xref: /freebsd/sys/kern/vfs_vnops.c (revision c98323078dede7579020518ec84cdcb478e5c142)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_vnops.c	8.2 (Berkeley) 1/21/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_mac.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/fcntl.h>
45 #include <sys/file.h>
46 #include <sys/kdb.h>
47 #include <sys/stat.h>
48 #include <sys/proc.h>
49 #include <sys/limits.h>
50 #include <sys/lock.h>
51 #include <sys/mac.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
55 #include <sys/vnode.h>
56 #include <sys/bio.h>
57 #include <sys/buf.h>
58 #include <sys/filio.h>
59 #include <sys/sx.h>
60 #include <sys/ttycom.h>
61 #include <sys/conf.h>
62 #include <sys/syslog.h>
63 #include <sys/unistd.h>
64 
65 static fo_rdwr_t	vn_read;
66 static fo_rdwr_t	vn_write;
67 static fo_ioctl_t	vn_ioctl;
68 static fo_poll_t	vn_poll;
69 static fo_kqfilter_t	vn_kqfilter;
70 static fo_stat_t	vn_statfile;
71 static fo_close_t	vn_closefile;
72 
73 struct 	fileops vnops = {
74 	.fo_read = vn_read,
75 	.fo_write = vn_write,
76 	.fo_ioctl = vn_ioctl,
77 	.fo_poll = vn_poll,
78 	.fo_kqfilter = vn_kqfilter,
79 	.fo_stat = vn_statfile,
80 	.fo_close = vn_closefile,
81 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
82 };
83 
84 int
85 vn_open(ndp, flagp, cmode, fdidx)
86 	struct nameidata *ndp;
87 	int *flagp, cmode, fdidx;
88 {
89 	struct thread *td = ndp->ni_cnd.cn_thread;
90 
91 	return (vn_open_cred(ndp, flagp, cmode, td->td_ucred, fdidx));
92 }
93 
94 /*
95  * Common code for vnode open operations.
96  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
97  *
98  * Note that this does NOT free nameidata for the successful case,
99  * due to the NDINIT being done elsewhere.
100  */
101 int
102 vn_open_cred(ndp, flagp, cmode, cred, fdidx)
103 	struct nameidata *ndp;
104 	int *flagp, cmode;
105 	struct ucred *cred;
106 	int fdidx;
107 {
108 	struct vnode *vp;
109 	struct mount *mp;
110 	struct thread *td = ndp->ni_cnd.cn_thread;
111 	struct vattr vat;
112 	struct vattr *vap = &vat;
113 	int mode, fmode, error;
114 #ifdef LOOKUP_SHARED
115 	int exclusive;	/* The current intended lock state */
116 
117 	exclusive = 0;
118 #endif
119 
120 restart:
121 	fmode = *flagp;
122 	if (fmode & O_CREAT) {
123 		ndp->ni_cnd.cn_nameiop = CREATE;
124 		ndp->ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
125 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
126 			ndp->ni_cnd.cn_flags |= FOLLOW;
127 		bwillwrite();
128 		if ((error = namei(ndp)) != 0)
129 			return (error);
130 		if (ndp->ni_vp == NULL) {
131 			VATTR_NULL(vap);
132 			vap->va_type = VREG;
133 			vap->va_mode = cmode;
134 			if (fmode & O_EXCL)
135 				vap->va_vaflags |= VA_EXCLUSIVE;
136 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
137 				NDFREE(ndp, NDF_ONLY_PNBUF);
138 				vput(ndp->ni_dvp);
139 				if ((error = vn_start_write(NULL, &mp,
140 				    V_XSLEEP | PCATCH)) != 0)
141 					return (error);
142 				goto restart;
143 			}
144 #ifdef MAC
145 			error = mac_check_vnode_create(cred, ndp->ni_dvp,
146 			    &ndp->ni_cnd, vap);
147 			if (error == 0) {
148 #endif
149 				VOP_LEASE(ndp->ni_dvp, td, cred, LEASE_WRITE);
150 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
151 						   &ndp->ni_cnd, vap);
152 #ifdef MAC
153 			}
154 #endif
155 			vput(ndp->ni_dvp);
156 			vn_finished_write(mp);
157 			if (error) {
158 				NDFREE(ndp, NDF_ONLY_PNBUF);
159 				return (error);
160 			}
161 			ASSERT_VOP_UNLOCKED(ndp->ni_dvp, "create");
162 			ASSERT_VOP_LOCKED(ndp->ni_vp, "create");
163 			fmode &= ~O_TRUNC;
