xref: /freebsd/sys/kern/vfs_vnops.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
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  * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11  * Copyright (c) 2013 The FreeBSD Foundation
12  *
13  * Portions of this software were developed by Konstantin Belousov
14  * under sponsorship from the FreeBSD Foundation.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)vfs_vnops.c	8.2 (Berkeley) 1/21/94
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/kdb.h>
51 #include <sys/stat.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/limits.h>
55 #include <sys/lock.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
58 #include <sys/namei.h>
59 #include <sys/vnode.h>
60 #include <sys/bio.h>
61 #include <sys/buf.h>
62 #include <sys/filio.h>
63 #include <sys/resourcevar.h>
64 #include <sys/rwlock.h>
65 #include <sys/sx.h>
66 #include <sys/sysctl.h>
67 #include <sys/ttycom.h>
68 #include <sys/conf.h>
69 #include <sys/syslog.h>
70 #include <sys/unistd.h>
71 
72 #include <security/audit/audit.h>
73 #include <security/mac/mac_framework.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_extern.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_page.h>
81 
82 static fo_rdwr_t	vn_read;
83 static fo_rdwr_t	vn_write;
84 static fo_rdwr_t	vn_io_fault;
85 static fo_truncate_t	vn_truncate;
86 static fo_ioctl_t	vn_ioctl;
87 static fo_poll_t	vn_poll;
88 static fo_kqfilter_t	vn_kqfilter;
89 static fo_stat_t	vn_statfile;
90 static fo_close_t	vn_closefile;
91 
92 struct 	fileops vnops = {
93 	.fo_read = vn_io_fault,
94 	.fo_write = vn_io_fault,
95 	.fo_truncate = vn_truncate,
96 	.fo_ioctl = vn_ioctl,
97 	.fo_poll = vn_poll,
98 	.fo_kqfilter = vn_kqfilter,
99 	.fo_stat = vn_statfile,
100 	.fo_close = vn_closefile,
101 	.fo_chmod = vn_chmod,
102 	.fo_chown = vn_chown,
103 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
104 };
105 
106 int
107 vn_open(ndp, flagp, cmode, fp)
108 	struct nameidata *ndp;
109 	int *flagp, cmode;
110 	struct file *fp;
111 {
112 	struct thread *td = ndp->ni_cnd.cn_thread;
113 
114 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
115 }
116 
117 /*
118  * Common code for vnode open operations via a name lookup.
119  * Lookup the vnode and invoke VOP_CREATE if needed.
120  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
121  *
122  * Note that this does NOT free nameidata for the successful case,
123  * due to the NDINIT being done elsewhere.
124  */
125 int
126 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
127     struct ucred *cred, struct file *fp)
128 {
129 	struct vnode *vp;
130 	struct mount *mp;
131 	struct thread *td = ndp->ni_cnd.cn_thread;
132 	struct vattr vat;
133 	struct vattr *vap = &vat;
134 	int fmode, error;
135 
136 restart:
137 	fmode = *flagp;
138 	if (fmode & O_CREAT) {
139 		ndp->ni_cnd.cn_nameiop = CREATE;
140 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
141 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
142 			ndp->ni_cnd.cn_flags |= FOLLOW;
143 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
144 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
145 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
146 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
147 		bwillwrite();
148 		if ((error = namei(ndp)) != 0)
149 			return (error);
150 		if (ndp->ni_vp == NULL) {
151 			VATTR_NULL(vap);
152 			vap->va_type = VREG;
153 			vap->va_mode = cmode;
154 			if (fmode & O_EXCL)
155 				vap->va_vaflags |= VA_EXCLUSIVE;
156 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
157 				NDFREE(ndp, NDF_ONLY_PNBUF);
158 				vput(ndp->ni_dvp);
159 				if ((error = vn_start_write(NULL, &mp,
160 				    V_XSLEEP | PCATCH)) != 0)
161 					return (error);
162 				goto restart;
163 			}
164 #ifdef MAC
165 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
166 			    &ndp->ni_cnd, vap);
167 			if (error == 0)
168 #endif
169 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
170 						   &ndp->ni_cnd, vap);
171 			vput(ndp->ni_dvp);
172 			vn_finished_write(mp);
173 			if (error) {
174 				NDFREE(ndp, NDF_ONLY_PNBUF);
175 				return (error);
176 			}
177 			fmode &= ~O_TRUNC;
178 			vp = ndp->ni_vp;
179 		} else {
180 			if (ndp->ni_dvp == ndp->ni_vp)
181 				vrele(ndp->ni_dvp);
182 			else
183 				vput(ndp->ni_dvp);
184 			ndp->ni_dvp = NULL;
185 			vp = ndp->ni_vp;
186 			if (fmode & O_EXCL) {
187 				error = EEXIST;
188 				goto bad;
189 			}
190 			fmode &= ~O_CREAT;
191 		}
192 	} else {
193 		ndp->ni_cnd.cn_nameiop = LOOKUP;
194 		ndp->ni_cnd.cn_flags = ISOPEN |
195 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
196 		if (!(fmode & FWRITE))
197 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
198 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
199 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
200 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
201 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
202 		if ((error = namei(ndp)) != 0)
203 			return (error);
204 		vp = ndp->ni_vp;
205 	}
206 	error = vn_open_vnode(vp, fmode, cred, td, fp);
207 	if (error)
208 		goto bad;
209 	*flagp = fmode;
210 	return (0);
211 bad:
212 	NDFREE(ndp, NDF_ONLY_PNBUF);
213 	vput(vp);
214 	*flagp = fmode;
215 	ndp->ni_vp = NULL;
216 	return (error);
217 }
218 
219 /*
220  * Common code for vnode open operations once a vnode is located.
221  * Check permissions, and call the VOP_OPEN routine.
222  */
223 int
224 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
225     struct thread *td, struct file *fp)
226 {
227 	struct mount *mp;
228 	accmode_t accmode;
229 	struct flock lf;
230 	int error, have_flock, lock_flags, type;
231 
232 	if (vp->v_type == VLNK)
233 		return (EMLINK);
234 	if (vp->v_type == VSOCK)
235 		return (EOPNOTSUPP);
236 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
237 		return (ENOTDIR);
238 	accmode = 0;
239 	if (fmode & (FWRITE | O_TRUNC)) {
240 		if (vp->v_type == VDIR)
241 			return (EISDIR);
242 		accmode |= VWRITE;
243 	}
244 	if (fmode & FREAD)
245 		accmode |= VREAD;
246 	if (fmode & FEXEC)
247 		accmode |= VEXEC;
248 	if ((fmode & O_APPEND) && (fmode & FWRITE))
249 		accmode |= VAPPEND;
250 #ifdef MAC
251 	error = mac_vnode_check_open(cred, vp, accmode);
252 	if (error)
253 		return (error);
254 #endif
255 	if ((fmode & O_CREAT) == 0) {
256 		if (accmode & VWRITE) {
257 			error = vn_writechk(vp);
258 			if (error)
259 				return (error);
260 		}
261 		if (accmode) {
262 		        error = VOP_ACCESS(vp, accmode, cred, td);
263 			if (error)
264 				return (error);
265 		}
266 	}
267 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
268 		return (error);
269 
270 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
271 		KASSERT(fp != NULL, ("open with flock requires fp"));
272 		lock_flags = VOP_ISLOCKED(vp);
273 		VOP_UNLOCK(vp, 0);
274 		lf.l_whence = SEEK_SET;
275 		lf.l_start = 0;
276 		lf.l_len = 0;
277 		if (fmode & O_EXLOCK)
278 			lf.l_type = F_WRLCK;
279 		else
280 			lf.l_type = F_RDLCK;
281 		type = F_FLOCK;
282 		if ((fmode & FNONBLOCK) == 0)
283 			type |= F_WAIT;
284 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
285 		have_flock = (error == 0);
286 		vn_lock(vp, lock_flags | LK_RETRY);
287 		if (error == 0 && vp->v_iflag & VI_DOOMED)
288 			error = ENOENT;
289 		/*
290 		 * Another thread might have used this vnode as an
291 		 * executable while the vnode lock was dropped.
