(bp); 2123 /* continue loop */ 2124 } 2125 2126 rf_lock_mutex2(raidPtr->mutex); 2127 } 2128 rf_unlock_mutex2(raidPtr->mutex); 2129 } 2130 2131 2132 2133 2134 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 2135 2136 int 2137 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req) 2138 { 2139 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 2140 struct buf *bp; 2141 2142 req->queue = queue; 2143 bp = req->bp; 2144 2145 switch (req->type) { 2146 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 2147 /* XXX need to do something extra here.. */ 2148 /* I'm leaving this in, as I've never actually seen it used, 2149 * and I'd like folks to report it... GO */ 2150 printf(("WAKEUP CALLED\n")); 2151 queue->numOutstanding++; 2152 2153 bp->b_flags = 0; 2154 bp->b_private = req; 2155 2156 KernelWakeupFunc(bp); 2157 break; 2158 2159 case RF_IO_TYPE_READ: 2160 case RF_IO_TYPE_WRITE: 2161 #if RF_ACC_TRACE > 0 2162 if (req->tracerec) { 2163 RF_ETIMER_START(req->tracerec->timer); 2164 } 2165 #endif 2166 InitBP(bp, queue->rf_cinfo->ci_vp, 2167 op, queue->rf_cinfo->ci_dev, 2168 req->sectorOffset, req->numSector, 2169 req->buf, KernelWakeupFunc, (void *) req, 2170 queue->raidPtr->logBytesPerSector, req->b_proc); 2171 2172 if (rf_debugKernelAccess) { 2173 db1_printf(("dispatch: bp->b_blkno = %ld\n", 2174 (long) bp->b_blkno)); 2175 } 2176 queue->numOutstanding++; 2177 queue->last_deq_sector = req->sectorOffset; 2178 /* acc wouldn't have been let in if there were any pending 2179 * reqs at any other priority */ 2180 queue->curPriority = req->priority; 2181 2182 db1_printf(("Going for %c to unit %d col %d\n", 2183 req->type, queue->raidPtr->raidid, 2184 queue->col)); 2185 db1_printf(("sector %d count %d (%d bytes) %d\n", 2186 (int) req->sectorOffset, (int) req->numSector, 2187 (int) (req->numSector << 2188 queue->raidPtr->logBytesPerSector), 2189 (int) queue->raidPtr->logBytesPerSector)); 2190 2191 /* 2192 * XXX: drop lock here since this can block at 2193 * least with backing SCSI devices. Retake it 2194 * to minimize fuss with calling interfaces. 2195 */ 2196 2197 RF_UNLOCK_QUEUE_MUTEX(queue, "unusedparam"); 2198 bdev_strategy(bp); 2199 RF_LOCK_QUEUE_MUTEX(queue, "unusedparam"); 2200 break; 2201 2202 default: 2203 panic("bad req->type in rf_DispatchKernelIO"); 2204 } 2205 db1_printf(("Exiting from DispatchKernelIO\n")); 2206 2207 return (0); 2208 } 2209 /* this is the callback function associated with a I/O invoked from 2210 kernel code. 2211 */ 2212 static void 2213 KernelWakeupFunc(struct buf *bp) 2214 { 2215 RF_DiskQueueData_t *req = NULL; 2216 RF_DiskQueue_t *queue; 2217 2218 db1_printf(("recovering the request queue:\n")); 2219 2220 req = bp->b_private; 2221 2222 queue = (RF_DiskQueue_t *) req->queue; 2223 2224 rf_lock_mutex2(queue->raidPtr->iodone_lock); 2225 2226 #if RF_ACC_TRACE > 0 2227 if (req->tracerec) { 2228 RF_ETIMER_STOP(req->tracerec->timer); 2229 RF_ETIMER_EVAL(req->tracerec->timer); 2230 rf_lock_mutex2(rf_tracing_mutex); 2231 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2232 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2233 req->tracerec->num_phys_ios++; 2234 rf_unlock_mutex2(rf_tracing_mutex); 2235 } 2236 #endif 2237 2238 /* XXX Ok, let's get aggressive... If b_error is set, let's go 2239 * ballistic, and mark the component as hosed... */ 2240 2241 if (bp->b_error != 0) { 2242 /* Mark the disk as dead */ 2243 /* but only mark it once... */ 2244 /* and only if it wouldn't leave this RAID set 2245 completely broken */ 2246 if (((queue->raidPtr->Disks[queue->col].status == 2247 rf_ds_optimal) || 2248 (queue->raidPtr->Disks[queue->col].status == 2249 rf_ds_used_spare)) && 2250 (queue->raidPtr->numFailures < 2251 queue->raidPtr->Layout.map->faultsTolerated)) { 2252 printf("raid%d: IO Error. Marking %s as failed.\n", 2253 queue->raidPtr->raidid, 2254 queue->raidPtr->Disks[queue->col].devname); 2255 queue->raidPtr->Disks[queue->col].status = 2256 rf_ds_failed; 2257 queue->raidPtr->status = rf_rs_degraded; 2258 queue->raidPtr->numFailures++; 2259 queue->raidPtr->numNewFailures++; 2260 } else { /* Disk is already dead... */ 2261 /* printf("Disk already marked as dead!\n"); */ 2262 } 2263 2264 } 2265 2266 /* Fill in the error value */ 2267 req->error = bp->b_error; 2268 2269 /* Drop this one on the "finished" queue... */ 2270 TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries); 2271 2272 /* Let the raidio thread know there is work to be done. */ 2273 rf_signal_cond2(queue->raidPtr->iodone_cv); 2274 2275 rf_unlock_mutex2(queue->raidPtr->iodone_lock); 2276 } 2277 2278 2279 /* 2280 * initialize a buf structure for doing an I/O in the kernel. 2281 */ 2282 static void 2283 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev, 2284 RF_SectorNum_t startSect, RF_SectorCount_t numSect, void *bf, 2285 void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector, 2286 struct proc *b_proc) 2287 { 2288 /* bp->b_flags = B_PHYS | rw_flag; */ 2289 bp->b_flags = rw_flag; /* XXX need B_PHYS here too??? */ 2290 bp->b_oflags = 0; 2291 bp->b_cflags = 0; 2292 bp->b_bcount = numSect << logBytesPerSector; 2293 bp->b_bufsize = bp->b_bcount; 2294 bp->b_error = 0; 2295 bp->b_dev = dev; 2296 bp->b_data = bf; 2297 bp->b_blkno = startSect << logBytesPerSector >> DEV_BSHIFT; 2298 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 2299 if (bp->b_bcount == 0) { 2300 panic("bp->b_bcount is zero in InitBP!!"); 2301 } 2302 bp->b_proc = b_proc; 2303 bp->b_iodone = cbFunc; 2304 bp->b_private = cbArg; 2305 } 2306 2307 static void 2308 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs, 2309 struct disklabel *lp) 2310 { 2311 memset(lp, 0, sizeof(*lp)); 2312 2313 /* fabricate a label... */ 2314 lp->d_secperunit = raidPtr->totalSectors; 2315 lp->d_secsize = raidPtr->bytesPerSector; 2316 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 2317 lp->d_ntracks = 4 * raidPtr->numCol; 2318 lp->d_ncylinders = raidPtr->totalSectors / 2319 (lp->d_nsectors * lp->d_ntracks); 2320 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 2321 2322 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 2323 lp->d_type = DTYPE_RAID; 2324 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 2325 lp->d_rpm = 3600; 2326 lp->d_interleave = 1; 2327 lp->d_flags = 0; 2328 2329 lp->d_partitions[RAW_PART].p_offset = 0; 2330 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 2331 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 2332 lp->d_npartitions = RAW_PART + 1; 2333 2334 lp->d_magic = DISKMAGIC; 2335 lp->d_magic2 = DISKMAGIC; 2336 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 2337 2338 } 2339 /* 2340 * Read the disklabel from the raid device. If one is not present, fake one 2341 * up. 2342 */ 2343 static void 2344 raidgetdisklabel(dev_t dev) 2345 { 2346 int unit = raidunit(dev); 2347 struct raid_softc *rs = &raid_softc[unit]; 2348 const char *errstring; 2349 struct disklabel *lp = rs->sc_dkdev.dk_label; 2350 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 2351 RF_Raid_t *raidPtr; 2352 2353 db1_printf(("Getting the disklabel...