xref: /netbsd-src/sys/kern/vfs_vnops.c (revision fad4c9f71477ae11cea2ee75ec82151ac770a534)
1 /*	$NetBSD: vfs_vnops.c,v 1.112 2006/05/27 23:46:49 simonb Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vfs_vnops.c	8.14 (Berkeley) 6/15/95
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: vfs_vnops.c,v 1.112 2006/05/27 23:46:49 simonb Exp $");
41 
42 #include "opt_verified_exec.h"
43 
44 #include "fs_union.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/file.h>
50 #include <sys/stat.h>
51 #include <sys/buf.h>
52 #include <sys/proc.h>
53 #include <sys/malloc.h>
54 #include <sys/mount.h>
55 #include <sys/namei.h>
56 #include <sys/vnode.h>
57 #include <sys/ioctl.h>
58 #include <sys/tty.h>
59 #include <sys/poll.h>
60 #include <sys/kauth.h>
61 
62 #include <miscfs/specfs/specdev.h>
63 
64 #include <uvm/uvm_extern.h>
65 #include <uvm/uvm_readahead.h>
66 
67 #ifdef UNION
68 #include <fs/union/union.h>
69 #endif
70 
71 #if defined(LKM) || defined(UNION)
72 int (*vn_union_readdir_hook) (struct vnode **, struct file *, struct lwp *);
73 #endif
74 
75 #ifdef VERIFIED_EXEC
76 #include <sys/verified_exec.h>
77 #endif
78 
79 static int vn_read(struct file *fp, off_t *offset, struct uio *uio,
80 	    kauth_cred_t cred, int flags);
81 static int vn_write(struct file *fp, off_t *offset, struct uio *uio,
82 	    kauth_cred_t cred, int flags);
83 static int vn_closefile(struct file *fp, struct lwp *l);
84 static int vn_poll(struct file *fp, int events, struct lwp *l);
85 static int vn_fcntl(struct file *fp, u_int com, void *data, struct lwp *l);
86 static int vn_statfile(struct file *fp, struct stat *sb, struct lwp *l);
87 static int vn_ioctl(struct file *fp, u_long com, void *data, struct lwp *l);
88 
89 const struct fileops vnops = {
90 	vn_read, vn_write, vn_ioctl, vn_fcntl, vn_poll,
91 	vn_statfile, vn_closefile, vn_kqfilter
92 };
93 
94 /*
95  * Common code for vnode open operations.
96  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
97  */
98 int
99 vn_open(struct nameidata *ndp, int fmode, int cmode)
100 {
101 	struct vnode *vp;
102 	struct mount *mp = NULL;	/* XXX: GCC */
103 	struct lwp *l = ndp->ni_cnd.cn_lwp;
104 	kauth_cred_t cred = l->l_proc->p_cred;
105 	struct vattr va;
106 	int error;
107 #ifdef VERIFIED_EXEC
108 	struct veriexec_hash_entry *vhe = NULL;
109 	char pathbuf[MAXPATHLEN];
110 	size_t pathlen;
111 	int (*copyfun)(const void *, void *, size_t, size_t *) =
112 	    ndp->ni_segflg == UIO_SYSSPACE ? copystr : copyinstr;
113 #endif /* VERIFIED_EXEC */
114 
115 #ifdef VERIFIED_EXEC
116 	error = (*copyfun)(ndp->ni_dirp, pathbuf, sizeof(pathbuf), &pathlen);
117 	if (error) {
118 		if (veriexec_verbose >= 1)
119 			printf("veriexec: Can't copy path. (error=%d)\n",
120 			    error);
121 
122 		return (error);
123 	}
124 #endif /* VERIFIED_EXEC */
125 
126 restart:
127 	if (fmode & O_CREAT) {
128 		ndp->ni_cnd.cn_nameiop = CREATE;
129 		ndp->ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
130 		if ((fmode & O_EXCL) == 0 &&
131 		    ((fmode & O_NOFOLLOW) == 0))
132 			ndp->ni_cnd.cn_flags |= FOLLOW;
133 		if ((error = namei(ndp)) != 0)
134 			return (error);
135 		if (ndp->ni_vp == NULL) {
136 #ifdef VERIFIED_EXEC
137 			/* Lockdown mode: Prevent creation of new files. */
138 			if (veriexec_strict >= 3) {
139 				VOP_ABORTOP(ndp->ni_dvp, &ndp->ni_cnd);
140 
141 				printf("Veriexec: vn_open: Preventing "
142 				       "new file creation in %s.\n",
143 				       pathbuf);
144 
145 				vp = ndp->ni_dvp;
146 				error = EPERM;
