xref: /openbsd-src/sys/kern/sys_generic.c (revision 25c4e8bd056e974b28f4a0ffd39d76c190a56013)
1 /*	$OpenBSD: sys_generic.c,v 1.148 2022/07/05 15:06:16 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/kernel.h>
54 #include <sys/stat.h>
55 #include <sys/time.h>
56 #include <sys/malloc.h>
57 #include <sys/poll.h>
58 #include <sys/eventvar.h>
59 #ifdef KTRACE
60 #include <sys/ktrace.h>
61 #endif
62 #include <sys/sched.h>
63 #include <sys/pledge.h>
64 
65 #include <sys/mount.h>
66 #include <sys/syscallargs.h>
67 
68 #include <uvm/uvm_extern.h>
69 
70 /*
71  * Debug values:
72  *  1 - print implementation errors, things that should not happen.
73  *  2 - print ppoll(2) information, somewhat verbose
74  *  3 - print pselect(2) and ppoll(2) information, very verbose
75  */
76 int kqpoll_debug = 0;
77 #define DPRINTFN(v, x...) if (kqpoll_debug > v) {			\
78 	printf("%s(%d): ", curproc->p_p->ps_comm, curproc->p_tid);	\
79 	printf(x);							\
80 }
81 
82 int pselregister(struct proc *, fd_set *[], fd_set *[], int, int *, int *);
83 int pselcollect(struct proc *, struct kevent *, fd_set *[], int *);
84 void ppollregister(struct proc *, struct pollfd *, int, int *, int *);
85 int ppollcollect(struct proc *, struct kevent *, struct pollfd *, u_int);
86 
87 int pollout(struct pollfd *, struct pollfd *, u_int);
88 int dopselect(struct proc *, int, fd_set *, fd_set *, fd_set *,
89     struct timespec *, const sigset_t *, register_t *);
90 int doppoll(struct proc *, struct pollfd *, u_int, struct timespec *,
91     const sigset_t *, register_t *);
92 
93 int
94 iovec_copyin(const struct iovec *uiov, struct iovec **iovp, struct iovec *aiov,
95     unsigned int iovcnt, size_t *residp)
96 {
97 #ifdef KTRACE
98 	struct proc *p = curproc;
99 #endif
100 	struct iovec *iov;
101 	int error, i;
102 	size_t resid = 0;
103 
104 	if (iovcnt > UIO_SMALLIOV) {
105 		if (iovcnt > IOV_MAX)
106 			return (EINVAL);
107 		iov = mallocarray(iovcnt, sizeof(*iov), M_IOV, M_WAITOK);
108 	} else if (iovcnt > 0) {
109 		iov = aiov;
110 	} else {
111 		return (EINVAL);
112 	}
113 	*iovp = iov;
114 
115 	if ((error = copyin(uiov, iov, iovcnt * sizeof(*iov))))
116 		return (error);
117 
118 #ifdef KTRACE
119 	if (KTRPOINT(p, KTR_STRUCT))
120 		ktriovec(p, iov, iovcnt);
121 #endif
122 
123 	for (i = 0; i < iovcnt; i++) {
124 		resid += iov->iov_len;
125 		/*
126 		 * Writes return ssize_t because -1 is returned on error.
127 		 * Therefore we must restrict the length to SSIZE_MAX to
128 		 * avoid garbage return values.  Note that the addition is
129 		 * guaranteed to not wrap because SSIZE_MAX * 2 < SIZE_MAX.
130 		 */
131 		if (iov->iov_len > SSIZE_MAX || resid > SSIZE_MAX)
132 			return (EINVAL);
133 		iov++;
134 	}
135 
136 	if (residp != NULL)
137 		*residp = resid;
138 
139 	return (0);
140 }
141 
142 void
143 iovec_free(struct iovec *iov, unsigned int iovcnt)
144 {
145 	if (iovcnt > UIO_SMALLIOV)
146 		free(iov, M_IOV, iovcnt * sizeof(*iov));
147 }
148 
149 /*
150  * Read system call.
151  */
152 int
153 sys_read(struct proc *p, void *v, register_t *retval)
154 {
155 	struct sys_read_args /* {
156 		syscallarg(int) fd;
157 		syscallarg(void *) buf;
158 		syscallarg(size_t) nbyte;
159 	} */ *uap = v;
160 	struct iovec iov;
161 	struct uio auio;
162 
163 	iov.iov_base = SCARG(uap, buf);
164 	iov.iov_len = SCARG(uap, nbyte);
165 	if (iov.iov_len > SSIZE_MAX)
166 		return (EINVAL);
167 
168 	auio.uio_iov = &iov;
169 	auio.uio_iovcnt = 1;
170 	auio.uio_resid = iov.iov_len;
171 
172 	return (dofilereadv(p, SCARG(uap, fd), &auio, 0, retval));
173 }
174 
175 /*
176  * Scatter read system call.
