vnd.c revision 1.92
1/* $NetBSD: vnd.c,v 1.92 2003/02/25 20:35:34 thorpej Exp $ */ 2 3/*- 4 * Copyright (c) 1996, 1997, 1998 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. 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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39/* 40 * Copyright (c) 1988 University of Utah. 41 * Copyright (c) 1990, 1993 42 * The Regents of the University of California. All rights reserved. 43 * 44 * This code is derived from software contributed to Berkeley by 45 * the Systems Programming Group of the University of Utah Computer 46 * Science Department. 47 * 48 * Redistribution and use in source and binary forms, with or without 49 * modification, are permitted provided that the following conditions 50 * are met: 51 * 1. Redistributions of source code must retain the above copyright 52 * notice, this list of conditions and the following disclaimer. 53 * 2. Redistributions in binary form must reproduce the above copyright 54 * notice, this list of conditions and the following disclaimer in the 55 * documentation and/or other materials provided with the distribution. 56 * 3. All advertising materials mentioning features or use of this software 57 * must display the following acknowledgement: 58 * This product includes software developed by the University of 59 * California, Berkeley and its contributors. 60 * 4. Neither the name of the University nor the names of its contributors 61 * may be used to endorse or promote products derived from this software 62 * without specific prior written permission. 63 * 64 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 65 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 66 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 67 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 68 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 69 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 70 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 71 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 72 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 73 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 74 * SUCH DAMAGE. 75 * 76 * from: Utah $Hdr: vn.c 1.13 94/04/02$ 77 * 78 * @(#)vn.c 8.9 (Berkeley) 5/14/95 79 */ 80 81/* 82 * Vnode disk driver. 83 * 84 * Block/character interface to a vnode. Allows one to treat a file 85 * as a disk (e.g. build a filesystem in it, mount it, etc.). 86 * 87 * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode 88 * instead of a simple VOP_RDWR. We do this to avoid distorting the 89 * local buffer cache. 90 * 91 * NOTE 2: There is a security issue involved with this driver. 92 * Once mounted all access to the contents of the "mapped" file via 93 * the special file is controlled by the permissions on the special 94 * file, the protection of the mapped file is ignored (effectively, 95 * by using root credentials in all transactions). 96 * 97 * NOTE 3: Doesn't interact with leases, should it? 98 */ 99 100#include <sys/cdefs.h> 101__KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.92 2003/02/25 20:35:34 thorpej Exp $"); 102 103#if defined(_KERNEL_OPT) 104#include "fs_nfs.h" 105#endif 106 107#include <sys/param.h> 108#include <sys/systm.h> 109#include <sys/namei.h> 110#include <sys/proc.h> 111#include <sys/errno.h> 112#include <sys/buf.h> 113#include <sys/malloc.h> 114#include <sys/ioctl.h> 115#include <sys/disklabel.h> 116#include <sys/device.h> 117#include <sys/disk.h> 118#include <sys/stat.h> 119#include <sys/mount.h> 120#include <sys/vnode.h> 121#include <sys/file.h> 122#include <sys/uio.h> 123#include <sys/conf.h> 124 125#include <miscfs/specfs/specdev.h> 126 127#include <dev/vndvar.h> 128 129#if defined(VNDDEBUG) && !defined(DEBUG) 130#define DEBUG 131#endif 132 133#ifdef DEBUG 134int dovndcluster = 1; 135#define VDB_FOLLOW 0x01 136#define VDB_INIT 0x02 137#define VDB_IO 0x04 138#define VDB_LABEL 0x08 139int vnddebug = 0x00; 140#endif 141 142#define vndunit(x) DISKUNIT(x) 143 144struct vndxfer { 145 struct buf *vx_bp; /* Pointer to parent buffer */ 146 int vx_error; 147 int vx_pending; /* # of