2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94
35 * $FreeBSD: src/sys/kern/sys_generic.c,v 1.55.2.10 2001/03/17 10:39:32 peter Exp $
38 #include "opt_ktrace.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/event.h>
44 #include <sys/filedesc.h>
45 #include <sys/filio.h>
46 #include <sys/fcntl.h>
49 #include <sys/signalvar.h>
50 #include <sys/socketvar.h>
52 #include <sys/kernel.h>
53 #include <sys/kern_syscall.h>
54 #include <sys/malloc.h>
55 #include <sys/mapped_ioctl.h>
57 #include <sys/queue.h>
58 #include <sys/resourcevar.h>
59 #include <sys/socketops.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
64 #include <sys/ktrace.h>
67 #include <vm/vm_page.h>
69 #include <sys/file2.h>
70 #include <sys/spinlock2.h>
72 #include <machine/limits.h>
74 static MALLOC_DEFINE(M_IOCTLOPS
, "ioctlops", "ioctl data buffer");
75 static MALLOC_DEFINE(M_IOCTLMAP
, "ioctlmap", "mapped ioctl handler buffer");
76 static MALLOC_DEFINE(M_SELECT
, "select", "select() buffer");
77 MALLOC_DEFINE(M_IOV
, "iov", "large iov's");
79 typedef struct kfd_set
{
83 enum select_copyin_states
{
84 COPYIN_READ
, COPYIN_WRITE
, COPYIN_EXCEPT
, COPYIN_DONE
};
86 struct select_kevent_copyin_args
{
90 kfd_set
*oread_set
; /* orig set, points into read_set */
91 kfd_set
*owrite_set
; /* orig set, points into write_set */
92 kfd_set
*oexcept_set
; /* orig set, points into except_set */
93 int active_set
; /* One of select_copyin_states */
94 struct lwp
*lwp
; /* Pointer to our lwp */
95 int num_fds
; /* Number of file descriptors (syscall arg) */
96 int proc_fds
; /* Processed fd's (wraps) */
97 int error
; /* Returned to userland */
100 struct poll_kevent_copyin_args
{
108 static struct lwkt_token mioctl_token
= LWKT_TOKEN_INITIALIZER(mioctl_token
);
110 static int doselect(int nd
, fd_set
*in
, fd_set
*ou
, fd_set
*ex
,
111 struct timespec
*ts
, int *res
);
112 static int dopoll(int nfds
, struct pollfd
*fds
, struct timespec
*ts
,
113 int *res
, int flags
);
114 static int dofileread(int, struct file
*, struct uio
*, int, size_t *);
115 static int dofilewrite(int, struct file
*, struct uio
*, int, size_t *);
123 sys_read(struct read_args
*uap
)
125 struct thread
*td
= curthread
;
130 if ((ssize_t
)uap
->nbyte
< 0)
133 aiov
.iov_base
= uap
->buf
;
134 aiov
.iov_len
= uap
->nbyte
;
135 auio
.uio_iov
= &aiov
;
137 auio
.uio_offset
= -1;
138 auio
.uio_resid
= uap
->nbyte
;
139 auio
.uio_rw
= UIO_READ
;
140 auio
.uio_segflg
= UIO_USERSPACE
;
143 error
= kern_preadv(uap
->fd
, &auio
, 0, &uap
->sysmsg_szresult
);
148 * Positioned (Pread) read system call
153 sys_extpread(struct extpread_args
*uap
)
155 struct thread
*td
= curthread
;
161 if ((ssize_t
)uap
->nbyte
< 0)
164 aiov
.iov_base
= uap
->buf
;
165 aiov
.iov_len
= uap
->nbyte
;
166 auio
.uio_iov
= &aiov
;
168 auio
.uio_offset
= uap
->offset
;
169 auio
.uio_resid
= uap
->nbyte
;
170 auio
.uio_rw
= UIO_READ
;
171 auio
.uio_segflg
= UIO_USERSPACE
;
174 flags
= uap
->flags
& O_FMASK
;
175 if (uap
->offset
!= (off_t
)-1)
178 error
= kern_preadv(uap
->fd
, &auio
, flags
, &uap
->sysmsg_szresult
);
183 * Scatter read system call.
188 sys_readv(struct readv_args
*uap
)
190 struct thread
*td
= curthread
;
192 struct iovec aiov
[UIO_SMALLIOV
], *iov
= NULL
;
195 error
= iovec_copyin(uap
->iovp
, &iov
, aiov
, uap
->iovcnt
,
200 auio
.uio_iovcnt
= uap
->iovcnt
;
201 auio
.uio_offset
= -1;
202 auio
.uio_rw
= UIO_READ
;
203 auio
.uio_segflg
= UIO_USERSPACE
;
206 error
= kern_preadv(uap
->fd
, &auio
, 0, &uap
->sysmsg_szresult
);
208 iovec_free(&iov
, aiov
);
214 * Scatter positioned read system call.
