xref: /freebsd/sys/kern/vfs_vnops.c (revision f5f7c05209ca2c3748fd8b27c5e80ffad49120eb)
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 <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/fcntl.h>
43 #include <sys/file.h>
44 #include <sys/kdb.h>
45 #include <sys/stat.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/limits.h>
49 #include <sys/lock.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/vnode.h>
54 #include <sys/bio.h>
55 #include <sys/buf.h>
56 #include <sys/filio.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sx.h>
59 #include <sys/sysctl.h>
60 #include <sys/ttycom.h>
61 #include <sys/conf.h>
62 #include <sys/syslog.h>
63 #include <sys/unistd.h>
64 
65 #include <security/audit/audit.h>
66 #include <security/mac/mac_framework.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70 #include <vm/pmap.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 
75 static fo_rdwr_t	vn_read;
76 static fo_rdwr_t	vn_write;
77 static fo_rdwr_t	vn_io_fault;
78 static fo_truncate_t	vn_truncate;
79 static fo_ioctl_t	vn_ioctl;
80 static fo_poll_t	vn_poll;
81 static fo_kqfilter_t	vn_kqfilter;
82 static fo_stat_t	vn_statfile;
83 static fo_close_t	vn_closefile;
84 
85 struct 	fileops vnops = {
86 	.fo_read = vn_io_fault,
87 	.fo_write = vn_io_fault,
88 	.fo_truncate = vn_truncate,
89 	.fo_ioctl = vn_ioctl,
90 	.fo_poll = vn_poll,
91 	.fo_kqfilter = vn_kqfilter,
92 	.fo_stat = vn_statfile,
93 	.fo_close = vn_closefile,
94 	.fo_chmod = vn_chmod,
95 	.fo_chown = vn_chown,
96 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
97 };
98 
99 int
100 vn_open(ndp, flagp, cmode, fp)
101 	struct nameidata *ndp;
102 	int *flagp, cmode;
103 	struct file *fp;
104 {
105 	struct thread *td = ndp->ni_cnd.cn_thread;
106 
107 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
108 }
109 
110 /*
111  * Common code for vnode open operations via a name lookup.
112  * Lookup the vnode and invoke VOP_CREATE if needed.
113  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
114  *
115  * Note that this does NOT free nameidata for the successful case,
116  * due to the NDINIT being done elsewhere.
117  */
118 int
119 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
120     struct ucred *cred, struct file *fp)
121 {
122 	struct vnode *vp;
123 	struct mount *mp;
124 	struct thread *td = ndp->ni_cnd.cn_thread;
125 	struct vattr vat;
126 	struct vattr *vap = &vat;
127 	int fmode, error;
128 
129 restart:
130 	fmode = *flagp;
131 	if (fmode & O_CREAT) {
132 		ndp->ni_cnd.cn_nameiop = CREATE;
133 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
134 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
135 			ndp->ni_cnd.cn_flags |= FOLLOW;
136 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
137 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
138 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
139 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
140 		bwillwrite();
141 		if ((error = namei(ndp)) != 0)
142 			return (error);
143 		if (ndp->ni_vp == NULL) {
144 			VATTR_NULL(vap);
145 			vap->va_type = VREG;
146 			vap->va_mode = cmode;
147 			if (fmode & O_EXCL)
148 				vap->va_vaflags |= VA_EXCLUSIVE;
149 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
150 				NDFREE(ndp, NDF_ONLY_PNBUF);
151 				vput(ndp->ni_dvp);
152 				if ((error = vn_start_write(NULL, &mp,
153 				    V_XSLEEP | PCATCH)) != 0)
154 					return (error);
155 				goto restart;
156 			}
157 #ifdef MAC
158 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
159 			    &ndp->ni_cnd, vap);
160 			if (error == 0)
161 #endif
162 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
163 						   &ndp->ni_cnd, vap);
164 			vput(ndp->ni_dvp);
165 			vn_finished_write(mp);
166 			if (error) {
167 				NDFREE(ndp, NDF_ONLY_PNBUF);
168 				return (error);
169 			}
170 			fmode &= ~O_TRUNC;
171 			vp = ndp->ni_vp;
172 		} else {
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 = ISOPEN |
188 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
189 		if (!(fmode & FWRITE))
190 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
191 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
192 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
193 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
194 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
195 		if ((error = namei(ndp)) != 0)
196 			return (error);
197 		vp = ndp->ni_vp;
198 	}
199 	error = vn_open_vnode(vp, fmode, cred, td, fp);
200 	if (error)
201 		goto bad;
202 	*flagp = fmode;
203 	return (0);
204 bad:
205 	NDFREE(ndp, NDF_ONLY_PNBUF);
206 	vput(vp);
207 	*flagp = fmode;
208 	ndp->ni_vp = NULL;
209 	return (error);
210 }
211 
212 /*
213  * Common code for vnode open operations once a vnode is located.
214  * Check permissions, and call the VOP_OPEN routine.
215  */
216 int
217 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
218     struct thread *td, struct file *fp)
219 {
220 	struct mount *mp;
221 	accmode_t accmode;
222 	struct flock lf;
223 	int error, have_flock, lock_flags, type;
224 
225 	if (vp->v_type == VLNK)
226 		return (EMLINK);
227 	if (vp->v_type == VSOCK)
228 		return (EOPNOTSUPP);
229 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
230 		return (ENOTDIR);
231 	accmode = 0;
232 	if (fmode & (FWRITE | O_TRUNC)) {
233 		if (vp->v_type == VDIR)
234 			return (EISDIR);
235 		accmode |= VWRITE;
236 	}
237 	if (fmode & FREAD)
238 		accmode |= VREAD;
239 	if (fmode & FEXEC)
240 		accmode |= VEXEC;
241 	if ((fmode & O_APPEND) && (fmode & FWRITE))
242 		accmode |= VAPPEND;
243 #ifdef MAC
244 	error = mac_vnode_check_open(cred, vp, accmode);
245 	if (error)
246 		return (error);
247 #endif
248 	if ((fmode & O_CREAT) == 0) {
249 		if (accmode & VWRITE) {
250 			error = vn_writechk(vp);
251 			if (error)
252 				return (error);
253 		}
254 		if (accmode) {
255 		        error = VOP_ACCESS(vp, accmode, cred, td);
256 			if (error)
257 				return (error);
258 		}
259 	}
260 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
261 		return (error);
262 
263 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
264 		KASSERT(fp != NULL, ("open with flock requires fp"));
265 		lock_flags = VOP_ISLOCKED(vp);
266 		VOP_UNLOCK(vp, 0);
267 		lf.l_whence = SEEK_SET;
268 		lf.l_start = 0;
269 		lf.l_len = 0;
270 		if (fmode & O_EXLOCK)
271 			lf.l_type = F_WRLCK;
272 		else
273 			lf.l_type = F_RDLCK;
274 		type = F_FLOCK;
275 		if ((fmode & FNONBLOCK) == 0)
276 			type |= F_WAIT;
277 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
278 		have_flock = (error == 0);
279 		vn_lock(vp, lock_flags | LK_RETRY);
280 		if (error == 0 && vp->v_iflag & VI_DOOMED)
281 			error = ENOENT;
282 		/*
283 		 * Another thread might have used this vnode as an
284 		 * executable while the vnode lock was dropped.
285 		 * Ensure the vnode is still able to be opened for
286 		 * writing after the lock has been obtained.
