Home | History | Annotate | Line # | Download | only in specfs
spec_vnops.c revision 1.85
      1 /*	$NetBSD: spec_vnops.c,v 1.85 2005/12/11 12:24:51 christos 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 1.85 2005/12/11 12:24:51 christos 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 = uio->uio_lwp;
    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_segflg == UIO_USERSPACE && uio->uio_lwp != curlwp)
    295 		panic("spec_read proc");
    296 #endif
    297 	if (uio->uio_resid == 0)
    298 		return (0);
    299 
    300 	switch (vp->v_type) {
    301 
    302 	case VCHR:
    303 		VOP_UNLOCK(vp, 0);
    304 		cdev = cdevsw_lookup(vp->v_rdev);
    305 		if (cdev != NULL)
    306 			error = (*cdev->d_read)(vp->v_rdev, uio, ap->a_ioflag);
    307 		else
    308 			error = ENXIO;
    309 		vn_lock(vp, LK_SHARED | LK_RETRY);
    310 		return (error);
    311 
    312 	case VBLK:
    313 		if (uio->uio_offset < 0)
    314 			return (EINVAL);
    315 		bsize = BLKDEV_IOSIZE;
    316 		bdev = bdevsw_lookup(vp->v_rdev);
    317 		if (bdev != NULL &&
    318 		    (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&dpart,
    319 				     FREAD, l) == 0) {
    320 			if (dpart.part->p_fstype == FS_BSDFFS &&
    321 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
    322 				bsize = dpart.part->p_frag *
    323 				    dpart.part->p_fsize;
    324 		}
    325 		bscale = bsize >> DEV_BSHIFT;
    326 		do {
    327 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
    328 			on = uio->uio_offset % bsize;
    329 			n = min((unsigned)(bsize - on), uio->uio_resid);
    330 			error = bread(vp, bn, bsize, NOCRED, &bp);
    331 			n = min(n, bsize - bp->b_resid);
    332 			if (error) {
    333 				brelse(bp);
    334 				return (error);
    335 			}
    336 			error = uiomove((char *)bp->b_data + on, n, uio);
    337 			brelse(bp);
    338 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
    339 		return (error);
    340 
    341 	default:
    342 		panic("spec_read type");
    343 	}
    344 	/* NOTREACHED */
    345 }
    346 
    347 /*
    348  * Vnode op for write
    349  */
    350 /* ARGSUSED */
    351 int
    352 spec_write(v)
    353 	void *v;
    354 {
    355 	struct vop_write_args /* {
    356 		struct vnode *a_vp;
    357 		struct uio *a_uio;
    358 		int  a_ioflag;
    359 		struct ucred *a_cred;
    360 	} */ *ap = v;
    361 	struct vnode *vp = ap->a_vp;
    362 	struct uio *uio = ap->a_uio;
    363 	struct lwp *l = uio->uio_lwp;
    364 	struct buf *bp;
    365 	const struct bdevsw *bdev;
    366 	const struct cdevsw *cdev;
    367 	daddr_t bn;
    368 	int bsize, bscale;
    369 	struct partinfo dpart;
    370 	int n, on;
    371 	int error = 0;
    372 
    373 #ifdef DIAGNOSTIC
    374 	if (uio->uio_rw != UIO_WRITE)
    375 		panic("spec_write mode");
    376 	if (uio->uio_segflg == UIO_USERSPACE && uio->uio_lwp != curlwp)
    377 		panic("spec_write proc");
    378 #endif
    379 
    380 	switch (vp->v_type) {
    381 
    382 	case VCHR:
    383 		VOP_UNLOCK(vp, 0);
    384 		cdev = cdevsw_lookup(vp->v_rdev);
    385 		if (cdev != NULL)
    386 			error = (*cdev->d_write)(vp->v_rdev, uio, ap->a_ioflag);
    387 		else
    388 			error = ENXIO;
    389 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    390 		return (error);
    391 
    392 	case VBLK:
    393 		if (uio->uio_resid == 0)
    394 			return (0);
    395 		if (uio->uio_offset < 0)
    396 			return (EINVAL);
    397 		bsize = BLKDEV_IOSIZE;
    398 		bdev = bdevsw_lookup(vp->v_rdev);
    399 		if (bdev != NULL &&
    400 		    (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&dpart,
    401 				    FREAD, l) == 0) {
    402 			if (dpart.part->p_fstype == FS_BSDFFS &&
    403 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
    404 				bsize = dpart.part->p_frag *
    405 				    dpart.part->p_fsize;
    406 		}
    407 		bscale = bsize >> DEV_BSHIFT;
    408 		do {
