Home | History | Annotate | Line # | Download | only in specfs
spec_vnops.c revision 1.81
      1 /*	$NetBSD: spec_vnops.c,v 1.81 2005/06/21 14:01:13 ws 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 1.81 2005/06/21 14:01:13 ws 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) __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_blkatoff_desc, spec_blkatoff },		/* blkatoff */
    121 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
    122 	{ &vop_vfree_desc, spec_vfree },		/* vfree */
    123 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
    124 	{ &vop_update_desc, spec_update },		/* update */
    125 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
    126 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
    127 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
    128 	{ NULL, NULL }
    129 };
    130 const struct vnodeopv_desc spec_vnodeop_opv_desc =
    131 	{ &spec_vnodeop_p, spec_vnodeop_entries };
    132 
    133 /*
    134  * Trivial lookup routine that always fails.
    135  */
    136 int
    137 spec_lookup(v)
    138 	void *v;
    139 {
    140 	struct vop_lookup_args /* {
    141 		struct vnode *a_dvp;
    142 		struct vnode **a_vpp;
    143 		struct componentname *a_cnp;
    144 	} */ *ap = v;
    145 
    146 	*ap->a_vpp = NULL;
    147 	return (ENOTDIR);
    148 }
    149 
    150 /*
    151  * Returns true if dev is /dev/mem or /dev/kmem.
    152  */
    153 static int
    154 iskmemdev(dev_t dev)
    155 {
    156 	/* mem_no is emitted by config(8) to generated devsw.c */
    157 	extern const int mem_no;
    158 
    159 	/* minor 14 is /dev/io on i386 with COMPAT_10 */
    160 	return (major(dev) == mem_no && (minor(dev) < 2 || minor(dev) == 14));
    161 }
    162 
    163 /*
    164  * Open a special file.
    165  */
    166 /* ARGSUSED */
    167 int
    168 spec_open(v)
    169 	void *v;
    170 {
    171 	struct vop_open_args /* {
    172 		struct vnode *a_vp;
    173 		int  a_mode;
    174 		struct ucred *a_cred;
    175 		struct proc *a_p;
    176 	} */ *ap = v;
    177 	struct proc *p = ap->a_p;
    178 	struct vnode *bvp, *vp = ap->a_vp;
    179 	const struct bdevsw *bdev;
    180 	const struct cdevsw *cdev;
    181 	dev_t blkdev, dev = (dev_t)vp->v_rdev;
    182 	int error;
    183 	struct partinfo pi;
    184 	int (*d_ioctl)(dev_t, u_long, caddr_t, int, struct proc *);
    185 
    186 	/*
    187 	 * Don't allow open if fs is mounted -nodev.
    188 	 */
    189 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
    190 		return (ENXIO);
    191 
    192 	switch (vp->v_type) {
    193 
    194 	case VCHR:
    195 		cdev = cdevsw_lookup(dev);
    196 		if (cdev == NULL)
    197 			return (ENXIO);
    198 		if (ap->a_cred != FSCRED && (ap->a_mode & FWRITE)) {
    199 			/*
    200 			 * When running in very secure mode, do not allow
    201 			 * opens for writing of any disk character devices.
    202 			 */
    203 			if (securelevel >= 2 && cdev->d_type == D_DISK)
    204 				return (EPERM);
    205 			/*
    206 			 * When running in secure mode, do not allow opens
    207 			 * for writing of /dev/mem, /dev/kmem, or character
    208 			 * devices whose corresponding block devices are
    209 			 * currently mounted.
    210 			 */
    211 			if (securelevel >= 1) {
    212 				blkdev = devsw_chr2blk(dev);
    213 				if (blkdev != (dev_t)NODEV &&
    214 				    vfinddev(blkdev, VBLK, &bvp) &&
    215 				    (error = vfs_mountedon(bvp)))
    216 					return (error);
    217 				if (iskmemdev(dev))
    218 					return (EPERM);
    219 			}
    220 		}
    221 		if (cdev->d_type == D_TTY)
    222 			vp->v_flag |= VISTTY;
    223 		VOP_UNLOCK(vp, 0);
    224 		error = (*cdev->d_open)(dev, ap->a_mode, S_IFCHR, p);
    225 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    226 		if (cdev->d_type != D_DISK)
    227 			return error;
    228 		d_ioctl = cdev->d_ioctl;
    229 		break;
    230 
    231 	case VBLK:
    232 		bdev = bdevsw_lookup(dev);
    233 		if (bdev == NULL)
    234 			return (ENXIO);
    235 		/*
    236 		 * When running in very secure mode, do not allow
    237 		 * opens for writing of any disk block devices.
