1 /* $NetBSD: dksubr.c,v 1.112 2020/03/01 03:21:54 riastradh Exp $ */ 2 3 /*- 4 * Copyright (c) 1996, 1997, 1998, 1999, 2002, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe and Roland C. Dowdeswell. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: dksubr.c,v 1.112 2020/03/01 03:21:54 riastradh Exp $"); 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/stat.h> 38 #include <sys/proc.h> 39 #include <sys/ioctl.h> 40 #include <sys/device.h> 41 #include <sys/disk.h> 42 #include <sys/disklabel.h> 43 #include <sys/buf.h> 44 #include <sys/bufq.h> 45 #include <sys/vnode.h> 46 #include <sys/fcntl.h> 47 #include <sys/namei.h> 48 #include <sys/module.h> 49 #include <sys/syslog.h> 50 51 #include <dev/dkvar.h> 52 #include <miscfs/specfs/specdev.h> /* for v_rdev */ 53 54 int dkdebug = 0; 55 56 #ifdef DEBUG 57 #define DKDB_FOLLOW 0x1 58 #define DKDB_INIT 0x2 59 #define DKDB_VNODE 0x4 60 #define DKDB_DUMP 0x8 61 62 #define IFDEBUG(x,y) if (dkdebug & (x)) y 63 #define DPRINTF(x,y) IFDEBUG(x, printf y) 64 #define DPRINTF_FOLLOW(y) DPRINTF(DKDB_FOLLOW, y) 65 #else 66 #define IFDEBUG(x,y) 67 #define DPRINTF(x,y) 68 #define DPRINTF_FOLLOW(y) 69 #endif 70 71 #define DKF_READYFORDUMP (DKF_INITED|DKF_TAKEDUMP) 72 73 static int dk_subr_modcmd(modcmd_t, void *); 74 75 #define DKLABELDEV(dev) \ 76 (MAKEDISKDEV(major((dev)), DISKUNIT((dev)), RAW_PART)) 77 78 static void dk_makedisklabel(struct dk_softc *); 79 static int dk_translate(struct dk_softc *, struct buf *); 80 static void dk_done1(struct dk_softc *, struct buf *, bool); 81 82 void 83 dk_init(struct dk_softc *dksc, device_t dev, int dtype) 84 { 85 86 memset(dksc, 0x0, sizeof(*dksc)); 87 dksc->sc_dtype = dtype; 88 dksc->sc_dev = dev; 89 90 strlcpy(dksc->sc_xname, device_xname(dev), DK_XNAME_SIZE); 91 dksc->sc_dkdev.dk_name = dksc->sc_xname; 92 } 93 94 void 95 dk_attach(struct dk_softc *dksc) 96 { 97 KASSERT(dksc->sc_dev != NULL); 98 99 mutex_init(&dksc->sc_iolock, MUTEX_DEFAULT, IPL_VM); 100 dksc->sc_flags |= DKF_READYFORDUMP; 101 #ifdef DIAGNOSTIC 102 dksc->sc_flags |= DKF_WARNLABEL | DKF_LABELSANITY; 103 #endif 104 105 if ((dksc->sc_flags & DKF_NO_RND) == 0) { 106 /* Attach the device into the rnd source list. */ 107 rnd_attach_source(&dksc->sc_rnd_source, dksc->sc_xname, 108 RND_TYPE_DISK, RND_FLAG_DEFAULT); 109 } 110 } 111 112 void 113 dk_detach(struct dk_softc *dksc) 114 { 115 if ((dksc->sc_flags & DKF_NO_RND) == 0) { 116 /* Unhook the entropy source. */ 117 rnd_detach_source(&dksc->sc_rnd_source); 118 } 119 120 dksc->sc_flags &= ~DKF_READYFORDUMP; 121 mutex_destroy(&dksc->sc_iolock); 122 } 123 124 /* ARGSUSED */ 125 int 126 dk_open(struct dk_softc *dksc, dev_t dev, 127 int flags, int fmt, struct lwp *l) 128 { 129 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 130 struct disklabel *lp = dksc->sc_dkdev.dk_label; 131 int part = DISKPART(dev); 132 int pmask = 1 << part; 133 int ret = 0; 134 struct disk *dk = &dksc->sc_dkdev; 135 136 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%x)\n", __func__, 137 dksc->sc_xname, dksc, dev, flags)); 138 139 mutex_enter(&dk->dk_openlock); 140 141 /* 142 * If there are wedges, and this is not RAW_PART, then