292 		 * Ensure the vnode is still able to be opened for
293 		 * writing after the lock has been obtained.
294 		 */
295 		if (error == 0 && accmode & VWRITE)
296 			error = vn_writechk(vp);
297 		if (error) {
298 			VOP_UNLOCK(vp, 0);
299 			if (have_flock) {
300 				lf.l_whence = SEEK_SET;
301 				lf.l_start = 0;
302 				lf.l_len = 0;
303 				lf.l_type = F_UNLCK;
304 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
305 				    F_FLOCK);
306 			}
307 			vn_start_write(vp, &mp, V_WAIT);
308 			vn_lock(vp, lock_flags | LK_RETRY);
309 			(void)VOP_CLOSE(vp, fmode, cred, td);
310 			vn_finished_write(mp);
311 			return (error);
312 		}
313 		fp->f_flag |= FHASLOCK;
314 	}
315 	if (fmode & FWRITE) {
316 		VOP_ADD_WRITECOUNT(vp, 1);
317 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
318 		    __func__, vp, vp->v_writecount);
319 	}
320 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
321 	return (0);
322 }
323 
324 /*
325  * Check for write permissions on the specified vnode.
326  * Prototype text segments cannot be written.
327  */
328 int
329 vn_writechk(vp)
330 	register struct vnode *vp;
331 {
332 
333 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
334 	/*
335 	 * If there's shared text associated with
336 	 * the vnode, try to free it up once.  If
337 	 * we fail, we can't allow writing.
338 	 */
339 	if (VOP_IS_TEXT(vp))
340 		return (ETXTBSY);
341 
342 	return (0);
343 }
344 
345 /*
346  * Vnode close call
347  */
348 int
349 vn_close(vp, flags, file_cred, td)
350 	register struct vnode *vp;
351 	int flags;
352 	struct ucred *file_cred;
353 	struct thread *td;
354 {
355 	struct mount *mp;
356 	int error, lock_flags;
357 
358 	if (!(flags & FWRITE) && vp->v_mount != NULL &&
359 	    vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
360 		lock_flags = LK_SHARED;
361 	else
362 		lock_flags = LK_EXCLUSIVE;
363 
364 	vn_start_write(vp, &mp, V_WAIT);
365 	vn_lock(vp, lock_flags | LK_RETRY);
366 	if (flags & FWRITE) {
367 		VNASSERT(vp->v_writecount > 0, vp,
368 		    ("vn_close: negative writecount"));
369 		VOP_ADD_WRITECOUNT(vp, -1);
370 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
371 		    __func__, vp, vp->v_writecount);
372 	}
373 	error = VOP_CLOSE(vp, flags, file_cred, td);
374 	vput(vp);
375 	vn_finished_write(mp);
376 	return (error);
377 }
378 
379 /*
380  * Heuristic to detect sequential operation.
381  */
382 static int
383 sequential_heuristic(struct uio *uio, struct file *fp)
384 {
385 
386 	if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
387 		return (fp->f_seqcount << IO_SEQSHIFT);
388 
389 	/*
390 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
391 	 * that the first I/O is normally considered to be slightly
392 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
393 	 * unless previous seeks have reduced f_seqcount to 0, in which
394 	 * case offset 0 is not special.
395 	 */
396 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
397 	    uio->uio_offset == fp->f_nextoff) {
398 		/*
399 		 * f_seqcount is in units of fixed-size blocks so that it
400 		 * depends mainly on the amount of sequential I/O and not
401 		 * much on the number of sequential I/O's.  The fixed size
402 		 * of 16384 is hard-coded here since it is (not quite) just
403 		 * a magic size that works well here.  This size is more
404 		 * closely related to the best I/O size for real disks than
405 		 * to any block size used by software.
406 		 */
407 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
408 		if (fp->f_seqcount > IO_SEQMAX)
409 			fp->f_seqcount = IO_SEQMAX;
410 		return (fp->f_seqcount << IO_SEQSHIFT);
411 	}
412 
413 	/* Not sequential.  Quickly draw-down sequentiality. */
414 	if (fp->f_seqcount > 1)
415 		fp->f_seqcount = 1;
416 	else
417 		fp->f_seqcount = 0;
418 	return (0);
419 }
420 
421 /*
422  * Package up an I/O request on a vnode into a uio and do it.
423  */
424 int
425 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
426     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
427     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
428 {
429 	struct uio auio;
430 	struct iovec aiov;
431 	struct mount *mp;
432 	struct ucred *cred;
433 	void *rl_cookie;
434 	int error, lock_flags;
435 
436 	auio.uio_iov = &aiov;
437 	auio.uio_iovcnt = 1;
438 	aiov.iov_base = base;
439 	aiov.iov_len = len;
440 	auio.uio_resid = len;
441 	auio.uio_offset = offset;
442 	auio.uio_segflg = segflg;
443 	auio.uio_rw = rw;
444 	auio.uio_td = td;
445 	error = 0;
446 
447 	if ((ioflg & IO_NODELOCKED) == 0) {
448 		if (rw == UIO_READ) {
449 			rl_cookie = vn_rangelock_rlock(vp, offset,
450 			    offset + len);
451 		} else {
452 			rl_cookie = vn_rangelock_wlock(vp, offset,
453 			    offset + len);
454 		}
455 		mp = NULL;
456 		if (rw == UIO_WRITE) {
457 			if (vp->v_type != VCHR &&
458 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
459 			    != 0)
460 				goto out;
461 			if (MNT_SHARED_WRITES(mp) ||
462 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
463 				lock_flags = LK_SHARED;
464 			else
465 				lock_flags = LK_EXCLUSIVE;
466 		} else
467 			lock_flags = LK_SHARED;
468 		vn_lock(vp, lock_flags | LK_RETRY);
469 	} else
470 		rl_cookie = NULL;
471 
472 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
473 #ifdef MAC
474 	if ((ioflg & IO_NOMACCHECK) == 0) {
475 		if (rw == UIO_READ)
476 			error = mac_vnode_check_read(active_cred, file_cred,
477 			    vp);
478 		else
479 			error = mac_vnode_check_write(active_cred, file_cred,
480 			    vp);
481 	}
482 #endif
483 	if (error == 0) {
484 		if (file_cred != NULL)
485 			cred = file_cred;
486 		else
487 			cred = active_cred;
488 		if (rw == UIO_READ)
489 			error = VOP_READ(vp, &auio, ioflg, cred);
490 		else
491 			error = VOP_WRITE(vp, &auio, ioflg, cred);
492 	}
493 	if (aresid)
494 		*aresid = auio.uio_resid;
495 	else
496 		if (auio.uio_resid && error == 0)
497 			error = EIO;
498 	if ((ioflg & IO_NODELOCKED) == 0) {
499 		VOP_UNLOCK(vp, 0);
500 		if (mp != NULL)
501 			vn_finished_write(mp);
502 	}
503  out:
504 	if (rl_cookie != NULL)
505 		vn_rangelock_unlock(vp, rl_cookie);
506 	return (error);
507 }
508 
509 /*
510  * Package up an I/O request on a vnode into a uio and do it.  The I/O
511  * request is split up into smaller chunks and we try to avoid saturating
512  * the buffer cache while potentially holding a vnode locked, so we
513  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
514  * to give other processes a chance to lock the vnode (either other processes
515  * core'ing the same binary, or unrelated processes scanning the directory).
