xref: /freebsd/sys/kern/sys_generic.c (revision 5b3e5c6ce3e552dae3e84d6bc781bc9f53c22d71)
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 
37 #include <sys/cdefs.h>
38 #include "opt_capsicum.h"
39 #include "opt_ktrace.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/sysproto.h>
44 #include <sys/capsicum.h>
45 #include <sys/filedesc.h>
46 #include <sys/filio.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/lock.h>
50 #include <sys/proc.h>
51 #include <sys/signalvar.h>
52 #include <sys/socketvar.h>
53 #include <sys/uio.h>
54 #include <sys/eventfd.h>
55 #include <sys/kernel.h>
56 #include <sys/ktr.h>
57 #include <sys/limits.h>
58 #include <sys/malloc.h>
59 #include <sys/poll.h>
60 #include <sys/resourcevar.h>
61 #include <sys/selinfo.h>
62 #include <sys/sleepqueue.h>
63 #include <sys/specialfd.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/vnode.h>
68 #include <sys/unistd.h>
69 #include <sys/bio.h>
70 #include <sys/buf.h>
71 #include <sys/condvar.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75 
76 #include <security/audit/audit.h>
77 
78 /*
79  * The following macro defines how many bytes will be allocated from
80  * the stack instead of memory allocated when passing the IOCTL data
81  * structures from userspace and to the kernel. Some IOCTLs having
82  * small data structures are used very frequently and this small
83  * buffer on the stack gives a significant speedup improvement for
84  * those requests. The value of this define should be greater or equal
85  * to 64 bytes and should also be power of two. The data structure is
86  * currently hard-aligned to a 8-byte boundary on the stack. This
87  * should currently be sufficient for all supported platforms.
88  */
89 #define	SYS_IOCTL_SMALL_SIZE	128	/* bytes */
90 #define	SYS_IOCTL_SMALL_ALIGN	8	/* bytes */
91 
92 #ifdef __LP64__
93 static int iosize_max_clamp = 0;
94 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
95     &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
96 static int devfs_iosize_max_clamp = 1;
97 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
98     &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
99 #endif
100 
101 /*
102  * Assert that the return value of read(2) and write(2) syscalls fits
103  * into a register.  If not, an architecture will need to provide the
104  * usermode wrappers to reconstruct the result.
105  */
106 CTASSERT(sizeof(register_t) >= sizeof(size_t));
107 
108 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
109 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
110 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
111 
112 static int	pollout(struct thread *, struct pollfd *, struct pollfd *,
113 		    u_int);
114 static int	pollscan(struct thread *, struct pollfd *, u_int);
115 static int	pollrescan(struct thread *);
116 static int	selscan(struct thread *, fd_mask **, fd_mask **, int);
117 static int	selrescan(struct thread *, fd_mask **, fd_mask **);
118 static void	selfdalloc(struct thread *, void *);
119 static void	selfdfree(struct seltd *, struct selfd *);
120 static int	dofileread(struct thread *, int, struct file *, struct uio *,
121 		    off_t, int);
122 static int	dofilewrite(struct thread *, int, struct file *, struct uio *,
123 		    off_t, int);
124 static void	doselwakeup(struct selinfo *, int);
125 static void	seltdinit(struct thread *);
126 static int	seltdwait(struct thread *, sbintime_t, sbintime_t);
127 static void	seltdclear(struct thread *);
128 
129 /*
130  * One seltd per-thread allocated on demand as needed.
131  *
132  *	t - protected by st_mtx
133  * 	k - Only accessed by curthread or read-only
134  */
135 struct seltd {
136 	STAILQ_HEAD(, selfd)	st_selq;	/* (k) List of selfds. */
137 	struct selfd		*st_free1;	/* (k) free fd for read set. */
138 	struct selfd		*st_free2;	/* (k) free fd for write set. */
139 	struct mtx		st_mtx;		/* Protects struct seltd */
140 	struct cv		st_wait;	/* (t) Wait channel. */
141 	int			st_flags;	/* (t) SELTD_ flags. */
142 };
143 
144 #define	SELTD_PENDING	0x0001			/* We have pending events. */
145 #define	SELTD_RESCAN	0x0002			/* Doing a rescan. */
146 
147 /*
148  * One selfd allocated per-thread per-file-descriptor.
149  *	f - protected by sf_mtx
150  */
151 struct selfd {
152 	STAILQ_ENTRY(selfd)	sf_link;	/* (k) fds owned by this td. */
153 	TAILQ_ENTRY(selfd)	sf_threads;	/* (f) fds on this selinfo. */
154 	struct selinfo		*sf_si;		/* (f) selinfo when linked. */
155 	struct mtx		*sf_mtx;	/* Pointer to selinfo mtx. */
156 	struct seltd		*sf_td;		/* (k) owning seltd. */
157 	void			*sf_cookie;	/* (k) fd or pollfd. */
158 };
159 
160 MALLOC_DEFINE(M_SELFD, "selfd", "selfd");
161 static struct mtx_pool *mtxpool_select;
162 
163 #ifdef __LP64__
164 size_t
165 devfs_iosize_max(void)
166 {
167 
168 	return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
169 	    INT_MAX : SSIZE_MAX);
170 }
171 
172 size_t
173 iosize_max(void)
174 {
175 
176 	return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
177 	    INT_MAX : SSIZE_MAX);
178 }
179 #endif
180 
181 #ifndef _SYS_SYSPROTO_H_
182 struct read_args {
183 	int	fd;
184 	void	*buf;
185 	size_t	nbyte;
186 };
187 #endif
188 int
189 sys_read(struct thread *td, struct read_args *uap)
190 {
191 	struct uio auio;
192 	struct iovec aiov;
193 	int error;
194 
195 	if (uap->nbyte > IOSIZE_MAX)
196 		return (EINVAL);
197 	aiov.iov_base = uap->buf;
198 	aiov.iov_len = uap->nbyte;
199 	auio.uio_iov = &aiov;
200 	auio.uio_iovcnt = 1;
201 	auio.uio_resid = uap->nbyte;
202 	auio.uio_segflg = UIO_USERSPACE;
203 	error = kern_readv(td, uap->fd, &auio);
204 	return (error);
205 }
206 
207 /*
208  * Positioned read system call
209  */
210 #ifndef _SYS_SYSPROTO_H_
211 struct pread_args {
212 	int	fd;
213 	void	*buf;
214 	size_t	nbyte;
215 	int	pad;
216 	off_t	offset;
217 };
218 #endif
219 int
220 sys_pread(struct thread *td, struct pread_args *uap)
221 {
222 
223 	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
224 }
225 
226 int
227 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset)
228 {
229 	struct uio auio;
230 	struct iovec aiov;
231 	int error;
232 
233 	if (nbyte > IOSIZE_MAX)
234 		return (EINVAL);
235 	aiov.iov_base = buf;
236 	aiov.iov_len = nbyte;
237 	auio.uio_iov = &aiov;
238 	auio.uio_iovcnt = 1;
239 	auio.uio_resid = nbyte;
240 	auio.uio_segflg = UIO_USERSPACE;
241 	error = kern_preadv(td, fd, &auio, offset);
242 	return (error);
243 }
244 
245 #if defined(COMPAT_FREEBSD6)
246 int
247 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap)
248 {
249 
250 	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
251 }
252 #endif
253 
254 /*
255  * Scatter read system call.
256  */
257 #ifndef _SYS_SYSPROTO_H_
258 struct readv_args {
259 	int	fd;
260 	struct	iovec *iovp;
261 	u_int	iovcnt;
262 };
263 #endif
264 int
265 sys_readv(struct thread *td, struct readv_args *uap)
266 {
267 	struct uio *auio;
268 	int error;
269 
270 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
271 	if (error)
272 		return (error);
273 	error = kern_readv(td, uap->fd, auio);
274 	freeuio(auio);
275 	return (error);
276 }
277 
278 int
279 kern_readv(struct thread *td, int fd, struct uio *auio)
280 {
281 	struct file *fp;
282 	int error;
283 
284 	error = fget_read(td, fd, &cap_read_rights, &fp);
285 	if (error)
286 		return (error);
287 	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
288 	fdrop(fp, td);
289 	return (error);
290 }
291 
292 /*
293  * Scatter positioned read system call.
