xref: /freebsd/sys/kern/vfs_vnops.c (revision d940bfec8c329dd82d8d54efebd81c8aa420503b)
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/disk.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/kdb.h>
52 #include <sys/stat.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/limits.h>
56 #include <sys/lock.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/vnode.h>
61 #include <sys/bio.h>
62 #include <sys/buf.h>
63 #include <sys/filio.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
66 #include <sys/sx.h>
67 #include <sys/sysctl.h>
68 #include <sys/ttycom.h>
69 #include <sys/conf.h>
70 #include <sys/syslog.h>
71 #include <sys/unistd.h>
72 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_extern.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
82 
83 static fo_rdwr_t	vn_read;
84 static fo_rdwr_t	vn_write;
85 static fo_rdwr_t	vn_io_fault;
86 static fo_truncate_t	vn_truncate;
87 static fo_ioctl_t	vn_ioctl;
88 static fo_poll_t	vn_poll;
89 static fo_kqfilter_t	vn_kqfilter;
90 static fo_stat_t	vn_statfile;
91 static fo_close_t	vn_closefile;
92 
93 struct 	fileops vnops = {
94 	.fo_read = vn_io_fault,
95 	.fo_write = vn_io_fault,
96 	.fo_truncate = vn_truncate,
97 	.fo_ioctl = vn_ioctl,
98 	.fo_poll = vn_poll,
99 	.fo_kqfilter = vn_kqfilter,
100 	.fo_stat = vn_statfile,
101 	.fo_close = vn_closefile,
102 	.fo_chmod = vn_chmod,
103 	.fo_chown = vn_chown,
104 	.fo_sendfile = vn_sendfile,
105 	.fo_seek = vn_seek,
106 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
107 };
108 
109 int
110 vn_open(ndp, flagp, cmode, fp)
111 	struct nameidata *ndp;
112 	int *flagp, cmode;
113 	struct file *fp;
114 {
115 	struct thread *td = ndp->ni_cnd.cn_thread;
116 
117 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
118 }
119 
120 /*
121  * Common code for vnode open operations via a name lookup.
122  * Lookup the vnode and invoke VOP_CREATE if needed.
123  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
124  *
125  * Note that this does NOT free nameidata for the successful case,
126  * due to the NDINIT being done elsewhere.
127  */
128 int
129 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
130     struct ucred *cred, struct file *fp)
131 {
132 	struct vnode *vp;
133 	struct mount *mp;
134 	struct thread *td = ndp->ni_cnd.cn_thread;
135 	struct vattr vat;
136 	struct vattr *vap = &vat;
137 	int fmode, error;
138 
139 restart:
140 	fmode = *flagp;
141 	if (fmode & O_CREAT) {
142 		ndp->ni_cnd.cn_nameiop = CREATE;
143 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
144 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
145 			ndp->ni_cnd.cn_flags |= FOLLOW;
146 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
147 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
148 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
149 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
150 		bwillwrite();
151 		if ((error = namei(ndp)) != 0)
152 			return (error);
153 		if (ndp->ni_vp == NULL) {
154 			VATTR_NULL(vap);
155 			vap->va_type = VREG;
156 			vap->va_mode = cmode;
157 			if (fmode & O_EXCL)
158 				vap->va_vaflags |= VA_EXCLUSIVE;
159 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
160 				NDFREE(ndp, NDF_ONLY_PNBUF);
161 				vput(ndp->ni_dvp);
162 				if ((error = vn_start_write(NULL, &mp,
163 				    V_XSLEEP | PCATCH)) != 0)
164 					return (error);
165 				goto restart;
166 			}
167 #ifdef MAC
168 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
169 			    &ndp->ni_cnd, vap);
170 			if (error == 0)
171 #endif
172 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
173 						   &ndp->ni_cnd, vap);
174 			vput(ndp->ni_dvp);
175 			vn_finished_write(mp);
176 			if (error) {
177 				NDFREE(ndp, NDF_ONLY_PNBUF);
178 				return (error);
179 			}
180 			fmode &= ~O_TRUNC;
181 			vp = ndp->ni_vp;
182 		} else {
183 			if (ndp->ni_dvp == ndp->ni_vp)
184 				vrele(ndp->ni_dvp);
185 			else
186 				vput(ndp->ni_dvp);
187 			ndp->ni_dvp = NULL;
188 			vp = ndp->ni_vp;
189 			if (fmode & O_EXCL) {
190 				error = EEXIST;
191 				goto bad;
192 			}
193 			fmode &= ~O_CREAT;
194 		}
195 	} else {
196 		ndp->ni_cnd.cn_nameiop = LOOKUP;
197 		ndp->ni_cnd.cn_flags = ISOPEN |
198 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
199 		if (!(fmode & FWRITE))
200 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
201 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
202 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
203 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
204 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
205 		if ((error = namei(ndp)) != 0)
206 			return (error);
207 		vp = ndp->ni_vp;
208 	}
209 	error = vn_open_vnode(vp, fmode, cred, td, fp);
210 	if (error)
211 		goto bad;
212 	*flagp = fmode;
213 	return (0);
214 bad:
215 	NDFREE(ndp, NDF_ONLY_PNBUF);
216 	vput(vp);
217 	*flagp = fmode;
218 	ndp->ni_vp = NULL;
219 	return (error);
220 }
221 
222 /*
223  * Common code for vnode open operations once a vnode is located.
224  * Check permissions, and call the VOP_OPEN routine.
225  */
226 int
227 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
228     struct thread *td, struct file *fp)
229 {
230 	struct mount *mp;
231 	accmode_t accmode;
232 	struct flock lf;
233 	int error, have_flock, lock_flags, type;
234 
235 	if (vp->v_type == VLNK)
236 		return (EMLINK);
237 	if (vp->v_type == VSOCK)
238 		return (EOPNOTSUPP);
239 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
240 		return (ENOTDIR);
241 	accmode = 0;
242 	if (fmode & (FWRITE | O_TRUNC)) {
243 		if (vp->v_type == VDIR)
244 			return (EISDIR);
245 		accmode |= VWRITE;
246 	}
247 	if (fmode & FREAD)
248 		accmode |= VREAD;
249 	if (fmode & FEXEC)
250 		accmode |= VEXEC;
251 	if ((fmode & O_APPEND) && (fmode & FWRITE))
252 		accmode |= VAPPEND;
253 #ifdef MAC
254 	error = mac_vnode_check_open(cred, vp, accmode);
255 	if (error)
256 		return (error);
257 #endif
258 	if ((fmode & O_CREAT) == 0) {
259 		if (accmode & VWRITE) {
260 			error = vn_writechk(vp);
261 			if (error)
262 				return (error);
263 		}
264 		if (accmode) {
265 		        error = VOP_ACCESS(vp, accmode, cred, td);
266 			if (error)
267 				return (error);
268 		}
269 	}
270 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
271 		return (error);
272 
273 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
274 		KASSERT(fp != NULL, ("open with flock requires fp"));
275 		lock_flags = VOP_ISLOCKED(vp);
276 		VOP_UNLOCK(vp, 0);
277 		lf.l_whence = SEEK_SET;
278 		lf.l_start = 0;
279 		lf.l_len = 0;
280 		if (fmode & O_EXLOCK)
281 			lf.l_type = F_WRLCK;
282 		else
283 			lf.l_type = F_RDLCK;
284 		type = F_FLOCK;
285 		if ((fmode & FNONBLOCK) == 0)
286 			type |= F_WAIT;
287 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
288 		have_flock = (error == 0);
289 		vn_lock(vp, lock_flags | LK_RETRY);
290 		if (error == 0 && vp->v_iflag & VI_DOOMED)
291 			error = ENOENT;
292 		/*
293 		 * Another thread might have used this vnode as an
294 		 * executable while the vnode lock was dropped.
295 		 * Ensure the vnode is still able to be opened for
296 		 * writing after the lock has been obtained.
