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/sysmsg.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>
51 #include <sys/malloc.h>
53 #include <sys/kernel.h>
54 #include <sys/kern_syscall.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 int active_set; /* One of select_copyin_states */
91 struct lwp *lwp; /* Pointer to our lwp */
92 int num_fds; /* Number of file descriptors (syscall arg) */
93 int proc_fds; /* Processed fd's (wraps) */
94 int error; /* Returned to userland */
97 struct poll_kevent_copyin_args {
105 static struct lwkt_token mioctl_token = LWKT_TOKEN_INITIALIZER(mioctl_token);
107 static int doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex,
108 struct timespec *ts, int *res);
109 static int dopoll(int nfds, struct pollfd *fds, struct timespec *ts,
110 int *res, int flags);
111 static int dofileread(int, struct file *, struct uio *, int, size_t *);
112 static int dofilewrite(int, struct file *, struct uio *, int, size_t *);
120 sys_read(struct sysmsg *sysmsg, const struct read_args *uap)
122 struct thread *td = curthread;
127 if ((ssize_t)uap->nbyte < 0)
130 aiov.iov_base = uap->buf;
131 aiov.iov_len = uap->nbyte;
132 auio.uio_iov = &aiov;
134 auio.uio_offset = -1;
135 auio.uio_resid = uap->nbyte;
136 auio.uio_rw = UIO_READ;
137 auio.uio_segflg = UIO_USERSPACE;
140 error = kern_preadv(uap->fd, &auio, 0, &sysmsg->sysmsg_szresult);
145 * Positioned (Pread) read system call
150 sys_extpread(struct sysmsg *sysmsg, const struct extpread_args *uap)
152 struct thread *td = curthread;
158 if ((ssize_t)uap->nbyte < 0)
161 aiov.iov_base = uap->buf;
162 aiov.iov_len = uap->nbyte;
163 auio.uio_iov = &aiov;
165 auio.uio_offset = uap->offset;
166 auio.uio_resid = uap->nbyte;
167 auio.uio_rw = UIO_READ;
168 auio.uio_segflg = UIO_USERSPACE;
171 flags = uap->flags & O_FMASK;
172 if (uap->offset != (off_t)-1)
175 error = kern_preadv(uap->fd, &auio, flags, &sysmsg->sysmsg_szresult);
180 * Scatter read system call.
185 sys_readv(struct sysmsg *sysmsg, const struct readv_args *uap)
187 struct thread *td = curthread;
189 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
192 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
197 auio.uio_iovcnt = uap->iovcnt;
198 auio.uio_offset = -1;
199 auio.uio_rw = UIO_READ;
200 auio.uio_segflg = UIO_USERSPACE;
203 error = kern_preadv(uap->fd, &auio, 0, &sysmsg->sysmsg_szresult);
205 iovec_free(&iov, aiov);
211 * Scatter positioned read system call.
216 sys_extpreadv(struct sysmsg *sysmsg, const struct extpreadv_args *uap)
218 struct thread *td = curthread;
220 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
224 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
229 auio.uio_iovcnt = uap->iovcnt;
230 auio.uio_offset = uap->offset;
231 auio.uio_rw = UIO_READ;
232 auio.uio_segflg = UIO_USERSPACE;
235 flags = uap->flags & O_FMASK;
236 if (uap->offset != (off_t)-1)
239 error = kern_preadv(uap->fd, &auio, flags, &sysmsg->sysmsg_szresult);
241 iovec_free(&iov, aiov);
249 kern_preadv(int fd, struct uio *auio, int flags, size_t *res)
251 struct thread *td = curthread;
255 fp = holdfp(td, fd, FREAD);
258 if (flags & O_FOFFSET && fp->f_type != DTYPE_VNODE) {
261 error = dofileread(fd, fp, auio, flags, res);
269 * Common code for readv and preadv that reads data in
270 * from a file using the passed in uio, offset, and flags.
