xref: /freebsd/sys/kern/vfs_vnops.c (revision 13de33a5dc2304b13d595d75d48c51793958474f)
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 (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
271 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
272 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
273 		return (error);
274 
275 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
276 		KASSERT(fp != NULL, ("open with flock requires fp"));
277 		lock_flags = VOP_ISLOCKED(vp);
278 		VOP_UNLOCK(vp, 0);
279 		lf.l_whence = SEEK_SET;
280 		lf.l_start = 0;
281 		lf.l_len = 0;
282 		if (fmode & O_EXLOCK)
283 			lf.l_type = F_WRLCK;
284 		else
285 			lf.l_type = F_RDLCK;
286 		type = F_FLOCK;
287 		if ((fmode & FNONBLOCK) == 0)
288 			type |= F_WAIT;
289 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
290 		have_flock = (error == 0);
291 		vn_lock(vp, lock_flags | LK_RETRY);
292 		if (error == 0 && vp->v_iflag & VI_DOOMED)
293 			error = ENOENT;
294 		/*
295 		 * Another thread might have used this vnode as an
296 		 * executable while the vnode lock was dropped.
297 		 * Ensure the vnode is still able to be opened for
298 		 * writing after the lock has been obtained.
299 		 */
300 		if (error == 0 && accmode & VWRITE)
301 			error = vn_writechk(vp);
302 		if (error) {
303 			VOP_UNLOCK(vp, 0);
304 			if (have_flock) {
305 				lf.l_whence = SEEK_SET;
306 				lf.l_start = 0;
307 				lf.l_len = 0;
308 				lf.l_type = F_UNLCK;
309 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
310 				    F_FLOCK);
311 			}
312 			vn_start_write(vp, &mp, V_WAIT);
313 			vn_lock(vp, lock_flags | LK_RETRY);
314 			(void)VOP_CLOSE(vp, fmode, cred, td);
315 			vn_finished_write(mp);
316 			return (error);
317 		}
318 		fp->f_flag |= FHASLOCK;
319 	}
320 	if (fmode & FWRITE) {
321 		VOP_ADD_WRITECOUNT(vp, 1);
322 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
323 		    __func__, vp, vp->v_writecount);
324 	}
325 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
326 	return (0);
327 }
328 
329 /*
330  * Check for write permissions on the specified vnode.
331  * Prototype text segments cannot be written.
332  */
333 int
334 vn_writechk(vp)
335 	register struct vnode *vp;
336 {
337 
338 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
339 	/*
340 	 * If there's shared text associated with
341 	 * the vnode, try to free it up once.  If
342 	 * we fail, we can't allow writing.
343 	 */
344 	if (VOP_IS_TEXT(vp))
345 		return (ETXTBSY);
346 
347 	return (0);
348 }
349 
350 /*
351  * Vnode close call
352  */
353 int
354 vn_close(vp, flags, file_cred, td)
355 	register struct vnode *vp;
356 	int flags;
357 	struct ucred *file_cred;
358 	struct thread *td;
359 {
360 	struct mount *mp;
361 	int error, lock_flags;
362 
363 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
364 	    MNT_EXTENDED_SHARED(vp->v_mount))
365 		lock_flags = LK_SHARED;
366 	else
367 		lock_flags = LK_EXCLUSIVE;
368 
369 	vn_start_write(vp, &mp, V_WAIT);
370 	vn_lock(vp, lock_flags | LK_RETRY);
371 	if (flags & FWRITE) {
372 		VNASSERT(vp->v_writecount > 0, vp,
373 		    ("vn_close: negative writecount"));
374 		VOP_ADD_WRITECOUNT(vp, -1);
375 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
376 		    __func__, vp, vp->v_writecount);
377 	}
378 	error = VOP_CLOSE(vp, flags, file_cred, td);
379 	vput(vp);
380 	vn_finished_write(mp);
381 	return (error);
382 }
383 
384 /*
385  * Heuristic to detect sequential operation.
386  */
387 static int
388 sequential_heuristic(struct uio *uio, struct file *fp)
389 {
390 
391 	if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
392 		return (fp->f_seqcount << IO_SEQSHIFT);
393 
394 	/*
395 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
396 	 * that the first I/O is normally considered to be slightly
397 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
398 	 * unless previous seeks have reduced f_seqcount to 0, in which
399 	 * case offset 0 is not special.
400 	 */
401 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
402 	    uio->uio_offset == fp->f_nextoff) {
403 		/*
404 		 * f_seqcount is in units of fixed-size blocks so that it
405 		 * depends mainly on the amount of sequential I/O and not
406 		 * much on the number of sequential I/O's.  The fixed size
407 		 * of 16384 is hard-coded here since it is (not quite) just
408 		 * a magic size that works well here.  This size is more
409 		 * closely related to the best I/O size for real disks than
410 		 * to any block size used by software.
411 		 */
412 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
413 		if (fp->f_seqcount > IO_SEQMAX)
414 			fp->f_seqcount = IO_SEQMAX;
415 		return (fp->f_seqcount << IO_SEQSHIFT);
416 	}
417 
418 	/* Not sequential.  Quickly draw-down sequentiality. */
419 	if (fp->f_seqcount > 1)
420 		fp->f_seqcount = 1;
421 	else
422 		fp->f_seqcount = 0;
423 	return (0);
424 }
425 
426 /*
427  * Package up an I/O request on a vnode into a uio and do it.
428  */
429 int
430 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
431     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
432     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
433 {
434 	struct uio auio;
435 	struct iovec aiov;
436 	struct mount *mp;
437 	struct ucred *cred;
438 	void *rl_cookie;
439 	int error, lock_flags;
440 
441 	auio.uio_iov = &aiov;
442 	auio.uio_iovcnt = 1;
443 	aiov.iov_base = base;
444 	aiov.iov_len = len;
445 	auio.uio_resid = len;
446 	auio.uio_offset = offset;
447 	auio.uio_segflg = segflg;
448 	auio.uio_rw = rw;
449 	auio.uio_td = td;
450 	error = 0;
451 
452 	if ((ioflg & IO_NODELOCKED) == 0) {
453 		if (rw == UIO_READ) {
454 			rl_cookie = vn_rangelock_rlock(vp, offset,
455 			    offset + len);
456 		} else {
457 			rl_cookie = vn_rangelock_wlock(vp, offset,
458 			    offset + len);
459 		}
460 		mp = NULL;
461 		if (rw == UIO_WRITE) {
462 			if (vp->v_type != VCHR &&
463 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
464 			    != 0)
465 				goto out;
466 			if (MNT_SHARED_WRITES(mp) ||
467 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
468 				lock_flags = LK_SHARED;
469 			else
470 				lock_flags = LK_EXCLUSIVE;
471 		} else
472 			lock_flags = LK_SHARED;
473 		vn_lock(vp, lock_flags | LK_RETRY);
474 	} else
475 		rl_cookie = NULL;
476 
477 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
478 #ifdef MAC
479 	if ((ioflg & IO_NOMACCHECK) == 0) {
480 		if (rw == UIO_READ)
481 			error = mac_vnode_check_read(active_cred, file_cred,
482 			    vp);
483 		else
484 			error = mac_vnode_check_write(active_cred, file_cred,
485 			    vp);
486 	}
487 #endif
488 	if (error == 0) {
489 		if (file_cred != NULL)
490 			cred = file_cred;
491 		else
492 			cred = active_cred;
493 		if (rw == UIO_READ)
494 			error = VOP_READ(vp, &auio, ioflg, cred);
495 		else
496 			error = VOP_WRITE(vp, &auio, ioflg, cred);
497 	}
498 	if (aresid)
499 		*aresid = auio.uio_resid;
500 	else
501 		if (auio.uio_resid && error == 0)
502 			error = EIO;
503 	if ((ioflg & IO_NODELOCKED) == 0) {
504 		VOP_UNLOCK(vp, 0);
505 		if (mp != NULL)
506 			vn_finished_write(mp);
507 	}
508  out:
509 	if (rl_cookie != NULL)
510 		vn_rangelock_unlock(vp, rl_cookie);
511 	return (error);
512 }
513 
514 /*
515  * Package up an I/O request on a vnode into a uio and do it.  The I/O
516  * request is split up into smaller chunks and we try to avoid saturating
517  * the buffer cache while potentially holding a vnode locked, so we
518  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
519  * to give other processes a chance to lock the vnode (either other processes
520  * core'ing the same binary, or unrelated processes scanning the directory).
