xref: /freebsd/sys/kern/vfs_vnops.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
193 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
194 			ndp->ni_cnd.cn_flags |= FOLLOW;
195 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
196 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
197 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
198 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
199 		bwillwrite();
200 		if ((error = namei(ndp)) != 0)
201 			return (error);
202 		if (ndp->ni_vp == NULL) {
203 			VATTR_NULL(vap);
204 			vap->va_type = VREG;
205 			vap->va_mode = cmode;
206 			if (fmode & O_EXCL)
207 				vap->va_vaflags |= VA_EXCLUSIVE;
208 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
209 				NDFREE(ndp, NDF_ONLY_PNBUF);
210 				vput(ndp->ni_dvp);
211 				if ((error = vn_start_write(NULL, &mp,
212 				    V_XSLEEP | PCATCH)) != 0)
213 					return (error);
214 				goto restart;
215 			}
216 #ifdef MAC
217 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
218 			    &ndp->ni_cnd, vap);
219 			if (error == 0)
220 #endif
221 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
222 						   &ndp->ni_cnd, vap);
223 			vput(ndp->ni_dvp);
224 			vn_finished_write(mp);
225 			if (error) {
226 				NDFREE(ndp, NDF_ONLY_PNBUF);
227 				return (error);
228 			}
229 			fmode &= ~O_TRUNC;
230 			vp = ndp->ni_vp;
231 		} else {
232 			if (ndp->ni_dvp == ndp->ni_vp)
233 				vrele(ndp->ni_dvp);
234 			else
235 				vput(ndp->ni_dvp);
236 			ndp->ni_dvp = NULL;
237 			vp = ndp->ni_vp;
238 			if (fmode & O_EXCL) {
239 				error = EEXIST;
240 				goto bad;
241 			}
242 			fmode &= ~O_CREAT;
243 		}
244 	} else {
245 		ndp->ni_cnd.cn_nameiop = LOOKUP;
246 		ndp->ni_cnd.cn_flags = ISOPEN |
247 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
248 		if (!(fmode & FWRITE))
249 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
250 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
251 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
252 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
253 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
254 		if ((error = namei(ndp)) != 0)
255 			return (error);
256 		vp = ndp->ni_vp;
257 	}
258 	error = vn_open_vnode(vp, fmode, cred, td, fp);
259 	if (error)
260 		goto bad;
261 	*flagp = fmode;
262 	return (0);
263 bad:
264 	NDFREE(ndp, NDF_ONLY_PNBUF);
265 	vput(vp);
266 	*flagp = fmode;
267 	ndp->ni_vp = NULL;
268 	return (error);
269 }
270 
271 /*
272  * Common code for vnode open operations once a vnode is located.
273  * Check permissions, and call the VOP_OPEN routine.
274  */
275 int
276 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
277     struct thread *td, struct file *fp)
278 {
279 	struct mount *mp;
280 	accmode_t accmode;
281 	struct flock lf;
282 	int error, have_flock, lock_flags, type;
283 
284 	if (vp->v_type == VLNK)
285 		return (EMLINK);
286 	if (vp->v_type == VSOCK)
287 		return (EOPNOTSUPP);
288 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
289 		return (ENOTDIR);
290 	accmode = 0;
291 	if (fmode & (FWRITE | O_TRUNC)) {
292 		if (vp->v_type == VDIR)
293 			return (EISDIR);
294 		accmode |= VWRITE;
295 	}
296 	if (fmode & FREAD)
297 		accmode |= VREAD;
298 	if (fmode & FEXEC)
299 		accmode |= VEXEC;
300 	if ((fmode & O_APPEND) && (fmode & FWRITE))
301 		accmode |= VAPPEND;
302 #ifdef MAC
303 	error = mac_vnode_check_open(cred, vp, accmode);
304 	if (error)
305 		return (error);
306 #endif
307 	if ((fmode & O_CREAT) == 0) {
308 		if (accmode & VWRITE) {
309 			error = vn_writechk(vp);
310 			if (error)
311 				return (error);
312 		}
313 		if (accmode) {
314 		        error = VOP_ACCESS(vp, accmode, cred, td);
315 			if (error)
316 				return (error);
317 		}
318 	}
319 	if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
320 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
321 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
322 		return (error);
323 
324 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
325 		KASSERT(fp != NULL, ("open with flock requires fp"));
326 		lock_flags = VOP_ISLOCKED(vp);
327 		VOP_UNLOCK(vp, 0);
328 		lf.l_whence = SEEK_SET;
329 		lf.l_start = 0;
330 		lf.l_len = 0;
331 		if (fmode & O_EXLOCK)
332 			lf.l_type = F_WRLCK;
333 		else
334 			lf.l_type = F_RDLCK;
335 		type = F_FLOCK;
336 		if ((fmode & FNONBLOCK) == 0)
337 			type |= F_WAIT;
338 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
339 		have_flock = (error == 0);
340 		vn_lock(vp, lock_flags | LK_RETRY);
341 		if (error == 0 && vp->v_iflag & VI_DOOMED)
342 			error = ENOENT;
343 		/*
344 		 * Another thread might have used this vnode as an
345 		 * executable while the vnode lock was dropped.
346 		 * Ensure the vnode is still able to be opened for
347 		 * writing after the lock has been obtained.
348 		 */
349 		if (error == 0 && accmode & VWRITE)
350 			error = vn_writechk(vp);
351 		if (error) {
352 			VOP_UNLOCK(vp, 0);
353 			if (have_flock) {
354 				lf.l_whence = SEEK_SET;
355 				lf.l_start = 0;
356 				lf.l_len = 0;
357 				lf.l_type = F_UNLCK;
358 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
359 				    F_FLOCK);
360 			}
361 			vn_start_write(vp, &mp, V_WAIT);
362 			vn_lock(vp, lock_flags | LK_RETRY);
363 			(void)VOP_CLOSE(vp, fmode, cred, td);
364 			vn_finished_write(mp);
365 			/* Prevent second close from fdrop()->vn_close(). */
366 			if (fp != NULL)
367 				fp->f_ops= &badfileops;
368 			return (error);
369 		}
370 		fp->f_flag |= FHASLOCK;
371 	}
372 	if (fmode & FWRITE) {
373 		VOP_ADD_WRITECOUNT(vp, 1);
374 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
375 		    __func__, vp, vp->v_writecount);
376 	}
377 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
378 	return (0);
379 }
380 
381 /*
382  * Check for write permissions on the specified vnode.
383  * Prototype text segments cannot be written.
384  */
385 int
386 vn_writechk(vp)
387 	register struct vnode *vp;
388 {
389 
390 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
391 	/*
392 	 * If there's shared text associated with
393 	 * the vnode, try to free it up once.  If
394 	 * we fail, we can't allow writing.
395 	 */
396 	if (VOP_IS_TEXT(vp))
397 		return (ETXTBSY);
398 
399 	return (0);
400 }
401 
402 /*
403  * Vnode close call
404  */
405 int
406 vn_close(vp, flags, file_cred, td)
407 	register struct vnode *vp;
408 	int flags;
409 	struct ucred *file_cred;
410 	struct thread *td;
411 {
412 	struct mount *mp;
413 	int error, lock_flags;
414 
415 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
416 	    MNT_EXTENDED_SHARED(vp->v_mount))
417 		lock_flags = LK_SHARED;
418 	else
419 		lock_flags = LK_EXCLUSIVE;
420 
421 	vn_start_write(vp, &mp, V_WAIT);
422 	vn_lock(vp, lock_flags | LK_RETRY);
423 	if (flags & FWRITE) {
424 		VNASSERT(vp->v_writecount > 0, vp,
425 		    ("vn_close: negative writecount"));
426 		VOP_ADD_WRITECOUNT(vp, -1);
427 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
428 		    __func__, vp, vp->v_writecount);
429 	}
430 	error = VOP_CLOSE(vp, flags, file_cred, td);
431 	vput(vp);
432 	vn_finished_write(mp);
433 	return (error);
434 }
435 
436 /*
437  * Heuristic to detect sequential operation.
438  */
439 static int
440 sequential_heuristic(struct uio *uio, struct file *fp)
441 {
442 
443 	ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
444 	if (fp->f_flag & FRDAHEAD)
445 		return (fp->f_seqcount << IO_SEQSHIFT);
446 
447 	/*
448 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
449 	 * that the first I/O is normally considered to be slightly
450 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
451 	 * unless previous seeks have reduced f_seqcount to 0, in which
452 	 * case offset 0 is not special.
