xref: /freebsd/sys/kern/vfs_vnops.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11  * Copyright (c) 2013, 2014 The FreeBSD Foundation
12  *
13  * Portions of this software were developed by Konstantin Belousov
14  * under sponsorship from the FreeBSD Foundation.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)vfs_vnops.c	8.2 (Berkeley) 1/21/94
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/disk.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/kdb.h>
52 #include <sys/stat.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/limits.h>
56 #include <sys/lock.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/vnode.h>
61 #include <sys/bio.h>
62 #include <sys/buf.h>
63 #include <sys/filio.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
66 #include <sys/sx.h>
67 #include <sys/sysctl.h>
68 #include <sys/ttycom.h>
69 #include <sys/conf.h>
70 #include <sys/syslog.h>
71 #include <sys/unistd.h>
72 #include <sys/user.h>
73 
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.h>
76 
77 #include <vm/vm.h>
78 #include <vm/vm_extern.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_page.h>
83 
84 static fo_rdwr_t	vn_read;
85 static fo_rdwr_t	vn_write;
86 static fo_rdwr_t	vn_io_fault;
87 static fo_truncate_t	vn_truncate;
88 static fo_ioctl_t	vn_ioctl;
89 static fo_poll_t	vn_poll;
90 static fo_kqfilter_t	vn_kqfilter;
91 static fo_stat_t	vn_statfile;
92 static fo_close_t	vn_closefile;
93 
94 struct 	fileops vnops = {
95 	.fo_read = vn_io_fault,
96 	.fo_write = vn_io_fault,
97 	.fo_truncate = vn_truncate,
98 	.fo_ioctl = vn_ioctl,
99 	.fo_poll = vn_poll,
100 	.fo_kqfilter = vn_kqfilter,
101 	.fo_stat = vn_statfile,
102 	.fo_close = vn_closefile,
103 	.fo_chmod = vn_chmod,
104 	.fo_chown = vn_chown,
105 	.fo_sendfile = vn_sendfile,
106 	.fo_seek = vn_seek,
107 	.fo_fill_kinfo = vn_fill_kinfo,
108 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
109 };
110 
111 static const int io_hold_cnt = 16;
112 static int vn_io_fault_enable = 1;
113 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
114     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
115 static u_long vn_io_faults_cnt;
116 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
117     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
118 
119 /*
120  * Returns true if vn_io_fault mode of handling the i/o request should
121  * be used.
122  */
123 static bool
124 do_vn_io_fault(struct vnode *vp, struct uio *uio)
125 {
126 	struct mount *mp;
127 
128 	return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
129 	    (mp = vp->v_mount) != NULL &&
130 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
131 }
132 
133 /*
134  * Structure used to pass arguments to vn_io_fault1(), to do either
135  * file- or vnode-based I/O calls.
136  */
137 struct vn_io_fault_args {
138 	enum {
139 		VN_IO_FAULT_FOP,
140 		VN_IO_FAULT_VOP
141 	} kind;
142 	struct ucred *cred;
143 	int flags;
144 	union {
145 		struct fop_args_tag {
146 			struct file *fp;
147 			fo_rdwr_t *doio;
148 		} fop_args;
149 		struct vop_args_tag {
150 			struct vnode *vp;
151 		} vop_args;
152 	} args;
153 };
154 
155 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
156     struct vn_io_fault_args *args, struct thread *td);
157 
158 int
159 vn_open(ndp, flagp, cmode, fp)
160 	struct nameidata *ndp;
161 	int *flagp, cmode;
162 	struct file *fp;
163 {
164 	struct thread *td = ndp->ni_cnd.cn_thread;
165 
166 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
167 }
168 
169 /*
170  * Common code for vnode open operations via a name lookup.
171  * Lookup the vnode and invoke VOP_CREATE if needed.
172  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
173  *
174  * Note that this does NOT free nameidata for the successful case,
175  * due to the NDINIT being done elsewhere.
176  */
177 int
178 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
179     struct ucred *cred, struct file *fp)
180 {
181 	struct vnode *vp;
182 	struct mount *mp;
183 	struct thread *td = ndp->ni_cnd.cn_thread;
184 	struct vattr vat;
185 	struct vattr *vap = &vat;
186 	int fmode, error;
187 
188 restart:
189 	fmode = *flagp;
190 	if (fmode & O_CREAT) {
191 		ndp->ni_cnd.cn_nameiop = CREATE;
192 		/*
193 		 * Set NOCACHE to avoid flushing the cache when
194 		 * rolling in many files at once.
195 		*/
196 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
197 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
198 			ndp->ni_cnd.cn_flags |= FOLLOW;
199 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
200 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
201 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
202 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
203 		bwillwrite();
204 		if ((error = namei(ndp)) != 0)
205 			return (error);
206 		if (ndp->ni_vp == NULL) {
207 			VATTR_NULL(vap);
208 			vap->va_type = VREG;
209 			vap->va_mode = cmode;
210 			if (fmode & O_EXCL)
211 				vap->va_vaflags |= VA_EXCLUSIVE;
212 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
213 				NDFREE(ndp, NDF_ONLY_PNBUF);
214 				vput(ndp->ni_dvp);
215 				if ((error = vn_start_write(NULL, &mp,
216 				    V_XSLEEP | PCATCH)) != 0)
217 					return (error);
218 				goto restart;
219 			}
220 			if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
221 				ndp->ni_cnd.cn_flags |= MAKEENTRY;
222 #ifdef MAC
223 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
224 			    &ndp->ni_cnd, vap);
225 			if (error == 0)
226 #endif
227 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
228 						   &ndp->ni_cnd, vap);
229 			vput(ndp->ni_dvp);
230 			vn_finished_write(mp);
231 			if (error) {
232 				NDFREE(ndp, NDF_ONLY_PNBUF);
233 				return (error);
234 			}
235 			fmode &= ~O_TRUNC;
236 			vp = ndp->ni_vp;
237 		} else {
238 			if (ndp->ni_dvp == ndp->ni_vp)
239 				vrele(ndp->ni_dvp);
240 			else
241 				vput(ndp->ni_dvp);
242 			ndp->ni_dvp = NULL;
243 			vp = ndp->ni_vp;
244 			if (fmode & O_EXCL) {
245 				error = EEXIST;
246 				goto bad;
247 			}
248 			fmode &= ~O_CREAT;
249 		}
250 	} else {
251 		ndp->ni_cnd.cn_nameiop = LOOKUP;
252 		ndp->ni_cnd.cn_flags = ISOPEN |
253 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
254 		if (!(fmode & FWRITE))
255 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
256 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
257 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
258 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
259 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
260 		if ((error = namei(ndp)) != 0)
261 			return (error);
262 		vp = ndp->ni_vp;
263 	}
264 	error = vn_open_vnode(vp, fmode, cred, td, fp);
265 	if (error)
266 		goto bad;
267 	*flagp = fmode;
268 	return (0);
269 bad:
270 	NDFREE(ndp, NDF_ONLY_PNBUF);
271 	vput(vp);
272 	*flagp = fmode;
273 	ndp->ni_vp = NULL;
274 	return (error);
275 }
276 
277 /*
278  * Common code for vnode open operations once a vnode is located.
279  * Check permissions, and call the VOP_OPEN routine.
280  */
281 int
282 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
283     struct thread *td, struct file *fp)
284 {
285 	struct mount *mp;
286 	accmode_t accmode;
287 	struct flock lf;
288 	int error, have_flock, lock_flags, type;
289 
290 	if (vp->v_type == VLNK)
291 		return (EMLINK);
292 	if (vp->v_type == VSOCK)
293 		return (EOPNOTSUPP);
294 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
295 		return (ENOTDIR);
296 	accmode = 0;
297 	if (fmode & (FWRITE | O_TRUNC)) {
298 		if (vp->v_type == VDIR)
299 			return (EISDIR);
300 		accmode |= VWRITE;
301 	}
302 	if (fmode & FREAD)
303 		accmode |= VREAD;
304 	if (fmode & FEXEC)
305 		accmode |= VEXEC;
306 	if ((fmode & O_APPEND) && (fmode & FWRITE))
307 		accmode |= VAPPEND;
308 #ifdef MAC
309 	error = mac_vnode_check_open(cred, vp, accmode);
310 	if (error)
311 		return (error);
312 #endif
313 	if ((fmode & O_CREAT) == 0) {
314 		if (accmode & VWRITE) {
315 			error = vn_writechk(vp);
316 			if (error)
317 				return (error);
318 		}
319 		if (accmode) {
320 		        error = VOP_ACCESS(vp, accmode, cred, td);
321 			if (error)
322 				return (error);
323 		}
324 	}
325 	if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
326 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
327 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
328 		return (error);
329 
330 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
331 		KASSERT(fp != NULL, ("open with flock requires fp"));
332 		lock_flags = VOP_ISLOCKED(vp);
333 		VOP_UNLOCK(vp, 0);
334 		lf.l_whence = SEEK_SET;
335 		lf.l_start = 0;
336 		lf.l_len = 0;
337 		if (fmode & O_EXLOCK)
338 			lf.l_type = F_WRLCK;
339 		else
340 			lf.l_type = F_RDLCK;
341 		type = F_FLOCK;
342 		if ((fmode & FNONBLOCK) == 0)
343 			type |= F_WAIT;
344 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
345 		have_flock = (error == 0);
346 		vn_lock(vp, lock_flags | LK_RETRY);
347 		if (error == 0 && vp->v_iflag & VI_DOOMED)
348 			error = ENOENT;
349 		/*
350 		 * Another thread might have used this vnode as an
351 		 * executable while the vnode lock was dropped.
