xref: /freebsd/sys/kern/vfs_vnops.c (revision d8000daa92a4cdf940348f483861d52fa3cbf91d)
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, 2014 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/mman.h>
58 #include <sys/mount.h>
59 #include <sys/mutex.h>
60 #include <sys/namei.h>
61 #include <sys/vnode.h>
62 #include <sys/bio.h>
63 #include <sys/buf.h>
64 #include <sys/filio.h>
65 #include <sys/resourcevar.h>
66 #include <sys/rwlock.h>
67 #include <sys/sx.h>
68 #include <sys/sysctl.h>
69 #include <sys/ttycom.h>
70 #include <sys/conf.h>
71 #include <sys/syslog.h>
72 #include <sys/unistd.h>
73 #include <sys/user.h>
74 
75 #include <security/audit/audit.h>
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/vm_extern.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vnode_pager.h>
85 
86 static fo_rdwr_t	vn_read;
87 static fo_rdwr_t	vn_write;
88 static fo_rdwr_t	vn_io_fault;
89 static fo_truncate_t	vn_truncate;
90 static fo_ioctl_t	vn_ioctl;
91 static fo_poll_t	vn_poll;
92 static fo_kqfilter_t	vn_kqfilter;
93 static fo_stat_t	vn_statfile;
94 static fo_close_t	vn_closefile;
95 static fo_mmap_t	vn_mmap;
96 
97 struct 	fileops vnops = {
98 	.fo_read = vn_io_fault,
99 	.fo_write = vn_io_fault,
100 	.fo_truncate = vn_truncate,
101 	.fo_ioctl = vn_ioctl,
102 	.fo_poll = vn_poll,
103 	.fo_kqfilter = vn_kqfilter,
104 	.fo_stat = vn_statfile,
105 	.fo_close = vn_closefile,
106 	.fo_chmod = vn_chmod,
107 	.fo_chown = vn_chown,
108 	.fo_sendfile = vn_sendfile,
109 	.fo_seek = vn_seek,
110 	.fo_fill_kinfo = vn_fill_kinfo,
111 	.fo_mmap = vn_mmap,
112 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
113 };
114 
115 static const int io_hold_cnt = 16;
116 static int vn_io_fault_enable = 1;
117 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
118     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
119 static int vn_io_fault_prefault = 0;
120 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
121     &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
122 static u_long vn_io_faults_cnt;
123 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
124     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
125 
126 /*
127  * Returns true if vn_io_fault mode of handling the i/o request should
128  * be used.
129  */
130 static bool
131 do_vn_io_fault(struct vnode *vp, struct uio *uio)
132 {
133 	struct mount *mp;
134 
135 	return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
136 	    (mp = vp->v_mount) != NULL &&
137 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
138 }
139 
140 /*
141  * Structure used to pass arguments to vn_io_fault1(), to do either
142  * file- or vnode-based I/O calls.
143  */
144 struct vn_io_fault_args {
145 	enum {
146 		VN_IO_FAULT_FOP,
147 		VN_IO_FAULT_VOP
148 	} kind;
149 	struct ucred *cred;
150 	int flags;
151 	union {
152 		struct fop_args_tag {
153 			struct file *fp;
154 			fo_rdwr_t *doio;
155 		} fop_args;
156 		struct vop_args_tag {
157 			struct vnode *vp;
158 		} vop_args;
159 	} args;
160 };
161 
162 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
163     struct vn_io_fault_args *args, struct thread *td);
164 
165 int
166 vn_open(ndp, flagp, cmode, fp)
167 	struct nameidata *ndp;
168 	int *flagp, cmode;
169 	struct file *fp;
170 {
171 	struct thread *td = ndp->ni_cnd.cn_thread;
172 
173 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
174 }
175 
176 /*
177  * Common code for vnode open operations via a name lookup.
178  * Lookup the vnode and invoke VOP_CREATE if needed.
179  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
180  *
181  * Note that this does NOT free nameidata for the successful case,
182  * due to the NDINIT being done elsewhere.
183  */
184 int
185 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
186     struct ucred *cred, struct file *fp)
187 {
188 	struct vnode *vp;
189 	struct mount *mp;
190 	struct thread *td = ndp->ni_cnd.cn_thread;
191 	struct vattr vat;
192 	struct vattr *vap = &vat;
193 	int fmode, error;
194 
195 restart:
196 	fmode = *flagp;
197 	if (fmode & O_CREAT) {
198 		ndp->ni_cnd.cn_nameiop = CREATE;
199 		/*
200 		 * Set NOCACHE to avoid flushing the cache when
201 		 * rolling in many files at once.
202 		*/
203 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
204 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
205 			ndp->ni_cnd.cn_flags |= FOLLOW;
206 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
207 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
208 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
209 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
210 		bwillwrite();
211 		if ((error = namei(ndp)) != 0)
212 			return (error);
213 		if (ndp->ni_vp == NULL) {
214 			VATTR_NULL(vap);
215 			vap->va_type = VREG;
216 			vap->va_mode = cmode;
217 			if (fmode & O_EXCL)
218 				vap->va_vaflags |= VA_EXCLUSIVE;
219 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
220 				NDFREE(ndp, NDF_ONLY_PNBUF);
221 				vput(ndp->ni_dvp);
222 				if ((error = vn_start_write(NULL, &mp,
223 				    V_XSLEEP | PCATCH)) != 0)
224 					return (error);
225 				goto restart;
226 			}
227 			if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
228 				ndp->ni_cnd.cn_flags |= MAKEENTRY;
229 #ifdef MAC
230 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
231 			    &ndp->ni_cnd, vap);
232 			if (error == 0)
233 #endif
234 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
235 						   &ndp->ni_cnd, vap);
236 			vput(ndp->ni_dvp);
237 			vn_finished_write(mp);
238 			if (error) {
239 				NDFREE(ndp, NDF_ONLY_PNBUF);
240 				return (error);
241 			}
242 			fmode &= ~O_TRUNC;
243 			vp = ndp->ni_vp;
244 		} else {
245 			if (ndp->ni_dvp == ndp->ni_vp)
246 				vrele(ndp->ni_dvp);
247 			else
248 				vput(ndp->ni_dvp);
249 			ndp->ni_dvp = NULL;
250 			vp = ndp->ni_vp;
251 			if (fmode & O_EXCL) {
252 				error = EEXIST;
253 				goto bad;
254 			}
255 			fmode &= ~O_CREAT;
256 		}
257 	} else {
258 		ndp->ni_cnd.cn_nameiop = LOOKUP;
259 		ndp->ni_cnd.cn_flags = ISOPEN |
260 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
261 		if (!(fmode & FWRITE))
262 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
263 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
264 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
265 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
266 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
267 		if ((error = namei(ndp)) != 0)
268 			return (error);
269 		vp = ndp->ni_vp;
270 	}
271 	error = vn_open_vnode(vp, fmode, cred, td, fp);
272 	if (error)
273 		goto bad;
274 	*flagp = fmode;
275 	return (0);
276 bad:
277 	NDFREE(ndp, NDF_ONLY_PNBUF);
278 	vput(vp);
279 	*flagp = fmode;
280 	ndp->ni_vp = NULL;
281 	return (error);
282 }
283 
284 /*
285  * Common code for vnode open operations once a vnode is located.
286  * Check permissions, and call the VOP_OPEN routine.
287  */
288 int
289 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
290     struct thread *td, struct file *fp)
291 {
292 	struct mount *mp;
293 	accmode_t accmode;
294 	struct flock lf;
295 	int error, have_flock, lock_flags, type;
296 
297 	if (vp->v_type == VLNK)
298 		return (EMLINK);
299 	if (vp->v_type == VSOCK)
300 		return (EOPNOTSUPP);
301 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
302 		return (ENOTDIR);
303 	accmode = 0;
304 	if (fmode & (FWRITE | O_TRUNC)) {
305 		if (vp->v_type == VDIR)
306 			return (EISDIR);
307 		accmode |= VWRITE;
308 	}
309 	if (fmode & FREAD)
310 		accmode |= VREAD;
311 	if (fmode & FEXEC)
312 		accmode |= VEXEC;
313 	if ((fmode & O_APPEND) && (fmode & FWRITE))
314 		accmode |= VAPPEND;
315 #ifdef MAC
316 	if (fmode & O_CREAT)
317 		accmode |= VCREAT;
318 	if (fmode & O_VERIFY)
319 		accmode |= VVERIFY;
320 	error = mac_vnode_check_open(cred, vp, accmode);
321 	if (error)
322 		return (error);
323 
324 	accmode &= ~(VCREAT | VVERIFY);
325 #endif
326 	if ((fmode & O_CREAT) == 0) {
327 		if (accmode & VWRITE) {
328 			error = vn_writechk(vp);
329 			if (error)
330 				return (error);
331 		}
332 		if (accmode) {
333 		        error = VOP_ACCESS(vp, accmode, cred, td);
334 			if (error)
335 				return (error);
336 		}
337 	}
338 	if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
339 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
340 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
341 		return (error);
342 
343 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
344 		KASSERT(fp != NULL, ("open with flock requires fp"));
345 		lock_flags = VOP_ISLOCKED(vp);
346 		VOP_UNLOCK(vp, 0);
347 		lf.l_whence = SEEK_SET;
348 		lf.l_start = 0;
349 		lf.l_len = 0;
350 		if (fmode & O_EXLOCK)
351 			lf.l_type = F_WRLCK;
352 		else
353 			lf.l_type = F_RDLCK;
354 		type = F_FLOCK;
355 		if ((fmode & FNONBLOCK) == 0)
356 			type |= F_WAIT;
357 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
358 		have_flock = (error == 0);
359 		vn_lock(vp, lock_flags | LK_RETRY);
360 		if (error == 0 && vp->v_iflag & VI_DOOMED)
361 			error = ENOENT;
362 		/*
363 		 * Another thread might have used this vnode as an
364 		 * executable while the vnode lock was dropped.