297 		 */
298 		if (error == 0 && accmode & VWRITE)
299 			error = vn_writechk(vp);
300 		if (error) {
301 			VOP_UNLOCK(vp, 0);
302 			if (have_flock) {
303 				lf.l_whence = SEEK_SET;
304 				lf.l_start = 0;
305 				lf.l_len = 0;
306 				lf.l_type = F_UNLCK;
307 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
308 				    F_FLOCK);
309 			}
310 			vn_start_write(vp, &mp, V_WAIT);
311 			vn_lock(vp, lock_flags | LK_RETRY);
312 			(void)VOP_CLOSE(vp, fmode, cred, td);
313 			vn_finished_write(mp);
314 			return (error);
315 		}
316 		fp->f_flag |= FHASLOCK;
317 	}
318 	if (fmode & FWRITE) {
319 		VOP_ADD_WRITECOUNT(vp, 1);
320 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
321 		    __func__, vp, vp->v_writecount);
322 	}
323 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
324 	return (0);
325 }
326 
327 /*
328  * Check for write permissions on the specified vnode.
329  * Prototype text segments cannot be written.
330  */
331 int
332 vn_writechk(vp)
333 	register struct vnode *vp;
334 {
335 
336 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
337 	/*
338 	 * If there's shared text associated with
339 	 * the vnode, try to free it up once.  If
340 	 * we fail, we can't allow writing.
341 	 */
342 	if (VOP_IS_TEXT(vp))
343 		return (ETXTBSY);
344 
345 	return (0);
346 }
347 
348 /*
349  * Vnode close call
350  */
351 int
352 vn_close(vp, flags, file_cred, td)
353 	register struct vnode *vp;
354 	int flags;
355 	struct ucred *file_cred;
356 	struct thread *td;
357 {
358 	struct mount *mp;
359 	int error, lock_flags;
360 
361 	if (!(flags & FWRITE) && vp->v_mount != NULL &&
362 	    vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
363 		lock_flags = LK_SHARED;
364 	else
365 		lock_flags = LK_EXCLUSIVE;
366 
367 	vn_start_write(vp, &mp, V_WAIT);
368 	vn_lock(vp, lock_flags | LK_RETRY);
369 	if (flags & FWRITE) {
370 		VNASSERT(vp->v_writecount > 0, vp,
371 		    ("vn_close: negative writecount"));
372 		VOP_ADD_WRITECOUNT(vp, -1);
373 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
374 		    __func__, vp, vp->v_writecount);
375 	}
376 	error = VOP_CLOSE(vp, flags, file_cred, td);
377 	vput(vp);
378 	vn_finished_write(mp);
379 	return (error);
380 }
381 
382 /*
383  * Heuristic to detect sequential operation.
384  */
385 static int
386 sequential_heuristic(struct uio *uio, struct file *fp)
387 {
388 
389 	if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
390 		return (fp->f_seqcount << IO_SEQSHIFT);
391 
392 	/*
393 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
394 	 * that the first I/O is normally considered to be slightly
395 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
396 	 * unless previous seeks have reduced f_seqcount to 0, in which
397 	 * case offset 0 is not special.
398 	 */
399 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
400 	    uio->uio_offset == fp->f_nextoff) {
401 		/*
402 		 * f_seqcount is in units of fixed-size blocks so that it
403 		 * depends mainly on the amount of sequential I/O and not
404 		 * much on the number of sequential I/O's.  The fixed size
405 		 * of 16384 is hard-coded here since it is (not quite) just
406 		 * a magic size that works well here.  This size is more
407 		 * closely related to the best I/O size for real disks than
408 		 * to any block size used by software.
409 		 */
410 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
411 		if (fp->f_seqcount > IO_SEQMAX)
412 			fp->f_seqcount = IO_SEQMAX;
413 		return (fp->f_seqcount << IO_SEQSHIFT);
414 	}
415 
416 	/* Not sequential.  Quickly draw-down sequentiality. */
417 	if (fp->f_seqcount > 1)
418 		fp->f_seqcount = 1;
419 	else
420 		fp->f_seqcount = 0;
421 	return (0);
422 }
423 
424 /*
425  * Package up an I/O request on a vnode into a uio and do it.
426  */
427 int
428 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
429     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
430     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
431 {
432 	struct uio auio;
433 	struct iovec aiov;
434 	struct mount *mp;
435 	struct ucred *cred;
436 	void *rl_cookie;
437 	int error, lock_flags;
438 
439 	auio.uio_iov = &aiov;
440 	auio.uio_iovcnt = 1;
441 	aiov.iov_base = base;
442 	aiov.iov_len = len;
443 	auio.uio_resid = len;
444 	auio.uio_offset = offset;
445 	auio.uio_segflg = segflg;
446 	auio.uio_rw = rw;
447 	auio.uio_td = td;
448 	error = 0;
449 
450 	if ((ioflg & IO_NODELOCKED) == 0) {
451 		if (rw == UIO_READ) {
452 			rl_cookie = vn_rangelock_rlock(vp, offset,
453 			    offset + len);
454 		} else {
455 			rl_cookie = vn_rangelock_wlock(vp, offset,
456 			    offset + len);
457 		}
458 		mp = NULL;
459 		if (rw == UIO_WRITE) {
460 			if (vp->v_type != VCHR &&
461 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
462 			    != 0)
463 				goto out;
464 			if (MNT_SHARED_WRITES(mp) ||
465 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
466 				lock_flags = LK_SHARED;
467 			else
468 				lock_flags = LK_EXCLUSIVE;
469 		} else
470 			lock_flags = LK_SHARED;
471 		vn_lock(vp, lock_flags | LK_RETRY);
472 	} else
473 		rl_cookie = NULL;
474 
475 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
476 #ifdef MAC
477 	if ((ioflg & IO_NOMACCHECK) == 0) {
478 		if (rw == UIO_READ)
479 			error = mac_vnode_check_read(active_cred, file_cred,
480 			    vp);
481 		else
482 			error = mac_vnode_check_write(active_cred, file_cred,
483 			    vp);
484 	}
485 #endif
486 	if (error == 0) {
487 		if (file_cred != NULL)
488 			cred = file_cred;
489 		else
490 			cred = active_cred;
491 		if (rw == UIO_READ)
492 			error = VOP_READ(vp, &auio, ioflg, cred);
493 		else
494 			error = VOP_WRITE(vp, &auio, ioflg, cred);
495 	}
496 	if (aresid)
497 		*aresid = auio.uio_resid;
498 	else
499 		if (auio.uio_resid && error == 0)
500 			error = EIO;
501 	if ((ioflg & IO_NODELOCKED) == 0) {
502 		VOP_UNLOCK(vp, 0);
503 		if (mp != NULL)
504 			vn_finished_write(mp);
505 	}
506  out:
507 	if (rl_cookie != NULL)
508 		vn_rangelock_unlock(vp, rl_cookie);
509 	return (error);
510 }
511 
512 /*
513  * Package up an I/O request on a vnode into a uio and do it.  The I/O
514  * request is split up into smaller chunks and we try to avoid saturating
515  * the buffer cache while potentially holding a vnode locked, so we
516  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
517  * to give other processes a chance to lock the vnode (either other processes
518  * core'ing the same binary, or unrelated processes scanning the directory).
519  */
520 int
521 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
522     file_cred, aresid, td)
523 	enum uio_rw rw;
524 	struct vnode *vp;
525 	void *base;
526 	size_t len;
527 	off_t offset;
528 	enum uio_seg segflg;
529 	int ioflg;
530 	struct ucred *active_cred;
531 	struct ucred *file_cred;
532 	size_t *aresid;
533 	struct thread *td;
534 {
535 	int error = 0;
536 	ssize_t iaresid;
537 
538 	do {
539 		int chunk;
540 
541 		/*
542 		 * Force `offset' to a multiple of MAXBSIZE except possibly
543 		 * for the first chunk, so that filesystems only need to
544 		 * write full blocks except possibly for the first and last
545 		 * chunks.
