vnd.c revision 1.267
1 /* $NetBSD: vnd.c,v 1.267 2018/10/07 11:51:26 mlelstv 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 1.267 2018/10/07 11:51:26 mlelstv 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/fstrans.h> 118 #include <sys/file.h> 119 #include <sys/uio.h> 120 #include <sys/conf.h> 121 #include <sys/kauth.h> 122 #include <sys/module.h> 123 124 #include <net/zlib.h> 125 126 #include <miscfs/genfs/genfs.h> 127 #include <miscfs/specfs/specdev.h> 128 129 #include <dev/dkvar.h> 130 #include <dev/vndvar.h> 131 132 #include "ioconf.h" 133 134 #if defined(VNDDEBUG) && !defined(DEBUG) 135 #define DEBUG 136 #endif 137 138 #ifdef DEBUG 139 int dovndcluster = 1; 140 #define VDB_FOLLOW 0x01 141 #define VDB_INIT 0x02 142 #define VDB_IO 0x04 143 #define VDB_LABEL 0x08 144 int vnddebug = 0; 145 #endif 146 147 #define vndunit(x) DISKUNIT(x) 148 149 struct vndxfer { 150 struct buf vx_buf; 151 struct vnd_softc *vx_vnd; 152 }; 153 #define VND_BUFTOXFER(bp) ((struct vndxfer *)(void *)bp) 154 155 #define VND_GETXFER(vnd) pool_get(&(vnd)->sc_vxpool, PR_WAITOK) 156 #define VND_PUTXFER(vnd, vx) pool_put(&(vnd)->sc_vxpool, (vx)) 157 158 #define VNDLABELDEV(dev) \ 159 (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART)) 160 161 #define VND_MAXPENDING(vnd) ((vnd)->sc_maxactive * 4) 162 163 164 static void vndclear(struct vnd_softc *, int); 165 static int vnddoclear(struct vnd_softc *, int, int, bool); 166 static int vndsetcred(struct vnd_softc *, kauth_cred_t); 167 static void vndthrottle(struct vnd_softc *, struct vnode *); 168 static void vndiodone(struct buf *); 169 #if 0 170 static void vndshutdown(void); 171 #endif 172 173 static void vndgetdefaultlabel(struct vnd_softc *, struct disklabel *); 174 static void vndgetdisklabel(dev_t, struct vnd_softc *); 175 176 static int vndlock(struct vnd_softc *); 177 static void vndunlock(struct vnd_softc *); 178 #ifdef VND_COMPRESSION 179 static void compstrategy(struct buf *, off_t); 180 static void *vnd_alloc(void *, u_int, u_int); 181 static void vnd_free(void *, void *); 182 #endif /* VND_COMPRESSION */ 183 184 static void vndthread(void *); 185 static bool vnode_has_op(const struct vnode *, int); 186 static void handle_with_rdwr(struct vnd_softc *, const struct buf *, 187 struct buf *); 188 static void handle_with_strategy(struct vnd_softc *, const struct buf *, 189 struct buf *); 190 static void vnd_set_geometry(struct vnd_softc *); 191 192 static dev_type_open(vndopen); 193 static dev_type_close(vndclose); 194 static dev_type_read(vndread); 195 static dev_type_write(vndwrite); 196 static dev_type_ioctl(vndioctl); 197 static dev_type_strategy(vndstrategy); 198 static dev_type_dump(vnddump); 199 static dev_type_size(vndsize); 200 201 const struct bdevsw vnd_bdevsw = { 202 .d_open = vndopen, 203 .d_close = vndclose, 204 .d_strategy = vndstrategy, 205 .d_ioctl = vndioctl, 206 .d_dump = vnddump, 207 .d_psize = vndsize, 208 .d_discard = nodiscard, 209 .d_flag = D_DISK 210 }; 211 212 const struct cdevsw vnd_cdevsw = { 213 .d_open = vndopen, 214 .d_close = vndclose, 215 .d_read = vndread, 216 .d_write = vndwrite, 217 .d_ioctl = vndioctl, 218 .d_stop = nostop, 219 .d_tty = notty, 220 .d_poll = nopoll, 221 .d_mmap = nommap, 222 .d_kqfilter = nokqfilter, 223 .d_discard = nodiscard, 224 .d_flag = D_DISK 225 }; 226 227 static int vnd_match(device_t, cfdata_t, void *); 228 static void vnd_attach(device_t, device_t, void *); 229 static int vnd_detach(device_t, int); 230 231 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc), 232 vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN); 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 /* Verify that I/O requests cannot be smaller than the 578 * smallest I/O size supported by the backend. 579 */ 580 static bool 581 vnode_has_large_blocks(struct vnd_softc *vnd) 582 { 583 u_int32_t vnd_secsize, iosize; 584 585 iosize = vnd->sc_iosize; 586 vnd_secsize = vnd->sc_geom.vng_secsize; 587 588 return vnd_secsize % iosize != 0; 589 } 590 591 /* XXX this function needs a reliable check to detect 592 * sparse files. Otherwise, bmap/strategy may be used 593 * and fail on non-allocated blocks. VOP_READ/VOP_WRITE 594 * works on sparse files. 595 */ 596 #if notyet 597 static bool 598 vnode_strategy_probe(struct vnd_softc *vnd) 599 { 600 int error; 601 daddr_t nbn; 602 603 if (!vnode_has_strategy(vnd)) 604 return false; 605 606 if (vnode_has_large_blocks(vnd)) 607 return false; 608 609 /* Convert the first logical block number to its 610 * physical block number. 611 */ 612 error = 0; 613 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 614 error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL); 615 VOP_UNLOCK(vnd->sc_vp); 616 617 /* Test if that worked. */ 618 if (error == 0 && (long)nbn == -1) 619 return false; 620 621 return true; 622 } 623 #endif 624 625 static void 626 vndthread(void *arg) 627 { 628 struct vnd_softc *vnd = arg; 629 int s; 630 631 /* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to 632 * directly access the backing vnode. If we can, use these two 633 * operations to avoid messing with the local buffer cache. 634 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations 635 * which are guaranteed to work with any file system. */ 636 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 && 637 ! vnode_has_strategy(vnd)) 638 vnd->sc_flags |= VNF_USE_VN_RDWR; 639 640 /* VOP_STRATEGY can only be used if the backing vnode allows 641 * to access blocks as small as defined by the vnd geometry. 642 */ 643 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 && 644 vnode_has_large_blocks(vnd)) 645 vnd->sc_flags |= VNF_USE_VN_RDWR; 646 647 #ifdef DEBUG 648 if (vnddebug & VDB_INIT) 649 printf("vndthread: vp %p, %s\n", vnd->sc_vp, 650 (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ? 