xref: /freebsd/sys/kern/vfs_vnops.c (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
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 "opt_hwpmc_hooks.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/disk.h>
51 #include <sys/fail.h>
52 #include <sys/fcntl.h>
53 #include <sys/file.h>
54 #include <sys/kdb.h>
55 #include <sys/stat.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/limits.h>
59 #include <sys/lock.h>
60 #include <sys/mman.h>
61 #include <sys/mount.h>
62 #include <sys/mutex.h>
63 #include <sys/namei.h>
64 #include <sys/vnode.h>
65 #include <sys/bio.h>
66 #include <sys/buf.h>
67 #include <sys/filio.h>
68 #include <sys/resourcevar.h>
69 #include <sys/rwlock.h>
70 #include <sys/sx.h>
71 #include <sys/sysctl.h>
72 #include <sys/ttycom.h>
73 #include <sys/conf.h>
74 #include <sys/syslog.h>
75 #include <sys/unistd.h>
76 #include <sys/user.h>
77 
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80 
81 #include <vm/vm.h>
82 #include <vm/vm_extern.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vnode_pager.h>
88 
89 #ifdef HWPMC_HOOKS
90 #include <sys/pmckern.h>
91 #endif
92 
93 static fo_rdwr_t	vn_read;
94 static fo_rdwr_t	vn_write;
95 static fo_rdwr_t	vn_io_fault;
96 static fo_truncate_t	vn_truncate;
97 static fo_ioctl_t	vn_ioctl;
98 static fo_poll_t	vn_poll;
99 static fo_kqfilter_t	vn_kqfilter;
100 static fo_stat_t	vn_statfile;
101 static fo_close_t	vn_closefile;
102 static fo_mmap_t	vn_mmap;
103 
104 struct 	fileops vnops = {
105 	.fo_read = vn_io_fault,
106 	.fo_write = vn_io_fault,
107 	.fo_truncate = vn_truncate,
108 	.fo_ioctl = vn_ioctl,
109 	.fo_poll = vn_poll,
110 	.fo_kqfilter = vn_kqfilter,
111 	.fo_stat = vn_statfile,
112 	.fo_close = vn_closefile,
113 	.fo_chmod = vn_chmod,
114 	.fo_chown = vn_chown,
115 	.fo_sendfile = vn_sendfile,
116 	.fo_seek = vn_seek,
117 	.fo_fill_kinfo = vn_fill_kinfo,
118 	.fo_mmap = vn_mmap,
119 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
120 };
121 
122 static const int io_hold_cnt = 16;
123 static int vn_io_fault_enable = 1;
124 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
125     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
126 static int vn_io_fault_prefault = 0;
127 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
128     &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
129 static u_long vn_io_faults_cnt;
130 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
131     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
132 
133 /*
134  * Returns true if vn_io_fault mode of handling the i/o request should
135  * be used.
136  */
137 static bool
138 do_vn_io_fault(struct vnode *vp, struct uio *uio)
139 {
140 	struct mount *mp;
141 
142 	return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
143 	    (mp = vp->v_mount) != NULL &&
144 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
145 }
146 
147 /*
148  * Structure used to pass arguments to vn_io_fault1(), to do either
149  * file- or vnode-based I/O calls.
150  */
151 struct vn_io_fault_args {
152 	enum {
153 		VN_IO_FAULT_FOP,
154 		VN_IO_FAULT_VOP
155 	} kind;
156 	struct ucred *cred;
157 	int flags;
158 	union {
159 		struct fop_args_tag {
160 			struct file *fp;
161 			fo_rdwr_t *doio;
162 		} fop_args;
163 		struct vop_args_tag {
164 			struct vnode *vp;
165 		} vop_args;
166 	} args;
167 };
168 
169 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
170     struct vn_io_fault_args *args, struct thread *td);
171 
172 int
173 vn_open(ndp, flagp, cmode, fp)
174 	struct nameidata *ndp;
175 	int *flagp, cmode;
176 	struct file *fp;
177 {
178 	struct thread *td = ndp->ni_cnd.cn_thread;
179 
180 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
181 }
182 
183 /*
184  * Common code for vnode open operations via a name lookup.
185  * Lookup the vnode and invoke VOP_CREATE if needed.
186  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
187  *
188  * Note that this does NOT free nameidata for the successful case,
189  * due to the NDINIT being done elsewhere.
190  */
191 int
192 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
193     struct ucred *cred, struct file *fp)
194 {
195 	struct vnode *vp;
196 	struct mount *mp;
197 	struct thread *td = ndp->ni_cnd.cn_thread;
198 	struct vattr vat;
199 	struct vattr *vap = &vat;
200 	int fmode, error;
201 
202 restart:
203 	fmode = *flagp;
204 	if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
205 	    O_EXCL | O_DIRECTORY))
206 		return (EINVAL);
207 	else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
208 		ndp->ni_cnd.cn_nameiop = CREATE;
209 		/*
210 		 * Set NOCACHE to avoid flushing the cache when
211 		 * rolling in many files at once.
212 		*/
213 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
214 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
215 			ndp->ni_cnd.cn_flags |= FOLLOW;
216 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
217 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
218 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
219 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
220 		bwillwrite();
221 		if ((error = namei(ndp)) != 0)
222 			return (error);
223 		if (ndp->ni_vp == NULL) {
224 			VATTR_NULL(vap);
225 			vap->va_type = VREG;
226 			vap->va_mode = cmode;
227 			if (fmode & O_EXCL)
228 				vap->va_vaflags |= VA_EXCLUSIVE;
229 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
230 				NDFREE(ndp, NDF_ONLY_PNBUF);
231 				vput(ndp->ni_dvp);
232 				if ((error = vn_start_write(NULL, &mp,
233 				    V_XSLEEP | PCATCH)) != 0)
234 					return (error);
235 				goto restart;
236 			}
237 			if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
238 				ndp->ni_cnd.cn_flags |= MAKEENTRY;
239 #ifdef MAC
240 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
241 			    &ndp->ni_cnd, vap);
242 			if (error == 0)
243 #endif
244 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
245 						   &ndp->ni_cnd, vap);
246 			vput(ndp->ni_dvp);
247 			vn_finished_write(mp);
248 			if (error) {
249 				NDFREE(ndp, NDF_ONLY_PNBUF);
250 				return (error);
251 			}
252 			fmode &= ~O_TRUNC;
253 			vp = ndp->ni_vp;
254 		} else {
255 			if (ndp->ni_dvp == ndp->ni_vp)
256 				vrele(ndp->ni_dvp);
257 			else
258 				vput(ndp->ni_dvp);
259 			ndp->ni_dvp = NULL;
260 			vp = ndp->ni_vp;
261 			if (fmode & O_EXCL) {
262 				error = EEXIST;
263 				goto bad;
264 			}
265 			fmode &= ~O_CREAT;
266 		}
267 	} else {
268 		ndp->ni_cnd.cn_nameiop = LOOKUP;
269 		ndp->ni_cnd.cn_flags = ISOPEN |
270 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
271 		if (!(fmode & FWRITE))
272 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
273 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
274 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
275 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
276 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
277 		if ((error = namei(ndp)) != 0)
278 			return (error);
279 		vp = ndp->ni_vp;
280 	}
281 	error = vn_open_vnode(vp, fmode, cred, td, fp);
282 	if (error)
283 		goto bad;
284 	*flagp = fmode;
285 	return (0);
286 bad:
287 	NDFREE(ndp, NDF_ONLY_PNBUF);
288 	vput(vp);
289 	*flagp = fmode;
290 	ndp->ni_vp = NULL;
291 	return (error);
292 }
293 
294 /*
295  * Common code for vnode open operations once a vnode is located.
296  * Check permissions, and call the VOP_OPEN routine.
297  */
298 int
299 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
300     struct thread *td, struct file *fp)
301 {
302 	accmode_t accmode;
303 	struct flock lf;
304 	int error, lock_flags, type;
305 
306 	if (vp->v_type == VLNK)
307 		return (EMLINK);
308 	if (vp->v_type == VSOCK)
309 		return (EOPNOTSUPP);
310 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
311 		return (ENOTDIR);
312 	accmode = 0;
313 	if (fmode & (FWRITE | O_TRUNC)) {
314 		if (vp->v_type == VDIR)
315 			return (EISDIR);
316 		accmode |= VWRITE;
317 	}
318 	if (fmode & FREAD)
319 		accmode |= VREAD;
320 	if (fmode & FEXEC)
321 		accmode |= VEXEC;
322 	if ((fmode & O_APPEND) && (fmode & FWRITE))
323 		accmode |= VAPPEND;
324 #ifdef MAC
325 	if (fmode & O_CREAT)
326 		accmode |= VCREAT;
327 	if (fmode & O_VERIFY)
328 		accmode |= VVERIFY;
329 	error = mac_vnode_check_open(cred, vp, accmode);
330 	if (error)
331 		return (error);
332 
333 	accmode &= ~(VCREAT | VVERIFY);
334 #endif
335 	if ((fmode & O_CREAT) == 0) {
336 		if (accmode & VWRITE) {
337 			error = vn_writechk(vp);
338 			if (error)
339 				return (error);
340 		}
341 		if (accmode) {
342 		        error = VOP_ACCESS(vp, accmode, cred, td);
343 			if (error)
344 				return (error);
345 		}
346 	}
347 	if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
348 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
349 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
350 		return (error);
351 
352 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
353 		KASSERT(fp != NULL, ("open with flock requires fp"));
354 		lock_flags = VOP_ISLOCKED(vp);
355 		VOP_UNLOCK(vp, 0);
356 		lf.l_whence = SEEK_SET;
357 		lf.l_start = 0;
358 		lf.l_len = 0;
359 		if (fmode & O_EXLOCK)
360 			lf.l_type = F_WRLCK;
361 		else
362 			lf.l_type = F_RDLCK;
363 		type = F_FLOCK;
364 		if ((fmode & FNONBLOCK) == 0)
365 			type |= F_WAIT;
366 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
367 		if (error == 0)
368 			fp->f_flag |= FHASLOCK;
369 		vn_lock(vp, lock_flags | LK_RETRY);
370 		if (error == 0 && vp->v_iflag & VI_DOOMED)
371 			error = ENOENT;
372 
373 		/*
374 		 * Another thread might have used this vnode as an
375 		 * executable while the vnode lock was dropped.
