1 /* $NetBSD: vnd.c,v 1.285 2022/03/31 19:30:15 pgoyette Exp $ */ 2 3 /*- 4 * Copyright (c) 1996, 1997, 1998, 2008, 2020 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 1.285 2022/03/31 19:30:15 pgoyette 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 #include <sys/compat_stub.h> 124 #include <sys/atomic.h> 125 126 #include <net/zlib.h> 127 128 #include <miscfs/genfs/genfs.h> 129 #include <miscfs/specfs/specdev.h> 130 131 #include <dev/dkvar.h> 132 #include <dev/vndvar.h> 133 134 #include "ioconf.h" 135 136 #if defined(VNDDEBUG) && !defined(DEBUG) 137 #define DEBUG 138 #endif 139 140 #ifdef DEBUG 141 int dovndcluster = 1; 142 #define VDB_FOLLOW 0x01 143 #define VDB_INIT 0x02 144 #define VDB_IO 0x04 145 #define VDB_LABEL 0x08 146 int vnddebug = 0; 147 #endif 148 149 #define vndunit(x) DISKUNIT(x) 150 151 struct vndxfer { 152 struct buf vx_buf; 153 struct vnd_softc *vx_vnd; 154 }; 155 #define VND_BUFTOXFER(bp) ((struct vndxfer *)(void *)bp) 156 157 #define VND_GETXFER(vnd) pool_get(&(vnd)->sc_vxpool, PR_WAITOK) 158 #define VND_PUTXFER(vnd, vx) pool_put(&(vnd)->sc_vxpool, (vx)) 159 160 #define VNDLABELDEV(dev) \ 161 (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART)) 162 163 #define VND_MAXPENDING(vnd) ((vnd)->sc_maxactive * 4) 164 #define VND_MAXPAGES(vnd) (1024 * 1024 / PAGE_SIZE) 165 166 167 static void vndclear(struct vnd_softc *, int); 168 static int vnddoclear(struct vnd_softc *, int, int, bool); 169 static int vndsetcred(struct vnd_softc *, kauth_cred_t); 170 static void vndthrottle(struct vnd_softc *, struct vnode *); 171 static void vndiodone(struct buf *); 172 #if 0 173 static void vndshutdown(void); 174 #endif 175 176 static void vndgetdefaultlabel(struct vnd_softc *, struct disklabel *); 177 static void vndgetdisklabel(dev_t, struct vnd_softc *); 178 179 static int vndlock(struct vnd_softc *); 180 static void vndunlock(struct vnd_softc *); 181 #ifdef VND_COMPRESSION 182 static void compstrategy(struct buf *, off_t); 183 static void *vnd_alloc(void *, u_int, u_int); 184 static void vnd_free(void *, void *); 185 #endif /* VND_COMPRESSION */ 186 187 static void vndthread(void *); 188 static bool vnode_has_op(const struct vnode *, int); 189 static void handle_with_rdwr(struct vnd_softc *, const struct buf *, 190 struct buf *); 191 static void handle_with_strategy(struct vnd_softc *, const struct buf *, 192 struct buf *); 193 static void vnd_set_geometry(struct vnd_softc *); 194 195 static dev_type_open(vndopen); 196 static dev_type_close(vndclose); 197 static dev_type_read(vndread); 198 static dev_type_write(vndwrite); 199 static dev_type_ioctl(vndioctl); 200 static dev_type_strategy(vndstrategy); 201 static dev_type_dump(vnddump); 202 static dev_type_size(vndsize); 203 204 const struct bdevsw vnd_bdevsw = { 205 .d_open = vndopen, 206 .d_close = vndclose, 207 .d_strategy = vndstrategy, 208 .d_ioctl = vndioctl, 209 .d_dump = vnddump, 210 .d_psize = vndsize, 211 .d_discard = nodiscard, 212 .d_flag = D_DISK 213 }; 214 215 const struct cdevsw vnd_cdevsw = { 216 .d_open = vndopen, 217 .d_close = vndclose, 218 .d_read = vndread, 219 .d_write = vndwrite, 220 .d_ioctl = vndioctl, 221 .d_stop = nostop, 222 .d_tty = notty, 223 .d_poll = nopoll, 224 .d_mmap = nommap, 225 .d_kqfilter = nokqfilter, 226 .d_discard = nodiscard, 227 .d_flag = D_DISK 228 }; 229 230 static int vnd_match(device_t, cfdata_t, void *); 231 static void vnd_attach(device_t, device_t, void *); 232 static int vnd_detach(device_t, int); 233 234 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc), 235 vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN); 236 237 static struct vnd_softc *vnd_spawn(int); 238 static int vnd_destroy(device_t); 239 240 static const struct dkdriver vnddkdriver = { 241 .d_strategy = vndstrategy, 242 .d_minphys = minphys 243 }; 244 245 void 246 vndattach(int num) 247 { 248 int error; 249 250 error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca); 251 if (error) 252 aprint_error("%s: unable to register cfattach, error = %d\n", 253 vnd_cd.cd_name, error); 254 } 255 256 static int 257 vnd_match(device_t self, cfdata_t cfdata, void *aux) 258 { 259 260 return 1; 261 } 262 263 static void 264 vnd_attach(device_t parent, device_t self, void *aux) 265 { 266 struct vnd_softc *sc = device_private(self); 267 268 sc->sc_dev = self; 269 sc->sc_comp_offsets = NULL; 270 sc->sc_comp_buff = NULL; 271 sc->sc_comp_decombuf = NULL; 272 bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK); 273 disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver); 274 if (!pmf_device_register(self, NULL, NULL)) 275 aprint_error_dev(self, "couldn't establish power handler\n"); 276 } 277 278 static int 279 vnd_detach(device_t self, int flags) 280 { 281 int error; 282 struct vnd_softc *sc = device_private(self); 283 284 if (sc->sc_flags & VNF_INITED) { 285 error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0); 286 if (error != 0) 287 return error; 288 } 289 290 pmf_device_deregister(self); 291 bufq_free(sc->sc_tab); 292 disk_destroy(&sc->sc_dkdev); 293 294 return 0; 295 } 296 297 static struct vnd_softc * 298 vnd_spawn(int unit) 299 { 300 cfdata_t cf; 301 302 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 303 cf->cf_name = vnd_cd.cd_name; 304 cf->cf_atname = vnd_cd.cd_name; 305 cf->cf_unit = unit; 306 cf->cf_fstate = FSTATE_STAR; 307 308 return device_private(config_attach_pseudo(cf)); 309 } 310 311 static int 312 vnd_destroy(device_t dev) 313 { 314 int error; 315 cfdata_t cf; 316 317 cf = device_cfdata(dev); 318 error = config_detach(dev, DETACH_QUIET); 319 if (error) 320 return error; 321 free(cf, M_DEVBUF); 322 return 0; 323 } 324 325 static int 326 vndopen(dev_t dev, int flags, int mode, struct lwp *l) 327 { 328 int unit = vndunit(dev); 329 struct vnd_softc *sc; 330 int error = 0, part, pmask; 331 struct disklabel *lp; 332 333 #ifdef DEBUG 334 if (vnddebug & VDB_FOLLOW) 335 printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l); 336 #endif 337 sc = device_lookup_private(&vnd_cd, unit); 338 if (sc == NULL) { 339 sc = vnd_spawn(unit); 340 if (sc == NULL) 341 return ENOMEM; 342 343 /* compatibility, keep disklabel after close */ 344 sc->sc_flags = VNF_KLABEL; 345 } 346 347 if ((error = vndlock(sc)) != 0) 348 return error; 349 350 mutex_enter(&sc->sc_dkdev.dk_openlock); 351 352 if ((sc->sc_flags & VNF_CLEARING) != 0) { 353 error = ENXIO; 354 goto done; 355 } 356 357 lp = sc->sc_dkdev.dk_label; 358 359 part = DISKPART(dev); 360 pmask = (1 << part); 361 362 if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) { 363 error = EBUSY; 364 goto done; 365 } 366 367 if ((flags & FWRITE) && (sc->sc_flags & VNF_READONLY)) { 368 error = EROFS; 369 goto done; 370 } 371 372 if (sc->sc_flags & VNF_INITED) { 373 if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) { 374 /* 375 * If any non-raw partition is open, but the disk 376 * has been invalidated, disallow further opens. 377 */ 378 if ((sc->sc_flags & VNF_VLABEL) == 0) { 379 error = EIO; 380 goto done; 381 } 382 } else { 383 /* 384 * Load the partition info if not already loaded. 