1 /* $NetBSD: spec_vnops.c,v 1.131 2010/08/21 13:19:40 pgoyette Exp $ */ 2 3 /*- 4 * Copyright (c) 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 17 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 18 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 19 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 20 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 21 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 22 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 23 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 24 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 25 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 26 * POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 /* 30 * Copyright (c) 1989, 1993 31 * The Regents of the University of California. All rights reserved. 32 * 33 * Redistribution and use in source and binary forms, with or without 34 * modification, are permitted provided that the following conditions 35 * are met: 36 * 1. Redistributions of source code must retain the above copyright 37 * notice, this list of conditions and the following disclaimer. 38 * 2. Redistributions in binary form must reproduce the above copyright 39 * notice, this list of conditions and the following disclaimer in the 40 * documentation and/or other materials provided with the distribution. 41 * 3. Neither the name of the University nor the names of its contributors 42 * may be used to endorse or promote products derived from this software 43 * without specific prior written permission. 44 * 45 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 48 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 55 * SUCH DAMAGE. 56 * 57 * @(#)spec_vnops.c 8.15 (Berkeley) 7/14/95 58 */ 59 60 #include <sys/cdefs.h> 61 __KERNEL_RCSID(0, "$NetBSD: spec_vnops.c,v 1.131 2010/08/21 13:19:40 pgoyette Exp $"); 62 63 #include <sys/param.h> 64 #include <sys/proc.h> 65 #include <sys/systm.h> 66 #include <sys/kernel.h> 67 #include <sys/conf.h> 68 #include <sys/buf.h> 69 #include <sys/mount.h> 70 #include <sys/namei.h> 71 #include <sys/vnode.h> 72 #include <sys/stat.h> 73 #include <sys/errno.h> 74 #include <sys/ioctl.h> 75 #include <sys/poll.h> 76 #include <sys/file.h> 77 #include <sys/disklabel.h> 78 #include <sys/lockf.h> 79 #include <sys/tty.h> 80 #include <sys/kauth.h> 81 #include <sys/fstrans.h> 82 #include <sys/module.h> 83 84 #include <miscfs/genfs/genfs.h> 85 #include <miscfs/specfs/specdev.h> 86 87 /* symbolic sleep message strings for devices */ 88 const char devopn[] = "devopn"; 89 const char devio[] = "devio"; 90 const char devwait[] = "devwait"; 91 const char devin[] = "devin"; 92 const char devout[] = "devout"; 93 const char devioc[] = "devioc"; 94 const char devcls[] = "devcls"; 95 96 vnode_t *specfs_hash[SPECHSZ]; 97 98 /* 99 * This vnode operations vector is used for special device nodes 100 * created from whole cloth by the kernel. For the ops vector for 101 * vnodes built from special devices found in a filesystem, see (e.g) 102 * ffs_specop_entries[] in ffs_vnops.c or the equivalent for other 103 * filesystems. 