1 /* $NetBSD: spec_vnops.c,v 1.137 2013/02/13 14:03:48 hannken 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 1.137 2013/02/13 14:03:48 hannken 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 #define SPECHSZ 64 97 #if ((SPECHSZ&(SPECHSZ-1)) == 0) 98 #define SPECHASH(rdev) (((rdev>>5)+(rdev))&(SPECHSZ-1)) 99 #else 100 #define SPECHASH(rdev) (((unsigned)((rdev>>5)+(rdev)))%SPECHSZ) 101 #endif 102 103 static vnode_t *specfs_hash[SPECHSZ]; 104 105 /* 106 * This vnode operations vector is used for special device nodes 107 * created from whole cloth by the kernel. For the ops vector for 108 * vnodes built from special devices found in a filesystem, see (e.g) 109 * ffs_specop_entries[] in ffs_vnops.c or the equivalent for other 110 * filesystems. 111 */ 112 113 int (**spec_vnodeop_p)(void *); 114 const struct vnodeopv_entry_desc spec_vnodeop_entries[] = { 115 { &vop_default_desc, vn_default_error }, 116 { &vop_lookup_desc, spec_lookup }, /* lookup */ 117 { &vop_create_desc, spec_create }, /* create */ 118 { &vop_mknod_desc, spec_mknod }, /* mknod */ 119 { &vop_open_desc, spec_open }, /* open */ 120 { &vop_close_desc, spec_close }, /* close */ 121 { &vop_access_desc, spec_access }, /* access */ 122 { &vop_getattr_desc, spec_getattr }, /* getattr */ 123 { &vop_setattr_desc, spec_setattr }, /* setattr */ 124 { &vop_read_desc, spec_read }, /* read */ 125 { &vop_write_desc, spec_write }, /* write */ 126 { &vop_fcntl_desc, spec_fcntl }, /* fcntl */ 127 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 128 { &vop_poll_desc, spec_poll }, /* poll */ 129 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 130 { &vop_revoke_desc, spec_revoke }, /* revoke */ 131 { &vop_mmap_desc, spec_mmap }, /* mmap */ 132 { &vop_fsync_desc, spec_fsync }, /* fsync */ 133 { &vop_seek_desc, spec_seek }, /* seek */ 134 { &vop_remove_desc, spec_remove }, /* remove */ 135 { &vop_link_desc, spec_link }, /* link */ 136 { &vop_rename_desc, spec_rename }, /* rename */ 137 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 138 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 139 { &vop_symlink_desc, spec_symlink }, /* symlink */ 140 { &vop_readdir_desc, spec_readdir }, /* readdir */ 141 { &vop_readlink_desc, spec_readlink }, /* readlink */ 142 { &vop_abortop_desc, spec_abortop }, /* abortop */ 143 { &vop_inactive_desc, spec_inactive }, /* inactive */ 144 { &vop_reclaim_desc, spec_reclaim }, /* reclaim */ 145 { &vop_lock_desc, spec_lock }, /* lock */ 146 { &vop_unlock_desc, spec_unlock }, /* unlock */ 147 { &vop_bmap_desc, spec_bmap }, /* bmap */ 148 { &vop_strategy_desc, spec_strategy }, /* strategy */ 149 { &vop_print_desc, spec_print }, /* print */ 150 { &vop_islocked_desc, spec_islocked }, /* islocked */ 151 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 152 { &vop_advlock_desc, spec_advlock }, /* advlock */ 153 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 154 { &vop_getpages_desc, spec_getpages }, /* getpages */ 155 { &vop_putpages_desc, spec_putpages }, /* putpages */ 156 { NULL, NULL } 157 }; 158 const struct vnodeopv_desc spec_vnodeop_opv_desc = 159 { &spec_vnodeop_p, spec_vnodeop_entries }; 160 161 static kauth_listener_t rawio_listener; 162 163 /* Returns true if vnode is /dev/mem or /dev/kmem. */ 164 bool 165 iskmemvp(struct vnode *vp) 166 { 167 return ((vp->v_type == VCHR) && iskmemdev(vp->v_rdev)); 168 } 169 170 /* 171 * Returns true if dev is /dev/mem or /dev/kmem. 172 */ 173 int 174 iskmemdev(dev_t dev) 175 { 176 /* mem_no is emitted by config(8) to generated