546 		 */
547 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
548 
549 		if (chunk > len)
550 			chunk = len;
551 		if (rw != UIO_READ && vp->v_type == VREG)
552 			bwillwrite();
553 		iaresid = 0;
554 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
555 		    ioflg, active_cred, file_cred, &iaresid, td);
556 		len -= chunk;	/* aresid calc already includes length */
557 		if (error)
558 			break;
559 		offset += chunk;
560 		base = (char *)base + chunk;
561 		kern_yield(PRI_USER);
562 	} while (len);
563 	if (aresid)
564 		*aresid = len + iaresid;
565 	return (error);
566 }
567 
568 off_t
569 foffset_lock(struct file *fp, int flags)
570 {
571 	struct mtx *mtxp;
572 	off_t res;
573 
574 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
575 
576 #if OFF_MAX <= LONG_MAX
577 	/*
578 	 * Caller only wants the current f_offset value.  Assume that
579 	 * the long and shorter integer types reads are atomic.
580 	 */
581 	if ((flags & FOF_NOLOCK) != 0)
582 		return (fp->f_offset);
583 #endif
584 
585 	/*
586 	 * According to McKusick the vn lock was protecting f_offset here.
587 	 * It is now protected by the FOFFSET_LOCKED flag.
588 	 */
589 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
590 	mtx_lock(mtxp);
591 	if ((flags & FOF_NOLOCK) == 0) {
592 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
593 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
594 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
595 			    "vofflock", 0);
596 		}
597 		fp->f_vnread_flags |= FOFFSET_LOCKED;
598 	}
599 	res = fp->f_offset;
600 	mtx_unlock(mtxp);
601 	return (res);
602 }
603 
604 void
605 foffset_unlock(struct file *fp, off_t val, int flags)
606 {
607 	struct mtx *mtxp;
608 
609 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
610 
611 #if OFF_MAX <= LONG_MAX
612 	if ((flags & FOF_NOLOCK) != 0) {
613 		if ((flags & FOF_NOUPDATE) == 0)
614 			fp->f_offset = val;
615 		if ((flags & FOF_NEXTOFF) != 0)
616 			fp->f_nextoff = val;
617 		return;
618 	}
619 #endif
620 
621 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
622 	mtx_lock(mtxp);
623 	if ((flags & FOF_NOUPDATE) == 0)
624 		fp->f_offset = val;
625 	if ((flags & FOF_NEXTOFF) != 0)
626 		fp->f_nextoff = val;
627 	if ((flags & FOF_NOLOCK) == 0) {
628 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
629 		    ("Lost FOFFSET_LOCKED"));
630 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
631 			wakeup(&fp->f_vnread_flags);
632 		fp->f_vnread_flags = 0;
633 	}
634 	mtx_unlock(mtxp);
635 }
636 
637 void
638 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
639 {
640 
641 	if ((flags & FOF_OFFSET) == 0)
642 		uio->uio_offset = foffset_lock(fp, flags);
643 }
644 
645 void
646 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
647 {
648 
649 	if ((flags & FOF_OFFSET) == 0)
650 		foffset_unlock(fp, uio->uio_offset, flags);
651 }
652 
653 static int
654 get_advice(struct file *fp, struct uio *uio)
655 {
656 	struct mtx *mtxp;
657 	int ret;
658 
659 	ret = POSIX_FADV_NORMAL;
660 	if (fp->f_advice == NULL)
661 		return (ret);
662 
663 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
664 	mtx_lock(mtxp);
665 	if (uio->uio_offset >= fp->f_advice->fa_start &&
666 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
667 		ret = fp->f_advice->fa_advice;
668 	mtx_unlock(mtxp);
669 	return (ret);
670 }
671 
672 /*
673  * File table vnode read routine.
674  */
675 static int
676 vn_read(fp, uio, active_cred, flags, td)
677 	struct file *fp;
678 	struct uio *uio;
679 	struct ucred *active_cred;
680 	int flags;
681 	struct thread *td;
682 {
683 	struct vnode *vp;
684 	struct mtx *mtxp;
685 	int error, ioflag;
686 	int advice;
687 	off_t offset, start, end;
688 
689 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
690 	    uio->uio_td, td));
691 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
692 	vp = fp->f_vnode;
693 	ioflag = 0;
694 	if (fp->f_flag & FNONBLOCK)
695 		ioflag |= IO_NDELAY;
696 	if (fp->f_flag & O_DIRECT)
697 		ioflag |= IO_DIRECT;
698 	advice = get_advice(fp, uio);
699 	vn_lock(vp, LK_SHARED | LK_RETRY);
700 
701 	switch (advice) {
702 	case POSIX_FADV_NORMAL:
703 	case POSIX_FADV_SEQUENTIAL:
704 	case POSIX_FADV_NOREUSE:
705 		ioflag |= sequential_heuristic(uio, fp);
706 		break;
707 	case POSIX_FADV_RANDOM:
708 		/* Disable read-ahead for random I/O. */
709 		break;
710 	}
711 	offset = uio->uio_offset;
712 
713 #ifdef MAC
714 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
715 	if (error == 0)
716 #endif
717 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
718 	fp->f_nextoff = uio->uio_offset;
719 	VOP_UNLOCK(vp, 0);
720 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
721 	    offset != uio->uio_offset) {
722 		/*
723 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
724 		 * buffers for the backing file after a
725 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
726 		 * case of using POSIX_FADV_NOREUSE with sequential
727 		 * access, track the previous implicit DONTNEED
728 		 * request and grow this request to include the
729 		 * current read(2) in addition to the previous
730 		 * DONTNEED.  With purely sequential access this will
731 		 * cause the DONTNEED requests to continously grow to
732 		 * cover all of the previously read regions of the
733 		 * file.  This allows filesystem blocks that are
734 		 * accessed by multiple calls to read(2) to be flushed
735 		 * once the last read(2) finishes.
736 		 */
737 		start = offset;
738 		end = uio->uio_offset - 1;
739 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
740 		mtx_lock(mtxp);
741 		if (fp->f_advice != NULL &&
742 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
743 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
744 				start = fp->f_advice->fa_prevstart;
745 			else if (fp->f_advice->fa_prevstart != 0 &&
746 			    fp->f_advice->fa_prevstart == end + 1)
747 				end = fp->f_advice->fa_prevend;
748 			fp->f_advice->fa_prevstart = start;
749 			fp->f_advice->fa_prevend = end;
750 		}
751 		mtx_unlock(mtxp);
752 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
753 	}
754 	return (error);
755 }
756 
757 /*
758  * File table vnode write routine.
759  */
760 static int
761 vn_write(fp, uio, active_cred, flags, td)
762 	struct file *fp;
763 	struct uio *uio;
764 	struct ucred *active_cred;
765 	int flags;
766 	struct thread *td;
767 {
768 	struct vnode *vp;
769 	struct mount *mp;
770 	struct mtx *mtxp;
771 	int error, ioflag, lock_flags;
772 	int advice;
773 	off_t offset, start, end;
774 
775 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
776 	    uio->uio_td, td));
777 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
778 	vp = fp->f_vnode;
779 	if (vp->v_type == VREG)
780 		bwillwrite();
781 	ioflag = IO_UNIT;
782 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
783 		ioflag |= IO_APPEND;
784 	if (fp->f_flag & FNONBLOCK)
785 		ioflag |= IO_NDELAY;
786 	if (fp->f_flag & O_DIRECT)
787 		ioflag |= IO_DIRECT;
788 	if ((fp->f_flag & O_FSYNC) ||
789 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
790 		ioflag |= IO_SYNC;
791 	mp = NULL;
792 	if (vp->v_type != VCHR &&
793 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
794 		goto unlock;
795 
796 	advice = get_advice(fp, uio);
797 
798 	if (MNT_SHARED_WRITES(mp) ||
799 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
800 		lock_flags = LK_SHARED;
801 	} else {
802 		lock_flags = LK_EXCLUSIVE;
803 	}
804 
805 	vn_lock(vp, lock_flags | LK_RETRY);
806 	switch (advice) {
807 	case POSIX_FADV_NORMAL:
808 	case POSIX_FADV_SEQUENTIAL:
809 	case POSIX_FADV_NOREUSE:
810 		ioflag |= sequential_heuristic(uio, fp);
811 		break;
812 	case POSIX_FADV_RANDOM:
813 		/* XXX: Is this correct? */
814 		break;
815 	}
816 	offset = uio->uio_offset;
817 
818 #ifdef MAC
819 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
820 	if (error == 0)
821 #endif
822 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
823 	fp->f_nextoff = uio->uio_offset;
824 	VOP_UNLOCK(vp, 0);
825 	if (vp->v_type != VCHR)
826 		vn_finished_write(mp);
827 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
828 	    offset != uio->uio_offset) {
829 		/*
830 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
831 		 * buffers for the backing file after a
832 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
833 		 * common case of using POSIX_FADV_NOREUSE with
834 		 * sequential access, track the previous implicit
835 		 * DONTNEED request and grow this request to include
836 		 * the current write(2) in addition to the previous
837 		 * DONTNEED.  With purely sequential access this will
838 		 * cause the DONTNEED requests to continously grow to
839 		 * cover all of the previously written regions of the
840 		 * file.