294  */
295 #ifndef _SYS_SYSPROTO_H_
296 struct preadv_args {
297 	int	fd;
298 	struct	iovec *iovp;
299 	u_int	iovcnt;
300 	off_t	offset;
301 };
302 #endif
303 int
304 sys_preadv(struct thread *td, struct preadv_args *uap)
305 {
306 	struct uio *auio;
307 	int error;
308 
309 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
310 	if (error)
311 		return (error);
312 	error = kern_preadv(td, uap->fd, auio, uap->offset);
313 	freeuio(auio);
314 	return (error);
315 }
316 
317 int
318 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset)
319 {
320 	struct file *fp;
321 	int error;
322 
323 	error = fget_read(td, fd, &cap_pread_rights, &fp);
324 	if (error)
325 		return (error);
326 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
327 		error = ESPIPE;
328 	else if (offset < 0 &&
329 	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
330 		error = EINVAL;
331 	else
332 		error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
333 	fdrop(fp, td);
334 	return (error);
335 }
336 
337 /*
338  * Common code for readv and preadv that reads data in
339  * from a file using the passed in uio, offset, and flags.
340  */
341 static int
342 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio,
343     off_t offset, int flags)
344 {
345 	ssize_t cnt;
346 	int error;
347 #ifdef KTRACE
348 	struct uio *ktruio = NULL;
349 #endif
350 
351 	AUDIT_ARG_FD(fd);
352 
353 	/* Finish zero length reads right here */
354 	if (auio->uio_resid == 0) {
355 		td->td_retval[0] = 0;
356 		return (0);
357 	}
358 	auio->uio_rw = UIO_READ;
359 	auio->uio_offset = offset;
360 	auio->uio_td = td;
361 #ifdef KTRACE
362 	if (KTRPOINT(td, KTR_GENIO))
363 		ktruio = cloneuio(auio);
364 #endif
365 	cnt = auio->uio_resid;
366 	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
367 		if (auio->uio_resid != cnt && (error == ERESTART ||
368 		    error == EINTR || error == EWOULDBLOCK))
369 			error = 0;
370 	}
371 	cnt -= auio->uio_resid;
372 #ifdef KTRACE
373 	if (ktruio != NULL) {
374 		ktruio->uio_resid = cnt;
375 		ktrgenio(fd, UIO_READ, ktruio, error);
376 	}
377 #endif
378 	td->td_retval[0] = cnt;
379 	return (error);
380 }
381 
382 #ifndef _SYS_SYSPROTO_H_
383 struct write_args {
384 	int	fd;
385 	const void *buf;
386 	size_t	nbyte;
387 };
388 #endif
389 int
390 sys_write(struct thread *td, struct write_args *uap)
391 {
392 	struct uio auio;
393 	struct iovec aiov;
394 	int error;
395 
396 	if (uap->nbyte > IOSIZE_MAX)
397 		return (EINVAL);
398 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
399 	aiov.iov_len = uap->nbyte;
400 	auio.uio_iov = &aiov;
401 	auio.uio_iovcnt = 1;
402 	auio.uio_resid = uap->nbyte;
403 	auio.uio_segflg = UIO_USERSPACE;
404 	error = kern_writev(td, uap->fd, &auio);
405 	return (error);
406 }
407 
408 /*
409  * Positioned write system call.
410  */
411 #ifndef _SYS_SYSPROTO_H_
412 struct pwrite_args {
413 	int	fd;
414 	const void *buf;
415 	size_t	nbyte;
416 	int	pad;
417 	off_t	offset;
418 };
419 #endif
420 int
421 sys_pwrite(struct thread *td, struct pwrite_args *uap)
422 {
423 
424 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
425 }
426 
427 int
428 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte,
429     off_t offset)
430 {
431 	struct uio auio;
432 	struct iovec aiov;
433 	int error;
434 
435 	if (nbyte > IOSIZE_MAX)
436 		return (EINVAL);
437 	aiov.iov_base = (void *)(uintptr_t)buf;
438 	aiov.iov_len = nbyte;
439 	auio.uio_iov = &aiov;
440 	auio.uio_iovcnt = 1;
441 	auio.uio_resid = nbyte;
442 	auio.uio_segflg = UIO_USERSPACE;
443 	error = kern_pwritev(td, fd, &auio, offset);
444 	return (error);
445 }
446 
447 #if defined(COMPAT_FREEBSD6)
448 int
449 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap)
450 {
451 
452 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
453 }
454 #endif
455 
456 /*
457  * Gather write system call.
458  */
459 #ifndef _SYS_SYSPROTO_H_
460 struct writev_args {
461 	int	fd;
462 	struct	iovec *iovp;
463 	u_int	iovcnt;
464 };
465 #endif
466 int
467 sys_writev(struct thread *td, struct writev_args *uap)
468 {
469 	struct uio *auio;
470 	int error;
471 
472 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
473 	if (error)
474 		return (error);
475 	error = kern_writev(td, uap->fd, auio);
476 	freeuio(auio);
477 	return (error);
478 }
479 
480 int
481 kern_writev(struct thread *td, int fd, struct uio *auio)
482 {
483 	struct file *fp;
484 	int error;
485 
486 	error = fget_write(td, fd, &cap_write_rights, &fp);
487 	if (error)
488 		return (error);
489 	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
490 	fdrop(fp, td);
491 	return (error);
492 }
493 
494 /*
495  * Gather positioned write system call.
496  */
497 #ifndef _SYS_SYSPROTO_H_
498 struct pwritev_args {
499 	int	fd;
500 	struct	iovec *iovp;
501 	u_int	iovcnt;
502 	off_t	offset;
503 };
504 #endif
505 int
506 sys_pwritev(struct thread *td, struct pwritev_args *uap)
507 {
508 	struct uio *auio;
509 	int error;
510 
511 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
512 	if (error)
513 		return (error);
514 	error = kern_pwritev(td, uap->fd, auio, uap->offset);
515 	freeuio(auio);
516 	return (error);
517 }
518 
519 int
520 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset)
521 {
522 	struct file *fp;
523 	int error;
524 
525 	error = fget_write(td, fd, &cap_pwrite_rights, &fp);
526 	if (error)
527 		return (error);
528 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
529 		error = ESPIPE;
530 	else if (offset < 0 &&
531 	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
532 		error = EINVAL;
533 	else
534 		error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
535 	fdrop(fp, td);
536 	return (error);
537 }
538 
539 /*
540  * Common code for writev and pwritev that writes data to
541  * a file using the passed in uio, offset, and flags.
542  */
543 static int
544 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio,
545     off_t offset, int flags)
546 {
547 	ssize_t cnt;
548 	int error;
549 #ifdef KTRACE
550 	struct uio *ktruio = NULL;
551 #endif
552 
553 	AUDIT_ARG_FD(fd);
554 	auio->uio_rw = UIO_WRITE;
555 	auio->uio_td = td;
556 	auio->uio_offset = offset;
557 #ifdef KTRACE
558 	if (KTRPOINT(td, KTR_GENIO))
559 		ktruio = cloneuio(auio);
560 #endif
561 	cnt = auio->uio_resid;
562 	error = fo_write(fp, auio, td->td_ucred, flags, td);
563 	/*
564 	 * Socket layer is responsible for special error handling,
565 	 * see sousrsend().
566 	 */
567 	if (error != 0 && fp->f_type != DTYPE_SOCKET) {
568 		if (auio->uio_resid != cnt && (error == ERESTART ||
569 		    error == EINTR || error == EWOULDBLOCK))
570 			error = 0;
571 		if (error == EPIPE) {
572 			PROC_LOCK(td->td_proc);
573 			tdsignal(td, SIGPIPE);
574 			PROC_UNLOCK(td->td_proc);
575 		}
576 	}
577 	cnt -= auio->uio_resid;
578 #ifdef KTRACE
579 	if (ktruio != NULL) {
580 		if (error == 0)
581 			ktruio->uio_resid = cnt;
582 		ktrgenio(fd, UIO_WRITE, ktruio, error);
583 	}
584 #endif
585 	td->td_retval[0] = cnt;
586 	return (error);
587 }
588 
589 /*
590  * Truncate a file given a file descriptor.
591  *
592  * Can't use fget_write() here, since must return EINVAL and not EBADF if the
593  * descriptor isn't writable.