385 */ 386 if ((sc->sc_flags & VNF_VLABEL) == 0) { 387 sc->sc_flags |= VNF_VLABEL; 388 vndgetdisklabel(dev, sc); 389 } 390 } 391 } 392 393 /* Check that the partitions exists. */ 394 if (part != RAW_PART) { 395 if (((sc->sc_flags & VNF_INITED) == 0) || 396 ((part >= lp->d_npartitions) || 397 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 398 error = ENXIO; 399 goto done; 400 } 401 } 402 403 /* Prevent our unit from being unconfigured while open. */ 404 switch (mode) { 405 case S_IFCHR: 406 sc->sc_dkdev.dk_copenmask |= pmask; 407 break; 408 409 case S_IFBLK: 410 sc->sc_dkdev.dk_bopenmask |= pmask; 411 break; 412 } 413 sc->sc_dkdev.dk_openmask = 414 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask; 415 416 done: 417 mutex_exit(&sc->sc_dkdev.dk_openlock); 418 vndunlock(sc); 419 return error; 420 } 421 422 static int 423 vndclose(dev_t dev, int flags, int mode, struct lwp *l) 424 { 425 int unit = vndunit(dev); 426 struct vnd_softc *sc; 427 int error = 0, part; 428 429 #ifdef DEBUG 430 if (vnddebug & VDB_FOLLOW) 431 printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l); 432 #endif 433 sc = device_lookup_private(&vnd_cd, unit); 434 if (sc == NULL) 435 return ENXIO; 436 437 if ((error = vndlock(sc)) != 0) 438 return error; 439 440 mutex_enter(&sc->sc_dkdev.dk_openlock); 441 442 part = DISKPART(dev); 443 444 /* ...that much closer to allowing unconfiguration... */ 445 switch (mode) { 446 case S_IFCHR: 447 sc->sc_dkdev.dk_copenmask &= ~(1 << part); 448 break; 449 450 case S_IFBLK: 451 sc->sc_dkdev.dk_bopenmask &= ~(1 << part); 452 break; 453 } 454 sc->sc_dkdev.dk_openmask = 455 sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask; 456 457 /* are we last opener ? */ 458 if (sc->sc_dkdev.dk_openmask == 0) { 459 if ((sc->sc_flags & VNF_KLABEL) == 0) 460 sc->sc_flags &= ~VNF_VLABEL; 461 } 462 463 mutex_exit(&sc->sc_dkdev.dk_openlock); 464 465 vndunlock(sc); 466 467 if ((sc->sc_flags & VNF_INITED) == 0) { 468 if ((error = vnd_destroy(sc->sc_dev)) != 0) { 469 aprint_error_dev(sc->sc_dev, 470 "unable to detach instance\n"); 471 return error; 472 } 473 } 474 475 return 0; 476 } 477 478 /* 479 * Queue the request, and wakeup the kernel thread to handle it. 480 */ 481 static void 482 vndstrategy(struct buf *bp) 483 { 484 int unit = vndunit(bp->b_dev); 485 struct vnd_softc *vnd = 486 device_lookup_private(&vnd_cd, unit); 487 struct disklabel *lp; 488 daddr_t blkno; 489 int s = splbio(); 490 491 if (vnd == NULL) { 492 bp->b_error = ENXIO; 493 goto done; 494 } 495 lp = vnd->sc_dkdev.dk_label; 496 497 if ((vnd->sc_flags & VNF_INITED) == 0) { 498 bp->b_error = ENXIO; 499 goto done; 500 } 501 502 /* 503 * The transfer must be a whole number of blocks. 504 */ 505 if ((bp->b_bcount % lp->d_secsize) != 0) { 506 bp->b_error = EINVAL; 507 goto done; 508 } 509 510 /* 511 * check if we're read-only. 512 */ 513 if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) { 514 bp->b_error = EACCES; 515 goto done; 516 } 517 518 /* If it's a nil transfer, wake up the top half now. */ 519 if (bp->b_bcount == 0) { 520 goto done; 521 } 522 523 /* 524 * Do bounds checking and adjust transfer. If there's an error, 525 * the bounds check will flag that for us. 526 */ 527 if (DISKPART(bp->b_dev) == RAW_PART) { 528 if (bounds_check_with_mediasize(bp, DEV_BSIZE, 529 vnd->sc_size) <= 0) 530 goto done; 531 } else { 532 if (bounds_check_with_label(&vnd->sc_dkdev, 533 bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0) 534 goto done; 535 } 536 537 /* 538 * Put the block number in terms of the logical blocksize 539 * of the "device". 540 */ 541 542 blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE); 543 544 /* 545 * Translate the partition-relative block number to an absolute. 546 */ 547 if (DISKPART(bp->b_dev) != RAW_PART) { 548 struct partition *pp; 549 550 pp = &vnd->sc_dkdev.dk_label->d_partitions[ 551 DISKPART(bp->b_dev)]; 552 blkno += pp->p_offset; 553 } 554 bp->b_rawblkno = blkno; 555 556 #ifdef DEBUG 557 if (vnddebug & VDB_FOLLOW) 558 printf("vndstrategy(%p): unit %d\n", bp, unit); 559 #endif 560 if ((vnd->sc_flags & VNF_USE_VN_RDWR)) { 561 KASSERT(vnd->sc_pending >= 0 && 562 vnd->sc_pending <= VND_MAXPENDING(vnd)); 563 while (vnd->sc_pending == VND_MAXPENDING(vnd)) 564 tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0); 565 vnd->sc_pending++; 566 } 567 bufq_put(vnd->sc_tab, bp); 568 wakeup(&vnd->sc_tab); 569 splx(s); 570 return; 571 572 done: 573 bp->b_resid = bp->b_bcount; 574 biodone(bp); 575 splx(s); 576 } 577 578 static bool 579 vnode_has_strategy(struct vnd_softc *vnd) 580 { 581 return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) && 582 vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy)); 583 } 584 585 /* Verify that I/O requests cannot be smaller than the 586 * smallest I/O size supported by the backend. 587 */ 588 static bool 589 vnode_has_large_blocks(struct vnd_softc *vnd) 590 { 591 u_int32_t vnd_secsize, iosize; 592 593 iosize = vnd->sc_iosize; 594 vnd_secsize = vnd->sc_geom.vng_secsize; 595 596 return vnd_secsize % iosize != 0; 597 } 598 599 /* XXX this function needs a reliable check to detect 600 * sparse files. Otherwise, bmap/strategy may be used 601 * and fail on non-allocated blocks. VOP_READ/VOP_WRITE 602 * works on sparse files. 603 */ 604 #if notyet 605 static bool 606 vnode_strategy_probe(struct vnd_softc *vnd) 607 { 608 int error; 609 daddr_t nbn; 610 611 if (!vnode_has_strategy(vnd)) 612 return false; 613 614 if (vnode_has_large_blocks(vnd)) 615 return false; 616 617 /* Convert the first logical block number to its 618 * physical block number. 619 */ 620 error = 0; 621 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 622 error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL); 623 VOP_UNLOCK(vnd->sc_vp); 624 625 /* Test if that worked. */ 626 if (error == 0 && (long)nbn == -1) 627 return false; 628 629 return true; 630 } 631 #endif 632 633 static void 634 vndthread(void *arg) 635 { 636 struct vnd_softc *vnd = arg; 637 int s; 638 639 /* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to 640 * directly access the backing vnode. If we can, use these two 641 * operations to avoid messing with the local buffer cache. 642 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations 643 * which are guaranteed to work with any file system. */ 644 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 && 645 ! vnode_has_strategy(vnd)) 646 vnd->sc_flags |= VNF_USE_VN_RDWR; 647 648 /* VOP_STRATEGY can only be used if the backing vnode allows 649 * to access blocks as small as defined by the vnd geometry. 650 */ 651 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 && 652 vnode_has_large_blocks(vnd)) 653 vnd->sc_flags |= VNF_USE_VN_RDWR; 654 655 #ifdef DEBUG 656 if (vnddebug & VDB_INIT) 657 printf("vndthread: vp %p, %s\n", vnd->sc_vp, 658 (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ? 