1 /* $NetBSD: rf_netbsdkintf.c,v 1.245 2008/04/05 18:54:10 cegger Exp $ */ 2 /*- 3 * Copyright (c) 1996, 1997, 1998, 2008 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Greg Oster; Jason R. Thorpe. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 38 /* 39 * Copyright (c) 1990, 1993 40 * The Regents of the University of California. All rights reserved. 41 * 42 * This code is derived from software contributed to Berkeley by 43 * the Systems Programming Group of the University of Utah Computer 44 * Science Department. 45 * 46 * Redistribution and use in source and binary forms, with or without 47 * modification, are permitted provided that the following conditions 48 * are met: 49 * 1. Redistributions of source code must retain the above copyright 50 * notice, this list of conditions and the following disclaimer. 51 * 2. Redistributions in binary form must reproduce the above copyright 52 * notice, this list of conditions and the following disclaimer in the 53 * documentation and/or other materials provided with the distribution. 54 * 3. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 71 * 72 * @(#)cd.c 8.2 (Berkeley) 11/16/93 73 */ 74 75 /* 76 * Copyright (c) 1988 University of Utah. 77 * 78 * This code is derived from software contributed to Berkeley by 79 * the Systems Programming Group of the University of Utah Computer 80 * Science Department. 81 * 82 * Redistribution and use in source and binary forms, with or without 83 * modification, are permitted provided that the following conditions 84 * are met: 85 * 1. Redistributions of source code must retain the above copyright 86 * notice, this list of conditions and the following disclaimer. 87 * 2. Redistributions in binary form must reproduce the above copyright 88 * notice, this list of conditions and the following disclaimer in the 89 * documentation and/or other materials provided with the distribution. 90 * 3. All advertising materials mentioning features or use of this software 91 * must display the following acknowledgement: 92 * This product includes software developed by the University of 93 * California, Berkeley and its contributors. 94 * 4. Neither the name of the University nor the names of its contributors 95 * may be used to endorse or promote products derived from this software 96 * without specific prior written permission. 97 * 98 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 99 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 100 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 101 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 102 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 103 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 104 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 105 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 106 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 107 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 108 * SUCH DAMAGE. 109 * 110 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 111 * 112 * @(#)cd.c 8.2 (Berkeley) 11/16/93 113 */ 114 115 /* 116 * Copyright (c) 1995 Carnegie-Mellon University. 117 * All rights reserved. 118 * 119 * Authors: Mark Holland, Jim Zelenka 120 * 121 * Permission to use, copy, modify and distribute this software and 122 * its documentation is hereby granted, provided that both the copyright 123 * notice and this permission notice appear in all copies of the 124 * software, derivative works or modified versions, and any portions 125 * thereof, and that both notices appear in supporting documentation. 126 * 127 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 128 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 129 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 130 * 131 * Carnegie Mellon requests users of this software to return to 132 * 133 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 134 * School of Computer Science 135 * Carnegie Mellon University 136 * Pittsburgh PA 15213-3890 137 * 138 * any improvements or extensions that they make and grant Carnegie the 139 * rights to redistribute these changes. 140 */ 141 142 /*********************************************************** 143 * 144 * rf_kintf.c -- the kernel interface routines for RAIDframe 145 * 146 ***********************************************************/ 147 148 #include <sys/cdefs.h> 149 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.245 2008/04/05 18:54:10 cegger Exp $"); 150 151 #include <sys/param.h> 152 #include <sys/errno.h> 153 #include <sys/pool.h> 154 #include <sys/proc.h> 155 #include <sys/queue.h> 156 #include <sys/disk.h> 157 #include <sys/device.h> 158 #include <sys/stat.h> 159 #include <sys/ioctl.h> 160 #include <sys/fcntl.h> 161 #include <sys/systm.h> 162 #include <sys/vnode.h> 163 #include <sys/disklabel.h> 164 #include <sys/conf.h> 165 #include <sys/buf.h> 166 #include <sys/bufq.h> 167 #include <sys/user.h> 168 #include <sys/reboot.h> 169 #include <sys/kauth.h> 170 171 #include <prop/proplib.h> 172 173 #include <dev/raidframe/raidframevar.h> 174 #include <dev/raidframe/raidframeio.h> 175 #include "raid.h" 176 #include "opt_raid_autoconfig.h" 177 #include "rf_raid.h" 178 #include "rf_copyback.h" 179 #include "rf_dag.h" 180 #include "rf_dagflags.h" 181 #include "rf_desc.h" 182 #include "rf_diskqueue.h" 183 #include "rf_etimer.h" 184 #include "rf_general.h" 185 #include "rf_kintf.h" 186 #include "rf_options.h" 187 #include "rf_driver.h" 188 #include "rf_parityscan.h" 189 #include "rf_threadstuff.h" 190 191 #ifdef DEBUG 192 int rf_kdebug_level = 0; 193 #define db1_printf(a) if (rf_kdebug_level > 0) printf a 194 #else /* DEBUG */ 195 #define db1_printf(a) { } 196 #endif /* DEBUG */ 197 198 static RF_Raid_t **raidPtrs; /* global raid device descriptors */ 199 200 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex) 201 202 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a 203 * spare table */ 204 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from 205 * installation process */ 206 207 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures"); 208 209 /* prototypes */ 210 static void KernelWakeupFunc(struct buf *); 211 static void InitBP(struct buf *, struct vnode *, unsigned, 212 dev_t, RF_SectorNum_t, RF_SectorCount_t, void *, void (*) (struct buf *), 213 void *, int, struct proc *); 214 static void raidinit(RF_Raid_t *); 215 216 void raidattach(int); 217 static int raid_match(struct device *, struct cfdata *, void *); 218 static void raid_attach(struct device *, struct device *, void *); 219 static int raid_detach(struct device *, int); 220 221 dev_type_open(raidopen); 222 dev_type_close(raidclose); 223 dev_type_read(raidread); 224 dev_type_write(raidwrite); 225 dev_type_ioctl(raidioctl); 226 dev_type_strategy(raidstrategy); 227 dev_type_dump(raiddump); 228 dev_type_size(raidsize); 229 230 const struct bdevsw raid_bdevsw = { 231 raidopen, raidclose, raidstrategy, raidioctl, 232 raiddump, raidsize, D_DISK 233 }; 234 235 const struct cdevsw raid_cdevsw = { 236 raidopen, raidclose, raidread, raidwrite, raidioctl, 237 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 238 }; 239 240 static struct dkdriver rf_dkdriver = { raidstrategy, minphys }; 241 242 /* XXX Not sure if the following should be replacing the raidPtrs above, 243 or if it should be used in conjunction with that... 244 */ 245 246 struct raid_softc { 247 struct device *sc_dev; 248 int sc_flags; /* flags */ 249 int sc_cflags; /* configuration flags */ 250 uint64_t sc_size; /* size of the raid device */ 251 char sc_xname[20]; /* XXX external name */ 252 struct disk sc_dkdev; /* generic disk device info */ 253 struct bufq_state *buf_queue; /* used for the device queue */ 254 }; 255 /* sc_flags */ 256 #define RAIDF_INITED 0x01 /* unit has been initialized */ 257 #define RAIDF_WLABEL 0x02 /* label area is writable */ 258 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 259 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 260 #define RAIDF_LOCKED 0x80 /* unit is locked */ 261 262 #define raidunit(x) DISKUNIT(x) 263 int numraid = 0; 264 265 extern struct cfdriver raid_cd; 266 CFATTACH_DECL_NEW(raid, sizeof(struct raid_softc), 267 raid_match, raid_attach, raid_detach, NULL); 268 269 /* 270 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 271 * Be aware that large numbers can allow the driver to consume a lot of 272 * kernel memory, especially on writes, and in degraded mode reads. 273 * 274 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 275 * a single 64K write will typically require 64K for the old data, 276 * 64K for the old parity, and 64K for the new parity, for a total 277 * of 192K (if the parity buffer is not re-used immediately). 278 * Even it if is used immediately, that's still 128K, which when multiplied 279 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 280 * 281 * Now in degraded mode, for example, a 64K read on the above setup may 282 * require data reconstruction, which will require *all* of the 4 remaining 283 * disks to participate -- 4 * 32K/disk == 128K again. 284 */ 285 286 #ifndef RAIDOUTSTANDING 287 #define RAIDOUTSTANDING 6 288 #endif 289 290 #define RAIDLABELDEV(dev) \ 291 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 292 293 /* declared here, and made public, for the benefit of KVM stuff.. */ 294 struct raid_softc *raid_softc; 295 296 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *, 297 struct disklabel *); 298 static void raidgetdisklabel(dev_t); 299 static void raidmakedisklabel(struct raid_softc *); 300 301 static int raidlock(struct raid_softc *); 302 static void raidunlock(struct raid_softc *); 303 304 static void rf_markalldirty(RF_Raid_t *); 305 static void rf_set_properties(struct raid_softc *, RF_Raid_t *); 306 307 void rf_ReconThread(struct rf_recon_req *); 308 void rf_RewriteParityThread(RF_Raid_t *raidPtr); 309 void rf_CopybackThread(RF_Raid_t *raidPtr); 310 void rf_ReconstructInPlaceThread(struct rf_recon_req *); 311 int rf_autoconfig(struct device *self); 312 void rf_buildroothack(RF_ConfigSet_t *); 313 314 RF_AutoConfig_t *rf_find_raid_components(void); 315 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *); 316 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *); 317 static int rf_reasonable_label(RF_ComponentLabel_t *); 318 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *); 319 int rf_set_autoconfig(RF_Raid_t *, int); 320 int rf_set_rootpartition(RF_Raid_t *, int); 321 void rf_release_all_vps(RF_ConfigSet_t *); 322 void rf_cleanup_config_set(RF_ConfigSet_t *); 323 int rf_have_enough_components(RF_ConfigSet_t *); 324 int rf_auto_config_set(RF_ConfigSet_t *, int *); 325 326 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not 327 allow autoconfig to take place. 328 Note that this is overridden by having 329 RAID_AUTOCONFIG as an option in the 330 kernel config file. */ 331 332 struct RF_Pools_s rf_pools; 333 334 void 335 raidattach(int num) 336 { 337 int raidID; 338 int i, rc; 339 340 #ifdef DEBUG 341 printf("raidattach: Asked for %d units\n", num); 342 #endif 343 344 if (num <= 0) { 345 #ifdef DIAGNOSTIC 346 panic("raidattach: count <= 0"); 347 #endif 348 return; 349 } 350 /* This is where all the initialization stuff gets done. */ 351 352 numraid = num; 353 354 /* Make some space for requested number of units... */ 355 356 RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **)); 357 if (raidPtrs == NULL) { 358 panic("raidPtrs is NULL!!"); 359 } 360 361 rf_mutex_init(&rf_sparet_wait_mutex); 362 363 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 364 365 for (i = 0; i < num; i++) 366 raidPtrs[i] = NULL; 367 rc = rf_BootRaidframe(); 368 if (rc == 0) 369 aprint_normal("Kernelized RAIDframe activated\n"); 370 else 371 panic("Serious error booting RAID!!"); 372 373 /* put together some datastructures like the CCD device does.. This 374 * lets us lock the device and what-not when it gets opened. */ 375 376 raid_softc = (struct raid_softc *) 377 malloc(num * sizeof(struct raid_softc), 378 M_RAIDFRAME, M_NOWAIT); 379 if (raid_softc == NULL) { 380 aprint_error("WARNING: no memory for RAIDframe driver\n"); 381 return; 382 } 383 384 memset(raid_softc, 0, num * sizeof(struct raid_softc)); 385 386 for (raidID = 0; raidID < num; raidID++) { 387 bufq_alloc(&raid_softc[raidID].buf_queue, "fcfs", 0); 388 389 RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t), 390 (RF_Raid_t *)); 391 if (raidPtrs[raidID] == NULL) { 392 aprint_error("WARNING: raidPtrs[%d] is NULL\n", raidID); 393 numraid = raidID; 394 return; 395 } 396 } 397 398 if (config_cfattach_attach(raid_cd.cd_name, &raid_ca)) { 399 aprint_error("raidattach: config_cfattach_attach failed?\n"); 400 } 401 402 #ifdef RAID_AUTOCONFIG 403 raidautoconfig = 1; 404 #endif 405 406 /* 407 * Register a finalizer which will be used to auto-config RAID 408 * sets once all real hardware devices have been found. 