xref: /netbsd-src/sys/kern/vfs_vnops.c (revision 7d62b00eb9ad855ffcd7da46b41e23feb5476fac)
1 /*	$NetBSD: vfs_vnops.c,v 1.237 2023/03/13 18:13:18 riastradh Exp $	*/
2 
3 /*-
4  * Copyright (c) 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1982, 1986, 1989, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  * (c) UNIX System Laboratories, Inc.
36  * All or some portions of this file are derived from material licensed
37  * to the University of California by American Telephone and Telegraph
38  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
39  * the permission of UNIX System Laboratories, Inc.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)vfs_vnops.c	8.14 (Berkeley) 6/15/95
66  */
67 
68 #include <sys/cdefs.h>
69 __KERNEL_RCSID(0, "$NetBSD: vfs_vnops.c,v 1.237 2023/03/13 18:13:18 riastradh Exp $");
70 
71 #include "veriexec.h"
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/kernel.h>
76 #include <sys/file.h>
77 #include <sys/stat.h>
78 #include <sys/buf.h>
79 #include <sys/proc.h>
80 #include <sys/mount.h>
81 #include <sys/namei.h>
82 #include <sys/vnode_impl.h>
83 #include <sys/ioctl.h>
84 #include <sys/tty.h>
85 #include <sys/poll.h>
86 #include <sys/kauth.h>
87 #include <sys/syslog.h>
88 #include <sys/fstrans.h>
89 #include <sys/atomic.h>
90 #include <sys/filedesc.h>
91 #include <sys/wapbl.h>
92 #include <sys/mman.h>
93 
94 #include <miscfs/specfs/specdev.h>
95 #include <miscfs/fifofs/fifo.h>
96 
97 #include <uvm/uvm_extern.h>
98 #include <uvm/uvm_readahead.h>
99 #include <uvm/uvm_device.h>
100 
101 #ifdef UNION
102 #include <fs/union/union.h>
103 #endif
104 
105 #ifndef COMPAT_ZERODEV
106 #define COMPAT_ZERODEV(dev)	(0)
107 #endif
108 
109 int (*vn_union_readdir_hook)(struct vnode **, struct file *, struct lwp *);
110 
111 #include <sys/verified_exec.h>
112 
113 static int vn_read(file_t *fp, off_t *offset, struct uio *uio,
114     kauth_cred_t cred, int flags);
115 static int vn_write(file_t *fp, off_t *offset, struct uio *uio,
116     kauth_cred_t cred, int flags);
117 static int vn_closefile(file_t *fp);
118 static int vn_poll(file_t *fp, int events);
119 static int vn_fcntl(file_t *fp, u_int com, void *data);
120 static int vn_statfile(file_t *fp, struct stat *sb);
121 static int vn_ioctl(file_t *fp, u_long com, void *data);
122 static int vn_mmap(struct file *, off_t *, size_t, int, int *, int *,
123     struct uvm_object **, int *);
124 static int vn_seek(struct file *, off_t, int, off_t *, int);
125 
126 const struct fileops vnops = {
127 	.fo_name = "vn",
128 	.fo_read = vn_read,
129 	.fo_write = vn_write,
130 	.fo_ioctl = vn_ioctl,
131 	.fo_fcntl = vn_fcntl,
132 	.fo_poll = vn_poll,
133 	.fo_stat = vn_statfile,
134 	.fo_close = vn_closefile,
135 	.fo_kqfilter = vn_kqfilter,
136 	.fo_restart = fnullop_restart,
137 	.fo_mmap = vn_mmap,
138 	.fo_seek = vn_seek,
139 };
140 
141 /*
142  * Common code for vnode open operations.
143  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
144  *
145  * at_dvp is the directory for openat(), if any.
146  * pb is the path.
147  * nmode is additional namei flags, restricted to TRYEMULROOT and NOCHROOT.
148  * fmode is the open flags, converted from O_* to F*
149  * cmode is the creation file permissions.
150  *
151  * XXX shouldn't cmode be mode_t?
152  *
153  * On success produces either a locked vnode in *ret_vp, or NULL in
154  * *ret_vp and a file descriptor number in *ret_fd.
155  *
156  * The caller may pass NULL for ret_fd (and ret_domove), in which case
157  * EOPNOTSUPP will be produced in the cases that would otherwise return
158  * a file descriptor.
159  *
160  * Note that callers that want no-follow behavior should pass
161  * O_NOFOLLOW in fmode. Neither FOLLOW nor NOFOLLOW in nmode is
162  * honored.
163  */
164 int
165 vn_open(struct vnode *at_dvp, struct pathbuf *pb,
166 	int nmode, int fmode, int cmode,
167 	struct vnode **ret_vp, bool *ret_domove, int *ret_fd)
168 {
169 	struct nameidata nd;
170 	struct vnode *vp = NULL;
171 	struct lwp *l = curlwp;
172 	kauth_cred_t cred = l->l_cred;
173 	struct vattr va;
174 	int error;
175 	const char *pathstring;
176 
177 	KASSERT((nmode & (TRYEMULROOT | NOCHROOT)) == nmode);
178 
179 	KASSERT(ret_vp != NULL);
180 	KASSERT((ret_domove == NULL) == (ret_fd == NULL));
181 
182 	if ((fmode & (O_CREAT | O_DIRECTORY)) == (O_CREAT | O_DIRECTORY))
183 		return EINVAL;
184 
185 	NDINIT(&nd, LOOKUP, nmode, pb);
186 	if (at_dvp != NULL)
187 		NDAT(&nd, at_dvp);
188 
189 	nd.ni_cnd.cn_flags &= TRYEMULROOT | NOCHROOT;
190 
191 	if (fmode & O_CREAT) {
192 		nd.ni_cnd.cn_nameiop = CREATE;
193 		nd.ni_cnd.cn_flags |= LOCKPARENT | LOCKLEAF;
194 		if ((fmode & O_EXCL) == 0 &&
195 		    ((fmode & O_NOFOLLOW) == 0))
196 			nd.ni_cnd.cn_flags |= FOLLOW;
197 		if ((fmode & O_EXCL) == 0)
198 			nd.ni_cnd.cn_flags |= NONEXCLHACK;
199 	} else {
200 		nd.ni_cnd.cn_nameiop = LOOKUP;
201 		nd.ni_cnd.cn_flags |= LOCKLEAF;
202 		if ((fmode & O_NOFOLLOW) == 0)
203 			nd.ni_cnd.cn_flags |= FOLLOW;
204 	}
205 
206 	pathstring = pathbuf_stringcopy_get(nd.ni_pathbuf);
207 	if (pathstring == NULL) {
208 		return ENOMEM;
209 	}
210 
211 	/*
212 	 * When this "interface" was exposed to do_open() it used
213 	 * to initialize l_dupfd to -newfd-1 (thus passing in the
214 	 * new file handle number to use)... but nothing in the
215 	 * kernel uses that value. So just send 0.
