Context Navigation

source: rtems-libbsd/freebsd/sys/kern/sys_generic.c @ 3c967ca

55-freebsd-126-freebsd-12

Last change on this file since 3c967ca was 3c967ca, checked in by Sebastian Huber <sebastian.huber@…>, on 06/08/17 at 11:15:12
Use <sys/lock.h> provided by Newlib
Property mode set to `100644`
File size: 45.4 KB

Line
1	#include <machine/rtems-bsd-kernel-space.h>
2
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 <rtems/bsd/local/opt_capsicum.h>
43	#include <rtems/bsd/local/opt_compat.h>
44	#include <rtems/bsd/local/opt_ktrace.h>
45
46	#include <sys/param.h>
47	#include <sys/systm.h>
48	#include <sys/sysproto.h>
49	#include <sys/capsicum.h>
50	#include <sys/filedesc.h>
51	#include <sys/filio.h>
52	#include <sys/fcntl.h>
53	#include <sys/file.h>
54	#include <sys/lock.h>
55	#include <sys/proc.h>
56	#include <sys/signalvar.h>
57	#include <sys/socketvar.h>
58	#include <sys/uio.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/syscallsubr.h>
68	#include <sys/sysctl.h>
69	#include <sys/sysent.h>
70	#include <sys/vnode.h>
71	#include <sys/bio.h>
72	#include <sys/buf.h>
73	#include <sys/condvar.h>
74	#ifdef KTRACE
75	#include <sys/ktrace.h>
76	#endif
77
78	#include <security/audit/audit.h>
79	#ifdef __rtems__
80	#include <machine/rtems-bsd-syscall-api.h>
81	#endif /* __rtems__ */
82
83	/*
84	* The following macro defines how many bytes will be allocated from
85	* the stack instead of memory allocated when passing the IOCTL data
86	* structures from userspace and to the kernel. Some IOCTLs having
87	* small data structures are used very frequently and this small
88	* buffer on the stack gives a significant speedup improvement for
89	* those requests. The value of this define should be greater or equal
90	* to 64 bytes and should also be power of two. The data structure is
91	* currently hard-aligned to a 8-byte boundary on the stack. This
92	* should currently be sufficient for all supported platforms.
93	*/
94	#define SYS_IOCTL_SMALL_SIZE 128 /* bytes */
95	#define SYS_IOCTL_SMALL_ALIGN 8 /* bytes */
96
97	#ifndef __rtems__
98	#ifdef __LP64__
99	static int iosize_max_clamp = 0;
100	SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
101	&iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
102	static int devfs_iosize_max_clamp = 1;
103	SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
104	&devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
105	#endif
106
107	/*
108	* Assert that the return value of read(2) and write(2) syscalls fits
109	* into a register. If not, an architecture will need to provide the
110	* usermode wrappers to reconstruct the result.
111	*/
112	CTASSERT(sizeof(register_t) >= sizeof(size_t));
113
114	static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
115	#endif /* __rtems__ */
116	static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
117	#ifndef __rtems__
118	MALLOC_DEFINE(M_IOV, "iov", "large iov's");
119	#endif /* __rtems__ */
120
121	static int pollout(struct thread , struct pollfd , struct pollfd *,
122	u_int);
123	static int pollscan(struct thread , struct pollfd , u_int);
124	static int pollrescan(struct thread *);
125	static int selscan(struct thread , fd_mask , fd_mask *, int);
126	static int selrescan(struct thread , fd_mask , fd_mask *);
127	static void selfdalloc(struct thread , void );
128	static void selfdfree(struct seltd , struct selfd );
129	#ifndef __rtems__
130	static int dofileread(struct thread , int, struct file , struct uio *,
131	off_t, int);
132	static int dofilewrite(struct thread , int, struct file , struct uio *,
133	off_t, int);
134	#endif /* __rtems__ */
135	static void doselwakeup(struct selinfo *, int);
136	static void seltdinit(struct thread *);
137	static int seltdwait(struct thread *, sbintime_t, sbintime_t);
138	static void seltdclear(struct thread *);
139
140	/*
141	* One seltd per-thread allocated on demand as needed.
142	*
143	* t - protected by st_mtx
144	* k - Only accessed by curthread or read-only
145	*/
146	struct seltd {
147	STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
148	struct selfd st_free1; / (k) free fd for read set. */
149	struct selfd st_free2; / (k) free fd for write set. */
150	struct mtx st_mtx; /* Protects struct seltd */
151	struct cv st_wait; /* (t) Wait channel. */
152	int st_flags; /* (t) SELTD_ flags. */
153	};
154
155	#define SELTD_PENDING 0x0001 /* We have pending events. */
156	#define SELTD_RESCAN 0x0002 /* Doing a rescan. */
157
158	/*
159	* One selfd allocated per-thread per-file-descriptor.
160	* f - protected by sf_mtx
161	*/
162	struct selfd {
163	STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
164	TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
165	struct selinfo sf_si; / (f) selinfo when linked. */
166	struct mtx sf_mtx; / Pointer to selinfo mtx. */
167	struct seltd sf_td; / (k) owning seltd. */
168	void sf_cookie; / (k) fd or pollfd. */
169	u_int sf_refs;
170	};
171
172	static uma_zone_t selfd_zone;
173	static struct mtx_pool *mtxpool_select;
174
175	#ifndef __rtems__
176	#ifdef __LP64__
177	size_t
178	devfs_iosize_max(void)
179	{
180
181	return (devfs_iosize_max_clamp \|\| SV_CURPROC_FLAG(SV_ILP32) ?
182	INT_MAX : SSIZE_MAX);
183	}
184
185	size_t
186	iosize_max(void)
187	{
188
189	return (iosize_max_clamp \|\| SV_CURPROC_FLAG(SV_ILP32) ?
190	INT_MAX : SSIZE_MAX);
191	}
192	#endif
193
194	#ifndef _SYS_SYSPROTO_H_
195	struct read_args {
196	int fd;
197	void *buf;
198	size_t nbyte;
199	};
200	#endif
201	int
202	sys_read(td, uap)
203	struct thread *td;
204	struct read_args *uap;
205	{
206	struct uio auio;
207	struct iovec aiov;
208	int error;
209
210	if (uap->nbyte > IOSIZE_MAX)
211	return (EINVAL);
212	aiov.iov_base = uap->buf;
213	aiov.iov_len = uap->nbyte;
214	auio.uio_iov = &aiov;
215	auio.uio_iovcnt = 1;
216	auio.uio_resid = uap->nbyte;
217	auio.uio_segflg = UIO_USERSPACE;
218	error = kern_readv(td, uap->fd, &auio);
219	return(error);
220	}
221
222	/*
223	* Positioned read system call
224	*/
225	#ifndef _SYS_SYSPROTO_H_
226	struct pread_args {
227	int fd;
228	void *buf;
229	size_t nbyte;
230	int pad;
231	off_t offset;
232	};
233	#endif
234	int
235	sys_pread(struct thread td, struct pread_args uap)
236	{
237
238	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
239	}
240
241	int
242	kern_pread(struct thread td, int fd, void buf, size_t nbyte, off_t offset)
243	{
244	struct uio auio;
245	struct iovec aiov;
246	int error;
247
248	if (nbyte > IOSIZE_MAX)
249	return (EINVAL);
250	aiov.iov_base = buf;
251	aiov.iov_len = nbyte;
252	auio.uio_iov = &aiov;
253	auio.uio_iovcnt = 1;
254	auio.uio_resid = nbyte;
255	auio.uio_segflg = UIO_USERSPACE;
256	error = kern_preadv(td, fd, &auio, offset);
257	return (error);
258	}
259
260	#if defined(COMPAT_FREEBSD6)
261	int
262	freebsd6_pread(struct thread td, struct freebsd6_pread_args uap)
263	{
264
265	return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
266	}
267	#endif
268
269	/*
270	* Scatter read system call.
