xref: /freebsd/sys/kern/sys_generic.c (revision 6e8394b8baa7d5d9153ab90de6824bcd19b3b4e1)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)sys_generic.c	8.5 (Berkeley) 1/21/94
39  * $Id: sys_generic.c,v 1.46 1999/03/27 21:16:33 alc Exp $
40  */
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/filedesc.h>
48 #include <sys/filio.h>
49 #include <sys/ttycom.h>
50 #include <sys/fcntl.h>
51 #include <sys/file.h>
52 #include <sys/proc.h>
53 #include <sys/signalvar.h>
54 #include <sys/socketvar.h>
55 #include <sys/uio.h>
56 #include <sys/kernel.h>
57 #include <sys/malloc.h>
58 #include <sys/poll.h>
59 #include <sys/sysent.h>
60 #ifdef KTRACE
61 #include <sys/ktrace.h>
62 #endif
63 
64 #include <machine/limits.h>
65 
66 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
67 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
68 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
69 
70 static int	pollscan __P((struct proc *, struct pollfd *, int));
71 static int	selscan __P((struct proc *, fd_mask **, fd_mask **, int));
72 static struct file* getfp __P((struct filedesc *, int, int));
73 static int	dofileread __P((struct proc *, struct file *, int, void *,
74 		    size_t, off_t, int));
75 static int	dofilewrite __P((struct proc *, struct file *, int,
76 		    const void *, size_t, off_t, int));
77 
78 static struct file*
79 getfp(fdp, fd, flag)
80 	struct filedesc* fdp;
81 	int fd, flag;
82 {
83 	struct file* fp;
84 
85 	if (((u_int)fd) >= fdp->fd_nfiles ||
86 	    (fp = fdp->fd_ofiles[fd]) == NULL ||
87 	    (fp->f_flag & flag) == 0)
88 		return (NULL);
89 	return (fp);
90 }
91 
92 /*
93  * Read system call.
94  */
95 #ifndef _SYS_SYSPROTO_H_
96 struct read_args {
97 	int	fd;
98 	void	*buf;
99 	size_t	nbyte;
100 };
101 #endif
102 int
103 read(p, uap)
104 	struct proc *p;
105 	register struct read_args *uap;
106 {
107 	register struct file *fp;
108 
109 	if ((fp = getfp(p->p_fd, uap->fd, FREAD)) == NULL)
110 		return (EBADF);
111 	return (dofileread(p, fp, uap->fd, uap->buf, uap->nbyte, (off_t)-1, 0));
112 }
113 
114 /*
115  * Pread system call
116  */
117 #ifndef _SYS_SYSPROTO_H_
118 struct pread_args {
119 	int	fd;
120 	void	*buf;
121 	size_t	nbyte;
122 	int	pad;
123 	off_t	offset;
124 };
125 #endif
126 int
127 pread(p, uap)
128 	struct proc *p;
129 	register struct pread_args *uap;
130 {
131 	register struct file *fp;
132 
133 	if ((fp = getfp(p->p_fd, uap->fd, FREAD)) == NULL)
134 		return (EBADF);
135 	if (fp->f_type != DTYPE_VNODE)
136 		return (ESPIPE);
137 	return (dofileread(p, fp, uap->fd, uap->buf, uap->nbyte, uap->offset,
138 	    FOF_OFFSET));
139 }
140 
141 /*
142  * Code common for read and pread
143  */
144 int
145 dofileread(p, fp, fd, buf, nbyte, offset, flags)
146 	struct proc *p;
147 	struct file *fp;
148 	int fd, flags;
149 	void *buf;
150 	size_t nbyte;
151 	off_t offset;
152 {
153 	struct uio auio;
154 	struct iovec aiov;
155 	long cnt, error = 0;
156 #ifdef KTRACE
157 	struct iovec ktriov;
158 #endif
159 
