1/*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1993
5 *	The Regents of the University of California.  All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 *    notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 *    notice, this list of conditions and the following disclaimer in the
19 *    documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 *    may be used to endorse or promote products derived from this software
22 *    without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 *	@(#)sys_generic.c	8.5 (Berkeley) 1/21/94
37 */
38
39#include <sys/cdefs.h>
40__FBSDID("$FreeBSD$");
41
42#include "opt_capsicum.h"
43#include "opt_ktrace.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/sysproto.h>
48#include <sys/capsicum.h>
49#include <sys/filedesc.h>
50#include <sys/filio.h>
51#include <sys/fcntl.h>
52#include <sys/file.h>
53#include <sys/lock.h>
54#include <sys/proc.h>
55#include <sys/signalvar.h>
56#include <sys/socketvar.h>
57#include <sys/uio.h>
58#include <sys/kernel.h>
59#include <sys/ktr.h>
60#include <sys/limits.h>
61#include <sys/malloc.h>
62#include <sys/poll.h>
63#include <sys/resourcevar.h>
64#include <sys/selinfo.h>
65#include <sys/sleepqueue.h>
66#include <sys/syscallsubr.h>
67#include <sys/sysctl.h>
68#include <sys/sysent.h>
69#include <sys/vnode.h>
70#include <sys/bio.h>
71#include <sys/buf.h>
72#include <sys/condvar.h>
73#ifdef KTRACE
74#include <sys/ktrace.h>
75#endif
76
77#include <security/audit/audit.h>
78
79/*
80 * The following macro defines how many bytes will be allocated from
81 * the stack instead of memory allocated when passing the IOCTL data
82 * structures from userspace and to the kernel. Some IOCTLs having
83 * small data structures are used very frequently and this small
84 * buffer on the stack gives a significant speedup improvement for
85 * those requests. The value of this define should be greater or equal
86 * to 64 bytes and should also be power of two. The data structure is
87 * currently hard-aligned to a 8-byte boundary on the stack. This
88 * should currently be sufficient for all supported platforms.
89 */
90#define	SYS_IOCTL_SMALL_SIZE	128	/* bytes */
91#define	SYS_IOCTL_SMALL_ALIGN	8	/* bytes */
92
93#ifdef __LP64__
94static int iosize_max_clamp = 0;
95SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
96    &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
97static int devfs_iosize_max_clamp = 1;
98SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
99    &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
100#endif
101
102/*
103 * Assert that the return value of read(2) and write(2) syscalls fits
104 * into a register.  If not, an architecture will need to provide the
105 * usermode wrappers to reconstruct the result.
106 */
107CTASSERT(sizeof(register_t) >= sizeof(size_t));
108
109static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
110static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
111MALLOC_DEFINE(M_IOV, "iov", "large iov's");
112
113static int	pollout(struct thread *, struct pollfd *, struct pollfd *,
114		    u_int);
115static int	pollscan(struct thread *, struct pollfd *, u_int);
116static int	pollrescan(struct thread *);
117static int	selscan(struct thread *, fd_mask **, fd_mask **, int);
118static int	selrescan(struct thread *, fd_mask **, fd_mask **);
119static void	selfdalloc(struct thread *, void *);
120static void	selfdfree(struct seltd *, struct selfd *);
121static int	dofileread(struct thread *, int, struct file *, struct uio *,
122		    off_t, int);
123static int	dofilewrite(struct thread *, int, struct file *, struct uio *,
124		    off_t, int);
125static void	doselwakeup(struct selinfo *, int);
126static void	seltdinit(struct thread *);
127static int	seltdwait(struct thread *, sbintime_t, sbintime_t);
128static void	seltdclear(struct thread *);
129
130/*
131 * One seltd per-thread allocated on demand as needed.
132 *
133 *	t - protected by st_mtx
134 * 	k - Only accessed by curthread or read-only
135 */
136struct seltd {
137	STAILQ_HEAD(, selfd)	st_selq;	/* (k) List of selfds. */
138	struct selfd		*st_free1;	/* (k) free fd for read set. */
139	struct selfd		*st_free2;	/* (k) free fd for write set. */
140	struct mtx		st_mtx;		/* Protects struct seltd */
141	struct cv		st_wait;	/* (t) Wait channel. */
142	int			st_flags;	/* (t) SELTD_ flags. */
143};
144
145#define	SELTD_PENDING	0x0001			/* We have pending events. */
146#define	SELTD_RESCAN	0x0002			/* Doing a rescan. */
147
148/*
149 * One selfd allocated per-thread per-file-descriptor.
150 *	f - protected by sf_mtx
151 */
152struct selfd {
153	STAILQ_ENTRY(selfd)	sf_link;	/* (k) fds owned by this td. */
154	TAILQ_ENTRY(selfd)	sf_threads;	/* (f) fds on this selinfo. */
155	struct selinfo		*sf_si;		/* (f) selinfo when linked. */
156	struct mtx		*sf_mtx;	/* Pointer to selinfo mtx. */
157	struct seltd		*sf_td;		/* (k) owning seltd. */
158	void			*sf_cookie;	/* (k) fd or pollfd. */
159	u_int			sf_refs;
160};
161
162static uma_zone_t selfd_zone;
163static struct mtx_pool *mtxpool_select;
164
165#ifdef __LP64__
166size_t
167devfs_iosize_max(void)
168{
169
170	return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
171	    INT_MAX : SSIZE_MAX);
172}
173
174size_t
175iosize_max(void)
176{
177
178	return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
179	    INT_MAX : SSIZE_MAX);
180}
181#endif
182
183#ifndef _SYS_SYSPROTO_H_
184struct read_args {
185	int	fd;
186	void	*buf;
187	size_t	nbyte;
188};
189#endif
190int
191sys_read(struct thread *td, struct read_args *uap)
192{
193	struct uio auio;
194	struct iovec aiov;
195	int error;
196
197	if (uap->nbyte > IOSIZE_MAX)
198		return (EINVAL);
199	aiov.iov_base = uap->buf;
200	aiov.iov_len = uap->nbyte;
201	auio.uio_iov = &aiov;
202	auio.uio_iovcnt = 1;
203	auio.uio_resid = uap->nbyte;
204	auio.uio_segflg = UIO_USERSPACE;
205	error = kern_readv(td, uap->fd, &auio);
206	return (error);
207}
208
209/*
210 * Positioned read system call
211 */
212#ifndef _SYS_SYSPROTO_H_
213struct pread_args {
214	int	fd;
215	void	*buf;
216	size_t	nbyte;
217	int	pad;
218	off_t	offset;
219};
220#endif
221int
222sys_pread(struct thread *td, struct pread_args *uap)
223{
224
225	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
226}
227
228int
229kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset)
230{
231	struct uio auio;
232	struct iovec aiov;
233	int error;
234
235	if (nbyte > IOSIZE_MAX)
236		return (EINVAL);
237	aiov.iov_base = buf;
238	aiov.iov_len = nbyte;
239	auio.uio_iov = &aiov;
240	auio.uio_iovcnt = 1;
241	auio.uio_resid = nbyte;
242	auio.uio_segflg = UIO_USERSPACE;
243	error = kern_preadv(td, fd, &auio, offset);
244	return (error);
245}
246
247#if defined(COMPAT_FREEBSD6)
248int
249freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap)
250{
251
252	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
253}
254#endif
255
256/*
257 * Scatter read system call.
