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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94
39 * $FreeBSD: src/sys/kern/sys_generic.c,v 1.55.2.10 2001/03/17 10:39:32 peter Exp $
42 #include "opt_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/event.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/fcntl.h>
53 #include <sys/signalvar.h>
54 #include <sys/socketvar.h>
56 #include <sys/kernel.h>
57 #include <sys/kern_syscall.h>
58 #include <sys/malloc.h>
59 #include <sys/mapped_ioctl.h>
61 #include <sys/queue.h>
62 #include <sys/resourcevar.h>
63 #include <sys/socketops.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
68 #include <sys/ktrace.h>
71 #include <vm/vm_page.h>
73 #include <sys/file2.h>
74 #include <sys/mplock2.h>
75 #include <sys/spinlock2.h>
77 #include <machine/limits.h>
79 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
80 static MALLOC_DEFINE(M_IOCTLMAP, "ioctlmap", "mapped ioctl handler buffer");
81 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
82 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
84 typedef struct kfd_set {
88 enum select_copyin_states {
89 COPYIN_READ, COPYIN_WRITE, COPYIN_EXCEPT, COPYIN_DONE };
91 struct select_kevent_copyin_args {
95 int active_set; /* One of select_copyin_states */
96 struct lwp *lwp; /* Pointer to our lwp */
97 int num_fds; /* Number of file descriptors (syscall arg) */
98 int proc_fds; /* Processed fd's (wraps) */
99 int error; /* Returned to userland */
102 struct poll_kevent_copyin_args {
110 static struct lwkt_token mioctl_token = LWKT_TOKEN_INITIALIZER(mioctl_token);
112 static int doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex,
113 struct timespec *ts, int *res);
114 static int dopoll(int nfds, struct pollfd *fds, struct timespec *ts,
116 static int dofileread(int, struct file *, struct uio *, int, size_t *);
117 static int dofilewrite(int, struct file *, struct uio *, int, size_t *);
125 sys_read(struct read_args *uap)
127 struct thread *td = curthread;
132 if ((ssize_t)uap->nbyte < 0)
135 aiov.iov_base = uap->buf;
136 aiov.iov_len = uap->nbyte;
137 auio.uio_iov = &aiov;
139 auio.uio_offset = -1;
140 auio.uio_resid = uap->nbyte;
141 auio.uio_rw = UIO_READ;
142 auio.uio_segflg = UIO_USERSPACE;
145 error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
150 * Positioned (Pread) read system call
155 sys_extpread(struct extpread_args *uap)
157 struct thread *td = curthread;
163 if ((ssize_t)uap->nbyte < 0)
166 aiov.iov_base = uap->buf;
167 aiov.iov_len = uap->nbyte;
168 auio.uio_iov = &aiov;
170 auio.uio_offset = uap->offset;
171 auio.uio_resid = uap->nbyte;
172 auio.uio_rw = UIO_READ;
173 auio.uio_segflg = UIO_USERSPACE;
176 flags = uap->flags & O_FMASK;
177 if (uap->offset != (off_t)-1)
180 error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
185 * Scatter read system call.
190 sys_readv(struct readv_args *uap)
192 struct thread *td = curthread;
194 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
197 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
202 auio.uio_iovcnt = uap->iovcnt;
203 auio.uio_offset = -1;
204 auio.uio_rw = UIO_READ;
205 auio.uio_segflg = UIO_USERSPACE;
208 error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
210 iovec_free(&iov, aiov);
216 * Scatter positioned read system call.
221 sys_extpreadv(struct extpreadv_args *uap)
223 struct thread *td = curthread;
225 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
229 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
234 auio.uio_iovcnt = uap->iovcnt;
235 auio.uio_offset = uap->offset;
236 auio.uio_rw = UIO_READ;
237 auio.uio_segflg = UIO_USERSPACE;
240 flags = uap->flags & O_FMASK;
241 if (uap->offset != (off_t)-1)
244 error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
246 iovec_free(&iov, aiov);
254 kern_preadv(int fd, struct uio *auio, int flags, size_t *res)
256 struct thread *td = curthread;
257 struct proc *p = td->td_proc;
263 fp = holdfp(p->p_fd, fd, FREAD);
266 if (flags & O_FOFFSET && fp->f_type != DTYPE_VNODE) {
269 error = dofileread(fd, fp, auio, flags, res);
276 * Common code for readv and preadv that reads data in
277 * from a file using the passed in uio, offset, and flags.