594  */
595 int
596 kern_ftruncate(struct thread *td, int fd, off_t length)
597 {
598 	struct file *fp;
599 	int error;
600 
601 	AUDIT_ARG_FD(fd);
602 	if (length < 0)
603 		return (EINVAL);
604 	error = fget(td, fd, &cap_ftruncate_rights, &fp);
605 	if (error)
606 		return (error);
607 	AUDIT_ARG_FILE(td->td_proc, fp);
608 	if (!(fp->f_flag & FWRITE)) {
609 		fdrop(fp, td);
610 		return (EINVAL);
611 	}
612 	error = fo_truncate(fp, length, td->td_ucred, td);
613 	fdrop(fp, td);
614 	return (error);
615 }
616 
617 #ifndef _SYS_SYSPROTO_H_
618 struct ftruncate_args {
619 	int	fd;
620 	int	pad;
621 	off_t	length;
622 };
623 #endif
624 int
625 sys_ftruncate(struct thread *td, struct ftruncate_args *uap)
626 {
627 
628 	return (kern_ftruncate(td, uap->fd, uap->length));
629 }
630 
631 #if defined(COMPAT_43)
632 #ifndef _SYS_SYSPROTO_H_
633 struct oftruncate_args {
634 	int	fd;
635 	long	length;
636 };
637 #endif
638 int
639 oftruncate(struct thread *td, struct oftruncate_args *uap)
640 {
641 
642 	return (kern_ftruncate(td, uap->fd, uap->length));
643 }
644 #endif /* COMPAT_43 */
645 
646 #ifndef _SYS_SYSPROTO_H_
647 struct ioctl_args {
648 	int	fd;
649 	u_long	com;
650 	caddr_t	data;
651 };
652 #endif
653 /* ARGSUSED */
654 int
655 sys_ioctl(struct thread *td, struct ioctl_args *uap)
656 {
657 	u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
658 	uint32_t com;
659 	int arg, error;
660 	u_int size;
661 	caddr_t data;
662 
663 #ifdef INVARIANTS
664 	if (uap->com > 0xffffffff) {
665 		printf(
666 		    "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
667 		    td->td_proc->p_pid, td->td_name, uap->com);
668 	}
669 #endif
670 	com = (uint32_t)uap->com;
671 
672 	/*
673 	 * Interpret high order word to find amount of data to be
674 	 * copied to/from the user's address space.
675 	 */
676 	size = IOCPARM_LEN(com);
677 	if ((size > IOCPARM_MAX) ||
678 	    ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
679 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
680 	    ((com & IOC_OUT) && size == 0) ||
681 #else
682 	    ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
683 #endif
684 	    ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
685 		return (ENOTTY);
686 
687 	if (size > 0) {
688 		if (com & IOC_VOID) {
689 			/* Integer argument. */
690 			arg = (intptr_t)uap->data;
691 			data = (void *)&arg;
692 			size = 0;
693 		} else {
694 			if (size > SYS_IOCTL_SMALL_SIZE)
695 				data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
696 			else
697 				data = smalldata;
698 		}
699 	} else
700 		data = (void *)&uap->data;
701 	if (com & IOC_IN) {
702 		error = copyin(uap->data, data, (u_int)size);
703 		if (error != 0)
704 			goto out;
705 	} else if (com & IOC_OUT) {
706 		/*
707 		 * Zero the buffer so the user always
708 		 * gets back something deterministic.
709 		 */
710 		bzero(data, size);
711 	}
712 
713 	error = kern_ioctl(td, uap->fd, com, data);
714 
715 	if (error == 0 && (com & IOC_OUT))
716 		error = copyout(data, uap->data, (u_int)size);
717 
718 out:
719 	if (size > SYS_IOCTL_SMALL_SIZE)
720 		free(data, M_IOCTLOPS);
721 	return (error);
722 }
723 
724 int
725 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
726 {
727 	struct file *fp;
728 	struct filedesc *fdp;
729 	int error, tmp, locked;
730 
731 	AUDIT_ARG_FD(fd);
732 	AUDIT_ARG_CMD(com);
733 
734 	fdp = td->td_proc->p_fd;
735 
736 	switch (com) {
737 	case FIONCLEX:
738 	case FIOCLEX:
739 		FILEDESC_XLOCK(fdp);
740 		locked = LA_XLOCKED;
741 		break;
742 	default:
743 #ifdef CAPABILITIES
744 		FILEDESC_SLOCK(fdp);
745 		locked = LA_SLOCKED;
746 #else
747 		locked = LA_UNLOCKED;
748 #endif
749 		break;
750 	}
751 
752 #ifdef CAPABILITIES
753 	if ((fp = fget_noref(fdp, fd)) == NULL) {
754 		error = EBADF;
755 		goto out;
756 	}
757 	if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
758 		fp = NULL;	/* fhold() was not called yet */
759 		goto out;
760 	}
761 	if (!fhold(fp)) {
762 		error = EBADF;
763 		fp = NULL;
764 		goto out;
765 	}
766 	if (locked == LA_SLOCKED) {
767 		FILEDESC_SUNLOCK(fdp);
768 		locked = LA_UNLOCKED;
769 	}
770 #else
771 	error = fget(td, fd, &cap_ioctl_rights, &fp);
772 	if (error != 0) {
773 		fp = NULL;
774 		goto out;
775 	}
776 #endif
777 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
778 		error = EBADF;
779 		goto out;
780 	}
781 
782 	switch (com) {
783 	case FIONCLEX:
784 		fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
785 		goto out;
786 	case FIOCLEX:
787 		fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
788 		goto out;
789 	case FIONBIO:
790 		if ((tmp = *(int *)data))
791 			atomic_set_int(&fp->f_flag, FNONBLOCK);
792 		else
793 			atomic_clear_int(&fp->f_flag, FNONBLOCK);
794 		data = (void *)&tmp;
795 		break;
796 	case FIOASYNC:
797 		if ((tmp = *(int *)data))
798 			atomic_set_int(&fp->f_flag, FASYNC);
799 		else
800 			atomic_clear_int(&fp->f_flag, FASYNC);
801 		data = (void *)&tmp;
802 		break;
803 	}
804 
805 	error = fo_ioctl(fp, com, data, td->td_ucred, td);
806 out:
807 	switch (locked) {
808 	case LA_XLOCKED:
809 		FILEDESC_XUNLOCK(fdp);
810 		break;
811 #ifdef CAPABILITIES
812 	case LA_SLOCKED:
813 		FILEDESC_SUNLOCK(fdp);
814 		break;
815 #endif
816 	default:
817 		FILEDESC_UNLOCK_ASSERT(fdp);
818 		break;
819 	}
820 	if (fp != NULL)
821 		fdrop(fp, td);
822 	return (error);
823 }
824 
825 int
826 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap)
827 {
828 	int error;
829 
830 	error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len);
831 	return (kern_posix_error(td, error));
832 }
833 
834 int
835 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len)
836 {
837 	struct file *fp;
838 	int error;
839 
840 	AUDIT_ARG_FD(fd);
841 	if (offset < 0 || len <= 0)
842 		return (EINVAL);
843 	/* Check for wrap. */
844 	if (offset > OFF_MAX - len)
845 		return (EFBIG);
846 	AUDIT_ARG_FD(fd);
847 	error = fget(td, fd, &cap_pwrite_rights, &fp);
848 	if (error != 0)
849 		return (error);
850 	AUDIT_ARG_FILE(td->td_proc, fp);
851 	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
852 		error = ESPIPE;
853 		goto out;
854 	}
855 	if ((fp->f_flag & FWRITE) == 0) {
856 		error = EBADF;
857 		goto out;
858 	}
859 
860 	error = fo_fallocate(fp, offset, len, td);
861  out:
862 	fdrop(fp, td);
863 	return (error);
864 }
865 
866 int
867 sys_fspacectl(struct thread *td, struct fspacectl_args *uap)
868 {
869 	struct spacectl_range rqsr, rmsr;
870 	int error, cerror;
871 
872 	error = copyin(uap->rqsr, &rqsr, sizeof(rqsr));
873 	if (error != 0)
874 		return (error);
875 
876 	error = kern_fspacectl(td, uap->fd, uap->cmd, &rqsr, uap->flags,
877 	    &rmsr);
878 	if (uap->rmsr != NULL) {
879 		cerror = copyout(&rmsr, uap->rmsr, sizeof(rmsr));
880 		if (error == 0)
881 			error = cerror;
882 	}
883 	return (error);
884 }
885 
886 int
887 kern_fspacectl(struct thread *td, int fd, int cmd,
888     const struct spacectl_range *rqsr, int flags, struct spacectl_range *rmsrp)
889 {
890 	struct file *fp;
891 	struct spacectl_range rmsr;
892 	int error;
893 
894 	AUDIT_ARG_FD(fd);
895 	AUDIT_ARG_CMD(cmd);
896 	AUDIT_ARG_FFLAGS(flags);
897 
898 	if (rqsr == NULL)
899 		return (EINVAL);
900 	rmsr = *rqsr;
901 	if (rmsrp != NULL)
902 		*rmsrp = rmsr;
903 
904 	if (cmd != SPACECTL_DEALLOC ||