pending aux buffers */ 148 int vx_flags; 149#define VX_BUSY 1 150}; 151 152struct vndbuf { 153 struct buf vb_buf; 154 struct vndxfer *vb_xfer; 155}; 156 157#define VND_GETXFER(vnd) pool_get(&(vnd)->sc_vxpool, PR_NOWAIT) 158#define VND_PUTXFER(vnd, vx) pool_put(&(vnd)->sc_vxpool, (vx)) 159 160#define VND_GETBUF(vnd) pool_get(&(vnd)->sc_vbpool, PR_NOWAIT) 161#define VND_PUTBUF(vnd, vb) pool_put(&(vnd)->sc_vbpool, (vb)) 162 163struct vnd_softc *vnd_softc; 164int numvnd = 0; 165 166#define VNDLABELDEV(dev) \ 167 (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART)) 168 169/* called by main() at boot time (XXX: and the LKM driver) */ 170void vndattach __P((int)); 171int vnddetach __P((void)); 172 173void vndclear __P((struct vnd_softc *)); 174void vndstart __P((struct vnd_softc *)); 175int vndsetcred __P((struct vnd_softc *, struct ucred *)); 176void vndthrottle __P((struct vnd_softc *, struct vnode *)); 177void vndiodone __P((struct buf *)); 178void vndshutdown __P((void)); 179 180void vndgetdefaultlabel __P((struct vnd_softc *, struct disklabel *)); 181void vndgetdisklabel __P((dev_t)); 182 183static int vndlock __P((struct vnd_softc *)); 184static void vndunlock __P((struct vnd_softc *)); 185 186dev_type_open(vndopen); 187dev_type_close(vndclose); 188dev_type_read(vndread); 189dev_type_write(vndwrite); 190dev_type_ioctl(vndioctl); 191dev_type_strategy(vndstrategy); 192dev_type_dump(vnddump); 193dev_type_size(vndsize); 194 195const struct bdevsw vnd_bdevsw = { 196 vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK 197}; 198 199const struct cdevsw vnd_cdevsw = { 200 vndopen, vndclose, vndread, vndwrite, vndioctl, 201 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 202}; 203 204int vndattached = 0; 205 206void 207vndattach(num) 208 int num; 209{ 210 int i; 211 char *mem; 212 213 if (vndattached) 214 return; 215 vndattached = 1; 216 if (num <= 0) 217 return; 218 i = num * sizeof(struct vnd_softc); 219 mem = malloc(i, M_DEVBUF, M_NOWAIT|M_ZERO); 220 if (mem == NULL) { 221 printf("WARNING: no memory for vnode disks\n"); 222 return; 223 } 224 vnd_softc = (struct vnd_softc *)mem; 225 numvnd = num; 226 227 for (i = 0; i < numvnd; i++) 228 bufq_alloc(&vnd_softc[i].sc_tab, 229 BUFQ_DISKSORT|BUFQ_SORT_RAWBLOCK); 230} 231 232int 233vnddetach() 234{ 235 int i; 236 237 /* First check we aren't in use. */ 238 for (i = 0; i < numvnd; i++) 239 if (vnd_softc[i].sc_flags & VNF_INITED) 240 return (EBUSY); 241 242 for (i = 0; i < numvnd; i++) 243 bufq_free(&vnd_softc[i].sc_tab); 244 245 free(vnd_softc, M_DEVBUF); 246 vndattached = 0; 247 248 return (0); 249} 250 251int 252vndopen(dev, flags, mode, p) 253 dev_t dev; 254 int flags, mode; 255 struct proc *p; 256{ 257 int unit = vndunit(dev); 258 struct vnd_softc *sc; 259 int error = 0, part, pmask; 260 struct disklabel *lp; 261 262#ifdef DEBUG 263 if (vnddebug & VDB_FOLLOW) 264 printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p); 265#endif 266 if (unit >= numvnd) 267 return (ENXIO); 268 sc = &vnd_softc[unit]; 269 270 if ((error = vndlock(sc)) != 0) 271 return (error); 272 273 lp = sc->sc_dkdev.dk_label; 274 275 part = DISKPART(dev); 276 pmask = (1 << part); 277 278 /* 279 * If we're initialized, check to see if there are any other 280 * open partitions. If not, then it's safe to update the 281 * in-core disklabel. 282 */ 283 if ((sc->sc_flags & VNF_INITED) && (sc->sc_dkdev.dk_openmask == 0)) 284 vndgetdisklabel(dev); 285 286 /* Check that the partitions exists. */ 287 if (part != RAW_PART) { 288 if (((sc->sc_flags & VNF_INITED) == 0) || 289 ((part >= lp->d_npartitions) || 290 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 291 error = ENXIO; 292 goto done; 293 } 294 } 295 296 /* Prevent our unit from being unconfigured while open. */ 297 switch (mode) { 298 case S_IFCHR: 299 sc->sc_dkdev.dk_copenmask |= pmask; 300 break; 301 302 case S_IFBLK: 303 sc->sc_dkdev.dk_bopenmask |= pmask; 304 break; 305 } 306 sc->sc_dkdev.dk_openmask = 307 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask; 308 309 done: 310 vndunlock(sc); 311 return (error); 312} 313 314int 315vndclose(dev, flags, mode, p) 