xref: /freebsd/sys/kern/vfs_vnops.c (revision 9ecd54f24fe9fa373e07c9fd7c052deb2188f545)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11  * Copyright (c) 2013 The FreeBSD Foundation
12  *
13  * Portions of this software were developed by Konstantin Belousov
14  * under sponsorship from the FreeBSD Foundation.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)vfs_vnops.c	8.2 (Berkeley) 1/21/94
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/disk.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/kdb.h>
52 #include <sys/stat.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/limits.h>
56 #include <sys/lock.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/vnode.h>
61 #include <sys/bio.h>
62 #include <sys/buf.h>
63 #include <sys/filio.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
66 #include <sys/sx.h>
67 #include <sys/sysctl.h>
68 #include <sys/ttycom.h>
69 #include <sys/conf.h>
70 #include <sys/syslog.h>
71 #include <sys/unistd.h>
72 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_extern.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
82 
83 static fo_rdwr_t	vn_read;
84 static fo_rdwr_t	vn_write;
85 static fo_rdwr_t	vn_io_fault;
86 static fo_truncate_t	vn_truncate;
87 static fo_ioctl_t	vn_ioctl;
88 static fo_poll_t	vn_poll;
89 static fo_kqfilter_t	vn_kqfilter;
90 static fo_stat_t	vn_statfile;
91 static fo_close_t	vn_closefile;
92 
93 struct 	fileops vnops = {
94 	.fo_read = vn_io_fault,
95 	.fo_write = vn_io_fault,
96 	.fo_truncate = vn_truncate,
97 	.fo_ioctl = vn_ioctl,
98 	.fo_poll = vn_poll,
99 	.fo_kqfilter = vn_kqfilter,
100 	.fo_stat = vn_statfile,
101 	.fo_close = vn_closefile,
102 	.fo_chmod = vn_chmod,
103 	.fo_chown = vn_chown,
104 	.fo_sendfile = vn_sendfile,
105 	.fo_seek = vn_seek,
106 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
107 };
108 
109 int
110 vn_open(ndp, flagp, cmode, fp)
111 	struct nameidata *ndp;
112 	int *flagp, cmode;
113 	struct file *fp;
114 {
115 	struct thread *td = ndp->ni_cnd.cn_thread;
116 
117 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
118 }
119 
120 /*
121  * Common code for vnode open operations via a name lookup.
122  * Lookup the vnode and invoke VOP_CREATE if needed.
123  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
124  *
125  * Note that this does NOT free nameidata for the successful case,
126  * due to the NDINIT being done elsewhere.
127  */
128 int
129 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
130     struct ucred *cred, struct file *fp)
131 {
132 	struct vnode *vp;
133 	struct mount *mp;
134 	struct thread *td = ndp->ni_cnd.cn_thread;
135 	struct vattr vat;
136 	struct vattr *vap = &vat;
137 	int fmode, error;
138 
139 restart:
140 	fmode = *flagp;
141 	if (fmode & O_CREAT) {
142 		ndp->ni_cnd.cn_nameiop = CREATE;
143 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
144 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
145 			ndp->ni_cnd.cn_flags |= FOLLOW;
146 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
147 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
148 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
149 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
150 		bwillwrite();
151 		if ((error = namei(ndp)) != 0)
152 			return (error);
153 		if (ndp->ni_vp == NULL) {
154 			VATTR_NULL(vap);
155 			vap->va_type = VREG;
156 			vap->va_mode = cmode;
157 			if (fmode & O_EXCL)
158 				vap->va_vaflags |= VA_EXCLUSIVE;
159 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
160 				NDFREE(ndp, NDF_ONLY_PNBUF);
161 				vput(ndp->ni_dvp);
162 				if ((error = vn_start_write(NULL, &mp,
163 				    V_XSLEEP | PCATCH)) != 0)
164 					return (error);
165 				goto restart;
166 			}
167 #ifdef MAC
168 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
169 			    &ndp->ni_cnd, vap);
170 			if (error == 0)
171 #endif
172 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
173 						   &ndp->ni_cnd, vap);
174 			vput(ndp->ni_dvp);
175 			vn_finished_write(mp);
176 			if (error) {
177 				NDFREE(ndp, NDF_ONLY_PNBUF);
178 				return (error);
179 			}
180 			fmode &= ~O_TRUNC;
181 			vp = ndp->ni_vp;
182 		} else {
183 			if (ndp->ni_dvp == ndp->ni_vp)
184 				vrele(ndp->ni_dvp);
185 			else
186 				vput(ndp->ni_dvp);
187 			ndp->ni_dvp = NULL;
188 			vp = ndp->ni_vp;
189 			if (fmode & O_EXCL) {
190 				error = EEXIST;
191 				goto bad;
192 			}
193 			fmode &= ~O_CREAT;
194 		}
195 	} else {
196 		ndp->ni_cnd.cn_nameiop = LOOKUP;
197 		ndp->ni_cnd.cn_flags = ISOPEN |
198 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
199 		if (!(fmode & FWRITE))
200 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
201 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
202 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
203 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
204 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
205 		if ((error = namei(ndp)) != 0)
206 			return (error);
207 		vp = ndp->ni_vp;
208 	}
209 	error = vn_open_vnode(vp, fmode, cred, td, fp);
210 	if (error)
211 		goto bad;
212 	*flagp = fmode;
213 	return (0);
214 bad:
215 	NDFREE(ndp, NDF_ONLY_PNBUF);
216 	vput(vp);
217 	*flagp = fmode;
218 	ndp->ni_vp = NULL;
219 	return (error);
220 }
221 
222 /*
223  * Common code for vnode open operations once a vnode is located.
224  * Check permissions, and call the VOP_OPEN routine.
225  */
226 int
227 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
228     struct thread *td, struct file *fp)
229 {
230 	struct mount *mp;
231 	accmode_t accmode;
232 	struct flock lf;
233 	int error, have_flock, lock_flags, type;
234 
235 	if (vp->v_type == VLNK)
236 		return (EMLINK);
237 	if (vp->v_type == VSOCK)
238 		return (EOPNOTSUPP);
239 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
240 		return (ENOTDIR);
241 	accmode = 0;
242 	if (fmode & (FWRITE | O_TRUNC)) {
243 		if (vp->v_type == VDIR)
244 			return (EISDIR);
245 		accmode |= VWRITE;
246 	}
247 	if (fmode & FREAD)
248 		accmode |= VREAD;
249 	if (fmode & FEXEC)
250 		accmode |= VEXEC;
251 	if ((fmode & O_APPEND) && (fmode & FWRITE))
252 		accmode |= VAPPEND;
253 #ifdef MAC
254 	error = mac_vnode_check_open(cred, vp, accmode);
255 	if (error)
256 		return (error);
257 #endif
258 	if ((fmode & O_CREAT) == 0) {
259 		if (accmode & VWRITE) {
260 			error = vn_writechk(vp);
261 			if (error)
262 				return (error);
263 		}
264 		if (accmode) {
265 		        error = VOP_ACCESS(vp, accmode, cred, td);
266 			if (error)
267 				return (error);
268 		}
269 	}
270 	if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
271 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
272 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
273 		return (error);
274 
275 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
276 		KASSERT(fp != NULL, ("open with flock requires fp"));
277 		lock_flags = VOP_ISLOCKED(vp);
278 		VOP_UNLOCK(vp, 0);
279 		lf.l_whence = SEEK_SET;
280 		lf.l_start = 0;
281 		lf.l_len = 0;
282 		if (fmode & O_EXLOCK)
283 			lf.l_type = F_WRLCK;
284 		else
285 			lf.l_type = F_RDLCK;
286 		type = F_FLOCK;
287 		if ((fmode & FNONBLOCK) == 0)
288 			type |= F_WAIT;
289 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
290 		have_flock = (error == 0);
291 		vn_lock(vp, lock_flags | LK_RETRY);
292 		if (error == 0 && vp->v_iflag & VI_DOOMED)
293 			error = ENOENT;
294 		/*
295 		 * Another thread might have used this vnode as an
296 		 * executable while the vnode lock was dropped.
