Home | History | Annotate | Line # | Download | only in specfs
spec_vnops.c revision 1.86
      1 /*	$NetBSD: spec_vnops.c,v 1.86 2006/03/01 12:38:32 yamt 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 1.86 2006/03/01 12:38:32 yamt 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 
     55 #include <miscfs/genfs/genfs.h>
     56 #include <miscfs/specfs/specdev.h>
     57 
     58 /* symbolic sleep message strings for devices */
     59 const char	devopn[] = "devopn";
     60 const char	devio[] = "devio";
     61 const char	devwait[] = "devwait";
     62 const char	devin[] = "devin";
     63 const char	devout[] = "devout";
     64 const char	devioc[] = "devioc";
     65 const char	devcls[] = "devcls";
     66 
     67 struct vnode	*speclisth[SPECHSZ];
     68 
     69 /*
     70  * This vnode operations vector is used for two things only:
     71  * - special device nodes created from whole cloth by the kernel.
     72  * - as a temporary vnodeops replacement for vnodes which were found to
     73  *	be aliased by callers of checkalias().
     74  * For the ops vector for vnodes built from special devices found in a
     75  * filesystem, see (e.g) ffs_specop_entries[] in ffs_vnops.c or the
     76  * equivalent for other filesystems.
     77  */
     78 
     79 int (**spec_vnodeop_p)(void *);
     80 const struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
     81 	{ &vop_default_desc, vn_default_error },
     82 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
     83 	{ &vop_create_desc, spec_create },		/* create */
     84 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
     85 	{ &vop_open_desc, spec_open },			/* open */
     86 	{ &vop_close_desc, spec_close },		/* close */
     87 	{ &vop_access_desc, spec_access },		/* access */
     88 	{ &vop_getattr_desc, spec_getattr },		/* getattr */
     89 	{ &vop_setattr_desc, spec_setattr },		/* setattr */
     90 	{ &vop_read_desc, spec_read },			/* read */
     91 	{ &vop_write_desc, spec_write },		/* write */
     92 	{ &vop_lease_desc, spec_lease_check },		/* lease */
     93 	{ &vop_fcntl_desc, spec_fcntl },		/* fcntl */
     94 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
     95 	{ &vop_poll_desc, spec_poll },			/* poll */
     96 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
     97 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
     98 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
     99 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
    100 	{ &vop_seek_desc, spec_seek },			/* seek */
    101 	{ &vop_remove_desc, spec_remove },		/* remove */
    102 	{ &vop_link_desc, spec_link },			/* link */
    103 	{ &vop_rename_desc, spec_rename },		/* rename */
    104 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
    105 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
    106 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
    107 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
    108 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
    109 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
    110 	{ &vop_inactive_desc, spec_inactive },		/* inactive */
    111 	{ &vop_reclaim_desc, spec_reclaim },		/* reclaim */
    112 	{ &vop_lock_desc, spec_lock },			/* lock */
    113 	{ &vop_unlock_desc, spec_unlock },		/* unlock */
    114 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
    115 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
    116 	{ &vop_print_desc, spec_print },		/* print */
    117 	{ &vop_islocked_desc, spec_islocked },		/* islocked */
    118 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
    119 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
    120 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
    121 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
    122 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
    123 	{ NULL, NULL }
    124 };
    125 const struct vnodeopv_desc spec_vnodeop_opv_desc =
    126 	{ &spec_vnodeop_p, spec_vnodeop_entries };
    127 
    128 /*
    129  * Trivial lookup routine that always fails.
    130  */
    131 int
    132 spec_lookup(v)
    133 	void *v;
    134 {
    135 	struct vop_lookup_args /* {
    136 		struct vnode *a_dvp;
    137 		struct vnode **a_vpp;
    138 		struct componentname *a_cnp;
    139 	} */ *ap = v;
    140 
    141 	*ap->a_vpp = NULL;
    142 	return (ENOTDIR);
    143 }
    144 
    145 /*
    146  * Returns true if dev is /dev/mem or /dev/kmem.
    147  */
    148 static int
    149 iskmemdev(dev_t dev)
    150 {
    151 	/* mem_no is emitted by config(8) to generated devsw.c */
    152 	extern const int mem_no;
    153 
    154 	/* minor 14 is /dev/io on i386 with COMPAT_10 */
    155 	return (major(dev) == mem_no && (minor(dev) < 2 || minor(dev) == 14));
    156 }
    157 
    158 /*
    159  * Open a special file.