651 "using bmap/strategy operations" : 652 "using read/write operations"); 653 #endif 654 655 s = splbio(); 656 vnd->sc_flags |= VNF_KTHREAD; 657 wakeup(&vnd->sc_kthread); 658 659 /* 660 * Dequeue requests and serve them depending on the available 661 * vnode operations. 662 */ 663 while ((vnd->sc_flags & VNF_VUNCONF) == 0) { 664 struct vndxfer *vnx; 665 struct buf *obp; 666 struct buf *bp; 667 668 obp = bufq_get(vnd->sc_tab); 669 if (obp == NULL) { 670 tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0); 671 continue; 672 }; 673 if ((vnd->sc_flags & VNF_USE_VN_RDWR)) { 674 KASSERT(vnd->sc_pending > 0 && 675 vnd->sc_pending <= VND_MAXPENDING(vnd)); 676 if (vnd->sc_pending-- == VND_MAXPENDING(vnd)) 677 wakeup(&vnd->sc_pending); 678 } 679 splx(s); 680 #ifdef DEBUG 681 if (vnddebug & VDB_FOLLOW) 682 printf("vndthread(%p)\n", obp); 683 #endif 684 685 if (vnd->sc_vp->v_mount == NULL) { 686 obp->b_error = ENXIO; 687 goto done; 688 } 689 #ifdef VND_COMPRESSION 690 /* handle a compressed read */ 691 if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) { 692 off_t bn; 693 694 /* Convert to a byte offset within the file. */ 695 bn = obp->b_rawblkno * 696 vnd->sc_dkdev.dk_label->d_secsize; 697 698 compstrategy(obp, bn); 699 goto done; 700 } 701 #endif /* VND_COMPRESSION */ 702 703 /* 704 * Allocate a header for this transfer and link it to the 705 * buffer 706 */ 707 s = splbio(); 708 vnx = VND_GETXFER(vnd); 709 splx(s); 710 vnx->vx_vnd = vnd; 711 712 s = splbio(); 713 while (vnd->sc_active >= vnd->sc_maxactive) { 714 tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0); 715 } 716 vnd->sc_active++; 717 splx(s); 718 719 /* Instrumentation. */ 720 disk_busy(&vnd->sc_dkdev); 721 722 bp = &vnx->vx_buf; 723 buf_init(bp); 724 bp->b_flags = (obp->b_flags & B_READ); 725 bp->b_oflags = obp->b_oflags; 726 bp->b_cflags = obp->b_cflags; 727 bp->b_iodone = vndiodone; 728 bp->b_private = obp; 729 bp->b_vp = vnd->sc_vp; 730 bp->b_objlock = bp->b_vp->v_interlock; 731 bp->b_data = obp->b_data; 732 bp->b_bcount = obp->b_bcount; 733 BIO_COPYPRIO(bp, obp); 734 735 /* Handle the request using the appropriate operations. */ 736 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0) 737 handle_with_strategy(vnd, obp, bp); 738 else 739 handle_with_rdwr(vnd, obp, bp); 740 741 s = splbio(); 742 continue; 743 744 done: 745 biodone(obp); 746 s = splbio(); 747 } 748 749 vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF); 750 wakeup(&vnd->sc_kthread); 751 splx(s); 752 kthread_exit(0); 753 } 754 755 /* 756 * Checks if the given vnode supports the requested operation. 757 * The operation is specified the offset returned by VOFFSET. 758 * 759 * XXX The test below used to determine this is quite fragile 760 * because it relies on the file system to use genfs to specify 761 * unimplemented operations. There might be another way to do 762 * it more cleanly. 763 */ 764 static bool 765 vnode_has_op(const struct vnode *vp, int opoffset) 766 { 767 int (*defaultp)(void *); 768 int (*opp)(void *); 769 770 defaultp = vp->v_op[VOFFSET(vop_default)]; 771 opp = vp->v_op[opoffset]; 772 773 return opp != defaultp && opp != genfs_eopnotsupp && 774 opp != genfs_badop && opp != genfs_nullop; 775 } 776 777 /* 778 * Handles the read/write request given in 'bp' using the vnode's VOP_READ 779 * and VOP_WRITE operations. 780 * 781 * 'obp' is a pointer to the original request fed to the vnd device. 782 */ 783 static void 784 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp) 785 { 786 bool doread; 787 off_t offset; 788 size_t len, resid; 789 struct vnode *vp; 790 791 doread = bp->b_flags & B_READ; 792 offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize; 793 len = bp->b_bcount; 794 vp = vnd->sc_vp; 795 796 #if defined(DEBUG) 797 if (vnddebug & VDB_IO) 798 printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64 799 ", secsize %d, offset %" PRIu64 800 ", bcount %d\n", 801 vp, doread ? "read" : "write", obp->b_rawblkno, 802 vnd->sc_dkdev.dk_label->d_secsize, offset, 803 bp->b_bcount); 804 #endif 805 806 /* Make sure the request succeeds while suspending this fs. */ 807 fstrans_start_lazy(vp->v_mount); 808 809 /* Issue the read or write operation. */ 810 bp->b_error = 811 vn_rdwr(doread ? UIO_READ : UIO_WRITE, 812 vp, bp->b_data, len, offset, UIO_SYSSPACE, 813 IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL); 814 bp->b_resid = resid; 815 816 mutex_enter(vp->v_interlock); 817 (void) VOP_PUTPAGES(vp, 0, 0, 818 PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 819 820 fstrans_done(vp->v_mount); 821 822 /* We need to increase the number of outputs on the vnode if 823 * there was any write to it. */ 824 if (!doread) { 825 mutex_enter(vp->v_interlock); 826 vp->v_numoutput++; 827 mutex_exit(vp->v_interlock); 828 } 829 830 biodone(bp); 831 } 832 833 /* 834 * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP 835 * and VOP_STRATEGY operations. 836 * 837 * 'obp' is a pointer to the original request fed to the vnd device. 838 */ 839 static void 840 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp, 841 struct buf *bp) 842 { 843 int bsize, error, flags, skipped; 844 size_t resid, sz; 845 off_t bn, offset; 846 struct vnode *vp; 847 struct buf *nbp = NULL; 848 849 flags = obp->b_flags; 850 851 852 /* convert to a byte offset within the file. */ 853 bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize; 854 855 bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize; 856 skipped = 0; 857 858 /* 859 * Break the request into bsize pieces and feed them 860 * sequentially using VOP_BMAP/VOP_STRATEGY. 861 * We do it this way to keep from flooding NFS servers if we 862 * are connected to an NFS file. This places the burden on 863 * the client rather than the server. 