xref: /freebsd/sys/kern/sys_generic.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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  * 4. 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  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_compat.h"
41 #include "opt_ktrace.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/filedesc.h>
47 #include <sys/filio.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/proc.h>
51 #include <sys/signalvar.h>
52 #include <sys/socketvar.h>
53 #include <sys/uio.h>
54 #include <sys/kernel.h>
55 #include <sys/ktr.h>
56 #include <sys/limits.h>
57 #include <sys/malloc.h>
58 #include <sys/poll.h>
59 #include <sys/resourcevar.h>
60 #include <sys/selinfo.h>
61 #include <sys/sleepqueue.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/vnode.h>
66 #include <sys/bio.h>
67 #include <sys/buf.h>
68 #include <sys/condvar.h>
69 #ifdef KTRACE
70 #include <sys/ktrace.h>
71 #endif
72 
73 #include <security/audit/audit.h>
74 
75 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
76 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
77 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
78 
79 static int	pollout(struct pollfd *, struct pollfd *, u_int);
80 static int	pollscan(struct thread *, struct pollfd *, u_int);
81 static int	pollrescan(struct thread *);
82 static int	selscan(struct thread *, fd_mask **, fd_mask **, int);
83 static int	selrescan(struct thread *, fd_mask **, fd_mask **);
84 static void	selfdalloc(struct thread *, void *);
85 static void	selfdfree(struct seltd *, struct selfd *);
86 static int	dofileread(struct thread *, int, struct file *, struct uio *,
87 		    off_t, int);
88 static int	dofilewrite(struct thread *, int, struct file *, struct uio *,
89 		    off_t, int);
90 static void	doselwakeup(struct selinfo *, int);
91 static void	seltdinit(struct thread *);
92 static int	seltdwait(struct thread *, int);
93 static void	seltdclear(struct thread *);
94 
95 /*
96  * One seltd per-thread allocated on demand as needed.
97  *
98  *	t - protected by st_mtx
99  * 	k - Only accessed by curthread or read-only
100  */
101 struct seltd {
102 	STAILQ_HEAD(, selfd)	st_selq;	/* (k) List of selfds. */
103 	struct selfd		*st_free1;	/* (k) free fd for read set. */
104 	struct selfd		*st_free2;	/* (k) free fd for write set. */
105 	struct mtx		st_mtx;		/* Protects struct seltd */
106 	struct cv		st_wait;	/* (t) Wait channel. */
107 	int			st_flags;	/* (t) SELTD_ flags. */
108 };
109 
110 #define	SELTD_PENDING	0x0001			/* We have pending events. */
111 #define	SELTD_RESCAN	0x0002			/* Doing a rescan. */
112 
113 /*
114  * One selfd allocated per-thread per-file-descriptor.
115  *	f - protected by sf_mtx
116  */
117 struct selfd {
118 	STAILQ_ENTRY(selfd)	sf_link;	/* (k) fds owned by this td. */
119 	TAILQ_ENTRY(selfd)	sf_threads;	/* (f) fds on this selinfo. */
120 	struct selinfo		*sf_si;		/* (f) selinfo when linked. */
121 	struct mtx		*sf_mtx;	/* Pointer to selinfo mtx. */
122 	struct seltd		*sf_td;		/* (k) owning seltd. */
123 	void			*sf_cookie;	/* (k) fd or pollfd. */
124 };
125 
126 static uma_zone_t selfd_zone;
127 static struct mtx_pool *mtxpool_select;
128 
129 #ifndef _SYS_SYSPROTO_H_
130 struct read_args {
131 	int	fd;
132 	void	*buf;
133 	size_t	nbyte;
134 };
135 #endif
136 int
137 read(td, uap)
138 	struct thread *td;
139 	struct read_args *uap;
140 {
141 	struct uio auio;
142 	struct iovec aiov;
143 	int error;
144 
145 	if (uap->nbyte > INT_MAX)
146 		return (EINVAL);
147 	aiov.iov_base = uap->buf;
148 	aiov.iov_len = uap->nbyte;
149 	auio.uio_iov = &aiov;
150 	auio.uio_iovcnt = 1;
151 	auio.uio_resid = uap->nbyte;
152 	auio.uio_segflg = UIO_USERSPACE;
153 	error = kern_readv(td, uap->fd, &auio);
154 	return(error);
155 }
156 
157 /*
158  * Positioned read system call
159  */
160 #ifndef _SYS_SYSPROTO_H_
161 struct pread_args {
162 	int	fd;
163 	void	*buf;
164 	size_t	nbyte;
165 	int	pad;
166 	off_t	offset;
167 };
168 #endif
169 int
170 pread(td, uap)
171 	struct thread *td;
172 	struct pread_args *uap;
173 {
174 	struct uio auio;
175 	struct iovec aiov;
176 	int error;
177 
178 	if (uap->nbyte > INT_MAX)
179 		return (EINVAL);
180 	aiov.iov_base = uap->buf;
181 	aiov.iov_len = uap->nbyte;
182 	auio.uio_iov = &aiov;
183 	auio.uio_iovcnt = 1;
184 	auio.uio_resid = uap->nbyte;
185 	auio.uio_segflg = UIO_USERSPACE;
186 	error = kern_preadv(td, uap->fd, &auio, uap->offset);
187 	return(error);
188 }
189 
190 int
191 freebsd6_pread(td, uap)
192 	struct thread *td;
193 	struct freebsd6_pread_args *uap;
194 {
195 	struct pread_args oargs;
196 
197 	oargs.fd = uap->fd;
198 	oargs.buf = uap->buf;
199 	oargs.nbyte = uap->nbyte;
200 	oargs.offset = uap->offset;
201 	return (pread(td, &oargs));
202 }
203 
204 /*
205  * Scatter read system call.
