xref: /dragonfly/sys/kern/sys_generic.c (revision 1a05b9d1)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)sys_generic.c	8.5 (Berkeley) 1/21/94
35  * $FreeBSD: src/sys/kern/sys_generic.c,v 1.55.2.10 2001/03/17 10:39:32 peter Exp $
36  */
37 
38 #include "opt_ktrace.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/event.h>
44 #include <sys/filedesc.h>
45 #include <sys/filio.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/proc.h>
49 #include <sys/signalvar.h>
50 #include <sys/socketvar.h>
51 #include <sys/malloc.h>
52 #include <sys/uio.h>
53 #include <sys/kernel.h>
54 #include <sys/kern_syscall.h>
55 #include <sys/mapped_ioctl.h>
56 #include <sys/poll.h>
57 #include <sys/queue.h>
58 #include <sys/resourcevar.h>
59 #include <sys/socketops.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
62 #include <sys/buf.h>
63 #ifdef KTRACE
64 #include <sys/ktrace.h>
65 #endif
66 #include <vm/vm.h>
67 #include <vm/vm_page.h>
68 
69 #include <sys/file2.h>
70 #include <sys/spinlock2.h>
71 
72 #include <machine/limits.h>
73 
74 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
75 static MALLOC_DEFINE(M_IOCTLMAP, "ioctlmap", "mapped ioctl handler buffer");
76 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
77 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
78 
79 typedef struct kfd_set {
80         fd_mask	fds_bits[2];
81 } kfd_set;
82 
83 enum select_copyin_states {
84     COPYIN_READ, COPYIN_WRITE, COPYIN_EXCEPT, COPYIN_DONE };
85 
86 struct select_kevent_copyin_args {
87 	kfd_set		*read_set;
88 	kfd_set		*write_set;
89 	kfd_set		*except_set;
90 	int		active_set;	/* One of select_copyin_states */
91 	struct lwp	*lwp;		/* Pointer to our lwp */
92 	int		num_fds;	/* Number of file descriptors (syscall arg) */
93 	int		proc_fds;	/* Processed fd's (wraps) */
94 	int		error;		/* Returned to userland */
95 };
96 
97 struct poll_kevent_copyin_args {
98 	struct lwp	*lwp;
99 	struct pollfd	*fds;
100 	int		nfds;
101 	int		pfds;
102 	int		error;
103 };
104 
105 static struct lwkt_token mioctl_token = LWKT_TOKEN_INITIALIZER(mioctl_token);
106 
107 static int 	doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex,
108 			 struct timespec *ts, int *res);
109 static int	dopoll(int nfds, struct pollfd *fds, struct timespec *ts,
110 		       int *res, int flags);
111 static int	dofileread(int, struct file *, struct uio *, int, size_t *);
112 static int	dofilewrite(int, struct file *, struct uio *, int, size_t *);
113 
114 /*
115  * Read system call.
116  *
117  * MPSAFE
118  */
119 int
120 sys_read(struct read_args *uap)
121 {
122 	struct thread *td = curthread;
123 	struct uio auio;
124 	struct iovec aiov;
125 	int error;
126 
127 	if ((ssize_t)uap->nbyte < 0)
128 		error = EINVAL;
129 
130 	aiov.iov_base = uap->buf;
131 	aiov.iov_len = uap->nbyte;
132 	auio.uio_iov = &aiov;
133 	auio.uio_iovcnt = 1;
134 	auio.uio_offset = -1;
135 	auio.uio_resid = uap->nbyte;
136 	auio.uio_rw = UIO_READ;
137 	auio.uio_segflg = UIO_USERSPACE;
138 	auio.uio_td = td;
139 
140 	error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
141 	return(error);
142 }
143 
144 /*
145  * Positioned (Pread) read system call
146  *
147  * MPSAFE
148  */
149 int
150 sys_extpread(struct extpread_args *uap)
151 {
152 	struct thread *td = curthread;
153 	struct uio auio;
154 	struct iovec aiov;
155 	int error;
156 	int flags;
157 
158 	if ((ssize_t)uap->nbyte < 0)
159 		return(EINVAL);
160 
161 	aiov.iov_base = uap->buf;
162 	aiov.iov_len = uap->nbyte;
163 	auio.uio_iov = &aiov;
164 	auio.uio_iovcnt = 1;
165 	auio.uio_offset = uap->offset;
166 	auio.uio_resid = uap->nbyte;
167 	auio.uio_rw = UIO_READ;
168 	auio.uio_segflg = UIO_USERSPACE;
169 	auio.uio_td = td;
170 
171 	flags = uap->flags & O_FMASK;
172 	if (uap->offset != (off_t)-1)
173 		flags |= O_FOFFSET;
174 
175 	error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
176 	return(error);
177 }
178 
179 /*
180  * Scatter read system call.
181  *
182  * MPSAFE
183  */
184 int
185 sys_readv(struct readv_args *uap)
186 {
187 	struct thread *td = curthread;
188 	struct uio auio;
189 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
190 	int error;
191 
192 	error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
193 			     &auio.uio_resid);
194 	if (error)
195 		return (error);
196 	auio.uio_iov = iov;
197 	auio.uio_iovcnt = uap->iovcnt;
198 	auio.uio_offset = -1;
199 	auio.uio_rw = UIO_READ;
200 	auio.uio_segflg = UIO_USERSPACE;
201 	auio.uio_td = td;
202 
203 	error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
204 
205 	iovec_free(&iov, aiov);
206 	return (error);
207 }
208 
209 
210 /*
211  * Scatter positioned read system call.
212  *
213  * MPSAFE
214  */
215 int
216 sys_extpreadv(struct extpreadv_args *uap)
217 {
218 	struct thread *td = curthread;
219 	struct uio auio;
220 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
221 	int error;
222 	int flags;
223 
224 	error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
225 			     &auio.uio_resid);
226 	if (error)
227 		return (error);
228 	auio.uio_iov = iov;
229 	auio.uio_iovcnt = uap->iovcnt;
230 	auio.uio_offset = uap->offset;
231 	auio.uio_rw = UIO_READ;
232 	auio.uio_segflg = UIO_USERSPACE;
233 	auio.uio_td = td;
234 
235 	flags = uap->flags & O_FMASK;
236 	if (uap->offset != (off_t)-1)
237 		flags |= O_FOFFSET;
238 
239 	error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
240 
241 	iovec_free(&iov, aiov);
242 	return(error);
243 }
244 
245 /*
246  * MPSAFE
247  */
248 int
249 kern_preadv(int fd, struct uio *auio, int flags, size_t *res)
250 {
251 	struct thread *td = curthread;
252 	struct file *fp;
253 	int error;
254 
255 	fp = holdfp(td, fd, FREAD);
256 	if (fp == NULL)
257 		return (EBADF);
258 	if (flags & O_FOFFSET && fp->f_type != DTYPE_VNODE) {
259 		error = ESPIPE;
260 	} else {
261 		error = dofileread(fd, fp, auio, flags, res);
262 	}
263 	dropfp(td, fd, fp);
264 
265 	return(error);
266 }
267 
268 /*
269  * Common code for readv and preadv that reads data in
270  * from a file using the passed in uio, offset, and flags.
