xref: /freebsd/sys/kern/sys_generic.c (revision 3cbb4cc200f8a0ad7ed08233425ea54524a21f1c)
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 };
160 
161 MALLOC_DEFINE(M_SELFD, "selfd", "selfd");
162 static struct mtx_pool *mtxpool_select;
163 
164 #ifdef __LP64__
165 size_t
166 devfs_iosize_max(void)
167 {
168 
169 	return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
170 	    INT_MAX : SSIZE_MAX);
171 }
172 
173 size_t
174 iosize_max(void)
175 {
176 
177 	return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
178 	    INT_MAX : SSIZE_MAX);
179 }
180 #endif
181 
182 #ifndef _SYS_SYSPROTO_H_
183 struct read_args {
184 	int	fd;
185 	void	*buf;
186 	size_t	nbyte;
187 };
188 #endif
189 int
190 sys_read(struct thread *td, struct read_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_readv(td, uap->fd, &auio);
205 	return (error);
206 }
207 
208 /*
209  * Positioned read system call
210  */
211 #ifndef _SYS_SYSPROTO_H_
212 struct pread_args {
213 	int	fd;
214 	void	*buf;
215 	size_t	nbyte;
216 	int	pad;
217 	off_t	offset;
218 };
219 #endif
220 int
221 sys_pread(struct thread *td, struct pread_args *uap)
222 {
223 
224 	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
225 }
226 
227 int
228 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset)
229 {
230 	struct uio auio;
231 	struct iovec aiov;
232 	int error;
233 
234 	if (nbyte > IOSIZE_MAX)
235 		return (EINVAL);
236 	aiov.iov_base = buf;
237 	aiov.iov_len = nbyte;
238 	auio.uio_iov = &aiov;
239 	auio.uio_iovcnt = 1;
240 	auio.uio_resid = nbyte;
241 	auio.uio_segflg = UIO_USERSPACE;
242 	error = kern_preadv(td, fd, &auio, offset);
243 	return (error);
244 }
245 
246 #if defined(COMPAT_FREEBSD6)
247 int
248 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap)
249 {
250 
251 	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
252 }
253 #endif
254 
255 /*
256  * Scatter read system call.
257  */
258 #ifndef _SYS_SYSPROTO_H_
259 struct readv_args {
260 	int	fd;
261 	struct	iovec *iovp;
262 	u_int	iovcnt;
263 };
264 #endif
265 int
266 sys_readv(struct thread *td, struct readv_args *uap)
267 {
268 	struct uio *auio;
269 	int error;
270 
271 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
272 	if (error)
273 		return (error);
274 	error = kern_readv(td, uap->fd, auio);
275 	free(auio, M_IOV);
276 	return (error);
277 }
278 
279 int
280 kern_readv(struct thread *td, int fd, struct uio *auio)
281 {
282 	struct file *fp;
283 	int error;
284 
285 	error = fget_read(td, fd, &cap_read_rights, &fp);
286 	if (error)
287 		return (error);
288 	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
289 	fdrop(fp, td);
290 	return (error);
291 }
292 
293 /*
294  * Scatter positioned read system call.
295  */
296 #ifndef _SYS_SYSPROTO_H_
297 struct preadv_args {
298 	int	fd;
299 	struct	iovec *iovp;
300 	u_int	iovcnt;
301 	off_t	offset;
302 };
303 #endif
304 int
305 sys_preadv(struct thread *td, struct preadv_args *uap)
306 {
307 	struct uio *auio;
308 	int error;
309 
310 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
311 	if (error)
312 		return (error);
313 	error = kern_preadv(td, uap->fd, auio, uap->offset);
314 	free(auio, M_IOV);
315 	return (error);
316 }
317 
318 int
319 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset)
320 {
321 	struct file *fp;
322 	int error;
323 
324 	error = fget_read(td, fd, &cap_pread_rights, &fp);
325 	if (error)
326 		return (error);
327 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
328 		error = ESPIPE;
329 	else if (offset < 0 &&
330 	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
331 		error = EINVAL;
332 	else
333 		error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
334 	fdrop(fp, td);
335 	return (error);
336 }
337 
338 /*
339  * Common code for readv and preadv that reads data in
340  * from a file using the passed in uio, offset, and flags.
341  */
342 static int
343 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio,
344     off_t offset, int flags)
345 {
346 	ssize_t cnt;
347 	int error;
348 #ifdef KTRACE
349 	struct uio *ktruio = NULL;
350 #endif
351 
352 	AUDIT_ARG_FD(fd);
353 
354 	/* Finish zero length reads right here */
355 	if (auio->uio_resid == 0) {
356 		td->td_retval[0] = 0;
357 		return (0);
358 	}
359 	auio->uio_rw = UIO_READ;
360 	auio->uio_offset = offset;
361 	auio->uio_td = td;
362 #ifdef KTRACE
363 	if (KTRPOINT(td, KTR_GENIO))
364 		ktruio = cloneuio(auio);
365 #endif
366 	cnt = auio->uio_resid;
367 	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
368 		if (auio->uio_resid != cnt && (error == ERESTART ||
369 		    error == EINTR || error == EWOULDBLOCK))
370 			error = 0;
371 	}
372 	cnt -= auio->uio_resid;
373 #ifdef KTRACE
374 	if (ktruio != NULL) {
375 		ktruio->uio_resid = cnt;
376 		ktrgenio(fd, UIO_READ, ktruio, error);
377 	}
378 #endif
379 	td->td_retval[0] = cnt;
380 	return (error);
381 }
382 
383 #ifndef _SYS_SYSPROTO_H_
384 struct write_args {
385 	int	fd;
386 	const void *buf;
387 	size_t	nbyte;
388 };
389 #endif
390 int
391 sys_write(struct thread *td, struct write_args *uap)
392 {
393 	struct uio auio;
394 	struct iovec aiov;
395 	int error;
396 
397 	if (uap->nbyte > IOSIZE_MAX)
398 		return (EINVAL);
399 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
400 	aiov.iov_len = uap->nbyte;
401 	auio.uio_iov = &aiov;
402 	auio.uio_iovcnt = 1;
403 	auio.uio_resid = uap->nbyte;
404 	auio.uio_segflg = UIO_USERSPACE;
405 	error = kern_writev(td, uap->fd, &auio);
406 	return (error);
407 }
408 
409 /*
410  * Positioned write system call.