\n")); 2354 2355 memset(clp, 0, sizeof(*clp)); 2356 2357 raidPtr = raidPtrs[unit]; 2358 2359 raidgetdefaultlabel(raidPtr, rs, lp); 2360 2361 /* 2362 * Call the generic disklabel extraction routine. 2363 */ 2364 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 2365 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2366 if (errstring) 2367 raidmakedisklabel(rs); 2368 else { 2369 int i; 2370 struct partition *pp; 2371 2372 /* 2373 * Sanity check whether the found disklabel is valid. 2374 * 2375 * This is necessary since total size of the raid device 2376 * may vary when an interleave is changed even though exactly 2377 * same components are used, and old disklabel may used 2378 * if that is found. 2379 */ 2380 if (lp->d_secperunit != rs->sc_size) 2381 printf("raid%d: WARNING: %s: " 2382 "total sector size in disklabel (%" PRIu32 ") != " 2383 "the size of raid (%" PRIu64 ")\n", unit, rs->sc_xname, 2384 lp->d_secperunit, rs->sc_size); 2385 for (i = 0; i < lp->d_npartitions; i++) { 2386 pp = &lp->d_partitions[i]; 2387 if (pp->p_offset + pp->p_size > rs->sc_size) 2388 printf("raid%d: WARNING: %s: end of partition `%c' " 2389 "exceeds the size of raid (%" PRIu64 ")\n", 2390 unit, rs->sc_xname, 'a' + i, rs->sc_size); 2391 } 2392 } 2393 2394 } 2395 /* 2396 * Take care of things one might want to take care of in the event 2397 * that a disklabel isn't present. 2398 */ 2399 static void 2400 raidmakedisklabel(struct raid_softc *rs) 2401 { 2402 struct disklabel *lp = rs->sc_dkdev.dk_label; 2403 db1_printf(("Making a label..\n")); 2404 2405 /* 2406 * For historical reasons, if there's no disklabel present 2407 * the raw partition must be marked FS_BSDFFS. 2408 */ 2409 2410 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2411 2412 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2413 2414 lp->d_checksum = dkcksum(lp); 2415 } 2416 /* 2417 * Wait interruptibly for an exclusive lock. 2418 * 2419 * XXX 2420 * Several drivers do this; it should be abstracted and made MP-safe. 2421 * (Hmm... where have we seen this warning before :-> GO ) 2422 */ 2423 static int 2424 raidlock(struct raid_softc *rs) 2425 { 2426 int error; 2427 2428 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2429 rs->sc_flags |= RAIDF_WANTED; 2430 if ((error = 2431 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2432 return (error); 2433 } 2434 rs->sc_flags |= RAIDF_LOCKED; 2435 return (0); 2436 } 2437 /* 2438 * Unlock and wake up any waiters. 2439 */ 2440 static void 2441 raidunlock(struct raid_softc *rs) 2442 { 2443 2444 rs->sc_flags &= ~RAIDF_LOCKED; 2445 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2446 rs->sc_flags &= ~RAIDF_WANTED; 2447 wakeup(rs); 2448 } 2449 } 2450 2451 2452 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2453 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2454 #define RF_PARITY_MAP_SIZE RF_PARITYMAP_NBYTE 2455 2456 static daddr_t 2457 rf_component_info_offset(void) 2458 { 2459 2460 return RF_COMPONENT_INFO_OFFSET; 2461 } 2462 2463 static daddr_t 2464 rf_component_info_size(unsigned secsize) 2465 { 2466 daddr_t info_size; 2467 2468 KASSERT(secsize); 2469 if (secsize > RF_COMPONENT_INFO_SIZE) 2470 info_size = secsize; 2471 else 2472 info_size = RF_COMPONENT_INFO_SIZE; 2473 2474 return info_size; 2475 } 2476 2477 static daddr_t 2478 rf_parity_map_offset(RF_Raid_t *raidPtr) 2479 { 2480 daddr_t map_offset; 2481 2482 KASSERT(raidPtr->bytesPerSector); 2483 if (raidPtr->bytesPerSector > RF_COMPONENT_INFO_SIZE) 2484 map_offset = raidPtr->bytesPerSector; 2485 else 2486 map_offset = RF_COMPONENT_INFO_SIZE; 2487 map_offset += rf_component_info_offset(); 2488 2489 return map_offset; 2490 } 2491 2492 static daddr_t 2493 rf_parity_map_size(RF_Raid_t *raidPtr) 2494 { 2495 daddr_t map_size; 2496 2497 if (raidPtr->bytesPerSector > RF_PARITY_MAP_SIZE) 2498 map_size = raidPtr->bytesPerSector; 2499 else 2500 map_size = RF_PARITY_MAP_SIZE; 2501 2502 return map_size; 2503 } 2504 2505 int 2506 raidmarkclean(RF_Raid_t *raidPtr, RF_RowCol_t col) 2507 { 2508 RF_ComponentLabel_t *clabel; 2509 2510 clabel = raidget_component_label(raidPtr, col); 2511 clabel->clean = RF_RAID_CLEAN; 2512 raidflush_component_label(raidPtr, col); 2513 return(0); 2514 } 2515 2516 2517 int 2518 raidmarkdirty(RF_Raid_t *raidPtr, RF_RowCol_t col) 2519 { 2520 RF_ComponentLabel_t *clabel; 2521 2522 clabel = raidget_component_label(raidPtr, col); 2523 clabel->clean = RF_RAID_DIRTY; 2524 raidflush_component_label(raidPtr, col); 2525 return(0); 2526 } 2527 2528 int 2529 raidfetch_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2530 { 2531 KASSERT(raidPtr->bytesPerSector); 2532 return raidread_component_label(raidPtr->bytesPerSector, 2533 raidPtr->Disks[col].dev, 2534 raidPtr->raid_cinfo[col].ci_vp, 2535 &raidPtr->raid_cinfo[col].ci_label); 2536 } 2537 2538 RF_ComponentLabel_t * 2539 raidget_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2540 { 2541 return &raidPtr->raid_cinfo[col].ci_label; 2542 } 2543 2544 int 2545 raidflush_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2546 { 2547 RF_ComponentLabel_t *label; 2548 2549 label = &raidPtr->raid_cinfo[col].ci_label; 2550 label->mod_counter = raidPtr->mod_counter; 2551 #ifndef RF_NO_PARITY_MAP 2552 label->parity_map_modcount = label->mod_counter; 2553 #endif 2554 return raidwrite_component_label(raidPtr->bytesPerSector, 2555 raidPtr->Disks[col].dev, 2556 raidPtr->raid_cinfo[col].ci_vp, label); 2557 } 2558 2559 2560 static int 2561 raidread_component_label(unsigned secsize, dev_t dev, struct vnode *b_vp, 2562 RF_ComponentLabel_t *clabel) 2563 { 2564 return raidread_component_area(dev, b_vp, clabel, 2565 sizeof(RF_ComponentLabel_t), 2566 rf_component_info_offset(), 2567 rf_component_info_size(secsize)); 2568 } 2569 2570 /* ARGSUSED */ 2571 static int 2572 raidread_component_area(dev_t dev, struct vnode *b_vp, void *data, 2573 size_t msize, daddr_t offset, daddr_t dsize) 2574 { 2575 struct buf *bp; 2576 const struct bdevsw *bdev; 2577 int error; 2578 2579 /* XXX should probably ensure that we don't try to do this if 2580 someone has changed rf_protected_sectors. */ 2581 2582 if (b_vp == NULL) { 2583 /* For whatever reason, this component is not valid. 