xref: /freebsd/sys/kern/sys_generic.c (revision aa64588d28258aef88cc33b8043112e8856948d0)
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 int
755 pselect(struct thread *td, struct pselect_args *uap)
756 {
757 	struct timespec ts;
758 	struct timeval tv, *tvp;
759 	sigset_t set, *uset;
760 	int error;
761 
762 	if (uap->ts != NULL) {
763 		error = copyin(uap->ts, &ts, sizeof(ts));
764 		if (error != 0)
765 		    return (error);
766 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
767 		tvp = &tv;
768 	} else
769 		tvp = NULL;
770 	if (uap->sm != NULL) {
771 		error = copyin(uap->sm, &set, sizeof(set));
772 		if (error != 0)
773 			return (error);
774 		uset = &set;
775 	} else
776 		uset = NULL;
777 	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
778 	    uset, NFDBITS));
779 }
780 
781 int
782 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
783     struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
784 {
785 	int error;
786 
787 	if (uset != NULL) {
788 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
789 		    &td->td_oldsigmask, 0);
790 		if (error != 0)
791 			return (error);
792 		td->td_pflags |= TDP_OLDMASK;
793 		/*
794 		 * Make sure that ast() is called on return to
795 		 * usermode and TDP_OLDMASK is cleared, restoring old
796 		 * sigmask.
797 		 */
798 		thread_lock(td);
799 		td->td_flags |= TDF_ASTPENDING;
800 		thread_unlock(td);
801 	}
802 	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
803 	return (error);
804 }
805 
806 #ifndef _SYS_SYSPROTO_H_
807 struct select_args {
808 	int	nd;
809 	fd_set	*in, *ou, *ex;
810 	struct	timeval *tv;
811 };
812 #endif
813 int
814 select(struct thread *td, struct select_args *uap)
815 {
816 	struct timeval tv, *tvp;
817 	int error;
818 
819 	if (uap->tv != NULL) {
820 		error = copyin(uap->tv, &tv, sizeof(tv));
821 		if (error)
822 			return (error);
823 		tvp = &tv;
824 	} else
825 		tvp = NULL;
826 
827 	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
828 	    NFDBITS));
829 }
830 
831 int
832 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
833     fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
834 {
835 	struct filedesc *fdp;
836 	/*
837 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
838 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
839 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
840 	 * of 256.
841 	 */
842 	fd_mask s_selbits[howmany(2048, NFDBITS)];
843 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
844 	struct timeval atv, rtv, ttv;
845 	int error, timo;
846 	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
847 
848 	if (nd < 0)
849 		return (EINVAL);
850 	fdp = td->td_proc->p_fd;
851 	if (nd > fdp->fd_lastfile + 1)
852 		nd = fdp->fd_lastfile + 1;
853 
854 	/*
855 	 * Allocate just enough bits for the non-null fd_sets.  Use the
856 	 * preallocated auto buffer if possible.
857 	 */
858 	nfdbits = roundup(nd, NFDBITS);
859 	ncpbytes = nfdbits / NBBY;
860 	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
861 	nbufbytes = 0;
862 	if (fd_in != NULL)
863 		nbufbytes += 2 * ncpbytes;
864 	if (fd_ou != NULL)
865 		nbufbytes += 2 * ncpbytes;
866 	if (fd_ex != NULL)
867 		nbufbytes += 2 * ncpbytes;
868 	if (nbufbytes <= sizeof s_selbits)
869 		selbits = &s_selbits[0];
870 	else
871 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
872 
873 	/*
874 	 * Assign pointers into the bit buffers and fetch the input bits.
875 	 * Put the output buffers together so that they can be bzeroed
876 	 * together.
