xref: /freebsd/sys/kern/vfs_vnops.c (revision 5405b282e1f319b6f3597bb77f68be903e7f248c)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
13  * Copyright (c) 2013, 2014 The FreeBSD Foundation
14  *
15  * Portions of this software were developed by Konstantin Belousov
16  * under sponsorship from the FreeBSD Foundation.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
21  * 1. Redistributions of source code must retain the above copyright
22  *    notice, this list of conditions and the following disclaimer.
23  * 2. Redistributions in binary form must reproduce the above copyright
24  *    notice, this list of conditions and the following disclaimer in the
25  *    documentation and/or other materials provided with the distribution.
26  * 3. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)vfs_vnops.c	8.2 (Berkeley) 1/21/94
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_hwpmc_hooks.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/disk.h>
53 #include <sys/fail.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/kdb.h>
57 #include <sys/ktr.h>
58 #include <sys/stat.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/limits.h>
62 #include <sys/lock.h>
63 #include <sys/mman.h>
64 #include <sys/mount.h>
65 #include <sys/mutex.h>
66 #include <sys/namei.h>
67 #include <sys/vnode.h>
68 #include <sys/bio.h>
69 #include <sys/buf.h>
70 #include <sys/filio.h>
71 #include <sys/resourcevar.h>
72 #include <sys/rwlock.h>
73 #include <sys/sx.h>
74 #include <sys/sysctl.h>
75 #include <sys/ttycom.h>
76 #include <sys/conf.h>
77 #include <sys/syslog.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 
81 #include <security/audit/audit.h>
82 #include <security/mac/mac_framework.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_extern.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vnode_pager.h>
91 
92 #ifdef HWPMC_HOOKS
93 #include <sys/pmckern.h>
94 #endif
95 
96 static fo_rdwr_t	vn_read;
97 static fo_rdwr_t	vn_write;
98 static fo_rdwr_t	vn_io_fault;
99 static fo_truncate_t	vn_truncate;
100 static fo_ioctl_t	vn_ioctl;
101 static fo_poll_t	vn_poll;
102 static fo_kqfilter_t	vn_kqfilter;
103 static fo_stat_t	vn_statfile;
104 static fo_close_t	vn_closefile;
105 static fo_mmap_t	vn_mmap;
106 
107 struct 	fileops vnops = {
108 	.fo_read = vn_io_fault,
109 	.fo_write = vn_io_fault,
110 	.fo_truncate = vn_truncate,
111 	.fo_ioctl = vn_ioctl,
112 	.fo_poll = vn_poll,
113 	.fo_kqfilter = vn_kqfilter,
114 	.fo_stat = vn_statfile,
115 	.fo_close = vn_closefile,
116 	.fo_chmod = vn_chmod,
117 	.fo_chown = vn_chown,
118 	.fo_sendfile = vn_sendfile,
119 	.fo_seek = vn_seek,
120 	.fo_fill_kinfo = vn_fill_kinfo,
121 	.fo_mmap = vn_mmap,
122 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
123 };
124 
125 static const int io_hold_cnt = 16;
126 static int vn_io_fault_enable = 1;
127 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
128     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
129 static int vn_io_fault_prefault = 0;
130 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
131     &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
132 static u_long vn_io_faults_cnt;
133 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
134     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
135 
136 /*
137  * Returns true if vn_io_fault mode of handling the i/o request should
138  * be used.
139  */
140 static bool
141 do_vn_io_fault(struct vnode *vp, struct uio *uio)
142 {
143 	struct mount *mp;
144 
145 	return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
146 	    (mp = vp->v_mount) != NULL &&
147 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
148 }
149 
150 /*
151  * Structure used to pass arguments to vn_io_fault1(), to do either
152  * file- or vnode-based I/O calls.
153  */
154 struct vn_io_fault_args {
155 	enum {
156 		VN_IO_FAULT_FOP,
157 		VN_IO_FAULT_VOP
158 	} kind;
159 	struct ucred *cred;
160 	int flags;
161 	union {
162 		struct fop_args_tag {
163 			struct file *fp;
164 			fo_rdwr_t *doio;
165 		} fop_args;
166 		struct vop_args_tag {
167 			struct vnode *vp;
168 		} vop_args;
169 	} args;
170 };
171 
172 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
173     struct vn_io_fault_args *args, struct thread *td);
174 
175 int
176 vn_open(struct nameidata *ndp, int *flagp, int cmode, 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 ((fmode & O_BENEATH) != 0)
217 			ndp->ni_cnd.cn_flags |= BENEATH;
218 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
219 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
220 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
221 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
222 		bwillwrite();
223 		if ((error = namei(ndp)) != 0)
224 			return (error);
225 		if (ndp->ni_vp == NULL) {
226 			VATTR_NULL(vap);
227 			vap->va_type = VREG;
228 			vap->va_mode = cmode;
229 			if (fmode & O_EXCL)
230 				vap->va_vaflags |= VA_EXCLUSIVE;
231 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
232 				NDFREE(ndp, NDF_ONLY_PNBUF);
233 				vput(ndp->ni_dvp);
234 				if ((error = vn_start_write(NULL, &mp,
235 				    V_XSLEEP | PCATCH)) != 0)
236 					return (error);
237 				goto restart;
238 			}
239 			if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
240 				ndp->ni_cnd.cn_flags |= MAKEENTRY;
241 #ifdef MAC
242 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
243 			    &ndp->ni_cnd, vap);
244 			if (error == 0)
245 #endif
246 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
247 						   &ndp->ni_cnd, vap);
248 			vput(ndp->ni_dvp);
249 			vn_finished_write(mp);
250 			if (error) {
251 				NDFREE(ndp, NDF_ONLY_PNBUF);
252 				return (error);
253 			}
254 			fmode &= ~O_TRUNC;
255 			vp = ndp->ni_vp;
256 		} else {
257 			if (ndp->ni_dvp == ndp->ni_vp)
258 				vrele(ndp->ni_dvp);
259 			else
260 				vput(ndp->ni_dvp);
261 			ndp->ni_dvp = NULL;
262 			vp = ndp->ni_vp;
263 			if (fmode & O_EXCL) {
264 				error = EEXIST;
265 				goto bad;
266 			}
267 			fmode &= ~O_CREAT;
268 		}
269 	} else {
270 		ndp->ni_cnd.cn_nameiop = LOOKUP;
271 		ndp->ni_cnd.cn_flags = ISOPEN |
272 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
273 		if (!(fmode & FWRITE))
274 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
275 		if ((fmode & O_BENEATH) != 0)
276 			ndp->ni_cnd.cn_flags |= BENEATH;
277 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
278 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
279 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
280 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
281 		if ((error = namei(ndp)) != 0)
282 			return (error);
283 		vp = ndp->ni_vp;
284 	}
285 	error = vn_open_vnode(vp, fmode, cred, td, fp);
286 	if (error)
287 		goto bad;
288 	*flagp = fmode;
289 	return (0);
290 bad:
291 	NDFREE(ndp, NDF_ONLY_PNBUF);
292 	vput(vp);
293 	*flagp = fmode;
294 	ndp->ni_vp = NULL;
295 	return (error);
296 }
297 
298 static int
299 vn_open_vnode_advlock(struct vnode *vp, int fmode, struct file *fp)
300 {
301 	struct flock lf;
302 	int error, lock_flags, type;
303 
304 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode_advlock");
305 	if ((fmode & (O_EXLOCK | O_SHLOCK)) == 0)
306 		return (0);
307 	KASSERT(fp != NULL, ("open with flock requires fp"));
308 	if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE)
309 		return (EOPNOTSUPP);
310 
311 	lock_flags = VOP_ISLOCKED(vp);
312 	VOP_UNLOCK(vp, 0);
313 
314 	lf.l_whence = SEEK_SET;
315 	lf.l_start = 0;
316 	lf.l_len = 0;
317 	lf.l_type = (fmode & O_EXLOCK) != 0 ? F_WRLCK : F_RDLCK;
318 	type = F_FLOCK;
319 	if ((fmode & FNONBLOCK) == 0)
320 		type |= F_WAIT;
321 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
322 	if (error == 0)
323 		fp->f_flag |= FHASLOCK;
324 
325 	vn_lock(vp, lock_flags | LK_RETRY);
326 	if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
327 		error = ENOENT;
328 	return (error);
329 }
330 
331 /*
332  * Common code for vnode open operations once a vnode is located.
333  * Check permissions, and call the VOP_OPEN routine.
334  */
335 int
336 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
337     struct thread *td, struct file *fp)
338 {
339 	accmode_t accmode;
340 	int error;
341 
342 	if (vp->v_type == VLNK)
343 		return (EMLINK);
344 	if (vp->v_type == VSOCK)
345 		return (EOPNOTSUPP);
346 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
347 		return (ENOTDIR);
348 	accmode = 0;
349 	if (fmode & (FWRITE | O_TRUNC)) {
350 		if (vp->v_type == VDIR)
351 			return (EISDIR);
352 		accmode |= VWRITE;
353 	}
354 	if (fmode & FREAD)
355 		accmode |= VREAD;
356 	if (fmode & FEXEC)
357 		accmode |= VEXEC;
358 	if ((fmode & O_APPEND) && (fmode & FWRITE))
359 		accmode |= VAPPEND;
360 #ifdef MAC
361 	if (fmode & O_CREAT)
362 		accmode |= VCREAT;
363 	if (fmode & O_VERIFY)
364 		accmode |= VVERIFY;
365 	error = mac_vnode_check_open(cred, vp, accmode);
366 	if (error)
367 		return (error);
368 
369 	accmode &= ~(VCREAT | VVERIFY);
370 #endif
371 	if ((fmode & O_CREAT) == 0 && accmode != 0) {
372 		error = VOP_ACCESS(vp, accmode, cred, td);
373 		if (error != 0)
374 			return (error);
375 	}
376 	if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
377 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
378 	error = VOP_OPEN(vp, fmode, cred, td, fp);
379 	if (error != 0)
380 		return (error);
381 
382 	error = vn_open_vnode_advlock(vp, fmode, fp);
383 	if (error == 0 && (fmode & FWRITE) != 0) {
384 		error = VOP_ADD_WRITECOUNT(vp, 1);
385 		if (error == 0) {
386 			CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
387 			     __func__, vp, vp->v_writecount);
388 		}
389 	}
390 
391 	/*
392 	 * Error from advlock or VOP_ADD_WRITECOUNT() still requires
393 	 * calling VOP_CLOSE() to pair with earlier VOP_OPEN().