2584 Don't try to read a component label from it. */ 2585 return(EINVAL); 2586 } 2587 2588 /* get a block of the appropriate size... */ 2589 bp = geteblk((int)dsize); 2590 bp->b_dev = dev; 2591 2592 /* get our ducks in a row for the read */ 2593 bp->b_blkno = offset / DEV_BSIZE; 2594 bp->b_bcount = dsize; 2595 bp->b_flags |= B_READ; 2596 bp->b_resid = dsize; 2597 2598 bdev = bdevsw_lookup(bp->b_dev); 2599 if (bdev == NULL) 2600 return (ENXIO); 2601 (*bdev->d_strategy)(bp); 2602 2603 error = biowait(bp); 2604 2605 if (!error) { 2606 memcpy(data, bp->b_data, msize); 2607 } 2608 2609 brelse(bp, 0); 2610 return(error); 2611 } 2612 2613 2614 static int 2615 raidwrite_component_label(unsigned secsize, dev_t dev, struct vnode *b_vp, 2616 RF_ComponentLabel_t *clabel) 2617 { 2618 return raidwrite_component_area(dev, b_vp, clabel, 2619 sizeof(RF_ComponentLabel_t), 2620 rf_component_info_offset(), 2621 rf_component_info_size(secsize), 0); 2622 } 2623 2624 /* ARGSUSED */ 2625 static int 2626 raidwrite_component_area(dev_t dev, struct vnode *b_vp, void *data, 2627 size_t msize, daddr_t offset, daddr_t dsize, int asyncp) 2628 { 2629 struct buf *bp; 2630 const struct bdevsw *bdev; 2631 int error; 2632 2633 /* get a block of the appropriate size... */ 2634 bp = geteblk((int)dsize); 2635 bp->b_dev = dev; 2636 2637 /* get our ducks in a row for the write */ 2638 bp->b_blkno = offset / DEV_BSIZE; 2639 bp->b_bcount = dsize; 2640 bp->b_flags |= B_WRITE | (asyncp ? B_ASYNC : 0); 2641 bp->b_resid = dsize; 2642 2643 memset(bp->b_data, 0, dsize); 2644 memcpy(bp->b_data, data, msize); 2645 2646 bdev = bdevsw_lookup(bp->b_dev); 2647 if (bdev == NULL) 2648 return (ENXIO); 2649 (*bdev->d_strategy)(bp); 2650 if (asyncp) 2651 return 0; 2652 error = biowait(bp); 2653 brelse(bp, 0); 2654 if (error) { 2655 #if 1 2656 printf("Failed to write RAID component info!\n"); 2657 #endif 2658 } 2659 2660 return(error); 2661 } 2662 2663 void 2664 rf_paritymap_kern_write(RF_Raid_t *raidPtr, struct rf_paritymap_ondisk *map) 2665 { 2666 int c; 2667 2668 for (c = 0; c < raidPtr->numCol; c++) { 2669 /* Skip dead disks. */ 2670 if (RF_DEAD_DISK(raidPtr->Disks[c].status)) 2671 continue; 2672 /* XXXjld: what if an error occurs here? */ 2673 raidwrite_component_area(raidPtr->Disks[c].dev, 2674 raidPtr->raid_cinfo[c].ci_vp, map, 2675 RF_PARITYMAP_NBYTE, 2676 rf_parity_map_offset(raidPtr), 2677 rf_parity_map_size(raidPtr), 0); 2678 } 2679 } 2680 2681 void 2682 rf_paritymap_kern_read(RF_Raid_t *raidPtr, struct rf_paritymap_ondisk *map) 2683 { 2684 struct rf_paritymap_ondisk tmp; 2685 int c,first; 2686 2687 first=1; 2688 for (c = 0; c < raidPtr->numCol; c++) { 2689 /* Skip dead disks. */ 2690 if (RF_DEAD_DISK(raidPtr->Disks[c].status)) 2691 continue; 2692 raidread_component_area(raidPtr->Disks[c].dev, 2693 raidPtr->raid_cinfo[c].ci_vp, &tmp, 2694 RF_PARITYMAP_NBYTE, 2695 rf_parity_map_offset(raidPtr), 2696 rf_parity_map_size(raidPtr)); 2697 if (first) { 2698 memcpy(map, &tmp, sizeof(*map)); 2699 first = 0; 2700 } else { 2701 rf_paritymap_merge(map, &tmp); 2702 } 2703 } 2704 } 2705 2706 void 2707 rf_markalldirty(RF_Raid_t *raidPtr) 2708 { 2709 RF_ComponentLabel_t *clabel; 2710 int sparecol; 2711 int c; 2712 int j; 2713 int scol = -1; 2714 2715 raidPtr->mod_counter++; 2716 for (c = 0; c < raidPtr->numCol; c++) { 2717 /* we don't want to touch (at all) a disk that has 2718 failed */ 2719 if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) { 2720 clabel = raidget_component_label(raidPtr, c); 2721 if (clabel->status == rf_ds_spared) { 2722 /* XXX do something special... 2723 but whatever you do, don't 2724 try to access it!! */ 2725 } else { 2726 raidmarkdirty(raidPtr, c); 2727 } 2728 } 2729 } 2730 2731 for( c = 0; c < raidPtr->numSpare ; c++) { 2732 sparecol = raidPtr->numCol + c; 2733 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2734 /* 2735 2736 we claim this disk is "optimal" if it's 2737 rf_ds_used_spare, as that means it should be 2738 directly substitutable for the disk it replaced. 2739 We note that too... 2740 2741 */ 2742 2743 for(j=0;j<raidPtr->numCol;j++) { 2744 if (raidPtr->Disks[j].spareCol == sparecol) { 2745 scol = j; 2746 break; 2747 } 2748 } 2749 2750 clabel = raidget_component_label(raidPtr, sparecol); 2751 /* make sure status is noted */ 2752 2753 raid_init_component_label(raidPtr, clabel); 2754 2755 clabel->row = 0; 2756 clabel->column = scol; 2757 /* Note: we *don't* change status from rf_ds_used_spare 2758 to rf_ds_optimal */ 2759 /* clabel.status = rf_ds_optimal; */ 2760 2761 raidmarkdirty(raidPtr, sparecol); 2762 } 2763 } 2764 } 2765 2766 2767 void 2768 rf_update_component_labels(RF_Raid_t *raidPtr, int final) 2769 { 2770 RF_ComponentLabel_t *clabel; 2771 int sparecol; 2772 int c; 2773 int j; 2774 int scol; 2775 2776 scol = -1; 2777 2778 /* XXX should do extra checks to make sure things really are clean, 2779 rather than blindly setting the clean bit... */ 2780 2781 raidPtr->mod_counter++; 2782 2783 for (c = 0; c < raidPtr->numCol; c++) { 2784 if (raidPtr->Disks[c].status == rf_ds_optimal) { 2785 clabel = raidget_component_label(raidPtr, c); 2786 /* make sure status is noted */ 2787 clabel->status = rf_ds_optimal; 2788 2789 /* note what unit we are configured as */ 2790 clabel->last_unit = raidPtr->raidid; 2791 2792 raidflush_component_label(raidPtr, c); 2793 if (final == RF_FINAL_COMPONENT_UPDATE) { 2794 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2795 raidmarkclean(raidPtr, c); 2796 } 2797 } 2798 } 2799 /* else we don't touch it.. */ 2800 } 2801 2802 for( c = 0; c < raidPtr->numSpare ; c++) { 2803 sparecol = raidPtr->numCol + c; 2804 /* Need to ensure that the reconstruct actually completed! */ 2805 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2806 /* 2807 2808 we claim this disk is "optimal" if it's 2809 rf_ds_used_spare, as that means it should be 2810 directly substitutable for the disk it replaced. 