376 		 * Ensure the vnode is still able to be opened for
377 		 * writing after the lock has been obtained.
378 		 */
379 		if (error == 0 && accmode & VWRITE)
380 			error = vn_writechk(vp);
381 
382 		if (error != 0) {
383 			fp->f_flag |= FOPENFAILED;
384 			fp->f_vnode = vp;
385 			if (fp->f_ops == &badfileops) {
386 				fp->f_type = DTYPE_VNODE;
387 				fp->f_ops = &vnops;
388 			}
389 			vref(vp);
390 		}
391 	}
392 	if (error == 0 && fmode & FWRITE) {
393 		VOP_ADD_WRITECOUNT(vp, 1);
394 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
395 		    __func__, vp, vp->v_writecount);
396 	}
397 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
398 	return (error);
399 }
400 
401 /*
402  * Check for write permissions on the specified vnode.
403  * Prototype text segments cannot be written.
404  */
405 int
406 vn_writechk(vp)
407 	register struct vnode *vp;
408 {
409 
410 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
411 	/*
412 	 * If there's shared text associated with
413 	 * the vnode, try to free it up once.  If
414 	 * we fail, we can't allow writing.
415 	 */
416 	if (VOP_IS_TEXT(vp))
417 		return (ETXTBSY);
418 
419 	return (0);
420 }
421 
422 /*
423  * Vnode close call
424  */
425 int
426 vn_close(vp, flags, file_cred, td)
427 	register struct vnode *vp;
428 	int flags;
429 	struct ucred *file_cred;
430 	struct thread *td;
431 {
432 	struct mount *mp;
433 	int error, lock_flags;
434 
435 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
436 	    MNT_EXTENDED_SHARED(vp->v_mount))
437 		lock_flags = LK_SHARED;
438 	else
439 		lock_flags = LK_EXCLUSIVE;
440 
441 	vn_start_write(vp, &mp, V_WAIT);
442 	vn_lock(vp, lock_flags | LK_RETRY);
443 	if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
444 		VNASSERT(vp->v_writecount > 0, vp,
445 		    ("vn_close: negative writecount"));
446 		VOP_ADD_WRITECOUNT(vp, -1);
447 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
448 		    __func__, vp, vp->v_writecount);
449 	}
450 	error = VOP_CLOSE(vp, flags, file_cred, td);
451 	vput(vp);
452 	vn_finished_write(mp);
453 	return (error);
454 }
455 
456 /*
457  * Heuristic to detect sequential operation.
458  */
459 static int
460 sequential_heuristic(struct uio *uio, struct file *fp)
461 {
462 
463 	ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
464 	if (fp->f_flag & FRDAHEAD)
465 		return (fp->f_seqcount << IO_SEQSHIFT);
466 
467 	/*
468 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
469 	 * that the first I/O is normally considered to be slightly
470 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
471 	 * unless previous seeks have reduced f_seqcount to 0, in which
472 	 * case offset 0 is not special.
473 	 */
474 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
475 	    uio->uio_offset == fp->f_nextoff) {
476 		/*
477 		 * f_seqcount is in units of fixed-size blocks so that it
478 		 * depends mainly on the amount of sequential I/O and not
479 		 * much on the number of sequential I/O's.  The fixed size
480 		 * of 16384 is hard-coded here since it is (not quite) just
481 		 * a magic size that works well here.  This size is more
482 		 * closely related to the best I/O size for real disks than
483 		 * to any block size used by software.
484 		 */
485 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
486 		if (fp->f_seqcount > IO_SEQMAX)
487 			fp->f_seqcount = IO_SEQMAX;
488 		return (fp->f_seqcount << IO_SEQSHIFT);
489 	}
490 
491 	/* Not sequential.  Quickly draw-down sequentiality. */
492 	if (fp->f_seqcount > 1)
493 		fp->f_seqcount = 1;
494 	else
495 		fp->f_seqcount = 0;
496 	return (0);
497 }
498 
499 /*
500  * Package up an I/O request on a vnode into a uio and do it.
501  */
502 int
503 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
504     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
505     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
506 {
507 	struct uio auio;
508 	struct iovec aiov;
509 	struct mount *mp;
510 	struct ucred *cred;
511 	void *rl_cookie;
512 	struct vn_io_fault_args args;
513 	int error, lock_flags;
514 
515 	auio.uio_iov = &aiov;
516 	auio.uio_iovcnt = 1;
517 	aiov.iov_base = base;
518 	aiov.iov_len = len;
519 	auio.uio_resid = len;
520 	auio.uio_offset = offset;
521 	auio.uio_segflg = segflg;
522 	auio.uio_rw = rw;
523 	auio.uio_td = td;
524 	error = 0;
525 
526 	if ((ioflg & IO_NODELOCKED) == 0) {
527 		if ((ioflg & IO_RANGELOCKED) == 0) {
528 			if (rw == UIO_READ) {
529 				rl_cookie = vn_rangelock_rlock(vp, offset,
530 				    offset + len);
531 			} else {
532 				rl_cookie = vn_rangelock_wlock(vp, offset,
533 				    offset + len);
534 			}
535 		} else
536 			rl_cookie = NULL;
537 		mp = NULL;
538 		if (rw == UIO_WRITE) {
539 			if (vp->v_type != VCHR &&
540 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
541 			    != 0)
542 				goto out;
543 			if (MNT_SHARED_WRITES(mp) ||
544 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
545 				lock_flags = LK_SHARED;
546 			else
547 				lock_flags = LK_EXCLUSIVE;
548 		} else
549 			lock_flags = LK_SHARED;
550 		vn_lock(vp, lock_flags | LK_RETRY);
551 	} else
552 		rl_cookie = NULL;
553 
554 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
555 #ifdef MAC
556 	if ((ioflg & IO_NOMACCHECK) == 0) {
557 		if (rw == UIO_READ)
558 			error = mac_vnode_check_read(active_cred, file_cred,
559 			    vp);
560 		else
561 			error = mac_vnode_check_write(active_cred, file_cred,
562 			    vp);
563 	}
564 #endif
565 	if (error == 0) {
566 		if (file_cred != NULL)
567 			cred = file_cred;
568 		else
569 			cred = active_cred;
570 		if (do_vn_io_fault(vp, &auio)) {
571 			args.kind = VN_IO_FAULT_VOP;
572 			args.cred = cred;
573 			args.flags = ioflg;
574 			args.args.vop_args.vp = vp;
575 			error = vn_io_fault1(vp, &auio, &args, td);
576 		} else if (rw == UIO_READ) {
577 			error = VOP_READ(vp, &auio, ioflg, cred);
578 		} else /* if (rw == UIO_WRITE) */ {
579 			error = VOP_WRITE(vp, &auio, ioflg, cred);
580 		}
581 	}
582 	if (aresid)
583 		*aresid = auio.uio_resid;
584 	else
585 		if (auio.uio_resid && error == 0)
586 			error = EIO;
587 	if ((ioflg & IO_NODELOCKED) == 0) {
588 		VOP_UNLOCK(vp, 0);
589 		if (mp != NULL)
590 			vn_finished_write(mp);
591 	}
592  out:
593 	if (rl_cookie != NULL)
594 		vn_rangelock_unlock(vp, rl_cookie);
595 	return (error);
596 }
597 
598 /*
599  * Package up an I/O request on a vnode into a uio and do it.  The I/O
600  * request is split up into smaller chunks and we try to avoid saturating
601  * the buffer cache while potentially holding a vnode locked, so we
602  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
603  * to give other processes a chance to lock the vnode (either other processes
604  * core'ing the same binary, or unrelated processes scanning the directory).
605  */
606 int
607 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
608     file_cred, aresid, td)
609 	enum uio_rw rw;
610 	struct vnode *vp;
611 	void *base;
612 	size_t len;
613 	off_t offset;
614 	enum uio_seg segflg;
615 	int ioflg;
616 	struct ucred *active_cred;
617 	struct ucred *file_cred;
618 	size_t *aresid;
619 	struct thread *td;
620 {
621 	int error = 0;
622 	ssize_t iaresid;
623 
624 	do {
625 		int chunk;
626 
627 		/*
628 		 * Force `offset' to a multiple of MAXBSIZE except possibly
629 		 * for the first chunk, so that filesystems only need to
630 		 * write full blocks except possibly for the first and last
631 		 * chunks.