352 		 * Ensure the vnode is still able to be opened for
353 		 * writing after the lock has been obtained.
354 		 */
355 		if (error == 0 && accmode & VWRITE)
356 			error = vn_writechk(vp);
357 		if (error) {
358 			VOP_UNLOCK(vp, 0);
359 			if (have_flock) {
360 				lf.l_whence = SEEK_SET;
361 				lf.l_start = 0;
362 				lf.l_len = 0;
363 				lf.l_type = F_UNLCK;
364 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
365 				    F_FLOCK);
366 			}
367 			vn_start_write(vp, &mp, V_WAIT);
368 			vn_lock(vp, lock_flags | LK_RETRY);
369 			(void)VOP_CLOSE(vp, fmode, cred, td);
370 			vn_finished_write(mp);
371 			/* Prevent second close from fdrop()->vn_close(). */
372 			if (fp != NULL)
373 				fp->f_ops= &badfileops;
374 			return (error);
375 		}
376 		fp->f_flag |= FHASLOCK;
377 	}
378 	if (fmode & FWRITE) {
379 		VOP_ADD_WRITECOUNT(vp, 1);
380 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
381 		    __func__, vp, vp->v_writecount);
382 	}
383 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
384 	return (0);
385 }
386 
387 /*
388  * Check for write permissions on the specified vnode.
389  * Prototype text segments cannot be written.
390  */
391 int
392 vn_writechk(vp)
393 	register struct vnode *vp;
394 {
395 
396 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
397 	/*
398 	 * If there's shared text associated with
399 	 * the vnode, try to free it up once.  If
400 	 * we fail, we can't allow writing.
401 	 */
402 	if (VOP_IS_TEXT(vp))
403 		return (ETXTBSY);
404 
405 	return (0);
406 }
407 
408 /*
409  * Vnode close call
410  */
411 int
412 vn_close(vp, flags, file_cred, td)
413 	register struct vnode *vp;
414 	int flags;
415 	struct ucred *file_cred;
416 	struct thread *td;
417 {
418 	struct mount *mp;
419 	int error, lock_flags;
420 
421 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
422 	    MNT_EXTENDED_SHARED(vp->v_mount))
423 		lock_flags = LK_SHARED;
424 	else
425 		lock_flags = LK_EXCLUSIVE;
426 
427 	vn_start_write(vp, &mp, V_WAIT);
428 	vn_lock(vp, lock_flags | LK_RETRY);
429 	if (flags & FWRITE) {
430 		VNASSERT(vp->v_writecount > 0, vp,
431 		    ("vn_close: negative writecount"));
432 		VOP_ADD_WRITECOUNT(vp, -1);
433 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
434 		    __func__, vp, vp->v_writecount);
435 	}
436 	error = VOP_CLOSE(vp, flags, file_cred, td);
437 	vput(vp);
438 	vn_finished_write(mp);
439 	return (error);
440 }
441 
442 /*
443  * Heuristic to detect sequential operation.
444  */
445 static int
446 sequential_heuristic(struct uio *uio, struct file *fp)
447 {
448 
449 	ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
450 	if (fp->f_flag & FRDAHEAD)
451 		return (fp->f_seqcount << IO_SEQSHIFT);
452 
453 	/*
454 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
455 	 * that the first I/O is normally considered to be slightly
456 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
457 	 * unless previous seeks have reduced f_seqcount to 0, in which
458 	 * case offset 0 is not special.
459 	 */
460 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
461 	    uio->uio_offset == fp->f_nextoff) {
462 		/*
463 		 * f_seqcount is in units of fixed-size blocks so that it
464 		 * depends mainly on the amount of sequential I/O and not
465 		 * much on the number of sequential I/O's.  The fixed size
466 		 * of 16384 is hard-coded here since it is (not quite) just
467 		 * a magic size that works well here.  This size is more
468 		 * closely related to the best I/O size for real disks than
469 		 * to any block size used by software.
470 		 */
471 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
472 		if (fp->f_seqcount > IO_SEQMAX)
473 			fp->f_seqcount = IO_SEQMAX;
474 		return (fp->f_seqcount << IO_SEQSHIFT);
475 	}
476 
477 	/* Not sequential.  Quickly draw-down sequentiality. */
478 	if (fp->f_seqcount > 1)
479 		fp->f_seqcount = 1;
480 	else
481 		fp->f_seqcount = 0;
482 	return (0);
483 }
484 
485 /*
486  * Package up an I/O request on a vnode into a uio and do it.
487  */
488 int
489 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
490     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
491     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
492 {
493 	struct uio auio;
494 	struct iovec aiov;
495 	struct mount *mp;
496 	struct ucred *cred;
497 	void *rl_cookie;
498 	struct vn_io_fault_args args;
499 	int error, lock_flags;
500 
501 	auio.uio_iov = &aiov;
502 	auio.uio_iovcnt = 1;
503 	aiov.iov_base = base;
504 	aiov.iov_len = len;
505 	auio.uio_resid = len;
506 	auio.uio_offset = offset;
507 	auio.uio_segflg = segflg;
508 	auio.uio_rw = rw;
509 	auio.uio_td = td;
510 	error = 0;
511 
512 	if ((ioflg & IO_NODELOCKED) == 0) {
513 		if ((ioflg & IO_RANGELOCKED) == 0) {
514 			if (rw == UIO_READ) {
515 				rl_cookie = vn_rangelock_rlock(vp, offset,
516 				    offset + len);
517 			} else {
518 				rl_cookie = vn_rangelock_wlock(vp, offset,
519 				    offset + len);
520 			}
521 		} else
522 			rl_cookie = NULL;
523 		mp = NULL;
524 		if (rw == UIO_WRITE) {
525 			if (vp->v_type != VCHR &&
526 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
527 			    != 0)
528 				goto out;
529 			if (MNT_SHARED_WRITES(mp) ||
530 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
531 				lock_flags = LK_SHARED;
532 			else
533 				lock_flags = LK_EXCLUSIVE;
534 		} else
535 			lock_flags = LK_SHARED;
536 		vn_lock(vp, lock_flags | LK_RETRY);
537 	} else
538 		rl_cookie = NULL;
539 
540 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
541 #ifdef MAC
542 	if ((ioflg & IO_NOMACCHECK) == 0) {
543 		if (rw == UIO_READ)
544 			error = mac_vnode_check_read(active_cred, file_cred,
545 			    vp);
546 		else
547 			error = mac_vnode_check_write(active_cred, file_cred,
548 			    vp);
549 	}
550 #endif
551 	if (error == 0) {
552 		if (file_cred != NULL)
553 			cred = file_cred;
554 		else
555 			cred = active_cred;
556 		if (do_vn_io_fault(vp, &auio)) {
557 			args.kind = VN_IO_FAULT_VOP;
558 			args.cred = cred;
559 			args.flags = ioflg;
560 			args.args.vop_args.vp = vp;
561 			error = vn_io_fault1(vp, &auio, &args, td);
562 		} else if (rw == UIO_READ) {
563 			error = VOP_READ(vp, &auio, ioflg, cred);
564 		} else /* if (rw == UIO_WRITE) */ {
565 			error = VOP_WRITE(vp, &auio, ioflg, cred);
566 		}
567 	}
568 	if (aresid)
569 		*aresid = auio.uio_resid;
570 	else
571 		if (auio.uio_resid && error == 0)
572 			error = EIO;
573 	if ((ioflg & IO_NODELOCKED) == 0) {
574 		VOP_UNLOCK(vp, 0);
575 		if (mp != NULL)
576 			vn_finished_write(mp);
577 	}
578  out:
579 	if (rl_cookie != NULL)
580 		vn_rangelock_unlock(vp, rl_cookie);
581 	return (error);
582 }
583 
584 /*
585  * Package up an I/O request on a vnode into a uio and do it.  The I/O
586  * request is split up into smaller chunks and we try to avoid saturating
587  * the buffer cache while potentially holding a vnode locked, so we
588  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
589  * to give other processes a chance to lock the vnode (either other processes
590  * core'ing the same binary, or unrelated processes scanning the directory).
591  */
592 int
593 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
594     file_cred, aresid, td)
595 	enum uio_rw rw;
596 	struct vnode *vp;
597 	void *base;
598 	size_t len;
599 	off_t offset;
600 	enum uio_seg segflg;
601 	int ioflg;
602 	struct ucred *active_cred;
603 	struct ucred *file_cred;
604 	size_t *aresid;
605 	struct thread *td;
606 {
607 	int error = 0;
608 	ssize_t iaresid;
609 
610 	do {
611 		int chunk;
612 
613 		/*
614 		 * Force `offset' to a multiple of MAXBSIZE except possibly
615 		 * for the first chunk, so that filesystems only need to
616 		 * write full blocks except possibly for the first and last
617 		 * chunks.