160 	aiov.iov_base = (caddr_t)buf;
161 	aiov.iov_len = nbyte;
162 	auio.uio_iov = &aiov;
163 	auio.uio_iovcnt = 1;
164 	auio.uio_offset = offset;
165 	if (nbyte > INT_MAX)
166 		return (EINVAL);
167 	auio.uio_resid = nbyte;
168 	auio.uio_rw = UIO_READ;
169 	auio.uio_segflg = UIO_USERSPACE;
170 	auio.uio_procp = p;
171 #ifdef KTRACE
172 	/*
173 	 * if tracing, save a copy of iovec
174 	 */
175 	if (KTRPOINT(p, KTR_GENIO))
176 		ktriov = aiov;
177 #endif
178 	cnt = nbyte;
179 	if ((error = (*fp->f_ops->fo_read)(fp, &auio, fp->f_cred, flags)))
180 		if (auio.uio_resid != cnt && (error == ERESTART ||
181 		    error == EINTR || error == EWOULDBLOCK))
182 			error = 0;
183 	cnt -= auio.uio_resid;
184 #ifdef KTRACE
185 	if (KTRPOINT(p, KTR_GENIO) && error == 0)
186 		ktrgenio(p->p_tracep, fd, UIO_READ, &ktriov, cnt, error);
187 #endif
188 	p->p_retval[0] = cnt;
189 	return (error);
190 }
191 
192 /*
193  * Scatter read system call.
194  */
195 #ifndef _SYS_SYSPROTO_H_
196 struct readv_args {
197 	int	fd;
198 	struct	iovec *iovp;
199 	u_int	iovcnt;
200 };
201 #endif
202 int
203 readv(p, uap)
204 	struct proc *p;
205 	register struct readv_args *uap;
206 {
207 	register struct file *fp;
208 	register struct filedesc *fdp = p->p_fd;
209 	struct uio auio;
210 	register struct iovec *iov;
211 	struct iovec *needfree;
212 	struct iovec aiov[UIO_SMALLIOV];
213 	long i, cnt, error = 0;
214 	u_int iovlen;
215 #ifdef KTRACE
216 	struct iovec *ktriov = NULL;
217 #endif
218 
219 	if ((fp = getfp(fdp, uap->fd, FREAD)) == NULL)
220 		return (EBADF);
221 	/* note: can't use iovlen until iovcnt is validated */
222 	iovlen = uap->iovcnt * sizeof (struct iovec);
223 	if (uap->iovcnt > UIO_SMALLIOV) {
224 		if (uap->iovcnt > UIO_MAXIOV)
225 			return (EINVAL);
226 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
227 		needfree = iov;
228 	} else {
229 		iov = aiov;
230 		needfree = NULL;
231 	}
232 	auio.uio_iov = iov;
233 	auio.uio_iovcnt = uap->iovcnt;
234 	auio.uio_rw = UIO_READ;
235 	auio.uio_segflg = UIO_USERSPACE;
236 	auio.uio_procp = p;
237 	auio.uio_offset = -1;
238 	if ((error = copyin((caddr_t)uap->iovp, (caddr_t)iov, iovlen)))
239 		goto done;
240 	auio.uio_resid = 0;
241 	for (i = 0; i < uap->iovcnt; i++) {
242 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
243 			error = EINVAL;
244 			goto done;
245 		}
246 		auio.uio_resid += iov->iov_len;
247 		iov++;
248 	}
249 #ifdef KTRACE
250 	/*
251 	 * if tracing, save a copy of iovec
252 	 */
253 	if (KTRPOINT(p, KTR_GENIO))  {
254 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
255 		bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen);
256 	}
257 #endif
258 	cnt = auio.uio_resid;
259 	if ((error = (*fp->f_ops->fo_read)(fp, &auio, fp->f_cred, 0)))
260 		if (auio.uio_resid != cnt && (error == ERESTART ||
261 		    error == EINTR || error == EWOULDBLOCK))
262 			error = 0;
263 	cnt -= auio.uio_resid;
264 #ifdef KTRACE
265 	if (ktriov != NULL) {
266 		if (error == 0)
267 			ktrgenio(p->p_tracep, uap->fd, UIO_READ, ktriov,