271  *
272  * MPALMOSTSAFE - ktrace needs help
273  */
274 static int
275 dofileread(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
276 {
277 	int error;
278 	size_t len;
279 #ifdef KTRACE
280 	struct thread *td = curthread;
281 	struct iovec *ktriov = NULL;
282 	struct uio ktruio;
283 #endif
284 
285 #ifdef KTRACE
286 	/*
287 	 * if tracing, save a copy of iovec
288 	 */
289 	if (KTRPOINT(td, KTR_GENIO))  {
290 		int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
291 
292 		ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
293 		bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
294 		ktruio = *auio;
295 	}
296 #endif
297 	len = auio->uio_resid;
298 	error = fo_read(fp, auio, fp->f_cred, flags);
299 	if (error) {
300 		if (auio->uio_resid != len && (error == ERESTART ||
301 		    error == EINTR || error == EWOULDBLOCK))
302 			error = 0;
303 	}
304 #ifdef KTRACE
305 	if (ktriov != NULL) {
306 		if (error == 0) {
307 			ktruio.uio_iov = ktriov;
308 			ktruio.uio_resid = len - auio->uio_resid;
309 			ktrgenio(td->td_lwp, fd, UIO_READ, &ktruio, error);
310 		}
311 		kfree(ktriov, M_TEMP);
312 	}
313 #endif
314 	if (error == 0)
315 		*res = len - auio->uio_resid;
316 
317 	return(error);
318 }
319 
320 /*
321  * Write system call
322  *
323  * MPSAFE
324  */
325 int
326 sys_write(struct write_args *uap)
327 {
328 	struct thread *td = curthread;
329 	struct uio auio;
330 	struct iovec aiov;
331 	int error;
332 
333 	if ((ssize_t)uap->nbyte < 0)
334 		error = EINVAL;
335 
336 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
337 	aiov.iov_len = uap->nbyte;
338 	auio.uio_iov = &aiov;
339 	auio.uio_iovcnt = 1;
340 	auio.uio_offset = -1;
341 	auio.uio_resid = uap->nbyte;
342 	auio.uio_rw = UIO_WRITE;
343 	auio.uio_segflg = UIO_USERSPACE;
344 	auio.uio_td = td;
345 
346 	error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
347 
348 	return(error);
349 }
350 
351 /*
352  * Pwrite system call
353  *
354  * MPSAFE
355  */
356 int
357 sys_extpwrite(struct extpwrite_args *uap)
358 {
359 	struct thread *td = curthread;
360 	struct uio auio;
361 	struct iovec aiov;
362 	int error;
363 	int flags;
364 
365 	if ((ssize_t)uap->nbyte < 0)
366 		error = EINVAL;
367 
368 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
369 	aiov.iov_len = uap->nbyte;
370 	auio.uio_iov = &aiov;
371 	auio.uio_iovcnt = 1;
372 	auio.uio_offset = uap->offset;
373 	auio.uio_resid = uap->nbyte;
374 	auio.uio_rw = UIO_WRITE;
375 	auio.uio_segflg = UIO_USERSPACE;
376 	auio.uio_td = td;
377 
378 	flags = uap->flags & O_FMASK;
379 	if (uap->offset != (off_t)-1)
380 		flags |= O_FOFFSET;
381 	error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
382 	return(error);
383 }
384 
385 /*
386  * MPSAFE
387  */
388 int
389 sys_writev(struct writev_args *uap)
390 {
391 	struct thread *td = curthread;
392 	struct uio auio;
393 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
394 	int error;
395 
396 	error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
397 			     &auio.uio_resid);
398 	if (error)
399 		return (error);
400 	auio.uio_iov = iov;
401 	auio.uio_iovcnt = uap->iovcnt;
402 	auio.uio_offset = -1;
403 	auio.uio_rw = UIO_WRITE;
404 	auio.uio_segflg = UIO_USERSPACE;
405 	auio.uio_td = td;
406 
407 	error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
408 
409 	iovec_free(&iov, aiov);
410 	return (error);
411 }
412 
413 
414 /*
415  * Gather positioned write system call
416  *
417  * MPSAFE
418  */
419 int
420 sys_extpwritev(struct extpwritev_args *uap)
421 {
422 	struct thread *td = curthread;
423 	struct uio auio;
424 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
425 	int error;
426 	int flags;
427 
428 	error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
429 			     &auio.uio_resid);
430 	if (error)
431 		return (error);
432 	auio.uio_iov = iov;
433 	auio.uio_iovcnt = uap->iovcnt;
434 	auio.uio_offset = uap->offset;
435 	auio.uio_rw = UIO_WRITE;
436 	auio.uio_segflg = UIO_USERSPACE;
437 	auio.uio_td = td;
438 
439 	flags = uap->flags & O_FMASK;
440 	if (uap->offset != (off_t)-1)
441 		flags |= O_FOFFSET;
442 
443 	error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
444 
445 	iovec_free(&iov, aiov);
446 	return(error);
447 }
448 
449 /*
450  * MPSAFE
451  */
452 int
453 kern_pwritev(int fd, struct uio *auio, int flags, size_t *res)
454 {
455 	struct thread *td = curthread;
456 	struct file *fp;
457 	int error;
458 
459 	fp = holdfp(td, fd, FWRITE);
460 	if (fp == NULL)
461 		return (EBADF);
462 	else if ((flags & O_FOFFSET) && fp->f_type != DTYPE_VNODE) {
463 		error = ESPIPE;
464 	} else {
465 		error = dofilewrite(fd, fp, auio, flags, res);
466 	}
467 	dropfp(td, fd, fp);
468 
469 	return(error);
470 }
471 
472 /*
473  * Common code for writev and pwritev that writes data to
474  * a file using the passed in uio, offset, and flags.