272 * MPALMOSTSAFE - ktrace needs help
275 dofileread(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
280 struct thread *td = curthread;
281 struct iovec *ktriov = NULL;
287 * if tracing, save a copy of iovec
289 if (KTRPOINT(td, KTR_GENIO)) {
290 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
292 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
293 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
297 len = auio->uio_resid;
298 error = fo_read(fp, auio, fp->f_cred, flags);
300 if (auio->uio_resid != len && (error == ERESTART ||
301 error == EINTR || error == EWOULDBLOCK))
305 if (ktriov != NULL) {
307 ktruio.uio_iov = ktriov;
308 ktruio.uio_resid = len - auio->uio_resid;
309 ktrgenio(td->td_lwp, fd, UIO_READ, &ktruio, error);
311 kfree(ktriov, M_TEMP);
315 *res = len - auio->uio_resid;
326 sys_write(struct sysmsg *sysmsg, const struct write_args *uap)
328 struct thread *td = curthread;
333 if ((ssize_t)uap->nbyte < 0)
336 aiov.iov_base = (void *)(uintptr_t)uap->buf;
337 aiov.iov_len = uap->nbyte;
338 auio.uio_iov = &aiov;
340 auio.uio_offset = -1;
341 auio.uio_resid = uap->nbyte;
342 auio.uio_rw = UIO_WRITE;
343 auio.uio_segflg = UIO_USERSPACE;
346 error = kern_pwritev(uap->fd, &auio, 0, &sysmsg->sysmsg_szresult);
357 sys_extpwrite(struct sysmsg *sysmsg, const struct extpwrite_args *uap)
359 struct thread *td = curthread;
365 if ((ssize_t)uap->nbyte < 0)
368 aiov.iov_base = (void *)(uintptr_t)uap->buf;
369 aiov.iov_len = uap->nbyte;
370 auio.uio_iov = &aiov;
372 auio.uio_offset = uap->offset;
373 auio.uio_resid = uap->nbyte;
374 auio.uio_rw = UIO_WRITE;
375 auio.uio_segflg = UIO_USERSPACE;
378 flags = uap->flags & O_FMASK;
379 if (uap->offset != (off_t)-1)
381 error = kern_pwritev(uap->fd, &auio, flags, &sysmsg->sysmsg_szresult);
389 sys_writev(struct sysmsg *sysmsg, const struct writev_args *uap)
391 struct thread *td = curthread;
393 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
396 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
401 auio.uio_iovcnt = uap->iovcnt;
402 auio.uio_offset = -1;
403 auio.uio_rw = UIO_WRITE;
404 auio.uio_segflg = UIO_USERSPACE;
407 error = kern_pwritev(uap->fd, &auio, 0, &sysmsg->sysmsg_szresult);
409 iovec_free(&iov, aiov);
415 * Gather positioned write system call
420 sys_extpwritev(struct sysmsg *sysmsg, const struct extpwritev_args *uap)
422 struct thread *td = curthread;
424 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
428 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
433 auio.uio_iovcnt = uap->iovcnt;
434 auio.uio_offset = uap->offset;
435 auio.uio_rw = UIO_WRITE;
436 auio.uio_segflg = UIO_USERSPACE;
439 flags = uap->flags & O_FMASK;
440 if (uap->offset != (off_t)-1)
443 error = kern_pwritev(uap->fd, &auio, flags, &sysmsg->sysmsg_szresult);
445 iovec_free(&iov, aiov);
453 kern_pwritev(int fd, struct uio *auio, int flags, size_t *res)
455 struct thread *td = curthread;
459 fp = holdfp(td, fd, FWRITE);
462 else if ((flags & O_FOFFSET) && fp->f_type != DTYPE_VNODE) {
465 error = dofilewrite(fd, fp, auio, flags, res);
473 * Common code for writev and pwritev that writes data to
474 * a file using the passed in uio, offset, and flags.
476 * MPALMOSTSAFE - ktrace needs help
479 dofilewrite(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
481 struct thread *td = curthread;
482 struct lwp *lp = td->td_lwp;
486 struct iovec *ktriov = NULL;
492 * if tracing, save a copy of iovec and uio
494 if (KTRPOINT(td, KTR_GENIO)) {
495 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
497 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
498 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
502 len = auio->uio_resid;
503 error = fo_write(fp, auio, fp->f_cred, flags);
505 if (auio->uio_resid != len && (error == ERESTART ||
506 error == EINTR || error == EWOULDBLOCK))
508 /* Socket layer is responsible for issuing SIGPIPE. */
509 if (error == EPIPE && fp->f_type != DTYPE_SOCKET)
510 lwpsignal(lp->lwp_proc, lp, SIGPIPE);
513 if (ktriov != NULL) {
515 ktruio.uio_iov = ktriov;
516 ktruio.uio_resid = len - auio->uio_resid;
517 ktrgenio(lp, fd, UIO_WRITE, &ktruio, error);
519 kfree(ktriov, M_TEMP);
523 *res = len - auio->uio_resid;
534 sys_ioctl(struct sysmsg *sysmsg, const struct ioctl_args *uap)
538 error = mapped_ioctl(uap->fd, uap->com, uap->data, NULL, sysmsg);
542 struct ioctl_map_entry {
544 struct ioctl_map_range *cmd_ranges;
545 LIST_ENTRY(ioctl_map_entry) entries;
549 * The true heart of all ioctl syscall handlers (native, emulation).