216 	 */
217 	l->l_dupfd = 0;
218 
219 	error = namei(&nd);
220 	if (error)
221 		goto out;
222 
223 	vp = nd.ni_vp;
224 
225 #if NVERIEXEC > 0
226 	error = veriexec_openchk(l, nd.ni_vp, pathstring, fmode);
227 	if (error) {
228 		/* We have to release the locks ourselves */
229 		/*
230 		 * 20210604 dholland passing NONEXCLHACK means we can
231 		 * get ni_dvp == NULL back if ni_vp exists, and we should
232 		 * treat that like the non-O_CREAT case.
233 		 */
234 		if ((fmode & O_CREAT) != 0 && nd.ni_dvp != NULL) {
235 			if (vp == NULL) {
236 				vput(nd.ni_dvp);
237 			} else {
238 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
239 				if (nd.ni_dvp == nd.ni_vp)
240 					vrele(nd.ni_dvp);
241 				else
242 					vput(nd.ni_dvp);
243 				nd.ni_dvp = NULL;
244 				vput(vp);
245 				vp = NULL;
246 			}
247 		} else {
248 			vput(vp);
249 			vp = NULL;
250 		}
251 		goto out;
252 	}
253 #endif /* NVERIEXEC > 0 */
254 
255 	/*
256 	 * 20210604 dholland ditto
257 	 */
258 	if ((fmode & O_CREAT) != 0 && nd.ni_dvp != NULL) {
259 		if (nd.ni_vp == NULL) {
260 			vattr_null(&va);
261 			va.va_type = VREG;
262 			va.va_mode = cmode;
263 			if (fmode & O_EXCL)
264 				 va.va_vaflags |= VA_EXCLUSIVE;
265 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp,
266 					   &nd.ni_cnd, &va);
267 			if (error) {
268 				vput(nd.ni_dvp);
269 				goto out;
270 			}
271 			fmode &= ~O_TRUNC;
272 			vp = nd.ni_vp;
273 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
274 			vput(nd.ni_dvp);
275 		} else {
276 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
277 			if (nd.ni_dvp == nd.ni_vp)
278 				vrele(nd.ni_dvp);
279 			else
280 				vput(nd.ni_dvp);
281 			nd.ni_dvp = NULL;
282 			vp = nd.ni_vp;
283 			if (fmode & O_EXCL) {
284 				error = EEXIST;
285 				goto bad;
286 			}
287 			fmode &= ~O_CREAT;
288 		}
289 	} else if ((fmode & O_CREAT) != 0) {
290 		/*
291 		 * 20210606 dholland passing NONEXCLHACK means this
292 		 * case exists; it is the same as the following one
293 		 * but also needs to do things in the second (exists)
294 		 * half of the following block. (Besides handle
295 		 * ni_dvp, anyway.)
296 		 */
297 		vp = nd.ni_vp;
298 		KASSERT((fmode & O_EXCL) == 0);
299 		fmode &= ~O_CREAT;
300 	} else {
301 		vp = nd.ni_vp;
302 	}
303 	if (vp->v_type == VSOCK) {
304 		error = EOPNOTSUPP;
305 		goto bad;
306 	}
307 	if (nd.ni_vp->v_type == VLNK) {
308 		error = EFTYPE;
309 		goto bad;
310 	}
311 
312 	if ((fmode & O_CREAT) == 0) {
313 		error = vn_openchk(vp, cred, fmode);
314 		if (error != 0)
315 			goto bad;
316 	}
317 
318 	if (fmode & O_TRUNC) {
319 		vattr_null(&va);
320 		va.va_size = 0;
321 		error = VOP_SETATTR(vp, &va, cred);
322 		if (error != 0)
323 			goto bad;
324 	}
325 	if ((error = VOP_OPEN(vp, fmode, cred)) != 0)
326 		goto bad;
327 	if (fmode & FWRITE) {
328 		mutex_enter(vp->v_interlock);
329 		vp->v_writecount++;
330 		mutex_exit(vp->v_interlock);
331 	}
332 
333 bad:
334 	if (error) {
335 		vput(vp);
336 		vp = NULL;
337 	}
338 out:
339 	pathbuf_stringcopy_put(nd.ni_pathbuf, pathstring);
340 
341 	switch (error) {
342 	case EDUPFD:
343 	case EMOVEFD:
344 		/* if the caller isn't prepared to handle fds, fail for them */
345 		if (ret_fd == NULL) {
346 			error = EOPNOTSUPP;
347 			break;
348 		}
349 		*ret_vp = NULL;
350 		*ret_domove = error == EMOVEFD;
351 		*ret_fd = l->l_dupfd;
352 		error = 0;
353 		break;
354 	case 0:
355 		KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
356 		*ret_vp = vp;
357 		break;
358 	}
359 	l->l_dupfd = 0;
360 	return error;
361 }
362 
363 /*
364  * Check for write permissions on the specified vnode.
365  * Prototype text segments cannot be written.
366  */
367 int
368 vn_writechk(struct vnode *vp)
369 {
370 
371 	/*
372 	 * If the vnode is in use as a process's text,
373 	 * we can't allow writing.
374 	 */
375 	if (vp->v_iflag & VI_TEXT)
376 		return ETXTBSY;
377 	return 0;
378 }
379 
380 int
381 vn_openchk(struct vnode *vp, kauth_cred_t cred, int fflags)
382 {
383 	int permbits = 0;
384 	int error;
385 
386 	if (vp->v_type == VNON || vp->v_type == VBAD)
387 		return ENXIO;
388 
389 	if ((fflags & O_DIRECTORY) != 0 && vp->v_type != VDIR)
390 		return ENOTDIR;
391 
392 	if ((fflags & O_REGULAR) != 0 && vp->v_type != VREG)
393 		return EFTYPE;
394 
395 	if ((fflags & FREAD) != 0) {
396 		permbits = VREAD;
397 	}
398 	if ((fflags & FEXEC) != 0) {
399 		permbits |= VEXEC;
400 	}
401 	if ((fflags & (FWRITE | O_TRUNC)) != 0) {
402 		permbits |= VWRITE;
403 		if (vp->v_type == VDIR) {
404 			error = EISDIR;
405 			goto bad;
406 		}
407 		error = vn_writechk(vp);
408 		if (error != 0)
409 			goto bad;
410 	}
411 	error = VOP_ACCESS(vp, permbits, cred);
412 bad:
413 	return error;
414 }
415 
416 /*
417  * Mark a vnode as having executable mappings.
