vnd.c revision 1.91
1/* $NetBSD: vnd.c,v 1.91 2003/02/05 21:38:39 pk 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 1.91 2003/02/05 21:38:39 pk 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 simple_lock_init(&nbp->vb_buf.b_interlock); 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 LIST_INIT(&nbp->vb_buf.b_dep); 505 506 nbp->vb_xfer = vnx; 507 508 /* 509 * Just sort by block number 510 */ 511 s = splbio(); 512 if (vnx->vx_error != 0) { 513 VND_PUTBUF(vnd, nbp); 514 goto out; 515 } 516 vnx->vx_pending++; 517 bgetvp(vp, &nbp->vb_buf); 518 BUFQ_PUT(&vnd->sc_tab, &nbp->vb_buf); 519 vndstart(vnd); 520 splx(s); 521 bn += sz; 522 addr += sz; 523 } 524 525 s = splbio(); 526 527out: /* Arrive here at splbio */ 528 vnx->vx_flags &= ~VX_BUSY; 529 if (vnx->vx_pending == 0) { 530 if (vnx->vx_error != 0) { 531 bp->b_error = vnx->vx_error; 532 bp->b_flags |= B_ERROR; 533 } 534 VND_PUTXFER(vnd, vnx); 535 biodone(bp); 536 } 537 splx(s); 538 return; 539 540 done: 541 biodone(bp); 542} 543 544/* 545 * Feed requests sequentially. 546 * We do it this way to keep from flooding NFS servers if we are connected 547 * to an NFS file. This places the burden on the client rather than the 548 * server. 549 */ 550void 551vndstart(vnd) 552 struct vnd_softc *vnd; 553{ 554 struct buf *bp; 555 556 /* 557 * Dequeue now since lower level strategy routine might 558 * queue using same links 559 */ 560 561 if ((vnd->sc_flags & VNF_BUSY) != 0) 562 return; 563 564 vnd->sc_flags |= VNF_BUSY; 565 566 while (vnd->sc_active < vnd->sc_maxactive) { 567 bp = BUFQ_GET(&vnd->sc_tab); 568 if (bp == NULL) 569 break; 570 vnd->sc_active++; 571#ifdef DEBUG 572 if (vnddebug & VDB_IO) 573 printf("vndstart(%ld): bp %p vp %p blkno 0x%" PRIx64 574 " flags %lx addr %p cnt 0x%lx\n", 575 (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno, 576 bp->b_flags, bp->b_data, bp->b_bcount); 577#endif 578 579 /* Instrumentation. */ 580 disk_busy(&vnd->sc_dkdev); 581 582 if ((bp->b_flags & B_READ) == 0) 583 bp->b_vp->v_numoutput++; 584 VOP_STRATEGY(bp); 585 } 586 vnd->sc_flags &= ~VNF_BUSY; 587} 588 589void 590vndiodone(bp) 591 struct buf *bp; 592{ 593 struct vndbuf *vbp = (struct vndbuf *) bp; 594 struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer; 595 struct buf *pbp = vnx->vx_bp; 596 struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)]; 597 int s, resid; 598 599 s = splbio(); 600#ifdef DEBUG 601 if (vnddebug & VDB_IO) 602 printf("vndiodone(%ld): vbp %p vp %p blkno 0x%" PRIx64 603 " addr %p cnt 0x%lx\n", 604 (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp, 605 vbp->vb_buf.b_blkno, vbp->vb_buf.b_data, 606 vbp->vb_buf.b_bcount); 607#endif 608 609 resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid; 610 pbp->b_resid -= resid; 611 disk_unbusy(&vnd->sc_dkdev, resid, (pbp->b_flags & B_READ)); 612 vnx->vx_pending--; 613 614 if (vbp->vb_buf.b_error) { 615#ifdef DEBUG 616 if (vnddebug & VDB_IO) 617 printf("vndiodone: vbp %p error %d\n", vbp, 618 vbp->vb_buf.b_error); 619#endif 620 vnx->vx_error = vbp->vb_buf.b_error; 621 } 622 623 if (vbp->vb_buf.b_vp != NULLVP) 624 brelvp(&vbp->vb_buf); 625 626 VND_PUTBUF(vnd, vbp); 627 628 /* 629 * Wrap up this transaction if it has run to completion or, in 630 * case of an error, when all auxiliary buffers have returned. 