xref: /netbsd-src/sys/kern/sys_generic.c (revision 21e37cc72a480a47828990a439cde7ac9ffaf0c6)
1 /*	$NetBSD: sys_generic.c,v 1.81 2004/03/23 13:22:04 junyoung Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)sys_generic.c	8.9 (Berkeley) 2/14/95
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: sys_generic.c,v 1.81 2004/03/23 13:22:04 junyoung Exp $");
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/filedesc.h>
47 #include <sys/ioctl.h>
48 #include <sys/file.h>
49 #include <sys/proc.h>
50 #include <sys/socketvar.h>
51 #include <sys/signalvar.h>
52 #include <sys/uio.h>
53 #include <sys/kernel.h>
54 #include <sys/stat.h>
55 #include <sys/malloc.h>
56 #include <sys/poll.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60 
61 #include <sys/mount.h>
62 #include <sys/sa.h>
63 #include <sys/syscallargs.h>
64 
65 int selscan(struct proc *, fd_mask *, fd_mask *, int, register_t *);
66 int pollscan(struct proc *, struct pollfd *, int, register_t *);
67 
68 /*
69  * Read system call.
70  */
71 /* ARGSUSED */
72 int
73 sys_read(struct lwp *l, void *v, register_t *retval)
74 {
75 	struct sys_read_args /* {
76 		syscallarg(int)		fd;
77 		syscallarg(void *)	buf;
78 		syscallarg(size_t)	nbyte;
79 	} */ *uap = v;
80 	int		fd;
81 	struct file	*fp;
82 	struct proc	*p;
83 	struct filedesc	*fdp;
84 
85 	fd = SCARG(uap, fd);
86 	p = l->l_proc;
87 	fdp = p->p_fd;
88 
89 	if ((fp = fd_getfile(fdp, fd)) == NULL)
90 		return (EBADF);
91 
92 	if ((fp->f_flag & FREAD) == 0) {
93 		simple_unlock(&fp->f_slock);
94 		return (EBADF);
95 	}
96 
97 	FILE_USE(fp);
98 
99 	/* dofileread() will unuse the descriptor for us */
100 	return (dofileread(p, fd, fp, SCARG(uap, buf), SCARG(uap, nbyte),
101 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
102 }
103 
104 int
105 dofileread(struct proc *p, int fd, struct file *fp, void *buf, size_t nbyte,
106 	off_t *offset, int flags, register_t *retval)
107 {
108 	struct uio	auio;
109 	struct iovec	aiov;
110 	size_t		cnt;
111 	int		error;
112 #ifdef KTRACE
113 	struct iovec	ktriov = {0};
114 #endif
115 	error = 0;
116 
117 	aiov.iov_base = (caddr_t)buf;
118 	aiov.iov_len = nbyte;
119 	auio.uio_iov = &aiov;
120 	auio.uio_iovcnt = 1;
121 	auio.uio_resid = nbyte;
122 	auio.uio_rw = UIO_READ;
123 	auio.uio_segflg = UIO_USERSPACE;
124 	auio.uio_procp = p;
125 
126 	/*
127 	 * Reads return ssize_t because -1 is returned on error.  Therefore
128 	 * we must restrict the length to SSIZE_MAX to avoid garbage return
129 	 * values.
130 	 */
131 	if (auio.uio_resid > SSIZE_MAX) {
132 		error = EINVAL;
133 		goto out;
134 	}
135 
136 #ifdef KTRACE
137 	/*
138 	 * if tracing, save a copy of iovec
139 	 */
140 	if (KTRPOINT(p, KTR_GENIO))
141 		ktriov = aiov;
142 #endif
143 	cnt = auio.uio_resid;
144 	error = (*fp->f_ops->fo_read)(fp, offset, &auio, fp->f_cred, flags);
145 	if (error)
146 		if (auio.uio_resid != cnt && (error == ERESTART ||
147 		    error == EINTR || error == EWOULDBLOCK))
148 			error = 0;
149 	cnt -= auio.uio_resid;
150 #ifdef KTRACE
151 	if (KTRPOINT(p, KTR_GENIO) && error == 0)
152 		ktrgenio(p, fd, UIO_READ, &ktriov, cnt, error);
153 #endif
154 	*retval = cnt;
155  out:
156 	FILE_UNUSE(fp, p);
157 	return (error);
158 }
159 
160 /*
161  * Scatter read system call.
