xref: /freebsd/sys/kern/sys_generic.c (revision b633e08c705fe43180567eae26923d6f6f98c8d9)
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/eventfd.h>
59 #include <sys/kernel.h>
60 #include <sys/ktr.h>
61 #include <sys/limits.h>
62 #include <sys/malloc.h>
63 #include <sys/poll.h>
64 #include <sys/resourcevar.h>
65 #include <sys/selinfo.h>
66 #include <sys/sleepqueue.h>
67 #include <sys/specialfd.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/vnode.h>
72 #include <sys/bio.h>
73 #include <sys/buf.h>
74 #include <sys/condvar.h>
75 #ifdef KTRACE
76 #include <sys/ktrace.h>
77 #endif
78 
79 #include <security/audit/audit.h>
80 
81 /*
82  * The following macro defines how many bytes will be allocated from
83  * the stack instead of memory allocated when passing the IOCTL data
84  * structures from userspace and to the kernel. Some IOCTLs having
85  * small data structures are used very frequently and this small
86  * buffer on the stack gives a significant speedup improvement for
87  * those requests. The value of this define should be greater or equal
88  * to 64 bytes and should also be power of two. The data structure is
89  * currently hard-aligned to a 8-byte boundary on the stack. This
90  * should currently be sufficient for all supported platforms.
91  */
92 #define	SYS_IOCTL_SMALL_SIZE	128	/* bytes */
93 #define	SYS_IOCTL_SMALL_ALIGN	8	/* bytes */
94 
95 #ifdef __LP64__
96 static int iosize_max_clamp = 0;
97 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
98     &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
99 static int devfs_iosize_max_clamp = 1;
100 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
101     &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
102 #endif
103 
104 /*
105  * Assert that the return value of read(2) and write(2) syscalls fits
106  * into a register.  If not, an architecture will need to provide the
107  * usermode wrappers to reconstruct the result.
108  */
109 CTASSERT(sizeof(register_t) >= sizeof(size_t));
110 
111 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
112 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
113 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
114 
115 static int	pollout(struct thread *, struct pollfd *, struct pollfd *,
116 		    u_int);
117 static int	pollscan(struct thread *, struct pollfd *, u_int);
118 static int	pollrescan(struct thread *);
119 static int	selscan(struct thread *, fd_mask **, fd_mask **, int);
120 static int	selrescan(struct thread *, fd_mask **, fd_mask **);
121 static void	selfdalloc(struct thread *, void *);
122 static void	selfdfree(struct seltd *, struct selfd *);
123 static int	dofileread(struct thread *, int, struct file *, struct uio *,
124 		    off_t, int);
125 static int	dofilewrite(struct thread *, int, struct file *, struct uio *,
126 		    off_t, int);
127 static void	doselwakeup(struct selinfo *, int);
128 static void	seltdinit(struct thread *);
129 static int	seltdwait(struct thread *, sbintime_t, sbintime_t);
130 static void	seltdclear(struct thread *);
131 
132 /*
133  * One seltd per-thread allocated on demand as needed.
134  *
135  *	t - protected by st_mtx
136  * 	k - Only accessed by curthread or read-only
137  */
138 struct seltd {
139 	STAILQ_HEAD(, selfd)	st_selq;	/* (k) List of selfds. */
140 	struct selfd		*st_free1;	/* (k) free fd for read set. */
141 	struct selfd		*st_free2;	/* (k) free fd for write set. */
142 	struct mtx		st_mtx;		/* Protects struct seltd */
143 	struct cv		st_wait;	/* (t) Wait channel. */
144 	int			st_flags;	/* (t) SELTD_ flags. */
145 };
146 
147 #define	SELTD_PENDING	0x0001			/* We have pending events. */
148 #define	SELTD_RESCAN	0x0002			/* Doing a rescan. */
149 
150 /*
151  * One selfd allocated per-thread per-file-descriptor.
152  *	f - protected by sf_mtx
153  */
154 struct selfd {
155 	STAILQ_ENTRY(selfd)	sf_link;	/* (k) fds owned by this td. */
156 	TAILQ_ENTRY(selfd)	sf_threads;	/* (f) fds on this selinfo. */
157 	struct selinfo		*sf_si;		/* (f) selinfo when linked. */
158 	struct mtx		*sf_mtx;	/* Pointer to selinfo mtx. */
159 	struct seltd		*sf_td;		/* (k) owning seltd. */
160 	void			*sf_cookie;	/* (k) fd or pollfd. */
161 };
162 
163 MALLOC_DEFINE(M_SELFD, "selfd", "selfd");
164 static struct mtx_pool *mtxpool_select;
165 
166 #ifdef __LP64__
167 size_t
168 devfs_iosize_max(void)
169 {
170 
171 	return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
172 	    INT_MAX : SSIZE_MAX);
173 }
174 
175 size_t
176 iosize_max(void)
177 {
178 
179 	return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
180 	    INT_MAX : SSIZE_MAX);
181 }
182 #endif
183 
184 #ifndef _SYS_SYSPROTO_H_
185 struct read_args {
186 	int	fd;
187 	void	*buf;
188 	size_t	nbyte;
189 };
190 #endif
191 int
192 sys_read(struct thread *td, struct read_args *uap)
193 {
194 	struct uio auio;
195 	struct iovec aiov;
196 	int error;
197 
198 	if (uap->nbyte > IOSIZE_MAX)
199 		return (EINVAL);
200 	aiov.iov_base = uap->buf;
201 	aiov.iov_len = uap->nbyte;
202 	auio.uio_iov = &aiov;
203 	auio.uio_iovcnt = 1;
204 	auio.uio_resid = uap->nbyte;
205 	auio.uio_segflg = UIO_USERSPACE;
206 	error = kern_readv(td, uap->fd, &auio);
207 	return (error);
208 }
209 
210 /*
211  * Positioned read system call
212  */
213 #ifndef _SYS_SYSPROTO_H_
214 struct pread_args {
215 	int	fd;
216 	void	*buf;
217 	size_t	nbyte;
218 	int	pad;
219 	off_t	offset;
220 };
221 #endif
222 int
223 sys_pread(struct thread *td, struct pread_args *uap)
224 {
225 
226 	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
227 }
228 
229 int
230 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset)
231 {
232 	struct uio auio;
233 	struct iovec aiov;
234 	int error;
235 
236 	if (nbyte > IOSIZE_MAX)
237 		return (EINVAL);
238 	aiov.iov_base = buf;
239 	aiov.iov_len = nbyte;
240 	auio.uio_iov = &aiov;
241 	auio.uio_iovcnt = 1;
242 	auio.uio_resid = nbyte;
243 	auio.uio_segflg = UIO_USERSPACE;
244 	error = kern_preadv(td, fd, &auio, offset);
245 	return (error);
246 }
247 
248 #if defined(COMPAT_FREEBSD6)
249 int
250 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap)
251 {
252 
253 	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
254 }
255 #endif
256 
257 /*
258  * Scatter read system call.
259  */
260 #ifndef _SYS_SYSPROTO_H_
261 struct readv_args {
262 	int	fd;
263 	struct	iovec *iovp;
264 	u_int	iovcnt;
265 };
266 #endif
267 int
268 sys_readv(struct thread *td, struct readv_args *uap)
269 {
270 	struct uio *auio;
271 	int error;
272 
273 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
274 	if (error)
275 		return (error);
276 	error = kern_readv(td, uap->fd, auio);
277 	free(auio, M_IOV);
278 	return (error);
279 }
280 
281 int
282 kern_readv(struct thread *td, int fd, struct uio *auio)
283 {
284 	struct file *fp;
285 	int error;
286 
287 	error = fget_read(td, fd, &cap_read_rights, &fp);
288 	if (error)
289 		return (error);
290 	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
291 	fdrop(fp, td);
292 	return (error);
293 }
294 
295 /*
296  * Scatter positioned read system call.
