Home | History | Annotate | Line # | Download | only in kern
vfs_vnops.c revision 1.45.2.10
      1  1.45.2.10   thorpej /*	$NetBSD: vfs_vnops.c,v 1.45.2.10 2002/12/19 00:50:43 thorpej Exp $	*/
      2       1.12       cgd 
      3       1.10       cgd /*
      4       1.11   mycroft  * Copyright (c) 1982, 1986, 1989, 1993
      5       1.11   mycroft  *	The Regents of the University of California.  All rights reserved.
      6       1.10       cgd  * (c) UNIX System Laboratories, Inc.
      7       1.10       cgd  * All or some portions of this file are derived from material licensed
      8       1.10       cgd  * to the University of California by American Telephone and Telegraph
      9       1.10       cgd  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     10       1.10       cgd  * the permission of UNIX System Laboratories, Inc.
     11       1.10       cgd  *
     12       1.10       cgd  * Redistribution and use in source and binary forms, with or without
     13       1.10       cgd  * modification, are permitted provided that the following conditions
     14       1.10       cgd  * are met:
     15       1.10       cgd  * 1. Redistributions of source code must retain the above copyright
     16       1.10       cgd  *    notice, this list of conditions and the following disclaimer.
     17       1.10       cgd  * 2. Redistributions in binary form must reproduce the above copyright
     18       1.10       cgd  *    notice, this list of conditions and the following disclaimer in the
     19       1.10       cgd  *    documentation and/or other materials provided with the distribution.
     20       1.10       cgd  * 3. All advertising materials mentioning features or use of this software
     21       1.10       cgd  *    must display the following acknowledgement:
     22       1.10       cgd  *	This product includes software developed by the University of
     23       1.10       cgd  *	California, Berkeley and its contributors.
     24       1.10       cgd  * 4. Neither the name of the University nor the names of its contributors
     25       1.10       cgd  *    may be used to endorse or promote products derived from this software
     26       1.10       cgd  *    without specific prior written permission.
     27       1.10       cgd  *
     28       1.10       cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29       1.10       cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30       1.10       cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31       1.10       cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32       1.10       cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33       1.10       cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34       1.10       cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35       1.10       cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36       1.10       cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37       1.10       cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38       1.10       cgd  * SUCH DAMAGE.
     39       1.10       cgd  *
     40       1.28      fvdl  *	@(#)vfs_vnops.c	8.14 (Berkeley) 6/15/95
     41       1.10       cgd  */
     42       1.26       mrg 
     43   1.45.2.4   nathanw #include <sys/cdefs.h>
     44  1.45.2.10   thorpej __KERNEL_RCSID(0, "$NetBSD: vfs_vnops.c,v 1.45.2.10 2002/12/19 00:50:43 thorpej Exp $");
     45   1.45.2.4   nathanw 
     46       1.27   thorpej #include "fs_union.h"
     47       1.10       cgd 
     48       1.10       cgd #include <sys/param.h>
     49       1.10       cgd #include <sys/systm.h>
     50       1.10       cgd #include <sys/kernel.h>
     51       1.10       cgd #include <sys/file.h>
     52       1.10       cgd #include <sys/stat.h>
     53       1.10       cgd #include <sys/buf.h>
     54       1.10       cgd #include <sys/proc.h>
     55       1.10       cgd #include <sys/mount.h>
     56       1.10       cgd #include <sys/namei.h>
     57       1.10       cgd #include <sys/vnode.h>
     58       1.10       cgd #include <sys/ioctl.h>
     59       1.10       cgd #include <sys/tty.h>
     60       1.21   mycroft #include <sys/poll.h>
     61       1.11   mycroft 
     62       1.25       mrg #include <uvm/uvm_extern.h>
     63       1.25       mrg 
     64       1.28      fvdl #ifdef UNION
     65       1.28      fvdl #include <miscfs/union/union.h>
     66       1.28      fvdl #endif
     67   1.45.2.8   nathanw #include <sys/verified_exec.h>
     68   1.45.2.8   nathanw 
     69   1.45.2.8   nathanw extern LIST_HEAD(veriexec_devhead, veriexec_dev_list) veriexec_dev_head;
     70   1.45.2.8   nathanw extern struct veriexec_devhead veriexec_file_dev_head;
     71       1.28      fvdl 
     72   1.45.2.7   nathanw static int vn_read(struct file *fp, off_t *offset, struct uio *uio,
     73   1.45.2.7   nathanw 	    struct ucred *cred, int flags);
     74   1.45.2.7   nathanw static int vn_write(struct file *fp, off_t *offset, struct uio *uio,
     75   1.45.2.7   nathanw 	    struct ucred *cred, int flags);
     76   1.45.2.7   nathanw static int vn_closefile(struct file *fp, struct proc *p);
     77   1.45.2.7   nathanw static int vn_poll(struct file *fp, int events, struct proc *p);
     78   1.45.2.7   nathanw static int vn_fcntl(struct file *fp, u_int com, caddr_t data, struct proc *p);
     79   1.45.2.7   nathanw static int vn_statfile(struct file *fp, struct stat *sb, struct proc *p);
     80   1.45.2.7   nathanw static int vn_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p);
     81   1.45.2.2   nathanw 
     82   1.45.2.2   nathanw struct 	fileops vnops = {
     83   1.45.2.2   nathanw 	vn_read, vn_write, vn_ioctl, vn_fcntl, vn_poll,
     84   1.45.2.8   nathanw 	vn_statfile, vn_closefile, vn_kqfilter
     85   1.45.2.2   nathanw };
     86       1.10       cgd 
     87       1.10       cgd /*
     88       1.10       cgd  * Common code for vnode open operations.