    160  */
    161 /* ARGSUSED */
    162 int
    163 spec_open(v)
    164 	void *v;
    165 {
    166 	struct vop_open_args /* {
    167 		struct vnode *a_vp;
    168 		int  a_mode;
    169 		struct ucred *a_cred;
    170 		struct lwp *a_l;
    171 	} */ *ap = v;
    172 	struct lwp *l = ap->a_l;
    173 	struct vnode *bvp, *vp = ap->a_vp;
    174 	const struct bdevsw *bdev;
    175 	const struct cdevsw *cdev;
    176 	dev_t blkdev, dev = (dev_t)vp->v_rdev;
    177 	int error;
    178 	struct partinfo pi;
    179 	int (*d_ioctl)(dev_t, u_long, caddr_t, int, struct lwp *);
    180 
    181 	/*
    182 	 * Don't allow open if fs is mounted -nodev.
    183 	 */
    184 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
    185 		return (ENXIO);
    186 
    187 	switch (vp->v_type) {
    188 
    189 	case VCHR:
    190 		cdev = cdevsw_lookup(dev);
    191 		if (cdev == NULL)
    192 			return (ENXIO);
    193 		if (ap->a_cred != FSCRED && (ap->a_mode & FWRITE)) {
    194 			/*
    195 			 * When running in very secure mode, do not allow
    196 			 * opens for writing of any disk character devices.
    197 			 */
    198 			if (securelevel >= 2 && cdev->d_type == D_DISK)
    199 				return (EPERM);
    200 			/*
    201 			 * When running in secure mode, do not allow opens
    202 			 * for writing of /dev/mem, /dev/kmem, or character
    203 			 * devices whose corresponding block devices are
    204 			 * currently mounted.
    205 			 */
    206 			if (securelevel >= 1) {
    207 				blkdev = devsw_chr2blk(dev);
    208 				if (blkdev != (dev_t)NODEV &&
    209 				    vfinddev(blkdev, VBLK, &bvp) &&
    210 				    (error = vfs_mountedon(bvp)))
    211 					return (error);
    212 				if (iskmemdev(dev))
    213 					return (EPERM);
    214 			}
    215 		}
    216 		if (cdev->d_type == D_TTY)
    217 			vp->v_flag |= VISTTY;
    218 		VOP_UNLOCK(vp, 0);
    219 		error = (*cdev->d_open)(dev, ap->a_mode, S_IFCHR, l);
    220 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    221 		if (cdev->d_type != D_DISK)
    222 			return error;
    223 		d_ioctl = cdev->d_ioctl;
    224 		break;
    225 
    226 	case VBLK:
    227 		bdev = bdevsw_lookup(dev);
    228 		if (bdev == NULL)
    229 			return (ENXIO);
    230 		/*
    231 		 * When running in very secure mode, do not allow
    232 		 * opens for writing of any disk block devices.
    233 		 */
    234 		if (securelevel >= 2 && ap->a_cred != FSCRED &&
    235 		    (ap->a_mode & FWRITE) && bdev->d_type == D_DISK)
    236 			return (EPERM);
    237 		/*
    238 		 * Do not allow opens of block devices that are
    239 		 * currently mounted.