206  */
207 #ifndef _SYS_SYSPROTO_H_
208 struct readv_args {
209 	int	fd;
210 	struct	iovec *iovp;
211 	u_int	iovcnt;
212 };
213 #endif
214 int
215 readv(struct thread *td, struct readv_args *uap)
216 {
217 	struct uio *auio;
218 	int error;
219 
220 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
221 	if (error)
222 		return (error);
223 	error = kern_readv(td, uap->fd, auio);
224 	free(auio, M_IOV);
225 	return (error);
226 }
227 
228 int
229 kern_readv(struct thread *td, int fd, struct uio *auio)
230 {
231 	struct file *fp;
232 	int error;
233 
234 	error = fget_read(td, fd, &fp);
235 	if (error)
236 		return (error);
237 	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
238 	fdrop(fp, td);
239 	return (error);
240 }
241 
242 /*
243  * Scatter positioned read system call.
244  */
245 #ifndef _SYS_SYSPROTO_H_
246 struct preadv_args {
247 	int	fd;
248 	struct	iovec *iovp;
249 	u_int	iovcnt;
250 	off_t	offset;
251 };
252 #endif
253 int
254 preadv(struct thread *td, struct preadv_args *uap)
255 {
256 	struct uio *auio;
257 	int error;
258 
259 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
260 	if (error)
261 		return (error);
262 	error = kern_preadv(td, uap->fd, auio, uap->offset);
263 	free(auio, M_IOV);
264 	return (error);
265 }
266 
267 int
268 kern_preadv(td, fd, auio, offset)
269 	struct thread *td;
270 	int fd;
271 	struct uio *auio;
272 	off_t offset;
273 {
274 	struct file *fp;
275 	int error;
276 
277 	error = fget_read(td, fd, &fp);
278 	if (error)
279 		return (error);
280 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
281 		error = ESPIPE;
282 	else if (offset < 0 && fp->f_vnode->v_type != VCHR)
283 		error = EINVAL;
284 	else
285 		error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
286 	fdrop(fp, td);
287 	return (error);
288 }
289 
290 /*
291  * Common code for readv and preadv that reads data in
292  * from a file using the passed in uio, offset, and flags.
293  */
294 static int
295 dofileread(td, fd, fp, auio, offset, flags)
296 	struct thread *td;
297 	int fd;
298 	struct file *fp;
299 	struct uio *auio;
300 	off_t offset;
301 	int flags;
302 {
303 	ssize_t cnt;
304 	int error;
305 #ifdef KTRACE
306 	struct uio *ktruio = NULL;
307 #endif
308 
309 	/* Finish zero length reads right here */
310 	if (auio->uio_resid == 0) {
311 		td->td_retval[0] = 0;
312 		return(0);
313 	}
314 	auio->uio_rw = UIO_READ;
315 	auio->uio_offset = offset;
316 	auio->uio_td = td;
317 #ifdef KTRACE
318 	if (KTRPOINT(td, KTR_GENIO))
319 		ktruio = cloneuio(auio);
320 #endif
321 	cnt = auio->uio_resid;
322 	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
323 		if (auio->uio_resid != cnt && (error == ERESTART ||
324 		    error == EINTR || error == EWOULDBLOCK))
325 			error = 0;
326 	}
327 	cnt -= auio->uio_resid;
328 #ifdef KTRACE
329 	if (ktruio != NULL) {
330 		ktruio->uio_resid = cnt;
331 		ktrgenio(fd, UIO_READ, ktruio, error);
332 	}
333 #endif
334 	td->td_retval[0] = cnt;
335 	return (error);
336 }
337 
338 #ifndef _SYS_SYSPROTO_H_
339 struct write_args {
340 	int	fd;
341 	const void *buf;
342 	size_t	nbyte;
343 };
344 #endif
345 int
346 write(td, uap)
347 	struct thread *td;
348 	struct write_args *uap;
349 {
350 	struct uio auio;
351 	struct iovec aiov;
352 	int error;
353 
354 	if (uap->nbyte > INT_MAX)
355 		return (EINVAL);
356 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
357 	aiov.iov_len = uap->nbyte;
358 	auio.uio_iov = &aiov;
359 	auio.uio_iovcnt = 1;
360 	auio.uio_resid = uap->nbyte;
361 	auio.uio_segflg = UIO_USERSPACE;
362 	error = kern_writev(td, uap->fd, &auio);
363 	return(error);
364 }
365 
366 /*
367  * Positioned write system call.
368  */
369 #ifndef _SYS_SYSPROTO_H_
370 struct pwrite_args {
371 	int	fd;
372 	const void *buf;
373 	size_t	nbyte;
374 	int	pad;
375 	off_t	offset;
376 };
377 #endif
378 int
379 pwrite(td, uap)
380 	struct thread *td;
381 	struct pwrite_args *uap;
382 {
383 	struct uio auio;
384 	struct iovec aiov;
385 	int error;
386 
387 	if (uap->nbyte > INT_MAX)
388 		return (EINVAL);
389 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
390 	aiov.iov_len = uap->nbyte;
391 	auio.uio_iov = &aiov;
392 	auio.uio_iovcnt = 1;
393 	auio.uio_resid = uap->nbyte;
394 	auio.uio_segflg = UIO_USERSPACE;
395 	error = kern_pwritev(td, uap->fd, &auio, uap->offset);
396 	return(error);
397 }
398 
399 int
400 freebsd6_pwrite(td, uap)
401 	struct thread *td;
402 	struct freebsd6_pwrite_args *uap;
403 {
404 	struct pwrite_args oargs;
405 
406 	oargs.fd = uap->fd;
407 	oargs.buf = uap->buf;
408 	oargs.nbyte = uap->nbyte;
409 	oargs.offset = uap->offset;
410 	return (pwrite(td, &oargs));
411 }
412 
413 /*
414  * Gather write system call.
