xref: /freebsd/sys/kern/vfs_vnops.c (revision 193d9e768ba63fcfb187cfd17f461f7d41345048)
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  * 3. 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 "opt_hwpmc_hooks.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/disk.h>
51 #include <sys/fail.h>
52 #include <sys/fcntl.h>
53 #include <sys/file.h>
54 #include <sys/kdb.h>
55 #include <sys/stat.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/limits.h>
59 #include <sys/lock.h>
60 #include <sys/mman.h>
61 #include <sys/mount.h>
62 #include <sys/mutex.h>
63 #include <sys/namei.h>
64 #include <sys/vnode.h>
65 #include <sys/bio.h>
66 #include <sys/buf.h>
67 #include <sys/filio.h>
68 #include <sys/resourcevar.h>
69 #include <sys/rwlock.h>
70 #include <sys/sx.h>
71 #include <sys/sysctl.h>
72 #include <sys/ttycom.h>
73 #include <sys/conf.h>
74 #include <sys/syslog.h>
75 #include <sys/unistd.h>
76 #include <sys/user.h>
77 
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80 
81 #include <vm/vm.h>
82 #include <vm/vm_extern.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vnode_pager.h>
88 
89 #ifdef HWPMC_HOOKS
90 #include <sys/pmckern.h>
91 #endif
92 
93 static fo_rdwr_t	vn_read;
94 static fo_rdwr_t	vn_write;
95 static fo_rdwr_t	vn_io_fault;
96 static fo_truncate_t	vn_truncate;
97 static fo_ioctl_t	vn_ioctl;
98 static fo_poll_t	vn_poll;
99 static fo_kqfilter_t	vn_kqfilter;
100 static fo_stat_t	vn_statfile;
101 static fo_close_t	vn_closefile;
102 static fo_mmap_t	vn_mmap;
103 
104 struct 	fileops vnops = {
105 	.fo_read = vn_io_fault,
106 	.fo_write = vn_io_fault,
107 	.fo_truncate = vn_truncate,
108 	.fo_ioctl = vn_ioctl,
109 	.fo_poll = vn_poll,
110 	.fo_kqfilter = vn_kqfilter,
111 	.fo_stat = vn_statfile,
112 	.fo_close = vn_closefile,
113 	.fo_chmod = vn_chmod,
114 	.fo_chown = vn_chown,
115 	.fo_sendfile = vn_sendfile,
116 	.fo_seek = vn_seek,
117 	.fo_fill_kinfo = vn_fill_kinfo,
118 	.fo_mmap = vn_mmap,
119 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
120 };
121 
122 static const int io_hold_cnt = 16;
123 static int vn_io_fault_enable = 1;
124 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
125     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
126 static int vn_io_fault_prefault = 0;
127 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
128     &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
129 static u_long vn_io_faults_cnt;
130 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
131     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
132 
133 /*
134  * Returns true if vn_io_fault mode of handling the i/o request should
135  * be used.
136  */
137 static bool
138 do_vn_io_fault(struct vnode *vp, struct uio *uio)
139 {
140 	struct mount *mp;
141 
142 	return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
143 	    (mp = vp->v_mount) != NULL &&
144 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
145 }
146 
147 /*
148  * Structure used to pass arguments to vn_io_fault1(), to do either
149  * file- or vnode-based I/O calls.
150  */
151 struct vn_io_fault_args {
152 	enum {
153 		VN_IO_FAULT_FOP,
154 		VN_IO_FAULT_VOP
155 	} kind;
156 	struct ucred *cred;
157 	int flags;
158 	union {
159 		struct fop_args_tag {
160 			struct file *fp;
161 			fo_rdwr_t *doio;
162 		} fop_args;
163 		struct vop_args_tag {
164 			struct vnode *vp;
165 		} vop_args;
166 	} args;
167 };
168 
169 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
170     struct vn_io_fault_args *args, struct thread *td);
171 
172 int
173 vn_open(ndp, flagp, cmode, fp)
174 	struct nameidata *ndp;
175 	int *flagp, cmode;
176 	struct file *fp;
177 {
178 	struct thread *td = ndp->ni_cnd.cn_thread;
179 
180 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
181 }
182 
183 /*
184  * Common code for vnode open operations via a name lookup.
185  * Lookup the vnode and invoke VOP_CREATE if needed.
186  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
187  *
188  * Note that this does NOT free nameidata for the successful case,
189  * due to the NDINIT being done elsewhere.
190  */
191 int
192 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
193     struct ucred *cred, struct file *fp)
194 {
195 	struct vnode *vp;
196 	struct mount *mp;
197 	struct thread *td = ndp->ni_cnd.cn_thread;
198 	struct vattr vat;
199 	struct vattr *vap = &vat;
200 	int fmode, error;
201 
202 restart:
203 	fmode = *flagp;
204 	if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
205 	    O_EXCL | O_DIRECTORY))
206 		return (EINVAL);
207 	else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
208 		ndp->ni_cnd.cn_nameiop = CREATE;
209 		/*
210 		 * Set NOCACHE to avoid flushing the cache when
211 		 * rolling in many files at once.
212 		*/
213 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
214 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
215 			ndp->ni_cnd.cn_flags |= FOLLOW;
216 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
217 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
218 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
219 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
220 		bwillwrite();
221 		if ((error = namei(ndp)) != 0)
222 			return (error);
223 		if (ndp->ni_vp == NULL) {
224 			VATTR_NULL(vap);
225 			vap->va_type = VREG;
226 			vap->va_mode = cmode;
227 			if (fmode & O_EXCL)
228 				vap->va_vaflags |= VA_EXCLUSIVE;
229 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
230 				NDFREE(ndp, NDF_ONLY_PNBUF);
231 				vput(ndp->ni_dvp);
232 				if ((error = vn_start_write(NULL, &mp,
233 				    V_XSLEEP | PCATCH)) != 0)
234 					return (error);
235 				goto restart;
236 			}
237 			if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
238 				ndp->ni_cnd.cn_flags |= MAKEENTRY;
239 #ifdef MAC
240 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
241 			    &ndp->ni_cnd, vap);
242 			if (error == 0)
243 #endif
244 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
245 						   &ndp->ni_cnd, vap);
246 			vput(ndp->ni_dvp);
247 			vn_finished_write(mp);
248 			if (error) {
249 				NDFREE(ndp, NDF_ONLY_PNBUF);
250 				return (error);
251 			}
252 			fmode &= ~O_TRUNC;
253 			vp = ndp->ni_vp;
254 		} else {
255 			if (ndp->ni_dvp == ndp->ni_vp)
256 				vrele(ndp->ni_dvp);
257 			else
258 				vput(ndp->ni_dvp);
259 			ndp->ni_dvp = NULL;
260 			vp = ndp->ni_vp;
261 			if (fmode & O_EXCL) {
262 				error = EEXIST;
263 				goto bad;
264 			}
265 			fmode &= ~O_CREAT;
266 		}
267 	} else {
268 		ndp->ni_cnd.cn_nameiop = LOOKUP;
269 		ndp->ni_cnd.cn_flags = ISOPEN |
270 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
271 		if (!(fmode & FWRITE))
272 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
273 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
274 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
275 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
276 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
277 		if ((error = namei(ndp)) != 0)
278 			return (error);
279 		vp = ndp->ni_vp;
280 	}
281 	error = vn_open_vnode(vp, fmode, cred, td, fp);
282 	if (error)
283 		goto bad;
284 	*flagp = fmode;
285 	return (0);
286 bad:
287 	NDFREE(ndp, NDF_ONLY_PNBUF);
288 	vput(vp);
289 	*flagp = fmode;
290 	ndp->ni_vp = NULL;
291 	return (error);
292 }
293 
294 /*
295  * Common code for vnode open operations once a vnode is located.
296  * Check permissions, and call the VOP_OPEN routine.
297  */
298 int
299 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
300     struct thread *td, struct file *fp)
301 {
302 	accmode_t accmode;
303 	struct flock lf;
304 	int error, lock_flags, type;
305 
306 	if (vp->v_type == VLNK)
307 		return (EMLINK);
308 	if (vp->v_type == VSOCK)
309 		return (EOPNOTSUPP);
310 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
311 		return (ENOTDIR);
312 	accmode = 0;
313 	if (fmode & (FWRITE | O_TRUNC)) {
314 		if (vp->v_type == VDIR)
315 			return (EISDIR);
316 		accmode |= VWRITE;
317 	}
318 	if (fmode & FREAD)
319 		accmode |= VREAD;
320 	if (fmode & FEXEC)
321 		accmode |= VEXEC;
322 	if ((fmode & O_APPEND) && (fmode & FWRITE))
323 		accmode |= VAPPEND;
324 #ifdef MAC
325 	if (fmode & O_CREAT)
326 		accmode |= VCREAT;
327 	if (fmode & O_VERIFY)
328 		accmode |= VVERIFY;
329 	error = mac_vnode_check_open(cred, vp, accmode);
330 	if (error)
331 		return (error);
332 
333 	accmode &= ~(VCREAT | VVERIFY);
334 #endif
335 	if ((fmode & O_CREAT) == 0) {
336 		if (accmode & VWRITE) {
337 			error = vn_writechk(vp);
338 			if (error)
339 				return (error);
340 		}
341 		if (accmode) {
342 		        error = VOP_ACCESS(vp, accmode, cred, td);
343 			if (error)
344 				return (error);
345 		}
346 	}
347 	if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
348 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
349 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
350 		return (error);
351 
352 	while ((fmode & (O_EXLOCK | O_SHLOCK)) != 0) {
353 		KASSERT(fp != NULL, ("open with flock requires fp"));
354 		if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) {
355 			error = EOPNOTSUPP;
356 			break;
357 		}
358 		lock_flags = VOP_ISLOCKED(vp);
359 		VOP_UNLOCK(vp, 0);
360 		lf.l_whence = SEEK_SET;
361 		lf.l_start = 0;
362 		lf.l_len = 0;
363 		if (fmode & O_EXLOCK)
364 			lf.l_type = F_WRLCK;
365 		else
366 			lf.l_type = F_RDLCK;
367 		type = F_FLOCK;
368 		if ((fmode & FNONBLOCK) == 0)
369 			type |= F_WAIT;
370 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
371 		if (error == 0)
372 			fp->f_flag |= FHASLOCK;
373 		vn_lock(vp, lock_flags | LK_RETRY);
374 		if (error != 0)
375 			break;
376 		if ((vp->v_iflag & VI_DOOMED) != 0) {
377 			error = ENOENT;
378 			break;
379 		}
380 
381 		/*
382 		 * Another thread might have used this vnode as an
383 		 * executable while the vnode lock was dropped.