162  */
163 int
164 sys_readv(struct lwp *l, void *v, register_t *retval)
165 {
166 	struct sys_readv_args /* {
167 		syscallarg(int)				fd;
168 		syscallarg(const struct iovec *)	iovp;
169 		syscallarg(int)				iovcnt;
170 	} */ *uap = v;
171 	int		fd;
172 	struct file	*fp;
173 	struct proc	*p;
174 	struct filedesc	*fdp;
175 
176 	fd = SCARG(uap, fd);
177 	p = l->l_proc;
178 	fdp = p->p_fd;
179 
180 	if ((fp = fd_getfile(fdp, fd)) == NULL)
181 		return (EBADF);
182 
183 	if ((fp->f_flag & FREAD) == 0) {
184 		simple_unlock(&fp->f_slock);
185 		return (EBADF);
186 	}
187 
188 	FILE_USE(fp);
189 
190 	/* dofilereadv() will unuse the descriptor for us */
191 	return (dofilereadv(p, fd, fp, SCARG(uap, iovp), SCARG(uap, iovcnt),
192 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
193 }
194 
195 int
196 dofilereadv(struct proc *p, int fd, struct file *fp, const struct iovec *iovp,
197 	int iovcnt, off_t *offset, int flags, register_t *retval)
198 {
199 	struct uio	auio;
200 	struct iovec	*iov, *needfree, aiov[UIO_SMALLIOV];
201 	int		i, error;
202 	size_t		cnt;
203 	u_int		iovlen;
204 #ifdef KTRACE
205 	struct iovec	*ktriov;
206 #endif
207 
208 	error = 0;
209 #ifdef KTRACE
210 	ktriov = NULL;
211 #endif
212 	/* note: can't use iovlen until iovcnt is validated */
213 	iovlen = iovcnt * sizeof(struct iovec);
214 	if ((u_int)iovcnt > UIO_SMALLIOV) {
215 		if ((u_int)iovcnt > IOV_MAX) {
216 			error = EINVAL;
217 			goto out;
218 		}
219 		iov = malloc(iovlen, M_IOV, M_WAITOK);
220 		needfree = iov;
221 	} else if ((u_int)iovcnt > 0) {
222 		iov = aiov;
223 		needfree = NULL;
224 	} else {
225 		error = EINVAL;
226 		goto out;
227 	}
228 
229 	auio.uio_iov = iov;
230 	auio.uio_iovcnt = iovcnt;
231 	auio.uio_rw = UIO_READ;
232 	auio.uio_segflg = UIO_USERSPACE;
233 	auio.uio_procp = p;
234 	error = copyin(iovp, iov, iovlen);
235 	if (error)
236 		goto done;
237 	auio.uio_resid = 0;
238 	for (i = 0; i < iovcnt; i++) {
239 		auio.uio_resid += iov->iov_len;
240 		/*
241 		 * Reads return ssize_t because -1 is returned on error.
242 		 * Therefore we must restrict the length to SSIZE_MAX to
243 		 * avoid garbage return values.