279 * MPALMOSTSAFE - ktrace needs help
282 dofileread(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
287 struct thread *td = curthread;
288 struct iovec *ktriov = NULL;
294 * if tracing, save a copy of iovec
296 if (KTRPOINT(td, KTR_GENIO)) {
297 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
299 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
300 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
304 len = auio->uio_resid;
305 error = fo_read(fp, auio, fp->f_cred, flags);
307 if (auio->uio_resid != len && (error == ERESTART ||
308 error == EINTR || error == EWOULDBLOCK))
312 if (ktriov != NULL) {
314 ktruio.uio_iov = ktriov;
315 ktruio.uio_resid = len - auio->uio_resid;
317 ktrgenio(td->td_lwp, fd, UIO_READ, &ktruio, error);
320 kfree(ktriov, M_TEMP);
324 *res = len - auio->uio_resid;
335 sys_write(struct write_args *uap)
337 struct thread *td = curthread;
342 if ((ssize_t)uap->nbyte < 0)
345 aiov.iov_base = (void *)(uintptr_t)uap->buf;
346 aiov.iov_len = uap->nbyte;
347 auio.uio_iov = &aiov;
349 auio.uio_offset = -1;
350 auio.uio_resid = uap->nbyte;
351 auio.uio_rw = UIO_WRITE;
352 auio.uio_segflg = UIO_USERSPACE;
355 error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
366 sys_extpwrite(struct extpwrite_args *uap)
368 struct thread *td = curthread;
374 if ((ssize_t)uap->nbyte < 0)
377 aiov.iov_base = (void *)(uintptr_t)uap->buf;
378 aiov.iov_len = uap->nbyte;
379 auio.uio_iov = &aiov;
381 auio.uio_offset = uap->offset;
382 auio.uio_resid = uap->nbyte;
383 auio.uio_rw = UIO_WRITE;
384 auio.uio_segflg = UIO_USERSPACE;
387 flags = uap->flags & O_FMASK;
388 if (uap->offset != (off_t)-1)
390 error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
398 sys_writev(struct writev_args *uap)
400 struct thread *td = curthread;
402 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
405 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
410 auio.uio_iovcnt = uap->iovcnt;
411 auio.uio_offset = -1;
412 auio.uio_rw = UIO_WRITE;
413 auio.uio_segflg = UIO_USERSPACE;
416 error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
418 iovec_free(&iov, aiov);
424 * Gather positioned write system call
429 sys_extpwritev(struct extpwritev_args *uap)
431 struct thread *td = curthread;
433 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
437 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
442 auio.uio_iovcnt = uap->iovcnt;
443 auio.uio_offset = uap->offset;
444 auio.uio_rw = UIO_WRITE;
445 auio.uio_segflg = UIO_USERSPACE;
448 flags = uap->flags & O_FMASK;
449 if (uap->offset != (off_t)-1)
452 error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
454 iovec_free(&iov, aiov);
462 kern_pwritev(int fd, struct uio *auio, int flags, size_t *res)
464 struct thread *td = curthread;
465 struct proc *p = td->td_proc;
471 fp = holdfp(p->p_fd, fd, FWRITE);
474 else if ((flags & O_FOFFSET) && fp->f_type != DTYPE_VNODE) {
477 error = dofilewrite(fd, fp, auio, flags, res);
485 * Common code for writev and pwritev that writes data to
486 * a file using the passed in uio, offset, and flags.