632 		 */
633 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
634 
635 		if (chunk > len)
636 			chunk = len;
637 		if (rw != UIO_READ && vp->v_type == VREG)
638 			bwillwrite();
639 		iaresid = 0;
640 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
641 		    ioflg, active_cred, file_cred, &iaresid, td);
642 		len -= chunk;	/* aresid calc already includes length */
643 		if (error)
644 			break;
645 		offset += chunk;
646 		base = (char *)base + chunk;
647 		kern_yield(PRI_USER);
648 	} while (len);
649 	if (aresid)
650 		*aresid = len + iaresid;
651 	return (error);
652 }
653 
654 off_t
655 foffset_lock(struct file *fp, int flags)
656 {
657 	struct mtx *mtxp;
658 	off_t res;
659 
660 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
661 
662 #if OFF_MAX <= LONG_MAX
663 	/*
664 	 * Caller only wants the current f_offset value.  Assume that
665 	 * the long and shorter integer types reads are atomic.
666 	 */
667 	if ((flags & FOF_NOLOCK) != 0)
668 		return (fp->f_offset);
669 #endif
670 
671 	/*
672 	 * According to McKusick the vn lock was protecting f_offset here.
673 	 * It is now protected by the FOFFSET_LOCKED flag.
674 	 */
675 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
676 	mtx_lock(mtxp);
677 	if ((flags & FOF_NOLOCK) == 0) {
678 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
679 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
680 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
681 			    "vofflock", 0);
682 		}
683 		fp->f_vnread_flags |= FOFFSET_LOCKED;
684 	}
685 	res = fp->f_offset;
686 	mtx_unlock(mtxp);
687 	return (res);
688 }
689 
690 void
691 foffset_unlock(struct file *fp, off_t val, int flags)
692 {
693 	struct mtx *mtxp;
694 
695 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
696 
697 #if OFF_MAX <= LONG_MAX
698 	if ((flags & FOF_NOLOCK) != 0) {
699 		if ((flags & FOF_NOUPDATE) == 0)
700 			fp->f_offset = val;
701 		if ((flags & FOF_NEXTOFF) != 0)
702 			fp->f_nextoff = val;
703 		return;
704 	}
705 #endif
706 
707 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
708 	mtx_lock(mtxp);
709 	if ((flags & FOF_NOUPDATE) == 0)
710 		fp->f_offset = val;
711 	if ((flags & FOF_NEXTOFF) != 0)
712 		fp->f_nextoff = val;
713 	if ((flags & FOF_NOLOCK) == 0) {
714 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
715 		    ("Lost FOFFSET_LOCKED"));
716 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
717 			wakeup(&fp->f_vnread_flags);
718 		fp->f_vnread_flags = 0;
719 	}
720 	mtx_unlock(mtxp);
721 }
722 
723 void
724 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
725 {
726 
727 	if ((flags & FOF_OFFSET) == 0)
728 		uio->uio_offset = foffset_lock(fp, flags);
729 }
730 
731 void
732 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
733 {
734 
735 	if ((flags & FOF_OFFSET) == 0)
736 		foffset_unlock(fp, uio->uio_offset, flags);
737 }
738 
739 static int
740 get_advice(struct file *fp, struct uio *uio)
741 {
742 	struct mtx *mtxp;
743 	int ret;
744 
745 	ret = POSIX_FADV_NORMAL;
746 	if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
747 		return (ret);
748 
749 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
750 	mtx_lock(mtxp);
751 	if (fp->f_advice != NULL &&
752 	    uio->uio_offset >= fp->f_advice->fa_start &&
753 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
754 		ret = fp->f_advice->fa_advice;
755 	mtx_unlock(mtxp);
756 	return (ret);
757 }
758 
759 /*
760  * File table vnode read routine.
761  */
762 static int
763 vn_read(fp, uio, active_cred, flags, td)
764 	struct file *fp;
765 	struct uio *uio;
766 	struct ucred *active_cred;
767 	int flags;
768 	struct thread *td;
769 {
770 	struct vnode *vp;
771 	off_t orig_offset;
772 	int error, ioflag;
773 	int advice;
774 
775 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
776 	    uio->uio_td, td));
777 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
778 	vp = fp->f_vnode;
779 	ioflag = 0;
780 	if (fp->f_flag & FNONBLOCK)
781 		ioflag |= IO_NDELAY;
782 	if (fp->f_flag & O_DIRECT)
783 		ioflag |= IO_DIRECT;
784 	advice = get_advice(fp, uio);
785 	vn_lock(vp, LK_SHARED | LK_RETRY);
786 
787 	switch (advice) {
788 	case POSIX_FADV_NORMAL:
789 	case POSIX_FADV_SEQUENTIAL:
790 	case POSIX_FADV_NOREUSE:
791 		ioflag |= sequential_heuristic(uio, fp);
792 		break;
793 	case POSIX_FADV_RANDOM:
794 		/* Disable read-ahead for random I/O. */
795 		break;
796 	}
797 	orig_offset = uio->uio_offset;
798 
799 #ifdef MAC
800 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
801 	if (error == 0)
802 #endif
803 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
804 	fp->f_nextoff = uio->uio_offset;
805 	VOP_UNLOCK(vp, 0);
806 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
807 	    orig_offset != uio->uio_offset)
808 		/*
809 		 * Use POSIX_FADV_DONTNEED to flush pages and buffers
810 		 * for the backing file after a POSIX_FADV_NOREUSE
811 		 * read(2).
812 		 */
813 		error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
814 		    POSIX_FADV_DONTNEED);
815 	return (error);
816 }
817 
818 /*
819  * File table vnode write routine.
820  */
821 static int
822 vn_write(fp, uio, active_cred, flags, td)
823 	struct file *fp;
824 	struct uio *uio;
825 	struct ucred *active_cred;
826 	int flags;
827 	struct thread *td;
828 {
829 	struct vnode *vp;
830 	struct mount *mp;
831 	off_t orig_offset;
832 	int error, ioflag, lock_flags;
833 	int advice;
834 
835 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
836 	    uio->uio_td, td));
837 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
838 	vp = fp->f_vnode;
839 	if (vp->v_type == VREG)
840 		bwillwrite();
841 	ioflag = IO_UNIT;
842 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
843 		ioflag |= IO_APPEND;
844 	if (fp->f_flag & FNONBLOCK)
845 		ioflag |= IO_NDELAY;
846 	if (fp->f_flag & O_DIRECT)
847 		ioflag |= IO_DIRECT;
848 	if ((fp->f_flag & O_FSYNC) ||
849 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
850 		ioflag |= IO_SYNC;
851 	mp = NULL;
852 	if (vp->v_type != VCHR &&
853 	    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
854 		goto unlock;
855 
856 	advice = get_advice(fp, uio);
857 
858 	if (MNT_SHARED_WRITES(mp) ||
859 	    (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
860 		lock_flags = LK_SHARED;
861 	} else {
862 		lock_flags = LK_EXCLUSIVE;
863 	}
864 
865 	vn_lock(vp, lock_flags | LK_RETRY);
866 	switch (advice) {
867 	case POSIX_FADV_NORMAL:
868 	case POSIX_FADV_SEQUENTIAL:
869 	case POSIX_FADV_NOREUSE:
870 		ioflag |= sequential_heuristic(uio, fp);
871 		break;
872 	case POSIX_FADV_RANDOM:
873 		/* XXX: Is this correct? */
874 		break;
875 	}
876 	orig_offset = uio->uio_offset;
877 
878 #ifdef MAC
879 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
880 	if (error == 0)
881 #endif
882 		error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
883 	fp->f_nextoff = uio->uio_offset;
884 	VOP_UNLOCK(vp, 0);
885 	if (vp->v_type != VCHR)
886 		vn_finished_write(mp);
887 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
888 	    orig_offset != uio->uio_offset)
889 		/*
890 		 * Use POSIX_FADV_DONTNEED to flush pages and buffers
891 		 * for the backing file after a POSIX_FADV_NOREUSE
892 		 * write(2).
893 		 */
894 		error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
895 		    POSIX_FADV_DONTNEED);
896 unlock:
897 	return (error);
898 }
899 
900 /*
901  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
902  * prevent the following deadlock:
903  *
904  * Assume that the thread A reads from the vnode vp1 into userspace
905  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
906  * currently not resident, then system ends up with the call chain
907  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
908  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
909  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
910  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
911  * backed by the pages of vnode vp1, and some page in buf2 is not
912  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
913  *
914  * To prevent the lock order reversal and deadlock, vn_io_fault() does
915  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
916  * Instead, it first tries to do the whole range i/o with pagefaults
917  * disabled. If all pages in the i/o buffer are resident and mapped,
918  * VOP will succeed (ignoring the genuine filesystem errors).
919  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
920  * i/o in chunks, with all pages in the chunk prefaulted and held
921  * using vm_fault_quick_hold_pages().
922  *
923  * Filesystems using this deadlock avoidance scheme should use the
924  * array of the held pages from uio, saved in the curthread->td_ma,
925  * instead of doing uiomove().  A helper function
926  * vn_io_fault_uiomove() converts uiomove request into
927  * uiomove_fromphys() over td_ma array.
928  *
929  * Since vnode locks do not cover the whole i/o anymore, rangelocks
930  * make the current i/o request atomic with respect to other i/os and
931  * truncations.
932  */
933 
934 /*
935  * Decode vn_io_fault_args and perform the corresponding i/o.
