xref: /freebsd/sys/kern/sys_generic.c (revision 57718be8fa0bd5edc11ab9a72e68cc71982939a6)
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 int iosize_max_clamp = 0;
79 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
80     &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
81 int devfs_iosize_max_clamp = 1;
82 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
83     &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
84 
85 /*
86  * Assert that the return value of read(2) and write(2) syscalls fits
87  * into a register.  If not, an architecture will need to provide the
88  * usermode wrappers to reconstruct the result.
89  */
90 CTASSERT(sizeof(register_t) >= sizeof(size_t));
91 
92 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
93 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
94 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
95 
96 static int	pollout(struct thread *, struct pollfd *, struct pollfd *,
97 		    u_int);
98 static int	pollscan(struct thread *, struct pollfd *, u_int);
99 static int	pollrescan(struct thread *);
100 static int	selscan(struct thread *, fd_mask **, fd_mask **, int);
101 static int	selrescan(struct thread *, fd_mask **, fd_mask **);
102 static void	selfdalloc(struct thread *, void *);
103 static void	selfdfree(struct seltd *, struct selfd *);
104 static int	dofileread(struct thread *, int, struct file *, struct uio *,
105 		    off_t, int);
106 static int	dofilewrite(struct thread *, int, struct file *, struct uio *,
107 		    off_t, int);
108 static void	doselwakeup(struct selinfo *, int);
109 static void	seltdinit(struct thread *);
110 static int	seltdwait(struct thread *, sbintime_t, sbintime_t);
111 static void	seltdclear(struct thread *);
112 
113 /*
114  * One seltd per-thread allocated on demand as needed.
115  *
116  *	t - protected by st_mtx
117  * 	k - Only accessed by curthread or read-only
118  */
119 struct seltd {
120 	STAILQ_HEAD(, selfd)	st_selq;	/* (k) List of selfds. */
121 	struct selfd		*st_free1;	/* (k) free fd for read set. */
122 	struct selfd		*st_free2;	/* (k) free fd for write set. */
123 	struct mtx		st_mtx;		/* Protects struct seltd */
124 	struct cv		st_wait;	/* (t) Wait channel. */
125 	int			st_flags;	/* (t) SELTD_ flags. */
126 };
127 
128 #define	SELTD_PENDING	0x0001			/* We have pending events. */
129 #define	SELTD_RESCAN	0x0002			/* Doing a rescan. */
130 
131 /*
132  * One selfd allocated per-thread per-file-descriptor.
133  *	f - protected by sf_mtx
134  */
135 struct selfd {
136 	STAILQ_ENTRY(selfd)	sf_link;	/* (k) fds owned by this td. */
137 	TAILQ_ENTRY(selfd)	sf_threads;	/* (f) fds on this selinfo. */
138 	struct selinfo		*sf_si;		/* (f) selinfo when linked. */
139 	struct mtx		*sf_mtx;	/* Pointer to selinfo mtx. */
140 	struct seltd		*sf_td;		/* (k) owning seltd. */
141 	void			*sf_cookie;	/* (k) fd or pollfd. */
142 };
143 
144 static uma_zone_t selfd_zone;
145 static struct mtx_pool *mtxpool_select;
146 
147 #ifndef _SYS_SYSPROTO_H_
148 struct read_args {
149 	int	fd;
150 	void	*buf;
151 	size_t	nbyte;
152 };
153 #endif
154 int
155 sys_read(td, uap)
156 	struct thread *td;
157 	struct read_args *uap;
158 {
159 	struct uio auio;
160 	struct iovec aiov;
161 	int error;
162 
163 	if (uap->nbyte > IOSIZE_MAX)
164 		return (EINVAL);
165 	aiov.iov_base = uap->buf;
166 	aiov.iov_len = uap->nbyte;
167 	auio.uio_iov = &aiov;
168 	auio.uio_iovcnt = 1;
169 	auio.uio_resid = uap->nbyte;
170 	auio.uio_segflg = UIO_USERSPACE;
171 	error = kern_readv(td, uap->fd, &auio);
172 	return(error);
173 }
174 
175 /*
176  * Positioned read system call
177  */
178 #ifndef _SYS_SYSPROTO_H_
179 struct pread_args {
180 	int	fd;
181 	void	*buf;
182 	size_t	nbyte;
183 	int	pad;
184 	off_t	offset;
185 };
186 #endif
187 int
188 sys_pread(td, uap)
189 	struct thread *td;
190 	struct pread_args *uap;
191 {
192 	struct uio auio;
193 	struct iovec aiov;
194 	int error;
195 
196 	if (uap->nbyte > IOSIZE_MAX)
197 		return (EINVAL);
198 	aiov.iov_base = uap->buf;
199 	aiov.iov_len = uap->nbyte;
200 	auio.uio_iov = &aiov;
201 	auio.uio_iovcnt = 1;
202 	auio.uio_resid = uap->nbyte;
203 	auio.uio_segflg = UIO_USERSPACE;
204 	error = kern_preadv(td, uap->fd, &auio, uap->offset);
205 	return(error);
206 }
207 
208 int
209 freebsd6_pread(td, uap)
210 	struct thread *td;
211 	struct freebsd6_pread_args *uap;
212 {
213 	struct pread_args oargs;
214 
215 	oargs.fd = uap->fd;
216 	oargs.buf = uap->buf;
217 	oargs.nbyte = uap->nbyte;
218 	oargs.offset = uap->offset;
219 	return (sys_pread(td, &oargs));
220 }
221 
222 /*
223  * Scatter read system call.
224  */
225 #ifndef _SYS_SYSPROTO_H_
226 struct readv_args {
227 	int	fd;
228 	struct	iovec *iovp;
229 	u_int	iovcnt;
230 };
231 #endif
232 int
233 sys_readv(struct thread *td, struct readv_args *uap)
234 {
235 	struct uio *auio;
236 	int error;
237 
238 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
239 	if (error)
240 		return (error);
241 	error = kern_readv(td, uap->fd, auio);
242 	free(auio, M_IOV);
243 	return (error);
244 }
245 
246 int
247 kern_readv(struct thread *td, int fd, struct uio *auio)
248 {
249 	struct file *fp;
250 	cap_rights_t rights;
251 	int error;
252 
253 	error = fget_read(td, fd, cap_rights_init(&rights, CAP_READ), &fp);
254 	if (error)
255 		return (error);
256 	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
257 	fdrop(fp, td);
258 	return (error);
259 }
260 
261 /*
262  * Scatter positioned read system call.
