vnd.c revision 1.262
1 /* $NetBSD: vnd.c,v 1.262 2017/07/28 16:30:41 riastradh Exp $ */ 2 3 /*- 4 * Copyright (c) 1996, 1997, 1998, 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. 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 /* 33 * Copyright (c) 1988 University of Utah. 34 * Copyright (c) 1990, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * This code is derived from software contributed to Berkeley by 38 * the Systems Programming Group of the University of Utah Computer 39 * Science Department. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * from: Utah $Hdr: vn.c 1.13 94/04/02$ 66 * 67 * @(#)vn.c 8.9 (Berkeley) 5/14/95 68 */ 69 70 /* 71 * Vnode disk driver. 72 * 73 * Block/character interface to a vnode. Allows one to treat a file 74 * as a disk (e.g. build a filesystem in it, mount it, etc.). 75 * 76 * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations, 77 * this uses them to avoid distorting the local buffer cache. If those 78 * block-level operations are not available, this falls back to the regular 79 * read and write calls. Using these may distort the cache in some cases 80 * but better have the driver working than preventing it to work on file 81 * systems where the block-level operations are not implemented for 82 * whatever reason. 83 * 84 * NOTE 2: There is a security issue involved with this driver. 85 * Once mounted all access to the contents of the "mapped" file via 86 * the special file is controlled by the permissions on the special 87 * file, the protection of the mapped file is ignored (effectively, 88 * by using root credentials in all transactions). 89 * 90 * NOTE 3: Doesn't interact with leases, should it? 91 */ 92 93 #include <sys/cdefs.h> 94 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.262 2017/07/28 16:30:41 riastradh Exp $"); 95 96 #if defined(_KERNEL_OPT) 97 #include "opt_vnd.h" 98 #include "opt_compat_netbsd.h" 99 #endif 100 101 #include <sys/param.h> 102 #include <sys/systm.h> 103 #include <sys/namei.h> 104 #include <sys/proc.h> 105 #include <sys/kthread.h> 106 #include <sys/errno.h> 107 #include <sys/buf.h> 108 #include <sys/bufq.h> 109 #include <sys/malloc.h> 110 #include <sys/ioctl.h> 111 #include <sys/disklabel.h> 112 #include <sys/device.h> 113 #include <sys/disk.h> 114 #include <sys/stat.h> 115 #include <sys/mount.h> 116 #include <sys/vnode.h> 117 #include <sys/file.h> 118 #include <sys/uio.h> 119 #include <sys/conf.h> 120 #include <sys/kauth.h> 121 #include <sys/module.h> 122 123 #include <net/zlib.h> 124 125 #include <miscfs/genfs/genfs.h> 126 #include <miscfs/specfs/specdev.h> 127 128 #include <dev/dkvar.h> 129 #include <dev/vndvar.h> 130 131 #include "ioconf.h" 132 133 #if defined(VNDDEBUG) && !defined(DEBUG) 134 #define DEBUG 135 #endif 136 137 #ifdef DEBUG 138 int dovndcluster = 1; 139 #define VDB_FOLLOW 0x01 140 #define VDB_INIT 0x02 141 #define VDB_IO 0x04 142 #define VDB_LABEL 0x08 143 int vnddebug = 0; 144 #endif 145 146 #define vndunit(x) DISKUNIT(x) 147 148 struct vndxfer { 149 struct buf vx_buf; 150 struct vnd_softc *vx_vnd; 151 }; 152 #define VND_BUFTOXFER(bp) ((struct vndxfer *)(void *)bp) 153 154 #define VND_GETXFER(vnd) pool_get(&(vnd)->sc_vxpool, PR_WAITOK) 155 #define VND_PUTXFER(vnd, vx) pool_put(&(vnd)->sc_vxpool, (vx)) 156 157 #define VNDLABELDEV(dev) \ 158 (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART)) 159 160 #define VND_MAXPENDING(vnd) ((vnd)->sc_maxactive * 4) 161 162 163 static void vndclear(struct vnd_softc *, int); 164 static int vnddoclear(struct vnd_softc *, int, int, bool); 165 static int vndsetcred(struct vnd_softc *, kauth_cred_t); 166 static void vndthrottle(struct vnd_softc *, struct vnode *); 167 static void vndiodone(struct buf *); 168 #if 0 169 static void vndshutdown(void); 170 #endif 171 172 static void vndgetdefaultlabel(struct vnd_softc *, struct disklabel *); 173 static void vndgetdisklabel(dev_t, struct vnd_softc *); 174 175 static int vndlock(struct vnd_softc *); 176 static void vndunlock(struct vnd_softc *); 177 #ifdef VND_COMPRESSION 178 static void compstrategy(struct buf *, off_t); 179 static void *vnd_alloc(void *, u_int, u_int); 180 static void vnd_free(void *, void *); 181 #endif /* VND_COMPRESSION */ 182 183 static void vndthread(void *); 184 static bool vnode_has_op(const struct vnode *, int); 185 static void handle_with_rdwr(struct vnd_softc *, const struct buf *, 186 struct buf *); 187 static void handle_with_strategy(struct vnd_softc *, const struct buf *, 188 struct buf *); 189 static void vnd_set_geometry(struct vnd_softc *); 190 191 static dev_type_open(vndopen); 192 static dev_type_close(vndclose); 193 static dev_type_read(vndread); 194 static dev_type_write(vndwrite); 195 static dev_type_ioctl(vndioctl); 196 static dev_type_strategy(vndstrategy); 197 static dev_type_dump(vnddump); 198 static dev_type_size(vndsize); 199 200 const struct bdevsw vnd_bdevsw = { 201 .d_open = vndopen, 202 .d_close = vndclose, 203 .d_strategy = vndstrategy, 204 .d_ioctl = vndioctl, 205 .d_dump = vnddump, 206 .d_psize = vndsize, 207 .d_discard = nodiscard, 208 .d_flag = D_DISK 209 }; 210 211 const struct cdevsw vnd_cdevsw = { 212 .d_open = vndopen, 213 .d_close = vndclose, 214 .d_read = vndread, 215 .d_write = vndwrite, 216 .d_ioctl = vndioctl, 217 .d_stop = nostop, 218 .d_tty = notty, 219 .d_poll = nopoll, 220 .d_mmap = nommap, 221 .d_kqfilter = nokqfilter, 222 .d_discard = nodiscard, 223 .d_flag = D_DISK 224 }; 225 226 static int vnd_match(device_t, cfdata_t, void *); 227 static void vnd_attach(device_t, device_t, void *); 228 static int vnd_detach(device_t, int); 229 230 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc), 231 vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN); 232 extern struct cfdriver vnd_cd; 233 234 static struct vnd_softc *vnd_spawn(int); 235 int vnd_destroy(device_t); 236 237 static struct dkdriver vnddkdriver = { 238 .d_strategy = vndstrategy, 239 .d_minphys = minphys 240 }; 241 242 void 243 vndattach(int num) 244 { 245 int error; 246 247 error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca); 248 if (error) 249 aprint_error("%s: unable to register cfattach, error = %d\n", 250 vnd_cd.cd_name, error); 251 } 252 253 static int 254 vnd_match(device_t self, cfdata_t cfdata, void *aux) 255 { 256 257 return 1; 258 } 259 260 static void 261 vnd_attach(device_t parent, device_t self, void *aux) 262 { 263 struct vnd_softc *sc = device_private(self); 264 265 sc->sc_dev = self; 266 sc->sc_comp_offsets = NULL; 267 sc->sc_comp_buff = NULL; 268 sc->sc_comp_decombuf = NULL; 269 bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK); 270 disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver); 271 if (!pmf_device_register(self, NULL, NULL)) 272 aprint_error_dev(self, "couldn't establish power handler\n"); 273 } 274 275 static int 276 vnd_detach(device_t self, int flags) 277 { 278 int error; 279 struct vnd_softc *sc = device_private(self); 280 281 if (sc->sc_flags & VNF_INITED) { 282 error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0); 283 if (error != 0) 284 return error; 285 } 286 287 pmf_device_deregister(self); 