244 		 */
245 		if (iov->iov_len > SSIZE_MAX || auio.uio_resid > SSIZE_MAX) {
246 			error = EINVAL;
247 			goto done;
248 		}
249 		iov++;
250 	}
251 #ifdef KTRACE
252 	/*
253 	 * if tracing, save a copy of iovec
254 	 */
255 	if (KTRPOINT(p, KTR_GENIO))  {
256 		ktriov = malloc(iovlen, M_TEMP, M_WAITOK);
257 		memcpy((caddr_t)ktriov, (caddr_t)auio.uio_iov, iovlen);
258 	}
259 #endif
260 	cnt = auio.uio_resid;
261 	error = (*fp->f_ops->fo_read)(fp, offset, &auio, fp->f_cred, flags);
262 	if (error)
263 		if (auio.uio_resid != cnt && (error == ERESTART ||
264 		    error == EINTR || error == EWOULDBLOCK))
265 			error = 0;
266 	cnt -= auio.uio_resid;
267 #ifdef KTRACE
268 	if (ktriov != NULL) {
269 		if (KTRPOINT(p, KTR_GENIO) && (error == 0))
270 			ktrgenio(p, fd, UIO_READ, ktriov, cnt, error);
271 		free(ktriov, M_TEMP);
272 	}
273 #endif
274 	*retval = cnt;
275  done:
276 	if (needfree)
277 		free(needfree, M_IOV);
278  out:
279 	FILE_UNUSE(fp, p);
280 	return (error);
281 }
282 
283 /*
284  * Write system call
285  */
286 int
287 sys_write(struct lwp *l, void *v, register_t *retval)
288 {
289 	struct sys_write_args /* {
290 		syscallarg(int)			fd;
291 		syscallarg(const void *)	buf;
292 		syscallarg(size_t)		nbyte;
293 	} */ *uap = v;
294 	int		fd;
295 	struct file	*fp;
296 	struct proc	*p;
297 	struct filedesc	*fdp;
298 
299 	fd = SCARG(uap, fd);
300 	p = l->l_proc;
301 	fdp = p->p_fd;
302 
303 	if ((fp = fd_getfile(fdp, fd)) == NULL)
304 		return (EBADF);
305 
306 	if ((fp->f_flag & FWRITE) == 0) {
307 		simple_unlock(&fp->f_slock);
308 		return (EBADF);
309 	}
310 
311 	FILE_USE(fp);
312 
313 	/* dofilewrite() will unuse the descriptor for us */
314 	return (dofilewrite(p, fd, fp, SCARG(uap, buf), SCARG(uap, nbyte),
315 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
316 }
317 
318 int
319 dofilewrite(struct proc *p, int fd, struct file *fp, const void *buf,
320 	size_t nbyte, off_t *offset, int flags, register_t *retval)
321 {
322 	struct uio	auio;
323 	struct iovec	aiov;
324 	size_t		cnt;
325 	int		error;
326 #ifdef KTRACE
327 	struct iovec	ktriov = {0};
328 #endif
329 
330 	error = 0;
331 	aiov.iov_base = (caddr_t)buf;		/* XXX kills const */
332 	aiov.iov_len = nbyte;
333 	auio.uio_iov = &aiov;
334 	auio.uio_iovcnt = 1;
335 	auio.uio_resid = nbyte;
336 	auio.uio_rw = UIO_WRITE;
337 	auio.uio_segflg = UIO_USERSPACE;
338 	auio.uio_procp = p;
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 #ifdef KTRACE
351 	/*
352 	 * if tracing, save a copy of iovec
353 	 */
354 	if (KTRPOINT(p, KTR_GENIO))
355 		ktriov = aiov;
356 #endif
357 	cnt = auio.uio_resid;
358 	error = (*fp->f_ops->fo_write)(fp, offset, &auio, fp->f_cred, flags);
359 	if (error) {
360 		if (auio.uio_resid != cnt && (error == ERESTART ||
361 		    error == EINTR || error == EWOULDBLOCK))
362 			error = 0;
363 		if (error == EPIPE)
364 			psignal(p, SIGPIPE);
365 	}
366 	cnt -= auio.uio_resid;
367 #ifdef KTRACE
368 	if (KTRPOINT(p, KTR_GENIO) && error == 0)
369 		ktrgenio(p, fd, UIO_WRITE, &ktriov, cnt, error);
370 #endif
371 	*retval = cnt;