263  */
264 #ifndef _SYS_SYSPROTO_H_
265 struct preadv_args {
266 	int	fd;
267 	struct	iovec *iovp;
268 	u_int	iovcnt;
269 	off_t	offset;
270 };
271 #endif
272 int
273 sys_preadv(struct thread *td, struct preadv_args *uap)
274 {
275 	struct uio *auio;
276 	int error;
277 
278 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
279 	if (error)
280 		return (error);
281 	error = kern_preadv(td, uap->fd, auio, uap->offset);
282 	free(auio, M_IOV);
283 	return (error);
284 }
285 
286 int
287 kern_preadv(td, fd, auio, offset)
288 	struct thread *td;
289 	int fd;
290 	struct uio *auio;
291 	off_t offset;
292 {
293 	struct file *fp;
294 	cap_rights_t rights;
295 	int error;
296 
297 	error = fget_read(td, fd, cap_rights_init(&rights, CAP_PREAD), &fp);
298 	if (error)
299 		return (error);
300 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
301 		error = ESPIPE;
302 	else if (offset < 0 && fp->f_vnode->v_type != VCHR)
303 		error = EINVAL;
304 	else
305 		error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
306 	fdrop(fp, td);
307 	return (error);
308 }
309 
310 /*
311  * Common code for readv and preadv that reads data in
312  * from a file using the passed in uio, offset, and flags.
313  */
314 static int
315 dofileread(td, fd, fp, auio, offset, flags)
316 	struct thread *td;
317 	int fd;
318 	struct file *fp;
319 	struct uio *auio;
320 	off_t offset;
321 	int flags;
322 {
323 	ssize_t cnt;
324 	int error;
325 #ifdef KTRACE
326 	struct uio *ktruio = NULL;
327 #endif
328 
329 	/* Finish zero length reads right here */
330 	if (auio->uio_resid == 0) {
331 		td->td_retval[0] = 0;
332 		return(0);
333 	}
334 	auio->uio_rw = UIO_READ;
335 	auio->uio_offset = offset;
336 	auio->uio_td = td;
337 #ifdef KTRACE
338 	if (KTRPOINT(td, KTR_GENIO))
339 		ktruio = cloneuio(auio);
340 #endif
341 	cnt = auio->uio_resid;
342 	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
343 		if (auio->uio_resid != cnt && (error == ERESTART ||
344 		    error == EINTR || error == EWOULDBLOCK))
345 			error = 0;
346 	}
347 	cnt -= auio->uio_resid;
348 #ifdef KTRACE
349 	if (ktruio != NULL) {
350 		ktruio->uio_resid = cnt;
351 		ktrgenio(fd, UIO_READ, ktruio, error);
352 	}
353 #endif
354 	td->td_retval[0] = cnt;
355 	return (error);
356 }
357 
358 #ifndef _SYS_SYSPROTO_H_
359 struct write_args {
360 	int	fd;
361 	const void *buf;
362 	size_t	nbyte;
363 };
364 #endif
365 int
366 sys_write(td, uap)
367 	struct thread *td;
368 	struct write_args *uap;
369 {
370 	struct uio auio;
371 	struct iovec aiov;
372 	int error;
373 
374 	if (uap->nbyte > IOSIZE_MAX)
375 		return (EINVAL);
376 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
377 	aiov.iov_len = uap->nbyte;
378 	auio.uio_iov = &aiov;
379 	auio.uio_iovcnt = 1;
380 	auio.uio_resid = uap->nbyte;
381 	auio.uio_segflg = UIO_USERSPACE;
382 	error = kern_writev(td, uap->fd, &auio);
383 	return(error);
384 }
385 
386 /*
387  * Positioned write system call.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct pwrite_args {
391 	int	fd;
392 	const void *buf;
393 	size_t	nbyte;
394 	int	pad;
395 	off_t	offset;
396 };
397 #endif
398 int
399 sys_pwrite(td, uap)
400 	struct thread *td;
401 	struct pwrite_args *uap;
402 {
403 	struct uio auio;
404 	struct iovec aiov;
405 	int error;
406 
407 	if (uap->nbyte > IOSIZE_MAX)
408 		return (EINVAL);
409 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
410 	aiov.iov_len = uap->nbyte;
411 	auio.uio_iov = &aiov;
412 	auio.uio_iovcnt = 1;
413 	auio.uio_resid = uap->nbyte;
414 	auio.uio_segflg = UIO_USERSPACE;
415 	error = kern_pwritev(td, uap->fd, &auio, uap->offset);
416 	return(error);
417 }
418 
419 int
420 freebsd6_pwrite(td, uap)
421 	struct thread *td;
422 	struct freebsd6_pwrite_args *uap;
423 {
424 	struct pwrite_args oargs;
425 
426 	oargs.fd = uap->fd;
427 	oargs.buf = uap->buf;
428 	oargs.nbyte = uap->nbyte;
429 	oargs.offset = uap->offset;
430 	return (sys_pwrite(td, &oargs));
431 }
432 
433 /*
434  * Gather write system call.
435  */
436 #ifndef _SYS_SYSPROTO_H_
437 struct writev_args {
438 	int	fd;
439 	struct	iovec *iovp;
440 	u_int	iovcnt;
441 };
442 #endif
443 int
444 sys_writev(struct thread *td, struct writev_args *uap)
445 {
446 	struct uio *auio;
447 	int error;
448 
449 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
450 	if (error)
451 		return (error);
452 	error = kern_writev(td, uap->fd, auio);
453 	free(auio, M_IOV);
454 	return (error);
455 }
456 
457 int
458 kern_writev(struct thread *td, int fd, struct uio *auio)
459 {
460 	struct file *fp;
461 	cap_rights_t rights;
462 	int error;
463 
464 	error = fget_write(td, fd, cap_rights_init(&rights, CAP_WRITE), &fp);
465 	if (error)
466 		return (error);
467 	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
468 	fdrop(fp, td);
469 	return (error);
470 }
471 
472 /*
473  * Gather positioned write system call.
474  */
475 #ifndef _SYS_SYSPROTO_H_
476 struct pwritev_args {
477 	int	fd;
478 	struct	iovec *iovp;
479 	u_int	iovcnt;
480 	off_t	offset;
481 };
482 #endif
483 int
484 sys_pwritev(struct thread *td, struct pwritev_args *uap)
485 {
486 	struct uio *auio;
487 	int error;
488 
489 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
490 	if (error)
491 		return (error);
492 	error = kern_pwritev(td, uap->fd, auio, uap->offset);
493 	free(auio, M_IOV);
494 	return (error);
495 }
496 
497 int
498 kern_pwritev(td, fd, auio, offset)
499 	struct thread *td;
500 	struct uio *auio;
501 	int fd;
502 	off_t offset;
503 {
504 	struct file *fp;
505 	cap_rights_t rights;
506 	int error;
507 
508 	error = fget_write(td, fd, cap_rights_init(&rights, CAP_PWRITE), &fp);
509 	if (error)
510 		return (error);
511 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
512 		error = ESPIPE;
513 	else if (offset < 0 && fp->f_vnode->v_type != VCHR)
514 		error = EINVAL;
515 	else
516 		error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
517 	fdrop(fp, td);
518 	return (error);
519 }
520 
521 /*
522  * Common code for writev and pwritev that writes data to
523  * a file using the passed in uio, offset, and flags.
