xref: /netbsd-src/sys/kern/vfs_vnops.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: vfs_vnops.c,v 1.102 2005/12/12 16:26:33 elad 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 1.102 2005/12/12 16:26:33 elad 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 
61 #include <miscfs/specfs/specdev.h>
62 
63 #include <uvm/uvm_extern.h>
64 #include <uvm/uvm_readahead.h>
65 
66 #ifdef UNION
67 #include <fs/union/union.h>
68 #endif
69 
70 #if defined(LKM) || defined(UNION)
71 int (*vn_union_readdir_hook) (struct vnode **, struct file *, struct lwp *);
72 #endif
73 
74 #ifdef VERIFIED_EXEC
75 #include <sys/verified_exec.h>
76 #endif
77 
78 static int vn_read(struct file *fp, off_t *offset, struct uio *uio,
79 	    struct ucred *cred, int flags);
80 static int vn_write(struct file *fp, off_t *offset, struct uio *uio,
81 	    struct ucred *cred, int flags);
82 static int vn_closefile(struct file *fp, struct lwp *l);
83 static int vn_poll(struct file *fp, int events, struct lwp *l);
84 static int vn_fcntl(struct file *fp, u_int com, void *data, struct lwp *l);
85 static int vn_statfile(struct file *fp, struct stat *sb, struct lwp *l);
86 static int vn_ioctl(struct file *fp, u_long com, void *data, struct lwp *l);
87 
88 const struct fileops vnops = {
89 	vn_read, vn_write, vn_ioctl, vn_fcntl, vn_poll,
90 	vn_statfile, vn_closefile, vn_kqfilter
91 };
92 
93 /*
94  * Common code for vnode open operations.
95  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
96  */
97 int
98 vn_open(struct nameidata *ndp, int fmode, int cmode)
99 {
100 	struct vnode *vp;
101 	struct mount *mp;
102 	struct lwp *l = ndp->ni_cnd.cn_lwp;
103 	struct ucred *cred = l->l_proc->p_ucred;
104 	struct vattr va;
105 	int error;
106 #ifdef VERIFIED_EXEC
107 	struct veriexec_hash_entry *vhe = NULL;
108 	char pathbuf[MAXPATHLEN];
109 	size_t pathlen;
110 	int (*copyfun)(const void *, void *, size_t, size_t *) =
111 	    ndp->ni_segflg == UIO_SYSSPACE ? copystr : copyinstr;
112 #endif /* VERIFIED_EXEC */
113 
114 #ifdef VERIFIED_EXEC
115 	error = (*copyfun)(ndp->ni_dirp, pathbuf, sizeof(pathbuf), &pathlen);
116 	if (error) {
117 		if (veriexec_verbose >= 1)
118 			printf("veriexec: Can't copy path. (error=%d)\n",
119 			    error);
120 
121 		return (error);
122 	}
123 #endif /* VERIFIED_EXEC */
124 
125 restart:
126 	if (fmode & O_CREAT) {
127 		ndp->ni_cnd.cn_nameiop = CREATE;
128 		ndp->ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
129 		if ((fmode & O_EXCL) == 0 &&
130 		    ((fmode & O_NOFOLLOW) == 0))
131 			ndp->ni_cnd.cn_flags |= FOLLOW;
132 		if ((error = namei(ndp)) != 0)
133 			return (error);
134 		if (ndp->ni_vp == NULL) {
135 #ifdef VERIFIED_EXEC
136 			/* Lockdown mode: Prevent creation of new files. */
137 			if (veriexec_strict >= 3) {
138 				VOP_ABORTOP(ndp->ni_dvp, &ndp->ni_cnd);
139 
140 				printf("Veriexec: vn_open: Preventing "
141 				       "new file creation in %s.\n",
142 				       pathbuf);
143 
144 				vp = ndp->ni_dvp;
145 				error = EPERM;
146 				goto bad;
147 			}
148 #endif /* VERIFIED_EXEC */
149 
150 			VATTR_NULL(&va);
151 			va.va_type = VREG;
152 			va.va_mode = cmode;
153 			if (fmode & O_EXCL)
154 				 va.va_vaflags |= VA_EXCLUSIVE;
155 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