409 */ 410 if (config_finalize_register(NULL, rf_autoconfig) != 0) 411 aprint_error("WARNING: unable to register RAIDframe finalizer\n"); 412 } 413 414 int 415 rf_autoconfig(struct device *self) 416 { 417 RF_AutoConfig_t *ac_list; 418 RF_ConfigSet_t *config_sets; 419 420 if (raidautoconfig == 0) 421 return (0); 422 423 /* XXX This code can only be run once. */ 424 raidautoconfig = 0; 425 426 /* 1. locate all RAID components on the system */ 427 #ifdef DEBUG 428 printf("Searching for RAID components...\n"); 429 #endif 430 ac_list = rf_find_raid_components(); 431 432 /* 2. Sort them into their respective sets. */ 433 config_sets = rf_create_auto_sets(ac_list); 434 435 /* 436 * 3. Evaluate each set andconfigure the valid ones. 437 * This gets done in rf_buildroothack(). 438 */ 439 rf_buildroothack(config_sets); 440 441 return 1; 442 } 443 444 void 445 rf_buildroothack(RF_ConfigSet_t *config_sets) 446 { 447 RF_ConfigSet_t *cset; 448 RF_ConfigSet_t *next_cset; 449 int retcode; 450 int raidID; 451 int rootID; 452 int col; 453 int num_root; 454 char *devname; 455 456 rootID = 0; 457 num_root = 0; 458 cset = config_sets; 459 while(cset != NULL ) { 460 next_cset = cset->next; 461 if (rf_have_enough_components(cset) && 462 cset->ac->clabel->autoconfigure==1) { 463 retcode = rf_auto_config_set(cset,&raidID); 464 if (!retcode) { 465 #ifdef DEBUG 466 printf("raid%d: configured ok\n", raidID); 467 #endif 468 if (cset->rootable) { 469 rootID = raidID; 470 num_root++; 471 } 472 } else { 473 /* The autoconfig didn't work :( */ 474 #ifdef DEBUG 475 printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 476 #endif 477 rf_release_all_vps(cset); 478 } 479 } else { 480 /* we're not autoconfiguring this set... 481 release the associated resources */ 482 rf_release_all_vps(cset); 483 } 484 /* cleanup */ 485 rf_cleanup_config_set(cset); 486 cset = next_cset; 487 } 488 489 /* if the user has specified what the root device should be 490 then we don't touch booted_device or boothowto... */ 491 492 if (rootspec != NULL) 493 return; 494 495 /* we found something bootable... */ 496 497 if (num_root == 1) { 498 booted_device = raid_softc[rootID].sc_dev; 499 } else if (num_root > 1) { 500 501 /* 502 * Maybe the MD code can help. If it cannot, then 503 * setroot() will discover that we have no 504 * booted_device and will ask the user if nothing was 505 * hardwired in the kernel config file 506 */ 507 508 if (booted_device == NULL) 509 cpu_rootconf(); 510 if (booted_device == NULL) 511 return; 512 513 num_root = 0; 514 for (raidID = 0; raidID < numraid; raidID++) { 515 if (raidPtrs[raidID]->valid == 0) 516 continue; 517 518 if (raidPtrs[raidID]->root_partition == 0) 519 continue; 520 521 for (col = 0; col < raidPtrs[raidID]->numCol; col++) { 522 devname = raidPtrs[raidID]->Disks[col].devname; 523 devname += sizeof("/dev/") - 1; 524 if (strncmp(devname, device_xname(booted_device), 525 strlen(device_xname(booted_device))) != 0) 526 continue; 527 #ifdef DEBUG 528 printf("raid%d includes boot device %s\n", 529 raidID, devname); 530 #endif 531 num_root++; 532 rootID = raidID; 533 } 534 } 535 536 if (num_root == 1) { 537 booted_device = raid_softc[rootID].sc_dev; 538 } else { 539 /* we can't guess.. require the user to answer... */ 540 boothowto |= RB_ASKNAME; 541 } 542 } 543 } 544 545 546 int 547 raidsize(dev_t dev) 548 { 549 struct raid_softc *rs; 550 struct disklabel *lp; 551 int part, unit, omask, size; 552 553 unit = raidunit(dev); 554 if (unit >= numraid) 555 return (-1); 556 rs = &raid_softc[unit]; 557 558 if ((rs->sc_flags & RAIDF_INITED) == 0) 559 return (-1); 560 561 part = DISKPART(dev); 562 omask = rs->sc_dkdev.dk_openmask & (1 << part); 563 lp = rs->sc_dkdev.dk_label; 564 565 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curlwp)) 566 return (-1); 567 568 if (lp->d_partitions[part].p_fstype != FS_SWAP) 569 size = -1; 570 else 571 size = lp->d_partitions[part].p_size * 572 (lp->d_secsize / DEV_BSIZE); 573 574 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curlwp)) 575 return (-1); 576 577 return (size); 578 579 } 580 581 int 582 raiddump(dev_t dev, daddr_t blkno, void *va, size_t size) 583 { 584 int unit = raidunit(dev); 585 struct raid_softc *rs; 586 const struct bdevsw *bdev; 587 struct disklabel *lp; 588 RF_Raid_t *raidPtr; 589 daddr_t offset; 590 int part, c, sparecol, j, scol, dumpto; 591 int error = 0; 592 593 if (unit >= numraid) 594 return (ENXIO); 595 596 rs = &raid_softc[unit]; 597 raidPtr = raidPtrs[unit]; 598 599 if ((rs->sc_flags & RAIDF_INITED) == 0) 600 return ENXIO; 601 602 /* we only support dumping to RAID 1 sets */ 603 if (raidPtr->Layout.numDataCol != 1 || 604 raidPtr->Layout.numParityCol != 1) 605 return EINVAL; 606 607 608 if ((error = raidlock(rs)) != 0) 609 return error; 610 611 if (size % DEV_BSIZE != 0) { 612 error = EINVAL; 613 goto out; 614 } 615 616 if (blkno + size / DEV_BSIZE > rs->sc_size) { 617 printf("%s: blkno (%" PRIu64 ") + size / DEV_BSIZE (%zu) > " 618 "sc->sc_size (%" PRIu64 ")\n", __func__, blkno, 619 size / DEV_BSIZE, rs->sc_size); 620 error = EINVAL; 621 goto out; 622 } 623 624 part = DISKPART(dev); 625 lp = rs->sc_dkdev.dk_label; 626 offset = lp->d_partitions[part].p_offset + RF_PROTECTED_SECTORS; 627 628 /* figure out what device is alive.. */ 629 630 /* 631 Look for a component to dump to. The preference for the 632 component to dump to is as follows: 633 1) the master 634 2) a used_spare of the master 635 3) the slave 636 4) a used_spare of the slave 637 */ 638 639 dumpto = -1; 640 for (c = 0; c < raidPtr->numCol; c++) { 641 if (raidPtr->Disks[c].status == rf_ds_optimal) { 642 /* this might be the one */ 643 dumpto = c; 644 break; 645 } 646 } 647 648 /* 649 At this point we have possibly selected a live master or a 650 live slave. We now check to see if there is a spared 651 master (or a spared slave), if we didn't find a live master 652 or a live slave. 653 */ 654 655 for (c = 0; c < raidPtr->numSpare; c++) { 656 sparecol = raidPtr->numCol + c; 657 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 658 /* How about this one? */ 659 scol = -1; 660 for(j=0;j<raidPtr->numCol;j++) { 661 if (raidPtr->Disks[j].spareCol == sparecol) { 662 scol = j; 663 break; 664 } 665 } 666 if (scol == 0) { 667 /* 668 We must have found a spared master! 669 We'll take that over anything else 670 found so far. (We couldn't have 671 found a real master before, since 672 this is a used spare, and it's 673 saying that it's replacing the 674 master.) On reboot (with 675 autoconfiguration turned on) 676 sparecol will become the 1st 677 component (component0) of this set. 678 */ 679 dumpto = sparecol; 680 break; 681 } else if (scol != -1) { 682 /* 683 Must be a spared slave. We'll dump 684 to that if we havn't found anything 685 else so far. 686 */ 687 if (dumpto == -1) 688 dumpto = sparecol; 689 } 690 } 691 } 692 693 if (dumpto == -1) { 694 /* we couldn't find any live components to dump to!?!? 695 */ 696 error = EINVAL; 697 goto out; 698 } 699 700 bdev = bdevsw_lookup(raidPtr->Disks[dumpto].dev); 701 702 /* 703 Note that blkno is relative to this particular partition. 704 By adding the offset of this partition in the RAID 705 set, and also adding RF_PROTECTED_SECTORS, we get a 706 value that is relative to the partition used for the 707 underlying component. 708 */ 709 710 error = (*bdev->d_dump)(raidPtr->Disks[dumpto].dev, 711 blkno + offset, va, size); 712 713 out: 714 raidunlock(rs); 715 716 return error; 717 } 718 /* ARGSUSED */ 719 int 720 raidopen(dev_t dev, int flags, int fmt, 721 struct lwp *l) 722 { 723 int unit = raidunit(dev); 724 struct raid_softc *rs; 725 struct disklabel *lp; 726 int part, pmask; 727 int error = 0; 728 729 if (unit >= numraid) 730 return (ENXIO); 731 rs = &raid_softc[unit]; 732 733 if ((error = raidlock(rs)) != 0) 734 return (error); 735 lp = rs->sc_dkdev.dk_label; 736 737 part = DISKPART(dev); 738 739 /* 740 * If there are wedges, and this is not RAW_PART, then we 741 * need to fail. 742 */ 743 if (rs->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) { 744 error = EBUSY; 745 goto bad; 746 } 747 pmask = (1 << part); 748 749 if ((rs->sc_flags & RAIDF_INITED) && 750 (rs->sc_dkdev.dk_openmask == 0)) 751 raidgetdisklabel(dev); 752 753 /* make sure that this partition exists */ 754 755 if (part != RAW_PART) { 756 if (((rs->sc_flags & RAIDF_INITED) == 0) || 757 ((part >= lp->d_npartitions) || 758 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 759 error = ENXIO; 760 goto bad; 761 } 762 } 763 /* Prevent this unit from being unconfigured while open. */ 764 switch (fmt) { 765 case S_IFCHR: 766 rs->sc_dkdev.dk_copenmask |= pmask; 767 break; 768 769 case S_IFBLK: 770 rs->sc_dkdev.dk_bopenmask |= pmask; 771 break; 772 } 773 774 if ((rs->sc_dkdev.dk_openmask == 0) && 775 ((rs->sc_flags & RAIDF_INITED) != 0)) { 776 /* First one... mark things as dirty... Note that we *MUST* 777 have done a configure before this. I DO NOT WANT TO BE 778 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 779 THAT THEY BELONG TOGETHER!!!!! */ 780 /* XXX should check to see if we're only open for reading 781 here... If so, we needn't do this, but then need some 782 other way of keeping track of what's happened.. */ 783 784 rf_markalldirty( raidPtrs[unit] ); 785 } 786 787 788 rs->sc_dkdev.dk_openmask = 789 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 790 791 bad: 792 raidunlock(rs); 793 794 return (error); 795 796 797 } 798 /* ARGSUSED */ 799 int 800 raidclose(dev_t dev, int flags, int fmt, struct lwp *l) 801 { 802 int unit = raidunit(dev); 803 struct cfdata *cf; 804 struct raid_softc *rs; 805 int error = 0; 806 int part; 807 808 if (unit >= numraid) 809 return (ENXIO); 810 rs = &raid_softc[unit]; 811 812 if ((error = raidlock(rs)) != 0) 813 return (error); 814 815 part = DISKPART(dev); 816 817 /* ...that much closer to allowing unconfiguration... */ 818 switch (fmt) { 819 case S_IFCHR: 820 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 821 break; 822 823 case S_IFBLK: 824 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 825 break; 826 } 827 rs->sc_dkdev.dk_openmask = 828 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 829 830 if ((rs->sc_dkdev.dk_openmask == 0) && 831 ((rs->sc_flags & RAIDF_INITED) != 0)) { 832 /* Last one... device is not unconfigured yet. 833 Device shutdown has taken care of setting the 834 clean bits if RAIDF_INITED is not set 835 mark things as clean... */ 836 837 rf_update_component_labels(raidPtrs[unit], 838 RF_FINAL_COMPONENT_UPDATE); 839 if (doing_shutdown) { 840 /* last one, and we're going down, so 841 lights out for this RAID set too. */ 842 error = rf_Shutdown(raidPtrs[unit]); 843 844 /* It's no longer initialized... */ 845 rs->sc_flags &= ~RAIDF_INITED; 846 847 /* detach the device */ 848 849 cf = device_cfdata(rs->sc_dev); 850 error = config_detach(rs->sc_dev, DETACH_QUIET); 851 free(cf, M_RAIDFRAME); 852 853 /* Detach the disk. */ 854 disk_detach(&rs->sc_dkdev); 855 disk_destroy(&rs->sc_dkdev); 856 } 857 } 858 859 raidunlock(rs); 860 return (0); 861 862 } 863 864 void 865 raidstrategy(struct buf *bp) 866 { 867 int s; 868 869 unsigned int raidID = raidunit(bp->b_dev); 870 RF_Raid_t *raidPtr; 871 struct raid_softc *rs = &raid_softc[raidID]; 872 int wlabel; 873 874 if ((rs->sc_flags & RAIDF_INITED) ==0) { 875 bp->b_error = ENXIO; 876 goto done; 877 } 878 if (raidID >= numraid || !raidPtrs[raidID]) { 879 bp->b_error = ENODEV; 880 goto done; 881 } 882 raidPtr = raidPtrs[raidID]; 883 if (!raidPtr->valid) { 884 bp->b_error = ENODEV; 885 goto done; 886 } 887 if (bp->b_bcount == 0) { 888 db1_printf(("b_bcount is zero..