524  */
525 static int
526 dofilewrite(td, fd, fp, auio, offset, flags)
527 	struct thread *td;
528 	int fd;
529 	struct file *fp;
530 	struct uio *auio;
531 	off_t offset;
532 	int flags;
533 {
534 	ssize_t cnt;
535 	int error;
536 #ifdef KTRACE
537 	struct uio *ktruio = NULL;
538 #endif
539 
540 	auio->uio_rw = UIO_WRITE;
541 	auio->uio_td = td;
542 	auio->uio_offset = offset;
543 #ifdef KTRACE
544 	if (KTRPOINT(td, KTR_GENIO))
545 		ktruio = cloneuio(auio);
546 #endif
547 	cnt = auio->uio_resid;
548 	if (fp->f_type == DTYPE_VNODE &&
549 	    (fp->f_vnread_flags & FDEVFS_VNODE) == 0)
550 		bwillwrite();
551 	if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
552 		if (auio->uio_resid != cnt && (error == ERESTART ||
553 		    error == EINTR || error == EWOULDBLOCK))
554 			error = 0;
555 		/* Socket layer is responsible for issuing SIGPIPE. */
556 		if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
557 			PROC_LOCK(td->td_proc);
558 			tdsignal(td, SIGPIPE);
559 			PROC_UNLOCK(td->td_proc);
560 		}
561 	}
562 	cnt -= auio->uio_resid;
563 #ifdef KTRACE
564 	if (ktruio != NULL) {
565 		ktruio->uio_resid = cnt;
566 		ktrgenio(fd, UIO_WRITE, ktruio, error);
567 	}
568 #endif
569 	td->td_retval[0] = cnt;
570 	return (error);
571 }
572 
573 /*
574  * Truncate a file given a file descriptor.
575  *
576  * Can't use fget_write() here, since must return EINVAL and not EBADF if the
577  * descriptor isn't writable.
578  */
579 int
580 kern_ftruncate(td, fd, length)
581 	struct thread *td;
582 	int fd;
583 	off_t length;
584 {
585 	struct file *fp;
586 	cap_rights_t rights;
587 	int error;
588 
589 	AUDIT_ARG_FD(fd);
590 	if (length < 0)
591 		return (EINVAL);
592 	error = fget(td, fd, cap_rights_init(&rights, CAP_FTRUNCATE), &fp);
593 	if (error)
594 		return (error);
595 	AUDIT_ARG_FILE(td->td_proc, fp);
596 	if (!(fp->f_flag & FWRITE)) {
597 		fdrop(fp, td);
598 		return (EINVAL);
599 	}
600 	error = fo_truncate(fp, length, td->td_ucred, td);
601 	fdrop(fp, td);
602 	return (error);
603 }
604 
605 #ifndef _SYS_SYSPROTO_H_
606 struct ftruncate_args {
607 	int	fd;
608 	int	pad;
609 	off_t	length;
610 };
611 #endif
612 int
613 sys_ftruncate(td, uap)
614 	struct thread *td;
615 	struct ftruncate_args *uap;
616 {
617 
618 	return (kern_ftruncate(td, uap->fd, uap->length));
619 }
620 
621 #if defined(COMPAT_43)
622 #ifndef _SYS_SYSPROTO_H_
623 struct oftruncate_args {
624 	int	fd;
625 	long	length;
626 };
627 #endif
628 int
629 oftruncate(td, uap)
630 	struct thread *td;
631 	struct oftruncate_args *uap;
632 {
633 
634 	return (kern_ftruncate(td, uap->fd, uap->length));
635 }
636 #endif /* COMPAT_43 */
637 
638 #ifndef _SYS_SYSPROTO_H_
639 struct ioctl_args {
640 	int	fd;
641 	u_long	com;
642 	caddr_t	data;
643 };
644 #endif
645 /* ARGSUSED */
646 int
647 sys_ioctl(struct thread *td, struct ioctl_args *uap)
648 {
649 	u_long com;
650 	int arg, error;
651 	u_int size;
652 	caddr_t data;
653 
654 	if (uap->com > 0xffffffff) {
655 		printf(
656 		    "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
657 		    td->td_proc->p_pid, td->td_name, uap->com);
658 		uap->com &= 0xffffffff;
659 	}
660 	com = uap->com;
661 
662 	/*
663 	 * Interpret high order word to find amount of data to be
664 	 * copied to/from the user's address space.
665 	 */
666 	size = IOCPARM_LEN(com);
667 	if ((size > IOCPARM_MAX) ||
668 	    ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
669 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
670 	    ((com & IOC_OUT) && size == 0) ||
671 #else
672 	    ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
673 #endif
674 	    ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
675 		return (ENOTTY);
676 
677 	if (size > 0) {
678 		if (com & IOC_VOID) {
679 			/* Integer argument. */
680 			arg = (intptr_t)uap->data;
681 			data = (void *)&arg;
682 			size = 0;
683 		} else
684 			data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
685 	} else
686 		data = (void *)&uap->data;
687 	if (com & IOC_IN) {
688 		error = copyin(uap->data, data, (u_int)size);
689 		if (error) {
690 			if (size > 0)
691 				free(data, M_IOCTLOPS);
692 			return (error);
693 		}
694 	} else if (com & IOC_OUT) {
695 		/*
696 		 * Zero the buffer so the user always
697 		 * gets back something deterministic.
