xref: /freebsd/sys/kern/sys_generic.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
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_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/capsicum.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/fcntl.h>
51 #include <sys/file.h>
52 #include <sys/lock.h>
53 #include <sys/proc.h>
54 #include <sys/signalvar.h>
55 #include <sys/socketvar.h>
56 #include <sys/uio.h>
57 #include <sys/kernel.h>
58 #include <sys/ktr.h>
59 #include <sys/limits.h>
60 #include <sys/malloc.h>
61 #include <sys/poll.h>
62 #include <sys/resourcevar.h>
63 #include <sys/selinfo.h>
64 #include <sys/sleepqueue.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/bio.h>
70 #include <sys/buf.h>
71 #include <sys/condvar.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75 
76 #include <security/audit/audit.h>
77 
78 /*
79  * The following macro defines how many bytes will be allocated from
80  * the stack instead of memory allocated when passing the IOCTL data
81  * structures from userspace and to the kernel. Some IOCTLs having
82  * small data structures are used very frequently and this small
83  * buffer on the stack gives a significant speedup improvement for
84  * those requests. The value of this define should be greater or equal
85  * to 64 bytes and should also be power of two. The data structure is
86  * currently hard-aligned to a 8-byte boundary on the stack. This
87  * should currently be sufficient for all supported platforms.
88  */
89 #define	SYS_IOCTL_SMALL_SIZE	128	/* bytes */
90 #define	SYS_IOCTL_SMALL_ALIGN	8	/* bytes */
91 
92 int iosize_max_clamp = 0;
93 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
94     &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
95 int devfs_iosize_max_clamp = 1;
96 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
97     &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
98 
99 /*
100  * Assert that the return value of read(2) and write(2) syscalls fits
101  * into a register.  If not, an architecture will need to provide the
102  * usermode wrappers to reconstruct the result.
103  */
104 CTASSERT(sizeof(register_t) >= sizeof(size_t));
105 
106 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
107 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
108 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
109 
110 static int	pollout(struct thread *, struct pollfd *, struct pollfd *,
111 		    u_int);
112 static int	pollscan(struct thread *, struct pollfd *, u_int);
113 static int	pollrescan(struct thread *);
114 static int	selscan(struct thread *, fd_mask **, fd_mask **, int);
115 static int	selrescan(struct thread *, fd_mask **, fd_mask **);
116 static void	selfdalloc(struct thread *, void *);
117 static void	selfdfree(struct seltd *, struct selfd *);
118 static int	dofileread(struct thread *, int, struct file *, struct uio *,
119 		    off_t, int);
120 static int	dofilewrite(struct thread *, int, struct file *, struct uio *,
121 		    off_t, int);
122 static void	doselwakeup(struct selinfo *, int);
123 static void	seltdinit(struct thread *);
124 static int	seltdwait(struct thread *, sbintime_t, sbintime_t);
125 static void	seltdclear(struct thread *);
126 
127 /*
128  * One seltd per-thread allocated on demand as needed.
129  *
130  *	t - protected by st_mtx
131  * 	k - Only accessed by curthread or read-only
132  */
133 struct seltd {
134 	STAILQ_HEAD(, selfd)	st_selq;	/* (k) List of selfds. */
135 	struct selfd		*st_free1;	/* (k) free fd for read set. */
136 	struct selfd		*st_free2;	/* (k) free fd for write set. */
137 	struct mtx		st_mtx;		/* Protects struct seltd */
138 	struct cv		st_wait;	/* (t) Wait channel. */
139 	int			st_flags;	/* (t) SELTD_ flags. */
140 };
141 
142 #define	SELTD_PENDING	0x0001			/* We have pending events. */
143 #define	SELTD_RESCAN	0x0002			/* Doing a rescan. */
144 
145 /*
146  * One selfd allocated per-thread per-file-descriptor.
147  *	f - protected by sf_mtx
148  */
149 struct selfd {
150 	STAILQ_ENTRY(selfd)	sf_link;	/* (k) fds owned by this td. */
151 	TAILQ_ENTRY(selfd)	sf_threads;	/* (f) fds on this selinfo. */
152 	struct selinfo		*sf_si;		/* (f) selinfo when linked. */
153 	struct mtx		*sf_mtx;	/* Pointer to selinfo mtx. */
154 	struct seltd		*sf_td;		/* (k) owning seltd. */
155 	void			*sf_cookie;	/* (k) fd or pollfd. */
156 };
157 
158 static uma_zone_t selfd_zone;
159 static struct mtx_pool *mtxpool_select;
160 
161 #ifndef _SYS_SYSPROTO_H_
162 struct read_args {
163 	int	fd;
164 	void	*buf;
165 	size_t	nbyte;
166 };
167 #endif
168 int
169 sys_read(td, uap)
170 	struct thread *td;
171 	struct read_args *uap;
172 {
173 	struct uio auio;
174 	struct iovec aiov;
175 	int error;
176 
177 	if (uap->nbyte > IOSIZE_MAX)
178 		return (EINVAL);
179 	aiov.iov_base = uap->buf;
180 	aiov.iov_len = uap->nbyte;
181 	auio.uio_iov = &aiov;
182 	auio.uio_iovcnt = 1;
183 	auio.uio_resid = uap->nbyte;
184 	auio.uio_segflg = UIO_USERSPACE;
185 	error = kern_readv(td, uap->fd, &auio);
186 	return(error);
187 }
188 
189 /*
190  * Positioned read system call
191  */
192 #ifndef _SYS_SYSPROTO_H_
193 struct pread_args {
194 	int	fd;
195 	void	*buf;
196 	size_t	nbyte;
197 	int	pad;
198 	off_t	offset;
199 };
200 #endif
201 int
202 sys_pread(td, uap)
203 	struct thread *td;
204 	struct pread_args *uap;
205 {
206 	struct uio auio;
207 	struct iovec aiov;
208 	int error;
209 
210 	if (uap->nbyte > IOSIZE_MAX)
211 		return (EINVAL);
212 	aiov.iov_base = uap->buf;
213 	aiov.iov_len = uap->nbyte;
214 	auio.uio_iov = &aiov;
215 	auio.uio_iovcnt = 1;
216 	auio.uio_resid = uap->nbyte;
217 	auio.uio_segflg = UIO_USERSPACE;
218 	error = kern_preadv(td, uap->fd, &auio, uap->offset);
219 	return(error);
220 }
221 
222 int
223 freebsd6_pread(td, uap)
224 	struct thread *td;
225 	struct freebsd6_pread_args *uap;
226 {
227 	struct pread_args oargs;
228 
229 	oargs.fd = uap->fd;
230 	oargs.buf = uap->buf;
231 	oargs.nbyte = uap->nbyte;
232 	oargs.offset = uap->offset;
233 	return (sys_pread(td, &oargs));
234 }
235 
236 /*
237  * Scatter read system call.
