xref: /netbsd-src/sys/kern/sys_generic.c (revision 53b02e147d4ed531c0d2a5ca9b3e8026ba3e99b5)
1 /*	$NetBSD: sys_generic.c,v 1.133 2021/09/11 10:08:55 riastradh Exp $	*/
2 
3 /*-
4  * Copyright (c) 2007, 2008, 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  *	@(#)sys_generic.c	8.9 (Berkeley) 2/14/95
66  */
67 
68 /*
69  * System calls relating to files.
70  */
71 
72 #include <sys/cdefs.h>
73 __KERNEL_RCSID(0, "$NetBSD: sys_generic.c,v 1.133 2021/09/11 10:08:55 riastradh Exp $");
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/filedesc.h>
78 #include <sys/ioctl.h>
79 #include <sys/file.h>
80 #include <sys/proc.h>
81 #include <sys/socketvar.h>
82 #include <sys/signalvar.h>
83 #include <sys/uio.h>
84 #include <sys/kernel.h>
85 #include <sys/stat.h>
86 #include <sys/kmem.h>
87 #include <sys/poll.h>
88 #include <sys/vnode.h>
89 #include <sys/mount.h>
90 #include <sys/syscallargs.h>
91 #include <sys/ktrace.h>
92 #include <sys/atomic.h>
93 #include <sys/disklabel.h>
94 
95 /*
96  * Read system call.
97  */
98 /* ARGSUSED */
99 int
100 sys_read(struct lwp *l, const struct sys_read_args *uap, register_t *retval)
101 {
102 	/* {
103 		syscallarg(int)		fd;
104 		syscallarg(void *)	buf;
105 		syscallarg(size_t)	nbyte;
106 	} */
107 	file_t *fp;
108 	int fd;
109 
110 	fd = SCARG(uap, fd);
111 
112 	if ((fp = fd_getfile(fd)) == NULL)
113 		return (EBADF);
114 
115 	if ((fp->f_flag & FREAD) == 0) {
116 		fd_putfile(fd);
117 		return (EBADF);
118 	}
119 
120 	/* dofileread() will unuse the descriptor for us */
121 	return (dofileread(fd, fp, SCARG(uap, buf), SCARG(uap, nbyte),
122 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
123 }
124 
125 int
126 dofileread(int fd, struct file *fp, void *buf, size_t nbyte,
127 	off_t *offset, int flags, register_t *retval)
128 {
129 	struct iovec aiov;
130 	struct uio auio;
131 	size_t cnt;
132 	int error;
133 	lwp_t *l;
134 
135 	l = curlwp;
136 
137 	aiov.iov_base = (void *)buf;
138 	aiov.iov_len = nbyte;
139 	auio.uio_iov = &aiov;
140 	auio.uio_iovcnt = 1;
141 	auio.uio_resid = nbyte;
142 	auio.uio_rw = UIO_READ;
143 	auio.uio_vmspace = l->l_proc->p_vmspace;
144 
145 	/*
146 	 * Reads return ssize_t because -1 is returned on error.  Therefore
147 	 * we must restrict the length to SSIZE_MAX to avoid garbage return
148 	 * values.
149 	 */
150 	if (auio.uio_resid > SSIZE_MAX) {
151 		error = EINVAL;
152 		goto out;
153 	}
154 
155 	cnt = auio.uio_resid;
156 	error = (*fp->f_ops->fo_read)(fp, offset, &auio, fp->f_cred, flags);
157 	if (error)
158 		if (auio.uio_resid != cnt && (error == ERESTART ||
159 		    error == EINTR || error == EWOULDBLOCK))
160 			error = 0;
161 	cnt -= auio.uio_resid;
162 	ktrgenio(fd, UIO_READ, buf, cnt, error);
163 	*retval = cnt;
164  out:
165 	fd_putfile(fd);
166 	return (error);
167 }
168 
169 /*
170  * Scatter read system call.