475  *
476  * MPALMOSTSAFE - ktrace needs help
477  */
478 static int
479 dofilewrite(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
480 {
481 	struct thread *td = curthread;
482 	struct lwp *lp = td->td_lwp;
483 	int error;
484 	size_t len;
485 #ifdef KTRACE
486 	struct iovec *ktriov = NULL;
487 	struct uio ktruio;
488 #endif
489 
490 #ifdef KTRACE
491 	/*
492 	 * if tracing, save a copy of iovec and uio
493 	 */
494 	if (KTRPOINT(td, KTR_GENIO))  {
495 		int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
496 
497 		ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
498 		bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
499 		ktruio = *auio;
500 	}
501 #endif
502 	len = auio->uio_resid;
503 	error = fo_write(fp, auio, fp->f_cred, flags);
504 	if (error) {
505 		if (auio->uio_resid != len && (error == ERESTART ||
506 		    error == EINTR || error == EWOULDBLOCK))
507 			error = 0;
508 		/* Socket layer is responsible for issuing SIGPIPE. */
509 		if (error == EPIPE && fp->f_type != DTYPE_SOCKET)
510 			lwpsignal(lp->lwp_proc, lp, SIGPIPE);
511 	}
512 #ifdef KTRACE
513 	if (ktriov != NULL) {
514 		if (error == 0) {
515 			ktruio.uio_iov = ktriov;
516 			ktruio.uio_resid = len - auio->uio_resid;
517 			ktrgenio(lp, fd, UIO_WRITE, &ktruio, error);
518 		}
519 		kfree(ktriov, M_TEMP);
520 	}
521 #endif
522 	if (error == 0)
523 		*res = len - auio->uio_resid;
524 
525 	return(error);
526 }
527 
528 /*
529  * Ioctl system call
530  *
531  * MPSAFE
532  */
533 int
534 sys_ioctl(struct ioctl_args *uap)
535 {
536 	int error;
537 
538 	error = mapped_ioctl(uap->fd, uap->com, uap->data, NULL, &uap->sysmsg);
539 	return (error);
540 }
541 
542 struct ioctl_map_entry {
543 	const char *subsys;
544 	struct ioctl_map_range *cmd_ranges;
545 	LIST_ENTRY(ioctl_map_entry) entries;
546 };
547 
548 /*
549  * The true heart of all ioctl syscall handlers (native, emulation).
550  * If map != NULL, it will be searched for a matching entry for com,
551  * and appropriate conversions/conversion functions will be utilized.
552  *
553  * MPSAFE
554  */
555 int
556 mapped_ioctl(int fd, u_long com, caddr_t uspc_data, struct ioctl_map *map,
557 	     struct sysmsg *msg)
558 {
559 	struct thread *td = curthread;
560 	struct proc *p = td->td_proc;
561 	struct ucred *cred;
562 	struct file *fp;
563 	struct ioctl_map_range *iomc = NULL;
564 	int error;
565 	u_int size;
566 	u_long ocom = com;
567 	caddr_t data, memp;
568 	int tmp;
569 #define STK_PARAMS	128
570 	union {
571 	    char stkbuf[STK_PARAMS];
572 	    long align;
573 	} ubuf;
574 
575 	KKASSERT(p);
576 	cred = td->td_ucred;
577 	memp = NULL;
578 
579 	fp = holdfp(td, fd, FREAD|FWRITE);
580 	if (fp == NULL)
581 		return(EBADF);
582 
583 	if (map != NULL) {	/* obey translation map */
584 		u_long maskcmd;
585 		struct ioctl_map_entry *e;
586 
587 		maskcmd = com & map->mask;
588 
589 		lwkt_gettoken(&mioctl_token);
590 		LIST_FOREACH(e, &map->mapping, entries) {
591 			for (iomc = e->cmd_ranges; iomc->start != 0 ||
592 			     iomc->maptocmd != 0 || iomc->wrapfunc != NULL ||
593 			     iomc->mapfunc != NULL;
594 			     iomc++) {
595 				if (maskcmd >= iomc->start &&
596 				    maskcmd <= iomc->end)
597 					break;
598 			}
599 
600 			/* Did we find a match? */
601 			if (iomc->start != 0 || iomc->maptocmd != 0 ||
602 			    iomc->wrapfunc != NULL || iomc->mapfunc != NULL)
603 				break;
604 		}
605 		lwkt_reltoken(&mioctl_token);
606 
607 		if (iomc == NULL ||
608 		    (iomc->start == 0 && iomc->maptocmd == 0
609 		     && iomc->wrapfunc == NULL && iomc->mapfunc == NULL)) {
610 			kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n",
611 			       map->sys, fd, maskcmd,
612 			       (int)((maskcmd >> 8) & 0xff),
613 			       (int)(maskcmd & 0xff));
614 			error = EINVAL;
615 			goto done;
616 		}
617 
618 		/*
619 		 * If it's a non-range one to one mapping, maptocmd should be
620 		 * correct. If it's a ranged one to one mapping, we pass the
621 		 * original value of com, and for a range mapped to a different
622 		 * range, we always need a mapping function to translate the
623 		 * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff
624 		 */
625 		if (iomc->start == iomc->end && iomc->maptocmd == iomc->maptoend) {
626 			com = iomc->maptocmd;
627 		} else if (iomc->start == iomc->maptocmd && iomc->end == iomc->maptoend) {
628 			if (iomc->mapfunc != NULL)
629 				com = iomc->mapfunc(iomc->start, iomc->end,
630 						    iomc->start, iomc->end,
631 						    com, com);
632 		} else {
633 			if (iomc->mapfunc != NULL) {
634 				com = iomc->mapfunc(iomc->start, iomc->end,
635 						    iomc->maptocmd, iomc->maptoend,
636 						    com, ocom);
637 			} else {
638 				kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n",
639 				       map->sys, fd, maskcmd,
640 				       (int)((maskcmd >> 8) & 0xff),
641 				       (int)(maskcmd & 0xff));
642 				error = EINVAL;
643 				goto done;
644 			}
645 		}
646 	}
647 
648 	switch (com) {
649 	case FIONCLEX:
650 		error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
651 		goto done;
652 	case FIOCLEX:
653 		error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
654 		goto done;
655 	}
656 
657 	/*
658 	 * Interpret high order word to find amount of data to be
659 	 * copied to/from the user's address space.
660 	 */
661 	size = IOCPARM_LEN(com);
662 	if (size > IOCPARM_MAX) {
663 		error = ENOTTY;
664 		goto done;
665 	}
666 
667 	if ((com & IOC_VOID) == 0 && size > sizeof(ubuf.stkbuf)) {
668 		memp = kmalloc(size, M_IOCTLOPS, M_WAITOK);
669 		data = memp;
670 	} else {
671 		memp = NULL;
672 		data = ubuf.stkbuf;
673 	}
674 	if (com & IOC_VOID) {
675 		*(caddr_t *)data = uspc_data;
676 	} else if (com & IOC_IN) {
677 		if (size != 0) {
678 			error = copyin(uspc_data, data, (size_t)size);
679 			if (error)
680 				goto done;
681 		} else {
682 			*(caddr_t *)data = uspc_data;
683 		}
684 	} else if ((com & IOC_OUT) != 0 && size) {
685 		/*
686 		 * Zero the buffer so the user always
687 		 * gets back something deterministic.