    409 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
    410 			on = uio->uio_offset % bsize;
    411 			n = min((unsigned)(bsize - on), uio->uio_resid);
    412 			if (n == bsize)
    413 				bp = getblk(vp, bn, bsize, 0, 0);
    414 			else
    415 				error = bread(vp, bn, bsize, NOCRED, &bp);
    416 			if (error) {
    417 				brelse(bp);
    418 				return (error);
    419 			}
    420 			n = min(n, bsize - bp->b_resid);
    421 			error = uiomove((char *)bp->b_data + on, n, uio);
    422 			if (error)
    423 				brelse(bp);
    424 			else {
    425 				if (n + on == bsize)
    426 					bawrite(bp);
    427 				else
    428 					bdwrite(bp);
    429 				if (bp->b_flags & B_ERROR)
    430 					error = bp->b_error;
    431 			}
    432 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
    433 		return (error);
    434 
    435 	default:
    436 		panic("spec_write type");
    437 	}
    438 	/* NOTREACHED */
    439 }
    440 
    441 /*
    442  * Device ioctl operation.
    443  */
    444 /* ARGSUSED */
    445 int
    446 spec_ioctl(v)
    447 	void *v;
    448 {
    449 	struct vop_ioctl_args /* {
    450 		struct vnode *a_vp;
    451 		u_long a_command;
    452 		void  *a_data;
    453 		int  a_fflag;
    454 		struct ucred *a_cred;
    455 		struct lwp *a_l;
    456 	} */ *ap = v;
    457 	const struct bdevsw *bdev;
    458 	const struct cdevsw *cdev;
    459 	struct vnode *vp;
    460 	dev_t dev;
    461 
    462 	/*
    463 	 * Extract all the info we need from the vnode, taking care to
    464 	 * avoid a race with VOP_REVOKE().
    465 	 */
    466 
    467 	vp = ap->a_vp;
    468 	dev = NODEV;
    469 	simple_lock(&vp->v_interlock);
    470 	if ((vp->v_flag & VXLOCK) == 0 && vp->v_specinfo) {
    471 		dev = vp->v_rdev;
    472 	}
    473 	simple_unlock(&vp->v_interlock);
    474 	if (dev == NODEV) {
    475 		return ENXIO;
    476 	}
    477 
    478 	switch (vp->v_type) {
    479 
    480 	case VCHR:
    481 		cdev = cdevsw_lookup(dev);
    482 		if (cdev == NULL)
    483 			return (ENXIO);
    484 		return ((*cdev->d_ioctl)(dev, ap->a_command, ap->a_data,
    485 		    ap->a_fflag, ap->a_l));
    486 
    487 	case VBLK:
    488 		bdev = bdevsw_lookup(dev);
    489 		if (bdev == NULL)
    490 			return (ENXIO);
    491 		if (ap->a_command == 0 && (long)ap->a_data == B_TAPE) {
    492 			if (bdev->d_type == D_TAPE)
    493 				return (0);
    494 			else
    495 				return (1);
    496 		}
    497 		return ((*bdev->d_ioctl)(dev, ap->a_command, ap->a_data,
    498 		   ap->a_fflag, ap->a_l));
    499 
    500 	default:
    501 		panic("spec_ioctl");
    502 		/* NOTREACHED */
    503 	}
    504 }
    505 
    506 /* ARGSUSED */
    507 int
    508 spec_poll(v)
    509 	void *v;
    510 {
    511 	struct vop_poll_args /* {
    512 		struct vnode *a_vp;
    513 		int a_events;
    514 		struct lwp *a_l;
    515 	} */ *ap = v;
    516 	const struct cdevsw *cdev;
    517 	dev_t dev;
    518 
    519 	switch (ap->a_vp->v_type) {
    520 
    521 	case VCHR:
    522 		dev = ap->a_vp->v_rdev;
    523 		cdev = cdevsw_lookup(dev);
    524 		if (cdev == NULL)
    525 			return (POLLERR);
    526 		return (*cdev->d_poll)(dev, ap->a_events, ap->a_l);
    527 
    528 	default:
    529 		return (genfs_poll(v));
    530 	}
    531 }
    532 
    533 /* ARGSUSED */
    534 int
    535 spec_kqfilter(v)
    536 	void *v;
    537 {
    538 	struct vop_kqfilter_args /* {
    539 		struct vnode	*a_vp;
    540 		struct proc	*a_kn;
    541 	} */ *ap = v;
    542 	const struct cdevsw *cdev;
    543 	dev_t dev;
    544 
    545 	switch (ap->a_vp->v_type) {
    546 
    547 	case VCHR:
    548 		dev = ap->a_vp->v_rdev;
    549 		cdev = cdevsw_lookup(dev);
    550 		if (cdev == NULL)
    551 			return (ENXIO);
    552 		return (*cdev->d_kqfilter)(dev, ap->a_kn);
    553 	default:
    554 		/*
    555 		 * Block devices don't support kqfilter, and refuse it
    556 		 * for any other files (like those vflush()ed) too.
