1 /* $NetBSD: rf_netbsdkintf.c,v 1.271 2009/12/01 01:03:54 dyoung 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 1.271 2009/12/01 01:03:54 dyoung 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_normal("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 rf_paritymap_status(raidPtr->parity_map, 1510 (struct rf_pmstat *)data); 1511 return 0; 1512 1513 case RAIDFRAME_PARITYMAP_SET_PARAMS: 1514 if (raidPtr->parity_map == NULL) 1515 return ENOENT; /* ??? */ 1516 if (0 != rf_paritymap_set_params(raidPtr->parity_map, 1517 (struct rf_pmparams *)data, 1)) 1518 return EINVAL; 1519 return 0; 1520 1521 case RAIDFRAME_PARITYMAP_GET_DISABLE: 1522 *(int *) data = rf_paritymap_get_disable(raidPtr); 1523 return 0; 1524 1525 case RAIDFRAME_PARITYMAP_SET_DISABLE: 1526 rf_paritymap_set_disable(raidPtr, *(int *)data); 1527 /* XXX should errors be passed up? */ 1528 return 0; 1529 1530 case RAIDFRAME_RESET_ACCTOTALS: 1531 memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals)); 1532 return (0); 1533 1534 case RAIDFRAME_GET_ACCTOTALS: 1535 totals = (RF_AccTotals_t *) data; 1536 *totals = raidPtr->acc_totals; 1537 return (0); 1538 1539 case RAIDFRAME_KEEP_ACCTOTALS: 1540 raidPtr->keep_acc_totals = *(int *)data; 1541 return (0); 1542 1543 case RAIDFRAME_GET_SIZE: 1544 *(int *) data = raidPtr->totalSectors; 1545 return (0); 1546 1547 /* fail a disk & optionally start reconstruction */ 1548 case RAIDFRAME_FAIL_DISK: 1549 1550 if (raidPtr->Layout.map->faultsTolerated == 0) { 1551 /* Can't do this on a RAID 0!! */ 1552 return(EINVAL); 1553 } 1554 1555 rr = (struct rf_recon_req *) data; 1556 rr->row = 0; 1557 if (rr->col < 0 || rr->col >= raidPtr->numCol) 1558 return (EINVAL); 1559 1560 1561 RF_LOCK_MUTEX(raidPtr->mutex); 1562 if (raidPtr->status == rf_rs_reconstructing) { 1563 /* you can't fail a disk while we're reconstructing! */ 1564 /* XXX wrong for RAID6 */ 1565 RF_UNLOCK_MUTEX(raidPtr->mutex); 1566 return (EINVAL); 1567 } 1568 if ((raidPtr->Disks[rr->col].status == 1569 rf_ds_optimal) && (raidPtr->numFailures > 0)) { 1570 /* some other component has failed. Let's not make 1571 things worse. XXX wrong for RAID6 */ 1572 RF_UNLOCK_MUTEX(raidPtr->mutex); 1573 return (EINVAL); 1574 } 1575 if (raidPtr->Disks[rr->col].status == rf_ds_spared) { 1576 /* Can't fail a spared disk! */ 1577 RF_UNLOCK_MUTEX(raidPtr->mutex); 1578 return (EINVAL); 1579 } 1580 RF_UNLOCK_MUTEX(raidPtr->mutex); 1581 1582 /* make a copy of the recon request so that we don't rely on 1583 * the user's buffer */ 1584 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *)); 1585 if (rrcopy == NULL) 1586 return(ENOMEM); 1587 memcpy(rrcopy, rr, sizeof(*rr)); 1588 rrcopy->raidPtr = (void *) raidPtr; 1589 1590 retcode = RF_CREATE_THREAD(raidPtr->recon_thread, 1591 rf_ReconThread, 1592 rrcopy,"raid_recon"); 1593 return (0); 1594 1595 /* invoke a copyback operation after recon on whatever disk 1596 * needs it, if any */ 1597 case RAIDFRAME_COPYBACK: 1598 1599 if (raidPtr->Layout.map->faultsTolerated == 0) { 1600 /* This makes no sense on a RAID 0!! */ 1601 return(EINVAL); 1602 } 1603 1604 if (raidPtr->copyback_in_progress == 1) { 1605 /* Copyback is already in progress! */ 1606 return(EINVAL); 1607 } 1608 1609 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread, 1610 rf_CopybackThread, 1611 raidPtr,"raid_copyback"); 1612 return (retcode); 1613 1614 /* return the percentage completion of reconstruction */ 1615 case RAIDFRAME_CHECK_RECON_STATUS: 1616 if (raidPtr->Layout.map->faultsTolerated == 0) { 1617 /* This makes no sense on a RAID 0, so tell the 1618 user it's done. */ 1619 *(int *) data = 100; 1620 return(0); 1621 } 1622 if (raidPtr->status != rf_rs_reconstructing) 1623 *(int *) data = 100; 1624 else { 1625 if (raidPtr->reconControl->numRUsTotal > 0) { 1626 *(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal); 1627 } else { 1628 *(int *) data = 0; 1629 } 1630 } 1631 return (0); 1632 case RAIDFRAME_CHECK_RECON_STATUS_EXT: 1633 progressInfoPtr = (RF_ProgressInfo_t **) data; 1634 if (raidPtr->status != rf_rs_reconstructing) { 1635 progressInfo.remaining = 0; 1636 progressInfo.completed = 100; 1637 progressInfo.total = 100; 1638 } else { 1639 progressInfo.total = 1640 raidPtr->reconControl->numRUsTotal; 1641 progressInfo.completed = 1642 raidPtr->reconControl->numRUsComplete; 1643 progressInfo.remaining = progressInfo.total - 1644 progressInfo.completed; 1645 } 1646 retcode = copyout(&progressInfo, *progressInfoPtr, 1647 sizeof(RF_ProgressInfo_t)); 1648 return (retcode); 1649 1650 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS: 1651 if (raidPtr->Layout.map->faultsTolerated == 0) { 1652 /* This makes no sense on a RAID 0, so tell the 1653 user it's done. */ 1654 *(int *) data = 100; 1655 return(0); 1656 } 1657 if (raidPtr->parity_rewrite_in_progress == 1) { 1658 *(int *) data = 100 * 1659 raidPtr->parity_rewrite_stripes_done / 1660 raidPtr->Layout.numStripe; 1661 } else { 1662 *(int *) data = 100; 1663 } 1664 return (0); 1665 1666 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT: 1667 progressInfoPtr = (RF_ProgressInfo_t **) data; 1668 if (raidPtr->parity_rewrite_in_progress == 1) { 1669 progressInfo.total = raidPtr->Layout.numStripe; 1670 progressInfo.completed = 1671 raidPtr->parity_rewrite_stripes_done; 1672 progressInfo.remaining = progressInfo.total - 1673 progressInfo.completed; 1674 } else { 1675 progressInfo.remaining = 0; 1676 progressInfo.completed = 100; 1677 progressInfo.total = 100; 1678 } 1679 retcode = copyout(&progressInfo, *progressInfoPtr, 1680 sizeof(RF_ProgressInfo_t)); 1681 return (retcode); 1682 1683 case RAIDFRAME_CHECK_COPYBACK_STATUS: 1684 if (raidPtr->Layout.map->faultsTolerated == 0) { 1685 /* This makes no sense on a RAID 0 */ 1686 *(int *) data = 100; 1687 return(0); 1688 } 1689 if (raidPtr->copyback_in_progress == 1) { 1690 *(int *) data = 100 * raidPtr->copyback_stripes_done / 1691 raidPtr->Layout.numStripe; 1692 } else { 1693 *(int *) data = 100; 1694 } 1695 return (0); 1696 1697 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT: 1698 progressInfoPtr = (RF_ProgressInfo_t **) data; 1699 if (raidPtr->copyback_in_progress == 1) { 1700 progressInfo.total = raidPtr->Layout.numStripe; 1701 progressInfo.completed = 1702 raidPtr->copyback_stripes_done; 1703 progressInfo.remaining = progressInfo.total - 1704 progressInfo.completed; 1705 } else { 1706 progressInfo.remaining = 0; 1707 progressInfo.completed = 100; 1708 progressInfo.total = 100; 1709 } 1710 retcode = copyout(&progressInfo, *progressInfoPtr, 1711 sizeof(RF_ProgressInfo_t)); 1712 return (retcode); 1713 1714 /* the sparetable daemon calls this to wait for the kernel to 1715 * need a spare table. this ioctl does not return until a 1716 * spare table is needed. XXX -- calling mpsleep here in the 1717 * ioctl code is almost certainly wrong and evil. -- XXX XXX 1718 * -- I should either compute the spare table in the kernel, 1719 * or have a different -- XXX XXX -- interface (a different 1720 * character device) for delivering the table -- XXX */ 1721 #if 0 1722 case RAIDFRAME_SPARET_WAIT: 1723 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1724 while (!rf_sparet_wait_queue) 1725 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE); 1726 waitreq = rf_sparet_wait_queue; 1727 rf_sparet_wait_queue = rf_sparet_wait_queue->next; 1728 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1729 1730 /* structure assignment */ 1731 *((RF_SparetWait_t *) data) = *waitreq; 1732 1733 RF_Free(waitreq, sizeof(*waitreq)); 1734 return (0); 1735 1736 /* wakes up a process waiting on SPARET_WAIT and puts an error 1737 * code in it that will cause the dameon to exit */ 1738 case RAIDFRAME_ABORT_SPARET_WAIT: 1739 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1740 waitreq->fcol = -1; 1741 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1742 waitreq->next = rf_sparet_wait_queue; 1743 rf_sparet_wait_queue = waitreq; 1744 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1745 wakeup(&rf_sparet_wait_queue); 1746 return (0); 1747 1748 /* used by the spare table daemon to deliver a spare table 1749 * into the kernel */ 1750 case RAIDFRAME_SEND_SPARET: 1751 1752 /* install the spare table */ 1753 retcode = rf_SetSpareTable(raidPtr, *(void **) data); 1754 1755 /* respond to the requestor. the return status of the spare 1756 * table installation is passed in the "fcol" field */ 1757 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *)); 1758 waitreq->fcol = retcode; 1759 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1760 waitreq->next = rf_sparet_resp_queue; 1761 rf_sparet_resp_queue = waitreq; 1762 wakeup(&rf_sparet_resp_queue); 1763 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1764 1765 return (retcode); 1766 #endif 1767 1768 default: 1769 break; /* fall through to the os-specific code below */ 1770 1771 } 1772 1773 if (!raidPtr->valid) 1774 return (EINVAL); 1775 1776 /* 1777 * Add support for "regular" device ioctls here. 1778 */ 1779 1780 error = disk_ioctl(&rs->sc_dkdev, cmd, data, flag, l); 1781 if (error != EPASSTHROUGH) 1782 return (error); 1783 1784 switch (cmd) { 1785 case DIOCGDINFO: 1786 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label); 1787 break; 1788 #ifdef __HAVE_OLD_DISKLABEL 1789 case ODIOCGDINFO: 1790 newlabel = *(rs->sc_dkdev.dk_label); 1791 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1792 return ENOTTY; 1793 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1794 break; 1795 #endif 1796 1797 case DIOCGPART: 1798 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label; 1799 ((struct partinfo *) data)->part = 1800 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 1801 break; 1802 1803 case DIOCWDINFO: 1804 case DIOCSDINFO: 1805 #ifdef __HAVE_OLD_DISKLABEL 1806 case ODIOCWDINFO: 1807 case ODIOCSDINFO: 1808 #endif 1809 { 1810 struct disklabel *lp; 1811 #ifdef __HAVE_OLD_DISKLABEL 1812 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1813 memset(&newlabel, 0, sizeof newlabel); 1814 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1815 lp = &newlabel; 1816 } else 1817 #endif 1818 lp = (struct disklabel *)data; 1819 1820 if ((error = raidlock(rs)) != 0) 1821 return (error); 1822 1823 rs->sc_flags |= RAIDF_LABELLING; 1824 1825 error = setdisklabel(rs->sc_dkdev.dk_label, 1826 lp, 0, rs->sc_dkdev.dk_cpulabel); 1827 if (error == 0) { 1828 if (cmd == DIOCWDINFO 1829 #ifdef __HAVE_OLD_DISKLABEL 1830 || cmd == ODIOCWDINFO 1831 #endif 1832 ) 1833 error = writedisklabel(RAIDLABELDEV(dev), 1834 raidstrategy, rs->sc_dkdev.dk_label, 1835 rs->sc_dkdev.dk_cpulabel); 1836 } 1837 rs->sc_flags &= ~RAIDF_LABELLING; 1838 1839 raidunlock(rs); 1840 1841 if (error) 1842 return (error); 1843 break; 1844 } 1845 1846 case DIOCWLABEL: 1847 if (*(int *) data != 0) 1848 rs->sc_flags |= RAIDF_WLABEL; 1849 else 1850 rs->sc_flags &= ~RAIDF_WLABEL; 1851 break; 1852 1853 case DIOCGDEFLABEL: 1854 raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data); 1855 break; 1856 1857 #ifdef __HAVE_OLD_DISKLABEL 1858 case ODIOCGDEFLABEL: 1859 raidgetdefaultlabel(raidPtr, rs, &newlabel); 1860 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1861 return ENOTTY; 1862 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1863 break; 1864 #endif 1865 1866 case DIOCAWEDGE: 1867 case DIOCDWEDGE: 1868 dkw = (void *)data; 1869 1870 /* If the ioctl happens here, the parent is us. */ 1871 (void)strcpy(dkw->dkw_parent, rs->sc_xname); 1872 return cmd == DIOCAWEDGE ? dkwedge_add(dkw) : dkwedge_del(dkw); 1873 1874 case DIOCLWEDGES: 1875 return dkwedge_list(&rs->sc_dkdev, 1876 (struct dkwedge_list *)data, l); 1877 case DIOCCACHESYNC: 1878 return rf_sync_component_caches(raidPtr); 1879 default: 1880 retcode = ENOTTY; 1881 } 1882 return (retcode); 1883 1884 } 1885 1886 1887 /* raidinit -- complete the rest of the initialization for the 1888 RAIDframe device. */ 1889 1890 1891 static void 1892 raidinit(RF_Raid_t *raidPtr) 1893 { 1894 cfdata_t cf; 1895 struct raid_softc *rs; 1896 int unit; 1897 1898 unit = raidPtr->raidid; 1899 1900 rs = &raid_softc[unit]; 1901 1902 /* XXX should check return code first... */ 1903 rs->sc_flags |= RAIDF_INITED; 1904 1905 /* XXX doesn't check bounds. */ 1906 snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit); 1907 1908 /* attach the pseudo device */ 1909 cf = malloc(sizeof(*cf), M_RAIDFRAME, M_WAITOK); 1910 cf->cf_name = raid_cd.cd_name; 1911 cf->cf_atname = raid_cd.cd_name; 1912 cf->cf_unit = unit; 1913 cf->cf_fstate = FSTATE_STAR; 1914 1915 rs->sc_dev = config_attach_pseudo(cf); 1916 1917 if (rs->sc_dev == NULL) { 1918 printf("raid%d: config_attach_pseudo failed\n", 1919 raidPtr->raidid); 1920 rs->sc_flags &= ~RAIDF_INITED; 1921 free(cf, M_RAIDFRAME); 1922 return; 1923 } 1924 1925 /* disk_attach actually creates space for the CPU disklabel, among 1926 * other things, so it's critical to call this *BEFORE* we try putzing 1927 * with disklabels. */ 1928 1929 disk_init(&rs->sc_dkdev, rs->sc_xname, &rf_dkdriver); 1930 disk_attach(&rs->sc_dkdev); 1931 1932 /* XXX There may be a weird interaction here between this, and 1933 * protectedSectors, as used in RAIDframe. */ 1934 1935 rs->sc_size = raidPtr->totalSectors; 1936 1937 dkwedge_discover(&rs->sc_dkdev); 1938 1939 rf_set_properties(rs, raidPtr); 1940 1941 } 1942 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0) 1943 /* wake up the daemon & tell it to get us a spare table 1944 * XXX 1945 * the entries in the queues should be tagged with the raidPtr 1946 * so that in the extremely rare case that two recons happen at once, 1947 * we know for which device were requesting a spare table 1948 * XXX 1949 * 1950 * XXX This code is not currently used. GO 1951 */ 1952 int 1953 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req) 1954 { 1955 int retcode; 1956 1957 RF_LOCK_MUTEX(rf_sparet_wait_mutex); 1958 req->next = rf_sparet_wait_queue; 1959 rf_sparet_wait_queue = req; 1960 wakeup(&rf_sparet_wait_queue); 1961 1962 /* mpsleep unlocks the mutex */ 1963 while (!rf_sparet_resp_queue) { 1964 tsleep(&rf_sparet_resp_queue, PRIBIO, 1965 "raidframe getsparetable", 0); 1966 } 1967 req = rf_sparet_resp_queue; 1968 rf_sparet_resp_queue = req->next; 1969 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex); 1970 1971 retcode = req->fcol; 1972 RF_Free(req, sizeof(*req)); /* this is not the same req as we 1973 * alloc'd */ 1974 return (retcode); 1975 } 1976 #endif 1977 1978 /* a wrapper around rf_DoAccess that extracts appropriate info from the 1979 * bp & passes it down. 1980 * any calls originating in the kernel must use non-blocking I/O 1981 * do some extra sanity checking to return "appropriate" error values for 1982 * certain conditions (to make some standard utilities work) 1983 * 1984 * Formerly known as: rf_DoAccessKernel 1985 */ 1986 void 1987 raidstart(RF_Raid_t *raidPtr) 1988 { 1989 RF_SectorCount_t num_blocks, pb, sum; 1990 RF_RaidAddr_t raid_addr; 1991 struct partition *pp; 1992 daddr_t blocknum; 1993 int unit; 1994 struct raid_softc *rs; 1995 int do_async; 1996 struct buf *bp; 1997 int rc; 1998 1999 unit = raidPtr->raidid; 2000 rs = &raid_softc[unit]; 2001 2002 /* quick check to see if anything has died recently */ 2003 RF_LOCK_MUTEX(raidPtr->mutex); 2004 if (raidPtr->numNewFailures > 0) { 2005 RF_UNLOCK_MUTEX(raidPtr->mutex); 2006 rf_update_component_labels(raidPtr, 2007 RF_NORMAL_COMPONENT_UPDATE); 2008 RF_LOCK_MUTEX(raidPtr->mutex); 2009 raidPtr->numNewFailures--; 2010 } 2011 2012 /* Check to see if we're at the limit... */ 2013 while (raidPtr->openings > 0) { 2014 RF_UNLOCK_MUTEX(raidPtr->mutex); 2015 2016 /* get the next item, if any, from the queue */ 2017 if ((bp = bufq_get(rs->buf_queue)) == NULL) { 2018 /* nothing more to do */ 2019 return; 2020 } 2021 2022 /* Ok, for the bp we have here, bp->b_blkno is relative to the 2023 * partition.. Need to make it absolute to the underlying 2024 * device.. */ 2025 2026 blocknum = bp->b_blkno; 2027 if (DISKPART(bp->b_dev) != RAW_PART) { 2028 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 2029 blocknum += pp->p_offset; 2030 } 2031 2032 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno, 2033 (int) blocknum)); 2034 2035 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount)); 2036 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid)); 2037 2038 /* *THIS* is where we adjust what block we're going to... 2039 * but DO NOT TOUCH bp->b_blkno!!! */ 2040 raid_addr = blocknum; 2041 2042 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector; 2043 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0; 2044 sum = raid_addr + num_blocks + pb; 2045 if (1 || rf_debugKernelAccess) { 2046 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n", 2047 (int) raid_addr, (int) sum, (int) num_blocks, 2048 (int) pb, (int) bp->b_resid)); 2049 } 2050 if ((sum > raidPtr->totalSectors) || (sum < raid_addr) 2051 || (sum < num_blocks) || (sum < pb)) { 2052 bp->b_error = ENOSPC; 2053 bp->b_resid = bp->b_bcount; 2054 biodone(bp); 2055 RF_LOCK_MUTEX(raidPtr->mutex); 2056 continue; 2057 } 2058 /* 2059 * XXX rf_DoAccess() should do this, not just DoAccessKernel() 2060 */ 2061 2062 if (bp->b_bcount & raidPtr->sectorMask) { 2063 bp->b_error = EINVAL; 2064 bp->b_resid = bp->b_bcount; 2065 biodone(bp); 2066 RF_LOCK_MUTEX(raidPtr->mutex); 2067 continue; 2068 2069 } 2070 db1_printf(("Calling DoAccess..