864 */ 865 error = 0; 866 bp->b_resid = bp->b_bcount; 867 for (offset = 0, resid = bp->b_resid; /* true */; 868 resid -= sz, offset += sz) { 869 daddr_t nbn; 870 int off, nra; 871 872 nra = 0; 873 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 874 error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra); 875 VOP_UNLOCK(vnd->sc_vp); 876 877 if (error == 0 && (long)nbn == -1) 878 error = EIO; 879 880 /* 881 * If there was an error or a hole in the file...punt. 882 * Note that we may have to wait for any operations 883 * that we have already fired off before releasing 884 * the buffer. 885 * 886 * XXX we could deal with holes here but it would be 887 * a hassle (in the write case). 888 */ 889 if (error) { 890 skipped += resid; 891 break; 892 } 893 894 #ifdef DEBUG 895 if (!dovndcluster) 896 nra = 0; 897 #endif 898 899 off = bn % bsize; 900 sz = MIN(((off_t)1 + nra) * bsize - off, resid); 901 #ifdef DEBUG 902 if (vnddebug & VDB_IO) 903 printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64 904 " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn, 905 nbn, sz); 906 #endif 907 908 nbp = getiobuf(vp, true); 909 nestiobuf_setup(bp, nbp, offset, sz); 910 nbp->b_blkno = nbn + btodb(off); 911 912 #if 0 /* XXX #ifdef DEBUG */ 913 if (vnddebug & VDB_IO) 914 printf("vndstart(%ld): bp %p vp %p blkno " 915 "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n", 916 (long) (vnd-vnd_softc), &nbp->vb_buf, 917 nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno, 918 nbp->vb_buf.b_flags, nbp->vb_buf.b_data, 919 nbp->vb_buf.b_bcount); 920 #endif 921 if (resid == sz) { 922 break; 923 } 924 VOP_STRATEGY(vp, nbp); 925 bn += sz; 926 } 927 if (!(flags & B_READ)) { 928 struct vnode *w_vp; 929 /* 930 * this is the last nested buf, account for 931 * the parent buf write too. 932 * This has to be done last, so that 933 * fsync won't wait for this write which 934 * has no chance to complete before all nested bufs 935 * have been queued. But it has to be done 936 * before the last VOP_STRATEGY() 937 * or the call to nestiobuf_done(). 938 */ 939 w_vp = bp->b_vp; 940 mutex_enter(w_vp->v_interlock); 941 w_vp->v_numoutput++; 942 mutex_exit(w_vp->v_interlock); 943 } 944 KASSERT(skipped != 0 || nbp != NULL); 945 if (skipped) 946 nestiobuf_done(bp, skipped, error); 947 else 948 VOP_STRATEGY(vp, nbp); 949 } 950 951 static void 952 vndiodone(struct buf *bp) 953 { 954 struct vndxfer *vnx = VND_BUFTOXFER(bp); 955 struct vnd_softc *vnd = vnx->vx_vnd; 956 struct buf *obp = bp->b_private; 957 int s = splbio(); 958 959 KASSERT(&vnx->vx_buf == bp); 960 KASSERT(vnd->sc_active > 0); 961 #ifdef DEBUG 962 if (vnddebug & VDB_IO) { 963 printf("vndiodone1: bp %p iodone: error %d\n", 964 bp, bp->b_error); 965 } 966 #endif 967 disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid, 968 (bp->b_flags & B_READ)); 969 vnd->sc_active--; 970 if (vnd->sc_active == 0) { 971 wakeup(&vnd->sc_tab); 972 } 973 splx(s); 974 obp->b_error = bp->b_error; 975 obp->b_resid = bp->b_resid; 976 buf_destroy(bp); 977 VND_PUTXFER(vnd, vnx); 978 biodone(obp); 979 } 980 981 /* ARGSUSED */ 982 static int 983 vndread(dev_t dev, struct uio *uio, int flags) 984 { 985 int unit = vndunit(dev); 986 struct vnd_softc *sc; 987 988 #ifdef DEBUG 989 if (vnddebug & VDB_FOLLOW) 990 printf("vndread(0x%"PRIx64", %p)\n", dev, uio); 991 #endif 992 993 sc = device_lookup_private(&vnd_cd, unit); 994 if (sc == NULL) 995 return ENXIO; 996 997 if ((sc->sc_flags & VNF_INITED) == 0) 998 return ENXIO; 999 1000 return physio(vndstrategy, NULL, dev, B_READ, minphys, uio); 1001 } 1002 1003 /* ARGSUSED */ 1004 static int 1005 vndwrite(dev_t dev, struct uio *uio, int flags) 1006 { 1007 int unit = vndunit(dev); 1008 struct vnd_softc *sc; 1009 1010 #ifdef DEBUG 1011 if (vnddebug & VDB_FOLLOW) 1012 printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio); 1013 #endif 1014 1015 sc = device_lookup_private(&vnd_cd, unit); 1016 if (sc == NULL) 1017 return ENXIO; 1018 1019 if ((sc->sc_flags & VNF_INITED) == 0) 1020 return ENXIO; 1021 1022 return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio); 1023 } 1024 1025 static int 1026 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va) 1027 { 1028 int error; 1029 struct vnd_softc *vnd; 1030 1031 if (*un == -1) 1032 *un = unit; 1033 if (*un < 0) 1034 return EINVAL; 1035 1036 vnd = device_lookup_private(&vnd_cd, *un); 1037 if (vnd == NULL) 1038 return -1; 1039 1040 if ((vnd->sc_flags & VNF_INITED) == 0) 1041 return -1; 1042 1043 vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY); 1044 error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred); 1045 VOP_UNLOCK(vnd->sc_vp); 1046 return error; 1047 } 1048 1049 static int 1050 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force) 1051 { 1052 int error; 1053 1054 if ((error = vndlock(vnd)) != 0) 1055 return error; 1056 1057 /* 1058 * Don't unconfigure if any other partitions are open 1059 * or if both the character and block flavors of this 1060 * partition are open. 1061 */ 1062 if (DK_BUSY(vnd, pmask) && !force) { 1063 vndunlock(vnd); 1064 return EBUSY; 1065 } 1066 1067 /* Delete all of our wedges */ 1068 dkwedge_delall(&vnd->sc_dkdev); 1069 1070 /* 1071 * XXX vndclear() might call vndclose() implicitly; 1072 * release lock to avoid recursion 1073 * 1074 * Set VNF_CLEARING to prevent vndopen() from 1075 * sneaking in after we vndunlock(). 