147 				goto bad;
148 			}
149 #endif /* VERIFIED_EXEC */
150 
151 			VATTR_NULL(&va);
152 			va.va_type = VREG;
153 			va.va_mode = cmode;
154 			if (fmode & O_EXCL)
155 				 va.va_vaflags |= VA_EXCLUSIVE;
156 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
157 				VOP_ABORTOP(ndp->ni_dvp, &ndp->ni_cnd);
158 				vput(ndp->ni_dvp);
159 				if ((error = vn_start_write(NULL, &mp,
160 				    V_WAIT | V_SLEEPONLY | V_PCATCH)) != 0)
161 					return (error);
162 				goto restart;
163 			}
164 			VOP_LEASE(ndp->ni_dvp, l, cred, LEASE_WRITE);
165 			error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
166 					   &ndp->ni_cnd, &va);
167 			vn_finished_write(mp, 0);
168 			if (error)
169 				return (error);
170 			fmode &= ~O_TRUNC;
171 			vp = ndp->ni_vp;
172 		} else {
173 			VOP_ABORTOP(ndp->ni_dvp, &ndp->ni_cnd);
174 			if (ndp->ni_dvp == ndp->ni_vp)
175 				vrele(ndp->ni_dvp);
176 			else
177 				vput(ndp->ni_dvp);
178 			ndp->ni_dvp = NULL;
179 			vp = ndp->ni_vp;
180 			if (fmode & O_EXCL) {
181 				error = EEXIST;
182 				goto bad;
183 			}
184 			fmode &= ~O_CREAT;
185 		}
186 	} else {
187 		ndp->ni_cnd.cn_nameiop = LOOKUP;
188 		ndp->ni_cnd.cn_flags = LOCKLEAF;
189 		if ((fmode & O_NOFOLLOW) == 0)
190 			ndp->ni_cnd.cn_flags |= FOLLOW;
191 		if ((error = namei(ndp)) != 0)
192 			return (error);
193 		vp = ndp->ni_vp;
194 	}
195 	if (vp->v_type == VSOCK) {
196 		error = EOPNOTSUPP;
197 		goto bad;
198 	}
199 	if (ndp->ni_vp->v_type == VLNK) {
200 		error = EFTYPE;
201 		goto bad;
202 	}
203 
204 #ifdef VERIFIED_EXEC
205 	if ((error = VOP_GETATTR(vp, &va, cred, l)) != 0)
206 		goto bad;
207 #endif
208 
209 	if ((fmode & O_CREAT) == 0) {
210 #ifdef VERIFIED_EXEC
211 		if ((error = veriexec_verify(l, vp, &va, pathbuf,
212 					     VERIEXEC_FILE, &vhe)) != 0)
213 			goto bad;
214 #endif
215 
216 		if (fmode & FREAD) {
217 			if ((error = VOP_ACCESS(vp, VREAD, cred, l)) != 0)
218 				goto bad;
219 		}
220 
221 		if (fmode & (FWRITE | O_TRUNC)) {
222 			if (vp->v_type == VDIR) {
223 				error = EISDIR;
224 				goto bad;
225 			}
226 			if ((error = vn_writechk(vp)) != 0 ||
227 			    (error = VOP_ACCESS(vp, VWRITE, cred, l)) != 0)
228 				goto bad;
229 #ifdef VERIFIED_EXEC
230 			if (vhe != NULL) {
231 				veriexec_report("Write access request.",
232 						pathbuf, &va, l,
233 						REPORT_NOVERBOSE,
234 						REPORT_ALARM,
235 						REPORT_NOPANIC);
236 
237 				/* IPS mode: Deny writing to monitored files. */
238 				if (veriexec_strict >= 2) {
239 					error = EPERM;
240 					goto bad;
241 				} else {
242 					vhe->status = FINGERPRINT_NOTEVAL;
243 				}
244 			}
245 #endif
246 		}
247 	}
248 
249 	if (fmode & O_TRUNC) {
250 		VOP_UNLOCK(vp, 0);			/* XXX */
251 		if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) {
252 			vrele(vp);
253 			return (error);
254 		}
255 		VOP_LEASE(vp, l, cred, LEASE_WRITE);
256 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
257 		VATTR_NULL(&va);
258 		va.va_size = 0;
259 		error = VOP_SETATTR(vp, &va, cred, l);
260 		vn_finished_write(mp, 0);
261 		if (error != 0)
262 			goto bad;
263 	}
264 	if ((error = VOP_OPEN(vp, fmode, cred, l)) != 0)
265 		goto bad;
266 	if (vp->v_type == VREG &&
267 	    uvn_attach(vp, fmode & FWRITE ? VM_PROT_WRITE : 0) == NULL) {
268 		error = EIO;
269 		goto bad;
270 	}
271 	if (fmode & FWRITE)
272 		vp->v_writecount++;
273 
274 	return (0);
275 bad:
276 	vput(vp);
277 	return (error);
278 }
279 
280 /*
281  * Check for write permissions on the specified vnode.