488 * MPALMOSTSAFE - ktrace needs help
491 dofilewrite(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
493 struct thread *td = curthread;
494 struct lwp *lp = td->td_lwp;
498 struct iovec *ktriov = NULL;
504 * if tracing, save a copy of iovec and uio
506 if (KTRPOINT(td, KTR_GENIO)) {
507 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
509 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
510 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
514 len = auio->uio_resid;
515 error = fo_write(fp, auio, fp->f_cred, flags);
517 if (auio->uio_resid != len && (error == ERESTART ||
518 error == EINTR || error == EWOULDBLOCK))
520 /* Socket layer is responsible for issuing SIGPIPE. */
522 lwpsignal(lp->lwp_proc, lp, SIGPIPE);
525 if (ktriov != NULL) {
527 ktruio.uio_iov = ktriov;
528 ktruio.uio_resid = len - auio->uio_resid;
530 ktrgenio(lp, fd, UIO_WRITE, &ktruio, error);
533 kfree(ktriov, M_TEMP);
537 *res = len - auio->uio_resid;
548 sys_ioctl(struct ioctl_args *uap)
552 error = mapped_ioctl(uap->fd, uap->com, uap->data, NULL, &uap->sysmsg);
556 struct ioctl_map_entry {
558 struct ioctl_map_range *cmd_ranges;
559 LIST_ENTRY(ioctl_map_entry) entries;
563 * The true heart of all ioctl syscall handlers (native, emulation).
564 * If map != NULL, it will be searched for a matching entry for com,
565 * and appropriate conversions/conversion functions will be utilized.
570 mapped_ioctl(int fd, u_long com, caddr_t uspc_data, struct ioctl_map *map,
573 struct thread *td = curthread;
574 struct proc *p = td->td_proc;
577 struct ioctl_map_range *iomc = NULL;
583 #define STK_PARAMS 128
585 char stkbuf[STK_PARAMS];
592 fp = holdfp(p->p_fd, fd, FREAD|FWRITE);
596 if (map != NULL) { /* obey translation map */
598 struct ioctl_map_entry *e;
600 maskcmd = com & map->mask;
602 lwkt_gettoken(&mioctl_token);
603 LIST_FOREACH(e, &map->mapping, entries) {
604 for (iomc = e->cmd_ranges; iomc->start != 0 ||
605 iomc->maptocmd != 0 || iomc->wrapfunc != NULL ||
606 iomc->mapfunc != NULL;
608 if (maskcmd >= iomc->start &&
609 maskcmd <= iomc->end)
613 /* Did we find a match? */
614 if (iomc->start != 0 || iomc->maptocmd != 0 ||
615 iomc->wrapfunc != NULL || iomc->mapfunc != NULL)
618 lwkt_reltoken(&mioctl_token);
621 (iomc->start == 0 && iomc->maptocmd == 0
622 && iomc->wrapfunc == NULL && iomc->mapfunc == NULL)) {
623 kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n",
624 map->sys, fd, maskcmd,
625 (int)((maskcmd >> 8) & 0xff),
626 (int)(maskcmd & 0xff));
632 * If it's a non-range one to one mapping, maptocmd should be
633 * correct. If it's a ranged one to one mapping, we pass the
634 * original value of com, and for a range mapped to a different
635 * range, we always need a mapping function to translate the
636 * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff
638 if (iomc->start == iomc->end && iomc->maptocmd == iomc->maptoend) {
639 com = iomc->maptocmd;
640 } else if (iomc->start == iomc->maptocmd && iomc->end == iomc->maptoend) {
641 if (iomc->mapfunc != NULL)
642 com = iomc->mapfunc(iomc->start, iomc->end,
643 iomc->start, iomc->end,
646 if (iomc->mapfunc != NULL) {
647 com = iomc->mapfunc(iomc->start, iomc->end,
648 iomc->maptocmd, iomc->maptoend,
651 kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n",
652 map->sys, fd, maskcmd,
653 (int)((maskcmd >> 8) & 0xff),
654 (int)(maskcmd & 0xff));
663 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
666 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
671 * Interpret high order word to find amount of data to be
672 * copied to/from the user's address space.