104 */ 105 106 int (**spec_vnodeop_p)(void *); 107 const struct vnodeopv_entry_desc spec_vnodeop_entries[] = { 108 { &vop_default_desc, vn_default_error }, 109 { &vop_lookup_desc, spec_lookup }, /* lookup */ 110 { &vop_create_desc, spec_create }, /* create */ 111 { &vop_mknod_desc, spec_mknod }, /* mknod */ 112 { &vop_open_desc, spec_open }, /* open */ 113 { &vop_close_desc, spec_close }, /* close */ 114 { &vop_access_desc, spec_access }, /* access */ 115 { &vop_getattr_desc, spec_getattr }, /* getattr */ 116 { &vop_setattr_desc, spec_setattr }, /* setattr */ 117 { &vop_read_desc, spec_read }, /* read */ 118 { &vop_write_desc, spec_write }, /* write */ 119 { &vop_fcntl_desc, spec_fcntl }, /* fcntl */ 120 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 121 { &vop_poll_desc, spec_poll }, /* poll */ 122 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 123 { &vop_revoke_desc, spec_revoke }, /* revoke */ 124 { &vop_mmap_desc, spec_mmap }, /* mmap */ 125 { &vop_fsync_desc, spec_fsync }, /* fsync */ 126 { &vop_seek_desc, spec_seek }, /* seek */ 127 { &vop_remove_desc, spec_remove }, /* remove */ 128 { &vop_link_desc, spec_link }, /* link */ 129 { &vop_rename_desc, spec_rename }, /* rename */ 130 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 131 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 132 { &vop_symlink_desc, spec_symlink }, /* symlink */ 133 { &vop_readdir_desc, spec_readdir }, /* readdir */ 134 { &vop_readlink_desc, spec_readlink }, /* readlink */ 135 { &vop_abortop_desc, spec_abortop }, /* abortop */ 136 { &vop_inactive_desc, spec_inactive }, /* inactive */ 137 { &vop_reclaim_desc, spec_reclaim }, /* reclaim */ 138 { &vop_lock_desc, spec_lock }, /* lock */ 139 { &vop_unlock_desc, spec_unlock }, /* unlock */ 140 { &vop_bmap_desc, spec_bmap }, /* bmap */ 141 { &vop_strategy_desc, spec_strategy }, /* strategy */ 142 { &vop_print_desc, spec_print }, /* print */ 143 { &vop_islocked_desc, spec_islocked }, /* islocked */ 144 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 145 { &vop_advlock_desc, spec_advlock }, /* advlock */ 146 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 147 { &vop_getpages_desc, spec_getpages }, /* getpages */ 148 { &vop_putpages_desc, spec_putpages }, /* putpages */ 149 { NULL, NULL } 150 }; 151 const struct vnodeopv_desc spec_vnodeop_opv_desc = 152 { &spec_vnodeop_p, spec_vnodeop_entries }; 153 154 static kauth_listener_t rawio_listener; 155 156 /* Returns true if vnode is /dev/mem or /dev/kmem. */ 157 bool 158 iskmemvp(struct vnode *vp) 159 { 160 return ((vp->v_type == VCHR) && iskmemdev(vp->v_rdev)); 161 } 162 163 /* 164 * Returns true if dev is /dev/mem or /dev/kmem. 165 */ 166 int 167 iskmemdev(dev_t dev) 168 { 169 /* mem_no is emitted by config(8) to generated devsw.c */ 170 extern const int mem_no; 171 172 /* minor 14 is /dev/io on i386 with COMPAT_10 */ 173 return (major(dev) == mem_no && (minor(dev) < 2 || minor(dev) == 14)); 174 } 175 176 static int 177 rawio_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 178 void *arg0, void *arg1, void *arg2, void *arg3) 179 { 180 int result; 181 182 result = KAUTH_RESULT_DEFER; 183 184 if ((action != KAUTH_DEVICE_RAWIO_SPEC) && 185 (action != KAUTH_DEVICE_RAWIO_PASSTHRU)) 186 return result; 187 188 /* Access is mandated by permissions. */ 189 result = KAUTH_RESULT_ALLOW; 190 191 return result; 192 } 193 194 void 195 spec_init(void) 196 { 197 198 rawio_listener = kauth_listen_scope(KAUTH_SCOPE_DEVICE, 199 rawio_listener_cb, NULL); 200 } 201 202 /* 203 * Initialize a vnode that represents a device. 204 */ 205 void 206 spec_node_init(vnode_t *vp, dev_t rdev) 207 { 208 specnode_t *sn; 209 specdev_t *sd; 210 vnode_t *vp2; 211 vnode_t **vpp; 212 213 KASSERT(vp->v_type == VBLK || vp->v_type == VCHR); 214 KASSERT(vp->v_specnode == NULL); 215 216 /* 217 * Search the hash table for this device. If known, add a 218 * reference to the device structure. If not known, create 219 * a new entry to represent the device. In all cases add 220 * the vnode to the hash table. 