415  */
416 #ifndef _SYS_SYSPROTO_H_
417 struct writev_args {
418 	int	fd;
419 	struct	iovec *iovp;
420 	u_int	iovcnt;
421 };
422 #endif
423 int
424 writev(struct thread *td, struct writev_args *uap)
425 {
426 	struct uio *auio;
427 	int error;
428 
429 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
430 	if (error)
431 		return (error);
432 	error = kern_writev(td, uap->fd, auio);
433 	free(auio, M_IOV);
434 	return (error);
435 }
436 
437 int
438 kern_writev(struct thread *td, int fd, struct uio *auio)
439 {
440 	struct file *fp;
441 	int error;
442 
443 	error = fget_write(td, fd, &fp);
444 	if (error)
445 		return (error);
446 	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
447 	fdrop(fp, td);
448 	return (error);
449 }
450 
451 /*
452  * Gather positioned write system call.
453  */
454 #ifndef _SYS_SYSPROTO_H_
455 struct pwritev_args {
456 	int	fd;
457 	struct	iovec *iovp;
458 	u_int	iovcnt;
459 	off_t	offset;
460 };
461 #endif
462 int
463 pwritev(struct thread *td, struct pwritev_args *uap)
464 {
465 	struct uio *auio;
466 	int error;
467 
468 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
469 	if (error)
470 		return (error);
471 	error = kern_pwritev(td, uap->fd, auio, uap->offset);
472 	free(auio, M_IOV);
473 	return (error);
474 }
475 
476 int
477 kern_pwritev(td, fd, auio, offset)
478 	struct thread *td;
479 	struct uio *auio;
480 	int fd;
481 	off_t offset;
482 {
483 	struct file *fp;
484 	int error;
485 
486 	error = fget_write(td, fd, &fp);
487 	if (error)
488 		return (error);
489 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
490 		error = ESPIPE;
491 	else if (offset < 0 && fp->f_vnode->v_type != VCHR)
492 		error = EINVAL;
493 	else
494 		error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
495 	fdrop(fp, td);
496 	return (error);
497 }
498 
499 /*
500  * Common code for writev and pwritev that writes data to
501  * a file using the passed in uio, offset, and flags.
502  */
503 static int
504 dofilewrite(td, fd, fp, auio, offset, flags)
505 	struct thread *td;
506 	int fd;
507 	struct file *fp;
508 	struct uio *auio;
509 	off_t offset;
510 	int flags;
511 {
512 	ssize_t cnt;
513 	int error;
514 #ifdef KTRACE
515 	struct uio *ktruio = NULL;
516 #endif
517 
518 	auio->uio_rw = UIO_WRITE;
519 	auio->uio_td = td;
520 	auio->uio_offset = offset;
521 #ifdef KTRACE
522 	if (KTRPOINT(td, KTR_GENIO))
523 		ktruio = cloneuio(auio);
524 #endif
525 	cnt = auio->uio_resid;
526 	if (fp->f_type == DTYPE_VNODE)
527 		bwillwrite();
528 	if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
529 		if (auio->uio_resid != cnt && (error == ERESTART ||
530 		    error == EINTR || error == EWOULDBLOCK))
531 			error = 0;
532 		/* Socket layer is responsible for issuing SIGPIPE. */
533 		if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
534 			PROC_LOCK(td->td_proc);
535 			psignal(td->td_proc, SIGPIPE);
536 			PROC_UNLOCK(td->td_proc);
537 		}
538 	}
539 	cnt -= auio->uio_resid;
540 #ifdef KTRACE
541 	if (ktruio != NULL) {
542 		ktruio->uio_resid = cnt;
543 		ktrgenio(fd, UIO_WRITE, ktruio, error);
544 	}
545 #endif
546 	td->td_retval[0] = cnt;
547 	return (error);
548 }
549 
550 /*
551  * Truncate a file given a file descriptor.
552  *
553  * Can't use fget_write() here, since must return EINVAL and not EBADF if the
554  * descriptor isn't writable.
555  */
556 int
557 kern_ftruncate(td, fd, length)
558 	struct thread *td;
559 	int fd;
560 	off_t length;
561 {
562 	struct file *fp;
563 	int error;
564 
565 	AUDIT_ARG_FD(fd);
566 	if (length < 0)
567 		return (EINVAL);
568 	error = fget(td, fd, &fp);
569 	if (error)
570 		return (error);
571 	AUDIT_ARG_FILE(td->td_proc, fp);
572 	if (!(fp->f_flag & FWRITE)) {
573 		fdrop(fp, td);
574 		return (EINVAL);
575 	}
576 	error = fo_truncate(fp, length, td->td_ucred, td);
577 	fdrop(fp, td);
578 	return (error);
579 }
580 
581 #ifndef _SYS_SYSPROTO_H_
582 struct ftruncate_args {
583 	int	fd;
584 	int	pad;
585 	off_t	length;
586 };
587 #endif
588 int
589 ftruncate(td, uap)
590 	struct thread *td;
591 	struct ftruncate_args *uap;
592 {
593 
594 	return (kern_ftruncate(td, uap->fd, uap->length));
595 }
596 
597 #if defined(COMPAT_43)
598 #ifndef _SYS_SYSPROTO_H_
599 struct oftruncate_args {
600 	int	fd;
601 	long	length;
602 };
603 #endif
604 int
605 oftruncate(td, uap)
606 	struct thread *td;
607 	struct oftruncate_args *uap;
608 {
609 
610 	return (kern_ftruncate(td, uap->fd, uap->length));
611 }
612 #endif /* COMPAT_43 */
613 
614 #ifndef _SYS_SYSPROTO_H_
615 struct ioctl_args {
616 	int	fd;
617 	u_long	com;
618 	caddr_t	data;
619 };
620 #endif
621 /* ARGSUSED */
622 int
623 ioctl(struct thread *td, struct ioctl_args *uap)
624 {
625 	u_long com;
626 	int arg, error;
627 	u_int size;
628 	caddr_t data;
629 
630 	if (uap->com > 0xffffffff) {
631 		printf(
632 		    "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
633 		    td->td_proc->p_pid, td->td_name, uap->com);
634 		uap->com &= 0xffffffff;
635 	}
636 	com = uap->com;
637 
638 	/*
639 	 * Interpret high order word to find amount of data to be
640 	 * copied to/from the user's address space.