297 		 * Ensure the vnode is still able to be opened for
298 		 * writing after the lock has been obtained.
299 		 */
300 		if (error == 0 && accmode & VWRITE)
301 			error = vn_writechk(vp);
302 		if (error) {
303 			VOP_UNLOCK(vp, 0);
304 			if (have_flock) {
305 				lf.l_whence = SEEK_SET;
306 				lf.l_start = 0;
307 				lf.l_len = 0;
308 				lf.l_type = F_UNLCK;
309 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
310 				    F_FLOCK);
311 			}
312 			vn_start_write(vp, &mp, V_WAIT);
313 			vn_lock(vp, lock_flags | LK_RETRY);
314 			(void)VOP_CLOSE(vp, fmode, cred, td);
315 			vn_finished_write(mp);
316 			/* Prevent second close from fdrop()->vn_close(). */
317 			if (fp != NULL)
318 				fp->f_ops= &badfileops;
319 			return (error);
320 		}
321 		fp->f_flag |= FHASLOCK;
322 	}
323 	if (fmode & FWRITE) {
324 		VOP_ADD_WRITECOUNT(vp, 1);
325 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
326 		    __func__, vp, vp->v_writecount);
327 	}
328 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
329 	return (0);
330 }
331 
332 /*
333  * Check for write permissions on the specified vnode.
334  * Prototype text segments cannot be written.
335  */
336 int
337 vn_writechk(vp)
338 	register struct vnode *vp;
339 {
340 
341 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
342 	/*
343 	 * If there's shared text associated with
344 	 * the vnode, try to free it up once.  If
345 	 * we fail, we can't allow writing.
346 	 */
347 	if (VOP_IS_TEXT(vp))
348 		return (ETXTBSY);
349 
350 	return (0);
351 }
352 
353 /*
354  * Vnode close call
355  */
356 int
357 vn_close(vp, flags, file_cred, td)
358 	register struct vnode *vp;
359 	int flags;
360 	struct ucred *file_cred;
361 	struct thread *td;
362 {
363 	struct mount *mp;
364 	int error, lock_flags;
365 
366 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
367 	    MNT_EXTENDED_SHARED(vp->v_mount))
368 		lock_flags = LK_SHARED;
369 	else
370 		lock_flags = LK_EXCLUSIVE;
371 
372 	vn_start_write(vp, &mp, V_WAIT);
373 	vn_lock(vp, lock_flags | LK_RETRY);
374 	if (flags & FWRITE) {
375 		VNASSERT(vp->v_writecount > 0, vp,
376 		    ("vn_close: negative writecount"));
377 		VOP_ADD_WRITECOUNT(vp, -1);
378 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
379 		    __func__, vp, vp->v_writecount);
380 	}
381 	error = VOP_CLOSE(vp, flags, file_cred, td);
382 	vput(vp);
383 	vn_finished_write(mp);
384 	return (error);
385 }
386 
387 /*
388  * Heuristic to detect sequential operation.
389  */
390 static int
391 sequential_heuristic(struct uio *uio, struct file *fp)
392 {
393 
394 	if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
395 		return (fp->f_seqcount << IO_SEQSHIFT);
396 
397 	/*
398 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
399 	 * that the first I/O is normally considered to be slightly
400 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
401 	 * unless previous seeks have reduced f_seqcount to 0, in which
402 	 * case offset 0 is not special.
403 	 */
404 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
405 	    uio->uio_offset == fp->f_nextoff) {
406 		/*
407 		 * f_seqcount is in units of fixed-size blocks so that it
408 		 * depends mainly on the amount of sequential I/O and not
409 		 * much on the number of sequential I/O's.  The fixed size
410 		 * of 16384 is hard-coded here since it is (not quite) just
411 		 * a magic size that works well here.  This size is more
412 		 * closely related to the best I/O size for real disks than
413 		 * to any block size used by software.
414 		 */
415 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
416 		if (fp->f_seqcount > IO_SEQMAX)
417 			fp->f_seqcount = IO_SEQMAX;
418 		return (fp->f_seqcount << IO_SEQSHIFT);
419 	}
420 
421 	/* Not sequential.  Quickly draw-down sequentiality. */
422 	if (fp->f_seqcount > 1)
423 		fp->f_seqcount = 1;
424 	else
425 		fp->f_seqcount = 0;
426 	return (0);
427 }
428 
429 /*
430  * Package up an I/O request on a vnode into a uio and do it.
431  */
432 int
433 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
434     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
435     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
436 {
437 	struct uio auio;
438 	struct iovec aiov;
439 	struct mount *mp;
440 	struct ucred *cred;
441 	void *rl_cookie;
442 	int error, lock_flags;
443 
444 	auio.uio_iov = &aiov;
445 	auio.uio_iovcnt = 1;
446 	aiov.iov_base = base;
447 	aiov.iov_len = len;
448 	auio.uio_resid = len;
449 	auio.uio_offset = offset;
450 	auio.uio_segflg = segflg;
451 	auio.uio_rw = rw;
452 	auio.uio_td = td;
453 	error = 0;
454 
455 	if ((ioflg & IO_NODELOCKED) == 0) {
456 		if (rw == UIO_READ) {
457 			rl_cookie = vn_rangelock_rlock(vp, offset,
458 			    offset + len);
459 		} else {
460 			rl_cookie = vn_rangelock_wlock(vp, offset,
461 			    offset + len);
462 		}
463 		mp = NULL;
464 		if (rw == UIO_WRITE) {
465 			if (vp->v_type != VCHR &&
466 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
467 			    != 0)
468 				goto out;
469 			if (MNT_SHARED_WRITES(mp) ||
470 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
471 				lock_flags = LK_SHARED;
472 			else
473 				lock_flags = LK_EXCLUSIVE;
474 		} else
475 			lock_flags = LK_SHARED;
476 		vn_lock(vp, lock_flags | LK_RETRY);
477 	} else
478 		rl_cookie = NULL;
479 
480 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
481 #ifdef MAC
482 	if ((ioflg & IO_NOMACCHECK) == 0) {
483 		if (rw == UIO_READ)
484 			error = mac_vnode_check_read(active_cred, file_cred,
485 			    vp);
486 		else
487 			error = mac_vnode_check_write(active_cred, file_cred,
488 			    vp);
489 	}
490 #endif
491 	if (error == 0) {
492 		if (file_cred != NULL)
493 			cred = file_cred;
494 		else
495 			cred = active_cred;
496 		if (rw == UIO_READ)
497 			error = VOP_READ(vp, &auio, ioflg, cred);
498 		else
499 			error = VOP_WRITE(vp, &auio, ioflg, cred);
500 	}
501 	if (aresid)
502 		*aresid = auio.uio_resid;
503 	else
504 		if (auio.uio_resid && error == 0)
505 			error = EIO;
506 	if ((ioflg & IO_NODELOCKED) == 0) {
507 		VOP_UNLOCK(vp, 0);
508 		if (mp != NULL)
509 			vn_finished_write(mp);
510 	}
511  out:
512 	if (rl_cookie != NULL)
513 		vn_rangelock_unlock(vp, rl_cookie);
514 	return (error);
515 }
516 
517 /*
518  * Package up an I/O request on a vnode into a uio and do it.  The I/O
519  * request is split up into smaller chunks and we try to avoid saturating
520  * the buffer cache while potentially holding a vnode locked, so we
521  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
522  * to give other processes a chance to lock the vnode (either other processes
523  * core'ing the same binary, or unrelated processes scanning the directory).
524  */
525 int
526 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
527     file_cred, aresid, td)
528 	enum uio_rw rw;
529 	struct vnode *vp;
530 	void *base;
531 	size_t len;
532 	off_t offset;
533 	enum uio_seg segflg;
534 	int ioflg;
535 	struct ucred *active_cred;
536 	struct ucred *file_cred;
537 	size_t *aresid;
538 	struct thread *td;
539 {
540 	int error = 0;
541 	ssize_t iaresid;
542 
543 	do {
544 		int chunk;
545 
546 		/*
547 		 * Force `offset' to a multiple of MAXBSIZE except possibly
548 		 * for the first chunk, so that filesystems only need to
549 		 * write full blocks except possibly for the first and last
550 		 * chunks.
