xref: /freebsd/sys/kern/sys_generic.c (revision 7ec2f6bce5d28e6662c29e63f6ab6b7ef57d98b2)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)sys_generic.c	8.5 (Berkeley) 1/21/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/capsicum.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/lock.h>
54 #include <sys/proc.h>
55 #include <sys/signalvar.h>
56 #include <sys/socketvar.h>
57 #include <sys/uio.h>
58 #include <sys/kernel.h>
59 #include <sys/ktr.h>
60 #include <sys/limits.h>
61 #include <sys/malloc.h>
62 #include <sys/poll.h>
63 #include <sys/resourcevar.h>
64 #include <sys/selinfo.h>
65 #include <sys/sleepqueue.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/vnode.h>
70 #include <sys/bio.h>
71 #include <sys/buf.h>
72 #include <sys/condvar.h>
73 #ifdef KTRACE
74 #include <sys/ktrace.h>
75 #endif
76 
77 #include <security/audit/audit.h>
78 
79 /*
80  * The following macro defines how many bytes will be allocated from
81  * the stack instead of memory allocated when passing the IOCTL data
82  * structures from userspace and to the kernel. Some IOCTLs having
83  * small data structures are used very frequently and this small
84  * buffer on the stack gives a significant speedup improvement for
85  * those requests. The value of this define should be greater or equal
86  * to 64 bytes and should also be power of two. The data structure is
87  * currently hard-aligned to a 8-byte boundary on the stack. This
88  * should currently be sufficient for all supported platforms.
89  */
90 #define	SYS_IOCTL_SMALL_SIZE	128	/* bytes */
91 #define	SYS_IOCTL_SMALL_ALIGN	8	/* bytes */
92 
93 #ifdef __LP64__
94 static int iosize_max_clamp = 0;
95 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
96     &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
97 static int devfs_iosize_max_clamp = 1;
98 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
99     &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
100 #endif
101 
102 /*
103  * Assert that the return value of read(2) and write(2) syscalls fits
104  * into a register.  If not, an architecture will need to provide the
105  * usermode wrappers to reconstruct the result.
106  */
107 CTASSERT(sizeof(register_t) >= sizeof(size_t));
108 
109 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
110 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
111 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
112 
113 static int	pollout(struct thread *, struct pollfd *, struct pollfd *,
114 		    u_int);
115 static int	pollscan(struct thread *, struct pollfd *, u_int);
116 static int	pollrescan(struct thread *);
117 static int	selscan(struct thread *, fd_mask **, fd_mask **, int);
118 static int	selrescan(struct thread *, fd_mask **, fd_mask **);
119 static void	selfdalloc(struct thread *, void *);
120 static void	selfdfree(struct seltd *, struct selfd *);
121 static int	dofileread(struct thread *, int, struct file *, struct uio *,
122 		    off_t, int);
123 static int	dofilewrite(struct thread *, int, struct file *, struct uio *,
124 		    off_t, int);
125 static void	doselwakeup(struct selinfo *, int);
126 static void	seltdinit(struct thread *);
127 static int	seltdwait(struct thread *, sbintime_t, sbintime_t);
128 static void	seltdclear(struct thread *);
129 
130 /*
131  * One seltd per-thread allocated on demand as needed.
132  *
133  *	t - protected by st_mtx
134  * 	k - Only accessed by curthread or read-only
135  */
136 struct seltd {
137 	STAILQ_HEAD(, selfd)	st_selq;	/* (k) List of selfds. */
138 	struct selfd		*st_free1;	/* (k) free fd for read set. */
139 	struct selfd		*st_free2;	/* (k) free fd for write set. */
140 	struct mtx		st_mtx;		/* Protects struct seltd */
141 	struct cv		st_wait;	/* (t) Wait channel. */
142 	int			st_flags;	/* (t) SELTD_ flags. */
143 };
144 
145 #define	SELTD_PENDING	0x0001			/* We have pending events. */
146 #define	SELTD_RESCAN	0x0002			/* Doing a rescan. */
147 
148 /*
149  * One selfd allocated per-thread per-file-descriptor.
150  *	f - protected by sf_mtx
151  */
152 struct selfd {
153 	STAILQ_ENTRY(selfd)	sf_link;	/* (k) fds owned by this td. */
154 	TAILQ_ENTRY(selfd)	sf_threads;	/* (f) fds on this selinfo. */
155 	struct selinfo		*sf_si;		/* (f) selinfo when linked. */
156 	struct mtx		*sf_mtx;	/* Pointer to selinfo mtx. */
157 	struct seltd		*sf_td;		/* (k) owning seltd. */
158 	void			*sf_cookie;	/* (k) fd or pollfd. */
159 	u_int			sf_refs;
160 };
161 
162 static uma_zone_t selfd_zone;
163 static struct mtx_pool *mtxpool_select;
164 
165 #ifdef __LP64__
166 size_t
167 devfs_iosize_max(void)
168 {
169 
170 	return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
171 	    INT_MAX : SSIZE_MAX);
172 }
173 
174 size_t
175 iosize_max(void)
176 {
177 
178 	return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
179 	    INT_MAX : SSIZE_MAX);
180 }
181 #endif
182 
183 #ifndef _SYS_SYSPROTO_H_
184 struct read_args {
185 	int	fd;
186 	void	*buf;
187 	size_t	nbyte;
188 };
189 #endif
190 int
191 sys_read(struct thread *td, struct read_args *uap)
192 {
193 	struct uio auio;
194 	struct iovec aiov;
195 	int error;
196 
197 	if (uap->nbyte > IOSIZE_MAX)
198 		return (EINVAL);
199 	aiov.iov_base = uap->buf;
200 	aiov.iov_len = uap->nbyte;
201 	auio.uio_iov = &aiov;
202 	auio.uio_iovcnt = 1;
203 	auio.uio_resid = uap->nbyte;
204 	auio.uio_segflg = UIO_USERSPACE;
205 	error = kern_readv(td, uap->fd, &auio);
206 	return (error);
207 }
208 
209 /*
210  * Positioned read system call
211  */
212 #ifndef _SYS_SYSPROTO_H_
213 struct pread_args {
214 	int	fd;
215 	void	*buf;
216 	size_t	nbyte;
217 	int	pad;
218 	off_t	offset;
219 };
220 #endif
221 int
222 sys_pread(struct thread *td, struct pread_args *uap)
223 {
224 
225 	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
226 }
227 
228 int
229 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset)
230 {
231 	struct uio auio;
232 	struct iovec aiov;
233 	int error;
234 
235 	if (nbyte > IOSIZE_MAX)
236 		return (EINVAL);
237 	aiov.iov_base = buf;
238 	aiov.iov_len = nbyte;
239 	auio.uio_iov = &aiov;
240 	auio.uio_iovcnt = 1;
241 	auio.uio_resid = nbyte;
242 	auio.uio_segflg = UIO_USERSPACE;
243 	error = kern_preadv(td, fd, &auio, offset);
244 	return (error);
245 }
246 
247 #if defined(COMPAT_FREEBSD6)
248 int
249 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap)
250 {
251 
252 	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
253 }
254 #endif
255 
256 /*
257  * Scatter read system call.