631 */ 632 if (vnx->vx_error != 0) { 633 pbp->b_flags |= B_ERROR; 634 pbp->b_error = vnx->vx_error; 635 if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) { 636 637#ifdef DEBUG 638 if (vnddebug & VDB_IO) 639 printf("vndiodone: pbp %p iodone: error %d\n", 640 pbp, vnx->vx_error); 641#endif 642 VND_PUTXFER(vnd, vnx); 643 biodone(pbp); 644 } 645 } else if (pbp->b_resid == 0) { 646 647#ifdef DIAGNOSTIC 648 if (vnx->vx_pending != 0) 649 panic("vndiodone: vnx pending: %d", vnx->vx_pending); 650#endif 651 652 if ((vnx->vx_flags & VX_BUSY) == 0) { 653#ifdef DEBUG 654 if (vnddebug & VDB_IO) 655 printf("vndiodone: pbp %p iodone\n", pbp); 656#endif 657 VND_PUTXFER(vnd, vnx); 658 biodone(pbp); 659 } 660 } 661 662 vnd->sc_active--; 663 vndstart(vnd); 664 splx(s); 665} 666 667/* ARGSUSED */ 668int 669vndread(dev, uio, flags) 670 dev_t dev; 671 struct uio *uio; 672 int flags; 673{ 674 int unit = vndunit(dev); 675 struct vnd_softc *sc; 676 677#ifdef DEBUG 678 if (vnddebug & VDB_FOLLOW) 679 printf("vndread(0x%x, %p)\n", dev, uio); 680#endif 681 682 if (unit >= numvnd) 683 return (ENXIO); 684 sc = &vnd_softc[unit]; 685 686 if ((sc->sc_flags & VNF_INITED) == 0) 687 return (ENXIO); 688 689 return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio)); 690} 691 692/* ARGSUSED */ 693int 694vndwrite(dev, uio, flags) 695 dev_t dev; 696 struct uio *uio; 697 int flags; 698{ 699 int unit = vndunit(dev); 700 struct vnd_softc *sc; 701 702#ifdef DEBUG 703 if (vnddebug & VDB_FOLLOW) 704 printf("vndwrite(0x%x, %p)\n", dev, uio); 705#endif 706 707 if (unit >= numvnd) 708 return (ENXIO); 709 sc = &vnd_softc[unit]; 710 711 if ((sc->sc_flags & VNF_INITED) == 0) 712 return (ENXIO); 713 714 return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio)); 715} 716 717/* ARGSUSED */ 718int 719vndioctl(dev, cmd, data, flag, p) 720 dev_t dev; 721 u_long cmd; 722 caddr_t data; 723 int flag; 724 struct proc *p; 725{ 726 int unit = vndunit(dev); 727 struct vnd_softc *vnd; 728 struct vnd_ioctl *vio; 729 struct vattr vattr; 730 struct nameidata nd; 731 int error, part, pmask; 732 size_t geomsize; 733#ifdef __HAVE_OLD_DISKLABEL 734 struct disklabel newlabel; 735#endif 736 737#ifdef DEBUG 738 if (vnddebug & VDB_FOLLOW) 739 printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n", 740 dev, cmd, data, flag, p, unit); 741#endif 742 error = suser(p->p_ucred, &p->p_acflag); 743 if (error) 744 return (error); 745 if (unit >= numvnd) 746 return (ENXIO); 747 748 vnd = &vnd_softc[unit]; 749 vio = (struct vnd_ioctl *)data; 750 751 /* Must be open for writes for these commands... */ 752 switch (cmd) { 753 case VNDIOCSET: 754 case VNDIOCCLR: 755 case DIOCSDINFO: 756 case DIOCWDINFO: 757#ifdef __HAVE_OLD_DISKLABEL 758 case ODIOCSDINFO: 759 case ODIOCWDINFO: 760#endif 761 case DIOCWLABEL: 762 if ((flag & FWRITE) == 0) 763 return (EBADF); 764 } 765 766 /* Must be initialized for these... */ 767 switch (cmd) { 768 case VNDIOCCLR: 769 case DIOCGDINFO: 770 case DIOCSDINFO: 771 case DIOCWDINFO: 772 case DIOCGPART: 773 case DIOCWLABEL: 774 case DIOCGDEFLABEL: 775#ifdef __HAVE_OLD_DISKLABEL 776 case ODIOCGDINFO: 777 case ODIOCSDINFO: 778 case ODIOCWDINFO: 779 case ODIOCGDEFLABEL: 780#endif 781 if ((vnd->sc_flags & VNF_INITED) == 0) 782 return (ENXIO); 783 } 784 785 switch (cmd) { 786 case VNDIOCSET: 787 if (vnd->sc_flags & VNF_INITED) 788 return (EBUSY); 789 790 if ((error = vndlock(vnd)) != 0) 791 return (error); 792 793 /* 794 * Always open for read and write. 