674 size = IOCPARM_LEN(com);
675 if (size > IOCPARM_MAX) {
680 if (size > sizeof (ubuf.stkbuf)) {
681 memp = kmalloc(size, M_IOCTLOPS, M_WAITOK);
687 if ((com & IOC_IN) != 0) {
689 error = copyin(uspc_data, data, (size_t)size);
692 kfree(memp, M_IOCTLOPS);
696 *(caddr_t *)data = uspc_data;
698 } else if ((com & IOC_OUT) != 0 && size) {
700 * Zero the buffer so the user always
701 * gets back something deterministic.
703 bzero(data, (size_t)size);
704 } else if ((com & IOC_VOID) != 0) {
705 *(caddr_t *)data = uspc_data;
710 if ((tmp = *(int *)data))
711 atomic_set_int(&fp->f_flag, FNONBLOCK);
713 atomic_clear_int(&fp->f_flag, FNONBLOCK);
718 if ((tmp = *(int *)data))
719 atomic_set_int(&fp->f_flag, FASYNC);
721 atomic_clear_int(&fp->f_flag, FASYNC);
722 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred, msg);
727 * If there is a override function,
728 * call it instead of directly routing the call
730 if (map != NULL && iomc->wrapfunc != NULL)
731 error = iomc->wrapfunc(fp, com, ocom, data, cred);
733 error = fo_ioctl(fp, com, data, cred, msg);
735 * Copy any data to user, size was
736 * already set and checked above.
738 if (error == 0 && (com & IOC_OUT) != 0 && size != 0)
739 error = copyout(data, uspc_data, (size_t)size);
743 kfree(memp, M_IOCTLOPS);
753 mapped_ioctl_register_handler(struct ioctl_map_handler *he)
755 struct ioctl_map_entry *ne;
757 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL &&
758 he->subsys != NULL && *he->subsys != '\0');
760 ne = kmalloc(sizeof(struct ioctl_map_entry), M_IOCTLMAP,
763 ne->subsys = he->subsys;
764 ne->cmd_ranges = he->cmd_ranges;
766 lwkt_gettoken(&mioctl_token);
767 LIST_INSERT_HEAD(&he->map->mapping, ne, entries);
768 lwkt_reltoken(&mioctl_token);
777 mapped_ioctl_unregister_handler(struct ioctl_map_handler *he)
779 struct ioctl_map_entry *ne;
782 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL);
784 lwkt_gettoken(&mioctl_token);
785 LIST_FOREACH(ne, &he->map->mapping, entries) {
786 if (ne->cmd_ranges == he->cmd_ranges) {
787 LIST_REMOVE(ne, entries);
788 kfree(ne, M_IOCTLMAP);
793 lwkt_reltoken(&mioctl_token);
797 static int nselcoll; /* Select collisions since boot */
799 SYSCTL_INT(_kern, OID_AUTO, nselcoll, CTLFLAG_RD, &nselcoll, 0, "");
800 static int nseldebug;
801 SYSCTL_INT(_kern, OID_AUTO, nseldebug, CTLFLAG_RW, &nseldebug, 0, "");
804 * Select system call.
809 sys_select(struct select_args *uap)
812 struct timespec *ktsp, kts;
816 * Get timeout if any.
818 if (uap->tv != NULL) {
819 error = copyin(uap->tv, &ktv, sizeof (ktv));
822 TIMEVAL_TO_TIMESPEC(&ktv, &kts);
831 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
832 &uap->sysmsg_result);
839 * Pselect system call.
842 sys_pselect(struct pselect_args *uap)
844 struct thread *td = curthread;
845 struct lwp *lp = td->td_lwp;
846 struct timespec *ktsp, kts;
851 * Get timeout if any.
853 if (uap->ts != NULL) {
854 error = copyin(uap->ts, &kts, sizeof (kts));
863 * Install temporary signal mask if any provided.