258  */
259 #ifndef _SYS_SYSPROTO_H_
260 struct readv_args {
261 	int	fd;
262 	struct	iovec *iovp;
263 	u_int	iovcnt;
264 };
265 #endif
266 int
267 sys_readv(struct thread *td, struct readv_args *uap)
268 {
269 	struct uio *auio;
270 	int error;
271 
272 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
273 	if (error)
274 		return (error);
275 	error = kern_readv(td, uap->fd, auio);
276 	free(auio, M_IOV);
277 	return (error);
278 }
279 
280 int
281 kern_readv(struct thread *td, int fd, struct uio *auio)
282 {
283 	struct file *fp;
284 	int error;
285 
286 	error = fget_read(td, fd, &cap_read_rights, &fp);
287 	if (error)
288 		return (error);
289 	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
290 	fdrop(fp, td);
291 	return (error);
292 }
293 
294 /*
295  * Scatter positioned read system call.
296  */
297 #ifndef _SYS_SYSPROTO_H_
298 struct preadv_args {
299 	int	fd;
300 	struct	iovec *iovp;
301 	u_int	iovcnt;
302 	off_t	offset;
303 };
304 #endif
305 int
306 sys_preadv(struct thread *td, struct preadv_args *uap)
307 {
308 	struct uio *auio;
309 	int error;
310 
311 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
312 	if (error)
313 		return (error);
314 	error = kern_preadv(td, uap->fd, auio, uap->offset);
315 	free(auio, M_IOV);
316 	return (error);
317 }
318 
319 int
320 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset)
321 {
322 	struct file *fp;
323 	int error;
324 
325 	error = fget_read(td, fd, &cap_pread_rights, &fp);
326 	if (error)
327 		return (error);
328 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
329 		error = ESPIPE;
330 	else if (offset < 0 &&
331 	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
332 		error = EINVAL;
333 	else
334 		error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
335 	fdrop(fp, td);
336 	return (error);
337 }
338 
339 /*
340  * Common code for readv and preadv that reads data in
341  * from a file using the passed in uio, offset, and flags.
342  */
343 static int
344 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio,
345     off_t offset, int flags)
346 {
347 	ssize_t cnt;
348 	int error;
349 #ifdef KTRACE
350 	struct uio *ktruio = NULL;
351 #endif
352 
353 	AUDIT_ARG_FD(fd);
354 
355 	/* Finish zero length reads right here */
356 	if (auio->uio_resid == 0) {
357 		td->td_retval[0] = 0;
358 		return (0);
359 	}
360 	auio->uio_rw = UIO_READ;
361 	auio->uio_offset = offset;
362 	auio->uio_td = td;
363 #ifdef KTRACE
364 	if (KTRPOINT(td, KTR_GENIO))
365 		ktruio = cloneuio(auio);
366 #endif
367 	cnt = auio->uio_resid;
368 	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
369 		if (auio->uio_resid != cnt && (error == ERESTART ||
370 		    error == EINTR || error == EWOULDBLOCK))
371 			error = 0;
372 	}
373 	cnt -= auio->uio_resid;
374 #ifdef KTRACE
375 	if (ktruio != NULL) {
376 		ktruio->uio_resid = cnt;
377 		ktrgenio(fd, UIO_READ, ktruio, error);
378 	}
379 #endif
380 	td->td_retval[0] = cnt;
381 	return (error);
382 }
383 
384 #ifndef _SYS_SYSPROTO_H_
385 struct write_args {
386 	int	fd;
387 	const void *buf;
388 	size_t	nbyte;
389 };
390 #endif
391 int
392 sys_write(struct thread *td, struct write_args *uap)
393 {
394 	struct uio auio;
395 	struct iovec aiov;
396 	int error;
397 
398 	if (uap->nbyte > IOSIZE_MAX)
399 		return (EINVAL);
400 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
401 	aiov.iov_len = uap->nbyte;
402 	auio.uio_iov = &aiov;
403 	auio.uio_iovcnt = 1;
404 	auio.uio_resid = uap->nbyte;
405 	auio.uio_segflg = UIO_USERSPACE;
406 	error = kern_writev(td, uap->fd, &auio);
407 	return (error);
408 }
409 
410 /*
411  * Positioned write system call.
412  */
413 #ifndef _SYS_SYSPROTO_H_
414 struct pwrite_args {
415 	int	fd;
416 	const void *buf;
417 	size_t	nbyte;
418 	int	pad;
419 	off_t	offset;
420 };
421 #endif
422 int
423 sys_pwrite(struct thread *td, struct pwrite_args *uap)
424 {
425 
426 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
427 }
428 
429 int
430 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte,
431     off_t offset)
432 {
433 	struct uio auio;
434 	struct iovec aiov;
435 	int error;
436 
437 	if (nbyte > IOSIZE_MAX)
438 		return (EINVAL);
439 	aiov.iov_base = (void *)(uintptr_t)buf;
440 	aiov.iov_len = nbyte;
441 	auio.uio_iov = &aiov;
442 	auio.uio_iovcnt = 1;
443 	auio.uio_resid = nbyte;
444 	auio.uio_segflg = UIO_USERSPACE;
445 	error = kern_pwritev(td, fd, &auio, offset);
446 	return (error);
447 }
448 
449 #if defined(COMPAT_FREEBSD6)
450 int
451 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap)
452 {
453 
454 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
455 }
456 #endif
457 
458 /*
459  * Gather write system call.
460  */
461 #ifndef _SYS_SYSPROTO_H_
462 struct writev_args {
463 	int	fd;
464 	struct	iovec *iovp;
465 	u_int	iovcnt;
466 };
467 #endif
468 int
469 sys_writev(struct thread *td, struct writev_args *uap)
470 {
471 	struct uio *auio;
472 	int error;
473 
474 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
475 	if (error)
476 		return (error);
477 	error = kern_writev(td, uap->fd, auio);
478 	free(auio, M_IOV);
479 	return (error);
480 }
481 
482 int
483 kern_writev(struct thread *td, int fd, struct uio *auio)
484 {
485 	struct file *fp;
486 	int error;
487 
488 	error = fget_write(td, fd, &cap_write_rights, &fp);
489 	if (error)
490 		return (error);
491 	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
492 	fdrop(fp, td);
493 	return (error);
494 }
495 
496 /*
497  * Gather positioned write system call.
498  */
499 #ifndef _SYS_SYSPROTO_H_
500 struct pwritev_args {
501 	int	fd;
502 	struct	iovec *iovp;
503 	u_int	iovcnt;
504 	off_t	offset;
505 };
506 #endif
507 int
508 sys_pwritev(struct thread *td, struct pwritev_args *uap)
509 {
510 	struct uio *auio;
511 	int error;
512 
513 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
514 	if (error)
515 		return (error);
516 	error = kern_pwritev(td, uap->fd, auio, uap->offset);
517 	free(auio, M_IOV);
518 	return (error);
519 }
520 
521 int
522 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset)
523 {
524 	struct file *fp;
525 	int error;
526 
527 	error = fget_write(td, fd, &cap_pwrite_rights, &fp);
528 	if (error)
529 		return (error);
530 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
531 		error = ESPIPE;
532 	else if (offset < 0 &&
533 	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
534 		error = EINVAL;
535 	else
536 		error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
537 	fdrop(fp, td);
538 	return (error);
539 }
540 
541 /*
542  * Common code for writev and pwritev that writes data to
543  * a file using the passed in uio, offset, and flags.