\n")); 2071 2072 2073 RF_LOCK_MUTEX(raidPtr->mutex); 2074 raidPtr->openings--; 2075 RF_UNLOCK_MUTEX(raidPtr->mutex); 2076 2077 /* 2078 * Everything is async. 2079 */ 2080 do_async = 1; 2081 2082 disk_busy(&rs->sc_dkdev); 2083 2084 /* XXX we're still at splbio() here... do we *really* 2085 need to be? */ 2086 2087 /* don't ever condition on bp->b_flags & B_WRITE. 2088 * always condition on B_READ instead */ 2089 2090 rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ? 2091 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE, 2092 do_async, raid_addr, num_blocks, 2093 bp->b_data, bp, RF_DAG_NONBLOCKING_IO); 2094 2095 if (rc) { 2096 bp->b_error = rc; 2097 bp->b_resid = bp->b_bcount; 2098 biodone(bp); 2099 /* continue loop */ 2100 } 2101 2102 RF_LOCK_MUTEX(raidPtr->mutex); 2103 } 2104 RF_UNLOCK_MUTEX(raidPtr->mutex); 2105 } 2106 2107 2108 2109 2110 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */ 2111 2112 int 2113 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req) 2114 { 2115 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE; 2116 struct buf *bp; 2117 2118 req->queue = queue; 2119 bp = req->bp; 2120 2121 switch (req->type) { 2122 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */ 2123 /* XXX need to do something extra here.. */ 2124 /* I'm leaving this in, as I've never actually seen it used, 2125 * and I'd like folks to report it... GO */ 2126 printf(("WAKEUP CALLED\n")); 2127 queue->numOutstanding++; 2128 2129 bp->b_flags = 0; 2130 bp->b_private = req; 2131 2132 KernelWakeupFunc(bp); 2133 break; 2134 2135 case RF_IO_TYPE_READ: 2136 case RF_IO_TYPE_WRITE: 2137 #if RF_ACC_TRACE > 0 2138 if (req->tracerec) { 2139 RF_ETIMER_START(req->tracerec->timer); 2140 } 2141 #endif 2142 InitBP(bp, queue->rf_cinfo->ci_vp, 2143 op, queue->rf_cinfo->ci_dev, 2144 req->sectorOffset, req->numSector, 2145 req->buf, KernelWakeupFunc, (void *) req, 2146 queue->raidPtr->logBytesPerSector, req->b_proc); 2147 2148 if (rf_debugKernelAccess) { 2149 db1_printf(("dispatch: bp->b_blkno = %ld\n", 2150 (long) bp->b_blkno)); 2151 } 2152 queue->numOutstanding++; 2153 queue->last_deq_sector = req->sectorOffset; 2154 /* acc wouldn't have been let in if there were any pending 2155 * reqs at any other priority */ 2156 queue->curPriority = req->priority; 2157 2158 db1_printf(("Going for %c to unit %d col %d\n", 2159 req->type, queue->raidPtr->raidid, 2160 queue->col)); 2161 db1_printf(("sector %d count %d (%d bytes) %d\n", 2162 (int) req->sectorOffset, (int) req->numSector, 2163 (int) (req->numSector << 2164 queue->raidPtr->logBytesPerSector), 2165 (int) queue->raidPtr->logBytesPerSector)); 2166 2167 /* 2168 * XXX: drop lock here since this can block at 2169 * least with backing SCSI devices. Retake it 2170 * to minimize fuss with calling interfaces. 2171 */ 2172 2173 RF_UNLOCK_QUEUE_MUTEX(queue, "unusedparam"); 2174 bdev_strategy(bp); 2175 RF_LOCK_QUEUE_MUTEX(queue, "unusedparam"); 2176 break; 2177 2178 default: 2179 panic("bad req->type in rf_DispatchKernelIO"); 2180 } 2181 db1_printf(("Exiting from DispatchKernelIO\n")); 2182 2183 return (0); 2184 } 2185 /* this is the callback function associated with a I/O invoked from 2186 kernel code. 2187 */ 2188 static void 2189 KernelWakeupFunc(struct buf *bp) 2190 { 2191 RF_DiskQueueData_t *req = NULL; 2192 RF_DiskQueue_t *queue; 2193 int s; 2194 2195 s = splbio(); 2196 db1_printf(("recovering the request queue:\n")); 2197 req = bp->b_private; 2198 2199 queue = (RF_DiskQueue_t *) req->queue; 2200 2201 #if RF_ACC_TRACE > 0 2202 if (req->tracerec) { 2203 RF_ETIMER_STOP(req->tracerec->timer); 2204 RF_ETIMER_EVAL(req->tracerec->timer); 2205 RF_LOCK_MUTEX(rf_tracing_mutex); 2206 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2207 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer); 2208 req->tracerec->num_phys_ios++; 2209 RF_UNLOCK_MUTEX(rf_tracing_mutex); 2210 } 2211 #endif 2212 2213 /* XXX Ok, let's get aggressive... If b_error is set, let's go 2214 * ballistic, and mark the component as hosed... */ 2215 2216 if (bp->b_error != 0) { 2217 /* Mark the disk as dead */ 2218 /* but only mark it once... */ 2219 /* and only if it wouldn't leave this RAID set 2220 completely broken */ 2221 if (((queue->raidPtr->Disks[queue->col].status == 2222 rf_ds_optimal) || 2223 (queue->raidPtr->Disks[queue->col].status == 2224 rf_ds_used_spare)) && 2225 (queue->raidPtr->numFailures < 2226 queue->raidPtr->Layout.map->faultsTolerated)) { 2227 printf("raid%d: IO Error. Marking %s as failed.\n", 2228 queue->raidPtr->raidid, 2229 queue->raidPtr->Disks[queue->col].devname); 2230 queue->raidPtr->Disks[queue->col].status = 2231 rf_ds_failed; 2232 queue->raidPtr->status = rf_rs_degraded; 2233 queue->raidPtr->numFailures++; 2234 queue->raidPtr->numNewFailures++; 2235 } else { /* Disk is already dead... */ 2236 /* printf("Disk already marked as dead!\n"); */ 2237 } 2238 2239 } 2240 2241 /* Fill in the error value */ 2242 2243 req->error = bp->b_error; 2244 2245 simple_lock(&queue->raidPtr->iodone_lock); 2246 2247 /* Drop this one on the "finished" queue... */ 2248 TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries); 2249 2250 /* Let the raidio thread know there is work to be done. */ 2251 wakeup(&(queue->raidPtr->iodone)); 2252 2253 simple_unlock(&queue->raidPtr->iodone_lock); 2254 2255 splx(s); 2256 } 2257 2258 2259 2260 /* 2261 * initialize a buf structure for doing an I/O in the kernel. 2262 */ 2263 static void 2264 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev, 2265 RF_SectorNum_t startSect, RF_SectorCount_t numSect, void *bf, 2266 void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector, 2267 struct proc *b_proc) 2268 { 2269 /* bp->b_flags = B_PHYS | rw_flag; */ 2270 bp->b_flags = rw_flag; /* XXX need B_PHYS here too??? */ 2271 bp->b_oflags = 0; 2272 bp->b_cflags = 0; 2273 bp->b_bcount = numSect << logBytesPerSector; 2274 bp->b_bufsize = bp->b_bcount; 2275 bp->b_error = 0; 2276 bp->b_dev = dev; 2277 bp->b_data = bf; 2278 bp->b_blkno = startSect; 2279 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */ 2280 if (bp->b_bcount == 0) { 2281 panic("bp->b_bcount is zero in InitBP!!"); 2282 } 2283 bp->b_proc = b_proc; 2284 bp->b_iodone = cbFunc; 2285 bp->b_private = cbArg; 2286 } 2287 2288 static void 2289 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs, 2290 struct disklabel *lp) 2291 { 2292 memset(lp, 0, sizeof(*lp)); 2293 2294 /* fabricate a label... */ 2295 lp->d_secperunit = raidPtr->totalSectors; 2296 lp->d_secsize = raidPtr->bytesPerSector; 2297 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe; 2298 lp->d_ntracks = 4 * raidPtr->numCol; 2299 lp->d_ncylinders = raidPtr->totalSectors / 2300 (lp->d_nsectors * lp->d_ntracks); 2301 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 2302 2303 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename)); 2304 lp->d_type = DTYPE_RAID; 2305 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 2306 lp->d_rpm = 3600; 2307 lp->d_interleave = 1; 2308 lp->d_flags = 0; 2309 2310 lp->d_partitions[RAW_PART].p_offset = 0; 2311 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors; 2312 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 2313 lp->d_npartitions = RAW_PART + 1; 2314 2315 lp->d_magic = DISKMAGIC; 2316 lp->d_magic2 = DISKMAGIC; 2317 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label); 2318 2319 } 2320 /* 2321 * Read the disklabel from the raid device. If one is not present, fake one 2322 * up. 2323 */ 2324 static void 2325 raidgetdisklabel(dev_t dev) 2326 { 2327 int unit = raidunit(dev); 2328 struct raid_softc *rs = &raid_softc[unit]; 2329 const char *errstring; 2330 struct disklabel *lp = rs->sc_dkdev.dk_label; 2331 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel; 2332 RF_Raid_t *raidPtr; 2333 2334 db1_printf(("Getting the disklabel...