795 * This is probably bogus, but it lets vn_open() 796 * weed out directories, sockets, etc. so we don't 797 * have to worry about them. 798 */ 799 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p); 800 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) 801 goto unlock_and_exit; 802 error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p); 803 VOP_UNLOCK(nd.ni_vp, 0); 804 if (error) 805 goto close_and_exit; 806 vnd->sc_vp = nd.ni_vp; 807 vnd->sc_size = btodb(vattr.va_size); /* note truncation */ 808 809 /* 810 * Use pseudo-geometry specified. If none was provided, 811 * use "standard" Adaptec fictitious geometry. 812 */ 813 if (vio->vnd_flags & VNDIOF_HASGEOM) { 814 815 memcpy(&vnd->sc_geom, &vio->vnd_geom, 816 sizeof(vio->vnd_geom)); 817 818 /* 819 * Sanity-check the sector size. 820 * XXX Don't allow secsize < DEV_BSIZE. Should 821 * XXX we? 822 */ 823 if (vnd->sc_geom.vng_secsize < DEV_BSIZE || 824 (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0) { 825 error = EINVAL; 826 goto close_and_exit; 827 } 828 829 /* 830 * Compute the size (in DEV_BSIZE blocks) specified 831 * by the geometry. 832 */ 833 geomsize = (vnd->sc_geom.vng_nsectors * 834 vnd->sc_geom.vng_ntracks * 835 vnd->sc_geom.vng_ncylinders) * 836 (vnd->sc_geom.vng_secsize / DEV_BSIZE); 837 838 /* 839 * Sanity-check the size against the specified 840 * geometry. 841 */ 842 if (vnd->sc_size < geomsize) { 843 error = EINVAL; 844 goto close_and_exit; 845 } 846 } else { 847 /* 848 * Size must be at least 2048 DEV_BSIZE blocks 849 * (1M) in order to use this geometry. 850 */ 851 if (vnd->sc_size < (32 * 64)) { 852 error = EINVAL; 853 goto close_and_exit; 854 } 855 856 vnd->sc_geom.vng_secsize = DEV_BSIZE; 857 vnd->sc_geom.vng_nsectors = 32; 858 vnd->sc_geom.vng_ntracks = 64; 859 vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32); 860 } 861 862 if ((error = vndsetcred(vnd, p->p_ucred)) != 0) 863 goto close_and_exit; 864 vndthrottle(vnd, vnd->sc_vp); 865 vio->vnd_size = dbtob(vnd->sc_size); 866 vnd->sc_flags |= VNF_INITED; 867#ifdef DEBUG 868 if (vnddebug & VDB_INIT) 869 printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n", 870 vnd->sc_vp, (unsigned long) vnd->sc_size, 871 vnd->sc_geom.vng_secsize, 872 vnd->sc_geom.vng_nsectors, 873 vnd->sc_geom.vng_ntracks, 874 vnd->sc_geom.vng_ncylinders); 875#endif 876 877 /* Attach the disk. */ 878 memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */ 879 sprintf(vnd->sc_xname, "vnd%d", unit); /* XXX */ 880 vnd->sc_dkdev.dk_name = vnd->sc_xname; 881 disk_attach(&vnd->sc_dkdev); 882 883 /* Initialize the xfer and buffer pools. */ 884 pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0, 885 0, 0, "vndxpl", NULL); 886 pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0, 887 0, 0, "vndbpl", NULL); 888 889 /* Try and read the disklabel. */ 890 vndgetdisklabel(dev); 891 892 vndunlock(vnd); 893 894 break; 895 896close_and_exit: 897 (void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p); 898unlock_and_exit: 899 vndunlock(vnd); 900 return (error); 901 902 case VNDIOCCLR: 903 if ((error = vndlock(vnd)) != 0) 904 return (error); 905 906 /* 907 * Don't unconfigure if any other partitions are open 908 * or if both the character and block flavors of this 909 * partition are open. 910 */ 911 part = DISKPART(dev); 912 pmask = (1 << part); 913 if ((vnd->sc_dkdev.dk_openmask & ~pmask) || 914 ((vnd->sc_dkdev.dk_bopenmask & pmask) && 915 (vnd->sc_dkdev.dk_copenmask & pmask))) { 916 vndunlock(vnd); 917 return (EBUSY); 918 } 919 920 vndclear(vnd); 921#ifdef DEBUG 922 if (vnddebug & VDB_INIT) 923 printf("vndioctl: CLRed\n"); 924#endif 925 926 /* Destroy the xfer and buffer pools. */ 927 pool_destroy(&vnd->sc_vxpool); 928 pool_destroy(&vnd->sc_vbpool); 929 930 /* Detatch the disk. */ 931 disk_detach(&vnd->sc_dkdev); 932 933 vndunlock(vnd); 934 935 break; 936 937 case VNDIOCGET: { 938 struct vnd_user *vnu; 939 struct vattr va; 940 941 vnu = (struct vnd_user *)data; 942 943 if (vnu->vnu_unit == -1) 944 vnu->vnu_unit = unit; 945 if (vnu->vnu_unit >= numvnd) 946 return (ENXIO); 947 if (vnu->vnu_unit < 0) 948 return (EINVAL); 949 950 vnd = &vnd_softc[vnu->vnu_unit]; 951 952 if (vnd->sc_flags & VNF_INITED) { 953 error = VOP_GETATTR(vnd->sc_vp, &va, p->p_ucred, p); 954 if (error) 955 return (error); 956 vnu->vnu_dev = va.va_fsid; 957 vnu->vnu_ino = va.va_fileid; 958 } 959 else { 960 /* unused is not an error */ 961 vnu->vnu_dev = 0; 962 vnu->vnu_ino = 0; 963 } 964 965 break; 966 } 967 968 case DIOCGDINFO: 969 *(struct disklabel *)data = *(vnd->sc_dkdev.dk_label); 970 break; 971 972#ifdef __HAVE_OLD_DISKLABEL 973 case ODIOCGDINFO: 974 newlabel = *(vnd->sc_dkdev.dk_label); 975 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 976 return ENOTTY; 977 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 978 break; 979#endif 980 981 case DIOCGPART: 982 ((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label; 983 ((struct partinfo *)data)->part = 984 &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 985 break; 986 987 case DIOCWDINFO: 988 case DIOCSDINFO: 989#ifdef __HAVE_OLD_DISKLABEL 990 case ODIOCWDINFO: 991 case ODIOCSDINFO: 992#endif 993 { 994 struct disklabel *lp; 995 996 if ((error = vndlock(vnd)) != 0) 997 return (error); 998 999 vnd->sc_flags |= VNF_LABELLING; 1000 1001#ifdef __HAVE_OLD_DISKLABEL 1002 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1003 memset(&newlabel, 0, sizeof newlabel); 1004 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1005 lp = &newlabel; 1006 } else 1007#endif 1008 lp = (struct disklabel *)data; 1009 1010 error = setdisklabel(vnd->sc_dkdev.dk_label, 1011 lp, 0, vnd->sc_dkdev.dk_cpulabel); 1012 if (error == 0) { 1013 if (cmd == DIOCWDINFO 1014#ifdef __HAVE_OLD_DISKLABEL 1015 || cmd == ODIOCWDINFO 1016#endif 1017 ) 1018 error = writedisklabel(VNDLABELDEV(dev), 1019 vndstrategy, vnd->sc_dkdev.dk_label, 1020 vnd->sc_dkdev.dk_cpulabel); 1021 } 1022 1023 vnd->sc_flags &= ~VNF_LABELLING; 1024 1025 vndunlock(vnd); 1026 1027 if (error) 1028 return (error); 1029 break; 1030 } 1031 1032 case DIOCWLABEL: 1033 if (*(int *)data != 0) 1034 vnd->sc_flags |= VNF_WLABEL; 1035 else 1036 vnd->sc_flags &= ~VNF_WLABEL; 1037 break; 1038 1039 case DIOCGDEFLABEL: 1040 vndgetdefaultlabel(vnd, (struct disklabel *)data); 1041 break; 1042 1043#ifdef __HAVE_OLD_DISKLABEL 1044 case ODIOCGDEFLABEL: 1045 vndgetdefaultlabel(vnd, &newlabel); 1046 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1047 return ENOTTY; 1048 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1049 break; 1050#endif 1051 1052 default: 1053 return (ENOTTY); 1054 } 1055 1056 return (0); 1057} 1058 1059/* 1060 * Duplicate the current processes' credentials. Since we are called only 1061 * as the result of a SET ioctl and only root can do that, any future access 1062 * to this "disk" is essentially as root. Note that credentials may change 1063 * if some other uid can write directly to the mapped file (NFS). 