544  */
545 static int
546 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio,
547     off_t offset, int flags)
548 {
549 	ssize_t cnt;
550 	int error;
551 #ifdef KTRACE
552 	struct uio *ktruio = NULL;
553 #endif
554 
555 	AUDIT_ARG_FD(fd);
556 	auio->uio_rw = UIO_WRITE;
557 	auio->uio_td = td;
558 	auio->uio_offset = offset;
559 #ifdef KTRACE
560 	if (KTRPOINT(td, KTR_GENIO))
561 		ktruio = cloneuio(auio);
562 #endif
563 	cnt = auio->uio_resid;
564 	if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
565 		if (auio->uio_resid != cnt && (error == ERESTART ||
566 		    error == EINTR || error == EWOULDBLOCK))
567 			error = 0;
568 		/* Socket layer is responsible for issuing SIGPIPE. */
569 		if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
570 			PROC_LOCK(td->td_proc);
571 			tdsignal(td, SIGPIPE);
572 			PROC_UNLOCK(td->td_proc);
573 		}
574 	}
575 	cnt -= auio->uio_resid;
576 #ifdef KTRACE
577 	if (ktruio != NULL) {
578 		ktruio->uio_resid = cnt;
579 		ktrgenio(fd, UIO_WRITE, ktruio, error);
580 	}
581 #endif
582 	td->td_retval[0] = cnt;
583 	return (error);
584 }
585 
586 /*
587  * Truncate a file given a file descriptor.
588  *
589  * Can't use fget_write() here, since must return EINVAL and not EBADF if the
590  * descriptor isn't writable.
591  */
592 int
593 kern_ftruncate(struct thread *td, int fd, off_t length)
594 {
595 	struct file *fp;
596 	int error;
597 
598 	AUDIT_ARG_FD(fd);
599 	if (length < 0)
600 		return (EINVAL);
601 	error = fget(td, fd, &cap_ftruncate_rights, &fp);
602 	if (error)
603 		return (error);
604 	AUDIT_ARG_FILE(td->td_proc, fp);
605 	if (!(fp->f_flag & FWRITE)) {
606 		fdrop(fp, td);
607 		return (EINVAL);
608 	}
609 	error = fo_truncate(fp, length, td->td_ucred, td);
610 	fdrop(fp, td);
611 	return (error);
612 }
613 
614 #ifndef _SYS_SYSPROTO_H_
615 struct ftruncate_args {
616 	int	fd;
617 	int	pad;
618 	off_t	length;
619 };
620 #endif
621 int
622 sys_ftruncate(struct thread *td, struct ftruncate_args *uap)
623 {
624 
625 	return (kern_ftruncate(td, uap->fd, uap->length));
626 }
627 
628 #if defined(COMPAT_43)
629 #ifndef _SYS_SYSPROTO_H_
630 struct oftruncate_args {
631 	int	fd;
632 	long	length;
633 };
634 #endif
635 int
636 oftruncate(struct thread *td, struct oftruncate_args *uap)
637 {
638 
639 	return (kern_ftruncate(td, uap->fd, uap->length));
640 }
641 #endif /* COMPAT_43 */
642 
643 #ifndef _SYS_SYSPROTO_H_
644 struct ioctl_args {
645 	int	fd;
646 	u_long	com;
647 	caddr_t	data;
648 };
649 #endif
650 /* ARGSUSED */
651 int
652 sys_ioctl(struct thread *td, struct ioctl_args *uap)
653 {
654 	u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
655 	uint32_t com;
656 	int arg, error;
657 	u_int size;
658 	caddr_t data;
659 
660 #ifdef INVARIANTS
661 	if (uap->com > 0xffffffff) {
662 		printf(
663 		    "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
664 		    td->td_proc->p_pid, td->td_name, uap->com);
665 	}
666 #endif
667 	com = (uint32_t)uap->com;
668 
669 	/*
670 	 * Interpret high order word to find amount of data to be
671 	 * copied to/from the user's address space.
672 	 */
673 	size = IOCPARM_LEN(com);
674 	if ((size > IOCPARM_MAX) ||
675 	    ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
676 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
677 	    ((com & IOC_OUT) && size == 0) ||
678 #else
679 	    ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
680 #endif
681 	    ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
682 		return (ENOTTY);
683 
684 	if (size > 0) {
685 		if (com & IOC_VOID) {
686 			/* Integer argument. */
687 			arg = (intptr_t)uap->data;
688 			data = (void *)&arg;
689 			size = 0;
690 		} else {
691 			if (size > SYS_IOCTL_SMALL_SIZE)
692 				data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
693 			else
694 				data = smalldata;
695 		}
696 	} else
697 		data = (void *)&uap->data;
698 	if (com & IOC_IN) {
699 		error = copyin(uap->data, data, (u_int)size);
700 		if (error != 0)
701 			goto out;
702 	} else if (com & IOC_OUT) {
703 		/*
704 		 * Zero the buffer so the user always
705 		 * gets back something deterministic.
706 		 */
707 		bzero(data, size);
708 	}
709 
710 	error = kern_ioctl(td, uap->fd, com, data);
711 
712 	if (error == 0 && (com & IOC_OUT))
713 		error = copyout(data, uap->data, (u_int)size);
714 
715 out:
716 	if (size > SYS_IOCTL_SMALL_SIZE)
717 		free(data, M_IOCTLOPS);
718 	return (error);
719 }
720 
721 int
722 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
723 {
724 	struct file *fp;
725 	struct filedesc *fdp;
726 	int error, tmp, locked;
727 
728 	AUDIT_ARG_FD(fd);
729 	AUDIT_ARG_CMD(com);
730 
731 	fdp = td->td_proc->p_fd;
732 
733 	switch (com) {
734 	case FIONCLEX:
735 	case FIOCLEX:
736 		FILEDESC_XLOCK(fdp);
737 		locked = LA_XLOCKED;
738 		break;
739 	default:
740 #ifdef CAPABILITIES
741 		FILEDESC_SLOCK(fdp);
742 		locked = LA_SLOCKED;
743 #else
744 		locked = LA_UNLOCKED;
745 #endif
746 		break;
747 	}
748 
749 #ifdef CAPABILITIES
750 	if ((fp = fget_locked(fdp, fd)) == NULL) {
751 		error = EBADF;
752 		goto out;
753 	}
754 	if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
755 		fp = NULL;	/* fhold() was not called yet */
756 		goto out;
757 	}
758 	if (!fhold(fp)) {
759 		error = EBADF;
760 		fp = NULL;
761 		goto out;
762 	}
763 	if (locked == LA_SLOCKED) {
764 		FILEDESC_SUNLOCK(fdp);
765 		locked = LA_UNLOCKED;
766 	}
767 #else
768 	error = fget(td, fd, &cap_ioctl_rights, &fp);
769 	if (error != 0) {
770 		fp = NULL;
771 		goto out;
772 	}
773 #endif
774 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
775 		error = EBADF;
776 		goto out;
777 	}
778 
779 	switch (com) {
780 	case FIONCLEX:
781 		fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
782 		goto out;
783 	case FIOCLEX:
784 		fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
785 		goto out;
786 	case FIONBIO:
787 		if ((tmp = *(int *)data))
788 			atomic_set_int(&fp->f_flag, FNONBLOCK);
789 		else
790 			atomic_clear_int(&fp->f_flag, FNONBLOCK);
791 		data = (void *)&tmp;
792 		break;
793 	case FIOASYNC:
794 		if ((tmp = *(int *)data))
795 			atomic_set_int(&fp->f_flag, FASYNC);
796 		else
797 			atomic_clear_int(&fp->f_flag, FASYNC);
798 		data = (void *)&tmp;
799 		break;
800 	}
801 
802 	error = fo_ioctl(fp, com, data, td->td_ucred, td);
803 out:
804 	switch (locked) {
805 	case LA_XLOCKED:
806 		FILEDESC_XUNLOCK(fdp);
807 		break;
808 #ifdef CAPABILITIES
809 	case LA_SLOCKED:
810 		FILEDESC_SUNLOCK(fdp);
811 		break;
812 #endif
813 	default:
814 		FILEDESC_UNLOCK_ASSERT(fdp);