1076 */ 1077 vnd->sc_flags |= VNF_CLEARING; 1078 vndunlock(vnd); 1079 vndclear(vnd, minor); 1080 #ifdef DEBUG 1081 if (vnddebug & VDB_INIT) 1082 printf("%s: CLRed\n", __func__); 1083 #endif 1084 1085 /* Destroy the xfer and buffer pools. */ 1086 pool_destroy(&vnd->sc_vxpool); 1087 1088 /* Detach the disk. */ 1089 disk_detach(&vnd->sc_dkdev); 1090 1091 return 0; 1092 } 1093 1094 static int 1095 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va) 1096 { 1097 int error; 1098 1099 KASSERT(l); 1100 1101 /* the first member is always int vnd_unit in all the versions */ 1102 if (*(int *)data >= vnd_cd.cd_ndevs) 1103 return ENXIO; 1104 1105 switch (error = vnd_cget(l, unit, (int *)data, va)) { 1106 case -1: 1107 /* unused is not an error */ 1108 memset(va, 0, sizeof(*va)); 1109 /*FALLTHROUGH*/ 1110 case 0: 1111 return 0; 1112 default: 1113 return error; 1114 } 1115 } 1116 1117 /* ARGSUSED */ 1118 static int 1119 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 1120 { 1121 bool force; 1122 int unit = vndunit(dev); 1123 struct vnd_softc *vnd; 1124 struct vnd_ioctl *vio; 1125 struct vattr vattr; 1126 struct pathbuf *pb; 1127 struct nameidata nd; 1128 int error, part, pmask; 1129 uint64_t geomsize; 1130 int fflags; 1131 #ifdef __HAVE_OLD_DISKLABEL 1132 struct disklabel newlabel; 1133 #endif 1134 1135 #ifdef DEBUG 1136 if (vnddebug & VDB_FOLLOW) 1137 printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n", 1138 dev, cmd, data, flag, l->l_proc, unit); 1139 #endif 1140 /* Do the get's first; they don't need initialization or verification */ 1141 switch (cmd) { 1142 #ifdef COMPAT_30 1143 case VNDIOCGET30: { 1144 if ((error = vndioctl_get(l, data, unit, &vattr)) != 0) 1145 return error; 1146 1147 struct vnd_user30 *vnu = data; 1148 vnu->vnu_dev = vattr.va_fsid; 1149 vnu->vnu_ino = vattr.va_fileid; 1150 return 0; 1151 } 1152 #endif 1153 #ifdef COMPAT_50 1154 case VNDIOCGET50: { 1155 if ((error = vndioctl_get(l, data, unit, &vattr)) != 0) 1156 return error; 1157 1158 struct vnd_user50 *vnu = data; 1159 vnu->vnu_dev = vattr.va_fsid; 1160 vnu->vnu_ino = vattr.va_fileid; 1161 return 0; 1162 } 1163 #endif 1164 1165 case VNDIOCGET: { 1166 if ((error = vndioctl_get(l, data, unit, &vattr)) != 0) 1167 return error; 1168 1169 struct vnd_user *vnu = data; 1170 vnu->vnu_dev = vattr.va_fsid; 1171 vnu->vnu_ino = vattr.va_fileid; 1172 return 0; 1173 } 1174 default: 1175 break; 1176 } 1177 1178 vnd = device_lookup_private(&vnd_cd, unit); 1179 if (vnd == NULL) 1180 return ENXIO; 1181 vio = (struct vnd_ioctl *)data; 1182 1183 /* Must be open for writes for these commands... */ 1184 switch (cmd) { 1185 case VNDIOCSET: 1186 case VNDIOCCLR: 1187 #ifdef COMPAT_50 1188 case VNDIOCSET50: 1189 case VNDIOCCLR50: 1190 #endif 1191 case DIOCSDINFO: 1192 case DIOCWDINFO: 1193 #ifdef __HAVE_OLD_DISKLABEL 1194 case ODIOCSDINFO: 1195 case ODIOCWDINFO: 1196 #endif 1197 case DIOCKLABEL: 1198 case DIOCWLABEL: 1199 if ((flag & FWRITE) == 0) 1200 return EBADF; 1201 } 1202 1203 /* Must be initialized for these... */ 1204 switch (cmd) { 1205 case VNDIOCCLR: 1206 #ifdef VNDIOCCLR50 1207 case VNDIOCCLR50: 1208 #endif 1209 case DIOCGDINFO: 1210 case DIOCSDINFO: 1211 case DIOCWDINFO: 1212 case DIOCGPARTINFO: 1213 case DIOCKLABEL: 1214 case DIOCWLABEL: 1215 case DIOCGDEFLABEL: 1216 case DIOCCACHESYNC: 1217 #ifdef __HAVE_OLD_DISKLABEL 1218 case ODIOCGDINFO: 1219 case ODIOCSDINFO: 1220 case ODIOCWDINFO: 1221 case ODIOCGDEFLABEL: 1222 #endif 1223 if ((vnd->sc_flags & VNF_INITED) == 0) 1224 return ENXIO; 1225 } 1226 1227 error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l); 1228 if (error != EPASSTHROUGH) 1229 return error; 1230 1231 1232 switch (cmd) { 1233 #ifdef VNDIOCSET50 1234 case VNDIOCSET50: 1235 #endif 1236 case VNDIOCSET: 1237 if (vnd->sc_flags & VNF_INITED) 1238 return EBUSY; 1239 1240 if ((error = vndlock(vnd)) != 0) 1241 return error; 1242 1243 fflags = FREAD; 1244 if ((vio->vnd_flags & VNDIOF_READONLY) == 0) 1245 fflags |= FWRITE; 1246 if ((vio->vnd_flags & VNDIOF_FILEIO) != 0) 1247 vnd->sc_flags |= VNF_USE_VN_RDWR; 1248 error = pathbuf_copyin(vio->vnd_file, &pb); 1249 if (error) { 1250 goto unlock_and_exit; 1251 } 1252 NDINIT(&nd, LOOKUP, FOLLOW, pb); 1253 if ((error = vn_open(&nd, fflags, 0)) != 0) { 1254 pathbuf_destroy(pb); 1255 goto unlock_and_exit; 1256 } 1257 KASSERT(l); 1258 error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred); 1259 if (!error && nd.ni_vp->v_type != VREG) 1260 error = EOPNOTSUPP; 1261 if (!error && vattr.va_bytes < vattr.va_size) 1262 /* File is definitely sparse, use vn_rdwr() */ 1263 vnd->sc_flags |= VNF_USE_VN_RDWR; 1264 if (error) { 1265 VOP_UNLOCK(nd.ni_vp); 1266 goto close_and_exit; 1267 } 1268 1269 /* If using a compressed file, initialize its info */ 1270 /* (or abort with an error if kernel has no compression) */ 1271 if (vio->vnd_flags & VNDIOF_COMP) { 1272 #ifdef VND_COMPRESSION 1273 struct vnd_comp_header *ch; 1274 int i; 1275 uint32_t comp_size; 1276 uint32_t comp_maxsize; 1277 1278 /* allocate space for compresed file header */ 1279 ch = malloc(sizeof(struct vnd_comp_header), 1280 M_TEMP, M_WAITOK); 1281 1282 /* read compressed file header */ 1283 error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch, 1284 sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE, 1285 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL); 1286 if (error) { 1287 free(ch, M_TEMP); 1288 VOP_UNLOCK(nd.ni_vp); 1289 goto close_and_exit; 1290 } 1291 1292 if (be32toh(ch->block_size) == 0 || 1293 be32toh(ch->num_blocks) > UINT32_MAX - 1) { 1294 free(ch, M_TEMP); 1295 VOP_UNLOCK(nd.ni_vp); 1296 goto close_and_exit; 1297 } 1298 1299 /* save some header info */ 1300 vnd->sc_comp_blksz = be32toh(ch->block_size); 1301 /* note last offset is the file byte size */ 1302 vnd->sc_comp_numoffs = be32toh(ch->num_blocks) + 1; 1303 free(ch, M_TEMP); 1304 if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) { 1305 VOP_UNLOCK(nd.ni_vp); 1306 error = EINVAL; 1307 goto close_and_exit; 1308 } 1309 KASSERT(0 < vnd->sc_comp_blksz); 1310 KASSERT(0 < vnd->sc_comp_numoffs); 1311 /* 1312 * @#^@!$& gcc -Wtype-limits refuses to let me 1313 * write SIZE_MAX/sizeof(uint64_t) < numoffs, 1314 * because the range of the type on amd64 makes 1315 * the comparisons always false. 