164 			vp = ndp->ni_vp;
165 #ifdef LOOKUP_SHARED
166 			exclusive = 1;
167 #endif
168 		} else {
169 			if (ndp->ni_dvp == ndp->ni_vp)
170 				vrele(ndp->ni_dvp);
171 			else
172 				vput(ndp->ni_dvp);
173 			ndp->ni_dvp = NULL;
174 			vp = ndp->ni_vp;
175 			if (fmode & O_EXCL) {
176 				error = EEXIST;
177 				goto bad;
178 			}
179 			fmode &= ~O_CREAT;
180 		}
181 	} else {
182 		ndp->ni_cnd.cn_nameiop = LOOKUP;
183 #ifdef LOOKUP_SHARED
184 		ndp->ni_cnd.cn_flags =
185 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
186 		    LOCKSHARED | LOCKLEAF;
187 #else
188 		ndp->ni_cnd.cn_flags =
189 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
190 #endif
191 		if ((error = namei(ndp)) != 0)
192 			return (error);
193 		vp = ndp->ni_vp;
194 	}
195 	if (vp->v_type == VLNK) {
196 		error = EMLINK;
197 		goto bad;
198 	}
199 	if (vp->v_type == VSOCK) {
200 		error = EOPNOTSUPP;
201 		goto bad;
202 	}
203 	mode = 0;
204 	if (fmode & (FWRITE | O_TRUNC)) {
205 		if (vp->v_type == VDIR) {
206 			error = EISDIR;
207 			goto bad;
208 		}
209 		mode |= VWRITE;
210 	}
211 	if (fmode & FREAD)
212 		mode |= VREAD;
213 	if (fmode & O_APPEND)
214 		mode |= VAPPEND;
215 #ifdef MAC
216 	error = mac_check_vnode_open(cred, vp, mode);
217 	if (error)
218 		goto bad;
219 #endif
220 	if ((fmode & O_CREAT) == 0) {
221 		if (mode & VWRITE) {
222 			error = vn_writechk(vp);
223 			if (error)
224 				goto bad;
225 		}
226 		if (mode) {
227 		        error = VOP_ACCESS(vp, mode, cred, td);
228 			if (error)
229 				goto bad;
230 		}
231 	}
232 	if ((error = VOP_GETATTR(vp, vap, cred, td)) == 0) {
233 		vp->v_cachedfs = vap->va_fsid;
234 		vp->v_cachedid = vap->va_fileid;
235 	}
236 	if ((error = VOP_OPEN(vp, fmode, cred, td, fdidx)) != 0)
237 		goto bad;
238 	/*
239 	 * Make sure that a VM object is created for VMIO support.
240 	 */
241 	if (vn_canvmio(vp) == TRUE) {
242 #ifdef LOOKUP_SHARED
243 		int flock;
244 
245 		if (!exclusive && VOP_GETVOBJECT(vp, NULL) != 0)
246 			VOP_LOCK(vp, LK_UPGRADE, td);
247 		/*
248 		 * In cases where the object is marked as dead object_create
249 		 * will unlock and relock exclusive.  It is safe to call in
250 		 * here with a shared lock because we only examine fields that
251 		 * the shared lock guarantees will be stable.  In the UPGRADE
252 		 * case it is not likely that anyone has used this vnode yet
253 		 * so there will be no contention.  The logic after this call
254 		 * restores the requested locking state.
255 		 */
256 #endif
257 		if ((error = vfs_object_create(vp, td, cred)) != 0) {
258 			VOP_UNLOCK(vp, 0, td);
259 			VOP_CLOSE(vp, fmode, cred, td);
260 			NDFREE(ndp, NDF_ONLY_PNBUF);
261 			vrele(vp);
262 			*flagp = fmode;
263 			return (error);
264 		}
265 #ifdef LOOKUP_SHARED
266 		flock = VOP_ISLOCKED(vp, td);
267 		if (!exclusive && flock == LK_EXCLUSIVE)
268 			VOP_LOCK(vp, LK_DOWNGRADE, td);
269 #endif
270 	}
271 
272 	if (fmode & FWRITE)
273 		vp->v_writecount++;
274 	*flagp = fmode;
275 	ASSERT_VOP_LOCKED(vp, "vn_open_cred");
276 	return (0);
277 bad:
278 	NDFREE(ndp, NDF_ONLY_PNBUF);
279 	vput(vp);
280 	*flagp = fmode;
281 	ndp->ni_vp = NULL;
282 	return (error);
283 }
284 
285 /*
286  * Check for write permissions on the specified vnode.
287  * Prototype text segments cannot be written.
288  */
289 int
290 vn_writechk(vp)
291 	register struct vnode *vp;
292 {
293 
294 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
295 	/*
296 	 * If there's shared text associated with
297 	 * the vnode, try to free it up once.  If
298 	 * we fail, we can't allow writing.
