Home | History | Annotate | Line # | Download | only in specfs
spec_vnops.c revision 1.107
      1 /*	$NetBSD: spec_vnops.c,v 1.107 2007/10/08 18:04:05 ad Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1989, 1993
      5  *	The Regents of the University of California.  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  * 3. Neither the name of the University nor the names of its contributors
     16  *    may be used to endorse or promote products derived from this software
     17  *    without specific prior written permission.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     29  * SUCH DAMAGE.
     30  *
     31  *	@(#)spec_vnops.c	8.15 (Berkeley) 7/14/95
     32  */
     33 
     34 #include <sys/cdefs.h>
     35 __KERNEL_RCSID(0, "$NetBSD: spec_vnops.c,v 1.107 2007/10/08 18:04:05 ad Exp $");
     36 
     37 #include <sys/param.h>
     38 #include <sys/proc.h>
     39 #include <sys/systm.h>
     40 #include <sys/kernel.h>
     41 #include <sys/conf.h>
     42 #include <sys/buf.h>
     43 #include <sys/mount.h>
     44 #include <sys/namei.h>
     45 #include <sys/vnode.h>
     46 #include <sys/stat.h>
     47 #include <sys/errno.h>
     48 #include <sys/ioctl.h>
     49 #include <sys/poll.h>
     50 #include <sys/file.h>
     51 #include <sys/disklabel.h>
     52 #include <sys/lockf.h>
     53 #include <sys/tty.h>
     54 #include <sys/kauth.h>
     55 #include <sys/fstrans.h>
     56 
     57 #include <miscfs/genfs/genfs.h>
     58 #include <miscfs/specfs/specdev.h>
     59 
     60 /* symbolic sleep message strings for devices */
     61 const char	devopn[] = "devopn";
     62 const char	devio[] = "devio";
     63 const char	devwait[] = "devwait";
     64 const char	devin[] = "devin";
     65 const char	devout[] = "devout";
     66 const char	devioc[] = "devioc";
     67 const char	devcls[] = "devcls";
     68 
     69 struct vnode	*speclisth[SPECHSZ];
     70 
     71 /*
     72  * This vnode operations vector is used for two things only:
     73  * - special device nodes created from whole cloth by the kernel.
     74  * - as a temporary vnodeops replacement for vnodes which were found to
     75  *	be aliased by callers of checkalias().
     76  * For the ops vector for vnodes built from special devices found in a
     77  * filesystem, see (e.g) ffs_specop_entries[] in ffs_vnops.c or the
     78  * equivalent for other filesystems.
     79  */
     80 
     81 int (**spec_vnodeop_p)(void *);
     82 const struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
     83 	{ &vop_default_desc, vn_default_error },
     84 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
     85 	{ &vop_create_desc, spec_create },		/* create */
     86 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
     87 	{ &vop_open_desc, spec_open },			/* open */
     88 	{ &vop_close_desc, spec_close },		/* close */
     89 	{ &vop_access_desc, spec_access },		/* access */
     90 	{ &vop_getattr_desc, spec_getattr },		/* getattr */
     91 	{ &vop_setattr_desc, spec_setattr },		/* setattr */
     92 	{ &vop_read_desc, spec_read },			/* read */
     93 	{ &vop_write_desc, spec_write },		/* write */
     94 	{ &vop_lease_desc, spec_lease_check },		/* lease */
     95 	{ &vop_fcntl_desc, spec_fcntl },		/* fcntl */
     96 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
     97 	{ &vop_poll_desc, spec_poll },			/* poll */
     98 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
     99 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
    100 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
    101 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
    102 	{ &vop_seek_desc, spec_seek },			/* seek */
    103 	{ &vop_remove_desc, spec_remove },		/* remove */
    104 	{ &vop_link_desc, spec_link },			/* link */
    105 	{ &vop_rename_desc, spec_rename },		/* rename */
    106 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
    107 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
    108 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
    109 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
    110 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
    111 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
    112 	{ &vop_inactive_desc, spec_inactive },		/* inactive */
    113 	{ &vop_reclaim_desc, spec_reclaim },		/* reclaim */
    114 	{ &vop_lock_desc, spec_lock },			/* lock */
    115 	{ &vop_unlock_desc, spec_unlock },		/* unlock */
    116 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
    117 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
    118 	{ &vop_print_desc, spec_print },		/* print */
    119 	{ &vop_islocked_desc, spec_islocked },		/* islocked */
    120 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
    121 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
    122 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
    123 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
    124 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
    125 	{ NULL, NULL }
    126 };
    127 const struct vnodeopv_desc spec_vnodeop_opv_desc =
    128 	{ &spec_vnodeop_p, spec_vnodeop_entries };
    129 
    130 /*
    131  * Trivial lookup routine that always fails.