394 	 * Arrange for that by having fdrop() to use vn_closefile().
395 	 */
396 	if (error != 0) {
397 		fp->f_flag |= FOPENFAILED;
398 		fp->f_vnode = vp;
399 		if (fp->f_ops == &badfileops) {
400 			fp->f_type = DTYPE_VNODE;
401 			fp->f_ops = &vnops;
402 		}
403 		vref(vp);
404 	}
405 
406 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
407 	return (error);
408 
409 }
410 
411 /*
412  * Check for write permissions on the specified vnode.
413  * Prototype text segments cannot be written.
414  * It is racy.
415  */
416 int
417 vn_writechk(struct vnode *vp)
418 {
419 
420 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
421 	/*
422 	 * If there's shared text associated with
423 	 * the vnode, try to free it up once.  If
424 	 * we fail, we can't allow writing.
425 	 */
426 	if (VOP_IS_TEXT(vp))
427 		return (ETXTBSY);
428 
429 	return (0);
430 }
431 
432 /*
433  * Vnode close call
434  */
435 static int
436 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred,
437     struct thread *td, bool keep_ref)
438 {
439 	struct mount *mp;
440 	int error, lock_flags;
441 
442 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
443 	    MNT_EXTENDED_SHARED(vp->v_mount))
444 		lock_flags = LK_SHARED;
445 	else
446 		lock_flags = LK_EXCLUSIVE;
447 
448 	vn_start_write(vp, &mp, V_WAIT);
449 	vn_lock(vp, lock_flags | LK_RETRY);
450 	AUDIT_ARG_VNODE1(vp);
451 	if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
452 		VOP_ADD_WRITECOUNT_CHECKED(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 	if (offset < 0 && vp->v_type != VCHR)
533 		return (EINVAL);
534 	auio.uio_iov = &aiov;
535 	auio.uio_iovcnt = 1;
536 	aiov.iov_base = base;
537 	aiov.iov_len = len;
538 	auio.uio_resid = len;
539 	auio.uio_offset = offset;
540 	auio.uio_segflg = segflg;
541 	auio.uio_rw = rw;
542 	auio.uio_td = td;
543 	error = 0;
544 
545 	if ((ioflg & IO_NODELOCKED) == 0) {
546 		if ((ioflg & IO_RANGELOCKED) == 0) {
547 			if (rw == UIO_READ) {
548 				rl_cookie = vn_rangelock_rlock(vp, offset,
549 				    offset + len);
550 			} else {
551 				rl_cookie = vn_rangelock_wlock(vp, offset,
552 				    offset + len);
553 			}
554 		} else
555 			rl_cookie = NULL;
556 		mp = NULL;
557 		if (rw == UIO_WRITE) {
558 			if (vp->v_type != VCHR &&
559 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
560 			    != 0)
561 				goto out;
562 			if (MNT_SHARED_WRITES(mp) ||
563 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
564 				lock_flags = LK_SHARED;
565 			else
566 				lock_flags = LK_EXCLUSIVE;
567 		} else
568 			lock_flags = LK_SHARED;
569 		vn_lock(vp, lock_flags | LK_RETRY);
570 	} else
571 		rl_cookie = NULL;
572 
573 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
574 #ifdef MAC
575 	if ((ioflg & IO_NOMACCHECK) == 0) {
576 		if (rw == UIO_READ)
577 			error = mac_vnode_check_read(active_cred, file_cred,
578 			    vp);
579 		else
580 			error = mac_vnode_check_write(active_cred, file_cred,
581 			    vp);
582 	}
583 #endif
584 	if (error == 0) {
585 		if (file_cred != NULL)
586 			cred = file_cred;
587 		else
588 			cred = active_cred;
589 		if (do_vn_io_fault(vp, &auio)) {
590 			args.kind = VN_IO_FAULT_VOP;
591 			args.cred = cred;
592 			args.flags = ioflg;
593 			args.args.vop_args.vp = vp;
594 			error = vn_io_fault1(vp, &auio, &args, td);
595 		} else if (rw == UIO_READ) {
596 			error = VOP_READ(vp, &auio, ioflg, cred);
597 		} else /* if (rw == UIO_WRITE) */ {
598 			error = VOP_WRITE(vp, &auio, ioflg, cred);
599 		}
600 	}
601 	if (aresid)
602 		*aresid = auio.uio_resid;
603 	else
604 		if (auio.uio_resid && error == 0)
605 			error = EIO;
606 	if ((ioflg & IO_NODELOCKED) == 0) {
607 		VOP_UNLOCK(vp, 0);
608 		if (mp != NULL)
609 			vn_finished_write(mp);
610 	}
611  out:
612 	if (rl_cookie != NULL)
613 		vn_rangelock_unlock(vp, rl_cookie);
614 	return (error);
615 }
616 
617 /*
618  * Package up an I/O request on a vnode into a uio and do it.  The I/O
619  * request is split up into smaller chunks and we try to avoid saturating
620  * the buffer cache while potentially holding a vnode locked, so we
621  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
622  * to give other processes a chance to lock the vnode (either other processes
623  * core'ing the same binary, or unrelated processes scanning the directory).
624  */
625 int
626 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len,
627     off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred,
628     struct ucred *file_cred, size_t *aresid, struct thread *td)
629 {
630 	int error = 0;
631 	ssize_t iaresid;
632 
633 	do {
634 		int chunk;
635 
636 		/*
637 		 * Force `offset' to a multiple of MAXBSIZE except possibly
638 		 * for the first chunk, so that filesystems only need to
639 		 * write full blocks except possibly for the first and last
640 		 * chunks.
641 		 */
642 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
643 
644 		if (chunk > len)
645 			chunk = len;
646 		if (rw != UIO_READ && vp->v_type == VREG)
647 			bwillwrite();
648 		iaresid = 0;
649 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
650 		    ioflg, active_cred, file_cred, &iaresid, td);
651 		len -= chunk;	/* aresid calc already includes length */
652 		if (error)
653 			break;
654 		offset += chunk;
655 		base = (char *)base + chunk;
656 		kern_yield(PRI_USER);
657 	} while (len);
658 	if (aresid)
659 		*aresid = len + iaresid;
660 	return (error);
661 }
662 
663 off_t
664 foffset_lock(struct file *fp, int flags)
665 {
666 	struct mtx *mtxp;
667 	off_t res;
668 
669 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
670 
671 #if OFF_MAX <= LONG_MAX
672 	/*
673 	 * Caller only wants the current f_offset value.  Assume that
674 	 * the long and shorter integer types reads are atomic.
675 	 */
676 	if ((flags & FOF_NOLOCK) != 0)
677 		return (fp->f_offset);
678 #endif
679 
680 	/*
681 	 * According to McKusick the vn lock was protecting f_offset here.
682 	 * It is now protected by the FOFFSET_LOCKED flag.
683 	 */
684 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
685 	mtx_lock(mtxp);
686 	if ((flags & FOF_NOLOCK) == 0) {
687 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
688 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
689 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
690 			    "vofflock", 0);
691 		}
692 		fp->f_vnread_flags |= FOFFSET_LOCKED;
693 	}
694 	res = fp->f_offset;
695 	mtx_unlock(mtxp);
696 	return (res);
697 }
698 
699 void
700 foffset_unlock(struct file *fp, off_t val, int flags)
701 {
702 	struct mtx *mtxp;
703 
704 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
705 
706 #if OFF_MAX <= LONG_MAX
707 	if ((flags & FOF_NOLOCK) != 0) {
708 		if ((flags & FOF_NOUPDATE) == 0)
709 			fp->f_offset = val;
710 		if ((flags & FOF_NEXTOFF) != 0)
711 			fp->f_nextoff = val;
712 		return;
713 	}
714 #endif
715 
716 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
717 	mtx_lock(mtxp);
718 	if ((flags & FOF_NOUPDATE) == 0)
719 		fp->f_offset = val;
720 	if ((flags & FOF_NEXTOFF) != 0)
721 		fp->f_nextoff = val;
722 	if ((flags & FOF_NOLOCK) == 0) {
723 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
724 		    ("Lost FOFFSET_LOCKED"));
725 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
726 			wakeup(&fp->f_vnread_flags);
727 		fp->f_vnread_flags = 0;
728 	}
729 	mtx_unlock(mtxp);
730 }
731 
732 void
733 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
734 {
735 
736 	if ((flags & FOF_OFFSET) == 0)
737 		uio->uio_offset = foffset_lock(fp, flags);
738 }
739 
740 void
741 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
742 {
743 
744 	if ((flags & FOF_OFFSET) == 0)
745 		foffset_unlock(fp, uio->uio_offset, flags);
746 }
747 
748 static int
749 get_advice(struct file *fp, struct uio *uio)
750 {
751 	struct mtx *mtxp;
752 	int ret;
753 
754 	ret = POSIX_FADV_NORMAL;
755 	if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
756 		return (ret);
757 
758 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
759 	mtx_lock(mtxp);
760 	if (fp->f_advice != NULL &&
761 	    uio->uio_offset >= fp->f_advice->fa_start &&
762 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
763 		ret = fp->f_advice->fa_advice;
764 	mtx_unlock(mtxp);
765 	return (ret);
766 }
767 
768 /*
769  * File table vnode read routine.