365 		 * Ensure the vnode is still able to be opened for
366 		 * writing after the lock has been obtained.
367 		 */
368 		if (error == 0 && accmode & VWRITE)
369 			error = vn_writechk(vp);
370 		if (error) {
371 			VOP_UNLOCK(vp, 0);
372 			if (have_flock) {
373 				lf.l_whence = SEEK_SET;
374 				lf.l_start = 0;
375 				lf.l_len = 0;
376 				lf.l_type = F_UNLCK;
377 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
378 				    F_FLOCK);
379 			}
380 			vn_start_write(vp, &mp, V_WAIT);
381 			vn_lock(vp, lock_flags | LK_RETRY);
382 			(void)VOP_CLOSE(vp, fmode, cred, td);
383 			vn_finished_write(mp);
384 			/* Prevent second close from fdrop()->vn_close(). */
385 			if (fp != NULL)
386 				fp->f_ops= &badfileops;
387 			return (error);
388 		}
389 		fp->f_flag |= FHASLOCK;
390 	}
391 	if (fmode & FWRITE) {
392 		VOP_ADD_WRITECOUNT(vp, 1);
393 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
394 		    __func__, vp, vp->v_writecount);
395 	}
396 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
397 	return (0);
398 }
399 
400 /*
401  * Check for write permissions on the specified vnode.
402  * Prototype text segments cannot be written.
403  */
404 int
405 vn_writechk(vp)
406 	register struct vnode *vp;
407 {
408 
409 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
410 	/*
411 	 * If there's shared text associated with
412 	 * the vnode, try to free it up once.  If
413 	 * we fail, we can't allow writing.
414 	 */
415 	if (VOP_IS_TEXT(vp))
416 		return (ETXTBSY);
417 
418 	return (0);
419 }
420 
421 /*
422  * Vnode close call
423  */
424 int
425 vn_close(vp, flags, file_cred, td)
426 	register struct vnode *vp;
427 	int flags;
428 	struct ucred *file_cred;
429 	struct thread *td;
430 {
431 	struct mount *mp;
432 	int error, lock_flags;
433 
434 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
435 	    MNT_EXTENDED_SHARED(vp->v_mount))
436 		lock_flags = LK_SHARED;
437 	else
438 		lock_flags = LK_EXCLUSIVE;
439 
440 	vn_start_write(vp, &mp, V_WAIT);
441 	vn_lock(vp, lock_flags | LK_RETRY);
442 	if (flags & FWRITE) {
443 		VNASSERT(vp->v_writecount > 0, vp,
444 		    ("vn_close: negative writecount"));
445 		VOP_ADD_WRITECOUNT(vp, -1);
446 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
447 		    __func__, vp, vp->v_writecount);
448 	}
449 	error = VOP_CLOSE(vp, flags, file_cred, td);
450 	vput(vp);
451 	vn_finished_write(mp);
452 	return (error);
453 }
454 
455 /*
456  * Heuristic to detect sequential operation.
457  */
458 static int
459 sequential_heuristic(struct uio *uio, struct file *fp)
460 {
461 
462 	ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
463 	if (fp->f_flag & FRDAHEAD)
464 		return (fp->f_seqcount << IO_SEQSHIFT);
465 
466 	/*
467 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
468 	 * that the first I/O is normally considered to be slightly
469 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
470 	 * unless previous seeks have reduced f_seqcount to 0, in which
471 	 * case offset 0 is not special.
472 	 */
473 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
474 	    uio->uio_offset == fp->f_nextoff) {
475 		/*
476 		 * f_seqcount is in units of fixed-size blocks so that it
477 		 * depends mainly on the amount of sequential I/O and not
478 		 * much on the number of sequential I/O's.  The fixed size
479 		 * of 16384 is hard-coded here since it is (not quite) just
480 		 * a magic size that works well here.  This size is more
481 		 * closely related to the best I/O size for real disks than
482 		 * to any block size used by software.
483 		 */
484 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
485 		if (fp->f_seqcount > IO_SEQMAX)
486 			fp->f_seqcount = IO_SEQMAX;
487 		return (fp->f_seqcount << IO_SEQSHIFT);
488 	}
489 
490 	/* Not sequential.  Quickly draw-down sequentiality. */
491 	if (fp->f_seqcount > 1)
492 		fp->f_seqcount = 1;
493 	else
494 		fp->f_seqcount = 0;
495 	return (0);
496 }
497 
498 /*
499  * Package up an I/O request on a vnode into a uio and do it.
500  */
501 int
502 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
503     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
504     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
505 {
506 	struct uio auio;
507 	struct iovec aiov;
508 	struct mount *mp;
509 	struct ucred *cred;
510 	void *rl_cookie;
511 	struct vn_io_fault_args args;
512 	int error, lock_flags;
513 
514 	auio.uio_iov = &aiov;
515 	auio.uio_iovcnt = 1;
516 	aiov.iov_base = base;
517 	aiov.iov_len = len;
518 	auio.uio_resid = len;
519 	auio.uio_offset = offset;
520 	auio.uio_segflg = segflg;
521 	auio.uio_rw = rw;
522 	auio.uio_td = td;
523 	error = 0;
524 
525 	if ((ioflg & IO_NODELOCKED) == 0) {
526 		if ((ioflg & IO_RANGELOCKED) == 0) {
527 			if (rw == UIO_READ) {
528 				rl_cookie = vn_rangelock_rlock(vp, offset,
529 				    offset + len);
530 			} else {
531 				rl_cookie = vn_rangelock_wlock(vp, offset,
532 				    offset + len);
533 			}
534 		} else
535 			rl_cookie = NULL;
536 		mp = NULL;
537 		if (rw == UIO_WRITE) {
538 			if (vp->v_type != VCHR &&
539 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
540 			    != 0)
541 				goto out;
542 			if (MNT_SHARED_WRITES(mp) ||
543 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
544 				lock_flags = LK_SHARED;
545 			else
546 				lock_flags = LK_EXCLUSIVE;
547 		} else
548 			lock_flags = LK_SHARED;
549 		vn_lock(vp, lock_flags | LK_RETRY);
550 	} else
551 		rl_cookie = NULL;
552 
553 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
554 #ifdef MAC
555 	if ((ioflg & IO_NOMACCHECK) == 0) {
556 		if (rw == UIO_READ)
557 			error = mac_vnode_check_read(active_cred, file_cred,
558 			    vp);
559 		else
560 			error = mac_vnode_check_write(active_cred, file_cred,
561 			    vp);
562 	}
563 #endif
564 	if (error == 0) {
565 		if (file_cred != NULL)
566 			cred = file_cred;
567 		else
568 			cred = active_cred;
569 		if (do_vn_io_fault(vp, &auio)) {
570 			args.kind = VN_IO_FAULT_VOP;
571 			args.cred = cred;
572 			args.flags = ioflg;
573 			args.args.vop_args.vp = vp;
574 			error = vn_io_fault1(vp, &auio, &args, td);
575 		} else if (rw == UIO_READ) {
576 			error = VOP_READ(vp, &auio, ioflg, cred);
577 		} else /* if (rw == UIO_WRITE) */ {
578 			error = VOP_WRITE(vp, &auio, ioflg, cred);
579 		}
580 	}
581 	if (aresid)
582 		*aresid = auio.uio_resid;
583 	else
584 		if (auio.uio_resid && error == 0)
585 			error = EIO;
586 	if ((ioflg & IO_NODELOCKED) == 0) {
587 		VOP_UNLOCK(vp, 0);
588 		if (mp != NULL)
589 			vn_finished_write(mp);
590 	}
591  out:
592 	if (rl_cookie != NULL)
593 		vn_rangelock_unlock(vp, rl_cookie);
594 	return (error);
595 }
596 
597 /*
598  * Package up an I/O request on a vnode into a uio and do it.  The I/O
599  * request is split up into smaller chunks and we try to avoid saturating
600  * the buffer cache while potentially holding a vnode locked, so we
601  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
602  * to give other processes a chance to lock the vnode (either other processes
603  * core'ing the same binary, or unrelated processes scanning the directory).
604  */
605 int
606 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
607     file_cred, aresid, td)
608 	enum uio_rw rw;
609 	struct vnode *vp;
610 	void *base;
611 	size_t len;
612 	off_t offset;
613 	enum uio_seg segflg;
614 	int ioflg;
615 	struct ucred *active_cred;
616 	struct ucred *file_cred;
617 	size_t *aresid;
618 	struct thread *td;
619 {
620 	int error = 0;
621 	ssize_t iaresid;
622 
623 	do {
624 		int chunk;
625 
626 		/*
627 		 * Force `offset' to a multiple of MAXBSIZE except possibly
628 		 * for the first chunk, so that filesystems only need to
629 		 * write full blocks except possibly for the first and last
630 		 * chunks.
631 		 */
632 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
633 
634 		if (chunk > len)
635 			chunk = len;
636 		if (rw != UIO_READ && vp->v_type == VREG)
637 			bwillwrite();
638 		iaresid = 0;
639 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
640 		    ioflg, active_cred, file_cred, &iaresid, td);
641 		len -= chunk;	/* aresid calc already includes length */
642 		if (error)
643 			break;
644 		offset += chunk;
645 		base = (char *)base + chunk;
646 		kern_yield(PRI_USER);
647 	} while (len);
648 	if (aresid)
649 		*aresid = len + iaresid;
650 	return (error);
651 }
652 
653 off_t
654 foffset_lock(struct file *fp, int flags)
655 {
656 	struct mtx *mtxp;
657 	off_t res;
658 
659 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
660 
661 #if OFF_MAX <= LONG_MAX
662 	/*
663 	 * Caller only wants the current f_offset value.  Assume that
664 	 * the long and shorter integer types reads are atomic.