905 	    rqsr->r_offset < 0 || rqsr->r_len <= 0 ||
906 	    rqsr->r_offset > OFF_MAX - rqsr->r_len ||
907 	    (flags & ~SPACECTL_F_SUPPORTED) != 0)
908 		return (EINVAL);
909 
910 	error = fget_write(td, fd, &cap_pwrite_rights, &fp);
911 	if (error != 0)
912 		return (error);
913 	AUDIT_ARG_FILE(td->td_proc, fp);
914 	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
915 		error = ESPIPE;
916 		goto out;
917 	}
918 	if ((fp->f_flag & FWRITE) == 0) {
919 		error = EBADF;
920 		goto out;
921 	}
922 
923 	error = fo_fspacectl(fp, cmd, &rmsr.r_offset, &rmsr.r_len, flags,
924 	    td->td_ucred, td);
925 	/* fspacectl is not restarted after signals if the file is modified. */
926 	if (rmsr.r_len != rqsr->r_len && (error == ERESTART ||
927 	    error == EINTR || error == EWOULDBLOCK))
928 		error = 0;
929 	if (rmsrp != NULL)
930 		*rmsrp = rmsr;
931 out:
932 	fdrop(fp, td);
933 	return (error);
934 }
935 
936 int
937 kern_specialfd(struct thread *td, int type, void *arg)
938 {
939 	struct file *fp;
940 	struct specialfd_eventfd *ae;
941 	int error, fd, fflags;
942 
943 	fflags = 0;
944 	error = falloc_noinstall(td, &fp);
945 	if (error != 0)
946 		return (error);
947 
948 	switch (type) {
949 	case SPECIALFD_EVENTFD:
950 		ae = arg;
951 		if ((ae->flags & EFD_CLOEXEC) != 0)
952 			fflags |= O_CLOEXEC;
953 		error = eventfd_create_file(td, fp, ae->initval, ae->flags);
954 		break;
955 	default:
956 		error = EINVAL;
957 		break;
958 	}
959 
960 	if (error == 0)
961 		error = finstall(td, fp, &fd, fflags, NULL);
962 	fdrop(fp, td);
963 	if (error == 0)
964 		td->td_retval[0] = fd;
965 	return (error);
966 }
967 
968 int
969 sys___specialfd(struct thread *td, struct __specialfd_args *args)
970 {
971 	struct specialfd_eventfd ae;
972 	int error;
973 
974 	switch (args->type) {
975 	case SPECIALFD_EVENTFD:
976 		if (args->len != sizeof(struct specialfd_eventfd)) {
977 			error = EINVAL;
978 			break;
979 		}
980 		error = copyin(args->req, &ae, sizeof(ae));
981 		if (error != 0)
982 			break;
983 		if ((ae.flags & ~(EFD_CLOEXEC | EFD_NONBLOCK |
984 		    EFD_SEMAPHORE)) != 0) {
985 			error = EINVAL;
986 			break;
987 		}
988 		error = kern_specialfd(td, args->type, &ae);
989 		break;
990 	default:
991 		error = EINVAL;
992 		break;
993 	}
994 	return (error);
995 }
996 
997 int
998 poll_no_poll(int events)
999 {
1000 	/*
1001 	 * Return true for read/write.  If the user asked for something
1002 	 * special, return POLLNVAL, so that clients have a way of
1003 	 * determining reliably whether or not the extended
1004 	 * functionality is present without hard-coding knowledge
1005 	 * of specific filesystem implementations.
1006 	 */
1007 	if (events & ~POLLSTANDARD)
1008 		return (POLLNVAL);
1009 
1010 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
1011 }
1012 
1013 int
1014 sys_pselect(struct thread *td, struct pselect_args *uap)
1015 {
1016 	struct timespec ts;
1017 	struct timeval tv, *tvp;
1018 	sigset_t set, *uset;
1019 	int error;
1020 
1021 	if (uap->ts != NULL) {
1022 		error = copyin(uap->ts, &ts, sizeof(ts));
1023 		if (error != 0)
1024 		    return (error);
1025 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1026 		tvp = &tv;
1027 	} else
1028 		tvp = NULL;
1029 	if (uap->sm != NULL) {
1030 		error = copyin(uap->sm, &set, sizeof(set));
1031 		if (error != 0)
1032 			return (error);
1033 		uset = &set;
1034 	} else
1035 		uset = NULL;
1036 	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
1037 	    uset, NFDBITS));
1038 }
1039 
1040 int
1041 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
1042     struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
1043 {
1044 	int error;
1045 
1046 	if (uset != NULL) {
1047 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
1048 		    &td->td_oldsigmask, 0);
1049 		if (error != 0)
1050 			return (error);
1051 		td->td_pflags |= TDP_OLDMASK;
1052 		/*
1053 		 * Make sure that ast() is called on return to
1054 		 * usermode and TDP_OLDMASK is cleared, restoring old
1055 		 * sigmask.
1056 		 */
1057 		ast_sched(td, TDA_SIGSUSPEND);
1058 	}
1059 	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
1060 	return (error);
1061 }
1062 
1063 #ifndef _SYS_SYSPROTO_H_
1064 struct select_args {
1065 	int	nd;
1066 	fd_set	*in, *ou, *ex;
1067 	struct	timeval *tv;
1068 };
1069 #endif
1070 int
1071 sys_select(struct thread *td, struct select_args *uap)
1072 {
1073 	struct timeval tv, *tvp;
1074 	int error;
1075 
1076 	if (uap->tv != NULL) {
1077 		error = copyin(uap->tv, &tv, sizeof(tv));
1078 		if (error)
1079 			return (error);
1080 		tvp = &tv;
1081 	} else
1082 		tvp = NULL;
1083 
1084 	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
1085 	    NFDBITS));
1086 }
1087 
1088 /*
1089  * In the unlikely case when user specified n greater then the last
1090  * open file descriptor, check that no bits are set after the last
1091  * valid fd.  We must return EBADF if any is set.
1092  *
1093  * There are applications that rely on the behaviour.
1094  *
1095  * nd is fd_nfiles.
1096  */
1097 static int
1098 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
1099 {
1100 	char *addr, *oaddr;
1101 	int b, i, res;
1102 	uint8_t bits;
1103 
1104 	if (nd >= ndu || fd_in == NULL)
1105 		return (0);
1106 
1107 	oaddr = NULL;
1108 	bits = 0; /* silence gcc */
1109 	for (i = nd; i < ndu; i++) {
1110 		b = i / NBBY;
1111 #if BYTE_ORDER == LITTLE_ENDIAN
1112 		addr = (char *)fd_in + b;
1113 #else
1114 		addr = (char *)fd_in;
1115 		if (abi_nfdbits == NFDBITS) {
1116 			addr += rounddown(b, sizeof(fd_mask)) +
1117 			    sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
1118 		} else {
1119 			addr += rounddown(b, sizeof(uint32_t)) +
1120 			    sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
1121 		}
1122 #endif
1123 		if (addr != oaddr) {
1124 			res = fubyte(addr);
1125 			if (res == -1)
1126 				return (EFAULT);
1127 			oaddr = addr;
1128 			bits = res;
1129 		}
1130 		if ((bits & (1 << (i % NBBY))) != 0)
1131 			return (EBADF);
1132 	}
1133 	return (0);
1134 }
1135 
1136 int
1137 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
1138     fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
1139 {
1140 	struct filedesc *fdp;
1141 	/*
1142 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
1143 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
1144 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
1145 	 * of 256.
1146 	 */
1147 	fd_mask s_selbits[howmany(2048, NFDBITS)];
1148 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
1149 	struct timeval rtv;
1150 	sbintime_t asbt, precision, rsbt;
1151 	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
1152 	int error, lf, ndu;
1153 
1154 	if (nd < 0)
1155 		return (EINVAL);
1156 	fdp = td->td_proc->p_fd;
1157 	ndu = nd;
1158 	lf = fdp->fd_nfiles;
1159 	if (nd > lf)
1160 		nd = lf;
1161 
1162 	error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1163 	if (error != 0)
1164 		return (error);
1165 	error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1166 	if (error != 0)
1167 		return (error);
1168 	error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1169 	if (error != 0)
1170 		return (error);
1171 
1172 	/*
1173 	 * Allocate just enough bits for the non-null fd_sets.  Use the
1174 	 * preallocated auto buffer if possible.