453 	 */
454 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
455 	    uio->uio_offset == fp->f_nextoff) {
456 		/*
457 		 * f_seqcount is in units of fixed-size blocks so that it
458 		 * depends mainly on the amount of sequential I/O and not
459 		 * much on the number of sequential I/O's.  The fixed size
460 		 * of 16384 is hard-coded here since it is (not quite) just
461 		 * a magic size that works well here.  This size is more
462 		 * closely related to the best I/O size for real disks than
463 		 * to any block size used by software.
464 		 */
465 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
466 		if (fp->f_seqcount > IO_SEQMAX)
467 			fp->f_seqcount = IO_SEQMAX;
468 		return (fp->f_seqcount << IO_SEQSHIFT);
469 	}
470 
471 	/* Not sequential.  Quickly draw-down sequentiality. */
472 	if (fp->f_seqcount > 1)
473 		fp->f_seqcount = 1;
474 	else
475 		fp->f_seqcount = 0;
476 	return (0);
477 }
478 
479 /*
480  * Package up an I/O request on a vnode into a uio and do it.
481  */
482 int
483 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
484     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
485     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
486 {
487 	struct uio auio;
488 	struct iovec aiov;
489 	struct mount *mp;
490 	struct ucred *cred;
491 	void *rl_cookie;
492 	struct vn_io_fault_args args;
493 	int error, lock_flags;
494 
495 	auio.uio_iov = &aiov;
496 	auio.uio_iovcnt = 1;
497 	aiov.iov_base = base;
498 	aiov.iov_len = len;
499 	auio.uio_resid = len;
500 	auio.uio_offset = offset;
501 	auio.uio_segflg = segflg;
502 	auio.uio_rw = rw;
503 	auio.uio_td = td;
504 	error = 0;
505 
506 	if ((ioflg & IO_NODELOCKED) == 0) {
507 		if ((ioflg & IO_RANGELOCKED) == 0) {
508 			if (rw == UIO_READ) {
509 				rl_cookie = vn_rangelock_rlock(vp, offset,
510 				    offset + len);
511 			} else {
512 				rl_cookie = vn_rangelock_wlock(vp, offset,
513 				    offset + len);
514 			}
515 		} else
516 			rl_cookie = NULL;
517 		mp = NULL;
518 		if (rw == UIO_WRITE) {
519 			if (vp->v_type != VCHR &&
520 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
521 			    != 0)
522 				goto out;
523 			if (MNT_SHARED_WRITES(mp) ||
524 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
525 				lock_flags = LK_SHARED;
526 			else
527 				lock_flags = LK_EXCLUSIVE;
528 		} else
529 			lock_flags = LK_SHARED;
530 		vn_lock(vp, lock_flags | LK_RETRY);
531 	} else
532 		rl_cookie = NULL;
533 
534 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
535 #ifdef MAC
536 	if ((ioflg & IO_NOMACCHECK) == 0) {
537 		if (rw == UIO_READ)
538 			error = mac_vnode_check_read(active_cred, file_cred,
539 			    vp);
540 		else
541 			error = mac_vnode_check_write(active_cred, file_cred,
542 			    vp);
543 	}
544 #endif
545 	if (error == 0) {
546 		if (file_cred != NULL)
547 			cred = file_cred;
548 		else
549 			cred = active_cred;
550 		if (do_vn_io_fault(vp, &auio)) {
551 			args.kind = VN_IO_FAULT_VOP;
552 			args.cred = cred;
553 			args.flags = ioflg;
554 			args.args.vop_args.vp = vp;
555 			error = vn_io_fault1(vp, &auio, &args, td);
556 		} else if (rw == UIO_READ) {
557 			error = VOP_READ(vp, &auio, ioflg, cred);
558 		} else /* if (rw == UIO_WRITE) */ {
559 			error = VOP_WRITE(vp, &auio, ioflg, cred);
560 		}
561 	}
562 	if (aresid)
563 		*aresid = auio.uio_resid;
564 	else
565 		if (auio.uio_resid && error == 0)
566 			error = EIO;
567 	if ((ioflg & IO_NODELOCKED) == 0) {
568 		VOP_UNLOCK(vp, 0);
569 		if (mp != NULL)
570 			vn_finished_write(mp);
571 	}
572  out:
573 	if (rl_cookie != NULL)
574 		vn_rangelock_unlock(vp, rl_cookie);
575 	return (error);
576 }
577 
578 /*
579  * Package up an I/O request on a vnode into a uio and do it.  The I/O
580  * request is split up into smaller chunks and we try to avoid saturating
581  * the buffer cache while potentially holding a vnode locked, so we
582  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
583  * to give other processes a chance to lock the vnode (either other processes
584  * core'ing the same binary, or unrelated processes scanning the directory).
585  */
586 int
587 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
588     file_cred, aresid, td)
589 	enum uio_rw rw;
590 	struct vnode *vp;
591 	void *base;
592 	size_t len;
593 	off_t offset;
594 	enum uio_seg segflg;
595 	int ioflg;
596 	struct ucred *active_cred;
597 	struct ucred *file_cred;
598 	size_t *aresid;
599 	struct thread *td;
600 {
601 	int error = 0;
602 	ssize_t iaresid;
603 
604 	do {
605 		int chunk;
606 
607 		/*
608 		 * Force `offset' to a multiple of MAXBSIZE except possibly
609 		 * for the first chunk, so that filesystems only need to
610 		 * write full blocks except possibly for the first and last
611 		 * chunks.
612 		 */
613 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
614 
615 		if (chunk > len)
616 			chunk = len;
617 		if (rw != UIO_READ && vp->v_type == VREG)
618 			bwillwrite();
619 		iaresid = 0;
620 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
621 		    ioflg, active_cred, file_cred, &iaresid, td);
622 		len -= chunk;	/* aresid calc already includes length */
623 		if (error)
624 			break;
625 		offset += chunk;
626 		base = (char *)base + chunk;
627 		kern_yield(PRI_USER);
628 	} while (len);
629 	if (aresid)
630 		*aresid = len + iaresid;
631 	return (error);
632 }
633 
634 off_t
635 foffset_lock(struct file *fp, int flags)
636 {
637 	struct mtx *mtxp;
638 	off_t res;
639 
640 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
641 
642 #if OFF_MAX <= LONG_MAX
643 	/*
644 	 * Caller only wants the current f_offset value.  Assume that
645 	 * the long and shorter integer types reads are atomic.
646 	 */
647 	if ((flags & FOF_NOLOCK) != 0)
648 		return (fp->f_offset);
649 #endif
650 
651 	/*
652 	 * According to McKusick the vn lock was protecting f_offset here.
653 	 * It is now protected by the FOFFSET_LOCKED flag.
654 	 */
655 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
656 	mtx_lock(mtxp);
657 	if ((flags & FOF_NOLOCK) == 0) {
658 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
659 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
660 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
661 			    "vofflock", 0);
662 		}
663 		fp->f_vnread_flags |= FOFFSET_LOCKED;
664 	}
665 	res = fp->f_offset;
666 	mtx_unlock(mtxp);
667 	return (res);
668 }
669 
670 void
671 foffset_unlock(struct file *fp, off_t val, int flags)
672 {
673 	struct mtx *mtxp;
674 
675 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
676 
677 #if OFF_MAX <= LONG_MAX
678 	if ((flags & FOF_NOLOCK) != 0) {
679 		if ((flags & FOF_NOUPDATE) == 0)
680 			fp->f_offset = val;
681 		if ((flags & FOF_NEXTOFF) != 0)
682 			fp->f_nextoff = val;
683 		return;
684 	}
685 #endif
686 
687 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
688 	mtx_lock(mtxp);
689 	if ((flags & FOF_NOUPDATE) == 0)
690 		fp->f_offset = val;
691 	if ((flags & FOF_NEXTOFF) != 0)
692 		fp->f_nextoff = val;
693 	if ((flags & FOF_NOLOCK) == 0) {
694 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
695 		    ("Lost FOFFSET_LOCKED"));
696 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
697 			wakeup(&fp->f_vnread_flags);
698 		fp->f_vnread_flags = 0;
699 	}
700 	mtx_unlock(mtxp);
701 }
702 
703 void
704 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
705 {
706 
707 	if ((flags & FOF_OFFSET) == 0)
708 		uio->uio_offset = foffset_lock(fp, flags);
709 }
710 
711 void
712 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
713 {
714 
715 	if ((flags & FOF_OFFSET) == 0)
716 		foffset_unlock(fp, uio->uio_offset, flags);
717 }
718 
719 static int
720 get_advice(struct file *fp, struct uio *uio)
721 {
722 	struct mtx *mtxp;
723 	int ret;
724 
725 	ret = POSIX_FADV_NORMAL;
726 	if (fp->f_advice == NULL)
727 		return (ret);
728 
729 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
730 	mtx_lock(mtxp);
731 	if (uio->uio_offset >= fp->f_advice->fa_start &&
732 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
733 		ret = fp->f_advice->fa_advice;
734 	mtx_unlock(mtxp);
735 	return (ret);
736 }
737 
738 /*
739  * File table vnode read routine.