    557 		 */
    558 		return (EOPNOTSUPP);
    559 	}
    560 }
    561 
    562 /*
    563  * Synch buffers associated with a block device
    564  */
    565 /* ARGSUSED */
    566 int
    567 spec_fsync(v)
    568 	void *v;
    569 {
    570 	struct vop_fsync_args /* {
    571 		struct vnode *a_vp;
    572 		struct ucred *a_cred;
    573 		int  a_flags;
    574 		off_t offlo;
    575 		off_t offhi;
    576 		struct lwp *a_l;
    577 	} */ *ap = v;
    578 	struct vnode *vp = ap->a_vp;
    579 
    580 	if (vp->v_type == VBLK)
    581 		vflushbuf(vp, (ap->a_flags & FSYNC_WAIT) != 0);
    582 	return (0);
    583 }
    584 
    585 /*
    586  * Just call the device strategy routine
    587  */
    588 int
    589 spec_strategy(v)
    590 	void *v;
    591 {
    592 	struct vop_strategy_args /* {
    593 		struct vnode *a_vp;
    594 		struct buf *a_bp;
    595 	} */ *ap = v;
    596 	struct vnode *vp = ap->a_vp;
    597 	struct buf *bp = ap->a_bp;
    598 	int error, s;
    599 	struct spec_cow_entry *e;
    600 
    601 	error = 0;
    602 	bp->b_dev = vp->v_rdev;
    603 	if (!(bp->b_flags & B_READ) &&
    604 	    (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
    605 		(*bioops.io_start)(bp);
    606 
    607 	if (!(bp->b_flags & B_READ) && !SLIST_EMPTY(&vp->v_spec_cow_head)) {
    608 		SPEC_COW_LOCK(vp->v_specinfo, s);
    609 		while (vp->v_spec_cow_req > 0)
    610 			ltsleep(&vp->v_spec_cow_req, PRIBIO, "cowlist", 0,
    611 			    &vp->v_spec_cow_slock);
    612 		vp->v_spec_cow_count++;
    613 		SPEC_COW_UNLOCK(vp->v_specinfo, s);
    614 
    615 		SLIST_FOREACH(e, &vp->v_spec_cow_head, ce_list) {
    616 			if ((error = (*e->ce_func)(e->ce_cookie, bp)) != 0)
    617 				break;
    618 		}
    619 
    620 		SPEC_COW_LOCK(vp->v_specinfo, s);
    621 		vp->v_spec_cow_count--;
    622 		if (vp->v_spec_cow_req && vp->v_spec_cow_count == 0)
    623 			wakeup(&vp->v_spec_cow_req);
    624 		SPEC_COW_UNLOCK(vp->v_specinfo, s);
    625 	}
    626 
    627 	if (error) {
    628 		bp->b_error = error;
    629 		bp->b_flags |= B_ERROR;
    630 		biodone(bp);
    631 		return (error);
    632 	}
    633 
    634 	DEV_STRATEGY(bp);
    635 
    636 	return (0);
    637 }
    638 
    639 int
    640 spec_inactive(v)
    641 	void *v;
    642 {
    643 	struct vop_inactive_args /* {
    644 		struct vnode *a_vp;
    645 		struct proc *a_l;
    646 	} */ *ap = v;
    647 
    648 	VOP_UNLOCK(ap->a_vp, 0);
    649 	return (0);
    650 }
    651 
    652 /*
    653  * This is a noop, simply returning what one has been given.