936  */
937 static int
938 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
939     struct thread *td)
940 {
941 
942 	switch (args->kind) {
943 	case VN_IO_FAULT_FOP:
944 		return ((args->args.fop_args.doio)(args->args.fop_args.fp,
945 		    uio, args->cred, args->flags, td));
946 	case VN_IO_FAULT_VOP:
947 		if (uio->uio_rw == UIO_READ) {
948 			return (VOP_READ(args->args.vop_args.vp, uio,
949 			    args->flags, args->cred));
950 		} else if (uio->uio_rw == UIO_WRITE) {
951 			return (VOP_WRITE(args->args.vop_args.vp, uio,
952 			    args->flags, args->cred));
953 		}
954 		break;
955 	}
956 	panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
957 	    uio->uio_rw);
958 }
959 
960 static int
961 vn_io_fault_touch(char *base, const struct uio *uio)
962 {
963 	int r;
964 
965 	r = fubyte(base);
966 	if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
967 		return (EFAULT);
968 	return (0);
969 }
970 
971 static int
972 vn_io_fault_prefault_user(const struct uio *uio)
973 {
974 	char *base;
975 	const struct iovec *iov;
976 	size_t len;
977 	ssize_t resid;
978 	int error, i;
979 
980 	KASSERT(uio->uio_segflg == UIO_USERSPACE,
981 	    ("vn_io_fault_prefault userspace"));
982 
983 	error = i = 0;
984 	iov = uio->uio_iov;
985 	resid = uio->uio_resid;
986 	base = iov->iov_base;
987 	len = iov->iov_len;
988 	while (resid > 0) {
989 		error = vn_io_fault_touch(base, uio);
990 		if (error != 0)
991 			break;
992 		if (len < PAGE_SIZE) {
993 			if (len != 0) {
994 				error = vn_io_fault_touch(base + len - 1, uio);
995 				if (error != 0)
996 					break;
997 				resid -= len;
998 			}
999 			if (++i >= uio->uio_iovcnt)
1000 				break;
1001 			iov = uio->uio_iov + i;
1002 			base = iov->iov_base;
1003 			len = iov->iov_len;
1004 		} else {
1005 			len -= PAGE_SIZE;
1006 			base += PAGE_SIZE;
1007 			resid -= PAGE_SIZE;
1008 		}
1009 	}
1010 	return (error);
1011 }
1012 
1013 /*
1014  * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1015  * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1016  * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1017  * into args and call vn_io_fault1() to handle faults during the user
1018  * mode buffer accesses.
1019  */
1020 static int
1021 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1022     struct thread *td)
1023 {
1024 	vm_page_t ma[io_hold_cnt + 2];
1025 	struct uio *uio_clone, short_uio;
1026 	struct iovec short_iovec[1];
1027 	vm_page_t *prev_td_ma;
1028 	vm_prot_t prot;
1029 	vm_offset_t addr, end;
1030 	size_t len, resid;
1031 	ssize_t adv;
1032 	int error, cnt, save, saveheld, prev_td_ma_cnt;
1033 
1034 	if (vn_io_fault_prefault) {
1035 		error = vn_io_fault_prefault_user(uio);
1036 		if (error != 0)
1037 			return (error); /* Or ignore ? */
1038 	}
1039 
1040 	prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1041 
1042 	/*
1043 	 * The UFS follows IO_UNIT directive and replays back both
1044 	 * uio_offset and uio_resid if an error is encountered during the
1045 	 * operation.  But, since the iovec may be already advanced,
1046 	 * uio is still in an inconsistent state.
1047 	 *
1048 	 * Cache a copy of the original uio, which is advanced to the redo
1049 	 * point using UIO_NOCOPY below.
1050 	 */
1051 	uio_clone = cloneuio(uio);
1052 	resid = uio->uio_resid;
1053 
1054 	short_uio.uio_segflg = UIO_USERSPACE;
1055 	short_uio.uio_rw = uio->uio_rw;
1056 	short_uio.uio_td = uio->uio_td;
1057 
1058 	save = vm_fault_disable_pagefaults();
1059 	error = vn_io_fault_doio(args, uio, td);
1060 	if (error != EFAULT)
1061 		goto out;
1062 
1063 	atomic_add_long(&vn_io_faults_cnt, 1);
1064 	uio_clone->uio_segflg = UIO_NOCOPY;
1065 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
1066 	uio_clone->uio_segflg = uio->uio_segflg;
1067 
1068 	saveheld = curthread_pflags_set(TDP_UIOHELD);
1069 	prev_td_ma = td->td_ma;
1070 	prev_td_ma_cnt = td->td_ma_cnt;
1071 
1072 	while (uio_clone->uio_resid != 0) {
1073 		len = uio_clone->uio_iov->iov_len;
1074 		if (len == 0) {
1075 			KASSERT(uio_clone->uio_iovcnt >= 1,
1076 			    ("iovcnt underflow"));
1077 			uio_clone->uio_iov++;
1078 			uio_clone->uio_iovcnt--;
1079 			continue;
1080 		}
1081 		if (len > io_hold_cnt * PAGE_SIZE)
1082 			len = io_hold_cnt * PAGE_SIZE;
1083 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1084 		end = round_page(addr + len);
1085 		if (end < addr) {
1086 			error = EFAULT;
1087 			break;
1088 		}
1089 		cnt = atop(end - trunc_page(addr));
1090 		/*
1091 		 * A perfectly misaligned address and length could cause
1092 		 * both the start and the end of the chunk to use partial
1093 		 * page.  +2 accounts for such a situation.
1094 		 */
1095 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1096 		    addr, len, prot, ma, io_hold_cnt + 2);
1097 		if (cnt == -1) {
1098 			error = EFAULT;
1099 			break;
1100 		}
1101 		short_uio.uio_iov = &short_iovec[0];
1102 		short_iovec[0].iov_base = (void *)addr;
1103 		short_uio.uio_iovcnt = 1;
1104 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1105 		short_uio.uio_offset = uio_clone->uio_offset;
1106 		td->td_ma = ma;
1107 		td->td_ma_cnt = cnt;
1108 
1109 		error = vn_io_fault_doio(args, &short_uio, td);
1110 		vm_page_unhold_pages(ma, cnt);
1111 		adv = len - short_uio.uio_resid;
1112 
1113 		uio_clone->uio_iov->iov_base =
1114 		    (char *)uio_clone->uio_iov->iov_base + adv;
1115 		uio_clone->uio_iov->iov_len -= adv;
1116 		uio_clone->uio_resid -= adv;
1117 		uio_clone->uio_offset += adv;
1118 
1119 		uio->uio_resid -= adv;
1120 		uio->uio_offset += adv;
1121 
1122 		if (error != 0 || adv == 0)
1123 			break;
1124 	}
1125 	td->td_ma = prev_td_ma;
1126 	td->td_ma_cnt = prev_td_ma_cnt;
1127 	curthread_pflags_restore(saveheld);
1128 out:
1129 	vm_fault_enable_pagefaults(save);
1130 	free(uio_clone, M_IOV);
1131 	return (error);
1132 }
1133 
1134 static int
1135 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1136     int flags, struct thread *td)
1137 {
1138 	fo_rdwr_t *doio;
1139 	struct vnode *vp;
1140 	void *rl_cookie;
1141 	struct vn_io_fault_args args;
1142 	int error;
1143 
1144 	doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1145 	vp = fp->f_vnode;
1146 	foffset_lock_uio(fp, uio, flags);
1147 	if (do_vn_io_fault(vp, uio)) {
1148 		args.kind = VN_IO_FAULT_FOP;
1149 		args.args.fop_args.fp = fp;
1150 		args.args.fop_args.doio = doio;
1151 		args.cred = active_cred;
1152 		args.flags = flags | FOF_OFFSET;
1153 		if (uio->uio_rw == UIO_READ) {
1154 			rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1155 			    uio->uio_offset + uio->uio_resid);
1156 		} else if ((fp->f_flag & O_APPEND) != 0 ||
1157 		    (flags & FOF_OFFSET) == 0) {
1158 			/* For appenders, punt and lock the whole range. */
1159 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1160 		} else {
1161 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1162 			    uio->uio_offset + uio->uio_resid);
1163 		}
1164 		error = vn_io_fault1(vp, uio, &args, td);
1165 		vn_rangelock_unlock(vp, rl_cookie);
1166 	} else {
1167 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1168 	}
1169 	foffset_unlock_uio(fp, uio, flags);
1170 	return (error);
1171 }
1172 
1173 /*
1174  * Helper function to perform the requested uiomove operation using
1175  * the held pages for io->uio_iov[0].iov_base buffer instead of
1176  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1177  * instead of iov_base prevents page faults that could occur due to
1178  * pmap_collect() invalidating the mapping created by
1179  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1180  * object cleanup revoking the write access from page mappings.
1181  *
1182  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1183  * instead of plain uiomove().
1184  */
1185 int
1186 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1187 {
1188 	struct uio transp_uio;
1189 	struct iovec transp_iov[1];
1190 	struct thread *td;
1191 	size_t adv;
1192 	int error, pgadv;
1193 
1194 	td = curthread;
1195 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1196 	    uio->uio_segflg != UIO_USERSPACE)
1197 		return (uiomove(data, xfersize, uio));
1198 
1199 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1200 	transp_iov[0].iov_base = data;
1201 	transp_uio.uio_iov = &transp_iov[0];
1202 	transp_uio.uio_iovcnt = 1;
1203 	if (xfersize > uio->uio_resid)
1204 		xfersize = uio->uio_resid;
1205 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1206 	transp_uio.uio_offset = 0;
1207 	transp_uio.uio_segflg = UIO_SYSSPACE;
1208 	/*
1209 	 * Since transp_iov points to data, and td_ma page array
1210 	 * corresponds to original uio->uio_iov, we need to invert the
1211 	 * direction of the i/o operation as passed to
1212 	 * uiomove_fromphys().