815 		break;
816 	}
817 	if (fp != NULL)
818 		fdrop(fp, td);
819 	return (error);
820 }
821 
822 int
823 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap)
824 {
825 	int error;
826 
827 	error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len);
828 	return (kern_posix_error(td, error));
829 }
830 
831 int
832 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len)
833 {
834 	struct file *fp;
835 	int error;
836 
837 	AUDIT_ARG_FD(fd);
838 	if (offset < 0 || len <= 0)
839 		return (EINVAL);
840 	/* Check for wrap. */
841 	if (offset > OFF_MAX - len)
842 		return (EFBIG);
843 	AUDIT_ARG_FD(fd);
844 	error = fget(td, fd, &cap_pwrite_rights, &fp);
845 	if (error != 0)
846 		return (error);
847 	AUDIT_ARG_FILE(td->td_proc, fp);
848 	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
849 		error = ESPIPE;
850 		goto out;
851 	}
852 	if ((fp->f_flag & FWRITE) == 0) {
853 		error = EBADF;
854 		goto out;
855 	}
856 
857 	error = fo_fallocate(fp, offset, len, td);
858  out:
859 	fdrop(fp, td);
860 	return (error);
861 }
862 
863 int
864 poll_no_poll(int events)
865 {
866 	/*
867 	 * Return true for read/write.  If the user asked for something
868 	 * special, return POLLNVAL, so that clients have a way of
869 	 * determining reliably whether or not the extended
870 	 * functionality is present without hard-coding knowledge
871 	 * of specific filesystem implementations.
872 	 */
873 	if (events & ~POLLSTANDARD)
874 		return (POLLNVAL);
875 
876 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
877 }
878 
879 int
880 sys_pselect(struct thread *td, struct pselect_args *uap)
881 {
882 	struct timespec ts;
883 	struct timeval tv, *tvp;
884 	sigset_t set, *uset;
885 	int error;
886 
887 	if (uap->ts != NULL) {
888 		error = copyin(uap->ts, &ts, sizeof(ts));
889 		if (error != 0)
890 		    return (error);
891 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
892 		tvp = &tv;
893 	} else
894 		tvp = NULL;
895 	if (uap->sm != NULL) {
896 		error = copyin(uap->sm, &set, sizeof(set));
897 		if (error != 0)
898 			return (error);
899 		uset = &set;
900 	} else
901 		uset = NULL;
902 	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
903 	    uset, NFDBITS));
904 }
905 
906 int
907 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
908     struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
909 {
910 	int error;
911 
912 	if (uset != NULL) {
913 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
914 		    &td->td_oldsigmask, 0);
915 		if (error != 0)
916 			return (error);
917 		td->td_pflags |= TDP_OLDMASK;
918 		/*
919 		 * Make sure that ast() is called on return to
920 		 * usermode and TDP_OLDMASK is cleared, restoring old
921 		 * sigmask.
922 		 */
923 		thread_lock(td);
924 		td->td_flags |= TDF_ASTPENDING;
925 		thread_unlock(td);
926 	}
927 	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
928 	return (error);
929 }
930 
931 #ifndef _SYS_SYSPROTO_H_
932 struct select_args {
933 	int	nd;
934 	fd_set	*in, *ou, *ex;
935 	struct	timeval *tv;
936 };
937 #endif
938 int
939 sys_select(struct thread *td, struct select_args *uap)
940 {
941 	struct timeval tv, *tvp;
942 	int error;
943 
944 	if (uap->tv != NULL) {
945 		error = copyin(uap->tv, &tv, sizeof(tv));
946 		if (error)
947 			return (error);
948 		tvp = &tv;
949 	} else
950 		tvp = NULL;
951 
952 	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
953 	    NFDBITS));
954 }
955 
956 /*
957  * In the unlikely case when user specified n greater then the last
958  * open file descriptor, check that no bits are set after the last
959  * valid fd.  We must return EBADF if any is set.
960  *
961  * There are applications that rely on the behaviour.
962  *
963  * nd is fd_nfiles.
964  */
965 static int
966 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
967 {
968 	char *addr, *oaddr;
969 	int b, i, res;
970 	uint8_t bits;
971 
972 	if (nd >= ndu || fd_in == NULL)
973 		return (0);
974 
975 	oaddr = NULL;
976 	bits = 0; /* silence gcc */
977 	for (i = nd; i < ndu; i++) {
978 		b = i / NBBY;
979 #if BYTE_ORDER == LITTLE_ENDIAN
980 		addr = (char *)fd_in + b;
981 #else
982 		addr = (char *)fd_in;
983 		if (abi_nfdbits == NFDBITS) {
984 			addr += rounddown(b, sizeof(fd_mask)) +
985 			    sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
986 		} else {
987 			addr += rounddown(b, sizeof(uint32_t)) +
988 			    sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
989 		}
990 #endif
991 		if (addr != oaddr) {
992 			res = fubyte(addr);
993 			if (res == -1)
994 				return (EFAULT);
995 			oaddr = addr;
996 			bits = res;
997 		}
998 		if ((bits & (1 << (i % NBBY))) != 0)
999 			return (EBADF);
1000 	}
1001 	return (0);
1002 }
1003 
1004 int
1005 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
1006     fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
1007 {
1008 	struct filedesc *fdp;
1009 	/*
1010 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
1011 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
1012 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
1013 	 * of 256.
1014 	 */
1015 	fd_mask s_selbits[howmany(2048, NFDBITS)];
1016 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
1017 	struct timeval rtv;
1018 	sbintime_t asbt, precision, rsbt;
1019 	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
1020 	int error, lf, ndu;
1021 
1022 	if (nd < 0)
1023 		return (EINVAL);
1024 	fdp = td->td_proc->p_fd;
1025 	ndu = nd;
1026 	lf = fdp->fd_nfiles;
1027 	if (nd > lf)
1028 		nd = lf;
1029 
1030 	error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1031 	if (error != 0)
1032 		return (error);
1033 	error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1034 	if (error != 0)
1035 		return (error);
1036 	error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1037 	if (error != 0)
1038 		return (error);
1039 
1040 	/*
1041 	 * Allocate just enough bits for the non-null fd_sets.  Use the
1042 	 * preallocated auto buffer if possible.
1043 	 */
1044 	nfdbits = roundup(nd, NFDBITS);
1045 	ncpbytes = nfdbits / NBBY;
1046 	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1047 	nbufbytes = 0;
1048 	if (fd_in != NULL)
1049 		nbufbytes += 2 * ncpbytes;
1050 	if (fd_ou != NULL)
1051 		nbufbytes += 2 * ncpbytes;
1052 	if (fd_ex != NULL)
1053 		nbufbytes += 2 * ncpbytes;
1054 	if (nbufbytes <= sizeof s_selbits)
1055 		selbits = &s_selbits[0];
1056 	else
1057 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1058 
1059 	/*
1060 	 * Assign pointers into the bit buffers and fetch the input bits.