372  out:
373 	FILE_UNUSE(fp, p);
374 	return (error);
375 }
376 
377 /*
378  * Gather write system call
379  */
380 int
381 sys_writev(struct lwp *l, void *v, register_t *retval)
382 {
383 	struct sys_writev_args /* {
384 		syscallarg(int)				fd;
385 		syscallarg(const struct iovec *)	iovp;
386 		syscallarg(int)				iovcnt;
387 	} */ *uap = v;
388 	int		fd;
389 	struct file	*fp;
390 	struct proc	*p;
391 	struct filedesc	*fdp;
392 
393 	fd = SCARG(uap, fd);
394 	p = l->l_proc;
395 	fdp = p->p_fd;
396 
397 	if ((fp = fd_getfile(fdp, fd)) == NULL)
398 		return (EBADF);
399 
400 	if ((fp->f_flag & FWRITE) == 0) {
401 		simple_unlock(&fp->f_slock);
402 		return (EBADF);
403 	}
404 
405 	FILE_USE(fp);
406 
407 	/* dofilewritev() will unuse the descriptor for us */
408 	return (dofilewritev(p, fd, fp, SCARG(uap, iovp), SCARG(uap, iovcnt),
409 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
410 }
411 
412 int
413 dofilewritev(struct proc *p, int fd, struct file *fp, const struct iovec *iovp,
414 	int iovcnt, off_t *offset, int flags, register_t *retval)
415 {
416 	struct uio	auio;
417 	struct iovec	*iov, *needfree, aiov[UIO_SMALLIOV];
418 	int		i, error;
419 	size_t		cnt;
420 	u_int		iovlen;
421 #ifdef KTRACE
422 	struct iovec	*ktriov;
423 #endif
424 
425 	error = 0;
426 #ifdef KTRACE
427 	ktriov = NULL;
428 #endif
429 	/* note: can't use iovlen until iovcnt is validated */
430 	iovlen = iovcnt * sizeof(struct iovec);
431 	if ((u_int)iovcnt > UIO_SMALLIOV) {
432 		if ((u_int)iovcnt > IOV_MAX) {
433 			error = EINVAL;
434 			goto out;
435 		}
436 		iov = malloc(iovlen, M_IOV, M_WAITOK);
437 		needfree = iov;
438 	} else if ((u_int)iovcnt > 0) {
439 		iov = aiov;
440 		needfree = NULL;
441 	} else {
442 		error = EINVAL;
443 		goto out;
444 	}
445 
446 	auio.uio_iov = iov;
447 	auio.uio_iovcnt = iovcnt;
448 	auio.uio_rw = UIO_WRITE;
449 	auio.uio_segflg = UIO_USERSPACE;
450 	auio.uio_procp = p;
451 	error = copyin(iovp, iov, iovlen);
452 	if (error)
453 		goto done;
454 	auio.uio_resid = 0;
455 	for (i = 0; i < iovcnt; i++) {
456 		auio.uio_resid += iov->iov_len;
457 		/*
458 		 * Writes return ssize_t because -1 is returned on error.
459 		 * Therefore we must restrict the length to SSIZE_MAX to
460 		 * avoid garbage return values.
461 		 */
462 		if (iov->iov_len > SSIZE_MAX || auio.uio_resid > SSIZE_MAX) {
463 			error = EINVAL;
464 			goto done;
465 		}
466 		iov++;
467 	}
468 #ifdef KTRACE
469 	/*
470 	 * if tracing, save a copy of iovec
471 	 */
472 	if (KTRPOINT(p, KTR_GENIO))  {
473 		ktriov = malloc(iovlen, M_TEMP, M_WAITOK);
474 		memcpy((caddr_t)ktriov, (caddr_t)auio.uio_iov, iovlen);
475 	}
476 #endif
477 	cnt = auio.uio_resid;
478 	error = (*fp->f_ops->fo_write)(fp, offset, &auio, fp->f_cred, flags);
479 	if (error) {
480 		if (auio.uio_resid != cnt && (error == ERESTART ||
481 		    error == EINTR || error == EWOULDBLOCK))
482 			error = 0;
483 		if (error == EPIPE)
484 			psignal(p, SIGPIPE);
485 	}
486 	cnt -= auio.uio_resid;
487 #ifdef KTRACE
488 	if (ktriov != NULL) {
489 		if (KTRPOINT(p, KTR_GENIO) && (error == 0))
