xref: /freebsd/sys/kern/vfs_vnops.c (revision 5608fd23c27fa1e8ee595d7b678cbfd35d657fbe)
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 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_extern.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
82 
83 static fo_rdwr_t	vn_read;
84 static fo_rdwr_t	vn_write;
85 static fo_rdwr_t	vn_io_fault;
86 static fo_truncate_t	vn_truncate;
87 static fo_ioctl_t	vn_ioctl;
88 static fo_poll_t	vn_poll;
89 static fo_kqfilter_t	vn_kqfilter;
90 static fo_stat_t	vn_statfile;
91 static fo_close_t	vn_closefile;
92 
93 struct 	fileops vnops = {
94 	.fo_read = vn_io_fault,
95 	.fo_write = vn_io_fault,
96 	.fo_truncate = vn_truncate,
97 	.fo_ioctl = vn_ioctl,
98 	.fo_poll = vn_poll,
99 	.fo_kqfilter = vn_kqfilter,
100 	.fo_stat = vn_statfile,
101 	.fo_close = vn_closefile,
102 	.fo_chmod = vn_chmod,
103 	.fo_chown = vn_chown,
104 	.fo_sendfile = vn_sendfile,
105 	.fo_seek = vn_seek,
106 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
107 };
108 
109 static const int io_hold_cnt = 16;
110 static int vn_io_fault_enable = 1;
111 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
112     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
113 static u_long vn_io_faults_cnt;
114 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
115     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
116 
117 /*
118  * Returns true if vn_io_fault mode of handling the i/o request should
119  * be used.
120  */
121 static bool
122 do_vn_io_fault(struct vnode *vp, struct uio *uio)
123 {
124 	struct mount *mp;
125 
126 	return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
127 	    (mp = vp->v_mount) != NULL &&
128 	    (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
129 }
130 
131 /*
132  * Structure used to pass arguments to vn_io_fault1(), to do either
133  * file- or vnode-based I/O calls.
134  */
135 struct vn_io_fault_args {
136 	enum {
137 		VN_IO_FAULT_FOP,
138 		VN_IO_FAULT_VOP
139 	} kind;
140 	struct ucred *cred;
141 	int flags;
142 	union {
143 		struct fop_args_tag {
144 			struct file *fp;
145 			fo_rdwr_t *doio;
146 		} fop_args;
147 		struct vop_args_tag {
148 			struct vnode *vp;
149 		} vop_args;
150 	} args;
151 };
152 
153 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
154     struct vn_io_fault_args *args, struct thread *td);
155 
156 int
157 vn_open(ndp, flagp, cmode, fp)
158 	struct nameidata *ndp;
159 	int *flagp, cmode;
160 	struct file *fp;
161 {
162 	struct thread *td = ndp->ni_cnd.cn_thread;
163 
164 	return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
165 }
166 
167 /*
168  * Common code for vnode open operations via a name lookup.
169  * Lookup the vnode and invoke VOP_CREATE if needed.
170  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
171  *
172  * Note that this does NOT free nameidata for the successful case,
173  * due to the NDINIT being done elsewhere.
174  */
175 int
176 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
177     struct ucred *cred, struct file *fp)
178 {
179 	struct vnode *vp;
180 	struct mount *mp;
181 	struct thread *td = ndp->ni_cnd.cn_thread;
182 	struct vattr vat;
183 	struct vattr *vap = &vat;
184 	int fmode, error;
185 
186 restart:
187 	fmode = *flagp;
188 	if (fmode & O_CREAT) {
189 		ndp->ni_cnd.cn_nameiop = CREATE;
190 		ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
191 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
192 			ndp->ni_cnd.cn_flags |= FOLLOW;
193 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
194 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
195 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
196 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
197 		bwillwrite();
198 		if ((error = namei(ndp)) != 0)
199 			return (error);
200 		if (ndp->ni_vp == NULL) {
201 			VATTR_NULL(vap);
202 			vap->va_type = VREG;
203 			vap->va_mode = cmode;
204 			if (fmode & O_EXCL)
205 				vap->va_vaflags |= VA_EXCLUSIVE;
206 			if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
207 				NDFREE(ndp, NDF_ONLY_PNBUF);
208 				vput(ndp->ni_dvp);
209 				if ((error = vn_start_write(NULL, &mp,
210 				    V_XSLEEP | PCATCH)) != 0)
211 					return (error);
212 				goto restart;
213 			}
214 #ifdef MAC
215 			error = mac_vnode_check_create(cred, ndp->ni_dvp,
216 			    &ndp->ni_cnd, vap);
217 			if (error == 0)
218 #endif
219 				error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
220 						   &ndp->ni_cnd, vap);
221 			vput(ndp->ni_dvp);
222 			vn_finished_write(mp);
223 			if (error) {
224 				NDFREE(ndp, NDF_ONLY_PNBUF);
225 				return (error);
226 			}
227 			fmode &= ~O_TRUNC;
228 			vp = ndp->ni_vp;
229 		} else {
230 			if (ndp->ni_dvp == ndp->ni_vp)
231 				vrele(ndp->ni_dvp);
232 			else
233 				vput(ndp->ni_dvp);
234 			ndp->ni_dvp = NULL;
235 			vp = ndp->ni_vp;
236 			if (fmode & O_EXCL) {
237 				error = EEXIST;
238 				goto bad;
239 			}
240 			fmode &= ~O_CREAT;
241 		}
242 	} else {
243 		ndp->ni_cnd.cn_nameiop = LOOKUP;
244 		ndp->ni_cnd.cn_flags = ISOPEN |
245 		    ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
246 		if (!(fmode & FWRITE))
247 			ndp->ni_cnd.cn_flags |= LOCKSHARED;
248 		if (!(vn_open_flags & VN_OPEN_NOAUDIT))
249 			ndp->ni_cnd.cn_flags |= AUDITVNODE1;
250 		if (vn_open_flags & VN_OPEN_NOCAPCHECK)
251 			ndp->ni_cnd.cn_flags |= NOCAPCHECK;
252 		if ((error = namei(ndp)) != 0)
253 			return (error);
254 		vp = ndp->ni_vp;
255 	}
256 	error = vn_open_vnode(vp, fmode, cred, td, fp);
257 	if (error)
258 		goto bad;
259 	*flagp = fmode;
260 	return (0);
261 bad:
262 	NDFREE(ndp, NDF_ONLY_PNBUF);
263 	vput(vp);
264 	*flagp = fmode;
265 	ndp->ni_vp = NULL;
266 	return (error);
267 }
268 
269 /*
270  * Common code for vnode open operations once a vnode is located.
271  * Check permissions, and call the VOP_OPEN routine.
272  */
273 int
274 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
275     struct thread *td, struct file *fp)
276 {
277 	struct mount *mp;
278 	accmode_t accmode;
279 	struct flock lf;
280 	int error, have_flock, lock_flags, type;
281 
282 	if (vp->v_type == VLNK)
283 		return (EMLINK);
284 	if (vp->v_type == VSOCK)
285 		return (EOPNOTSUPP);
286 	if (vp->v_type != VDIR && fmode & O_DIRECTORY)
287 		return (ENOTDIR);
288 	accmode = 0;
289 	if (fmode & (FWRITE | O_TRUNC)) {
290 		if (vp->v_type == VDIR)
291 			return (EISDIR);
292 		accmode |= VWRITE;
293 	}
294 	if (fmode & FREAD)
295 		accmode |= VREAD;
296 	if (fmode & FEXEC)
297 		accmode |= VEXEC;
298 	if ((fmode & O_APPEND) && (fmode & FWRITE))
299 		accmode |= VAPPEND;
300 #ifdef MAC
301 	error = mac_vnode_check_open(cred, vp, accmode);
302 	if (error)
303 		return (error);
304 #endif
305 	if ((fmode & O_CREAT) == 0) {
306 		if (accmode & VWRITE) {
307 			error = vn_writechk(vp);
308 			if (error)
309 				return (error);
310 		}
311 		if (accmode) {
312 		        error = VOP_ACCESS(vp, accmode, cred, td);
313 			if (error)
314 				return (error);
315 		}
316 	}
317 	if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
318 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
319 	if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
320 		return (error);
321 
322 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
323 		KASSERT(fp != NULL, ("open with flock requires fp"));
324 		lock_flags = VOP_ISLOCKED(vp);
325 		VOP_UNLOCK(vp, 0);
326 		lf.l_whence = SEEK_SET;
327 		lf.l_start = 0;
328 		lf.l_len = 0;
329 		if (fmode & O_EXLOCK)
330 			lf.l_type = F_WRLCK;
331 		else
332 			lf.l_type = F_RDLCK;
333 		type = F_FLOCK;
334 		if ((fmode & FNONBLOCK) == 0)
335 			type |= F_WAIT;
336 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
337 		have_flock = (error == 0);
338 		vn_lock(vp, lock_flags | LK_RETRY);
339 		if (error == 0 && vp->v_iflag & VI_DOOMED)
340 			error = ENOENT;
341 		/*
342 		 * Another thread might have used this vnode as an
343 		 * executable while the vnode lock was dropped.