268 			    cnt, error);
269 		FREE(ktriov, M_TEMP);
270 	}
271 #endif
272 	p->p_retval[0] = cnt;
273 done:
274 	if (needfree)
275 		FREE(needfree, M_IOV);
276 	return (error);
277 }
278 
279 /*
280  * Write system call
281  */
282 #ifndef _SYS_SYSPROTO_H_
283 struct write_args {
284 	int	fd;
285 	const void *buf;
286 	size_t	nbyte;
287 };
288 #endif
289 int
290 write(p, uap)
291 	struct proc *p;
292 	register struct write_args *uap;
293 {
294 	register struct file *fp;
295 
296 	if ((fp = getfp(p->p_fd, uap->fd, FWRITE)) == NULL)
297 		return (EBADF);
298 	return (dofilewrite(p, fp, uap->fd, uap->buf, uap->nbyte, (off_t)-1, 0));
299 }
300 
301 /*
302  * Pwrite system call
303  */
304 #ifndef _SYS_SYSPROTO_H_
305 struct pwrite_args {
306 	int	fd;
307 	const void *buf;
308 	size_t	nbyte;
309 	int	pad;
310 	off_t	offset;
311 };
312 #endif
313 int
314 pwrite(p, uap)
315 	struct proc *p;
316 	register struct pwrite_args *uap;
317 {
318 	register struct file *fp;
319 
320 	if ((fp = getfp(p->p_fd, uap->fd, FWRITE)) == NULL)
321 		return (EBADF);
322 	if (fp->f_type != DTYPE_VNODE)
323 		return (ESPIPE);
324 	return (dofilewrite(p, fp, uap->fd, uap->buf, uap->nbyte, uap->offset,
325 	    FOF_OFFSET));
326 }
327 
328 static int
329 dofilewrite(p, fp, fd, buf, nbyte, offset, flags)
330 	struct proc *p;
331 	struct file *fp;
332 	int fd, flags;
333 	const void *buf;
334 	size_t nbyte;
335 	off_t offset;
336 {
337 	struct uio auio;
338 	struct iovec aiov;
339 	long cnt, error = 0;
340 #ifdef KTRACE
341 	struct iovec ktriov;
342 #endif
343 
344 	aiov.iov_base = (void *)buf;
345 	aiov.iov_len = nbyte;
346 	auio.uio_iov = &aiov;
347 	auio.uio_iovcnt = 1;
348 	auio.uio_offset = offset;
349 	if (nbyte > INT_MAX)
350 		return (EINVAL);
351 	auio.uio_resid = nbyte;
352 	auio.uio_rw = UIO_WRITE;
353 	auio.uio_segflg = UIO_USERSPACE;
354 	auio.uio_procp = p;
355 #ifdef KTRACE
356 	/*
357 	 * if tracing, save a copy of iovec
358 	 */
359 	if (KTRPOINT(p, KTR_GENIO))
360 		ktriov = aiov;
361 #endif
362 	cnt = nbyte;
363 	if ((error = (*fp->f_ops->fo_write)(fp, &auio, fp->f_cred, flags))) {
364 		if (auio.uio_resid != cnt && (error == ERESTART ||
365 		    error == EINTR || error == EWOULDBLOCK))
366 			error = 0;
367 		if (error == EPIPE)
368 			psignal(p, SIGPIPE);
369 	}
370 	cnt -= auio.uio_resid;
371 #ifdef KTRACE
372 	if (KTRPOINT(p, KTR_GENIO) && error == 0)
373 		ktrgenio(p->p_tracep, fd, UIO_WRITE,
374 		    &ktriov, cnt, error);
375 #endif
376 	p->p_retval[0] = cnt;
377 	return (error);
378 }
379 
380 /*
381  * Gather write system call
382  */
383 #ifndef _SYS_SYSPROTO_H_
384 struct writev_args {
385 	int	fd;
386 	struct	iovec *iovp;
387 	u_int	iovcnt;
388 };
389 #endif
390 int
391 writev(p, uap)
392 	struct proc *p;
393 	register struct writev_args *uap;
394 {
395 	register struct file *fp;
396 	register struct filedesc *fdp = p->p_fd;
397 	struct uio auio;
398 	register struct iovec *iov;
399 	struct iovec *needfree;