841 		 *
842 		 * Note that the blocks just written are almost
843 		 * certainly still dirty, so this only works when
844 		 * VOP_ADVISE() calls from subsequent writes push out
845 		 * the data written by this write(2) once the backing
846 		 * buffers are clean.  However, as compared to forcing
847 		 * IO_DIRECT, this gives much saner behavior.  Write
848 		 * clustering is still allowed, and clean pages are
849 		 * merely moved to the cache page queue rather than
850 		 * outright thrown away.  This means a subsequent
851 		 * read(2) can still avoid hitting the disk if the
852 		 * pages have not been reclaimed.
853 		 *
854 		 * This does make POSIX_FADV_NOREUSE largely useless
855 		 * with non-sequential access.  However, sequential
856 		 * access is the more common use case and the flag is
857 		 * merely advisory.
858 		 */
859 		start = offset;
860 		end = uio->uio_offset - 1;
861 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
862 		mtx_lock(mtxp);
863 		if (fp->f_advice != NULL &&
864 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
865 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
866 				start = fp->f_advice->fa_prevstart;
867 			else if (fp->f_advice->fa_prevstart != 0 &&
868 			    fp->f_advice->fa_prevstart == end + 1)
869 				end = fp->f_advice->fa_prevend;
870 			fp->f_advice->fa_prevstart = start;
871 			fp->f_advice->fa_prevend = end;
872 		}
873 		mtx_unlock(mtxp);
874 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
875 	}
876 
877 unlock:
878 	return (error);
879 }
880 
881 static const int io_hold_cnt = 16;
882 static int vn_io_fault_enable = 1;
883 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
884     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
885 static u_long vn_io_faults_cnt;
886 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
887     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
888 
889 /*
890  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
891  * prevent the following deadlock:
892  *
893  * Assume that the thread A reads from the vnode vp1 into userspace
894  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
895  * currently not resident, then system ends up with the call chain
896  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
897  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
898  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
899  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
900  * backed by the pages of vnode vp1, and some page in buf2 is not
901  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
902  *
903  * To prevent the lock order reversal and deadlock, vn_io_fault() does
904  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
905  * Instead, it first tries to do the whole range i/o with pagefaults
906  * disabled. If all pages in the i/o buffer are resident and mapped,
907  * VOP will succeed (ignoring the genuine filesystem errors).
908  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
909  * i/o in chunks, with all pages in the chunk prefaulted and held
910  * using vm_fault_quick_hold_pages().
911  *
912  * Filesystems using this deadlock avoidance scheme should use the
913  * array of the held pages from uio, saved in the curthread->td_ma,
914  * instead of doing uiomove().  A helper function
915  * vn_io_fault_uiomove() converts uiomove request into
916  * uiomove_fromphys() over td_ma array.
917  *
918  * Since vnode locks do not cover the whole i/o anymore, rangelocks
919  * make the current i/o request atomic with respect to other i/os and
920  * truncations.
921  */
922 static int
923 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
924     int flags, struct thread *td)
925 {
926 	vm_page_t ma[io_hold_cnt + 2];
927 	struct uio *uio_clone, short_uio;
928 	struct iovec short_iovec[1];
929 	fo_rdwr_t *doio;
930 	struct vnode *vp;
931 	void *rl_cookie;
932 	struct mount *mp;
933 	vm_page_t *prev_td_ma;
934 	int cnt, error, save, saveheld, prev_td_ma_cnt;
935 	vm_offset_t addr, end;
936 	vm_prot_t prot;
937 	size_t len, resid;
938 	ssize_t adv;
939 
940 	if (uio->uio_rw == UIO_READ)
941 		doio = vn_read;
942 	else
943 		doio = vn_write;
944 	vp = fp->f_vnode;
945 	foffset_lock_uio(fp, uio, flags);
946 
947 	if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
948 	    ((mp = vp->v_mount) != NULL &&
949 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
950 	    !vn_io_fault_enable) {
951 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
952 		goto out_last;
953 	}
954 
955 	/*
956 	 * The UFS follows IO_UNIT directive and replays back both
957 	 * uio_offset and uio_resid if an error is encountered during the
958 	 * operation.  But, since the iovec may be already advanced,
959 	 * uio is still in an inconsistent state.
960 	 *
961 	 * Cache a copy of the original uio, which is advanced to the redo
962 	 * point using UIO_NOCOPY below.
963 	 */
964 	uio_clone = cloneuio(uio);
965 	resid = uio->uio_resid;
966 
967 	short_uio.uio_segflg = UIO_USERSPACE;
968 	short_uio.uio_rw = uio->uio_rw;
969 	short_uio.uio_td = uio->uio_td;
970 
971 	if (uio->uio_rw == UIO_READ) {
972 		prot = VM_PROT_WRITE;
973 		rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
974 		    uio->uio_offset + uio->uio_resid);
975 	} else {
976 		prot = VM_PROT_READ;
977 		if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
978 			/* For appenders, punt and lock the whole range. */
979 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
980 		else
981 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
982 			    uio->uio_offset + uio->uio_resid);
983 	}
984 
985 	save = vm_fault_disable_pagefaults();
986 	error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
987 	if (error != EFAULT)
988 		goto out;
989 
990 	atomic_add_long(&vn_io_faults_cnt, 1);
991 	uio_clone->uio_segflg = UIO_NOCOPY;
992 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
993 	uio_clone->uio_segflg = uio->uio_segflg;
994 
995 	saveheld = curthread_pflags_set(TDP_UIOHELD);
996 	prev_td_ma = td->td_ma;
997 	prev_td_ma_cnt = td->td_ma_cnt;
998 
999 	while (uio_clone->uio_resid != 0) {
1000 		len = uio_clone->uio_iov->iov_len;
1001 		if (len == 0) {
1002 			KASSERT(uio_clone->uio_iovcnt >= 1,
1003 			    ("iovcnt underflow"));
1004 			uio_clone->uio_iov++;
1005 			uio_clone->uio_iovcnt--;
1006 			continue;
1007 		}
1008 
1009 		addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
1010 		end = round_page(addr + len);
1011 		cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
1012 		/*
1013 		 * A perfectly misaligned address and length could cause
1014 		 * both the start and the end of the chunk to use partial
1015 		 * page.  +2 accounts for such a situation.