258 */
259#ifndef _SYS_SYSPROTO_H_
260struct readv_args {
261	int	fd;
262	struct	iovec *iovp;
263	u_int	iovcnt;
264};
265#endif
266int
267sys_readv(struct thread *td, struct readv_args *uap)
268{
269	struct uio *auio;
270	int error;
271
272	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
273	if (error)
274		return (error);
275	error = kern_readv(td, uap->fd, auio);
276	free(auio, M_IOV);
277	return (error);
278}
279
280int
281kern_readv(struct thread *td, int fd, struct uio *auio)
282{
283	struct file *fp;
284	int error;
285
286	error = fget_read(td, fd, &cap_read_rights, &fp);
287	if (error)
288		return (error);
289	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
290	fdrop(fp, td);
291	return (error);
292}
293
294/*
295 * Scatter positioned read system call.
296 */
297#ifndef _SYS_SYSPROTO_H_
298struct preadv_args {
299	int	fd;
300	struct	iovec *iovp;
301	u_int	iovcnt;
302	off_t	offset;
303};
304#endif
305int
306sys_preadv(struct thread *td, struct preadv_args *uap)
307{
308	struct uio *auio;
309	int error;
310
311	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
312	if (error)
313		return (error);
314	error = kern_preadv(td, uap->fd, auio, uap->offset);
315	free(auio, M_IOV);
316	return (error);
317}
318
319int
320kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset)
321{
322	struct file *fp;
323	int error;
324
325	error = fget_read(td, fd, &cap_pread_rights, &fp);
326	if (error)
327		return (error);
328	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
329		error = ESPIPE;
330	else if (offset < 0 &&
331	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
332		error = EINVAL;
333	else
334		error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
335	fdrop(fp, td);
336	return (error);
337}
338
339/*
340 * Common code for readv and preadv that reads data in
341 * from a file using the passed in uio, offset, and flags.
342 */
343static int
344dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio,
345    off_t offset, int flags)
346{
347	ssize_t cnt;
348	int error;
349#ifdef KTRACE
350	struct uio *ktruio = NULL;
351#endif
352
353	AUDIT_ARG_FD(fd);
354
355	/* Finish zero length reads right here */
356	if (auio->uio_resid == 0) {
357		td->td_retval[0] = 0;
358		return (0);
359	}
360	auio->uio_rw = UIO_READ;
361	auio->uio_offset = offset;
362	auio->uio_td = td;
363#ifdef KTRACE
364	if (KTRPOINT(td, KTR_GENIO))
365		ktruio = cloneuio(auio);
366#endif
367	cnt = auio->uio_resid;
368	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
369		if (auio->uio_resid != cnt && (error == ERESTART ||
370		    error == EINTR || error == EWOULDBLOCK))
371			error = 0;
372	}
373	cnt -= auio->uio_resid;
374#ifdef KTRACE
375	if (ktruio != NULL) {
376		ktruio->uio_resid = cnt;
377		ktrgenio(fd, UIO_READ, ktruio, error);
378	}
379#endif
380	td->td_retval[0] = cnt;
381	return (error);
382}
383
384#ifndef _SYS_SYSPROTO_H_
385struct write_args {
386	int	fd;
387	const void *buf;
388	size_t	nbyte;
389};
390#endif
391int
392sys_write(struct thread *td, struct write_args *uap)
393{
394	struct uio auio;
395	struct iovec aiov;
396	int error;
397
398	if (uap->nbyte > IOSIZE_MAX)
399		return (EINVAL);
400	aiov.iov_base = (void *)(uintptr_t)uap->buf;
401	aiov.iov_len = uap->nbyte;
402	auio.uio_iov = &aiov;
403	auio.uio_iovcnt = 1;
404	auio.uio_resid = uap->nbyte;
405	auio.uio_segflg = UIO_USERSPACE;
406	error = kern_writev(td, uap->fd, &auio);
407	return (error);
408}
409
410/*
411 * Positioned write system call.
412 */
413#ifndef _SYS_SYSPROTO_H_
414struct pwrite_args {
415	int	fd;
416	const void *buf;
417	size_t	nbyte;
418	int	pad;
419	off_t	offset;
420};
421#endif
422int
423sys_pwrite(struct thread *td, struct pwrite_args *uap)
424{
425
426	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
427}
428
429int
430kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte,
431    off_t offset)
432{
433	struct uio auio;
434	struct iovec aiov;
435	int error;
436
437	if (nbyte > IOSIZE_MAX)
438		return (EINVAL);
439	aiov.iov_base = (void *)(uintptr_t)buf;
440	aiov.iov_len = nbyte;
441	auio.uio_iov = &aiov;
442	auio.uio_iovcnt = 1;
443	auio.uio_resid = nbyte;
444	auio.uio_segflg = UIO_USERSPACE;
445	error = kern_pwritev(td, fd, &auio, offset);
446	return (error);
447}
448
449#if defined(COMPAT_FREEBSD6)
450int
451freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap)
452{
453
454	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
455}
456#endif
457
458/*
459 * Gather write system call.
460 */
461#ifndef _SYS_SYSPROTO_H_
462struct writev_args {
463	int	fd;
464	struct	iovec *iovp;
465	u_int	iovcnt;
466};
467#endif
468int
469sys_writev(struct thread *td, struct writev_args *uap)
470{
471	struct uio *auio;
472	int error;
473
474	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
475	if (error)
476		return (error);
477	error = kern_writev(td, uap->fd, auio);
478	free(auio, M_IOV);
479	return (error);
480}
481
482int
483kern_writev(struct thread *td, int fd, struct uio *auio)
484{
485	struct file *fp;
486	int error;
487
488	error = fget_write(td, fd, &cap_write_rights, &fp);
489	if (error)
490		return (error);
491	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
492	fdrop(fp, td);
493	return (error);
494}
495
496/*
497 * Gather positioned write system call.