\n")); 2335 2336 memset(clp, 0, sizeof(*clp)); 2337 2338 raidPtr = raidPtrs[unit]; 2339 2340 raidgetdefaultlabel(raidPtr, rs, lp); 2341 2342 /* 2343 * Call the generic disklabel extraction routine. 2344 */ 2345 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy, 2346 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel); 2347 if (errstring) 2348 raidmakedisklabel(rs); 2349 else { 2350 int i; 2351 struct partition *pp; 2352 2353 /* 2354 * Sanity check whether the found disklabel is valid. 2355 * 2356 * This is necessary since total size of the raid device 2357 * may vary when an interleave is changed even though exactly 2358 * same components are used, and old disklabel may used 2359 * if that is found. 2360 */ 2361 if (lp->d_secperunit != rs->sc_size) 2362 printf("raid%d: WARNING: %s: " 2363 "total sector size in disklabel (%" PRIu32 ") != " 2364 "the size of raid (%" PRIu64 ")\n", unit, rs->sc_xname, 2365 lp->d_secperunit, rs->sc_size); 2366 for (i = 0; i < lp->d_npartitions; i++) { 2367 pp = &lp->d_partitions[i]; 2368 if (pp->p_offset + pp->p_size > rs->sc_size) 2369 printf("raid%d: WARNING: %s: end of partition `%c' " 2370 "exceeds the size of raid (%" PRIu64 ")\n", 2371 unit, rs->sc_xname, 'a' + i, rs->sc_size); 2372 } 2373 } 2374 2375 } 2376 /* 2377 * Take care of things one might want to take care of in the event 2378 * that a disklabel isn't present. 2379 */ 2380 static void 2381 raidmakedisklabel(struct raid_softc *rs) 2382 { 2383 struct disklabel *lp = rs->sc_dkdev.dk_label; 2384 db1_printf(("Making a label..\n")); 2385 2386 /* 2387 * For historical reasons, if there's no disklabel present 2388 * the raw partition must be marked FS_BSDFFS. 2389 */ 2390 2391 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 2392 2393 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 2394 2395 lp->d_checksum = dkcksum(lp); 2396 } 2397 /* 2398 * Wait interruptibly for an exclusive lock. 2399 * 2400 * XXX 2401 * Several drivers do this; it should be abstracted and made MP-safe. 2402 * (Hmm... where have we seen this warning before :-> GO ) 2403 */ 2404 static int 2405 raidlock(struct raid_softc *rs) 2406 { 2407 int error; 2408 2409 while ((rs->sc_flags & RAIDF_LOCKED) != 0) { 2410 rs->sc_flags |= RAIDF_WANTED; 2411 if ((error = 2412 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0) 2413 return (error); 2414 } 2415 rs->sc_flags |= RAIDF_LOCKED; 2416 return (0); 2417 } 2418 /* 2419 * Unlock and wake up any waiters. 2420 */ 2421 static void 2422 raidunlock(struct raid_softc *rs) 2423 { 2424 2425 rs->sc_flags &= ~RAIDF_LOCKED; 2426 if ((rs->sc_flags & RAIDF_WANTED) != 0) { 2427 rs->sc_flags &= ~RAIDF_WANTED; 2428 wakeup(rs); 2429 } 2430 } 2431 2432 2433 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */ 2434 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */ 2435 #define RF_PARITY_MAP_OFFSET \ 2436 (RF_COMPONENT_INFO_OFFSET + RF_COMPONENT_INFO_SIZE) 2437 #define RF_PARITY_MAP_SIZE RF_PARITYMAP_NBYTE 2438 2439 int 2440 raidmarkclean(RF_Raid_t *raidPtr, RF_RowCol_t col) 2441 { 2442 RF_ComponentLabel_t *clabel; 2443 2444 clabel = raidget_component_label(raidPtr, col); 2445 clabel->clean = RF_RAID_CLEAN; 2446 raidflush_component_label(raidPtr, col); 2447 return(0); 2448 } 2449 2450 2451 int 2452 raidmarkdirty(RF_Raid_t *raidPtr, RF_RowCol_t col) 2453 { 2454 RF_ComponentLabel_t *clabel; 2455 2456 clabel = raidget_component_label(raidPtr, col); 2457 clabel->clean = RF_RAID_DIRTY; 2458 raidflush_component_label(raidPtr, col); 2459 return(0); 2460 } 2461 2462 int 2463 raidfetch_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2464 { 2465 return raidread_component_label(raidPtr->Disks[col].dev, 2466 raidPtr->raid_cinfo[col].ci_vp, 2467 &raidPtr->raid_cinfo[col].ci_label); 2468 } 2469 2470 RF_ComponentLabel_t * 2471 raidget_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2472 { 2473 return &raidPtr->raid_cinfo[col].ci_label; 2474 } 2475 2476 int 2477 raidflush_component_label(RF_Raid_t *raidPtr, RF_RowCol_t col) 2478 { 2479 RF_ComponentLabel_t *label; 2480 2481 label = &raidPtr->raid_cinfo[col].ci_label; 2482 label->mod_counter = raidPtr->mod_counter; 2483 #ifndef RF_NO_PARITY_MAP 2484 label->parity_map_modcount = label->mod_counter; 2485 #endif 2486 return raidwrite_component_label(raidPtr->Disks[col].dev, 2487 raidPtr->raid_cinfo[col].ci_vp, label); 2488 } 2489 2490 2491 static int 2492 raidread_component_label(dev_t dev, struct vnode *b_vp, 2493 RF_ComponentLabel_t *clabel) 2494 { 2495 return raidread_component_area(dev, b_vp, clabel, 2496 sizeof(RF_ComponentLabel_t), 2497 RF_COMPONENT_INFO_OFFSET, RF_COMPONENT_INFO_SIZE); 2498 } 2499 2500 /* ARGSUSED */ 2501 static int 2502 raidread_component_area(dev_t dev, struct vnode *b_vp, void *data, 2503 size_t msize, daddr_t offset, daddr_t dsize) 2504 { 2505 struct buf *bp; 2506 const struct bdevsw *bdev; 2507 int error; 2508 2509 /* XXX should probably ensure that we don't try to do this if 2510 someone has changed rf_protected_sectors. */ 2511 2512 if (b_vp == NULL) { 2513 /* For whatever reason, this component is not valid. 2514 Don't try to read a component label from it. */ 2515 return(EINVAL); 2516 } 2517 2518 /* get a block of the appropriate size... */ 2519 bp = geteblk((int)dsize); 2520 bp->b_dev = dev; 2521 2522 /* get our ducks in a row for the read */ 2523 bp->b_blkno = offset / DEV_BSIZE; 2524 bp->b_bcount = dsize; 2525 bp->b_flags |= B_READ; 2526 bp->b_resid = dsize; 2527 2528 bdev = bdevsw_lookup(bp->b_dev); 2529 if (bdev == NULL) 2530 return (ENXIO); 2531 (*bdev->d_strategy)(bp); 2532 2533 error = biowait(bp); 2534 2535 if (!error) { 2536 memcpy(data, bp->b_data, msize); 2537 } 2538 2539 brelse(bp, 0); 2540 return(error); 2541 } 2542 2543 2544 static int 2545 raidwrite_component_label(dev_t dev, struct vnode *b_vp, 2546 RF_ComponentLabel_t *clabel) 2547 { 2548 return raidwrite_component_area(dev, b_vp, clabel, 2549 sizeof(RF_ComponentLabel_t), 2550 RF_COMPONENT_INFO_OFFSET, RF_COMPONENT_INFO_SIZE, 0); 2551 } 2552 2553 /* ARGSUSED */ 2554 static int 2555 raidwrite_component_area(dev_t dev, struct vnode *b_vp, void *data, 2556 size_t msize, daddr_t offset, daddr_t dsize, int asyncp) 2557 { 2558 struct buf *bp; 2559 const struct bdevsw *bdev; 2560 int error; 2561 2562 /* get a block of the appropriate size... */ 2563 bp = geteblk((int)dsize); 2564 bp->b_dev = dev; 2565 2566 /* get our ducks in a row for the write */ 2567 bp->b_blkno = offset / DEV_BSIZE; 2568 bp->b_bcount = dsize; 2569 bp->b_flags |= B_WRITE | (asyncp ? B_ASYNC : 0); 2570 bp->b_resid = dsize; 2571 2572 memset(bp->b_data, 0, dsize); 2573 memcpy(bp->b_data, data, msize); 2574 2575 bdev = bdevsw_lookup(bp->b_dev); 2576 if (bdev == NULL) 2577 return (ENXIO); 2578 (*bdev->d_strategy)(bp); 2579 if (asyncp) 2580 return 0; 2581 error = biowait(bp); 2582 brelse(bp, 0); 2583 if (error) { 2584 #if 1 2585 printf("Failed to write RAID component info!\n"); 2586 #endif 2587 } 2588 2589 return(error); 2590 } 2591 2592 void 2593 rf_paritymap_kern_write(RF_Raid_t *raidPtr, struct rf_paritymap_ondisk *map) 2594 { 2595 int c; 2596 2597 for (c = 0; c < raidPtr->numCol; c++) { 2598 /* Skip dead disks. */ 2599 if (RF_DEAD_DISK(raidPtr->Disks[c].status)) 2600 continue; 2601 /* XXXjld: what if an error occurs here? */ 2602 raidwrite_component_area(raidPtr->Disks[c].dev, 2603 raidPtr->raid_cinfo[c].ci_vp, map, 2604 RF_PARITYMAP_NBYTE, 2605 RF_PARITY_MAP_OFFSET, RF_PARITY_MAP_SIZE, 0); 2606 } 2607 } 2608 2609 void 2610 rf_paritymap_kern_read(RF_Raid_t *raidPtr, struct rf_paritymap_ondisk *map) 2611 { 2612 struct rf_paritymap_ondisk tmp; 2613 int c; 2614 2615 for (c = 0; c < raidPtr->numCol; c++) { 2616 /* Skip dead disks. */ 2617 if (RF_DEAD_DISK(raidPtr->Disks[c].status)) 2618 continue; 2619 raidread_component_area(raidPtr->Disks[c].dev, 2620 raidPtr->raid_cinfo[c].ci_vp, &tmp, 2621 RF_PARITYMAP_NBYTE, 2622 RF_PARITY_MAP_OFFSET, RF_PARITY_MAP_SIZE); 2623 if (c == 0) { 2624 memcpy(map, &tmp, sizeof(*map)); 2625 } else { 2626 rf_paritymap_merge(map, &tmp); 2627 } 2628 } 2629 } 2630 2631 void 2632 rf_markalldirty(RF_Raid_t *raidPtr) 2633 { 2634 RF_ComponentLabel_t *clabel; 2635 int sparecol; 2636 int c; 2637 int j; 2638 int scol = -1; 2639 2640 raidPtr->mod_counter++; 2641 for (c = 0; c < raidPtr->numCol; c++) { 2642 /* we don't want to touch (at all) a disk that has 2643 failed */ 2644 if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) { 2645 clabel = raidget_component_label(raidPtr, c); 2646 if (clabel->status == rf_ds_spared) { 2647 /* XXX do something special... 