400 	struct iovec aiov[UIO_SMALLIOV];
401 	long i, cnt, error = 0;
402 	u_int iovlen;
403 #ifdef KTRACE
404 	struct iovec *ktriov = NULL;
405 #endif
406 
407 	if ((fp = getfp(fdp, uap->fd, FWRITE)) == NULL)
408 		return (EBADF);
409 	/* note: can't use iovlen until iovcnt is validated */
410 	iovlen = uap->iovcnt * sizeof (struct iovec);
411 	if (uap->iovcnt > UIO_SMALLIOV) {
412 		if (uap->iovcnt > UIO_MAXIOV)
413 			return (EINVAL);
414 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
415 		needfree = iov;
416 	} else {
417 		iov = aiov;
418 		needfree = NULL;
419 	}
420 	auio.uio_iov = iov;
421 	auio.uio_iovcnt = uap->iovcnt;
422 	auio.uio_rw = UIO_WRITE;
423 	auio.uio_segflg = UIO_USERSPACE;
424 	auio.uio_procp = p;
425 	auio.uio_offset = -1;
426 	if ((error = copyin((caddr_t)uap->iovp, (caddr_t)iov, iovlen)))
427 		goto done;
428 	auio.uio_resid = 0;
429 	for (i = 0; i < uap->iovcnt; i++) {
430 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
431 			error = EINVAL;
432 			goto done;
433 		}
434 		auio.uio_resid += iov->iov_len;
435 		iov++;
436 	}
437 #ifdef KTRACE
438 	/*
439 	 * if tracing, save a copy of iovec
440 	 */
441 	if (KTRPOINT(p, KTR_GENIO))  {
442 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
443 		bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen);
444 	}
445 #endif
446 	cnt = auio.uio_resid;
447 	if ((error = (*fp->f_ops->fo_write)(fp, &auio, fp->f_cred, 0))) {
448 		if (auio.uio_resid != cnt && (error == ERESTART ||
449 		    error == EINTR || error == EWOULDBLOCK))
450 			error = 0;
451 		if (error == EPIPE)
452 			psignal(p, SIGPIPE);
453 	}
454 	cnt -= auio.uio_resid;
455 #ifdef KTRACE
456 	if (ktriov != NULL) {
457 		if (error == 0)
458 			ktrgenio(p->p_tracep, uap->fd, UIO_WRITE,
459 				ktriov, cnt, error);
460 		FREE(ktriov, M_TEMP);
461 	}
462 #endif
463 	p->p_retval[0] = cnt;
464 done:
465 	if (needfree)
466 		FREE(needfree, M_IOV);
467 	return (error);
468 }
469 
470 /*
471  * Ioctl system call
472  */
473 #ifndef _SYS_SYSPROTO_H_
474 struct ioctl_args {
475 	int	fd;
476 	u_long	com;
477 	caddr_t	data;
478 };
479 #endif
480 /* ARGSUSED */
481 int
482 ioctl(p, uap)
483 	struct proc *p;
484 	register struct ioctl_args *uap;
485 {
486 	register struct file *fp;
487 	register struct filedesc *fdp;
488 	register u_long com;
489 	int error;
490 	register u_int size;
491 	caddr_t data, memp;
492 	int tmp;
493 #define STK_PARAMS	128
494 	char stkbuf[STK_PARAMS];
495 
496 	fdp = p->p_fd;
497 	if ((u_int)uap->fd >= fdp->fd_nfiles ||
498 	    (fp = fdp->fd_ofiles[uap->fd]) == NULL)
499 		return (EBADF);
500 
501 	if ((fp->f_flag & (FREAD | FWRITE)) == 0)
502 		return (EBADF);
503 
504 	switch (com = uap->com) {
505 	case FIONCLEX:
506 		fdp->fd_ofileflags[uap->fd] &= ~UF_EXCLOSE;
507 		return (0);
508 	case FIOCLEX:
509 		fdp->fd_ofileflags[uap->fd] |= UF_EXCLOSE;
510 		return (0);
511 	}
512 
513 	/*
514 	 * Interpret high order word to find amount of data to be
515 	 * copied to/from the user's address space.