698 		 */
699 		bzero(data, size);
700 	}
701 
702 	error = kern_ioctl(td, uap->fd, com, data);
703 
704 	if (error == 0 && (com & IOC_OUT))
705 		error = copyout(data, uap->data, (u_int)size);
706 
707 	if (size > 0)
708 		free(data, M_IOCTLOPS);
709 	return (error);
710 }
711 
712 int
713 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
714 {
715 	struct file *fp;
716 	struct filedesc *fdp;
717 #ifndef CAPABILITIES
718 	cap_rights_t rights;
719 #endif
720 	int error, tmp, locked;
721 
722 	AUDIT_ARG_FD(fd);
723 	AUDIT_ARG_CMD(com);
724 
725 	fdp = td->td_proc->p_fd;
726 
727 	switch (com) {
728 	case FIONCLEX:
729 	case FIOCLEX:
730 		FILEDESC_XLOCK(fdp);
731 		locked = LA_XLOCKED;
732 		break;
733 	default:
734 #ifdef CAPABILITIES
735 		FILEDESC_SLOCK(fdp);
736 		locked = LA_SLOCKED;
737 #else
738 		locked = LA_UNLOCKED;
739 #endif
740 		break;
741 	}
742 
743 #ifdef CAPABILITIES
744 	if ((fp = fget_locked(fdp, fd)) == NULL) {
745 		error = EBADF;
746 		goto out;
747 	}
748 	if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
749 		fp = NULL;	/* fhold() was not called yet */
750 		goto out;
751 	}
752 	fhold(fp);
753 	if (locked == LA_SLOCKED) {
754 		FILEDESC_SUNLOCK(fdp);
755 		locked = LA_UNLOCKED;
756 	}
757 #else
758 	error = fget(td, fd, cap_rights_init(&rights, CAP_IOCTL), &fp);
759 	if (error != 0) {
760 		fp = NULL;
761 		goto out;
762 	}
763 #endif
764 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
765 		error = EBADF;
766 		goto out;
767 	}
768 
769 	switch (com) {
770 	case FIONCLEX:
771 		fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
772 		goto out;
773 	case FIOCLEX:
774 		fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
775 		goto out;
776 	case FIONBIO:
777 		if ((tmp = *(int *)data))
778 			atomic_set_int(&fp->f_flag, FNONBLOCK);
779 		else
780 			atomic_clear_int(&fp->f_flag, FNONBLOCK);
781 		data = (void *)&tmp;
782 		break;
783 	case FIOASYNC:
784 		if ((tmp = *(int *)data))
785 			atomic_set_int(&fp->f_flag, FASYNC);
786 		else
787 			atomic_clear_int(&fp->f_flag, FASYNC);
788 		data = (void *)&tmp;
789 		break;
790 	}
791 
792 	error = fo_ioctl(fp, com, data, td->td_ucred, td);
793 out:
794 	switch (locked) {
795 	case LA_XLOCKED:
796 		FILEDESC_XUNLOCK(fdp);
797 		break;
798 #ifdef CAPABILITIES
799 	case LA_SLOCKED:
800 		FILEDESC_SUNLOCK(fdp);
801 		break;
802 #endif
803 	default:
804 		FILEDESC_UNLOCK_ASSERT(fdp);
805 		break;
806 	}
807 	if (fp != NULL)
808 		fdrop(fp, td);
809 	return (error);
810 }
811 
812 int
813 poll_no_poll(int events)
814 {
815 	/*
816 	 * Return true for read/write.  If the user asked for something
817 	 * special, return POLLNVAL, so that clients have a way of
818 	 * determining reliably whether or not the extended
819 	 * functionality is present without hard-coding knowledge
820 	 * of specific filesystem implementations.
821 	 */
822 	if (events & ~POLLSTANDARD)
823 		return (POLLNVAL);
824 
825 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
826 }
827 
828 int
829 sys_pselect(struct thread *td, struct pselect_args *uap)
830 {
831 	struct timespec ts;
832 	struct timeval tv, *tvp;
833 	sigset_t set, *uset;
834 	int error;
835 
836 	if (uap->ts != NULL) {
837 		error = copyin(uap->ts, &ts, sizeof(ts));
838 		if (error != 0)
839 		    return (error);
840 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
841 		tvp = &tv;
842 	} else
843 		tvp = NULL;
844 	if (uap->sm != NULL) {
845 		error = copyin(uap->sm, &set, sizeof(set));
846 		if (error != 0)
847 			return (error);
848 		uset = &set;
849 	} else
850 		uset = NULL;
851 	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
852 	    uset, NFDBITS));
853 }
854 
855 int
856 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
857     struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
858 {
859 	int error;
860 
861 	if (uset != NULL) {
862 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
863 		    &td->td_oldsigmask, 0);
864 		if (error != 0)
865 			return (error);
866 		td->td_pflags |= TDP_OLDMASK;
867 		/*
868 		 * Make sure that ast() is called on return to
869 		 * usermode and TDP_OLDMASK is cleared, restoring old
870 		 * sigmask.
871 		 */
872 		thread_lock(td);
873 		td->td_flags |= TDF_ASTPENDING;
874 		thread_unlock(td);
875 	}
876 	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
877 	return (error);
878 }
879 
880 #ifndef _SYS_SYSPROTO_H_
881 struct select_args {
882 	int	nd;
883 	fd_set	*in, *ou, *ex;
884 	struct	timeval *tv;
885 };
886 #endif
887 int
888 sys_select(struct thread *td, struct select_args *uap)
889 {
890 	struct timeval tv, *tvp;
891 	int error;
892 
893 	if (uap->tv != NULL) {
894 		error = copyin(uap->tv, &tv, sizeof(tv));
895 		if (error)
896 			return (error);
897 		tvp = &tv;
898 	} else
899 		tvp = NULL;
900 
901 	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
902 	    NFDBITS));
903 }
904 
905 /*
906  * In the unlikely case when user specified n greater then the last
907  * open file descriptor, check that no bits are set after the last
908  * valid fd.  We must return EBADF if any is set.
909  *
910  * There are applications that rely on the behaviour.
911  *
912  * nd is fd_lastfile + 1.
913  */
914 static int
915 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
916 {
917 	char *addr, *oaddr;
918 	int b, i, res;
919 	uint8_t bits;
920 
921 	if (nd >= ndu || fd_in == NULL)
922 		return (0);
923 
924 	oaddr = NULL;
925 	bits = 0; /* silence gcc */
926 	for (i = nd; i < ndu; i++) {
927 		b = i / NBBY;
928 #if BYTE_ORDER == LITTLE_ENDIAN
929 		addr = (char *)fd_in + b;
930 #else
931 		addr = (char *)fd_in;
932 		if (abi_nfdbits == NFDBITS) {
933 			addr += rounddown(b, sizeof(fd_mask)) +
934 			    sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
935 		} else {
936 			addr += rounddown(b, sizeof(uint32_t)) +
937 			    sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
938 		}
939 #endif
940 		if (addr != oaddr) {
941 			res = fubyte(addr);
942 			if (res == -1)
943 				return (EFAULT);
944 			oaddr = addr;
945 			bits = res;
946 		}
947 		if ((bits & (1 << (i % NBBY))) != 0)
948 			return (EBADF);
949 	}
950 	return (0);
951 }
952 
953 int
954 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
955     fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
956 {
957 	struct filedesc *fdp;
958 	/*
959 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
960 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
961 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
962 	 * of 256.
963 	 */
964 	fd_mask s_selbits[howmany(2048, NFDBITS)];
965 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
966 	struct timeval rtv;
967 	sbintime_t asbt, precision, rsbt;
968 	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
969 	int error, lf, ndu;
970 
971 	if (nd < 0)
972 		return (EINVAL);
973 	fdp = td->td_proc->p_fd;
974 	ndu = nd;
975 	lf = fdp->fd_lastfile;
976 	if (nd > lf + 1)
977 		nd = lf + 1;
978 
979 	error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
980 	if (error != 0)
981 		return (error);
982 	error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
983 	if (error != 0)
984 		return (error);
985 	error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
986 	if (error != 0)
987 		return (error);
988 
989 	/*
990 	 * Allocate just enough bits for the non-null fd_sets.  Use the
991 	 * preallocated auto buffer if possible.