    240 		 */
    241 		if ((error = vfs_mountedon(vp)) != 0)
    242 			return (error);
    243 		error = (*bdev->d_open)(dev, ap->a_mode, S_IFBLK, l);
    244 		d_ioctl = bdev->d_ioctl;
    245 		break;
    246 
    247 	case VNON:
    248 	case VLNK:
    249 	case VDIR:
    250 	case VREG:
    251 	case VBAD:
    252 	case VFIFO:
    253 	case VSOCK:
    254 	default:
    255 		return 0;
    256 	}
    257 
    258 	if (error)
    259 		return error;
    260 	if (!(*d_ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&pi, FREAD, curlwp))
    261 		vp->v_size = (voff_t)pi.disklab->d_secsize * pi.part->p_size;
    262 	return 0;
    263 }
    264 
    265 /*
    266  * Vnode op for read
    267  */
    268 /* ARGSUSED */
    269 int
    270 spec_read(v)
    271 	void *v;
    272 {
    273 	struct vop_read_args /* {
    274 		struct vnode *a_vp;
    275 		struct uio *a_uio;
    276 		int  a_ioflag;
    277 		struct ucred *a_cred;
    278 	} */ *ap = v;
    279 	struct vnode *vp = ap->a_vp;
    280 	struct uio *uio = ap->a_uio;
    281  	struct lwp *l = curlwp;
    282 	struct buf *bp;
    283 	const struct bdevsw *bdev;
    284 	const struct cdevsw *cdev;
    285 	daddr_t bn;
    286 	int bsize, bscale;
    287 	struct partinfo dpart;
    288 	int n, on;
    289 	int error = 0;
    290 
    291 #ifdef DIAGNOSTIC
    292 	if (uio->uio_rw != UIO_READ)
    293 		panic("spec_read mode");
    294 	if (&uio->uio_vmspace->vm_map != kernel_map &&
    295 	    uio->uio_vmspace != curproc->p_vmspace)
    296 		panic("spec_read proc");
    297 #endif
    298 	if (uio->uio_resid == 0)
    299 		return (0);
    300 
    301 	switch (vp->v_type) {
    302 
    303 	case VCHR:
    304 		VOP_UNLOCK(vp, 0);
    305 		cdev = cdevsw_lookup(vp->v_rdev);
    306 		if (cdev != NULL)
    307 			error = (*cdev->d_read)(vp->v_rdev, uio, ap->a_ioflag);
    308 		else
    309 			error = ENXIO;
    310 		vn_lock(vp, LK_SHARED | LK_RETRY);
    311 		return (error);
    312 
    313 	case VBLK:
    314 		if (uio->uio_offset < 0)
    315 			return (EINVAL);
    316 		bsize = BLKDEV_IOSIZE;
    317 		bdev = bdevsw_lookup(vp->v_rdev);
    318 		if (bdev != NULL &&
    319 		    (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&dpart,
    320 				     FREAD, l) == 0) {
    321 			if (dpart.part->p_fstype == FS_BSDFFS &&
    322 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
    323 				bsize = dpart.part->p_frag *
    324 				    dpart.part->p_fsize;
    325 		}
    326 		bscale = bsize >> DEV_BSHIFT;
    327 		do {
    328 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
    329 			on = uio->uio_offset % bsize;
    330 			n = min((unsigned)(bsize - on), uio->uio_resid);
    331 			error = bread(vp, bn, bsize, NOCRED, &bp);
    332 			n = min(n, bsize - bp->b_resid);
    333 			if (error) {
    334 				brelse(bp);
    335 				return (error);
    336 			}
    337 			error = uiomove((char *)bp->b_data + on, n, uio);
    338 			brelse(bp);
    339 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
    340 		return (error);
    341 
    342 	default:
    343 		panic("spec_read type");
    344 	}
    345 	/* NOTREACHED */
    346 }
    347 
    348 /*
    349  * Vnode op for write
    350  */
    351 /* ARGSUSED */
    352 int
    353 spec_write(v)
    354 	void *v;
    355 {
    356 	struct vop_write_args /* {
    357 		struct vnode *a_vp;
    358 		struct uio *a_uio;
    359 		int  a_ioflag;
    360 		struct ucred *a_cred;
    361 	} */ *ap = v;
    362 	struct vnode *vp = ap->a_vp;
    363 	struct uio *uio = ap->a_uio;
    364 	struct lwp *l = curlwp;
    365 	struct buf *bp;
    366 	const struct bdevsw *bdev;
    367 	const struct cdevsw *cdev;
    368 	daddr_t bn;
    369 	int bsize, bscale;
    370 	struct partinfo dpart;
    371 	int n, on;
    372 	int error = 0;
    373 
    374 #ifdef DIAGNOSTIC
    375 	if (uio->uio_rw != UIO_WRITE)