1016 		 */
1017 		if (cnt > io_hold_cnt + 2) {
1018 			len = io_hold_cnt * PAGE_SIZE;
1019 			KASSERT(howmany(round_page(addr + len) -
1020 			    trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
1021 			    ("cnt overflow"));
1022 		}
1023 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1024 		    addr, len, prot, ma, io_hold_cnt + 2);
1025 		if (cnt == -1) {
1026 			error = EFAULT;
1027 			break;
1028 		}
1029 		short_uio.uio_iov = &short_iovec[0];
1030 		short_iovec[0].iov_base = (void *)addr;
1031 		short_uio.uio_iovcnt = 1;
1032 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1033 		short_uio.uio_offset = uio_clone->uio_offset;
1034 		td->td_ma = ma;
1035 		td->td_ma_cnt = cnt;
1036 
1037 		error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
1038 		    td);
1039 		vm_page_unhold_pages(ma, cnt);
1040 		adv = len - short_uio.uio_resid;
1041 
1042 		uio_clone->uio_iov->iov_base =
1043 		    (char *)uio_clone->uio_iov->iov_base + adv;
1044 		uio_clone->uio_iov->iov_len -= adv;
1045 		uio_clone->uio_resid -= adv;
1046 		uio_clone->uio_offset += adv;
1047 
1048 		uio->uio_resid -= adv;
1049 		uio->uio_offset += adv;
1050 
1051 		if (error != 0 || adv == 0)
1052 			break;
1053 	}
1054 	td->td_ma = prev_td_ma;
1055 	td->td_ma_cnt = prev_td_ma_cnt;
1056 	curthread_pflags_restore(saveheld);
1057 out:
1058 	vm_fault_enable_pagefaults(save);
1059 	vn_rangelock_unlock(vp, rl_cookie);
1060 	free(uio_clone, M_IOV);
1061 out_last:
1062 	foffset_unlock_uio(fp, uio, flags);
1063 	return (error);
1064 }
1065 
1066 /*
1067  * Helper function to perform the requested uiomove operation using
1068  * the held pages for io->uio_iov[0].iov_base buffer instead of
1069  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1070  * instead of iov_base prevents page faults that could occur due to
1071  * pmap_collect() invalidating the mapping created by
1072  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1073  * object cleanup revoking the write access from page mappings.
1074  *
1075  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1076  * instead of plain uiomove().
1077  */
1078 int
1079 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1080 {
1081 	struct uio transp_uio;
1082 	struct iovec transp_iov[1];
1083 	struct thread *td;
1084 	size_t adv;
1085 	int error, pgadv;
1086 
1087 	td = curthread;
1088 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1089 	    uio->uio_segflg != UIO_USERSPACE)
1090 		return (uiomove(data, xfersize, uio));
1091 
1092 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1093 	transp_iov[0].iov_base = data;
1094 	transp_uio.uio_iov = &transp_iov[0];
1095 	transp_uio.uio_iovcnt = 1;
1096 	if (xfersize > uio->uio_resid)
1097 		xfersize = uio->uio_resid;
1098 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1099 	transp_uio.uio_offset = 0;
1100 	transp_uio.uio_segflg = UIO_SYSSPACE;
1101 	/*
1102 	 * Since transp_iov points to data, and td_ma page array
1103 	 * corresponds to original uio->uio_iov, we need to invert the
1104 	 * direction of the i/o operation as passed to
1105 	 * uiomove_fromphys().
1106 	 */
1107 	switch (uio->uio_rw) {
1108 	case UIO_WRITE:
1109 		transp_uio.uio_rw = UIO_READ;
1110 		break;
1111 	case UIO_READ:
1112 		transp_uio.uio_rw = UIO_WRITE;
1113 		break;
1114 	}
1115 	transp_uio.uio_td = uio->uio_td;
1116 	error = uiomove_fromphys(td->td_ma,
1117 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1118 	    xfersize, &transp_uio);
1119 	adv = xfersize - transp_uio.uio_resid;
1120 	pgadv =
1121 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1122 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1123 	td->td_ma += pgadv;
1124 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1125 	    pgadv));
1126 	td->td_ma_cnt -= pgadv;
1127 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1128 	uio->uio_iov->iov_len -= adv;
1129 	uio->uio_resid -= adv;
1130 	uio->uio_offset += adv;
1131 	return (error);
1132 }
1133 
1134 int
1135 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1136     struct uio *uio)
1137 {
1138 	struct thread *td;
1139 	vm_offset_t iov_base;
1140 	int cnt, pgadv;
1141 
1142 	td = curthread;
1143 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1144 	    uio->uio_segflg != UIO_USERSPACE)
1145 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1146 
1147 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1148 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1149 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1150 	switch (uio->uio_rw) {
1151 	case UIO_WRITE:
1152 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1153 		    offset, cnt);
1154 		break;
1155 	case UIO_READ:
1156 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1157 		    cnt);
1158 		break;
1159 	}
1160 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1161 	td->td_ma += pgadv;
1162 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1163 	    pgadv));
1164 	td->td_ma_cnt -= pgadv;
1165 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1166 	uio->uio_iov->iov_len -= cnt;
1167 	uio->uio_resid -= cnt;
1168 	uio->uio_offset += cnt;
1169 	return (0);
1170 }
1171 
1172 
1173 /*
1174  * File table truncate routine.
1175  */
1176 static int
1177 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1178     struct thread *td)
1179 {
1180 	struct vattr vattr;
1181 	struct mount *mp;
1182 	struct vnode *vp;
1183 	void *rl_cookie;
1184 	int error;
1185 
1186 	vp = fp->f_vnode;
1187 
1188 	/*
1189 	 * Lock the whole range for truncation.  Otherwise split i/o
1190 	 * might happen partly before and partly after the truncation.
1191 	 */
1192 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1193 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1194 	if (error)
1195 		goto out1;
1196 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1197 	if (vp->v_type == VDIR) {
1198 		error = EISDIR;
1199 		goto out;
1200 	}
1201 #ifdef MAC
1202 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1203 	if (error)
1204 		goto out;
1205 #endif
1206 	error = vn_writechk(vp);
1207 	if (error == 0) {
1208 		VATTR_NULL(&vattr);
1209 		vattr.va_size = length;
1210 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1211 	}
1212 out:
1213 	VOP_UNLOCK(vp, 0);
1214 	vn_finished_write(mp);
1215 out1:
1216 	vn_rangelock_unlock(vp, rl_cookie);
1217 	return (error);
1218 }
1219 
1220 /*
1221  * File table vnode stat routine.
1222  */
1223 static int
1224 vn_statfile(fp, sb, active_cred, td)
1225 	struct file *fp;
1226 	struct stat *sb;
1227 	struct ucred *active_cred;
1228 	struct thread *td;
1229 {
1230 	struct vnode *vp = fp->f_vnode;
1231 	int error;
1232 
1233 	vn_lock(vp, LK_SHARED | LK_RETRY);
1234 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1235 	VOP_UNLOCK(vp, 0);
1236 
1237 	return (error);
1238 }
1239 
1240 /*
1241  * Stat a vnode; implementation for the stat syscall
1242  */
1243 int
1244 vn_stat(vp, sb, active_cred, file_cred, td)
1245 	struct vnode *vp;
1246 	register struct stat *sb;
1247 	struct ucred *active_cred;
1248 	struct ucred *file_cred;
1249 	struct thread *td;
1250 {
1251 	struct vattr vattr;
1252 	register struct vattr *vap;
1253 	int error;
1254 	u_short mode;
1255 
1256 #ifdef MAC
1257 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1258 	if (error)
1259 		return (error);
1260 #endif
1261 
1262 	vap = &vattr;
1263 
1264 	/*
1265 	 * Initialize defaults for new and unusual fields, so that file
1266 	 * systems which don't support these fields don't need to know
1267 	 * about them.