665 	 */
666 	if ((flags & FOF_NOLOCK) != 0)
667 		return (fp->f_offset);
668 #endif
669 
670 	/*
671 	 * According to McKusick the vn lock was protecting f_offset here.
672 	 * It is now protected by the FOFFSET_LOCKED flag.
673 	 */
674 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
675 	mtx_lock(mtxp);
676 	if ((flags & FOF_NOLOCK) == 0) {
677 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
678 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
679 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
680 			    "vofflock", 0);
681 		}
682 		fp->f_vnread_flags |= FOFFSET_LOCKED;
683 	}
684 	res = fp->f_offset;
685 	mtx_unlock(mtxp);
686 	return (res);
687 }
688 
689 void
690 foffset_unlock(struct file *fp, off_t val, int flags)
691 {
692 	struct mtx *mtxp;
693 
694 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
695 
696 #if OFF_MAX <= LONG_MAX
697 	if ((flags & FOF_NOLOCK) != 0) {
698 		if ((flags & FOF_NOUPDATE) == 0)
699 			fp->f_offset = val;
700 		if ((flags & FOF_NEXTOFF) != 0)
701 			fp->f_nextoff = val;
702 		return;
703 	}
704 #endif
705 
706 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
707 	mtx_lock(mtxp);
708 	if ((flags & FOF_NOUPDATE) == 0)
709 		fp->f_offset = val;
710 	if ((flags & FOF_NEXTOFF) != 0)
711 		fp->f_nextoff = val;
712 	if ((flags & FOF_NOLOCK) == 0) {
713 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
714 		    ("Lost FOFFSET_LOCKED"));
715 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
716 			wakeup(&fp->f_vnread_flags);
717 		fp->f_vnread_flags = 0;
718 	}
719 	mtx_unlock(mtxp);
720 }
721 
722 void
723 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
724 {
725 
726 	if ((flags & FOF_OFFSET) == 0)
727 		uio->uio_offset = foffset_lock(fp, flags);
728 }
729 
730 void
731 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
732 {
733 
734 	if ((flags & FOF_OFFSET) == 0)
735 		foffset_unlock(fp, uio->uio_offset, flags);
736 }
737 
738 static int
739 get_advice(struct file *fp, struct uio *uio)
740 {
741 	struct mtx *mtxp;
742 	int ret;
743 
744 	ret = POSIX_FADV_NORMAL;
745 	if (fp->f_advice == NULL)
746 		return (ret);
747 
748 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
749 	mtx_lock(mtxp);
750 	if (uio->uio_offset >= fp->f_advice->fa_start &&
751 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
752 		ret = fp->f_advice->fa_advice;
753 	mtx_unlock(mtxp);
754 	return (ret);
755 }
756 
757 /*
758  * File table vnode read routine.
759  */
760 static int
761 vn_read(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 mtx *mtxp;
770 	int error, ioflag;
771 	int advice;
772 	off_t offset, start, end;
773 
774 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
775 	    uio->uio_td, td));
776 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
777 	vp = fp->f_vnode;
778 	ioflag = 0;
779 	if (fp->f_flag & FNONBLOCK)
780 		ioflag |= IO_NDELAY;
781 	if (fp->f_flag & O_DIRECT)
782 		ioflag |= IO_DIRECT;
783 	advice = get_advice(fp, uio);
784 	vn_lock(vp, LK_SHARED | LK_RETRY);
785 
786 	switch (advice) {
787 	case POSIX_FADV_NORMAL:
788 	case POSIX_FADV_SEQUENTIAL:
789 	case POSIX_FADV_NOREUSE:
790 		ioflag |= sequential_heuristic(uio, fp);
791 		break;
792 	case POSIX_FADV_RANDOM:
793 		/* Disable read-ahead for random I/O. */
794 		break;
795 	}
796 	offset = uio->uio_offset;
797 
798 #ifdef MAC
799 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
800 	if (error == 0)
801 #endif
802 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
803 	fp->f_nextoff = uio->uio_offset;
804 	VOP_UNLOCK(vp, 0);
805 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
806 	    offset != uio->uio_offset) {
807 		/*
808 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
809 		 * buffers for the backing file after a
810 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
811 		 * case of using POSIX_FADV_NOREUSE with sequential
812 		 * access, track the previous implicit DONTNEED
813 		 * request and grow this request to include the
814 		 * current read(2) in addition to the previous
815 		 * DONTNEED.  With purely sequential access this will
816 		 * cause the DONTNEED requests to continously grow to
817 		 * cover all of the previously read regions of the
818 		 * file.  This allows filesystem blocks that are
819 		 * accessed by multiple calls to read(2) to be flushed
820 		 * once the last read(2) finishes.
821 		 */
822 		start = offset;
823 		end = uio->uio_offset - 1;
824 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
825 		mtx_lock(mtxp);
826 		if (fp->f_advice != NULL &&
827 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
828 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
829 				start = fp->f_advice->fa_prevstart;
830 			else if (fp->f_advice->fa_prevstart != 0 &&
831 			    fp->f_advice->fa_prevstart == end + 1)
832 				end = fp->f_advice->fa_prevend;
833 			fp->f_advice->fa_prevstart = start;
834 			fp->f_advice->fa_prevend = end;
835 		}
836 		mtx_unlock(mtxp);
837 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
838 	}
839 	return (error);
840 }
841 
842 /*
843  * File table vnode write routine.
844  */
845 static int
846 vn_write(fp, uio, active_cred, flags, td)
847 	struct file *fp;
848 	struct uio *uio;
849 	struct ucred *active_cred;
850 	int flags;
851 	struct thread *td;
852 {
853 	struct vnode *vp;
854 	struct mount *mp;
855 	struct mtx *mtxp;
856 	int error, ioflag, lock_flags;
857 	int advice;
858 	off_t offset, start, end;
859 
860 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
861 	    uio->uio_td, td));
862 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
863 	vp = fp->f_vnode;
864 	if (vp->v_type == VREG)
865 		bwillwrite();
866 	ioflag = IO_UNIT;
867 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
868 		ioflag |= IO_APPEND;
869 	if (fp->f_flag & FNONBLOCK)
870 		ioflag |= IO_NDELAY;
871 	if (fp->f_flag & O_DIRECT)
872 		ioflag |= IO_DIRECT;
873 	if ((fp->f_flag & O_FSYNC) ||
874 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
875 		ioflag |= IO_SYNC;
876 	mp = NULL;
877 	if (vp->v_type != VCHR &&
878 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
879 		goto unlock;
880 
881 	advice = get_advice(fp, uio);
882 
883 	if (MNT_SHARED_WRITES(mp) ||
884 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
885 		lock_flags = LK_SHARED;
886 	} else {
887 		lock_flags = LK_EXCLUSIVE;
888 	}
889 
890 	vn_lock(vp, lock_flags | LK_RETRY);
891 	switch (advice) {
892 	case POSIX_FADV_NORMAL:
893 	case POSIX_FADV_SEQUENTIAL:
894 	case POSIX_FADV_NOREUSE:
895 		ioflag |= sequential_heuristic(uio, fp);
896 		break;
897 	case POSIX_FADV_RANDOM:
898 		/* XXX: Is this correct? */
899 		break;
900 	}
901 	offset = uio->uio_offset;
902 
903 #ifdef MAC
904 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
905 	if (error == 0)
906 #endif
907 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
908 	fp->f_nextoff = uio->uio_offset;
909 	VOP_UNLOCK(vp, 0);
910 	if (vp->v_type != VCHR)
911 		vn_finished_write(mp);
912 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
913 	    offset != uio->uio_offset) {
914 		/*
915 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
916 		 * buffers for the backing file after a
917 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
918 		 * common case of using POSIX_FADV_NOREUSE with
919 		 * sequential access, track the previous implicit
920 		 * DONTNEED request and grow this request to include
921 		 * the current write(2) in addition to the previous
922 		 * DONTNEED.  With purely sequential access this will
923 		 * cause the DONTNEED requests to continously grow to
924 		 * cover all of the previously written regions of the
925 		 * file.
926 		 *
927 		 * Note that the blocks just written are almost
928 		 * certainly still dirty, so this only works when
929 		 * VOP_ADVISE() calls from subsequent writes push out
930 		 * the data written by this write(2) once the backing
931 		 * buffers are clean.  However, as compared to forcing
932 		 * IO_DIRECT, this gives much saner behavior.  Write
933 		 * clustering is still allowed, and clean pages are
934 		 * merely moved to the cache page queue rather than
935 		 * outright thrown away.  This means a subsequent
936 		 * read(2) can still avoid hitting the disk if the
937 		 * pages have not been reclaimed.
938 		 *
939 		 * This does make POSIX_FADV_NOREUSE largely useless
940 		 * with non-sequential access.  However, sequential
941 		 * access is the more common use case and the flag is
942 		 * merely advisory.
943 		 */
944 		start = offset;
945 		end = uio->uio_offset - 1;
946 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
947 		mtx_lock(mtxp);
948 		if (fp->f_advice != NULL &&
949 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
950 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
951 				start = fp->f_advice->fa_prevstart;
952 			else if (fp->f_advice->fa_prevstart != 0 &&
953 			    fp->f_advice->fa_prevstart == end + 1)
954 				end = fp->f_advice->fa_prevend;
955 			fp->f_advice->fa_prevstart = start;
956 			fp->f_advice->fa_prevend = end;
957 		}
958 		mtx_unlock(mtxp);
959 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
960 	}
961 
962 unlock:
963 	return (error);
964 }
965 
966 /*
967  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
968  * prevent the following deadlock:
969  *
970  * Assume that the thread A reads from the vnode vp1 into userspace
971  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
972  * currently not resident, then system ends up with the call chain
973  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
974  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
975  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
976  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
977  * backed by the pages of vnode vp1, and some page in buf2 is not
978  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
979  *
980  * To prevent the lock order reversal and deadlock, vn_io_fault() does
981  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
982  * Instead, it first tries to do the whole range i/o with pagefaults
983  * disabled. If all pages in the i/o buffer are resident and mapped,
984  * VOP will succeed (ignoring the genuine filesystem errors).