550 * If map != NULL, it will be searched for a matching entry for com,
551 * and appropriate conversions/conversion functions will be utilized.
556 mapped_ioctl(int fd, u_long com, caddr_t uspc_data, struct ioctl_map *map,
559 struct thread *td = curthread;
560 struct proc *p = td->td_proc;
563 struct ioctl_map_range *iomc = NULL;
569 #define STK_PARAMS 128
571 char stkbuf[STK_PARAMS];
579 fp = holdfp(td, fd, FREAD|FWRITE);
583 if (map != NULL) { /* obey translation map */
585 struct ioctl_map_entry *e;
587 maskcmd = com & map->mask;
589 lwkt_gettoken(&mioctl_token);
590 LIST_FOREACH(e, &map->mapping, entries) {
591 for (iomc = e->cmd_ranges; iomc->start != 0 ||
592 iomc->maptocmd != 0 || iomc->wrapfunc != NULL ||
593 iomc->mapfunc != NULL;
595 if (maskcmd >= iomc->start &&
596 maskcmd <= iomc->end)
600 /* Did we find a match? */
601 if (iomc->start != 0 || iomc->maptocmd != 0 ||
602 iomc->wrapfunc != NULL || iomc->mapfunc != NULL)
605 lwkt_reltoken(&mioctl_token);
608 (iomc->start == 0 && iomc->maptocmd == 0
609 && iomc->wrapfunc == NULL && iomc->mapfunc == NULL)) {
610 kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n",
611 map->sys, fd, maskcmd,
612 (int)((maskcmd >> 8) & 0xff),
613 (int)(maskcmd & 0xff));
619 * If it's a non-range one to one mapping, maptocmd should be
620 * correct. If it's a ranged one to one mapping, we pass the
621 * original value of com, and for a range mapped to a different
622 * range, we always need a mapping function to translate the
623 * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff
625 if (iomc->start == iomc->end && iomc->maptocmd == iomc->maptoend) {
626 com = iomc->maptocmd;
627 } else if (iomc->start == iomc->maptocmd && iomc->end == iomc->maptoend) {
628 if (iomc->mapfunc != NULL)
629 com = iomc->mapfunc(iomc->start, iomc->end,
630 iomc->start, iomc->end,
633 if (iomc->mapfunc != NULL) {
634 com = iomc->mapfunc(iomc->start, iomc->end,
635 iomc->maptocmd, iomc->maptoend,
638 kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n",
639 map->sys, fd, maskcmd,
640 (int)((maskcmd >> 8) & 0xff),
641 (int)(maskcmd & 0xff));
650 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
653 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
658 * Interpret high order word to find amount of data to be
659 * copied to/from the user's address space.
661 size = IOCPARM_LEN(com);
662 if (size > IOCPARM_MAX) {
667 if ((com & IOC_VOID) == 0 && size > sizeof(ubuf.stkbuf)) {
668 memp = kmalloc(size, M_IOCTLOPS, M_WAITOK);
674 if (com & IOC_VOID) {
675 *(caddr_t *)data = uspc_data;
676 } else if (com & IOC_IN) {
678 error = copyin(uspc_data, data, (size_t)size);
682 *(caddr_t *)data = uspc_data;
684 } else if ((com & IOC_OUT) != 0 && size) {
686 * Zero the buffer so the user always
687 * gets back something deterministic.
689 bzero(data, (size_t)size);
694 if ((tmp = *(int *)data))
695 atomic_set_int(&fp->f_flag, FNONBLOCK);
697 atomic_clear_int(&fp->f_flag, FNONBLOCK);
702 if ((tmp = *(int *)data))
703 atomic_set_int(&fp->f_flag, FASYNC);
705 atomic_clear_int(&fp->f_flag, FASYNC);
706 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred, msg);
711 * If there is a override function,
712 * call it instead of directly routing the call
714 if (map != NULL && iomc->wrapfunc != NULL)
715 error = iomc->wrapfunc(fp, com, ocom, data, cred);
717 error = fo_ioctl(fp, com, data, cred, msg);
719 * Copy any data to user, size was
720 * already set and checked above.