740  */
741 static int
742 vn_read(fp, uio, active_cred, flags, td)
743 	struct file *fp;
744 	struct uio *uio;
745 	struct ucred *active_cred;
746 	int flags;
747 	struct thread *td;
748 {
749 	struct vnode *vp;
750 	struct mtx *mtxp;
751 	int error, ioflag;
752 	int advice;
753 	off_t offset, start, end;
754 
755 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
756 	    uio->uio_td, td));
757 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
758 	vp = fp->f_vnode;
759 	ioflag = 0;
760 	if (fp->f_flag & FNONBLOCK)
761 		ioflag |= IO_NDELAY;
762 	if (fp->f_flag & O_DIRECT)
763 		ioflag |= IO_DIRECT;
764 	advice = get_advice(fp, uio);
765 	vn_lock(vp, LK_SHARED | LK_RETRY);
766 
767 	switch (advice) {
768 	case POSIX_FADV_NORMAL:
769 	case POSIX_FADV_SEQUENTIAL:
770 	case POSIX_FADV_NOREUSE:
771 		ioflag |= sequential_heuristic(uio, fp);
772 		break;
773 	case POSIX_FADV_RANDOM:
774 		/* Disable read-ahead for random I/O. */
775 		break;
776 	}
777 	offset = uio->uio_offset;
778 
779 #ifdef MAC
780 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
781 	if (error == 0)
782 #endif
783 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
784 	fp->f_nextoff = uio->uio_offset;
785 	VOP_UNLOCK(vp, 0);
786 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
787 	    offset != uio->uio_offset) {
788 		/*
789 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
790 		 * buffers for the backing file after a
791 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
792 		 * case of using POSIX_FADV_NOREUSE with sequential
793 		 * access, track the previous implicit DONTNEED
794 		 * request and grow this request to include the
795 		 * current read(2) in addition to the previous
796 		 * DONTNEED.  With purely sequential access this will
797 		 * cause the DONTNEED requests to continously grow to
798 		 * cover all of the previously read regions of the
799 		 * file.  This allows filesystem blocks that are
800 		 * accessed by multiple calls to read(2) to be flushed
801 		 * once the last read(2) finishes.
802 		 */
803 		start = offset;
804 		end = uio->uio_offset - 1;
805 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
806 		mtx_lock(mtxp);
807 		if (fp->f_advice != NULL &&
808 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
809 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
810 				start = fp->f_advice->fa_prevstart;
811 			else if (fp->f_advice->fa_prevstart != 0 &&
812 			    fp->f_advice->fa_prevstart == end + 1)
813 				end = fp->f_advice->fa_prevend;
814 			fp->f_advice->fa_prevstart = start;
815 			fp->f_advice->fa_prevend = end;
816 		}
817 		mtx_unlock(mtxp);
818 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
819 	}
820 	return (error);
821 }
822 
823 /*
824  * File table vnode write routine.
825  */
826 static int
827 vn_write(fp, uio, active_cred, flags, td)
828 	struct file *fp;
829 	struct uio *uio;
830 	struct ucred *active_cred;
831 	int flags;
832 	struct thread *td;
833 {
834 	struct vnode *vp;
835 	struct mount *mp;
836 	struct mtx *mtxp;
837 	int error, ioflag, lock_flags;
838 	int advice;
839 	off_t offset, start, end;
840 
841 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
842 	    uio->uio_td, td));
843 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
844 	vp = fp->f_vnode;
845 	if (vp->v_type == VREG)
846 		bwillwrite();
847 	ioflag = IO_UNIT;
848 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
849 		ioflag |= IO_APPEND;
850 	if (fp->f_flag & FNONBLOCK)
851 		ioflag |= IO_NDELAY;
852 	if (fp->f_flag & O_DIRECT)
853 		ioflag |= IO_DIRECT;
854 	if ((fp->f_flag & O_FSYNC) ||
855 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
856 		ioflag |= IO_SYNC;
857 	mp = NULL;
858 	if (vp->v_type != VCHR &&
859 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
860 		goto unlock;
861 
862 	advice = get_advice(fp, uio);
863 
864 	if (MNT_SHARED_WRITES(mp) ||
865 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
866 		lock_flags = LK_SHARED;
867 	} else {
868 		lock_flags = LK_EXCLUSIVE;
869 	}
870 
871 	vn_lock(vp, lock_flags | LK_RETRY);
872 	switch (advice) {
873 	case POSIX_FADV_NORMAL:
874 	case POSIX_FADV_SEQUENTIAL:
875 	case POSIX_FADV_NOREUSE:
876 		ioflag |= sequential_heuristic(uio, fp);
877 		break;
878 	case POSIX_FADV_RANDOM:
879 		/* XXX: Is this correct? */
880 		break;
881 	}
882 	offset = uio->uio_offset;
883 
884 #ifdef MAC
885 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
886 	if (error == 0)
887 #endif
888 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
889 	fp->f_nextoff = uio->uio_offset;
890 	VOP_UNLOCK(vp, 0);
891 	if (vp->v_type != VCHR)
892 		vn_finished_write(mp);
893 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
894 	    offset != uio->uio_offset) {
895 		/*
896 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
897 		 * buffers for the backing file after a
898 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
899 		 * common case of using POSIX_FADV_NOREUSE with
900 		 * sequential access, track the previous implicit
901 		 * DONTNEED request and grow this request to include
902 		 * the current write(2) in addition to the previous
903 		 * DONTNEED.  With purely sequential access this will
904 		 * cause the DONTNEED requests to continously grow to
905 		 * cover all of the previously written regions of the
906 		 * file.
907 		 *
908 		 * Note that the blocks just written are almost
909 		 * certainly still dirty, so this only works when
910 		 * VOP_ADVISE() calls from subsequent writes push out
911 		 * the data written by this write(2) once the backing
912 		 * buffers are clean.  However, as compared to forcing
913 		 * IO_DIRECT, this gives much saner behavior.  Write
914 		 * clustering is still allowed, and clean pages are
915 		 * merely moved to the cache page queue rather than
916 		 * outright thrown away.  This means a subsequent
917 		 * read(2) can still avoid hitting the disk if the
918 		 * pages have not been reclaimed.
919 		 *
920 		 * This does make POSIX_FADV_NOREUSE largely useless
921 		 * with non-sequential access.  However, sequential
922 		 * access is the more common use case and the flag is
923 		 * merely advisory.
924 		 */
925 		start = offset;
926 		end = uio->uio_offset - 1;
927 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
928 		mtx_lock(mtxp);
929 		if (fp->f_advice != NULL &&
930 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
931 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
932 				start = fp->f_advice->fa_prevstart;
933 			else if (fp->f_advice->fa_prevstart != 0 &&
934 			    fp->f_advice->fa_prevstart == end + 1)
935 				end = fp->f_advice->fa_prevend;
936 			fp->f_advice->fa_prevstart = start;
937 			fp->f_advice->fa_prevend = end;
938 		}
939 		mtx_unlock(mtxp);
940 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
941 	}
942 
943 unlock:
944 	return (error);
945 }
946 
947 /*
948  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
949  * prevent the following deadlock:
950  *
951  * Assume that the thread A reads from the vnode vp1 into userspace
952  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
953  * currently not resident, then system ends up with the call chain
954  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
955  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
956  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
957  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
958  * backed by the pages of vnode vp1, and some page in buf2 is not
959  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
960  *
961  * To prevent the lock order reversal and deadlock, vn_io_fault() does
962  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
963  * Instead, it first tries to do the whole range i/o with pagefaults
964  * disabled. If all pages in the i/o buffer are resident and mapped,
965  * VOP will succeed (ignoring the genuine filesystem errors).
966  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
967  * i/o in chunks, with all pages in the chunk prefaulted and held
968  * using vm_fault_quick_hold_pages().
969  *
970  * Filesystems using this deadlock avoidance scheme should use the
971  * array of the held pages from uio, saved in the curthread->td_ma,
972  * instead of doing uiomove().  A helper function
973  * vn_io_fault_uiomove() converts uiomove request into
974  * uiomove_fromphys() over td_ma array.