384 		 * Ensure the vnode is still able to be opened for
385 		 * writing after the lock has been obtained.
386 		 */
387 		if ((accmode & VWRITE) != 0)
388 			error = vn_writechk(vp);
389 		break;
390 	}
391 
392 	if (error != 0) {
393 		fp->f_flag |= FOPENFAILED;
394 		fp->f_vnode = vp;
395 		if (fp->f_ops == &badfileops) {
396 			fp->f_type = DTYPE_VNODE;
397 			fp->f_ops = &vnops;
398 		}
399 		vref(vp);
400 	} else if  ((fmode & FWRITE) != 0) {
401 		VOP_ADD_WRITECOUNT(vp, 1);
402 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
403 		    __func__, vp, vp->v_writecount);
404 	}
405 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
406 	return (error);
407 }
408 
409 /*
410  * Check for write permissions on the specified vnode.
411  * Prototype text segments cannot be written.
412  */
413 int
414 vn_writechk(vp)
415 	register struct vnode *vp;
416 {
417 
418 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
419 	/*
420 	 * If there's shared text associated with
421 	 * the vnode, try to free it up once.  If
422 	 * we fail, we can't allow writing.
423 	 */
424 	if (VOP_IS_TEXT(vp))
425 		return (ETXTBSY);
426 
427 	return (0);
428 }
429 
430 /*
431  * Vnode close call
432  */
433 static int
434 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred,
435     struct thread *td, bool keep_ref)
436 {
437 	struct mount *mp;
438 	int error, lock_flags;
439 
440 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
441 	    MNT_EXTENDED_SHARED(vp->v_mount))
442 		lock_flags = LK_SHARED;
443 	else
444 		lock_flags = LK_EXCLUSIVE;
445 
446 	vn_start_write(vp, &mp, V_WAIT);
447 	vn_lock(vp, lock_flags | LK_RETRY);
448 	AUDIT_ARG_VNODE1(vp);
449 	if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
450 		VNASSERT(vp->v_writecount > 0, vp,
451 		    ("vn_close: negative writecount"));
452 		VOP_ADD_WRITECOUNT(vp, -1);
453 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
454 		    __func__, vp, vp->v_writecount);
455 	}
456 	error = VOP_CLOSE(vp, flags, file_cred, td);
457 	if (keep_ref)
458 		VOP_UNLOCK(vp, 0);
459 	else
460 		vput(vp);
461 	vn_finished_write(mp);
462 	return (error);
463 }
464 
465 int
466 vn_close(struct vnode *vp, int flags, struct ucred *file_cred,
467     struct thread *td)
468 {
469 
470 	return (vn_close1(vp, flags, file_cred, td, false));
471 }
472 
473 /*
474  * Heuristic to detect sequential operation.
475  */
476 static int
477 sequential_heuristic(struct uio *uio, struct file *fp)
478 {
479 
480 	ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
481 	if (fp->f_flag & FRDAHEAD)
482 		return (fp->f_seqcount << IO_SEQSHIFT);
483 
484 	/*
485 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
486 	 * that the first I/O is normally considered to be slightly
487 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
488 	 * unless previous seeks have reduced f_seqcount to 0, in which
489 	 * case offset 0 is not special.
490 	 */
491 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
492 	    uio->uio_offset == fp->f_nextoff) {
493 		/*
494 		 * f_seqcount is in units of fixed-size blocks so that it
495 		 * depends mainly on the amount of sequential I/O and not
496 		 * much on the number of sequential I/O's.  The fixed size
497 		 * of 16384 is hard-coded here since it is (not quite) just
498 		 * a magic size that works well here.  This size is more
499 		 * closely related to the best I/O size for real disks than
500 		 * to any block size used by software.
501 		 */
502 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
503 		if (fp->f_seqcount > IO_SEQMAX)
504 			fp->f_seqcount = IO_SEQMAX;
505 		return (fp->f_seqcount << IO_SEQSHIFT);
506 	}
507 
508 	/* Not sequential.  Quickly draw-down sequentiality. */
509 	if (fp->f_seqcount > 1)
510 		fp->f_seqcount = 1;
511 	else
512 		fp->f_seqcount = 0;
513 	return (0);
514 }
515 
516 /*
517  * Package up an I/O request on a vnode into a uio and do it.
518  */
519 int
520 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
521     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
522     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
523 {
524 	struct uio auio;
525 	struct iovec aiov;
526 	struct mount *mp;
527 	struct ucred *cred;
528 	void *rl_cookie;
529 	struct vn_io_fault_args args;
530 	int error, lock_flags;
531 
532 	auio.uio_iov = &aiov;
533 	auio.uio_iovcnt = 1;
534 	aiov.iov_base = base;
535 	aiov.iov_len = len;
536 	auio.uio_resid = len;
537 	auio.uio_offset = offset;
538 	auio.uio_segflg = segflg;
539 	auio.uio_rw = rw;
540 	auio.uio_td = td;
541 	error = 0;
542 
543 	if ((ioflg & IO_NODELOCKED) == 0) {
544 		if ((ioflg & IO_RANGELOCKED) == 0) {
545 			if (rw == UIO_READ) {
546 				rl_cookie = vn_rangelock_rlock(vp, offset,
547 				    offset + len);
548 			} else {
549 				rl_cookie = vn_rangelock_wlock(vp, offset,
550 				    offset + len);
551 			}
552 		} else
553 			rl_cookie = NULL;
554 		mp = NULL;
555 		if (rw == UIO_WRITE) {
556 			if (vp->v_type != VCHR &&
557 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
558 			    != 0)
559 				goto out;
560 			if (MNT_SHARED_WRITES(mp) ||
561 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
562 				lock_flags = LK_SHARED;
563 			else
564 				lock_flags = LK_EXCLUSIVE;
565 		} else
566 			lock_flags = LK_SHARED;
567 		vn_lock(vp, lock_flags | LK_RETRY);
568 	} else
569 		rl_cookie = NULL;
570 
571 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
572 #ifdef MAC
573 	if ((ioflg & IO_NOMACCHECK) == 0) {
574 		if (rw == UIO_READ)
575 			error = mac_vnode_check_read(active_cred, file_cred,
576 			    vp);
577 		else
578 			error = mac_vnode_check_write(active_cred, file_cred,
579 			    vp);
580 	}
581 #endif
582 	if (error == 0) {
583 		if (file_cred != NULL)
584 			cred = file_cred;
585 		else
586 			cred = active_cred;
587 		if (do_vn_io_fault(vp, &auio)) {
588 			args.kind = VN_IO_FAULT_VOP;
589 			args.cred = cred;
590 			args.flags = ioflg;
591 			args.args.vop_args.vp = vp;
592 			error = vn_io_fault1(vp, &auio, &args, td);
593 		} else if (rw == UIO_READ) {
594 			error = VOP_READ(vp, &auio, ioflg, cred);
595 		} else /* if (rw == UIO_WRITE) */ {
596 			error = VOP_WRITE(vp, &auio, ioflg, cred);
597 		}
598 	}
599 	if (aresid)
600 		*aresid = auio.uio_resid;
601 	else
602 		if (auio.uio_resid && error == 0)
603 			error = EIO;
604 	if ((ioflg & IO_NODELOCKED) == 0) {
605 		VOP_UNLOCK(vp, 0);
606 		if (mp != NULL)
607 			vn_finished_write(mp);
608 	}
609  out:
610 	if (rl_cookie != NULL)
611 		vn_rangelock_unlock(vp, rl_cookie);
612 	return (error);
613 }
614 
615 /*
616  * Package up an I/O request on a vnode into a uio and do it.  The I/O
617  * request is split up into smaller chunks and we try to avoid saturating
618  * the buffer cache while potentially holding a vnode locked, so we
619  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
620  * to give other processes a chance to lock the vnode (either other processes
621  * core'ing the same binary, or unrelated processes scanning the directory).
622  */
623 int
624 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
625     file_cred, aresid, td)
626 	enum uio_rw rw;
627 	struct vnode *vp;
628 	void *base;
629 	size_t len;
630 	off_t offset;
631 	enum uio_seg segflg;
632 	int ioflg;
633 	struct ucred *active_cred;
634 	struct ucred *file_cred;
635 	size_t *aresid;
636 	struct thread *td;
637 {
638 	int error = 0;
639 	ssize_t iaresid;
640 
641 	do {
642 		int chunk;
643 
644 		/*
645 		 * Force `offset' to a multiple of MAXBSIZE except possibly
646 		 * for the first chunk, so that filesystems only need to
647 		 * write full blocks except possibly for the first and last
648 		 * chunks.