156 				VOP_ABORTOP(ndp->ni_dvp, &ndp->ni_cnd);
157 				vput(ndp->ni_dvp);
158 				if ((error = vn_start_write(NULL, &mp,
159 				    V_WAIT | V_SLEEPONLY | V_PCATCH)) != 0)
160 					return (error);
161 				goto restart;
162 			}
163 			VOP_LEASE(ndp->ni_dvp, l, cred, LEASE_WRITE);
164 			error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
165 					   &ndp->ni_cnd, &va);
166 			vn_finished_write(mp, 0);
167 			if (error)
168 				return (error);
169 			fmode &= ~O_TRUNC;
170 			vp = ndp->ni_vp;
171 		} else {
172 			VOP_ABORTOP(ndp->ni_dvp, &ndp->ni_cnd);
173 			if (ndp->ni_dvp == ndp->ni_vp)
174 				vrele(ndp->ni_dvp);
175 			else
176 				vput(ndp->ni_dvp);
177 			ndp->ni_dvp = NULL;
178 			vp = ndp->ni_vp;
179 			if (fmode & O_EXCL) {
180 				error = EEXIST;
181 				goto bad;
182 			}
183 			fmode &= ~O_CREAT;
184 		}
185 	} else {
186 		ndp->ni_cnd.cn_nameiop = LOOKUP;
187 		ndp->ni_cnd.cn_flags = LOCKLEAF;
188 		if ((fmode & O_NOFOLLOW) == 0)
189 			ndp->ni_cnd.cn_flags |= FOLLOW;
190 		if ((error = namei(ndp)) != 0)
191 			return (error);
192 		vp = ndp->ni_vp;
193 	}
194 	if (vp->v_type == VSOCK) {
195 		error = EOPNOTSUPP;
196 		goto bad;
197 	}
198 	if (ndp->ni_vp->v_type == VLNK) {
199 		error = EFTYPE;
200 		goto bad;
201 	}
202 
203 #ifdef VERIFIED_EXEC
204 	if ((error = VOP_GETATTR(vp, &va, cred, l)) != 0)
205 		goto bad;
206 #endif
207 
208 	if ((fmode & O_CREAT) == 0) {
209 #ifdef VERIFIED_EXEC
210 		if ((error = veriexec_verify(l, vp, &va, pathbuf,
211 					     VERIEXEC_FILE, &vhe)) != 0)
212 			goto bad;
213 #endif
214 
215 		if (fmode & FREAD) {
216 			if ((error = VOP_ACCESS(vp, VREAD, cred, l)) != 0)
217 				goto bad;
218 		}
219 
220 		if (fmode & (FWRITE | O_TRUNC)) {
221 			if (vp->v_type == VDIR) {
222 				error = EISDIR;
223 				goto bad;
224 			}
225 			if ((error = vn_writechk(vp)) != 0 ||
226 			    (error = VOP_ACCESS(vp, VWRITE, cred, l)) != 0)
227 				goto bad;
228 #ifdef VERIFIED_EXEC
229 			if (vhe != NULL) {
230 				veriexec_report("Write access request.",
231 						pathbuf, &va, l,
232 						REPORT_NOVERBOSE,
233 						REPORT_ALARM,
234 						REPORT_NOPANIC);
235 
236 				/* IPS mode: Deny writing to monitored files. */
237 				if (veriexec_strict >= 2) {
238 					error = EPERM;
239 					goto bad;
240 				} else {
241 					vhe->status = FINGERPRINT_NOTEVAL;
242 				}
243 			}
244 #endif
245 		}
246 	}
247 
248 	if (fmode & O_TRUNC) {
249 		VOP_UNLOCK(vp, 0);			/* XXX */
250 		if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) {
251 			vrele(vp);
252 			return (error);
253 		}
254 		VOP_LEASE(vp, l, cred, LEASE_WRITE);
255 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
256 		VATTR_NULL(&va);
257 		va.va_size = 0;
258 		error = VOP_SETATTR(vp, &va, cred, l);
259 		vn_finished_write(mp, 0);
260 		if (error != 0)
261 			goto bad;
262 	}
263 	if ((error = VOP_OPEN(vp, fmode, cred, l)) != 0)
264 		goto bad;
265 	if (vp->v_type == VREG &&
266 	    uvn_attach(vp, fmode & FWRITE ? VM_PROT_WRITE : 0) == NULL) {
267 		error = EIO;
268 		goto bad;
269 	}
270 	if (fmode & FWRITE)
271 		vp->v_writecount++;
272 
273 	return (0);
274 bad:
275 	vput(vp);
276 	return (error);
277 }
278 
279 /*
280  * Check for write permissions on the specified vnode.