1061 	 * Put the output buffers together so that they can be bzeroed
1062 	 * together.
1063 	 */
1064 	sbp = selbits;
1065 #define	getbits(name, x) \
1066 	do {								\
1067 		if (name == NULL) {					\
1068 			ibits[x] = NULL;				\
1069 			obits[x] = NULL;				\
1070 		} else {						\
1071 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
1072 			obits[x] = sbp;					\
1073 			sbp += ncpbytes / sizeof *sbp;			\
1074 			error = copyin(name, ibits[x], ncpubytes);	\
1075 			if (error != 0)					\
1076 				goto done;				\
1077 			if (ncpbytes != ncpubytes)			\
1078 				bzero((char *)ibits[x] + ncpubytes,	\
1079 				    ncpbytes - ncpubytes);		\
1080 		}							\
1081 	} while (0)
1082 	getbits(fd_in, 0);
1083 	getbits(fd_ou, 1);
1084 	getbits(fd_ex, 2);
1085 #undef	getbits
1086 
1087 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1088 	/*
1089 	 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1090 	 * we are running under 32-bit emulation. This should be more
1091 	 * generic.
1092 	 */
1093 #define swizzle_fdset(bits)						\
1094 	if (abi_nfdbits != NFDBITS && bits != NULL) {			\
1095 		int i;							\
1096 		for (i = 0; i < ncpbytes / sizeof *sbp; i++)		\
1097 			bits[i] = (bits[i] >> 32) | (bits[i] << 32);	\
1098 	}
1099 #else
1100 #define swizzle_fdset(bits)
1101 #endif
1102 
1103 	/* Make sure the bit order makes it through an ABI transition */
1104 	swizzle_fdset(ibits[0]);
1105 	swizzle_fdset(ibits[1]);
1106 	swizzle_fdset(ibits[2]);
1107 
1108 	if (nbufbytes != 0)
1109 		bzero(selbits, nbufbytes / 2);
1110 
1111 	precision = 0;
1112 	if (tvp != NULL) {
1113 		rtv = *tvp;
1114 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1115 		    rtv.tv_usec >= 1000000) {
1116 			error = EINVAL;
1117 			goto done;
1118 		}
1119 		if (!timevalisset(&rtv))
1120 			asbt = 0;
1121 		else if (rtv.tv_sec <= INT32_MAX) {
1122 			rsbt = tvtosbt(rtv);
1123 			precision = rsbt;
1124 			precision >>= tc_precexp;
1125 			if (TIMESEL(&asbt, rsbt))
1126 				asbt += tc_tick_sbt;
1127 			if (asbt <= SBT_MAX - rsbt)
1128 				asbt += rsbt;
1129 			else
1130 				asbt = -1;
1131 		} else
1132 			asbt = -1;
1133 	} else
1134 		asbt = -1;
1135 	seltdinit(td);
1136 	/* Iterate until the timeout expires or descriptors become ready. */
1137 	for (;;) {
1138 		error = selscan(td, ibits, obits, nd);
1139 		if (error || td->td_retval[0] != 0)
1140 			break;
1141 		error = seltdwait(td, asbt, precision);
1142 		if (error)
1143 			break;
1144 		error = selrescan(td, ibits, obits);
1145 		if (error || td->td_retval[0] != 0)
1146 			break;
1147 	}
1148 	seltdclear(td);
1149 
1150 done:
1151 	/* select is not restarted after signals... */
1152 	if (error == ERESTART)
1153 		error = EINTR;
1154 	if (error == EWOULDBLOCK)
1155 		error = 0;
1156 
1157 	/* swizzle bit order back, if necessary */
1158 	swizzle_fdset(obits[0]);
1159 	swizzle_fdset(obits[1]);
1160 	swizzle_fdset(obits[2]);
1161 #undef swizzle_fdset
1162 
1163 #define	putbits(name, x) \
1164 	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1165 		error = error2;
1166 	if (error == 0) {
1167 		int error2;
1168 
1169 		putbits(fd_in, 0);
1170 		putbits(fd_ou, 1);
1171 		putbits(fd_ex, 2);
1172 #undef putbits
1173 	}
1174 	if (selbits != &s_selbits[0])
1175 		free(selbits, M_SELECT);
1176 
1177 	return (error);
1178 }
1179 /*
1180  * Convert a select bit set to poll flags.
1181  *
1182  * The backend always returns POLLHUP/POLLERR if appropriate and we
1183  * return this as a set bit in any set.
1184  */
1185 static int select_flags[3] = {
1186     POLLRDNORM | POLLHUP | POLLERR,
1187     POLLWRNORM | POLLHUP | POLLERR,
1188     POLLRDBAND | POLLERR
1189 };
1190 
1191 /*
1192  * Compute the fo_poll flags required for a fd given by the index and
1193  * bit position in the fd_mask array.
1194  */
1195 static __inline int
1196 selflags(fd_mask **ibits, int idx, fd_mask bit)
1197 {
1198 	int flags;
1199 	int msk;
1200 
1201 	flags = 0;
1202 	for (msk = 0; msk < 3; msk++) {
1203 		if (ibits[msk] == NULL)
1204 			continue;
1205 		if ((ibits[msk][idx] & bit) == 0)
1206 			continue;
1207 		flags |= select_flags[msk];
1208 	}
1209 	return (flags);
1210 }
1211 
1212 /*
1213  * Set the appropriate output bits given a mask of fired events and the
1214  * input bits originally requested.
1215  */
1216 static __inline int
1217 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1218 {
1219 	int msk;
1220 	int n;
1221 
1222 	n = 0;
1223 	for (msk = 0; msk < 3; msk++) {
1224 		if ((events & select_flags[msk]) == 0)
1225 			continue;
1226 		if (ibits[msk] == NULL)
1227 			continue;
1228 		if ((ibits[msk][idx] & bit) == 0)
1229 			continue;
1230 		/*
1231 		 * XXX Check for a duplicate set.  This can occur because a
1232 		 * socket calls selrecord() twice for each poll() call
1233 		 * resulting in two selfds per real fd.  selrescan() will
1234 		 * call selsetbits twice as a result.
1235 		 */
1236 		if ((obits[msk][idx] & bit) != 0)
1237 			continue;
1238 		obits[msk][idx] |= bit;
1239 		n++;
1240 	}
1241 
1242 	return (n);
1243 }
1244 
1245 static __inline int
1246 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1247 {
1248 
1249 	return (fget_unlocked(fdp, fd, &cap_event_rights, fpp));
1250 }
1251 
1252 /*
1253  * Traverse the list of fds attached to this thread's seltd and check for
1254  * completion.