221 */ 222 sn = kmem_alloc(sizeof(*sn), KM_SLEEP); 223 if (sn == NULL) { 224 /* XXX */ 225 panic("spec_node_init: unable to allocate memory"); 226 } 227 sd = kmem_alloc(sizeof(*sd), KM_SLEEP); 228 if (sd == NULL) { 229 /* XXX */ 230 panic("spec_node_init: unable to allocate memory"); 231 } 232 mutex_enter(&device_lock); 233 vpp = &specfs_hash[SPECHASH(rdev)]; 234 for (vp2 = *vpp; vp2 != NULL; vp2 = vp2->v_specnext) { 235 KASSERT(vp2->v_specnode != NULL); 236 if (rdev == vp2->v_rdev && vp->v_type == vp2->v_type) { 237 break; 238 } 239 } 240 if (vp2 == NULL) { 241 /* No existing record, create a new one. */ 242 sd->sd_rdev = rdev; 243 sd->sd_mountpoint = NULL; 244 sd->sd_lockf = NULL; 245 sd->sd_refcnt = 1; 246 sd->sd_opencnt = 0; 247 sd->sd_bdevvp = NULL; 248 sn->sn_dev = sd; 249 sd = NULL; 250 } else { 251 /* Use the existing record. */ 252 sn->sn_dev = vp2->v_specnode->sn_dev; 253 sn->sn_dev->sd_refcnt++; 254 } 255 /* Insert vnode into the hash chain. */ 256 sn->sn_opencnt = 0; 257 sn->sn_rdev = rdev; 258 sn->sn_gone = false; 259 vp->v_specnode = sn; 260 vp->v_specnext = *vpp; 261 *vpp = vp; 262 mutex_exit(&device_lock); 263 264 /* Free the record we allocated if unused. */ 265 if (sd != NULL) { 266 kmem_free(sd, sizeof(*sd)); 267 } 268 } 269 270 /* 271 * A vnode representing a special device is going away. Close 272 * the device if the vnode holds it open. 273 */ 274 void 275 spec_node_revoke(vnode_t *vp) 276 { 277 specnode_t *sn; 278 specdev_t *sd; 279 280 sn = vp->v_specnode; 281 sd = sn->sn_dev; 282 283 KASSERT(vp->v_type == VBLK || vp->v_type == VCHR); 284 KASSERT(vp->v_specnode != NULL); 285 KASSERT((vp->v_iflag & VI_XLOCK) != 0); 286 KASSERT(sn->sn_gone == false); 287 288 mutex_enter(&device_lock); 289 KASSERT(sn->sn_opencnt <= sd->sd_opencnt); 290 if (sn->sn_opencnt != 0) { 291 sd->sd_opencnt -= (sn->sn_opencnt - 1); 292 sn->sn_opencnt = 1; 293 sn->sn_gone = true; 294 mutex_exit(&device_lock); 295 296 VOP_CLOSE(vp, FNONBLOCK, NOCRED); 297 298 mutex_enter(&device_lock); 299 KASSERT(sn->sn_opencnt == 0); 300 } 301 mutex_exit(&device_lock); 302 } 303 304 /* 305 * A vnode representing a special device is being recycled. 306 * Destroy the specfs component. 307 */ 308 void 309 spec_node_destroy(vnode_t *vp) 310 { 311 specnode_t *sn; 312 specdev_t *sd; 313 vnode_t **vpp, *vp2; 314 int refcnt; 315 316 sn = vp->v_specnode; 317 sd = sn->sn_dev; 318 319 KASSERT(vp->v_type == VBLK || vp->v_type == VCHR); 320 KASSERT(vp->v_specnode != NULL); 321 KASSERT(sn->sn_opencnt == 0); 322 323 mutex_enter(&device_lock); 324 /* Remove from the hash and destroy the node. */ 325 vpp = &specfs_hash[SPECHASH(vp->v_rdev)]; 326 for (vp2 = *vpp;; vp2 = vp2->v_specnext) { 327 if (vp2 == NULL) { 328 panic("spec_node_destroy: corrupt hash"); 329 } 330 if (vp2 == vp) { 331 KASSERT(vp == *vpp); 332 *vpp = vp->v_specnext; 333 break; 334 } 335 if (vp2->v_specnext == vp) { 336 vp2->v_specnext = vp->v_specnext; 337 break; 338 } 339 } 340 sn = vp->v_specnode; 341 vp->v_specnode = NULL; 342 refcnt = sd->sd_refcnt--; 343 KASSERT(refcnt > 0); 344 mutex_exit(&device_lock); 345 346 /* If the device is no longer in use, destroy our record. */ 347 if (refcnt == 1) { 348 KASSERT(sd->sd_opencnt == 0); 349 KASSERT(sd->sd_bdevvp == NULL); 350 kmem_free(sd, sizeof(*sd)); 351 } 352 kmem_free(sn, sizeof(*sn)); 353 } 354 355 /* 356 * Trivial lookup routine that always fails. 