975  *
976  * Since vnode locks do not cover the whole i/o anymore, rangelocks
977  * make the current i/o request atomic with respect to other i/os and
978  * truncations.
979  */
980 
981 /*
982  * Decode vn_io_fault_args and perform the corresponding i/o.
983  */
984 static int
985 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
986     struct thread *td)
987 {
988 
989 	switch (args->kind) {
990 	case VN_IO_FAULT_FOP:
991 		return ((args->args.fop_args.doio)(args->args.fop_args.fp,
992 		    uio, args->cred, args->flags, td));
993 	case VN_IO_FAULT_VOP:
994 		if (uio->uio_rw == UIO_READ) {
995 			return (VOP_READ(args->args.vop_args.vp, uio,
996 			    args->flags, args->cred));
997 		} else if (uio->uio_rw == UIO_WRITE) {
998 			return (VOP_WRITE(args->args.vop_args.vp, uio,
999 			    args->flags, args->cred));
1000 		}
1001 		break;
1002 	}
1003 	panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
1004 	    uio->uio_rw);
1005 }
1006 
1007 /*
1008  * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1009  * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1010  * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1011  * into args and call vn_io_fault1() to handle faults during the user
1012  * mode buffer accesses.
1013  */
1014 static int
1015 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1016     struct thread *td)
1017 {
1018 	vm_page_t ma[io_hold_cnt + 2];
1019 	struct uio *uio_clone, short_uio;
1020 	struct iovec short_iovec[1];
1021 	vm_page_t *prev_td_ma;
1022 	vm_prot_t prot;
1023 	vm_offset_t addr, end;
1024 	size_t len, resid;
1025 	ssize_t adv;
1026 	int error, cnt, save, saveheld, prev_td_ma_cnt;
1027 
1028 	prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1029 
1030 	/*
1031 	 * The UFS follows IO_UNIT directive and replays back both
1032 	 * uio_offset and uio_resid if an error is encountered during the
1033 	 * operation.  But, since the iovec may be already advanced,
1034 	 * uio is still in an inconsistent state.
1035 	 *
1036 	 * Cache a copy of the original uio, which is advanced to the redo
1037 	 * point using UIO_NOCOPY below.
1038 	 */
1039 	uio_clone = cloneuio(uio);
1040 	resid = uio->uio_resid;
1041 
1042 	short_uio.uio_segflg = UIO_USERSPACE;
1043 	short_uio.uio_rw = uio->uio_rw;
1044 	short_uio.uio_td = uio->uio_td;
1045 
1046 	save = vm_fault_disable_pagefaults();
1047 	error = vn_io_fault_doio(args, uio, td);
1048 	if (error != EFAULT)
1049 		goto out;
1050 
1051 	atomic_add_long(&vn_io_faults_cnt, 1);
1052 	uio_clone->uio_segflg = UIO_NOCOPY;
1053 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
1054 	uio_clone->uio_segflg = uio->uio_segflg;
1055 
1056 	saveheld = curthread_pflags_set(TDP_UIOHELD);
1057 	prev_td_ma = td->td_ma;
1058 	prev_td_ma_cnt = td->td_ma_cnt;
1059 
1060 	while (uio_clone->uio_resid != 0) {
1061 		len = uio_clone->uio_iov->iov_len;
1062 		if (len == 0) {
1063 			KASSERT(uio_clone->uio_iovcnt >= 1,
1064 			    ("iovcnt underflow"));
1065 			uio_clone->uio_iov++;
1066 			uio_clone->uio_iovcnt--;
1067 			continue;
1068 		}
1069 		if (len > io_hold_cnt * PAGE_SIZE)
1070 			len = io_hold_cnt * PAGE_SIZE;
1071 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1072 		end = round_page(addr + len);
1073 		if (end < addr) {
1074 			error = EFAULT;
1075 			break;
1076 		}
1077 		cnt = atop(end - trunc_page(addr));
1078 		/*
1079 		 * A perfectly misaligned address and length could cause
1080 		 * both the start and the end of the chunk to use partial
1081 		 * page.  +2 accounts for such a situation.
1082 		 */
1083 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1084 		    addr, len, prot, ma, io_hold_cnt + 2);
1085 		if (cnt == -1) {
1086 			error = EFAULT;
1087 			break;
1088 		}
1089 		short_uio.uio_iov = &short_iovec[0];
1090 		short_iovec[0].iov_base = (void *)addr;
1091 		short_uio.uio_iovcnt = 1;
1092 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1093 		short_uio.uio_offset = uio_clone->uio_offset;
1094 		td->td_ma = ma;
1095 		td->td_ma_cnt = cnt;
1096 
1097 		error = vn_io_fault_doio(args, &short_uio, td);
1098 		vm_page_unhold_pages(ma, cnt);
1099 		adv = len - short_uio.uio_resid;
1100 
1101 		uio_clone->uio_iov->iov_base =
1102 		    (char *)uio_clone->uio_iov->iov_base + adv;
1103 		uio_clone->uio_iov->iov_len -= adv;
1104 		uio_clone->uio_resid -= adv;
1105 		uio_clone->uio_offset += adv;
1106 
1107 		uio->uio_resid -= adv;
1108 		uio->uio_offset += adv;
1109 
1110 		if (error != 0 || adv == 0)
1111 			break;
1112 	}
1113 	td->td_ma = prev_td_ma;
1114 	td->td_ma_cnt = prev_td_ma_cnt;
1115 	curthread_pflags_restore(saveheld);
1116 out:
1117 	vm_fault_enable_pagefaults(save);
1118 	free(uio_clone, M_IOV);
1119 	return (error);
1120 }
1121 
1122 static int
1123 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1124     int flags, struct thread *td)
1125 {
1126 	fo_rdwr_t *doio;
1127 	struct vnode *vp;
1128 	void *rl_cookie;
1129 	struct vn_io_fault_args args;
1130 	int error;
1131 
1132 	doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1133 	vp = fp->f_vnode;
1134 	foffset_lock_uio(fp, uio, flags);
1135 	if (do_vn_io_fault(vp, uio)) {
1136 		args.kind = VN_IO_FAULT_FOP;
1137 		args.args.fop_args.fp = fp;
1138 		args.args.fop_args.doio = doio;
1139 		args.cred = active_cred;
1140 		args.flags = flags | FOF_OFFSET;
1141 		if (uio->uio_rw == UIO_READ) {
1142 			rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1143 			    uio->uio_offset + uio->uio_resid);
1144 		} else if ((fp->f_flag & O_APPEND) != 0 ||
1145 		    (flags & FOF_OFFSET) == 0) {
1146 			/* For appenders, punt and lock the whole range. */
1147 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1148 		} else {
1149 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1150 			    uio->uio_offset + uio->uio_resid);
1151 		}
1152 		error = vn_io_fault1(vp, uio, &args, td);
1153 		vn_rangelock_unlock(vp, rl_cookie);
1154 	} else {
1155 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1156 	}
1157 	foffset_unlock_uio(fp, uio, flags);
1158 	return (error);
1159 }
1160 
1161 /*
1162  * Helper function to perform the requested uiomove operation using
1163  * the held pages for io->uio_iov[0].iov_base buffer instead of
1164  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1165  * instead of iov_base prevents page faults that could occur due to
1166  * pmap_collect() invalidating the mapping created by
1167  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1168  * object cleanup revoking the write access from page mappings.
1169  *
1170  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1171  * instead of plain uiomove().
1172  */
1173 int
1174 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1175 {
1176 	struct uio transp_uio;
1177 	struct iovec transp_iov[1];
1178 	struct thread *td;
1179 	size_t adv;
1180 	int error, pgadv;
1181 
1182 	td = curthread;
1183 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1184 	    uio->uio_segflg != UIO_USERSPACE)
1185 		return (uiomove(data, xfersize, uio));
1186 
1187 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1188 	transp_iov[0].iov_base = data;
1189 	transp_uio.uio_iov = &transp_iov[0];
1190 	transp_uio.uio_iovcnt = 1;
1191 	if (xfersize > uio->uio_resid)
1192 		xfersize = uio->uio_resid;
1193 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1194 	transp_uio.uio_offset = 0;
1195 	transp_uio.uio_segflg = UIO_SYSSPACE;
1196 	/*
1197 	 * Since transp_iov points to data, and td_ma page array
1198 	 * corresponds to original uio->uio_iov, we need to invert the
1199 	 * direction of the i/o operation as passed to
1200 	 * uiomove_fromphys().