722 if (error == 0 && (com & IOC_OUT) != 0 && size != 0)
723 error = copyout(data, uspc_data, (size_t)size);
728 kfree(memp, M_IOCTLOPS);
738 mapped_ioctl_register_handler(struct ioctl_map_handler *he)
740 struct ioctl_map_entry *ne;
742 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL &&
743 he->subsys != NULL && *he->subsys != '\0');
745 ne = kmalloc(sizeof(struct ioctl_map_entry), M_IOCTLMAP,
748 ne->subsys = he->subsys;
749 ne->cmd_ranges = he->cmd_ranges;
751 lwkt_gettoken(&mioctl_token);
752 LIST_INSERT_HEAD(&he->map->mapping, ne, entries);
753 lwkt_reltoken(&mioctl_token);
762 mapped_ioctl_unregister_handler(struct ioctl_map_handler *he)
764 struct ioctl_map_entry *ne;
767 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL);
769 lwkt_gettoken(&mioctl_token);
770 LIST_FOREACH(ne, &he->map->mapping, entries) {
771 if (ne->cmd_ranges == he->cmd_ranges) {
772 LIST_REMOVE(ne, entries);
773 kfree(ne, M_IOCTLMAP);
778 lwkt_reltoken(&mioctl_token);
782 static int nseldebug;
783 SYSCTL_INT(_kern, OID_AUTO, nseldebug, CTLFLAG_RW, &nseldebug, 0, "");
786 * Select system call.
791 sys_select(struct sysmsg *sysmsg, const struct select_args *uap)
794 struct timespec *ktsp, kts;
798 * Get timeout if any.
800 if (uap->tv != NULL) {
801 error = copyin(uap->tv, &ktv, sizeof (ktv));
804 TIMEVAL_TO_TIMESPEC(&ktv, &kts);
813 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
814 &sysmsg->sysmsg_result);
821 * Pselect system call.
824 sys_pselect(struct sysmsg *sysmsg, const struct pselect_args *uap)
826 struct thread *td = curthread;
827 struct lwp *lp = td->td_lwp;
828 struct timespec *ktsp, kts;
833 * Get timeout if any.
835 if (uap->ts != NULL) {
836 error = copyin(uap->ts, &kts, sizeof (kts));
845 * Install temporary signal mask if any provided.
847 if (uap->sigmask != NULL) {
848 error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
851 lwkt_gettoken(&lp->lwp_proc->p_token);
852 lp->lwp_oldsigmask = lp->lwp_sigmask;
853 SIG_CANTMASK(sigmask);
854 lp->lwp_sigmask = sigmask;
855 lwkt_reltoken(&lp->lwp_proc->p_token);
861 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
862 &sysmsg->sysmsg_result);
864 if (uap->sigmask != NULL) {
865 lwkt_gettoken(&lp->lwp_proc->p_token);
866 /* doselect() responsible for turning ERESTART into EINTR */
867 KKASSERT(error != ERESTART);
868 if (error == EINTR) {
870 * We can't restore the previous signal mask now
871 * because it could block the signal that interrupted
872 * us. So make a note to restore it after executing
875 lp->lwp_flags |= LWP_OLDMASK;
878 * No handler to run. Restore previous mask immediately.
