xref: /openbsd-src/sys/kern/sys_generic.c (revision 5c389b79544373bccfce668b646e62e7ba9802a3)
1 /*	$OpenBSD: sys_generic.c,v 1.156 2023/05/09 14:22:17 visa Exp $	*/
2 /*	$NetBSD: sys_generic.c,v 1.24 1996/03/29 00:25:32 cgd Exp $	*/
3 
4 /*
5  * Copyright (c) 1996 Theo de Raadt
6  * Copyright (c) 1982, 1986, 1989, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. 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  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/filedesc.h>
44 #include <sys/ioctl.h>
45 #include <sys/fcntl.h>
46 #include <sys/vnode.h>
47 #include <sys/file.h>
48 #include <sys/proc.h>
49 #include <sys/resourcevar.h>
50 #include <sys/socketvar.h>
51 #include <sys/signalvar.h>
52 #include <sys/uio.h>
53 #include <sys/time.h>
54 #include <sys/malloc.h>
55 #include <sys/poll.h>
56 #include <sys/eventvar.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60 #include <sys/pledge.h>
61 
62 #include <sys/mount.h>
63 #include <sys/syscallargs.h>
64 
65 /*
66  * Debug values:
67  *  1 - print implementation errors, things that should not happen.
68  *  2 - print ppoll(2) information, somewhat verbose
69  *  3 - print pselect(2) and ppoll(2) information, very verbose
70  */
71 int kqpoll_debug = 0;
72 #define DPRINTFN(v, x...) if (kqpoll_debug > v) {			\
73 	printf("%s(%d): ", curproc->p_p->ps_comm, curproc->p_tid);	\
74 	printf(x);							\
75 }
76 
77 int pselregister(struct proc *, fd_set **, fd_set **, int, int *, int *);
78 int pselcollect(struct proc *, struct kevent *, fd_set **, int *);
79 void ppollregister(struct proc *, struct pollfd *, int, int *, int *);
80 int ppollcollect(struct proc *, struct kevent *, struct pollfd *, u_int);
81 
82 int pollout(struct pollfd *, struct pollfd *, u_int);
83 int dopselect(struct proc *, int, fd_set *, fd_set *, fd_set *,
84     struct timespec *, const sigset_t *, register_t *);
85 int doppoll(struct proc *, struct pollfd *, u_int, struct timespec *,
86     const sigset_t *, register_t *);
87 
88 int
89 iovec_copyin(const struct iovec *uiov, struct iovec **iovp, struct iovec *aiov,
90     unsigned int iovcnt, size_t *residp)
91 {
92 #ifdef KTRACE
93 	struct proc *p = curproc;
94 #endif
95 	struct iovec *iov;
96 	int error, i;
97 	size_t resid = 0;
98 
99 	if (iovcnt > UIO_SMALLIOV) {
100 		if (iovcnt > IOV_MAX)
101 			return (EINVAL);
102 		iov = mallocarray(iovcnt, sizeof(*iov), M_IOV, M_WAITOK);
103 	} else if (iovcnt > 0) {
104 		iov = aiov;
105 	} else {
106 		return (EINVAL);
107 	}
108 	*iovp = iov;
109 
110 	if ((error = copyin(uiov, iov, iovcnt * sizeof(*iov))))
111 		return (error);
112 
113 #ifdef KTRACE
114 	if (KTRPOINT(p, KTR_STRUCT))
115 		ktriovec(p, iov, iovcnt);
116 #endif
117 
118 	for (i = 0; i < iovcnt; i++) {
119 		resid += iov->iov_len;
120 		/*
121 		 * Writes return ssize_t because -1 is returned on error.
122 		 * Therefore we must restrict the length to SSIZE_MAX to
123 		 * avoid garbage return values.  Note that the addition is
124 		 * guaranteed to not wrap because SSIZE_MAX * 2 < SIZE_MAX.
125 		 */
126 		if (iov->iov_len > SSIZE_MAX || resid > SSIZE_MAX)
127 			return (EINVAL);
128 		iov++;
129 	}
130 
131 	if (residp != NULL)
132 		*residp = resid;
133 
134 	return (0);
135 }
136 
137 void
138 iovec_free(struct iovec *iov, unsigned int iovcnt)
139 {
140 	if (iovcnt > UIO_SMALLIOV)
141 		free(iov, M_IOV, iovcnt * sizeof(*iov));
142 }
143 
144 /*
145  * Read system call.
146  */
147 int
148 sys_read(struct proc *p, void *v, register_t *retval)
149 {
150 	struct sys_read_args /* {
151 		syscallarg(int) fd;
152 		syscallarg(void *) buf;
153 		syscallarg(size_t) nbyte;
154 	} */ *uap = v;
155 	struct iovec iov;
156 	struct uio auio;
157 
158 	iov.iov_base = SCARG(uap, buf);
159 	iov.iov_len = SCARG(uap, nbyte);
160 	if (iov.iov_len > SSIZE_MAX)
161 		return (EINVAL);
162 
163 	auio.uio_iov = &iov;
164 	auio.uio_iovcnt = 1;
165 	auio.uio_resid = iov.iov_len;
166 
167 	return (dofilereadv(p, SCARG(uap, fd), &auio, 0, retval));
168 }
169 
170 /*
171  * Scatter read system call.