618 		 */
619 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
620 
621 		if (chunk > len)
622 			chunk = len;
623 		if (rw != UIO_READ && vp->v_type == VREG)
624 			bwillwrite();
625 		iaresid = 0;
626 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
627 		    ioflg, active_cred, file_cred, &iaresid, td);
628 		len -= chunk;	/* aresid calc already includes length */
629 		if (error)
630 			break;
631 		offset += chunk;
632 		base = (char *)base + chunk;
633 		kern_yield(PRI_USER);
634 	} while (len);
635 	if (aresid)
636 		*aresid = len + iaresid;
637 	return (error);
638 }
639 
640 off_t
641 foffset_lock(struct file *fp, int flags)
642 {
643 	struct mtx *mtxp;
644 	off_t res;
645 
646 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
647 
648 #if OFF_MAX <= LONG_MAX
649 	/*
650 	 * Caller only wants the current f_offset value.  Assume that
651 	 * the long and shorter integer types reads are atomic.
652 	 */
653 	if ((flags & FOF_NOLOCK) != 0)
654 		return (fp->f_offset);
655 #endif
656 
657 	/*
658 	 * According to McKusick the vn lock was protecting f_offset here.
659 	 * It is now protected by the FOFFSET_LOCKED flag.
660 	 */
661 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
662 	mtx_lock(mtxp);
663 	if ((flags & FOF_NOLOCK) == 0) {
664 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
665 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
666 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
667 			    "vofflock", 0);
668 		}
669 		fp->f_vnread_flags |= FOFFSET_LOCKED;
670 	}
671 	res = fp->f_offset;
672 	mtx_unlock(mtxp);
673 	return (res);
674 }
675 
676 void
677 foffset_unlock(struct file *fp, off_t val, int flags)
678 {
679 	struct mtx *mtxp;
680 
681 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
682 
683 #if OFF_MAX <= LONG_MAX
684 	if ((flags & FOF_NOLOCK) != 0) {
685 		if ((flags & FOF_NOUPDATE) == 0)
686 			fp->f_offset = val;
687 		if ((flags & FOF_NEXTOFF) != 0)
688 			fp->f_nextoff = val;
689 		return;
690 	}
691 #endif
692 
693 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
694 	mtx_lock(mtxp);
695 	if ((flags & FOF_NOUPDATE) == 0)
696 		fp->f_offset = val;
697 	if ((flags & FOF_NEXTOFF) != 0)
698 		fp->f_nextoff = val;
699 	if ((flags & FOF_NOLOCK) == 0) {
700 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
701 		    ("Lost FOFFSET_LOCKED"));
702 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
703 			wakeup(&fp->f_vnread_flags);
704 		fp->f_vnread_flags = 0;
705 	}
706 	mtx_unlock(mtxp);
707 }
708 
709 void
710 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
711 {
712 
713 	if ((flags & FOF_OFFSET) == 0)
714 		uio->uio_offset = foffset_lock(fp, flags);
715 }
716 
717 void
718 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
719 {
720 
721 	if ((flags & FOF_OFFSET) == 0)
722 		foffset_unlock(fp, uio->uio_offset, flags);
723 }
724 
725 static int
726 get_advice(struct file *fp, struct uio *uio)
727 {
728 	struct mtx *mtxp;
729 	int ret;
730 
731 	ret = POSIX_FADV_NORMAL;
732 	if (fp->f_advice == NULL)
733 		return (ret);
734 
735 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
736 	mtx_lock(mtxp);
737 	if (uio->uio_offset >= fp->f_advice->fa_start &&
738 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
739 		ret = fp->f_advice->fa_advice;
740 	mtx_unlock(mtxp);
741 	return (ret);
742 }
743 
744 /*
745  * File table vnode read routine.
746  */
747 static int
748 vn_read(fp, uio, active_cred, flags, td)
749 	struct file *fp;
750 	struct uio *uio;
751 	struct ucred *active_cred;
752 	int flags;
753 	struct thread *td;
754 {
755 	struct vnode *vp;
756 	struct mtx *mtxp;
757 	int error, ioflag;
758 	int advice;
759 	off_t offset, start, end;
760 
761 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
762 	    uio->uio_td, td));
763 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
764 	vp = fp->f_vnode;
765 	ioflag = 0;
766 	if (fp->f_flag & FNONBLOCK)
767 		ioflag |= IO_NDELAY;
768 	if (fp->f_flag & O_DIRECT)
769 		ioflag |= IO_DIRECT;
770 	advice = get_advice(fp, uio);
771 	vn_lock(vp, LK_SHARED | LK_RETRY);
772 
773 	switch (advice) {
774 	case POSIX_FADV_NORMAL:
775 	case POSIX_FADV_SEQUENTIAL:
776 	case POSIX_FADV_NOREUSE:
777 		ioflag |= sequential_heuristic(uio, fp);
778 		break;
779 	case POSIX_FADV_RANDOM:
780 		/* Disable read-ahead for random I/O. */
781 		break;
782 	}
783 	offset = uio->uio_offset;
784 
785 #ifdef MAC
786 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
787 	if (error == 0)
788 #endif
789 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
790 	fp->f_nextoff = uio->uio_offset;
791 	VOP_UNLOCK(vp, 0);
792 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
793 	    offset != uio->uio_offset) {
794 		/*
795 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
796 		 * buffers for the backing file after a
797 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
798 		 * case of using POSIX_FADV_NOREUSE with sequential
799 		 * access, track the previous implicit DONTNEED
800 		 * request and grow this request to include the
801 		 * current read(2) in addition to the previous
802 		 * DONTNEED.  With purely sequential access this will
803 		 * cause the DONTNEED requests to continously grow to
804 		 * cover all of the previously read regions of the
805 		 * file.  This allows filesystem blocks that are
806 		 * accessed by multiple calls to read(2) to be flushed
807 		 * once the last read(2) finishes.
808 		 */
809 		start = offset;
810 		end = uio->uio_offset - 1;
811 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
812 		mtx_lock(mtxp);
813 		if (fp->f_advice != NULL &&
814 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
815 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
816 				start = fp->f_advice->fa_prevstart;
817 			else if (fp->f_advice->fa_prevstart != 0 &&
818 			    fp->f_advice->fa_prevstart == end + 1)
819 				end = fp->f_advice->fa_prevend;
820 			fp->f_advice->fa_prevstart = start;
821 			fp->f_advice->fa_prevend = end;
822 		}
823 		mtx_unlock(mtxp);
824 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
825 	}
826 	return (error);
827 }
828 
829 /*
830  * File table vnode write routine.
831  */
832 static int
833 vn_write(fp, uio, active_cred, flags, td)
834 	struct file *fp;
835 	struct uio *uio;
836 	struct ucred *active_cred;
837 	int flags;
838 	struct thread *td;
839 {
840 	struct vnode *vp;
841 	struct mount *mp;
842 	struct mtx *mtxp;
843 	int error, ioflag, lock_flags;
844 	int advice;
845 	off_t offset, start, end;
846 
847 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
848 	    uio->uio_td, td));
849 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
850 	vp = fp->f_vnode;
851 	if (vp->v_type == VREG)
852 		bwillwrite();
853 	ioflag = IO_UNIT;
854 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
855 		ioflag |= IO_APPEND;
856 	if (fp->f_flag & FNONBLOCK)
857 		ioflag |= IO_NDELAY;
858 	if (fp->f_flag & O_DIRECT)
859 		ioflag |= IO_DIRECT;
860 	if ((fp->f_flag & O_FSYNC) ||
861 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
862 		ioflag |= IO_SYNC;
863 	mp = NULL;
864 	if (vp->v_type != VCHR &&
865 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
866 		goto unlock;
867 
868 	advice = get_advice(fp, uio);
869 
870 	if (MNT_SHARED_WRITES(mp) ||
871 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
872 		lock_flags = LK_SHARED;
873 	} else {
874 		lock_flags = LK_EXCLUSIVE;
875 	}
876 
877 	vn_lock(vp, lock_flags | LK_RETRY);
878 	switch (advice) {
879 	case POSIX_FADV_NORMAL:
880 	case POSIX_FADV_SEQUENTIAL:
881 	case POSIX_FADV_NOREUSE:
882 		ioflag |= sequential_heuristic(uio, fp);
883 		break;
884 	case POSIX_FADV_RANDOM:
885 		/* XXX: Is this correct? */
886 		break;
887 	}
888 	offset = uio->uio_offset;
889 
890 #ifdef MAC
891 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
892 	if (error == 0)
893 #endif
894 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
895 	fp->f_nextoff = uio->uio_offset;
896 	VOP_UNLOCK(vp, 0);
897 	if (vp->v_type != VCHR)
898 		vn_finished_write(mp);
899 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
900 	    offset != uio->uio_offset) {
901 		/*
902 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
903 		 * buffers for the backing file after a
904 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
905 		 * common case of using POSIX_FADV_NOREUSE with
906 		 * sequential access, track the previous implicit
907 		 * DONTNEED request and grow this request to include
908 		 * the current write(2) in addition to the previous
909 		 * DONTNEED.  With purely sequential access this will
910 		 * cause the DONTNEED requests to continously grow to
911 		 * cover all of the previously written regions of the
912 		 * file.