    132  */
    133 int
    134 spec_lookup(void *v)
    135 {
    136 	struct vop_lookup_args /* {
    137 		struct vnode *a_dvp;
    138 		struct vnode **a_vpp;
    139 		struct componentname *a_cnp;
    140 	} */ *ap = v;
    141 
    142 	*ap->a_vpp = NULL;
    143 	return (ENOTDIR);
    144 }
    145 
    146 /*
    147  * Returns true if dev is /dev/mem or /dev/kmem.
    148  */
    149 int
    150 iskmemdev(dev_t dev)
    151 {
    152 	/* mem_no is emitted by config(8) to generated devsw.c */
    153 	extern const int mem_no;
    154 
    155 	/* minor 14 is /dev/io on i386 with COMPAT_10 */
    156 	return (major(dev) == mem_no && (minor(dev) < 2 || minor(dev) == 14));
    157 }
    158 
    159 /*
    160  * Open a special file.
    161  */
    162 /* ARGSUSED */
    163 int
    164 spec_open(void *v)
    165 {
    166 	struct vop_open_args /* {
    167 		struct vnode *a_vp;
    168 		int  a_mode;
    169 		kauth_cred_t a_cred;
    170 		struct lwp *a_l;
    171 	} */ *ap = v;
    172 	struct lwp *l = ap->a_l;
    173 	struct vnode *vp = ap->a_vp;
    174 	dev_t dev = (dev_t)vp->v_rdev;
    175 	int error;
    176 	struct partinfo pi;
    177 	enum kauth_device_req req;
    178 
    179 	/*
    180 	 * Don't allow open if fs is mounted -nodev.
    181 	 */
    182 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
    183 		return (ENXIO);
    184 
    185 #define M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
    186 		 KAUTH_REQ_DEVICE_RAWIO_SPEC_RW : \
    187 		 (m) & FWRITE ? KAUTH_REQ_DEVICE_RAWIO_SPEC_WRITE : \
    188 		 KAUTH_REQ_DEVICE_RAWIO_SPEC_READ)
    189 
    190 	switch (vp->v_type) {
    191 
    192 	case VCHR:
    193 		req = M2K(ap->a_mode);
    194 		error = kauth_authorize_device_spec(ap->a_cred, req, vp);
    195 		if (error)
    196 			return (error);
    197 
    198 		if (cdev_type(dev) == D_TTY)
    199 			vp->v_flag |= VISTTY;
    200 		VOP_UNLOCK(vp, 0);
    201 		error = cdev_open(dev, ap->a_mode, S_IFCHR, l);
    202 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    203 		if (cdev_type(dev) != D_DISK)
    204 			return error;
    205 		break;
    206 
    207 	case VBLK:
    208 		req = M2K(ap->a_mode);
    209 		error = kauth_authorize_device_spec(ap->a_cred, req, vp);
    210 		if (error)
    211 			return (error);
    212 		error = bdev_open(dev, ap->a_mode, S_IFBLK, l);
    213 		break;
    214 
    215 	case VNON:
    216 	case VLNK:
    217 	case VDIR:
    218 	case VREG:
    219 	case VBAD:
    220 	case VFIFO:
    221 	case VSOCK:
    222 	default:
    223 		return 0;
    224 	}
    225 
    226 #undef M2K
    227 
    228 	if (error)
    229 		return error;
    230 	if (vp->v_type == VCHR)
    231 		error = cdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp);
    232 	else
    233 		error = bdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp);
    234 	if (error == 0)
    235 		uvm_vnp_setsize(vp,
    236 		    (voff_t)pi.disklab->d_secsize * pi.part->p_size);
    237 	return 0;
    238 }
    239 
    240 /*
    241  * Vnode op for read
    242  */
    243 /* ARGSUSED */
    244 int
    245 spec_read(void *v)