344 		 * Ensure the vnode is still able to be opened for
345 		 * writing after the lock has been obtained.
346 		 */
347 		if (error == 0 && accmode & VWRITE)
348 			error = vn_writechk(vp);
349 		if (error) {
350 			VOP_UNLOCK(vp, 0);
351 			if (have_flock) {
352 				lf.l_whence = SEEK_SET;
353 				lf.l_start = 0;
354 				lf.l_len = 0;
355 				lf.l_type = F_UNLCK;
356 				(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
357 				    F_FLOCK);
358 			}
359 			vn_start_write(vp, &mp, V_WAIT);
360 			vn_lock(vp, lock_flags | LK_RETRY);
361 			(void)VOP_CLOSE(vp, fmode, cred, td);
362 			vn_finished_write(mp);
363 			/* Prevent second close from fdrop()->vn_close(). */
364 			if (fp != NULL)
365 				fp->f_ops= &badfileops;
366 			return (error);
367 		}
368 		fp->f_flag |= FHASLOCK;
369 	}
370 	if (fmode & FWRITE) {
371 		VOP_ADD_WRITECOUNT(vp, 1);
372 		CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
373 		    __func__, vp, vp->v_writecount);
374 	}
375 	ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
376 	return (0);
377 }
378 
379 /*
380  * Check for write permissions on the specified vnode.
381  * Prototype text segments cannot be written.
382  */
383 int
384 vn_writechk(vp)
385 	register struct vnode *vp;
386 {
387 
388 	ASSERT_VOP_LOCKED(vp, "vn_writechk");
389 	/*
390 	 * If there's shared text associated with
391 	 * the vnode, try to free it up once.  If
392 	 * we fail, we can't allow writing.
393 	 */
394 	if (VOP_IS_TEXT(vp))
395 		return (ETXTBSY);
396 
397 	return (0);
398 }
399 
400 /*
401  * Vnode close call
402  */
403 int
404 vn_close(vp, flags, file_cred, td)
405 	register struct vnode *vp;
406 	int flags;
407 	struct ucred *file_cred;
408 	struct thread *td;
409 {
410 	struct mount *mp;
411 	int error, lock_flags;
412 
413 	if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
414 	    MNT_EXTENDED_SHARED(vp->v_mount))
415 		lock_flags = LK_SHARED;
416 	else
417 		lock_flags = LK_EXCLUSIVE;
418 
419 	vn_start_write(vp, &mp, V_WAIT);
420 	vn_lock(vp, lock_flags | LK_RETRY);
421 	if (flags & FWRITE) {
422 		VNASSERT(vp->v_writecount > 0, vp,
423 		    ("vn_close: negative writecount"));
424 		VOP_ADD_WRITECOUNT(vp, -1);
425 		CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
426 		    __func__, vp, vp->v_writecount);
427 	}
428 	error = VOP_CLOSE(vp, flags, file_cred, td);
429 	vput(vp);
430 	vn_finished_write(mp);
431 	return (error);
432 }
433 
434 /*
435  * Heuristic to detect sequential operation.
436  */
437 static int
438 sequential_heuristic(struct uio *uio, struct file *fp)
439 {
440 
441 	if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
442 		return (fp->f_seqcount << IO_SEQSHIFT);
443 
444 	/*
445 	 * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
446 	 * that the first I/O is normally considered to be slightly
447 	 * sequential.  Seeking to offset 0 doesn't change sequentiality
448 	 * unless previous seeks have reduced f_seqcount to 0, in which
449 	 * case offset 0 is not special.
450 	 */
451 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
452 	    uio->uio_offset == fp->f_nextoff) {
453 		/*
454 		 * f_seqcount is in units of fixed-size blocks so that it
455 		 * depends mainly on the amount of sequential I/O and not
456 		 * much on the number of sequential I/O's.  The fixed size
457 		 * of 16384 is hard-coded here since it is (not quite) just
458 		 * a magic size that works well here.  This size is more
459 		 * closely related to the best I/O size for real disks than
460 		 * to any block size used by software.
461 		 */
462 		fp->f_seqcount += howmany(uio->uio_resid, 16384);
463 		if (fp->f_seqcount > IO_SEQMAX)
464 			fp->f_seqcount = IO_SEQMAX;
465 		return (fp->f_seqcount << IO_SEQSHIFT);
466 	}
467 
468 	/* Not sequential.  Quickly draw-down sequentiality. */
469 	if (fp->f_seqcount > 1)
470 		fp->f_seqcount = 1;
471 	else
472 		fp->f_seqcount = 0;
473 	return (0);
474 }
475 
476 /*
477  * Package up an I/O request on a vnode into a uio and do it.
478  */
479 int
480 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
481     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
482     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
483 {
484 	struct uio auio;
485 	struct iovec aiov;
486 	struct mount *mp;
487 	struct ucred *cred;
488 	void *rl_cookie;
489 	struct vn_io_fault_args args;
490 	int error, lock_flags;
491 
492 	auio.uio_iov = &aiov;
493 	auio.uio_iovcnt = 1;
494 	aiov.iov_base = base;
495 	aiov.iov_len = len;
496 	auio.uio_resid = len;
497 	auio.uio_offset = offset;
498 	auio.uio_segflg = segflg;
499 	auio.uio_rw = rw;
500 	auio.uio_td = td;
501 	error = 0;
502 
503 	if ((ioflg & IO_NODELOCKED) == 0) {
504 		if (rw == UIO_READ) {
505 			rl_cookie = vn_rangelock_rlock(vp, offset,
506 			    offset + len);
507 		} else {
508 			rl_cookie = vn_rangelock_wlock(vp, offset,
509 			    offset + len);
510 		}
511 		mp = NULL;
512 		if (rw == UIO_WRITE) {
513 			if (vp->v_type != VCHR &&
514 			    (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
515 			    != 0)
516 				goto out;
517 			if (MNT_SHARED_WRITES(mp) ||
518 			    ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
519 				lock_flags = LK_SHARED;
520 			else
521 				lock_flags = LK_EXCLUSIVE;
522 		} else
523 			lock_flags = LK_SHARED;
524 		vn_lock(vp, lock_flags | LK_RETRY);
525 	} else
526 		rl_cookie = NULL;
527 
528 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
529 #ifdef MAC
530 	if ((ioflg & IO_NOMACCHECK) == 0) {
531 		if (rw == UIO_READ)
532 			error = mac_vnode_check_read(active_cred, file_cred,
533 			    vp);
534 		else
535 			error = mac_vnode_check_write(active_cred, file_cred,
536 			    vp);
537 	}
538 #endif
539 	if (error == 0) {
540 		if (file_cred != NULL)
541 			cred = file_cred;
542 		else
543 			cred = active_cred;
544 		if (do_vn_io_fault(vp, &auio)) {
545 			args.kind = VN_IO_FAULT_VOP;
546 			args.cred = cred;
547 			args.flags = ioflg;
548 			args.args.vop_args.vp = vp;
549 			error = vn_io_fault1(vp, &auio, &args, td);
550 		} else if (rw == UIO_READ) {
551 			error = VOP_READ(vp, &auio, ioflg, cred);
552 		} else /* if (rw == UIO_WRITE) */ {
553 			error = VOP_WRITE(vp, &auio, ioflg, cred);
554 		}
555 	}
556 	if (aresid)
557 		*aresid = auio.uio_resid;
558 	else
559 		if (auio.uio_resid && error == 0)
560 			error = EIO;
561 	if ((ioflg & IO_NODELOCKED) == 0) {
562 		VOP_UNLOCK(vp, 0);
563 		if (mp != NULL)
564 			vn_finished_write(mp);
565 	}
566  out:
567 	if (rl_cookie != NULL)
568 		vn_rangelock_unlock(vp, rl_cookie);
569 	return (error);
570 }
571 
572 /*
573  * Package up an I/O request on a vnode into a uio and do it.  The I/O
574  * request is split up into smaller chunks and we try to avoid saturating
575  * the buffer cache while potentially holding a vnode locked, so we
576  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
577  * to give other processes a chance to lock the vnode (either other processes
578  * core'ing the same binary, or unrelated processes scanning the directory).
