1 /* $NetBSD: rf_netbsdkintf.c,v 1.274 2010/08/08 18:25:14 chs Exp $ */ 2 /*- 3 * Copyright (c) 1996, 1997, 1998, 2008 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Greg Oster; Jason R. Thorpe. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28 * POSSIBILITY OF SUCH DAMAGE. 29 */ 30 31 /* 32 * Copyright (c) 1990, 1993 33 * The Regents of the University of California. All rights reserved. 34 * 35 * This code is derived from software contributed to Berkeley by 36 * the Systems Programming Group of the University of Utah Computer 37 * Science Department. 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. Neither the name of the University nor the names of its contributors 48 * may be used to endorse or promote products derived from this software 49 * without specific prior written permission. 50 * 51 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 52 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 53 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 54 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 55 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 56 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 57 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 58 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 59 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 60 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 61 * SUCH DAMAGE. 62 * 63 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 64 * 65 * @(#)cd.c 8.2 (Berkeley) 11/16/93 66 */ 67 68 /* 69 * Copyright (c) 1988 University of Utah. 70 * 71 * This code is derived from software contributed to Berkeley by 72 * the Systems Programming Group of the University of Utah Computer 73 * Science Department. 74 * 75 * Redistribution and use in source and binary forms, with or without 76 * modification, are permitted provided that the following conditions 77 * are met: 78 * 1. Redistributions of source code must retain the above copyright 79 * notice, this list of conditions and the following disclaimer. 80 * 2. Redistributions in binary form must reproduce the above copyright 81 * notice, this list of conditions and the following disclaimer in the 82 * documentation and/or other materials provided with the distribution. 83 * 3. All advertising materials mentioning features or use of this software 84 * must display the following acknowledgement: 85 * This product includes software developed by the University of 86 * California, Berkeley and its contributors. 87 * 4. Neither the name of the University nor the names of its contributors 88 * may be used to endorse or promote products derived from this software 89 * without specific prior written permission. 90 * 91 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 92 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 93 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 94 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 95 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 96 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 97 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 98 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 99 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 100 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 101 * SUCH DAMAGE. 102 * 103 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 104 * 105 * @(#)cd.c 8.2 (Berkeley) 11/16/93 106 */ 107 108 /* 109 * Copyright (c) 1995 Carnegie-Mellon University. 110 * All rights reserved. 111 * 112 * Authors: Mark Holland, Jim Zelenka 113 * 114 * Permission to use, copy, modify and distribute this software and 115 * its documentation is hereby granted, provided that both the copyright 116 * notice and this permission notice appear in all copies of the 117 * software, derivative works or modified versions, and any portions 118 * thereof, and that both notices appear in supporting documentation. 119 * 120 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 121 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 122 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 123 * 124 * Carnegie Mellon requests users of this software to return to 125 * 126 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 127 * School of Computer Science 128 * Carnegie Mellon University 129 * Pittsburgh PA 15213-3890 130 * 131 * any improvements or extensions that they make and grant Carnegie the 132 * rights to redistribute these changes. 133 */ 134 135 /*********************************************************** 136 * 137 * rf_kintf.c -- the kernel interface routines for RAIDframe 138 * 139 ***********************************************************/ 140 141 #include <sys/cdefs.h> 142 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.274 2010/08/08 18:25:14 chs Exp $"); 143 144 #ifdef _KERNEL_OPT 145 #include "opt_compat_netbsd.h" 146 #include "opt_raid_autoconfig.h" 147 #include "raid.h" 148 #endif 149 150 #include <sys/param.h> 151 #include <sys/errno.h> 152 #include <sys/pool.h> 153 #include <sys/proc.h> 154 #include <sys/queue.h> 155 #include <sys/disk.h> 156 #include <sys/device.h> 157 #include <sys/stat.h> 158 #include <sys/ioctl.h> 159 #include <sys/fcntl.h> 160 #include <sys/systm.h> 161 #include <sys/vnode.h> 162 #include <sys/disklabel.h> 163 #include <sys/conf.h> 164 #include <sys/buf.h> 165 #include <sys/bufq.h> 166 #include <sys/reboot.h> 167 #include <sys/kauth.h> 168 169 #include <prop/proplib.h> 170 171 #include <dev/raidframe/raidframevar.h> 172 #include <dev/raidframe/raidframeio.h> 173 #include <dev/raidframe/rf_paritymap.h> 174 175 #include "rf_raid.h" 176 #include "rf_copyback.h" 177 #include "rf_dag.h" 178 #include "rf_dagflags.h" 179 #include "rf_desc.h" 180 #include "rf_diskqueue.h" 181 #include "rf_etimer.h" 182 #include "rf_general.h" 183 #include "rf_kintf.h" 184 #include "rf_options.h" 185 #include "rf_driver.h" 186 #include "rf_parityscan.h" 187 #include "rf_threadstuff.h" 188 189 #ifdef COMPAT_50 190 #include "rf_compat50.h" 191 #endif 192 193 #ifdef DEBUG 194 int rf_kdebug_level = 0; 195 #define db1_printf(a) if (rf_kdebug_level > 0) printf a 196 #else /* DEBUG */ 197 #define db1_printf(a) { } 198 #endif /* DEBUG */ 199 200 static RF_Raid_t **raidPtrs; /* global raid device descriptors */ 201 202 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 203 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex) 204 205 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a 206 * spare table */ 207 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from 208 * installation process */ 209 #endif 210 211 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures"); 212 213 /* prototypes */ 214 static void KernelWakeupFunc(struct buf *); 215 static void InitBP(struct buf *, struct vnode *, unsigned, 216 dev_t, RF_SectorNum_t, RF_SectorCount_t, void *, void (*) (struct buf *), 217 void *, int, struct proc *); 218 static void raidinit(RF_Raid_t *); 219 220 void raidattach(int); 221 static int raid_match(device_t, cfdata_t, void *); 222 static void raid_attach(device_t, device_t, void *); 223 static int raid_detach(device_t, int); 224 225 static int raidread_component_area(dev_t, struct vnode *, void *, size_t, 226 daddr_t, daddr_t); 227 static int raidwrite_component_area(dev_t, struct vnode *, void *, size_t, 228 daddr_t, daddr_t, int); 229 230 static int raidwrite_component_label(dev_t, struct vnode *, 231 RF_ComponentLabel_t *); 232 static int raidread_component_label(dev_t, struct vnode *, 233 RF_ComponentLabel_t *); 234 235 236 dev_type_open(raidopen); 237 dev_type_close(raidclose); 238 dev_type_read(raidread); 239 dev_type_write(raidwrite); 240 dev_type_ioctl(raidioctl); 241 dev_type_strategy(raidstrategy); 242 dev_type_dump(raiddump); 243 dev_type_size(raidsize); 244 245 const struct bdevsw raid_bdevsw = { 246 raidopen, raidclose, raidstrategy, raidioctl, 247 raiddump, raidsize, D_DISK 248 }; 249 250 const struct cdevsw raid_cdevsw = { 251 raidopen, raidclose, raidread, raidwrite, raidioctl, 252 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 253 }; 254 255 static struct dkdriver rf_dkdriver = { raidstrategy, minphys }; 256 257 /* XXX Not sure if the following should be replacing the raidPtrs above, 258 or if it should be used in conjunction with that... 259 */ 260 261 struct raid_softc { 262 device_t sc_dev; 263 int sc_flags; /* flags */ 264 int sc_cflags; /* configuration flags */ 265 uint64_t sc_size; /* size of the raid device */ 266 char sc_xname[20]; /* XXX external name */ 267 struct disk sc_dkdev; /* generic disk device info */ 268 struct bufq_state *buf_queue; /* used for the device queue */ 269 }; 270 /* sc_flags */ 271 #define RAIDF_INITED 0x01 /* unit has been initialized */ 272 #define RAIDF_WLABEL 0x02 /* label area is writable */ 273 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */ 274 #define RAIDF_SHUTDOWN 0x08 /* unit is being shutdown */ 275 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */ 276 #define RAIDF_LOCKED 0x80 /* unit is locked */ 277 278 #define raidunit(x) DISKUNIT(x) 279 int numraid = 0; 280 281 extern struct cfdriver raid_cd; 282 CFATTACH_DECL3_NEW(raid, sizeof(struct raid_softc), 283 raid_match, raid_attach, raid_detach, NULL, NULL, NULL, 284 DVF_DETACH_SHUTDOWN); 285 286 /* 287 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device. 288 * Be aware that large numbers can allow the driver to consume a lot of 289 * kernel memory, especially on writes, and in degraded mode reads. 290 * 291 * For example: with a stripe width of 64 blocks (32k) and 5 disks, 292 * a single 64K write will typically require 64K for the old data, 293 * 64K for the old parity, and 64K for the new parity, for a total 294 * of 192K (if the parity buffer is not re-used immediately). 295 * Even it if is used immediately, that's still 128K, which when multiplied 296 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data. 297 * 298 * Now in degraded mode, for example, a 64K read on the above setup may 299 * require data reconstruction, which will require *all* of the 4 remaining 300 * disks to participate -- 4 * 32K/disk == 128K again. 301 */ 302 303 #ifndef RAIDOUTSTANDING 304 #define RAIDOUTSTANDING 6 305 #endif 306 307 #define RAIDLABELDEV(dev) \ 308 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART)) 309 310 /* declared here, and made public, for the benefit of KVM stuff.. */ 311 struct raid_softc *raid_softc; 312 313 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *, 314 struct disklabel *); 315 static void raidgetdisklabel(dev_t); 316 static void raidmakedisklabel(struct raid_softc *); 317 318 static int raidlock(struct raid_softc *); 319 static void raidunlock(struct raid_softc *); 320 321 static int raid_detach_unlocked(struct raid_softc *); 322 323 static void rf_markalldirty(RF_Raid_t *); 324 static void rf_set_properties(struct raid_softc *, RF_Raid_t *); 325 326 void rf_ReconThread(struct rf_recon_req *); 327 void rf_RewriteParityThread(RF_Raid_t *raidPtr); 328 void rf_CopybackThread(RF_Raid_t *raidPtr); 329 void rf_ReconstructInPlaceThread(struct rf_recon_req *); 330 int rf_autoconfig(device_t); 331 void rf_buildroothack(RF_ConfigSet_t *); 332 333 RF_AutoConfig_t *rf_find_raid_components(void); 334 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *); 335 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *); 336 static int rf_reasonable_label(RF_ComponentLabel_t *); 337 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *); 338 int rf_set_autoconfig(RF_Raid_t *, int); 339 int rf_set_rootpartition(RF_Raid_t *, int); 340 void rf_release_all_vps(RF_ConfigSet_t *); 341 void rf_cleanup_config_set(RF_ConfigSet_t *); 342 int rf_have_enough_components(RF_ConfigSet_t *); 343 int rf_auto_config_set(RF_ConfigSet_t *, int *); 344 345 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not 346 allow autoconfig to take place. 347 Note that this is overridden by having 348 RAID_AUTOCONFIG as an option in the 349 kernel config file. */ 350 351 struct RF_Pools_s rf_pools; 352 353 void 354 raidattach(int num) 355 { 356 int raidID; 357 int i, rc; 358 359 aprint_debug("raidattach: Asked for %d units\n", num); 360 361 if (num <= 0) { 362 #ifdef DIAGNOSTIC 363 panic("raidattach: count <= 0"); 364 #endif 365 return; 366 } 367 /* This is where all the initialization stuff gets done. */ 368 369 numraid = num; 370 371 /* Make some space for requested number of units... */ 372 373 RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **)); 374 if (raidPtrs == NULL) { 375 panic("raidPtrs is NULL!!"); 376 } 377 378 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 379 rf_mutex_init(&rf_sparet_wait_mutex); 380 381 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL; 382 #endif 383 384 for (i = 0; i < num; i++) 385 raidPtrs[i] = NULL; 386 rc = rf_BootRaidframe(); 387 if (rc == 0) 388 aprint_verbose("Kernelized RAIDframe activated\n"); 389 else 390 panic("Serious error booting RAID!!"); 391 392 /* put together some datastructures like the CCD device does.. This 393 * lets us lock the device and what-not when it gets opened. */ 394 395 raid_softc = (struct raid_softc *) 396 malloc(num * sizeof(struct raid_softc), 397 M_RAIDFRAME, M_NOWAIT); 398 if (raid_softc == NULL) { 399 aprint_error("WARNING: no memory for RAIDframe driver\n"); 400 return; 401 } 402 403 memset(raid_softc, 0, num * sizeof(struct raid_softc)); 404 405 for (raidID = 0; raidID < num; raidID++) { 406 bufq_alloc(&raid_softc[raidID].buf_queue, "fcfs", 0); 407 408 RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t), 409 (RF_Raid_t *)); 410 if (raidPtrs[raidID] == NULL) { 411 aprint_error("WARNING: raidPtrs[%d] is NULL\n", raidID); 412 numraid = raidID; 413 return; 414 } 415 } 416 417 if (config_cfattach_attach(raid_cd.cd_name, &raid_ca)) { 418 aprint_error("raidattach: config_cfattach_attach failed?