521  */
522 int
523 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
524     file_cred, aresid, td)
525 	enum uio_rw rw;
526 	struct vnode *vp;
527 	void *base;
528 	size_t len;
529 	off_t offset;
530 	enum uio_seg segflg;
531 	int ioflg;
532 	struct ucred *active_cred;
533 	struct ucred *file_cred;
534 	size_t *aresid;
535 	struct thread *td;
536 {
537 	int error = 0;
538 	ssize_t iaresid;
539 
540 	do {
541 		int chunk;
542 
543 		/*
544 		 * Force `offset' to a multiple of MAXBSIZE except possibly
545 		 * for the first chunk, so that filesystems only need to
546 		 * write full blocks except possibly for the first and last
547 		 * chunks.
548 		 */
549 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
550 
551 		if (chunk > len)
552 			chunk = len;
553 		if (rw != UIO_READ && vp->v_type == VREG)
554 			bwillwrite();
555 		iaresid = 0;
556 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
557 		    ioflg, active_cred, file_cred, &iaresid, td);
558 		len -= chunk;	/* aresid calc already includes length */
559 		if (error)
560 			break;
561 		offset += chunk;
562 		base = (char *)base + chunk;
563 		kern_yield(PRI_USER);
564 	} while (len);
565 	if (aresid)
566 		*aresid = len + iaresid;
567 	return (error);
568 }
569 
570 off_t
571 foffset_lock(struct file *fp, int flags)
572 {
573 	struct mtx *mtxp;
574 	off_t res;
575 
576 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
577 
578 #if OFF_MAX <= LONG_MAX
579 	/*
580 	 * Caller only wants the current f_offset value.  Assume that
581 	 * the long and shorter integer types reads are atomic.
582 	 */
583 	if ((flags & FOF_NOLOCK) != 0)
584 		return (fp->f_offset);
585 #endif
586 
587 	/*
588 	 * According to McKusick the vn lock was protecting f_offset here.
589 	 * It is now protected by the FOFFSET_LOCKED flag.
590 	 */
591 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
592 	mtx_lock(mtxp);
593 	if ((flags & FOF_NOLOCK) == 0) {
594 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
595 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
596 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
597 			    "vofflock", 0);
598 		}
599 		fp->f_vnread_flags |= FOFFSET_LOCKED;
600 	}
601 	res = fp->f_offset;
602 	mtx_unlock(mtxp);
603 	return (res);
604 }
605 
606 void
607 foffset_unlock(struct file *fp, off_t val, int flags)
608 {
609 	struct mtx *mtxp;
610 
611 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
612 
613 #if OFF_MAX <= LONG_MAX
614 	if ((flags & FOF_NOLOCK) != 0) {
615 		if ((flags & FOF_NOUPDATE) == 0)
616 			fp->f_offset = val;
617 		if ((flags & FOF_NEXTOFF) != 0)
618 			fp->f_nextoff = val;
619 		return;
620 	}
621 #endif
622 
623 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
624 	mtx_lock(mtxp);
625 	if ((flags & FOF_NOUPDATE) == 0)
626 		fp->f_offset = val;
627 	if ((flags & FOF_NEXTOFF) != 0)
628 		fp->f_nextoff = val;
629 	if ((flags & FOF_NOLOCK) == 0) {
630 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
631 		    ("Lost FOFFSET_LOCKED"));
632 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
633 			wakeup(&fp->f_vnread_flags);
634 		fp->f_vnread_flags = 0;
635 	}
636 	mtx_unlock(mtxp);
637 }
638 
639 void
640 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
641 {
642 
643 	if ((flags & FOF_OFFSET) == 0)
644 		uio->uio_offset = foffset_lock(fp, flags);
645 }
646 
647 void
648 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
649 {
650 
651 	if ((flags & FOF_OFFSET) == 0)
652 		foffset_unlock(fp, uio->uio_offset, flags);
653 }
654 
655 static int
656 get_advice(struct file *fp, struct uio *uio)
657 {
658 	struct mtx *mtxp;
659 	int ret;
660 
661 	ret = POSIX_FADV_NORMAL;
662 	if (fp->f_advice == NULL)
663 		return (ret);
664 
665 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
666 	mtx_lock(mtxp);
667 	if (uio->uio_offset >= fp->f_advice->fa_start &&
668 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
669 		ret = fp->f_advice->fa_advice;
670 	mtx_unlock(mtxp);
671 	return (ret);
672 }
673 
674 /*
675  * File table vnode read routine.
676  */
677 static int
678 vn_read(fp, uio, active_cred, flags, td)
679 	struct file *fp;
680 	struct uio *uio;
681 	struct ucred *active_cred;
682 	int flags;
683 	struct thread *td;
684 {
685 	struct vnode *vp;
686 	struct mtx *mtxp;
687 	int error, ioflag;
688 	int advice;
689 	off_t offset, start, end;
690 
691 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
692 	    uio->uio_td, td));
693 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
694 	vp = fp->f_vnode;
695 	ioflag = 0;
696 	if (fp->f_flag & FNONBLOCK)
697 		ioflag |= IO_NDELAY;
698 	if (fp->f_flag & O_DIRECT)
699 		ioflag |= IO_DIRECT;
700 	advice = get_advice(fp, uio);
701 	vn_lock(vp, LK_SHARED | LK_RETRY);
702 
703 	switch (advice) {
704 	case POSIX_FADV_NORMAL:
705 	case POSIX_FADV_SEQUENTIAL:
706 	case POSIX_FADV_NOREUSE:
707 		ioflag |= sequential_heuristic(uio, fp);
708 		break;
709 	case POSIX_FADV_RANDOM:
710 		/* Disable read-ahead for random I/O. */
711 		break;
712 	}
713 	offset = uio->uio_offset;
714 
715 #ifdef MAC
716 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
717 	if (error == 0)
718 #endif
719 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
720 	fp->f_nextoff = uio->uio_offset;
721 	VOP_UNLOCK(vp, 0);
722 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
723 	    offset != uio->uio_offset) {
724 		/*
725 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
726 		 * buffers for the backing file after a
727 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
728 		 * case of using POSIX_FADV_NOREUSE with sequential
729 		 * access, track the previous implicit DONTNEED
730 		 * request and grow this request to include the
731 		 * current read(2) in addition to the previous
732 		 * DONTNEED.  With purely sequential access this will
733 		 * cause the DONTNEED requests to continously grow to
734 		 * cover all of the previously read regions of the
735 		 * file.  This allows filesystem blocks that are
736 		 * accessed by multiple calls to read(2) to be flushed
737 		 * once the last read(2) finishes.