418  */
419 void
420 vn_markexec(struct vnode *vp)
421 {
422 
423 	if ((vp->v_iflag & VI_EXECMAP) != 0) {
424 		/* Safe unlocked, as long as caller holds a reference. */
425 		return;
426 	}
427 
428 	rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
429 	mutex_enter(vp->v_interlock);
430 	if ((vp->v_iflag & VI_EXECMAP) == 0) {
431 		cpu_count(CPU_COUNT_EXECPAGES, vp->v_uobj.uo_npages);
432 		vp->v_iflag |= VI_EXECMAP;
433 	}
434 	mutex_exit(vp->v_interlock);
435 	rw_exit(vp->v_uobj.vmobjlock);
436 }
437 
438 /*
439  * Mark a vnode as being the text of a process.
440  * Fail if the vnode is currently writable.
441  */
442 int
443 vn_marktext(struct vnode *vp)
444 {
445 
446 	if ((vp->v_iflag & (VI_TEXT|VI_EXECMAP)) == (VI_TEXT|VI_EXECMAP)) {
447 		/* Safe unlocked, as long as caller holds a reference. */
448 		return 0;
449 	}
450 
451 	rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
452 	mutex_enter(vp->v_interlock);
453 	if (vp->v_writecount != 0) {
454 		KASSERT((vp->v_iflag & VI_TEXT) == 0);
455 		mutex_exit(vp->v_interlock);
456 		rw_exit(vp->v_uobj.vmobjlock);
457 		return ETXTBSY;
458 	}
459 	if ((vp->v_iflag & VI_EXECMAP) == 0) {
460 		cpu_count(CPU_COUNT_EXECPAGES, vp->v_uobj.uo_npages);
461 	}
462 	vp->v_iflag |= (VI_TEXT | VI_EXECMAP);
463 	mutex_exit(vp->v_interlock);
464 	rw_exit(vp->v_uobj.vmobjlock);
465 	return 0;
466 }
467 
468 /*
469  * Vnode close call
470  *
471  * Note: takes an unlocked vnode, while VOP_CLOSE takes a locked node.
472  */
473 int
474 vn_close(struct vnode *vp, int flags, kauth_cred_t cred)
475 {
476 	int error;
477 
478 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
479 	if (flags & FWRITE) {
480 		mutex_enter(vp->v_interlock);
481 		KASSERT(vp->v_writecount > 0);
482 		vp->v_writecount--;
483 		mutex_exit(vp->v_interlock);
484 	}
485 	error = VOP_CLOSE(vp, flags, cred);
486 	vput(vp);
487 	return error;
488 }
489 
490 static int
491 enforce_rlimit_fsize(struct vnode *vp, struct uio *uio, int ioflag)
492 {
493 	struct lwp *l = curlwp;
494 	off_t testoff;
495 
496 	if (uio->uio_rw != UIO_WRITE || vp->v_type != VREG)
497 		return 0;
498 
499 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
500 	if (ioflag & IO_APPEND)
501 		testoff = vp->v_size;
502 	else
503 		testoff = uio->uio_offset;
504 
505 	if (testoff + uio->uio_resid >
506 	    l->l_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
507 		mutex_enter(&proc_lock);
508 		psignal(l->l_proc, SIGXFSZ);
509 		mutex_exit(&proc_lock);
510 		return EFBIG;
511 	}
512 
513 	return 0;
514 }
515 
516 /*
517  * Package up an I/O request on a vnode into a uio and do it.
518  */
519 int
520 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
521     enum uio_seg segflg, int ioflg, kauth_cred_t cred, size_t *aresid,
522     struct lwp *l)
523 {
524 	struct uio auio;
525 	struct iovec aiov;
526 	int error;
527 
528 	if ((ioflg & IO_NODELOCKED) == 0) {
529 		if (rw == UIO_READ) {
530 			vn_lock(vp, LK_SHARED | LK_RETRY);
531 		} else /* UIO_WRITE */ {
532 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
533 		}
534 	}
535 	auio.uio_iov = &aiov;
536 	auio.uio_iovcnt = 1;
537 	aiov.iov_base = base;
538 	aiov.iov_len = len;
539 	auio.uio_resid = len;
540 	auio.uio_offset = offset;
541 	auio.uio_rw = rw;
542 	if (segflg == UIO_SYSSPACE) {
543 		UIO_SETUP_SYSSPACE(&auio);
544 	} else {
545 		auio.uio_vmspace = l->l_proc->p_vmspace;
546 	}
547 
548 	if ((error = enforce_rlimit_fsize(vp, &auio, ioflg)) != 0)
549 		goto out;
550 
551 	if (rw == UIO_READ) {
552 		error = VOP_READ(vp, &auio, ioflg, cred);
553 	} else {
554 		error = VOP_WRITE(vp, &auio, ioflg, cred);
555 	}
556 
557 	if (aresid)
558 		*aresid = auio.uio_resid;
559 	else
560 		if (auio.uio_resid && error == 0)
561 			error = EIO;
562 
563  out:
564 	if ((ioflg & IO_NODELOCKED) == 0) {
565 		VOP_UNLOCK(vp);
566 	}
567 	return error;
568 }
569 
570 int
571 vn_readdir(file_t *fp, char *bf, int segflg, u_int count, int *done,
572     struct lwp *l, off_t **cookies, int *ncookies)
573 {
574 	struct vnode *vp = fp->f_vnode;
575 	struct iovec aiov;
576 	struct uio auio;
577 	int error, eofflag;
578 
579 	/* Limit the size on any kernel buffers used by VOP_READDIR */
580 	count = uimin(MAXBSIZE, count);
581 
582 unionread:
583 	if (vp->v_type != VDIR)
584 		return EINVAL;
585 	aiov.iov_base = bf;
586 	aiov.iov_len = count;
587 	auio.uio_iov = &aiov;
588 	auio.uio_iovcnt = 1;
589 	auio.uio_rw = UIO_READ;
590 	if (segflg == UIO_SYSSPACE) {
591 		UIO_SETUP_SYSSPACE(&auio);
592 	} else {
593 		KASSERT(l == curlwp);
594 		auio.uio_vmspace = l->l_proc->p_vmspace;
595 	}
596 	auio.uio_resid = count;
597 	vn_lock(vp, LK_SHARED | LK_RETRY);
598 	auio.uio_offset = fp->f_offset;
599 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, cookies,
600 		    ncookies);
601 	mutex_enter(&fp->f_lock);
602 	fp->f_offset = auio.uio_offset;
603 	mutex_exit(&fp->f_lock);
604 	VOP_UNLOCK(vp);
605 	if (error)
606 		return error;
607 
608 	if (count == auio.uio_resid && vn_union_readdir_hook) {
609 		struct vnode *ovp = vp;
610 
611 		error = (*vn_union_readdir_hook)(&vp, fp, l);
612 		if (error)
613 			return error;
614 		if (vp != ovp)
615 			goto unionread;
616 	}
617 
618 	if (count == auio.uio_resid && (vp->v_vflag & VV_ROOT) &&
619 	    (vp->v_mount->mnt_flag & MNT_UNION)) {
620 		struct vnode *tvp = vp;
621 		vp = vp->v_mount->mnt_vnodecovered;
622 		vref(vp);
623 		mutex_enter(&fp->f_lock);
624 		fp->f_vnode = vp;
625 		fp->f_offset = 0;
626 		mutex_exit(&fp->f_lock);
627 		vrele(tvp);
628 		goto unionread;
629 	}
630 	*done = count - auio.uio_resid;
631 	return error;
632 }
633 
634 /*
635  * File table vnode read routine.