641 	 */
642 	size = IOCPARM_LEN(com);
643 	if ((size > IOCPARM_MAX) ||
644 	    ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
645 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
646 	    ((com & IOC_OUT) && size == 0) ||
647 #else
648 	    ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
649 #endif
650 	    ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
651 		return (ENOTTY);
652 
653 	if (size > 0) {
654 		if (com & IOC_VOID) {
655 			/* Integer argument. */
656 			arg = (intptr_t)uap->data;
657 			data = (void *)&arg;
658 			size = 0;
659 		} else
660 			data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
661 	} else
662 		data = (void *)&uap->data;
663 	if (com & IOC_IN) {
664 		error = copyin(uap->data, data, (u_int)size);
665 		if (error) {
666 			if (size > 0)
667 				free(data, M_IOCTLOPS);
668 			return (error);
669 		}
670 	} else if (com & IOC_OUT) {
671 		/*
672 		 * Zero the buffer so the user always
673 		 * gets back something deterministic.
674 		 */
675 		bzero(data, size);
676 	}
677 
678 	error = kern_ioctl(td, uap->fd, com, data);
679 
680 	if (error == 0 && (com & IOC_OUT))
681 		error = copyout(data, uap->data, (u_int)size);
682 
683 	if (size > 0)
684 		free(data, M_IOCTLOPS);
685 	return (error);
686 }
687 
688 int
689 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
690 {
691 	struct file *fp;
692 	struct filedesc *fdp;
693 	int error;
694 	int tmp;
695 
696 	AUDIT_ARG_FD(fd);
697 	AUDIT_ARG_CMD(com);
698 	if ((error = fget(td, fd, &fp)) != 0)
699 		return (error);
700 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
701 		fdrop(fp, td);
702 		return (EBADF);
703 	}
704 	fdp = td->td_proc->p_fd;
705 	switch (com) {
706 	case FIONCLEX:
707 		FILEDESC_XLOCK(fdp);
708 		fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
709 		FILEDESC_XUNLOCK(fdp);
710 		goto out;
711 	case FIOCLEX:
712 		FILEDESC_XLOCK(fdp);
713 		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
714 		FILEDESC_XUNLOCK(fdp);
715 		goto out;
716 	case FIONBIO:
717 		if ((tmp = *(int *)data))
718 			atomic_set_int(&fp->f_flag, FNONBLOCK);
719 		else
720 			atomic_clear_int(&fp->f_flag, FNONBLOCK);
721 		data = (void *)&tmp;
722 		break;
723 	case FIOASYNC:
724 		if ((tmp = *(int *)data))
725 			atomic_set_int(&fp->f_flag, FASYNC);
726 		else
727 			atomic_clear_int(&fp->f_flag, FASYNC);
728 		data = (void *)&tmp;
729 		break;
730 	}
731 
732 	error = fo_ioctl(fp, com, data, td->td_ucred, td);
733 out:
734 	fdrop(fp, td);
735 	return (error);
736 }
737 
738 int
739 poll_no_poll(int events)
740 {
741 	/*
742 	 * Return true for read/write.  If the user asked for something
743 	 * special, return POLLNVAL, so that clients have a way of
744 	 * determining reliably whether or not the extended
745 	 * functionality is present without hard-coding knowledge
746 	 * of specific filesystem implementations.
747 	 */
748 	if (events & ~POLLSTANDARD)
749 		return (POLLNVAL);
750 
751 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
752 }
753 
754 #ifndef _SYS_SYSPROTO_H_
755 struct select_args {
756 	int	nd;
757 	fd_set	*in, *ou, *ex;
758 	struct	timeval *tv;
759 };
760 #endif
761 int
762 select(td, uap)
763 	register struct thread *td;
764 	register struct select_args *uap;
765 {
766 	struct timeval tv, *tvp;
767 	int error;
768 
769 	if (uap->tv != NULL) {
770 		error = copyin(uap->tv, &tv, sizeof(tv));
771 		if (error)
772 			return (error);
773 		tvp = &tv;
774 	} else
775 		tvp = NULL;
776 
777 	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp));
778 }
779 
780 int
781 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
782     fd_set *fd_ex, struct timeval *tvp)
783 {
784 	struct filedesc *fdp;
785 	/*
786 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
787 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
788 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
789 	 * of 256.
790 	 */
791 	fd_mask s_selbits[howmany(2048, NFDBITS)];
792 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
793 	struct timeval atv, rtv, ttv;
794 	int error, timo;
795 	u_int nbufbytes, ncpbytes, nfdbits;
796 
797 	if (nd < 0)
798 		return (EINVAL);
799 	fdp = td->td_proc->p_fd;
800 	if (nd > fdp->fd_lastfile + 1)
801 		nd = fdp->fd_lastfile + 1;
802 
803 	/*
804 	 * Allocate just enough bits for the non-null fd_sets.  Use the
805 	 * preallocated auto buffer if possible.
806 	 */
807 	nfdbits = roundup(nd, NFDBITS);
808 	ncpbytes = nfdbits / NBBY;
809 	nbufbytes = 0;
810 	if (fd_in != NULL)
811 		nbufbytes += 2 * ncpbytes;
812 	if (fd_ou != NULL)
813 		nbufbytes += 2 * ncpbytes;
814 	if (fd_ex != NULL)
815 		nbufbytes += 2 * ncpbytes;
816 	if (nbufbytes <= sizeof s_selbits)
817 		selbits = &s_selbits[0];
818 	else
819 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
820 
821 	/*
822 	 * Assign pointers into the bit buffers and fetch the input bits.