297  */
298 #ifndef _SYS_SYSPROTO_H_
299 struct preadv_args {
300 	int	fd;
301 	struct	iovec *iovp;
302 	u_int	iovcnt;
303 	off_t	offset;
304 };
305 #endif
306 int
307 sys_preadv(struct thread *td, struct preadv_args *uap)
308 {
309 	struct uio *auio;
310 	int error;
311 
312 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
313 	if (error)
314 		return (error);
315 	error = kern_preadv(td, uap->fd, auio, uap->offset);
316 	free(auio, M_IOV);
317 	return (error);
318 }
319 
320 int
321 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset)
322 {
323 	struct file *fp;
324 	int error;
325 
326 	error = fget_read(td, fd, &cap_pread_rights, &fp);
327 	if (error)
328 		return (error);
329 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
330 		error = ESPIPE;
331 	else if (offset < 0 &&
332 	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
333 		error = EINVAL;
334 	else
335 		error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
336 	fdrop(fp, td);
337 	return (error);
338 }
339 
340 /*
341  * Common code for readv and preadv that reads data in
342  * from a file using the passed in uio, offset, and flags.
343  */
344 static int
345 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio,
346     off_t offset, int flags)
347 {
348 	ssize_t cnt;
349 	int error;
350 #ifdef KTRACE
351 	struct uio *ktruio = NULL;
352 #endif
353 
354 	AUDIT_ARG_FD(fd);
355 
356 	/* Finish zero length reads right here */
357 	if (auio->uio_resid == 0) {
358 		td->td_retval[0] = 0;
359 		return (0);
360 	}
361 	auio->uio_rw = UIO_READ;
362 	auio->uio_offset = offset;
363 	auio->uio_td = td;
364 #ifdef KTRACE
365 	if (KTRPOINT(td, KTR_GENIO))
366 		ktruio = cloneuio(auio);
367 #endif
368 	cnt = auio->uio_resid;
369 	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
370 		if (auio->uio_resid != cnt && (error == ERESTART ||
371 		    error == EINTR || error == EWOULDBLOCK))
372 			error = 0;
373 	}
374 	cnt -= auio->uio_resid;
375 #ifdef KTRACE
376 	if (ktruio != NULL) {
377 		ktruio->uio_resid = cnt;
378 		ktrgenio(fd, UIO_READ, ktruio, error);
379 	}
380 #endif
381 	td->td_retval[0] = cnt;
382 	return (error);
383 }
384 
385 #ifndef _SYS_SYSPROTO_H_
386 struct write_args {
387 	int	fd;
388 	const void *buf;
389 	size_t	nbyte;
390 };
391 #endif
392 int
393 sys_write(struct thread *td, struct write_args *uap)
394 {
395 	struct uio auio;
396 	struct iovec aiov;
397 	int error;
398 
399 	if (uap->nbyte > IOSIZE_MAX)
400 		return (EINVAL);
401 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
402 	aiov.iov_len = uap->nbyte;
403 	auio.uio_iov = &aiov;
404 	auio.uio_iovcnt = 1;
405 	auio.uio_resid = uap->nbyte;
406 	auio.uio_segflg = UIO_USERSPACE;
407 	error = kern_writev(td, uap->fd, &auio);
408 	return (error);
409 }
410 
411 /*
412  * Positioned write system call.
413  */
414 #ifndef _SYS_SYSPROTO_H_
415 struct pwrite_args {
416 	int	fd;
417 	const void *buf;
418 	size_t	nbyte;
419 	int	pad;
420 	off_t	offset;
421 };
422 #endif
423 int
424 sys_pwrite(struct thread *td, struct pwrite_args *uap)
425 {
426 
427 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
428 }
429 
430 int
431 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte,
432     off_t offset)
433 {
434 	struct uio auio;
435 	struct iovec aiov;
436 	int error;
437 
438 	if (nbyte > IOSIZE_MAX)
439 		return (EINVAL);
440 	aiov.iov_base = (void *)(uintptr_t)buf;
441 	aiov.iov_len = nbyte;
442 	auio.uio_iov = &aiov;
443 	auio.uio_iovcnt = 1;
444 	auio.uio_resid = nbyte;
445 	auio.uio_segflg = UIO_USERSPACE;
446 	error = kern_pwritev(td, fd, &auio, offset);
447 	return (error);
448 }
449 
450 #if defined(COMPAT_FREEBSD6)
451 int
452 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap)
453 {
454 
455 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
456 }
457 #endif
458 
459 /*
460  * Gather write system call.
461  */
462 #ifndef _SYS_SYSPROTO_H_
463 struct writev_args {
464 	int	fd;
465 	struct	iovec *iovp;
466 	u_int	iovcnt;
467 };
468 #endif
469 int
470 sys_writev(struct thread *td, struct writev_args *uap)
471 {
472 	struct uio *auio;
473 	int error;
474 
475 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
476 	if (error)
477 		return (error);
478 	error = kern_writev(td, uap->fd, auio);
479 	free(auio, M_IOV);
480 	return (error);
481 }
482 
483 int
484 kern_writev(struct thread *td, int fd, struct uio *auio)
485 {
486 	struct file *fp;
487 	int error;
488 
489 	error = fget_write(td, fd, &cap_write_rights, &fp);
490 	if (error)
491 		return (error);
492 	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
493 	fdrop(fp, td);
494 	return (error);
495 }
496 
497 /*
498  * Gather positioned write system call.
499  */
500 #ifndef _SYS_SYSPROTO_H_
501 struct pwritev_args {
502 	int	fd;
503 	struct	iovec *iovp;
504 	u_int	iovcnt;
505 	off_t	offset;
506 };
507 #endif
508 int
509 sys_pwritev(struct thread *td, struct pwritev_args *uap)
510 {
511 	struct uio *auio;
512 	int error;
513 
514 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
515 	if (error)
516 		return (error);
517 	error = kern_pwritev(td, uap->fd, auio, uap->offset);
518 	free(auio, M_IOV);
519 	return (error);
520 }
521 
522 int
523 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset)
524 {
525 	struct file *fp;
526 	int error;
527 
528 	error = fget_write(td, fd, &cap_pwrite_rights, &fp);
529 	if (error)
530 		return (error);
531 	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
532 		error = ESPIPE;
533 	else if (offset < 0 &&
534 	    (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
535 		error = EINVAL;
536 	else
537 		error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
538 	fdrop(fp, td);
539 	return (error);
540 }
541 
542 /*
543  * Common code for writev and pwritev that writes data to
544  * a file using the passed in uio, offset, and flags.
545  */
546 static int
547 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio,
548     off_t offset, int flags)
549 {
550 	ssize_t cnt;
551 	int error;
552 #ifdef KTRACE
553 	struct uio *ktruio = NULL;
554 #endif
555 
556 	AUDIT_ARG_FD(fd);
557 	auio->uio_rw = UIO_WRITE;
558 	auio->uio_td = td;
559 	auio->uio_offset = offset;
560 #ifdef KTRACE
561 	if (KTRPOINT(td, KTR_GENIO))
562 		ktruio = cloneuio(auio);
563 #endif
564 	cnt = auio->uio_resid;
565 	if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
566 		if (auio->uio_resid != cnt && (error == ERESTART ||
567 		    error == EINTR || error == EWOULDBLOCK))
568 			error = 0;
569 		/* Socket layer is responsible for issuing SIGPIPE. */
570 		if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
571 			PROC_LOCK(td->td_proc);
572 			tdsignal(td, SIGPIPE);
573 			PROC_UNLOCK(td->td_proc);
574 		}
575 	}
576 	cnt -= auio->uio_resid;
577 #ifdef KTRACE
578 	if (ktruio != NULL) {
579 		ktruio->uio_resid = cnt;
580 		ktrgenio(fd, UIO_WRITE, ktruio, error);
581 	}
582 #endif
583 	td->td_retval[0] = cnt;
584 	return (error);
585 }
586 
587 /*
588  * Truncate a file given a file descriptor.