     89       1.10       cgd  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
     90       1.10       cgd  */
     91       1.20  christos int
     92       1.18   mycroft vn_open(ndp, fmode, cmode)
     93       1.41  augustss 	struct nameidata *ndp;
     94       1.10       cgd 	int fmode, cmode;
     95       1.10       cgd {
     96       1.41  augustss 	struct vnode *vp;
     97       1.41  augustss 	struct proc *p = ndp->ni_cnd.cn_proc;
     98       1.41  augustss 	struct ucred *cred = p->p_ucred;
     99       1.19   mycroft 	struct vattr va;
    100       1.10       cgd 	int error;
    101   1.45.2.8   nathanw #ifdef VERIFIED_EXEC
    102   1.45.2.8   nathanw 	char got_dev;
    103   1.45.2.8   nathanw 	struct veriexec_inode_list *veriexec_node;
    104   1.45.2.8   nathanw 	char fingerprint[MAXFINGERPRINTLEN];
    105   1.45.2.8   nathanw #endif
    106   1.45.2.8   nathanw 
    107       1.10       cgd 	if (fmode & O_CREAT) {
    108       1.11   mycroft 		ndp->ni_cnd.cn_nameiop = CREATE;
    109       1.11   mycroft 		ndp->ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
    110       1.38    bouyer 		if ((fmode & O_EXCL) == 0 &&
    111   1.45.2.9   thorpej 		    ((fmode & O_NOFOLLOW) == 0))
    112       1.11   mycroft 			ndp->ni_cnd.cn_flags |= FOLLOW;
    113       1.20  christos 		if ((error = namei(ndp)) != 0)
    114       1.10       cgd 			return (error);
    115       1.10       cgd 		if (ndp->ni_vp == NULL) {
    116       1.19   mycroft 			VATTR_NULL(&va);
    117       1.19   mycroft 			va.va_type = VREG;
    118       1.19   mycroft 			va.va_mode = cmode;
    119       1.28      fvdl 			if (fmode & O_EXCL)
    120       1.28      fvdl 				 va.va_vaflags |= VA_EXCLUSIVE;
    121       1.17   mycroft 			VOP_LEASE(ndp->ni_dvp, p, cred, LEASE_WRITE);
    122       1.20  christos 			error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
    123       1.20  christos 					   &ndp->ni_cnd, &va);
    124       1.20  christos 			if (error)
    125       1.10       cgd 				return (error);
    126       1.10       cgd 			fmode &= ~O_TRUNC;
    127       1.10       cgd 			vp = ndp->ni_vp;
    128       1.10       cgd 		} else {
    129       1.11   mycroft 			VOP_ABORTOP(ndp->ni_dvp, &ndp->ni_cnd);
    130       1.10       cgd 			if (ndp->ni_dvp == ndp->ni_vp)
    131       1.10       cgd 				vrele(ndp->ni_dvp);
    132       1.10       cgd 			else
    133       1.10       cgd 				vput(ndp->ni_dvp);
    134       1.10       cgd 			ndp->ni_dvp = NULL;
    135       1.10       cgd 			vp = ndp->ni_vp;
    136       1.10       cgd 			if (fmode & O_EXCL) {
    137       1.10       cgd 				error = EEXIST;
    138       1.38    bouyer 				goto bad;
    139       1.38    bouyer 			}
    140       1.38    bouyer 			if (ndp->ni_vp->v_type == VLNK) {
    141       1.38    bouyer 				error = EFTYPE;
    142       1.10       cgd 				goto bad;
    143       1.10       cgd 			}
    144       1.10       cgd 			fmode &= ~O_CREAT;
    145       1.10       cgd 		}
    146       1.10       cgd 	} else {
    147       1.11   mycroft 		ndp->ni_cnd.cn_nameiop = LOOKUP;
    148       1.11   mycroft 		ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF;
    149       1.20  christos 		if ((error = namei(ndp)) != 0)
    150       1.10       cgd 			return (error);
    151       1.10       cgd 		vp = ndp->ni_vp;
    152       1.10       cgd 	}
    153       1.10       cgd 	if (vp->v_type == VSOCK) {
    154       1.10       cgd 		error = EOPNOTSUPP;
    155       1.10       cgd 		goto bad;
    156       1.10       cgd 	}
    157   1.45.2.8   nathanw 
    158   1.45.2.8   nathanw #ifdef VERIFIED_EXEC
    159   1.45.2.8   nathanw 	veriexec_node = NULL;
    160   1.45.2.8   nathanw 
    161   1.45.2.8   nathanw 	if ((error = VOP_GETATTR(vp, &va, cred, p)) != 0)
    162   1.45.2.8   nathanw 		goto bad;
    163   1.45.2.8   nathanw #endif
    164   1.45.2.8   nathanw 
    165       1.10       cgd 	if ((fmode & O_CREAT) == 0) {
    166   1.45.2.8   nathanw #ifdef VERIFIED_EXEC
    167   1.45.2.8   nathanw 		  /*
    168   1.45.2.8   nathanw 		   * Look for the file on the fingerprint lists iff
    169   1.45.2.8   nathanw 		   * it has not been seen before.