1255  */
1256 static int
1257 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1258 {
1259 	struct filedesc *fdp;
1260 	struct selinfo *si;
1261 	struct seltd *stp;
1262 	struct selfd *sfp;
1263 	struct selfd *sfn;
1264 	struct file *fp;
1265 	fd_mask bit;
1266 	int fd, ev, n, idx;
1267 	int error;
1268 
1269 	fdp = td->td_proc->p_fd;
1270 	stp = td->td_sel;
1271 	n = 0;
1272 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1273 		fd = (int)(uintptr_t)sfp->sf_cookie;
1274 		si = sfp->sf_si;
1275 		selfdfree(stp, sfp);
1276 		/* If the selinfo wasn't cleared the event didn't fire. */
1277 		if (si != NULL)
1278 			continue;
1279 		error = getselfd_cap(fdp, fd, &fp);
1280 		if (error)
1281 			return (error);
1282 		idx = fd / NFDBITS;
1283 		bit = (fd_mask)1 << (fd % NFDBITS);
1284 		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1285 		fdrop(fp, td);
1286 		if (ev != 0)
1287 			n += selsetbits(ibits, obits, idx, bit, ev);
1288 	}
1289 	stp->st_flags = 0;
1290 	td->td_retval[0] = n;
1291 	return (0);
1292 }
1293 
1294 /*
1295  * Perform the initial filedescriptor scan and register ourselves with
1296  * each selinfo.
1297  */
1298 static int
1299 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd)
1300 {
1301 	struct filedesc *fdp;
1302 	struct file *fp;
1303 	fd_mask bit;
1304 	int ev, flags, end, fd;
1305 	int n, idx;
1306 	int error;
1307 
1308 	fdp = td->td_proc->p_fd;
1309 	n = 0;
1310 	for (idx = 0, fd = 0; fd < nfd; idx++) {
1311 		end = imin(fd + NFDBITS, nfd);
1312 		for (bit = 1; fd < end; bit <<= 1, fd++) {
1313 			/* Compute the list of events we're interested in. */
1314 			flags = selflags(ibits, idx, bit);
1315 			if (flags == 0)
1316 				continue;
1317 			error = getselfd_cap(fdp, fd, &fp);
1318 			if (error)
1319 				return (error);
1320 			selfdalloc(td, (void *)(uintptr_t)fd);
1321 			ev = fo_poll(fp, flags, td->td_ucred, td);
1322 			fdrop(fp, td);
1323 			if (ev != 0)
1324 				n += selsetbits(ibits, obits, idx, bit, ev);
1325 		}
1326 	}
1327 
1328 	td->td_retval[0] = n;
1329 	return (0);
1330 }
1331 
1332 int
1333 sys_poll(struct thread *td, struct poll_args *uap)
1334 {
1335 	struct timespec ts, *tsp;
1336 
1337 	if (uap->timeout != INFTIM) {
1338 		if (uap->timeout < 0)
1339 			return (EINVAL);
1340 		ts.tv_sec = uap->timeout / 1000;
1341 		ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1342 		tsp = &ts;
1343 	} else
1344 		tsp = NULL;
1345 
1346 	return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1347 }
1348 
1349 int
1350 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds,
1351     struct timespec *tsp, sigset_t *uset)
1352 {
1353 	struct pollfd *kfds;
1354 	struct pollfd stackfds[32];
1355 	sbintime_t sbt, precision, tmp;
1356 	time_t over;
1357 	struct timespec ts;
1358 	int error;
1359 
1360 	precision = 0;
1361 	if (tsp != NULL) {
1362 		if (tsp->tv_sec < 0)
1363 			return (EINVAL);
1364 		if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000)
1365 			return (EINVAL);
1366 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1367 			sbt = 0;
1368 		else {
1369 			ts = *tsp;
1370 			if (ts.tv_sec > INT32_MAX / 2) {
1371 				over = ts.tv_sec - INT32_MAX / 2;
1372 				ts.tv_sec -= over;
1373 			} else
1374 				over = 0;
1375 			tmp = tstosbt(ts);
1376 			precision = tmp;
1377 			precision >>= tc_precexp;
1378 			if (TIMESEL(&sbt, tmp))
1379 				sbt += tc_tick_sbt;
1380 			sbt += tmp;
1381 		}
1382 	} else
1383 		sbt = -1;
1384 
1385 	/*
1386 	 * This is kinda bogus.  We have fd limits, but that is not
1387 	 * really related to the size of the pollfd array.  Make sure
1388 	 * we let the process use at least FD_SETSIZE entries and at
1389 	 * least enough for the system-wide limits.  We want to be reasonably
1390 	 * safe, but not overly restrictive.
1391 	 */
1392 	if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1393 		return (EINVAL);
1394 	if (nfds > nitems(stackfds))
1395 		kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK);
1396 	else
1397 		kfds = stackfds;
1398 	error = copyin(ufds, kfds, nfds * sizeof(*kfds));
1399 	if (error)
1400 		goto done;
1401 
1402 	if (uset != NULL) {
1403 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
1404 		    &td->td_oldsigmask, 0);
1405 		if (error)
1406 			goto done;
1407 		td->td_pflags |= TDP_OLDMASK;
1408 		/*
1409 		 * Make sure that ast() is called on return to
1410 		 * usermode and TDP_OLDMASK is cleared, restoring old
1411 		 * sigmask.
1412 		 */
1413 		thread_lock(td);
1414 		td->td_flags |= TDF_ASTPENDING;
1415 		thread_unlock(td);
1416 	}
1417 
1418 	seltdinit(td);
1419 	/* Iterate until the timeout expires or descriptors become ready. */
1420 	for (;;) {
1421 		error = pollscan(td, kfds, nfds);
1422 		if (error || td->td_retval[0] != 0)
1423 			break;
1424 		error = seltdwait(td, sbt, precision);
1425 		if (error)
1426 			break;
1427 		error = pollrescan(td);
1428 		if (error || td->td_retval[0] != 0)
1429 			break;
1430 	}
1431 	seltdclear(td);
1432 
1433 done:
1434 	/* poll is not restarted after signals... */
1435 	if (error == ERESTART)
1436 		error = EINTR;
1437 	if (error == EWOULDBLOCK)
1438 		error = 0;
1439 	if (error == 0) {
1440 		error = pollout(td, kfds, ufds, nfds);
1441 		if (error)
1442 			goto out;
1443 	}
1444 out:
1445 	if (nfds > nitems(stackfds))
1446 		free(kfds, M_TEMP);
1447 	return (error);
1448 }
1449 
1450 int
1451 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1452 {
1453 	struct timespec ts, *tsp;
1454 	sigset_t set, *ssp;
1455 	int error;
1456 
1457 	if (uap->ts != NULL) {
1458 		error = copyin(uap->ts, &ts, sizeof(ts));
1459 		if (error)
1460 			return (error);
1461 		tsp = &ts;
1462 	} else
1463 		tsp = NULL;
1464 	if (uap->set != NULL) {
1465 		error = copyin(uap->set, &set, sizeof(set));
1466 		if (error)
1467 			return (error);
1468 		ssp = &set;
1469 	} else
1470 		ssp = NULL;
1471 	/*
1472 	 * fds is still a pointer to user space. kern_poll() will
1473 	 * take care of copyin that array to the kernel space.