172  */
173 int
174 sys_readv(struct proc *p, void *v, register_t *retval)
175 {
176 	struct sys_readv_args /* {
177 		syscallarg(int) fd;
178 		syscallarg(const struct iovec *) iovp;
179 		syscallarg(int) iovcnt;
180 	} */ *uap = v;
181 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
182 	int error, iovcnt = SCARG(uap, iovcnt);
183 	struct uio auio;
184 	size_t resid;
185 
186 	error = iovec_copyin(SCARG(uap, iovp), &iov, aiov, iovcnt, &resid);
187 	if (error)
188 		goto done;
189 
190 	auio.uio_iov = iov;
191 	auio.uio_iovcnt = iovcnt;
192 	auio.uio_resid = resid;
193 
194 	error = dofilereadv(p, SCARG(uap, fd), &auio, 0, retval);
195  done:
196 	iovec_free(iov, iovcnt);
197 	return (error);
198 }
199 
200 int
201 dofilereadv(struct proc *p, int fd, struct uio *uio, int flags,
202     register_t *retval)
203 {
204 	struct filedesc *fdp = p->p_fd;
205 	struct file *fp;
206 	long cnt, error = 0;
207 	u_int iovlen;
208 #ifdef KTRACE
209 	struct iovec *ktriov = NULL;
210 #endif
211 
212 	KASSERT(uio->uio_iov != NULL && uio->uio_iovcnt > 0);
213 	iovlen = uio->uio_iovcnt * sizeof(struct iovec);
214 
215 	if ((fp = fd_getfile_mode(fdp, fd, FREAD)) == NULL)
216 		return (EBADF);
217 
218 	/* Checks for positioned read. */
219 	if (flags & FO_POSITION) {
220 		struct vnode *vp = fp->f_data;
221 
222 		if (fp->f_type != DTYPE_VNODE || vp->v_type == VFIFO ||
223 		    (vp->v_flag & VISTTY)) {
224 			error = ESPIPE;
225 			goto done;
226 		}
227 
228 		if (uio->uio_offset < 0 && vp->v_type != VCHR) {
229 			error = EINVAL;
230 			goto done;
231 		}
232 	}
233 
234 	uio->uio_rw = UIO_READ;
235 	uio->uio_segflg = UIO_USERSPACE;
236 	uio->uio_procp = p;
237 #ifdef KTRACE
238 	/*
239 	 * if tracing, save a copy of iovec
240 	 */
241 	if (KTRPOINT(p, KTR_GENIO)) {
242 		ktriov = malloc(iovlen, M_TEMP, M_WAITOK);
243 		memcpy(ktriov, uio->uio_iov, iovlen);
244 	}
245 #endif
246 	cnt = uio->uio_resid;
247 	error = (*fp->f_ops->fo_read)(fp, uio, flags);
248 	if (error) {
249 		if (uio->uio_resid != cnt && (error == ERESTART ||
250 		    error == EINTR || error == EWOULDBLOCK))
251 			error = 0;
252 	}
253 	cnt -= uio->uio_resid;
254 
255 	mtx_enter(&fp->f_mtx);
256 	fp->f_rxfer++;
257 	fp->f_rbytes += cnt;
258 	mtx_leave(&fp->f_mtx);
259 #ifdef KTRACE
260 	if (ktriov != NULL) {
261 		if (error == 0)
262 			ktrgenio(p, fd, UIO_READ, ktriov, cnt);
263 		free(ktriov, M_TEMP, iovlen);
264 	}
265 #endif
266 	*retval = cnt;
267  done:
268 	FRELE(fp, p);
269 	return (error);
270 }
271 
272 /*
273  * Write system call
274  */
275 int
276 sys_write(struct proc *p, void *v, register_t *retval)
277 {
278 	struct sys_write_args /* {
279 		syscallarg(int) fd;
280 		syscallarg(const void *) buf;
281 		syscallarg(size_t) nbyte;
282 	} */ *uap = v;
283 	struct iovec iov;
284 	struct uio auio;
285 
286 	iov.iov_base = (void *)SCARG(uap, buf);
287 	iov.iov_len = SCARG(uap, nbyte);
288 	if (iov.iov_len > SSIZE_MAX)
289 		return (EINVAL);
290 
291 	auio.uio_iov = &iov;
292 	auio.uio_iovcnt = 1;
293 	auio.uio_resid = iov.iov_len;
294 
295 	return (dofilewritev(p, SCARG(uap, fd), &auio, 0, retval));
296 }
297 
298 /*
299  * Gather write system call
300  */
301 int
302 sys_writev(struct proc *p, void *v, register_t *retval)
303 {
304 	struct sys_writev_args /* {
305 		syscallarg(int) fd;
306 		syscallarg(const struct iovec *) iovp;
307 		syscallarg(int) iovcnt;
308 	} */ *uap = v;
309 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
310 	int error, iovcnt = SCARG(uap, iovcnt);
311 	struct uio auio;
312 	size_t resid;
313 
314 	error = iovec_copyin(SCARG(uap, iovp), &iov, aiov, iovcnt, &resid);
315 	if (error)
316 		goto done;
317 
318 	auio.uio_iov = iov;
319 	auio.uio_iovcnt = iovcnt;
320 	auio.uio_resid = resid;
321 
322 	error = dofilewritev(p, SCARG(uap, fd), &auio, 0, retval);
323  done:
324 	iovec_free(iov, iovcnt);
325  	return (error);
326 }
327 
328 int
329 dofilewritev(struct proc *p, int fd, struct uio *uio, int flags,
330     register_t *retval)
331 {
332 	struct filedesc *fdp = p->p_fd;
333 	struct file *fp;
334 	long cnt, error = 0;
335 	u_int iovlen;
336 #ifdef KTRACE
337 	struct iovec *ktriov = NULL;
338 #endif
339 
340 	KASSERT(uio->uio_iov != NULL && uio->uio_iovcnt > 0);
341 	iovlen = uio->uio_iovcnt * sizeof(struct iovec);
342 
343 	if ((fp = fd_getfile_mode(fdp, fd, FWRITE)) == NULL)