2648 but whatever you do, don't 2649 try to access it!! */ 2650 } else { 2651 raidmarkdirty(raidPtr, c); 2652 } 2653 } 2654 } 2655 2656 for( c = 0; c < raidPtr->numSpare ; c++) { 2657 sparecol = raidPtr->numCol + c; 2658 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2659 /* 2660 2661 we claim this disk is "optimal" if it's 2662 rf_ds_used_spare, as that means it should be 2663 directly substitutable for the disk it replaced. 2664 We note that too... 2665 2666 */ 2667 2668 for(j=0;j<raidPtr->numCol;j++) { 2669 if (raidPtr->Disks[j].spareCol == sparecol) { 2670 scol = j; 2671 break; 2672 } 2673 } 2674 2675 clabel = raidget_component_label(raidPtr, sparecol); 2676 /* make sure status is noted */ 2677 2678 raid_init_component_label(raidPtr, clabel); 2679 2680 clabel->row = 0; 2681 clabel->column = scol; 2682 /* Note: we *don't* change status from rf_ds_used_spare 2683 to rf_ds_optimal */ 2684 /* clabel.status = rf_ds_optimal; */ 2685 2686 raidmarkdirty(raidPtr, sparecol); 2687 } 2688 } 2689 } 2690 2691 2692 void 2693 rf_update_component_labels(RF_Raid_t *raidPtr, int final) 2694 { 2695 RF_ComponentLabel_t *clabel; 2696 int sparecol; 2697 int c; 2698 int j; 2699 int scol; 2700 2701 scol = -1; 2702 2703 /* XXX should do extra checks to make sure things really are clean, 2704 rather than blindly setting the clean bit... */ 2705 2706 raidPtr->mod_counter++; 2707 2708 for (c = 0; c < raidPtr->numCol; c++) { 2709 if (raidPtr->Disks[c].status == rf_ds_optimal) { 2710 clabel = raidget_component_label(raidPtr, c); 2711 /* make sure status is noted */ 2712 clabel->status = rf_ds_optimal; 2713 2714 /* note what unit we are configured as */ 2715 clabel->last_unit = raidPtr->raidid; 2716 2717 raidflush_component_label(raidPtr, c); 2718 if (final == RF_FINAL_COMPONENT_UPDATE) { 2719 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2720 raidmarkclean(raidPtr, c); 2721 } 2722 } 2723 } 2724 /* else we don't touch it.. */ 2725 } 2726 2727 for( c = 0; c < raidPtr->numSpare ; c++) { 2728 sparecol = raidPtr->numCol + c; 2729 /* Need to ensure that the reconstruct actually completed! */ 2730 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 2731 /* 2732 2733 we claim this disk is "optimal" if it's 2734 rf_ds_used_spare, as that means it should be 2735 directly substitutable for the disk it replaced. 2736 We note that too... 2737 2738 */ 2739 2740 for(j=0;j<raidPtr->numCol;j++) { 2741 if (raidPtr->Disks[j].spareCol == sparecol) { 2742 scol = j; 2743 break; 2744 } 2745 } 2746 2747 /* XXX shouldn't *really* need this... */ 2748 clabel = raidget_component_label(raidPtr, sparecol); 2749 /* make sure status is noted */ 2750 2751 raid_init_component_label(raidPtr, clabel); 2752 2753 clabel->column = scol; 2754 clabel->status = rf_ds_optimal; 2755 clabel->last_unit = raidPtr->raidid; 2756 2757 raidflush_component_label(raidPtr, sparecol); 2758 if (final == RF_FINAL_COMPONENT_UPDATE) { 2759 if (raidPtr->parity_good == RF_RAID_CLEAN) { 2760 raidmarkclean(raidPtr, sparecol); 2761 } 2762 } 2763 } 2764 } 2765 } 2766 2767 void 2768 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured) 2769 { 2770 2771 if (vp != NULL) { 2772 if (auto_configured == 1) { 2773 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2774 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2775 vput(vp); 2776 2777 } else { 2778 (void) vn_close(vp, FREAD | FWRITE, curlwp->l_cred); 2779 } 2780 } 2781 } 2782 2783 2784 void 2785 rf_UnconfigureVnodes(RF_Raid_t *raidPtr) 2786 { 2787 int r,c; 2788 struct vnode *vp; 2789 int acd; 2790 2791 2792 /* We take this opportunity to close the vnodes like we should.. */ 2793 2794 for (c = 0; c < raidPtr->numCol; c++) { 2795 vp = raidPtr->raid_cinfo[c].ci_vp; 2796 acd = raidPtr->Disks[c].auto_configured; 2797 rf_close_component(raidPtr, vp, acd); 2798 raidPtr->raid_cinfo[c].ci_vp = NULL; 2799 raidPtr->Disks[c].auto_configured = 0; 2800 } 2801 2802 for (r = 0; r < raidPtr->numSpare; r++) { 2803 vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp; 2804 acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured; 2805 rf_close_component(raidPtr, vp, acd); 2806 raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL; 2807 raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0; 2808 } 2809 } 2810 2811 2812 void 2813 rf_ReconThread(struct rf_recon_req *req) 2814 { 2815 int s; 2816 RF_Raid_t *raidPtr; 2817 2818 s = splbio(); 2819 raidPtr = (RF_Raid_t *) req->raidPtr; 2820 raidPtr->recon_in_progress = 1; 2821 2822 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col, 2823 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0)); 2824 2825 RF_Free(req, sizeof(*req)); 2826 2827 raidPtr->recon_in_progress = 0; 2828 splx(s); 2829 2830 /* That's all... */ 2831 kthread_exit(0); /* does not return */ 2832 } 2833 2834 void 2835 rf_RewriteParityThread(RF_Raid_t *raidPtr) 2836 { 2837 int retcode; 2838 int s; 2839 2840 raidPtr->parity_rewrite_stripes_done = 0; 2841 raidPtr->parity_rewrite_in_progress = 1; 2842 s = splbio(); 2843 retcode = rf_RewriteParity(raidPtr); 2844 splx(s); 2845 if (retcode) { 2846 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid); 2847 } else { 2848 /* set the clean bit! If we shutdown correctly, 2849 the clean bit on each component label will get 2850 set */ 2851 raidPtr->parity_good = RF_RAID_CLEAN; 2852 } 2853 raidPtr->parity_rewrite_in_progress = 0; 2854 2855 /* Anyone waiting for us to stop? If so, inform them... */ 2856 if (raidPtr->waitShutdown) { 2857 wakeup(&raidPtr->parity_rewrite_in_progress); 2858 } 2859 2860 /* That's all... */ 2861 kthread_exit(0); /* does not return */ 2862 } 2863 2864 2865 void 2866 rf_CopybackThread(RF_Raid_t *raidPtr) 2867 { 2868 int s; 2869 2870 raidPtr->copyback_in_progress = 1; 2871 s = splbio(); 2872 rf_CopybackReconstructedData(raidPtr); 2873 splx(s); 2874 raidPtr->copyback_in_progress = 0; 2875 2876 /* That's all... */ 2877 kthread_exit(0); /* does not return */ 2878 } 2879 2880 2881 void 2882 rf_ReconstructInPlaceThread(struct rf_recon_req *req) 2883 { 2884 int s; 2885 RF_Raid_t *raidPtr; 2886 2887 s = splbio(); 2888 raidPtr = req->raidPtr; 2889 raidPtr->recon_in_progress = 1; 2890 rf_ReconstructInPlace(raidPtr, req->col); 2891 RF_Free(req, sizeof(*req)); 2892 raidPtr->recon_in_progress = 0; 2893 splx(s); 2894 2895 /* That's all... */ 2896 kthread_exit(0); /* does not return */ 2897 } 2898 2899 static RF_AutoConfig_t * 2900 rf_get_component(RF_AutoConfig_t *ac_list, dev_t dev, struct vnode *vp, 2901 const char *cname, RF_SectorCount_t size) 2902 { 2903 int good_one = 0; 2904 RF_ComponentLabel_t *clabel; 2905 RF_AutoConfig_t *ac; 2906 2907 clabel = malloc(sizeof(RF_ComponentLabel_t), M_RAIDFRAME, M_NOWAIT); 2908 if (clabel == NULL) { 2909 oomem: 2910 while(ac_list) { 2911 ac = ac_list; 2912 if (ac->clabel) 2913 free(ac->clabel, M_RAIDFRAME); 2914 ac_list = ac_list->next; 2915 free(ac, M_RAIDFRAME); 2916 } 2917 printf("RAID auto config: out of memory!