498 */
499#ifndef _SYS_SYSPROTO_H_
500struct pwritev_args {
501	int	fd;
502	struct	iovec *iovp;
503	u_int	iovcnt;
504	off_t	offset;
505};
506#endif
507int
508sys_pwritev(struct thread *td, struct pwritev_args *uap)
509{
510	struct uio *auio;
511	int error;
512
513	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
514	if (error)
515		return (error);
516	error = kern_pwritev(td, uap->fd, auio, uap->offset);
517	free(auio, M_IOV);
518	return (error);
519}
520
521int
522kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset)
523{
524	struct file *fp;
525	int error;
526
527	error = fget_write(td, fd, &cap_pwrite_rights, &fp);
528	if (error)
529		return (error);
530	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
531		error = ESPIPE;
532	else if (offset < 0 &&
533	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
534		error = EINVAL;
535	else
536		error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
537	fdrop(fp, td);
538	return (error);
539}
540
541/*
542 * Common code for writev and pwritev that writes data to
543 * a file using the passed in uio, offset, and flags.
544 */
545static int
546dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio,
547    off_t offset, int flags)
548{
549	ssize_t cnt;
550	int error;
551#ifdef KTRACE
552	struct uio *ktruio = NULL;
553#endif
554
555	AUDIT_ARG_FD(fd);
556	auio->uio_rw = UIO_WRITE;
557	auio->uio_td = td;
558	auio->uio_offset = offset;
559#ifdef KTRACE
560	if (KTRPOINT(td, KTR_GENIO))
561		ktruio = cloneuio(auio);
562#endif
563	cnt = auio->uio_resid;
564	if (fp->f_type == DTYPE_VNODE &&
565	    (fp->f_vnread_flags & FDEVFS_VNODE) == 0)
566		bwillwrite();
567	if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
568		if (auio->uio_resid != cnt && (error == ERESTART ||
569		    error == EINTR || error == EWOULDBLOCK))
570			error = 0;
571		/* Socket layer is responsible for issuing SIGPIPE. */
572		if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
573			PROC_LOCK(td->td_proc);
574			tdsignal(td, SIGPIPE);
575			PROC_UNLOCK(td->td_proc);
576		}
577	}
578	cnt -= auio->uio_resid;
579#ifdef KTRACE
580	if (ktruio != NULL) {
581		ktruio->uio_resid = cnt;
582		ktrgenio(fd, UIO_WRITE, ktruio, error);
583	}
584#endif
585	td->td_retval[0] = cnt;
586	return (error);
587}
588
589/*
590 * Truncate a file given a file descriptor.
591 *
592 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
593 * descriptor isn't writable.
594 */
595int
596kern_ftruncate(struct thread *td, int fd, off_t length)
597{
598	struct file *fp;
599	int error;
600
601	AUDIT_ARG_FD(fd);
602	if (length < 0)
603		return (EINVAL);
604	error = fget(td, fd, &cap_ftruncate_rights, &fp);
605	if (error)
606		return (error);
607	AUDIT_ARG_FILE(td->td_proc, fp);
608	if (!(fp->f_flag & FWRITE)) {
609		fdrop(fp, td);
610		return (EINVAL);
611	}
612	error = fo_truncate(fp, length, td->td_ucred, td);
613	fdrop(fp, td);
614	return (error);
615}
616
617#ifndef _SYS_SYSPROTO_H_
618struct ftruncate_args {
619	int	fd;
620	int	pad;
621	off_t	length;
622};
623#endif
624int
625sys_ftruncate(struct thread *td, struct ftruncate_args *uap)
626{
627
628	return (kern_ftruncate(td, uap->fd, uap->length));
629}
630
631#if defined(COMPAT_43)
632#ifndef _SYS_SYSPROTO_H_
633struct oftruncate_args {
634	int	fd;
635	long	length;
636};
637#endif
638int
639oftruncate(struct thread *td, struct oftruncate_args *uap)
640{
641
642	return (kern_ftruncate(td, uap->fd, uap->length));
643}
644#endif /* COMPAT_43 */
645
646#ifndef _SYS_SYSPROTO_H_
647struct ioctl_args {
648	int	fd;
649	u_long	com;
650	caddr_t	data;
651};
652#endif
653/* ARGSUSED */
654int
655sys_ioctl(struct thread *td, struct ioctl_args *uap)
656{
657	u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
658	u_long com;
659	int arg, error;
660	u_int size;
661	caddr_t data;
662
663	if (uap->com > 0xffffffff) {
664		printf(
665		    "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
666		    td->td_proc->p_pid, td->td_name, uap->com);
667		uap->com &= 0xffffffff;
668	}
669	com = uap->com;
670
671	/*
672	 * Interpret high order word to find amount of data to be
673	 * copied to/from the user's address space.
674	 */
675	size = IOCPARM_LEN(com);
676	if ((size > IOCPARM_MAX) ||
677	    ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
678#if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
679	    ((com & IOC_OUT) && size == 0) ||
680#else
681	    ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
682#endif
683	    ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
684		return (ENOTTY);
685
686	if (size > 0) {
687		if (com & IOC_VOID) {
688			/* Integer argument. */
689			arg = (intptr_t)uap->data;
690			data = (void *)&arg;
691			size = 0;
692		} else {
693			if (size > SYS_IOCTL_SMALL_SIZE)
694				data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
695			else
696				data = smalldata;
697		}
698	} else
699		data = (void *)&uap->data;
700	if (com & IOC_IN) {
701		error = copyin(uap->data, data, (u_int)size);
702		if (error != 0)
703			goto out;
704	} else if (com & IOC_OUT) {
705		/*
706		 * Zero the buffer so the user always
707		 * gets back something deterministic.