516 	 */
517 	size = IOCPARM_LEN(com);
518 	if (size > IOCPARM_MAX)
519 		return (ENOTTY);
520 	memp = NULL;
521 	if (size > sizeof (stkbuf)) {
522 		memp = (caddr_t)malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
523 		data = memp;
524 	} else
525 		data = stkbuf;
526 	if (com&IOC_IN) {
527 		if (size) {
528 			error = copyin(uap->data, data, (u_int)size);
529 			if (error) {
530 				if (memp)
531 					free(memp, M_IOCTLOPS);
532 				return (error);
533 			}
534 		} else
535 			*(caddr_t *)data = uap->data;
536 	} else if ((com&IOC_OUT) && size)
537 		/*
538 		 * Zero the buffer so the user always
539 		 * gets back something deterministic.
540 		 */
541 		bzero(data, size);
542 	else if (com&IOC_VOID)
543 		*(caddr_t *)data = uap->data;
544 
545 	switch (com) {
546 
547 	case FIONBIO:
548 		if ((tmp = *(int *)data))
549 			fp->f_flag |= FNONBLOCK;
550 		else
551 			fp->f_flag &= ~FNONBLOCK;
552 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
553 		break;
554 
555 	case FIOASYNC:
556 		if ((tmp = *(int *)data))
557 			fp->f_flag |= FASYNC;
558 		else
559 			fp->f_flag &= ~FASYNC;
560 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p);
561 		break;
562 
563 	default:
564 		error = (*fp->f_ops->fo_ioctl)(fp, com, data, p);
565 		/*
566 		 * Copy any data to user, size was
567 		 * already set and checked above.
568 		 */
569 		if (error == 0 && (com&IOC_OUT) && size)
570 			error = copyout(data, uap->data, (u_int)size);
571 		break;
572 	}
573 	if (memp)
574 		free(memp, M_IOCTLOPS);
575 	return (error);
576 }
577 
578 static int	nselcoll;
579 int	selwait;
580 
581 /*
582  * Select system call.
583  */
584 #ifndef _SYS_SYSPROTO_H_
585 struct select_args {
586 	int	nd;
587 	fd_set	*in, *ou, *ex;
588 	struct	timeval *tv;
589 };
590 #endif
591 int
592 select(p, uap)
593 	register struct proc *p;
594 	register struct select_args *uap;
595 {
596 	/*
597 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
598 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
599 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
600 	 * of 256.
601 	 */
602 	fd_mask s_selbits[howmany(2048, NFDBITS)];
603 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
604 	struct timeval atv, rtv, ttv;
605 	int s, ncoll, error, timo;
606 	u_int nbufbytes, ncpbytes, nfdbits;
607 
608 	if (uap->nd < 0)
609 		return (EINVAL);
610 	if (uap->nd > p->p_fd->fd_nfiles)
611 		uap->nd = p->p_fd->fd_nfiles;   /* forgiving; slightly wrong */
612 
613 	/*
614 	 * Allocate just enough bits for the non-null fd_sets.  Use the
615 	 * preallocated auto buffer if possible.