659 "using bmap/strategy operations" : 660 "using read/write operations"); 661 #endif 662 663 s = splbio(); 664 vnd->sc_flags |= VNF_KTHREAD; 665 wakeup(&vnd->sc_kthread); 666 667 /* 668 * Dequeue requests and serve them depending on the available 669 * vnode operations. 670 */ 671 while ((vnd->sc_flags & VNF_VUNCONF) == 0) { 672 struct vndxfer *vnx; 673 struct buf *obp; 674 struct buf *bp; 675 676 obp = bufq_get(vnd->sc_tab); 677 if (obp == NULL) { 678 tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0); 679 continue; 680 }; 681 if ((vnd->sc_flags & VNF_USE_VN_RDWR)) { 682 KASSERT(vnd->sc_pending > 0 && 683 vnd->sc_pending <= VND_MAXPENDING(vnd)); 684 if (vnd->sc_pending-- == VND_MAXPENDING(vnd)) 685 wakeup(&vnd->sc_pending); 686 } 687 splx(s); 688 #ifdef DEBUG 689 if (vnddebug & VDB_FOLLOW) 690 printf("vndthread(%p)\n", obp); 691 #endif 692 693 if (vnd->sc_vp->v_mount == NULL) { 694 obp->b_error = ENXIO; 695 goto done; 696 } 697 #ifdef VND_COMPRESSION 698 /* handle a compressed read */ 699 if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) { 700 off_t bn; 701 702 /* Convert to a byte offset within the file. */ 703 bn = obp->b_rawblkno * 704 vnd->sc_dkdev.dk_label->d_secsize; 705 706 compstrategy(obp, bn); 707 goto done; 708 } 709 #endif /* VND_COMPRESSION */ 710 711 /* 712 * Allocate a header for this transfer and link it to the 713 * buffer 714 */ 715 s = splbio(); 716 vnx = VND_GETXFER(vnd); 717 splx(s); 718 vnx->vx_vnd = vnd; 719 720 s = splbio(); 721 while (vnd->sc_active >= vnd->sc_maxactive) { 722 tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0); 723 } 724 vnd->sc_active++; 725 splx(s); 726 727 /* Instrumentation. */ 728 disk_busy(&vnd->sc_dkdev); 729 730 bp = &vnx->vx_buf; 731 buf_init(bp); 732 bp->b_flags = (obp->b_flags & (B_READ | B_PHYS | B_RAW)); 733 bp->b_oflags = obp->b_oflags; 734 bp->b_cflags = obp->b_cflags; 735 bp->b_iodone = vndiodone; 736 bp->b_private = obp; 737 bp->b_vp = vnd->sc_vp; 738 bp->b_objlock = bp->b_vp->v_interlock; 739 bp->b_data = obp->b_data; 740 bp->b_bcount = obp->b_bcount; 741 BIO_COPYPRIO(bp, obp); 742 743 /* Make sure the request succeeds while suspending this fs. */ 744 fstrans_start_lazy(vnd->sc_vp->v_mount); 745 746 /* Handle the request using the appropriate operations. */ 747 if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0) 748 handle_with_strategy(vnd, obp, bp); 749 else 750 handle_with_rdwr(vnd, obp, bp); 751 752 fstrans_done(vnd->sc_vp->v_mount); 753 754 s = splbio(); 755 continue; 756 757 done: 758 biodone(obp); 759 s = splbio(); 760 } 761 762 vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF); 763 wakeup(&vnd->sc_kthread); 764 splx(s); 765 kthread_exit(0); 766 } 767 768 /* 769 * Checks if the given vnode supports the requested operation. 770 * The operation is specified the offset returned by VOFFSET. 771 * 772 * XXX The test below used to determine this is quite fragile 773 * because it relies on the file system to use genfs to specify 774 * unimplemented operations. There might be another way to do 775 * it more cleanly. 776 */ 777 static bool 778 vnode_has_op(const struct vnode *vp, int opoffset) 779 { 780 int (*defaultp)(void *); 781 int (*opp)(void *); 782 783 defaultp = vp->v_op[VOFFSET(vop_default)]; 784 opp = vp->v_op[opoffset]; 785 786 return opp != defaultp && opp != genfs_eopnotsupp && 787 opp != genfs_badop && opp != genfs_nullop; 788 } 789 790 /* 791 * Handles the read/write request given in 'bp' using the vnode's VOP_READ 792 * and VOP_WRITE operations. 793 * 794 * 'obp' is a pointer to the original request fed to the vnd device. 795 */ 796 static void 797 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp) 798 { 799 bool doread; 800 off_t offset; 801 size_t len, resid; 802 struct vnode *vp; 803 int npages; 804 805 doread = bp->b_flags & B_READ; 806 offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize; 807 len = bp->b_bcount; 808 vp = vnd->sc_vp; 809 810 #if defined(DEBUG) 811 if (vnddebug & VDB_IO) 812 printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64 813 ", secsize %d, offset %" PRIu64 814 ", bcount %d\n", 815 vp, doread ? "read" : "write", obp->b_rawblkno, 816 vnd->sc_dkdev.dk_label->d_secsize, offset, 817 bp->b_bcount); 818 #endif 819 820 /* Issue the read or write operation. */ 821 bp->b_error = 822 vn_rdwr(doread ? UIO_READ : UIO_WRITE, 823 vp, bp->b_data, len, offset, UIO_SYSSPACE, 824 IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_DIRECT, 825 vnd->sc_cred, &resid, NULL); 826 bp->b_resid = resid; 827 828 /* 829 * Avoid caching too many pages, the vnd user 830 * is usually a filesystem and caches itself. 831 * We need some amount of caching to not hinder 832 * read-ahead and write-behind operations. 833 */ 834 npages = atomic_load_relaxed(&vp->v_uobj.uo_npages); 835 if (npages > VND_MAXPAGES(vnd)) { 836 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 837 (void) VOP_PUTPAGES(vp, 0, 0, 838 PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE); 839 } 840 841 /* We need to increase the number of outputs on the vnode if 842 * there was any write to it. */ 843 if (!doread) { 844 mutex_enter(vp->v_interlock); 845 vp->v_numoutput++; 846 mutex_exit(vp->v_interlock); 847 } 848 849 biodone(bp); 850 } 851 852 /* 853 * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP 854 * and VOP_STRATEGY operations. 855 * 856 * 'obp' is a pointer to the original request fed to the vnd device. 857 */ 858 static void 859 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp, 860 struct buf *bp) 861 { 862 int bsize, error, flags, skipped; 863 size_t resid, sz; 864 off_t bn, offset; 865 struct vnode *vp; 866 struct buf *nbp = NULL; 867 868 flags = obp->b_flags; 869 870 871 /* convert to a byte offset within the file. */ 872 bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize; 873 874 bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize; 875 skipped = 0; 876 877 /* 878 * Break the request into bsize pieces and feed them 879 * sequentially using VOP_BMAP/VOP_STRATEGY. 880 * We do it this way to keep from flooding NFS servers if we 881 * are connected to an NFS file. This places the burden on 882 * the client rather than the server. 883 */ 884 error = 0; 885 bp->b_resid = bp->b_bcount; 886 for (offset = 0, resid = bp->b_resid; /* true */; 887 resid -= sz, offset += sz) { 888 daddr_t nbn; 889 int off, nra; 890 891 nra = 0; 892 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 893 error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra); 894 VOP_UNLOCK(vnd->sc_vp); 895 896 if (error == 0 && (long)nbn == -1) 897 error = EIO; 898 899 /* 900 * If there was an error or a hole in the file...punt. 901 * Note that we may have to wait for any operations 902 * that we have already fired off before releasing 903 * the buffer. 904 * 905 * XXX we could deal with holes here but it would be 906 * a hassle (in the write case). 