287 		 */
288 		if (error == 0 && accmode & VWRITE)
289 			error = vn_writechk(vp);
290 		if (error) {
291 			VOP_UNLOCK(vp, 0);
292 			if (have_flock) {
293 				lf.l_whence = SEEK_SET;
294 				lf.l_start = 0;
295 				lf.l_len = 0;
296 				lf.l_type = F_UNLCK;
297 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
298 				    F_FLOCK);
299 			}
300 			vn_start_write(vp, &mp, V_WAIT);
301 			vn_lock(vp, lock_flags | LK_RETRY);
302 			(void)VOP_CLOSE(vp, fmode, cred, td);
303 			vn_finished_write(mp);
304 			return (error);
305 		}
306 		fp->f_flag |= FHASLOCK;
307 	}
308 	if (fmode & FWRITE) {
309 		VOP_ADD_WRITECOUNT(vp, 1);
310 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
311 		    __func__, vp, vp->v_writecount);
312 	}
313 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
314 	return (0);
315 }
316 
317 /*
318  * Check for write permissions on the specified vnode.
319  * Prototype text segments cannot be written.
320  */
321 int
322 vn_writechk(vp)
323 	register struct vnode *vp;
324 {
325 
326 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
327 	/*
328 	 * If there's shared text associated with
329 	 * the vnode, try to free it up once.  If
330 	 * we fail, we can't allow writing.
331 	 */
332 	if (VOP_IS_TEXT(vp))
333 		return (ETXTBSY);
334 
335 	return (0);
336 }
337 
338 /*
339  * Vnode close call
340  */
341 int
342 vn_close(vp, flags, file_cred, td)
343 	register struct vnode *vp;
344 	int flags;
345 	struct ucred *file_cred;
346 	struct thread *td;
347 {
348 	struct mount *mp;
349 	int error, lock_flags;
350 
351 	if (!(flags & FWRITE) && vp->v_mount != NULL &&
352 	    vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
353 		lock_flags = LK_SHARED;
354 	else
355 		lock_flags = LK_EXCLUSIVE;
356 
357 	vn_start_write(vp, &mp, V_WAIT);
358 	vn_lock(vp, lock_flags | LK_RETRY);
359 	if (flags & FWRITE) {
360 		VNASSERT(vp->v_writecount > 0, vp,
361 		    ("vn_close: negative writecount"));
362 		VOP_ADD_WRITECOUNT(vp, -1);
363 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
364 		    __func__, vp, vp->v_writecount);
365 	}
366 	error = VOP_CLOSE(vp, flags, file_cred, td);
367 	vput(vp);
368 	vn_finished_write(mp);
369 	return (error);
370 }
371 
372 /*
373  * Heuristic to detect sequential operation.
374  */
375 static int
376 sequential_heuristic(struct uio *uio, struct file *fp)
377 {
378 
379 	if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
380 		return (fp->f_seqcount << IO_SEQSHIFT);
381 
382 	/*
383 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
384 	 * that the first I/O is normally considered to be slightly
385 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
386 	 * unless previous seeks have reduced f_seqcount to 0, in which
387 	 * case offset 0 is not special.
388 	 */
389 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
390 	    uio->uio_offset == fp->f_nextoff) {
391 		/*
392 		 * f_seqcount is in units of fixed-size blocks so that it
393 		 * depends mainly on the amount of sequential I/O and not
394 		 * much on the number of sequential I/O's.  The fixed size
395 		 * of 16384 is hard-coded here since it is (not quite) just
396 		 * a magic size that works well here.  This size is more
397 		 * closely related to the best I/O size for real disks than
398 		 * to any block size used by software.
399 		 */
400 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
401 		if (fp->f_seqcount > IO_SEQMAX)
402 			fp->f_seqcount = IO_SEQMAX;
403 		return (fp->f_seqcount << IO_SEQSHIFT);
404 	}
405 
406 	/* Not sequential.  Quickly draw-down sequentiality. */
407 	if (fp->f_seqcount > 1)
408 		fp->f_seqcount = 1;
409 	else
410 		fp->f_seqcount = 0;
411 	return (0);
412 }
413 
414 /*
415  * Package up an I/O request on a vnode into a uio and do it.
416  */
417 int
418 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
419     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
420     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
421 {
422 	struct uio auio;
423 	struct iovec aiov;
424 	struct mount *mp;
425 	struct ucred *cred;
426 	void *rl_cookie;
427 	int error, lock_flags;
428 
429 	auio.uio_iov = &aiov;
430 	auio.uio_iovcnt = 1;
431 	aiov.iov_base = base;
432 	aiov.iov_len = len;
433 	auio.uio_resid = len;
434 	auio.uio_offset = offset;
435 	auio.uio_segflg = segflg;
436 	auio.uio_rw = rw;
437 	auio.uio_td = td;
438 	error = 0;
439 
440 	if ((ioflg & IO_NODELOCKED) == 0) {
441 		if (rw == UIO_READ) {
442 			rl_cookie = vn_rangelock_rlock(vp, offset,
443 			    offset + len);
444 		} else {
445 			rl_cookie = vn_rangelock_wlock(vp, offset,
446 			    offset + len);
447 		}
448 		mp = NULL;
449 		if (rw == UIO_WRITE) {
450 			if (vp->v_type != VCHR &&
451 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
452 			    != 0)
453 				goto out;
454 			if (MNT_SHARED_WRITES(mp) ||
455 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
456 				lock_flags = LK_SHARED;
457 			else
458 				lock_flags = LK_EXCLUSIVE;
459 		} else
460 			lock_flags = LK_SHARED;
461 		vn_lock(vp, lock_flags | LK_RETRY);
462 	} else
463 		rl_cookie = NULL;
464 
465 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
466 #ifdef MAC
467 	if ((ioflg & IO_NOMACCHECK) == 0) {
468 		if (rw == UIO_READ)
469 			error = mac_vnode_check_read(active_cred, file_cred,
470 			    vp);
471 		else
472 			error = mac_vnode_check_write(active_cred, file_cred,
473 			    vp);
474 	}
475 #endif
476 	if (error == 0) {
477 		if (file_cred != NULL)
478 			cred = file_cred;
479 		else
480 			cred = active_cred;
481 		if (rw == UIO_READ)
482 			error = VOP_READ(vp, &auio, ioflg, cred);
483 		else
484 			error = VOP_WRITE(vp, &auio, ioflg, cred);
485 	}
486 	if (aresid)
487 		*aresid = auio.uio_resid;
488 	else
489 		if (auio.uio_resid && error == 0)
490 			error = EIO;
491 	if ((ioflg & IO_NODELOCKED) == 0) {
492 		VOP_UNLOCK(vp, 0);
493 		if (mp != NULL)
494 			vn_finished_write(mp);
495 	}
496  out:
497 	if (rl_cookie != NULL)
498 		vn_rangelock_unlock(vp, rl_cookie);
499 	return (error);
500 }
501 
502 /*
503  * Package up an I/O request on a vnode into a uio and do it.  The I/O
504  * request is split up into smaller chunks and we try to avoid saturating
505  * the buffer cache while potentially holding a vnode locked, so we
506  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
507  * to give other processes a chance to lock the vnode (either other processes
508  * core'ing the same binary, or unrelated processes scanning the directory).
509  */
510 int
511 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
512     file_cred, aresid, td)
513 	enum uio_rw rw;
514 	struct vnode *vp;
515 	void *base;
516 	size_t len;
517 	off_t offset;
518 	enum uio_seg segflg;
519 	int ioflg;
520 	struct ucred *active_cred;
521 	struct ucred *file_cred;
522 	size_t *aresid;
523 	struct thread *td;
524 {
525 	int error = 0;
526 	ssize_t iaresid;
527 
528 	do {
529 		int chunk;
530 
531 		/*
532 		 * Force `offset' to a multiple of MAXBSIZE except possibly
533 		 * for the first chunk, so that filesystems only need to
534 		 * write full blocks except possibly for the first and last
535 		 * chunks.