649 		 */
650 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
651 
652 		if (chunk > len)
653 			chunk = len;
654 		if (rw != UIO_READ && vp->v_type == VREG)
655 			bwillwrite();
656 		iaresid = 0;
657 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
658 		    ioflg, active_cred, file_cred, &iaresid, td);
659 		len -= chunk;	/* aresid calc already includes length */
660 		if (error)
661 			break;
662 		offset += chunk;
663 		base = (char *)base + chunk;
664 		kern_yield(PRI_USER);
665 	} while (len);
666 	if (aresid)
667 		*aresid = len + iaresid;
668 	return (error);
669 }
670 
671 off_t
672 foffset_lock(struct file *fp, int flags)
673 {
674 	struct mtx *mtxp;
675 	off_t res;
676 
677 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
678 
679 #if OFF_MAX <= LONG_MAX
680 	/*
681 	 * Caller only wants the current f_offset value.  Assume that
682 	 * the long and shorter integer types reads are atomic.
683 	 */
684 	if ((flags & FOF_NOLOCK) != 0)
685 		return (fp->f_offset);
686 #endif
687 
688 	/*
689 	 * According to McKusick the vn lock was protecting f_offset here.
690 	 * It is now protected by the FOFFSET_LOCKED flag.
691 	 */
692 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
693 	mtx_lock(mtxp);
694 	if ((flags & FOF_NOLOCK) == 0) {
695 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
696 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
697 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
698 			    "vofflock", 0);
699 		}
700 		fp->f_vnread_flags |= FOFFSET_LOCKED;
701 	}
702 	res = fp->f_offset;
703 	mtx_unlock(mtxp);
704 	return (res);
705 }
706 
707 void
708 foffset_unlock(struct file *fp, off_t val, int flags)
709 {
710 	struct mtx *mtxp;
711 
712 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
713 
714 #if OFF_MAX <= LONG_MAX
715 	if ((flags & FOF_NOLOCK) != 0) {
716 		if ((flags & FOF_NOUPDATE) == 0)
717 			fp->f_offset = val;
718 		if ((flags & FOF_NEXTOFF) != 0)
719 			fp->f_nextoff = val;
720 		return;
721 	}
722 #endif
723 
724 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
725 	mtx_lock(mtxp);
726 	if ((flags & FOF_NOUPDATE) == 0)
727 		fp->f_offset = val;
728 	if ((flags & FOF_NEXTOFF) != 0)
729 		fp->f_nextoff = val;
730 	if ((flags & FOF_NOLOCK) == 0) {
731 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
732 		    ("Lost FOFFSET_LOCKED"));
733 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
734 			wakeup(&fp->f_vnread_flags);
735 		fp->f_vnread_flags = 0;
736 	}
737 	mtx_unlock(mtxp);
738 }
739 
740 void
741 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
742 {
743 
744 	if ((flags & FOF_OFFSET) == 0)
745 		uio->uio_offset = foffset_lock(fp, flags);
746 }
747 
748 void
749 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
750 {
751 
752 	if ((flags & FOF_OFFSET) == 0)
753 		foffset_unlock(fp, uio->uio_offset, flags);
754 }
755 
756 static int
757 get_advice(struct file *fp, struct uio *uio)
758 {
759 	struct mtx *mtxp;
760 	int ret;
761 
762 	ret = POSIX_FADV_NORMAL;
763 	if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
764 		return (ret);
765 
766 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
767 	mtx_lock(mtxp);
768 	if (fp->f_advice != NULL &&
769 	    uio->uio_offset >= fp->f_advice->fa_start &&
770 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
771 		ret = fp->f_advice->fa_advice;
772 	mtx_unlock(mtxp);
773 	return (ret);
774 }
775 
776 /*
777  * File table vnode read routine.
778  */
779 static int
780 vn_read(fp, uio, active_cred, flags, td)
781 	struct file *fp;
782 	struct uio *uio;
783 	struct ucred *active_cred;
784 	int flags;
785 	struct thread *td;
786 {
787 	struct vnode *vp;
788 	off_t orig_offset;
789 	int error, ioflag;
790 	int advice;
791 
792 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
793 	    uio->uio_td, td));
794 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
795 	vp = fp->f_vnode;
796 	ioflag = 0;
797 	if (fp->f_flag & FNONBLOCK)
798 		ioflag |= IO_NDELAY;
799 	if (fp->f_flag & O_DIRECT)
800 		ioflag |= IO_DIRECT;
801 	advice = get_advice(fp, uio);
802 	vn_lock(vp, LK_SHARED | LK_RETRY);
803 
804 	switch (advice) {
805 	case POSIX_FADV_NORMAL:
806 	case POSIX_FADV_SEQUENTIAL:
807 	case POSIX_FADV_NOREUSE:
808 		ioflag |= sequential_heuristic(uio, fp);
809 		break;
810 	case POSIX_FADV_RANDOM:
811 		/* Disable read-ahead for random I/O. */
812 		break;
813 	}
814 	orig_offset = uio->uio_offset;
815 
816 #ifdef MAC
817 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
818 	if (error == 0)
819 #endif
820 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
821 	fp->f_nextoff = uio->uio_offset;
822 	VOP_UNLOCK(vp, 0);
823 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
824 	    orig_offset != uio->uio_offset)
825 		/*
826 		 * Use POSIX_FADV_DONTNEED to flush pages and buffers
827 		 * for the backing file after a POSIX_FADV_NOREUSE
828 		 * read(2).
829 		 */
830 		error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
831 		    POSIX_FADV_DONTNEED);
832 	return (error);
833 }
834 
835 /*
836  * File table vnode write routine.
837  */
838 static int
839 vn_write(fp, uio, active_cred, flags, td)
840 	struct file *fp;
841 	struct uio *uio;
842 	struct ucred *active_cred;
843 	int flags;
844 	struct thread *td;
845 {
846 	struct vnode *vp;
847 	struct mount *mp;
848 	off_t orig_offset;
849 	int error, ioflag, lock_flags;
850 	int advice;
851 
852 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
853 	    uio->uio_td, td));
854 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
855 	vp = fp->f_vnode;
856 	if (vp->v_type == VREG)
857 		bwillwrite();
858 	ioflag = IO_UNIT;
859 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
860 		ioflag |= IO_APPEND;
861 	if (fp->f_flag & FNONBLOCK)
862 		ioflag |= IO_NDELAY;
863 	if (fp->f_flag & O_DIRECT)
864 		ioflag |= IO_DIRECT;
865 	if ((fp->f_flag & O_FSYNC) ||
866 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
867 		ioflag |= IO_SYNC;
868 	mp = NULL;
869 	if (vp->v_type != VCHR &&
870 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
871 		goto unlock;
872 
873 	advice = get_advice(fp, uio);
874 
875 	if (MNT_SHARED_WRITES(mp) ||
876 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
877 		lock_flags = LK_SHARED;
878 	} else {
879 		lock_flags = LK_EXCLUSIVE;
880 	}
881 
882 	vn_lock(vp, lock_flags | LK_RETRY);
883 	switch (advice) {
884 	case POSIX_FADV_NORMAL:
885 	case POSIX_FADV_SEQUENTIAL:
886 	case POSIX_FADV_NOREUSE:
887 		ioflag |= sequential_heuristic(uio, fp);
888 		break;
889 	case POSIX_FADV_RANDOM:
890 		/* XXX: Is this correct? */
891 		break;
892 	}
893 	orig_offset = uio->uio_offset;
894 
895 #ifdef MAC
896 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
897 	if (error == 0)
898 #endif
899 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
900 	fp->f_nextoff = uio->uio_offset;
901 	VOP_UNLOCK(vp, 0);
902 	if (vp->v_type != VCHR)
903 		vn_finished_write(mp);
904 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
905 	    orig_offset != uio->uio_offset)
906 		/*
907 		 * Use POSIX_FADV_DONTNEED to flush pages and buffers
908 		 * for the backing file after a POSIX_FADV_NOREUSE
909 		 * write(2).
910 		 */
911 		error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
912 		    POSIX_FADV_DONTNEED);
913 unlock:
914 	return (error);
915 }
916 
917 /*
918  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
919  * prevent the following deadlock:
920  *
921  * Assume that the thread A reads from the vnode vp1 into userspace
922  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
923  * currently not resident, then system ends up with the call chain
924  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
925  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
926  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
927  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
928  * backed by the pages of vnode vp1, and some page in buf2 is not
929  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
930  *
931  * To prevent the lock order reversal and deadlock, vn_io_fault() does
932  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
933  * Instead, it first tries to do the whole range i/o with pagefaults
934  * disabled. If all pages in the i/o buffer are resident and mapped,
935  * VOP will succeed (ignoring the genuine filesystem errors).
936  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
937  * i/o in chunks, with all pages in the chunk prefaulted and held
938  * using vm_fault_quick_hold_pages().
939  *
940  * Filesystems using this deadlock avoidance scheme should use the
941  * array of the held pages from uio, saved in the curthread->td_ma,
942  * instead of doing uiomove().  A helper function
943  * vn_io_fault_uiomove() converts uiomove request into
944  * uiomove_fromphys() over td_ma array.
945  *
946  * Since vnode locks do not cover the whole i/o anymore, rangelocks
947  * make the current i/o request atomic with respect to other i/os and
948  * truncations.
949  */
950 
951 /*
952  * Decode vn_io_fault_args and perform the corresponding i/o.