219 sys_extpreadv(struct extpreadv_args
*uap
)
221 struct thread
*td
= curthread
;
223 struct iovec aiov
[UIO_SMALLIOV
], *iov
= NULL
;
227 error
= iovec_copyin(uap
->iovp
, &iov
, aiov
, uap
->iovcnt
,
232 auio
.uio_iovcnt
= uap
->iovcnt
;
233 auio
.uio_offset
= uap
->offset
;
234 auio
.uio_rw
= UIO_READ
;
235 auio
.uio_segflg
= UIO_USERSPACE
;
238 flags
= uap
->flags
& O_FMASK
;
239 if (uap
->offset
!= (off_t
)-1)
242 error
= kern_preadv(uap
->fd
, &auio
, flags
, &uap
->sysmsg_szresult
);
244 iovec_free(&iov
, aiov
);
252 kern_preadv(int fd
, struct uio
*auio
, int flags
, size_t *res
)
254 struct thread
*td
= curthread
;
255 struct proc
*p
= td
->td_proc
;
261 fp
= holdfp(p
->p_fd
, fd
, FREAD
);
264 if (flags
& O_FOFFSET
&& fp
->f_type
!= DTYPE_VNODE
) {
267 error
= dofileread(fd
, fp
, auio
, flags
, res
);
274 * Common code for readv and preadv that reads data in
275 * from a file using the passed in uio, offset, and flags.
277 * MPALMOSTSAFE - ktrace needs help
280 dofileread(int fd
, struct file
*fp
, struct uio
*auio
, int flags
, size_t *res
)
285 struct thread
*td
= curthread
;
286 struct iovec
*ktriov
= NULL
;
292 * if tracing, save a copy of iovec
294 if (KTRPOINT(td
, KTR_GENIO
)) {
295 int iovlen
= auio
->uio_iovcnt
* sizeof(struct iovec
);
297 ktriov
= kmalloc(iovlen
, M_TEMP
, M_WAITOK
);
298 bcopy((caddr_t
)auio
->uio_iov
, (caddr_t
)ktriov
, iovlen
);
302 len
= auio
->uio_resid
;
303 error
= fo_read(fp
, auio
, fp
->f_cred
, flags
);
305 if (auio
->uio_resid
!= len
&& (error
== ERESTART
||
306 error
== EINTR
|| error
== EWOULDBLOCK
))
310 if (ktriov
!= NULL
) {
312 ktruio
.uio_iov
= ktriov
;
313 ktruio
.uio_resid
= len
- auio
->uio_resid
;
314 ktrgenio(td
->td_lwp
, fd
, UIO_READ
, &ktruio
, error
);
316 kfree(ktriov
, M_TEMP
);
320 *res
= len
- auio
->uio_resid
;
331 sys_write(struct write_args
*uap
)
333 struct thread
*td
= curthread
;
338 if ((ssize_t
)uap
->nbyte
< 0)
341 aiov
.iov_base
= (void *)(uintptr_t)uap
->buf
;
342 aiov
.iov_len
= uap
->nbyte
;
343 auio
.uio_iov
= &aiov
;
345 auio
.uio_offset
= -1;
346 auio
.uio_resid
= uap
->nbyte
;
347 auio
.uio_rw
= UIO_WRITE
;
348 auio
.uio_segflg
= UIO_USERSPACE
;
351 error
= kern_pwritev(uap
->fd
, &auio
, 0, &uap
->sysmsg_szresult
);
362 sys_extpwrite(struct extpwrite_args
*uap
)
364 struct thread
*td
= curthread
;
370 if ((ssize_t
)uap
->nbyte
< 0)
373 aiov
.iov_base
= (void *)(uintptr_t)uap
->buf
;
374 aiov
.iov_len
= uap
->nbyte
;
375 auio
.uio_iov
= &aiov
;
377 auio
.uio_offset
= uap
->offset
;
378 auio
.uio_resid
= uap
->nbyte
;
379 auio
.uio_rw
= UIO_WRITE
;
380 auio
.uio_segflg
= UIO_USERSPACE
;
383 flags
= uap
->flags
& O_FMASK
;
384 if (uap
->offset
!= (off_t
)-1)
386 error
= kern_pwritev(uap
->fd
, &auio
, flags
, &uap
->sysmsg_szresult
);
394 sys_writev(struct writev_args
*uap
)
396 struct thread
*td
= curthread
;
398 struct iovec aiov
[UIO_SMALLIOV
], *iov
= NULL
;
401 error
= iovec_copyin(uap
->iovp
, &iov
, aiov
, uap
->iovcnt
,
406 auio
.uio_iovcnt
= uap
->iovcnt
;
407 auio
.uio_offset
= -1;
408 auio
.uio_rw
= UIO_WRITE
;
409 auio
.uio_segflg
= UIO_USERSPACE
;
412 error
= kern_pwritev(uap
->fd
, &auio
, 0, &uap
->sysmsg_szresult
);
414 iovec_free(&iov
, aiov
);
420 * Gather positioned write system call
425 sys_extpwritev(struct extpwritev_args
*uap
)
427 struct thread
*td
= curthread
;
429 struct iovec aiov
[UIO_SMALLIOV
], *iov
= NULL
;
433 error
= iovec_copyin(uap
->iovp
, &iov
, aiov
, uap
->iovcnt
,
438 auio
.uio_iovcnt
= uap
->iovcnt
;
439 auio
.uio_offset
= uap
->offset
;
440 auio
.uio_rw
= UIO_WRITE
;
441 auio
.uio_segflg
= UIO_USERSPACE
;
444 flags
= uap
->flags
& O_FMASK
;
445 if (uap
->offset
!= (off_t
)-1)
448 error
= kern_pwritev(uap
->fd
, &auio
, flags
, &uap
->sysmsg_szresult
);
450 iovec_free(&iov
, aiov
);
458 kern_pwritev(int fd
, struct uio
*auio
, int flags
, size_t *res
)
460 struct thread
*td
= curthread
;
461 struct proc
*p
= td
->td_proc
;
467 fp
= holdfp(p
->p_fd
, fd
, FWRITE
);
470 else if ((flags
& O_FOFFSET
) && fp
->f_type
!= DTYPE_VNODE
) {
473 error
= dofilewrite(fd
, fp
, auio
, flags
, res
);
481 * Common code for writev and pwritev that writes data to
482 * a file using the passed in uio, offset, and flags.