2811 We note that too... 2812 2813 */ 2814 2815 for(j=0;j<raidPtr->numCol;j++) { 2816 if (raidPtr->Disks[j].spareCol == sparecol) { 2817 scol = j; 2818 break; 2819 } 2820 } 2821 2822 /* XXX shouldn't *really* need this... */ 2823 clabel = raidget_component_label(raidPtr, sparecol); 2824 /* make sure status is noted */ 2825 2826 raid_init_component_label(raidPtr, clabel); 2827 2828 clabel->column = scol; 2829 clabel->status = rf_ds_optimal; 2830 clabel->last_unit = raidPtr->raidid; 2831 2832 raidflush_component_label(raidPtr, sparecol); 2833 if (final == RF_FINAL_COMPONENT_UPDATE) { 2834 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2835 raidmarkclean(raidPtr, sparecol); 2836 } 2837 } 2838 } 2839 } 2840 } 2841 2842 void 2843 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured) 2844 { 2845 2846 if (vp != NULL) { 2847 if (auto_configured == 1) { 2848 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2849 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2850 vput(vp); 2851 2852 } else { 2853 (void) vn_close(vp, FREAD | FWRITE, curlwp->l_cred); 2854 } 2855 } 2856 } 2857 2858 2859 void 2860 rf_UnconfigureVnodes(RF_Raid_t *raidPtr) 2861 { 2862 int r,c; 2863 struct vnode *vp; 2864 int acd; 2865 2866 2867 /* We take this opportunity to close the vnodes like we should.. */ 2868 2869 for (c = 0; c < raidPtr->numCol; c++) { 2870 vp = raidPtr->raid_cinfo[c].ci_vp; 2871 acd = raidPtr->Disks[c].auto_configured; 2872 rf_close_component(raidPtr, vp, acd); 2873 raidPtr->raid_cinfo[c].ci_vp = NULL; 2874 raidPtr->Disks[c].auto_configured = 0; 2875 } 2876 2877 for (r = 0; r < raidPtr->numSpare; r++) { 2878 vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp; 2879 acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured; 2880 rf_close_component(raidPtr, vp, acd); 2881 raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL; 2882 raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0; 2883 } 2884 } 2885 2886 2887 void 2888 rf_ReconThread(struct rf_recon_req *req) 2889 { 2890 int s; 2891 RF_Raid_t *raidPtr; 2892 2893 s = splbio(); 2894 raidPtr = (RF_Raid_t *) req->raidPtr; 2895 raidPtr->recon_in_progress = 1; 2896 2897 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col, 2898 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2899 2900 RF_Free(req, sizeof(*req)); 2901 2902 raidPtr->recon_in_progress = 0; 2903 splx(s); 2904 2905 /* That's all... */ 2906 kthread_exit(0); /* does not return */ 2907 } 2908 2909 void 2910 rf_RewriteParityThread(RF_Raid_t *raidPtr) 2911 { 2912 int retcode; 2913 int s; 2914 2915 raidPtr->parity_rewrite_stripes_done = 0; 2916 raidPtr->parity_rewrite_in_progress = 1; 2917 s = splbio(); 2918 retcode = rf_RewriteParity(raidPtr); 2919 splx(s); 2920 if (retcode) { 2921 printf("raid%d: Error re-writing parity (%d)!\n", 2922 raidPtr->raidid, retcode); 2923 } else { 2924 /* set the clean bit! If we shutdown correctly, 2925 the clean bit on each component label will get 2926 set */ 2927 raidPtr->parity_good = RF_RAID_CLEAN; 2928 } 2929 raidPtr->parity_rewrite_in_progress = 0; 2930 2931 /* Anyone waiting for us to stop? If so, inform them... */ 2932 if (raidPtr->waitShutdown) { 2933 wakeup(&raidPtr->parity_rewrite_in_progress); 2934 } 2935 2936 /* That's all... */ 2937 kthread_exit(0); /* does not return */ 2938 } 2939 2940 2941 void 2942 rf_CopybackThread(RF_Raid_t *raidPtr) 2943 { 2944 int s; 2945 2946 raidPtr->copyback_in_progress = 1; 2947 s = splbio(); 2948 rf_CopybackReconstructedData(raidPtr); 2949 splx(s); 2950 raidPtr->copyback_in_progress = 0; 2951 2952 /* That's all... */ 2953 kthread_exit(0); /* does not return */ 2954 } 2955 2956 2957 void 2958 rf_ReconstructInPlaceThread(struct rf_recon_req *req) 2959 { 2960 int s; 2961 RF_Raid_t *raidPtr; 2962 2963 s = splbio(); 2964 raidPtr = req->raidPtr; 2965 raidPtr->recon_in_progress = 1; 2966 rf_ReconstructInPlace(raidPtr, req->col); 2967 RF_Free(req, sizeof(*req)); 2968 raidPtr->recon_in_progress = 0; 2969 splx(s); 2970 2971 /* That's all... */ 2972 kthread_exit(0); /* does not return */ 2973 } 2974 2975 static RF_AutoConfig_t * 2976 rf_get_component(RF_AutoConfig_t *ac_list, dev_t dev, struct vnode *vp, 2977 const char *cname, RF_SectorCount_t size, uint64_t numsecs, 2978 unsigned secsize) 2979 { 2980 int good_one = 0; 2981 RF_ComponentLabel_t *clabel; 2982 RF_AutoConfig_t *ac; 2983 2984 clabel = malloc(sizeof(RF_ComponentLabel_t), M_RAIDFRAME, M_NOWAIT); 2985 if (clabel == NULL) { 2986 oomem: 2987 while(ac_list) { 2988 ac = ac_list; 2989 if (ac->clabel) 2990 free(ac->clabel, M_RAIDFRAME); 2991 ac_list = ac_list->next; 2992 free(ac, M_RAIDFRAME); 2993 } 2994 printf("RAID auto config: out of memory!\n"); 2995 return NULL; /* XXX probably should panic? */ 2996 } 2997 2998 if (!raidread_component_label(secsize, dev, vp, clabel)) { 2999 /* Got the label. Does it look reasonable? */ 3000 if (rf_reasonable_label(clabel, numsecs) && 3001 (rf_component_label_partitionsize(clabel) <= size)) { 3002 #ifdef DEBUG 3003 printf("Component on: %s: %llu\n", 3004 cname, (unsigned long long)size); 3005 rf_print_component_label(clabel); 3006 #endif 3007 /* if it's reasonable, add it, else ignore it. */ 3008 ac = malloc(sizeof(RF_AutoConfig_t), M_RAIDFRAME, 3009 M_NOWAIT); 3010 if (ac == NULL) { 3011 free(clabel, M_RAIDFRAME); 3012 goto oomem; 3013 } 3014 strlcpy(ac->devname, cname, sizeof(ac->devname)); 3015 ac->dev = dev; 3016 ac->vp = vp; 3017 ac->clabel = clabel; 3018 ac->next = ac_list; 3019 ac_list = ac; 3020 good_one = 1; 3021 } 3022 } 3023 if (!good_one) { 3024 /* cleanup */ 3025 free(clabel, M_RAIDFRAME); 3026 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3027 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3028 vput(vp); 3029 } 3030 return ac_list; 3031 } 3032 3033 RF_AutoConfig_t * 3034 rf_find_raid_components(void) 3035 { 3036 struct vnode *vp; 3037 struct disklabel label; 3038 device_t dv; 3039 deviter_t di; 3040 dev_t dev; 3041 int bmajor, bminor, wedge, rf_part_found; 3042 int error; 3043 int i; 3044 RF_AutoConfig_t *ac_list; 3045 uint64_t numsecs; 3046 unsigned secsize; 3047 3048 /* initialize the AutoConfig list */ 3049 ac_list = NULL; 3050 3051 /* we begin by trolling through *all* the devices on the system */ 3052 3053 for (dv = deviter_first(&di, DEVITER_F_ROOT_FIRST); dv != NULL; 3054 dv = deviter_next(&di)) { 3055 3056 /* we are only interested in disks... */ 3057 if (device_class(dv) != DV_DISK) 3058 continue; 3059 3060 /* we don't care about floppies... */ 3061 if (device_is_a(dv, "fd")) { 3062 continue; 3063 } 3064 3065 /* we don't care about CD's... */ 3066 if (device_is_a(dv, "cd")) { 3067 continue; 3068 } 3069 