516  */
517 int
518 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
519     file_cred, aresid, td)
520 	enum uio_rw rw;
521 	struct vnode *vp;
522 	void *base;
523 	size_t len;
524 	off_t offset;
525 	enum uio_seg segflg;
526 	int ioflg;
527 	struct ucred *active_cred;
528 	struct ucred *file_cred;
529 	size_t *aresid;
530 	struct thread *td;
531 {
532 	int error = 0;
533 	ssize_t iaresid;
534 
535 	do {
536 		int chunk;
537 
538 		/*
539 		 * Force `offset' to a multiple of MAXBSIZE except possibly
540 		 * for the first chunk, so that filesystems only need to
541 		 * write full blocks except possibly for the first and last
542 		 * chunks.
543 		 */
544 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
545 
546 		if (chunk > len)
547 			chunk = len;
548 		if (rw != UIO_READ && vp->v_type == VREG)
549 			bwillwrite();
550 		iaresid = 0;
551 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
552 		    ioflg, active_cred, file_cred, &iaresid, td);
553 		len -= chunk;	/* aresid calc already includes length */
554 		if (error)
555 			break;
556 		offset += chunk;
557 		base = (char *)base + chunk;
558 		kern_yield(PRI_USER);
559 	} while (len);
560 	if (aresid)
561 		*aresid = len + iaresid;
562 	return (error);
563 }
564 
565 off_t
566 foffset_lock(struct file *fp, int flags)
567 {
568 	struct mtx *mtxp;
569 	off_t res;
570 
571 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
572 
573 #if OFF_MAX <= LONG_MAX
574 	/*
575 	 * Caller only wants the current f_offset value.  Assume that
576 	 * the long and shorter integer types reads are atomic.
577 	 */
578 	if ((flags & FOF_NOLOCK) != 0)
579 		return (fp->f_offset);
580 #endif
581 
582 	/*
583 	 * According to McKusick the vn lock was protecting f_offset here.
584 	 * It is now protected by the FOFFSET_LOCKED flag.
585 	 */
586 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
587 	mtx_lock(mtxp);
588 	if ((flags & FOF_NOLOCK) == 0) {
589 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
590 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
591 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
592 			    "vofflock", 0);
593 		}
594 		fp->f_vnread_flags |= FOFFSET_LOCKED;
595 	}
596 	res = fp->f_offset;
597 	mtx_unlock(mtxp);
598 	return (res);
599 }
600 
601 void
602 foffset_unlock(struct file *fp, off_t val, int flags)
603 {
604 	struct mtx *mtxp;
605 
606 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
607 
608 #if OFF_MAX <= LONG_MAX
609 	if ((flags & FOF_NOLOCK) != 0) {
610 		if ((flags & FOF_NOUPDATE) == 0)
611 			fp->f_offset = val;
612 		if ((flags & FOF_NEXTOFF) != 0)
613 			fp->f_nextoff = val;
614 		return;
615 	}
616 #endif
617 
618 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
619 	mtx_lock(mtxp);
620 	if ((flags & FOF_NOUPDATE) == 0)
621 		fp->f_offset = val;
622 	if ((flags & FOF_NEXTOFF) != 0)
623 		fp->f_nextoff = val;
624 	if ((flags & FOF_NOLOCK) == 0) {
625 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
626 		    ("Lost FOFFSET_LOCKED"));
627 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
628 			wakeup(&fp->f_vnread_flags);
629 		fp->f_vnread_flags = 0;
630 	}
631 	mtx_unlock(mtxp);
632 }
633 
634 void
635 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
636 {
637 
638 	if ((flags & FOF_OFFSET) == 0)
639 		uio->uio_offset = foffset_lock(fp, flags);
640 }
641 
642 void
643 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
644 {
645 
646 	if ((flags & FOF_OFFSET) == 0)
647 		foffset_unlock(fp, uio->uio_offset, flags);
648 }
649 
650 static int
651 get_advice(struct file *fp, struct uio *uio)
652 {
653 	struct mtx *mtxp;
654 	int ret;
655 
656 	ret = POSIX_FADV_NORMAL;
657 	if (fp->f_advice == NULL)
658 		return (ret);
659 
660 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
661 	mtx_lock(mtxp);
662 	if (uio->uio_offset >= fp->f_advice->fa_start &&
663 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
664 		ret = fp->f_advice->fa_advice;
665 	mtx_unlock(mtxp);
666 	return (ret);
667 }
668 
669 /*
670  * File table vnode read routine.
671  */
672 static int
673 vn_read(fp, uio, active_cred, flags, td)
674 	struct file *fp;
675 	struct uio *uio;
676 	struct ucred *active_cred;
677 	int flags;
678 	struct thread *td;
679 {
680 	struct vnode *vp;
681 	struct mtx *mtxp;
682 	int error, ioflag;
683 	int advice;
684 	off_t offset, start, end;
685 
686 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
687 	    uio->uio_td, td));
688 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
689 	vp = fp->f_vnode;
690 	ioflag = 0;
691 	if (fp->f_flag & FNONBLOCK)
692 		ioflag |= IO_NDELAY;
693 	if (fp->f_flag & O_DIRECT)
694 		ioflag |= IO_DIRECT;
695 	advice = get_advice(fp, uio);
696 	vn_lock(vp, LK_SHARED | LK_RETRY);
697 
698 	switch (advice) {
699 	case POSIX_FADV_NORMAL:
700 	case POSIX_FADV_SEQUENTIAL:
701 	case POSIX_FADV_NOREUSE:
702 		ioflag |= sequential_heuristic(uio, fp);
703 		break;
704 	case POSIX_FADV_RANDOM:
705 		/* Disable read-ahead for random I/O. */
706 		break;
707 	}
708 	offset = uio->uio_offset;
709 
710 #ifdef MAC
711 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
712 	if (error == 0)
713 #endif
714 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
715 	fp->f_nextoff = uio->uio_offset;
716 	VOP_UNLOCK(vp, 0);
717 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
718 	    offset != uio->uio_offset) {
719 		/*
720 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
721 		 * buffers for the backing file after a
722 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
723 		 * case of using POSIX_FADV_NOREUSE with sequential
724 		 * access, track the previous implicit DONTNEED
725 		 * request and grow this request to include the
726 		 * current read(2) in addition to the previous
727 		 * DONTNEED.  With purely sequential access this will
728 		 * cause the DONTNEED requests to continously grow to
729 		 * cover all of the previously read regions of the
730 		 * file.  This allows filesystem blocks that are
731 		 * accessed by multiple calls to read(2) to be flushed
732 		 * once the last read(2) finishes.
733 		 */
734 		start = offset;
735 		end = uio->uio_offset - 1;
736 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
737 		mtx_lock(mtxp);
738 		if (fp->f_advice != NULL &&
739 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
740 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
741 				start = fp->f_advice->fa_prevstart;
742 			else if (fp->f_advice->fa_prevstart != 0 &&
743 			    fp->f_advice->fa_prevstart == end + 1)
744 				end = fp->f_advice->fa_prevend;
745 			fp->f_advice->fa_prevstart = start;
746 			fp->f_advice->fa_prevend = end;
747 		}
748 		mtx_unlock(mtxp);
749 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
750 	}
751 	return (error);
752 }
753 
754 /*
755  * File table vnode write routine.