    170   1.45.2.8   nathanw 		   */
    171   1.45.2.8   nathanw 		if ((vp->fp_status == FINGERPRINT_INVALID) ||
    172   1.45.2.8   nathanw 		    (vp->fp_status == FINGERPRINT_NODEV)) {
    173   1.45.2.8   nathanw 			  /* check the file list for the finger print */
    174   1.45.2.8   nathanw 			veriexec_node = get_veriexec_inode(&veriexec_file_dev_head,
    175   1.45.2.8   nathanw 						     va.va_fsid,
    176   1.45.2.8   nathanw 						     va.va_fileid,
    177   1.45.2.8   nathanw 						     &got_dev);
    178   1.45.2.8   nathanw 			if (veriexec_node == NULL) {
    179   1.45.2.8   nathanw 				/* failing that, check the exec list */
    180   1.45.2.8   nathanw 				veriexec_node = get_veriexec_inode(
    181   1.45.2.8   nathanw 					&veriexec_dev_head, va.va_fsid,
    182   1.45.2.8   nathanw 					va.va_fileid, &got_dev);
    183   1.45.2.8   nathanw 			}
    184   1.45.2.8   nathanw 
    185   1.45.2.8   nathanw 			if ((veriexec_node == NULL) && (got_dev == 1))
    186   1.45.2.8   nathanw 				vp->fp_status = FINGERPRINT_NOENTRY;
    187   1.45.2.8   nathanw 
    188   1.45.2.8   nathanw 			if (veriexec_node != NULL) {
    189   1.45.2.8   nathanw 				if ((error = evaluate_fingerprint(vp,
    190   1.45.2.8   nathanw 						veriexec_node, p, va.va_size,
    191   1.45.2.8   nathanw 						fingerprint)) != 0)
    192   1.45.2.8   nathanw 					goto bad;
    193   1.45.2.8   nathanw 
    194   1.45.2.8   nathanw 				if (fingerprintcmp(veriexec_node,
    195   1.45.2.8   nathanw 						   fingerprint) == 0) {
    196   1.45.2.8   nathanw 					  /* fingerprint ok */
    197   1.45.2.8   nathanw 					vp->fp_status =	FINGERPRINT_VALID;
    198   1.45.2.8   nathanw #ifdef VERIFIED_EXEC_DEBUG
    199   1.45.2.8   nathanw 					printf(
    200   1.45.2.8   nathanw 			"file fingerprint matches for dev %lu, file %lu\n",
    201   1.45.2.8   nathanw 						va.va_fsid, va.va_fileid);
    202   1.45.2.8   nathanw #endif
    203   1.45.2.8   nathanw 				} else {
    204   1.45.2.8   nathanw 					vp->fp_status =	FINGERPRINT_NOMATCH;
    205   1.45.2.8   nathanw 				}
    206   1.45.2.8   nathanw 			}
    207   1.45.2.8   nathanw 		}
    208   1.45.2.8   nathanw #endif
    209   1.45.2.8   nathanw 
    210       1.10       cgd 		if (fmode & FREAD) {
    211       1.20  christos 			if ((error = VOP_ACCESS(vp, VREAD, cred, p)) != 0)
    212       1.10       cgd 				goto bad;
    213   1.45.2.8   nathanw 
    214   1.45.2.8   nathanw #ifdef VERIFIED_EXEC
    215   1.45.2.8   nathanw 				/* file is on finger print list */
    216   1.45.2.8   nathanw 			if (vp->fp_status == FINGERPRINT_NOMATCH) {
    217   1.45.2.8   nathanw 				  /* fingerprint bad */
    218   1.45.2.8   nathanw 				printf(
    219   1.45.2.8   nathanw 		"file fingerprint does not match on dev %lu, file %lu\n",
    220   1.45.2.8   nathanw 					va.va_fsid, va.va_fileid);
    221   1.45.2.8   nathanw 				if (securelevel > 2) {
    222   1.45.2.8   nathanw 					error = EPERM;
    223   1.45.2.8   nathanw 					goto bad;
    224   1.45.2.8   nathanw 				}
    225   1.45.2.8   nathanw 			}
    226   1.45.2.8   nathanw #endif
    227       1.10       cgd 		}
    228       1.10       cgd 		if (fmode & (FWRITE | O_TRUNC)) {
    229       1.10       cgd 			if (vp->v_type == VDIR) {
    230       1.10       cgd 				error = EISDIR;
    231       1.10       cgd 				goto bad;
    232       1.10       cgd 			}
    233       1.20  christos 			if ((error = vn_writechk(vp)) != 0 ||
    234       1.20  christos 			    (error = VOP_ACCESS(vp, VWRITE, cred, p)) != 0)
    235       1.10       cgd 				goto bad;
    236   1.45.2.8   nathanw #ifdef VERIFIED_EXEC
    237   1.45.2.8   nathanw 			  /*
    238   1.45.2.8   nathanw 			   * If file has a fingerprint then
    239   1.45.2.8   nathanw 			   * deny the write request, otherwise
    240   1.45.2.8   nathanw 			   * invalidate the status so we don't
    241   1.45.2.8   nathanw 			   * keep checking for the file having
    242   1.45.2.8   nathanw 			   * a fingerprint.
    243   1.45.2.8   nathanw 			   */
    244   1.45.2.8   nathanw 			if (vp->fp_status == FINGERPRINT_VALID) {
    245   1.45.2.8   nathanw 				printf(
    246   1.45.2.8   nathanw 		      "writing to fingerprinted file for dev %lu, file %lu\n",
    247   1.45.2.8   nathanw 		      va.va_fsid, va.va_fileid);
    248   1.45.2.8   nathanw 				if (securelevel > 2) {
    249   1.45.2.8   nathanw 					error = EPERM;
    250   1.45.2.8   nathanw 					goto bad;
    251   1.45.2.8   nathanw 				} else {
    252   1.45.2.8   nathanw 					vp->fp_status =	FINGERPRINT_INVALID;
    253   1.45.2.8   nathanw 				}
    254   1.45.2.8   nathanw 			}
    255   1.45.2.8   nathanw #endif
    256       1.10       cgd 		}
    257       1.10       cgd 	}
    258       1.10       cgd 	if (fmode & O_TRUNC) {
    259       1.28      fvdl 		VOP_UNLOCK(vp, 0);			/* XXX */
    260       1.17   mycroft 		VOP_LEASE(vp, p, cred, LEASE_WRITE);
    261       1.28      fvdl 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
    262       1.19   mycroft 		VATTR_NULL(&va);
    263       1.19   mycroft 		va.va_size = 0;
    264       1.20  christos 		if ((error = VOP_SETATTR(vp, &va, cred, p)) != 0)
    265       1.10       cgd 			goto bad;
    266       1.10       cgd 	}
    267       1.20  christos 	if ((error = VOP_OPEN(vp, fmode, cred, p)) != 0)
    268       1.10       cgd 		goto bad;
    269       1.45       chs 	if (vp->v_type == VREG &&
    270       1.45       chs 	    uvn_attach(vp, fmode & FWRITE ? VM_PROT_WRITE : 0) == NULL) {
    271       1.45       chs 		error = EIO;
    272       1.45       chs 		goto bad;
    273       1.45       chs 	}
    274       1.10       cgd 	if (fmode & FWRITE)
    275       1.10       cgd 		vp->v_writecount++;
    276       1.45       chs 
    277       1.10       cgd 	return (0);
    278       1.10       cgd bad:
    279       1.10       cgd 	vput(vp);
    280       1.10       cgd 	return (error);
    281       1.10       cgd }
    282       1.10       cgd 
    283       1.10       cgd /*
    284       1.10       cgd  * Check for write permissions on the specified vnode.