    376 		panic("spec_write mode");
    377 	if (&uio->uio_vmspace->vm_map != kernel_map &&
    378 	    uio->uio_vmspace != curproc->p_vmspace)
    379 		panic("spec_write proc");
    380 #endif
    381 
    382 	switch (vp->v_type) {
    383 
    384 	case VCHR:
    385 		VOP_UNLOCK(vp, 0);
    386 		cdev = cdevsw_lookup(vp->v_rdev);
    387 		if (cdev != NULL)
    388 			error = (*cdev->d_write)(vp->v_rdev, uio, ap->a_ioflag);
    389 		else
    390 			error = ENXIO;
    391 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    392 		return (error);
    393 
    394 	case VBLK:
    395 		if (uio->uio_resid == 0)
    396 			return (0);
    397 		if (uio->uio_offset < 0)
    398 			return (EINVAL);
    399 		bsize = BLKDEV_IOSIZE;
    400 		bdev = bdevsw_lookup(vp->v_rdev);
    401 		if (bdev != NULL &&
    402 		    (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&dpart,
    403 				    FREAD, l) == 0) {
    404 			if (dpart.part->p_fstype == FS_BSDFFS &&
    405 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
    406 				bsize = dpart.part->p_frag *
    407 				    dpart.part->p_fsize;
    408 		}
    409 		bscale = bsize >> DEV_BSHIFT;
    410 		do {
    411 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
    412 			on = uio->uio_offset % bsize;
    413 			n = min((unsigned)(bsize - on), uio->uio_resid);
    414 			if (n == bsize)
    415 				bp = getblk(vp, bn, bsize, 0, 0);
    416 			else
    417 				error = bread(vp, bn, bsize, NOCRED, &bp);
    418 			if (error) {
    419 				brelse(bp);
    420 				return (error);
    421 			}
    422 			n = min(n, bsize - bp->b_resid);
    423 			error = uiomove((char *)bp->b_data + on, n, uio);
    424 			if (error)
    425 				brelse(bp);
    426 			else {
    427 				if (n + on == bsize)
    428 					bawrite(bp);
    429 				else
    430 					bdwrite(bp);
    431 				if (bp->b_flags & B_ERROR)
    432 					error = bp->b_error;
    433 			}
    434 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
    435 		return (error);
    436 
    437 	default:
    438 		panic("spec_write type");
    439 	}
    440 	/* NOTREACHED */
    441 }
    442 
    443 /*
    444  * Device ioctl operation.
    445  */
    446 /* ARGSUSED */
    447 int
    448 spec_ioctl(v)
    449 	void *v;
    450 {
    451 	struct vop_ioctl_args /* {
    452 		struct vnode *a_vp;
    453 		u_long a_command;
    454 		void  *a_data;
    455 		int  a_fflag;
    456 		struct ucred *a_cred;
    457 		struct lwp *a_l;
    458 	} */ *ap = v;
    459 	const struct bdevsw *bdev;
    460 	const struct cdevsw *cdev;
    461 	struct vnode *vp;
    462 	dev_t dev;
    463 
    464 	/*
    465 	 * Extract all the info we need from the vnode, taking care to
    466 	 * avoid a race with VOP_REVOKE().
    467 	 */
    468 
    469 	vp = ap->a_vp;
    470 	dev = NODEV;
    471 	simple_lock(&vp->v_interlock);
    472 	if ((vp->v_flag & VXLOCK) == 0 && vp->v_specinfo) {
    473 		dev = vp->v_rdev;
    474 	}
    475 	simple_unlock(&vp->v_interlock);
    476 	if (dev == NODEV) {
    477 		return ENXIO;
    478 	}
    479 
    480 	switch (vp->v_type) {
    481 
    482 	case VCHR:
    483 		cdev = cdevsw_lookup(dev);
    484 		if (cdev == NULL)
    485 			return (ENXIO);
    486 		return ((*cdev->d_ioctl)(dev, ap->a_command, ap->a_data,
    487 		    ap->a_fflag, ap->a_l));
    488 
    489 	case VBLK:
    490 		bdev = bdevsw_lookup(dev);
    491 		if (bdev == NULL)
    492 			return (ENXIO);
    493 		if (ap->a_command == 0 && (long)ap->a_data == B_TAPE) {
    494 			if (bdev->d_type == D_TAPE)
    495 				return (0);
    496 			else
    497 				return (1);
    498 		}
    499 		return ((*bdev->d_ioctl)(dev, ap->a_command, ap->a_data,
    500 		   ap->a_fflag, ap->a_l));
    501 
    502 	default:
    503 		panic("spec_ioctl");
    504 		/* NOTREACHED */
    505 	}
    506 }
    507 