770  */
771 static int
772 vn_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags,
773     struct thread *td)
774 {
775 	struct vnode *vp;
776 	off_t orig_offset;
777 	int error, ioflag;
778 	int advice;
779 
780 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
781 	    uio->uio_td, td));
782 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
783 	vp = fp->f_vnode;
784 	ioflag = 0;
785 	if (fp->f_flag & FNONBLOCK)
786 		ioflag |= IO_NDELAY;
787 	if (fp->f_flag & O_DIRECT)
788 		ioflag |= IO_DIRECT;
789 	advice = get_advice(fp, uio);
790 	vn_lock(vp, LK_SHARED | LK_RETRY);
791 
792 	switch (advice) {
793 	case POSIX_FADV_NORMAL:
794 	case POSIX_FADV_SEQUENTIAL:
795 	case POSIX_FADV_NOREUSE:
796 		ioflag |= sequential_heuristic(uio, fp);
797 		break;
798 	case POSIX_FADV_RANDOM:
799 		/* Disable read-ahead for random I/O. */
800 		break;
801 	}
802 	orig_offset = uio->uio_offset;
803 
804 #ifdef MAC
805 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
806 	if (error == 0)
807 #endif
808 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
809 	fp->f_nextoff = uio->uio_offset;
810 	VOP_UNLOCK(vp, 0);
811 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
812 	    orig_offset != uio->uio_offset)
813 		/*
814 		 * Use POSIX_FADV_DONTNEED to flush pages and buffers
815 		 * for the backing file after a POSIX_FADV_NOREUSE
816 		 * read(2).
817 		 */
818 		error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
819 		    POSIX_FADV_DONTNEED);
820 	return (error);
821 }
822 
823 /*
824  * File table vnode write routine.
825  */
826 static int
827 vn_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags,
828     struct thread *td)
829 {
830 	struct vnode *vp;
831 	struct mount *mp;
832 	off_t orig_offset;
833 	int error, ioflag, lock_flags;
834 	int advice;
835 
836 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
837 	    uio->uio_td, td));
838 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
839 	vp = fp->f_vnode;
840 	if (vp->v_type == VREG)
841 		bwillwrite();
842 	ioflag = IO_UNIT;
843 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
844 		ioflag |= IO_APPEND;
845 	if (fp->f_flag & FNONBLOCK)
846 		ioflag |= IO_NDELAY;
847 	if (fp->f_flag & O_DIRECT)
848 		ioflag |= IO_DIRECT;
849 	if ((fp->f_flag & O_FSYNC) ||
850 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
851 		ioflag |= IO_SYNC;
852 	mp = NULL;
853 	if (vp->v_type != VCHR &&
854 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
855 		goto unlock;
856 
857 	advice = get_advice(fp, uio);
858 
859 	if (MNT_SHARED_WRITES(mp) ||
860 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
861 		lock_flags = LK_SHARED;
862 	} else {
863 		lock_flags = LK_EXCLUSIVE;
864 	}
865 
866 	vn_lock(vp, lock_flags | LK_RETRY);
867 	switch (advice) {
868 	case POSIX_FADV_NORMAL:
869 	case POSIX_FADV_SEQUENTIAL:
870 	case POSIX_FADV_NOREUSE:
871 		ioflag |= sequential_heuristic(uio, fp);
872 		break;
873 	case POSIX_FADV_RANDOM:
874 		/* XXX: Is this correct? */
875 		break;
876 	}
877 	orig_offset = uio->uio_offset;
878 
879 #ifdef MAC
880 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
881 	if (error == 0)
882 #endif
883 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
884 	fp->f_nextoff = uio->uio_offset;
885 	VOP_UNLOCK(vp, 0);
886 	if (vp->v_type != VCHR)
887 		vn_finished_write(mp);
888 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
889 	    orig_offset != uio->uio_offset)
890 		/*
891 		 * Use POSIX_FADV_DONTNEED to flush pages and buffers
892 		 * for the backing file after a POSIX_FADV_NOREUSE
893 		 * write(2).
894 		 */
895 		error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
896 		    POSIX_FADV_DONTNEED);
897 unlock:
898 	return (error);
899 }
900 
901 /*
902  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
903  * prevent the following deadlock:
904  *
905  * Assume that the thread A reads from the vnode vp1 into userspace
906  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
907  * currently not resident, then system ends up with the call chain
908  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
909  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
910  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
911  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
912  * backed by the pages of vnode vp1, and some page in buf2 is not
913  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
914  *
915  * To prevent the lock order reversal and deadlock, vn_io_fault() does
916  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
917  * Instead, it first tries to do the whole range i/o with pagefaults
918  * disabled. If all pages in the i/o buffer are resident and mapped,
919  * VOP will succeed (ignoring the genuine filesystem errors).
920  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
921  * i/o in chunks, with all pages in the chunk prefaulted and held
922  * using vm_fault_quick_hold_pages().
923  *
924  * Filesystems using this deadlock avoidance scheme should use the
925  * array of the held pages from uio, saved in the curthread->td_ma,
926  * instead of doing uiomove().  A helper function
927  * vn_io_fault_uiomove() converts uiomove request into
928  * uiomove_fromphys() over td_ma array.
929  *
930  * Since vnode locks do not cover the whole i/o anymore, rangelocks
931  * make the current i/o request atomic with respect to other i/os and
932  * truncations.
933  */
934 
935 /*
936  * Decode vn_io_fault_args and perform the corresponding i/o.
937  */
938 static int
939 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
940     struct thread *td)
941 {
942 	int error, save;
943 
944 	error = 0;
945 	save = vm_fault_disable_pagefaults();
946 	switch (args->kind) {
947 	case VN_IO_FAULT_FOP:
948 		error = (args->args.fop_args.doio)(args->args.fop_args.fp,
949 		    uio, args->cred, args->flags, td);
950 		break;
951 	case VN_IO_FAULT_VOP:
952 		if (uio->uio_rw == UIO_READ) {
953 			error = VOP_READ(args->args.vop_args.vp, uio,
954 			    args->flags, args->cred);
955 		} else if (uio->uio_rw == UIO_WRITE) {
956 			error = VOP_WRITE(args->args.vop_args.vp, uio,
957 			    args->flags, args->cred);
958 		}
959 		break;
960 	default:
961 		panic("vn_io_fault_doio: unknown kind of io %d %d",
962 		    args->kind, uio->uio_rw);
963 	}
964 	vm_fault_enable_pagefaults(save);
965 	return (error);
966 }
967 
968 static int
969 vn_io_fault_touch(char *base, const struct uio *uio)
970 {
971 	int r;
972 
973 	r = fubyte(base);
974 	if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
975 		return (EFAULT);
976 	return (0);
977 }
978 
979 static int
980 vn_io_fault_prefault_user(const struct uio *uio)
981 {
982 	char *base;
983 	const struct iovec *iov;
984 	size_t len;
985 	ssize_t resid;
986 	int error, i;
987 
988 	KASSERT(uio->uio_segflg == UIO_USERSPACE,
989 	    ("vn_io_fault_prefault userspace"));
990 
991 	error = i = 0;
992 	iov = uio->uio_iov;
993 	resid = uio->uio_resid;
994 	base = iov->iov_base;
995 	len = iov->iov_len;
996 	while (resid > 0) {
997 		error = vn_io_fault_touch(base, uio);
998 		if (error != 0)
999 			break;
1000 		if (len < PAGE_SIZE) {
1001 			if (len != 0) {
1002 				error = vn_io_fault_touch(base + len - 1, uio);
1003 				if (error != 0)
1004 					break;
1005 				resid -= len;
1006 			}
1007 			if (++i >= uio->uio_iovcnt)
1008 				break;
1009 			iov = uio->uio_iov + i;
1010 			base = iov->iov_base;
1011 			len = iov->iov_len;
1012 		} else {
1013 			len -= PAGE_SIZE;
1014 			base += PAGE_SIZE;
1015 			resid -= PAGE_SIZE;
1016 		}
1017 	}
1018 	return (error);
1019 }
1020 
1021 /*
1022  * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1023  * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1024  * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1025  * into args and call vn_io_fault1() to handle faults during the user
1026  * mode buffer accesses.
1027  */
1028 static int
1029 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1030     struct thread *td)
1031 {
1032 	vm_page_t ma[io_hold_cnt + 2];
1033 	struct uio *uio_clone, short_uio;
1034 	struct iovec short_iovec[1];
1035 	vm_page_t *prev_td_ma;
1036 	vm_prot_t prot;
1037 	vm_offset_t addr, end;
1038 	size_t len, resid;
1039 	ssize_t adv;
1040 	int error, cnt, saveheld, prev_td_ma_cnt;
1041 
1042 	if (vn_io_fault_prefault) {
1043 		error = vn_io_fault_prefault_user(uio);
1044 		if (error != 0)
1045 			return (error); /* Or ignore ? */
1046 	}
1047 
1048 	prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1049 
1050 	/*
1051 	 * The UFS follows IO_UNIT directive and replays back both
1052 	 * uio_offset and uio_resid if an error is encountered during the
1053 	 * operation.  But, since the iovec may be already advanced,
1054 	 * uio is still in an inconsistent state.
1055 	 *
1056 	 * Cache a copy of the original uio, which is advanced to the redo
1057 	 * point using UIO_NOCOPY below.