    654  */
    655 int
    656 spec_bmap(v)
    657 	void *v;
    658 {
    659 	struct vop_bmap_args /* {
    660 		struct vnode *a_vp;
    661 		daddr_t  a_bn;
    662 		struct vnode **a_vpp;
    663 		daddr_t *a_bnp;
    664 		int *a_runp;
    665 	} */ *ap = v;
    666 
    667 	if (ap->a_vpp != NULL)
    668 		*ap->a_vpp = ap->a_vp;
    669 	if (ap->a_bnp != NULL)
    670 		*ap->a_bnp = ap->a_bn;
    671 	if (ap->a_runp != NULL)
    672 		*ap->a_runp = (MAXBSIZE >> DEV_BSHIFT) - 1;
    673 	return (0);
    674 }
    675 
    676 /*
    677  * Device close routine
    678  */
    679 /* ARGSUSED */
    680 int
    681 spec_close(v)
    682 	void *v;
    683 {
    684 	struct vop_close_args /* {
    685 		struct vnode *a_vp;
    686 		int  a_fflag;
    687 		struct ucred *a_cred;
    688 		struct lwp *a_l;
    689 	} */ *ap = v;
    690 	struct vnode *vp = ap->a_vp;
    691 	const struct bdevsw *bdev;
    692 	const struct cdevsw *cdev;
    693 	struct session *sess;
    694 	dev_t dev = vp->v_rdev;
    695 	int (*devclose)(dev_t, int, int, struct lwp *);
    696 	int mode, error, count, flags, flags1;
    697 
    698 	count = vcount(vp);
    699 	flags = vp->v_flag;
    700 
    701 	switch (vp->v_type) {
    702 
    703 	case VCHR:
    704 		/*
    705 		 * Hack: a tty device that is a controlling terminal
    706 		 * has a reference from the session structure.
    707 		 * We cannot easily tell that a character device is
    708 		 * a controlling terminal, unless it is the closing
    709 		 * process' controlling terminal.  In that case,
    710 		 * if the reference count is 2 (this last descriptor
    711 		 * plus the session), release the reference from the session.
    712 		 * Also remove the link from the tty back to the session
    713 		 * and pgrp - due to the way consoles are handled we cannot
    714 		 * guarantee that the vrele() will do the final close on the
    715 		 * actual tty device.
    716 		 */
    717 		if (count == 2 && ap->a_l &&
    718 		    vp == (sess = ap->a_l->l_proc->p_session)->s_ttyvp) {
    719 			sess->s_ttyvp = NULL;
    720 			if (sess->s_ttyp->t_session != NULL) {
    721 				sess->s_ttyp->t_pgrp = NULL;
    722 				sess->s_ttyp->t_session = NULL;
    723 				SESSRELE(sess);
    724 			} else if (sess->s_ttyp->t_pgrp != NULL)
    725 				panic("spec_close: spurious pgrp ref");
    726 			vrele(vp);
    727 			count--;
    728 		}
    729 		/*
    730 		 * If the vnode is locked, then we are in the midst
    731 		 * of forcably closing the device, otherwise we only
    732 		 * close on last reference.
    733 		 */
    734 		if (count > 1 && (flags & VXLOCK) == 0)
    735 			return (0);
    736 		cdev = cdevsw_lookup(dev);
    737 		if (cdev != NULL)
    738 			devclose = cdev->d_close;
    739 		else
    740 			devclose = NULL;
    741 		mode = S_IFCHR;
    742 		break;
    743 
    744 	case VBLK:
    745 		/*
    746 		 * On last close of a block device (that isn't mounted)
    747 		 * we must invalidate any in core blocks, so that
    748 		 * we can, for instance, change floppy disks.