238  */
239 #ifndef _SYS_SYSPROTO_H_
240 struct readv_args {
241 	int	fd;
242 	struct	iovec *iovp;
243 	u_int	iovcnt;
244 };
245 #endif
246 int
247 sys_readv(struct thread *td, struct readv_args *uap)
248 {
249 	struct uio *auio;
250 	int error;
251 
252 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
253 	if (error)
254 		return (error);
255 	error = kern_readv(td, uap->fd, auio);
256 	free(auio, M_IOV);
257 	return (error);
258 }
259 
260 int
261 kern_readv(struct thread *td, int fd, struct uio *auio)
262 {
263 	struct file *fp;
264 	cap_rights_t rights;
265 	int error;
266 
267 	error = fget_read(td, fd, cap_rights_init(&rights, CAP_READ), &fp);
268 	if (error)
269 		return (error);
270 	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
271 	fdrop(fp, td);
272 	return (error);
273 }
274 
275 /*
276  * Scatter positioned read system call.
277  */
278 #ifndef _SYS_SYSPROTO_H_
279 struct preadv_args {
280 	int	fd;
281 	struct	iovec *iovp;
282 	u_int	iovcnt;
283 	off_t	offset;
284 };
285 #endif
286 int
287 sys_preadv(struct thread *td, struct preadv_args *uap)
288 {
289 	struct uio *auio;
290 	int error;
291 
292 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
293 	if (error)
294 		return (error);
295 	error = kern_preadv(td, uap->fd, auio, uap->offset);
296 	free(auio, M_IOV);
297 	return (error);
298 }
299 
300 int
301 kern_preadv(td, fd, auio, offset)
302 	struct thread *td;
303 	int fd;
304 	struct uio *auio;
305 	off_t offset;
306 {
307 	struct file *fp;
308 	cap_rights_t rights;
309 	int error;
310 
311 	error = fget_read(td, fd, cap_rights_init(&rights, CAP_PREAD), &fp);
312 	if (error)
313 		return (error);
314 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
315 		error = ESPIPE;
316 	else if (offset < 0 && fp->f_vnode->v_type != VCHR)
317 		error = EINVAL;
318 	else
319 		error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
320 	fdrop(fp, td);
321 	return (error);
322 }
323 
324 /*
325  * Common code for readv and preadv that reads data in
326  * from a file using the passed in uio, offset, and flags.
327  */
328 static int
329 dofileread(td, fd, fp, auio, offset, flags)
330 	struct thread *td;
331 	int fd;
332 	struct file *fp;
333 	struct uio *auio;
334 	off_t offset;
335 	int flags;
336 {
337 	ssize_t cnt;
338 	int error;
339 #ifdef KTRACE
340 	struct uio *ktruio = NULL;
341 #endif
342 
343 	/* Finish zero length reads right here */
344 	if (auio->uio_resid == 0) {
345 		td->td_retval[0] = 0;
346 		return(0);
347 	}
348 	auio->uio_rw = UIO_READ;
349 	auio->uio_offset = offset;
350 	auio->uio_td = td;
351 #ifdef KTRACE
352 	if (KTRPOINT(td, KTR_GENIO))
353 		ktruio = cloneuio(auio);
354 #endif
355 	cnt = auio->uio_resid;
356 	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
357 		if (auio->uio_resid != cnt && (error == ERESTART ||
358 		    error == EINTR || error == EWOULDBLOCK))
359 			error = 0;
360 	}
361 	cnt -= auio->uio_resid;
362 #ifdef KTRACE
363 	if (ktruio != NULL) {
364 		ktruio->uio_resid = cnt;
365 		ktrgenio(fd, UIO_READ, ktruio, error);
366 	}
367 #endif
368 	td->td_retval[0] = cnt;
369 	return (error);
370 }
371 
372 #ifndef _SYS_SYSPROTO_H_
373 struct write_args {
374 	int	fd;
375 	const void *buf;
376 	size_t	nbyte;
377 };
378 #endif
379 int
380 sys_write(td, uap)
381 	struct thread *td;
382 	struct write_args *uap;
383 {
384 	struct uio auio;
385 	struct iovec aiov;
386 	int error;
387 
388 	if (uap->nbyte > IOSIZE_MAX)
389 		return (EINVAL);
390 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
391 	aiov.iov_len = uap->nbyte;
392 	auio.uio_iov = &aiov;
393 	auio.uio_iovcnt = 1;
394 	auio.uio_resid = uap->nbyte;
395 	auio.uio_segflg = UIO_USERSPACE;
396 	error = kern_writev(td, uap->fd, &auio);
397 	return(error);
398 }
399 
400 /*
401  * Positioned write system call.
402  */
403 #ifndef _SYS_SYSPROTO_H_
404 struct pwrite_args {
405 	int	fd;
406 	const void *buf;
407 	size_t	nbyte;
408 	int	pad;
409 	off_t	offset;
410 };
411 #endif
412 int
413 sys_pwrite(td, uap)
414 	struct thread *td;
415 	struct pwrite_args *uap;
416 {
417 	struct uio auio;
418 	struct iovec aiov;
419 	int error;
420 
421 	if (uap->nbyte > IOSIZE_MAX)
422 		return (EINVAL);
423 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
424 	aiov.iov_len = uap->nbyte;
425 	auio.uio_iov = &aiov;
426 	auio.uio_iovcnt = 1;
427 	auio.uio_resid = uap->nbyte;
428 	auio.uio_segflg = UIO_USERSPACE;
429 	error = kern_pwritev(td, uap->fd, &auio, uap->offset);
430 	return(error);
431 }
432 
433 int
434 freebsd6_pwrite(td, uap)
435 	struct thread *td;
436 	struct freebsd6_pwrite_args *uap;
437 {
438 	struct pwrite_args oargs;
439 
440 	oargs.fd = uap->fd;
441 	oargs.buf = uap->buf;
442 	oargs.nbyte = uap->nbyte;
443 	oargs.offset = uap->offset;
444 	return (sys_pwrite(td, &oargs));
445 }
446 
447 /*
448  * Gather write system call.
449  */
450 #ifndef _SYS_SYSPROTO_H_
451 struct writev_args {
452 	int	fd;
453 	struct	iovec *iovp;
454 	u_int	iovcnt;
455 };
456 #endif
457 int
458 sys_writev(struct thread *td, struct writev_args *uap)
459 {
460 	struct uio *auio;
461 	int error;
462 
463 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
464 	if (error)
465 		return (error);
466 	error = kern_writev(td, uap->fd, auio);
467 	free(auio, M_IOV);
468 	return (error);
469 }
470 
471 int
472 kern_writev(struct thread *td, int fd, struct uio *auio)
473 {
474 	struct file *fp;
475 	cap_rights_t rights;
476 	int error;
477 
478 	error = fget_write(td, fd, cap_rights_init(&rights, CAP_WRITE), &fp);
479 	if (error)
480 		return (error);
481 	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
482 	fdrop(fp, td);
483 	return (error);
484 }
485 
486 /*
487  * Gather positioned write system call.
488  */
489 #ifndef _SYS_SYSPROTO_H_
490 struct pwritev_args {
491 	int	fd;
492 	struct	iovec *iovp;
493 	u_int	iovcnt;
494 	off_t	offset;
495 };
496 #endif
497 int
498 sys_pwritev(struct thread *td, struct pwritev_args *uap)
499 {
500 	struct uio *auio;
501 	int error;
502 
503 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
504 	if (error)
505 		return (error);
506 	error = kern_pwritev(td, uap->fd, auio, uap->offset);
507 	free(auio, M_IOV);
508 	return (error);
509 }
510 
511 int
512 kern_pwritev(td, fd, auio, offset)
513 	struct thread *td;
514 	struct uio *auio;
515 	int fd;
516 	off_t offset;
517 {
518 	struct file *fp;
519 	cap_rights_t rights;
520 	int error;
521 
522 	error = fget_write(td, fd, cap_rights_init(&rights, CAP_PWRITE), &fp);
523 	if (error)
524 		return (error);
525 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
526 		error = ESPIPE;
527 	else if (offset < 0 && fp->f_vnode->v_type != VCHR)
528 		error = EINVAL;
529 	else
530 		error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
531 	fdrop(fp, td);
532 	return (error);
533 }
534 
535 /*
536  * Common code for writev and pwritev that writes data to
537  * a file using the passed in uio, offset, and flags.