738 		 */
739 		start = offset;
740 		end = uio->uio_offset - 1;
741 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
742 		mtx_lock(mtxp);
743 		if (fp->f_advice != NULL &&
744 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
745 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
746 				start = fp->f_advice->fa_prevstart;
747 			else if (fp->f_advice->fa_prevstart != 0 &&
748 			    fp->f_advice->fa_prevstart == end + 1)
749 				end = fp->f_advice->fa_prevend;
750 			fp->f_advice->fa_prevstart = start;
751 			fp->f_advice->fa_prevend = end;
752 		}
753 		mtx_unlock(mtxp);
754 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
755 	}
756 	return (error);
757 }
758 
759 /*
760  * File table vnode write routine.
761  */
762 static int
763 vn_write(fp, uio, active_cred, flags, td)
764 	struct file *fp;
765 	struct uio *uio;
766 	struct ucred *active_cred;
767 	int flags;
768 	struct thread *td;
769 {
770 	struct vnode *vp;
771 	struct mount *mp;
772 	struct mtx *mtxp;
773 	int error, ioflag, lock_flags;
774 	int advice;
775 	off_t offset, start, end;
776 
777 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
778 	    uio->uio_td, td));
779 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
780 	vp = fp->f_vnode;
781 	if (vp->v_type == VREG)
782 		bwillwrite();
783 	ioflag = IO_UNIT;
784 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
785 		ioflag |= IO_APPEND;
786 	if (fp->f_flag & FNONBLOCK)
787 		ioflag |= IO_NDELAY;
788 	if (fp->f_flag & O_DIRECT)
789 		ioflag |= IO_DIRECT;
790 	if ((fp->f_flag & O_FSYNC) ||
791 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
792 		ioflag |= IO_SYNC;
793 	mp = NULL;
794 	if (vp->v_type != VCHR &&
795 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
796 		goto unlock;
797 
798 	advice = get_advice(fp, uio);
799 
800 	if (MNT_SHARED_WRITES(mp) ||
801 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
802 		lock_flags = LK_SHARED;
803 	} else {
804 		lock_flags = LK_EXCLUSIVE;
805 	}
806 
807 	vn_lock(vp, lock_flags | LK_RETRY);
808 	switch (advice) {
809 	case POSIX_FADV_NORMAL:
810 	case POSIX_FADV_SEQUENTIAL:
811 	case POSIX_FADV_NOREUSE:
812 		ioflag |= sequential_heuristic(uio, fp);
813 		break;
814 	case POSIX_FADV_RANDOM:
815 		/* XXX: Is this correct? */
816 		break;
817 	}
818 	offset = uio->uio_offset;
819 
820 #ifdef MAC
821 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
822 	if (error == 0)
823 #endif
824 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
825 	fp->f_nextoff = uio->uio_offset;
826 	VOP_UNLOCK(vp, 0);
827 	if (vp->v_type != VCHR)
828 		vn_finished_write(mp);
829 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
830 	    offset != uio->uio_offset) {
831 		/*
832 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
833 		 * buffers for the backing file after a
834 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
835 		 * common case of using POSIX_FADV_NOREUSE with
836 		 * sequential access, track the previous implicit
837 		 * DONTNEED request and grow this request to include
838 		 * the current write(2) in addition to the previous
839 		 * DONTNEED.  With purely sequential access this will
840 		 * cause the DONTNEED requests to continously grow to
841 		 * cover all of the previously written regions of the
842 		 * file.
843 		 *
844 		 * Note that the blocks just written are almost
845 		 * certainly still dirty, so this only works when
846 		 * VOP_ADVISE() calls from subsequent writes push out
847 		 * the data written by this write(2) once the backing
848 		 * buffers are clean.  However, as compared to forcing
849 		 * IO_DIRECT, this gives much saner behavior.  Write
850 		 * clustering is still allowed, and clean pages are
851 		 * merely moved to the cache page queue rather than
852 		 * outright thrown away.  This means a subsequent
853 		 * read(2) can still avoid hitting the disk if the
854 		 * pages have not been reclaimed.
855 		 *
856 		 * This does make POSIX_FADV_NOREUSE largely useless
857 		 * with non-sequential access.  However, sequential
858 		 * access is the more common use case and the flag is
859 		 * merely advisory.
860 		 */
861 		start = offset;
862 		end = uio->uio_offset - 1;
863 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
864 		mtx_lock(mtxp);
865 		if (fp->f_advice != NULL &&
866 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
867 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
868 				start = fp->f_advice->fa_prevstart;
869 			else if (fp->f_advice->fa_prevstart != 0 &&
870 			    fp->f_advice->fa_prevstart == end + 1)
871 				end = fp->f_advice->fa_prevend;
872 			fp->f_advice->fa_prevstart = start;
873 			fp->f_advice->fa_prevend = end;
874 		}
875 		mtx_unlock(mtxp);
876 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
877 	}
878 
879 unlock:
880 	return (error);
881 }
882 
883 static const int io_hold_cnt = 16;
884 static int vn_io_fault_enable = 1;
885 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
886     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
887 static u_long vn_io_faults_cnt;
888 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
889     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
890 
891 /*
892  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
893  * prevent the following deadlock:
894  *
895  * Assume that the thread A reads from the vnode vp1 into userspace
896  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
897  * currently not resident, then system ends up with the call chain
898  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
899  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
900  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
901  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
902  * backed by the pages of vnode vp1, and some page in buf2 is not
903  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
904  *
905  * To prevent the lock order reversal and deadlock, vn_io_fault() does
906  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
907  * Instead, it first tries to do the whole range i/o with pagefaults
908  * disabled. If all pages in the i/o buffer are resident and mapped,
909  * VOP will succeed (ignoring the genuine filesystem errors).
910  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
911  * i/o in chunks, with all pages in the chunk prefaulted and held
912  * using vm_fault_quick_hold_pages().
913  *
914  * Filesystems using this deadlock avoidance scheme should use the
915  * array of the held pages from uio, saved in the curthread->td_ma,
916  * instead of doing uiomove().  A helper function
917  * vn_io_fault_uiomove() converts uiomove request into
918  * uiomove_fromphys() over td_ma array.
919  *
920  * Since vnode locks do not cover the whole i/o anymore, rangelocks
921  * make the current i/o request atomic with respect to other i/os and
922  * truncations.