we 143 * need to fail. 144 */ 145 if (dk->dk_nwedges != 0 && part != RAW_PART) { 146 ret = EBUSY; 147 goto done; 148 } 149 150 /* If no dkdriver attached, bail */ 151 if (dkd == NULL) { 152 ret = ENXIO; 153 goto done; 154 } 155 156 /* 157 * initialize driver for the first opener 158 */ 159 if (dk->dk_openmask == 0 && dkd->d_firstopen != NULL) { 160 ret = (*dkd->d_firstopen)(dksc->sc_dev, dev, flags, fmt); 161 if (ret) 162 goto done; 163 } 164 165 /* 166 * If we're init'ed and there are no other open partitions then 167 * update the in-core disklabel. 168 */ 169 if ((dksc->sc_flags & DKF_INITED)) { 170 if ((dksc->sc_flags & DKF_VLABEL) == 0) { 171 dksc->sc_flags |= DKF_VLABEL; 172 dk_getdisklabel(dksc, dev); 173 } 174 } 175 176 /* Fail if we can't find the partition. */ 177 if (part != RAW_PART && 178 ((dksc->sc_flags & DKF_VLABEL) == 0 || 179 part >= lp->d_npartitions || 180 lp->d_partitions[part].p_fstype == FS_UNUSED)) { 181 ret = ENXIO; 182 goto done; 183 } 184 185 /* Mark our unit as open. */ 186 switch (fmt) { 187 case S_IFCHR: 188 dk->dk_copenmask |= pmask; 189 break; 190 case S_IFBLK: 191 dk->dk_bopenmask |= pmask; 192 break; 193 } 194 195 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask; 196 197 done: 198 mutex_exit(&dk->dk_openlock); 199 return ret; 200 } 201 202 /* ARGSUSED */ 203 int 204 dk_close(struct dk_softc *dksc, dev_t dev, 205 int flags, int fmt, struct lwp *l) 206 { 207 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 208 int part = DISKPART(dev); 209 int pmask = 1 << part; 210 struct disk *dk = &dksc->sc_dkdev; 211 212 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%x)\n", __func__, 213 dksc->sc_xname, dksc, dev, flags)); 214 215 mutex_enter(&dk->dk_openlock); 216 217 switch (fmt) { 218 case S_IFCHR: 219 dk->dk_copenmask &= ~pmask; 220 break; 221 case S_IFBLK: 222 dk->dk_bopenmask &= ~pmask; 223 break; 224 } 225 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask; 226 227 if (dk->dk_openmask == 0) { 228 if (dkd->d_lastclose != NULL) 229 (*dkd->d_lastclose)(dksc->sc_dev); 230 if ((dksc->sc_flags & DKF_KLABEL) == 0) 231 dksc->sc_flags &= ~DKF_VLABEL; 232 } 233 234 mutex_exit(&dk->dk_openlock); 235 return 0; 236 } 237 238 static int 239 dk_translate(struct dk_softc *dksc, struct buf *bp) 240 { 241 int part; 242 int wlabel; 243 daddr_t blkno; 244 struct disklabel *lp; 245 struct disk *dk; 246 uint64_t numsecs; 247 unsigned secsize; 248 249 lp = dksc->sc_dkdev.dk_label; 250 dk = &dksc->sc_dkdev; 251 252 part = DISKPART(bp->b_dev); 253 numsecs = dk->dk_geom.dg_secperunit; 254 secsize = dk->dk_geom.dg_secsize; 255 256 /* 257 * The transfer must be a whole number of blocks and the offset must 258 * not be negative. 