1 /* $NetBSD: dksubr.c,v 1.113 2021/04/15 00:32:50 rin 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 1.113 2021/04/15 00:32:50 rin 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 || error == ENOMEM) { 432 /* 433 * Not a disk error. Retry later. 434 */ 435 KASSERT(dksc->sc_deferred == NULL); 436 dksc->sc_deferred = bp; 437 disk_unbusy(&dksc->sc_dkdev, 0, (bp->b_flags & B_READ)); 438 disk_wait(&dksc->sc_dkdev); 439 break; 440 } 441 442 if (error != 0) { 443 bp->b_error = error; 444 bp->b_resid = bp->b_bcount; 445 dk_done1(dksc, bp, false); 446 } 447 448 bp = bufq_get(dksc->sc_bufq); 449 } 450 451 dksc->sc_busy--; 452 } 453 done: 454 mutex_exit(&dksc->sc_iolock); 455 } 456 457 static void 458 dk_done1(struct dk_softc *dksc, struct buf *bp, bool lock) 459 { 460 struct disk *dk = &dksc->sc_dkdev; 461 462 if (bp->b_error != 0) { 463 struct cfdriver *cd = device_cfdriver(dksc->sc_dev); 464 465 diskerr(bp, cd->cd_name, "error", LOG_PRINTF, 0, 466 dk->dk_label); 467 printf("\n"); 468 } 469 470 if (lock) 471 mutex_enter(&dksc->sc_iolock); 472 disk_unbusy(dk, bp->b_bcount - bp->b_resid, (bp->b_flags & B_READ)); 473 474 if ((dksc->sc_flags & DKF_NO_RND) == 0) 475 rnd_add_uint32(&dksc->sc_rnd_source, bp->b_rawblkno); 476 if (lock) 477 mutex_exit(&dksc->sc_iolock); 478 479 biodone(bp); 480 } 481 482 void 483 dk_done(struct dk_softc *dksc, struct buf *bp) 484 { 485 dk_done1(dksc, bp, true); 486 } 487 488 void 489 dk_drain(struct dk_softc *dksc) 490 { 491 struct buf *bp; 492 493 mutex_enter(&dksc->sc_iolock); 494 bp = dksc->sc_deferred; 495 dksc->sc_deferred = NULL; 496 if (bp != NULL) { 497 bp->b_error = EIO; 498 bp->b_resid = bp->b_bcount; 499 biodone(bp); 500 } 501 bufq_drain(dksc->sc_bufq); 502 mutex_exit(&dksc->sc_iolock); 503 } 504 505 int 506 dk_discard(struct dk_softc *dksc, dev_t dev, off_t pos, off_t len) 507 { 508 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 509 unsigned secsize = dksc->sc_dkdev.dk_geom.dg_secsize; 510 struct buf tmp, *bp = &tmp; 511 int maxsz; 512 int error = 0; 513 514 KASSERT(len >= 0); 515 516 DPRINTF_FOLLOW(("%s(%s, %p, 0x"PRIx64", %jd, %jd)\n", __func__, 517 dksc->sc_xname, dksc, (intmax_t)pos, (intmax_t)len)); 518 519 if (!(dksc->sc_flags & DKF_INITED)) { 520 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 521 return ENXIO; 522 } 523 524 if (secsize == 0 || (pos % secsize) != 0 || (len % secsize) != 0) 525 return EINVAL; 526 527 /* largest value that b_bcount can store */ 528 maxsz = rounddown(INT_MAX, secsize); 529 530 while (len > 0) { 531 /* enough data to please the bounds checking code */ 532 bp->b_dev = dev; 533 bp->b_blkno = (daddr_t)(pos / secsize); 534 bp->b_bcount = uimin(len, maxsz); 535 bp->b_flags = B_WRITE; 536 537 error = dk_translate(dksc, bp); 538 if (error >= 0) 539 break; 540 541 error = dkd->d_discard(dksc->sc_dev, 542 (off_t)bp->b_rawblkno * secsize, 543 (off_t)bp->b_bcount); 544 if (error) 545 break; 546 547 pos += bp->b_bcount; 548 len -= bp->b_bcount; 549 } 550 551 return error; 552 } 553 554 int 555 dk_size(struct dk_softc *dksc, dev_t dev) 556 { 557 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 558 struct