708		 */
709		bzero(data, size);
710	}
711
712	error = kern_ioctl(td, uap->fd, com, data);
713
714	if (error == 0 && (com & IOC_OUT))
715		error = copyout(data, uap->data, (u_int)size);
716
717out:
718	if (size > SYS_IOCTL_SMALL_SIZE)
719		free(data, M_IOCTLOPS);
720	return (error);
721}
722
723int
724kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
725{
726	struct file *fp;
727	struct filedesc *fdp;
728	int error, tmp, locked;
729
730	AUDIT_ARG_FD(fd);
731	AUDIT_ARG_CMD(com);
732
733	fdp = td->td_proc->p_fd;
734
735	switch (com) {
736	case FIONCLEX:
737	case FIOCLEX:
738		FILEDESC_XLOCK(fdp);
739		locked = LA_XLOCKED;
740		break;
741	default:
742#ifdef CAPABILITIES
743		FILEDESC_SLOCK(fdp);
744		locked = LA_SLOCKED;
745#else
746		locked = LA_UNLOCKED;
747#endif
748		break;
749	}
750
751#ifdef CAPABILITIES
752	if ((fp = fget_locked(fdp, fd)) == NULL) {
753		error = EBADF;
754		goto out;
755	}
756	if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
757		fp = NULL;	/* fhold() was not called yet */
758		goto out;
759	}
760	fhold(fp);
761	if (locked == LA_SLOCKED) {
762		FILEDESC_SUNLOCK(fdp);
763		locked = LA_UNLOCKED;
764	}
765#else
766	error = fget(td, fd, &cap_ioctl_rights, &fp);
767	if (error != 0) {
768		fp = NULL;
769		goto out;
770	}
771#endif
772	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
773		error = EBADF;
774		goto out;
775	}
776
777	switch (com) {
778	case FIONCLEX:
779		fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
780		goto out;
781	case FIOCLEX:
782		fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
783		goto out;
784	case FIONBIO:
785		if ((tmp = *(int *)data))
786			atomic_set_int(&fp->f_flag, FNONBLOCK);
787		else
788			atomic_clear_int(&fp->f_flag, FNONBLOCK);
789		data = (void *)&tmp;
790		break;
791	case FIOASYNC:
792		if ((tmp = *(int *)data))
793			atomic_set_int(&fp->f_flag, FASYNC);
794		else
795			atomic_clear_int(&fp->f_flag, FASYNC);
796		data = (void *)&tmp;
797		break;
798	}
799
800	error = fo_ioctl(fp, com, data, td->td_ucred, td);
801out:
802	switch (locked) {
803	case LA_XLOCKED:
804		FILEDESC_XUNLOCK(fdp);
805		break;
806#ifdef CAPABILITIES
807	case LA_SLOCKED:
808		FILEDESC_SUNLOCK(fdp);
809		break;
810#endif
811	default:
812		FILEDESC_UNLOCK_ASSERT(fdp);
813		break;
814	}
815	if (fp != NULL)
816		fdrop(fp, td);
817	return (error);
818}
819
820int
821poll_no_poll(int events)
822{
823	/*
824	 * Return true for read/write.  If the user asked for something
825	 * special, return POLLNVAL, so that clients have a way of
826	 * determining reliably whether or not the extended
827	 * functionality is present without hard-coding knowledge
828	 * of specific filesystem implementations.
829	 */
830	if (events & ~POLLSTANDARD)
831		return (POLLNVAL);
832
833	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
834}
835
836int
837sys_pselect(struct thread *td, struct pselect_args *uap)
838{
839	struct timespec ts;
840	struct timeval tv, *tvp;
841	sigset_t set, *uset;
842	int error;
843
844	if (uap->ts != NULL) {
845		error = copyin(uap->ts, &ts, sizeof(ts));
846		if (error != 0)
847		    return (error);
848		TIMESPEC_TO_TIMEVAL(&tv, &ts);
849		tvp = &tv;
850	} else
851		tvp = NULL;
852	if (uap->sm != NULL) {
853		error = copyin(uap->sm, &set, sizeof(set));
854		if (error != 0)
855			return (error);
856		uset = &set;
857	} else
858		uset = NULL;
859	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
860	    uset, NFDBITS));
861}
862
863int
864kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
865    struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
866{
867	int error;
868
869	if (uset != NULL) {
870		error = kern_sigprocmask(td, SIG_SETMASK, uset,
871		    &td->td_oldsigmask, 0);
872		if (error != 0)
873			return (error);
874		td->td_pflags |= TDP_OLDMASK;
875		/*
876		 * Make sure that ast() is called on return to
877		 * usermode and TDP_OLDMASK is cleared, restoring old
878		 * sigmask.
879		 */
880		thread_lock(td);
881		td->td_flags |= TDF_ASTPENDING;
882		thread_unlock(td);
883	}
884	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
885	return (error);
886}
887
888#ifndef _SYS_SYSPROTO_H_
889struct select_args {
890	int	nd;
891	fd_set	*in, *ou, *ex;
892	struct	timeval *tv;
893};
894#endif
895int
896sys_select(struct thread *td, struct select_args *uap)
897{
898	struct timeval tv, *tvp;
899	int error;
900
901	if (uap->tv != NULL) {
902		error = copyin(uap->tv, &tv, sizeof(tv));
903		if (error)
904			return (error);
905		tvp = &tv;
906	} else
907		tvp = NULL;
908
909	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
910	    NFDBITS));
911}
912
913/*
914 * In the unlikely case when user specified n greater then the last
915 * open file descriptor, check that no bits are set after the last
916 * valid fd.  We must return EBADF if any is set.
917 *
918 * There are applications that rely on the behaviour.
919 *
920 * nd is fd_lastfile + 1.
921 */
922static int
923select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
924{
925	char *addr, *oaddr;
926	int b, i, res;
927	uint8_t bits;
928
929	if (nd >= ndu || fd_in == NULL)
930		return (0);
931
932	oaddr = NULL;
933	bits = 0; /* silence gcc */
934	for (i = nd; i < ndu; i++) {
935		b = i / NBBY;
936#if BYTE_ORDER == LITTLE_ENDIAN
937		addr = (char *)fd_in + b;
938#else
939		addr = (char *)fd_in;
940		if (abi_nfdbits == NFDBITS) {
941			addr += rounddown(b, sizeof(fd_mask)) +
942			    sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
943		} else {
944			addr += rounddown(b, sizeof(uint32_t)) +
945			    sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
946		}
947#endif
948		if (addr != oaddr) {
949			res = fubyte(addr);
950			if (res == -1)
951				return (EFAULT);
952			oaddr = addr;
953			bits = res;
954		}
955		if ((bits & (1 << (i % NBBY))) != 0)
956			return (EBADF);
957	}
958	return (0);
959}
960
961int
962kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
963    fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
964{
965	struct filedesc *fdp;
966	/*
967	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
968	 * infds with the new FD_SETSIZE of 1024, and more than enough for
969	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
970	 * of 256.
971	 */
972	fd_mask s_selbits[howmany(2048, NFDBITS)];
973	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
974	struct timeval rtv;
975	sbintime_t asbt, precision, rsbt;
976	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
977	int error, lf, ndu;
978
979	if (nd < 0)
980		return (EINVAL);
981	fdp = td->td_proc->p_fd;
982	ndu = nd;
983	lf = fdp->fd_lastfile;
984	if (nd > lf + 1)
985		nd = lf + 1;
986
987	error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
988	if (error != 0)
989		return (error);
990	error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
991	if (error != 0)
992		return (error);
993	error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
994	if (error != 0)
995		return (error);
996
997	/*
998	 * Allocate just enough bits for the non-null fd_sets.  Use the
999	 * preallocated auto buffer if possible.