1201 	 */
1202 	switch (uio->uio_rw) {
1203 	case UIO_WRITE:
1204 		transp_uio.uio_rw = UIO_READ;
1205 		break;
1206 	case UIO_READ:
1207 		transp_uio.uio_rw = UIO_WRITE;
1208 		break;
1209 	}
1210 	transp_uio.uio_td = uio->uio_td;
1211 	error = uiomove_fromphys(td->td_ma,
1212 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1213 	    xfersize, &transp_uio);
1214 	adv = xfersize - transp_uio.uio_resid;
1215 	pgadv =
1216 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1217 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1218 	td->td_ma += pgadv;
1219 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1220 	    pgadv));
1221 	td->td_ma_cnt -= pgadv;
1222 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1223 	uio->uio_iov->iov_len -= adv;
1224 	uio->uio_resid -= adv;
1225 	uio->uio_offset += adv;
1226 	return (error);
1227 }
1228 
1229 int
1230 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1231     struct uio *uio)
1232 {
1233 	struct thread *td;
1234 	vm_offset_t iov_base;
1235 	int cnt, pgadv;
1236 
1237 	td = curthread;
1238 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1239 	    uio->uio_segflg != UIO_USERSPACE)
1240 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1241 
1242 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1243 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1244 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1245 	switch (uio->uio_rw) {
1246 	case UIO_WRITE:
1247 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1248 		    offset, cnt);
1249 		break;
1250 	case UIO_READ:
1251 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1252 		    cnt);
1253 		break;
1254 	}
1255 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1256 	td->td_ma += pgadv;
1257 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1258 	    pgadv));
1259 	td->td_ma_cnt -= pgadv;
1260 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1261 	uio->uio_iov->iov_len -= cnt;
1262 	uio->uio_resid -= cnt;
1263 	uio->uio_offset += cnt;
1264 	return (0);
1265 }
1266 
1267 
1268 /*
1269  * File table truncate routine.
1270  */
1271 static int
1272 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1273     struct thread *td)
1274 {
1275 	struct vattr vattr;
1276 	struct mount *mp;
1277 	struct vnode *vp;
1278 	void *rl_cookie;
1279 	int error;
1280 
1281 	vp = fp->f_vnode;
1282 
1283 	/*
1284 	 * Lock the whole range for truncation.  Otherwise split i/o
1285 	 * might happen partly before and partly after the truncation.
1286 	 */
1287 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1288 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1289 	if (error)
1290 		goto out1;
1291 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1292 	if (vp->v_type == VDIR) {
1293 		error = EISDIR;
1294 		goto out;
1295 	}
1296 #ifdef MAC
1297 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1298 	if (error)
1299 		goto out;
1300 #endif
1301 	error = vn_writechk(vp);
1302 	if (error == 0) {
1303 		VATTR_NULL(&vattr);
1304 		vattr.va_size = length;
1305 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1306 	}
1307 out:
1308 	VOP_UNLOCK(vp, 0);
1309 	vn_finished_write(mp);
1310 out1:
1311 	vn_rangelock_unlock(vp, rl_cookie);
1312 	return (error);
1313 }
1314 
1315 /*
1316  * File table vnode stat routine.
1317  */
1318 static int
1319 vn_statfile(fp, sb, active_cred, td)
1320 	struct file *fp;
1321 	struct stat *sb;
1322 	struct ucred *active_cred;
1323 	struct thread *td;
1324 {
1325 	struct vnode *vp = fp->f_vnode;
1326 	int error;
1327 
1328 	vn_lock(vp, LK_SHARED | LK_RETRY);
1329 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1330 	VOP_UNLOCK(vp, 0);
1331 
1332 	return (error);
1333 }
1334 
1335 /*
1336  * Stat a vnode; implementation for the stat syscall
1337  */
1338 int
1339 vn_stat(vp, sb, active_cred, file_cred, td)
1340 	struct vnode *vp;
1341 	register struct stat *sb;
1342 	struct ucred *active_cred;
1343 	struct ucred *file_cred;
1344 	struct thread *td;
1345 {
1346 	struct vattr vattr;
1347 	register struct vattr *vap;
1348 	int error;
1349 	u_short mode;
1350 
1351 #ifdef MAC
1352 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1353 	if (error)
1354 		return (error);
1355 #endif
1356 
1357 	vap = &vattr;
1358 
1359 	/*
1360 	 * Initialize defaults for new and unusual fields, so that file
1361 	 * systems which don't support these fields don't need to know
1362 	 * about them.
1363 	 */
1364 	vap->va_birthtime.tv_sec = -1;
1365 	vap->va_birthtime.tv_nsec = 0;
1366 	vap->va_fsid = VNOVAL;
1367 	vap->va_rdev = NODEV;
1368 
1369 	error = VOP_GETATTR(vp, vap, active_cred);
1370 	if (error)
1371 		return (error);
1372 
1373 	/*
1374 	 * Zero the spare stat fields
1375 	 */
1376 	bzero(sb, sizeof *sb);
1377 
1378 	/*
1379 	 * Copy from vattr table
1380 	 */
1381 	if (vap->va_fsid != VNOVAL)
1382 		sb->st_dev = vap->va_fsid;
1383 	else
1384 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1385 	sb->st_ino = vap->va_fileid;
1386 	mode = vap->va_mode;
1387 	switch (vap->va_type) {
1388 	case VREG:
1389 		mode |= S_IFREG;
1390 		break;
1391 	case VDIR:
1392 		mode |= S_IFDIR;
1393 		break;
1394 	case VBLK:
1395 		mode |= S_IFBLK;
1396 		break;
1397 	case VCHR:
1398 		mode |= S_IFCHR;
1399 		break;
1400 	case VLNK:
1401 		mode |= S_IFLNK;
1402 		break;
1403 	case VSOCK:
1404 		mode |= S_IFSOCK;
1405 		break;
1406 	case VFIFO:
1407 		mode |= S_IFIFO;
1408 		break;
1409 	default:
1410 		return (EBADF);
1411 	};
1412 	sb->st_mode = mode;
1413 	sb->st_nlink = vap->va_nlink;
1414 	sb->st_uid = vap->va_uid;
1415 	sb->st_gid = vap->va_gid;
1416 	sb->st_rdev = vap->va_rdev;
1417 	if (vap->va_size > OFF_MAX)
1418 		return (EOVERFLOW);
1419 	sb->st_size = vap->va_size;
1420 	sb->st_atim = vap->va_atime;
1421 	sb->st_mtim = vap->va_mtime;
1422 	sb->st_ctim = vap->va_ctime;
1423 	sb->st_birthtim = vap->va_birthtime;
1424 
1425         /*
1426 	 * According to www.opengroup.org, the meaning of st_blksize is
1427 	 *   "a filesystem-specific preferred I/O block size for this
1428 	 *    object.  In some filesystem types, this may vary from file
1429 	 *    to file"
1430 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1431 	 */
1432 
1433 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1434 
1435 	sb->st_flags = vap->va_flags;
1436 	if (priv_check(td, PRIV_VFS_GENERATION))
1437 		sb->st_gen = 0;
1438 	else
1439 		sb->st_gen = vap->va_gen;
1440 
1441 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1442 	return (0);
1443 }
1444 
1445 /*
1446  * File table vnode ioctl routine.
1447  */
1448 static int
1449 vn_ioctl(fp, com, data, active_cred, td)
1450 	struct file *fp;
1451 	u_long com;
1452 	void *data;
1453 	struct ucred *active_cred;
1454 	struct thread *td;
1455 {
1456 	struct vattr vattr;
1457 	struct vnode *vp;
1458 	int error;
1459 
1460 	vp = fp->f_vnode;
1461 	switch (vp->v_type) {
1462 	case VDIR:
1463 	case VREG:
1464 		switch (com) {
1465 		case FIONREAD:
1466 			vn_lock(vp, LK_SHARED | LK_RETRY);
1467 			error = VOP_GETATTR(vp, &vattr, active_cred);
1468 			VOP_UNLOCK(vp, 0);
1469 			if (error == 0)
1470 				*(int *)data = vattr.va_size - fp->f_offset;
1471 			return (error);
1472 		case FIONBIO:
1473 		case FIOASYNC:
1474 			return (0);
1475 		default:
1476 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1477 			    active_cred, td));
1478 		}
1479 	default:
1480 		return (ENOTTY);
1481 	}
1482 }
1483 
1484 /*
1485  * File table vnode poll routine.