1213 	 */
1214 	switch (uio->uio_rw) {
1215 	case UIO_WRITE:
1216 		transp_uio.uio_rw = UIO_READ;
1217 		break;
1218 	case UIO_READ:
1219 		transp_uio.uio_rw = UIO_WRITE;
1220 		break;
1221 	}
1222 	transp_uio.uio_td = uio->uio_td;
1223 	error = uiomove_fromphys(td->td_ma,
1224 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1225 	    xfersize, &transp_uio);
1226 	adv = xfersize - transp_uio.uio_resid;
1227 	pgadv =
1228 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1229 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1230 	td->td_ma += pgadv;
1231 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1232 	    pgadv));
1233 	td->td_ma_cnt -= pgadv;
1234 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1235 	uio->uio_iov->iov_len -= adv;
1236 	uio->uio_resid -= adv;
1237 	uio->uio_offset += adv;
1238 	return (error);
1239 }
1240 
1241 int
1242 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1243     struct uio *uio)
1244 {
1245 	struct thread *td;
1246 	vm_offset_t iov_base;
1247 	int cnt, pgadv;
1248 
1249 	td = curthread;
1250 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1251 	    uio->uio_segflg != UIO_USERSPACE)
1252 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1253 
1254 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1255 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1256 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1257 	switch (uio->uio_rw) {
1258 	case UIO_WRITE:
1259 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1260 		    offset, cnt);
1261 		break;
1262 	case UIO_READ:
1263 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1264 		    cnt);
1265 		break;
1266 	}
1267 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1268 	td->td_ma += pgadv;
1269 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1270 	    pgadv));
1271 	td->td_ma_cnt -= pgadv;
1272 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1273 	uio->uio_iov->iov_len -= cnt;
1274 	uio->uio_resid -= cnt;
1275 	uio->uio_offset += cnt;
1276 	return (0);
1277 }
1278 
1279 
1280 /*
1281  * File table truncate routine.
1282  */
1283 static int
1284 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1285     struct thread *td)
1286 {
1287 	struct vattr vattr;
1288 	struct mount *mp;
1289 	struct vnode *vp;
1290 	void *rl_cookie;
1291 	int error;
1292 
1293 	vp = fp->f_vnode;
1294 
1295 	/*
1296 	 * Lock the whole range for truncation.  Otherwise split i/o
1297 	 * might happen partly before and partly after the truncation.
1298 	 */
1299 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1300 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1301 	if (error)
1302 		goto out1;
1303 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1304 	if (vp->v_type == VDIR) {
1305 		error = EISDIR;
1306 		goto out;
1307 	}
1308 #ifdef MAC
1309 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1310 	if (error)
1311 		goto out;
1312 #endif
1313 	error = vn_writechk(vp);
1314 	if (error == 0) {
1315 		VATTR_NULL(&vattr);
1316 		vattr.va_size = length;
1317 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1318 	}
1319 out:
1320 	VOP_UNLOCK(vp, 0);
1321 	vn_finished_write(mp);
1322 out1:
1323 	vn_rangelock_unlock(vp, rl_cookie);
1324 	return (error);
1325 }
1326 
1327 /*
1328  * File table vnode stat routine.
1329  */
1330 static int
1331 vn_statfile(fp, sb, active_cred, td)
1332 	struct file *fp;
1333 	struct stat *sb;
1334 	struct ucred *active_cred;
1335 	struct thread *td;
1336 {
1337 	struct vnode *vp = fp->f_vnode;
1338 	int error;
1339 
1340 	vn_lock(vp, LK_SHARED | LK_RETRY);
1341 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1342 	VOP_UNLOCK(vp, 0);
1343 
1344 	return (error);
1345 }
1346 
1347 /*
1348  * Stat a vnode; implementation for the stat syscall
1349  */
1350 int
1351 vn_stat(vp, sb, active_cred, file_cred, td)
1352 	struct vnode *vp;
1353 	register struct stat *sb;
1354 	struct ucred *active_cred;
1355 	struct ucred *file_cred;
1356 	struct thread *td;
1357 {
1358 	struct vattr vattr;
1359 	register struct vattr *vap;
1360 	int error;
1361 	u_short mode;
1362 
1363 #ifdef MAC
1364 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1365 	if (error)
1366 		return (error);
1367 #endif
1368 
1369 	vap = &vattr;
1370 
1371 	/*
1372 	 * Initialize defaults for new and unusual fields, so that file
1373 	 * systems which don't support these fields don't need to know
1374 	 * about them.
1375 	 */
1376 	vap->va_birthtime.tv_sec = -1;
1377 	vap->va_birthtime.tv_nsec = 0;
1378 	vap->va_fsid = VNOVAL;
1379 	vap->va_rdev = NODEV;
1380 
1381 	error = VOP_GETATTR(vp, vap, active_cred);
1382 	if (error)
1383 		return (error);
1384 
1385 	/*
1386 	 * Zero the spare stat fields
1387 	 */
1388 	bzero(sb, sizeof *sb);
1389 
1390 	/*
1391 	 * Copy from vattr table
1392 	 */
1393 	if (vap->va_fsid != VNOVAL)
1394 		sb->st_dev = vap->va_fsid;
1395 	else
1396 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1397 	sb->st_ino = vap->va_fileid;
1398 	mode = vap->va_mode;
1399 	switch (vap->va_type) {
1400 	case VREG:
1401 		mode |= S_IFREG;
1402 		break;
1403 	case VDIR:
1404 		mode |= S_IFDIR;
1405 		break;
1406 	case VBLK:
1407 		mode |= S_IFBLK;
1408 		break;
1409 	case VCHR:
1410 		mode |= S_IFCHR;
1411 		break;
1412 	case VLNK:
1413 		mode |= S_IFLNK;
1414 		break;
1415 	case VSOCK:
1416 		mode |= S_IFSOCK;
1417 		break;
1418 	case VFIFO:
1419 		mode |= S_IFIFO;
1420 		break;
1421 	default:
1422 		return (EBADF);
1423 	};
1424 	sb->st_mode = mode;
1425 	sb->st_nlink = vap->va_nlink;
1426 	sb->st_uid = vap->va_uid;
1427 	sb->st_gid = vap->va_gid;
1428 	sb->st_rdev = vap->va_rdev;
1429 	if (vap->va_size > OFF_MAX)
1430 		return (EOVERFLOW);
1431 	sb->st_size = vap->va_size;
1432 	sb->st_atim = vap->va_atime;
1433 	sb->st_mtim = vap->va_mtime;
1434 	sb->st_ctim = vap->va_ctime;
1435 	sb->st_birthtim = vap->va_birthtime;
1436 
1437         /*
1438 	 * According to www.opengroup.org, the meaning of st_blksize is
1439 	 *   "a filesystem-specific preferred I/O block size for this
1440 	 *    object.  In some filesystem types, this may vary from file
1441 	 *    to file"
1442 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1443 	 */
1444 
1445 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1446 
1447 	sb->st_flags = vap->va_flags;
1448 	if (priv_check(td, PRIV_VFS_GENERATION))
1449 		sb->st_gen = 0;
1450 	else
1451 		sb->st_gen = vap->va_gen;
1452 
1453 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1454 	return (0);
1455 }
1456 
1457 /*
1458  * File table vnode ioctl routine.
1459  */
1460 static int
1461 vn_ioctl(fp, com, data, active_cred, td)
1462 	struct file *fp;
1463 	u_long com;
1464 	void *data;
1465 	struct ucred *active_cred;
1466 	struct thread *td;
1467 {
1468 	struct vattr vattr;
1469 	struct vnode *vp;
1470 	int error;
1471 
1472 	vp = fp->f_vnode;
1473 	switch (vp->v_type) {
1474 	case VDIR:
1475 	case VREG:
1476 		switch (com) {
1477 		case FIONREAD:
1478 			vn_lock(vp, LK_SHARED | LK_RETRY);
1479 			error = VOP_GETATTR(vp, &vattr, active_cred);
1480 			VOP_UNLOCK(vp, 0);
1481 			if (error == 0)
1482 				*(int *)data = vattr.va_size - fp->f_offset;
1483 			return (error);
1484 		case FIONBIO:
1485 		case FIOASYNC:
1486 			return (0);
1487 		default:
1488 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1489 			    active_cred, td));
1490 		}
1491 	default:
1492 		return (ENOTTY);
1493 	}
1494 }
1495 
1496 /*
1497  * File table vnode poll routine.
1498  */
1499 static int
1500 vn_poll(fp, events, active_cred, td)
1501 	struct file *fp;
1502 	int events;
1503 	struct ucred *active_cred;
1504 	struct thread *td;
1505 {
1506 	struct vnode *vp;
1507 	int error;
1508 
1509 	vp = fp->f_vnode;
1510 #ifdef MAC
1511 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1512 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1513 	VOP_UNLOCK(vp, 0);
1514 	if (!error)
1515 #endif
1516 
1517 	error = VOP_POLL(vp, events, fp->f_cred, td);
1518 	return (error);
1519 }
1520 
1521 /*
1522  * Acquire the requested lock and then check for validity.  LK_RETRY
1523  * permits vn_lock to return doomed vnodes.
