2 * Copyright (c) 1982, 1986, 1989, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
5 * sendfile(2) and related extensions:
6 * Copyright (c) 1998, David Greenman. All rights reserved.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. 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.
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
36 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94
37 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $
40 #include "opt_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/sysproto.h>
47 #include <sys/malloc.h>
48 #include <sys/filedesc.h>
49 #include <sys/event.h>
51 #include <sys/fcntl.h>
53 #include <sys/filio.h>
54 #include <sys/kern_syscall.h>
56 #include <sys/protosw.h>
57 #include <sys/sfbuf.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/socketops.h>
62 #include <sys/vnode.h>
64 #include <sys/mount.h>
66 #include <sys/ktrace.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_pageout.h>
72 #include <vm/vm_kern.h>
73 #include <vm/vm_extern.h>
74 #include <sys/file2.h>
75 #include <sys/signalvar.h>
76 #include <sys/serialize.h>
78 #include <sys/thread2.h>
79 #include <sys/msgport2.h>
80 #include <sys/socketvar2.h>
81 #include <net/netmsg2.h>
84 #include <netinet/sctp_peeloff.h>
87 extern int use_soaccept_pred_fast;
88 extern int use_sendfile_async;
91 * System call interface to the socket abstraction.
94 extern struct fileops socketops;
97 * socket_args(int domain, int type, int protocol)
100 kern_socket(int domain, int type, int protocol, int *res)
102 struct thread *td = curthread;
103 struct filedesc *fdp = td->td_proc->p_fd;
108 KKASSERT(td->td_lwp);
110 error = falloc(td->td_lwp, &fp, &fd);
113 error = socreate(domain, &so, type, protocol, td);
115 fsetfd(fdp, NULL, fd);
117 fp->f_type = DTYPE_SOCKET;
118 fp->f_flag = FREAD | FWRITE;
119 fp->f_ops = &socketops;
132 sys_socket(struct socket_args *uap)
136 error = kern_socket(uap->domain, uap->type, uap->protocol,
137 &uap->sysmsg_iresult);
143 kern_bind(int s, struct sockaddr *sa)
145 struct thread *td = curthread;
146 struct proc *p = td->td_proc;
151 error = holdsock(p->p_fd, s, &fp);
154 error = sobind((struct socket *)fp->f_data, sa, td);
160 * bind_args(int s, caddr_t name, int namelen)
165 sys_bind(struct bind_args *uap)
170 error = getsockaddr(&sa, uap->name, uap->namelen);
173 error = kern_bind(uap->s, sa);
180 kern_listen(int s, int backlog)
182 struct thread *td = curthread;
183 struct proc *p = td->td_proc;
188 error = holdsock(p->p_fd, s, &fp);
191 error = solisten((struct socket *)fp->f_data, backlog, td);
197 * listen_args(int s, int backlog)
202 sys_listen(struct listen_args *uap)
206 error = kern_listen(uap->s, uap->backlog);
211 * Returns the accepted socket as well.
213 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the
214 * pool token is absolutely required to avoid a sofree() race,
215 * as well as to avoid tailq handling races.
218 soaccept_predicate(struct netmsg_so_notify *msg)
220 struct socket *head = msg->base.nm_so;
223 if (head->so_error != 0) {
224 msg->base.lmsg.ms_error = head->so_error;
227 lwkt_getpooltoken(head);
228 if (!TAILQ_EMPTY(&head->so_comp)) {
229 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
230 so = TAILQ_FIRST(&head->so_comp);
231 TAILQ_REMOVE(&head->so_comp, so, so_list);
233 soclrstate(so, SS_COMP);
237 lwkt_relpooltoken(head);
239 msg->base.lmsg.ms_error = 0;
240 msg->base.nm_so = so;
243 lwkt_relpooltoken(head);
244 if (head->so_state & SS_CANTRCVMORE) {
245 msg->base.lmsg.ms_error = ECONNABORTED;
248 if (msg->nm_fflags & FNONBLOCK) {
249 msg->base.lmsg.ms_error = EWOULDBLOCK;
257 * The second argument to kern_accept() is a handle to a struct sockaddr.
258 * This allows kern_accept() to return a pointer to an allocated struct
259 * sockaddr which must be freed later with FREE(). The caller must
260 * initialize *name to NULL.