    238 		 */
    239 		if (securelevel >= 2 && ap->a_cred != FSCRED &&
    240 		    (ap->a_mode & FWRITE) && bdev->d_type == D_DISK)
    241 			return (EPERM);
    242 		/*
    243 		 * Do not allow opens of block devices that are
    244 		 * currently mounted.
    245 		 */
    246 		if ((error = vfs_mountedon(vp)) != 0)
    247 			return (error);
    248 		error = (*bdev->d_open)(dev, ap->a_mode, S_IFBLK, p);
    249 		d_ioctl = bdev->d_ioctl;
    250 		break;
    251 
    252 	case VNON:
    253 	case VLNK:
    254 	case VDIR:
    255 	case VREG:
    256 	case VBAD:
    257 	case VFIFO:
    258 	case VSOCK:
    259 	default:
    260 		return 0;
    261 	}
    262 
    263 	if (error)
    264 		return error;
    265 	if (!(*d_ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&pi, FREAD, curproc))
    266 		vp->v_size = (voff_t)pi.disklab->d_secsize * pi.part->p_size;
    267 	return 0;
    268 }
    269 
    270 /*
    271  * Vnode op for read
    272  */
    273 /* ARGSUSED */
    274 int
    275 spec_read(v)
    276 	void *v;
    277 {
    278 	struct vop_read_args /* {
    279 		struct vnode *a_vp;
    280 		struct uio *a_uio;
    281 		int  a_ioflag;
    282 		struct ucred *a_cred;
    283 	} */ *ap = v;
    284 	struct vnode *vp = ap->a_vp;
    285 	struct uio *uio = ap->a_uio;
    286  	struct proc *p = uio->uio_procp;
    287 	struct buf *bp;
    288 	const struct bdevsw *bdev;
    289 	const struct cdevsw *cdev;
    290 	daddr_t bn;
    291 	int bsize, bscale;
    292 	struct partinfo dpart;
    293 	int n, on;
    294 	int error = 0;
    295 
    296 #ifdef DIAGNOSTIC
    297 	if (uio->uio_rw != UIO_READ)
    298 		panic("spec_read mode");
    299 	if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc)
    300 		panic("spec_read proc");
    301 #endif
    302 	if (uio->uio_resid == 0)
    303 		return (0);
    304 
    305 	switch (vp->v_type) {
    306 
    307 	case VCHR:
    308 		VOP_UNLOCK(vp, 0);
    309 		cdev = cdevsw_lookup(vp->v_rdev);
    310 		if (cdev != NULL)
    311 			error = (*cdev->d_read)(vp->v_rdev, uio, ap->a_ioflag);
    312 		else
    313 			error = ENXIO;
    314 		vn_lock(vp, LK_SHARED | LK_RETRY);
    315 		return (error);
    316 
    317 	case VBLK:
    318 		if (uio->uio_offset < 0)
    319 			return (EINVAL);
    320 		bsize = BLKDEV_IOSIZE;
    321 		bdev = bdevsw_lookup(vp->v_rdev);
    322 		if (bdev != NULL &&
    323 		    (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&dpart,
    324 				     FREAD, p) == 0) {
    325 			if (dpart.part->p_fstype == FS_BSDFFS &&
    326 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
    327 				bsize = dpart.part->p_frag *
    328 				    dpart.part->p_fsize;
    329 		}
    330 		bscale = bsize >> DEV_BSHIFT;
    331 		do {
    332 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
    333 			on = uio->uio_offset % bsize;
    334 			n = min((unsigned)(bsize - on), uio->uio_resid);
    335 			error = bread(vp, bn, bsize, NOCRED, &bp);
    336 			n = min(n, bsize - bp->b_resid);
    337 			if (error) {
    338 				brelse(bp);
    339 				return (error);
    340 			}
    341 			error = uiomove((char *)bp->b_data + on, n, uio);
    342 			brelse(bp);
    343 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
    344 		return (error);
    345 
    346 	default:
    347 		panic("spec_read type");
    348 	}
    349 	/* NOTREACHED */
    350 }
    351 
    352 /*
    353  * Vnode op for write
    354  */
    355 /* ARGSUSED */
    356 int
    357 spec_write(v)
    358 	void *v;
    359 {
    360 	struct vop_write_args /* {
    361 		struct vnode *a_vp;
    362 		struct uio *a_uio;
    363 		int  a_ioflag;
    364 		struct ucred *a_cred;
    365 	} */ *ap = v;
    366 	struct vnode *vp = ap->a_vp;
    367 	struct uio *uio = ap->a_uio;