538  */
539 static int
540 dofilewrite(td, fd, fp, auio, offset, flags)
541 	struct thread *td;
542 	int fd;
543 	struct file *fp;
544 	struct uio *auio;
545 	off_t offset;
546 	int flags;
547 {
548 	ssize_t cnt;
549 	int error;
550 #ifdef KTRACE
551 	struct uio *ktruio = NULL;
552 #endif
553 
554 	auio->uio_rw = UIO_WRITE;
555 	auio->uio_td = td;
556 	auio->uio_offset = offset;
557 #ifdef KTRACE
558 	if (KTRPOINT(td, KTR_GENIO))
559 		ktruio = cloneuio(auio);
560 #endif
561 	cnt = auio->uio_resid;
562 	if (fp->f_type == DTYPE_VNODE &&
563 	    (fp->f_vnread_flags & FDEVFS_VNODE) == 0)
564 		bwillwrite();
565 	if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
566 		if (auio->uio_resid != cnt && (error == ERESTART ||
567 		    error == EINTR || error == EWOULDBLOCK))
568 			error = 0;
569 		/* Socket layer is responsible for issuing SIGPIPE. */
570 		if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
571 			PROC_LOCK(td->td_proc);
572 			tdsignal(td, SIGPIPE);
573 			PROC_UNLOCK(td->td_proc);
574 		}
575 	}
576 	cnt -= auio->uio_resid;
577 #ifdef KTRACE
578 	if (ktruio != NULL) {
579 		ktruio->uio_resid = cnt;
580 		ktrgenio(fd, UIO_WRITE, ktruio, error);
581 	}
582 #endif
583 	td->td_retval[0] = cnt;
584 	return (error);
585 }
586 
587 /*
588  * Truncate a file given a file descriptor.
589  *
590  * Can't use fget_write() here, since must return EINVAL and not EBADF if the
591  * descriptor isn't writable.
592  */
593 int
594 kern_ftruncate(td, fd, length)
595 	struct thread *td;
596 	int fd;
597 	off_t length;
598 {
599 	struct file *fp;
600 	cap_rights_t rights;
601 	int error;
602 
603 	AUDIT_ARG_FD(fd);
604 	if (length < 0)
605 		return (EINVAL);
606 	error = fget(td, fd, cap_rights_init(&rights, CAP_FTRUNCATE), &fp);
607 	if (error)
608 		return (error);
609 	AUDIT_ARG_FILE(td->td_proc, fp);
610 	if (!(fp->f_flag & FWRITE)) {
611 		fdrop(fp, td);
612 		return (EINVAL);
613 	}
614 	error = fo_truncate(fp, length, td->td_ucred, td);
615 	fdrop(fp, td);
616 	return (error);
617 }
618 
619 #ifndef _SYS_SYSPROTO_H_
620 struct ftruncate_args {
621 	int	fd;
622 	int	pad;
623 	off_t	length;
624 };
625 #endif
626 int
627 sys_ftruncate(td, uap)
628 	struct thread *td;
629 	struct ftruncate_args *uap;
630 {
631 
632 	return (kern_ftruncate(td, uap->fd, uap->length));
633 }
634 
635 #if defined(COMPAT_43)
636 #ifndef _SYS_SYSPROTO_H_
637 struct oftruncate_args {
638 	int	fd;
639 	long	length;
640 };
641 #endif
642 int
643 oftruncate(td, uap)
644 	struct thread *td;
645 	struct oftruncate_args *uap;
646 {
647 
648 	return (kern_ftruncate(td, uap->fd, uap->length));
649 }
650 #endif /* COMPAT_43 */
651 
652 #ifndef _SYS_SYSPROTO_H_
653 struct ioctl_args {
654 	int	fd;
655 	u_long	com;
656 	caddr_t	data;
657 };
658 #endif
659 /* ARGSUSED */
660 int
661 sys_ioctl(struct thread *td, struct ioctl_args *uap)
662 {
663 	u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
664 	u_long com;
665 	int arg, error;
666 	u_int size;
667 	caddr_t data;
668 
669 	if (uap->com > 0xffffffff) {
670 		printf(
671 		    "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
672 		    td->td_proc->p_pid, td->td_name, uap->com);
673 		uap->com &= 0xffffffff;
674 	}
675 	com = uap->com;
676 
677 	/*
678 	 * Interpret high order word to find amount of data to be
679 	 * copied to/from the user's address space.
680 	 */
681 	size = IOCPARM_LEN(com);
682 	if ((size > IOCPARM_MAX) ||
683 	    ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
684 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
685 	    ((com & IOC_OUT) && size == 0) ||
686 #else
687 	    ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
688 #endif
689 	    ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
690 		return (ENOTTY);
691 
692 	if (size > 0) {
693 		if (com & IOC_VOID) {
694 			/* Integer argument. */
695 			arg = (intptr_t)uap->data;
696 			data = (void *)&arg;
697 			size = 0;
698 		} else {
699 			if (size > SYS_IOCTL_SMALL_SIZE)
700 				data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
701 			else
702 				data = smalldata;
703 		}
704 	} else
705 		data = (void *)&uap->data;
706 	if (com & IOC_IN) {
707 		error = copyin(uap->data, data, (u_int)size);
708 		if (error != 0)
709 			goto out;
710 	} else if (com & IOC_OUT) {
711 		/*
712 		 * Zero the buffer so the user always
713 		 * gets back something deterministic.