263 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res)
265 struct thread *td = curthread;
266 struct filedesc *fdp = td->td_proc->p_fd;
267 struct file *lfp = NULL;
268 struct file *nfp = NULL;
270 struct socket *head, *so;
271 struct netmsg_so_notify msg;
273 u_int fflag; /* type must match fp->f_flag */
277 if (name && namelen && *namelen < 0)
280 error = holdsock(td->td_proc->p_fd, s, &lfp);
284 error = falloc(td->td_lwp, &nfp, &fd);
285 if (error) { /* Probably ran out of file descriptors. */
289 head = (struct socket *)lfp->f_data;
290 if ((head->so_options & SO_ACCEPTCONN) == 0) {
295 if (fflags & O_FBLOCKING)
296 fflags |= lfp->f_flag & ~FNONBLOCK;
297 else if (fflags & O_FNONBLOCKING)
298 fflags |= lfp->f_flag | FNONBLOCK;
300 fflags = lfp->f_flag;
302 if (use_soaccept_pred_fast) {
305 /* Initialize necessary parts for soaccept_predicate() */
306 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL);
307 msg.nm_fflags = fflags;
309 lwkt_getpooltoken(head);
310 pred = soaccept_predicate(&msg);
311 lwkt_relpooltoken(head);
314 error = msg.base.lmsg.ms_error;
322 /* optimize for uniprocessor case later XXX JH */
323 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport,
324 0, netmsg_so_notify, netmsg_so_notify_doabort);
325 msg.nm_predicate = soaccept_predicate;
326 msg.nm_fflags = fflags;
327 msg.nm_etype = NM_REVENT;
328 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH);
334 * At this point we have the connection that's ready to be accepted.
336 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects
337 * to eat the ref and turn it into a descriptor.
343 /* connection has been removed from the listen queue */
344 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0);
346 if (head->so_sigio != NULL)
347 fsetown(fgetown(&head->so_sigio), &so->so_sigio);
349 nfp->f_type = DTYPE_SOCKET;
351 nfp->f_ops = &socketops;
353 /* Sync socket nonblocking/async state with file flags */
354 tmp = fflag & FNONBLOCK;
355 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL);
356 tmp = fflag & FASYNC;
357 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
360 if (so->so_faddr != NULL) {
364 soaccept_generic(so);
367 error = soaccept(so, &sa);
371 * Set the returned name and namelen as applicable. Set the returned
372 * namelen to 0 for older code which might ignore the return value
376 if (sa && name && namelen) {
377 if (*namelen > sa->sa_len)
378 *namelen = sa->sa_len;
388 * If an error occured clear the reserved descriptor, else associate
391 * Note that *res is normally ignored if an error is returned but
392 * a syscall message will still have access to the result code.
395 fsetfd(fdp, NULL, fd);
398 fsetfd(fdp, nfp, fd);
406 * accept(int s, caddr_t name, int *anamelen)
411 sys_accept(struct accept_args *uap)
413 struct sockaddr *sa = NULL;
418 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
422 error = kern_accept(uap->s, 0, &sa, &sa_len,
423 &uap->sysmsg_iresult);
426 error = copyout(sa, uap->name, sa_len);
428 error = copyout(&sa_len, uap->anamelen,
429 sizeof(*uap->anamelen));
434 error = kern_accept(uap->s, 0, NULL, 0,
435 &uap->sysmsg_iresult);
441 * extaccept(int s, int fflags, caddr_t name, int *anamelen)
446 sys_extaccept(struct extaccept_args *uap)
448 struct sockaddr *sa = NULL;
451 int fflags = uap->flags & O_FMASK;
454 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
458 error = kern_accept(uap->s, fflags, &sa, &sa_len,
459 &uap->sysmsg_iresult);
462 error = copyout(sa, uap->name, sa_len);
464 error = copyout(&sa_len, uap->anamelen,
465 sizeof(*uap->anamelen));
470 error = kern_accept(uap->s, fflags, NULL, 0,
471 &uap->sysmsg_iresult);
478 * Returns TRUE if predicate satisfied.