1486  */
1487 static int
1488 vn_poll(fp, events, active_cred, td)
1489 	struct file *fp;
1490 	int events;
1491 	struct ucred *active_cred;
1492 	struct thread *td;
1493 {
1494 	struct vnode *vp;
1495 	int error;
1496 
1497 	vp = fp->f_vnode;
1498 #ifdef MAC
1499 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1500 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1501 	VOP_UNLOCK(vp, 0);
1502 	if (!error)
1503 #endif
1504 
1505 	error = VOP_POLL(vp, events, fp->f_cred, td);
1506 	return (error);
1507 }
1508 
1509 /*
1510  * Acquire the requested lock and then check for validity.  LK_RETRY
1511  * permits vn_lock to return doomed vnodes.
1512  */
1513 int
1514 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1515 {
1516 	int error;
1517 
1518 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1519 	    ("vn_lock called with no locktype."));
1520 	do {
1521 #ifdef DEBUG_VFS_LOCKS
1522 		KASSERT(vp->v_holdcnt != 0,
1523 		    ("vn_lock %p: zero hold count", vp));
1524 #endif
1525 		error = VOP_LOCK1(vp, flags, file, line);
1526 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1527 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1528 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1529 		    flags, error));
1530 		/*
1531 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1532 		 * If RETRY is not set, we return ENOENT instead.
1533 		 */
1534 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1535 		    (flags & LK_RETRY) == 0) {
1536 			VOP_UNLOCK(vp, 0);
1537 			error = ENOENT;
1538 			break;
1539 		}
1540 	} while (flags & LK_RETRY && error != 0);
1541 	return (error);
1542 }
1543 
1544 /*
1545  * File table vnode close routine.
1546  */
1547 static int
1548 vn_closefile(fp, td)
1549 	struct file *fp;
1550 	struct thread *td;
1551 {
1552 	struct vnode *vp;
1553 	struct flock lf;
1554 	int error;
1555 
1556 	vp = fp->f_vnode;
1557 	fp->f_ops = &badfileops;
1558 
1559 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1560 		vref(vp);
1561 
1562 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1563 
1564 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1565 		lf.l_whence = SEEK_SET;
1566 		lf.l_start = 0;
1567 		lf.l_len = 0;
1568 		lf.l_type = F_UNLCK;
1569 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1570 		vrele(vp);
1571 	}
1572 	return (error);
1573 }
1574 
1575 static bool
1576 vn_suspendable_mp(struct mount *mp)
1577 {
1578 
1579 	return ((mp->mnt_kern_flag & MNTK_SUSPENDABLE) != 0);
1580 }
1581 
1582 static bool
1583 vn_suspendable(struct vnode *vp, struct mount **mpp)
1584 {
1585 
1586 	if (vp != NULL)
1587 		*mpp = vp->v_mount;
1588 	if (*mpp == NULL)
1589 		return (false);
1590 
1591 	return (vn_suspendable_mp(*mpp));
1592 }
1593 
1594 /*
1595  * Preparing to start a filesystem write operation. If the operation is
1596  * permitted, then we bump the count of operations in progress and
1597  * proceed. If a suspend request is in progress, we wait until the
1598  * suspension is over, and then proceed.
1599  */
1600 static int
1601 vn_start_write_locked(struct mount *mp, int flags)
1602 {
1603 	int error;
1604 
1605 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1606 	error = 0;
1607 
1608 	/*
1609 	 * Check on status of suspension.
1610 	 */
1611 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1612 	    mp->mnt_susp_owner != curthread) {
1613 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1614 			if (flags & V_NOWAIT) {
1615 				error = EWOULDBLOCK;
1616 				goto unlock;
1617 			}
1618 			error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1619 			    (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1620 			if (error)
1621 				goto unlock;
1622 		}
1623 	}
1624 	if (flags & V_XSLEEP)
1625 		goto unlock;
1626 	mp->mnt_writeopcount++;
1627 unlock:
1628 	if (error != 0 || (flags & V_XSLEEP) != 0)
1629 		MNT_REL(mp);
1630 	MNT_IUNLOCK(mp);
1631 	return (error);
1632 }
1633 
1634 int
1635 vn_start_write(vp, mpp, flags)
1636 	struct vnode *vp;
1637 	struct mount **mpp;
1638 	int flags;
1639 {
1640 	struct mount *mp;
1641 	int error;
1642 
1643 	if (!vn_suspendable(vp, mpp))
1644 		return (0);
1645 
1646 	error = 0;
1647 	/*
1648 	 * If a vnode is provided, get and return the mount point that
1649 	 * to which it will write.
1650 	 */
1651 	if (vp != NULL) {
1652 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1653 			*mpp = NULL;
1654 			if (error != EOPNOTSUPP)
1655 				return (error);
1656 			return (0);
1657 		}
1658 	}
1659 	if ((mp = *mpp) == NULL)
1660 		return (0);
1661 
1662 	/*
1663 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1664 	 * a vfs_ref().
1665 	 * As long as a vnode is not provided we need to acquire a
1666 	 * refcount for the provided mountpoint too, in order to
1667 	 * emulate a vfs_ref().
1668 	 */
1669 	MNT_ILOCK(mp);
1670 	if (vp == NULL)
1671 		MNT_REF(mp);
1672 
1673 	return (vn_start_write_locked(mp, flags));
1674 }
1675 
1676 /*
1677  * Secondary suspension. Used by operations such as vop_inactive
1678  * routines that are needed by the higher level functions. These
1679  * are allowed to proceed until all the higher level functions have
1680  * completed (indicated by mnt_writeopcount dropping to zero). At that
1681  * time, these operations are halted until the suspension is over.
1682  */
1683 int
1684 vn_start_secondary_write(vp, mpp, flags)
1685 	struct vnode *vp;
1686 	struct mount **mpp;
1687 	int flags;
1688 {
1689 	struct mount *mp;
1690 	int error;
1691 
1692 	if (!vn_suspendable(vp, mpp))
1693 		return (0);
1694 
1695  retry:
1696 	if (vp != NULL) {
1697 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1698 			*mpp = NULL;
1699 			if (error != EOPNOTSUPP)
1700 				return (error);
1701 			return (0);
1702 		}
1703 	}
1704 	/*
1705 	 * If we are not suspended or have not yet reached suspended
1706 	 * mode, then let the operation proceed.
1707 	 */
1708 	if ((mp = *mpp) == NULL)
1709 		return (0);
1710 
1711 	/*
1712 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1713 	 * a vfs_ref().
1714 	 * As long as a vnode is not provided we need to acquire a
1715 	 * refcount for the provided mountpoint too, in order to
1716 	 * emulate a vfs_ref().
1717 	 */
1718 	MNT_ILOCK(mp);
1719 	if (vp == NULL)
1720 		MNT_REF(mp);
1721 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1722 		mp->mnt_secondary_writes++;
1723 		mp->mnt_secondary_accwrites++;
1724 		MNT_IUNLOCK(mp);
1725 		return (0);
1726 	}
1727 	if (flags & V_NOWAIT) {
1728 		MNT_REL(mp);
1729 		MNT_IUNLOCK(mp);
1730 		return (EWOULDBLOCK);
1731 	}
1732 	/*
1733 	 * Wait for the suspension to finish.
1734 	 */
1735 	error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1736 		       (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1737 	vfs_rel(mp);
1738 	if (error == 0)
1739 		goto retry;
1740 	return (error);
1741 }
1742 
1743 /*
1744  * Filesystem write operation has completed. If we are suspending and this
1745  * operation is the last one, notify the suspender that the suspension is
1746  * now in effect.
1747  */
1748 void
1749 vn_finished_write(mp)
1750 	struct mount *mp;
1751 {
1752 	if (mp == NULL || !vn_suspendable_mp(mp))
1753 		return;
1754 	MNT_ILOCK(mp);
1755 	MNT_REL(mp);
1756 	mp->mnt_writeopcount--;
1757 	if (mp->mnt_writeopcount < 0)
1758 		panic("vn_finished_write: neg cnt");
1759 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1760 	    mp->mnt_writeopcount <= 0)
1761 		wakeup(&mp->mnt_writeopcount);
1762 	MNT_IUNLOCK(mp);
1763 }
1764 
1765 
1766 /*
1767  * Filesystem secondary write operation has completed. If we are
1768  * suspending and this operation is the last one, notify the suspender
1769  * that the suspension is now in effect.