devsw.c */ 177 extern const int mem_no; 178 179 /* minor 14 is /dev/io on i386 with COMPAT_10 */ 180 return (major(dev) == mem_no && (minor(dev) < 2 || minor(dev) == 14)); 181 } 182 183 static int 184 rawio_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 185 void *arg0, void *arg1, void *arg2, void *arg3) 186 { 187 int result; 188 189 result = KAUTH_RESULT_DEFER; 190 191 if ((action != KAUTH_DEVICE_RAWIO_SPEC) && 192 (action != KAUTH_DEVICE_RAWIO_PASSTHRU)) 193 return result; 194 195 /* Access is mandated by permissions. */ 196 result = KAUTH_RESULT_ALLOW; 197 198 return result; 199 } 200 201 void 202 spec_init(void) 203 { 204 205 rawio_listener = kauth_listen_scope(KAUTH_SCOPE_DEVICE, 206 rawio_listener_cb, NULL); 207 } 208 209 /* 210 * Initialize a vnode that represents a device. 211 */ 212 void 213 spec_node_init(vnode_t *vp, dev_t rdev) 214 { 215 specnode_t *sn; 216 specdev_t *sd; 217 vnode_t *vp2; 218 vnode_t **vpp; 219 220 KASSERT(vp->v_type == VBLK || vp->v_type == VCHR); 221 KASSERT(vp->v_specnode == NULL); 222 223 /* 224 * Search the hash table for this device. If known, add a 225 * reference to the device structure. If not known, create 226 * a new entry to represent the device. In all cases add 227 * the vnode to the hash table. 228 */ 229 sn = kmem_alloc(sizeof(*sn), KM_SLEEP); 230 if (sn == NULL) { 231 /* XXX */ 232 panic("spec_node_init: unable to allocate memory"); 233 } 234 sd = kmem_alloc(sizeof(*sd), KM_SLEEP); 235 if (sd == NULL) { 236 /* XXX */ 237 panic("spec_node_init: unable to allocate memory"); 238 } 239 mutex_enter(&device_lock); 240 vpp = &specfs_hash[SPECHASH(rdev)]; 241 for (vp2 = *vpp; vp2 != NULL; vp2 = vp2->v_specnext) { 242 KASSERT(vp2->v_specnode != NULL); 243 if (rdev == vp2->v_rdev && vp->v_type == vp2->v_type) { 244 break; 245 } 246 } 247 if (vp2 == NULL) { 248 /* No existing record, create a new one. */ 249 sd->sd_rdev = rdev; 250 sd->sd_mountpoint = NULL; 251 sd->sd_lockf = NULL; 252 sd->sd_refcnt = 1; 253 sd->sd_opencnt = 0; 254 sd->sd_bdevvp = NULL; 255 sn->sn_dev = sd; 256 sd = NULL; 257 } else { 258 /* Use the existing record. */ 259 sn->sn_dev = vp2->v_specnode->sn_dev; 260 sn->sn_dev->sd_refcnt++; 261 } 262 /* Insert vnode into the hash chain. */ 263 sn->sn_opencnt = 0; 264 sn->sn_rdev = rdev; 265 sn->sn_gone = false; 266 vp->v_specnode = sn; 267 vp->v_specnext = *vpp; 268 *vpp = vp; 269 mutex_exit(&device_lock); 270 271 /* Free the record we allocated if unused. */ 272 if (sd != NULL) { 273 kmem_free(sd, sizeof(*sd)); 274 } 275 } 276 277 /* 278 * Lookup a vnode by device number and return it referenced. 279 */ 280 int 281 spec_node_lookup_by_dev(enum vtype type, dev_t dev, vnode_t **vpp) 282 { 283 int error; 284 vnode_t *vp; 285 286 mutex_enter(&device_lock); 287 for (vp = specfs_hash[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 288 if (type == vp->v_type && dev == vp->v_rdev) { 289 mutex_enter(vp->v_interlock); 290 /* If clean or being cleaned, then ignore it. */ 291 if ((vp->v_iflag & (VI_CLEAN | VI_XLOCK)) == 0) 292 break; 293 mutex_exit(vp->v_interlock); 294 } 295 } 296 KASSERT(vp == NULL || mutex_owned(vp->v_interlock)); 297 if (vp == NULL) { 298 mutex_exit(&device_lock); 299 return ENOENT; 300 } 301 /* 302 * If it is an opened block device return the opened vnode. 303 */ 304 if (type == VBLK && vp->v_specnode->sn_dev->sd_bdevvp != NULL) { 305 mutex_exit(vp->v_interlock); 306 vp = vp->v_specnode->sn_dev->sd_bdevvp; 307 mutex_enter(vp->v_interlock); 308 } 309 mutex_exit(&device_lock); 310 error = vget(vp, 0); 311 if (error != 0) 312 return error; 313 *vpp = vp; 314 315 return 0; 316 } 317 318 /* 319 * Lookup a vnode by file system mounted on and return it referenced. 