823 	 * Put the output buffers together so that they can be bzeroed
824 	 * together.
825 	 */
826 	sbp = selbits;
827 #define	getbits(name, x) \
828 	do {								\
829 		if (name == NULL)					\
830 			ibits[x] = NULL;				\
831 		else {							\
832 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
833 			obits[x] = sbp;					\
834 			sbp += ncpbytes / sizeof *sbp;			\
835 			error = copyin(name, ibits[x], ncpbytes);	\
836 			if (error != 0)					\
837 				goto done;				\
838 		}							\
839 	} while (0)
840 	getbits(fd_in, 0);
841 	getbits(fd_ou, 1);
842 	getbits(fd_ex, 2);
843 #undef	getbits
844 	if (nbufbytes != 0)
845 		bzero(selbits, nbufbytes / 2);
846 
847 	if (tvp != NULL) {
848 		atv = *tvp;
849 		if (itimerfix(&atv)) {
850 			error = EINVAL;
851 			goto done;
852 		}
853 		getmicrouptime(&rtv);
854 		timevaladd(&atv, &rtv);
855 	} else {
856 		atv.tv_sec = 0;
857 		atv.tv_usec = 0;
858 	}
859 	timo = 0;
860 	seltdinit(td);
861 	/* Iterate until the timeout expires or descriptors become ready. */
862 	for (;;) {
863 		error = selscan(td, ibits, obits, nd);
864 		if (error || td->td_retval[0] != 0)
865 			break;
866 		if (atv.tv_sec || atv.tv_usec) {
867 			getmicrouptime(&rtv);
868 			if (timevalcmp(&rtv, &atv, >=))
869 				break;
870 			ttv = atv;
871 			timevalsub(&ttv, &rtv);
872 			timo = ttv.tv_sec > 24 * 60 * 60 ?
873 			    24 * 60 * 60 * hz : tvtohz(&ttv);
874 		}
875 		error = seltdwait(td, timo);
876 		if (error)
877 			break;
878 		error = selrescan(td, ibits, obits);
879 		if (error || td->td_retval[0] != 0)
880 			break;
881 	}
882 	seltdclear(td);
883 
884 done:
885 	/* select is not restarted after signals... */
886 	if (error == ERESTART)
887 		error = EINTR;
888 	if (error == EWOULDBLOCK)
889 		error = 0;
890 #define	putbits(name, x) \
891 	if (name && (error2 = copyout(obits[x], name, ncpbytes))) \
892 		error = error2;
893 	if (error == 0) {
894 		int error2;
895 
896 		putbits(fd_in, 0);
897 		putbits(fd_ou, 1);
898 		putbits(fd_ex, 2);
899 #undef putbits
900 	}
901 	if (selbits != &s_selbits[0])
902 		free(selbits, M_SELECT);
903 
904 	return (error);
905 }
906 /*
907  * Convert a select bit set to poll flags.
908  *
909  * The backend always returns POLLHUP/POLLERR if appropriate and we
910  * return this as a set bit in any set.
911  */
912 static int select_flags[3] = {
913     POLLRDNORM | POLLHUP | POLLERR,
914     POLLWRNORM | POLLHUP | POLLERR,
915     POLLRDBAND | POLLHUP | POLLERR
916 };
917 
918 /*
919  * Compute the fo_poll flags required for a fd given by the index and
920  * bit position in the fd_mask array.
921  */
922 static __inline int
923 selflags(fd_mask **ibits, int idx, fd_mask bit)
924 {
925 	int flags;
926 	int msk;
927 
928 	flags = 0;
929 	for (msk = 0; msk < 3; msk++) {
930 		if (ibits[msk] == NULL)
931 			continue;
932 		if ((ibits[msk][idx] & bit) == 0)
933 			continue;
934 		flags |= select_flags[msk];
935 	}
936 	return (flags);
937 }
938 
939 /*
940  * Set the appropriate output bits given a mask of fired events and the
941  * input bits originally requested.
942  */
943 static __inline int
944 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
945 {
946 	int msk;
947 	int n;
948 
949 	n = 0;
950 	for (msk = 0; msk < 3; msk++) {
951 		if ((events & select_flags[msk]) == 0)
952 			continue;
953 		if (ibits[msk] == NULL)
954 			continue;
955 		if ((ibits[msk][idx] & bit) == 0)
956 			continue;
957 		/*
958 		 * XXX Check for a duplicate set.  This can occur because a
959 		 * socket calls selrecord() twice for each poll() call
960 		 * resulting in two selfds per real fd.  selrescan() will
961 		 * call selsetbits twice as a result.
962 		 */
963 		if ((obits[msk][idx] & bit) != 0)
964 			continue;
965 		obits[msk][idx] |= bit;
966 		n++;
967 	}
968 
969 	return (n);
970 }
971 
972 /*
973  * Traverse the list of fds attached to this thread's seltd and check for
974  * completion.
975  */
976 static int
977 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
978 {
979 	struct filedesc *fdp;
980 	struct selinfo *si;
981 	struct seltd *stp;
982 	struct selfd *sfp;
983 	struct selfd *sfn;
984 	struct file *fp;
985 	fd_mask bit;
986 	int fd, ev, n, idx;
987 
988 	fdp = td->td_proc->p_fd;
989 	stp = td->td_sel;
990 	n = 0;
991 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
992 		fd = (int)(uintptr_t)sfp->sf_cookie;
993 		si = sfp->sf_si;
994 		selfdfree(stp, sfp);
995 		/* If the selinfo wasn't cleared the event didn't fire. */
996 		if (si != NULL)
997 			continue;
998 		if ((fp = fget_unlocked(fdp, fd)) == NULL)
999 			return (EBADF);
1000 		idx = fd / NFDBITS;
1001 		bit = (fd_mask)1 << (fd % NFDBITS);
1002 		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1003 		fdrop(fp, td);
1004 		if (ev != 0)
1005 			n += selsetbits(ibits, obits, idx, bit, ev);
1006 	}
1007 	stp->st_flags = 0;
1008 	td->td_retval[0] = n;
1009 	return (0);
1010 }
1011 
1012 /*
1013  * Perform the initial filedescriptor scan and register ourselves with
1014  * each selinfo.