357 */ 358 int 359 spec_lookup(void *v) 360 { 361 struct vop_lookup_args /* { 362 struct vnode *a_dvp; 363 struct vnode **a_vpp; 364 struct componentname *a_cnp; 365 } */ *ap = v; 366 367 *ap->a_vpp = NULL; 368 return (ENOTDIR); 369 } 370 371 /* 372 * Open a special file. 373 */ 374 /* ARGSUSED */ 375 int 376 spec_open(void *v) 377 { 378 struct vop_open_args /* { 379 struct vnode *a_vp; 380 int a_mode; 381 kauth_cred_t a_cred; 382 } */ *ap = v; 383 struct lwp *l; 384 struct vnode *vp; 385 dev_t dev; 386 int error; 387 struct partinfo pi; 388 enum kauth_device_req req; 389 specnode_t *sn; 390 specdev_t *sd; 391 392 u_int gen; 393 const char *name; 394 395 l = curlwp; 396 vp = ap->a_vp; 397 dev = vp->v_rdev; 398 sn = vp->v_specnode; 399 sd = sn->sn_dev; 400 name = NULL; 401 gen = 0; 402 403 /* 404 * Don't allow open if fs is mounted -nodev. 405 */ 406 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV)) 407 return (ENXIO); 408 409 switch (ap->a_mode & (FREAD | FWRITE)) { 410 case FREAD | FWRITE: 411 req = KAUTH_REQ_DEVICE_RAWIO_SPEC_RW; 412 break; 413 case FWRITE: 414 req = KAUTH_REQ_DEVICE_RAWIO_SPEC_WRITE; 415 break; 416 default: 417 req = KAUTH_REQ_DEVICE_RAWIO_SPEC_READ; 418 break; 419 } 420 421 switch (vp->v_type) { 422 case VCHR: 423 error = kauth_authorize_device_spec(ap->a_cred, req, vp); 424 if (error != 0) 425 return (error); 426 427 /* 428 * Character devices can accept opens from multiple 429 * vnodes. 430 */ 431 mutex_enter(&device_lock); 432 if (sn->sn_gone) { 433 mutex_exit(&device_lock); 434 return (EBADF); 435 } 436 sd->sd_opencnt++; 437 sn->sn_opencnt++; 438 mutex_exit(&device_lock); 439 if (cdev_type(dev) == D_TTY) 440 vp->v_vflag |= VV_ISTTY; 441 VOP_UNLOCK(vp); 442 do { 443 const struct cdevsw *cdev; 444 445 gen = module_gen; 446 error = cdev_open(dev, ap->a_mode, S_IFCHR, l); 447 if (error != ENXIO) 448 break; 449 450 /* Check if we already have a valid driver */ 451 mutex_enter(&device_lock); 452 cdev = cdevsw_lookup(dev); 453 mutex_exit(&device_lock); 454 if (cdev != NULL) 455 break; 456 457 /* Get device name from devsw_conv array */ 458 if ((name = cdevsw_getname(major(dev))) == NULL) 459 break; 460 461 /* Try to autoload device module */ 462 (void) module_autoload(name, MODULE_CLASS_DRIVER); 463 } while (gen != module_gen); 464 465 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 466 break; 467 468 case VBLK: 469 error = kauth_authorize_device_spec(ap->a_cred, req, vp); 470 if (error != 0) 471 return (error); 472 473 /* 474 * For block devices, permit only one open. The buffer 475 * cache cannot remain self-consistent with multiple 476 * vnodes holding a block device open. 477 */ 478 mutex_enter(&device_lock); 479 if (sn->sn_gone) { 480 mutex_exit(&device_lock); 481 return (EBADF); 482 } 483 if (sd->sd_opencnt != 0) { 484 mutex_exit(&device_lock); 485 return EBUSY; 486 } 487 sn->sn_opencnt = 1; 488 sd->sd_opencnt = 1; 489 sd->sd_bdevvp = vp; 490 mutex_exit(&device_lock); 491 do { 492 const struct bdevsw *bdev; 493 494 gen = module_gen; 495 error = bdev_open(dev, ap->a_mode, S_IFBLK, l); 496 if (error != ENXIO) 497 break; 498 499 /* Check if we already have a valid driver */ 500 mutex_enter(&device_lock); 501 bdev = bdevsw_lookup(dev); 502 mutex_exit(&device_lock); 503 if (bdev != NULL) 504 break; 505 506 /* Get device name from devsw_conv array */ 507 if ((name = bdevsw_getname(major(dev))) == NULL) 508 break; 509 510 VOP_UNLOCK(vp); 511 512 /* Try to autoload device module */ 513 (void) module_autoload(name, MODULE_CLASS_DRIVER); 514 515 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 516 } while (gen != module_gen); 517 518 break; 519 520 case VNON: 521 case VLNK: 522 case VDIR: 523 case VREG: 524 case VBAD: 525 case VFIFO: 526 case VSOCK: 527 default: 528 return 0; 529 } 530 531 mutex_enter(&device_lock); 532 