953  */
954 static int
955 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
956     struct thread *td)
957 {
958 
959 	switch (args->kind) {
960 	case VN_IO_FAULT_FOP:
961 		return ((args->args.fop_args.doio)(args->args.fop_args.fp,
962 		    uio, args->cred, args->flags, td));
963 	case VN_IO_FAULT_VOP:
964 		if (uio->uio_rw == UIO_READ) {
965 			return (VOP_READ(args->args.vop_args.vp, uio,
966 			    args->flags, args->cred));
967 		} else if (uio->uio_rw == UIO_WRITE) {
968 			return (VOP_WRITE(args->args.vop_args.vp, uio,
969 			    args->flags, args->cred));
970 		}
971 		break;
972 	}
973 	panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
974 	    uio->uio_rw);
975 }
976 
977 static int
978 vn_io_fault_touch(char *base, const struct uio *uio)
979 {
980 	int r;
981 
982 	r = fubyte(base);
983 	if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
984 		return (EFAULT);
985 	return (0);
986 }
987 
988 static int
989 vn_io_fault_prefault_user(const struct uio *uio)
990 {
991 	char *base;
992 	const struct iovec *iov;
993 	size_t len;
994 	ssize_t resid;
995 	int error, i;
996 
997 	KASSERT(uio->uio_segflg == UIO_USERSPACE,
998 	    ("vn_io_fault_prefault userspace"));
999 
1000 	error = i = 0;
1001 	iov = uio->uio_iov;
1002 	resid = uio->uio_resid;
1003 	base = iov->iov_base;
1004 	len = iov->iov_len;
1005 	while (resid > 0) {
1006 		error = vn_io_fault_touch(base, uio);
1007 		if (error != 0)
1008 			break;
1009 		if (len < PAGE_SIZE) {
1010 			if (len != 0) {
1011 				error = vn_io_fault_touch(base + len - 1, uio);
1012 				if (error != 0)
1013 					break;
1014 				resid -= len;
1015 			}
1016 			if (++i >= uio->uio_iovcnt)
1017 				break;
1018 			iov = uio->uio_iov + i;
1019 			base = iov->iov_base;
1020 			len = iov->iov_len;
1021 		} else {
1022 			len -= PAGE_SIZE;
1023 			base += PAGE_SIZE;
1024 			resid -= PAGE_SIZE;
1025 		}
1026 	}
1027 	return (error);
1028 }
1029 
1030 /*
1031  * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1032  * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1033  * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1034  * into args and call vn_io_fault1() to handle faults during the user
1035  * mode buffer accesses.
1036  */
1037 static int
1038 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1039     struct thread *td)
1040 {
1041 	vm_page_t ma[io_hold_cnt + 2];
1042 	struct uio *uio_clone, short_uio;
1043 	struct iovec short_iovec[1];
1044 	vm_page_t *prev_td_ma;
1045 	vm_prot_t prot;
1046 	vm_offset_t addr, end;
1047 	size_t len, resid;
1048 	ssize_t adv;
1049 	int error, cnt, save, saveheld, prev_td_ma_cnt;
1050 
1051 	if (vn_io_fault_prefault) {
1052 		error = vn_io_fault_prefault_user(uio);
1053 		if (error != 0)
1054 			return (error); /* Or ignore ? */
1055 	}
1056 
1057 	prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1058 
1059 	/*
1060 	 * The UFS follows IO_UNIT directive and replays back both
1061 	 * uio_offset and uio_resid if an error is encountered during the
1062 	 * operation.  But, since the iovec may be already advanced,
1063 	 * uio is still in an inconsistent state.
1064 	 *
1065 	 * Cache a copy of the original uio, which is advanced to the redo
1066 	 * point using UIO_NOCOPY below.
1067 	 */
1068 	uio_clone = cloneuio(uio);
1069 	resid = uio->uio_resid;
1070 
1071 	short_uio.uio_segflg = UIO_USERSPACE;
1072 	short_uio.uio_rw = uio->uio_rw;
1073 	short_uio.uio_td = uio->uio_td;
1074 
1075 	save = vm_fault_disable_pagefaults();
1076 	error = vn_io_fault_doio(args, uio, td);
1077 	if (error != EFAULT)
1078 		goto out;
1079 
1080 	atomic_add_long(&vn_io_faults_cnt, 1);
1081 	uio_clone->uio_segflg = UIO_NOCOPY;
1082 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
1083 	uio_clone->uio_segflg = uio->uio_segflg;
1084 
1085 	saveheld = curthread_pflags_set(TDP_UIOHELD);
1086 	prev_td_ma = td->td_ma;
1087 	prev_td_ma_cnt = td->td_ma_cnt;
1088 
1089 	while (uio_clone->uio_resid != 0) {
1090 		len = uio_clone->uio_iov->iov_len;
1091 		if (len == 0) {
1092 			KASSERT(uio_clone->uio_iovcnt >= 1,
1093 			    ("iovcnt underflow"));
1094 			uio_clone->uio_iov++;
1095 			uio_clone->uio_iovcnt--;
1096 			continue;
1097 		}
1098 		if (len > io_hold_cnt * PAGE_SIZE)
1099 			len = io_hold_cnt * PAGE_SIZE;
1100 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1101 		end = round_page(addr + len);
1102 		if (end < addr) {
1103 			error = EFAULT;
1104 			break;
1105 		}
1106 		cnt = atop(end - trunc_page(addr));
1107 		/*
1108 		 * A perfectly misaligned address and length could cause
1109 		 * both the start and the end of the chunk to use partial
1110 		 * page.  +2 accounts for such a situation.
1111 		 */
1112 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1113 		    addr, len, prot, ma, io_hold_cnt + 2);
1114 		if (cnt == -1) {
1115 			error = EFAULT;
1116 			break;
1117 		}
1118 		short_uio.uio_iov = &short_iovec[0];
1119 		short_iovec[0].iov_base = (void *)addr;
1120 		short_uio.uio_iovcnt = 1;
1121 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1122 		short_uio.uio_offset = uio_clone->uio_offset;
1123 		td->td_ma = ma;
1124 		td->td_ma_cnt = cnt;
1125 
1126 		error = vn_io_fault_doio(args, &short_uio, td);
1127 		vm_page_unhold_pages(ma, cnt);
1128 		adv = len - short_uio.uio_resid;
1129 
1130 		uio_clone->uio_iov->iov_base =
1131 		    (char *)uio_clone->uio_iov->iov_base + adv;
1132 		uio_clone->uio_iov->iov_len -= adv;
1133 		uio_clone->uio_resid -= adv;
1134 		uio_clone->uio_offset += adv;
1135 
1136 		uio->uio_resid -= adv;
1137 		uio->uio_offset += adv;
1138 
1139 		if (error != 0 || adv == 0)
1140 			break;
1141 	}
1142 	td->td_ma = prev_td_ma;
1143 	td->td_ma_cnt = prev_td_ma_cnt;
1144 	curthread_pflags_restore(saveheld);
1145 out:
1146 	vm_fault_enable_pagefaults(save);
1147 	free(uio_clone, M_IOV);
1148 	return (error);
1149 }
1150 
1151 static int
1152 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1153     int flags, struct thread *td)
1154 {
1155 	fo_rdwr_t *doio;
1156 	struct vnode *vp;
1157 	void *rl_cookie;
1158 	struct vn_io_fault_args args;
1159 	int error;
1160 
1161 	doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1162 	vp = fp->f_vnode;
1163 	foffset_lock_uio(fp, uio, flags);
1164 	if (do_vn_io_fault(vp, uio)) {
1165 		args.kind = VN_IO_FAULT_FOP;
1166 		args.args.fop_args.fp = fp;
1167 		args.args.fop_args.doio = doio;
1168 		args.cred = active_cred;
1169 		args.flags = flags | FOF_OFFSET;
1170 		if (uio->uio_rw == UIO_READ) {
1171 			rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1172 			    uio->uio_offset + uio->uio_resid);
1173 		} else if ((fp->f_flag & O_APPEND) != 0 ||
1174 		    (flags & FOF_OFFSET) == 0) {
1175 			/* For appenders, punt and lock the whole range. */
1176 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1177 		} else {
1178 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1179 			    uio->uio_offset + uio->uio_resid);
1180 		}
1181 		error = vn_io_fault1(vp, uio, &args, td);
1182 		vn_rangelock_unlock(vp, rl_cookie);
1183 	} else {
1184 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1185 	}
1186 	foffset_unlock_uio(fp, uio, flags);
1187 	return (error);
1188 }
1189 
1190 /*
1191  * Helper function to perform the requested uiomove operation using
1192  * the held pages for io->uio_iov[0].iov_base buffer instead of
1193  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1194  * instead of iov_base prevents page faults that could occur due to
1195  * pmap_collect() invalidating the mapping created by
1196  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1197  * object cleanup revoking the write access from page mappings.
1198  *
1199  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1200  * instead of plain uiomove().
1201  */
1202 int
1203 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1204 {
1205 	struct uio transp_uio;
1206 	struct iovec transp_iov[1];
1207 	struct thread *td;
1208 	size_t adv;
1209 	int error, pgadv;
1210 
1211 	td = curthread;
1212 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1213 	    uio->uio_segflg != UIO_USERSPACE)
1214 		return (uiomove(data, xfersize, uio));
1215 
1216 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1217 	transp_iov[0].iov_base = data;
1218 	transp_uio.uio_iov = &transp_iov[0];
1219 	transp_uio.uio_iovcnt = 1;
1220 	if (xfersize > uio->uio_resid)
1221 		xfersize = uio->uio_resid;
1222 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1223 	transp_uio.uio_offset = 0;
1224 	transp_uio.uio_segflg = UIO_SYSSPACE;
1225 	/*
1226 	 * Since transp_iov points to data, and td_ma page array
1227 	 * corresponds to original uio->uio_iov, we need to invert the
1228 	 * direction of the i/o operation as passed to
1229 	 * uiomove_fromphys().