344 		return (EBADF);
345 
346 	/* Checks for positioned write. */
347 	if (flags & FO_POSITION) {
348 		struct vnode *vp = fp->f_data;
349 
350 		if (fp->f_type != DTYPE_VNODE || vp->v_type == VFIFO ||
351 		    (vp->v_flag & VISTTY)) {
352 			error = ESPIPE;
353 			goto done;
354 		}
355 
356 		if (uio->uio_offset < 0 && vp->v_type != VCHR) {
357 			error = EINVAL;
358 			goto done;
359 		}
360 	}
361 
362 	uio->uio_rw = UIO_WRITE;
363 	uio->uio_segflg = UIO_USERSPACE;
364 	uio->uio_procp = p;
365 #ifdef KTRACE
366 	/*
367 	 * if tracing, save a copy of iovec
368 	 */
369 	if (KTRPOINT(p, KTR_GENIO)) {
370 		ktriov = malloc(iovlen, M_TEMP, M_WAITOK);
371 		memcpy(ktriov, uio->uio_iov, iovlen);
372 	}
373 #endif
374 	cnt = uio->uio_resid;
375 	error = (*fp->f_ops->fo_write)(fp, uio, flags);
376 	if (error) {
377 		if (uio->uio_resid != cnt && (error == ERESTART ||
378 		    error == EINTR || error == EWOULDBLOCK))
379 			error = 0;
380 		if (error == EPIPE) {
381 			KERNEL_LOCK();
382 			ptsignal(p, SIGPIPE, STHREAD);
383 			KERNEL_UNLOCK();
384 		}
385 	}
386 	cnt -= uio->uio_resid;
387 
388 	mtx_enter(&fp->f_mtx);
389 	fp->f_wxfer++;
390 	fp->f_wbytes += cnt;
391 	mtx_leave(&fp->f_mtx);
392 #ifdef KTRACE
393 	if (ktriov != NULL) {
394 		if (error == 0)
395 			ktrgenio(p, fd, UIO_WRITE, ktriov, cnt);
396 		free(ktriov, M_TEMP, iovlen);
397 	}
398 #endif
399 	*retval = cnt;
400  done:
401 	FRELE(fp, p);
402 	return (error);
403 }
404 
405 /*
406  * Ioctl system call
407  */
408 int
409 sys_ioctl(struct proc *p, void *v, register_t *retval)
410 {
411 	struct sys_ioctl_args /* {
412 		syscallarg(int) fd;
413 		syscallarg(u_long) com;
414 		syscallarg(void *) data;
415 	} */ *uap = v;
416 	struct file *fp;
417 	struct filedesc *fdp = p->p_fd;
418 	u_long com = SCARG(uap, com);
419 	int error = 0;
420 	u_int size = 0;
421 	caddr_t data, memp = NULL;
422 	int tmp;
423 #define STK_PARAMS	128
424 	long long stkbuf[STK_PARAMS / sizeof(long long)];
425 
426 	if ((fp = fd_getfile_mode(fdp, SCARG(uap, fd), FREAD|FWRITE)) == NULL)
427 		return (EBADF);
428 
429 	if (fp->f_type == DTYPE_SOCKET) {
430 		struct socket *so = fp->f_data;
431 
432 		if (so->so_state & SS_DNS) {
433 			error = EINVAL;
434 			goto out;
435 		}
436 	}
437 
438 	error = pledge_ioctl(p, com, fp);
439 	if (error)
440 		goto out;
441 
442 	switch (com) {
443 	case FIONCLEX:
444 	case FIOCLEX:
445 		fdplock(fdp);
446 		if (com == FIONCLEX)
447 			fdp->fd_ofileflags[SCARG(uap, fd)] &= ~UF_EXCLOSE;
448 		else
449 			fdp->fd_ofileflags[SCARG(uap, fd)] |= UF_EXCLOSE;
450 		fdpunlock(fdp);
451 		goto out;
452 	}
453 
454 	/*
455 	 * Interpret high order word to find amount of data to be
456 	 * copied to/from the user's address space.
457 	 */
458 	size = IOCPARM_LEN(com);
459 	if (size > IOCPARM_MAX) {
460 		error = ENOTTY;
461 		goto out;
462 	}
463 	if (size > sizeof (stkbuf)) {
464 		memp = malloc(size, M_IOCTLOPS, M_WAITOK);
465 		data = memp;
466 	} else
467 		data = (caddr_t)stkbuf;
468 	if (com&IOC_IN) {
469 		if (size) {
470 			error = copyin(SCARG(uap, data), data, size);
471 			if (error) {
472 				goto out;
473 			}
474 		} else
475 			*(caddr_t *)data = SCARG(uap, data);
476 	} else if ((com&IOC_OUT) && size)
477 		/*
478 		 * Zero the buffer so the user always
479 		 * gets back something deterministic.
480 		 */
481 		memset(data, 0, size);
482 	else if (com&IOC_VOID)
483 		*(caddr_t *)data = SCARG(uap, data);
484 
485 	switch (com) {
486 
487 	case FIONBIO:
488 		if ((tmp = *(int *)data) != 0)
489 			atomic_setbits_int(&fp->f_flag, FNONBLOCK);
490 		else
491 			atomic_clearbits_int(&fp->f_flag, FNONBLOCK);
492 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
493 		break;
494 
495 	case FIOASYNC:
496 		if ((tmp = *(int *)data) != 0)
497 			atomic_setbits_int(&fp->f_flag, FASYNC);
498 		else
499 			atomic_clearbits_int(&fp->f_flag, FASYNC);
500 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p);
501 		break;
502 
503 	default:
504 		error = (*fp->f_ops->fo_ioctl)(fp, com, data, p);
505 		break;
506 	}
507 	/*
508 	 * Copy any data to user, size was
509 	 * already set and checked above.
510 	 */
511 	if (error == 0 && (com&IOC_OUT) && size)
512 		error = copyout(data, SCARG(uap, data), size);
513 out:
514 	FRELE(fp, p);
515 	free(memp, M_IOCTLOPS, size);
516 	return (error);
517 }
518 
519 /*
520  * Select system call.