589  *
590  * Can't use fget_write() here, since must return EINVAL and not EBADF if the
591  * descriptor isn't writable.
592  */
593 int
594 kern_ftruncate(struct thread *td, int fd, off_t length)
595 {
596 	struct file *fp;
597 	int error;
598 
599 	AUDIT_ARG_FD(fd);
600 	if (length < 0)
601 		return (EINVAL);
602 	error = fget(td, fd, &cap_ftruncate_rights, &fp);
603 	if (error)
604 		return (error);
605 	AUDIT_ARG_FILE(td->td_proc, fp);
606 	if (!(fp->f_flag & FWRITE)) {
607 		fdrop(fp, td);
608 		return (EINVAL);
609 	}
610 	error = fo_truncate(fp, length, td->td_ucred, td);
611 	fdrop(fp, td);
612 	return (error);
613 }
614 
615 #ifndef _SYS_SYSPROTO_H_
616 struct ftruncate_args {
617 	int	fd;
618 	int	pad;
619 	off_t	length;
620 };
621 #endif
622 int
623 sys_ftruncate(struct thread *td, struct ftruncate_args *uap)
624 {
625 
626 	return (kern_ftruncate(td, uap->fd, uap->length));
627 }
628 
629 #if defined(COMPAT_43)
630 #ifndef _SYS_SYSPROTO_H_
631 struct oftruncate_args {
632 	int	fd;
633 	long	length;
634 };
635 #endif
636 int
637 oftruncate(struct thread *td, struct oftruncate_args *uap)
638 {
639 
640 	return (kern_ftruncate(td, uap->fd, uap->length));
641 }
642 #endif /* COMPAT_43 */
643 
644 #ifndef _SYS_SYSPROTO_H_
645 struct ioctl_args {
646 	int	fd;
647 	u_long	com;
648 	caddr_t	data;
649 };
650 #endif
651 /* ARGSUSED */
652 int
653 sys_ioctl(struct thread *td, struct ioctl_args *uap)
654 {
655 	u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
656 	uint32_t com;
657 	int arg, error;
658 	u_int size;
659 	caddr_t data;
660 
661 #ifdef INVARIANTS
662 	if (uap->com > 0xffffffff) {
663 		printf(
664 		    "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
665 		    td->td_proc->p_pid, td->td_name, uap->com);
666 	}
667 #endif
668 	com = (uint32_t)uap->com;
669 
670 	/*
671 	 * Interpret high order word to find amount of data to be
672 	 * copied to/from the user's address space.
673 	 */
674 	size = IOCPARM_LEN(com);
675 	if ((size > IOCPARM_MAX) ||
676 	    ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
677 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
678 	    ((com & IOC_OUT) && size == 0) ||
679 #else
680 	    ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
681 #endif
682 	    ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
683 		return (ENOTTY);
684 
685 	if (size > 0) {
686 		if (com & IOC_VOID) {
687 			/* Integer argument. */
688 			arg = (intptr_t)uap->data;
689 			data = (void *)&arg;
690 			size = 0;
691 		} else {
692 			if (size > SYS_IOCTL_SMALL_SIZE)
693 				data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
694 			else
695 				data = smalldata;
696 		}
697 	} else
698 		data = (void *)&uap->data;
699 	if (com & IOC_IN) {
700 		error = copyin(uap->data, data, (u_int)size);
701 		if (error != 0)
702 			goto out;
703 	} else if (com & IOC_OUT) {
704 		/*
705 		 * Zero the buffer so the user always
706 		 * gets back something deterministic.
707 		 */
708 		bzero(data, size);
709 	}
710 
711 	error = kern_ioctl(td, uap->fd, com, data);
712 
713 	if (error == 0 && (com & IOC_OUT))
714 		error = copyout(data, uap->data, (u_int)size);
715 
716 out:
717 	if (size > SYS_IOCTL_SMALL_SIZE)
718 		free(data, M_IOCTLOPS);
719 	return (error);
720 }
721 
722 int
723 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
724 {
725 	struct file *fp;
726 	struct filedesc *fdp;
727 	int error, tmp, locked;
728 
729 	AUDIT_ARG_FD(fd);
730 	AUDIT_ARG_CMD(com);
731 
732 	fdp = td->td_proc->p_fd;
733 
734 	switch (com) {
735 	case FIONCLEX:
736 	case FIOCLEX:
737 		FILEDESC_XLOCK(fdp);
738 		locked = LA_XLOCKED;
739 		break;
740 	default:
741 #ifdef CAPABILITIES
742 		FILEDESC_SLOCK(fdp);
743 		locked = LA_SLOCKED;
744 #else
745 		locked = LA_UNLOCKED;
746 #endif
747 		break;
748 	}
749 
750 #ifdef CAPABILITIES
751 	if ((fp = fget_locked(fdp, fd)) == NULL) {
752 		error = EBADF;
753 		goto out;
754 	}
755 	if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
756 		fp = NULL;	/* fhold() was not called yet */
757 		goto out;
758 	}
759 	if (!fhold(fp)) {
760 		error = EBADF;
761 		fp = NULL;
762 		goto out;
763 	}
764 	if (locked == LA_SLOCKED) {
765 		FILEDESC_SUNLOCK(fdp);
766 		locked = LA_UNLOCKED;
767 	}
768 #else
769 	error = fget(td, fd, &cap_ioctl_rights, &fp);
770 	if (error != 0) {
771 		fp = NULL;
772 		goto out;
773 	}
774 #endif
775 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
776 		error = EBADF;
777 		goto out;
778 	}
779 
780 	switch (com) {
781 	case FIONCLEX:
782 		fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
783 		goto out;
784 	case FIOCLEX:
785 		fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
786 		goto out;
787 	case FIONBIO:
788 		if ((tmp = *(int *)data))
789 			atomic_set_int(&fp->f_flag, FNONBLOCK);
790 		else
791 			atomic_clear_int(&fp->f_flag, FNONBLOCK);
792 		data = (void *)&tmp;
793 		break;
794 	case FIOASYNC:
795 		if ((tmp = *(int *)data))
796 			atomic_set_int(&fp->f_flag, FASYNC);
797 		else
798 			atomic_clear_int(&fp->f_flag, FASYNC);
799 		data = (void *)&tmp;
800 		break;
801 	}
802 
803 	error = fo_ioctl(fp, com, data, td->td_ucred, td);
804 out:
805 	switch (locked) {
806 	case LA_XLOCKED:
807 		FILEDESC_XUNLOCK(fdp);
808 		break;
809 #ifdef CAPABILITIES
810 	case LA_SLOCKED:
811 		FILEDESC_SUNLOCK(fdp);
812 		break;
813 #endif
814 	default:
815 		FILEDESC_UNLOCK_ASSERT(fdp);
816 		break;
817 	}
818 	if (fp != NULL)
819 		fdrop(fp, td);
820 	return (error);
821 }
822 
823 int
824 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap)
825 {
826 	int error;
827 
828 	error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len);
829 	return (kern_posix_error(td, error));
830 }
831 
832 int
833 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len)
834 {
835 	struct file *fp;
836 	int error;
837 
838 	AUDIT_ARG_FD(fd);
839 	if (offset < 0 || len <= 0)
840 		return (EINVAL);
841 	/* Check for wrap. */
842 	if (offset > OFF_MAX - len)
843 		return (EFBIG);
844 	AUDIT_ARG_FD(fd);
845 	error = fget(td, fd, &cap_pwrite_rights, &fp);
846 	if (error != 0)
847 		return (error);
848 	AUDIT_ARG_FILE(td->td_proc, fp);
849 	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
850 		error = ESPIPE;
851 		goto out;
852 	}
853 	if ((fp->f_flag & FWRITE) == 0) {
854 		error = EBADF;
855 		goto out;
856 	}
857 
858 	error = fo_fallocate(fp, offset, len, td);
859  out:
860 	fdrop(fp, td);