1000	 */
1001	nfdbits = roundup(nd, NFDBITS);
1002	ncpbytes = nfdbits / NBBY;
1003	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1004	nbufbytes = 0;
1005	if (fd_in != NULL)
1006		nbufbytes += 2 * ncpbytes;
1007	if (fd_ou != NULL)
1008		nbufbytes += 2 * ncpbytes;
1009	if (fd_ex != NULL)
1010		nbufbytes += 2 * ncpbytes;
1011	if (nbufbytes <= sizeof s_selbits)
1012		selbits = &s_selbits[0];
1013	else
1014		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1015
1016	/*
1017	 * Assign pointers into the bit buffers and fetch the input bits.
1018	 * Put the output buffers together so that they can be bzeroed
1019	 * together.
1020	 */
1021	sbp = selbits;
1022#define	getbits(name, x) \
1023	do {								\
1024		if (name == NULL) {					\
1025			ibits[x] = NULL;				\
1026			obits[x] = NULL;				\
1027		} else {						\
1028			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
1029			obits[x] = sbp;					\
1030			sbp += ncpbytes / sizeof *sbp;			\
1031			error = copyin(name, ibits[x], ncpubytes);	\
1032			if (error != 0)					\
1033				goto done;				\
1034			if (ncpbytes != ncpubytes)			\
1035				bzero((char *)ibits[x] + ncpubytes,	\
1036				    ncpbytes - ncpubytes);		\
1037		}							\
1038	} while (0)
1039	getbits(fd_in, 0);
1040	getbits(fd_ou, 1);
1041	getbits(fd_ex, 2);
1042#undef	getbits
1043
1044#if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1045	/*
1046	 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1047	 * we are running under 32-bit emulation. This should be more
1048	 * generic.
1049	 */
1050#define swizzle_fdset(bits)						\
1051	if (abi_nfdbits != NFDBITS && bits != NULL) {			\
1052		int i;							\
1053		for (i = 0; i < ncpbytes / sizeof *sbp; i++)		\
1054			bits[i] = (bits[i] >> 32) | (bits[i] << 32);	\
1055	}
1056#else
1057#define swizzle_fdset(bits)
1058#endif
1059
1060	/* Make sure the bit order makes it through an ABI transition */
1061	swizzle_fdset(ibits[0]);
1062	swizzle_fdset(ibits[1]);
1063	swizzle_fdset(ibits[2]);
1064
1065	if (nbufbytes != 0)
1066		bzero(selbits, nbufbytes / 2);
1067
1068	precision = 0;
1069	if (tvp != NULL) {
1070		rtv = *tvp;
1071		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1072		    rtv.tv_usec >= 1000000) {
1073			error = EINVAL;
1074			goto done;
1075		}
1076		if (!timevalisset(&rtv))
1077			asbt = 0;
1078		else if (rtv.tv_sec <= INT32_MAX) {
1079			rsbt = tvtosbt(rtv);
1080			precision = rsbt;
1081			precision >>= tc_precexp;
1082			if (TIMESEL(&asbt, rsbt))
1083				asbt += tc_tick_sbt;
1084			if (asbt <= SBT_MAX - rsbt)
1085				asbt += rsbt;
1086			else
1087				asbt = -1;
1088		} else
1089			asbt = -1;
1090	} else
1091		asbt = -1;
1092	seltdinit(td);
1093	/* Iterate until the timeout expires or descriptors become ready. */
1094	for (;;) {
1095		error = selscan(td, ibits, obits, nd);
1096		if (error || td->td_retval[0] != 0)
1097			break;
1098		error = seltdwait(td, asbt, precision);
1099		if (error)
1100			break;
1101		error = selrescan(td, ibits, obits);
1102		if (error || td->td_retval[0] != 0)
1103			break;
1104	}
1105	seltdclear(td);
1106
1107done:
1108	/* select is not restarted after signals... */
1109	if (error == ERESTART)
1110		error = EINTR;
1111	if (error == EWOULDBLOCK)
1112		error = 0;
1113
1114	/* swizzle bit order back, if necessary */
1115	swizzle_fdset(obits[0]);
1116	swizzle_fdset(obits[1]);
1117	swizzle_fdset(obits[2]);
1118#undef swizzle_fdset
1119
1120#define	putbits(name, x) \
1121	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1122		error = error2;
1123	if (error == 0) {
1124		int error2;
1125
1126		putbits(fd_in, 0);
1127		putbits(fd_ou, 1);
1128		putbits(fd_ex, 2);
1129#undef putbits
1130	}
1131	if (selbits != &s_selbits[0])
1132		free(selbits, M_SELECT);
1133
1134	return (error);
1135}
1136/*
1137 * Convert a select bit set to poll flags.
1138 *
1139 * The backend always returns POLLHUP/POLLERR if appropriate and we
1140 * return this as a set bit in any set.
1141 */
1142static int select_flags[3] = {
1143    POLLRDNORM | POLLHUP | POLLERR,
1144    POLLWRNORM | POLLHUP | POLLERR,
1145    POLLRDBAND | POLLERR
1146};
1147
1148/*
1149 * Compute the fo_poll flags required for a fd given by the index and
1150 * bit position in the fd_mask array.
1151 */
1152static __inline int
1153selflags(fd_mask **ibits, int idx, fd_mask bit)
1154{
1155	int flags;
1156	int msk;
1157
1158	flags = 0;
1159	for (msk = 0; msk < 3; msk++) {
1160		if (ibits[msk] == NULL)
1161			continue;
1162		if ((ibits[msk][idx] & bit) == 0)
1163			continue;
1164		flags |= select_flags[msk];
1165	}
1166	return (flags);
1167}
1168
1169/*
1170 * Set the appropriate output bits given a mask of fired events and the
1171 * input bits originally requested.
1172 */
1173static __inline int
1174selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1175{
1176	int msk;
1177	int n;
1178
1179	n = 0;
1180	for (msk = 0; msk < 3; msk++) {
1181		if ((events & select_flags[msk]) == 0)
1182			continue;
1183		if (ibits[msk] == NULL)
1184			continue;
1185		if ((ibits[msk][idx] & bit) == 0)
1186			continue;
1187		/*
1188		 * XXX Check for a duplicate set.  This can occur because a
1189		 * socket calls selrecord() twice for each poll() call
1190		 * resulting in two selfds per real fd.  selrescan() will
1191		 * call selsetbits twice as a result.
1192		 */
1193		if ((obits[msk][idx] & bit) != 0)
1194			continue;
1195		obits[msk][idx] |= bit;
1196		n++;
1197	}
1198
1199	return (n);
1200}
1201
1202static __inline int
1203getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1204{
1205
1206	return (fget_unlocked(fdp, fd, &cap_event_rights, fpp, NULL));
1207}
1208
1209/*
1210 * Traverse the list of fds attached to this thread's seltd and check for
1211 * completion.