636  */
637 static int
638 vn_read(file_t *fp, off_t *offset, struct uio *uio, kauth_cred_t cred,
639     int flags)
640 {
641 	struct vnode *vp = fp->f_vnode;
642 	int error, ioflag, fflag;
643 	size_t count;
644 
645 	ioflag = IO_ADV_ENCODE(fp->f_advice);
646 	fflag = fp->f_flag;
647 	if (fflag & FNONBLOCK)
648 		ioflag |= IO_NDELAY;
649 	if ((fflag & (FFSYNC | FRSYNC)) == (FFSYNC | FRSYNC))
650 		ioflag |= IO_SYNC;
651 	if (fflag & FALTIO)
652 		ioflag |= IO_ALTSEMANTICS;
653 	if (fflag & FDIRECT)
654 		ioflag |= IO_DIRECT;
655 	if (offset == &fp->f_offset && (flags & FOF_UPDATE_OFFSET) != 0)
656 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
657 	else
658 		vn_lock(vp, LK_SHARED | LK_RETRY);
659 	uio->uio_offset = *offset;
660 	count = uio->uio_resid;
661 	error = VOP_READ(vp, uio, ioflag, cred);
662 	if (flags & FOF_UPDATE_OFFSET)
663 		*offset += count - uio->uio_resid;
664 	VOP_UNLOCK(vp);
665 	return error;
666 }
667 
668 /*
669  * File table vnode write routine.
670  */
671 static int
672 vn_write(file_t *fp, off_t *offset, struct uio *uio, kauth_cred_t cred,
673     int flags)
674 {
675 	struct vnode *vp = fp->f_vnode;
676 	int error, ioflag, fflag;
677 	size_t count;
678 
679 	ioflag = IO_ADV_ENCODE(fp->f_advice) | IO_UNIT;
680 	fflag = fp->f_flag;
681 	if (vp->v_type == VREG && (fflag & O_APPEND))
682 		ioflag |= IO_APPEND;
683 	if (fflag & FNONBLOCK)
684 		ioflag |= IO_NDELAY;
685 	if (fflag & FFSYNC ||
686 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
687 		ioflag |= IO_SYNC;
688 	else if (fflag & FDSYNC)
689 		ioflag |= IO_DSYNC;
690 	if (fflag & FALTIO)
691 		ioflag |= IO_ALTSEMANTICS;
692 	if (fflag & FDIRECT)
693 		ioflag |= IO_DIRECT;
694 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
695 	uio->uio_offset = *offset;
696 	count = uio->uio_resid;
697 
698 	if ((error = enforce_rlimit_fsize(vp, uio, ioflag)) != 0)
699 		goto out;
700 
701 	error = VOP_WRITE(vp, uio, ioflag, cred);
702 
703 	if (flags & FOF_UPDATE_OFFSET) {
704 		if (ioflag & IO_APPEND) {
705 			/*
706 			 * SUSv3 describes behaviour for count = 0 as following:
707 			 * "Before any action ... is taken, and if nbyte is zero
708 			 * and the file is a regular file, the write() function
709 			 * ... in the absence of errors ... shall return zero
710 			 * and have no other results."
711 			 */
712 			if (count)
713 				*offset = uio->uio_offset;
714 		} else
715 			*offset += count - uio->uio_resid;
716 	}
717 
718  out:
719 	VOP_UNLOCK(vp);
720 	return error;
721 }
722 
723 /*
724  * File table vnode stat routine.
725  */
726 static int
727 vn_statfile(file_t *fp, struct stat *sb)
728 {
729 	struct vnode *vp = fp->f_vnode;
730 	int error;
731 
732 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
733 	error = vn_stat(vp, sb);
734 	VOP_UNLOCK(vp);
735 	return error;
736 }
737 
738 int
739 vn_stat(struct vnode *vp, struct stat *sb)
740 {
741 	struct vattr va;
742 	int error;
743 	mode_t mode;
744 
745 	memset(&va, 0, sizeof(va));
746 	error = VOP_GETATTR(vp, &va, kauth_cred_get());
747 	if (error)
748 		return error;
749 	/*
750 	 * Copy from vattr table
751 	 */
752 	memset(sb, 0, sizeof(*sb));
753 	sb->st_dev = va.va_fsid;
754 	sb->st_ino = va.va_fileid;
755 	mode = va.va_mode;
756 	switch (vp->v_type) {
757 	case VREG:
758 		mode |= S_IFREG;
759 		break;
760 	case VDIR:
761 		mode |= S_IFDIR;
762 		break;
763 	case VBLK:
764 		mode |= S_IFBLK;
765 		break;
766 	case VCHR:
767 		mode |= S_IFCHR;
768 		break;
769 	case VLNK:
770 		mode |= S_IFLNK;
771 		break;
772 	case VSOCK:
773 		mode |= S_IFSOCK;
774 		break;
775 	case VFIFO:
776 		mode |= S_IFIFO;
777 		break;
778 	default:
779 		return EBADF;
780 	}
781 	sb->st_mode = mode;
782 	sb->st_nlink = va.va_nlink;
783 	sb->st_uid = va.va_uid;
784 	sb->st_gid = va.va_gid;
785 	sb->st_rdev = va.va_rdev;
786 	sb->st_size = va.va_size;
787 	sb->st_atimespec = va.va_atime;
788 	sb->st_mtimespec = va.va_mtime;
789 	sb->st_ctimespec = va.va_ctime;
790 	sb->st_birthtimespec = va.va_birthtime;
791 	sb->st_blksize = va.va_blocksize;
792 	sb->st_flags = va.va_flags;
793 	sb->st_gen = 0;
794 	sb->st_blocks = va.va_bytes / S_BLKSIZE;
795 	return 0;
796 }
797 
798 /*
799  * File table vnode fcntl routine.
800  */
801 static int
802 vn_fcntl(file_t *fp, u_int com, void *data)
803 {
804 	struct vnode *vp = fp->f_vnode;
805 	int error;
806 
807 	error = VOP_FCNTL(vp, com, data, fp->f_flag, kauth_cred_get());
808 	return error;
809 }
810 
811 /*
812  * File table vnode ioctl routine.
813  */
814 static int
815 vn_ioctl(file_t *fp, u_long com, void *data)
816 {
817 	struct vnode *vp = fp->f_vnode, *ovp;
818 	struct vattr vattr;
819 	int error;
820 
821 	switch (vp->v_type) {
822 
823 	case VREG:
824 	case VDIR:
825 		if (com == FIONREAD) {
826 			vn_lock(vp, LK_SHARED | LK_RETRY);
827 			error = VOP_GETATTR(vp, &vattr, kauth_cred_get());
828 			if (error == 0)
829 				*(int *)data = vattr.va_size - fp->f_offset;
830 			VOP_UNLOCK(vp);
831 			if (error)
832 				return error;
833 			return 0;
834 		}
835 		if ((com == FIONWRITE) || (com == FIONSPACE)) {
836 			/*
837 			 * Files don't have send queues, so there never
838 			 * are any bytes in them, nor is there any
839 			 * open space in them.
840 			 */
841 			*(int *)data = 0;
842 			return 0;
843 		}
844 		if (com == FIOGETBMAP) {
845 			daddr_t *block;
846 
847 			if (*(daddr_t *)data < 0)
848 				return EINVAL;
849 			block = (daddr_t *)data;
850 			vn_lock(vp, LK_SHARED | LK_RETRY);
851 			error = VOP_BMAP(vp, *block, NULL, block, NULL);
852 			VOP_UNLOCK(vp);
853 			return error;
854 		}
855 		if (com == OFIOGETBMAP) {
856 			daddr_t ibn, obn;
857 
858 			if (*(int32_t *)data < 0)
859 				return EINVAL;
860 			ibn = (daddr_t)*(int32_t *)data;
861 			vn_lock(vp, LK_SHARED | LK_RETRY);
862 			error = VOP_BMAP(vp, ibn, NULL, &obn, NULL);
863 			VOP_UNLOCK(vp);
864 			*(int32_t *)data = (int32_t)obn;
865 			return error;
866 		}
867 		if (com == FIONBIO || com == FIOASYNC)	/* XXX */
868 			return 0;			/* XXX */
869 		/* FALLTHROUGH */
870 	case VFIFO:
871 	case VCHR:
872 	case VBLK:
873 		error = VOP_IOCTL(vp, com, data, fp->f_flag,
874 		    kauth_cred_get());
875 		if (error == 0 && com == TIOCSCTTY) {
876 			vref(vp);
877 			mutex_enter(&proc_lock);
878 			ovp = curproc->p_session->s_ttyvp;
879 			curproc->p_session->s_ttyvp = vp;
880 			mutex_exit(&proc_lock);
881 			if (ovp != NULL)
882 				vrele(ovp);
883 		}
884 		return error;
885 
886 	default:
887 		return EPASSTHROUGH;
888 	}
889 }
890 
891 /*
892  * File table vnode poll routine.
893  */
894 static int
895 vn_poll(file_t *fp, int events)
896 {
897 
898 	return VOP_POLL(fp->f_vnode, events);
899 }
900 
901 /*
902  * File table vnode kqfilter routine.
903  */
904 int
905 vn_kqfilter(file_t *fp, struct knote *kn)
906 {
907 
908 	return VOP_KQFILTER(fp->f_vnode, kn);
909 }
910 
911 static int
912 vn_mmap(struct file *fp, off_t *offp, size_t size, int prot, int *flagsp,
913     int *advicep, struct uvm_object **uobjp, int *maxprotp)
914 {
915 	struct uvm_object *uobj;
916 	struct vnode *vp;
917 	struct vattr va;
918 	struct lwp *l;
919 	vm_prot_t maxprot;
920 	off_t off;
921 	int error, flags;
922 	bool needwritemap;
923 
924 	l = curlwp;
925 
926 	off = *offp;
927 	flags = *flagsp;
928 	maxprot = VM_PROT_EXECUTE;
929 
930 	KASSERT(size > 0);
931 
932 	vp = fp->f_vnode;
933 	if (vp->v_type != VREG && vp->v_type != VCHR &&
934 	    vp->v_type != VBLK) {
935 		/* only REG/CHR/BLK support mmap */
936 		return ENODEV;
937 	}
938 	if (vp->v_type != VCHR && off < 0) {
939 		return EINVAL;
940 	}
941 #if SIZE_MAX > UINT32_MAX	/* XXX -Wtype-limits */
942 	if (vp->v_type != VCHR && size > __type_max(off_t)) {
943 		return EOVERFLOW;
944 	}
945 #endif
946 	if (vp->v_type != VCHR && off > __type_max(off_t) - size) {
947 		/* no offset wrapping */
948 		return EOVERFLOW;
949 	}
950 
951 	/* special case: catch SunOS style /dev/zero */
952 	if (vp->v_type == VCHR &&
953 	    (vp->v_rdev == zerodev || COMPAT_ZERODEV(vp->v_rdev))) {
954 		*uobjp = NULL;
955 		*maxprotp = VM_PROT_ALL;
956 		return 0;
957 	}
958 
959 	/*
960 	 * Old programs may not select a specific sharing type, so
961 	 * default to an appropriate one.
962 	 *
963 	 * XXX: how does MAP_ANON fit in the picture?
964 	 */
965 	if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
966 #if defined(DEBUG)
967 		struct proc *p = l->l_proc;
968 		printf("WARNING: defaulted mmap() share type to "
969 		       "%s (pid %d command %s)\n", vp->v_type == VCHR ?
970 		       "MAP_SHARED" : "MAP_PRIVATE", p->p_pid,
971 		       p->p_comm);
972 #endif
973 		if (vp->v_type == VCHR)
974 			flags |= MAP_SHARED;	/* for a device */
975 		else
976 			flags |= MAP_PRIVATE;	/* for a file */
977 	}
978 
979 	/*
980 	 * MAP_PRIVATE device mappings don't make sense (and aren't
981 	 * supported anyway).  However, some programs rely on this,
982 	 * so just change it to MAP_SHARED.
983 	 */
984 	if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
985 		flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
986 	}
987 
988 	/*
989 	 * now check protection
990 	 */
991 
992 	/* check read access */
993 	if (fp->f_flag & FREAD)
994 		maxprot |= VM_PROT_READ;
995 	else if (prot & PROT_READ) {
996 		return EACCES;
997 	}
998 
999 	/* check write access, shared case first */
1000 	if (flags & MAP_SHARED) {
1001 		/*
1002 		 * if the file is writable, only add PROT_WRITE to
1003 		 * maxprot if the file is not immutable, append-only.
1004 		 * otherwise, if we have asked for PROT_WRITE, return
1005 		 * EPERM.
1006 		 */
1007 		if (fp->f_flag & FWRITE) {
1008 			vn_lock(vp, LK_SHARED | LK_RETRY);
1009 			error = VOP_GETATTR(vp, &va, l->l_cred);
1010 			VOP_UNLOCK(vp);
1011 			if (error) {
1012 				return error;
1013 			}
1014 			if ((va.va_flags &
1015 			     (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0)
1016 				maxprot |= VM_PROT_WRITE;
1017 			else if (prot & PROT_WRITE) {
1018 				return EPERM;
1019 			}
1020 		} else if (prot & PROT_WRITE) {
1021 			return EACCES;
1022 		}
1023 	} else {
1024 		/* MAP_PRIVATE mappings can always write to */
1025 		maxprot |= VM_PROT_WRITE;
1026 	}
1027 
1028 	/*
1029 	 * Don't allow mmap for EXEC if the file system
1030 	 * is mounted NOEXEC.
1031 	 */
1032 	if ((prot & PROT_EXEC) != 0 &&
1033 	    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
1034 		return EACCES;
1035 	}
1036 
1037 	if (vp->v_type != VCHR) {
1038 		error = VOP_MMAP(vp, prot, curlwp->l_cred);
1039 		if (error) {
1040 			return error;
1041 		}
1042 		vref(vp);
1043 		uobj = &vp->v_uobj;
1044 
1045 		/*
1046 		 * If the vnode is being mapped with PROT_EXEC,
1047 		 * then mark it as text.
1048 		 */
1049 		if (prot & PROT_EXEC) {
1050 			vn_markexec(vp);
1051 		}
1052 	} else {
1053 		int i = maxprot;
1054 
1055 		/*
1056 		 * XXX Some devices don't like to be mapped with
1057 		 * XXX PROT_EXEC or PROT_WRITE, but we don't really
1058 		 * XXX have a better way of handling this, right now
1059 		 */
1060 		do {
1061 			uobj = udv_attach(vp->v_rdev,
1062 					  (flags & MAP_SHARED) ? i :
1063 					  (i & ~VM_PROT_WRITE), off, size);
1064 			i--;
1065 		} while ((uobj == NULL) && (i > 0));
1066 		if (uobj == NULL) {
1067 			return EINVAL;
1068 		}
1069 		*advicep = UVM_ADV_RANDOM;
1070 	}
1071 
1072 	/*
1073 	 * Set vnode flags to indicate the new kinds of mapping.
1074 	 * We take the vnode lock in exclusive mode here to serialize
1075 	 * with direct I/O.
1076 	 *
1077 	 * Safe to check for these flag values without a lock, as
1078 	 * long as a reference to the vnode is held.
1079 	 */
1080 	needwritemap = (vp->v_iflag & VI_WRMAP) == 0 &&
1081 		(flags & MAP_SHARED) != 0 &&
1082 		(maxprot & VM_PROT_WRITE) != 0;
1083 	if ((vp->v_vflag & VV_MAPPED) == 0 || needwritemap) {
1084 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1085 		vp->v_vflag |= VV_MAPPED;
1086 		if (needwritemap) {
1087 			rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
1088 			mutex_enter(vp->v_interlock);
1089 			vp->v_iflag |= VI_WRMAP;
1090 			mutex_exit(vp->v_interlock);
1091 			rw_exit(vp->v_uobj.vmobjlock);
1092 		}
1093 		VOP_UNLOCK(vp);
1094 	}
1095 
1096 #if NVERIEXEC > 0
1097 
1098 	/*
1099 	 * Check if the file can be executed indirectly.
1100 	 *
1101 	 * XXX: This gives false warnings about "Incorrect access type"
1102 	 * XXX: if the mapping is not executable. Harmless, but will be
1103 	 * XXX: fixed as part of other changes.
1104 	 */
1105 	if (veriexec_verify(l, vp, "(mmap)", VERIEXEC_INDIRECT,
1106 			    NULL)) {
1107 
1108 		/*
1109 		 * Don't allow executable mappings if we can't
1110 		 * indirectly execute the file.
1111 		 */
1112 		if (prot & VM_PROT_EXECUTE) {
1113 			return EPERM;
1114 		}
1115 
1116 		/*
1117 		 * Strip the executable bit from 'maxprot' to make sure
1118 		 * it can't be made executable later.