1316 */ 1317 #if SIZE_MAX <= UINT32_MAX*(64/CHAR_BIT) 1318 if (SIZE_MAX/sizeof(uint64_t) < vnd->sc_comp_numoffs) { 1319 VOP_UNLOCK(nd.ni_vp); 1320 error = EINVAL; 1321 goto close_and_exit; 1322 } 1323 #endif 1324 if ((vattr.va_size < sizeof(struct vnd_comp_header)) || 1325 (vattr.va_size - sizeof(struct vnd_comp_header) < 1326 sizeof(uint64_t)*vnd->sc_comp_numoffs) || 1327 (UQUAD_MAX/vnd->sc_comp_blksz < 1328 vnd->sc_comp_numoffs - 1)) { 1329 VOP_UNLOCK(nd.ni_vp); 1330 error = EINVAL; 1331 goto close_and_exit; 1332 } 1333 1334 /* set decompressed file size */ 1335 KASSERT(vnd->sc_comp_numoffs - 1 <= 1336 UQUAD_MAX/vnd->sc_comp_blksz); 1337 vattr.va_size = 1338 ((u_quad_t)vnd->sc_comp_numoffs - 1) * 1339 (u_quad_t)vnd->sc_comp_blksz; 1340 1341 /* allocate space for all the compressed offsets */ 1342 __CTASSERT(UINT32_MAX <= UQUAD_MAX/sizeof(uint64_t)); 1343 vnd->sc_comp_offsets = 1344 malloc(sizeof(uint64_t) * vnd->sc_comp_numoffs, 1345 M_DEVBUF, M_WAITOK); 1346 1347 /* read in the offsets */ 1348 error = vn_rdwr(UIO_READ, nd.ni_vp, 1349 (void *)vnd->sc_comp_offsets, 1350 sizeof(uint64_t) * vnd->sc_comp_numoffs, 1351 sizeof(struct vnd_comp_header), UIO_SYSSPACE, 1352 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL); 1353 if (error) { 1354 VOP_UNLOCK(nd.ni_vp); 1355 goto close_and_exit; 1356 } 1357 /* 1358 * find largest block size (used for allocation limit). 1359 * Also convert offset to native byte order. 1360 */ 1361 comp_maxsize = 0; 1362 for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) { 1363 vnd->sc_comp_offsets[i] = 1364 be64toh(vnd->sc_comp_offsets[i]); 1365 comp_size = 1366 be64toh(vnd->sc_comp_offsets[i + 1]) 1367 - vnd->sc_comp_offsets[i]; 1368 if (comp_size > comp_maxsize) 1369 comp_maxsize = comp_size; 1370 } 1371 vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] = 1372 be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs 1373 - 1]); 1374 1375 /* create compressed data buffer */ 1376 vnd->sc_comp_buff = malloc(comp_maxsize, 1377 M_DEVBUF, M_WAITOK); 1378 1379 /* create decompressed buffer */ 1380 vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz, 1381 M_DEVBUF, M_WAITOK); 1382 vnd->sc_comp_buffblk = -1; 1383 1384 /* Initialize decompress stream */ 1385 memset(&vnd->sc_comp_stream, 0, sizeof(z_stream)); 1386 vnd->sc_comp_stream.zalloc = vnd_alloc; 1387 vnd->sc_comp_stream.zfree = vnd_free; 1388 error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS); 1389 if (error) { 1390 if (vnd->sc_comp_stream.msg) 1391 printf("vnd%d: compressed file, %s\n", 1392 unit, vnd->sc_comp_stream.msg); 1393 VOP_UNLOCK(nd.ni_vp); 1394 error = EINVAL; 1395 goto close_and_exit; 1396 } 1397 1398 vnd->sc_flags |= VNF_COMP | VNF_READONLY; 1399 #else /* !VND_COMPRESSION */ 1400 VOP_UNLOCK(nd.ni_vp); 1401 error = EOPNOTSUPP; 1402 goto close_and_exit; 1403 #endif /* VND_COMPRESSION */ 1404 } 1405 1406 VOP_UNLOCK(nd.ni_vp); 1407 vnd->sc_vp = nd.ni_vp; 1408 vnd->sc_size = btodb(vattr.va_size); /* note truncation */ 1409 1410 /* get smallest I/O size for underlying device, fall back to 1411 * fundamental I/O size of underlying filesystem 1412 */ 1413 error = bdev_ioctl(vattr.va_fsid, DIOCGSECTORSIZE, &vnd->sc_iosize, FKIOCTL, l); 1414 if (error) 1415 vnd->sc_iosize = vnd->sc_vp->v_mount->mnt_stat.f_frsize; 1416 1417 /* 1418 * Use pseudo-geometry specified. If none was provided, 1419 * use "standard" Adaptec fictitious geometry. 1420 */ 1421 if (vio->vnd_flags & VNDIOF_HASGEOM) { 1422 1423 memcpy(&vnd->sc_geom, &vio->vnd_geom, 1424 sizeof(vio->vnd_geom)); 1425 1426 /* 1427 * Sanity-check the sector size. 1428 */ 1429 if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) || 1430 vnd->sc_geom.vng_ncylinders == 0 || 1431 vnd->sc_geom.vng_ntracks == 0 || 1432 vnd->sc_geom.vng_nsectors == 0) { 1433 error = EINVAL; 1434 goto close_and_exit; 1435 } 1436 1437 /* 1438 * Compute the size (in DEV_BSIZE blocks) specified 1439 * by the geometry. 1440 */ 1441 geomsize = (int64_t)vnd->sc_geom.vng_nsectors * 1442 vnd->sc_geom.vng_ntracks * 1443 vnd->sc_geom.vng_ncylinders * 1444 (vnd->sc_geom.vng_secsize / DEV_BSIZE); 1445 1446 /* 1447 * Sanity-check the size against the specified 1448 * geometry. 1449 */ 1450 if (vnd->sc_size < geomsize) { 1451 error = EINVAL; 1452 goto close_and_exit; 1453 } 1454 } else if (vnd->sc_size >= (32 * 64)) { 1455 /* 1456 * Size must be at least 2048 DEV_BSIZE blocks 1457 * (1M) in order to use this geometry. 1458 */ 1459 vnd->sc_geom.vng_secsize = DEV_BSIZE; 1460 vnd->sc_geom.vng_nsectors = 32; 1461 vnd->sc_geom.vng_ntracks = 64; 1462 vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32); 1463 } else { 1464 vnd->sc_geom.vng_secsize = DEV_BSIZE; 1465 vnd->sc_geom.vng_nsectors = 1; 1466 vnd->sc_geom.vng_ntracks = 1; 1467 vnd->sc_geom.vng_ncylinders = vnd->sc_size; 1468 } 1469 1470 vnd_set_geometry(vnd); 1471 1472 if (vio->vnd_flags & VNDIOF_READONLY) { 1473 vnd->sc_flags |= VNF_READONLY; 1474 } 1475 1476 if ((error = vndsetcred(vnd, l->l_cred)) != 0) 1477 goto close_and_exit; 1478 1479 vndthrottle(vnd, vnd->sc_vp); 1480 vio->vnd_osize = dbtob(vnd->sc_size); 1481 #ifdef VNDIOCSET50 1482 if (cmd != VNDIOCSET50) 1483 #endif 1484 vio->vnd_size = dbtob(vnd->sc_size); 1485 vnd->sc_flags |= VNF_INITED; 1486 1487 /* create the kernel thread, wait for it to be up */ 1488 error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd, 1489 &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev)); 1490 if (error) 1491 goto close_and_exit; 1492 while ((vnd->sc_flags & VNF_KTHREAD) == 0) { 1493 tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0); 1494 } 1495 #ifdef DEBUG 1496 if (vnddebug & VDB_INIT) 1497 printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n", 1498 vnd->sc_vp, (unsigned long) vnd->sc_size, 1499 vnd->sc_geom.vng_secsize, 1500 vnd->sc_geom.vng_nsectors, 1501 vnd->sc_geom.vng_ntracks, 1502 vnd->sc_geom.vng_ncylinders); 1503 #endif 1504 1505 /* Attach the disk. */ 1506 disk_attach(&vnd->sc_dkdev); 1507 1508 /* Initialize the xfer and buffer pools. */ 1509 pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0, 1510 0, 0, "vndxpl", NULL, IPL_BIO); 1511 1512 vndunlock(vnd); 1513 1514 pathbuf_destroy(pb); 1515 1516 /* Discover wedges on this disk */ 1517 dkwedge_discover(&vnd->sc_dkdev); 1518 1519 break; 1520 1521 close_and_exit: 1522 (void) vn_close(nd.ni_vp, fflags, l->l_cred); 1523 