1058 	 */
1059 	uio_clone = cloneuio(uio);
1060 	resid = uio->uio_resid;
1061 
1062 	short_uio.uio_segflg = UIO_USERSPACE;
1063 	short_uio.uio_rw = uio->uio_rw;
1064 	short_uio.uio_td = uio->uio_td;
1065 
1066 	error = vn_io_fault_doio(args, uio, td);
1067 	if (error != EFAULT)
1068 		goto out;
1069 
1070 	atomic_add_long(&vn_io_faults_cnt, 1);
1071 	uio_clone->uio_segflg = UIO_NOCOPY;
1072 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
1073 	uio_clone->uio_segflg = uio->uio_segflg;
1074 
1075 	saveheld = curthread_pflags_set(TDP_UIOHELD);
1076 	prev_td_ma = td->td_ma;
1077 	prev_td_ma_cnt = td->td_ma_cnt;
1078 
1079 	while (uio_clone->uio_resid != 0) {
1080 		len = uio_clone->uio_iov->iov_len;
1081 		if (len == 0) {
1082 			KASSERT(uio_clone->uio_iovcnt >= 1,
1083 			    ("iovcnt underflow"));
1084 			uio_clone->uio_iov++;
1085 			uio_clone->uio_iovcnt--;
1086 			continue;
1087 		}
1088 		if (len > io_hold_cnt * PAGE_SIZE)
1089 			len = io_hold_cnt * PAGE_SIZE;
1090 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1091 		end = round_page(addr + len);
1092 		if (end < addr) {
1093 			error = EFAULT;
1094 			break;
1095 		}
1096 		cnt = atop(end - trunc_page(addr));
1097 		/*
1098 		 * A perfectly misaligned address and length could cause
1099 		 * both the start and the end of the chunk to use partial
1100 		 * page.  +2 accounts for such a situation.
1101 		 */
1102 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1103 		    addr, len, prot, ma, io_hold_cnt + 2);
1104 		if (cnt == -1) {
1105 			error = EFAULT;
1106 			break;
1107 		}
1108 		short_uio.uio_iov = &short_iovec[0];
1109 		short_iovec[0].iov_base = (void *)addr;
1110 		short_uio.uio_iovcnt = 1;
1111 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1112 		short_uio.uio_offset = uio_clone->uio_offset;
1113 		td->td_ma = ma;
1114 		td->td_ma_cnt = cnt;
1115 
1116 		error = vn_io_fault_doio(args, &short_uio, td);
1117 		vm_page_unhold_pages(ma, cnt);
1118 		adv = len - short_uio.uio_resid;
1119 
1120 		uio_clone->uio_iov->iov_base =
1121 		    (char *)uio_clone->uio_iov->iov_base + adv;
1122 		uio_clone->uio_iov->iov_len -= adv;
1123 		uio_clone->uio_resid -= adv;
1124 		uio_clone->uio_offset += adv;
1125 
1126 		uio->uio_resid -= adv;
1127 		uio->uio_offset += adv;
1128 
1129 		if (error != 0 || adv == 0)
1130 			break;
1131 	}
1132 	td->td_ma = prev_td_ma;
1133 	td->td_ma_cnt = prev_td_ma_cnt;
1134 	curthread_pflags_restore(saveheld);
1135 out:
1136 	free(uio_clone, M_IOV);
1137 	return (error);
1138 }
1139 
1140 static int
1141 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1142     int flags, struct thread *td)
1143 {
1144 	fo_rdwr_t *doio;
1145 	struct vnode *vp;
1146 	void *rl_cookie;
1147 	struct vn_io_fault_args args;
1148 	int error;
1149 
1150 	doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1151 	vp = fp->f_vnode;
1152 	foffset_lock_uio(fp, uio, flags);
1153 	if (do_vn_io_fault(vp, uio)) {
1154 		args.kind = VN_IO_FAULT_FOP;
1155 		args.args.fop_args.fp = fp;
1156 		args.args.fop_args.doio = doio;
1157 		args.cred = active_cred;
1158 		args.flags = flags | FOF_OFFSET;
1159 		if (uio->uio_rw == UIO_READ) {
1160 			rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1161 			    uio->uio_offset + uio->uio_resid);
1162 		} else if ((fp->f_flag & O_APPEND) != 0 ||
1163 		    (flags & FOF_OFFSET) == 0) {
1164 			/* For appenders, punt and lock the whole range. */
1165 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1166 		} else {
1167 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1168 			    uio->uio_offset + uio->uio_resid);
1169 		}
1170 		error = vn_io_fault1(vp, uio, &args, td);
1171 		vn_rangelock_unlock(vp, rl_cookie);
1172 	} else {
1173 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1174 	}
1175 	foffset_unlock_uio(fp, uio, flags);
1176 	return (error);
1177 }
1178 
1179 /*
1180  * Helper function to perform the requested uiomove operation using
1181  * the held pages for io->uio_iov[0].iov_base buffer instead of
1182  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1183  * instead of iov_base prevents page faults that could occur due to
1184  * pmap_collect() invalidating the mapping created by
1185  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1186  * object cleanup revoking the write access from page mappings.
1187  *
1188  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1189  * instead of plain uiomove().
1190  */
1191 int
1192 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1193 {
1194 	struct uio transp_uio;
1195 	struct iovec transp_iov[1];
1196 	struct thread *td;
1197 	size_t adv;
1198 	int error, pgadv;
1199 
1200 	td = curthread;
1201 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1202 	    uio->uio_segflg != UIO_USERSPACE)
1203 		return (uiomove(data, xfersize, uio));
1204 
1205 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1206 	transp_iov[0].iov_base = data;
1207 	transp_uio.uio_iov = &transp_iov[0];
1208 	transp_uio.uio_iovcnt = 1;
1209 	if (xfersize > uio->uio_resid)
1210 		xfersize = uio->uio_resid;
1211 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1212 	transp_uio.uio_offset = 0;
1213 	transp_uio.uio_segflg = UIO_SYSSPACE;
1214 	/*
1215 	 * Since transp_iov points to data, and td_ma page array
1216 	 * corresponds to original uio->uio_iov, we need to invert the
1217 	 * direction of the i/o operation as passed to
1218 	 * uiomove_fromphys().
1219 	 */
1220 	switch (uio->uio_rw) {
1221 	case UIO_WRITE:
1222 		transp_uio.uio_rw = UIO_READ;
1223 		break;
1224 	case UIO_READ:
1225 		transp_uio.uio_rw = UIO_WRITE;
1226 		break;
1227 	}
1228 	transp_uio.uio_td = uio->uio_td;
1229 	error = uiomove_fromphys(td->td_ma,
1230 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1231 	    xfersize, &transp_uio);
1232 	adv = xfersize - transp_uio.uio_resid;
1233 	pgadv =
1234 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1235 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1236 	td->td_ma += pgadv;
1237 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1238 	    pgadv));
1239 	td->td_ma_cnt -= pgadv;
1240 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1241 	uio->uio_iov->iov_len -= adv;
1242 	uio->uio_resid -= adv;
1243 	uio->uio_offset += adv;
1244 	return (error);
1245 }
1246 
1247 int
1248 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1249     struct uio *uio)
1250 {
1251 	struct thread *td;
1252 	vm_offset_t iov_base;
1253 	int cnt, pgadv;
1254 
1255 	td = curthread;
1256 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1257 	    uio->uio_segflg != UIO_USERSPACE)
1258 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1259 
1260 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1261 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1262 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1263 	switch (uio->uio_rw) {
1264 	case UIO_WRITE:
1265 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1266 		    offset, cnt);
1267 		break;
1268 	case UIO_READ:
1269 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1270 		    cnt);
1271 		break;
1272 	}
1273 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1274 	td->td_ma += pgadv;
1275 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1276 	    pgadv));
1277 	td->td_ma_cnt -= pgadv;
1278 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1279 	uio->uio_iov->iov_len -= cnt;
1280 	uio->uio_resid -= cnt;
1281 	uio->uio_offset += cnt;
1282 	return (0);
1283 }
1284 
1285 
1286 /*
1287  * File table truncate routine.
1288  */
1289 static int
1290 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1291     struct thread *td)
1292 {
1293 	struct vattr vattr;
1294 	struct mount *mp;
1295 	struct vnode *vp;
1296 	void *rl_cookie;
1297 	int error;
1298 
1299 	vp = fp->f_vnode;
1300 
1301 	/*
1302 	 * Lock the whole range for truncation.  Otherwise split i/o
1303 	 * might happen partly before and partly after the truncation.
1304 	 */
1305 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1306 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1307 	if (error)
1308 		goto out1;
1309 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1310 	AUDIT_ARG_VNODE1(vp);
1311 	if (vp->v_type == VDIR) {
1312 		error = EISDIR;
1313 		goto out;
1314 	}
1315 #ifdef MAC
1316 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1317 	if (error)
1318 		goto out;
1319 #endif
1320 	error = VOP_ADD_WRITECOUNT(vp, 1);
1321 	if (error == 0) {
1322 		VATTR_NULL(&vattr);
1323 		vattr.va_size = length;
1324 		if ((fp->f_flag & O_FSYNC) != 0)
1325 			vattr.va_vaflags |= VA_SYNC;
1326 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1327 		VOP_ADD_WRITECOUNT_CHECKED(vp, -1);
1328 	}
1329 out:
1330 	VOP_UNLOCK(vp, 0);
1331 	vn_finished_write(mp);
1332 out1:
1333 	vn_rangelock_unlock(vp, rl_cookie);
1334 	return (error);
1335 }
1336 
1337 /*
1338  * File table vnode stat routine.