1524  */
1525 int
1526 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1527 {
1528 	int error;
1529 
1530 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1531 	    ("vn_lock called with no locktype."));
1532 	do {
1533 #ifdef DEBUG_VFS_LOCKS
1534 		KASSERT(vp->v_holdcnt != 0,
1535 		    ("vn_lock %p: zero hold count", vp));
1536 #endif
1537 		error = VOP_LOCK1(vp, flags, file, line);
1538 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1539 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1540 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1541 		    flags, error));
1542 		/*
1543 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1544 		 * If RETRY is not set, we return ENOENT instead.
1545 		 */
1546 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1547 		    (flags & LK_RETRY) == 0) {
1548 			VOP_UNLOCK(vp, 0);
1549 			error = ENOENT;
1550 			break;
1551 		}
1552 	} while (flags & LK_RETRY && error != 0);
1553 	return (error);
1554 }
1555 
1556 /*
1557  * File table vnode close routine.
1558  */
1559 static int
1560 vn_closefile(fp, td)
1561 	struct file *fp;
1562 	struct thread *td;
1563 {
1564 	struct vnode *vp;
1565 	struct flock lf;
1566 	int error;
1567 
1568 	vp = fp->f_vnode;
1569 	fp->f_ops = &badfileops;
1570 
1571 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1572 		vref(vp);
1573 
1574 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1575 
1576 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1577 		lf.l_whence = SEEK_SET;
1578 		lf.l_start = 0;
1579 		lf.l_len = 0;
1580 		lf.l_type = F_UNLCK;
1581 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1582 		vrele(vp);
1583 	}
1584 	return (error);
1585 }
1586 
1587 static bool
1588 vn_suspendable(struct mount *mp)
1589 {
1590 
1591 	return (mp->mnt_op->vfs_susp_clean != NULL);
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, mflags;
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 		mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1614 		    (flags & PCATCH) : 0) | (PUSER - 1);
1615 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1616 			if (flags & V_NOWAIT) {
1617 				error = EWOULDBLOCK;
1618 				goto unlock;
1619 			}
1620 			error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1621 			    "suspfs", 0);
1622 			if (error)
1623 				goto unlock;
1624 		}
1625 	}
1626 	if (flags & V_XSLEEP)
1627 		goto unlock;
1628 	mp->mnt_writeopcount++;
1629 unlock:
1630 	if (error != 0 || (flags & V_XSLEEP) != 0)
1631 		MNT_REL(mp);
1632 	MNT_IUNLOCK(mp);
1633 	return (error);
1634 }
1635 
1636 int
1637 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1638 {
1639 	struct mount *mp;
1640 	int error;
1641 
1642 	KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1643 	    ("V_MNTREF requires mp"));
1644 
1645 	error = 0;
1646 	/*
1647 	 * If a vnode is provided, get and return the mount point that
1648 	 * to which it will write.
1649 	 */
1650 	if (vp != NULL) {
1651 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1652 			*mpp = NULL;
1653 			if (error != EOPNOTSUPP)
1654 				return (error);
1655 			return (0);
1656 		}
1657 	}
1658 	if ((mp = *mpp) == NULL)
1659 		return (0);
1660 
1661 	if (!vn_suspendable(mp)) {
1662 		if (vp != NULL || (flags & V_MNTREF) != 0)
1663 			vfs_rel(mp);
1664 		return (0);
1665 	}
1666 
1667 	/*
1668 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1669 	 * a vfs_ref().
1670 	 * As long as a vnode is not provided we need to acquire a
1671 	 * refcount for the provided mountpoint too, in order to
1672 	 * emulate a vfs_ref().
1673 	 */
1674 	MNT_ILOCK(mp);
1675 	if (vp == NULL && (flags & V_MNTREF) == 0)
1676 		MNT_REF(mp);
1677 
1678 	return (vn_start_write_locked(mp, flags));
1679 }
1680 
1681 /*
1682  * Secondary suspension. Used by operations such as vop_inactive
1683  * routines that are needed by the higher level functions. These
1684  * are allowed to proceed until all the higher level functions have
1685  * completed (indicated by mnt_writeopcount dropping to zero). At that
1686  * time, these operations are halted until the suspension is over.
1687  */
1688 int
1689 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1690 {
1691 	struct mount *mp;
1692 	int error;
1693 
1694 	KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1695 	    ("V_MNTREF requires mp"));
1696 
1697  retry:
1698 	if (vp != NULL) {
1699 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1700 			*mpp = NULL;
1701 			if (error != EOPNOTSUPP)
1702 				return (error);
1703 			return (0);
1704 		}
1705 	}
1706 	/*
1707 	 * If we are not suspended or have not yet reached suspended
1708 	 * mode, then let the operation proceed.
1709 	 */
1710 	if ((mp = *mpp) == NULL)
1711 		return (0);
1712 
1713 	if (!vn_suspendable(mp)) {
1714 		if (vp != NULL || (flags & V_MNTREF) != 0)
1715 			vfs_rel(mp);
1716 		return (0);
1717 	}
1718 
1719 	/*
1720 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1721 	 * a vfs_ref().
1722 	 * As long as a vnode is not provided we need to acquire a
1723 	 * refcount for the provided mountpoint too, in order to
1724 	 * emulate a vfs_ref().
1725 	 */
1726 	MNT_ILOCK(mp);
1727 	if (vp == NULL && (flags & V_MNTREF) == 0)
1728 		MNT_REF(mp);
1729 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1730 		mp->mnt_secondary_writes++;
1731 		mp->mnt_secondary_accwrites++;
1732 		MNT_IUNLOCK(mp);
1733 		return (0);
1734 	}
1735 	if (flags & V_NOWAIT) {
1736 		MNT_REL(mp);
1737 		MNT_IUNLOCK(mp);
1738 		return (EWOULDBLOCK);
1739 	}
1740 	/*
1741 	 * Wait for the suspension to finish.
1742 	 */
1743 	error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1744 	    ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1745 	    "suspfs", 0);
1746 	vfs_rel(mp);
1747 	if (error == 0)
1748 		goto retry;
1749 	return (error);
1750 }
1751 
1752 /*
1753  * Filesystem write operation has completed. If we are suspending and this
1754  * operation is the last one, notify the suspender that the suspension is
1755  * now in effect.
1756  */
1757 void
1758 vn_finished_write(mp)
1759 	struct mount *mp;
1760 {
1761 	if (mp == NULL || !vn_suspendable(mp))
1762 		return;
1763 	MNT_ILOCK(mp);
1764 	MNT_REL(mp);
1765 	mp->mnt_writeopcount--;
1766 	if (mp->mnt_writeopcount < 0)
1767 		panic("vn_finished_write: neg cnt");
1768 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1769 	    mp->mnt_writeopcount <= 0)
1770 		wakeup(&mp->mnt_writeopcount);
1771 	MNT_IUNLOCK(mp);
1772 }
1773 
1774 
1775 /*
1776  * Filesystem secondary write operation has completed. If we are
1777  * suspending and this operation is the last one, notify the suspender
1778  * that the suspension is now in effect.
1779  */
1780 void
1781 vn_finished_secondary_write(mp)
1782 	struct mount *mp;
1783 {
1784 	if (mp == NULL || !vn_suspendable(mp))
1785 		return;
1786 	MNT_ILOCK(mp);
1787 	MNT_REL(mp);
1788 	mp->mnt_secondary_writes--;
1789 	if (mp->mnt_secondary_writes < 0)
1790 		panic("vn_finished_secondary_write: neg cnt");
1791 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1792 	    mp->mnt_secondary_writes <= 0)
1793 		wakeup(&mp->mnt_secondary_writes);
1794 	MNT_IUNLOCK(mp);
1795 }
1796 
1797 
1798 
1799 /*
1800  * Request a filesystem to suspend write operations.
1801  */
1802 int
1803 vfs_write_suspend(struct mount *mp, int flags)
1804 {
1805 	int error;
1806 
1807 	MPASS(vn_suspendable(mp));
1808 
1809 	MNT_ILOCK(mp);
1810 	if (mp->mnt_susp_owner == curthread) {
1811 		MNT_IUNLOCK(mp);
1812 		return (EALREADY);
1813 	}
1814 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1815 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1816 
1817 	/*
1818 	 * Unmount holds a write reference on the mount point.  If we
1819 	 * own busy reference and drain for writers, we deadlock with
1820 	 * the reference draining in the unmount path.  Callers of
1821 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1822 	 * vfs_busy() reference is owned and caller is not in the
1823 	 * unmount context.
1824 	 */
1825 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1826 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1827 		MNT_IUNLOCK(mp);
1828 		return (EBUSY);
1829 	}
1830 
1831 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1832 	mp->mnt_susp_owner = curthread;
1833 	if (mp->mnt_writeopcount > 0)
1834 		(void) msleep(&mp->mnt_writeopcount,
1835 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1836 	else
1837 		MNT_IUNLOCK(mp);
1838 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1839 		vfs_write_resume(mp, 0);
1840 	return (error);
1841 }
1842 
1843 /*
1844  * Request a filesystem to resume write operations.