171  */
172 int
173 sys_readv(struct lwp *l, const struct sys_readv_args *uap, register_t *retval)
174 {
175 	/* {
176 		syscallarg(int)				fd;
177 		syscallarg(const struct iovec *)	iovp;
178 		syscallarg(int)				iovcnt;
179 	} */
180 
181 	return do_filereadv(SCARG(uap, fd), SCARG(uap, iovp),
182 	    SCARG(uap, iovcnt), NULL, FOF_UPDATE_OFFSET, retval);
183 }
184 
185 int
186 do_filereadv(int fd, const struct iovec *iovp, int iovcnt,
187     off_t *offset, int flags, register_t *retval)
188 {
189 	struct uio	auio;
190 	struct iovec	*iov, *needfree = NULL, aiov[UIO_SMALLIOV];
191 	int		i, error;
192 	size_t		cnt;
193 	u_int		iovlen;
194 	struct file	*fp;
195 	struct iovec	*ktriov = NULL;
196 
197 	if (iovcnt == 0)
198 		return EINVAL;
199 
200 	if ((fp = fd_getfile(fd)) == NULL)
201 		return EBADF;
202 
203 	if ((fp->f_flag & FREAD) == 0) {
204 		fd_putfile(fd);
205 		return EBADF;
206 	}
207 
208 	if (offset == NULL)
209 		offset = &fp->f_offset;
210 	else {
211 		/*
212 		 * Caller must not specify &fp->f_offset -- we can't
213 		 * safely dereference it for the call to fo_seek
214 		 * without holding some underlying object lock.
215 		 */
216 		KASSERT(offset != &fp->f_offset);
217 		if (fp->f_ops->fo_seek == NULL) {
218 			error = ESPIPE;
219 			goto out;
220 		}
221 		error = (*fp->f_ops->fo_seek)(fp, *offset, SEEK_SET, NULL,
222 		    0);
223 		if (error != 0)
224 			goto out;
225 	}
226 
227 	iovlen = iovcnt * sizeof(struct iovec);
228 	if (flags & FOF_IOV_SYSSPACE)
229 		iov = __UNCONST(iovp);
230 	else {
231 		iov = aiov;
232 		if ((u_int)iovcnt > UIO_SMALLIOV) {
233 			if ((u_int)iovcnt > IOV_MAX) {
234 				error = EINVAL;
235 				goto out;
236 			}
237 			iov = kmem_alloc(iovlen, KM_SLEEP);
238 			needfree = iov;
239 		}
240 		error = copyin(iovp, iov, iovlen);
241 		if (error)
242 			goto done;
243 	}
244 
245 	auio.uio_iov = iov;
246 	auio.uio_iovcnt = iovcnt;
247 	auio.uio_rw = UIO_READ;
248 	auio.uio_vmspace = curproc->p_vmspace;
249 
250 	auio.uio_resid = 0;
251 	for (i = 0; i < iovcnt; i++, iov++) {
252 		auio.uio_resid += iov->iov_len;
253 		/*
254 		 * Reads return ssize_t because -1 is returned on error.
255 		 * Therefore we must restrict the length to SSIZE_MAX to
256 		 * avoid garbage return values.