    285       1.28      fvdl  * Prototype text segments cannot be written.
    286       1.10       cgd  */
    287       1.20  christos int
    288       1.10       cgd vn_writechk(vp)
    289       1.41  augustss 	struct vnode *vp;
    290       1.10       cgd {
    291       1.10       cgd 
    292       1.10       cgd 	/*
    293       1.45       chs 	 * If the vnode is in use as a process's text,
    294       1.45       chs 	 * we can't allow writing.
    295       1.10       cgd 	 */
    296       1.45       chs 	if (vp->v_flag & VTEXT)
    297       1.25       mrg 		return (ETXTBSY);
    298       1.10       cgd 	return (0);
    299       1.42       chs }
    300       1.42       chs 
    301       1.42       chs /*
    302   1.45.2.4   nathanw  * Mark a vnode as having executable mappings.
    303       1.42       chs  */
    304       1.42       chs void
    305   1.45.2.4   nathanw vn_markexec(vp)
    306       1.42       chs 	struct vnode *vp;
    307       1.42       chs {
    308   1.45.2.4   nathanw 	if ((vp->v_flag & VEXECMAP) == 0) {
    309   1.45.2.5   nathanw 		uvmexp.filepages -= vp->v_uobj.uo_npages;
    310   1.45.2.5   nathanw 		uvmexp.execpages += vp->v_uobj.uo_npages;
    311   1.45.2.1   nathanw 	}
    312   1.45.2.4   nathanw 	vp->v_flag |= VEXECMAP;
    313       1.10       cgd }
    314       1.10       cgd 
    315       1.10       cgd /*
    316   1.45.2.7   nathanw  * Mark a vnode as being the text of a process.
    317   1.45.2.7   nathanw  * Fail if the vnode is currently writable.
    318   1.45.2.7   nathanw  */
    319   1.45.2.7   nathanw int
    320   1.45.2.7   nathanw vn_marktext(vp)
    321   1.45.2.7   nathanw 	struct vnode *vp;
    322   1.45.2.7   nathanw {
    323   1.45.2.7   nathanw 
    324   1.45.2.7   nathanw 	if (vp->v_writecount != 0) {
    325   1.45.2.7   nathanw 		KASSERT((vp->v_flag & VTEXT) == 0);
    326   1.45.2.7   nathanw 		return (ETXTBSY);
    327   1.45.2.7   nathanw 	}
    328   1.45.2.7   nathanw 	vp->v_flag |= VTEXT;
    329   1.45.2.7   nathanw 	vn_markexec(vp);
    330   1.45.2.7   nathanw 	return (0);
    331   1.45.2.7   nathanw }
    332   1.45.2.7   nathanw 
    333   1.45.2.7   nathanw /*
    334       1.10       cgd  * Vnode close call
    335       1.32  wrstuden  *
    336       1.32  wrstuden  * Note: takes an unlocked vnode, while VOP_CLOSE takes a locked node.
    337       1.10       cgd  */
    338       1.20  christos int
    339       1.10       cgd vn_close(vp, flags, cred, p)
    340       1.41  augustss 	struct vnode *vp;
    341       1.10       cgd 	int flags;
    342       1.10       cgd 	struct ucred *cred;
    343       1.10       cgd 	struct proc *p;
    344       1.10       cgd {
    345       1.10       cgd 	int error;
    346       1.10       cgd 
    347       1.10       cgd 	if (flags & FWRITE)
    348       1.10       cgd 		vp->v_writecount--;
    349       1.32  wrstuden 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    350       1.10       cgd 	error = VOP_CLOSE(vp, flags, cred, p);
    351       1.32  wrstuden 	vput(vp);
    352       1.10       cgd 	return (error);
    353       1.10       cgd }
    354       1.10       cgd 
    355       1.10       cgd /*
    356       1.10       cgd  * Package up an I/O request on a vnode into a uio and do it.