1474 	 */
1475 
1476 	return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1477 }
1478 
1479 #ifdef CAPABILITIES
1480 static int
1481 poll_fget(struct filedesc *fdp, int fd, struct file **fpp)
1482 {
1483 	const struct filedescent *fde;
1484 	const struct fdescenttbl *fdt;
1485 	const cap_rights_t *haverights;
1486 	struct file *fp;
1487 	int error;
1488 
1489 	if (__predict_false(fd >= fdp->fd_nfiles))
1490 		return (EBADF);
1491 
1492 	fdt = fdp->fd_files;
1493 	fde = &fdt->fdt_ofiles[fd];
1494 	fp = fde->fde_file;
1495 	if (__predict_false(fp == NULL))
1496 		return (EBADF);
1497 	haverights = cap_rights_fde_inline(fde);
1498 	error = cap_check_inline(haverights, &cap_event_rights);
1499 	if (__predict_false(error != 0))
1500 		return (EBADF);
1501 	*fpp = fp;
1502 	return (0);
1503 }
1504 #else
1505 static int
1506 poll_fget(struct filedesc *fdp, int fd, struct file **fpp)
1507 {
1508 	struct file *fp;
1509 
1510 	if (__predict_false(fd >= fdp->fd_nfiles))
1511 		return (EBADF);
1512 
1513 	fp = fdp->fd_ofiles[fd].fde_file;
1514 	if (__predict_false(fp == NULL))
1515 		return (EBADF);
1516 
1517 	*fpp = fp;
1518 	return (0);
1519 }
1520 #endif
1521 
1522 static int
1523 pollrescan(struct thread *td)
1524 {
1525 	struct seltd *stp;
1526 	struct selfd *sfp;
1527 	struct selfd *sfn;
1528 	struct selinfo *si;
1529 	struct filedesc *fdp;
1530 	struct file *fp;
1531 	struct pollfd *fd;
1532 	int n;
1533 
1534 	n = 0;
1535 	fdp = td->td_proc->p_fd;
1536 	stp = td->td_sel;
1537 	FILEDESC_SLOCK(fdp);
1538 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1539 		fd = (struct pollfd *)sfp->sf_cookie;
1540 		si = sfp->sf_si;
1541 		selfdfree(stp, sfp);
1542 		/* If the selinfo wasn't cleared the event didn't fire. */
1543 		if (si != NULL)
1544 			continue;
1545 		if (poll_fget(fdp, fd->fd, &fp) != 0) {
1546 			fd->revents = POLLNVAL;
1547 			n++;
1548 			continue;
1549 		}
1550 		/*
1551 		 * Note: backend also returns POLLHUP and
1552 		 * POLLERR if appropriate.
1553 		 */
1554 		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1555 		if (fd->revents != 0)
1556 			n++;
1557 	}
1558 	FILEDESC_SUNLOCK(fdp);
1559 	stp->st_flags = 0;
1560 	td->td_retval[0] = n;
1561 	return (0);
1562 }
1563 
1564 static int
1565 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd)
1566 {
1567 	int error = 0;
1568 	u_int i = 0;
1569 	u_int n = 0;
1570 
1571 	for (i = 0; i < nfd; i++) {
1572 		error = copyout(&fds->revents, &ufds->revents,
1573 		    sizeof(ufds->revents));
1574 		if (error)
1575 			return (error);
1576 		if (fds->revents != 0)
1577 			n++;
1578 		fds++;
1579 		ufds++;
1580 	}
1581 	td->td_retval[0] = n;
1582 	return (0);
1583 }
1584 
1585 static int
1586 pollscan(struct thread *td, struct pollfd *fds, u_int nfd)
1587 {
1588 	struct filedesc *fdp;
1589 	struct file *fp;
1590 	int i, n;
1591 
1592 	n = 0;
1593 	fdp = td->td_proc->p_fd;
1594 	FILEDESC_SLOCK(fdp);
1595 	for (i = 0; i < nfd; i++, fds++) {
1596 		if (fds->fd < 0) {
1597 			fds->revents = 0;
1598 			continue;
1599 		}
1600 		if (poll_fget(fdp, fds->fd, &fp) != 0) {
1601 			fds->revents = POLLNVAL;
1602 			n++;
1603 			continue;
1604 		}
1605 		/*
1606 		 * Note: backend also returns POLLHUP and
1607 		 * POLLERR if appropriate.
1608 		 */
1609 		selfdalloc(td, fds);
1610 		fds->revents = fo_poll(fp, fds->events,
1611 		    td->td_ucred, td);
1612 		/*
1613 		 * POSIX requires POLLOUT to be never
1614 		 * set simultaneously with POLLHUP.
1615 		 */
1616 		if ((fds->revents & POLLHUP) != 0)
1617 			fds->revents &= ~POLLOUT;
1618 
1619 		if (fds->revents != 0)
1620 			n++;
1621 	}
1622 	FILEDESC_SUNLOCK(fdp);
1623 	td->td_retval[0] = n;
1624 	return (0);
1625 }
1626 
1627 /*
1628  * XXX This was created specifically to support netncp and netsmb.  This
1629  * allows the caller to specify a socket to wait for events on.  It returns
1630  * 0 if any events matched and an error otherwise.  There is no way to
1631  * determine which events fired.
1632  */
1633 int
1634 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1635 {
1636 	struct timeval rtv;
1637 	sbintime_t asbt, precision, rsbt;
1638 	int error;
1639 
1640 	precision = 0;	/* stupid gcc! */
1641 	if (tvp != NULL) {
1642 		rtv = *tvp;
1643 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1644 		    rtv.tv_usec >= 1000000)
1645 			return (EINVAL);
1646 		if (!timevalisset(&rtv))
1647 			asbt = 0;
1648 		else if (rtv.tv_sec <= INT32_MAX) {
1649 			rsbt = tvtosbt(rtv);
1650 			precision = rsbt;
1651 			precision >>= tc_precexp;
1652 			if (TIMESEL(&asbt, rsbt))
1653 				asbt += tc_tick_sbt;
1654 			if (asbt <= SBT_MAX - rsbt)
1655 				asbt += rsbt;
1656 			else
1657 				asbt = -1;
1658 		} else
1659 			asbt = -1;
1660 	} else
1661 		asbt = -1;
1662 	seltdinit(td);
1663 	/*
1664 	 * Iterate until the timeout expires or the socket becomes ready.
1665 	 */
1666 	for (;;) {
1667 		selfdalloc(td, NULL);
1668 		error = sopoll(so, events, NULL, td);
1669 		/* error here is actually the ready events. */
1670 		if (error)
1671 			return (0);
1672 		error = seltdwait(td, asbt, precision);
1673 		if (error)
1674 			break;
1675 	}
1676 	seltdclear(td);
1677 	/* XXX Duplicates ncp/smb behavior. */
1678 	if (error == ERESTART)
1679 		error = 0;
1680 	return (error);
1681 }
1682 
1683 /*
1684  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1685  * have two select sets, one for read and another for write.
1686  */
1687 static void
1688 selfdalloc(struct thread *td, void *cookie)
1689 {
1690 	struct seltd *stp;
1691 
1692 	stp = td->td_sel;
1693 	if (stp->st_free1 == NULL)
1694 		stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1695 	stp->st_free1->sf_td = stp;
1696 	stp->st_free1->sf_cookie = cookie;
1697 	if (stp->st_free2 == NULL)
1698 		stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1699 	stp->st_free2->sf_td = stp;
1700 	stp->st_free2->sf_cookie = cookie;
1701 }
1702 
1703 static void
1704 selfdfree(struct seltd *stp, struct selfd *sfp)
1705 {
1706 	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1707 	if (sfp->sf_si != NULL) {
1708 		mtx_lock(sfp->sf_mtx);
1709 		if (sfp->sf_si != NULL) {
1710 			TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1711 			refcount_release(&sfp->sf_refs);
1712 		}
1713 		mtx_unlock(sfp->sf_mtx);
1714 	}
1715 	if (refcount_release(&sfp->sf_refs))
1716 		uma_zfree(selfd_zone, sfp);
1717 }
1718 
1719 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1720 void
1721 seldrain(struct selinfo *sip)
1722 {
1723 
1724 	/*
1725 	 * This feature is already provided by doselwakeup(), thus it is
1726 	 * enough to go for it.