271	*/
272	#ifndef _SYS_SYSPROTO_H_
273	struct readv_args {
274	int fd;
275	struct iovec *iovp;
276	u_int iovcnt;
277	};
278	#endif
279	int
280	sys_readv(struct thread td, struct readv_args uap)
281	{
282	struct uio *auio;
283	int error;
284
285	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
286	if (error)
287	return (error);
288	error = kern_readv(td, uap->fd, auio);
289	free(auio, M_IOV);
290	return (error);
291	}
292
293	int
294	kern_readv(struct thread td, int fd, struct uio auio)
295	{
296	struct file *fp;
297	cap_rights_t rights;
298	int error;
299
300	error = fget_read(td, fd, cap_rights_init(&rights, CAP_READ), &fp);
301	if (error)
302	return (error);
303	error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
304	fdrop(fp, td);
305	return (error);
306	}
307
308	/*
309	* Scatter positioned read system call.
310	*/
311	#ifndef _SYS_SYSPROTO_H_
312	struct preadv_args {
313	int fd;
314	struct iovec *iovp;
315	u_int iovcnt;
316	off_t offset;
317	};
318	#endif
319	int
320	sys_preadv(struct thread td, struct preadv_args uap)
321	{
322	struct uio *auio;
323	int error;
324
325	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
326	if (error)
327	return (error);
328	error = kern_preadv(td, uap->fd, auio, uap->offset);
329	free(auio, M_IOV);
330	return (error);
331	}
332
333	int
334	kern_preadv(td, fd, auio, offset)
335	struct thread *td;
336	int fd;
337	struct uio *auio;
338	off_t offset;
339	{
340	struct file *fp;
341	cap_rights_t rights;
342	int error;
343
344	error = fget_read(td, fd, cap_rights_init(&rights, CAP_PREAD), &fp);
345	if (error)
346	return (error);
347	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
348	error = ESPIPE;
349	else if (offset < 0 &&
350	(fp->f_vnode == NULL \|\| fp->f_vnode->v_type != VCHR))
351	error = EINVAL;
352	else
353	error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
354	fdrop(fp, td);
355	return (error);
356	}
357
358	/*
359	* Common code for readv and preadv that reads data in
360	* from a file using the passed in uio, offset, and flags.
361	*/
362	static int
363	dofileread(td, fd, fp, auio, offset, flags)
364	struct thread *td;
365	int fd;
366	struct file *fp;
367	struct uio *auio;
368	off_t offset;
369	int flags;
370	{
371	ssize_t cnt;
372	int error;
373	#ifdef KTRACE
374	struct uio *ktruio = NULL;
375	#endif
376
377	AUDIT_ARG_FD(fd);
378
379	/* Finish zero length reads right here */
380	if (auio->uio_resid == 0) {
381	td->td_retval[0] = 0;
382	return(0);
383	}
384	auio->uio_rw = UIO_READ;
385	auio->uio_offset = offset;
386	auio->uio_td = td;
387	#ifdef KTRACE
388	if (KTRPOINT(td, KTR_GENIO))
389	ktruio = cloneuio(auio);
390	#endif
391	cnt = auio->uio_resid;
392	if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
393	if (auio->uio_resid != cnt && (error == ERESTART \|\|
394	error == EINTR \|\| error == EWOULDBLOCK))
395	error = 0;
396	}
397	cnt -= auio->uio_resid;
398	#ifdef KTRACE
399	if (ktruio != NULL) {
400	ktruio->uio_resid = cnt;
401	ktrgenio(fd, UIO_READ, ktruio, error);
402	}
403	#endif
404	td->td_retval[0] = cnt;
405	return (error);
406	}
407
408	#ifndef _SYS_SYSPROTO_H_
409	struct write_args {
410	int fd;
411	const void *buf;
412	size_t nbyte;
413	};
414	#endif
415	int
416	sys_write(td, uap)
417	struct thread *td;
418	struct write_args *uap;
419	{
420	struct uio auio;
421	struct iovec aiov;
422	int error;
423
424	if (uap->nbyte > IOSIZE_MAX)
425	return (EINVAL);
426	aiov.iov_base = (void *)(uintptr_t)uap->buf;
427	aiov.iov_len = uap->nbyte;
428	auio.uio_iov = &aiov;
429	auio.uio_iovcnt = 1;
430	auio.uio_resid = uap->nbyte;
431	auio.uio_segflg = UIO_USERSPACE;
432	error = kern_writev(td, uap->fd, &auio);
433	return(error);
434	}
435
436	/*
437	* Positioned write system call.
438	*/
439	#ifndef _SYS_SYSPROTO_H_
440	struct pwrite_args {
441	int fd;
442	const void *buf;
443	size_t nbyte;
444	int pad;
445	off_t offset;
446	};
447	#endif
448	int
449	sys_pwrite(struct thread td, struct pwrite_args uap)
450	{
451
452	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
453	}
454
455	int
456	kern_pwrite(struct thread td, int fd, const void buf, size_t nbyte,
457	off_t offset)
458	{
459	struct uio auio;
460	struct iovec aiov;
461	int error;
462
463	if (nbyte > IOSIZE_MAX)
464	return (EINVAL);
465	aiov.iov_base = (void *)(uintptr_t)buf;
466	aiov.iov_len = nbyte;
467	auio.uio_iov = &aiov;
468	auio.uio_iovcnt = 1;
469	auio.uio_resid = nbyte;
470	auio.uio_segflg = UIO_USERSPACE;
471	error = kern_pwritev(td, fd, &auio, offset);
472	return(error);
473	}
474
475	#if defined(COMPAT_FREEBSD6)
476	int
477	freebsd6_pwrite(struct thread td, struct freebsd6_pwrite_args uap)
478	{
479
480	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
481	}
482	#endif
483
484	/*
485	* Gather write system call.
486	*/
487	#ifndef _SYS_SYSPROTO_H_
488	struct writev_args {
489	int fd;
490	struct iovec *iovp;
491	u_int iovcnt;
492	};
493	#endif
494	int
495	sys_writev(struct thread td, struct writev_args uap)
496	{
497	struct uio *auio;
498	int error;
499
500	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
501	if (error)
502	return (error);
503	error = kern_writev(td, uap->fd, auio);
504	free(auio, M_IOV);
505	return (error);
506	}
507
508	int
509	kern_writev(struct thread td, int fd, struct uio auio)
510	{
511	struct file *fp;
512	cap_rights_t rights;
513	int error;
514
515	error = fget_write(td, fd, cap_rights_init(&rights, CAP_WRITE), &fp);
516	if (error)
517	return (error);
518	error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
519	fdrop(fp, td);
520	return (error);
521	}
522
523	/*
524	* Gather positioned write system call.
525	*/
526	#ifndef _SYS_SYSPROTO_H_
527	struct pwritev_args {
528	int fd;
529	struct iovec *iovp;
530	u_int iovcnt;
531	off_t offset;
532	};
533	#endif
534	int
535	sys_pwritev(struct thread td, struct pwritev_args uap)
536	{
537	struct uio *auio;
538	int error;
539
540	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
541	if (error)
542	return (error);
543	error = kern_pwritev(td, uap->fd, auio, uap->offset);
544	free(auio, M_IOV);
545	return (error);
546	}
547
548	int
549	kern_pwritev(td, fd, auio, offset)
550	struct thread *td;
551	struct uio *auio;
552	int fd;
553	off_t offset;
554	{
555	struct file *fp;
556	cap_rights_t rights;
557	int error;
558
559	error = fget_write(td, fd, cap_rights_init(&rights, CAP_PWRITE), &fp);
560	if (error)
561	return (error);
562	if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
563	error = ESPIPE;
564	else if (offset < 0 &&
565	(fp->f_vnode == NULL \|\| fp->f_vnode->v_type != VCHR))
566	error = EINVAL;
567	else
568	error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
569	fdrop(fp, td);
570	return (error);
571	}
572
573	/*
574	* Common code for writev and pwritev that writes data to
575	* a file using the passed in uio, offset, and flags.