if (sn->sn_gone) { 533 if (error == 0) 534 error = EBADF; 535 } else if (error != 0) { 536 sd->sd_opencnt--; 537 sn->sn_opencnt--; 538 if (vp->v_type == VBLK) 539 sd->sd_bdevvp = NULL; 540 541 } 542 mutex_exit(&device_lock); 543 544 if (cdev_type(dev) != D_DISK || error != 0) 545 return error; 546 547 if (vp->v_type == VCHR) 548 error = cdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp); 549 else 550 error = bdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp); 551 if (error == 0) 552 uvm_vnp_setsize(vp, 553 (voff_t)pi.disklab->d_secsize * pi.part->p_size); 554 return 0; 555 } 556 557 /* 558 * Vnode op for read 559 */ 560 /* ARGSUSED */ 561 int 562 spec_read(void *v) 563 { 564 struct vop_read_args /* { 565 struct vnode *a_vp; 566 struct uio *a_uio; 567 int a_ioflag; 568 kauth_cred_t a_cred; 569 } */ *ap = v; 570 struct vnode *vp = ap->a_vp; 571 struct uio *uio = ap->a_uio; 572 struct lwp *l = curlwp; 573 struct buf *bp; 574 daddr_t bn; 575 int bsize, bscale; 576 struct partinfo dpart; 577 int n, on; 578 int error = 0; 579 580 #ifdef DIAGNOSTIC 581 if (uio->uio_rw != UIO_READ) 582 panic("spec_read mode"); 583 if (&uio->uio_vmspace->vm_map != kernel_map && 584 uio->uio_vmspace != curproc->p_vmspace) 585 panic("spec_read proc"); 586 #endif 587 if (uio->uio_resid == 0) 588 return (0); 589 590 switch (vp->v_type) { 591 592 case VCHR: 593 VOP_UNLOCK(vp); 594 error = cdev_read(vp->v_rdev, uio, ap->a_ioflag); 595 vn_lock(vp, LK_SHARED | LK_RETRY); 596 return (error); 597 598 case VBLK: 599 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 600 if (uio->uio_offset < 0) 601 return (EINVAL); 602 bsize = BLKDEV_IOSIZE; 603 if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) { 604 if (dpart.part->p_fstype == FS_BSDFFS && 605 dpart.part->p_frag != 0 && dpart.part->p_fsize != 0) 606 bsize = dpart.part->p_frag * 607 dpart.part->p_fsize; 608 } 609 bscale = bsize >> DEV_BSHIFT; 610 do { 611 bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1); 612 on = uio->uio_offset % bsize; 613 n = min((unsigned)(bsize - on), uio->uio_resid); 614 error = bread(vp, bn, bsize, NOCRED, 0, &bp); 615 n = min(n, bsize - bp->b_resid); 616 if (error) { 617 brelse(bp, 0); 618 return (error); 619 } 620 error = uiomove((char *)bp->b_data + on, n, uio); 621 brelse(bp, 0); 622 } while (error == 0 && uio->uio_resid > 0 && n != 0); 623 return (error); 624 625 default: 626 panic("spec_read type"); 627 } 628 /* NOTREACHED */ 629 } 630 631 /* 632 * Vnode op for write 633 */ 634 /* ARGSUSED */ 635 int 636 spec_write(void *v) 637 { 638 struct vop_write_args /* { 639 struct vnode *a_vp; 640 struct uio *a_uio; 641 int a_ioflag; 642 kauth_cred_t a_cred; 643 } */ *ap = v; 644 struct vnode *vp = ap->a_vp; 645 struct uio *uio = ap->a_uio; 646 struct lwp *l = curlwp; 647 struct buf *bp; 648 daddr_t bn; 649 int bsize, bscale; 650 struct partinfo dpart; 651 int n, on; 652 int error = 0; 653 654 #ifdef DIAGNOSTIC 655 if (uio->uio_rw != UIO_WRITE) 656 panic("spec_write mode"); 657 if (&uio->uio_vmspace->vm_map != kernel_map && 658 uio->uio_vmspace != curproc->p_vmspace) 659 panic("spec_write proc"); 660 #endif 661 662 switch (vp->v_type) { 663 664 case VCHR: 665 VOP_UNLOCK(vp); 666 error = cdev_write(vp->v_rdev, uio, ap->a_ioflag); 667 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 668 return (error); 669 670 case VBLK: 671 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 672 if (uio->uio_resid == 0) 673 return (0); 674 if (uio->uio_offset < 0) 675 return (EINVAL); 676 bsize = BLKDEV_IOSIZE; 677 if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) { 