579  */
580 int
581 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
582     file_cred, aresid, td)
583 	enum uio_rw rw;
584 	struct vnode *vp;
585 	void *base;
586 	size_t len;
587 	off_t offset;
588 	enum uio_seg segflg;
589 	int ioflg;
590 	struct ucred *active_cred;
591 	struct ucred *file_cred;
592 	size_t *aresid;
593 	struct thread *td;
594 {
595 	int error = 0;
596 	ssize_t iaresid;
597 
598 	do {
599 		int chunk;
600 
601 		/*
602 		 * Force `offset' to a multiple of MAXBSIZE except possibly
603 		 * for the first chunk, so that filesystems only need to
604 		 * write full blocks except possibly for the first and last
605 		 * chunks.
606 		 */
607 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
608 
609 		if (chunk > len)
610 			chunk = len;
611 		if (rw != UIO_READ && vp->v_type == VREG)
612 			bwillwrite();
613 		iaresid = 0;
614 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
615 		    ioflg, active_cred, file_cred, &iaresid, td);
616 		len -= chunk;	/* aresid calc already includes length */
617 		if (error)
618 			break;
619 		offset += chunk;
620 		base = (char *)base + chunk;
621 		kern_yield(PRI_USER);
622 	} while (len);
623 	if (aresid)
624 		*aresid = len + iaresid;
625 	return (error);
626 }
627 
628 off_t
629 foffset_lock(struct file *fp, int flags)
630 {
631 	struct mtx *mtxp;
632 	off_t res;
633 
634 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
635 
636 #if OFF_MAX <= LONG_MAX
637 	/*
638 	 * Caller only wants the current f_offset value.  Assume that
639 	 * the long and shorter integer types reads are atomic.
640 	 */
641 	if ((flags & FOF_NOLOCK) != 0)
642 		return (fp->f_offset);
643 #endif
644 
645 	/*
646 	 * According to McKusick the vn lock was protecting f_offset here.
647 	 * It is now protected by the FOFFSET_LOCKED flag.
648 	 */
649 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
650 	mtx_lock(mtxp);
651 	if ((flags & FOF_NOLOCK) == 0) {
652 		while (fp->f_vnread_flags & FOFFSET_LOCKED) {
653 			fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
654 			msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
655 			    "vofflock", 0);
656 		}
657 		fp->f_vnread_flags |= FOFFSET_LOCKED;
658 	}
659 	res = fp->f_offset;
660 	mtx_unlock(mtxp);
661 	return (res);
662 }
663 
664 void
665 foffset_unlock(struct file *fp, off_t val, int flags)
666 {
667 	struct mtx *mtxp;
668 
669 	KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
670 
671 #if OFF_MAX <= LONG_MAX
672 	if ((flags & FOF_NOLOCK) != 0) {
673 		if ((flags & FOF_NOUPDATE) == 0)
674 			fp->f_offset = val;
675 		if ((flags & FOF_NEXTOFF) != 0)
676 			fp->f_nextoff = val;
677 		return;
678 	}
679 #endif
680 
681 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
682 	mtx_lock(mtxp);
683 	if ((flags & FOF_NOUPDATE) == 0)
684 		fp->f_offset = val;
685 	if ((flags & FOF_NEXTOFF) != 0)
686 		fp->f_nextoff = val;
687 	if ((flags & FOF_NOLOCK) == 0) {
688 		KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
689 		    ("Lost FOFFSET_LOCKED"));
690 		if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
691 			wakeup(&fp->f_vnread_flags);
692 		fp->f_vnread_flags = 0;
693 	}
694 	mtx_unlock(mtxp);
695 }
696 
697 void
698 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
699 {
700 
701 	if ((flags & FOF_OFFSET) == 0)
702 		uio->uio_offset = foffset_lock(fp, flags);
703 }
704 
705 void
706 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
707 {
708 
709 	if ((flags & FOF_OFFSET) == 0)
710 		foffset_unlock(fp, uio->uio_offset, flags);
711 }
712 
713 static int
714 get_advice(struct file *fp, struct uio *uio)
715 {
716 	struct mtx *mtxp;
717 	int ret;
718 
719 	ret = POSIX_FADV_NORMAL;
720 	if (fp->f_advice == NULL)
721 		return (ret);
722 
723 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
724 	mtx_lock(mtxp);
725 	if (uio->uio_offset >= fp->f_advice->fa_start &&
726 	    uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
727 		ret = fp->f_advice->fa_advice;
728 	mtx_unlock(mtxp);
729 	return (ret);
730 }
731 
732 /*
733  * File table vnode read routine.
734  */
735 static int
736 vn_read(fp, uio, active_cred, flags, td)
737 	struct file *fp;
738 	struct uio *uio;
739 	struct ucred *active_cred;
740 	int flags;
741 	struct thread *td;
742 {
743 	struct vnode *vp;
744 	struct mtx *mtxp;
745 	int error, ioflag;
746 	int advice;
747 	off_t offset, start, end;
748 
749 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
750 	    uio->uio_td, td));
751 	KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
752 	vp = fp->f_vnode;
753 	ioflag = 0;
754 	if (fp->f_flag & FNONBLOCK)
755 		ioflag |= IO_NDELAY;
756 	if (fp->f_flag & O_DIRECT)
757 		ioflag |= IO_DIRECT;
758 	advice = get_advice(fp, uio);
759 	vn_lock(vp, LK_SHARED | LK_RETRY);
760 
761 	switch (advice) {
762 	case POSIX_FADV_NORMAL:
763 	case POSIX_FADV_SEQUENTIAL:
764 	case POSIX_FADV_NOREUSE:
765 		ioflag |= sequential_heuristic(uio, fp);
766 		break;
767 	case POSIX_FADV_RANDOM:
768 		/* Disable read-ahead for random I/O. */
769 		break;
770 	}
771 	offset = uio->uio_offset;
772 
773 #ifdef MAC
774 	error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
775 	if (error == 0)
776 #endif
777 		error = VOP_READ(vp, uio, ioflag, fp->f_cred);
778 	fp->f_nextoff = uio->uio_offset;
779 	VOP_UNLOCK(vp, 0);
780 	if (error == 0 && advice == POSIX_FADV_NOREUSE &&
781 	    offset != uio->uio_offset) {
782 		/*
783 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
784 		 * buffers for the backing file after a
785 		 * POSIX_FADV_NOREUSE read(2).  To optimize the common
786 		 * case of using POSIX_FADV_NOREUSE with sequential
787 		 * access, track the previous implicit DONTNEED
788 		 * request and grow this request to include the
789 		 * current read(2) in addition to the previous
790 		 * DONTNEED.  With purely sequential access this will
791 		 * cause the DONTNEED requests to continously grow to
792 		 * cover all of the previously read regions of the
793 		 * file.  This allows filesystem blocks that are
794 		 * accessed by multiple calls to read(2) to be flushed
795 		 * once the last read(2) finishes.
796 		 */
797 		start = offset;
798 		end = uio->uio_offset - 1;
799 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
800 		mtx_lock(mtxp);
801 		if (fp->f_advice != NULL &&
802 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
803 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
804 				start = fp->f_advice->fa_prevstart;
805 			else if (fp->f_advice->fa_prevstart != 0 &&
806 			    fp->f_advice->fa_prevstart == end + 1)
807 				end = fp->f_advice->fa_prevend;
808 			fp->f_advice->fa_prevstart = start;
809 			fp->f_advice->fa_prevend = end;
810 		}
811 		mtx_unlock(mtxp);
812 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
813 	}
814 	return (error);
815 }
816 
817 /*
818  * File table vnode write routine.
819  */
820 static int
821 vn_write(fp, uio, active_cred, flags, td)
822 	struct file *fp;
823 	struct uio *uio;
824 	struct ucred *active_cred;
825 	int flags;
826 	struct thread *td;
827 {
828 	struct vnode *vp;
829 	struct mount *mp;
830 	struct mtx *mtxp;
831 	int error, ioflag, lock_flags;
832 	int advice;
833 	off_t offset, start, end;
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 	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 	    offset != uio->uio_offset) {
889 		/*
890 		 * Use POSIX_FADV_DONTNEED to flush clean pages and
891 		 * buffers for the backing file after a
892 		 * POSIX_FADV_NOREUSE write(2).  To optimize the
893 		 * common case of using POSIX_FADV_NOREUSE with
894 		 * sequential access, track the previous implicit
895 		 * DONTNEED request and grow this request to include
896 		 * the current write(2) in addition to the previous
897 		 * DONTNEED.  With purely sequential access this will
898 		 * cause the DONTNEED requests to continously grow to
899 		 * cover all of the previously written regions of the
900 		 * file.