411  */
412 #ifndef _SYS_SYSPROTO_H_
413 struct pwrite_args {
414 	int	fd;
415 	const void *buf;
416 	size_t	nbyte;
417 	int	pad;
418 	off_t	offset;
419 };
420 #endif
421 int
422 sys_pwrite(struct thread *td, struct pwrite_args *uap)
423 {
424 
425 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
426 }
427 
428 int
429 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte,
430     off_t offset)
431 {
432 	struct uio auio;
433 	struct iovec aiov;
434 	int error;
435 
436 	if (nbyte > IOSIZE_MAX)
437 		return (EINVAL);
438 	aiov.iov_base = (void *)(uintptr_t)buf;
439 	aiov.iov_len = nbyte;
440 	auio.uio_iov = &aiov;
441 	auio.uio_iovcnt = 1;
442 	auio.uio_resid = nbyte;
443 	auio.uio_segflg = UIO_USERSPACE;
444 	error = kern_pwritev(td, fd, &auio, offset);
445 	return (error);
446 }
447 
448 #if defined(COMPAT_FREEBSD6)
449 int
450 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap)
451 {
452 
453 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
454 }
455 #endif
456 
457 /*
458  * Gather write system call.
459  */
460 #ifndef _SYS_SYSPROTO_H_
461 struct writev_args {
462 	int	fd;
463 	struct	iovec *iovp;
464 	u_int	iovcnt;
465 };
466 #endif
467 int
468 sys_writev(struct thread *td, struct writev_args *uap)
469 {
470 	struct uio *auio;
471 	int error;
472 
473 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
474 	if (error)
475 		return (error);
476 	error = kern_writev(td, uap->fd, auio);
477 	free(auio, M_IOV);
478 	return (error);
479 }
480 
481 int
482 kern_writev(struct thread *td, int fd, struct uio *auio)
483 {
484 	struct file *fp;
485 	int error;
486 
487 	error = fget_write(td, fd, &cap_write_rights, &fp);
488 	if (error)
489 		return (error);
490 	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
491 	fdrop(fp, td);
492 	return (error);
493 }
494 
495 /*
496  * Gather positioned write system call.
497  */
498 #ifndef _SYS_SYSPROTO_H_
499 struct pwritev_args {
500 	int	fd;
501 	struct	iovec *iovp;
502 	u_int	iovcnt;
503 	off_t	offset;
504 };
505 #endif
506 int
507 sys_pwritev(struct thread *td, struct pwritev_args *uap)
508 {
509 	struct uio *auio;
510 	int error;
511 
512 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
513 	if (error)
514 		return (error);
515 	error = kern_pwritev(td, uap->fd, auio, uap->offset);
516 	free(auio, M_IOV);
517 	return (error);
518 }
519 
520 int
521 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset)
522 {
523 	struct file *fp;
524 	int error;
525 
526 	error = fget_write(td, fd, &cap_pwrite_rights, &fp);
527 	if (error)
528 		return (error);
529 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
530 		error = ESPIPE;
531 	else if (offset < 0 &&
532 	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
533 		error = EINVAL;
534 	else
535 		error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
536 	fdrop(fp, td);
537 	return (error);
538 }
539 
540 /*
541  * Common code for writev and pwritev that writes data to
542  * a file using the passed in uio, offset, and flags.
543  */
544 static int
545 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio,
546     off_t offset, int flags)
547 {
548 	ssize_t cnt;
549 	int error;
550 #ifdef KTRACE
551 	struct uio *ktruio = NULL;
552 #endif
553 
554 	AUDIT_ARG_FD(fd);
555 	auio->uio_rw = UIO_WRITE;
556 	auio->uio_td = td;
557 	auio->uio_offset = offset;
558 #ifdef KTRACE
559 	if (KTRPOINT(td, KTR_GENIO))
560 		ktruio = cloneuio(auio);
561 #endif
562 	cnt = auio->uio_resid;
563 	if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
564 		if (auio->uio_resid != cnt && (error == ERESTART ||
565 		    error == EINTR || error == EWOULDBLOCK))
566 			error = 0;
567 		/* Socket layer is responsible for issuing SIGPIPE. */
568 		if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
569 			PROC_LOCK(td->td_proc);
570 			tdsignal(td, SIGPIPE);
571 			PROC_UNLOCK(td->td_proc);
572 		}
573 	}
574 	cnt -= auio->uio_resid;
575 #ifdef KTRACE
576 	if (ktruio != NULL) {
577 		ktruio->uio_resid = cnt;
578 		ktrgenio(fd, UIO_WRITE, ktruio, error);
579 	}
580 #endif
581 	td->td_retval[0] = cnt;
582 	return (error);
583 }
584 
585 /*
586  * Truncate a file given a file descriptor.
587  *
588  * Can't use fget_write() here, since must return EINVAL and not EBADF if the
589  * descriptor isn't writable.
590  */
591 int
592 kern_ftruncate(struct thread *td, int fd, off_t length)
593 {
594 	struct file *fp;
595 	int error;
596 
597 	AUDIT_ARG_FD(fd);
598 	if (length < 0)
599 		return (EINVAL);
600 	error = fget(td, fd, &cap_ftruncate_rights, &fp);
601 	if (error)
602 		return (error);
603 	AUDIT_ARG_FILE(td->td_proc, fp);
604 	if (!(fp->f_flag & FWRITE)) {
605 		fdrop(fp, td);
606 		return (EINVAL);
607 	}
608 	error = fo_truncate(fp, length, td->td_ucred, td);
609 	fdrop(fp, td);
610 	return (error);
611 }
612 
613 #ifndef _SYS_SYSPROTO_H_
614 struct ftruncate_args {
615 	int	fd;
616 	int	pad;
617 	off_t	length;
618 };
619 #endif
620 int
621 sys_ftruncate(struct thread *td, struct ftruncate_args *uap)
622 {
623 
624 	return (kern_ftruncate(td, uap->fd, uap->length));
625 }
626 
627 #if defined(COMPAT_43)
628 #ifndef _SYS_SYSPROTO_H_
629 struct oftruncate_args {
630 	int	fd;
631 	long	length;
632 };
633 #endif
634 int
635 oftruncate(struct thread *td, struct oftruncate_args *uap)
636 {
637 
638 	return (kern_ftruncate(td, uap->fd, uap->length));
639 }
640 #endif /* COMPAT_43 */
641 
642 #ifndef _SYS_SYSPROTO_H_
643 struct ioctl_args {
644 	int	fd;
645 	u_long	com;
646 	caddr_t	data;
647 };
648 #endif
649 /* ARGSUSED */
650 int
651 sys_ioctl(struct thread *td, struct ioctl_args *uap)
652 {
653 	u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
654 	uint32_t com;
655 	int arg, error;
656 	u_int size;
657 	caddr_t data;
658 
659 #ifdef INVARIANTS
660 	if (uap->com > 0xffffffff) {
661 		printf(
662 		    "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
663 		    td->td_proc->p_pid, td->td_name, uap->com);
664 	}
665 #endif
666 	com = (uint32_t)uap->com;
667 
668 	/*
669 	 * Interpret high order word to find amount of data to be
670 	 * copied to/from the user's address space.