\n")); 889 goto done; 890 } 891 892 /* 893 * Do bounds checking and adjust transfer. If there's an 894 * error, the bounds check will flag that for us. 895 */ 896 897 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 898 if (DISKPART(bp->b_dev) == RAW_PART) { 899 uint64_t size; /* device size in DEV_BSIZE unit */ 900 901 if (raidPtr->logBytesPerSector > DEV_BSHIFT) { 902 size = raidPtr->totalSectors << 903 (raidPtr->logBytesPerSector - DEV_BSHIFT); 904 } else { 905 size = raidPtr->totalSectors >> 906 (DEV_BSHIFT - raidPtr->logBytesPerSector); 907 } 908 if (bounds_check_with_mediasize(bp, DEV_BSIZE, size) <= 0) { 909 goto done; 910 } 911 } else { 912 if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) { 913 db1_printf(("Bounds check failed!!:%d %d\n", 914 (int) bp->b_blkno, (int) wlabel)); 915 goto done; 916 } 917 } 918 s = splbio(); 919 920 bp->b_resid = 0; 921 922 /* stuff it onto our queue */ 923 BUFQ_PUT(rs->buf_queue, bp); 924 925 /* scheduled the IO to happen at the next convenient time */ 926 wakeup(&(raidPtrs[raidID]->iodone)); 927 928 splx(s); 929 return; 930 931 done: 932 bp->b_resid = bp->b_bcount; 933 biodone(bp); 934 } 935 /* ARGSUSED */ 936 int 937 raidread(dev_t dev, struct uio *uio, int flags) 938 { 939 int unit = raidunit(dev); 940 struct raid_softc *rs; 941 942 if (unit >= numraid) 943 return (ENXIO); 944 rs = &raid_softc[unit]; 945 946 if ((rs->sc_flags & RAIDF_INITED) == 0) 947 return (ENXIO); 948 949 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio)); 950 951 } 952 /* ARGSUSED */ 953 int 954 raidwrite(dev_t dev, struct uio *uio, int flags) 955 { 956 int unit = raidunit(dev); 957 struct raid_softc *rs; 958 959 if (unit >= numraid) 960 return (ENXIO); 961 rs = &raid_softc[unit]; 962 963 if ((rs->sc_flags & RAIDF_INITED) == 0) 964 return (ENXIO); 965 966 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio)); 967 968 } 969 970 int 971 raidioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 972 { 973 int unit = raidunit(dev); 974 int error = 0; 975 int part, pmask; 976 struct cfdata *cf; 977 struct raid_softc *rs; 978 RF_Config_t *k_cfg, *u_cfg; 979 RF_Raid_t *raidPtr; 980 RF_RaidDisk_t *diskPtr; 981 RF_AccTotals_t *totals; 982 RF_DeviceConfig_t *d_cfg, **ucfgp; 983 u_char *specific_buf; 984 int retcode = 0; 985 int column; 986 int raidid; 987 struct rf_recon_req *rrcopy, *rr; 988 RF_ComponentLabel_t *clabel; 989 RF_ComponentLabel_t *ci_label; 990 RF_ComponentLabel_t **clabel_ptr; 991 RF_SingleComponent_t *sparePtr,*componentPtr; 992 RF_SingleComponent_t component; 993 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 994 int i, j, d; 995 #ifdef __HAVE_OLD_DISKLABEL 996 struct disklabel newlabel; 997 #endif 998 struct dkwedge_info *dkw; 999 1000 if (unit >= numraid) 1001 return (ENXIO); 1002 rs = &raid_softc[unit]; 1003 raidPtr = raidPtrs[unit]; 1004 1005 db1_printf(("raidioctl: %d %d %d %d\n", (int) dev, 1006 (int) DISKPART(dev), (int) unit, (int) cmd)); 1007 1008 /* Must be open for writes for these commands... */ 1009 switch (cmd) { 1010 #ifdef DIOCGSECTORSIZE 1011 case DIOCGSECTORSIZE: 1012 *(u_int *)data = raidPtr->bytesPerSector; 1013 return 0; 1014 case DIOCGMEDIASIZE: 1015 *(off_t *)data = 1016 (off_t)raidPtr->totalSectors * raidPtr->bytesPerSector; 1017 return 0; 1018 #endif 1019 case DIOCSDINFO: 1020 case DIOCWDINFO: 1021 #ifdef __HAVE_OLD_DISKLABEL 1022 case ODIOCWDINFO: 1023 case ODIOCSDINFO: 1024 #endif 1025 case DIOCWLABEL: 1026 case DIOCAWEDGE: 1027 case DIOCDWEDGE: 1028 if ((flag & FWRITE) == 0) 1029 return (EBADF); 1030 } 1031 1032 /* Must be initialized for these... */ 1033 switch (cmd) { 1034 case DIOCGDINFO: 1035 case DIOCSDINFO: 1036 case DIOCWDINFO: 1037 #ifdef __HAVE_OLD_DISKLABEL 1038 case ODIOCGDINFO: 1039 case ODIOCWDINFO: 1040 case ODIOCSDINFO: 1041 case ODIOCGDEFLABEL: 1042 #endif 1043 case DIOCGPART: 1044 case DIOCWLABEL: 1045 case DIOCGDEFLABEL: 1046 case DIOCAWEDGE: 1047 case DIOCDWEDGE: 1048 case DIOCLWEDGES: 1049 case RAIDFRAME_SHUTDOWN: 1050 case RAIDFRAME_REWRITEPARITY: 1051 case RAIDFRAME_GET_INFO: 1052 case RAIDFRAME_RESET_ACCTOTALS: 1053 case RAIDFRAME_GET_ACCTOTALS: 1054 case RAIDFRAME_KEEP_ACCTOTALS: 1055 case RAIDFRAME_GET_SIZE: 1056 case RAIDFRAME_FAIL_DISK: 1057 case RAIDFRAME_COPYBACK: 1058 case RAIDFRAME_CHECK_RECON_STATUS: 1059 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1060 case RAIDFRAME_GET_COMPONENT_LABEL: 1061 case RAIDFRAME_SET_COMPONENT_LABEL: 1062 case RAIDFRAME_ADD_HOT_SPARE: 1063 case RAIDFRAME_REMOVE_HOT_SPARE: 1064 case RAIDFRAME_INIT_LABELS: 1065 case RAIDFRAME_REBUILD_IN_PLACE: 1066 case RAIDFRAME_CHECK_PARITY: 1067 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1068 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1069 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1070 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1071 case RAIDFRAME_SET_AUTOCONFIG: 1072 case RAIDFRAME_SET_ROOT: 1073 case RAIDFRAME_DELETE_COMPONENT: 1074 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1075 if ((rs->sc_flags & RAIDF_INITED) == 0) 1076 return (ENXIO); 1077 } 1078 1079 switch (cmd) { 1080 1081 /* configure the system */ 1082 case RAIDFRAME_CONFIGURE: 1083 1084 if (raidPtr->valid) { 1085 /* There is a valid RAID set running on this unit! */ 1086 printf("raid%d: Device already configured!\n",unit); 1087 return(EINVAL); 1088 } 1089 1090 /* copy-in the configuration information */ 1091 /* data points to a pointer to the configuration structure */ 1092 1093 u_cfg = *((RF_Config_t **) data); 1094 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *)); 1095 if (k_cfg == NULL) { 1096 return (ENOMEM); 1097 } 1098 retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t)); 1099 if (retcode) { 1100 RF_Free(k_cfg, sizeof(RF_Config_t)); 1101 db1_printf(("rf_ioctl: retcode=%d copyin.1\n", 1102 retcode)); 1103 return (retcode); 1104 } 1105 /* allocate a buffer for the layout-specific data, and copy it 1106 * in */ 1107 if (k_cfg->layoutSpecificSize) { 1108 if (k_cfg->layoutSpecificSize > 10000) { 1109 /* sanity check */ 1110 RF_Free(k_cfg, sizeof(RF_Config_t)); 1111 return (EINVAL); 1112 } 1113 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize, 1114 (u_char *)); 1115 if (specific_buf == NULL) { 1116 RF_Free(k_cfg, sizeof(RF_Config_t)); 1117 return (ENOMEM); 1118 } 1119 retcode = copyin(k_cfg->layoutSpecific, specific_buf, 1120 k_cfg->layoutSpecificSize); 1121 if (retcode) { 1122 RF_Free(k_cfg, sizeof(RF_Config_t)); 1123 RF_Free(specific_buf, 1124 k_cfg->layoutSpecificSize); 1125 db1_printf(("rf_ioctl: retcode=%d copyin.2\n", 1126 retcode)); 1127 return (retcode); 1128 } 1129 } else 1130 specific_buf = NULL; 1131 k_cfg->layoutSpecific = specific_buf; 1132 1133 /* should do some kind of sanity check on the configuration. 1134 * Store the sum of all the bytes in the last byte? */ 1135 1136 /* configure the system */ 1137 1138 /* 1139 * Clear the entire RAID descriptor, just to make sure 1140 * there is no stale data left in the case of a 1141 * reconfiguration 1142 */ 1143 memset((char *) raidPtr, 0, sizeof(RF_Raid_t)); 1144 raidPtr->raidid = unit; 1145 1146 retcode = rf_Configure(raidPtr, k_cfg, NULL); 1147 1148 if (retcode == 0) { 1149 1150 /* allow this many simultaneous IO's to 1151 this RAID device */ 1152 raidPtr->openings = RAIDOUTSTANDING; 1153 1154 raidinit(raidPtr); 1155 rf_markalldirty(raidPtr); 1156 } 1157 /* free the buffers. No return code here. */ 1158 if (k_cfg->layoutSpecificSize) { 1159 RF_Free(specific_buf, k_cfg->layoutSpecificSize); 1160 } 1161 RF_Free(k_cfg, sizeof(RF_Config_t)); 1162 1163 return (retcode); 1164 1165 /* shutdown the system */ 1166 case RAIDFRAME_SHUTDOWN: 1167 1168 if ((error = raidlock(rs)) != 0) 1169 return (error); 1170 1171 /* 1172 * If somebody has a partition mounted, we shouldn't 1173 * shutdown. 1174 */ 1175 1176 part = DISKPART(dev); 1177 pmask = (1 << part); 1178 if ((rs->sc_dkdev.dk_openmask & ~pmask) || 1179 ((rs->sc_dkdev.dk_bopenmask & pmask) && 1180 (rs->sc_dkdev.dk_copenmask & pmask))) { 1181 raidunlock(rs); 1182 return (EBUSY); 1183 } 1184 1185 retcode = rf_Shutdown(raidPtr); 1186 1187 /* It's no longer initialized... */ 1188 rs->sc_flags &= ~RAIDF_INITED; 1189 1190 /* free the pseudo device attach bits */ 1191 1192 cf = device_cfdata(rs->sc_dev); 1193 /* XXX this causes us to not return any errors 1194 from the above call to rf_Shutdown() */ 1195 retcode = config_detach(rs->sc_dev, DETACH_QUIET); 1196 free(cf, M_RAIDFRAME); 1197 1198 /* Detach the disk. */ 1199 disk_detach(&rs->sc_dkdev); 1200 disk_destroy(&rs->sc_dkdev); 1201 1202 raidunlock(rs); 1203 1204 return (retcode); 1205 case RAIDFRAME_GET_COMPONENT_LABEL: 1206 clabel_ptr = (RF_ComponentLabel_t **) data; 1207 /* need to read the component label for the disk indicated 1208 by row,column in clabel */ 1209 1210 /* For practice, let's get it directly fromdisk, rather 1211 than from the in-core copy */ 1212 RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ), 1213 (RF_ComponentLabel_t *)); 1214 if (clabel == NULL) 1215 return (ENOMEM); 1216 1217 retcode = copyin( *clabel_ptr, clabel, 1218 sizeof(RF_ComponentLabel_t)); 1219 1220 if (retcode) { 1221 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1222 return(retcode); 1223 } 1224 1225 clabel->row = 0; /* Don't allow looking at anything else.*/ 1226 1227 column = clabel->column; 1228 1229 if ((column < 0) || (column >= raidPtr->numCol + 1230 raidPtr->numSpare)) { 1231 RF_Free( clabel, sizeof(RF_ComponentLabel_t)); 1232 return(EINVAL); 1233 } 1234 1235 retcode = raidread_component_label(raidPtr->Disks[column].dev, 1236 raidPtr->raid_cinfo[column].ci_vp, 1237 clabel ); 1238 1239 if (retcode == 0) { 1240 retcode = copyout(clabel, *clabel_ptr, 1241 sizeof(RF_ComponentLabel_t)); 1242 } 1243 RF_Free(clabel, sizeof(RF_ComponentLabel_t)); 1244 return (retcode); 1245 1246 case RAIDFRAME_SET_COMPONENT_LABEL: 1247 clabel = (RF_ComponentLabel_t *) data; 1248 1249 /* XXX check the label for valid stuff... */ 1250 /* Note that some things *should not* get modified -- 1251 the user should be re-initing the labels instead of 1252 trying to patch things. 1253 */ 1254 1255 raidid = raidPtr->raidid; 1256 #ifdef DEBUG 1257 printf("raid%d: Got component label:\n", raidid); 1258 printf("raid%d: Version: %d\n", raidid, clabel->version); 1259 printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number); 1260 printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter); 1261 printf("raid%d: Column: %d\n", raidid, clabel->column); 1262 printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns); 1263 printf("raid%d: Clean: %d\n", raidid, clabel->clean); 1264 printf("raid%d: Status: %d\n", raidid, clabel->status); 1265 #endif 1266 clabel->row = 0; 1267 column = clabel->column; 1268 1269 if ((column < 0) || (column >= raidPtr->numCol)) { 1270 return(EINVAL); 1271 } 1272 1273 /* XXX this isn't allowed to do anything for now :-) */ 1274 1275 /* XXX and before it is, we need to fill in the rest 1276 of the fields!?!?!?! */ 1277 #if 0 1278 raidwrite_component_label( 1279 raidPtr->Disks[column].dev, 1280 raidPtr->raid_cinfo[column].ci_vp, 1281 clabel ); 1282 #endif 1283 return (0); 1284 1285 case RAIDFRAME_INIT_LABELS: 1286 clabel = (RF_ComponentLabel_t *) data; 1287 /* 1288 we only want the serial number from 1289 the above. We get all the rest of the information 1290 from the config that was used to create this RAID 1291 set. 1292 */ 1293 1294 raidPtr->serial_number = clabel->serial_number; 1295 1296 RF_Malloc(ci_label, sizeof(RF_ComponentLabel_t), 1297 (RF_ComponentLabel_t *)); 1298 if (ci_label == NULL) 1299 return (ENOMEM); 1300 1301 raid_init_component_label(raidPtr, ci_label); 1302 ci_label->serial_number = clabel->serial_number; 1303 ci_label->row = 0; /* we dont' pretend to support more */ 1304 1305 for(column=0;column<raidPtr->numCol;column++) { 1306 diskPtr = &raidPtr->Disks[column]; 1307 if (!RF_DEAD_DISK(diskPtr->status)) { 1308 ci_label->partitionSize = diskPtr->partitionSize; 1309 ci_label->column = column; 1310 raidwrite_component_label( 1311 raidPtr->Disks[column].dev, 1312 raidPtr->raid_cinfo[column].ci_vp, 1313 ci_label ); 1314 } 1315 } 1316 RF_Free(ci_label, sizeof(RF_ComponentLabel_t)); 1317 1318 return (retcode); 1319 case RAIDFRAME_SET_AUTOCONFIG: 1320 d = rf_set_autoconfig(raidPtr, *(int *) data); 1321 printf("raid%d: New autoconfig value is: %d\n", 1322 raidPtr->raidid, d); 1323 *(int *) data = d; 1324 return (retcode); 1325 1326 case RAIDFRAME_SET_ROOT: 1327 d = rf_set_rootpartition(raidPtr, *(int *) data); 1328 printf("raid%d: New rootpartition value is: %d\n", 1329 raidPtr->raidid, d); 1330 *(int *) data = d; 1331 return (retcode); 1332 1333 /* initialize all parity */ 1334 case RAIDFRAME_REWRITEPARITY: 1335 1336 if (raidPtr->Layout.map->faultsTolerated == 0) { 1337 /* Parity for RAID 0 is trivially correct */ 1338 raidPtr->parity_good = RF_RAID_CLEAN; 1339 return(0); 1340 } 1341 1342 if (raidPtr->parity_rewrite_in_progress == 1) { 1343 /* Re-write is already in progress! */ 1344 return(EINVAL); 1345 } 1346 1347 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1348 rf_RewriteParityThread, 1349 raidPtr,"raid_parity"); 1350 return (retcode); 1351 1352 1353 case RAIDFRAME_ADD_HOT_SPARE: 1354 sparePtr = (RF_SingleComponent_t *) data; 1355 memcpy( &component, sparePtr, sizeof(RF_SingleComponent_t)); 1356 retcode = rf_add_hot_spare(raidPtr, &component); 1357 return(retcode); 1358 1359 case RAIDFRAME_REMOVE_HOT_SPARE: 1360 return(retcode); 1361 1362 case RAIDFRAME_DELETE_COMPONENT: 1363 componentPtr = (RF_SingleComponent_t *)data; 1364 memcpy( &component, componentPtr, 1365 sizeof(RF_SingleComponent_t)); 1366 retcode = rf_delete_component(raidPtr, &component); 1367 return(retcode); 1368 1369 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1370 componentPtr = (RF_SingleComponent_t *)data; 1371 memcpy( &component, componentPtr, 1372 sizeof(RF_SingleComponent_t)); 1373 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1374 return(retcode); 1375 1376 case RAIDFRAME_REBUILD_IN_PLACE: 1377 1378 if (raidPtr->Layout.map->faultsTolerated == 0) { 1379 /* Can't do this on a RAID 0!! */ 1380 return(EINVAL); 1381 } 1382 1383 if (raidPtr->recon_in_progress == 1) { 1384 /* a reconstruct is already in progress! */ 1385 return(EINVAL); 1386 } 1387 1388 componentPtr = (RF_SingleComponent_t *) data; 1389 memcpy( &component, componentPtr, 1390 sizeof(RF_SingleComponent_t)); 1391 component.row = 0; /* we don't support any more */ 1392 column = component.column; 1393 1394 if ((column < 0) || (column >= raidPtr->numCol)) { 1395 return(EINVAL); 1396 } 1397 1398 RF_LOCK_MUTEX(raidPtr->mutex); 1399 if ((raidPtr->Disks[column].status == rf_ds_optimal) && 1400 (raidPtr->numFailures > 0)) { 1401 /* XXX 0 above shouldn't be constant!!! */ 1402 /* some component other than this has failed. 