259 */ 260 if ((bp->b_bcount % secsize) != 0 || bp->b_blkno < 0) { 261 bp->b_error = EINVAL; 262 goto done; 263 } 264 265 /* If there is nothing to do, then we are done */ 266 if (bp->b_bcount == 0) 267 goto done; 268 269 wlabel = dksc->sc_flags & (DKF_WLABEL|DKF_LABELLING); 270 if (part == RAW_PART) { 271 uint64_t numblocks = btodb(numsecs * secsize); 272 if (bounds_check_with_mediasize(bp, DEV_BSIZE, numblocks) <= 0) 273 goto done; 274 } else { 275 if (bounds_check_with_label(&dksc->sc_dkdev, bp, wlabel) <= 0) 276 goto done; 277 } 278 279 /* 280 * Convert the block number to absolute and put it in terms 281 * of the device's logical block size. 282 */ 283 if (secsize >= DEV_BSIZE) 284 blkno = bp->b_blkno / (secsize / DEV_BSIZE); 285 else 286 blkno = bp->b_blkno * (DEV_BSIZE / secsize); 287 288 if (part != RAW_PART) 289 blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset; 290 bp->b_rawblkno = blkno; 291 292 return -1; 293 294 done: 295 bp->b_resid = bp->b_bcount; 296 return bp->b_error; 297 } 298 299 static int 300 dk_strategy1(struct dk_softc *dksc, struct buf *bp) 301 { 302 int error; 303 304 DPRINTF_FOLLOW(("%s(%s, %p, %p)\n", __func__, 305 dksc->sc_xname, dksc, bp)); 306 307 if (!(dksc->sc_flags & DKF_INITED)) { 308 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 309 bp->b_error = ENXIO; 310 bp->b_resid = bp->b_bcount; 311 biodone(bp); 312 return 1; 313 } 314 315 error = dk_translate(dksc, bp); 316 if (error >= 0) { 317 biodone(bp); 318 return 1; 319 } 320 321 return 0; 322 } 323 324 void 325 dk_strategy(struct dk_softc *dksc, struct buf *bp) 326 { 327 int error; 328 329 error = dk_strategy1(dksc, bp); 330 if (error) 331 return; 332 333 /* 334 * Queue buffer and start unit 335 */ 336 dk_start(dksc, bp); 337 } 338 339 int 340 dk_strategy_defer(struct dk_softc *dksc, struct buf *bp) 341 { 342 int error; 343 344 error = dk_strategy1(dksc, bp); 345 if (error) 346 return error; 347 348 /* 349 * Queue buffer only 350 */ 351 mutex_enter(&dksc->sc_iolock); 352 disk_wait(&dksc->sc_dkdev); 353 bufq_put(dksc->sc_bufq, bp); 354 mutex_exit(&dksc->sc_iolock); 355 356 return 0; 357 } 358 359 int 360 dk_strategy_pending(struct dk_softc *dksc) 361 { 362 struct buf *bp; 363 364 if (!(dksc->sc_flags & DKF_INITED)) { 365 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 366 return 0; 367 } 368 369 mutex_enter(&dksc->sc_iolock); 370 bp = bufq_peek(dksc->sc_bufq); 371 mutex_exit(&dksc->sc_iolock); 372 373 return bp != NULL; 374 } 375 376 void 377 dk_start(struct dk_softc *dksc, struct buf *bp) 378 { 379 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 380 int error; 381 382 if (!(dksc->sc_flags & DKF_INITED)) { 383 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 384 return; 385 } 386 387 mutex_enter(&dksc->sc_iolock); 388 389 if (bp != NULL) { 390 bp->b_ci = curcpu(); 391 disk_wait(&dksc->sc_dkdev); 392 bufq_put(dksc->sc_bufq, bp); 393 } 394 395 /* 396 * If another thread is running the queue, increment 397 * busy counter to 2 so that the queue is retried, 398 * because the driver may now accept additional 399 * requests. 400 */ 401 if (dksc->sc_busy < 2) 402 dksc->sc_busy++; 403 if (dksc->sc_busy > 1) 404 goto done; 405 406 /* 407 * Peeking at the buffer queue and committing the operation 408 * only after success isn't atomic. 409 * 410 * So when a diskstart fails, the buffer is saved 411 * and tried again before the next buffer is fetched. 412 * dk_drain() handles flushing of a saved buffer. 