1268 	 */
1269 	vap->va_birthtime.tv_sec = -1;
1270 	vap->va_birthtime.tv_nsec = 0;
1271 	vap->va_fsid = VNOVAL;
1272 	vap->va_rdev = NODEV;
1273 
1274 	error = VOP_GETATTR(vp, vap, active_cred);
1275 	if (error)
1276 		return (error);
1277 
1278 	/*
1279 	 * Zero the spare stat fields
1280 	 */
1281 	bzero(sb, sizeof *sb);
1282 
1283 	/*
1284 	 * Copy from vattr table
1285 	 */
1286 	if (vap->va_fsid != VNOVAL)
1287 		sb->st_dev = vap->va_fsid;
1288 	else
1289 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1290 	sb->st_ino = vap->va_fileid;
1291 	mode = vap->va_mode;
1292 	switch (vap->va_type) {
1293 	case VREG:
1294 		mode |= S_IFREG;
1295 		break;
1296 	case VDIR:
1297 		mode |= S_IFDIR;
1298 		break;
1299 	case VBLK:
1300 		mode |= S_IFBLK;
1301 		break;
1302 	case VCHR:
1303 		mode |= S_IFCHR;
1304 		break;
1305 	case VLNK:
1306 		mode |= S_IFLNK;
1307 		break;
1308 	case VSOCK:
1309 		mode |= S_IFSOCK;
1310 		break;
1311 	case VFIFO:
1312 		mode |= S_IFIFO;
1313 		break;
1314 	default:
1315 		return (EBADF);
1316 	};
1317 	sb->st_mode = mode;
1318 	sb->st_nlink = vap->va_nlink;
1319 	sb->st_uid = vap->va_uid;
1320 	sb->st_gid = vap->va_gid;
1321 	sb->st_rdev = vap->va_rdev;
1322 	if (vap->va_size > OFF_MAX)
1323 		return (EOVERFLOW);
1324 	sb->st_size = vap->va_size;
1325 	sb->st_atim = vap->va_atime;
1326 	sb->st_mtim = vap->va_mtime;
1327 	sb->st_ctim = vap->va_ctime;
1328 	sb->st_birthtim = vap->va_birthtime;
1329 
1330         /*
1331 	 * According to www.opengroup.org, the meaning of st_blksize is
1332 	 *   "a filesystem-specific preferred I/O block size for this
1333 	 *    object.  In some filesystem types, this may vary from file
1334 	 *    to file"
1335 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1336 	 */
1337 
1338 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1339 
1340 	sb->st_flags = vap->va_flags;
1341 	if (priv_check(td, PRIV_VFS_GENERATION))
1342 		sb->st_gen = 0;
1343 	else
1344 		sb->st_gen = vap->va_gen;
1345 
1346 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1347 	return (0);
1348 }
1349 
1350 /*
1351  * File table vnode ioctl routine.
1352  */
1353 static int
1354 vn_ioctl(fp, com, data, active_cred, td)
1355 	struct file *fp;
1356 	u_long com;
1357 	void *data;
1358 	struct ucred *active_cred;
1359 	struct thread *td;
1360 {
1361 	struct vattr vattr;
1362 	struct vnode *vp;
1363 	int error;
1364 
1365 	vp = fp->f_vnode;
1366 	switch (vp->v_type) {
1367 	case VDIR:
1368 	case VREG:
1369 		switch (com) {
1370 		case FIONREAD:
1371 			vn_lock(vp, LK_SHARED | LK_RETRY);
1372 			error = VOP_GETATTR(vp, &vattr, active_cred);
1373 			VOP_UNLOCK(vp, 0);
1374 			if (error == 0)
1375 				*(int *)data = vattr.va_size - fp->f_offset;
1376 			return (error);
1377 		case FIONBIO:
1378 		case FIOASYNC:
1379 			return (0);
1380 		default:
1381 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1382 			    active_cred, td));
1383 		}
1384 	default:
1385 		return (ENOTTY);
1386 	}
1387 }
1388 
1389 /*
1390  * File table vnode poll routine.
1391  */
1392 static int
1393 vn_poll(fp, events, active_cred, td)
1394 	struct file *fp;
1395 	int events;
1396 	struct ucred *active_cred;
1397 	struct thread *td;
1398 {
1399 	struct vnode *vp;
1400 	int error;
1401 
1402 	vp = fp->f_vnode;
1403 #ifdef MAC
1404 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1405 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1406 	VOP_UNLOCK(vp, 0);
1407 	if (!error)
1408 #endif
1409 
1410 	error = VOP_POLL(vp, events, fp->f_cred, td);
1411 	return (error);
1412 }
1413 
1414 /*
1415  * Acquire the requested lock and then check for validity.  LK_RETRY
1416  * permits vn_lock to return doomed vnodes.
1417  */
1418 int
1419 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1420 {
1421 	int error;
1422 
1423 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1424 	    ("vn_lock called with no locktype."));
1425 	do {
1426 #ifdef DEBUG_VFS_LOCKS
1427 		KASSERT(vp->v_holdcnt != 0,
1428 		    ("vn_lock %p: zero hold count", vp));
1429 #endif
1430 		error = VOP_LOCK1(vp, flags, file, line);
1431 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1432 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1433 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1434 		    flags, error));
1435 		/*
1436 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1437 		 * If RETRY is not set, we return ENOENT instead.
1438 		 */
1439 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1440 		    (flags & LK_RETRY) == 0) {
1441 			VOP_UNLOCK(vp, 0);
1442 			error = ENOENT;
1443 			break;
1444 		}
1445 	} while (flags & LK_RETRY && error != 0);
1446 	return (error);
1447 }
1448 
1449 /*
1450  * File table vnode close routine.
1451  */
1452 static int
1453 vn_closefile(fp, td)
1454 	struct file *fp;
1455 	struct thread *td;
1456 {
1457 	struct vnode *vp;
1458 	struct flock lf;
1459 	int error;
1460 
1461 	vp = fp->f_vnode;
1462 	fp->f_ops = &badfileops;
1463 
1464 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1465 		vref(vp);
1466 
1467 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1468 
1469 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1470 		lf.l_whence = SEEK_SET;
1471 		lf.l_start = 0;
1472 		lf.l_len = 0;
1473 		lf.l_type = F_UNLCK;
1474 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1475 		vrele(vp);
1476 	}
1477 	return (error);
1478 }
1479 
1480 /*
1481  * Preparing to start a filesystem write operation. If the operation is
1482  * permitted, then we bump the count of operations in progress and
1483  * proceed. If a suspend request is in progress, we wait until the
1484  * suspension is over, and then proceed.
1485  */
1486 static int
1487 vn_start_write_locked(struct mount *mp, int flags)
1488 {
1489 	int error;
1490 
1491 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1492 	error = 0;
1493 
1494 	/*
1495 	 * Check on status of suspension.
1496 	 */
1497 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1498 	    mp->mnt_susp_owner != curthread) {
1499 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1500 			if (flags & V_NOWAIT) {
1501 				error = EWOULDBLOCK;
1502 				goto unlock;
1503 			}
1504 			error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1505 			    (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1506 			if (error)
1507 				goto unlock;
1508 		}
1509 	}
1510 	if (flags & V_XSLEEP)
1511 		goto unlock;
1512 	mp->mnt_writeopcount++;
1513 unlock:
1514 	if (error != 0 || (flags & V_XSLEEP) != 0)
1515 		MNT_REL(mp);
1516 	MNT_IUNLOCK(mp);
1517 	return (error);
1518 }
1519 
1520 int
1521 vn_start_write(vp, mpp, flags)
1522 	struct vnode *vp;
1523 	struct mount **mpp;
1524 	int flags;
1525 {
1526 	struct mount *mp;
1527 	int error;
1528 
1529 	error = 0;
1530 	/*
1531 	 * If a vnode is provided, get and return the mount point that
1532 	 * to which it will write.
1533 	 */
1534 	if (vp != NULL) {
1535 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1536 			*mpp = NULL;
1537 			if (error != EOPNOTSUPP)
1538 				return (error);
1539 			return (0);
1540 		}
1541 	}
1542 	if ((mp = *mpp) == NULL)
1543 		return (0);
1544 
1545 	/*
1546 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1547 	 * a vfs_ref().
1548 	 * As long as a vnode is not provided we need to acquire a
1549 	 * refcount for the provided mountpoint too, in order to
1550 	 * emulate a vfs_ref().
1551 	 */
1552 	MNT_ILOCK(mp);
1553 	if (vp == NULL)
1554 		MNT_REF(mp);
1555 
1556 	return (vn_start_write_locked(mp, flags));
1557 }
1558 
1559 /*
1560  * Secondary suspension. Used by operations such as vop_inactive
1561  * routines that are needed by the higher level functions. These
1562  * are allowed to proceed until all the higher level functions have
1563  * completed (indicated by mnt_writeopcount dropping to zero). At that
1564  * time, these operations are halted until the suspension is over.