320 */ 321 int 322 spec_node_lookup_by_mount(struct mount *mp, vnode_t **vpp) 323 { 324 int i, error; 325 vnode_t *vp, *vq; 326 327 mutex_enter(&device_lock); 328 for (i = 0, vq = NULL; i < SPECHSZ && vq == NULL; i++) { 329 for (vp = specfs_hash[i]; vp; vp = vp->v_specnext) { 330 if (vp->v_type != VBLK) 331 continue; 332 vq = vp->v_specnode->sn_dev->sd_bdevvp; 333 if (vq != NULL && vq->v_specmountpoint == mp) 334 break; 335 vq = NULL; 336 } 337 } 338 if (vq == NULL) { 339 mutex_exit(&device_lock); 340 return ENOENT; 341 } 342 mutex_enter(vq->v_interlock); 343 mutex_exit(&device_lock); 344 error = vget(vq, 0); 345 if (error != 0) 346 return error; 347 *vpp = vq; 348 349 return 0; 350 351 } 352 353 /* 354 * A vnode representing a special device is going away. Close 355 * the device if the vnode holds it open. 356 */ 357 void 358 spec_node_revoke(vnode_t *vp) 359 { 360 specnode_t *sn; 361 specdev_t *sd; 362 363 sn = vp->v_specnode; 364 sd = sn->sn_dev; 365 366 KASSERT(vp->v_type == VBLK || vp->v_type == VCHR); 367 KASSERT(vp->v_specnode != NULL); 368 KASSERT((vp->v_iflag & VI_XLOCK) != 0); 369 KASSERT(sn->sn_gone == false); 370 371 mutex_enter(&device_lock); 372 KASSERT(sn->sn_opencnt <= sd->sd_opencnt); 373 if (sn->sn_opencnt != 0) { 374 sd->sd_opencnt -= (sn->sn_opencnt - 1); 375 sn->sn_opencnt = 1; 376 sn->sn_gone = true; 377 mutex_exit(&device_lock); 378 379 VOP_CLOSE(vp, FNONBLOCK, NOCRED); 380 381 mutex_enter(&device_lock); 382 KASSERT(sn->sn_opencnt == 0); 383 } 384 mutex_exit(&device_lock); 385 } 386 387 /* 388 * A vnode representing a special device is being recycled. 389 * Destroy the specfs component. 390 */ 391 void 392 spec_node_destroy(vnode_t *vp) 393 { 394 specnode_t *sn; 395 specdev_t *sd; 396 vnode_t **vpp, *vp2; 397 int refcnt; 398 399 sn = vp->v_specnode; 400 sd = sn->sn_dev; 401 402 KASSERT(vp->v_type == VBLK || vp->v_type == VCHR); 403 KASSERT(vp->v_specnode != NULL); 404 KASSERT(sn->sn_opencnt == 0); 405 406 mutex_enter(&device_lock); 407 /* Remove from the hash and destroy the node. */ 408 vpp = &specfs_hash[SPECHASH(vp->v_rdev)]; 409 for (vp2 = *vpp;; vp2 = vp2->v_specnext) { 410 if (vp2 == NULL) { 411 panic("spec_node_destroy: corrupt hash"); 412 } 413 if (vp2 == vp) { 414 KASSERT(vp == *vpp); 415 *vpp = vp->v_specnext; 416 break; 417 } 418 if (vp2->v_specnext == vp) { 419 vp2->v_specnext = vp->v_specnext; 420 break; 421 } 422 } 423 sn = vp->v_specnode; 424 vp->v_specnode = NULL; 425 refcnt = sd->sd_refcnt--; 426 KASSERT(refcnt > 0); 427 mutex_exit(&device_lock); 428 429 /* If the device is no longer in use, destroy our record. */ 430 if (refcnt == 1) { 431 KASSERT(sd->sd_opencnt == 0); 432 KASSERT(sd->sd_bdevvp == NULL); 433 kmem_free(sd, sizeof(*sd)); 434 } 435 kmem_free(sn, sizeof(*sn)); 436 } 437 438 /* 439 * Trivial lookup routine that always fails. 440 */ 441 int 442 spec_lookup(void *v) 443 { 444 struct vop_lookup_args /* { 445 struct vnode *a_dvp; 446 struct vnode **a_vpp; 447 struct componentname *a_cnp; 448 } */ *ap = v; 449 450 *ap->a_vpp = NULL; 451 return (ENOTDIR); 452 } 453 454 /* 455 * Open a special file. 456 */ 457 /* ARGSUSED */ 458 int 459 spec_open(void *v) 460 { 461 struct vop_open_args /* { 462 struct vnode *a_vp; 463 int a_mode; 464 kauth_cred_t a_cred; 465 } */ *ap = v; 466 struct lwp *l; 467 struct vnode *vp; 468 dev_t dev; 469 int error; 470 struct partinfo pi; 471 enum kauth_device_req req; 472 specnode_t *sn; 473 specdev_t *sd; 474 475 u_int gen; 476 const char *name; 477 478 l = curlwp; 479 vp = ap->a_vp; 480 dev = vp->v_rdev; 481 sn = vp->v_specnode; 482 sd = sn->sn_dev; 483 name = NULL; 484 gen = 0; 485 486 /* 487 * Don't allow open if fs is mounted -nodev. 