316 dev_t dev; 317 int flags, mode; 318 struct proc *p; 319{ 320 int unit = vndunit(dev); 321 struct vnd_softc *sc; 322 int error = 0, part; 323 324#ifdef DEBUG 325 if (vnddebug & VDB_FOLLOW) 326 printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p); 327#endif 328 329 if (unit >= numvnd) 330 return (ENXIO); 331 sc = &vnd_softc[unit]; 332 333 if ((error = vndlock(sc)) != 0) 334 return (error); 335 336 part = DISKPART(dev); 337 338 /* ...that much closer to allowing unconfiguration... */ 339 switch (mode) { 340 case S_IFCHR: 341 sc->sc_dkdev.dk_copenmask &= ~(1 << part); 342 break; 343 344 case S_IFBLK: 345 sc->sc_dkdev.dk_bopenmask &= ~(1 << part); 346 break; 347 } 348 sc->sc_dkdev.dk_openmask = 349 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask; 350 351 vndunlock(sc); 352 return (0); 353} 354 355/* 356 * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY. 357 */ 358void 359vndstrategy(bp) 360 struct buf *bp; 361{ 362 int unit = vndunit(bp->b_dev); 363 struct vnd_softc *vnd = &vnd_softc[unit]; 364 struct vndxfer *vnx; 365 int s, bsize, resid; 366 off_t bn; 367 caddr_t addr; 368 int sz, flags, error, wlabel; 369 struct disklabel *lp; 370 struct partition *pp; 371 372#ifdef DEBUG 373 if (vnddebug & VDB_FOLLOW) 374 printf("vndstrategy(%p): unit %d\n", bp, unit); 375#endif 376 if ((vnd->sc_flags & VNF_INITED) == 0) { 377 bp->b_error = ENXIO; 378 bp->b_flags |= B_ERROR; 379 goto done; 380 } 381 382 /* If it's a nil transfer, wake up the top half now. */ 383 if (bp->b_bcount == 0) 384 goto done; 385 386 lp = vnd->sc_dkdev.dk_label; 387 388 /* 389 * The transfer must be a whole number of blocks. 390 */ 391 if ((bp->b_bcount % lp->d_secsize) != 0) { 392 bp->b_error = EINVAL; 393 bp->b_flags |= B_ERROR; 394 goto done; 395 } 396 397 /* 398 * Do bounds checking and adjust transfer. If there's an error, 399 * the bounds check will flag that for us. 400 */ 401 wlabel = vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING); 402 if (DISKPART(bp->b_dev) != RAW_PART) 403 if (bounds_check_with_label(bp, lp, wlabel) <= 0) 404 goto done; 405 406 bp->b_resid = bp->b_bcount; 407 408 /* 409 * Put the block number in terms of the logical blocksize 410 * of the "device". 411 */ 412 bn = bp->b_blkno / (lp->d_secsize / DEV_BSIZE); 413 414 /* 415 * Translate the partition-relative block number to an absolute. 416 */ 417 if (DISKPART(bp->b_dev) != RAW_PART) { 418 pp = &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)]; 419 bn += pp->p_offset; 420 } 421 422 /* ...and convert to a byte offset within the file. */ 423 bn *= lp->d_secsize; 424 425 if (vnd->sc_vp->v_mount == NULL) { 426 bp->b_error = ENXIO; 427 bp->b_flags |= B_ERROR; 428 goto done; 429 } 430 bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize; 431 addr = bp->b_data; 432 flags = (bp->b_flags & (B_READ|B_ASYNC)) | B_CALL; 433 434 /* Allocate a header for this transfer and link it to the buffer */ 435 s = splbio(); 436 vnx = VND_GETXFER(vnd); 437 splx(s); 438 vnx->vx_flags = VX_BUSY; 439 vnx->vx_error = 0; 440 vnx->vx_pending = 0; 441 vnx->vx_bp = bp; 442 443 for (resid = bp->b_resid; resid; resid -= sz) { 444 struct vndbuf *nbp; 445 struct vnode *vp; 446 daddr_t nbn; 447 int off, nra; 448 449 nra = 0; 450 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE); 451 error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra); 452 VOP_UNLOCK(vnd->sc_vp, 0); 453 454 if (error == 0 && (long)nbn == -1) 455 error = EIO; 456 457 /* 458 * If there was an error or a hole in the file...punt. 459 * Note that we may have to wait for any operations 460 * that we have already fired off before releasing 461 * the buffer. 462 * 463 * XXX we could deal with holes here but it would be 464 * a hassle (in the write case). 