    357       1.10       cgd  */
    358       1.20  christos int
    359       1.10       cgd vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, p)
    360       1.10       cgd 	enum uio_rw rw;
    361       1.10       cgd 	struct vnode *vp;
    362       1.10       cgd 	caddr_t base;
    363       1.10       cgd 	int len;
    364       1.10       cgd 	off_t offset;
    365       1.10       cgd 	enum uio_seg segflg;
    366       1.10       cgd 	int ioflg;
    367       1.10       cgd 	struct ucred *cred;
    368       1.30   thorpej 	size_t *aresid;
    369       1.10       cgd 	struct proc *p;
    370       1.10       cgd {
    371       1.10       cgd 	struct uio auio;
    372       1.10       cgd 	struct iovec aiov;
    373       1.10       cgd 	int error;
    374       1.10       cgd 
    375   1.45.2.3   nathanw 	if ((ioflg & IO_NODELOCKED) == 0) {
    376   1.45.2.3   nathanw 		if (rw == UIO_READ) {
    377   1.45.2.3   nathanw 			vn_lock(vp, LK_SHARED | LK_RETRY);
    378   1.45.2.3   nathanw 		} else {
    379   1.45.2.3   nathanw 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    380   1.45.2.3   nathanw 		}
    381   1.45.2.3   nathanw 	}
    382       1.10       cgd 	auio.uio_iov = &aiov;
    383       1.10       cgd 	auio.uio_iovcnt = 1;
    384       1.10       cgd 	aiov.iov_base = base;
    385       1.10       cgd 	aiov.iov_len = len;
    386       1.10       cgd 	auio.uio_resid = len;
    387       1.10       cgd 	auio.uio_offset = offset;
    388       1.10       cgd 	auio.uio_segflg = segflg;
    389       1.10       cgd 	auio.uio_rw = rw;
    390       1.10       cgd 	auio.uio_procp = p;
    391       1.11   mycroft 	if (rw == UIO_READ) {
    392       1.10       cgd 		error = VOP_READ(vp, &auio, ioflg, cred);
    393       1.11   mycroft 	} else {
    394       1.10       cgd 		error = VOP_WRITE(vp, &auio, ioflg, cred);
    395       1.11   mycroft 	}
    396       1.10       cgd 	if (aresid)
    397       1.10       cgd 		*aresid = auio.uio_resid;
    398       1.10       cgd 	else
    399       1.10       cgd 		if (auio.uio_resid && error == 0)
    400       1.10       cgd 			error = EIO;
    401       1.10       cgd 	if ((ioflg & IO_NODELOCKED) == 0)
    402       1.28      fvdl 		VOP_UNLOCK(vp, 0);
    403       1.10       cgd 	return (error);
    404       1.23      fvdl }
    405       1.23      fvdl 
    406       1.23      fvdl int
    407       1.23      fvdl vn_readdir(fp, buf, segflg, count, done, p, cookies, ncookies)
    408       1.23      fvdl 	struct file *fp;
    409       1.23      fvdl 	char *buf;
    410       1.28      fvdl 	int segflg, *done, *ncookies;
    411       1.23      fvdl 	u_int count;
    412       1.23      fvdl 	struct proc *p;
    413       1.28      fvdl 	off_t **cookies;
    414       1.23      fvdl {
    415       1.23      fvdl 	struct vnode *vp = (struct vnode *)fp->f_data;
    416       1.23      fvdl 	struct iovec aiov;
    417       1.23      fvdl 	struct uio auio;
    418       1.23      fvdl 	int error, eofflag;
    419       1.23      fvdl 
    420       1.23      fvdl unionread:
    421       1.23      fvdl 	if (vp->v_type != VDIR)
    422       1.23      fvdl 		return (EINVAL);
    423       1.23      fvdl 	aiov.iov_base = buf;
    424       1.23      fvdl 	aiov.iov_len = count;
    425       1.23      fvdl 	auio.uio_iov = &aiov;
    426       1.23      fvdl 	auio.uio_iovcnt = 1;
    427       1.23      fvdl 	auio.uio_rw = UIO_READ;
    428       1.23      fvdl 	auio.uio_segflg = segflg;
    429       1.23      fvdl 	auio.uio_procp = p;
    430       1.23      fvdl 	auio.uio_resid = count;
    431   1.45.2.3   nathanw 	vn_lock(vp, LK_SHARED | LK_RETRY);
    432       1.23      fvdl 	auio.uio_offset = fp->f_offset;
    433       1.28      fvdl 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, cookies,
    434       1.23      fvdl 		    ncookies);
    435       1.23      fvdl 	fp->f_offset = auio.uio_offset;
    436       1.28      fvdl 	VOP_UNLOCK(vp, 0);
    437       1.23      fvdl 	if (error)
    438       1.23      fvdl 		return (error);
    439       1.23      fvdl 
    440       1.23      fvdl #ifdef UNION
    441       1.23      fvdl {
    442       1.23      fvdl 	extern struct vnode *union_dircache __P((struct vnode *));
    443       1.23      fvdl 
    444       1.23      fvdl 	if (count == auio.uio_resid && (vp->v_op == union_vnodeop_p)) {
    445       1.23      fvdl 		struct vnode *lvp;
    446       1.23      fvdl 
    447       1.23      fvdl 		lvp = union_dircache(vp);
    448       1.23      fvdl 		if (lvp != NULLVP) {
    449       1.23      fvdl 			struct vattr va;
    450       1.23      fvdl 
    451       1.23      fvdl 			/*
    452       1.23      fvdl 			 * If the directory is opaque,
    453       1.23      fvdl 			 * then don't show lower entries
    454       1.23      fvdl 			 */
    455       1.23      fvdl 			error = VOP_GETATTR(vp, &va, fp->f_cred, p);
    456       1.23      fvdl 			if (va.va_flags & OPAQUE) {
    457       1.23      fvdl 				vput(lvp);
    458       1.23      fvdl 				lvp = NULL;
    459       1.23      fvdl 			}
    460       1.23      fvdl 		}
    461       1.23      fvdl 
    462       1.23      fvdl 		if (lvp != NULLVP) {
    463       1.23      fvdl 			error = VOP_OPEN(lvp, FREAD, fp->f_cred, p);
    464       1.23      fvdl 			if (error) {
    465       1.28      fvdl 				vput(lvp);
    466       1.23      fvdl 				return (error);
    467       1.23      fvdl 			}
    468       1.28      fvdl 			VOP_UNLOCK(lvp, 0);
    469       1.23      fvdl 			fp->f_data = (caddr_t) lvp;
    470       1.23      fvdl 			fp->f_offset = 0;
    471       1.23      fvdl 			error = vn_close(vp, FREAD, fp->f_cred, p);
    472       1.23      fvdl 			if (error)
    473       1.23      fvdl 				return (error);
    474       1.23      fvdl 			vp = lvp;
    475       1.23      fvdl 			goto unionread;
    476       1.23      fvdl 		}
    477       1.23      fvdl 	}
    478       1.23      fvdl }
    479       1.23      fvdl #endif /* UNION */
    480       1.23      fvdl 
    481       1.23      fvdl 	if (count == auio.uio_resid && (vp->v_flag & VROOT) &&
    482       1.23      fvdl 	    (vp->v_mount->mnt_flag & MNT_UNION)) {
    483       1.23      fvdl 		struct vnode *tvp = vp;
    484       1.23      fvdl 		vp = vp->v_mount->mnt_vnodecovered;
    485       1.23      fvdl 		VREF(vp);
    486       1.23      fvdl 		fp->f_data = (caddr_t) vp;
    487       1.23      fvdl 		fp->f_offset = 0;
    488       1.23      fvdl 		vrele(tvp);
    489       1.23      fvdl 		goto unionread;
    490       1.23      fvdl 	}
    491       1.23      fvdl 	*done = count - auio.uio_resid;
    492       1.23      fvdl 	return error;
    493       1.10       cgd }
    494       1.10       cgd 
    495       1.10       cgd /*
    496       1.10       cgd  * File table vnode read routine.