\n"); 419 } 420 421 #ifdef RAID_AUTOCONFIG 422 raidautoconfig = 1; 423 #endif 424 425 /* 426 * Register a finalizer which will be used to auto-config RAID 427 * sets once all real hardware devices have been found. 428 */ 429 if (config_finalize_register(NULL, rf_autoconfig) != 0) 430 aprint_error("WARNING: unable to register RAIDframe finalizer\n"); 431 } 432 433 int 434 rf_autoconfig(device_t self) 435 { 436 RF_AutoConfig_t *ac_list; 437 RF_ConfigSet_t *config_sets; 438 439 if (raidautoconfig == 0) 440 return (0); 441 442 /* XXX This code can only be run once. */ 443 raidautoconfig = 0; 444 445 /* 1. locate all RAID components on the system */ 446 aprint_debug("Searching for RAID components...\n"); 447 ac_list = rf_find_raid_components(); 448 449 /* 2. Sort them into their respective sets. */ 450 config_sets = rf_create_auto_sets(ac_list); 451 452 /* 453 * 3. Evaluate each set andconfigure the valid ones. 454 * This gets done in rf_buildroothack(). 455 */ 456 rf_buildroothack(config_sets); 457 458 return 1; 459 } 460 461 void 462 rf_buildroothack(RF_ConfigSet_t *config_sets) 463 { 464 RF_ConfigSet_t *cset; 465 RF_ConfigSet_t *next_cset; 466 int retcode; 467 int raidID; 468 int rootID; 469 int col; 470 int num_root; 471 char *devname; 472 473 rootID = 0; 474 num_root = 0; 475 cset = config_sets; 476 while (cset != NULL) { 477 next_cset = cset->next; 478 if (rf_have_enough_components(cset) && 479 cset->ac->clabel->autoconfigure==1) { 480 retcode = rf_auto_config_set(cset,&raidID); 481 if (!retcode) { 482 aprint_debug("raid%d: configured ok\n", raidID); 483 if (cset->rootable) { 484 rootID = raidID; 485 num_root++; 486 } 487 } else { 488 /* The autoconfig didn't work :( */ 489 aprint_debug("Autoconfig failed with code %d for raid%d\n", retcode, raidID); 490 rf_release_all_vps(cset); 491 } 492 } else { 493 /* we're not autoconfiguring this set... 494 release the associated resources */ 495 rf_release_all_vps(cset); 496 } 497 /* cleanup */ 498 rf_cleanup_config_set(cset); 499 cset = next_cset; 500 } 501 502 /* if the user has specified what the root device should be 503 then we don't touch booted_device or boothowto... */ 504 505 if (rootspec != NULL) 506 return; 507 508 /* we found something bootable... */ 509 510 if (num_root == 1) { 511 booted_device = raid_softc[rootID].sc_dev; 512 } else if (num_root > 1) { 513 514 /* 515 * Maybe the MD code can help. If it cannot, then 516 * setroot() will discover that we have no 517 * booted_device and will ask the user if nothing was 518 * hardwired in the kernel config file 519 */ 520 521 if (booted_device == NULL) 522 cpu_rootconf(); 523 if (booted_device == NULL) 524 return; 525 526 num_root = 0; 527 for (raidID = 0; raidID < numraid; raidID++) { 528 if (raidPtrs[raidID]->valid == 0) 529 continue; 530 531 if (raidPtrs[raidID]->root_partition == 0) 532 continue; 533 534 for (col = 0; col < raidPtrs[raidID]->numCol; col++) { 535 devname = raidPtrs[raidID]->Disks[col].devname; 536 devname += sizeof("/dev/") - 1; 537 if (strncmp(devname, device_xname(booted_device), 538 strlen(device_xname(booted_device))) != 0) 539 continue; 540 aprint_debug("raid%d includes boot device %s\n", 541 raidID, devname); 542 num_root++; 543 rootID = raidID; 544 } 545 } 546 547 if (num_root == 1) { 548 booted_device = raid_softc[rootID].sc_dev; 549 } else { 550 /* we can't guess.. require the user to answer... */ 551 boothowto |= RB_ASKNAME; 552 } 553 } 554 } 555 556 557 int 558 raidsize(dev_t dev) 559 { 560 struct raid_softc *rs; 561 struct disklabel *lp; 562 int part, unit, omask, size; 563 564 unit = raidunit(dev); 565 if (unit >= numraid) 566 return (-1); 567 rs = &raid_softc[unit]; 568 569 if ((rs->sc_flags & RAIDF_INITED) == 0) 570 return (-1); 571 572 part = DISKPART(dev); 573 omask = rs->sc_dkdev.dk_openmask & (1 << part); 574 lp = rs->sc_dkdev.dk_label; 575 576 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curlwp)) 577 return (-1); 578 579 if (lp->d_partitions[part].p_fstype != FS_SWAP) 580 size = -1; 581 else 582 size = lp->d_partitions[part].p_size * 583 (lp->d_secsize / DEV_BSIZE); 584 585 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curlwp)) 586 return (-1); 587 588 return (size); 589 590 } 591 592 int 593 raiddump(dev_t dev, daddr_t blkno, void *va, size_t size) 594 { 595 int unit = raidunit(dev); 596 struct raid_softc *rs; 597 const struct bdevsw *bdev; 598 struct disklabel *lp; 599 RF_Raid_t *raidPtr; 600 daddr_t offset; 601 int part, c, sparecol, j, scol, dumpto; 602 int error = 0; 603 604 if (unit >= numraid) 605 return (ENXIO); 606 607 rs = &raid_softc[unit]; 608 raidPtr = raidPtrs[unit]; 609 610 if ((rs->sc_flags & RAIDF_INITED) == 0) 611 return ENXIO; 612 613 /* we only support dumping to RAID 1 sets */ 614 if (raidPtr->Layout.numDataCol != 1 || 615 raidPtr->Layout.numParityCol != 1) 616 return EINVAL; 617 618 619 if ((error = raidlock(rs)) != 0) 620 return error; 621 622 if (size % DEV_BSIZE != 0) { 623 error = EINVAL; 624 goto out; 625 } 626 627 if (blkno + size / DEV_BSIZE > rs->sc_size) { 628 printf("%s: blkno (%" PRIu64 ") + size / DEV_BSIZE (%zu) > " 629 "sc->sc_size (%" PRIu64 ")\n", __func__, blkno, 630 size / DEV_BSIZE, rs->sc_size); 631 error = EINVAL; 632 goto out; 633 } 634 635 part = DISKPART(dev); 636 lp = rs->sc_dkdev.dk_label; 637 offset = lp->d_partitions[part].p_offset + RF_PROTECTED_SECTORS; 638 639 /* figure out what device is alive.. */ 640 641 /* 642 Look for a component to dump to. The preference for the 643 component to dump to is as follows: 644 1) the master 645 2) a used_spare of the master 646 3) the slave 647 4) a used_spare of the slave 648 */ 649 650 dumpto = -1; 651 for (c = 0; c < raidPtr->numCol; c++) { 652 if (raidPtr->Disks[c].status == rf_ds_optimal) { 653 /* this might be the one */ 654 dumpto = c; 655 break; 656 } 657 } 658 659 /* 660 At this point we have possibly selected a live master or a 661 live slave. We now check to see if there is a spared 662 master (or a spared slave), if we didn't find a live master 663 or a live slave. 664 */ 665 666 for (c = 0; c < raidPtr->numSpare; c++) { 667 sparecol = raidPtr->numCol + c; 668 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 669 /* How about this one? */ 670 scol = -1; 671 for(j=0;j<raidPtr->numCol;j++) { 672 if (raidPtr->Disks[j].spareCol == sparecol) { 673 scol = j; 674 break; 675 } 676 } 677 if (scol == 0) { 678 /* 679 We must have found a spared master! 680 We'll take that over anything else 681 found so far. (We couldn't have 682 found a real master before, since 683 this is a used spare, and it's 684 saying that it's replacing the 685 master.) On reboot (with 686 autoconfiguration turned on) 687 sparecol will become the 1st 688 component (component0) of this set. 689 */ 690 dumpto = sparecol; 691 break; 692 } else if (scol != -1) { 693 /* 694 Must be a spared slave. We'll dump 695 to that if we havn't found anything 696 else so far. 697 */ 698 if (dumpto == -1) 699 dumpto = sparecol; 700 } 701 } 702 } 703 704 if (dumpto == -1) { 705 /* we couldn't find any live components to dump to!?!? 706 */ 707 error = EINVAL; 708 goto out; 709 } 710 711 bdev = bdevsw_lookup(raidPtr->Disks[dumpto].dev); 712 713 /* 714 Note that blkno is relative to this particular partition. 715 By adding the offset of this partition in the RAID 716 set, and also adding RF_PROTECTED_SECTORS, we get a 717 value that is relative to the partition used for the 718 underlying component. 719 */ 720 721 error = (*bdev->d_dump)(raidPtr->Disks[dumpto].dev, 722 blkno + offset, va, size); 723 724 out: 725 raidunlock(rs); 726 727 return error; 728 } 729 /* ARGSUSED */ 730 int 731 raidopen(dev_t dev, int flags, int fmt, 732 struct lwp *l) 733 { 734 int unit = raidunit(dev); 735 struct raid_softc *rs; 736 struct disklabel *lp; 737 int part, pmask; 738 int error = 0; 739 740 if (unit >= numraid) 741 return (ENXIO); 742 rs = &raid_softc[unit]; 743 744 if ((error = raidlock(rs)) != 0) 745 return (error); 746 747 if ((rs->sc_flags & RAIDF_SHUTDOWN) != 0) { 748 error = EBUSY; 749 goto bad; 750 } 751 752 lp = rs->sc_dkdev.dk_label; 753 754 part = DISKPART(dev); 755 756 /* 757 * If there are wedges, and this is not RAW_PART, then we 758 * need to fail. 759 */ 760 if (rs->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) { 761 error = EBUSY; 762 goto bad; 763 } 764 pmask = (1 << part); 765 766 if ((rs->sc_flags & RAIDF_INITED) && 767 (rs->sc_dkdev.dk_openmask == 0)) 768 raidgetdisklabel(dev); 769 770 /* make sure that this partition exists */ 771 772 if (part != RAW_PART) { 773 if (((rs->sc_flags & RAIDF_INITED) == 0) || 774 ((part >= lp->d_npartitions) || 775 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 776 error = ENXIO; 777 goto bad; 778 } 779 } 780 /* Prevent this unit from being unconfigured while open. */ 781 switch (fmt) { 782 case S_IFCHR: 783 rs->sc_dkdev.dk_copenmask |= pmask; 784 break; 785 786 case S_IFBLK: 787 rs->sc_dkdev.dk_bopenmask |= pmask; 788 break; 789 } 790 791 if ((rs->sc_dkdev.dk_openmask == 0) && 792 ((rs->sc_flags & RAIDF_INITED) != 0)) { 793 /* First one... mark things as dirty... Note that we *MUST* 794 have done a configure before this. I DO NOT WANT TO BE 795 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED 796 THAT THEY BELONG TOGETHER!!!!! */ 797 /* XXX should check to see if we're only open for reading 798 here... If so, we needn't do this, but then need some 799 other way of keeping track of what's happened.. */ 800 801 rf_markalldirty(raidPtrs[unit]); 802 } 803 804 805 rs->sc_dkdev.dk_openmask = 806 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 807 808 bad: 809 raidunlock(rs); 810 811 return (error); 812 813 814 } 815 /* ARGSUSED */ 816 int 817 raidclose(dev_t dev, int flags, int fmt, struct lwp *l) 818 { 819 int unit = raidunit(dev); 820 struct raid_softc *rs; 821 int error = 0; 822 int part; 823 824 if (unit >= numraid) 825 return (ENXIO); 826 rs = &raid_softc[unit]; 827 828 if ((error = raidlock(rs)) != 0) 829 return (error); 830 831 part = DISKPART(dev); 832 833 /* ...that much closer to allowing unconfiguration... */ 834 switch (fmt) { 835 case S_IFCHR: 836 rs->sc_dkdev.dk_copenmask &= ~(1 << part); 837 break; 838 839 case S_IFBLK: 840 rs->sc_dkdev.dk_bopenmask &= ~(1 << part); 841 break; 842 } 843 rs->sc_dkdev.dk_openmask = 844 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask; 845 846 if ((rs->sc_dkdev.dk_openmask == 0) && 847 ((rs->sc_flags & RAIDF_INITED) != 0)) { 848 /* Last one... device is not unconfigured yet. 849 Device shutdown has taken care of setting the 850 clean bits if RAIDF_INITED is not set 851 mark things as clean... */ 852 853 rf_update_component_labels(raidPtrs[unit], 854 RF_FINAL_COMPONENT_UPDATE); 855 856 /* If the kernel is shutting down, it will detach 857 * this RAID set soon enough. 858 */ 859 } 860 861 raidunlock(rs); 862 return (0); 863 864 } 865 866 void 867 raidstrategy(struct buf *bp) 868 { 869 int s; 870 871 unsigned int raidID = raidunit(bp->b_dev); 872 RF_Raid_t *raidPtr; 873 struct raid_softc *rs = &raid_softc[raidID]; 874 int wlabel; 875 876 if ((rs->sc_flags & RAIDF_INITED) ==0) { 877 bp->b_error = ENXIO; 878 goto done; 879 } 880 if (raidID >= numraid || !raidPtrs[raidID]) { 881 bp->b_error = ENODEV; 882 goto done; 883 } 884 raidPtr = raidPtrs[raidID]; 885 if (!raidPtr->valid) { 886 bp->b_error = ENODEV; 887 goto done; 888 } 889 if (bp->b_bcount == 0) { 890 db1_printf(("b_bcount is zero..\n")); 891 goto done; 892 } 893 894 /* 895 * Do bounds checking and adjust transfer. If there's an 896 * error, the bounds check will flag that for us. 897 */ 898 899 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING); 900 if (DISKPART(bp->b_dev) == RAW_PART) { 901 uint64_t size; /* device size in DEV_BSIZE unit */ 902 903 if (raidPtr->logBytesPerSector > DEV_BSHIFT) { 904 size = raidPtr->totalSectors << 905 (raidPtr->logBytesPerSector - DEV_BSHIFT); 906 } else { 907 size = raidPtr->totalSectors >> 908 (DEV_BSHIFT - raidPtr->logBytesPerSector); 909 } 910 if (bounds_check_with_mediasize(bp, DEV_BSIZE, size) <= 0) { 911 goto done; 912 } 913 } else { 914 if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) { 915 db1_printf(("Bounds check failed!!:%d %d\n", 916 (int) bp->b_blkno, (int) wlabel)); 917 goto done; 918 } 919 } 920 s = splbio(); 921 922 bp->b_resid = 0; 923 924 /* stuff it onto our queue */ 925 bufq_put(rs->buf_queue, bp); 926 927 /* scheduled the IO to happen at the next convenient time */ 928 wakeup(&(raidPtrs[raidID]->iodone)); 929 930 splx(s); 931 return; 932 933 done: 934 bp->b_resid = bp->b_bcount; 935 biodone(bp); 936 } 937 /* ARGSUSED */ 938 int 939 raidread(dev_t dev, struct uio *uio, int flags) 940 { 941 int unit = raidunit(dev); 942 struct raid_softc *rs; 943 944 if (unit >= numraid) 945 return (ENXIO); 946 rs = &raid_softc[unit]; 947 948 if ((rs->sc_flags & RAIDF_INITED) == 0) 949 return (ENXIO); 950 951 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio)); 952 953 } 954 /* ARGSUSED */ 955 int 956 raidwrite(dev_t dev, struct uio *uio, int flags) 957 { 958 int unit = raidunit(dev); 959 struct raid_softc *rs; 960 961 if (unit >= numraid) 962 return (ENXIO); 963 rs = &raid_softc[unit]; 964 965 if ((rs->sc_flags & RAIDF_INITED) == 0) 966 return (ENXIO); 967 968 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio)); 969 970 } 971 972 static int 973 raid_detach_unlocked(struct raid_softc *rs) 974 { 975 int error; 976 RF_Raid_t *raidPtr; 977 978 raidPtr = raidPtrs[device_unit(rs->sc_dev)]; 979 980 /* 981 * If somebody has a partition mounted, we shouldn't 982 * shutdown. 