985  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
986  * i/o in chunks, with all pages in the chunk prefaulted and held
987  * using vm_fault_quick_hold_pages().
988  *
989  * Filesystems using this deadlock avoidance scheme should use the
990  * array of the held pages from uio, saved in the curthread->td_ma,
991  * instead of doing uiomove().  A helper function
992  * vn_io_fault_uiomove() converts uiomove request into
993  * uiomove_fromphys() over td_ma array.
994  *
995  * Since vnode locks do not cover the whole i/o anymore, rangelocks
996  * make the current i/o request atomic with respect to other i/os and
997  * truncations.
998  */
999 
1000 /*
1001  * Decode vn_io_fault_args and perform the corresponding i/o.
1002  */
1003 static int
1004 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
1005     struct thread *td)
1006 {
1007 
1008 	switch (args->kind) {
1009 	case VN_IO_FAULT_FOP:
1010 		return ((args->args.fop_args.doio)(args->args.fop_args.fp,
1011 		    uio, args->cred, args->flags, td));
1012 	case VN_IO_FAULT_VOP:
1013 		if (uio->uio_rw == UIO_READ) {
1014 			return (VOP_READ(args->args.vop_args.vp, uio,
1015 			    args->flags, args->cred));
1016 		} else if (uio->uio_rw == UIO_WRITE) {
1017 			return (VOP_WRITE(args->args.vop_args.vp, uio,
1018 			    args->flags, args->cred));
1019 		}
1020 		break;
1021 	}
1022 	panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
1023 	    uio->uio_rw);
1024 }
1025 
1026 static int
1027 vn_io_fault_touch(char *base, const struct uio *uio)
1028 {
1029 	int r;
1030 
1031 	r = fubyte(base);
1032 	if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
1033 		return (EFAULT);
1034 	return (0);
1035 }
1036 
1037 static int
1038 vn_io_fault_prefault_user(const struct uio *uio)
1039 {
1040 	char *base;
1041 	const struct iovec *iov;
1042 	size_t len;
1043 	ssize_t resid;
1044 	int error, i;
1045 
1046 	KASSERT(uio->uio_segflg == UIO_USERSPACE,
1047 	    ("vn_io_fault_prefault userspace"));
1048 
1049 	error = i = 0;
1050 	iov = uio->uio_iov;
1051 	resid = uio->uio_resid;
1052 	base = iov->iov_base;
1053 	len = iov->iov_len;
1054 	while (resid > 0) {
1055 		error = vn_io_fault_touch(base, uio);
1056 		if (error != 0)
1057 			break;
1058 		if (len < PAGE_SIZE) {
1059 			if (len != 0) {
1060 				error = vn_io_fault_touch(base + len - 1, uio);
1061 				if (error != 0)
1062 					break;
1063 				resid -= len;
1064 			}
1065 			if (++i >= uio->uio_iovcnt)
1066 				break;
1067 			iov = uio->uio_iov + i;
1068 			base = iov->iov_base;
1069 			len = iov->iov_len;
1070 		} else {
1071 			len -= PAGE_SIZE;
1072 			base += PAGE_SIZE;
1073 			resid -= PAGE_SIZE;
1074 		}
1075 	}
1076 	return (error);
1077 }
1078 
1079 /*
1080  * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1081  * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1082  * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1083  * into args and call vn_io_fault1() to handle faults during the user
1084  * mode buffer accesses.
1085  */
1086 static int
1087 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1088     struct thread *td)
1089 {
1090 	vm_page_t ma[io_hold_cnt + 2];
1091 	struct uio *uio_clone, short_uio;
1092 	struct iovec short_iovec[1];
1093 	vm_page_t *prev_td_ma;
1094 	vm_prot_t prot;
1095 	vm_offset_t addr, end;
1096 	size_t len, resid;
1097 	ssize_t adv;
1098 	int error, cnt, save, saveheld, prev_td_ma_cnt;
1099 
1100 	if (vn_io_fault_prefault) {
1101 		error = vn_io_fault_prefault_user(uio);
1102 		if (error != 0)
1103 			return (error); /* Or ignore ? */
1104 	}
1105 
1106 	prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1107 
1108 	/*
1109 	 * The UFS follows IO_UNIT directive and replays back both
1110 	 * uio_offset and uio_resid if an error is encountered during the
1111 	 * operation.  But, since the iovec may be already advanced,
1112 	 * uio is still in an inconsistent state.
1113 	 *
1114 	 * Cache a copy of the original uio, which is advanced to the redo
1115 	 * point using UIO_NOCOPY below.
1116 	 */
1117 	uio_clone = cloneuio(uio);
1118 	resid = uio->uio_resid;
1119 
1120 	short_uio.uio_segflg = UIO_USERSPACE;
1121 	short_uio.uio_rw = uio->uio_rw;
1122 	short_uio.uio_td = uio->uio_td;
1123 
1124 	save = vm_fault_disable_pagefaults();
1125 	error = vn_io_fault_doio(args, uio, td);
1126 	if (error != EFAULT)
1127 		goto out;
1128 
1129 	atomic_add_long(&vn_io_faults_cnt, 1);
1130 	uio_clone->uio_segflg = UIO_NOCOPY;
1131 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
1132 	uio_clone->uio_segflg = uio->uio_segflg;
1133 
1134 	saveheld = curthread_pflags_set(TDP_UIOHELD);
1135 	prev_td_ma = td->td_ma;
1136 	prev_td_ma_cnt = td->td_ma_cnt;
1137 
1138 	while (uio_clone->uio_resid != 0) {
1139 		len = uio_clone->uio_iov->iov_len;
1140 		if (len == 0) {
1141 			KASSERT(uio_clone->uio_iovcnt >= 1,
1142 			    ("iovcnt underflow"));
1143 			uio_clone->uio_iov++;
1144 			uio_clone->uio_iovcnt--;
1145 			continue;
1146 		}
1147 		if (len > io_hold_cnt * PAGE_SIZE)
1148 			len = io_hold_cnt * PAGE_SIZE;
1149 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1150 		end = round_page(addr + len);
1151 		if (end < addr) {
1152 			error = EFAULT;
1153 			break;
1154 		}
1155 		cnt = atop(end - trunc_page(addr));
1156 		/*
1157 		 * A perfectly misaligned address and length could cause
1158 		 * both the start and the end of the chunk to use partial
1159 		 * page.  +2 accounts for such a situation.
1160 		 */
1161 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1162 		    addr, len, prot, ma, io_hold_cnt + 2);
1163 		if (cnt == -1) {
1164 			error = EFAULT;
1165 			break;
1166 		}
1167 		short_uio.uio_iov = &short_iovec[0];
1168 		short_iovec[0].iov_base = (void *)addr;
1169 		short_uio.uio_iovcnt = 1;
1170 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1171 		short_uio.uio_offset = uio_clone->uio_offset;
1172 		td->td_ma = ma;
1173 		td->td_ma_cnt = cnt;
1174 
1175 		error = vn_io_fault_doio(args, &short_uio, td);
1176 		vm_page_unhold_pages(ma, cnt);
1177 		adv = len - short_uio.uio_resid;
1178 
1179 		uio_clone->uio_iov->iov_base =
1180 		    (char *)uio_clone->uio_iov->iov_base + adv;
1181 		uio_clone->uio_iov->iov_len -= adv;
1182 		uio_clone->uio_resid -= adv;
1183 		uio_clone->uio_offset += adv;
1184 
1185 		uio->uio_resid -= adv;
1186 		uio->uio_offset += adv;
1187 
1188 		if (error != 0 || adv == 0)
1189 			break;
1190 	}
1191 	td->td_ma = prev_td_ma;
1192 	td->td_ma_cnt = prev_td_ma_cnt;
1193 	curthread_pflags_restore(saveheld);
1194 out:
1195 	vm_fault_enable_pagefaults(save);
1196 	free(uio_clone, M_IOV);
1197 	return (error);
1198 }
1199 
1200 static int
1201 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1202     int flags, struct thread *td)
1203 {
1204 	fo_rdwr_t *doio;
1205 	struct vnode *vp;
1206 	void *rl_cookie;
1207 	struct vn_io_fault_args args;
1208 	int error;
1209 
1210 	doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1211 	vp = fp->f_vnode;
1212 	foffset_lock_uio(fp, uio, flags);
1213 	if (do_vn_io_fault(vp, uio)) {
1214 		args.kind = VN_IO_FAULT_FOP;
1215 		args.args.fop_args.fp = fp;
1216 		args.args.fop_args.doio = doio;
1217 		args.cred = active_cred;
1218 		args.flags = flags | FOF_OFFSET;
1219 		if (uio->uio_rw == UIO_READ) {
1220 			rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1221 			    uio->uio_offset + uio->uio_resid);
1222 		} else if ((fp->f_flag & O_APPEND) != 0 ||
1223 		    (flags & FOF_OFFSET) == 0) {
1224 			/* For appenders, punt and lock the whole range. */
1225 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1226 		} else {
1227 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1228 			    uio->uio_offset + uio->uio_resid);
1229 		}
1230 		error = vn_io_fault1(vp, uio, &args, td);
1231 		vn_rangelock_unlock(vp, rl_cookie);
1232 	} else {
1233 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1234 	}
1235 	foffset_unlock_uio(fp, uio, flags);
1236 	return (error);
1237 }
1238 
1239 /*
1240  * Helper function to perform the requested uiomove operation using
1241  * the held pages for io->uio_iov[0].iov_base buffer instead of
1242  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1243  * instead of iov_base prevents page faults that could occur due to
1244  * pmap_collect() invalidating the mapping created by
1245  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1246  * object cleanup revoking the write access from page mappings.