177  */
178 int
179 sys_readv(struct proc *p, void *v, register_t *retval)
180 {
181 	struct sys_readv_args /* {
182 		syscallarg(int) fd;
183 		syscallarg(const struct iovec *) iovp;
184 		syscallarg(int) iovcnt;
185 	} */ *uap = v;
186 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
187 	int error, iovcnt = SCARG(uap, iovcnt);
188 	struct uio auio;
189 	size_t resid;
190 
191 	error = iovec_copyin(SCARG(uap, iovp), &iov, aiov, iovcnt, &resid);
192 	if (error)
193 		goto done;
194 
195 	auio.uio_iov = iov;
196 	auio.uio_iovcnt = iovcnt;
197 	auio.uio_resid = resid;
198 
199 	error = dofilereadv(p, SCARG(uap, fd), &auio, 0, retval);
200  done:
201 	iovec_free(iov, iovcnt);
202 	return (error);
203 }
204 
205 int
206 dofilereadv(struct proc *p, int fd, struct uio *uio, int flags,
207     register_t *retval)
208 {
209 	struct filedesc *fdp = p->p_fd;
210 	struct file *fp;
211 	long cnt, error = 0;
212 	u_int iovlen;
213 #ifdef KTRACE
214 	struct iovec *ktriov = NULL;
215 #endif
216 
217 	KASSERT(uio->uio_iov != NULL && uio->uio_iovcnt > 0);
218 	iovlen = uio->uio_iovcnt * sizeof(struct iovec);
219 
220 	if ((fp = fd_getfile_mode(fdp, fd, FREAD)) == NULL)
221 		return (EBADF);
222 
223 	/* Checks for positioned read. */
224 	if (flags & FO_POSITION) {
225 		struct vnode *vp = fp->f_data;
226 
227 		if (fp->f_type != DTYPE_VNODE || vp->v_type == VFIFO ||
228 		    (vp->v_flag & VISTTY)) {
229 			error = ESPIPE;
230 			goto done;
231 		}
232 
233 		if (uio->uio_offset < 0 && vp->v_type != VCHR) {
234 			error = EINVAL;
235 			goto done;
236 		}
237 	}
238 
239 	uio->uio_rw = UIO_READ;
240 	uio->uio_segflg = UIO_USERSPACE;
241 	uio->uio_procp = p;
242 #ifdef KTRACE
243 	/*
244 	 * if tracing, save a copy of iovec
245 	 */
246 	if (KTRPOINT(p, KTR_GENIO)) {
247 		ktriov = malloc(iovlen, M_TEMP, M_WAITOK);
248 		memcpy(ktriov, uio->uio_iov, iovlen);
249 	}
250 #endif
251 	cnt = uio->uio_resid;
252 	error = (*fp->f_ops->fo_read)(fp, uio, flags);
253 	if (error) {
254 		if (uio->uio_resid != cnt && (error == ERESTART ||
255 		    error == EINTR || error == EWOULDBLOCK))
256 			error = 0;
257 	}
258 	cnt -= uio->uio_resid;
259 
260 	mtx_enter(&fp->f_mtx);
261 	fp->f_rxfer++;
262 	fp->f_rbytes += cnt;
263 	mtx_leave(&fp->f_mtx);
264 #ifdef KTRACE
265 	if (ktriov != NULL) {
266 		if (error == 0)
267 			ktrgenio(p, fd, UIO_READ, ktriov, cnt);
268 		free(ktriov, M_TEMP, iovlen);
269 	}
270 #endif
271 	*retval = cnt;
272  done:
273 	FRELE(fp, p);
274 	return (error);
275 }
276 
277 /*
278  * Write system call
279  */
280 int
281 sys_write(struct proc *p, void *v, register_t *retval)
282 {
283 	struct sys_write_args /* {
284 		syscallarg(int) fd;
285 		syscallarg(const void *) buf;
286 		syscallarg(size_t) nbyte;
287 	} */ *uap = v;
288 	struct iovec iov;
289 	struct uio auio;
290 
291 	iov.iov_base = (void *)SCARG(uap, buf);
292 	iov.iov_len = SCARG(uap, nbyte);
293 	if (iov.iov_len > SSIZE_MAX)
294 		return (EINVAL);
295 
296 	auio.uio_iov = &iov;
297 	auio.uio_iovcnt = 1;
298 	auio.uio_resid = iov.iov_len;
299 
300 	return (dofilewritev(p, SCARG(uap, fd), &auio, 0, retval));
301 }
302 
303 /*
304  * Gather write system call
305  */
306 int
307 sys_writev(struct proc *p, void *v, register_t *retval)
308 {
309 	struct sys_writev_args /* {
310 		syscallarg(int) fd;
311 		syscallarg(const struct iovec *) iovp;
312 		syscallarg(int) iovcnt;
313 	} */ *uap = v;
314 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
315 	int error, iovcnt = SCARG(uap, iovcnt);
316 	struct uio auio;
317 	size_t resid;
318 
319 	error = iovec_copyin(SCARG(uap, iovp), &iov, aiov, iovcnt, &resid);
320 	if (error)
321 		goto done;
322 
323 	auio.uio_iov = iov;
324 	auio.uio_iovcnt = iovcnt;
325 	auio.uio_resid = resid;
326 
327 	error = dofilewritev(p, SCARG(uap, fd), &auio, 0, retval);
328  done:
329 	iovec_free(iov, iovcnt);
330  	return (error);
331 }
332 
333 int
334 dofilewritev(struct proc *p, int fd, struct uio *uio, int flags,
335     register_t *retval)
336 {
337 	struct filedesc *fdp = p->p_fd;
338 	struct file *fp;
339 	long cnt, error = 0;
340 	u_int iovlen;
341 #ifdef KTRACE
342 	struct iovec *ktriov = NULL;
343 #endif
344 
345 	KASSERT(uio->uio_iov != NULL && uio->uio_iovcnt > 0);
346 	iovlen = uio->uio_iovcnt * sizeof(struct iovec);
347 
348 	if ((fp = fd_getfile_mode(fdp, fd, FWRITE)) == NULL)