465 */ 466 if (error) { 467 s = splbio(); 468 vnx->vx_error = error; 469 goto out; 470 } 471 472#ifdef DEBUG 473 if (!dovndcluster) 474 nra = 0; 475#endif 476 477 if ((off = bn % bsize) != 0) 478 sz = bsize - off; 479 else 480 sz = (1 + nra) * bsize; 481 if (resid < sz) 482 sz = resid; 483#ifdef DEBUG 484 if (vnddebug & VDB_IO) 485 printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64 486 " sz 0x%x\n", 487 vnd->sc_vp, vp, (long long)bn, nbn, sz); 488#endif 489 490 s = splbio(); 491 nbp = VND_GETBUF(vnd); 492 splx(s); 493 BUF_INIT(&nbp->vb_buf); 494 nbp->vb_buf.b_flags = flags; 495 nbp->vb_buf.b_bcount = sz; 496 nbp->vb_buf.b_bufsize = round_page((ulong)addr + sz) 497 - trunc_page((ulong) addr); 498 nbp->vb_buf.b_error = 0; 499 nbp->vb_buf.b_data = addr; 500 nbp->vb_buf.b_blkno = nbp->vb_buf.b_rawblkno = nbn + btodb(off); 501 nbp->vb_buf.b_proc = bp->b_proc; 502 nbp->vb_buf.b_iodone = vndiodone; 503 nbp->vb_buf.b_vp = NULLVP; 504 505 nbp->vb_xfer = vnx; 506 507 /* 508 * Just sort by block number 509 */ 510 s = splbio(); 511 if (vnx->vx_error != 0) { 512 VND_PUTBUF(vnd, nbp); 513 goto out; 514 } 515 vnx->vx_pending++; 516 bgetvp(vp, &nbp->vb_buf); 517 BUFQ_PUT(&vnd->sc_tab, &nbp->vb_buf); 518 vndstart(vnd); 519 splx(s); 520 bn += sz; 521 addr += sz; 522 } 523 524 s = splbio(); 525 526out: /* Arrive here at splbio */ 527 vnx->vx_flags &= ~VX_BUSY; 528 if (vnx->vx_pending == 0) { 529 if (vnx->vx_error != 0) { 530 bp->b_error = vnx->vx_error; 531 bp->b_flags |= B_ERROR; 532 } 533 VND_PUTXFER(vnd, vnx); 534 biodone(bp); 535 } 536 splx(s); 537 return; 538 539 done: 540 biodone(bp); 541} 542 543/* 544 * Feed requests sequentially. 545 * We do it this way to keep from flooding NFS servers if we are connected 546 * to an NFS file. This places the burden on the client rather than the 547 * server. 548 */ 549void 550vndstart(vnd) 551 struct vnd_softc *vnd; 552{ 553 struct buf *bp; 554 555 /* 556 * Dequeue now since lower level strategy routine might 557 * queue using same links 558 */ 559 560 if ((vnd->sc_flags & VNF_BUSY) != 0) 561 return; 562 563 vnd->sc_flags |= VNF_BUSY; 564 565 while (vnd->sc_active < vnd->sc_maxactive) { 566 bp = BUFQ_GET(&vnd->sc_tab); 567 if (bp == NULL) 568 break; 569 vnd->sc_active++; 570#ifdef DEBUG 571 if (vnddebug & VDB_IO) 572 printf("vndstart(%ld): bp %p vp %p blkno 0x%" PRIx64 573 " flags %lx addr %p cnt 0x%lx\n", 574 (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno, 575 bp->b_flags, bp->b_data, bp->b_bcount); 576#endif 577 578 /* Instrumentation. */ 579 disk_busy(&vnd->sc_dkdev); 580 581 if ((bp->b_flags & B_READ) == 0) 582 bp->b_vp->v_numoutput++; 583 VOP_STRATEGY(bp); 584 } 585 vnd->sc_flags &= ~VNF_BUSY; 586} 587 588void 589vndiodone(bp) 590 struct buf *bp; 591{ 592 struct vndbuf *vbp = (struct vndbuf *) bp; 593 struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer; 594 struct buf *pbp = vnx->vx_bp; 595 struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)]; 596 int s, resid; 597 598 s = splbio(); 599#ifdef DEBUG 600 if (vnddebug & VDB_IO) 601 printf("vndiodone(%ld): vbp %p vp %p blkno 0x%" PRIx64 602 " addr %p cnt 0x%lx\n", 603 (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp, 604 vbp->vb_buf.b_blkno, vbp->vb_buf.b_data, 605 vbp->vb_buf.b_bcount); 606#endif 607 608 resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid; 609 pbp->b_resid -= resid; 610 disk_unbusy(&vnd->sc_dkdev, resid, (pbp->b_flags & B_READ)); 611 vnx->vx_pending--; 612 613 if (vbp->vb_buf.b_error) { 614#ifdef DEBUG 615 if (vnddebug & VDB_IO) 616 printf("vndiodone: vbp %p error %d\n", vbp, 617 vbp->vb_buf.b_error); 618#endif 619 vnx->vx_error = vbp->vb_buf.b_error; 620 } 621 622 if (vbp->vb_buf.b_vp != NULLVP) 623 brelvp(&vbp->vb_buf); 624 625 VND_PUTBUF(vnd, vbp); 626 627 /* 628 * Wrap up this transaction if it has run to completion or, in 629 * case of an error, when all auxiliary buffers have returned. 