576	*/
577	static int
578	dofilewrite(td, fd, fp, auio, offset, flags)
579	struct thread *td;
580	int fd;
581	struct file *fp;
582	struct uio *auio;
583	off_t offset;
584	int flags;
585	{
586	ssize_t cnt;
587	int error;
588	#ifdef KTRACE
589	struct uio *ktruio = NULL;
590	#endif
591
592	AUDIT_ARG_FD(fd);
593	auio->uio_rw = UIO_WRITE;
594	auio->uio_td = td;
595	auio->uio_offset = offset;
596	#ifdef KTRACE
597	if (KTRPOINT(td, KTR_GENIO))
598	ktruio = cloneuio(auio);
599	#endif
600	cnt = auio->uio_resid;
601	if (fp->f_type == DTYPE_VNODE &&
602	(fp->f_vnread_flags & FDEVFS_VNODE) == 0)
603	bwillwrite();
604	if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
605	if (auio->uio_resid != cnt && (error == ERESTART \|\|
606	error == EINTR \|\| error == EWOULDBLOCK))
607	error = 0;
608	/* Socket layer is responsible for issuing SIGPIPE. */
609	if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
610	PROC_LOCK(td->td_proc);
611	tdsignal(td, SIGPIPE);
612	PROC_UNLOCK(td->td_proc);
613	}
614	}
615	cnt -= auio->uio_resid;
616	#ifdef KTRACE
617	if (ktruio != NULL) {
618	ktruio->uio_resid = cnt;
619	ktrgenio(fd, UIO_WRITE, ktruio, error);
620	}
621	#endif
622	td->td_retval[0] = cnt;
623	return (error);
624	}
625
626	/*
627	* Truncate a file given a file descriptor.
628	*
629	* Can't use fget_write() here, since must return EINVAL and not EBADF if the
630	* descriptor isn't writable.
631	*/
632	int
633	kern_ftruncate(td, fd, length)
634	struct thread *td;
635	int fd;
636	off_t length;
637	{
638	struct file *fp;
639	cap_rights_t rights;
640	int error;
641
642	AUDIT_ARG_FD(fd);
643	if (length < 0)
644	return (EINVAL);
645	error = fget(td, fd, cap_rights_init(&rights, CAP_FTRUNCATE), &fp);
646	if (error)
647	return (error);
648	AUDIT_ARG_FILE(td->td_proc, fp);
649	if (!(fp->f_flag & FWRITE)) {
650	fdrop(fp, td);
651	return (EINVAL);
652	}
653	error = fo_truncate(fp, length, td->td_ucred, td);
654	fdrop(fp, td);
655	return (error);
656	}
657
658	#ifndef _SYS_SYSPROTO_H_
659	struct ftruncate_args {
660	int fd;
661	int pad;
662	off_t length;
663	};
664	#endif
665	int
666	sys_ftruncate(td, uap)
667	struct thread *td;
668	struct ftruncate_args *uap;
669	{
670
671	return (kern_ftruncate(td, uap->fd, uap->length));
672	}
673
674	#if defined(COMPAT_43)
675	#ifndef _SYS_SYSPROTO_H_
676	struct oftruncate_args {
677	int fd;
678	long length;
679	};
680	#endif
681	int
682	oftruncate(td, uap)
683	struct thread *td;
684	struct oftruncate_args *uap;
685	{
686
687	return (kern_ftruncate(td, uap->fd, uap->length));
688	}
689	#endif /* COMPAT_43 */
690
691	#ifndef _SYS_SYSPROTO_H_
692	struct ioctl_args {
693	int fd;
694	u_long com;
695	caddr_t data;
696	};
697	#endif
698	/* ARGSUSED */
699	int
700	sys_ioctl(struct thread td, struct ioctl_args uap)
701	{
702	u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
703	u_long com;
704	int arg, error;
705	u_int size;
706	caddr_t data;
707
708	if (uap->com > 0xffffffff) {
709	printf(
710	"WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
711	td->td_proc->p_pid, td->td_name, uap->com);
712	uap->com &= 0xffffffff;
713	}
714	com = uap->com;
715
716	/*
717	* Interpret high order word to find amount of data to be
718	* copied to/from the user's address space.
719	*/
720	size = IOCPARM_LEN(com);
721	if ((size > IOCPARM_MAX) \|\|
722	((com & (IOC_VOID \| IOC_IN \| IOC_OUT)) == 0) \|\|
723	#if defined(COMPAT_FREEBSD5) \|\| defined(COMPAT_FREEBSD4) \|\| defined(COMPAT_43)
724	((com & IOC_OUT) && size == 0) \|\|
725	#else
726	((com & (IOC_IN \| IOC_OUT)) && size == 0) \|\|
727	#endif
728	((com & IOC_VOID) && size > 0 && size != sizeof(int)))
729	return (ENOTTY);
730
731	if (size > 0) {
732	if (com & IOC_VOID) {
733	/* Integer argument. */
734	arg = (intptr_t)uap->data;
735	data = (void *)&arg;
736	size = 0;
737	} else {
738	if (size > SYS_IOCTL_SMALL_SIZE)
739	data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
740	else
741	data = smalldata;
742	}
743	} else
744	data = (void *)&uap->data;
745	if (com & IOC_IN) {
746	error = copyin(uap->data, data, (u_int)size);
747	if (error != 0)
748	goto out;
749	} else if (com & IOC_OUT) {
750	/*
751	* Zero the buffer so the user always
752	* gets back something deterministic.
753	*/
754	bzero(data, size);
755	}
756
757	error = kern_ioctl(td, uap->fd, com, data);
758
759	if (error == 0 && (com & IOC_OUT))
760	error = copyout(data, uap->data, (u_int)size);
761
762	out:
763	if (size > SYS_IOCTL_SMALL_SIZE)
764	free(data, M_IOCTLOPS);
765	return (error);
766	}
767
768	int
769	kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
770	{
771	struct file *fp;
772	struct filedesc *fdp;
773	#ifndef CAPABILITIES
774	cap_rights_t rights;
775	#endif
776	int error, tmp, locked;
777
778	AUDIT_ARG_FD(fd);
779	AUDIT_ARG_CMD(com);
780
781	fdp = td->td_proc->p_fd;
782
783	switch (com) {
784	case FIONCLEX:
785	case FIOCLEX:
786	FILEDESC_XLOCK(fdp);
787	locked = LA_XLOCKED;
788	break;
789	default:
790	#ifdef CAPABILITIES
791	FILEDESC_SLOCK(fdp);
792	locked = LA_SLOCKED;
793	#else
794	locked = LA_UNLOCKED;
795	#endif
796	break;
797	}
798
799	#ifdef CAPABILITIES
800	if ((fp = fget_locked(fdp, fd)) == NULL) {
801	error = EBADF;
802	goto out;
803	}
804	if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
805	fp = NULL; /* fhold() was not called yet */
806	goto out;
807	}
808	fhold(fp);
809	if (locked == LA_SLOCKED) {
810	FILEDESC_SUNLOCK(fdp);
811	locked = LA_UNLOCKED;
812	}
813	#else
814	error = fget(td, fd, cap_rights_init(&rights, CAP_IOCTL), &fp);
815	if (error != 0) {
816	fp = NULL;
817	goto out;
818	}
819	#endif
820	if ((fp->f_flag & (FREAD \| FWRITE)) == 0) {
821	error = EBADF;
822	goto out;
823	}
824
825	switch (com) {
826	case FIONCLEX:
827	fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
828	goto out;
829	case FIOCLEX:
830	fdp->fd_ofiles[fd].fde_flags \|= UF_EXCLOSE;
831	goto out;
832	case FIONBIO:
833	if ((tmp = (int )data))
834	atomic_set_int(&fp->f_flag, FNONBLOCK);
835	else
836	atomic_clear_int(&fp->f_flag, FNONBLOCK);
837	data = (void *)&tmp;
838	break;
839	case FIOASYNC:
840	if ((tmp = (int )data))
841	atomic_set_int(&fp->f_flag, FASYNC);
842	else
843	atomic_clear_int(&fp->f_flag, FASYNC);
844	data = (void *)&tmp;
845	break;
846	}
847
848	error = fo_ioctl(fp, com, data, td->td_ucred, td);
849	out:
850	switch (locked) {
851	case LA_XLOCKED:
852	FILEDESC_XUNLOCK(fdp);
853	break;
854	#ifdef CAPABILITIES
855	case LA_SLOCKED:
856	FILEDESC_SUNLOCK(fdp);
857	break;
858	#endif
859	default:
860	FILEDESC_UNLOCK_ASSERT(fdp);
861	break;
862	}
863	if (fp != NULL)
864	fdrop(fp, td);
865	return (error);
866	}
867	#endif /* __rtems__ */
868
869	int
870	poll_no_poll(int events)
871	{
872	/*
873	* Return true for read/write. If the user asked for something
874	* special, return POLLNVAL, so that clients have a way of
875	* determining reliably whether or not the extended
876	* functionality is present without hard-coding knowledge
877	* of specific filesystem implementations.