257 		 */
258 		if (iov->iov_len > SSIZE_MAX || auio.uio_resid > SSIZE_MAX) {
259 			error = EINVAL;
260 			goto done;
261 		}
262 	}
263 
264 	/*
265 	 * if tracing, save a copy of iovec
266 	 */
267 	if (ktrpoint(KTR_GENIO))  {
268 		ktriov = kmem_alloc(iovlen, KM_SLEEP);
269 		memcpy(ktriov, auio.uio_iov, iovlen);
270 	}
271 
272 	cnt = auio.uio_resid;
273 	error = (*fp->f_ops->fo_read)(fp, offset, &auio, fp->f_cred, flags);
274 	if (error)
275 		if (auio.uio_resid != cnt && (error == ERESTART ||
276 		    error == EINTR || error == EWOULDBLOCK))
277 			error = 0;
278 	cnt -= auio.uio_resid;
279 	*retval = cnt;
280 
281 	if (ktriov != NULL) {
282 		ktrgeniov(fd, UIO_READ, ktriov, cnt, error);
283 		kmem_free(ktriov, iovlen);
284 	}
285 
286  done:
287 	if (needfree)
288 		kmem_free(needfree, iovlen);
289  out:
290 	fd_putfile(fd);
291 	return (error);
292 }
293 
294 /*
295  * Write system call
296  */
297 int
298 sys_write(struct lwp *l, const struct sys_write_args *uap, register_t *retval)
299 {
300 	/* {
301 		syscallarg(int)			fd;
302 		syscallarg(const void *)	buf;
303 		syscallarg(size_t)		nbyte;
304 	} */
305 	file_t *fp;
306 	int fd;
307 
308 	fd = SCARG(uap, fd);
309 
310 	if ((fp = fd_getfile(fd)) == NULL)
311 		return (EBADF);
312 
313 	if ((fp->f_flag & FWRITE) == 0) {
314 		fd_putfile(fd);
315 		return (EBADF);
316 	}
317 
318 	/* dofilewrite() will unuse the descriptor for us */
319 	return (dofilewrite(fd, fp, SCARG(uap, buf), SCARG(uap, nbyte),
320 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
321 }
322 
323 int
324 dofilewrite(int fd, struct file *fp, const void *buf,
325 	size_t nbyte, off_t *offset, int flags, register_t *retval)
326 {
327 	struct iovec aiov;
328 	struct uio auio;
329 	size_t cnt;
330 	int error;
331 
332 	aiov.iov_base = __UNCONST(buf);		/* XXXUNCONST kills const */
333 	aiov.iov_len = nbyte;
334 	auio.uio_iov = &aiov;
335 	auio.uio_iovcnt = 1;
336 	auio.uio_resid = nbyte;
337 	auio.uio_rw = UIO_WRITE;
338 	auio.uio_vmspace = curproc->p_vmspace;
339 
340 	/*
341 	 * Writes return ssize_t because -1 is returned on error.  Therefore
342 	 * we must restrict the length to SSIZE_MAX to avoid garbage return
343 	 * values.
344 	 */
345 	if (auio.uio_resid > SSIZE_MAX) {
346 		error = EINVAL;
347 		goto out;
348 	}
349 
350 	cnt = auio.uio_resid;
351 	error = (*fp->f_ops->fo_write)(fp, offset, &auio, fp->f_cred, flags);
352 	if (error) {
353 		if (auio.uio_resid != cnt && (error == ERESTART ||
354 		    error == EINTR || error == EWOULDBLOCK))
355 			error = 0;
356 		if (error == EPIPE && !(fp->f_flag & FNOSIGPIPE)) {
357 			mutex_enter(&proc_lock);
358 			psignal(curproc, SIGPIPE);
359 			mutex_exit(&proc_lock);
360 		}
361 	}
362 	cnt -= auio.uio_resid;
363 	ktrgenio(fd, UIO_WRITE, buf, cnt, error);
364 	*retval = cnt;
365  out:
366 	fd_putfile(fd);
367 	return (error);
368 }
369 
370 /*
371  * Gather write system call
372  */
373 int
374 sys_writev(struct lwp *l, const struct sys_writev_args *uap, register_t *retval)
375 {
376 	/* {
377 		syscallarg(int)				fd;
378 		syscallarg(const struct iovec *)	iovp;
379 		syscallarg(int)				iovcnt;
380 	} */
381 
382 	return do_filewritev(SCARG(uap, fd), SCARG(uap, iovp),
383 	    SCARG(uap, iovcnt), NULL, FOF_UPDATE_OFFSET, retval);
384 }
385 
386 int
387 do_filewritev(int fd, const struct iovec *iovp, int iovcnt,
388     off_t *offset, int flags, register_t *retval)
389 {
390 	struct uio	auio;
391 	struct iovec	*iov, *needfree = NULL, aiov[UIO_SMALLIOV];
392 	int		i, error;
393 	size_t		cnt;
394 	u_int		iovlen;
395 	struct file	*fp;
396 	struct iovec	*ktriov = NULL;
397 
398 	if (iovcnt == 0)
399 		return EINVAL;
400 
401 	if ((fp = fd_getfile(fd)) == NULL)
402 		return EBADF;
403 
404 	if ((fp->f_flag & FWRITE) == 0) {
405 		fd_putfile(fd);
406 		return EBADF;
407 	}
408 
409 	if (offset == NULL)
410 		offset = &fp->f_offset;
411 	else {
412 		/*
413 		 * Caller must not specify &fp->f_offset -- we can't
414 		 * safely dereference it for the call to fo_seek
415 		 * without holding some underlying object lock.