521  */
522 int
523 sys_select(struct proc *p, void *v, register_t *retval)
524 {
525 	struct sys_select_args /* {
526 		syscallarg(int) nd;
527 		syscallarg(fd_set *) in;
528 		syscallarg(fd_set *) ou;
529 		syscallarg(fd_set *) ex;
530 		syscallarg(struct timeval *) tv;
531 	} */ *uap = v;
532 
533 	struct timespec ts, *tsp = NULL;
534 	int error;
535 
536 	if (SCARG(uap, tv) != NULL) {
537 		struct timeval tv;
538 		if ((error = copyin(SCARG(uap, tv), &tv, sizeof tv)) != 0)
539 			return (error);
540 #ifdef KTRACE
541 		if (KTRPOINT(p, KTR_STRUCT))
542 			ktrreltimeval(p, &tv);
543 #endif
544 		if (tv.tv_sec < 0 || !timerisvalid(&tv))
545 			return (EINVAL);
546 		TIMEVAL_TO_TIMESPEC(&tv, &ts);
547 		tsp = &ts;
548 	}
549 
550 	return (dopselect(p, SCARG(uap, nd), SCARG(uap, in), SCARG(uap, ou),
551 	    SCARG(uap, ex), tsp, NULL, retval));
552 }
553 
554 int
555 sys_pselect(struct proc *p, void *v, register_t *retval)
556 {
557 	struct sys_pselect_args /* {
558 		syscallarg(int) nd;
559 		syscallarg(fd_set *) in;
560 		syscallarg(fd_set *) ou;
561 		syscallarg(fd_set *) ex;
562 		syscallarg(const struct timespec *) ts;
563 		syscallarg(const sigset_t *) mask;
564 	} */ *uap = v;
565 
566 	struct timespec ts, *tsp = NULL;
567 	sigset_t ss, *ssp = NULL;
568 	int error;
569 
570 	if (SCARG(uap, ts) != NULL) {
571 		if ((error = copyin(SCARG(uap, ts), &ts, sizeof ts)) != 0)
572 			return (error);
573 #ifdef KTRACE
574 		if (KTRPOINT(p, KTR_STRUCT))
575 			ktrreltimespec(p, &ts);
576 #endif
577 		if (ts.tv_sec < 0 || !timespecisvalid(&ts))
578 			return (EINVAL);
579 		tsp = &ts;
580 	}
581 	if (SCARG(uap, mask) != NULL) {
582 		if ((error = copyin(SCARG(uap, mask), &ss, sizeof ss)) != 0)
583 			return (error);
584 		ssp = &ss;
585 	}
586 
587 	return (dopselect(p, SCARG(uap, nd), SCARG(uap, in), SCARG(uap, ou),
588 	    SCARG(uap, ex), tsp, ssp, retval));
589 }
590 
591 int
592 dopselect(struct proc *p, int nd, fd_set *in, fd_set *ou, fd_set *ex,
593     struct timespec *timeout, const sigset_t *sigmask, register_t *retval)
594 {
595 	struct kqueue_scan_state scan;
596 	struct timespec zerots = {};
597 	fd_mask bits[6];
598 	fd_set *pibits[3], *pobits[3];
599 	int error, nfiles, ncollected = 0, nevents = 0;
600 	u_int ni;
601 
602 	if (nd < 0)
603 		return (EINVAL);
604 
605 	nfiles = READ_ONCE(p->p_fd->fd_nfiles);
606 	if (nd > nfiles)
607 		nd = nfiles;
608 
609 	ni = howmany(nd, NFDBITS) * sizeof(fd_mask);
610 	if (ni > sizeof(bits[0])) {
611 		caddr_t mbits;
612 
613 		mbits = mallocarray(6, ni, M_TEMP, M_WAITOK|M_ZERO);
614 		pibits[0] = (fd_set *)&mbits[ni * 0];
615 		pibits[1] = (fd_set *)&mbits[ni * 1];
616 		pibits[2] = (fd_set *)&mbits[ni * 2];
617 		pobits[0] = (fd_set *)&mbits[ni * 3];
618 		pobits[1] = (fd_set *)&mbits[ni * 4];
619 		pobits[2] = (fd_set *)&mbits[ni * 5];
620 	} else {
621 		memset(bits, 0, sizeof(bits));
622 		pibits[0] = (fd_set *)&bits[0];
623 		pibits[1] = (fd_set *)&bits[1];
624 		pibits[2] = (fd_set *)&bits[2];
625 		pobits[0] = (fd_set *)&bits[3];
626 		pobits[1] = (fd_set *)&bits[4];
627 		pobits[2] = (fd_set *)&bits[5];
628 	}
629 
630 	kqpoll_init(nd);
631 
632 #define	getbits(name, x) \
633 	if (name && (error = copyin(name, pibits[x], ni))) \
634 		goto done;
635 	getbits(in, 0);
636 	getbits(ou, 1);
637 	getbits(ex, 2);
638 #undef	getbits
639 #ifdef KTRACE
640 	if (ni > 0 && KTRPOINT(p, KTR_STRUCT)) {
641 		if (in) ktrfdset(p, pibits[0], ni);
642 		if (ou) ktrfdset(p, pibits[1], ni);
643 		if (ex) ktrfdset(p, pibits[2], ni);
644 	}
645 #endif
646 
647 	if (sigmask) {
648 		KERNEL_LOCK();
649 		dosigsuspend(p, *sigmask &~ sigcantmask);
650 		KERNEL_UNLOCK();
651 	}
652 
653 	/* Register kqueue events */
654 	error = pselregister(p, pibits, pobits, nd, &nevents, &ncollected);
655 	if (error != 0)
656 		goto done;
657 
658 	/*
659 	 * The poll/select family of syscalls has been designed to
660 	 * block when file descriptors are not available, even if
661 	 * there's nothing to wait for.