349 		return (EBADF);
350 
351 	/* Checks for positioned write. */
352 	if (flags & FO_POSITION) {
353 		struct vnode *vp = fp->f_data;
354 
355 		if (fp->f_type != DTYPE_VNODE || vp->v_type == VFIFO ||
356 		    (vp->v_flag & VISTTY)) {
357 			error = ESPIPE;
358 			goto done;
359 		}
360 
361 		if (uio->uio_offset < 0 && vp->v_type != VCHR) {
362 			error = EINVAL;
363 			goto done;
364 		}
365 	}
366 
367 	uio->uio_rw = UIO_WRITE;
368 	uio->uio_segflg = UIO_USERSPACE;
369 	uio->uio_procp = p;
370 #ifdef KTRACE
371 	/*
372 	 * if tracing, save a copy of iovec
373 	 */
374 	if (KTRPOINT(p, KTR_GENIO)) {
375 		ktriov = malloc(iovlen, M_TEMP, M_WAITOK);
376 		memcpy(ktriov, uio->uio_iov, iovlen);
377 	}
378 #endif
379 	cnt = uio->uio_resid;
380 	error = (*fp->f_ops->fo_write)(fp, uio, flags);
381 	if (error) {
382 		if (uio->uio_resid != cnt && (error == ERESTART ||
383 		    error == EINTR || error == EWOULDBLOCK))
384 			error = 0;
385 		if (error == EPIPE) {
386 			KERNEL_LOCK();
387 			ptsignal(p, SIGPIPE, STHREAD);
388 			KERNEL_UNLOCK();
389 		}
390 	}
391 	cnt -= uio->uio_resid;
392 
393 	mtx_enter(&fp->f_mtx);
394 	fp->f_wxfer++;
395 	fp->f_wbytes += cnt;
396 	mtx_leave(&fp->f_mtx);
397 #ifdef KTRACE
398 	if (ktriov != NULL) {
399 		if (error == 0)
400 			ktrgenio(p, fd, UIO_WRITE, ktriov, cnt);
401 		free(ktriov, M_TEMP, iovlen);
402 	}
403 #endif
404 	*retval = cnt;
405  done:
406 	FRELE(fp, p);
407 	return (error);
408 }
409 
410 /*
411  * Ioctl system call
412  */
413 int
414 sys_ioctl(struct proc *p, void *v, register_t *retval)
415 {
416 	struct sys_ioctl_args /* {
417 		syscallarg(int) fd;
418 		syscallarg(u_long) com;
419 		syscallarg(void *) data;
420 	} */ *uap = v;
421 	struct file *fp;
422 	struct filedesc *fdp = p->p_fd;
423 	u_long com = SCARG(uap, com);
424 	int error = 0;
425 	u_int size = 0;
426 	caddr_t data, memp = NULL;
427 	int tmp;
428 #define STK_PARAMS	128
429 	long long stkbuf[STK_PARAMS / sizeof(long long)];
430 
431 	if ((fp = fd_getfile_mode(fdp, SCARG(uap, fd), FREAD|FWRITE)) == NULL)
432 		return (EBADF);
433 
434 	if (fp->f_type == DTYPE_SOCKET) {
435 		struct socket *so = fp->f_data;
436 
437 		if (so->so_state & SS_DNS) {
438 			error = EINVAL;
439 			goto out;
440 		}
441 	}
442 
443 	error = pledge_ioctl(p, com, fp);
444 	if (error)
445 		goto out;
446 
447 	switch (com) {
448 	case FIONCLEX:
449 	case FIOCLEX:
450 		fdplock(fdp);
451 		if (com == FIONCLEX)
452 			fdp->fd_ofileflags[SCARG(uap, fd)] &= ~UF_EXCLOSE;
453 		else
454 			fdp->fd_ofileflags[SCARG(uap, fd)] |= UF_EXCLOSE;
455 		fdpunlock(fdp);
456 		goto out;
457 	}
458 
459 	/*
460 	 * Interpret high order word to find amount of data to be
461 	 * copied to/from the user's address space.
462 	 */
463 	size = IOCPARM_LEN(com);
464 	if (size > IOCPARM_MAX) {
465 		error = ENOTTY;
466 		goto out;
467 	}
468 	if (size > sizeof (stkbuf)) {
469 		memp = malloc(size, M_IOCTLOPS, M_WAITOK);
470 		data = memp;
471 	} else
472 		data = (caddr_t)stkbuf;
473 	if (com&IOC_IN) {
474 		if (size) {
475 			error = copyin(SCARG(uap, data), data, size);
476 			if (error) {
477 				goto out;
478 			}
479 		} else
480 			*(caddr_t *)data = SCARG(uap, data);
481 	} else if ((com&IOC_OUT) && size)
482 		/*
483 		 * Zero the buffer so the user always
484 		 * gets back something deterministic.
485 		 */
486 		memset(data, 0, size);
487 	else if (com&IOC_VOID)
488 		*(caddr_t *)data = SCARG(uap, data);
489 
490 	switch (com) {
491 
492 	case FIONBIO:
493 		if ((tmp = *(int *)data) != 0)
494 			atomic_setbits_int(&fp->f_flag, FNONBLOCK);
495 		else
496 			atomic_clearbits_int(&fp->f_flag, FNONBLOCK);
497 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
498 		break;
499 
500 	case FIOASYNC:
501 		if ((tmp = *(int *)data) != 0)
502 			atomic_setbits_int(&fp->f_flag, FASYNC);
503 		else
504 			atomic_clearbits_int(&fp->f_flag, FASYNC);
505 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p);
506 		break;
507 
508 	default:
509 		error = (*fp->f_ops->fo_ioctl)(fp, com, data, p);
510 		break;
511 	}
512 	/*
513 	 * Copy any data to user, size was
514 	 * already set and checked above.
515 	 */
516 	if (error == 0 && (com&IOC_OUT) && size)
517 		error = copyout(data, SCARG(uap, data), size);
518 out:
519 	FRELE(fp, p);
520 	free(memp, M_IOCTLOPS, size);
521 	return (error);
522 }
523 
524 /*
525  * Select system call.