    246 {
    247 	struct vop_read_args /* {
    248 		struct vnode *a_vp;
    249 		struct uio *a_uio;
    250 		int  a_ioflag;
    251 		kauth_cred_t a_cred;
    252 	} */ *ap = v;
    253 	struct vnode *vp = ap->a_vp;
    254 	struct uio *uio = ap->a_uio;
    255  	struct lwp *l = curlwp;
    256 	struct buf *bp;
    257 	daddr_t bn;
    258 	int bsize, bscale;
    259 	struct partinfo dpart;
    260 	int n, on;
    261 	int error = 0;
    262 
    263 #ifdef DIAGNOSTIC
    264 	if (uio->uio_rw != UIO_READ)
    265 		panic("spec_read mode");
    266 	if (&uio->uio_vmspace->vm_map != kernel_map &&
    267 	    uio->uio_vmspace != curproc->p_vmspace)
    268 		panic("spec_read proc");
    269 #endif
    270 	if (uio->uio_resid == 0)
    271 		return (0);
    272 
    273 	switch (vp->v_type) {
    274 
    275 	case VCHR:
    276 		VOP_UNLOCK(vp, 0);
    277 		error = cdev_read(vp->v_rdev, uio, ap->a_ioflag);
    278 		vn_lock(vp, LK_SHARED | LK_RETRY);
    279 		return (error);
    280 
    281 	case VBLK:
    282 		if (uio->uio_offset < 0)
    283 			return (EINVAL);
    284 		bsize = BLKDEV_IOSIZE;
    285 		if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) {
    286 			if (dpart.part->p_fstype == FS_BSDFFS &&
    287 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
    288 				bsize = dpart.part->p_frag *
    289 				    dpart.part->p_fsize;
    290 		}
    291 		bscale = bsize >> DEV_BSHIFT;
    292 		do {
    293 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
    294 			on = uio->uio_offset % bsize;
    295 			n = min((unsigned)(bsize - on), uio->uio_resid);
    296 			error = bread(vp, bn, bsize, NOCRED, &bp);
    297 			n = min(n, bsize - bp->b_resid);
    298 			if (error) {
    299 				brelse(bp, 0);
    300 				return (error);
    301 			}
    302 			error = uiomove((char *)bp->b_data + on, n, uio);
    303 			brelse(bp, 0);
    304 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
    305 		return (error);
    306 
    307 	default:
    308 		panic("spec_read type");
    309 	}
    310 	/* NOTREACHED */
    311 }
    312 
    313 /*
    314  * Vnode op for write
    315  */
    316 /* ARGSUSED */
    317 int
    318 spec_write(void *v)
    319 {
    320 	struct vop_write_args /* {
    321 		struct vnode *a_vp;
    322 		struct uio *a_uio;
    323 		int  a_ioflag;
    324 		kauth_cred_t a_cred;
    325 	} */ *ap = v;
    326 	struct vnode *vp = ap->a_vp;
    327 	struct uio *uio = ap->a_uio;
    328 	struct lwp *l = curlwp;
    329 	struct buf *bp;
    330 	daddr_t bn;
    331 	int bsize, bscale;
    332 	struct partinfo dpart;
    333 	int n, on;
    334 	int error = 0;
    335 
    336 #ifdef DIAGNOSTIC
    337 	if (uio->uio_rw != UIO_WRITE)
    338 		panic("spec_write mode");
    339 	if (&uio->uio_vmspace->vm_map != kernel_map &&
    340 	    uio->uio_vmspace != curproc->p_vmspace)
    341 		panic("spec_write proc");
    342 #endif
    343 
    344 	switch (vp->v_type) {
    345 
    346 	case VCHR:
    347 		VOP_UNLOCK(vp, 0);
    348 		error = cdev_write(vp->v_rdev, uio, ap->a_ioflag);