913 		 *
914 		 * Note that the blocks just written are almost
915 		 * certainly still dirty, so this only works when
916 		 * VOP_ADVISE() calls from subsequent writes push out
917 		 * the data written by this write(2) once the backing
918 		 * buffers are clean.  However, as compared to forcing
919 		 * IO_DIRECT, this gives much saner behavior.  Write
920 		 * clustering is still allowed, and clean pages are
921 		 * merely moved to the cache page queue rather than
922 		 * outright thrown away.  This means a subsequent
923 		 * read(2) can still avoid hitting the disk if the
924 		 * pages have not been reclaimed.
925 		 *
926 		 * This does make POSIX_FADV_NOREUSE largely useless
927 		 * with non-sequential access.  However, sequential
928 		 * access is the more common use case and the flag is
929 		 * merely advisory.
930 		 */
931 		start = offset;
932 		end = uio->uio_offset - 1;
933 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
934 		mtx_lock(mtxp);
935 		if (fp->f_advice != NULL &&
936 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
937 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
938 				start = fp->f_advice->fa_prevstart;
939 			else if (fp->f_advice->fa_prevstart != 0 &&
940 			    fp->f_advice->fa_prevstart == end + 1)
941 				end = fp->f_advice->fa_prevend;
942 			fp->f_advice->fa_prevstart = start;
943 			fp->f_advice->fa_prevend = end;
944 		}
945 		mtx_unlock(mtxp);
946 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
947 	}
948 
949 unlock:
950 	return (error);
951 }
952 
953 /*
954  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
955  * prevent the following deadlock:
956  *
957  * Assume that the thread A reads from the vnode vp1 into userspace
958  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
959  * currently not resident, then system ends up with the call chain
960  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
961  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
962  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
963  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
964  * backed by the pages of vnode vp1, and some page in buf2 is not
965  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
966  *
967  * To prevent the lock order reversal and deadlock, vn_io_fault() does
968  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
969  * Instead, it first tries to do the whole range i/o with pagefaults
970  * disabled. If all pages in the i/o buffer are resident and mapped,
971  * VOP will succeed (ignoring the genuine filesystem errors).
972  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
973  * i/o in chunks, with all pages in the chunk prefaulted and held
974  * using vm_fault_quick_hold_pages().
975  *
976  * Filesystems using this deadlock avoidance scheme should use the
977  * array of the held pages from uio, saved in the curthread->td_ma,
978  * instead of doing uiomove().  A helper function
979  * vn_io_fault_uiomove() converts uiomove request into
980  * uiomove_fromphys() over td_ma array.
981  *
982  * Since vnode locks do not cover the whole i/o anymore, rangelocks
983  * make the current i/o request atomic with respect to other i/os and
984  * truncations.
985  */
986 
987 /*
988  * Decode vn_io_fault_args and perform the corresponding i/o.
989  */
990 static int
991 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
992     struct thread *td)
993 {
994 
995 	switch (args->kind) {
996 	case VN_IO_FAULT_FOP:
997 		return ((args->args.fop_args.doio)(args->args.fop_args.fp,
998 		    uio, args->cred, args->flags, td));
999 	case VN_IO_FAULT_VOP:
1000 		if (uio->uio_rw == UIO_READ) {
1001 			return (VOP_READ(args->args.vop_args.vp, uio,
1002 			    args->flags, args->cred));
1003 		} else if (uio->uio_rw == UIO_WRITE) {
1004 			return (VOP_WRITE(args->args.vop_args.vp, uio,
1005 			    args->flags, args->cred));
1006 		}
1007 		break;
1008 	}
1009 	panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
1010 	    uio->uio_rw);
1011 }
1012 
1013 /*
1014  * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1015  * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1016  * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1017  * into args and call vn_io_fault1() to handle faults during the user
1018  * mode buffer accesses.
1019  */
1020 static int
1021 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1022     struct thread *td)
1023 {
1024 	vm_page_t ma[io_hold_cnt + 2];
1025 	struct uio *uio_clone, short_uio;
1026 	struct iovec short_iovec[1];
1027 	vm_page_t *prev_td_ma;
1028 	vm_prot_t prot;
1029 	vm_offset_t addr, end;
1030 	size_t len, resid;
1031 	ssize_t adv;
1032 	int error, cnt, save, saveheld, prev_td_ma_cnt;
1033 
1034 	prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1035 
1036 	/*
1037 	 * The UFS follows IO_UNIT directive and replays back both
1038 	 * uio_offset and uio_resid if an error is encountered during the
1039 	 * operation.  But, since the iovec may be already advanced,
1040 	 * uio is still in an inconsistent state.
1041 	 *
1042 	 * Cache a copy of the original uio, which is advanced to the redo
1043 	 * point using UIO_NOCOPY below.
1044 	 */
1045 	uio_clone = cloneuio(uio);
1046 	resid = uio->uio_resid;
1047 
1048 	short_uio.uio_segflg = UIO_USERSPACE;
1049 	short_uio.uio_rw = uio->uio_rw;
1050 	short_uio.uio_td = uio->uio_td;
1051 
1052 	save = vm_fault_disable_pagefaults();
1053 	error = vn_io_fault_doio(args, uio, td);
1054 	if (error != EFAULT)
1055 		goto out;
1056 
1057 	atomic_add_long(&vn_io_faults_cnt, 1);
1058 	uio_clone->uio_segflg = UIO_NOCOPY;
1059 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
1060 	uio_clone->uio_segflg = uio->uio_segflg;
1061 
1062 	saveheld = curthread_pflags_set(TDP_UIOHELD);
1063 	prev_td_ma = td->td_ma;
1064 	prev_td_ma_cnt = td->td_ma_cnt;
1065 
1066 	while (uio_clone->uio_resid != 0) {
1067 		len = uio_clone->uio_iov->iov_len;
1068 		if (len == 0) {
1069 			KASSERT(uio_clone->uio_iovcnt >= 1,
1070 			    ("iovcnt underflow"));
1071 			uio_clone->uio_iov++;
1072 			uio_clone->uio_iovcnt--;
1073 			continue;
1074 		}
1075 		if (len > io_hold_cnt * PAGE_SIZE)
1076 			len = io_hold_cnt * PAGE_SIZE;
1077 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1078 		end = round_page(addr + len);
1079 		if (end < addr) {
1080 			error = EFAULT;
1081 			break;
1082 		}
1083 		cnt = atop(end - trunc_page(addr));
1084 		/*
1085 		 * A perfectly misaligned address and length could cause
1086 		 * both the start and the end of the chunk to use partial
1087 		 * page.  +2 accounts for such a situation.
1088 		 */
1089 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1090 		    addr, len, prot, ma, io_hold_cnt + 2);
1091 		if (cnt == -1) {
1092 			error = EFAULT;
1093 			break;
1094 		}
1095 		short_uio.uio_iov = &short_iovec[0];
1096 		short_iovec[0].iov_base = (void *)addr;
1097 		short_uio.uio_iovcnt = 1;
1098 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1099 		short_uio.uio_offset = uio_clone->uio_offset;
1100 		td->td_ma = ma;
1101 		td->td_ma_cnt = cnt;
1102 
1103 		error = vn_io_fault_doio(args, &short_uio, td);
1104 		vm_page_unhold_pages(ma, cnt);
1105 		adv = len - short_uio.uio_resid;
1106 
1107 		uio_clone->uio_iov->iov_base =
1108 		    (char *)uio_clone->uio_iov->iov_base + adv;
1109 		uio_clone->uio_iov->iov_len -= adv;
1110 		uio_clone->uio_resid -= adv;
1111 		uio_clone->uio_offset += adv;
1112 
1113 		uio->uio_resid -= adv;
1114 		uio->uio_offset += adv;
1115 
1116 		if (error != 0 || adv == 0)
1117 			break;
1118 	}
1119 	td->td_ma = prev_td_ma;
1120 	td->td_ma_cnt = prev_td_ma_cnt;
1121 	curthread_pflags_restore(saveheld);
1122 out:
1123 	vm_fault_enable_pagefaults(save);
1124 	free(uio_clone, M_IOV);
1125 	return (error);
1126 }
1127 
1128 static int
1129 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1130     int flags, struct thread *td)
1131 {
1132 	fo_rdwr_t *doio;
1133 	struct vnode *vp;
1134 	void *rl_cookie;
1135 	struct vn_io_fault_args args;
1136 	int error;
1137 
1138 	doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1139 	vp = fp->f_vnode;
1140 	foffset_lock_uio(fp, uio, flags);
1141 	if (do_vn_io_fault(vp, uio)) {
1142 		args.kind = VN_IO_FAULT_FOP;
1143 		args.args.fop_args.fp = fp;
1144 		args.args.fop_args.doio = doio;
1145 		args.cred = active_cred;
1146 		args.flags = flags | FOF_OFFSET;
1147 		if (uio->uio_rw == UIO_READ) {
1148 			rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1149 			    uio->uio_offset + uio->uio_resid);
1150 		} else if ((fp->f_flag & O_APPEND) != 0 ||
1151 		    (flags & FOF_OFFSET) == 0) {
1152 			/* For appenders, punt and lock the whole range. */
1153 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1154 		} else {
1155 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1156 			    uio->uio_offset + uio->uio_resid);
1157 		}
1158 		error = vn_io_fault1(vp, uio, &args, td);
1159 		vn_rangelock_unlock(vp, rl_cookie);
1160 	} else {
1161 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1162 	}
1163 	foffset_unlock_uio(fp, uio, flags);
1164 	return (error);
1165 }
1166 
1167 /*
1168  * Helper function to perform the requested uiomove operation using
1169  * the held pages for io->uio_iov[0].iov_base buffer instead of
1170  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1171  * instead of iov_base prevents page faults that could occur due to
1172  * pmap_collect() invalidating the mapping created by
1173  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1174  * object cleanup revoking the write access from page mappings.