    497       1.10       cgd  */
    498   1.45.2.7   nathanw static int
    499       1.29   thorpej vn_read(fp, offset, uio, cred, flags)
    500       1.10       cgd 	struct file *fp;
    501       1.29   thorpej 	off_t *offset;
    502       1.10       cgd 	struct uio *uio;
    503       1.10       cgd 	struct ucred *cred;
    504       1.29   thorpej 	int flags;
    505       1.10       cgd {
    506       1.28      fvdl 	struct vnode *vp = (struct vnode *)fp->f_data;
    507       1.31    kleink 	int count, error, ioflag = 0;
    508       1.10       cgd 
    509       1.17   mycroft 	VOP_LEASE(vp, uio->uio_procp, cred, LEASE_READ);
    510       1.31    kleink 	if (fp->f_flag & FNONBLOCK)
    511       1.31    kleink 		ioflag |= IO_NDELAY;
    512       1.31    kleink 	if ((fp->f_flag & (FFSYNC | FRSYNC)) == (FFSYNC | FRSYNC))
    513       1.31    kleink 		ioflag |= IO_SYNC;
    514       1.37  wrstuden 	if (fp->f_flag & FALTIO)
    515       1.37  wrstuden 		ioflag |= IO_ALTSEMANTICS;
    516   1.45.2.3   nathanw 	vn_lock(vp, LK_SHARED | LK_RETRY);
    517       1.29   thorpej 	uio->uio_offset = *offset;
    518       1.10       cgd 	count = uio->uio_resid;
    519       1.31    kleink 	error = VOP_READ(vp, uio, ioflag, cred);
    520       1.29   thorpej 	if (flags & FOF_UPDATE_OFFSET)
    521       1.29   thorpej 		*offset += count - uio->uio_resid;
    522       1.28      fvdl 	VOP_UNLOCK(vp, 0);
    523       1.10       cgd 	return (error);
    524       1.10       cgd }
    525       1.10       cgd 
    526       1.10       cgd /*
    527       1.10       cgd  * File table vnode write routine.
    528       1.10       cgd  */
    529   1.45.2.7   nathanw static int
    530       1.29   thorpej vn_write(fp, offset, uio, cred, flags)
    531       1.10       cgd 	struct file *fp;
    532       1.29   thorpej 	off_t *offset;
    533       1.10       cgd 	struct uio *uio;
    534       1.10       cgd 	struct ucred *cred;
    535       1.29   thorpej 	int flags;
    536       1.10       cgd {
    537       1.28      fvdl 	struct vnode *vp = (struct vnode *)fp->f_data;
    538       1.17   mycroft 	int count, error, ioflag = IO_UNIT;
    539       1.10       cgd 
    540       1.10       cgd 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
    541       1.10       cgd 		ioflag |= IO_APPEND;
    542       1.10       cgd 	if (fp->f_flag & FNONBLOCK)
    543       1.10       cgd 		ioflag |= IO_NDELAY;
    544       1.24   thorpej 	if (fp->f_flag & FFSYNC ||
    545       1.24   thorpej 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
    546       1.24   thorpej 		ioflag |= IO_SYNC;
    547       1.31    kleink 	else if (fp->f_flag & FDSYNC)
    548       1.31    kleink 		ioflag |= IO_DSYNC;
    549       1.37  wrstuden 	if (fp->f_flag & FALTIO)
    550       1.37  wrstuden 		ioflag |= IO_ALTSEMANTICS;
    551       1.17   mycroft 	VOP_LEASE(vp, uio->uio_procp, cred, LEASE_WRITE);
    552       1.28      fvdl 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    553       1.29   thorpej 	uio->uio_offset = *offset;
    554       1.10       cgd 	count = uio->uio_resid;
    555       1.10       cgd 	error = VOP_WRITE(vp, uio, ioflag, cred);
    556       1.29   thorpej 	if (flags & FOF_UPDATE_OFFSET) {
    557       1.29   thorpej 		if (ioflag & IO_APPEND)
    558       1.29   thorpej 			*offset = uio->uio_offset;
    559       1.29   thorpej 		else
    560       1.29   thorpej 			*offset += count - uio->uio_resid;
    561       1.29   thorpej 	}
    562       1.28      fvdl 	VOP_UNLOCK(vp, 0);
    563       1.10       cgd 	return (error);
    564       1.10       cgd }
    565       1.10       cgd 
    566       1.10       cgd /*
    567       1.10       cgd  * File table vnode stat routine.