481 soconnected_predicate(struct netmsg_so_notify *msg)
483 struct socket *so = msg->base.nm_so;
485 /* check predicate */
486 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
487 msg->base.lmsg.ms_error = so->so_error;
495 kern_connect(int s, int fflags, struct sockaddr *sa)
497 struct thread *td = curthread;
498 struct proc *p = td->td_proc;
501 int error, interrupted = 0;
503 error = holdsock(p->p_fd, s, &fp);
506 so = (struct socket *)fp->f_data;
508 if (fflags & O_FBLOCKING)
509 /* fflags &= ~FNONBLOCK; */;
510 else if (fflags & O_FNONBLOCKING)
515 if (so->so_state & SS_ISCONNECTING) {
519 error = soconnect(so, sa, td);
522 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
526 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
527 struct netmsg_so_notify msg;
529 netmsg_init_abortable(&msg.base, so,
530 &curthread->td_msgport,
533 netmsg_so_notify_doabort);
534 msg.nm_predicate = soconnected_predicate;
535 msg.nm_etype = NM_REVENT;
536 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH);
537 if (error == EINTR || error == ERESTART)
541 error = so->so_error;
546 soclrstate(so, SS_ISCONNECTING);
547 if (error == ERESTART)
555 * connect_args(int s, caddr_t name, int namelen)
560 sys_connect(struct connect_args *uap)
565 error = getsockaddr(&sa, uap->name, uap->namelen);
568 error = kern_connect(uap->s, 0, sa);
575 * connect_args(int s, int fflags, caddr_t name, int namelen)
580 sys_extconnect(struct extconnect_args *uap)
584 int fflags = uap->flags & O_FMASK;
586 error = getsockaddr(&sa, uap->name, uap->namelen);
589 error = kern_connect(uap->s, fflags, sa);
596 kern_socketpair(int domain, int type, int protocol, int *sv)
598 struct thread *td = curthread;
599 struct filedesc *fdp;
600 struct file *fp1, *fp2;
601 struct socket *so1, *so2;
604 fdp = td->td_proc->p_fd;
605 error = socreate(domain, &so1, type, protocol, td);
608 error = socreate(domain, &so2, type, protocol, td);
611 error = falloc(td->td_lwp, &fp1, &fd1);
616 error = falloc(td->td_lwp, &fp2, &fd2);
621 error = soconnect2(so1, so2);
624 if (type == SOCK_DGRAM) {
626 * Datagram socket connection is asymmetric.
628 error = soconnect2(so2, so1);
632 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
633 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
634 fp1->f_ops = fp2->f_ops = &socketops;
635 fsetfd(fdp, fp1, fd1);
636 fsetfd(fdp, fp2, fd2);
641 fsetfd(fdp, NULL, fd2);
644 fsetfd(fdp, NULL, fd1);
647 (void)soclose(so2, 0);
649 (void)soclose(so1, 0);
654 * socketpair(int domain, int type, int protocol, int *rsv)
657 sys_socketpair(struct socketpair_args *uap)
661 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
664 error = copyout(sockv, uap->rsv, sizeof(sockv));
667 kern_close(sockv[0]);
668 kern_close(sockv[1]);
676 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
677 struct mbuf *control, int flags, size_t *res)
679 struct thread *td = curthread;
680 struct lwp *lp = td->td_lwp;
681 struct proc *p = td->td_proc;
687 struct iovec *ktriov = NULL;
691 error = holdsock(p->p_fd, s, &fp);
695 if (KTRPOINT(td, KTR_GENIO)) {
696 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
698 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
699 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
703 len = auio->uio_resid;
704 so = (struct socket *)fp->f_data;
705 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
706 if (fp->f_flag & FNONBLOCK)
707 flags |= MSG_FNONBLOCKING;
709 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
711 if (auio->uio_resid != len && (error == ERESTART ||
712 error == EINTR || error == EWOULDBLOCK))
714 if (error == EPIPE && !(flags & MSG_NOSIGNAL) &&
715 !(so->so_options & SO_NOSIGPIPE))
716 lwpsignal(p, lp, SIGPIPE);
719 if (ktriov != NULL) {
721 ktruio.uio_iov = ktriov;
722 ktruio.uio_resid = len - auio->uio_resid;
723 ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
725 kfree(ktriov, M_TEMP);
729 *res = len - auio->uio_resid;
735 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
740 sys_sendto(struct sendto_args *uap)
742 struct thread *td = curthread;
745 struct sockaddr *sa = NULL;
749 error = getsockaddr(&sa, uap->to, uap->tolen);
753 aiov.iov_base = uap->buf;
754 aiov.iov_len = uap->len;
755 auio.uio_iov = &aiov;
758 auio.uio_resid = uap->len;
759 auio.uio_segflg = UIO_USERSPACE;
760 auio.uio_rw = UIO_WRITE;
763 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
764 &uap->sysmsg_szresult);
772 * sendmsg_args(int s, caddr_t msg, int flags)
777 sys_sendmsg(struct sendmsg_args *uap)
779 struct thread *td = curthread;
782 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
783 struct sockaddr *sa = NULL;
784 struct mbuf *control = NULL;
787 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
792 * Conditionally copyin msg.msg_name.