1230 	 */
1231 	switch (uio->uio_rw) {
1232 	case UIO_WRITE:
1233 		transp_uio.uio_rw = UIO_READ;
1234 		break;
1235 	case UIO_READ:
1236 		transp_uio.uio_rw = UIO_WRITE;
1237 		break;
1238 	}
1239 	transp_uio.uio_td = uio->uio_td;
1240 	error = uiomove_fromphys(td->td_ma,
1241 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1242 	    xfersize, &transp_uio);
1243 	adv = xfersize - transp_uio.uio_resid;
1244 	pgadv =
1245 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1246 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1247 	td->td_ma += pgadv;
1248 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1249 	    pgadv));
1250 	td->td_ma_cnt -= pgadv;
1251 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1252 	uio->uio_iov->iov_len -= adv;
1253 	uio->uio_resid -= adv;
1254 	uio->uio_offset += adv;
1255 	return (error);
1256 }
1257 
1258 int
1259 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1260     struct uio *uio)
1261 {
1262 	struct thread *td;
1263 	vm_offset_t iov_base;
1264 	int cnt, pgadv;
1265 
1266 	td = curthread;
1267 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1268 	    uio->uio_segflg != UIO_USERSPACE)
1269 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1270 
1271 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1272 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1273 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1274 	switch (uio->uio_rw) {
1275 	case UIO_WRITE:
1276 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1277 		    offset, cnt);
1278 		break;
1279 	case UIO_READ:
1280 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1281 		    cnt);
1282 		break;
1283 	}
1284 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1285 	td->td_ma += pgadv;
1286 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1287 	    pgadv));
1288 	td->td_ma_cnt -= pgadv;
1289 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1290 	uio->uio_iov->iov_len -= cnt;
1291 	uio->uio_resid -= cnt;
1292 	uio->uio_offset += cnt;
1293 	return (0);
1294 }
1295 
1296 
1297 /*
1298  * File table truncate routine.
1299  */
1300 static int
1301 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1302     struct thread *td)
1303 {
1304 	struct vattr vattr;
1305 	struct mount *mp;
1306 	struct vnode *vp;
1307 	void *rl_cookie;
1308 	int error;
1309 
1310 	vp = fp->f_vnode;
1311 
1312 	/*
1313 	 * Lock the whole range for truncation.  Otherwise split i/o
1314 	 * might happen partly before and partly after the truncation.
1315 	 */
1316 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1317 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1318 	if (error)
1319 		goto out1;
1320 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1321 	AUDIT_ARG_VNODE1(vp);
1322 	if (vp->v_type == VDIR) {
1323 		error = EISDIR;
1324 		goto out;
1325 	}
1326 #ifdef MAC
1327 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1328 	if (error)
1329 		goto out;
1330 #endif
1331 	error = vn_writechk(vp);
1332 	if (error == 0) {
1333 		VATTR_NULL(&vattr);
1334 		vattr.va_size = length;
1335 		if ((fp->f_flag & O_FSYNC) != 0)
1336 			vattr.va_vaflags |= VA_SYNC;
1337 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1338 	}
1339 out:
1340 	VOP_UNLOCK(vp, 0);
1341 	vn_finished_write(mp);
1342 out1:
1343 	vn_rangelock_unlock(vp, rl_cookie);
1344 	return (error);
1345 }
1346 
1347 /*
1348  * File table vnode stat routine.
1349  */
1350 static int
1351 vn_statfile(fp, sb, active_cred, td)
1352 	struct file *fp;
1353 	struct stat *sb;
1354 	struct ucred *active_cred;
1355 	struct thread *td;
1356 {
1357 	struct vnode *vp = fp->f_vnode;
1358 	int error;
1359 
1360 	vn_lock(vp, LK_SHARED | LK_RETRY);
1361 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1362 	VOP_UNLOCK(vp, 0);
1363 
1364 	return (error);
1365 }
1366 
1367 /*
1368  * Stat a vnode; implementation for the stat syscall
1369  */
1370 int
1371 vn_stat(vp, sb, active_cred, file_cred, td)
1372 	struct vnode *vp;
1373 	register struct stat *sb;
1374 	struct ucred *active_cred;
1375 	struct ucred *file_cred;
1376 	struct thread *td;
1377 {
1378 	struct vattr vattr;
1379 	register struct vattr *vap;
1380 	int error;
1381 	u_short mode;
1382 
1383 	AUDIT_ARG_VNODE1(vp);
1384 #ifdef MAC
1385 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1386 	if (error)
1387 		return (error);
1388 #endif
1389 
1390 	vap = &vattr;
1391 
1392 	/*
1393 	 * Initialize defaults for new and unusual fields, so that file
1394 	 * systems which don't support these fields don't need to know
1395 	 * about them.
1396 	 */
1397 	vap->va_birthtime.tv_sec = -1;
1398 	vap->va_birthtime.tv_nsec = 0;
1399 	vap->va_fsid = VNOVAL;
1400 	vap->va_rdev = NODEV;
1401 
1402 	error = VOP_GETATTR(vp, vap, active_cred);
1403 	if (error)
1404 		return (error);
1405 
1406 	/*
1407 	 * Zero the spare stat fields
1408 	 */
1409 	bzero(sb, sizeof *sb);
1410 
1411 	/*
1412 	 * Copy from vattr table
1413 	 */
1414 	if (vap->va_fsid != VNOVAL)
1415 		sb->st_dev = vap->va_fsid;
1416 	else
1417 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1418 	sb->st_ino = vap->va_fileid;
1419 	mode = vap->va_mode;
1420 	switch (vap->va_type) {
1421 	case VREG:
1422 		mode |= S_IFREG;
1423 		break;
1424 	case VDIR:
1425 		mode |= S_IFDIR;
1426 		break;
1427 	case VBLK:
1428 		mode |= S_IFBLK;
1429 		break;
1430 	case VCHR:
1431 		mode |= S_IFCHR;
1432 		break;
1433 	case VLNK:
1434 		mode |= S_IFLNK;
1435 		break;
1436 	case VSOCK:
1437 		mode |= S_IFSOCK;
1438 		break;
1439 	case VFIFO:
1440 		mode |= S_IFIFO;
1441 		break;
1442 	default:
1443 		return (EBADF);
1444 	}
1445 	sb->st_mode = mode;
1446 	sb->st_nlink = vap->va_nlink;
1447 	sb->st_uid = vap->va_uid;
1448 	sb->st_gid = vap->va_gid;
1449 	sb->st_rdev = vap->va_rdev;
1450 	if (vap->va_size > OFF_MAX)
1451 		return (EOVERFLOW);
1452 	sb->st_size = vap->va_size;
1453 	sb->st_atim = vap->va_atime;
1454 	sb->st_mtim = vap->va_mtime;
1455 	sb->st_ctim = vap->va_ctime;
1456 	sb->st_birthtim = vap->va_birthtime;
1457 
1458         /*
1459 	 * According to www.opengroup.org, the meaning of st_blksize is
1460 	 *   "a filesystem-specific preferred I/O block size for this
1461 	 *    object.  In some filesystem types, this may vary from file
1462 	 *    to file"
1463 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1464 	 */
1465 
1466 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1467 
1468 	sb->st_flags = vap->va_flags;
1469 	if (priv_check(td, PRIV_VFS_GENERATION))
1470 		sb->st_gen = 0;
1471 	else
1472 		sb->st_gen = vap->va_gen;
1473 
1474 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1475 	return (0);
1476 }
1477 
1478 /*
1479  * File table vnode ioctl routine.
1480  */
1481 static int
1482 vn_ioctl(fp, com, data, active_cred, td)
1483 	struct file *fp;
1484 	u_long com;
1485 	void *data;
1486 	struct ucred *active_cred;
1487 	struct thread *td;
1488 {
1489 	struct vattr vattr;
1490 	struct vnode *vp;
1491 	int error;
1492 
1493 	vp = fp->f_vnode;
1494 	switch (vp->v_type) {
1495 	case VDIR:
1496 	case VREG:
1497 		switch (com) {
1498 		case FIONREAD:
1499 			vn_lock(vp, LK_SHARED | LK_RETRY);
1500 			error = VOP_GETATTR(vp, &vattr, active_cred);
1501 			VOP_UNLOCK(vp, 0);
1502 			if (error == 0)
1503 				*(int *)data = vattr.va_size - fp->f_offset;
1504 			return (error);
1505 		case FIONBIO:
1506 		case FIOASYNC:
1507 			return (0);
1508 		default:
1509 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1510 			    active_cred, td));
1511 		}
1512 		break;
1513 	case VCHR:
1514 		return (VOP_IOCTL(vp, com, data, fp->f_flag,
1515 		    active_cred, td));
1516 	default:
1517 		return (ENOTTY);
1518 	}
1519 }
1520 
1521 /*
1522  * File table vnode poll routine.
1523  */
1524 static int
1525 vn_poll(fp, events, active_cred, td)
1526 	struct file *fp;
1527 	int events;
1528 	struct ucred *active_cred;
1529 	struct thread *td;
1530 {
1531 	struct vnode *vp;
1532 	int error;
1533 
1534 	vp = fp->f_vnode;
1535 #ifdef MAC
1536 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1537 	AUDIT_ARG_VNODE1(vp);
1538 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1539 	VOP_UNLOCK(vp, 0);
1540 	if (!error)
1541 #endif
1542 
1543 	error = VOP_POLL(vp, events, fp->f_cred, td);
1544 	return (error);
1545 }
1546 
1547 /*
1548  * Acquire the requested lock and then check for validity.  LK_RETRY
1549  * permits vn_lock to return doomed vnodes.