    568       1.10       cgd  */
    569   1.45.2.2   nathanw static int
    570   1.45.2.2   nathanw vn_statfile(fp, sb, p)
    571   1.45.2.2   nathanw 	struct file *fp;
    572   1.45.2.2   nathanw 	struct stat *sb;
    573   1.45.2.2   nathanw 	struct proc *p;
    574   1.45.2.2   nathanw {
    575   1.45.2.2   nathanw 	struct vnode *vp = (struct vnode *)fp->f_data;
    576   1.45.2.2   nathanw 
    577   1.45.2.2   nathanw 	return vn_stat(vp, sb, p);
    578   1.45.2.2   nathanw }
    579   1.45.2.2   nathanw 
    580       1.20  christos int
    581   1.45.2.7   nathanw vn_stat(vp, sb, p)
    582   1.45.2.7   nathanw 	struct vnode *vp;
    583       1.41  augustss 	struct stat *sb;
    584       1.10       cgd 	struct proc *p;
    585       1.10       cgd {
    586       1.19   mycroft 	struct vattr va;
    587       1.10       cgd 	int error;
    588       1.35  wrstuden 	mode_t mode;
    589       1.10       cgd 
    590       1.19   mycroft 	error = VOP_GETATTR(vp, &va, p->p_ucred, p);
    591       1.10       cgd 	if (error)
    592       1.10       cgd 		return (error);
    593       1.10       cgd 	/*
    594       1.10       cgd 	 * Copy from vattr table
    595       1.10       cgd 	 */
    596       1.19   mycroft 	sb->st_dev = va.va_fsid;
    597       1.19   mycroft 	sb->st_ino = va.va_fileid;
    598       1.19   mycroft 	mode = va.va_mode;
    599       1.10       cgd 	switch (vp->v_type) {
    600       1.10       cgd 	case VREG:
    601       1.10       cgd 		mode |= S_IFREG;
    602       1.10       cgd 		break;
    603       1.10       cgd 	case VDIR:
    604       1.10       cgd 		mode |= S_IFDIR;
    605       1.10       cgd 		break;
    606       1.10       cgd 	case VBLK:
    607       1.10       cgd 		mode |= S_IFBLK;
    608       1.10       cgd 		break;
    609       1.10       cgd 	case VCHR:
    610       1.10       cgd 		mode |= S_IFCHR;
    611       1.10       cgd 		break;
    612       1.10       cgd 	case VLNK:
    613       1.10       cgd 		mode |= S_IFLNK;
    614       1.10       cgd 		break;
    615       1.10       cgd 	case VSOCK:
    616       1.10       cgd 		mode |= S_IFSOCK;
    617       1.10       cgd 		break;
    618       1.10       cgd 	case VFIFO:
    619       1.10       cgd 		mode |= S_IFIFO;
    620       1.10       cgd 		break;
    621       1.10       cgd 	default:
    622       1.10       cgd 		return (EBADF);
    623       1.10       cgd 	};
    624       1.10       cgd 	sb->st_mode = mode;
    625       1.19   mycroft 	sb->st_nlink = va.va_nlink;
    626       1.19   mycroft 	sb->st_uid = va.va_uid;
    627       1.19   mycroft 	sb->st_gid = va.va_gid;
    628       1.19   mycroft 	sb->st_rdev = va.va_rdev;
    629       1.19   mycroft 	sb->st_size = va.va_size;
    630       1.19   mycroft 	sb->st_atimespec = va.va_atime;
    631       1.19   mycroft 	sb->st_mtimespec = va.va_mtime;
    632       1.19   mycroft 	sb->st_ctimespec = va.va_ctime;
    633       1.19   mycroft 	sb->st_blksize = va.va_blocksize;
    634       1.19   mycroft 	sb->st_flags = va.va_flags;
    635       1.22   mycroft 	sb->st_gen = 0;
    636       1.19   mycroft 	sb->st_blocks = va.va_bytes / S_BLKSIZE;
    637       1.10       cgd 	return (0);
    638       1.37  wrstuden }
    639       1.37  wrstuden 
    640       1.37  wrstuden /*
    641       1.37  wrstuden  * File table vnode fcntl routine.
    642       1.37  wrstuden  */
    643   1.45.2.7   nathanw static int
    644       1.37  wrstuden vn_fcntl(fp, com, data, p)
    645       1.37  wrstuden 	struct file *fp;
    646       1.37  wrstuden 	u_int com;
    647       1.37  wrstuden 	caddr_t data;
    648       1.37  wrstuden 	struct proc *p;
    649       1.37  wrstuden {
    650       1.41  augustss 	struct vnode *vp = ((struct vnode *)fp->f_data);
    651       1.37  wrstuden 	int error;
    652       1.37  wrstuden 
    653       1.37  wrstuden 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    654       1.37  wrstuden 	error = VOP_FCNTL(vp, com, data, fp->f_flag, p->p_ucred, p);
    655       1.37  wrstuden 	VOP_UNLOCK(vp, 0);
    656       1.37  wrstuden 	return (error);
    657       1.10       cgd }
    658       1.10       cgd 
    659       1.10       cgd /*
    660       1.10       cgd  * File table vnode ioctl routine.
    661       1.10       cgd  */
    662   1.45.2.7   nathanw static int
    663       1.10       cgd vn_ioctl(fp, com, data, p)
    664       1.10       cgd 	struct file *fp;
    665       1.15       cgd 	u_long com;
    666       1.10       cgd 	caddr_t data;
    667       1.10       cgd 	struct proc *p;
    668       1.10       cgd {
    669       1.41  augustss 	struct vnode *vp = ((struct vnode *)fp->f_data);
    670       1.10       cgd 	struct vattr vattr;
    671       1.10       cgd 	int error;
    672       1.10       cgd 
    673       1.10       cgd 	switch (vp->v_type) {
    674       1.10       cgd 
    675       1.10       cgd 	case VREG:
    676       1.10       cgd 	case VDIR:
    677       1.10       cgd 		if (com == FIONREAD) {
    678       1.20  christos 			error = VOP_GETATTR(vp, &vattr, p->p_ucred, p);
    679       1.20  christos 			if (error)
    680       1.10       cgd 				return (error);
    681       1.10       cgd 			*(int *)data = vattr.va_size - fp->f_offset;
    682       1.10       cgd 			return (0);
    683  1.45.2.10   thorpej 		}
    684  1.45.2.10   thorpej 		if (com == FIOGETBMAP) {
    685  1.45.2.10   thorpej 			daddr_t *block;
    686  1.45.2.10   thorpej 
    687  1.45.2.10   thorpej 			block = (daddr_t *)data;
    688  1.45.2.10   thorpej 			return (VOP_BMAP(vp, *block, NULL, block, NULL));
    689       1.10       cgd 		}
    690       1.10       cgd 		if (com == FIONBIO || com == FIOASYNC)	/* XXX */
    691       1.10       cgd 			return (0);			/* XXX */
    692       1.10       cgd 		/* fall into ... */
    693       1.10       cgd 
    694       1.10       cgd 	default:
    695   1.45.2.6   nathanw 		return (EPASSTHROUGH);
    696       1.10       cgd 
    697       1.10       cgd 	case VFIFO:
    698       1.10       cgd 	case VCHR:
    699       1.10       cgd 	case VBLK:
    700       1.10       cgd 		error = VOP_IOCTL(vp, com, data, fp->f_flag, p->p_ucred, p);
    701       1.10       cgd 		if (error == 0 && com == TIOCSCTTY) {
    702       1.13       cgd 			if (p->p_session->s_ttyvp)
    703       1.13       cgd 				vrele(p->p_session->s_ttyvp);
    704       1.10       cgd 			p->p_session->s_ttyvp = vp;
    705       1.10       cgd 			VREF(vp);
    706       1.10       cgd 		}
    707       1.10       cgd 		return (error);
    708       1.10       cgd 	}
    709       1.10       cgd }
    710       1.10       cgd 
    711       1.10       cgd /*
    712       1.21   mycroft  * File table vnode poll routine.