907 */ 908 if (error) { 909 skipped += resid; 910 break; 911 } 912 913 #ifdef DEBUG 914 if (!dovndcluster) 915 nra = 0; 916 #endif 917 918 off = bn % bsize; 919 sz = MIN(((off_t)1 + nra) * bsize - off, resid); 920 #ifdef DEBUG 921 if (vnddebug & VDB_IO) 922 printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64 923 " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn, 924 nbn, sz); 925 #endif 926 927 nbp = getiobuf(vp, true); 928 nestiobuf_setup(bp, nbp, offset, sz); 929 nbp->b_blkno = nbn + btodb(off); 930 931 #if 0 /* XXX #ifdef DEBUG */ 932 if (vnddebug & VDB_IO) 933 printf("vndstart(%ld): bp %p vp %p blkno " 934 "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n", 935 (long) (vnd-vnd_softc), &nbp->vb_buf, 936 nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno, 937 nbp->vb_buf.b_flags, nbp->vb_buf.b_data, 938 nbp->vb_buf.b_bcount); 939 #endif 940 if (resid == sz) { 941 break; 942 } 943 VOP_STRATEGY(vp, nbp); 944 bn += sz; 945 } 946 if (!(flags & B_READ)) { 947 struct vnode *w_vp; 948 /* 949 * this is the last nested buf, account for 950 * the parent buf write too. 951 * This has to be done last, so that 952 * fsync won't wait for this write which 953 * has no chance to complete before all nested bufs 954 * have been queued. But it has to be done 955 * before the last VOP_STRATEGY() 956 * or the call to nestiobuf_done(). 957 */ 958 w_vp = bp->b_vp; 959 mutex_enter(w_vp->v_interlock); 960 w_vp->v_numoutput++; 961 mutex_exit(w_vp->v_interlock); 962 } 963 KASSERT(skipped != 0 || nbp != NULL); 964 if (skipped) 965 nestiobuf_done(bp, skipped, error); 966 else 967 VOP_STRATEGY(vp, nbp); 968 } 969 970 static void 971 vndiodone(struct buf *bp) 972 { 973 struct vndxfer *vnx = VND_BUFTOXFER(bp); 974 struct vnd_softc *vnd = vnx->vx_vnd; 975 struct buf *obp = bp->b_private; 976 int s = splbio(); 977 978 KERNEL_LOCK(1, NULL); /* XXXSMP */ 979 KASSERT(&vnx->vx_buf == bp); 980 KASSERT(vnd->sc_active > 0); 981 #ifdef DEBUG 982 if (vnddebug & VDB_IO) { 983 printf("vndiodone1: bp %p iodone: error %d\n", 984 bp, bp->b_error); 985 } 986 #endif 987 disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid, 988 (bp->b_flags & B_READ)); 989 vnd->sc_active--; 990 if (vnd->sc_active == 0) { 991 wakeup(&vnd->sc_tab); 992 } 993 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 994 splx(s); 995 obp->b_error = bp->b_error; 996 obp->b_resid = bp->b_resid; 997 buf_destroy(bp); 998 VND_PUTXFER(vnd, vnx); 999 biodone(obp); 1000 } 1001 1002 /* ARGSUSED */ 1003 static int 1004 vndread(dev_t dev, struct uio *uio, int flags) 1005 { 1006 int unit = vndunit(dev); 1007 struct vnd_softc *sc; 1008 1009 #ifdef DEBUG 1010 if (vnddebug & VDB_FOLLOW) 1011 printf("vndread(0x%"PRIx64", %p)\n", dev, uio); 1012 #endif 1013 1014 sc = device_lookup_private(&vnd_cd, unit); 1015 if (sc == NULL) 1016 return ENXIO; 1017 1018 if ((sc->sc_flags & VNF_INITED) == 0) 1019 return ENXIO; 1020 1021 return physio(vndstrategy, NULL, dev, B_READ, minphys, uio); 1022 } 1023 1024 /* ARGSUSED */ 1025 static int 1026 vndwrite(dev_t dev, struct uio *uio, int flags) 1027 { 1028 int unit = vndunit(dev); 1029 struct vnd_softc *sc; 1030 1031 #ifdef DEBUG 1032 if (vnddebug & VDB_FOLLOW) 1033 printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio); 1034 #endif 1035 1036 sc = device_lookup_private(&vnd_cd, unit); 1037 if (sc == NULL) 1038 return ENXIO; 1039 1040 if ((sc->sc_flags & VNF_INITED) == 0) 1041 return ENXIO; 1042 1043 return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio); 1044 } 1045 1046 static int 1047 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va) 1048 { 1049 int error; 1050 struct vnd_softc *vnd; 1051 1052 if (*un == -1) 1053 *un = unit; 1054 if (*un < 0) 1055 return EINVAL; 1056 1057 vnd = device_lookup_private(&vnd_cd, *un); 1058 if (vnd == NULL) 1059 return -1; 1060 1061 if ((vnd->sc_flags & VNF_INITED) == 0) 1062 return -1; 1063 1064 vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY); 1065 error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred); 1066 VOP_UNLOCK(vnd->sc_vp); 1067 return error; 1068 } 1069 1070 static int 1071 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force) 1072 { 1073 int error; 1074 1075 if ((error = vndlock(vnd)) != 0) 1076 return error; 1077 1078 /* 1079 * Don't unconfigure if any other partitions are open 1080 * or if both the character and block flavors of this 1081 * partition are open. 1082 */ 1083 if (DK_BUSY(vnd, pmask) && !force) { 1084 vndunlock(vnd); 1085 return EBUSY; 1086 } 1087 1088 /* Delete all of our wedges */ 1089 dkwedge_delall(&vnd->sc_dkdev); 1090 1091 /* 1092 * XXX vndclear() might call vndclose() implicitly; 1093 * release lock to avoid recursion 1094 * 1095 * Set VNF_CLEARING to prevent vndopen() from 1096 * sneaking in after we vndunlock(). 1097 */ 1098 vnd->sc_flags |= VNF_CLEARING; 1099 vndunlock(vnd); 1100 vndclear(vnd, minor); 1101 #ifdef DEBUG 1102 if (vnddebug & VDB_INIT) 1103 printf("%s: CLRed\n", __func__); 1104 #endif 1105 1106 /* Destroy the xfer and buffer pools. */ 1107 pool_destroy(&vnd->sc_vxpool); 1108 1109 /* Detach the disk. */ 1110 disk_detach(&vnd->sc_dkdev); 1111 1112 return 0; 1113 } 1114 1115 static int 1116 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va) 1117 { 1118 int error; 1119 1120 KASSERT(l); 1121 1122 /* the first member is always int vnd_unit in all the versions */ 1123 if (*(int *)data >= vnd_cd.cd_ndevs) 1124 return ENXIO; 1125 1126 switch (error = vnd_cget(l, unit, (int *)data, va)) { 1127 case -1: 1128 /* unused is not an error */ 1129 memset(va, 0, sizeof(*va)); 1130 /*FALLTHROUGH*/ 1131 case 0: 1132 return 0; 1133 default: 1134 return error; 1135 } 1136 } 1137 1138 /* ARGSUSED */ 1139 static int 1140 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 1141 { 1142 bool force; 1143 int unit = vndunit(dev); 1144 struct vnd_softc *vnd; 1145 struct vnd_ioctl *vio; 1146 struct vattr vattr; 1147 struct pathbuf *pb; 1148 struct vnode *vp; 1149 int error, part, pmask; 1150 uint64_t geomsize; 1151 int fflags; 1152 #ifdef __HAVE_OLD_DISKLABEL 1153 struct disklabel newlabel; 1154 #endif 1155 1156 #ifdef DEBUG 1157 if (vnddebug & VDB_FOLLOW) 1158 printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n", 1159 dev, cmd, data, flag, l->l_proc, unit); 1160 #endif 1161 /* Do the get's first; they don't need initialization or verification */ 1162 switch (cmd) { 1163 case VNDIOCGET: 1164 if ((error = vndioctl_get(l, data, unit, &vattr)) != 0) 1165 return error; 1166 1167 struct vnd_user *vnu = data; 1168 vnu->vnu_dev = vattr.va_fsid; 1169 vnu->vnu_ino = vattr.va_fileid; 1170 return 0; 1171 1172 default: 1173 /* First check for COMPAT_50 hook */ 1174 MODULE_HOOK_CALL(compat_vndioctl_50_hook, 1175 (cmd, l, data, unit, &vattr, vndioctl_get), 1176 enosys(), error); 1177 1178 /* 1179 * If not present, then COMPAT_30 hook also not 1180 * present, so just continue with checks for the 1181 * "write" commands 1182 */ 1183 if (error == ENOSYS) { 1184 error = 0; 1185 break; 1186 } 1187 1188 /* If not already handled, try the COMPAT_30 hook */ 1189 if (error == EPASSTHROUGH) 1190 MODULE_HOOK_CALL(compat_vndioctl_30_hook, 1191 (cmd, l, data, unit, &vattr, vndioctl_get), 1192 enosys(), error); 1193 1194 /* If no COMPAT_30 