877 	 */
878 	sbp = selbits;
879 #define	getbits(name, x) \
880 	do {								\
881 		if (name == NULL) {					\
882 			ibits[x] = NULL;				\
883 			obits[x] = NULL;				\
884 		} else {						\
885 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
886 			obits[x] = sbp;					\
887 			sbp += ncpbytes / sizeof *sbp;			\
888 			error = copyin(name, ibits[x], ncpubytes);	\
889 			if (error != 0)					\
890 				goto done;				\
891 			bzero((char *)ibits[x] + ncpubytes,		\
892 			    ncpbytes - ncpubytes);			\
893 		}							\
894 	} while (0)
895 	getbits(fd_in, 0);
896 	getbits(fd_ou, 1);
897 	getbits(fd_ex, 2);
898 #undef	getbits
899 
900 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
901 	/*
902 	 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
903 	 * we are running under 32-bit emulation. This should be more
904 	 * generic.
905 	 */
906 #define swizzle_fdset(bits)						\
907 	if (abi_nfdbits != NFDBITS && bits != NULL) {			\
908 		int i;							\
909 		for (i = 0; i < ncpbytes / sizeof *sbp; i++)		\
910 			bits[i] = (bits[i] >> 32) | (bits[i] << 32);	\
911 	}
912 #else
913 #define swizzle_fdset(bits)
914 #endif
915 
916 	/* Make sure the bit order makes it through an ABI transition */
917 	swizzle_fdset(ibits[0]);
918 	swizzle_fdset(ibits[1]);
919 	swizzle_fdset(ibits[2]);
920 
921 	if (nbufbytes != 0)
922 		bzero(selbits, nbufbytes / 2);
923 
924 	if (tvp != NULL) {
925 		atv = *tvp;
926 		if (itimerfix(&atv)) {
927 			error = EINVAL;
928 			goto done;
929 		}
930 		getmicrouptime(&rtv);
931 		timevaladd(&atv, &rtv);
932 	} else {
933 		atv.tv_sec = 0;
934 		atv.tv_usec = 0;
935 	}
936 	timo = 0;
937 	seltdinit(td);
938 	/* Iterate until the timeout expires or descriptors become ready. */
939 	for (;;) {
940 		error = selscan(td, ibits, obits, nd);
941 		if (error || td->td_retval[0] != 0)
942 			break;
943 		if (atv.tv_sec || atv.tv_usec) {
944 			getmicrouptime(&rtv);
945 			if (timevalcmp(&rtv, &atv, >=))
946 				break;
947 			ttv = atv;
948 			timevalsub(&ttv, &rtv);
949 			timo = ttv.tv_sec > 24 * 60 * 60 ?
950 			    24 * 60 * 60 * hz : tvtohz(&ttv);
951 		}
952 		error = seltdwait(td, timo);
953 		if (error)
954 			break;
955 		error = selrescan(td, ibits, obits);
956 		if (error || td->td_retval[0] != 0)
957 			break;
958 	}
959 	seltdclear(td);
960 
961 done:
962 	/* select is not restarted after signals... */
963 	if (error == ERESTART)
964 		error = EINTR;
965 	if (error == EWOULDBLOCK)
966 		error = 0;
967 
968 	/* swizzle bit order back, if necessary */
969 	swizzle_fdset(obits[0]);
970 	swizzle_fdset(obits[1]);
971 	swizzle_fdset(obits[2]);
972 #undef swizzle_fdset
973 
974 #define	putbits(name, x) \
975 	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
976 		error = error2;
977 	if (error == 0) {
978 		int error2;
979 
980 		putbits(fd_in, 0);
981 		putbits(fd_ou, 1);
982 		putbits(fd_ex, 2);
983 #undef putbits
984 	}
985 	if (selbits != &s_selbits[0])
986 		free(selbits, M_SELECT);
987 
988 	return (error);
989 }
990 /*
991  * Convert a select bit set to poll flags.
992  *
993  * The backend always returns POLLHUP/POLLERR if appropriate and we
994  * return this as a set bit in any set.
995  */
996 static int select_flags[3] = {
997     POLLRDNORM | POLLHUP | POLLERR,
998     POLLWRNORM | POLLHUP | POLLERR,
999     POLLRDBAND | POLLERR
1000 };
1001 
1002 /*
1003  * Compute the fo_poll flags required for a fd given by the index and
1004  * bit position in the fd_mask array.