526  */
527 int
528 sys_select(struct proc *p, void *v, register_t *retval)
529 {
530 	struct sys_select_args /* {
531 		syscallarg(int) nd;
532 		syscallarg(fd_set *) in;
533 		syscallarg(fd_set *) ou;
534 		syscallarg(fd_set *) ex;
535 		syscallarg(struct timeval *) tv;
536 	} */ *uap = v;
537 
538 	struct timespec ts, *tsp = NULL;
539 	int error;
540 
541 	if (SCARG(uap, tv) != NULL) {
542 		struct timeval tv;
543 		if ((error = copyin(SCARG(uap, tv), &tv, sizeof tv)) != 0)
544 			return (error);
545 #ifdef KTRACE
546 		if (KTRPOINT(p, KTR_STRUCT))
547 			ktrreltimeval(p, &tv);
548 #endif
549 		if (tv.tv_sec < 0 || !timerisvalid(&tv))
550 			return (EINVAL);
551 		TIMEVAL_TO_TIMESPEC(&tv, &ts);
552 		tsp = &ts;
553 	}
554 
555 	return (dopselect(p, SCARG(uap, nd), SCARG(uap, in), SCARG(uap, ou),
556 	    SCARG(uap, ex), tsp, NULL, retval));
557 }
558 
559 int
560 sys_pselect(struct proc *p, void *v, register_t *retval)
561 {
562 	struct sys_pselect_args /* {
563 		syscallarg(int) nd;
564 		syscallarg(fd_set *) in;
565 		syscallarg(fd_set *) ou;
566 		syscallarg(fd_set *) ex;
567 		syscallarg(const struct timespec *) ts;
568 		syscallarg(const sigset_t *) mask;
569 	} */ *uap = v;
570 
571 	struct timespec ts, *tsp = NULL;
572 	sigset_t ss, *ssp = NULL;
573 	int error;
574 
575 	if (SCARG(uap, ts) != NULL) {
576 		if ((error = copyin(SCARG(uap, ts), &ts, sizeof ts)) != 0)
577 			return (error);
578 #ifdef KTRACE
579 		if (KTRPOINT(p, KTR_STRUCT))
580 			ktrreltimespec(p, &ts);
581 #endif
582 		if (ts.tv_sec < 0 || !timespecisvalid(&ts))
583 			return (EINVAL);
584 		tsp = &ts;
585 	}
586 	if (SCARG(uap, mask) != NULL) {
587 		if ((error = copyin(SCARG(uap, mask), &ss, sizeof ss)) != 0)
588 			return (error);
589 		ssp = &ss;
590 	}
591 
592 	return (dopselect(p, SCARG(uap, nd), SCARG(uap, in), SCARG(uap, ou),
593 	    SCARG(uap, ex), tsp, ssp, retval));
594 }
595 
596 int
597 dopselect(struct proc *p, int nd, fd_set *in, fd_set *ou, fd_set *ex,
598     struct timespec *timeout, const sigset_t *sigmask, register_t *retval)
599 {
600 	struct kqueue_scan_state scan;
601 	struct timespec zerots = {};
602 	fd_mask bits[6];
603 	fd_set *pibits[3], *pobits[3];
604 	int error, ncollected = 0, nevents = 0;
605 	u_int ni;
606 
607 	if (nd < 0)
608 		return (EINVAL);
609 	if (nd > p->p_fd->fd_nfiles) {
610 		/* forgiving; slightly wrong */
611 		nd = p->p_fd->fd_nfiles;
612 	}
613 	ni = howmany(nd, NFDBITS) * sizeof(fd_mask);
614 	if (ni > sizeof(bits[0])) {
615 		caddr_t mbits;
616 
617 		mbits = mallocarray(6, ni, M_TEMP, M_WAITOK|M_ZERO);
618 		pibits[0] = (fd_set *)&mbits[ni * 0];
619 		pibits[1] = (fd_set *)&mbits[ni * 1];
620 		pibits[2] = (fd_set *)&mbits[ni * 2];
621 		pobits[0] = (fd_set *)&mbits[ni * 3];
622 		pobits[1] = (fd_set *)&mbits[ni * 4];
623 		pobits[2] = (fd_set *)&mbits[ni * 5];
624 	} else {
625 		memset(bits, 0, sizeof(bits));
626 		pibits[0] = (fd_set *)&bits[0];
627 		pibits[1] = (fd_set *)&bits[1];
628 		pibits[2] = (fd_set *)&bits[2];
629 		pobits[0] = (fd_set *)&bits[3];
630 		pobits[1] = (fd_set *)&bits[4];
631 		pobits[2] = (fd_set *)&bits[5];
632 	}
633 
634 	kqpoll_init(nd);
635 
636 #define	getbits(name, x) \
637 	if (name && (error = copyin(name, pibits[x], ni))) \
638 		goto done;
639 	getbits(in, 0);
640 	getbits(ou, 1);
641 	getbits(ex, 2);
642 #undef	getbits
643 #ifdef KTRACE
644 	if (ni > 0 && KTRPOINT(p, KTR_STRUCT)) {
645 		if (in) ktrfdset(p, pibits[0], ni);
646 		if (ou) ktrfdset(p, pibits[1], ni);
647 		if (ex) ktrfdset(p, pibits[2], ni);
648 	}
649 #endif
650 
651 	if (sigmask)
652 		dosigsuspend(p, *sigmask &~ sigcantmask);
653 
654 	/* Register kqueue events */
655 	error = pselregister(p, pibits, pobits, nd, &nevents, &ncollected);
656 	if (error != 0)
657 		goto done;
658 
659 	/*
660 	 * The poll/select family of syscalls has been designed to
661 	 * block when file descriptors are not available, even if
662 	 * there's nothing to wait for.