756  */
757 static int
758 vn_write(fp, uio, active_cred, flags, td)
759 	struct file *fp;
760 	struct uio *uio;
761 	struct ucred *active_cred;
762 	int flags;
763 	struct thread *td;
764 {
765 	struct vnode *vp;
766 	struct mount *mp;
767 	struct mtx *mtxp;
768 	int error, ioflag, lock_flags;
769 	int advice;
770 	off_t offset, start, end;
771 
772 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
773 	    uio->uio_td, td));
774 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
775 	vp = fp->f_vnode;
776 	if (vp->v_type == VREG)
777 		bwillwrite();
778 	ioflag = IO_UNIT;
779 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
780 		ioflag |= IO_APPEND;
781 	if (fp->f_flag & FNONBLOCK)
782 		ioflag |= IO_NDELAY;
783 	if (fp->f_flag & O_DIRECT)
784 		ioflag |= IO_DIRECT;
785 	if ((fp->f_flag & O_FSYNC) ||
786 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
787 		ioflag |= IO_SYNC;
788 	mp = NULL;
789 	if (vp->v_type != VCHR &&
790 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
791 		goto unlock;
792 
793 	advice = get_advice(fp, uio);
794 
795 	if (MNT_SHARED_WRITES(mp) ||
796 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
797 		lock_flags = LK_SHARED;
798 	} else {
799 		lock_flags = LK_EXCLUSIVE;
800 	}
801 
802 	vn_lock(vp, lock_flags | LK_RETRY);
803 	switch (advice) {
804 	case POSIX_FADV_NORMAL:
805 	case POSIX_FADV_SEQUENTIAL:
806 	case POSIX_FADV_NOREUSE:
807 		ioflag |= sequential_heuristic(uio, fp);
808 		break;
809 	case POSIX_FADV_RANDOM:
810 		/* XXX: Is this correct? */
811 		break;
812 	}
813 	offset = uio->uio_offset;
814 
815 #ifdef MAC
816 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
817 	if (error == 0)
818 #endif
819 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
820 	fp->f_nextoff = uio->uio_offset;
821 	VOP_UNLOCK(vp, 0);
822 	if (vp->v_type != VCHR)
823 		vn_finished_write(mp);
824 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
825 	    offset != uio->uio_offset) {
826 		/*
827 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
828 		 * buffers for the backing file after a
829 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
830 		 * common case of using POSIX_FADV_NOREUSE with
831 		 * sequential access, track the previous implicit
832 		 * DONTNEED request and grow this request to include
833 		 * the current write(2) in addition to the previous
834 		 * DONTNEED.  With purely sequential access this will
835 		 * cause the DONTNEED requests to continously grow to
836 		 * cover all of the previously written regions of the
837 		 * file.
838 		 *
839 		 * Note that the blocks just written are almost
840 		 * certainly still dirty, so this only works when
841 		 * VOP_ADVISE() calls from subsequent writes push out
842 		 * the data written by this write(2) once the backing
843 		 * buffers are clean.  However, as compared to forcing
844 		 * IO_DIRECT, this gives much saner behavior.  Write
845 		 * clustering is still allowed, and clean pages are
846 		 * merely moved to the cache page queue rather than
847 		 * outright thrown away.  This means a subsequent
848 		 * read(2) can still avoid hitting the disk if the
849 		 * pages have not been reclaimed.
850 		 *
851 		 * This does make POSIX_FADV_NOREUSE largely useless
852 		 * with non-sequential access.  However, sequential
853 		 * access is the more common use case and the flag is
854 		 * merely advisory.
855 		 */
856 		start = offset;
857 		end = uio->uio_offset - 1;
858 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
859 		mtx_lock(mtxp);
860 		if (fp->f_advice != NULL &&
861 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
862 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
863 				start = fp->f_advice->fa_prevstart;
864 			else if (fp->f_advice->fa_prevstart != 0 &&
865 			    fp->f_advice->fa_prevstart == end + 1)
866 				end = fp->f_advice->fa_prevend;
867 			fp->f_advice->fa_prevstart = start;
868 			fp->f_advice->fa_prevend = end;
869 		}
870 		mtx_unlock(mtxp);
871 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
872 	}
873 
874 unlock:
875 	return (error);
876 }
877 
878 static const int io_hold_cnt = 16;
879 static int vn_io_fault_enable = 1;
880 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
881     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
882 static u_long vn_io_faults_cnt;
883 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
884     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
885 
886 /*
887  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
888  * prevent the following deadlock:
889  *
890  * Assume that the thread A reads from the vnode vp1 into userspace
891  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
892  * currently not resident, then system ends up with the call chain
893  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
894  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
895  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
896  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
897  * backed by the pages of vnode vp1, and some page in buf2 is not
898  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
899  *
900  * To prevent the lock order reversal and deadlock, vn_io_fault() does
901  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
902  * Instead, it first tries to do the whole range i/o with pagefaults
903  * disabled. If all pages in the i/o buffer are resident and mapped,
904  * VOP will succeed (ignoring the genuine filesystem errors).
905  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
906  * i/o in chunks, with all pages in the chunk prefaulted and held
907  * using vm_fault_quick_hold_pages().
908  *
909  * Filesystems using this deadlock avoidance scheme should use the
910  * array of the held pages from uio, saved in the curthread->td_ma,
911  * instead of doing uiomove().  A helper function
912  * vn_io_fault_uiomove() converts uiomove request into
913  * uiomove_fromphys() over td_ma array.
914  *
915  * Since vnode locks do not cover the whole i/o anymore, rangelocks
916  * make the current i/o request atomic with respect to other i/os and
917  * truncations.
918  */
919 static int
920 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
921     int flags, struct thread *td)
922 {
923 	vm_page_t ma[io_hold_cnt + 2];
924 	struct uio *uio_clone, short_uio;
925 	struct iovec short_iovec[1];
926 	fo_rdwr_t *doio;
927 	struct vnode *vp;
928 	void *rl_cookie;
929 	struct mount *mp;
930 	vm_page_t *prev_td_ma;
931 	int cnt, error, save, saveheld, prev_td_ma_cnt;
932 	vm_offset_t addr, end;
933 	vm_prot_t prot;
934 	size_t len, resid;
935 	ssize_t adv;
936 
937 	if (uio->uio_rw == UIO_READ)
938 		doio = vn_read;
939 	else
940 		doio = vn_write;
941 	vp = fp->f_vnode;
942 	foffset_lock_uio(fp, uio, flags);
943 
944 	if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
945 	    ((mp = vp->v_mount) != NULL &&
946 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
947 	    !vn_io_fault_enable) {
948 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
949 		goto out_last;
950 	}
951 
952 	/*
953 	 * The UFS follows IO_UNIT directive and replays back both
954 	 * uio_offset and uio_resid if an error is encountered during the
955 	 * operation.  But, since the iovec may be already advanced,
956 	 * uio is still in an inconsistent state.
957 	 *
958 	 * Cache a copy of the original uio, which is advanced to the redo
959 	 * point using UIO_NOCOPY below.