1005  */
1006 static __inline int
1007 selflags(fd_mask **ibits, int idx, fd_mask bit)
1008 {
1009 	int flags;
1010 	int msk;
1011 
1012 	flags = 0;
1013 	for (msk = 0; msk < 3; msk++) {
1014 		if (ibits[msk] == NULL)
1015 			continue;
1016 		if ((ibits[msk][idx] & bit) == 0)
1017 			continue;
1018 		flags |= select_flags[msk];
1019 	}
1020 	return (flags);
1021 }
1022 
1023 /*
1024  * Set the appropriate output bits given a mask of fired events and the
1025  * input bits originally requested.
1026  */
1027 static __inline int
1028 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1029 {
1030 	int msk;
1031 	int n;
1032 
1033 	n = 0;
1034 	for (msk = 0; msk < 3; msk++) {
1035 		if ((events & select_flags[msk]) == 0)
1036 			continue;
1037 		if (ibits[msk] == NULL)
1038 			continue;
1039 		if ((ibits[msk][idx] & bit) == 0)
1040 			continue;
1041 		/*
1042 		 * XXX Check for a duplicate set.  This can occur because a
1043 		 * socket calls selrecord() twice for each poll() call
1044 		 * resulting in two selfds per real fd.  selrescan() will
1045 		 * call selsetbits twice as a result.
1046 		 */
1047 		if ((obits[msk][idx] & bit) != 0)
1048 			continue;
1049 		obits[msk][idx] |= bit;
1050 		n++;
1051 	}
1052 
1053 	return (n);
1054 }
1055 
1056 /*
1057  * Traverse the list of fds attached to this thread's seltd and check for
1058  * completion.
1059  */
1060 static int
1061 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1062 {
1063 	struct filedesc *fdp;
1064 	struct selinfo *si;
1065 	struct seltd *stp;
1066 	struct selfd *sfp;
1067 	struct selfd *sfn;
1068 	struct file *fp;
1069 	fd_mask bit;
1070 	int fd, ev, n, idx;
1071 
1072 	fdp = td->td_proc->p_fd;
1073 	stp = td->td_sel;
1074 	n = 0;
1075 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1076 		fd = (int)(uintptr_t)sfp->sf_cookie;
1077 		si = sfp->sf_si;
1078 		selfdfree(stp, sfp);
1079 		/* If the selinfo wasn't cleared the event didn't fire. */
1080 		if (si != NULL)
1081 			continue;
1082 		if ((fp = fget_unlocked(fdp, fd)) == NULL)
1083 			return (EBADF);
1084 		idx = fd / NFDBITS;
1085 		bit = (fd_mask)1 << (fd % NFDBITS);
1086 		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1087 		fdrop(fp, td);
1088 		if (ev != 0)
1089 			n += selsetbits(ibits, obits, idx, bit, ev);
1090 	}
1091 	stp->st_flags = 0;
1092 	td->td_retval[0] = n;
1093 	return (0);
1094 }
1095 
1096 /*
1097  * Perform the initial filedescriptor scan and register ourselves with
1098  * each selinfo.