678 if (dpart.part->p_fstype == FS_BSDFFS && 679 dpart.part->p_frag != 0 && dpart.part->p_fsize != 0) 680 bsize = dpart.part->p_frag * 681 dpart.part->p_fsize; 682 } 683 bscale = bsize >> DEV_BSHIFT; 684 do { 685 bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1); 686 on = uio->uio_offset % bsize; 687 n = min((unsigned)(bsize - on), uio->uio_resid); 688 if (n == bsize) 689 bp = getblk(vp, bn, bsize, 0, 0); 690 else 691 error = bread(vp, bn, bsize, NOCRED, 692 B_MODIFY, &bp); 693 if (error) { 694 brelse(bp, 0); 695 return (error); 696 } 697 n = min(n, bsize - bp->b_resid); 698 error = uiomove((char *)bp->b_data + on, n, uio); 699 if (error) 700 brelse(bp, 0); 701 else { 702 if (n + on == bsize) 703 bawrite(bp); 704 else 705 bdwrite(bp); 706 error = bp->b_error; 707 } 708 } while (error == 0 && uio->uio_resid > 0 && n != 0); 709 return (error); 710 711 default: 712 panic("spec_write type"); 713 } 714 /* NOTREACHED */ 715 } 716 717 /* 718 * Device ioctl operation. 719 */ 720 /* ARGSUSED */ 721 int 722 spec_ioctl(void *v) 723 { 724 struct vop_ioctl_args /* { 725 struct vnode *a_vp; 726 u_long a_command; 727 void *a_data; 728 int a_fflag; 729 kauth_cred_t a_cred; 730 } */ *ap = v; 731 struct vnode *vp; 732 dev_t dev; 733 734 /* 735 * Extract all the info we need from the vnode, taking care to 736 * avoid a race with VOP_REVOKE(). 737 */ 738 739 vp = ap->a_vp; 740 dev = NODEV; 741 mutex_enter(&vp->v_interlock); 742 if ((vp->v_iflag & VI_XLOCK) == 0 && vp->v_specnode) { 743 dev = vp->v_rdev; 744 } 745 mutex_exit(&vp->v_interlock); 746 if (dev == NODEV) { 747 return ENXIO; 748 } 749 750 switch (vp->v_type) { 751 752 case VCHR: 753 return cdev_ioctl(dev, ap->a_command, ap->a_data, 754 ap->a_fflag, curlwp); 755 756 case VBLK: 757 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 758 return bdev_ioctl(dev, ap->a_command, ap->a_data, 759 ap->a_fflag, curlwp); 760 761 default: 762 panic("spec_ioctl"); 763 /* NOTREACHED */ 764 } 765 } 766 767 /* ARGSUSED */ 768 int 769 spec_poll(void *v) 770 { 771 struct vop_poll_args /* { 772 struct vnode *a_vp; 773 int a_events; 774 } */ *ap = v; 775 struct vnode *vp; 776 dev_t dev; 777 778 /* 779 * Extract all the info we need from the vnode, taking care to 780 * avoid a race with VOP_REVOKE(). 781 */ 782 783 vp = ap->a_vp; 784 dev = NODEV; 785 mutex_enter(&vp->v_interlock); 786 if ((vp->v_iflag & VI_XLOCK) == 0 && vp->v_specnode) { 787 dev = vp->v_rdev; 788 } 789 mutex_exit(&vp->v_interlock); 790 if (dev == NODEV) { 791 return POLLERR; 792 } 793 794 switch (vp->v_type) { 795 796 case VCHR: 797 return cdev_poll(dev, ap->a_events, curlwp); 798 799 default: 800 return (genfs_poll(v)); 801 } 802 } 803 804 /* ARGSUSED */ 805 int 806 spec_kqfilter(void *v) 807 { 808 struct vop_kqfilter_args /* { 809 struct vnode *a_vp; 810 struct proc *a_kn; 811 } */ *ap = v; 812 dev_t dev; 813 814 switch (ap->a_vp->v_type) { 815 816 case VCHR: 817 dev = ap->a_vp->v_rdev; 818 return cdev_kqfilter(dev, ap->a_kn); 819 default: 820 /* 821 * Block devices don't support kqfilter, and refuse it 822 * for any other files (like those vflush()ed) too. 823 */ 824 return (EOPNOTSUPP); 825 } 826 } 827 828 /* 829 * Allow mapping of only D_DISK. This is called only for VBLK. 830 */ 831 int 832 spec_mmap(void *v) 833 { 834 struct vop_mmap_args /* { 835 struct vnode *a_vp; 836 vm_prot_t a_prot; 837 kauth_cred_t a_cred; 838 } */ *ap = v; 839 struct vnode *vp = ap->a_vp; 840 841 KASSERT(vp->v_type == VBLK); 842 if (bdev_type(vp->v_rdev) != D_DISK) 843 return EINVAL; 844 845 return 0; 846 } 847 848 /* 849 * Synch buffers associated