1403 Let's not make things worse than they already 1404 are... */ 1405 printf("raid%d: Unable to reconstruct to disk at:\n", 1406 raidPtr->raidid); 1407 printf("raid%d: Col: %d Too many failures.\n", 1408 raidPtr->raidid, column); 1409 RF_UNLOCK_MUTEX(raidPtr->mutex); 1410 return (EINVAL); 1411 } 1412 if (raidPtr->Disks[column].status == 1413 rf_ds_reconstructing) { 1414 printf("raid%d: Unable to reconstruct to disk at:\n", 1415 raidPtr->raidid); 1416 printf("raid%d: Col: %d Reconstruction already occuring!\n", raidPtr->raidid, column); 1417 1418 RF_UNLOCK_MUTEX(raidPtr->mutex); 1419 return (EINVAL); 1420 } 1421 if (raidPtr->Disks[column].status == rf_ds_spared) { 1422 RF_UNLOCK_MUTEX(raidPtr->mutex); 1423 return (EINVAL); 1424 } 1425 RF_UNLOCK_MUTEX(raidPtr->mutex); 1426 1427 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1428 if (rrcopy == NULL) 1429 return(ENOMEM); 1430 1431 rrcopy->raidPtr = (void *) raidPtr; 1432 rrcopy->col = column; 1433 1434 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1435 rf_ReconstructInPlaceThread, 1436 rrcopy,"raid_reconip"); 1437 return(retcode); 1438 1439 case RAIDFRAME_GET_INFO: 1440 if (!raidPtr->valid) 1441 return (ENODEV); 1442 ucfgp = (RF_DeviceConfig_t **) data; 1443 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1444 (RF_DeviceConfig_t *)); 1445 if (d_cfg == NULL) 1446 return (ENOMEM); 1447 d_cfg->rows = 1; /* there is only 1 row now */ 1448 d_cfg->cols = raidPtr->numCol; 1449 d_cfg->ndevs = raidPtr->numCol; 1450 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1451 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1452 return (ENOMEM); 1453 } 1454 d_cfg->nspares = raidPtr->numSpare; 1455 if (d_cfg->nspares >= RF_MAX_DISKS) { 1456 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1457 return (ENOMEM); 1458 } 1459 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1460 d = 0; 1461 for (j = 0; j < d_cfg->cols; j++) { 1462 d_cfg->devs[d] = raidPtr->Disks[j]; 1463 d++; 1464 } 1465 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1466 d_cfg->spares[i] = raidPtr->Disks[j]; 1467 } 1468 retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t)); 1469 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1470 1471 return (retcode); 1472 1473 case RAIDFRAME_CHECK_PARITY: 1474 *(int *) data = raidPtr->parity_good; 1475 return (0); 1476 1477 case RAIDFRAME_RESET_ACCTOTALS: 1478 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1479 return (0); 1480 1481 case RAIDFRAME_GET_ACCTOTALS: 1482 totals = (RF_AccTotals_t *) data; 1483 *totals = raidPtr->acc_totals; 1484 return (0); 1485 1486 case RAIDFRAME_KEEP_ACCTOTALS: 1487 raidPtr->keep_acc_totals = *(int *)data; 1488 return (0); 1489 1490 case RAIDFRAME_GET_SIZE: 1491 *(int *) data = raidPtr->totalSectors; 1492 return (0); 1493 1494 /* fail a disk & optionally start reconstruction */ 1495 case RAIDFRAME_FAIL_DISK: 1496 1497 if (raidPtr->Layout.map->faultsTolerated == 0) { 1498 /* Can't do this on a RAID 0!! */ 1499 return(EINVAL); 1500 } 1501 1502 rr = (struct rf_recon_req *) data; 1503 rr->row = 0; 1504 if (rr->col < 0 || rr->col >= raidPtr->numCol) 1505 return (EINVAL); 1506 1507 1508 RF_LOCK_MUTEX(raidPtr->mutex); 1509 if (raidPtr->status == rf_rs_reconstructing) { 1510 /* you can't fail a disk while we're reconstructing! */ 1511 /* XXX wrong for RAID6 */ 1512 RF_UNLOCK_MUTEX(raidPtr->mutex); 1513 return (EINVAL); 1514 } 1515 if ((raidPtr->Disks[rr->col].status == 1516 rf_ds_optimal) && (raidPtr->numFailures > 0)) { 1517 /* some other component has failed. Let's not make 1518 things worse. XXX wrong for RAID6 */ 1519 RF_UNLOCK_MUTEX(raidPtr->mutex); 1520 return (EINVAL); 1521 } 1522 if (raidPtr->Disks[rr->col].status == rf_ds_spared) { 1523 /* Can't fail a spared disk! */ 1524 RF_UNLOCK_MUTEX(raidPtr->mutex); 1525 return (EINVAL); 1526 } 1527 RF_UNLOCK_MUTEX(raidPtr->mutex); 1528 1529 /* make a copy of the recon request so that we don't rely on 1530 * the user's buffer */ 1531 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1532 if (rrcopy == NULL) 1533 return(ENOMEM); 1534 memcpy(rrcopy, rr, sizeof(*rr)); 1535 rrcopy->raidPtr = (void *) raidPtr; 1536 1537 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1538 rf_ReconThread, 1539 rrcopy,"raid_recon"); 1540 return (0); 1541 1542 /* invoke a copyback operation after recon on whatever disk 1543 * needs it, if any */ 1544 case RAIDFRAME_COPYBACK: 1545 1546 if (raidPtr->Layout.map->faultsTolerated == 0) { 1547 /* This makes no sense on a RAID 0!! */ 1548 return(EINVAL); 1549 } 1550 1551 if (raidPtr->copyback_in_progress == 1) { 1552 /* Copyback is already in progress! */ 1553 return(EINVAL); 1554 } 1555 1556 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1557 rf_CopybackThread, 1558 raidPtr,"raid_copyback"); 1559 return (retcode); 1560 1561 /* return the percentage completion of reconstruction */ 1562 case RAIDFRAME_CHECK_RECON_STATUS: 1563 if (raidPtr->Layout.map->faultsTolerated == 0) { 1564 /* This makes no sense on a RAID 0, so tell the 1565 user it's done. */ 1566 *(int *) data = 100; 1567 return(0); 1568 } 1569 if (raidPtr->status != rf_rs_reconstructing) 1570 *(int *) data = 100; 1571 else { 1572 if (raidPtr->reconControl->numRUsTotal > 0) { 1573 *(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal); 1574 } else { 1575 *(int *) data = 0; 1576 } 1577 } 1578 return (0); 1579 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1580 progressInfoPtr = (RF_ProgressInfo_t **) data; 1581 if (raidPtr->status != rf_rs_reconstructing) { 1582 progressInfo.remaining = 0; 1583 progressInfo.completed = 100; 1584 progressInfo.total = 100; 1585 } else { 1586 progressInfo.total = 1587 raidPtr->reconControl->numRUsTotal; 1588 progressInfo.completed = 1589 raidPtr->reconControl->numRUsComplete; 1590 progressInfo.remaining = progressInfo.total - 1591 progressInfo.completed; 1592 } 1593 retcode = copyout(&progressInfo, *progressInfoPtr, 1594 sizeof(RF_ProgressInfo_t)); 1595 return (retcode); 1596 1597 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1598 if (raidPtr->Layout.map->faultsTolerated == 0) { 1599 /* This makes no sense on a RAID 0, so tell the 1600 user it's done. */ 1601 *(int *) data = 100; 1602 return(0); 1603 } 1604 if (raidPtr->parity_rewrite_in_progress == 1) { 1605 *(int *) data = 100 * 1606 raidPtr->parity_rewrite_stripes_done / 1607 raidPtr->Layout.numStripe; 1608 } else { 1609 *(int *) data = 100; 1610 } 1611 return (0); 1612 1613 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1614 progressInfoPtr = (RF_ProgressInfo_t **) data; 1615 if (raidPtr->parity_rewrite_in_progress == 1) { 1616 progressInfo.total = raidPtr->Layout.numStripe; 1617 progressInfo.completed = 1618 raidPtr->parity_rewrite_stripes_done; 1619 progressInfo.remaining = progressInfo.total - 1620 progressInfo.completed; 1621 } else { 1622 progressInfo.remaining = 0; 1623 progressInfo.completed = 100; 1624 progressInfo.total = 100; 1625 } 1626 retcode = copyout(&progressInfo, *progressInfoPtr, 1627 sizeof(RF_ProgressInfo_t)); 1628 return (retcode); 1629 1630 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1631 if (raidPtr->Layout.map->faultsTolerated == 0) { 1632 /* This makes no sense on a RAID 0 */ 1633 *(int *) data = 100; 1634 return(0); 1635 } 1636 if (raidPtr->copyback_in_progress == 1) { 1637 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1638 raidPtr->Layout.numStripe; 1639 } else { 1640 *(int *) data = 100; 1641 } 1642 return (0); 1643 1644 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1645 progressInfoPtr = (RF_ProgressInfo_t **) data; 1646 if (raidPtr->copyback_in_progress == 1) { 1647 progressInfo.total = raidPtr->Layout.numStripe; 1648 progressInfo.completed = 1649 raidPtr->copyback_stripes_done; 1650 progressInfo.remaining = progressInfo.total - 1651 progressInfo.completed; 1652 } else { 1653 progressInfo.remaining = 0; 1654 progressInfo.completed = 100; 1655 progressInfo.total = 100; 1656 } 1657 retcode = copyout(&progressInfo, *progressInfoPtr, 1658 sizeof(RF_ProgressInfo_t)); 1659 return (retcode); 1660 1661 /* the sparetable daemon calls this to wait for the kernel to 1662 * need a spare table. this ioctl does not return until a 1663 * spare table is needed. XXX -- calling mpsleep here in the 1664 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1665 * -- I should either compute the spare table in the kernel, 1666 * or have a different -- XXX XXX -- interface (a different 1667 * character device) for delivering the table -- XXX */ 1668 #if 0 1669 case RAIDFRAME_SPARET_WAIT: 1670 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1671 while (!rf_sparet_wait_queue) 1672 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE); 1673 waitreq = rf_sparet_wait_queue; 1674 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1675 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1676 1677 /* structure assignment */ 1678 *((RF_SparetWait_t *) data) = *waitreq; 1679 1680 RF_Free(waitreq, sizeof(*waitreq)); 1681 return (0); 1682 1683 /* wakes up a process waiting on SPARET_WAIT and puts an error 1684 * code in it that will cause the dameon to exit */ 1685 case RAIDFRAME_ABORT_SPARET_WAIT: 1686 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1687 waitreq->fcol = -1; 1688 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1689 waitreq->next = rf_sparet_wait_queue; 1690 rf_sparet_wait_queue = waitreq; 1691 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1692 wakeup(&rf_sparet_wait_queue); 1693 return (0); 1694 1695 /* used by the spare table daemon to deliver a spare table 1696 * into the kernel */ 1697 case RAIDFRAME_SEND_SPARET: 1698 1699 /* install the spare table */ 1700 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1701 1702 /* respond to the requestor. the return status of the spare 1703 * table installation is passed in the "fcol" field */ 1704 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1705 waitreq->fcol = retcode; 1706 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1707 waitreq->next = rf_sparet_resp_queue; 1708 rf_sparet_resp_queue = waitreq; 1709 wakeup(&rf_sparet_resp_queue); 1710 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1711 1712 return (retcode); 1713 #endif 1714 1715 default: 1716 break; /* fall through to the os-specific code below */ 1717 1718 } 1719 1720 if (!raidPtr->valid) 1721 return (EINVAL); 1722 1723 /* 1724 * Add support for "regular" device ioctls here. 1725 */ 1726 1727 switch (cmd) { 1728 case DIOCGDINFO: 1729 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1730 break; 1731 #ifdef __HAVE_OLD_DISKLABEL 1732 case ODIOCGDINFO: 1733 newlabel = *(rs->sc_dkdev.dk_label); 1734 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1735 return ENOTTY; 1736 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1737 break; 1738 #endif 1739 1740 case DIOCGPART: 1741 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1742 ((struct partinfo *) data)->part = 1743 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1744 break; 1745 1746 case DIOCWDINFO: 1747 case DIOCSDINFO: 1748 #ifdef __HAVE_OLD_DISKLABEL 1749 case ODIOCWDINFO: 1750 case ODIOCSDINFO: 1751 #endif 1752 { 1753 struct disklabel *lp; 1754 #ifdef __HAVE_OLD_DISKLABEL 1755 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1756 memset(&newlabel, 0, sizeof newlabel); 1757 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1758 lp = &newlabel; 1759 } else 1760 #endif 1761 lp = (struct disklabel *)data; 1762 1763 if ((error = raidlock(rs)) != 0) 1764 return (error); 1765 1766 rs->sc_flags |= RAIDF_LABELLING; 1767 1768 error = setdisklabel(rs->sc_dkdev.dk_label, 1769 lp, 0, rs->sc_dkdev.dk_cpulabel); 1770 if (error == 0) { 1771 if (cmd == DIOCWDINFO 1772 #ifdef __HAVE_OLD_DISKLABEL 1773 || cmd == ODIOCWDINFO 1774 #endif 1775 ) 1776 error = writedisklabel(RAIDLABELDEV(dev), 1777 raidstrategy, rs->sc_dkdev.dk_label, 1778 rs->sc_dkdev.dk_cpulabel); 1779 } 1780 rs->sc_flags &= ~RAIDF_LABELLING; 1781 1782 raidunlock(rs); 1783 1784 if (error) 1785 return (error); 1786 break; 1787 } 1788 1789 case DIOCWLABEL: 1790 if (*(int *) data != 0) 1791 rs->sc_flags |= RAIDF_WLABEL; 1792 else 1793 rs->sc_flags &= ~RAIDF_WLABEL; 1794 break; 1795 1796 case DIOCGDEFLABEL: 1797 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1798 break; 1799 1800 #ifdef __HAVE_OLD_DISKLABEL 1801 case ODIOCGDEFLABEL: 1802 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1803 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1804 return ENOTTY; 1805 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1806 