    508 /* ARGSUSED */
    509 int
    510 spec_poll(v)
    511 	void *v;
    512 {
    513 	struct vop_poll_args /* {
    514 		struct vnode *a_vp;
    515 		int a_events;
    516 		struct lwp *a_l;
    517 	} */ *ap = v;
    518 	const struct cdevsw *cdev;
    519 	dev_t dev;
    520 
    521 	switch (ap->a_vp->v_type) {
    522 
    523 	case VCHR:
    524 		dev = ap->a_vp->v_rdev;
    525 		cdev = cdevsw_lookup(dev);
    526 		if (cdev == NULL)
    527 			return (POLLERR);
    528 		return (*cdev->d_poll)(dev, ap->a_events, ap->a_l);
    529 
    530 	default:
    531 		return (genfs_poll(v));
    532 	}
    533 }
    534 
    535 /* ARGSUSED */
    536 int
    537 spec_kqfilter(v)
    538 	void *v;
    539 {
    540 	struct vop_kqfilter_args /* {
    541 		struct vnode	*a_vp;
    542 		struct proc	*a_kn;
    543 	} */ *ap = v;
    544 	const struct cdevsw *cdev;
    545 	dev_t dev;
    546 
    547 	switch (ap->a_vp->v_type) {
    548 
    549 	case VCHR:
    550 		dev = ap->a_vp->v_rdev;
    551 		cdev = cdevsw_lookup(dev);
    552 		if (cdev == NULL)
    553 			return (ENXIO);
    554 		return (*cdev->d_kqfilter)(dev, ap->a_kn);
    555 	default:
    556 		/*
    557 		 * Block devices don't support kqfilter, and refuse it
    558 		 * for any other files (like those vflush()ed) too.
    559 		 */
    560 		return (EOPNOTSUPP);
    561 	}
    562 }
    563 
    564 /*
    565  * Synch buffers associated with a block device
    566  */
    567 /* ARGSUSED */
    568 int
    569 spec_fsync(v)
    570 	void *v;
    571 {
    572 	struct vop_fsync_args /* {
    573 		struct vnode *a_vp;
    574 		struct ucred *a_cred;
    575 		int  a_flags;
    576 		off_t offlo;
    577 		off_t offhi;
    578 		struct lwp *a_l;
    579 	} */ *ap = v;
    580 	struct vnode *vp = ap->a_vp;
    581 
    582 	if (vp->v_type == VBLK)
    583 		vflushbuf(vp, (ap->a_flags & FSYNC_WAIT) != 0);
    584 	return (0);
    585 }
    586 
    587 /*
    588  * Just call the device strategy routine
    589  */
    590 int
    591 spec_strategy(v)
    592 	void *v;
    593 {
    594 	struct vop_strategy_args /* {
    595 		struct vnode *a_vp;
    596 		struct buf *a_bp;
    597 	} */ *ap = v;
    598 	struct vnode *vp = ap->a_vp;
    599 	struct buf *bp = ap->a_bp;
    600 	int error, s;
    601 	struct spec_cow_entry *e;
    602 
    603 	error = 0;
    604 	bp->b_dev = vp->v_rdev;
    605 	if (!(bp->b_flags & B_READ) &&
    606 	    (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
    607 		(*bioops.io_start)(bp);
    608 
    609 	if (!(bp->b_flags & B_READ) && !SLIST_EMPTY(&vp->v_spec_cow_head)) {
    610 		SPEC_COW_LOCK(vp->v_specinfo, s);
    611 		while (vp->v_spec_cow_req > 0)
    612 			ltsleep(&vp->v_spec_cow_req, PRIBIO, "cowlist", 0,
    613 			    &vp->v_spec_cow_slock);
    614 		vp->v_spec_cow_count++;
    615 		SPEC_COW_UNLOCK(vp->v_specinfo, s);
    616 
    617 		SLIST_FOREACH(e, &vp->v_spec_cow_head, ce_list) {
    618 			if ((error = (*e->ce_func)(e->ce_cookie, bp)) != 0)
    619 				break;
    620 		}
    621 
    622 		SPEC_COW_LOCK(vp->v_specinfo, s);
    623 		vp->v_spec_cow_count--;
    624 		if (vp->v_spec_cow_req && vp->v_spec_cow_count == 0)
    625 			wakeup(&vp->v_spec_cow_req);
    626 		SPEC_COW_UNLOCK(vp->v_specinfo, s);
    627 	}
    628 
    629 	if (error) {
    630 		bp->b_error = error;
    631 		bp->b_flags |= B_ERROR;
    632 		biodone(bp);
    633 		return (error);
    634 	}
    635 
    636 	DEV_STRATEGY(bp);
    637 
    638 	return (0);
    639 }
    640 
    641 int
    642 spec_inactive(v)
    643 	void *v;
    644 {
    645 	struct vop_inactive_args /* {
    646 		struct vnode *a_vp;
    647 		struct proc *a_l;
    648 	} */ *ap = v;
    649 
    650 	VOP_UNLOCK(ap->a_vp, 0);
    651 	return (0);
    652 }
    653 
    654 /*
    655  * This is a noop, simply returning what one has been given.