688 		 */
689 		bzero(data, (size_t)size);
690 	}
691 
692 	switch (com) {
693 	case FIONBIO:
694 		if ((tmp = *(int *)data))
695 			atomic_set_int(&fp->f_flag, FNONBLOCK);
696 		else
697 			atomic_clear_int(&fp->f_flag, FNONBLOCK);
698 		error = 0;
699 		break;
700 
701 	case FIOASYNC:
702 		if ((tmp = *(int *)data))
703 			atomic_set_int(&fp->f_flag, FASYNC);
704 		else
705 			atomic_clear_int(&fp->f_flag, FASYNC);
706 		error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred, msg);
707 		break;
708 
709 	default:
710 		/*
711 		 *  If there is a override function,
712 		 *  call it instead of directly routing the call
713 		 */
714 		if (map != NULL && iomc->wrapfunc != NULL)
715 			error = iomc->wrapfunc(fp, com, ocom, data, cred);
716 		else
717 			error = fo_ioctl(fp, com, data, cred, msg);
718 		/*
719 		 * Copy any data to user, size was
720 		 * already set and checked above.
721 		 */
722 		if (error == 0 && (com & IOC_OUT) != 0 && size != 0)
723 			error = copyout(data, uspc_data, (size_t)size);
724 		break;
725 	}
726 done:
727 	if (memp != NULL)
728 		kfree(memp, M_IOCTLOPS);
729 	dropfp(td, fd, fp);
730 
731 	return(error);
732 }
733 
734 /*
735  * MPSAFE
736  */
737 int
738 mapped_ioctl_register_handler(struct ioctl_map_handler *he)
739 {
740 	struct ioctl_map_entry *ne;
741 
742 	KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL &&
743 		 he->subsys != NULL && *he->subsys != '\0');
744 
745 	ne = kmalloc(sizeof(struct ioctl_map_entry), M_IOCTLMAP,
746 		     M_WAITOK | M_ZERO);
747 
748 	ne->subsys = he->subsys;
749 	ne->cmd_ranges = he->cmd_ranges;
750 
751 	lwkt_gettoken(&mioctl_token);
752 	LIST_INSERT_HEAD(&he->map->mapping, ne, entries);
753 	lwkt_reltoken(&mioctl_token);
754 
755 	return(0);
756 }
757 
758 /*
759  * MPSAFE
760  */
761 int
762 mapped_ioctl_unregister_handler(struct ioctl_map_handler *he)
763 {
764 	struct ioctl_map_entry *ne;
765 	int error = EINVAL;
766 
767 	KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL);
768 
769 	lwkt_gettoken(&mioctl_token);
770 	LIST_FOREACH(ne, &he->map->mapping, entries) {
771 		if (ne->cmd_ranges == he->cmd_ranges) {
772 			LIST_REMOVE(ne, entries);
773 			kfree(ne, M_IOCTLMAP);
774 			error = 0;
775 			break;
776 		}
777 	}
778 	lwkt_reltoken(&mioctl_token);
779 	return(error);
780 }
781 
782 static int	nselcoll;	/* Select collisions since boot */
783 int	selwait;
784 SYSCTL_INT(_kern, OID_AUTO, nselcoll, CTLFLAG_RD, &nselcoll, 0, "");
785 static int	nseldebug;
786 SYSCTL_INT(_kern, OID_AUTO, nseldebug, CTLFLAG_RW, &nseldebug, 0, "");
787 
788 /*
789  * Select system call.
790  *
791  * MPSAFE
792  */
793 int
794 sys_select(struct select_args *uap)
795 {
796 	struct timeval ktv;
797 	struct timespec *ktsp, kts;
798 	int error;
799 
800 	/*
801 	 * Get timeout if any.
802 	 */
803 	if (uap->tv != NULL) {
804 		error = copyin(uap->tv, &ktv, sizeof (ktv));
805 		if (error)
806 			return (error);
807 		TIMEVAL_TO_TIMESPEC(&ktv, &kts);
808 		ktsp = &kts;
809 	} else {
810 		ktsp = NULL;
811 	}
812 
813 	/*
814 	 * Do real work.
815 	 */
816 	error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
817 			 &uap->sysmsg_result);
818 
819 	return (error);
820 }
821 
822 
823 /*
824  * Pselect system call.
825  */
826 int
827 sys_pselect(struct pselect_args *uap)
828 {
829 	struct thread *td = curthread;
830 	struct lwp *lp = td->td_lwp;
831 	struct timespec *ktsp, kts;
832 	sigset_t sigmask;
833 	int error;
834 
835 	/*
836 	 * Get timeout if any.
837 	 */
838 	if (uap->ts != NULL) {
839 		error = copyin(uap->ts, &kts, sizeof (kts));
840 		if (error)
841 			return (error);
842 		ktsp = &kts;
843 	} else {
844 		ktsp = NULL;
845 	}
846 
847 	/*
848 	 * Install temporary signal mask if any provided.
849 	 */
850 	if (uap->sigmask != NULL) {
851 		error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
852 		if (error)
853 			return (error);
854 		lwkt_gettoken(&lp->lwp_proc->p_token);
855 		lp->lwp_oldsigmask = lp->lwp_sigmask;
856 		SIG_CANTMASK(sigmask);
857 		lp->lwp_sigmask = sigmask;
858 		lwkt_reltoken(&lp->lwp_proc->p_token);
859 	}
860 
861 	/*
862 	 * Do real job.
863 	 */
864 	error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
865 			 &uap->sysmsg_result);
866 
867 	if (uap->sigmask != NULL) {
868 		lwkt_gettoken(&lp->lwp_proc->p_token);
869 		/* doselect() responsible for turning ERESTART into EINTR */
870 		KKASSERT(error != ERESTART);
871 		if (error == EINTR) {
872 			/*
873 			 * We can't restore the previous signal mask now
874 			 * because it could block the signal that interrupted
875 			 * us.  So make a note to restore it after executing
876 			 * the handler.
877 			 */
878 			lp->lwp_flags |= LWP_OLDMASK;
879 		} else {
880 			/*
881 			 * No handler to run. Restore previous mask immediately.