281  * Prototype text segments cannot be written.
282  */
283 int
284 vn_writechk(struct vnode *vp)
285 {
286 
287 	/*
288 	 * If the vnode is in use as a process's text,
289 	 * we can't allow writing.
290 	 */
291 	if (vp->v_flag & VTEXT)
292 		return (ETXTBSY);
293 	return (0);
294 }
295 
296 /*
297  * Mark a vnode as having executable mappings.
298  */
299 void
300 vn_markexec(struct vnode *vp)
301 {
302 	if ((vp->v_flag & VEXECMAP) == 0) {
303 		uvmexp.filepages -= vp->v_uobj.uo_npages;
304 		uvmexp.execpages += vp->v_uobj.uo_npages;
305 	}
306 	vp->v_flag |= VEXECMAP;
307 }
308 
309 /*
310  * Mark a vnode as being the text of a process.
311  * Fail if the vnode is currently writable.
312  */
313 int
314 vn_marktext(struct vnode *vp)
315 {
316 
317 	if (vp->v_writecount != 0) {
318 		KASSERT((vp->v_flag & VTEXT) == 0);
319 		return (ETXTBSY);
320 	}
321 	vp->v_flag |= VTEXT;
322 	vn_markexec(vp);
323 	return (0);
324 }
325 
326 /*
327  * Vnode close call
328  *
329  * Note: takes an unlocked vnode, while VOP_CLOSE takes a locked node.
330  */
331 int
332 vn_close(struct vnode *vp, int flags, struct ucred *cred, struct lwp *l)
333 {
334 	int error;
335 
336 	if (flags & FWRITE)
337 		vp->v_writecount--;
338 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
339 	error = VOP_CLOSE(vp, flags, cred, l);
340 	vput(vp);
341 	return (error);
342 }
343 
344 /*
345  * Package up an I/O request on a vnode into a uio and do it.
346  */
347 int
348 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, off_t offset,
349     enum uio_seg segflg, int ioflg, struct ucred *cred, size_t *aresid,
350     struct lwp *l)
351 {
352 	struct uio auio;
353 	struct iovec aiov;
354 	struct mount *mp;
355 	int error;
356 
357 	if ((ioflg & IO_NODELOCKED) == 0) {
358 		if (rw == UIO_READ) {
359 			vn_lock(vp, LK_SHARED | LK_RETRY);
360 		} else /* UIO_WRITE */ {
361 			if (vp->v_type != VCHR &&
362 			    (error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH))
363 			    != 0)
364 				return (error);
365 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
366 		}
367 	}
368 	auio.uio_iov = &aiov;
369 	auio.uio_iovcnt = 1;
370 	aiov.iov_base = base;
371 	aiov.iov_len = len;
372 	auio.uio_resid = len;
373 	auio.uio_offset = offset;
374 	auio.uio_segflg = segflg;
375 	auio.uio_rw = rw;
376 	auio.uio_lwp = l;
377 	if (rw == UIO_READ) {
378 		error = VOP_READ(vp, &auio, ioflg, cred);
379 	} else {
380 		error = VOP_WRITE(vp, &auio, ioflg, cred);
381 	}
382 	if (aresid)
383 		*aresid = auio.uio_resid;
384 	else
385 		if (auio.uio_resid && error == 0)
386 			error = EIO;
387 	if ((ioflg & IO_NODELOCKED) == 0) {
388 		if (rw == UIO_WRITE)
389 			vn_finished_write(mp, 0);
390 		VOP_UNLOCK(vp, 0);
391 	}
392 	return (error);
393 }
394 
395 int
396 vn_readdir(struct file *fp, char *bf, int segflg, u_int count, int *done,
397     struct lwp *l, off_t **cookies, int *ncookies)
398 {
399 	struct vnode *vp = (struct vnode *)fp->f_data;
400 	struct iovec aiov;
401 	struct uio auio;
402 	int error, eofflag;
403 
404 unionread:
405 	if (vp->v_type != VDIR)
406 		return (EINVAL);
407 	aiov.iov_base = bf;
408 	aiov.iov_len = count;
409 	auio.uio_iov = &aiov;
410 	auio.uio_iovcnt = 1;
411 	auio.uio_rw = UIO_READ;
412 	auio.uio_segflg = segflg;
413 	auio.uio_lwp = l;
414 	auio.uio_resid = count;
415 	vn_lock(vp, LK_SHARED | LK_RETRY);
416 	auio.uio_offset = fp->f_offset;
417 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, cookies,
418 		    ncookies);
419 	fp->f_offset = auio.uio_offset;
420 	VOP_UNLOCK(vp, 0);
421 	if (error)
422 		return (error);
423 
424 #if defined(UNION) || defined(LKM)
425 	if (count == auio.uio_resid && vn_union_readdir_hook) {
426 		struct vnode *ovp = vp;
427 
428 		error = (*vn_union_readdir_hook)(&vp, fp, l);
429 		if (error)
430 			return (error);
431 		if (vp != ovp)
432 			goto unionread;
433 	}
434 #endif /* UNION || LKM */
435 
436 	if (count == auio.uio_resid && (vp->v_flag & VROOT) &&
437 	    (vp->v_mount->mnt_flag & MNT_UNION)) {
438 		struct vnode *tvp = vp;
439 		vp = vp->v_mount->mnt_vnodecovered;
440 		VREF(vp);
441 		fp->f_data = vp;
442 		fp->f_offset = 0;
443 		vrele(tvp);
444 		goto unionread;
445 	}
446 	*done = count - auio.uio_resid;
447 	return error;
448 }
449 
450 /*
451  * File table vnode read routine.