983 */ 984 if (rs->sc_dkdev.dk_openmask != 0) 985 return EBUSY; 986 987 if ((rs->sc_flags & RAIDF_INITED) == 0) 988 ; /* not initialized: nothing to do */ 989 else if ((error = rf_Shutdown(raidPtr)) != 0) 990 return error; 991 else 992 rs->sc_flags &= ~(RAIDF_INITED|RAIDF_SHUTDOWN); 993 994 /* Detach the disk. */ 995 disk_detach(&rs->sc_dkdev); 996 disk_destroy(&rs->sc_dkdev); 997 998 return 0; 999 } 1000 1001 int 1002 raidioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 1003 { 1004 int unit = raidunit(dev); 1005 int error = 0; 1006 int part, pmask; 1007 cfdata_t cf; 1008 struct raid_softc *rs; 1009 RF_Config_t *k_cfg, *u_cfg; 1010 RF_Raid_t *raidPtr; 1011 RF_RaidDisk_t *diskPtr; 1012 RF_AccTotals_t *totals; 1013 RF_DeviceConfig_t *d_cfg, **ucfgp; 1014 u_char *specific_buf; 1015 int retcode = 0; 1016 int column; 1017 /* int raidid; */ 1018 struct rf_recon_req *rrcopy, *rr; 1019 RF_ComponentLabel_t *clabel; 1020 RF_ComponentLabel_t *ci_label; 1021 RF_ComponentLabel_t **clabel_ptr; 1022 RF_SingleComponent_t *sparePtr,*componentPtr; 1023 RF_SingleComponent_t component; 1024 RF_ProgressInfo_t progressInfo, **progressInfoPtr; 1025 int i, j, d; 1026 #ifdef __HAVE_OLD_DISKLABEL 1027 struct disklabel newlabel; 1028 #endif 1029 struct dkwedge_info *dkw; 1030 1031 if (unit >= numraid) 1032 return (ENXIO); 1033 rs = &raid_softc[unit]; 1034 raidPtr = raidPtrs[unit]; 1035 1036 db1_printf(("raidioctl: %d %d %d %d\n", (int) dev, 1037 (int) DISKPART(dev), (int) unit, (int) cmd)); 1038 1039 /* Must be open for writes for these commands... */ 1040 switch (cmd) { 1041 #ifdef DIOCGSECTORSIZE 1042 case DIOCGSECTORSIZE: 1043 *(u_int *)data = raidPtr->bytesPerSector; 1044 return 0; 1045 case DIOCGMEDIASIZE: 1046 *(off_t *)data = 1047 (off_t)raidPtr->totalSectors * raidPtr->bytesPerSector; 1048 return 0; 1049 #endif 1050 case DIOCSDINFO: 1051 case DIOCWDINFO: 1052 #ifdef __HAVE_OLD_DISKLABEL 1053 case ODIOCWDINFO: 1054 case ODIOCSDINFO: 1055 #endif 1056 case DIOCWLABEL: 1057 case DIOCAWEDGE: 1058 case DIOCDWEDGE: 1059 if ((flag & FWRITE) == 0) 1060 return (EBADF); 1061 } 1062 1063 /* Must be initialized for these... */ 1064 switch (cmd) { 1065 case DIOCGDINFO: 1066 case DIOCSDINFO: 1067 case DIOCWDINFO: 1068 #ifdef __HAVE_OLD_DISKLABEL 1069 case ODIOCGDINFO: 1070 case ODIOCWDINFO: 1071 case ODIOCSDINFO: 1072 case ODIOCGDEFLABEL: 1073 #endif 1074 case DIOCGPART: 1075 case DIOCWLABEL: 1076 case DIOCGDEFLABEL: 1077 case DIOCAWEDGE: 1078 case DIOCDWEDGE: 1079 case DIOCLWEDGES: 1080 case DIOCCACHESYNC: 1081 case RAIDFRAME_SHUTDOWN: 1082 case RAIDFRAME_REWRITEPARITY: 1083 case RAIDFRAME_GET_INFO: 1084 case RAIDFRAME_RESET_ACCTOTALS: 1085 case RAIDFRAME_GET_ACCTOTALS: 1086 case RAIDFRAME_KEEP_ACCTOTALS: 1087 case RAIDFRAME_GET_SIZE: 1088 case RAIDFRAME_FAIL_DISK: 1089 case RAIDFRAME_COPYBACK: 1090 case RAIDFRAME_CHECK_RECON_STATUS: 1091 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1092 case RAIDFRAME_GET_COMPONENT_LABEL: 1093 case RAIDFRAME_SET_COMPONENT_LABEL: 1094 case RAIDFRAME_ADD_HOT_SPARE: 1095 case RAIDFRAME_REMOVE_HOT_SPARE: 1096 case RAIDFRAME_INIT_LABELS: 1097 case RAIDFRAME_REBUILD_IN_PLACE: 1098 case RAIDFRAME_CHECK_PARITY: 1099 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1100 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1101 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1102 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1103 case RAIDFRAME_SET_AUTOCONFIG: 1104 case RAIDFRAME_SET_ROOT: 1105 case RAIDFRAME_DELETE_COMPONENT: 1106 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1107 case RAIDFRAME_PARITYMAP_STATUS: 1108 case RAIDFRAME_PARITYMAP_GET_DISABLE: 1109 case RAIDFRAME_PARITYMAP_SET_DISABLE: 1110 case RAIDFRAME_PARITYMAP_SET_PARAMS: 1111 if ((rs->sc_flags & RAIDF_INITED) == 0) 1112 return (ENXIO); 1113 } 1114 1115 switch (cmd) { 1116 #ifdef COMPAT_50 1117 case RAIDFRAME_GET_INFO50: 1118 return rf_get_info50(raidPtr, data); 1119 1120 case RAIDFRAME_CONFIGURE50: 1121 if ((retcode = rf_config50(raidPtr, unit, data, &k_cfg)) != 0) 1122 return retcode; 1123 goto config; 1124 #endif 1125 /* configure the system */ 1126 case RAIDFRAME_CONFIGURE: 1127 1128 if (raidPtr->valid) { 1129 /* There is a valid RAID set running on this unit! */ 1130 printf("raid%d: Device already configured!\n",unit); 1131 return(EINVAL); 1132 } 1133 1134 /* copy-in the configuration information */ 1135 /* data points to a pointer to the configuration structure */ 1136 1137 u_cfg = *((RF_Config_t **) data); 1138 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *)); 1139 if (k_cfg == NULL) { 1140 return (ENOMEM); 1141 } 1142 retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t)); 1143 if (retcode) { 1144 RF_Free(k_cfg, sizeof(RF_Config_t)); 1145 db1_printf(("rf_ioctl: retcode=%d copyin.1\n", 1146 retcode)); 1147 return (retcode); 1148 } 1149 goto config; 1150 config: 1151 /* allocate a buffer for the layout-specific data, and copy it 1152 * in */ 1153 if (k_cfg->layoutSpecificSize) { 1154 if (k_cfg->layoutSpecificSize > 10000) { 1155 /* sanity check */ 1156 RF_Free(k_cfg, sizeof(RF_Config_t)); 1157 return (EINVAL); 1158 } 1159 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize, 1160 (u_char *)); 1161 if (specific_buf == NULL) { 1162 RF_Free(k_cfg, sizeof(RF_Config_t)); 1163 return (ENOMEM); 1164 } 1165 retcode = copyin(k_cfg->layoutSpecific, specific_buf, 1166 k_cfg->layoutSpecificSize); 1167 if (retcode) { 1168 RF_Free(k_cfg, sizeof(RF_Config_t)); 1169 RF_Free(specific_buf, 1170 k_cfg->layoutSpecificSize); 1171 db1_printf(("rf_ioctl: retcode=%d copyin.2\n", 1172 retcode)); 1173 return (retcode); 1174 } 1175 } else 1176 specific_buf = NULL; 1177 k_cfg->layoutSpecific = specific_buf; 1178 1179 /* should do some kind of sanity check on the configuration. 1180 * Store the sum of all the bytes in the last byte? */ 1181 1182 /* configure the system */ 1183 1184 /* 1185 * Clear the entire RAID descriptor, just to make sure 1186 * there is no stale data left in the case of a 1187 * reconfiguration 1188 */ 1189 memset((char *) raidPtr, 0, sizeof(RF_Raid_t)); 1190 raidPtr->raidid = unit; 1191 1192 retcode = rf_Configure(raidPtr, k_cfg, NULL); 1193 1194 if (retcode == 0) { 1195 1196 /* allow this many simultaneous IO's to 1197 this RAID device */ 1198 raidPtr->openings = RAIDOUTSTANDING; 1199 1200 raidinit(raidPtr); 1201 rf_markalldirty(raidPtr); 1202 } 1203 /* free the buffers. No return code here. */ 1204 if (k_cfg->layoutSpecificSize) { 1205 RF_Free(specific_buf, k_cfg->layoutSpecificSize); 1206 } 1207 RF_Free(k_cfg, sizeof(RF_Config_t)); 1208 1209 return (retcode); 1210 1211 /* shutdown the system */ 1212 case RAIDFRAME_SHUTDOWN: 1213 1214 part = DISKPART(dev); 1215 pmask = (1 << part); 1216 1217 if ((error = raidlock(rs)) != 0) 1218 return (error); 1219 1220 if ((rs->sc_dkdev.dk_openmask & ~pmask) || 1221 ((rs->sc_dkdev.dk_bopenmask & pmask) && 1222 (rs->sc_dkdev.dk_copenmask & pmask))) 1223 retcode = EBUSY; 1224 else { 1225 rs->sc_flags |= RAIDF_SHUTDOWN; 1226 rs->sc_dkdev.dk_copenmask &= ~pmask; 1227 rs->sc_dkdev.dk_bopenmask &= ~pmask; 1228 rs->sc_dkdev.dk_openmask &= ~pmask; 1229 retcode = 0; 1230 } 1231 1232 raidunlock(rs); 1233 1234 if (retcode != 0) 1235 return retcode; 1236 1237 /* free the pseudo device attach bits */ 1238 1239 cf = device_cfdata(rs->sc_dev); 1240 if ((retcode = config_detach(rs->sc_dev, DETACH_QUIET)) == 0) 1241 free(cf, M_RAIDFRAME); 1242 1243 return (retcode); 1244 case RAIDFRAME_GET_COMPONENT_LABEL: 1245 clabel_ptr = (RF_ComponentLabel_t **) data; 1246 /* need to read the component label for the disk indicated 1247 by row,column in clabel */ 1248 1249 /* 1250 * Perhaps there should be an option to skip the in-core 1251 * copy and hit the disk, as with disklabel(8). 1252 */ 1253 RF_Malloc(clabel, sizeof(*clabel), (RF_ComponentLabel_t *)); 1254 1255 retcode = copyin( *clabel_ptr, clabel, 1256 sizeof(RF_ComponentLabel_t)); 1257 1258 if (retcode) { 1259 return(retcode); 1260 } 1261 1262 clabel->row = 0; /* Don't allow looking at anything else.*/ 1263 1264 column = clabel->column; 1265 1266 if ((column < 0) || (column >= raidPtr->numCol + 1267 raidPtr->numSpare)) { 1268 return(EINVAL); 1269 } 1270 1271 RF_Free(clabel, sizeof(*clabel)); 1272 1273 clabel = raidget_component_label(raidPtr, column); 1274 1275 if (retcode == 0) { 1276 retcode = copyout(clabel, *clabel_ptr, 1277 sizeof(RF_ComponentLabel_t)); 1278 } 1279 return (retcode); 1280 1281 #if 0 1282 case RAIDFRAME_SET_COMPONENT_LABEL: 1283 clabel = (RF_ComponentLabel_t *) data; 1284 1285 /* XXX check the label for valid stuff... */ 1286 /* Note that some things *should not* get modified -- 1287 the user should be re-initing the labels instead of 1288 trying to patch things. 1289 */ 1290 1291 raidid = raidPtr->raidid; 1292 #ifdef DEBUG 1293 printf("raid%d: Got component label:\n", raidid); 1294 printf("raid%d: Version: %d\n", raidid, clabel->version); 1295 printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number); 1296 printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter); 1297 printf("raid%d: Column: %d\n", raidid, clabel->column); 1298 printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns); 1299 printf("raid%d: Clean: %d\n", raidid, clabel->clean); 1300 printf("raid%d: Status: %d\n", raidid, clabel->status); 1301 #endif 1302 clabel->row = 0; 1303 column = clabel->column; 1304 1305 if ((column < 0) || (column >= raidPtr->numCol)) { 1306 return(EINVAL); 1307 } 1308 1309 /* XXX this isn't allowed to do anything for now :-) */ 1310 1311 /* XXX and before it is, we need to fill in the rest 1312 of the fields!?!?!?! */ 1313 memcpy(raidget_component_label(raidPtr, column), 1314 clabel, sizeof(*clabel)); 1315 raidflush_component_label(raidPtr, column); 1316 return (0); 1317 #endif 1318 1319 case RAIDFRAME_INIT_LABELS: 1320 clabel = (RF_ComponentLabel_t *) data; 1321 /* 1322 we only want the serial number from 1323 the above. We get all the rest of the information 1324 from the config that was used to create this RAID 1325 set. 1326 */ 1327 1328 raidPtr->serial_number = clabel->serial_number; 1329 1330 for(column=0;column<raidPtr->numCol;column++) { 1331 diskPtr = &raidPtr->Disks[column]; 1332 if (!RF_DEAD_DISK(diskPtr->status)) { 1333 ci_label = raidget_component_label(raidPtr, 1334 column); 1335 /* Zeroing this is important. */ 1336 memset(ci_label, 0, sizeof(*ci_label)); 1337 raid_init_component_label(raidPtr, ci_label); 1338 ci_label->serial_number = 1339 raidPtr->serial_number; 1340 ci_label->row = 0; /* we dont' pretend to support more */ 1341 ci_label->partitionSize = 1342 diskPtr->partitionSize; 1343 ci_label->column = column; 1344 raidflush_component_label(raidPtr, column); 1345 } 1346 /* XXXjld what about the spares? */ 1347 } 1348 1349 return (retcode); 1350 case RAIDFRAME_SET_AUTOCONFIG: 1351 d = rf_set_autoconfig(raidPtr, *(int *) data); 1352 printf("raid%d: New autoconfig value is: %d\n", 1353 raidPtr->raidid, d); 1354 *(int *) data = d; 1355 return (retcode); 1356 1357 case RAIDFRAME_SET_ROOT: 1358 d = rf_set_rootpartition(raidPtr, *(int *) data); 1359 printf("raid%d: New rootpartition value is: %d\n", 1360 raidPtr->raidid, d); 1361 *(int *) data = d; 1362 return (retcode); 1363 1364 /* initialize all parity */ 1365 case RAIDFRAME_REWRITEPARITY: 1366 1367 if (raidPtr->Layout.map->faultsTolerated == 0) { 1368 /* Parity for RAID 0 is trivially correct */ 1369 raidPtr->parity_good = RF_RAID_CLEAN; 1370 return(0); 1371 } 1372 1373 if (raidPtr->parity_rewrite_in_progress == 1) { 1374 /* Re-write is already in progress! */ 1375 return(EINVAL); 1376 } 1377 1378 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread, 1379 rf_RewriteParityThread, 1380 raidPtr,"raid_parity"); 1381 return (retcode); 1382 1383 1384 case RAIDFRAME_ADD_HOT_SPARE: 1385 sparePtr = (RF_SingleComponent_t *) data; 1386 memcpy( &component, sparePtr, sizeof(RF_SingleComponent_t)); 1387 retcode = rf_add_hot_spare(raidPtr, &component); 1388 return(retcode); 1389 1390 case RAIDFRAME_REMOVE_HOT_SPARE: 1391 return(retcode); 1392 1393 case RAIDFRAME_DELETE_COMPONENT: 1394 componentPtr = (RF_SingleComponent_t *)data; 1395 memcpy( &component, componentPtr, 1396 sizeof(RF_SingleComponent_t)); 1397 retcode = rf_delete_component(raidPtr, &component); 1398 return(retcode); 1399 1400 case RAIDFRAME_INCORPORATE_HOT_SPARE: 1401 componentPtr = (RF_SingleComponent_t *)data; 1402 memcpy( &component, componentPtr, 1403 sizeof(RF_SingleComponent_t)); 1404 retcode = rf_incorporate_hot_spare(raidPtr, &component); 1405 return(retcode); 1406 1407 case RAIDFRAME_REBUILD_IN_PLACE: 1408 1409 if (raidPtr->Layout.map->faultsTolerated == 0) { 1410 /* Can't do this on a RAID 0!! */ 1411 return(EINVAL); 1412 } 1413 1414 if (raidPtr->recon_in_progress == 1) { 1415 /* a reconstruct is already in progress! */ 1416 return(EINVAL); 1417 } 1418 1419 componentPtr = (RF_SingleComponent_t *) data; 1420 memcpy( &component, componentPtr, 1421 sizeof(RF_SingleComponent_t)); 1422 component.row = 0; /* we don't support any more */ 1423 column = component.column; 1424 1425 if ((column < 0) || (column >= raidPtr->numCol)) { 1426 return(EINVAL); 1427 } 1428 1429 RF_LOCK_MUTEX(raidPtr->mutex); 1430 if ((raidPtr->Disks[column].status == rf_ds_optimal) && 1431 (raidPtr->numFailures > 0)) { 1432 /* XXX 0 above shouldn't be constant!!! */ 1433 /* some component other than this has failed. 