865 if (uap->sigmask != NULL) {
866 error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
869 lwkt_gettoken(&lp->lwp_proc->p_token);
870 lp->lwp_oldsigmask = lp->lwp_sigmask;
871 SIG_CANTMASK(sigmask);
872 lp->lwp_sigmask = sigmask;
873 lwkt_reltoken(&lp->lwp_proc->p_token);
879 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
880 &uap->sysmsg_result);
882 if (uap->sigmask != NULL) {
883 lwkt_gettoken(&lp->lwp_proc->p_token);
884 /* doselect() responsible for turning ERESTART into EINTR */
885 KKASSERT(error != ERESTART);
886 if (error == EINTR) {
888 * We can't restore the previous signal mask now
889 * because it could block the signal that interrupted
890 * us. So make a note to restore it after executing
893 lp->lwp_flags |= LWP_OLDMASK;
896 * No handler to run. Restore previous mask immediately.
898 lp->lwp_sigmask = lp->lwp_oldsigmask;
900 lwkt_reltoken(&lp->lwp_proc->p_token);
907 select_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
909 struct select_kevent_copyin_args *skap = NULL;
916 skap = (struct select_kevent_copyin_args *)arg;
918 if (*events == maxevents)
921 while (skap->active_set < COPYIN_DONE) {
922 switch (skap->active_set) {
925 * Register descriptors for the read filter
927 fdp = skap->read_set;
928 filter = EVFILT_READ;
929 fflags = NOTE_OLDAPI;
937 * Register descriptors for the write filter
939 fdp = skap->write_set;
940 filter = EVFILT_WRITE;
941 fflags = NOTE_OLDAPI;
949 * Register descriptors for the exception filter
951 fdp = skap->except_set;
952 filter = EVFILT_EXCEPT;
953 fflags = NOTE_OLDAPI | NOTE_OOB;
961 * Nothing left to register
967 while (skap->proc_fds < skap->num_fds) {
969 if (FD_ISSET(fd, fdp)) {
970 kev = &kevp[*events];
971 EV_SET(kev, fd, filter,
975 skap->lwp->lwp_kqueue_serial);
980 kprintf("select fd %d filter %d serial %d\n",
981 fd, filter, skap->lwp->lwp_kqueue_serial);
984 if (*events == maxevents)
995 select_copyout(void *arg, struct kevent *kevp, int count, int *res)
997 struct select_kevent_copyin_args *skap;
1001 skap = (struct select_kevent_copyin_args *)arg;
1003 for (i = 0; i < count; ++i) {
1005 * Filter out and delete spurious events
1007 if ((u_int)(uintptr_t)kevp[i].udata !=
1008 skap->lwp->lwp_kqueue_serial) {
1010 kev.flags = EV_DISABLE|EV_DELETE;
1011 kqueue_register(&skap->lwp->lwp_kqueue, &kev);
1013 kprintf("select fd %ju mismatched serial %d\n",
1014 (uintmax_t)kevp[i].ident,
1015 skap->lwp->lwp_kqueue_serial);
1022 if (kevp[i].flags & EV_ERROR) {
1023 int error = kevp[i].data;
1028 * A bad file descriptor is considered a
1029 * fatal error for select, bail out.
1031 skap->error = error;
1037 * Select silently swallows any unknown errors
1038 * for descriptors in the read or write sets.
1040 * ALWAYS filter out EOPNOTSUPP errors from
1041 * filters (at least until all filters support
1044 if (kevp[i].filter != EVFILT_READ &&
1045 kevp[i].filter != EVFILT_WRITE &&
1046 error != EOPNOTSUPP) {
1047 skap->error = error;
1054 kprintf("select fd %ju filter %d error %d\n",
1055 (uintmax_t)kevp[i].ident,
1056 kevp[i].filter, error);
1060 switch (kevp[i].filter) {
1062 FD_SET(kevp[i].ident, skap->read_set);
1065 FD_SET(kevp[i].ident, skap->write_set);
1068 FD_SET(kevp[i].ident, skap->except_set);
1079 * Copy select bits in from userland. Allocate kernel memory if the
1083 getbits(int bytes, fd_set *in_set, kfd_set **out_set, kfd_set *tmp_set)
1088 if (bytes < sizeof(*tmp_set))
1091 *out_set = kmalloc(bytes, M_SELECT, M_WAITOK);
1092 error = copyin(in_set, *out_set, bytes);
1101 * Copy returned select bits back out to userland.