1015  */
1016 static int
1017 selscan(td, ibits, obits, nfd)
1018 	struct thread *td;
1019 	fd_mask **ibits, **obits;
1020 	int nfd;
1021 {
1022 	struct filedesc *fdp;
1023 	struct file *fp;
1024 	fd_mask bit;
1025 	int ev, flags, end, fd;
1026 	int n, idx;
1027 
1028 	fdp = td->td_proc->p_fd;
1029 	n = 0;
1030 	for (idx = 0, fd = 0; fd < nfd; idx++) {
1031 		end = imin(fd + NFDBITS, nfd);
1032 		for (bit = 1; fd < end; bit <<= 1, fd++) {
1033 			/* Compute the list of events we're interested in. */
1034 			flags = selflags(ibits, idx, bit);
1035 			if (flags == 0)
1036 				continue;
1037 			if ((fp = fget_unlocked(fdp, fd)) == NULL)
1038 				return (EBADF);
1039 			selfdalloc(td, (void *)(uintptr_t)fd);
1040 			ev = fo_poll(fp, flags, td->td_ucred, td);
1041 			fdrop(fp, td);
1042 			if (ev != 0)
1043 				n += selsetbits(ibits, obits, idx, bit, ev);
1044 		}
1045 	}
1046 
1047 	td->td_retval[0] = n;
1048 	return (0);
1049 }
1050 
1051 #ifndef _SYS_SYSPROTO_H_
1052 struct poll_args {
1053 	struct pollfd *fds;
1054 	u_int	nfds;
1055 	int	timeout;
1056 };
1057 #endif
1058 int
1059 poll(td, uap)
1060 	struct thread *td;
1061 	struct poll_args *uap;
1062 {
1063 	struct pollfd *bits;
1064 	struct pollfd smallbits[32];
1065 	struct timeval atv, rtv, ttv;
1066 	int error = 0, timo;
1067 	u_int nfds;
1068 	size_t ni;
1069 
1070 	nfds = uap->nfds;
1071 	if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1072 		return (EINVAL);
1073 	ni = nfds * sizeof(struct pollfd);
1074 	if (ni > sizeof(smallbits))
1075 		bits = malloc(ni, M_TEMP, M_WAITOK);
1076 	else
1077 		bits = smallbits;
1078 	error = copyin(uap->fds, bits, ni);
1079 	if (error)
1080 		goto done;
1081 	if (uap->timeout != INFTIM) {
1082 		atv.tv_sec = uap->timeout / 1000;
1083 		atv.tv_usec = (uap->timeout % 1000) * 1000;
1084 		if (itimerfix(&atv)) {
1085 			error = EINVAL;
1086 			goto done;
1087 		}
1088 		getmicrouptime(&rtv);
1089 		timevaladd(&atv, &rtv);
1090 	} else {
1091 		atv.tv_sec = 0;
1092 		atv.tv_usec = 0;
1093 	}
1094 	timo = 0;
1095 	seltdinit(td);
1096 	/* Iterate until the timeout expires or descriptors become ready. */
1097 	for (;;) {
1098 		error = pollscan(td, bits, nfds);
1099 		if (error || td->td_retval[0] != 0)
1100 			break;
1101 		if (atv.tv_sec || atv.tv_usec) {
1102 			getmicrouptime(&rtv);
1103 			if (timevalcmp(&rtv, &atv, >=))
1104 				break;
1105 			ttv = atv;
1106 			timevalsub(&ttv, &rtv);
1107 			timo = ttv.tv_sec > 24 * 60 * 60 ?
1108 			    24 * 60 * 60 * hz : tvtohz(&ttv);
1109 		}
1110 		error = seltdwait(td, timo);
1111 		if (error)
1112 			break;
1113 		error = pollrescan(td);
1114 		if (error || td->td_retval[0] != 0)
1115 			break;
1116 	}
1117 	seltdclear(td);
1118 
1119 done:
1120 	/* poll is not restarted after signals... */
1121 	if (error == ERESTART)
1122 		error = EINTR;
1123 	if (error == EWOULDBLOCK)
1124 		error = 0;
1125 	if (error == 0) {
1126 		error = pollout(bits, uap->fds, nfds);
1127 		if (error)
1128 			goto out;
1129 	}
1130 out:
1131 	if (ni > sizeof(smallbits))
1132 		free(bits, M_TEMP);
1133 	return (error);
1134 }
1135 
1136 static int
1137 pollrescan(struct thread *td)
1138 {
1139 	struct seltd *stp;
1140 	struct selfd *sfp;
1141 	struct selfd *sfn;
1142 	struct selinfo *si;
1143 	struct filedesc *fdp;
1144 	struct file *fp;
1145 	struct pollfd *fd;
1146 	int n;
1147 
1148 	n = 0;
1149 	fdp = td->td_proc->p_fd;
1150 	stp = td->td_sel;
1151 	FILEDESC_SLOCK(fdp);
1152 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1153 		fd = (struct pollfd *)sfp->sf_cookie;
1154 		si = sfp->sf_si;
1155 		selfdfree(stp, sfp);
1156 		/* If the selinfo wasn't cleared the event didn't fire. */
1157 		if (si != NULL)
1158 			continue;
1159 		fp = fdp->fd_ofiles[fd->fd];
1160 		if (fp == NULL) {
1161 			fd->revents = POLLNVAL;
1162 			n++;
1163 			continue;
1164 		}
1165 		/*
1166 		 * Note: backend also returns POLLHUP and
1167 		 * POLLERR if appropriate.