module, or not handled, check writes */ 1195 if (error == ENOSYS || error == EPASSTHROUGH) { 1196 error = 0; 1197 break; 1198 } 1199 return error; 1200 } 1201 1202 vnd = device_lookup_private(&vnd_cd, unit); 1203 if (vnd == NULL) 1204 return ENXIO; 1205 vio = (struct vnd_ioctl *)data; 1206 1207 /* Must be open for writes for these commands... */ 1208 switch (cmd) { 1209 case VNDIOCSET50: 1210 case VNDIOCCLR50: 1211 if (!compat_vndioctl_50_hook.hooked) 1212 return EINVAL; 1213 /* FALLTHROUGH */ 1214 case VNDIOCSET: 1215 case VNDIOCCLR: 1216 case DIOCSDINFO: 1217 case DIOCWDINFO: 1218 #ifdef __HAVE_OLD_DISKLABEL 1219 case ODIOCSDINFO: 1220 case ODIOCWDINFO: 1221 #endif 1222 case DIOCKLABEL: 1223 case DIOCWLABEL: 1224 case DIOCCACHESYNC: 1225 if ((flag & FWRITE) == 0) 1226 return EBADF; 1227 } 1228 1229 switch (cmd) { 1230 case VNDIOCSET50: 1231 case VNDIOCSET: 1232 /* Must not be initialized */ 1233 if (vnd->sc_flags & VNF_INITED) 1234 return EBUSY; 1235 break; 1236 default: 1237 /* Must be initialized */ 1238 if ((vnd->sc_flags & VNF_INITED) == 0) 1239 return ENXIO; 1240 break; 1241 } 1242 1243 error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l); 1244 if (error != EPASSTHROUGH) 1245 return error; 1246 1247 switch (cmd) { 1248 case VNDIOCSET50: 1249 case VNDIOCSET: 1250 if ((error = vndlock(vnd)) != 0) 1251 return error; 1252 1253 fflags = FREAD; 1254 if ((vio->vnd_flags & VNDIOF_READONLY) == 0) 1255 fflags |= FWRITE; 1256 if ((vio->vnd_flags & VNDIOF_FILEIO) != 0) 1257 vnd->sc_flags |= VNF_USE_VN_RDWR; 1258 error = pathbuf_copyin(vio->vnd_file, &pb); 1259 if (error) { 1260 goto unlock_and_exit; 1261 } 1262 error = vn_open(NULL, pb, 0, fflags, 0, &vp, NULL, NULL); 1263 if (error != 0) { 1264 pathbuf_destroy(pb); 1265 goto unlock_and_exit; 1266 } 1267 KASSERT(l); 1268 error = VOP_GETATTR(vp, &vattr, l->l_cred); 1269 if (!error && vp->v_type != VREG) 1270 error = EOPNOTSUPP; 1271 if (!error && vattr.va_bytes < vattr.va_size) 1272 /* File is definitely sparse, use vn_rdwr() */ 1273 vnd->sc_flags |= VNF_USE_VN_RDWR; 1274 if (error) { 1275 VOP_UNLOCK(vp); 1276 goto close_and_exit; 1277 } 1278 1279 /* If using a compressed file, initialize its info */ 1280 /* (or abort with an error if kernel has no compression) */ 1281 if (vio->vnd_flags & VNDIOF_COMP) { 1282 #ifdef VND_COMPRESSION 1283 struct vnd_comp_header *ch; 1284 int i; 1285 uint32_t comp_size; 1286 uint32_t comp_maxsize; 1287 1288 /* allocate space for compressed file header */ 1289 ch = malloc(sizeof(struct vnd_comp_header), 1290 M_TEMP, M_WAITOK); 1291 1292 /* read compressed file header */ 1293 error = vn_rdwr(UIO_READ, vp, (void *)ch, 1294 sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE, 1295 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL); 1296 if (error) { 1297 free(ch, M_TEMP); 1298 VOP_UNLOCK(vp); 1299 goto close_and_exit; 1300 } 1301 1302 if (be32toh(ch->block_size) == 0 || 1303 be32toh(ch->num_blocks) > UINT32_MAX - 1) { 1304 free(ch, M_TEMP); 1305 VOP_UNLOCK(vp); 1306 goto close_and_exit; 1307 } 1308 1309 /* save some header info */ 1310 vnd->sc_comp_blksz = be32toh(ch->block_size); 1311 /* note last offset is the file byte size */ 1312 vnd->sc_comp_numoffs = be32toh(ch->num_blocks) + 1; 1313 free(ch, M_TEMP); 1314 if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) { 1315 VOP_UNLOCK(vp); 1316 error = EINVAL; 1317 goto close_and_exit; 1318 } 1319 KASSERT(0 < vnd->sc_comp_blksz); 1320 KASSERT(0 < vnd->sc_comp_numoffs); 1321 /* 1322 * @#^@!$& gcc -Wtype-limits refuses to let me 1323 * write SIZE_MAX/sizeof(uint64_t) < numoffs, 1324 * because the range of the type on amd64 makes 1325 * the comparisons always false. 1326 */ 1327 #if SIZE_MAX <= UINT32_MAX*(64/CHAR_BIT) 1328 if (SIZE_MAX/sizeof(uint64_t) < vnd->sc_comp_numoffs) { 1329 VOP_UNLOCK(vp); 1330 error = EINVAL; 1331 goto close_and_exit; 1332 } 1333 #endif 1334 if ((vattr.va_size < sizeof(struct vnd_comp_header)) || 1335 (vattr.va_size - sizeof(struct vnd_comp_header) < 1336 sizeof(uint64_t)*vnd->sc_comp_numoffs) || 1337 (UQUAD_MAX/vnd->sc_comp_blksz < 1338 vnd->sc_comp_numoffs - 1)) { 1339 VOP_UNLOCK(vp); 1340 error = EINVAL; 1341 goto close_and_exit; 1342 } 1343 1344 /* set decompressed file size */ 1345 KASSERT(vnd->sc_comp_numoffs - 1 <= 1346 UQUAD_MAX/vnd->sc_comp_blksz); 1347 vattr.va_size = 1348 ((u_quad_t)vnd->sc_comp_numoffs - 1) * 1349 (u_quad_t)vnd->sc_comp_blksz; 1350 1351 /* allocate space for all the compressed offsets */ 1352 __CTASSERT(UINT32_MAX <= UQUAD_MAX/sizeof(uint64_t)); 1353 vnd->sc_comp_offsets = 1354 malloc(sizeof(uint64_t) * vnd->sc_comp_numoffs, 1355 M_DEVBUF, M_WAITOK); 1356 1357 /* read in the offsets */ 1358 error = vn_rdwr(UIO_READ, vp, 1359 (void *)vnd->sc_comp_offsets, 1360 sizeof(uint64_t) * vnd->sc_comp_numoffs, 1361 sizeof(struct vnd_comp_header), UIO_SYSSPACE, 1362 IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL); 1363 if (error) { 1364 VOP_UNLOCK(vp); 1365 goto close_and_exit; 1366 } 1367 /* 1368 * find largest block size (used for allocation limit). 1369 * Also convert offset to native byte order. 1370 */ 1371 comp_maxsize = 0; 1372 for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) { 1373 vnd->sc_comp_offsets[i] = 1374 be64toh(vnd->sc_comp_offsets[i]); 1375 comp_size = 1376 be64toh(vnd->sc_comp_offsets[i + 1]) 1377 - vnd->sc_comp_offsets[i]; 1378 if (comp_size > comp_maxsize) 1379 comp_maxsize = comp_size; 1380 } 1381 vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] = 1382 be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs 1383 - 1]); 1384 1385 /* create compressed data buffer */ 1386 vnd->sc_comp_buff = malloc(comp_maxsize, 1387 M_DEVBUF, M_WAITOK); 1388 1389 /* create decompressed buffer */ 1390 vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz, 1391 M_DEVBUF, M_WAITOK); 1392 vnd->sc_comp_buffblk = -1; 1393 1394 /* Initialize decompress stream */ 1395 memset(&vnd->sc_comp_stream, 0, sizeof(z_stream)); 1396 vnd->sc_comp_stream.zalloc = vnd_alloc; 1397 vnd->sc_comp_stream.zfree = vnd_free; 1398 error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS); 1399 if (error) { 1400 if (vnd->sc_comp_stream.msg) 1401 printf("vnd%d: compressed file, %s\n", 1402 unit, vnd->sc_comp_stream.msg); 1403 VOP_UNLOCK(vp); 1404 error = EINVAL; 1405 goto close_and_exit; 1406 } 1407 1408 vnd->sc_flags |= VNF_COMP | VNF_READONLY; 1409 #else /* !VND_COMPRESSION */ 1410 VOP_UNLOCK(vp); 1411 error = EOPNOTSUPP; 1412 goto close_and_exit; 1413 #endif /* VND_COMPRESSION */ 1414 } 1415 1416 VOP_UNLOCK(vp); 1417 vnd->sc_vp = vp; 1418 vnd->sc_size = btodb(vattr.va_size); /* note truncation */ 1419 1420 /* get smallest I/O size for underlying device, fall back to 1421 * fundamental I/O size of underlying filesystem 1422 */ 1423 error = bdev_ioctl(vattr.va_fsid, DIOCGSECTORSIZE, &vnd->sc_iosize, FKIOCTL, l); 1424 if (error) 1425 vnd->sc_iosize = vnd->sc_vp->v_mount->mnt_stat.f_frsize; 1426 1427 /* Default I/O size to DEV_BSIZE */ 1428 if (vnd->sc_iosize == 0) 1429 vnd->sc_iosize = DEV_BSIZE; 1430 1431 /* 1432 * Use pseudo-geometry specified. If none was provided, 1433 * use "standard" Adaptec fictitious geometry. 