    349 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    350 		return (error);
    351 
    352 	case VBLK:
    353 		if (uio->uio_resid == 0)
    354 			return (0);
    355 		if (uio->uio_offset < 0)
    356 			return (EINVAL);
    357 		bsize = BLKDEV_IOSIZE;
    358 		if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) {
    359 			if (dpart.part->p_fstype == FS_BSDFFS &&
    360 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
    361 				bsize = dpart.part->p_frag *
    362 				    dpart.part->p_fsize;
    363 		}
    364 		bscale = bsize >> DEV_BSHIFT;
    365 		do {
    366 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
    367 			on = uio->uio_offset % bsize;
    368 			n = min((unsigned)(bsize - on), uio->uio_resid);
    369 			if (n == bsize)
    370 				bp = getblk(vp, bn, bsize, 0, 0);
    371 			else
    372 				error = bread(vp, bn, bsize, NOCRED, &bp);
    373 			if (error) {
    374 				brelse(bp, 0);
    375 				return (error);
    376 			}
    377 			n = min(n, bsize - bp->b_resid);
    378 			error = uiomove((char *)bp->b_data + on, n, uio);
    379 			if (error)
    380 				brelse(bp, 0);
    381 			else {
    382 				if (n + on == bsize)
    383 					bawrite(bp);
    384 				else
    385 					bdwrite(bp);
    386 				error = bp->b_error;
    387 			}
    388 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
    389 		return (error);
    390 
    391 	default:
    392 		panic("spec_write type");
    393 	}
    394 	/* NOTREACHED */
    395 }
    396 
    397 /*
    398  * Device ioctl operation.
    399  */
    400 /* ARGSUSED */
    401 int
    402 spec_ioctl(void *v)
    403 {
    404 	struct vop_ioctl_args /* {
    405 		struct vnode *a_vp;
    406 		u_long a_command;
    407 		void  *a_data;
    408 		int  a_fflag;
    409 		kauth_cred_t a_cred;
    410 		struct lwp *a_l;
    411 	} */ *ap = v;
    412 	struct vnode *vp;
    413 	dev_t dev;
    414 
    415 	/*
    416 	 * Extract all the info we need from the vnode, taking care to
    417 	 * avoid a race with VOP_REVOKE().
    418 	 */
    419 
    420 	vp = ap->a_vp;
    421 	dev = NODEV;
    422 	simple_lock(&vp->v_interlock);
    423 	if ((vp->v_flag & VXLOCK) == 0 && vp->v_specinfo) {
    424 		dev = vp->v_rdev;
    425 	}
    426 	simple_unlock(&vp->v_interlock);
    427 	if (dev == NODEV) {
    428 		return ENXIO;
    429 	}
    430 
    431 	switch (vp->v_type) {
    432 
    433 	case VCHR:
    434 		return cdev_ioctl(dev, ap->a_command, ap->a_data,
    435 		    ap->a_fflag, ap->a_l);
    436 
    437 	case VBLK:
    438 		return bdev_ioctl(dev, ap->a_command, ap->a_data,
    439 		   ap->a_fflag, ap->a_l);
    440 
    441 	default:
    442 		panic("spec_ioctl");
    443 		/* NOTREACHED */
    444 	}
    445 }
    446 
    447 /* ARGSUSED */
    448 int
    449 spec_poll(void *v)
    450 {
    451 	struct vop_poll_args /* {
    452 		struct vnode *a_vp;
    453 		int a_events;
    454 		struct lwp *a_l;
    455 	} */ *ap = v;
    456 	struct vnode *vp;
    457 	dev_t dev;
    458 
    459 	/*
    460 	 * Extract all the info we need from the vnode, taking care to
    461 	 * avoid a race with VOP_REVOKE().