1104 putbits(int bytes, kfd_set *in_set, fd_set *out_set)
1109 error = copyout(in_set, out_set, bytes);
1117 dotimeout_only(struct timespec *ts)
1119 return(nanosleep1(ts, NULL));
1123 * Common code for sys_select() and sys_pselect().
1125 * in, out and ex are userland pointers. ts must point to validated
1126 * kernel-side timeout value or NULL for infinite timeout. res must
1127 * point to syscall return value.
1130 doselect(int nd, fd_set *read, fd_set *write, fd_set *except,
1131 struct timespec *ts, int *res)
1133 struct proc *p = curproc;
1134 struct select_kevent_copyin_args *kap, ka;
1144 return (dotimeout_only(ts));
1146 if (nd > p->p_fd->fd_nfiles) /* limit kmalloc */
1147 nd = p->p_fd->fd_nfiles;
1150 kap->lwp = curthread->td_lwp;
1154 kap->active_set = COPYIN_READ;
1157 * Calculate bytes based on the number of __fd_mask[] array entries
1158 * multiplied by the size of __fd_mask.
1160 bytes = howmany(nd, __NFDBITS) * sizeof(__fd_mask);
1162 /* kap->read_set = NULL; not needed */
1163 kap->write_set = NULL;
1164 kap->except_set = NULL;
1166 error = getbits(bytes, read, &kap->read_set, &read_tmp);
1168 error = getbits(bytes, write, &kap->write_set, &write_tmp);
1170 error = getbits(bytes, except, &kap->except_set, &except_tmp);
1175 * NOTE: Make sure the max events passed to kern_kevent() is
1176 * effectively unlimited. (nd * 3) accomplishes this.
1178 * (*res) continues to increment as returned events are
1181 error = kern_kevent(&kap->lwp->lwp_kqueue, 0x7FFFFFFF, res, kap,
1182 select_copyin, select_copyout, ts);
1184 error = putbits(bytes, kap->read_set, read);
1186 error = putbits(bytes, kap->write_set, write);
1188 error = putbits(bytes, kap->except_set, except);
1191 * An error from an individual event that should be passed
1192 * back to userland (EBADF)
1201 if (kap->read_set && kap->read_set != &read_tmp)
1202 kfree(kap->read_set, M_SELECT);
1203 if (kap->write_set && kap->write_set != &write_tmp)
1204 kfree(kap->write_set, M_SELECT);
1205 if (kap->except_set && kap->except_set != &except_tmp)
1206 kfree(kap->except_set, M_SELECT);
1208 kap->lwp->lwp_kqueue_serial += kap->num_fds;
1219 sys_poll(struct poll_args *uap)
1221 struct timespec ts, *tsp;
1224 if (uap->timeout != INFTIM) {
1225 ts.tv_sec = uap->timeout / 1000;
1226 ts.tv_nsec = (uap->timeout % 1000) * 1000 * 1000;
1232 error = dopoll(uap->nfds, uap->fds, tsp, &uap->sysmsg_result);
1238 poll_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1240 struct poll_kevent_copyin_args *pkap;
1245 pkap = (struct poll_kevent_copyin_args *)arg;
1247 while (pkap->pfds < pkap->nfds) {
1248 pfd = &pkap->fds[pkap->pfds];
1250 /* Clear return events */
1253 /* Do not check if fd is equal to -1 */
1254 if (pfd->fd == -1) {
1260 if (pfd->events & (POLLIN | POLLRDNORM))
1262 if (pfd->events & (POLLOUT | POLLWRNORM))
1264 if (pfd->events & (POLLPRI | POLLRDBAND))
1267 if (*events + kev_count > maxevents)
1271 * NOTE: A combined serial number and poll array index is
1272 * stored in kev->udata.