536 		 */
537 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
538 
539 		if (chunk > len)
540 			chunk = len;
541 		if (rw != UIO_READ && vp->v_type == VREG)
542 			bwillwrite();
543 		iaresid = 0;
544 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
545 		    ioflg, active_cred, file_cred, &iaresid, td);
546 		len -= chunk;	/* aresid calc already includes length */
547 		if (error)
548 			break;
549 		offset += chunk;
550 		base = (char *)base + chunk;
551 		kern_yield(PRI_USER);
552 	} while (len);
553 	if (aresid)
554 		*aresid = len + iaresid;
555 	return (error);
556 }
557 
558 off_t
559 foffset_lock(struct file *fp, int flags)
560 {
561 	struct mtx *mtxp;
562 	off_t res;
563 
564 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
565 
566 #if OFF_MAX <= LONG_MAX
567 	/*
568 	 * Caller only wants the current f_offset value.  Assume that
569 	 * the long and shorter integer types reads are atomic.
570 	 */
571 	if ((flags & FOF_NOLOCK) != 0)
572 		return (fp->f_offset);
573 #endif
574 
575 	/*
576 	 * According to McKusick the vn lock was protecting f_offset here.
577 	 * It is now protected by the FOFFSET_LOCKED flag.
578 	 */
579 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
580 	mtx_lock(mtxp);
581 	if ((flags & FOF_NOLOCK) == 0) {
582 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
583 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
584 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
585 			    "vofflock", 0);
586 		}
587 		fp->f_vnread_flags |= FOFFSET_LOCKED;
588 	}
589 	res = fp->f_offset;
590 	mtx_unlock(mtxp);
591 	return (res);
592 }
593 
594 void
595 foffset_unlock(struct file *fp, off_t val, int flags)
596 {
597 	struct mtx *mtxp;
598 
599 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
600 
601 #if OFF_MAX <= LONG_MAX
602 	if ((flags & FOF_NOLOCK) != 0) {
603 		if ((flags & FOF_NOUPDATE) == 0)
604 			fp->f_offset = val;
605 		if ((flags & FOF_NEXTOFF) != 0)
606 			fp->f_nextoff = val;
607 		return;
608 	}
609 #endif
610 
611 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
612 	mtx_lock(mtxp);
613 	if ((flags & FOF_NOUPDATE) == 0)
614 		fp->f_offset = val;
615 	if ((flags & FOF_NEXTOFF) != 0)
616 		fp->f_nextoff = val;
617 	if ((flags & FOF_NOLOCK) == 0) {
618 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
619 		    ("Lost FOFFSET_LOCKED"));
620 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
621 			wakeup(&fp->f_vnread_flags);
622 		fp->f_vnread_flags = 0;
623 	}
624 	mtx_unlock(mtxp);
625 }
626 
627 void
628 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
629 {
630 
631 	if ((flags & FOF_OFFSET) == 0)
632 		uio->uio_offset = foffset_lock(fp, flags);
633 }
634 
635 void
636 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
637 {
638 
639 	if ((flags & FOF_OFFSET) == 0)
640 		foffset_unlock(fp, uio->uio_offset, flags);
641 }
642 
643 static int
644 get_advice(struct file *fp, struct uio *uio)
645 {
646 	struct mtx *mtxp;
647 	int ret;
648 
649 	ret = POSIX_FADV_NORMAL;
650 	if (fp->f_advice == NULL)
651 		return (ret);
652 
653 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
654 	mtx_lock(mtxp);
655 	if (uio->uio_offset >= fp->f_advice->fa_start &&
656 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
657 		ret = fp->f_advice->fa_advice;
658 	mtx_unlock(mtxp);
659 	return (ret);
660 }
661 
662 /*
663  * File table vnode read routine.
664  */
665 static int
666 vn_read(fp, uio, active_cred, flags, td)
667 	struct file *fp;
668 	struct uio *uio;
669 	struct ucred *active_cred;
670 	int flags;
671 	struct thread *td;
672 {
673 	struct vnode *vp;
674 	struct mtx *mtxp;
675 	int error, ioflag;
676 	int advice;
677 	off_t offset, start, end;
678 
679 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
680 	    uio->uio_td, td));
681 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
682 	vp = fp->f_vnode;
683 	ioflag = 0;
684 	if (fp->f_flag & FNONBLOCK)
685 		ioflag |= IO_NDELAY;
686 	if (fp->f_flag & O_DIRECT)
687 		ioflag |= IO_DIRECT;
688 	advice = get_advice(fp, uio);
689 	vn_lock(vp, LK_SHARED | LK_RETRY);
690 
691 	switch (advice) {
692 	case POSIX_FADV_NORMAL:
693 	case POSIX_FADV_SEQUENTIAL:
694 	case POSIX_FADV_NOREUSE:
695 		ioflag |= sequential_heuristic(uio, fp);
696 		break;
697 	case POSIX_FADV_RANDOM:
698 		/* Disable read-ahead for random I/O. */
699 		break;
700 	}
701 	offset = uio->uio_offset;
702 
703 #ifdef MAC
704 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
705 	if (error == 0)
706 #endif
707 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
708 	fp->f_nextoff = uio->uio_offset;
709 	VOP_UNLOCK(vp, 0);
710 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
711 	    offset != uio->uio_offset) {
712 		/*
713 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
714 		 * buffers for the backing file after a
715 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
716 		 * case of using POSIX_FADV_NOREUSE with sequential
717 		 * access, track the previous implicit DONTNEED
718 		 * request and grow this request to include the
719 		 * current read(2) in addition to the previous
720 		 * DONTNEED.  With purely sequential access this will
721 		 * cause the DONTNEED requests to continously grow to
722 		 * cover all of the previously read regions of the
723 		 * file.  This allows filesystem blocks that are
724 		 * accessed by multiple calls to read(2) to be flushed
725 		 * once the last read(2) finishes.
726 		 */
727 		start = offset;
728 		end = uio->uio_offset - 1;
729 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
730 		mtx_lock(mtxp);
731 		if (fp->f_advice != NULL &&
732 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
733 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
734 				start = fp->f_advice->fa_prevstart;
735 			else if (fp->f_advice->fa_prevstart != 0 &&
736 			    fp->f_advice->fa_prevstart == end + 1)
737 				end = fp->f_advice->fa_prevend;
738 			fp->f_advice->fa_prevstart = start;
739 			fp->f_advice->fa_prevend = end;
740 		}
741 		mtx_unlock(mtxp);
742 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
743 	}
744 	return (error);
745 }
746 
747 /*
748  * File table vnode write routine.