1212 */
1213static int
1214selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1215{
1216	struct filedesc *fdp;
1217	struct selinfo *si;
1218	struct seltd *stp;
1219	struct selfd *sfp;
1220	struct selfd *sfn;
1221	struct file *fp;
1222	fd_mask bit;
1223	int fd, ev, n, idx;
1224	int error;
1225
1226	fdp = td->td_proc->p_fd;
1227	stp = td->td_sel;
1228	n = 0;
1229	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1230		fd = (int)(uintptr_t)sfp->sf_cookie;
1231		si = sfp->sf_si;
1232		selfdfree(stp, sfp);
1233		/* If the selinfo wasn't cleared the event didn't fire. */
1234		if (si != NULL)
1235			continue;
1236		error = getselfd_cap(fdp, fd, &fp);
1237		if (error)
1238			return (error);
1239		idx = fd / NFDBITS;
1240		bit = (fd_mask)1 << (fd % NFDBITS);
1241		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1242		fdrop(fp, td);
1243		if (ev != 0)
1244			n += selsetbits(ibits, obits, idx, bit, ev);
1245	}
1246	stp->st_flags = 0;
1247	td->td_retval[0] = n;
1248	return (0);
1249}
1250
1251/*
1252 * Perform the initial filedescriptor scan and register ourselves with
1253 * each selinfo.
1254 */
1255static int
1256selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd)
1257{
1258	struct filedesc *fdp;
1259	struct file *fp;
1260	fd_mask bit;
1261	int ev, flags, end, fd;
1262	int n, idx;
1263	int error;
1264
1265	fdp = td->td_proc->p_fd;
1266	n = 0;
1267	for (idx = 0, fd = 0; fd < nfd; idx++) {
1268		end = imin(fd + NFDBITS, nfd);
1269		for (bit = 1; fd < end; bit <<= 1, fd++) {
1270			/* Compute the list of events we're interested in. */
1271			flags = selflags(ibits, idx, bit);
1272			if (flags == 0)
1273				continue;
1274			error = getselfd_cap(fdp, fd, &fp);
1275			if (error)
1276				return (error);
1277			selfdalloc(td, (void *)(uintptr_t)fd);
1278			ev = fo_poll(fp, flags, td->td_ucred, td);
1279			fdrop(fp, td);
1280			if (ev != 0)
1281				n += selsetbits(ibits, obits, idx, bit, ev);
1282		}
1283	}
1284
1285	td->td_retval[0] = n;
1286	return (0);
1287}
1288
1289int
1290sys_poll(struct thread *td, struct poll_args *uap)
1291{
1292	struct timespec ts, *tsp;
1293
1294	if (uap->timeout != INFTIM) {
1295		if (uap->timeout < 0)
1296			return (EINVAL);
1297		ts.tv_sec = uap->timeout / 1000;
1298		ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1299		tsp = &ts;
1300	} else
1301		tsp = NULL;
1302
1303	return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1304}
1305
1306int
1307kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds,
1308    struct timespec *tsp, sigset_t *uset)
1309{
1310	struct pollfd *kfds;
1311	struct pollfd stackfds[32];
1312	sbintime_t sbt, precision, tmp;
1313	time_t over;
1314	struct timespec ts;
1315	int error;
1316
1317	precision = 0;
1318	if (tsp != NULL) {
1319		if (tsp->tv_sec < 0)
1320			return (EINVAL);
1321		if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000)
1322			return (EINVAL);
1323		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1324			sbt = 0;
1325		else {
1326			ts = *tsp;
1327			if (ts.tv_sec > INT32_MAX / 2) {
1328				over = ts.tv_sec - INT32_MAX / 2;
1329				ts.tv_sec -= over;
1330			} else
1331				over = 0;
1332			tmp = tstosbt(ts);
1333			precision = tmp;
1334			precision >>= tc_precexp;
1335			if (TIMESEL(&sbt, tmp))
1336				sbt += tc_tick_sbt;
1337			sbt += tmp;
1338		}
1339	} else
1340		sbt = -1;
1341
1342	/*
1343	 * This is kinda bogus.  We have fd limits, but that is not
1344	 * really related to the size of the pollfd array.  Make sure
1345	 * we let the process use at least FD_SETSIZE entries and at
1346	 * least enough for the system-wide limits.  We want to be reasonably
1347	 * safe, but not overly restrictive.
1348	 */
1349	if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1350		return (EINVAL);
1351	if (nfds > nitems(stackfds))
1352		kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK);
1353	else
1354		kfds = stackfds;
1355	error = copyin(ufds, kfds, nfds * sizeof(*kfds));
1356	if (error)
1357		goto done;
1358
1359	if (uset != NULL) {
1360		error = kern_sigprocmask(td, SIG_SETMASK, uset,
1361		    &td->td_oldsigmask, 0);
1362		if (error)
1363			goto done;
1364		td->td_pflags |= TDP_OLDMASK;
1365		/*
1366		 * Make sure that ast() is called on return to
1367		 * usermode and TDP_OLDMASK is cleared, restoring old
1368		 * sigmask.
1369		 */
1370		thread_lock(td);
1371		td->td_flags |= TDF_ASTPENDING;
1372		thread_unlock(td);
1373	}
1374
1375	seltdinit(td);
1376	/* Iterate until the timeout expires or descriptors become ready. */
1377	for (;;) {
1378		error = pollscan(td, kfds, nfds);
1379		if (error || td->td_retval[0] != 0)
1380			break;
1381		error = seltdwait(td, sbt, precision);
1382		if (error)
1383			break;
1384		error = pollrescan(td);
1385		if (error || td->td_retval[0] != 0)
1386			break;
1387	}
1388	seltdclear(td);
1389
1390done:
1391	/* poll is not restarted after signals... */
1392	if (error == ERESTART)
1393		error = EINTR;
1394	if (error == EWOULDBLOCK)
1395		error = 0;
1396	if (error == 0) {
1397		error = pollout(td, kfds, ufds, nfds);
1398		if (error)
1399			goto out;
1400	}
1401out:
1402	if (nfds > nitems(stackfds))
1403		free(kfds, M_TEMP);
1404	return (error);
1405}
1406
1407int
1408sys_ppoll(struct thread *td, struct ppoll_args *uap)
1409{
1410	struct timespec ts, *tsp;
1411	sigset_t set, *ssp;
1412	int error;
1413
1414	if (uap->ts != NULL) {
1415		error = copyin(uap->ts, &ts, sizeof(ts));
1416		if (error)
1417			return (error);
1418		tsp = &ts;
1419	} else
1420		tsp = NULL;
1421	if (uap->set != NULL) {
1422		error = copyin(uap->set, &set, sizeof(set));
1423		if (error)
1424			return (error);
1425		ssp = &set;
1426	} else
1427		ssp = NULL;
1428	/*
1429	 * fds is still a pointer to user space. kern_poll() will
1430	 * take care of copyin that array to the kernel space.