    713       1.10       cgd  */
    714   1.45.2.7   nathanw static int
    715       1.21   mycroft vn_poll(fp, events, p)
    716       1.10       cgd 	struct file *fp;
    717       1.21   mycroft 	int events;
    718       1.10       cgd 	struct proc *p;
    719       1.10       cgd {
    720       1.10       cgd 
    721       1.21   mycroft 	return (VOP_POLL(((struct vnode *)fp->f_data), events, p));
    722   1.45.2.8   nathanw }
    723   1.45.2.8   nathanw 
    724   1.45.2.8   nathanw /*
    725   1.45.2.8   nathanw  * File table vnode kqfilter routine.
    726   1.45.2.8   nathanw  */
    727   1.45.2.8   nathanw int
    728   1.45.2.8   nathanw vn_kqfilter(fp, kn)
    729   1.45.2.8   nathanw 	struct file *fp;
    730   1.45.2.8   nathanw 	struct knote *kn;
    731   1.45.2.8   nathanw {
    732   1.45.2.8   nathanw 
    733   1.45.2.8   nathanw 	return (VOP_KQFILTER((struct vnode *)fp->f_data, kn));
    734       1.28      fvdl }
    735       1.28      fvdl 
    736       1.28      fvdl /*
    737       1.28      fvdl  * Check that the vnode is still valid, and if so
    738       1.28      fvdl  * acquire requested lock.
    739       1.28      fvdl  */
    740       1.28      fvdl int
    741       1.28      fvdl vn_lock(vp, flags)
    742       1.28      fvdl 	struct vnode *vp;
    743       1.28      fvdl 	int flags;
    744       1.28      fvdl {
    745       1.28      fvdl 	int error;
    746       1.34  sommerfe 
    747       1.28      fvdl 	do {
    748       1.36   mycroft 		if ((flags & LK_INTERLOCK) == 0)
    749       1.28      fvdl 			simple_lock(&vp->v_interlock);
    750       1.28      fvdl 		if (vp->v_flag & VXLOCK) {
    751   1.45.2.3   nathanw 			if (flags & LK_NOWAIT) {
    752   1.45.2.3   nathanw 				simple_unlock(&vp->v_interlock);
    753   1.45.2.3   nathanw 				return EBUSY;
    754   1.45.2.3   nathanw 			}
    755       1.28      fvdl 			vp->v_flag |= VXWANT;
    756   1.45.2.3   nathanw 			ltsleep(vp, PINOD | PNORELOCK,
    757       1.44  sommerfe 			    "vn_lock", 0, &vp->v_interlock);
    758       1.28      fvdl 			error = ENOENT;
    759       1.28      fvdl 		} else {
    760       1.28      fvdl 			error = VOP_LOCK(vp, flags | LK_INTERLOCK);
    761   1.45.2.3   nathanw 			if (error == 0 || error == EDEADLK || error == EBUSY)
    762       1.28      fvdl 				return (error);
    763       1.28      fvdl 		}
    764       1.28      fvdl 		flags &= ~LK_INTERLOCK;
    765       1.28      fvdl 	} while (flags & LK_RETRY);
    766       1.28      fvdl 	return (error);
    767       1.10       cgd }
    768       1.10       cgd 
    769       1.10       cgd /*
    770       1.10       cgd  * File table vnode close routine.
    771       1.10       cgd  */
    772   1.45.2.7   nathanw static int
    773       1.10       cgd vn_closefile(fp, p)
    774       1.10       cgd 	struct file *fp;
    775       1.10       cgd 	struct proc *p;
    776       1.10       cgd {
    777       1.10       cgd 
    778       1.10       cgd 	return (vn_close(((struct vnode *)fp->f_data), fp->f_flag,
    779       1.10       cgd 		fp->f_cred, p));
    780       1.39      fvdl }
    781       1.39      fvdl 
    782       1.39      fvdl /*
    783       1.39      fvdl  * Enable LK_CANRECURSE on lock. Return prior status.
    784       1.39      fvdl  */
    785       1.39      fvdl u_int
    786       1.39      fvdl vn_setrecurse(vp)
    787       1.39      fvdl 	struct vnode *vp;
    788       1.39      fvdl {
    789       1.39      fvdl 	struct lock *lkp = &vp->v_lock;
    790       1.39      fvdl 	u_int retval = lkp->lk_flags & LK_CANRECURSE;
    791       1.39      fvdl 
    792       1.39      fvdl 	lkp->lk_flags |= LK_CANRECURSE;
    793       1.39      fvdl 	return retval;
    794       1.39      fvdl }
    795       1.39      fvdl 
    796       1.39      fvdl /*
    797       1.39      fvdl  * Called when done with locksetrecurse.
    798       1.39      fvdl  */
    799       1.39      fvdl void
    800       1.39      fvdl vn_restorerecurse(vp, flags)
    801       1.39      fvdl 	struct vnode *vp;
    802       1.39      fvdl 	u_int flags;
    803       1.39      fvdl {
    804       1.39      fvdl 	struct lock *lkp = &vp->v_lock;
    805       1.39      fvdl 
    806       1.39      fvdl 	lkp->lk_flags &= ~LK_CANRECURSE;
    807       1.39      fvdl 	lkp->lk_flags |= flags;
    808       1.10       cgd }
    809