484 * MPALMOSTSAFE - ktrace needs help
487 dofilewrite(int fd
, struct file
*fp
, struct uio
*auio
, int flags
, size_t *res
)
489 struct thread
*td
= curthread
;
490 struct lwp
*lp
= td
->td_lwp
;
494 struct iovec
*ktriov
= NULL
;
500 * if tracing, save a copy of iovec and uio
502 if (KTRPOINT(td
, KTR_GENIO
)) {
503 int iovlen
= auio
->uio_iovcnt
* sizeof(struct iovec
);
505 ktriov
= kmalloc(iovlen
, M_TEMP
, M_WAITOK
);
506 bcopy((caddr_t
)auio
->uio_iov
, (caddr_t
)ktriov
, iovlen
);
510 len
= auio
->uio_resid
;
511 error
= fo_write(fp
, auio
, fp
->f_cred
, flags
);
513 if (auio
->uio_resid
!= len
&& (error
== ERESTART
||
514 error
== EINTR
|| error
== EWOULDBLOCK
))
516 /* Socket layer is responsible for issuing SIGPIPE. */
517 if (error
== EPIPE
&& fp
->f_type
!= DTYPE_SOCKET
)
518 lwpsignal(lp
->lwp_proc
, lp
, SIGPIPE
);
521 if (ktriov
!= NULL
) {
523 ktruio
.uio_iov
= ktriov
;
524 ktruio
.uio_resid
= len
- auio
->uio_resid
;
525 ktrgenio(lp
, fd
, UIO_WRITE
, &ktruio
, error
);
527 kfree(ktriov
, M_TEMP
);
531 *res
= len
- auio
->uio_resid
;
542 sys_ioctl(struct ioctl_args
*uap
)
546 error
= mapped_ioctl(uap
->fd
, uap
->com
, uap
->data
, NULL
, &uap
->sysmsg
);
550 struct ioctl_map_entry
{
552 struct ioctl_map_range
*cmd_ranges
;
553 LIST_ENTRY(ioctl_map_entry
) entries
;
557 * The true heart of all ioctl syscall handlers (native, emulation).
558 * If map != NULL, it will be searched for a matching entry for com,
559 * and appropriate conversions/conversion functions will be utilized.
564 mapped_ioctl(int fd
, u_long com
, caddr_t uspc_data
, struct ioctl_map
*map
,
567 struct thread
*td
= curthread
;
568 struct proc
*p
= td
->td_proc
;
571 struct ioctl_map_range
*iomc
= NULL
;
577 #define STK_PARAMS 128
579 char stkbuf
[STK_PARAMS
];
587 fp
= holdfp(p
->p_fd
, fd
, FREAD
|FWRITE
);
591 if (map
!= NULL
) { /* obey translation map */
593 struct ioctl_map_entry
*e
;
595 maskcmd
= com
& map
->mask
;
597 lwkt_gettoken(&mioctl_token
);
598 LIST_FOREACH(e
, &map
->mapping
, entries
) {
599 for (iomc
= e
->cmd_ranges
; iomc
->start
!= 0 ||
600 iomc
->maptocmd
!= 0 || iomc
->wrapfunc
!= NULL
||
601 iomc
->mapfunc
!= NULL
;
603 if (maskcmd
>= iomc
->start
&&
604 maskcmd
<= iomc
->end
)
608 /* Did we find a match? */
609 if (iomc
->start
!= 0 || iomc
->maptocmd
!= 0 ||
610 iomc
->wrapfunc
!= NULL
|| iomc
->mapfunc
!= NULL
)
613 lwkt_reltoken(&mioctl_token
);
616 (iomc
->start
== 0 && iomc
->maptocmd
== 0
617 && iomc
->wrapfunc
== NULL
&& iomc
->mapfunc
== NULL
)) {
618 kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n",
619 map
->sys
, fd
, maskcmd
,
620 (int)((maskcmd
>> 8) & 0xff),
621 (int)(maskcmd
& 0xff));
627 * If it's a non-range one to one mapping, maptocmd should be
628 * correct. If it's a ranged one to one mapping, we pass the
629 * original value of com, and for a range mapped to a different
630 * range, we always need a mapping function to translate the
631 * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff
633 if (iomc
->start
== iomc
->end
&& iomc
->maptocmd
== iomc
->maptoend
) {
634 com
= iomc
->maptocmd
;
635 } else if (iomc
->start
== iomc
->maptocmd
&& iomc
->end
== iomc
->maptoend
) {
636 if (iomc
->mapfunc
!= NULL
)
637 com
= iomc
->mapfunc(iomc
->start
, iomc
->end
,
638 iomc
->start
, iomc
->end
,
641 if (iomc
->mapfunc
!= NULL
) {
642 com
= iomc
->mapfunc(iomc
->start
, iomc
->end
,
643 iomc
->maptocmd
, iomc
->maptoend
,
646 kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n",
647 map
->sys
, fd
, maskcmd
,
648 (int)((maskcmd
>> 8) & 0xff),
649 (int)(maskcmd
& 0xff));
658 error
= fclrfdflags(p
->p_fd
, fd
, UF_EXCLOSE
);
661 error
= fsetfdflags(p
->p_fd
, fd
, UF_EXCLOSE
);
666 * Interpret high order word to find amount of data to be
667 * copied to/from the user's address space.