490 			ktrgenio(p, fd, UIO_WRITE, ktriov, cnt, error);
491 		free(ktriov, M_TEMP);
492 	}
493 #endif
494 	*retval = cnt;
495  done:
496 	if (needfree)
497 		free(needfree, M_IOV);
498  out:
499 	FILE_UNUSE(fp, p);
500 	return (error);
501 }
502 
503 /*
504  * Ioctl system call
505  */
506 /* ARGSUSED */
507 int
508 sys_ioctl(struct lwp *l, void *v, register_t *retval)
509 {
510 	struct sys_ioctl_args /* {
511 		syscallarg(int)		fd;
512 		syscallarg(u_long)	com;
513 		syscallarg(caddr_t)	data;
514 	} */ *uap = v;
515 	struct file	*fp;
516 	struct proc	*p;
517 	struct filedesc	*fdp;
518 	u_long		com;
519 	int		error;
520 	u_int		size;
521 	caddr_t		data, memp;
522 #define	STK_PARAMS	128
523 	u_long		stkbuf[STK_PARAMS/sizeof(u_long)];
524 
525 	error = 0;
526 	p = l->l_proc;
527 	fdp = p->p_fd;
528 
529 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL)
530 		return (EBADF);
531 
532 	FILE_USE(fp);
533 
534 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
535 		error = EBADF;
536 		com = 0;
537 		goto out;
538 	}
539 
540 	switch (com = SCARG(uap, com)) {
541 	case FIONCLEX:
542 		fdp->fd_ofileflags[SCARG(uap, fd)] &= ~UF_EXCLOSE;
543 		goto out;
544 
545 	case FIOCLEX:
546 		fdp->fd_ofileflags[SCARG(uap, fd)] |= UF_EXCLOSE;
547 		goto out;
548 	}
549 
550 	/*
551 	 * Interpret high order word to find amount of data to be
552 	 * copied to/from the user's address space.
553 	 */
554 	size = IOCPARM_LEN(com);
555 	if (size > IOCPARM_MAX) {
556 		error = ENOTTY;
557 		goto out;
558 	}
559 	memp = NULL;
560 	if (size > sizeof(stkbuf)) {
561 		memp = (caddr_t)malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
562 		data = memp;
563 	} else
564 		data = (caddr_t)stkbuf;
565 	if (com&IOC_IN) {
566 		if (size) {
567 			error = copyin(SCARG(uap, data), data, size);
568 			if (error) {
569 				if (memp)
570 					free(memp, M_IOCTLOPS);
571 				goto out;
572 			}
573 #ifdef KTRACE
574 			if (KTRPOINT(p, KTR_GENIO)) {
575 				struct iovec iov;
576 				iov.iov_base = SCARG(uap, data);
577 				iov.iov_len = size;
578 				ktrgenio(p, SCARG(uap, fd), UIO_WRITE, &iov,
579 					size, 0);
580 			}
581 #endif
582 		} else
583 			*(caddr_t *)data = SCARG(uap, data);
584 	} else if ((com&IOC_OUT) && size)
585 		/*
586 		 * Zero the buffer so the user always
587 		 * gets back something deterministic.
588 		 */
589 		memset(data, 0, size);
590 	else if (com&IOC_VOID)
591 		*(caddr_t *)data = SCARG(uap, data);
592 
593 	switch (com) {
594 
595 	case FIONBIO:
596 		if (*(int *)data != 0)
597 			fp->f_flag |= FNONBLOCK;
598 		else
599 			fp->f_flag &= ~FNONBLOCK;
600 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, data, p);
601 		break;
602 
603 	case FIOASYNC:
604 		if (*(int *)data != 0)
605 			fp->f_flag |= FASYNC;
606 		else
607 			fp->f_flag &= ~FASYNC;
608 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, data, p);
609 		break;
610 
611 	default:
612 		error = (*fp->f_ops->fo_ioctl)(fp, com, data, p);
613 		/*
614 		 * Copy any data to user, size was
615 		 * already set and checked above.