1550  */
1551 int
1552 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1553 {
1554 	int error;
1555 
1556 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1557 	    ("vn_lock: no locktype"));
1558 	VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count"));
1559 retry:
1560 	error = VOP_LOCK1(vp, flags, file, line);
1561 	flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1562 	KASSERT((flags & LK_RETRY) == 0 || error == 0,
1563 	    ("vn_lock: error %d incompatible with flags %#x", error, flags));
1564 
1565 	if ((flags & LK_RETRY) == 0) {
1566 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) {
1567 			VOP_UNLOCK(vp, 0);
1568 			error = ENOENT;
1569 		}
1570 	} else if (error != 0)
1571 		goto retry;
1572 	return (error);
1573 }
1574 
1575 /*
1576  * File table vnode close routine.
1577  */
1578 static int
1579 vn_closefile(struct file *fp, struct thread *td)
1580 {
1581 	struct vnode *vp;
1582 	struct flock lf;
1583 	int error;
1584 	bool ref;
1585 
1586 	vp = fp->f_vnode;
1587 	fp->f_ops = &badfileops;
1588 	ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE;
1589 
1590 	error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref);
1591 
1592 	if (__predict_false(ref)) {
1593 		lf.l_whence = SEEK_SET;
1594 		lf.l_start = 0;
1595 		lf.l_len = 0;
1596 		lf.l_type = F_UNLCK;
1597 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1598 		vrele(vp);
1599 	}
1600 	return (error);
1601 }
1602 
1603 static bool
1604 vn_suspendable(struct mount *mp)
1605 {
1606 
1607 	return (mp->mnt_op->vfs_susp_clean != NULL);
1608 }
1609 
1610 /*
1611  * Preparing to start a filesystem write operation. If the operation is
1612  * permitted, then we bump the count of operations in progress and
1613  * proceed. If a suspend request is in progress, we wait until the
1614  * suspension is over, and then proceed.
1615  */
1616 static int
1617 vn_start_write_locked(struct mount *mp, int flags)
1618 {
1619 	int error, mflags;
1620 
1621 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1622 	error = 0;
1623 
1624 	/*
1625 	 * Check on status of suspension.
1626 	 */
1627 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1628 	    mp->mnt_susp_owner != curthread) {
1629 		mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1630 		    (flags & PCATCH) : 0) | (PUSER - 1);
1631 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1632 			if (flags & V_NOWAIT) {
1633 				error = EWOULDBLOCK;
1634 				goto unlock;
1635 			}
1636 			error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1637 			    "suspfs", 0);
1638 			if (error)
1639 				goto unlock;
1640 		}
1641 	}
1642 	if (flags & V_XSLEEP)
1643 		goto unlock;
1644 	mp->mnt_writeopcount++;
1645 unlock:
1646 	if (error != 0 || (flags & V_XSLEEP) != 0)
1647 		MNT_REL(mp);
1648 	MNT_IUNLOCK(mp);
1649 	return (error);
1650 }
1651 
1652 int
1653 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1654 {
1655 	struct mount *mp;
1656 	int error;
1657 
1658 	KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1659 	    ("V_MNTREF requires mp"));
1660 
1661 	error = 0;
1662 	/*
1663 	 * If a vnode is provided, get and return the mount point that
1664 	 * to which it will write.
1665 	 */
1666 	if (vp != NULL) {
1667 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1668 			*mpp = NULL;
1669 			if (error != EOPNOTSUPP)
1670 				return (error);
1671 			return (0);
1672 		}
1673 	}
1674 	if ((mp = *mpp) == NULL)
1675 		return (0);
1676 
1677 	if (!vn_suspendable(mp)) {
1678 		if (vp != NULL || (flags & V_MNTREF) != 0)
1679 			vfs_rel(mp);
1680 		return (0);
1681 	}
1682 
1683 	/*
1684 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1685 	 * a vfs_ref().
1686 	 * As long as a vnode is not provided we need to acquire a
1687 	 * refcount for the provided mountpoint too, in order to
1688 	 * emulate a vfs_ref().
1689 	 */
1690 	MNT_ILOCK(mp);
1691 	if (vp == NULL && (flags & V_MNTREF) == 0)
1692 		MNT_REF(mp);
1693 
1694 	return (vn_start_write_locked(mp, flags));
1695 }
1696 
1697 /*
1698  * Secondary suspension. Used by operations such as vop_inactive
1699  * routines that are needed by the higher level functions. These
1700  * are allowed to proceed until all the higher level functions have
1701  * completed (indicated by mnt_writeopcount dropping to zero). At that
1702  * time, these operations are halted until the suspension is over.
1703  */
1704 int
1705 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1706 {
1707 	struct mount *mp;
1708 	int error;
1709 
1710 	KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1711 	    ("V_MNTREF requires mp"));
1712 
1713  retry:
1714 	if (vp != NULL) {
1715 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1716 			*mpp = NULL;
1717 			if (error != EOPNOTSUPP)
1718 				return (error);
1719 			return (0);
1720 		}
1721 	}
1722 	/*
1723 	 * If we are not suspended or have not yet reached suspended
1724 	 * mode, then let the operation proceed.
1725 	 */
1726 	if ((mp = *mpp) == NULL)
1727 		return (0);
1728 
1729 	if (!vn_suspendable(mp)) {
1730 		if (vp != NULL || (flags & V_MNTREF) != 0)
1731 			vfs_rel(mp);
1732 		return (0);
1733 	}
1734 
1735 	/*
1736 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1737 	 * a vfs_ref().
1738 	 * As long as a vnode is not provided we need to acquire a
1739 	 * refcount for the provided mountpoint too, in order to
1740 	 * emulate a vfs_ref().
1741 	 */
1742 	MNT_ILOCK(mp);
1743 	if (vp == NULL && (flags & V_MNTREF) == 0)
1744 		MNT_REF(mp);
1745 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1746 		mp->mnt_secondary_writes++;
1747 		mp->mnt_secondary_accwrites++;
1748 		MNT_IUNLOCK(mp);
1749 		return (0);
1750 	}
1751 	if (flags & V_NOWAIT) {
1752 		MNT_REL(mp);
1753 		MNT_IUNLOCK(mp);
1754 		return (EWOULDBLOCK);
1755 	}
1756 	/*
1757 	 * Wait for the suspension to finish.
1758 	 */
1759 	error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1760 	    ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1761 	    "suspfs", 0);
1762 	vfs_rel(mp);
1763 	if (error == 0)
1764 		goto retry;
1765 	return (error);
1766 }
1767 
1768 /*
1769  * Filesystem write operation has completed. If we are suspending and this
1770  * operation is the last one, notify the suspender that the suspension is
1771  * now in effect.
1772  */
1773 void
1774 vn_finished_write(mp)
1775 	struct mount *mp;
1776 {
1777 	if (mp == NULL || !vn_suspendable(mp))
1778 		return;
1779 	MNT_ILOCK(mp);
1780 	MNT_REL(mp);
1781 	mp->mnt_writeopcount--;
1782 	if (mp->mnt_writeopcount < 0)
1783 		panic("vn_finished_write: neg cnt");
1784 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1785 	    mp->mnt_writeopcount <= 0)
1786 		wakeup(&mp->mnt_writeopcount);
1787 	MNT_IUNLOCK(mp);
1788 }
1789 
1790 
1791 /*
1792  * Filesystem secondary write operation has completed. If we are
1793  * suspending and this operation is the last one, notify the suspender
1794  * that the suspension is now in effect.
1795  */
1796 void
1797 vn_finished_secondary_write(mp)
1798 	struct mount *mp;
1799 {
1800 	if (mp == NULL || !vn_suspendable(mp))
1801 		return;
1802 	MNT_ILOCK(mp);
1803 	MNT_REL(mp);
1804 	mp->mnt_secondary_writes--;
1805 	if (mp->mnt_secondary_writes < 0)
1806 		panic("vn_finished_secondary_write: neg cnt");
1807 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1808 	    mp->mnt_secondary_writes <= 0)
1809 		wakeup(&mp->mnt_secondary_writes);
1810 	MNT_IUNLOCK(mp);
1811 }
1812 
1813 
1814 
1815 /*
1816  * Request a filesystem to suspend write operations.
1817  */
1818 int
1819 vfs_write_suspend(struct mount *mp, int flags)
1820 {
1821 	int error;
1822 
1823 	MPASS(vn_suspendable(mp));
1824 
1825 	MNT_ILOCK(mp);
1826 	if (mp->mnt_susp_owner == curthread) {
1827 		MNT_IUNLOCK(mp);
1828 		return (EALREADY);
1829 	}
1830 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1831 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1832 
1833 	/*
1834 	 * Unmount holds a write reference on the mount point.  If we
1835 	 * own busy reference and drain for writers, we deadlock with
1836 	 * the reference draining in the unmount path.  Callers of
1837 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1838 	 * vfs_busy() reference is owned and caller is not in the
1839 	 * unmount context.
1840 	 */
1841 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1842 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1843 		MNT_IUNLOCK(mp);
1844 		return (EBUSY);
1845 	}
1846 
1847 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1848 	mp->mnt_susp_owner = curthread;
1849 	if (mp->mnt_writeopcount > 0)
1850 		(void) msleep(&mp->mnt_writeopcount,
1851 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1852 	else
1853 		MNT_IUNLOCK(mp);
1854 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1855 		vfs_write_resume(mp, 0);
1856 	return (error);
1857 }
1858 
1859 /*
1860  * Request a filesystem to resume write operations.