616 	 */
617 	nfdbits = roundup(uap->nd, NFDBITS);
618 	ncpbytes = nfdbits / NBBY;
619 	nbufbytes = 0;
620 	if (uap->in != NULL)
621 		nbufbytes += 2 * ncpbytes;
622 	if (uap->ou != NULL)
623 		nbufbytes += 2 * ncpbytes;
624 	if (uap->ex != NULL)
625 		nbufbytes += 2 * ncpbytes;
626 	if (nbufbytes <= sizeof s_selbits)
627 		selbits = &s_selbits[0];
628 	else
629 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
630 
631 	/*
632 	 * Assign pointers into the bit buffers and fetch the input bits.
633 	 * Put the output buffers together so that they can be bzeroed
634 	 * together.
635 	 */
636 	sbp = selbits;
637 #define	getbits(name, x) \
638 	do {								\
639 		if (uap->name == NULL)					\
640 			ibits[x] = NULL;				\
641 		else {							\
642 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
643 			obits[x] = sbp;					\
644 			sbp += ncpbytes / sizeof *sbp;			\
645 			error = copyin(uap->name, ibits[x], ncpbytes);	\
646 			if (error != 0)					\
647 				goto done;				\
648 		}							\
649 	} while (0)
650 	getbits(in, 0);
651 	getbits(ou, 1);
652 	getbits(ex, 2);
653 #undef	getbits
654 	if (nbufbytes != 0)
655 		bzero(selbits, nbufbytes / 2);
656 
657 	if (uap->tv) {
658 		error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
659 			sizeof (atv));
660 		if (error)
661 			goto done;
662 		if (itimerfix(&atv)) {
663 			error = EINVAL;
664 			goto done;
665 		}
666 		getmicrouptime(&rtv);
667 		timevaladd(&atv, &rtv);
668 	} else
669 		atv.tv_sec = 0;
670 	timo = 0;
671 retry:
672 	ncoll = nselcoll;
673 	p->p_flag |= P_SELECT;
674 	error = selscan(p, ibits, obits, uap->nd);
675 	if (error || p->p_retval[0])
676 		goto done;
677 	if (atv.tv_sec) {
678 		getmicrouptime(&rtv);
679 		if (timevalcmp(&rtv, &atv, >=))
680 			goto done;
681 		ttv = atv;
682 		timevalsub(&ttv, &rtv);
683 		timo = ttv.tv_sec > 24 * 60 * 60 ?
684 		    24 * 60 * 60 * hz : tvtohz(&ttv);
685 	}
686 	s = splhigh();
687 	if ((p->p_flag & P_SELECT) == 0 || nselcoll != ncoll) {
688 		splx(s);
689 		goto retry;
690 	}
691 	p->p_flag &= ~P_SELECT;
692 	error = tsleep((caddr_t)&selwait, PSOCK | PCATCH, "select", timo);
693 	splx(s);
694 	if (error == 0)
695 		goto retry;
696 done:
697 	p->p_flag &= ~P_SELECT;
698 	/* select is not restarted after signals... */
699 	if (error == ERESTART)
700 		error = EINTR;
701 	if (error == EWOULDBLOCK)
702 		error = 0;
703 #define	putbits(name, x) \
704 	if (uap->name && (error2 = copyout(obits[x], uap->name, ncpbytes))) \
705 		error = error2;
706 	if (error == 0) {
707 		int error2;
708 
709 		putbits(in, 0);
710 		putbits(ou, 1);
711 		putbits(ex, 2);
712 #undef putbits
713 	}
714 	if (selbits != &s_selbits[0])
715 		free(selbits, M_SELECT);
716 	return (error);
717 }
718 
719 static int
720 selscan(p, ibits, obits, nfd)
721 	struct proc *p;
722 	fd_mask **ibits, **obits;
723 	int nfd;
724 {
725 	register struct filedesc *fdp = p->p_fd;
726 	register int msk, i, j, fd;
727 	register fd_mask bits;
728 	struct file *fp;
729 	int n = 0;
730 	/* Note: backend also returns POLLHUP/POLLERR if appropriate. */
731 	static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND };
732 
733 	for (msk = 0; msk < 3; msk++) {
734 		if (ibits[msk] == NULL)
735 			continue;
736 		for (i = 0; i < nfd; i += NFDBITS) {
737 			bits = ibits[msk][i/NFDBITS];
738 			while ((j = ffs(bits)) && (fd = i + --j) < nfd) {
739 				bits &= ~(1 << j);
740 				fp = fdp->fd_ofiles[fd];
741 				if (fp == NULL)
742 					return (EBADF);
743 				if ((*fp->f_ops->fo_poll)(fp, flag[msk],
744 				    fp->f_cred, p)) {
745 					obits[msk][(fd)/NFDBITS] |=
746 						(1 << ((fd) % NFDBITS));
747 					n++;
748 				}
749 			}
750 		}
751 	}
752 	p->p_retval[0] = n;
753 	return (0);
754 }
755 
756 /*
757  * Poll system call.