299 	 */
300 	if (vp->v_vflag & VV_TEXT)
301 		return (ETXTBSY);
302 
303 	return (0);
304 }
305 
306 /*
307  * Vnode close call
308  */
309 int
310 vn_close(vp, flags, file_cred, td)
311 	register struct vnode *vp;
312 	int flags;
313 	struct ucred *file_cred;
314 	struct thread *td;
315 {
316 	int error;
317 
318 	if (flags & FWRITE)
319 		vp->v_writecount--;
320 	error = VOP_CLOSE(vp, flags, file_cred, td);
321 	/*
322 	 * XXX - In certain instances VOP_CLOSE has to do the vrele
323 	 * itself. If the vrele has been done, it will return EAGAIN
324 	 * to indicate that the vrele should not be done again. When
325 	 * this happens, we just return success. The correct thing to
326 	 * do would be to have all VOP_CLOSE instances do the vrele.
327 	 */
328 	if (error == EAGAIN)
329 		return (0);
330 	vrele(vp);
331 	return (error);
332 }
333 
334 /*
335  * Sequential heuristic - detect sequential operation
336  */
337 static __inline
338 int
339 sequential_heuristic(struct uio *uio, struct file *fp)
340 {
341 
342 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
343 	    uio->uio_offset == fp->f_nextoff) {
344 		/*
345 		 * XXX we assume that the filesystem block size is
346 		 * the default.  Not true, but still gives us a pretty
347 		 * good indicator of how sequential the read operations
348 		 * are.
349 		 */
350 		fp->f_seqcount += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
351 		if (fp->f_seqcount > IO_SEQMAX)
352 			fp->f_seqcount = IO_SEQMAX;
353 		return(fp->f_seqcount << IO_SEQSHIFT);
354 	}
355 
356 	/*
357 	 * Not sequential, quick draw-down of seqcount
358 	 */
359 	if (fp->f_seqcount > 1)
360 		fp->f_seqcount = 1;
361 	else
362 		fp->f_seqcount = 0;
363 	return(0);
364 }
365 
366 /*
367  * Package up an I/O request on a vnode into a uio and do it.
368  */
369 int
370 vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred,
371     aresid, td)
372 	enum uio_rw rw;
373 	struct vnode *vp;
374 	caddr_t base;
375 	int len;
376 	off_t offset;
377 	enum uio_seg segflg;
378 	int ioflg;
379 	struct ucred *active_cred;
380 	struct ucred *file_cred;
381 	int *aresid;
382 	struct thread *td;
383 {
384 	struct uio auio;
385 	struct iovec aiov;
386 	struct mount *mp;
387 	struct ucred *cred;
388 	int error;
389 
390 	if ((ioflg & IO_NODELOCKED) == 0) {
391 		mp = NULL;
392 		if (rw == UIO_WRITE) {
393 			if (vp->v_type != VCHR &&
394 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
395 			    != 0)
396 				return (error);
397 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
398 		} else {
399 			/*
400 			 * XXX This should be LK_SHARED but I don't trust VFS
401 			 * enough to leave it like that until it has been
402 			 * reviewed further.
403 			 */
404 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
405 		}
406 
407 	}
408 	auio.uio_iov = &aiov;
409 	auio.uio_iovcnt = 1;
410 	aiov.iov_base = base;
411 	aiov.iov_len = len;
412 	auio.uio_resid = len;
413 	auio.uio_offset = offset;
414 	auio.uio_segflg = segflg;
415 	auio.uio_rw = rw;
416 	auio.uio_td = td;
417 	error = 0;
418 #ifdef MAC
419 	if ((ioflg & IO_NOMACCHECK) == 0) {
420 		if (rw == UIO_READ)
421 			error = mac_check_vnode_read(active_cred, file_cred,
422 			    vp);
423 		else
424 			error = mac_check_vnode_write(active_cred, file_cred,
425 			    vp);
426 	}
427 #endif
428 	if (error == 0) {
429 		if (file_cred)
430 			cred = file_cred;
431 		else
432 			cred = active_cred;
433 		if (rw == UIO_READ)
434 			error = VOP_READ(vp, &auio, ioflg, cred);
435 		else
436 			error = VOP_WRITE(vp, &auio, ioflg, cred);
437 	}
438 	if (aresid)
439 		*aresid = auio.uio_resid;
440 	else
441 		if (auio.uio_resid && error == 0)
442 			error = EIO;
443 	if ((ioflg & IO_NODELOCKED) == 0) {
444 		if (rw == UIO_WRITE)
445 			vn_finished_write(mp);
446 		VOP_UNLOCK(vp, 0, td);
447 	}
448 	return (error);
449 }
450 
451 /*
452  * Package up an I/O request on a vnode into a uio and do it.  The I/O
453  * request is split up into smaller chunks and we try to avoid saturating
454  * the buffer cache while potentially holding a vnode locked, so we
455  * check bwillwrite() before calling vn_rdwr().  We also call uio_yield()
456  * to give other processes a chance to lock the vnode (either other processes
457  * core'ing the same binary, or unrelated processes scanning the directory).