1434 */ 1435 if (vio->vnd_flags & VNDIOF_HASGEOM) { 1436 1437 memcpy(&vnd->sc_geom, &vio->vnd_geom, 1438 sizeof(vio->vnd_geom)); 1439 1440 /* 1441 * Sanity-check the sector size. 1442 */ 1443 if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) || 1444 vnd->sc_geom.vng_ntracks == 0 || 1445 vnd->sc_geom.vng_nsectors == 0) { 1446 error = EINVAL; 1447 goto close_and_exit; 1448 } 1449 1450 /* 1451 * Compute missing cylinder count from size 1452 */ 1453 if (vnd->sc_geom.vng_ncylinders == 0) 1454 vnd->sc_geom.vng_ncylinders = vnd->sc_size / ( 1455 (vnd->sc_geom.vng_secsize / DEV_BSIZE) * 1456 vnd->sc_geom.vng_ntracks * 1457 vnd->sc_geom.vng_nsectors); 1458 1459 /* 1460 * Compute the size (in DEV_BSIZE blocks) specified 1461 * by the geometry. 1462 */ 1463 geomsize = (int64_t)vnd->sc_geom.vng_nsectors * 1464 vnd->sc_geom.vng_ntracks * 1465 vnd->sc_geom.vng_ncylinders * 1466 (vnd->sc_geom.vng_secsize / DEV_BSIZE); 1467 1468 /* 1469 * Sanity-check the size against the specified 1470 * geometry. 1471 */ 1472 if (vnd->sc_size < geomsize) { 1473 error = EINVAL; 1474 goto close_and_exit; 1475 } 1476 } else if (vnd->sc_size >= (32 * 64)) { 1477 /* 1478 * Size must be at least 2048 DEV_BSIZE blocks 1479 * (1M) in order to use this geometry. 1480 */ 1481 vnd->sc_geom.vng_secsize = DEV_BSIZE; 1482 vnd->sc_geom.vng_nsectors = 32; 1483 vnd->sc_geom.vng_ntracks = 64; 1484 vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32); 1485 } else { 1486 vnd->sc_geom.vng_secsize = DEV_BSIZE; 1487 vnd->sc_geom.vng_nsectors = 1; 1488 vnd->sc_geom.vng_ntracks = 1; 1489 vnd->sc_geom.vng_ncylinders = vnd->sc_size; 1490 } 1491 1492 vnd_set_geometry(vnd); 1493 1494 if (vio->vnd_flags & VNDIOF_READONLY) { 1495 vnd->sc_flags |= VNF_READONLY; 1496 } 1497 1498 if ((error = vndsetcred(vnd, l->l_cred)) != 0) 1499 goto close_and_exit; 1500 1501 vndthrottle(vnd, vnd->sc_vp); 1502 vio->vnd_osize = dbtob(vnd->sc_size); 1503 if (cmd != VNDIOCSET50) 1504 vio->vnd_size = dbtob(vnd->sc_size); 1505 vnd->sc_flags |= VNF_INITED; 1506 1507 /* create the kernel thread, wait for it to be up */ 1508 error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd, 1509 &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev)); 1510 if (error) 1511 goto close_and_exit; 1512 while ((vnd->sc_flags & VNF_KTHREAD) == 0) { 1513 tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0); 1514 } 1515 #ifdef DEBUG 1516 if (vnddebug & VDB_INIT) 1517 printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n", 1518 vnd->sc_vp, (unsigned long) vnd->sc_size, 1519 vnd->sc_geom.vng_secsize, 1520 vnd->sc_geom.vng_nsectors, 1521 vnd->sc_geom.vng_ntracks, 1522 vnd->sc_geom.vng_ncylinders); 1523 #endif 1524 1525 /* Attach the disk. */ 1526 disk_attach(&vnd->sc_dkdev); 1527 1528 /* Initialize the xfer and buffer pools. */ 1529 pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0, 1530 0, 0, "vndxpl", NULL, IPL_BIO); 1531 1532 vndunlock(vnd); 1533 1534 pathbuf_destroy(pb); 1535 1536 /* Discover wedges on this disk */ 1537 dkwedge_discover(&vnd->sc_dkdev); 1538 1539 break; 1540 1541 close_and_exit: 1542 (void) vn_close(vp, fflags, l->l_cred); 1543 pathbuf_destroy(pb); 1544 unlock_and_exit: 1545 #ifdef VND_COMPRESSION 1546 /* free any allocated memory (for compressed file) */ 1547 if (vnd->sc_comp_offsets) { 1548 free(vnd->sc_comp_offsets, M_DEVBUF); 1549 vnd->sc_comp_offsets = NULL; 1550 } 1551 if (vnd->sc_comp_buff) { 1552 free(vnd->sc_comp_buff, M_DEVBUF); 1553 vnd->sc_comp_buff = NULL; 1554 } 1555 if (vnd->sc_comp_decombuf) { 1556 free(vnd->sc_comp_decombuf, M_DEVBUF); 1557 vnd->sc_comp_decombuf = NULL; 1558 } 1559 #endif /* VND_COMPRESSION */ 1560 vndunlock(vnd); 1561 return error; 1562 1563 case VNDIOCCLR50: 1564 case VNDIOCCLR: 1565 part = DISKPART(dev); 1566 pmask = (1 << part); 1567 force = (vio->vnd_flags & VNDIOF_FORCE) != 0; 1568 1569 if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0) 1570 return error; 1571 1572 break; 1573 1574 1575 case DIOCWDINFO: 1576 case DIOCSDINFO: 1577 #ifdef __HAVE_OLD_DISKLABEL 1578 case ODIOCWDINFO: 1579 case ODIOCSDINFO: 1580 #endif 1581 { 1582 struct disklabel *lp; 1583 1584 if ((error = vndlock(vnd)) != 0) 1585 return error; 1586 1587 vnd->sc_flags |= VNF_LABELLING; 1588 1589 #ifdef __HAVE_OLD_DISKLABEL 1590 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1591 memset(&newlabel, 0, sizeof newlabel); 1592 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1593 lp = &newlabel; 1594 } else 1595 #endif 1596 lp = (struct disklabel *)data; 1597 1598 error = setdisklabel(vnd->sc_dkdev.dk_label, 1599 lp, 0, vnd->sc_dkdev.dk_cpulabel); 1600 if (error == 0) { 1601 if (cmd == DIOCWDINFO 1602 #ifdef __HAVE_OLD_DISKLABEL 1603 || cmd == ODIOCWDINFO 1604 #endif 1605 ) 1606 error = writedisklabel(VNDLABELDEV(dev), 1607 vndstrategy, vnd->sc_dkdev.dk_label, 1608 vnd->sc_dkdev.dk_cpulabel); 1609 } 1610 1611 vnd->sc_flags &= ~VNF_LABELLING; 1612 1613 vndunlock(vnd); 1614 1615 if (error) 1616 return error; 1617 break; 1618 } 1619 1620 case DIOCKLABEL: 1621 if (*(int *)data != 0) 1622 vnd->sc_flags |= VNF_KLABEL; 1623 else 1624 vnd->sc_flags &= ~VNF_KLABEL; 1625 break; 1626 1627 case DIOCWLABEL: 1628 if (*(int *)data != 0) 1629 vnd->sc_flags |= VNF_WLABEL; 1630 else 1631 vnd->sc_flags &= ~VNF_WLABEL; 1632 break; 1633 1634 case DIOCGDEFLABEL: 1635 vndgetdefaultlabel(vnd, (struct disklabel *)data); 1636 break; 1637 1638 #ifdef __HAVE_OLD_DISKLABEL 1639 case ODIOCGDEFLABEL: 1640 vndgetdefaultlabel(vnd, &newlabel); 1641 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1642 return ENOTTY; 1643 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1644 break; 1645 #endif 1646 1647 case DIOCGSTRATEGY: 1648 { 1649 struct disk_strategy *dks = (void *)data; 1650 1651 /* No lock needed, never changed */ 1652 strlcpy(dks->dks_name, 1653 bufq_getstrategyname(vnd->sc_tab), 1654 sizeof(dks->dks_name)); 1655 dks->dks_paramlen = 0; 1656 break; 1657 } 1658 case DIOCGCACHE: 1659 { 1660 int *bits = (int *)data; 1661 *bits |= DKCACHE_READ | DKCACHE_WRITE; 1662 break; 1663 } 1664 case DIOCCACHESYNC: 1665 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1666 error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred, 1667 FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0); 1668 VOP_UNLOCK(vnd->sc_vp); 1669 return error; 1670 1671 default: 1672 return ENOTTY; 1673 } 1674 1675 return 0; 1676 } 1677 1678 /* 1679 * Duplicate the current processes' credentials. Since we are called only 1680 * as the result of a SET ioctl and only root can do that, any future access 1681 * to this "disk" is essentially as root. Note that credentials may change 1682 * if some other uid can write directly to the mapped file (NFS). 