861 	return (error);
862 }
863 
864 int
865 kern_specialfd(struct thread *td, int type, void *arg)
866 {
867 	struct file *fp;
868 	struct specialfd_eventfd *ae;
869 	int error, fd, fflags;
870 
871 	fflags = 0;
872 	error = falloc_noinstall(td, &fp);
873 	if (error != 0)
874 		return (error);
875 
876 	switch (type) {
877 	case SPECIALFD_EVENTFD:
878 		ae = arg;
879 		if ((ae->flags & EFD_CLOEXEC) != 0)
880 			fflags |= O_CLOEXEC;
881 		error = eventfd_create_file(td, fp, ae->initval, ae->flags);
882 		break;
883 	default:
884 		error = EINVAL;
885 		break;
886 	}
887 
888 	if (error == 0)
889 		error = finstall(td, fp, &fd, fflags, NULL);
890 	fdrop(fp, td);
891 	if (error == 0)
892 		td->td_retval[0] = fd;
893 	return (error);
894 }
895 
896 int
897 sys___specialfd(struct thread *td, struct __specialfd_args *args)
898 {
899 	struct specialfd_eventfd ae;
900 	int error;
901 
902 	switch (args->type) {
903 	case SPECIALFD_EVENTFD:
904 		if (args->len != sizeof(struct specialfd_eventfd)) {
905 			error = EINVAL;
906 			break;
907 		}
908 		error = copyin(args->req, &ae, sizeof(ae));
909 		if (error != 0)
910 			break;
911 		if ((ae.flags & ~(EFD_CLOEXEC | EFD_NONBLOCK |
912 		    EFD_SEMAPHORE)) != 0) {
913 			error = EINVAL;
914 			break;
915 		}
916 		error = kern_specialfd(td, args->type, &ae);
917 		break;
918 	default:
919 		error = EINVAL;
920 		break;
921 	}
922 	return (error);
923 }
924 
925 int
926 poll_no_poll(int events)
927 {
928 	/*
929 	 * Return true for read/write.  If the user asked for something
930 	 * special, return POLLNVAL, so that clients have a way of
931 	 * determining reliably whether or not the extended
932 	 * functionality is present without hard-coding knowledge
933 	 * of specific filesystem implementations.
934 	 */
935 	if (events & ~POLLSTANDARD)
936 		return (POLLNVAL);
937 
938 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
939 }
940 
941 int
942 sys_pselect(struct thread *td, struct pselect_args *uap)
943 {
944 	struct timespec ts;
945 	struct timeval tv, *tvp;
946 	sigset_t set, *uset;
947 	int error;
948 
949 	if (uap->ts != NULL) {
950 		error = copyin(uap->ts, &ts, sizeof(ts));
951 		if (error != 0)
952 		    return (error);
953 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
954 		tvp = &tv;
955 	} else
956 		tvp = NULL;
957 	if (uap->sm != NULL) {
958 		error = copyin(uap->sm, &set, sizeof(set));
959 		if (error != 0)
960 			return (error);
961 		uset = &set;
962 	} else
963 		uset = NULL;
964 	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
965 	    uset, NFDBITS));
966 }
967 
968 int
969 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
970     struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
971 {
972 	int error;
973 
974 	if (uset != NULL) {
975 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
976 		    &td->td_oldsigmask, 0);
977 		if (error != 0)
978 			return (error);
979 		td->td_pflags |= TDP_OLDMASK;
980 		/*
981 		 * Make sure that ast() is called on return to
982 		 * usermode and TDP_OLDMASK is cleared, restoring old
983 		 * sigmask.
984 		 */
985 		thread_lock(td);
986 		td->td_flags |= TDF_ASTPENDING;
987 		thread_unlock(td);
988 	}
989 	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
990 	return (error);
991 }
992 
993 #ifndef _SYS_SYSPROTO_H_
994 struct select_args {
995 	int	nd;
996 	fd_set	*in, *ou, *ex;
997 	struct	timeval *tv;
998 };
999 #endif
1000 int
1001 sys_select(struct thread *td, struct select_args *uap)
1002 {
1003 	struct timeval tv, *tvp;
1004 	int error;
1005 
1006 	if (uap->tv != NULL) {
1007 		error = copyin(uap->tv, &tv, sizeof(tv));
1008 		if (error)
1009 			return (error);
1010 		tvp = &tv;
1011 	} else
1012 		tvp = NULL;
1013 
1014 	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
1015 	    NFDBITS));
1016 }
1017 
1018 /*
1019  * In the unlikely case when user specified n greater then the last
1020  * open file descriptor, check that no bits are set after the last
1021  * valid fd.  We must return EBADF if any is set.
1022  *
1023  * There are applications that rely on the behaviour.
1024  *
1025  * nd is fd_nfiles.
1026  */
1027 static int
1028 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
1029 {
1030 	char *addr, *oaddr;
1031 	int b, i, res;
1032 	uint8_t bits;
1033 
1034 	if (nd >= ndu || fd_in == NULL)
1035 		return (0);
1036 
1037 	oaddr = NULL;
1038 	bits = 0; /* silence gcc */
1039 	for (i = nd; i < ndu; i++) {
1040 		b = i / NBBY;
1041 #if BYTE_ORDER == LITTLE_ENDIAN
1042 		addr = (char *)fd_in + b;
1043 #else
1044 		addr = (char *)fd_in;
1045 		if (abi_nfdbits == NFDBITS) {
1046 			addr += rounddown(b, sizeof(fd_mask)) +
1047 			    sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
1048 		} else {
1049 			addr += rounddown(b, sizeof(uint32_t)) +
1050 			    sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
1051 		}
1052 #endif
1053 		if (addr != oaddr) {
1054 			res = fubyte(addr);
1055 			if (res == -1)
1056 				return (EFAULT);
1057 			oaddr = addr;
1058 			bits = res;
1059 		}
1060 		if ((bits & (1 << (i % NBBY))) != 0)
1061 			return (EBADF);
1062 	}
1063 	return (0);
1064 }
1065 
1066 int
1067 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
1068     fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
1069 {
1070 	struct filedesc *fdp;
1071 	/*
1072 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
1073 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
1074 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
1075 	 * of 256.
1076 	 */
1077 	fd_mask s_selbits[howmany(2048, NFDBITS)];
1078 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
1079 	struct timeval rtv;
1080 	sbintime_t asbt, precision, rsbt;
1081 	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
1082 	int error, lf, ndu;
1083 
1084 	if (nd < 0)
1085 		return (EINVAL);
1086 	fdp = td->td_proc->p_fd;
1087 	ndu = nd;
1088 	lf = fdp->fd_nfiles;
1089 	if (nd > lf)
1090 		nd = lf;
1091 
1092 	error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1093 	if (error != 0)
1094 		return (error);
1095 	error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1096 	if (error != 0)
1097 		return (error);
1098 	error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1099 	if (error != 0)
1100 		return (error);
1101 
1102 	/*
1103 	 * Allocate just enough bits for the non-null fd_sets.  Use the
1104 	 * preallocated auto buffer if possible.