551 		 */
552 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
553 
554 		if (chunk > len)
555 			chunk = len;
556 		if (rw != UIO_READ && vp->v_type == VREG)
557 			bwillwrite();
558 		iaresid = 0;
559 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
560 		    ioflg, active_cred, file_cred, &iaresid, td);
561 		len -= chunk;	/* aresid calc already includes length */
562 		if (error)
563 			break;
564 		offset += chunk;
565 		base = (char *)base + chunk;
566 		kern_yield(PRI_USER);
567 	} while (len);
568 	if (aresid)
569 		*aresid = len + iaresid;
570 	return (error);
571 }
572 
573 off_t
574 foffset_lock(struct file *fp, int flags)
575 {
576 	struct mtx *mtxp;
577 	off_t res;
578 
579 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
580 
581 #if OFF_MAX <= LONG_MAX
582 	/*
583 	 * Caller only wants the current f_offset value.  Assume that
584 	 * the long and shorter integer types reads are atomic.
585 	 */
586 	if ((flags & FOF_NOLOCK) != 0)
587 		return (fp->f_offset);
588 #endif
589 
590 	/*
591 	 * According to McKusick the vn lock was protecting f_offset here.
592 	 * It is now protected by the FOFFSET_LOCKED flag.
593 	 */
594 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
595 	mtx_lock(mtxp);
596 	if ((flags & FOF_NOLOCK) == 0) {
597 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
598 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
599 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
600 			    "vofflock", 0);
601 		}
602 		fp->f_vnread_flags |= FOFFSET_LOCKED;
603 	}
604 	res = fp->f_offset;
605 	mtx_unlock(mtxp);
606 	return (res);
607 }
608 
609 void
610 foffset_unlock(struct file *fp, off_t val, int flags)
611 {
612 	struct mtx *mtxp;
613 
614 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
615 
616 #if OFF_MAX <= LONG_MAX
617 	if ((flags & FOF_NOLOCK) != 0) {
618 		if ((flags & FOF_NOUPDATE) == 0)
619 			fp->f_offset = val;
620 		if ((flags & FOF_NEXTOFF) != 0)
621 			fp->f_nextoff = val;
622 		return;
623 	}
624 #endif
625 
626 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
627 	mtx_lock(mtxp);
628 	if ((flags & FOF_NOUPDATE) == 0)
629 		fp->f_offset = val;
630 	if ((flags & FOF_NEXTOFF) != 0)
631 		fp->f_nextoff = val;
632 	if ((flags & FOF_NOLOCK) == 0) {
633 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
634 		    ("Lost FOFFSET_LOCKED"));
635 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
636 			wakeup(&fp->f_vnread_flags);
637 		fp->f_vnread_flags = 0;
638 	}
639 	mtx_unlock(mtxp);
640 }
641 
642 void
643 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
644 {
645 
646 	if ((flags & FOF_OFFSET) == 0)
647 		uio->uio_offset = foffset_lock(fp, flags);
648 }
649 
650 void
651 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
652 {
653 
654 	if ((flags & FOF_OFFSET) == 0)
655 		foffset_unlock(fp, uio->uio_offset, flags);
656 }
657 
658 static int
659 get_advice(struct file *fp, struct uio *uio)
660 {
661 	struct mtx *mtxp;
662 	int ret;
663 
664 	ret = POSIX_FADV_NORMAL;
665 	if (fp->f_advice == NULL)
666 		return (ret);
667 
668 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
669 	mtx_lock(mtxp);
670 	if (uio->uio_offset >= fp->f_advice->fa_start &&
671 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
672 		ret = fp->f_advice->fa_advice;
673 	mtx_unlock(mtxp);
674 	return (ret);
675 }
676 
677 /*
678  * File table vnode read routine.
679  */
680 static int
681 vn_read(fp, uio, active_cred, flags, td)
682 	struct file *fp;
683 	struct uio *uio;
684 	struct ucred *active_cred;
685 	int flags;
686 	struct thread *td;
687 {
688 	struct vnode *vp;
689 	struct mtx *mtxp;
690 	int error, ioflag;
691 	int advice;
692 	off_t offset, start, end;
693 
694 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
695 	    uio->uio_td, td));
696 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
697 	vp = fp->f_vnode;
698 	ioflag = 0;
699 	if (fp->f_flag & FNONBLOCK)
700 		ioflag |= IO_NDELAY;
701 	if (fp->f_flag & O_DIRECT)
702 		ioflag |= IO_DIRECT;
703 	advice = get_advice(fp, uio);
704 	vn_lock(vp, LK_SHARED | LK_RETRY);
705 
706 	switch (advice) {
707 	case POSIX_FADV_NORMAL:
708 	case POSIX_FADV_SEQUENTIAL:
709 	case POSIX_FADV_NOREUSE:
710 		ioflag |= sequential_heuristic(uio, fp);
711 		break;
712 	case POSIX_FADV_RANDOM:
713 		/* Disable read-ahead for random I/O. */
714 		break;
715 	}
716 	offset = uio->uio_offset;
717 
718 #ifdef MAC
719 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
720 	if (error == 0)
721 #endif
722 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
723 	fp->f_nextoff = uio->uio_offset;
724 	VOP_UNLOCK(vp, 0);
725 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
726 	    offset != uio->uio_offset) {
727 		/*
728 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
729 		 * buffers for the backing file after a
730 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
731 		 * case of using POSIX_FADV_NOREUSE with sequential
732 		 * access, track the previous implicit DONTNEED
733 		 * request and grow this request to include the
734 		 * current read(2) in addition to the previous
735 		 * DONTNEED.  With purely sequential access this will
736 		 * cause the DONTNEED requests to continously grow to
737 		 * cover all of the previously read regions of the
738 		 * file.  This allows filesystem blocks that are
739 		 * accessed by multiple calls to read(2) to be flushed
740 		 * once the last read(2) finishes.
741 		 */
742 		start = offset;
743 		end = uio->uio_offset - 1;
744 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
745 		mtx_lock(mtxp);
746 		if (fp->f_advice != NULL &&
747 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
748 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
749 				start = fp->f_advice->fa_prevstart;
750 			else if (fp->f_advice->fa_prevstart != 0 &&
751 			    fp->f_advice->fa_prevstart == end + 1)
752 				end = fp->f_advice->fa_prevend;
753 			fp->f_advice->fa_prevstart = start;
754 			fp->f_advice->fa_prevend = end;
755 		}
756 		mtx_unlock(mtxp);
757 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
758 	}
759 	return (error);
760 }
761 
762 /*
763  * File table vnode write routine.