1565  */
1566 int
1567 vn_start_secondary_write(vp, mpp, flags)
1568 	struct vnode *vp;
1569 	struct mount **mpp;
1570 	int flags;
1571 {
1572 	struct mount *mp;
1573 	int error;
1574 
1575  retry:
1576 	if (vp != NULL) {
1577 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1578 			*mpp = NULL;
1579 			if (error != EOPNOTSUPP)
1580 				return (error);
1581 			return (0);
1582 		}
1583 	}
1584 	/*
1585 	 * If we are not suspended or have not yet reached suspended
1586 	 * mode, then let the operation proceed.
1587 	 */
1588 	if ((mp = *mpp) == NULL)
1589 		return (0);
1590 
1591 	/*
1592 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1593 	 * a vfs_ref().
1594 	 * As long as a vnode is not provided we need to acquire a
1595 	 * refcount for the provided mountpoint too, in order to
1596 	 * emulate a vfs_ref().
1597 	 */
1598 	MNT_ILOCK(mp);
1599 	if (vp == NULL)
1600 		MNT_REF(mp);
1601 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1602 		mp->mnt_secondary_writes++;
1603 		mp->mnt_secondary_accwrites++;
1604 		MNT_IUNLOCK(mp);
1605 		return (0);
1606 	}
1607 	if (flags & V_NOWAIT) {
1608 		MNT_REL(mp);
1609 		MNT_IUNLOCK(mp);
1610 		return (EWOULDBLOCK);
1611 	}
1612 	/*
1613 	 * Wait for the suspension to finish.
1614 	 */
1615 	error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1616 		       (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1617 	vfs_rel(mp);
1618 	if (error == 0)
1619 		goto retry;
1620 	return (error);
1621 }
1622 
1623 /*
1624  * Filesystem write operation has completed. If we are suspending and this
1625  * operation is the last one, notify the suspender that the suspension is
1626  * now in effect.
1627  */
1628 void
1629 vn_finished_write(mp)
1630 	struct mount *mp;
1631 {
1632 	if (mp == NULL)
1633 		return;
1634 	MNT_ILOCK(mp);
1635 	MNT_REL(mp);
1636 	mp->mnt_writeopcount--;
1637 	if (mp->mnt_writeopcount < 0)
1638 		panic("vn_finished_write: neg cnt");
1639 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1640 	    mp->mnt_writeopcount <= 0)
1641 		wakeup(&mp->mnt_writeopcount);
1642 	MNT_IUNLOCK(mp);
1643 }
1644 
1645 
1646 /*
1647  * Filesystem secondary write operation has completed. If we are
1648  * suspending and this operation is the last one, notify the suspender
1649  * that the suspension is now in effect.
1650  */
1651 void
1652 vn_finished_secondary_write(mp)
1653 	struct mount *mp;
1654 {
1655 	if (mp == NULL)
1656 		return;
1657 	MNT_ILOCK(mp);
1658 	MNT_REL(mp);
1659 	mp->mnt_secondary_writes--;
1660 	if (mp->mnt_secondary_writes < 0)
1661 		panic("vn_finished_secondary_write: neg cnt");
1662 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1663 	    mp->mnt_secondary_writes <= 0)
1664 		wakeup(&mp->mnt_secondary_writes);
1665 	MNT_IUNLOCK(mp);
1666 }
1667 
1668 
1669 
1670 /*
1671  * Request a filesystem to suspend write operations.
1672  */
1673 int
1674 vfs_write_suspend(struct mount *mp, int flags)
1675 {
1676 	int error;
1677 
1678 	MNT_ILOCK(mp);
1679 	if (mp->mnt_susp_owner == curthread) {
1680 		MNT_IUNLOCK(mp);
1681 		return (EALREADY);
1682 	}
1683 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1684 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1685 
1686 	/*
1687 	 * Unmount holds a write reference on the mount point.  If we
1688 	 * own busy reference and drain for writers, we deadlock with
1689 	 * the reference draining in the unmount path.  Callers of
1690 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1691 	 * vfs_busy() reference is owned and caller is not in the
1692 	 * unmount context.
1693 	 */
1694 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1695 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1696 		MNT_IUNLOCK(mp);
1697 		return (EBUSY);
1698 	}
1699 
1700 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1701 	mp->mnt_susp_owner = curthread;
1702 	if (mp->mnt_writeopcount > 0)
1703 		(void) msleep(&mp->mnt_writeopcount,
1704 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1705 	else
1706 		MNT_IUNLOCK(mp);
1707 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1708 		vfs_write_resume(mp, 0);
1709 	return (error);
1710 }
1711 
1712 /*
1713  * Request a filesystem to resume write operations.
1714  */
1715 void
1716 vfs_write_resume(struct mount *mp, int flags)
1717 {
1718 
1719 	MNT_ILOCK(mp);
1720 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1721 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1722 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1723 				       MNTK_SUSPENDED);
1724 		mp->mnt_susp_owner = NULL;
1725 		wakeup(&mp->mnt_writeopcount);
1726 		wakeup(&mp->mnt_flag);
1727 		curthread->td_pflags &= ~TDP_IGNSUSP;
1728 		if ((flags & VR_START_WRITE) != 0) {
1729 			MNT_REF(mp);
1730 			mp->mnt_writeopcount++;
1731 		}
1732 		MNT_IUNLOCK(mp);
1733 		if ((flags & VR_NO_SUSPCLR) == 0)
1734 			VFS_SUSP_CLEAN(mp);
1735 	} else if ((flags & VR_START_WRITE) != 0) {
1736 		MNT_REF(mp);
1737 		vn_start_write_locked(mp, 0);
1738 	} else {
1739 		MNT_IUNLOCK(mp);
1740 	}
1741 }
1742 
1743 /*
1744  * Implement kqueues for files by translating it to vnode operation.
1745  */
1746 static int
1747 vn_kqfilter(struct file *fp, struct knote *kn)
1748 {
1749 
1750 	return (VOP_KQFILTER(fp->f_vnode, kn));
1751 }
1752 
1753 /*
1754  * Simplified in-kernel wrapper calls for extended attribute access.
1755  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1756  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1757  */
1758 int
1759 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1760     const char *attrname, int *buflen, char *buf, struct thread *td)
1761 {
1762 	struct uio	auio;
1763 	struct iovec	iov;
1764 	int	error;
1765 
1766 	iov.iov_len = *buflen;
1767 	iov.iov_base = buf;
1768 
1769 	auio.uio_iov = &iov;
1770 	auio.uio_iovcnt = 1;
1771 	auio.uio_rw = UIO_READ;
1772 	auio.uio_segflg = UIO_SYSSPACE;
1773 	auio.uio_td = td;
1774 	auio.uio_offset = 0;
1775 	auio.uio_resid = *buflen;
1776 
1777 	if ((ioflg & IO_NODELOCKED) == 0)
1778 		vn_lock(vp, LK_SHARED | LK_RETRY);
1779 
1780 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1781 
1782 	/* authorize attribute retrieval as kernel */
1783 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1784 	    td);
1785 
1786 	if ((ioflg & IO_NODELOCKED) == 0)
1787 		VOP_UNLOCK(vp, 0);
1788 
1789 	if (error == 0) {
1790 		*buflen = *buflen - auio.uio_resid;
1791 	}
1792 
1793 	return (error);
1794 }
1795 
1796 /*
1797  * XXX failure mode if partially written?