1339  */
1340 static int
1341 vn_statfile(struct file *fp, struct stat *sb, struct ucred *active_cred,
1342     struct thread *td)
1343 {
1344 	struct vnode *vp = fp->f_vnode;
1345 	int error;
1346 
1347 	vn_lock(vp, LK_SHARED | LK_RETRY);
1348 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1349 	VOP_UNLOCK(vp, 0);
1350 
1351 	return (error);
1352 }
1353 
1354 /*
1355  * Stat a vnode; implementation for the stat syscall
1356  */
1357 int
1358 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred,
1359     struct ucred *file_cred, struct thread *td)
1360 {
1361 	struct vattr vattr;
1362 	struct vattr *vap;
1363 	int error;
1364 	u_short mode;
1365 
1366 	AUDIT_ARG_VNODE1(vp);
1367 #ifdef MAC
1368 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1369 	if (error)
1370 		return (error);
1371 #endif
1372 
1373 	vap = &vattr;
1374 
1375 	/*
1376 	 * Initialize defaults for new and unusual fields, so that file
1377 	 * systems which don't support these fields don't need to know
1378 	 * about them.
1379 	 */
1380 	vap->va_birthtime.tv_sec = -1;
1381 	vap->va_birthtime.tv_nsec = 0;
1382 	vap->va_fsid = VNOVAL;
1383 	vap->va_rdev = NODEV;
1384 
1385 	error = VOP_GETATTR(vp, vap, active_cred);
1386 	if (error)
1387 		return (error);
1388 
1389 	/*
1390 	 * Zero the spare stat fields
1391 	 */
1392 	bzero(sb, sizeof *sb);
1393 
1394 	/*
1395 	 * Copy from vattr table
1396 	 */
1397 	if (vap->va_fsid != VNOVAL)
1398 		sb->st_dev = vap->va_fsid;
1399 	else
1400 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1401 	sb->st_ino = vap->va_fileid;
1402 	mode = vap->va_mode;
1403 	switch (vap->va_type) {
1404 	case VREG:
1405 		mode |= S_IFREG;
1406 		break;
1407 	case VDIR:
1408 		mode |= S_IFDIR;
1409 		break;
1410 	case VBLK:
1411 		mode |= S_IFBLK;
1412 		break;
1413 	case VCHR:
1414 		mode |= S_IFCHR;
1415 		break;
1416 	case VLNK:
1417 		mode |= S_IFLNK;
1418 		break;
1419 	case VSOCK:
1420 		mode |= S_IFSOCK;
1421 		break;
1422 	case VFIFO:
1423 		mode |= S_IFIFO;
1424 		break;
1425 	default:
1426 		return (EBADF);
1427 	}
1428 	sb->st_mode = mode;
1429 	sb->st_nlink = vap->va_nlink;
1430 	sb->st_uid = vap->va_uid;
1431 	sb->st_gid = vap->va_gid;
1432 	sb->st_rdev = vap->va_rdev;
1433 	if (vap->va_size > OFF_MAX)
1434 		return (EOVERFLOW);
1435 	sb->st_size = vap->va_size;
1436 	sb->st_atim = vap->va_atime;
1437 	sb->st_mtim = vap->va_mtime;
1438 	sb->st_ctim = vap->va_ctime;
1439 	sb->st_birthtim = vap->va_birthtime;
1440 
1441         /*
1442 	 * According to www.opengroup.org, the meaning of st_blksize is
1443 	 *   "a filesystem-specific preferred I/O block size for this
1444 	 *    object.  In some filesystem types, this may vary from file
1445 	 *    to file"
1446 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1447 	 */
1448 
1449 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1450 
1451 	sb->st_flags = vap->va_flags;
1452 	if (priv_check(td, PRIV_VFS_GENERATION))
1453 		sb->st_gen = 0;
1454 	else
1455 		sb->st_gen = vap->va_gen;
1456 
1457 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1458 	return (0);
1459 }
1460 
1461 /*
1462  * File table vnode ioctl routine.
1463  */
1464 static int
1465 vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
1466     struct thread *td)
1467 {
1468 	struct vattr vattr;
1469 	struct vnode *vp;
1470 	int error;
1471 
1472 	vp = fp->f_vnode;
1473 	switch (vp->v_type) {
1474 	case VDIR:
1475 	case VREG:
1476 		switch (com) {
1477 		case FIONREAD:
1478 			vn_lock(vp, LK_SHARED | LK_RETRY);
1479 			error = VOP_GETATTR(vp, &vattr, active_cred);
1480 			VOP_UNLOCK(vp, 0);
1481 			if (error == 0)
1482 				*(int *)data = vattr.va_size - fp->f_offset;
1483 			return (error);
1484 		case FIONBIO:
1485 		case FIOASYNC:
1486 			return (0);
1487 		default:
1488 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1489 			    active_cred, td));
1490 		}
1491 		break;
1492 	case VCHR:
1493 		return (VOP_IOCTL(vp, com, data, fp->f_flag,
1494 		    active_cred, td));
1495 	default:
1496 		return (ENOTTY);
1497 	}
1498 }
1499 
1500 /*
1501  * File table vnode poll routine.
1502  */
1503 static int
1504 vn_poll(struct file *fp, int events, struct ucred *active_cred,
1505     struct thread *td)
1506 {
1507 	struct vnode *vp;
1508 	int error;
1509 
1510 	vp = fp->f_vnode;
1511 #ifdef MAC
1512 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1513 	AUDIT_ARG_VNODE1(vp);
1514 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1515 	VOP_UNLOCK(vp, 0);
1516 	if (!error)
1517 #endif
1518 
1519 	error = VOP_POLL(vp, events, fp->f_cred, td);
1520 	return (error);
1521 }
1522 
1523 /*
1524  * Acquire the requested lock and then check for validity.  LK_RETRY
1525  * permits vn_lock to return doomed vnodes.
1526  */
1527 int
1528 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1529 {
1530 	int error;
1531 
1532 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1533 	    ("vn_lock: no locktype"));
1534 	VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count"));
1535 retry:
1536 	error = VOP_LOCK1(vp, flags, file, line);
1537 	flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1538 	KASSERT((flags & LK_RETRY) == 0 || error == 0,
1539 	    ("vn_lock: error %d incompatible with flags %#x", error, flags));
1540 
1541 	if ((flags & LK_RETRY) == 0) {
1542 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) {
1543 			VOP_UNLOCK(vp, 0);
1544 			error = ENOENT;
1545 		}
1546 	} else if (error != 0)
1547 		goto retry;
1548 	return (error);
1549 }
1550 
1551 /*
1552  * File table vnode close routine.
1553  */
1554 static int
1555 vn_closefile(struct file *fp, struct thread *td)
1556 {
1557 	struct vnode *vp;
1558 	struct flock lf;
1559 	int error;
1560 	bool ref;
1561 
1562 	vp = fp->f_vnode;
1563 	fp->f_ops = &badfileops;
1564 	ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE;
1565 
1566 	error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref);
1567 
1568 	if (__predict_false(ref)) {
1569 		lf.l_whence = SEEK_SET;
1570 		lf.l_start = 0;
1571 		lf.l_len = 0;
1572 		lf.l_type = F_UNLCK;
1573 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1574 		vrele(vp);
1575 	}
1576 	return (error);
1577 }
1578 
1579 static bool
1580 vn_suspendable(struct mount *mp)
1581 {
1582 
1583 	return (mp->mnt_op->vfs_susp_clean != NULL);
1584 }
1585 
1586 /*
1587  * Preparing to start a filesystem write operation. If the operation is
1588  * permitted, then we bump the count of operations in progress and
1589  * proceed. If a suspend request is in progress, we wait until the
1590  * suspension is over, and then proceed.
1591  */
1592 static int
1593 vn_start_write_locked(struct mount *mp, int flags)
1594 {
1595 	int error, mflags;
1596 
1597 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1598 	error = 0;
1599 
1600 	/*
1601 	 * Check on status of suspension.
1602 	 */
1603 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1604 	    mp->mnt_susp_owner != curthread) {
1605 		mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1606 		    (flags & PCATCH) : 0) | (PUSER - 1);
1607 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1608 			if (flags & V_NOWAIT) {
1609 				error = EWOULDBLOCK;
1610 				goto unlock;
1611 			}
1612 			error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1613 			    "suspfs", 0);
1614 			if (error)
1615 				goto unlock;
1616 		}
1617 	}
1618 	if (flags & V_XSLEEP)
1619 		goto unlock;
1620 	mp->mnt_writeopcount++;
1621 unlock:
1622 	if (error != 0 || (flags & V_XSLEEP) != 0)
1623 		MNT_REL(mp);
1624 	MNT_IUNLOCK(mp);
1625 	return (error);
1626 }
1627 
1628 int
1629 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1630 {
1631 	struct mount *mp;
1632 	int error;
1633 
1634 	KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1635 	    ("V_MNTREF requires mp"));
1636 
1637 	error = 0;
1638 	/*
1639 	 * If a vnode is provided, get and return the mount point that
1640 	 * to which it will write.
1641 	 */
1642 	if (vp != NULL) {
1643 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1644 			*mpp = NULL;
1645 			if (error != EOPNOTSUPP)
1646 				return (error);
1647 			return (0);
1648 		}
1649 	}
1650 	if ((mp = *mpp) == NULL)
1651 		return (0);
1652 
1653 	if (!vn_suspendable(mp)) {
1654 		if (vp != NULL || (flags & V_MNTREF) != 0)
1655 			vfs_rel(mp);
1656 		return (0);
1657 	}
1658 
1659 	/*
1660 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1661 	 * a vfs_ref().
1662 	 * As long as a vnode is not provided we need to acquire a
1663 	 * refcount for the provided mountpoint too, in order to
1664 	 * emulate a vfs_ref().
1665 	 */
1666 	MNT_ILOCK(mp);
1667 	if (vp == NULL && (flags & V_MNTREF) == 0)
1668 		MNT_REF(mp);
1669 
1670 	return (vn_start_write_locked(mp, flags));
1671 }
1672 
1673 /*
1674  * Secondary suspension. Used by operations such as vop_inactive
1675  * routines that are needed by the higher level functions. These
1676  * are allowed to proceed until all the higher level functions have
1677  * completed (indicated by mnt_writeopcount dropping to zero). At that
1678  * time, these operations are halted until the suspension is over.