714 		 */
715 		bzero(data, size);
716 	}
717 
718 	error = kern_ioctl(td, uap->fd, com, data);
719 
720 	if (error == 0 && (com & IOC_OUT))
721 		error = copyout(data, uap->data, (u_int)size);
722 
723 out:
724 	if (size > SYS_IOCTL_SMALL_SIZE)
725 		free(data, M_IOCTLOPS);
726 	return (error);
727 }
728 
729 int
730 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
731 {
732 	struct file *fp;
733 	struct filedesc *fdp;
734 #ifndef CAPABILITIES
735 	cap_rights_t rights;
736 #endif
737 	int error, tmp, locked;
738 
739 	AUDIT_ARG_FD(fd);
740 	AUDIT_ARG_CMD(com);
741 
742 	fdp = td->td_proc->p_fd;
743 
744 	switch (com) {
745 	case FIONCLEX:
746 	case FIOCLEX:
747 		FILEDESC_XLOCK(fdp);
748 		locked = LA_XLOCKED;
749 		break;
750 	default:
751 #ifdef CAPABILITIES
752 		FILEDESC_SLOCK(fdp);
753 		locked = LA_SLOCKED;
754 #else
755 		locked = LA_UNLOCKED;
756 #endif
757 		break;
758 	}
759 
760 #ifdef CAPABILITIES
761 	if ((fp = fget_locked(fdp, fd)) == NULL) {
762 		error = EBADF;
763 		goto out;
764 	}
765 	if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
766 		fp = NULL;	/* fhold() was not called yet */
767 		goto out;
768 	}
769 	fhold(fp);
770 	if (locked == LA_SLOCKED) {
771 		FILEDESC_SUNLOCK(fdp);
772 		locked = LA_UNLOCKED;
773 	}
774 #else
775 	error = fget(td, fd, cap_rights_init(&rights, CAP_IOCTL), &fp);
776 	if (error != 0) {
777 		fp = NULL;
778 		goto out;
779 	}
780 #endif
781 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
782 		error = EBADF;
783 		goto out;
784 	}
785 
786 	switch (com) {
787 	case FIONCLEX:
788 		fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
789 		goto out;
790 	case FIOCLEX:
791 		fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
792 		goto out;
793 	case FIONBIO:
794 		if ((tmp = *(int *)data))
795 			atomic_set_int(&fp->f_flag, FNONBLOCK);
796 		else
797 			atomic_clear_int(&fp->f_flag, FNONBLOCK);
798 		data = (void *)&tmp;
799 		break;
800 	case FIOASYNC:
801 		if ((tmp = *(int *)data))
802 			atomic_set_int(&fp->f_flag, FASYNC);
803 		else
804 			atomic_clear_int(&fp->f_flag, FASYNC);
805 		data = (void *)&tmp;
806 		break;
807 	}
808 
809 	error = fo_ioctl(fp, com, data, td->td_ucred, td);
810 out:
811 	switch (locked) {
812 	case LA_XLOCKED:
813 		FILEDESC_XUNLOCK(fdp);
814 		break;
815 #ifdef CAPABILITIES
816 	case LA_SLOCKED:
817 		FILEDESC_SUNLOCK(fdp);
818 		break;
819 #endif
820 	default:
821 		FILEDESC_UNLOCK_ASSERT(fdp);
822 		break;
823 	}
824 	if (fp != NULL)
825 		fdrop(fp, td);
826 	return (error);
827 }
828 
829 int
830 poll_no_poll(int events)
831 {
832 	/*
833 	 * Return true for read/write.  If the user asked for something
834 	 * special, return POLLNVAL, so that clients have a way of
835 	 * determining reliably whether or not the extended
836 	 * functionality is present without hard-coding knowledge
837 	 * of specific filesystem implementations.
838 	 */
839 	if (events & ~POLLSTANDARD)
840 		return (POLLNVAL);
841 
842 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
843 }
844 
845 int
846 sys_pselect(struct thread *td, struct pselect_args *uap)
847 {
848 	struct timespec ts;
849 	struct timeval tv, *tvp;
850 	sigset_t set, *uset;
851 	int error;
852 
853 	if (uap->ts != NULL) {
854 		error = copyin(uap->ts, &ts, sizeof(ts));
855 		if (error != 0)
856 		    return (error);
857 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
858 		tvp = &tv;
859 	} else
860 		tvp = NULL;
861 	if (uap->sm != NULL) {
862 		error = copyin(uap->sm, &set, sizeof(set));
863 		if (error != 0)
864 			return (error);
865 		uset = &set;
866 	} else
867 		uset = NULL;
868 	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
869 	    uset, NFDBITS));
870 }
871 
872 int
873 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
874     struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
875 {
876 	int error;
877 
878 	if (uset != NULL) {
879 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
880 		    &td->td_oldsigmask, 0);
881 		if (error != 0)
882 			return (error);
883 		td->td_pflags |= TDP_OLDMASK;
884 		/*
885 		 * Make sure that ast() is called on return to
886 		 * usermode and TDP_OLDMASK is cleared, restoring old
887 		 * sigmask.
888 		 */
889 		thread_lock(td);
890 		td->td_flags |= TDF_ASTPENDING;
891 		thread_unlock(td);
892 	}
893 	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
894 	return (error);
895 }
896 
897 #ifndef _SYS_SYSPROTO_H_
898 struct select_args {
899 	int	nd;
900 	fd_set	*in, *ou, *ex;
901 	struct	timeval *tv;
902 };
903 #endif
904 int
905 sys_select(struct thread *td, struct select_args *uap)
906 {
907 	struct timeval tv, *tvp;
908 	int error;
909 
910 	if (uap->tv != NULL) {
911 		error = copyin(uap->tv, &tv, sizeof(tv));
912 		if (error)
913 			return (error);
914 		tvp = &tv;
915 	} else
916 		tvp = NULL;
917 
918 	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
919 	    NFDBITS));
920 }
921 
922 /*
923  * In the unlikely case when user specified n greater then the last
924  * open file descriptor, check that no bits are set after the last
925  * valid fd.  We must return EBADF if any is set.
926  *
927  * There are applications that rely on the behaviour.
928  *
929  * nd is fd_lastfile + 1.
930  */
931 static int
932 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
933 {
934 	char *addr, *oaddr;
935 	int b, i, res;
936 	uint8_t bits;
937 
938 	if (nd >= ndu || fd_in == NULL)
939 		return (0);
940 
941 	oaddr = NULL;
942 	bits = 0; /* silence gcc */
943 	for (i = nd; i < ndu; i++) {
944 		b = i / NBBY;
945 #if BYTE_ORDER == LITTLE_ENDIAN
946 		addr = (char *)fd_in + b;
947 #else
948 		addr = (char *)fd_in;
949 		if (abi_nfdbits == NFDBITS) {
950 			addr += rounddown(b, sizeof(fd_mask)) +
951 			    sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
952 		} else {
953 			addr += rounddown(b, sizeof(uint32_t)) +
954 			    sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
955 		}
956 #endif
957 		if (addr != oaddr) {
958 			res = fubyte(addr);
959 			if (res == -1)
960 				return (EFAULT);
961 			oaddr = addr;
962 			bits = res;
963 		}
964 		if ((bits & (1 << (i % NBBY))) != 0)
965 			return (EBADF);
966 	}
967 	return (0);
968 }
969 
970 int
971 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
972     fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
973 {
974 	struct filedesc *fdp;
975 	/*
976 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
977 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
978 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
979 	 * of 256.
980 	 */
981 	fd_mask s_selbits[howmany(2048, NFDBITS)];
982 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
983 	struct timeval rtv;
984 	sbintime_t asbt, precision, rsbt;
985 	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
986 	int error, lf, ndu;
987 
988 	if (nd < 0)
989 		return (EINVAL);
990 	fdp = td->td_proc->p_fd;
991 	ndu = nd;
992 	lf = fdp->fd_lastfile;
993 	if (nd > lf + 1)
994 		nd = lf + 1;
995 
996 	error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
997 	if (error != 0)
998 		return (error);
999 	error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1000 	if (error != 0)
1001 		return (error);
1002 	error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1003 	if (error != 0)
1004 		return (error);
1005 
1006 	/*
1007 	 * Allocate just enough bits for the non-null fd_sets.  Use the
1008 	 * preallocated auto buffer if possible.
