1 /* $NetBSD: dksubr.c,v 1.102 2018/05/12 10:33:06 mlelstv 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 1.102 2018/05/12 10:33:06 mlelstv 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 /* 151 * initialize driver for the first opener 152 */ 153 if (dk->dk_openmask == 0 && dkd->d_firstopen != NULL) { 154 ret = (*dkd->d_firstopen)(dksc->sc_dev, dev, flags, fmt); 155 if (ret) 156 goto done; 157 } 158 159 /* 160 * If we're init'ed and there are no other open partitions then 161 * update the in-core disklabel. 162 */ 163 if ((dksc->sc_flags & DKF_INITED)) { 164 if ((dksc->sc_flags & DKF_VLABEL) == 0) { 165 dksc->sc_flags |= DKF_VLABEL; 166 dk_getdisklabel(dksc, dev); 167 } 168 } 169 170 /* Fail if we can't find the partition. */ 171 if (part != RAW_PART && 172 ((dksc->sc_flags & DKF_VLABEL) == 0 || 173 part >= lp->d_npartitions || 174 lp->d_partitions[part].p_fstype == FS_UNUSED)) { 175 ret = ENXIO; 176 goto done; 177 } 178 179 /* Mark our unit as open. */ 180 switch (fmt) { 181 case S_IFCHR: 182 dk->dk_copenmask |= pmask; 183 break; 184 case S_IFBLK: 185 dk->dk_bopenmask |= pmask; 186 break; 187 } 188 189 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask; 190 191 done: 192 mutex_exit(&dk->dk_openlock); 193 return ret; 194 } 195 196 /* ARGSUSED */ 197 int 198 dk_close(struct dk_softc *dksc, dev_t dev, 199 int flags, int fmt, struct lwp *l) 200 { 201 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 202 int part = DISKPART(dev); 203 int pmask = 1 << part; 204 struct disk *dk = &dksc->sc_dkdev; 205 206 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%x)\n", __func__, 207 dksc->sc_xname, dksc, dev, flags)); 208 209 mutex_enter(&dk->dk_openlock); 210 211 switch (fmt) { 212 case S_IFCHR: 213 dk->dk_copenmask &= ~pmask; 214 break; 215 case S_IFBLK: 216 dk->dk_bopenmask &= ~pmask; 217 break; 218 } 219 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask; 220 221 if (dk->dk_openmask == 0) { 222 if (dkd->d_lastclose != NULL) 223 (*dkd->d_lastclose)(dksc->sc_dev); 224 if ((dksc->sc_flags & DKF_KLABEL) == 0) 225 dksc->sc_flags &= ~DKF_VLABEL; 226 } 227 228 mutex_exit(&dk->dk_openlock); 229 return 0; 230 } 231 232 static int 233 dk_translate(struct dk_softc *dksc, struct buf *bp) 234 { 235 int part; 236 int wlabel; 237 daddr_t blkno; 238 struct disklabel *lp; 239 struct disk *dk; 240 uint64_t numsecs; 241 unsigned secsize; 242 243 lp = dksc->sc_dkdev.dk_label; 244 dk = &dksc->sc_dkdev; 245 246 part = DISKPART(bp->b_dev); 247 numsecs = dk->dk_geom.dg_secperunit; 248 secsize = dk->dk_geom.dg_secsize; 249 250 /* 251 * The transfer must be a whole number of blocks and the offset must 252 * not be negative. 253 */ 254 if ((bp->b_bcount % secsize) != 0 || bp->b_blkno < 0) { 255 bp->b_error = EINVAL; 256 goto done; 257 } 258 259 /* If there is nothing to do, then we are done */ 260 if (bp->b_bcount == 0) 261 goto done; 262 263 wlabel = dksc->sc_flags & (DKF_WLABEL|DKF_LABELLING); 264 if (part == RAW_PART) { 265 uint64_t numblocks = btodb(numsecs * secsize); 266 if (bounds_check_with_mediasize(bp, DEV_BSIZE, numblocks) <= 0) 267 goto done; 268 } else { 269 if (bounds_check_with_label(&dksc->sc_dkdev, bp, wlabel) <= 0) 270 goto done; 271 } 272 273 /* 274 * Convert the block number to absolute and put it in terms 275 * of the device's logical block size. 276 */ 277 if (secsize >= DEV_BSIZE) 278 blkno = bp->b_blkno / (secsize / DEV_BSIZE); 279 else 280 blkno = bp->b_blkno * (DEV_BSIZE / secsize); 281 282 if (part != RAW_PART) 283 blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset; 284 bp->b_rawblkno = blkno; 285 286 return -1; 287 288 done: 289 bp->b_resid = bp->b_bcount; 290 return bp->b_error; 291 } 292 293 static int 294 dk_strategy1(struct dk_softc *dksc, struct buf *bp) 295 { 296 int error; 297 298 DPRINTF_FOLLOW(("%s(%s, %p, %p)\n", __func__, 299 dksc->sc_xname, dksc, bp)); 300 301 if (!(dksc->sc_flags & DKF_INITED)) { 302 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 303 bp->b_error = ENXIO; 304 bp->b_resid = bp->b_bcount; 305 biodone(bp); 306 return 1; 307 } 308 309 error = dk_translate(dksc, bp); 310 if (error >= 0) { 311 biodone(bp); 312 return 1; 313 } 314 315 return 0; 316 } 317 318 void 319 dk_strategy(struct dk_softc *dksc, struct buf *bp) 320 { 321 int error; 322 323 error = dk_strategy1(dksc, bp); 324 if (error) 325 return; 326 327 /* 328 * Queue buffer and start unit 329 */ 330 dk_start(dksc, bp); 331 } 332 333 int 334 dk_strategy_defer(struct dk_softc *dksc, struct buf *bp) 335 { 336 int error; 337 338 error = dk_strategy1(dksc, bp); 339 if (error) 340 return error; 341 342 /* 343 * Queue buffer only 344 */ 345 mutex_enter(&dksc->sc_iolock); 346 disk_wait(&dksc->sc_dkdev); 347 bufq_put(dksc->sc_bufq, bp); 348 mutex_exit(&dksc->sc_iolock); 349 350 return 0; 351 } 352 353 int 354 dk_strategy_pending(struct dk_softc *dksc) 355 { 356 struct buf *bp; 357 358 if (!(dksc->sc_flags & DKF_INITED)) { 359 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 360 return 0; 361 } 362 363 mutex_enter(&dksc->sc_iolock); 364 bp = bufq_peek(dksc->sc_bufq); 365 mutex_exit(&dksc->sc_iolock); 366 367 return bp != NULL; 368 } 369 370 void 371 dk_start(struct dk_softc *dksc, struct buf *bp) 372 { 373 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 374 int error; 375 376 if (!(dksc->sc_flags & DKF_INITED)) { 377 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 378 return; 379 } 380 381 mutex_enter(&dksc->sc_iolock); 382 383 if (bp != NULL) { 384 disk_wait(&dksc->sc_dkdev); 385 bufq_put(dksc->sc_bufq, bp); 386 } 387 388 /* 389 * If another thread is running the queue, increment 390 * busy counter to 2 so that the queue is retried, 391 * because the driver may now accept additional 392 * requests. 393 */ 394 if (dksc->sc_busy < 2) 395 dksc->sc_busy++; 396 if (dksc->sc_busy > 1) 397 goto done; 398 399 /* 400 * Peeking at the buffer queue and committing the operation 401 * only after success isn't atomic. 402 * 403 * So when a diskstart fails, the buffer is saved 404 * and tried again before the next buffer is fetched. 405 * dk_drain() handles flushing of a saved buffer. 