416 		 */
417 		KASSERT(offset != &fp->f_offset);
418 		if (fp->f_ops->fo_seek == NULL) {
419 			error = ESPIPE;
420 			goto out;
421 		}
422 		error = (*fp->f_ops->fo_seek)(fp, *offset, SEEK_SET, NULL,
423 		    0);
424 		if (error != 0)
425 			goto out;
426 	}
427 
428 	iovlen = iovcnt * sizeof(struct iovec);
429 	if (flags & FOF_IOV_SYSSPACE)
430 		iov = __UNCONST(iovp);
431 	else {
432 		iov = aiov;
433 		if ((u_int)iovcnt > UIO_SMALLIOV) {
434 			if ((u_int)iovcnt > IOV_MAX) {
435 				error = EINVAL;
436 				goto out;
437 			}
438 			iov = kmem_alloc(iovlen, KM_SLEEP);
439 			needfree = iov;
440 		}
441 		error = copyin(iovp, iov, iovlen);
442 		if (error)
443 			goto done;
444 	}
445 
446 	auio.uio_iov = iov;
447 	auio.uio_iovcnt = iovcnt;
448 	auio.uio_rw = UIO_WRITE;
449 	auio.uio_vmspace = curproc->p_vmspace;
450 
451 	auio.uio_resid = 0;
452 	for (i = 0; i < iovcnt; i++, iov++) {
453 		auio.uio_resid += iov->iov_len;
454 		/*
455 		 * Writes return ssize_t because -1 is returned on error.
456 		 * Therefore we must restrict the length to SSIZE_MAX to
457 		 * avoid garbage return values.
458 		 */
459 		if (iov->iov_len > SSIZE_MAX || auio.uio_resid > SSIZE_MAX) {
460 			error = EINVAL;
461 			goto done;
462 		}
463 	}
464 
465 	/*
466 	 * if tracing, save a copy of iovec
467 	 */
468 	if (ktrpoint(KTR_GENIO))  {
469 		ktriov = kmem_alloc(iovlen, KM_SLEEP);
470 		memcpy(ktriov, auio.uio_iov, iovlen);
471 	}
472 
473 	cnt = auio.uio_resid;
474 	error = (*fp->f_ops->fo_write)(fp, offset, &auio, fp->f_cred, flags);
475 	if (error) {
476 		if (auio.uio_resid != cnt && (error == ERESTART ||
477 		    error == EINTR || error == EWOULDBLOCK))
478 			error = 0;
479 		if (error == EPIPE && !(fp->f_flag & FNOSIGPIPE)) {
480 			mutex_enter(&proc_lock);
481 			psignal(curproc, SIGPIPE);
482 			mutex_exit(&proc_lock);
483 		}
484 	}
485 	cnt -= auio.uio_resid;
486 	*retval = cnt;
487 
488 	if (ktriov != NULL) {
489 		ktrgeniov(fd, UIO_WRITE, ktriov, cnt, error);
490 		kmem_free(ktriov, iovlen);
491 	}
492 
493  done:
494 	if (needfree)
495 		kmem_free(needfree, iovlen);