878	*/
879	if (events & ~POLLSTANDARD)
880	return (POLLNVAL);
881
882	return (events & (POLLIN \| POLLOUT \| POLLRDNORM \| POLLWRNORM));
883	}
884
885	#ifndef __rtems__
886	int
887	sys_pselect(struct thread td, struct pselect_args uap)
888	{
889	struct timespec ts;
890	struct timeval tv, *tvp;
891	sigset_t set, *uset;
892	int error;
893
894	if (uap->ts != NULL) {
895	error = copyin(uap->ts, &ts, sizeof(ts));
896	if (error != 0)
897	return (error);
898	TIMESPEC_TO_TIMEVAL(&tv, &ts);
899	tvp = &tv;
900	} else
901	tvp = NULL;
902	if (uap->sm != NULL) {
903	error = copyin(uap->sm, &set, sizeof(set));
904	if (error != 0)
905	return (error);
906	uset = &set;
907	} else
908	uset = NULL;
909	return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
910	uset, NFDBITS));
911	}
912
913	int
914	kern_pselect(struct thread td, int nd, fd_set in, fd_set ou, fd_set ex,
915	struct timeval tvp, sigset_t uset, int abi_nfdbits)
916	{
917	int error;
918
919	if (uset != NULL) {
920	error = kern_sigprocmask(td, SIG_SETMASK, uset,
921	&td->td_oldsigmask, 0);
922	if (error != 0)
923	return (error);
924	td->td_pflags \|= TDP_OLDMASK;
925	/*
926	* Make sure that ast() is called on return to
927	* usermode and TDP_OLDMASK is cleared, restoring old
928	* sigmask.
929	*/
930	thread_lock(td);
931	td->td_flags \|= TDF_ASTPENDING;
932	thread_unlock(td);
933	}
934	error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
935	return (error);
936	}
937
938	#ifndef _SYS_SYSPROTO_H_
939	struct select_args {
940	int nd;
941	fd_set in, ou, *ex;
942	struct timeval *tv;
943	};
944	#endif
945	int
946	sys_select(struct thread td, struct select_args uap)
947	{
948	struct timeval tv, *tvp;
949	int error;
950
951	if (uap->tv != NULL) {
952	error = copyin(uap->tv, &tv, sizeof(tv));
953	if (error)
954	return (error);
955	tvp = &tv;
956	} else
957	tvp = NULL;
958
959	return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
960	NFDBITS));
961	}
962	#endif /* __rtems__ */
963
964	/*
965	* In the unlikely case when user specified n greater then the last
966	* open file descriptor, check that no bits are set after the last
967	* valid fd. We must return EBADF if any is set.
968	*
969	* There are applications that rely on the behaviour.
970	*
971	* nd is fd_lastfile + 1.
972	*/
973	static int
974	select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
975	{
976	char addr, oaddr;
977	int b, i, res;
978	uint8_t bits;
979
980	if (nd >= ndu \|\| fd_in == NULL)
981	return (0);
982
983	oaddr = NULL;
984	bits = 0; /* silence gcc */
985	for (i = nd; i < ndu; i++) {
986	b = i / NBBY;
987	#if BYTE_ORDER == LITTLE_ENDIAN
988	addr = (char *)fd_in + b;
989	#else
990	addr = (char *)fd_in;
991	if (abi_nfdbits == NFDBITS) {
992	addr += rounddown(b, sizeof(fd_mask)) +
993	sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
994	} else {
995	addr += rounddown(b, sizeof(uint32_t)) +
996	sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
997	}
998	#endif
999	if (addr != oaddr) {
1000	res = fubyte(addr);
1001	if (res == -1)
1002	return (EFAULT);
1003	oaddr = addr;
1004	bits = res;
1005	}
1006	if ((bits & (1 << (i % NBBY))) != 0)
1007	return (EBADF);
1008	}
1009	return (0);
1010	}
1011
1012	int
1013	kern_select(struct thread td, int nd, fd_set fd_in, fd_set *fd_ou,
1014	fd_set fd_ex, struct timeval tvp, int abi_nfdbits)
1015	{
1016	struct filedesc *fdp;
1017	/*
1018	* The magic 2048 here is chosen to be just enough for FD_SETSIZE
1019	* infds with the new FD_SETSIZE of 1024, and more than enough for
1020	* FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
1021	* of 256.
1022	*/
1023	fd_mask s_selbits[howmany(2048, NFDBITS)];
1024	fd_mask ibits[3], obits[3], selbits, sbp;
1025	struct timeval rtv;
1026	sbintime_t asbt, precision, rsbt;
1027	u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
1028	int error, lf, ndu;
1029
1030	if (nd < 0)
1031	return (EINVAL);
1032	#ifndef __rtems__
1033	fdp = td->td_proc->p_fd;
1034	#endif /* __rtems__ */
1035	ndu = nd;
1036	#ifndef __rtems__
1037	lf = fdp->fd_lastfile;
1038	#else /* __rtems__ */
1039	(void) fdp;
1040	lf = rtems_libio_number_iops;
1041	#endif /* __rtems__ */
1042	if (nd > lf + 1)
1043	nd = lf + 1;
1044
1045	error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1046	if (error != 0)
1047	return (error);
1048	error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1049	if (error != 0)
1050	return (error);
1051	error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1052	if (error != 0)
1053	return (error);
1054
1055	/*
1056	* Allocate just enough bits for the non-null fd_sets. Use the
1057	* preallocated auto buffer if possible.
1058	*/
1059	nfdbits = roundup(nd, NFDBITS);
1060	ncpbytes = nfdbits / NBBY;
1061	ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1062	nbufbytes = 0;
1063	if (fd_in != NULL)
1064	nbufbytes += 2 * ncpbytes;
1065	if (fd_ou != NULL)
1066	nbufbytes += 2 * ncpbytes;
1067	if (fd_ex != NULL)
1068	nbufbytes += 2 * ncpbytes;
1069	if (nbufbytes <= sizeof s_selbits)
1070	selbits = &s_selbits[0];
1071	else
1072	selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1073
1074	/*
1075	* Assign pointers into the bit buffers and fetch the input bits.
1076	* Put the output buffers together so that they can be bzeroed
1077	* together.
1078	*/
1079	sbp = selbits;
1080	#define getbits(name, x) \
1081	do { \
1082	if (name == NULL) { \
1083	ibits[x] = NULL; \
1084	obits[x] = NULL; \
1085	} else { \
1086	ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
1087	obits[x] = sbp; \
1088	sbp += ncpbytes / sizeof *sbp; \
1089	error = copyin(name, ibits[x], ncpubytes); \
1090	if (error != 0) \
1091	goto done; \
1092	bzero((char *)ibits[x] + ncpubytes, \
1093	ncpbytes - ncpubytes); \
1094	} \
1095	} while (0)
1096	getbits(fd_in, 0);
1097	getbits(fd_ou, 1);
1098	getbits(fd_ex, 2);
1099	#undef getbits
1100
1101	#if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1102	/*
1103	* XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1104	* we are running under 32-bit emulation. This should be more
1105	* generic.