880 lp->lwp_sigmask = lp->lwp_oldsigmask;
882 lwkt_reltoken(&lp->lwp_proc->p_token);
889 select_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
891 struct select_kevent_copyin_args *skap = NULL;
898 skap = (struct select_kevent_copyin_args *)arg;
900 if (*events == maxevents)
903 while (skap->active_set < COPYIN_DONE) {
904 switch (skap->active_set) {
907 * Register descriptors for the read filter
909 fdp = skap->read_set;
910 filter = EVFILT_READ;
911 fflags = NOTE_OLDAPI;
919 * Register descriptors for the write filter
921 fdp = skap->write_set;
922 filter = EVFILT_WRITE;
923 fflags = NOTE_OLDAPI;
931 * Register descriptors for the exception filter
933 fdp = skap->except_set;
934 filter = EVFILT_EXCEPT;
935 fflags = NOTE_OLDAPI | NOTE_OOB;
943 * Nothing left to register
949 while (skap->proc_fds < skap->num_fds) {
951 if (FD_ISSET(fd, fdp)) {
952 kev = &kevp[*events];
953 EV_SET(kev, fd, filter,
957 skap->lwp->lwp_kqueue_serial);
962 kprintf("select fd %d filter %d "
963 "serial %ju\n", fd, filter,
965 skap->lwp->lwp_kqueue_serial);
969 if (*events == maxevents)
980 select_copyout(void *arg, struct kevent *kevp, int count, int *res)
982 struct select_kevent_copyin_args *skap;
987 skap = (struct select_kevent_copyin_args *)arg;
989 for (i = 0; i < count; ++i) {
991 * Filter out and delete spurious events
993 if ((uint64_t)(uintptr_t)kevp[i].udata !=
994 skap->lwp->lwp_kqueue_serial) {
997 kev.flags = EV_DISABLE|EV_DELETE;
999 kqueue_register(&skap->lwp->lwp_kqueue, &kev, &n);
1001 kprintf("select fd %ju mismatched serial %ju\n",
1002 (uintmax_t)kevp[i].ident,
1003 (uintmax_t)skap->lwp->lwp_kqueue_serial);
1011 if (kevp[i].flags & EV_ERROR) {
1012 int error = kevp[i].data;
1017 * A bad file descriptor is considered a
1018 * fatal error for select, bail out.
1020 skap->error = error;
1026 * Select silently swallows any unknown errors
1027 * for descriptors in the read or write sets.
1029 * ALWAYS filter out EOPNOTSUPP errors from
1030 * filters (at least until all filters support
1033 * We also filter out ENODEV since dev_dkqfilter
1034 * returns ENODEV if EOPNOTSUPP is returned in an
1039 if (kevp[i].filter != EVFILT_READ &&
1040 kevp[i].filter != EVFILT_WRITE &&
1041 error != EOPNOTSUPP &&
1043 skap->error = error;
1051 * We must deregister any unsupported select events
1052 * to avoid a live-lock.
1055 kprintf("select fd %ju filter %d error %d\n",
1056 (uintmax_t)kevp[i].ident,
1057 kevp[i].filter, error);
1062 switch (kevp[i].filter) {
1064 FD_SET(kevp[i].ident, skap->read_set);
1067 FD_SET(kevp[i].ident, skap->write_set);
1070 FD_SET(kevp[i].ident, skap->except_set);
1081 * Copy select bits in from userland. Allocate kernel memory if the
1085 getbits(int bytes, fd_set *in_set, kfd_set **out_set, kfd_set *tmp_set)
1090 if (bytes < sizeof(*tmp_set))
1093 *out_set = kmalloc(bytes, M_SELECT, M_WAITOK);
1094 error = copyin(in_set, *out_set, bytes);
1103 * Copy returned select bits back out to userland.
1106 putbits(int bytes, kfd_set *in_set, fd_set *out_set)
1111 error = copyout(in_set, out_set, bytes);
1119 dotimeout_only(struct timespec *ts)
1121 return(nanosleep1(ts, NULL));
1125 * Common code for sys_select() and sys_pselect().
1127 * in, out and ex are userland pointers. ts must point to validated
1128 * kernel-side timeout value or NULL for infinite timeout. res must
1129 * point to syscall return value.
1132 doselect(int nd, fd_set *read, fd_set *write, fd_set *except,
1133 struct timespec *ts, int *res)
1135 struct proc *p = curproc;
1136 struct select_kevent_copyin_args *kap, ka;
1146 return (dotimeout_only(ts));
1148 if (nd > p->p_fd->fd_nfiles) /* limit kmalloc */
1149 nd = p->p_fd->fd_nfiles;
1152 kap->lwp = curthread->td_lwp;
1156 kap->active_set = COPYIN_READ;
1159 * Calculate bytes based on the number of __fd_mask[] array entries
1160 * multiplied by the size of __fd_mask.
1162 bytes = howmany(nd, __NFDBITS) * sizeof(__fd_mask);
1164 /* kap->read_set = NULL; not needed */
1165 kap->write_set = NULL;
1166 kap->except_set = NULL;
1168 error = getbits(bytes, read, &kap->read_set, &read_tmp);
1170 error = getbits(bytes, write, &kap->write_set, &write_tmp);
1172 error = getbits(bytes, except, &kap->except_set, &except_tmp);
1177 * NOTE: Make sure the max events passed to kern_kevent() is
1178 * effectively unlimited. (nd * 3) accomplishes this.