1105 	 */
1106 	nfdbits = roundup(nd, NFDBITS);
1107 	ncpbytes = nfdbits / NBBY;
1108 	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1109 	nbufbytes = 0;
1110 	if (fd_in != NULL)
1111 		nbufbytes += 2 * ncpbytes;
1112 	if (fd_ou != NULL)
1113 		nbufbytes += 2 * ncpbytes;
1114 	if (fd_ex != NULL)
1115 		nbufbytes += 2 * ncpbytes;
1116 	if (nbufbytes <= sizeof s_selbits)
1117 		selbits = &s_selbits[0];
1118 	else
1119 		selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1120 
1121 	/*
1122 	 * Assign pointers into the bit buffers and fetch the input bits.
1123 	 * Put the output buffers together so that they can be bzeroed
1124 	 * together.
1125 	 */
1126 	sbp = selbits;
1127 #define	getbits(name, x) \
1128 	do {								\
1129 		if (name == NULL) {					\
1130 			ibits[x] = NULL;				\
1131 			obits[x] = NULL;				\
1132 		} else {						\
1133 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
1134 			obits[x] = sbp;					\
1135 			sbp += ncpbytes / sizeof *sbp;			\
1136 			error = copyin(name, ibits[x], ncpubytes);	\
1137 			if (error != 0)					\
1138 				goto done;				\
1139 			if (ncpbytes != ncpubytes)			\
1140 				bzero((char *)ibits[x] + ncpubytes,	\
1141 				    ncpbytes - ncpubytes);		\
1142 		}							\
1143 	} while (0)
1144 	getbits(fd_in, 0);
1145 	getbits(fd_ou, 1);
1146 	getbits(fd_ex, 2);
1147 #undef	getbits
1148 
1149 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1150 	/*
1151 	 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1152 	 * we are running under 32-bit emulation. This should be more
1153 	 * generic.
1154 	 */
1155 #define swizzle_fdset(bits)						\
1156 	if (abi_nfdbits != NFDBITS && bits != NULL) {			\
1157 		int i;							\
1158 		for (i = 0; i < ncpbytes / sizeof *sbp; i++)		\
1159 			bits[i] = (bits[i] >> 32) | (bits[i] << 32);	\
1160 	}
1161 #else
1162 #define swizzle_fdset(bits)
1163 #endif
1164 
1165 	/* Make sure the bit order makes it through an ABI transition */
1166 	swizzle_fdset(ibits[0]);
1167 	swizzle_fdset(ibits[1]);
1168 	swizzle_fdset(ibits[2]);
1169 
1170 	if (nbufbytes != 0)
1171 		bzero(selbits, nbufbytes / 2);
1172 
1173 	precision = 0;
1174 	if (tvp != NULL) {
1175 		rtv = *tvp;
1176 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1177 		    rtv.tv_usec >= 1000000) {
1178 			error = EINVAL;
1179 			goto done;
1180 		}
1181 		if (!timevalisset(&rtv))
1182 			asbt = 0;
1183 		else if (rtv.tv_sec <= INT32_MAX) {
1184 			rsbt = tvtosbt(rtv);
1185 			precision = rsbt;
1186 			precision >>= tc_precexp;
1187 			if (TIMESEL(&asbt, rsbt))
1188 				asbt += tc_tick_sbt;
1189 			if (asbt <= SBT_MAX - rsbt)
1190 				asbt += rsbt;
1191 			else
1192 				asbt = -1;
1193 		} else
1194 			asbt = -1;
1195 	} else
1196 		asbt = -1;
1197 	seltdinit(td);
1198 	/* Iterate until the timeout expires or descriptors become ready. */
1199 	for (;;) {
1200 		error = selscan(td, ibits, obits, nd);
1201 		if (error || td->td_retval[0] != 0)
1202 			break;
1203 		error = seltdwait(td, asbt, precision);
1204 		if (error)
1205 			break;
1206 		error = selrescan(td, ibits, obits);
1207 		if (error || td->td_retval[0] != 0)
1208 			break;
1209 	}
1210 	seltdclear(td);
1211 
1212 done:
1213 	/* select is not restarted after signals... */
1214 	if (error == ERESTART)
1215 		error = EINTR;
1216 	if (error == EWOULDBLOCK)
1217 		error = 0;
1218 
1219 	/* swizzle bit order back, if necessary */
1220 	swizzle_fdset(obits[0]);
1221 	swizzle_fdset(obits[1]);
1222 	swizzle_fdset(obits[2]);
1223 #undef swizzle_fdset
1224 
1225 #define	putbits(name, x) \
1226 	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1227 		error = error2;
1228 	if (error == 0) {
1229 		int error2;
1230 
1231 		putbits(fd_in, 0);
1232 		putbits(fd_ou, 1);
1233 		putbits(fd_ex, 2);
1234 #undef putbits
1235 	}
1236 	if (selbits != &s_selbits[0])
1237 		free(selbits, M_SELECT);
1238 
1239 	return (error);
1240 }
1241 /*
1242  * Convert a select bit set to poll flags.
1243  *
1244  * The backend always returns POLLHUP/POLLERR if appropriate and we
1245  * return this as a set bit in any set.
1246  */
1247 static const int select_flags[3] = {
1248     POLLRDNORM | POLLHUP | POLLERR,
1249     POLLWRNORM | POLLHUP | POLLERR,
1250     POLLRDBAND | POLLERR
1251 };
1252 
1253 /*
1254  * Compute the fo_poll flags required for a fd given by the index and
1255  * bit position in the fd_mask array.
1256  */
1257 static __inline int
1258 selflags(fd_mask **ibits, int idx, fd_mask bit)
1259 {
1260 	int flags;
1261 	int msk;
1262 
1263 	flags = 0;
1264 	for (msk = 0; msk < 3; msk++) {
1265 		if (ibits[msk] == NULL)
1266 			continue;
1267 		if ((ibits[msk][idx] & bit) == 0)
1268 			continue;
1269 		flags |= select_flags[msk];
1270 	}
1271 	return (flags);
1272 }
1273 
1274 /*
1275  * Set the appropriate output bits given a mask of fired events and the
1276  * input bits originally requested.
1277  */
1278 static __inline int
1279 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1280 {
1281 	int msk;
1282 	int n;
1283 
1284 	n = 0;
1285 	for (msk = 0; msk < 3; msk++) {
1286 		if ((events & select_flags[msk]) == 0)
1287 			continue;
1288 		if (ibits[msk] == NULL)
1289 			continue;
1290 		if ((ibits[msk][idx] & bit) == 0)
1291 			continue;
1292 		/*
1293 		 * XXX Check for a duplicate set.  This can occur because a
1294 		 * socket calls selrecord() twice for each poll() call
1295 		 * resulting in two selfds per real fd.  selrescan() will
1296 		 * call selsetbits twice as a result.
1297 		 */
1298 		if ((obits[msk][idx] & bit) != 0)
1299 			continue;
1300 		obits[msk][idx] |= bit;
1301 		n++;
1302 	}
1303 
1304 	return (n);
1305 }
1306 
1307 /*
1308  * Traverse the list of fds attached to this thread's seltd and check for
1309  * completion.