282  * Prototype text segments cannot be written.
283  */
284 int
285 vn_writechk(struct vnode *vp)
286 {
287 
288 	/*
289 	 * If the vnode is in use as a process's text,
290 	 * we can't allow writing.
291 	 */
292 	if (vp->v_flag & VTEXT)
293 		return (ETXTBSY);
294 	return (0);
295 }
296 
297 /*
298  * Mark a vnode as having executable mappings.
299  */
300 void
301 vn_markexec(struct vnode *vp)
302 {
303 	if ((vp->v_flag & VEXECMAP) == 0) {
304 		uvmexp.filepages -= vp->v_uobj.uo_npages;
305 		uvmexp.execpages += vp->v_uobj.uo_npages;
306 	}
307 	vp->v_flag |= VEXECMAP;
308 }
309 
310 /*
311  * Mark a vnode as being the text of a process.
312  * Fail if the vnode is currently writable.
313  */
314 int
315 vn_marktext(struct vnode *vp)
316 {
317 
318 	if (vp->v_writecount != 0) {
319 		KASSERT((vp->v_flag & VTEXT) == 0);
320 		return (ETXTBSY);
321 	}
322 	vp->v_flag |= VTEXT;
323 	vn_markexec(vp);
324 	return (0);
325 }
326 
327 /*
328  * Vnode close call
329  *
330  * Note: takes an unlocked vnode, while VOP_CLOSE takes a locked node.
331  */
332 int
333 vn_close(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l)
334 {
335 	int error;
336 
337 	if (flags & FWRITE)
338 		vp->v_writecount--;
339 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
340 	error = VOP_CLOSE(vp, flags, cred, l);
341 	vput(vp);
342 	return (error);
343 }
344 
345 /*
346  * Package up an I/O request on a vnode into a uio and do it.
347  */
348 int
349 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, off_t offset,
350     enum uio_seg segflg, int ioflg, kauth_cred_t cred, size_t *aresid,
351     struct lwp *l)
352 {
353 	struct uio auio;
354 	struct iovec aiov;
355 	struct mount *mp = NULL;
356 	int error;
357 
358 	if ((ioflg & IO_NODELOCKED) == 0) {
359 		if (rw == UIO_READ) {
360 			vn_lock(vp, LK_SHARED | LK_RETRY);
361 		} else /* UIO_WRITE */ {
362 			if (vp->v_type != VCHR &&
363 			    (error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH))
364 			    != 0)
365 				return (error);
366 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
367 		}
368 	}
369 	auio.uio_iov = &aiov;
370 	auio.uio_iovcnt = 1;
371 	aiov.iov_base = base;
372 	aiov.iov_len = len;
373 	auio.uio_resid = len;
374 	auio.uio_offset = offset;
375 	auio.uio_rw = rw;
376 	if (segflg == UIO_SYSSPACE) {
377 		UIO_SETUP_SYSSPACE(&auio);
378 	} else {
379 		auio.uio_vmspace = l->l_proc->p_vmspace;
380 	}
381 	if (rw == UIO_READ) {
382 		error = VOP_READ(vp, &auio, ioflg, cred);
383 	} else {
384 		error = VOP_WRITE(vp, &auio, ioflg, cred);
385 	}
386 	if (aresid)
387 		*aresid = auio.uio_resid;
388 	else
389 		if (auio.uio_resid && error == 0)
390 			error = EIO;
391 	if ((ioflg & IO_NODELOCKED) == 0) {
392 		if (rw == UIO_WRITE)
393 			vn_finished_write(mp, 0);
394 		VOP_UNLOCK(vp, 0);
395 	}
396 	return (error);
397 }
398 
399 int
400 vn_readdir(struct file *fp, char *bf, int segflg, u_int count, int *done,
401     struct lwp *l, off_t **cookies, int *ncookies)
402 {
403 	struct vnode *vp = (struct vnode *)fp->f_data;
404 	struct iovec aiov;
405 	struct uio auio;
406 	int error, eofflag;
407 
408 	/* Limit the size on any kernel buffers used by VOP_READDIR */
409 	count = min(MAXBSIZE, count);
410 
411 unionread:
412 	if (vp->v_type != VDIR)
413 		return (EINVAL);
414 	aiov.iov_base = bf;
415 	aiov.iov_len = count;
416 	auio.uio_iov = &aiov;
417 	auio.uio_iovcnt = 1;