795 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
803 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
808 auio.uio_iovcnt = msg.msg_iovlen;
810 auio.uio_segflg = UIO_USERSPACE;
811 auio.uio_rw = UIO_WRITE;
815 * Conditionally copyin msg.msg_control.
817 if (msg.msg_control) {
818 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
819 msg.msg_controllen > MLEN) {
823 control = m_get(MB_WAIT, MT_CONTROL);
824 if (control == NULL) {
828 control->m_len = msg.msg_controllen;
829 error = copyin(msg.msg_control, mtod(control, caddr_t),
837 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
838 &uap->sysmsg_szresult);
841 iovec_free(&iov, aiov);
849 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
850 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
851 * Don't forget to FREE() and m_free() these if they are returned.
854 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
855 struct mbuf **control, int *flags, size_t *res)
857 struct thread *td = curthread;
858 struct proc *p = td->td_proc;
865 struct iovec *ktriov = NULL;
869 error = holdsock(p->p_fd, s, &fp);
873 if (KTRPOINT(td, KTR_GENIO)) {
874 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
876 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
877 bcopy(auio->uio_iov, ktriov, iovlen);
881 len = auio->uio_resid;
882 so = (struct socket *)fp->f_data;
884 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
885 if (fp->f_flag & FNONBLOCK) {
887 *flags |= MSG_FNONBLOCKING;
889 lflags = MSG_FNONBLOCKING;
895 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
897 if (auio->uio_resid != len && (error == ERESTART ||
898 error == EINTR || error == EWOULDBLOCK))
902 if (ktriov != NULL) {
904 ktruio.uio_iov = ktriov;
905 ktruio.uio_resid = len - auio->uio_resid;
906 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
908 kfree(ktriov, M_TEMP);
912 *res = len - auio->uio_resid;
918 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
919 * caddr_t from, int *fromlenaddr)
924 sys_recvfrom(struct recvfrom_args *uap)
926 struct thread *td = curthread;
929 struct sockaddr *sa = NULL;
932 if (uap->from && uap->fromlenaddr) {
933 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
941 aiov.iov_base = uap->buf;
942 aiov.iov_len = uap->len;
943 auio.uio_iov = &aiov;
946 auio.uio_resid = uap->len;
947 auio.uio_segflg = UIO_USERSPACE;
948 auio.uio_rw = UIO_READ;
951 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
952 &uap->flags, &uap->sysmsg_szresult);
954 if (error == 0 && uap->from) {
955 /* note: sa may still be NULL */
957 fromlen = MIN(fromlen, sa->sa_len);
958 error = copyout(sa, uap->from, fromlen);
963 error = copyout(&fromlen, uap->fromlenaddr,
974 * recvmsg_args(int s, struct msghdr *msg, int flags)
979 sys_recvmsg(struct recvmsg_args *uap)
981 struct thread *td = curthread;
984 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
985 struct mbuf *m, *control = NULL;
986 struct sockaddr *sa = NULL;
988 socklen_t *ufromlenp, *ucontrollenp;
989 int error, fromlen, controllen, len, flags, *uflagsp;
992 * This copyin handles everything except the iovec.
994 error = copyin(uap->msg, &msg, sizeof(msg));
998 if (msg.msg_name && msg.msg_namelen < 0)
1000 if (msg.msg_control && msg.msg_controllen < 0)
1003 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1005 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1007 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
1013 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1018 auio.uio_iovcnt = msg.msg_iovlen;
1019 auio.uio_offset = 0;
1020 auio.uio_segflg = UIO_USERSPACE;
1021 auio.uio_rw = UIO_READ;
1026 error = kern_recvmsg(uap->s,
1027 (msg.msg_name ? &sa : NULL), &auio,
1028 (msg.msg_control ? &control : NULL), &flags,
1029 &uap->sysmsg_szresult);
1032 * Conditionally copyout the name and populate the namelen field.