458  */
459 int
460 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
461     file_cred, aresid, td)
462 	enum uio_rw rw;
463 	struct vnode *vp;
464 	caddr_t base;
465 	size_t len;
466 	off_t offset;
467 	enum uio_seg segflg;
468 	int ioflg;
469 	struct ucred *active_cred;
470 	struct ucred *file_cred;
471 	size_t *aresid;
472 	struct thread *td;
473 {
474 	int error = 0;
475 	int iaresid;
476 
477 	do {
478 		int chunk;
479 
480 		/*
481 		 * Force `offset' to a multiple of MAXBSIZE except possibly
482 		 * for the first chunk, so that filesystems only need to
483 		 * write full blocks except possibly for the first and last
484 		 * chunks.
485 		 */
486 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
487 
488 		if (chunk > len)
489 			chunk = len;
490 		if (rw != UIO_READ && vp->v_type == VREG)
491 			bwillwrite();
492 		iaresid = 0;
493 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
494 		    ioflg, active_cred, file_cred, &iaresid, td);
495 		len -= chunk;	/* aresid calc already includes length */
496 		if (error)
497 			break;
498 		offset += chunk;
499 		base += chunk;
500 		uio_yield();
501 	} while (len);
502 	if (aresid)
503 		*aresid = len + iaresid;
504 	return (error);
505 }
506 
507 /*
508  * File table vnode read routine.
509  */
510 static int
511 vn_read(fp, uio, active_cred, flags, td)
512 	struct file *fp;
513 	struct uio *uio;
514 	struct ucred *active_cred;
515 	struct thread *td;
516 	int flags;
517 {
518 	struct vnode *vp;
519 	int error, ioflag;
520 
521 	mtx_lock(&Giant);
522 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
523 	    uio->uio_td, td));
524 	vp = fp->f_vnode;
525 	ioflag = 0;
526 	if (fp->f_flag & FNONBLOCK)
527 		ioflag |= IO_NDELAY;
528 	if (fp->f_flag & O_DIRECT)
529 		ioflag |= IO_DIRECT;
530 	VOP_LEASE(vp, td, fp->f_cred, LEASE_READ);
531 	/*
532 	 * According to McKusick the vn lock is protecting f_offset here.
533 	 * Once this field has it's own lock we can acquire this shared.
534 	 */
535 	if ((flags & FOF_OFFSET) == 0) {
536 		vn_lock(vp, LK_EXCLUSIVE | LK_NOPAUSE | LK_RETRY, td);
537 		uio->uio_offset = fp->f_offset;
538 	} else
539 		vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
540 
541 	ioflag |= sequential_heuristic(uio, fp);
542 
543 #ifdef MAC
544 	error = mac_check_vnode_read(active_cred, fp->f_cred, vp);
545 	if (error == 0)
546 #endif
547 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
548 	if ((flags & FOF_OFFSET) == 0)
549 		fp->f_offset = uio->uio_offset;
550 	fp->f_nextoff = uio->uio_offset;
551 	VOP_UNLOCK(vp, 0, td);
552 	mtx_unlock(&Giant);
553 	return (error);
554 }
555 
556 /*
557  * File table vnode write routine.
558  */
559 static int
560 vn_write(fp, uio, active_cred, flags, td)
561 	struct file *fp;
562 	struct uio *uio;
563 	struct ucred *active_cred;
564 	struct thread *td;
565 	int flags;
566 {
567 	struct vnode *vp;
568 	struct mount *mp;
569 	int error, ioflag;
570 
571 	mtx_lock(&Giant);
572 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
573 	    uio->uio_td, td));
574 	vp = fp->f_vnode;
575 	if (vp->v_type == VREG)
576 		bwillwrite();
577 	ioflag = IO_UNIT;
578 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
579 		ioflag |= IO_APPEND;
580 	if (fp->f_flag & FNONBLOCK)
581 		ioflag |= IO_NDELAY;
582 	if (fp->f_flag & O_DIRECT)
583 		ioflag |= IO_DIRECT;
584 	if ((fp->f_flag & O_FSYNC) ||
585 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
586 		ioflag |= IO_SYNC;
587 	mp = NULL;
588 	if (vp->v_type != VCHR &&
589 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) {
590 		mtx_unlock(&Giant);
591 		return (error);
592 	}
593 	VOP_LEASE(vp, td, fp->f_cred, LEASE_WRITE);
594 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
595 	if ((flags & FOF_OFFSET) == 0)
596 		uio->uio_offset = fp->f_offset;
597 	ioflag |= sequential_heuristic(uio, fp);
598 #ifdef MAC
599 	error = mac_check_vnode_write(active_cred, fp->f_cred, vp);
600 	if (error == 0)
601 #endif
602 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
603 	if ((flags & FOF_OFFSET) == 0)
604 		fp->f_offset = uio->uio_offset;
605 	fp->f_nextoff = uio->uio_offset;
606 	VOP_UNLOCK(vp, 0, td);
607 	vn_finished_write(mp);
608 	mtx_unlock(&Giant);
609 	return (error);
610 }
611 
612 /*
613  * File table vnode stat routine.