669 size
= IOCPARM_LEN(com
);
670 if (size
> IOCPARM_MAX
) {
675 if ((com
& IOC_VOID
) == 0 && size
> sizeof(ubuf
.stkbuf
)) {
676 memp
= kmalloc(size
, M_IOCTLOPS
, M_WAITOK
);
682 if (com
& IOC_VOID
) {
683 *(caddr_t
*)data
= uspc_data
;
684 } else if (com
& IOC_IN
) {
686 error
= copyin(uspc_data
, data
, (size_t)size
);
690 *(caddr_t
*)data
= uspc_data
;
692 } else if ((com
& IOC_OUT
) != 0 && size
) {
694 * Zero the buffer so the user always
695 * gets back something deterministic.
697 bzero(data
, (size_t)size
);
702 if ((tmp
= *(int *)data
))
703 atomic_set_int(&fp
->f_flag
, FNONBLOCK
);
705 atomic_clear_int(&fp
->f_flag
, FNONBLOCK
);
710 if ((tmp
= *(int *)data
))
711 atomic_set_int(&fp
->f_flag
, FASYNC
);
713 atomic_clear_int(&fp
->f_flag
, FASYNC
);
714 error
= fo_ioctl(fp
, FIOASYNC
, (caddr_t
)&tmp
, cred
, msg
);
719 * If there is a override function,
720 * call it instead of directly routing the call
722 if (map
!= NULL
&& iomc
->wrapfunc
!= NULL
)
723 error
= iomc
->wrapfunc(fp
, com
, ocom
, data
, cred
);
725 error
= fo_ioctl(fp
, com
, data
, cred
, msg
);
727 * Copy any data to user, size was
728 * already set and checked above.
730 if (error
== 0 && (com
& IOC_OUT
) != 0 && size
!= 0)
731 error
= copyout(data
, uspc_data
, (size_t)size
);
736 kfree(memp
, M_IOCTLOPS
);
745 mapped_ioctl_register_handler(struct ioctl_map_handler
*he
)
747 struct ioctl_map_entry
*ne
;
749 KKASSERT(he
!= NULL
&& he
->map
!= NULL
&& he
->cmd_ranges
!= NULL
&&
750 he
->subsys
!= NULL
&& *he
->subsys
!= '\0');
752 ne
= kmalloc(sizeof(struct ioctl_map_entry
), M_IOCTLMAP
,
755 ne
->subsys
= he
->subsys
;
756 ne
->cmd_ranges
= he
->cmd_ranges
;
758 lwkt_gettoken(&mioctl_token
);
759 LIST_INSERT_HEAD(&he
->map
->mapping
, ne
, entries
);
760 lwkt_reltoken(&mioctl_token
);
769 mapped_ioctl_unregister_handler(struct ioctl_map_handler
*he
)
771 struct ioctl_map_entry
*ne
;
774 KKASSERT(he
!= NULL
&& he
->map
!= NULL
&& he
->cmd_ranges
!= NULL
);
776 lwkt_gettoken(&mioctl_token
);
777 LIST_FOREACH(ne
, &he
->map
->mapping
, entries
) {
778 if (ne
->cmd_ranges
== he
->cmd_ranges
) {
779 LIST_REMOVE(ne
, entries
);
780 kfree(ne
, M_IOCTLMAP
);
785 lwkt_reltoken(&mioctl_token
);
789 static int nselcoll
; /* Select collisions since boot */
791 SYSCTL_INT(_kern
, OID_AUTO
, nselcoll
, CTLFLAG_RD
, &nselcoll
, 0, "");
792 static int nseldebug
;
793 SYSCTL_INT(_kern
, OID_AUTO
, nseldebug
, CTLFLAG_RW
, &nseldebug
, 0, "");
796 * Select system call.
801 sys_select(struct select_args
*uap
)
804 struct timespec
*ktsp
, kts
;
808 * Get timeout if any.
810 if (uap
->tv
!= NULL
) {
811 error
= copyin(uap
->tv
, &ktv
, sizeof (ktv
));
814 TIMEVAL_TO_TIMESPEC(&ktv
, &kts
);
823 error
= doselect(uap
->nd
, uap
->in
, uap
->ou
, uap
->ex
, ktsp
,
824 &uap
->sysmsg_result
);
831 * Pselect system call.
834 sys_pselect(struct pselect_args
*uap
)
836 struct thread
*td
= curthread
;
837 struct lwp
*lp
= td
->td_lwp
;
838 struct timespec
*ktsp
, kts
;
843 * Get timeout if any.
845 if (uap
->ts
!= NULL
) {
846 error
= copyin(uap
->ts
, &kts
, sizeof (kts
));
855 * Install temporary signal mask if any provided.