pathbuf_destroy(pb); 1524 unlock_and_exit: 1525 #ifdef VND_COMPRESSION 1526 /* free any allocated memory (for compressed file) */ 1527 if (vnd->sc_comp_offsets) { 1528 free(vnd->sc_comp_offsets, M_DEVBUF); 1529 vnd->sc_comp_offsets = NULL; 1530 } 1531 if (vnd->sc_comp_buff) { 1532 free(vnd->sc_comp_buff, M_DEVBUF); 1533 vnd->sc_comp_buff = NULL; 1534 } 1535 if (vnd->sc_comp_decombuf) { 1536 free(vnd->sc_comp_decombuf, M_DEVBUF); 1537 vnd->sc_comp_decombuf = NULL; 1538 } 1539 #endif /* VND_COMPRESSION */ 1540 vndunlock(vnd); 1541 return error; 1542 1543 #ifdef VNDIOCCLR50 1544 case VNDIOCCLR50: 1545 #endif 1546 case VNDIOCCLR: 1547 part = DISKPART(dev); 1548 pmask = (1 << part); 1549 force = (vio->vnd_flags & VNDIOF_FORCE) != 0; 1550 1551 if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0) 1552 return error; 1553 1554 break; 1555 1556 1557 case DIOCWDINFO: 1558 case DIOCSDINFO: 1559 #ifdef __HAVE_OLD_DISKLABEL 1560 case ODIOCWDINFO: 1561 case ODIOCSDINFO: 1562 #endif 1563 { 1564 struct disklabel *lp; 1565 1566 if ((error = vndlock(vnd)) != 0) 1567 return error; 1568 1569 vnd->sc_flags |= VNF_LABELLING; 1570 1571 #ifdef __HAVE_OLD_DISKLABEL 1572 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1573 memset(&newlabel, 0, sizeof newlabel); 1574 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1575 lp = &newlabel; 1576 } else 1577 #endif 1578 lp = (struct disklabel *)data; 1579 1580 error = setdisklabel(vnd->sc_dkdev.dk_label, 1581 lp, 0, vnd->sc_dkdev.dk_cpulabel); 1582 if (error == 0) { 1583 if (cmd == DIOCWDINFO 1584 #ifdef __HAVE_OLD_DISKLABEL 1585 || cmd == ODIOCWDINFO 1586 #endif 1587 ) 1588 error = writedisklabel(VNDLABELDEV(dev), 1589 vndstrategy, vnd->sc_dkdev.dk_label, 1590 vnd->sc_dkdev.dk_cpulabel); 1591 } 1592 1593 vnd->sc_flags &= ~VNF_LABELLING; 1594 1595 vndunlock(vnd); 1596 1597 if (error) 1598 return error; 1599 break; 1600 } 1601 1602 case DIOCKLABEL: 1603 if (*(int *)data != 0) 1604 vnd->sc_flags |= VNF_KLABEL; 1605 else 1606 vnd->sc_flags &= ~VNF_KLABEL; 1607 break; 1608 1609 case DIOCWLABEL: 1610 if (*(int *)data != 0) 1611 vnd->sc_flags |= VNF_WLABEL; 1612 else 1613 vnd->sc_flags &= ~VNF_WLABEL; 1614 break; 1615 1616 case DIOCGDEFLABEL: 1617 vndgetdefaultlabel(vnd, (struct disklabel *)data); 1618 break; 1619 1620 #ifdef __HAVE_OLD_DISKLABEL 1621 case ODIOCGDEFLABEL: 1622 vndgetdefaultlabel(vnd, &newlabel); 1623 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1624 return ENOTTY; 1625 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1626 break; 1627 #endif 1628 1629 case DIOCCACHESYNC: 1630 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1631 error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred, 1632 FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0); 1633 VOP_UNLOCK(vnd->sc_vp); 1634 return error; 1635 1636 default: 1637 return ENOTTY; 1638 } 1639 1640 return 0; 1641 } 1642 1643 /* 1644 * Duplicate the current processes' credentials. Since we are called only 1645 * as the result of a SET ioctl and only root can do that, any future access 1646 * to this "disk" is essentially as root. Note that credentials may change 1647 * if some other uid can write directly to the mapped file (NFS). 1648 */ 1649 static int 1650 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred) 1651 { 1652 struct uio auio; 1653 struct iovec aiov; 1654 char *tmpbuf; 1655 int error; 1656 1657 vnd->sc_cred = kauth_cred_dup(cred); 1658 tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK); 1659 1660 /* XXX: Horrible kludge to establish credentials for NFS */ 1661 aiov.iov_base = tmpbuf; 1662 aiov.iov_len = uimin(DEV_BSIZE, dbtob(vnd->sc_size)); 1663 auio.uio_iov = &aiov; 1664 auio.uio_iovcnt = 1; 1665 auio.uio_offset = 0; 1666 auio.uio_rw = UIO_READ; 1667 auio.uio_resid = aiov.iov_len; 1668 UIO_SETUP_SYSSPACE(&auio); 1669 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1670 error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred); 1671 if (error == 0) { 1672 /* 1673 * Because vnd does all IO directly through the vnode 1674 * we need to flush (at least) the buffer from the above 1675 * VOP_READ from the buffer cache to prevent cache 1676 * incoherencies. Also, be careful to write dirty 1677 * buffers back to stable storage. 1678 */ 1679 error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred, 1680 curlwp, 0, 0); 1681 } 1682 VOP_UNLOCK(vnd->sc_vp); 1683 1684 free(tmpbuf, M_TEMP); 1685 return error; 1686 } 1687 1688 /* 1689 * Set maxactive based on FS type 1690 */ 1691 static void 1692 vndthrottle(struct vnd_softc *vnd, struct vnode *vp) 1693 { 1694 1695 if (vp->v_tag == VT_NFS) 1696 vnd->sc_maxactive = 2; 1697 else 1698 vnd->sc_maxactive = 8; 1699 1700 if (vnd->sc_maxactive < 1) 1701 vnd->sc_maxactive = 1; 1702 } 1703 1704 #if 0 1705 static void 1706 vndshutdown(void) 1707 { 1708 struct vnd_softc *vnd; 1709 1710 for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++) 1711 if (vnd->sc_flags & VNF_INITED) 1712 vndclear(vnd); 1713 } 1714 #endif 1715 1716 static void 1717 vndclear(struct vnd_softc *vnd, int myminor) 1718 { 1719 struct vnode *vp = vnd->sc_vp; 1720 int fflags = FREAD; 1721 int bmaj, cmaj, i, mn; 1722 int s; 1723 1724 #ifdef DEBUG 1725 if (vnddebug & VDB_FOLLOW) 1726 printf("vndclear(%p): vp %p\n", vnd, vp); 1727 #endif 1728 /* locate the major number */ 1729 bmaj = bdevsw_lookup_major(&vnd_bdevsw); 1730 cmaj = cdevsw_lookup_major(&vnd_cdevsw); 1731 1732 /* Nuke the vnodes for any open instances */ 1733 for (i = 0; i < MAXPARTITIONS; i++) { 1734 mn = DISKMINOR(device_unit(vnd->sc_dev), i); 1735 vdevgone(bmaj, mn, mn, VBLK); 1736 if (mn != myminor) /* XXX avoid to kill own vnode */ 1737 vdevgone(cmaj, mn, mn, VCHR); 1738 } 1739 1740 if ((vnd->sc_flags & VNF_READONLY) == 0) 1741 fflags |= FWRITE; 1742 1743 s = splbio(); 1744 bufq_drain(vnd->sc_tab); 1745 splx(s); 1746 1747 vnd->sc_flags |= VNF_VUNCONF; 1748 wakeup(&vnd->sc_tab); 1749 while (vnd->sc_flags & VNF_KTHREAD) 1750 tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0); 1751 1752 #ifdef VND_COMPRESSION 1753 /* free the compressed file buffers */ 1754 if (vnd->sc_flags & VNF_COMP) { 1755 if (vnd->sc_comp_offsets) { 1756 free(vnd->sc_comp_offsets, M_DEVBUF); 1757 vnd->sc_comp_offsets = NULL; 1758 } 1759 if (vnd->sc_comp_buff) { 1760 free(vnd->sc_comp_buff, M_DEVBUF); 1761 vnd->sc_comp_buff = NULL; 1762 } 1763 if (vnd->sc_comp_decombuf) { 1764 free(vnd->sc_comp_decombuf, M_DEVBUF); 1765 vnd->sc_comp_decombuf = NULL; 1766 } 1767 } 1768 #endif /* VND_COMPRESSION */ 1769 vnd->sc_flags &= 1770 ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL 1771 | VNF_VUNCONF | VNF_COMP | VNF_CLEARING); 1772 if (vp == NULL) 1773 panic("vndclear: null vp"); 1774 (void) vn_close(vp, fflags, vnd->sc_cred); 1775 kauth_cred_free(vnd->sc_cred); 1776 vnd->sc_vp = NULL; 1777 vnd->sc_cred = NULL; 1778 vnd->sc_size = 0; 1779 } 1780 1781 static int 1782 vndsize(dev_t dev) 1783 { 1784 struct vnd_softc *sc; 1785 struct disklabel *lp; 1786 int part, unit, omask; 1787 int size; 1788 1789 unit = vndunit(dev); 1790 sc = device_lookup_private(&vnd_cd, unit); 1791 if (sc == NULL) 1792 return -1; 1793 1794 if ((sc->sc_flags & VNF_INITED) == 0) 1795 return -1; 1796 1797 part = DISKPART(dev); 1798 omask = sc->sc_dkdev.dk_openmask & (1 << part); 1799 lp = sc->sc_dkdev.dk_label; 1800 1801 if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp)) /* XXX */ 1802 return -1; 1803 1804 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1805 size = -1; 1806 else 1807 size = lp->d_partitions[part].p_size * 1808 (lp->d_secsize / DEV_BSIZE); 1809 1810 if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp)) /* XXX */ 1811 return -1; 1812 1813 return size; 1814 } 1815 1816 static int 1817 vnddump(dev_t dev, daddr_t blkno, void *va, 1818 size_t size) 1819 { 1820 1821 /* Not implemented. */ 1822 return ENXIO; 1823 } 1824 1825 static void 1826 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp) 1827 { 1828 struct vndgeom *vng = &sc->sc_geom; 1829 struct partition *pp; 1830 unsigned spb; 1831 1832 memset(lp, 0, sizeof(*lp)); 1833 1834 spb = vng->vng_secsize / DEV_BSIZE; 1835 if (sc->sc_size / spb > UINT32_MAX) 1836 lp->d_secperunit = UINT32_MAX; 1837 else 1838 lp->d_secperunit = sc->sc_size / spb; 1839 lp->d_secsize = vng->vng_secsize; 1840 lp->d_nsectors = vng->vng_nsectors; 1841 lp->d_ntracks = vng->vng_ntracks; 1842 lp->d_ncylinders = vng->vng_ncylinders; 1843 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1844 1845 strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename)); 1846 lp->d_type = DKTYPE_VND; 1847 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1848 lp->d_rpm = 3600; 1849 lp->d_interleave = 1; 1850 lp->d_flags = 0; 1851 1852 pp = &lp->d_partitions[RAW_PART]; 1853 pp->p_offset = 0; 1854 pp->p_size = lp->d_secperunit; 1855 pp->p_fstype = FS_UNUSED; 1856 lp->d_npartitions = RAW_PART + 1; 1857 1858 lp->d_magic = DISKMAGIC; 1859 lp->d_magic2 = DISKMAGIC; 1860 lp->d_checksum = dkcksum(lp); 1861 } 1862 1863 /* 1864 * Read the disklabel from a vnd. If one is not present, create a fake one. 1865 */ 1866 static void 1867 vndgetdisklabel(dev_t dev, struct vnd_softc *sc) 1868 { 1869 const char *errstring; 1870 struct disklabel *lp = sc->sc_dkdev.dk_label; 1871 struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel; 1872 int i; 1873 1874 memset(clp, 0, sizeof(*clp)); 1875 1876 vndgetdefaultlabel(sc, lp); 1877 1878 /* 1879 * Call the generic disklabel extraction routine. 1880 */ 1881 errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp); 1882 if (errstring) { 1883 /* 1884 * Lack of disklabel is common, but we print the warning 1885 * anyway, since it might contain other useful information. 1886 */ 1887 aprint_normal_dev(sc->sc_dev, "%s\n", errstring); 1888 1889 /* 1890 * For historical reasons, if there's no disklabel 1891 * present, all partitions must be FS_BSDFFS and 1892 * occupy the entire disk. 1893 */ 1894 for (i = 0; i < MAXPARTITIONS; i++) { 1895 /* 1896 * Don't wipe out port specific hack (such as 1897 * dos partition hack of i386 port). 1898 */ 1899 if (lp->d_partitions[i].p_size != 0) 1900 continue; 1901 1902 lp->d_partitions[i].p_size = lp->d_secperunit; 1903 lp->d_partitions[i].p_offset = 0; 1904 lp->d_partitions[i].p_fstype = FS_BSDFFS; 1905 } 1906 1907 strncpy(lp->d_packname, "default label", 1908 sizeof(lp->d_packname)); 1909 1910 lp->d_npartitions = MAXPARTITIONS; 1911 lp->d_checksum = dkcksum(lp); 1912 } 1913 } 1914 1915 /* 1916 * Wait interruptibly for an exclusive lock. 1917 * 1918 * XXX 1919 * Several drivers do this; it should be abstracted and made MP-safe. 1920 */ 1921 static int 1922 vndlock(struct vnd_softc *sc) 1923 { 1924 int error; 1925 1926 while ((sc->sc_flags & VNF_LOCKED) != 0) { 1927 sc->sc_flags |= VNF_WANTED; 1928 if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0) 1929 return error; 1930 } 1931 sc->sc_flags |= VNF_LOCKED; 1932 return 0; 1933 } 1934 1935 /* 1936 * Unlock and wake up any waiters. 