630 */ 631 if (vnx->vx_error != 0) { 632 pbp->b_flags |= B_ERROR; 633 pbp->b_error = vnx->vx_error; 634 if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) { 635 636#ifdef DEBUG 637 if (vnddebug & VDB_IO) 638 printf("vndiodone: pbp %p iodone: error %d\n", 639 pbp, vnx->vx_error); 640#endif 641 VND_PUTXFER(vnd, vnx); 642 biodone(pbp); 643 } 644 } else if (pbp->b_resid == 0) { 645 646#ifdef DIAGNOSTIC 647 if (vnx->vx_pending != 0) 648 panic("vndiodone: vnx pending: %d", vnx->vx_pending); 649#endif 650 651 if ((vnx->vx_flags & VX_BUSY) == 0) { 652#ifdef DEBUG 653 if (vnddebug & VDB_IO) 654 printf("vndiodone: pbp %p iodone\n", pbp); 655#endif 656 VND_PUTXFER(vnd, vnx); 657 biodone(pbp); 658 } 659 } 660 661 vnd->sc_active--; 662 vndstart(vnd); 663 splx(s); 664} 665 666/* ARGSUSED */ 667int 668vndread(dev, uio, flags) 669 dev_t dev; 670 struct uio *uio; 671 int flags; 672{ 673 int unit = vndunit(dev); 674 struct vnd_softc *sc; 675 676#ifdef DEBUG 677 if (vnddebug & VDB_FOLLOW) 678 printf("vndread(0x%x, %p)\n", dev, uio); 679#endif 680 681 if (unit >= numvnd) 682 return (ENXIO); 683 sc = &vnd_softc[unit]; 684 685 if ((sc->sc_flags & VNF_INITED) == 0) 686 return (ENXIO); 687 688 return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio)); 689} 690 691/* ARGSUSED */ 692int 693vndwrite(dev, uio, flags) 694 dev_t dev; 695 struct uio *uio; 696 int flags; 697{ 698 int unit = vndunit(dev); 699 struct vnd_softc *sc; 700 701#ifdef DEBUG 702 if (vnddebug & VDB_FOLLOW) 703 printf("vndwrite(0x%x, %p)\n", dev, uio); 704#endif 705 706 if (unit >= numvnd) 707 return (ENXIO); 708 sc = &vnd_softc[unit]; 709 710 if ((sc->sc_flags & VNF_INITED) == 0) 711 return (ENXIO); 712 713 return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio)); 714} 715 716/* ARGSUSED */ 717int 718vndioctl(dev, cmd, data, flag, p) 719 dev_t dev; 720 u_long cmd; 721 caddr_t data; 722 int flag; 723 struct proc *p; 724{ 725 int unit = vndunit(dev); 726 struct vnd_softc *vnd; 727 struct vnd_ioctl *vio; 728 struct vattr vattr; 729 struct nameidata nd; 730 int error, part, pmask; 731 size_t geomsize; 732#ifdef __HAVE_OLD_DISKLABEL 733 struct disklabel newlabel; 734#endif 735 736#ifdef DEBUG 737 if (vnddebug & VDB_FOLLOW) 738 printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n", 739 dev, cmd, data, flag, p, unit); 740#endif 741 error = suser(p->p_ucred, &p->p_acflag); 742 if (error) 743 return (error); 744 if (unit >= numvnd) 745 return (ENXIO); 746 747 vnd = &vnd_softc[unit]; 748 vio = (struct vnd_ioctl *)data; 749 750 /* Must be open for writes for these commands... */ 751 switch (cmd) { 752 case VNDIOCSET: 753 case VNDIOCCLR: 754 case DIOCSDINFO: 755 case DIOCWDINFO: 756#ifdef __HAVE_OLD_DISKLABEL 757 case ODIOCSDINFO: 758 case ODIOCWDINFO: 759#endif 760 case DIOCWLABEL: 761 if ((flag & FWRITE) == 0) 762 return (EBADF); 763 } 764 765 /* Must be initialized for these... */ 766 switch (cmd) { 767 case VNDIOCCLR: 768 case DIOCGDINFO: 769 case DIOCSDINFO: 770 case DIOCWDINFO: 771 case DIOCGPART: 772 case DIOCWLABEL: 773 case DIOCGDEFLABEL: 774#ifdef __HAVE_OLD_DISKLABEL 775 case ODIOCGDINFO: 776 case ODIOCSDINFO: 777 case ODIOCWDINFO: 778 case ODIOCGDEFLABEL: 779#endif 780 if ((vnd->sc_flags & VNF_INITED) == 0) 781 return (ENXIO); 782 } 783 784 switch (cmd) { 785 case VNDIOCSET: 786 if (vnd->sc_flags & VNF_INITED) 787 return (EBUSY); 788 789 if ((error = vndlock(vnd)) != 0) 790 return (error); 791 792 /* 793 * Always open for read and write. 794 * This is probably bogus, but it lets vn_open() 795 * weed out directories, sockets, etc. so we don't 796 * have to worry about them. 797 */ 798 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p); 799 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) 800 goto unlock_and_exit; 801 error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p); 802 VOP_UNLOCK(nd.ni_vp, 0); 803 if (error) 804 goto close_and_exit; 805 vnd->sc_vp = nd.ni_vp; 806 vnd->sc_size = btodb(vattr.va_size); /* note truncation */ 807 808 /* 809 * Use pseudo-geometry specified. If none was provided, 810 * use "standard" Adaptec fictitious geometry. 811 */ 812 if (vio->vnd_flags & VNDIOF_HASGEOM) { 813 814 memcpy(&vnd->sc_geom, &vio->vnd_geom, 815 sizeof(vio->vnd_geom)); 816 817 /* 818 * Sanity-check the sector size. 819 * XXX Don't allow secsize < DEV_BSIZE. Should 820 * XXX we? 821 */ 822 if (vnd->sc_geom.vng_secsize < DEV_BSIZE || 823 (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0) { 824 error = EINVAL; 825 goto close_and_exit; 826 } 827 828 /* 829 * Compute the size (in DEV_BSIZE blocks) specified 830 * by the geometry. 