749  */
750 static int
751 vn_write(fp, uio, active_cred, flags, td)
752 	struct file *fp;
753 	struct uio *uio;
754 	struct ucred *active_cred;
755 	int flags;
756 	struct thread *td;
757 {
758 	struct vnode *vp;
759 	struct mount *mp;
760 	struct mtx *mtxp;
761 	int error, ioflag, lock_flags;
762 	int advice;
763 	off_t offset, start, end;
764 
765 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
766 	    uio->uio_td, td));
767 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
768 	vp = fp->f_vnode;
769 	if (vp->v_type == VREG)
770 		bwillwrite();
771 	ioflag = IO_UNIT;
772 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
773 		ioflag |= IO_APPEND;
774 	if (fp->f_flag & FNONBLOCK)
775 		ioflag |= IO_NDELAY;
776 	if (fp->f_flag & O_DIRECT)
777 		ioflag |= IO_DIRECT;
778 	if ((fp->f_flag & O_FSYNC) ||
779 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
780 		ioflag |= IO_SYNC;
781 	mp = NULL;
782 	if (vp->v_type != VCHR &&
783 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
784 		goto unlock;
785 
786 	advice = get_advice(fp, uio);
787 
788 	if (MNT_SHARED_WRITES(mp) ||
789 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
790 		lock_flags = LK_SHARED;
791 	} else {
792 		lock_flags = LK_EXCLUSIVE;
793 	}
794 
795 	vn_lock(vp, lock_flags | LK_RETRY);
796 	switch (advice) {
797 	case POSIX_FADV_NORMAL:
798 	case POSIX_FADV_SEQUENTIAL:
799 	case POSIX_FADV_NOREUSE:
800 		ioflag |= sequential_heuristic(uio, fp);
801 		break;
802 	case POSIX_FADV_RANDOM:
803 		/* XXX: Is this correct? */
804 		break;
805 	}
806 	offset = uio->uio_offset;
807 
808 #ifdef MAC
809 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
810 	if (error == 0)
811 #endif
812 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
813 	fp->f_nextoff = uio->uio_offset;
814 	VOP_UNLOCK(vp, 0);
815 	if (vp->v_type != VCHR)
816 		vn_finished_write(mp);
817 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
818 	    offset != uio->uio_offset) {
819 		/*
820 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
821 		 * buffers for the backing file after a
822 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
823 		 * common case of using POSIX_FADV_NOREUSE with
824 		 * sequential access, track the previous implicit
825 		 * DONTNEED request and grow this request to include
826 		 * the current write(2) in addition to the previous
827 		 * DONTNEED.  With purely sequential access this will
828 		 * cause the DONTNEED requests to continously grow to
829 		 * cover all of the previously written regions of the
830 		 * file.
831 		 *
832 		 * Note that the blocks just written are almost
833 		 * certainly still dirty, so this only works when
834 		 * VOP_ADVISE() calls from subsequent writes push out
835 		 * the data written by this write(2) once the backing
836 		 * buffers are clean.  However, as compared to forcing
837 		 * IO_DIRECT, this gives much saner behavior.  Write
838 		 * clustering is still allowed, and clean pages are
839 		 * merely moved to the cache page queue rather than
840 		 * outright thrown away.  This means a subsequent
841 		 * read(2) can still avoid hitting the disk if the
842 		 * pages have not been reclaimed.
843 		 *
844 		 * This does make POSIX_FADV_NOREUSE largely useless
845 		 * with non-sequential access.  However, sequential
846 		 * access is the more common use case and the flag is
847 		 * merely advisory.
848 		 */
849 		start = offset;
850 		end = uio->uio_offset - 1;
851 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
852 		mtx_lock(mtxp);
853 		if (fp->f_advice != NULL &&
854 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
855 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
856 				start = fp->f_advice->fa_prevstart;
857 			else if (fp->f_advice->fa_prevstart != 0 &&
858 			    fp->f_advice->fa_prevstart == end + 1)
859 				end = fp->f_advice->fa_prevend;
860 			fp->f_advice->fa_prevstart = start;
861 			fp->f_advice->fa_prevend = end;
862 		}
863 		mtx_unlock(mtxp);
864 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
865 	}
866 
867 unlock:
868 	return (error);
869 }
870 
871 static const int io_hold_cnt = 16;
872 static int vn_io_fault_enable = 1;
873 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
874     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
875 static unsigned long vn_io_faults_cnt;
876 SYSCTL_LONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
877     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
878 
879 /*
880  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
881  * prevent the following deadlock:
882  *
883  * Assume that the thread A reads from the vnode vp1 into userspace
884  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
885  * currently not resident, then system ends up with the call chain
886  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
887  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
888  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
889  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
890  * backed by the pages of vnode vp1, and some page in buf2 is not
891  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
892  *
893  * To prevent the lock order reversal and deadlock, vn_io_fault() does
894  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
895  * Instead, it first tries to do the whole range i/o with pagefaults
896  * disabled. If all pages in the i/o buffer are resident and mapped,
897  * VOP will succeed (ignoring the genuine filesystem errors).
898  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
899  * i/o in chunks, with all pages in the chunk prefaulted and held
900  * using vm_fault_quick_hold_pages().
901  *
902  * Filesystems using this deadlock avoidance scheme should use the
903  * array of the held pages from uio, saved in the curthread->td_ma,
904  * instead of doing uiomove().  A helper function
905  * vn_io_fault_uiomove() converts uiomove request into
906  * uiomove_fromphys() over td_ma array.
907  *
908  * Since vnode locks do not cover the whole i/o anymore, rangelocks
909  * make the current i/o request atomic with respect to other i/os and
910  * truncations.
911  */
912 static int
913 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
914     int flags, struct thread *td)
915 {
916 	vm_page_t ma[io_hold_cnt + 2];
917 	struct uio *uio_clone, short_uio;
918 	struct iovec short_iovec[1];
919 	fo_rdwr_t *doio;
920 	struct vnode *vp;
921 	void *rl_cookie;
922 	struct mount *mp;
923 	vm_page_t *prev_td_ma;
924 	int cnt, error, save, saveheld, prev_td_ma_cnt;
925 	vm_offset_t addr, end;
926 	vm_prot_t prot;
927 	size_t len, resid;
928 	ssize_t adv;
929 
930 	if (uio->uio_rw == UIO_READ)
931 		doio = vn_read;
932 	else
933 		doio = vn_write;
934 	vp = fp->f_vnode;
935 	foffset_lock_uio(fp, uio, flags);
936 
937 	if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
938 	    ((mp = vp->v_mount) != NULL &&
939 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
940 	    !vn_io_fault_enable) {
941 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
942 		goto out_last;
943 	}
944 
945 	/*
946 	 * The UFS follows IO_UNIT directive and replays back both
947 	 * uio_offset and uio_resid if an error is encountered during the
948 	 * operation.  But, since the iovec may be already advanced,
949 	 * uio is still in an inconsistent state.
950 	 *
951 	 * Cache a copy of the original uio, which is advanced to the redo
952 	 * point using UIO_NOCOPY below.
953 	 */
954 	uio_clone = cloneuio(uio);
955 	resid = uio->uio_resid;
956 
957 	short_uio.uio_segflg = UIO_USERSPACE;
958 	short_uio.uio_rw = uio->uio_rw;
959 	short_uio.uio_td = uio->uio_td;
960 
961 	if (uio->uio_rw == UIO_READ) {
962 		prot = VM_PROT_WRITE;
963 		rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
964 		    uio->uio_offset + uio->uio_resid);
965 	} else {
966 		prot = VM_PROT_READ;
967 		if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
968 			/* For appenders, punt and lock the whole range. */
969 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
970 		else
971 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
972 			    uio->uio_offset + uio->uio_resid);
973 	}
974 
975 	save = vm_fault_disable_pagefaults();
976 	error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
977 	if (error != EFAULT)
978 		goto out;
979 
980 	atomic_add_long(&vn_io_faults_cnt, 1);
981 	uio_clone->uio_segflg = UIO_NOCOPY;
982 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
983 	uio_clone->uio_segflg = uio->uio_segflg;
984 
985 	saveheld = curthread_pflags_set(TDP_UIOHELD);
986 	prev_td_ma = td->td_ma;
987 	prev_td_ma_cnt = td->td_ma_cnt;
988 
989 	while (uio_clone->uio_resid != 0) {
990 		len = uio_clone->uio_iov->iov_len;
991 		if (len == 0) {
992 			KASSERT(uio_clone->uio_iovcnt >= 1,
993 			    ("iovcnt underflow"));
994 			uio_clone->uio_iov++;
995 			uio_clone->uio_iovcnt--;
996 			continue;
997 		}
998 
999 		addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
1000 		end = round_page(addr + len);
1001 		cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
1002 		/*
1003 		 * A perfectly misaligned address and length could cause
1004 		 * both the start and the end of the chunk to use partial
1005 		 * page.  +2 accounts for such a situation.