1679  */
1680 int
1681 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1682 {
1683 	struct mount *mp;
1684 	int error;
1685 
1686 	KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1687 	    ("V_MNTREF requires mp"));
1688 
1689  retry:
1690 	if (vp != NULL) {
1691 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1692 			*mpp = NULL;
1693 			if (error != EOPNOTSUPP)
1694 				return (error);
1695 			return (0);
1696 		}
1697 	}
1698 	/*
1699 	 * If we are not suspended or have not yet reached suspended
1700 	 * mode, then let the operation proceed.
1701 	 */
1702 	if ((mp = *mpp) == NULL)
1703 		return (0);
1704 
1705 	if (!vn_suspendable(mp)) {
1706 		if (vp != NULL || (flags & V_MNTREF) != 0)
1707 			vfs_rel(mp);
1708 		return (0);
1709 	}
1710 
1711 	/*
1712 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1713 	 * a vfs_ref().
1714 	 * As long as a vnode is not provided we need to acquire a
1715 	 * refcount for the provided mountpoint too, in order to
1716 	 * emulate a vfs_ref().
1717 	 */
1718 	MNT_ILOCK(mp);
1719 	if (vp == NULL && (flags & V_MNTREF) == 0)
1720 		MNT_REF(mp);
1721 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1722 		mp->mnt_secondary_writes++;
1723 		mp->mnt_secondary_accwrites++;
1724 		MNT_IUNLOCK(mp);
1725 		return (0);
1726 	}
1727 	if (flags & V_NOWAIT) {
1728 		MNT_REL(mp);
1729 		MNT_IUNLOCK(mp);
1730 		return (EWOULDBLOCK);
1731 	}
1732 	/*
1733 	 * Wait for the suspension to finish.
1734 	 */
1735 	error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1736 	    ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1737 	    "suspfs", 0);
1738 	vfs_rel(mp);
1739 	if (error == 0)
1740 		goto retry;
1741 	return (error);
1742 }
1743 
1744 /*
1745  * Filesystem write operation has completed. If we are suspending and this
1746  * operation is the last one, notify the suspender that the suspension is
1747  * now in effect.
1748  */
1749 void
1750 vn_finished_write(struct mount *mp)
1751 {
1752 	if (mp == NULL || !vn_suspendable(mp))
1753 		return;
1754 	MNT_ILOCK(mp);
1755 	MNT_REL(mp);
1756 	mp->mnt_writeopcount--;
1757 	if (mp->mnt_writeopcount < 0)
1758 		panic("vn_finished_write: neg cnt");
1759 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1760 	    mp->mnt_writeopcount <= 0)
1761 		wakeup(&mp->mnt_writeopcount);
1762 	MNT_IUNLOCK(mp);
1763 }
1764 
1765 
1766 /*
1767  * Filesystem secondary write operation has completed. If we are
1768  * suspending and this operation is the last one, notify the suspender
1769  * that the suspension is now in effect.
1770  */
1771 void
1772 vn_finished_secondary_write(struct mount *mp)
1773 {
1774 	if (mp == NULL || !vn_suspendable(mp))
1775 		return;
1776 	MNT_ILOCK(mp);
1777 	MNT_REL(mp);
1778 	mp->mnt_secondary_writes--;
1779 	if (mp->mnt_secondary_writes < 0)
1780 		panic("vn_finished_secondary_write: neg cnt");
1781 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1782 	    mp->mnt_secondary_writes <= 0)
1783 		wakeup(&mp->mnt_secondary_writes);
1784 	MNT_IUNLOCK(mp);
1785 }
1786 
1787 
1788 
1789 /*
1790  * Request a filesystem to suspend write operations.
1791  */
1792 int
1793 vfs_write_suspend(struct mount *mp, int flags)
1794 {
1795 	int error;
1796 
1797 	MPASS(vn_suspendable(mp));
1798 
1799 	MNT_ILOCK(mp);
1800 	if (mp->mnt_susp_owner == curthread) {
1801 		MNT_IUNLOCK(mp);
1802 		return (EALREADY);
1803 	}
1804 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1805 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1806 
1807 	/*
1808 	 * Unmount holds a write reference on the mount point.  If we
1809 	 * own busy reference and drain for writers, we deadlock with
1810 	 * the reference draining in the unmount path.  Callers of
1811 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1812 	 * vfs_busy() reference is owned and caller is not in the
1813 	 * unmount context.
1814 	 */
1815 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1816 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1817 		MNT_IUNLOCK(mp);
1818 		return (EBUSY);
1819 	}
1820 
1821 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1822 	mp->mnt_susp_owner = curthread;
1823 	if (mp->mnt_writeopcount > 0)
1824 		(void) msleep(&mp->mnt_writeopcount,
1825 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1826 	else
1827 		MNT_IUNLOCK(mp);
1828 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1829 		vfs_write_resume(mp, 0);
1830 	return (error);
1831 }
1832 
1833 /*
1834  * Request a filesystem to resume write operations.
1835  */
1836 void
1837 vfs_write_resume(struct mount *mp, int flags)
1838 {
1839 
1840 	MPASS(vn_suspendable(mp));
1841 
1842 	MNT_ILOCK(mp);
1843 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1844 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1845 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1846 				       MNTK_SUSPENDED);
1847 		mp->mnt_susp_owner = NULL;
1848 		wakeup(&mp->mnt_writeopcount);
1849 		wakeup(&mp->mnt_flag);
1850 		curthread->td_pflags &= ~TDP_IGNSUSP;
1851 		if ((flags & VR_START_WRITE) != 0) {
1852 			MNT_REF(mp);
1853 			mp->mnt_writeopcount++;
1854 		}
1855 		MNT_IUNLOCK(mp);
1856 		if ((flags & VR_NO_SUSPCLR) == 0)
1857 			VFS_SUSP_CLEAN(mp);
1858 	} else if ((flags & VR_START_WRITE) != 0) {
1859 		MNT_REF(mp);
1860 		vn_start_write_locked(mp, 0);
1861 	} else {
1862 		MNT_IUNLOCK(mp);
1863 	}
1864 }
1865 
1866 /*
1867  * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1868  * methods.
1869  */
1870 int
1871 vfs_write_suspend_umnt(struct mount *mp)
1872 {
1873 	int error;
1874 
1875 	MPASS(vn_suspendable(mp));
1876 	KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1877 	    ("vfs_write_suspend_umnt: recursed"));
1878 
1879 	/* dounmount() already called vn_start_write(). */
1880 	for (;;) {
1881 		vn_finished_write(mp);
1882 		error = vfs_write_suspend(mp, 0);
1883 		if (error != 0) {
1884 			vn_start_write(NULL, &mp, V_WAIT);
1885 			return (error);
1886 		}
1887 		MNT_ILOCK(mp);
1888 		if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1889 			break;
1890 		MNT_IUNLOCK(mp);
1891 		vn_start_write(NULL, &mp, V_WAIT);
1892 	}
1893 	mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1894 	wakeup(&mp->mnt_flag);
1895 	MNT_IUNLOCK(mp);
1896 	curthread->td_pflags |= TDP_IGNSUSP;
1897 	return (0);
1898 }
1899 
1900 /*
1901  * Implement kqueues for files by translating it to vnode operation.
1902  */
1903 static int
1904 vn_kqfilter(struct file *fp, struct knote *kn)
1905 {
1906 
1907 	return (VOP_KQFILTER(fp->f_vnode, kn));
1908 }
1909 
1910 /*
1911  * Simplified in-kernel wrapper calls for extended attribute access.
1912  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1913  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1914  */
1915 int
1916 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1917     const char *attrname, int *buflen, char *buf, struct thread *td)
1918 {
1919 	struct uio	auio;
1920 	struct iovec	iov;
1921 	int	error;
1922 
1923 	iov.iov_len = *buflen;
1924 	iov.iov_base = buf;
1925 
1926 	auio.uio_iov = &iov;
1927 	auio.uio_iovcnt = 1;
1928 	auio.uio_rw = UIO_READ;
1929 	auio.uio_segflg = UIO_SYSSPACE;
1930 	auio.uio_td = td;
1931 	auio.uio_offset = 0;
1932 	auio.uio_resid = *buflen;
1933 
1934 	if ((ioflg & IO_NODELOCKED) == 0)
1935 		vn_lock(vp, LK_SHARED | LK_RETRY);
1936 
1937 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1938 
1939 	/* authorize attribute retrieval as kernel */
1940 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1941 	    td);
1942 
1943 	if ((ioflg & IO_NODELOCKED) == 0)
1944 		VOP_UNLOCK(vp, 0);
1945 
1946 	if (error == 0) {
1947 		*buflen = *buflen - auio.uio_resid;
1948 	}
1949 
1950 	return (error);
1951 }
1952 
1953 /*
1954  * XXX failure mode if partially written?