\n"); 2918 return NULL; /* XXX probably should panic? */ 2919 } 2920 2921 if (!raidread_component_label(dev, vp, clabel)) { 2922 /* Got the label. Does it look reasonable? */ 2923 if (rf_reasonable_label(clabel) && 2924 (clabel->partitionSize <= size)) { 2925 #ifdef DEBUG 2926 printf("Component on: %s: %llu\n", 2927 cname, (unsigned long long)size); 2928 rf_print_component_label(clabel); 2929 #endif 2930 /* if it's reasonable, add it, else ignore it. */ 2931 ac = malloc(sizeof(RF_AutoConfig_t), M_RAIDFRAME, 2932 M_NOWAIT); 2933 if (ac == NULL) { 2934 free(clabel, M_RAIDFRAME); 2935 goto oomem; 2936 } 2937 strlcpy(ac->devname, cname, sizeof(ac->devname)); 2938 ac->dev = dev; 2939 ac->vp = vp; 2940 ac->clabel = clabel; 2941 ac->next = ac_list; 2942 ac_list = ac; 2943 good_one = 1; 2944 } 2945 } 2946 if (!good_one) { 2947 /* cleanup */ 2948 free(clabel, M_RAIDFRAME); 2949 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2950 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 2951 vput(vp); 2952 } 2953 return ac_list; 2954 } 2955 2956 RF_AutoConfig_t * 2957 rf_find_raid_components(void) 2958 { 2959 struct vnode *vp; 2960 struct disklabel label; 2961 device_t dv; 2962 deviter_t di; 2963 dev_t dev; 2964 int bmajor, bminor, wedge; 2965 int error; 2966 int i; 2967 RF_AutoConfig_t *ac_list; 2968 2969 2970 /* initialize the AutoConfig list */ 2971 ac_list = NULL; 2972 2973 /* we begin by trolling through *all* the devices on the system */ 2974 2975 for (dv = deviter_first(&di, DEVITER_F_ROOT_FIRST); dv != NULL; 2976 dv = deviter_next(&di)) { 2977 2978 /* we are only interested in disks... */ 2979 if (device_class(dv) != DV_DISK) 2980 continue; 2981 2982 /* we don't care about floppies... */ 2983 if (device_is_a(dv, "fd")) { 2984 continue; 2985 } 2986 2987 /* we don't care about CD's... */ 2988 if (device_is_a(dv, "cd")) { 2989 continue; 2990 } 2991 2992 /* we don't care about md's... */ 2993 if (device_is_a(dv, "md")) { 2994 continue; 2995 } 2996 2997 /* hdfd is the Atari/Hades floppy driver */ 2998 if (device_is_a(dv, "hdfd")) { 2999 continue; 3000 } 3001 3002 /* fdisa is the Atari/Milan floppy driver */ 3003 if (device_is_a(dv, "fdisa")) { 3004 continue; 3005 } 3006 3007 /* need to find the device_name_to_block_device_major stuff */ 3008 bmajor = devsw_name2blk(device_xname(dv), NULL, 0); 3009 3010 /* get a vnode for the raw partition of this disk */ 3011 3012 wedge = device_is_a(dv, "dk"); 3013 bminor = minor(device_unit(dv)); 3014 dev = wedge ? makedev(bmajor, bminor) : 3015 MAKEDISKDEV(bmajor, bminor, RAW_PART); 3016 if (bdevvp(dev, &vp)) 3017 panic("RAID can't alloc vnode"); 3018 3019 error = VOP_OPEN(vp, FREAD, NOCRED); 3020 3021 if (error) { 3022 /* "Who cares." Continue looking 3023 for something that exists*/ 3024 vput(vp); 3025 continue; 3026 } 3027 3028 if (wedge) { 3029 struct dkwedge_info dkw; 3030 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, 3031 NOCRED); 3032 if (error) { 3033 printf("RAIDframe: can't get wedge info for " 3034 "dev %s (%d)\n", device_xname(dv), error); 3035 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3036 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3037 vput(vp); 3038 continue; 3039 } 3040 3041 if (strcmp(dkw.dkw_ptype, DKW_PTYPE_RAIDFRAME) != 0) { 3042 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3043 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3044 vput(vp); 3045 continue; 3046 } 3047 3048 ac_list = rf_get_component(ac_list, dev, vp, 3049 device_xname(dv), dkw.dkw_size); 3050 continue; 3051 } 3052 3053 /* Ok, the disk exists. Go get the disklabel. */ 3054 error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED); 3055 if (error) { 3056 /* 3057 * XXX can't happen - open() would 3058 * have errored out (or faked up one) 3059 */ 3060 if (error != ENOTTY) 3061 printf("RAIDframe: can't get label for dev " 3062 "%s (%d)\n", device_xname(dv), error); 3063 } 3064 3065 /* don't need this any more. We'll allocate it again 3066 a little later if we really do... */ 3067 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3068 VOP_CLOSE(vp, FREAD | FWRITE, NOCRED); 3069 vput(vp); 3070 3071 if (error) 3072 continue; 3073 3074 for (i = 0; i < label.d_npartitions; i++) { 3075 char cname[sizeof(ac_list->devname)]; 3076 3077 /* We only support partitions marked as RAID */ 3078 if (label.d_partitions[i].p_fstype != FS_RAID) 3079 continue; 3080 3081 dev = MAKEDISKDEV(bmajor, device_unit(dv), i); 3082 if (bdevvp(dev, &vp)) 3083 panic("RAID can't alloc vnode"); 3084 3085 error = VOP_OPEN(vp, FREAD, NOCRED); 3086 if (error) { 3087 /* Whatever... */ 3088 vput(vp); 3089 continue; 3090 } 3091 snprintf(cname, sizeof(cname), "%s%c", 3092 device_xname(dv), 'a' + i); 3093 ac_list = rf_get_component(ac_list, dev, vp, cname, 3094 label.d_partitions[i].p_size); 3095 } 3096 } 3097 deviter_release(&di); 3098 return ac_list; 3099 } 3100 3101 3102 static int 3103 rf_reasonable_label(RF_ComponentLabel_t *clabel) 3104 { 3105 3106 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) || 3107 (clabel->version==RF_COMPONENT_LABEL_VERSION)) && 3108 ((clabel->clean == RF_RAID_CLEAN) || 3109 (clabel->clean == RF_RAID_DIRTY)) && 3110 clabel->row >=0 && 3111 clabel->column >= 0 && 3112 clabel->num_rows > 0 && 3113 clabel->num_columns > 0 && 3114 clabel->row < clabel->num_rows && 3115 clabel->column < clabel->num_columns && 3116 clabel->blockSize > 0 && 3117 clabel->numBlocks > 0) { 3118 /* label looks reasonable enough... */ 3119 return(1); 3120 } 3121 return(0); 3122 } 3123 3124 3125 #ifdef DEBUG 3126 void 3127 rf_print_component_label(RF_ComponentLabel_t *clabel) 3128 { 3129 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n", 3130 clabel->row, clabel->column, 3131 clabel->num_rows, clabel->num_columns); 3132 printf(" Version: %d Serial Number: %d Mod Counter: %d\n", 3133 clabel->version, clabel->serial_number, 3134 clabel->mod_counter); 3135 printf(" Clean: %s Status: %d\n", 3136 clabel->clean ? "Yes" : "No", clabel->status); 3137 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n", 3138 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU); 3139 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n", 3140 (char) clabel->parityConfig, clabel->blockSize, 3141 clabel->numBlocks); 3142 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No"); 3143 printf(" Contains root partition: %s\n", 3144 clabel->root_partition ? "Yes" : "No"); 3145 printf(" Last configured as: raid%d\n", clabel->last_unit); 3146 #if 0 3147 printf(" Config order: %d\n", clabel->config_order); 3148 #endif 3149 3150 } 3151 #endif 3152 3153 RF_ConfigSet_t * 3154 rf_create_auto_sets(RF_AutoConfig_t *ac_list) 3155 { 3156 RF_AutoConfig_t *ac; 3157 RF_ConfigSet_t *config_sets; 3158 RF_ConfigSet_t *cset; 3159 RF_AutoConfig_t *ac_next; 3160 3161 3162 config_sets = NULL; 3163 3164 /* Go through the AutoConfig list, and figure out which components 3165 belong to what sets. */ 3166 ac = ac_list; 3167 while(ac!=NULL) { 3168 /* we're going to putz with ac->next, so save it here 3169 for use at the end of the loop */ 3170 ac_next = ac->next; 3171 3172 if (config_sets == NULL) { 3173 /* will need at least this one... */ 3174 config_sets = (RF_ConfigSet_t *) 3175 malloc(sizeof(RF_ConfigSet_t), 3176 M_RAIDFRAME, M_NOWAIT); 3177 if (config_sets == NULL) { 3178 panic("rf_create_auto_sets: No memory!"); 3179 } 3180 /* this one is easy :) */ 3181 config_sets->ac = ac; 3182 config_sets->next = NULL; 3183 config_sets->rootable = 0; 3184 ac->next = NULL; 3185 } else { 3186 /* which set does this component fit into? */ 3187 cset = config_sets; 3188 while(cset!=NULL) { 3189 if (rf_does_it_fit(cset, ac)) { 3190 /* looks like it matches... */ 3191 ac->next = cset->ac; 3192 cset->ac = ac; 3193 break; 3194 } 3195 cset = cset->next; 3196 } 3197 if (cset==NULL) { 3198 /* didn't find a match above... new set..*/ 3199 cset = (RF_ConfigSet_t *) 3200 malloc(sizeof(RF_ConfigSet_t), 3201 M_RAIDFRAME, M_NOWAIT); 3202 if (cset == NULL) { 3203 panic("rf_create_auto_sets: No memory!"); 3204 } 3205 cset->ac = ac; 3206 ac->next = NULL; 3207 cset->next = config_sets; 3208 cset->rootable = 0; 3209 config_sets = cset; 3210 } 3211 } 3212 ac = ac_next; 3213 } 3214 3215 3216 return(config_sets); 3217 } 3218 3219 static int 3220 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac) 3221 { 3222 RF_ComponentLabel_t *clabel1, *clabel2; 3223 3224 /* If this one matches the *first* one in the set, that's good 3225 enough, since the other members of the set would have been 3226 through here too... */ 3227 /* note that we are not checking partitionSize here.. 3228 3229 Note that we are also not checking the mod_counters here. 3230 If everything else matches execpt the mod_counter, that's 3231 good enough for this test. We will deal with the mod_counters 3232 a little later in the autoconfiguration process. 