923  */
924 static int
925 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
926     int flags, struct thread *td)
927 {
928 	vm_page_t ma[io_hold_cnt + 2];
929 	struct uio *uio_clone, short_uio;
930 	struct iovec short_iovec[1];
931 	fo_rdwr_t *doio;
932 	struct vnode *vp;
933 	void *rl_cookie;
934 	struct mount *mp;
935 	vm_page_t *prev_td_ma;
936 	int error, cnt, save, saveheld, prev_td_ma_cnt;
937 	vm_offset_t addr, end;
938 	vm_prot_t prot;
939 	size_t len, resid;
940 	ssize_t adv;
941 
942 	if (uio->uio_rw == UIO_READ)
943 		doio = vn_read;
944 	else
945 		doio = vn_write;
946 	vp = fp->f_vnode;
947 	foffset_lock_uio(fp, uio, flags);
948 
949 	if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
950 	    ((mp = vp->v_mount) != NULL &&
951 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
952 	    !vn_io_fault_enable) {
953 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
954 		goto out_last;
955 	}
956 
957 	/*
958 	 * The UFS follows IO_UNIT directive and replays back both
959 	 * uio_offset and uio_resid if an error is encountered during the
960 	 * operation.  But, since the iovec may be already advanced,
961 	 * uio is still in an inconsistent state.
962 	 *
963 	 * Cache a copy of the original uio, which is advanced to the redo
964 	 * point using UIO_NOCOPY below.
965 	 */
966 	uio_clone = cloneuio(uio);
967 	resid = uio->uio_resid;
968 
969 	short_uio.uio_segflg = UIO_USERSPACE;
970 	short_uio.uio_rw = uio->uio_rw;
971 	short_uio.uio_td = uio->uio_td;
972 
973 	if (uio->uio_rw == UIO_READ) {
974 		prot = VM_PROT_WRITE;
975 		rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
976 		    uio->uio_offset + uio->uio_resid);
977 	} else {
978 		prot = VM_PROT_READ;
979 		if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
980 			/* For appenders, punt and lock the whole range. */
981 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
982 		else
983 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
984 			    uio->uio_offset + uio->uio_resid);
985 	}
986 
987 	save = vm_fault_disable_pagefaults();
988 	error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
989 	if (error != EFAULT)
990 		goto out;
991 
992 	atomic_add_long(&vn_io_faults_cnt, 1);
993 	uio_clone->uio_segflg = UIO_NOCOPY;
994 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
995 	uio_clone->uio_segflg = uio->uio_segflg;
996 
997 	saveheld = curthread_pflags_set(TDP_UIOHELD);
998 	prev_td_ma = td->td_ma;
999 	prev_td_ma_cnt = td->td_ma_cnt;
1000 
1001 	while (uio_clone->uio_resid != 0) {
1002 		len = uio_clone->uio_iov->iov_len;
1003 		if (len == 0) {
1004 			KASSERT(uio_clone->uio_iovcnt >= 1,
1005 			    ("iovcnt underflow"));
1006 			uio_clone->uio_iov++;
1007 			uio_clone->uio_iovcnt--;
1008 			continue;
1009 		}
1010 		if (len > io_hold_cnt * PAGE_SIZE)
1011 			len = io_hold_cnt * PAGE_SIZE;
1012 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1013 		end = round_page(addr + len);
1014 		if (end < addr) {
1015 			error = EFAULT;
1016 			break;
1017 		}
1018 		cnt = atop(end - trunc_page(addr));
1019 		/*
1020 		 * A perfectly misaligned address and length could cause
1021 		 * both the start and the end of the chunk to use partial
1022 		 * page.  +2 accounts for such a situation.
1023 		 */
1024 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1025 		    addr, len, prot, ma, io_hold_cnt + 2);
1026 		if (cnt == -1) {
1027 			error = EFAULT;
1028 			break;
1029 		}
1030 		short_uio.uio_iov = &short_iovec[0];
1031 		short_iovec[0].iov_base = (void *)addr;
1032 		short_uio.uio_iovcnt = 1;
1033 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1034 		short_uio.uio_offset = uio_clone->uio_offset;
1035 		td->td_ma = ma;
1036 		td->td_ma_cnt = cnt;
1037 
1038 		error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
1039 		    td);
1040 		vm_page_unhold_pages(ma, cnt);
1041 		adv = len - short_uio.uio_resid;
1042 
1043 		uio_clone->uio_iov->iov_base =
1044 		    (char *)uio_clone->uio_iov->iov_base + adv;
1045 		uio_clone->uio_iov->iov_len -= adv;
1046 		uio_clone->uio_resid -= adv;
1047 		uio_clone->uio_offset += adv;
1048 
1049 		uio->uio_resid -= adv;
1050 		uio->uio_offset += adv;
1051 
1052 		if (error != 0 || adv == 0)
1053 			break;
1054 	}
1055 	td->td_ma = prev_td_ma;
1056 	td->td_ma_cnt = prev_td_ma_cnt;
1057 	curthread_pflags_restore(saveheld);
1058 out:
1059 	vm_fault_enable_pagefaults(save);
1060 	vn_rangelock_unlock(vp, rl_cookie);
1061 	free(uio_clone, M_IOV);
1062 out_last:
1063 	foffset_unlock_uio(fp, uio, flags);
1064 	return (error);
1065 }
1066 
1067 /*
1068  * Helper function to perform the requested uiomove operation using
1069  * the held pages for io->uio_iov[0].iov_base buffer instead of
1070  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1071  * instead of iov_base prevents page faults that could occur due to
1072  * pmap_collect() invalidating the mapping created by
1073  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1074  * object cleanup revoking the write access from page mappings.
1075  *
1076  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1077  * instead of plain uiomove().
1078  */
1079 int
1080 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1081 {
1082 	struct uio transp_uio;
1083 	struct iovec transp_iov[1];
1084 	struct thread *td;
1085 	size_t adv;
1086 	int error, pgadv;
1087 
1088 	td = curthread;
1089 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1090 	    uio->uio_segflg != UIO_USERSPACE)
1091 		return (uiomove(data, xfersize, uio));
1092 
1093 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1094 	transp_iov[0].iov_base = data;
1095 	transp_uio.uio_iov = &transp_iov[0];
1096 	transp_uio.uio_iovcnt = 1;
1097 	if (xfersize > uio->uio_resid)
1098 		xfersize = uio->uio_resid;
1099 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1100 	transp_uio.uio_offset = 0;
1101 	transp_uio.uio_segflg = UIO_SYSSPACE;
1102 	/*
1103 	 * Since transp_iov points to data, and td_ma page array
1104 	 * corresponds to original uio->uio_iov, we need to invert the
1105 	 * direction of the i/o operation as passed to
1106 	 * uiomove_fromphys().