857 if (uap
->sigmask
!= NULL
) {
858 error
= copyin(uap
->sigmask
, &sigmask
, sizeof(sigmask
));
861 lwkt_gettoken(&lp
->lwp_proc
->p_token
);
862 lp
->lwp_oldsigmask
= lp
->lwp_sigmask
;
863 SIG_CANTMASK(sigmask
);
864 lp
->lwp_sigmask
= sigmask
;
865 lwkt_reltoken(&lp
->lwp_proc
->p_token
);
871 error
= doselect(uap
->nd
, uap
->in
, uap
->ou
, uap
->ex
, ktsp
,
872 &uap
->sysmsg_result
);
874 if (uap
->sigmask
!= NULL
) {
875 lwkt_gettoken(&lp
->lwp_proc
->p_token
);
876 /* doselect() responsible for turning ERESTART into EINTR */
877 KKASSERT(error
!= ERESTART
);
878 if (error
== EINTR
) {
880 * We can't restore the previous signal mask now
881 * because it could block the signal that interrupted
882 * us. So make a note to restore it after executing
885 lp
->lwp_flags
|= LWP_OLDMASK
;
888 * No handler to run. Restore previous mask immediately.
890 lp
->lwp_sigmask
= lp
->lwp_oldsigmask
;
892 lwkt_reltoken(&lp
->lwp_proc
->p_token
);
899 select_copyin(void *arg
, struct kevent
*kevp
, int maxevents
, int *events
)
901 struct select_kevent_copyin_args
*skap
= NULL
;
908 skap
= (struct select_kevent_copyin_args
*)arg
;
910 if (*events
== maxevents
)
913 while (skap
->active_set
< COPYIN_DONE
) {
914 switch (skap
->active_set
) {
917 * Register descriptors for the read filter
919 fdp
= skap
->read_set
;
920 filter
= EVFILT_READ
;
921 fflags
= NOTE_OLDAPI
;
929 * Register descriptors for the write filter
931 fdp
= skap
->write_set
;
932 filter
= EVFILT_WRITE
;
933 fflags
= NOTE_OLDAPI
;
941 * Register descriptors for the exception filter
943 fdp
= skap
->except_set
;
944 filter
= EVFILT_EXCEPT
;
945 fflags
= NOTE_OLDAPI
| NOTE_OOB
;
953 * Nothing left to register
959 while (skap
->proc_fds
< skap
->num_fds
) {
961 if (FD_ISSET(fd
, fdp
)) {
962 kev
= &kevp
[*events
];
963 EV_SET(kev
, fd
, filter
,
967 skap
->lwp
->lwp_kqueue_serial
);
972 kprintf("select fd %d filter %d serial %d\n",
973 fd
, filter
, skap
->lwp
->lwp_kqueue_serial
);
976 if (*events
== maxevents
)
987 select_copyout(void *arg
, struct kevent
*kevp
, int count
, int *res
)
989 struct select_kevent_copyin_args
*skap
;
993 skap
= (struct select_kevent_copyin_args
*)arg
;
995 for (i
= 0; i
< count
; ++i
) {
996 kfd_set
*fd_set
= NULL
;
999 * Filter out and delete spurious events
1001 if ((u_int
)(uintptr_t)kevp
[i
].udata
!=
1002 skap
->lwp
->lwp_kqueue_serial
) {
1004 kprintf("select fd %ju mismatched serial %d\n",
1005 (uintmax_t)kevp
[i
].ident
,
1006 skap
->lwp
->lwp_kqueue_serial
);
1009 switch (kevp
[i
].filter
) {
1011 if (__predict_true(skap
->oread_set
!= NULL
&&
1012 FD_ISSET(kevp
[i
].ident
, skap
->oread_set
)))
1013 fd_set
= skap
->read_set
;
1017 if (__predict_true(skap
->owrite_set
!= NULL
&&
1018 FD_ISSET(kevp
[i
].ident
, skap
->owrite_set
)))
1019 fd_set
= skap
->write_set
;
1023 if (__predict_true(skap
->oexcept_set
!= NULL
&&
1024 FD_ISSET(kevp
[i
].ident
, skap
->oexcept_set
)))
1025 fd_set
= skap
->except_set
;
1028 if (__predict_false(fd_set
== NULL
) && nseldebug
) {
1029 kprintf("select leftover fd %ju, "
1031 (uintmax_t)kevp
[i
].ident
);
1034 if (fd_set
== NULL
) {
1036 kev
.flags
= EV_DISABLE
|EV_DELETE
;
1037 kqueue_register(&skap
->lwp
->lwp_kqueue
, &kev
);
1044 if (kevp
[i
].flags
& EV_ERROR
) {
1045 int error
= kevp
[i
].data
;
1050 * A bad file descriptor is considered a
1051 * fatal error for select, bail out.
1053 skap
->error
= error
;
1059 * Select silently swallows any unknown errors
1060 * for descriptors in the read or write sets.
1062 * ALWAYS filter out EOPNOTSUPP errors from
1063 * filters (at least until all filters support
1066 * We also filter out ENODEV since dev_dkqfilter
1067 * returns ENODEV if EOPNOTSUPP is returned in an
1072 if (kevp
[i
].filter
!= EVFILT_READ
&&
1073 kevp
[i
].filter
!= EVFILT_WRITE
&&
1074 error
!= EOPNOTSUPP
&&
1076 skap
->error
= error
;
1083 kprintf("select fd %ju filter %d error %d\n",
1084 (uintmax_t)kevp
[i
].ident
,
1085 kevp
[i
].filter
, error
);
1088 FD_SET(kevp
[i
].ident
, fd_set
);
1097 * Copy select bits in from userland. Allocate kernel memory if the
1101 getbits(int bytes
, fd_set
*in_set
, kfd_set
**out_set0
, kfd_set
**orig_set0
,
1104 kfd_set
*out_set
= NULL
, *orig_set
= NULL
;
1108 if (bytes
< sizeof(kfd_set
)) {
1110 orig_set
= tmp_set
+ 1;
1112 out_set
= kmalloc(bytes
* 2, M_SELECT
, M_WAITOK
);
1113 orig_set
= (kfd_set
*)(((uint8_t *)out_set
) + bytes
);
1115 error
= copyin(in_set
, out_set
, bytes
);
1117 memcpy(orig_set
, out_set
, bytes
);
1119 *out_set0
= out_set
;
1120 *orig_set0
= orig_set
;
1125 * Copy returned select bits back out to userland.