3233 3234 (clabel1->mod_counter == clabel2->mod_counter) && 3235 3236 The reason we don't check for this is that failed disks 3237 will have lower modification counts. If those disks are 3238 not added to the set they used to belong to, then they will 3239 form their own set, which may result in 2 different sets, 3240 for example, competing to be configured at raid0, and 3241 perhaps competing to be the root filesystem set. If the 3242 wrong ones get configured, or both attempt to become /, 3243 weird behaviour and or serious lossage will occur. Thus we 3244 need to bring them into the fold here, and kick them out at 3245 a later point. 3246 3247 */ 3248 3249 clabel1 = cset->ac->clabel; 3250 clabel2 = ac->clabel; 3251 if ((clabel1->version == clabel2->version) && 3252 (clabel1->serial_number == clabel2->serial_number) && 3253 (clabel1->num_rows == clabel2->num_rows) && 3254 (clabel1->num_columns == clabel2->num_columns) && 3255 (clabel1->sectPerSU == clabel2->sectPerSU) && 3256 (clabel1->SUsPerPU == clabel2->SUsPerPU) && 3257 (clabel1->SUsPerRU == clabel2->SUsPerRU) && 3258 (clabel1->parityConfig == clabel2->parityConfig) && 3259 (clabel1->maxOutstanding == clabel2->maxOutstanding) && 3260 (clabel1->blockSize == clabel2->blockSize) && 3261 (clabel1->numBlocks == clabel2->numBlocks) && 3262 (clabel1->autoconfigure == clabel2->autoconfigure) && 3263 (clabel1->root_partition == clabel2->root_partition) && 3264 (clabel1->last_unit == clabel2->last_unit) && 3265 (clabel1->config_order == clabel2->config_order)) { 3266 /* if it get's here, it almost *has* to be a match */ 3267 } else { 3268 /* it's not consistent with somebody in the set.. 3269 punt */ 3270 return(0); 3271 } 3272 /* all was fine.. it must fit... */ 3273 return(1); 3274 } 3275 3276 int 3277 rf_have_enough_components(RF_ConfigSet_t *cset) 3278 { 3279 RF_AutoConfig_t *ac; 3280 RF_AutoConfig_t *auto_config; 3281 RF_ComponentLabel_t *clabel; 3282 int c; 3283 int num_cols; 3284 int num_missing; 3285 int mod_counter; 3286 int mod_counter_found; 3287 int even_pair_failed; 3288 char parity_type; 3289 3290 3291 /* check to see that we have enough 'live' components 3292 of this set. If so, we can configure it if necessary */ 3293 3294 num_cols = cset->ac->clabel->num_columns; 3295 parity_type = cset->ac->clabel->parityConfig; 3296 3297 /* XXX Check for duplicate components!?!?!? */ 3298 3299 /* Determine what the mod_counter is supposed to be for this set. */ 3300 3301 mod_counter_found = 0; 3302 mod_counter = 0; 3303 ac = cset->ac; 3304 while(ac!=NULL) { 3305 if (mod_counter_found==0) { 3306 mod_counter = ac->clabel->mod_counter; 3307 mod_counter_found = 1; 3308 } else { 3309 if (ac->clabel->mod_counter > mod_counter) { 3310 mod_counter = ac->clabel->mod_counter; 3311 } 3312 } 3313 ac = ac->next; 3314 } 3315 3316 num_missing = 0; 3317 auto_config = cset->ac; 3318 3319 even_pair_failed = 0; 3320 for(c=0; c<num_cols; c++) { 3321 ac = auto_config; 3322 while(ac!=NULL) { 3323 if ((ac->clabel->column == c) && 3324 (ac->clabel->mod_counter == mod_counter)) { 3325 /* it's this one... */ 3326 #ifdef DEBUG 3327 printf("Found: %s at %d\n", 3328 ac->devname,c); 3329 #endif 3330 break; 3331 } 3332 ac=ac->next; 3333 } 3334 if (ac==NULL) { 3335 /* Didn't find one here! */ 3336 /* special case for RAID 1, especially 3337 where there are more than 2 3338 components (where RAIDframe treats 3339 things a little differently :( ) */ 3340 if (parity_type == '1') { 3341 if (c%2 == 0) { /* even component */ 3342 even_pair_failed = 1; 3343 } else { /* odd component. If 3344 we're failed, and 3345 so is the even 3346 component, it's 3347 "Good Night, Charlie" */ 3348 if (even_pair_failed == 1) { 3349 return(0); 3350 } 3351 } 3352 } else { 3353 /* normal accounting */ 3354 num_missing++; 3355 } 3356 } 3357 if ((parity_type == '1') && (c%2 == 1)) { 3358 /* Just did an even component, and we didn't 3359 bail.. reset the even_pair_failed flag, 3360 and go on to the next component.... */ 3361 even_pair_failed = 0; 3362 } 3363 } 3364 3365 clabel = cset->ac->clabel; 3366 3367 if (((clabel->parityConfig == '0') && (num_missing > 0)) || 3368 ((clabel->parityConfig == '4') && (num_missing > 1)) || 3369 ((clabel->parityConfig == '5') && (num_missing > 1))) { 3370 /* XXX this needs to be made *much* more general */ 3371 /* Too many failures */ 3372 return(0); 3373 } 3374 /* otherwise, all is well, and we've got enough to take a kick 3375 at autoconfiguring this set */ 3376 return(1); 3377 } 3378 3379 void 3380 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config, 3381 RF_Raid_t *raidPtr) 3382 { 3383 RF_ComponentLabel_t *clabel; 3384 int i; 3385 3386 clabel = ac->clabel; 3387 3388 /* 1. Fill in the common stuff */ 3389 config->numRow = clabel->num_rows = 1; 3390 config->numCol = clabel->num_columns; 3391 config->numSpare = 0; /* XXX should this be set here? */ 3392 config->sectPerSU = clabel->sectPerSU; 3393 config->SUsPerPU = clabel->SUsPerPU; 3394 config->SUsPerRU = clabel->SUsPerRU; 3395 config->parityConfig = clabel->parityConfig; 3396 /* XXX... */ 3397 strcpy(config->diskQueueType,"fifo"); 3398 config->maxOutstandingDiskReqs = clabel->maxOutstanding; 3399 config->layoutSpecificSize = 0; /* XXX ?? */ 3400 3401 while(ac!=NULL) { 3402 /* row/col values will be in range due to the checks 3403 in reasonable_label() */ 3404 strcpy(config->devnames[0][ac->clabel->column], 3405 ac->devname); 3406 ac = ac->next; 3407 } 3408 3409 for(i=0;i<RF_MAXDBGV;i++) { 3410 config->debugVars[i][0] = 0; 3411 } 3412 } 3413 3414 int 3415 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value) 3416 { 3417 RF_ComponentLabel_t *clabel; 3418 int column; 3419 int sparecol; 3420 3421 raidPtr->autoconfigure = new_value; 3422 3423 for(column=0; column<raidPtr->numCol; column++) { 3424 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3425 clabel = raidget_component_label(raidPtr, column); 3426 clabel->autoconfigure = new_value; 3427 raidflush_component_label(raidPtr, column); 3428 } 3429 } 3430 for(column = 0; column < raidPtr->numSpare ; column++) { 3431 sparecol = raidPtr->numCol + column; 3432 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3433 clabel = raidget_component_label(raidPtr, sparecol); 3434 clabel->autoconfigure = new_value; 3435 raidflush_component_label(raidPtr, sparecol); 3436 } 3437 } 3438 return(new_value); 3439 } 3440 3441 int 3442 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value) 3443 { 3444 RF_ComponentLabel_t *clabel; 3445 int column; 3446 int sparecol; 3447 3448 raidPtr->root_partition = new_value; 3449 for(column=0; column<raidPtr->numCol; column++) { 3450 if (raidPtr->Disks[column].status == rf_ds_optimal) { 3451 clabel = raidget_component_label(raidPtr, column); 3452 clabel->root_partition = new_value; 3453 raidflush_component_label(raidPtr, column); 3454 } 3455 } 3456 for(column = 0; column < raidPtr->numSpare ; column++) { 3457 sparecol = raidPtr->numCol + column; 3458 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3459 clabel = raidget_component_label(raidPtr, sparecol); 3460 clabel->root_partition = new_value; 3461 raidflush_component_label(raidPtr, sparecol); 3462 } 3463 } 3464 return(new_value); 3465 } 3466 3467 void 3468 rf_release_all_vps(RF_ConfigSet_t *cset) 3469 { 3470 RF_AutoConfig_t *ac; 3471 3472 ac = cset->ac; 3473 while(ac!=NULL) { 3474 /* Close the vp, and give it back */ 3475 if (ac->vp) { 3476 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY); 3477 VOP_CLOSE(ac->vp, FREAD, NOCRED); 3478 vput(ac->vp); 3479 ac->vp = NULL; 3480 } 3481 ac = ac->next; 3482 } 3483 } 3484 3485 3486 void 3487 rf_cleanup_config_set(RF_ConfigSet_t *cset) 3488 { 3489 RF_AutoConfig_t *ac; 3490 RF_AutoConfig_t *next_ac; 3491 3492 ac = cset->ac; 3493 while(ac!