1683 */ 1684 static int 1685 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred) 1686 { 1687 struct uio auio; 1688 struct iovec aiov; 1689 char *tmpbuf; 1690 int error; 1691 1692 vnd->sc_cred = kauth_cred_dup(cred); 1693 tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK); 1694 1695 /* XXX: Horrible kludge to establish credentials for NFS */ 1696 aiov.iov_base = tmpbuf; 1697 aiov.iov_len = uimin(DEV_BSIZE, dbtob(vnd->sc_size)); 1698 auio.uio_iov = &aiov; 1699 auio.uio_iovcnt = 1; 1700 auio.uio_offset = 0; 1701 auio.uio_rw = UIO_READ; 1702 auio.uio_resid = aiov.iov_len; 1703 UIO_SETUP_SYSSPACE(&auio); 1704 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 1705 error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred); 1706 if (error == 0) { 1707 /* 1708 * Because vnd does all IO directly through the vnode 1709 * we need to flush (at least) the buffer from the above 1710 * VOP_READ from the buffer cache to prevent cache 1711 * incoherencies. Also, be careful to write dirty 1712 * buffers back to stable storage. 1713 */ 1714 error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred, 1715 curlwp, 0, 0); 1716 } 1717 VOP_UNLOCK(vnd->sc_vp); 1718 1719 free(tmpbuf, M_TEMP); 1720 return error; 1721 } 1722 1723 /* 1724 * Set maxactive based on FS type 1725 */ 1726 static void 1727 vndthrottle(struct vnd_softc *vnd, struct vnode *vp) 1728 { 1729 1730 if (vp->v_tag == VT_NFS) 1731 vnd->sc_maxactive = 2; 1732 else 1733 vnd->sc_maxactive = 8; 1734 1735 if (vnd->sc_maxactive < 1) 1736 vnd->sc_maxactive = 1; 1737 } 1738 1739 #if 0 1740 static void 1741 vndshutdown(void) 1742 { 1743 struct vnd_softc *vnd; 1744 1745 for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++) 1746 if (vnd->sc_flags & VNF_INITED) 1747 vndclear(vnd); 1748 } 1749 #endif 1750 1751 static void 1752 vndclear(struct vnd_softc *vnd, int myminor) 1753 { 1754 struct vnode *vp = vnd->sc_vp; 1755 int fflags = FREAD; 1756 int bmaj, cmaj, i, mn; 1757 int s; 1758 1759 #ifdef DEBUG 1760 if (vnddebug & VDB_FOLLOW) 1761 printf("vndclear(%p): vp %p\n", vnd, vp); 1762 #endif 1763 /* locate the major number */ 1764 bmaj = bdevsw_lookup_major(&vnd_bdevsw); 1765 cmaj = cdevsw_lookup_major(&vnd_cdevsw); 1766 1767 /* Nuke the vnodes for any open instances */ 1768 for (i = 0; i < MAXPARTITIONS; i++) { 1769 mn = DISKMINOR(device_unit(vnd->sc_dev), i); 1770 vdevgone(bmaj, mn, mn, VBLK); 1771 if (mn != myminor) /* XXX avoid to kill own vnode */ 1772 vdevgone(cmaj, mn, mn, VCHR); 1773 } 1774 1775 if ((vnd->sc_flags & VNF_READONLY) == 0) 1776 fflags |= FWRITE; 1777 1778 s = splbio(); 1779 bufq_drain(vnd->sc_tab); 1780 splx(s); 1781 1782 vnd->sc_flags |= VNF_VUNCONF; 1783 wakeup(&vnd->sc_tab); 1784 while (vnd->sc_flags & VNF_KTHREAD) 1785 tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0); 1786 1787 #ifdef VND_COMPRESSION 1788 /* free the compressed file buffers */ 1789 if (vnd->sc_flags & VNF_COMP) { 1790 if (vnd->sc_comp_offsets) { 1791 free(vnd->sc_comp_offsets, M_DEVBUF); 1792 vnd->sc_comp_offsets = NULL; 1793 } 1794 if (vnd->sc_comp_buff) { 1795 free(vnd->sc_comp_buff, M_DEVBUF); 1796 vnd->sc_comp_buff = NULL; 1797 } 1798 if (vnd->sc_comp_decombuf) { 1799 free(vnd->sc_comp_decombuf, M_DEVBUF); 1800 vnd->sc_comp_decombuf = NULL; 1801 } 1802 } 1803 #endif /* VND_COMPRESSION */ 1804 vnd->sc_flags &= 1805 ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL 1806 | VNF_VUNCONF | VNF_COMP | VNF_CLEARING); 1807 if (vp == NULL) 1808 panic("vndclear: null vp"); 1809 (void) vn_close(vp, fflags, vnd->sc_cred); 1810 kauth_cred_free(vnd->sc_cred); 1811 vnd->sc_vp = NULL; 1812 vnd->sc_cred = NULL; 1813 vnd->sc_size = 0; 1814 } 1815 1816 static int 1817 vndsize(dev_t dev) 1818 { 1819 struct vnd_softc *sc; 1820 struct disklabel *lp; 1821 int part, unit, omask; 1822 int size; 1823 1824 unit = vndunit(dev); 1825 sc = device_lookup_private(&vnd_cd, unit); 1826 if (sc == NULL) 1827 return -1; 1828 1829 if ((sc->sc_flags & VNF_INITED) == 0) 1830 return -1; 1831 1832 part = DISKPART(dev); 1833 omask = sc->sc_dkdev.dk_openmask & (1 << part); 1834 lp = sc->sc_dkdev.dk_label; 1835 1836 if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp)) /* XXX */ 1837 return -1; 1838 1839 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1840 size = -1; 1841 else 1842 size = lp->d_partitions[part].p_size * 1843 (lp->d_secsize / DEV_BSIZE); 1844 1845 if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp)) /* XXX */ 1846 return -1; 1847 1848 return size; 1849 } 1850 1851 static int 1852 vnddump(dev_t dev, daddr_t blkno, void *va, 1853 size_t size) 1854 { 1855 1856 /* Not implemented. */ 1857 return ENXIO; 1858 } 1859 1860 static void 1861 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp) 1862 { 1863 struct vndgeom *vng = &sc->sc_geom; 1864 struct partition *pp; 1865 unsigned spb; 1866 1867 memset(lp, 0, sizeof(*lp)); 1868 1869 spb = vng->vng_secsize / DEV_BSIZE; 1870 if (sc->sc_size / spb > UINT32_MAX) 1871 lp->d_secperunit = UINT32_MAX; 1872 else 1873 lp->d_secperunit = sc->sc_size / spb; 1874 lp->d_secsize = vng->vng_secsize; 1875 lp->d_nsectors = vng->vng_nsectors; 1876 lp->d_ntracks = vng->vng_ntracks; 1877 lp->d_ncylinders = vng->vng_ncylinders; 1878 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1879 1880 strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename)); 1881 lp->d_type = DKTYPE_VND; 1882 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1883 lp->d_rpm = 3600; 1884 lp->d_interleave = 1; 1885 lp->d_flags = 0; 1886 1887 pp = &lp->d_partitions[RAW_PART]; 1888 pp->p_offset = 0; 1889 pp->p_size = lp->d_secperunit; 1890 pp->p_fstype = FS_UNUSED; 1891 lp->d_npartitions = RAW_PART + 1; 1892 1893 lp->d_magic = DISKMAGIC; 1894 lp->d_magic2 = DISKMAGIC; 1895 lp->d_checksum = dkcksum(lp); 1896 } 1897 1898 /* 1899 * Read the disklabel from a vnd. If one is not present, create a fake one. 1900 */ 1901 static void 1902 vndgetdisklabel(dev_t dev, struct vnd_softc *sc) 1903 { 1904 const char *errstring; 1905 struct disklabel *lp = sc->sc_dkdev.dk_label; 1906 struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel; 1907 int i; 1908 1909 memset(clp, 0, sizeof(*clp)); 1910 1911 vndgetdefaultlabel(sc, lp); 1912 1913 /* 1914 * Call the generic disklabel extraction routine. 1915 */ 1916 errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp); 1917 if (errstring) { 1918 /* 1919 * Lack of disklabel is common, but we print the warning 1920 * anyway, since it might contain other useful information. 1921 */ 1922 aprint_normal_dev(sc->sc_dev, "%s\n", errstring); 1923 1924 /* 1925 * For historical reasons, if there's no disklabel 1926 * present, all partitions must be FS_BSDFFS and 1927 * occupy the entire disk. 1928 */ 1929 for (i = 0; i < MAXPARTITIONS; i++) { 1930 /* 1931 * Don't wipe out port specific hack (such as 1932 * dos partition hack of i386 port). 