288 bufq_free(sc->sc_tab); 289 disk_destroy(&sc->sc_dkdev); 290 291 return 0; 292 } 293 294 static struct vnd_softc * 295 vnd_spawn(int unit) 296 { 297 cfdata_t cf; 298 299 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 300 cf->cf_name = vnd_cd.cd_name; 301 cf->cf_atname = vnd_cd.cd_name; 302 cf->cf_unit = unit; 303 cf->cf_fstate = FSTATE_STAR; 304 305 return device_private(config_attach_pseudo(cf)); 306 } 307 308 int 309 vnd_destroy(device_t dev) 310 { 311 int error; 312 cfdata_t cf; 313 314 cf = device_cfdata(dev); 315 error = config_detach(dev, DETACH_QUIET); 316 if (error) 317 return error; 318 free(cf, M_DEVBUF); 319 return 0; 320 } 321 322 static int 323 vndopen(dev_t dev, int flags, int mode, struct lwp *l) 324 { 325 int unit = vndunit(dev); 326 struct vnd_softc *sc; 327 int error = 0, part, pmask; 328 struct disklabel *lp; 329 330 #ifdef DEBUG 331 if (vnddebug & VDB_FOLLOW) 332 printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l); 333 #endif 334 sc = device_lookup_private(&vnd_cd, unit); 335 if (sc == NULL) { 336 sc = vnd_spawn(unit); 337 if (sc == NULL) 338 return ENOMEM; 339 340 /* compatibility, keep disklabel after close */ 341 sc->sc_flags = VNF_KLABEL; 342 } 343 344 if ((error = vndlock(sc)) != 0) 345 return error; 346 347 mutex_enter(&sc->sc_dkdev.dk_openlock); 348 349 if ((sc->sc_flags & VNF_CLEARING) != 0) { 350 error = ENXIO; 351 goto done; 352 } 353 354 lp = sc->sc_dkdev.dk_label; 355 356 part = DISKPART(dev); 357 pmask = (1 << part); 358 359 if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) { 360 error = EBUSY; 361 goto done; 362 } 363 364 if (sc->sc_flags & VNF_INITED) { 365 if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) { 366 /* 367 * If any non-raw partition is open, but the disk 368 * has been invalidated, disallow further opens. 369 */ 370 if ((sc->sc_flags & VNF_VLABEL) == 0) { 371 error = EIO; 372 goto done; 373 } 374 } else { 375 /* 376 * Load the partition info if not already loaded. 377 */ 378 if ((sc->sc_flags & VNF_VLABEL) == 0) { 379 sc->sc_flags |= VNF_VLABEL; 380 vndgetdisklabel(dev, sc); 381 } 382 } 383 } 384 385 /* Check that the partitions exists. */ 386 if (part != RAW_PART) { 387 if (((sc->sc_flags & VNF_INITED) == 0) || 388 ((part >= lp->d_npartitions) || 389 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 390 error = ENXIO; 391 goto done; 392 } 393 } 394 395 /* Prevent our unit from being unconfigured while open. */ 396 switch (mode) { 397 case S_IFCHR: 398 sc->sc_dkdev.dk_copenmask |= pmask; 399 break; 400 401 case S_IFBLK: 402 sc->sc_dkdev.dk_bopenmask |= pmask; 403 break; 404 } 405 sc->sc_dkdev.dk_openmask = 406 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask; 407 408 done: 409 mutex_exit(&sc->sc_dkdev.dk_openlock); 410 vndunlock(sc); 411 return error; 412 } 413 414 static int 415 vndclose(dev_t dev, int flags, int mode, struct lwp *l) 416 { 417 int unit = vndunit(dev); 418 struct vnd_softc *sc; 419 int error = 0, part; 420 421 #ifdef DEBUG 422 if (vnddebug & VDB_FOLLOW) 423 printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l); 424 #endif 425 sc = device_lookup_private(&vnd_cd, unit); 426 if (sc == NULL) 427 return ENXIO; 428 429 if ((error = vndlock(sc)) != 0) 430 return error; 431 432 mutex_enter(&sc->sc_dkdev.dk_openlock); 433 434 part = DISKPART(dev); 435 436 /* ...that much closer to allowing unconfiguration... */ 437 switch (mode) { 438 case S_IFCHR: 439 sc->sc_dkdev.dk_copenmask &= ~(1 << part); 440 break; 441 442 case S_IFBLK: 443 sc->sc_dkdev.dk_bopenmask &= ~(1 << part); 444 break; 445 } 446 sc->sc_dkdev.dk_openmask = 447 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask; 448 449 /* are we last opener ? */ 450 if (sc->sc_dkdev.dk_openmask == 0) { 451 if ((sc->sc_flags & VNF_KLABEL) == 0) 452 sc->sc_flags &= ~VNF_VLABEL; 453 } 454 455 mutex_exit(&sc->sc_dkdev.dk_openlock); 456 457 vndunlock(sc); 458 459 if ((sc->sc_flags & VNF_INITED) == 0) { 460 if ((error = vnd_destroy(sc->sc_dev)) != 0) { 461 aprint_error_dev(sc->sc_dev, 462 "unable to detach instance\n"); 463 return error; 464 } 465 } 466 467 return 0; 468 } 469 470 /* 471 * Queue the request, and wakeup the kernel thread to handle it. 472 */ 473 static void 474 vndstrategy(struct buf *bp) 475 { 476 int unit = vndunit(bp->b_dev); 477 struct vnd_softc *vnd = 478 device_lookup_private(&vnd_cd, unit); 479 struct disklabel *lp; 480 daddr_t blkno; 481 int s = splbio(); 482 483 if (vnd == NULL) { 484 bp->b_error = ENXIO; 485 goto done; 486 } 487 lp = vnd->sc_dkdev.dk_label; 488 489 if ((vnd->sc_flags & VNF_INITED) == 0) { 490 bp->b_error = ENXIO; 491 goto done; 492 } 493 494 /* 495 * The transfer must be a whole number of blocks. 496 */ 497 if ((bp->b_bcount % lp->d_secsize) != 0) { 498 bp->b_error = EINVAL; 499 goto done; 500 } 501 502 /* 503 * check if we're read-only. 504 */ 505 if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) { 506 bp->b_error = EACCES; 507 goto done; 508 } 509 510 /* If it's a nil transfer, wake up the top half now. */ 511 if (bp->b_bcount == 0) { 512 goto done; 513 } 514 515 /* 516 * Do bounds checking and adjust transfer. If there's an error, 517 * the bounds check will flag that for us. 518 */ 519 if (DISKPART(bp->b_dev) == RAW_PART) { 520 if (bounds_check_with_mediasize(bp, DEV_BSIZE, 521 vnd->sc_size) <= 0) 522 goto done; 523 } else { 524 if (bounds_check_with_label(&vnd->sc_dkdev, 525 bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0) 526 goto done; 527 } 528 529 /* 530 * Put the block number in terms of the logical blocksize 531 * of the "device". 532 */ 533 534 blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE); 535 536 /* 537 * Translate the partition-relative block number to an absolute. 538 */ 539 if (DISKPART(bp->b_dev) != RAW_PART) { 540 struct partition *pp; 541 542 pp = &vnd->sc_dkdev.dk_label->d_partitions[ 543 DISKPART(bp->b_dev)]; 544 blkno += pp->p_offset; 545 } 546 bp->b_rawblkno = blkno; 547 548 #ifdef DEBUG 549 if (vnddebug & VDB_FOLLOW) 550 printf("vndstrategy(%p): unit %d\n", bp, unit); 551 #endif 552 if ((vnd->sc_flags & VNF_USE_VN_RDWR)) { 553 KASSERT(vnd->sc_pending >= 0 && 554 vnd->sc_pending <= VND_MAXPENDING(vnd)); 555 while (vnd->sc_pending == VND_MAXPENDING(vnd)) 556 tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0); 557 vnd->sc_pending++; 558 } 559 bufq_put(vnd->sc_tab, bp); 560 wakeup(&vnd->sc_tab); 561 splx(s); 562 return; 563 564 done: 565 bp->b_resid = bp->b_bcount; 566 biodone(bp); 567 splx(s); 568 } 569 570 static bool 571 vnode_has_strategy(struct vnd_softc *vnd) 572 { 573 return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) && 574 vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy)); 575 } 576 577 static bool 578 vnode_has_large_blocks(struct vnd_softc *vnd) 579 { 580 u_int32_t vnd_secsize, mnt_secsize; 581 uint64_t numsec; 582 unsigned secsize; 583 584 if (getdisksize(vnd->sc_vp, &numsec, &secsize)) 585 return true; 586 587 vnd_secsize = vnd->sc_geom.vng_secsize; 588 mnt_secsize = secsize; 589 590 return vnd_secsize % mnt_secsize != 0; 591 } 592 593 /* XXX this function needs a reliable check to detect 594 * sparse files. Otherwise, bmap/strategy may be used 595 * and fail on non-allocated blocks. VOP_READ/VOP_WRITE 596 * works on sparse files. 