1845  */
1846 void
1847 vfs_write_resume(struct mount *mp, int flags)
1848 {
1849 
1850 	MPASS(vn_suspendable(mp));
1851 
1852 	MNT_ILOCK(mp);
1853 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1854 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1855 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1856 				       MNTK_SUSPENDED);
1857 		mp->mnt_susp_owner = NULL;
1858 		wakeup(&mp->mnt_writeopcount);
1859 		wakeup(&mp->mnt_flag);
1860 		curthread->td_pflags &= ~TDP_IGNSUSP;
1861 		if ((flags & VR_START_WRITE) != 0) {
1862 			MNT_REF(mp);
1863 			mp->mnt_writeopcount++;
1864 		}
1865 		MNT_IUNLOCK(mp);
1866 		if ((flags & VR_NO_SUSPCLR) == 0)
1867 			VFS_SUSP_CLEAN(mp);
1868 	} else if ((flags & VR_START_WRITE) != 0) {
1869 		MNT_REF(mp);
1870 		vn_start_write_locked(mp, 0);
1871 	} else {
1872 		MNT_IUNLOCK(mp);
1873 	}
1874 }
1875 
1876 /*
1877  * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1878  * methods.
1879  */
1880 int
1881 vfs_write_suspend_umnt(struct mount *mp)
1882 {
1883 	int error;
1884 
1885 	MPASS(vn_suspendable(mp));
1886 	KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1887 	    ("vfs_write_suspend_umnt: recursed"));
1888 
1889 	/* dounmount() already called vn_start_write(). */
1890 	for (;;) {
1891 		vn_finished_write(mp);
1892 		error = vfs_write_suspend(mp, 0);
1893 		if (error != 0) {
1894 			vn_start_write(NULL, &mp, V_WAIT);
1895 			return (error);
1896 		}
1897 		MNT_ILOCK(mp);
1898 		if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1899 			break;
1900 		MNT_IUNLOCK(mp);
1901 		vn_start_write(NULL, &mp, V_WAIT);
1902 	}
1903 	mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1904 	wakeup(&mp->mnt_flag);
1905 	MNT_IUNLOCK(mp);
1906 	curthread->td_pflags |= TDP_IGNSUSP;
1907 	return (0);
1908 }
1909 
1910 /*
1911  * Implement kqueues for files by translating it to vnode operation.
1912  */
1913 static int
1914 vn_kqfilter(struct file *fp, struct knote *kn)
1915 {
1916 
1917 	return (VOP_KQFILTER(fp->f_vnode, kn));
1918 }
1919 
1920 /*
1921  * Simplified in-kernel wrapper calls for extended attribute access.
1922  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1923  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1924  */
1925 int
1926 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1927     const char *attrname, int *buflen, char *buf, struct thread *td)
1928 {
1929 	struct uio	auio;
1930 	struct iovec	iov;
1931 	int	error;
1932 
1933 	iov.iov_len = *buflen;
1934 	iov.iov_base = buf;
1935 
1936 	auio.uio_iov = &iov;
1937 	auio.uio_iovcnt = 1;
1938 	auio.uio_rw = UIO_READ;
1939 	auio.uio_segflg = UIO_SYSSPACE;
1940 	auio.uio_td = td;
1941 	auio.uio_offset = 0;
1942 	auio.uio_resid = *buflen;
1943 
1944 	if ((ioflg & IO_NODELOCKED) == 0)
1945 		vn_lock(vp, LK_SHARED | LK_RETRY);
1946 
1947 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1948 
1949 	/* authorize attribute retrieval as kernel */
1950 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1951 	    td);
1952 
1953 	if ((ioflg & IO_NODELOCKED) == 0)
1954 		VOP_UNLOCK(vp, 0);
1955 
1956 	if (error == 0) {
1957 		*buflen = *buflen - auio.uio_resid;
1958 	}
1959 
1960 	return (error);
1961 }
1962 
1963 /*
1964  * XXX failure mode if partially written?
1965  */
1966 int
1967 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1968     const char *attrname, int buflen, char *buf, struct thread *td)
1969 {
1970 	struct uio	auio;
1971 	struct iovec	iov;
1972 	struct mount	*mp;
1973 	int	error;
1974 
1975 	iov.iov_len = buflen;
1976 	iov.iov_base = buf;
1977 
1978 	auio.uio_iov = &iov;
1979 	auio.uio_iovcnt = 1;
1980 	auio.uio_rw = UIO_WRITE;
1981 	auio.uio_segflg = UIO_SYSSPACE;
1982 	auio.uio_td = td;
1983 	auio.uio_offset = 0;
1984 	auio.uio_resid = buflen;
1985 
1986 	if ((ioflg & IO_NODELOCKED) == 0) {
1987 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1988 			return (error);
1989 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1990 	}
1991 
1992 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1993 
1994 	/* authorize attribute setting as kernel */
1995 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1996 
1997 	if ((ioflg & IO_NODELOCKED) == 0) {
1998 		vn_finished_write(mp);
1999 		VOP_UNLOCK(vp, 0);
2000 	}
2001 
2002 	return (error);
2003 }
2004 
2005 int
2006 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2007     const char *attrname, struct thread *td)
2008 {
2009 	struct mount	*mp;
2010 	int	error;
2011 
2012 	if ((ioflg & IO_NODELOCKED) == 0) {
2013 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2014 			return (error);
2015 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2016 	}
2017 
2018 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2019 
2020 	/* authorize attribute removal as kernel */
2021 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2022 	if (error == EOPNOTSUPP)
2023 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2024 		    NULL, td);
2025 
2026 	if ((ioflg & IO_NODELOCKED) == 0) {
2027 		vn_finished_write(mp);
2028 		VOP_UNLOCK(vp, 0);
2029 	}
2030 
2031 	return (error);
2032 }
2033 
2034 static int
2035 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2036     struct vnode **rvp)
2037 {
2038 
2039 	return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2040 }
2041 
2042 int
2043 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2044 {
2045 
2046 	return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2047 	    lkflags, rvp));
2048 }
2049 
2050 int
2051 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2052     int lkflags, struct vnode **rvp)
2053 {
2054 	struct mount *mp;
2055 	int ltype, error;
2056 
2057 	ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2058 	mp = vp->v_mount;
2059 	ltype = VOP_ISLOCKED(vp);
2060 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2061 	    ("vn_vget_ino: vp not locked"));
2062 	error = vfs_busy(mp, MBF_NOWAIT);
2063 	if (error != 0) {
2064 		vfs_ref(mp);
2065 		VOP_UNLOCK(vp, 0);
2066 		error = vfs_busy(mp, 0);
2067 		vn_lock(vp, ltype | LK_RETRY);
2068 		vfs_rel(mp);
2069 		if (error != 0)
2070 			return (ENOENT);
2071 		if (vp->v_iflag & VI_DOOMED) {
2072 			vfs_unbusy(mp);
2073 			return (ENOENT);
2074 		}
2075 	}
2076 	VOP_UNLOCK(vp, 0);
2077 	error = alloc(mp, alloc_arg, lkflags, rvp);
2078 	vfs_unbusy(mp);
2079 	if (*rvp != vp)
2080 		vn_lock(vp, ltype | LK_RETRY);
2081 	if (vp->v_iflag & VI_DOOMED) {
2082 		if (error == 0) {
2083 			if (*rvp == vp)
2084 				vunref(vp);
2085 			else
2086 				vput(*rvp);
2087 		}
2088 		error = ENOENT;
2089 	}
2090 	return (error);
2091 }
2092 
2093 int
2094 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2095     struct thread *td)
2096 {
2097 
2098 	if (vp->v_type != VREG || td == NULL)
2099 		return (0);
2100 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
2101 	    lim_cur(td, RLIMIT_FSIZE)) {
2102 		PROC_LOCK(td->td_proc);
2103 		kern_psignal(td->td_proc, SIGXFSZ);
2104 		PROC_UNLOCK(td->td_proc);
2105 		return (EFBIG);
2106 	}
2107 	return (0);
2108 }
2109 
2110 int
2111 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2112     struct thread *td)
2113 {
2114 	struct vnode *vp;
2115 
2116 	vp = fp->f_vnode;
2117 #ifdef AUDIT
2118 	vn_lock(vp, LK_SHARED | LK_RETRY);
2119 	AUDIT_ARG_VNODE1(vp);
2120 	VOP_UNLOCK(vp, 0);
2121 #endif
2122 	return (setfmode(td, active_cred, vp, mode));
2123 }
2124 
2125 int
2126 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2127     struct thread *td)
2128 {
2129 	struct vnode *vp;
2130 
2131 	vp = fp->f_vnode;
2132 #ifdef AUDIT
2133 	vn_lock(vp, LK_SHARED | LK_RETRY);
2134 	AUDIT_ARG_VNODE1(vp);
2135 	VOP_UNLOCK(vp, 0);
2136 #endif
2137 	return (setfown(td, active_cred, vp, uid, gid));
2138 }
2139 
2140 void
2141 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2142 {
2143 	vm_object_t object;
2144 
2145 	if ((object = vp->v_object) == NULL)
2146 		return;
2147 	VM_OBJECT_WLOCK(object);
2148 	vm_object_page_remove(object, start, end, 0);
2149 	VM_OBJECT_WUNLOCK(object);
2150 }
2151 
2152 int
2153 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2154 {
2155 	struct vattr va;
2156 	daddr_t bn, bnp;
2157 	uint64_t bsize;
2158 	off_t noff;
2159 	int error;
2160 
2161 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2162 	    ("Wrong command %lu", cmd));
2163 
2164 	if (vn_lock(vp, LK_SHARED) != 0)
2165 		return (EBADF);
2166 	if (vp->v_type != VREG) {
2167 		error = ENOTTY;
2168 		goto unlock;
2169 	}
2170 	error = VOP_GETATTR(vp, &va, cred);
2171 	if (error != 0)
2172 		goto unlock;
2173 	noff = *off;
2174 	if (noff >= va.va_size) {
2175 		error = ENXIO;
2176 		goto unlock;
2177 	}
2178 	bsize = vp->v_mount->mnt_stat.f_iosize;
2179 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2180 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2181 		if (error == EOPNOTSUPP) {
2182 			error = ENOTTY;
2183 			goto unlock;
2184 		}
2185 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2186 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
2187 			noff = bn * bsize;
2188 			if (noff < *off)
2189 				noff = *off;
2190 			goto unlock;
2191 		}
2192 	}
2193 	if (noff > va.va_size)
2194 		noff = va.va_size;
2195 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2196 	if (cmd == FIOSEEKDATA)
2197 		error = ENXIO;
2198 unlock:
2199 	VOP_UNLOCK(vp, 0);
2200 	if (error == 0)
2201 		*off = noff;
2202 	return (error);
2203 }
2204 
2205 int
2206 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2207 {
2208 	struct ucred *cred;
2209 	struct vnode *vp;
2210 	struct vattr vattr;
2211 	off_t foffset, size;
2212 	int error, noneg;
2213 
2214 	cred = td->td_ucred;
2215 	vp = fp->f_vnode;
2216 	foffset = foffset_lock(fp, 0);
2217 	noneg = (vp->v_type != VCHR);
2218 	error = 0;
2219 	switch (whence) {
2220 	case L_INCR:
2221 		if (noneg &&
2222 		    (foffset < 0 ||
2223 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2224 			error = EOVERFLOW;
2225 			break;
2226 		}
2227 		offset += foffset;
2228 		break;
2229 	case L_XTND:
2230 		vn_lock(vp, LK_SHARED | LK_RETRY);
2231 		error = VOP_GETATTR(vp, &vattr, cred);
2232 		VOP_UNLOCK(vp, 0);
2233 		if (error)
2234 			break;
2235 
2236 		/*
2237 		 * If the file references a disk device, then fetch
2238 		 * the media size and use that to determine the ending
2239 		 * offset.