413 * 414 * This keeps order of I/O operations, unlike bufq_put. 415 */ 416 417 while (dksc->sc_busy > 0) { 418 419 bp = dksc->sc_deferred; 420 dksc->sc_deferred = NULL; 421 422 if (bp == NULL) 423 bp = bufq_get(dksc->sc_bufq); 424 425 while (bp != NULL) { 426 427 disk_busy(&dksc->sc_dkdev); 428 mutex_exit(&dksc->sc_iolock); 429 error = dkd->d_diskstart(dksc->sc_dev, bp); 430 mutex_enter(&dksc->sc_iolock); 431 if (error == EAGAIN) { 432 KASSERT(dksc->sc_deferred == NULL); 433 dksc->sc_deferred = bp; 434 disk_unbusy(&dksc->sc_dkdev, 0, (bp->b_flags & B_READ)); 435 disk_wait(&dksc->sc_dkdev); 436 break; 437 } 438 439 if (error != 0) { 440 bp->b_error = error; 441 bp->b_resid = bp->b_bcount; 442 dk_done1(dksc, bp, false); 443 } 444 445 bp = bufq_get(dksc->sc_bufq); 446 } 447 448 dksc->sc_busy--; 449 } 450 done: 451 mutex_exit(&dksc->sc_iolock); 452 } 453 454 static void 455 dk_done1(struct dk_softc *dksc, struct buf *bp, bool lock) 456 { 457 struct disk *dk = &dksc->sc_dkdev; 458 459 if (bp->b_error != 0) { 460 struct cfdriver *cd = device_cfdriver(dksc->sc_dev); 461 462 diskerr(bp, cd->cd_name, "error", LOG_PRINTF, 0, 463 dk->dk_label); 464 printf("\n"); 465 } 466 467 if (lock) 468 mutex_enter(&dksc->sc_iolock); 469 disk_unbusy(dk, bp->b_bcount - bp->b_resid, (bp->b_flags & B_READ)); 470 471 if ((dksc->sc_flags & DKF_NO_RND) == 0) 472 rnd_add_uint32(&dksc->sc_rnd_source, bp->b_rawblkno); 473 if (lock) 474 mutex_exit(&dksc->sc_iolock); 475 476 biodone(bp); 477 } 478 479 void 480 dk_done(struct dk_softc *dksc, struct buf *bp) 481 { 482 dk_done1(dksc, bp, true); 483 } 484 485 void 486 dk_drain(struct dk_softc *dksc) 487 { 488 struct buf *bp; 489 490 mutex_enter(&dksc->sc_iolock); 491 bp = dksc->sc_deferred; 492 dksc->sc_deferred = NULL; 493 if (bp != NULL) { 494 bp->b_error = EIO; 495 bp->b_resid = bp->b_bcount; 496 biodone(bp); 497 } 498 bufq_drain(dksc->sc_bufq); 499 mutex_exit(&dksc->sc_iolock); 500 } 501 502 int 503 dk_discard(struct dk_softc *dksc, dev_t dev, off_t pos, off_t len) 504 { 505 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 506 unsigned secsize = dksc->sc_dkdev.dk_geom.dg_secsize; 507 struct buf tmp, *bp = &tmp; 508 int maxsz; 509 int error = 0; 510 511 KASSERT(len >= 0); 512 513 DPRINTF_FOLLOW(("%s(%s, %p, 0x"PRIx64", %jd, %jd)\n", __func__, 514 dksc->sc_xname, dksc, (intmax_t)pos, (intmax_t)len)); 515 516 if (!(dksc->sc_flags & DKF_INITED)) { 517 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 518 return ENXIO; 519 } 520 521 if (secsize == 0 || (pos % secsize) != 0 || (len % secsize) != 0) 522 return EINVAL; 523 524 /* largest value that b_bcount can store */ 525 maxsz = rounddown(INT_MAX, secsize); 526 527 while (len > 0) { 528 /* enough data to please the bounds checking code */ 529 bp->b_dev = dev; 530 bp->b_blkno = (daddr_t)(pos / secsize); 531 bp->b_bcount = uimin(len, maxsz); 532 bp->b_flags = B_WRITE; 533 534 error = dk_translate(dksc, bp); 535 if (error >= 0) 536 break; 537 538 error = dkd->d_discard(dksc->sc_dev, 539 (off_t)bp->b_rawblkno * secsize, 540 (off_t)bp->b_bcount); 541 if (error) 542 break; 543 544 pos += bp->b_bcount; 545 len -= bp->b_bcount; 546 } 547 548 return error; 549 } 550 551 int 552 dk_size(struct dk_softc *dksc, dev_t dev) 553 { 554 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 