    462 	 */
    463 
    464 	vp = ap->a_vp;
    465 	dev = NODEV;
    466 	simple_lock(&vp->v_interlock);
    467 	if ((vp->v_flag & VXLOCK) == 0 && vp->v_specinfo) {
    468 		dev = vp->v_rdev;
    469 	}
    470 	simple_unlock(&vp->v_interlock);
    471 	if (dev == NODEV) {
    472 		return POLLERR;
    473 	}
    474 
    475 	switch (vp->v_type) {
    476 
    477 	case VCHR:
    478 		return cdev_poll(dev, ap->a_events, ap->a_l);
    479 
    480 	default:
    481 		return (genfs_poll(v));
    482 	}
    483 }
    484 
    485 /* ARGSUSED */
    486 int
    487 spec_kqfilter(void *v)
    488 {
    489 	struct vop_kqfilter_args /* {
    490 		struct vnode	*a_vp;
    491 		struct proc	*a_kn;
    492 	} */ *ap = v;
    493 	dev_t dev;
    494 
    495 	switch (ap->a_vp->v_type) {
    496 
    497 	case VCHR:
    498 		dev = ap->a_vp->v_rdev;
    499 		return cdev_kqfilter(dev, ap->a_kn);
    500 	default:
    501 		/*
    502 		 * Block devices don't support kqfilter, and refuse it
    503 		 * for any other files (like those vflush()ed) too.
    504 		 */
    505 		return (EOPNOTSUPP);
    506 	}
    507 }
    508 
    509 /*
    510  * Allow mapping of only D_DISK.  This is called only for VBLK.
    511  */
    512 int
    513 spec_mmap(void *v)
    514 {
    515 	struct vop_mmap_args /* {
    516 		struct vnode *a_vp;
    517 		vm_prot_t a_prot;
    518 		kauth_cred_t a_cred;
    519 		struct lwp *a_l;
    520 	} */ *ap = v;
    521 	struct vnode *vp = ap->a_vp;
    522 
    523 	KASSERT(vp->v_type == VBLK);
    524 	if (bdev_type(vp->v_rdev) != D_DISK)
    525 		return EINVAL;
    526 
    527 	return 0;
    528 }
    529 
    530 /*
    531  * Synch buffers associated with a block device
    532  */
    533 /* ARGSUSED */
    534 int
    535 spec_fsync(void *v)
    536 {
    537 	struct vop_fsync_args /* {
    538 		struct vnode *a_vp;
    539 		kauth_cred_t a_cred;
    540 		int  a_flags;
    541 		off_t offlo;
    542 		off_t offhi;
    543 		struct lwp *a_l;
    544 	} */ *ap = v;
    545 	struct vnode *vp = ap->a_vp;
    546 
    547 	if (vp->v_type == VBLK)
    548 		vflushbuf(vp, (ap->a_flags & FSYNC_WAIT) != 0);
    549 	return (0);
    550 }
    551 
    552 /*
    553  * Just call the device strategy routine
    554  */
    555 int
    556 spec_strategy(void *v)
    557 {
    558 	struct vop_strategy_args /* {
    559 		struct vnode *a_vp;
    560 		struct buf *a_bp;
    561 	} */ *ap = v;
    562 	struct vnode *vp = ap->a_vp;
    563 	struct buf *bp = ap->a_bp;
    564 	int error;
    565 
    566 	error = 0;
    567 	bp->b_dev = vp->v_rdev;
    568 	if (!(bp->b_flags & B_READ) &&
    569 	    (LIST_FIRST(&bp->b_dep)) != NULL && bioopsp)
    570 		bioopsp->io_start(bp);
    571 
    572 	if (!(bp->b_flags & B_READ))
    573 		error = fscow_run(bp);
    574 
    575 	if (error) {
    576 		bp->b_error = error;
    577 		biodone(bp);
    578 		return (error);
    579 	}
    580 
    581 	bdev_strategy(bp);
    582 
    583 	return (0);
    584 }
    585 
    586 int
    587 spec_inactive(void *v)
    588 {
    589 	struct vop_inactive_args /* {
    590 		struct vnode *a_vp;
    591 		struct proc *a_l;
    592 	} */ *ap = v;
    593 
    594 	VOP_UNLOCK(ap->a_vp, 0);
    595 	return (0);
    596 }
    597 
    598 /*
    599  * This is a noop, simply returning what one has been given.