1431	 */
1432
1433	return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1434}
1435
1436static int
1437pollrescan(struct thread *td)
1438{
1439	struct seltd *stp;
1440	struct selfd *sfp;
1441	struct selfd *sfn;
1442	struct selinfo *si;
1443	struct filedesc *fdp;
1444	struct file *fp;
1445	struct pollfd *fd;
1446	int n;
1447
1448	n = 0;
1449	fdp = td->td_proc->p_fd;
1450	stp = td->td_sel;
1451	FILEDESC_SLOCK(fdp);
1452	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1453		fd = (struct pollfd *)sfp->sf_cookie;
1454		si = sfp->sf_si;
1455		selfdfree(stp, sfp);
1456		/* If the selinfo wasn't cleared the event didn't fire. */
1457		if (si != NULL)
1458			continue;
1459		fp = fdp->fd_ofiles[fd->fd].fde_file;
1460#ifdef CAPABILITIES
1461		if (fp == NULL ||
1462		    cap_check(cap_rights(fdp, fd->fd), &cap_event_rights) != 0)
1463#else
1464		if (fp == NULL)
1465#endif
1466		{
1467			fd->revents = POLLNVAL;
1468			n++;
1469			continue;
1470		}
1471
1472		/*
1473		 * Note: backend also returns POLLHUP and
1474		 * POLLERR if appropriate.
1475		 */
1476		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1477		if (fd->revents != 0)
1478			n++;
1479	}
1480	FILEDESC_SUNLOCK(fdp);
1481	stp->st_flags = 0;
1482	td->td_retval[0] = n;
1483	return (0);
1484}
1485
1486
1487static int
1488pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd)
1489{
1490	int error = 0;
1491	u_int i = 0;
1492	u_int n = 0;
1493
1494	for (i = 0; i < nfd; i++) {
1495		error = copyout(&fds->revents, &ufds->revents,
1496		    sizeof(ufds->revents));
1497		if (error)
1498			return (error);
1499		if (fds->revents != 0)
1500			n++;
1501		fds++;
1502		ufds++;
1503	}
1504	td->td_retval[0] = n;
1505	return (0);
1506}
1507
1508static int
1509pollscan(struct thread *td, struct pollfd *fds, u_int nfd)
1510{
1511	struct filedesc *fdp = td->td_proc->p_fd;
1512	struct file *fp;
1513	int i, n = 0;
1514
1515	FILEDESC_SLOCK(fdp);
1516	for (i = 0; i < nfd; i++, fds++) {
1517		if (fds->fd > fdp->fd_lastfile) {
1518			fds->revents = POLLNVAL;
1519			n++;
1520		} else if (fds->fd < 0) {
1521			fds->revents = 0;
1522		} else {
1523			fp = fdp->fd_ofiles[fds->fd].fde_file;
1524#ifdef CAPABILITIES
1525			if (fp == NULL ||
1526			    cap_check(cap_rights(fdp, fds->fd), &cap_event_rights) != 0)
1527#else
1528			if (fp == NULL)
1529#endif
1530			{
1531				fds->revents = POLLNVAL;
1532				n++;
1533			} else {
1534				/*
1535				 * Note: backend also returns POLLHUP and
1536				 * POLLERR if appropriate.
1537				 */
1538				selfdalloc(td, fds);
1539				fds->revents = fo_poll(fp, fds->events,
1540				    td->td_ucred, td);
1541				/*
1542				 * POSIX requires POLLOUT to be never
1543				 * set simultaneously with POLLHUP.
1544				 */
1545				if ((fds->revents & POLLHUP) != 0)
1546					fds->revents &= ~POLLOUT;
1547
1548				if (fds->revents != 0)
1549					n++;
1550			}
1551		}
1552	}
1553	FILEDESC_SUNLOCK(fdp);
1554	td->td_retval[0] = n;
1555	return (0);
1556}
1557
1558/*
1559 * XXX This was created specifically to support netncp and netsmb.  This
1560 * allows the caller to specify a socket to wait for events on.  It returns
1561 * 0 if any events matched and an error otherwise.  There is no way to
1562 * determine which events fired.
1563 */
1564int
1565selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1566{
1567	struct timeval rtv;
1568	sbintime_t asbt, precision, rsbt;
1569	int error;
1570
1571	precision = 0;	/* stupid gcc! */
1572	if (tvp != NULL) {
1573		rtv = *tvp;
1574		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1575		    rtv.tv_usec >= 1000000)
1576			return (EINVAL);
1577		if (!timevalisset(&rtv))
1578			asbt = 0;
1579		else if (rtv.tv_sec <= INT32_MAX) {
1580			rsbt = tvtosbt(rtv);
1581			precision = rsbt;
1582			precision >>= tc_precexp;
1583			if (TIMESEL(&asbt, rsbt))
1584				asbt += tc_tick_sbt;
1585			if (asbt <= SBT_MAX - rsbt)
1586				asbt += rsbt;
1587			else
1588				asbt = -1;
1589		} else
1590			asbt = -1;
1591	} else
1592		asbt = -1;
1593	seltdinit(td);
1594	/*
1595	 * Iterate until the timeout expires or the socket becomes ready.
1596	 */
1597	for (;;) {
1598		selfdalloc(td, NULL);
1599		error = sopoll(so, events, NULL, td);
1600		/* error here is actually the ready events. */
1601		if (error)
1602			return (0);
1603		error = seltdwait(td, asbt, precision);
1604		if (error)
1605			break;
1606	}
1607	seltdclear(td);
1608	/* XXX Duplicates ncp/smb behavior. */
1609	if (error == ERESTART)
1610		error = 0;
1611	return (error);
1612}
1613
1614/*
1615 * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1616 * have two select sets, one for read and another for write.
1617 */
1618static void
1619selfdalloc(struct thread *td, void *cookie)
1620{
1621	struct seltd *stp;
1622
1623	stp = td->td_sel;
1624	if (stp->st_free1 == NULL)
1625		stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1626	stp->st_free1->sf_td = stp;
1627	stp->st_free1->sf_cookie = cookie;
1628	if (stp->st_free2 == NULL)
1629		stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1630	stp->st_free2->sf_td = stp;
1631	stp->st_free2->sf_cookie = cookie;
1632}
1633
1634static void
1635selfdfree(struct seltd *stp, struct selfd *sfp)
1636{
1637	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1638	if (sfp->sf_si != NULL) {
1639		mtx_lock(sfp->sf_mtx);
1640		if (sfp->sf_si != NULL) {
1641			TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1642			refcount_release(&sfp->sf_refs);
1643		}
1644		mtx_unlock(sfp->sf_mtx);
1645	}
1646	if (refcount_release(&sfp->sf_refs))
1647		uma_zfree(selfd_zone, sfp);
1648}
1649
1650/* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1651void
1652seldrain(struct selinfo *sip)
1653{
1654
1655	/*
1656	 * This feature is already provided by doselwakeup(), thus it is
1657	 * enough to go for it.