1099  */
1100 static int
1101 selscan(td, ibits, obits, nfd)
1102 	struct thread *td;
1103 	fd_mask **ibits, **obits;
1104 	int nfd;
1105 {
1106 	struct filedesc *fdp;
1107 	struct file *fp;
1108 	fd_mask bit;
1109 	int ev, flags, end, fd;
1110 	int n, idx;
1111 
1112 	fdp = td->td_proc->p_fd;
1113 	n = 0;
1114 	for (idx = 0, fd = 0; fd < nfd; idx++) {
1115 		end = imin(fd + NFDBITS, nfd);
1116 		for (bit = 1; fd < end; bit <<= 1, fd++) {
1117 			/* Compute the list of events we're interested in. */
1118 			flags = selflags(ibits, idx, bit);
1119 			if (flags == 0)
1120 				continue;
1121 			if ((fp = fget_unlocked(fdp, fd)) == NULL)
1122 				return (EBADF);
1123 			selfdalloc(td, (void *)(uintptr_t)fd);
1124 			ev = fo_poll(fp, flags, td->td_ucred, td);
1125 			fdrop(fp, td);
1126 			if (ev != 0)
1127 				n += selsetbits(ibits, obits, idx, bit, ev);
1128 		}
1129 	}
1130 
1131 	td->td_retval[0] = n;
1132 	return (0);
1133 }
1134 
1135 #ifndef _SYS_SYSPROTO_H_
1136 struct poll_args {
1137 	struct pollfd *fds;
1138 	u_int	nfds;
1139 	int	timeout;
1140 };
1141 #endif
1142 int
1143 poll(td, uap)
1144 	struct thread *td;
1145 	struct poll_args *uap;
1146 {
1147 	struct pollfd *bits;
1148 	struct pollfd smallbits[32];
1149 	struct timeval atv, rtv, ttv;
1150 	int error = 0, timo;
1151 	u_int nfds;
1152 	size_t ni;
1153 
1154 	nfds = uap->nfds;
1155 	if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1156 		return (EINVAL);
1157 	ni = nfds * sizeof(struct pollfd);
1158 	if (ni > sizeof(smallbits))
1159 		bits = malloc(ni, M_TEMP, M_WAITOK);
1160 	else
1161 		bits = smallbits;
1162 	error = copyin(uap->fds, bits, ni);
1163 	if (error)
1164 		goto done;
1165 	if (uap->timeout != INFTIM) {
1166 		atv.tv_sec = uap->timeout / 1000;
1167 		atv.tv_usec = (uap->timeout % 1000) * 1000;
1168 		if (itimerfix(&atv)) {
1169 			error = EINVAL;
1170 			goto done;
1171 		}
1172 		getmicrouptime(&rtv);
1173 		timevaladd(&atv, &rtv);
1174 	} else {
1175 		atv.tv_sec = 0;
1176 		atv.tv_usec = 0;
1177 	}
1178 	timo = 0;
1179 	seltdinit(td);
1180 	/* Iterate until the timeout expires or descriptors become ready. */
1181 	for (;;) {
1182 		error = pollscan(td, bits, nfds);
1183 		if (error || td->td_retval[0] != 0)
1184 			break;
1185 		if (atv.tv_sec || atv.tv_usec) {
1186 			getmicrouptime(&rtv);
1187 			if (timevalcmp(&rtv, &atv, >=))
1188 				break;
1189 			ttv = atv;
1190 			timevalsub(&ttv, &rtv);
1191 			timo = ttv.tv_sec > 24 * 60 * 60 ?
1192 			    24 * 60 * 60 * hz : tvtohz(&ttv);
1193 		}
1194 		error = seltdwait(td, timo);
1195 		if (error)
1196 			break;
1197 		error = pollrescan(td);
1198 		if (error || td->td_retval[0] != 0)
1199 			break;
1200 	}
1201 	seltdclear(td);
1202 
1203 done:
1204 	/* poll is not restarted after signals... */
1205 	if (error == ERESTART)
1206 		error = EINTR;
1207 	if (error == EWOULDBLOCK)
1208 		error = 0;
1209 	if (error == 0) {
1210 		error = pollout(bits, uap->fds, nfds);
1211 		if (error)
1212 			goto out;
1213 	}
1214 out:
1215 	if (ni > sizeof(smallbits))
1216 		free(bits, M_TEMP);
1217 	return (error);
1218 }
1219 
1220 static int
1221 pollrescan(struct thread *td)
1222 {
1223 	struct seltd *stp;
1224 	struct selfd *sfp;
1225 	struct selfd *sfn;
1226 	struct selinfo *si;
1227 	struct filedesc *fdp;
1228 	struct file *fp;
1229 	struct pollfd *fd;
1230 	int n;
1231 
1232 	n = 0;
1233 	fdp = td->td_proc->p_fd;
1234 	stp = td->td_sel;
1235 	FILEDESC_SLOCK(fdp);
1236 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1237 		fd = (struct pollfd *)sfp->sf_cookie;
1238 		si = sfp->sf_si;
1239 		selfdfree(stp, sfp);
1240 		/* If the selinfo wasn't cleared the event didn't fire. */
1241 		if (si != NULL)
1242 			continue;
1243 		fp = fdp->fd_ofiles[fd->fd];
1244 		if (fp == NULL) {
1245 			fd->revents = POLLNVAL;
1246 			n++;
1247 			continue;
1248 		}
1249 		/*
1250 		 * Note: backend also returns POLLHUP and
1251 		 * POLLERR if appropriate.