1175 	 */
1176 	nfdbits = roundup(nd, NFDBITS);
1177 	ncpbytes = nfdbits / NBBY;
1178 	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1179 	nbufbytes = 0;
1180 	if (fd_in != NULL)
1181 		nbufbytes += 2 * ncpbytes;
1182 	if (fd_ou != NULL)
1183 		nbufbytes += 2 * ncpbytes;
1184 	if (fd_ex != NULL)
1185 		nbufbytes += 2 * ncpbytes;
1186 	if (nbufbytes <= sizeof s_selbits)
1187 		selbits = &s_selbits[0];
1188 	else
1189 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1190 
1191 	/*
1192 	 * Assign pointers into the bit buffers and fetch the input bits.
1193 	 * Put the output buffers together so that they can be bzeroed
1194 	 * together.
1195 	 */
1196 	sbp = selbits;
1197 #define	getbits(name, x) \
1198 	do {								\
1199 		if (name == NULL) {					\
1200 			ibits[x] = NULL;				\
1201 			obits[x] = NULL;				\
1202 		} else {						\
1203 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
1204 			obits[x] = sbp;					\
1205 			sbp += ncpbytes / sizeof *sbp;			\
1206 			error = copyin(name, ibits[x], ncpubytes);	\
1207 			if (error != 0)					\
1208 				goto done;				\
1209 			if (ncpbytes != ncpubytes)			\
1210 				bzero((char *)ibits[x] + ncpubytes,	\
1211 				    ncpbytes - ncpubytes);		\
1212 		}							\
1213 	} while (0)
1214 	getbits(fd_in, 0);
1215 	getbits(fd_ou, 1);
1216 	getbits(fd_ex, 2);
1217 #undef	getbits
1218 
1219 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1220 	/*
1221 	 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1222 	 * we are running under 32-bit emulation. This should be more
1223 	 * generic.
1224 	 */
1225 #define swizzle_fdset(bits)						\
1226 	if (abi_nfdbits != NFDBITS && bits != NULL) {			\
1227 		int i;							\
1228 		for (i = 0; i < ncpbytes / sizeof *sbp; i++)		\
1229 			bits[i] = (bits[i] >> 32) | (bits[i] << 32);	\
1230 	}
1231 #else
1232 #define swizzle_fdset(bits)
1233 #endif
1234 
1235 	/* Make sure the bit order makes it through an ABI transition */
1236 	swizzle_fdset(ibits[0]);
1237 	swizzle_fdset(ibits[1]);
1238 	swizzle_fdset(ibits[2]);
1239 
1240 	if (nbufbytes != 0)
1241 		bzero(selbits, nbufbytes / 2);
1242 
1243 	precision = 0;
1244 	if (tvp != NULL) {
1245 		rtv = *tvp;
1246 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1247 		    rtv.tv_usec >= 1000000) {
1248 			error = EINVAL;
1249 			goto done;
1250 		}
1251 		if (!timevalisset(&rtv))
1252 			asbt = 0;
1253 		else if (rtv.tv_sec <= INT32_MAX) {
1254 			rsbt = tvtosbt(rtv);
1255 			precision = rsbt;
1256 			precision >>= tc_precexp;
1257 			if (TIMESEL(&asbt, rsbt))
1258 				asbt += tc_tick_sbt;
1259 			if (asbt <= SBT_MAX - rsbt)
1260 				asbt += rsbt;
1261 			else
1262 				asbt = -1;
1263 		} else
1264 			asbt = -1;
1265 	} else
1266 		asbt = -1;
1267 	seltdinit(td);
1268 	/* Iterate until the timeout expires or descriptors become ready. */
1269 	for (;;) {
1270 		error = selscan(td, ibits, obits, nd);
1271 		if (error || td->td_retval[0] != 0)
1272 			break;
1273 		error = seltdwait(td, asbt, precision);
1274 		if (error)
1275 			break;
1276 		error = selrescan(td, ibits, obits);
1277 		if (error || td->td_retval[0] != 0)
1278 			break;
1279 	}
1280 	seltdclear(td);
1281 
1282 done:
1283 	/* select is not restarted after signals... */
1284 	if (error == ERESTART)
1285 		error = EINTR;
1286 	if (error == EWOULDBLOCK)
1287 		error = 0;
1288 
1289 	/* swizzle bit order back, if necessary */
1290 	swizzle_fdset(obits[0]);
1291 	swizzle_fdset(obits[1]);
1292 	swizzle_fdset(obits[2]);
1293 #undef swizzle_fdset
1294 
1295 #define	putbits(name, x) \
1296 	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1297 		error = error2;
1298 	if (error == 0) {
1299 		int error2;
1300 
1301 		putbits(fd_in, 0);
1302 		putbits(fd_ou, 1);
1303 		putbits(fd_ex, 2);
1304 #undef putbits
1305 	}
1306 	if (selbits != &s_selbits[0])
1307 		free(selbits, M_SELECT);
1308 
1309 	return (error);
1310 }
1311 /*
1312  * Convert a select bit set to poll flags.
1313  *
1314  * The backend always returns POLLHUP/POLLERR if appropriate and we
1315  * return this as a set bit in any set.
1316  */
1317 static const int select_flags[3] = {
1318     POLLRDNORM | POLLHUP | POLLERR,
1319     POLLWRNORM | POLLHUP | POLLERR,
1320     POLLRDBAND | POLLERR
1321 };
1322 
1323 /*
1324  * Compute the fo_poll flags required for a fd given by the index and
1325  * bit position in the fd_mask array.
1326  */
1327 static __inline int
1328 selflags(fd_mask **ibits, int idx, fd_mask bit)
1329 {
1330 	int flags;
1331 	int msk;
1332 
1333 	flags = 0;
1334 	for (msk = 0; msk < 3; msk++) {
1335 		if (ibits[msk] == NULL)
1336 			continue;
1337 		if ((ibits[msk][idx] & bit) == 0)
1338 			continue;
1339 		flags |= select_flags[msk];
1340 	}
1341 	return (flags);
1342 }
1343 
1344 /*
1345  * Set the appropriate output bits given a mask of fired events and the
1346  * input bits originally requested.
1347  */
1348 static __inline int
1349 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1350 {
1351 	int msk;
1352 	int n;
1353 
1354 	n = 0;
1355 	for (msk = 0; msk < 3; msk++) {
1356 		if ((events & select_flags[msk]) == 0)
1357 			continue;
1358 		if (ibits[msk] == NULL)
1359 			continue;
1360 		if ((ibits[msk][idx] & bit) == 0)
1361 			continue;
1362 		/*
1363 		 * XXX Check for a duplicate set.  This can occur because a
1364 		 * socket calls selrecord() twice for each poll() call
1365 		 * resulting in two selfds per real fd.  selrescan() will
1366 		 * call selsetbits twice as a result.
1367 		 */
1368 		if ((obits[msk][idx] & bit) != 0)
1369 			continue;
1370 		obits[msk][idx] |= bit;
1371 		n++;
1372 	}
1373 
1374 	return (n);
1375 }
1376 
1377 /*
1378  * Traverse the list of fds attached to this thread's seltd and check for
1379  * completion.
1380  */
1381 static int
1382 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1383 {
1384 	struct filedesc *fdp;
1385 	struct selinfo *si;
1386 	struct seltd *stp;
1387 	struct selfd *sfp;
1388 	struct selfd *sfn;
1389 	struct file *fp;
1390 	fd_mask bit;
1391 	int fd, ev, n, idx;
1392 	int error;
1393 	bool only_user;
1394 
1395 	fdp = td->td_proc->p_fd;
1396 	stp = td->td_sel;
1397 	n = 0;
1398 	only_user = FILEDESC_IS_ONLY_USER(fdp);
1399 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1400 		fd = (int)(uintptr_t)sfp->sf_cookie;
1401 		si = sfp->sf_si;
1402 		selfdfree(stp, sfp);
1403 		/* If the selinfo wasn't cleared the event didn't fire. */
1404 		if (si != NULL)
1405 			continue;
1406 		if (only_user)
1407 			error = fget_only_user(fdp, fd, &cap_event_rights, &fp);
1408 		else
1409 			error = fget_unlocked(td, fd, &cap_event_rights, &fp);
1410 		if (__predict_false(error != 0))
1411 			return (error);
1412 		idx = fd / NFDBITS;
1413 		bit = (fd_mask)1 << (fd % NFDBITS);
1414 		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1415 		if (only_user)
1416 			fput_only_user(fdp, fp);
1417 		else
1418 			fdrop(fp, td);
1419 		if (ev != 0)
1420 			n += selsetbits(ibits, obits, idx, bit, ev);
1421 	}
1422 	stp->st_flags = 0;
1423 	td->td_retval[0] = n;
1424 	return (0);
1425 }
1426 
1427 /*
1428  * Perform the initial filedescriptor scan and register ourselves with
1429  * each selinfo.