662 	 */
663 	if (nevents == 0 && ncollected == 0) {
664 		uint64_t nsecs = INFSLP;
665 
666 		if (timeout != NULL) {
667 			if (!timespecisset(timeout))
668 				goto done;
669 			nsecs = MAX(1, MIN(TIMESPEC_TO_NSEC(timeout), MAXTSLP));
670 		}
671 		error = tsleep_nsec(&nowake, PSOCK | PCATCH, "kqsel", nsecs);
672 		/* select is not restarted after signals... */
673 		if (error == ERESTART)
674 			error = EINTR;
675 		if (error == EWOULDBLOCK)
676 			error = 0;
677 		goto done;
678 	}
679 
680 	/* Do not block if registering found pending events. */
681 	if (ncollected > 0)
682 		timeout = &zerots;
683 
684 	/* Collect at most `nevents' possibly waiting in kqueue_scan() */
685 	kqueue_scan_setup(&scan, p->p_kq);
686 	while (nevents > 0) {
687 		struct kevent kev[KQ_NEVENTS];
688 		int i, ready, count;
689 
690 		/* Maximum number of events per iteration */
691 		count = MIN(nitems(kev), nevents);
692 		ready = kqueue_scan(&scan, count, kev, timeout, p, &error);
693 
694 		/* Convert back events that are ready. */
695 		for (i = 0; i < ready && error == 0; i++)
696 			error = pselcollect(p, &kev[i], pobits, &ncollected);
697 		/*
698 		 * Stop if there was an error or if we had enough
699 		 * space to collect all events that were ready.
700 		 */
701 		if (error || ready < count)
702 			break;
703 
704 		nevents -= ready;
705 	}
706 	kqueue_scan_finish(&scan);
707 	*retval = ncollected;
708 done:
709 #define	putbits(name, x) \
710 	if (name && (error2 = copyout(pobits[x], name, ni))) \
711 		error = error2;
712 	if (error == 0) {
713 		int error2;
714 
715 		putbits(in, 0);
716 		putbits(ou, 1);
717 		putbits(ex, 2);
718 #undef putbits
719 #ifdef KTRACE
720 		if (ni > 0 && KTRPOINT(p, KTR_STRUCT)) {
721 			if (in) ktrfdset(p, pobits[0], ni);
722 			if (ou) ktrfdset(p, pobits[1], ni);
723 			if (ex) ktrfdset(p, pobits[2], ni);
724 		}
725 #endif
726 	}
727 
728 	if (pibits[0] != (fd_set *)&bits[0])
729 		free(pibits[0], M_TEMP, 6 * ni);
730 
731 	kqpoll_done(nd);
732 
733 	return (error);
734 }
735 
736 /*
737  * Convert fd_set into kqueue events and register them on the
738  * per-thread queue.
739  */
740 int
741 pselregister(struct proc *p, fd_set *pibits[3], fd_set *pobits[3], int nfd,
742     int *nregistered, int *ncollected)
743 {
744 	static const int evf[] = { EVFILT_READ, EVFILT_WRITE, EVFILT_EXCEPT };
745 	static const int evff[] = { 0, 0, NOTE_OOB };
746 	int msk, i, j, fd, nevents = 0, error = 0;
747 	struct kevent kev;
748 	fd_mask bits;
749 
750 	for (msk = 0; msk < 3; msk++) {
751 		for (i = 0; i < nfd; i += NFDBITS) {
752 			bits = pibits[msk]->fds_bits[i / NFDBITS];
753 			while ((j = ffs(bits)) && (fd = i + --j) < nfd) {
754 				bits &= ~(1 << j);
755 
756 				DPRINTFN(2, "select fd %d mask %d serial %lu\n",
757 				    fd, msk, p->p_kq_serial);
758 				EV_SET(&kev, fd, evf[msk],
759 				    EV_ADD|EV_ENABLE|__EV_SELECT,
760 				    evff[msk], 0, (void *)(p->p_kq_serial));
761 				error = kqueue_register(p->p_kq, &kev, 0, p);
762 				switch (error) {
763 				case 0:
764 					nevents++;
765 				/* FALLTHROUGH */
766 				case EOPNOTSUPP:/* No underlying kqfilter */
767 				case EINVAL:	/* Unimplemented filter */
768 				case EPERM:	/* Specific to FIFO and
769 						 * __EV_SELECT */
770 					error = 0;
771 					break;
772 				case ENXIO:	/* Device has been detached */
773 				default:
774 					goto bad;
775 				}
776 			}
777 		}
778 	}
779 
780 	*nregistered = nevents;
781 	return (0);
782 bad:
783 	DPRINTFN(0, "select fd %u filt %d error %d\n", (int)kev.ident,
784 	    kev.filter, error);
785 	return (error);
786 }
787 
788 /*
789  * Convert given kqueue event into corresponding select(2) bit.
790  */
791 int
792 pselcollect(struct proc *p, struct kevent *kevp, fd_set *pobits[3],
793     int *ncollected)
794 {
795 	if ((unsigned long)kevp->udata != p->p_kq_serial) {
796 		panic("%s: spurious kevp %p fd %d udata 0x%lx serial 0x%lx",
797 		    __func__, kevp, (int)kevp->ident,
798 		    (unsigned long)kevp->udata, p->p_kq_serial);
799 	}
800 
801 	if (kevp->flags & EV_ERROR) {
802 		DPRINTFN(2, "select fd %d filt %d error %d\n",
803 		    (int)kevp->ident, kevp->filter, (int)kevp->data);
804 		return (kevp->data);
805 	}
806 
807 	switch (kevp->filter) {
808 	case EVFILT_READ:
809 		FD_SET(kevp->ident, pobits[0]);
810 		break;
811 	case EVFILT_WRITE:
812 		FD_SET(kevp->ident, pobits[1]);
813 		break;
814 	case EVFILT_EXCEPT:
815 		FD_SET(kevp->ident, pobits[2]);
816 		break;
817 	default:
818 		KASSERT(0);
819 	}
820 	(*ncollected)++;
821 
822 	DPRINTFN(2, "select fd %d filt %d\n", (int)kevp->ident, kevp->filter);
823 	return (0);
824 }
825 
826 /*
827  * Do a wakeup when a selectable event occurs.