1955  */
1956 int
1957 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1958     const char *attrname, int buflen, char *buf, struct thread *td)
1959 {
1960 	struct uio	auio;
1961 	struct iovec	iov;
1962 	struct mount	*mp;
1963 	int	error;
1964 
1965 	iov.iov_len = buflen;
1966 	iov.iov_base = buf;
1967 
1968 	auio.uio_iov = &iov;
1969 	auio.uio_iovcnt = 1;
1970 	auio.uio_rw = UIO_WRITE;
1971 	auio.uio_segflg = UIO_SYSSPACE;
1972 	auio.uio_td = td;
1973 	auio.uio_offset = 0;
1974 	auio.uio_resid = buflen;
1975 
1976 	if ((ioflg & IO_NODELOCKED) == 0) {
1977 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1978 			return (error);
1979 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1980 	}
1981 
1982 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1983 
1984 	/* authorize attribute setting as kernel */
1985 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1986 
1987 	if ((ioflg & IO_NODELOCKED) == 0) {
1988 		vn_finished_write(mp);
1989 		VOP_UNLOCK(vp, 0);
1990 	}
1991 
1992 	return (error);
1993 }
1994 
1995 int
1996 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1997     const char *attrname, struct thread *td)
1998 {
1999 	struct mount	*mp;
2000 	int	error;
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 removal as kernel */
2011 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2012 	if (error == EOPNOTSUPP)
2013 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2014 		    NULL, td);
2015 
2016 	if ((ioflg & IO_NODELOCKED) == 0) {
2017 		vn_finished_write(mp);
2018 		VOP_UNLOCK(vp, 0);
2019 	}
2020 
2021 	return (error);
2022 }
2023 
2024 static int
2025 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2026     struct vnode **rvp)
2027 {
2028 
2029 	return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2030 }
2031 
2032 int
2033 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2034 {
2035 
2036 	return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2037 	    lkflags, rvp));
2038 }
2039 
2040 int
2041 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2042     int lkflags, struct vnode **rvp)
2043 {
2044 	struct mount *mp;
2045 	int ltype, error;
2046 
2047 	ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2048 	mp = vp->v_mount;
2049 	ltype = VOP_ISLOCKED(vp);
2050 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2051 	    ("vn_vget_ino: vp not locked"));
2052 	error = vfs_busy(mp, MBF_NOWAIT);
2053 	if (error != 0) {
2054 		vfs_ref(mp);
2055 		VOP_UNLOCK(vp, 0);
2056 		error = vfs_busy(mp, 0);
2057 		vn_lock(vp, ltype | LK_RETRY);
2058 		vfs_rel(mp);
2059 		if (error != 0)
2060 			return (ENOENT);
2061 		if (vp->v_iflag & VI_DOOMED) {
2062 			vfs_unbusy(mp);
2063 			return (ENOENT);
2064 		}
2065 	}
2066 	VOP_UNLOCK(vp, 0);
2067 	error = alloc(mp, alloc_arg, lkflags, rvp);
2068 	vfs_unbusy(mp);
2069 	if (*rvp != vp)
2070 		vn_lock(vp, ltype | LK_RETRY);
2071 	if (vp->v_iflag & VI_DOOMED) {
2072 		if (error == 0) {
2073 			if (*rvp == vp)
2074 				vunref(vp);
2075 			else
2076 				vput(*rvp);
2077 		}
2078 		error = ENOENT;
2079 	}
2080 	return (error);
2081 }
2082 
2083 int
2084 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2085     struct thread *td)
2086 {
2087 
2088 	if (vp->v_type != VREG || td == NULL)
2089 		return (0);
2090 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
2091 	    lim_cur(td, RLIMIT_FSIZE)) {
2092 		PROC_LOCK(td->td_proc);
2093 		kern_psignal(td->td_proc, SIGXFSZ);
2094 		PROC_UNLOCK(td->td_proc);
2095 		return (EFBIG);
2096 	}
2097 	return (0);
2098 }
2099 
2100 int
2101 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2102     struct thread *td)
2103 {
2104 	struct vnode *vp;
2105 
2106 	vp = fp->f_vnode;
2107 #ifdef AUDIT
2108 	vn_lock(vp, LK_SHARED | LK_RETRY);
2109 	AUDIT_ARG_VNODE1(vp);
2110 	VOP_UNLOCK(vp, 0);
2111 #endif
2112 	return (setfmode(td, active_cred, vp, mode));
2113 }
2114 
2115 int
2116 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2117     struct thread *td)
2118 {
2119 	struct vnode *vp;
2120 
2121 	vp = fp->f_vnode;
2122 #ifdef AUDIT
2123 	vn_lock(vp, LK_SHARED | LK_RETRY);
2124 	AUDIT_ARG_VNODE1(vp);
2125 	VOP_UNLOCK(vp, 0);
2126 #endif
2127 	return (setfown(td, active_cred, vp, uid, gid));
2128 }
2129 
2130 void
2131 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2132 {
2133 	vm_object_t object;
2134 
2135 	if ((object = vp->v_object) == NULL)
2136 		return;
2137 	VM_OBJECT_WLOCK(object);
2138 	vm_object_page_remove(object, start, end, 0);
2139 	VM_OBJECT_WUNLOCK(object);
2140 }
2141 
2142 int
2143 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2144 {
2145 	struct vattr va;
2146 	daddr_t bn, bnp;
2147 	uint64_t bsize;
2148 	off_t noff;
2149 	int error;
2150 
2151 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2152 	    ("Wrong command %lu", cmd));
2153 
2154 	if (vn_lock(vp, LK_SHARED) != 0)
2155 		return (EBADF);
2156 	if (vp->v_type != VREG) {
2157 		error = ENOTTY;
2158 		goto unlock;
2159 	}
2160 	error = VOP_GETATTR(vp, &va, cred);
2161 	if (error != 0)
2162 		goto unlock;
2163 	noff = *off;
2164 	if (noff >= va.va_size) {
2165 		error = ENXIO;
2166 		goto unlock;
2167 	}
2168 	bsize = vp->v_mount->mnt_stat.f_iosize;
2169 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize -
2170 	    noff % bsize) {
2171 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2172 		if (error == EOPNOTSUPP) {
2173 			error = ENOTTY;
2174 			goto unlock;
2175 		}
2176 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2177 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
2178 			noff = bn * bsize;
2179 			if (noff < *off)
2180 				noff = *off;
2181 			goto unlock;
2182 		}
2183 	}
2184 	if (noff > va.va_size)
2185 		noff = va.va_size;
2186 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2187 	if (cmd == FIOSEEKDATA)
2188 		error = ENXIO;
2189 unlock:
2190 	VOP_UNLOCK(vp, 0);
2191 	if (error == 0)
2192 		*off = noff;
2193 	return (error);
2194 }
2195 
2196 int
2197 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2198 {
2199 	struct ucred *cred;
2200 	struct vnode *vp;
2201 	struct vattr vattr;
2202 	off_t foffset, size;
2203 	int error, noneg;
2204 
2205 	cred = td->td_ucred;
2206 	vp = fp->f_vnode;
2207 	foffset = foffset_lock(fp, 0);
2208 	noneg = (vp->v_type != VCHR);
2209 	error = 0;
2210 	switch (whence) {
2211 	case L_INCR:
2212 		if (noneg &&
2213 		    (foffset < 0 ||
2214 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2215 			error = EOVERFLOW;
2216 			break;
2217 		}
2218 		offset += foffset;
2219 		break;
2220 	case L_XTND:
2221 		vn_lock(vp, LK_SHARED | LK_RETRY);
2222 		error = VOP_GETATTR(vp, &vattr, cred);
2223 		VOP_UNLOCK(vp, 0);
2224 		if (error)
2225 			break;
2226 
2227 		/*
2228 		 * If the file references a disk device, then fetch
2229 		 * the media size and use that to determine the ending
2230 		 * offset.
2231 		 */
2232 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2233 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2234 			vattr.va_size = size;
2235 		if (noneg &&
2236 		    (vattr.va_size > OFF_MAX ||
2237 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2238 			error = EOVERFLOW;
2239 			break;
2240 		}
2241 		offset += vattr.va_size;
2242 		break;
2243 	case L_SET:
2244 		break;
2245 	case SEEK_DATA:
2246 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2247 		break;
2248 	case SEEK_HOLE:
2249 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2250 		break;
2251 	default:
2252 		error = EINVAL;
2253 	}
2254 	if (error == 0 && noneg && offset < 0)
2255 		error = EINVAL;
2256 	if (error != 0)
2257 		goto drop;
2258 	VFS_KNOTE_UNLOCKED(vp, 0);
2259 	td->td_uretoff.tdu_off = offset;
2260 drop:
2261 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2262 	return (error);
2263 }
2264 
2265 int
2266 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2267     struct thread *td)
2268 {
2269 	int error;
2270 
2271 	/*
2272 	 * Grant permission if the caller is the owner of the file, or
2273 	 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2274 	 * on the file.  If the time pointer is null, then write
2275 	 * permission on the file is also sufficient.
2276 	 *
2277 	 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2278 	 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2279 	 * will be allowed to set the times [..] to the current
2280 	 * server time.
2281 	 */
2282 	error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2283 	if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2284 		error = VOP_ACCESS(vp, VWRITE, cred, td);
2285 	return (error);
2286 }
2287 
2288 int
2289 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2290 {
2291 	struct vnode *vp;
2292 	int error;
2293 
2294 	if (fp->f_type == DTYPE_FIFO)
2295 		kif->kf_type = KF_TYPE_FIFO;
2296 	else
2297 		kif->kf_type = KF_TYPE_VNODE;
2298 	vp = fp->f_vnode;
2299 	vref(vp);
2300 	FILEDESC_SUNLOCK(fdp);
2301 	error = vn_fill_kinfo_vnode(vp, kif);
2302 	vrele(vp);
2303 	FILEDESC_SLOCK(fdp);
2304 	return (error);
2305 }
2306 
2307 static inline void
2308 vn_fill_junk(struct kinfo_file *kif)
2309 {
2310 	size_t len, olen;
2311 
2312 	/*
2313 	 * Simulate vn_fullpath returning changing values for a given
2314 	 * vp during e.g. coredump.