406 * 407 * This keeps order of I/O operations, unlike bufq_put. 408 */ 409 410 while (dksc->sc_busy > 0) { 411 412 bp = dksc->sc_deferred; 413 dksc->sc_deferred = NULL; 414 415 if (bp == NULL) 416 bp = bufq_get(dksc->sc_bufq); 417 418 while (bp != NULL) { 419 420 disk_busy(&dksc->sc_dkdev); 421 mutex_exit(&dksc->sc_iolock); 422 error = dkd->d_diskstart(dksc->sc_dev, bp); 423 mutex_enter(&dksc->sc_iolock); 424 if (error == EAGAIN) { 425 KASSERT(dksc->sc_deferred == NULL); 426 dksc->sc_deferred = bp; 427 disk_unbusy(&dksc->sc_dkdev, 0, (bp->b_flags & B_READ)); 428 disk_wait(&dksc->sc_dkdev); 429 break; 430 } 431 432 if (error != 0) { 433 bp->b_error = error; 434 bp->b_resid = bp->b_bcount; 435 dk_done1(dksc, bp, false); 436 } 437 438 bp = bufq_get(dksc->sc_bufq); 439 } 440 441 dksc->sc_busy--; 442 } 443 done: 444 mutex_exit(&dksc->sc_iolock); 445 } 446 447 static void 448 dk_done1(struct dk_softc *dksc, struct buf *bp, bool lock) 449 { 450 struct disk *dk = &dksc->sc_dkdev; 451 452 if (bp->b_error != 0) { 453 struct cfdriver *cd = device_cfdriver(dksc->sc_dev); 454 455 diskerr(bp, cd->cd_name, "error", LOG_PRINTF, 0, 456 dk->dk_label); 457 printf("\n"); 458 } 459 460 if (lock) 461 mutex_enter(&dksc->sc_iolock); 462 disk_unbusy(dk, bp->b_bcount - bp->b_resid, (bp->b_flags & B_READ)); 463 if (lock) 464 mutex_exit(&dksc->sc_iolock); 465 466 if ((dksc->sc_flags & DKF_NO_RND) == 0) 467 rnd_add_uint32(&dksc->sc_rnd_source, bp->b_rawblkno); 468 469 biodone(bp); 470 } 471 472 void 473 dk_done(struct dk_softc *dksc, struct buf *bp) 474 { 475 dk_done1(dksc, bp, true); 476 } 477 478 void 479 dk_drain(struct dk_softc *dksc) 480 { 481 struct buf *bp; 482 483 mutex_enter(&dksc->sc_iolock); 484 bp = dksc->sc_deferred; 485 dksc->sc_deferred = NULL; 486 if (bp != NULL) { 487 bp->b_error = EIO; 488 bp->b_resid = bp->b_bcount; 489 biodone(bp); 490 } 491 bufq_drain(dksc->sc_bufq); 492 mutex_exit(&dksc->sc_iolock); 493 } 494 495 int 496 dk_discard(struct dk_softc *dksc, dev_t dev, off_t pos, off_t len) 497 { 498 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 499 unsigned secsize = dksc->sc_dkdev.dk_geom.dg_secsize; 500 struct buf tmp, *bp = &tmp; 501 int maxsz; 502 int error = 0; 503 504 KASSERT(len >= 0); 505 506 DPRINTF_FOLLOW(("%s(%s, %p, 0x"PRIx64", %jd, %jd)\n", __func__, 507 dksc->sc_xname, dksc, (intmax_t)pos, (intmax_t)len)); 508 509 if (!(dksc->sc_flags & DKF_INITED)) { 510 DPRINTF_FOLLOW(("%s: not inited\n", __func__)); 511 return ENXIO; 512 } 513 514 if (secsize == 0 || (pos % secsize) != 0 || (len % secsize) != 0) 515 return EINVAL; 516 517 /* largest value that b_bcount can store */ 518 maxsz = rounddown(INT_MAX, secsize); 519 520 while (len > 0) { 521 /* enough data to please the bounds checking code */ 522 bp->b_dev = dev; 523 bp->b_blkno = (daddr_t)(pos / secsize); 524 bp->b_bcount = min(len, maxsz); 525 bp->b_flags = B_WRITE; 526 527 error = dk_translate(dksc, bp); 528 if (error >= 0) 529 break; 530 531 error = dkd->d_discard(dksc->sc_dev, 532 (off_t)bp->b_rawblkno * secsize, 533 (off_t)bp->b_bcount); 534 if (error) 535 break; 536 537 pos += bp->b_bcount; 538 len -= bp->b_bcount; 539 } 540 541 return error; 542 } 543 544 int 545 dk_size(struct dk_softc *dksc, dev_t dev) 546 { 547 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 548 struct disklabel *lp; 549 int is_open; 550 int part; 551 int size; 552 553 if ((dksc->sc_flags & DKF_INITED) == 0) 554 return -1; 555 556 part = DISKPART(dev); 557 is_open = dksc->sc_dkdev.dk_openmask & (1 << part); 558 559 if (!is_open && dkd->d_open(dev, 0, S_IFBLK, curlwp)) 560 return -1; 561 562 lp = dksc->sc_dkdev.dk_label; 563 if (lp->d_partitions[part].p_fstype != FS_SWAP) 564 size = -1; 565 else 566 size = lp->d_partitions[part].p_size * 567 (lp->d_secsize / DEV_BSIZE); 568 569 if (!is_open && dkd->d_close(dev, 0, S_IFBLK, curlwp)) 570 return -1; 571 572 return size; 573 } 574 575 int 576 dk_ioctl(struct dk_softc *dksc, dev_t dev, 577 u_long cmd, void *data, int flag, struct lwp *l) 578 { 579 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 580 struct disklabel *lp; 581 struct disk *dk = &dksc->sc_dkdev; 582 #ifdef __HAVE_OLD_DISKLABEL 583 struct disklabel newlabel; 584 #endif 585 int error; 586 587 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%lx)\n", __func__, 588 dksc->sc_xname, dksc, dev, cmd)); 589 590 /* ensure that the pseudo disk is open for writes for these commands */ 591 switch (cmd) { 592 case DIOCSDINFO: 593 case DIOCWDINFO: 594 #ifdef __HAVE_OLD_DISKLABEL 595 case ODIOCSDINFO: 596 case ODIOCWDINFO: 597 #endif 598 case DIOCKLABEL: 599 case DIOCWLABEL: 600 case DIOCAWEDGE: 601 case DIOCDWEDGE: 602 case DIOCSSTRATEGY: 603 if ((flag & FWRITE) == 0) 604 return EBADF; 605 } 606 607 /* ensure that the pseudo-disk is initialized for these */ 608 switch (cmd) { 609 case DIOCGDINFO: 610 case DIOCSDINFO: 611 case DIOCWDINFO: 612 case DIOCGPARTINFO: 613 case DIOCKLABEL: 614 case DIOCWLABEL: 615 case DIOCGDEFLABEL: 616 case DIOCAWEDGE: 617 case DIOCDWEDGE: 618 case DIOCLWEDGES: 619 case DIOCMWEDGES: 620 case DIOCCACHESYNC: 621 #ifdef __HAVE_OLD_DISKLABEL 622 case ODIOCGDINFO: 623 case ODIOCSDINFO: 624 case ODIOCWDINFO: 625 case ODIOCGDEFLABEL: 626 #endif 627 if ((dksc->sc_flags & DKF_INITED) == 0) 628 return ENXIO; 629 } 630 631 error = disk_ioctl(dk, dev, cmd, data, flag, l); 632 if (error != EPASSTHROUGH) 633 return error; 634 else 635 error = 0; 636 637 switch (cmd) { 638 case DIOCWDINFO: 639 case DIOCSDINFO: 640 #ifdef __HAVE_OLD_DISKLABEL 641 case ODIOCWDINFO: 642 case ODIOCSDINFO: 643 #endif 644 #ifdef __HAVE_OLD_DISKLABEL 645 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 646 memset(&newlabel, 0, sizeof newlabel); 647 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 648 lp = &newlabel; 649 } else 650 #endif 651 lp = (struct disklabel *)data; 652 653 mutex_enter(&dk->dk_openlock); 654 dksc->sc_flags |= DKF_LABELLING; 655 656 error = setdisklabel(dksc->sc_dkdev.dk_label, 657 lp, 0, dksc->sc_dkdev.dk_cpulabel); 658 if (error == 0) { 659 if (cmd == DIOCWDINFO 660 #ifdef __HAVE_OLD_DISKLABEL 661 || cmd == ODIOCWDINFO 662 #endif 663 ) 664 error = writedisklabel(DKLABELDEV(dev), 665 dkd->d_strategy, dksc->sc_dkdev.dk_label, 666 dksc->sc_dkdev.dk_cpulabel); 667 } 668 669 dksc->sc_flags &= ~DKF_LABELLING; 670 mutex_exit(&dk->dk_openlock); 671 break; 672 673 case DIOCKLABEL: 674 if (*(int *)data != 0) 675 dksc->sc_flags |= DKF_KLABEL; 676 else 677 dksc->sc_flags &= ~DKF_KLABEL; 678 break; 679 680 case DIOCWLABEL: 681 if (*(int *)data != 0) 682 dksc->sc_flags |= DKF_WLABEL; 683 else 684 dksc->sc_flags &= ~DKF_WLABEL; 685 break; 686 687 case DIOCGDEFLABEL: 688 dk_getdefaultlabel(dksc, (struct disklabel *)data); 689 break; 690 691 #ifdef __HAVE_OLD_DISKLABEL 692 case ODIOCGDEFLABEL: 693 dk_getdefaultlabel(dksc, &newlabel); 694 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 695 return ENOTTY; 696 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 697 break; 698 #endif 699 700 case DIOCGSTRATEGY: 701 { 702 struct disk_strategy *dks = (void *)data; 703 704 mutex_enter(&dksc->sc_iolock); 705 if (dksc->sc_bufq != NULL) 706 strlcpy(dks->dks_name, 707 bufq_getstrategyname(dksc->sc_bufq), 708 sizeof(dks->dks_name)); 709 else 710 error = EINVAL; 711 mutex_exit(&dksc->sc_iolock); 712 dks->dks_paramlen = 0; 713 break; 714 } 715 716 case DIOCSSTRATEGY: 717 { 718 struct disk_strategy *dks = (void *)data; 719 struct bufq_state *new; 720 struct bufq_state *old; 721 722 if (dks->dks_param != NULL) { 723 return EINVAL; 724 } 725 dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */ 726 error = bufq_alloc(&new, dks->dks_name, 727 BUFQ_EXACT|BUFQ_SORT_RAWBLOCK); 728 if (error) { 729 return error; 730 } 731 mutex_enter(&dksc->sc_iolock); 732 old = dksc->sc_bufq; 733 if (old) 734 bufq_move(new, old); 735 dksc->sc_bufq = new; 736 mutex_exit(&dksc->sc_iolock); 737 if (old) 738 bufq_free(old); 739 break; 740 } 741 742 default: 743 error = ENOTTY; 744 } 745 746 return error; 747 } 748 749 /* 750 * dk_dump dumps all of physical memory into the partition specified. 751 * This requires substantially more framework than {s,w}ddump, and hence 752 * is probably much more fragile. 753 * 754 */ 755 756 #define DKFF_READYFORDUMP(x) (((x) & DKF_READYFORDUMP) == DKF_READYFORDUMP) 757 static volatile int dk_dumping = 0; 758 759 /* ARGSUSED */ 760 int 761 dk_dump(struct dk_softc *dksc, dev_t dev, 762 daddr_t blkno, void *vav, size_t size) 763 { 764 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 765 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 766 char *va = vav; 767 struct disklabel *lp; 768 struct partition *p; 769 int part, towrt, maxblkcnt, nblk; 770 int maxxfer, rv = 0; 771 772 /* 773 * ensure that we consider this device to be safe for dumping, 774 * and that the device is configured. 775 */ 776 if (!DKFF_READYFORDUMP(dksc->sc_flags)) { 777 DPRINTF(DKDB_DUMP, ("%s: bad dump flags 0x%x\n", __func__, 778 dksc->sc_flags)); 779 return ENXIO; 780 } 781 782 /* ensure that we are not already dumping */ 783 if (dk_dumping) 784 return EFAULT; 785 dk_dumping = 1; 786 787 if (dkd->d_dumpblocks == NULL) { 788 DPRINTF(DKDB_DUMP, ("%s: no dumpblocks\n", __func__)); 789 return ENXIO; 790 } 791 792 /* device specific max transfer size */ 793 maxxfer = MAXPHYS; 794 if (dkd->d_iosize != NULL) 795 (*dkd->d_iosize)(dksc->sc_dev, &maxxfer); 796 797 /* Convert to disk sectors. Request must be a multiple of size. */ 798 part = DISKPART(dev); 799 lp = dksc->sc_dkdev.dk_label; 800 if ((size % lp->d_secsize) != 0) { 801 DPRINTF(DKDB_DUMP, ("%s: odd size %zu\n", __func__, size)); 802 return EFAULT; 803 } 804 towrt = size / lp->d_secsize; 805 blkno = dbtob(blkno) / lp->d_secsize; /* blkno in secsize units */ 806 807 p = &lp->d_partitions[part]; 808 if (part == RAW_PART) { 809 if (p->p_fstype != FS_UNUSED) { 810 DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__, 811 p->p_fstype)); 812 return ENXIO; 813 } 814 /* Check wether dump goes to a wedge */ 815 if (dksc->sc_dkdev.dk_nwedges == 0) { 816 DPRINTF(DKDB_DUMP, ("%s: dump to raw\n", __func__)); 817 return ENXIO; 818 } 819 /* Check transfer bounds against media size */ 820 if (blkno < 0 || (blkno + towrt) > dg->dg_secperunit) { 821 DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, " 822 "nsects=%jd\n", __func__, (intmax_t)blkno, towrt, dg->dg_secperunit)); 