418 	auio.uio_rw = UIO_READ;
419 	if (segflg == UIO_SYSSPACE) {
420 		UIO_SETUP_SYSSPACE(&auio);
421 	} else {
422 		KASSERT(l == curlwp);
423 		auio.uio_vmspace = l->l_proc->p_vmspace;
424 	}
425 	auio.uio_resid = count;
426 	vn_lock(vp, LK_SHARED | LK_RETRY);
427 	auio.uio_offset = fp->f_offset;
428 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, cookies,
429 		    ncookies);
430 	fp->f_offset = auio.uio_offset;
431 	VOP_UNLOCK(vp, 0);
432 	if (error)
433 		return (error);
434 
435 #if defined(UNION) || defined(LKM)
436 	if (count == auio.uio_resid && vn_union_readdir_hook) {
437 		struct vnode *ovp = vp;
438 
439 		error = (*vn_union_readdir_hook)(&vp, fp, l);
440 		if (error)
441 			return (error);
442 		if (vp != ovp)
443 			goto unionread;
444 	}
445 #endif /* UNION || LKM */
446 
447 	if (count == auio.uio_resid && (vp->v_flag & VROOT) &&
448 	    (vp->v_mount->mnt_flag & MNT_UNION)) {
449 		struct vnode *tvp = vp;
450 		vp = vp->v_mount->mnt_vnodecovered;
451 		VREF(vp);
452 		fp->f_data = vp;
453 		fp->f_offset = 0;
454 		vrele(tvp);
455 		goto unionread;
456 	}
457 	*done = count - auio.uio_resid;
458 	return error;
459 }
460 
461 /*
462  * File table vnode read routine.
463  */
464 static int
465 vn_read(struct file *fp, off_t *offset, struct uio *uio, kauth_cred_t cred,
466     int flags)
467 {
468 	struct vnode *vp = (struct vnode *)fp->f_data;
469 	int count, error, ioflag;
470 	struct lwp *l = curlwp;
471 
472 	VOP_LEASE(vp, l, cred, LEASE_READ);
473 	ioflag = IO_ADV_ENCODE(fp->f_advice);
474 	if (fp->f_flag & FNONBLOCK)
475 		ioflag |= IO_NDELAY;
476 	if ((fp->f_flag & (FFSYNC | FRSYNC)) == (FFSYNC | FRSYNC))
477 		ioflag |= IO_SYNC;
478 	if (fp->f_flag & FALTIO)
479 		ioflag |= IO_ALTSEMANTICS;
480 	vn_lock(vp, LK_SHARED | LK_RETRY);
481 	uio->uio_offset = *offset;
482 	count = uio->uio_resid;
483 	error = VOP_READ(vp, uio, ioflag, cred);
484 	if (flags & FOF_UPDATE_OFFSET)
485 		*offset += count - uio->uio_resid;
486 	VOP_UNLOCK(vp, 0);
487 	return (error);
488 }
489 
490 /*
491  * File table vnode write routine.
492  */
493 static int
494 vn_write(struct file *fp, off_t *offset, struct uio *uio, kauth_cred_t cred,
495     int flags)
496 {
497 	struct vnode *vp = (struct vnode *)fp->f_data;
498 	struct mount *mp;
499 	int count, error, ioflag = IO_UNIT;
500 	struct lwp *l = curlwp;
501 
502 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
503 		ioflag |= IO_APPEND;
504 	if (fp->f_flag & FNONBLOCK)
505 		ioflag |= IO_NDELAY;
506 	if (fp->f_flag & FFSYNC ||
507 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
508 		ioflag |= IO_SYNC;
509 	else if (fp->f_flag & FDSYNC)
510 		ioflag |= IO_DSYNC;
511 	if (fp->f_flag & FALTIO)
512 		ioflag |= IO_ALTSEMANTICS;
513 	mp = NULL;
514 	if (vp->v_type != VCHR &&
515 	    (error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0)
516 		return (error);
517 	VOP_LEASE(vp, l, cred, LEASE_WRITE);
518 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
519 	uio->uio_offset = *offset;
520 	count = uio->uio_resid;
521 	error = VOP_WRITE(vp, uio, ioflag, cred);
522 	if (flags & FOF_UPDATE_OFFSET) {
523 		if (ioflag & IO_APPEND)
524 			*offset = uio->uio_offset;
525 		else
526 			*offset += count - uio->uio_resid;
527 	}
528 	VOP_UNLOCK(vp, 0);
529 	vn_finished_write(mp, 0);
530 	return (error);
531 }
532 
533 /*
534  * File table vnode stat routine.