992 	 */
993 	nfdbits = roundup(nd, NFDBITS);
994 	ncpbytes = nfdbits / NBBY;
995 	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
996 	nbufbytes = 0;
997 	if (fd_in != NULL)
998 		nbufbytes += 2 * ncpbytes;
999 	if (fd_ou != NULL)
1000 		nbufbytes += 2 * ncpbytes;
1001 	if (fd_ex != NULL)
1002 		nbufbytes += 2 * ncpbytes;
1003 	if (nbufbytes <= sizeof s_selbits)
1004 		selbits = &s_selbits[0];
1005 	else
1006 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1007 
1008 	/*
1009 	 * Assign pointers into the bit buffers and fetch the input bits.
1010 	 * Put the output buffers together so that they can be bzeroed
1011 	 * together.
1012 	 */
1013 	sbp = selbits;
1014 #define	getbits(name, x) \
1015 	do {								\
1016 		if (name == NULL) {					\
1017 			ibits[x] = NULL;				\
1018 			obits[x] = NULL;				\
1019 		} else {						\
1020 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
1021 			obits[x] = sbp;					\
1022 			sbp += ncpbytes / sizeof *sbp;			\
1023 			error = copyin(name, ibits[x], ncpubytes);	\
1024 			if (error != 0)					\
1025 				goto done;				\
1026 			bzero((char *)ibits[x] + ncpubytes,		\
1027 			    ncpbytes - ncpubytes);			\
1028 		}							\
1029 	} while (0)
1030 	getbits(fd_in, 0);
1031 	getbits(fd_ou, 1);
1032 	getbits(fd_ex, 2);
1033 #undef	getbits
1034 
1035 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1036 	/*
1037 	 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1038 	 * we are running under 32-bit emulation. This should be more
1039 	 * generic.
1040 	 */
1041 #define swizzle_fdset(bits)						\
1042 	if (abi_nfdbits != NFDBITS && bits != NULL) {			\
1043 		int i;							\
1044 		for (i = 0; i < ncpbytes / sizeof *sbp; i++)		\
1045 			bits[i] = (bits[i] >> 32) | (bits[i] << 32);	\
1046 	}
1047 #else
1048 #define swizzle_fdset(bits)
1049 #endif
1050 
1051 	/* Make sure the bit order makes it through an ABI transition */
1052 	swizzle_fdset(ibits[0]);
1053 	swizzle_fdset(ibits[1]);
1054 	swizzle_fdset(ibits[2]);
1055 
1056 	if (nbufbytes != 0)
1057 		bzero(selbits, nbufbytes / 2);
1058 
1059 	precision = 0;
1060 	if (tvp != NULL) {
1061 		rtv = *tvp;
1062 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1063 		    rtv.tv_usec >= 1000000) {
1064 			error = EINVAL;
1065 			goto done;
1066 		}
1067 		if (!timevalisset(&rtv))
1068 			asbt = 0;
1069 		else if (rtv.tv_sec <= INT32_MAX) {
1070 			rsbt = tvtosbt(rtv);
1071 			precision = rsbt;
1072 			precision >>= tc_precexp;
1073 			if (TIMESEL(&asbt, rsbt))
1074 				asbt += tc_tick_sbt;
1075 			if (asbt <= SBT_MAX - rsbt)
1076 				asbt += rsbt;
1077 			else
1078 				asbt = -1;
1079 		} else
1080 			asbt = -1;
1081 	} else
1082 		asbt = -1;
1083 	seltdinit(td);
1084 	/* Iterate until the timeout expires or descriptors become ready. */
1085 	for (;;) {
1086 		error = selscan(td, ibits, obits, nd);
1087 		if (error || td->td_retval[0] != 0)
1088 			break;
1089 		error = seltdwait(td, asbt, precision);
1090 		if (error)
1091 			break;
1092 		error = selrescan(td, ibits, obits);
1093 		if (error || td->td_retval[0] != 0)
1094 			break;
1095 	}
1096 	seltdclear(td);
1097 
1098 done:
1099 	/* select is not restarted after signals... */
1100 	if (error == ERESTART)
1101 		error = EINTR;
1102 	if (error == EWOULDBLOCK)
1103 		error = 0;
1104 
1105 	/* swizzle bit order back, if necessary */
1106 	swizzle_fdset(obits[0]);
1107 	swizzle_fdset(obits[1]);
1108 	swizzle_fdset(obits[2]);
1109 #undef swizzle_fdset
1110 
1111 #define	putbits(name, x) \
1112 	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1113 		error = error2;
1114 	if (error == 0) {
1115 		int error2;
1116 
1117 		putbits(fd_in, 0);
1118 		putbits(fd_ou, 1);
1119 		putbits(fd_ex, 2);
1120 #undef putbits
1121 	}
1122 	if (selbits != &s_selbits[0])
1123 		free(selbits, M_SELECT);
1124 
1125 	return (error);
1126 }
1127 /*
1128  * Convert a select bit set to poll flags.
1129  *
1130  * The backend always returns POLLHUP/POLLERR if appropriate and we
1131  * return this as a set bit in any set.
1132  */
1133 static int select_flags[3] = {
1134     POLLRDNORM | POLLHUP | POLLERR,
1135     POLLWRNORM | POLLHUP | POLLERR,
1136     POLLRDBAND | POLLERR
1137 };
1138 
1139 /*
1140  * Compute the fo_poll flags required for a fd given by the index and
1141  * bit position in the fd_mask array.
1142  */
1143 static __inline int
1144 selflags(fd_mask **ibits, int idx, fd_mask bit)
1145 {
1146 	int flags;
1147 	int msk;
1148 
1149 	flags = 0;
1150 	for (msk = 0; msk < 3; msk++) {
1151 		if (ibits[msk] == NULL)
1152 			continue;
1153 		if ((ibits[msk][idx] & bit) == 0)
1154 			continue;
1155 		flags |= select_flags[msk];
1156 	}
1157 	return (flags);
1158 }
1159 
1160 /*
1161  * Set the appropriate output bits given a mask of fired events and the
1162  * input bits originally requested.
1163  */
1164 static __inline int
1165 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1166 {
1167 	int msk;
1168 	int n;
1169 
1170 	n = 0;
1171 	for (msk = 0; msk < 3; msk++) {
1172 		if ((events & select_flags[msk]) == 0)
1173 			continue;
1174 		if (ibits[msk] == NULL)
1175 			continue;
1176 		if ((ibits[msk][idx] & bit) == 0)
1177 			continue;
1178 		/*
1179 		 * XXX Check for a duplicate set.  This can occur because a
1180 		 * socket calls selrecord() twice for each poll() call
1181 		 * resulting in two selfds per real fd.  selrescan() will
1182 		 * call selsetbits twice as a result.