1933 */ 1934 if (lp->d_partitions[i].p_size != 0) 1935 continue; 1936 1937 lp->d_partitions[i].p_size = lp->d_secperunit; 1938 lp->d_partitions[i].p_offset = 0; 1939 lp->d_partitions[i].p_fstype = FS_BSDFFS; 1940 } 1941 1942 strncpy(lp->d_packname, "default label", 1943 sizeof(lp->d_packname)); 1944 1945 lp->d_npartitions = MAXPARTITIONS; 1946 lp->d_checksum = dkcksum(lp); 1947 } 1948 } 1949 1950 /* 1951 * Wait interruptibly for an exclusive lock. 1952 * 1953 * XXX 1954 * Several drivers do this; it should be abstracted and made MP-safe. 1955 */ 1956 static int 1957 vndlock(struct vnd_softc *sc) 1958 { 1959 int error; 1960 1961 while ((sc->sc_flags & VNF_LOCKED) != 0) { 1962 sc->sc_flags |= VNF_WANTED; 1963 if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0) 1964 return error; 1965 } 1966 sc->sc_flags |= VNF_LOCKED; 1967 return 0; 1968 } 1969 1970 /* 1971 * Unlock and wake up any waiters. 1972 */ 1973 static void 1974 vndunlock(struct vnd_softc *sc) 1975 { 1976 1977 sc->sc_flags &= ~VNF_LOCKED; 1978 if ((sc->sc_flags & VNF_WANTED) != 0) { 1979 sc->sc_flags &= ~VNF_WANTED; 1980 wakeup(sc); 1981 } 1982 } 1983 1984 #ifdef VND_COMPRESSION 1985 /* compressed file read */ 1986 static void 1987 compstrategy(struct buf *bp, off_t bn) 1988 { 1989 int error; 1990 int unit = vndunit(bp->b_dev); 1991 struct vnd_softc *vnd = 1992 device_lookup_private(&vnd_cd, unit); 1993 u_int32_t comp_block; 1994 struct uio auio; 1995 char *addr; 1996 int s; 1997 1998 /* set up constants for data move */ 1999 auio.uio_rw = UIO_READ; 2000 UIO_SETUP_SYSSPACE(&auio); 2001 2002 /* read, and transfer the data */ 2003 addr = bp->b_data; 2004 bp->b_resid = bp->b_bcount; 2005 s = splbio(); 2006 while (bp->b_resid > 0) { 2007 unsigned length; 2008 size_t length_in_buffer; 2009 u_int32_t offset_in_buffer; 2010 struct iovec aiov; 2011 2012 /* calculate the compressed block number */ 2013 comp_block = bn / (off_t)vnd->sc_comp_blksz; 2014 2015 /* check for good block number */ 2016 if (comp_block >= vnd->sc_comp_numoffs) { 2017 bp->b_error = EINVAL; 2018 splx(s); 2019 return; 2020 } 2021 2022 /* read in the compressed block, if not in buffer */ 2023 if (comp_block != vnd->sc_comp_buffblk) { 2024 length = vnd->sc_comp_offsets[comp_block + 1] - 2025 vnd->sc_comp_offsets[comp_block]; 2026 vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY); 2027 error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff, 2028 length, vnd->sc_comp_offsets[comp_block], 2029 UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred, 2030 NULL, NULL); 2031 if (error) { 2032 bp->b_error = error; 2033 VOP_UNLOCK(vnd->sc_vp); 2034 splx(s); 2035 return; 2036 } 2037 /* uncompress the buffer */ 2038 vnd->sc_comp_stream.next_in = vnd->sc_comp_buff; 2039 vnd->sc_comp_stream.avail_in = length; 2040 vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf; 2041 vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz; 2042 inflateReset(&vnd->sc_comp_stream); 2043 error = inflate(&vnd->sc_comp_stream, Z_FINISH); 2044 if (error != Z_STREAM_END) { 2045 if (vnd->sc_comp_stream.msg) 2046 aprint_normal_dev(vnd->sc_dev, 2047 "compressed file, %s\n", 2048 vnd->sc_comp_stream.msg); 2049 bp->b_error = EBADMSG; 2050 VOP_UNLOCK(vnd->sc_vp); 2051 splx(s); 2052 return; 2053 } 2054 vnd->sc_comp_buffblk = comp_block; 2055 VOP_UNLOCK(vnd->sc_vp); 2056 } 2057 2058 /* transfer the usable uncompressed data */ 2059 offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz; 2060 length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer; 2061 if (length_in_buffer > bp->b_resid) 2062 length_in_buffer = bp->b_resid; 2063 auio.uio_iov = &aiov; 2064 auio.uio_iovcnt = 1; 2065 aiov.iov_base = addr; 2066 aiov.iov_len = length_in_buffer; 2067 auio.uio_resid = aiov.iov_len; 2068 auio.uio_offset = 0; 2069 error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer, 2070 length_in_buffer, &auio); 2071 if (error) { 2072 bp->b_error = error; 2073 splx(s); 2074 return; 2075 } 2076 2077 bn += length_in_buffer; 2078 addr += length_in_buffer; 2079 bp->b_resid -= length_in_buffer; 2080 } 2081 splx(s); 2082 } 2083 2084 /* compression memory allocation routines */ 2085 static void * 2086 vnd_alloc(void *aux, u_int items, u_int siz) 2087 { 2088 return malloc(items * siz, M_TEMP, M_NOWAIT); 2089 } 2090 2091 static void 2092 vnd_free(void *aux, void *ptr) 2093 { 2094 free(ptr, M_TEMP); 2095 } 2096 #endif /* VND_COMPRESSION */ 2097 2098 static void 2099 vnd_set_geometry(struct vnd_softc *vnd) 2100 { 2101 struct disk_geom *dg = &vnd->sc_dkdev.dk_geom; 2102 unsigned spb; 2103 2104 memset(dg, 0, sizeof(*dg)); 2105 2106 spb = vnd->sc_geom.vng_secsize / DEV_BSIZE; 2107 dg->dg_secperunit = vnd->sc_size / spb; 2108 dg->dg_secsize = vnd->sc_geom.vng_secsize; 2109 dg->dg_nsectors = vnd->sc_geom.vng_nsectors; 2110 dg->dg_ntracks = vnd->sc_geom.vng_ntracks; 2111 dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders; 2112 2113 #ifdef DEBUG 2114 if (vnddebug & VDB_LABEL) { 2115 printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit); 2116 printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders); 2117 } 2118 #endif 2119 disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL); 2120 } 2121 2122 #ifdef VND_COMPRESSION 2123 #define VND_DEPENDS "zlib" 2124 #else 2125 #define VND_DEPENDS NULL 2126 #endif 2127 2128 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS); 2129 2130 #ifdef _MODULE 2131 int vnd_bmajor = -1, vnd_cmajor = -1; 2132 2133 CFDRIVER_DECL(vnd, DV_DISK, NULL); 2134 #endif 2135 2136 static int 2137 vnd_modcmd(modcmd_t cmd, void *arg) 2138 { 2139 int error = 0; 2140 2141 switch (cmd) { 2142 case MODULE_CMD_INIT: 2143 #ifdef _MODULE 2144 /* 2145 * Attach the {b,c}devsw's 2146 */ 2147 error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor, 2148 &vnd_cdevsw, &vnd_cmajor); 2149 if (error) { 2150 #ifdef DIAGNOSTIC 2151 aprint_error("%s: unable to attach %s devsw, " 2152 "error %d", __func__, vnd_cd.cd_name, error); 2153 #endif 2154 break; 2155 } 2156 2157 error = config_cfdriver_attach(&vnd_cd); 2158 if (error) { 2159 devsw_detach(&vnd_bdevsw, &vnd_cdevsw); 2160 break; 2161 } 2162 2163 error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca); 2164 if (error) { 2165 config_cfdriver_detach(&vnd_cd); 2166 devsw_detach(&vnd_bdevsw, &vnd_cdevsw); 2167 #ifdef DIAGNOSTIC 2168 aprint_error("%s: unable to register cfattach for \n" 2169 "%s, error %d", __func__, vnd_cd.cd_name, error); 2170 #endif 2171 break; 2172 } 2173 #endif 2174 break; 2175 2176 case MODULE_CMD_FINI: 2177 #ifdef _MODULE 2178 /* 2179 * Remove device from autoconf database 2180 */ 2181 error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca); 2182 if (error) { 2183 #ifdef DIAGNOSTIC 2184 aprint_error("%s: failed to detach %s cfattach, " 2185 "error %d\n", __func__, vnd_cd.cd_name, error); 2186 #endif 2187 break; 2188 } 2189 error = config_cfdriver_detach(&vnd_cd); 2190 if (error) { 2191 (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca); 2192 #ifdef DIAGNOSTIC 2193 aprint_error("%s: failed to detach %s cfdriver, " 2194 "error %d\n", __func__, vnd_cd.cd_name, error); 2195 break; 2196 #endif 2197 } 2198 /* 2199 * Remove {b,c}devsw's 2200 */ 2201 devsw_detach(&vnd_bdevsw, &vnd_cdevsw); 2202 2203 #endif 2204 break; 2205 2206 case MODULE_CMD_STAT: 2207 return ENOTTY; 2208 2209 default: 2210 return ENOTTY; 2211 } 2212 2213 return error; 2214 } 2215