1034 if (error == 0 && msg.msg_name) {
1035 /* note: sa may still be NULL */
1037 fromlen = MIN(msg.msg_namelen, sa->sa_len);
1038 error = copyout(sa, msg.msg_name, fromlen);
1043 error = copyout(&fromlen, ufromlenp,
1044 sizeof(*ufromlenp));
1048 * Copyout msg.msg_control and msg.msg_controllen.
1050 if (error == 0 && msg.msg_control) {
1051 len = msg.msg_controllen;
1053 ctlbuf = (caddr_t)msg.msg_control;
1055 while(m && len > 0) {
1056 unsigned int tocopy;
1058 if (len >= m->m_len) {
1061 msg.msg_flags |= MSG_CTRUNC;
1065 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1073 controllen = ctlbuf - (caddr_t)msg.msg_control;
1074 error = copyout(&controllen, ucontrollenp,
1075 sizeof(*ucontrollenp));
1079 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1083 kfree(sa, M_SONAME);
1084 iovec_free(&iov, aiov);
1091 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1092 * in kernel pointer instead of a userland pointer. This allows us
1093 * to manipulate socket options in the emulation code.
1096 kern_setsockopt(int s, struct sockopt *sopt)
1098 struct thread *td = curthread;
1099 struct proc *p = td->td_proc;
1103 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1105 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1107 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1110 error = holdsock(p->p_fd, s, &fp);
1114 error = sosetopt((struct socket *)fp->f_data, sopt);
1120 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1125 sys_setsockopt(struct setsockopt_args *uap)
1127 struct thread *td = curthread;
1128 struct sockopt sopt;
1131 sopt.sopt_level = uap->level;
1132 sopt.sopt_name = uap->name;
1133 sopt.sopt_valsize = uap->valsize;
1135 sopt.sopt_val = NULL;
1137 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1140 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1141 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1146 error = kern_setsockopt(uap->s, &sopt);
1149 kfree(sopt.sopt_val, M_TEMP);
1154 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1155 * in kernel pointer instead of a userland pointer. This allows us
1156 * to manipulate socket options in the emulation code.
1159 kern_getsockopt(int s, struct sockopt *sopt)
1161 struct thread *td = curthread;
1162 struct proc *p = td->td_proc;
1166 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1168 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1170 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1173 error = holdsock(p->p_fd, s, &fp);
1177 error = sogetopt((struct socket *)fp->f_data, sopt);
1183 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1188 sys_getsockopt(struct getsockopt_args *uap)
1190 struct thread *td = curthread;
1191 struct sockopt sopt;
1195 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1202 sopt.sopt_level = uap->level;
1203 sopt.sopt_name = uap->name;
1204 sopt.sopt_valsize = valsize;
1206 sopt.sopt_val = NULL;
1208 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1211 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1212 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1217 error = kern_getsockopt(uap->s, &sopt);
1220 valsize = sopt.sopt_valsize;
1221 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1225 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1228 kfree(sopt.sopt_val, M_TEMP);
1233 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1234 * This allows kern_getsockname() to return a pointer to an allocated struct
1235 * sockaddr which must be freed later with FREE(). The caller must
1236 * initialize *name to NULL.
1239 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1241 struct thread *td = curthread;
1242 struct proc *p = td->td_proc;
1245 struct sockaddr *sa = NULL;
1248 error = holdsock(p->p_fd, s, &fp);
1255 so = (struct socket *)fp->f_data;
1256 error = so_pru_sockaddr(so, &sa);
1261 *namelen = MIN(*namelen, sa->sa_len);
1271 * getsockname_args(int fdes, caddr_t asa, int *alen)
1278 sys_getsockname(struct getsockname_args *uap)
1280 struct sockaddr *sa = NULL;
1283 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1287 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1290 error = copyout(sa, uap->asa, sa_len);
1292 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1294 kfree(sa, M_SONAME);
1299 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1300 * This allows kern_getpeername() to return a pointer to an allocated struct
1301 * sockaddr which must be freed later with FREE(). The caller must
1302 * initialize *name to NULL.