828  */
829 void
830 selwakeup(struct selinfo *sip)
831 {
832 	KERNEL_LOCK();
833 	knote_locked(&sip->si_note, NOTE_SUBMIT);
834 	KERNEL_UNLOCK();
835 }
836 
837 /*
838  * Only copyout the revents field.
839  */
840 int
841 pollout(struct pollfd *pl, struct pollfd *upl, u_int nfds)
842 {
843 	int error = 0;
844 	u_int i = 0;
845 
846 	while (!error && i++ < nfds) {
847 		error = copyout(&pl->revents, &upl->revents,
848 		    sizeof(upl->revents));
849 		pl++;
850 		upl++;
851 	}
852 
853 	return (error);
854 }
855 
856 /*
857  * We are using the same mechanism as select only we encode/decode args
858  * differently.
859  */
860 int
861 sys_poll(struct proc *p, void *v, register_t *retval)
862 {
863 	struct sys_poll_args /* {
864 		syscallarg(struct pollfd *) fds;
865 		syscallarg(u_int) nfds;
866 		syscallarg(int) timeout;
867 	} */ *uap = v;
868 
869 	struct timespec ts, *tsp = NULL;
870 	int msec = SCARG(uap, timeout);
871 
872 	if (msec != INFTIM) {
873 		if (msec < 0)
874 			return (EINVAL);
875 		ts.tv_sec = msec / 1000;
876 		ts.tv_nsec = (msec - (ts.tv_sec * 1000)) * 1000000;
877 		tsp = &ts;
878 	}
879 
880 	return (doppoll(p, SCARG(uap, fds), SCARG(uap, nfds), tsp, NULL,
881 	    retval));
882 }
883 
884 int
885 sys_ppoll(struct proc *p, void *v, register_t *retval)
886 {
887 	struct sys_ppoll_args /* {
888 		syscallarg(struct pollfd *) fds;
889 		syscallarg(u_int) nfds;
890 		syscallarg(const struct timespec *) ts;
891 		syscallarg(const sigset_t *) mask;
892 	} */ *uap = v;
893 
894 	int error;
895 	struct timespec ts, *tsp = NULL;
896 	sigset_t ss, *ssp = NULL;
897 
898 	if (SCARG(uap, ts) != NULL) {
899 		if ((error = copyin(SCARG(uap, ts), &ts, sizeof ts)) != 0)
900 			return (error);
901 #ifdef KTRACE
902 		if (KTRPOINT(p, KTR_STRUCT))
903 			ktrreltimespec(p, &ts);
904 #endif
905 		if (ts.tv_sec < 0 || !timespecisvalid(&ts))
906 			return (EINVAL);
907 		tsp = &ts;
908 	}
909 
910 	if (SCARG(uap, mask) != NULL) {
911 		if ((error = copyin(SCARG(uap, mask), &ss, sizeof ss)) != 0)
912 			return (error);
913 		ssp = &ss;
914 	}
915 
916 	return (doppoll(p, SCARG(uap, fds), SCARG(uap, nfds), tsp, ssp,
917 	    retval));
918 }
919 
920 int
921 doppoll(struct proc *p, struct pollfd *fds, u_int nfds,
922     struct timespec *timeout, const sigset_t *sigmask, register_t *retval)
923 {
924 	struct kqueue_scan_state scan;
925 	struct timespec zerots = {};
926 	struct pollfd pfds[4], *pl = pfds;
927 	int error, ncollected = 0, nevents = 0;
928 	size_t sz;
929 
930 	/* Standards say no more than MAX_OPEN; this is possibly better. */
931 	if (nfds > min((int)lim_cur(RLIMIT_NOFILE), maxfiles))
932 		return (EINVAL);
933 
934 	/* optimize for the default case, of a small nfds value */
935 	if (nfds > nitems(pfds)) {
936 		pl = mallocarray(nfds, sizeof(*pl), M_TEMP,
937 		    M_WAITOK | M_CANFAIL);
938 		if (pl == NULL)
939 			return (EINVAL);
940 	}
941 
942 	kqpoll_init(nfds);
943 
944 	sz = nfds * sizeof(*pl);
945 
946 	if ((error = copyin(fds, pl, sz)) != 0)
947 		goto bad;
948 
949 	if (sigmask) {
950 		KERNEL_LOCK();
951 		dosigsuspend(p, *sigmask &~ sigcantmask);
952 		KERNEL_UNLOCK();
953 	}
954 
955 	/* Register kqueue events */
956 	ppollregister(p, pl, nfds, &nevents, &ncollected);
957 
958 	/*
959 	 * The poll/select family of syscalls has been designed to
960 	 * block when file descriptors are not available, even if
961 	 * there's nothing to wait for.
962 	 */
963 	if (nevents == 0 && ncollected == 0) {
964 		uint64_t nsecs = INFSLP;
965 
966 		if (timeout != NULL) {
967 			if (!timespecisset(timeout))
968 				goto done;
969 			nsecs = MAX(1, MIN(TIMESPEC_TO_NSEC(timeout), MAXTSLP));
970 		}
971 
972 		error = tsleep_nsec(&nowake, PSOCK | PCATCH, "kqpoll", nsecs);
973 		if (error == ERESTART)
974 			error = EINTR;
975 		if (error == EWOULDBLOCK)
976 			error = 0;
977 		goto done;
978 	}
979 
980 	/* Do not block if registering found pending events. */
981 	if (ncollected > 0)
982 		timeout = &zerots;
983 
984 	/* Collect at most `nevents' possibly waiting in kqueue_scan() */
985 	kqueue_scan_setup(&scan, p->p_kq);
986 	while (nevents > 0) {
987 		struct kevent kev[KQ_NEVENTS];
988 		int i, ready, count;
989 
990 		/* Maximum number of events per iteration */
991 		count = MIN(nitems(kev), nevents);
992 		ready = kqueue_scan(&scan, count, kev, timeout, p, &error);
993 
994 		/* Convert back events that are ready. */
995 		for (i = 0; i < ready; i++)
996 			ncollected += ppollcollect(p, &kev[i], pl, nfds);
997 
998 		/*
999 		 * Stop if there was an error or if we had enough
1000 		 * place to collect all events that were ready.