1430  */
1431 static int
1432 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd)
1433 {
1434 	struct filedesc *fdp;
1435 	struct file *fp;
1436 	fd_mask bit;
1437 	int ev, flags, end, fd;
1438 	int n, idx;
1439 	int error;
1440 	bool only_user;
1441 
1442 	fdp = td->td_proc->p_fd;
1443 	n = 0;
1444 	only_user = FILEDESC_IS_ONLY_USER(fdp);
1445 	for (idx = 0, fd = 0; fd < nfd; idx++) {
1446 		end = imin(fd + NFDBITS, nfd);
1447 		for (bit = 1; fd < end; bit <<= 1, fd++) {
1448 			/* Compute the list of events we're interested in. */
1449 			flags = selflags(ibits, idx, bit);
1450 			if (flags == 0)
1451 				continue;
1452 			if (only_user)
1453 				error = fget_only_user(fdp, fd, &cap_event_rights, &fp);
1454 			else
1455 				error = fget_unlocked(td, fd, &cap_event_rights, &fp);
1456 			if (__predict_false(error != 0))
1457 				return (error);
1458 			selfdalloc(td, (void *)(uintptr_t)fd);
1459 			ev = fo_poll(fp, flags, td->td_ucred, td);
1460 			if (only_user)
1461 				fput_only_user(fdp, fp);
1462 			else
1463 				fdrop(fp, td);
1464 			if (ev != 0)
1465 				n += selsetbits(ibits, obits, idx, bit, ev);
1466 		}
1467 	}
1468 
1469 	td->td_retval[0] = n;
1470 	return (0);
1471 }
1472 
1473 int
1474 sys_poll(struct thread *td, struct poll_args *uap)
1475 {
1476 	struct timespec ts, *tsp;
1477 
1478 	if (uap->timeout != INFTIM) {
1479 		if (uap->timeout < 0)
1480 			return (EINVAL);
1481 		ts.tv_sec = uap->timeout / 1000;
1482 		ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1483 		tsp = &ts;
1484 	} else
1485 		tsp = NULL;
1486 
1487 	return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1488 }
1489 
1490 /*
1491  * kfds points to an array in the kernel.
1492  */
1493 int
1494 kern_poll_kfds(struct thread *td, struct pollfd *kfds, u_int nfds,
1495     struct timespec *tsp, sigset_t *uset)
1496 {
1497 	sbintime_t sbt, precision, tmp;
1498 	time_t over;
1499 	struct timespec ts;
1500 	int error;
1501 
1502 	precision = 0;
1503 	if (tsp != NULL) {
1504 		if (!timespecvalid_interval(tsp))
1505 			return (EINVAL);
1506 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1507 			sbt = 0;
1508 		else {
1509 			ts = *tsp;
1510 			if (ts.tv_sec > INT32_MAX / 2) {
1511 				over = ts.tv_sec - INT32_MAX / 2;
1512 				ts.tv_sec -= over;
1513 			} else
1514 				over = 0;
1515 			tmp = tstosbt(ts);
1516 			precision = tmp;
1517 			precision >>= tc_precexp;
1518 			if (TIMESEL(&sbt, tmp))
1519 				sbt += tc_tick_sbt;
1520 			sbt += tmp;
1521 		}
1522 	} else
1523 		sbt = -1;
1524 
1525 	if (uset != NULL) {
1526 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
1527 		    &td->td_oldsigmask, 0);
1528 		if (error)
1529 			return (error);
1530 		td->td_pflags |= TDP_OLDMASK;
1531 		/*
1532 		 * Make sure that ast() is called on return to
1533 		 * usermode and TDP_OLDMASK is cleared, restoring old
1534 		 * sigmask.
1535 		 */
1536 		ast_sched(td, TDA_SIGSUSPEND);
1537 	}
1538 
1539 	seltdinit(td);
1540 	/* Iterate until the timeout expires or descriptors become ready. */
1541 	for (;;) {
1542 		error = pollscan(td, kfds, nfds);
1543 		if (error || td->td_retval[0] != 0)
1544 			break;
1545 		error = seltdwait(td, sbt, precision);
1546 		if (error)
1547 			break;
1548 		error = pollrescan(td);
1549 		if (error || td->td_retval[0] != 0)
1550 			break;
1551 	}
1552 	seltdclear(td);
1553 
1554 	/* poll is not restarted after signals... */
1555 	if (error == ERESTART)
1556 		error = EINTR;
1557 	if (error == EWOULDBLOCK)
1558 		error = 0;
1559 	return (error);
1560 }
1561 
1562 int
1563 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1564 {
1565 	struct timespec ts, *tsp;
1566 	sigset_t set, *ssp;
1567 	int error;
1568 
1569 	if (uap->ts != NULL) {
1570 		error = copyin(uap->ts, &ts, sizeof(ts));
1571 		if (error)
1572 			return (error);
1573 		tsp = &ts;
1574 	} else
1575 		tsp = NULL;
1576 	if (uap->set != NULL) {
1577 		error = copyin(uap->set, &set, sizeof(set));
1578 		if (error)
1579 			return (error);
1580 		ssp = &set;
1581 	} else
1582 		ssp = NULL;
1583 	return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1584 }
1585 
1586 /*
1587  * ufds points to an array in user space.
1588  */
1589 int
1590 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds,
1591     struct timespec *tsp, sigset_t *set)
1592 {
1593 	struct pollfd *kfds;
1594 	struct pollfd stackfds[32];
1595 	int error;
1596 
1597 	if (kern_poll_maxfds(nfds))
1598 		return (EINVAL);
1599 	if (nfds > nitems(stackfds))
1600 		kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK);
1601 	else
1602 		kfds = stackfds;
1603 	error = copyin(ufds, kfds, nfds * sizeof(*kfds));
1604 	if (error != 0)
1605 		goto out;
1606 
1607 	error = kern_poll_kfds(td, kfds, nfds, tsp, set);
1608 	if (error == 0)
1609 		error = pollout(td, kfds, ufds, nfds);
1610 #ifdef KTRACE
1611 	if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY))
1612 		ktrstructarray("pollfd", UIO_USERSPACE, ufds, nfds,
1613 		    sizeof(*ufds));
1614 #endif
1615 
1616 out:
1617 	if (nfds > nitems(stackfds))
1618 		free(kfds, M_TEMP);
1619 	return (error);
1620 }
1621 
1622 bool
1623 kern_poll_maxfds(u_int nfds)
1624 {
1625 
1626 	/*
1627 	 * This is kinda bogus.  We have fd limits, but that is not
1628 	 * really related to the size of the pollfd array.  Make sure
1629 	 * we let the process use at least FD_SETSIZE entries and at
1630 	 * least enough for the system-wide limits.  We want to be reasonably
1631 	 * safe, but not overly restrictive.
1632 	 */
1633 	return (nfds > maxfilesperproc && nfds > FD_SETSIZE);
1634 }
1635 
1636 static int
1637 pollrescan(struct thread *td)
1638 {
1639 	struct seltd *stp;
1640 	struct selfd *sfp;
1641 	struct selfd *sfn;
1642 	struct selinfo *si;
1643 	struct filedesc *fdp;
1644 	struct file *fp;
1645 	struct pollfd *fd;
1646 	int n, error;
1647 	bool only_user;
1648 
1649 	n = 0;
1650 	fdp = td->td_proc->p_fd;
1651 	stp = td->td_sel;
1652 	only_user = FILEDESC_IS_ONLY_USER(fdp);
1653 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1654 		fd = (struct pollfd *)sfp->sf_cookie;
1655 		si = sfp->sf_si;
1656 		selfdfree(stp, sfp);
1657 		/* If the selinfo wasn't cleared the event didn't fire. */
1658 		if (si != NULL)
1659 			continue;
1660 		if (only_user)
1661 			error = fget_only_user(fdp, fd->fd, &cap_event_rights, &fp);
1662 		else
1663 			error = fget_unlocked(td, fd->fd, &cap_event_rights, &fp);
1664 		if (__predict_false(error != 0)) {
1665 			fd->revents = POLLNVAL;
1666 			n++;
1667 			continue;
1668 		}
1669 		/*
1670 		 * Note: backend also returns POLLHUP and
1671 		 * POLLERR if appropriate.