1128 putbits(int bytes
, kfd_set
*in_set
, fd_set
*out_set
)
1133 error
= copyout(in_set
, out_set
, bytes
);
1141 dotimeout_only(struct timespec
*ts
)
1143 return(nanosleep1(ts
, NULL
));
1147 * Common code for sys_select() and sys_pselect().
1149 * in, out and ex are userland pointers. ts must point to validated
1150 * kernel-side timeout value or NULL for infinite timeout. res must
1151 * point to syscall return value.
1154 doselect(int nd
, fd_set
*read
, fd_set
*write
, fd_set
*except
,
1155 struct timespec
*ts
, int *res
)
1157 struct proc
*p
= curproc
;
1158 struct select_kevent_copyin_args
*kap
, ka
;
1160 kfd_set read_tmp
[2];
1161 kfd_set write_tmp
[2];
1162 kfd_set except_tmp
[2];
1168 return (dotimeout_only(ts
));
1170 if (nd
> p
->p_fd
->fd_nfiles
) /* limit kmalloc */
1171 nd
= p
->p_fd
->fd_nfiles
;
1174 kap
->lwp
= curthread
->td_lwp
;
1178 kap
->active_set
= COPYIN_READ
;
1181 * Calculate bytes based on the number of __fd_mask[] array entries
1182 * multiplied by the size of __fd_mask.
1184 bytes
= howmany(nd
, __NFDBITS
) * sizeof(__fd_mask
);
1186 /* kap->read_set = NULL; not needed */
1187 kap
->write_set
= NULL
;
1188 kap
->except_set
= NULL
;
1190 error
= getbits(bytes
, read
, &kap
->read_set
, &kap
->oread_set
, read_tmp
);
1192 error
= getbits(bytes
, write
, &kap
->write_set
, &kap
->owrite_set
,
1196 error
= getbits(bytes
, except
, &kap
->except_set
,
1197 &kap
->oexcept_set
, except_tmp
);
1203 * NOTE: Make sure the max events passed to kern_kevent() is
1204 * effectively unlimited. (nd * 3) accomplishes this.
1206 * (*res) continues to increment as returned events are
1209 error
= kern_kevent(&kap
->lwp
->lwp_kqueue
, 0x7FFFFFFF, res
, kap
,
1210 select_copyin
, select_copyout
, ts
, 0);
1212 error
= putbits(bytes
, kap
->read_set
, read
);
1214 error
= putbits(bytes
, kap
->write_set
, write
);
1216 error
= putbits(bytes
, kap
->except_set
, except
);
1219 * An error from an individual event that should be passed
1220 * back to userland (EBADF)
1229 if (kap
->read_set
&& kap
->read_set
!= read_tmp
)
1230 kfree(kap
->read_set
, M_SELECT
);
1231 if (kap
->write_set
&& kap
->write_set
!= write_tmp
)
1232 kfree(kap
->write_set
, M_SELECT
);
1233 if (kap
->except_set
&& kap
->except_set
!= except_tmp
)
1234 kfree(kap
->except_set
, M_SELECT
);
1236 kap
->lwp
->lwp_kqueue_serial
+= kap
->num_fds
;
1247 sys_poll(struct poll_args
*uap
)
1249 struct timespec ts
, *tsp
;
1252 if (uap
->timeout
!= INFTIM
) {
1253 if (uap
->timeout
< 0)
1255 ts
.tv_sec
= uap
->timeout
/ 1000;
1256 ts
.tv_nsec
= (uap
->timeout
% 1000) * 1000 * 1000;
1262 error
= dopoll(uap
->nfds
, uap
->fds
, tsp
, &uap
->sysmsg_result
, 0);
1268 * Ppoll system call.
1273 sys_ppoll(struct ppoll_args
*uap
)
1275 struct thread
*td
= curthread
;
1276 struct lwp
*lp
= td
->td_lwp
;
1277 struct timespec
*ktsp
, kts
;
1282 * Get timeout if any.
1284 if (uap
->ts
!= NULL
) {
1285 error
= copyin(uap
->ts
, &kts
, sizeof (kts
));
1294 * Install temporary signal mask if any provided.
1296 if (uap
->sigmask
!= NULL
) {
1297 error
= copyin(uap
->sigmask
, &sigmask
, sizeof(sigmask
));
1300 lwkt_gettoken(&lp
->lwp_proc
->p_token
);
1301 lp
->lwp_oldsigmask
= lp
->lwp_sigmask
;
1302 SIG_CANTMASK(sigmask
);
1303 lp
->lwp_sigmask
= sigmask
;
1304 lwkt_reltoken(&lp
->lwp_proc
->p_token
);
1307 error
= dopoll(uap
->nfds
, uap
->fds
, ktsp
, &uap
->sysmsg_result
,
1308 ktsp
!= NULL
? KEVENT_TIMEOUT_PRECISE
: 0);
1310 if (uap
->sigmask
!= NULL
) {
1311 lwkt_gettoken(&lp
->lwp_proc
->p_token
);
1312 /* dopoll() responsible for turning ERESTART into EINTR */
1313 KKASSERT(error
!= ERESTART
);
1314 if (error
== EINTR
) {
1316 * We can't restore the previous signal mask now
1317 * because it could block the signal that interrupted
1318 * us. So make a note to restore it after executing
1321 lp
->lwp_flags
|= LWP_OLDMASK
;
1324 * No handler to run. Restore previous mask immediately.