960 	 */
961 	uio_clone = cloneuio(uio);
962 	resid = uio->uio_resid;
963 
964 	short_uio.uio_segflg = UIO_USERSPACE;
965 	short_uio.uio_rw = uio->uio_rw;
966 	short_uio.uio_td = uio->uio_td;
967 
968 	if (uio->uio_rw == UIO_READ) {
969 		prot = VM_PROT_WRITE;
970 		rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
971 		    uio->uio_offset + uio->uio_resid);
972 	} else {
973 		prot = VM_PROT_READ;
974 		if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
975 			/* For appenders, punt and lock the whole range. */
976 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
977 		else
978 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
979 			    uio->uio_offset + uio->uio_resid);
980 	}
981 
982 	save = vm_fault_disable_pagefaults();
983 	error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
984 	if (error != EFAULT)
985 		goto out;
986 
987 	atomic_add_long(&vn_io_faults_cnt, 1);
988 	uio_clone->uio_segflg = UIO_NOCOPY;
989 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
990 	uio_clone->uio_segflg = uio->uio_segflg;
991 
992 	saveheld = curthread_pflags_set(TDP_UIOHELD);
993 	prev_td_ma = td->td_ma;
994 	prev_td_ma_cnt = td->td_ma_cnt;
995 
996 	while (uio_clone->uio_resid != 0) {
997 		len = uio_clone->uio_iov->iov_len;
998 		if (len == 0) {
999 			KASSERT(uio_clone->uio_iovcnt >= 1,
1000 			    ("iovcnt underflow"));
1001 			uio_clone->uio_iov++;
1002 			uio_clone->uio_iovcnt--;
1003 			continue;
1004 		}
1005 
1006 		addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
1007 		end = round_page(addr + len);
1008 		cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
1009 		/*
1010 		 * A perfectly misaligned address and length could cause
1011 		 * both the start and the end of the chunk to use partial
1012 		 * page.  +2 accounts for such a situation.
1013 		 */
1014 		if (cnt > io_hold_cnt + 2) {
1015 			len = io_hold_cnt * PAGE_SIZE;
1016 			KASSERT(howmany(round_page(addr + len) -
1017 			    trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
1018 			    ("cnt overflow"));
1019 		}
1020 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1021 		    addr, len, prot, ma, io_hold_cnt + 2);
1022 		if (cnt == -1) {
1023 			error = EFAULT;
1024 			break;
1025 		}
1026 		short_uio.uio_iov = &short_iovec[0];
1027 		short_iovec[0].iov_base = (void *)addr;
1028 		short_uio.uio_iovcnt = 1;
1029 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1030 		short_uio.uio_offset = uio_clone->uio_offset;
1031 		td->td_ma = ma;
1032 		td->td_ma_cnt = cnt;
1033 
1034 		error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
1035 		    td);
1036 		vm_page_unhold_pages(ma, cnt);
1037 		adv = len - short_uio.uio_resid;
1038 
1039 		uio_clone->uio_iov->iov_base =
1040 		    (char *)uio_clone->uio_iov->iov_base + adv;
1041 		uio_clone->uio_iov->iov_len -= adv;
1042 		uio_clone->uio_resid -= adv;
1043 		uio_clone->uio_offset += adv;
1044 
1045 		uio->uio_resid -= adv;
1046 		uio->uio_offset += adv;
1047 
1048 		if (error != 0 || adv == 0)
1049 			break;
1050 	}
1051 	td->td_ma = prev_td_ma;
1052 	td->td_ma_cnt = prev_td_ma_cnt;
1053 	curthread_pflags_restore(saveheld);
1054 out:
1055 	vm_fault_enable_pagefaults(save);
1056 	vn_rangelock_unlock(vp, rl_cookie);
1057 	free(uio_clone, M_IOV);
1058 out_last:
1059 	foffset_unlock_uio(fp, uio, flags);
1060 	return (error);
1061 }
1062 
1063 /*
1064  * Helper function to perform the requested uiomove operation using
1065  * the held pages for io->uio_iov[0].iov_base buffer instead of
1066  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1067  * instead of iov_base prevents page faults that could occur due to
1068  * pmap_collect() invalidating the mapping created by
1069  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1070  * object cleanup revoking the write access from page mappings.
1071  *
1072  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1073  * instead of plain uiomove().
1074  */
1075 int
1076 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1077 {
1078 	struct uio transp_uio;
1079 	struct iovec transp_iov[1];
1080 	struct thread *td;
1081 	size_t adv;
1082 	int error, pgadv;
1083 
1084 	td = curthread;
1085 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1086 	    uio->uio_segflg != UIO_USERSPACE)
1087 		return (uiomove(data, xfersize, uio));
1088 
1089 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1090 	transp_iov[0].iov_base = data;
1091 	transp_uio.uio_iov = &transp_iov[0];
1092 	transp_uio.uio_iovcnt = 1;
1093 	if (xfersize > uio->uio_resid)
1094 		xfersize = uio->uio_resid;
1095 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1096 	transp_uio.uio_offset = 0;
1097 	transp_uio.uio_segflg = UIO_SYSSPACE;
1098 	/*
1099 	 * Since transp_iov points to data, and td_ma page array
1100 	 * corresponds to original uio->uio_iov, we need to invert the
1101 	 * direction of the i/o operation as passed to
1102 	 * uiomove_fromphys().
1103 	 */
1104 	switch (uio->uio_rw) {
1105 	case UIO_WRITE:
1106 		transp_uio.uio_rw = UIO_READ;
1107 		break;
1108 	case UIO_READ:
1109 		transp_uio.uio_rw = UIO_WRITE;
1110 		break;
1111 	}
1112 	transp_uio.uio_td = uio->uio_td;
1113 	error = uiomove_fromphys(td->td_ma,
1114 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1115 	    xfersize, &transp_uio);
1116 	adv = xfersize - transp_uio.uio_resid;
1117 	pgadv =
1118 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1119 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1120 	td->td_ma += pgadv;
1121 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1122 	    pgadv));
1123 	td->td_ma_cnt -= pgadv;
1124 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1125 	uio->uio_iov->iov_len -= adv;
1126 	uio->uio_resid -= adv;
1127 	uio->uio_offset += adv;
1128 	return (error);
1129 }
1130 
1131 int
1132 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1133     struct uio *uio)
1134 {
1135 	struct thread *td;
1136 	vm_offset_t iov_base;
1137 	int cnt, pgadv;
1138 
1139 	td = curthread;
1140 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1141 	    uio->uio_segflg != UIO_USERSPACE)
1142 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1143 
1144 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1145 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1146 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1147 	switch (uio->uio_rw) {
1148 	case UIO_WRITE:
1149 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1150 		    offset, cnt);
1151 		break;
1152 	case UIO_READ:
1153 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1154 		    cnt);
1155 		break;
1156 	}
1157 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1158 	td->td_ma += pgadv;
1159 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1160 	    pgadv));
1161 	td->td_ma_cnt -= pgadv;
1162 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1163 	uio->uio_iov->iov_len -= cnt;
1164 	uio->uio_resid -= cnt;
1165 	uio->uio_offset += cnt;
1166 	return (0);
1167 }
1168 
1169 
1170 /*
1171  * File table truncate routine.
1172  */
1173 static int
1174 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1175     struct thread *td)
1176 {
1177 	struct vattr vattr;
1178 	struct mount *mp;
1179 	struct vnode *vp;
1180 	void *rl_cookie;
1181 	int error;
1182 
1183 	vp = fp->f_vnode;
1184 
1185 	/*
1186 	 * Lock the whole range for truncation.  Otherwise split i/o
1187 	 * might happen partly before and partly after the truncation.
1188 	 */
1189 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1190 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1191 	if (error)
1192 		goto out1;
1193 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1194 	if (vp->v_type == VDIR) {
1195 		error = EISDIR;
1196 		goto out;
1197 	}
1198 #ifdef MAC
1199 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1200 	if (error)
1201 		goto out;
1202 #endif
1203 	error = vn_writechk(vp);
1204 	if (error == 0) {
1205 		VATTR_NULL(&vattr);
1206 		vattr.va_size = length;
1207 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1208 	}
1209 out:
1210 	VOP_UNLOCK(vp, 0);
1211 	vn_finished_write(mp);
1212 out1:
1213 	vn_rangelock_unlock(vp, rl_cookie);
1214 	return (error);
1215 }
1216 
1217 /*
1218  * File table vnode stat routine.