1001 		 */
1002 		if (error || ready < count)
1003 			break;
1004 
1005 		nevents -= ready;
1006 	}
1007 	kqueue_scan_finish(&scan);
1008 	*retval = ncollected;
1009 done:
1010 	/*
1011 	 * NOTE: poll(2) is not restarted after a signal and EWOULDBLOCK is
1012 	 *       ignored (since the whole point is to see what would block).
1013 	 */
1014 	switch (error) {
1015 	case EINTR:
1016 		error = pollout(pl, fds, nfds);
1017 		if (error == 0)
1018 			error = EINTR;
1019 		break;
1020 	case EWOULDBLOCK:
1021 	case 0:
1022 		error = pollout(pl, fds, nfds);
1023 		break;
1024 	}
1025 #ifdef KTRACE
1026 	if (KTRPOINT(p, KTR_STRUCT))
1027 		ktrpollfd(p, pl, nfds);
1028 #endif /* KTRACE */
1029 bad:
1030 	if (pl != pfds)
1031 		free(pl, M_TEMP, sz);
1032 
1033 	kqpoll_done(nfds);
1034 
1035 	return (error);
1036 }
1037 
1038 int
1039 ppollregister_evts(struct proc *p, struct kevent *kevp, int nkev,
1040     struct pollfd *pl, unsigned int pollid)
1041 {
1042 	int i, error, nevents = 0;
1043 
1044 	KASSERT(pl->revents == 0);
1045 
1046 	for (i = 0; i < nkev; i++, kevp++) {
1047 again:
1048 		error = kqueue_register(p->p_kq, kevp, pollid, p);
1049 		switch (error) {
1050 		case 0:
1051 			nevents++;
1052 			break;
1053 		case EOPNOTSUPP:/* No underlying kqfilter */
1054 		case EINVAL:	/* Unimplemented filter */
1055 			break;
1056 		case EBADF:	/* Bad file descriptor */
1057 			pl->revents |= POLLNVAL;
1058 			break;
1059 		case EPERM:	/* Specific to FIFO */
1060 			KASSERT(kevp->filter == EVFILT_WRITE);
1061 			if (nkev == 1) {
1062 				/*
1063 				 * If this is the only filter make sure
1064 				 * POLLHUP is passed to userland.
1065 				 */
1066 				kevp->filter = EVFILT_EXCEPT;
1067 				goto again;
1068 			}
1069 			break;
1070 		default:
1071 			DPRINTFN(0, "poll err %lu fd %d revents %02x serial"
1072 			    " %lu filt %d ERROR=%d\n",
1073 			    ((unsigned long)kevp->udata - p->p_kq_serial),
1074 			    pl->fd, pl->revents, p->p_kq_serial, kevp->filter,
1075 			    error);
1076 			/* FALLTHROUGH */
1077 		case ENXIO:	/* Device has been detached */
1078 			pl->revents |= POLLERR;
1079 			break;
1080 		}
1081 	}
1082 
1083 	return (nevents);
1084 }
1085 
1086 /*
1087  * Convert pollfd into kqueue events and register them on the
1088  * per-thread queue.
1089  *
1090  * At most 3 events can correspond to a single pollfd.
1091  */
1092 void
1093 ppollregister(struct proc *p, struct pollfd *pl, int nfds, int *nregistered,
1094     int *ncollected)
1095 {
1096 	int i, nkev, nevt, forcehup;
1097 	struct kevent kev[3], *kevp;
1098 
1099 	for (i = 0; i < nfds; i++) {
1100 		pl[i].events &= ~POLL_NOHUP;
1101 		pl[i].revents = 0;
1102 
1103 		if (pl[i].fd < 0)
1104 			continue;
1105 
1106 		/*
1107 		 * POLLHUP checking is implicit in the event filters.
1108 		 * However, the checking must be even if no events are
1109 		 * requested.
1110 		 */
1111 		forcehup = ((pl[i].events & ~POLLHUP) == 0);
1112 
1113 		DPRINTFN(1, "poll set %d/%d fd %d events %02x serial %lu\n",
1114 		    i+1, nfds, pl[i].fd, pl[i].events, p->p_kq_serial);
1115 
1116 		nevt = 0;
1117 		nkev = 0;
1118 		kevp = kev;
1119 		if (pl[i].events & (POLLIN | POLLRDNORM)) {
1120 			EV_SET(kevp, pl[i].fd, EVFILT_READ,
1121 			    EV_ADD|EV_ENABLE|__EV_POLL, 0, 0,
1122 			    (void *)(p->p_kq_serial + i));
1123 			nkev++;
1124 			kevp++;
1125 		}
1126 		if (pl[i].events & (POLLOUT | POLLWRNORM)) {
1127 			EV_SET(kevp, pl[i].fd, EVFILT_WRITE,
1128 			    EV_ADD|EV_ENABLE|__EV_POLL, 0, 0,
1129 			    (void *)(p->p_kq_serial + i));
1130 			nkev++;
1131 			kevp++;
1132 		}
1133 		if ((pl[i].events & (POLLPRI | POLLRDBAND)) || forcehup) {
1134 			int evff = forcehup ? 0 : NOTE_OOB;
1135 
1136 			EV_SET(kevp, pl[i].fd, EVFILT_EXCEPT,
1137 			    EV_ADD|EV_ENABLE|__EV_POLL, evff, 0,
1138 			    (void *)(p->p_kq_serial + i));
1139 			nkev++;
1140 			kevp++;
1141 		}
1142 
1143 		if (nkev == 0)
1144 			continue;
1145 
1146 		*nregistered += ppollregister_evts(p, kev, nkev, &pl[i], i);
1147 
1148 		if (pl[i].revents != 0)
1149 			(*ncollected)++;
1150 	}
1151 
1152 	DPRINTFN(1, "poll registered = %d, collected = %d\n", *nregistered,
1153 	    *ncollected);
1154 }
1155 
1156 /*
1157  * Convert given kqueue event into corresponding poll(2) revents bit.