1168 		 */
1169 		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1170 		if (fd->revents != 0)
1171 			n++;
1172 	}
1173 	FILEDESC_SUNLOCK(fdp);
1174 	stp->st_flags = 0;
1175 	td->td_retval[0] = n;
1176 	return (0);
1177 }
1178 
1179 
1180 static int
1181 pollout(fds, ufds, nfd)
1182 	struct pollfd *fds;
1183 	struct pollfd *ufds;
1184 	u_int nfd;
1185 {
1186 	int error = 0;
1187 	u_int i = 0;
1188 
1189 	for (i = 0; i < nfd; i++) {
1190 		error = copyout(&fds->revents, &ufds->revents,
1191 		    sizeof(ufds->revents));
1192 		if (error)
1193 			return (error);
1194 		fds++;
1195 		ufds++;
1196 	}
1197 	return (0);
1198 }
1199 
1200 static int
1201 pollscan(td, fds, nfd)
1202 	struct thread *td;
1203 	struct pollfd *fds;
1204 	u_int nfd;
1205 {
1206 	struct filedesc *fdp = td->td_proc->p_fd;
1207 	int i;
1208 	struct file *fp;
1209 	int n = 0;
1210 
1211 	FILEDESC_SLOCK(fdp);
1212 	for (i = 0; i < nfd; i++, fds++) {
1213 		if (fds->fd >= fdp->fd_nfiles) {
1214 			fds->revents = POLLNVAL;
1215 			n++;
1216 		} else if (fds->fd < 0) {
1217 			fds->revents = 0;
1218 		} else {
1219 			fp = fdp->fd_ofiles[fds->fd];
1220 			if (fp == NULL) {
1221 				fds->revents = POLLNVAL;
1222 				n++;
1223 			} else {
1224 				/*
1225 				 * Note: backend also returns POLLHUP and
1226 				 * POLLERR if appropriate.
1227 				 */
1228 				selfdalloc(td, fds);
1229 				fds->revents = fo_poll(fp, fds->events,
1230 				    td->td_ucred, td);
1231 				if (fds->revents != 0)
1232 					n++;
1233 			}
1234 		}
1235 	}
1236 	FILEDESC_SUNLOCK(fdp);
1237 	td->td_retval[0] = n;
1238 	return (0);
1239 }
1240 
1241 /*
1242  * OpenBSD poll system call.
1243  *
1244  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
1245  */
1246 #ifndef _SYS_SYSPROTO_H_
1247 struct openbsd_poll_args {
1248 	struct pollfd *fds;
1249 	u_int	nfds;
1250 	int	timeout;
1251 };
1252 #endif
1253 int
1254 openbsd_poll(td, uap)
1255 	register struct thread *td;
1256 	register struct openbsd_poll_args *uap;
1257 {
1258 	return (poll(td, (struct poll_args *)uap));
1259 }
1260 
1261 /*
1262  * XXX This was created specifically to support netncp and netsmb.  This
1263  * allows the caller to specify a socket to wait for events on.  It returns
1264  * 0 if any events matched and an error otherwise.  There is no way to
1265  * determine which events fired.
1266  */
1267 int
1268 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1269 {
1270 	struct timeval atv, rtv, ttv;
1271 	int error, timo;
1272 
1273 	if (tvp != NULL) {
1274 		atv = *tvp;
1275 		if (itimerfix(&atv))
1276 			return (EINVAL);
1277 		getmicrouptime(&rtv);
1278 		timevaladd(&atv, &rtv);
1279 	} else {
1280 		atv.tv_sec = 0;
1281 		atv.tv_usec = 0;
1282 	}
1283 
1284 	timo = 0;
1285 	seltdinit(td);
1286 	/*
1287 	 * Iterate until the timeout expires or the socket becomes ready.
1288 	 */
1289 	for (;;) {
1290 		selfdalloc(td, NULL);
1291 		error = sopoll(so, events, NULL, td);
1292 		/* error here is actually the ready events. */
1293 		if (error)
1294 			return (0);
1295 		if (atv.tv_sec || atv.tv_usec) {
1296 			getmicrouptime(&rtv);
1297 			if (timevalcmp(&rtv, &atv, >=)) {
1298 				seltdclear(td);
1299 				return (EWOULDBLOCK);
1300 			}
1301 			ttv = atv;
1302 			timevalsub(&ttv, &rtv);
1303 			timo = ttv.tv_sec > 24 * 60 * 60 ?
1304 			    24 * 60 * 60 * hz : tvtohz(&ttv);
1305 		}
1306 		error = seltdwait(td, timo);
1307 		seltdclear(td);
1308 		if (error)
1309 			break;
1310 	}
1311 	/* XXX Duplicates ncp/smb behavior. */
1312 	if (error == ERESTART)
1313 		error = 0;
1314 	return (error);
1315 }
1316 
1317 /*
1318  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1319  * have two select sets, one for read and another for write.
1320  */
1321 static void
1322 selfdalloc(struct thread *td, void *cookie)
1323 {
1324 	struct seltd *stp;
1325 
1326 	stp = td->td_sel;
1327 	if (stp->st_free1 == NULL)
1328 		stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1329 	stp->st_free1->sf_td = stp;
1330 	stp->st_free1->sf_cookie = cookie;
1331 	if (stp->st_free2 == NULL)
1332 		stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1333 	stp->st_free2->sf_td = stp;
1334 	stp->st_free2->sf_cookie = cookie;
1335 }
1336 
1337 static void
1338 selfdfree(struct seltd *stp, struct selfd *sfp)
1339 {
1340 	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1341 	mtx_lock(sfp->sf_mtx);
1342 	if (sfp->sf_si)
1343 		TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1344 	mtx_unlock(sfp->sf_mtx);
1345 	uma_zfree(selfd_zone, sfp);
1346 }
1347 
1348 /*
1349  * Record a select request.