1252 		 */
1253 		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1254 		if (fd->revents != 0)
1255 			n++;
1256 	}
1257 	FILEDESC_SUNLOCK(fdp);
1258 	stp->st_flags = 0;
1259 	td->td_retval[0] = n;
1260 	return (0);
1261 }
1262 
1263 
1264 static int
1265 pollout(fds, ufds, nfd)
1266 	struct pollfd *fds;
1267 	struct pollfd *ufds;
1268 	u_int nfd;
1269 {
1270 	int error = 0;
1271 	u_int i = 0;
1272 
1273 	for (i = 0; i < nfd; i++) {
1274 		error = copyout(&fds->revents, &ufds->revents,
1275 		    sizeof(ufds->revents));
1276 		if (error)
1277 			return (error);
1278 		fds++;
1279 		ufds++;
1280 	}
1281 	return (0);
1282 }
1283 
1284 static int
1285 pollscan(td, fds, nfd)
1286 	struct thread *td;
1287 	struct pollfd *fds;
1288 	u_int nfd;
1289 {
1290 	struct filedesc *fdp = td->td_proc->p_fd;
1291 	int i;
1292 	struct file *fp;
1293 	int n = 0;
1294 
1295 	FILEDESC_SLOCK(fdp);
1296 	for (i = 0; i < nfd; i++, fds++) {
1297 		if (fds->fd >= fdp->fd_nfiles) {
1298 			fds->revents = POLLNVAL;
1299 			n++;
1300 		} else if (fds->fd < 0) {
1301 			fds->revents = 0;
1302 		} else {
1303 			fp = fdp->fd_ofiles[fds->fd];
1304 			if (fp == NULL) {
1305 				fds->revents = POLLNVAL;
1306 				n++;
1307 			} else {
1308 				/*
1309 				 * Note: backend also returns POLLHUP and
1310 				 * POLLERR if appropriate.
1311 				 */
1312 				selfdalloc(td, fds);
1313 				fds->revents = fo_poll(fp, fds->events,
1314 				    td->td_ucred, td);
1315 				/*
1316 				 * POSIX requires POLLOUT to be never
1317 				 * set simultaneously with POLLHUP.
1318 				 */
1319 				if ((fds->revents & POLLHUP) != 0)
1320 					fds->revents &= ~POLLOUT;
1321 
1322 				if (fds->revents != 0)
1323 					n++;
1324 			}
1325 		}
1326 	}
1327 	FILEDESC_SUNLOCK(fdp);
1328 	td->td_retval[0] = n;
1329 	return (0);
1330 }
1331 
1332 /*
1333  * OpenBSD poll system call.
1334  *
1335  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
1336  */
1337 #ifndef _SYS_SYSPROTO_H_
1338 struct openbsd_poll_args {
1339 	struct pollfd *fds;
1340 	u_int	nfds;
1341 	int	timeout;
1342 };
1343 #endif
1344 int
1345 openbsd_poll(td, uap)
1346 	register struct thread *td;
1347 	register struct openbsd_poll_args *uap;
1348 {
1349 	return (poll(td, (struct poll_args *)uap));
1350 }
1351 
1352 /*
1353  * XXX This was created specifically to support netncp and netsmb.  This
1354  * allows the caller to specify a socket to wait for events on.  It returns
1355  * 0 if any events matched and an error otherwise.  There is no way to
1356  * determine which events fired.
1357  */
1358 int
1359 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1360 {
1361 	struct timeval atv, rtv, ttv;
1362 	int error, timo;
1363 
1364 	if (tvp != NULL) {
1365 		atv = *tvp;
1366 		if (itimerfix(&atv))
1367 			return (EINVAL);
1368 		getmicrouptime(&rtv);
1369 		timevaladd(&atv, &rtv);
1370 	} else {
1371 		atv.tv_sec = 0;
1372 		atv.tv_usec = 0;
1373 	}
1374 
1375 	timo = 0;
1376 	seltdinit(td);
1377 	/*
1378 	 * Iterate until the timeout expires or the socket becomes ready.