671 	 */
672 	size = IOCPARM_LEN(com);
673 	if ((size > IOCPARM_MAX) ||
674 	    ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
675 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
676 	    ((com & IOC_OUT) && size == 0) ||
677 #else
678 	    ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
679 #endif
680 	    ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
681 		return (ENOTTY);
682 
683 	if (size > 0) {
684 		if (com & IOC_VOID) {
685 			/* Integer argument. */
686 			arg = (intptr_t)uap->data;
687 			data = (void *)&arg;
688 			size = 0;
689 		} else {
690 			if (size > SYS_IOCTL_SMALL_SIZE)
691 				data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
692 			else
693 				data = smalldata;
694 		}
695 	} else
696 		data = (void *)&uap->data;
697 	if (com & IOC_IN) {
698 		error = copyin(uap->data, data, (u_int)size);
699 		if (error != 0)
700 			goto out;
701 	} else if (com & IOC_OUT) {
702 		/*
703 		 * Zero the buffer so the user always
704 		 * gets back something deterministic.
705 		 */
706 		bzero(data, size);
707 	}
708 
709 	error = kern_ioctl(td, uap->fd, com, data);
710 
711 	if (error == 0 && (com & IOC_OUT))
712 		error = copyout(data, uap->data, (u_int)size);
713 
714 out:
715 	if (size > SYS_IOCTL_SMALL_SIZE)
716 		free(data, M_IOCTLOPS);
717 	return (error);
718 }
719 
720 int
721 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
722 {
723 	struct file *fp;
724 	struct filedesc *fdp;
725 	int error, tmp, locked;
726 
727 	AUDIT_ARG_FD(fd);
728 	AUDIT_ARG_CMD(com);
729 
730 	fdp = td->td_proc->p_fd;
731 
732 	switch (com) {
733 	case FIONCLEX:
734 	case FIOCLEX:
735 		FILEDESC_XLOCK(fdp);
736 		locked = LA_XLOCKED;
737 		break;
738 	default:
739 #ifdef CAPABILITIES
740 		FILEDESC_SLOCK(fdp);
741 		locked = LA_SLOCKED;
742 #else
743 		locked = LA_UNLOCKED;
744 #endif
745 		break;
746 	}
747 
748 #ifdef CAPABILITIES
749 	if ((fp = fget_locked(fdp, fd)) == NULL) {
750 		error = EBADF;
751 		goto out;
752 	}
753 	if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
754 		fp = NULL;	/* fhold() was not called yet */
755 		goto out;
756 	}
757 	if (!fhold(fp)) {
758 		error = EBADF;
759 		fp = NULL;
760 		goto out;
761 	}
762 	if (locked == LA_SLOCKED) {
763 		FILEDESC_SUNLOCK(fdp);
764 		locked = LA_UNLOCKED;
765 	}
766 #else
767 	error = fget(td, fd, &cap_ioctl_rights, &fp);
768 	if (error != 0) {
769 		fp = NULL;
770 		goto out;
771 	}
772 #endif
773 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
774 		error = EBADF;
775 		goto out;
776 	}
777 
778 	switch (com) {
779 	case FIONCLEX:
780 		fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
781 		goto out;
782 	case FIOCLEX:
783 		fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
784 		goto out;
785 	case FIONBIO:
786 		if ((tmp = *(int *)data))
787 			atomic_set_int(&fp->f_flag, FNONBLOCK);
788 		else
789 			atomic_clear_int(&fp->f_flag, FNONBLOCK);
790 		data = (void *)&tmp;
791 		break;
792 	case FIOASYNC:
793 		if ((tmp = *(int *)data))
794 			atomic_set_int(&fp->f_flag, FASYNC);
795 		else
796 			atomic_clear_int(&fp->f_flag, FASYNC);
797 		data = (void *)&tmp;
798 		break;
799 	}
800 
801 	error = fo_ioctl(fp, com, data, td->td_ucred, td);
802 out:
803 	switch (locked) {
804 	case LA_XLOCKED:
805 		FILEDESC_XUNLOCK(fdp);
806 		break;
807 #ifdef CAPABILITIES
808 	case LA_SLOCKED:
809 		FILEDESC_SUNLOCK(fdp);
810 		break;
811 #endif
812 	default:
813 		FILEDESC_UNLOCK_ASSERT(fdp);
814 		break;
815 	}
816 	if (fp != NULL)
817 		fdrop(fp, td);
818 	return (error);
819 }
820 
821 int
822 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap)
823 {
824 	int error;
825 
826 	error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len);
827 	return (kern_posix_error(td, error));
828 }
829 
830 int
831 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len)
832 {
833 	struct file *fp;
834 	int error;
835 
836 	AUDIT_ARG_FD(fd);
837 	if (offset < 0 || len <= 0)
838 		return (EINVAL);
839 	/* Check for wrap. */
840 	if (offset > OFF_MAX - len)
841 		return (EFBIG);
842 	AUDIT_ARG_FD(fd);
843 	error = fget(td, fd, &cap_pwrite_rights, &fp);
844 	if (error != 0)
845 		return (error);
846 	AUDIT_ARG_FILE(td->td_proc, fp);
847 	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
848 		error = ESPIPE;
849 		goto out;
850 	}
851 	if ((fp->f_flag & FWRITE) == 0) {
852 		error = EBADF;
853 		goto out;
854 	}
855 
856 	error = fo_fallocate(fp, offset, len, td);
857  out:
858 	fdrop(fp, td);
859 	return (error);
860 }
861 
862 int
863 poll_no_poll(int events)
864 {
865 	/*
866 	 * Return true for read/write.  If the user asked for something
867 	 * special, return POLLNVAL, so that clients have a way of
868 	 * determining reliably whether or not the extended
869 	 * functionality is present without hard-coding knowledge
870 	 * of specific filesystem implementations.