1006 		 */
1007 		if (cnt > io_hold_cnt + 2) {
1008 			len = io_hold_cnt * PAGE_SIZE;
1009 			KASSERT(howmany(round_page(addr + len) -
1010 			    trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
1011 			    ("cnt overflow"));
1012 		}
1013 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1014 		    addr, len, prot, ma, io_hold_cnt + 2);
1015 		if (cnt == -1) {
1016 			error = EFAULT;
1017 			break;
1018 		}
1019 		short_uio.uio_iov = &short_iovec[0];
1020 		short_iovec[0].iov_base = (void *)addr;
1021 		short_uio.uio_iovcnt = 1;
1022 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1023 		short_uio.uio_offset = uio_clone->uio_offset;
1024 		td->td_ma = ma;
1025 		td->td_ma_cnt = cnt;
1026 
1027 		error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
1028 		    td);
1029 		vm_page_unhold_pages(ma, cnt);
1030 		adv = len - short_uio.uio_resid;
1031 
1032 		uio_clone->uio_iov->iov_base =
1033 		    (char *)uio_clone->uio_iov->iov_base + adv;
1034 		uio_clone->uio_iov->iov_len -= adv;
1035 		uio_clone->uio_resid -= adv;
1036 		uio_clone->uio_offset += adv;
1037 
1038 		uio->uio_resid -= adv;
1039 		uio->uio_offset += adv;
1040 
1041 		if (error != 0 || adv == 0)
1042 			break;
1043 	}
1044 	td->td_ma = prev_td_ma;
1045 	td->td_ma_cnt = prev_td_ma_cnt;
1046 	curthread_pflags_restore(saveheld);
1047 out:
1048 	vm_fault_enable_pagefaults(save);
1049 	vn_rangelock_unlock(vp, rl_cookie);
1050 	free(uio_clone, M_IOV);
1051 out_last:
1052 	foffset_unlock_uio(fp, uio, flags);
1053 	return (error);
1054 }
1055 
1056 /*
1057  * Helper function to perform the requested uiomove operation using
1058  * the held pages for io->uio_iov[0].iov_base buffer instead of
1059  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1060  * instead of iov_base prevents page faults that could occur due to
1061  * pmap_collect() invalidating the mapping created by
1062  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1063  * object cleanup revoking the write access from page mappings.
1064  *
1065  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1066  * instead of plain uiomove().
1067  */
1068 int
1069 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1070 {
1071 	struct uio transp_uio;
1072 	struct iovec transp_iov[1];
1073 	struct thread *td;
1074 	size_t adv;
1075 	int error, pgadv;
1076 
1077 	td = curthread;
1078 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1079 	    uio->uio_segflg != UIO_USERSPACE)
1080 		return (uiomove(data, xfersize, uio));
1081 
1082 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1083 	transp_iov[0].iov_base = data;
1084 	transp_uio.uio_iov = &transp_iov[0];
1085 	transp_uio.uio_iovcnt = 1;
1086 	if (xfersize > uio->uio_resid)
1087 		xfersize = uio->uio_resid;
1088 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1089 	transp_uio.uio_offset = 0;
1090 	transp_uio.uio_segflg = UIO_SYSSPACE;
1091 	/*
1092 	 * Since transp_iov points to data, and td_ma page array
1093 	 * corresponds to original uio->uio_iov, we need to invert the
1094 	 * direction of the i/o operation as passed to
1095 	 * uiomove_fromphys().
1096 	 */
1097 	switch (uio->uio_rw) {
1098 	case UIO_WRITE:
1099 		transp_uio.uio_rw = UIO_READ;
1100 		break;
1101 	case UIO_READ:
1102 		transp_uio.uio_rw = UIO_WRITE;
1103 		break;
1104 	}
1105 	transp_uio.uio_td = uio->uio_td;
1106 	error = uiomove_fromphys(td->td_ma,
1107 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1108 	    xfersize, &transp_uio);
1109 	adv = xfersize - transp_uio.uio_resid;
1110 	pgadv =
1111 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1112 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1113 	td->td_ma += pgadv;
1114 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1115 	    pgadv));
1116 	td->td_ma_cnt -= pgadv;
1117 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1118 	uio->uio_iov->iov_len -= adv;
1119 	uio->uio_resid -= adv;
1120 	uio->uio_offset += adv;
1121 	return (error);
1122 }
1123 
1124 /*
1125  * File table truncate routine.
1126  */
1127 static int
1128 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1129     struct thread *td)
1130 {
1131 	struct vattr vattr;
1132 	struct mount *mp;
1133 	struct vnode *vp;
1134 	void *rl_cookie;
1135 	int error;
1136 
1137 	vp = fp->f_vnode;
1138 
1139 	/*
1140 	 * Lock the whole range for truncation.  Otherwise split i/o
1141 	 * might happen partly before and partly after the truncation.
1142 	 */
1143 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1144 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1145 	if (error)
1146 		goto out1;
1147 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1148 	if (vp->v_type == VDIR) {
1149 		error = EISDIR;
1150 		goto out;
1151 	}
1152 #ifdef MAC
1153 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1154 	if (error)
1155 		goto out;
1156 #endif
1157 	error = vn_writechk(vp);
1158 	if (error == 0) {
1159 		VATTR_NULL(&vattr);
1160 		vattr.va_size = length;
1161 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1162 	}
1163 out:
1164 	VOP_UNLOCK(vp, 0);
1165 	vn_finished_write(mp);
1166 out1:
1167 	vn_rangelock_unlock(vp, rl_cookie);
1168 	return (error);
1169 }
1170 
1171 /*
1172  * File table vnode stat routine.
1173  */
1174 static int
1175 vn_statfile(fp, sb, active_cred, td)
1176 	struct file *fp;
1177 	struct stat *sb;
1178 	struct ucred *active_cred;
1179 	struct thread *td;
1180 {
1181 	struct vnode *vp = fp->f_vnode;
1182 	int error;
1183 
1184 	vn_lock(vp, LK_SHARED | LK_RETRY);
1185 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1186 	VOP_UNLOCK(vp, 0);
1187 
1188 	return (error);
1189 }
1190 
1191 /*
1192  * Stat a vnode; implementation for the stat syscall
1193  */
1194 int
1195 vn_stat(vp, sb, active_cred, file_cred, td)
1196 	struct vnode *vp;
1197 	register struct stat *sb;
1198 	struct ucred *active_cred;
1199 	struct ucred *file_cred;
1200 	struct thread *td;
1201 {
1202 	struct vattr vattr;
1203 	register struct vattr *vap;
1204 	int error;
1205 	u_short mode;
1206 
1207 #ifdef MAC
1208 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1209 	if (error)
1210 		return (error);
1211 #endif
1212 
1213 	vap = &vattr;
1214 
1215 	/*
1216 	 * Initialize defaults for new and unusual fields, so that file
1217 	 * systems which don't support these fields don't need to know
1218 	 * about them.