1107 	 */
1108 	switch (uio->uio_rw) {
1109 	case UIO_WRITE:
1110 		transp_uio.uio_rw = UIO_READ;
1111 		break;
1112 	case UIO_READ:
1113 		transp_uio.uio_rw = UIO_WRITE;
1114 		break;
1115 	}
1116 	transp_uio.uio_td = uio->uio_td;
1117 	error = uiomove_fromphys(td->td_ma,
1118 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1119 	    xfersize, &transp_uio);
1120 	adv = xfersize - transp_uio.uio_resid;
1121 	pgadv =
1122 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1123 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1124 	td->td_ma += pgadv;
1125 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1126 	    pgadv));
1127 	td->td_ma_cnt -= pgadv;
1128 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1129 	uio->uio_iov->iov_len -= adv;
1130 	uio->uio_resid -= adv;
1131 	uio->uio_offset += adv;
1132 	return (error);
1133 }
1134 
1135 int
1136 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1137     struct uio *uio)
1138 {
1139 	struct thread *td;
1140 	vm_offset_t iov_base;
1141 	int cnt, pgadv;
1142 
1143 	td = curthread;
1144 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1145 	    uio->uio_segflg != UIO_USERSPACE)
1146 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1147 
1148 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1149 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1150 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1151 	switch (uio->uio_rw) {
1152 	case UIO_WRITE:
1153 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1154 		    offset, cnt);
1155 		break;
1156 	case UIO_READ:
1157 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1158 		    cnt);
1159 		break;
1160 	}
1161 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1162 	td->td_ma += pgadv;
1163 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1164 	    pgadv));
1165 	td->td_ma_cnt -= pgadv;
1166 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1167 	uio->uio_iov->iov_len -= cnt;
1168 	uio->uio_resid -= cnt;
1169 	uio->uio_offset += cnt;
1170 	return (0);
1171 }
1172 
1173 
1174 /*
1175  * File table truncate routine.
1176  */
1177 static int
1178 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1179     struct thread *td)
1180 {
1181 	struct vattr vattr;
1182 	struct mount *mp;
1183 	struct vnode *vp;
1184 	void *rl_cookie;
1185 	int error;
1186 
1187 	vp = fp->f_vnode;
1188 
1189 	/*
1190 	 * Lock the whole range for truncation.  Otherwise split i/o
1191 	 * might happen partly before and partly after the truncation.
1192 	 */
1193 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1194 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1195 	if (error)
1196 		goto out1;
1197 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1198 	if (vp->v_type == VDIR) {
1199 		error = EISDIR;
1200 		goto out;
1201 	}
1202 #ifdef MAC
1203 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1204 	if (error)
1205 		goto out;
1206 #endif
1207 	error = vn_writechk(vp);
1208 	if (error == 0) {
1209 		VATTR_NULL(&vattr);
1210 		vattr.va_size = length;
1211 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1212 	}
1213 out:
1214 	VOP_UNLOCK(vp, 0);
1215 	vn_finished_write(mp);
1216 out1:
1217 	vn_rangelock_unlock(vp, rl_cookie);
1218 	return (error);
1219 }
1220 
1221 /*
1222  * File table vnode stat routine.
1223  */
1224 static int
1225 vn_statfile(fp, sb, active_cred, td)
1226 	struct file *fp;
1227 	struct stat *sb;
1228 	struct ucred *active_cred;
1229 	struct thread *td;
1230 {
1231 	struct vnode *vp = fp->f_vnode;
1232 	int error;
1233 
1234 	vn_lock(vp, LK_SHARED | LK_RETRY);
1235 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1236 	VOP_UNLOCK(vp, 0);
1237 
1238 	return (error);
1239 }
1240 
1241 /*
1242  * Stat a vnode; implementation for the stat syscall
1243  */
1244 int
1245 vn_stat(vp, sb, active_cred, file_cred, td)
1246 	struct vnode *vp;
1247 	register struct stat *sb;
1248 	struct ucred *active_cred;
1249 	struct ucred *file_cred;
1250 	struct thread *td;
1251 {
1252 	struct vattr vattr;
1253 	register struct vattr *vap;
1254 	int error;
1255 	u_short mode;
1256 
1257 #ifdef MAC
1258 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1259 	if (error)
1260 		return (error);
1261 #endif
1262 
1263 	vap = &vattr;
1264 
1265 	/*
1266 	 * Initialize defaults for new and unusual fields, so that file
1267 	 * systems which don't support these fields don't need to know
1268 	 * about them.
1269 	 */
1270 	vap->va_birthtime.tv_sec = -1;
1271 	vap->va_birthtime.tv_nsec = 0;
1272 	vap->va_fsid = VNOVAL;
1273 	vap->va_rdev = NODEV;
1274 
1275 	error = VOP_GETATTR(vp, vap, active_cred);
1276 	if (error)
1277 		return (error);
1278 
1279 	/*
1280 	 * Zero the spare stat fields
1281 	 */
1282 	bzero(sb, sizeof *sb);
1283 
1284 	/*
1285 	 * Copy from vattr table
1286 	 */
1287 	if (vap->va_fsid != VNOVAL)
1288 		sb->st_dev = vap->va_fsid;
1289 	else
1290 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1291 	sb->st_ino = vap->va_fileid;
1292 	mode = vap->va_mode;
1293 	switch (vap->va_type) {
1294 	case VREG:
1295 		mode |= S_IFREG;
1296 		break;
1297 	case VDIR:
1298 		mode |= S_IFDIR;
1299 		break;
1300 	case VBLK:
1301 		mode |= S_IFBLK;
1302 		break;
1303 	case VCHR:
1304 		mode |= S_IFCHR;
1305 		break;
1306 	case VLNK:
1307 		mode |= S_IFLNK;
1308 		break;
1309 	case VSOCK:
1310 		mode |= S_IFSOCK;
1311 		break;
1312 	case VFIFO:
1313 		mode |= S_IFIFO;
1314 		break;
1315 	default:
1316 		return (EBADF);
1317 	};
1318 	sb->st_mode = mode;
1319 	sb->st_nlink = vap->va_nlink;
1320 	sb->st_uid = vap->va_uid;
1321 	sb->st_gid = vap->va_gid;
1322 	sb->st_rdev = vap->va_rdev;
1323 	if (vap->va_size > OFF_MAX)
1324 		return (EOVERFLOW);
1325 	sb->st_size = vap->va_size;
1326 	sb->st_atim = vap->va_atime;
1327 	sb->st_mtim = vap->va_mtime;
1328 	sb->st_ctim = vap->va_ctime;
1329 	sb->st_birthtim = vap->va_birthtime;
1330 
1331         /*
1332 	 * According to www.opengroup.org, the meaning of st_blksize is
1333 	 *   "a filesystem-specific preferred I/O block size for this
1334 	 *    object.  In some filesystem types, this may vary from file
1335 	 *    to file"
1336 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1337 	 */
1338 
1339 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1340 
1341 	sb->st_flags = vap->va_flags;
1342 	if (priv_check(td, PRIV_VFS_GENERATION))
1343 		sb->st_gen = 0;
1344 	else
1345 		sb->st_gen = vap->va_gen;
1346 
1347 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1348 	return (0);
1349 }
1350 
1351 /*
1352  * File table vnode ioctl routine.