1379 	 */
1380 	for (;;) {
1381 		selfdalloc(td, NULL);
1382 		error = sopoll(so, events, NULL, td);
1383 		/* error here is actually the ready events. */
1384 		if (error)
1385 			return (0);
1386 		if (atv.tv_sec || atv.tv_usec) {
1387 			getmicrouptime(&rtv);
1388 			if (timevalcmp(&rtv, &atv, >=)) {
1389 				seltdclear(td);
1390 				return (EWOULDBLOCK);
1391 			}
1392 			ttv = atv;
1393 			timevalsub(&ttv, &rtv);
1394 			timo = ttv.tv_sec > 24 * 60 * 60 ?
1395 			    24 * 60 * 60 * hz : tvtohz(&ttv);
1396 		}
1397 		error = seltdwait(td, timo);
1398 		seltdclear(td);
1399 		if (error)
1400 			break;
1401 	}
1402 	/* XXX Duplicates ncp/smb behavior. */
1403 	if (error == ERESTART)
1404 		error = 0;
1405 	return (error);
1406 }
1407 
1408 /*
1409  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1410  * have two select sets, one for read and another for write.
1411  */
1412 static void
1413 selfdalloc(struct thread *td, void *cookie)
1414 {
1415 	struct seltd *stp;
1416 
1417 	stp = td->td_sel;
1418 	if (stp->st_free1 == NULL)
1419 		stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1420 	stp->st_free1->sf_td = stp;
1421 	stp->st_free1->sf_cookie = cookie;
1422 	if (stp->st_free2 == NULL)
1423 		stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1424 	stp->st_free2->sf_td = stp;
1425 	stp->st_free2->sf_cookie = cookie;
1426 }
1427 
1428 static void
1429 selfdfree(struct seltd *stp, struct selfd *sfp)
1430 {
1431 	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1432 	mtx_lock(sfp->sf_mtx);
1433 	if (sfp->sf_si)
1434 		TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1435 	mtx_unlock(sfp->sf_mtx);
1436 	uma_zfree(selfd_zone, sfp);
1437 }
1438 
1439 /*
1440  * Record a select request.
1441  */
1442 void
1443 selrecord(selector, sip)
1444 	struct thread *selector;
1445 	struct selinfo *sip;
1446 {
1447 	struct selfd *sfp;
1448 	struct seltd *stp;
1449 	struct mtx *mtxp;
1450 
1451 	stp = selector->td_sel;
1452 	/*
1453 	 * Don't record when doing a rescan.
1454 	 */
1455 	if (stp->st_flags & SELTD_RESCAN)
1456 		return;
1457 	/*
1458 	 * Grab one of the preallocated descriptors.
1459 	 */
1460 	sfp = NULL;
1461 	if ((sfp = stp->st_free1) != NULL)
1462 		stp->st_free1 = NULL;
1463 	else if ((sfp = stp->st_free2) != NULL)
1464 		stp->st_free2 = NULL;
1465 	else
1466 		panic("selrecord: No free selfd on selq");
1467 	mtxp = sip->si_mtx;
1468 	if (mtxp == NULL)
1469 		mtxp = mtx_pool_find(mtxpool_select, sip);
1470 	/*
1471 	 * Initialize the sfp and queue it in the thread.
1472 	 */
1473 	sfp->sf_si = sip;
1474 	sfp->sf_mtx = mtxp;
1475 	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1476 	/*
1477 	 * Now that we've locked the sip, check for initialization.
1478 	 */
1479 	mtx_lock(mtxp);
1480 	if (sip->si_mtx == NULL) {
1481 		sip->si_mtx = mtxp;
1482 		TAILQ_INIT(&sip->si_tdlist);
1483 	}
1484 	/*
1485 	 * Add this thread to the list of selfds listening on this selinfo.