1326 lp
->lwp_sigmask
= lp
->lwp_oldsigmask
;
1328 lwkt_reltoken(&lp
->lwp_proc
->p_token
);
1335 poll_copyin(void *arg
, struct kevent
*kevp
, int maxevents
, int *events
)
1337 struct poll_kevent_copyin_args
*pkap
;
1342 pkap
= (struct poll_kevent_copyin_args
*)arg
;
1344 while (pkap
->pfds
< pkap
->nfds
) {
1345 pfd
= &pkap
->fds
[pkap
->pfds
];
1347 /* Clear return events */
1350 /* Do not check if fd is equal to -1 */
1351 if (pfd
->fd
== -1) {
1357 if (pfd
->events
& (POLLIN
| POLLRDNORM
))
1359 if (pfd
->events
& (POLLOUT
| POLLWRNORM
))
1361 if (pfd
->events
& (POLLPRI
| POLLRDBAND
))
1364 if (*events
+ kev_count
> maxevents
)
1368 * NOTE: A combined serial number and poll array index is
1369 * stored in kev->udata.
1371 kev
= &kevp
[*events
];
1372 if (pfd
->events
& (POLLIN
| POLLRDNORM
)) {
1373 EV_SET(kev
++, pfd
->fd
, EVFILT_READ
, EV_ADD
|EV_ENABLE
,
1374 NOTE_OLDAPI
, 0, (void *)(uintptr_t)
1375 (pkap
->lwp
->lwp_kqueue_serial
+ pkap
->pfds
));
1377 if (pfd
->events
& (POLLOUT
| POLLWRNORM
)) {
1378 EV_SET(kev
++, pfd
->fd
, EVFILT_WRITE
, EV_ADD
|EV_ENABLE
,
1379 NOTE_OLDAPI
, 0, (void *)(uintptr_t)
1380 (pkap
->lwp
->lwp_kqueue_serial
+ pkap
->pfds
));
1382 if (pfd
->events
& (POLLPRI
| POLLRDBAND
)) {
1383 EV_SET(kev
++, pfd
->fd
, EVFILT_EXCEPT
, EV_ADD
|EV_ENABLE
,
1384 NOTE_OLDAPI
| NOTE_OOB
, 0,
1386 (pkap
->lwp
->lwp_kqueue_serial
+ pkap
->pfds
));
1390 kprintf("poll index %d/%d fd %d events %08x serial %d\n",
1391 pkap
->pfds
, pkap
->nfds
-1, pfd
->fd
, pfd
->events
,
1392 pkap
->lwp
->lwp_kqueue_serial
);
1396 (*events
) += kev_count
;
1403 poll_copyout(void *arg
, struct kevent
*kevp
, int count
, int *res
)
1405 struct poll_kevent_copyin_args
*pkap
;
1412 pkap
= (struct poll_kevent_copyin_args
*)arg
;
1414 for (i
= 0; i
< count
; ++i
) {
1416 * Extract the poll array index and delete spurious events.
1417 * We can easily tell if the serial number is incorrect
1418 * by checking whether the extracted index is out of range.
1420 pi
= (u_int
)(uintptr_t)kevp
[i
].udata
-
1421 (u_int
)pkap
->lwp
->lwp_kqueue_serial
;
1423 if (pi
>= pkap
->nfds
) {
1425 kev
.flags
= EV_DISABLE
|EV_DELETE
;
1426 kqueue_register(&pkap
->lwp
->lwp_kqueue
, &kev
);
1428 kprintf("poll index %d out of range against serial %d\n",
1429 pi
, pkap
->lwp
->lwp_kqueue_serial
);
1432 pfd
= &pkap
->fds
[pi
];
1433 if (kevp
[i
].ident
== pfd
->fd
) {
1435 * A single descriptor may generate an error against
1436 * more than one filter, make sure to set the
1437 * appropriate flags but do not increment (*res)
1440 count_res
= (pfd
->revents
== 0);
1441 if (kevp
[i
].flags
& EV_ERROR
) {
1442 switch(kevp
[i
].data
) {
1445 /* Bad file descriptor */
1448 pfd
->revents
|= POLLNVAL
;
1452 * Poll silently swallows any unknown
1453 * errors except in the case of POLLPRI
1454 * (OOB/urgent data).
1456 * ALWAYS filter out EOPNOTSUPP errors
1457 * from filters, common applications
1458 * set POLLPRI|POLLRDBAND and most
1459 * filters do not support EVFILT_EXCEPT.
1461 * We also filter out ENODEV since dev_dkqfilter
1462 * returns ENODEV if EOPNOTSUPP is returned in an
1467 if (kevp
[i
].filter
!= EVFILT_READ
&&
1468 kevp
[i
].filter
!= EVFILT_WRITE
&&
1469 kevp
[i
].data
!= EOPNOTSUPP
&&
1470 kevp
[i
].data
!= ENODEV
) {
1473 pfd
->revents
|= POLLERR
;
1478 kprintf("poll index %d fd %d "
1479 "filter %d error %jd\n",
1482 (intmax_t)kevp
[i
].data
);
1487 switch (kevp
[i
].filter
) {
1491 * NODATA on the read side can indicate a
1492 * half-closed situation and not necessarily
1493 * a disconnect, so depend on the user
1494 * issuing a read() and getting 0 bytes back.