1353  */
1354 static int
1355 vn_ioctl(fp, com, data, active_cred, td)
1356 	struct file *fp;
1357 	u_long com;
1358 	void *data;
1359 	struct ucred *active_cred;
1360 	struct thread *td;
1361 {
1362 	struct vattr vattr;
1363 	struct vnode *vp;
1364 	int error;
1365 
1366 	vp = fp->f_vnode;
1367 	switch (vp->v_type) {
1368 	case VDIR:
1369 	case VREG:
1370 		switch (com) {
1371 		case FIONREAD:
1372 			vn_lock(vp, LK_SHARED | LK_RETRY);
1373 			error = VOP_GETATTR(vp, &vattr, active_cred);
1374 			VOP_UNLOCK(vp, 0);
1375 			if (error == 0)
1376 				*(int *)data = vattr.va_size - fp->f_offset;
1377 			return (error);
1378 		case FIONBIO:
1379 		case FIOASYNC:
1380 			return (0);
1381 		default:
1382 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1383 			    active_cred, td));
1384 		}
1385 	default:
1386 		return (ENOTTY);
1387 	}
1388 }
1389 
1390 /*
1391  * File table vnode poll routine.
1392  */
1393 static int
1394 vn_poll(fp, events, active_cred, td)
1395 	struct file *fp;
1396 	int events;
1397 	struct ucred *active_cred;
1398 	struct thread *td;
1399 {
1400 	struct vnode *vp;
1401 	int error;
1402 
1403 	vp = fp->f_vnode;
1404 #ifdef MAC
1405 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1406 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1407 	VOP_UNLOCK(vp, 0);
1408 	if (!error)
1409 #endif
1410 
1411 	error = VOP_POLL(vp, events, fp->f_cred, td);
1412 	return (error);
1413 }
1414 
1415 /*
1416  * Acquire the requested lock and then check for validity.  LK_RETRY
1417  * permits vn_lock to return doomed vnodes.
1418  */
1419 int
1420 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1421 {
1422 	int error;
1423 
1424 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1425 	    ("vn_lock called with no locktype."));
1426 	do {
1427 #ifdef DEBUG_VFS_LOCKS
1428 		KASSERT(vp->v_holdcnt != 0,
1429 		    ("vn_lock %p: zero hold count", vp));
1430 #endif
1431 		error = VOP_LOCK1(vp, flags, file, line);
1432 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1433 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1434 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1435 		    flags, error));
1436 		/*
1437 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1438 		 * If RETRY is not set, we return ENOENT instead.
1439 		 */
1440 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1441 		    (flags & LK_RETRY) == 0) {
1442 			VOP_UNLOCK(vp, 0);
1443 			error = ENOENT;
1444 			break;
1445 		}
1446 	} while (flags & LK_RETRY && error != 0);
1447 	return (error);
1448 }
1449 
1450 /*
1451  * File table vnode close routine.
1452  */
1453 static int
1454 vn_closefile(fp, td)
1455 	struct file *fp;
1456 	struct thread *td;
1457 {
1458 	struct vnode *vp;
1459 	struct flock lf;
1460 	int error;
1461 
1462 	vp = fp->f_vnode;
1463 	fp->f_ops = &badfileops;
1464 
1465 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1466 		vref(vp);
1467 
1468 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1469 
1470 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1471 		lf.l_whence = SEEK_SET;
1472 		lf.l_start = 0;
1473 		lf.l_len = 0;
1474 		lf.l_type = F_UNLCK;
1475 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1476 		vrele(vp);
1477 	}
1478 	return (error);
1479 }
1480 
1481 /*
1482  * Preparing to start a filesystem write operation. If the operation is
1483  * permitted, then we bump the count of operations in progress and
1484  * proceed. If a suspend request is in progress, we wait until the
1485  * suspension is over, and then proceed.
1486  */
1487 static int
1488 vn_start_write_locked(struct mount *mp, int flags)
1489 {
1490 	int error;
1491 
1492 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1493 	error = 0;
1494 
1495 	/*
1496 	 * Check on status of suspension.
1497 	 */
1498 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1499 	    mp->mnt_susp_owner != curthread) {
1500 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1501 			if (flags & V_NOWAIT) {
1502 				error = EWOULDBLOCK;
1503 				goto unlock;
1504 			}
1505 			error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1506 			    (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1507 			if (error)
1508 				goto unlock;
1509 		}
1510 	}
1511 	if (flags & V_XSLEEP)
1512 		goto unlock;
1513 	mp->mnt_writeopcount++;
1514 unlock:
1515 	if (error != 0 || (flags & V_XSLEEP) != 0)
1516 		MNT_REL(mp);
1517 	MNT_IUNLOCK(mp);
1518 	return (error);
1519 }
1520 
1521 int
1522 vn_start_write(vp, mpp, flags)
1523 	struct vnode *vp;
1524 	struct mount **mpp;
1525 	int flags;
1526 {
1527 	struct mount *mp;
1528 	int error;
1529 
1530 	error = 0;
1531 	/*
1532 	 * If a vnode is provided, get and return the mount point that
1533 	 * to which it will write.
1534 	 */
1535 	if (vp != NULL) {
1536 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1537 			*mpp = NULL;
1538 			if (error != EOPNOTSUPP)
1539 				return (error);
1540 			return (0);
1541 		}
1542 	}
1543 	if ((mp = *mpp) == NULL)
1544 		return (0);
1545 
1546 	/*
1547 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1548 	 * a vfs_ref().
1549 	 * As long as a vnode is not provided we need to acquire a
1550 	 * refcount for the provided mountpoint too, in order to
1551 	 * emulate a vfs_ref().
1552 	 */
1553 	MNT_ILOCK(mp);
1554 	if (vp == NULL)
1555 		MNT_REF(mp);
1556 
1557 	return (vn_start_write_locked(mp, flags));
1558 }
1559 
1560 /*
1561  * Secondary suspension. Used by operations such as vop_inactive
1562  * routines that are needed by the higher level functions. These
1563  * are allowed to proceed until all the higher level functions have
1564  * completed (indicated by mnt_writeopcount dropping to zero). At that
1565  * time, these operations are halted until the suspension is over.
1566  */
1567 int
1568 vn_start_secondary_write(vp, mpp, flags)
1569 	struct vnode *vp;
1570 	struct mount **mpp;
1571 	int flags;
1572 {
1573 	struct mount *mp;
1574 	int error;
1575 
1576  retry:
1577 	if (vp != NULL) {
1578 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1579 			*mpp = NULL;
1580 			if (error != EOPNOTSUPP)
1581 				return (error);
1582 			return (0);
1583 		}
1584 	}
1585 	/*
1586 	 * If we are not suspended or have not yet reached suspended
1587 	 * mode, then let the operation proceed.
1588 	 */
1589 	if ((mp = *mpp) == NULL)
1590 		return (0);
1591 
1592 	/*
1593 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1594 	 * a vfs_ref().
1595 	 * As long as a vnode is not provided we need to acquire a
1596 	 * refcount for the provided mountpoint too, in order to
1597 	 * emulate a vfs_ref().
1598 	 */
1599 	MNT_ILOCK(mp);
1600 	if (vp == NULL)
1601 		MNT_REF(mp);
1602 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1603 		mp->mnt_secondary_writes++;
1604 		mp->mnt_secondary_accwrites++;
1605 		MNT_IUNLOCK(mp);
1606 		return (0);
1607 	}
1608 	if (flags & V_NOWAIT) {
1609 		MNT_REL(mp);
1610 		MNT_IUNLOCK(mp);
1611 		return (EWOULDBLOCK);
1612 	}
1613 	/*
1614 	 * Wait for the suspension to finish.