616 		 */
617 		if (error == 0 && (com&IOC_OUT) && size) {
618 			error = copyout(data, SCARG(uap, data), size);
619 #ifdef KTRACE
620 			if (KTRPOINT(p, KTR_GENIO)) {
621 				struct iovec iov;
622 				iov.iov_base = SCARG(uap, data);
623 				iov.iov_len = size;
624 				ktrgenio(p, SCARG(uap, fd), UIO_READ, &iov,
625 					size, error);
626 			}
627 #endif
628 		}
629 		break;
630 	}
631 	if (memp)
632 		free(memp, M_IOCTLOPS);
633  out:
634 	FILE_UNUSE(fp, p);
635 	switch (error) {
636 	case -1:
637 		printf("sys_ioctl: _IO%s%s('%c', %lu, %lu) returned -1: "
638 		    "pid=%d comm=%s\n",
639 		    (com & IOC_IN) ? "W" : "", (com & IOC_OUT) ? "R" : "",
640 		    (char)IOCGROUP(com), (com & 0xff), IOCPARM_LEN(com),
641 		    p->p_pid, p->p_comm);
642 		/* FALLTHROUGH */
643 	case EPASSTHROUGH:
644 		error = ENOTTY;
645 		/* FALLTHROUGH */
646 	default:
647 		return (error);
648 	}
649 }
650 
651 int	selwait, nselcoll;
652 
653 /*
654  * Select system call.
655  */
656 int
657 sys_select(struct lwp *l, void *v, register_t *retval)
658 {
659 	struct sys_select_args /* {
660 		syscallarg(int)			nd;
661 		syscallarg(fd_set *)		in;
662 		syscallarg(fd_set *)		ou;
663 		syscallarg(fd_set *)		ex;
664 		syscallarg(struct timeval *)	tv;
665 	} */ *uap = v;
666 	struct proc	*p;
667 	caddr_t		bits;
668 	char		smallbits[howmany(FD_SETSIZE, NFDBITS) *
669 			    sizeof(fd_mask) * 6];
670 	struct		timeval atv;
671 	int		s, ncoll, error, timo;
672 	size_t		ni;
673 
674 	error = 0;
675 	p = l->l_proc;
676 	if (SCARG(uap, nd) < 0)
677 		return (EINVAL);
678 	if (SCARG(uap, nd) > p->p_fd->fd_nfiles) {
679 		/* forgiving; slightly wrong */
680 		SCARG(uap, nd) = p->p_fd->fd_nfiles;
681 	}
682 	ni = howmany(SCARG(uap, nd), NFDBITS) * sizeof(fd_mask);
683 	if (ni * 6 > sizeof(smallbits))
684 		bits = malloc(ni * 6, M_TEMP, M_WAITOK);
685 	else
686 		bits = smallbits;
687 
688 #define	getbits(name, x)						\
689 	if (SCARG(uap, name)) {						\
690 		error = copyin(SCARG(uap, name), bits + ni * x, ni);	\
691 		if (error)						\
692 			goto done;					\
693 	} else								\
694 		memset(bits + ni * x, 0, ni);
695 	getbits(in, 0);
696 	getbits(ou, 1);
697 	getbits(ex, 2);
698 #undef	getbits
699 
700 	timo = 0;
701 	if (SCARG(uap, tv)) {
702 		error = copyin(SCARG(uap, tv), (caddr_t)&atv,
703 			sizeof(atv));
704 		if (error)
705 			goto done;
706 		if (itimerfix(&atv)) {
707 			error = EINVAL;
708 			goto done;
709 		}
710 		s = splclock();
711 		timeradd(&atv, &time, &atv);
712 		splx(s);
713 	}
714 
715  retry:
716 	ncoll = nselcoll;
717 	l->l_flag |= L_SELECT;
718 	error = selscan(p, (fd_mask *)(bits + ni * 0),
719 			   (fd_mask *)(bits + ni * 3), SCARG(uap, nd), retval);
720 	if (error || *retval)
721 		goto done;
722 	if (SCARG(uap, tv)) {
723 		/*
724 		 * We have to recalculate the timeout on every retry.