1219 	 */
1220 	vap->va_birthtime.tv_sec = -1;
1221 	vap->va_birthtime.tv_nsec = 0;
1222 	vap->va_fsid = VNOVAL;
1223 	vap->va_rdev = NODEV;
1224 
1225 	error = VOP_GETATTR(vp, vap, active_cred);
1226 	if (error)
1227 		return (error);
1228 
1229 	/*
1230 	 * Zero the spare stat fields
1231 	 */
1232 	bzero(sb, sizeof *sb);
1233 
1234 	/*
1235 	 * Copy from vattr table
1236 	 */
1237 	if (vap->va_fsid != VNOVAL)
1238 		sb->st_dev = vap->va_fsid;
1239 	else
1240 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1241 	sb->st_ino = vap->va_fileid;
1242 	mode = vap->va_mode;
1243 	switch (vap->va_type) {
1244 	case VREG:
1245 		mode |= S_IFREG;
1246 		break;
1247 	case VDIR:
1248 		mode |= S_IFDIR;
1249 		break;
1250 	case VBLK:
1251 		mode |= S_IFBLK;
1252 		break;
1253 	case VCHR:
1254 		mode |= S_IFCHR;
1255 		break;
1256 	case VLNK:
1257 		mode |= S_IFLNK;
1258 		break;
1259 	case VSOCK:
1260 		mode |= S_IFSOCK;
1261 		break;
1262 	case VFIFO:
1263 		mode |= S_IFIFO;
1264 		break;
1265 	default:
1266 		return (EBADF);
1267 	};
1268 	sb->st_mode = mode;
1269 	sb->st_nlink = vap->va_nlink;
1270 	sb->st_uid = vap->va_uid;
1271 	sb->st_gid = vap->va_gid;
1272 	sb->st_rdev = vap->va_rdev;
1273 	if (vap->va_size > OFF_MAX)
1274 		return (EOVERFLOW);
1275 	sb->st_size = vap->va_size;
1276 	sb->st_atim = vap->va_atime;
1277 	sb->st_mtim = vap->va_mtime;
1278 	sb->st_ctim = vap->va_ctime;
1279 	sb->st_birthtim = vap->va_birthtime;
1280 
1281         /*
1282 	 * According to www.opengroup.org, the meaning of st_blksize is
1283 	 *   "a filesystem-specific preferred I/O block size for this
1284 	 *    object.  In some filesystem types, this may vary from file
1285 	 *    to file"
1286 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1287 	 */
1288 
1289 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1290 
1291 	sb->st_flags = vap->va_flags;
1292 	if (priv_check(td, PRIV_VFS_GENERATION))
1293 		sb->st_gen = 0;
1294 	else
1295 		sb->st_gen = vap->va_gen;
1296 
1297 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1298 	return (0);
1299 }
1300 
1301 /*
1302  * File table vnode ioctl routine.
1303  */
1304 static int
1305 vn_ioctl(fp, com, data, active_cred, td)
1306 	struct file *fp;
1307 	u_long com;
1308 	void *data;
1309 	struct ucred *active_cred;
1310 	struct thread *td;
1311 {
1312 	struct vnode *vp = fp->f_vnode;
1313 	struct vattr vattr;
1314 	int error;
1315 
1316 	error = ENOTTY;
1317 	switch (vp->v_type) {
1318 	case VREG:
1319 	case VDIR:
1320 		if (com == FIONREAD) {
1321 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1322 			error = VOP_GETATTR(vp, &vattr, active_cred);
1323 			VOP_UNLOCK(vp, 0);
1324 			if (!error)
1325 				*(int *)data = vattr.va_size - fp->f_offset;
1326 		}
1327 		if (com == FIONBIO || com == FIOASYNC)	/* XXX */
1328 			error = 0;
1329 		else
1330 			error = VOP_IOCTL(vp, com, data, fp->f_flag,
1331 			    active_cred, td);
1332 		break;
1333 
1334 	default:
1335 		break;
1336 	}
1337 	return (error);
1338 }
1339 
1340 /*
1341  * File table vnode poll routine.
1342  */
1343 static int
1344 vn_poll(fp, events, active_cred, td)
1345 	struct file *fp;
1346 	int events;
1347 	struct ucred *active_cred;
1348 	struct thread *td;
1349 {
1350 	struct vnode *vp;
1351 	int error;
1352 
1353 	vp = fp->f_vnode;
1354 #ifdef MAC
1355 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1356 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1357 	VOP_UNLOCK(vp, 0);
1358 	if (!error)
1359 #endif
1360 
1361 	error = VOP_POLL(vp, events, fp->f_cred, td);
1362 	return (error);
1363 }
1364 
1365 /*
1366  * Acquire the requested lock and then check for validity.  LK_RETRY
1367  * permits vn_lock to return doomed vnodes.
1368  */
1369 int
1370 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1371 {
1372 	int error;
1373 
1374 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1375 	    ("vn_lock called with no locktype."));
1376 	do {
1377 #ifdef DEBUG_VFS_LOCKS
1378 		KASSERT(vp->v_holdcnt != 0,
1379 		    ("vn_lock %p: zero hold count", vp));
1380 #endif
1381 		error = VOP_LOCK1(vp, flags, file, line);
1382 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1383 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1384 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1385 		    flags, error));
1386 		/*
1387 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1388 		 * If RETRY is not set, we return ENOENT instead.
1389 		 */
1390 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1391 		    (flags & LK_RETRY) == 0) {
1392 			VOP_UNLOCK(vp, 0);
1393 			error = ENOENT;
1394 			break;
1395 		}
1396 	} while (flags & LK_RETRY && error != 0);
1397 	return (error);
1398 }
1399 
1400 /*
1401  * File table vnode close routine.
1402  */
1403 static int
1404 vn_closefile(fp, td)
1405 	struct file *fp;
1406 	struct thread *td;
1407 {
1408 	struct vnode *vp;
1409 	struct flock lf;
1410 	int error;
1411 
1412 	vp = fp->f_vnode;
1413 	fp->f_ops = &badfileops;
1414 
1415 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1416 		vref(vp);
1417 
1418 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1419 
1420 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1421 		lf.l_whence = SEEK_SET;
1422 		lf.l_start = 0;
1423 		lf.l_len = 0;
1424 		lf.l_type = F_UNLCK;
1425 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1426 		vrele(vp);
1427 	}
1428 	return (error);
1429 }
1430 
1431 /*
1432  * Preparing to start a filesystem write operation. If the operation is
1433  * permitted, then we bump the count of operations in progress and
1434  * proceed. If a suspend request is in progress, we wait until the
1435  * suspension is over, and then proceed.
1436  */
1437 static int
1438 vn_start_write_locked(struct mount *mp, int flags)
1439 {
1440 	int error;
1441 
1442 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1443 	error = 0;
1444 
1445 	/*
1446 	 * Check on status of suspension.
1447 	 */
1448 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1449 	    mp->mnt_susp_owner != curthread) {
1450 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1451 			if (flags & V_NOWAIT) {
1452 				error = EWOULDBLOCK;
1453 				goto unlock;
1454 			}
1455 			error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1456 			    (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1457 			if (error)
1458 				goto unlock;
1459 		}
1460 	}
1461 	if (flags & V_XSLEEP)
1462 		goto unlock;
1463 	mp->mnt_writeopcount++;
1464 unlock:
1465 	if (error != 0 || (flags & V_XSLEEP) != 0)
1466 		MNT_REL(mp);
1467 	MNT_IUNLOCK(mp);
1468 	return (error);
1469 }
1470 
1471 int
1472 vn_start_write(vp, mpp, flags)
1473 	struct vnode *vp;
1474 	struct mount **mpp;
1475 	int flags;
1476 {
1477 	struct mount *mp;
1478 	int error;
1479 
1480 	error = 0;
1481 	/*
1482 	 * If a vnode is provided, get and return the mount point that
1483 	 * to which it will write.