1615 	 */
1616 	error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1617 		       (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1618 	vfs_rel(mp);
1619 	if (error == 0)
1620 		goto retry;
1621 	return (error);
1622 }
1623 
1624 /*
1625  * Filesystem write operation has completed. If we are suspending and this
1626  * operation is the last one, notify the suspender that the suspension is
1627  * now in effect.
1628  */
1629 void
1630 vn_finished_write(mp)
1631 	struct mount *mp;
1632 {
1633 	if (mp == NULL)
1634 		return;
1635 	MNT_ILOCK(mp);
1636 	MNT_REL(mp);
1637 	mp->mnt_writeopcount--;
1638 	if (mp->mnt_writeopcount < 0)
1639 		panic("vn_finished_write: neg cnt");
1640 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1641 	    mp->mnt_writeopcount <= 0)
1642 		wakeup(&mp->mnt_writeopcount);
1643 	MNT_IUNLOCK(mp);
1644 }
1645 
1646 
1647 /*
1648  * Filesystem secondary write operation has completed. If we are
1649  * suspending and this operation is the last one, notify the suspender
1650  * that the suspension is now in effect.
1651  */
1652 void
1653 vn_finished_secondary_write(mp)
1654 	struct mount *mp;
1655 {
1656 	if (mp == NULL)
1657 		return;
1658 	MNT_ILOCK(mp);
1659 	MNT_REL(mp);
1660 	mp->mnt_secondary_writes--;
1661 	if (mp->mnt_secondary_writes < 0)
1662 		panic("vn_finished_secondary_write: neg cnt");
1663 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1664 	    mp->mnt_secondary_writes <= 0)
1665 		wakeup(&mp->mnt_secondary_writes);
1666 	MNT_IUNLOCK(mp);
1667 }
1668 
1669 
1670 
1671 /*
1672  * Request a filesystem to suspend write operations.
1673  */
1674 int
1675 vfs_write_suspend(struct mount *mp, int flags)
1676 {
1677 	int error;
1678 
1679 	MNT_ILOCK(mp);
1680 	if (mp->mnt_susp_owner == curthread) {
1681 		MNT_IUNLOCK(mp);
1682 		return (EALREADY);
1683 	}
1684 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1685 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1686 
1687 	/*
1688 	 * Unmount holds a write reference on the mount point.  If we
1689 	 * own busy reference and drain for writers, we deadlock with
1690 	 * the reference draining in the unmount path.  Callers of
1691 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1692 	 * vfs_busy() reference is owned and caller is not in the
1693 	 * unmount context.
1694 	 */
1695 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1696 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1697 		MNT_IUNLOCK(mp);
1698 		return (EBUSY);
1699 	}
1700 
1701 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1702 	mp->mnt_susp_owner = curthread;
1703 	if (mp->mnt_writeopcount > 0)
1704 		(void) msleep(&mp->mnt_writeopcount,
1705 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1706 	else
1707 		MNT_IUNLOCK(mp);
1708 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1709 		vfs_write_resume(mp, 0);
1710 	return (error);
1711 }
1712 
1713 /*
1714  * Request a filesystem to resume write operations.
1715  */
1716 void
1717 vfs_write_resume(struct mount *mp, int flags)
1718 {
1719 
1720 	MNT_ILOCK(mp);
1721 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1722 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1723 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1724 				       MNTK_SUSPENDED);
1725 		mp->mnt_susp_owner = NULL;
1726 		wakeup(&mp->mnt_writeopcount);
1727 		wakeup(&mp->mnt_flag);
1728 		curthread->td_pflags &= ~TDP_IGNSUSP;
1729 		if ((flags & VR_START_WRITE) != 0) {
1730 			MNT_REF(mp);
1731 			mp->mnt_writeopcount++;
1732 		}
1733 		MNT_IUNLOCK(mp);
1734 		if ((flags & VR_NO_SUSPCLR) == 0)
1735 			VFS_SUSP_CLEAN(mp);
1736 	} else if ((flags & VR_START_WRITE) != 0) {
1737 		MNT_REF(mp);
1738 		vn_start_write_locked(mp, 0);
1739 	} else {
1740 		MNT_IUNLOCK(mp);
1741 	}
1742 }
1743 
1744 /*
1745  * Implement kqueues for files by translating it to vnode operation.
1746  */
1747 static int
1748 vn_kqfilter(struct file *fp, struct knote *kn)
1749 {
1750 
1751 	return (VOP_KQFILTER(fp->f_vnode, kn));
1752 }
1753 
1754 /*
1755  * Simplified in-kernel wrapper calls for extended attribute access.
1756  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1757  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1758  */
1759 int
1760 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1761     const char *attrname, int *buflen, char *buf, struct thread *td)
1762 {
1763 	struct uio	auio;
1764 	struct iovec	iov;
1765 	int	error;
1766 
1767 	iov.iov_len = *buflen;
1768 	iov.iov_base = buf;
1769 
1770 	auio.uio_iov = &iov;
1771 	auio.uio_iovcnt = 1;
1772 	auio.uio_rw = UIO_READ;
1773 	auio.uio_segflg = UIO_SYSSPACE;
1774 	auio.uio_td = td;
1775 	auio.uio_offset = 0;
1776 	auio.uio_resid = *buflen;
1777 
1778 	if ((ioflg & IO_NODELOCKED) == 0)
1779 		vn_lock(vp, LK_SHARED | LK_RETRY);
1780 
1781 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1782 
1783 	/* authorize attribute retrieval as kernel */
1784 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1785 	    td);
1786 
1787 	if ((ioflg & IO_NODELOCKED) == 0)
1788 		VOP_UNLOCK(vp, 0);
1789 
1790 	if (error == 0) {
1791 		*buflen = *buflen - auio.uio_resid;
1792 	}
1793 
1794 	return (error);
1795 }
1796 
1797 /*
1798  * XXX failure mode if partially written?