764  */
765 static int
766 vn_write(fp, uio, active_cred, flags, td)
767 	struct file *fp;
768 	struct uio *uio;
769 	struct ucred *active_cred;
770 	int flags;
771 	struct thread *td;
772 {
773 	struct vnode *vp;
774 	struct mount *mp;
775 	struct mtx *mtxp;
776 	int error, ioflag, lock_flags;
777 	int advice;
778 	off_t offset, start, end;
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 	if (vp->v_type == VREG)
785 		bwillwrite();
786 	ioflag = IO_UNIT;
787 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
788 		ioflag |= IO_APPEND;
789 	if (fp->f_flag & FNONBLOCK)
790 		ioflag |= IO_NDELAY;
791 	if (fp->f_flag & O_DIRECT)
792 		ioflag |= IO_DIRECT;
793 	if ((fp->f_flag & O_FSYNC) ||
794 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
795 		ioflag |= IO_SYNC;
796 	mp = NULL;
797 	if (vp->v_type != VCHR &&
798 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
799 		goto unlock;
800 
801 	advice = get_advice(fp, uio);
802 
803 	if (MNT_SHARED_WRITES(mp) ||
804 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
805 		lock_flags = LK_SHARED;
806 	} else {
807 		lock_flags = LK_EXCLUSIVE;
808 	}
809 
810 	vn_lock(vp, lock_flags | LK_RETRY);
811 	switch (advice) {
812 	case POSIX_FADV_NORMAL:
813 	case POSIX_FADV_SEQUENTIAL:
814 	case POSIX_FADV_NOREUSE:
815 		ioflag |= sequential_heuristic(uio, fp);
816 		break;
817 	case POSIX_FADV_RANDOM:
818 		/* XXX: Is this correct? */
819 		break;
820 	}
821 	offset = uio->uio_offset;
822 
823 #ifdef MAC
824 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
825 	if (error == 0)
826 #endif
827 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
828 	fp->f_nextoff = uio->uio_offset;
829 	VOP_UNLOCK(vp, 0);
830 	if (vp->v_type != VCHR)
831 		vn_finished_write(mp);
832 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
833 	    offset != uio->uio_offset) {
834 		/*
835 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
836 		 * buffers for the backing file after a
837 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
838 		 * common case of using POSIX_FADV_NOREUSE with
839 		 * sequential access, track the previous implicit
840 		 * DONTNEED request and grow this request to include
841 		 * the current write(2) in addition to the previous
842 		 * DONTNEED.  With purely sequential access this will
843 		 * cause the DONTNEED requests to continously grow to
844 		 * cover all of the previously written regions of the
845 		 * file.
846 		 *
847 		 * Note that the blocks just written are almost
848 		 * certainly still dirty, so this only works when
849 		 * VOP_ADVISE() calls from subsequent writes push out
850 		 * the data written by this write(2) once the backing
851 		 * buffers are clean.  However, as compared to forcing
852 		 * IO_DIRECT, this gives much saner behavior.  Write
853 		 * clustering is still allowed, and clean pages are
854 		 * merely moved to the cache page queue rather than
855 		 * outright thrown away.  This means a subsequent
856 		 * read(2) can still avoid hitting the disk if the
857 		 * pages have not been reclaimed.
858 		 *
859 		 * This does make POSIX_FADV_NOREUSE largely useless
860 		 * with non-sequential access.  However, sequential
861 		 * access is the more common use case and the flag is
862 		 * merely advisory.
863 		 */
864 		start = offset;
865 		end = uio->uio_offset - 1;
866 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
867 		mtx_lock(mtxp);
868 		if (fp->f_advice != NULL &&
869 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
870 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
871 				start = fp->f_advice->fa_prevstart;
872 			else if (fp->f_advice->fa_prevstart != 0 &&
873 			    fp->f_advice->fa_prevstart == end + 1)
874 				end = fp->f_advice->fa_prevend;
875 			fp->f_advice->fa_prevstart = start;
876 			fp->f_advice->fa_prevend = end;
877 		}
878 		mtx_unlock(mtxp);
879 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
880 	}
881 
882 unlock:
883 	return (error);
884 }
885 
886 static const int io_hold_cnt = 16;
887 static int vn_io_fault_enable = 1;
888 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
889     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
890 static u_long vn_io_faults_cnt;
891 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
892     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
893 
894 /*
895  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
896  * prevent the following deadlock:
897  *
898  * Assume that the thread A reads from the vnode vp1 into userspace
899  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
900  * currently not resident, then system ends up with the call chain
901  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
902  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
903  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
904  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
905  * backed by the pages of vnode vp1, and some page in buf2 is not
906  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
907  *
908  * To prevent the lock order reversal and deadlock, vn_io_fault() does
909  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
910  * Instead, it first tries to do the whole range i/o with pagefaults
911  * disabled. If all pages in the i/o buffer are resident and mapped,
912  * VOP will succeed (ignoring the genuine filesystem errors).
913  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
914  * i/o in chunks, with all pages in the chunk prefaulted and held
915  * using vm_fault_quick_hold_pages().
916  *
917  * Filesystems using this deadlock avoidance scheme should use the
918  * array of the held pages from uio, saved in the curthread->td_ma,
919  * instead of doing uiomove().  A helper function
920  * vn_io_fault_uiomove() converts uiomove request into
921  * uiomove_fromphys() over td_ma array.
922  *
923  * Since vnode locks do not cover the whole i/o anymore, rangelocks
924  * make the current i/o request atomic with respect to other i/os and
925  * truncations.
926  */
927 static int
928 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
929     int flags, struct thread *td)
930 {
931 	vm_page_t ma[io_hold_cnt + 2];
932 	struct uio *uio_clone, short_uio;
933 	struct iovec short_iovec[1];
934 	fo_rdwr_t *doio;
935 	struct vnode *vp;
936 	void *rl_cookie;
937 	struct mount *mp;
938 	vm_page_t *prev_td_ma;
939 	int error, cnt, save, saveheld, prev_td_ma_cnt;
940 	vm_offset_t addr, end;
941 	vm_prot_t prot;
942 	size_t len, resid;
943 	ssize_t adv;
944 
945 	if (uio->uio_rw == UIO_READ)
946 		doio = vn_read;
947 	else
948 		doio = vn_write;
949 	vp = fp->f_vnode;
950 	foffset_lock_uio(fp, uio, flags);
951 
952 	if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
953 	    ((mp = vp->v_mount) != NULL &&
954 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
955 	    !vn_io_fault_enable) {
956 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
957 		goto out_last;
958 	}
959 
960 	/*
961 	 * The UFS follows IO_UNIT directive and replays back both
962 	 * uio_offset and uio_resid if an error is encountered during the
963 	 * operation.  But, since the iovec may be already advanced,
964 	 * uio is still in an inconsistent state.
965 	 *
966 	 * Cache a copy of the original uio, which is advanced to the redo
967 	 * point using UIO_NOCOPY below.
968 	 */
969 	uio_clone = cloneuio(uio);
970 	resid = uio->uio_resid;
971 
972 	short_uio.uio_segflg = UIO_USERSPACE;
973 	short_uio.uio_rw = uio->uio_rw;
974 	short_uio.uio_td = uio->uio_td;
975 
976 	if (uio->uio_rw == UIO_READ) {
977 		prot = VM_PROT_WRITE;
978 		rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
979 		    uio->uio_offset + uio->uio_resid);
980 	} else {
981 		prot = VM_PROT_READ;
982 		if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
983 			/* For appenders, punt and lock the whole range. */
984 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
985 		else
986 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
987 			    uio->uio_offset + uio->uio_resid);
988 	}
989 
990 	save = vm_fault_disable_pagefaults();
991 	error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
992 	if (error != EFAULT)
993 		goto out;
994 
995 	atomic_add_long(&vn_io_faults_cnt, 1);
996 	uio_clone->uio_segflg = UIO_NOCOPY;
997 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
998 	uio_clone->uio_segflg = uio->uio_segflg;
999 
1000 	saveheld = curthread_pflags_set(TDP_UIOHELD);
1001 	prev_td_ma = td->td_ma;
1002 	prev_td_ma_cnt = td->td_ma_cnt;
1003 
1004 	while (uio_clone->uio_resid != 0) {
1005 		len = uio_clone->uio_iov->iov_len;
1006 		if (len == 0) {
1007 			KASSERT(uio_clone->uio_iovcnt >= 1,
1008 			    ("iovcnt underflow"));
1009 			uio_clone->uio_iov++;
1010 			uio_clone->uio_iovcnt--;
1011 			continue;
1012 		}
1013 		if (len > io_hold_cnt * PAGE_SIZE)
1014 			len = io_hold_cnt * PAGE_SIZE;
1015 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1016 		end = round_page(addr + len);
1017 		if (end < addr) {
1018 			error = EFAULT;
1019 			break;
1020 		}
1021 		cnt = atop(end - trunc_page(addr));
1022 		/*
1023 		 * A perfectly misaligned address and length could cause
1024 		 * both the start and the end of the chunk to use partial
1025 		 * page.  +2 accounts for such a situation.