1247  *
1248  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1249  * instead of plain uiomove().
1250  */
1251 int
1252 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1253 {
1254 	struct uio transp_uio;
1255 	struct iovec transp_iov[1];
1256 	struct thread *td;
1257 	size_t adv;
1258 	int error, pgadv;
1259 
1260 	td = curthread;
1261 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1262 	    uio->uio_segflg != UIO_USERSPACE)
1263 		return (uiomove(data, xfersize, uio));
1264 
1265 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1266 	transp_iov[0].iov_base = data;
1267 	transp_uio.uio_iov = &transp_iov[0];
1268 	transp_uio.uio_iovcnt = 1;
1269 	if (xfersize > uio->uio_resid)
1270 		xfersize = uio->uio_resid;
1271 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1272 	transp_uio.uio_offset = 0;
1273 	transp_uio.uio_segflg = UIO_SYSSPACE;
1274 	/*
1275 	 * Since transp_iov points to data, and td_ma page array
1276 	 * corresponds to original uio->uio_iov, we need to invert the
1277 	 * direction of the i/o operation as passed to
1278 	 * uiomove_fromphys().
1279 	 */
1280 	switch (uio->uio_rw) {
1281 	case UIO_WRITE:
1282 		transp_uio.uio_rw = UIO_READ;
1283 		break;
1284 	case UIO_READ:
1285 		transp_uio.uio_rw = UIO_WRITE;
1286 		break;
1287 	}
1288 	transp_uio.uio_td = uio->uio_td;
1289 	error = uiomove_fromphys(td->td_ma,
1290 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1291 	    xfersize, &transp_uio);
1292 	adv = xfersize - transp_uio.uio_resid;
1293 	pgadv =
1294 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1295 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1296 	td->td_ma += pgadv;
1297 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1298 	    pgadv));
1299 	td->td_ma_cnt -= pgadv;
1300 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1301 	uio->uio_iov->iov_len -= adv;
1302 	uio->uio_resid -= adv;
1303 	uio->uio_offset += adv;
1304 	return (error);
1305 }
1306 
1307 int
1308 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1309     struct uio *uio)
1310 {
1311 	struct thread *td;
1312 	vm_offset_t iov_base;
1313 	int cnt, pgadv;
1314 
1315 	td = curthread;
1316 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1317 	    uio->uio_segflg != UIO_USERSPACE)
1318 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1319 
1320 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1321 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1322 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1323 	switch (uio->uio_rw) {
1324 	case UIO_WRITE:
1325 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1326 		    offset, cnt);
1327 		break;
1328 	case UIO_READ:
1329 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1330 		    cnt);
1331 		break;
1332 	}
1333 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1334 	td->td_ma += pgadv;
1335 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1336 	    pgadv));
1337 	td->td_ma_cnt -= pgadv;
1338 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1339 	uio->uio_iov->iov_len -= cnt;
1340 	uio->uio_resid -= cnt;
1341 	uio->uio_offset += cnt;
1342 	return (0);
1343 }
1344 
1345 
1346 /*
1347  * File table truncate routine.
1348  */
1349 static int
1350 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1351     struct thread *td)
1352 {
1353 	struct vattr vattr;
1354 	struct mount *mp;
1355 	struct vnode *vp;
1356 	void *rl_cookie;
1357 	int error;
1358 
1359 	vp = fp->f_vnode;
1360 
1361 	/*
1362 	 * Lock the whole range for truncation.  Otherwise split i/o
1363 	 * might happen partly before and partly after the truncation.
1364 	 */
1365 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1366 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1367 	if (error)
1368 		goto out1;
1369 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1370 	if (vp->v_type == VDIR) {
1371 		error = EISDIR;
1372 		goto out;
1373 	}
1374 #ifdef MAC
1375 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1376 	if (error)
1377 		goto out;
1378 #endif
1379 	error = vn_writechk(vp);
1380 	if (error == 0) {
1381 		VATTR_NULL(&vattr);
1382 		vattr.va_size = length;
1383 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1384 	}
1385 out:
1386 	VOP_UNLOCK(vp, 0);
1387 	vn_finished_write(mp);
1388 out1:
1389 	vn_rangelock_unlock(vp, rl_cookie);
1390 	return (error);
1391 }
1392 
1393 /*
1394  * File table vnode stat routine.
1395  */
1396 static int
1397 vn_statfile(fp, sb, active_cred, td)
1398 	struct file *fp;
1399 	struct stat *sb;
1400 	struct ucred *active_cred;
1401 	struct thread *td;
1402 {
1403 	struct vnode *vp = fp->f_vnode;
1404 	int error;
1405 
1406 	vn_lock(vp, LK_SHARED | LK_RETRY);
1407 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1408 	VOP_UNLOCK(vp, 0);
1409 
1410 	return (error);
1411 }
1412 
1413 /*
1414  * Stat a vnode; implementation for the stat syscall
1415  */
1416 int
1417 vn_stat(vp, sb, active_cred, file_cred, td)
1418 	struct vnode *vp;
1419 	register struct stat *sb;
1420 	struct ucred *active_cred;
1421 	struct ucred *file_cred;
1422 	struct thread *td;
1423 {
1424 	struct vattr vattr;
1425 	register struct vattr *vap;
1426 	int error;
1427 	u_short mode;
1428 
1429 #ifdef MAC
1430 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1431 	if (error)
1432 		return (error);
1433 #endif
1434 
1435 	vap = &vattr;
1436 
1437 	/*
1438 	 * Initialize defaults for new and unusual fields, so that file
1439 	 * systems which don't support these fields don't need to know
1440 	 * about them.
1441 	 */
1442 	vap->va_birthtime.tv_sec = -1;
1443 	vap->va_birthtime.tv_nsec = 0;
1444 	vap->va_fsid = VNOVAL;
1445 	vap->va_rdev = NODEV;
1446 
1447 	error = VOP_GETATTR(vp, vap, active_cred);
1448 	if (error)
1449 		return (error);
1450 
1451 	/*
1452 	 * Zero the spare stat fields
1453 	 */
1454 	bzero(sb, sizeof *sb);
1455 
1456 	/*
1457 	 * Copy from vattr table
1458 	 */
1459 	if (vap->va_fsid != VNOVAL)
1460 		sb->st_dev = vap->va_fsid;
1461 	else
1462 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1463 	sb->st_ino = vap->va_fileid;
1464 	mode = vap->va_mode;
1465 	switch (vap->va_type) {
1466 	case VREG:
1467 		mode |= S_IFREG;
1468 		break;
1469 	case VDIR:
1470 		mode |= S_IFDIR;
1471 		break;
1472 	case VBLK:
1473 		mode |= S_IFBLK;
1474 		break;
1475 	case VCHR:
1476 		mode |= S_IFCHR;
1477 		break;
1478 	case VLNK:
1479 		mode |= S_IFLNK;
1480 		break;
1481 	case VSOCK:
1482 		mode |= S_IFSOCK;
1483 		break;
1484 	case VFIFO:
1485 		mode |= S_IFIFO;
1486 		break;
1487 	default:
1488 		return (EBADF);
1489 	};
1490 	sb->st_mode = mode;
1491 	sb->st_nlink = vap->va_nlink;
1492 	sb->st_uid = vap->va_uid;
1493 	sb->st_gid = vap->va_gid;
1494 	sb->st_rdev = vap->va_rdev;
1495 	if (vap->va_size > OFF_MAX)
1496 		return (EOVERFLOW);
1497 	sb->st_size = vap->va_size;
1498 	sb->st_atim = vap->va_atime;
1499 	sb->st_mtim = vap->va_mtime;
1500 	sb->st_ctim = vap->va_ctime;
1501 	sb->st_birthtim = vap->va_birthtime;
1502 
1503         /*
1504 	 * According to www.opengroup.org, the meaning of st_blksize is
1505 	 *   "a filesystem-specific preferred I/O block size for this
1506 	 *    object.  In some filesystem types, this may vary from file
1507 	 *    to file"
1508 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1509 	 */
1510 
1511 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1512 
1513 	sb->st_flags = vap->va_flags;
1514 	if (priv_check(td, PRIV_VFS_GENERATION))
1515 		sb->st_gen = 0;
1516 	else
1517 		sb->st_gen = vap->va_gen;
1518 
1519 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1520 	return (0);
1521 }
1522 
1523 /*
1524  * File table vnode ioctl routine.
1525  */
1526 static int
1527 vn_ioctl(fp, com, data, active_cred, td)
1528 	struct file *fp;
1529 	u_long com;
1530 	void *data;
1531 	struct ucred *active_cred;
1532 	struct thread *td;
1533 {
1534 	struct vattr vattr;
1535 	struct vnode *vp;
1536 	int error;
1537 
1538 	vp = fp->f_vnode;
1539 	switch (vp->v_type) {
1540 	case VDIR:
1541 	case VREG:
1542 		switch (com) {
1543 		case FIONREAD:
1544 			vn_lock(vp, LK_SHARED | LK_RETRY);
1545 			error = VOP_GETATTR(vp, &vattr, active_cred);
1546 			VOP_UNLOCK(vp, 0);
1547 			if (error == 0)
1548 				*(int *)data = vattr.va_size - fp->f_offset;
1549 			return (error);
1550 		case FIONBIO:
1551 		case FIOASYNC:
1552 			return (0);
1553 		default:
1554 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1555 			    active_cred, td));
1556 		}
1557 	default:
1558 		return (ENOTTY);
1559 	}
1560 }
1561 
1562 /*
1563  * File table vnode poll routine.