901 		 *
902 		 * Note that the blocks just written are almost
903 		 * certainly still dirty, so this only works when
904 		 * VOP_ADVISE() calls from subsequent writes push out
905 		 * the data written by this write(2) once the backing
906 		 * buffers are clean.  However, as compared to forcing
907 		 * IO_DIRECT, this gives much saner behavior.  Write
908 		 * clustering is still allowed, and clean pages are
909 		 * merely moved to the cache page queue rather than
910 		 * outright thrown away.  This means a subsequent
911 		 * read(2) can still avoid hitting the disk if the
912 		 * pages have not been reclaimed.
913 		 *
914 		 * This does make POSIX_FADV_NOREUSE largely useless
915 		 * with non-sequential access.  However, sequential
916 		 * access is the more common use case and the flag is
917 		 * merely advisory.
918 		 */
919 		start = offset;
920 		end = uio->uio_offset - 1;
921 		mtxp = mtx_pool_find(mtxpool_sleep, fp);
922 		mtx_lock(mtxp);
923 		if (fp->f_advice != NULL &&
924 		    fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
925 			if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
926 				start = fp->f_advice->fa_prevstart;
927 			else if (fp->f_advice->fa_prevstart != 0 &&
928 			    fp->f_advice->fa_prevstart == end + 1)
929 				end = fp->f_advice->fa_prevend;
930 			fp->f_advice->fa_prevstart = start;
931 			fp->f_advice->fa_prevend = end;
932 		}
933 		mtx_unlock(mtxp);
934 		error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
935 	}
936 
937 unlock:
938 	return (error);
939 }
940 
941 /*
942  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
943  * prevent the following deadlock:
944  *
945  * Assume that the thread A reads from the vnode vp1 into userspace
946  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
947  * currently not resident, then system ends up with the call chain
948  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
949  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
950  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
951  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
952  * backed by the pages of vnode vp1, and some page in buf2 is not
953  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
954  *
955  * To prevent the lock order reversal and deadlock, vn_io_fault() does
956  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
957  * Instead, it first tries to do the whole range i/o with pagefaults
958  * disabled. If all pages in the i/o buffer are resident and mapped,
959  * VOP will succeed (ignoring the genuine filesystem errors).
960  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
961  * i/o in chunks, with all pages in the chunk prefaulted and held
962  * using vm_fault_quick_hold_pages().
963  *
964  * Filesystems using this deadlock avoidance scheme should use the
965  * array of the held pages from uio, saved in the curthread->td_ma,
966  * instead of doing uiomove().  A helper function
967  * vn_io_fault_uiomove() converts uiomove request into
968  * uiomove_fromphys() over td_ma array.
969  *
970  * Since vnode locks do not cover the whole i/o anymore, rangelocks
971  * make the current i/o request atomic with respect to other i/os and
972  * truncations.
973  */
974 
975 /*
976  * Decode vn_io_fault_args and perform the corresponding i/o.
977  */
978 static int
979 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
980     struct thread *td)
981 {
982 
983 	switch (args->kind) {
984 	case VN_IO_FAULT_FOP:
985 		return ((args->args.fop_args.doio)(args->args.fop_args.fp,
986 		    uio, args->cred, args->flags, td));
987 	case VN_IO_FAULT_VOP:
988 		if (uio->uio_rw == UIO_READ) {
989 			return (VOP_READ(args->args.vop_args.vp, uio,
990 			    args->flags, args->cred));
991 		} else if (uio->uio_rw == UIO_WRITE) {
992 			return (VOP_WRITE(args->args.vop_args.vp, uio,
993 			    args->flags, args->cred));
994 		}
995 		break;
996 	}
997 	panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
998 	    uio->uio_rw);
999 }
1000 
1001 /*
1002  * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1003  * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1004  * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1005  * into args and call vn_io_fault1() to handle faults during the user
1006  * mode buffer accesses.
1007  */
1008 static int
1009 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1010     struct thread *td)
1011 {
1012 	vm_page_t ma[io_hold_cnt + 2];
1013 	struct uio *uio_clone, short_uio;
1014 	struct iovec short_iovec[1];
1015 	vm_page_t *prev_td_ma;
1016 	vm_prot_t prot;
1017 	vm_offset_t addr, end;
1018 	size_t len, resid;
1019 	ssize_t adv;
1020 	int error, cnt, save, saveheld, prev_td_ma_cnt;
1021 
1022 	prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1023 
1024 	/*
1025 	 * The UFS follows IO_UNIT directive and replays back both
1026 	 * uio_offset and uio_resid if an error is encountered during the
1027 	 * operation.  But, since the iovec may be already advanced,
1028 	 * uio is still in an inconsistent state.
1029 	 *
1030 	 * Cache a copy of the original uio, which is advanced to the redo
1031 	 * point using UIO_NOCOPY below.
1032 	 */
1033 	uio_clone = cloneuio(uio);
1034 	resid = uio->uio_resid;
1035 
1036 	short_uio.uio_segflg = UIO_USERSPACE;
1037 	short_uio.uio_rw = uio->uio_rw;
1038 	short_uio.uio_td = uio->uio_td;
1039 
1040 	save = vm_fault_disable_pagefaults();
1041 	error = vn_io_fault_doio(args, uio, td);
1042 	if (error != EFAULT)
1043 		goto out;
1044 
1045 	atomic_add_long(&vn_io_faults_cnt, 1);
1046 	uio_clone->uio_segflg = UIO_NOCOPY;
1047 	uiomove(NULL, resid - uio->uio_resid, uio_clone);
1048 	uio_clone->uio_segflg = uio->uio_segflg;
1049 
1050 	saveheld = curthread_pflags_set(TDP_UIOHELD);
1051 	prev_td_ma = td->td_ma;
1052 	prev_td_ma_cnt = td->td_ma_cnt;
1053 
1054 	while (uio_clone->uio_resid != 0) {
1055 		len = uio_clone->uio_iov->iov_len;
1056 		if (len == 0) {
1057 			KASSERT(uio_clone->uio_iovcnt >= 1,
1058 			    ("iovcnt underflow"));
1059 			uio_clone->uio_iov++;
1060 			uio_clone->uio_iovcnt--;
1061 			continue;
1062 		}
1063 		if (len > io_hold_cnt * PAGE_SIZE)
1064 			len = io_hold_cnt * PAGE_SIZE;
1065 		addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1066 		end = round_page(addr + len);
1067 		if (end < addr) {
1068 			error = EFAULT;
1069 			break;
1070 		}
1071 		cnt = atop(end - trunc_page(addr));
1072 		/*
1073 		 * A perfectly misaligned address and length could cause
1074 		 * both the start and the end of the chunk to use partial
1075 		 * page.  +2 accounts for such a situation.
1076 		 */
1077 		cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1078 		    addr, len, prot, ma, io_hold_cnt + 2);
1079 		if (cnt == -1) {
1080 			error = EFAULT;
1081 			break;
1082 		}
1083 		short_uio.uio_iov = &short_iovec[0];
1084 		short_iovec[0].iov_base = (void *)addr;
1085 		short_uio.uio_iovcnt = 1;
1086 		short_uio.uio_resid = short_iovec[0].iov_len = len;
1087 		short_uio.uio_offset = uio_clone->uio_offset;
1088 		td->td_ma = ma;
1089 		td->td_ma_cnt = cnt;
1090 
1091 		error = vn_io_fault_doio(args, &short_uio, td);
1092 		vm_page_unhold_pages(ma, cnt);
1093 		adv = len - short_uio.uio_resid;
1094 
1095 		uio_clone->uio_iov->iov_base =
1096 		    (char *)uio_clone->uio_iov->iov_base + adv;
1097 		uio_clone->uio_iov->iov_len -= adv;
1098 		uio_clone->uio_resid -= adv;
1099 		uio_clone->uio_offset += adv;
1100 
1101 		uio->uio_resid -= adv;
1102 		uio->uio_offset += adv;
1103 
1104 		if (error != 0 || adv == 0)
1105 			break;
1106 	}
1107 	td->td_ma = prev_td_ma;
1108 	td->td_ma_cnt = prev_td_ma_cnt;
1109 	curthread_pflags_restore(saveheld);
1110 out:
1111 	vm_fault_enable_pagefaults(save);
1112 	free(uio_clone, M_IOV);
1113 	return (error);
1114 }
1115 
1116 static int
1117 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1118     int flags, struct thread *td)
1119 {
1120 	fo_rdwr_t *doio;
1121 	struct vnode *vp;
1122 	void *rl_cookie;
1123 	struct vn_io_fault_args args;
1124 	int error;
1125 
1126 	doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1127 	vp = fp->f_vnode;
1128 	foffset_lock_uio(fp, uio, flags);
1129 	if (do_vn_io_fault(vp, uio)) {
1130 		args.kind = VN_IO_FAULT_FOP;
1131 		args.args.fop_args.fp = fp;
1132 		args.args.fop_args.doio = doio;
1133 		args.cred = active_cred;
1134 		args.flags = flags | FOF_OFFSET;
1135 		if (uio->uio_rw == UIO_READ) {
1136 			rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1137 			    uio->uio_offset + uio->uio_resid);
1138 		} else if ((fp->f_flag & O_APPEND) != 0 ||
1139 		    (flags & FOF_OFFSET) == 0) {
1140 			/* For appenders, punt and lock the whole range. */
1141 			rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1142 		} else {
1143 			rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1144 			    uio->uio_offset + uio->uio_resid);
1145 		}
1146 		error = vn_io_fault1(vp, uio, &args, td);
1147 		vn_rangelock_unlock(vp, rl_cookie);
1148 	} else {
1149 		error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1150 	}
1151 	foffset_unlock_uio(fp, uio, flags);
1152 	return (error);
1153 }
1154 
1155 /*
1156  * Helper function to perform the requested uiomove operation using
1157  * the held pages for io->uio_iov[0].iov_base buffer instead of
1158  * copyin/copyout.  Access to the pages with uiomove_fromphys()
1159  * instead of iov_base prevents page faults that could occur due to
1160  * pmap_collect() invalidating the mapping created by
1161  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1162  * object cleanup revoking the write access from page mappings.