disklabel *lp; 559 int is_open; 560 int part; 561 int size; 562 563 if ((dksc->sc_flags & DKF_INITED) == 0) 564 return -1; 565 566 part = DISKPART(dev); 567 is_open = dksc->sc_dkdev.dk_openmask & (1 << part); 568 569 if (!is_open && dkd->d_open(dev, 0, S_IFBLK, curlwp)) 570 return -1; 571 572 lp = dksc->sc_dkdev.dk_label; 573 if (lp->d_partitions[part].p_fstype != FS_SWAP) 574 size = -1; 575 else 576 size = lp->d_partitions[part].p_size * 577 (lp->d_secsize / DEV_BSIZE); 578 579 if (!is_open && dkd->d_close(dev, 0, S_IFBLK, curlwp)) 580 return -1; 581 582 return size; 583 } 584 585 int 586 dk_ioctl(struct dk_softc *dksc, dev_t dev, 587 u_long cmd, void *data, int flag, struct lwp *l) 588 { 589 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 590 struct disklabel *lp; 591 struct disk *dk = &dksc->sc_dkdev; 592 #ifdef __HAVE_OLD_DISKLABEL 593 struct disklabel newlabel; 594 #endif 595 int error; 596 597 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%lx)\n", __func__, 598 dksc->sc_xname, dksc, dev, cmd)); 599 600 /* ensure that the pseudo disk is open for writes for these commands */ 601 switch (cmd) { 602 case DIOCSDINFO: 603 case DIOCWDINFO: 604 #ifdef __HAVE_OLD_DISKLABEL 605 case ODIOCSDINFO: 606 case ODIOCWDINFO: 607 #endif 608 case DIOCKLABEL: 609 case DIOCWLABEL: 610 case DIOCAWEDGE: 611 case DIOCDWEDGE: 612 case DIOCSSTRATEGY: 613 if ((flag & FWRITE) == 0) 614 return EBADF; 615 } 616 617 /* ensure that the pseudo-disk is initialized for these */ 618 switch (cmd) { 619 case DIOCGDINFO: 620 case DIOCSDINFO: 621 case DIOCWDINFO: 622 case DIOCGPARTINFO: 623 case DIOCKLABEL: 624 case DIOCWLABEL: 625 case DIOCGDEFLABEL: 626 case DIOCAWEDGE: 627 case DIOCDWEDGE: 628 case DIOCLWEDGES: 629 case DIOCMWEDGES: 630 case DIOCRMWEDGES: 631 case DIOCCACHESYNC: 632 #ifdef __HAVE_OLD_DISKLABEL 633 case ODIOCGDINFO: 634 case ODIOCSDINFO: 635 case ODIOCWDINFO: 636 case ODIOCGDEFLABEL: 637 #endif 638 if ((dksc->sc_flags & DKF_INITED) == 0) 639 return ENXIO; 640 } 641 642 error = disk_ioctl(dk, dev, cmd, data, flag, l); 643 if (error != EPASSTHROUGH) 644 return error; 645 else 646 error = 0; 647 648 switch (cmd) { 649 case DIOCWDINFO: 650 case DIOCSDINFO: 651 #ifdef __HAVE_OLD_DISKLABEL 652 case ODIOCWDINFO: 653 case ODIOCSDINFO: 654 #endif 655 #ifdef __HAVE_OLD_DISKLABEL 656 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 657 memset(&newlabel, 0, sizeof newlabel); 658 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 659 lp = &newlabel; 660 } else 661 #endif 662 lp = (struct disklabel *)data; 663 664 mutex_enter(&dk->dk_openlock); 665 dksc->sc_flags |= DKF_LABELLING; 666 667 error = setdisklabel(dksc->sc_dkdev.dk_label, 668 lp, 0, dksc->sc_dkdev.dk_cpulabel); 669 if (error == 0) { 670 if (cmd == DIOCWDINFO 671 #ifdef __HAVE_OLD_DISKLABEL 672 || cmd == ODIOCWDINFO 673 #endif 674 ) 675 error = writedisklabel(DKLABELDEV(dev), 676 dkd->d_strategy, dksc->sc_dkdev.dk_label, 677 dksc->sc_dkdev.dk_cpulabel); 678 } 679 680 dksc->sc_flags &= ~DKF_LABELLING; 681 mutex_exit(&dk->dk_openlock); 682 break; 683 684 case DIOCKLABEL: 685 if (*(int *)data != 0) 686 dksc->sc_flags |= DKF_KLABEL; 687 