496  out:
497 	fd_putfile(fd);
498 	return (error);
499 }
500 
501 /*
502  * Ioctl system call
503  */
504 /* ARGSUSED */
505 int
506 sys_ioctl(struct lwp *l, const struct sys_ioctl_args *uap, register_t *retval)
507 {
508 	/* {
509 		syscallarg(int)		fd;
510 		syscallarg(u_long)	com;
511 		syscallarg(void *)	data;
512 	} */
513 	struct file	*fp;
514 	proc_t		*p;
515 	u_long		com;
516 	int		error;
517 	size_t		size, alloc_size;
518 	void 		*data, *memp;
519 #define	STK_PARAMS	128
520 	u_long		stkbuf[STK_PARAMS/sizeof(u_long)];
521 #if  __TMPBIGMAXPARTITIONS > MAXPARTITIONS
522 	size_t		zero_last = 0;
523 #define	zero_size(SZ)	((SZ)+zero_last)
524 #else
525 #define	zero_size(SZ)	(SZ)
526 #endif
527 
528 	memp = NULL;
529 	alloc_size = 0;
530 	error = 0;
531 	p = l->l_proc;
532 
533 	if ((fp = fd_getfile(SCARG(uap, fd))) == NULL)
534 		return (EBADF);
535 
536 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
537 		error = EBADF;
538 		com = 0;
539 		goto out;
540 	}
541 
542 	switch (com = SCARG(uap, com)) {
543 	case FIONCLEX:
544 	case FIOCLEX:
545 		fd_set_exclose(l, SCARG(uap, fd), com == FIOCLEX);
546 		goto out;
547 	}
548 
549 	/*
550 	 * Interpret high order word to find amount of data to be
551 	 * copied to/from the user's address space.
552 	 */
553 	size = IOCPARM_LEN(com);
554 	alloc_size = size;
555 
556 	/*
557 	 * The disklabel is now padded to a multiple of 8 bytes however the old
558 	 * disklabel on 32bit platforms wasn't.  This leaves a difference in
559 	 * size of 4 bytes between the two but are otherwise identical.
560 	 * To deal with this, we allocate enough space for the new disklabel
561 	 * but only copyin/out the smaller amount.
562 	 */
563 	if (IOCGROUP(com) == 'd') {
564 #if  __TMPBIGMAXPARTITIONS > MAXPARTITIONS
565 		u_long ocom = com;
566 #endif
567 		u_long ncom = com ^ (DIOCGDINFO ^ DIOCGDINFO32);
568 
569 #if  __TMPBIGMAXPARTITIONS > MAXPARTITIONS
570 	/*
571 	 * Userland might use struct disklabel that is bigger than the
572 	 * the kernel version (historic accident) - alloc userland
573 	 * size and zero unused part on copyout.