535  */
536 static int
537 vn_statfile(struct file *fp, struct stat *sb, struct lwp *l)
538 {
539 	struct vnode *vp = (struct vnode *)fp->f_data;
540 
541 	return vn_stat(vp, sb, l);
542 }
543 
544 int
545 vn_stat(struct vnode *vp, struct stat *sb, struct lwp *l)
546 {
547 	struct vattr va;
548 	int error;
549 	mode_t mode;
550 
551 	error = VOP_GETATTR(vp, &va, l->l_proc->p_cred, l);
552 	if (error)
553 		return (error);
554 	/*
555 	 * Copy from vattr table
556 	 */
557 	sb->st_dev = va.va_fsid;
558 	sb->st_ino = va.va_fileid;
559 	mode = va.va_mode;
560 	switch (vp->v_type) {
561 	case VREG:
562 		mode |= S_IFREG;
563 		break;
564 	case VDIR:
565 		mode |= S_IFDIR;
566 		break;
567 	case VBLK:
568 		mode |= S_IFBLK;
569 		break;
570 	case VCHR:
571 		mode |= S_IFCHR;
572 		break;
573 	case VLNK:
574 		mode |= S_IFLNK;
575 		break;
576 	case VSOCK:
577 		mode |= S_IFSOCK;
578 		break;
579 	case VFIFO:
580 		mode |= S_IFIFO;
581 		break;
582 	default:
583 		return (EBADF);
584 	};
585 	sb->st_mode = mode;
586 	sb->st_nlink = va.va_nlink;
587 	sb->st_uid = va.va_uid;
588 	sb->st_gid = va.va_gid;
589 	sb->st_rdev = va.va_rdev;
590 	sb->st_size = va.va_size;
591 	sb->st_atimespec = va.va_atime;
592 	sb->st_mtimespec = va.va_mtime;
593 	sb->st_ctimespec = va.va_ctime;
594 	sb->st_birthtimespec = va.va_birthtime;
595 	sb->st_blksize = va.va_blocksize;
596 	sb->st_flags = va.va_flags;
597 	sb->st_gen = 0;
598 	sb->st_blocks = va.va_bytes / S_BLKSIZE;
599 	return (0);
600 }
601 
602 /*
603  * File table vnode fcntl routine.
604  */
605 static int
606 vn_fcntl(struct file *fp, u_int com, void *data, struct lwp *l)
607 {
608 	struct vnode *vp = ((struct vnode *)fp->f_data);
609 	int error;
610 
611 	error = VOP_FCNTL(vp, com, data, fp->f_flag, l->l_proc->p_cred, l);
612 	return (error);
613 }
614 
615 /*
616  * File table vnode ioctl routine.
617  */
618 static int
619 vn_ioctl(struct file *fp, u_long com, void *data, struct lwp *l)
620 {
621 	struct vnode *vp = ((struct vnode *)fp->f_data);
622 	struct proc *p = l->l_proc;
623 	struct vattr vattr;
624 	int error;
625 
626 	switch (vp->v_type) {
627 
628 	case VREG:
629 	case VDIR:
630 		if (com == FIONREAD) {
631 			error = VOP_GETATTR(vp, &vattr, l->l_proc->p_cred, l);
632 			if (error)
633 				return (error);
634 			*(int *)data = vattr.va_size - fp->f_offset;
635 			return (0);
636 		}
637 		if ((com == FIONWRITE) || (com == FIONSPACE)) {
638 			/*
639 			 * Files don't have send queues, so there never
640 			 * are any bytes in them, nor is there any
641 			 * open space in them.
642 			 */
643 			*(int *)data = 0;
644 			return (0);
645 		}
646 		if (com == FIOGETBMAP) {
647 			daddr_t *block;
648 
649 			if (*(daddr_t *)data < 0)
650 				return (EINVAL);
651 			block = (daddr_t *)data;
652 			return (VOP_BMAP(vp, *block, NULL, block, NULL));
653 		}
654 		if (com == OFIOGETBMAP) {
655 			daddr_t ibn, obn;
656 
657 			if (*(int32_t *)data < 0)
658 				return (EINVAL);
659 			ibn = (daddr_t)*(int32_t *)data;
660 			error = VOP_BMAP(vp, ibn, NULL, &obn, NULL);
661 			*(int32_t *)data = (int32_t)obn;
662 			return error;
663 		}
664 		if (com == FIONBIO || com == FIOASYNC)	/* XXX */
665 			return (0);			/* XXX */
666 		/* fall into ... */
667 	case VFIFO:
668 	case VCHR:
669 	case VBLK:
670 		error = VOP_IOCTL(vp, com, data, fp->f_flag,
671 		    l->l_proc->p_cred, l);
672 		if (error == 0 && com == TIOCSCTTY) {
673 			if (p->p_session->s_ttyvp)
674 				vrele(p->p_session->s_ttyvp);
675 			p->p_session->s_ttyvp = vp;
676 			VREF(vp);
677 		}
678 		return (error);
679 
680 	default:
681 		return (EPASSTHROUGH);
682 	}
683 }
684 
685 /*
686  * File table vnode poll routine.