1183 		 */
1184 		if ((obits[msk][idx] & bit) != 0)
1185 			continue;
1186 		obits[msk][idx] |= bit;
1187 		n++;
1188 	}
1189 
1190 	return (n);
1191 }
1192 
1193 static __inline int
1194 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1195 {
1196 	cap_rights_t rights;
1197 
1198 	cap_rights_init(&rights, CAP_EVENT);
1199 
1200 	return (fget_unlocked(fdp, fd, &rights, 0, fpp, NULL));
1201 }
1202 
1203 /*
1204  * Traverse the list of fds attached to this thread's seltd and check for
1205  * completion.
1206  */
1207 static int
1208 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1209 {
1210 	struct filedesc *fdp;
1211 	struct selinfo *si;
1212 	struct seltd *stp;
1213 	struct selfd *sfp;
1214 	struct selfd *sfn;
1215 	struct file *fp;
1216 	fd_mask bit;
1217 	int fd, ev, n, idx;
1218 	int error;
1219 
1220 	fdp = td->td_proc->p_fd;
1221 	stp = td->td_sel;
1222 	n = 0;
1223 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1224 		fd = (int)(uintptr_t)sfp->sf_cookie;
1225 		si = sfp->sf_si;
1226 		selfdfree(stp, sfp);
1227 		/* If the selinfo wasn't cleared the event didn't fire. */
1228 		if (si != NULL)
1229 			continue;
1230 		error = getselfd_cap(fdp, fd, &fp);
1231 		if (error)
1232 			return (error);
1233 		idx = fd / NFDBITS;
1234 		bit = (fd_mask)1 << (fd % NFDBITS);
1235 		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1236 		fdrop(fp, td);
1237 		if (ev != 0)
1238 			n += selsetbits(ibits, obits, idx, bit, ev);
1239 	}
1240 	stp->st_flags = 0;
1241 	td->td_retval[0] = n;
1242 	return (0);
1243 }
1244 
1245 /*
1246  * Perform the initial filedescriptor scan and register ourselves with
1247  * each selinfo.
1248  */
1249 static int
1250 selscan(td, ibits, obits, nfd)
1251 	struct thread *td;
1252 	fd_mask **ibits, **obits;
1253 	int nfd;
1254 {
1255 	struct filedesc *fdp;
1256 	struct file *fp;
1257 	fd_mask bit;
1258 	int ev, flags, end, fd;
1259 	int n, idx;
1260 	int error;
1261 
1262 	fdp = td->td_proc->p_fd;
1263 	n = 0;
1264 	for (idx = 0, fd = 0; fd < nfd; idx++) {
1265 		end = imin(fd + NFDBITS, nfd);
1266 		for (bit = 1; fd < end; bit <<= 1, fd++) {
1267 			/* Compute the list of events we're interested in. */
1268 			flags = selflags(ibits, idx, bit);
1269 			if (flags == 0)
1270 				continue;
1271 			error = getselfd_cap(fdp, fd, &fp);
1272 			if (error)
1273 				return (error);
1274 			selfdalloc(td, (void *)(uintptr_t)fd);
1275 			ev = fo_poll(fp, flags, td->td_ucred, td);
1276 			fdrop(fp, td);
1277 			if (ev != 0)
1278 				n += selsetbits(ibits, obits, idx, bit, ev);
1279 		}
1280 	}
1281 
1282 	td->td_retval[0] = n;
1283 	return (0);
1284 }
1285 
1286 #ifndef _SYS_SYSPROTO_H_
1287 struct poll_args {
1288 	struct pollfd *fds;
1289 	u_int	nfds;
1290 	int	timeout;
1291 };
1292 #endif
1293 int
1294 sys_poll(td, uap)
1295 	struct thread *td;
1296 	struct poll_args *uap;
1297 {
1298 	struct pollfd *bits;
1299 	struct pollfd smallbits[32];
1300 	sbintime_t asbt, precision, rsbt;
1301 	u_int nfds;
1302 	int error;
1303 	size_t ni;
1304 
1305 	nfds = uap->nfds;
1306 	if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1307 		return (EINVAL);
1308 	ni = nfds * sizeof(struct pollfd);
1309 	if (ni > sizeof(smallbits))
1310 		bits = malloc(ni, M_TEMP, M_WAITOK);
1311 	else
1312 		bits = smallbits;
1313 	error = copyin(uap->fds, bits, ni);
1314 	if (error)
1315 		goto done;
1316 	precision = 0;
1317 	if (uap->timeout != INFTIM) {
1318 		if (uap->timeout < 0) {
1319 			error = EINVAL;
1320 			goto done;
1321 		}
1322 		if (uap->timeout == 0)
1323 			asbt = 0;
1324 		else {
1325 			rsbt = SBT_1MS * uap->timeout;
1326 			precision = rsbt;
1327 			precision >>= tc_precexp;
1328 			if (TIMESEL(&asbt, rsbt))
1329 				asbt += tc_tick_sbt;
1330 			asbt += rsbt;
1331 		}
1332 	} else
1333 		asbt = -1;
1334 	seltdinit(td);
1335 	/* Iterate until the timeout expires or descriptors become ready. */
1336 	for (;;) {
1337 		error = pollscan(td, bits, nfds);
1338 		if (error || td->td_retval[0] != 0)
1339 			break;
1340 		error = seltdwait(td, asbt, precision);
1341 		if (error)
1342 			break;
1343 		error = pollrescan(td);
1344 		if (error || td->td_retval[0] != 0)
1345 			break;
1346 	}
1347 	seltdclear(td);
1348 
1349 done:
1350 	/* poll is not restarted after signals... */
1351 	if (error == ERESTART)
1352 		error = EINTR;
1353 	if (error == EWOULDBLOCK)
1354 		error = 0;
1355 	if (error == 0) {
1356 		error = pollout(td, bits, uap->fds, nfds);
1357 		if (error)
1358 			goto out;
1359 	}
1360 out:
1361 	if (ni > sizeof(smallbits))
1362 		free(bits, M_TEMP);
1363 	return (error);
1364 }
1365 
1366 static int
1367 pollrescan(struct thread *td)
1368 {
1369 	struct seltd *stp;
1370 	struct selfd *sfp;
1371 	struct selfd *sfn;
1372 	struct selinfo *si;
1373 	struct filedesc *fdp;
1374 	struct file *fp;
1375 	struct pollfd *fd;
1376 #ifdef CAPABILITIES
1377 	cap_rights_t rights;
1378 #endif
1379 	int n;
1380 
1381 	n = 0;
1382 	fdp = td->td_proc->p_fd;
1383 	stp = td->td_sel;
1384 	FILEDESC_SLOCK(fdp);
1385 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1386 		fd = (struct pollfd *)sfp->sf_cookie;
1387 		si = sfp->sf_si;
1388 		selfdfree(stp, sfp);
1389 		/* If the selinfo wasn't cleared the event didn't fire. */
1390 		if (si != NULL)
1391 			continue;
1392 		fp = fdp->fd_ofiles[fd->fd].fde_file;
1393 #ifdef CAPABILITIES
1394 		if (fp == NULL ||
1395 		    cap_check(cap_rights(fdp, fd->fd),
1396 		    cap_rights_init(&rights, CAP_EVENT)) != 0)
1397 #else
1398 		if (fp == NULL)
1399 #endif
1400 		{
1401 			fd->revents = POLLNVAL;
1402 			n++;
1403 			continue;
1404 		}
1405 
1406 		/*
1407 		 * Note: backend also returns POLLHUP and
1408 		 * POLLERR if appropriate.