1799  */
1800 int
1801 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1802     const char *attrname, int buflen, char *buf, struct thread *td)
1803 {
1804 	struct uio	auio;
1805 	struct iovec	iov;
1806 	struct mount	*mp;
1807 	int	error;
1808 
1809 	iov.iov_len = buflen;
1810 	iov.iov_base = buf;
1811 
1812 	auio.uio_iov = &iov;
1813 	auio.uio_iovcnt = 1;
1814 	auio.uio_rw = UIO_WRITE;
1815 	auio.uio_segflg = UIO_SYSSPACE;
1816 	auio.uio_td = td;
1817 	auio.uio_offset = 0;
1818 	auio.uio_resid = buflen;
1819 
1820 	if ((ioflg & IO_NODELOCKED) == 0) {
1821 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1822 			return (error);
1823 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1824 	}
1825 
1826 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1827 
1828 	/* authorize attribute setting as kernel */
1829 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1830 
1831 	if ((ioflg & IO_NODELOCKED) == 0) {
1832 		vn_finished_write(mp);
1833 		VOP_UNLOCK(vp, 0);
1834 	}
1835 
1836 	return (error);
1837 }
1838 
1839 int
1840 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1841     const char *attrname, struct thread *td)
1842 {
1843 	struct mount	*mp;
1844 	int	error;
1845 
1846 	if ((ioflg & IO_NODELOCKED) == 0) {
1847 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1848 			return (error);
1849 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1850 	}
1851 
1852 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1853 
1854 	/* authorize attribute removal as kernel */
1855 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1856 	if (error == EOPNOTSUPP)
1857 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1858 		    NULL, td);
1859 
1860 	if ((ioflg & IO_NODELOCKED) == 0) {
1861 		vn_finished_write(mp);
1862 		VOP_UNLOCK(vp, 0);
1863 	}
1864 
1865 	return (error);
1866 }
1867 
1868 int
1869 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1870 {
1871 	struct mount *mp;
1872 	int ltype, error;
1873 
1874 	mp = vp->v_mount;
1875 	ltype = VOP_ISLOCKED(vp);
1876 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1877 	    ("vn_vget_ino: vp not locked"));
1878 	error = vfs_busy(mp, MBF_NOWAIT);
1879 	if (error != 0) {
1880 		vfs_ref(mp);
1881 		VOP_UNLOCK(vp, 0);
1882 		error = vfs_busy(mp, 0);
1883 		vn_lock(vp, ltype | LK_RETRY);
1884 		vfs_rel(mp);
1885 		if (error != 0)
1886 			return (ENOENT);
1887 		if (vp->v_iflag & VI_DOOMED) {
1888 			vfs_unbusy(mp);
1889 			return (ENOENT);
1890 		}
1891 	}
1892 	VOP_UNLOCK(vp, 0);
1893 	error = VFS_VGET(mp, ino, lkflags, rvp);
1894 	vfs_unbusy(mp);
1895 	vn_lock(vp, ltype | LK_RETRY);
1896 	if (vp->v_iflag & VI_DOOMED) {
1897 		if (error == 0)
1898 			vput(*rvp);
1899 		error = ENOENT;
1900 	}
1901 	return (error);
1902 }
1903 
1904 int
1905 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1906     const struct thread *td)
1907 {
1908 
1909 	if (vp->v_type != VREG || td == NULL)
1910 		return (0);
1911 	PROC_LOCK(td->td_proc);
1912 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1913 	    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1914 		kern_psignal(td->td_proc, SIGXFSZ);
1915 		PROC_UNLOCK(td->td_proc);
1916 		return (EFBIG);
1917 	}
1918 	PROC_UNLOCK(td->td_proc);
1919 	return (0);
1920 }
1921 
1922 int
1923 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1924     struct thread *td)
1925 {
1926 	struct vnode *vp;
1927 
1928 	vp = fp->f_vnode;
1929 #ifdef AUDIT
1930 	vn_lock(vp, LK_SHARED | LK_RETRY);
1931 	AUDIT_ARG_VNODE1(vp);
1932 	VOP_UNLOCK(vp, 0);
1933 #endif
1934 	return (setfmode(td, active_cred, vp, mode));
1935 }
1936 
1937 int
1938 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1939     struct thread *td)
1940 {
1941 	struct vnode *vp;
1942 
1943 	vp = fp->f_vnode;
1944 #ifdef AUDIT
1945 	vn_lock(vp, LK_SHARED | LK_RETRY);
1946 	AUDIT_ARG_VNODE1(vp);
1947 	VOP_UNLOCK(vp, 0);
1948 #endif
1949 	return (setfown(td, active_cred, vp, uid, gid));
1950 }
1951 
1952 void
1953 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1954 {
1955 	vm_object_t object;
1956 
1957 	if ((object = vp->v_object) == NULL)
1958 		return;
1959 	VM_OBJECT_WLOCK(object);
1960 	vm_object_page_remove(object, start, end, 0);
1961 	VM_OBJECT_WUNLOCK(object);
1962 }
1963 
1964 int
1965 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1966 {
1967 	struct vattr va;
1968 	daddr_t bn, bnp;
1969 	uint64_t bsize;
1970 	off_t noff;
1971 	int error;
1972 
1973 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1974 	    ("Wrong command %lu", cmd));
1975 
1976 	if (vn_lock(vp, LK_SHARED) != 0)
1977 		return (EBADF);
1978 	if (vp->v_type != VREG) {
1979 		error = ENOTTY;
1980 		goto unlock;
1981 	}
1982 	error = VOP_GETATTR(vp, &va, cred);
1983 	if (error != 0)
1984 		goto unlock;
1985 	noff = *off;
1986 	if (noff >= va.va_size) {
1987 		error = ENXIO;
1988 		goto unlock;
1989 	}
1990 	bsize = vp->v_mount->mnt_stat.f_iosize;
1991 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1992 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1993 		if (error == EOPNOTSUPP) {
1994 			error = ENOTTY;
1995 			goto unlock;
1996 		}
1997 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1998 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
1999 			noff = bn * bsize;
2000 			if (noff < *off)
2001 				noff = *off;
2002 			goto unlock;
2003 		}
2004 	}
2005 	if (noff > va.va_size)
2006 		noff = va.va_size;
2007 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2008 	if (cmd == FIOSEEKDATA)
2009 		error = ENXIO;
2010 unlock:
2011 	VOP_UNLOCK(vp, 0);
2012 	if (error == 0)
2013 		*off = noff;
2014 	return (error);
2015 }
2016 
2017 int
2018 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2019 {
2020 	struct ucred *cred;
2021 	struct vnode *vp;
2022 	struct vattr vattr;
2023 	off_t foffset, size;
2024 	int error, noneg;
2025 
2026 	cred = td->td_ucred;
2027 	vp = fp->f_vnode;
2028 	foffset = foffset_lock(fp, 0);
2029 	noneg = (vp->v_type != VCHR);
2030 	error = 0;
2031 	switch (whence) {
2032 	case L_INCR:
2033 		if (noneg &&
2034 		    (foffset < 0 ||
2035 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2036 			error = EOVERFLOW;
2037 			break;
2038 		}
2039 		offset += foffset;
2040 		break;
2041 	case L_XTND:
2042 		vn_lock(vp, LK_SHARED | LK_RETRY);
2043 		error = VOP_GETATTR(vp, &vattr, cred);
2044 		VOP_UNLOCK(vp, 0);
2045 		if (error)
2046 			break;
2047 
2048 		/*
2049 		 * If the file references a disk device, then fetch
2050 		 * the media size and use that to determine the ending
2051 		 * offset.
2052 		 */
2053 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2054 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2055 			vattr.va_size = size;
2056 		if (noneg &&
2057 		    (vattr.va_size > OFF_MAX ||
2058 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2059 			error = EOVERFLOW;
2060 			break;
2061 		}
2062 		offset += vattr.va_size;
2063 		break;
2064 	case L_SET:
2065 		break;
2066 	case SEEK_DATA:
2067 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2068 		break;
2069 	case SEEK_HOLE:
2070 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2071 		break;
2072 	default:
2073 		error = EINVAL;
2074 	}
2075 	if (error == 0 && noneg && offset < 0)
2076 		error = EINVAL;
2077 	if (error != 0)
2078 		goto drop;
2079 	VFS_KNOTE_UNLOCKED(vp, 0);
2080 	*(off_t *)(td->td_retval) = offset;
2081 drop:
2082 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2083 	return (error);
2084 }
2085