1106	*/
1107	#define swizzle_fdset(bits) \
1108	if (abi_nfdbits != NFDBITS && bits != NULL) { \
1109	int i; \
1110	for (i = 0; i < ncpbytes / sizeof *sbp; i++) \
1111	bits[i] = (bits[i] >> 32) \| (bits[i] << 32); \
1112	}
1113	#else
1114	#define swizzle_fdset(bits)
1115	#endif
1116
1117	/* Make sure the bit order makes it through an ABI transition */
1118	swizzle_fdset(ibits[0]);
1119	swizzle_fdset(ibits[1]);
1120	swizzle_fdset(ibits[2]);
1121
1122	if (nbufbytes != 0)
1123	bzero(selbits, nbufbytes / 2);
1124
1125	precision = 0;
1126	if (tvp != NULL) {
1127	rtv = *tvp;
1128	if (rtv.tv_sec < 0 \|\| rtv.tv_usec < 0 \|\|
1129	rtv.tv_usec >= 1000000) {
1130	error = EINVAL;
1131	goto done;
1132	}
1133	if (!timevalisset(&rtv))
1134	asbt = 0;
1135	else if (rtv.tv_sec <= INT32_MAX) {
1136	rsbt = tvtosbt(rtv);
1137	precision = rsbt;
1138	precision >>= tc_precexp;
1139	if (TIMESEL(&asbt, rsbt))
1140	asbt += tc_tick_sbt;
1141	if (asbt <= SBT_MAX - rsbt)
1142	asbt += rsbt;
1143	else
1144	asbt = -1;
1145	} else
1146	asbt = -1;
1147	} else
1148	asbt = -1;
1149	seltdinit(td);
1150	/* Iterate until the timeout expires or descriptors become ready. */
1151	for (;;) {
1152	error = selscan(td, ibits, obits, nd);
1153	if (error \|\| td->td_retval[0] != 0)
1154	break;
1155	error = seltdwait(td, asbt, precision);
1156	if (error)
1157	break;
1158	error = selrescan(td, ibits, obits);
1159	if (error \|\| td->td_retval[0] != 0)
1160	break;
1161	}
1162	seltdclear(td);
1163
1164	done:
1165	/* select is not restarted after signals... */
1166	if (error == ERESTART)
1167	error = EINTR;
1168	if (error == EWOULDBLOCK)
1169	error = 0;
1170
1171	/* swizzle bit order back, if necessary */
1172	swizzle_fdset(obits[0]);
1173	swizzle_fdset(obits[1]);
1174	swizzle_fdset(obits[2]);
1175	#undef swizzle_fdset
1176
1177	#define putbits(name, x) \
1178	if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1179	error = error2;
1180	if (error == 0) {
1181	int error2;
1182
1183	putbits(fd_in, 0);
1184	putbits(fd_ou, 1);
1185	putbits(fd_ex, 2);
1186	#undef putbits
1187	}
1188	if (selbits != &s_selbits[0])
1189	free(selbits, M_SELECT);
1190
1191	return (error);
1192	}
1193	#ifdef __rtems__
1194	int
1195	select(int nfds, fd_set readfds, fd_set writefds, fd_set *errorfds,
1196	struct timeval *timeout)
1197	{
1198	struct thread *td = rtems_bsd_get_curthread_or_null();
1199	int error;
1200
1201	if (td != NULL) {
1202	error = kern_select(td, nfds, readfds, writefds, errorfds,
1203	timeout, NFDBITS);
1204	} else {
1205	error = ENOMEM;
1206	}
1207
1208	if (error == 0) {
1209	return td->td_retval[0];
1210	} else {
1211	rtems_set_errno_and_return_minus_one(error);
1212	}
1213	}
1214	#endif /* __rtems__ */
1215
1216	/*
1217	* Convert a select bit set to poll flags.
1218	*
1219	* The backend always returns POLLHUP/POLLERR if appropriate and we
1220	* return this as a set bit in any set.
1221	*/
1222	static int select_flags[3] = {
1223	POLLRDNORM \| POLLHUP \| POLLERR,
1224	POLLWRNORM \| POLLHUP \| POLLERR,
1225	POLLRDBAND \| POLLERR
1226	};
1227
1228	/*
1229	* Compute the fo_poll flags required for a fd given by the index and
1230	* bit position in the fd_mask array.
1231	*/
1232	static __inline int
1233	selflags(fd_mask **ibits, int idx, fd_mask bit)
1234	{
1235	int flags;
1236	int msk;
1237
1238	flags = 0;
1239	for (msk = 0; msk < 3; msk++) {
1240	if (ibits[msk] == NULL)
1241	continue;
1242	if ((ibits[msk][idx] & bit) == 0)
1243	continue;
1244	flags \|= select_flags[msk];
1245	}
1246	return (flags);
1247	}
1248
1249	/*
1250	* Set the appropriate output bits given a mask of fired events and the
1251	* input bits originally requested.
1252	*/
1253	static __inline int
1254	selsetbits(fd_mask ibits, fd_mask obits, int idx, fd_mask bit, int events)
1255	{
1256	int msk;
1257	int n;
1258
1259	n = 0;
1260	for (msk = 0; msk < 3; msk++) {
1261	if ((events & select_flags[msk]) == 0)
1262	continue;
1263	if (ibits[msk] == NULL)
1264	continue;
1265	if ((ibits[msk][idx] & bit) == 0)
1266	continue;
1267	/*
1268	* XXX Check for a duplicate set. This can occur because a
1269	* socket calls selrecord() twice for each poll() call
1270	* resulting in two selfds per real fd. selrescan() will
1271	* call selsetbits twice as a result.
1272	*/
1273	if ((obits[msk][idx] & bit) != 0)
1274	continue;
1275	obits[msk][idx] \|= bit;
1276	n++;
1277	}
1278
1279	return (n);
1280	}
1281
1282	static __inline int
1283	getselfd_cap(struct filedesc fdp, int fd, struct file *fpp)
1284	{
1285	cap_rights_t rights;
1286
1287	cap_rights_init(&rights, CAP_EVENT);
1288
1289	return (fget_unlocked(fdp, fd, &rights, fpp, NULL));
1290	}
1291
1292	/*
1293	* Traverse the list of fds attached to this thread's seltd and check for
1294	* completion.
1295	*/
1296	static int
1297	selrescan(struct thread td, fd_mask ibits, fd_mask *obits)
1298	{
1299	struct filedesc *fdp;
1300	struct selinfo *si;
1301	struct seltd *stp;
1302	struct selfd *sfp;
1303	struct selfd *sfn;
1304	struct file *fp;
1305	fd_mask bit;
1306	int fd, ev, n, idx;
1307	int error;
1308
1309	#ifndef __rtems__
1310	fdp = td->td_proc->p_fd;
1311	#else /* __rtems__ */
1312	fdp = NULL;
1313	#endif /* __rtems__ */
1314	stp = td->td_sel;
1315	n = 0;
1316	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1317	fd = (int)(uintptr_t)sfp->sf_cookie;
1318	si = sfp->sf_si;
1319	selfdfree(stp, sfp);
1320	/* If the selinfo wasn't cleared the event didn't fire. */
1321	if (si != NULL)
1322	continue;
1323	error = getselfd_cap(fdp, fd, &fp);
1324	if (error)
1325	return (error);
1326	idx = fd / NFDBITS;
1327	bit = (fd_mask)1 << (fd % NFDBITS);
1328	ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1329	fdrop(fp, td);
1330	if (ev != 0)
1331	n += selsetbits(ibits, obits, idx, bit, ev);
1332	}
1333	stp->st_flags = 0;
1334	td->td_retval[0] = n;
1335	return (0);
1336	}
1337
1338	/*
1339	* Perform the initial filedescriptor scan and register ourselves with
1340	* each selinfo.