1163  *
1164  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1165  * instead of plain uiomove().
1166  */
1167 int
1168 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1169 {
1170 	struct uio transp_uio;
1171 	struct iovec transp_iov[1];
1172 	struct thread *td;
1173 	size_t adv;
1174 	int error, pgadv;
1175 
1176 	td = curthread;
1177 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1178 	    uio->uio_segflg != UIO_USERSPACE)
1179 		return (uiomove(data, xfersize, uio));
1180 
1181 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1182 	transp_iov[0].iov_base = data;
1183 	transp_uio.uio_iov = &transp_iov[0];
1184 	transp_uio.uio_iovcnt = 1;
1185 	if (xfersize > uio->uio_resid)
1186 		xfersize = uio->uio_resid;
1187 	transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1188 	transp_uio.uio_offset = 0;
1189 	transp_uio.uio_segflg = UIO_SYSSPACE;
1190 	/*
1191 	 * Since transp_iov points to data, and td_ma page array
1192 	 * corresponds to original uio->uio_iov, we need to invert the
1193 	 * direction of the i/o operation as passed to
1194 	 * uiomove_fromphys().
1195 	 */
1196 	switch (uio->uio_rw) {
1197 	case UIO_WRITE:
1198 		transp_uio.uio_rw = UIO_READ;
1199 		break;
1200 	case UIO_READ:
1201 		transp_uio.uio_rw = UIO_WRITE;
1202 		break;
1203 	}
1204 	transp_uio.uio_td = uio->uio_td;
1205 	error = uiomove_fromphys(td->td_ma,
1206 	    ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1207 	    xfersize, &transp_uio);
1208 	adv = xfersize - transp_uio.uio_resid;
1209 	pgadv =
1210 	    (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1211 	    (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1212 	td->td_ma += pgadv;
1213 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1214 	    pgadv));
1215 	td->td_ma_cnt -= pgadv;
1216 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1217 	uio->uio_iov->iov_len -= adv;
1218 	uio->uio_resid -= adv;
1219 	uio->uio_offset += adv;
1220 	return (error);
1221 }
1222 
1223 int
1224 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1225     struct uio *uio)
1226 {
1227 	struct thread *td;
1228 	vm_offset_t iov_base;
1229 	int cnt, pgadv;
1230 
1231 	td = curthread;
1232 	if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1233 	    uio->uio_segflg != UIO_USERSPACE)
1234 		return (uiomove_fromphys(ma, offset, xfersize, uio));
1235 
1236 	KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1237 	cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1238 	iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1239 	switch (uio->uio_rw) {
1240 	case UIO_WRITE:
1241 		pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1242 		    offset, cnt);
1243 		break;
1244 	case UIO_READ:
1245 		pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1246 		    cnt);
1247 		break;
1248 	}
1249 	pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1250 	td->td_ma += pgadv;
1251 	KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1252 	    pgadv));
1253 	td->td_ma_cnt -= pgadv;
1254 	uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1255 	uio->uio_iov->iov_len -= cnt;
1256 	uio->uio_resid -= cnt;
1257 	uio->uio_offset += cnt;
1258 	return (0);
1259 }
1260 
1261 
1262 /*
1263  * File table truncate routine.
1264  */
1265 static int
1266 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1267     struct thread *td)
1268 {
1269 	struct vattr vattr;
1270 	struct mount *mp;
1271 	struct vnode *vp;
1272 	void *rl_cookie;
1273 	int error;
1274 
1275 	vp = fp->f_vnode;
1276 
1277 	/*
1278 	 * Lock the whole range for truncation.  Otherwise split i/o
1279 	 * might happen partly before and partly after the truncation.
1280 	 */
1281 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1282 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1283 	if (error)
1284 		goto out1;
1285 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1286 	if (vp->v_type == VDIR) {
1287 		error = EISDIR;
1288 		goto out;
1289 	}
1290 #ifdef MAC
1291 	error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1292 	if (error)
1293 		goto out;
1294 #endif
1295 	error = vn_writechk(vp);
1296 	if (error == 0) {
1297 		VATTR_NULL(&vattr);
1298 		vattr.va_size = length;
1299 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1300 	}
1301 out:
1302 	VOP_UNLOCK(vp, 0);
1303 	vn_finished_write(mp);
1304 out1:
1305 	vn_rangelock_unlock(vp, rl_cookie);
1306 	return (error);
1307 }
1308 
1309 /*
1310  * File table vnode stat routine.
1311  */
1312 static int
1313 vn_statfile(fp, sb, active_cred, td)
1314 	struct file *fp;
1315 	struct stat *sb;
1316 	struct ucred *active_cred;
1317 	struct thread *td;
1318 {
1319 	struct vnode *vp = fp->f_vnode;
1320 	int error;
1321 
1322 	vn_lock(vp, LK_SHARED | LK_RETRY);
1323 	error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1324 	VOP_UNLOCK(vp, 0);
1325 
1326 	return (error);
1327 }
1328 
1329 /*
1330  * Stat a vnode; implementation for the stat syscall
1331  */
1332 int
1333 vn_stat(vp, sb, active_cred, file_cred, td)
1334 	struct vnode *vp;
1335 	register struct stat *sb;
1336 	struct ucred *active_cred;
1337 	struct ucred *file_cred;
1338 	struct thread *td;
1339 {
1340 	struct vattr vattr;
1341 	register struct vattr *vap;
1342 	int error;
1343 	u_short mode;
1344 
1345 #ifdef MAC
1346 	error = mac_vnode_check_stat(active_cred, file_cred, vp);
1347 	if (error)
1348 		return (error);
1349 #endif
1350 
1351 	vap = &vattr;
1352 
1353 	/*
1354 	 * Initialize defaults for new and unusual fields, so that file
1355 	 * systems which don't support these fields don't need to know
1356 	 * about them.
1357 	 */
1358 	vap->va_birthtime.tv_sec = -1;
1359 	vap->va_birthtime.tv_nsec = 0;
1360 	vap->va_fsid = VNOVAL;
1361 	vap->va_rdev = NODEV;
1362 
1363 	error = VOP_GETATTR(vp, vap, active_cred);
1364 	if (error)
1365 		return (error);
1366 
1367 	/*
1368 	 * Zero the spare stat fields
1369 	 */
1370 	bzero(sb, sizeof *sb);
1371 
1372 	/*
1373 	 * Copy from vattr table
1374 	 */
1375 	if (vap->va_fsid != VNOVAL)
1376 		sb->st_dev = vap->va_fsid;
1377 	else
1378 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1379 	sb->st_ino = vap->va_fileid;
1380 	mode = vap->va_mode;
1381 	switch (vap->va_type) {
1382 	case VREG:
1383 		mode |= S_IFREG;
1384 		break;
1385 	case VDIR:
1386 		mode |= S_IFDIR;
1387 		break;
1388 	case VBLK:
1389 		mode |= S_IFBLK;
1390 		break;
1391 	case VCHR:
1392 		mode |= S_IFCHR;
1393 		break;
1394 	case VLNK:
1395 		mode |= S_IFLNK;
1396 		break;
1397 	case VSOCK:
1398 		mode |= S_IFSOCK;
1399 		break;
1400 	case VFIFO:
1401 		mode |= S_IFIFO;
1402 		break;
1403 	default:
1404 		return (EBADF);
1405 	};
1406 	sb->st_mode = mode;
1407 	sb->st_nlink = vap->va_nlink;
1408 	sb->st_uid = vap->va_uid;
1409 	sb->st_gid = vap->va_gid;
1410 	sb->st_rdev = vap->va_rdev;
1411 	if (vap->va_size > OFF_MAX)
1412 		return (EOVERFLOW);
1413 	sb->st_size = vap->va_size;
1414 	sb->st_atim = vap->va_atime;
1415 	sb->st_mtim = vap->va_mtime;
1416 	sb->st_ctim = vap->va_ctime;
1417 	sb->st_birthtim = vap->va_birthtime;
1418 
1419         /*
1420 	 * According to www.opengroup.org, the meaning of st_blksize is
1421 	 *   "a filesystem-specific preferred I/O block size for this
1422 	 *    object.  In some filesystem types, this may vary from file
1423 	 *    to file"
1424 	 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1425 	 */
1426 
1427 	sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1428 
1429 	sb->st_flags = vap->va_flags;
1430 	if (priv_check(td, PRIV_VFS_GENERATION))
1431 		sb->st_gen = 0;
1432 	else
1433 		sb->st_gen = vap->va_gen;
1434 
1435 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1436 	return (0);
1437 }
1438 
1439 /*
1440  * File table vnode ioctl routine.
1441  */
1442 static int
1443 vn_ioctl(fp, com, data, active_cred, td)
1444 	struct file *fp;
1445 	u_long com;
1446 	void *data;
1447 	struct ucred *active_cred;
1448 	struct thread *td;
1449 {
1450 	struct vattr vattr;
1451 	struct vnode *vp;
1452 	int error;
1453 
1454 	vp = fp->f_vnode;
1455 	switch (vp->v_type) {
1456 	case VDIR:
1457 	case VREG:
1458 		switch (com) {
1459 		case FIONREAD:
1460 			vn_lock(vp, LK_SHARED | LK_RETRY);
1461 			error = VOP_GETATTR(vp, &vattr, active_cred);
1462 			VOP_UNLOCK(vp, 0);
1463 			if (error == 0)
1464 				*(int *)data = vattr.va_size - fp->f_offset;
1465 			return (error);
1466 		case FIONBIO:
1467 		case FIOASYNC:
1468 			return (0);
1469 		default:
1470 			return (VOP_IOCTL(vp, com, data, fp->f_flag,
1471 			    active_cred, td));
1472 		}
1473 	default:
1474 		return (ENOTTY);
1475 	}
1476 }
1477 
1478 /*
1479  * File table vnode poll routine.
1480  */
1481 static int
1482 vn_poll(fp, events, active_cred, td)
1483 	struct file *fp;
1484 	int events;
1485 	struct ucred *active_cred;
1486 	struct thread *td;
1487 {
1488 	struct vnode *vp;
1489 	int error;
1490 
1491 	vp = fp->f_vnode;
1492 #ifdef MAC
1493 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1494 	error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1495 	VOP_UNLOCK(vp, 0);
1496 	if (!error)
1497 #endif
1498 
1499 	error = VOP_POLL(vp, events, fp->f_cred, td);
1500 	return (error);
1501 }
1502 
1503 /*
1504  * Acquire the requested lock and then check for validity.  LK_RETRY
1505  * permits vn_lock to return doomed vnodes.
1506  */
1507 int
1508 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1509 {
1510 	int error;
1511 
1512 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1513 	    ("vn_lock called with no locktype."));
1514 	do {
1515 #ifdef DEBUG_VFS_LOCKS
1516 		KASSERT(vp->v_holdcnt != 0,
1517 		    ("vn_lock %p: zero hold count", vp));
1518 #endif
1519 		error = VOP_LOCK1(vp, flags, file, line);
1520 		flags &= ~LK_INTERLOCK;	/* Interlock is always dropped. */
1521 		KASSERT((flags & LK_RETRY) == 0 || error == 0,
1522 		    ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1523 		    flags, error));
1524 		/*
1525 		 * Callers specify LK_RETRY if they wish to get dead vnodes.
1526 		 * If RETRY is not set, we return ENOENT instead.
1527 		 */
1528 		if (error == 0 && vp->v_iflag & VI_DOOMED &&
1529 		    (flags & LK_RETRY) == 0) {
1530 			VOP_UNLOCK(vp, 0);
1531 			error = ENOENT;
1532 			break;
1533 		}
1534 	} while (flags & LK_RETRY && error != 0);
1535 	return (error);
1536 }
1537 
1538 /*
1539  * File table vnode close routine.
1540  */
1541 static int
1542 vn_closefile(fp, td)
1543 	struct file *fp;
1544 	struct thread *td;
1545 {
1546 	struct vnode *vp;
1547 	struct flock lf;
1548 	int error;
1549 
1550 	vp = fp->f_vnode;
1551 	fp->f_ops = &badfileops;
1552 
1553 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1554 		vref(vp);
1555 
1556 	error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1557 
1558 	if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1559 		lf.l_whence = SEEK_SET;
1560 		lf.l_start = 0;
1561 		lf.l_len = 0;
1562 		lf.l_type = F_UNLCK;
1563 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1564 		vrele(vp);
1565 	}
1566 	return (error);
1567 }
1568 
1569 /*
1570  * Preparing to start a filesystem write operation. If the operation is
1571  * permitted, then we bump the count of operations in progress and
1572  * proceed. If a suspend request is in progress, we wait until the
1573  * suspension is over, and then proceed.
1574  */
1575 static int
1576 vn_start_write_locked(struct mount *mp, int flags)
1577 {
1578 	int error;
1579 
1580 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1581 	error = 0;
1582 
1583 	/*
1584 	 * Check on status of suspension.
1585 	 */
1586 	if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1587 	    mp->mnt_susp_owner != curthread) {
1588 		while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1589 			if (flags & V_NOWAIT) {
1590 				error = EWOULDBLOCK;
1591 				goto unlock;
1592 			}
1593 			error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1594 			    (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1595 			if (error)
1596 				goto unlock;
1597 		}
1598 	}
1599 	if (flags & V_XSLEEP)
1600 		goto unlock;
1601 	mp->mnt_writeopcount++;
1602 unlock:
1603 	if (error != 0 || (flags & V_XSLEEP) != 0)
1604 		MNT_REL(mp);
1605 	MNT_IUNLOCK(mp);
1606 	return (error);
1607 }
1608 
1609 int
1610 vn_start_write(vp, mpp, flags)
1611 	struct vnode *vp;
1612 	struct mount **mpp;
1613 	int flags;
1614 {
1615 	struct mount *mp;
1616 	int error;
1617 
1618 	error = 0;
1619 	/*
1620 	 * If a vnode is provided, get and return the mount point that
1621 	 * to which it will write.
1622 	 */
1623 	if (vp != NULL) {
1624 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1625 			*mpp = NULL;
1626 			if (error != EOPNOTSUPP)
1627 				return (error);
1628 			return (0);
1629 		}
1630 	}
1631 	if ((mp = *mpp) == NULL)
1632 		return (0);
1633 
1634 	/*
1635 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1636 	 * a vfs_ref().
1637 	 * As long as a vnode is not provided we need to acquire a
1638 	 * refcount for the provided mountpoint too, in order to
1639 	 * emulate a vfs_ref().
1640 	 */
1641 	MNT_ILOCK(mp);
1642 	if (vp == NULL)
1643 		MNT_REF(mp);
1644 
1645 	return (vn_start_write_locked(mp, flags));
1646 }
1647 
1648 /*
1649  * Secondary suspension. Used by operations such as vop_inactive
1650  * routines that are needed by the higher level functions. These
1651  * are allowed to proceed until all the higher level functions have
1652  * completed (indicated by mnt_writeopcount dropping to zero). At that
1653  * time, these operations are halted until the suspension is over.
1654  */
1655 int
1656 vn_start_secondary_write(vp, mpp, flags)
1657 	struct vnode *vp;
1658 	struct mount **mpp;
1659 	int flags;
1660 {
1661 	struct mount *mp;
1662 	int error;
1663 
1664  retry:
1665 	if (vp != NULL) {
1666 		if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1667 			*mpp = NULL;
1668 			if (error != EOPNOTSUPP)
1669 				return (error);
1670 			return (0);
1671 		}
1672 	}
1673 	/*
1674 	 * If we are not suspended or have not yet reached suspended
1675 	 * mode, then let the operation proceed.
1676 	 */
1677 	if ((mp = *mpp) == NULL)
1678 		return (0);
1679 
1680 	/*
1681 	 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1682 	 * a vfs_ref().
1683 	 * As long as a vnode is not provided we need to acquire a
1684 	 * refcount for the provided mountpoint too, in order to
1685 	 * emulate a vfs_ref().