882 			 */
883 			lp->lwp_sigmask = lp->lwp_oldsigmask;
884 		}
885 		lwkt_reltoken(&lp->lwp_proc->p_token);
886 	}
887 
888 	return (error);
889 }
890 
891 static int
892 select_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
893 {
894 	struct select_kevent_copyin_args *skap = NULL;
895 	struct kevent *kev;
896 	int fd;
897 	kfd_set *fdp = NULL;
898 	short filter = 0;
899 	u_int fflags = 0;
900 
901 	skap = (struct select_kevent_copyin_args *)arg;
902 
903 	if (*events == maxevents)
904 		return (0);
905 
906 	while (skap->active_set < COPYIN_DONE) {
907 		switch (skap->active_set) {
908 		case COPYIN_READ:
909 			/*
910 			 * Register descriptors for the read filter
911 			 */
912 			fdp = skap->read_set;
913 			filter = EVFILT_READ;
914 			fflags = NOTE_OLDAPI;
915 			if (fdp)
916 				break;
917 			++skap->active_set;
918 			skap->proc_fds = 0;
919 			/* fall through */
920 		case COPYIN_WRITE:
921 			/*
922 			 * Register descriptors for the write filter
923 			 */
924 			fdp = skap->write_set;
925 			filter = EVFILT_WRITE;
926 			fflags = NOTE_OLDAPI;
927 			if (fdp)
928 				break;
929 			++skap->active_set;
930 			skap->proc_fds = 0;
931 			/* fall through */
932 		case COPYIN_EXCEPT:
933 			/*
934 			 * Register descriptors for the exception filter
935 			 */
936 			fdp = skap->except_set;
937 			filter = EVFILT_EXCEPT;
938 			fflags = NOTE_OLDAPI | NOTE_OOB;
939 			if (fdp)
940 				break;
941 			++skap->active_set;
942 			skap->proc_fds = 0;
943 			/* fall through */
944 		case COPYIN_DONE:
945 			/*
946 			 * Nothing left to register
947 			 */
948 			return(0);
949 			/* NOT REACHED */
950 		}
951 
952 		while (skap->proc_fds < skap->num_fds) {
953 			fd = skap->proc_fds;
954 			if (FD_ISSET(fd, fdp)) {
955 				kev = &kevp[*events];
956 				EV_SET(kev, fd, filter,
957 				       EV_ADD|EV_ENABLE,
958 				       fflags, 0,
959 				       (void *)(uintptr_t)
960 					skap->lwp->lwp_kqueue_serial);
961 				FD_CLR(fd, fdp);
962 				++*events;
963 
964 				if (nseldebug) {
965 					kprintf("select fd %d filter %d "
966 					    "serial %ju\n", fd, filter,
967 					    (uintmax_t)
968 					    skap->lwp->lwp_kqueue_serial);
969 				}
970 			}
971 			++skap->proc_fds;
972 			if (*events == maxevents)
973 				return (0);
974 		}
975 		skap->active_set++;
976 		skap->proc_fds = 0;
977 	}
978 
979 	return (0);
980 }
981 
982 static int
983 select_copyout(void *arg, struct kevent *kevp, int count, int *res)
984 {
985 	struct select_kevent_copyin_args *skap;
986 	struct kevent kev;
987 	int i;
988 	int n;
989 
990 	skap = (struct select_kevent_copyin_args *)arg;
991 
992 	for (i = 0; i < count; ++i) {
993 		/*
994 		 * Filter out and delete spurious events
995 		 */
996 		if ((uint64_t)(uintptr_t)kevp[i].udata !=
997 		    skap->lwp->lwp_kqueue_serial) {
998 			kev = kevp[i];
999 			kev.flags = EV_DISABLE|EV_DELETE;
1000 			n = 1;
1001 			kqueue_register(&skap->lwp->lwp_kqueue, &kev, &n);
1002 			if (nseldebug) {
1003 				kprintf("select fd %ju mismatched serial %ju\n",
1004 				    (uintmax_t)kevp[i].ident,
1005 				    (uintmax_t)skap->lwp->lwp_kqueue_serial);
1006 			}
1007 			continue;
1008 		}
1009 
1010 		/*
1011 		 * Handle errors
1012 		 */
1013 		if (kevp[i].flags & EV_ERROR) {
1014 			int error = kevp[i].data;
1015 
1016 			switch (error) {
1017 			case EBADF:
1018 				/*
1019 				 * A bad file descriptor is considered a
1020 				 * fatal error for select, bail out.
1021 				 */
1022 				skap->error = error;
1023 				*res = -1;
1024 				return error;
1025 
1026 			default:
1027 				/*
1028 				 * Select silently swallows any unknown errors
1029 				 * for descriptors in the read or write sets.
1030 				 *
1031 				 * ALWAYS filter out EOPNOTSUPP errors from
1032 				 * filters (at least until all filters support
1033 				 * EVFILT_EXCEPT)
1034 				 *
1035 				 * We also filter out ENODEV since dev_dkqfilter
1036 				 * returns ENODEV if EOPNOTSUPP is returned in an
1037 				 * inner call.
1038 				 *
1039 				 * XXX: fix this
1040 				 */
1041 				if (kevp[i].filter != EVFILT_READ &&
1042 				    kevp[i].filter != EVFILT_WRITE &&
1043 				    error != EOPNOTSUPP &&
1044 				    error != ENODEV) {
1045 					skap->error = error;
1046 					*res = -1;
1047 					return error;
1048 				}
1049 				break;
1050 			}
1051 			if (nseldebug)
1052 				kprintf("select fd %ju filter %d error %d\n",
1053 					(uintmax_t)kevp[i].ident,
1054 					kevp[i].filter, error);
1055 			continue;
1056 		}
1057 
1058 		switch (kevp[i].filter) {
1059 		case EVFILT_READ:
1060 			FD_SET(kevp[i].ident, skap->read_set);
1061 			break;
1062 		case EVFILT_WRITE:
1063 			FD_SET(kevp[i].ident, skap->write_set);
1064 			break;
1065 		case EVFILT_EXCEPT:
1066 			FD_SET(kevp[i].ident, skap->except_set);
1067 			break;
1068 		}
1069 
1070 		++*res;
1071 	}
1072 
1073 	return (0);
1074 }
1075 
1076 /*
1077  * Copy select bits in from userland.  Allocate kernel memory if the
1078  * set is large.
1079  */
1080 static int
1081 getbits(int bytes, fd_set *in_set, kfd_set **out_set, kfd_set *tmp_set)
1082 {
1083 	int error;
1084 
1085 	if (in_set) {
1086 		if (bytes < sizeof(*tmp_set))
1087 			*out_set = tmp_set;
1088 		else
1089 			*out_set = kmalloc(bytes, M_SELECT, M_WAITOK);
1090 		error = copyin(in_set, *out_set, bytes);
1091 	} else {
1092 		*out_set = NULL;
1093 		error = 0;
1094 	}
1095 	return (error);
1096 }
1097 
1098 /*
1099  * Copy returned select bits back out to userland.
1100  */
1101 static int
1102 putbits(int bytes, kfd_set *in_set, fd_set *out_set)
1103 {
1104 	int error;
1105 
1106 	if (in_set) {
1107 		error = copyout(in_set, out_set, bytes);
1108 	} else {
1109 		error = 0;
1110 	}
1111 	return (error);
1112 }
1113 
1114 static int
1115 dotimeout_only(struct timespec *ts)
1116 {
1117 	return(nanosleep1(ts, NULL));
1118 }
1119 
1120 /*
1121  * Common code for sys_select() and sys_pselect().