871 	 */
872 	if (events & ~POLLSTANDARD)
873 		return (POLLNVAL);
874 
875 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
876 }
877 
878 int
879 sys_pselect(struct thread *td, struct pselect_args *uap)
880 {
881 	struct timespec ts;
882 	struct timeval tv, *tvp;
883 	sigset_t set, *uset;
884 	int error;
885 
886 	if (uap->ts != NULL) {
887 		error = copyin(uap->ts, &ts, sizeof(ts));
888 		if (error != 0)
889 		    return (error);
890 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
891 		tvp = &tv;
892 	} else
893 		tvp = NULL;
894 	if (uap->sm != NULL) {
895 		error = copyin(uap->sm, &set, sizeof(set));
896 		if (error != 0)
897 			return (error);
898 		uset = &set;
899 	} else
900 		uset = NULL;
901 	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
902 	    uset, NFDBITS));
903 }
904 
905 int
906 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
907     struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
908 {
909 	int error;
910 
911 	if (uset != NULL) {
912 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
913 		    &td->td_oldsigmask, 0);
914 		if (error != 0)
915 			return (error);
916 		td->td_pflags |= TDP_OLDMASK;
917 		/*
918 		 * Make sure that ast() is called on return to
919 		 * usermode and TDP_OLDMASK is cleared, restoring old
920 		 * sigmask.
921 		 */
922 		thread_lock(td);
923 		td->td_flags |= TDF_ASTPENDING;
924 		thread_unlock(td);
925 	}
926 	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
927 	return (error);
928 }
929 
930 #ifndef _SYS_SYSPROTO_H_
931 struct select_args {
932 	int	nd;
933 	fd_set	*in, *ou, *ex;
934 	struct	timeval *tv;
935 };
936 #endif
937 int
938 sys_select(struct thread *td, struct select_args *uap)
939 {
940 	struct timeval tv, *tvp;
941 	int error;
942 
943 	if (uap->tv != NULL) {
944 		error = copyin(uap->tv, &tv, sizeof(tv));
945 		if (error)
946 			return (error);
947 		tvp = &tv;
948 	} else
949 		tvp = NULL;
950 
951 	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
952 	    NFDBITS));
953 }
954 
955 /*
956  * In the unlikely case when user specified n greater then the last
957  * open file descriptor, check that no bits are set after the last
958  * valid fd.  We must return EBADF if any is set.
959  *
960  * There are applications that rely on the behaviour.
961  *
962  * nd is fd_nfiles.
963  */
964 static int
965 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
966 {
967 	char *addr, *oaddr;
968 	int b, i, res;
969 	uint8_t bits;
970 
971 	if (nd >= ndu || fd_in == NULL)
972 		return (0);
973 
974 	oaddr = NULL;
975 	bits = 0; /* silence gcc */
976 	for (i = nd; i < ndu; i++) {
977 		b = i / NBBY;
978 #if BYTE_ORDER == LITTLE_ENDIAN
979 		addr = (char *)fd_in + b;
980 #else
981 		addr = (char *)fd_in;
982 		if (abi_nfdbits == NFDBITS) {
983 			addr += rounddown(b, sizeof(fd_mask)) +
984 			    sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
985 		} else {
986 			addr += rounddown(b, sizeof(uint32_t)) +
987 			    sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
988 		}
989 #endif
990 		if (addr != oaddr) {
991 			res = fubyte(addr);
992 			if (res == -1)
993 				return (EFAULT);
994 			oaddr = addr;
995 			bits = res;
996 		}
997 		if ((bits & (1 << (i % NBBY))) != 0)
998 			return (EBADF);
999 	}
1000 	return (0);
1001 }
1002 
1003 int
1004 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
1005     fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
1006 {
1007 	struct filedesc *fdp;
1008 	/*
1009 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
1010 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
1011 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
1012 	 * of 256.
1013 	 */
1014 	fd_mask s_selbits[howmany(2048, NFDBITS)];
1015 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
1016 	struct timeval rtv;
1017 	sbintime_t asbt, precision, rsbt;
1018 	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
1019 	int error, lf, ndu;
1020 
1021 	if (nd < 0)
1022 		return (EINVAL);
1023 	fdp = td->td_proc->p_fd;
1024 	ndu = nd;
1025 	lf = fdp->fd_nfiles;
1026 	if (nd > lf)
1027 		nd = lf;
1028 
1029 	error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1030 	if (error != 0)
1031 		return (error);
1032 	error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1033 	if (error != 0)
1034 		return (error);
1035 	error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1036 	if (error != 0)
1037 		return (error);
1038 
1039 	/*
1040 	 * Allocate just enough bits for the non-null fd_sets.  Use the
1041 	 * preallocated auto buffer if possible.
1042 	 */
1043 	nfdbits = roundup(nd, NFDBITS);
1044 	ncpbytes = nfdbits / NBBY;
1045 	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1046 	nbufbytes = 0;
1047 	if (fd_in != NULL)
1048 		nbufbytes += 2 * ncpbytes;
1049 	if (fd_ou != NULL)
1050 		nbufbytes += 2 * ncpbytes;
1051 	if (fd_ex != NULL)
1052 		nbufbytes += 2 * ncpbytes;
1053 	if (nbufbytes <= sizeof s_selbits)
1054 		selbits = &s_selbits[0];
1055 	else
1056 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1057 
1058 	/*
1059 	 * Assign pointers into the bit buffers and fetch the input bits.
1060 	 * Put the output buffers together so that they can be bzeroed
1061 	 * together.