1026 		 */
1027 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1028 		    addr, len, prot, ma, io_hold_cnt + 2);
1029 		if (cnt == -1) {
1030 			error = EFAULT;
1031 			break;
1032 		}
1033 		short_uio.uio_iov = &short_iovec[0];
1034 		short_iovec[0].iov_base = (void *)addr;
1035 		short_uio.uio_iovcnt = 1;
1036 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1037 		short_uio.uio_offset = uio_clone->uio_offset;
1038 		td->td_ma = ma;
1039 		td->td_ma_cnt = cnt;
1040 
1041 		error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
1042 		    td);
1043 		vm_page_unhold_pages(ma, cnt);
1044 		adv = len - short_uio.uio_resid;
1045 
1046 		uio_clone->uio_iov->iov_base =
1047 		    (char *)uio_clone->uio_iov->iov_base + adv;
1048 		uio_clone->uio_iov->iov_len -= adv;
1049 		uio_clone->uio_resid -= adv;
1050 		uio_clone->uio_offset += adv;
1051 
1052 		uio->uio_resid -= adv;
1053 		uio->uio_offset += adv;
1054 
1055 		if (error != 0 || adv == 0)
1056 			break;
1057 	}
1058 	td->td_ma = prev_td_ma;
1059 	td->td_ma_cnt = prev_td_ma_cnt;
1060 	curthread_pflags_restore(saveheld);
1061 out:
1062 	vm_fault_enable_pagefaults(save);
1063 	vn_rangelock_unlock(vp, rl_cookie);
1064 	free(uio_clone, M_IOV);
1065 out_last:
1066 	foffset_unlock_uio(fp, uio, flags);
1067 	return (error);
1068 }
1069 
1070 /*
1071  * Helper function to perform the requested uiomove operation using
1072  * the held pages for io->uio_iov[0].iov_base buffer instead of
1073  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1074  * instead of iov_base prevents page faults that could occur due to
1075  * pmap_collect() invalidating the mapping created by
1076  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1077  * object cleanup revoking the write access from page mappings.
1078  *
1079  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1080  * instead of plain uiomove().
1081  */
1082 int
1083 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1084 {
1085 	struct uio transp_uio;
1086 	struct iovec transp_iov[1];
1087 	struct thread *td;
1088 	size_t adv;
1089 	int error, pgadv;
1090 
1091 	td = curthread;
1092 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1093 	    uio->uio_segflg != UIO_USERSPACE)
1094 		return (uiomove(data, xfersize, uio));
1095 
1096 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1097 	transp_iov[0].iov_base = data;
1098 	transp_uio.uio_iov = &transp_iov[0];
1099 	transp_uio.uio_iovcnt = 1;
1100 	if (xfersize > uio->uio_resid)
1101 		xfersize = uio->uio_resid;
1102 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1103 	transp_uio.uio_offset = 0;
1104 	transp_uio.uio_segflg = UIO_SYSSPACE;
1105 	/*
1106 	 * Since transp_iov points to data, and td_ma page array
1107 	 * corresponds to original uio->uio_iov, we need to invert the
1108 	 * direction of the i/o operation as passed to
1109 	 * uiomove_fromphys().
1110 	 */
1111 	switch (uio->uio_rw) {
1112 	case UIO_WRITE:
1113 		transp_uio.uio_rw = UIO_READ;
1114 		break;
1115 	case UIO_READ:
1116 		transp_uio.uio_rw = UIO_WRITE;
1117 		break;
1118 	}
1119 	transp_uio.uio_td = uio->uio_td;
1120 	error = uiomove_fromphys(td->td_ma,
1121 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1122 	    xfersize, &transp_uio);
1123 	adv = xfersize - transp_uio.uio_resid;
1124 	pgadv =
1125 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1126 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1127 	td->td_ma += pgadv;
1128 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1129 	    pgadv));
1130 	td->td_ma_cnt -= pgadv;
1131 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1132 	uio->uio_iov->iov_len -= adv;
1133 	uio->uio_resid -= adv;
1134 	uio->uio_offset += adv;
1135 	return (error);
1136 }
1137 
1138 int
1139 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1140     struct uio *uio)
1141 {
1142 	struct thread *td;
1143 	vm_offset_t iov_base;
1144 	int cnt, pgadv;
1145 
1146 	td = curthread;
1147 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1148 	    uio->uio_segflg != UIO_USERSPACE)
1149 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1150 
1151 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1152 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1153 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1154 	switch (uio->uio_rw) {
1155 	case UIO_WRITE:
1156 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1157 		    offset, cnt);
1158 		break;
1159 	case UIO_READ:
1160 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1161 		    cnt);
1162 		break;
1163 	}
1164 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1165 	td->td_ma += pgadv;
1166 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1167 	    pgadv));
1168 	td->td_ma_cnt -= pgadv;
1169 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1170 	uio->uio_iov->iov_len -= cnt;
1171 	uio->uio_resid -= cnt;
1172 	uio->uio_offset += cnt;
1173 	return (0);
1174 }
1175 
1176 
1177 /*
1178  * File table truncate routine.
1179  */
1180 static int
1181 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1182     struct thread *td)
1183 {
1184 	struct vattr vattr;
1185 	struct mount *mp;
1186 	struct vnode *vp;
1187 	void *rl_cookie;
1188 	int error;
1189 
1190 	vp = fp->f_vnode;
1191 
1192 	/*
1193 	 * Lock the whole range for truncation.  Otherwise split i/o
1194 	 * might happen partly before and partly after the truncation.
1195 	 */
1196 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1197 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1198 	if (error)
1199 		goto out1;
1200 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1201 	if (vp->v_type == VDIR) {
1202 		error = EISDIR;
1203 		goto out;
1204 	}
1205 #ifdef MAC
1206 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1207 	if (error)
1208 		goto out;
1209 #endif
1210 	error = vn_writechk(vp);
1211 	if (error == 0) {
1212 		VATTR_NULL(&vattr);
1213 		vattr.va_size = length;
1214 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1215 	}
1216 out:
1217 	VOP_UNLOCK(vp, 0);
1218 	vn_finished_write(mp);
1219 out1:
1220 	vn_rangelock_unlock(vp, rl_cookie);
1221 	return (error);
1222 }
1223 
1224 /*
1225  * File table vnode stat routine.
1226  */
1227 static int
1228 vn_statfile(fp, sb, active_cred, td)
1229 	struct file *fp;
1230 	struct stat *sb;
1231 	struct ucred *active_cred;
1232 	struct thread *td;
1233 {
1234 	struct vnode *vp = fp->f_vnode;
1235 	int error;
1236 
1237 	vn_lock(vp, LK_SHARED | LK_RETRY);
1238 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1239 	VOP_UNLOCK(vp, 0);
1240 
1241 	return (error);
1242 }
1243 
1244 /*
1245  * Stat a vnode; implementation for the stat syscall
1246  */
1247 int
1248 vn_stat(vp, sb, active_cred, file_cred, td)
1249 	struct vnode *vp;
1250 	register struct stat *sb;
1251 	struct ucred *active_cred;
1252 	struct ucred *file_cred;
1253 	struct thread *td;
1254 {
1255 	struct vattr vattr;
1256 	register struct vattr *vap;
1257 	int error;
1258 	u_short mode;
1259 
1260 #ifdef MAC
1261 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1262 	if (error)
1263 		return (error);
1264 #endif
1265 
1266 	vap = &vattr;
1267 
1268 	/*
1269 	 * Initialize defaults for new and unusual fields, so that file
1270 	 * systems which don't support these fields don't need to know
1271 	 * about them.