1409 		 */
1410 		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1411 		if (fd->revents != 0)
1412 			n++;
1413 	}
1414 	FILEDESC_SUNLOCK(fdp);
1415 	stp->st_flags = 0;
1416 	td->td_retval[0] = n;
1417 	return (0);
1418 }
1419 
1420 
1421 static int
1422 pollout(td, fds, ufds, nfd)
1423 	struct thread *td;
1424 	struct pollfd *fds;
1425 	struct pollfd *ufds;
1426 	u_int nfd;
1427 {
1428 	int error = 0;
1429 	u_int i = 0;
1430 	u_int n = 0;
1431 
1432 	for (i = 0; i < nfd; i++) {
1433 		error = copyout(&fds->revents, &ufds->revents,
1434 		    sizeof(ufds->revents));
1435 		if (error)
1436 			return (error);
1437 		if (fds->revents != 0)
1438 			n++;
1439 		fds++;
1440 		ufds++;
1441 	}
1442 	td->td_retval[0] = n;
1443 	return (0);
1444 }
1445 
1446 static int
1447 pollscan(td, fds, nfd)
1448 	struct thread *td;
1449 	struct pollfd *fds;
1450 	u_int nfd;
1451 {
1452 	struct filedesc *fdp = td->td_proc->p_fd;
1453 	struct file *fp;
1454 #ifdef CAPABILITIES
1455 	cap_rights_t rights;
1456 #endif
1457 	int i, n = 0;
1458 
1459 	FILEDESC_SLOCK(fdp);
1460 	for (i = 0; i < nfd; i++, fds++) {
1461 		if (fds->fd > fdp->fd_lastfile) {
1462 			fds->revents = POLLNVAL;
1463 			n++;
1464 		} else if (fds->fd < 0) {
1465 			fds->revents = 0;
1466 		} else {
1467 			fp = fdp->fd_ofiles[fds->fd].fde_file;
1468 #ifdef CAPABILITIES
1469 			if (fp == NULL ||
1470 			    cap_check(cap_rights(fdp, fds->fd),
1471 			    cap_rights_init(&rights, CAP_EVENT)) != 0)
1472 #else
1473 			if (fp == NULL)
1474 #endif
1475 			{
1476 				fds->revents = POLLNVAL;
1477 				n++;
1478 			} else {
1479 				/*
1480 				 * Note: backend also returns POLLHUP and
1481 				 * POLLERR if appropriate.
1482 				 */
1483 				selfdalloc(td, fds);
1484 				fds->revents = fo_poll(fp, fds->events,
1485 				    td->td_ucred, td);
1486 				/*
1487 				 * POSIX requires POLLOUT to be never
1488 				 * set simultaneously with POLLHUP.
1489 				 */
1490 				if ((fds->revents & POLLHUP) != 0)
1491 					fds->revents &= ~POLLOUT;
1492 
1493 				if (fds->revents != 0)
1494 					n++;
1495 			}
1496 		}
1497 	}
1498 	FILEDESC_SUNLOCK(fdp);
1499 	td->td_retval[0] = n;
1500 	return (0);
1501 }
1502 
1503 /*
1504  * OpenBSD poll system call.
1505  *
1506  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
1507  */
1508 #ifndef _SYS_SYSPROTO_H_
1509 struct openbsd_poll_args {
1510 	struct pollfd *fds;
1511 	u_int	nfds;
1512 	int	timeout;
1513 };
1514 #endif
1515 int
1516 sys_openbsd_poll(td, uap)
1517 	register struct thread *td;
1518 	register struct openbsd_poll_args *uap;
1519 {
1520 	return (sys_poll(td, (struct poll_args *)uap));
1521 }
1522 
1523 /*
1524  * XXX This was created specifically to support netncp and netsmb.  This
1525  * allows the caller to specify a socket to wait for events on.  It returns
1526  * 0 if any events matched and an error otherwise.  There is no way to
1527  * determine which events fired.
1528  */
1529 int
1530 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1531 {
1532 	struct timeval rtv;
1533 	sbintime_t asbt, precision, rsbt;
1534 	int error;
1535 
1536 	precision = 0;	/* stupid gcc! */
1537 	if (tvp != NULL) {
1538 		rtv = *tvp;
1539 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1540 		    rtv.tv_usec >= 1000000)
1541 			return (EINVAL);
1542 		if (!timevalisset(&rtv))
1543 			asbt = 0;
1544 		else if (rtv.tv_sec <= INT32_MAX) {
1545 			rsbt = tvtosbt(rtv);
1546 			precision = rsbt;
1547 			precision >>= tc_precexp;
1548 			if (TIMESEL(&asbt, rsbt))
1549 				asbt += tc_tick_sbt;
1550 			if (asbt <= SBT_MAX - rsbt)
1551 				asbt += rsbt;
1552 			else
1553 				asbt = -1;
1554 		} else
1555 			asbt = -1;
1556 	} else
1557 		asbt = -1;
1558 	seltdinit(td);
1559 	/*
1560 	 * Iterate until the timeout expires or the socket becomes ready.
1561 	 */
1562 	for (;;) {
1563 		selfdalloc(td, NULL);
1564 		error = sopoll(so, events, NULL, td);
1565 		/* error here is actually the ready events. */
1566 		if (error)
1567 			return (0);
1568 		error = seltdwait(td, asbt, precision);
1569 		if (error)
1570 			break;
1571 	}
1572 	seltdclear(td);
1573 	/* XXX Duplicates ncp/smb behavior. */
1574 	if (error == ERESTART)
1575 		error = 0;
1576 	return (error);
1577 }
1578 
1579 /*
1580  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1581  * have two select sets, one for read and another for write.
1582  */
1583 static void
1584 selfdalloc(struct thread *td, void *cookie)
1585 {
1586 	struct seltd *stp;
1587 
1588 	stp = td->td_sel;
1589 	if (stp->st_free1 == NULL)
1590 		stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1591 	stp->st_free1->sf_td = stp;
1592 	stp->st_free1->sf_cookie = cookie;
1593 	if (stp->st_free2 == NULL)
1594 		stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1595 	stp->st_free2->sf_td = stp;
1596 	stp->st_free2->sf_cookie = cookie;
1597 }
1598 
1599 static void
1600 selfdfree(struct seltd *stp, struct selfd *sfp)
1601 {
1602 	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1603 	mtx_lock(sfp->sf_mtx);
1604 	if (sfp->sf_si)
1605 		TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1606 	mtx_unlock(sfp->sf_mtx);
1607 	uma_zfree(selfd_zone, sfp);
1608 }
1609 
1610 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1611 void
1612 seldrain(sip)
1613         struct selinfo *sip;
1614 {
1615 
1616 	/*
1617 	 * This feature is already provided by doselwakeup(), thus it is
1618 	 * enough to go for it.