614  */
615 static int
616 vn_statfile(fp, sb, active_cred, td)
617 	struct file *fp;
618 	struct stat *sb;
619 	struct ucred *active_cred;
620 	struct thread *td;
621 {
622 	struct vnode *vp = fp->f_vnode;
623 	int error;
624 
625 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
626 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
627 	VOP_UNLOCK(vp, 0, td);
628 
629 	return (error);
630 }
631 
632 /*
633  * Stat a vnode; implementation for the stat syscall
634  */
635 int
636 vn_stat(vp, sb, active_cred, file_cred, td)
637 	struct vnode *vp;
638 	register struct stat *sb;
639 	struct ucred *active_cred;
640 	struct ucred *file_cred;
641 	struct thread *td;
642 {
643 	struct vattr vattr;
644 	register struct vattr *vap;
645 	int error;
646 	u_short mode;
647 
648 #ifdef MAC
649 	error = mac_check_vnode_stat(active_cred, file_cred, vp);
650 	if (error)
651 		return (error);
652 #endif
653 
654 	vap = &vattr;
655 	error = VOP_GETATTR(vp, vap, active_cred, td);
656 	if (error)
657 		return (error);
658 
659 	vp->v_cachedfs = vap->va_fsid;
660 	vp->v_cachedid = vap->va_fileid;
661 
662 	/*
663 	 * Zero the spare stat fields
664 	 */
665 	bzero(sb, sizeof *sb);
666 
667 	/*
668 	 * Copy from vattr table
669 	 */
670 	if (vap->va_fsid != VNOVAL)
671 		sb->st_dev = vap->va_fsid;
672 	else
673 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
674 	sb->st_ino = vap->va_fileid;
675 	mode = vap->va_mode;
676 	switch (vap->va_type) {
677 	case VREG:
678 		mode |= S_IFREG;
679 		break;
680 	case VDIR:
681 		mode |= S_IFDIR;
682 		break;
683 	case VBLK:
684 		mode |= S_IFBLK;
685 		break;
686 	case VCHR:
687 		mode |= S_IFCHR;
688 		break;
689 	case VLNK:
690 		mode |= S_IFLNK;
691 		/* This is a cosmetic change, symlinks do not have a mode. */
692 		if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW)
693 			sb->st_mode &= ~ACCESSPERMS;	/* 0000 */
694 		else
695 			sb->st_mode |= ACCESSPERMS;	/* 0777 */
696 		break;
697 	case VSOCK:
698 		mode |= S_IFSOCK;
699 		break;
700 	case VFIFO:
701 		mode |= S_IFIFO;
702 		break;
703 	default:
704 		return (EBADF);
705 	};
706 	sb->st_mode = mode;
707 	sb->st_nlink = vap->va_nlink;
708 	sb->st_uid = vap->va_uid;
709 	sb->st_gid = vap->va_gid;
710 	sb->st_rdev = vap->va_rdev;
711 	if (vap->va_size > OFF_MAX)
712 		return (EOVERFLOW);
713 	sb->st_size = vap->va_size;
714 	sb->st_atimespec = vap->va_atime;
715 	sb->st_mtimespec = vap->va_mtime;
716 	sb->st_ctimespec = vap->va_ctime;
717 	sb->st_birthtimespec = vap->va_birthtime;
718 
719         /*
720 	 * According to www.opengroup.org, the meaning of st_blksize is
721 	 *   "a filesystem-specific preferred I/O block size for this
722 	 *    object.  In some filesystem types, this may vary from file
723 	 *    to file"
724 	 * Default to PAGE_SIZE after much discussion.
725 	 */
726 
727 	if (vap->va_type == VREG) {
728 		sb->st_blksize = vap->va_blocksize;
729 	} else if (vn_isdisk(vp, NULL)) {
730 		sb->st_blksize = vp->v_rdev->si_bsize_best;
731 		if (sb->st_blksize < vp->v_rdev->si_bsize_phys)
732 			sb->st_blksize = vp->v_rdev->si_bsize_phys;
733 		if (sb->st_blksize < BLKDEV_IOSIZE)
734 			sb->st_blksize = BLKDEV_IOSIZE;
735 	} else {
736 		sb->st_blksize = PAGE_SIZE;
737 	}
738 
739 	sb->st_flags = vap->va_flags;
740 	if (suser(td))
741 		sb->st_gen = 0;
742 	else
743 		sb->st_gen = vap->va_gen;
744 
745 #if (S_BLKSIZE == 512)
746 	/* Optimize this case */
747 	sb->st_blocks = vap->va_bytes >> 9;
748 #else
749 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
750 #endif
751 	return (0);
752 }
753 
754 /*
755  * File table vnode ioctl routine.