663 	 */
664 	if (nevents == 0 && ncollected == 0) {
665 		uint64_t nsecs = INFSLP;
666 
667 		if (timeout != NULL) {
668 			if (!timespecisset(timeout))
669 				goto done;
670 			nsecs = MAX(1, MIN(TIMESPEC_TO_NSEC(timeout), MAXTSLP));
671 		}
672 		error = tsleep_nsec(&nowake, PSOCK | PCATCH, "kqsel", nsecs);
673 		/* select is not restarted after signals... */
674 		if (error == ERESTART)
675 			error = EINTR;
676 		if (error == EWOULDBLOCK)
677 			error = 0;
678 		goto done;
679 	}
680 
681 	/* Do not block if registering found pending events. */
682 	if (ncollected > 0)
683 		timeout = &zerots;
684 
685 	/* Collect at most `nevents' possibly waiting in kqueue_scan() */
686 	kqueue_scan_setup(&scan, p->p_kq);
687 	while (nevents > 0) {
688 		struct kevent kev[KQ_NEVENTS];
689 		int i, ready, count;
690 
691 		/* Maximum number of events per iteration */
692 		count = MIN(nitems(kev), nevents);
693 		ready = kqueue_scan(&scan, count, kev, timeout, p, &error);
694 #ifdef KTRACE
695 		if (KTRPOINT(p, KTR_STRUCT))
696 			ktrevent(p, kev, ready);
697 #endif
698 		/* Convert back events that are ready. */
699 		for (i = 0; i < ready && error == 0; i++)
700 			error = pselcollect(p, &kev[i], pobits, &ncollected);
701 		/*
702 		 * Stop if there was an error or if we had enough
703 		 * space to collect all events that were ready.
704 		 */
705 		if (error || ready < count)
706 			break;
707 
708 		nevents -= ready;
709 	}
710 	kqueue_scan_finish(&scan);
711 	*retval = ncollected;
712 done:
713 #define	putbits(name, x) \
714 	if (name && (error2 = copyout(pobits[x], name, ni))) \
715 		error = error2;
716 	if (error == 0) {
717 		int error2;
718 
719 		putbits(in, 0);
720 		putbits(ou, 1);
721 		putbits(ex, 2);
722 #undef putbits
723 #ifdef KTRACE
724 		if (ni > 0 && KTRPOINT(p, KTR_STRUCT)) {
725 			if (in) ktrfdset(p, pobits[0], ni);
726 			if (ou) ktrfdset(p, pobits[1], ni);
727 			if (ex) ktrfdset(p, pobits[2], ni);
728 		}
729 #endif
730 	}
731 
732 	if (pibits[0] != (fd_set *)&bits[0])
733 		free(pibits[0], M_TEMP, 6 * ni);
734 
735 	kqpoll_done(nd);
736 
737 	return (error);
738 }
739 
740 /*
741  * Convert fd_set into kqueue events and register them on the
742  * per-thread queue.
743  */
744 int
745 pselregister(struct proc *p, fd_set *pibits[3], fd_set *pobits[3], int nfd,
746     int *nregistered, int *ncollected)
747 {
748 	static const int evf[] = { EVFILT_READ, EVFILT_WRITE, EVFILT_EXCEPT };
749 	static const int evff[] = { 0, 0, NOTE_OOB };
750 	int msk, i, j, fd, nevents = 0, error = 0;
751 	struct kevent kev;
752 	fd_mask bits;
753 
754 	for (msk = 0; msk < 3; msk++) {
755 		for (i = 0; i < nfd; i += NFDBITS) {
756 			bits = pibits[msk]->fds_bits[i / NFDBITS];
757 			while ((j = ffs(bits)) && (fd = i + --j) < nfd) {
758 				bits &= ~(1 << j);
759 
760 				DPRINTFN(2, "select fd %d mask %d serial %lu\n",
761 				    fd, msk, p->p_kq_serial);
762 				EV_SET(&kev, fd, evf[msk],
763 				    EV_ADD|EV_ENABLE|__EV_SELECT,
764 				    evff[msk], 0, (void *)(p->p_kq_serial));
765 #ifdef KTRACE
766 				if (KTRPOINT(p, KTR_STRUCT))
767 					ktrevent(p, &kev, 1);
768 #endif
769 				error = kqueue_register(p->p_kq, &kev, 0, p);
770 				switch (error) {
771 				case 0:
772 					nevents++;
773 				/* FALLTHROUGH */
774 				case EOPNOTSUPP:/* No underlying kqfilter */
775 				case EINVAL:	/* Unimplemented filter */
776 				case EPERM:	/* Specific to FIFO and
777 						 * __EV_SELECT */
778 					error = 0;
779 					break;
780 				case EPIPE:	/* Specific to pipes */
781 					KASSERT(kev.filter == EVFILT_WRITE);
782 					FD_SET(kev.ident, pobits[1]);
783 					(*ncollected)++;
784 					error = 0;
785 					break;
786 				case ENXIO:	/* Device has been detached */
787 				default:
788 					goto bad;
789 				}
790 			}
791 		}
792 	}
793 
794 	*nregistered = nevents;
795 	return (0);
796 bad:
797 	DPRINTFN(0, "select fd %u filt %d error %d\n", (int)kev.ident,
798 	    kev.filter, error);
799 	return (error);
800 }
801 
802 /*
803  * Convert given kqueue event into corresponding select(2) bit.
804  */
805 int
806 pselcollect(struct proc *p, struct kevent *kevp, fd_set *pobits[3],
807     int *ncollected)
808 {
809 	if ((unsigned long)kevp->udata != p->p_kq_serial) {
810 		panic("%s: spurious kevp %p fd %d udata 0x%lx serial 0x%lx",
811 		    __func__, kevp, (int)kevp->ident,
812 		    (unsigned long)kevp->udata, p->p_kq_serial);
813 	}
814 
815 	if (kevp->flags & EV_ERROR) {
816 		DPRINTFN(2, "select fd %d filt %d error %d\n",
817 		    (int)kevp->ident, kevp->filter, (int)kevp->data);
818 		return (kevp->data);
819 	}
820 
821 	switch (kevp->filter) {
822 	case EVFILT_READ:
823 		FD_SET(kevp->ident, pobits[0]);
824 		break;
825 	case EVFILT_WRITE:
826 		FD_SET(kevp->ident, pobits[1]);
827 		break;
828 	case EVFILT_EXCEPT:
829 		FD_SET(kevp->ident, pobits[2]);
830 		break;
831 	default:
832 		KASSERT(0);
833 	}
834 	(*ncollected)++;
835 
836 	DPRINTFN(2, "select fd %d filt %d\n", (int)kevp->ident, kevp->filter);
837 	return (0);
838 }
839 
840 /*
841  * Do a wakeup when a selectable event occurs.