555 struct disklabel *lp; 556 int is_open; 557 int part; 558 int size; 559 560 if ((dksc->sc_flags & DKF_INITED) == 0) 561 return -1; 562 563 part = DISKPART(dev); 564 is_open = dksc->sc_dkdev.dk_openmask & (1 << part); 565 566 if (!is_open && dkd->d_open(dev, 0, S_IFBLK, curlwp)) 567 return -1; 568 569 lp = dksc->sc_dkdev.dk_label; 570 if (lp->d_partitions[part].p_fstype != FS_SWAP) 571 size = -1; 572 else 573 size = lp->d_partitions[part].p_size * 574 (lp->d_secsize / DEV_BSIZE); 575 576 if (!is_open && dkd->d_close(dev, 0, S_IFBLK, curlwp)) 577 return -1; 578 579 return size; 580 } 581 582 int 583 dk_ioctl(struct dk_softc *dksc, dev_t dev, 584 u_long cmd, void *data, int flag, struct lwp *l) 585 { 586 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 587 struct disklabel *lp; 588 struct disk *dk = &dksc->sc_dkdev; 589 #ifdef __HAVE_OLD_DISKLABEL 590 struct disklabel newlabel; 591 #endif 592 int error; 593 594 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%lx)\n", __func__, 595 dksc->sc_xname, dksc, dev, cmd)); 596 597 /* ensure that the pseudo disk is open for writes for these commands */ 598 switch (cmd) { 599 case DIOCSDINFO: 600 case DIOCWDINFO: 601 #ifdef __HAVE_OLD_DISKLABEL 602 case ODIOCSDINFO: 603 case ODIOCWDINFO: 604 #endif 605 case DIOCKLABEL: 606 case DIOCWLABEL: 607 case DIOCAWEDGE: 608 case DIOCDWEDGE: 609 case DIOCSSTRATEGY: 610 if ((flag & FWRITE) == 0) 611 return EBADF; 612 } 613 614 /* ensure that the pseudo-disk is initialized for these */ 615 switch (cmd) { 616 case DIOCGDINFO: 617 case DIOCSDINFO: 618 case DIOCWDINFO: 619 case DIOCGPARTINFO: 620 case DIOCKLABEL: 621 case DIOCWLABEL: 622 case DIOCGDEFLABEL: 623 case DIOCAWEDGE: 624 case DIOCDWEDGE: 625 case DIOCLWEDGES: 626 case DIOCMWEDGES: 627 case DIOCRMWEDGES: 628 case DIOCCACHESYNC: 629 #ifdef __HAVE_OLD_DISKLABEL 630 case ODIOCGDINFO: 631 case ODIOCSDINFO: 632 case ODIOCWDINFO: 633 case ODIOCGDEFLABEL: 634 #endif 635 if ((dksc->sc_flags & DKF_INITED) == 0) 636 return ENXIO; 637 } 638 639 error = disk_ioctl(dk, dev, cmd, data, flag, l); 640 if (error != EPASSTHROUGH) 641 return error; 642 else 643 error = 0; 644 645 switch (cmd) { 646 case DIOCWDINFO: 647 case DIOCSDINFO: 648 #ifdef __HAVE_OLD_DISKLABEL 649 case ODIOCWDINFO: 650 case ODIOCSDINFO: 651 #endif 652 #ifdef __HAVE_OLD_DISKLABEL 653 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 654 memset(&newlabel, 0, sizeof newlabel); 655 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 656 lp = &newlabel; 657 } else 658 #endif 659 lp = (struct disklabel *)data; 660 661 mutex_enter(&dk->dk_openlock); 662 dksc->sc_flags |= DKF_LABELLING; 663 664 error = setdisklabel(dksc->sc_dkdev.dk_label, 665 lp, 0, dksc->sc_dkdev.dk_cpulabel); 666 if (error == 0) { 667 if (cmd == DIOCWDINFO 668 #ifdef __HAVE_OLD_DISKLABEL 669 || cmd == ODIOCWDINFO 670 #endif 671 ) 672 error = writedisklabel(DKLABELDEV(dev), 673 dkd->d_strategy, dksc->sc_dkdev.dk_label, 674 dksc->sc_dkdev.dk_cpulabel); 675 } 676 677 dksc->sc_flags &= ~DKF_LABELLING; 678 mutex_exit(&dk->dk_openlock); 679 break; 680 681 case DIOCKLABEL: 682 if (*(int *)data != 0) 683 