597 */ 598 #if notyet 599 static bool 600 vnode_strategy_probe(struct vnd_softc *vnd) 601 { 602 int error; 603 daddr_t nbn; 604 605 if (!vnode_has_strategy(vnd)) 606 return false; 607 608 if (vnode_has_large_blocks(vnd)) 609 return false; 610 611 /* Convert the first logical block number to its 612 * physical block number. 613 */ 614 error = 0; 615 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 616 error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL); 617 VOP_UNLOCK(vnd->sc_vp); 618 619 /* Test if that worked. */ 620 if (error == 0 && (long)nbn == -1) 621 return false; 622 623 return true; 624 } 625 #endif 626 627 static void 628 vndthread(void *arg) 629 { 630 struct vnd_softc *vnd = arg; 631 int s; 632 633 /* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to 634 * directly access the backing vnode. If we can, use these two 635 * operations to avoid messing with the local buffer cache. 636 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations 637 * which are guaranteed to work with any file system. */ 638 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 && 639 ! vnode_has_strategy(vnd)) 640 vnd->sc_flags |= VNF_USE_VN_RDWR; 641 642 /* VOP_STRATEGY can only be used if the backing vnode allows 643 * to access blocks as small as defined by the vnd geometry. 644 */ 645 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 && 646 vnode_has_large_blocks(vnd)) 647 vnd->sc_flags |= VNF_USE_VN_RDWR; 648 649 #ifdef DEBUG 650 if (vnddebug & VDB_INIT) 651 printf("vndthread: vp %p, %s\n", vnd->sc_vp, 652 (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ? 653 "using bmap/strategy operations" : 654 "using read/write operations"); 655 #endif 656 657 s = splbio(); 658 vnd->sc_flags |= VNF_KTHREAD; 659 wakeup(&vnd->sc_kthread); 660 661 /* 662 * Dequeue requests and serve them depending on the available 663 * vnode operations. 664 */ 665 while ((vnd->sc_flags & VNF_VUNCONF) == 0) { 666 struct vndxfer *vnx; 667 struct buf *obp; 668 struct buf *bp; 669 670 obp = bufq_get(vnd->sc_tab); 671 if (obp == NULL) { 672 tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0); 673 continue; 674 }; 675 if ((vnd->sc_flags & VNF_USE_VN_RDWR)) { 676 KASSERT(vnd->sc_pending > 0 && 677 vnd->sc_pending <= VND_MAXPENDING(vnd)); 678 if (vnd->sc_pending-- == VND_MAXPENDING(vnd)) 679 wakeup(&vnd->sc_pending); 680 } 681 splx(s); 682 #ifdef DEBUG 683 if (vnddebug & VDB_FOLLOW) 684 printf("vndthread(%p)\n", obp); 685 #endif 686 687 if (vnd->sc_vp->v_mount == NULL) { 688 obp->b_error = ENXIO; 689 goto done; 690 } 691 #ifdef VND_COMPRESSION 692 /* handle a compressed read */ 693 if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) { 694 off_t bn; 695 696 /* Convert to a byte offset within the file. */ 697 bn = obp->b_rawblkno * 698 vnd->sc_dkdev.dk_label->d_secsize; 699 700 compstrategy(obp, bn); 701 goto done; 702 } 703 #endif /* VND_COMPRESSION */ 704 705 /* 706 * Allocate a header for this transfer and link it to the 707 * buffer 708 */ 709 s = splbio(); 710 vnx = VND_GETXFER(vnd); 711 splx(s); 712 vnx->vx_vnd = vnd; 713 714 s = splbio(); 715 while (vnd->sc_active >= vnd->sc_maxactive) { 716 tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0); 717 } 718 vnd->sc_active++; 719 splx(s); 720 721 /* Instrumentation. */ 722 disk_busy(&vnd->sc_dkdev); 723 724 bp = &vnx->vx_buf; 725 buf_init(bp); 726 bp->b_flags = (obp->b_flags & B_READ); 727 bp->b_oflags = obp->b_oflags; 728 bp->b_cflags = obp->b_cflags; 729 bp->b_iodone = vndiodone; 730 bp->b_private = obp; 731 bp->b_vp = vnd->sc_vp; 732 bp->b_objlock = bp->b_vp->v_interlock; 733 bp->b_data = obp->b_data; 734 bp->b_bcount = obp->b_bcount; 735 BIO_COPYPRIO(bp, obp); 736 737 /* Handle the request using the appropriate operations. */ 738 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0) 739 handle_with_strategy(vnd, obp, bp); 740 else 741 handle_with_rdwr(vnd, obp, bp); 742 743 s = splbio(); 744 continue; 745 746 done: 747 biodone(obp); 748 s = splbio(); 749 } 750 751 vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF); 752 wakeup(&vnd->sc_kthread); 753 splx(s); 754 kthread_exit(0); 755 } 756 757 /* 758 * Checks if the given vnode supports the requested operation. 759 * The operation is specified the offset returned by VOFFSET. 760 * 761 * XXX The test below used to determine this is quite fragile 762 * because it relies on the file system to use genfs to specify 763 * unimplemented operations. There might be another way to do 764 * it more cleanly. 765 */ 766 static bool 767 vnode_has_op(const struct vnode *vp, int opoffset) 768 { 769 int (*defaultp)(void *); 770 int (*opp)(void *); 771 772 defaultp = vp->v_op[VOFFSET(vop_default)]; 773 opp = vp->v_op[opoffset]; 774 775 return opp != defaultp && opp != genfs_eopnotsupp && 776 opp != genfs_badop && opp != genfs_nullop; 777 } 778 779 /* 780 * Handles the read/write request given in 'bp' using the vnode's VOP_READ 781 * and VOP_WRITE operations. 782 * 783 * 'obp' is a pointer to the original request fed to the vnd device. 784 */ 785 static void 786 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp) 787 { 788 bool doread; 789 off_t offset; 790 size_t len, resid; 791 struct vnode *vp; 792 793 doread = bp->b_flags & B_READ; 794 offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize; 795 len = bp->b_bcount; 796 vp = vnd->sc_vp; 797 798 #if defined(DEBUG) 799 if (vnddebug & VDB_IO) 800 printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64 801 ", secsize %d, offset %" PRIu64 802 ", bcount %d\n", 803 vp, doread ? "read" : "write", obp->b_rawblkno, 804 vnd->sc_dkdev.dk_label->d_secsize, offset, 805 bp->b_bcount); 806 #endif 807 808 /* Issue the read or write operation. */ 809 bp->b_error = 810 vn_rdwr(doread ? UIO_READ : UIO_WRITE, 811 vp, bp->b_data, len, offset, UIO_SYSSPACE, 812 IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL); 813 bp->b_resid = resid; 814 815 mutex_enter(vp->v_interlock); 816 (void) VOP_PUTPAGES(vp, 0, 0, 817 PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 818 819 /* We need to increase the number of outputs on the vnode if 820 * there was any write to it. */ 821 if (!doread) { 822 mutex_enter(vp->v_interlock); 823 vp->v_numoutput++; 824 mutex_exit(vp->v_interlock); 825 } 826 827 biodone(bp); 828 } 829 830 /* 831 * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP 832 * and VOP_STRATEGY operations. 833 * 834 * 'obp' is a pointer to the original request fed to the vnd device. 835 */ 836 static void 837 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp, 838 struct buf *bp) 839 { 840 int bsize, error, flags, skipped; 841 size_t resid, sz; 842 off_t bn, offset; 843 struct vnode *vp; 844 struct buf *nbp = NULL; 845 846 flags = obp->b_flags; 847 848 849 /* convert to a byte offset within the file. */ 850 bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize; 851 852 bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize; 853 skipped = 0; 854 855 /* 856 * Break the request into bsize pieces and feed them 857 * sequentially using VOP_BMAP/VOP_STRATEGY. 