1158  */
1159 int
1160 ppollcollect(struct proc *p, struct kevent *kevp, struct pollfd *pl, u_int nfds)
1161 {
1162 	static struct timeval poll_errintvl = { 5, 0 };
1163 	static struct timeval poll_lasterr;
1164 	int already_seen;
1165 	unsigned long i;
1166 
1167 	/*  Extract poll array index */
1168 	i = (unsigned long)kevp->udata - p->p_kq_serial;
1169 
1170 	if (i >= nfds) {
1171 		panic("%s: spurious kevp %p nfds %u udata 0x%lx serial 0x%lx",
1172 		    __func__, kevp, nfds,
1173 		    (unsigned long)kevp->udata, p->p_kq_serial);
1174 	}
1175 	if ((int)kevp->ident != pl[i].fd) {
1176 		panic("%s: kevp %p %lu/%d mismatch fd %d!=%d serial 0x%lx",
1177 		    __func__, kevp, i + 1, nfds, (int)kevp->ident, pl[i].fd,
1178 		    p->p_kq_serial);
1179 	}
1180 
1181 	/*
1182 	 * A given descriptor may already have generated an error
1183 	 * against another filter during kqueue_register().
1184 	 *
1185 	 * Make sure to set the appropriate flags but do not
1186 	 * increment `*retval' more than once.
1187 	 */
1188 	already_seen = (pl[i].revents != 0);
1189 
1190 	/* POLLNVAL preempts other events. */
1191 	if ((kevp->flags & EV_ERROR) && kevp->data == EBADF) {
1192 		pl[i].revents = POLLNVAL;
1193 		goto done;
1194 	} else if (pl[i].revents & POLLNVAL) {
1195 		goto done;
1196 	}
1197 
1198 	switch (kevp->filter) {
1199 	case EVFILT_READ:
1200 		if (kevp->flags & __EV_HUP)
1201 			pl[i].revents |= POLLHUP;
1202 		if (pl[i].events & (POLLIN | POLLRDNORM))
1203 			pl[i].revents |= pl[i].events & (POLLIN | POLLRDNORM);
1204 		break;
1205 	case EVFILT_WRITE:
1206 		/* POLLHUP and POLLOUT/POLLWRNORM are mutually exclusive */
1207 		if (kevp->flags & __EV_HUP) {
1208 			pl[i].revents |= POLLHUP;
1209 		} else if (pl[i].events & (POLLOUT | POLLWRNORM)) {
1210 			pl[i].revents |= pl[i].events & (POLLOUT | POLLWRNORM);
1211 		}
1212 		break;
1213 	case EVFILT_EXCEPT:
1214 		if (kevp->flags & __EV_HUP) {
1215 			if (pl[i].events != 0 && pl[i].events != POLLOUT)
1216 				DPRINTFN(0, "weird events %x\n", pl[i].events);
1217 			pl[i].revents |= POLLHUP;
1218 			break;
1219 		}
1220 		if (pl[i].events & (POLLPRI | POLLRDBAND))
1221 			pl[i].revents |= pl[i].events & (POLLPRI | POLLRDBAND);
1222 		break;
1223 	default:
1224 		KASSERT(0);
1225 	}
1226 
1227 done:
1228 	DPRINTFN(1, "poll get %lu/%d fd %d revents %02x serial %lu filt %d\n",
1229 	    i+1, nfds, pl[i].fd, pl[i].revents, (unsigned long)kevp->udata,
1230 	    kevp->filter);
1231 
1232 	/*
1233 	 * Make noise about unclaimed events as they might indicate a bug
1234 	 * and can result in spurious-looking wakeups of poll(2).
1235 	 *
1236 	 * Live-locking within the system call should not happen because
1237 	 * the scan loop in doppoll() has an upper limit for the number
1238 	 * of events to process.
1239 	 */
1240 	if (pl[i].revents == 0 && ratecheck(&poll_lasterr, &poll_errintvl)) {
1241 		printf("%s[%d]: poll index %lu fd %d events 0x%x "
1242 		    "filter %d/0x%x unclaimed\n",
1243 		    p->p_p->ps_comm, p->p_tid, i, pl[i].fd,
1244 		    pl[i].events, kevp->filter, kevp->flags);
1245 	}
1246 
1247 	if (!already_seen && (pl[i].revents != 0))
1248 		return (1);
1249 
1250 	return (0);
1251 }
1252 
1253 /*
1254  * utrace system call
1255  */
1256 int
1257 sys_utrace(struct proc *curp, void *v, register_t *retval)
1258 {
1259 #ifdef KTRACE
1260 	struct sys_utrace_args /* {
1261 		syscallarg(const char *) label;
1262 		syscallarg(const void *) addr;
1263 		syscallarg(size_t) len;
1264 	} */ *uap = v;
1265 
1266 	return (ktruser(curp, SCARG(uap, label), SCARG(uap, addr),
1267 	    SCARG(uap, len)));
1268 #else
1269 	return (0);
1270 #endif
1271 }
1272