1496 if (kevp
[i
].flags
& EV_NODATA
)
1497 pfd
->revents
|= POLLHUP
;
1499 if ((kevp
[i
].flags
& EV_EOF
) &&
1500 kevp
[i
].fflags
!= 0)
1501 pfd
->revents
|= POLLERR
;
1502 if (pfd
->events
& POLLIN
)
1503 pfd
->revents
|= POLLIN
;
1504 if (pfd
->events
& POLLRDNORM
)
1505 pfd
->revents
|= POLLRDNORM
;
1509 * As per the OpenGroup POLLHUP is mutually
1510 * exclusive with the writability flags. I
1511 * consider this a bit broken but...
1513 * In this case a disconnect is implied even
1514 * for a half-closed (write side) situation.
1516 if (kevp
[i
].flags
& EV_EOF
) {
1517 pfd
->revents
|= POLLHUP
;
1518 if (kevp
[i
].fflags
!= 0)
1519 pfd
->revents
|= POLLERR
;
1521 if (pfd
->events
& POLLOUT
)
1522 pfd
->revents
|= POLLOUT
;
1523 if (pfd
->events
& POLLWRNORM
)
1524 pfd
->revents
|= POLLWRNORM
;
1529 * EV_NODATA should never be tagged for this
1532 if (pfd
->events
& POLLPRI
)
1533 pfd
->revents
|= POLLPRI
;
1534 if (pfd
->events
& POLLRDBAND
)
1535 pfd
->revents
|= POLLRDBAND
;
1540 kprintf("poll index %d/%d fd %d revents %08x\n",
1541 pi
, pkap
->nfds
, pfd
->fd
, pfd
->revents
);
1544 if (count_res
&& pfd
->revents
)
1548 kprintf("poll index %d mismatch %ju/%d\n",
1549 pi
, (uintmax_t)kevp
[i
].ident
, pfd
->fd
);
1558 dopoll(int nfds
, struct pollfd
*fds
, struct timespec
*ts
, int *res
, int flags
)
1560 struct poll_kevent_copyin_args ka
;
1561 struct pollfd sfds
[64];
1569 if (nfds
== 0 && ts
)
1570 return (dotimeout_only(ts
));
1573 * This is a bit arbitrary but we need to limit internal kmallocs.
1575 if (nfds
> maxfilesperproc
* 2)
1576 nfds
= maxfilesperproc
* 2;
1577 bytes
= sizeof(struct pollfd
) * nfds
;
1579 ka
.lwp
= curthread
->td_lwp
;
1587 ka
.fds
= kmalloc(bytes
, M_SELECT
, M_WAITOK
);
1589 error
= copyin(fds
, ka
.fds
, bytes
);
1591 error
= kern_kevent(&ka
.lwp
->lwp_kqueue
, 0x7FFFFFFF, res
, &ka
,
1592 poll_copyin
, poll_copyout
, ts
, flags
);
1595 error
= copyout(ka
.fds
, fds
, bytes
);
1598 kfree(ka
.fds
, M_SELECT
);
1600 ka
.lwp
->lwp_kqueue_serial
+= nfds
;
1606 socket_wait_copyin(void *arg
, struct kevent
*kevp
, int maxevents
, int *events
)
1612 socket_wait_copyout(void *arg
, struct kevent
*kevp
, int count
, int *res
)
1618 extern struct fileops socketops
;
1621 * NOTE: Callers of socket_wait() must already have a reference on the
1625 socket_wait(struct socket
*so
, struct timespec
*ts
, int *res
)
1627 struct thread
*td
= curthread
;
1633 if ((error
= falloc(td
->td_lwp
, &fp
, &fd
)) != 0)
1636 fp
->f_type
= DTYPE_SOCKET
;
1637 fp
->f_flag
= FREAD
| FWRITE
;
1638 fp
->f_ops
= &socketops
;
1640 fsetfd(td
->td_lwp
->lwp_proc
->p_fd
, fp
, fd
);
1641 fsetfdflags(td
->td_proc
->p_fd
, fd
, UF_EXCLOSE
);
1643 bzero(&kq
, sizeof(kq
));
1644 kqueue_init(&kq
, td
->td_lwp
->lwp_proc
->p_fd
);
1645 EV_SET(&kev
, fd
, EVFILT_READ
, EV_ADD
|EV_ENABLE
, 0, 0, NULL
);
1646 if ((error
= kqueue_register(&kq
, &kev
)) != 0) {
1651 error
= kern_kevent(&kq
, 1, res
, NULL
, socket_wait_copyin
,
1652 socket_wait_copyout
, ts
, 0);
1654 EV_SET(&kev
, fd
, EVFILT_READ
, EV_DELETE
|EV_DISABLE
, 0, 0, NULL
);
1655 kqueue_register(&kq
, &kev
);
1656 fp
->f_ops
= &badfileops
;
1663 * OpenBSD poll system call.
1664 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1669 sys_openbsd_poll(struct openbsd_poll_args
*uap
)
1671 return (sys_poll((struct poll_args
*)uap
));
1676 seltrue(cdev_t dev
, int events
)
1678 return (events
& (POLLIN
| POLLOUT
| POLLRDNORM
| POLLWRNORM
));