1310  */
1311 static int
1312 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1313 {
1314 	struct filedesc *fdp;
1315 	struct selinfo *si;
1316 	struct seltd *stp;
1317 	struct selfd *sfp;
1318 	struct selfd *sfn;
1319 	struct file *fp;
1320 	fd_mask bit;
1321 	int fd, ev, n, idx;
1322 	int error;
1323 	bool only_user;
1324 
1325 	fdp = td->td_proc->p_fd;
1326 	stp = td->td_sel;
1327 	n = 0;
1328 	only_user = FILEDESC_IS_ONLY_USER(fdp);
1329 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1330 		fd = (int)(uintptr_t)sfp->sf_cookie;
1331 		si = sfp->sf_si;
1332 		selfdfree(stp, sfp);
1333 		/* If the selinfo wasn't cleared the event didn't fire. */
1334 		if (si != NULL)
1335 			continue;
1336 		if (only_user)
1337 			error = fget_only_user(fdp, fd, &cap_event_rights, &fp);
1338 		else
1339 			error = fget_unlocked(fdp, fd, &cap_event_rights, &fp);
1340 		if (__predict_false(error != 0))
1341 			return (error);
1342 		idx = fd / NFDBITS;
1343 		bit = (fd_mask)1 << (fd % NFDBITS);
1344 		ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1345 		if (only_user)
1346 			fput_only_user(fdp, fp);
1347 		else
1348 			fdrop(fp, td);
1349 		if (ev != 0)
1350 			n += selsetbits(ibits, obits, idx, bit, ev);
1351 	}
1352 	stp->st_flags = 0;
1353 	td->td_retval[0] = n;
1354 	return (0);
1355 }
1356 
1357 /*
1358  * Perform the initial filedescriptor scan and register ourselves with
1359  * each selinfo.
1360  */
1361 static int
1362 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd)
1363 {
1364 	struct filedesc *fdp;
1365 	struct file *fp;
1366 	fd_mask bit;
1367 	int ev, flags, end, fd;
1368 	int n, idx;
1369 	int error;
1370 	bool only_user;
1371 
1372 	fdp = td->td_proc->p_fd;
1373 	n = 0;
1374 	only_user = FILEDESC_IS_ONLY_USER(fdp);
1375 	for (idx = 0, fd = 0; fd < nfd; idx++) {
1376 		end = imin(fd + NFDBITS, nfd);
1377 		for (bit = 1; fd < end; bit <<= 1, fd++) {
1378 			/* Compute the list of events we're interested in. */
1379 			flags = selflags(ibits, idx, bit);
1380 			if (flags == 0)
1381 				continue;
1382 			if (only_user)
1383 				error = fget_only_user(fdp, fd, &cap_event_rights, &fp);
1384 			else
1385 				error = fget_unlocked(fdp, fd, &cap_event_rights, &fp);
1386 			if (__predict_false(error != 0))
1387 				return (error);
1388 			selfdalloc(td, (void *)(uintptr_t)fd);
1389 			ev = fo_poll(fp, flags, td->td_ucred, td);
1390 			if (only_user)
1391 				fput_only_user(fdp, fp);
1392 			else
1393 				fdrop(fp, td);
1394 			if (ev != 0)
1395 				n += selsetbits(ibits, obits, idx, bit, ev);
1396 		}
1397 	}
1398 
1399 	td->td_retval[0] = n;
1400 	return (0);
1401 }
1402 
1403 int
1404 sys_poll(struct thread *td, struct poll_args *uap)
1405 {
1406 	struct timespec ts, *tsp;
1407 
1408 	if (uap->timeout != INFTIM) {
1409 		if (uap->timeout < 0)
1410 			return (EINVAL);
1411 		ts.tv_sec = uap->timeout / 1000;
1412 		ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1413 		tsp = &ts;
1414 	} else
1415 		tsp = NULL;
1416 
1417 	return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1418 }
1419 
1420 /*
1421  * kfds points to an array in the kernel.
1422  */
1423 int
1424 kern_poll_kfds(struct thread *td, struct pollfd *kfds, u_int nfds,
1425     struct timespec *tsp, sigset_t *uset)
1426 {
1427 	sbintime_t sbt, precision, tmp;
1428 	time_t over;
1429 	struct timespec ts;
1430 	int error;
1431 
1432 	precision = 0;
1433 	if (tsp != NULL) {
1434 		if (tsp->tv_sec < 0)
1435 			return (EINVAL);
1436 		if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000)
1437 			return (EINVAL);
1438 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1439 			sbt = 0;
1440 		else {
1441 			ts = *tsp;
1442 			if (ts.tv_sec > INT32_MAX / 2) {
1443 				over = ts.tv_sec - INT32_MAX / 2;
1444 				ts.tv_sec -= over;
1445 			} else
1446 				over = 0;
1447 			tmp = tstosbt(ts);
1448 			precision = tmp;
1449 			precision >>= tc_precexp;
1450 			if (TIMESEL(&sbt, tmp))
1451 				sbt += tc_tick_sbt;
1452 			sbt += tmp;
1453 		}
1454 	} else
1455 		sbt = -1;
1456 
1457 	if (uset != NULL) {
1458 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
1459 		    &td->td_oldsigmask, 0);
1460 		if (error)
1461 			return (error);
1462 		td->td_pflags |= TDP_OLDMASK;
1463 		/*
1464 		 * Make sure that ast() is called on return to
1465 		 * usermode and TDP_OLDMASK is cleared, restoring old
1466 		 * sigmask.
1467 		 */
1468 		thread_lock(td);
1469 		td->td_flags |= TDF_ASTPENDING;
1470 		thread_unlock(td);
1471 	}
1472 
1473 	seltdinit(td);
1474 	/* Iterate until the timeout expires or descriptors become ready. */
1475 	for (;;) {
1476 		error = pollscan(td, kfds, nfds);
1477 		if (error || td->td_retval[0] != 0)
1478 			break;
1479 		error = seltdwait(td, sbt, precision);
1480 		if (error)
1481 			break;
1482 		error = pollrescan(td);
1483 		if (error || td->td_retval[0] != 0)
1484 			break;
1485 	}
1486 	seltdclear(td);
1487 
1488 	/* poll is not restarted after signals... */
1489 	if (error == ERESTART)
1490 		error = EINTR;
1491 	if (error == EWOULDBLOCK)
1492 		error = 0;
1493 	return (error);
1494 }
1495 
1496 int
1497 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1498 {
1499 	struct timespec ts, *tsp;
1500 	sigset_t set, *ssp;
1501 	int error;
1502 
1503 	if (uap->ts != NULL) {
1504 		error = copyin(uap->ts, &ts, sizeof(ts));
1505 		if (error)
1506 			return (error);
1507 		tsp = &ts;
1508 	} else
1509 		tsp = NULL;
1510 	if (uap->set != NULL) {
1511 		error = copyin(uap->set, &set, sizeof(set));
1512 		if (error)
1513 			return (error);
1514 		ssp = &set;
1515 	} else
1516 		ssp = NULL;
1517 	return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1518 }
1519 
1520 /*
1521  * ufds points to an array in user space.
1522  */
1523 int
1524 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds,
1525     struct timespec *tsp, sigset_t *set)
1526 {
1527 	struct pollfd *kfds;
1528 	struct pollfd stackfds[32];
1529 	int error;
1530 
1531 	if (kern_poll_maxfds(nfds))
1532 		return (EINVAL);
1533 	if (nfds > nitems(stackfds))
1534 		kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK);
1535 	else
1536 		kfds = stackfds;
1537 	error = copyin(ufds, kfds, nfds * sizeof(*kfds));
1538 	if (error != 0)
1539 		goto out;
1540 
1541 	error = kern_poll_kfds(td, kfds, nfds, tsp, set);
1542 	if (error == 0)
1543 		error = pollout(td, kfds, ufds, nfds);
1544 
1545 out:
1546 	if (nfds > nitems(stackfds))
1547 		free(kfds, M_TEMP);
1548 	return (error);
1549 }
1550 
1551 bool
1552 kern_poll_maxfds(u_int nfds)
1553 {
1554 
1555 	/*
1556 	 * This is kinda bogus.  We have fd limits, but that is not
1557 	 * really related to the size of the pollfd array.  Make sure
1558 	 * we let the process use at least FD_SETSIZE entries and at
1559 	 * least enough for the system-wide limits.  We want to be reasonably
1560 	 * safe, but not overly restrictive.