1274 kev = &kevp[*events];
1275 if (pfd->events & (POLLIN | POLLRDNORM)) {
1276 EV_SET(kev++, pfd->fd, EVFILT_READ, EV_ADD|EV_ENABLE,
1277 NOTE_OLDAPI, 0, (void *)(uintptr_t)
1278 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1280 if (pfd->events & (POLLOUT | POLLWRNORM)) {
1281 EV_SET(kev++, pfd->fd, EVFILT_WRITE, EV_ADD|EV_ENABLE,
1282 NOTE_OLDAPI, 0, (void *)(uintptr_t)
1283 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1285 if (pfd->events & (POLLPRI | POLLRDBAND)) {
1286 EV_SET(kev++, pfd->fd, EVFILT_EXCEPT, EV_ADD|EV_ENABLE,
1287 NOTE_OLDAPI | NOTE_OOB, 0,
1289 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1293 kprintf("poll index %d/%d fd %d events %08x serial %d\n",
1294 pkap->pfds, pkap->nfds-1, pfd->fd, pfd->events,
1295 pkap->lwp->lwp_kqueue_serial);
1299 (*events) += kev_count;
1306 poll_copyout(void *arg, struct kevent *kevp, int count, int *res)
1308 struct poll_kevent_copyin_args *pkap;
1315 pkap = (struct poll_kevent_copyin_args *)arg;
1317 for (i = 0; i < count; ++i) {
1319 * Extract the poll array index and delete spurious events.
1320 * We can easily tell if the serial number is incorrect
1321 * by checking whether the extracted index is out of range.
1323 pi = (u_int)(uintptr_t)kevp[i].udata -
1324 (u_int)pkap->lwp->lwp_kqueue_serial;
1326 if (pi >= pkap->nfds) {
1328 kev.flags = EV_DISABLE|EV_DELETE;
1329 kqueue_register(&pkap->lwp->lwp_kqueue, &kev);
1331 kprintf("poll index %d out of range against serial %d\n",
1332 pi, pkap->lwp->lwp_kqueue_serial);
1335 pfd = &pkap->fds[pi];
1336 if (kevp[i].ident == pfd->fd) {
1338 * A single descriptor may generate an error against
1339 * more than one filter, make sure to set the
1340 * appropriate flags but do not increment (*res)
1343 count_res = (pfd->revents == 0);
1344 if (kevp[i].flags & EV_ERROR) {
1345 switch(kevp[i].data) {
1347 /* Bad file descriptor */
1350 pfd->revents |= POLLNVAL;
1354 * Poll silently swallows any unknown
1355 * errors except in the case of POLLPRI
1356 * (OOB/urgent data).
1358 * ALWAYS filter out EOPNOTSUPP errors
1359 * from filters, common applications
1360 * set POLLPRI|POLLRDBAND and most
1361 * filters do not support EVFILT_EXCEPT.
1363 if (kevp[i].filter != EVFILT_READ &&
1364 kevp[i].filter != EVFILT_WRITE &&
1365 kevp[i].data != EOPNOTSUPP) {
1368 pfd->revents |= POLLERR;
1373 kprintf("poll index %d fd %d "
1374 "filter %d error %jd\n",
1377 (intmax_t)kevp[i].data);
1382 switch (kevp[i].filter) {
1386 * NODATA on the read side can indicate a
1387 * half-closed situation and not necessarily
1388 * a disconnect, so depend on the user
1389 * issuing a read() and getting 0 bytes back.
1391 if (kevp[i].flags & EV_NODATA)
1392 pfd->revents |= POLLHUP;
1394 if ((kevp[i].flags & EV_EOF) &&
1395 kevp[i].fflags != 0)
1396 pfd->revents |= POLLERR;
1397 if (pfd->events & POLLIN)
1398 pfd->revents |= POLLIN;
1399 if (pfd->events & POLLRDNORM)
1400 pfd->revents |= POLLRDNORM;
1404 * As per the OpenGroup POLLHUP is mutually
1405 * exclusive with the writability flags. I
1406 * consider this a bit broken but...