1122  *
1123  * in, out and ex are userland pointers.  ts must point to validated
1124  * kernel-side timeout value or NULL for infinite timeout.  res must
1125  * point to syscall return value.
1126  */
1127 static int
1128 doselect(int nd, fd_set *read, fd_set *write, fd_set *except,
1129 	 struct timespec *ts, int *res)
1130 {
1131 	struct proc *p = curproc;
1132 	struct select_kevent_copyin_args *kap, ka;
1133 	int bytes, error;
1134 	kfd_set read_tmp;
1135 	kfd_set write_tmp;
1136 	kfd_set except_tmp;
1137 
1138 	*res = 0;
1139 	if (nd < 0)
1140 		return (EINVAL);
1141 	if (nd == 0 && ts)
1142 		return (dotimeout_only(ts));
1143 
1144 	if (nd > p->p_fd->fd_nfiles)		/* limit kmalloc */
1145 		nd = p->p_fd->fd_nfiles;
1146 
1147 	kap = &ka;
1148 	kap->lwp = curthread->td_lwp;
1149 	kap->num_fds = nd;
1150 	kap->proc_fds = 0;
1151 	kap->error = 0;
1152 	kap->active_set = COPYIN_READ;
1153 
1154 	/*
1155 	 * Calculate bytes based on the number of __fd_mask[] array entries
1156 	 * multiplied by the size of __fd_mask.
1157 	 */
1158 	bytes = howmany(nd, __NFDBITS) * sizeof(__fd_mask);
1159 
1160 	/* kap->read_set = NULL; not needed */
1161 	kap->write_set = NULL;
1162 	kap->except_set = NULL;
1163 
1164 	error = getbits(bytes, read, &kap->read_set, &read_tmp);
1165 	if (error == 0)
1166 		error = getbits(bytes, write, &kap->write_set, &write_tmp);
1167 	if (error == 0)
1168 		error = getbits(bytes, except, &kap->except_set, &except_tmp);
1169 	if (error)
1170 		goto done;
1171 
1172 	/*
1173 	 * NOTE: Make sure the max events passed to kern_kevent() is
1174 	 *	 effectively unlimited.  (nd * 3) accomplishes this.
1175 	 *
1176 	 *	 (*res) continues to increment as returned events are
1177 	 *	 loaded in.
1178 	 */
1179 	error = kern_kevent(&kap->lwp->lwp_kqueue, 0x7FFFFFFF, res, kap,
1180 			    select_copyin, select_copyout, ts, 0);
1181 	if (error == 0)
1182 		error = putbits(bytes, kap->read_set, read);
1183 	if (error == 0)
1184 		error = putbits(bytes, kap->write_set, write);
1185 	if (error == 0)
1186 		error = putbits(bytes, kap->except_set, except);
1187 
1188 	/*
1189 	 * An error from an individual event that should be passed
1190 	 * back to userland (EBADF)
1191 	 */
1192 	if (kap->error)
1193 		error = kap->error;
1194 
1195 	/*
1196 	 * Clean up.
1197 	 */
1198 done:
1199 	if (kap->read_set && kap->read_set != &read_tmp)
1200 		kfree(kap->read_set, M_SELECT);
1201 	if (kap->write_set && kap->write_set != &write_tmp)
1202 		kfree(kap->write_set, M_SELECT);
1203 	if (kap->except_set && kap->except_set != &except_tmp)
1204 		kfree(kap->except_set, M_SELECT);
1205 
1206 	kap->lwp->lwp_kqueue_serial += kap->num_fds;
1207 
1208 	return (error);
1209 }
1210 
1211 /*
1212  * Poll system call.
1213  *
1214  * MPSAFE
1215  */
1216 int
1217 sys_poll(struct poll_args *uap)
1218 {
1219 	struct timespec ts, *tsp;
1220 	int error;
1221 
1222 	if (uap->timeout != INFTIM) {
1223 		if (uap->timeout < 0)
1224 			return (EINVAL);
1225 		ts.tv_sec = uap->timeout / 1000;
1226 		ts.tv_nsec = (uap->timeout % 1000) * 1000 * 1000;
1227 		tsp = &ts;
1228 	} else {
1229 		tsp = NULL;
1230 	}
1231 
1232 	error = dopoll(uap->nfds, uap->fds, tsp, &uap->sysmsg_result, 0);
1233 
1234 	return (error);
1235 }
1236 
1237 /*
1238  * Ppoll system call.
1239  *
1240  * MPSAFE
1241  */
1242 int
1243 sys_ppoll(struct ppoll_args *uap)
1244 {
1245 	struct thread *td = curthread;
1246 	struct lwp *lp = td->td_lwp;
1247 	struct timespec *ktsp, kts;
1248 	sigset_t sigmask;
1249 	int error;
1250 
1251 	/*
1252 	 * Get timeout if any.
1253 	 */
1254 	if (uap->ts != NULL) {
1255 		error = copyin(uap->ts, &kts, sizeof (kts));
1256 		if (error)
1257 			return (error);
1258 		ktsp = &kts;
1259 	} else {
1260 		ktsp = NULL;
1261 	}
1262 
1263 	/*
1264 	 * Install temporary signal mask if any provided.
1265 	 */
1266 	if (uap->sigmask != NULL) {
1267 		error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
1268 		if (error)
1269 			return (error);
1270 		lwkt_gettoken(&lp->lwp_proc->p_token);
1271 		lp->lwp_oldsigmask = lp->lwp_sigmask;
1272 		SIG_CANTMASK(sigmask);
1273 		lp->lwp_sigmask = sigmask;
1274 		lwkt_reltoken(&lp->lwp_proc->p_token);
1275 	}
1276 
1277 	error = dopoll(uap->nfds, uap->fds, ktsp, &uap->sysmsg_result,
1278 	    ktsp != NULL ? KEVENT_TIMEOUT_PRECISE : 0);
1279 
1280 	if (uap->sigmask != NULL) {
1281 		lwkt_gettoken(&lp->lwp_proc->p_token);
1282 		/* dopoll() responsible for turning ERESTART into EINTR */
1283 		KKASSERT(error != ERESTART);
1284 		if (error == EINTR) {
1285 			/*
1286 			 * We can't restore the previous signal mask now
1287 			 * because it could block the signal that interrupted
1288 			 * us.  So make a note to restore it after executing
1289 			 * the handler.
1290 			 */
1291 			lp->lwp_flags |= LWP_OLDMASK;
1292 		} else {
1293 			/*
1294 			 * No handler to run. Restore previous mask immediately.