756  */
757 static int
758 vn_ioctl(fp, com, data, active_cred, td)
759 	struct file *fp;
760 	u_long com;
761 	void *data;
762 	struct ucred *active_cred;
763 	struct thread *td;
764 {
765 	struct vnode *vp = fp->f_vnode;
766 	struct vnode *vpold;
767 	struct vattr vattr;
768 	int error;
769 
770 	switch (vp->v_type) {
771 
772 	case VREG:
773 	case VDIR:
774 		if (com == FIONREAD) {
775 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
776 			error = VOP_GETATTR(vp, &vattr, active_cred, td);
777 			VOP_UNLOCK(vp, 0, td);
778 			if (error)
779 				return (error);
780 			*(int *)data = vattr.va_size - fp->f_offset;
781 			return (0);
782 		}
783 		if (com == FIONBIO || com == FIOASYNC)	/* XXX */
784 			return (0);			/* XXX */
785 		/* FALLTHROUGH */
786 
787 	default:
788 #if 0
789 		return (ENOTTY);
790 #endif
791 	case VFIFO:
792 	case VCHR:
793 	case VBLK:
794 		if (com == FIODTYPE) {
795 			if (vp->v_type != VCHR && vp->v_type != VBLK)
796 				return (ENOTTY);
797 			*(int *)data = devsw(vp->v_rdev)->d_flags & D_TYPEMASK;
798 			return (0);
799 		}
800 		error = VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td);
801 		if (error == ENOIOCTL) {
802 #ifdef DIAGNOSTIC
803 			kdb_enter("ENOIOCTL leaked through");
804 #endif
805 			error = ENOTTY;
806 		}
807 		if (error == 0 && com == TIOCSCTTY) {
808 
809 			/* Do nothing if reassigning same control tty */
810 			sx_slock(&proctree_lock);
811 			if (td->td_proc->p_session->s_ttyvp == vp) {
812 				sx_sunlock(&proctree_lock);
813 				return (0);
814 			}
815 
816 			vpold = td->td_proc->p_session->s_ttyvp;
817 			VREF(vp);
818 			SESS_LOCK(td->td_proc->p_session);
819 			td->td_proc->p_session->s_ttyvp = vp;
820 			SESS_UNLOCK(td->td_proc->p_session);
821 
822 			sx_sunlock(&proctree_lock);
823 
824 			/* Get rid of reference to old control tty */
825 			if (vpold)
826 				vrele(vpold);
827 		}
828 		return (error);
829 	}
830 }
831 
832 /*
833  * File table vnode poll routine.
834  */
835 static int
836 vn_poll(fp, events, active_cred, td)
837 	struct file *fp;
838 	int events;
839 	struct ucred *active_cred;
840 	struct thread *td;
841 {
842 	struct vnode *vp;
843 #ifdef MAC
844 	int error;
845 #endif
846 
847 	vp = fp->f_vnode;
848 #ifdef MAC
849 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
850 	error = mac_check_vnode_poll(active_cred, fp->f_cred, vp);
851 	VOP_UNLOCK(vp, 0, td);
852 	if (error)
853 		return (error);
854 #endif
855 
856 	return (VOP_POLL(vp, events, fp->f_cred, td));
857 }
858 
859 /*
860  * Check that the vnode is still valid, and if so
861  * acquire requested lock.
862  */
863 int
864 #ifndef	DEBUG_LOCKS
865 vn_lock(vp, flags, td)
866 #else
867 debug_vn_lock(vp, flags, td, filename, line)
868 #endif
869 	struct vnode *vp;
870 	int flags;
871 	struct thread *td;
872 #ifdef	DEBUG_LOCKS
873 	const char *filename;
874 	int line;
875 #endif
876 {
877 	int error;
878 
879 	do {
880 		if ((flags & LK_INTERLOCK) == 0)
881 			VI_LOCK(vp);
882 		if ((vp->v_iflag & VI_XLOCK) && vp->v_vxthread != curthread) {
883 			if ((flags & LK_NOWAIT) != 0) {
884 				VI_UNLOCK(vp);
885 				return (ENOENT);
886 			}
887 			vp->v_iflag |= VI_XWANT;
888 			msleep(vp, VI_MTX(vp), PINOD, "vn_lock", 0);
889 			if ((flags & LK_RETRY) == 0) {
890 				VI_UNLOCK(vp);
891 				return (ENOENT);
892 			}
893 		}
894 #ifdef	DEBUG_LOCKS
895 		vp->filename = filename;
896 		vp->line = line;
897 #endif
898 		/*
899 		 * lockmgr drops interlock before it will return for
900 		 * any reason.  So force the code above to relock it.