1658	 * Eventually, the context, should take care to avoid races
1659	 * between thread calling select()/poll() and file descriptor
1660	 * detaching, but, again, the races are just the same as
1661	 * selwakeup().
1662	 */
1663        doselwakeup(sip, -1);
1664}
1665
1666/*
1667 * Record a select request.
1668 */
1669void
1670selrecord(struct thread *selector, struct selinfo *sip)
1671{
1672	struct selfd *sfp;
1673	struct seltd *stp;
1674	struct mtx *mtxp;
1675
1676	stp = selector->td_sel;
1677	/*
1678	 * Don't record when doing a rescan.
1679	 */
1680	if (stp->st_flags & SELTD_RESCAN)
1681		return;
1682	/*
1683	 * Grab one of the preallocated descriptors.
1684	 */
1685	sfp = NULL;
1686	if ((sfp = stp->st_free1) != NULL)
1687		stp->st_free1 = NULL;
1688	else if ((sfp = stp->st_free2) != NULL)
1689		stp->st_free2 = NULL;
1690	else
1691		panic("selrecord: No free selfd on selq");
1692	mtxp = sip->si_mtx;
1693	if (mtxp == NULL)
1694		mtxp = mtx_pool_find(mtxpool_select, sip);
1695	/*
1696	 * Initialize the sfp and queue it in the thread.
1697	 */
1698	sfp->sf_si = sip;
1699	sfp->sf_mtx = mtxp;
1700	refcount_init(&sfp->sf_refs, 2);
1701	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1702	/*
1703	 * Now that we've locked the sip, check for initialization.
1704	 */
1705	mtx_lock(mtxp);
1706	if (sip->si_mtx == NULL) {
1707		sip->si_mtx = mtxp;
1708		TAILQ_INIT(&sip->si_tdlist);
1709	}
1710	/*
1711	 * Add this thread to the list of selfds listening on this selinfo.
1712	 */
1713	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1714	mtx_unlock(sip->si_mtx);
1715}
1716
1717/* Wake up a selecting thread. */
1718void
1719selwakeup(struct selinfo *sip)
1720{
1721	doselwakeup(sip, -1);
1722}
1723
1724/* Wake up a selecting thread, and set its priority. */
1725void
1726selwakeuppri(struct selinfo *sip, int pri)
1727{
1728	doselwakeup(sip, pri);
1729}
1730
1731/*
1732 * Do a wakeup when a selectable event occurs.
1733 */
1734static void
1735doselwakeup(struct selinfo *sip, int pri)
1736{
1737	struct selfd *sfp;
1738	struct selfd *sfn;
1739	struct seltd *stp;
1740
1741	/* If it's not initialized there can't be any waiters. */
1742	if (sip->si_mtx == NULL)
1743		return;
1744	/*
1745	 * Locking the selinfo locks all selfds associated with it.
1746	 */
1747	mtx_lock(sip->si_mtx);
1748	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1749		/*
1750		 * Once we remove this sfp from the list and clear the
1751		 * sf_si seltdclear will know to ignore this si.
1752		 */
1753		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1754		sfp->sf_si = NULL;
1755		stp = sfp->sf_td;
1756		mtx_lock(&stp->st_mtx);
1757		stp->st_flags |= SELTD_PENDING;
1758		cv_broadcastpri(&stp->st_wait, pri);
1759		mtx_unlock(&stp->st_mtx);
1760		if (refcount_release(&sfp->sf_refs))
1761			uma_zfree(selfd_zone, sfp);
1762	}
1763	mtx_unlock(sip->si_mtx);
1764}
1765
1766static void
1767seltdinit(struct thread *td)
1768{
1769	struct seltd *stp;
1770
1771	if ((stp = td->td_sel) != NULL)
1772		goto out;
1773	td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1774	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1775	cv_init(&stp->st_wait, "select");
1776out:
1777	stp->st_flags = 0;
1778	STAILQ_INIT(&stp->st_selq);
1779}
1780
1781static int
1782seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1783{
1784	struct seltd *stp;
1785	int error;
1786
1787	stp = td->td_sel;
1788	/*
1789	 * An event of interest may occur while we do not hold the seltd
1790	 * locked so check the pending flag before we sleep.
1791	 */
1792	mtx_lock(&stp->st_mtx);
1793	/*
1794	 * Any further calls to selrecord will be a rescan.
1795	 */
1796	stp->st_flags |= SELTD_RESCAN;
1797	if (stp->st_flags & SELTD_PENDING) {
1798		mtx_unlock(&stp->st_mtx);
1799		return (0);
1800	}
1801	if (sbt == 0)
1802		error = EWOULDBLOCK;
1803	else if (sbt != -1)
1804		error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
1805		    sbt, precision, C_ABSOLUTE);
1806	else
1807		error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1808	mtx_unlock(&stp->st_mtx);
1809
1810	return (error);
1811}
1812
1813void
1814seltdfini(struct thread *td)
1815{
1816	struct seltd *stp;
1817
1818	stp = td->td_sel;
1819	if (stp == NULL)
1820		return;
1821	if (stp->st_free1)
1822		uma_zfree(selfd_zone, stp->st_free1);
1823	if (stp->st_free2)
1824		uma_zfree(selfd_zone, stp->st_free2);
1825	td->td_sel = NULL;
1826	cv_destroy(&stp->st_wait);
1827	mtx_destroy(&stp->st_mtx);
1828	free(stp, M_SELECT);
1829}
1830
1831/*
1832 * Remove the references to the thread from all of the objects we were
1833 * polling.
1834 */
1835static void
1836seltdclear(struct thread *td)
1837{
1838	struct seltd *stp;
1839	struct selfd *sfp;
1840	struct selfd *sfn;
1841
1842	stp = td->td_sel;
1843	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1844		selfdfree(stp, sfp);
1845	stp->st_flags = 0;
1846}
1847
1848static void selectinit(void *);
1849SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1850static void
1851selectinit(void *dummy __unused)
1852{
1853
1854	selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1855	    NULL, NULL, UMA_ALIGN_PTR, 0);
1856	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1857}
1858
1859/*
1860 * Set up a syscall return value that follows the convention specified for
1861 * posix_* functions.
1862 */
1863int
1864kern_posix_error(struct thread *td, int error)
1865{
1866
1867	if (error <= 0)
1868		return (error);
1869	td->td_errno = error;
1870	td->td_pflags |= TDP_NERRNO;
1871	td->td_retval[0] = error;
1872	return (0);
1873}
1874