488 */ 489 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV)) 490 return (ENXIO); 491 492 switch (ap->a_mode & (FREAD | FWRITE)) { 493 case FREAD | FWRITE: 494 req = KAUTH_REQ_DEVICE_RAWIO_SPEC_RW; 495 break; 496 case FWRITE: 497 req = KAUTH_REQ_DEVICE_RAWIO_SPEC_WRITE; 498 break; 499 default: 500 req = KAUTH_REQ_DEVICE_RAWIO_SPEC_READ; 501 break; 502 } 503 504 switch (vp->v_type) { 505 case VCHR: 506 error = kauth_authorize_device_spec(ap->a_cred, req, vp); 507 if (error != 0) 508 return (error); 509 510 /* 511 * Character devices can accept opens from multiple 512 * vnodes. 513 */ 514 mutex_enter(&device_lock); 515 if (sn->sn_gone) { 516 mutex_exit(&device_lock); 517 return (EBADF); 518 } 519 sd->sd_opencnt++; 520 sn->sn_opencnt++; 521 mutex_exit(&device_lock); 522 if (cdev_type(dev) == D_TTY) 523 vp->v_vflag |= VV_ISTTY; 524 VOP_UNLOCK(vp); 525 do { 526 const struct cdevsw *cdev; 527 528 gen = module_gen; 529 error = cdev_open(dev, ap->a_mode, S_IFCHR, l); 530 if (error != ENXIO) 531 break; 532 533 /* Check if we already have a valid driver */ 534 mutex_enter(&device_lock); 535 cdev = cdevsw_lookup(dev); 536 mutex_exit(&device_lock); 537 if (cdev != NULL) 538 break; 539 540 /* Get device name from devsw_conv array */ 541 if ((name = cdevsw_getname(major(dev))) == NULL) 542 break; 543 544 /* Try to autoload device module */ 545 (void) module_autoload(name, MODULE_CLASS_DRIVER); 546 } while (gen != module_gen); 547 548 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 549 break; 550 551 case VBLK: 552 error = kauth_authorize_device_spec(ap->a_cred, req, vp); 553 if (error != 0) 554 return (error); 555 556 /* 557 * For block devices, permit only one open. The buffer 558 * cache cannot remain self-consistent with multiple 559 * vnodes holding a block device open. 560 */ 561 mutex_enter(&device_lock); 562 if (sn->sn_gone) { 563 mutex_exit(&device_lock); 564 return (EBADF); 565 } 566 if (sd->sd_opencnt != 0) { 567 mutex_exit(&device_lock); 568 return EBUSY; 569 } 570 sn->sn_opencnt = 1; 571 sd->sd_opencnt = 1; 572 sd->sd_bdevvp = vp; 573 mutex_exit(&device_lock); 574 do { 575 const struct bdevsw *bdev; 576 577 gen = module_gen; 578 error = bdev_open(dev, ap->a_mode, S_IFBLK, l); 579 if (error != ENXIO) 580 break; 581 582 /* Check if we already have a valid driver */ 583 mutex_enter(&device_lock); 584 bdev = bdevsw_lookup(dev); 585 mutex_exit(&device_lock); 586 if (bdev != NULL) 587 break; 588 589 /* Get device name from devsw_conv array */ 590 if ((name = bdevsw_getname(major(dev))) == NULL) 591 break; 592 593 VOP_UNLOCK(vp); 594 595 /* Try to autoload device module */ 596 (void) module_autoload(name, MODULE_CLASS_DRIVER); 597 598 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 599 } while (gen != module_gen); 600 601 break; 602 603 case VNON: 604 case VLNK: 605 case VDIR: 606 case VREG: 607 case VBAD: 608 case VFIFO: 609 case VSOCK: 610 default: 611 return 0; 612 } 613 614 mutex_enter(&device_lock); 615 if (sn->sn_gone) { 616 if (error == 0) 617 error = EBADF; 618 } else if (error != 0) { 619 sd->sd_opencnt--; 620 sn->sn_opencnt--; 621 if (vp->v_type == VBLK) 622 sd->sd_bdevvp = NULL; 623 624 } 625 mutex_exit(&device_lock); 626 627 if (cdev_type(dev) != D_DISK || error != 0) 628 return error; 629 630 if (vp->v_type == VCHR) 631 error = cdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp); 632 else 633 error = bdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp); 634 if (error == 0) 635 uvm_vnp_setsize(vp, 636 (voff_t)pi.disklab->d_secsize * pi.part->p_size); 637 return 0; 638 } 639 640 /* 641 * Vnode op for read 642 */ 643 /* ARGSUSED */ 644 int 645 spec_read(void *v) 646 { 647 struct vop_read_args /* { 648 struct vnode *a_vp; 649 struct uio *a_uio; 650 int a_ioflag; 651 kauth_cred_t a_cred; 652 } */ *ap = v; 653 struct vnode *vp = ap->a_vp; 654 struct uio *uio = ap->a_uio; 655 struct lwp *l = curlwp; 656 struct buf *bp; 657 daddr_t bn; 658 int bsize, bscale; 659 struct partinfo dpart; 660 int n, on; 661 int error = 0; 662 663 #ifdef DIAGNOSTIC 664 if (uio->uio_rw != UIO_READ) 665 panic("spec_read mode"); 666 if (&uio->uio_vmspace->vm_map != kernel_map && 667 uio->uio_vmspace != curproc->p_vmspace) 668 panic("spec_read proc"); 669 #endif 670 if (uio->uio_resid == 0) 671 return (0); 672 673 switch (vp->v_type) { 674 675 case VCHR: 676 VOP_UNLOCK(vp); 677 error = cdev_read(vp->v_rdev, uio, ap->a_ioflag); 678 vn_lock(vp, LK_SHARED | LK_RETRY); 679 return (error); 680 681 case VBLK: 682 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 683 if (uio->uio_offset < 0) 684 return (EINVAL); 685 bsize = BLKDEV_IOSIZE; 686 if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) { 687 if (dpart.part->p_fstype == FS_BSDFFS && 688 dpart.part->p_frag != 0 && dpart.part->p_fsize != 0) 689 bsize = dpart.part->p_frag * 690 dpart.part->p_fsize; 691 } 692 bscale = bsize >> DEV_BSHIFT; 693 do { 694 bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1); 695 on = uio->uio_offset % bsize; 696 n = min((unsigned)(bsize - on), uio->uio_resid); 697 error = bread(vp, bn, bsize, NOCRED, 0, &bp); 698 if (error) { 699 return (error); 700 } 701 n = min(n, bsize - bp->b_resid); 702 error = uiomove((char *)bp->b_data + on, n, uio); 703 brelse(bp, 0); 704 } while (error == 0 && uio->uio_resid > 0 && n != 0); 705 return (error); 706 707 default: 708 panic("spec_read type"); 709 } 710 /* NOTREACHED */ 711 } 712 713 /* 714 * Vnode op for write 715 */ 716 /* ARGSUSED */ 717 int 718 spec_write(void *v) 719 { 720 struct vop_write_args /* { 721 struct vnode *a_vp; 722 struct uio *a_uio; 723 int a_ioflag; 724 kauth_cred_t a_cred; 725 } */ *ap = v; 726 struct vnode *vp = ap->a_vp; 727 struct uio *uio = ap->a_uio; 728 struct lwp *l = curlwp; 729 struct buf *bp; 730 daddr_t bn; 731 int bsize, bscale; 732 struct partinfo dpart; 733 int n, on; 734 int error = 0; 735 736 #ifdef DIAGNOSTIC 737 if (uio->uio_rw != UIO_WRITE) 738 panic("spec_write mode"); 739 if (&uio->uio_vmspace->vm_map != kernel_map && 740 uio->uio_vmspace != curproc->p_vmspace) 741 panic("spec_write proc"); 742 #endif 743 744 switch (vp->v_type) { 745 746 case VCHR: 747 VOP_UNLOCK(vp); 748 error = cdev_write(vp->v_rdev, uio, ap->a_ioflag); 749 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 750 return (error); 751 752 case VBLK: 753 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 754 if (uio->uio_resid == 0) 755 return (0); 756 if (uio->uio_offset < 0) 757 return (EINVAL); 758 bsize = BLKDEV_IOSIZE; 759 if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) { 760 if (dpart.part->p_fstype == FS_BSDFFS && 761 dpart.part->p_frag != 0 && dpart.part->p_fsize != 0) 762 bsize = dpart.part->p_frag * 763 dpart.part->p_fsize; 764 } 765 bscale = bsize >> DEV_BSHIFT; 766 