    656  */
    657 int
    658 spec_bmap(v)
    659 	void *v;
    660 {
    661 	struct vop_bmap_args /* {
    662 		struct vnode *a_vp;
    663 		daddr_t  a_bn;
    664 		struct vnode **a_vpp;
    665 		daddr_t *a_bnp;
    666 		int *a_runp;
    667 	} */ *ap = v;
    668 
    669 	if (ap->a_vpp != NULL)
    670 		*ap->a_vpp = ap->a_vp;
    671 	if (ap->a_bnp != NULL)
    672 		*ap->a_bnp = ap->a_bn;
    673 	if (ap->a_runp != NULL)
    674 		*ap->a_runp = (MAXBSIZE >> DEV_BSHIFT) - 1;
    675 	return (0);
    676 }
    677 
    678 /*
    679  * Device close routine
    680  */
    681 /* ARGSUSED */
    682 int
    683 spec_close(v)
    684 	void *v;
    685 {
    686 	struct vop_close_args /* {
    687 		struct vnode *a_vp;
    688 		int  a_fflag;
    689 		struct ucred *a_cred;
    690 		struct lwp *a_l;
    691 	} */ *ap = v;
    692 	struct vnode *vp = ap->a_vp;
    693 	const struct bdevsw *bdev;
    694 	const struct cdevsw *cdev;
    695 	struct session *sess;
    696 	dev_t dev = vp->v_rdev;
    697 	int (*devclose)(dev_t, int, int, struct lwp *);
    698 	int mode, error, count, flags, flags1;
    699 
    700 	count = vcount(vp);
    701 	flags = vp->v_flag;
    702 
    703 	switch (vp->v_type) {
    704 
    705 	case VCHR:
    706 		/*
    707 		 * Hack: a tty device that is a controlling terminal
    708 		 * has a reference from the session structure.
    709 		 * We cannot easily tell that a character device is
    710 		 * a controlling terminal, unless it is the closing
    711 		 * process' controlling terminal.  In that case,
    712 		 * if the reference count is 2 (this last descriptor
    713 		 * plus the session), release the reference from the session.
    714 		 * Also remove the link from the tty back to the session
    715 		 * and pgrp - due to the way consoles are handled we cannot
    716 		 * guarantee that the vrele() will do the final close on the
    717 		 * actual tty device.
    718 		 */
    719 		if (count == 2 && ap->a_l &&
    720 		    vp == (sess = ap->a_l->l_proc->p_session)->s_ttyvp) {
    721 			sess->s_ttyvp = NULL;
    722 			if (sess->s_ttyp->t_session != NULL) {
    723 				sess->s_ttyp->t_pgrp = NULL;
    724 				sess->s_ttyp->t_session = NULL;
    725 				SESSRELE(sess);
    726 			} else if (sess->s_ttyp->t_pgrp != NULL)
    727 				panic("spec_close: spurious pgrp ref");
    728 			vrele(vp);
    729 			count--;
    730 		}
    731 		/*
    732 		 * If the vnode is locked, then we are in the midst
    733 		 * of forcably closing the device, otherwise we only
    734 		 * close on last reference.
    735 		 */
    736 		if (count > 1 && (flags & VXLOCK) == 0)
    737 			return (0);
    738 		cdev = cdevsw_lookup(dev);
    739 		if (cdev != NULL)
    740 			devclose = cdev->d_close;
    741 		else
    742 			devclose = NULL;
    743 		mode = S_IFCHR;
    744 		break;
    745 
    746 	case VBLK:
    747 		/*
    748 		 * On last close of a block device (that isn't mounted)
    749 		 * we must invalidate any in core blocks, so that
    750 		 * we can, for instance, change floppy disks.