858 * We do it this way to keep from flooding NFS servers if we 859 * are connected to an NFS file. This places the burden on 860 * the client rather than the server. 861 */ 862 error = 0; 863 bp->b_resid = bp->b_bcount; 864 for (offset = 0, resid = bp->b_resid; /* true */; 865 resid -= sz, offset += sz) { 866 daddr_t nbn; 867 int off, nra; 868 869 nra = 0; 870 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 871 error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra); 872 VOP_UNLOCK(vnd->sc_vp); 873 874 if (error == 0 && (long)nbn == -1) 875 error = EIO; 876 877 /* 878 * If there was an error or a hole in the file...punt. 879 * Note that we may have to wait for any operations 880 * that we have already fired off before releasing 881 * the buffer. 882 * 883 * XXX we could deal with holes here but it would be 884 * a hassle (in the write case). 885 */ 886 if (error) { 887 skipped += resid; 888 break; 889 } 890 891 #ifdef DEBUG 892 if (!dovndcluster) 893 nra = 0; 894 #endif 895 896 off = bn % bsize; 897 sz = MIN(((off_t)1 + nra) * bsize - off, resid); 898 #ifdef DEBUG 899 if (vnddebug & VDB_IO) 900 printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64 901 " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn, 902 nbn, sz); 903 #endif 904 905 nbp = getiobuf(vp, true); 906 nestiobuf_setup(bp, nbp, offset, sz); 907 nbp->b_blkno = nbn + btodb(off); 908 909 #if 0 /* XXX #ifdef DEBUG */ 910 if (vnddebug & VDB_IO) 911 printf("vndstart(%ld): bp %p vp %p blkno " 912 "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n", 913 (long) (vnd-vnd_softc), &nbp->vb_buf, 914 nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno, 915 nbp->vb_buf.b_flags, nbp->vb_buf.b_data, 916 nbp->vb_buf.b_bcount); 917 #endif 918 if (resid == sz) { 919 break; 920 } 921 VOP_STRATEGY(vp, nbp); 922 bn += sz; 923 } 924 if (!(flags & B_READ)) { 925 struct vnode *w_vp; 926 /* 927 * this is the last nested buf, account for 928 * the parent buf write too. 929 * This has to be done last, so that 930 * fsync won't wait for this write which 931 * has no chance to complete before all nested bufs 932 * have been queued. But it has to be done 933 * before the last VOP_STRATEGY() 934 * or the call to nestiobuf_done(). 935 */ 936 w_vp = bp->b_vp; 937 mutex_enter(w_vp->v_interlock); 938 w_vp->v_numoutput++; 939 mutex_exit(w_vp->v_interlock); 940 } 941 KASSERT(skipped != 0 || nbp != NULL); 942 if (skipped) 943 nestiobuf_done(bp, skipped, error); 944 else 945 VOP_STRATEGY(vp, nbp); 946 } 947 948 static void 949 vndiodone(struct buf *bp) 950 { 951 struct vndxfer *vnx = VND_BUFTOXFER(bp); 952 struct vnd_softc *vnd = vnx->vx_vnd; 953 struct buf *obp = bp->b_private; 954 int s = splbio(); 955 956 KASSERT(&vnx->vx_buf == bp); 957 KASSERT(vnd->sc_active > 0); 958 #ifdef DEBUG 959 if (vnddebug & VDB_IO) { 960 printf("vndiodone1: bp %p iodone: error %d\n", 961 bp, bp->b_error); 962 } 963 #endif 964 disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid, 965 (bp->b_flags & B_READ)); 966 vnd->sc_active--; 967 if (vnd->sc_active == 0) { 968 wakeup(&vnd->sc_tab); 969 } 970 splx(s); 971 obp->b_error = bp->b_error; 972 obp->b_resid = bp->b_resid; 973 buf_destroy(bp); 974 VND_PUTXFER(vnd, vnx); 975 biodone(obp); 976 } 977 978 /* ARGSUSED */ 979 static int 980 vndread(dev_t dev, struct uio *uio, int flags) 981 { 982 int unit = vndunit(dev); 983 struct vnd_softc *sc; 984 985 #ifdef DEBUG 986 if (vnddebug & VDB_FOLLOW) 987 printf("vndread(0x%"PRIx64", %p)\n", dev, uio); 988 #endif 989 990 sc = device_lookup_private(&vnd_cd, unit); 991 if (sc == NULL) 992 return ENXIO; 993 994 if ((sc->sc_flags & VNF_INITED) == 0) 995 return ENXIO; 996 997 return physio(vndstrategy, NULL, dev, B_READ, minphys, uio); 998 } 999 1000 /* ARGSUSED */ 1001 static int 1002 vndwrite(dev_t dev, struct uio *uio, int flags) 1003 { 1004 int unit = vndunit(dev); 1005 struct vnd_softc *sc; 1006 1007 #ifdef DEBUG 1008 if (vnddebug & VDB_FOLLOW) 1009 printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio); 1010 #endif 1011 1012 sc = device_lookup_private(&vnd_cd, unit); 1013 if (sc == NULL) 1014 return ENXIO; 1015 1016 if ((sc->sc_flags & VNF_INITED) == 0) 1017 return ENXIO; 1018 1019 return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio); 1020 } 1021 1022 static int 1023 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va) 1024 { 1025 int error; 1026 struct vnd_softc *vnd; 1027 1028 if (*un == -1) 1029 *un = unit; 1030 if (*un < 0) 1031 return EINVAL; 1032 1033 vnd = device_lookup_private(&vnd_cd, *un); 1034 if (vnd == NULL) 1035 return -1; 1036 1037 if ((vnd->sc_flags & VNF_INITED) == 0) 1038 return -1; 1039 1040 vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY); 1041 error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred); 1042 VOP_UNLOCK(vnd->sc_vp); 1043 return error; 1044 } 1045 1046 static int 1047 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force) 1048 { 1049 int error; 1050 1051 if ((error = vndlock(vnd)) != 0) 1052 return error; 1053 1054 /* 1055 * Don't unconfigure if any other partitions are open 1056 * or if both the character and block flavors of this 1057 * partition are open. 1058 */ 1059 if (DK_BUSY(vnd, pmask) && !force) { 1060 vndunlock(vnd); 1061 return EBUSY; 1062 } 1063 1064 /* Delete all of our wedges */ 1065 dkwedge_delall(&vnd->sc_dkdev); 1066 1067 /* 1068 * XXX vndclear() might call vndclose() implicitly; 1069 * release lock to avoid recursion 1070 * 1071 * Set VNF_CLEARING to prevent vndopen() from 1072 * sneaking in after we vndunlock(). 1073 */ 1074 vnd->sc_flags |= VNF_CLEARING; 1075 vndunlock(vnd); 1076 vndclear(vnd, minor); 1077 #ifdef DEBUG 1078 if (vnddebug & VDB_INIT) 1079 printf("%s: CLRed\n", __func__); 1080 #endif 1081 1082 /* Destroy the xfer and buffer pools. */ 1083 pool_destroy(&vnd->sc_vxpool); 1084 1085 /* Detach the disk. */ 1086 disk_detach(&vnd->sc_dkdev); 1087 1088 return 0; 1089 } 1090 1091 static int 1092 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va) 1093 { 1094 int error; 1095 1096 KASSERT(l); 1097 1098 /* the first member is always int vnd_unit in all the versions */ 1099 if (*(int *)data >= vnd_cd.cd_ndevs) 1100 return ENXIO; 1101 1102 switch (error = vnd_cget(l, unit, (int *)data, va)) { 1103 case -1: 1104 /* unused is not an error */ 1105 memset(va, 0, sizeof(*va)); 1106 /*FALLTHROUGH*/ 1107 case 0: 1108 return 0; 1109 default: 1110 return error; 1111 } 1112 } 1113 1114 /* ARGSUSED */ 1115 static int 1116 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 1117 { 1118 bool force; 1119 int unit = vndunit(dev); 1120 struct vnd_softc *vnd; 1121 struct vnd_ioctl *vio; 1122 struct vattr vattr; 1123 struct pathbuf *pb; 1124 struct nameidata nd; 1125 int error, part, pmask; 1126 uint64_t geomsize; 1127 int fflags; 1128 #ifdef __HAVE_OLD_DISKLABEL 1129 struct disklabel newlabel; 1130 #endif 1131 1132 #ifdef DEBUG 1133 if (vnddebug & VDB_FOLLOW) 1134 printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n", 1135 dev, cmd, data, flag, l->l_proc, unit); 1136 #endif 1137 /* Do the get's