842  */
843 void
844 selwakeup(struct selinfo *sip)
845 {
846 	KERNEL_LOCK();
847 	KNOTE(&sip->si_note, NOTE_SUBMIT);
848 	KERNEL_UNLOCK();
849 }
850 
851 /*
852  * Only copyout the revents field.
853  */
854 int
855 pollout(struct pollfd *pl, struct pollfd *upl, u_int nfds)
856 {
857 	int error = 0;
858 	u_int i = 0;
859 
860 	while (!error && i++ < nfds) {
861 		error = copyout(&pl->revents, &upl->revents,
862 		    sizeof(upl->revents));
863 		pl++;
864 		upl++;
865 	}
866 
867 	return (error);
868 }
869 
870 /*
871  * We are using the same mechanism as select only we encode/decode args
872  * differently.
873  */
874 int
875 sys_poll(struct proc *p, void *v, register_t *retval)
876 {
877 	struct sys_poll_args /* {
878 		syscallarg(struct pollfd *) fds;
879 		syscallarg(u_int) nfds;
880 		syscallarg(int) timeout;
881 	} */ *uap = v;
882 
883 	struct timespec ts, *tsp = NULL;
884 	int msec = SCARG(uap, timeout);
885 
886 	if (msec != INFTIM) {
887 		if (msec < 0)
888 			return (EINVAL);
889 		ts.tv_sec = msec / 1000;
890 		ts.tv_nsec = (msec - (ts.tv_sec * 1000)) * 1000000;
891 		tsp = &ts;
892 	}
893 
894 	return (doppoll(p, SCARG(uap, fds), SCARG(uap, nfds), tsp, NULL,
895 	    retval));
896 }
897 
898 int
899 sys_ppoll(struct proc *p, void *v, register_t *retval)
900 {
901 	struct sys_ppoll_args /* {
902 		syscallarg(struct pollfd *) fds;
903 		syscallarg(u_int) nfds;
904 		syscallarg(const struct timespec *) ts;
905 		syscallarg(const sigset_t *) mask;
906 	} */ *uap = v;
907 
908 	int error;
909 	struct timespec ts, *tsp = NULL;
910 	sigset_t ss, *ssp = NULL;
911 
912 	if (SCARG(uap, ts) != NULL) {
913 		if ((error = copyin(SCARG(uap, ts), &ts, sizeof ts)) != 0)
914 			return (error);
915 #ifdef KTRACE
916 		if (KTRPOINT(p, KTR_STRUCT))
917 			ktrreltimespec(p, &ts);
918 #endif
919 		if (ts.tv_sec < 0 || !timespecisvalid(&ts))
920 			return (EINVAL);
921 		tsp = &ts;
922 	}
923 
924 	if (SCARG(uap, mask) != NULL) {
925 		if ((error = copyin(SCARG(uap, mask), &ss, sizeof ss)) != 0)
926 			return (error);
927 		ssp = &ss;
928 	}
929 
930 	return (doppoll(p, SCARG(uap, fds), SCARG(uap, nfds), tsp, ssp,
931 	    retval));
932 }
933 
934 int
935 doppoll(struct proc *p, struct pollfd *fds, u_int nfds,
936     struct timespec *timeout, const sigset_t *sigmask, register_t *retval)
937 {
938 	struct kqueue_scan_state scan;
939 	struct timespec zerots = {};
940 	struct pollfd pfds[4], *pl = pfds;
941 	int error, ncollected = 0, nevents = 0;
942 	size_t sz;
943 
944 	/* Standards say no more than MAX_OPEN; this is possibly better. */
945 	if (nfds > min((int)lim_cur(RLIMIT_NOFILE), maxfiles))
946 		return (EINVAL);
947 
948 	/* optimize for the default case, of a small nfds value */
949 	if (nfds > nitems(pfds)) {
950 		pl = mallocarray(nfds, sizeof(*pl), M_TEMP,
951 		    M_WAITOK | M_CANFAIL);
952 		if (pl == NULL)
953 			return (EINVAL);
954 	}
955 
956 	kqpoll_init(nfds);
957 
958 	sz = nfds * sizeof(*pl);
959 
960 	if ((error = copyin(fds, pl, sz)) != 0)
961 		goto bad;
962 
963 	if (sigmask)
964 		dosigsuspend(p, *sigmask &~ sigcantmask);
965 
966 	/* Register kqueue events */
967 	ppollregister(p, pl, nfds, &nevents, &ncollected);
968 
969 	/*
970 	 * The poll/select family of syscalls has been designed to
971 	 * block when file descriptors are not available, even if
972 	 * there's nothing to wait for.
973 	 */
974 	if (nevents == 0 && ncollected == 0) {
975 		uint64_t nsecs = INFSLP;
976 
977 		if (timeout != NULL) {
978 			if (!timespecisset(timeout))
979 				goto done;
980 			nsecs = MAX(1, MIN(TIMESPEC_TO_NSEC(timeout), MAXTSLP));
981 		}
982 
983 		error = tsleep_nsec(&nowake, PSOCK | PCATCH, "kqpoll", nsecs);
984 		if (error == ERESTART)
985 			error = EINTR;
986 		if (error == EWOULDBLOCK)
987 			error = 0;
988 		goto done;
989 	}
990 
991 	/* Do not block if registering found pending events. */
992 	if (ncollected > 0)
993 		timeout = &zerots;
994 
995 	/* Collect at most `nevents' possibly waiting in kqueue_scan() */
996 	kqueue_scan_setup(&scan, p->p_kq);
997 	while (nevents > 0) {
998 		struct kevent kev[KQ_NEVENTS];
999 		int i, ready, count;
1000 
1001 		/* Maximum number of events per iteration */
1002 		count = MIN(nitems(kev), nevents);
1003 		ready = kqueue_scan(&scan, count, kev, timeout, p, &error);
1004 #ifdef KTRACE
1005 		if (KTRPOINT(p, KTR_STRUCT))
1006 			ktrevent(p, kev, ready);
1007 #endif
1008 		/* Convert back events that are ready. */
1009 		for (i = 0; i < ready; i++)
1010 			ncollected += ppollcollect(p, &kev[i], pl, nfds);
1011 
1012 		/*
1013 		 * Stop if there was an error or if we had enough
1014 		 * place to collect all events that were ready.