758  */
759 #ifndef _SYS_SYSPROTO_H_
760 struct poll_args {
761 	struct pollfd *fds;
762 	u_int	nfds;
763 	int	timeout;
764 };
765 #endif
766 int
767 poll(p, uap)
768 	register struct proc *p;
769 	register struct poll_args *uap;
770 {
771 	caddr_t bits;
772 	char smallbits[32 * sizeof(struct pollfd)];
773 	struct timeval atv, rtv, ttv;
774 	int s, ncoll, error = 0, timo;
775 	size_t ni;
776 
777 	if (SCARG(uap, nfds) > p->p_fd->fd_nfiles) {
778 		/* forgiving; slightly wrong */
779 		SCARG(uap, nfds) = p->p_fd->fd_nfiles;
780 	}
781 	ni = SCARG(uap, nfds) * sizeof(struct pollfd);
782 	if (ni > sizeof(smallbits))
783 		bits = malloc(ni, M_TEMP, M_WAITOK);
784 	else
785 		bits = smallbits;
786 	error = copyin(SCARG(uap, fds), bits, ni);
787 	if (error)
788 		goto done;
789 	if (SCARG(uap, timeout) != INFTIM) {
790 		atv.tv_sec = SCARG(uap, timeout) / 1000;
791 		atv.tv_usec = (SCARG(uap, timeout) % 1000) * 1000;
792 		if (itimerfix(&atv)) {
793 			error = EINVAL;
794 			goto done;
795 		}
796 		getmicrouptime(&rtv);
797 		timevaladd(&atv, &rtv);
798 	} else
799 		atv.tv_sec = 0;
800 	timo = 0;
801 retry:
802 	ncoll = nselcoll;
803 	p->p_flag |= P_SELECT;
804 	error = pollscan(p, (struct pollfd *)bits, SCARG(uap, nfds));
805 	if (error || p->p_retval[0])
806 		goto done;
807 	if (atv.tv_sec) {
808 		getmicrouptime(&rtv);
809 		if (timevalcmp(&rtv, &atv, >=))
810 			goto done;
811 		ttv = atv;
812 		timevalsub(&ttv, &rtv);
813 		timo = ttv.tv_sec > 24 * 60 * 60 ?