first; they don't need initialization or verification */ 1138 switch (cmd) { 1139 #ifdef COMPAT_30 1140 case VNDIOCGET30: { 1141 if ((error = vndioctl_get(l, data, unit, &vattr)) != 0) 1142 return error; 1143 1144 struct vnd_user30 *vnu = data; 1145 vnu->vnu_dev = vattr.va_fsid; 1146 vnu->vnu_ino = vattr.va_fileid; 1147 return 0; 1148 } 1149 #endif 1150 #ifdef COMPAT_50 1151 case VNDIOCGET50: { 1152 if ((error = vndioctl_get(l, data, unit, &vattr)) != 0) 1153 return error; 1154 1155 struct vnd_user50 *vnu = data; 1156 vnu->vnu_dev = vattr.va_fsid; 1157 vnu->vnu_ino = vattr.va_fileid; 1158 return 0; 1159 } 1160 #endif 1161 1162 case VNDIOCGET: { 1163 if ((error = vndioctl_get(l, data, unit, &vattr)) != 0) 1164 return error; 1165 1166 struct vnd_user *vnu = data; 1167 vnu->vnu_dev = vattr.va_fsid; 1168 vnu->vnu_ino = vattr.va_fileid; 1169 return 0; 1170 } 1171 default: 1172 break; 1173 } 1174 1175 vnd = device_lookup_private(&vnd_cd, unit); 1176 if (vnd == NULL) 1177 return ENXIO; 1178 vio = (struct vnd_ioctl *)data; 1179 1180 /* Must be open for writes for these commands... */ 1181 switch (cmd) { 1182 case VNDIOCSET: 1183 case VNDIOCCLR: 1184 #ifdef COMPAT_50 1185 case VNDIOCSET50: 1186 case VNDIOCCLR50: 1187 #endif 1188 case DIOCSDINFO: 1189 case DIOCWDINFO: 1190 #ifdef __HAVE_OLD_DISKLABEL 1191 case ODIOCSDINFO: 1192 case ODIOCWDINFO: 1193 #endif 1194 case DIOCKLABEL: 1195 case DIOCWLABEL: 1196 if ((flag & FWRITE) == 0) 1197 return EBADF; 1198 } 1199 1200 /* Must be initialized for these... */ 1201 switch (cmd) { 1202 case VNDIOCCLR: 1203 #ifdef VNDIOCCLR50 1204 case VNDIOCCLR50: 1205 #endif 1206 case DIOCGDINFO: 1207 case DIOCSDINFO: 1208 case DIOCWDINFO: 1209 case DIOCGPARTINFO: 1210 case DIOCKLABEL: 1211 case DIOCWLABEL: 1212 case DIOCGDEFLABEL: 1213 case DIOCCACHESYNC: 1214 #ifdef __HAVE_OLD_DISKLABEL 1215 case ODIOCGDINFO: 1216 case ODIOCSDINFO: 1217 case ODIOCWDINFO: 1218 case ODIOCGDEFLABEL: 1219 #endif 1220 if ((vnd->sc_flags & VNF_INITED) == 0) 1221 return ENXIO; 1222 } 1223 1224 error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l); 1225 if (error != EPASSTHROUGH) 1226 return error; 1227 1228 1229 switch (cmd) { 1230 #ifdef VNDIOCSET50 1231 case VNDIOCSET50: 1232 #endif 1233 case VNDIOCSET: 1234 if (vnd->sc_flags & VNF_INITED) 1235 return EBUSY; 1236 1237 if ((error = vndlock(vnd)) != 0) 1238 return error; 1239 1240 fflags = FREAD; 1241 if ((vio->vnd_flags & VNDIOF_READONLY) == 0) 1242 fflags |= FWRITE; 1243 error = pathbuf_copyin(vio->vnd_file, &pb); 1244 if (error) { 1245 goto unlock_and_exit; 1246 } 1247 NDINIT(&nd, LOOKUP, FOLLOW, pb); 1248 if ((error = vn_open(&nd, fflags, 0)) != 0) { 1249 pathbuf_destroy(pb); 1250 goto unlock_and_exit; 1251 } 1252 KASSERT(l); 1253 error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred); 1254 if (!error && nd.ni_vp->v_type != VREG) 1255 error = EOPNOTSUPP; 1256 if (!error && vattr.va_bytes < vattr.va_size) 1257 /* File is definitely sparse, use vn_rdwr() */ 1258 vnd->sc_flags |= VNF_USE_VN_RDWR; 1259 if (error) { 1260 VOP_UNLOCK(nd.ni_vp); 1261 goto close_and_exit; 1262 } 1263 1264 /* If using a compressed file, initialize its info */ 1265 /* (or abort with an error if kernel has no compression) */ 1266 if (vio->vnd_flags & VNF_COMP) { 1267 #ifdef VND_COMPRESSION 1268 struct vnd_comp_header *ch; 1269 int i; 1270 uint32_t comp_size; 1271 uint32_t comp_maxsize; 1272 1273 /* allocate space for compresed file header */ 1274 ch = malloc(sizeof(struct vnd_comp_header), 1275 M_TEMP, M_WAITOK); 1276 1277 /* read compressed file header */ 1278 error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch, 1279 sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE, 1280 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL); 1281 if (error) { 1282 free(ch, M_TEMP); 1283 VOP_UNLOCK(nd.ni_vp); 1284 goto close_and_exit; 1285 } 1286 1287 if (be32toh(ch->block_size) == 0 || 1288 be32toh(ch->num_blocks) > UINT32_MAX - 1) { 1289 free(ch, M_TEMP); 1290 VOP_UNLOCK(nd.ni_vp); 1291 goto close_and_exit; 1292 } 1293 1294 /* save some header info */ 1295 vnd->sc_comp_blksz = be32toh(ch->block_size); 1296 /* note last offset is the file byte size */ 1297 vnd->sc_comp_numoffs = be32toh(ch->num_blocks) + 1; 1298 free(ch, M_TEMP); 1299 if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) { 1300 VOP_UNLOCK(nd.ni_vp); 1301 error = EINVAL; 1302 goto close_and_exit; 1303 } 1304 KASSERT(0 < vnd->sc_comp_blksz); 1305 KASSERT(0 < vnd->sc_comp_numoffs); 1306 /* 1307 * @#^@!$& gcc -Wtype-limits refuses to let me 1308 * write SIZE_MAX/sizeof(uint64_t) < numoffs, 1309 * because the range of the type on amd64 makes 1310 * the comparisons always false. 1311 */ 1312 #if SIZE_MAX <= UINT32_MAX*(64/CHAR_BIT) 1313 if (SIZE_MAX/sizeof(uint64_t) < vnd->sc_comp_numoffs) { 1314 VOP_UNLOCK(nd.ni_vp); 1315 error = EINVAL; 1316 goto close_and_exit; 1317 } 1318 #endif 1319 if ((vattr.va_size < sizeof(struct vnd_comp_header)) || 1320 (vattr.va_size - sizeof(struct vnd_comp_header) < 1321 sizeof(uint64_t)*vnd->sc_comp_numoffs) || 1322 (UQUAD_MAX/vnd->sc_comp_blksz < 1323 vnd->sc_comp_numoffs - 1)) { 1324 VOP_UNLOCK(nd.ni_vp); 1325 error = EINVAL; 1326 goto close_and_exit; 1327 } 1328 1329 /* set decompressed file size */ 1330 KASSERT(vnd->sc_comp_numoffs - 1 <= 1331 UQUAD_MAX/vnd->sc_comp_blksz); 1332 vattr.va_size = 1333 ((u_quad_t)vnd->sc_comp_numoffs - 1) * 1334 (u_quad_t)vnd->sc_comp_blksz; 1335 1336 /* allocate space for all the compressed offsets */ 1337 __CTASSERT(UINT32_MAX <= UQUAD_MAX/sizeof(uint64_t)); 1338 vnd->sc_comp_offsets = 1339 malloc(sizeof(uint64_t) * vnd->sc_comp_numoffs, 1340 M_DEVBUF, M_WAITOK); 1341 1342 /* read in the offsets */ 1343 error = vn_rdwr(UIO_READ, nd.ni_vp, 1344 (void *)vnd->sc_comp_offsets, 1345 sizeof(uint64_t) * vnd->sc_comp_numoffs, 1346 sizeof(struct vnd_comp_header), UIO_SYSSPACE, 1347 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL); 1348 if (error) { 1349 VOP_UNLOCK(nd.ni_vp); 1350 goto close_and_exit; 1351 } 1352 /* 1353 * find largest block size (used for allocation limit). 1354 * Also convert offset to native byte order. 1355 */ 1356 comp_maxsize = 0; 1357 for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) { 1358 vnd->sc_comp_offsets[i] = 1359 be64toh(vnd->sc_comp_offsets[i]); 1360 comp_size = 1361 be64toh(vnd->sc_comp_offsets[i + 1]) 1362 - vnd->sc_comp_offsets[i]; 1363 if (comp_size > comp_maxsize) 1364 comp_maxsize = comp_size; 1365 } 1366 vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] = 1367 be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs 1368 - 1]); 1369 1370 /* create compressed data buffer */ 1371 vnd->sc_comp_buff = malloc(comp_maxsize, 1372 M_DEVBUF, M_WAITOK); 1373 1374 /* create decompressed buffer */ 1375 vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz, 1376 M_DEVBUF, M_WAITOK); 1377 vnd->sc_comp_buffblk = -1; 1378 1379 /* Initialize decompress stream */ 1380 memset(&vnd->sc_comp_stream, 0, sizeof(z_stream)); 1381 vnd->sc_comp_stream.zalloc = vnd_alloc; 1382 vnd->sc_comp_stream.zfree = vnd_free; 1383 error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS); 1384 if (error) { 1385 if (vnd->sc_comp_stream.msg) 1386 printf("vnd%d: compressed file, %s\n", 1387 unit, vnd->sc_comp_stream.msg); 1388 VOP_UNLOCK(nd.ni_vp); 1389 error = EINVAL; 1390 goto close_and_exit; 1391 } 1392 1393 vnd->sc_flags |= VNF_COMP | VNF_READONLY; 1394 #else /* !VND_COMPRESSION */ 1395 VOP_UNLOCK(nd.ni_vp); 1396 error = EOPNOTSUPP; 1397 goto close_and_exit; 1398 #endif /* VND_COMPRESSION */ 1399 } 1400 1401 VOP_UNLOCK(nd.ni_vp); 1402 vnd->sc_vp = nd.ni_vp; 1403 vnd->sc_size = btodb(vattr.va_size); /* note truncation */ 1404 1405 /* 1406 * Use pseudo-geometry specified. If none was provided, 1407 * use "standard" Adaptec fictitious geometry. 