452  */
453 static int
454 vn_read(struct file *fp, off_t *offset, struct uio *uio, struct ucred *cred,
455     int flags)
456 {
457 	struct vnode *vp = (struct vnode *)fp->f_data;
458 	int count, error, ioflag;
459 
460 	VOP_LEASE(vp, uio->uio_lwp, cred, LEASE_READ);
461 	ioflag = IO_ADV_ENCODE(fp->f_advice);
462 	if (fp->f_flag & FNONBLOCK)
463 		ioflag |= IO_NDELAY;
464 	if ((fp->f_flag & (FFSYNC | FRSYNC)) == (FFSYNC | FRSYNC))
465 		ioflag |= IO_SYNC;
466 	if (fp->f_flag & FALTIO)
467 		ioflag |= IO_ALTSEMANTICS;
468 	vn_lock(vp, LK_SHARED | LK_RETRY);
469 	vn_ra_allocctx(vp);
470 	uio->uio_offset = *offset;
471 	count = uio->uio_resid;
472 	error = VOP_READ(vp, uio, ioflag, cred);
473 	if (flags & FOF_UPDATE_OFFSET)
474 		*offset += count - uio->uio_resid;
475 	VOP_UNLOCK(vp, 0);
476 	return (error);
477 }
478 
479 /*
480  * File table vnode write routine.
481  */
482 static int
483 vn_write(struct file *fp, off_t *offset, struct uio *uio, struct ucred *cred,
484     int flags)
485 {
486 	struct vnode *vp = (struct vnode *)fp->f_data;
487 	struct mount *mp;
488 	int count, error, ioflag = IO_UNIT;
489 
490 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
491 		ioflag |= IO_APPEND;
492 	if (fp->f_flag & FNONBLOCK)
493 		ioflag |= IO_NDELAY;
494 	if (fp->f_flag & FFSYNC ||
495 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
496 		ioflag |= IO_SYNC;
497 	else if (fp->f_flag & FDSYNC)
498 		ioflag |= IO_DSYNC;
499 	if (fp->f_flag & FALTIO)
500 		ioflag |= IO_ALTSEMANTICS;
501 	mp = NULL;
502 	if (vp->v_type != VCHR &&
503 	    (error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0)
504 		return (error);
505 	VOP_LEASE(vp, uio->uio_lwp, cred, LEASE_WRITE);
506 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
507 	uio->uio_offset = *offset;
508 	count = uio->uio_resid;
509 	error = VOP_WRITE(vp, uio, ioflag, cred);
510 	if (flags & FOF_UPDATE_OFFSET) {
511 		if (ioflag & IO_APPEND)
512 			*offset = uio->uio_offset;
513 		else
514 			*offset += count - uio->uio_resid;
515 	}
516 	VOP_UNLOCK(vp, 0);
517 	vn_finished_write(mp, 0);
518 	return (error);
519 }
520 
521 /*
522  * File table vnode stat routine.