823 return EINVAL; 824 } 825 } else { 826 int nsects, sectoff; 827 828 if (p->p_fstype != FS_SWAP) { 829 DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__, 830 p->p_fstype)); 831 return ENXIO; 832 } 833 nsects = p->p_size; 834 sectoff = p->p_offset; 835 836 /* Check transfer bounds against partition size. */ 837 if ((blkno < 0) || ((blkno + towrt) > nsects)) { 838 DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, " 839 "nsects=%d\n", __func__, (intmax_t)blkno, towrt, nsects)); 840 return EINVAL; 841 } 842 843 /* Offset block number to start of partition. */ 844 blkno += sectoff; 845 } 846 847 /* Start dumping and return when done. */ 848 maxblkcnt = howmany(maxxfer, lp->d_secsize); 849 while (towrt > 0) { 850 nblk = min(maxblkcnt, towrt); 851 852 if ((rv = (*dkd->d_dumpblocks)(dksc->sc_dev, va, blkno, nblk)) 853 != 0) { 854 DPRINTF(DKDB_DUMP, ("%s: dumpblocks %d\n", __func__, 855 rv)); 856 return rv; 857 } 858 859 towrt -= nblk; 860 blkno += nblk; 861 va += nblk * lp->d_secsize; 862 } 863 864 dk_dumping = 0; 865 866 return 0; 867 } 868 869 /* ARGSUSED */ 870 void 871 dk_getdefaultlabel(struct dk_softc *dksc, struct disklabel *lp) 872 { 873 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 874 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 875 876 memset(lp, 0, sizeof(*lp)); 877 878 if (dg->dg_secperunit > UINT32_MAX) 879 lp->d_secperunit = UINT32_MAX; 880 else 881 lp->d_secperunit = dg->dg_secperunit; 882 lp->d_secsize = dg->dg_secsize; 883 lp->d_nsectors = dg->dg_nsectors; 884 lp->d_ntracks = dg->dg_ntracks; 885 lp->d_ncylinders = dg->dg_ncylinders; 886 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 887 888 strlcpy(lp->d_typename, dksc->sc_xname, sizeof(lp->d_typename)); 889 lp->d_type = dksc->sc_dtype; 890 strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 891 lp->d_rpm = 3600; 892 lp->d_interleave = 1; 893 lp->d_flags = 0; 894 895 lp->d_partitions[RAW_PART].p_offset = 0; 896 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit; 897 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 898 lp->d_npartitions = RAW_PART + 1; 899 900 lp->d_magic = DISKMAGIC; 901 lp->d_magic2 = DISKMAGIC; 902 903 if (dkd->d_label) 904 dkd->d_label(dksc->sc_dev, lp); 905 906 lp->d_checksum = dkcksum(lp); 907 } 908 909 /* ARGSUSED */ 910 void 911 dk_getdisklabel(struct dk_softc *dksc, dev_t dev) 912 { 913 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 914 struct disklabel *lp = dksc->sc_dkdev.dk_label; 915 struct cpu_disklabel *clp = dksc->sc_dkdev.dk_cpulabel; 916 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom; 917 struct partition *pp; 918 int i; 919 const char *errstring; 920 921 memset(clp, 0x0, sizeof(*clp)); 922 dk_getdefaultlabel(dksc, lp); 923 errstring = readdisklabel(DKLABELDEV(dev), dkd->d_strategy, 924 dksc->sc_dkdev.dk_label, dksc->sc_dkdev.dk_cpulabel); 925 if (errstring) { 926 dk_makedisklabel(dksc); 927 if (dksc->sc_flags & DKF_WARNLABEL) 928 printf("%s: %s\n", dksc->sc_xname, errstring); 929 return; 930 } 931 932 if ((dksc->sc_flags & DKF_LABELSANITY) == 0) 933 return; 934 935 /* Sanity check */ 936 if (lp->d_secperunit > dg->dg_secperunit) 937 printf("WARNING: %s: total sector size in disklabel (%ju) " 938 "!