1305 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1307 struct thread *td = curthread;
1308 struct proc *p = td->td_proc;
1311 struct sockaddr *sa = NULL;
1314 error = holdsock(p->p_fd, s, &fp);
1321 so = (struct socket *)fp->f_data;
1322 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1326 error = so_pru_peeraddr(so, &sa);
1331 *namelen = MIN(*namelen, sa->sa_len);
1341 * getpeername_args(int fdes, caddr_t asa, int *alen)
1343 * Get name of peer for connected socket.
1348 sys_getpeername(struct getpeername_args *uap)
1350 struct sockaddr *sa = NULL;
1353 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1357 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1360 error = copyout(sa, uap->asa, sa_len);
1362 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1364 kfree(sa, M_SONAME);
1369 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1371 struct sockaddr *sa;
1375 if (len > SOCK_MAXADDRLEN)
1376 return ENAMETOOLONG;
1377 if (len < offsetof(struct sockaddr, sa_data[0]))
1379 sa = kmalloc(len, M_SONAME, M_WAITOK);
1380 error = copyin(uaddr, sa, len);
1382 kfree(sa, M_SONAME);
1384 #if BYTE_ORDER != BIG_ENDIAN
1386 * The bind(), connect(), and sendto() syscalls were not
1387 * versioned for COMPAT_43. Thus, this check must stay.
1389 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1390 sa->sa_family = sa->sa_len;
1399 * Detach a mapped page and release resources back to the system.
1400 * We must release our wiring and if the object is ripped out
1401 * from under the vm_page we become responsible for freeing the
1407 sf_buf_mfree(void *arg)
1409 struct sf_buf *sf = arg;
1412 m = sf_buf_page(sf);
1413 if (sf_buf_free(sf)) {
1414 /* sf invalid now */
1415 vm_page_busy_wait(m, FALSE, "sockpgf");
1416 vm_page_unwire(m, 0);
1417 if (m->object == NULL &&
1418 m->wire_count == 0 &&
1419 (m->flags & PG_NEED_COMMIT) == 0) {
1429 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1430 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1432 * Send a file specified by 'fd' and starting at 'offset' to a socket
1433 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1434 * nbytes == 0. Optionally add a header and/or trailer to the socket
1435 * output. If specified, write the total number of bytes sent into *sbytes.
1437 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1438 * the headers to count against the remaining bytes to be sent from
1439 * the file descriptor. We may wish to implement a compatibility syscall
1445 sys_sendfile(struct sendfile_args *uap)
1447 struct thread *td = curthread;
1448 struct proc *p = td->td_proc;
1450 struct vnode *vp = NULL;
1451 struct sf_hdtr hdtr;
1452 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1454 struct mbuf *mheader = NULL;
1457 off_t hdtr_size = 0;
1464 * Do argument checking. Must be a regular file in, stream
1465 * type and connected socket out, positive offset.
1467 fp = holdfp(p->p_fd, uap->fd, FREAD);
1471 if (fp->f_type != DTYPE_VNODE) {
1475 vp = (struct vnode *)fp->f_data;
1480 * If specified, get the pointer to the sf_hdtr struct for
1481 * any headers/trailers.
1484 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1491 error = iovec_copyin(hdtr.headers, &iov, aiov,
1492 hdtr.hdr_cnt, &hbytes);
1496 auio.uio_iovcnt = hdtr.hdr_cnt;
1497 auio.uio_offset = 0;
1498 auio.uio_segflg = UIO_USERSPACE;
1499 auio.uio_rw = UIO_WRITE;
1501 auio.uio_resid = hbytes;
1503 mheader = m_uiomove(&auio);
1505 iovec_free(&iov, aiov);
1506 if (mheader == NULL)
1511 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1512 &sbytes, uap->flags);
1517 * Send trailers. Wimp out and use writev(2).