687  */
688 static int
689 vn_poll(struct file *fp, int events, struct lwp *l)
690 {
691 
692 	return (VOP_POLL(((struct vnode *)fp->f_data), events, l));
693 }
694 
695 /*
696  * File table vnode kqfilter routine.
697  */
698 int
699 vn_kqfilter(struct file *fp, struct knote *kn)
700 {
701 
702 	return (VOP_KQFILTER((struct vnode *)fp->f_data, kn));
703 }
704 
705 /*
706  * Check that the vnode is still valid, and if so
707  * acquire requested lock.
708  */
709 int
710 vn_lock(struct vnode *vp, int flags)
711 {
712 	int error;
713 
714 #if 0
715 	KASSERT(vp->v_usecount > 0 || (flags & LK_INTERLOCK) != 0
716 	    || (vp->v_flag & VONWORKLST) != 0);
717 #endif
718 	KASSERT((flags &
719 	    ~(LK_INTERLOCK|LK_SHARED|LK_EXCLUSIVE|LK_DRAIN|LK_NOWAIT|LK_RETRY|
720 	    LK_SETRECURSE|LK_CANRECURSE))
721 	    == 0);
722 
723 	do {
724 		if ((flags & LK_INTERLOCK) == 0)
725 			simple_lock(&vp->v_interlock);
726 		if (vp->v_flag & VXLOCK) {
727 			if (flags & LK_NOWAIT) {
728 				simple_unlock(&vp->v_interlock);
729 				return EBUSY;
730 			}
731 			vp->v_flag |= VXWANT;
732 			ltsleep(vp, PINOD | PNORELOCK,
733 			    "vn_lock", 0, &vp->v_interlock);
734 			error = ENOENT;
735 		} else {
736 			error = VOP_LOCK(vp,
737 			    (flags & ~LK_RETRY) | LK_INTERLOCK);
738 			if (error == 0 || error == EDEADLK || error == EBUSY)
739 				return (error);
740 		}
741 		flags &= ~LK_INTERLOCK;
742 	} while (flags & LK_RETRY);
743 	return (error);
744 }
745 
746 /*
747  * File table vnode close routine.
748  */
749 static int
750 vn_closefile(struct file *fp, struct lwp *l)
751 {
752 
753 	return (vn_close(((struct vnode *)fp->f_data), fp->f_flag,
754 		fp->f_cred, l));
755 }
756 
757 /*
758  * Enable LK_CANRECURSE on lock. Return prior status.
759  */
760 u_int
761 vn_setrecurse(struct vnode *vp)
762 {
763 	struct lock *lkp = &vp->v_lock;
764 	u_int retval = lkp->lk_flags & LK_CANRECURSE;
765 
766 	lkp->lk_flags |= LK_CANRECURSE;
767 	return retval;
768 }
769 
770 /*
771  * Called when done with locksetrecurse.