523  */
524 static int
525 vn_statfile(struct file *fp, struct stat *sb, struct lwp *l)
526 {
527 	struct vnode *vp = (struct vnode *)fp->f_data;
528 
529 	return vn_stat(vp, sb, l);
530 }
531 
532 int
533 vn_stat(struct vnode *vp, struct stat *sb, struct lwp *l)
534 {
535 	struct vattr va;
536 	int error;
537 	mode_t mode;
538 
539 	error = VOP_GETATTR(vp, &va, l->l_proc->p_ucred, l);
540 	if (error)
541 		return (error);
542 	/*
543 	 * Copy from vattr table
544 	 */
545 	sb->st_dev = va.va_fsid;
546 	sb->st_ino = va.va_fileid;
547 	mode = va.va_mode;
548 	switch (vp->v_type) {
549 	case VREG:
550 		mode |= S_IFREG;
551 		break;
552 	case VDIR:
553 		mode |= S_IFDIR;
554 		break;
555 	case VBLK:
556 		mode |= S_IFBLK;
557 		break;
558 	case VCHR:
559 		mode |= S_IFCHR;
560 		break;
561 	case VLNK:
562 		mode |= S_IFLNK;
563 		break;
564 	case VSOCK:
565 		mode |= S_IFSOCK;
566 		break;
567 	case VFIFO:
568 		mode |= S_IFIFO;
569 		break;
570 	default:
571 		return (EBADF);
572 	};
573 	sb->st_mode = mode;
574 	sb->st_nlink = va.va_nlink;
575 	sb->st_uid = va.va_uid;
576 	sb->st_gid = va.va_gid;
577 	sb->st_rdev = va.va_rdev;
578 	sb->st_size = va.va_size;
579 	sb->st_atimespec = va.va_atime;
580 	sb->st_mtimespec = va.va_mtime;
581 	sb->st_ctimespec = va.va_ctime;
582 	sb->st_birthtimespec = va.va_birthtime;
583 	sb->st_blksize = va.va_blocksize;
584 	sb->st_flags = va.va_flags;
585 	sb->st_gen = 0;
586 	sb->st_blocks = va.va_bytes / S_BLKSIZE;
587 	return (0);
588 }
589 
590 /*
591  * File table vnode fcntl routine.
592  */
593 static int
594 vn_fcntl(struct file *fp, u_int com, void *data, struct lwp *l)
595 {
596 	struct vnode *vp = ((struct vnode *)fp->f_data);
597 	int error;
598 
599 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
600 	error = VOP_FCNTL(vp, com, data, fp->f_flag, l->l_proc->p_ucred, l);
601 	VOP_UNLOCK(vp, 0);
602 	return (error);
603 }
604 
605 /*
606  * File table vnode ioctl routine.
607  */
608 static int
609 vn_ioctl(struct file *fp, u_long com, void *data, struct lwp *l)
610 {
611 	struct vnode *vp = ((struct vnode *)fp->f_data);
612 	struct proc *p = l->l_proc;
613 	struct vattr vattr;
614 	int error;
615 
616 	switch (vp->v_type) {
617 
618 	case VREG:
619 	case VDIR:
620 		if (com == FIONREAD) {
621 			error = VOP_GETATTR(vp, &vattr, l->l_proc->p_ucred, l);
622 			if (error)
623 				return (error);
624 			*(int *)data = vattr.va_size - fp->f_offset;
625 			return (0);
626 		}
627 		if ((com == FIONWRITE) || (com == FIONSPACE)) {
628 			/*
629 			 * Files don't have send queues, so there never
630 			 * are any bytes in them, nor is there any
631 			 * open space in them.
632 			 */
633 			*(int *)data = 0;
634 			return (0);
635 		}
636 		if (com == FIOGETBMAP) {
637 			daddr_t *block;
638 
639 			if (*(daddr_t *)data < 0)
640 				return (EINVAL);
641 			block = (daddr_t *)data;
642 			return (VOP_BMAP(vp, *block, NULL, block, NULL));
643 		}
644 		if (com == OFIOGETBMAP) {
645 			daddr_t ibn, obn;
646 
647 			if (*(int32_t *)data < 0)
648 				return (EINVAL);
649 			ibn = (daddr_t)*(int32_t *)data;
650 			error = VOP_BMAP(vp, ibn, NULL, &obn, NULL);
651 			*(int32_t *)data = (int32_t)obn;
652 			return error;
653 		}
654 		if (com == FIONBIO || com == FIOASYNC)	/* XXX */
655 			return (0);			/* XXX */
656 		/* fall into ... */
657 	case VFIFO:
658 	case VCHR:
659 	case VBLK:
660 		error = VOP_IOCTL(vp, com, data, fp->f_flag,
661 		    l->l_proc->p_ucred, l);
662 		if (error == 0 && com == TIOCSCTTY) {
663 			if (p->p_session->s_ttyvp)
664 				vrele(p->p_session->s_ttyvp);
665 			p->p_session->s_ttyvp = vp;
666 			VREF(vp);
667 		}
668 		return (error);
669 
670 	default:
671 		return (EPASSTHROUGH);
672 	}
673 }
674 
675 /*
676  * File table vnode poll routine.