else 688 dksc->sc_flags &= ~DKF_KLABEL; 689 break; 690 691 case DIOCWLABEL: 692 if (*(int *)data != 0) 693 dksc->sc_flags |= DKF_WLABEL; 694 else 695 dksc->sc_flags &= ~DKF_WLABEL; 696 break; 697 698 case DIOCGDEFLABEL: 699 dk_getdefaultlabel(dksc, (struct disklabel *)data); 700 break; 701 702 #ifdef __HAVE_OLD_DISKLABEL 703 case ODIOCGDEFLABEL: 704 dk_getdefaultlabel(dksc, &newlabel); 705 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 706 return ENOTTY; 707 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 708 break; 709 #endif 710 711 case DIOCGSTRATEGY: 712 { 713 struct disk_strategy *dks = (void *)data; 714 715 mutex_enter(&dksc->sc_iolock); 716 if (dksc->sc_bufq != NULL) 717 strlcpy(dks->dks_name, 718 bufq_getstrategyname(dksc->sc_bufq), 719 sizeof(dks->dks_name)); 720 else 721 error = EINVAL; 722 mutex_exit(&dksc->sc_iolock); 723 dks->dks_paramlen = 0; 724 break; 725 } 726 727 case DIOCSSTRATEGY: 728 { 729 struct disk_strategy *dks = (void *)data; 730 struct bufq_state *new; 731 struct bufq_state *old; 732 733 if (dks->dks_param != NULL) { 734 return EINVAL; 735 } 736 dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */ 737 error = bufq_alloc(&new, dks->dks_name, 738 BUFQ_EXACT|BUFQ_SORT_RAWBLOCK); 739 if (error) { 740 return error; 741 } 742 mutex_enter(&dksc->sc_iolock); 743 old = dksc->sc_bufq; 744 if (old) 745 bufq_move(new, old); 746 dksc->sc_bufq = new; 747 mutex_exit(&dksc->sc_iolock); 748 if (old) 749 bufq_free(old); 750 break; 751 } 752 753 default: 754 error = ENOTTY; 755 } 756 757 return error; 758 } 759 760 /* 761 * dk_dump dumps all of physical memory into the partition specified. 762 * This requires substantially more framework than {s,w}ddump, and hence 763 * is probably much more fragile. 764 * 765 */ 766 767 #define DKFF_READYFORDUMP(x) (((x) & DKF_READYFORDUMP) == DKF_READYFORDUMP) 768 static volatile int dk_dumping = 0; 769 770 /* ARGSUSED */ 771 int 772 dk_dump(struct dk_softc *dksc, dev_t dev, 773 daddr_t blkno, void *vav, size_t size, int flags) 774 { 775 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 776 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 777 char *va = vav; 778 struct disklabel *lp; 779 struct partition *p; 780 int part, towrt, maxblkcnt, nblk; 781 int maxxfer, rv = 0; 782 783 /* 784 * ensure that we consider this device to be safe for dumping, 785 * and that the device is configured. 786 */ 787 if (!DKFF_READYFORDUMP(dksc->sc_flags)) { 788 DPRINTF(DKDB_DUMP, ("%s: bad dump flags 0x%x\n", __func__, 789 dksc->sc_flags)); 790 return ENXIO; 791 } 792 793 /* ensure that we are not already dumping */ 794 if (dk_dumping) 795 return EFAULT; 796 if ((flags & DK_DUMP_RECURSIVE) == 0) 797 dk_dumping = 1; 798 799 if (dkd->d_dumpblocks == NULL) { 800 DPRINTF(DKDB_DUMP, ("%s: no dumpblocks\n", __func__)); 801 return ENXIO; 802 } 803 804 /* device specific max transfer size */ 805 maxxfer = MAXPHYS; 806 if (dkd->d_iosize != NULL) 807 (*dkd->d_iosize)(dksc->sc_dev, &maxxfer); 808 809 /* Convert to disk sectors. Request must be a multiple of size. */ 810 part = DISKPART(dev); 811 lp = dksc->sc_dkdev.dk_label; 812 if ((size % lp->d_secsize) != 0) { 813 DPRINTF(DKDB_DUMP, ("%s: odd size %zu\n", __func__, size)); 814 return EFAULT; 815 } 816 towrt = size / lp->d_secsize; 817 blkno = dbtob(blkno) / lp->d_secsize; /* blkno in secsize units */ 818 819 p = &lp->d_partitions[part]; 820 if (part == RAW_PART) { 821 if (p->p_fstype != FS_UNUSED) { 822 DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__, 823 p->p_fstype)); 824 return ENXIO; 825 } 826 /* Check whether dump goes to a wedge */ 827 if (dksc->sc_dkdev.dk_nwedges == 0) { 828 DPRINTF(DKDB_DUMP, ("%s: dump to raw\n", __func__)); 829 return ENXIO; 830 } 831 /* Check transfer bounds against media size */ 832 if (blkno < 0 || (blkno + towrt) > dg->dg_secperunit) { 833 DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, " 834 "nsects=%jd\n", __func__, (intmax_t)blkno, towrt, dg->dg_secperunit)); 835 return EINVAL; 836 } 837 } else { 838 int nsects, sectoff; 839 840 if (p->p_fstype != FS_SWAP) { 841 DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__, 842 p->p_fstype)); 843 return ENXIO; 844 } 845 nsects = p->p_size; 846 sectoff = p->p_offset; 847 848 /* Check transfer bounds against partition size. */ 849 if ((blkno < 0) || ((blkno + towrt) > nsects)) { 850 DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, " 851 "nsects=%d\n", __func__, (intmax_t)blkno, towrt, nsects)); 852 return EINVAL; 853 } 854 855 /* Offset block number to start of partition. */ 856 blkno += sectoff; 857 } 858 859 /* Start dumping and return when done. */ 860 maxblkcnt = howmany(maxxfer, lp->d_secsize); 861 while (towrt > 0) { 862 nblk = uimin(maxblkcnt, towrt); 863 864 if ((rv = (*dkd->d_dumpblocks)(dksc->sc_dev, va, blkno, nblk)) 865 != 0) { 866 DPRINTF(DKDB_DUMP, ("%s: dumpblocks %d\n", __func__, 867 rv)); 868 return rv; 869 } 870 871 towrt -= nblk; 872 blkno += nblk; 873 va += nblk * lp->d_secsize; 874 } 875 876 if ((flags & DK_DUMP_RECURSIVE) == 0) 877 dk_dumping = 0; 878 879 return 0; 880 } 881 882 /* ARGSUSED */ 883 void 884 dk_getdefaultlabel(struct dk_softc *dksc, struct disklabel *lp) 885 { 886 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 887 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 888 889 memset(lp, 0, sizeof(*lp)); 890 891 if (dg->dg_secperunit > UINT32_MAX) 892 lp->d_secperunit = UINT32_MAX; 893 else 894 lp->d_secperunit = dg->dg_secperunit; 895 lp->d_secsize = dg->dg_secsize; 896 lp->d_nsectors = dg->dg_nsectors; 897 lp->d_ntracks = dg->dg_ntracks; 898 lp->d_ncylinders = dg->dg_ncylinders; 899 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 900 901 strlcpy(lp->d_typename, dksc->sc_xname, sizeof(lp->d_typename)); 902 lp->d_type = dksc->sc_dtype; 903 strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 904 lp->d_rpm = 3600; 905 lp->d_interleave = 1; 906 lp->d_flags = 0; 907 908 lp->d_partitions[RAW_PART].p_offset = 0; 909 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit; 910 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 911 lp->d_npartitions = RAW_PART + 1; 912 913 lp->d_magic = DISKMAGIC; 914 lp->d_magic2 = DISKMAGIC; 915 916 if (dkd->d_label) 917 dkd->d_label(dksc->sc_dev, lp); 918 919 lp->d_checksum = dkcksum(lp); 