1341	*/
1342	static int
1343	selscan(td, ibits, obits, nfd)
1344	struct thread *td;
1345	fd_mask ibits, obits;
1346	int nfd;
1347	{
1348	struct filedesc *fdp;
1349	struct file *fp;
1350	fd_mask bit;
1351	int ev, flags, end, fd;
1352	int n, idx;
1353	int error;
1354
1355	#ifndef __rtems__
1356	fdp = td->td_proc->p_fd;
1357	#else /* __rtems__ */
1358	fdp = NULL;
1359	#endif /* __rtems__ */
1360	n = 0;
1361	for (idx = 0, fd = 0; fd < nfd; idx++) {
1362	end = imin(fd + NFDBITS, nfd);
1363	for (bit = 1; fd < end; bit <<= 1, fd++) {
1364	/* Compute the list of events we're interested in. */
1365	flags = selflags(ibits, idx, bit);
1366	if (flags == 0)
1367	continue;
1368	error = getselfd_cap(fdp, fd, &fp);
1369	if (error)
1370	return (error);
1371	selfdalloc(td, (void *)(uintptr_t)fd);
1372	ev = fo_poll(fp, flags, td->td_ucred, td);
1373	fdrop(fp, td);
1374	if (ev != 0)
1375	n += selsetbits(ibits, obits, idx, bit, ev);
1376	}
1377	}
1378
1379	td->td_retval[0] = n;
1380	return (0);
1381	}
1382
1383	#ifdef __rtems__
1384	static int kern_poll(struct thread td, struct pollfd fds, u_int nfds,
1385	struct timespec tsp, sigset_t uset);
1386
1387	static
1388	#endif /* __rtems__ */
1389	int
1390	sys_poll(struct thread td, struct poll_args uap)
1391	{
1392	struct timespec ts, *tsp;
1393
1394	if (uap->timeout != INFTIM) {
1395	if (uap->timeout < 0)
1396	return (EINVAL);
1397	ts.tv_sec = uap->timeout / 1000;
1398	ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1399	tsp = &ts;
1400	} else
1401	tsp = NULL;
1402
1403	return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1404	}
1405
1406	int
1407	kern_poll(struct thread td, struct pollfd fds, u_int nfds,
1408	struct timespec tsp, sigset_t uset)
1409	{
1410	struct pollfd *bits;
1411	struct pollfd smallbits[32];
1412	sbintime_t sbt, precision, tmp;
1413	time_t over;
1414	struct timespec ts;
1415	int error;
1416	size_t ni;
1417
1418	precision = 0;
1419	if (tsp != NULL) {
1420	if (tsp->tv_sec < 0)
1421	return (EINVAL);
1422	if (tsp->tv_nsec < 0 \|\| tsp->tv_nsec >= 1000000000)
1423	return (EINVAL);
1424	if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1425	sbt = 0;
1426	else {
1427	ts = *tsp;
1428	if (ts.tv_sec > INT32_MAX / 2) {
1429	over = ts.tv_sec - INT32_MAX / 2;
1430	ts.tv_sec -= over;
1431	} else
1432	over = 0;
1433	tmp = tstosbt(ts);
1434	precision = tmp;
1435	precision >>= tc_precexp;
1436	if (TIMESEL(&sbt, tmp))
1437	sbt += tc_tick_sbt;
1438	sbt += tmp;
1439	}
1440	} else
1441	sbt = -1;
1442
1443	#ifndef __rtems__
1444	if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1445	#else /* __rtems__ */
1446	if (nfds > rtems_libio_number_iops)
1447	#endif /* __rtems__ */
1448	return (EINVAL);
1449	ni = nfds * sizeof(struct pollfd);
1450	if (ni > sizeof(smallbits))
1451	bits = malloc(ni, M_TEMP, M_WAITOK);
1452	else
1453	bits = smallbits;
1454	error = copyin(fds, bits, ni);
1455	if (error)
1456	goto done;
1457
1458	#ifndef __rtems__
1459	if (uset != NULL) {
1460	error = kern_sigprocmask(td, SIG_SETMASK, uset,
1461	&td->td_oldsigmask, 0);
1462	if (error)
1463	goto done;
1464	td->td_pflags \|= TDP_OLDMASK;
1465	/*
1466	* Make sure that ast() is called on return to
1467	* usermode and TDP_OLDMASK is cleared, restoring old
1468	* sigmask.
1469	*/
1470	thread_lock(td);
1471	td->td_flags \|= TDF_ASTPENDING;
1472	thread_unlock(td);
1473	}
1474	#endif /* __rtems__ */
1475
1476	seltdinit(td);
1477	/* Iterate until the timeout expires or descriptors become ready. */
1478	for (;;) {
1479	error = pollscan(td, bits, nfds);
1480	if (error \|\| td->td_retval[0] != 0)
1481	break;
1482	error = seltdwait(td, sbt, precision);
1483	if (error)
1484	break;
1485	error = pollrescan(td);
1486	if (error \|\| td->td_retval[0] != 0)
1487	break;
1488	}
1489	seltdclear(td);
1490
1491	done:
1492	/* poll is not restarted after signals... */
1493	if (error == ERESTART)
1494	error = EINTR;
1495	if (error == EWOULDBLOCK)
1496	error = 0;
1497	if (error == 0) {
1498	error = pollout(td, bits, fds, nfds);
1499	if (error)
1500	goto out;
1501	}
1502	out:
1503	if (ni > sizeof(smallbits))
1504	free(bits, M_TEMP);
1505	return (error);
1506	}
1507	#ifdef __rtems__
1508	int
1509	poll(struct pollfd fds[], nfds_t nfds, int timeout)
1510	{
1511	struct thread *td = rtems_bsd_get_curthread_or_null();
1512	struct poll_args ua = {
1513	.fds = &fds[0],
1514	.nfds = nfds,
1515	.timeout = timeout
1516	};
1517	int error;
1518
1519	if (td != NULL) {
1520	error = sys_poll(td, &ua);
1521	} else {
1522	error = ENOMEM;
1523	}
1524
1525	if (error == 0) {
1526	return td->td_retval[0];
1527	} else {
1528	rtems_set_errno_and_return_minus_one(error);
1529	}
1530	}
1531	#endif /* __rtems__ */
1532
1533	#ifndef __rtems__
1534	int
1535	sys_ppoll(struct thread td, struct ppoll_args uap)
1536	{
1537	struct timespec ts, *tsp;
1538	sigset_t set, *ssp;
1539	int error;
1540
1541	if (uap->ts != NULL) {
1542	error = copyin(uap->ts, &ts, sizeof(ts));
1543	if (error)
1544	return (error);
1545	tsp = &ts;
1546	} else
1547	tsp = NULL;
1548	if (uap->set != NULL) {
1549	error = copyin(uap->set, &set, sizeof(set));
1550	if (error)
1551	return (error);
1552	ssp = &set;
1553	} else
1554	ssp = NULL;
1555	/*
1556	* fds is still a pointer to user space. kern_poll() will
1557	* take care of copyin that array to the kernel space.
1558	*/
1559
1560	return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1561	}
1562	#endif /* __rtems__ */
1563
1564	static int
1565	pollrescan(struct thread *td)
1566	{
1567	struct seltd *stp;
1568	struct selfd *sfp;
1569	struct selfd *sfn;
1570	struct selinfo *si;
1571	struct filedesc *fdp;
1572	struct file *fp;
1573	struct pollfd *fd;
1574	#ifdef CAPABILITIES
1575	cap_rights_t rights;
1576	#endif
1577	int n;
1578
1579	n = 0;
1580	#ifndef __rtems__
1581	fdp = td->td_proc->p_fd;
1582	#else /* __rtems__ */
1583	fdp = NULL;
1584	#endif /* __rtems__ */
1585	stp = td->td_sel;
1586	FILEDESC_SLOCK(fdp);
1587	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1588	fd = (struct pollfd *)sfp->sf_cookie;
1589	si = sfp->sf_si;
1590	selfdfree(stp, sfp);
1591	/* If the selinfo wasn't cleared the event didn't fire. */
1592	if (si != NULL)
1593	continue;
1594	#ifndef __rtems__
1595	fp = fdp->fd_ofiles[fd->fd].fde_file;
1596	#else /* __rtems__ */
1597	fget_unlocked(fdp, fd->fd, NULL, &fp, NULL);
1598	#endif /* __rtems__ */
1599	#ifdef CAPABILITIES
1600	if (fp == NULL \|\|
1601	cap_check(cap_rights(fdp, fd->fd),
1602	cap_rights_init(&rights, CAP_EVENT)) != 0)
1603	#else
1604	if (fp == NULL)
1605	#endif
1606	{
1607	fd->revents = POLLNVAL;
1608	n++;
1609	continue;
1610	}
1611
1612	/*
1613	* Note: backend also returns POLLHUP and
1614	* POLLERR if appropriate.