    368 	struct proc *p = uio->uio_procp;
    369 	struct buf *bp;
    370 	const struct bdevsw *bdev;
    371 	const struct cdevsw *cdev;
    372 	daddr_t bn;
    373 	int bsize, bscale;
    374 	struct partinfo dpart;
    375 	int n, on;
    376 	int error = 0;
    377 
    378 #ifdef DIAGNOSTIC
    379 	if (uio->uio_rw != UIO_WRITE)
    380 		panic("spec_write mode");
    381 	if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc)
    382 		panic("spec_write proc");
    383 #endif
    384 
    385 	switch (vp->v_type) {
    386 
    387 	case VCHR:
    388 		VOP_UNLOCK(vp, 0);
    389 		cdev = cdevsw_lookup(vp->v_rdev);
    390 		if (cdev != NULL)
    391 			error = (*cdev->d_write)(vp->v_rdev, uio, ap->a_ioflag);
    392 		else
    393 			error = ENXIO;
    394 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    395 		return (error);
    396 
    397 	case VBLK:
    398 		if (uio->uio_resid == 0)
    399 			return (0);
    400 		if (uio->uio_offset < 0)
    401 			return (EINVAL);
    402 		bsize = BLKDEV_IOSIZE;
    403 		bdev = bdevsw_lookup(vp->v_rdev);
    404 		if (bdev != NULL &&
    405 		    (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&dpart,
    406 				    FREAD, p) == 0) {
    407 			if (dpart.part->p_fstype == FS_BSDFFS &&
    408 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
    409 				bsize = dpart.part->p_frag *
    410 				    dpart.part->p_fsize;
    411 		}
    412 		bscale = bsize >> DEV_BSHIFT;
    413 		do {
    414 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
    415 			on = uio->uio_offset % bsize;
    416 			n = min((unsigned)(bsize - on), uio->uio_resid);
    417 			if (n == bsize)
    418 				bp = getblk(vp, bn, bsize, 0, 0);
    419 			else
    420 				error = bread(vp, bn, bsize, NOCRED, &bp);
    421 			if (error) {
    422 				brelse(bp);
    423 				return (error);
    424 			}
    425 			n = min(n, bsize - bp->b_resid);
    426 			error = uiomove((char *)bp->b_data + on, n, uio);
    427 			if (error)
    428 				brelse(bp);
    429 			else {
    430 				if (n + on == bsize)
    431 					bawrite(bp);
    432 				else
    433 					bdwrite(bp);
    434 				if (bp->b_flags & B_ERROR)
    435 					error = bp->b_error;
    436 			}
    437 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
    438 		return (error);
    439 
    440 	default:
    441 		panic("spec_write type");
    442 	}
    443 	/* NOTREACHED */
    444 }
    445 
    446 /*
    447  * Device ioctl operation.
    448  */
    449 /* ARGSUSED */
    450 int
    451 spec_ioctl(v)
    452 	void *v;
    453 {
    454 	struct vop_ioctl_args /* {
    455 		struct vnode *a_vp;
    456 		u_long a_command;
    457 		void  *a_data;
    458 		int  a_fflag;
    459 		struct ucred *a_cred;
    460 		struct proc *a_p;
    461 	} */ *ap = v;
    462 	const struct bdevsw *bdev;
    463 	const struct cdevsw *cdev;
    464 	dev_t dev = ap->a_vp->v_rdev;
    465 
    466 	switch (ap->a_vp->v_type) {
    467 
    468 	case VCHR:
    469 		cdev = cdevsw_lookup(dev);
    470 		if (cdev == NULL)
    471 			return (ENXIO);
    472 		return ((*cdev->d_ioctl)(dev, ap->a_command, ap->a_data,
    473 		    ap->a_fflag, ap->a_p));
    474 
    475 	case VBLK:
    476 		bdev = bdevsw_lookup(dev);
    477 		if (bdev == NULL)
    478 			return (ENXIO);
    479 		if (ap->a_command == 0 && (long)ap->a_data == B_TAPE) {
    480 			if (bdev->d_type == D_TAPE)
    481 				return (0);
    482 			else
    483 				return (1);
    484 		}
    485 		return ((*bdev->d_ioctl)(dev, ap->a_command, ap->a_data,
    486 		   ap->a_fflag, ap->a_p));
    487 
    488 	default:
    489 		panic("spec_ioctl");
    490 		/* NOTREACHED */
    491 	}
    492 }
    493 
    494 /* ARGSUSED */
    495 int
    496 spec_poll(v)
    497 	void *v;
    498 {