1686 	 */
1687 	MNT_ILOCK(mp);
1688 	if (vp == NULL)
1689 		MNT_REF(mp);
1690 	if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1691 		mp->mnt_secondary_writes++;
1692 		mp->mnt_secondary_accwrites++;
1693 		MNT_IUNLOCK(mp);
1694 		return (0);
1695 	}
1696 	if (flags & V_NOWAIT) {
1697 		MNT_REL(mp);
1698 		MNT_IUNLOCK(mp);
1699 		return (EWOULDBLOCK);
1700 	}
1701 	/*
1702 	 * Wait for the suspension to finish.
1703 	 */
1704 	error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1705 		       (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1706 	vfs_rel(mp);
1707 	if (error == 0)
1708 		goto retry;
1709 	return (error);
1710 }
1711 
1712 /*
1713  * Filesystem write operation has completed. If we are suspending and this
1714  * operation is the last one, notify the suspender that the suspension is
1715  * now in effect.
1716  */
1717 void
1718 vn_finished_write(mp)
1719 	struct mount *mp;
1720 {
1721 	if (mp == NULL)
1722 		return;
1723 	MNT_ILOCK(mp);
1724 	MNT_REL(mp);
1725 	mp->mnt_writeopcount--;
1726 	if (mp->mnt_writeopcount < 0)
1727 		panic("vn_finished_write: neg cnt");
1728 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1729 	    mp->mnt_writeopcount <= 0)
1730 		wakeup(&mp->mnt_writeopcount);
1731 	MNT_IUNLOCK(mp);
1732 }
1733 
1734 
1735 /*
1736  * Filesystem secondary write operation has completed. If we are
1737  * suspending and this operation is the last one, notify the suspender
1738  * that the suspension is now in effect.
1739  */
1740 void
1741 vn_finished_secondary_write(mp)
1742 	struct mount *mp;
1743 {
1744 	if (mp == NULL)
1745 		return;
1746 	MNT_ILOCK(mp);
1747 	MNT_REL(mp);
1748 	mp->mnt_secondary_writes--;
1749 	if (mp->mnt_secondary_writes < 0)
1750 		panic("vn_finished_secondary_write: neg cnt");
1751 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1752 	    mp->mnt_secondary_writes <= 0)
1753 		wakeup(&mp->mnt_secondary_writes);
1754 	MNT_IUNLOCK(mp);
1755 }
1756 
1757 
1758 
1759 /*
1760  * Request a filesystem to suspend write operations.
1761  */
1762 int
1763 vfs_write_suspend(struct mount *mp, int flags)
1764 {
1765 	int error;
1766 
1767 	MNT_ILOCK(mp);
1768 	if (mp->mnt_susp_owner == curthread) {
1769 		MNT_IUNLOCK(mp);
1770 		return (EALREADY);
1771 	}
1772 	while (mp->mnt_kern_flag & MNTK_SUSPEND)
1773 		msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1774 
1775 	/*
1776 	 * Unmount holds a write reference on the mount point.  If we
1777 	 * own busy reference and drain for writers, we deadlock with
1778 	 * the reference draining in the unmount path.  Callers of
1779 	 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1780 	 * vfs_busy() reference is owned and caller is not in the
1781 	 * unmount context.
1782 	 */
1783 	if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1784 	    (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1785 		MNT_IUNLOCK(mp);
1786 		return (EBUSY);
1787 	}
1788 
1789 	mp->mnt_kern_flag |= MNTK_SUSPEND;
1790 	mp->mnt_susp_owner = curthread;
1791 	if (mp->mnt_writeopcount > 0)
1792 		(void) msleep(&mp->mnt_writeopcount,
1793 		    MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1794 	else
1795 		MNT_IUNLOCK(mp);
1796 	if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1797 		vfs_write_resume(mp, 0);
1798 	return (error);
1799 }
1800 
1801 /*
1802  * Request a filesystem to resume write operations.
1803  */
1804 void
1805 vfs_write_resume(struct mount *mp, int flags)
1806 {
1807 
1808 	MNT_ILOCK(mp);
1809 	if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1810 		KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1811 		mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1812 				       MNTK_SUSPENDED);
1813 		mp->mnt_susp_owner = NULL;
1814 		wakeup(&mp->mnt_writeopcount);
1815 		wakeup(&mp->mnt_flag);
1816 		curthread->td_pflags &= ~TDP_IGNSUSP;
1817 		if ((flags & VR_START_WRITE) != 0) {
1818 			MNT_REF(mp);
1819 			mp->mnt_writeopcount++;
1820 		}
1821 		MNT_IUNLOCK(mp);
1822 		if ((flags & VR_NO_SUSPCLR) == 0)
1823 			VFS_SUSP_CLEAN(mp);
1824 	} else if ((flags & VR_START_WRITE) != 0) {
1825 		MNT_REF(mp);
1826 		vn_start_write_locked(mp, 0);
1827 	} else {
1828 		MNT_IUNLOCK(mp);
1829 	}
1830 }
1831 
1832 /*
1833  * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1834  * methods.
1835  */
1836 int
1837 vfs_write_suspend_umnt(struct mount *mp)
1838 {
1839 	int error;
1840 
1841 	KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1842 	    ("vfs_write_suspend_umnt: recursed"));
1843 
1844 	/* dounmount() already called vn_start_write(). */
1845 	for (;;) {
1846 		vn_finished_write(mp);
1847 		error = vfs_write_suspend(mp, 0);
1848 		if (error != 0)
1849 			return (error);
1850 		MNT_ILOCK(mp);
1851 		if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1852 			break;
1853 		MNT_IUNLOCK(mp);
1854 		vn_start_write(NULL, &mp, V_WAIT);
1855 	}
1856 	mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1857 	wakeup(&mp->mnt_flag);
1858 	MNT_IUNLOCK(mp);
1859 	curthread->td_pflags |= TDP_IGNSUSP;
1860 	return (0);
1861 }
1862 
1863 /*
1864  * Implement kqueues for files by translating it to vnode operation.
1865  */
1866 static int
1867 vn_kqfilter(struct file *fp, struct knote *kn)
1868 {
1869 
1870 	return (VOP_KQFILTER(fp->f_vnode, kn));
1871 }
1872 
1873 /*
1874  * Simplified in-kernel wrapper calls for extended attribute access.
1875  * Both calls pass in a NULL credential, authorizing as "kernel" access.
1876  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1877  */
1878 int
1879 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1880     const char *attrname, int *buflen, char *buf, struct thread *td)
1881 {
1882 	struct uio	auio;
1883 	struct iovec	iov;
1884 	int	error;
1885 
1886 	iov.iov_len = *buflen;
1887 	iov.iov_base = buf;
1888 
1889 	auio.uio_iov = &iov;
1890 	auio.uio_iovcnt = 1;
1891 	auio.uio_rw = UIO_READ;
1892 	auio.uio_segflg = UIO_SYSSPACE;
1893 	auio.uio_td = td;
1894 	auio.uio_offset = 0;
1895 	auio.uio_resid = *buflen;
1896 
1897 	if ((ioflg & IO_NODELOCKED) == 0)
1898 		vn_lock(vp, LK_SHARED | LK_RETRY);
1899 
1900 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1901 
1902 	/* authorize attribute retrieval as kernel */
1903 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1904 	    td);
1905 
1906 	if ((ioflg & IO_NODELOCKED) == 0)
1907 		VOP_UNLOCK(vp, 0);
1908 
1909 	if (error == 0) {
1910 		*buflen = *buflen - auio.uio_resid;
1911 	}
1912 
1913 	return (error);
1914 }
1915 
1916 /*
1917  * XXX failure mode if partially written?