dksc->sc_flags |= DKF_KLABEL; 684 else 685 dksc->sc_flags &= ~DKF_KLABEL; 686 break; 687 688 case DIOCWLABEL: 689 if (*(int *)data != 0) 690 dksc->sc_flags |= DKF_WLABEL; 691 else 692 dksc->sc_flags &= ~DKF_WLABEL; 693 break; 694 695 case DIOCGDEFLABEL: 696 dk_getdefaultlabel(dksc, (struct disklabel *)data); 697 break; 698 699 #ifdef __HAVE_OLD_DISKLABEL 700 case ODIOCGDEFLABEL: 701 dk_getdefaultlabel(dksc, &newlabel); 702 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 703 return ENOTTY; 704 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 705 break; 706 #endif 707 708 case DIOCGSTRATEGY: 709 { 710 struct disk_strategy *dks = (void *)data; 711 712 mutex_enter(&dksc->sc_iolock); 713 if (dksc->sc_bufq != NULL) 714 strlcpy(dks->dks_name, 715 bufq_getstrategyname(dksc->sc_bufq), 716 sizeof(dks->dks_name)); 717 else 718 error = EINVAL; 719 mutex_exit(&dksc->sc_iolock); 720 dks->dks_paramlen = 0; 721 break; 722 } 723 724 case DIOCSSTRATEGY: 725 { 726 struct disk_strategy *dks = (void *)data; 727 struct bufq_state *new; 728 struct bufq_state *old; 729 730 if (dks->dks_param != NULL) { 731 return EINVAL; 732 } 733 dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */ 734 error = bufq_alloc(&new, dks->dks_name, 735 BUFQ_EXACT|BUFQ_SORT_RAWBLOCK); 736 if (error) { 737 return error; 738 } 739 mutex_enter(&dksc->sc_iolock); 740 old = dksc->sc_bufq; 741 if (old) 742 bufq_move(new, old); 743 dksc->sc_bufq = new; 744 mutex_exit(&dksc->sc_iolock); 745 if (old) 746 bufq_free(old); 747 break; 748 } 749 750 default: 751 error = ENOTTY; 752 } 753 754 return error; 755 } 756 757 /* 758 * dk_dump dumps all of physical memory into the partition specified. 759 * This requires substantially more framework than {s,w}ddump, and hence 760 * is probably much more fragile. 761 * 762 */ 763 764 #define DKFF_READYFORDUMP(x) (((x) & DKF_READYFORDUMP) == DKF_READYFORDUMP) 765 static volatile int dk_dumping = 0; 766 767 /* ARGSUSED */ 768 int 769 dk_dump(struct dk_softc *dksc, dev_t dev, 770 daddr_t blkno, void *vav, size_t size, int flags) 771 { 772 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 773 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 774 char *va = vav; 775 struct disklabel *lp; 776 struct partition *p; 777 int part, towrt, maxblkcnt, nblk; 778 int maxxfer, rv = 0; 779 780 /* 781 * ensure that we consider this device to be safe for dumping, 782 * and that the device is configured. 783 */ 784 if (!DKFF_READYFORDUMP(dksc->sc_flags)) { 785 DPRINTF(DKDB_DUMP, ("%s: bad dump flags 0x%x\n", __func__, 786 dksc->sc_flags)); 787 return ENXIO; 788 } 789 790 /* ensure that we are not already dumping */ 791 if (dk_dumping) 792 return EFAULT; 793 if ((flags & DK_DUMP_RECURSIVE) == 0) 794 dk_dumping = 1; 795 796 if (dkd->d_dumpblocks == NULL) { 797 DPRINTF(DKDB_DUMP, ("%s: no dumpblocks\n", __func__)); 798 return ENXIO; 799 } 800 801 /* device specific max transfer size */ 802 maxxfer = MAXPHYS; 803 if (dkd->d_iosize != NULL) 804 (*dkd->d_iosize)(dksc->sc_dev, &maxxfer); 805 806 /* Convert to disk sectors. Request must be a multiple of size. */ 807 part = DISKPART(dev); 808 lp = dksc->sc_dkdev.dk_label; 809 