1727 	 * Eventually, the context, should take care to avoid races
1728 	 * between thread calling select()/poll() and file descriptor
1729 	 * detaching, but, again, the races are just the same as
1730 	 * selwakeup().
1731 	 */
1732         doselwakeup(sip, -1);
1733 }
1734 
1735 /*
1736  * Record a select request.
1737  */
1738 void
1739 selrecord(struct thread *selector, struct selinfo *sip)
1740 {
1741 	struct selfd *sfp;
1742 	struct seltd *stp;
1743 	struct mtx *mtxp;
1744 
1745 	stp = selector->td_sel;
1746 	/*
1747 	 * Don't record when doing a rescan.
1748 	 */
1749 	if (stp->st_flags & SELTD_RESCAN)
1750 		return;
1751 	/*
1752 	 * Grab one of the preallocated descriptors.
1753 	 */
1754 	sfp = NULL;
1755 	if ((sfp = stp->st_free1) != NULL)
1756 		stp->st_free1 = NULL;
1757 	else if ((sfp = stp->st_free2) != NULL)
1758 		stp->st_free2 = NULL;
1759 	else
1760 		panic("selrecord: No free selfd on selq");
1761 	mtxp = sip->si_mtx;
1762 	if (mtxp == NULL)
1763 		mtxp = mtx_pool_find(mtxpool_select, sip);
1764 	/*
1765 	 * Initialize the sfp and queue it in the thread.
1766 	 */
1767 	sfp->sf_si = sip;
1768 	sfp->sf_mtx = mtxp;
1769 	refcount_init(&sfp->sf_refs, 2);
1770 	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1771 	/*
1772 	 * Now that we've locked the sip, check for initialization.
1773 	 */
1774 	mtx_lock(mtxp);
1775 	if (sip->si_mtx == NULL) {
1776 		sip->si_mtx = mtxp;
1777 		TAILQ_INIT(&sip->si_tdlist);
1778 	}
1779 	/*
1780 	 * Add this thread to the list of selfds listening on this selinfo.
1781 	 */
1782 	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1783 	mtx_unlock(sip->si_mtx);
1784 }
1785 
1786 /* Wake up a selecting thread. */
1787 void
1788 selwakeup(struct selinfo *sip)
1789 {
1790 	doselwakeup(sip, -1);
1791 }
1792 
1793 /* Wake up a selecting thread, and set its priority. */
1794 void
1795 selwakeuppri(struct selinfo *sip, int pri)
1796 {
1797 	doselwakeup(sip, pri);
1798 }
1799 
1800 /*
1801  * Do a wakeup when a selectable event occurs.
1802  */
1803 static void
1804 doselwakeup(struct selinfo *sip, int pri)
1805 {
1806 	struct selfd *sfp;
1807 	struct selfd *sfn;
1808 	struct seltd *stp;
1809 
1810 	/* If it's not initialized there can't be any waiters. */
1811 	if (sip->si_mtx == NULL)
1812 		return;
1813 	/*
1814 	 * Locking the selinfo locks all selfds associated with it.
1815 	 */
1816 	mtx_lock(sip->si_mtx);
1817 	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1818 		/*
1819 		 * Once we remove this sfp from the list and clear the
1820 		 * sf_si seltdclear will know to ignore this si.
1821 		 */
1822 		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1823 		sfp->sf_si = NULL;
1824 		stp = sfp->sf_td;
1825 		mtx_lock(&stp->st_mtx);
1826 		stp->st_flags |= SELTD_PENDING;
1827 		cv_broadcastpri(&stp->st_wait, pri);
1828 		mtx_unlock(&stp->st_mtx);
1829 		if (refcount_release(&sfp->sf_refs))
1830 			uma_zfree(selfd_zone, sfp);
1831 	}
1832 	mtx_unlock(sip->si_mtx);
1833 }
1834 
1835 static void
1836 seltdinit(struct thread *td)
1837 {
1838 	struct seltd *stp;
1839 
1840 	if ((stp = td->td_sel) != NULL)
1841 		goto out;
1842 	td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1843 	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1844 	cv_init(&stp->st_wait, "select");
1845 out:
1846 	stp->st_flags = 0;
1847 	STAILQ_INIT(&stp->st_selq);
1848 }
1849 
1850 static int
1851 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1852 {
1853 	struct seltd *stp;
1854 	int error;
1855 
1856 	stp = td->td_sel;
1857 	/*
1858 	 * An event of interest may occur while we do not hold the seltd
1859 	 * locked so check the pending flag before we sleep.
1860 	 */
1861 	mtx_lock(&stp->st_mtx);
1862 	/*
1863 	 * Any further calls to selrecord will be a rescan.
1864 	 */
1865 	stp->st_flags |= SELTD_RESCAN;
1866 	if (stp->st_flags & SELTD_PENDING) {
1867 		mtx_unlock(&stp->st_mtx);
1868 		return (0);
1869 	}
1870 	if (sbt == 0)
1871 		error = EWOULDBLOCK;
1872 	else if (sbt != -1)
1873 		error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
1874 		    sbt, precision, C_ABSOLUTE);
1875 	else
1876 		error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1877 	mtx_unlock(&stp->st_mtx);
1878 
1879 	return (error);
1880 }
1881 
1882 void
1883 seltdfini(struct thread *td)
1884 {
1885 	struct seltd *stp;
1886 
1887 	stp = td->td_sel;
1888 	if (stp == NULL)
1889 		return;
1890 	if (stp->st_free1)
1891 		uma_zfree(selfd_zone, stp->st_free1);
1892 	if (stp->st_free2)
1893 		uma_zfree(selfd_zone, stp->st_free2);
1894 	td->td_sel = NULL;
1895 	cv_destroy(&stp->st_wait);
1896 	mtx_destroy(&stp->st_mtx);
1897 	free(stp, M_SELECT);
1898 }
1899 
1900 /*
1901  * Remove the references to the thread from all of the objects we were
1902  * polling.
1903  */
1904 static void
1905 seltdclear(struct thread *td)
1906 {
1907 	struct seltd *stp;
1908 	struct selfd *sfp;
1909 	struct selfd *sfn;
1910 
1911 	stp = td->td_sel;
1912 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1913 		selfdfree(stp, sfp);
1914 	stp->st_flags = 0;
1915 }
1916 
1917 static void selectinit(void *);
1918 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1919 static void
1920 selectinit(void *dummy __unused)
1921 {
1922 
1923 	selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1924 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1925 	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1926 }
1927 
1928 /*
1929  * Set up a syscall return value that follows the convention specified for
1930  * posix_* functions.
1931  */
1932 int
1933 kern_posix_error(struct thread *td, int error)
1934 {
1935 
1936 	if (error <= 0)
1937 		return (error);
1938 	td->td_errno = error;
1939 	td->td_pflags |= TDP_NERRNO;
1940 	td->td_retval[0] = error;
1941 	return (0);
1942 }
1943