725 		 */
726 		timo = hzto(&atv);
727 		if (timo <= 0)
728 			goto done;
729 	}
730 	s = splsched();
731 	if ((l->l_flag & L_SELECT) == 0 || nselcoll != ncoll) {
732 		splx(s);
733 		goto retry;
734 	}
735 	l->l_flag &= ~L_SELECT;
736 	error = tsleep((caddr_t)&selwait, PSOCK | PCATCH, "select", timo);
737 	splx(s);
738 	if (error == 0)
739 		goto retry;
740  done:
741 	l->l_flag &= ~L_SELECT;
742 	/* select is not restarted after signals... */
743 	if (error == ERESTART)
744 		error = EINTR;
745 	if (error == EWOULDBLOCK)
746 		error = 0;
747 	if (error == 0) {
748 
749 #define	putbits(name, x)						\
750 		if (SCARG(uap, name)) {					\
751 			error = copyout(bits + ni * x, SCARG(uap, name), ni); \
752 			if (error)					\
753 				goto out;				\
754 		}
755 		putbits(in, 3);
756 		putbits(ou, 4);
757 		putbits(ex, 5);
758 #undef putbits
759 	}
760  out:
761 	if (ni * 6 > sizeof(smallbits))
762 		free(bits, M_TEMP);
763 	return (error);
764 }
765 
766 int
767 selscan(struct proc *p, fd_mask *ibitp, fd_mask *obitp, int nfd,
768 	register_t *retval)
769 {
770 	struct filedesc	*fdp;
771 	int		msk, i, j, fd, n;
772 	fd_mask		ibits, obits;
773 	struct file	*fp;
774 	static const int flag[3] = { POLLRDNORM | POLLHUP | POLLERR,
775 			       POLLWRNORM | POLLHUP | POLLERR,
776 			       POLLRDBAND };
777 
778 	fdp = p->p_fd;
779 	n = 0;
780 	for (msk = 0; msk < 3; msk++) {
781 		for (i = 0; i < nfd; i += NFDBITS) {
782 			ibits = *ibitp++;
783 			obits = 0;
784 			while ((j = ffs(ibits)) && (fd = i + --j) < nfd) {
785 				ibits &= ~(1 << j);
786 				if ((fp = fd_getfile(fdp, fd)) == NULL)
787 					return (EBADF);
788 				FILE_USE(fp);
789 				if ((*fp->f_ops->fo_poll)(fp, flag[msk], p)) {
790 					obits |= (1 << j);
791 					n++;
792 				}
793 				FILE_UNUSE(fp, p);
794 			}
795 			*obitp++ = obits;
796 		}
797 	}
798 	*retval = n;
799 	return (0);
800 }
801 
802 /*
803  * Poll system call.
804  */
805 int
806 sys_poll(struct lwp *l, void *v, register_t *retval)
807 {
808 	struct sys_poll_args /* {
809 		syscallarg(struct pollfd *)	fds;
810 		syscallarg(u_int)		nfds;
811 		syscallarg(int)			timeout;
812 	} */ *uap = v;
813 	struct proc	*p;
814 	caddr_t		bits;
815 	char		smallbits[32 * sizeof(struct pollfd)];
816 	struct timeval	atv;
817 	int		s, ncoll, error, timo;
818 	size_t		ni;
819 
820 	error = 0;
821 	p = l->l_proc;
822 	if (SCARG(uap, nfds) > p->p_fd->fd_nfiles) {
823 		/* forgiving; slightly wrong */
824 		SCARG(uap, nfds) = p->p_fd->fd_nfiles;
825 	}
826 	ni = SCARG(uap, nfds) * sizeof(struct pollfd);
827 	if (ni > sizeof(smallbits))
828 		bits = malloc(ni, M_TEMP, M_WAITOK);
829 	else
830 		bits = smallbits;
831 
832 	error = copyin(SCARG(uap, fds), bits, ni);
833 	if (error)
834 		goto done;
835 
836 	timo = 0;
837 	if (SCARG(uap, timeout) != INFTIM) {
838 		atv.tv_sec = SCARG(uap, timeout) / 1000;
839 		atv.tv_usec = (SCARG(uap, timeout) % 1000) * 1000;
840 		if (itimerfix(&atv)) {
841 			error = EINVAL;
842 			goto done;
843 		}
844 		s = splclock();
845 		timeradd(&atv, &time, &atv);
846 		splx(s);
847 	}
848 
849  retry:
850 	ncoll = nselcoll;
851 	l->l_flag |= L_SELECT;
852 	error = pollscan(p, (struct pollfd *)bits, SCARG(uap, nfds), retval);
853 	if (error || *retval)
854 		goto done;
855 	if (SCARG(uap, timeout) != INFTIM) {
856 		/*
857 		 * We have to recalculate the timeout on every retry.