3070 /* we don't care about md's... */ 3071 if (device_is_a(dv, "md")) { 3072 continue; 3073 } 3074 3075 /* hdfd is the Atari/Hades floppy driver */ 3076 if (device_is_a(dv, "hdfd")) { 3077 continue; 3078 } 3079 3080 /* fdisa is the Atari/Milan floppy driver */ 3081 if (device_is_a(dv, "fdisa")) { 3082 continue; 3083 } 3084 3085 /* need to find the device_name_to_block_device_major stuff */ 3086 bmajor = devsw_name2blk(device_xname(dv), NULL, 0); 3087 3088 rf_part_found = 0; /*No raid partition as yet*/ 3089 3090 /* get a vnode for the raw partition of this disk */ 3091 3092 wedge = device_is_a(dv, "dk"); 3093 bminor = minor(device_unit(dv)); 3094 dev = wedge ? makedev(bmajor, bminor) : 3095 MAKEDISKDEV(bmajor, bminor, RAW_PART); 3096 if (bdevvp(dev, &vp)) 3097 panic("RAID can't alloc vnode"); 3098 3099 error = VOP_OPEN(vp, FREAD | FSILENT, NOCRED); 3100 3101 if (error) { 3102 /* "Who cares." Continue looking 3103 for something that exists*/ 3104 vput(vp); 3105 continue; 3106 } 3107 3108 error = getdisksize(vp, &numsecs, &secsize); 3109 if (error) { 3110 vput(vp); 3111 continue; 3112 } 3113 if (wedge) { 3114 struct dkwedge_info dkw; 3115 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, 3116 NOCRED); 3117 if (error) { 3118 printf("RAIDframe: can't get wedge info for " 3119 "dev %s (%d)\n", device_xname(dv), error); 3120 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3121 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3122 vput(vp); 3123 continue; 3124 } 3125 3126 if (strcmp(dkw.dkw_ptype, DKW_PTYPE_RAIDFRAME) != 0) { 3127 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3128 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3129 vput(vp); 3130 continue; 3131 } 3132 3133 ac_list = rf_get_component(ac_list, dev, vp, 3134 device_xname(dv), dkw.dkw_size, numsecs, secsize); 3135 rf_part_found = 1; /*There is a raid component on this disk*/ 3136 continue; 3137 } 3138 3139 /* Ok, the disk exists. Go get the disklabel. */ 3140 error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED); 3141 if (error) { 3142 /* 3143 * XXX can't happen - open() would 3144 * have errored out (or faked up one) 3145 */ 3146 if (error != ENOTTY) 3147 printf("RAIDframe: can't get label for dev " 3148 "%s (%d)\n", device_xname(dv), error); 3149 } 3150 3151 /* don't need this any more. We'll allocate it again 3152 a little later if we really do... */ 3153 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3154 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3155 vput(vp); 3156 3157 if (error) 3158 continue; 3159 3160 rf_part_found = 0; /*No raid partitions yet*/ 3161 for (i = 0; i < label.d_npartitions; i++) { 3162 char cname[sizeof(ac_list->devname)]; 3163 3164 /* We only support partitions marked as RAID */ 3165 if (label.d_partitions[i].p_fstype != FS_RAID) 3166 continue; 3167 3168 dev = MAKEDISKDEV(bmajor, device_unit(dv), i); 3169 if (bdevvp(dev, &vp)) 3170 panic("RAID can't alloc vnode"); 3171 3172 error = VOP_OPEN(vp, FREAD, NOCRED); 3173 if (error) { 3174 /* Whatever... */ 3175 vput(vp); 3176 continue; 3177 } 3178 snprintf(cname, sizeof(cname), "%s%c", 3179 device_xname(dv), 'a' + i); 3180 ac_list = rf_get_component(ac_list, dev, vp, cname, 3181 label.d_partitions[i].p_size, numsecs, secsize); 3182 rf_part_found = 1; /*There is at least one raid partition on this disk*/ 3183 } 3184 3185 /* 3186 *If there is no raid component on this disk, either in a 3187 *disklabel or inside a wedge, check the raw partition as well, 3188 *as it is possible to configure raid components on raw disk 3189 *devices. 3190 */ 3191 3192 if (!rf_part_found) { 3193 char cname[sizeof(ac_list->devname)]; 3194 3195 dev = MAKEDISKDEV(bmajor, device_unit(dv), RAW_PART); 3196 if (bdevvp(dev, &vp)) 3197 panic("RAID can't alloc vnode"); 3198 3199 error = VOP_OPEN(vp, FREAD, NOCRED); 3200 if (error) { 3201 /* Whatever... */ 3202 vput(vp); 3203 continue; 3204 } 3205 snprintf(cname, sizeof(cname), "%s%c", 3206 device_xname(dv), 'a' + RAW_PART); 3207 ac_list = rf_get_component(ac_list, dev, vp, cname, 3208 label.d_partitions[RAW_PART].p_size, numsecs, secsize); 3209 } 3210 } 3211 deviter_release(&di); 3212 return ac_list; 3213 } 3214 3215 3216 int 3217 rf_reasonable_label(RF_ComponentLabel_t *clabel, uint64_t numsecs) 3218 { 3219 3220 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 3221 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 3222 ((clabel->clean == RF_RAID_CLEAN) || 3223 (clabel->clean == RF_RAID_DIRTY)) && 3224 clabel->row >=0 && 3225 clabel->column >= 0 && 3226 clabel->num_rows > 0 && 3227 clabel->num_columns > 0 && 3228 clabel->row < clabel->num_rows && 3229 clabel->column < clabel->num_columns && 3230 clabel->blockSize > 0 && 3231 /* 3232 * numBlocksHi may contain garbage, but it is ok since 3233 * the type is unsigned. If it is really garbage, 3234 * rf_fix_old_label_size() will fix it. 3235 */ 3236 rf_component_label_numblocks(clabel) > 0) { 3237 /* 3238 * label looks reasonable enough... 3239 * let's make sure it has no old garbage. 3240 */ 3241 if (numsecs) 3242 rf_fix_old_label_size(clabel, numsecs); 3243 return(1); 3244 } 3245 return(0); 3246 } 3247 3248 3249 /* 3250 * For reasons yet unknown, some old component labels have garbage in 3251 * the newer numBlocksHi region, and this causes lossage. Since those 3252 * disks will also have numsecs set to less than 32 bits of sectors, 3253 * we can determine when this corruption has occured, and fix it. 3254 * 3255 * The exact same problem, with the same unknown reason, happens to 3256 * the partitionSizeHi member as well. 3257 */ 3258 static void 3259 rf_fix_old_label_size(RF_ComponentLabel_t *clabel, uint64_t numsecs) 3260 { 3261 3262 if (numsecs < ((uint64_t)1 << 32)) { 3263 if (clabel->numBlocksHi) { 3264 printf("WARNING: total sectors < 32 bits, yet " 3265 "numBlocksHi set\n" 3266 "WARNING: resetting numBlocksHi to zero.\n"); 3267 clabel->numBlocksHi = 0; 3268 } 3269 3270 if (clabel->partitionSizeHi) { 3271 printf("WARNING: total sectors < 32 bits, yet " 3272 "partitionSizeHi set\n" 3273 "WARNING: resetting partitionSizeHi to zero.