1615	*/
1616	fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1617	if (fd->revents != 0)
1618	n++;
1619	}
1620	FILEDESC_SUNLOCK(fdp);
1621	stp->st_flags = 0;
1622	td->td_retval[0] = n;
1623	return (0);
1624	}
1625
1626
1627	static int
1628	pollout(td, fds, ufds, nfd)
1629	struct thread *td;
1630	struct pollfd *fds;
1631	struct pollfd *ufds;
1632	u_int nfd;
1633	{
1634	int error = 0;
1635	u_int i = 0;
1636	u_int n = 0;
1637
1638	for (i = 0; i < nfd; i++) {
1639	error = copyout(&fds->revents, &ufds->revents,
1640	sizeof(ufds->revents));
1641	if (error)
1642	return (error);
1643	if (fds->revents != 0)
1644	n++;
1645	fds++;
1646	ufds++;
1647	}
1648	td->td_retval[0] = n;
1649	return (0);
1650	}
1651
1652	static int
1653	pollscan(td, fds, nfd)
1654	struct thread *td;
1655	struct pollfd *fds;
1656	u_int nfd;
1657	{
1658	#ifndef __rtems__
1659	struct filedesc *fdp = td->td_proc->p_fd;
1660	#else /* __rtems__ */
1661	struct filedesc *fdp = NULL;
1662	#endif /* __rtems__ */
1663	struct file *fp;
1664	#ifdef CAPABILITIES
1665	cap_rights_t rights;
1666	#endif
1667	int i, n = 0;
1668
1669	FILEDESC_SLOCK(fdp);
1670	for (i = 0; i < nfd; i++, fds++) {
1671	#ifndef __rtems__
1672	if (fds->fd > fdp->fd_lastfile) {
1673	#else /* __rtems__ */
1674	if (fds->fd >= rtems_libio_number_iops) {
1675	#endif /* __rtems__ */
1676	fds->revents = POLLNVAL;
1677	n++;
1678	} else if (fds->fd < 0) {
1679	fds->revents = 0;
1680	} else {
1681	#ifndef __rtems__
1682	fp = fdp->fd_ofiles[fds->fd].fde_file;
1683	#else /* __rtems__ */
1684	fget_unlocked(fdp, fds->fd, NULL, &fp, NULL);
1685	#endif /* __rtems__ */
1686	#ifdef CAPABILITIES
1687	if (fp == NULL \|\|
1688	cap_check(cap_rights(fdp, fds->fd),
1689	cap_rights_init(&rights, CAP_EVENT)) != 0)
1690	#else
1691	if (fp == NULL)
1692	#endif
1693	{
1694	fds->revents = POLLNVAL;
1695	n++;
1696	} else {
1697	/*
1698	* Note: backend also returns POLLHUP and
1699	* POLLERR if appropriate.
1700	*/
1701	selfdalloc(td, fds);
1702	fds->revents = fo_poll(fp, fds->events,
1703	td->td_ucred, td);
1704	/*
1705	* POSIX requires POLLOUT to be never
1706	* set simultaneously with POLLHUP.
1707	*/
1708	if ((fds->revents & POLLHUP) != 0)
1709	fds->revents &= ~POLLOUT;
1710
1711	if (fds->revents != 0)
1712	n++;
1713	}
1714	}
1715	}
1716	FILEDESC_SUNLOCK(fdp);
1717	td->td_retval[0] = n;
1718	return (0);
1719	}
1720
1721	#ifndef __rtems__
1722	/*
1723	* XXX This was created specifically to support netncp and netsmb. This
1724	* allows the caller to specify a socket to wait for events on. It returns
1725	* 0 if any events matched and an error otherwise. There is no way to
1726	* determine which events fired.
1727	*/
1728	int
1729	selsocket(struct socket so, int events, struct timeval tvp, struct thread *td)
1730	{
1731	struct timeval rtv;
1732	sbintime_t asbt, precision, rsbt;
1733	int error;
1734
1735	precision = 0; /* stupid gcc! */
1736	if (tvp != NULL) {
1737	rtv = *tvp;
1738	if (rtv.tv_sec < 0 \|\| rtv.tv_usec < 0 \|\|
1739	rtv.tv_usec >= 1000000)
1740	return (EINVAL);
1741	if (!timevalisset(&rtv))
1742	asbt = 0;
1743	else if (rtv.tv_sec <= INT32_MAX) {
1744	rsbt = tvtosbt(rtv);
1745	precision = rsbt;
1746	precision >>= tc_precexp;
1747	if (TIMESEL(&asbt, rsbt))
1748	asbt += tc_tick_sbt;
1749	if (asbt <= SBT_MAX - rsbt)
1750	asbt += rsbt;
1751	else
1752	asbt = -1;
1753	} else
1754	asbt = -1;
1755	} else
1756	asbt = -1;
1757	seltdinit(td);
1758	/*
1759	* Iterate until the timeout expires or the socket becomes ready.
1760	*/
1761	for (;;) {
1762	selfdalloc(td, NULL);
1763	error = sopoll(so, events, NULL, td);
1764	/* error here is actually the ready events. */
1765	if (error)
1766	return (0);
1767	error = seltdwait(td, asbt, precision);
1768	if (error)
1769	break;
1770	}
1771	seltdclear(td);
1772	/* XXX Duplicates ncp/smb behavior. */
1773	if (error == ERESTART)
1774	error = 0;
1775	return (error);
1776	}
1777	#endif /* __rtems__ */
1778
1779	/*
1780	* Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1781	* have two select sets, one for read and another for write.
1782	*/
1783	static void
1784	selfdalloc(struct thread td, void cookie)
1785	{
1786	struct seltd *stp;
1787
1788	stp = td->td_sel;
1789	if (stp->st_free1 == NULL)
1790	stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK\|M_ZERO);
1791	stp->st_free1->sf_td = stp;
1792	stp->st_free1->sf_cookie = cookie;
1793	if (stp->st_free2 == NULL)
1794	stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK\|M_ZERO);
1795	stp->st_free2->sf_td = stp;
1796	stp->st_free2->sf_cookie = cookie;
1797	}
1798
1799	static void
1800	selfdfree(struct seltd stp, struct selfd sfp)
1801	{
1802	STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1803	if (sfp->sf_si != NULL) {
1804	mtx_lock(sfp->sf_mtx);
1805	if (sfp->sf_si != NULL) {
1806	TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1807	refcount_release(&sfp->sf_refs);
1808	}
1809	mtx_unlock(sfp->sf_mtx);
1810	}
1811	if (refcount_release(&sfp->sf_refs))
1812	uma_zfree(selfd_zone, sfp);
1813	}
1814
1815	/* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1816	void
1817	seldrain(sip)
1818	struct selinfo *sip;
1819	{
1820
1821	/*
1822	* This feature is already provided by doselwakeup(), thus it is
1823	* enough to go for it.
1824	* Eventually, the context, should take care to avoid races
1825	* between thread calling select()/poll() and file descriptor
1826	* detaching, but, again, the races are just the same as
1827	* selwakeup().
1828	*/
1829	doselwakeup(sip, -1);
1830	}
1831
1832	/*
1833	* Record a select request.