1062 	 */
1063 	sbp = selbits;
1064 #define	getbits(name, x) \
1065 	do {								\
1066 		if (name == NULL) {					\
1067 			ibits[x] = NULL;				\
1068 			obits[x] = NULL;				\
1069 		} else {						\
1070 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
1071 			obits[x] = sbp;					\
1072 			sbp += ncpbytes / sizeof *sbp;			\
1073 			error = copyin(name, ibits[x], ncpubytes);	\
1074 			if (error != 0)					\
1075 				goto done;				\
1076 			if (ncpbytes != ncpubytes)			\
1077 				bzero((char *)ibits[x] + ncpubytes,	\
1078 				    ncpbytes - ncpubytes);		\
1079 		}							\
1080 	} while (0)
1081 	getbits(fd_in, 0);
1082 	getbits(fd_ou, 1);
1083 	getbits(fd_ex, 2);
1084 #undef	getbits
1085 
1086 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1087 	/*
1088 	 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1089 	 * we are running under 32-bit emulation. This should be more
1090 	 * generic.
1091 	 */
1092 #define swizzle_fdset(bits)						\
1093 	if (abi_nfdbits != NFDBITS && bits != NULL) {			\
1094 		int i;							\
1095 		for (i = 0; i < ncpbytes / sizeof *sbp; i++)		\
1096 			bits[i] = (bits[i] >> 32) | (bits[i] << 32);	\
1097 	}
1098 #else
1099 #define swizzle_fdset(bits)
1100 #endif
1101 
1102 	/* Make sure the bit order makes it through an ABI transition */
1103 	swizzle_fdset(ibits[0]);
1104 	swizzle_fdset(ibits[1]);
1105 	swizzle_fdset(ibits[2]);
1106 
1107 	if (nbufbytes != 0)
1108 		bzero(selbits, nbufbytes / 2);
1109 
1110 	precision = 0;
1111 	if (tvp != NULL) {
1112 		rtv = *tvp;
1113 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1114 		    rtv.tv_usec >= 1000000) {
1115 			error = EINVAL;
1116 			goto done;
1117 		}
1118 		if (!timevalisset(&rtv))
1119 			asbt = 0;
1120 		else if (rtv.tv_sec <= INT32_MAX) {
1121 			rsbt = tvtosbt(rtv);
1122 			precision = rsbt;
1123 			precision >>= tc_precexp;
1124 			if (TIMESEL(&asbt, rsbt))
1125 				asbt += tc_tick_sbt;
1126 			if (asbt <= SBT_MAX - rsbt)
1127 				asbt += rsbt;
1128 			else
1129 				asbt = -1;
1130 		} else
1131 			asbt = -1;
1132 	} else
1133 		asbt = -1;
1134 	seltdinit(td);
1135 	/* Iterate until the timeout expires or descriptors become ready. */
1136 	for (;;) {
1137 		error = selscan(td, ibits, obits, nd);
1138 		if (error || td->td_retval[0] != 0)
1139 			break;
1140 		error = seltdwait(td, asbt, precision);
1141 		if (error)
1142 			break;
1143 		error = selrescan(td, ibits, obits);
1144 		if (error || td->td_retval[0] != 0)
1145 			break;
1146 	}
1147 	seltdclear(td);
1148 
1149 done:
1150 	/* select is not restarted after signals... */
1151 	if (error == ERESTART)
1152 		error = EINTR;
1153 	if (error == EWOULDBLOCK)
1154 		error = 0;
1155 
1156 	/* swizzle bit order back, if necessary */
1157 	swizzle_fdset(obits[0]);
1158 	swizzle_fdset(obits[1]);
1159 	swizzle_fdset(obits[2]);
1160 #undef swizzle_fdset
1161 
1162 #define	putbits(name, x) \
1163 	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1164 		error = error2;
1165 	if (error == 0) {
1166 		int error2;
1167 
1168 		putbits(fd_in, 0);
1169 		putbits(fd_ou, 1);
1170 		putbits(fd_ex, 2);
1171 #undef putbits
1172 	}
1173 	if (selbits != &s_selbits[0])
1174 		free(selbits, M_SELECT);
1175 
1176 	return (error);
1177 }
1178 /*
1179  * Convert a select bit set to poll flags.
1180  *
1181  * The backend always returns POLLHUP/POLLERR if appropriate and we
1182  * return this as a set bit in any set.
1183  */
1184 static int select_flags[3] = {
1185     POLLRDNORM | POLLHUP | POLLERR,
1186     POLLWRNORM | POLLHUP | POLLERR,
1187     POLLRDBAND | POLLERR
1188 };
1189 
1190 /*
1191  * Compute the fo_poll flags required for a fd given by the index and
1192  * bit position in the fd_mask array.
1193  */
1194 static __inline int
1195 selflags(fd_mask **ibits, int idx, fd_mask bit)
1196 {
1197 	int flags;
1198 	int msk;
1199 
1200 	flags = 0;
1201 	for (msk = 0; msk < 3; msk++) {
1202 		if (ibits[msk] == NULL)
1203 			continue;
1204 		if ((ibits[msk][idx] & bit) == 0)
1205 			continue;
1206 		flags |= select_flags[msk];
1207 	}
1208 	return (flags);
1209 }
1210 
1211 /*
1212  * Set the appropriate output bits given a mask of fired events and the
1213  * input bits originally requested.
1214  */
1215 static __inline int
1216 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1217 {
1218 	int msk;
1219 	int n;
1220 
1221 	n = 0;
1222 	for (msk = 0; msk < 3; msk++) {
1223 		if ((events & select_flags[msk]) == 0)
1224 			continue;
1225 		if (ibits[msk] == NULL)
1226 			continue;
1227 		if ((ibits[msk][idx] & bit) == 0)
1228 			continue;
1229 		/*
1230 		 * XXX Check for a duplicate set.  This can occur because a
1231 		 * socket calls selrecord() twice for each poll() call
1232 		 * resulting in two selfds per real fd.  selrescan() will
1233 		 * call selsetbits twice as a result.
1234 		 */
1235 		if ((obits[msk][idx] & bit) != 0)
1236 			continue;
1237 		obits[msk][idx] |= bit;
1238 		n++;
1239 	}
1240 
1241 	return (n);
1242 }
1243 
1244 static __inline int
1245 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1246 {
1247 
1248 	return (fget_unlocked(fdp, fd, &cap_event_rights, fpp));
1249 }
1250 
1251 /*
1252  * Traverse the list of fds attached to this thread's seltd and check for
1253  * completion.