1009 	 */
1010 	nfdbits = roundup(nd, NFDBITS);
1011 	ncpbytes = nfdbits / NBBY;
1012 	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1013 	nbufbytes = 0;
1014 	if (fd_in != NULL)
1015 		nbufbytes += 2 * ncpbytes;
1016 	if (fd_ou != NULL)
1017 		nbufbytes += 2 * ncpbytes;
1018 	if (fd_ex != NULL)
1019 		nbufbytes += 2 * ncpbytes;
1020 	if (nbufbytes <= sizeof s_selbits)
1021 		selbits = &s_selbits[0];
1022 	else
1023 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1024 
1025 	/*
1026 	 * Assign pointers into the bit buffers and fetch the input bits.
1027 	 * Put the output buffers together so that they can be bzeroed
1028 	 * together.
1029 	 */
1030 	sbp = selbits;
1031 #define	getbits(name, x) \
1032 	do {								\
1033 		if (name == NULL) {					\
1034 			ibits[x] = NULL;				\
1035 			obits[x] = NULL;				\
1036 		} else {						\
1037 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
1038 			obits[x] = sbp;					\
1039 			sbp += ncpbytes / sizeof *sbp;			\
1040 			error = copyin(name, ibits[x], ncpubytes);	\
1041 			if (error != 0)					\
1042 				goto done;				\
1043 			bzero((char *)ibits[x] + ncpubytes,		\
1044 			    ncpbytes - ncpubytes);			\
1045 		}							\
1046 	} while (0)
1047 	getbits(fd_in, 0);
1048 	getbits(fd_ou, 1);
1049 	getbits(fd_ex, 2);
1050 #undef	getbits
1051 
1052 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1053 	/*
1054 	 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1055 	 * we are running under 32-bit emulation. This should be more
1056 	 * generic.
1057 	 */
1058 #define swizzle_fdset(bits)						\
1059 	if (abi_nfdbits != NFDBITS && bits != NULL) {			\
1060 		int i;							\
1061 		for (i = 0; i < ncpbytes / sizeof *sbp; i++)		\
1062 			bits[i] = (bits[i] >> 32) | (bits[i] << 32);	\
1063 	}
1064 #else
1065 #define swizzle_fdset(bits)
1066 #endif
1067 
1068 	/* Make sure the bit order makes it through an ABI transition */
1069 	swizzle_fdset(ibits[0]);
1070 	swizzle_fdset(ibits[1]);
1071 	swizzle_fdset(ibits[2]);
1072 
1073 	if (nbufbytes != 0)
1074 		bzero(selbits, nbufbytes / 2);
1075 
1076 	precision = 0;
1077 	if (tvp != NULL) {
1078 		rtv = *tvp;
1079 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1080 		    rtv.tv_usec >= 1000000) {
1081 			error = EINVAL;
1082 			goto done;
1083 		}
1084 		if (!timevalisset(&rtv))
1085 			asbt = 0;
1086 		else if (rtv.tv_sec <= INT32_MAX) {
1087 			rsbt = tvtosbt(rtv);
1088 			precision = rsbt;
1089 			precision >>= tc_precexp;
1090 			if (TIMESEL(&asbt, rsbt))
1091 				asbt += tc_tick_sbt;
1092 			if (asbt <= SBT_MAX - rsbt)
1093 				asbt += rsbt;
1094 			else
1095 				asbt = -1;
1096 		} else
1097 			asbt = -1;
1098 	} else
1099 		asbt = -1;
1100 	seltdinit(td);
1101 	/* Iterate until the timeout expires or descriptors become ready. */
1102 	for (;;) {
1103 		error = selscan(td, ibits, obits, nd);
1104 		if (error || td->td_retval[0] != 0)
1105 			break;
1106 		error = seltdwait(td, asbt, precision);
1107 		if (error)
1108 			break;
1109 		error = selrescan(td, ibits, obits);
1110 		if (error || td->td_retval[0] != 0)
1111 			break;
1112 	}
1113 	seltdclear(td);
1114 
1115 done:
1116 	/* select is not restarted after signals... */
1117 	if (error == ERESTART)
1118 		error = EINTR;
1119 	if (error == EWOULDBLOCK)
1120 		error = 0;
1121 
1122 	/* swizzle bit order back, if necessary */
1123 	swizzle_fdset(obits[0]);
1124 	swizzle_fdset(obits[1]);
1125 	swizzle_fdset(obits[2]);
1126 #undef swizzle_fdset
1127 
1128 #define	putbits(name, x) \
1129 	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1130 		error = error2;
1131 	if (error == 0) {
1132 		int error2;
1133 
1134 		putbits(fd_in, 0);
1135 		putbits(fd_ou, 1);
1136 		putbits(fd_ex, 2);
1137 #undef putbits
1138 	}
1139 	if (selbits != &s_selbits[0])
1140 		free(selbits, M_SELECT);
1141 
1142 	return (error);
1143 }
1144 /*
1145  * Convert a select bit set to poll flags.
1146  *
1147  * The backend always returns POLLHUP/POLLERR if appropriate and we
1148  * return this as a set bit in any set.
1149  */
1150 static int select_flags[3] = {
1151     POLLRDNORM | POLLHUP | POLLERR,
1152     POLLWRNORM | POLLHUP | POLLERR,
1153     POLLRDBAND | POLLERR
1154 };
1155 
1156 /*
1157  * Compute the fo_poll flags required for a fd given by the index and
1158  * bit position in the fd_mask array.
1159  */
1160 static __inline int
1161 selflags(fd_mask **ibits, int idx, fd_mask bit)
1162 {
1163 	int flags;
1164 	int msk;
1165 
1166 	flags = 0;
1167 	for (msk = 0; msk < 3; msk++) {
1168 		if (ibits[msk] == NULL)
1169 			continue;
1170 		if ((ibits[msk][idx] & bit) == 0)
1171 			continue;
1172 		flags |= select_flags[msk];
1173 	}
1174 	return (flags);
1175 }
1176 
1177 /*
1178  * Set the appropriate output bits given a mask of fired events and the
1179  * input bits originally requested.
1180  */
1181 static __inline int
1182 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1183 {
1184 	int msk;
1185 	int n;
1186 
1187 	n = 0;
1188 	for (msk = 0; msk < 3; msk++) {
1189 		if ((events & select_flags[msk]) == 0)
1190 			continue;
1191 		if (ibits[msk] == NULL)
1192 			continue;
1193 		if ((ibits[msk][idx] & bit) == 0)
1194 			continue;
1195 		/*
1196 		 * XXX Check for a duplicate set.  This can occur because a
1197 		 * socket calls selrecord() twice for each poll() call
1198 		 * resulting in two selfds per real fd.  selrescan() will
1199 		 * call selsetbits twice as a result.
1200 		 */
1201 		if ((obits[msk][idx] & bit) != 0)
1202 			continue;
1203 		obits[msk][idx] |= bit;
1204 		n++;
1205 	}
1206 
1207 	return (n);
1208 }
1209 
1210 static __inline int
1211 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1212 {
1213 	cap_rights_t rights;
1214 
1215 	cap_rights_init(&rights, CAP_EVENT);
1216 
1217 	return (fget_unlocked(fdp, fd, &rights, 0, fpp, NULL));
1218 }
1219 
1220 /*
1221  * Traverse the list of fds attached to this thread's seltd and check for
1222  * completion.