920 } 921 922 /* ARGSUSED */ 923 void 924 dk_getdisklabel(struct dk_softc *dksc, dev_t dev) 925 { 926 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 927 struct disklabel *lp = dksc->sc_dkdev.dk_label; 928 struct cpu_disklabel *clp = dksc->sc_dkdev.dk_cpulabel; 929 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 930 struct partition *pp; 931 int i, lpratio, dgratio; 932 const char *errstring; 933 934 memset(clp, 0x0, sizeof(*clp)); 935 dk_getdefaultlabel(dksc, lp); 936 errstring = readdisklabel(DKLABELDEV(dev), dkd->d_strategy, 937 dksc->sc_dkdev.dk_label, dksc->sc_dkdev.dk_cpulabel); 938 if (errstring) { 939 dk_makedisklabel(dksc); 940 if (dksc->sc_flags & DKF_WARNLABEL) 941 printf("%s: %s\n", dksc->sc_xname, errstring); 942 return; 943 } 944 945 if ((dksc->sc_flags & DKF_LABELSANITY) == 0) 946 return; 947 948 /* Convert sector counts to multiple of DEV_BSIZE for comparison */ 949 lpratio = dgratio = 1; 950 if (lp->d_secsize > DEV_BSIZE) 951 lpratio = lp->d_secsize / DEV_BSIZE; 952 if (dg->dg_secsize > DEV_BSIZE) 953 dgratio = dg->dg_secsize / DEV_BSIZE; 954 955 /* Sanity check */ 956 if ((uint64_t)lp->d_secperunit * lpratio > dg->dg_secperunit * dgratio) 957 printf("WARNING: %s: " 958 "total unit size in disklabel (%" PRIu64 ") " 959 "!= the size of %s (%" PRIu64 ")\n", dksc->sc_xname, 960 (uint64_t)lp->d_secperunit * lpratio, dksc->sc_xname, 961 dg->dg_secperunit * dgratio); 962 else if (lp->d_secperunit < UINT32_MAX && 963 (uint64_t)lp->d_secperunit * lpratio < dg->dg_secperunit * dgratio) 964 printf("%s: %" PRIu64 " trailing sectors not covered" 965 " by disklabel\n", dksc->sc_xname, 966 (dg->dg_secperunit * dgratio) 967 - (lp->d_secperunit * lpratio)); 968 969 for (i=0; i < lp->d_npartitions; i++) { 970 uint64_t pend; 971 972 pp = &lp->d_partitions[i]; 973 pend = pp->p_offset + pp->p_size; 974 if (pend * lpratio > dg->dg_secperunit * dgratio) 975 printf("WARNING: %s: end of partition `%c' exceeds " 976 "the size of %s (%" PRIu64 ")\n", dksc->sc_xname, 977 'a' + i, dksc->sc_xname, 978 dg->dg_secperunit * dgratio); 979 } 980 } 981 982 /* 983 * Heuristic to conjure a disklabel if reading a disklabel failed. 984 * 985 * This is to allow the raw partition to be used for a filesystem 986 * without caring about the write protected label sector. 987 * 988 * If the driver provides it's own callback, use that instead. 989 */ 990 /* ARGSUSED */ 991 static void 992 dk_makedisklabel(struct dk_softc *dksc) 993 { 994 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 995 struct disklabel *lp = dksc->sc_dkdev.dk_label; 996 997 strlcpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 998 999 if (dkd->d_label) 1000 dkd->d_label(dksc->sc_dev, lp); 1001 else 1002 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 1003 1004 lp->d_checksum = dkcksum(lp); 1005 } 1006 1007 MODULE(MODULE_CLASS_MISC, dk_subr, NULL); 1008 1009 static int 1010 dk_subr_modcmd(modcmd_t cmd, void *arg) 1011 { 1012 switch (cmd) { 1013 case MODULE_CMD_INIT: 1014 case MODULE_CMD_FINI: 1015 return 0; 1016 case MODULE_CMD_STAT: 1017 case MODULE_CMD_AUTOUNLOAD: 1018 default: 1019 return ENOTTY; 1020 } 1021 } 1022