2315 	 */
2316 	len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2317 	olen = strlen(kif->kf_path);
2318 	if (len < olen)
2319 		strcpy(&kif->kf_path[len - 1], "$");
2320 	else
2321 		for (; olen < len; olen++)
2322 			strcpy(&kif->kf_path[olen], "A");
2323 }
2324 
2325 int
2326 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2327 {
2328 	struct vattr va;
2329 	char *fullpath, *freepath;
2330 	int error;
2331 
2332 	kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type);
2333 	freepath = NULL;
2334 	fullpath = "-";
2335 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2336 	if (error == 0) {
2337 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2338 	}
2339 	if (freepath != NULL)
2340 		free(freepath, M_TEMP);
2341 
2342 	KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2343 		vn_fill_junk(kif);
2344 	);
2345 
2346 	/*
2347 	 * Retrieve vnode attributes.
2348 	 */
2349 	va.va_fsid = VNOVAL;
2350 	va.va_rdev = NODEV;
2351 	vn_lock(vp, LK_SHARED | LK_RETRY);
2352 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2353 	VOP_UNLOCK(vp, 0);
2354 	if (error != 0)
2355 		return (error);
2356 	if (va.va_fsid != VNOVAL)
2357 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2358 	else
2359 		kif->kf_un.kf_file.kf_file_fsid =
2360 		    vp->v_mount->mnt_stat.f_fsid.val[0];
2361 	kif->kf_un.kf_file.kf_file_fsid_freebsd11 =
2362 	    kif->kf_un.kf_file.kf_file_fsid; /* truncate */
2363 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2364 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2365 	kif->kf_un.kf_file.kf_file_size = va.va_size;
2366 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2367 	kif->kf_un.kf_file.kf_file_rdev_freebsd11 =
2368 	    kif->kf_un.kf_file.kf_file_rdev; /* truncate */
2369 	return (0);
2370 }
2371 
2372 int
2373 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2374     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2375     struct thread *td)
2376 {
2377 #ifdef HWPMC_HOOKS
2378 	struct pmckern_map_in pkm;
2379 #endif
2380 	struct mount *mp;
2381 	struct vnode *vp;
2382 	vm_object_t object;
2383 	vm_prot_t maxprot;
2384 	boolean_t writecounted;
2385 	int error;
2386 
2387 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2388     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2389 	/*
2390 	 * POSIX shared-memory objects are defined to have
2391 	 * kernel persistence, and are not defined to support
2392 	 * read(2)/write(2) -- or even open(2).  Thus, we can
2393 	 * use MAP_ASYNC to trade on-disk coherence for speed.
2394 	 * The shm_open(3) library routine turns on the FPOSIXSHM
2395 	 * flag to request this behavior.
2396 	 */
2397 	if ((fp->f_flag & FPOSIXSHM) != 0)
2398 		flags |= MAP_NOSYNC;
2399 #endif
2400 	vp = fp->f_vnode;
2401 
2402 	/*
2403 	 * Ensure that file and memory protections are
2404 	 * compatible.  Note that we only worry about
2405 	 * writability if mapping is shared; in this case,
2406 	 * current and max prot are dictated by the open file.
2407 	 * XXX use the vnode instead?  Problem is: what
2408 	 * credentials do we use for determination? What if
2409 	 * proc does a setuid?
2410 	 */
2411 	mp = vp->v_mount;
2412 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
2413 		maxprot = VM_PROT_NONE;
2414 		if ((prot & VM_PROT_EXECUTE) != 0)
2415 			return (EACCES);
2416 	} else
2417 		maxprot = VM_PROT_EXECUTE;
2418 	if ((fp->f_flag & FREAD) != 0)
2419 		maxprot |= VM_PROT_READ;
2420 	else if ((prot & VM_PROT_READ) != 0)
2421 		return (EACCES);
2422 
2423 	/*
2424 	 * If we are sharing potential changes via MAP_SHARED and we
2425 	 * are trying to get write permission although we opened it
2426 	 * without asking for it, bail out.
2427 	 */
2428 	if ((flags & MAP_SHARED) != 0) {
2429 		if ((fp->f_flag & FWRITE) != 0)
2430 			maxprot |= VM_PROT_WRITE;
2431 		else if ((prot & VM_PROT_WRITE) != 0)
2432 			return (EACCES);
2433 	} else {
2434 		maxprot |= VM_PROT_WRITE;
2435 		cap_maxprot |= VM_PROT_WRITE;
2436 	}
2437 	maxprot &= cap_maxprot;
2438 
2439 	/*
2440 	 * For regular files and shared memory, POSIX requires that
2441 	 * the value of foff be a legitimate offset within the data
2442 	 * object.  In particular, negative offsets are invalid.
2443 	 * Blocking negative offsets and overflows here avoids
2444 	 * possible wraparound or user-level access into reserved
2445 	 * ranges of the data object later.  In contrast, POSIX does
2446 	 * not dictate how offsets are used by device drivers, so in
2447 	 * the case of a device mapping a negative offset is passed
2448 	 * on.
2449 	 */
2450 	if (
2451 #ifdef _LP64
2452 	    size > OFF_MAX ||
2453 #endif
2454 	    foff < 0 || foff > OFF_MAX - size)
2455 		return (EINVAL);
2456 
2457 	writecounted = FALSE;
2458 	error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2459 	    &foff, &object, &writecounted);
2460 	if (error != 0)
2461 		return (error);
2462 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2463 	    foff, writecounted, td);
2464 	if (error != 0) {
2465 		/*
2466 		 * If this mapping was accounted for in the vnode's
2467 		 * writecount, then undo that now.
2468 		 */
2469 		if (writecounted)
2470 			vnode_pager_release_writecount(object, 0, size);
2471 		vm_object_deallocate(object);
2472 	}
2473 #ifdef HWPMC_HOOKS
2474 	/* Inform hwpmc(4) if an executable is being mapped. */
2475 	if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) {
2476 		if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) {
2477 			pkm.pm_file = vp;
2478 			pkm.pm_address = (uintptr_t) *addr;
2479 			PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_MMAP, (void *) &pkm);
2480 		}
2481 	}
2482 #endif
2483 	return (error);
2484 }
2485 
2486 void
2487 vn_fsid(struct vnode *vp, struct vattr *va)
2488 {
2489 	fsid_t *f;
2490 
2491 	f = &vp->v_mount->mnt_stat.f_fsid;
2492 	va->va_fsid = (uint32_t)f->val[1];
2493 	va->va_fsid <<= sizeof(f->val[1]) * NBBY;
2494 	va->va_fsid += (uint32_t)f->val[0];
2495 }
2496 
2497 int
2498 vn_fsync_buf(struct vnode *vp, int waitfor)
2499 {
2500 	struct buf *bp, *nbp;
2501 	struct bufobj *bo;
2502 	struct mount *mp;
2503 	int error, maxretry;
2504 
2505 	error = 0;
2506 	maxretry = 10000;     /* large, arbitrarily chosen */
2507 	mp = NULL;
2508 	if (vp->v_type == VCHR) {
2509 		VI_LOCK(vp);
2510 		mp = vp->v_rdev->si_mountpt;
2511 		VI_UNLOCK(vp);
2512 	}
2513 	bo = &vp->v_bufobj;
2514 	BO_LOCK(bo);
2515 loop1:
2516 	/*
2517 	 * MARK/SCAN initialization to avoid infinite loops.
2518 	 */
2519         TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
2520 		bp->b_vflags &= ~BV_SCANNED;
2521 		bp->b_error = 0;
2522 	}
2523 
2524 	/*
2525 	 * Flush all dirty buffers associated with a vnode.
2526 	 */
2527 loop2:
2528 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2529 		if ((bp->b_vflags & BV_SCANNED) != 0)
2530 			continue;
2531 		bp->b_vflags |= BV_SCANNED;
2532 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2533 			if (waitfor != MNT_WAIT)
2534 				continue;
2535 			if (BUF_LOCK(bp,
2536 			    LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL,
2537 			    BO_LOCKPTR(bo)) != 0) {
2538 				BO_LOCK(bo);
2539 				goto loop1;
2540 			}
2541 			BO_LOCK(bo);
2542 		}
2543 		BO_UNLOCK(bo);
2544 		KASSERT(bp->b_bufobj == bo,
2545 		    ("bp %p wrong b_bufobj %p should be %p",
2546 		    bp, bp->b_bufobj, bo));
2547 		if ((bp->b_flags & B_DELWRI) == 0)
2548 			panic("fsync: not dirty");
2549 		if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) {
2550 			vfs_bio_awrite(bp);
2551 		} else {
2552 			bremfree(bp);
2553 			bawrite(bp);
2554 		}
2555 		if (maxretry < 1000)
2556 			pause("dirty", hz < 1000 ? 1 : hz / 1000);
2557 		BO_LOCK(bo);
2558 		goto loop2;
2559 	}
2560 
2561 	/*
2562 	 * If synchronous the caller expects us to completely resolve all
2563 	 * dirty buffers in the system.  Wait for in-progress I/O to
2564 	 * complete (which could include background bitmap writes), then
2565 	 * retry if dirty blocks still exist.
2566 	 */
2567 	if (waitfor == MNT_WAIT) {
2568 		bufobj_wwait(bo, 0, 0);
2569 		if (bo->bo_dirty.bv_cnt > 0) {
2570 			/*
2571 			 * If we are unable to write any of these buffers
2572 			 * then we fail now rather than trying endlessly
2573 			 * to write them out.
2574 			 */
2575 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
2576 				if ((error = bp->b_error) != 0)
2577 					break;
2578 			if ((mp != NULL && mp->mnt_secondary_writes > 0) ||
2579 			    (error == 0 && --maxretry >= 0))
2580 				goto loop1;
2581 			if (error == 0)
2582 				error = EAGAIN;
2583 		}
2584 	}
2585 	BO_UNLOCK(bo);
2586 	if (error != 0)
2587 		vn_printf(vp, "fsync: giving up on dirty (error = %d) ", error);
2588 
2589 	return (error);
2590 }
2591