1484 	 */
1485 	if (vp != NULL) {
1486 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1487 			*mpp = NULL;
1488 			if (error != EOPNOTSUPP)
1489 				return (error);
1490 			return (0);
1491 		}
1492 	}
1493 	if ((mp = *mpp) == NULL)
1494 		return (0);
1495 
1496 	/*
1497 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1498 	 * a vfs_ref().
1499 	 * As long as a vnode is not provided we need to acquire a
1500 	 * refcount for the provided mountpoint too, in order to
1501 	 * emulate a vfs_ref().
1502 	 */
1503 	MNT_ILOCK(mp);
1504 	if (vp == NULL)
1505 		MNT_REF(mp);
1506 
1507 	return (vn_start_write_locked(mp, flags));
1508 }
1509 
1510 /*
1511  * Secondary suspension. Used by operations such as vop_inactive
1512  * routines that are needed by the higher level functions. These
1513  * are allowed to proceed until all the higher level functions have
1514  * completed (indicated by mnt_writeopcount dropping to zero). At that
1515  * time, these operations are halted until the suspension is over.
1516  */
1517 int
1518 vn_start_secondary_write(vp, mpp, flags)
1519 	struct vnode *vp;
1520 	struct mount **mpp;
1521 	int flags;
1522 {
1523 	struct mount *mp;
1524 	int error;
1525 
1526  retry:
1527 	if (vp != NULL) {
1528 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1529 			*mpp = NULL;
1530 			if (error != EOPNOTSUPP)
1531 				return (error);
1532 			return (0);
1533 		}
1534 	}
1535 	/*
1536 	 * If we are not suspended or have not yet reached suspended
1537 	 * mode, then let the operation proceed.
1538 	 */
1539 	if ((mp = *mpp) == NULL)
1540 		return (0);
1541 
1542 	/*
1543 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1544 	 * a vfs_ref().
1545 	 * As long as a vnode is not provided we need to acquire a
1546 	 * refcount for the provided mountpoint too, in order to
1547 	 * emulate a vfs_ref().
1548 	 */
1549 	MNT_ILOCK(mp);
1550 	if (vp == NULL)
1551 		MNT_REF(mp);
1552 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1553 		mp->mnt_secondary_writes++;
1554 		mp->mnt_secondary_accwrites++;
1555 		MNT_IUNLOCK(mp);
1556 		return (0);
1557 	}
1558 	if (flags & V_NOWAIT) {
1559 		MNT_REL(mp);
1560 		MNT_IUNLOCK(mp);
1561 		return (EWOULDBLOCK);
1562 	}
1563 	/*
1564 	 * Wait for the suspension to finish.
1565 	 */
1566 	error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1567 		       (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1568 	vfs_rel(mp);
1569 	if (error == 0)
1570 		goto retry;
1571 	return (error);
1572 }
1573 
1574 /*
1575  * Filesystem write operation has completed. If we are suspending and this
1576  * operation is the last one, notify the suspender that the suspension is
1577  * now in effect.
1578  */
1579 void
1580 vn_finished_write(mp)
1581 	struct mount *mp;
1582 {
1583 	if (mp == NULL)
1584 		return;
1585 	MNT_ILOCK(mp);
1586 	MNT_REL(mp);
1587 	mp->mnt_writeopcount--;
1588 	if (mp->mnt_writeopcount < 0)
1589 		panic("vn_finished_write: neg cnt");
1590 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1591 	    mp->mnt_writeopcount <= 0)
1592 		wakeup(&mp->mnt_writeopcount);
1593 	MNT_IUNLOCK(mp);
1594 }
1595 
1596 
1597 /*
1598  * Filesystem secondary write operation has completed. If we are
1599  * suspending and this operation is the last one, notify the suspender
1600  * that the suspension is now in effect.
1601  */
1602 void
1603 vn_finished_secondary_write(mp)
1604 	struct mount *mp;
1605 {
1606 	if (mp == NULL)
1607 		return;
1608 	MNT_ILOCK(mp);
1609 	MNT_REL(mp);
1610 	mp->mnt_secondary_writes--;
1611 	if (mp->mnt_secondary_writes < 0)
1612 		panic("vn_finished_secondary_write: neg cnt");
1613 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1614 	    mp->mnt_secondary_writes <= 0)
1615 		wakeup(&mp->mnt_secondary_writes);
1616 	MNT_IUNLOCK(mp);
1617 }
1618 
1619 
1620 
1621 /*
1622  * Request a filesystem to suspend write operations.
1623  */
1624 int
1625 vfs_write_suspend(mp)
1626 	struct mount *mp;
1627 {
1628 	int error;
1629 
1630 	MNT_ILOCK(mp);
1631 	if (mp->mnt_susp_owner == curthread) {
1632 		MNT_IUNLOCK(mp);
1633 		return (EALREADY);
1634 	}
1635 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1636 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1637 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1638 	mp->mnt_susp_owner = curthread;
1639 	if (mp->mnt_writeopcount > 0)
1640 		(void) msleep(&mp->mnt_writeopcount,
1641 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1642 	else
1643 		MNT_IUNLOCK(mp);
1644 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1645 		vfs_write_resume(mp, 0);
1646 	return (error);
1647 }
1648 
1649 /*
1650  * Request a filesystem to resume write operations.
1651  */
1652 void
1653 vfs_write_resume(struct mount *mp, int flags)
1654 {
1655 
1656 	MNT_ILOCK(mp);
1657 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1658 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1659 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1660 				       MNTK_SUSPENDED);
1661 		mp->mnt_susp_owner = NULL;
1662 		wakeup(&mp->mnt_writeopcount);
1663 		wakeup(&mp->mnt_flag);
1664 		curthread->td_pflags &= ~TDP_IGNSUSP;
1665 		if ((flags & VR_START_WRITE) != 0) {
1666 			MNT_REF(mp);
1667 			mp->mnt_writeopcount++;
1668 		}
1669 		MNT_IUNLOCK(mp);
1670 		if ((flags & VR_NO_SUSPCLR) == 0)
1671 			VFS_SUSP_CLEAN(mp);
1672 	} else if ((flags & VR_START_WRITE) != 0) {
1673 		MNT_REF(mp);
1674 		vn_start_write_locked(mp, 0);
1675 	} else {
1676 		MNT_IUNLOCK(mp);
1677 	}
1678 }
1679 
1680 /*
1681  * Implement kqueues for files by translating it to vnode operation.
1682  */
1683 static int
1684 vn_kqfilter(struct file *fp, struct knote *kn)
1685 {
1686 
1687 	return (VOP_KQFILTER(fp->f_vnode, kn));
1688 }
1689 
1690 /*
1691  * Simplified in-kernel wrapper calls for extended attribute access.