772  */
773 void
774 vn_restorerecurse(struct vnode *vp, u_int flags)
775 {
776 	struct lock *lkp = &vp->v_lock;
777 
778 	lkp->lk_flags &= ~LK_CANRECURSE;
779 	lkp->lk_flags |= flags;
780 }
781 
782 int
783 vn_cow_establish(struct vnode *vp,
784     int (*func)(void *, struct buf *), void *cookie)
785 {
786 	int s;
787 	struct spec_cow_entry *e;
788 
789 	MALLOC(e, struct spec_cow_entry *, sizeof(struct spec_cow_entry),
790 	    M_DEVBUF, M_WAITOK);
791 	e->ce_func = func;
792 	e->ce_cookie = cookie;
793 
794 	SPEC_COW_LOCK(vp->v_specinfo, s);
795 	vp->v_spec_cow_req++;
796 	while (vp->v_spec_cow_count > 0)
797 		ltsleep(&vp->v_spec_cow_req, PRIBIO, "cowlist", 0,
798 		    &vp->v_spec_cow_slock);
799 
800 	SLIST_INSERT_HEAD(&vp->v_spec_cow_head, e, ce_list);
801 
802 	vp->v_spec_cow_req--;
803 	if (vp->v_spec_cow_req == 0)
804 		wakeup(&vp->v_spec_cow_req);
805 	SPEC_COW_UNLOCK(vp->v_specinfo, s);
806 
807 	return 0;
808 }
809 
810 int
811 vn_cow_disestablish(struct vnode *vp,
812     int (*func)(void *, struct buf *), void *cookie)
813 {
814 	int s;
815 	struct spec_cow_entry *e;
816 
817 	SPEC_COW_LOCK(vp->v_specinfo, s);
818 	vp->v_spec_cow_req++;
819 	while (vp->v_spec_cow_count > 0)
820 		ltsleep(&vp->v_spec_cow_req, PRIBIO, "cowlist", 0,
821 		    &vp->v_spec_cow_slock);
822 
823 	SLIST_FOREACH(e, &vp->v_spec_cow_head, ce_list)
824 		if (e->ce_func == func && e->ce_cookie == cookie) {
825 			SLIST_REMOVE(&vp->v_spec_cow_head, e,
826 			    spec_cow_entry, ce_list);
827 			FREE(e, M_DEVBUF);
828 			break;
829 		}
830 
831 	vp->v_spec_cow_req--;
832 	if (vp->v_spec_cow_req == 0)
833 		wakeup(&vp->v_spec_cow_req);
834 	SPEC_COW_UNLOCK(vp->v_specinfo, s);
835 
836 	return e ? 0 : EINVAL;
837 }
838 
839 /*
840  * Simplified in-kernel wrapper calls for extended attribute access.
841  * Both calls pass in a NULL credential, authorizing a "kernel" access.
842  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
843  */
844 int
845 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
846     const char *attrname, size_t *buflen, void *bf, struct lwp *l)
847 {
848 	struct uio auio;
849 	struct iovec aiov;
850 	int error;
851 
852 	aiov.iov_len = *buflen;
853 	aiov.iov_base = bf;
854 
855 	auio.uio_iov = &aiov;
856 	auio.uio_iovcnt = 1;
857 	auio.uio_rw = UIO_READ;
858 	auio.uio_offset = 0;
859 	auio.uio_resid = *buflen;
860 	UIO_SETUP_SYSSPACE(&auio);
861 
862 	if ((ioflg & IO_NODELOCKED) == 0)
863 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
864 
865 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
866 	    l);
867 
868 	if ((ioflg & IO_NODELOCKED) == 0)
869 		VOP_UNLOCK(vp, 0);
870 
871 	if (error == 0)
872 		*buflen = *buflen - auio.uio_resid;
873 
874 	return (error);
875 }
876 
877 /*
878  * XXX Failure mode if partially written?
879  */
880 int
881 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
882     const char *attrname, size_t buflen, const void *bf, struct lwp *l)
883 {
884 	struct uio auio;
885 	struct iovec aiov;
886 	struct mount *mp = NULL;	/* XXX: GCC */
887 	int error;
888 
889 	aiov.iov_len = buflen;
890 	aiov.iov_base = __UNCONST(bf);		/* XXXUNCONST kills const */
891 
892 	auio.uio_iov = &aiov;
893 	auio.uio_iovcnt = 1;
894 	auio.uio_rw = UIO_WRITE;
895 	auio.uio_offset = 0;
896 	auio.uio_resid = buflen;
897 	UIO_SETUP_SYSSPACE(&auio);
898 
899 	if ((ioflg & IO_NODELOCKED) == 0) {
900 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
901 			return (error);
902 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
903 	}
904 
905 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, l);
906 
907 	if ((ioflg & IO_NODELOCKED) == 0) {
908 		vn_finished_write(mp, 0);
909 		VOP_UNLOCK(vp, 0);
910 	}
911 
912 	return (error);
913 }
914 
915 int
916 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
917     const char *attrname, struct lwp *l)
918 {
919 	struct mount *mp = NULL;	/* XXX: GCC */
920 	int error;
921 
922 	if ((ioflg & IO_NODELOCKED) == 0) {
923 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
924 			return (error);
925 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
926 	}
927 
928 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, l);
929 	if (error == EOPNOTSUPP)
930 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
931 		    NULL, l);
932 
933 	if ((ioflg & IO_NODELOCKED) == 0) {
934 		vn_finished_write(mp, 0);
935 		VOP_UNLOCK(vp, 0);
936 	}
937 
938 	return (error);
939 }
940 
941 /*
942  * Preparing to start a filesystem write operation. If the operation is
943  * permitted, then we bump the count of operations in progress and
944  * proceed. If a suspend request is in progress, we wait until the
945  * suspension is over, and then proceed.