1672 		 */
1673 		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1674 		if (only_user)
1675 			fput_only_user(fdp, fp);
1676 		else
1677 			fdrop(fp, td);
1678 		if (fd->revents != 0)
1679 			n++;
1680 	}
1681 	stp->st_flags = 0;
1682 	td->td_retval[0] = n;
1683 	return (0);
1684 }
1685 
1686 static int
1687 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd)
1688 {
1689 	int error = 0;
1690 	u_int i = 0;
1691 	u_int n = 0;
1692 
1693 	for (i = 0; i < nfd; i++) {
1694 		error = copyout(&fds->revents, &ufds->revents,
1695 		    sizeof(ufds->revents));
1696 		if (error)
1697 			return (error);
1698 		if (fds->revents != 0)
1699 			n++;
1700 		fds++;
1701 		ufds++;
1702 	}
1703 	td->td_retval[0] = n;
1704 	return (0);
1705 }
1706 
1707 static int
1708 pollscan(struct thread *td, struct pollfd *fds, u_int nfd)
1709 {
1710 	struct filedesc *fdp;
1711 	struct file *fp;
1712 	int i, n, error;
1713 	bool only_user;
1714 
1715 	n = 0;
1716 	fdp = td->td_proc->p_fd;
1717 	only_user = FILEDESC_IS_ONLY_USER(fdp);
1718 	for (i = 0; i < nfd; i++, fds++) {
1719 		if (fds->fd < 0) {
1720 			fds->revents = 0;
1721 			continue;
1722 		}
1723 		if (only_user)
1724 			error = fget_only_user(fdp, fds->fd, &cap_event_rights, &fp);
1725 		else
1726 			error = fget_unlocked(td, fds->fd, &cap_event_rights, &fp);
1727 		if (__predict_false(error != 0)) {
1728 			fds->revents = POLLNVAL;
1729 			n++;
1730 			continue;
1731 		}
1732 		/*
1733 		 * Note: backend also returns POLLHUP and
1734 		 * POLLERR if appropriate.
1735 		 */
1736 		selfdalloc(td, fds);
1737 		fds->revents = fo_poll(fp, fds->events,
1738 		    td->td_ucred, td);
1739 		if (only_user)
1740 			fput_only_user(fdp, fp);
1741 		else
1742 			fdrop(fp, td);
1743 		/*
1744 		 * POSIX requires POLLOUT to be never
1745 		 * set simultaneously with POLLHUP.
1746 		 */
1747 		if ((fds->revents & POLLHUP) != 0)
1748 			fds->revents &= ~POLLOUT;
1749 
1750 		if (fds->revents != 0)
1751 			n++;
1752 	}
1753 	td->td_retval[0] = n;
1754 	return (0);
1755 }
1756 
1757 /*
1758  * XXX This was created specifically to support netncp and netsmb.  This
1759  * allows the caller to specify a socket to wait for events on.  It returns
1760  * 0 if any events matched and an error otherwise.  There is no way to
1761  * determine which events fired.
1762  */
1763 int
1764 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1765 {
1766 	struct timeval rtv;
1767 	sbintime_t asbt, precision, rsbt;
1768 	int error;
1769 
1770 	precision = 0;	/* stupid gcc! */
1771 	if (tvp != NULL) {
1772 		rtv = *tvp;
1773 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1774 		    rtv.tv_usec >= 1000000)
1775 			return (EINVAL);
1776 		if (!timevalisset(&rtv))
1777 			asbt = 0;
1778 		else if (rtv.tv_sec <= INT32_MAX) {
1779 			rsbt = tvtosbt(rtv);
1780 			precision = rsbt;
1781 			precision >>= tc_precexp;
1782 			if (TIMESEL(&asbt, rsbt))
1783 				asbt += tc_tick_sbt;
1784 			if (asbt <= SBT_MAX - rsbt)
1785 				asbt += rsbt;
1786 			else
1787 				asbt = -1;
1788 		} else
1789 			asbt = -1;
1790 	} else
1791 		asbt = -1;
1792 	seltdinit(td);
1793 	/*
1794 	 * Iterate until the timeout expires or the socket becomes ready.
1795 	 */
1796 	for (;;) {
1797 		selfdalloc(td, NULL);
1798 		if (sopoll(so, events, NULL, td) != 0) {
1799 			error = 0;
1800 			break;
1801 		}
1802 		error = seltdwait(td, asbt, precision);
1803 		if (error)
1804 			break;
1805 	}
1806 	seltdclear(td);
1807 	/* XXX Duplicates ncp/smb behavior. */
1808 	if (error == ERESTART)
1809 		error = 0;
1810 	return (error);
1811 }
1812 
1813 /*
1814  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1815  * have two select sets, one for read and another for write.
1816  */
1817 static void
1818 selfdalloc(struct thread *td, void *cookie)
1819 {
1820 	struct seltd *stp;
1821 
1822 	stp = td->td_sel;
1823 	if (stp->st_free1 == NULL)
1824 		stp->st_free1 = malloc(sizeof(*stp->st_free1), M_SELFD, M_WAITOK|M_ZERO);
1825 	stp->st_free1->sf_td = stp;
1826 	stp->st_free1->sf_cookie = cookie;
1827 	if (stp->st_free2 == NULL)
1828 		stp->st_free2 = malloc(sizeof(*stp->st_free2), M_SELFD, M_WAITOK|M_ZERO);
1829 	stp->st_free2->sf_td = stp;
1830 	stp->st_free2->sf_cookie = cookie;
1831 }
1832 
1833 static void
1834 selfdfree(struct seltd *stp, struct selfd *sfp)
1835 {
1836 	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1837 	/*
1838 	 * Paired with doselwakeup.
1839 	 */
1840 	if (atomic_load_acq_ptr((uintptr_t *)&sfp->sf_si) != (uintptr_t)NULL) {
1841 		mtx_lock(sfp->sf_mtx);
1842 		if (sfp->sf_si != NULL) {
1843 			TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1844 		}
1845 		mtx_unlock(sfp->sf_mtx);
1846 	}
1847 	free(sfp, M_SELFD);
1848 }
1849 
1850 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1851 void
1852 seldrain(struct selinfo *sip)
1853 {
1854 
1855 	/*
1856 	 * This feature is already provided by doselwakeup(), thus it is
1857 	 * enough to go for it.
1858 	 * Eventually, the context, should take care to avoid races
1859 	 * between thread calling select()/poll() and file descriptor
1860 	 * detaching, but, again, the races are just the same as
1861 	 * selwakeup().
1862 	 */
1863         doselwakeup(sip, -1);
1864 }
1865 
1866 /*
1867  * Record a select request.
1868  */
1869 void
1870 selrecord(struct thread *selector, struct selinfo *sip)
1871 {
1872 	struct selfd *sfp;
1873 	struct seltd *stp;
1874 	struct mtx *mtxp;
1875 
1876 	stp = selector->td_sel;
1877 	/*
1878 	 * Don't record when doing a rescan.
1879 	 */
1880 	if (stp->st_flags & SELTD_RESCAN)
1881 		return;
1882 	/*
1883 	 * Grab one of the preallocated descriptors.
1884 	 */
1885 	sfp = NULL;
1886 	if ((sfp = stp->st_free1) != NULL)
1887 		stp->st_free1 = NULL;
1888 	else if ((sfp = stp->st_free2) != NULL)
1889 		stp->st_free2 = NULL;
1890 	else
1891 		panic("selrecord: No free selfd on selq");
1892 	mtxp = sip->si_mtx;
1893 	if (mtxp == NULL)
1894 		mtxp = mtx_pool_find(mtxpool_select, sip);
1895 	/*
1896 	 * Initialize the sfp and queue it in the thread.
1897 	 */
1898 	sfp->sf_si = sip;
1899 	sfp->sf_mtx = mtxp;
1900 	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1901 	/*
1902 	 * Now that we've locked the sip, check for initialization.
1903 	 */
1904 	mtx_lock(mtxp);
1905 	if (sip->si_mtx == NULL) {
1906 		sip->si_mtx = mtxp;
1907 		TAILQ_INIT(&sip->si_tdlist);
1908 	}
1909 	/*
1910 	 * Add this thread to the list of selfds listening on this selinfo.
1911 	 */
1912 	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1913 	mtx_unlock(sip->si_mtx);
1914 }
1915 
1916 /* Wake up a selecting thread. */
1917 void
1918 selwakeup(struct selinfo *sip)
1919 {
1920 	doselwakeup(sip, -1);
1921 }
1922 
1923 /* Wake up a selecting thread, and set its priority. */
1924 void
1925 selwakeuppri(struct selinfo *sip, int pri)
1926 {
1927 	doselwakeup(sip, pri);
1928 }
1929 
1930 /*
1931  * Do a wakeup when a selectable event occurs.