1175  *
1176  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1177  * instead of plain uiomove().
1178  */
1179 int
1180 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1181 {
1182 	struct uio transp_uio;
1183 	struct iovec transp_iov[1];
1184 	struct thread *td;
1185 	size_t adv;
1186 	int error, pgadv;
1187 
1188 	td = curthread;
1189 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1190 	    uio->uio_segflg != UIO_USERSPACE)
1191 		return (uiomove(data, xfersize, uio));
1192 
1193 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1194 	transp_iov[0].iov_base = data;
1195 	transp_uio.uio_iov = &transp_iov[0];
1196 	transp_uio.uio_iovcnt = 1;
1197 	if (xfersize > uio->uio_resid)
1198 		xfersize = uio->uio_resid;
1199 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1200 	transp_uio.uio_offset = 0;
1201 	transp_uio.uio_segflg = UIO_SYSSPACE;
1202 	/*
1203 	 * Since transp_iov points to data, and td_ma page array
1204 	 * corresponds to original uio->uio_iov, we need to invert the
1205 	 * direction of the i/o operation as passed to
1206 	 * uiomove_fromphys().
1207 	 */
1208 	switch (uio->uio_rw) {
1209 	case UIO_WRITE:
1210 		transp_uio.uio_rw = UIO_READ;
1211 		break;
1212 	case UIO_READ:
1213 		transp_uio.uio_rw = UIO_WRITE;
1214 		break;
1215 	}
1216 	transp_uio.uio_td = uio->uio_td;
1217 	error = uiomove_fromphys(td->td_ma,
1218 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1219 	    xfersize, &transp_uio);
1220 	adv = xfersize - transp_uio.uio_resid;
1221 	pgadv =
1222 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1223 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1224 	td->td_ma += pgadv;
1225 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1226 	    pgadv));
1227 	td->td_ma_cnt -= pgadv;
1228 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1229 	uio->uio_iov->iov_len -= adv;
1230 	uio->uio_resid -= adv;
1231 	uio->uio_offset += adv;
1232 	return (error);
1233 }
1234 
1235 int
1236 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1237     struct uio *uio)
1238 {
1239 	struct thread *td;
1240 	vm_offset_t iov_base;
1241 	int cnt, pgadv;
1242 
1243 	td = curthread;
1244 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1245 	    uio->uio_segflg != UIO_USERSPACE)
1246 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1247 
1248 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1249 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1250 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1251 	switch (uio->uio_rw) {
1252 	case UIO_WRITE:
1253 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1254 		    offset, cnt);
1255 		break;
1256 	case UIO_READ:
1257 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1258 		    cnt);
1259 		break;
1260 	}
1261 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1262 	td->td_ma += pgadv;
1263 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1264 	    pgadv));
1265 	td->td_ma_cnt -= pgadv;
1266 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1267 	uio->uio_iov->iov_len -= cnt;
1268 	uio->uio_resid -= cnt;
1269 	uio->uio_offset += cnt;
1270 	return (0);
1271 }
1272 
1273 
1274 /*
1275  * File table truncate routine.
1276  */
1277 static int
1278 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1279     struct thread *td)
1280 {
1281 	struct vattr vattr;
1282 	struct mount *mp;
1283 	struct vnode *vp;
1284 	void *rl_cookie;
1285 	int error;
1286 
1287 	vp = fp->f_vnode;
1288 
1289 	/*
1290 	 * Lock the whole range for truncation.  Otherwise split i/o
1291 	 * might happen partly before and partly after the truncation.
1292 	 */
1293 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1294 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1295 	if (error)
1296 		goto out1;
1297 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1298 	if (vp->v_type == VDIR) {
1299 		error = EISDIR;
1300 		goto out;
1301 	}
1302 #ifdef MAC
1303 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1304 	if (error)
1305 		goto out;
1306 #endif
1307 	error = vn_writechk(vp);
1308 	if (error == 0) {
1309 		VATTR_NULL(&vattr);
1310 		vattr.va_size = length;
1311 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1312 	}
1313 out:
1314 	VOP_UNLOCK(vp, 0);
1315 	vn_finished_write(mp);
1316 out1:
1317 	vn_rangelock_unlock(vp, rl_cookie);
1318 	return (error);
1319 }
1320 
1321 /*
1322  * File table vnode stat routine.
1323  */
1324 static int
1325 vn_statfile(fp, sb, active_cred, td)
1326 	struct file *fp;
1327 	struct stat *sb;
1328 	struct ucred *active_cred;
1329 	struct thread *td;
1330 {
1331 	struct vnode *vp = fp->f_vnode;
1332 	int error;
1333 
1334 	vn_lock(vp, LK_SHARED | LK_RETRY);
1335 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1336 	VOP_UNLOCK(vp, 0);
1337 
1338 	return (error);
1339 }
1340 
1341 /*
1342  * Stat a vnode; implementation for the stat syscall
1343  */
1344 int
1345 vn_stat(vp, sb, active_cred, file_cred, td)
1346 	struct vnode *vp;
1347 	register struct stat *sb;
1348 	struct ucred *active_cred;
1349 	struct ucred *file_cred;
1350 	struct thread *td;
1351 {
1352 	struct vattr vattr;
1353 	register struct vattr *vap;
1354 	int error;
1355 	u_short mode;
1356 
1357 #ifdef MAC
1358 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1359 	if (error)
1360 		return (error);
1361 #endif
1362 
1363 	vap = &vattr;
1364 
1365 	/*
1366 	 * Initialize defaults for new and unusual fields, so that file
1367 	 * systems which don't support these fields don't need to know
1368 	 * about them.
1369 	 */
1370 	vap->va_birthtime.tv_sec = -1;
1371 	vap->va_birthtime.tv_nsec = 0;
1372 	vap->va_fsid = VNOVAL;
1373 	vap->va_rdev = NODEV;
1374 
1375 	error = VOP_GETATTR(vp, vap, active_cred);
1376 	if (error)
1377 		return (error);
1378 
1379 	/*
1380 	 * Zero the spare stat fields
1381 	 */
1382 	bzero(sb, sizeof *sb);
1383 
1384 	/*
1385 	 * Copy from vattr table
1386 	 */
1387 	if (vap->va_fsid != VNOVAL)
1388 		sb->st_dev = vap->va_fsid;
1389 	else
1390 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1391 	sb->st_ino = vap->va_fileid;
1392 	mode = vap->va_mode;
1393 	switch (vap->va_type) {
1394 	case VREG:
1395 		mode |= S_IFREG;
1396 		break;
1397 	case VDIR:
1398 		mode |= S_IFDIR;
1399 		break;
1400 	case VBLK:
1401 		mode |= S_IFBLK;
1402 		break;
1403 	case VCHR:
1404 		mode |= S_IFCHR;
1405 		break;
1406 	case VLNK:
1407 		mode |= S_IFLNK;
1408 		break;
1409 	case VSOCK:
1410 		mode |= S_IFSOCK;
1411 		break;
1412 	case VFIFO:
1413 		mode |= S_IFIFO;
1414 		break;
1415 	default:
1416 		return (EBADF);
1417 	};
1418 	sb->st_mode = mode;
1419 	sb->st_nlink = vap->va_nlink;
1420 	sb->st_uid = vap->va_uid;
1421 	sb->st_gid = vap->va_gid;
1422 	sb->st_rdev = vap->va_rdev;
1423 	if (vap->va_size > OFF_MAX)
1424 		return (EOVERFLOW);
1425 	sb->st_size = vap->va_size;
1426 	sb->st_atim = vap->va_atime;
1427 	sb->st_mtim = vap->va_mtime;
1428 	sb->st_ctim = vap->va_ctime;
1429 	sb->st_birthtim = vap->va_birthtime;
1430 
1431         /*
1432 	 * According to www.opengroup.org, the meaning of st_blksize is
1433 	 *   "a filesystem-specific preferred I/O block size for this
1434 	 *    object.  In some filesystem types, this may vary from file
1435 	 *    to file"
1436 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1437 	 */
1438 
1439 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1440 
1441 	sb->st_flags = vap->va_flags;
1442 	if (priv_check(td, PRIV_VFS_GENERATION))
1443 		sb->st_gen = 0;
1444 	else
1445 		sb->st_gen = vap->va_gen;
1446 
1447 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1448 	return (0);
1449 }
1450 
1451 /*
1452  * File table vnode ioctl routine.