946  * V_PCATCH    adds PCATCH to the tsleep flags.
947  * V_WAIT      waits until suspension is over. Otherwise returns EWOULDBLOCK.
948  * V_SLEEPONLY wait, but do not bump the operations count.
949  * V_LOWER     this is a lower level operation. No further vnodes should be
950  *             locked. Otherwise it is a upper level operation. No vnodes
951  *             should be locked.
952  */
953 int
954 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
955 {
956 	struct mount *mp;
957 	int error, mask, prio;
958 
959 	/*
960 	 * If a vnode is provided, get and return the mount point that
961 	 * to which it will write.
962 	 */
963 	if (vp != NULL) {
964 		*mpp = vp->v_mount;
965 	}
966 	if ((mp = *mpp) == NULL)
967 		return (0);
968 	mp = mp->mnt_leaf;
969 	/*
970 	 * Check on status of suspension.
971 	 */
972 	prio = PUSER - 1;
973 	if (flags & V_PCATCH)
974 		prio |= PCATCH;
975 
976 	if ((flags & V_LOWER) == 0)
977 		mask = IMNT_SUSPEND;
978 	else
979 		mask = IMNT_SUSPENDLOW;
980 
981 	while ((mp->mnt_iflag & mask) != 0) {
982 		if ((flags & V_WAIT) == 0)
983 			return (EWOULDBLOCK);
984 		error = tsleep(&mp->mnt_flag, prio, "suspfs", 0);
985 		if (error)
986 			return (error);
987 	}
988 	if (flags & V_SLEEPONLY)
989 		return (0);
990 	simple_lock(&mp->mnt_slock);
991 	if ((flags & V_LOWER) == 0)
992 		mp->mnt_writeopcountupper++;
993 	else
994 		mp->mnt_writeopcountlower++;
995 	simple_unlock(&mp->mnt_slock);
996 	return (0);
997 }
998 
999 /*
1000  * Filesystem write operation has completed. If we are suspending and this
1001  * operation is the last one, notify the suspender that the suspension is
1002  * now in effect.
1003  */
1004 void
1005 vn_finished_write(struct mount *mp, int flags)
1006 {
1007 	if (mp == NULL)
1008 		return;
1009 	mp = mp->mnt_leaf;
1010 	simple_lock(&mp->mnt_slock);
1011 	if ((flags & V_LOWER) == 0) {
1012 		mp->mnt_writeopcountupper--;
1013 		if (mp->mnt_writeopcountupper < 0)
1014 			printf("vn_finished_write: neg cnt upper=%d\n",
1015 			       mp->mnt_writeopcountupper);
1016 		if ((mp->mnt_iflag & IMNT_SUSPEND) != 0 &&
1017 		    mp->mnt_writeopcountupper <= 0)
1018 			wakeup(&mp->mnt_writeopcountupper);
1019 	} else {
1020 		mp->mnt_writeopcountlower--;
1021 		if (mp->mnt_writeopcountlower < 0)
1022 			printf("vn_finished_write: neg cnt lower=%d\n",
1023 			       mp->mnt_writeopcountlower);
1024 		if ((mp->mnt_iflag & IMNT_SUSPENDLOW) != 0 &&
1025 		    mp->mnt_writeopcountupper <= 0)
1026 			wakeup(&mp->mnt_writeopcountlower);
1027 	}
1028 	simple_unlock(&mp->mnt_slock);
1029 }
1030 
1031 void
1032 vn_ra_allocctx(struct vnode *vp)
1033 {
1034 	struct uvm_ractx *ra = NULL;
1035 
1036 	if (vp->v_type != VREG) {
1037 		return;
1038 	}
1039 	if (vp->v_ractx != NULL) {
1040 		return;
1041 	}
1042 	simple_lock(&vp->v_interlock);
1043 	if (vp->v_ractx == NULL) {
1044 		simple_unlock(&vp->v_interlock);
1045 		ra = uvm_ra_allocctx();
1046 		simple_lock(&vp->v_interlock);
1047 		if (ra != NULL && vp->v_ractx == NULL) {
1048 			vp->v_ractx = ra;
1049 			ra = NULL;
1050 		}
1051 	}
1052 	simple_unlock(&vp->v_interlock);
1053 	if (ra != NULL) {
1054 		uvm_ra_freectx(ra);
1055 	}
1056 }
1057