1434 Let's not make things worse than they already 1435 are... */ 1436 printf("raid%d: Unable to reconstruct to disk at:\n", 1437 raidPtr->raidid); 1438 printf("raid%d: Col: %d Too many failures.\n", 1439 raidPtr->raidid, column); 1440 RF_UNLOCK_MUTEX(raidPtr->mutex); 1441 return (EINVAL); 1442 } 1443 if (raidPtr->Disks[column].status == 1444 rf_ds_reconstructing) { 1445 printf("raid%d: Unable to reconstruct to disk at:\n", 1446 raidPtr->raidid); 1447 printf("raid%d: Col: %d Reconstruction already occuring!\n", raidPtr->raidid, column); 1448 1449 RF_UNLOCK_MUTEX(raidPtr->mutex); 1450 return (EINVAL); 1451 } 1452 if (raidPtr->Disks[column].status == rf_ds_spared) { 1453 RF_UNLOCK_MUTEX(raidPtr->mutex); 1454 return (EINVAL); 1455 } 1456 RF_UNLOCK_MUTEX(raidPtr->mutex); 1457 1458 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1459 if (rrcopy == NULL) 1460 return(ENOMEM); 1461 1462 rrcopy->raidPtr = (void *) raidPtr; 1463 rrcopy->col = column; 1464 1465 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1466 rf_ReconstructInPlaceThread, 1467 rrcopy,"raid_reconip"); 1468 return(retcode); 1469 1470 case RAIDFRAME_GET_INFO: 1471 if (!raidPtr->valid) 1472 return (ENODEV); 1473 ucfgp = (RF_DeviceConfig_t **) data; 1474 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t), 1475 (RF_DeviceConfig_t *)); 1476 if (d_cfg == NULL) 1477 return (ENOMEM); 1478 d_cfg->rows = 1; /* there is only 1 row now */ 1479 d_cfg->cols = raidPtr->numCol; 1480 d_cfg->ndevs = raidPtr->numCol; 1481 if (d_cfg->ndevs >= RF_MAX_DISKS) { 1482 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1483 return (ENOMEM); 1484 } 1485 d_cfg->nspares = raidPtr->numSpare; 1486 if (d_cfg->nspares >= RF_MAX_DISKS) { 1487 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1488 return (ENOMEM); 1489 } 1490 d_cfg->maxqdepth = raidPtr->maxQueueDepth; 1491 d = 0; 1492 for (j = 0; j < d_cfg->cols; j++) { 1493 d_cfg->devs[d] = raidPtr->Disks[j]; 1494 d++; 1495 } 1496 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) { 1497 d_cfg->spares[i] = raidPtr->Disks[j]; 1498 } 1499 retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t)); 1500 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t)); 1501 1502 return (retcode); 1503 1504 case RAIDFRAME_CHECK_PARITY: 1505 *(int *) data = raidPtr->parity_good; 1506 return (0); 1507 1508 case RAIDFRAME_PARITYMAP_STATUS: 1509 if (rf_paritymap_ineligible(raidPtr)) 1510 return EINVAL; 1511 rf_paritymap_status(raidPtr->parity_map, 1512 (struct rf_pmstat *)data); 1513 return 0; 1514 1515 case RAIDFRAME_PARITYMAP_SET_PARAMS: 1516 if (rf_paritymap_ineligible(raidPtr)) 1517 return EINVAL; 1518 if (raidPtr->parity_map == NULL) 1519 return ENOENT; /* ??? */ 1520 if (0 != rf_paritymap_set_params(raidPtr->parity_map, 1521 (struct rf_pmparams *)data, 1)) 1522 return EINVAL; 1523 return 0; 1524 1525 case RAIDFRAME_PARITYMAP_GET_DISABLE: 1526 if (rf_paritymap_ineligible(raidPtr)) 1527 return EINVAL; 1528 *(int *) data = rf_paritymap_get_disable(raidPtr); 1529 return 0; 1530 1531 case RAIDFRAME_PARITYMAP_SET_DISABLE: 1532 if (rf_paritymap_ineligible(raidPtr)) 1533 return EINVAL; 1534 rf_paritymap_set_disable(raidPtr, *(int *)data); 1535 /* XXX should errors be passed up? */ 1536 return 0; 1537 1538 case RAIDFRAME_RESET_ACCTOTALS: 1539 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1540 return (0); 1541 1542 case RAIDFRAME_GET_ACCTOTALS: 1543 totals = (RF_AccTotals_t *) data; 1544 *totals = raidPtr->acc_totals; 1545 return (0); 1546 1547 case RAIDFRAME_KEEP_ACCTOTALS: 1548 raidPtr->keep_acc_totals = *(int *)data; 1549 return (0); 1550 1551 case RAIDFRAME_GET_SIZE: 1552 *(int *) data = raidPtr->totalSectors; 1553 return (0); 1554 1555 /* fail a disk & optionally start reconstruction */ 1556 case RAIDFRAME_FAIL_DISK: 1557 1558 if (raidPtr->Layout.map->faultsTolerated == 0) { 1559 /* Can't do this on a RAID 0!! */ 1560 return(EINVAL); 1561 } 1562 1563 rr = (struct rf_recon_req *) data; 1564 rr->row = 0; 1565 if (rr->col < 0 || rr->col >= raidPtr->numCol) 1566 return (EINVAL); 1567 1568 1569 RF_LOCK_MUTEX(raidPtr->mutex); 1570 if (raidPtr->status == rf_rs_reconstructing) { 1571 /* you can't fail a disk while we're reconstructing! */ 1572 /* XXX wrong for RAID6 */ 1573 RF_UNLOCK_MUTEX(raidPtr->mutex); 1574 return (EINVAL); 1575 } 1576 if ((raidPtr->Disks[rr->col].status == 1577 rf_ds_optimal) && (raidPtr->numFailures > 0)) { 1578 /* some other component has failed. Let's not make 1579 things worse. XXX wrong for RAID6 */ 1580 RF_UNLOCK_MUTEX(raidPtr->mutex); 1581 return (EINVAL); 1582 } 1583 if (raidPtr->Disks[rr->col].status == rf_ds_spared) { 1584 /* Can't fail a spared disk! */ 1585 RF_UNLOCK_MUTEX(raidPtr->mutex); 1586 return (EINVAL); 1587 } 1588 RF_UNLOCK_MUTEX(raidPtr->mutex); 1589 1590 /* make a copy of the recon request so that we don't rely on 1591 * the user's buffer */ 1592 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1593 if (rrcopy == NULL) 1594 return(ENOMEM); 1595 memcpy(rrcopy, rr, sizeof(*rr)); 1596 rrcopy->raidPtr = (void *) raidPtr; 1597 1598 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1599 rf_ReconThread, 1600 rrcopy,"raid_recon"); 1601 return (0); 1602 1603 /* invoke a copyback operation after recon on whatever disk 1604 * needs it, if any */ 1605 case RAIDFRAME_COPYBACK: 1606 1607 if (raidPtr->Layout.map->faultsTolerated == 0) { 1608 /* This makes no sense on a RAID 0!! */ 1609 return(EINVAL); 1610 } 1611 1612 if (raidPtr->copyback_in_progress == 1) { 1613 /* Copyback is already in progress! */ 1614 return(EINVAL); 1615 } 1616 1617 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1618 rf_CopybackThread, 1619 raidPtr,"raid_copyback"); 1620 return (retcode); 1621 1622 /* return the percentage completion of reconstruction */ 1623 case RAIDFRAME_CHECK_RECON_STATUS: 1624 if (raidPtr->Layout.map->faultsTolerated == 0) { 1625 /* This makes no sense on a RAID 0, so tell the 1626 user it's done. */ 1627 *(int *) data = 100; 1628 return(0); 1629 } 1630 if (raidPtr->status != rf_rs_reconstructing) 1631 *(int *) data = 100; 1632 else { 1633 if (raidPtr->reconControl->numRUsTotal > 0) { 1634 *(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal); 1635 } else { 1636 *(int *) data = 0; 1637 } 1638 } 1639 return (0); 1640 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1641 progressInfoPtr = (RF_ProgressInfo_t **) data; 1642 if (raidPtr->status != rf_rs_reconstructing) { 1643 progressInfo.remaining = 0; 1644 progressInfo.completed = 100; 1645 progressInfo.total = 100; 1646 } else { 1647 progressInfo.total = 1648 raidPtr->reconControl->numRUsTotal; 1649 progressInfo.completed = 1650 raidPtr->reconControl->numRUsComplete; 1651 progressInfo.remaining = progressInfo.total - 1652 progressInfo.completed; 1653 } 1654 retcode = copyout(&progressInfo, *progressInfoPtr, 1655 sizeof(RF_ProgressInfo_t)); 1656 return (retcode); 1657 1658 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1659 if (raidPtr->Layout.map->faultsTolerated == 0) { 1660 /* This makes no sense on a RAID 0, so tell the 1661 user it's done. */ 1662 *(int *) data = 100; 1663 return(0); 1664 } 1665 if (raidPtr->parity_rewrite_in_progress == 1) { 1666 *(int *) data = 100 * 1667 raidPtr->parity_rewrite_stripes_done / 1668 raidPtr->Layout.numStripe; 1669 } else { 1670 *(int *) data = 100; 1671 } 1672 return (0); 1673 1674 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1675 progressInfoPtr = (RF_ProgressInfo_t **) data; 1676 if (raidPtr->parity_rewrite_in_progress == 1) { 1677 progressInfo.total = raidPtr->Layout.numStripe; 1678 progressInfo.completed = 1679 raidPtr->parity_rewrite_stripes_done; 1680 progressInfo.remaining = progressInfo.total - 1681 progressInfo.completed; 1682 } else { 1683 progressInfo.remaining = 0; 1684 progressInfo.completed = 100; 1685 progressInfo.total = 100; 1686 } 1687 retcode = copyout(&progressInfo, *progressInfoPtr, 1688 sizeof(RF_ProgressInfo_t)); 1689 return (retcode); 1690 1691 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1692 if (raidPtr->Layout.map->faultsTolerated == 0) { 1693 /* This makes no sense on a RAID 0 */ 1694 *(int *) data = 100; 1695 return(0); 1696 } 1697 if (raidPtr->copyback_in_progress == 1) { 1698 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1699 raidPtr->Layout.numStripe; 1700 } else { 1701 *(int *) data = 100; 1702 } 1703 return (0); 1704 1705 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1706 progressInfoPtr = (RF_ProgressInfo_t **) data; 1707 if (raidPtr->copyback_in_progress == 1) { 1708 progressInfo.total = raidPtr->Layout.numStripe; 1709 progressInfo.completed = 1710 raidPtr->copyback_stripes_done; 1711 progressInfo.remaining = progressInfo.total - 1712 progressInfo.completed; 1713 } else { 1714 progressInfo.remaining = 0; 1715 progressInfo.completed = 100; 1716 progressInfo.total = 100; 1717 } 1718 retcode = copyout(&progressInfo, *progressInfoPtr, 1719 sizeof(RF_ProgressInfo_t)); 1720 return (retcode); 1721 1722 /* the sparetable daemon calls this to wait for the kernel to 1723 * need a spare table. this ioctl does not return until a 1724 * spare table is needed. XXX -- calling mpsleep here in the 1725 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1726 * -- I should either compute the spare table in the kernel, 1727 * or have a different -- XXX XXX -- interface (a different 1728 * character device) for delivering the table -- XXX */ 1729 #if 0 1730 case RAIDFRAME_SPARET_WAIT: 1731 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1732 while (!rf_sparet_wait_queue) 1733 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE); 1734 waitreq = rf_sparet_wait_queue; 1735 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1736 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1737 1738 /* structure assignment */ 1739 *((RF_SparetWait_t *) data) = *waitreq; 1740 1741 RF_Free(waitreq, sizeof(*waitreq)); 1742 return (0); 1743 1744 /* wakes up a process waiting on SPARET_WAIT and puts an error 1745 * code in it that will cause the dameon to exit */ 1746 case RAIDFRAME_ABORT_SPARET_WAIT: 1747 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1748 waitreq->fcol = -1; 1749 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1750 waitreq->next = rf_sparet_wait_queue; 1751 rf_sparet_wait_queue = waitreq; 1752 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1753 wakeup(&rf_sparet_wait_queue); 1754 return (0); 1755 1756 /* used by the spare table daemon to deliver a spare table 1757 * into the kernel */ 1758 case RAIDFRAME_SEND_SPARET: 1759 1760 /* install the spare table */ 1761 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1762 1763 /* respond to the requestor. the return status of the spare 1764 * table installation is passed in the "fcol" field */ 1765 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1766 waitreq->fcol = retcode; 1767 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1768 waitreq->next = rf_sparet_resp_queue; 1769 rf_sparet_resp_queue = waitreq; 1770 wakeup(&rf_sparet_resp_queue); 1771 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1772 1773 return (retcode); 1774 #endif 1775 1776 default: 1777 break; /* fall through to the os-specific code below */ 1778 1779 } 1780 1781 if (!raidPtr->valid) 1782 return (EINVAL); 1783 1784 /* 1785 * Add support for "regular" device ioctls here. 1786 */ 1787 1788 error = disk_ioctl(&rs->sc_dkdev, cmd, data, flag, l); 1789 if (error != EPASSTHROUGH) 1790 return (error); 1791 1792 switch (cmd) { 1793 case DIOCGDINFO: 1794 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1795 break; 1796 #ifdef __HAVE_OLD_DISKLABEL 1797 case ODIOCGDINFO: 1798 newlabel = *(rs->sc_dkdev.dk_label); 1799 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1800 return ENOTTY; 1801 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1802 break; 1803 #endif 1804 1805 case DIOCGPART: 1806 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1807 ((struct partinfo *) data)->part = 1808 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1809 break; 1810 1811 case DIOCWDINFO: 1812 case DIOCSDINFO: 1813 #ifdef __HAVE_OLD_DISKLABEL 1814 case ODIOCWDINFO: 1815 case ODIOCSDINFO: 1816 #endif 1817 { 1818 struct disklabel *lp; 1819 #ifdef __HAVE_OLD_DISKLABEL 1820 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1821 memset(&newlabel, 0, sizeof newlabel); 1822 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1823 lp = &newlabel; 1824 } else 1825 #endif 1826 lp = (struct disklabel *)data; 1827 1828 if ((error = raidlock(rs)) != 0) 1829 return (error); 1830 1831 rs->sc_flags |= RAIDF_LABELLING; 1832 1833 error = setdisklabel(rs->sc_dkdev.dk_label, 1834 lp, 0, rs->sc_dkdev.dk_cpulabel); 1835 if (error == 0) { 1836 if (cmd == DIOCWDINFO 1837 #ifdef __HAVE_OLD_DISKLABEL 1838 || cmd == ODIOCWDINFO 1839 #endif 1840 ) 1841 error = writedisklabel(RAIDLABELDEV(dev), 1842 raidstrategy, rs->sc_dkdev.dk_label, 1843 rs->sc_dkdev.dk_cpulabel); 1844 } 1845 rs->sc_flags &= ~RAIDF_LABELLING; 1846 1847 raidunlock(rs); 1848 1849 if (error) 1850 return (error); 1851 break; 1852 } 1853 1854 case DIOCWLABEL: 1855 if (*(int *) data != 0) 1856 rs->sc_flags |= RAIDF_WLABEL; 1857 else 1858 rs->sc_flags &= ~RAIDF_WLABEL; 1859 break; 1860 1861 case DIOCGDEFLABEL: 1862 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1863 break; 1864 1865 #ifdef __HAVE_OLD_DISKLABEL 1866 case ODIOCGDEFLABEL: 1867 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1868 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1869 return ENOTTY; 1870 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1871 break; 1872 #endif 1873 1874 case DIOCAWEDGE: 1875 case DIOCDWEDGE: 1876 dkw = (void *)data; 1877 1878 /* If the ioctl happens here, the parent is us. */ 1879 (void)strcpy(dkw->dkw_parent, rs->sc_xname); 1880 return cmd == DIOCAWEDGE ? dkwedge_add(dkw) : dkwedge_del(dkw); 1881 1882 case DIOCLWEDGES: 1883 return dkwedge_list(&rs->sc_dkdev, 1884 (struct dkwedge_list *)data, l); 1885 case DIOCCACHESYNC: 1886 return rf_sync_component_caches(raidPtr); 1887 default: 1888 retcode = ENOTTY; 1889 } 1890 return (retcode); 1891 1892 } 1893 1894 1895 /* raidinit -- complete the rest of the initialization for the 1896 RAIDframe device. */ 1897 1898 1899 static void 1900 raidinit(RF_Raid_t *raidPtr) 1901 { 1902 cfdata_t cf; 1903 struct raid_softc *rs; 1904 int unit; 1905 1906 unit = raidPtr->raidid; 1907 1908 rs = &raid_softc[unit]; 1909 1910 /* XXX should check return code first... */ 1911 rs->sc_flags |= RAIDF_INITED; 1912 1913 /* XXX doesn't check bounds. */ 1914 snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit); 1915 1916 /* attach the pseudo device */ 1917 cf = malloc(sizeof(*cf), M_RAIDFRAME, M_WAITOK); 1918 cf->cf_name = raid_cd.cd_name; 1919 cf->cf_atname = raid_cd.cd_name; 1920 cf->cf_unit = unit; 1921 cf->cf_fstate = FSTATE_STAR; 1922 1923 rs->sc_dev = config_attach_pseudo(cf); 1924 1925 if (rs->sc_dev == NULL) { 1926 printf("raid%d: config_attach_pseudo failed\n", 1927 raidPtr->raidid); 1928 rs->sc_flags &= ~RAIDF_INITED; 1929 free(cf, M_RAIDFRAME); 1930 return; 1931 } 1932 1933 /* disk_attach actually creates space for the CPU disklabel, among 1934 * other things, so it's critical to call this *BEFORE* we try putzing 1935 * with disklabels. */ 1936 1937 disk_init(&rs->sc_dkdev, rs->sc_xname, &rf_dkdriver); 1938 disk_attach(&rs->sc_dkdev); 1939 1940 /* XXX There may be a weird interaction here between this, and 1941 * protectedSectors, as used in RAIDframe. */ 1942 1943 rs->sc_size = raidPtr->totalSectors; 1944 1945 dkwedge_discover(&rs->sc_dkdev); 1946 1947 rf_set_properties(rs, raidPtr); 1948 1949 } 1950 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 1951 /* wake up the daemon & tell it to get us a spare table 1952 * XXX 1953 * the entries in the queues should be tagged with the raidPtr 1954 * so that in the extremely rare case that two recons happen at once, 1955 * we know for which device were requesting a spare table 1956 * XXX 1957 * 1958 * XXX This code is not currently used. GO 1959 */ 1960 int 1961 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req) 1962 { 1963 int retcode; 1964 1965 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1966 req->next = rf_sparet_wait_queue; 1967 rf_sparet_wait_queue = req; 1968 wakeup(&rf_sparet_wait_queue); 1969 1970 /* mpsleep unlocks the mutex */ 1971 while (!rf_sparet_resp_queue) { 1972 tsleep(&rf_sparet_resp_queue, PRIBIO, 1973 "raidframe getsparetable", 0); 1974 } 1975 req = rf_sparet_resp_queue; 1976 rf_sparet_resp_queue = req->next; 1977 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1978 1979 retcode = req->fcol; 1980 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1981 * alloc'd */ 1982 return (retcode); 1983 } 1984 #endif 1985 1986 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1987 * bp & passes it down. 1988 * any calls originating in the kernel must use non-blocking I/O 1989 * do some extra sanity checking to return "appropriate" error values for 1990 * certain conditions (to make some standard utilities work) 1991 * 1992 * Formerly known as: rf_DoAccessKernel 1993 */ 1994 void 1995 raidstart(RF_Raid_t *raidPtr) 1996 { 1997 RF_SectorCount_t num_blocks, pb, sum; 1998 RF_RaidAddr_t raid_addr; 1999 struct partition *pp; 2000 daddr_t blocknum; 2001 int unit; 2002 struct raid_softc *rs; 2003 int do_async; 2004 struct buf *bp; 2005 int rc; 2006 2007 unit = raidPtr->raidid; 2008 rs = &raid_softc[unit]; 2009 2010 /* quick check to see if anything has died recently */ 2011 RF_LOCK_MUTEX(raidPtr->mutex); 2012 if (raidPtr->numNewFailures > 0) { 2013 RF_UNLOCK_MUTEX(raidPtr->mutex); 2014 rf_update_component_labels(raidPtr, 2015 RF_NORMAL_COMPONENT_UPDATE); 2016 RF_LOCK_MUTEX(raidPtr->mutex); 2017 raidPtr->numNewFailures--; 2018 } 2019 2020 /* Check to see if we're at the limit... */ 2021 while (raidPtr->openings > 0) { 2022 RF_UNLOCK_MUTEX(raidPtr->mutex); 2023 2024 /* get the next item, if any, from the queue */ 2025 if ((bp = bufq_get(rs->buf_queue)) == NULL) { 2026 /* nothing more to do */ 2027 return; 2028 } 2029 2030 /* Ok, for the bp we have here, bp->b_blkno is relative to the 2031 * partition.. Need to make it absolute to the underlying 2032 * device.. */ 2033 2034 blocknum = bp->b_blkno; 2035 if (DISKPART(bp->b_dev) != RAW_PART) { 2036 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 2037 blocknum += pp->p_offset; 2038 } 2039 2040 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 2041 (int) blocknum)); 2042 2043 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 2044 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 2045 2046 /* *THIS* is where we adjust what block we're going to... 2047 * but DO NOT TOUCH bp->b_blkno!!! */ 2048 raid_addr = blocknum; 2049 2050 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 2051 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 2052 sum = raid_addr + num_blocks + pb; 2053 if (1 || rf_debugKernelAccess) { 2054 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 2055 (int) raid_addr, (int) sum, (int) num_blocks, 2056 (int) pb, (int) bp->b_resid)); 2057 } 2058 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 2059 || (sum < num_blocks) || (sum < pb)) { 2060 bp->b_error = ENOSPC; 2061 bp->b_resid = bp->b_bcount; 2062 biodone(bp); 2063 RF_LOCK_MUTEX(raidPtr->mutex); 2064 continue; 2065 } 2066 /* 2067 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 2068 */ 2069 2070 if (bp->b_bcount & raidPtr->sectorMask) { 2071 bp->b_error = EINVAL; 2072 bp->b_resid = bp->b_bcount; 2073 biodone(bp); 2074 RF_LOCK_MUTEX(raidPtr->mutex); 2075 continue; 2076 2077 } 2078 db1_printf(("Calling DoAccess..\n")); 2079 2080 2081 RF_LOCK_MUTEX(raidPtr->mutex); 2082 raidPtr->openings--; 2083 RF_UNLOCK_MUTEX(raidPtr->mutex); 2084 2085 /* 2086 * Everything is async. 2087 */ 2088 do_async = 1; 2089 2090 disk_busy(&rs->sc_dkdev); 2091 2092 /* XXX we're still at splbio() here... do we *really* 2093 need to be? */ 2094 2095 /* don't ever condition on bp->b_flags & B_WRITE. 2096 * always condition on B_READ instead */ 2097 2098 rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 2099 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 2100 do_async, raid_addr, num_blocks, 2101 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 2102 2103 if (rc) { 2104 bp->b_error = rc; 2105 bp->b_resid = bp->b_bcount; 2106 biodone(bp); 2107 /* continue loop */ 2108 } 2109 2110 RF_LOCK_MUTEX(raidPtr->mutex); 2111 } 2112 RF_UNLOCK_MUTEX(raidPtr->mutex); 2113 } 2114 2115 2116 2117 2118 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 2119 2120 int 2121 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req) 2122 { 2123 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 2124 struct buf *bp; 2125 2126 req->queue = queue; 2127 bp = req->bp; 2128 2129 switch (req->type) { 2130 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 2131 /* XXX need to do something extra here.. */ 2132 /* I'm leaving this in, as I've never actually seen it used, 2133 * and I'd like folks to report it... GO */ 2134 printf(("WAKEUP CALLED\n")); 2135 queue->numOutstanding++; 2136 2137 bp->b_flags = 0; 2138 bp->b_private = req; 2139 2140 KernelWakeupFunc(bp); 2141 break; 2142 2143 case RF_IO_TYPE_READ: 2144 case RF_IO_TYPE_WRITE: 2145 #if RF_ACC_TRACE > 0 2146 if (req->tracerec) { 2147 RF_ETIMER_START(req->tracerec->timer); 2148 } 2149 #endif 2150 InitBP(bp, queue->rf_cinfo->ci_vp, 2151 op, queue->rf_cinfo->ci_dev, 2152 req->sectorOffset, req->numSector, 2153 req->buf, KernelWakeupFunc, (void *) req, 2154 queue->raidPtr->logBytesPerSector, req->b_proc); 2155 2156 if (rf_debugKernelAccess) { 2157 db1_printf(("dispatch: bp->b_blkno = %ld\n", 2158 (long) bp->b_blkno)); 2159 } 2160 queue->numOutstanding++; 2161 queue->last_deq_sector = req->sectorOffset; 2162 /* acc wouldn't have been let in if there were any pending 2163 * reqs at any other priority */ 2164 queue->curPriority = req->priority; 2165 2166 db1_printf(("Going for %c to unit %d col %d\n", 2167 req->type, queue->raidPtr->raidid, 2168 queue->col)); 2169 db1_printf(("sector %d count %d (%d bytes) %d\n", 2170 (int) req->sectorOffset, (int) req->numSector, 2171 (int) (req->numSector << 2172 queue->raidPtr->logBytesPerSector), 2173 (int) queue->raidPtr->logBytesPerSector)); 2174 2175 /* 2176 * XXX: drop lock here since this can block at 2177 * least with backing SCSI devices. Retake it 2178 * to minimize fuss with calling interfaces. 2179 */ 2180 2181 RF_UNLOCK_QUEUE_MUTEX(queue, "unusedparam"); 2182 bdev_strategy(bp); 2183 RF_LOCK_QUEUE_MUTEX(queue, "unusedparam"); 2184 break; 2185 2186 default: 2187 panic("bad req->type in rf_DispatchKernelIO"); 2188 } 2189 db1_printf(("Exiting from DispatchKernelIO\n")); 2190 2191 return (0); 2192 } 2193 /* this is the callback function associated with a I/O invoked from 2194 kernel code. 2195 */ 2196 static void 2197 KernelWakeupFunc(struct buf *bp) 2198 { 2199 RF_DiskQueueData_t *req = NULL; 2200 RF_DiskQueue_t *queue; 2201 int s; 2202 2203 s = splbio(); 2204 db1_printf(("recovering the request queue:\n")); 2205 req = bp->b_private; 2206 2207 queue = (RF_DiskQueue_t *) req->queue; 2208 2209 #if RF_ACC_TRACE > 0 2210 if (req->tracerec) { 2211 RF_ETIMER_STOP(req->tracerec->timer); 2212 RF_ETIMER_EVAL(req->tracerec->timer); 2213 RF_LOCK_MUTEX(rf_tracing_mutex); 2214 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2215 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2216 req->tracerec->num_phys_ios++; 2217 RF_UNLOCK_MUTEX(rf_tracing_mutex); 2218 } 2219 #endif 2220 2221 /* XXX Ok, let's get aggressive... If b_error is set, let's go 2222 * ballistic, and mark the component as hosed... */ 2223 2224 if (bp->b_error != 0) { 2225 /* Mark the disk as dead */ 2226 /* but only mark it once... */ 2227 /* and only if it wouldn't leave this RAID set 2228 completely broken */ 2229 if (((queue->raidPtr->Disks[queue->col].status == 2230 rf_ds_optimal) || 2231 (queue->raidPtr->Disks[queue->col].status == 2232 rf_ds_used_spare)) && 2233 (queue->raidPtr->numFailures < 2234 queue->raidPtr->Layout.map->faultsTolerated)) { 2235 printf("raid%d: IO Error. Marking %s as failed.\n", 2236 queue->raidPtr->raidid, 2237 queue->raidPtr->Disks[queue->col].devname); 2238 queue->raidPtr->Disks[queue->col].status = 2239 rf_ds_failed; 2240 queue->raidPtr->status = rf_rs_degraded; 2241 queue->raidPtr->numFailures++; 2242 queue->raidPtr->numNewFailures++; 2243 } else { /* Disk is already dead... */ 2244 /* printf("Disk already marked as dead!\n"); */ 2245 } 2246 2247 } 2248 2249 /* Fill in the error value */ 2250 2251 req->error = bp->b_error; 2252 2253 simple_lock(&queue->raidPtr->iodone_lock); 2254 2255 /* Drop this one on the "finished" queue... */ 2256 TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries); 2257 2258 /* Let the raidio thread know there is work to be done. */ 2259 wakeup(&(queue->raidPtr->iodone)); 2260 2261 simple_unlock(&queue->raidPtr->iodone_lock); 2262 2263 splx(s); 2264 } 2265 2266 2267 2268 /* 2269 * initialize a buf structure for doing an I/O in the kernel. 2270 */ 2271 static void 2272 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev, 2273 RF_SectorNum_t startSect, RF_SectorCount_t numSect, void *bf, 2274 void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector, 2275 struct proc *b_proc) 2276 { 2277 /* bp->b_flags = B_PHYS | rw_flag; */ 2278 bp->b_flags = rw_flag; /* XXX need B_PHYS here too??? */ 2279 bp->b_oflags = 0; 2280 bp->b_cflags = 0; 2281 bp->b_bcount = numSect << logBytesPerSector; 2282 bp->b_bufsize = bp->b_bcount; 2283 bp->b_error = 0; 2284 bp->b_dev = dev; 2285 bp->b_data = bf; 2286 bp->b_blkno = startSect; 2287 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 2288 if (bp->b_bcount == 0) { 2289 panic("bp->b_bcount is zero in InitBP!!"); 2290 } 2291 bp->b_proc = b_proc; 2292 bp->b_iodone = cbFunc; 2293 bp->b_private = cbArg; 2294 } 2295 2296 static void 2297 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs, 2298 struct disklabel *lp) 2299 { 2300 memset(lp, 0, sizeof(*lp)); 2301 2302 /* fabricate a label... */ 2303 lp->d_secperunit = raidPtr->totalSectors; 2304 lp->d_secsize = raidPtr->bytesPerSector; 2305 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 2306 lp->d_ntracks = 4 * raidPtr->numCol; 2307 lp->d_ncylinders = raidPtr->totalSectors / 2308 (lp->d_nsectors * lp->d_ntracks); 2309 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 2310 2311 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 2312 lp->d_type = DTYPE_RAID; 2313 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 2314 lp->d_rpm = 3600; 2315 lp->d_interleave = 1; 2316 lp->d_flags = 0; 2317 2318 lp->d_partitions[RAW_PART].p_offset = 0; 2319 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 2320 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 2321 lp->d_npartitions = RAW_PART + 1; 2322 2323 lp->d_magic = DISKMAGIC; 2324 lp->d_magic2 = DISKMAGIC; 2325 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 2326 2327 } 2328 /* 2329 * Read the disklabel from the raid device. If one is not present, fake one 2330 * up. 2331 */ 2332 static void 2333 raidgetdisklabel(dev_t dev) 2334 { 2335 int unit = raidunit(dev); 2336 struct raid_softc *rs = &raid_softc[unit]; 2337 const char *errstring; 2338 struct disklabel *lp = rs->sc_dkdev.dk_label; 2339 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 2340 RF_Raid_t *raidPtr; 2341 2342 db1_printf(("Getting the disklabel...