\n"); 3274 clabel->partitionSizeHi = 0; 3275 } 3276 } 3277 } 3278 3279 3280 #ifdef DEBUG 3281 void 3282 rf_print_component_label(RF_ComponentLabel_t *clabel) 3283 { 3284 uint64_t numBlocks; 3285 3286 numBlocks = rf_component_label_numblocks(clabel); 3287 3288 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 3289 clabel->row, clabel->column, 3290 clabel->num_rows, clabel->num_columns); 3291 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 3292 clabel->version, clabel->serial_number, 3293 clabel->mod_counter); 3294 printf(" Clean: %s Status: %d\n", 3295 clabel->clean ? "Yes" : "No", clabel->status); 3296 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 3297 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 3298 printf(" RAID Level: %c blocksize: %d numBlocks: %"PRIu64"\n", 3299 (char) clabel->parityConfig, clabel->blockSize, numBlocks); 3300 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No"); 3301 printf(" Contains root partition: %s\n", 3302 clabel->root_partition ? "Yes" : "No"); 3303 printf(" Last configured as: raid%d\n", clabel->last_unit); 3304 #if 0 3305 printf(" Config order: %d\n", clabel->config_order); 3306 #endif 3307 3308 } 3309 #endif 3310 3311 RF_ConfigSet_t * 3312 rf_create_auto_sets(RF_AutoConfig_t *ac_list) 3313 { 3314 RF_AutoConfig_t *ac; 3315 RF_ConfigSet_t *config_sets; 3316 RF_ConfigSet_t *cset; 3317 RF_AutoConfig_t *ac_next; 3318 3319 3320 config_sets = NULL; 3321 3322 /* Go through the AutoConfig list, and figure out which components 3323 belong to what sets. */ 3324 ac = ac_list; 3325 while(ac!=NULL) { 3326 /* we're going to putz with ac->next, so save it here 3327 for use at the end of the loop */ 3328 ac_next = ac->next; 3329 3330 if (config_sets == NULL) { 3331 /* will need at least this one... */ 3332 config_sets = (RF_ConfigSet_t *) 3333 malloc(sizeof(RF_ConfigSet_t), 3334 M_RAIDFRAME, M_NOWAIT); 3335 if (config_sets == NULL) { 3336 panic("rf_create_auto_sets: No memory!"); 3337 } 3338 /* this one is easy :) */ 3339 config_sets->ac = ac; 3340 config_sets->next = NULL; 3341 config_sets->rootable = 0; 3342 ac->next = NULL; 3343 } else { 3344 /* which set does this component fit into? */ 3345 cset = config_sets; 3346 while(cset!=NULL) { 3347 if (rf_does_it_fit(cset, ac)) { 3348 /* looks like it matches... */ 3349 ac->next = cset->ac; 3350 cset->ac = ac; 3351 break; 3352 } 3353 cset = cset->next; 3354 } 3355 if (cset==NULL) { 3356 /* didn't find a match above... new set..*/ 3357 cset = (RF_ConfigSet_t *) 3358 malloc(sizeof(RF_ConfigSet_t), 3359 M_RAIDFRAME, M_NOWAIT); 3360 if (cset == NULL) { 3361 panic("rf_create_auto_sets: No memory!"); 3362 } 3363 cset->ac = ac; 3364 ac->next = NULL; 3365 cset->next = config_sets; 3366 cset->rootable = 0; 3367 config_sets = cset; 3368 } 3369 } 3370 ac = ac_next; 3371 } 3372 3373 3374 return(config_sets); 3375 } 3376 3377 static int 3378 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac) 3379 { 3380 RF_ComponentLabel_t *clabel1, *clabel2; 3381 3382 /* If this one matches the *first* one in the set, that's good 3383 enough, since the other members of the set would have been 3384 through here too... */ 3385 /* note that we are not checking partitionSize here.. 3386 3387 Note that we are also not checking the mod_counters here. 3388 If everything else matches execpt the mod_counter, that's 3389 good enough for this test. We will deal with the mod_counters 3390 a little later in the autoconfiguration process. 3391 3392 (clabel1->mod_counter == clabel2->mod_counter) && 3393 3394 The reason we don't check for this is that failed disks 3395 will have lower modification counts. If those disks are 3396 not added to the set they used to belong to, then they will 3397 form their own set, which may result in 2 different sets, 3398 for example, competing to be configured at raid0, and 3399 perhaps competing to be the root filesystem set. If the 3400 wrong ones get configured, or both attempt to become /, 3401 weird behaviour and or serious lossage will occur. Thus we 3402 need to bring them into the fold here, and kick them out at 3403 a later point. 3404 3405 */ 3406 3407 clabel1 = cset->ac->clabel; 3408 clabel2 = ac->clabel; 3409 if ((clabel1->version == clabel2->version) && 3410 (clabel1->serial_number == clabel2->serial_number) && 3411 (clabel1->num_rows == clabel2->num_rows) && 3412 (clabel1->num_columns == clabel2->num_columns) && 3413 (clabel1->sectPerSU == clabel2->sectPerSU) && 3414 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 3415 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 3416 (clabel1->parityConfig == clabel2->parityConfig) && 3417 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 3418 (clabel1->blockSize == clabel2->blockSize) && 3419 rf_component_label_numblocks(clabel1) == 3420 rf_component_label_numblocks(clabel2) && 3421 (clabel1->autoconfigure == clabel2->autoconfigure) && 3422 (clabel1->root_partition == clabel2->root_partition) && 3423 (clabel1->last_unit == clabel2->last_unit) && 3424 (clabel1->config_order == clabel2->config_order)) { 3425 /* if it get's here, it almost *has* to be a match */ 3426 } else { 3427 /* it's not consistent with somebody in the set.. 3428 punt */ 3429 return(0); 3430 } 3431 /* all was fine.. it must fit... */ 3432 return(1); 3433 } 3434 3435 int 3436 rf_have_enough_components(RF_ConfigSet_t *cset) 3437 { 3438 RF_AutoConfig_t *ac; 3439 RF_AutoConfig_t *auto_config; 3440 RF_ComponentLabel_t *clabel; 3441 int c; 3442 int num_cols; 3443 int num_missing; 3444 int mod_counter; 3445 int mod_counter_found; 3446 int even_pair_failed; 3447 char parity_type; 3448 3449 3450 /* check to see that we have enough 'live' components 3451 of this set. If so, we can configure it if necessary */ 3452 3453 num_cols = cset->ac->clabel->num_columns; 3454 parity_type = cset->ac->clabel->parityConfig; 3455 3456 /* XXX Check for duplicate components!?!?!? */ 3457 3458 /* Determine what the mod_counter is supposed to be for this set. */ 3459 3460 mod_counter_found = 0; 3461 mod_counter = 0; 3462 ac = cset->ac; 3463 while(ac!=NULL) { 3464 if (mod_counter_found==0) { 3465 mod_counter = ac->clabel->mod_counter; 3466 mod_counter_found = 1; 3467 } else { 3468 if (ac->clabel->mod_counter > mod_counter) { 3469 mod_counter = ac->clabel->mod_counter; 3470 } 3471 } 3472 ac = ac->next; 3473 } 3474 3475 num_missing = 0; 3476 auto_config = cset->ac; 3477 3478 even_pair_failed = 0; 3479 for(c=0; c<num_cols; c++) { 3480 ac = auto_config; 3481 while(ac!