    499 	struct vop_poll_args /* {
    500 		struct vnode *a_vp;
    501 		int a_events;
    502 		struct proc *a_p;
    503 	} */ *ap = v;
    504 	const struct cdevsw *cdev;
    505 	dev_t dev;
    506 
    507 	switch (ap->a_vp->v_type) {
    508 
    509 	case VCHR:
    510 		dev = ap->a_vp->v_rdev;
    511 		cdev = cdevsw_lookup(dev);
    512 		if (cdev == NULL)
    513 			return (POLLERR);
    514 		return (*cdev->d_poll)(dev, ap->a_events, ap->a_p);
    515 
    516 	default:
    517 		return (genfs_poll(v));
    518 	}
    519 }
    520 
    521 /* ARGSUSED */
    522 int
    523 spec_kqfilter(v)
    524 	void *v;
    525 {
    526 	struct vop_kqfilter_args /* {
    527 		struct vnode	*a_vp;
    528 		struct proc	*a_kn;
    529 	} */ *ap = v;
    530 	const struct cdevsw *cdev;
    531 	dev_t dev;
    532 
    533 	switch (ap->a_vp->v_type) {
    534 
    535 	case VCHR:
    536 		dev = ap->a_vp->v_rdev;
    537 		cdev = cdevsw_lookup(dev);
    538 		if (cdev == NULL)
    539 			return (ENXIO);
    540 		return (*cdev->d_kqfilter)(dev, ap->a_kn);
    541 	default:
    542 		/*
    543 		 * Block devices don't support kqfilter, and refuse it
    544 		 * for any other files (like those vflush()ed) too.
    545 		 */
    546 		return (EOPNOTSUPP);
    547 	}
    548 }
    549 
    550 /*
    551  * Synch buffers associated with a block device
    552  */
    553 /* ARGSUSED */
    554 int
    555 spec_fsync(v)
    556 	void *v;
    557 {
    558 	struct vop_fsync_args /* {
    559 		struct vnode *a_vp;
    560 		struct ucred *a_cred;
    561 		int  a_flags;
    562 		off_t offlo;
    563 		off_t offhi;
    564 		struct proc *a_p;
    565 	} */ *ap = v;
    566 	struct vnode *vp = ap->a_vp;
    567 
    568 	if (vp->v_type == VBLK)
    569 		vflushbuf(vp, (ap->a_flags & FSYNC_WAIT) != 0);
    570 	return (0);
    571 }
    572 
    573 /*
    574  * Just call the device strategy routine
    575  */
    576 int
    577 spec_strategy(v)
    578 	void *v;
    579 {
    580 	struct vop_strategy_args /* {
    581 		struct vnode *a_vp;
    582 		struct buf *a_bp;
    583 	} */ *ap = v;
    584 	struct vnode *vp = ap->a_vp;
    585 	struct buf *bp = ap->a_bp;
    586 	int error, s;
    587 	struct spec_cow_entry *e;
    588 
    589 	error = 0;
    590 	bp->b_dev = vp->v_rdev;
    591 	if (!(bp->b_flags & B_READ) &&
    592 	    (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
    593 		(*bioops.io_start)(bp);
    594 
    595 	if (!(bp->b_flags & B_READ) && !SLIST_EMPTY(&vp->v_spec_cow_head)) {
    596 		SPEC_COW_LOCK(vp->v_specinfo, s);
    597 		while (vp->v_spec_cow_req > 0)
    598 			ltsleep(&vp->v_spec_cow_req, PRIBIO, "cowlist", 0,
    599 			    &vp->v_spec_cow_slock);
    600 		vp->v_spec_cow_count++;
    601 		SPEC_COW_UNLOCK(vp->v_specinfo, s);
    602 
    603 		SLIST_FOREACH(e, &vp->v_spec_cow_head, ce_list) {
    604 			if ((error = (*e->ce_func)(e->ce_cookie, bp)) != 0)
    605 				break;
    606 		}
    607 
    608 		SPEC_COW_LOCK(vp->v_specinfo, s);
    609 		vp->v_spec_cow_count--;
    610 		if (vp->v_spec_cow_req && vp->v_spec_cow_count == 0)
    611 			wakeup(&vp->v_spec_cow_req);
    612 		SPEC_COW_UNLOCK(vp->v_specinfo, s);
    613 	}
    614 
    615 	if (error) {
    616 		bp->b_error = error;
    617 		bp->b_flags |= B_ERROR;
    618 		biodone(bp);
    619 		return (error);
    620 	}
    621 
    622 	DEV_STRATEGY(bp);
    623 
    624 	return (0);
    625 }
    626 
    627 int
    628 spec_inactive(v)
    629 	void *v;
    630 {
    631 	struct vop_inactive_args /* {
    632 		struct vnode *a_vp;
    633 		struct proc *a_p;
    634 	} */ *ap = v;
    635 
    636 	VOP_UNLOCK(ap->a_vp, 0);
    637 	return (0);
    638 }
    639 
    640 /*
    641  * This is a noop, simply returning what one has been given.