814 		    24 * 60 * 60 * hz : tvtohz(&ttv);
815 	}
816 	s = splhigh();
817 	if ((p->p_flag & P_SELECT) == 0 || nselcoll != ncoll) {
818 		splx(s);
819 		goto retry;
820 	}
821 	p->p_flag &= ~P_SELECT;
822 	error = tsleep((caddr_t)&selwait, PSOCK | PCATCH, "poll", timo);
823 	splx(s);
824 	if (error == 0)
825 		goto retry;
826 done:
827 	p->p_flag &= ~P_SELECT;
828 	/* poll is not restarted after signals... */
829 	if (error == ERESTART)
830 		error = EINTR;
831 	if (error == EWOULDBLOCK)
832 		error = 0;
833 	if (error == 0) {
834 		error = copyout(bits, SCARG(uap, fds), ni);
835 		if (error)
836 			goto out;
837 	}
838 out:
839 	if (ni > sizeof(smallbits))
840 		free(bits, M_TEMP);
841 	return (error);
842 }
843 
844 static int
845 pollscan(p, fds, nfd)
846 	struct proc *p;
847 	struct pollfd *fds;
848 	int nfd;
849 {
850 	register struct filedesc *fdp = p->p_fd;
851 	int i;
852 	struct file *fp;
853 	int n = 0;
854 
855 	for (i = 0; i < nfd; i++, fds++) {
856 		if (fds->fd >= fdp->fd_nfiles) {
857 			fds->revents = POLLNVAL;
858 			n++;
859 		} else if (fds->fd < 0) {
860 			fds->revents = 0;
861 		} else {
862 			fp = fdp->fd_ofiles[fds->fd];
863 			if (fp == 0) {
864 				fds->revents = POLLNVAL;
865 				n++;
866 			} else {
867 				/*
868 				 * Note: backend also returns POLLHUP and
869 				 * POLLERR if appropriate.
870 				 */
871 				fds->revents = (*fp->f_ops->fo_poll)(fp,
872 				    fds->events, fp->f_cred, p);
873 				if (fds->revents != 0)
874 					n++;
875 			}
876 		}
877 	}
878 	p->p_retval[0] = n;
879 	return (0);
880 }
881 
882 /*
883  * OpenBSD poll system call.
884  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
885  */
886 #ifndef _SYS_SYSPROTO_H_
887 struct openbsd_poll_args {
888 	struct pollfd *fds;
889 	u_int	nfds;
890 	int	timeout;
891 };
892 #endif
893 int
894 openbsd_poll(p, uap)
895 	register struct proc *p;
896 	register struct openbsd_poll_args *uap;
897 {
898 	return (poll(p, (struct poll_args *)uap));
899 }
900 
901 /*ARGSUSED*/
902 int
903 seltrue(dev, events, p)
904 	dev_t dev;
905 	int events;
906 	struct proc *p;
907 {
908 
909 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
910 }
911 
912 /*
913  * Record a select request.
914  */
915 void
916 selrecord(selector, sip)
917 	struct proc *selector;
918 	struct selinfo *sip;
919 {
920 	struct proc *p;
921 	pid_t mypid;
922 
923 	mypid = selector->p_pid;
924 	if (sip->si_pid == mypid)
925 		return;
926 	if (sip->si_pid && (p = pfind(sip->si_pid)) &&
927 	    p->p_wchan == (caddr_t)&selwait)
928 		sip->si_flags |= SI_COLL;
929 	else
930 		sip->si_pid = mypid;
931 }
932 
933 /*
934  * Do a wakeup when a selectable event occurs.
935  */
936 void
937 selwakeup(sip)
938 	register struct selinfo *sip;
939 {
940 	register struct proc *p;
941 	int s;
942 
943 	if (sip->si_pid == 0)
944 		return;
945 	if (sip->si_flags & SI_COLL) {
946 		nselcoll++;
947 		sip->si_flags &= ~SI_COLL;
948 		wakeup((caddr_t)&selwait);
949 	}
950 	p = pfind(sip->si_pid);
951 	sip->si_pid = 0;
952 	if (p != NULL) {
953 		s = splhigh();
954 		if (p->p_wchan == (caddr_t)&selwait) {
955 			if (p->p_stat == SSLEEP)
956 				setrunnable(p);
957 			else
958 				unsleep(p);
959 		} else if (p->p_flag & P_SELECT)
960 			p->p_flag &= ~P_SELECT;
961 		splx(s);
962 	}
963 }
964