831 */ 832 geomsize = (vnd->sc_geom.vng_nsectors * 833 vnd->sc_geom.vng_ntracks * 834 vnd->sc_geom.vng_ncylinders) * 835 (vnd->sc_geom.vng_secsize / DEV_BSIZE); 836 837 /* 838 * Sanity-check the size against the specified 839 * geometry. 840 */ 841 if (vnd->sc_size < geomsize) { 842 error = EINVAL; 843 goto close_and_exit; 844 } 845 } else { 846 /* 847 * Size must be at least 2048 DEV_BSIZE blocks 848 * (1M) in order to use this geometry. 849 */ 850 if (vnd->sc_size < (32 * 64)) { 851 error = EINVAL; 852 goto close_and_exit; 853 } 854 855 vnd->sc_geom.vng_secsize = DEV_BSIZE; 856 vnd->sc_geom.vng_nsectors = 32; 857 vnd->sc_geom.vng_ntracks = 64; 858 vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32); 859 } 860 861 if ((error = vndsetcred(vnd, p->p_ucred)) != 0) 862 goto close_and_exit; 863 vndthrottle(vnd, vnd->sc_vp); 864 vio->vnd_size = dbtob(vnd->sc_size); 865 vnd->sc_flags |= VNF_INITED; 866#ifdef DEBUG 867 if (vnddebug & VDB_INIT) 868 printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n", 869 vnd->sc_vp, (unsigned long) vnd->sc_size, 870 vnd->sc_geom.vng_secsize, 871 vnd->sc_geom.vng_nsectors, 872 vnd->sc_geom.vng_ntracks, 873 vnd->sc_geom.vng_ncylinders); 874#endif 875 876 /* Attach the disk. */ 877 memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */ 878 sprintf(vnd->sc_xname, "vnd%d", unit); /* XXX */ 879 vnd->sc_dkdev.dk_name = vnd->sc_xname; 880 disk_attach(&vnd->sc_dkdev); 881 882 /* Initialize the xfer and buffer pools. */ 883 pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0, 884 0, 0, "vndxpl", NULL); 885 pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0, 886 0, 0, "vndbpl", NULL); 887 888 /* Try and read the disklabel. */ 889 vndgetdisklabel(dev); 890 891 vndunlock(vnd); 892 893 break; 894 895close_and_exit: 896 (void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p); 897unlock_and_exit: 898 vndunlock(vnd); 899 return (error); 900 901 case VNDIOCCLR: 902 if ((error = vndlock(vnd)) != 0) 903 return (error); 904 905 /* 906 * Don't unconfigure if any other partitions are open 907 * or if both the character and block flavors of this 908 * partition are open. 909 */ 910 part = DISKPART(dev); 911 pmask = (1 << part); 912 if ((vnd->sc_dkdev.dk_openmask & ~pmask) || 913 ((vnd->sc_dkdev.dk_bopenmask & pmask) && 914 (vnd->sc_dkdev.dk_copenmask & pmask))) { 915 vndunlock(vnd); 916 return (EBUSY); 917 } 918 919 vndclear(vnd); 920#ifdef DEBUG 921 if (vnddebug & VDB_INIT) 922 printf("vndioctl: CLRed\n"); 923#endif 924 925 /* Destroy the xfer and buffer pools. */ 926 pool_destroy(&vnd->sc_vxpool); 927 pool_destroy(&vnd->sc_vbpool); 928 929 /* Detatch the disk. */ 930 disk_detach(&vnd->sc_dkdev); 931 932 vndunlock(vnd); 933 934 break; 935 936 case VNDIOCGET: { 937 struct vnd_user *vnu; 938 struct vattr va; 939 940 vnu = (struct vnd_user *)data; 941 942 if (vnu->vnu_unit == -1) 943 vnu->vnu_unit = unit; 944 if (vnu->vnu_unit >= numvnd) 945 return (ENXIO); 946 if (vnu->vnu_unit < 0) 947 return (EINVAL); 948 949 vnd = &vnd_softc[vnu->vnu_unit]; 950 951 if (vnd->sc_flags & VNF_INITED) { 952 error = VOP_GETATTR(vnd->sc_vp, &va, p->p_ucred, p); 953 if (error) 954 return (error); 955 vnu->vnu_dev = va.va_fsid; 956 vnu->vnu_ino = va.va_fileid; 957 } 958 else { 959 /* unused is not an error */ 960 vnu->vnu_dev = 0; 961 vnu->vnu_ino = 0; 962 } 963 964 break; 965 } 966 967 case DIOCGDINFO: 968 *(struct disklabel *)data = *(vnd->sc_dkdev.dk_label); 969 break; 970 971#ifdef __HAVE_OLD_DISKLABEL 972 case ODIOCGDINFO: 973 newlabel = *(vnd->sc_dkdev.dk_label); 974 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 975 return ENOTTY; 976 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 977 break; 978#endif 979 980 case DIOCGPART: 981 ((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label; 982 ((struct partinfo *)data)->part = 983 &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 984 break; 985 986 case DIOCWDINFO: 987 case DIOCSDINFO: 988#ifdef __HAVE_OLD_DISKLABEL 989 case ODIOCWDINFO: 990 case ODIOCSDINFO: 991#endif 992 { 993 struct disklabel *lp; 994 995 if ((error = vndlock(vnd)) != 0) 996 return (error); 997 998 vnd->sc_flags |= VNF_LABELLING; 999 1000#ifdef __HAVE_OLD_DISKLABEL 1001 