1834	*/
1835	void
1836	selrecord(selector, sip)
1837	struct thread *selector;
1838	struct selinfo *sip;
1839	{
1840	struct selfd *sfp;
1841	struct seltd *stp;
1842	struct mtx *mtxp;
1843
1844	stp = selector->td_sel;
1845	/*
1846	* Don't record when doing a rescan.
1847	*/
1848	if (stp->st_flags & SELTD_RESCAN)
1849	return;
1850	/*
1851	* Grab one of the preallocated descriptors.
1852	*/
1853	sfp = NULL;
1854	if ((sfp = stp->st_free1) != NULL)
1855	stp->st_free1 = NULL;
1856	else if ((sfp = stp->st_free2) != NULL)
1857	stp->st_free2 = NULL;
1858	else
1859	panic("selrecord: No free selfd on selq");
1860	mtxp = sip->si_mtx;
1861	if (mtxp == NULL)
1862	mtxp = mtx_pool_find(mtxpool_select, sip);
1863	/*
1864	* Initialize the sfp and queue it in the thread.
1865	*/
1866	sfp->sf_si = sip;
1867	sfp->sf_mtx = mtxp;
1868	refcount_init(&sfp->sf_refs, 2);
1869	STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1870	/*
1871	* Now that we've locked the sip, check for initialization.
1872	*/
1873	mtx_lock(mtxp);
1874	if (sip->si_mtx == NULL) {
1875	sip->si_mtx = mtxp;
1876	TAILQ_INIT(&sip->si_tdlist);
1877	}
1878	/*
1879	* Add this thread to the list of selfds listening on this selinfo.
1880	*/
1881	TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1882	mtx_unlock(sip->si_mtx);
1883	}
1884
1885	/* Wake up a selecting thread. */
1886	void
1887	selwakeup(sip)
1888	struct selinfo *sip;
1889	{
1890	doselwakeup(sip, -1);
1891	}
1892
1893	/* Wake up a selecting thread, and set its priority. */
1894	void
1895	selwakeuppri(sip, pri)
1896	struct selinfo *sip;
1897	int pri;
1898	{
1899	doselwakeup(sip, pri);
1900	}
1901
1902	/*
1903	* Do a wakeup when a selectable event occurs.
1904	*/
1905	static void
1906	doselwakeup(sip, pri)
1907	struct selinfo *sip;
1908	int pri;
1909	{
1910	struct selfd *sfp;
1911	struct selfd *sfn;
1912	struct seltd *stp;
1913
1914	/* If it's not initialized there can't be any waiters. */
1915	if (sip->si_mtx == NULL)
1916	return;
1917	/*
1918	* Locking the selinfo locks all selfds associated with it.
1919	*/
1920	mtx_lock(sip->si_mtx);
1921	TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1922	/*
1923	* Once we remove this sfp from the list and clear the
1924	* sf_si seltdclear will know to ignore this si.
1925	*/
1926	TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1927	sfp->sf_si = NULL;
1928	stp = sfp->sf_td;
1929	mtx_lock(&stp->st_mtx);
1930	stp->st_flags \|= SELTD_PENDING;
1931	cv_broadcastpri(&stp->st_wait, pri);
1932	mtx_unlock(&stp->st_mtx);
1933	if (refcount_release(&sfp->sf_refs))
1934	uma_zfree(selfd_zone, sfp);
1935	}
1936	mtx_unlock(sip->si_mtx);
1937	}
1938
1939	static void
1940	seltdinit(struct thread *td)
1941	{
1942	struct seltd *stp;
1943
1944	if ((stp = td->td_sel) != NULL)
1945	goto out;
1946	td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK\|M_ZERO);
1947	mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1948	cv_init(&stp->st_wait, "select");
1949	out:
1950	stp->st_flags = 0;
1951	STAILQ_INIT(&stp->st_selq);
1952	}
1953
1954	static int
1955	seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1956	{
1957	struct seltd *stp;
1958	int error;
1959
1960	stp = td->td_sel;
1961	/*
1962	* An event of interest may occur while we do not hold the seltd
1963	* locked so check the pending flag before we sleep.
1964	*/
1965	mtx_lock(&stp->st_mtx);
1966	/*
1967	* Any further calls to selrecord will be a rescan.
1968	*/
1969	stp->st_flags \|= SELTD_RESCAN;
1970	if (stp->st_flags & SELTD_PENDING) {
1971	mtx_unlock(&stp->st_mtx);
1972	return (0);
1973	}
1974	if (sbt == 0)
1975	error = EWOULDBLOCK;
1976	else if (sbt != -1)
1977	error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
1978	sbt, precision, C_ABSOLUTE);
1979	else
1980	error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1981	mtx_unlock(&stp->st_mtx);
1982
1983	return (error);
1984	}
1985
1986	void
1987	seltdfini(struct thread *td)
1988	{
1989	struct seltd *stp;
1990
1991	stp = td->td_sel;
1992	if (stp == NULL)
1993	return;
1994	if (stp->st_free1)
1995	uma_zfree(selfd_zone, stp->st_free1);
1996	if (stp->st_free2)
1997	uma_zfree(selfd_zone, stp->st_free2);
1998	td->td_sel = NULL;
1999	free(stp, M_SELECT);
2000	}
2001
2002	/*
2003	* Remove the references to the thread from all of the objects we were
2004	* polling.
2005	*/
2006	static void
2007	seltdclear(struct thread *td)
2008	{
2009	struct seltd *stp;
2010	struct selfd *sfp;
2011	struct selfd *sfn;
2012
2013	stp = td->td_sel;
2014	STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
2015	selfdfree(stp, sfp);
2016	stp->st_flags = 0;
2017	}
2018
2019	static void selectinit(void *);
2020	SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
2021	static void
2022	selectinit(void *dummy __unused)
2023	{
2024
2025	selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
2026	NULL, NULL, UMA_ALIGN_PTR, 0);
2027	mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
2028	}
2029
2030	#ifndef __rtems__
2031	/*
2032	* Set up a syscall return value that follows the convention specified for
2033	* posix_* functions.
2034	*/
2035	int
2036	kern_posix_error(struct thread *td, int error)
2037	{
2038
2039	if (error <= 0)
2040	return (error);
2041	td->td_errno = error;
2042	td->td_pflags \|= TDP_NERRNO;
2043	td->td_retval[0] = error;
2044	return (0);
2045	}
2046	#endif /* __rtems__ */
2047	#ifdef __rtems__
2048	#include <machine/rtems-bsd-thread.h>
2049
2050	#undef ticks
2051
2052	#include <rtems/score/objectimpl.h>
2053	#include <rtems/score/threadimpl.h>
2054
2055	#include <rtems/bsd/util.h>
2056
2057	static void
2058	force_select_timeout(Thread_Control *thread)
2059	{
2060	struct thread *td = rtems_bsd_get_thread(thread);
2061
2062	if (td != NULL) {
2063	struct seltd *stp = td->td_sel;
2064
2065	cv_broadcastpri(&stp->st_wait, 0);
2066	}
2067	}
2068
2069	rtems_status_code rtems_bsd_force_select_timeout(rtems_id task_id)
2070	{
2071	Thread_Control *thread;
2072	ISR_lock_Context lock_context;
2073
2074	thread = _Thread_Get(task_id, &lock_context);
2075	if (thread == NULL) {
2076	#if defined(RTEMS_MULTIPROCESSING)
2077	if (_Thread_MP_Is_remote(id)) {
2078	return (RTEMS_ILLEGAL_ON_REMOTE_OBJECT);
2079	}
2080	#endif
2081
2082	return (RTEMS_INVALID_ID);
2083	}
2084
2085	_ISR_lock_ISR_enable(&lock_context);
2086	force_select_timeout(thread);
2087	return (RTEMS_SUCCESSFUL);
2088	}
2089	#endif /* __rtems__ */

Note: See TracBrowser for help on using the repository browser.

Download in other formats: