xref: /netbsd-src/sys/kern/vfs_vnops.c (revision e94a5d02693120d4ad9d909e488894e9fcf0eb76)
1 /*	$NetBSD: vfs_vnops.c,v 1.244 2024/12/07 02:27:38 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 1.244 2024/12/07 02:27:38 riastradh Exp $");
70 
71 #include "veriexec.h"
72 
73 #include <sys/param.h>
74 #include <sys/types.h>
75 
76 #include <sys/atomic.h>
77 #include <sys/buf.h>
78 #include <sys/file.h>
79 #include <sys/filedesc.h>
80 #include <sys/fstrans.h>
81 #include <sys/ioctl.h>
82 #include <sys/kauth.h>
83 #include <sys/kernel.h>
84 #include <sys/mman.h>
85 #include <sys/mount.h>
86 #include <sys/namei.h>
87 #include <sys/poll.h>
88 #include <sys/proc.h>
89 #include <sys/sdt.h>
90 #include <sys/stat.h>
91 #include <sys/syslog.h>
92 #include <sys/systm.h>
93 #include <sys/tty.h>
94 #include <sys/verified_exec.h>
95 #include <sys/vnode_impl.h>
96 #include <sys/wapbl.h>
97 
98 #include <miscfs/fifofs/fifo.h>
99 #include <miscfs/specfs/specdev.h>
100 
101 #include <uvm/uvm_device.h>
102 #include <uvm/uvm_extern.h>
103 #include <uvm/uvm_readahead.h>
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 static int vn_read(file_t *fp, off_t *offset, struct uio *uio,
112     kauth_cred_t cred, int flags);
113 static int vn_write(file_t *fp, off_t *offset, struct uio *uio,
114     kauth_cred_t cred, int flags);
115 static int vn_closefile(file_t *fp);
116 static int vn_poll(file_t *fp, int events);
117 static int vn_fcntl(file_t *fp, u_int com, void *data);
118 static int vn_statfile(file_t *fp, struct stat *sb);
119 static int vn_ioctl(file_t *fp, u_long com, void *data);
120 static int vn_mmap(struct file *, off_t *, size_t, int, int *, int *,
121     struct uvm_object **, int *);
122 static int vn_seek(struct file *, off_t, int, off_t *, int);
123 static int vn_advlock(struct file *, void *, int, struct flock *, int);
124 static int vn_fpathconf(struct file *, int, register_t *);
125 static int vn_posix_fadvise(struct file *, off_t, off_t, int);
126 static int vn_truncate(file_t *, off_t);
127 
128 const struct fileops vnops = {
129 	.fo_name = "vn",
130 	.fo_read = vn_read,
131 	.fo_write = vn_write,
132 	.fo_ioctl = vn_ioctl,
133 	.fo_fcntl = vn_fcntl,
134 	.fo_poll = vn_poll,
135 	.fo_stat = vn_statfile,
136 	.fo_close = vn_closefile,
137 	.fo_kqfilter = vn_kqfilter,
138 	.fo_restart = fnullop_restart,
139 	.fo_mmap = vn_mmap,
140 	.fo_seek = vn_seek,
141 	.fo_advlock = vn_advlock,
142 	.fo_fpathconf = vn_fpathconf,
143 	.fo_posix_fadvise = vn_posix_fadvise,
144 	.fo_truncate = vn_truncate,
145 };
146 
147 /*
148  * Common code for vnode open operations.
149  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
150  *
151  * at_dvp is the directory for openat(), if any.
152  * pb is the path.
153  * nmode is additional namei flags, restricted to TRYEMULROOT and NOCHROOT.
154  * fmode is the open flags, converted from O_* to F*
155  * cmode is the creation file permissions.
156  *
157  * XXX shouldn't cmode be mode_t?
158  *
159  * On success produces either a locked vnode in *ret_vp, or NULL in
160  * *ret_vp and a file descriptor number in *ret_fd.
161  *
162  * The caller may pass NULL for ret_fd (and ret_domove), in which case
163  * EOPNOTSUPP will be produced in the cases that would otherwise return
164  * a file descriptor.
165  *
166  * Note that callers that want no-follow behavior should pass
167  * O_NOFOLLOW in fmode. Neither FOLLOW nor NOFOLLOW in nmode is
168  * honored.
169  */
170 int
171 vn_open(struct vnode *at_dvp, struct pathbuf *pb,
172     int nmode, int fmode, int cmode,
173     struct vnode **ret_vp, bool *ret_domove, int *ret_fd)
174 {
175 	struct nameidata nd;
176 	struct vnode *vp = NULL;
177 	struct lwp *l = curlwp;
178 	kauth_cred_t cred = l->l_cred;
179 	struct vattr va;
180 	int error;
181 	const char *pathstring;
182 
183 	KASSERT((nmode & (TRYEMULROOT | NOCHROOT)) == nmode);
184 
185 	KASSERT(ret_vp != NULL);
186 	KASSERT((ret_domove == NULL) == (ret_fd == NULL));
187 
188 	if ((fmode & (O_CREAT | O_DIRECTORY)) == (O_CREAT | O_DIRECTORY))
189 		return SET_ERROR(EINVAL);
190 
191 	NDINIT(&nd, LOOKUP, nmode, pb);
192 	if (at_dvp != NULL)
193 		NDAT(&nd, at_dvp);
194 
195 	nd.ni_cnd.cn_flags &= TRYEMULROOT | NOCHROOT;
196 
197 	if (fmode & O_CREAT) {
198 		nd.ni_cnd.cn_nameiop = CREATE;
199 		nd.ni_cnd.cn_flags |= LOCKPARENT | LOCKLEAF;
200 		if ((fmode & O_EXCL) == 0 &&
201 		    ((fmode & O_NOFOLLOW) == 0))
202 			nd.ni_cnd.cn_flags |= FOLLOW;
203 		if ((fmode & O_EXCL) == 0)
204 			nd.ni_cnd.cn_flags |= NONEXCLHACK;
205 	} else {
206 		nd.ni_cnd.cn_nameiop = LOOKUP;
207 		nd.ni_cnd.cn_flags |= LOCKLEAF;
208 		if ((fmode & O_NOFOLLOW) == 0)
209 			nd.ni_cnd.cn_flags |= FOLLOW;
210 	}
211 
212 	pathstring = pathbuf_stringcopy_get(nd.ni_pathbuf);
213 	if (pathstring == NULL) {
214 		return SET_ERROR(ENOMEM);
215 	}
216 
217 	/*
218 	 * When this "interface" was exposed to do_open() it used
219 	 * to initialize l_dupfd to -newfd-1 (thus passing in the
220 	 * new file handle number to use)... but nothing in the
221 	 * kernel uses that value. So just send 0.
222 	 */
223 	l->l_dupfd = 0;
224 
225 	error = namei(&nd);
226 	if (error)
227 		goto out;
228 
229 	vp = nd.ni_vp;
230 
231 #if NVERIEXEC > 0
232 	error = veriexec_openchk(l, nd.ni_vp, pathstring, fmode);
233 	if (error) {
234 		/* We have to release the locks ourselves */
235 		/*
236 		 * 20210604 dholland passing NONEXCLHACK means we can
237 		 * get ni_dvp == NULL back if ni_vp exists, and we should
238 		 * treat that like the non-O_CREAT case.
239 		 */
240 		if ((fmode & O_CREAT) != 0 && nd.ni_dvp != NULL) {
241 			if (vp == NULL) {
242 				vput(nd.ni_dvp);
243 			} else {
244 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
245 				if (nd.ni_dvp == nd.ni_vp)
246 					vrele(nd.ni_dvp);
247 				else
248 					vput(nd.ni_dvp);
249 				nd.ni_dvp = NULL;
250 				vput(vp);
251 				vp = NULL;
252 			}
253 		} else {
254 			vput(vp);
255 			vp = NULL;
256 		}
257 		goto out;
258 	}
259 #endif /* NVERIEXEC > 0 */
260 
261 	/*
262 	 * 20210604 dholland ditto
263 	 */
264 	if ((fmode & O_CREAT) != 0 && nd.ni_dvp != NULL) {
265 		if (nd.ni_vp == NULL) {
266 			vattr_null(&va);
267 			va.va_type = VREG;
268 			va.va_mode = cmode;
269 			if (fmode & O_EXCL)
270 				va.va_vaflags |= VA_EXCLUSIVE;
271 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp,
272 			    &nd.ni_cnd, &va);
273 			if (error) {
274 				vput(nd.ni_dvp);
275 				goto out;
276 			}
277 			fmode &= ~O_TRUNC;
278 			vp = nd.ni_vp;
279 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
280 			vput(nd.ni_dvp);
281 		} else {
282 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
283 			if (nd.ni_dvp == nd.ni_vp)
284 				vrele(nd.ni_dvp);
285 			else
286 				vput(nd.ni_dvp);
287 			nd.ni_dvp = NULL;
288 			vp = nd.ni_vp;
289 			if (fmode & O_EXCL) {
290 				error = SET_ERROR(EEXIST);
291 				goto bad;
292 			}
293 			fmode &= ~O_CREAT;
294 		}
295 	} else if ((fmode & O_CREAT) != 0) {
296 		/*
297 		 * 20210606 dholland passing NONEXCLHACK means this
298 		 * case exists; it is the same as the following one
299 		 * but also needs to do things in the second (exists)
300 		 * half of the following block. (Besides handle
301 		 * ni_dvp, anyway.)
302 		 */
303 		vp = nd.ni_vp;
304 		KASSERT((fmode & O_EXCL) == 0);
305 		fmode &= ~O_CREAT;
306 	} else {
307 		vp = nd.ni_vp;
308 	}
309 	if (vp->v_type == VSOCK) {
310 		error = SET_ERROR(EOPNOTSUPP);
311 		goto bad;
312 	}
313 	if (nd.ni_vp->v_type == VLNK) {
314 		error = SET_ERROR(EFTYPE);
315 		goto bad;
316 	}
317 
318 	if ((fmode & O_CREAT) == 0) {
319 		error = vn_openchk(vp, cred, fmode);
320 		if (error != 0)
321 			goto bad;
322 	}
323 
324 	if (fmode & O_TRUNC) {
325 		vattr_null(&va);
326 		va.va_size = 0;
327 		error = VOP_SETATTR(vp, &va, cred);
328 		if (error != 0)
329 			goto bad;
330 	}
331 	if ((error = VOP_OPEN(vp, fmode, cred)) != 0)
332 		goto bad;
333 	if (fmode & FWRITE) {
334 		mutex_enter(vp->v_interlock);
335 		vp->v_writecount++;
336 		mutex_exit(vp->v_interlock);
337 	}
338 
339 bad:
340 	if (error) {
341 		vput(vp);
342 		vp = NULL;
343 	}
344 out:
345 	pathbuf_stringcopy_put(nd.ni_pathbuf, pathstring);
346 
347 	switch (error) {
348 	case EDUPFD:
349 	case EMOVEFD:
350 		/* if the caller isn't prepared to handle fds, fail for them */
351 		if (ret_fd == NULL) {
352 			error = SET_ERROR(EOPNOTSUPP);
353 			break;
354 		}
355 		*ret_vp = NULL;
356 		*ret_domove = error == EMOVEFD;
357 		*ret_fd = l->l_dupfd;
358 		error = 0;
359 		break;
360 	case 0:
361 		KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
362 		*ret_vp = vp;
363 		break;
364 	}
365 	l->l_dupfd = 0;
366 	return error;
367 }
368 
369 /*
370  * Check for write permissions on the specified vnode.
371  * Prototype text segments cannot be written.
372  */
373 int
374 vn_writechk(struct vnode *vp)
375 {
376 
377 	/*
378 	 * If the vnode is in use as a process's text,
379 	 * we can't allow writing.
380 	 */
381 	if (vp->v_iflag & VI_TEXT)
382 		return SET_ERROR(ETXTBSY);
383 	return 0;
384 }
385 
386 int
387 vn_openchk(struct vnode *vp, kauth_cred_t cred, int fflags)
388 {
389 	int permbits = 0;
390 	int error;
391 
392 	if (vp->v_type == VNON || vp->v_type == VBAD)
393 		return SET_ERROR(ENXIO);
394 
395 	if ((fflags & O_DIRECTORY) != 0 && vp->v_type != VDIR)
396 		return SET_ERROR(ENOTDIR);
397 
398 	if ((fflags & O_REGULAR) != 0 && vp->v_type != VREG)
399 		return SET_ERROR(EFTYPE);
400 
401 	if ((fflags & FREAD) != 0) {
402 		permbits = VREAD;
403 	}
404 	if ((fflags & FEXEC) != 0) {
405 		permbits |= VEXEC;
406 	}
407 	if ((fflags & (FWRITE | O_TRUNC)) != 0) {
408 		permbits |= VWRITE;
409 		if (vp->v_type == VDIR) {
410 			error = SET_ERROR(EISDIR);
411 			goto bad;
412 		}
413 		error = vn_writechk(vp);
414 		if (error != 0)
415 			goto bad;
416 	}
417 	error = VOP_ACCESS(vp, permbits, cred);
418 bad:
419 	return error;
420 }
421 
422 /*
423  * Mark a vnode as having executable mappings.
424  */
425 void
426 vn_markexec(struct vnode *vp)
427 {
428 
429 	if ((vp->v_iflag & VI_EXECMAP) != 0) {
430 		/* Safe unlocked, as long as caller holds a reference. */
431 		return;
432 	}
433 
434 	rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
435 	mutex_enter(vp->v_interlock);
436 	if ((vp->v_iflag & VI_EXECMAP) == 0) {
437 		cpu_count(CPU_COUNT_EXECPAGES, vp->v_uobj.uo_npages);
438 		vp->v_iflag |= VI_EXECMAP;
439 	}
440 	mutex_exit(vp->v_interlock);
441 	rw_exit(vp->v_uobj.vmobjlock);
442 }
443 
444 /*
445  * Mark a vnode as being the text of a process.
446  * Fail if the vnode is currently writable.
447  */
448 int
449 vn_marktext(struct vnode *vp)
450 {
451 
452 	if ((vp->v_iflag & (VI_TEXT|VI_EXECMAP)) == (VI_TEXT|VI_EXECMAP)) {
453 		/* Safe unlocked, as long as caller holds a reference. */
454 		return 0;
455 	}
456 
457 	rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
458 	mutex_enter(vp->v_interlock);
459 	if (vp->v_writecount != 0) {
460 		KASSERT((vp->v_iflag & VI_TEXT) == 0);
461 		mutex_exit(vp->v_interlock);
462 		rw_exit(vp->v_uobj.vmobjlock);
463 		return SET_ERROR(ETXTBSY);
464 	}
465 	if ((vp->v_iflag & VI_EXECMAP) == 0) {
466 		cpu_count(CPU_COUNT_EXECPAGES, vp->v_uobj.uo_npages);
467 	}
468 	vp->v_iflag |= (VI_TEXT | VI_EXECMAP);
469 	mutex_exit(vp->v_interlock);
470 	rw_exit(vp->v_uobj.vmobjlock);
471 	return 0;
472 }
473 
474 /*
475  * Vnode close call
476  *
477  * Note: takes an unlocked vnode, while VOP_CLOSE takes a locked node.
478  */
479 int
480 vn_close(struct vnode *vp, int flags, kauth_cred_t cred)
481 {
482 	int error;
483 
484 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
485 	if (flags & FWRITE) {
486 		mutex_enter(vp->v_interlock);
487 		KASSERT(vp->v_writecount > 0);
488 		vp->v_writecount--;
489 		mutex_exit(vp->v_interlock);
490 	}
491 	error = VOP_CLOSE(vp, flags, cred);
492 	vput(vp);
493 	return error;
494 }
495 
496 static int
497 enforce_rlimit_fsize(struct vnode *vp, struct uio *uio, int ioflag)
498 {
499 	struct lwp *l = curlwp;
500 	off_t testoff;
501 
502 	if (uio->uio_rw != UIO_WRITE || vp->v_type != VREG)
503 		return 0;
504 
505 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
506 	if (ioflag & IO_APPEND)
507 		testoff = vp->v_size;
508 	else
509 		testoff = uio->uio_offset;
510 
511 	if (testoff + uio->uio_resid >
512 	    l->l_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
513 		mutex_enter(&proc_lock);
514 		psignal(l->l_proc, SIGXFSZ);
515 		mutex_exit(&proc_lock);
516 		return SET_ERROR(EFBIG);
517 	}
518 
519 	return 0;
520 }
521 
522 /*
523  * Package up an I/O request on a vnode into a uio and do it.
524  */
525 int
526 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
527     enum uio_seg segflg, int ioflg, kauth_cred_t cred, size_t *aresid,
528     struct lwp *l)
529 {
530 	struct uio auio;
531 	struct iovec aiov;
532 	int error;
533 
534 	if ((ioflg & IO_NODELOCKED) == 0) {
535 		if (rw == UIO_READ) {
536 			vn_lock(vp, LK_SHARED | LK_RETRY);
537 		} else /* UIO_WRITE */ {
538 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
539 		}
540 	}
541 	auio.uio_iov = &aiov;
542 	auio.uio_iovcnt = 1;
543 	aiov.iov_base = base;
544 	aiov.iov_len = len;
545 	auio.uio_resid = len;
546 	auio.uio_offset = offset;
547 	auio.uio_rw = rw;
548 	if (segflg == UIO_SYSSPACE) {
549 		UIO_SETUP_SYSSPACE(&auio);
550 	} else {
551 		auio.uio_vmspace = l->l_proc->p_vmspace;
552 	}
553 
554 	if ((error = enforce_rlimit_fsize(vp, &auio, ioflg)) != 0)
555 		goto out;
556 
557 	if (rw == UIO_READ) {
558 		error = VOP_READ(vp, &auio, ioflg, cred);
559 	} else {
560 		error = VOP_WRITE(vp, &auio, ioflg, cred);
561 	}
562 
563 	if (aresid)
564 		*aresid = auio.uio_resid;
565 	else
566 		if (auio.uio_resid && error == 0)
567 			error = SET_ERROR(EIO);
568 
569 out:
570 	if ((ioflg & IO_NODELOCKED) == 0) {
571 		VOP_UNLOCK(vp);
572 	}
573 	return error;
574 }
575 
576 int
577 vn_readdir(file_t *fp, char *bf, int segflg, u_int count, int *done,
578     struct lwp *l, off_t **cookies, int *ncookies)
579 {
580 	struct vnode *vp = fp->f_vnode;
581 	struct iovec aiov;
582 	struct uio auio;
583 	int error, eofflag;
584 
585 	/* Limit the size on any kernel buffers used by VOP_READDIR */
586 	count = uimin(MAXBSIZE, count);
587 
588 unionread:
589 	if (vp->v_type != VDIR)
590 		return SET_ERROR(EINVAL);
591 	aiov.iov_base = bf;
592 	aiov.iov_len = count;
593 	auio.uio_iov = &aiov;
594 	auio.uio_iovcnt = 1;
595 	auio.uio_rw = UIO_READ;
596 	if (segflg == UIO_SYSSPACE) {
597 		UIO_SETUP_SYSSPACE(&auio);
598 	} else {
599 		KASSERT(l == curlwp);
600 		auio.uio_vmspace = l->l_proc->p_vmspace;
601 	}
602 	auio.uio_resid = count;
603 	vn_lock(vp, LK_SHARED | LK_RETRY);
604 	mutex_enter(&fp->f_lock);
605 	auio.uio_offset = fp->f_offset;
606 	mutex_exit(&fp->f_lock);
607 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, cookies,
608 	    ncookies);
609 	mutex_enter(&fp->f_lock);
610 	fp->f_offset = auio.uio_offset;
611 	mutex_exit(&fp->f_lock);
612 	VOP_UNLOCK(vp);
613 	if (error)
614 		return error;
615 
616 	if (count == auio.uio_resid && vn_union_readdir_hook) {
617 		struct vnode *ovp = vp;
618 
619 		error = (*vn_union_readdir_hook)(&vp, fp, l);
620 		if (error)
621 			return error;
622 		if (vp != ovp)
623 			goto unionread;
624 	}
625 
626 	if (count == auio.uio_resid && (vp->v_vflag & VV_ROOT) &&
627 	    (vp->v_mount->mnt_flag & MNT_UNION)) {
628 		struct vnode *tvp = vp;
629 		vp = vp->v_mount->mnt_vnodecovered;
630 		vref(vp);
631 		mutex_enter(&fp->f_lock);
632 		fp->f_vnode = vp;
633 		fp->f_offset = 0;
634 		mutex_exit(&fp->f_lock);
635 		vrele(tvp);
636 		goto unionread;
637 	}
638 	*done = count - auio.uio_resid;
639 	return error;
640 }
641 
642 /*
643  * File table vnode read routine.
644  */
645 static int
646 vn_read(file_t *fp, off_t *offset, struct uio *uio, kauth_cred_t cred,
647     int flags)
648 {
649 	struct vnode *vp = fp->f_vnode;
650 	int error, ioflag, fflag;
651 	size_t count;
652 
653 	ioflag = IO_ADV_ENCODE(fp->f_advice);
654 	fflag = fp->f_flag;
655 	if (fflag & FNONBLOCK)
656 		ioflag |= IO_NDELAY;
657 	if ((fflag & (FFSYNC | FRSYNC)) == (FFSYNC | FRSYNC))
658 		ioflag |= IO_SYNC;
659 	if (fflag & FALTIO)
660 		ioflag |= IO_ALTSEMANTICS;
661 	if (fflag & FDIRECT)
662 		ioflag |= IO_DIRECT;
663 	if (offset == &fp->f_offset && (flags & FOF_UPDATE_OFFSET) != 0)
664 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
665 	else
666 		vn_lock(vp, LK_SHARED | LK_RETRY);
667 	if (__predict_false(vp->v_type == VDIR) &&
668 	    offset == &fp->f_offset && (flags & FOF_UPDATE_OFFSET) == 0)
669 		mutex_enter(&fp->f_lock);
670 	uio->uio_offset = *offset;
671 	if (__predict_false(vp->v_type == VDIR) &&
672 	    offset == &fp->f_offset && (flags & FOF_UPDATE_OFFSET) == 0)
673 		mutex_enter(&fp->f_lock);
674 	count = uio->uio_resid;
675 	error = VOP_READ(vp, uio, ioflag, cred);
676 	if (flags & FOF_UPDATE_OFFSET)
677 		*offset += count - uio->uio_resid;
678 	VOP_UNLOCK(vp);
679 	return error;
680 }
681 
682 /*
683  * File table vnode write routine.
684  */
685 static int
686 vn_write(file_t *fp, off_t *offset, struct uio *uio, kauth_cred_t cred,
687     int flags)
688 {
689 	struct vnode *vp = fp->f_vnode;
690 	int error, ioflag, fflag;
691 	size_t count;
692 
693 	ioflag = IO_ADV_ENCODE(fp->f_advice) | IO_UNIT;
694 	fflag = fp->f_flag;
695 	if (vp->v_type == VREG && (fflag & O_APPEND))
696 		ioflag |= IO_APPEND;
697 	if (fflag & FNONBLOCK)
698 		ioflag |= IO_NDELAY;
699 	if (fflag & FFSYNC ||
700 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
701 		ioflag |= IO_SYNC;
702 	else if (fflag & FDSYNC)
703 		ioflag |= IO_DSYNC;
704 	if (fflag & FALTIO)
705 		ioflag |= IO_ALTSEMANTICS;
706 	if (fflag & FDIRECT)
707 		ioflag |= IO_DIRECT;
708 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
709 	uio->uio_offset = *offset;
710 	count = uio->uio_resid;
711 
712 	if ((error = enforce_rlimit_fsize(vp, uio, ioflag)) != 0)
713 		goto out;
714 
715 	error = VOP_WRITE(vp, uio, ioflag, cred);
716 
717 	if (flags & FOF_UPDATE_OFFSET) {
718 		if (ioflag & IO_APPEND) {
719 			/*
720 			 * SUSv3 describes behaviour for count = 0 as
721 			 * following: "Before any action ... is taken,
722 			 * and if nbyte is zero and the file is a
723 			 * regular file, the write() function ... in
724 			 * the absence of errors ... shall return zero
725 			 * and have no other results."
726 			 */
727 			if (count)
728 				*offset = uio->uio_offset;
729 		} else
730 			*offset += count - uio->uio_resid;
731 	}
732 
733 out:
734 	VOP_UNLOCK(vp);
735 	return error;
736 }
737 
738 /*
739  * File table vnode stat routine.
740  */
741 static int
742 vn_statfile(file_t *fp, struct stat *sb)
743 {
744 	struct vnode *vp = fp->f_vnode;
745 	int error;
746 
747 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
748 	error = vn_stat(vp, sb);
749 	VOP_UNLOCK(vp);
750 	return error;
751 }
752 
753 int
754 vn_stat(struct vnode *vp, struct stat *sb)
755 {
756 	struct vattr va;
757 	int error;
758 	mode_t mode;
759 
760 	memset(&va, 0, sizeof(va));
761 	error = VOP_GETATTR(vp, &va, kauth_cred_get());
762 	if (error)
763 		return error;
764 	/*
765 	 * Copy from vattr table
766 	 */
767 	memset(sb, 0, sizeof(*sb));
768 	sb->st_dev = va.va_fsid;
769 	sb->st_ino = va.va_fileid;
770 	mode = va.va_mode;
771 	switch (vp->v_type) {
772 	case VREG:
773 		mode |= S_IFREG;
774 		break;
775 	case VDIR:
776 		mode |= S_IFDIR;
777 		break;
778 	case VBLK:
779 		mode |= S_IFBLK;
780 		break;
781 	case VCHR:
782 		mode |= S_IFCHR;
783 		break;
784 	case VLNK:
785 		mode |= S_IFLNK;
786 		break;
787 	case VSOCK:
788 		mode |= S_IFSOCK;
789 		break;
790 	case VFIFO:
791 		mode |= S_IFIFO;
792 		break;
793 	default:
794 		return SET_ERROR(EBADF);
795 	}
796 	sb->st_mode = mode;
797 	sb->st_nlink = va.va_nlink;
798 	sb->st_uid = va.va_uid;
799 	sb->st_gid = va.va_gid;
800 	sb->st_rdev = va.va_rdev;
801 	sb->st_size = va.va_size;
802 	sb->st_atimespec = va.va_atime;
803 	sb->st_mtimespec = va.va_mtime;
804 	sb->st_ctimespec = va.va_ctime;
805 	sb->st_birthtimespec = va.va_birthtime;
806 	sb->st_blksize = va.va_blocksize;
807 	sb->st_flags = va.va_flags;
808 	sb->st_gen = 0;
809 	sb->st_blocks = va.va_bytes / S_BLKSIZE;
810 	return 0;
811 }
812 
813 /*
814  * File table vnode fcntl routine.
815  */
816 static int
817 vn_fcntl(file_t *fp, u_int com, void *data)
818 {
819 	struct vnode *vp = fp->f_vnode;
820 	int error;
821 
822 	error = VOP_FCNTL(vp, com, data, fp->f_flag, kauth_cred_get());
823 	return error;
824 }
825 
826 /*
827  * File table vnode ioctl routine.
828  */
829 static int
830 vn_ioctl(file_t *fp, u_long com, void *data)
831 {
832 	struct vnode *vp = fp->f_vnode, *ovp;
833 	struct vattr vattr;
834 	int error;
835 
836 	switch (vp->v_type) {
837 
838 	case VREG:
839 	case VDIR:
840 		if (com == FIONREAD) {
841 			vn_lock(vp, LK_SHARED | LK_RETRY);
842 			error = VOP_GETATTR(vp, &vattr, kauth_cred_get());
843 			if (error == 0) {
844 				if (vp->v_type == VDIR)
845 					mutex_enter(&fp->f_lock);
846 				*(int *)data = vattr.va_size - fp->f_offset;
847 				if (vp->v_type == VDIR)
848 					mutex_exit(&fp->f_lock);
849 			}
850 			VOP_UNLOCK(vp);
851 			if (error)
852 				return error;
853 			return 0;
854 		}
855 		if ((com == FIONWRITE) || (com == FIONSPACE)) {
856 			/*
857 			 * Files don't have send queues, so there never
858 			 * are any bytes in them, nor is there any
859 			 * open space in them.
860 			 */
861 			*(int *)data = 0;
862 			return 0;
863 		}
864 		if (com == FIOGETBMAP) {
865 			daddr_t *block;
866 
867 			if (*(daddr_t *)data < 0)
868 				return SET_ERROR(EINVAL);
869 			block = (daddr_t *)data;
870 			vn_lock(vp, LK_SHARED | LK_RETRY);
871 			error = VOP_BMAP(vp, *block, NULL, block, NULL);
872 			VOP_UNLOCK(vp);
873 			return error;
874 		}
875 		if (com == OFIOGETBMAP) {
876 			daddr_t ibn, obn;
877 
878 			if (*(int32_t *)data < 0)
879 				return SET_ERROR(EINVAL);
880 			ibn = (daddr_t)*(int32_t *)data;
881 			vn_lock(vp, LK_SHARED | LK_RETRY);
882 			error = VOP_BMAP(vp, ibn, NULL, &obn, NULL);
883 			VOP_UNLOCK(vp);
884 			*(int32_t *)data = (int32_t)obn;
885 			return error;
886 		}
887 		if (com == FIONBIO || com == FIOASYNC)	/* XXX */
888 			return 0;			/* XXX */
889 		/* FALLTHROUGH */
890 	case VFIFO:
891 	case VCHR:
892 	case VBLK:
893 		error = VOP_IOCTL(vp, com, data, fp->f_flag, kauth_cred_get());
894 		if (error == 0 && com == TIOCSCTTY) {
895 			vref(vp);
896 			mutex_enter(&proc_lock);
897 			ovp = curproc->p_session->s_ttyvp;
898 			curproc->p_session->s_ttyvp = vp;
899 			mutex_exit(&proc_lock);
900 			if (ovp != NULL)
901 				vrele(ovp);
902 		}
903 		return error;
904 
905 	default:
906 		return SET_ERROR(EPASSTHROUGH);
907 	}
908 }
909 
910 /*
911  * File table vnode poll routine.
912  */
913 static int
914 vn_poll(file_t *fp, int events)
915 {
916 
917 	return VOP_POLL(fp->f_vnode, events);
918 }
919 
920 /*
921  * File table vnode kqfilter routine.
922  */
923 int
924 vn_kqfilter(file_t *fp, struct knote *kn)
925 {
926 
927 	return VOP_KQFILTER(fp->f_vnode, kn);
928 }
929 
930 static int
931 vn_mmap(struct file *fp, off_t *offp, size_t size, int prot, int *flagsp,
932     int *advicep, struct uvm_object **uobjp, int *maxprotp)
933 {
934 	struct uvm_object *uobj;
935 	struct vnode *vp;
936 	struct vattr va;
937 	struct lwp *l;
938 	vm_prot_t maxprot;
939 	off_t off;
940 	int error, flags;
941 	bool needwritemap;
942 
943 	l = curlwp;
944 
945 	off = *offp;
946 	flags = *flagsp;
947 	maxprot = VM_PROT_EXECUTE;
948 
949 	KASSERT(size > 0);
950 
951 	vp = fp->f_vnode;
952 	if (vp->v_type != VREG && vp->v_type != VCHR &&
953 	    vp->v_type != VBLK) {
954 		/* only REG/CHR/BLK support mmap */
955 		return SET_ERROR(ENODEV);
956 	}
957 	if (vp->v_type != VCHR && off < 0) {
958 		return SET_ERROR(EINVAL);
959 	}
960 #if SIZE_MAX > UINT32_MAX	/* XXX -Wtype-limits */
961 	if (vp->v_type != VCHR && size > __type_max(off_t)) {
962 		return SET_ERROR(EOVERFLOW);
963 	}
964 #endif
965 	if (vp->v_type != VCHR && off > __type_max(off_t) - size) {
966 		/* no offset wrapping */
967 		return SET_ERROR(EOVERFLOW);
968 	}
969 
970 	/* special case: catch SunOS style /dev/zero */
971 	if (vp->v_type == VCHR &&
972 	    (vp->v_rdev == zerodev || COMPAT_ZERODEV(vp->v_rdev))) {
973 		*uobjp = NULL;
974 		*maxprotp = VM_PROT_ALL;
975 		return 0;
976 	}
977 
978 	/*
979 	 * Old programs may not select a specific sharing type, so
980 	 * default to an appropriate one.
981 	 *
982 	 * XXX: how does MAP_ANON fit in the picture?
983 	 */
984 	if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) {
985 #if defined(DEBUG)
986 		struct proc *p = l->l_proc;
987 		printf("WARNING: defaulted mmap() share type to "
988 		    "%s (pid %d command %s)\n",
989 		    vp->v_type == VCHR ? "MAP_SHARED" : "MAP_PRIVATE",
990 		    p->p_pid,
991 		    p->p_comm);
992 #endif
993 		if (vp->v_type == VCHR)
994 			flags |= MAP_SHARED;	/* for a device */
995 		else
996 			flags |= MAP_PRIVATE;	/* for a file */
997 	}
998 
999 	/*
1000 	 * MAP_PRIVATE device mappings don't make sense (and aren't
1001 	 * supported anyway).  However, some programs rely on this,
1002 	 * so just change it to MAP_SHARED.
1003 	 */
1004 	if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) {
1005 		flags = (flags & ~MAP_PRIVATE) | MAP_SHARED;
1006 	}
1007 
1008 	/*
1009 	 * now check protection
1010 	 */
1011 
1012 	/* check read access */
1013 	if (fp->f_flag & FREAD)
1014 		maxprot |= VM_PROT_READ;
1015 	else if (prot & PROT_READ) {
1016 		return SET_ERROR(EACCES);
1017 	}
1018 
1019 	/* check write access, shared case first */
1020 	if (flags & MAP_SHARED) {
1021 		/*
1022 		 * if the file is writable, only add PROT_WRITE to
1023 		 * maxprot if the file is not immutable, append-only.
1024 		 * otherwise, if we have asked for PROT_WRITE, return
1025 		 * EPERM.
1026 		 */
1027 		if (fp->f_flag & FWRITE) {
1028 			vn_lock(vp, LK_SHARED | LK_RETRY);
1029 			error = VOP_GETATTR(vp, &va, l->l_cred);
1030 			VOP_UNLOCK(vp);
1031 			if (error) {
1032 				return error;
1033 			}
1034 			if ((va.va_flags &
1035 				(SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0)
1036 				maxprot |= VM_PROT_WRITE;
1037 			else if (prot & PROT_WRITE) {
1038 				return SET_ERROR(EPERM);
1039 			}
1040 		} else if (prot & PROT_WRITE) {
1041 			return SET_ERROR(EACCES);
1042 		}
1043 	} else {
1044 		/* MAP_PRIVATE mappings can always write to */
1045 		maxprot |= VM_PROT_WRITE;
1046 	}
1047 
1048 	/*
1049 	 * Don't allow mmap for EXEC if the file system
1050 	 * is mounted NOEXEC.
1051 	 */
1052 	if ((prot & PROT_EXEC) != 0 &&
1053 	    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
1054 		return SET_ERROR(EACCES);
1055 	}
1056 
1057 	if (vp->v_type != VCHR) {
1058 		error = VOP_MMAP(vp, prot, curlwp->l_cred);
1059 		if (error) {
1060 			return error;
1061 		}
1062 		vref(vp);
1063 		uobj = &vp->v_uobj;
1064 
1065 		/*
1066 		 * If the vnode is being mapped with PROT_EXEC,
1067 		 * then mark it as text.
1068 		 */
1069 		if (prot & PROT_EXEC) {
1070 			vn_markexec(vp);
1071 		}
1072 	} else {
1073 		int i = maxprot;
1074 
1075 		/*
1076 		 * XXX Some devices don't like to be mapped with
1077 		 * XXX PROT_EXEC or PROT_WRITE, but we don't really
1078 		 * XXX have a better way of handling this, right now
1079 		 */
1080 		do {
1081 			uobj = udv_attach(vp->v_rdev,
1082 			    (flags & MAP_SHARED) ? i : (i & ~VM_PROT_WRITE),
1083 			    off, size);
1084 			i--;
1085 		} while ((uobj == NULL) && (i > 0));
1086 		if (uobj == NULL) {
1087 			return SET_ERROR(EINVAL);
1088 		}
1089 		*advicep = UVM_ADV_RANDOM;
1090 	}
1091 
1092 	/*
1093 	 * Set vnode flags to indicate the new kinds of mapping.
1094 	 * We take the vnode lock in exclusive mode here to serialize
1095 	 * with direct I/O.
1096 	 *
1097 	 * Safe to check for these flag values without a lock, as
1098 	 * long as a reference to the vnode is held.
1099 	 */
1100 	needwritemap = (vp->v_iflag & VI_WRMAP) == 0 &&
1101 	    (flags & MAP_SHARED) != 0 &&
1102 	    (maxprot & VM_PROT_WRITE) != 0;
1103 	if ((vp->v_vflag & VV_MAPPED) == 0 || needwritemap) {
1104 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1105 		vp->v_vflag |= VV_MAPPED;
1106 		if (needwritemap) {
1107 			rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
1108 			mutex_enter(vp->v_interlock);
1109 			vp->v_iflag |= VI_WRMAP;
1110 			mutex_exit(vp->v_interlock);
1111 			rw_exit(vp->v_uobj.vmobjlock);
1112 		}
1113 		VOP_UNLOCK(vp);
1114 	}
1115 
1116 #if NVERIEXEC > 0
1117 	/*
1118 	 * Check if the file can be executed indirectly.
1119 	 *
1120 	 * XXX: This gives false warnings about "Incorrect access type"
1121 	 * XXX: if the mapping is not executable. Harmless, but will be
1122 	 * XXX: fixed as part of other changes.
1123 	 */
1124 	if (veriexec_verify(l, vp, "(mmap)", VERIEXEC_INDIRECT,
1125 		NULL)) {
1126 
1127 		/*
1128 		 * Don't allow executable mappings if we can't
1129 		 * indirectly execute the file.
1130 		 */
1131 		if (prot & VM_PROT_EXECUTE) {
1132 			return SET_ERROR(EPERM);
1133 		}
1134 
1135 		/*
1136 		 * Strip the executable bit from 'maxprot' to make sure
1137 		 * it can't be made executable later.
1138 		 */
1139 		maxprot &= ~VM_PROT_EXECUTE;
1140 	}
1141 #endif /* NVERIEXEC > 0 */
1142 
1143 	*uobjp = uobj;
1144 	*maxprotp = maxprot;
1145 	*flagsp = flags;
1146 
1147 	return 0;
1148 }
1149 
1150 static int
1151 vn_seek(struct file *fp, off_t delta, int whence, off_t *newoffp, int flags)
1152 {
1153 	const off_t OFF_MIN = __type_min(off_t);
1154 	const off_t OFF_MAX = __type_max(off_t);
1155 	kauth_cred_t cred = fp->f_cred;
1156 	off_t oldoff, newoff;
1157 	struct vnode *vp = fp->f_vnode;
1158 	struct vattr vattr;
1159 	int error;
1160 
1161 	if (vp->v_type == VFIFO)
1162 		return SET_ERROR(ESPIPE);
1163 
1164 	if (flags & FOF_UPDATE_OFFSET)
1165 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1166 	else
1167 		vn_lock(vp, LK_SHARED | LK_RETRY);
1168 
1169 	/* Compute the old and new offsets.  */
1170 	if (vp->v_type == VDIR && (flags & FOF_UPDATE_OFFSET) == 0)
1171 		mutex_enter(&fp->f_lock);
1172 	oldoff = fp->f_offset;
1173 	if (vp->v_type == VDIR && (flags & FOF_UPDATE_OFFSET) == 0)
1174 		mutex_exit(&fp->f_lock);
1175 	switch (whence) {
1176 	case SEEK_CUR:
1177 		if (delta > 0) {
1178 			if (oldoff > 0 && delta > OFF_MAX - oldoff) {
1179 				newoff = OFF_MAX;
1180 				break;
1181 			}
1182 		} else {
1183 			if (oldoff < 0 && delta < OFF_MIN - oldoff) {
1184 				newoff = OFF_MIN;
1185 				break;
1186 			}
1187 		}
1188 		newoff = oldoff + delta;
1189 		break;
1190 	case SEEK_END:
1191 		error = VOP_GETATTR(vp, &vattr, cred);
1192 		if (error)
1193 			goto out;
1194 		if (vattr.va_size > OFF_MAX ||
1195 		    delta > OFF_MAX - (off_t)vattr.va_size) {
1196 			newoff = OFF_MAX;
1197 			break;
1198 		}
1199 		newoff = delta + vattr.va_size;
1200 		break;
1201 	case SEEK_SET:
1202 		newoff = delta;
1203 		break;
1204 	default:
1205 		error = SET_ERROR(EINVAL);
1206 		goto out;
1207 	}
1208 
1209 	/* Pass the proposed change to the file system to audit.  */
1210 	error = VOP_SEEK(vp, oldoff, newoff, cred);
1211 	if (error)
1212 		goto out;
1213 
1214 	/* Success!  */
1215 	if (newoffp)
1216 		*newoffp = newoff;
1217 	if (flags & FOF_UPDATE_OFFSET)
1218 		fp->f_offset = newoff;
1219 	error = 0;
1220 
1221 out:	VOP_UNLOCK(vp);
1222 	return error;
1223 }
1224 
1225 static int
1226 vn_advlock(struct file *fp, void *id, int op, struct flock *fl, int flags)
1227 {
1228 	struct vnode *const vp = fp->f_vnode;
1229 
1230 	if (fl->l_whence == SEEK_CUR) {
1231 		vn_lock(vp, LK_SHARED | LK_RETRY);
1232 		fl->l_start += fp->f_offset;
1233 		VOP_UNLOCK(vp);
1234 	}
1235 
1236 	return VOP_ADVLOCK(vp, id, op, fl, flags);
1237 }
1238 
1239 static int
1240 vn_fpathconf(struct file *fp, int name, register_t *retval)
1241 {
1242 	struct vnode *const vp = fp->f_vnode;
1243 	int error;
1244 
1245 	vn_lock(vp, LK_SHARED | LK_RETRY);
1246 	error = VOP_PATHCONF(vp, name, retval);
1247 	VOP_UNLOCK(vp);
1248 
1249 	return error;
1250 }
1251 
1252 static int
1253 vn_posix_fadvise(struct file *fp, off_t offset, off_t len, int advice)
1254 {
1255 	const off_t OFF_MAX = __type_max(off_t);
1256 	struct vnode *vp = fp->f_vnode;
1257 	off_t endoffset;
1258 	int error;
1259 
1260 	if (offset < 0) {
1261 		return SET_ERROR(EINVAL);
1262 	}
1263 	if (len == 0) {
1264 		endoffset = OFF_MAX;
1265 	} else if (len > 0 && (OFF_MAX - offset) >= len) {
1266 		endoffset = offset + len;
1267 	} else {
1268 		return SET_ERROR(EINVAL);
1269 	}
1270 
1271 	CTASSERT(POSIX_FADV_NORMAL == UVM_ADV_NORMAL);
1272 	CTASSERT(POSIX_FADV_RANDOM == UVM_ADV_RANDOM);
1273 	CTASSERT(POSIX_FADV_SEQUENTIAL == UVM_ADV_SEQUENTIAL);
1274 
1275 	switch (advice) {
1276 	case POSIX_FADV_WILLNEED:
1277 	case POSIX_FADV_DONTNEED:
1278 		if (vp->v_type != VREG && vp->v_type != VBLK)
1279 			return 0;
1280 		break;
1281 	}
1282 
1283 	switch (advice) {
1284 	case POSIX_FADV_NORMAL:
1285 	case POSIX_FADV_RANDOM:
1286 	case POSIX_FADV_SEQUENTIAL:
1287 		/*
1288 		 * We ignore offset and size.  Must lock the file to
1289 		 * do this, as f_advice is sub-word sized.
1290 		 */
1291 		mutex_enter(&fp->f_lock);
1292 		fp->f_advice = (u_char)advice;
1293 		mutex_exit(&fp->f_lock);
1294 		error = 0;
1295 		break;
1296 
1297 	case POSIX_FADV_WILLNEED:
1298 		error = uvm_readahead(&vp->v_uobj, offset, endoffset - offset);
1299 		break;
1300 
1301 	case POSIX_FADV_DONTNEED:
1302 		/*
1303 		 * Align the region to page boundaries as VOP_PUTPAGES expects
1304 		 * by shrinking it.  We shrink instead of expand because we
1305 		 * do not want to deactivate cache outside of the requested
1306 		 * region.  It means that if the specified region is smaller
1307 		 * than PAGE_SIZE, we do nothing.
1308 		 */
1309 		if (offset <= trunc_page(OFF_MAX) &&
1310 		    round_page(offset) < trunc_page(endoffset)) {
1311 			rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
1312 			error = VOP_PUTPAGES(vp,
1313 			    round_page(offset), trunc_page(endoffset),
1314 			    PGO_DEACTIVATE | PGO_CLEANIT);
1315 		} else {
1316 			error = 0;
1317 		}
1318 		break;
1319 
1320 	case POSIX_FADV_NOREUSE:
1321 		/* Not implemented yet. */
1322 		error = 0;
1323 		break;
1324 	default:
1325 		error = SET_ERROR(EINVAL);
1326 		break;
1327 	}
1328 
1329 	return error;
1330 }
1331 
1332 static int
1333 vn_truncate(file_t *fp, off_t length)
1334 {
1335 	struct vattr vattr;
1336 	struct vnode *vp;
1337 	int error = 0;
1338 
1339 	if (length < 0)
1340 		return SET_ERROR(EINVAL);
1341 
1342 	if ((fp->f_flag & FWRITE) == 0)
1343 		return SET_ERROR(EINVAL);
1344 	vp = fp->f_vnode;
1345 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1346 	if (vp->v_type == VDIR)
1347 		error = SET_ERROR(EISDIR);
1348 	else if ((error = vn_writechk(vp)) == 0) {
1349 		vattr_null(&vattr);
1350 		vattr.va_size = length;
1351 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1352 	}
1353 	VOP_UNLOCK(vp);
1354 
1355 	return error;
1356 }
1357 
1358 
1359 /*
1360  * Check that the vnode is still valid, and if so
1361  * acquire requested lock.
1362  */
1363 int
1364 vn_lock(struct vnode *vp, int flags)
1365 {
1366 	struct lwp *l;
1367 	int error;
1368 
1369 	KASSERT(vrefcnt(vp) > 0);
1370 	KASSERT((flags & ~(LK_SHARED|LK_EXCLUSIVE|LK_NOWAIT|LK_RETRY|
1371 	    LK_UPGRADE|LK_DOWNGRADE)) == 0);
1372 	KASSERT((flags & LK_NOWAIT) != 0 || !mutex_owned(vp->v_interlock));
1373 
1374 #ifdef DIAGNOSTIC
1375 	if (wapbl_vphaswapbl(vp))
1376 		WAPBL_JUNLOCK_ASSERT(wapbl_vptomp(vp));
1377 #endif
1378 
1379 	/* Get a more useful report for lockstat. */
1380 	l = curlwp;
1381 	KASSERT(l->l_rwcallsite == 0);
1382 	l->l_rwcallsite = (uintptr_t)__builtin_return_address(0);
1383 
1384 	error = VOP_LOCK(vp, flags);
1385 
1386 	l->l_rwcallsite = 0;
1387 
1388 	switch (flags & (LK_RETRY | LK_NOWAIT)) {
1389 	case 0:
1390 		KASSERT(error == 0 || error == ENOENT);
1391 		break;
1392 	case LK_RETRY:
1393 		KASSERT(error == 0);
1394 		break;
1395 	case LK_NOWAIT:
1396 		KASSERT(error == 0 || error == EBUSY || error == ENOENT);
1397 		break;
1398 	case LK_RETRY | LK_NOWAIT:
1399 		KASSERT(error == 0 || error == EBUSY);
1400 		break;
1401 	}
1402 
1403 	return error;
1404 }
1405 
1406 /*
1407  * File table vnode close routine.
1408  */
1409 static int
1410 vn_closefile(file_t *fp)
1411 {
1412 
1413 	return vn_close(fp->f_vnode, fp->f_flag, fp->f_cred);
1414 }
1415 
1416 /*
1417  * Simplified in-kernel wrapper calls for extended attribute access.
1418  * Both calls pass in a NULL credential, authorizing a "kernel" access.
1419  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1420  */
1421 int
1422 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1423     const char *attrname, size_t *buflen, void *bf, struct lwp *l)
1424 {
1425 	struct uio auio;
1426 	struct iovec aiov;
1427 	int error;
1428 
1429 	aiov.iov_len = *buflen;
1430 	aiov.iov_base = bf;
1431 
1432 	auio.uio_iov = &aiov;
1433 	auio.uio_iovcnt = 1;
1434 	auio.uio_rw = UIO_READ;
1435 	auio.uio_offset = 0;
1436 	auio.uio_resid = *buflen;
1437 	UIO_SETUP_SYSSPACE(&auio);
1438 
1439 	if ((ioflg & IO_NODELOCKED) == 0)
1440 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1441 
1442 	error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL,
1443 	    NOCRED);
1444 
1445 	if ((ioflg & IO_NODELOCKED) == 0)
1446 		VOP_UNLOCK(vp);
1447 
1448 	if (error == 0)
1449 		*buflen = *buflen - auio.uio_resid;
1450 
1451 	return error;
1452 }
1453 
1454 /*
1455  * XXX Failure mode if partially written?
1456  */
1457 int
1458 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1459     const char *attrname, size_t buflen, const void *bf, struct lwp *l)
1460 {
1461 	struct uio auio;
1462 	struct iovec aiov;
1463 	int error;
1464 
1465 	aiov.iov_len = buflen;
1466 	aiov.iov_base = __UNCONST(bf);		/* XXXUNCONST kills const */
1467 
1468 	auio.uio_iov = &aiov;
1469 	auio.uio_iovcnt = 1;
1470 	auio.uio_rw = UIO_WRITE;
1471 	auio.uio_offset = 0;
1472 	auio.uio_resid = buflen;
1473 	UIO_SETUP_SYSSPACE(&auio);
1474 
1475 	if ((ioflg & IO_NODELOCKED) == 0) {
1476 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1477 	}
1478 
1479 	error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NOCRED);
1480 
1481 	if ((ioflg & IO_NODELOCKED) == 0) {
1482 		VOP_UNLOCK(vp);
1483 	}
1484 
1485 	return error;
1486 }
1487 
1488 int
1489 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1490     const char *attrname, struct lwp *l)
1491 {
1492 	int error;
1493 
1494 	if ((ioflg & IO_NODELOCKED) == 0) {
1495 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1496 	}
1497 
1498 	error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NOCRED);
1499 	if (error == EOPNOTSUPP)
1500 		error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1501 		    NOCRED);
1502 
1503 	if ((ioflg & IO_NODELOCKED) == 0) {
1504 		VOP_UNLOCK(vp);
1505 	}
1506 
1507 	return error;
1508 }
1509 
1510 int
1511 vn_fifo_bypass(void *v)
1512 {
1513 	struct vop_generic_args *ap = v;
1514 
1515 	return VOCALL(fifo_vnodeop_p, ap->a_desc->vdesc_offset, v);
1516 }
1517 
1518 /*
1519  * Open block device by device number
1520  */
1521 int
1522 vn_bdev_open(dev_t dev, struct vnode **vpp, struct lwp *l)
1523 {
1524 	int     error;
1525 
1526 	if ((error = bdevvp(dev, vpp)) != 0)
1527 		return error;
1528 
1529 	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
1530 	if ((error = VOP_OPEN(*vpp, FREAD | FWRITE, l->l_cred)) != 0) {
1531 		vput(*vpp);
1532 		return error;
1533 	}
1534 	mutex_enter((*vpp)->v_interlock);
1535 	(*vpp)->v_writecount++;
1536 	mutex_exit((*vpp)->v_interlock);
1537 	VOP_UNLOCK(*vpp);
1538 
1539 	return 0;
1540 }
1541 
1542 /*
1543  * Lookup the provided name in the filesystem.  If the file exists,
1544  * is a valid block device, and isn't being used by anyone else,
1545  * set *vpp to the file's vnode.
1546  */
1547 int
1548 vn_bdev_openpath(struct pathbuf *pb, struct vnode **vpp, struct lwp *l)
1549 {
1550 	struct vnode *vp;
1551 	dev_t dev;
1552 	enum vtype vt;
1553 	int     error;
1554 
1555 	error = vn_open(NULL, pb, 0, FREAD | FWRITE, 0, &vp, NULL, NULL);
1556 	if (error != 0)
1557 		return error;
1558 
1559 	dev = vp->v_rdev;
1560 	vt = vp->v_type;
1561 
1562 	VOP_UNLOCK(vp);
1563 	(void) vn_close(vp, FREAD | FWRITE, l->l_cred);
1564 
1565 	if (vt != VBLK)
1566 		return SET_ERROR(ENOTBLK);
1567 
1568 	return vn_bdev_open(dev, vpp, l);
1569 }
1570 
1571 static long
1572 vn_knote_to_interest(const struct knote *kn)
1573 {
1574 
1575 	switch (kn->kn_filter) {
1576 	case EVFILT_READ:
1577 		/*
1578 		 * Writing to the file or changing its attributes can
1579 		 * set the file size, which impacts the readability
1580 		 * filter.
1581 		 *
1582 		 * (No need to set NOTE_EXTEND here; it's only ever
1583 		 * send with other hints; see vnode_if.c.)
1584 		 */
1585 		return NOTE_WRITE | NOTE_ATTRIB;
1586 
1587 	case EVFILT_VNODE:
1588 		return kn->kn_sfflags;
1589 
1590 	case EVFILT_WRITE:
1591 	default:
1592 		return 0;
1593 	}
1594 }
1595 
1596 void
1597 vn_knote_attach(struct vnode *vp, struct knote *kn)
1598 {
1599 	struct vnode_klist *vk = vp->v_klist;
1600 	long interest = 0;
1601 
1602 	/*
1603 	 * In the case of layered / stacked file systems, knotes
1604 	 * should only ever be associated with the base vnode.
1605 	 */
1606 	KASSERT(kn->kn_hook == vp);
1607 	KASSERT(vp->v_klist == &VNODE_TO_VIMPL(vp)->vi_klist);
1608 
1609 	/*
1610 	 * We maintain a bitmask of the kevents that there is interest in,
1611 	 * to minimize the impact of having watchers.  It's silly to have
1612 	 * to traverse vn_klist every time a read or write happens simply
1613 	 * because there is someone interested in knowing when the file
1614 	 * is deleted, for example.
1615 	 */
1616 
1617 	mutex_enter(vp->v_interlock);
1618 	SLIST_INSERT_HEAD(&vk->vk_klist, kn, kn_selnext);
1619 	SLIST_FOREACH(kn, &vk->vk_klist, kn_selnext) {
1620 		interest |= vn_knote_to_interest(kn);
1621 	}
1622 	vk->vk_interest = interest;
1623 	mutex_exit(vp->v_interlock);
1624 }
1625 
1626 void
1627 vn_knote_detach(struct vnode *vp, struct knote *kn)
1628 {
1629 	struct vnode_klist *vk = vp->v_klist;
1630 	long interest = 0;
1631 
1632 	/* See above. */
1633 	KASSERT(kn->kn_hook == vp);
1634 	KASSERT(vp->v_klist == &VNODE_TO_VIMPL(vp)->vi_klist);
1635 
1636 	/*
1637 	 * We special case removing the head of the list, because:
1638 	 *
1639 	 * 1. It's extremely likely that we're detaching the only
1640 	 *    knote.
1641 	 *
1642 	 * 2. We're already traversing the whole list, so we don't
1643 	 *    want to use the generic SLIST_REMOVE() which would
1644 	 *    traverse it *again*.
1645 	 */
1646 
1647 	mutex_enter(vp->v_interlock);
1648 	if (__predict_true(kn == SLIST_FIRST(&vk->vk_klist))) {
1649 		SLIST_REMOVE_HEAD(&vk->vk_klist, kn_selnext);
1650 		SLIST_FOREACH(kn, &vk->vk_klist, kn_selnext) {
1651 			interest |= vn_knote_to_interest(kn);
1652 		}
1653 		vk->vk_interest = interest;
1654 	} else {
1655 		struct knote *thiskn, *nextkn, *prevkn = NULL;
1656 
1657 		SLIST_FOREACH_SAFE(thiskn, &vk->vk_klist, kn_selnext, nextkn) {
1658 			if (thiskn == kn) {
1659 				KASSERT(kn != NULL);
1660 				KASSERT(prevkn != NULL);
1661 				SLIST_REMOVE_AFTER(prevkn, kn_selnext);
1662 				kn = NULL;
1663 			} else {
1664 				interest |= vn_knote_to_interest(thiskn);
1665 				prevkn = thiskn;
1666 			}
1667 		}
1668 		vk->vk_interest = interest;
1669 	}
1670 	mutex_exit(vp->v_interlock);
1671 }
1672