1453  */
1454 static int
1455 vn_ioctl(fp, com, data, active_cred, td)
1456 	struct file *fp;
1457 	u_long com;
1458 	void *data;
1459 	struct ucred *active_cred;
1460 	struct thread *td;
1461 {
1462 	struct vattr vattr;
1463 	struct vnode *vp;
1464 	int error;
1465 
1466 	vp = fp->f_vnode;
1467 	switch (vp->v_type) {
1468 	case VDIR:
1469 	case VREG:
1470 		switch (com) {
1471 		case FIONREAD:
1472 			vn_lock(vp, LK_SHARED | LK_RETRY);
1473 			error = VOP_GETATTR(vp, &vattr, active_cred);
1474 			VOP_UNLOCK(vp, 0);
1475 			if (error == 0)
1476 				*(int *)data = vattr.va_size - fp->f_offset;
1477 			return (error);
1478 		case FIONBIO:
1479 		case FIOASYNC:
1480 			return (0);
1481 		default:
1482 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1483 			    active_cred, td));
1484 		}
1485 	default:
1486 		return (ENOTTY);
1487 	}
1488 }
1489 
1490 /*
1491  * File table vnode poll routine.
1492  */
1493 static int
1494 vn_poll(fp, events, active_cred, td)
1495 	struct file *fp;
1496 	int events;
1497 	struct ucred *active_cred;
1498 	struct thread *td;
1499 {
1500 	struct vnode *vp;
1501 	int error;
1502 
1503 	vp = fp->f_vnode;
1504 #ifdef MAC
1505 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1506 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1507 	VOP_UNLOCK(vp, 0);
1508 	if (!error)
1509 #endif
1510 
1511 	error = VOP_POLL(vp, events, fp->f_cred, td);
1512 	return (error);
1513 }
1514 
1515 /*
1516  * Acquire the requested lock and then check for validity.  LK_RETRY
1517  * permits vn_lock to return doomed vnodes.
1518  */
1519 int
1520 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1521 {
1522 	int error;
1523 
1524 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1525 	    ("vn_lock called with no locktype."));
1526 	do {
1527 #ifdef DEBUG_VFS_LOCKS
1528 		KASSERT(vp->v_holdcnt != 0,
1529 		    ("vn_lock %p: zero hold count", vp));
1530 #endif
1531 		error = VOP_LOCK1(vp, flags, file, line);
1532 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1533 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1534 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1535 		    flags, error));
1536 		/*
1537 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1538 		 * If RETRY is not set, we return ENOENT instead.
1539 		 */
1540 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1541 		    (flags & LK_RETRY) == 0) {
1542 			VOP_UNLOCK(vp, 0);
1543 			error = ENOENT;
1544 			break;
1545 		}
1546 	} while (flags & LK_RETRY && error != 0);
1547 	return (error);
1548 }
1549 
1550 /*
1551  * File table vnode close routine.
1552  */
1553 static int
1554 vn_closefile(fp, td)
1555 	struct file *fp;
1556 	struct thread *td;
1557 {
1558 	struct vnode *vp;
1559 	struct flock lf;
1560 	int error;
1561 
1562 	vp = fp->f_vnode;
1563 	fp->f_ops = &badfileops;
1564 
1565 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1566 		vref(vp);
1567 
1568 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1569 
1570 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1571 		lf.l_whence = SEEK_SET;
1572 		lf.l_start = 0;
1573 		lf.l_len = 0;
1574 		lf.l_type = F_UNLCK;
1575 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1576 		vrele(vp);
1577 	}
1578 	return (error);
1579 }
1580 
1581 static bool
1582 vn_suspendable_mp(struct mount *mp)
1583 {
1584 
1585 	return ((mp->mnt_kern_flag & MNTK_SUSPENDABLE) != 0);
1586 }
1587 
1588 static bool
1589 vn_suspendable(struct vnode *vp, struct mount **mpp)
1590 {
1591 
1592 	if (vp != NULL)
1593 		*mpp = vp->v_mount;
1594 	if (*mpp == NULL)
1595 		return (false);
1596 
1597 	return (vn_suspendable_mp(*mpp));
1598 }
1599 
1600 /*
1601  * Preparing to start a filesystem write operation. If the operation is
1602  * permitted, then we bump the count of operations in progress and
1603  * proceed. If a suspend request is in progress, we wait until the
1604  * suspension is over, and then proceed.
1605  */
1606 static int
1607 vn_start_write_locked(struct mount *mp, int flags)
1608 {
1609 	int error, mflags;
1610 
1611 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1612 	error = 0;
1613 
1614 	/*
1615 	 * Check on status of suspension.
1616 	 */
1617 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1618 	    mp->mnt_susp_owner != curthread) {
1619 		mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1620 		    (flags & PCATCH) : 0) | (PUSER - 1);
1621 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1622 			if (flags & V_NOWAIT) {
1623 				error = EWOULDBLOCK;
1624 				goto unlock;
1625 			}
1626 			error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1627 			    "suspfs", 0);
1628 			if (error)
1629 				goto unlock;
1630 		}
1631 	}
1632 	if (flags & V_XSLEEP)
1633 		goto unlock;
1634 	mp->mnt_writeopcount++;
1635 unlock:
1636 	if (error != 0 || (flags & V_XSLEEP) != 0)
1637 		MNT_REL(mp);
1638 	MNT_IUNLOCK(mp);
1639 	return (error);
1640 }
1641 
1642 int
1643 vn_start_write(vp, mpp, flags)
1644 	struct vnode *vp;
1645 	struct mount **mpp;
1646 	int flags;
1647 {
1648 	struct mount *mp;
1649 	int error;
1650 
1651 	if (!vn_suspendable(vp, mpp))
1652 		return (0);
1653 
1654 	error = 0;
1655 	/*
1656 	 * If a vnode is provided, get and return the mount point that
1657 	 * to which it will write.
1658 	 */
1659 	if (vp != NULL) {
1660 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1661 			*mpp = NULL;
1662 			if (error != EOPNOTSUPP)
1663 				return (error);
1664 			return (0);
1665 		}
1666 	}
1667 	if ((mp = *mpp) == NULL)
1668 		return (0);
1669 
1670 	/*
1671 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1672 	 * a vfs_ref().
1673 	 * As long as a vnode is not provided we need to acquire a
1674 	 * refcount for the provided mountpoint too, in order to
1675 	 * emulate a vfs_ref().
1676 	 */
1677 	MNT_ILOCK(mp);
1678 	if (vp == NULL)
1679 		MNT_REF(mp);
1680 
1681 	return (vn_start_write_locked(mp, flags));
1682 }
1683 
1684 /*
1685  * Secondary suspension. Used by operations such as vop_inactive
1686  * routines that are needed by the higher level functions. These
1687  * are allowed to proceed until all the higher level functions have
1688  * completed (indicated by mnt_writeopcount dropping to zero). At that
1689  * time, these operations are halted until the suspension is over.
1690  */
1691 int
1692 vn_start_secondary_write(vp, mpp, flags)
1693 	struct vnode *vp;
1694 	struct mount **mpp;
1695 	int flags;
1696 {
1697 	struct mount *mp;
1698 	int error;
1699 
1700 	if (!vn_suspendable(vp, mpp))
1701 		return (0);
1702 
1703  retry:
1704 	if (vp != NULL) {
1705 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1706 			*mpp = NULL;
1707 			if (error != EOPNOTSUPP)
1708 				return (error);
1709 			return (0);
1710 		}
1711 	}
1712 	/*
1713 	 * If we are not suspended or have not yet reached suspended
1714 	 * mode, then let the operation proceed.
1715 	 */
1716 	if ((mp = *mpp) == NULL)
1717 		return (0);
1718 
1719 	/*
1720 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1721 	 * a vfs_ref().
1722 	 * As long as a vnode is not provided we need to acquire a
1723 	 * refcount for the provided mountpoint too, in order to
1724 	 * emulate a vfs_ref().
1725 	 */
1726 	MNT_ILOCK(mp);
1727 	if (vp == NULL)
1728 		MNT_REF(mp);
1729 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1730 		mp->mnt_secondary_writes++;
1731 		mp->mnt_secondary_accwrites++;
1732 		MNT_IUNLOCK(mp);
1733 		return (0);
1734 	}
1735 	if (flags & V_NOWAIT) {
1736 		MNT_REL(mp);
1737 		MNT_IUNLOCK(mp);
1738 		return (EWOULDBLOCK);
1739 	}
1740 	/*
1741 	 * Wait for the suspension to finish.