1692  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1693  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1694  */
1695 int
1696 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1697     const char *attrname, int *buflen, char *buf, struct thread *td)
1698 {
1699 	struct uio	auio;
1700 	struct iovec	iov;
1701 	int	error;
1702 
1703 	iov.iov_len = *buflen;
1704 	iov.iov_base = buf;
1705 
1706 	auio.uio_iov = &iov;
1707 	auio.uio_iovcnt = 1;
1708 	auio.uio_rw = UIO_READ;
1709 	auio.uio_segflg = UIO_SYSSPACE;
1710 	auio.uio_td = td;
1711 	auio.uio_offset = 0;
1712 	auio.uio_resid = *buflen;
1713 
1714 	if ((ioflg & IO_NODELOCKED) == 0)
1715 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1716 
1717 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1718 
1719 	/* authorize attribute retrieval as kernel */
1720 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1721 	    td);
1722 
1723 	if ((ioflg & IO_NODELOCKED) == 0)
1724 		VOP_UNLOCK(vp, 0);
1725 
1726 	if (error == 0) {
1727 		*buflen = *buflen - auio.uio_resid;
1728 	}
1729 
1730 	return (error);
1731 }
1732 
1733 /*
1734  * XXX failure mode if partially written?
1735  */
1736 int
1737 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1738     const char *attrname, int buflen, char *buf, struct thread *td)
1739 {
1740 	struct uio	auio;
1741 	struct iovec	iov;
1742 	struct mount	*mp;
1743 	int	error;
1744 
1745 	iov.iov_len = buflen;
1746 	iov.iov_base = buf;
1747 
1748 	auio.uio_iov = &iov;
1749 	auio.uio_iovcnt = 1;
1750 	auio.uio_rw = UIO_WRITE;
1751 	auio.uio_segflg = UIO_SYSSPACE;
1752 	auio.uio_td = td;
1753 	auio.uio_offset = 0;
1754 	auio.uio_resid = buflen;
1755 
1756 	if ((ioflg & IO_NODELOCKED) == 0) {
1757 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1758 			return (error);
1759 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1760 	}
1761 
1762 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1763 
1764 	/* authorize attribute setting as kernel */
1765 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1766 
1767 	if ((ioflg & IO_NODELOCKED) == 0) {
1768 		vn_finished_write(mp);
1769 		VOP_UNLOCK(vp, 0);
1770 	}
1771 
1772 	return (error);
1773 }
1774 
1775 int
1776 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1777     const char *attrname, struct thread *td)
1778 {
1779 	struct mount	*mp;
1780 	int	error;
1781 
1782 	if ((ioflg & IO_NODELOCKED) == 0) {
1783 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1784 			return (error);
1785 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1786 	}
1787 
1788 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1789 
1790 	/* authorize attribute removal as kernel */
1791 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1792 	if (error == EOPNOTSUPP)
1793 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1794 		    NULL, td);
1795 
1796 	if ((ioflg & IO_NODELOCKED) == 0) {
1797 		vn_finished_write(mp);
1798 		VOP_UNLOCK(vp, 0);
1799 	}
1800 
1801 	return (error);
1802 }
1803 
1804 int
1805 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1806 {
1807 	struct mount *mp;
1808 	int ltype, error;
1809 
1810 	mp = vp->v_mount;
1811 	ltype = VOP_ISLOCKED(vp);
1812 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1813 	    ("vn_vget_ino: vp not locked"));
1814 	error = vfs_busy(mp, MBF_NOWAIT);
1815 	if (error != 0) {
1816 		vfs_ref(mp);
1817 		VOP_UNLOCK(vp, 0);
1818 		error = vfs_busy(mp, 0);
1819 		vn_lock(vp, ltype | LK_RETRY);
1820 		vfs_rel(mp);
1821 		if (error != 0)
1822 			return (ENOENT);
1823 		if (vp->v_iflag & VI_DOOMED) {
1824 			vfs_unbusy(mp);
1825 			return (ENOENT);
1826 		}
1827 	}
1828 	VOP_UNLOCK(vp, 0);
1829 	error = VFS_VGET(mp, ino, lkflags, rvp);
1830 	vfs_unbusy(mp);
1831 	vn_lock(vp, ltype | LK_RETRY);
1832 	if (vp->v_iflag & VI_DOOMED) {
1833 		if (error == 0)
1834 			vput(*rvp);
1835 		error = ENOENT;
1836 	}
1837 	return (error);
1838 }
1839 
1840 int
1841 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1842     const struct thread *td)
1843 {
1844 
1845 	if (vp->v_type != VREG || td == NULL)
1846 		return (0);
1847 	PROC_LOCK(td->td_proc);
1848 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1849 	    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1850 		kern_psignal(td->td_proc, SIGXFSZ);
1851 		PROC_UNLOCK(td->td_proc);
1852 		return (EFBIG);
1853 	}
1854 	PROC_UNLOCK(td->td_proc);
1855 	return (0);
1856 }
1857 
1858 int
1859 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1860     struct thread *td)
1861 {
1862 	struct vnode *vp;
1863 	int error;
1864 
1865 	vp = fp->f_vnode;
1866 #ifdef AUDIT
1867 	vn_lock(vp, LK_SHARED | LK_RETRY);
1868 	AUDIT_ARG_VNODE1(vp);
1869 	VOP_UNLOCK(vp, 0);
1870 #endif
1871 	error = setfmode(td, active_cred, vp, mode);
1872 	return (error);
1873 }
1874 
1875 int
1876 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1877     struct thread *td)
1878 {
1879 	struct vnode *vp;
1880 	int error;
1881 
1882 	vp = fp->f_vnode;
1883 #ifdef AUDIT
1884 	vn_lock(vp, LK_SHARED | LK_RETRY);
1885 	AUDIT_ARG_VNODE1(vp);
1886 	VOP_UNLOCK(vp, 0);
1887 #endif
1888 	error = setfown(td, active_cred, vp, uid, gid);
1889 	return (error);
1890 }
1891 
1892 void
1893 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1894 {
1895 	vm_object_t object;
1896 
1897 	if ((object = vp->v_object) == NULL)
1898 		return;
1899 	VM_OBJECT_LOCK(object);
1900 	vm_object_page_remove(object, start, end, 0);
1901 	VM_OBJECT_UNLOCK(object);
1902 }
1903 
1904 int
1905 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1906 {
1907 	struct vattr va;
1908 	daddr_t bn, bnp;
1909 	uint64_t bsize;
1910 	off_t noff;
1911 	int error;
1912 
1913 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1914 	    ("Wrong command %lu", cmd));
1915 
1916 	if (vn_lock(vp, LK_SHARED) != 0)
1917 		return (EBADF);
1918 	if (vp->v_type != VREG) {
1919 		error = ENOTTY;
1920 		goto unlock;
1921 	}
1922 	error = VOP_GETATTR(vp, &va, cred);
1923 	if (error != 0)
1924 		goto unlock;
1925 	noff = *off;
1926 	if (noff >= va.va_size) {
1927 		error = ENXIO;
1928 		goto unlock;
1929 	}
1930 	bsize = vp->v_mount->mnt_stat.f_iosize;
1931 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1932 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1933 		if (error == EOPNOTSUPP) {
1934 			error = ENOTTY;
1935 			goto unlock;
1936 		}
1937 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1938 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
1939 			noff = bn * bsize;
1940 			if (noff < *off)
1941 				noff = *off;
1942 			goto unlock;
1943 		}
1944 	}
1945 	if (noff > va.va_size)
1946 		noff = va.va_size;
1947 	/* noff == va.va_size. There is an implicit hole at the end of file. */
1948 	if (cmd == FIOSEEKDATA)
1949 		error = ENXIO;
1950 unlock:
1951 	VOP_UNLOCK(vp, 0);
1952 	if (error == 0)
1953 		*off = noff;
1954 	return (error);
1955 }
1956