=NULL) { 3482 if ((ac->clabel->column == c) && 3483 (ac->clabel->mod_counter == mod_counter)) { 3484 /* it's this one... */ 3485 #ifdef DEBUG 3486 printf("Found: %s at %d\n", 3487 ac->devname,c); 3488 #endif 3489 break; 3490 } 3491 ac=ac->next; 3492 } 3493 if (ac==NULL) { 3494 /* Didn't find one here! */ 3495 /* special case for RAID 1, especially 3496 where there are more than 2 3497 components (where RAIDframe treats 3498 things a little differently :( ) */ 3499 if (parity_type == '1') { 3500 if (c%2 == 0) { /* even component */ 3501 even_pair_failed = 1; 3502 } else { /* odd component. If 3503 we're failed, and 3504 so is the even 3505 component, it's 3506 "Good Night, Charlie" */ 3507 if (even_pair_failed == 1) { 3508 return(0); 3509 } 3510 } 3511 } else { 3512 /* normal accounting */ 3513 num_missing++; 3514 } 3515 } 3516 if ((parity_type == '1') && (c%2 == 1)) { 3517 /* Just did an even component, and we didn't 3518 bail.. reset the even_pair_failed flag, 3519 and go on to the next component.... */ 3520 even_pair_failed = 0; 3521 } 3522 } 3523 3524 clabel = cset->ac->clabel; 3525 3526 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3527 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3528 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3529 /* XXX this needs to be made *much* more general */ 3530 /* Too many failures */ 3531 return(0); 3532 } 3533 /* otherwise, all is well, and we've got enough to take a kick 3534 at autoconfiguring this set */ 3535 return(1); 3536 } 3537 3538 void 3539 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config, 3540 RF_Raid_t *raidPtr) 3541 { 3542 RF_ComponentLabel_t *clabel; 3543 int i; 3544 3545 clabel = ac->clabel; 3546 3547 /* 1. Fill in the common stuff */ 3548 config->numRow = clabel->num_rows = 1; 3549 config->numCol = clabel->num_columns; 3550 config->numSpare = 0; /* XXX should this be set here? */ 3551 config->sectPerSU = clabel->sectPerSU; 3552 config->SUsPerPU = clabel->SUsPerPU; 3553 config->SUsPerRU = clabel->SUsPerRU; 3554 config->parityConfig = clabel->parityConfig; 3555 /* XXX... */ 3556 strcpy(config->diskQueueType,"fifo"); 3557 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3558 config->layoutSpecificSize = 0; /* XXX ?? */ 3559 3560 while(ac!=NULL) { 3561 /* row/col values will be in range due to the checks 3562 in reasonable_label() */ 3563 strcpy(config->devnames[0][ac->clabel->column], 3564 ac->devname); 3565 ac = ac->next; 3566 } 3567 3568 for(i=0;i<RF_MAXDBGV;i++) { 3569 config->debugVars[i][0] = 0; 3570 } 3571 } 3572 3573 int 3574 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value) 3575 { 3576 RF_ComponentLabel_t *clabel; 3577 int column; 3578 int sparecol; 3579 3580 raidPtr->autoconfigure = new_value; 3581 3582 for(column=0; column<raidPtr->numCol; column++) { 3583 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3584 clabel = raidget_component_label(raidPtr, column); 3585 clabel->autoconfigure = new_value; 3586 raidflush_component_label(raidPtr, column); 3587 } 3588 } 3589 for(column = 0; column < raidPtr->numSpare ; column++) { 3590 sparecol = raidPtr->numCol + column; 3591 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3592 clabel = raidget_component_label(raidPtr, sparecol); 3593 clabel->autoconfigure = new_value; 3594 raidflush_component_label(raidPtr, sparecol); 3595 } 3596 } 3597 return(new_value); 3598 } 3599 3600 int 3601 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value) 3602 { 3603 RF_ComponentLabel_t *clabel; 3604 int column; 3605 int sparecol; 3606 3607 raidPtr->root_partition = new_value; 3608 for(column=0; column<raidPtr->numCol; column++) { 3609 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3610 clabel = raidget_component_label(raidPtr, column); 3611 clabel->root_partition = new_value; 3612 raidflush_component_label(raidPtr, column); 3613 } 3614 } 3615 for(column = 0; column < raidPtr->numSpare ; column++) { 3616 sparecol = raidPtr->numCol + column; 3617 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3618 clabel = raidget_component_label(raidPtr, sparecol); 3619 clabel->root_partition = new_value; 3620 raidflush_component_label(raidPtr, sparecol); 3621 } 3622 } 3623 return(new_value); 3624 } 3625 3626 void 3627 rf_release_all_vps(RF_ConfigSet_t *cset) 3628 { 3629 RF_AutoConfig_t *ac; 3630 3631 ac = cset->ac; 3632 while(ac!=NULL) { 3633 /* Close the vp, and give it back */ 3634 if (ac->vp) { 3635 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3636 VOP_CLOSE(ac->vp, FREAD, NOCRED); 3637 vput(ac->vp); 3638 ac->vp = NULL; 3639 } 3640 ac = ac->next; 3641 } 3642 } 3643 3644 3645 void 3646 rf_cleanup_config_set(RF_ConfigSet_t *cset) 3647 { 3648 RF_AutoConfig_t *ac; 3649 RF_AutoConfig_t *next_ac; 3650 3651 ac = cset->ac; 3652 while(ac!=NULL) { 3653 next_ac = ac->next; 3654 /* nuke the label */ 3655 free(ac->clabel, M_RAIDFRAME); 3656 /* cleanup the config structure */ 3657 free(ac, M_RAIDFRAME); 3658 /* "next.." */ 3659 ac = next_ac; 3660 } 3661 /* and, finally, nuke the config set */ 3662 free(cset, M_RAIDFRAME); 3663 } 3664 3665 3666 void 3667 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel) 3668 { 3669 /* current version number */ 3670 clabel->version = RF_COMPONENT_LABEL_VERSION; 3671 clabel->serial_number = raidPtr->serial_number; 3672 clabel->mod_counter = raidPtr->mod_counter; 3673 3674 clabel->num_rows = 1; 3675 clabel->num_columns = raidPtr->numCol; 3676 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3677 clabel->status = rf_ds_optimal; /* "It's good!" */ 3678 3679 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3680 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3681 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3682 3683 clabel->blockSize = raidPtr->bytesPerSector; 3684 rf_component_label_set_numblocks(clabel, raidPtr->sectorsPerDisk); 3685 3686 /* XXX not portable */ 3687 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3688 clabel->maxOutstanding = raidPtr->maxOutstanding; 3689 clabel->autoconfigure = raidPtr->autoconfigure; 3690 clabel->root_partition = raidPtr->root_partition; 3691 clabel->last_unit = raidPtr->raidid; 3692 clabel->config_order = raidPtr->config_order; 3693 3694 #ifndef RF_NO_PARITY_MAP 3695 rf_paritymap_init_label(raidPtr->parity_map, clabel); 3696 #endif 3697 } 3698 3699 int 3700 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit) 3701 { 3702 RF_Raid_t *raidPtr; 3703 RF_Config_t *config; 3704 int raidID; 3705 int retcode; 3706 3707 #ifdef DEBUG 3708 printf("RAID autoconfigure\n"); 3709 #endif 3710 3711 retcode = 0; 3712 *unit = -1; 3713 3714 /* 1. Create a config structure */ 3715 3716 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3717 M_RAIDFRAME, 3718 M_NOWAIT); 3719 if (config==NULL) { 3720 printf("Out of mem!?!?