1564  */
1565 static int
1566 vn_poll(fp, events, active_cred, td)
1567 	struct file *fp;
1568 	int events;
1569 	struct ucred *active_cred;
1570 	struct thread *td;
1571 {
1572 	struct vnode *vp;
1573 	int error;
1574 
1575 	vp = fp->f_vnode;
1576 #ifdef MAC
1577 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1578 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1579 	VOP_UNLOCK(vp, 0);
1580 	if (!error)
1581 #endif
1582 
1583 	error = VOP_POLL(vp, events, fp->f_cred, td);
1584 	return (error);
1585 }
1586 
1587 /*
1588  * Acquire the requested lock and then check for validity.  LK_RETRY
1589  * permits vn_lock to return doomed vnodes.
1590  */
1591 int
1592 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1593 {
1594 	int error;
1595 
1596 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1597 	    ("vn_lock called with no locktype."));
1598 	do {
1599 #ifdef DEBUG_VFS_LOCKS
1600 		KASSERT(vp->v_holdcnt != 0,
1601 		    ("vn_lock %p: zero hold count", vp));
1602 #endif
1603 		error = VOP_LOCK1(vp, flags, file, line);
1604 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1605 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1606 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1607 		    flags, error));
1608 		/*
1609 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1610 		 * If RETRY is not set, we return ENOENT instead.
1611 		 */
1612 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1613 		    (flags & LK_RETRY) == 0) {
1614 			VOP_UNLOCK(vp, 0);
1615 			error = ENOENT;
1616 			break;
1617 		}
1618 	} while (flags & LK_RETRY && error != 0);
1619 	return (error);
1620 }
1621 
1622 /*
1623  * File table vnode close routine.
1624  */
1625 static int
1626 vn_closefile(fp, td)
1627 	struct file *fp;
1628 	struct thread *td;
1629 {
1630 	struct vnode *vp;
1631 	struct flock lf;
1632 	int error;
1633 
1634 	vp = fp->f_vnode;
1635 	fp->f_ops = &badfileops;
1636 
1637 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1638 		vref(vp);
1639 
1640 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1641 
1642 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1643 		lf.l_whence = SEEK_SET;
1644 		lf.l_start = 0;
1645 		lf.l_len = 0;
1646 		lf.l_type = F_UNLCK;
1647 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1648 		vrele(vp);
1649 	}
1650 	return (error);
1651 }
1652 
1653 static bool
1654 vn_suspendable(struct mount *mp)
1655 {
1656 
1657 	return (mp->mnt_op->vfs_susp_clean != NULL);
1658 }
1659 
1660 /*
1661  * Preparing to start a filesystem write operation. If the operation is
1662  * permitted, then we bump the count of operations in progress and
1663  * proceed. If a suspend request is in progress, we wait until the
1664  * suspension is over, and then proceed.
1665  */
1666 static int
1667 vn_start_write_locked(struct mount *mp, int flags)
1668 {
1669 	int error, mflags;
1670 
1671 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1672 	error = 0;
1673 
1674 	/*
1675 	 * Check on status of suspension.
1676 	 */
1677 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1678 	    mp->mnt_susp_owner != curthread) {
1679 		mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1680 		    (flags & PCATCH) : 0) | (PUSER - 1);
1681 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1682 			if (flags & V_NOWAIT) {
1683 				error = EWOULDBLOCK;
1684 				goto unlock;
1685 			}
1686 			error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1687 			    "suspfs", 0);
1688 			if (error)
1689 				goto unlock;
1690 		}
1691 	}
1692 	if (flags & V_XSLEEP)
1693 		goto unlock;
1694 	mp->mnt_writeopcount++;
1695 unlock:
1696 	if (error != 0 || (flags & V_XSLEEP) != 0)
1697 		MNT_REL(mp);
1698 	MNT_IUNLOCK(mp);
1699 	return (error);
1700 }
1701 
1702 int
1703 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1704 {
1705 	struct mount *mp;
1706 	int error;
1707 
1708 	KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1709 	    ("V_MNTREF requires mp"));
1710 
1711 	error = 0;
1712 	/*
1713 	 * If a vnode is provided, get and return the mount point that
1714 	 * to which it will write.
1715 	 */
1716 	if (vp != NULL) {
1717 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1718 			*mpp = NULL;
1719 			if (error != EOPNOTSUPP)
1720 				return (error);
1721 			return (0);
1722 		}
1723 	}
1724 	if ((mp = *mpp) == NULL)
1725 		return (0);
1726 
1727 	if (!vn_suspendable(mp)) {
1728 		if (vp != NULL || (flags & V_MNTREF) != 0)
1729 			vfs_rel(mp);
1730 		return (0);
1731 	}
1732 
1733 	/*
1734 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1735 	 * a vfs_ref().
1736 	 * As long as a vnode is not provided we need to acquire a
1737 	 * refcount for the provided mountpoint too, in order to
1738 	 * emulate a vfs_ref().
1739 	 */
1740 	MNT_ILOCK(mp);
1741 	if (vp == NULL && (flags & V_MNTREF) == 0)
1742 		MNT_REF(mp);
1743 
1744 	return (vn_start_write_locked(mp, flags));
1745 }
1746 
1747 /*
1748  * Secondary suspension. Used by operations such as vop_inactive
1749  * routines that are needed by the higher level functions. These
1750  * are allowed to proceed until all the higher level functions have
1751  * completed (indicated by mnt_writeopcount dropping to zero). At that
1752  * time, these operations are halted until the suspension is over.
1753  */
1754 int
1755 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1756 {
1757 	struct mount *mp;
1758 	int error;
1759 
1760 	KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1761 	    ("V_MNTREF requires mp"));
1762 
1763  retry:
1764 	if (vp != NULL) {
1765 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1766 			*mpp = NULL;
1767 			if (error != EOPNOTSUPP)
1768 				return (error);
1769 			return (0);
1770 		}
1771 	}
1772 	/*
1773 	 * If we are not suspended or have not yet reached suspended
1774 	 * mode, then let the operation proceed.
1775 	 */
1776 	if ((mp = *mpp) == NULL)
1777 		return (0);
1778 
1779 	if (!vn_suspendable(mp)) {
1780 		if (vp != NULL || (flags & V_MNTREF) != 0)
1781 			vfs_rel(mp);
1782 		return (0);
1783 	}
1784 
1785 	/*
1786 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1787 	 * a vfs_ref().
1788 	 * As long as a vnode is not provided we need to acquire a
1789 	 * refcount for the provided mountpoint too, in order to
1790 	 * emulate a vfs_ref().
1791 	 */
1792 	MNT_ILOCK(mp);
1793 	if (vp == NULL && (flags & V_MNTREF) == 0)
1794 		MNT_REF(mp);
1795 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1796 		mp->mnt_secondary_writes++;
1797 		mp->mnt_secondary_accwrites++;
1798 		MNT_IUNLOCK(mp);
1799 		return (0);
1800 	}
1801 	if (flags & V_NOWAIT) {
1802 		MNT_REL(mp);
1803 		MNT_IUNLOCK(mp);
1804 		return (EWOULDBLOCK);
1805 	}
1806 	/*
1807 	 * Wait for the suspension to finish.
1808 	 */
1809 	error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1810 	    ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1811 	    "suspfs", 0);
1812 	vfs_rel(mp);
1813 	if (error == 0)
1814 		goto retry;
1815 	return (error);
1816 }
1817 
1818 /*
1819  * Filesystem write operation has completed. If we are suspending and this
1820  * operation is the last one, notify the suspender that the suspension is
1821  * now in effect.
1822  */
1823 void
1824 vn_finished_write(mp)
1825 	struct mount *mp;
1826 {
1827 	if (mp == NULL || !vn_suspendable(mp))
1828 		return;
1829 	MNT_ILOCK(mp);
1830 	MNT_REL(mp);
1831 	mp->mnt_writeopcount--;
1832 	if (mp->mnt_writeopcount < 0)
1833 		panic("vn_finished_write: neg cnt");
1834 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1835 	    mp->mnt_writeopcount <= 0)
1836 		wakeup(&mp->mnt_writeopcount);
1837 	MNT_IUNLOCK(mp);
1838 }
1839 
1840 
1841 /*
1842  * Filesystem secondary write operation has completed. If we are
1843  * suspending and this operation is the last one, notify the suspender
1844  * that the suspension is now in effect.
1845  */
1846 void
1847 vn_finished_secondary_write(mp)
1848 	struct mount *mp;
1849 {
1850 	if (mp == NULL || !vn_suspendable(mp))
1851 		return;
1852 	MNT_ILOCK(mp);
1853 	MNT_REL(mp);
1854 	mp->mnt_secondary_writes--;
1855 	if (mp->mnt_secondary_writes < 0)
1856 		panic("vn_finished_secondary_write: neg cnt");
1857 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1858 	    mp->mnt_secondary_writes <= 0)
1859 		wakeup(&mp->mnt_secondary_writes);
1860 	MNT_IUNLOCK(mp);
1861 }
1862 
1863 
1864 
1865 /*
1866  * Request a filesystem to suspend write operations.