677  */
678 static int
679 vn_poll(struct file *fp, int events, struct lwp *l)
680 {
681 
682 	return (VOP_POLL(((struct vnode *)fp->f_data), events, l));
683 }
684 
685 /*
686  * File table vnode kqfilter routine.
687  */
688 int
689 vn_kqfilter(struct file *fp, struct knote *kn)
690 {
691 
692 	return (VOP_KQFILTER((struct vnode *)fp->f_data, kn));
693 }
694 
695 /*
696  * Check that the vnode is still valid, and if so
697  * acquire requested lock.
698  */
699 int
700 vn_lock(struct vnode *vp, int flags)
701 {
702 	int error;
703 
704 #if 0
705 	KASSERT(vp->v_usecount > 0 || (flags & LK_INTERLOCK) != 0
706 	    || (vp->v_flag & VONWORKLST) != 0);
707 #endif
708 
709 	do {
710 		if ((flags & LK_INTERLOCK) == 0)
711 			simple_lock(&vp->v_interlock);
712 		if (vp->v_flag & VXLOCK) {
713 			if (flags & LK_NOWAIT) {
714 				simple_unlock(&vp->v_interlock);
715 				return EBUSY;
716 			}
717 			vp->v_flag |= VXWANT;
718 			ltsleep(vp, PINOD | PNORELOCK,
719 			    "vn_lock", 0, &vp->v_interlock);
720 			error = ENOENT;
721 		} else {
722 			error = VOP_LOCK(vp,
723 			    (flags & ~LK_RETRY) | LK_INTERLOCK);
724 			if (error == 0 || error == EDEADLK || error == EBUSY)
725 				return (error);
726 		}
727 		flags &= ~LK_INTERLOCK;
728 	} while (flags & LK_RETRY);
729 	return (error);
730 }
731 
732 /*
733  * File table vnode close routine.
734  */
735 static int
736 vn_closefile(struct file *fp, struct lwp *l)
737 {
738 
739 	return (vn_close(((struct vnode *)fp->f_data), fp->f_flag,
740 		fp->f_cred, l));
741 }
742 
743 /*
744  * Enable LK_CANRECURSE on lock. Return prior status.
745  */
746 u_int
747 vn_setrecurse(struct vnode *vp)
748 {
749 	struct lock *lkp = &vp->v_lock;
750 	u_int retval = lkp->lk_flags & LK_CANRECURSE;
751 
752 	lkp->lk_flags |= LK_CANRECURSE;
753 	return retval;
754 }
755 
756 /*
757  * Called when done with locksetrecurse.