1219  */
1220 static int
1221 vn_statfile(fp, sb, active_cred, td)
1222 	struct file *fp;
1223 	struct stat *sb;
1224 	struct ucred *active_cred;
1225 	struct thread *td;
1226 {
1227 	struct vnode *vp = fp->f_vnode;
1228 	int error;
1229 
1230 	vn_lock(vp, LK_SHARED | LK_RETRY);
1231 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1232 	VOP_UNLOCK(vp, 0);
1233 
1234 	return (error);
1235 }
1236 
1237 /*
1238  * Stat a vnode; implementation for the stat syscall
1239  */
1240 int
1241 vn_stat(vp, sb, active_cred, file_cred, td)
1242 	struct vnode *vp;
1243 	register struct stat *sb;
1244 	struct ucred *active_cred;
1245 	struct ucred *file_cred;
1246 	struct thread *td;
1247 {
1248 	struct vattr vattr;
1249 	register struct vattr *vap;
1250 	int error;
1251 	u_short mode;
1252 
1253 #ifdef MAC
1254 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1255 	if (error)
1256 		return (error);
1257 #endif
1258 
1259 	vap = &vattr;
1260 
1261 	/*
1262 	 * Initialize defaults for new and unusual fields, so that file
1263 	 * systems which don't support these fields don't need to know
1264 	 * about them.
1265 	 */
1266 	vap->va_birthtime.tv_sec = -1;
1267 	vap->va_birthtime.tv_nsec = 0;
1268 	vap->va_fsid = VNOVAL;
1269 	vap->va_rdev = NODEV;
1270 
1271 	error = VOP_GETATTR(vp, vap, active_cred);
1272 	if (error)
1273 		return (error);
1274 
1275 	/*
1276 	 * Zero the spare stat fields
1277 	 */
1278 	bzero(sb, sizeof *sb);
1279 
1280 	/*
1281 	 * Copy from vattr table
1282 	 */
1283 	if (vap->va_fsid != VNOVAL)
1284 		sb->st_dev = vap->va_fsid;
1285 	else
1286 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1287 	sb->st_ino = vap->va_fileid;
1288 	mode = vap->va_mode;
1289 	switch (vap->va_type) {
1290 	case VREG:
1291 		mode |= S_IFREG;
1292 		break;
1293 	case VDIR:
1294 		mode |= S_IFDIR;
1295 		break;
1296 	case VBLK:
1297 		mode |= S_IFBLK;
1298 		break;
1299 	case VCHR:
1300 		mode |= S_IFCHR;
1301 		break;
1302 	case VLNK:
1303 		mode |= S_IFLNK;
1304 		break;
1305 	case VSOCK:
1306 		mode |= S_IFSOCK;
1307 		break;
1308 	case VFIFO:
1309 		mode |= S_IFIFO;
1310 		break;
1311 	default:
1312 		return (EBADF);
1313 	};
1314 	sb->st_mode = mode;
1315 	sb->st_nlink = vap->va_nlink;
1316 	sb->st_uid = vap->va_uid;
1317 	sb->st_gid = vap->va_gid;
1318 	sb->st_rdev = vap->va_rdev;
1319 	if (vap->va_size > OFF_MAX)
1320 		return (EOVERFLOW);
1321 	sb->st_size = vap->va_size;
1322 	sb->st_atim = vap->va_atime;
1323 	sb->st_mtim = vap->va_mtime;
1324 	sb->st_ctim = vap->va_ctime;
1325 	sb->st_birthtim = vap->va_birthtime;
1326 
1327         /*
1328 	 * According to www.opengroup.org, the meaning of st_blksize is
1329 	 *   "a filesystem-specific preferred I/O block size for this
1330 	 *    object.  In some filesystem types, this may vary from file
1331 	 *    to file"
1332 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1333 	 */
1334 
1335 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1336 
1337 	sb->st_flags = vap->va_flags;
1338 	if (priv_check(td, PRIV_VFS_GENERATION))
1339 		sb->st_gen = 0;
1340 	else
1341 		sb->st_gen = vap->va_gen;
1342 
1343 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1344 	return (0);
1345 }
1346 
1347 /*
1348  * File table vnode ioctl routine.
1349  */
1350 static int
1351 vn_ioctl(fp, com, data, active_cred, td)
1352 	struct file *fp;
1353 	u_long com;
1354 	void *data;
1355 	struct ucred *active_cred;
1356 	struct thread *td;
1357 {
1358 	struct vattr vattr;
1359 	struct vnode *vp;
1360 	int error;
1361 
1362 	vp = fp->f_vnode;
1363 	switch (vp->v_type) {
1364 	case VDIR:
1365 	case VREG:
1366 		switch (com) {
1367 		case FIONREAD:
1368 			vn_lock(vp, LK_SHARED | LK_RETRY);
1369 			error = VOP_GETATTR(vp, &vattr, active_cred);
1370 			VOP_UNLOCK(vp, 0);
1371 			if (error == 0)
1372 				*(int *)data = vattr.va_size - fp->f_offset;
1373 			return (error);
1374 		case FIONBIO:
1375 		case FIOASYNC:
1376 			return (0);
1377 		default:
1378 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1379 			    active_cred, td));
1380 		}
1381 	default:
1382 		return (ENOTTY);
1383 	}
1384 }
1385 
1386 /*
1387  * File table vnode poll routine.
1388  */
1389 static int
1390 vn_poll(fp, events, active_cred, td)
1391 	struct file *fp;
1392 	int events;
1393 	struct ucred *active_cred;
1394 	struct thread *td;
1395 {
1396 	struct vnode *vp;
1397 	int error;
1398 
1399 	vp = fp->f_vnode;
1400 #ifdef MAC
1401 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1402 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1403 	VOP_UNLOCK(vp, 0);
1404 	if (!error)
1405 #endif
1406 
1407 	error = VOP_POLL(vp, events, fp->f_cred, td);
1408 	return (error);
1409 }
1410 
1411 /*
1412  * Acquire the requested lock and then check for validity.  LK_RETRY
1413  * permits vn_lock to return doomed vnodes.
1414  */
1415 int
1416 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1417 {
1418 	int error;
1419 
1420 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1421 	    ("vn_lock called with no locktype."));
1422 	do {
1423 #ifdef DEBUG_VFS_LOCKS
1424 		KASSERT(vp->v_holdcnt != 0,
1425 		    ("vn_lock %p: zero hold count", vp));
1426 #endif
1427 		error = VOP_LOCK1(vp, flags, file, line);
1428 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1429 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1430 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1431 		    flags, error));
1432 		/*
1433 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1434 		 * If RETRY is not set, we return ENOENT instead.
1435 		 */
1436 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1437 		    (flags & LK_RETRY) == 0) {
1438 			VOP_UNLOCK(vp, 0);
1439 			error = ENOENT;
1440 			break;
1441 		}
1442 	} while (flags & LK_RETRY && error != 0);
1443 	return (error);
1444 }
1445 
1446 /*
1447  * File table vnode close routine.
1448  */
1449 static int
1450 vn_closefile(fp, td)
1451 	struct file *fp;
1452 	struct thread *td;
1453 {
1454 	struct vnode *vp;
1455 	struct flock lf;
1456 	int error;
1457 
1458 	vp = fp->f_vnode;
1459 	fp->f_ops = &badfileops;
1460 
1461 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1462 		vref(vp);
1463 
1464 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1465 
1466 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1467 		lf.l_whence = SEEK_SET;
1468 		lf.l_start = 0;
1469 		lf.l_len = 0;
1470 		lf.l_type = F_UNLCK;
1471 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1472 		vrele(vp);
1473 	}
1474 	return (error);
1475 }
1476 
1477 /*
1478  * Preparing to start a filesystem write operation. If the operation is
1479  * permitted, then we bump the count of operations in progress and
1480  * proceed. If a suspend request is in progress, we wait until the
1481  * suspension is over, and then proceed.