break; 1807 #endif 1808 1809 case DIOCAWEDGE: 1810 case DIOCDWEDGE: 1811 dkw = (void *)data; 1812 1813 /* If the ioctl happens here, the parent is us. */ 1814 (void)strcpy(dkw->dkw_parent, rs->sc_xname); 1815 return cmd == DIOCAWEDGE ? dkwedge_add(dkw) : dkwedge_del(dkw); 1816 1817 case DIOCLWEDGES: 1818 return dkwedge_list(&rs->sc_dkdev, 1819 (struct dkwedge_list *)data, l); 1820 1821 default: 1822 retcode = ENOTTY; 1823 } 1824 return (retcode); 1825 1826 } 1827 1828 1829 /* raidinit -- complete the rest of the initialization for the 1830 RAIDframe device. */ 1831 1832 1833 static void 1834 raidinit(RF_Raid_t *raidPtr) 1835 { 1836 struct cfdata *cf; 1837 struct raid_softc *rs; 1838 int unit; 1839 1840 unit = raidPtr->raidid; 1841 1842 rs = &raid_softc[unit]; 1843 1844 /* XXX should check return code first... */ 1845 rs->sc_flags |= RAIDF_INITED; 1846 1847 /* XXX doesn't check bounds. */ 1848 snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit); 1849 1850 /* attach the pseudo device */ 1851 cf = malloc(sizeof(*cf), M_RAIDFRAME, M_WAITOK); 1852 cf->cf_name = raid_cd.cd_name; 1853 cf->cf_atname = raid_cd.cd_name; 1854 cf->cf_unit = unit; 1855 cf->cf_fstate = FSTATE_STAR; 1856 1857 rs->sc_dev = config_attach_pseudo(cf); 1858 1859 if (rs->sc_dev==NULL) { 1860 printf("raid%d: config_attach_pseudo failed\n", 1861 raidPtr->raidid); 1862 } 1863 1864 /* disk_attach actually creates space for the CPU disklabel, among 1865 * other things, so it's critical to call this *BEFORE* we try putzing 1866 * with disklabels. */ 1867 1868 disk_init(&rs->sc_dkdev, rs->sc_xname, &rf_dkdriver); 1869 disk_attach(&rs->sc_dkdev); 1870 1871 /* XXX There may be a weird interaction here between this, and 1872 * protectedSectors, as used in RAIDframe. */ 1873 1874 rs->sc_size = raidPtr->totalSectors; 1875 1876 dkwedge_discover(&rs->sc_dkdev); 1877 1878 rf_set_properties(rs, raidPtr); 1879 1880 } 1881 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 1882 /* wake up the daemon & tell it to get us a spare table 1883 * XXX 1884 * the entries in the queues should be tagged with the raidPtr 1885 * so that in the extremely rare case that two recons happen at once, 1886 * we know for which device were requesting a spare table 1887 * XXX 1888 * 1889 * XXX This code is not currently used. GO 1890 */ 1891 int 1892 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req) 1893 { 1894 int retcode; 1895 1896 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1897 req->next = rf_sparet_wait_queue; 1898 rf_sparet_wait_queue = req; 1899 wakeup(&rf_sparet_wait_queue); 1900 1901 /* mpsleep unlocks the mutex */ 1902 while (!rf_sparet_resp_queue) { 1903 tsleep(&rf_sparet_resp_queue, PRIBIO, 1904 "raidframe getsparetable", 0); 1905 } 1906 req = rf_sparet_resp_queue; 1907 rf_sparet_resp_queue = req->next; 1908 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1909 1910 retcode = req->fcol; 1911 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1912 * alloc'd */ 1913 return (retcode); 1914 } 1915 #endif 1916 1917 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1918 * bp & passes it down. 1919 * any calls originating in the kernel must use non-blocking I/O 1920 * do some extra sanity checking to return "appropriate" error values for 1921 * certain conditions (to make some standard utilities work) 1922 * 1923 * Formerly known as: rf_DoAccessKernel 1924 */ 1925 void 1926 raidstart(RF_Raid_t *raidPtr) 1927 { 1928 RF_SectorCount_t num_blocks, pb, sum; 1929 RF_RaidAddr_t raid_addr; 1930 struct partition *pp; 1931 daddr_t blocknum; 1932 int unit; 1933 struct raid_softc *rs; 1934 int do_async; 1935 struct buf *bp; 1936 int rc; 1937 1938 unit = raidPtr->raidid; 1939 rs = &raid_softc[unit]; 1940 1941 /* quick check to see if anything has died recently */ 1942 RF_LOCK_MUTEX(raidPtr->mutex); 1943 if (raidPtr->numNewFailures > 0) { 1944 RF_UNLOCK_MUTEX(raidPtr->mutex); 1945 rf_update_component_labels(raidPtr, 1946 RF_NORMAL_COMPONENT_UPDATE); 1947 RF_LOCK_MUTEX(raidPtr->mutex); 1948 raidPtr->numNewFailures--; 1949 } 1950 1951 /* Check to see if we're at the limit... */ 1952 while (raidPtr->openings > 0) { 1953 RF_UNLOCK_MUTEX(raidPtr->mutex); 1954 1955 /* get the next item, if any, from the queue */ 1956 if ((bp = BUFQ_GET(rs->buf_queue)) == NULL) { 1957 /* nothing more to do */ 1958 return; 1959 } 1960 1961 /* Ok, for the bp we have here, bp->b_blkno is relative to the 1962 * partition.. Need to make it absolute to the underlying 1963 * device.. */ 1964 1965 blocknum = bp->b_blkno; 1966 if (DISKPART(bp->b_dev) != RAW_PART) { 1967 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 1968 blocknum += pp->p_offset; 1969 } 1970 1971 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 1972 (int) blocknum)); 1973 1974 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 1975 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 1976 1977 /* *THIS* is where we adjust what block we're going to... 1978 * but DO NOT TOUCH bp->b_blkno!!! */ 1979 raid_addr = blocknum; 1980 1981 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 1982 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 1983 sum = raid_addr + num_blocks + pb; 1984 if (1 || rf_debugKernelAccess) { 1985 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 1986 (int) raid_addr, (int) sum, (int) num_blocks, 1987 (int) pb, (int) bp->b_resid)); 1988 } 1989 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 1990 || (sum < num_blocks) || (sum < pb)) { 1991 bp->b_error = ENOSPC; 1992 bp->b_resid = bp->b_bcount; 1993 biodone(bp); 1994 RF_LOCK_MUTEX(raidPtr->mutex); 1995 continue; 1996 } 1997 /* 1998 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 1999 */ 2000 2001 if (bp->b_bcount & raidPtr->sectorMask) { 2002 bp->b_error = EINVAL; 2003 bp->b_resid = bp->b_bcount; 2004 biodone(bp); 2005 RF_LOCK_MUTEX(raidPtr->mutex); 2006 continue; 2007 2008 } 2009 db1_printf(("Calling DoAccess..\n")); 2010 2011 2012 RF_LOCK_MUTEX(raidPtr->mutex); 2013 raidPtr->openings--; 2014 RF_UNLOCK_MUTEX(raidPtr->mutex); 2015 2016 /* 2017 * Everything is async. 2018 */ 2019 do_async = 1; 2020 2021 disk_busy(&rs->sc_dkdev); 2022 2023 /* XXX we're still at splbio() here... do we *really* 2024 need to be? */ 2025 2026 /* don't ever condition on bp->b_flags & B_WRITE. 2027 * always condition on B_READ instead */ 2028 2029 rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 2030 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 2031 do_async, raid_addr, num_blocks, 2032 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 2033 2034 if (rc) { 2035 bp->b_error = rc; 2036 bp->b_resid = bp->b_bcount; 2037 biodone(bp); 2038 /* continue loop */ 2039 } 2040 2041 RF_LOCK_MUTEX(raidPtr->mutex); 2042 } 2043 RF_UNLOCK_MUTEX(raidPtr->mutex); 2044 } 2045 2046 2047 2048 2049 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 2050 2051 int 2052 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req) 2053 { 2054 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 2055 struct buf *bp; 2056 2057 req->queue = queue; 2058 2059 #if DIAGNOSTIC 2060 if (queue->raidPtr->raidid >= numraid) { 2061 printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid, 2062 numraid); 2063 panic("Invalid Unit number in rf_DispatchKernelIO"); 2064 } 2065 #endif 2066 2067 bp = req->bp; 2068 2069 switch (req->type) { 2070 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 2071 /* XXX need to do something extra here.. */ 2072 /* I'm leaving this in, as I've never actually seen it used, 2073 * and I'd like folks to report it... GO */ 2074 printf(("WAKEUP CALLED\n")); 2075 queue->numOutstanding++; 2076 2077 bp->b_flags = 0; 2078 bp->b_private = req; 2079 2080 KernelWakeupFunc(bp); 2081 break; 2082 2083 case RF_IO_TYPE_READ: 2084 case RF_IO_TYPE_WRITE: 2085 #if RF_ACC_TRACE > 0 2086 if (req->tracerec) { 2087 RF_ETIMER_START(req->tracerec->timer); 2088 } 2089 #endif 2090 InitBP(bp, queue->rf_cinfo->ci_vp, 2091 op, queue->rf_cinfo->ci_dev, 2092 req->sectorOffset, req->numSector, 2093 req->buf, KernelWakeupFunc, (void *) req, 2094 queue->raidPtr->logBytesPerSector, req->b_proc); 2095 2096 if (rf_debugKernelAccess) { 2097 db1_printf(("dispatch: bp->b_blkno = %ld\n", 2098 (long) bp->b_blkno)); 2099 } 2100 queue->numOutstanding++; 2101 queue->last_deq_sector = req->sectorOffset; 2102 /* acc wouldn't have been let in if there were any pending 2103 * reqs at any other priority */ 2104 queue->curPriority = req->priority; 2105 2106 db1_printf(("Going for %c to unit %d col %d\n", 2107 req->type, queue->raidPtr->raidid, 2108 queue->col)); 2109 db1_printf(("sector %d count %d (%d bytes) %d\n", 2110 (int) req->sectorOffset, (int) req->numSector, 2111 (int) (req->numSector << 2112 queue->raidPtr->logBytesPerSector), 2113 (int) queue->raidPtr->logBytesPerSector)); 2114 VOP_STRATEGY(bp->b_vp, bp); 2115 2116 break; 2117 2118 default: 2119 panic("bad req->type in rf_DispatchKernelIO"); 2120 } 2121 db1_printf(("Exiting from DispatchKernelIO\n")); 2122 2123 return (0); 2124 } 2125 /* this is the callback function associated with a I/O invoked from 2126 kernel code. 2127 */ 2128 static void 2129 KernelWakeupFunc(struct buf *bp) 2130 { 2131 RF_DiskQueueData_t *req = NULL; 2132 RF_DiskQueue_t *queue; 2133 int s; 2134 2135 s = splbio(); 2136 db1_printf(("recovering the request queue:\n")); 2137 req = bp->b_private; 2138 2139 queue = (RF_DiskQueue_t *) req->queue; 2140 2141 #if RF_ACC_TRACE > 0 2142 if (req->tracerec) { 2143 RF_ETIMER_STOP(req->tracerec->timer); 2144 RF_ETIMER_EVAL(req->tracerec->timer); 2145 RF_LOCK_MUTEX(rf_tracing_mutex); 2146 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2147 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2148 req->tracerec->num_phys_ios++; 2149 RF_UNLOCK_MUTEX(rf_tracing_mutex); 2150 } 2151 #endif 2152 2153 /* XXX Ok, let's get aggressive... If b_error is set, let's go 2154 * ballistic, and mark the component as hosed... */ 2155 2156 if (bp->b_error != 0) { 2157 /* Mark the disk as dead */ 2158 /* but only mark it once... */ 2159 /* and only if it wouldn't leave this RAID set 2160 completely broken */ 2161 if (((queue->raidPtr->Disks[queue->col].status == 2162 rf_ds_optimal) || 2163 (queue->raidPtr->Disks[queue->col].status == 2164 rf_ds_used_spare)) && 2165 (queue->raidPtr->numFailures < 2166 queue->raidPtr->Layout.map->faultsTolerated)) { 2167 printf("raid%d: IO Error. Marking %s as failed.\n", 2168 queue->raidPtr->raidid, 2169 queue->raidPtr->Disks[queue->col].devname); 2170 queue->raidPtr->Disks[queue->col].status = 2171 rf_ds_failed; 2172 queue->raidPtr->status = rf_rs_degraded; 2173 queue->raidPtr->numFailures++; 2174 queue->raidPtr->numNewFailures++; 2175 } else { /* Disk is already dead... */ 2176 /* printf("Disk already marked as dead!\n"); */ 2177 } 2178 2179 } 2180 2181 /* Fill in the error value */ 2182 2183 req->error = bp->b_error; 2184 2185 simple_lock(&queue->raidPtr->iodone_lock); 2186 2187 /* Drop this one on the "finished" queue... */ 2188 TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries); 2189 2190 /* Let the raidio thread know there is work to be done. */ 2191 wakeup(&(queue->raidPtr->iodone)); 2192 2193 simple_unlock(&queue->raidPtr->iodone_lock); 2194 2195 splx(s); 2196 } 2197 2198 2199 2200 /* 2201 * initialize a buf structure for doing an I/O in the kernel. 2202 */ 2203 static void 2204 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev, 2205 RF_SectorNum_t startSect, RF_SectorCount_t numSect, void *bf, 2206 void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector, 2207 struct proc *b_proc) 2208 { 2209 /* bp->b_flags = B_PHYS | rw_flag; */ 2210 bp->b_flags = rw_flag; /* XXX need B_PHYS here too??? */ 2211 bp->b_oflags = 0; 2212 bp->b_cflags = 0; 2213 bp->b_bcount = numSect << logBytesPerSector; 2214 bp->b_bufsize = bp->b_bcount; 2215 bp->b_error = 0; 2216 bp->b_dev = dev; 2217 bp->b_data = bf; 2218 bp->b_blkno = startSect; 2219 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 2220 if (bp->b_bcount == 0) { 2221 panic("bp->b_bcount is zero in InitBP!!"); 2222 } 2223 bp->b_proc = b_proc; 2224 bp->b_iodone = cbFunc; 2225 bp->b_private = cbArg; 2226 bp->b_vp = b_vp; 2227 bp->b_objlock = &b_vp->v_interlock; 2228 if ((bp->b_flags & B_READ) == 0) { 2229 mutex_enter(&b_vp->v_interlock); 2230 b_vp->v_numoutput++; 2231 mutex_exit(&b_vp->v_interlock); 2232 } 2233 2234 } 2235 2236 static void 2237 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs, 2238 struct disklabel *lp) 2239 { 2240 memset(lp, 0, sizeof(*lp)); 2241 2242 /* fabricate a label... */ 2243 lp->d_secperunit = raidPtr->totalSectors; 2244 lp->d_secsize = raidPtr->bytesPerSector; 2245 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 2246 lp->d_ntracks = 4 * raidPtr->numCol; 2247 lp->d_ncylinders = raidPtr->totalSectors / 2248 (lp->d_nsectors * lp->d_ntracks); 2249 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 2250 2251 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 2252 lp->d_type = DTYPE_RAID; 2253 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 2254 lp->d_rpm = 3600; 2255 lp->d_interleave = 1; 2256 lp->d_flags = 0; 2257 2258 lp->d_partitions[RAW_PART].p_offset = 0; 2259 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 2260 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 2261 lp->d_npartitions = RAW_PART + 1; 2262 2263 lp->d_magic = DISKMAGIC; 2264 lp->d_magic2 = DISKMAGIC; 2265 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 2266 2267 } 2268 /* 2269 * Read the disklabel from the raid device. If one is not present, fake one 2270 * up. 2271 */ 2272 static void 2273 raidgetdisklabel(dev_t dev) 2274 { 2275 int unit = raidunit(dev); 2276 struct raid_softc *rs = &raid_softc[unit]; 2277 const char *errstring; 2278 struct disklabel *lp = rs->sc_dkdev.dk_label; 2279 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 2280 RF_Raid_t *raidPtr; 2281 2282 db1_printf(("Getting the disklabel...