do { 767 bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1); 768 on = uio->uio_offset % bsize; 769 n = min((unsigned)(bsize - on), uio->uio_resid); 770 if (n == bsize) 771 bp = getblk(vp, bn, bsize, 0, 0); 772 else 773 error = bread(vp, bn, bsize, NOCRED, 774 B_MODIFY, &bp); 775 if (error) { 776 return (error); 777 } 778 n = min(n, bsize - bp->b_resid); 779 error = uiomove((char *)bp->b_data + on, n, uio); 780 if (error) 781 brelse(bp, 0); 782 else { 783 if (n + on == bsize) 784 bawrite(bp); 785 else 786 bdwrite(bp); 787 error = bp->b_error; 788 } 789 } while (error == 0 && uio->uio_resid > 0 && n != 0); 790 return (error); 791 792 default: 793 panic("spec_write type"); 794 } 795 /* NOTREACHED */ 796 } 797 798 /* 799 * Device ioctl operation. 800 */ 801 /* ARGSUSED */ 802 int 803 spec_ioctl(void *v) 804 { 805 struct vop_ioctl_args /* { 806 struct vnode *a_vp; 807 u_long a_command; 808 void *a_data; 809 int a_fflag; 810 kauth_cred_t a_cred; 811 } */ *ap = v; 812 struct vnode *vp; 813 dev_t dev; 814 815 /* 816 * Extract all the info we need from the vnode, taking care to 817 * avoid a race with VOP_REVOKE(). 818 */ 819 820 vp = ap->a_vp; 821 dev = NODEV; 822 mutex_enter(vp->v_interlock); 823 if ((vp->v_iflag & VI_XLOCK) == 0 && vp->v_specnode) { 824 dev = vp->v_rdev; 825 } 826 mutex_exit(vp->v_interlock); 827 if (dev == NODEV) { 828 return ENXIO; 829 } 830 831 switch (vp->v_type) { 832 833 case VCHR: 834 return cdev_ioctl(dev, ap->a_command, ap->a_data, 835 ap->a_fflag, curlwp); 836 837 case VBLK: 838 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 839 return bdev_ioctl(dev, ap->a_command, ap->a_data, 840 ap->a_fflag, curlwp); 841 842 default: 843 panic("spec_ioctl"); 844 /* NOTREACHED */ 845 } 846 } 847 848 /* ARGSUSED */ 849 int 850 spec_poll(void *v) 851 { 852 struct vop_poll_args /* { 853 struct vnode *a_vp; 854 int a_events; 855 } */ *ap = v; 856 struct vnode *vp; 857 dev_t dev; 858 859 /* 860 * Extract all the info we need from the vnode, taking care to 861 * avoid a race with VOP_REVOKE(). 862 */ 863 864 vp = ap->a_vp; 865 dev = NODEV; 866 mutex_enter(vp->v_interlock); 867 if ((vp->v_iflag & VI_XLOCK) == 0 && vp->v_specnode) { 868 dev = vp->v_rdev; 869 } 870 mutex_exit(vp->v_interlock); 871 if (dev == NODEV) { 872 return POLLERR; 873 } 874 875 switch (vp->v_type) { 876 877 case VCHR: 878 return cdev_poll(dev, ap->a_events, curlwp); 879 880 default: 881 return (genfs_poll(v)); 882 } 883 } 884 885 /* ARGSUSED */ 886 int 887 spec_kqfilter(void *v) 888 { 889 struct vop_kqfilter_args /* { 890 struct vnode *a_vp; 891 struct proc *a_kn; 892 } */ *ap = v; 893 dev_t dev; 894 895 switch (ap->a_vp->v_type) { 896 897 case VCHR: 898 dev = ap->a_vp->v_rdev; 899 return cdev_kqfilter(dev, ap->a_kn); 900 default: 901 /* 902 * Block devices don't support kqfilter, and refuse it 903 * for any other files (like those vflush()ed) too. 904 */ 905 return (EOPNOTSUPP); 906 } 907 } 908 909 /* 910 * Allow mapping of only D_DISK. This is called only for VBLK. 911 */ 912 int 913 spec_mmap(void *v) 914 { 915 struct vop_mmap_args /* { 916 struct vnode *a_vp; 917 vm_prot_t a_prot; 918 kauth_cred_t a_cred; 919 } */ *ap = v; 920 struct vnode *vp = ap->a_vp; 921 922 KASSERT(vp->v_type == VBLK); 923 if (bdev_type(vp->v_rdev) != D_DISK) 924 return EINVAL; 925 926 return 0; 927 } 928 929 /* 930 * Synch buffers associated with a block device 931 */ 932 /* ARGSUSED */ 933 int 934 spec_fsync(void *v) 935 { 936 struct vop_fsync_args /* { 937 struct vnode *a_vp; 938 kauth_cred_t a_cred; 939 