1482  */
1483 static int
1484 vn_start_write_locked(struct mount *mp, int flags)
1485 {
1486 	int error;
1487 
1488 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1489 	error = 0;
1490 
1491 	/*
1492 	 * Check on status of suspension.
1493 	 */
1494 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1495 	    mp->mnt_susp_owner != curthread) {
1496 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1497 			if (flags & V_NOWAIT) {
1498 				error = EWOULDBLOCK;
1499 				goto unlock;
1500 			}
1501 			error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1502 			    (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1503 			if (error)
1504 				goto unlock;
1505 		}
1506 	}
1507 	if (flags & V_XSLEEP)
1508 		goto unlock;
1509 	mp->mnt_writeopcount++;
1510 unlock:
1511 	if (error != 0 || (flags & V_XSLEEP) != 0)
1512 		MNT_REL(mp);
1513 	MNT_IUNLOCK(mp);
1514 	return (error);
1515 }
1516 
1517 int
1518 vn_start_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 	error = 0;
1527 	/*
1528 	 * If a vnode is provided, get and return the mount point that
1529 	 * to which it will write.
1530 	 */
1531 	if (vp != NULL) {
1532 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1533 			*mpp = NULL;
1534 			if (error != EOPNOTSUPP)
1535 				return (error);
1536 			return (0);
1537 		}
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 
1553 	return (vn_start_write_locked(mp, flags));
1554 }
1555 
1556 /*
1557  * Secondary suspension. Used by operations such as vop_inactive
1558  * routines that are needed by the higher level functions. These
1559  * are allowed to proceed until all the higher level functions have
1560  * completed (indicated by mnt_writeopcount dropping to zero). At that
1561  * time, these operations are halted until the suspension is over.
1562  */
1563 int
1564 vn_start_secondary_write(vp, mpp, flags)
1565 	struct vnode *vp;
1566 	struct mount **mpp;
1567 	int flags;
1568 {
1569 	struct mount *mp;
1570 	int error;
1571 
1572  retry:
1573 	if (vp != NULL) {
1574 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1575 			*mpp = NULL;
1576 			if (error != EOPNOTSUPP)
1577 				return (error);
1578 			return (0);
1579 		}
1580 	}
1581 	/*
1582 	 * If we are not suspended or have not yet reached suspended
1583 	 * mode, then let the operation proceed.
1584 	 */
1585 	if ((mp = *mpp) == NULL)
1586 		return (0);
1587 
1588 	/*
1589 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1590 	 * a vfs_ref().
1591 	 * As long as a vnode is not provided we need to acquire a
1592 	 * refcount for the provided mountpoint too, in order to
1593 	 * emulate a vfs_ref().
1594 	 */
1595 	MNT_ILOCK(mp);
1596 	if (vp == NULL)
1597 		MNT_REF(mp);
1598 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1599 		mp->mnt_secondary_writes++;
1600 		mp->mnt_secondary_accwrites++;
1601 		MNT_IUNLOCK(mp);
1602 		return (0);
1603 	}
1604 	if (flags & V_NOWAIT) {
1605 		MNT_REL(mp);
1606 		MNT_IUNLOCK(mp);
1607 		return (EWOULDBLOCK);
1608 	}
1609 	/*
1610 	 * Wait for the suspension to finish.
1611 	 */
1612 	error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1613 		       (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1614 	vfs_rel(mp);
1615 	if (error == 0)
1616 		goto retry;
1617 	return (error);
1618 }
1619 
1620 /*
1621  * Filesystem write operation has completed. If we are suspending and this
1622  * operation is the last one, notify the suspender that the suspension is
1623  * now in effect.
1624  */
1625 void
1626 vn_finished_write(mp)
1627 	struct mount *mp;
1628 {
1629 	if (mp == NULL)
1630 		return;
1631 	MNT_ILOCK(mp);
1632 	MNT_REL(mp);
1633 	mp->mnt_writeopcount--;
1634 	if (mp->mnt_writeopcount < 0)
1635 		panic("vn_finished_write: neg cnt");
1636 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1637 	    mp->mnt_writeopcount <= 0)
1638 		wakeup(&mp->mnt_writeopcount);
1639 	MNT_IUNLOCK(mp);
1640 }
1641 
1642 
1643 /*
1644  * Filesystem secondary write operation has completed. If we are
1645  * suspending and this operation is the last one, notify the suspender
1646  * that the suspension is now in effect.
1647  */
1648 void
1649 vn_finished_secondary_write(mp)
1650 	struct mount *mp;
1651 {
1652 	if (mp == NULL)
1653 		return;
1654 	MNT_ILOCK(mp);
1655 	MNT_REL(mp);
1656 	mp->mnt_secondary_writes--;
1657 	if (mp->mnt_secondary_writes < 0)
1658 		panic("vn_finished_secondary_write: neg cnt");
1659 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1660 	    mp->mnt_secondary_writes <= 0)
1661 		wakeup(&mp->mnt_secondary_writes);
1662 	MNT_IUNLOCK(mp);
1663 }
1664 
1665 
1666 
1667 /*
1668  * Request a filesystem to suspend write operations.
1669  */
1670 int
1671 vfs_write_suspend(mp)
1672 	struct mount *mp;
1673 {
1674 	int error;
1675 
1676 	MNT_ILOCK(mp);
1677 	if (mp->mnt_susp_owner == curthread) {
1678 		MNT_IUNLOCK(mp);
1679 		return (EALREADY);
1680 	}
1681 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1682 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1683 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1684 	mp->mnt_susp_owner = curthread;
1685 	if (mp->mnt_writeopcount > 0)
1686 		(void) msleep(&mp->mnt_writeopcount,
1687 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1688 	else
1689 		MNT_IUNLOCK(mp);
1690 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1691 		vfs_write_resume(mp, 0);
1692 	return (error);
1693 }
1694 
1695 /*
1696  * Request a filesystem to resume write operations.
1697  */
1698 void
1699 vfs_write_resume(struct mount *mp, int flags)
1700 {
1701 
1702 	MNT_ILOCK(mp);
1703 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1704 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1705 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1706 				       MNTK_SUSPENDED);
1707 		mp->mnt_susp_owner = NULL;
1708 		wakeup(&mp->mnt_writeopcount);
1709 		wakeup(&mp->mnt_flag);
1710 		curthread->td_pflags &= ~TDP_IGNSUSP;
1711 		if ((flags & VR_START_WRITE) != 0) {
1712 			MNT_REF(mp);
1713 			mp->mnt_writeopcount++;
1714 		}
1715 		MNT_IUNLOCK(mp);
1716 		if ((flags & VR_NO_SUSPCLR) == 0)
1717 			VFS_SUSP_CLEAN(mp);
1718 	} else if ((flags & VR_START_WRITE) != 0) {
1719 		MNT_REF(mp);
1720 		vn_start_write_locked(mp, 0);
1721 	} else {
1722 		MNT_IUNLOCK(mp);
1723 	}
1724 }
1725 
1726 /*
1727  * Implement kqueues for files by translating it to vnode operation.
1728  */
1729 static int
1730 vn_kqfilter(struct file *fp, struct knote *kn)
1731 {
1732 
1733 	return (VOP_KQFILTER(fp->f_vnode, kn));
1734 }
1735 
1736 /*
1737  * Simplified in-kernel wrapper calls for extended attribute access.