    749 		 */
    750 		error = vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_l, 0, 0);
    751 		if (error)
    752 			return (error);
    753 		/*
    754 		 * We do not want to really close the device if it
    755 		 * is still in use unless we are trying to close it
    756 		 * forcibly. Since every use (buffer, vnode, swap, cmap)
    757 		 * holds a reference to the vnode, and because we mark
    758 		 * any other vnodes that alias this device, when the
    759 		 * sum of the reference counts on all the aliased
    760 		 * vnodes descends to one, we are on last close.
    761 		 */
    762 		if (count > 1 && (flags & VXLOCK) == 0)
    763 			return (0);
    764 		bdev = bdevsw_lookup(dev);
    765 		if (bdev != NULL)
    766 			devclose = bdev->d_close;
    767 		else
    768 			devclose = NULL;
    769 		mode = S_IFBLK;
    770 		break;
    771 
    772 	default:
    773 		panic("spec_close: not special");
    774 	}
    775 
    776 	flags1 = ap->a_fflag;
    777 
    778 	/*
    779 	 * if VXLOCK is set, then we're going away soon, so make this
    780 	 * non-blocking. Also ensures that we won't wedge in vn_lock below.
    781 	 */
    782 	if (flags & VXLOCK)
    783 		flags1 |= FNONBLOCK;
    784 
    785 	/*
    786 	 * If we're able to block, release the vnode lock & reacquire. We
    787 	 * might end up sleeping for someone else who wants our queues. They
    788 	 * won't get them if we hold the vnode locked. Also, if VXLOCK is set,
    789 	 * don't release the lock as we won't be able to regain it.
    790 	 */
    791 	if (!(flags1 & FNONBLOCK))
    792 		VOP_UNLOCK(vp, 0);
    793 
    794 	if (devclose != NULL)
    795 		error = (*devclose)(dev, flags1, mode, ap->a_l);
    796 	else
    797 		error = ENXIO;
    798 
    799 	if (!(flags1 & FNONBLOCK))
    800 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    801 
    802 	return (error);
    803 }
    804 
    805 /*
    806  * Print out the contents of a special device vnode.
    807  */
    808 int
    809 spec_print(v)
    810 	void *v;
    811 {
    812 	struct vop_print_args /* {
    813 		struct vnode *a_vp;
    814 	} */ *ap = v;
    815 
    816 	printf("tag VT_NON, dev %d, %d\n", major(ap->a_vp->v_rdev),
    817 	    minor(ap->a_vp->v_rdev));
    818 	return 0;
    819 }
    820 
    821 /*
    822  * Return POSIX pathconf information applicable to special devices.
    823  */
    824 int
    825 spec_pathconf(v)
    826 	void *v;
    827 {
    828 	struct vop_pathconf_args /* {
    829 		struct vnode *a_vp;
    830 		int a_name;
    831 		register_t *a_retval;
    832 	} */ *ap = v;
    833 
    834 	switch (ap->a_name) {
    835 	case _PC_LINK_MAX:
    836 		*ap->a_retval = LINK_MAX;
    837 		return (0);
    838 	case _PC_MAX_CANON:
    839 		*ap->a_retval = MAX_CANON;
    840 		return (0);
    841 	case _PC_MAX_INPUT:
    842 		*ap->a_retval = MAX_INPUT;
    843 		return (0);
    844 	case _PC_PIPE_BUF:
    845 		*ap->a_retval = PIPE_BUF;
    846 		return (0);
    847 	case _PC_CHOWN_RESTRICTED:
    848 		*ap->a_retval = 1;
    849 		return (0);
    850 	case _PC_VDISABLE:
    851 		*ap->a_retval = _POSIX_VDISABLE;
    852 		return (0);
    853 	case _PC_SYNC_IO:
    854 		*ap->a_retval = 1;
    855 		return (0);
    856 	default:
    857 		return (EINVAL);
    858 	}
    859 	/* NOTREACHED */
    860 }
    861 
    862 /*
    863  * Advisory record locking support.
    864  */
    865 int
    866 spec_advlock(v)
    867 	void *v;
    868 {
    869 	struct vop_advlock_args /* {
    870 		struct vnode *a_vp;
    871 		void *a_id;
    872 		int a_op;
    873 		struct flock *a_fl;
    874 		int a_flags;
    875 	} */ *ap = v;
    876 	struct vnode *vp = ap->a_vp;
    877 
    878 	return lf_advlock(ap, &vp->v_speclockf, (off_t)0);
    879 }
    880