    642  */
    643 int
    644 spec_bmap(v)
    645 	void *v;
    646 {
    647 	struct vop_bmap_args /* {
    648 		struct vnode *a_vp;
    649 		daddr_t  a_bn;
    650 		struct vnode **a_vpp;
    651 		daddr_t *a_bnp;
    652 		int *a_runp;
    653 	} */ *ap = v;
    654 
    655 	if (ap->a_vpp != NULL)
    656 		*ap->a_vpp = ap->a_vp;
    657 	if (ap->a_bnp != NULL)
    658 		*ap->a_bnp = ap->a_bn;
    659 	if (ap->a_runp != NULL)
    660 		*ap->a_runp = (MAXBSIZE >> DEV_BSHIFT) - 1;
    661 	return (0);
    662 }
    663 
    664 /*
    665  * Device close routine
    666  */
    667 /* ARGSUSED */
    668 int
    669 spec_close(v)
    670 	void *v;
    671 {
    672 	struct vop_close_args /* {
    673 		struct vnode *a_vp;
    674 		int  a_fflag;
    675 		struct ucred *a_cred;
    676 		struct proc *a_p;
    677 	} */ *ap = v;
    678 	struct vnode *vp = ap->a_vp;
    679 	const struct bdevsw *bdev;
    680 	const struct cdevsw *cdev;
    681 	struct session *sess;
    682 	dev_t dev = vp->v_rdev;
    683 	int (*devclose) __P((dev_t, int, int, struct proc *));
    684 	int mode, error, count, flags, flags1;
    685 
    686 	count = vcount(vp);
    687 	flags = vp->v_flag;
    688 
    689 	switch (vp->v_type) {
    690 
    691 	case VCHR:
    692 		/*
    693 		 * Hack: a tty device that is a controlling terminal
    694 		 * has a reference from the session structure.
    695 		 * We cannot easily tell that a character device is
    696 		 * a controlling terminal, unless it is the closing
    697 		 * process' controlling terminal.  In that case,
    698 		 * if the reference count is 2 (this last descriptor
    699 		 * plus the session), release the reference from the session.
    700 		 * Also remove the link from the tty back to the session
    701 		 * and pgrp - due to the way consoles are handled we cannot
    702 		 * guarantee that the vrele() will do the final close on the
    703 		 * actual tty device.
    704 		 */
    705 		if (count == 2 && ap->a_p &&
    706 		    vp == (sess = ap->a_p->p_session)->s_ttyvp) {
    707 			sess->s_ttyvp = NULL;
    708 			if (sess->s_ttyp->t_session != NULL) {
    709 				sess->s_ttyp->t_pgrp = NULL;
    710 				sess->s_ttyp->t_session = NULL;
    711 				SESSRELE(sess);
    712 			} else if (sess->s_ttyp->t_pgrp != NULL)
    713 				panic("spec_close: spurious pgrp ref");
    714 			vrele(vp);
    715 			count--;
    716 		}
    717 		/*
    718 		 * If the vnode is locked, then we are in the midst
    719 		 * of forcably closing the device, otherwise we only
    720 		 * close on last reference.
    721 		 */
    722 		if (count > 1 && (flags & VXLOCK) == 0)
    723 			return (0);
    724 		cdev = cdevsw_lookup(dev);
    725 		if (cdev != NULL)
    726 			devclose = cdev->d_close;
    727 		else
    728 			devclose = NULL;
    729 		mode = S_IFCHR;
    730 		break;
    731 
    732 	case VBLK:
    733 		/*
    734 		 * On last close of a block device (that isn't mounted)
    735 		 * we must invalidate any in core blocks, so that
    736 		 * we can, for instance, change floppy disks.