1223  */
1224 static int
1225 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1226 {
1227 	struct filedesc *fdp;
1228 	struct selinfo *si;
1229 	struct seltd *stp;
1230 	struct selfd *sfp;
1231 	struct selfd *sfn;
1232 	struct file *fp;
1233 	fd_mask bit;
1234 	int fd, ev, n, idx;
1235 	int error;
1236 
1237 	fdp = td->td_proc->p_fd;
1238 	stp = td->td_sel;
1239 	n = 0;
1240 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1241 		fd = (int)(uintptr_t)sfp->sf_cookie;
1242 		si = sfp->sf_si;
1243 		selfdfree(stp, sfp);
1244 		/* If the selinfo wasn't cleared the event didn't fire. */
1245 		if (si != NULL)
1246 			continue;
1247 		error = getselfd_cap(fdp, fd, &fp);
1248 		if (error)
1249 			return (error);
1250 		idx = fd / NFDBITS;
1251 		bit = (fd_mask)1 << (fd % NFDBITS);
1252 		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1253 		fdrop(fp, td);
1254 		if (ev != 0)
1255 			n += selsetbits(ibits, obits, idx, bit, ev);
1256 	}
1257 	stp->st_flags = 0;
1258 	td->td_retval[0] = n;
1259 	return (0);
1260 }
1261 
1262 /*
1263  * Perform the initial filedescriptor scan and register ourselves with
1264  * each selinfo.
1265  */
1266 static int
1267 selscan(td, ibits, obits, nfd)
1268 	struct thread *td;
1269 	fd_mask **ibits, **obits;
1270 	int nfd;
1271 {
1272 	struct filedesc *fdp;
1273 	struct file *fp;
1274 	fd_mask bit;
1275 	int ev, flags, end, fd;
1276 	int n, idx;
1277 	int error;
1278 
1279 	fdp = td->td_proc->p_fd;
1280 	n = 0;
1281 	for (idx = 0, fd = 0; fd < nfd; idx++) {
1282 		end = imin(fd + NFDBITS, nfd);
1283 		for (bit = 1; fd < end; bit <<= 1, fd++) {
1284 			/* Compute the list of events we're interested in. */
1285 			flags = selflags(ibits, idx, bit);
1286 			if (flags == 0)
1287 				continue;
1288 			error = getselfd_cap(fdp, fd, &fp);
1289 			if (error)
1290 				return (error);
1291 			selfdalloc(td, (void *)(uintptr_t)fd);
1292 			ev = fo_poll(fp, flags, td->td_ucred, td);
1293 			fdrop(fp, td);
1294 			if (ev != 0)
1295 				n += selsetbits(ibits, obits, idx, bit, ev);
1296 		}
1297 	}
1298 
1299 	td->td_retval[0] = n;
1300 	return (0);
1301 }
1302 
1303 int
1304 sys_poll(struct thread *td, struct poll_args *uap)
1305 {
1306 	struct timespec ts, *tsp;
1307 
1308 	if (uap->timeout != INFTIM) {
1309 		if (uap->timeout < 0)
1310 			return (EINVAL);
1311 		ts.tv_sec = uap->timeout / 1000;
1312 		ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1313 		tsp = &ts;
1314 	} else
1315 		tsp = NULL;
1316 
1317 	return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1318 }
1319 
1320 int
1321 kern_poll(struct thread *td, struct pollfd *fds, u_int nfds,
1322     struct timespec *tsp, sigset_t *uset)
1323 {
1324 	struct pollfd *bits;
1325 	struct pollfd smallbits[32];
1326 	sbintime_t sbt, precision, tmp;
1327 	time_t over;
1328 	struct timespec ts;
1329 	int error;
1330 	size_t ni;
1331 
1332 	precision = 0;
1333 	if (tsp != NULL) {
1334 		if (tsp->tv_sec < 0)
1335 			return (EINVAL);
1336 		if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000)
1337 			return (EINVAL);
1338 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1339 			sbt = 0;
1340 		else {
1341 			ts = *tsp;
1342 			if (ts.tv_sec > INT32_MAX / 2) {
1343 				over = ts.tv_sec - INT32_MAX / 2;
1344 				ts.tv_sec -= over;
1345 			} else
1346 				over = 0;
1347 			tmp = tstosbt(ts);
1348 			precision = tmp;
1349 			precision >>= tc_precexp;
1350 			if (TIMESEL(&sbt, tmp))
1351 				sbt += tc_tick_sbt;
1352 			sbt += tmp;
1353 		}
1354 	} else
1355 		sbt = -1;
1356 
1357 	if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1358 		return (EINVAL);
1359 	ni = nfds * sizeof(struct pollfd);
1360 	if (ni > sizeof(smallbits))
1361 		bits = malloc(ni, M_TEMP, M_WAITOK);
1362 	else
1363 		bits = smallbits;
1364 	error = copyin(fds, bits, ni);
1365 	if (error)
1366 		goto done;
1367 
1368 	if (uset != NULL) {
1369 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
1370 		    &td->td_oldsigmask, 0);
1371 		if (error)
1372 			goto done;
1373 		td->td_pflags |= TDP_OLDMASK;
1374 		/*
1375 		 * Make sure that ast() is called on return to
1376 		 * usermode and TDP_OLDMASK is cleared, restoring old
1377 		 * sigmask.
1378 		 */
1379 		thread_lock(td);
1380 		td->td_flags |= TDF_ASTPENDING;
1381 		thread_unlock(td);
1382 	}
1383 
1384 	seltdinit(td);
1385 	/* Iterate until the timeout expires or descriptors become ready. */
1386 	for (;;) {
1387 		error = pollscan(td, bits, nfds);
1388 		if (error || td->td_retval[0] != 0)
1389 			break;
1390 		error = seltdwait(td, sbt, precision);
1391 		if (error)
1392 			break;
1393 		error = pollrescan(td);
1394 		if (error || td->td_retval[0] != 0)
1395 			break;
1396 	}
1397 	seltdclear(td);
1398 
1399 done:
1400 	/* poll is not restarted after signals... */
1401 	if (error == ERESTART)
1402 		error = EINTR;
1403 	if (error == EWOULDBLOCK)
1404 		error = 0;
1405 	if (error == 0) {
1406 		error = pollout(td, bits, fds, nfds);
1407 		if (error)
1408 			goto out;
1409 	}
1410 out:
1411 	if (ni > sizeof(smallbits))
1412 		free(bits, M_TEMP);
1413 	return (error);
1414 }
1415 
1416 int
1417 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1418 {
1419 	struct timespec ts, *tsp;
1420 	sigset_t set, *ssp;
1421 	int error;
1422 
1423 	if (uap->ts != NULL) {
1424 		error = copyin(uap->ts, &ts, sizeof(ts));
1425 		if (error)
1426 			return (error);
1427 		tsp = &ts;
1428 	} else
1429 		tsp = NULL;
1430 	if (uap->set != NULL) {
1431 		error = copyin(uap->set, &set, sizeof(set));
1432 		if (error)
1433 			return (error);
1434 		ssp = &set;
1435 	} else
1436 		ssp = NULL;
1437 	/*
1438 	 * fds is still a pointer to user space. kern_poll() will
1439 	 * take care of copyin that array to the kernel space.