1738  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1739  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1740  */
1741 int
1742 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1743     const char *attrname, int *buflen, char *buf, struct thread *td)
1744 {
1745 	struct uio	auio;
1746 	struct iovec	iov;
1747 	int	error;
1748 
1749 	iov.iov_len = *buflen;
1750 	iov.iov_base = buf;
1751 
1752 	auio.uio_iov = &iov;
1753 	auio.uio_iovcnt = 1;
1754 	auio.uio_rw = UIO_READ;
1755 	auio.uio_segflg = UIO_SYSSPACE;
1756 	auio.uio_td = td;
1757 	auio.uio_offset = 0;
1758 	auio.uio_resid = *buflen;
1759 
1760 	if ((ioflg & IO_NODELOCKED) == 0)
1761 		vn_lock(vp, LK_SHARED | LK_RETRY);
1762 
1763 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1764 
1765 	/* authorize attribute retrieval as kernel */
1766 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1767 	    td);
1768 
1769 	if ((ioflg & IO_NODELOCKED) == 0)
1770 		VOP_UNLOCK(vp, 0);
1771 
1772 	if (error == 0) {
1773 		*buflen = *buflen - auio.uio_resid;
1774 	}
1775 
1776 	return (error);
1777 }
1778 
1779 /*
1780  * XXX failure mode if partially written?
1781  */
1782 int
1783 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1784     const char *attrname, int buflen, char *buf, struct thread *td)
1785 {
1786 	struct uio	auio;
1787 	struct iovec	iov;
1788 	struct mount	*mp;
1789 	int	error;
1790 
1791 	iov.iov_len = buflen;
1792 	iov.iov_base = buf;
1793 
1794 	auio.uio_iov = &iov;
1795 	auio.uio_iovcnt = 1;
1796 	auio.uio_rw = UIO_WRITE;
1797 	auio.uio_segflg = UIO_SYSSPACE;
1798 	auio.uio_td = td;
1799 	auio.uio_offset = 0;
1800 	auio.uio_resid = buflen;
1801 
1802 	if ((ioflg & IO_NODELOCKED) == 0) {
1803 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1804 			return (error);
1805 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1806 	}
1807 
1808 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1809 
1810 	/* authorize attribute setting as kernel */
1811 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1812 
1813 	if ((ioflg & IO_NODELOCKED) == 0) {
1814 		vn_finished_write(mp);
1815 		VOP_UNLOCK(vp, 0);
1816 	}
1817 
1818 	return (error);
1819 }
1820 
1821 int
1822 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1823     const char *attrname, struct thread *td)
1824 {
1825 	struct mount	*mp;
1826 	int	error;
1827 
1828 	if ((ioflg & IO_NODELOCKED) == 0) {
1829 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1830 			return (error);
1831 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1832 	}
1833 
1834 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1835 
1836 	/* authorize attribute removal as kernel */
1837 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1838 	if (error == EOPNOTSUPP)
1839 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1840 		    NULL, td);
1841 
1842 	if ((ioflg & IO_NODELOCKED) == 0) {
1843 		vn_finished_write(mp);
1844 		VOP_UNLOCK(vp, 0);
1845 	}
1846 
1847 	return (error);
1848 }
1849 
1850 int
1851 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1852 {
1853 	struct mount *mp;
1854 	int ltype, error;
1855 
1856 	mp = vp->v_mount;
1857 	ltype = VOP_ISLOCKED(vp);
1858 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1859 	    ("vn_vget_ino: vp not locked"));
1860 	error = vfs_busy(mp, MBF_NOWAIT);
1861 	if (error != 0) {
1862 		vfs_ref(mp);
1863 		VOP_UNLOCK(vp, 0);
1864 		error = vfs_busy(mp, 0);
1865 		vn_lock(vp, ltype | LK_RETRY);
1866 		vfs_rel(mp);
1867 		if (error != 0)
1868 			return (ENOENT);
1869 		if (vp->v_iflag & VI_DOOMED) {
1870 			vfs_unbusy(mp);
1871 			return (ENOENT);
1872 		}
1873 	}
1874 	VOP_UNLOCK(vp, 0);
1875 	error = VFS_VGET(mp, ino, lkflags, rvp);
1876 	vfs_unbusy(mp);
1877 	vn_lock(vp, ltype | LK_RETRY);
1878 	if (vp->v_iflag & VI_DOOMED) {
1879 		if (error == 0)
1880 			vput(*rvp);
1881 		error = ENOENT;
1882 	}
1883 	return (error);
1884 }
1885 
1886 int
1887 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1888     const struct thread *td)
1889 {
1890 
1891 	if (vp->v_type != VREG || td == NULL)
1892 		return (0);
1893 	PROC_LOCK(td->td_proc);
1894 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1895 	    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1896 		kern_psignal(td->td_proc, SIGXFSZ);
1897 		PROC_UNLOCK(td->td_proc);
1898 		return (EFBIG);
1899 	}
1900 	PROC_UNLOCK(td->td_proc);
1901 	return (0);
1902 }
1903 
1904 int
1905 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1906     struct thread *td)
1907 {
1908 	struct vnode *vp;
1909 
1910 	vp = fp->f_vnode;
1911 #ifdef AUDIT
1912 	vn_lock(vp, LK_SHARED | LK_RETRY);
1913 	AUDIT_ARG_VNODE1(vp);
1914 	VOP_UNLOCK(vp, 0);
1915 #endif
1916 	return (setfmode(td, active_cred, vp, mode));
1917 }
1918 
1919 int
1920 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1921     struct thread *td)
1922 {
1923 	struct vnode *vp;
1924 
1925 	vp = fp->f_vnode;
1926 #ifdef AUDIT
1927 	vn_lock(vp, LK_SHARED | LK_RETRY);
1928 	AUDIT_ARG_VNODE1(vp);
1929 	VOP_UNLOCK(vp, 0);
1930 #endif
1931 	return (setfown(td, active_cred, vp, uid, gid));
1932 }
1933 
1934 void
1935 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1936 {
1937 	vm_object_t object;
1938 
1939 	if ((object = vp->v_object) == NULL)
1940 		return;
1941 	VM_OBJECT_WLOCK(object);
1942 	vm_object_page_remove(object, start, end, 0);
1943 	VM_OBJECT_WUNLOCK(object);
1944 }
1945 
1946 int
1947 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1948 {
1949 	struct vattr va;
1950 	daddr_t bn, bnp;
1951 	uint64_t bsize;
1952 	off_t noff;
1953 	int error;
1954 
1955 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1956 	    ("Wrong command %lu", cmd));
1957 
1958 	if (vn_lock(vp, LK_SHARED) != 0)
1959 		return (EBADF);
1960 	if (vp->v_type != VREG) {
1961 		error = ENOTTY;
1962 		goto unlock;
1963 	}
1964 	error = VOP_GETATTR(vp, &va, cred);
1965 	if (error != 0)
1966 		goto unlock;
1967 	noff = *off;
1968 	if (noff >= va.va_size) {
1969 		error = ENXIO;
1970 		goto unlock;
1971 	}
1972 	bsize = vp->v_mount->mnt_stat.f_iosize;
1973 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1974 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1975 		if (error == EOPNOTSUPP) {
1976 			error = ENOTTY;
1977 			goto unlock;
1978 		}
1979 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1980 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
1981 			noff = bn * bsize;
1982 			if (noff < *off)
1983 				noff = *off;
1984 			goto unlock;
1985 		}
1986 	}
1987 	if (noff > va.va_size)
1988 		noff = va.va_size;
1989 	/* noff == va.va_size. There is an implicit hole at the end of file. */
1990 	if (cmd == FIOSEEKDATA)
1991 		error = ENXIO;
1992 unlock:
1993 	VOP_UNLOCK(vp, 0);
1994 	if (error == 0)
1995 		*off = noff;
1996 	return (error);
1997 }
1998