int a_flags; 940 off_t offlo; 941 off_t offhi; 942 } */ *ap = v; 943 struct vnode *vp = ap->a_vp; 944 struct mount *mp; 945 int error; 946 947 if (vp->v_type == VBLK) { 948 if ((mp = vp->v_specmountpoint) != NULL) { 949 error = VFS_FSYNC(mp, vp, ap->a_flags); 950 if (error != EOPNOTSUPP) 951 return error; 952 } 953 return vflushbuf(vp, ap->a_flags); 954 } 955 return (0); 956 } 957 958 /* 959 * Just call the device strategy routine 960 */ 961 int 962 spec_strategy(void *v) 963 { 964 struct vop_strategy_args /* { 965 struct vnode *a_vp; 966 struct buf *a_bp; 967 } */ *ap = v; 968 struct vnode *vp = ap->a_vp; 969 struct buf *bp = ap->a_bp; 970 int error; 971 972 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 973 974 error = 0; 975 bp->b_dev = vp->v_rdev; 976 977 if (!(bp->b_flags & B_READ)) 978 error = fscow_run(bp, false); 979 980 if (error) { 981 bp->b_error = error; 982 biodone(bp); 983 return (error); 984 } 985 986 bdev_strategy(bp); 987 988 return (0); 989 } 990 991 int 992 spec_inactive(void *v) 993 { 994 struct vop_inactive_args /* { 995 struct vnode *a_vp; 996 struct proc *a_l; 997 } */ *ap = v; 998 999 VOP_UNLOCK(ap->a_vp); 1000 return (0); 1001 } 1002 1003 /* 1004 * This is a noop, simply returning what one has been given. 1005 */ 1006 int 1007 spec_bmap(void *v) 1008 { 1009 struct vop_bmap_args /* { 1010 struct vnode *a_vp; 1011 daddr_t a_bn; 1012 struct vnode **a_vpp; 1013 daddr_t *a_bnp; 1014 int *a_runp; 1015 } */ *ap = v; 1016 1017 if (ap->a_vpp != NULL) 1018 *ap->a_vpp = ap->a_vp; 1019 if (ap->a_bnp != NULL) 1020 *ap->a_bnp = ap->a_bn; 1021 if (ap->a_runp != NULL) 1022 *ap->a_runp = (MAXBSIZE >> DEV_BSHIFT) - 1; 1023 return (0); 1024 } 1025 1026 /* 1027 * Device close routine 1028 */ 1029 /* ARGSUSED */ 1030 int 1031 spec_close(void *v) 1032 { 1033 struct vop_close_args /* { 1034 struct vnode *a_vp; 1035 int a_fflag; 1036 kauth_cred_t a_cred; 1037 } */ *ap = v; 1038 struct vnode *vp = ap->a_vp; 1039 struct session *sess; 1040 dev_t dev = vp->v_rdev; 1041 int mode, error, flags, flags1, count; 1042 specnode_t *sn; 1043 specdev_t *sd; 1044 1045 flags = vp->v_iflag; 1046 sn = vp->v_specnode; 1047 sd = sn->sn_dev; 1048 1049 switch (vp->v_type) { 1050 1051 case VCHR: 1052 /* 1053 * Hack: a tty device that is a controlling terminal 1054 * has a reference from the session structure. We 1055 * cannot easily tell that a character device is a 1056 * controlling terminal, unless it is the closing 1057 * process' controlling terminal. In that case, if the 1058 * open count is 1 release the reference from the 1059 * session. Also, remove the link from the tty back to 1060 * the session and pgrp. 1061 * 1062 * XXX V. fishy. 1063 */ 1064 mutex_enter(proc_lock); 1065 sess = curlwp->l_proc->p_session; 1066 if (sn->sn_opencnt == 1 && vp == sess->s_ttyvp) { 1067 mutex_spin_enter(&tty_lock); 1068 sess->s_ttyvp = NULL; 1069 if (sess->s_ttyp->t_session != NULL) { 1070 sess->s_ttyp->t_pgrp = NULL; 1071 sess->s_ttyp->t_session = NULL; 1072 mutex_spin_exit(&tty_lock); 1073 /* Releases proc_lock. */ 1074 proc_sessrele(sess); 1075 } else { 1076 mutex_spin_exit(&tty_lock); 1077 if (sess->s_ttyp->t_pgrp != NULL) 1078 panic("spec_close: spurious pgrp ref"); 1079 mutex_exit(proc_lock); 1080 } 1081 vrele(vp); 1082 } else 1083 mutex_exit(proc_lock); 1084 1085 /* 1086 * If the vnode is locked, then we are in the midst 1087 * of forcably closing the device, otherwise we only 1088 * close on last reference. 