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1002 memset(&newlabel, 0, sizeof newlabel); 1003 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1004 lp = &newlabel; 1005 } else 1006#endif 1007 lp = (struct disklabel *)data; 1008 1009 error = setdisklabel(vnd->sc_dkdev.dk_label, 1010 lp, 0, vnd->sc_dkdev.dk_cpulabel); 1011 if (error == 0) { 1012 if (cmd == DIOCWDINFO 1013#ifdef __HAVE_OLD_DISKLABEL 1014 || cmd == ODIOCWDINFO 1015#endif 1016 ) 1017 error = writedisklabel(VNDLABELDEV(dev), 1018 vndstrategy, vnd->sc_dkdev.dk_label, 1019 vnd->sc_dkdev.dk_cpulabel); 1020 } 1021 1022 vnd->sc_flags &= ~VNF_LABELLING; 1023 1024 vndunlock(vnd); 1025 1026 if (error) 1027 return (error); 1028 break; 1029 } 1030 1031 case DIOCWLABEL: 1032 if (*(int *)data != 0) 1033 vnd->sc_flags |= VNF_WLABEL; 1034 else 1035 vnd->sc_flags &= ~VNF_WLABEL; 1036 break; 1037 1038 case DIOCGDEFLABEL: 1039 vndgetdefaultlabel(vnd, (struct disklabel *)data); 1040 break; 1041 1042#ifdef __HAVE_OLD_DISKLABEL 1043 case ODIOCGDEFLABEL: 1044 vndgetdefaultlabel(vnd, &newlabel); 1045 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1046 return ENOTTY; 1047 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1048 break; 1049#endif 1050 1051 default: 1052 return (ENOTTY); 1053 } 1054 1055 return (0); 1056} 1057 1058/* 1059 * Duplicate the current processes' credentials. Since we are called only 1060 * as the result of a SET ioctl and only root can do that, any future access 1061 * to this "disk" is essentially as root. Note that credentials may change 1062 * if some other uid can write directly to the mapped file (NFS). 1063 */ 1064int 1065vndsetcred(vnd, cred) 1066 struct vnd_softc *vnd; 1067 struct ucred *cred; 1068{ 1069 struct uio auio; 1070 struct iovec aiov; 1071 char *tmpbuf; 1072 int error; 1073 1074 vnd->sc_cred = crdup(cred); 1075 tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK); 1076 1077 /* XXX: Horrible kludge to establish credentials for NFS */ 1078 aiov.iov_base = tmpbuf; 1079 aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size)); 1080 auio.uio_iov = &aiov; 1081 auio.uio_iovcnt = 1; 1082 auio.uio_offset = 0; 1083 auio.uio_rw = UIO_READ; 1084 auio.uio_segflg = UIO_SYSSPACE; 1085 auio.uio_resid = aiov.iov_len; 1086 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1087 error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred); 1088 if (error == 0) { 1089 /* 1090 * Because vnd does all IO directly through the vnode 1091 * we need to flush (at least) the buffer from the above 1092 * VOP_READ from the buffer cache to prevent cache 1093 * incoherencies. Also, be careful to write dirty 1094 * buffers back to stable storage. 1095 */ 1096 error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred, 1097 curproc, 0, 0); 1098 } 1099 VOP_UNLOCK(vnd->sc_vp, 0); 1100 1101 free(tmpbuf, M_TEMP); 1102 return (error); 1103} 1104 1105/* 1106 * Set maxactive based on FS type 1107 */ 1108void 1109vndthrottle(vnd, vp) 1110 struct vnd_softc *vnd; 1111 struct vnode *vp; 1112{ 1113#ifdef NFS 1114 extern int (**nfsv2_vnodeop_p) __P((void *)); 1115 1116 if (vp->v_op == nfsv2_vnodeop_p) 1117 vnd->sc_maxactive = 2; 1118 else 1119#endif 1120 vnd->sc_maxactive = 8; 1121 1122 if (vnd->sc_maxactive < 1) 1123 vnd->sc_maxactive = 1; 1124} 1125 1126void 1127vndshutdown() 1128{ 1129 struct vnd_softc *vnd; 1130 1131 for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++) 1132 if (vnd->sc_flags & VNF_INITED) 1133 vndclear(vnd); 1134} 1135 1136void 1137vndclear(vnd) 1138 struct vnd_softc *vnd; 1139{ 1140 struct vnode *vp = vnd->sc_vp; 1141 struct proc *p = curproc; /* XXX */ 1142 1143#ifdef DEBUG 1144 if (vnddebug & VDB_FOLLOW) 1145 printf("vndclear(%p): vp %p\n", vnd, vp); 1146#endif 1147 vnd->sc_flags &= ~VNF_INITED; 1148 if (vp == (struct vnode *)0) 1149 panic("vndioctl: null vp"); 1150 (void) vn_close(vp, FREAD|FWRITE, vnd->sc_cred, p); 1151 crfree(vnd->sc_cred); 1152 vnd->sc_vp = (struct vnode *)0; 1153 vnd->sc_cred = (struct ucred *)0; 1154 vnd->sc_size = 0; 1155} 1156 1157int 1158vndsize(dev) 1159 dev_t dev; 1160{ 1161 struct vnd_softc *sc; 1162 struct