1937 */ 1938 static void 1939 vndunlock(struct vnd_softc *sc) 1940 { 1941 1942 sc->sc_flags &= ~VNF_LOCKED; 1943 if ((sc->sc_flags & VNF_WANTED) != 0) { 1944 sc->sc_flags &= ~VNF_WANTED; 1945 wakeup(sc); 1946 } 1947 } 1948 1949 #ifdef VND_COMPRESSION 1950 /* compressed file read */ 1951 static void 1952 compstrategy(struct buf *bp, off_t bn) 1953 { 1954 int error; 1955 int unit = vndunit(bp->b_dev); 1956 struct vnd_softc *vnd = 1957 device_lookup_private(&vnd_cd, unit); 1958 u_int32_t comp_block; 1959 struct uio auio; 1960 char *addr; 1961 int s; 1962 1963 /* set up constants for data move */ 1964 auio.uio_rw = UIO_READ; 1965 UIO_SETUP_SYSSPACE(&auio); 1966 1967 /* read, and transfer the data */ 1968 addr = bp->b_data; 1969 bp->b_resid = bp->b_bcount; 1970 s = splbio(); 1971 while (bp->b_resid > 0) { 1972 unsigned length; 1973 size_t length_in_buffer; 1974 u_int32_t offset_in_buffer; 1975 struct iovec aiov; 1976 1977 /* calculate the compressed block number */ 1978 comp_block = bn / (off_t)vnd->sc_comp_blksz; 1979 1980 /* check for good block number */ 1981 if (comp_block >= vnd->sc_comp_numoffs) { 1982 bp->b_error = EINVAL; 1983 splx(s); 1984 return; 1985 } 1986 1987 /* read in the compressed block, if not in buffer */ 1988 if (comp_block != vnd->sc_comp_buffblk) { 1989 length = vnd->sc_comp_offsets[comp_block + 1] - 1990 vnd->sc_comp_offsets[comp_block]; 1991 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1992 error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff, 1993 length, vnd->sc_comp_offsets[comp_block], 1994 UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred, 1995 NULL, NULL); 1996 if (error) { 1997 bp->b_error = error; 1998 VOP_UNLOCK(vnd->sc_vp); 1999 splx(s); 2000 return; 2001 } 2002 /* uncompress the buffer */ 2003 vnd->sc_comp_stream.next_in = vnd->sc_comp_buff; 2004 vnd->sc_comp_stream.avail_in = length; 2005 vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf; 2006 vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz; 2007 inflateReset(&vnd->sc_comp_stream); 2008 error = inflate(&vnd->sc_comp_stream, Z_FINISH); 2009 if (error != Z_STREAM_END) { 2010 if (vnd->sc_comp_stream.msg) 2011 aprint_normal_dev(vnd->sc_dev, 2012 "compressed file, %s\n", 2013 vnd->sc_comp_stream.msg); 2014 bp->b_error = EBADMSG; 2015 VOP_UNLOCK(vnd->sc_vp); 2016 splx(s); 2017 return; 2018 } 2019 vnd->sc_comp_buffblk = comp_block; 2020 VOP_UNLOCK(vnd->sc_vp); 2021 } 2022 2023 /* transfer the usable uncompressed data */ 2024 offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz; 2025 length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer; 2026 if (length_in_buffer > bp->b_resid) 2027 length_in_buffer = bp->b_resid; 2028 auio.uio_iov = &aiov; 2029 auio.uio_iovcnt = 1; 2030 aiov.iov_base = addr; 2031 aiov.iov_len = length_in_buffer; 2032 auio.uio_resid = aiov.iov_len; 2033 auio.uio_offset = 0; 2034 error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer, 2035 length_in_buffer, &auio); 2036 if (error) { 2037 bp->b_error = error; 2038 splx(s); 2039 return; 2040 } 2041 2042 bn += length_in_buffer; 2043 addr += length_in_buffer; 2044 bp->b_resid -= length_in_buffer; 2045 } 2046 splx(s); 2047 } 2048 2049 /* compression memory allocation routines */ 2050 static void * 2051 vnd_alloc(void *aux, u_int items, u_int siz) 2052 { 2053 return malloc(items * siz, M_TEMP, M_NOWAIT); 2054 } 2055 2056 static void 2057 vnd_free(void *aux, void *ptr) 2058 { 2059 free(ptr, M_TEMP); 2060 } 2061 #endif /* VND_COMPRESSION */ 2062 2063 static void 2064 vnd_set_geometry(struct vnd_softc *vnd) 2065 { 2066 struct disk_geom *dg = &vnd->sc_dkdev.dk_geom; 2067 2068 memset(dg, 0, sizeof(*dg)); 2069 2070 dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors * 2071 vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders; 2072 dg->dg_secsize = vnd->sc_geom.vng_secsize; 2073 dg->dg_nsectors = vnd->sc_geom.vng_nsectors; 2074 dg->dg_ntracks = vnd->sc_geom.vng_ntracks; 2075 dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders; 2076 2077 #ifdef DEBUG 2078 if (vnddebug & VDB_LABEL) { 2079 printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit); 2080 printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders); 2081 } 2082 #endif 2083 disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL); 2084 } 2085 2086 #ifdef VND_COMPRESSION 2087 #define VND_DEPENDS "zlib" 2088 #else 2089 #define VND_DEPENDS NULL 2090 #endif 2091 2092 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS); 2093 2094 #ifdef _MODULE 2095 int vnd_bmajor = -1, vnd_cmajor = -1; 2096 2097 CFDRIVER_DECL(vnd, DV_DISK, NULL); 2098 #endif 2099 2100 static int 2101 vnd_modcmd(modcmd_t cmd, void *arg) 2102 { 2103 int error = 0; 2104 2105 switch (cmd) { 2106 case MODULE_CMD_INIT: 2107 #ifdef _MODULE 2108 error = config_cfdriver_attach(&vnd_cd); 2109 if (error) 2110 break; 2111 2112 error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca); 2113 if (error) { 2114 config_cfdriver_detach(&vnd_cd); 2115 #ifdef DIAGNOSTIC 2116 aprint_error("%s: unable to register cfattach for \n" 2117 "%s, error %d", __func__, vnd_cd.cd_name, error); 2118 #endif 2119 break; 2120 } 2121 2122 /* 2123 * Attach the {b,c}devsw's 2124 */ 2125 error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor, 2126 &vnd_cdevsw, &vnd_cmajor); 2127 /* 2128 * If devsw_attach fails, remove from autoconf database 2129 */ 2130 if (error) { 2131 config_cfattach_detach(vnd_cd.cd_name, &vnd_ca); 2132 config_cfdriver_detach(&vnd_cd); 2133 #ifdef DIAGNOSTIC 2134 aprint_error("%s: unable to attach %s devsw, " 2135 "error %d", __func__, vnd_cd.cd_name, error); 2136 #endif 2137 break; 2138 } 2139 #endif 2140 break; 2141 2142 case MODULE_CMD_FINI: 2143 #ifdef _MODULE 2144 /* 2145 * Remove {b,c}devsw's 2146 */ 2147 devsw_detach(&vnd_bdevsw, &vnd_cdevsw); 2148 2149 /* 2150 * Now remove device from autoconf database 2151 */ 2152 error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca); 2153 if (error) { 2154 (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor, 2155 &vnd_cdevsw, &vnd_cmajor); 2156 #ifdef DIAGNOSTIC 2157 aprint_error("%s: failed to detach %s cfattach, " 2158 "error %d\n", __func__, vnd_cd.cd_name, error); 2159 #endif 2160 break; 2161 } 2162 error = config_cfdriver_detach(&vnd_cd); 2163 if (error) { 2164 (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca); 2165 (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor, 2166 &vnd_cdevsw, &vnd_cmajor); 2167 #ifdef DIAGNOSTIC 2168 aprint_error("%s: failed to detach %s cfdriver, " 2169 "error %d\n", __func__, vnd_cd.cd_name, error); 2170 break; 2171 #endif 2172 } 2173 #endif 2174 break; 2175 2176 case MODULE_CMD_STAT: 2177 return ENOTTY; 2178 2179 default: 2180 return ENOTTY; 2181 } 2182 2183 return error; 2184 } 2185
Indexes created Tue Sep 30 06:09:59 GMT 2025