1798  */
1799 int
1800 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1801     const char *attrname, int buflen, char *buf, struct thread *td)
1802 {
1803 	struct uio	auio;
1804 	struct iovec	iov;
1805 	struct mount	*mp;
1806 	int	error;
1807 
1808 	iov.iov_len = buflen;
1809 	iov.iov_base = buf;
1810 
1811 	auio.uio_iov = &iov;
1812 	auio.uio_iovcnt = 1;
1813 	auio.uio_rw = UIO_WRITE;
1814 	auio.uio_segflg = UIO_SYSSPACE;
1815 	auio.uio_td = td;
1816 	auio.uio_offset = 0;
1817 	auio.uio_resid = buflen;
1818 
1819 	if ((ioflg & IO_NODELOCKED) == 0) {
1820 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1821 			return (error);
1822 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1823 	}
1824 
1825 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1826 
1827 	/* authorize attribute setting as kernel */
1828 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1829 
1830 	if ((ioflg & IO_NODELOCKED) == 0) {
1831 		vn_finished_write(mp);
1832 		VOP_UNLOCK(vp, 0);
1833 	}
1834 
1835 	return (error);
1836 }
1837 
1838 int
1839 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1840     const char *attrname, struct thread *td)
1841 {
1842 	struct mount	*mp;
1843 	int	error;
1844 
1845 	if ((ioflg & IO_NODELOCKED) == 0) {
1846 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1847 			return (error);
1848 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1849 	}
1850 
1851 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1852 
1853 	/* authorize attribute removal as kernel */
1854 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1855 	if (error == EOPNOTSUPP)
1856 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1857 		    NULL, td);
1858 
1859 	if ((ioflg & IO_NODELOCKED) == 0) {
1860 		vn_finished_write(mp);
1861 		VOP_UNLOCK(vp, 0);
1862 	}
1863 
1864 	return (error);
1865 }
1866 
1867 int
1868 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1869 {
1870 	struct mount *mp;
1871 	int ltype, error;
1872 
1873 	mp = vp->v_mount;
1874 	ltype = VOP_ISLOCKED(vp);
1875 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1876 	    ("vn_vget_ino: vp not locked"));
1877 	error = vfs_busy(mp, MBF_NOWAIT);
1878 	if (error != 0) {
1879 		vfs_ref(mp);
1880 		VOP_UNLOCK(vp, 0);
1881 		error = vfs_busy(mp, 0);
1882 		vn_lock(vp, ltype | LK_RETRY);
1883 		vfs_rel(mp);
1884 		if (error != 0)
1885 			return (ENOENT);
1886 		if (vp->v_iflag & VI_DOOMED) {
1887 			vfs_unbusy(mp);
1888 			return (ENOENT);
1889 		}
1890 	}
1891 	VOP_UNLOCK(vp, 0);
1892 	error = VFS_VGET(mp, ino, lkflags, rvp);
1893 	vfs_unbusy(mp);
1894 	vn_lock(vp, ltype | LK_RETRY);
1895 	if (vp->v_iflag & VI_DOOMED) {
1896 		if (error == 0)
1897 			vput(*rvp);
1898 		error = ENOENT;
1899 	}
1900 	return (error);
1901 }
1902 
1903 int
1904 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1905     const struct thread *td)
1906 {
1907 
1908 	if (vp->v_type != VREG || td == NULL)
1909 		return (0);
1910 	PROC_LOCK(td->td_proc);
1911 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1912 	    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1913 		kern_psignal(td->td_proc, SIGXFSZ);
1914 		PROC_UNLOCK(td->td_proc);
1915 		return (EFBIG);
1916 	}
1917 	PROC_UNLOCK(td->td_proc);
1918 	return (0);
1919 }
1920 
1921 int
1922 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1923     struct thread *td)
1924 {
1925 	struct vnode *vp;
1926 
1927 	vp = fp->f_vnode;
1928 #ifdef AUDIT
1929 	vn_lock(vp, LK_SHARED | LK_RETRY);
1930 	AUDIT_ARG_VNODE1(vp);
1931 	VOP_UNLOCK(vp, 0);
1932 #endif
1933 	return (setfmode(td, active_cred, vp, mode));
1934 }
1935 
1936 int
1937 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1938     struct thread *td)
1939 {
1940 	struct vnode *vp;
1941 
1942 	vp = fp->f_vnode;
1943 #ifdef AUDIT
1944 	vn_lock(vp, LK_SHARED | LK_RETRY);
1945 	AUDIT_ARG_VNODE1(vp);
1946 	VOP_UNLOCK(vp, 0);
1947 #endif
1948 	return (setfown(td, active_cred, vp, uid, gid));
1949 }
1950 
1951 void
1952 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1953 {
1954 	vm_object_t object;
1955 
1956 	if ((object = vp->v_object) == NULL)
1957 		return;
1958 	VM_OBJECT_WLOCK(object);
1959 	vm_object_page_remove(object, start, end, 0);
1960 	VM_OBJECT_WUNLOCK(object);
1961 }
1962 
1963 int
1964 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1965 {
1966 	struct vattr va;
1967 	daddr_t bn, bnp;
1968 	uint64_t bsize;
1969 	off_t noff;
1970 	int error;
1971 
1972 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1973 	    ("Wrong command %lu", cmd));
1974 
1975 	if (vn_lock(vp, LK_SHARED) != 0)
1976 		return (EBADF);
1977 	if (vp->v_type != VREG) {
1978 		error = ENOTTY;
1979 		goto unlock;
1980 	}
1981 	error = VOP_GETATTR(vp, &va, cred);
1982 	if (error != 0)
1983 		goto unlock;
1984 	noff = *off;
1985 	if (noff >= va.va_size) {
1986 		error = ENXIO;
1987 		goto unlock;
1988 	}
1989 	bsize = vp->v_mount->mnt_stat.f_iosize;
1990 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1991 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1992 		if (error == EOPNOTSUPP) {
1993 			error = ENOTTY;
1994 			goto unlock;
1995 		}
1996 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1997 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
1998 			noff = bn * bsize;
1999 			if (noff < *off)
2000 				noff = *off;
2001 			goto unlock;
2002 		}
2003 	}
2004 	if (noff > va.va_size)
2005 		noff = va.va_size;
2006 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2007 	if (cmd == FIOSEEKDATA)
2008 		error = ENXIO;
2009 unlock:
2010 	VOP_UNLOCK(vp, 0);
2011 	if (error == 0)
2012 		*off = noff;
2013 	return (error);
2014 }
2015 
2016 int
2017 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2018 {
2019 	struct ucred *cred;
2020 	struct vnode *vp;
2021 	struct vattr vattr;
2022 	off_t foffset, size;
2023 	int error, noneg;
2024 
2025 	cred = td->td_ucred;
2026 	vp = fp->f_vnode;
2027 	foffset = foffset_lock(fp, 0);
2028 	noneg = (vp->v_type != VCHR);
2029 	error = 0;
2030 	switch (whence) {
2031 	case L_INCR:
2032 		if (noneg &&
2033 		    (foffset < 0 ||
2034 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2035 			error = EOVERFLOW;
2036 			break;
2037 		}
2038 		offset += foffset;
2039 		break;
2040 	case L_XTND:
2041 		vn_lock(vp, LK_SHARED | LK_RETRY);
2042 		error = VOP_GETATTR(vp, &vattr, cred);
2043 		VOP_UNLOCK(vp, 0);
2044 		if (error)
2045 			break;
2046 
2047 		/*
2048 		 * If the file references a disk device, then fetch
2049 		 * the media size and use that to determine the ending
2050 		 * offset.
2051 		 */
2052 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2053 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2054 			vattr.va_size = size;
2055 		if (noneg &&
2056 		    (vattr.va_size > OFF_MAX ||
2057 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2058 			error = EOVERFLOW;
2059 			break;
2060 		}
2061 		offset += vattr.va_size;
2062 		break;
2063 	case L_SET:
2064 		break;
2065 	case SEEK_DATA:
2066 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2067 		break;
2068 	case SEEK_HOLE:
2069 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2070 		break;
2071 	default:
2072 		error = EINVAL;
2073 	}
2074 	if (error == 0 && noneg && offset < 0)
2075 		error = EINVAL;
2076 	if (error != 0)
2077 		goto drop;
2078 	VFS_KNOTE_UNLOCKED(vp, 0);
2079 	*(off_t *)(td->td_retval) = offset;
2080 drop:
2081 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2082 	return (error);
2083 }
2084