\n")); 2283 2284 memset(clp, 0, sizeof(*clp)); 2285 2286 raidPtr = raidPtrs[unit]; 2287 2288 raidgetdefaultlabel(raidPtr, rs, lp); 2289 2290 /* 2291 * Call the generic disklabel extraction routine. 2292 */ 2293 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 2294 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2295 if (errstring) 2296 raidmakedisklabel(rs); 2297 else { 2298 int i; 2299 struct partition *pp; 2300 2301 /* 2302 * Sanity check whether the found disklabel is valid. 2303 * 2304 * This is necessary since total size of the raid device 2305 * may vary when an interleave is changed even though exactly 2306 * same components are used, and old disklabel may used 2307 * if that is found. 2308 */ 2309 if (lp->d_secperunit != rs->sc_size) 2310 printf("raid%d: WARNING: %s: " 2311 "total sector size in disklabel (%d) != " 2312 "the size of raid (%ld)\n", unit, rs->sc_xname, 2313 lp->d_secperunit, (long) rs->sc_size); 2314 for (i = 0; i < lp->d_npartitions; i++) { 2315 pp = &lp->d_partitions[i]; 2316 if (pp->p_offset + pp->p_size > rs->sc_size) 2317 printf("raid%d: WARNING: %s: end of partition `%c' " 2318 "exceeds the size of raid (%ld)\n", 2319 unit, rs->sc_xname, 'a' + i, (long) rs->sc_size); 2320 } 2321 } 2322 2323 } 2324 /* 2325 * Take care of things one might want to take care of in the event 2326 * that a disklabel isn't present. 2327 */ 2328 static void 2329 raidmakedisklabel(struct raid_softc *rs) 2330 { 2331 struct disklabel *lp = rs->sc_dkdev.dk_label; 2332 db1_printf(("Making a label..\n")); 2333 2334 /* 2335 * For historical reasons, if there's no disklabel present 2336 * the raw partition must be marked FS_BSDFFS. 2337 */ 2338 2339 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2340 2341 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2342 2343 lp->d_checksum = dkcksum(lp); 2344 } 2345 /* 2346 * Wait interruptibly for an exclusive lock. 2347 * 2348 * XXX 2349 * Several drivers do this; it should be abstracted and made MP-safe. 2350 * (Hmm... where have we seen this warning before :-> GO ) 2351 */ 2352 static int 2353 raidlock(struct raid_softc *rs) 2354 { 2355 int error; 2356 2357 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2358 rs->sc_flags |= RAIDF_WANTED; 2359 if ((error = 2360 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2361 return (error); 2362 } 2363 rs->sc_flags |= RAIDF_LOCKED; 2364 return (0); 2365 } 2366 /* 2367 * Unlock and wake up any waiters. 2368 */ 2369 static void 2370 raidunlock(struct raid_softc *rs) 2371 { 2372 2373 rs->sc_flags &= ~RAIDF_LOCKED; 2374 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2375 rs->sc_flags &= ~RAIDF_WANTED; 2376 wakeup(rs); 2377 } 2378 } 2379 2380 2381 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2382 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2383 2384 int 2385 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter) 2386 { 2387 RF_ComponentLabel_t clabel; 2388 raidread_component_label(dev, b_vp, &clabel); 2389 clabel.mod_counter = mod_counter; 2390 clabel.clean = RF_RAID_CLEAN; 2391 raidwrite_component_label(dev, b_vp, &clabel); 2392 return(0); 2393 } 2394 2395 2396 int 2397 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter) 2398 { 2399 RF_ComponentLabel_t clabel; 2400 raidread_component_label(dev, b_vp, &clabel); 2401 clabel.mod_counter = mod_counter; 2402 clabel.clean = RF_RAID_DIRTY; 2403 raidwrite_component_label(dev, b_vp, &clabel); 2404 return(0); 2405 } 2406 2407 /* ARGSUSED */ 2408 int 2409 raidread_component_label(dev_t dev, struct vnode *b_vp, 2410 RF_ComponentLabel_t *clabel) 2411 { 2412 struct buf *bp; 2413 const struct bdevsw *bdev; 2414 int error; 2415 2416 /* XXX should probably ensure that we don't try to do this if 2417 someone has changed rf_protected_sectors. */ 2418 2419 if (b_vp == NULL) { 2420 /* For whatever reason, this component is not valid. 2421 Don't try to read a component label from it. */ 2422 return(EINVAL); 2423 } 2424 2425 /* get a block of the appropriate size... */ 2426 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2427 bp->b_dev = dev; 2428 2429 /* get our ducks in a row for the read */ 2430 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2431 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2432 bp->b_flags |= B_READ; 2433 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2434 2435 bdev = bdevsw_lookup(bp->b_dev); 2436 if (bdev == NULL) 2437 return (ENXIO); 2438 (*bdev->d_strategy)(bp); 2439 2440 error = biowait(bp); 2441 2442 if (!error) { 2443 memcpy(clabel, bp->b_data, 2444 sizeof(RF_ComponentLabel_t)); 2445 } 2446 2447 brelse(bp, 0); 2448 return(error); 2449 } 2450 /* ARGSUSED */ 2451 int 2452 raidwrite_component_label(dev_t dev, struct vnode *b_vp, 2453 RF_ComponentLabel_t *clabel) 2454 { 2455 struct buf *bp; 2456 const struct bdevsw *bdev; 2457 int error; 2458 2459 /* get a block of the appropriate size... */ 2460 bp = geteblk((int)RF_COMPONENT_INFO_SIZE); 2461 bp->b_dev = dev; 2462 2463 /* get our ducks in a row for the write */ 2464 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE; 2465 bp->b_bcount = RF_COMPONENT_INFO_SIZE; 2466 bp->b_flags |= B_WRITE; 2467 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE; 2468 2469 memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE ); 2470 2471 memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t)); 2472 2473 bdev = bdevsw_lookup(bp->b_dev); 2474 if (bdev == NULL) 2475 return (ENXIO); 2476 (*bdev->d_strategy)(bp); 2477 error = biowait(bp); 2478 brelse(bp, 0); 2479 if (error) { 2480 #if 1 2481 printf("Failed to write RAID component info!\n"); 2482 #endif 2483 } 2484 2485 return(error); 2486 } 2487 2488 void 2489 rf_markalldirty(RF_Raid_t *raidPtr) 2490 { 2491 RF_ComponentLabel_t clabel; 2492 int sparecol; 2493 int c; 2494 int j; 2495 int scol = -1; 2496 2497 raidPtr->mod_counter++; 2498 for (c = 0; c < raidPtr->numCol; c++) { 2499 /* we don't want to touch (at all) a disk that has 2500 failed */ 2501 if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) { 2502 raidread_component_label( 2503 raidPtr->Disks[c].dev, 2504 raidPtr->raid_cinfo[c].ci_vp, 2505 &clabel); 2506 if (clabel.status == rf_ds_spared) { 2507 /* XXX do something special... 2508 but whatever you do, don't 2509 try to access it!! */ 2510 } else { 2511 raidmarkdirty( 2512 raidPtr->Disks[c].dev, 2513 raidPtr->raid_cinfo[c].ci_vp, 2514 raidPtr->mod_counter); 2515 } 2516 } 2517 } 2518 2519 for( c = 0; c < raidPtr->numSpare ; c++) { 2520 sparecol = raidPtr->numCol + c; 2521 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2522 /* 2523 2524 we claim this disk is "optimal" if it's 2525 rf_ds_used_spare, as that means it should be 2526 directly substitutable for the disk it replaced. 2527 We note that too... 2528 2529 */ 2530 2531 for(j=0;j<raidPtr->numCol;j++) { 2532 if (raidPtr->Disks[j].spareCol == sparecol) { 2533 scol = j; 2534 break; 2535 } 2536 } 2537 2538 raidread_component_label( 2539 raidPtr->Disks[sparecol].dev, 2540 raidPtr->raid_cinfo[sparecol].ci_vp, 2541 &clabel); 2542 /* make sure status is noted */ 2543 2544 raid_init_component_label(raidPtr, &clabel); 2545 2546 clabel.row = 0; 2547 clabel.column = scol; 2548 /* Note: we *don't* change status from rf_ds_used_spare 2549 to rf_ds_optimal */ 2550 /* clabel.status = rf_ds_optimal; */ 2551 2552 raidmarkdirty(raidPtr->Disks[sparecol].dev, 2553 raidPtr->raid_cinfo[sparecol].ci_vp, 2554 raidPtr->mod_counter); 2555 } 2556 } 2557 } 2558 2559 2560 void 2561 rf_update_component_labels(RF_Raid_t *raidPtr, int final) 2562 { 2563 RF_ComponentLabel_t clabel; 2564 int sparecol; 2565 int c; 2566 int j; 2567 int scol; 2568 2569 scol = -1; 2570 2571 /* XXX should do extra checks to make sure things really are clean, 2572 rather than blindly setting the clean bit... */ 2573 2574 raidPtr->mod_counter++; 2575 2576 for (c = 0; c < raidPtr->numCol; c++) { 2577 if (raidPtr->Disks[c].status == rf_ds_optimal) { 2578 raidread_component_label( 2579 raidPtr->Disks[c].dev, 2580 raidPtr->raid_cinfo[c].ci_vp, 2581 &clabel); 2582 /* make sure status is noted */ 2583 clabel.status = rf_ds_optimal; 2584 2585 /* bump the counter */ 2586 clabel.mod_counter = raidPtr->mod_counter; 2587 2588 /* note what unit we are configured as */ 2589 clabel.last_unit = raidPtr->raidid; 2590 2591 raidwrite_component_label( 2592 raidPtr->Disks[c].dev, 2593 raidPtr->raid_cinfo[c].ci_vp, 2594 &clabel); 2595 if (final == RF_FINAL_COMPONENT_UPDATE) { 2596 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2597 raidmarkclean( 2598 raidPtr->Disks[c].dev, 2599 raidPtr->raid_cinfo[c].ci_vp, 2600 raidPtr->mod_counter); 2601 } 2602 } 2603 } 2604 /* else we don't touch it.. */ 2605 } 2606 2607 for( c = 0; c < raidPtr->numSpare ; c++) { 2608 sparecol = raidPtr->numCol + c; 2609 /* Need to ensure that the reconstruct actually completed! */ 2610 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2611 /* 2612 2613 we claim this disk is "optimal" if it's 2614 rf_ds_used_spare, as that means it should be 2615 directly substitutable for the disk it replaced. 2616 We note that too... 2617 2618 */ 2619 2620 for(j=0;j<raidPtr->numCol;j++) { 2621 if (raidPtr->Disks[j].spareCol == sparecol) { 2622 scol = j; 2623 break; 2624 } 2625 } 2626 2627 /* XXX shouldn't *really* need this... */ 2628 raidread_component_label( 2629 raidPtr->Disks[sparecol].dev, 2630 raidPtr->raid_cinfo[sparecol].ci_vp, 2631 &clabel); 2632 /* make sure status is noted */ 2633 2634 raid_init_component_label(raidPtr, &clabel); 2635 2636 clabel.mod_counter = raidPtr->mod_counter; 2637 clabel.column = scol; 2638 clabel.status = rf_ds_optimal; 2639 clabel.last_unit = raidPtr->raidid; 2640 2641 raidwrite_component_label( 2642 raidPtr->Disks[sparecol].dev, 2643 raidPtr->raid_cinfo[sparecol].ci_vp, 2644 &clabel); 2645 if (final == RF_FINAL_COMPONENT_UPDATE) { 2646 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2647 raidmarkclean( raidPtr->Disks[sparecol].dev, 2648 raidPtr->raid_cinfo[sparecol].ci_vp, 2649 raidPtr->mod_counter); 2650 } 2651 } 2652 } 2653 } 2654 } 2655 2656 void 2657 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured) 2658 { 2659 2660 if (vp != NULL) { 2661 if (auto_configured == 1) { 2662 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2663 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2664 vput(vp); 2665 2666 } else { 2667 (void) vn_close(vp, FREAD | FWRITE, curlwp->l_cred); 2668 } 2669 } 2670 } 2671 2672 2673 void 2674 rf_UnconfigureVnodes(RF_Raid_t *raidPtr) 2675 { 2676 int r,c; 2677 struct vnode *vp; 2678 int acd; 2679 2680 2681 /* We take this opportunity to close the vnodes like we should.. */ 2682 2683 for (c = 0; c < raidPtr->numCol; c++) { 2684 vp = raidPtr->raid_cinfo[c].ci_vp; 2685 acd = raidPtr->Disks[c].auto_configured; 2686 rf_close_component(raidPtr, vp, acd); 2687 raidPtr->raid_cinfo[c].ci_vp = NULL; 2688 raidPtr->Disks[c].auto_configured = 0; 2689 } 2690 2691 for (r = 0; r < raidPtr->numSpare; r++) { 2692 vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp; 2693 acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured; 2694 rf_close_component(raidPtr, vp, acd); 2695 raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL; 2696 raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0; 2697 } 2698 } 2699 2700 2701 void 2702 rf_ReconThread(struct rf_recon_req *req) 2703 { 2704 int s; 2705 RF_Raid_t *raidPtr; 2706 2707 s = splbio(); 2708 raidPtr = (RF_Raid_t *) req->raidPtr; 2709 raidPtr->recon_in_progress = 1; 2710 2711 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col, 2712 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2713 2714 RF_Free(req, sizeof(*req)); 2715 2716 raidPtr->recon_in_progress = 0; 2717 splx(s); 2718 2719 /* That's all... */ 2720 kthread_exit(0); /* does not return */ 2721 } 2722 2723 void 2724 rf_RewriteParityThread(RF_Raid_t *raidPtr) 2725 { 2726 int retcode; 2727 int s; 2728 2729 raidPtr->parity_rewrite_stripes_done = 0; 2730 raidPtr->parity_rewrite_in_progress = 1; 2731 s = splbio(); 2732 retcode = rf_RewriteParity(raidPtr); 2733 splx(s); 2734 if (retcode) { 2735 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2736 } else { 2737 /* set the clean bit! If we shutdown correctly, 2738 the clean bit on each component label will get 2739 set */ 2740 raidPtr->parity_good = RF_RAID_CLEAN; 2741 } 2742 raidPtr->parity_rewrite_in_progress = 0; 2743 2744 /* Anyone waiting for us to stop? If so, inform them... */ 2745 if (raidPtr->waitShutdown) { 2746 wakeup(&raidPtr->parity_rewrite_in_progress); 2747 } 2748 2749 /* That's all... */ 2750 kthread_exit(0); /* does not return */ 2751 } 2752 2753 2754 void 2755 rf_CopybackThread(RF_Raid_t *raidPtr) 2756 { 2757 int s; 2758 2759 raidPtr->copyback_in_progress = 1; 2760 s = splbio(); 2761 rf_CopybackReconstructedData(raidPtr); 2762 splx(s); 2763 raidPtr->copyback_in_progress = 0; 2764 2765 /* That's all... */ 2766 kthread_exit(0); /* does not return */ 2767 } 2768 2769 2770 void 2771 rf_ReconstructInPlaceThread(struct rf_recon_req *req) 2772 { 2773 int s; 2774 RF_Raid_t *raidPtr; 2775 2776 s = splbio(); 2777 raidPtr = req->raidPtr; 2778 raidPtr->recon_in_progress = 1; 2779 rf_ReconstructInPlace(raidPtr, req->col); 2780 RF_Free(req, sizeof(*req)); 2781 raidPtr->recon_in_progress = 0; 2782 splx(s); 2783 2784 /* That's all... */ 2785 kthread_exit(0); /* does not return */ 2786 } 2787 2788 static RF_AutoConfig_t * 2789 rf_get_component(RF_AutoConfig_t *ac_list, dev_t dev, struct vnode *vp, 2790 const char *cname, RF_SectorCount_t size) 2791 { 2792 int good_one = 0; 2793 RF_ComponentLabel_t *clabel; 2794 RF_AutoConfig_t *ac; 2795 2796 clabel = malloc(sizeof(RF_ComponentLabel_t), M_RAIDFRAME, M_NOWAIT); 2797 if (clabel == NULL) { 2798 oomem: 2799 while(ac_list) { 2800 ac = ac_list; 2801 if (ac->clabel) 2802 free(ac->clabel, M_RAIDFRAME); 2803 ac_list = ac_list->next; 2804 free(ac, M_RAIDFRAME); 2805 } 2806 printf("RAID auto config: out of memory!