1254  */
1255 static int
1256 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1257 {
1258 	struct filedesc *fdp;
1259 	struct selinfo *si;
1260 	struct seltd *stp;
1261 	struct selfd *sfp;
1262 	struct selfd *sfn;
1263 	struct file *fp;
1264 	fd_mask bit;
1265 	int fd, ev, n, idx;
1266 	int error;
1267 
1268 	fdp = td->td_proc->p_fd;
1269 	stp = td->td_sel;
1270 	n = 0;
1271 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1272 		fd = (int)(uintptr_t)sfp->sf_cookie;
1273 		si = sfp->sf_si;
1274 		selfdfree(stp, sfp);
1275 		/* If the selinfo wasn't cleared the event didn't fire. */
1276 		if (si != NULL)
1277 			continue;
1278 		error = getselfd_cap(fdp, fd, &fp);
1279 		if (error)
1280 			return (error);
1281 		idx = fd / NFDBITS;
1282 		bit = (fd_mask)1 << (fd % NFDBITS);
1283 		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1284 		fdrop(fp, td);
1285 		if (ev != 0)
1286 			n += selsetbits(ibits, obits, idx, bit, ev);
1287 	}
1288 	stp->st_flags = 0;
1289 	td->td_retval[0] = n;
1290 	return (0);
1291 }
1292 
1293 /*
1294  * Perform the initial filedescriptor scan and register ourselves with
1295  * each selinfo.
1296  */
1297 static int
1298 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd)
1299 {
1300 	struct filedesc *fdp;
1301 	struct file *fp;
1302 	fd_mask bit;
1303 	int ev, flags, end, fd;
1304 	int n, idx;
1305 	int error;
1306 
1307 	fdp = td->td_proc->p_fd;
1308 	n = 0;
1309 	for (idx = 0, fd = 0; fd < nfd; idx++) {
1310 		end = imin(fd + NFDBITS, nfd);
1311 		for (bit = 1; fd < end; bit <<= 1, fd++) {
1312 			/* Compute the list of events we're interested in. */
1313 			flags = selflags(ibits, idx, bit);
1314 			if (flags == 0)
1315 				continue;
1316 			error = getselfd_cap(fdp, fd, &fp);
1317 			if (error)
1318 				return (error);
1319 			selfdalloc(td, (void *)(uintptr_t)fd);
1320 			ev = fo_poll(fp, flags, td->td_ucred, td);
1321 			fdrop(fp, td);
1322 			if (ev != 0)
1323 				n += selsetbits(ibits, obits, idx, bit, ev);
1324 		}
1325 	}
1326 
1327 	td->td_retval[0] = n;
1328 	return (0);
1329 }
1330 
1331 int
1332 sys_poll(struct thread *td, struct poll_args *uap)
1333 {
1334 	struct timespec ts, *tsp;
1335 
1336 	if (uap->timeout != INFTIM) {
1337 		if (uap->timeout < 0)
1338 			return (EINVAL);
1339 		ts.tv_sec = uap->timeout / 1000;
1340 		ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1341 		tsp = &ts;
1342 	} else
1343 		tsp = NULL;
1344 
1345 	return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1346 }
1347 
1348 int
1349 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds,
1350     struct timespec *tsp, sigset_t *uset)
1351 {
1352 	struct pollfd *kfds;
1353 	struct pollfd stackfds[32];
1354 	sbintime_t sbt, precision, tmp;
1355 	time_t over;
1356 	struct timespec ts;
1357 	int error;
1358 
1359 	precision = 0;
1360 	if (tsp != NULL) {
1361 		if (tsp->tv_sec < 0)
1362 			return (EINVAL);
1363 		if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000)
1364 			return (EINVAL);
1365 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1366 			sbt = 0;
1367 		else {
1368 			ts = *tsp;
1369 			if (ts.tv_sec > INT32_MAX / 2) {
1370 				over = ts.tv_sec - INT32_MAX / 2;
1371 				ts.tv_sec -= over;
1372 			} else
1373 				over = 0;
1374 			tmp = tstosbt(ts);
1375 			precision = tmp;
1376 			precision >>= tc_precexp;
1377 			if (TIMESEL(&sbt, tmp))
1378 				sbt += tc_tick_sbt;
1379 			sbt += tmp;
1380 		}
1381 	} else
1382 		sbt = -1;
1383 
1384 	/*
1385 	 * This is kinda bogus.  We have fd limits, but that is not
1386 	 * really related to the size of the pollfd array.  Make sure
1387 	 * we let the process use at least FD_SETSIZE entries and at
1388 	 * least enough for the system-wide limits.  We want to be reasonably
1389 	 * safe, but not overly restrictive.
1390 	 */
1391 	if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1392 		return (EINVAL);
1393 	if (nfds > nitems(stackfds))
1394 		kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK);
1395 	else
1396 		kfds = stackfds;
1397 	error = copyin(ufds, kfds, nfds * sizeof(*kfds));
1398 	if (error)
1399 		goto done;
1400 
1401 	if (uset != NULL) {
1402 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
1403 		    &td->td_oldsigmask, 0);
1404 		if (error)
1405 			goto done;
1406 		td->td_pflags |= TDP_OLDMASK;
1407 		/*
1408 		 * Make sure that ast() is called on return to
1409 		 * usermode and TDP_OLDMASK is cleared, restoring old
1410 		 * sigmask.