1408 * In this case a disconnect is implied even
1409 * for a half-closed (write side) situation.
1411 if (kevp[i].flags & EV_EOF) {
1412 pfd->revents |= POLLHUP;
1413 if (kevp[i].fflags != 0)
1414 pfd->revents |= POLLERR;
1416 if (pfd->events & POLLOUT)
1417 pfd->revents |= POLLOUT;
1418 if (pfd->events & POLLWRNORM)
1419 pfd->revents |= POLLWRNORM;
1424 * EV_NODATA should never be tagged for this
1427 if (pfd->events & POLLPRI)
1428 pfd->revents |= POLLPRI;
1429 if (pfd->events & POLLRDBAND)
1430 pfd->revents |= POLLRDBAND;
1435 kprintf("poll index %d/%d fd %d revents %08x\n",
1436 pi, pkap->nfds, pfd->fd, pfd->revents);
1439 if (count_res && pfd->revents)
1443 kprintf("poll index %d mismatch %ju/%d\n",
1444 pi, (uintmax_t)kevp[i].ident, pfd->fd);
1453 dopoll(int nfds, struct pollfd *fds, struct timespec *ts, int *res)
1455 struct poll_kevent_copyin_args ka;
1456 struct pollfd sfds[64];
1464 if (nfds == 0 && ts)
1465 return (dotimeout_only(ts));
1468 * This is a bit arbitrary but we need to limit internal kmallocs.
1470 if (nfds > maxfilesperproc * 2)
1471 nfds = maxfilesperproc * 2;
1472 bytes = sizeof(struct pollfd) * nfds;
1474 ka.lwp = curthread->td_lwp;
1482 ka.fds = kmalloc(bytes, M_SELECT, M_WAITOK);
1484 error = copyin(fds, ka.fds, bytes);
1486 error = kern_kevent(&ka.lwp->lwp_kqueue, 0x7FFFFFFF, res, &ka,
1487 poll_copyin, poll_copyout, ts);
1490 error = copyout(ka.fds, fds, bytes);
1493 kfree(ka.fds, M_SELECT);
1495 ka.lwp->lwp_kqueue_serial += nfds;
1501 socket_wait_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1507 socket_wait_copyout(void *arg, struct kevent *kevp, int count, int *res)
1513 extern struct fileops socketops;
1516 * NOTE: Callers of socket_wait() must already have a reference on the
1520 socket_wait(struct socket *so, struct timespec *ts, int *res)
1522 struct thread *td = curthread;
1528 if ((error = falloc(td->td_lwp, &fp, &fd)) != 0)
1531 fp->f_type = DTYPE_SOCKET;
1532 fp->f_flag = FREAD | FWRITE;
1533 fp->f_ops = &socketops;
1535 fsetfd(td->td_lwp->lwp_proc->p_fd, fp, fd);
1537 kqueue_init(&kq, td->td_lwp->lwp_proc->p_fd);
1538 EV_SET(&kev, fd, EVFILT_READ, EV_ADD|EV_ENABLE, 0, 0, NULL);
1539 if ((error = kqueue_register(&kq, &kev)) != 0) {
1544 error = kern_kevent(&kq, 1, res, NULL, socket_wait_copyin,
1545 socket_wait_copyout, ts);
1547 EV_SET(&kev, fd, EVFILT_READ, EV_DELETE, 0, 0, NULL);
1548 kqueue_register(&kq, &kev);
1549 fp->f_ops = &badfileops;
1556 * OpenBSD poll system call.
1557 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1562 sys_openbsd_poll(struct openbsd_poll_args *uap)
1564 return (sys_poll((struct poll_args *)uap));
1569 seltrue(cdev_t dev, int events)
1571 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));