\n"); 2807 return NULL; /* XXX probably should panic? */ 2808 } 2809 2810 if (!raidread_component_label(dev, vp, clabel)) { 2811 /* Got the label. Does it look reasonable? */ 2812 if (rf_reasonable_label(clabel) && 2813 (clabel->partitionSize <= size)) { 2814 #ifdef DEBUG 2815 printf("Component on: %s: %llu\n", 2816 cname, (unsigned long long)size); 2817 rf_print_component_label(clabel); 2818 #endif 2819 /* if it's reasonable, add it, else ignore it. */ 2820 ac = malloc(sizeof(RF_AutoConfig_t), M_RAIDFRAME, 2821 M_NOWAIT); 2822 if (ac == NULL) { 2823 free(clabel, M_RAIDFRAME); 2824 goto oomem; 2825 } 2826 strlcpy(ac->devname, cname, sizeof(ac->devname)); 2827 ac->dev = dev; 2828 ac->vp = vp; 2829 ac->clabel = clabel; 2830 ac->next = ac_list; 2831 ac_list = ac; 2832 good_one = 1; 2833 } 2834 } 2835 if (!good_one) { 2836 /* cleanup */ 2837 free(clabel, M_RAIDFRAME); 2838 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2839 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2840 vput(vp); 2841 } 2842 return ac_list; 2843 } 2844 2845 RF_AutoConfig_t * 2846 rf_find_raid_components() 2847 { 2848 struct vnode *vp; 2849 struct disklabel label; 2850 struct device *dv; 2851 dev_t dev; 2852 int bmajor, bminor, wedge; 2853 int error; 2854 int i; 2855 RF_AutoConfig_t *ac_list; 2856 2857 2858 /* initialize the AutoConfig list */ 2859 ac_list = NULL; 2860 2861 /* we begin by trolling through *all* the devices on the system */ 2862 2863 for (dv = alldevs.tqh_first; dv != NULL; 2864 dv = dv->dv_list.tqe_next) { 2865 2866 /* we are only interested in disks... */ 2867 if (device_class(dv) != DV_DISK) 2868 continue; 2869 2870 /* we don't care about floppies... */ 2871 if (device_is_a(dv, "fd")) { 2872 continue; 2873 } 2874 2875 /* we don't care about CD's... */ 2876 if (device_is_a(dv, "cd")) { 2877 continue; 2878 } 2879 2880 /* hdfd is the Atari/Hades floppy driver */ 2881 if (device_is_a(dv, "hdfd")) { 2882 continue; 2883 } 2884 2885 /* fdisa is the Atari/Milan floppy driver */ 2886 if (device_is_a(dv, "fdisa")) { 2887 continue; 2888 } 2889 2890 /* need to find the device_name_to_block_device_major stuff */ 2891 bmajor = devsw_name2blk(device_xname(dv), NULL, 0); 2892 2893 /* get a vnode for the raw partition of this disk */ 2894 2895 wedge = device_is_a(dv, "dk"); 2896 bminor = minor(device_unit(dv)); 2897 dev = wedge ? makedev(bmajor, bminor) : 2898 MAKEDISKDEV(bmajor, bminor, RAW_PART); 2899 if (bdevvp(dev, &vp)) 2900 panic("RAID can't alloc vnode"); 2901 2902 error = VOP_OPEN(vp, FREAD, NOCRED); 2903 2904 if (error) { 2905 /* "Who cares." Continue looking 2906 for something that exists*/ 2907 vput(vp); 2908 continue; 2909 } 2910 2911 if (wedge) { 2912 struct dkwedge_info dkw; 2913 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, 2914 NOCRED); 2915 if (error) { 2916 printf("RAIDframe: can't get wedge info for " 2917 "dev %s (%d)\n", device_xname(dv), error); 2918 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2919 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2920 vput(vp); 2921 continue; 2922 } 2923 2924 if (strcmp(dkw.dkw_ptype, DKW_PTYPE_RAIDFRAME) != 0) { 2925 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2926 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2927 vput(vp); 2928 continue; 2929 } 2930 2931 ac_list = rf_get_component(ac_list, dev, vp, 2932 device_xname(dv), dkw.dkw_size); 2933 continue; 2934 } 2935 2936 /* Ok, the disk exists. Go get the disklabel. */ 2937 error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED); 2938 if (error) { 2939 /* 2940 * XXX can't happen - open() would 2941 * have errored out (or faked up one) 2942 */ 2943 if (error != ENOTTY) 2944 printf("RAIDframe: can't get label for dev " 2945 "%s (%d)\n", device_xname(dv), error); 2946 } 2947 2948 /* don't need this any more. We'll allocate it again 2949 a little later if we really do... */ 2950 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2951 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2952 vput(vp); 2953 2954 if (error) 2955 continue; 2956 2957 for (i = 0; i < label.d_npartitions; i++) { 2958 char cname[sizeof(ac_list->devname)]; 2959 2960 /* We only support partitions marked as RAID */ 2961 if (label.d_partitions[i].p_fstype != FS_RAID) 2962 continue; 2963 2964 dev = MAKEDISKDEV(bmajor, device_unit(dv), i); 2965 if (bdevvp(dev, &vp)) 2966 panic("RAID can't alloc vnode"); 2967 2968 error = VOP_OPEN(vp, FREAD, NOCRED); 2969 if (error) { 2970 /* Whatever... */ 2971 vput(vp); 2972 continue; 2973 } 2974 snprintf(cname, sizeof(cname), "%s%c", 2975 device_xname(dv), 'a' + i); 2976 ac_list = rf_get_component(ac_list, dev, vp, cname, 2977 label.d_partitions[i].p_size); 2978 } 2979 } 2980 return ac_list; 2981 } 2982 2983 2984 static int 2985 rf_reasonable_label(RF_ComponentLabel_t *clabel) 2986 { 2987 2988 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 2989 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 2990 ((clabel->clean == RF_RAID_CLEAN) || 2991 (clabel->clean == RF_RAID_DIRTY)) && 2992 clabel->row >=0 && 2993 clabel->column >= 0 && 2994 clabel->num_rows > 0 && 2995 clabel->num_columns > 0 && 2996 clabel->row < clabel->num_rows && 2997 clabel->column < clabel->num_columns && 2998 clabel->blockSize > 0 && 2999 clabel->numBlocks > 0) { 3000 /* label looks reasonable enough... */ 3001 return(1); 3002 } 3003 return(0); 3004 } 3005 3006 3007 #ifdef DEBUG 3008 void 3009 rf_print_component_label(RF_ComponentLabel_t *clabel) 3010 { 3011 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 3012 clabel->row, clabel->column, 3013 clabel->num_rows, clabel->num_columns); 3014 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 3015 clabel->version, clabel->serial_number, 3016 clabel->mod_counter); 3017 printf(" Clean: %s Status: %d\n", 3018 clabel->clean ? "Yes" : "No", clabel->status ); 3019 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 3020 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 3021 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n", 3022 (char) clabel->parityConfig, clabel->blockSize, 3023 clabel->numBlocks); 3024 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" ); 3025 printf(" Contains root partition: %s\n", 3026 clabel->root_partition ? "Yes" : "No" ); 3027 printf(" Last configured as: raid%d\n", clabel->last_unit ); 3028 #if 0 3029 printf(" Config order: %d\n", clabel->config_order); 3030 #endif 3031 3032 } 3033 #endif 3034 3035 RF_ConfigSet_t * 3036 rf_create_auto_sets(RF_AutoConfig_t *ac_list) 3037 { 3038 RF_AutoConfig_t *ac; 3039 RF_ConfigSet_t *config_sets; 3040 RF_ConfigSet_t *cset; 3041 RF_AutoConfig_t *ac_next; 3042 3043 3044 config_sets = NULL; 3045 3046 /* Go through the AutoConfig list, and figure out which components 3047 belong to what sets. */ 3048 ac = ac_list; 3049 while(ac!=NULL) { 3050 /* we're going to putz with ac->next, so save it here 3051 for use at the end of the loop */ 3052 ac_next = ac->next; 3053 3054 if (config_sets == NULL) { 3055 /* will need at least this one... */ 3056 config_sets = (RF_ConfigSet_t *) 3057 malloc(sizeof(RF_ConfigSet_t), 3058 M_RAIDFRAME, M_NOWAIT); 3059 if (config_sets == NULL) { 3060 panic("rf_create_auto_sets: No memory!"); 3061 } 3062 /* this one is easy :) */ 3063 config_sets->ac = ac; 3064 config_sets->next = NULL; 3065 config_sets->rootable = 0; 3066 ac->next = NULL; 3067 } else { 3068 /* which set does this component fit into? */ 3069 cset = config_sets; 3070 while(cset!=NULL) { 3071 if (rf_does_it_fit(cset, ac)) { 3072 /* looks like it matches... */ 3073 ac->next = cset->ac; 3074 cset->ac = ac; 3075 break; 3076 } 3077 cset = cset->next; 3078 } 3079 if (cset==NULL) { 3080 /* didn't find a match above... new set..*/ 3081 cset = (RF_ConfigSet_t *) 3082 malloc(sizeof(RF_ConfigSet_t), 3083 M_RAIDFRAME, M_NOWAIT); 3084 if (cset == NULL) { 3085 panic("rf_create_auto_sets: No memory!"); 3086 } 3087 cset->ac = ac; 3088 ac->next = NULL; 3089 cset->next = config_sets; 3090 cset->rootable = 0; 3091 config_sets = cset; 3092 } 3093 } 3094 ac = ac_next; 3095 } 3096 3097 3098 return(config_sets); 3099 } 3100 3101 static int 3102 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac) 3103 { 3104 RF_ComponentLabel_t *clabel1, *clabel2; 3105 3106 /* If this one matches the *first* one in the set, that's good 3107 enough, since the other members of the set would have been 3108 through here too... */ 3109 /* note that we are not checking partitionSize here.. 3110 3111 Note that we are also not checking the mod_counters here. 3112 If everything else matches execpt the mod_counter, that's 3113 good enough for this test. We will deal with the mod_counters 3114 a little later in the autoconfiguration process. 3115 3116 (clabel1->mod_counter == clabel2->mod_counter) && 3117 3118 The reason we don't check for this is that failed disks 3119 will have lower modification counts. If those disks are 3120 not added to the set they used to belong to, then they will 3121 form their own set, which may result in 2 different sets, 3122 for example, competing to be configured at raid0, and 3123 perhaps competing to be the root filesystem set. If the 3124 wrong ones get configured, or both attempt to become /, 3125 weird behaviour and or serious lossage will occur. Thus we 3126 need to bring them into the fold here, and kick them out at 3127 a later point. 3128 3129 */ 3130 3131 clabel1 = cset->ac->clabel; 3132 clabel2 = ac->clabel; 3133 if ((clabel1->version == clabel2->version) && 3134 (clabel1->serial_number == clabel2->serial_number) && 3135 (clabel1->num_rows == clabel2->num_rows) && 3136 (clabel1->num_columns == clabel2->num_columns) && 3137 (clabel1->sectPerSU == clabel2->sectPerSU) && 3138 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 3139 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 3140 (clabel1->parityConfig == clabel2->parityConfig) && 3141 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 3142 (clabel1->blockSize == clabel2->blockSize) && 3143 (clabel1->numBlocks == clabel2->numBlocks) && 3144 (clabel1->autoconfigure == clabel2->autoconfigure) && 3145 (clabel1->root_partition == clabel2->root_partition) && 3146 (clabel1->last_unit == clabel2->last_unit) && 3147 (clabel1->config_order == clabel2->config_order)) { 3148 /* if it get's here, it almost *has* to be a match */ 3149 } else { 3150 /* it's not consistent with somebody in the set.. 3151 punt */ 3152 return(0); 3153 } 3154 /* all was fine.. it must fit... */ 3155 return(1); 3156 } 3157 3158 int 3159 rf_have_enough_components(RF_ConfigSet_t *cset) 3160 { 3161 RF_AutoConfig_t *ac; 3162 RF_AutoConfig_t *auto_config; 3163 RF_ComponentLabel_t *clabel; 3164 int c; 3165 int num_cols; 3166 int num_missing; 3167 int mod_counter; 3168 int mod_counter_found; 3169 int even_pair_failed; 3170 char parity_type; 3171 3172 3173 /* check to see that we have enough 'live' components 3174 of this set. If so, we can configure it if necessary */ 3175 3176 num_cols = cset->ac->clabel->num_columns; 3177 parity_type = cset->ac->clabel->parityConfig; 3178 3179 /* XXX Check for duplicate components!?!?!? */ 3180 3181 /* Determine what the mod_counter is supposed to be for this set. */ 3182 3183 mod_counter_found = 0; 3184 mod_counter = 0; 3185 ac = cset->ac; 3186 while(ac!