1295 			 */
1296 			lp->lwp_sigmask = lp->lwp_oldsigmask;
1297 		}
1298 		lwkt_reltoken(&lp->lwp_proc->p_token);
1299 	}
1300 
1301 	return (error);
1302 }
1303 
1304 static int
1305 poll_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1306 {
1307 	struct poll_kevent_copyin_args *pkap;
1308 	struct pollfd *pfd;
1309 	struct kevent *kev;
1310 	int kev_count;
1311 
1312 	pkap = (struct poll_kevent_copyin_args *)arg;
1313 
1314 	while (pkap->pfds < pkap->nfds) {
1315 		pfd = &pkap->fds[pkap->pfds];
1316 
1317 		/* Clear return events */
1318 		pfd->revents = 0;
1319 
1320 		/* Do not check if fd is equal to -1 */
1321 		if (pfd->fd == -1) {
1322 			++pkap->pfds;
1323 			continue;
1324 		}
1325 
1326 		kev_count = 0;
1327 		if (pfd->events & (POLLIN | POLLRDNORM))
1328 			kev_count++;
1329 		if (pfd->events & (POLLOUT | POLLWRNORM))
1330 			kev_count++;
1331 		if (pfd->events & (POLLPRI | POLLRDBAND))
1332 			kev_count++;
1333 
1334 		if (*events + kev_count > maxevents)
1335 			return (0);
1336 
1337 		/*
1338 		 * NOTE: A combined serial number and poll array index is
1339 		 * stored in kev->udata.
1340 		 */
1341 		kev = &kevp[*events];
1342 		if (pfd->events & (POLLIN | POLLRDNORM)) {
1343 			EV_SET(kev++, pfd->fd, EVFILT_READ, EV_ADD|EV_ENABLE,
1344 			       NOTE_OLDAPI, 0, (void *)(uintptr_t)
1345 				(pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1346 		}
1347 		if (pfd->events & (POLLOUT | POLLWRNORM)) {
1348 			EV_SET(kev++, pfd->fd, EVFILT_WRITE, EV_ADD|EV_ENABLE,
1349 			       NOTE_OLDAPI, 0, (void *)(uintptr_t)
1350 				(pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1351 		}
1352 		if (pfd->events & (POLLPRI | POLLRDBAND)) {
1353 			EV_SET(kev++, pfd->fd, EVFILT_EXCEPT, EV_ADD|EV_ENABLE,
1354 			       NOTE_OLDAPI | NOTE_OOB, 0,
1355 			       (void *)(uintptr_t)
1356 				(pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1357 		}
1358 
1359 		if (nseldebug) {
1360 			kprintf("poll index %d/%d fd %d events %08x "
1361 			    "serial %ju\n", pkap->pfds, pkap->nfds-1,
1362 			    pfd->fd, pfd->events,
1363 			    (uintmax_t)pkap->lwp->lwp_kqueue_serial);
1364 		}
1365 
1366 		++pkap->pfds;
1367 		(*events) += kev_count;
1368 	}
1369 
1370 	return (0);
1371 }
1372 
1373 static int
1374 poll_copyout(void *arg, struct kevent *kevp, int count, int *res)
1375 {
1376 	struct poll_kevent_copyin_args *pkap;
1377 	struct pollfd *pfd;
1378 	struct kevent kev;
1379 	int count_res;
1380 	int i;
1381 	int n;
1382 	uint64_t pi;
1383 
1384 	pkap = (struct poll_kevent_copyin_args *)arg;
1385 
1386 	for (i = 0; i < count; ++i) {
1387 		/*
1388 		 * Extract the poll array index and delete spurious events.
1389 		 * We can easily tell if the serial number is incorrect
1390 		 * by checking whether the extracted index is out of range.
1391 		 */
1392 		pi = (uint64_t)(uintptr_t)kevp[i].udata -
1393 		    pkap->lwp->lwp_kqueue_serial;
1394 
1395 		if (pi >= pkap->nfds) {
1396 			kev = kevp[i];
1397 			kev.flags = EV_DISABLE|EV_DELETE;
1398 			n = 1;
1399 			kqueue_register(&pkap->lwp->lwp_kqueue, &kev, &n);
1400 			if (nseldebug) {
1401 				kprintf("poll index %ju out of range against "
1402 				    "serial %ju\n", (uintmax_t)pi,
1403 				    (uintmax_t)pkap->lwp->lwp_kqueue_serial);
1404 			}
1405 			continue;
1406 		}
1407 		pfd = &pkap->fds[pi];
1408 		if (kevp[i].ident == pfd->fd) {
1409 			/*
1410 			 * A single descriptor may generate an error against
1411 			 * more than one filter, make sure to set the
1412 			 * appropriate flags but do not increment (*res)
1413 			 * more than once.
1414 			 */
1415 			count_res = (pfd->revents == 0);
1416 			if (kevp[i].flags & EV_ERROR) {
1417 				switch(kevp[i].data) {
1418 				case EBADF:
1419 				case POLLNVAL:
1420 					/* Bad file descriptor */
1421 					if (count_res)
1422 						++*res;
1423 					pfd->revents |= POLLNVAL;
1424 					break;
1425 				default:
1426 					/*
1427 					 * Poll silently swallows any unknown
1428 					 * errors except in the case of POLLPRI
1429 					 * (OOB/urgent data).
1430 					 *
1431 					 * ALWAYS filter out EOPNOTSUPP errors
1432 					 * from filters, common applications
1433 					 * set POLLPRI|POLLRDBAND and most
1434 					 * filters do not support EVFILT_EXCEPT.
1435 					 *
1436 					 * We also filter out ENODEV since dev_dkqfilter
1437 					 * returns ENODEV if EOPNOTSUPP is returned in an
1438 					 * inner call.
1439 					 *
1440 					 * XXX: fix this
1441 					 */
1442 					if (kevp[i].filter != EVFILT_READ &&
1443 					    kevp[i].filter != EVFILT_WRITE &&
1444 					    kevp[i].data != EOPNOTSUPP &&
1445 					    kevp[i].data != ENODEV) {
1446 						if (count_res)
1447 							++*res;
1448 						pfd->revents |= POLLERR;
1449 					}
1450 					break;
1451 				}
1452 				if (nseldebug) {
1453 					kprintf("poll index %ju fd %d "
1454 						"filter %d error %jd\n",
1455 						(uintmax_t)pi, pfd->fd,
1456 						kevp[i].filter,
1457 						(intmax_t)kevp[i].data);
1458 				}
1459 				continue;
1460 			}
1461 
1462 			switch (kevp[i].filter) {
1463 			case EVFILT_READ:
1464 				/*
1465 				 * NODATA on the read side can indicate a
1466 				 * half-closed situation and not necessarily
1467 				 * a disconnect, so depend on the user
1468 				 * issuing a read() and getting 0 bytes back.
1469 				 *
1470 				 * If EV_HUP is set the peer completely
1471 				 * disconnected and we can set POLLHUP
1472 				 * once data is exhausted.