1861  */
1862 void
1863 vfs_write_resume(struct mount *mp, int flags)
1864 {
1865 
1866 	MPASS(vn_suspendable(mp));
1867 
1868 	MNT_ILOCK(mp);
1869 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1870 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1871 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1872 				       MNTK_SUSPENDED);
1873 		mp->mnt_susp_owner = NULL;
1874 		wakeup(&mp->mnt_writeopcount);
1875 		wakeup(&mp->mnt_flag);
1876 		curthread->td_pflags &= ~TDP_IGNSUSP;
1877 		if ((flags & VR_START_WRITE) != 0) {
1878 			MNT_REF(mp);
1879 			mp->mnt_writeopcount++;
1880 		}
1881 		MNT_IUNLOCK(mp);
1882 		if ((flags & VR_NO_SUSPCLR) == 0)
1883 			VFS_SUSP_CLEAN(mp);
1884 	} else if ((flags & VR_START_WRITE) != 0) {
1885 		MNT_REF(mp);
1886 		vn_start_write_locked(mp, 0);
1887 	} else {
1888 		MNT_IUNLOCK(mp);
1889 	}
1890 }
1891 
1892 /*
1893  * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1894  * methods.
1895  */
1896 int
1897 vfs_write_suspend_umnt(struct mount *mp)
1898 {
1899 	int error;
1900 
1901 	MPASS(vn_suspendable(mp));
1902 	KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1903 	    ("vfs_write_suspend_umnt: recursed"));
1904 
1905 	/* dounmount() already called vn_start_write(). */
1906 	for (;;) {
1907 		vn_finished_write(mp);
1908 		error = vfs_write_suspend(mp, 0);
1909 		if (error != 0) {
1910 			vn_start_write(NULL, &mp, V_WAIT);
1911 			return (error);
1912 		}
1913 		MNT_ILOCK(mp);
1914 		if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1915 			break;
1916 		MNT_IUNLOCK(mp);
1917 		vn_start_write(NULL, &mp, V_WAIT);
1918 	}
1919 	mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1920 	wakeup(&mp->mnt_flag);
1921 	MNT_IUNLOCK(mp);
1922 	curthread->td_pflags |= TDP_IGNSUSP;
1923 	return (0);
1924 }
1925 
1926 /*
1927  * Implement kqueues for files by translating it to vnode operation.
1928  */
1929 static int
1930 vn_kqfilter(struct file *fp, struct knote *kn)
1931 {
1932 
1933 	return (VOP_KQFILTER(fp->f_vnode, kn));
1934 }
1935 
1936 /*
1937  * Simplified in-kernel wrapper calls for extended attribute access.
1938  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1939  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1940  */
1941 int
1942 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1943     const char *attrname, int *buflen, char *buf, struct thread *td)
1944 {
1945 	struct uio	auio;
1946 	struct iovec	iov;
1947 	int	error;
1948 
1949 	iov.iov_len = *buflen;
1950 	iov.iov_base = buf;
1951 
1952 	auio.uio_iov = &iov;
1953 	auio.uio_iovcnt = 1;
1954 	auio.uio_rw = UIO_READ;
1955 	auio.uio_segflg = UIO_SYSSPACE;
1956 	auio.uio_td = td;
1957 	auio.uio_offset = 0;
1958 	auio.uio_resid = *buflen;
1959 
1960 	if ((ioflg & IO_NODELOCKED) == 0)
1961 		vn_lock(vp, LK_SHARED | LK_RETRY);
1962 
1963 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1964 
1965 	/* authorize attribute retrieval as kernel */
1966 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1967 	    td);
1968 
1969 	if ((ioflg & IO_NODELOCKED) == 0)
1970 		VOP_UNLOCK(vp, 0);
1971 
1972 	if (error == 0) {
1973 		*buflen = *buflen - auio.uio_resid;
1974 	}
1975 
1976 	return (error);
1977 }
1978 
1979 /*
1980  * XXX failure mode if partially written?
1981  */
1982 int
1983 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1984     const char *attrname, int buflen, char *buf, struct thread *td)
1985 {
1986 	struct uio	auio;
1987 	struct iovec	iov;
1988 	struct mount	*mp;
1989 	int	error;
1990 
1991 	iov.iov_len = buflen;
1992 	iov.iov_base = buf;
1993 
1994 	auio.uio_iov = &iov;
1995 	auio.uio_iovcnt = 1;
1996 	auio.uio_rw = UIO_WRITE;
1997 	auio.uio_segflg = UIO_SYSSPACE;
1998 	auio.uio_td = td;
1999 	auio.uio_offset = 0;
2000 	auio.uio_resid = buflen;
2001 
2002 	if ((ioflg & IO_NODELOCKED) == 0) {
2003 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2004 			return (error);
2005 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2006 	}
2007 
2008 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2009 
2010 	/* authorize attribute setting as kernel */
2011 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2012 
2013 	if ((ioflg & IO_NODELOCKED) == 0) {
2014 		vn_finished_write(mp);
2015 		VOP_UNLOCK(vp, 0);
2016 	}
2017 
2018 	return (error);
2019 }
2020 
2021 int
2022 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2023     const char *attrname, struct thread *td)
2024 {
2025 	struct mount	*mp;
2026 	int	error;
2027 
2028 	if ((ioflg & IO_NODELOCKED) == 0) {
2029 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2030 			return (error);
2031 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2032 	}
2033 
2034 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2035 
2036 	/* authorize attribute removal as kernel */
2037 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2038 	if (error == EOPNOTSUPP)
2039 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2040 		    NULL, td);
2041 
2042 	if ((ioflg & IO_NODELOCKED) == 0) {
2043 		vn_finished_write(mp);
2044 		VOP_UNLOCK(vp, 0);
2045 	}
2046 
2047 	return (error);
2048 }
2049 
2050 static int
2051 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2052     struct vnode **rvp)
2053 {
2054 
2055 	return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2056 }
2057 
2058 int
2059 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2060 {
2061 
2062 	return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2063 	    lkflags, rvp));
2064 }
2065 
2066 int
2067 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2068     int lkflags, struct vnode **rvp)
2069 {
2070 	struct mount *mp;
2071 	int ltype, error;
2072 
2073 	ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2074 	mp = vp->v_mount;
2075 	ltype = VOP_ISLOCKED(vp);
2076 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2077 	    ("vn_vget_ino: vp not locked"));
2078 	error = vfs_busy(mp, MBF_NOWAIT);
2079 	if (error != 0) {
2080 		vfs_ref(mp);
2081 		VOP_UNLOCK(vp, 0);
2082 		error = vfs_busy(mp, 0);
2083 		vn_lock(vp, ltype | LK_RETRY);
2084 		vfs_rel(mp);
2085 		if (error != 0)
2086 			return (ENOENT);
2087 		if (vp->v_iflag & VI_DOOMED) {
2088 			vfs_unbusy(mp);
2089 			return (ENOENT);
2090 		}
2091 	}
2092 	VOP_UNLOCK(vp, 0);
2093 	error = alloc(mp, alloc_arg, lkflags, rvp);
2094 	vfs_unbusy(mp);
2095 	if (*rvp != vp)
2096 		vn_lock(vp, ltype | LK_RETRY);
2097 	if (vp->v_iflag & VI_DOOMED) {
2098 		if (error == 0) {
2099 			if (*rvp == vp)
2100 				vunref(vp);
2101 			else
2102 				vput(*rvp);
2103 		}
2104 		error = ENOENT;
2105 	}
2106 	return (error);
2107 }
2108 
2109 int
2110 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2111     struct thread *td)
2112 {
2113 
2114 	if (vp->v_type != VREG || td == NULL)
2115 		return (0);
2116 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
2117 	    lim_cur(td, RLIMIT_FSIZE)) {
2118 		PROC_LOCK(td->td_proc);
2119 		kern_psignal(td->td_proc, SIGXFSZ);
2120 		PROC_UNLOCK(td->td_proc);
2121 		return (EFBIG);
2122 	}
2123 	return (0);
2124 }
2125 
2126 int
2127 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2128     struct thread *td)
2129 {
2130 	struct vnode *vp;
2131 
2132 	vp = fp->f_vnode;
2133 #ifdef AUDIT
2134 	vn_lock(vp, LK_SHARED | LK_RETRY);
2135 	AUDIT_ARG_VNODE1(vp);
2136 	VOP_UNLOCK(vp, 0);
2137 #endif
2138 	return (setfmode(td, active_cred, vp, mode));
2139 }
2140 
2141 int
2142 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2143     struct thread *td)
2144 {
2145 	struct vnode *vp;
2146 
2147 	vp = fp->f_vnode;
2148 #ifdef AUDIT
2149 	vn_lock(vp, LK_SHARED | LK_RETRY);
2150 	AUDIT_ARG_VNODE1(vp);
2151 	VOP_UNLOCK(vp, 0);
2152 #endif
2153 	return (setfown(td, active_cred, vp, uid, gid));
2154 }
2155 
2156 void
2157 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2158 {
2159 	vm_object_t object;
2160 
2161 	if ((object = vp->v_object) == NULL)
2162 		return;
2163 	VM_OBJECT_WLOCK(object);
2164 	vm_object_page_remove(object, start, end, 0);
2165 	VM_OBJECT_WUNLOCK(object);
2166 }
2167 
2168 int
2169 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2170 {
2171 	struct vattr va;
2172 	daddr_t bn, bnp;
2173 	uint64_t bsize;
2174 	off_t noff;
2175 	int error;
2176 
2177 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2178 	    ("Wrong command %lu", cmd));
2179 
2180 	if (vn_lock(vp, LK_SHARED) != 0)
2181 		return (EBADF);
2182 	if (vp->v_type != VREG) {
2183 		error = ENOTTY;
2184 		goto unlock;
2185 	}
2186 	error = VOP_GETATTR(vp, &va, cred);
2187 	if (error != 0)
2188 		goto unlock;
2189 	noff = *off;
2190 	if (noff >= va.va_size) {
2191 		error = ENXIO;
2192 		goto unlock;
2193 	}
2194 	bsize = vp->v_mount->mnt_stat.f_iosize;
2195 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2196 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2197 		if (error == EOPNOTSUPP) {
2198 			error = ENOTTY;
2199 			goto unlock;
2200 		}
2201 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2202 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
2203 			noff = bn * bsize;
2204 			if (noff < *off)
2205 				noff = *off;
2206 			goto unlock;
2207 		}
2208 	}
2209 	if (noff > va.va_size)
2210 		noff = va.va_size;
2211 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2212 	if (cmd == FIOSEEKDATA)
2213 		error = ENXIO;
2214 unlock:
2215 	VOP_UNLOCK(vp, 0);
2216 	if (error == 0)
2217 		*off = noff;
2218 	return (error);
2219 }
2220 
2221 int
2222 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2223 {
2224 	struct ucred *cred;
2225 	struct vnode *vp;
2226 	struct vattr vattr;
2227 	off_t foffset, size;
2228 	int error, noneg;
2229 
2230 	cred = td->td_ucred;
2231 	vp = fp->f_vnode;
2232 	foffset = foffset_lock(fp, 0);
2233 	noneg = (vp->v_type != VCHR);
2234 	error = 0;
2235 	switch (whence) {
2236 	case L_INCR:
2237 		if (noneg &&
2238 		    (foffset < 0 ||
2239 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2240 			error = EOVERFLOW;
2241 			break;
2242 		}
2243 		offset += foffset;
2244 		break;
2245 	case L_XTND:
2246 		vn_lock(vp, LK_SHARED | LK_RETRY);
2247 		error = VOP_GETATTR(vp, &vattr, cred);
2248 		VOP_UNLOCK(vp, 0);
2249 		if (error)
2250 			break;
2251 
2252 		/*
2253 		 * If the file references a disk device, then fetch
2254 		 * the media size and use that to determine the ending
2255 		 * offset.