1519 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1520 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1521 hdtr.trl_cnt, &auio.uio_resid);
1525 auio.uio_iovcnt = hdtr.trl_cnt;
1526 auio.uio_offset = 0;
1527 auio.uio_segflg = UIO_USERSPACE;
1528 auio.uio_rw = UIO_WRITE;
1531 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1533 iovec_free(&iov, aiov);
1536 hdtr_size += tbytes; /* trailer bytes successfully sent */
1542 if (uap->sbytes != NULL) {
1543 sbytes += hdtr_size;
1544 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1550 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1551 struct mbuf *mheader, off_t *sbytes, int flags)
1553 struct thread *td = curthread;
1554 struct proc *p = td->td_proc;
1555 struct vm_object *obj;
1558 struct mbuf *m, *mp;
1565 if (vp->v_type != VREG) {
1569 if ((obj = vp->v_object) == NULL) {
1573 error = holdsock(p->p_fd, sfd, &fp);
1576 so = (struct socket *)fp->f_data;
1577 if (so->so_type != SOCK_STREAM) {
1581 if ((so->so_state & SS_ISCONNECTED) == 0) {
1592 * Protect against multiple writers to the socket.
1594 ssb_lock(&so->so_snd, M_WAITOK);
1597 * Loop through the pages in the file, starting with the requested
1598 * offset. Get a file page (do I/O if necessary), map the file page
1599 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1602 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1607 pindex = OFF_TO_IDX(off);
1610 * Calculate the amount to transfer. Not to exceed a page,
1611 * the EOF, or the passed in nbytes.
1613 xfsize = vp->v_filesize - off;
1614 if (xfsize > PAGE_SIZE)
1616 pgoff = (vm_offset_t)(off & PAGE_MASK);
1617 if (PAGE_SIZE - pgoff < xfsize)
1618 xfsize = PAGE_SIZE - pgoff;
1619 if (nbytes && xfsize > (nbytes - *sbytes))
1620 xfsize = nbytes - *sbytes;
1624 * Optimize the non-blocking case by looking at the socket space
1625 * before going to the extra work of constituting the sf_buf.
1627 if ((fp->f_flag & FNONBLOCK) &&
1628 ssb_space_prealloc(&so->so_snd) <= 0) {
1629 if (so->so_state & SS_CANTSENDMORE)
1633 ssb_unlock(&so->so_snd);
1637 * Attempt to look up the page.
1639 * Allocate if not found, wait and loop if busy, then
1640 * wire the page. critical section protection is
1641 * required to maintain the object association (an
1642 * interrupt can free the page) through to the
1643 * vm_page_wire() call.
1645 vm_object_hold(obj);
1646 pg = vm_page_lookup_busy_try(obj, pindex, TRUE, &error);
1648 vm_page_sleep_busy(pg, TRUE, "sfpbsy");
1649 vm_object_drop(obj);
1653 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL |
1657 vm_object_drop(obj);
1662 vm_object_drop(obj);
1665 * If page is not valid for what we need, initiate I/O
1668 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1674 * Ensure that our page is still around when the I/O
1677 vm_page_io_start(pg);
1681 * Get the page from backing store.
1683 bsize = vp->v_mount->mnt_stat.f_iosize;
1684 auio.uio_iov = &aiov;
1685 auio.uio_iovcnt = 1;
1687 aiov.iov_len = MAXBSIZE;
1688 auio.uio_resid = MAXBSIZE;
1689 auio.uio_offset = trunc_page(off);
1690 auio.uio_segflg = UIO_NOCOPY;
1691 auio.uio_rw = UIO_READ;
1693 vn_lock(vp, LK_SHARED | LK_RETRY);
1694 error = VOP_READ(vp, &auio,
1695 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1698 vm_page_flag_clear(pg, PG_ZERO);
1699 vm_page_busy_wait(pg, FALSE, "sockpg");
1700 vm_page_io_finish(pg);
1702 vm_page_unwire(pg, 0);
1704 vm_page_try_to_free(pg);
1705 ssb_unlock(&so->so_snd);
1712 * Get a sendfile buf. We usually wait as long as necessary,
1713 * but this wait can be interrupted.
1715 if ((sf = sf_buf_alloc(pg)) == NULL) {
1716 vm_page_unwire(pg, 0);
1718 vm_page_try_to_free(pg);
1719 ssb_unlock(&so->so_snd);
1726 * Get an mbuf header and set it up as having external storage.