1770  */
1771 void
1772 vn_finished_secondary_write(mp)
1773 	struct mount *mp;
1774 {
1775 	if (mp == NULL || !vn_suspendable_mp(mp))
1776 		return;
1777 	MNT_ILOCK(mp);
1778 	MNT_REL(mp);
1779 	mp->mnt_secondary_writes--;
1780 	if (mp->mnt_secondary_writes < 0)
1781 		panic("vn_finished_secondary_write: neg cnt");
1782 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1783 	    mp->mnt_secondary_writes <= 0)
1784 		wakeup(&mp->mnt_secondary_writes);
1785 	MNT_IUNLOCK(mp);
1786 }
1787 
1788 
1789 
1790 /*
1791  * Request a filesystem to suspend write operations.
1792  */
1793 int
1794 vfs_write_suspend(struct mount *mp, int flags)
1795 {
1796 	int error;
1797 
1798 	MPASS(vn_suspendable_mp(mp));
1799 
1800 	MNT_ILOCK(mp);
1801 	if (mp->mnt_susp_owner == curthread) {
1802 		MNT_IUNLOCK(mp);
1803 		return (EALREADY);
1804 	}
1805 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1806 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1807 
1808 	/*
1809 	 * Unmount holds a write reference on the mount point.  If we
1810 	 * own busy reference and drain for writers, we deadlock with
1811 	 * the reference draining in the unmount path.  Callers of
1812 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1813 	 * vfs_busy() reference is owned and caller is not in the
1814 	 * unmount context.
1815 	 */
1816 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1817 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1818 		MNT_IUNLOCK(mp);
1819 		return (EBUSY);
1820 	}
1821 
1822 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1823 	mp->mnt_susp_owner = curthread;
1824 	if (mp->mnt_writeopcount > 0)
1825 		(void) msleep(&mp->mnt_writeopcount,
1826 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1827 	else
1828 		MNT_IUNLOCK(mp);
1829 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1830 		vfs_write_resume(mp, 0);
1831 	return (error);
1832 }
1833 
1834 /*
1835  * Request a filesystem to resume write operations.
1836  */
1837 void
1838 vfs_write_resume(struct mount *mp, int flags)
1839 {
1840 
1841 	MPASS(vn_suspendable_mp(mp));
1842 
1843 	MNT_ILOCK(mp);
1844 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1845 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1846 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1847 				       MNTK_SUSPENDED);
1848 		mp->mnt_susp_owner = NULL;
1849 		wakeup(&mp->mnt_writeopcount);
1850 		wakeup(&mp->mnt_flag);
1851 		curthread->td_pflags &= ~TDP_IGNSUSP;
1852 		if ((flags & VR_START_WRITE) != 0) {
1853 			MNT_REF(mp);
1854 			mp->mnt_writeopcount++;
1855 		}
1856 		MNT_IUNLOCK(mp);
1857 		if ((flags & VR_NO_SUSPCLR) == 0)
1858 			VFS_SUSP_CLEAN(mp);
1859 	} else if ((flags & VR_START_WRITE) != 0) {
1860 		MNT_REF(mp);
1861 		vn_start_write_locked(mp, 0);
1862 	} else {
1863 		MNT_IUNLOCK(mp);
1864 	}
1865 }
1866 
1867 /*
1868  * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1869  * methods.
1870  */
1871 int
1872 vfs_write_suspend_umnt(struct mount *mp)
1873 {
1874 	int error;
1875 
1876 	MPASS(vn_suspendable_mp(mp));
1877 	KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1878 	    ("vfs_write_suspend_umnt: recursed"));
1879 
1880 	/* dounmount() already called vn_start_write(). */
1881 	for (;;) {
1882 		vn_finished_write(mp);
1883 		error = vfs_write_suspend(mp, 0);
1884 		if (error != 0) {
1885 			vn_start_write(NULL, &mp, V_WAIT);
1886 			return (error);
1887 		}
1888 		MNT_ILOCK(mp);
1889 		if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1890 			break;
1891 		MNT_IUNLOCK(mp);
1892 		vn_start_write(NULL, &mp, V_WAIT);
1893 	}
1894 	mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1895 	wakeup(&mp->mnt_flag);
1896 	MNT_IUNLOCK(mp);
1897 	curthread->td_pflags |= TDP_IGNSUSP;
1898 	return (0);
1899 }
1900 
1901 /*
1902  * Implement kqueues for files by translating it to vnode operation.
1903  */
1904 static int
1905 vn_kqfilter(struct file *fp, struct knote *kn)
1906 {
1907 
1908 	return (VOP_KQFILTER(fp->f_vnode, kn));
1909 }
1910 
1911 /*
1912  * Simplified in-kernel wrapper calls for extended attribute access.
1913  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1914  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1915  */
1916 int
1917 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1918     const char *attrname, int *buflen, char *buf, struct thread *td)
1919 {
1920 	struct uio	auio;
1921 	struct iovec	iov;
1922 	int	error;
1923 
1924 	iov.iov_len = *buflen;
1925 	iov.iov_base = buf;
1926 
1927 	auio.uio_iov = &iov;
1928 	auio.uio_iovcnt = 1;
1929 	auio.uio_rw = UIO_READ;
1930 	auio.uio_segflg = UIO_SYSSPACE;
1931 	auio.uio_td = td;
1932 	auio.uio_offset = 0;
1933 	auio.uio_resid = *buflen;
1934 
1935 	if ((ioflg & IO_NODELOCKED) == 0)
1936 		vn_lock(vp, LK_SHARED | LK_RETRY);
1937 
1938 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1939 
1940 	/* authorize attribute retrieval as kernel */
1941 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1942 	    td);
1943 
1944 	if ((ioflg & IO_NODELOCKED) == 0)
1945 		VOP_UNLOCK(vp, 0);
1946 
1947 	if (error == 0) {
1948 		*buflen = *buflen - auio.uio_resid;
1949 	}
1950 
1951 	return (error);
1952 }
1953 
1954 /*
1955  * XXX failure mode if partially written?
1956  */
1957 int
1958 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1959     const char *attrname, int buflen, char *buf, struct thread *td)
1960 {
1961 	struct uio	auio;
1962 	struct iovec	iov;
1963 	struct mount	*mp;
1964 	int	error;
1965 
1966 	iov.iov_len = buflen;
1967 	iov.iov_base = buf;
1968 
1969 	auio.uio_iov = &iov;
1970 	auio.uio_iovcnt = 1;
1971 	auio.uio_rw = UIO_WRITE;
1972 	auio.uio_segflg = UIO_SYSSPACE;
1973 	auio.uio_td = td;
1974 	auio.uio_offset = 0;
1975 	auio.uio_resid = buflen;
1976 
1977 	if ((ioflg & IO_NODELOCKED) == 0) {
1978 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1979 			return (error);
1980 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1981 	}
1982 
1983 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1984 
1985 	/* authorize attribute setting as kernel */
1986 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1987 
1988 	if ((ioflg & IO_NODELOCKED) == 0) {
1989 		vn_finished_write(mp);
1990 		VOP_UNLOCK(vp, 0);
1991 	}
1992 
1993 	return (error);
1994 }
1995 
1996 int
1997 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1998     const char *attrname, struct thread *td)
1999 {
2000 	struct mount	*mp;
2001 	int	error;
2002 
2003 	if ((ioflg & IO_NODELOCKED) == 0) {
2004 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2005 			return (error);
2006 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2007 	}
2008 
2009 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2010 
2011 	/* authorize attribute removal as kernel */
2012 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2013 	if (error == EOPNOTSUPP)
2014 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2015 		    NULL, td);
2016 
2017 	if ((ioflg & IO_NODELOCKED) == 0) {
2018 		vn_finished_write(mp);
2019 		VOP_UNLOCK(vp, 0);
2020 	}
2021 
2022 	return (error);
2023 }
2024 
2025 static int
2026 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2027     struct vnode **rvp)
2028 {
2029 
2030 	return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2031 }
2032 
2033 int
2034 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2035 {
2036 
2037 	return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2038 	    lkflags, rvp));
2039 }
2040 
2041 int
2042 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2043     int lkflags, struct vnode **rvp)
2044 {
2045 	struct mount *mp;
2046 	int ltype, error;
2047 
2048 	ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2049 	mp = vp->v_mount;
2050 	ltype = VOP_ISLOCKED(vp);
2051 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2052 	    ("vn_vget_ino: vp not locked"));