1408 */ 1409 if (vio->vnd_flags & VNDIOF_HASGEOM) { 1410 1411 memcpy(&vnd->sc_geom, &vio->vnd_geom, 1412 sizeof(vio->vnd_geom)); 1413 1414 /* 1415 * Sanity-check the sector size. 1416 */ 1417 if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) || 1418 vnd->sc_geom.vng_ncylinders == 0 || 1419 vnd->sc_geom.vng_ntracks == 0 || 1420 vnd->sc_geom.vng_nsectors == 0) { 1421 error = EINVAL; 1422 goto close_and_exit; 1423 } 1424 1425 /* 1426 * Compute the size (in DEV_BSIZE blocks) specified 1427 * by the geometry. 1428 */ 1429 geomsize = (int64_t)vnd->sc_geom.vng_nsectors * 1430 vnd->sc_geom.vng_ntracks * 1431 vnd->sc_geom.vng_ncylinders * 1432 (vnd->sc_geom.vng_secsize / DEV_BSIZE); 1433 1434 /* 1435 * Sanity-check the size against the specified 1436 * geometry. 1437 */ 1438 if (vnd->sc_size < geomsize) { 1439 error = EINVAL; 1440 goto close_and_exit; 1441 } 1442 } else if (vnd->sc_size >= (32 * 64)) { 1443 /* 1444 * Size must be at least 2048 DEV_BSIZE blocks 1445 * (1M) in order to use this geometry. 1446 */ 1447 vnd->sc_geom.vng_secsize = DEV_BSIZE; 1448 vnd->sc_geom.vng_nsectors = 32; 1449 vnd->sc_geom.vng_ntracks = 64; 1450 vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32); 1451 } else { 1452 vnd->sc_geom.vng_secsize = DEV_BSIZE; 1453 vnd->sc_geom.vng_nsectors = 1; 1454 vnd->sc_geom.vng_ntracks = 1; 1455 vnd->sc_geom.vng_ncylinders = vnd->sc_size; 1456 } 1457 1458 vnd_set_geometry(vnd); 1459 1460 if (vio->vnd_flags & VNDIOF_READONLY) { 1461 vnd->sc_flags |= VNF_READONLY; 1462 } 1463 1464 if ((error = vndsetcred(vnd, l->l_cred)) != 0) 1465 goto close_and_exit; 1466 1467 vndthrottle(vnd, vnd->sc_vp); 1468 vio->vnd_osize = dbtob(vnd->sc_size); 1469 #ifdef VNDIOCSET50 1470 if (cmd != VNDIOCSET50) 1471 #endif 1472 vio->vnd_size = dbtob(vnd->sc_size); 1473 vnd->sc_flags |= VNF_INITED; 1474 1475 /* create the kernel thread, wait for it to be up */ 1476 error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd, 1477 &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev)); 1478 if (error) 1479 goto close_and_exit; 1480 while ((vnd->sc_flags & VNF_KTHREAD) == 0) { 1481 tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0); 1482 } 1483 #ifdef DEBUG 1484 if (vnddebug & VDB_INIT) 1485 printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n", 1486 vnd->sc_vp, (unsigned long) vnd->sc_size, 1487 vnd->sc_geom.vng_secsize, 1488 vnd->sc_geom.vng_nsectors, 1489 vnd->sc_geom.vng_ntracks, 1490 vnd->sc_geom.vng_ncylinders); 1491 #endif 1492 1493 /* Attach the disk. */ 1494 disk_attach(&vnd->sc_dkdev); 1495 1496 /* Initialize the xfer and buffer pools. */ 1497 pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0, 1498 0, 0, "vndxpl", NULL, IPL_BIO); 1499 1500 vndunlock(vnd); 1501 1502 pathbuf_destroy(pb); 1503 1504 /* Discover wedges on this disk */ 1505 dkwedge_discover(&vnd->sc_dkdev); 1506 1507 break; 1508 1509 close_and_exit: 1510 (void) vn_close(nd.ni_vp, fflags, l->l_cred); 1511 pathbuf_destroy(pb); 1512 unlock_and_exit: 1513 #ifdef VND_COMPRESSION 1514 /* free any allocated memory (for compressed file) */ 1515 if (vnd->sc_comp_offsets) { 1516 free(vnd->sc_comp_offsets, M_DEVBUF); 1517 vnd->sc_comp_offsets = NULL; 1518 } 1519 if (vnd->sc_comp_buff) { 1520 free(vnd->sc_comp_buff, M_DEVBUF); 1521 vnd->sc_comp_buff = NULL; 1522 } 1523 if (vnd->sc_comp_decombuf) { 1524 free(vnd->sc_comp_decombuf, M_DEVBUF); 1525 vnd->sc_comp_decombuf = NULL; 1526 } 1527 #endif /* VND_COMPRESSION */ 1528 vndunlock(vnd); 1529 return error; 1530 1531 #ifdef VNDIOCCLR50 1532 case VNDIOCCLR50: 1533 #endif 1534 case VNDIOCCLR: 1535 part = DISKPART(dev); 1536 pmask = (1 << part); 1537 force = (vio->vnd_flags & VNDIOF_FORCE) != 0; 1538 1539 if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0) 1540 return error; 1541 1542 break; 1543 1544 1545 case DIOCWDINFO: 1546 case DIOCSDINFO: 1547 #ifdef __HAVE_OLD_DISKLABEL 1548 case ODIOCWDINFO: 1549 case ODIOCSDINFO: 1550 #endif 1551 { 1552 struct disklabel *lp; 1553 1554 if ((error = vndlock(vnd)) != 0) 1555 return error; 1556 1557 vnd->sc_flags |= VNF_LABELLING; 1558 1559 #ifdef __HAVE_OLD_DISKLABEL 1560 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1561 memset(&newlabel, 0, sizeof newlabel); 1562 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1563 lp = &newlabel; 1564 } else 1565 #endif 1566 lp = (struct disklabel *)data; 1567 1568 error = setdisklabel(vnd->sc_dkdev.dk_label, 1569 lp, 0, vnd->sc_dkdev.dk_cpulabel); 1570 if (error == 0) { 1571 if (cmd == DIOCWDINFO 1572 #ifdef __HAVE_OLD_DISKLABEL 1573 || cmd == ODIOCWDINFO 1574 #endif 1575 ) 1576 error = writedisklabel(VNDLABELDEV(dev), 1577 vndstrategy, vnd->sc_dkdev.dk_label, 1578 vnd->sc_dkdev.dk_cpulabel); 1579 } 1580 1581 vnd->sc_flags &= ~VNF_LABELLING; 1582 1583 vndunlock(vnd); 1584 1585 if (error) 1586 return error; 1587 break; 1588 } 1589 1590 case DIOCKLABEL: 1591 if (*(int *)data != 0) 1592 vnd->sc_flags |= VNF_KLABEL; 1593 else 1594 vnd->sc_flags &= ~VNF_KLABEL; 1595 break; 1596 1597 case DIOCWLABEL: 1598 if (*(int *)data != 0) 1599 vnd->sc_flags |= VNF_WLABEL; 1600 else 1601 vnd->sc_flags &= ~VNF_WLABEL; 1602 break; 1603 1604 case DIOCGDEFLABEL: 1605 vndgetdefaultlabel(vnd, (struct disklabel *)data); 1606 break; 1607 1608 #ifdef __HAVE_OLD_DISKLABEL 1609 case ODIOCGDEFLABEL: 1610 vndgetdefaultlabel(vnd, &newlabel); 1611 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1612 return ENOTTY; 1613 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1614 break; 1615 #endif 1616 1617 case DIOCCACHESYNC: 1618 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1619 error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred, 1620 FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0); 1621 VOP_UNLOCK(vnd->sc_vp); 1622 return error; 1623 1624 default: 1625 return ENOTTY; 1626 } 1627 1628 return 0; 1629 } 1630 1631 /* 1632 * Duplicate the current processes' credentials. Since we are called only 1633 * as the result of a SET ioctl and only root can do that, any future access 1634 * to this "disk" is essentially as root. Note that credentials may change 1635 * if some other uid can write directly to the mapped file (NFS). 