1728 MGETHDR(m, MB_WAIT, MT_DATA);
1732 ssb_unlock(&so->so_snd);
1736 m->m_ext.ext_free = sf_buf_mfree;
1737 m->m_ext.ext_ref = sf_buf_ref;
1738 m->m_ext.ext_arg = sf;
1739 m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1740 m->m_ext.ext_size = PAGE_SIZE;
1741 m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1742 m->m_flags |= M_EXT;
1743 m->m_pkthdr.len = m->m_len = xfsize;
1744 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1746 if (mheader != NULL) {
1747 hbytes = mheader->m_pkthdr.len;
1748 mheader->m_pkthdr.len += m->m_pkthdr.len;
1756 * Add the buffer to the socket buffer chain.
1761 * Make sure that the socket is still able to take more data.
1762 * CANTSENDMORE being true usually means that the connection
1763 * was closed. so_error is true when an error was sensed after
1765 * The state is checked after the page mapping and buffer
1766 * allocation above since those operations may block and make
1767 * any socket checks stale. From this point forward, nothing
1768 * blocks before the pru_send (or more accurately, any blocking
1769 * results in a loop back to here to re-check).
1771 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1772 if (so->so_state & SS_CANTSENDMORE) {
1775 error = so->so_error;
1779 ssb_unlock(&so->so_snd);
1784 * Wait for socket space to become available. We do this just
1785 * after checking the connection state above in order to avoid
1786 * a race condition with ssb_wait().
1788 space = ssb_space_prealloc(&so->so_snd);
1789 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) {
1790 if (fp->f_flag & FNONBLOCK) {
1792 ssb_unlock(&so->so_snd);
1797 error = ssb_wait(&so->so_snd);
1799 * An error from ssb_wait usually indicates that we've
1800 * been interrupted by a signal. If we've sent anything
1801 * then return bytes sent, otherwise return the error.
1805 ssb_unlock(&so->so_snd);
1812 for (mp = m; mp != NULL; mp = mp->m_next)
1813 ssb_preallocstream(&so->so_snd, mp);
1814 if (use_sendfile_async)
1815 error = so_pru_senda(so, 0, m, NULL, NULL, td);
1817 error = so_pru_send(so, 0, m, NULL, NULL, td);
1821 ssb_unlock(&so->so_snd);
1825 if (mheader != NULL) {
1826 *sbytes += mheader->m_pkthdr.len;
1828 for (mp = mheader; mp != NULL; mp = mp->m_next)
1829 ssb_preallocstream(&so->so_snd, mp);
1830 if (use_sendfile_async)
1831 error = so_pru_senda(so, 0, mheader, NULL, NULL, td);
1833 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1837 ssb_unlock(&so->so_snd);
1842 if (mheader != NULL)
1851 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1854 struct thread *td = curthread;
1855 struct filedesc *fdp = td->td_proc->p_fd;
1856 struct file *lfp = NULL;
1857 struct file *nfp = NULL;
1859 struct socket *head, *so;
1862 short fflag; /* type must match fp->f_flag */
1864 assoc_id = uap->name;
1865 error = holdsock(td->td_proc->p_fd, uap->sd, &lfp);
1870 head = (struct socket *)lfp->f_data;
1871 error = sctp_can_peel_off(head, assoc_id);
1877 * At this point we know we do have a assoc to pull
1878 * we proceed to get the fd setup. This may block
1882 fflag = lfp->f_flag;
1883 error = falloc(td->td_lwp, &nfp, &fd);
1886 * Probably ran out of file descriptors. Put the
1887 * unaccepted connection back onto the queue and
1888 * do another wakeup so some other process might
1889 * have a chance at it.
1894 uap->sysmsg_iresult = fd;
1896 so = sctp_get_peeloff(head, assoc_id, &error);
1899 * Either someone else peeled it off OR
1900 * we can't get a socket.
1904 soreference(so); /* reference needed */
1905 soclrstate(so, SS_NOFDREF | SS_COMP); /* when clearing NOFDREF */
1907 if (head->so_sigio != NULL)
1908 fsetown(fgetown(&head->so_sigio), &so->so_sigio);
1910 nfp->f_type = DTYPE_SOCKET;
1911 nfp->f_flag = fflag;
1912 nfp->f_ops = &socketops;
1917 * Assign the file pointer to the reserved descriptor, or clear
1918 * the reserved descriptor if an error occured.
1921 fsetfd(fdp, NULL, fd);
1923 fsetfd(fdp, nfp, fd);
1926 * Release explicitly held references before returning.