1350  */
1351 void
1352 selrecord(selector, sip)
1353 	struct thread *selector;
1354 	struct selinfo *sip;
1355 {
1356 	struct selfd *sfp;
1357 	struct seltd *stp;
1358 	struct mtx *mtxp;
1359 
1360 	stp = selector->td_sel;
1361 	/*
1362 	 * Don't record when doing a rescan.
1363 	 */
1364 	if (stp->st_flags & SELTD_RESCAN)
1365 		return;
1366 	/*
1367 	 * Grab one of the preallocated descriptors.
1368 	 */
1369 	sfp = NULL;
1370 	if ((sfp = stp->st_free1) != NULL)
1371 		stp->st_free1 = NULL;
1372 	else if ((sfp = stp->st_free2) != NULL)
1373 		stp->st_free2 = NULL;
1374 	else
1375 		panic("selrecord: No free selfd on selq");
1376 	mtxp = sip->si_mtx;
1377 	if (mtxp == NULL)
1378 		mtxp = mtx_pool_find(mtxpool_select, sip);
1379 	/*
1380 	 * Initialize the sfp and queue it in the thread.
1381 	 */
1382 	sfp->sf_si = sip;
1383 	sfp->sf_mtx = mtxp;
1384 	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1385 	/*
1386 	 * Now that we've locked the sip, check for initialization.
1387 	 */
1388 	mtx_lock(mtxp);
1389 	if (sip->si_mtx == NULL) {
1390 		sip->si_mtx = mtxp;
1391 		TAILQ_INIT(&sip->si_tdlist);
1392 	}
1393 	/*
1394 	 * Add this thread to the list of selfds listening on this selinfo.
1395 	 */
1396 	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1397 	mtx_unlock(sip->si_mtx);
1398 }
1399 
1400 /* Wake up a selecting thread. */
1401 void
1402 selwakeup(sip)
1403 	struct selinfo *sip;
1404 {
1405 	doselwakeup(sip, -1);
1406 }
1407 
1408 /* Wake up a selecting thread, and set its priority. */
1409 void
1410 selwakeuppri(sip, pri)
1411 	struct selinfo *sip;
1412 	int pri;
1413 {
1414 	doselwakeup(sip, pri);
1415 }
1416 
1417 /*
1418  * Do a wakeup when a selectable event occurs.
1419  */
1420 static void
1421 doselwakeup(sip, pri)
1422 	struct selinfo *sip;
1423 	int pri;
1424 {
1425 	struct selfd *sfp;
1426 	struct selfd *sfn;
1427 	struct seltd *stp;
1428 
1429 	/* If it's not initialized there can't be any waiters. */
1430 	if (sip->si_mtx == NULL)
1431 		return;
1432 	/*
1433 	 * Locking the selinfo locks all selfds associated with it.
1434 	 */
1435 	mtx_lock(sip->si_mtx);
1436 	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1437 		/*
1438 		 * Once we remove this sfp from the list and clear the
1439 		 * sf_si seltdclear will know to ignore this si.
1440 		 */
1441 		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1442 		sfp->sf_si = NULL;
1443 		stp = sfp->sf_td;
1444 		mtx_lock(&stp->st_mtx);
1445 		stp->st_flags |= SELTD_PENDING;
1446 		cv_broadcastpri(&stp->st_wait, pri);
1447 		mtx_unlock(&stp->st_mtx);
1448 	}
1449 	mtx_unlock(sip->si_mtx);
1450 }
1451 
1452 static void
1453 seltdinit(struct thread *td)
1454 {
1455 	struct seltd *stp;
1456 
1457 	if ((stp = td->td_sel) != NULL)
1458 		goto out;
1459 	td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1460 	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1461 	cv_init(&stp->st_wait, "select");
1462 out:
1463 	stp->st_flags = 0;
1464 	STAILQ_INIT(&stp->st_selq);
1465 }
1466 
1467 static int
1468 seltdwait(struct thread *td, int timo)
1469 {
1470 	struct seltd *stp;
1471 	int error;
1472 
1473 	stp = td->td_sel;
1474 	/*
1475 	 * An event of interest may occur while we do not hold the seltd
1476 	 * locked so check the pending flag before we sleep.
1477 	 */
1478 	mtx_lock(&stp->st_mtx);
1479 	/*
1480 	 * Any further calls to selrecord will be a rescan.
1481 	 */
1482 	stp->st_flags |= SELTD_RESCAN;
1483 	if (stp->st_flags & SELTD_PENDING) {
1484 		mtx_unlock(&stp->st_mtx);
1485 		return (0);
1486 	}
1487 	if (timo > 0)
1488 		error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1489 	else
1490 		error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1491 	mtx_unlock(&stp->st_mtx);
1492 
1493 	return (error);
1494 }
1495 
1496 void
1497 seltdfini(struct thread *td)
1498 {
1499 	struct seltd *stp;
1500 
1501 	stp = td->td_sel;
1502 	if (stp == NULL)
1503 		return;
1504 	if (stp->st_free1)
1505 		uma_zfree(selfd_zone, stp->st_free1);
1506 	if (stp->st_free2)
1507 		uma_zfree(selfd_zone, stp->st_free2);
1508 	td->td_sel = NULL;
1509 	free(stp, M_SELECT);
1510 }
1511 
1512 /*
1513  * Remove the references to the thread from all of the objects we were
1514  * polling.
1515  */
1516 static void
1517 seltdclear(struct thread *td)
1518 {
1519 	struct seltd *stp;
1520 	struct selfd *sfp;
1521 	struct selfd *sfn;
1522 
1523 	stp = td->td_sel;
1524 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1525 		selfdfree(stp, sfp);
1526 	stp->st_flags = 0;
1527 }
1528 
1529 static void selectinit(void *);
1530 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1531 static void
1532 selectinit(void *dummy __unused)
1533 {
1534 
1535 	selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1536 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1537 	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1538 }
1539