1119 		 */
1120 		maxprot &= ~VM_PROT_EXECUTE;
1121 	}
1122 #endif /* NVERIEXEC > 0 */
1123 
1124 	*uobjp = uobj;
1125 	*maxprotp = maxprot;
1126 	*flagsp = flags;
1127 
1128 	return 0;
1129 }
1130 
1131 static int
1132 vn_seek(struct file *fp, off_t delta, int whence, off_t *newoffp,
1133     int flags)
1134 {
1135 	const off_t OFF_MIN = __type_min(off_t);
1136 	const off_t OFF_MAX = __type_max(off_t);
1137 	kauth_cred_t cred = fp->f_cred;
1138 	off_t oldoff, newoff;
1139 	struct vnode *vp = fp->f_vnode;
1140 	struct vattr vattr;
1141 	int error;
1142 
1143 	if (vp->v_type == VFIFO)
1144 		return ESPIPE;
1145 
1146 	if (flags & FOF_UPDATE_OFFSET)
1147 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1148 	else
1149 		vn_lock(vp, LK_SHARED | LK_RETRY);
1150 
1151 	/* Compute the old and new offsets.  */
1152 	oldoff = fp->f_offset;
1153 	switch (whence) {
1154 	case SEEK_CUR:
1155 		if (delta > 0) {
1156 			if (oldoff > 0 && delta > OFF_MAX - oldoff) {
1157 				newoff = OFF_MAX;
1158 				break;
1159 			}
1160 		} else {
1161 			if (oldoff < 0 && delta < OFF_MIN - oldoff) {
1162 				newoff = OFF_MIN;
1163 				break;
1164 			}
1165 		}
1166 		newoff = oldoff + delta;
1167 		break;
1168 	case SEEK_END:
1169 		error = VOP_GETATTR(vp, &vattr, cred);
1170 		if (error)
1171 			goto out;
1172 		if (vattr.va_size > OFF_MAX ||
1173 		    delta > OFF_MAX - (off_t)vattr.va_size) {
1174 			newoff = OFF_MAX;
1175 			break;
1176 		}
1177 		newoff = delta + vattr.va_size;
1178 		break;
1179 	case SEEK_SET:
1180 		newoff = delta;
1181 		break;
1182 	default:
1183 		error = EINVAL;
1184 		goto out;
1185 	}
1186 
1187 	/* Pass the proposed change to the file system to audit.  */
1188 	error = VOP_SEEK(vp, oldoff, newoff, cred);
1189 	if (error)
1190 		goto out;
1191 
1192 	/* Success!  */
1193 	if (newoffp)
1194 		*newoffp = newoff;
1195 	if (flags & FOF_UPDATE_OFFSET)
1196 		fp->f_offset = newoff;
1197 	error = 0;
1198 
1199 out:	VOP_UNLOCK(vp);
1200 	return error;
1201 }
1202 
1203 /*
1204  * Check that the vnode is still valid, and if so
1205  * acquire requested lock.
1206  */
1207 int
1208 vn_lock(struct vnode *vp, int flags)
1209 {
1210 	struct lwp *l;
1211 	int error;
1212 
1213 	KASSERT(vrefcnt(vp) > 0);
1214 	KASSERT((flags & ~(LK_SHARED|LK_EXCLUSIVE|LK_NOWAIT|LK_RETRY|
1215 	    LK_UPGRADE|LK_DOWNGRADE)) == 0);
1216 	KASSERT((flags & LK_NOWAIT) != 0 || !mutex_owned(vp->v_interlock));
1217 
1218 #ifdef DIAGNOSTIC
1219 	if (wapbl_vphaswapbl(vp))
1220 		WAPBL_JUNLOCK_ASSERT(wapbl_vptomp(vp));
1221 #endif
1222 
1223 	/* Get a more useful report for lockstat. */
1224 	l = curlwp;
1225 	KASSERT(l->l_rwcallsite == 0);
1226 	l->l_rwcallsite = (uintptr_t)__builtin_return_address(0);
1227 
1228 	error = VOP_LOCK(vp, flags);
1229 
1230 	l->l_rwcallsite = 0;
1231 
1232 	switch (flags & (LK_RETRY | LK_NOWAIT)) {
1233 	case 0:
1234 		KASSERT(error == 0 || error == ENOENT);
1235 		break;
1236 	case LK_RETRY:
1237 		KASSERT(error == 0);
1238 		break;
1239 	case LK_NOWAIT:
1240 		KASSERT(error == 0 || error == EBUSY || error == ENOENT);
1241 		break;
1242 	case LK_RETRY | LK_NOWAIT:
1243 		KASSERT(error == 0 || error == EBUSY);
1244 		break;
1245 	}
1246 
1247 	return error;
1248 }
1249 
1250 /*
1251  * File table vnode close routine.
1252  */
1253 static int
1254 vn_closefile(file_t *fp)
1255 {
1256 
1257 	return vn_close(fp->f_vnode, fp->f_flag, fp->f_cred);
1258 }
1259 
1260 /*
1261  * Simplified in-kernel wrapper calls for extended attribute access.
1262  * Both calls pass in a NULL credential, authorizing a "kernel" access.
1263  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1264  */
1265 int
1266 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1267     const char *attrname, size_t *buflen, void *bf, struct lwp *l)
1268 {
1269 	struct uio auio;
1270 	struct iovec aiov;
1271 	int error;
1272 
1273 	aiov.iov_len = *buflen;
1274 	aiov.iov_base = bf;
1275 
1276 	auio.uio_iov = &aiov;
1277 	auio.uio_iovcnt = 1;
1278 	auio.uio_rw = UIO_READ;
1279 	auio.uio_offset = 0;
1280 	auio.uio_resid = *buflen;
1281 	UIO_SETUP_SYSSPACE(&auio);
1282 
1283 	if ((ioflg & IO_NODELOCKED) == 0)
1284 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1285 
1286 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL,
1287 	    NOCRED);
1288 
1289 	if ((ioflg & IO_NODELOCKED) == 0)
1290 		VOP_UNLOCK(vp);
1291 
1292 	if (error == 0)
1293 		*buflen = *buflen - auio.uio_resid;
1294 
1295 	return error;
1296 }
1297 
1298 /*
1299  * XXX Failure mode if partially written?