1440 	 */
1441 
1442 	return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1443 }
1444 
1445 static int
1446 pollrescan(struct thread *td)
1447 {
1448 	struct seltd *stp;
1449 	struct selfd *sfp;
1450 	struct selfd *sfn;
1451 	struct selinfo *si;
1452 	struct filedesc *fdp;
1453 	struct file *fp;
1454 	struct pollfd *fd;
1455 #ifdef CAPABILITIES
1456 	cap_rights_t rights;
1457 #endif
1458 	int n;
1459 
1460 	n = 0;
1461 	fdp = td->td_proc->p_fd;
1462 	stp = td->td_sel;
1463 	FILEDESC_SLOCK(fdp);
1464 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1465 		fd = (struct pollfd *)sfp->sf_cookie;
1466 		si = sfp->sf_si;
1467 		selfdfree(stp, sfp);
1468 		/* If the selinfo wasn't cleared the event didn't fire. */
1469 		if (si != NULL)
1470 			continue;
1471 		fp = fdp->fd_ofiles[fd->fd].fde_file;
1472 #ifdef CAPABILITIES
1473 		if (fp == NULL ||
1474 		    cap_check(cap_rights(fdp, fd->fd),
1475 		    cap_rights_init(&rights, CAP_EVENT)) != 0)
1476 #else
1477 		if (fp == NULL)
1478 #endif
1479 		{
1480 			fd->revents = POLLNVAL;
1481 			n++;
1482 			continue;
1483 		}
1484 
1485 		/*
1486 		 * Note: backend also returns POLLHUP and
1487 		 * POLLERR if appropriate.
1488 		 */
1489 		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1490 		if (fd->revents != 0)
1491 			n++;
1492 	}
1493 	FILEDESC_SUNLOCK(fdp);
1494 	stp->st_flags = 0;
1495 	td->td_retval[0] = n;
1496 	return (0);
1497 }
1498 
1499 
1500 static int
1501 pollout(td, fds, ufds, nfd)
1502 	struct thread *td;
1503 	struct pollfd *fds;
1504 	struct pollfd *ufds;
1505 	u_int nfd;
1506 {
1507 	int error = 0;
1508 	u_int i = 0;
1509 	u_int n = 0;
1510 
1511 	for (i = 0; i < nfd; i++) {
1512 		error = copyout(&fds->revents, &ufds->revents,
1513 		    sizeof(ufds->revents));
1514 		if (error)
1515 			return (error);
1516 		if (fds->revents != 0)
1517 			n++;
1518 		fds++;
1519 		ufds++;
1520 	}
1521 	td->td_retval[0] = n;
1522 	return (0);
1523 }
1524 
1525 static int
1526 pollscan(td, fds, nfd)
1527 	struct thread *td;
1528 	struct pollfd *fds;
1529 	u_int nfd;
1530 {
1531 	struct filedesc *fdp = td->td_proc->p_fd;
1532 	struct file *fp;
1533 #ifdef CAPABILITIES
1534 	cap_rights_t rights;
1535 #endif
1536 	int i, n = 0;
1537 
1538 	FILEDESC_SLOCK(fdp);
1539 	for (i = 0; i < nfd; i++, fds++) {
1540 		if (fds->fd > fdp->fd_lastfile) {
1541 			fds->revents = POLLNVAL;
1542 			n++;
1543 		} else if (fds->fd < 0) {
1544 			fds->revents = 0;
1545 		} else {
1546 			fp = fdp->fd_ofiles[fds->fd].fde_file;
1547 #ifdef CAPABILITIES
1548 			if (fp == NULL ||
1549 			    cap_check(cap_rights(fdp, fds->fd),
1550 			    cap_rights_init(&rights, CAP_EVENT)) != 0)
1551 #else
1552 			if (fp == NULL)
1553 #endif
1554 			{
1555 				fds->revents = POLLNVAL;
1556 				n++;
1557 			} else {
1558 				/*
1559 				 * Note: backend also returns POLLHUP and
1560 				 * POLLERR if appropriate.
1561 				 */
1562 				selfdalloc(td, fds);
1563 				fds->revents = fo_poll(fp, fds->events,
1564 				    td->td_ucred, td);
1565 				/*
1566 				 * POSIX requires POLLOUT to be never
1567 				 * set simultaneously with POLLHUP.
1568 				 */
1569 				if ((fds->revents & POLLHUP) != 0)
1570 					fds->revents &= ~POLLOUT;
1571 
1572 				if (fds->revents != 0)
1573 					n++;
1574 			}
1575 		}
1576 	}
1577 	FILEDESC_SUNLOCK(fdp);
1578 	td->td_retval[0] = n;
1579 	return (0);
1580 }
1581 
1582 /*
1583  * OpenBSD poll system call.
1584  *
1585  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
1586  */
1587 #ifndef _SYS_SYSPROTO_H_
1588 struct openbsd_poll_args {
1589 	struct pollfd *fds;
1590 	u_int	nfds;
1591 	int	timeout;
1592 };
1593 #endif
1594 int
1595 sys_openbsd_poll(td, uap)
1596 	register struct thread *td;
1597 	register struct openbsd_poll_args *uap;
1598 {
1599 	return (sys_poll(td, (struct poll_args *)uap));
1600 }
1601 
1602 /*
1603  * XXX This was created specifically to support netncp and netsmb.  This
1604  * allows the caller to specify a socket to wait for events on.  It returns
1605  * 0 if any events matched and an error otherwise.  There is no way to
1606  * determine which events fired.
1607  */
1608 int
1609 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1610 {
1611 	struct timeval rtv;
1612 	sbintime_t asbt, precision, rsbt;
1613 	int error;
1614 
1615 	precision = 0;	/* stupid gcc! */
1616 	if (tvp != NULL) {
1617 		rtv = *tvp;
1618 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1619 		    rtv.tv_usec >= 1000000)
1620 			return (EINVAL);
1621 		if (!timevalisset(&rtv))
1622 			asbt = 0;
1623 		else if (rtv.tv_sec <= INT32_MAX) {
1624 			rsbt = tvtosbt(rtv);
1625 			precision = rsbt;
1626 			precision >>= tc_precexp;
1627 			if (TIMESEL(&asbt, rsbt))
1628 				asbt += tc_tick_sbt;
1629 			if (asbt <= SBT_MAX - rsbt)
1630 				asbt += rsbt;
1631 			else
1632 				asbt = -1;
1633 		} else
1634 			asbt = -1;
1635 	} else
1636 		asbt = -1;
1637 	seltdinit(td);
1638 	/*
1639 	 * Iterate until the timeout expires or the socket becomes ready.
1640 	 */
1641 	for (;;) {
1642 		selfdalloc(td, NULL);
1643 		error = sopoll(so, events, NULL, td);
1644 		/* error here is actually the ready events. */
1645 		if (error)
1646 			return (0);
1647 		error = seltdwait(td, asbt, precision);
1648 		if (error)
1649 			break;
1650 	}
1651 	seltdclear(td);
1652 	/* XXX Duplicates ncp/smb behavior. */
1653 	if (error == ERESTART)
1654 		error = 0;
1655 	return (error);
1656 }
1657 
1658 /*
1659  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1660  * have two select sets, one for read and another for write.