1064 */ 1065int 1066vndsetcred(vnd, cred) 1067 struct vnd_softc *vnd; 1068 struct ucred *cred; 1069{ 1070 struct uio auio; 1071 struct iovec aiov; 1072 char *tmpbuf; 1073 int error; 1074 1075 vnd->sc_cred = crdup(cred); 1076 tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK); 1077 1078 /* XXX: Horrible kludge to establish credentials for NFS */ 1079 aiov.iov_base = tmpbuf; 1080 aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size)); 1081 auio.uio_iov = &aiov; 1082 auio.uio_iovcnt = 1; 1083 auio.uio_offset = 0; 1084 auio.uio_rw = UIO_READ; 1085 auio.uio_segflg = UIO_SYSSPACE; 1086 auio.uio_resid = aiov.iov_len; 1087 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1088 error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred); 1089 if (error == 0) { 1090 /* 1091 * Because vnd does all IO directly through the vnode 1092 * we need to flush (at least) the buffer from the above 1093 * VOP_READ from the buffer cache to prevent cache 1094 * incoherencies. Also, be careful to write dirty 1095 * buffers back to stable storage. 1096 */ 1097 error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred, 1098 curproc, 0, 0); 1099 } 1100 VOP_UNLOCK(vnd->sc_vp, 0); 1101 1102 free(tmpbuf, M_TEMP); 1103 return (error); 1104} 1105 1106/* 1107 * Set maxactive based on FS type 1108 */ 1109void 1110vndthrottle(vnd, vp) 1111 struct vnd_softc *vnd; 1112 struct vnode *vp; 1113{ 1114#ifdef NFS 1115 extern int (**nfsv2_vnodeop_p) __P((void *)); 1116 1117 if (vp->v_op == nfsv2_vnodeop_p) 1118 vnd->sc_maxactive = 2; 1119 else 1120#endif 1121 vnd->sc_maxactive = 8; 1122 1123 if (vnd->sc_maxactive < 1) 1124 vnd->sc_maxactive = 1; 1125} 1126 1127void 1128vndshutdown() 1129{ 1130 struct vnd_softc *vnd; 1131 1132 for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++) 1133 if (vnd->sc_flags & VNF_INITED) 1134 vndclear(vnd); 1135} 1136 1137void 1138vndclear(vnd) 1139 struct vnd_softc *vnd; 1140{ 1141 struct vnode *vp = vnd->sc_vp; 1142 struct proc *p = curproc; /* XXX */ 1143 1144#ifdef DEBUG 1145 if (vnddebug & VDB_FOLLOW) 1146 printf("vndclear(%p): vp %p\n", vnd, vp); 1147#endif 1148 vnd->sc_flags &= ~VNF_INITED; 1149 if (vp == (struct vnode *)0) 1150 panic("vndioctl: null vp"); 1151 (void) vn_close(vp, FREAD|FWRITE, vnd->sc_cred, p); 1152 crfree(vnd->sc_cred); 1153 vnd->sc_vp = (struct vnode *)0; 1154 vnd->sc_cred = (struct ucred *)0; 1155 vnd->sc_size = 0; 1156} 1157 1158int 1159vndsize(dev) 1160 dev_t dev; 1161{ 1162 struct vnd_softc *sc; 1163 struct disklabel *lp; 1164 int part, unit, omask; 1165 int size; 1166 1167 unit = vndunit(dev); 1168 if (unit >= numvnd) 1169 return (-1); 1170 sc = &vnd_softc[unit]; 1171 1172 if ((sc->sc_flags & VNF_INITED) == 0) 1173 return (-1); 1174 1175 part = DISKPART(dev); 1176 omask = sc->sc_dkdev.dk_openmask & (1 << part); 1177 lp = sc->sc_dkdev.dk_label; 1178 1179 if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc)) 1180 return (-1); 1181 1182 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1183 size = -1; 1184 else 1185 size = lp->d_partitions[part].p_size * 1186 (lp->d_secsize / DEV_BSIZE); 1187 1188 if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc)) 1189 return (-1); 1190 1191 return (size); 1192} 1193 1194int 