\n")); 2343 2344 memset(clp, 0, sizeof(*clp)); 2345 2346 raidPtr = raidPtrs[unit]; 2347 2348 raidgetdefaultlabel(raidPtr, rs, lp); 2349 2350 /* 2351 * Call the generic disklabel extraction routine. 2352 */ 2353 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 2354 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2355 if (errstring) 2356 raidmakedisklabel(rs); 2357 else { 2358 int i; 2359 struct partition *pp; 2360 2361 /* 2362 * Sanity check whether the found disklabel is valid. 2363 * 2364 * This is necessary since total size of the raid device 2365 * may vary when an interleave is changed even though exactly 2366 * same components are used, and old disklabel may used 2367 * if that is found. 2368 */ 2369 if (lp->d_secperunit != rs->sc_size) 2370 printf("raid%d: WARNING: %s: " 2371 "total sector size in disklabel (%" PRIu32 ") != " 2372 "the size of raid (%" PRIu64 ")\n", unit, rs->sc_xname, 2373 lp->d_secperunit, rs->sc_size); 2374 for (i = 0; i < lp->d_npartitions; i++) { 2375 pp = &lp->d_partitions[i]; 2376 if (pp->p_offset + pp->p_size > rs->sc_size) 2377 printf("raid%d: WARNING: %s: end of partition `%c' " 2378 "exceeds the size of raid (%" PRIu64 ")\n", 2379 unit, rs->sc_xname, 'a' + i, rs->sc_size); 2380 } 2381 } 2382 2383 } 2384 /* 2385 * Take care of things one might want to take care of in the event 2386 * that a disklabel isn't present. 2387 */ 2388 static void 2389 raidmakedisklabel(struct raid_softc *rs) 2390 { 2391 struct disklabel *lp = rs->sc_dkdev.dk_label; 2392 db1_printf(("Making a label..\n")); 2393 2394 /* 2395 * For historical reasons, if there's no disklabel present 2396 * the raw partition must be marked FS_BSDFFS. 2397 */ 2398 2399 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2400 2401 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2402 2403 lp->d_checksum = dkcksum(lp); 2404 } 2405 /* 2406 * Wait interruptibly for an exclusive lock. 2407 * 2408 * XXX 2409 * Several drivers do this; it should be abstracted and made MP-safe. 2410 * (Hmm... where have we seen this warning before :-> GO ) 2411 */ 2412 static int 2413 raidlock(struct raid_softc *rs) 2414 { 2415 int error; 2416 2417 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2418 rs->sc_flags |= RAIDF_WANTED; 2419 if ((error = 2420 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2421 return (error); 2422 } 2423 rs->sc_flags |= RAIDF_LOCKED; 2424 return (0); 2425 } 2426 /* 2427 * Unlock and wake up any waiters. 2428 */ 2429 static void 2430 raidunlock(struct raid_softc *rs) 2431 { 2432 2433 rs->sc_flags &= ~RAIDF_LOCKED; 2434 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2435 rs->sc_flags &= ~RAIDF_WANTED; 2436 wakeup(rs); 2437 } 2438 } 2439 2440 2441 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2442 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2443 #define RF_PARITY_MAP_OFFSET \ 2444 (RF_COMPONENT_INFO_OFFSET + RF_COMPONENT_INFO_SIZE) 2445 #define RF_PARITY_MAP_SIZE RF_PARITYMAP_NBYTE 2446 2447 int 2448 raidmarkclean(RF_Raid_t *raidPtr, RF_RowCol_t col) 2449 { 2450 RF_ComponentLabel_t *clabel; 2451 2452 clabel = raidget_component_label(raidPtr, col); 2453 clabel->clean = RF_RAID_CLEAN; 2454 raidflush_component_label(raidPtr, col); 2455 return(0); 2456 } 2457 2458 2459 int 2460 raidmarkdirty(RF_Raid_t *raidPtr, RF_RowCol_t col) 2461 { 2462 RF_ComponentLabel_t *clabel; 2463 2464 clabel = raidget_component_label(raidPtr, col); 2465 clabel->clean = RF_RAID_DIRTY; 2466 raidflush_component_label(raidPtr, col); 2467 return(0); 2468 } 2469 2470 int 2471 raidfetch_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2472 { 2473 return raidread_component_label(raidPtr->Disks[col].dev, 2474 raidPtr->raid_cinfo[col].ci_vp, 2475 &raidPtr->raid_cinfo[col].ci_label); 2476 } 2477 2478 RF_ComponentLabel_t * 2479 raidget_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2480 { 2481 return &raidPtr->raid_cinfo[col].ci_label; 2482 } 2483 2484 int 2485 raidflush_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2486 { 2487 RF_ComponentLabel_t *label; 2488 2489 label = &raidPtr->raid_cinfo[col].ci_label; 2490 label->mod_counter = raidPtr->mod_counter; 2491 #ifndef RF_NO_PARITY_MAP 2492 label->parity_map_modcount = label->mod_counter; 2493 #endif 2494 return raidwrite_component_label(raidPtr->Disks[col].dev, 2495 raidPtr->raid_cinfo[col].ci_vp, label); 2496 } 2497 2498 2499 static int 2500 raidread_component_label(dev_t dev, struct vnode *b_vp, 2501 RF_ComponentLabel_t *clabel) 2502 { 2503 return raidread_component_area(dev, b_vp, clabel, 2504 sizeof(RF_ComponentLabel_t), 2505 RF_COMPONENT_INFO_OFFSET, RF_COMPONENT_INFO_SIZE); 2506 } 2507 2508 /* ARGSUSED */ 2509 static int 2510 raidread_component_area(dev_t dev, struct vnode *b_vp, void *data, 2511 size_t msize, daddr_t offset, daddr_t dsize) 2512 { 2513 struct buf *bp; 2514 const struct bdevsw *bdev; 2515 int error; 2516 2517 /* XXX should probably ensure that we don't try to do this if 2518 someone has changed rf_protected_sectors. */ 2519 2520 if (b_vp == NULL) { 2521 /* For whatever reason, this component is not valid. 2522 Don't try to read a component label from it. */ 2523 return(EINVAL); 2524 } 2525 2526 /* get a block of the appropriate size... */ 2527 bp = geteblk((int)dsize); 2528 bp->b_dev = dev; 2529 2530 /* get our ducks in a row for the read */ 2531 bp->b_blkno = offset / DEV_BSIZE; 2532 bp->b_bcount = dsize; 2533 bp->b_flags |= B_READ; 2534 bp->b_resid = dsize; 2535 2536 bdev = bdevsw_lookup(bp->b_dev); 2537 if (bdev == NULL) 2538 return (ENXIO); 2539 (*bdev->d_strategy)(bp); 2540 2541 error = biowait(bp); 2542 2543 if (!error) { 2544 memcpy(data, bp->b_data, msize); 2545 } 2546 2547 brelse(bp, 0); 2548 return(error); 2549 } 2550 2551 2552 static int 2553 raidwrite_component_label(dev_t dev, struct vnode *b_vp, 2554 RF_ComponentLabel_t *clabel) 2555 { 2556 return raidwrite_component_area(dev, b_vp, clabel, 2557 sizeof(RF_ComponentLabel_t), 2558 RF_COMPONENT_INFO_OFFSET, RF_COMPONENT_INFO_SIZE, 0); 2559 } 2560 2561 /* ARGSUSED */ 2562 static int 2563 raidwrite_component_area(dev_t dev, struct vnode *b_vp, void *data, 2564 size_t msize, daddr_t offset, daddr_t dsize, int asyncp) 2565 { 2566 struct buf *bp; 2567 const struct bdevsw *bdev; 2568 int error; 2569 2570 /* get a block of the appropriate size... */ 2571 bp = geteblk((int)dsize); 2572 bp->b_dev = dev; 2573 2574 /* get our ducks in a row for the write */ 2575 bp->b_blkno = offset / DEV_BSIZE; 2576 bp->b_bcount = dsize; 2577 bp->b_flags |= B_WRITE | (asyncp ? B_ASYNC : 0); 2578 bp->b_resid = dsize; 2579 2580 memset(bp->b_data, 0, dsize); 2581 memcpy(bp->b_data, data, msize); 2582 2583 bdev = bdevsw_lookup(bp->b_dev); 2584 if (bdev == NULL) 2585 return (ENXIO); 2586 (*bdev->d_strategy)(bp); 2587 if (asyncp) 2588 return 0; 2589 error = biowait(bp); 2590 brelse(bp, 0); 2591 if (error) { 2592 #if 1 2593 printf("Failed to write RAID component info!\n"); 2594 #endif 2595 } 2596 2597 return(error); 2598 } 2599 2600 void 2601 rf_paritymap_kern_write(RF_Raid_t *raidPtr, struct rf_paritymap_ondisk *map) 2602 { 2603 int c; 2604 2605 for (c = 0; c < raidPtr->numCol; c++) { 2606 /* Skip dead disks. */ 2607 if (RF_DEAD_DISK(raidPtr->Disks[c].status)) 2608 continue; 2609 /* XXXjld: what if an error occurs here? */ 2610 raidwrite_component_area(raidPtr->Disks[c].dev, 2611 raidPtr->raid_cinfo[c].ci_vp, map, 2612 RF_PARITYMAP_NBYTE, 2613 RF_PARITY_MAP_OFFSET, RF_PARITY_MAP_SIZE, 0); 2614 } 2615 } 2616 2617 void 2618 rf_paritymap_kern_read(RF_Raid_t *raidPtr, struct rf_paritymap_ondisk *map) 2619 { 2620 struct rf_paritymap_ondisk tmp; 2621 int c,first; 2622 2623 first=1; 2624 for (c = 0; c < raidPtr->numCol; c++) { 2625 /* Skip dead disks. */ 2626 if (RF_DEAD_DISK(raidPtr->Disks[c].status)) 2627 continue; 2628 raidread_component_area(raidPtr->Disks[c].dev, 2629 raidPtr->raid_cinfo[c].ci_vp, &tmp, 2630 RF_PARITYMAP_NBYTE, 2631 RF_PARITY_MAP_OFFSET, RF_PARITY_MAP_SIZE); 2632 if (first) { 2633 memcpy(map, &tmp, sizeof(*map)); 2634 first = 0; 2635 } else { 2636 rf_paritymap_merge(map, &tmp); 2637 } 2638 } 2639 } 2640 2641 void 2642 rf_markalldirty(RF_Raid_t *raidPtr) 2643 { 2644 RF_ComponentLabel_t *clabel; 2645 int sparecol; 2646 int c; 2647 int j; 2648 int scol = -1; 2649 2650 raidPtr->mod_counter++; 2651 for (c = 0; c < raidPtr->numCol; c++) { 2652 /* we don't want to touch (at all) a disk that has 2653 failed */ 2654 if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) { 2655 clabel = raidget_component_label(raidPtr, c); 2656 if (clabel->status == rf_ds_spared) { 2657 /* XXX do something special... 2658 but whatever you do, don't 2659 try to access it!! */ 2660 } else { 2661 raidmarkdirty(raidPtr, c); 2662 } 2663 } 2664 } 2665 2666 for( c = 0; c < raidPtr->numSpare ; c++) { 2667 sparecol = raidPtr->numCol + c; 2668 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2669 /* 2670 2671 we claim this disk is "optimal" if it's 2672 rf_ds_used_spare, as that means it should be 2673 directly substitutable for the disk it replaced. 2674 We note that too... 2675 2676 */ 2677 2678 for(j=0;j<raidPtr->numCol;j++) { 2679 if (raidPtr->Disks[j].spareCol == sparecol) { 2680 scol = j; 2681 break; 2682 } 2683 } 2684 2685 clabel = raidget_component_label(raidPtr, sparecol); 2686 /* make sure status is noted */ 2687 2688 raid_init_component_label(raidPtr, clabel); 2689 2690 clabel->row = 0; 2691 clabel->column = scol; 2692 /* Note: we *don't* change status from rf_ds_used_spare 2693 to rf_ds_optimal */ 2694 /* clabel.status = rf_ds_optimal; */ 2695 2696 raidmarkdirty(raidPtr, sparecol); 2697 } 2698 } 2699 } 2700 2701 2702 void 2703 rf_update_component_labels(RF_Raid_t *raidPtr, int final) 2704 { 2705 RF_ComponentLabel_t *clabel; 2706 int sparecol; 2707 int c; 2708 int j; 2709 int scol; 2710 2711 scol = -1; 2712 2713 /* XXX should do extra checks to make sure things really are clean, 2714 rather than blindly setting the clean bit... */ 2715 2716 raidPtr->mod_counter++; 2717 2718 for (c = 0; c < raidPtr->numCol; c++) { 2719 if (raidPtr->Disks[c].status == rf_ds_optimal) { 2720 clabel = raidget_component_label(raidPtr, c); 2721 /* make sure status is noted */ 2722 clabel->status = rf_ds_optimal; 2723 2724 /* note what unit we are configured as */ 2725 clabel->last_unit = raidPtr->raidid; 2726 2727 raidflush_component_label(raidPtr, c); 2728 if (final == RF_FINAL_COMPONENT_UPDATE) { 2729 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2730 raidmarkclean(raidPtr, c); 2731 } 2732 } 2733 } 2734 /* else we don't touch it.. */ 2735 } 2736 2737 for( c = 0; c < raidPtr->numSpare ; c++) { 2738 sparecol = raidPtr->numCol + c; 2739 /* Need to ensure that the reconstruct actually completed! */ 2740 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2741 /* 2742 2743 we claim this disk is "optimal" if it's 2744 rf_ds_used_spare, as that means it should be 2745 directly substitutable for the disk it replaced. 2746 We note that too... 2747 2748 */ 2749 2750 for(j=0;j<raidPtr->numCol;j++) { 2751 if (raidPtr->Disks[j].spareCol == sparecol) { 2752 scol = j; 2753 break; 2754 } 2755 } 2756 2757 /* XXX shouldn't *really* need this... */ 2758 clabel = raidget_component_label(raidPtr, sparecol); 2759 /* make sure status is noted */ 2760 2761 raid_init_component_label(raidPtr, clabel); 2762 2763 clabel->column = scol; 2764 clabel->status = rf_ds_optimal; 2765 clabel->last_unit = raidPtr->raidid; 2766 2767 raidflush_component_label(raidPtr, sparecol); 2768 if (final == RF_FINAL_COMPONENT_UPDATE) { 2769 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2770 raidmarkclean(raidPtr, sparecol); 2771 } 2772 } 2773 } 2774 } 2775 } 2776 2777 void 2778 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured) 2779 { 2780 2781 if (vp != NULL) { 2782 if (auto_configured == 1) { 2783 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2784 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2785 vput(vp); 2786 2787 } else { 2788 (void) vn_close(vp, FREAD | FWRITE, curlwp->l_cred); 2789 } 2790 } 2791 } 2792 2793 2794 void 2795 rf_UnconfigureVnodes(RF_Raid_t *raidPtr) 2796 { 2797 int r,c; 2798 struct vnode *vp; 2799 int acd; 2800 2801 2802 /* We take this opportunity to close the vnodes like we should.. */ 2803 2804 for (c = 0; c < raidPtr->numCol; c++) { 2805 vp = raidPtr->raid_cinfo[c].ci_vp; 2806 acd = raidPtr->Disks[c].auto_configured; 2807 rf_close_component(raidPtr, vp, acd); 2808 raidPtr->raid_cinfo[c].ci_vp = NULL; 2809 raidPtr->Disks[c].auto_configured = 0; 2810 } 2811 2812 for (r = 0; r < raidPtr->numSpare; r++) { 2813 vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp; 2814 acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured; 2815 rf_close_component(raidPtr, vp, acd); 2816 raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL; 2817 raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0; 2818 } 2819 } 2820 2821 2822 void 2823 rf_ReconThread(struct rf_recon_req *req) 2824 { 2825 int s; 2826 RF_Raid_t *raidPtr; 2827 2828 s = splbio(); 2829 raidPtr = (RF_Raid_t *) req->raidPtr; 2830 raidPtr->recon_in_progress = 1; 2831 2832 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col, 2833 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2834 2835 RF_Free(req, sizeof(*req)); 2836 2837 raidPtr->recon_in_progress = 0; 2838 splx(s); 2839 2840 /* That's all... */ 2841 kthread_exit(0); /* does not return */ 2842 } 2843 2844 void 2845 rf_RewriteParityThread(RF_Raid_t *raidPtr) 2846 { 2847 int retcode; 2848 int s; 2849 2850 raidPtr->parity_rewrite_stripes_done = 0; 2851 raidPtr->parity_rewrite_in_progress = 1; 2852 s = splbio(); 2853 retcode = rf_RewriteParity(raidPtr); 2854 splx(s); 2855 if (retcode) { 2856 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2857 } else { 2858 /* set the clean bit! If we shutdown correctly, 2859 the clean bit on each component label will get 2860 set */ 2861 raidPtr->parity_good = RF_RAID_CLEAN; 2862 } 2863 raidPtr->parity_rewrite_in_progress = 0; 2864 2865 /* Anyone waiting for us to stop? If so, inform them... */ 2866 if (raidPtr->waitShutdown) { 2867 wakeup(&raidPtr->parity_rewrite_in_progress); 2868 } 2869 2870 /* That's all... */ 2871 kthread_exit(0); /* does not return */ 2872 } 2873 2874 2875 void 2876 rf_CopybackThread(RF_Raid_t *raidPtr) 2877 { 2878 int s; 2879 2880 raidPtr->copyback_in_progress = 1; 2881 s = splbio(); 2882 rf_CopybackReconstructedData(raidPtr); 2883 splx(s); 2884 raidPtr->copyback_in_progress = 0; 2885 2886 /* That's all... */ 2887 kthread_exit(0); /* does not return */ 2888 } 2889 2890 2891 void 2892 rf_ReconstructInPlaceThread(struct rf_recon_req *req) 2893 { 2894 int s; 2895 RF_Raid_t *raidPtr; 2896 2897 s = splbio(); 2898 raidPtr = req->raidPtr; 2899 raidPtr->recon_in_progress = 1; 2900 rf_ReconstructInPlace(raidPtr, req->col); 2901 RF_Free(req, sizeof(*req)); 2902 raidPtr->recon_in_progress = 0; 2903 splx(s); 2904 2905 /* That's all... */ 2906 kthread_exit(0); /* does not return */ 2907 } 2908 2909 static RF_AutoConfig_t * 2910 rf_get_component(RF_AutoConfig_t *ac_list, dev_t dev, struct vnode *vp, 2911 const char *cname, RF_SectorCount_t size) 2912 { 2913 int good_one = 0; 2914 RF_ComponentLabel_t *clabel; 2915 RF_AutoConfig_t *ac; 2916 2917 clabel = malloc(sizeof(RF_ComponentLabel_t), M_RAIDFRAME, M_NOWAIT); 2918 if (clabel == NULL) { 2919 oomem: 2920 while(ac_list) { 2921 ac = ac_list; 2922 if (ac->clabel) 2923 free(ac->clabel, M_RAIDFRAME); 2924 ac_list = ac_list->next; 2925 free(ac, M_RAIDFRAME); 2926 } 2927 printf("RAID auto config: out of memory!