2240 		 */
2241 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2242 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2243 			vattr.va_size = size;
2244 		if (noneg &&
2245 		    (vattr.va_size > OFF_MAX ||
2246 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2247 			error = EOVERFLOW;
2248 			break;
2249 		}
2250 		offset += vattr.va_size;
2251 		break;
2252 	case L_SET:
2253 		break;
2254 	case SEEK_DATA:
2255 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2256 		break;
2257 	case SEEK_HOLE:
2258 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2259 		break;
2260 	default:
2261 		error = EINVAL;
2262 	}
2263 	if (error == 0 && noneg && offset < 0)
2264 		error = EINVAL;
2265 	if (error != 0)
2266 		goto drop;
2267 	VFS_KNOTE_UNLOCKED(vp, 0);
2268 	td->td_uretoff.tdu_off = offset;
2269 drop:
2270 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2271 	return (error);
2272 }
2273 
2274 int
2275 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2276     struct thread *td)
2277 {
2278 	int error;
2279 
2280 	/*
2281 	 * Grant permission if the caller is the owner of the file, or
2282 	 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2283 	 * on the file.  If the time pointer is null, then write
2284 	 * permission on the file is also sufficient.
2285 	 *
2286 	 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2287 	 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2288 	 * will be allowed to set the times [..] to the current
2289 	 * server time.
2290 	 */
2291 	error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2292 	if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2293 		error = VOP_ACCESS(vp, VWRITE, cred, td);
2294 	return (error);
2295 }
2296 
2297 int
2298 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2299 {
2300 	struct vnode *vp;
2301 	int error;
2302 
2303 	if (fp->f_type == DTYPE_FIFO)
2304 		kif->kf_type = KF_TYPE_FIFO;
2305 	else
2306 		kif->kf_type = KF_TYPE_VNODE;
2307 	vp = fp->f_vnode;
2308 	vref(vp);
2309 	FILEDESC_SUNLOCK(fdp);
2310 	error = vn_fill_kinfo_vnode(vp, kif);
2311 	vrele(vp);
2312 	FILEDESC_SLOCK(fdp);
2313 	return (error);
2314 }
2315 
2316 static inline void
2317 vn_fill_junk(struct kinfo_file *kif)
2318 {
2319 	size_t len, olen;
2320 
2321 	/*
2322 	 * Simulate vn_fullpath returning changing values for a given
2323 	 * vp during e.g. coredump.
2324 	 */
2325 	len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2326 	olen = strlen(kif->kf_path);
2327 	if (len < olen)
2328 		strcpy(&kif->kf_path[len - 1], "$");
2329 	else
2330 		for (; olen < len; olen++)
2331 			strcpy(&kif->kf_path[olen], "A");
2332 }
2333 
2334 int
2335 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2336 {
2337 	struct vattr va;
2338 	char *fullpath, *freepath;
2339 	int error;
2340 
2341 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2342 	freepath = NULL;
2343 	fullpath = "-";
2344 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2345 	if (error == 0) {
2346 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2347 	}
2348 	if (freepath != NULL)
2349 		free(freepath, M_TEMP);
2350 
2351 	KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2352 		vn_fill_junk(kif);
2353 	);
2354 
2355 	/*
2356 	 * Retrieve vnode attributes.
2357 	 */
2358 	va.va_fsid = VNOVAL;
2359 	va.va_rdev = NODEV;
2360 	vn_lock(vp, LK_SHARED | LK_RETRY);
2361 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2362 	VOP_UNLOCK(vp, 0);
2363 	if (error != 0)
2364 		return (error);
2365 	if (va.va_fsid != VNOVAL)
2366 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2367 	else
2368 		kif->kf_un.kf_file.kf_file_fsid =
2369 		    vp->v_mount->mnt_stat.f_fsid.val[0];
2370 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2371 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2372 	kif->kf_un.kf_file.kf_file_size = va.va_size;
2373 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2374 	return (0);
2375 }
2376 
2377 int
2378 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2379     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2380     struct thread *td)
2381 {
2382 #ifdef HWPMC_HOOKS
2383 	struct pmckern_map_in pkm;
2384 #endif
2385 	struct mount *mp;
2386 	struct vnode *vp;
2387 	vm_object_t object;
2388 	vm_prot_t maxprot;
2389 	boolean_t writecounted;
2390 	int error;
2391 
2392 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2393     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2394 	/*
2395 	 * POSIX shared-memory objects are defined to have
2396 	 * kernel persistence, and are not defined to support
2397 	 * read(2)/write(2) -- or even open(2).  Thus, we can
2398 	 * use MAP_ASYNC to trade on-disk coherence for speed.
2399 	 * The shm_open(3) library routine turns on the FPOSIXSHM
2400 	 * flag to request this behavior.
2401 	 */
2402 	if ((fp->f_flag & FPOSIXSHM) != 0)
2403 		flags |= MAP_NOSYNC;
2404 #endif
2405 	vp = fp->f_vnode;
2406 
2407 	/*
2408 	 * Ensure that file and memory protections are
2409 	 * compatible.  Note that we only worry about
2410 	 * writability if mapping is shared; in this case,
2411 	 * current and max prot are dictated by the open file.
2412 	 * XXX use the vnode instead?  Problem is: what
2413 	 * credentials do we use for determination? What if
2414 	 * proc does a setuid?
2415 	 */
2416 	mp = vp->v_mount;
2417 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0)
2418 		maxprot = VM_PROT_NONE;
2419 	else
2420 		maxprot = VM_PROT_EXECUTE;
2421 	if ((fp->f_flag & FREAD) != 0)
2422 		maxprot |= VM_PROT_READ;
2423 	else if ((prot & VM_PROT_READ) != 0)
2424 		return (EACCES);
2425 
2426 	/*
2427 	 * If we are sharing potential changes via MAP_SHARED and we
2428 	 * are trying to get write permission although we opened it
2429 	 * without asking for it, bail out.
2430 	 */
2431 	if ((flags & MAP_SHARED) != 0) {
2432 		if ((fp->f_flag & FWRITE) != 0)
2433 			maxprot |= VM_PROT_WRITE;
2434 		else if ((prot & VM_PROT_WRITE) != 0)
2435 			return (EACCES);
2436 	} else {
2437 		maxprot |= VM_PROT_WRITE;
2438 		cap_maxprot |= VM_PROT_WRITE;
2439 	}
2440 	maxprot &= cap_maxprot;
2441 
2442 	writecounted = FALSE;
2443 	error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2444 	    &foff, &object, &writecounted);
2445 	if (error != 0)
2446 		return (error);
2447 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2448 	    foff, writecounted, td);
2449 	if (error != 0) {
2450 		/*
2451 		 * If this mapping was accounted for in the vnode's
2452 		 * writecount, then undo that now.
2453 		 */
2454 		if (writecounted)
2455 			vnode_pager_release_writecount(object, 0, size);
2456 		vm_object_deallocate(object);
2457 	}
2458 #ifdef HWPMC_HOOKS
2459 	/* Inform hwpmc(4) if an executable is being mapped. */
2460 	if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) {
2461 		pkm.pm_file = vp;
2462 		pkm.pm_address = (uintptr_t) *addr;
2463 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
2464 	}
2465 #endif
2466 	return (error);
2467 }
2468