1272 	 */
1273 	vap->va_birthtime.tv_sec = -1;
1274 	vap->va_birthtime.tv_nsec = 0;
1275 	vap->va_fsid = VNOVAL;
1276 	vap->va_rdev = NODEV;
1277 
1278 	error = VOP_GETATTR(vp, vap, active_cred);
1279 	if (error)
1280 		return (error);
1281 
1282 	/*
1283 	 * Zero the spare stat fields
1284 	 */
1285 	bzero(sb, sizeof *sb);
1286 
1287 	/*
1288 	 * Copy from vattr table
1289 	 */
1290 	if (vap->va_fsid != VNOVAL)
1291 		sb->st_dev = vap->va_fsid;
1292 	else
1293 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1294 	sb->st_ino = vap->va_fileid;
1295 	mode = vap->va_mode;
1296 	switch (vap->va_type) {
1297 	case VREG:
1298 		mode |= S_IFREG;
1299 		break;
1300 	case VDIR:
1301 		mode |= S_IFDIR;
1302 		break;
1303 	case VBLK:
1304 		mode |= S_IFBLK;
1305 		break;
1306 	case VCHR:
1307 		mode |= S_IFCHR;
1308 		break;
1309 	case VLNK:
1310 		mode |= S_IFLNK;
1311 		break;
1312 	case VSOCK:
1313 		mode |= S_IFSOCK;
1314 		break;
1315 	case VFIFO:
1316 		mode |= S_IFIFO;
1317 		break;
1318 	default:
1319 		return (EBADF);
1320 	};
1321 	sb->st_mode = mode;
1322 	sb->st_nlink = vap->va_nlink;
1323 	sb->st_uid = vap->va_uid;
1324 	sb->st_gid = vap->va_gid;
1325 	sb->st_rdev = vap->va_rdev;
1326 	if (vap->va_size > OFF_MAX)
1327 		return (EOVERFLOW);
1328 	sb->st_size = vap->va_size;
1329 	sb->st_atim = vap->va_atime;
1330 	sb->st_mtim = vap->va_mtime;
1331 	sb->st_ctim = vap->va_ctime;
1332 	sb->st_birthtim = vap->va_birthtime;
1333 
1334         /*
1335 	 * According to www.opengroup.org, the meaning of st_blksize is
1336 	 *   "a filesystem-specific preferred I/O block size for this
1337 	 *    object.  In some filesystem types, this may vary from file
1338 	 *    to file"
1339 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1340 	 */
1341 
1342 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1343 
1344 	sb->st_flags = vap->va_flags;
1345 	if (priv_check(td, PRIV_VFS_GENERATION))
1346 		sb->st_gen = 0;
1347 	else
1348 		sb->st_gen = vap->va_gen;
1349 
1350 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1351 	return (0);
1352 }
1353 
1354 /*
1355  * File table vnode ioctl routine.
1356  */
1357 static int
1358 vn_ioctl(fp, com, data, active_cred, td)
1359 	struct file *fp;
1360 	u_long com;
1361 	void *data;
1362 	struct ucred *active_cred;
1363 	struct thread *td;
1364 {
1365 	struct vattr vattr;
1366 	struct vnode *vp;
1367 	int error;
1368 
1369 	vp = fp->f_vnode;
1370 	switch (vp->v_type) {
1371 	case VDIR:
1372 	case VREG:
1373 		switch (com) {
1374 		case FIONREAD:
1375 			vn_lock(vp, LK_SHARED | LK_RETRY);
1376 			error = VOP_GETATTR(vp, &vattr, active_cred);
1377 			VOP_UNLOCK(vp, 0);
1378 			if (error == 0)
1379 				*(int *)data = vattr.va_size - fp->f_offset;
1380 			return (error);
1381 		case FIONBIO:
1382 		case FIOASYNC:
1383 			return (0);
1384 		default:
1385 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1386 			    active_cred, td));
1387 		}
1388 	default:
1389 		return (ENOTTY);
1390 	}
1391 }
1392 
1393 /*
1394  * File table vnode poll routine.
1395  */
1396 static int
1397 vn_poll(fp, events, active_cred, td)
1398 	struct file *fp;
1399 	int events;
1400 	struct ucred *active_cred;
1401 	struct thread *td;
1402 {
1403 	struct vnode *vp;
1404 	int error;
1405 
1406 	vp = fp->f_vnode;
1407 #ifdef MAC
1408 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1409 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1410 	VOP_UNLOCK(vp, 0);
1411 	if (!error)
1412 #endif
1413 
1414 	error = VOP_POLL(vp, events, fp->f_cred, td);
1415 	return (error);
1416 }
1417 
1418 /*
1419  * Acquire the requested lock and then check for validity.  LK_RETRY
1420  * permits vn_lock to return doomed vnodes.
1421  */
1422 int
1423 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1424 {
1425 	int error;
1426 
1427 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1428 	    ("vn_lock called with no locktype."));
1429 	do {
1430 #ifdef DEBUG_VFS_LOCKS
1431 		KASSERT(vp->v_holdcnt != 0,
1432 		    ("vn_lock %p: zero hold count", vp));
1433 #endif
1434 		error = VOP_LOCK1(vp, flags, file, line);
1435 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1436 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1437 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1438 		    flags, error));
1439 		/*
1440 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1441 		 * If RETRY is not set, we return ENOENT instead.
1442 		 */
1443 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1444 		    (flags & LK_RETRY) == 0) {
1445 			VOP_UNLOCK(vp, 0);
1446 			error = ENOENT;
1447 			break;
1448 		}
1449 	} while (flags & LK_RETRY && error != 0);
1450 	return (error);
1451 }
1452 
1453 /*
1454  * File table vnode close routine.
1455  */
1456 static int
1457 vn_closefile(fp, td)
1458 	struct file *fp;
1459 	struct thread *td;
1460 {
1461 	struct vnode *vp;
1462 	struct flock lf;
1463 	int error;
1464 
1465 	vp = fp->f_vnode;
1466 	fp->f_ops = &badfileops;
1467 
1468 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1469 		vref(vp);
1470 
1471 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1472 
1473 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1474 		lf.l_whence = SEEK_SET;
1475 		lf.l_start = 0;
1476 		lf.l_len = 0;
1477 		lf.l_type = F_UNLCK;
1478 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1479 		vrele(vp);
1480 	}
1481 	return (error);
1482 }
1483 
1484 /*
1485  * Preparing to start a filesystem write operation. If the operation is
1486  * permitted, then we bump the count of operations in progress and
1487  * proceed. If a suspend request is in progress, we wait until the
1488  * suspension is over, and then proceed.
1489  */
1490 static int
1491 vn_start_write_locked(struct mount *mp, int flags)
1492 {
1493 	int error;
1494 
1495 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1496 	error = 0;
1497 
1498 	/*
1499 	 * Check on status of suspension.
1500 	 */
1501 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1502 	    mp->mnt_susp_owner != curthread) {
1503 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1504 			if (flags & V_NOWAIT) {
1505 				error = EWOULDBLOCK;
1506 				goto unlock;
1507 			}
1508 			error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1509 			    (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1510 			if (error)
1511 				goto unlock;
1512 		}
1513 	}
1514 	if (flags & V_XSLEEP)
1515 		goto unlock;
1516 	mp->mnt_writeopcount++;
1517 unlock:
1518 	if (error != 0 || (flags & V_XSLEEP) != 0)
1519 		MNT_REL(mp);
1520 	MNT_IUNLOCK(mp);
1521 	return (error);
1522 }
1523 
1524 int
1525 vn_start_write(vp, mpp, flags)
1526 	struct vnode *vp;
1527 	struct mount **mpp;
1528 	int flags;
1529 {
1530 	struct mount *mp;
1531 	int error;
1532 
1533 	error = 0;
1534 	/*
1535 	 * If a vnode is provided, get and return the mount point that
1536 	 * to which it will write.
1537 	 */
1538 	if (vp != NULL) {
1539 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1540 			*mpp = NULL;
1541 			if (error != EOPNOTSUPP)
1542 				return (error);
1543 			return (0);
1544 		}
1545 	}
1546 	if ((mp = *mpp) == NULL)
1547 		return (0);
1548 
1549 	/*
1550 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1551 	 * a vfs_ref().
1552 	 * As long as a vnode is not provided we need to acquire a
1553 	 * refcount for the provided mountpoint too, in order to
1554 	 * emulate a vfs_ref().
1555 	 */
1556 	MNT_ILOCK(mp);
1557 	if (vp == NULL)
1558 		MNT_REF(mp);
1559 
1560 	return (vn_start_write_locked(mp, flags));
1561 }
1562 
1563 /*
1564  * Secondary suspension. Used by operations such as vop_inactive
1565  * routines that are needed by the higher level functions. These
1566  * are allowed to proceed until all the higher level functions have
1567  * completed (indicated by mnt_writeopcount dropping to zero). At that
1568  * time, these operations are halted until the suspension is over.