1473 				 */
1474 				if (kevp[i].flags & EV_NODATA) {
1475 					if (kevp[i].flags & EV_HUP)
1476 						pfd->revents |= POLLHUP;
1477 				}
1478 				if ((kevp[i].flags & EV_EOF) &&
1479 				    kevp[i].fflags != 0)
1480 					pfd->revents |= POLLERR;
1481 				if (pfd->events & POLLIN)
1482 					pfd->revents |= POLLIN;
1483 				if (pfd->events & POLLRDNORM)
1484 					pfd->revents |= POLLRDNORM;
1485 				break;
1486 			case EVFILT_WRITE:
1487 				/*
1488 				 * As per the OpenGroup POLLHUP is mutually
1489 				 * exclusive with the writability flags.  I
1490 				 * consider this a bit broken but...
1491 				 *
1492 				 * In this case a disconnect is implied even
1493 				 * for a half-closed (write side) situation.
1494 				 */
1495 				if (kevp[i].flags & EV_EOF) {
1496 					pfd->revents |= POLLHUP;
1497 					if (kevp[i].fflags != 0)
1498 						pfd->revents |= POLLERR;
1499 				} else {
1500 					if (pfd->events & POLLOUT)
1501 						pfd->revents |= POLLOUT;
1502 					if (pfd->events & POLLWRNORM)
1503 						pfd->revents |= POLLWRNORM;
1504 				}
1505 				break;
1506 			case EVFILT_EXCEPT:
1507 				/*
1508 				 * EV_NODATA should never be tagged for this
1509 				 * filter.
1510 				 */
1511 				if (pfd->events & POLLPRI)
1512 					pfd->revents |= POLLPRI;
1513 				if (pfd->events & POLLRDBAND)
1514 					pfd->revents |= POLLRDBAND;
1515 				break;
1516 			}
1517 
1518 			if (nseldebug) {
1519 				kprintf("poll index %ju/%d fd %d "
1520 				    "revents %08x\n", (uintmax_t)pi, pkap->nfds,
1521 				    pfd->fd, pfd->revents);
1522 			}
1523 
1524 			if (count_res && pfd->revents)
1525 				++*res;
1526 		} else {
1527 			if (nseldebug) {
1528 				kprintf("poll index %ju mismatch %ju/%d\n",
1529 				    (uintmax_t)pi, (uintmax_t)kevp[i].ident,
1530 				    pfd->fd);
1531 			}
1532 		}
1533 	}
1534 
1535 	return (0);
1536 }
1537 
1538 static int
1539 dopoll(int nfds, struct pollfd *fds, struct timespec *ts, int *res, int flags)
1540 {
1541 	struct poll_kevent_copyin_args ka;
1542 	struct pollfd sfds[64];
1543 	int bytes;
1544 	int error;
1545 
1546         *res = 0;
1547         if (nfds < 0)
1548                 return (EINVAL);
1549 
1550 	if (nfds == 0 && ts)
1551 		return (dotimeout_only(ts));
1552 
1553 	/*
1554 	 * This is a bit arbitrary but we need to limit internal kmallocs.
1555 	 */
1556         if (nfds > maxfilesperproc * 2)
1557                 nfds = maxfilesperproc * 2;
1558 	bytes = sizeof(struct pollfd) * nfds;
1559 
1560 	ka.lwp = curthread->td_lwp;
1561 	ka.nfds = nfds;
1562 	ka.pfds = 0;
1563 	ka.error = 0;
1564 
1565 	if (ka.nfds < 64)
1566 		ka.fds = sfds;
1567 	else
1568 		ka.fds = kmalloc(bytes, M_SELECT, M_WAITOK);
1569 
1570 	error = copyin(fds, ka.fds, bytes);
1571 	if (error == 0)
1572 		error = kern_kevent(&ka.lwp->lwp_kqueue, 0x7FFFFFFF, res, &ka,
1573 				    poll_copyin, poll_copyout, ts, flags);
1574 
1575 	if (error == 0)
1576 		error = copyout(ka.fds, fds, bytes);
1577 
1578 	if (ka.fds != sfds)
1579 		kfree(ka.fds, M_SELECT);
1580 
1581 	ka.lwp->lwp_kqueue_serial += nfds;
1582 
1583 	return (error);
1584 }
1585 
1586 static int
1587 socket_wait_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1588 {
1589 	return (0);
1590 }
1591 
1592 static int
1593 socket_wait_copyout(void *arg, struct kevent *kevp, int count, int *res)
1594 {
1595 	++*res;
1596 	return (0);
1597 }
1598 
1599 extern	struct fileops socketops;
1600 
1601 /*
1602  * NOTE: Callers of socket_wait() must already have a reference on the
1603  *	 socket.
1604  */
1605 int
1606 socket_wait(struct socket *so, struct timespec *ts, int *res)
1607 {
1608 	struct thread *td = curthread;
1609 	struct file *fp;
1610 	struct kqueue kq;
1611 	struct kevent kev;
1612 	int error, fd;
1613 	int n;
1614 
1615 	if ((error = falloc(td->td_lwp, &fp, &fd)) != 0)
1616 		return (error);
1617 
1618 	fp->f_type = DTYPE_SOCKET;
1619 	fp->f_flag = FREAD | FWRITE;
1620 	fp->f_ops = &socketops;
1621 	fp->f_data = so;
1622 	fsetfd(td->td_lwp->lwp_proc->p_fd, fp, fd);
1623 	fsetfdflags(td->td_proc->p_fd, fd, UF_EXCLOSE);
1624 
1625 	bzero(&kq, sizeof(kq));
1626 	kqueue_init(&kq, td->td_lwp->lwp_proc->p_fd);
1627 	EV_SET(&kev, fd, EVFILT_READ, EV_ADD|EV_ENABLE, 0, 0, NULL);
1628 	n = 1;
1629 	if ((error = kqueue_register(&kq, &kev, &n)) != 0) {
1630 		fdrop(fp);
1631 		return (error);
1632 	}
1633 
1634 	error = kern_kevent(&kq, 1, res, NULL, socket_wait_copyin,
1635 			    socket_wait_copyout, ts, 0);
1636 
1637 	EV_SET(&kev, fd, EVFILT_READ, EV_DELETE|EV_DISABLE, 0, 0, NULL);
1638 	n = 1;
1639 	kqueue_register(&kq, &kev, &n);
1640 	fp->f_ops = &badfileops;
1641 	fdrop(fp);
1642 
1643 	return (error);
1644 }
1645 
1646 /*
1647  * OpenBSD poll system call.
1648  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
1649  *
1650  * MPSAFE
1651  */
1652 int
1653 sys_openbsd_poll(struct openbsd_poll_args *uap)
1654 {
1655 	return (sys_poll((struct poll_args *)uap));
1656 }
1657 
1658 /*ARGSUSED*/
1659 int
1660 seltrue(cdev_t dev, int events)
1661 {
1662 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
1663 }
1664