=NULL) { 3494 next_ac = ac->next; 3495 /* nuke the label */ 3496 free(ac->clabel, M_RAIDFRAME); 3497 /* cleanup the config structure */ 3498 free(ac, M_RAIDFRAME); 3499 /* "next.." */ 3500 ac = next_ac; 3501 } 3502 /* and, finally, nuke the config set */ 3503 free(cset, M_RAIDFRAME); 3504 } 3505 3506 3507 void 3508 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel) 3509 { 3510 /* current version number */ 3511 clabel->version = RF_COMPONENT_LABEL_VERSION; 3512 clabel->serial_number = raidPtr->serial_number; 3513 clabel->mod_counter = raidPtr->mod_counter; 3514 3515 clabel->num_rows = 1; 3516 clabel->num_columns = raidPtr->numCol; 3517 clabel->clean = RF_RAID_DIRTY; /* not clean */ 3518 clabel->status = rf_ds_optimal; /* "It's good!" */ 3519 3520 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit; 3521 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU; 3522 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU; 3523 3524 clabel->blockSize = raidPtr->bytesPerSector; 3525 clabel->numBlocks = raidPtr->sectorsPerDisk; 3526 3527 /* XXX not portable */ 3528 clabel->parityConfig = raidPtr->Layout.map->parityConfig; 3529 clabel->maxOutstanding = raidPtr->maxOutstanding; 3530 clabel->autoconfigure = raidPtr->autoconfigure; 3531 clabel->root_partition = raidPtr->root_partition; 3532 clabel->last_unit = raidPtr->raidid; 3533 clabel->config_order = raidPtr->config_order; 3534 3535 #ifndef RF_NO_PARITY_MAP 3536 rf_paritymap_init_label(raidPtr->parity_map, clabel); 3537 #endif 3538 } 3539 3540 int 3541 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit) 3542 { 3543 RF_Raid_t *raidPtr; 3544 RF_Config_t *config; 3545 int raidID; 3546 int retcode; 3547 3548 #ifdef DEBUG 3549 printf("RAID autoconfigure\n"); 3550 #endif 3551 3552 retcode = 0; 3553 *unit = -1; 3554 3555 /* 1. Create a config structure */ 3556 3557 config = (RF_Config_t *)malloc(sizeof(RF_Config_t), 3558 M_RAIDFRAME, 3559 M_NOWAIT); 3560 if (config==NULL) { 3561 printf("Out of mem!?!?\n"); 3562 /* XXX do something more intelligent here. */ 3563 return(1); 3564 } 3565 3566 memset(config, 0, sizeof(RF_Config_t)); 3567 3568 /* 3569 2. Figure out what RAID ID this one is supposed to live at 3570 See if we can get the same RAID dev that it was configured 3571 on last time.. 3572 */ 3573 3574 raidID = cset->ac->clabel->last_unit; 3575 if ((raidID < 0) || (raidID >= numraid)) { 3576 /* let's not wander off into lala land. */ 3577 raidID = numraid - 1; 3578 } 3579 if (raidPtrs[raidID]->valid != 0) { 3580 3581 /* 3582 Nope... Go looking for an alternative... 3583 Start high so we don't immediately use raid0 if that's 3584 not taken. 3585 */ 3586 3587 for(raidID = numraid - 1; raidID >= 0; raidID--) { 3588 if (raidPtrs[raidID]->valid == 0) { 3589 /* can use this one! */ 3590 break; 3591 } 3592 } 3593 } 3594 3595 if (raidID < 0) { 3596 /* punt... */ 3597 printf("Unable to auto configure this set!\n"); 3598 printf("(Out of RAID devs!)\n"); 3599 free(config, M_RAIDFRAME); 3600 return(1); 3601 } 3602 3603 #ifdef DEBUG 3604 printf("Configuring raid%d:\n",raidID); 3605 #endif 3606 3607 raidPtr = raidPtrs[raidID]; 3608 3609 /* XXX all this stuff should be done SOMEWHERE ELSE! */ 3610 raidPtr->raidid = raidID; 3611 raidPtr->openings = RAIDOUTSTANDING; 3612 3613 /* 3. Build the configuration structure */ 3614 rf_create_configuration(cset->ac, config, raidPtr); 3615 3616 /* 4. Do the configuration */ 3617 retcode = rf_Configure(raidPtr, config, cset->ac); 3618 3619 if (retcode == 0) { 3620 3621 raidinit(raidPtrs[raidID]); 3622 3623 rf_markalldirty(raidPtrs[raidID]); 3624 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */ 3625 if (cset->ac->clabel->root_partition==1) { 3626 /* everything configured just fine. Make a note 3627 that this set is eligible to be root. */ 3628 cset->rootable = 1; 3629 /* XXX do this here? */ 3630 raidPtrs[raidID]->root_partition = 1; 3631 } 3632 } 3633 3634 /* 5. Cleanup */ 3635 free(config, M_RAIDFRAME); 3636 3637 *unit = raidID; 3638 return(retcode); 3639 } 3640 3641 void 3642 rf_disk_unbusy(RF_RaidAccessDesc_t *desc) 3643 { 3644 struct buf *bp; 3645 3646 bp = (struct buf *)desc->bp; 3647 disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev, 3648 (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ)); 3649 } 3650 3651 void 3652 rf_pool_init(struct pool *p, size_t size, const char *w_chan, 3653 size_t xmin, size_t xmax) 3654 { 3655 pool_init(p, size, 0, 0, 0, w_chan, NULL, IPL_BIO); 3656 pool_sethiwat(p, xmax); 3657 pool_prime(p, xmin); 3658 pool_setlowat(p, xmin); 3659 } 3660 3661 /* 3662 * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see 3663 * if there is IO pending and if that IO could possibly be done for a 3664 * given RAID set. Returns 0 if IO is waiting and can be done, 1 3665 * otherwise. 3666 * 3667 */ 3668 3669 int 3670 rf_buf_queue_check(int raidid) 3671 { 3672 if ((bufq_peek(raid_softc[raidid].buf_queue) != NULL) && 3673 raidPtrs[raidid]->openings > 0) { 3674 /* there is work to do */ 3675 return 0; 3676 } 3677 /* default is nothing to do */ 3678 return 1; 3679 } 3680 3681 int 3682 rf_getdisksize(struct vnode *vp, struct lwp *l, RF_RaidDisk_t *diskPtr) 3683 { 3684 struct partinfo dpart; 3685 struct dkwedge_info dkw; 3686 int error; 3687 3688 error = VOP_IOCTL(vp, DIOCGPART, &dpart, FREAD, l->l_cred); 3689 if (error == 0) { 3690 diskPtr->blockSize = dpart.disklab->d_secsize; 3691 diskPtr->numBlocks = dpart.part->p_size - rf_protectedSectors; 3692 diskPtr->partitionSize = dpart.part->p_size; 3693 return 0; 3694 } 3695 3696 error = VOP_IOCTL(vp, DIOCGWEDGEINFO, &dkw, FREAD, l->l_cred); 3697 if (error == 0) { 3698 diskPtr->blockSize = 512; /* XXX */ 3699 diskPtr->numBlocks = dkw.dkw_size - rf_protectedSectors; 3700 diskPtr->partitionSize = dkw.dkw_size; 3701 return 0; 3702 } 3703 return error; 3704 } 3705 3706 static int 3707 raid_match(device_t self, cfdata_t cfdata, void *aux) 3708 { 3709 return 1; 3710 } 3711 3712 static void 3713 raid_attach(device_t parent, device_t self, void *aux) 3714 { 3715 3716 } 3717 3718 3719 static int 3720 raid_detach(device_t self, int flags) 3721 { 3722 int error; 3723 struct raid_softc *rs = &raid_softc[device_unit(self)]; 3724 3725 if ((error = raidlock(rs)) != 0) 3726 return (error); 3727 3728 error = raid_detach_unlocked(rs); 3729 3730 raidunlock(rs); 3731 3732 return error; 3733 } 3734 3735 static void 3736 rf_set_properties(struct raid_softc *rs, RF_Raid_t *raidPtr) 3737 { 3738 prop_dictionary_t disk_info, odisk_info, geom; 3739 disk_info = prop_dictionary_create(); 3740 geom = prop_dictionary_create(); 3741 prop_dictionary_set_uint64(geom, "sectors-per-unit", 3742 raidPtr->totalSectors); 3743 prop_dictionary_set_uint32(geom, "sector-size", 3744 raidPtr->bytesPerSector); 3745 3746 prop_dictionary_set_uint16(geom, "sectors-per-track", 3747 raidPtr->Layout.dataSectorsPerStripe); 3748 prop_dictionary_set_uint16(geom, "tracks-per-cylinder", 3749 4 * raidPtr->numCol); 3750 3751 prop_dictionary_set_uint64(geom, "cylinders-per-unit", 3752 raidPtr->totalSectors / (raidPtr->Layout.dataSectorsPerStripe * 3753 (4 * raidPtr->numCol))); 3754 3755 prop_dictionary_set(disk_info, "geometry", geom); 3756 prop_object_release(geom); 3757 prop_dictionary_set(device_properties(rs->sc_dev), 3758 "disk-info", disk_info); 3759 odisk_info = rs->sc_dkdev.dk_info; 3760 rs->sc_dkdev.dk_info = disk_info; 3761 if (odisk_info) 3762 prop_object_release(odisk_info); 3763 } 3764 3765 /* 3766 * Implement forwarding of the DIOCCACHESYNC ioctl to each of the components. 3767 * We end up returning whatever error was returned by the first cache flush 3768 * that fails. 3769 */ 3770 3771 int 3772 rf_sync_component_caches(RF_Raid_t *raidPtr) 3773 { 3774 int c, sparecol; 3775 int e,error; 3776 int force = 1; 3777 3778 error = 0; 3779 for (c = 0; c < raidPtr->numCol; c++) { 3780 if (raidPtr->Disks[c].status == rf_ds_optimal) { 3781 e = VOP_IOCTL(raidPtr->raid_cinfo[c].ci_vp, DIOCCACHESYNC, 3782 &force, FWRITE, NOCRED); 3783 if (e) { 3784 if (e != ENODEV) 3785 printf("raid%d: cache flush to component %s failed.\n", 3786 raidPtr->raidid, raidPtr->Disks[c].devname); 3787 if (error == 0) { 3788 error = e; 3789 } 3790 } 3791 } 3792 } 3793 3794 for( c = 0; c < raidPtr->numSpare ; c++) { 3795 sparecol = raidPtr->numCol + c; 3796 /* Need to ensure that the reconstruct actually completed! */ 3797 if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) { 3798 e = VOP_IOCTL(raidPtr->raid_cinfo[sparecol].ci_vp, 3799 DIOCCACHESYNC, &force, FWRITE, NOCRED); 3800 if (e) { 3801 if (e != ENODEV) 3802 printf("raid%d: cache flush to component %s failed.\n", 3803 raidPtr->raidid, raidPtr->Disks[sparecol].devname); 3804 if (error == 0) { 3805 error = e; 3806 } 3807 } 3808 } 3809 } 3810 return error; 3811 } 3812