1089 */ 1090 mode = S_IFCHR; 1091 break; 1092 1093 case VBLK: 1094 KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp); 1095 /* 1096 * On last close of a block device (that isn't mounted) 1097 * we must invalidate any in core blocks, so that 1098 * we can, for instance, change floppy disks. 1099 */ 1100 error = vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 0, 0); 1101 if (error) 1102 return (error); 1103 /* 1104 * We do not want to really close the device if it 1105 * is still in use unless we are trying to close it 1106 * forcibly. Since every use (buffer, vnode, swap, cmap) 1107 * holds a reference to the vnode, and because we mark 1108 * any other vnodes that alias this device, when the 1109 * sum of the reference counts on all the aliased 1110 * vnodes descends to one, we are on last close. 1111 */ 1112 mode = S_IFBLK; 1113 break; 1114 1115 default: 1116 panic("spec_close: not special"); 1117 } 1118 1119 mutex_enter(&device_lock); 1120 sn->sn_opencnt--; 1121 count = --sd->sd_opencnt; 1122 if (vp->v_type == VBLK) 1123 sd->sd_bdevvp = NULL; 1124 mutex_exit(&device_lock); 1125 1126 if (count != 0) 1127 return 0; 1128 1129 flags1 = ap->a_fflag; 1130 1131 /* 1132 * if VI_XLOCK is set, then we're going away soon, so make this 1133 * non-blocking. Also ensures that we won't wedge in vn_lock below. 1134 */ 1135 if (flags & VI_XLOCK) 1136 flags1 |= FNONBLOCK; 1137 1138 /* 1139 * If we're able to block, release the vnode lock & reacquire. We 1140 * might end up sleeping for someone else who wants our queues. They 1141 * won't get them if we hold the vnode locked. Also, if VI_XLOCK is 1142 * set, don't release the lock as we won't be able to regain it. 1143 */ 1144 if (!(flags1 & FNONBLOCK)) 1145 VOP_UNLOCK(vp); 1146 1147 if (vp->v_type == VBLK) 1148 error = bdev_close(dev, flags1, mode, curlwp); 1149 else 1150 error = cdev_close(dev, flags1, mode, curlwp); 1151 1152 if (!(flags1 & FNONBLOCK)) 1153 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1154 1155 return (error); 1156 } 1157 1158 /* 1159 * Print out the contents of a special device vnode. 1160 */ 1161 int 1162 spec_print(void *v) 1163 { 1164 struct vop_print_args /* { 1165 struct vnode *a_vp; 1166 } */ *ap = v; 1167 1168 printf("dev %llu, %llu\n", (unsigned long long)major(ap->a_vp->v_rdev), 1169 (unsigned long long)minor(ap->a_vp->v_rdev)); 1170 return 0; 1171 } 1172 1173 /* 1174 * Return POSIX pathconf information applicable to special devices. 1175 */ 1176 int 1177 spec_pathconf(void *v) 1178 { 1179 struct vop_pathconf_args /* { 1180 struct vnode *a_vp; 1181 int a_name; 1182 register_t *a_retval; 1183 } */ *ap = v; 1184 1185 switch (ap->a_name) { 1186 case _PC_LINK_MAX: 1187 *ap->a_retval = LINK_MAX; 1188 return (0); 1189 case _PC_MAX_CANON: 1190 *ap->a_retval = MAX_CANON; 1191 return (0); 1192 case _PC_MAX_INPUT: 1193 *ap->a_retval = MAX_INPUT; 1194 return (0); 1195 case _PC_PIPE_BUF: 1196 *ap->a_retval = PIPE_BUF; 1197 return (0); 1198 case _PC_CHOWN_RESTRICTED: 1199 *ap->a_retval = 1; 1200 return (0); 1201 case _PC_VDISABLE: 1202 *ap->a_retval = _POSIX_VDISABLE; 1203 return (0); 1204 case _PC_SYNC_IO: 1205 *ap->a_retval = 1; 1206 return (0); 1207 default: 1208 return (EINVAL); 1209 } 1210 /* NOTREACHED */ 1211 } 1212 1213 /* 1214 * Advisory record locking support. 1215 */ 1216 int 1217 spec_advlock(void *v) 1218 { 1219 struct vop_advlock_args /* { 1220 struct vnode *a_vp; 1221 void *a_id; 1222 int a_op; 1223 struct flock *a_fl; 1224 int a_flags; 1225 } */ *ap = v; 1226 struct vnode *vp = ap->a_vp; 1227 1228 return lf_advlock(ap, &vp->v_speclockf, (off_t)0); 1229 } 1230