1918  */
1919 int
1920 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1921     const char *attrname, int buflen, char *buf, struct thread *td)
1922 {
1923 	struct uio	auio;
1924 	struct iovec	iov;
1925 	struct mount	*mp;
1926 	int	error;
1927 
1928 	iov.iov_len = buflen;
1929 	iov.iov_base = buf;
1930 
1931 	auio.uio_iov = &iov;
1932 	auio.uio_iovcnt = 1;
1933 	auio.uio_rw = UIO_WRITE;
1934 	auio.uio_segflg = UIO_SYSSPACE;
1935 	auio.uio_td = td;
1936 	auio.uio_offset = 0;
1937 	auio.uio_resid = buflen;
1938 
1939 	if ((ioflg & IO_NODELOCKED) == 0) {
1940 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1941 			return (error);
1942 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1943 	}
1944 
1945 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1946 
1947 	/* authorize attribute setting as kernel */
1948 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1949 
1950 	if ((ioflg & IO_NODELOCKED) == 0) {
1951 		vn_finished_write(mp);
1952 		VOP_UNLOCK(vp, 0);
1953 	}
1954 
1955 	return (error);
1956 }
1957 
1958 int
1959 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1960     const char *attrname, struct thread *td)
1961 {
1962 	struct mount	*mp;
1963 	int	error;
1964 
1965 	if ((ioflg & IO_NODELOCKED) == 0) {
1966 		if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1967 			return (error);
1968 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1969 	}
1970 
1971 	ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1972 
1973 	/* authorize attribute removal as kernel */
1974 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1975 	if (error == EOPNOTSUPP)
1976 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1977 		    NULL, td);
1978 
1979 	if ((ioflg & IO_NODELOCKED) == 0) {
1980 		vn_finished_write(mp);
1981 		VOP_UNLOCK(vp, 0);
1982 	}
1983 
1984 	return (error);
1985 }
1986 
1987 static int
1988 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
1989     struct vnode **rvp)
1990 {
1991 
1992 	return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
1993 }
1994 
1995 int
1996 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1997 {
1998 
1999 	return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2000 	    lkflags, rvp));
2001 }
2002 
2003 int
2004 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2005     int lkflags, struct vnode **rvp)
2006 {
2007 	struct mount *mp;
2008 	int ltype, error;
2009 
2010 	ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2011 	mp = vp->v_mount;
2012 	ltype = VOP_ISLOCKED(vp);
2013 	KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2014 	    ("vn_vget_ino: vp not locked"));
2015 	error = vfs_busy(mp, MBF_NOWAIT);
2016 	if (error != 0) {
2017 		vfs_ref(mp);
2018 		VOP_UNLOCK(vp, 0);
2019 		error = vfs_busy(mp, 0);
2020 		vn_lock(vp, ltype | LK_RETRY);
2021 		vfs_rel(mp);
2022 		if (error != 0)
2023 			return (ENOENT);
2024 		if (vp->v_iflag & VI_DOOMED) {
2025 			vfs_unbusy(mp);
2026 			return (ENOENT);
2027 		}
2028 	}
2029 	VOP_UNLOCK(vp, 0);
2030 	error = alloc(mp, alloc_arg, lkflags, rvp);
2031 	vfs_unbusy(mp);
2032 	if (*rvp != vp)
2033 		vn_lock(vp, ltype | LK_RETRY);
2034 	if (vp->v_iflag & VI_DOOMED) {
2035 		if (error == 0) {
2036 			if (*rvp == vp)
2037 				vunref(vp);
2038 			else
2039 				vput(*rvp);
2040 		}
2041 		error = ENOENT;
2042 	}
2043 	return (error);
2044 }
2045 
2046 int
2047 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2048     const struct thread *td)
2049 {
2050 
2051 	if (vp->v_type != VREG || td == NULL)
2052 		return (0);
2053 	PROC_LOCK(td->td_proc);
2054 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
2055 	    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
2056 		kern_psignal(td->td_proc, SIGXFSZ);
2057 		PROC_UNLOCK(td->td_proc);
2058 		return (EFBIG);
2059 	}
2060 	PROC_UNLOCK(td->td_proc);
2061 	return (0);
2062 }
2063 
2064 int
2065 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2066     struct thread *td)
2067 {
2068 	struct vnode *vp;
2069 
2070 	vp = fp->f_vnode;
2071 #ifdef AUDIT
2072 	vn_lock(vp, LK_SHARED | LK_RETRY);
2073 	AUDIT_ARG_VNODE1(vp);
2074 	VOP_UNLOCK(vp, 0);
2075 #endif
2076 	return (setfmode(td, active_cred, vp, mode));
2077 }
2078 
2079 int
2080 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2081     struct thread *td)
2082 {
2083 	struct vnode *vp;
2084 
2085 	vp = fp->f_vnode;
2086 #ifdef AUDIT
2087 	vn_lock(vp, LK_SHARED | LK_RETRY);
2088 	AUDIT_ARG_VNODE1(vp);
2089 	VOP_UNLOCK(vp, 0);
2090 #endif
2091 	return (setfown(td, active_cred, vp, uid, gid));
2092 }
2093 
2094 void
2095 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2096 {
2097 	vm_object_t object;
2098 
2099 	if ((object = vp->v_object) == NULL)
2100 		return;
2101 	VM_OBJECT_WLOCK(object);
2102 	vm_object_page_remove(object, start, end, 0);
2103 	VM_OBJECT_WUNLOCK(object);
2104 }
2105 
2106 int
2107 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2108 {
2109 	struct vattr va;
2110 	daddr_t bn, bnp;
2111 	uint64_t bsize;
2112 	off_t noff;
2113 	int error;
2114 
2115 	KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2116 	    ("Wrong command %lu", cmd));
2117 
2118 	if (vn_lock(vp, LK_SHARED) != 0)
2119 		return (EBADF);
2120 	if (vp->v_type != VREG) {
2121 		error = ENOTTY;
2122 		goto unlock;
2123 	}
2124 	error = VOP_GETATTR(vp, &va, cred);
2125 	if (error != 0)
2126 		goto unlock;
2127 	noff = *off;
2128 	if (noff >= va.va_size) {
2129 		error = ENXIO;
2130 		goto unlock;
2131 	}
2132 	bsize = vp->v_mount->mnt_stat.f_iosize;
2133 	for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2134 		error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2135 		if (error == EOPNOTSUPP) {
2136 			error = ENOTTY;
2137 			goto unlock;
2138 		}
2139 		if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2140 		    (bnp != -1 && cmd == FIOSEEKDATA)) {
2141 			noff = bn * bsize;
2142 			if (noff < *off)
2143 				noff = *off;
2144 			goto unlock;
2145 		}
2146 	}
2147 	if (noff > va.va_size)
2148 		noff = va.va_size;
2149 	/* noff == va.va_size. There is an implicit hole at the end of file. */
2150 	if (cmd == FIOSEEKDATA)
2151 		error = ENXIO;
2152 unlock:
2153 	VOP_UNLOCK(vp, 0);
2154 	if (error == 0)
2155 		*off = noff;
2156 	return (error);
2157 }
2158 
2159 int
2160 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2161 {
2162 	struct ucred *cred;
2163 	struct vnode *vp;
2164 	struct vattr vattr;
2165 	off_t foffset, size;
2166 	int error, noneg;
2167 
2168 	cred = td->td_ucred;
2169 	vp = fp->f_vnode;
2170 	foffset = foffset_lock(fp, 0);
2171 	noneg = (vp->v_type != VCHR);
2172 	error = 0;
2173 	switch (whence) {
2174 	case L_INCR:
2175 		if (noneg &&
2176 		    (foffset < 0 ||
2177 		    (offset > 0 && foffset > OFF_MAX - offset))) {
2178 			error = EOVERFLOW;
2179 			break;
2180 		}
2181 		offset += foffset;
2182 		break;
2183 	case L_XTND:
2184 		vn_lock(vp, LK_SHARED | LK_RETRY);
2185 		error = VOP_GETATTR(vp, &vattr, cred);
2186 		VOP_UNLOCK(vp, 0);
2187 		if (error)
2188 			break;
2189 
2190 		/*
2191 		 * If the file references a disk device, then fetch
2192 		 * the media size and use that to determine the ending
2193 		 * offset.
2194 		 */
2195 		if (vattr.va_size == 0 && vp->v_type == VCHR &&
2196 		    fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2197 			vattr.va_size = size;
2198 		if (noneg &&
2199 		    (vattr.va_size > OFF_MAX ||
2200 		    (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2201 			error = EOVERFLOW;
2202 			break;
2203 		}
2204 		offset += vattr.va_size;
2205 		break;
2206 	case L_SET:
2207 		break;
2208 	case SEEK_DATA:
2209 		error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2210 		break;
2211 	case SEEK_HOLE:
2212 		error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2213 		break;
2214 	default:
2215 		error = EINVAL;
2216 	}
2217 	if (error == 0 && noneg && offset < 0)
2218 		error = EINVAL;
2219 	if (error != 0)
2220 		goto drop;
2221 	VFS_KNOTE_UNLOCKED(vp, 0);
2222 	td->td_uretoff.tdu_off = offset;
2223 drop:
2224 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2225 	return (error);
2226 }
2227 
2228 int
2229 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2230     struct thread *td)
2231 {
2232 	int error;
2233 
2234 	error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2235 
2236 	/*
2237 	 * From utimes(2):
2238 	 * Grant permission if the caller is the owner of the file or
2239 	 * the super-user.  If the time pointer is null, then write
2240 	 * permission on the file is also sufficient.
2241 	 *
2242 	 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2243 	 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2244 	 * will be allowed to set the times [..] to the current
2245 	 * server time.
2246 	 */
2247 	if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2248 		error = VOP_ACCESS(vp, VWRITE, cred, td);
2249 	return (error);
2250 }
2251