1015 		 */
1016 		if (error || ready < count)
1017 			break;
1018 
1019 		nevents -= ready;
1020 	}
1021 	kqueue_scan_finish(&scan);
1022 	*retval = ncollected;
1023 done:
1024 	/*
1025 	 * NOTE: poll(2) is not restarted after a signal and EWOULDBLOCK is
1026 	 *       ignored (since the whole point is to see what would block).
1027 	 */
1028 	switch (error) {
1029 	case EINTR:
1030 		error = pollout(pl, fds, nfds);
1031 		if (error == 0)
1032 			error = EINTR;
1033 		break;
1034 	case EWOULDBLOCK:
1035 	case 0:
1036 		error = pollout(pl, fds, nfds);
1037 		break;
1038 	}
1039 #ifdef KTRACE
1040 	if (KTRPOINT(p, KTR_STRUCT))
1041 		ktrpollfd(p, pl, nfds);
1042 #endif /* KTRACE */
1043 bad:
1044 	if (pl != pfds)
1045 		free(pl, M_TEMP, sz);
1046 
1047 	kqpoll_done(nfds);
1048 
1049 	return (error);
1050 }
1051 
1052 int
1053 ppollregister_evts(struct proc *p, struct kevent *kevp, int nkev,
1054     struct pollfd *pl, unsigned int pollid)
1055 {
1056 	int i, error, nevents = 0;
1057 
1058 	KASSERT(pl->revents == 0);
1059 
1060 #ifdef KTRACE
1061 	if (KTRPOINT(p, KTR_STRUCT))
1062 		ktrevent(p, kevp, nkev);
1063 #endif
1064 	for (i = 0; i < nkev; i++, kevp++) {
1065 again:
1066 		error = kqueue_register(p->p_kq, kevp, pollid, p);
1067 		switch (error) {
1068 		case 0:
1069 			nevents++;
1070 			break;
1071 		case EOPNOTSUPP:/* No underlying kqfilter */
1072 		case EINVAL:	/* Unimplemented filter */
1073 			break;
1074 		case EBADF:	/* Bad file descriptor */
1075 			pl->revents |= POLLNVAL;
1076 			break;
1077 		case EPERM:	/* Specific to FIFO */
1078 			KASSERT(kevp->filter == EVFILT_WRITE);
1079 			if (nkev == 1) {
1080 				/*
1081 				 * If this is the only filter make sure
1082 				 * POLLHUP is passed to userland.
1083 				 */
1084 				kevp->filter = EVFILT_EXCEPT;
1085 				goto again;
1086 			}
1087 			break;
1088 		case EPIPE:	/* Specific to pipes */
1089 			KASSERT(kevp->filter == EVFILT_WRITE);
1090 			pl->revents |= POLLHUP;
1091 			break;
1092 		default:
1093 			DPRINTFN(0, "poll err %lu fd %d revents %02x serial"
1094 			    " %lu filt %d ERROR=%d\n",
1095 			    ((unsigned long)kevp->udata - p->p_kq_serial),
1096 			    pl->fd, pl->revents, p->p_kq_serial, kevp->filter,
1097 			    error);
1098 			/* FALLTHROUGH */
1099 		case ENXIO:	/* Device has been detached */
1100 			pl->revents |= POLLERR;
1101 			break;
1102 		}
1103 	}
1104 
1105 	return (nevents);
1106 }
1107 
1108 /*
1109  * Convert pollfd into kqueue events and register them on the
1110  * per-thread queue.
1111  *
1112  * At most 3 events can correspond to a single pollfd.
1113  */
1114 void
1115 ppollregister(struct proc *p, struct pollfd *pl, int nfds, int *nregistered,
1116     int *ncollected)
1117 {
1118 	int i, nkev, nevt, forcehup;
1119 	struct kevent kev[3], *kevp;
1120 
1121 	for (i = 0; i < nfds; i++) {
1122 		pl[i].events &= ~POLL_NOHUP;
1123 		pl[i].revents = 0;
1124 
1125 		if (pl[i].fd < 0)
1126 			continue;
1127 
1128 		/*
1129 		 * POLLHUP checking is implicit in the event filters.
1130 		 * However, the checking must be even if no events are
1131 		 * requested.
1132 		 */
1133 		forcehup = ((pl[i].events & ~POLLHUP) == 0);
1134 
1135 		DPRINTFN(1, "poll set %d/%d fd %d events %02x serial %lu\n",
1136 		    i+1, nfds, pl[i].fd, pl[i].events, p->p_kq_serial);
1137 
1138 		nevt = 0;
1139 		nkev = 0;
1140 		kevp = kev;
1141 		if (pl[i].events & (POLLIN | POLLRDNORM)) {
1142 			EV_SET(kevp, pl[i].fd, EVFILT_READ,
1143 			    EV_ADD|EV_ENABLE|__EV_POLL, 0, 0,
1144 			    (void *)(p->p_kq_serial + i));
1145 			nkev++;
1146 			kevp++;
1147 		}
1148 		if (pl[i].events & (POLLOUT | POLLWRNORM)) {
1149 			EV_SET(kevp, pl[i].fd, EVFILT_WRITE,
1150 			    EV_ADD|EV_ENABLE|__EV_POLL, 0, 0,
1151 			    (void *)(p->p_kq_serial + i));
1152 			nkev++;
1153 			kevp++;
1154 		}
1155 		if ((pl[i].events & (POLLPRI | POLLRDBAND)) || forcehup) {
1156 			int evff = forcehup ? 0 : NOTE_OOB;
1157 
1158 			EV_SET(kevp, pl[i].fd, EVFILT_EXCEPT,
1159 			    EV_ADD|EV_ENABLE|__EV_POLL, evff, 0,
1160 			    (void *)(p->p_kq_serial + i));
1161 			nkev++;
1162 			kevp++;
1163 		}
1164 
1165 		if (nkev == 0)
1166 			continue;
1167 
1168 		*nregistered += ppollregister_evts(p, kev, nkev, &pl[i], i);
1169 
1170 		if (pl[i].revents != 0)
1171 			(*ncollected)++;
1172 	}
1173 
1174 	DPRINTFN(1, "poll registered = %d, collected = %d\n", *nregistered,
1175 	    *ncollected);
1176 }
1177 
1178 /*
1179  * Convert given kqueue event into corresponding poll(2) revents bit.