with a block device 850 */ 851 /* ARGSUSED */ 852 int 853 spec_fsync(void *v) 854 { 855 struct vop_fsync_args /* { 856 struct vnode *a_vp; 857 kauth_cred_t a_cred; 858 int a_flags; 859 off_t offlo; 860 off_t offhi; 861 } */ *ap = v; 862 struct vnode *vp = ap->a_vp; 863 struct mount *mp; 864 int error; 865 866 if (vp->v_type == VBLK) { 867 if ((mp = vp->v_specmountpoint) != NULL) { 868 error = VFS_FSYNC(mp, vp, ap->a_flags | FSYNC_VFS); 869 if (error != EOPNOTSUPP) 870 return error; 871 } 872 vflushbuf(vp, (ap->a_flags & FSYNC_WAIT) != 0); 873 } 874 return (0); 875 } 876 877 /* 878 * Just call the device strategy routine 879 */ 880 int 881 spec_strategy(void *v) 882 { 883 struct vop_strategy_args /* { 884 struct vnode *a_vp; 885 struct buf *a_bp; 886 } */ *ap = v; 887 struct vnode *vp = ap->a_vp; 888 struct buf *bp = ap->a_bp; 889 int error; 890 891 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 892 893 error = 0; 894 bp->b_dev = vp->v_rdev; 895 896 if (!(bp->b_flags & B_READ)) 897 error = fscow_run(bp, false); 898 899 if (error) { 900 bp->b_error = error; 901 biodone(bp); 902 return (error); 903 } 904 905 bdev_strategy(bp); 906 907 return (0); 908 } 909 910 int 911 spec_inactive(void *v) 912 { 913 struct vop_inactive_args /* { 914 struct vnode *a_vp; 915 struct proc *a_l; 916 } */ *ap = v; 917 918 VOP_UNLOCK(ap->a_vp); 919 return (0); 920 } 921 922 /* 923 * This is a noop, simply returning what one has been given. 924 */ 925 int 926 spec_bmap(void *v) 927 { 928 struct vop_bmap_args /* { 929 struct vnode *a_vp; 930 daddr_t a_bn; 931 struct vnode **a_vpp; 932 daddr_t *a_bnp; 933 int *a_runp; 934 } */ *ap = v; 935 936 if (ap->a_vpp != NULL) 937 *ap->a_vpp = ap->a_vp; 938 if (ap->a_bnp != NULL) 939 *ap->a_bnp = ap->a_bn; 940 if (ap->a_runp != NULL) 941 *ap->a_runp = (MAXBSIZE >> DEV_BSHIFT) - 1; 942 return (0); 943 } 944 945 /* 946 * Device close routine 947 */ 948 /* ARGSUSED */ 949 int 950 spec_close(void *v) 951 { 952 struct vop_close_args /* { 953 struct vnode *a_vp; 954 int a_fflag; 955 kauth_cred_t a_cred; 956 } */ *ap = v; 957 struct vnode *vp = ap->a_vp; 958 struct session *sess; 959 dev_t dev = vp->v_rdev; 960 int mode, error, flags, flags1, count; 961 specnode_t *sn; 962 specdev_t *sd; 963 964 flags = vp->v_iflag; 965 sn = vp->v_specnode; 966 sd = sn->sn_dev; 967 968 switch (vp->v_type) { 969 970 case VCHR: 971 /* 972 * Hack: a tty device that is a controlling terminal 973 * has a reference from the session structure. We 974 * cannot easily tell that a character device is a 975 * controlling terminal, unless it is the closing 976 * process' controlling terminal. In that case, if the 977 * open count is 1 release the reference from the 978 * session. Also, remove the link from the tty back to 979 * the session and pgrp. 980 * 981 * XXX V. fishy. 982 */ 983 mutex_enter(proc_lock); 984 sess = curlwp->l_proc->p_session; 985 if (sn->sn_opencnt == 1 && vp == sess->s_ttyvp) { 986 mutex_spin_enter(&tty_lock); 987 sess->s_ttyvp = NULL; 988 if (sess->s_ttyp->t_session != NULL) { 989 sess->s_ttyp->t_pgrp = NULL; 990 sess->s_ttyp->t_session = NULL; 991 mutex_spin_exit(&tty_lock); 992 /* Releases proc_lock. */ 993 proc_sessrele(sess); 994 } else { 995 mutex_spin_exit(&tty_lock); 996 if (sess->s_ttyp->t_pgrp != NULL) 997 panic("spec_close: spurious pgrp ref"); 998 mutex_exit(proc_lock); 999 } 1000 vrele(vp); 1001 } else 1002 mutex_exit(proc_lock); 1003 1004 /* 1005 * If the vnode is locked, then we are in the midst 1006 * of forcably closing the device, otherwise we only 1007 * close on last reference. 