1661  */
1662 static void
1663 selfdalloc(struct thread *td, void *cookie)
1664 {
1665 	struct seltd *stp;
1666 
1667 	stp = td->td_sel;
1668 	if (stp->st_free1 == NULL)
1669 		stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1670 	stp->st_free1->sf_td = stp;
1671 	stp->st_free1->sf_cookie = cookie;
1672 	if (stp->st_free2 == NULL)
1673 		stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1674 	stp->st_free2->sf_td = stp;
1675 	stp->st_free2->sf_cookie = cookie;
1676 }
1677 
1678 static void
1679 selfdfree(struct seltd *stp, struct selfd *sfp)
1680 {
1681 	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1682 	if (sfp->sf_si != NULL) {
1683 		mtx_lock(sfp->sf_mtx);
1684 		if (sfp->sf_si != NULL)
1685 			TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1686 		mtx_unlock(sfp->sf_mtx);
1687 	}
1688 	uma_zfree(selfd_zone, sfp);
1689 }
1690 
1691 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1692 void
1693 seldrain(sip)
1694         struct selinfo *sip;
1695 {
1696 
1697 	/*
1698 	 * This feature is already provided by doselwakeup(), thus it is
1699 	 * enough to go for it.
1700 	 * Eventually, the context, should take care to avoid races
1701 	 * between thread calling select()/poll() and file descriptor
1702 	 * detaching, but, again, the races are just the same as
1703 	 * selwakeup().
1704 	 */
1705         doselwakeup(sip, -1);
1706 }
1707 
1708 /*
1709  * Record a select request.
1710  */
1711 void
1712 selrecord(selector, sip)
1713 	struct thread *selector;
1714 	struct selinfo *sip;
1715 {
1716 	struct selfd *sfp;
1717 	struct seltd *stp;
1718 	struct mtx *mtxp;
1719 
1720 	stp = selector->td_sel;
1721 	/*
1722 	 * Don't record when doing a rescan.
1723 	 */
1724 	if (stp->st_flags & SELTD_RESCAN)
1725 		return;
1726 	/*
1727 	 * Grab one of the preallocated descriptors.
1728 	 */
1729 	sfp = NULL;
1730 	if ((sfp = stp->st_free1) != NULL)
1731 		stp->st_free1 = NULL;
1732 	else if ((sfp = stp->st_free2) != NULL)
1733 		stp->st_free2 = NULL;
1734 	else
1735 		panic("selrecord: No free selfd on selq");
1736 	mtxp = sip->si_mtx;
1737 	if (mtxp == NULL)
1738 		mtxp = mtx_pool_find(mtxpool_select, sip);
1739 	/*
1740 	 * Initialize the sfp and queue it in the thread.
1741 	 */
1742 	sfp->sf_si = sip;
1743 	sfp->sf_mtx = mtxp;
1744 	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1745 	/*
1746 	 * Now that we've locked the sip, check for initialization.
1747 	 */
1748 	mtx_lock(mtxp);
1749 	if (sip->si_mtx == NULL) {
1750 		sip->si_mtx = mtxp;
1751 		TAILQ_INIT(&sip->si_tdlist);
1752 	}
1753 	/*
1754 	 * Add this thread to the list of selfds listening on this selinfo.
1755 	 */
1756 	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1757 	mtx_unlock(sip->si_mtx);
1758 }
1759 
1760 /* Wake up a selecting thread. */
1761 void
1762 selwakeup(sip)
1763 	struct selinfo *sip;
1764 {
1765 	doselwakeup(sip, -1);
1766 }
1767 
1768 /* Wake up a selecting thread, and set its priority. */
1769 void
1770 selwakeuppri(sip, pri)
1771 	struct selinfo *sip;
1772 	int pri;
1773 {
1774 	doselwakeup(sip, pri);
1775 }
1776 
1777 /*
1778  * Do a wakeup when a selectable event occurs.
1779  */
1780 static void
1781 doselwakeup(sip, pri)
1782 	struct selinfo *sip;
1783 	int pri;
1784 {
1785 	struct selfd *sfp;
1786 	struct selfd *sfn;
1787 	struct seltd *stp;
1788 
1789 	/* If it's not initialized there can't be any waiters. */
1790 	if (sip->si_mtx == NULL)
1791 		return;
1792 	/*
1793 	 * Locking the selinfo locks all selfds associated with it.
1794 	 */
1795 	mtx_lock(sip->si_mtx);
1796 	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1797 		/*
1798 		 * Once we remove this sfp from the list and clear the
1799 		 * sf_si seltdclear will know to ignore this si.
1800 		 */
1801 		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1802 		sfp->sf_si = NULL;
1803 		stp = sfp->sf_td;
1804 		mtx_lock(&stp->st_mtx);
1805 		stp->st_flags |= SELTD_PENDING;
1806 		cv_broadcastpri(&stp->st_wait, pri);
1807 		mtx_unlock(&stp->st_mtx);
1808 	}
1809 	mtx_unlock(sip->si_mtx);
1810 }
1811 
1812 static void
1813 seltdinit(struct thread *td)
1814 {
1815 	struct seltd *stp;
1816 
1817 	if ((stp = td->td_sel) != NULL)
1818 		goto out;
1819 	td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1820 	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1821 	cv_init(&stp->st_wait, "select");
1822 out:
1823 	stp->st_flags = 0;
1824 	STAILQ_INIT(&stp->st_selq);
1825 }
1826 
1827 static int
1828 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1829 {
1830 	struct seltd *stp;
1831 	int error;
1832 
1833 	stp = td->td_sel;
1834 	/*
1835 	 * An event of interest may occur while we do not hold the seltd
1836 	 * locked so check the pending flag before we sleep.
1837 	 */
1838 	mtx_lock(&stp->st_mtx);
1839 	/*
1840 	 * Any further calls to selrecord will be a rescan.
1841 	 */
1842 	stp->st_flags |= SELTD_RESCAN;
1843 	if (stp->st_flags & SELTD_PENDING) {
1844 		mtx_unlock(&stp->st_mtx);
1845 		return (0);
1846 	}
1847 	if (sbt == 0)
1848 		error = EWOULDBLOCK;
1849 	else if (sbt != -1)
1850 		error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
1851 		    sbt, precision, C_ABSOLUTE);
1852 	else
1853 		error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1854 	mtx_unlock(&stp->st_mtx);
1855 
1856 	return (error);
1857 }
1858 
1859 void
1860 seltdfini(struct thread *td)
1861 {
1862 	struct seltd *stp;
1863 
1864 	stp = td->td_sel;
1865 	if (stp == NULL)
1866 		return;
1867 	if (stp->st_free1)
1868 		uma_zfree(selfd_zone, stp->st_free1);
1869 	if (stp->st_free2)
1870 		uma_zfree(selfd_zone, stp->st_free2);
1871 	td->td_sel = NULL;
1872 	free(stp, M_SELECT);
1873 }
1874 
1875 /*
1876  * Remove the references to the thread from all of the objects we were
1877  * polling.
1878  */
1879 static void
1880 seltdclear(struct thread *td)
1881 {
1882 	struct seltd *stp;
1883 	struct selfd *sfp;
1884 	struct selfd *sfn;
1885 
1886 	stp = td->td_sel;
1887 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1888 		selfdfree(stp, sfp);
1889 	stp->st_flags = 0;
1890 }
1891 
1892 static void selectinit(void *);
1893 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1894 static void
1895 selectinit(void *dummy __unused)
1896 {
1897 
1898 	selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1899 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1900 	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1901 }
1902