1636 */ 1637 static int 1638 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred) 1639 { 1640 struct uio auio; 1641 struct iovec aiov; 1642 char *tmpbuf; 1643 int error; 1644 1645 vnd->sc_cred = kauth_cred_dup(cred); 1646 tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK); 1647 1648 /* XXX: Horrible kludge to establish credentials for NFS */ 1649 aiov.iov_base = tmpbuf; 1650 aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size)); 1651 auio.uio_iov = &aiov; 1652 auio.uio_iovcnt = 1; 1653 auio.uio_offset = 0; 1654 auio.uio_rw = UIO_READ; 1655 auio.uio_resid = aiov.iov_len; 1656 UIO_SETUP_SYSSPACE(&auio); 1657 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1658 error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred); 1659 if (error == 0) { 1660 /* 1661 * Because vnd does all IO directly through the vnode 1662 * we need to flush (at least) the buffer from the above 1663 * VOP_READ from the buffer cache to prevent cache 1664 * incoherencies. Also, be careful to write dirty 1665 * buffers back to stable storage. 1666 */ 1667 error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred, 1668 curlwp, 0, 0); 1669 } 1670 VOP_UNLOCK(vnd->sc_vp); 1671 1672 free(tmpbuf, M_TEMP); 1673 return error; 1674 } 1675 1676 /* 1677 * Set maxactive based on FS type 1678 */ 1679 static void 1680 vndthrottle(struct vnd_softc *vnd, struct vnode *vp) 1681 { 1682 1683 if (vp->v_tag == VT_NFS) 1684 vnd->sc_maxactive = 2; 1685 else 1686 vnd->sc_maxactive = 8; 1687 1688 if (vnd->sc_maxactive < 1) 1689 vnd->sc_maxactive = 1; 1690 } 1691 1692 #if 0 1693 static void 1694 vndshutdown(void) 1695 { 1696 struct vnd_softc *vnd; 1697 1698 for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++) 1699 if (vnd->sc_flags & VNF_INITED) 1700 vndclear(vnd); 1701 } 1702 #endif 1703 1704 static void 1705 vndclear(struct vnd_softc *vnd, int myminor) 1706 { 1707 struct vnode *vp = vnd->sc_vp; 1708 int fflags = FREAD; 1709 int bmaj, cmaj, i, mn; 1710 int s; 1711 1712 #ifdef DEBUG 1713 if (vnddebug & VDB_FOLLOW) 1714 printf("vndclear(%p): vp %p\n", vnd, vp); 1715 #endif 1716 /* locate the major number */ 1717 bmaj = bdevsw_lookup_major(&vnd_bdevsw); 1718 cmaj = cdevsw_lookup_major(&vnd_cdevsw); 1719 1720 /* Nuke the vnodes for any open instances */ 1721 for (i = 0; i < MAXPARTITIONS; i++) { 1722 mn = DISKMINOR(device_unit(vnd->sc_dev), i); 1723 vdevgone(bmaj, mn, mn, VBLK); 1724 if (mn != myminor) /* XXX avoid to kill own vnode */ 1725 vdevgone(cmaj, mn, mn, VCHR); 1726 } 1727 1728 if ((vnd->sc_flags & VNF_READONLY) == 0) 1729 fflags |= FWRITE; 1730 1731 s = splbio(); 1732 bufq_drain(vnd->sc_tab); 1733 splx(s); 1734 1735 vnd->sc_flags |= VNF_VUNCONF; 1736 wakeup(&vnd->sc_tab); 1737 while (vnd->sc_flags & VNF_KTHREAD) 1738 tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0); 1739 1740 #ifdef VND_COMPRESSION 1741 /* free the compressed file buffers */ 1742 if (vnd->sc_flags & VNF_COMP) { 1743 if (vnd->sc_comp_offsets) { 1744 free(vnd->sc_comp_offsets, M_DEVBUF); 1745 vnd->sc_comp_offsets = NULL; 1746 } 1747 if (vnd->sc_comp_buff) { 1748 free(vnd->sc_comp_buff, M_DEVBUF); 1749 vnd->sc_comp_buff = NULL; 1750 } 1751 if (vnd->sc_comp_decombuf) { 1752 free(vnd->sc_comp_decombuf, M_DEVBUF); 1753 vnd->sc_comp_decombuf = NULL; 1754 } 1755 } 1756 #endif /* VND_COMPRESSION */ 1757 vnd->sc_flags &= 1758 ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL 1759 | VNF_VUNCONF | VNF_COMP | VNF_CLEARING); 1760 if (vp == NULL) 1761 panic("vndclear: null vp"); 1762 (void) vn_close(vp, fflags, vnd->sc_cred); 1763 kauth_cred_free(vnd->sc_cred); 1764 vnd->sc_vp = NULL; 1765 vnd->sc_cred = NULL; 1766 vnd->sc_size = 0; 1767 } 1768 1769 static int 1770 vndsize(dev_t dev) 1771 { 1772 struct vnd_softc *sc; 1773 struct disklabel *lp; 1774 int part, unit, omask; 1775 int size; 1776 1777 unit = vndunit(dev); 1778 sc = device_lookup_private(&vnd_cd, unit); 1779 if (sc == NULL) 1780 return -1; 1781 1782 if ((sc->sc_flags & VNF_INITED) == 0) 1783 return -1; 1784 1785 part = DISKPART(dev); 1786 omask = sc->sc_dkdev.dk_openmask & (1 << part); 1787 lp = sc->sc_dkdev.dk_label; 1788 1789 if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp)) /* XXX */ 1790 return -1; 1791 1792 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1793 size = -1; 1794 else 1795 size = lp->d_partitions[part].p_size * 1796 (lp->d_secsize / DEV_BSIZE); 1797 1798 if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp)) /* XXX */ 1799 return -1; 1800 1801 return size; 1802 } 1803 1804 static int 1805 vnddump(dev_t dev, daddr_t blkno, void *va, 1806 size_t size) 1807 { 1808 1809 /* Not implemented. */ 1810 return ENXIO; 1811 } 1812 1813 static void 1814 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp) 1815 { 1816 struct vndgeom *vng = &sc->sc_geom; 1817 struct partition *pp; 1818 unsigned spb; 1819 1820 memset(lp, 0, sizeof(*lp)); 1821 1822 spb = vng->vng_secsize / DEV_BSIZE; 1823 if (sc->sc_size / spb > UINT32_MAX) 1824 lp->d_secperunit = UINT32_MAX; 1825 else 1826 lp->d_secperunit = sc->sc_size / spb; 1827 lp->d_secsize = vng->vng_secsize; 1828 lp->d_nsectors = vng->vng_nsectors; 1829 lp->d_ntracks = vng->vng_ntracks; 1830 lp->d_ncylinders = vng->vng_ncylinders; 1831 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1832 1833 strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename)); 1834 lp->d_type = DKTYPE_VND; 1835 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1836 lp->d_rpm = 3600; 1837 lp->d_interleave = 1; 1838 lp->d_flags = 0; 1839 1840 pp = &lp->d_partitions[RAW_PART]; 1841 pp->p_offset = 0; 1842 pp->p_size = lp->d_secperunit; 1843 pp->p_fstype = FS_UNUSED; 1844 lp->d_npartitions = RAW_PART + 1; 1845 1846 lp->d_magic = DISKMAGIC; 1847 lp->d_magic2 = DISKMAGIC; 1848 lp->d_checksum = dkcksum(lp); 1849 } 1850 1851 /* 1852 * Read the disklabel from a vnd. If one is not present, create a fake one. 1853 */ 1854 static void 1855 vndgetdisklabel(dev_t dev, struct vnd_softc *sc) 1856 { 1857 const char *errstring; 1858 struct disklabel *lp = sc->sc_dkdev.dk_label; 1859 struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel; 1860 int i; 1861 1862 memset(clp, 0, sizeof(*clp)); 1863 1864 vndgetdefaultlabel(sc, lp); 1865 1866 /* 1867 * Call the generic disklabel extraction routine. 1868 */ 1869 errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp); 1870 if (errstring) { 1871 /* 1872 * Lack of disklabel is common, but we print the warning 1873 * anyway, since it might contain other useful information. 1874 */ 1875 aprint_normal_dev(sc->sc_dev, "%s\n", errstring); 1876 1877 /* 1878 * For historical reasons, if there's no disklabel 1879 * present, all partitions must be FS_BSDFFS and 1880 * occupy the entire disk. 1881 */ 1882 for (i = 0; i < MAXPARTITIONS; i++) { 1883 /* 1884 * Don't wipe out port specific hack (such as 1885 * dos partition hack of i386 port). 1886 */ 1887 if (lp->d_partitions[i].p_size != 0) 1888 continue; 1889 1890 lp->d_partitions[i].p_size = lp->d_secperunit; 1891 lp->d_partitions[i].p_offset = 0; 1892 lp->d_partitions[i].p_fstype = FS_BSDFFS; 1893 } 1894 1895 strncpy(lp->d_packname, "default label", 1896 sizeof(lp->d_packname)); 1897 1898 lp->d_npartitions = MAXPARTITIONS; 1899 lp->d_checksum = dkcksum(lp); 1900 } 1901 } 1902 1903 /* 1904 * Wait interruptibly for an exclusive lock. 1905 * 1906 * XXX 1907 * Several drivers do this; it should be abstracted and made MP-safe. 1908 */ 1909 static int 1910 vndlock(struct vnd_softc *sc) 1911 { 1912 int error; 1913 1914 while ((sc->sc_flags & VNF_LOCKED) != 0) { 1915 sc->sc_flags |= VNF_WANTED; 1916 if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0) 1917 return error; 1918 } 1919 sc->sc_flags |= VNF_LOCKED; 1920 return 0; 1921 } 1922 1923 /* 1924 * Unlock and wake up any waiters. 