1180  */
1181 int
1182 ppollcollect(struct proc *p, struct kevent *kevp, struct pollfd *pl, u_int nfds)
1183 {
1184 	static struct timeval poll_errintvl = { 5, 0 };
1185 	static struct timeval poll_lasterr;
1186 	int already_seen;
1187 	unsigned long i;
1188 
1189 	/*  Extract poll array index */
1190 	i = (unsigned long)kevp->udata - p->p_kq_serial;
1191 
1192 	if (i >= nfds) {
1193 		panic("%s: spurious kevp %p nfds %u udata 0x%lx serial 0x%lx",
1194 		    __func__, kevp, nfds,
1195 		    (unsigned long)kevp->udata, p->p_kq_serial);
1196 	}
1197 	if ((int)kevp->ident != pl[i].fd) {
1198 		panic("%s: kevp %p %lu/%d mismatch fd %d!=%d serial 0x%lx",
1199 		    __func__, kevp, i + 1, nfds, (int)kevp->ident, pl[i].fd,
1200 		    p->p_kq_serial);
1201 	}
1202 
1203 	/*
1204 	 * A given descriptor may already have generated an error
1205 	 * against another filter during kqueue_register().
1206 	 *
1207 	 * Make sure to set the appropriate flags but do not
1208 	 * increment `*retval' more than once.
1209 	 */
1210 	already_seen = (pl[i].revents != 0);
1211 
1212 	/* POLLNVAL preempts other events. */
1213 	if ((kevp->flags & EV_ERROR) && kevp->data == EBADF) {
1214 		pl[i].revents = POLLNVAL;
1215 		goto done;
1216 	} else if (pl[i].revents & POLLNVAL) {
1217 		goto done;
1218 	}
1219 
1220 	switch (kevp->filter) {
1221 	case EVFILT_READ:
1222 		if (kevp->flags & __EV_HUP)
1223 			pl[i].revents |= POLLHUP;
1224 		if (pl[i].events & (POLLIN | POLLRDNORM))
1225 			pl[i].revents |= pl[i].events & (POLLIN | POLLRDNORM);
1226 		break;
1227 	case EVFILT_WRITE:
1228 		/* POLLHUP and POLLOUT/POLLWRNORM are mutually exclusive */
1229 		if (kevp->flags & __EV_HUP) {
1230 			pl[i].revents |= POLLHUP;
1231 		} else if (pl[i].events & (POLLOUT | POLLWRNORM)) {
1232 			pl[i].revents |= pl[i].events & (POLLOUT | POLLWRNORM);
1233 		}
1234 		break;
1235 	case EVFILT_EXCEPT:
1236 		if (kevp->flags & __EV_HUP) {
1237 			if (pl[i].events != 0 && pl[i].events != POLLOUT)
1238 				DPRINTFN(0, "weird events %x\n", pl[i].events);
1239 			pl[i].revents |= POLLHUP;
1240 			break;
1241 		}
1242 		if (pl[i].events & (POLLPRI | POLLRDBAND))
1243 			pl[i].revents |= pl[i].events & (POLLPRI | POLLRDBAND);
1244 		break;
1245 	default:
1246 		KASSERT(0);
1247 	}
1248 
1249 done:
1250 	DPRINTFN(1, "poll get %lu/%d fd %d revents %02x serial %lu filt %d\n",
1251 	    i+1, nfds, pl[i].fd, pl[i].revents, (unsigned long)kevp->udata,
1252 	    kevp->filter);
1253 
1254 	/*
1255 	 * Make noise about unclaimed events as they might indicate a bug
1256 	 * and can result in spurious-looking wakeups of poll(2).
1257 	 *
1258 	 * Live-locking within the system call should not happen because
1259 	 * the scan loop in doppoll() has an upper limit for the number
1260 	 * of events to process.
1261 	 */
1262 	if (pl[i].revents == 0 && ratecheck(&poll_lasterr, &poll_errintvl)) {
1263 		printf("%s[%d]: poll index %lu fd %d events 0x%x "
1264 		    "filter %d/0x%x unclaimed\n",
1265 		    p->p_p->ps_comm, p->p_tid, i, pl[i].fd,
1266 		    pl[i].events, kevp->filter, kevp->flags);
1267 	}
1268 
1269 	if (!already_seen && (pl[i].revents != 0))
1270 		return (1);
1271 
1272 	return (0);
1273 }
1274 
1275 /*
1276  * utrace system call
1277  */
1278 int
1279 sys_utrace(struct proc *curp, void *v, register_t *retval)
1280 {
1281 #ifdef KTRACE
1282 	struct sys_utrace_args /* {
1283 		syscallarg(const char *) label;
1284 		syscallarg(const void *) addr;
1285 		syscallarg(size_t) len;
1286 	} */ *uap = v;
1287 
1288 	return (ktruser(curp, SCARG(uap, label), SCARG(uap, addr),
1289 	    SCARG(uap, len)));
1290 #else
1291 	return (0);
1292 #endif
1293 }
1294