1569  */
1570 int
1571 vn_start_secondary_write(vp, mpp, flags)
1572 	struct vnode *vp;
1573 	struct mount **mpp;
1574 	int flags;
1575 {
1576 	struct mount *mp;
1577 	int error;
1578 
1579  retry:
1580 	if (vp != NULL) {
1581 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1582 			*mpp = NULL;
1583 			if (error != EOPNOTSUPP)
1584 				return (error);
1585 			return (0);
1586 		}
1587 	}
1588 	/*
1589 	 * If we are not suspended or have not yet reached suspended
1590 	 * mode, then let the operation proceed.
1591 	 */
1592 	if ((mp = *mpp) == NULL)
1593 		return (0);
1594 
1595 	/*
1596 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1597 	 * a vfs_ref().
1598 	 * As long as a vnode is not provided we need to acquire a
1599 	 * refcount for the provided mountpoint too, in order to
1600 	 * emulate a vfs_ref().
1601 	 */
1602 	MNT_ILOCK(mp);
1603 	if (vp == NULL)
1604 		MNT_REF(mp);
1605 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1606 		mp->mnt_secondary_writes++;
1607 		mp->mnt_secondary_accwrites++;
1608 		MNT_IUNLOCK(mp);
1609 		return (0);
1610 	}
1611 	if (flags & V_NOWAIT) {
1612 		MNT_REL(mp);
1613 		MNT_IUNLOCK(mp);
1614 		return (EWOULDBLOCK);
1615 	}
1616 	/*
1617 	 * Wait for the suspension to finish.
1618 	 */
1619 	error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1620 		       (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1621 	vfs_rel(mp);
1622 	if (error == 0)
1623 		goto retry;
1624 	return (error);
1625 }
1626 
1627 /*
1628  * Filesystem write operation has completed. If we are suspending and this
1629  * operation is the last one, notify the suspender that the suspension is
1630  * now in effect.
1631  */
1632 void
1633 vn_finished_write(mp)
1634 	struct mount *mp;
1635 {
1636 	if (mp == NULL)
1637 		return;
1638 	MNT_ILOCK(mp);
1639 	MNT_REL(mp);
1640 	mp->mnt_writeopcount--;
1641 	if (mp->mnt_writeopcount < 0)
1642 		panic("vn_finished_write: neg cnt");
1643 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1644 	    mp->mnt_writeopcount <= 0)
1645 		wakeup(&mp->mnt_writeopcount);
1646 	MNT_IUNLOCK(mp);
1647 }
1648 
1649 
1650 /*
1651  * Filesystem secondary write operation has completed. If we are
1652  * suspending and this operation is the last one, notify the suspender
1653  * that the suspension is now in effect.
1654  */
1655 void
1656 vn_finished_secondary_write(mp)
1657 	struct mount *mp;
1658 {
1659 	if (mp == NULL)
1660 		return;
1661 	MNT_ILOCK(mp);
1662 	MNT_REL(mp);
1663 	mp->mnt_secondary_writes--;
1664 	if (mp->mnt_secondary_writes < 0)
1665 		panic("vn_finished_secondary_write: neg cnt");
1666 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1667 	    mp->mnt_secondary_writes <= 0)
1668 		wakeup(&mp->mnt_secondary_writes);
1669 	MNT_IUNLOCK(mp);
1670 }
1671 
1672 
1673 
1674 /*
1675  * Request a filesystem to suspend write operations.
1676  */
1677 int
1678 vfs_write_suspend(struct mount *mp, int flags)
1679 {
1680 	int error;
1681 
1682 	MNT_ILOCK(mp);
1683 	if (mp->mnt_susp_owner == curthread) {
1684 		MNT_IUNLOCK(mp);
1685 		return (EALREADY);
1686 	}
1687 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1688 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1689 
1690 	/*
1691 	 * Unmount holds a write reference on the mount point.  If we
1692 	 * own busy reference and drain for writers, we deadlock with
1693 	 * the reference draining in the unmount path.  Callers of
1694 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1695 	 * vfs_busy() reference is owned and caller is not in the
1696 	 * unmount context.
1697 	 */
1698 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1699 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1700 		MNT_IUNLOCK(mp);
1701 		return (EBUSY);
1702 	}
1703 
1704 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1705 	mp->mnt_susp_owner = curthread;
1706 	if (mp->mnt_writeopcount > 0)
1707 		(void) msleep(&mp->mnt_writeopcount,
1708 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1709 	else
1710 		MNT_IUNLOCK(mp);
1711 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1712 		vfs_write_resume(mp, 0);
1713 	return (error);
1714 }
1715 
1716 /*
1717  * Request a filesystem to resume write operations.
1718  */
1719 void
1720 vfs_write_resume(struct mount *mp, int flags)
1721 {
1722 
1723 	MNT_ILOCK(mp);
1724 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1725 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1726 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1727 				       MNTK_SUSPENDED);
1728 		mp->mnt_susp_owner = NULL;
1729 		wakeup(&mp->mnt_writeopcount);
1730 		wakeup(&mp->mnt_flag);
1731 		curthread->td_pflags &= ~TDP_IGNSUSP;
1732 		if ((flags & VR_START_WRITE) != 0) {
1733 			MNT_REF(mp);
1734 			mp->mnt_writeopcount++;
1735 		}
1736 		MNT_IUNLOCK(mp);
1737 		if ((flags & VR_NO_SUSPCLR) == 0)
1738 			VFS_SUSP_CLEAN(mp);
1739 	} else if ((flags & VR_START_WRITE) != 0) {
1740 		MNT_REF(mp);
1741 		vn_start_write_locked(mp, 0);
1742 	} else {
1743 		MNT_IUNLOCK(mp);
1744 	}
1745 }
1746 
1747 /*
1748  * Implement kqueues for files by translating it to vnode operation.
1749  */
1750 static int
1751 vn_kqfilter(struct file *fp, struct knote *kn)
1752 {
1753 
1754 	return (VOP_KQFILTER(fp->f_vnode, kn));
1755 }
1756 
1757 /*
1758  * Simplified in-kernel wrapper calls for extended attribute access.
1759  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1760  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1761  */
1762 int
1763 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1764     const char *attrname, int *buflen, char *buf, struct thread *td)
1765 {
1766 	struct uio	auio;
1767 	struct iovec	iov;
1768 	int	error;
1769 
1770 	iov.iov_len = *buflen;
1771 	iov.iov_base = buf;
1772 
1773 	auio.uio_iov = &iov;
1774 	auio.uio_iovcnt = 1;
1775 	auio.uio_rw = UIO_READ;
1776 	auio.uio_segflg = UIO_SYSSPACE;
1777 	auio.uio_td = td;
1778 	auio.uio_offset = 0;
1779 	auio.uio_resid = *buflen;
1780 
1781 	if ((ioflg & IO_NODELOCKED) == 0)
1782 		vn_lock(vp, LK_SHARED | LK_RETRY);
1783 
1784 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1785 
1786 	/* authorize attribute retrieval as kernel */
1787 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1788 	    td);
1789 
1790 	if ((ioflg & IO_NODELOCKED) == 0)
1791 		VOP_UNLOCK(vp, 0);
1792 
1793 	if (error == 0) {
1794 		*buflen = *buflen - auio.uio_resid;
1795 	}
1796 
1797 	return (error);
1798 }
1799 
1800 /*
1801  * XXX failure mode if partially written?