\n"); 2928 return NULL; /* XXX probably should panic? */ 2929 } 2930 2931 if (!raidread_component_label(dev, vp, clabel)) { 2932 /* Got the label. Does it look reasonable? */ 2933 if (rf_reasonable_label(clabel) && 2934 (clabel->partitionSize <= size)) { 2935 #ifdef DEBUG 2936 printf("Component on: %s: %llu\n", 2937 cname, (unsigned long long)size); 2938 rf_print_component_label(clabel); 2939 #endif 2940 /* if it's reasonable, add it, else ignore it. */ 2941 ac = malloc(sizeof(RF_AutoConfig_t), M_RAIDFRAME, 2942 M_NOWAIT); 2943 if (ac == NULL) { 2944 free(clabel, M_RAIDFRAME); 2945 goto oomem; 2946 } 2947 strlcpy(ac->devname, cname, sizeof(ac->devname)); 2948 ac->dev = dev; 2949 ac->vp = vp; 2950 ac->clabel = clabel; 2951 ac->next = ac_list; 2952 ac_list = ac; 2953 good_one = 1; 2954 } 2955 } 2956 if (!good_one) { 2957 /* cleanup */ 2958 free(clabel, M_RAIDFRAME); 2959 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2960 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2961 vput(vp); 2962 } 2963 return ac_list; 2964 } 2965 2966 RF_AutoConfig_t * 2967 rf_find_raid_components(void) 2968 { 2969 struct vnode *vp; 2970 struct disklabel label; 2971 device_t dv; 2972 deviter_t di; 2973 dev_t dev; 2974 int bmajor, bminor, wedge; 2975 int error; 2976 int i; 2977 RF_AutoConfig_t *ac_list; 2978 2979 2980 /* initialize the AutoConfig list */ 2981 ac_list = NULL; 2982 2983 /* we begin by trolling through *all* the devices on the system */ 2984 2985 for (dv = deviter_first(&di, DEVITER_F_ROOT_FIRST); dv != NULL; 2986 dv = deviter_next(&di)) { 2987 2988 /* we are only interested in disks... */ 2989 if (device_class(dv) != DV_DISK) 2990 continue; 2991 2992 /* we don't care about floppies... */ 2993 if (device_is_a(dv, "fd")) { 2994 continue; 2995 } 2996 2997 /* we don't care about CD's... */ 2998 if (device_is_a(dv, "cd")) { 2999 continue; 3000 } 3001 3002 /* we don't care about md's... */ 3003 if (device_is_a(dv, "md")) { 3004 continue; 3005 } 3006 3007 /* hdfd is the Atari/Hades floppy driver */ 3008 if (device_is_a(dv, "hdfd")) { 3009 continue; 3010 } 3011 3012 /* fdisa is the Atari/Milan floppy driver */ 3013 if (device_is_a(dv, "fdisa")) { 3014 continue; 3015 } 3016 3017 /* need to find the device_name_to_block_device_major stuff */ 3018 bmajor = devsw_name2blk(device_xname(dv), NULL, 0); 3019 3020 /* get a vnode for the raw partition of this disk */ 3021 3022 wedge = device_is_a(dv, "dk"); 3023 bminor = minor(device_unit(dv)); 3024 dev = wedge ? makedev(bmajor, bminor) : 3025 MAKEDISKDEV(bmajor, bminor, RAW_PART); 3026 if (bdevvp(dev, &vp)) 3027 panic("RAID can't alloc vnode"); 3028 3029 error = VOP_OPEN(vp, FREAD, NOCRED); 3030 3031 if (error) { 3032 /* "Who cares." Continue looking 3033 for something that exists*/ 3034 vput(vp); 3035 continue; 3036 } 3037 3038 if (wedge) { 3039 struct dkwedge_info dkw; 3040 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, 3041 NOCRED); 3042 if (error) { 3043 printf("RAIDframe: can't get wedge info for " 3044 "dev %s (%d)\n", device_xname(dv), error); 3045 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3046 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3047 vput(vp); 3048 continue; 3049 } 3050 3051 if (strcmp(dkw.dkw_ptype, DKW_PTYPE_RAIDFRAME) != 0) { 3052 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3053 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3054 vput(vp); 3055 continue; 3056 } 3057 3058 ac_list = rf_get_component(ac_list, dev, vp, 3059 device_xname(dv), dkw.dkw_size); 3060 continue; 3061 } 3062 3063 /* Ok, the disk exists. Go get the disklabel. */ 3064 error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED); 3065 if (error) { 3066 /* 3067 * XXX can't happen - open() would 3068 * have errored out (or faked up one) 3069 */ 3070 if (error != ENOTTY) 3071 printf("RAIDframe: can't get label for dev " 3072 "%s (%d)\n", device_xname(dv), error); 3073 } 3074 3075 /* don't need this any more. We'll allocate it again 3076 a little later if we really do... */ 3077 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3078 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3079 vput(vp); 3080 3081 if (error) 3082 continue; 3083 3084 for (i = 0; i < label.d_npartitions; i++) { 3085 char cname[sizeof(ac_list->devname)]; 3086 3087 /* We only support partitions marked as RAID */ 3088 if (label.d_partitions[i].p_fstype != FS_RAID) 3089 continue; 3090 3091 dev = MAKEDISKDEV(bmajor, device_unit(dv), i); 3092 if (bdevvp(dev, &vp)) 3093 panic("RAID can't alloc vnode"); 3094 3095 error = VOP_OPEN(vp, FREAD, NOCRED); 3096 if (error) { 3097 /* Whatever... */ 3098 vput(vp); 3099 continue; 3100 } 3101 snprintf(cname, sizeof(cname), "%s%c", 3102 device_xname(dv), 'a' + i); 3103 ac_list = rf_get_component(ac_list, dev, vp, cname, 3104 label.d_partitions[i].p_size); 3105 } 3106 } 3107 deviter_release(&di); 3108 return ac_list; 3109 } 3110 3111 3112 static int 3113 rf_reasonable_label(RF_ComponentLabel_t *clabel) 3114 { 3115 3116 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 3117 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 3118 ((clabel->clean == RF_RAID_CLEAN) || 3119 (clabel->clean == RF_RAID_DIRTY)) && 3120 clabel->row >=0 && 3121 clabel->column >= 0 && 3122 clabel->num_rows > 0 && 3123 clabel->num_columns > 0 && 3124 clabel->row < clabel->num_rows && 3125 clabel->column < clabel->num_columns && 3126 clabel->blockSize > 0 && 3127 clabel->numBlocks > 0) { 3128 /* label looks reasonable enough... */ 3129 return(1); 3130 } 3131 return(0); 3132 } 3133 3134 3135 #ifdef DEBUG 3136 void 3137 rf_print_component_label(RF_ComponentLabel_t *clabel) 3138 { 3139 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 3140 clabel->row, clabel->column, 3141 clabel->num_rows, clabel->num_columns); 3142 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 3143 clabel->version, clabel->serial_number, 3144 clabel->mod_counter); 3145 printf(" Clean: %s Status: %d\n", 3146 clabel->clean ? "Yes" : "No", clabel->status); 3147 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 3148 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 3149 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n", 3150 (char) clabel->parityConfig, clabel->blockSize, 3151 clabel->numBlocks); 3152 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No"); 3153 printf(" Contains root partition: %s\n", 3154 clabel->root_partition ? "Yes" : "No"); 3155 printf(" Last configured as: raid%d\n", clabel->last_unit); 3156 #if 0 3157 printf(" Config order: %d\n", clabel->config_order); 3158 #endif 3159 3160 } 3161 #endif 3162 3163 RF_ConfigSet_t * 3164 rf_create_auto_sets(RF_AutoConfig_t *ac_list) 3165 { 3166 RF_AutoConfig_t *ac; 3167 RF_ConfigSet_t *config_sets; 3168 RF_ConfigSet_t *cset; 3169 RF_AutoConfig_t *ac_next; 3170 3171 3172 config_sets = NULL; 3173 3174 /* Go through the AutoConfig list, and figure out which components 3175 belong to what sets. */ 3176 ac = ac_list; 3177 while(ac!=NULL) { 3178 /* we're going to putz with ac->next, so save it here 3179 for use at the end of the loop */ 3180 ac_next = ac->next; 3181 3182 if (config_sets == NULL) { 3183 /* will need at least this one... */ 3184 config_sets = (RF_ConfigSet_t *) 3185 malloc(sizeof(RF_ConfigSet_t), 3186 M_RAIDFRAME, M_NOWAIT); 3187 if (config_sets == NULL) { 3188 panic("rf_create_auto_sets: No memory!"); 3189 } 3190 /* this one is easy :) */ 3191 config_sets->ac = ac; 3192 config_sets->next = NULL; 3193 config_sets->rootable = 0; 3194 ac->next = NULL; 3195 } else { 3196 /* which set does this component fit into? */ 3197 cset = config_sets; 3198 while(cset!=NULL) { 3199 if (rf_does_it_fit(cset, ac)) { 3200 /* looks like it matches... */ 3201 ac->next = cset->ac; 3202 cset->ac = ac; 3203 break; 3204 } 3205 cset = cset->next; 3206 } 3207 if (cset==NULL) { 3208 /* didn't find a match above... new set..*/ 3209 cset = (RF_ConfigSet_t *) 3210 malloc(sizeof(RF_ConfigSet_t), 3211 M_RAIDFRAME, M_NOWAIT); 3212 if (cset == NULL) { 3213 panic("rf_create_auto_sets: No memory!"); 3214 } 3215 cset->ac = ac; 3216 ac->next = NULL; 3217 cset->next = config_sets; 3218 cset->rootable = 0; 3219 config_sets = cset; 3220 } 3221 } 3222 ac = ac_next; 3223 } 3224 3225 3226 return(config_sets); 3227 } 3228 3229 static int 3230 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac) 3231 { 3232 RF_ComponentLabel_t *clabel1, *clabel2; 3233 3234 /* If this one matches the *first* one in the set, that's good 3235 enough, since the other members of the set would have been 3236 through here too... */ 3237 /* note that we are not checking partitionSize here.. 3238 3239 Note that we are also not checking the mod_counters here. 3240 If everything else matches execpt the mod_counter, that's 3241 good enough for this test. We will deal with the mod_counters 3242 a little later in the autoconfiguration process. 3243 3244 (clabel1->mod_counter == clabel2->mod_counter) && 3245 3246 The reason we don't check for this is that failed disks 3247 will have lower modification counts. If those disks are 3248 not added to the set they used to belong to, then they will 3249 form their own set, which may result in 2 different sets, 3250 for example, competing to be configured at raid0, and 3251 perhaps competing to be the root filesystem set. If the 3252 wrong ones get configured, or both attempt to become /, 3253 weird behaviour and or serious lossage will occur. Thus we 3254 need to bring them into the fold here, and kick them out at 3255 a later point. 3256 3257 */ 3258 3259 clabel1 = cset->ac->clabel; 3260 clabel2 = ac->clabel; 3261 if ((clabel1->version == clabel2->version) && 3262 (clabel1->serial_number == clabel2->serial_number) && 3263 (clabel1->num_rows == clabel2->num_rows) && 3264 (clabel1->num_columns == clabel2->num_columns) && 3265 (clabel1->sectPerSU == clabel2->sectPerSU) && 3266 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 3267 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 3268 (clabel1->parityConfig == clabel2->parityConfig) && 3269 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 3270 (clabel1->blockSize == clabel2->blockSize) && 3271 (clabel1->numBlocks == clabel2->numBlocks) && 3272 (clabel1->autoconfigure == clabel2->autoconfigure) && 3273 (clabel1->root_partition == clabel2->root_partition) && 3274 (clabel1->last_unit == clabel2->last_unit) && 3275 (clabel1->config_order == clabel2->config_order)) { 3276 /* if it get's here, it almost *has* to be a match */ 3277 } else { 3278 /* it's not consistent with somebody in the set.. 3279 punt */ 3280 return(0); 3281 } 3282 /* all was fine.. it must fit... */ 3283 return(1); 3284 } 3285 3286 int 3287 rf_have_enough_components(RF_ConfigSet_t *cset) 3288 { 3289 RF_AutoConfig_t *ac; 3290 RF_AutoConfig_t *auto_config; 3291 RF_ComponentLabel_t *clabel; 3292 int c; 3293 int num_cols; 3294 int num_missing; 3295 int mod_counter; 3296 int mod_counter_found; 3297 int even_pair_failed; 3298 char parity_type; 3299 3300 3301 /* check to see that we have enough 'live' components 3302 of this set. If so, we can configure it if necessary */ 3303 3304 num_cols = cset->ac->clabel->num_columns; 3305 parity_type = cset->ac->clabel->parityConfig; 3306 3307 /* XXX Check for duplicate components!?!?!? */ 3308 3309 /* Determine what the mod_counter is supposed to be for this set. */ 3310 3311 mod_counter_found = 0; 3312 mod_counter = 0; 3313 ac = cset->ac; 3314 while(ac!=NULL) { 3315 if (mod_counter_found==0) { 3316 mod_counter = ac->clabel->mod_counter; 3317 mod_counter_found = 1; 3318 } else { 3319 if (ac->clabel->mod_counter > mod_counter) { 3320 mod_counter = ac->clabel->mod_counter; 3321 } 3322 } 3323 ac = ac->next; 3324 } 3325 3326 num_missing = 0; 3327 auto_config = cset->ac; 3328 3329 even_pair_failed = 0; 3330 for(c=0; c<num_cols; c++) { 3331 ac = auto_config; 3332 while(ac!