901 		 */
902 		error = VOP_LOCK(vp, flags | LK_NOPAUSE | LK_INTERLOCK, td);
903 		flags &= ~LK_INTERLOCK;
904 	} while (flags & LK_RETRY && error != 0);
905 	return (error);
906 }
907 
908 /*
909  * File table vnode close routine.
910  */
911 static int
912 vn_closefile(fp, td)
913 	struct file *fp;
914 	struct thread *td;
915 {
916 	struct vnode *vp;
917 	struct flock lf;
918 
919 	GIANT_REQUIRED;
920 
921 	vp = fp->f_vnode;
922 
923 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
924 		lf.l_whence = SEEK_SET;
925 		lf.l_start = 0;
926 		lf.l_len = 0;
927 		lf.l_type = F_UNLCK;
928 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
929 	}
930 
931 	fp->f_ops = &badfileops;
932 
933 	return (vn_close(vp, fp->f_flag, fp->f_cred, td));
934 }
935 
936 /*
937  * Preparing to start a filesystem write operation. If the operation is
938  * permitted, then we bump the count of operations in progress and
939  * proceed. If a suspend request is in progress, we wait until the
940  * suspension is over, and then proceed.
941  */
942 int
943 vn_start_write(vp, mpp, flags)
944 	struct vnode *vp;
945 	struct mount **mpp;
946 	int flags;
947 {
948 	struct mount *mp;
949 	int error;
950 
951 	GIANT_REQUIRED;
952 
953 	/*
954 	 * If a vnode is provided, get and return the mount point that
955 	 * to which it will write.
956 	 */
957 	if (vp != NULL) {
958 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
959 			*mpp = NULL;
960 			if (error != EOPNOTSUPP)
961 				return (error);
962 			return (0);
963 		}
964 	}
965 	if ((mp = *mpp) == NULL)
966 		return (0);
967 	/*
968 	 * Check on status of suspension.
969 	 */
970 	while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
971 		if (flags & V_NOWAIT)
972 			return (EWOULDBLOCK);
973 		error = tsleep(&mp->mnt_flag, (PUSER - 1) | (flags & PCATCH),
974 		    "suspfs", 0);
975 		if (error)
976 			return (error);
977 	}
978 	if (flags & V_XSLEEP)
979 		return (0);
980 	mp->mnt_writeopcount++;
981 	return (0);
982 }
983 
984 /*
985  * Secondary suspension. Used by operations such as vop_inactive
986  * routines that are needed by the higher level functions. These
987  * are allowed to proceed until all the higher level functions have
988  * completed (indicated by mnt_writeopcount dropping to zero). At that
989  * time, these operations are halted until the suspension is over.
990  */
991 int
992 vn_write_suspend_wait(vp, mp, flags)
993 	struct vnode *vp;
994 	struct mount *mp;
995 	int flags;
996 {
997 	int error;
998 
999 	if (vp != NULL) {
1000 		if ((error = VOP_GETWRITEMOUNT(vp, &mp)) != 0) {
1001 			if (error != EOPNOTSUPP)
1002 				return (error);
1003 			return (0);
1004 		}
1005 	}
1006 	/*
1007 	 * If we are not suspended or have not yet reached suspended
1008 	 * mode, then let the operation proceed.
1009 	 */
1010 	if (mp == NULL || (mp->mnt_kern_flag & MNTK_SUSPENDED) == 0)
1011 		return (0);
1012 	if (flags & V_NOWAIT)
1013 		return (EWOULDBLOCK);
1014 	/*
1015 	 * Wait for the suspension to finish.
1016 	 */
1017 	return (tsleep(&mp->mnt_flag, (PUSER - 1) | (flags & PCATCH),
1018 	    "suspfs", 0));
1019 }
1020 
1021 /*
1022  * Filesystem write operation has completed. If we are suspending and this
1023  * operation is the last one, notify the suspender that the suspension is
1024  * now in effect.
1025  */
1026 void
1027 vn_finished_write(mp)
1028 	struct mount *mp;
1029 {
1030 
1031 	GIANT_REQUIRED;
1032 
1033 	if (mp == NULL)
1034 		return;
1035 	mp->mnt_writeopcount--;
1036 	if (mp->mnt_writeopcount < 0)
1037 		panic("vn_finished_write: neg cnt");
1038 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1039 	    mp->mnt_writeopcount <= 0)
1040 		wakeup(&mp->mnt_writeopcount);
1041 }
1042 
1043 /*
1044  * Request a filesystem to suspend write operations.