1008 */ 1009 mode = S_IFCHR; 1010 break; 1011 1012 case VBLK: 1013 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 1014 /* 1015 * On last close of a block device (that isn't mounted) 1016 * we must invalidate any in core blocks, so that 1017 * we can, for instance, change floppy disks. 1018 */ 1019 error = vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 0, 0); 1020 if (error) 1021 return (error); 1022 /* 1023 * We do not want to really close the device if it 1024 * is still in use unless we are trying to close it 1025 * forcibly. Since every use (buffer, vnode, swap, cmap) 1026 * holds a reference to the vnode, and because we mark 1027 * any other vnodes that alias this device, when the 1028 * sum of the reference counts on all the aliased 1029 * vnodes descends to one, we are on last close. 1030 */ 1031 mode = S_IFBLK; 1032 break; 1033 1034 default: 1035 panic("spec_close: not special"); 1036 } 1037 1038 mutex_enter(&device_lock); 1039 sn->sn_opencnt--; 1040 count = --sd->sd_opencnt; 1041 if (vp->v_type == VBLK) 1042 sd->sd_bdevvp = NULL; 1043 mutex_exit(&device_lock); 1044 1045 if (count != 0) 1046 return 0; 1047 1048 flags1 = ap->a_fflag; 1049 1050 /* 1051 * if VI_XLOCK is set, then we're going away soon, so make this 1052 * non-blocking. Also ensures that we won't wedge in vn_lock below. 1053 */ 1054 if (flags & VI_XLOCK) 1055 flags1 |= FNONBLOCK; 1056 1057 /* 1058 * If we're able to block, release the vnode lock & reacquire. We 1059 * might end up sleeping for someone else who wants our queues. They 1060 * won't get them if we hold the vnode locked. Also, if VI_XLOCK is 1061 * set, don't release the lock as we won't be able to regain it. 1062 */ 1063 if (!(flags1 & FNONBLOCK)) 1064 VOP_UNLOCK(vp); 1065 1066 if (vp->v_type == VBLK) 1067 error = bdev_close(dev, flags1, mode, curlwp); 1068 else 1069 error = cdev_close(dev, flags1, mode, curlwp); 1070 1071 if (!(flags1 & FNONBLOCK)) 1072 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1073 1074 return (error); 1075 } 1076 1077 /* 1078 * Print out the contents of a special device vnode. 1079 */ 1080 int 1081 spec_print(void *v) 1082 { 1083 struct vop_print_args /* { 1084 struct vnode *a_vp; 1085 } */ *ap = v; 1086 1087 printf("dev %llu, %llu\n", (unsigned long long)major(ap->a_vp->v_rdev), 1088 (unsigned long long)minor(ap->a_vp->v_rdev)); 1089 return 0; 1090 } 1091 1092 /* 1093 * Return POSIX pathconf information applicable to special devices. 1094 */ 1095 int 1096 spec_pathconf(void *v) 1097 { 1098 struct vop_pathconf_args /* { 1099 struct vnode *a_vp; 1100 int a_name; 1101 register_t *a_retval; 1102 } */ *ap = v; 1103 1104 switch (ap->a_name) { 1105 case _PC_LINK_MAX: 1106 *ap->a_retval = LINK_MAX; 1107 return (0); 1108 case _PC_MAX_CANON: 1109 *ap->a_retval = MAX_CANON; 1110 return (0); 1111 case _PC_MAX_INPUT: 1112 *ap->a_retval = MAX_INPUT; 1113 return (0); 1114 case _PC_PIPE_BUF: 1115 *ap->a_retval = PIPE_BUF; 1116 return (0); 1117 case _PC_CHOWN_RESTRICTED: 1118 *ap->a_retval = 1; 1119 return (0); 1120 case _PC_VDISABLE: 1121 *ap->a_retval = _POSIX_VDISABLE; 1122 return (0); 1123 case _PC_SYNC_IO: 1124 *ap->a_retval = 1; 1125 return (0); 1126 default: 1127 return (EINVAL); 1128 } 1129 /* NOTREACHED */ 1130 } 1131 1132 /* 1133 * Advisory record locking support. 1134 */ 1135 int 1136 spec_advlock(void *v) 1137 { 1138 struct vop_advlock_args /* { 1139 struct vnode *a_vp; 1140 void *a_id; 1141 int a_op; 1142 struct flock *a_fl; 1143 int a_flags; 1144 } */ *ap = v; 1145 struct vnode *vp = ap->a_vp; 1146 1147 return lf_advlock(ap, &vp->v_speclockf, (off_t)0); 1148 } 1149