\n"); 3721 /* XXX do something more intelligent here. */ 3722 return(1); 3723 } 3724 3725 memset(config, 0, sizeof(RF_Config_t)); 3726 3727 /* 3728 2. Figure out what RAID ID this one is supposed to live at 3729 See if we can get the same RAID dev that it was configured 3730 on last time.. 3731 */ 3732 3733 raidID = cset->ac->clabel->last_unit; 3734 if ((raidID < 0) || (raidID >= numraid)) { 3735 /* let's not wander off into lala land. */ 3736 raidID = numraid - 1; 3737 } 3738 if (raidPtrs[raidID]->valid != 0) { 3739 3740 /* 3741 Nope... Go looking for an alternative... 3742 Start high so we don't immediately use raid0 if that's 3743 not taken. 3744 */ 3745 3746 for(raidID = numraid - 1; raidID >= 0; raidID--) { 3747 if (raidPtrs[raidID]->valid == 0) { 3748 /* can use this one! */ 3749 break; 3750 } 3751 } 3752 } 3753 3754 if (raidID < 0) { 3755 /* punt... */ 3756 printf("Unable to auto configure this set!\n"); 3757 printf("(Out of RAID devs!)\n"); 3758 free(config, M_RAIDFRAME); 3759 return(1); 3760 } 3761 3762 #ifdef DEBUG 3763 printf("Configuring raid%d:\n",raidID); 3764 #endif 3765 3766 raidPtr = raidPtrs[raidID]; 3767 3768 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3769 raidPtr->raidid = raidID; 3770 raidPtr->openings = RAIDOUTSTANDING; 3771 3772 /* 3. Build the configuration structure */ 3773 rf_create_configuration(cset->ac, config, raidPtr); 3774 3775 /* 4. Do the configuration */ 3776 retcode = rf_Configure(raidPtr, config, cset->ac); 3777 3778 if (retcode == 0) { 3779 3780 raidinit(raidPtrs[raidID]); 3781 3782 rf_markalldirty(raidPtrs[raidID]); 3783 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3784 if (cset->ac->clabel->root_partition==1) { 3785 /* everything configured just fine. Make a note 3786 that this set is eligible to be root. */ 3787 cset->rootable = 1; 3788 /* XXX do this here? */ 3789 raidPtrs[raidID]->root_partition = 1; 3790 } 3791 } 3792 3793 /* 5. Cleanup */ 3794 free(config, M_RAIDFRAME); 3795 3796 *unit = raidID; 3797 return(retcode); 3798 } 3799 3800 void 3801 rf_disk_unbusy(RF_RaidAccessDesc_t *desc) 3802 { 3803 struct buf *bp; 3804 3805 bp = (struct buf *)desc->bp; 3806 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3807 (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ)); 3808 } 3809 3810 void 3811 rf_pool_init(struct pool *p, size_t size, const char *w_chan, 3812 size_t xmin, size_t xmax) 3813 { 3814 pool_init(p, size, 0, 0, 0, w_chan, NULL, IPL_BIO); 3815 pool_sethiwat(p, xmax); 3816 pool_prime(p, xmin); 3817 pool_setlowat(p, xmin); 3818 } 3819 3820 /* 3821 * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see 3822 * if there is IO pending and if that IO could possibly be done for a 3823 * given RAID set. Returns 0 if IO is waiting and can be done, 1 3824 * otherwise. 3825 * 3826 */ 3827 3828 int 3829 rf_buf_queue_check(int raidid) 3830 { 3831 if ((bufq_peek(raid_softc[raidid].buf_queue) != NULL) && 3832 raidPtrs[raidid]->openings > 0) { 3833 /* there is work to do */ 3834 return 0; 3835 } 3836 /* default is nothing to do */ 3837 return 1; 3838 } 3839 3840 int 3841 rf_getdisksize(struct vnode *vp, RF_RaidDisk_t *diskPtr) 3842 { 3843 uint64_t numsecs; 3844 unsigned secsize; 3845 int error; 3846 3847 error = getdisksize(vp, &numsecs, &secsize); 3848 if (error == 0) { 3849 diskPtr->blockSize = secsize; 3850 diskPtr->numBlocks = numsecs - rf_protectedSectors; 3851 diskPtr->partitionSize = numsecs; 3852 return 0; 3853 } 3854 return error; 3855 } 3856 3857 static int 3858 raid_match(device_t self, cfdata_t cfdata, void *aux) 3859 { 3860 return 1; 3861 } 3862 3863 static void 3864 raid_attach(device_t parent, device_t self, void *aux) 3865 { 3866 3867 } 3868 3869 3870 static int 3871 raid_detach(device_t self, int flags) 3872 { 3873 int error; 3874 struct raid_softc *rs = &raid_softc[device_unit(self)]; 3875 3876 if ((error = raidlock(rs)) != 0) 3877 return (error); 3878 3879 error = raid_detach_unlocked(rs); 3880 3881 raidunlock(rs); 3882 3883 return error; 3884 } 3885 3886 static void 3887 rf_set_properties(struct raid_softc *rs, RF_Raid_t *raidPtr) 3888 { 3889 prop_dictionary_t disk_info, odisk_info, geom; 3890 disk_info = prop_dictionary_create(); 3891 geom = prop_dictionary_create(); 3892 prop_dictionary_set_uint64(geom, "sectors-per-unit", 3893 raidPtr->totalSectors); 3894 prop_dictionary_set_uint32(geom, "sector-size", 3895 raidPtr->bytesPerSector); 3896 3897 prop_dictionary_set_uint16(geom, "sectors-per-track", 3898 raidPtr->Layout.dataSectorsPerStripe); 3899 prop_dictionary_set_uint16(geom, "tracks-per-cylinder", 3900 4 * raidPtr->numCol); 3901 3902 prop_dictionary_set_uint64(geom, "cylinders-per-unit", 3903 raidPtr->totalSectors / (raidPtr->Layout.dataSectorsPerStripe * 3904 (4 * raidPtr->numCol))); 3905 3906 prop_dictionary_set(disk_info, "geometry", geom); 3907 prop_object_release(geom); 3908 prop_dictionary_set(device_properties(rs->sc_dev), 3909 "disk-info", disk_info); 3910 odisk_info = rs->sc_dkdev.dk_info; 3911 rs->sc_dkdev.dk_info = disk_info; 3912 if (odisk_info) 3913 prop_object_release(odisk_info); 3914 } 3915 3916 /* 3917 * Implement forwarding of the DIOCCACHESYNC ioctl to each of the components. 3918 * We end up returning whatever error was returned by the first cache flush 3919 * that fails. 3920 */ 3921 3922 int 3923 rf_sync_component_caches(RF_Raid_t *raidPtr) 3924 { 3925 int c, sparecol; 3926 int e,error; 3927 int force = 1; 3928 3929 error = 0; 3930 for (c = 0; c < raidPtr->numCol; c++) { 3931 if (raidPtr->Disks[c].status == rf_ds_optimal) { 3932 e = VOP_IOCTL(raidPtr->raid_cinfo[c].ci_vp, DIOCCACHESYNC, 3933 &force, FWRITE, NOCRED); 3934 if (e) { 3935 if (e != ENODEV) 3936 printf("raid%d: cache flush to component %s failed.\n", 3937 raidPtr->raidid, raidPtr->Disks[c].devname); 3938 if (error == 0) { 3939 error = e; 3940 } 3941 } 3942 } 3943 } 3944 3945 for( c = 0; c < raidPtr->numSpare ; c++) { 3946 sparecol = raidPtr->numCol + c; 3947 /* Need to ensure that the reconstruct actually completed! */ 3948 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3949 e = VOP_IOCTL(raidPtr->raid_cinfo[sparecol].ci_vp, 3950 DIOCCACHESYNC, &force, FWRITE, NOCRED); 3951 if (e) { 3952 if (e != ENODEV) 3953 printf("raid%d: cache flush to component %s failed.\n", 3954 raidPtr->raidid, raidPtr->Disks[sparecol].devname); 3955 if (error == 0) { 3956 error = e; 3957 } 3958 } 3959 } 3960 } 3961 return error; 3962 } 3963
Indexes created Tue Sep 30 20:09:53 GMT 2025