    751 		 */
    752 		error = vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_l, 0, 0);
    753 		if (error)
    754 			return (error);
    755 		/*
    756 		 * We do not want to really close the device if it
    757 		 * is still in use unless we are trying to close it
    758 		 * forcibly. Since every use (buffer, vnode, swap, cmap)
    759 		 * holds a reference to the vnode, and because we mark
    760 		 * any other vnodes that alias this device, when the
    761 		 * sum of the reference counts on all the aliased
    762 		 * vnodes descends to one, we are on last close.
    763 		 */
    764 		if (count > 1 && (flags & VXLOCK) == 0)
    765 			return (0);
    766 		bdev = bdevsw_lookup(dev);
    767 		if (bdev != NULL)
    768 			devclose = bdev->d_close;
    769 		else
    770 			devclose = NULL;
    771 		mode = S_IFBLK;
    772 		break;
    773 
    774 	default:
    775 		panic("spec_close: not special");
    776 	}
    777 
    778 	flags1 = ap->a_fflag;
    779 
    780 	/*
    781 	 * if VXLOCK is set, then we're going away soon, so make this
    782 	 * non-blocking. Also ensures that we won't wedge in vn_lock below.
    783 	 */
    784 	if (flags & VXLOCK)
    785 		flags1 |= FNONBLOCK;
    786 
    787 	/*
    788 	 * If we're able to block, release the vnode lock & reacquire. We
    789 	 * might end up sleeping for someone else who wants our queues. They
    790 	 * won't get them if we hold the vnode locked. Also, if VXLOCK is set,
    791 	 * don't release the lock as we won't be able to regain it.
    792 	 */
    793 	if (!(flags1 & FNONBLOCK))
    794 		VOP_UNLOCK(vp, 0);
    795 
    796 	if (devclose != NULL)
    797 		error = (*devclose)(dev, flags1, mode, ap->a_l);
    798 	else
    799 		error = ENXIO;
    800 
    801 	if (!(flags1 & FNONBLOCK))
    802 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    803 
    804 	return (error);
    805 }
    806 
    807 /*
    808  * Print out the contents of a special device vnode.
    809  */
    810 int
    811 spec_print(v)
    812 	void *v;
    813 {
    814 	struct vop_print_args /* {
    815 		struct vnode *a_vp;
    816 	} */ *ap = v;
    817 
    818 	printf("tag VT_NON, dev %d, %d\n", major(ap->a_vp->v_rdev),
    819 	    minor(ap->a_vp->v_rdev));
    820 	return 0;
    821 }
    822 
    823 /*
    824  * Return POSIX pathconf information applicable to special devices.
    825  */
    826 int
    827 spec_pathconf(v)
    828 	void *v;
    829 {
    830 	struct vop_pathconf_args /* {
    831 		struct vnode *a_vp;
    832 		int a_name;
    833 		register_t *a_retval;
    834 	} */ *ap = v;
    835 
    836 	switch (ap->a_name) {
    837 	case _PC_LINK_MAX:
    838 		*ap->a_retval = LINK_MAX;
    839 		return (0);
    840 	case _PC_MAX_CANON:
    841 		*ap->a_retval = MAX_CANON;
    842 		return (0);
    843 	case _PC_MAX_INPUT:
    844 		*ap->a_retval = MAX_INPUT;
    845 		return (0);
    846 	case _PC_PIPE_BUF:
    847 		*ap->a_retval = PIPE_BUF;
    848 		return (0);
    849 	case _PC_CHOWN_RESTRICTED:
    850 		*ap->a_retval = 1;
    851 		return (0);
    852 	case _PC_VDISABLE:
    853 		*ap->a_retval = _POSIX_VDISABLE;
    854 		return (0);
    855 	case _PC_SYNC_IO:
    856 		*ap->a_retval = 1;
    857 		return (0);
    858 	default:
    859 		return (EINVAL);
    860 	}
    861 	/* NOTREACHED */
    862 }
    863 
    864 /*
    865  * Advisory record locking support.
    866  */
    867 int
    868 spec_advlock(v)
    869 	void *v;
    870 {
    871 	struct vop_advlock_args /* {
    872 		struct vnode *a_vp;
    873 		void *a_id;
    874 		int a_op;
    875 		struct flock *a_fl;
    876 		int a_flags;
    877 	} */ *ap = v;
    878 	struct vnode *vp = ap->a_vp;
    879 
    880 	return lf_advlock(ap, &vp->v_speclockf, (off_t)0);
    881 }
    882