1802  */
1803 int
1804 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1805     const char *attrname, int buflen, char *buf, struct thread *td)
1806 {
1807 	struct uio	auio;
1808 	struct iovec	iov;
1809 	struct mount	*mp;
1810 	int	error;
1811 
1812 	iov.iov_len = buflen;
1813 	iov.iov_base = buf;
1814 
1815 	auio.uio_iov = &iov;
1816 	auio.uio_iovcnt = 1;
1817 	auio.uio_rw = UIO_WRITE;
1818 	auio.uio_segflg = UIO_SYSSPACE;
1819 	auio.uio_td = td;
1820 	auio.uio_offset = 0;
1821 	auio.uio_resid = buflen;
1822 
1823 	if ((ioflg & IO_NODELOCKED) == 0) {
1824 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1825 			return (error);
1826 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1827 	}
1828 
1829 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1830 
1831 	/* authorize attribute setting as kernel */
1832 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1833 
1834 	if ((ioflg & IO_NODELOCKED) == 0) {
1835 		vn_finished_write(mp);
1836 		VOP_UNLOCK(vp, 0);
1837 	}
1838 
1839 	return (error);
1840 }
1841 
1842 int
1843 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1844     const char *attrname, struct thread *td)
1845 {
1846 	struct mount	*mp;
1847 	int	error;
1848 
1849 	if ((ioflg & IO_NODELOCKED) == 0) {
1850 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1851 			return (error);
1852 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1853 	}
1854 
1855 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1856 
1857 	/* authorize attribute removal as kernel */
1858 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1859 	if (error == EOPNOTSUPP)
1860 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1861 		    NULL, td);
1862 
1863 	if ((ioflg & IO_NODELOCKED) == 0) {
1864 		vn_finished_write(mp);
1865 		VOP_UNLOCK(vp, 0);
1866 	}
1867 
1868 	return (error);
1869 }
1870 
1871 int
1872 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1873 {
1874 	struct mount *mp;
1875 	int ltype, error;
1876 
1877 	mp = vp->v_mount;
1878 	ltype = VOP_ISLOCKED(vp);
1879 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1880 	    ("vn_vget_ino: vp not locked"));
1881 	error = vfs_busy(mp, MBF_NOWAIT);
1882 	if (error != 0) {
1883 		vfs_ref(mp);
1884 		VOP_UNLOCK(vp, 0);
1885 		error = vfs_busy(mp, 0);
1886 		vn_lock(vp, ltype | LK_RETRY);
1887 		vfs_rel(mp);
1888 		if (error != 0)
1889 			return (ENOENT);
1890 		if (vp->v_iflag & VI_DOOMED) {
1891 			vfs_unbusy(mp);
1892 			return (ENOENT);
1893 		}
1894 	}
1895 	VOP_UNLOCK(vp, 0);
1896 	error = VFS_VGET(mp, ino, lkflags, rvp);
1897 	vfs_unbusy(mp);
1898 	vn_lock(vp, ltype | LK_RETRY);
1899 	if (vp->v_iflag & VI_DOOMED) {
1900 		if (error == 0)
1901 			vput(*rvp);
1902 		error = ENOENT;
1903 	}
1904 	return (error);
1905 }
1906 
1907 int
1908 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1909     const struct thread *td)
1910 {
1911 
1912 	if (vp->v_type != VREG || td == NULL)
1913 		return (0);
1914 	PROC_LOCK(td->td_proc);
1915 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1916 	    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1917 		kern_psignal(td->td_proc, SIGXFSZ);
1918 		PROC_UNLOCK(td->td_proc);
1919 		return (EFBIG);
1920 	}
1921 	PROC_UNLOCK(td->td_proc);
1922 	return (0);
1923 }
1924 
1925 int
1926 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1927     struct thread *td)
1928 {
1929 	struct vnode *vp;
1930 
1931 	vp = fp->f_vnode;
1932 #ifdef AUDIT
1933 	vn_lock(vp, LK_SHARED | LK_RETRY);
1934 	AUDIT_ARG_VNODE1(vp);
1935 	VOP_UNLOCK(vp, 0);
1936 #endif
1937 	return (setfmode(td, active_cred, vp, mode));
1938 }
1939 
1940 int
1941 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1942     struct thread *td)
1943 {
1944 	struct vnode *vp;
1945 
1946 	vp = fp->f_vnode;
1947 #ifdef AUDIT
1948 	vn_lock(vp, LK_SHARED | LK_RETRY);
1949 	AUDIT_ARG_VNODE1(vp);
1950 	VOP_UNLOCK(vp, 0);
1951 #endif
1952 	return (setfown(td, active_cred, vp, uid, gid));
1953 }
1954 
1955 void
1956 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1957 {
1958 	vm_object_t object;
1959 
1960 	if ((object = vp->v_object) == NULL)
1961 		return;
1962 	VM_OBJECT_WLOCK(object);
1963 	vm_object_page_remove(object, start, end, 0);
1964 	VM_OBJECT_WUNLOCK(object);
1965 }
1966 
1967 int
1968 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1969 {
1970 	struct vattr va;
1971 	daddr_t bn, bnp;
1972 	uint64_t bsize;
1973 	off_t noff;
1974 	int error;
1975 
1976 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1977 	    ("Wrong command %lu", cmd));
1978 
1979 	if (vn_lock(vp, LK_SHARED) != 0)
1980 		return (EBADF);
1981 	if (vp->v_type != VREG) {
1982 		error = ENOTTY;
1983 		goto unlock;
1984 	}
1985 	error = VOP_GETATTR(vp, &va, cred);
1986 	if (error != 0)
1987 		goto unlock;
1988 	noff = *off;
1989 	if (noff >= va.va_size) {
1990 		error = ENXIO;
1991 		goto unlock;
1992 	}
1993 	bsize = vp->v_mount->mnt_stat.f_iosize;
1994 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1995 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1996 		if (error == EOPNOTSUPP) {
1997 			error = ENOTTY;
1998 			goto unlock;
1999 		}
2000 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2001 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
2002 			noff = bn * bsize;
2003 			if (noff < *off)
2004 				noff = *off;
2005 			goto unlock;
2006 		}
2007 	}
2008 	if (noff > va.va_size)
2009 		noff = va.va_size;
2010 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2011 	if (cmd == FIOSEEKDATA)
2012 		error = ENXIO;
2013 unlock:
2014 	VOP_UNLOCK(vp, 0);
2015 	if (error == 0)
2016 		*off = noff;
2017 	return (error);
2018 }
2019 
2020 int
2021 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2022 {
2023 	struct ucred *cred;
2024 	struct vnode *vp;
2025 	struct vattr vattr;
2026 	off_t foffset, size;
2027 	int error, noneg;
2028 
2029 	cred = td->td_ucred;
2030 	vp = fp->f_vnode;
2031 	foffset = foffset_lock(fp, 0);
2032 	noneg = (vp->v_type != VCHR);
2033 	error = 0;
2034 	switch (whence) {
2035 	case L_INCR:
2036 		if (noneg &&
2037 		    (foffset < 0 ||
2038 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2039 			error = EOVERFLOW;
2040 			break;
2041 		}
2042 		offset += foffset;
2043 		break;
2044 	case L_XTND:
2045 		vn_lock(vp, LK_SHARED | LK_RETRY);
2046 		error = VOP_GETATTR(vp, &vattr, cred);
2047 		VOP_UNLOCK(vp, 0);
2048 		if (error)
2049 			break;
2050 
2051 		/*
2052 		 * If the file references a disk device, then fetch
2053 		 * the media size and use that to determine the ending
2054 		 * offset.
2055 		 */
2056 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2057 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2058 			vattr.va_size = size;
2059 		if (noneg &&
2060 		    (vattr.va_size > OFF_MAX ||
2061 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2062 			error = EOVERFLOW;
2063 			break;
2064 		}
2065 		offset += vattr.va_size;
2066 		break;
2067 	case L_SET:
2068 		break;
2069 	case SEEK_DATA:
2070 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2071 		break;
2072 	case SEEK_HOLE:
2073 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2074 		break;
2075 	default:
2076 		error = EINVAL;
2077 	}
2078 	if (error == 0 && noneg && offset < 0)
2079 		error = EINVAL;
2080 	if (error != 0)
2081 		goto drop;
2082 	VFS_KNOTE_UNLOCKED(vp, 0);
2083 	td->td_uretoff.tdu_off = offset;
2084 drop:
2085 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2086 	return (error);
2087 }
2088