1300  */
1301 int
1302 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1303     const char *attrname, size_t buflen, const void *bf, struct lwp *l)
1304 {
1305 	struct uio auio;
1306 	struct iovec aiov;
1307 	int error;
1308 
1309 	aiov.iov_len = buflen;
1310 	aiov.iov_base = __UNCONST(bf);		/* XXXUNCONST kills const */
1311 
1312 	auio.uio_iov = &aiov;
1313 	auio.uio_iovcnt = 1;
1314 	auio.uio_rw = UIO_WRITE;
1315 	auio.uio_offset = 0;
1316 	auio.uio_resid = buflen;
1317 	UIO_SETUP_SYSSPACE(&auio);
1318 
1319 	if ((ioflg & IO_NODELOCKED) == 0) {
1320 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1321 	}
1322 
1323 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NOCRED);
1324 
1325 	if ((ioflg & IO_NODELOCKED) == 0) {
1326 		VOP_UNLOCK(vp);
1327 	}
1328 
1329 	return error;
1330 }
1331 
1332 int
1333 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1334     const char *attrname, struct lwp *l)
1335 {
1336 	int error;
1337 
1338 	if ((ioflg & IO_NODELOCKED) == 0) {
1339 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1340 	}
1341 
1342 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NOCRED);
1343 	if (error == EOPNOTSUPP)
1344 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1345 		    NOCRED);
1346 
1347 	if ((ioflg & IO_NODELOCKED) == 0) {
1348 		VOP_UNLOCK(vp);
1349 	}
1350 
1351 	return error;
1352 }
1353 
1354 int
1355 vn_fifo_bypass(void *v)
1356 {
1357 	struct vop_generic_args *ap = v;
1358 
1359 	return VOCALL(fifo_vnodeop_p, ap->a_desc->vdesc_offset, v);
1360 }
1361 
1362 /*
1363  * Open block device by device number
1364  */
1365 int
1366 vn_bdev_open(dev_t dev, struct vnode **vpp, struct lwp *l)
1367 {
1368 	int     error;
1369 
1370 	if ((error = bdevvp(dev, vpp)) != 0)
1371 		return error;
1372 
1373 	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
1374 	if ((error = VOP_OPEN(*vpp, FREAD | FWRITE, l->l_cred)) != 0) {
1375 		vput(*vpp);
1376 		return error;
1377 	}
1378 	mutex_enter((*vpp)->v_interlock);
1379 	(*vpp)->v_writecount++;
1380 	mutex_exit((*vpp)->v_interlock);
1381 	VOP_UNLOCK(*vpp);
1382 
1383 	return 0;
1384 }
1385 
1386 /*
1387  * Lookup the provided name in the filesystem.  If the file exists,
1388  * is a valid block device, and isn't being used by anyone else,
1389  * set *vpp to the file's vnode.
1390  */
1391 int
1392 vn_bdev_openpath(struct pathbuf *pb, struct vnode **vpp, struct lwp *l)
1393 {
1394 	struct vnode *vp;
1395 	dev_t dev;
1396 	enum vtype vt;
1397 	int     error;
1398 
1399 	error = vn_open(NULL, pb, 0, FREAD | FWRITE, 0, &vp, NULL, NULL);
1400 	if (error != 0)
1401 		return error;
1402 
1403 	dev = vp->v_rdev;
1404 	vt = vp->v_type;
1405 
1406 	VOP_UNLOCK(vp);
1407 	(void) vn_close(vp, FREAD | FWRITE, l->l_cred);
1408 
1409 	if (vt != VBLK)
1410 		return ENOTBLK;
1411 
1412 	return vn_bdev_open(dev, vpp, l);
1413 }
1414 
1415 static long
1416 vn_knote_to_interest(const struct knote *kn)
1417 {
1418 	switch (kn->kn_filter) {
1419 	case EVFILT_READ:
1420 		/*
1421 		 * Writing to the file or changing its attributes can
1422 		 * set the file size, which impacts the readability
1423 		 * filter.
1424 		 *
1425 		 * (No need to set NOTE_EXTEND here; it's only ever
1426 		 * send with other hints; see vnode_if.c.)
1427 		 */
1428 		return NOTE_WRITE | NOTE_ATTRIB;
1429 
1430 	case EVFILT_VNODE:
1431 		return kn->kn_sfflags;
1432 
1433 	case EVFILT_WRITE:
1434 	default:
1435 		return 0;
1436 	}
1437 }
1438 
1439 void
1440 vn_knote_attach(struct vnode *vp, struct knote *kn)
1441 {
1442 	struct vnode_klist *vk = vp->v_klist;
1443 	long interest = 0;
1444 
1445 	/*
1446 	 * In the case of layered / stacked file systems, knotes
1447 	 * should only ever be associated with the base vnode.
1448 	 */
1449 	KASSERT(kn->kn_hook == vp);
1450 	KASSERT(vp->v_klist == &VNODE_TO_VIMPL(vp)->vi_klist);
1451 
1452 	/*
1453 	 * We maintain a bitmask of the kevents that there is interest in,
1454 	 * to minimize the impact of having watchers.  It's silly to have
1455 	 * to traverse vn_klist every time a read or write happens simply
1456 	 * because there is someone interested in knowing when the file
1457 	 * is deleted, for example.
1458 	 */
1459 
1460 	mutex_enter(vp->v_interlock);
1461 	SLIST_INSERT_HEAD(&vk->vk_klist, kn, kn_selnext);
1462 	SLIST_FOREACH(kn, &vk->vk_klist, kn_selnext) {
1463 		interest |= vn_knote_to_interest(kn);
1464 	}
1465 	vk->vk_interest = interest;
1466 	mutex_exit(vp->v_interlock);
1467 }
1468 
1469 void
1470 vn_knote_detach(struct vnode *vp, struct knote *kn)
1471 {
1472 	struct vnode_klist *vk = vp->v_klist;
1473 	long interest = 0;
1474 
1475 	/* See above. */
1476 	KASSERT(kn->kn_hook == vp);
1477 	KASSERT(vp->v_klist == &VNODE_TO_VIMPL(vp)->vi_klist);
1478 
1479 	/*
1480 	 * We special case removing the head of the list, because:
1481 	 *
1482 	 * 1. It's extremely likely that we're detaching the only
1483 	 *    knote.
1484 	 *
1485 	 * 2. We're already traversing the whole list, so we don't
1486 	 *    want to use the generic SLIST_REMOVE() which would
1487 	 *    traverse it *again*.
1488 	 */
1489 
1490 	mutex_enter(vp->v_interlock);
1491 	if (__predict_true(kn == SLIST_FIRST(&vk->vk_klist))) {
1492 		SLIST_REMOVE_HEAD(&vk->vk_klist, kn_selnext);
1493 		SLIST_FOREACH(kn, &vk->vk_klist, kn_selnext) {
1494 			interest |= vn_knote_to_interest(kn);
1495 		}
1496 		vk->vk_interest = interest;
1497 	} else {
1498 		struct knote *thiskn, *nextkn, *prevkn = NULL;
1499 
1500 		SLIST_FOREACH_SAFE(thiskn, &vk->vk_klist, kn_selnext, nextkn) {
1501 			if (thiskn == kn) {
1502 				KASSERT(kn != NULL);
1503 				KASSERT(prevkn != NULL);
1504 				SLIST_REMOVE_AFTER(prevkn, kn_selnext);
1505 				kn = NULL;
1506 			} else {
1507 				interest |= vn_knote_to_interest(thiskn);
1508 				prevkn = thiskn;
1509 			}
1510 		}
1511 		vk->vk_interest = interest;
1512 	}
1513 	mutex_exit(vp->v_interlock);
1514 }
1515