858 		 */
859 		timo = hzto(&atv);
860 		if (timo <= 0)
861 			goto done;
862 	}
863 	s = splsched();
864 	if ((l->l_flag & L_SELECT) == 0 || nselcoll != ncoll) {
865 		splx(s);
866 		goto retry;
867 	}
868 	l->l_flag &= ~L_SELECT;
869 	error = tsleep((caddr_t)&selwait, PSOCK | PCATCH, "poll", timo);
870 	splx(s);
871 	if (error == 0)
872 		goto retry;
873  done:
874 	l->l_flag &= ~L_SELECT;
875 	/* poll is not restarted after signals... */
876 	if (error == ERESTART)
877 		error = EINTR;
878 	if (error == EWOULDBLOCK)
879 		error = 0;
880 	if (error == 0) {
881 		error = copyout(bits, SCARG(uap, fds), ni);
882 		if (error)
883 			goto out;
884 	}
885  out:
886 	if (ni > sizeof(smallbits))
887 		free(bits, M_TEMP);
888 	return (error);
889 }
890 
891 int
892 pollscan(struct proc *p, struct pollfd *fds, int nfd, register_t *retval)
893 {
894 	struct filedesc	*fdp;
895 	int		i, n;
896 	struct file	*fp;
897 
898 	fdp = p->p_fd;
899 	n = 0;
900 	for (i = 0; i < nfd; i++, fds++) {
901 		if (fds->fd >= fdp->fd_nfiles) {
902 			fds->revents = POLLNVAL;
903 			n++;
904 		} else if (fds->fd < 0) {
905 			fds->revents = 0;
906 		} else {
907 			if ((fp = fd_getfile(fdp, fds->fd)) == NULL) {
908 				fds->revents = POLLNVAL;
909 				n++;
910 			} else {
911 				FILE_USE(fp);
912 				fds->revents = (*fp->f_ops->fo_poll)(fp,
913 				    fds->events | POLLERR | POLLHUP, p);
914 				if (fds->revents != 0)
915 					n++;
916 				FILE_UNUSE(fp, p);
917 			}
918 		}
919 	}
920 	*retval = n;
921 	return (0);
922 }
923 
924 /*ARGSUSED*/
925 int
926 seltrue(dev_t dev, int events, struct proc *p)
927 {
928 
929 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
930 }
931 
932 /*
933  * Record a select request.
934  */
935 void
936 selrecord(struct proc *selector, struct selinfo *sip)
937 {
938 	struct lwp	*l;
939 	struct proc	*p;
940 	pid_t		mypid;
941 
942 	mypid = selector->p_pid;
943 	if (sip->sel_pid == mypid)
944 		return;
945 	if (sip->sel_pid && (p = pfind(sip->sel_pid))) {
946 		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
947 			if (l->l_wchan == (caddr_t)&selwait) {
948 				sip->sel_collision = 1;
949 				return;
950 			}
951 		}
952 	}
953 
954 	sip->sel_pid = mypid;
955 }
956 
957 /*
958  * Do a wakeup when a selectable event occurs.
959  */
960 void
961 selwakeup(sip)
962 	struct selinfo *sip;
963 {
964 	struct lwp *l;
965 	struct proc *p;
966 	int s;
967 
968 	if (sip->sel_pid == 0)
969 		return;
970 	if (sip->sel_collision) {
971 		sip->sel_pid = 0;
972 		nselcoll++;
973 		sip->sel_collision = 0;
974 		wakeup((caddr_t)&selwait);
975 		return;
976 	}
977 	p = pfind(sip->sel_pid);
978 	sip->sel_pid = 0;
979 	if (p != NULL) {
980 		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
981 			SCHED_LOCK(s);
982 			if (l->l_wchan == (caddr_t)&selwait) {
983 				if (l->l_stat == LSSLEEP)
984 					setrunnable(l);
985 				else
986 					unsleep(l);
987 			} else if (l->l_flag & L_SELECT)
988 				l->l_flag &= ~L_SELECT;
989 			SCHED_UNLOCK(s);
990 		}
991 	}
992 }
993