1561 	 */
1562 	return (nfds > maxfilesperproc && nfds > FD_SETSIZE);
1563 }
1564 
1565 static int
1566 pollrescan(struct thread *td)
1567 {
1568 	struct seltd *stp;
1569 	struct selfd *sfp;
1570 	struct selfd *sfn;
1571 	struct selinfo *si;
1572 	struct filedesc *fdp;
1573 	struct file *fp;
1574 	struct pollfd *fd;
1575 	int n, error;
1576 	bool only_user;
1577 
1578 	n = 0;
1579 	fdp = td->td_proc->p_fd;
1580 	stp = td->td_sel;
1581 	only_user = FILEDESC_IS_ONLY_USER(fdp);
1582 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1583 		fd = (struct pollfd *)sfp->sf_cookie;
1584 		si = sfp->sf_si;
1585 		selfdfree(stp, sfp);
1586 		/* If the selinfo wasn't cleared the event didn't fire. */
1587 		if (si != NULL)
1588 			continue;
1589 		if (only_user)
1590 			error = fget_only_user(fdp, fd->fd, &cap_event_rights, &fp);
1591 		else
1592 			error = fget_unlocked(fdp, fd->fd, &cap_event_rights, &fp);
1593 		if (__predict_false(error != 0)) {
1594 			fd->revents = POLLNVAL;
1595 			n++;
1596 			continue;
1597 		}
1598 		/*
1599 		 * Note: backend also returns POLLHUP and
1600 		 * POLLERR if appropriate.
1601 		 */
1602 		fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1603 		if (only_user)
1604 			fput_only_user(fdp, fp);
1605 		else
1606 			fdrop(fp, td);
1607 		if (fd->revents != 0)
1608 			n++;
1609 	}
1610 	stp->st_flags = 0;
1611 	td->td_retval[0] = n;
1612 	return (0);
1613 }
1614 
1615 static int
1616 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd)
1617 {
1618 	int error = 0;
1619 	u_int i = 0;
1620 	u_int n = 0;
1621 
1622 	for (i = 0; i < nfd; i++) {
1623 		error = copyout(&fds->revents, &ufds->revents,
1624 		    sizeof(ufds->revents));
1625 		if (error)
1626 			return (error);
1627 		if (fds->revents != 0)
1628 			n++;
1629 		fds++;
1630 		ufds++;
1631 	}
1632 	td->td_retval[0] = n;
1633 	return (0);
1634 }
1635 
1636 static int
1637 pollscan(struct thread *td, struct pollfd *fds, u_int nfd)
1638 {
1639 	struct filedesc *fdp;
1640 	struct file *fp;
1641 	int i, n, error;
1642 	bool only_user;
1643 
1644 	n = 0;
1645 	fdp = td->td_proc->p_fd;
1646 	only_user = FILEDESC_IS_ONLY_USER(fdp);
1647 	for (i = 0; i < nfd; i++, fds++) {
1648 		if (fds->fd < 0) {
1649 			fds->revents = 0;
1650 			continue;
1651 		}
1652 		if (only_user)
1653 			error = fget_only_user(fdp, fds->fd, &cap_event_rights, &fp);
1654 		else
1655 			error = fget_unlocked(fdp, fds->fd, &cap_event_rights, &fp);
1656 		if (__predict_false(error != 0)) {
1657 			fds->revents = POLLNVAL;
1658 			n++;
1659 			continue;
1660 		}
1661 		/*
1662 		 * Note: backend also returns POLLHUP and
1663 		 * POLLERR if appropriate.
1664 		 */
1665 		selfdalloc(td, fds);
1666 		fds->revents = fo_poll(fp, fds->events,
1667 		    td->td_ucred, td);
1668 		if (only_user)
1669 			fput_only_user(fdp, fp);
1670 		else
1671 			fdrop(fp, td);
1672 		/*
1673 		 * POSIX requires POLLOUT to be never
1674 		 * set simultaneously with POLLHUP.
1675 		 */
1676 		if ((fds->revents & POLLHUP) != 0)
1677 			fds->revents &= ~POLLOUT;
1678 
1679 		if (fds->revents != 0)
1680 			n++;
1681 	}
1682 	td->td_retval[0] = n;
1683 	return (0);
1684 }
1685 
1686 /*
1687  * XXX This was created specifically to support netncp and netsmb.  This
1688  * allows the caller to specify a socket to wait for events on.  It returns
1689  * 0 if any events matched and an error otherwise.  There is no way to
1690  * determine which events fired.
1691  */
1692 int
1693 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1694 {
1695 	struct timeval rtv;
1696 	sbintime_t asbt, precision, rsbt;
1697 	int error;
1698 
1699 	precision = 0;	/* stupid gcc! */
1700 	if (tvp != NULL) {
1701 		rtv = *tvp;
1702 		if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1703 		    rtv.tv_usec >= 1000000)
1704 			return (EINVAL);
1705 		if (!timevalisset(&rtv))
1706 			asbt = 0;
1707 		else if (rtv.tv_sec <= INT32_MAX) {
1708 			rsbt = tvtosbt(rtv);
1709 			precision = rsbt;
1710 			precision >>= tc_precexp;
1711 			if (TIMESEL(&asbt, rsbt))
1712 				asbt += tc_tick_sbt;
1713 			if (asbt <= SBT_MAX - rsbt)
1714 				asbt += rsbt;
1715 			else
1716 				asbt = -1;
1717 		} else
1718 			asbt = -1;
1719 	} else
1720 		asbt = -1;
1721 	seltdinit(td);
1722 	/*
1723 	 * Iterate until the timeout expires or the socket becomes ready.
1724 	 */
1725 	for (;;) {
1726 		selfdalloc(td, NULL);
1727 		error = sopoll(so, events, NULL, td);
1728 		/* error here is actually the ready events. */
1729 		if (error)
1730 			return (0);
1731 		error = seltdwait(td, asbt, precision);
1732 		if (error)
1733 			break;
1734 	}
1735 	seltdclear(td);
1736 	/* XXX Duplicates ncp/smb behavior. */
1737 	if (error == ERESTART)
1738 		error = 0;
1739 	return (error);
1740 }
1741 
1742 /*
1743  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
1744  * have two select sets, one for read and another for write.
1745  */
1746 static void
1747 selfdalloc(struct thread *td, void *cookie)
1748 {
1749 	struct seltd *stp;
1750 
1751 	stp = td->td_sel;
1752 	if (stp->st_free1 == NULL)
1753 		stp->st_free1 = malloc(sizeof(*stp->st_free1), M_SELFD, M_WAITOK|M_ZERO);
1754 	stp->st_free1->sf_td = stp;
1755 	stp->st_free1->sf_cookie = cookie;
1756 	if (stp->st_free2 == NULL)
1757 		stp->st_free2 = malloc(sizeof(*stp->st_free2), M_SELFD, M_WAITOK|M_ZERO);
1758 	stp->st_free2->sf_td = stp;
1759 	stp->st_free2->sf_cookie = cookie;
1760 }
1761 
1762 static void
1763 selfdfree(struct seltd *stp, struct selfd *sfp)
1764 {
1765 	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1766 	/*
1767 	 * Paired with doselwakeup.