    600  */
    601 int
    602 spec_bmap(void *v)
    603 {
    604 	struct vop_bmap_args /* {
    605 		struct vnode *a_vp;
    606 		daddr_t  a_bn;
    607 		struct vnode **a_vpp;
    608 		daddr_t *a_bnp;
    609 		int *a_runp;
    610 	} */ *ap = v;
    611 
    612 	if (ap->a_vpp != NULL)
    613 		*ap->a_vpp = ap->a_vp;
    614 	if (ap->a_bnp != NULL)
    615 		*ap->a_bnp = ap->a_bn;
    616 	if (ap->a_runp != NULL)
    617 		*ap->a_runp = (MAXBSIZE >> DEV_BSHIFT) - 1;
    618 	return (0);
    619 }
    620 
    621 /*
    622  * Device close routine
    623  */
    624 /* ARGSUSED */
    625 int
    626 spec_close(void *v)
    627 {
    628 	struct vop_close_args /* {
    629 		struct vnode *a_vp;
    630 		int  a_fflag;
    631 		kauth_cred_t a_cred;
    632 		struct lwp *a_l;
    633 	} */ *ap = v;
    634 	struct vnode *vp = ap->a_vp;
    635 	struct session *sess;
    636 	dev_t dev = vp->v_rdev;
    637 	int mode, error, count, flags, flags1;
    638 
    639 	count = vcount(vp);
    640 	flags = vp->v_flag;
    641 
    642 	switch (vp->v_type) {
    643 
    644 	case VCHR:
    645 		/*
    646 		 * Hack: a tty device that is a controlling terminal
    647 		 * has a reference from the session structure.
    648 		 * We cannot easily tell that a character device is
    649 		 * a controlling terminal, unless it is the closing
    650 		 * process' controlling terminal.  In that case,
    651 		 * if the reference count is 2 (this last descriptor
    652 		 * plus the session), release the reference from the session.
    653 		 * Also remove the link from the tty back to the session
    654 		 * and pgrp - due to the way consoles are handled we cannot
    655 		 * guarantee that the vrele() will do the final close on the
    656 		 * actual tty device.
    657 		 */
    658 		mutex_enter(&proclist_lock);
    659 		if (count == 2 && ap->a_l &&
    660 		    vp == (sess = ap->a_l->l_proc->p_session)->s_ttyvp) {
    661 			sess->s_ttyvp = NULL;
    662 			if (sess->s_ttyp->t_session != NULL) {
    663 				sess->s_ttyp->t_pgrp = NULL;
    664 				sess->s_ttyp->t_session = NULL;
    665 				mutex_exit(&proclist_lock);
    666 				SESSRELE(sess);
    667 			} else {
    668 				if (sess->s_ttyp->t_pgrp != NULL)
    669 					panic("spec_close: spurious pgrp ref");
    670 				mutex_exit(&proclist_lock);
    671 			}
    672 			vrele(vp);
    673 			count--;
    674 		} else
    675 			mutex_exit(&proclist_lock);
    676 
    677 		/*
    678 		 * If the vnode is locked, then we are in the midst
    679 		 * of forcably closing the device, otherwise we only
    680 		 * close on last reference.
    681 		 */
    682 		if (count > 1 && (flags & VXLOCK) == 0)
    683 			return (0);
    684 		mode = S_IFCHR;
    685 		break;
    686 
    687 	case VBLK:
    688 		/*
    689 		 * On last close of a block device (that isn't mounted)
    690 		 * we must invalidate any in core blocks, so that
    691 		 * we can, for instance, change floppy disks.
    692 		 */
    693 		error = vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_l, 0, 0);
    694 		if (error)
    695 			return (error);
    696 		/*
    697 		 * We do not want to really close the device if it
    698 		 * is still in use unless we are trying to close it
    699 		 * forcibly. Since every use (buffer, vnode, swap, cmap)
    700 		 * holds a reference to the vnode, and because we mark
    701 		 * any other vnodes that alias this device, when the
    702 		 * sum of the reference counts on all the aliased
    703 		 * vnodes descends to one, we are on last close.