    737 		 */
    738 		error = vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 0, 0);
    739 		if (error)
    740 			return (error);
    741 		/*
    742 		 * We do not want to really close the device if it
    743 		 * is still in use unless we are trying to close it
    744 		 * forcibly. Since every use (buffer, vnode, swap, cmap)
    745 		 * holds a reference to the vnode, and because we mark
    746 		 * any other vnodes that alias this device, when the
    747 		 * sum of the reference counts on all the aliased
    748 		 * vnodes descends to one, we are on last close.
    749 		 */
    750 		if (count > 1 && (flags & VXLOCK) == 0)
    751 			return (0);
    752 		bdev = bdevsw_lookup(dev);
    753 		if (bdev != NULL)
    754 			devclose = bdev->d_close;
    755 		else
    756 			devclose = NULL;
    757 		mode = S_IFBLK;
    758 		break;
    759 
    760 	default:
    761 		panic("spec_close: not special");
    762 	}
    763 
    764 	flags1 = ap->a_fflag;
    765 
    766 	/*
    767 	 * if VXLOCK is set, then we're going away soon, so make this
    768 	 * non-blocking. Also ensures that we won't wedge in vn_lock below.
    769 	 */
    770 	if (flags & VXLOCK)
    771 		flags1 |= FNONBLOCK;
    772 
    773 	/*
    774 	 * If we're able to block, release the vnode lock & reacquire. We
    775 	 * might end up sleeping for someone else who wants our queues. They
    776 	 * won't get them if we hold the vnode locked. Also, if VXLOCK is set,
    777 	 * don't release the lock as we won't be able to regain it.
    778 	 */
    779 	if (!(flags1 & FNONBLOCK))
    780 		VOP_UNLOCK(vp, 0);
    781 
    782 	if (devclose != NULL)
    783 		error = (*devclose)(dev, flags1, mode, ap->a_p);
    784 	else
    785 		error = ENXIO;
    786 
    787 	if (!(flags1 & FNONBLOCK))
    788 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    789 
    790 	return (error);
    791 }
    792 
    793 /*
    794  * Print out the contents of a special device vnode.
    795  */
    796 int
    797 spec_print(v)
    798 	void *v;
    799 {
    800 	struct vop_print_args /* {
    801 		struct vnode *a_vp;
    802 	} */ *ap = v;
    803 
    804 	printf("tag VT_NON, dev %d, %d\n", major(ap->a_vp->v_rdev),
    805 	    minor(ap->a_vp->v_rdev));
    806 	return 0;
    807 }
    808 
    809 /*
    810  * Return POSIX pathconf information applicable to special devices.
    811  */
    812 int
    813 spec_pathconf(v)
    814 	void *v;
    815 {
    816 	struct vop_pathconf_args /* {
    817 		struct vnode *a_vp;
    818 		int a_name;
    819 		register_t *a_retval;
    820 	} */ *ap = v;
    821 
    822 	switch (ap->a_name) {
    823 	case _PC_LINK_MAX:
    824 		*ap->a_retval = LINK_MAX;
    825 		return (0);
    826 	case _PC_MAX_CANON:
    827 		*ap->a_retval = MAX_CANON;
    828 		return (0);
    829 	case _PC_MAX_INPUT:
    830 		*ap->a_retval = MAX_INPUT;
    831 		return (0);
    832 	case _PC_PIPE_BUF:
    833 		*ap->a_retval = PIPE_BUF;
    834 		return (0);
    835 	case _PC_CHOWN_RESTRICTED:
    836 		*ap->a_retval = 1;
    837 		return (0);
    838 	case _PC_VDISABLE:
    839 		*ap->a_retval = _POSIX_VDISABLE;
    840 		return (0);
    841 	case _PC_SYNC_IO:
    842 		*ap->a_retval = 1;
    843 		return (0);
    844 	default:
    845 		return (EINVAL);
    846 	}
    847 	/* NOTREACHED */
    848 }
    849 
    850 /*
    851  * Advisory record locking support.
    852  */
    853 int
    854 spec_advlock(v)
    855 	void *v;
    856 {
    857 	struct vop_advlock_args /* {
    858 		struct vnode *a_vp;
    859 		void *a_id;
    860 		int a_op;
    861 		struct flock *a_fl;
    862 		int a_flags;
    863 	} */ *ap = v;
    864 	struct vnode *vp = ap->a_vp;
    865 
    866 	return lf_advlock(ap, &vp->v_speclockf, (off_t)0);
    867 }
    868