1411 		 */
1412 		thread_lock(td);
1413 		td->td_flags |= TDF_ASTPENDING;
1414 		thread_unlock(td);
1415 	}
1416 
1417 	seltdinit(td);
1418 	/* Iterate until the timeout expires or descriptors become ready. */
1419 	for (;;) {
1420 		error = pollscan(td, kfds, nfds);
1421 		if (error || td->td_retval[0] != 0)
1422 			break;
1423 		error = seltdwait(td, sbt, precision);
1424 		if (error)
1425 			break;
1426 		error = pollrescan(td);
1427 		if (error || td->td_retval[0] != 0)
1428 			break;
1429 	}
1430 	seltdclear(td);
1431 
1432 done:
1433 	/* poll is not restarted after signals... */
1434 	if (error == ERESTART)
1435 		error = EINTR;
1436 	if (error == EWOULDBLOCK)
1437 		error = 0;
1438 	if (error == 0) {
1439 		error = pollout(td, kfds, ufds, nfds);
1440 		if (error)
1441 			goto out;
1442 	}
1443 out:
1444 	if (nfds > nitems(stackfds))
1445 		free(kfds, M_TEMP);
1446 	return (error);
1447 }
1448 
1449 int
1450 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1451 {
1452 	struct timespec ts, *tsp;
1453 	sigset_t set, *ssp;
1454 	int error;
1455 
1456 	if (uap->ts != NULL) {
1457 		error = copyin(uap->ts, &ts, sizeof(ts));
1458 		if (error)
1459 			return (error);
1460 		tsp = &ts;
1461 	} else
1462 		tsp = NULL;
1463 	if (uap->set != NULL) {
1464 		error = copyin(uap->set, &set, sizeof(set));
1465 		if (error)
1466 			return (error);
1467 		ssp = &set;
1468 	} else
1469 		ssp = NULL;
1470 	/*
1471 	 * fds is still a pointer to user space. kern_poll() will
1472 	 * take care of copyin that array to the kernel space.
1473 	 */
1474 
1475 	return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1476 }
1477 
1478 #ifdef CAPABILITIES
1479 static int
1480 poll_fget(struct filedesc *fdp, int fd, struct file **fpp)
1481 {
1482 	const struct filedescent *fde;
1483 	const struct fdescenttbl *fdt;
1484 	const cap_rights_t *haverights;
1485 	struct file *fp;
1486 	int error;
1487 
1488 	if (__predict_false(fd >= fdp->fd_nfiles))
1489 		return (EBADF);
1490 
1491 	fdt = fdp->fd_files;
1492 	fde = &fdt->fdt_ofiles[fd];
1493 	fp = fde->fde_file;
1494 	if (__predict_false(fp == NULL))
1495 		return (EBADF);
1496 	haverights = cap_rights_fde_inline(fde);
1497 	error = cap_check_inline(haverights, &cap_event_rights);
1498 	if (__predict_false(error != 0))
1499 		return (EBADF);
1500 	*fpp = fp;
1501 	return (0);
1502 }
1503 #else
1504 static int
1505 poll_fget(struct filedesc *fdp, int fd, struct file **fpp)
1506 {
1507 	struct file *fp;
1508 
1509 	if (__predict_false(fd >= fdp->fd_nfiles))
1510 		return (EBADF);
1511 
1512 	fp = fdp->fd_ofiles[fd].fde_file;
1513 	if (__predict_false(fp == NULL))
1514 		return (EBADF);
1515 
1516 	*fpp = fp;
1517 	return (0);
1518 }
1519 #endif
1520 
1521 static int
1522 pollrescan(struct thread *td)
1523 {
1524 	struct seltd *stp;
1525 	struct selfd *sfp;
1526 	struct selfd *sfn;
1527 	struct selinfo *si;
1528 	struct filedesc *fdp;
1529 	struct file *fp;
1530 	struct pollfd *fd;
1531 	int n;
1532 
1533 	n = 0;
1534 	fdp = td->td_proc->p_fd;
1535 	stp = td->td_sel;
1536 	FILEDESC_SLOCK(fdp);
1537 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1538 		fd = (struct pollfd *)sfp->sf_cookie;
1539 		si = sfp->sf_si;
1540 		selfdfree(stp, sfp);
1541 		/* If the selinfo wasn't cleared the event didn't fire. */
1542 		if (si != NULL)
1543 			continue;
1544 		if (poll_fget(fdp, fd->fd, &fp) != 0) {
1545 			fd->revents = POLLNVAL;
1546 			n++;
1547 			continue;
1548 		}
1549 		/*
1550 		 * Note: backend also returns POLLHUP and
1551 		 * POLLERR if appropriate.
1552 		 */
1553 		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1554 		if (fd->revents != 0)
1555 			n++;
1556 	}
1557 	FILEDESC_SUNLOCK(fdp);
1558 	stp->st_flags = 0;
1559 	td->td_retval[0] = n;
1560 	return (0);
1561 }
1562 
1563 static int
1564 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd)
1565 {
1566 	int error = 0;
1567 	u_int i = 0;
1568 	u_int n = 0;
1569 
1570 	for (i = 0; i < nfd; i++) {
1571 		error = copyout(&fds->revents, &ufds->revents,
1572 		    sizeof(ufds->revents));
1573 		if (error)
1574 			return (error);
1575 		if (fds->revents != 0)
1576 			n++;
1577 		fds++;
1578 		ufds++;
1579 	}
1580 	td->td_retval[0] = n;
1581 	return (0);
1582 }
1583 
1584 static int
1585 pollscan(struct thread *td, struct pollfd *fds, u_int nfd)
1586 {
1587 	struct filedesc *fdp;
1588 	struct file *fp;
1589 	int i, n;
1590 
1591 	n = 0;
1592 	fdp = td->td_proc->p_fd;
1593 	FILEDESC_SLOCK(fdp);
1594 	for (i = 0; i < nfd; i++, fds++) {
1595 		if (fds->fd < 0) {
1596 			fds->revents = 0;
1597 			continue;
1598 		}
1599 		if (poll_fget(fdp, fds->fd, &fp) != 0) {
1600 			fds->revents = POLLNVAL;
1601 			n++;
1602 			continue;
1603 		}
1604 		/*
1605 		 * Note: backend also returns POLLHUP and
1606 		 * POLLERR if appropriate.
1607 		 */
1608 		selfdalloc(td, fds);
1609 		fds->revents = fo_poll(fp, fds->events,
1610 		    td->td_ucred, td);
1611 		/*
1612 		 * POSIX requires POLLOUT to be never
1613 		 * set simultaneously with POLLHUP.