1180 * (*res) continues to increment as returned events are
1183 error = kern_kevent(&kap->lwp->lwp_kqueue, 0x7FFFFFFF, res, kap,
1184 select_copyin, select_copyout, ts, 0);
1186 error = putbits(bytes, kap->read_set, read);
1188 error = putbits(bytes, kap->write_set, write);
1190 error = putbits(bytes, kap->except_set, except);
1193 * An error from an individual event that should be passed
1194 * back to userland (EBADF)
1203 if (kap->read_set && kap->read_set != &read_tmp)
1204 kfree(kap->read_set, M_SELECT);
1205 if (kap->write_set && kap->write_set != &write_tmp)
1206 kfree(kap->write_set, M_SELECT);
1207 if (kap->except_set && kap->except_set != &except_tmp)
1208 kfree(kap->except_set, M_SELECT);
1210 kap->lwp->lwp_kqueue_serial += kap->num_fds;
1221 sys_poll(struct sysmsg *sysmsg, const struct poll_args *uap)
1223 struct timespec ts, *tsp;
1226 if (uap->timeout != INFTIM) {
1227 if (uap->timeout < 0)
1229 ts.tv_sec = uap->timeout / 1000;
1230 ts.tv_nsec = (uap->timeout % 1000) * 1000 * 1000;
1236 error = dopoll(uap->nfds, uap->fds, tsp, &sysmsg->sysmsg_result, 0);
1242 * Ppoll system call.
1247 sys_ppoll(struct sysmsg *sysmsg, const struct ppoll_args *uap)
1249 struct thread *td = curthread;
1250 struct lwp *lp = td->td_lwp;
1251 struct timespec *ktsp, kts;
1256 * Get timeout if any.
1258 if (uap->ts != NULL) {
1259 error = copyin(uap->ts, &kts, sizeof (kts));
1268 * Install temporary signal mask if any provided.
1270 if (uap->sigmask != NULL) {
1271 error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
1274 lwkt_gettoken(&lp->lwp_proc->p_token);
1275 lp->lwp_oldsigmask = lp->lwp_sigmask;
1276 SIG_CANTMASK(sigmask);
1277 lp->lwp_sigmask = sigmask;
1278 lwkt_reltoken(&lp->lwp_proc->p_token);
1281 error = dopoll(uap->nfds, uap->fds, ktsp, &sysmsg->sysmsg_result,
1282 ktsp != NULL ? KEVENT_TIMEOUT_PRECISE : 0);
1284 if (uap->sigmask != NULL) {
1285 lwkt_gettoken(&lp->lwp_proc->p_token);
1286 /* dopoll() responsible for turning ERESTART into EINTR */
1287 KKASSERT(error != ERESTART);
1288 if (error == EINTR) {
1290 * We can't restore the previous signal mask now
1291 * because it could block the signal that interrupted
1292 * us. So make a note to restore it after executing
1295 lp->lwp_flags |= LWP_OLDMASK;
1298 * No handler to run. Restore previous mask immediately.
1300 lp->lwp_sigmask = lp->lwp_oldsigmask;
1302 lwkt_reltoken(&lp->lwp_proc->p_token);
1309 poll_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1311 struct poll_kevent_copyin_args *pkap;
1316 pkap = (struct poll_kevent_copyin_args *)arg;
1318 while (pkap->pfds < pkap->nfds) {
1319 pfd = &pkap->fds[pkap->pfds];
1321 /* Clear return events */
1324 /* Do not check if fd is equal to -1 */
1325 if (pfd->fd == -1) {
1331 if (pfd->events & (POLLIN | POLLHUP | POLLRDNORM))
1333 if (pfd->events & (POLLOUT | POLLWRNORM))
1335 if (pfd->events & (POLLPRI | POLLRDBAND))
1338 if (*events + kev_count > maxevents)
1342 * NOTE: A combined serial number and poll array index is
1343 * stored in kev->udata.