1768 	 */
1769 	if (atomic_load_acq_ptr((uintptr_t *)&sfp->sf_si) != (uintptr_t)NULL) {
1770 		mtx_lock(sfp->sf_mtx);
1771 		if (sfp->sf_si != NULL) {
1772 			TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1773 		}
1774 		mtx_unlock(sfp->sf_mtx);
1775 	}
1776 	free(sfp, M_SELFD);
1777 }
1778 
1779 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1780 void
1781 seldrain(struct selinfo *sip)
1782 {
1783 
1784 	/*
1785 	 * This feature is already provided by doselwakeup(), thus it is
1786 	 * enough to go for it.
1787 	 * Eventually, the context, should take care to avoid races
1788 	 * between thread calling select()/poll() and file descriptor
1789 	 * detaching, but, again, the races are just the same as
1790 	 * selwakeup().
1791 	 */
1792         doselwakeup(sip, -1);
1793 }
1794 
1795 /*
1796  * Record a select request.
1797  */
1798 void
1799 selrecord(struct thread *selector, struct selinfo *sip)
1800 {
1801 	struct selfd *sfp;
1802 	struct seltd *stp;
1803 	struct mtx *mtxp;
1804 
1805 	stp = selector->td_sel;
1806 	/*
1807 	 * Don't record when doing a rescan.
1808 	 */
1809 	if (stp->st_flags & SELTD_RESCAN)
1810 		return;
1811 	/*
1812 	 * Grab one of the preallocated descriptors.
1813 	 */
1814 	sfp = NULL;
1815 	if ((sfp = stp->st_free1) != NULL)
1816 		stp->st_free1 = NULL;
1817 	else if ((sfp = stp->st_free2) != NULL)
1818 		stp->st_free2 = NULL;
1819 	else
1820 		panic("selrecord: No free selfd on selq");
1821 	mtxp = sip->si_mtx;
1822 	if (mtxp == NULL)
1823 		mtxp = mtx_pool_find(mtxpool_select, sip);
1824 	/*
1825 	 * Initialize the sfp and queue it in the thread.
1826 	 */
1827 	sfp->sf_si = sip;
1828 	sfp->sf_mtx = mtxp;
1829 	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1830 	/*
1831 	 * Now that we've locked the sip, check for initialization.
1832 	 */
1833 	mtx_lock(mtxp);
1834 	if (sip->si_mtx == NULL) {
1835 		sip->si_mtx = mtxp;
1836 		TAILQ_INIT(&sip->si_tdlist);
1837 	}
1838 	/*
1839 	 * Add this thread to the list of selfds listening on this selinfo.
1840 	 */
1841 	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1842 	mtx_unlock(sip->si_mtx);
1843 }
1844 
1845 /* Wake up a selecting thread. */
1846 void
1847 selwakeup(struct selinfo *sip)
1848 {
1849 	doselwakeup(sip, -1);
1850 }
1851 
1852 /* Wake up a selecting thread, and set its priority. */
1853 void
1854 selwakeuppri(struct selinfo *sip, int pri)
1855 {
1856 	doselwakeup(sip, pri);
1857 }
1858 
1859 /*
1860  * Do a wakeup when a selectable event occurs.
1861  */
1862 static void
1863 doselwakeup(struct selinfo *sip, int pri)
1864 {
1865 	struct selfd *sfp;
1866 	struct selfd *sfn;
1867 	struct seltd *stp;
1868 
1869 	/* If it's not initialized there can't be any waiters. */
1870 	if (sip->si_mtx == NULL)
1871 		return;
1872 	/*
1873 	 * Locking the selinfo locks all selfds associated with it.
1874 	 */
1875 	mtx_lock(sip->si_mtx);
1876 	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1877 		/*
1878 		 * Once we remove this sfp from the list and clear the
1879 		 * sf_si seltdclear will know to ignore this si.
1880 		 */
1881 		TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1882 		stp = sfp->sf_td;
1883 		mtx_lock(&stp->st_mtx);
1884 		stp->st_flags |= SELTD_PENDING;
1885 		cv_broadcastpri(&stp->st_wait, pri);
1886 		mtx_unlock(&stp->st_mtx);
1887 		/*
1888 		 * Paired with selfdfree.
1889 		 *
1890 		 * Storing this only after the wakeup provides an invariant that
1891 		 * stp is not used after selfdfree returns.
1892 		 */
1893 		atomic_store_rel_ptr((uintptr_t *)&sfp->sf_si, (uintptr_t)NULL);
1894 	}
1895 	mtx_unlock(sip->si_mtx);
1896 }
1897 
1898 static void
1899 seltdinit(struct thread *td)
1900 {
1901 	struct seltd *stp;
1902 
1903 	stp = td->td_sel;
1904 	if (stp != NULL) {
1905 		MPASS(stp->st_flags == 0);
1906 		MPASS(STAILQ_EMPTY(&stp->st_selq));
1907 		return;
1908 	}
1909 	stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1910 	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1911 	cv_init(&stp->st_wait, "select");
1912 	stp->st_flags = 0;
1913 	STAILQ_INIT(&stp->st_selq);
1914 	td->td_sel = stp;
1915 }
1916 
1917 static int
1918 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1919 {
1920 	struct seltd *stp;
1921 	int error;
1922 
1923 	stp = td->td_sel;
1924 	/*
1925 	 * An event of interest may occur while we do not hold the seltd
1926 	 * locked so check the pending flag before we sleep.
1927 	 */
1928 	mtx_lock(&stp->st_mtx);
1929 	/*
1930 	 * Any further calls to selrecord will be a rescan.
1931 	 */
1932 	stp->st_flags |= SELTD_RESCAN;
1933 	if (stp->st_flags & SELTD_PENDING) {
1934 		mtx_unlock(&stp->st_mtx);
1935 		return (0);
1936 	}
1937 	if (sbt == 0)
1938 		error = EWOULDBLOCK;
1939 	else if (sbt != -1)
1940 		error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
1941 		    sbt, precision, C_ABSOLUTE);
1942 	else
1943 		error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1944 	mtx_unlock(&stp->st_mtx);
1945 
1946 	return (error);
1947 }
1948 
1949 void
1950 seltdfini(struct thread *td)
1951 {
1952 	struct seltd *stp;
1953 
1954 	stp = td->td_sel;
1955 	if (stp == NULL)
1956 		return;
1957 	MPASS(stp->st_flags == 0);
1958 	MPASS(STAILQ_EMPTY(&stp->st_selq));
1959 	if (stp->st_free1)
1960 		free(stp->st_free1, M_SELFD);
1961 	if (stp->st_free2)
1962 		free(stp->st_free2, M_SELFD);
1963 	td->td_sel = NULL;
1964 	cv_destroy(&stp->st_wait);
1965 	mtx_destroy(&stp->st_mtx);
1966 	free(stp, M_SELECT);
1967 }
1968 
1969 /*
1970  * Remove the references to the thread from all of the objects we were
1971  * polling.
1972  */
1973 static void
1974 seltdclear(struct thread *td)
1975 {
1976 	struct seltd *stp;
1977 	struct selfd *sfp;
1978 	struct selfd *sfn;
1979 
1980 	stp = td->td_sel;
1981 	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1982 		selfdfree(stp, sfp);
1983 	stp->st_flags = 0;
1984 }
1985 
1986 static void selectinit(void *);
1987 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1988 static void
1989 selectinit(void *dummy __unused)
1990 {
1991 
1992 	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1993 }
1994 
1995 /*
1996  * Set up a syscall return value that follows the convention specified for
1997  * posix_* functions.
1998  */
1999 int
2000 kern_posix_error(struct thread *td, int error)
2001 {
2002 
2003 	if (error <= 0)
2004 		return (error);
2005 	td->td_errno = error;
2006 	td->td_pflags |= TDP_NERRNO;
2007 	td->td_retval[0] = error;
2008 	return (0);
2009 }
2010