2053 	error = vfs_busy(mp, MBF_NOWAIT);
2054 	if (error != 0) {
2055 		vfs_ref(mp);
2056 		VOP_UNLOCK(vp, 0);
2057 		error = vfs_busy(mp, 0);
2058 		vn_lock(vp, ltype | LK_RETRY);
2059 		vfs_rel(mp);
2060 		if (error != 0)
2061 			return (ENOENT);
2062 		if (vp->v_iflag & VI_DOOMED) {
2063 			vfs_unbusy(mp);
2064 			return (ENOENT);
2065 		}
2066 	}
2067 	VOP_UNLOCK(vp, 0);
2068 	error = alloc(mp, alloc_arg, lkflags, rvp);
2069 	vfs_unbusy(mp);
2070 	if (*rvp != vp)
2071 		vn_lock(vp, ltype | LK_RETRY);
2072 	if (vp->v_iflag & VI_DOOMED) {
2073 		if (error == 0) {
2074 			if (*rvp == vp)
2075 				vunref(vp);
2076 			else
2077 				vput(*rvp);
2078 		}
2079 		error = ENOENT;
2080 	}
2081 	return (error);
2082 }
2083 
2084 int
2085 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2086     const struct thread *td)
2087 {
2088 
2089 	if (vp->v_type != VREG || td == NULL)
2090 		return (0);
2091 	PROC_LOCK(td->td_proc);
2092 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
2093 	    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
2094 		kern_psignal(td->td_proc, SIGXFSZ);
2095 		PROC_UNLOCK(td->td_proc);
2096 		return (EFBIG);
2097 	}
2098 	PROC_UNLOCK(td->td_proc);
2099 	return (0);
2100 }
2101 
2102 int
2103 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2104     struct thread *td)
2105 {
2106 	struct vnode *vp;
2107 
2108 	vp = fp->f_vnode;
2109 #ifdef AUDIT
2110 	vn_lock(vp, LK_SHARED | LK_RETRY);
2111 	AUDIT_ARG_VNODE1(vp);
2112 	VOP_UNLOCK(vp, 0);
2113 #endif
2114 	return (setfmode(td, active_cred, vp, mode));
2115 }
2116 
2117 int
2118 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2119     struct thread *td)
2120 {
2121 	struct vnode *vp;
2122 
2123 	vp = fp->f_vnode;
2124 #ifdef AUDIT
2125 	vn_lock(vp, LK_SHARED | LK_RETRY);
2126 	AUDIT_ARG_VNODE1(vp);
2127 	VOP_UNLOCK(vp, 0);
2128 #endif
2129 	return (setfown(td, active_cred, vp, uid, gid));
2130 }
2131 
2132 void
2133 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2134 {
2135 	vm_object_t object;
2136 
2137 	if ((object = vp->v_object) == NULL)
2138 		return;
2139 	VM_OBJECT_WLOCK(object);
2140 	vm_object_page_remove(object, start, end, 0);
2141 	VM_OBJECT_WUNLOCK(object);
2142 }
2143 
2144 int
2145 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2146 {
2147 	struct vattr va;
2148 	daddr_t bn, bnp;
2149 	uint64_t bsize;
2150 	off_t noff;
2151 	int error;
2152 
2153 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2154 	    ("Wrong command %lu", cmd));
2155 
2156 	if (vn_lock(vp, LK_SHARED) != 0)
2157 		return (EBADF);
2158 	if (vp->v_type != VREG) {
2159 		error = ENOTTY;
2160 		goto unlock;
2161 	}
2162 	error = VOP_GETATTR(vp, &va, cred);
2163 	if (error != 0)
2164 		goto unlock;
2165 	noff = *off;
2166 	if (noff >= va.va_size) {
2167 		error = ENXIO;
2168 		goto unlock;
2169 	}
2170 	bsize = vp->v_mount->mnt_stat.f_iosize;
2171 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2172 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2173 		if (error == EOPNOTSUPP) {
2174 			error = ENOTTY;
2175 			goto unlock;
2176 		}
2177 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2178 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
2179 			noff = bn * bsize;
2180 			if (noff < *off)
2181 				noff = *off;
2182 			goto unlock;
2183 		}
2184 	}
2185 	if (noff > va.va_size)
2186 		noff = va.va_size;
2187 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2188 	if (cmd == FIOSEEKDATA)
2189 		error = ENXIO;
2190 unlock:
2191 	VOP_UNLOCK(vp, 0);
2192 	if (error == 0)
2193 		*off = noff;
2194 	return (error);
2195 }
2196 
2197 int
2198 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2199 {
2200 	struct ucred *cred;
2201 	struct vnode *vp;
2202 	struct vattr vattr;
2203 	off_t foffset, size;
2204 	int error, noneg;
2205 
2206 	cred = td->td_ucred;
2207 	vp = fp->f_vnode;
2208 	foffset = foffset_lock(fp, 0);
2209 	noneg = (vp->v_type != VCHR);
2210 	error = 0;
2211 	switch (whence) {
2212 	case L_INCR:
2213 		if (noneg &&
2214 		    (foffset < 0 ||
2215 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2216 			error = EOVERFLOW;
2217 			break;
2218 		}
2219 		offset += foffset;
2220 		break;
2221 	case L_XTND:
2222 		vn_lock(vp, LK_SHARED | LK_RETRY);
2223 		error = VOP_GETATTR(vp, &vattr, cred);
2224 		VOP_UNLOCK(vp, 0);
2225 		if (error)
2226 			break;
2227 
2228 		/*
2229 		 * If the file references a disk device, then fetch
2230 		 * the media size and use that to determine the ending
2231 		 * offset.
2232 		 */
2233 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2234 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2235 			vattr.va_size = size;
2236 		if (noneg &&
2237 		    (vattr.va_size > OFF_MAX ||
2238 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2239 			error = EOVERFLOW;
2240 			break;
2241 		}
2242 		offset += vattr.va_size;
2243 		break;
2244 	case L_SET:
2245 		break;
2246 	case SEEK_DATA:
2247 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2248 		break;
2249 	case SEEK_HOLE:
2250 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2251 		break;
2252 	default:
2253 		error = EINVAL;
2254 	}
2255 	if (error == 0 && noneg && offset < 0)
2256 		error = EINVAL;
2257 	if (error != 0)
2258 		goto drop;
2259 	VFS_KNOTE_UNLOCKED(vp, 0);
2260 	td->td_uretoff.tdu_off = offset;
2261 drop:
2262 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2263 	return (error);
2264 }
2265 
2266 int
2267 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2268     struct thread *td)
2269 {
2270 	int error;
2271 
2272 	/*
2273 	 * Grant permission if the caller is the owner of the file, or
2274 	 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2275 	 * on the file.  If the time pointer is null, then write
2276 	 * permission on the file is also sufficient.
2277 	 *
2278 	 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2279 	 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2280 	 * will be allowed to set the times [..] to the current
2281 	 * server time.
2282 	 */
2283 	error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2284 	if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2285 		error = VOP_ACCESS(vp, VWRITE, cred, td);
2286 	return (error);
2287 }
2288 
2289 int
2290 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2291 {
2292 	struct vnode *vp;
2293 	int error;
2294 
2295 	if (fp->f_type == DTYPE_FIFO)
2296 		kif->kf_type = KF_TYPE_FIFO;
2297 	else
2298 		kif->kf_type = KF_TYPE_VNODE;
2299 	vp = fp->f_vnode;
2300 	vref(vp);
2301 	FILEDESC_SUNLOCK(fdp);
2302 	error = vn_fill_kinfo_vnode(vp, kif);
2303 	vrele(vp);
2304 	FILEDESC_SLOCK(fdp);
2305 	return (error);
2306 }
2307 
2308 int
2309 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2310 {
2311 	struct vattr va;
2312 	char *fullpath, *freepath;
2313 	int error;
2314 
2315 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2316 	freepath = NULL;
2317 	fullpath = "-";
2318 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2319 	if (error == 0) {
2320 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2321 	}
2322 	if (freepath != NULL)
2323 		free(freepath, M_TEMP);
2324 
2325 	/*
2326 	 * Retrieve vnode attributes.
2327 	 */
2328 	va.va_fsid = VNOVAL;
2329 	va.va_rdev = NODEV;
2330 	vn_lock(vp, LK_SHARED | LK_RETRY);
2331 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2332 	VOP_UNLOCK(vp, 0);
2333 	if (error != 0)
2334 		return (error);
2335 	if (va.va_fsid != VNOVAL)
2336 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2337 	else
2338 		kif->kf_un.kf_file.kf_file_fsid =
2339 		    vp->v_mount->mnt_stat.f_fsid.val[0];
2340 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2341 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2342 	kif->kf_un.kf_file.kf_file_size = va.va_size;
2343 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2344 	return (0);
2345 }
2346