1345 kev = &kevp[*events];
1346 if (pfd->events & (POLLIN | POLLHUP | POLLRDNORM)) {
1347 int notes = NOTE_OLDAPI;
1348 if ((pfd->events & (POLLIN | POLLRDNORM)) == 0)
1349 notes |= NOTE_HUPONLY;
1351 EV_SET(kev++, pfd->fd, EVFILT_READ, EV_ADD|EV_ENABLE,
1352 notes, 0, (void *)(uintptr_t)
1353 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1355 if (pfd->events & (POLLOUT | POLLWRNORM)) {
1356 EV_SET(kev++, pfd->fd, EVFILT_WRITE, EV_ADD|EV_ENABLE,
1357 NOTE_OLDAPI, 0, (void *)(uintptr_t)
1358 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1360 if (pfd->events & (POLLPRI | POLLRDBAND)) {
1361 EV_SET(kev++, pfd->fd, EVFILT_EXCEPT, EV_ADD|EV_ENABLE,
1362 NOTE_OLDAPI | NOTE_OOB, 0,
1364 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1368 kprintf("poll index %d/%d fd %d events %08x "
1369 "serial %ju\n", pkap->pfds, pkap->nfds-1,
1370 pfd->fd, pfd->events,
1371 (uintmax_t)pkap->lwp->lwp_kqueue_serial);
1375 (*events) += kev_count;
1382 poll_copyout(void *arg, struct kevent *kevp, int count, int *res)
1384 struct poll_kevent_copyin_args *pkap;
1392 pkap = (struct poll_kevent_copyin_args *)arg;
1394 for (i = 0; i < count; ++i) {
1396 * Extract the poll array index and delete spurious events.
1397 * We can easily tell if the serial number is incorrect
1398 * by checking whether the extracted index is out of range.
1400 pi = (uint64_t)(uintptr_t)kevp[i].udata -
1401 pkap->lwp->lwp_kqueue_serial;
1403 if (pi >= pkap->nfds) {
1406 kev.flags = EV_DISABLE|EV_DELETE;
1408 kqueue_register(&pkap->lwp->lwp_kqueue, &kev, &n);
1410 kprintf("poll index %ju out of range against "
1411 "serial %ju\n", (uintmax_t)pi,
1412 (uintmax_t)pkap->lwp->lwp_kqueue_serial);
1418 * Locate the pollfd and process events
1420 pfd = &pkap->fds[pi];
1421 if (kevp[i].ident == pfd->fd) {
1423 * A single descriptor may generate an error against
1424 * more than one filter, make sure to set the
1425 * appropriate flags but do not increment (*res)
1428 count_res = (pfd->revents == 0);
1429 if (kevp[i].flags & EV_ERROR) {
1430 switch(kevp[i].data) {
1433 /* Bad file descriptor */
1436 pfd->revents |= POLLNVAL;
1440 * Poll silently swallows any unknown
1441 * errors except in the case of POLLPRI
1442 * (OOB/urgent data).
1444 * ALWAYS filter out EOPNOTSUPP errors
1445 * from filters, common applications
1446 * set POLLPRI|POLLRDBAND and most
1447 * filters do not support EVFILT_EXCEPT.
1449 * We also filter out ENODEV since
1450 * dev_dkqfilter returns ENODEV if
1451 * EOPNOTSUPP is returned in an
1456 if (kevp[i].filter != EVFILT_READ &&
1457 kevp[i].filter != EVFILT_WRITE &&
1458 kevp[i].data != EOPNOTSUPP &&
1459 kevp[i].data != ENODEV) {
1462 pfd->revents |= POLLERR;
1466 if (pfd->revents == 0 && nseldebug) {
1467 kprintf("poll index EV_ERROR %ju fd %d "
1468 "filter %d error %jd\n",
1469 (uintmax_t)pi, pfd->fd,
1471 (intmax_t)kevp[i].data);
1475 * Silently deregister any unhandled EV_ERROR
1476 * condition (usually EOPNOTSUPP).
1478 if (pfd->revents == 0)
1483 switch (kevp[i].filter) {
1486 * NODATA on the read side can indicate a
1487 * half-closed situation and not necessarily
1488 * a disconnect, so depend on the user
1489 * issuing a read() and getting 0 bytes back.
1491 * If EV_HUP is set the peer completely
1492 * disconnected and we can set POLLHUP.
1493 * Linux can return POLLHUP even if read
1494 * data has not been drained, so we should
1497 /* if (kevp[i].flags & EV_NODATA) */ {
1498 if (kevp[i].flags & EV_HUP)
1499 pfd->revents |= POLLHUP;
1501 if ((kevp[i].flags & EV_EOF) &&
1502 kevp[i].fflags != 0)
1503 pfd->revents |= POLLERR;
1504 if (pfd->events & POLLIN)
1505 pfd->revents |= POLLIN;
1506 if (pfd->events & POLLRDNORM)
1507 pfd->revents |= POLLRDNORM;
1511 * As per the OpenGroup POLLHUP is mutually
1512 * exclusive with the writability flags. I
1513 * consider this a bit broken but...