758  */
759 void
760 vn_restorerecurse(struct vnode *vp, u_int flags)
761 {
762 	struct lock *lkp = &vp->v_lock;
763 
764 	lkp->lk_flags &= ~LK_CANRECURSE;
765 	lkp->lk_flags |= flags;
766 }
767 
768 int
769 vn_cow_establish(struct vnode *vp,
770     int (*func)(void *, struct buf *), void *cookie)
771 {
772 	int s;
773 	struct spec_cow_entry *e;
774 
775 	MALLOC(e, struct spec_cow_entry *, sizeof(struct spec_cow_entry),
776 	    M_DEVBUF, M_WAITOK);
777 	e->ce_func = func;
778 	e->ce_cookie = cookie;
779 
780 	SPEC_COW_LOCK(vp->v_specinfo, s);
781 	vp->v_spec_cow_req++;
782 	while (vp->v_spec_cow_count > 0)
783 		ltsleep(&vp->v_spec_cow_req, PRIBIO, "cowlist", 0,
784 		    &vp->v_spec_cow_slock);
785 
786 	SLIST_INSERT_HEAD(&vp->v_spec_cow_head, e, ce_list);
787 
788 	vp->v_spec_cow_req--;
789 	if (vp->v_spec_cow_req == 0)
790 		wakeup(&vp->v_spec_cow_req);
791 	SPEC_COW_UNLOCK(vp->v_specinfo, s);
792 
793 	return 0;
794 }
795 
796 int
797 vn_cow_disestablish(struct vnode *vp,
798     int (*func)(void *, struct buf *), void *cookie)
799 {
800 	int s;
801 	struct spec_cow_entry *e;
802 
803 	SPEC_COW_LOCK(vp->v_specinfo, s);
804 	vp->v_spec_cow_req++;
805 	while (vp->v_spec_cow_count > 0)
806 		ltsleep(&vp->v_spec_cow_req, PRIBIO, "cowlist", 0,
807 		    &vp->v_spec_cow_slock);
808 
809 	SLIST_FOREACH(e, &vp->v_spec_cow_head, ce_list)
810 		if (e->ce_func == func && e->ce_cookie == cookie) {
811 			SLIST_REMOVE(&vp->v_spec_cow_head, e,
812 			    spec_cow_entry, ce_list);
813 			FREE(e, M_DEVBUF);
814 			break;
815 		}
816 
817 	vp->v_spec_cow_req--;
818 	if (vp->v_spec_cow_req == 0)
819 		wakeup(&vp->v_spec_cow_req);
820 	SPEC_COW_UNLOCK(vp->v_specinfo, s);
821 
822 	return e ? 0 : EINVAL;
823 }
824 
825 /*
826  * Simplified in-kernel wrapper calls for extended attribute access.
827  * Both calls pass in a NULL credential, authorizing a "kernel" access.
828  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
829  */
830 int
831 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
832     const char *attrname, size_t *buflen, void *bf, struct lwp *l)
833 {
834 	struct uio auio;
835 	struct iovec aiov;
836 	int error;
837 
838 	aiov.iov_len = *buflen;
839 	aiov.iov_base = bf;
840 
841 	auio.uio_iov = &aiov;
842 	auio.uio_iovcnt = 1;
843 	auio.uio_rw = UIO_READ;
844 	auio.uio_segflg = UIO_SYSSPACE;
845 	auio.uio_lwp = l;
846 	auio.uio_offset = 0;
847 	auio.uio_resid = *buflen;
848 
849 	if ((ioflg & IO_NODELOCKED) == 0)
850 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
851 
852 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
853 	    l);
854 
855 	if ((ioflg & IO_NODELOCKED) == 0)
856 		VOP_UNLOCK(vp, 0);
857 
858 	if (error == 0)
859 		*buflen = *buflen - auio.uio_resid;
860 
861 	return (error);
862 }
863 
864 /*
865  * XXX Failure mode if partially written?
866  */
867 int
868 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
869     const char *attrname, size_t buflen, const void *bf, struct lwp *l)
870 {
871 	struct uio auio;
872 	struct iovec aiov;
873 	struct mount *mp;
874 	int error;
875 
876 	aiov.iov_len = buflen;
877 	aiov.iov_base = __UNCONST(bf);		/* XXXUNCONST kills const */
878 
879 	auio.uio_iov = &aiov;
880 	auio.uio_iovcnt = 1;
881 	auio.uio_rw = UIO_WRITE;
882 	auio.uio_segflg = UIO_SYSSPACE;
883 	auio.uio_lwp = l;
884 	auio.uio_offset = 0;
885 	auio.uio_resid = buflen;
886 
887 	if ((ioflg & IO_NODELOCKED) == 0) {
888 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
889 			return (error);
890 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
891 	}
892 
893 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, l);
894 
895 	if ((ioflg & IO_NODELOCKED) == 0) {
896 		vn_finished_write(mp, 0);
897 		VOP_UNLOCK(vp, 0);
898 	}
899 
900 	return (error);
901 }
902 
903 int
904 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
905     const char *attrname, struct lwp *l)
906 {
907 	struct mount *mp;
908 	int error;
909 
910 	if ((ioflg & IO_NODELOCKED) == 0) {
911 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
912 			return (error);
913 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
914 	}
915 
916 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, l);
917 	if (error == EOPNOTSUPP)
918 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
919 		    NULL, l);
920 
921 	if ((ioflg & IO_NODELOCKED) == 0) {
922 		vn_finished_write(mp, 0);
923 		VOP_UNLOCK(vp, 0);
924 	}
925 
926 	return (error);
927 }
928 
929 /*
930  * Preparing to start a filesystem write operation. If the operation is
931  * permitted, then we bump the count of operations in progress and
932  * proceed. If a suspend request is in progress, we wait until the
933  * suspension is over, and then proceed.