=NULL) { 3333 if ((ac->clabel->column == c) && 3334 (ac->clabel->mod_counter == mod_counter)) { 3335 /* it's this one... */ 3336 #ifdef DEBUG 3337 printf("Found: %s at %d\n", 3338 ac->devname,c); 3339 #endif 3340 break; 3341 } 3342 ac=ac->next; 3343 } 3344 if (ac==NULL) { 3345 /* Didn't find one here! */ 3346 /* special case for RAID 1, especially 3347 where there are more than 2 3348 components (where RAIDframe treats 3349 things a little differently :( ) */ 3350 if (parity_type == '1') { 3351 if (c%2 == 0) { /* even component */ 3352 even_pair_failed = 1; 3353 } else { /* odd component. If 3354 we're failed, and 3355 so is the even 3356 component, it's 3357 "Good Night, Charlie" */ 3358 if (even_pair_failed == 1) { 3359 return(0); 3360 } 3361 } 3362 } else { 3363 /* normal accounting */ 3364 num_missing++; 3365 } 3366 } 3367 if ((parity_type == '1') && (c%2 == 1)) { 3368 /* Just did an even component, and we didn't 3369 bail.. reset the even_pair_failed flag, 3370 and go on to the next component.... */ 3371 even_pair_failed = 0; 3372 } 3373 } 3374 3375 clabel = cset->ac->clabel; 3376 3377 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3378 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3379 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3380 /* XXX this needs to be made *much* more general */ 3381 /* Too many failures */ 3382 return(0); 3383 } 3384 /* otherwise, all is well, and we've got enough to take a kick 3385 at autoconfiguring this set */ 3386 return(1); 3387 } 3388 3389 void 3390 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config, 3391 RF_Raid_t *raidPtr) 3392 { 3393 RF_ComponentLabel_t *clabel; 3394 int i; 3395 3396 clabel = ac->clabel; 3397 3398 /* 1. Fill in the common stuff */ 3399 config->numRow = clabel->num_rows = 1; 3400 config->numCol = clabel->num_columns; 3401 config->numSpare = 0; /* XXX should this be set here? */ 3402 config->sectPerSU = clabel->sectPerSU; 3403 config->SUsPerPU = clabel->SUsPerPU; 3404 config->SUsPerRU = clabel->SUsPerRU; 3405 config->parityConfig = clabel->parityConfig; 3406 /* XXX... */ 3407 strcpy(config->diskQueueType,"fifo"); 3408 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3409 config->layoutSpecificSize = 0; /* XXX ?? */ 3410 3411 while(ac!=NULL) { 3412 /* row/col values will be in range due to the checks 3413 in reasonable_label() */ 3414 strcpy(config->devnames[0][ac->clabel->column], 3415 ac->devname); 3416 ac = ac->next; 3417 } 3418 3419 for(i=0;i<RF_MAXDBGV;i++) { 3420 config->debugVars[i][0] = 0; 3421 } 3422 } 3423 3424 int 3425 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value) 3426 { 3427 RF_ComponentLabel_t *clabel; 3428 int column; 3429 int sparecol; 3430 3431 raidPtr->autoconfigure = new_value; 3432 3433 for(column=0; column<raidPtr->numCol; column++) { 3434 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3435 clabel = raidget_component_label(raidPtr, column); 3436 clabel->autoconfigure = new_value; 3437 raidflush_component_label(raidPtr, column); 3438 } 3439 } 3440 for(column = 0; column < raidPtr->numSpare ; column++) { 3441 sparecol = raidPtr->numCol + column; 3442 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3443 clabel = raidget_component_label(raidPtr, sparecol); 3444 clabel->autoconfigure = new_value; 3445 raidflush_component_label(raidPtr, sparecol); 3446 } 3447 } 3448 return(new_value); 3449 } 3450 3451 int 3452 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value) 3453 { 3454 RF_ComponentLabel_t *clabel; 3455 int column; 3456 int sparecol; 3457 3458 raidPtr->root_partition = new_value; 3459 for(column=0; column<raidPtr->numCol; column++) { 3460 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3461 clabel = raidget_component_label(raidPtr, column); 3462 clabel->root_partition = new_value; 3463 raidflush_component_label(raidPtr, column); 3464 } 3465 } 3466 for(column = 0; column < raidPtr->numSpare ; column++) { 3467 sparecol = raidPtr->numCol + column; 3468 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3469 clabel = raidget_component_label(raidPtr, sparecol); 3470 clabel->root_partition = new_value; 3471 raidflush_component_label(raidPtr, sparecol); 3472 } 3473 } 3474 return(new_value); 3475 } 3476 3477 void 3478 rf_release_all_vps(RF_ConfigSet_t *cset) 3479 { 3480 RF_AutoConfig_t *ac; 3481 3482 ac = cset->ac; 3483 while(ac!=NULL) { 3484 /* Close the vp, and give it back */ 3485 if (ac->vp) { 3486 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3487 VOP_CLOSE(ac->vp, FREAD, NOCRED); 3488 vput(ac->vp); 3489 ac->vp = NULL; 3490 } 3491 ac = ac->next; 3492 } 3493 } 3494 3495 3496 void 3497 rf_cleanup_config_set(RF_ConfigSet_t *cset) 3498 { 3499 RF_AutoConfig_t *ac; 3500 RF_AutoConfig_t *next_ac; 3501 3502 ac = cset->ac; 3503 while(ac!=NULL) { 3504 next_ac = ac->next; 3505 /* nuke the label */ 3506 free(ac->clabel, M_RAIDFRAME); 3507 /* cleanup the config structure */ 3508 free(ac, M_RAIDFRAME); 3509 /* "next.." */ 3510 ac = next_ac; 3511 } 3512 /* and, finally, nuke the config set */ 3513 free(cset, M_RAIDFRAME); 3514 } 3515 3516 3517 void 3518 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel) 3519 { 3520 /* current version number */ 3521 clabel->version = RF_COMPONENT_LABEL_VERSION; 3522 clabel->serial_number = raidPtr->serial_number; 3523 clabel->mod_counter = raidPtr->mod_counter; 3524 3525 clabel->num_rows = 1; 3526 clabel->num_columns = raidPtr->numCol; 3527 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3528 clabel->status = rf_ds_optimal; /* "It's good!" */ 3529 3530 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3531 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3532 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3533 3534 clabel->blockSize = raidPtr->bytesPerSector; 3535 clabel->numBlocks = raidPtr->sectorsPerDisk; 3536 3537 /* XXX not portable */ 3538 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3539 clabel->maxOutstanding = raidPtr->maxOutstanding; 3540 clabel->autoconfigure = raidPtr->autoconfigure; 3541 clabel->root_partition = raidPtr->root_partition; 3542 clabel->last_unit = raidPtr->raidid; 3543 clabel->config_order = raidPtr->config_order; 3544 3545 #ifndef RF_NO_PARITY_MAP 3546 rf_paritymap_init_label(raidPtr->parity_map, clabel); 3547 #endif 3548 } 3549 3550 int 3551 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit) 3552 { 3553 RF_Raid_t *raidPtr; 3554 RF_Config_t *config; 3555 int raidID; 3556 int retcode; 3557 3558 #ifdef DEBUG 3559 printf("RAID autoconfigure\n"); 3560 #endif 3561 3562 retcode = 0; 3563 *unit = -1; 3564 3565 /* 1. Create a config structure */ 3566 3567 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3568 M_RAIDFRAME, 3569 M_NOWAIT); 3570 if (config==NULL) { 3571 printf("Out of mem!?!?\n"); 3572 /* XXX do something more intelligent here. */ 3573 return(1); 3574 } 3575 3576 memset(config, 0, sizeof(RF_Config_t)); 3577 3578 /* 3579 2. Figure out what RAID ID this one is supposed to live at 3580 See if we can get the same RAID dev that it was configured 3581 on last time.. 3582 */ 3583 3584 raidID = cset->ac->clabel->last_unit; 3585 if ((raidID < 0) || (raidID >= numraid)) { 3586 /* let's not wander off into lala land. */ 3587 raidID = numraid - 1; 3588 } 3589 if (raidPtrs[raidID]->valid != 0) { 3590 3591 /* 3592 Nope... Go looking for an alternative... 3593 Start high so we don't immediately use raid0 if that's 3594 not taken. 3595 */ 3596 3597 for(raidID = numraid - 1; raidID >= 0; raidID--) { 3598 if (raidPtrs[raidID]->valid == 0) { 3599 /* can use this one! */ 3600 break; 3601 } 3602 } 3603 } 3604 3605 if (raidID < 0) { 3606 /* punt... */ 3607 printf("Unable to auto configure this set!\n"); 3608 printf("(Out of RAID devs!)\n"); 3609 free(config, M_RAIDFRAME); 3610 return(1); 3611 } 3612 3613 #ifdef DEBUG 3614 printf("Configuring raid%d:\n",raidID); 3615 #endif 3616 3617 raidPtr = raidPtrs[raidID]; 3618 3619 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3620 raidPtr->raidid = raidID; 3621 raidPtr->openings = RAIDOUTSTANDING; 3622 3623 /* 3. Build the configuration structure */ 3624 rf_create_configuration(cset->ac, config, raidPtr); 3625 3626 /* 4. Do the configuration */ 3627 retcode = rf_Configure(raidPtr, config, cset->ac); 3628 3629 if (retcode == 0) { 3630 3631 raidinit(raidPtrs[raidID]); 3632 3633 rf_markalldirty(raidPtrs[raidID]); 3634 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3635 if (cset->ac->clabel->root_partition==1) { 3636 /* everything configured just fine. Make a note 3637 that this set is eligible to be root. */ 3638 cset->rootable = 1; 3639 /* XXX do this here? */ 3640 raidPtrs[raidID]->root_partition = 1; 3641 } 3642 } 3643 3644 /* 5. Cleanup */ 3645 free(config, M_RAIDFRAME); 3646 3647 *unit = raidID; 3648 return(retcode); 3649 } 3650 3651 void 3652 rf_disk_unbusy(RF_RaidAccessDesc_t *desc) 3653 { 3654 struct buf *bp; 3655 3656 bp = (struct buf *)desc->bp; 3657 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3658 (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ)); 3659 } 3660 3661 void 3662 rf_pool_init(struct pool *p, size_t size, const char *w_chan, 3663 size_t xmin, size_t xmax) 3664 { 3665 pool_init(p, size, 0, 0, 0, w_chan, NULL, IPL_BIO); 3666 pool_sethiwat(p, xmax); 3667 pool_prime(p, xmin); 3668 pool_setlowat(p, xmin); 3669 } 3670 3671 /* 3672 * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see 3673 * if there is IO pending and if that IO could possibly be done for a 3674 * given RAID set. Returns 0 if IO is waiting and can be done, 1 3675 * otherwise. 3676 * 3677 */ 3678 3679 int 3680 rf_buf_queue_check(int raidid) 3681 { 3682 if ((bufq_peek(raid_softc[raidid].buf_queue) != NULL) && 3683 raidPtrs[raidid]->openings > 0) { 3684 /* there is work to do */ 3685 return 0; 3686 } 3687 /* default is nothing to do */ 3688 return 1; 3689 } 3690 3691 int 3692 rf_getdisksize(struct vnode *vp, struct lwp *l, RF_RaidDisk_t *diskPtr) 3693 { 3694 struct partinfo dpart; 3695 struct dkwedge_info dkw; 3696 int error; 3697 3698 error = VOP_IOCTL(vp, DIOCGPART, &dpart, FREAD, l->l_cred); 3699 if (error == 0) { 3700 diskPtr->blockSize = dpart.disklab->d_secsize; 3701 diskPtr->numBlocks = dpart.part->p_size - rf_protectedSectors; 3702 diskPtr->partitionSize = dpart.part->p_size; 3703 return 0; 3704 } 3705 3706 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, l->l_cred); 3707 if (error == 0) { 3708 diskPtr->blockSize = 512; /* XXX */ 3709 diskPtr->numBlocks = dkw.dkw_size - rf_protectedSectors; 3710 diskPtr->partitionSize = dkw.dkw_size; 3711 return 0; 3712 } 3713 return error; 3714 } 3715 3716 static int 3717 raid_match(device_t self, cfdata_t cfdata, void *aux) 3718 { 3719 return 1; 3720 } 3721 3722 static void 3723 raid_attach(device_t parent, device_t self, void *aux) 3724 { 3725 3726 } 3727 3728 3729 static int 3730 raid_detach(device_t self, int flags) 3731 { 3732 int error; 3733 struct raid_softc *rs = &raid_softc[device_unit(self)]; 3734 3735 if ((error = raidlock(rs)) != 0) 3736 return (error); 3737 3738 error = raid_detach_unlocked(rs); 3739 3740 raidunlock(rs); 3741 3742 return error; 3743 } 3744 3745 static void 3746 rf_set_properties(struct raid_softc *rs, RF_Raid_t *raidPtr) 3747 { 3748 prop_dictionary_t disk_info, odisk_info, geom; 3749 disk_info = prop_dictionary_create(); 3750 geom = prop_dictionary_create(); 3751 prop_dictionary_set_uint64(geom, "sectors-per-unit", 3752 raidPtr->totalSectors); 3753 prop_dictionary_set_uint32(geom, "sector-size", 3754 raidPtr->bytesPerSector); 3755 3756 prop_dictionary_set_uint16(geom, "sectors-per-track", 3757 raidPtr->Layout.dataSectorsPerStripe); 3758 prop_dictionary_set_uint16(geom, "tracks-per-cylinder", 3759 4 * raidPtr->numCol); 3760 3761 prop_dictionary_set_uint64(geom, "cylinders-per-unit", 3762 raidPtr->totalSectors / (raidPtr->Layout.dataSectorsPerStripe * 3763 (4 * raidPtr->numCol))); 3764 3765 prop_dictionary_set(disk_info, "geometry", geom); 3766 prop_object_release(geom); 3767 prop_dictionary_set(device_properties(rs->sc_dev), 3768 "disk-info", disk_info); 3769 odisk_info = rs->sc_dkdev.dk_info; 3770 rs->sc_dkdev.dk_info = disk_info; 3771 if (odisk_info) 3772 prop_object_release(odisk_info); 3773 } 3774 3775 /* 3776 * Implement forwarding of the DIOCCACHESYNC ioctl to each of the components. 3777 * We end up returning whatever error was returned by the first cache flush 3778 * that fails. 3779 */ 3780 3781 int 3782 rf_sync_component_caches(RF_Raid_t *raidPtr) 3783 { 3784 int c, sparecol; 3785 int e,error; 3786 int force = 1; 3787 3788 error = 0; 3789 for (c = 0; c < raidPtr->numCol; c++) { 3790 if (raidPtr->Disks[c].status == rf_ds_optimal) { 3791 e = VOP_IOCTL(raidPtr->raid_cinfo[c].ci_vp, DIOCCACHESYNC, 3792 &force, FWRITE, NOCRED); 3793 if (e) { 3794 if (e != ENODEV) 3795 printf("raid%d: cache flush to component %s failed.\n", 3796 raidPtr->raidid, raidPtr->Disks[c].devname); 3797 if (error == 0) { 3798 error = e; 3799 } 3800 } 3801 } 3802 } 3803 3804 for( c = 0; c < raidPtr->numSpare ; c++) { 3805 sparecol = raidPtr->numCol + c; 3806 /* Need to ensure that the reconstruct actually completed! */ 3807 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3808 e = VOP_IOCTL(raidPtr->raid_cinfo[sparecol].ci_vp, 3809 DIOCCACHESYNC, &force, FWRITE, NOCRED); 3810 if (e) { 3811 if (e != ENODEV) 3812 printf("raid%d: cache flush to component %s failed.\n", 3813 raidPtr->raidid, raidPtr->Disks[sparecol].devname); 3814 if (error == 0) { 3815 error = e; 3816 } 3817 } 3818 } 3819 } 3820 return error; 3821 } 3822