1867  */
1868 int
1869 vfs_write_suspend(struct mount *mp, int flags)
1870 {
1871 	int error;
1872 
1873 	MPASS(vn_suspendable(mp));
1874 
1875 	MNT_ILOCK(mp);
1876 	if (mp->mnt_susp_owner == curthread) {
1877 		MNT_IUNLOCK(mp);
1878 		return (EALREADY);
1879 	}
1880 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1881 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1882 
1883 	/*
1884 	 * Unmount holds a write reference on the mount point.  If we
1885 	 * own busy reference and drain for writers, we deadlock with
1886 	 * the reference draining in the unmount path.  Callers of
1887 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1888 	 * vfs_busy() reference is owned and caller is not in the
1889 	 * unmount context.
1890 	 */
1891 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1892 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1893 		MNT_IUNLOCK(mp);
1894 		return (EBUSY);
1895 	}
1896 
1897 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1898 	mp->mnt_susp_owner = curthread;
1899 	if (mp->mnt_writeopcount > 0)
1900 		(void) msleep(&mp->mnt_writeopcount,
1901 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1902 	else
1903 		MNT_IUNLOCK(mp);
1904 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1905 		vfs_write_resume(mp, 0);
1906 	return (error);
1907 }
1908 
1909 /*
1910  * Request a filesystem to resume write operations.
1911  */
1912 void
1913 vfs_write_resume(struct mount *mp, int flags)
1914 {
1915 
1916 	MPASS(vn_suspendable(mp));
1917 
1918 	MNT_ILOCK(mp);
1919 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1920 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1921 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1922 				       MNTK_SUSPENDED);
1923 		mp->mnt_susp_owner = NULL;
1924 		wakeup(&mp->mnt_writeopcount);
1925 		wakeup(&mp->mnt_flag);
1926 		curthread->td_pflags &= ~TDP_IGNSUSP;
1927 		if ((flags & VR_START_WRITE) != 0) {
1928 			MNT_REF(mp);
1929 			mp->mnt_writeopcount++;
1930 		}
1931 		MNT_IUNLOCK(mp);
1932 		if ((flags & VR_NO_SUSPCLR) == 0)
1933 			VFS_SUSP_CLEAN(mp);
1934 	} else if ((flags & VR_START_WRITE) != 0) {
1935 		MNT_REF(mp);
1936 		vn_start_write_locked(mp, 0);
1937 	} else {
1938 		MNT_IUNLOCK(mp);
1939 	}
1940 }
1941 
1942 /*
1943  * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1944  * methods.
1945  */
1946 int
1947 vfs_write_suspend_umnt(struct mount *mp)
1948 {
1949 	int error;
1950 
1951 	MPASS(vn_suspendable(mp));
1952 	KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1953 	    ("vfs_write_suspend_umnt: recursed"));
1954 
1955 	/* dounmount() already called vn_start_write(). */
1956 	for (;;) {
1957 		vn_finished_write(mp);
1958 		error = vfs_write_suspend(mp, 0);
1959 		if (error != 0) {
1960 			vn_start_write(NULL, &mp, V_WAIT);
1961 			return (error);
1962 		}
1963 		MNT_ILOCK(mp);
1964 		if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1965 			break;
1966 		MNT_IUNLOCK(mp);
1967 		vn_start_write(NULL, &mp, V_WAIT);
1968 	}
1969 	mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1970 	wakeup(&mp->mnt_flag);
1971 	MNT_IUNLOCK(mp);
1972 	curthread->td_pflags |= TDP_IGNSUSP;
1973 	return (0);
1974 }
1975 
1976 /*
1977  * Implement kqueues for files by translating it to vnode operation.
1978  */
1979 static int
1980 vn_kqfilter(struct file *fp, struct knote *kn)
1981 {
1982 
1983 	return (VOP_KQFILTER(fp->f_vnode, kn));
1984 }
1985 
1986 /*
1987  * Simplified in-kernel wrapper calls for extended attribute access.
1988  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1989  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1990  */
1991 int
1992 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1993     const char *attrname, int *buflen, char *buf, struct thread *td)
1994 {
1995 	struct uio	auio;
1996 	struct iovec	iov;
1997 	int	error;
1998 
1999 	iov.iov_len = *buflen;
2000 	iov.iov_base = buf;
2001 
2002 	auio.uio_iov = &iov;
2003 	auio.uio_iovcnt = 1;
2004 	auio.uio_rw = UIO_READ;
2005 	auio.uio_segflg = UIO_SYSSPACE;
2006 	auio.uio_td = td;
2007 	auio.uio_offset = 0;
2008 	auio.uio_resid = *buflen;
2009 
2010 	if ((ioflg & IO_NODELOCKED) == 0)
2011 		vn_lock(vp, LK_SHARED | LK_RETRY);
2012 
2013 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2014 
2015 	/* authorize attribute retrieval as kernel */
2016 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
2017 	    td);
2018 
2019 	if ((ioflg & IO_NODELOCKED) == 0)
2020 		VOP_UNLOCK(vp, 0);
2021 
2022 	if (error == 0) {
2023 		*buflen = *buflen - auio.uio_resid;
2024 	}
2025 
2026 	return (error);
2027 }
2028 
2029 /*
2030  * XXX failure mode if partially written?
2031  */
2032 int
2033 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
2034     const char *attrname, int buflen, char *buf, struct thread *td)
2035 {
2036 	struct uio	auio;
2037 	struct iovec	iov;
2038 	struct mount	*mp;
2039 	int	error;
2040 
2041 	iov.iov_len = buflen;
2042 	iov.iov_base = buf;
2043 
2044 	auio.uio_iov = &iov;
2045 	auio.uio_iovcnt = 1;
2046 	auio.uio_rw = UIO_WRITE;
2047 	auio.uio_segflg = UIO_SYSSPACE;
2048 	auio.uio_td = td;
2049 	auio.uio_offset = 0;
2050 	auio.uio_resid = buflen;
2051 
2052 	if ((ioflg & IO_NODELOCKED) == 0) {
2053 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2054 			return (error);
2055 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2056 	}
2057 
2058 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2059 
2060 	/* authorize attribute setting as kernel */
2061 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2062 
2063 	if ((ioflg & IO_NODELOCKED) == 0) {
2064 		vn_finished_write(mp);
2065 		VOP_UNLOCK(vp, 0);
2066 	}
2067 
2068 	return (error);
2069 }
2070 
2071 int
2072 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2073     const char *attrname, struct thread *td)
2074 {
2075 	struct mount	*mp;
2076 	int	error;
2077 
2078 	if ((ioflg & IO_NODELOCKED) == 0) {
2079 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2080 			return (error);
2081 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2082 	}
2083 
2084 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2085 
2086 	/* authorize attribute removal as kernel */
2087 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2088 	if (error == EOPNOTSUPP)
2089 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2090 		    NULL, td);
2091 
2092 	if ((ioflg & IO_NODELOCKED) == 0) {
2093 		vn_finished_write(mp);
2094 		VOP_UNLOCK(vp, 0);
2095 	}
2096 
2097 	return (error);
2098 }
2099 
2100 static int
2101 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2102     struct vnode **rvp)
2103 {
2104 
2105 	return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2106 }
2107 
2108 int
2109 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2110 {
2111 
2112 	return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2113 	    lkflags, rvp));
2114 }
2115 
2116 int
2117 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2118     int lkflags, struct vnode **rvp)
2119 {
2120 	struct mount *mp;
2121 	int ltype, error;
2122 
2123 	ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2124 	mp = vp->v_mount;
2125 	ltype = VOP_ISLOCKED(vp);
2126 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2127 	    ("vn_vget_ino: vp not locked"));
2128 	error = vfs_busy(mp, MBF_NOWAIT);
2129 	if (error != 0) {
2130 		vfs_ref(mp);
2131 		VOP_UNLOCK(vp, 0);
2132 		error = vfs_busy(mp, 0);
2133 		vn_lock(vp, ltype | LK_RETRY);
2134 		vfs_rel(mp);
2135 		if (error != 0)
2136 			return (ENOENT);
2137 		if (vp->v_iflag & VI_DOOMED) {
2138 			vfs_unbusy(mp);
2139 			return (ENOENT);
2140 		}
2141 	}
2142 	VOP_UNLOCK(vp, 0);
2143 	error = alloc(mp, alloc_arg, lkflags, rvp);
2144 	vfs_unbusy(mp);
2145 	if (*rvp != vp)
2146 		vn_lock(vp, ltype | LK_RETRY);
2147 	if (vp->v_iflag & VI_DOOMED) {
2148 		if (error == 0) {
2149 			if (*rvp == vp)
2150 				vunref(vp);
2151 			else
2152 				vput(*rvp);
2153 		}
2154 		error = ENOENT;
2155 	}
2156 	return (error);
2157 }
2158 
2159 int
2160 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2161     struct thread *td)
2162 {
2163 
2164 	if (vp->v_type != VREG || td == NULL)
2165 		return (0);
2166 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
2167 	    lim_cur(td, RLIMIT_FSIZE)) {
2168 		PROC_LOCK(td->td_proc);
2169 		kern_psignal(td->td_proc, SIGXFSZ);
2170 		PROC_UNLOCK(td->td_proc);
2171 		return (EFBIG);
2172 	}
2173 	return (0);
2174 }
2175 
2176 int
2177 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2178     struct thread *td)
2179 {
2180 	struct vnode *vp;
2181 
2182 	vp = fp->f_vnode;
2183 #ifdef AUDIT
2184 	vn_lock(vp, LK_SHARED | LK_RETRY);
2185 	AUDIT_ARG_VNODE1(vp);
2186 	VOP_UNLOCK(vp, 0);
2187 #endif
2188 	return (setfmode(td, active_cred, vp, mode));
2189 }
2190 
2191 int
2192 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2193     struct thread *td)
2194 {
2195 	struct vnode *vp;
2196 
2197 	vp = fp->f_vnode;
2198 #ifdef AUDIT
2199 	vn_lock(vp, LK_SHARED | LK_RETRY);
2200 	AUDIT_ARG_VNODE1(vp);
2201 	VOP_UNLOCK(vp, 0);
2202 #endif