2256 		 */
2257 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2258 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2259 			vattr.va_size = size;
2260 		if (noneg &&
2261 		    (vattr.va_size > OFF_MAX ||
2262 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2263 			error = EOVERFLOW;
2264 			break;
2265 		}
2266 		offset += vattr.va_size;
2267 		break;
2268 	case L_SET:
2269 		break;
2270 	case SEEK_DATA:
2271 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2272 		break;
2273 	case SEEK_HOLE:
2274 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2275 		break;
2276 	default:
2277 		error = EINVAL;
2278 	}
2279 	if (error == 0 && noneg && offset < 0)
2280 		error = EINVAL;
2281 	if (error != 0)
2282 		goto drop;
2283 	VFS_KNOTE_UNLOCKED(vp, 0);
2284 	td->td_uretoff.tdu_off = offset;
2285 drop:
2286 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2287 	return (error);
2288 }
2289 
2290 int
2291 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2292     struct thread *td)
2293 {
2294 	int error;
2295 
2296 	/*
2297 	 * Grant permission if the caller is the owner of the file, or
2298 	 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2299 	 * on the file.  If the time pointer is null, then write
2300 	 * permission on the file is also sufficient.
2301 	 *
2302 	 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2303 	 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2304 	 * will be allowed to set the times [..] to the current
2305 	 * server time.
2306 	 */
2307 	error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2308 	if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2309 		error = VOP_ACCESS(vp, VWRITE, cred, td);
2310 	return (error);
2311 }
2312 
2313 int
2314 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2315 {
2316 	struct vnode *vp;
2317 	int error;
2318 
2319 	if (fp->f_type == DTYPE_FIFO)
2320 		kif->kf_type = KF_TYPE_FIFO;
2321 	else
2322 		kif->kf_type = KF_TYPE_VNODE;
2323 	vp = fp->f_vnode;
2324 	vref(vp);
2325 	FILEDESC_SUNLOCK(fdp);
2326 	error = vn_fill_kinfo_vnode(vp, kif);
2327 	vrele(vp);
2328 	FILEDESC_SLOCK(fdp);
2329 	return (error);
2330 }
2331 
2332 static inline void
2333 vn_fill_junk(struct kinfo_file *kif)
2334 {
2335 	size_t len, olen;
2336 
2337 	/*
2338 	 * Simulate vn_fullpath returning changing values for a given
2339 	 * vp during e.g. coredump.
2340 	 */
2341 	len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2342 	olen = strlen(kif->kf_path);
2343 	if (len < olen)
2344 		strcpy(&kif->kf_path[len - 1], "$");
2345 	else
2346 		for (; olen < len; olen++)
2347 			strcpy(&kif->kf_path[olen], "A");
2348 }
2349 
2350 int
2351 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2352 {
2353 	struct vattr va;
2354 	char *fullpath, *freepath;
2355 	int error;
2356 
2357 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2358 	freepath = NULL;
2359 	fullpath = "-";
2360 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2361 	if (error == 0) {
2362 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2363 	}
2364 	if (freepath != NULL)
2365 		free(freepath, M_TEMP);
2366 
2367 	KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2368 		vn_fill_junk(kif);
2369 	);
2370 
2371 	/*
2372 	 * Retrieve vnode attributes.
2373 	 */
2374 	va.va_fsid = VNOVAL;
2375 	va.va_rdev = NODEV;
2376 	vn_lock(vp, LK_SHARED | LK_RETRY);
2377 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2378 	VOP_UNLOCK(vp, 0);
2379 	if (error != 0)
2380 		return (error);
2381 	if (va.va_fsid != VNOVAL)
2382 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2383 	else
2384 		kif->kf_un.kf_file.kf_file_fsid =
2385 		    vp->v_mount->mnt_stat.f_fsid.val[0];
2386 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2387 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2388 	kif->kf_un.kf_file.kf_file_size = va.va_size;
2389 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2390 	return (0);
2391 }
2392 
2393 int
2394 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2395     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2396     struct thread *td)
2397 {
2398 #ifdef HWPMC_HOOKS
2399 	struct pmckern_map_in pkm;
2400 #endif
2401 	struct mount *mp;
2402 	struct vnode *vp;
2403 	vm_object_t object;
2404 	vm_prot_t maxprot;
2405 	boolean_t writecounted;
2406 	int error;
2407 
2408 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2409     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2410 	/*
2411 	 * POSIX shared-memory objects are defined to have
2412 	 * kernel persistence, and are not defined to support
2413 	 * read(2)/write(2) -- or even open(2).  Thus, we can
2414 	 * use MAP_ASYNC to trade on-disk coherence for speed.
2415 	 * The shm_open(3) library routine turns on the FPOSIXSHM
2416 	 * flag to request this behavior.
2417 	 */
2418 	if ((fp->f_flag & FPOSIXSHM) != 0)
2419 		flags |= MAP_NOSYNC;
2420 #endif
2421 	vp = fp->f_vnode;
2422 
2423 	/*
2424 	 * Ensure that file and memory protections are
2425 	 * compatible.  Note that we only worry about
2426 	 * writability if mapping is shared; in this case,
2427 	 * current and max prot are dictated by the open file.
2428 	 * XXX use the vnode instead?  Problem is: what
2429 	 * credentials do we use for determination? What if
2430 	 * proc does a setuid?
2431 	 */
2432 	mp = vp->v_mount;
2433 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
2434 		maxprot = VM_PROT_NONE;
2435 		if ((prot & VM_PROT_EXECUTE) != 0)
2436 			return (EACCES);
2437 	} else
2438 		maxprot = VM_PROT_EXECUTE;
2439 	if ((fp->f_flag & FREAD) != 0)
2440 		maxprot |= VM_PROT_READ;
2441 	else if ((prot & VM_PROT_READ) != 0)
2442 		return (EACCES);
2443 
2444 	/*
2445 	 * If we are sharing potential changes via MAP_SHARED and we
2446 	 * are trying to get write permission although we opened it
2447 	 * without asking for it, bail out.
2448 	 */
2449 	if ((flags & MAP_SHARED) != 0) {
2450 		if ((fp->f_flag & FWRITE) != 0)
2451 			maxprot |= VM_PROT_WRITE;
2452 		else if ((prot & VM_PROT_WRITE) != 0)
2453 			return (EACCES);
2454 	} else {
2455 		maxprot |= VM_PROT_WRITE;
2456 		cap_maxprot |= VM_PROT_WRITE;
2457 	}
2458 	maxprot &= cap_maxprot;
2459 
2460 	/*
2461 	 * For regular files and shared memory, POSIX requires that
2462 	 * the value of foff be a legitimate offset within the data
2463 	 * object.  In particular, negative offsets are invalid.
2464 	 * Blocking negative offsets and overflows here avoids
2465 	 * possible wraparound or user-level access into reserved
2466 	 * ranges of the data object later.  In contrast, POSIX does
2467 	 * not dictate how offsets are used by device drivers, so in
2468 	 * the case of a device mapping a negative offset is passed
2469 	 * on.
2470 	 */
2471 	if (
2472 #ifdef _LP64
2473 	    size > OFF_MAX ||
2474 #endif
2475 	    foff < 0 || foff > OFF_MAX - size)
2476 		return (EINVAL);
2477 
2478 	writecounted = FALSE;
2479 	error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2480 	    &foff, &object, &writecounted);
2481 	if (error != 0)
2482 		return (error);
2483 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2484 	    foff, writecounted, td);
2485 	if (error != 0) {
2486 		/*
2487 		 * If this mapping was accounted for in the vnode's
2488 		 * writecount, then undo that now.
2489 		 */
2490 		if (writecounted)
2491 			vnode_pager_release_writecount(object, 0, size);
2492 		vm_object_deallocate(object);
2493 	}
2494 #ifdef HWPMC_HOOKS
2495 	/* Inform hwpmc(4) if an executable is being mapped. */
2496 	if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) {
2497 		if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) {
2498 			pkm.pm_file = vp;
2499 			pkm.pm_address = (uintptr_t) *addr;
2500 			PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
2501 		}
2502 	}
2503 #endif
2504 	return (error);
2505 }
2506