=NULL) { 3187 if (mod_counter_found==0) { 3188 mod_counter = ac->clabel->mod_counter; 3189 mod_counter_found = 1; 3190 } else { 3191 if (ac->clabel->mod_counter > mod_counter) { 3192 mod_counter = ac->clabel->mod_counter; 3193 } 3194 } 3195 ac = ac->next; 3196 } 3197 3198 num_missing = 0; 3199 auto_config = cset->ac; 3200 3201 even_pair_failed = 0; 3202 for(c=0; c<num_cols; c++) { 3203 ac = auto_config; 3204 while(ac!=NULL) { 3205 if ((ac->clabel->column == c) && 3206 (ac->clabel->mod_counter == mod_counter)) { 3207 /* it's this one... */ 3208 #ifdef DEBUG 3209 printf("Found: %s at %d\n", 3210 ac->devname,c); 3211 #endif 3212 break; 3213 } 3214 ac=ac->next; 3215 } 3216 if (ac==NULL) { 3217 /* Didn't find one here! */ 3218 /* special case for RAID 1, especially 3219 where there are more than 2 3220 components (where RAIDframe treats 3221 things a little differently :( ) */ 3222 if (parity_type == '1') { 3223 if (c%2 == 0) { /* even component */ 3224 even_pair_failed = 1; 3225 } else { /* odd component. If 3226 we're failed, and 3227 so is the even 3228 component, it's 3229 "Good Night, Charlie" */ 3230 if (even_pair_failed == 1) { 3231 return(0); 3232 } 3233 } 3234 } else { 3235 /* normal accounting */ 3236 num_missing++; 3237 } 3238 } 3239 if ((parity_type == '1') && (c%2 == 1)) { 3240 /* Just did an even component, and we didn't 3241 bail.. reset the even_pair_failed flag, 3242 and go on to the next component.... */ 3243 even_pair_failed = 0; 3244 } 3245 } 3246 3247 clabel = cset->ac->clabel; 3248 3249 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3250 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3251 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3252 /* XXX this needs to be made *much* more general */ 3253 /* Too many failures */ 3254 return(0); 3255 } 3256 /* otherwise, all is well, and we've got enough to take a kick 3257 at autoconfiguring this set */ 3258 return(1); 3259 } 3260 3261 void 3262 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config, 3263 RF_Raid_t *raidPtr) 3264 { 3265 RF_ComponentLabel_t *clabel; 3266 int i; 3267 3268 clabel = ac->clabel; 3269 3270 /* 1. Fill in the common stuff */ 3271 config->numRow = clabel->num_rows = 1; 3272 config->numCol = clabel->num_columns; 3273 config->numSpare = 0; /* XXX should this be set here? */ 3274 config->sectPerSU = clabel->sectPerSU; 3275 config->SUsPerPU = clabel->SUsPerPU; 3276 config->SUsPerRU = clabel->SUsPerRU; 3277 config->parityConfig = clabel->parityConfig; 3278 /* XXX... */ 3279 strcpy(config->diskQueueType,"fifo"); 3280 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3281 config->layoutSpecificSize = 0; /* XXX ?? */ 3282 3283 while(ac!=NULL) { 3284 /* row/col values will be in range due to the checks 3285 in reasonable_label() */ 3286 strcpy(config->devnames[0][ac->clabel->column], 3287 ac->devname); 3288 ac = ac->next; 3289 } 3290 3291 for(i=0;i<RF_MAXDBGV;i++) { 3292 config->debugVars[i][0] = 0; 3293 } 3294 } 3295 3296 int 3297 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value) 3298 { 3299 RF_ComponentLabel_t clabel; 3300 struct vnode *vp; 3301 dev_t dev; 3302 int column; 3303 int sparecol; 3304 3305 raidPtr->autoconfigure = new_value; 3306 3307 for(column=0; column<raidPtr->numCol; column++) { 3308 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3309 dev = raidPtr->Disks[column].dev; 3310 vp = raidPtr->raid_cinfo[column].ci_vp; 3311 raidread_component_label(dev, vp, &clabel); 3312 clabel.autoconfigure = new_value; 3313 raidwrite_component_label(dev, vp, &clabel); 3314 } 3315 } 3316 for(column = 0; column < raidPtr->numSpare ; column++) { 3317 sparecol = raidPtr->numCol + column; 3318 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3319 dev = raidPtr->Disks[sparecol].dev; 3320 vp = raidPtr->raid_cinfo[sparecol].ci_vp; 3321 raidread_component_label(dev, vp, &clabel); 3322 clabel.autoconfigure = new_value; 3323 raidwrite_component_label(dev, vp, &clabel); 3324 } 3325 } 3326 return(new_value); 3327 } 3328 3329 int 3330 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value) 3331 { 3332 RF_ComponentLabel_t clabel; 3333 struct vnode *vp; 3334 dev_t dev; 3335 int column; 3336 int sparecol; 3337 3338 raidPtr->root_partition = new_value; 3339 for(column=0; column<raidPtr->numCol; column++) { 3340 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3341 dev = raidPtr->Disks[column].dev; 3342 vp = raidPtr->raid_cinfo[column].ci_vp; 3343 raidread_component_label(dev, vp, &clabel); 3344 clabel.root_partition = new_value; 3345 raidwrite_component_label(dev, vp, &clabel); 3346 } 3347 } 3348 for(column = 0; column < raidPtr->numSpare ; column++) { 3349 sparecol = raidPtr->numCol + column; 3350 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3351 dev = raidPtr->Disks[sparecol].dev; 3352 vp = raidPtr->raid_cinfo[sparecol].ci_vp; 3353 raidread_component_label(dev, vp, &clabel); 3354 clabel.root_partition = new_value; 3355 raidwrite_component_label(dev, vp, &clabel); 3356 } 3357 } 3358 return(new_value); 3359 } 3360 3361 void 3362 rf_release_all_vps(RF_ConfigSet_t *cset) 3363 { 3364 RF_AutoConfig_t *ac; 3365 3366 ac = cset->ac; 3367 while(ac!=NULL) { 3368 /* Close the vp, and give it back */ 3369 if (ac->vp) { 3370 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3371 VOP_CLOSE(ac->vp, FREAD, NOCRED); 3372 vput(ac->vp); 3373 ac->vp = NULL; 3374 } 3375 ac = ac->next; 3376 } 3377 } 3378 3379 3380 void 3381 rf_cleanup_config_set(RF_ConfigSet_t *cset) 3382 { 3383 RF_AutoConfig_t *ac; 3384 RF_AutoConfig_t *next_ac; 3385 3386 ac = cset->ac; 3387 while(ac!=NULL) { 3388 next_ac = ac->next; 3389 /* nuke the label */ 3390 free(ac->clabel, M_RAIDFRAME); 3391 /* cleanup the config structure */ 3392 free(ac, M_RAIDFRAME); 3393 /* "next.." */ 3394 ac = next_ac; 3395 } 3396 /* and, finally, nuke the config set */ 3397 free(cset, M_RAIDFRAME); 3398 } 3399 3400 3401 void 3402 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel) 3403 { 3404 /* current version number */ 3405 clabel->version = RF_COMPONENT_LABEL_VERSION; 3406 clabel->serial_number = raidPtr->serial_number; 3407 clabel->mod_counter = raidPtr->mod_counter; 3408 clabel->num_rows = 1; 3409 clabel->num_columns = raidPtr->numCol; 3410 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3411 clabel->status = rf_ds_optimal; /* "It's good!" */ 3412 3413 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3414 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3415 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3416 3417 clabel->blockSize = raidPtr->bytesPerSector; 3418 clabel->numBlocks = raidPtr->sectorsPerDisk; 3419 3420 /* XXX not portable */ 3421 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3422 clabel->maxOutstanding = raidPtr->maxOutstanding; 3423 clabel->autoconfigure = raidPtr->autoconfigure; 3424 clabel->root_partition = raidPtr->root_partition; 3425 clabel->last_unit = raidPtr->raidid; 3426 clabel->config_order = raidPtr->config_order; 3427 } 3428 3429 int 3430 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit) 3431 { 3432 RF_Raid_t *raidPtr; 3433 RF_Config_t *config; 3434 int raidID; 3435 int retcode; 3436 3437 #ifdef DEBUG 3438 printf("RAID autoconfigure\n"); 3439 #endif 3440 3441 retcode = 0; 3442 *unit = -1; 3443 3444 /* 1. Create a config structure */ 3445 3446 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3447 M_RAIDFRAME, 3448 M_NOWAIT); 3449 if (config==NULL) { 3450 printf("Out of mem!?!?\n"); 3451 /* XXX do something more intelligent here. */ 3452 return(1); 3453 } 3454 3455 memset(config, 0, sizeof(RF_Config_t)); 3456 3457 /* 3458 2. Figure out what RAID ID this one is supposed to live at 3459 See if we can get the same RAID dev that it was configured 3460 on last time.. 3461 */ 3462 3463 raidID = cset->ac->clabel->last_unit; 3464 if ((raidID < 0) || (raidID >= numraid)) { 3465 /* let's not wander off into lala land. */ 3466 raidID = numraid - 1; 3467 } 3468 if (raidPtrs[raidID]->valid != 0) { 3469 3470 /* 3471 Nope... Go looking for an alternative... 3472 Start high so we don't immediately use raid0 if that's 3473 not taken. 3474 */ 3475 3476 for(raidID = numraid - 1; raidID >= 0; raidID--) { 3477 if (raidPtrs[raidID]->valid == 0) { 3478 /* can use this one! */ 3479 break; 3480 } 3481 } 3482 } 3483 3484 if (raidID < 0) { 3485 /* punt... */ 3486 printf("Unable to auto configure this set!\n"); 3487 printf("(Out of RAID devs!)\n"); 3488 free(config, M_RAIDFRAME); 3489 return(1); 3490 } 3491 3492 #ifdef DEBUG 3493 printf("Configuring raid%d:\n",raidID); 3494 #endif 3495 3496 raidPtr = raidPtrs[raidID]; 3497 3498 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3499 raidPtr->raidid = raidID; 3500 raidPtr->openings = RAIDOUTSTANDING; 3501 3502 /* 3. Build the configuration structure */ 3503 rf_create_configuration(cset->ac, config, raidPtr); 3504 3505 /* 4. Do the configuration */ 3506 retcode = rf_Configure(raidPtr, config, cset->ac); 3507 3508 if (retcode == 0) { 3509 3510 raidinit(raidPtrs[raidID]); 3511 3512 rf_markalldirty(raidPtrs[raidID]); 3513 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3514 if (cset->ac->clabel->root_partition==1) { 3515 /* everything configured just fine. Make a note 3516 that this set is eligible to be root. */ 3517 cset->rootable = 1; 3518 /* XXX do this here? */ 3519 raidPtrs[raidID]->root_partition = 1; 3520 } 3521 } 3522 3523 /* 5. Cleanup */ 3524 free(config, M_RAIDFRAME); 3525 3526 *unit = raidID; 3527 return(retcode); 3528 } 3529 3530 void 3531 rf_disk_unbusy(RF_RaidAccessDesc_t *desc) 3532 { 3533 struct buf *bp; 3534 3535 bp = (struct buf *)desc->bp; 3536 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3537 (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ)); 3538 } 3539 3540 void 3541 rf_pool_init(struct pool *p, size_t size, const char *w_chan, 3542 size_t xmin, size_t xmax) 3543 { 3544 pool_init(p, size, 0, 0, 0, w_chan, NULL, IPL_BIO); 3545 pool_sethiwat(p, xmax); 3546 pool_prime(p, xmin); 3547 pool_setlowat(p, xmin); 3548 } 3549 3550 /* 3551 * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see 3552 * if there is IO pending and if that IO could possibly be done for a 3553 * given RAID set. Returns 0 if IO is waiting and can be done, 1 3554 * otherwise. 3555 * 3556 */ 3557 3558 int 3559 rf_buf_queue_check(int raidid) 3560 { 3561 if ((BUFQ_PEEK(raid_softc[raidid].buf_queue) != NULL) && 3562 raidPtrs[raidid]->openings > 0) { 3563 /* there is work to do */ 3564 return 0; 3565 } 3566 /* default is nothing to do */ 3567 return 1; 3568 } 3569 3570 int 3571 rf_getdisksize(struct vnode *vp, struct lwp *l, RF_RaidDisk_t *diskPtr) 3572 { 3573 struct partinfo dpart; 3574 struct dkwedge_info dkw; 3575 int error; 3576 3577 error = VOP_IOCTL(vp, DIOCGPART, &dpart, FREAD, l->l_cred); 3578 if (error == 0) { 3579 diskPtr->blockSize = dpart.disklab->d_secsize; 3580 diskPtr->numBlocks = dpart.part->p_size - rf_protectedSectors; 3581 diskPtr->partitionSize = dpart.part->p_size; 3582 return 0; 3583 } 3584 3585 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, l->l_cred); 3586 if (error == 0) { 3587 diskPtr->blockSize = 512; /* XXX */ 3588 diskPtr->numBlocks = dkw.dkw_size - rf_protectedSectors; 3589 diskPtr->partitionSize = dkw.dkw_size; 3590 return 0; 3591 } 3592 return error; 3593 } 3594 3595 static int 3596 raid_match(struct device *self, struct cfdata *cfdata, 3597 void *aux) 3598 { 3599 return 1; 3600 } 3601 3602 static void 3603 raid_attach(struct device *parent, struct device *self, 3604 void *aux) 3605 { 3606 3607 } 3608 3609 3610 static int 3611 raid_detach(struct device *self, int flags) 3612 { 3613 struct raid_softc *rs = (struct raid_softc *)self; 3614 3615 if (rs->sc_flags & RAIDF_INITED) 3616 return EBUSY; 3617 3618 return 0; 3619 } 3620 3621 static void 3622 rf_set_properties(struct raid_softc *rs, RF_Raid_t *raidPtr) 3623 { 3624 prop_dictionary_t disk_info, odisk_info, geom; 3625 disk_info = prop_dictionary_create(); 3626 geom = prop_dictionary_create(); 3627 prop_dictionary_set_uint64(geom, "sectors-per-unit", 3628 raidPtr->totalSectors); 3629 prop_dictionary_set_uint32(geom, "sector-size", 3630 raidPtr->bytesPerSector); 3631 3632 prop_dictionary_set_uint16(geom, "sectors-per-track", 3633 raidPtr->Layout.dataSectorsPerStripe); 3634 prop_dictionary_set_uint16(geom, "tracks-per-cylinder", 3635 4 * raidPtr->numCol); 3636 3637 prop_dictionary_set_uint64(geom, "cylinders-per-unit", 3638 raidPtr->totalSectors / (raidPtr->Layout.dataSectorsPerStripe * 3639 (4 * raidPtr->numCol))); 3640 3641 prop_dictionary_set(disk_info, "geometry", geom); 3642 prop_object_release(geom); 3643 prop_dictionary_set(device_properties(rs->sc_dev), 3644 "disk-info", disk_info); 3645 odisk_info = rs->sc_dkdev.dk_info; 3646 rs->sc_dkdev.dk_info = disk_info; 3647 if (odisk_info) 3648 prop_object_release(odisk_info); 3649 } 3650