1195vnddump(dev, blkno, va, size) 1196 dev_t dev; 1197 daddr_t blkno; 1198 caddr_t va; 1199 size_t size; 1200{ 1201 1202 /* Not implemented. */ 1203 return ENXIO; 1204} 1205 1206void 1207vndgetdefaultlabel(sc, lp) 1208 struct vnd_softc *sc; 1209 struct disklabel *lp; 1210{ 1211 struct vndgeom *vng = &sc->sc_geom; 1212 struct partition *pp; 1213 1214 memset(lp, 0, sizeof(*lp)); 1215 1216 lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE); 1217 lp->d_secsize = vng->vng_secsize; 1218 lp->d_nsectors = vng->vng_nsectors; 1219 lp->d_ntracks = vng->vng_ntracks; 1220 lp->d_ncylinders = vng->vng_ncylinders; 1221 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1222 1223 strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename)); 1224 lp->d_type = DTYPE_VND; 1225 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1226 lp->d_rpm = 3600; 1227 lp->d_interleave = 1; 1228 lp->d_flags = 0; 1229 1230 pp = &lp->d_partitions[RAW_PART]; 1231 pp->p_offset = 0; 1232 pp->p_size = lp->d_secperunit; 1233 pp->p_fstype = FS_UNUSED; 1234 lp->d_npartitions = RAW_PART + 1; 1235 1236 lp->d_magic = DISKMAGIC; 1237 lp->d_magic2 = DISKMAGIC; 1238 lp->d_checksum = dkcksum(lp); 1239} 1240 1241/* 1242 * Read the disklabel from a vnd. If one is not present, create a fake one. 1243 */ 1244void 1245vndgetdisklabel(dev) 1246 dev_t dev; 1247{ 1248 struct vnd_softc *sc = &vnd_softc[vndunit(dev)]; 1249 char *errstring; 1250 struct disklabel *lp = sc->sc_dkdev.dk_label; 1251 struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel; 1252 int i; 1253 1254 memset(clp, 0, sizeof(*clp)); 1255 1256 vndgetdefaultlabel(sc, lp); 1257 1258 /* 1259 * Call the generic disklabel extraction routine. 1260 */ 1261 errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp); 1262 if (errstring) { 1263 /* 1264 * Lack of disklabel is common, but we print the warning 1265 * anyway, since it might contain other useful information. 1266 */ 1267 printf("%s: %s\n", sc->sc_xname, errstring); 1268 1269 /* 1270 * For historical reasons, if there's no disklabel 1271 * present, all partitions must be FS_BSDFFS and 1272 * occupy the entire disk. 1273 */ 1274 for (i = 0; i < MAXPARTITIONS; i++) { 1275 /* 1276 * Don't wipe out port specific hack (such as 1277 * dos partition hack of i386 port). 1278 */ 1279 if (lp->d_partitions[i].p_fstype != FS_UNUSED) 1280 continue; 1281 1282 lp->d_partitions[i].p_size = lp->d_secperunit; 1283 lp->d_partitions[i].p_offset = 0; 1284 lp->d_partitions[i].p_fstype = FS_BSDFFS; 1285 } 1286 1287 strncpy(lp->d_packname, "default label", 1288 sizeof(lp->d_packname)); 1289 1290 lp->d_checksum = dkcksum(lp); 1291 } 1292} 1293 1294/* 1295 * Wait interruptibly for an exclusive lock. 1296 * 1297 * XXX 1298 * Several drivers do this; it should be abstracted and made MP-safe. 1299 */ 1300static int 1301vndlock(sc) 1302 struct vnd_softc *sc; 1303{ 1304 int error; 1305 1306 while ((sc->sc_flags & VNF_LOCKED) != 0) { 1307 sc->sc_flags |= VNF_WANTED; 1308 if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0) 1309 return (error); 1310 } 1311 sc->sc_flags |= VNF_LOCKED; 1312 return (0); 1313} 1314 1315/* 1316 * Unlock and wake up any waiters. 1317 */ 1318static void 1319vndunlock(sc) 1320 struct vnd_softc *sc; 1321{ 1322 1323 sc->sc_flags &= ~VNF_LOCKED; 1324 if ((sc->sc_flags & VNF_WANTED) != 0) { 1325 sc->sc_flags &= ~VNF_WANTED; 1326 wakeup(sc); 1327 } 1328} 1329