    704 		 */
    705 		if (count > 1 && (flags & VXLOCK) == 0)
    706 			return (0);
    707 		mode = S_IFBLK;
    708 		break;
    709 
    710 	default:
    711 		panic("spec_close: not special");
    712 	}
    713 
    714 	flags1 = ap->a_fflag;
    715 
    716 	/*
    717 	 * if VXLOCK is set, then we're going away soon, so make this
    718 	 * non-blocking. Also ensures that we won't wedge in vn_lock below.
    719 	 */
    720 	if (flags & VXLOCK)
    721 		flags1 |= FNONBLOCK;
    722 
    723 	/*
    724 	 * If we're able to block, release the vnode lock & reacquire. We
    725 	 * might end up sleeping for someone else who wants our queues. They
    726 	 * won't get them if we hold the vnode locked. Also, if VXLOCK is
    727 	 * set, don't release the lock as we won't be able to regain it.
    728 	 */
    729 	if (!(flags1 & FNONBLOCK))
    730 		VOP_UNLOCK(vp, 0);
    731 
    732 	if (vp->v_type == VBLK)
    733 		error = bdev_close(dev, flags1, mode, ap->a_l);
    734 	else
    735 		error = cdev_close(dev, flags1, mode, ap->a_l);
    736 
    737 	if (!(flags1 & FNONBLOCK))
    738 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    739 
    740 	return (error);
    741 }
    742 
    743 /*
    744  * Print out the contents of a special device vnode.
    745  */
    746 int
    747 spec_print(void *v)
    748 {
    749 	struct vop_print_args /* {
    750 		struct vnode *a_vp;
    751 	} */ *ap = v;
    752 
    753 	printf("tag VT_NON, dev %d, %d\n", major(ap->a_vp->v_rdev),
    754 	    minor(ap->a_vp->v_rdev));
    755 	return 0;
    756 }
    757 
    758 /*
    759  * Return POSIX pathconf information applicable to special devices.
    760  */
    761 int
    762 spec_pathconf(void *v)
    763 {
    764 	struct vop_pathconf_args /* {
    765 		struct vnode *a_vp;
    766 		int a_name;
    767 		register_t *a_retval;
    768 	} */ *ap = v;
    769 
    770 	switch (ap->a_name) {
    771 	case _PC_LINK_MAX:
    772 		*ap->a_retval = LINK_MAX;
    773 		return (0);
    774 	case _PC_MAX_CANON:
    775 		*ap->a_retval = MAX_CANON;
    776 		return (0);
    777 	case _PC_MAX_INPUT:
    778 		*ap->a_retval = MAX_INPUT;
    779 		return (0);
    780 	case _PC_PIPE_BUF:
    781 		*ap->a_retval = PIPE_BUF;
    782 		return (0);
    783 	case _PC_CHOWN_RESTRICTED:
    784 		*ap->a_retval = 1;
    785 		return (0);
    786 	case _PC_VDISABLE:
    787 		*ap->a_retval = _POSIX_VDISABLE;
    788 		return (0);
    789 	case _PC_SYNC_IO:
    790 		*ap->a_retval = 1;
    791 		return (0);
    792 	default:
    793 		return (EINVAL);
    794 	}
    795 	/* NOTREACHED */
    796 }
    797 
    798 /*
    799  * Advisory record locking support.
    800  */
    801 int
    802 spec_advlock(void *v)
    803 {
    804 	struct vop_advlock_args /* {
    805 		struct vnode *a_vp;
    806 		void *a_id;
    807 		int a_op;
    808 		struct flock *a_fl;
    809 		int a_flags;
    810 	} */ *ap = v;
    811 	struct vnode *vp = ap->a_vp;
    812 
    813 	return lf_advlock(ap, &vp->v_speclockf, (off_t)0);
    814 }
    815