1486 	 */
1487 	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1488 	mtx_unlock(sip->si_mtx);
1489 }
1490 
1491 /* Wake up a selecting thread. */
1492 void
1493 selwakeup(sip)
1494 	struct selinfo *sip;
1495 {
1496 	doselwakeup(sip, -1);
1497 }
1498 
1499 /* Wake up a selecting thread, and set its priority. */
1500 void
1501 selwakeuppri(sip, pri)
1502 	struct selinfo *sip;
1503 	int pri;
1504 {
1505 	doselwakeup(sip, pri);
1506 }
1507 
1508 /*
1509  * Do a wakeup when a selectable event occurs.
1510  */
1511 static void
1512 doselwakeup(sip, pri)
1513 	struct selinfo *sip;
1514 	int pri;
1515 {
1516 	struct selfd *sfp;
1517 	struct selfd *sfn;
1518 	struct seltd *stp;
1519 
1520 	/* If it's not initialized there can't be any waiters. */
1521 	if (sip->si_mtx == NULL)
1522 		return;
1523 	/*
1524 	 * Locking the selinfo locks all selfds associated with it.
1525 	 */
1526 	mtx_lock(sip->si_mtx);
1527 	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1528 		/*
1529 		 * Once we remove this sfp from the list and clear the
1530 		 * sf_si seltdclear will know to ignore this si.
1531 		 */
1532 		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1533 		sfp->sf_si = NULL;
1534 		stp = sfp->sf_td;
1535 		mtx_lock(&stp->st_mtx);
1536 		stp->st_flags |= SELTD_PENDING;
1537 		cv_broadcastpri(&stp->st_wait, pri);
1538 		mtx_unlock(&stp->st_mtx);
1539 	}
1540 	mtx_unlock(sip->si_mtx);
1541 }
1542 
1543 static void
1544 seltdinit(struct thread *td)
1545 {
1546 	struct seltd *stp;
1547 
1548 	if ((stp = td->td_sel) != NULL)
1549 		goto out;
1550 	td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1551 	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1552 	cv_init(&stp->st_wait, "select");
1553 out:
1554 	stp->st_flags = 0;
1555 	STAILQ_INIT(&stp->st_selq);
1556 }
1557 
1558 static int
1559 seltdwait(struct thread *td, int timo)
1560 {
1561 	struct seltd *stp;
1562 	int error;
1563 
1564 	stp = td->td_sel;
1565 	/*
1566 	 * An event of interest may occur while we do not hold the seltd
1567 	 * locked so check the pending flag before we sleep.
1568 	 */
1569 	mtx_lock(&stp->st_mtx);
1570 	/*
1571 	 * Any further calls to selrecord will be a rescan.
1572 	 */
1573 	stp->st_flags |= SELTD_RESCAN;
1574 	if (stp->st_flags & SELTD_PENDING) {
1575 		mtx_unlock(&stp->st_mtx);
1576 		return (0);
1577 	}
1578 	if (timo > 0)
1579 		error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1580 	else
1581 		error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1582 	mtx_unlock(&stp->st_mtx);
1583 
1584 	return (error);
1585 }
1586 
1587 void
1588 seltdfini(struct thread *td)
1589 {
1590 	struct seltd *stp;
1591 
1592 	stp = td->td_sel;
1593 	if (stp == NULL)
1594 		return;
1595 	if (stp->st_free1)
1596 		uma_zfree(selfd_zone, stp->st_free1);
1597 	if (stp->st_free2)
1598 		uma_zfree(selfd_zone, stp->st_free2);
1599 	td->td_sel = NULL;
1600 	free(stp, M_SELECT);
1601 }
1602 
1603 /*
1604  * Remove the references to the thread from all of the objects we were
1605  * polling.
1606  */
1607 static void
1608 seltdclear(struct thread *td)
1609 {
1610 	struct seltd *stp;
1611 	struct selfd *sfp;
1612 	struct selfd *sfn;
1613 
1614 	stp = td->td_sel;
1615 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1616 		selfdfree(stp, sfp);
1617 	stp->st_flags = 0;
1618 }
1619 
1620 static void selectinit(void *);
1621 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1622 static void
1623 selectinit(void *dummy __unused)
1624 {
1625 
1626 	selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1627 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1628 	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1629 }
1630