1619 	 * Eventually, the context, should take care to avoid races
1620 	 * between thread calling select()/poll() and file descriptor
1621 	 * detaching, but, again, the races are just the same as
1622 	 * selwakeup().
1623 	 */
1624         doselwakeup(sip, -1);
1625 }
1626 
1627 /*
1628  * Record a select request.
1629  */
1630 void
1631 selrecord(selector, sip)
1632 	struct thread *selector;
1633 	struct selinfo *sip;
1634 {
1635 	struct selfd *sfp;
1636 	struct seltd *stp;
1637 	struct mtx *mtxp;
1638 
1639 	stp = selector->td_sel;
1640 	/*
1641 	 * Don't record when doing a rescan.
1642 	 */
1643 	if (stp->st_flags & SELTD_RESCAN)
1644 		return;
1645 	/*
1646 	 * Grab one of the preallocated descriptors.
1647 	 */
1648 	sfp = NULL;
1649 	if ((sfp = stp->st_free1) != NULL)
1650 		stp->st_free1 = NULL;
1651 	else if ((sfp = stp->st_free2) != NULL)
1652 		stp->st_free2 = NULL;
1653 	else
1654 		panic("selrecord: No free selfd on selq");
1655 	mtxp = sip->si_mtx;
1656 	if (mtxp == NULL)
1657 		mtxp = mtx_pool_find(mtxpool_select, sip);
1658 	/*
1659 	 * Initialize the sfp and queue it in the thread.
1660 	 */
1661 	sfp->sf_si = sip;
1662 	sfp->sf_mtx = mtxp;
1663 	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1664 	/*
1665 	 * Now that we've locked the sip, check for initialization.
1666 	 */
1667 	mtx_lock(mtxp);
1668 	if (sip->si_mtx == NULL) {
1669 		sip->si_mtx = mtxp;
1670 		TAILQ_INIT(&sip->si_tdlist);
1671 	}
1672 	/*
1673 	 * Add this thread to the list of selfds listening on this selinfo.
1674 	 */
1675 	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1676 	mtx_unlock(sip->si_mtx);
1677 }
1678 
1679 /* Wake up a selecting thread. */
1680 void
1681 selwakeup(sip)
1682 	struct selinfo *sip;
1683 {
1684 	doselwakeup(sip, -1);
1685 }
1686 
1687 /* Wake up a selecting thread, and set its priority. */
1688 void
1689 selwakeuppri(sip, pri)
1690 	struct selinfo *sip;
1691 	int pri;
1692 {
1693 	doselwakeup(sip, pri);
1694 }
1695 
1696 /*
1697  * Do a wakeup when a selectable event occurs.
1698  */
1699 static void
1700 doselwakeup(sip, pri)
1701 	struct selinfo *sip;
1702 	int pri;
1703 {
1704 	struct selfd *sfp;
1705 	struct selfd *sfn;
1706 	struct seltd *stp;
1707 
1708 	/* If it's not initialized there can't be any waiters. */
1709 	if (sip->si_mtx == NULL)
1710 		return;
1711 	/*
1712 	 * Locking the selinfo locks all selfds associated with it.
1713 	 */
1714 	mtx_lock(sip->si_mtx);
1715 	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1716 		/*
1717 		 * Once we remove this sfp from the list and clear the
1718 		 * sf_si seltdclear will know to ignore this si.
1719 		 */
1720 		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1721 		sfp->sf_si = NULL;
1722 		stp = sfp->sf_td;
1723 		mtx_lock(&stp->st_mtx);
1724 		stp->st_flags |= SELTD_PENDING;
1725 		cv_broadcastpri(&stp->st_wait, pri);
1726 		mtx_unlock(&stp->st_mtx);
1727 	}
1728 	mtx_unlock(sip->si_mtx);
1729 }
1730 
1731 static void
1732 seltdinit(struct thread *td)
1733 {
1734 	struct seltd *stp;
1735 
1736 	if ((stp = td->td_sel) != NULL)
1737 		goto out;
1738 	td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1739 	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1740 	cv_init(&stp->st_wait, "select");
1741 out:
1742 	stp->st_flags = 0;
1743 	STAILQ_INIT(&stp->st_selq);
1744 }
1745 
1746 static int
1747 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1748 {
1749 	struct seltd *stp;
1750 	int error;
1751 
1752 	stp = td->td_sel;
1753 	/*
1754 	 * An event of interest may occur while we do not hold the seltd
1755 	 * locked so check the pending flag before we sleep.
1756 	 */
1757 	mtx_lock(&stp->st_mtx);
1758 	/*
1759 	 * Any further calls to selrecord will be a rescan.
1760 	 */
1761 	stp->st_flags |= SELTD_RESCAN;
1762 	if (stp->st_flags & SELTD_PENDING) {
1763 		mtx_unlock(&stp->st_mtx);
1764 		return (0);
1765 	}
1766 	if (sbt == 0)
1767 		error = EWOULDBLOCK;
1768 	else if (sbt != -1)
1769 		error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
1770 		    sbt, precision, C_ABSOLUTE);
1771 	else
1772 		error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1773 	mtx_unlock(&stp->st_mtx);
1774 
1775 	return (error);
1776 }
1777 
1778 void
1779 seltdfini(struct thread *td)
1780 {
1781 	struct seltd *stp;
1782 
1783 	stp = td->td_sel;
1784 	if (stp == NULL)
1785 		return;
1786 	if (stp->st_free1)
1787 		uma_zfree(selfd_zone, stp->st_free1);
1788 	if (stp->st_free2)
1789 		uma_zfree(selfd_zone, stp->st_free2);
1790 	td->td_sel = NULL;
1791 	free(stp, M_SELECT);
1792 }
1793 
1794 /*
1795  * Remove the references to the thread from all of the objects we were
1796  * polling.
1797  */
1798 static void
1799 seltdclear(struct thread *td)
1800 {
1801 	struct seltd *stp;
1802 	struct selfd *sfp;
1803 	struct selfd *sfn;
1804 
1805 	stp = td->td_sel;
1806 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1807 		selfdfree(stp, sfp);
1808 	stp->st_flags = 0;
1809 }
1810 
1811 static void selectinit(void *);
1812 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1813 static void
1814 selectinit(void *dummy __unused)
1815 {
1816 
1817 	selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1818 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1819 	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1820 }
1821