2203 	return (setfown(td, active_cred, vp, uid, gid));
2204 }
2205 
2206 void
2207 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2208 {
2209 	vm_object_t object;
2210 
2211 	if ((object = vp->v_object) == NULL)
2212 		return;
2213 	VM_OBJECT_WLOCK(object);
2214 	vm_object_page_remove(object, start, end, 0);
2215 	VM_OBJECT_WUNLOCK(object);
2216 }
2217 
2218 int
2219 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2220 {
2221 	struct vattr va;
2222 	daddr_t bn, bnp;
2223 	uint64_t bsize;
2224 	off_t noff;
2225 	int error;
2226 
2227 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2228 	    ("Wrong command %lu", cmd));
2229 
2230 	if (vn_lock(vp, LK_SHARED) != 0)
2231 		return (EBADF);
2232 	if (vp->v_type != VREG) {
2233 		error = ENOTTY;
2234 		goto unlock;
2235 	}
2236 	error = VOP_GETATTR(vp, &va, cred);
2237 	if (error != 0)
2238 		goto unlock;
2239 	noff = *off;
2240 	if (noff >= va.va_size) {
2241 		error = ENXIO;
2242 		goto unlock;
2243 	}
2244 	bsize = vp->v_mount->mnt_stat.f_iosize;
2245 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2246 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2247 		if (error == EOPNOTSUPP) {
2248 			error = ENOTTY;
2249 			goto unlock;
2250 		}
2251 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2252 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
2253 			noff = bn * bsize;
2254 			if (noff < *off)
2255 				noff = *off;
2256 			goto unlock;
2257 		}
2258 	}
2259 	if (noff > va.va_size)
2260 		noff = va.va_size;
2261 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2262 	if (cmd == FIOSEEKDATA)
2263 		error = ENXIO;
2264 unlock:
2265 	VOP_UNLOCK(vp, 0);
2266 	if (error == 0)
2267 		*off = noff;
2268 	return (error);
2269 }
2270 
2271 int
2272 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2273 {
2274 	struct ucred *cred;
2275 	struct vnode *vp;
2276 	struct vattr vattr;
2277 	off_t foffset, size;
2278 	int error, noneg;
2279 
2280 	cred = td->td_ucred;
2281 	vp = fp->f_vnode;
2282 	foffset = foffset_lock(fp, 0);
2283 	noneg = (vp->v_type != VCHR);
2284 	error = 0;
2285 	switch (whence) {
2286 	case L_INCR:
2287 		if (noneg &&
2288 		    (foffset < 0 ||
2289 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2290 			error = EOVERFLOW;
2291 			break;
2292 		}
2293 		offset += foffset;
2294 		break;
2295 	case L_XTND:
2296 		vn_lock(vp, LK_SHARED | LK_RETRY);
2297 		error = VOP_GETATTR(vp, &vattr, cred);
2298 		VOP_UNLOCK(vp, 0);
2299 		if (error)
2300 			break;
2301 
2302 		/*
2303 		 * If the file references a disk device, then fetch
2304 		 * the media size and use that to determine the ending
2305 		 * offset.
2306 		 */
2307 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2308 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2309 			vattr.va_size = size;
2310 		if (noneg &&
2311 		    (vattr.va_size > OFF_MAX ||
2312 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2313 			error = EOVERFLOW;
2314 			break;
2315 		}
2316 		offset += vattr.va_size;
2317 		break;
2318 	case L_SET:
2319 		break;
2320 	case SEEK_DATA:
2321 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2322 		break;
2323 	case SEEK_HOLE:
2324 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2325 		break;
2326 	default:
2327 		error = EINVAL;
2328 	}
2329 	if (error == 0 && noneg && offset < 0)
2330 		error = EINVAL;
2331 	if (error != 0)
2332 		goto drop;
2333 	VFS_KNOTE_UNLOCKED(vp, 0);
2334 	td->td_uretoff.tdu_off = offset;
2335 drop:
2336 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2337 	return (error);
2338 }
2339 
2340 int
2341 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2342     struct thread *td)
2343 {
2344 	int error;
2345 
2346 	/*
2347 	 * Grant permission if the caller is the owner of the file, or
2348 	 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2349 	 * on the file.  If the time pointer is null, then write
2350 	 * permission on the file is also sufficient.
2351 	 *
2352 	 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2353 	 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2354 	 * will be allowed to set the times [..] to the current
2355 	 * server time.
2356 	 */
2357 	error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2358 	if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2359 		error = VOP_ACCESS(vp, VWRITE, cred, td);
2360 	return (error);
2361 }
2362 
2363 int
2364 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2365 {
2366 	struct vnode *vp;
2367 	int error;
2368 
2369 	if (fp->f_type == DTYPE_FIFO)
2370 		kif->kf_type = KF_TYPE_FIFO;
2371 	else
2372 		kif->kf_type = KF_TYPE_VNODE;
2373 	vp = fp->f_vnode;
2374 	vref(vp);
2375 	FILEDESC_SUNLOCK(fdp);
2376 	error = vn_fill_kinfo_vnode(vp, kif);
2377 	vrele(vp);
2378 	FILEDESC_SLOCK(fdp);
2379 	return (error);
2380 }
2381 
2382 int
2383 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2384 {
2385 	struct vattr va;
2386 	char *fullpath, *freepath;
2387 	int error;
2388 
2389 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2390 	freepath = NULL;
2391 	fullpath = "-";
2392 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2393 	if (error == 0) {
2394 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2395 	}
2396 	if (freepath != NULL)
2397 		free(freepath, M_TEMP);
2398 
2399 	/*
2400 	 * Retrieve vnode attributes.
2401 	 */
2402 	va.va_fsid = VNOVAL;
2403 	va.va_rdev = NODEV;
2404 	vn_lock(vp, LK_SHARED | LK_RETRY);
2405 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2406 	VOP_UNLOCK(vp, 0);
2407 	if (error != 0)
2408 		return (error);
2409 	if (va.va_fsid != VNOVAL)
2410 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2411 	else
2412 		kif->kf_un.kf_file.kf_file_fsid =
2413 		    vp->v_mount->mnt_stat.f_fsid.val[0];
2414 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2415 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2416 	kif->kf_un.kf_file.kf_file_size = va.va_size;
2417 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2418 	return (0);
2419 }
2420 
2421 int
2422 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2423     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2424     struct thread *td)
2425 {
2426 #ifdef HWPMC_HOOKS
2427 	struct pmckern_map_in pkm;
2428 #endif
2429 	struct mount *mp;
2430 	struct vnode *vp;
2431 	vm_object_t object;
2432 	vm_prot_t maxprot;
2433 	boolean_t writecounted;
2434 	int error;
2435 
2436 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2437     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2438 	/*
2439 	 * POSIX shared-memory objects are defined to have
2440 	 * kernel persistence, and are not defined to support
2441 	 * read(2)/write(2) -- or even open(2).  Thus, we can
2442 	 * use MAP_ASYNC to trade on-disk coherence for speed.
2443 	 * The shm_open(3) library routine turns on the FPOSIXSHM
2444 	 * flag to request this behavior.
2445 	 */
2446 	if ((fp->f_flag & FPOSIXSHM) != 0)
2447 		flags |= MAP_NOSYNC;
2448 #endif
2449 	vp = fp->f_vnode;
2450 
2451 	/*
2452 	 * Ensure that file and memory protections are
2453 	 * compatible.  Note that we only worry about
2454 	 * writability if mapping is shared; in this case,
2455 	 * current and max prot are dictated by the open file.
2456 	 * XXX use the vnode instead?  Problem is: what
2457 	 * credentials do we use for determination? What if
2458 	 * proc does a setuid?
2459 	 */
2460 	mp = vp->v_mount;
2461 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0)
2462 		maxprot = VM_PROT_NONE;
2463 	else
2464 		maxprot = VM_PROT_EXECUTE;
2465 	if ((fp->f_flag & FREAD) != 0)
2466 		maxprot |= VM_PROT_READ;
2467 	else if ((prot & VM_PROT_READ) != 0)
2468 		return (EACCES);
2469 
2470 	/*
2471 	 * If we are sharing potential changes via MAP_SHARED and we
2472 	 * are trying to get write permission although we opened it
2473 	 * without asking for it, bail out.
2474 	 */
2475 	if ((flags & MAP_SHARED) != 0) {
2476 		if ((fp->f_flag & FWRITE) != 0)
2477 			maxprot |= VM_PROT_WRITE;
2478 		else if ((prot & VM_PROT_WRITE) != 0)
2479 			return (EACCES);
2480 	} else {
2481 		maxprot |= VM_PROT_WRITE;
2482 		cap_maxprot |= VM_PROT_WRITE;
2483 	}
2484 	maxprot &= cap_maxprot;
2485 
2486 	writecounted = FALSE;
2487 	error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2488 	    &foff, &object, &writecounted);
2489 	if (error != 0)
2490 		return (error);
2491 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2492 	    foff, writecounted, td);
2493 	if (error != 0) {
2494 		/*
2495 		 * If this mapping was accounted for in the vnode's
2496 		 * writecount, then undo that now.
2497 		 */
2498 		if (writecounted)
2499 			vnode_pager_release_writecount(object, 0, size);
2500 		vm_object_deallocate(object);
2501 	}
2502 #ifdef HWPMC_HOOKS
2503 	/* Inform hwpmc(4) if an executable is being mapped. */
2504 	if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) {
2505 		pkm.pm_file = vp;
2506 		pkm.pm_address = (uintptr_t) addr;
2507 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
2508 	}
2509 #endif
2510 	return (error);
2511 }
2512