934  * V_PCATCH    adds PCATCH to the tsleep flags.
935  * V_WAIT      waits until suspension is over. Otherwise returns EWOULDBLOCK.
936  * V_SLEEPONLY wait, but do not bump the operations count.
937  * V_LOWER     this is a lower level operation. No further vnodes should be
938  *             locked. Otherwise it is a upper level operation. No vnodes
939  *             should be locked.
940  */
941 int
942 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
943 {
944 	struct mount *mp;
945 	int error, mask, prio;
946 
947 	/*
948 	 * If a vnode is provided, get and return the mount point that
949 	 * to which it will write.
950 	 */
951 	if (vp != NULL) {
952 		*mpp = vp->v_mount;
953 	}
954 	if ((mp = *mpp) == NULL)
955 		return (0);
956 	mp = mp->mnt_leaf;
957 	/*
958 	 * Check on status of suspension.
959 	 */
960 	prio = PUSER - 1;
961 	if (flags & V_PCATCH)
962 		prio |= PCATCH;
963 
964 	if ((flags & V_LOWER) == 0)
965 		mask = IMNT_SUSPEND;
966 	else
967 		mask = IMNT_SUSPENDLOW;
968 
969 	while ((mp->mnt_iflag & mask) != 0) {
970 		if ((flags & V_WAIT) == 0)
971 			return (EWOULDBLOCK);
972 		error = tsleep(&mp->mnt_flag, prio, "suspfs", 0);
973 		if (error)
974 			return (error);
975 	}
976 	if (flags & V_SLEEPONLY)
977 		return (0);
978 	simple_lock(&mp->mnt_slock);
979 	if ((flags & V_LOWER) == 0)
980 		mp->mnt_writeopcountupper++;
981 	else
982 		mp->mnt_writeopcountlower++;
983 	simple_unlock(&mp->mnt_slock);
984 	return (0);
985 }
986 
987 /*
988  * Filesystem write operation has completed. If we are suspending and this
989  * operation is the last one, notify the suspender that the suspension is
990  * now in effect.
991  */
992 void
993 vn_finished_write(struct mount *mp, int flags)
994 {
995 	if (mp == NULL)
996 		return;
997 	mp = mp->mnt_leaf;
998 	simple_lock(&mp->mnt_slock);
999 	if ((flags & V_LOWER) == 0) {
1000 		mp->mnt_writeopcountupper--;
1001 		if (mp->mnt_writeopcountupper < 0)
1002 			printf("vn_finished_write: neg cnt upper=%d\n",
1003 			       mp->mnt_writeopcountupper);
1004 		if ((mp->mnt_iflag & IMNT_SUSPEND) != 0 &&
1005 		    mp->mnt_writeopcountupper <= 0)
1006 			wakeup(&mp->mnt_writeopcountupper);
1007 	} else {
1008 		mp->mnt_writeopcountlower--;
1009 		if (mp->mnt_writeopcountlower < 0)
1010 			printf("vn_finished_write: neg cnt lower=%d\n",
1011 			       mp->mnt_writeopcountlower);
1012 		if ((mp->mnt_iflag & IMNT_SUSPENDLOW) != 0 &&
1013 		    mp->mnt_writeopcountupper <= 0)
1014 			wakeup(&mp->mnt_writeopcountlower);
1015 	}
1016 	simple_unlock(&mp->mnt_slock);
1017 }
1018 
1019 void
1020 vn_ra_allocctx(struct vnode *vp)
1021 {
1022 	struct uvm_ractx *ra = NULL;
1023 
1024 	if (vp->v_type != VREG) {
1025 		return;
1026 	}
1027 	if (vp->v_ractx != NULL) {
1028 		return;
1029 	}
1030 	simple_lock(&vp->v_interlock);
1031 	if (vp->v_ractx == NULL) {
1032 		simple_unlock(&vp->v_interlock);
1033 		ra = uvm_ra_allocctx();
1034 		simple_lock(&vp->v_interlock);
1035 		if (ra != NULL && vp->v_ractx == NULL) {
1036 			vp->v_ractx = ra;
1037 			ra = NULL;
1038 		}
1039 	}
1040 	simple_unlock(&vp->v_interlock);
1041 	if (ra != NULL) {
1042 		uvm_ra_freectx(ra);
1043 	}
1044 }
1045