1515 * In this case a disconnect is implied even
1516 * for a half-closed (write side) situation.
1518 if (kevp[i].flags & EV_EOF) {
1519 pfd->revents |= POLLHUP;
1520 if (kevp[i].fflags != 0)
1521 pfd->revents |= POLLERR;
1523 if (pfd->events & POLLOUT)
1524 pfd->revents |= POLLOUT;
1525 if (pfd->events & POLLWRNORM)
1526 pfd->revents |= POLLWRNORM;
1531 * EV_NODATA should never be tagged for this
1534 if (pfd->events & POLLPRI)
1535 pfd->revents |= POLLPRI;
1536 if (pfd->events & POLLRDBAND)
1537 pfd->revents |= POLLRDBAND;
1542 kprintf("poll index %ju/%d fd %d "
1543 "revents %08x\n", (uintmax_t)pi, pkap->nfds,
1544 pfd->fd, pfd->revents);
1547 if (count_res && pfd->revents)
1552 * We must deregister any kqueue poll event that does not
1553 * set poll return bits to prevent a live-lock.
1555 if (pfd->revents == 0) {
1556 kprintf("poll index %ju no-action %ju/%d "
1557 "events=%08x kevpfilt=%d/%08x\n",
1558 (uintmax_t)pi, (uintmax_t)kevp[i].ident,
1559 pfd->fd, pfd->events,
1560 kevp[i].filter, kevp[i].flags);
1569 dopoll(int nfds, struct pollfd *fds, struct timespec *ts, int *res, int flags)
1571 struct poll_kevent_copyin_args ka;
1572 struct pollfd sfds[64];
1580 if (nfds == 0 && ts)
1581 return (dotimeout_only(ts));
1584 * This is a bit arbitrary but we need to limit internal kmallocs.
1586 if (nfds > maxfilesperproc * 2)
1587 nfds = maxfilesperproc * 2;
1588 bytes = sizeof(struct pollfd) * nfds;
1590 ka.lwp = curthread->td_lwp;
1598 ka.fds = kmalloc(bytes, M_SELECT, M_WAITOK);
1600 error = copyin(fds, ka.fds, bytes);
1602 error = kern_kevent(&ka.lwp->lwp_kqueue, 0x7FFFFFFF, res, &ka,
1603 poll_copyin, poll_copyout, ts, flags);
1606 error = copyout(ka.fds, fds, bytes);
1609 kfree(ka.fds, M_SELECT);
1611 ka.lwp->lwp_kqueue_serial += nfds;
1617 socket_wait_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1623 socket_wait_copyout(void *arg, struct kevent *kevp, int count, int *res)
1629 extern struct fileops socketops;
1632 * NOTE: Callers of socket_wait() must already have a reference on the
1636 socket_wait(struct socket *so, struct timespec *ts, int *res)
1638 struct thread *td = curthread;
1645 if ((error = falloc(td->td_lwp, &fp, &fd)) != 0)
1648 fp->f_type = DTYPE_SOCKET;
1649 fp->f_flag = FREAD | FWRITE;
1650 fp->f_ops = &socketops;
1652 fsetfd(td->td_lwp->lwp_proc->p_fd, fp, fd);
1653 fsetfdflags(td->td_proc->p_fd, fd, UF_EXCLOSE);
1655 bzero(&kq, sizeof(kq));
1656 kqueue_init(&kq, td->td_lwp->lwp_proc->p_fd);
1657 EV_SET(&kev, fd, EVFILT_READ, EV_ADD|EV_ENABLE, 0, 0, NULL);
1659 if ((error = kqueue_register(&kq, &kev, &n)) != 0) {
1664 error = kern_kevent(&kq, 1, res, NULL, socket_wait_copyin,
1665 socket_wait_copyout, ts, 0);
1667 EV_SET(&kev, fd, EVFILT_READ, EV_DELETE|EV_DISABLE, 0, 0, NULL);
1669 kqueue_register(&kq, &kev, &n);
1670 fp->f_ops = &badfileops;
1677 * OpenBSD poll system call.
1678 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1683 sys_openbsd_poll(struct sysmsg *sysmsg, const struct openbsd_poll_args *uap)
1685 return (sys_poll(sysmsg, (const struct poll_args *)uap));
1690 seltrue(cdev_t dev, int events)
1692 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));