1742 	 */
1743 	error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1744 	    ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1745 	    "suspfs", 0);
1746 	vfs_rel(mp);
1747 	if (error == 0)
1748 		goto retry;
1749 	return (error);
1750 }
1751 
1752 /*
1753  * Filesystem write operation has completed. If we are suspending and this
1754  * operation is the last one, notify the suspender that the suspension is
1755  * now in effect.
1756  */
1757 void
1758 vn_finished_write(mp)
1759 	struct mount *mp;
1760 {
1761 	if (mp == NULL || !vn_suspendable_mp(mp))
1762 		return;
1763 	MNT_ILOCK(mp);
1764 	MNT_REL(mp);
1765 	mp->mnt_writeopcount--;
1766 	if (mp->mnt_writeopcount < 0)
1767 		panic("vn_finished_write: neg cnt");
1768 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1769 	    mp->mnt_writeopcount <= 0)
1770 		wakeup(&mp->mnt_writeopcount);
1771 	MNT_IUNLOCK(mp);
1772 }
1773 
1774 
1775 /*
1776  * Filesystem secondary write operation has completed. If we are
1777  * suspending and this operation is the last one, notify the suspender
1778  * that the suspension is now in effect.
1779  */
1780 void
1781 vn_finished_secondary_write(mp)
1782 	struct mount *mp;
1783 {
1784 	if (mp == NULL || !vn_suspendable_mp(mp))
1785 		return;
1786 	MNT_ILOCK(mp);
1787 	MNT_REL(mp);
1788 	mp->mnt_secondary_writes--;
1789 	if (mp->mnt_secondary_writes < 0)
1790 		panic("vn_finished_secondary_write: neg cnt");
1791 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1792 	    mp->mnt_secondary_writes <= 0)
1793 		wakeup(&mp->mnt_secondary_writes);
1794 	MNT_IUNLOCK(mp);
1795 }
1796 
1797 
1798 
1799 /*
1800  * Request a filesystem to suspend write operations.
1801  */
1802 int
1803 vfs_write_suspend(struct mount *mp, int flags)
1804 {
1805 	int error;
1806 
1807 	MPASS(vn_suspendable_mp(mp));
1808 
1809 	MNT_ILOCK(mp);
1810 	if (mp->mnt_susp_owner == curthread) {
1811 		MNT_IUNLOCK(mp);
1812 		return (EALREADY);
1813 	}
1814 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1815 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1816 
1817 	/*
1818 	 * Unmount holds a write reference on the mount point.  If we
1819 	 * own busy reference and drain for writers, we deadlock with
1820 	 * the reference draining in the unmount path.  Callers of
1821 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1822 	 * vfs_busy() reference is owned and caller is not in the
1823 	 * unmount context.
1824 	 */
1825 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1826 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1827 		MNT_IUNLOCK(mp);
1828 		return (EBUSY);
1829 	}
1830 
1831 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1832 	mp->mnt_susp_owner = curthread;
1833 	if (mp->mnt_writeopcount > 0)
1834 		(void) msleep(&mp->mnt_writeopcount,
1835 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1836 	else
1837 		MNT_IUNLOCK(mp);
1838 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1839 		vfs_write_resume(mp, 0);
1840 	return (error);
1841 }
1842 
1843 /*
1844  * Request a filesystem to resume write operations.
1845  */
1846 void
1847 vfs_write_resume(struct mount *mp, int flags)
1848 {
1849 
1850 	MPASS(vn_suspendable_mp(mp));
1851 
1852 	MNT_ILOCK(mp);
1853 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1854 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1855 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1856 				       MNTK_SUSPENDED);
1857 		mp->mnt_susp_owner = NULL;
1858 		wakeup(&mp->mnt_writeopcount);
1859 		wakeup(&mp->mnt_flag);
1860 		curthread->td_pflags &= ~TDP_IGNSUSP;
1861 		if ((flags & VR_START_WRITE) != 0) {
1862 			MNT_REF(mp);
1863 			mp->mnt_writeopcount++;
1864 		}
1865 		MNT_IUNLOCK(mp);
1866 		if ((flags & VR_NO_SUSPCLR) == 0)
1867 			VFS_SUSP_CLEAN(mp);
1868 	} else if ((flags & VR_START_WRITE) != 0) {
1869 		MNT_REF(mp);
1870 		vn_start_write_locked(mp, 0);
1871 	} else {
1872 		MNT_IUNLOCK(mp);
1873 	}
1874 }
1875 
1876 /*
1877  * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1878  * methods.
1879  */
1880 int
1881 vfs_write_suspend_umnt(struct mount *mp)
1882 {
1883 	int error;
1884 
1885 	MPASS(vn_suspendable_mp(mp));
1886 	KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1887 	    ("vfs_write_suspend_umnt: recursed"));
1888 
1889 	/* dounmount() already called vn_start_write(). */
1890 	for (;;) {
1891 		vn_finished_write(mp);
1892 		error = vfs_write_suspend(mp, 0);
1893 		if (error != 0) {
1894 			vn_start_write(NULL, &mp, V_WAIT);
1895 			return (error);
1896 		}
1897 		MNT_ILOCK(mp);
1898 		if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1899 			break;
1900 		MNT_IUNLOCK(mp);
1901 		vn_start_write(NULL, &mp, V_WAIT);
1902 	}
1903 	mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1904 	wakeup(&mp->mnt_flag);
1905 	MNT_IUNLOCK(mp);
1906 	curthread->td_pflags |= TDP_IGNSUSP;
1907 	return (0);
1908 }
1909 
1910 /*
1911  * Implement kqueues for files by translating it to vnode operation.
1912  */
1913 static int
1914 vn_kqfilter(struct file *fp, struct knote *kn)
1915 {
1916 
1917 	return (VOP_KQFILTER(fp->f_vnode, kn));
1918 }
1919 
1920 /*
1921  * Simplified in-kernel wrapper calls for extended attribute access.
1922  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1923  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1924  */
1925 int
1926 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1927     const char *attrname, int *buflen, char *buf, struct thread *td)
1928 {
1929 	struct uio	auio;
1930 	struct iovec	iov;
1931 	int	error;
1932 
1933 	iov.iov_len = *buflen;
1934 	iov.iov_base = buf;
1935 
1936 	auio.uio_iov = &iov;
1937 	auio.uio_iovcnt = 1;
1938 	auio.uio_rw = UIO_READ;
1939 	auio.uio_segflg = UIO_SYSSPACE;
1940 	auio.uio_td = td;
1941 	auio.uio_offset = 0;
1942 	auio.uio_resid = *buflen;
1943 
1944 	if ((ioflg & IO_NODELOCKED) == 0)
1945 		vn_lock(vp, LK_SHARED | LK_RETRY);
1946 
1947 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1948 
1949 	/* authorize attribute retrieval as kernel */
1950 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1951 	    td);
1952 
1953 	if ((ioflg & IO_NODELOCKED) == 0)
1954 		VOP_UNLOCK(vp, 0);
1955 
1956 	if (error == 0) {
1957 		*buflen = *buflen - auio.uio_resid;
1958 	}
1959 
1960 	return (error);
1961 }
1962 
1963 /*
1964  * XXX failure mode if partially written?