= the size of %s (%ju)\n", dksc->sc_xname, 939 (uintmax_t)lp->d_secperunit, dksc->sc_xname, 940 (uintmax_t)dg->dg_secperunit); 941 else if (lp->d_secperunit < UINT32_MAX && 942 lp->d_secperunit < dg->dg_secperunit) 943 printf("%s: %ju trailing sectors not covered by disklabel\n", 944 dksc->sc_xname, 945 (uintmax_t)dg->dg_secperunit - lp->d_secperunit); 946 947 for (i=0; i < lp->d_npartitions; i++) { 948 pp = &lp->d_partitions[i]; 949 if (pp->p_offset + pp->p_size > dg->dg_secperunit) 950 printf("WARNING: %s: end of partition `%c' exceeds " 951 "the size of %s (%ju)\n", dksc->sc_xname, 952 'a' + i, dksc->sc_xname, 953 (uintmax_t)dg->dg_secperunit); 954 } 955 } 956 957 /* 958 * Heuristic to conjure a disklabel if reading a disklabel failed. 959 * 960 * This is to allow the raw partition to be used for a filesystem 961 * without caring about the write protected label sector. 962 * 963 * If the driver provides it's own callback, use that instead. 964 */ 965 /* ARGSUSED */ 966 static void 967 dk_makedisklabel(struct dk_softc *dksc) 968 { 969 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver; 970 struct disklabel *lp = dksc->sc_dkdev.dk_label; 971 972 strlcpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 973 974 if (dkd->d_label) 975 dkd->d_label(dksc->sc_dev, lp); 976 else 977 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 978 979 lp->d_checksum = dkcksum(lp); 980 } 981 982 /* This function is taken from ccd.c:1.76 --rcd */ 983 984 /* 985 * XXX this function looks too generic for dksubr.c, shouldn't we 986 * put it somewhere better? 987 */ 988 989 /* 990 * Lookup the provided name in the filesystem. If the file exists, 991 * is a valid block device, and isn't being used by anyone else, 992 * set *vpp to the file's vnode. 993 */ 994 int 995 dk_lookup(struct pathbuf *pb, struct lwp *l, struct vnode **vpp) 996 { 997 struct nameidata nd; 998 struct vnode *vp; 999 int error; 1000 1001 if (l == NULL) 1002 return ESRCH; /* Is ESRCH the best choice? */ 1003 1004 NDINIT(&nd, LOOKUP, FOLLOW, pb); 1005 if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) { 1006 DPRINTF((DKDB_FOLLOW|DKDB_INIT), 1007 ("%s: vn_open error = %d\n", __func__, error)); 1008 return error; 1009 } 1010 1011 vp = nd.ni_vp; 1012 if (vp->v_type != VBLK) { 1013 error = ENOTBLK; 1014 goto out; 1015 } 1016 1017 /* Reopen as anonymous vnode to protect against forced unmount. */ 1018 if ((error = bdevvp(vp->v_rdev, vpp)) != 0) 1019 goto out; 1020 VOP_UNLOCK(vp); 1021 if ((error = vn_close(vp, FREAD | FWRITE, l->l_cred)) != 0) { 1022 vrele(*vpp); 1023 return error; 1024 } 1025 if ((error = VOP_OPEN(*vpp, FREAD | FWRITE, l->l_cred)) != 0) { 1026 vrele(*vpp); 1027 return error; 1028 } 1029 mutex_enter((*vpp)->v_interlock); 1030 (*vpp)->v_writecount++; 1031 mutex_exit((*vpp)->v_interlock); 1032 1033 IFDEBUG(DKDB_VNODE, vprint("dk_lookup: vnode info", *vpp)); 1034 1035 return 0; 1036 out: 1037 VOP_UNLOCK(vp); 1038 (void) vn_close(vp, FREAD | FWRITE, l->l_cred); 1039 return error; 1040 } 1041 1042 MODULE(MODULE_CLASS_MISC, dk_subr, NULL); 1043 1044 static int 1045 dk_subr_modcmd(modcmd_t cmd, void *arg) 1046 { 1047 switch (cmd) { 1048 case MODULE_CMD_INIT: 1049 case MODULE_CMD_FINI: 1050 return 0; 1051 case MODULE_CMD_STAT: 1052 case MODULE_CMD_AUTOUNLOAD: 1053 default: 1054 return ENOTTY; 1055 } 1056 } 1057