1614 		 */
1615 		if ((fds->revents & POLLHUP) != 0)
1616 			fds->revents &= ~POLLOUT;
1617 
1618 		if (fds->revents != 0)
1619 			n++;
1620 	}
1621 	FILEDESC_SUNLOCK(fdp);
1622 	td->td_retval[0] = n;
1623 	return (0);
1624 }
1625 
1626 /*
1627  * XXX This was created specifically to support netncp and netsmb.  This
1628  * allows the caller to specify a socket to wait for events on.  It returns
1629  * 0 if any events matched and an error otherwise.  There is no way to
1630  * determine which events fired.
1631  */
1632 int
1633 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1634 {
1635 	struct timeval rtv;
1636 	sbintime_t asbt, precision, rsbt;
1637 	int error;
1638 
1639 	precision = 0;	/* stupid gcc! */
1640 	if (tvp != NULL) {
1641 		rtv = *tvp;
1642 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1643 		    rtv.tv_usec >= 1000000)
1644 			return (EINVAL);
1645 		if (!timevalisset(&rtv))
1646 			asbt = 0;
1647 		else if (rtv.tv_sec <= INT32_MAX) {
1648 			rsbt = tvtosbt(rtv);
1649 			precision = rsbt;
1650 			precision >>= tc_precexp;
1651 			if (TIMESEL(&asbt, rsbt))
1652 				asbt += tc_tick_sbt;
1653 			if (asbt <= SBT_MAX - rsbt)
1654 				asbt += rsbt;
1655 			else
1656 				asbt = -1;
1657 		} else
1658 			asbt = -1;
1659 	} else
1660 		asbt = -1;
1661 	seltdinit(td);
1662 	/*
1663 	 * Iterate until the timeout expires or the socket becomes ready.
1664 	 */
1665 	for (;;) {
1666 		selfdalloc(td, NULL);
1667 		error = sopoll(so, events, NULL, td);
1668 		/* error here is actually the ready events. */
1669 		if (error)
1670 			return (0);
1671 		error = seltdwait(td, asbt, precision);
1672 		if (error)
1673 			break;
1674 	}
1675 	seltdclear(td);
1676 	/* XXX Duplicates ncp/smb behavior. */
1677 	if (error == ERESTART)
1678 		error = 0;
1679 	return (error);
1680 }
1681 
1682 /*
1683  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1684  * have two select sets, one for read and another for write.
1685  */
1686 static void
1687 selfdalloc(struct thread *td, void *cookie)
1688 {
1689 	struct seltd *stp;
1690 
1691 	stp = td->td_sel;
1692 	if (stp->st_free1 == NULL)
1693 		stp->st_free1 = malloc(sizeof(*stp->st_free1), M_SELFD, M_WAITOK|M_ZERO);
1694 	stp->st_free1->sf_td = stp;
1695 	stp->st_free1->sf_cookie = cookie;
1696 	if (stp->st_free2 == NULL)
1697 		stp->st_free2 = malloc(sizeof(*stp->st_free2), M_SELFD, M_WAITOK|M_ZERO);
1698 	stp->st_free2->sf_td = stp;
1699 	stp->st_free2->sf_cookie = cookie;
1700 }
1701 
1702 static void
1703 selfdfree(struct seltd *stp, struct selfd *sfp)
1704 {
1705 	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1706 	/*
1707 	 * Paired with doselwakeup.
1708 	 */
1709 	if (atomic_load_acq_ptr((uintptr_t *)&sfp->sf_si) != (uintptr_t)NULL) {
1710 		mtx_lock(sfp->sf_mtx);
1711 		if (sfp->sf_si != NULL) {
1712 			TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1713 		}
1714 		mtx_unlock(sfp->sf_mtx);
1715 	}
1716 	free(sfp, M_SELFD);
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 	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1770 	/*
1771 	 * Now that we've locked the sip, check for initialization.
1772 	 */
1773 	mtx_lock(mtxp);
1774 	if (sip->si_mtx == NULL) {
1775 		sip->si_mtx = mtxp;
1776 		TAILQ_INIT(&sip->si_tdlist);
1777 	}
1778 	/*
1779 	 * Add this thread to the list of selfds listening on this selinfo.
1780 	 */
1781 	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1782 	mtx_unlock(sip->si_mtx);
1783 }
1784 
1785 /* Wake up a selecting thread. */
1786 void
1787 selwakeup(struct selinfo *sip)
1788 {
1789 	doselwakeup(sip, -1);
1790 }
1791 
1792 /* Wake up a selecting thread, and set its priority. */
1793 void
1794 selwakeuppri(struct selinfo *sip, int pri)
1795 {
1796 	doselwakeup(sip, pri);
1797 }
1798 
1799 /*
1800  * Do a wakeup when a selectable event occurs.
1801  */
1802 static void
1803 doselwakeup(struct selinfo *sip, int pri)
1804 {
1805 	struct selfd *sfp;
1806 	struct selfd *sfn;
1807 	struct seltd *stp;
1808 
1809 	/* If it's not initialized there can't be any waiters. */
1810 	if (sip->si_mtx == NULL)
1811 		return;
1812 	/*
1813 	 * Locking the selinfo locks all selfds associated with it.
1814 	 */
1815 	mtx_lock(sip->si_mtx);
1816 	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1817 		/*
1818 		 * Once we remove this sfp from the list and clear the
1819 		 * sf_si seltdclear will know to ignore this si.
1820 		 */
1821 		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1822 		stp = sfp->sf_td;
1823 		/*
1824 		 * Paired with selfdfree.
1825 		 */
1826 		atomic_store_rel_ptr((uintptr_t *)&sfp->sf_si, (uintptr_t)NULL);
1827 		mtx_lock(&stp->st_mtx);
1828 		stp->st_flags |= SELTD_PENDING;
1829 		cv_broadcastpri(&stp->st_wait, pri);
1830 		mtx_unlock(&stp->st_mtx);
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 		free(stp->st_free1, M_SELFD);
1892 	if (stp->st_free2)
1893 		free(stp->st_free2, M_SELFD);
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 	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1924 }
1925 
1926 /*
1927  * Set up a syscall return value that follows the convention specified for
1928  * posix_* functions.
1929  */
1930 int
1931 kern_posix_error(struct thread *td, int error)
1932 {
1933 
1934 	if (error <= 0)
1935 		return (error);
1936 	td->td_errno = error;
1937 	td->td_pflags |= TDP_NERRNO;
1938 	td->td_retval[0] = error;
1939 	return (0);
1940 }
1941