1965  */
1966 int
1967 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1968     const char *attrname, int buflen, char *buf, struct thread *td)
1969 {
1970 	struct uio	auio;
1971 	struct iovec	iov;
1972 	struct mount	*mp;
1973 	int	error;
1974 
1975 	iov.iov_len = buflen;
1976 	iov.iov_base = buf;
1977 
1978 	auio.uio_iov = &iov;
1979 	auio.uio_iovcnt = 1;
1980 	auio.uio_rw = UIO_WRITE;
1981 	auio.uio_segflg = UIO_SYSSPACE;
1982 	auio.uio_td = td;
1983 	auio.uio_offset = 0;
1984 	auio.uio_resid = buflen;
1985 
1986 	if ((ioflg & IO_NODELOCKED) == 0) {
1987 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1988 			return (error);
1989 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1990 	}
1991 
1992 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1993 
1994 	/* authorize attribute setting as kernel */
1995 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1996 
1997 	if ((ioflg & IO_NODELOCKED) == 0) {
1998 		vn_finished_write(mp);
1999 		VOP_UNLOCK(vp, 0);
2000 	}
2001 
2002 	return (error);
2003 }
2004 
2005 int
2006 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2007     const char *attrname, struct thread *td)
2008 {
2009 	struct mount	*mp;
2010 	int	error;
2011 
2012 	if ((ioflg & IO_NODELOCKED) == 0) {
2013 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2014 			return (error);
2015 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2016 	}
2017 
2018 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2019 
2020 	/* authorize attribute removal as kernel */
2021 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2022 	if (error == EOPNOTSUPP)
2023 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2024 		    NULL, td);
2025 
2026 	if ((ioflg & IO_NODELOCKED) == 0) {
2027 		vn_finished_write(mp);
2028 		VOP_UNLOCK(vp, 0);
2029 	}
2030 
2031 	return (error);
2032 }
2033 
2034 static int
2035 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2036     struct vnode **rvp)
2037 {
2038 
2039 	return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2040 }
2041 
2042 int
2043 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2044 {
2045 
2046 	return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2047 	    lkflags, rvp));
2048 }
2049 
2050 int
2051 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2052     int lkflags, struct vnode **rvp)
2053 {
2054 	struct mount *mp;
2055 	int ltype, error;
2056 
2057 	ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2058 	mp = vp->v_mount;
2059 	ltype = VOP_ISLOCKED(vp);
2060 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2061 	    ("vn_vget_ino: vp not locked"));
2062 	error = vfs_busy(mp, MBF_NOWAIT);
2063 	if (error != 0) {
2064 		vfs_ref(mp);
2065 		VOP_UNLOCK(vp, 0);
2066 		error = vfs_busy(mp, 0);
2067 		vn_lock(vp, ltype | LK_RETRY);
2068 		vfs_rel(mp);
2069 		if (error != 0)
2070 			return (ENOENT);
2071 		if (vp->v_iflag & VI_DOOMED) {
2072 			vfs_unbusy(mp);
2073 			return (ENOENT);
2074 		}
2075 	}
2076 	VOP_UNLOCK(vp, 0);
2077 	error = alloc(mp, alloc_arg, lkflags, rvp);
2078 	vfs_unbusy(mp);
2079 	if (*rvp != vp)
2080 		vn_lock(vp, ltype | LK_RETRY);
2081 	if (vp->v_iflag & VI_DOOMED) {
2082 		if (error == 0) {
2083 			if (*rvp == vp)
2084 				vunref(vp);
2085 			else
2086 				vput(*rvp);
2087 		}
2088 		error = ENOENT;
2089 	}
2090 	return (error);
2091 }
2092 
2093 int
2094 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2095     const struct thread *td)
2096 {
2097 
2098 	if (vp->v_type != VREG || td == NULL)
2099 		return (0);
2100 	PROC_LOCK(td->td_proc);
2101 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
2102 	    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
2103 		kern_psignal(td->td_proc, SIGXFSZ);
2104 		PROC_UNLOCK(td->td_proc);
2105 		return (EFBIG);
2106 	}
2107 	PROC_UNLOCK(td->td_proc);
2108 	return (0);
2109 }
2110 
2111 int
2112 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2113     struct thread *td)
2114 {
2115 	struct vnode *vp;
2116 
2117 	vp = fp->f_vnode;
2118 #ifdef AUDIT
2119 	vn_lock(vp, LK_SHARED | LK_RETRY);
2120 	AUDIT_ARG_VNODE1(vp);
2121 	VOP_UNLOCK(vp, 0);
2122 #endif
2123 	return (setfmode(td, active_cred, vp, mode));
2124 }
2125 
2126 int
2127 vn_chown(struct file *fp, uid_t uid, gid_t gid, 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 (setfown(td, active_cred, vp, uid, gid));
2139 }
2140 
2141 void
2142 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2143 {
2144 	vm_object_t object;
2145 
2146 	if ((object = vp->v_object) == NULL)
2147 		return;
2148 	VM_OBJECT_WLOCK(object);
2149 	vm_object_page_remove(object, start, end, 0);
2150 	VM_OBJECT_WUNLOCK(object);
2151 }
2152 
2153 int
2154 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2155 {
2156 	struct vattr va;
2157 	daddr_t bn, bnp;
2158 	uint64_t bsize;
2159 	off_t noff;
2160 	int error;
2161 
2162 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2163 	    ("Wrong command %lu", cmd));
2164 
2165 	if (vn_lock(vp, LK_SHARED) != 0)
2166 		return (EBADF);
2167 	if (vp->v_type != VREG) {
2168 		error = ENOTTY;
2169 		goto unlock;
2170 	}
2171 	error = VOP_GETATTR(vp, &va, cred);
2172 	if (error != 0)
2173 		goto unlock;
2174 	noff = *off;
2175 	if (noff >= va.va_size) {
2176 		error = ENXIO;
2177 		goto unlock;
2178 	}
2179 	bsize = vp->v_mount->mnt_stat.f_iosize;
2180 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2181 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2182 		if (error == EOPNOTSUPP) {
2183 			error = ENOTTY;
2184 			goto unlock;
2185 		}
2186 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2187 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
2188 			noff = bn * bsize;
2189 			if (noff < *off)
2190 				noff = *off;
2191 			goto unlock;
2192 		}
2193 	}
2194 	if (noff > va.va_size)
2195 		noff = va.va_size;
2196 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2197 	if (cmd == FIOSEEKDATA)
2198 		error = ENXIO;
2199 unlock:
2200 	VOP_UNLOCK(vp, 0);
2201 	if (error == 0)
2202 		*off = noff;
2203 	return (error);
2204 }
2205 
2206 int
2207 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2208 {
2209 	struct ucred *cred;
2210 	struct vnode *vp;
2211 	struct vattr vattr;
2212 	off_t foffset, size;
2213 	int error, noneg;
2214 
2215 	cred = td->td_ucred;
2216 	vp = fp->f_vnode;
2217 	foffset = foffset_lock(fp, 0);
2218 	noneg = (vp->v_type != VCHR);
2219 	error = 0;
2220 	switch (whence) {
2221 	case L_INCR:
2222 		if (noneg &&
2223 		    (foffset < 0 ||
2224 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2225 			error = EOVERFLOW;
2226 			break;
2227 		}
2228 		offset += foffset;
2229 		break;
2230 	case L_XTND:
2231 		vn_lock(vp, LK_SHARED | LK_RETRY);
2232 		error = VOP_GETATTR(vp, &vattr, cred);
2233 		VOP_UNLOCK(vp, 0);
2234 		if (error)
2235 			break;
2236 
2237 		/*
2238 		 * If the file references a disk device, then fetch
2239 		 * the media size and use that to determine the ending
2240 		 * offset.
2241 		 */
2242 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2243 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2244 			vattr.va_size = size;
2245 		if (noneg &&
2246 		    (vattr.va_size > OFF_MAX ||
2247 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2248 			error = EOVERFLOW;
2249 			break;
2250 		}
2251 		offset += vattr.va_size;
2252 		break;
2253 	case L_SET:
2254 		break;
2255 	case SEEK_DATA:
2256 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2257 		break;
2258 	case SEEK_HOLE:
2259 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2260 		break;
2261 	default:
2262 		error = EINVAL;
2263 	}
2264 	if (error == 0 && noneg && offset < 0)
2265 		error = EINVAL;
2266 	if (error != 0)
2267 		goto drop;
2268 	VFS_KNOTE_UNLOCKED(vp, 0);
2269 	td->td_uretoff.tdu_off = offset;
2270 drop:
2271 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2272 	return (error);
2273 }
2274 
2275 int
2276 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2277     struct thread *td)
2278 {
2279 	int error;
2280 
2281 	/*
2282 	 * Grant permission if the caller is the owner of the file, or
2283 	 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2284 	 * on the file.  If the time pointer is null, then write
2285 	 * permission on the file is also sufficient.
2286 	 *
2287 	 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2288 	 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2289 	 * will be allowed to set the times [..] to the current
2290 	 * server time.
2291 	 */
2292 	error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2293 	if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2294 		error = VOP_ACCESS(vp, VWRITE, cred, td);
2295 	return (error);
2296 }
2297 
2298 int
2299 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2300 {
2301 	struct vnode *vp;
2302 	int error;
2303 
2304 	if (fp->f_type == DTYPE_FIFO)
2305 		kif->kf_type = KF_TYPE_FIFO;
2306 	else
2307 		kif->kf_type = KF_TYPE_VNODE;
2308 	vp = fp->f_vnode;
2309 	vref(vp);
2310 	FILEDESC_SUNLOCK(fdp);
2311 	error = vn_fill_kinfo_vnode(vp, kif);
2312 	vrele(vp);
2313 	FILEDESC_SLOCK(fdp);
2314 	return (error);
2315 }
2316 
2317 int
2318 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2319 {
2320 	struct vattr va;
2321 	char *fullpath, *freepath;
2322 	int error;
2323 
2324 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2325 	freepath = NULL;
2326 	fullpath = "-";
2327 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2328 	if (error == 0) {
2329 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2330 	}
2331 	if (freepath != NULL)
2332 		free(freepath, M_TEMP);
2333 
2334 	/*
2335 	 * Retrieve vnode attributes.
2336 	 */
2337 	va.va_fsid = VNOVAL;
2338 	va.va_rdev = NODEV;
2339 	vn_lock(vp, LK_SHARED | LK_RETRY);
2340 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2341 	VOP_UNLOCK(vp, 0);
2342 	if (error != 0)
2343 		return (error);
2344 	if (va.va_fsid != VNOVAL)
2345 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2346 	else
2347 		kif->kf_un.kf_file.kf_file_fsid =
2348 		    vp->v_mount->mnt_stat.f_fsid.val[0];
2349 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2350 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2351 	kif->kf_un.kf_file.kf_file_size = va.va_size;
2352 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2353 	return (0);
2354 }
2355