if ((size % lp->d_secsize) != 0) { 810 DPRINTF(DKDB_DUMP, ("%s: odd size %zu\n", __func__, size)); 811 return EFAULT; 812 } 813 towrt = size / lp->d_secsize; 814 blkno = dbtob(blkno) / lp->d_secsize; /* blkno in secsize units */ 815 816 p = &lp->d_partitions[part]; 817 if (part == RAW_PART) { 818 if (p->p_fstype != FS_UNUSED) { 819 DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__, 820 p->p_fstype)); 821 return ENXIO; 822 } 823 /* Check whether dump goes to a wedge */ 824 if (dksc->sc_dkdev.dk_nwedges == 0) { 825 DPRINTF(DKDB_DUMP, ("%s: dump to raw\n", __func__)); 826 return ENXIO; 827 } 828 /* Check transfer bounds against media size */ 829 if (blkno < 0 || (blkno + towrt) > dg->dg_secperunit) { 830 DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, " 831 "nsects=%jd\n", __func__, (intmax_t)blkno, towrt, dg->dg_secperunit)); 832 return EINVAL; 833 } 834 } else { 835 int nsects, sectoff; 836 837 if (p->p_fstype != FS_SWAP) { 838 DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__, 839 p->p_fstype)); 840 return ENXIO; 841 } 842 nsects = p->p_size; 843 sectoff = p->p_offset; 844 845 /* Check transfer bounds against partition size. */ 846 if ((blkno < 0) || ((blkno + towrt) > nsects)) { 847 DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, " 848 "nsects=%d\n", __func__, (intmax_t)blkno, towrt, nsects)); 849 return EINVAL; 850 } 851 852 /* Offset block number to start of partition. */ 853 blkno += sectoff; 854 } 855 856 /* Start dumping and return when done. */ 857 maxblkcnt = howmany(maxxfer, lp->d_secsize); 858 while (towrt > 0) { 859 nblk = uimin(maxblkcnt, towrt); 860 861 if ((rv = (*dkd->d_dumpblocks)(dksc->sc_dev, va, blkno, nblk)) 862 != 0) { 863 DPRINTF(DKDB_DUMP, ("%s: dumpblocks %d\n", __func__, 864 rv)); 865 return rv; 866 } 867 868 towrt -= nblk; 869 blkno += nblk; 870 va += nblk * lp->d_secsize; 871 } 872 873 if ((flags & DK_DUMP_RECURSIVE) == 0) 874 dk_dumping = 0; 875 876 return 0; 877 } 878 879 /* ARGSUSED */ 880 void 881 dk_getdefaultlabel(struct dk_softc *dksc, struct disklabel *lp) 882 { 883 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 884 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 885 886 memset(lp, 0, sizeof(*lp)); 887 888 if (dg->dg_secperunit > UINT32_MAX) 889 lp->d_secperunit = UINT32_MAX; 890 else 891 lp->d_secperunit = dg->dg_secperunit; 892 lp->d_secsize = dg->dg_secsize; 893 lp->d_nsectors = dg->dg_nsectors; 894 lp->d_ntracks = dg->dg_ntracks; 895 lp->d_ncylinders = dg->dg_ncylinders; 896 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 897 898 strlcpy(lp->d_typename, dksc->sc_xname, sizeof(lp->d_typename)); 899 lp->d_type = dksc->sc_dtype; 900 strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 901 lp->d_rpm = 3600; 902 lp->d_interleave = 1; 903 lp->d_flags = 0; 904 905 lp->d_partitions[RAW_PART].p_offset = 0; 906 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit; 907 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 908 lp->d_npartitions = RAW_PART + 1; 909 910 lp->d_magic = DISKMAGIC; 911 lp->d_magic2 = DISKMAGIC; 912 913 if (dkd->d_label) 914 dkd->d_label(dksc->sc_dev, lp); 915 916 lp->d_checksum = dkcksum(lp); 917 } 918 919 /* ARGSUSED */ 920 void 921 dk_getdisklabel(struct dk_softc *dksc, dev_t dev) 922 { 923 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 924 struct disklabel *lp = dksc->sc_dkdev.dk_label; 925 struct cpu_disklabel *clp = dksc->sc_dkdev.dk_cpulabel; 926 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 927 struct partition *pp; 928 int i, lpratio, dgratio; 929 const char *errstring; 930 931 memset(clp, 0x0, sizeof(*clp)); 932 dk_getdefaultlabel(dksc, lp); 933 errstring = readdisklabel(DKLABELDEV(dev), dkd->d_strategy, 934 dksc->sc_dkdev.dk_label, dksc->sc_dkdev.dk_cpulabel); 935 if (errstring) { 936 dk_makedisklabel(dksc); 937 if (dksc->sc_flags & DKF_WARNLABEL) 938 printf("%s: %s\n", dksc->sc_xname, errstring); 939 return; 940 } 941 942 if ((dksc->sc_flags & DKF_LABELSANITY) == 0) 943 return; 944 945 /* Convert sector counts to multiple of DEV_BSIZE for comparison */ 946 lpratio = dgratio = 1; 947 if (lp->d_secsize > DEV_BSIZE) 948 lpratio = lp->d_secsize / DEV_BSIZE; 949 if (dg->dg_secsize > DEV_BSIZE) 950 dgratio = dg->dg_secsize / DEV_BSIZE; 951 952 /* Sanity check */ 953 if ((uint64_t)lp->d_secperunit * lpratio > dg->dg_secperunit * dgratio) 954 printf("WARNING: %s: " 955 "total unit size in disklabel (%" PRIu64 ") " 956 "!= the size of %s (%" PRIu64 ")\n", dksc->sc_xname, 957 (uint64_t)lp->d_secperunit * lpratio, dksc->sc_xname, 958 dg->dg_secperunit * dgratio); 959 else if (lp->d_secperunit < UINT32_MAX && 960 (uint64_t)lp->d_secperunit * lpratio < dg->dg_secperunit * dgratio) 961 printf("%s: %" PRIu64 " trailing sectors not covered" 962 " by disklabel\n", dksc->sc_xname, 963 (dg->dg_secperunit * dgratio) 964 - (lp->d_secperunit * lpratio)); 965 966 for (i=0; i < lp->d_npartitions; i++) { 967 uint64_t pend; 968 969 pp = &lp->d_partitions[i]; 970 pend = pp->p_offset + pp->p_size; 971 if (pend * lpratio > dg->dg_secperunit * dgratio) 972 printf("WARNING: %s: end of partition `%c' exceeds " 973 "the size of %s (%" PRIu64 ")\n", dksc->sc_xname, 974 'a' + i, dksc->sc_xname, 975 dg->dg_secperunit * dgratio); 976 } 977 } 978 979 /* 980 * Heuristic to conjure a disklabel if reading a disklabel failed. 981 * 982 * This is to allow the raw partition to be used for a filesystem 983 * without caring about the write protected label sector. 984 * 985 * If the driver provides it's own callback, use that instead. 986 */ 987 /* ARGSUSED */ 988 static void 989 dk_makedisklabel(struct dk_softc *dksc) 990 { 991 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 992 struct disklabel *lp = dksc->sc_dkdev.dk_label; 993 994 strlcpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 995 996 if (dkd->d_label) 997 dkd->d_label(dksc->sc_dev, lp); 998 else 999 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 1000 1001 lp->d_checksum = dkcksum(lp); 1002 } 1003 1004 MODULE(MODULE_CLASS_MISC, dk_subr, NULL); 1005 1006 static int 1007 dk_subr_modcmd(modcmd_t cmd, void *arg) 1008 { 1009 switch (cmd) { 1010 case MODULE_CMD_INIT: 1011 case MODULE_CMD_FINI: 1012 return 0; 1013 case MODULE_CMD_STAT: 1014 case MODULE_CMD_AUTOUNLOAD: 1015 default: 1016 return ENOTTY; 1017 } 1018 } 1019