1045  */
1046 int
1047 vfs_write_suspend(mp)
1048 	struct mount *mp;
1049 {
1050 	struct thread *td = curthread;
1051 	int error;
1052 
1053 	if (mp->mnt_kern_flag & MNTK_SUSPEND)
1054 		return (0);
1055 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1056 	if (mp->mnt_writeopcount > 0)
1057 		(void) tsleep(&mp->mnt_writeopcount, PUSER - 1, "suspwt", 0);
1058 	if ((error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) != 0) {
1059 		vfs_write_resume(mp);
1060 		return (error);
1061 	}
1062 	mp->mnt_kern_flag |= MNTK_SUSPENDED;
1063 	return (0);
1064 }
1065 
1066 /*
1067  * Request a filesystem to resume write operations.
1068  */
1069 void
1070 vfs_write_resume(mp)
1071 	struct mount *mp;
1072 {
1073 
1074 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) == 0)
1075 		return;
1076 	mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPENDED);
1077 	wakeup(&mp->mnt_writeopcount);
1078 	wakeup(&mp->mnt_flag);
1079 }
1080 
1081 /*
1082  * Implement kqueues for files by translating it to vnode operation.
1083  */
1084 static int
1085 vn_kqfilter(struct file *fp, struct knote *kn)
1086 {
1087 
1088 	return (VOP_KQFILTER(fp->f_vnode, kn));
1089 }
1090 
1091 /*
1092  * Simplified in-kernel wrapper calls for extended attribute access.
1093  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1094  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1095  */
1096 int
1097 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1098     const char *attrname, int *buflen, char *buf, struct thread *td)
1099 {
1100 	struct uio	auio;
1101 	struct iovec	iov;
1102 	int	error;
1103 
1104 	iov.iov_len = *buflen;
1105 	iov.iov_base = buf;
1106 
1107 	auio.uio_iov = &iov;
1108 	auio.uio_iovcnt = 1;
1109 	auio.uio_rw = UIO_READ;
1110 	auio.uio_segflg = UIO_SYSSPACE;
1111 	auio.uio_td = td;
1112 	auio.uio_offset = 0;
1113 	auio.uio_resid = *buflen;
1114 
1115 	if ((ioflg & IO_NODELOCKED) == 0)
1116 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1117 
1118 	/* authorize attribute retrieval as kernel */
1119 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1120 	    td);
1121 
1122 	if ((ioflg & IO_NODELOCKED) == 0)
1123 		VOP_UNLOCK(vp, 0, td);
1124 
1125 	if (error == 0) {
1126 		*buflen = *buflen - auio.uio_resid;
1127 	}
1128 
1129 	return (error);
1130 }
1131 
1132 /*
1133  * XXX failure mode if partially written?
1134  */
1135 int
1136 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1137     const char *attrname, int buflen, char *buf, struct thread *td)
1138 {
1139 	struct uio	auio;
1140 	struct iovec	iov;
1141 	struct mount	*mp;
1142 	int	error;
1143 
1144 	iov.iov_len = buflen;
1145 	iov.iov_base = buf;
1146 
1147 	auio.uio_iov = &iov;
1148 	auio.uio_iovcnt = 1;
1149 	auio.uio_rw = UIO_WRITE;
1150 	auio.uio_segflg = UIO_SYSSPACE;
1151 	auio.uio_td = td;
1152 	auio.uio_offset = 0;
1153 	auio.uio_resid = buflen;
1154 
1155 	if ((ioflg & IO_NODELOCKED) == 0) {
1156 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1157 			return (error);
1158 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1159 	}
1160 
1161 	/* authorize attribute setting as kernel */
1162 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1163 
1164 	if ((ioflg & IO_NODELOCKED) == 0) {
1165 		vn_finished_write(mp);
1166 		VOP_UNLOCK(vp, 0, td);
1167 	}
1168 
1169 	return (error);
1170 }
1171 
1172 int
1173 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1174     const char *attrname, struct thread *td)
1175 {
1176 	struct mount	*mp;
1177 	int	error;
1178 
1179 	if ((ioflg & IO_NODELOCKED) == 0) {
1180 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1181 			return (error);
1182 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1183 	}
1184 
1185 	/* authorize attribute removal as kernel */
1186 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1187 	if (error == EOPNOTSUPP)
1188 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1189 		    NULL, td);
1190 
1191 	if ((ioflg & IO_NODELOCKED) == 0) {
1192 		vn_finished_write(mp);
1193 		VOP_UNLOCK(vp, 0, td);
1194 	}
1195 
1196 	return (error);
1197 }
1198