1932  */
1933 static void
1934 doselwakeup(struct selinfo *sip, int pri)
1935 {
1936 	struct selfd *sfp;
1937 	struct selfd *sfn;
1938 	struct seltd *stp;
1939 
1940 	/* If it's not initialized there can't be any waiters. */
1941 	if (sip->si_mtx == NULL)
1942 		return;
1943 	/*
1944 	 * Locking the selinfo locks all selfds associated with it.
1945 	 */
1946 	mtx_lock(sip->si_mtx);
1947 	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1948 		/*
1949 		 * Once we remove this sfp from the list and clear the
1950 		 * sf_si seltdclear will know to ignore this si.
1951 		 */
1952 		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1953 		stp = sfp->sf_td;
1954 		mtx_lock(&stp->st_mtx);
1955 		stp->st_flags |= SELTD_PENDING;
1956 		cv_broadcastpri(&stp->st_wait, pri);
1957 		mtx_unlock(&stp->st_mtx);
1958 		/*
1959 		 * Paired with selfdfree.
1960 		 *
1961 		 * Storing this only after the wakeup provides an invariant that
1962 		 * stp is not used after selfdfree returns.
1963 		 */
1964 		atomic_store_rel_ptr((uintptr_t *)&sfp->sf_si, (uintptr_t)NULL);
1965 	}
1966 	mtx_unlock(sip->si_mtx);
1967 }
1968 
1969 static void
1970 seltdinit(struct thread *td)
1971 {
1972 	struct seltd *stp;
1973 
1974 	stp = td->td_sel;
1975 	if (stp != NULL) {
1976 		MPASS(stp->st_flags == 0);
1977 		MPASS(STAILQ_EMPTY(&stp->st_selq));
1978 		return;
1979 	}
1980 	stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1981 	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1982 	cv_init(&stp->st_wait, "select");
1983 	stp->st_flags = 0;
1984 	STAILQ_INIT(&stp->st_selq);
1985 	td->td_sel = stp;
1986 }
1987 
1988 static int
1989 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1990 {
1991 	struct seltd *stp;
1992 	int error;
1993 
1994 	stp = td->td_sel;
1995 	/*
1996 	 * An event of interest may occur while we do not hold the seltd
1997 	 * locked so check the pending flag before we sleep.
1998 	 */
1999 	mtx_lock(&stp->st_mtx);
2000 	/*
2001 	 * Any further calls to selrecord will be a rescan.
2002 	 */
2003 	stp->st_flags |= SELTD_RESCAN;
2004 	if (stp->st_flags & SELTD_PENDING) {
2005 		mtx_unlock(&stp->st_mtx);
2006 		return (0);
2007 	}
2008 	if (sbt == 0)
2009 		error = EWOULDBLOCK;
2010 	else if (sbt != -1)
2011 		error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
2012 		    sbt, precision, C_ABSOLUTE);
2013 	else
2014 		error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
2015 	mtx_unlock(&stp->st_mtx);
2016 
2017 	return (error);
2018 }
2019 
2020 void
2021 seltdfini(struct thread *td)
2022 {
2023 	struct seltd *stp;
2024 
2025 	stp = td->td_sel;
2026 	if (stp == NULL)
2027 		return;
2028 	MPASS(stp->st_flags == 0);
2029 	MPASS(STAILQ_EMPTY(&stp->st_selq));
2030 	if (stp->st_free1)
2031 		free(stp->st_free1, M_SELFD);
2032 	if (stp->st_free2)
2033 		free(stp->st_free2, M_SELFD);
2034 	td->td_sel = NULL;
2035 	cv_destroy(&stp->st_wait);
2036 	mtx_destroy(&stp->st_mtx);
2037 	free(stp, M_SELECT);
2038 }
2039 
2040 /*
2041  * Remove the references to the thread from all of the objects we were
2042  * polling.
2043  */
2044 static void
2045 seltdclear(struct thread *td)
2046 {
2047 	struct seltd *stp;
2048 	struct selfd *sfp;
2049 	struct selfd *sfn;
2050 
2051 	stp = td->td_sel;
2052 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
2053 		selfdfree(stp, sfp);
2054 	stp->st_flags = 0;
2055 }
2056 
2057 static void selectinit(void *);
2058 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
2059 static void
2060 selectinit(void *dummy __unused)
2061 {
2062 
2063 	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
2064 }
2065 
2066 /*
2067  * Set up a syscall return value that follows the convention specified for
2068  * posix_* functions.
2069  */
2070 int
2071 kern_posix_error(struct thread *td, int error)
2072 {
2073 
2074 	if (error <= 0)
2075 		return (error);
2076 	td->td_errno = error;
2077 	td->td_pflags |= TDP_NERRNO;
2078 	td->td_retval[0] = error;
2079 	return (0);
2080 }
2081 
2082 int
2083 kcmp_cmp(uintptr_t a, uintptr_t b)
2084 {
2085 	if (a == b)
2086 		return (0);
2087 	else if (a < b)
2088 		return (1);
2089 	return (2);
2090 }
2091 
2092 static int
2093 kcmp_pget(struct thread *td, pid_t pid, struct proc **pp)
2094 {
2095 	int error;
2096 
2097 	if (pid == td->td_proc->p_pid) {
2098 		*pp = td->td_proc;
2099 		return (0);
2100 	}
2101 	error = pget(pid, PGET_NOTID | PGET_CANDEBUG | PGET_NOTWEXIT |
2102 	    PGET_HOLD, pp);
2103 	MPASS(*pp != td->td_proc);
2104 	return (error);
2105 }
2106 
2107 int
2108 kern_kcmp(struct thread *td, pid_t pid1, pid_t pid2, int type,
2109     uintptr_t idx1, uintptr_t idx2)
2110 {
2111 	struct proc *p1, *p2;
2112 	struct file *fp1, *fp2;
2113 	int error, res;
2114 
2115 	res = -1;
2116 	p1 = p2 = NULL;
2117 	error = kcmp_pget(td, pid1, &p1);
2118 	if (error == 0)
2119 		error = kcmp_pget(td, pid2, &p2);
2120 	if (error != 0)
2121 		goto out;
2122 
2123 	switch (type) {
2124 	case KCMP_FILE:
2125 	case KCMP_FILEOBJ:
2126 		error = fget_remote(td, p1, idx1, &fp1);
2127 		if (error == 0) {
2128 			error = fget_remote(td, p2, idx2, &fp2);
2129 			if (error == 0) {
2130 				if (type == KCMP_FILEOBJ)
2131 					res = fo_cmp(fp1, fp2, td);
2132 				else
2133 					res = kcmp_cmp((uintptr_t)fp1,
2134 					    (uintptr_t)fp2);
2135 				fdrop(fp2, td);
2136 			}
2137 			fdrop(fp1, td);
2138 		}
2139 		break;
2140 	case KCMP_FILES:
2141 		res = kcmp_cmp((uintptr_t)p1->p_fd, (uintptr_t)p2->p_fd);
2142 		break;
2143 	case KCMP_SIGHAND:
2144 		res = kcmp_cmp((uintptr_t)p1->p_sigacts,
2145 		    (uintptr_t)p2->p_sigacts);
2146 		break;
2147 	case KCMP_VM:
2148 		res = kcmp_cmp((uintptr_t)p1->p_vmspace,
2149 		    (uintptr_t)p2->p_vmspace);
2150 		break;
2151 	default:
2152 		error = EINVAL;
2153 		break;
2154 	}
2155 
2156 out:
2157 	if (p1 != NULL && p1 != td->td_proc)
2158 		PRELE(p1);
2159 	if (p2 != NULL && p2 != td->td_proc)
2160 		PRELE(p2);
2161 
2162 	td->td_retval[0] = res;
2163 	return (error);
2164 }
2165 
2166 int
2167 sys_kcmp(struct thread *td, struct kcmp_args *uap)
2168 {
2169 	return (kern_kcmp(td, uap->pid1, uap->pid2, uap->type,
2170 	    uap->idx1, uap->idx2));
2171 }
2172 
2173 int
2174 file_kcmp_generic(struct file *fp1, struct file *fp2, struct thread *td)
2175 {
2176 	if (fp1->f_type != fp2->f_type)
2177 		return (3);
2178 	return (kcmp_cmp((uintptr_t)fp1->f_data, (uintptr_t)fp2->f_data));
2179 }
2180