1925 */ 1926 static void 1927 vndunlock(struct vnd_softc *sc) 1928 { 1929 1930 sc->sc_flags &= ~VNF_LOCKED; 1931 if ((sc->sc_flags & VNF_WANTED) != 0) { 1932 sc->sc_flags &= ~VNF_WANTED; 1933 wakeup(sc); 1934 } 1935 } 1936 1937 #ifdef VND_COMPRESSION 1938 /* compressed file read */ 1939 static void 1940 compstrategy(struct buf *bp, off_t bn) 1941 { 1942 int error; 1943 int unit = vndunit(bp->b_dev); 1944 struct vnd_softc *vnd = 1945 device_lookup_private(&vnd_cd, unit); 1946 u_int32_t comp_block; 1947 struct uio auio; 1948 char *addr; 1949 int s; 1950 1951 /* set up constants for data move */ 1952 auio.uio_rw = UIO_READ; 1953 UIO_SETUP_SYSSPACE(&auio); 1954 1955 /* read, and transfer the data */ 1956 addr = bp->b_data; 1957 bp->b_resid = bp->b_bcount; 1958 s = splbio(); 1959 while (bp->b_resid > 0) { 1960 unsigned length; 1961 size_t length_in_buffer; 1962 u_int32_t offset_in_buffer; 1963 struct iovec aiov; 1964 1965 /* calculate the compressed block number */ 1966 comp_block = bn / (off_t)vnd->sc_comp_blksz; 1967 1968 /* check for good block number */ 1969 if (comp_block >= vnd->sc_comp_numoffs) { 1970 bp->b_error = EINVAL; 1971 splx(s); 1972 return; 1973 } 1974 1975 /* read in the compressed block, if not in buffer */ 1976 if (comp_block != vnd->sc_comp_buffblk) { 1977 length = vnd->sc_comp_offsets[comp_block + 1] - 1978 vnd->sc_comp_offsets[comp_block]; 1979 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1980 error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff, 1981 length, vnd->sc_comp_offsets[comp_block], 1982 UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred, 1983 NULL, NULL); 1984 if (error) { 1985 bp->b_error = error; 1986 VOP_UNLOCK(vnd->sc_vp); 1987 splx(s); 1988 return; 1989 } 1990 /* uncompress the buffer */ 1991 vnd->sc_comp_stream.next_in = vnd->sc_comp_buff; 1992 vnd->sc_comp_stream.avail_in = length; 1993 vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf; 1994 vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz; 1995 inflateReset(&vnd->sc_comp_stream); 1996 error = inflate(&vnd->sc_comp_stream, Z_FINISH); 1997 if (error != Z_STREAM_END) { 1998 if (vnd->sc_comp_stream.msg) 1999 aprint_normal_dev(vnd->sc_dev, 2000 "compressed file, %s\n", 2001 vnd->sc_comp_stream.msg); 2002 bp->b_error = EBADMSG; 2003 VOP_UNLOCK(vnd->sc_vp); 2004 splx(s); 2005 return; 2006 } 2007 vnd->sc_comp_buffblk = comp_block; 2008 VOP_UNLOCK(vnd->sc_vp); 2009 } 2010 2011 /* transfer the usable uncompressed data */ 2012 offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz; 2013 length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer; 2014 if (length_in_buffer > bp->b_resid) 2015 length_in_buffer = bp->b_resid; 2016 auio.uio_iov = &aiov; 2017 auio.uio_iovcnt = 1; 2018 aiov.iov_base = addr; 2019 aiov.iov_len = length_in_buffer; 2020 auio.uio_resid = aiov.iov_len; 2021 auio.uio_offset = 0; 2022 error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer, 2023 length_in_buffer, &auio); 2024 if (error) { 2025 bp->b_error = error; 2026 splx(s); 2027 return; 2028 } 2029 2030 bn += length_in_buffer; 2031 addr += length_in_buffer; 2032 bp->b_resid -= length_in_buffer; 2033 } 2034 splx(s); 2035 } 2036 2037 /* compression memory allocation routines */ 2038 static void * 2039 vnd_alloc(void *aux, u_int items, u_int siz) 2040 { 2041 return malloc(items * siz, M_TEMP, M_NOWAIT); 2042 } 2043 2044 static void 2045 vnd_free(void *aux, void *ptr) 2046 { 2047 free(ptr, M_TEMP); 2048 } 2049 #endif /* VND_COMPRESSION */ 2050 2051 static void 2052 vnd_set_geometry(struct vnd_softc *vnd) 2053 { 2054 struct disk_geom *dg = &vnd->sc_dkdev.dk_geom; 2055 2056 memset(dg, 0, sizeof(*dg)); 2057 2058 dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors * 2059 vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders; 2060 dg->dg_secsize = vnd->sc_geom.vng_secsize; 2061 dg->dg_nsectors = vnd->sc_geom.vng_nsectors; 2062 dg->dg_ntracks = vnd->sc_geom.vng_ntracks; 2063 dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders; 2064 2065 #ifdef DEBUG 2066 if (vnddebug & VDB_LABEL) { 2067 printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit); 2068 printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders); 2069 } 2070 #endif 2071 disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL); 2072 } 2073 2074 #ifdef VND_COMPRESSION 2075 #define VND_DEPENDS "zlib" 2076 #else 2077 #define VND_DEPENDS NULL 2078 #endif 2079 2080 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS); 2081 2082 #ifdef _MODULE 2083 int vnd_bmajor = -1, vnd_cmajor = -1; 2084 2085 CFDRIVER_DECL(vnd, DV_DISK, NULL); 2086 #endif 2087 2088 static int 2089 vnd_modcmd(modcmd_t cmd, void *arg) 2090 { 2091 int error = 0; 2092 2093 switch (cmd) { 2094 case MODULE_CMD_INIT: 2095 #ifdef _MODULE 2096 error = config_cfdriver_attach(&vnd_cd); 2097 if (error) 2098 break; 2099 2100 error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca); 2101 if (error) { 2102 config_cfdriver_detach(&vnd_cd); 2103 #ifdef DIAGNOSTIC 2104 aprint_error("%s: unable to register cfattach for \n" 2105 "%s, error %d", __func__, vnd_cd.cd_name, error); 2106 #endif 2107 break; 2108 } 2109 2110 /* 2111 * Attach the {b,c}devsw's 2112 */ 2113 error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor, 2114 &vnd_cdevsw, &vnd_cmajor); 2115 /* 2116 * If devsw_attach fails, remove from autoconf database 2117 */ 2118 if (error) { 2119 config_cfattach_detach(vnd_cd.cd_name, &vnd_ca); 2120 config_cfdriver_detach(&vnd_cd); 2121 #ifdef DIAGNOSTIC 2122 aprint_error("%s: unable to attach %s devsw, " 2123 "error %d", __func__, vnd_cd.cd_name, error); 2124 #endif 2125 break; 2126 } 2127 #endif 2128 break; 2129 2130 case MODULE_CMD_FINI: 2131 #ifdef _MODULE 2132 /* 2133 * Remove {b,c}devsw's 2134 */ 2135 devsw_detach(&vnd_bdevsw, &vnd_cdevsw); 2136 2137 /* 2138 * Now remove device from autoconf database 2139 */ 2140 error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca); 2141 if (error) { 2142 (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor, 2143 &vnd_cdevsw, &vnd_cmajor); 2144 #ifdef DIAGNOSTIC 2145 aprint_error("%s: failed to detach %s cfattach, " 2146 "error %d\n", __func__, vnd_cd.cd_name, error); 2147 #endif 2148 break; 2149 } 2150 error = config_cfdriver_detach(&vnd_cd); 2151 if (error) { 2152 (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca); 2153 (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor, 2154 &vnd_cdevsw, &vnd_cmajor); 2155 #ifdef DIAGNOSTIC 2156 aprint_error("%s: failed to detach %s cfdriver, " 2157 "error %d\n", __func__, vnd_cd.cd_name, error); 2158 break; 2159 #endif 2160 } 2161 #endif 2162 break; 2163 2164 case MODULE_CMD_STAT: 2165 return ENOTTY; 2166 2167 default: 2168 return ENOTTY; 2169 } 2170 2171 return error; 2172 } 2173
Indexes created Mon Sep 29 21:09:56 GMT 2025