574 	 */
575 #define	DISKLABELLENDIFF	(sizeof(struct partition)	\
576 				       *(__TMPBIGMAXPARTITIONS-MAXPARTITIONS))
577 #define	IOCFIXUP(NIOC)	((NIOC&~(IOCPARM_MASK<<IOCPARM_SHIFT))	| \
578 			   (IOCPARM_LEN(NIOC)-DISKLABELLENDIFF)<<IOCPARM_SHIFT)
579 
580 		switch (IOCFIXUP(ocom)) {
581 		case DIOCGDINFO:
582 		case DIOCWDINFO:
583 		case DIOCSDINFO:
584 		case DIOCGDEFLABEL:
585 			com = ncom = IOCFIXUP(ocom);
586 			zero_last = DISKLABELLENDIFF;
587 			size -= DISKLABELLENDIFF;
588 			goto done;
589 		}
590 #endif
591 
592 		switch (ncom) {
593 		case DIOCGDINFO:
594 		case DIOCWDINFO:
595 		case DIOCSDINFO:
596 		case DIOCGDEFLABEL:
597 			com = ncom;
598 			if (IOCPARM_LEN(DIOCGDINFO32) < IOCPARM_LEN(DIOCGDINFO))
599 				alloc_size = IOCPARM_LEN(DIOCGDINFO);
600 			break;
601 		}
602 #if  __TMPBIGMAXPARTITIONS > MAXPARTITIONS
603 		done: ;
604 #endif
605 	}
606 	if (size > IOCPARM_MAX) {
607 		error = ENOTTY;
608 		goto out;
609 	}
610 	memp = NULL;
611 	if ((com >> IOCPARM_SHIFT) == 0)  {
612 		/* UNIX-style ioctl. */
613 		data = SCARG(uap, data);
614 	} else {
615 		if (alloc_size > sizeof(stkbuf)) {
616 			memp = kmem_alloc(alloc_size, KM_SLEEP);
617 			data = memp;
618 		} else {
619 			data = (void *)stkbuf;
620 		}
621 		if (com&IOC_IN) {
622 			if (size) {
623 				error = copyin(SCARG(uap, data), data, size);
624 				if (error) {
625 					goto out;
626 				}
627 				/*
628 				 * The data between size and alloc_size has
629 				 * not been overwritten.  It shouldn't matter
630 				 * but let's clear that anyway.
631 				 */
632 				if (__predict_false(size < alloc_size)) {
633 					memset((char *)data+size, 0,
634 					    alloc_size - size);
635 				}
636 				ktrgenio(SCARG(uap, fd), UIO_WRITE,
637 				    SCARG(uap, data), size, 0);
638 			} else {
639 				*(void **)data = SCARG(uap, data);
640 			}
641 		} else if ((com&IOC_OUT) && size) {
642 			/*
643 			 * Zero the buffer so the user always
644 			 * gets back something deterministic.
645 			 */
646 			memset(data, 0, zero_size(size));
647 		} else if (com&IOC_VOID) {
648 			*(void **)data = SCARG(uap, data);
649 		}
650 	}
651 
652 	switch (com) {
653 
654 	case FIONBIO:
655 		/* XXX Code block is not atomic */
656 		if (*(int *)data != 0)
657 			atomic_or_uint(&fp->f_flag, FNONBLOCK);
658 		else
659 			atomic_and_uint(&fp->f_flag, ~FNONBLOCK);
660 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, data);
661 		break;
662 
663 	case FIOASYNC:
664 		/* XXX Code block is not atomic */
665 		if (*(int *)data != 0)
666 			atomic_or_uint(&fp->f_flag, FASYNC);
667 		else
668 			atomic_and_uint(&fp->f_flag, ~FASYNC);
669 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, data);
670 		break;
671 
672 	default:
673 		error = (*fp->f_ops->fo_ioctl)(fp, com, data);
674 		/*
675 		 * Copy any data to user, size was
676 		 * already set and checked above.
677 		 */
678 		if (error == 0 && (com&IOC_OUT) && size) {
679 			error = copyout(data, SCARG(uap, data),
680 			    zero_size(size));
681 			ktrgenio(SCARG(uap, fd), UIO_READ, SCARG(uap, data),
682 			    size, error);
683 		}
684 		break;
685 	}
686  out:
687 	if (memp)
688 		kmem_free(memp, alloc_size);
689 	fd_putfile(SCARG(uap, fd));
690 	switch (error) {
691 	case -1:
692 		printf("sys_ioctl: _IO%s%s('%c', %lu, %lu) returned -1: "
693 		    "pid=%d comm=%s\n",
694 		    (com & IOC_IN) ? "W" : "", (com & IOC_OUT) ? "R" : "",
695 		    (char)IOCGROUP(com), (com & 0xff), IOCPARM_LEN(com),
696 		    p->p_pid, p->p_comm);
697 		/* FALLTHROUGH */
698 	case EPASSTHROUGH:
699 		error = ENOTTY;
700 		/* FALLTHROUGH */
701 	default:
702 		return (error);
703 	}
704 }
705