disklabel *lp; 1163 int part, unit, omask; 1164 int size; 1165 1166 unit = vndunit(dev); 1167 if (unit >= numvnd) 1168 return (-1); 1169 sc = &vnd_softc[unit]; 1170 1171 if ((sc->sc_flags & VNF_INITED) == 0) 1172 return (-1); 1173 1174 part = DISKPART(dev); 1175 omask = sc->sc_dkdev.dk_openmask & (1 << part); 1176 lp = sc->sc_dkdev.dk_label; 1177 1178 if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc)) 1179 return (-1); 1180 1181 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1182 size = -1; 1183 else 1184 size = lp->d_partitions[part].p_size * 1185 (lp->d_secsize / DEV_BSIZE); 1186 1187 if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc)) 1188 return (-1); 1189 1190 return (size); 1191} 1192 1193int 1194vnddump(dev, blkno, va, size) 1195 dev_t dev; 1196 daddr_t blkno; 1197 caddr_t va; 1198 size_t size; 1199{ 1200 1201 /* Not implemented. */ 1202 return ENXIO; 1203} 1204 1205void 1206vndgetdefaultlabel(sc, lp) 1207 struct vnd_softc *sc; 1208 struct disklabel *lp; 1209{ 1210 struct vndgeom *vng = &sc->sc_geom; 1211 struct partition *pp; 1212 1213 memset(lp, 0, sizeof(*lp)); 1214 1215 lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE); 1216 lp->d_secsize = vng->vng_secsize; 1217 lp->d_nsectors = vng->vng_nsectors; 1218 lp->d_ntracks = vng->vng_ntracks; 1219 lp->d_ncylinders = vng->vng_ncylinders; 1220 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1221 1222 strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename)); 1223 lp->d_type = DTYPE_VND; 1224 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1225 lp->d_rpm = 3600; 1226 lp->d_interleave = 1; 1227 lp->d_flags = 0; 1228 1229 pp = &lp->d_partitions[RAW_PART]; 1230 pp->p_offset = 0; 1231 pp->p_size = lp->d_secperunit; 1232 pp->p_fstype = FS_UNUSED; 1233 lp->d_npartitions = RAW_PART + 1; 1234 1235 lp->d_magic = DISKMAGIC; 1236 lp->d_magic2 = DISKMAGIC; 1237 lp->d_checksum = dkcksum(lp); 1238} 1239 1240/* 1241 * Read the disklabel from a vnd. If one is not present, create a fake one. 1242 */ 1243void 1244vndgetdisklabel(dev) 1245 dev_t dev; 1246{ 1247 struct vnd_softc *sc = &vnd_softc[vndunit(dev)]; 1248 char *errstring; 1249 struct disklabel *lp = sc->sc_dkdev.dk_label; 1250 struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel; 1251 int i; 1252 1253 memset(clp, 0, sizeof(*clp)); 1254 1255 vndgetdefaultlabel(sc, lp); 1256 1257 /* 1258 * Call the generic disklabel extraction routine. 1259 */ 1260 errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp); 1261 if (errstring) { 1262 /* 1263 * Lack of disklabel is common, but we print the warning 1264 * anyway, since it might contain other useful information. 1265 */ 1266 printf("%s: %s\n", sc->sc_xname, errstring); 1267 1268 /* 1269 * For historical reasons, if there's no disklabel 1270 * present, all partitions must be FS_BSDFFS and 1271 * occupy the entire disk. 1272 */ 1273 for (i = 0; i < MAXPARTITIONS; i++) { 1274 /* 1275 * Don't wipe out port specific hack (such as 1276 * dos partition hack of i386 port). 1277 */ 1278 if (lp->d_partitions[i].p_fstype != FS_UNUSED) 1279 continue; 1280 1281 lp->d_partitions[i].p_size = lp->d_secperunit; 1282 lp->d_partitions[i].p_offset = 0; 1283 lp->d_partitions[i].p_fstype = FS_BSDFFS; 1284 } 1285 1286 strncpy(lp->d_packname, "default label", 1287 sizeof(lp->d_packname)); 1288 1289 lp->d_checksum = dkcksum(lp); 1290 } 1291} 1292 1293/* 1294 * Wait interruptibly for an exclusive lock. 1295 * 1296 * XXX 1297 * Several drivers do this; it should be abstracted and made MP-safe. 1298 */ 1299static int 1300vndlock(sc) 1301 struct vnd_softc *sc; 1302{ 1303 int error; 1304 1305 while ((sc->sc_flags & VNF_LOCKED) != 0) { 1306 sc->sc_flags |= VNF_WANTED; 1307 if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0) 1308 return (error); 1309 } 1310 sc->sc_flags |= VNF_LOCKED; 1311 return (0); 1312} 1313 1314/* 1315 * Unlock and wake up any waiters. 1316 */ 1317static void 1318vndunlock(sc) 1319 struct vnd_softc *sc; 1320{ 1321 1322 sc->sc_flags &= ~VNF_LOCKED; 1323 if ((sc->sc_flags & VNF_WANTED) != 0) { 1324 sc->sc_flags &= ~VNF_WANTED; 1325 wakeup(sc); 1326 } 1327} 1328