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 $
38 * $DragonFly: src/sys/kern/uipc_syscalls.c,v 1.92 2008/11/26 13:10:56 sephe Exp $
41 #include "opt_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysproto.h>
48 #include <sys/malloc.h>
49 #include <sys/filedesc.h>
50 #include <sys/event.h>
52 #include <sys/fcntl.h>
54 #include <sys/filio.h>
55 #include <sys/kern_syscall.h>
57 #include <sys/protosw.h>
58 #include <sys/sfbuf.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/socketops.h>
63 #include <sys/vnode.h>
65 #include <sys/mount.h>
67 #include <sys/ktrace.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_pageout.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_extern.h>
75 #include <sys/file2.h>
76 #include <sys/signalvar.h>
77 #include <sys/serialize.h>
79 #include <sys/thread2.h>
80 #include <sys/msgport2.h>
81 #include <sys/socketvar2.h>
82 #include <sys/mplock2.h>
83 #include <net/netmsg2.h>
86 #include <netinet/sctp_peeloff.h>
94 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures");
97 * System call interface to the socket abstraction.
100 extern struct fileops socketops;
103 * socket_args(int domain, int type, int protocol)
106 kern_socket(int domain, int type, int protocol, int *res)
108 struct thread *td = curthread;
109 struct filedesc *fdp = td->td_proc->p_fd;
114 KKASSERT(td->td_lwp);
116 error = falloc(td->td_lwp, &fp, &fd);
119 error = socreate(domain, &so, type, protocol, td);
121 fsetfd(fdp, NULL, fd);
123 fp->f_type = DTYPE_SOCKET;
124 fp->f_flag = FREAD | FWRITE;
125 fp->f_ops = &socketops;
138 sys_socket(struct socket_args *uap)
143 error = kern_socket(uap->domain, uap->type, uap->protocol,
144 &uap->sysmsg_iresult);
151 kern_bind(int s, struct sockaddr *sa)
153 struct thread *td = curthread;
154 struct proc *p = td->td_proc;
159 error = holdsock(p->p_fd, s, &fp);
162 error = sobind((struct socket *)fp->f_data, sa, td);
168 * bind_args(int s, caddr_t name, int namelen)
173 sys_bind(struct bind_args *uap)
178 error = getsockaddr(&sa, uap->name, uap->namelen);
182 error = kern_bind(uap->s, sa);
190 kern_listen(int s, int backlog)
192 struct thread *td = curthread;
193 struct proc *p = td->td_proc;
198 error = holdsock(p->p_fd, s, &fp);
201 error = solisten((struct socket *)fp->f_data, backlog, td);
207 * listen_args(int s, int backlog)
212 sys_listen(struct listen_args *uap)
217 error = kern_listen(uap->s, uap->backlog);
223 * Returns the accepted socket as well.
226 soaccept_predicate(struct netmsg *msg0)
228 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
229 struct socket *head = msg->nm_so;
231 if (head->so_error != 0) {
232 msg->nm_netmsg.nm_lmsg.ms_error = head->so_error;
235 if (!TAILQ_EMPTY(&head->so_comp)) {
236 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
237 msg->nm_so = TAILQ_FIRST(&head->so_comp);
238 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
241 msg->nm_netmsg.nm_lmsg.ms_error = 0;
244 if (head->so_state & SS_CANTRCVMORE) {
245 msg->nm_netmsg.nm_lmsg.ms_error = ECONNABORTED;
248 if (msg->nm_fflags & FNONBLOCK) {
249 msg->nm_netmsg.nm_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 /* optimize for uniprocessor case later XXX JH */
303 netmsg_init_abortable(&msg.nm_netmsg, head, &curthread->td_msgport,
304 0, netmsg_so_notify, netmsg_so_notify_doabort);
305 msg.nm_predicate = soaccept_predicate;
306 msg.nm_fflags = fflags;
308 msg.nm_etype = NM_REVENT;
309 error = lwkt_domsg(head->so_port, &msg.nm_netmsg.nm_lmsg, PCATCH);
314 * At this point we have the connection that's ready to be accepted.
320 /* connection has been removed from the listen queue */
321 KNOTE(&head->so_rcv.ssb_sel.si_note, 0);
323 so->so_state &= ~SS_COMP;
325 if (head->so_sigio != NULL)
326 fsetown(fgetown(head->so_sigio), &so->so_sigio);
328 nfp->f_type = DTYPE_SOCKET;
330 nfp->f_ops = &socketops;
332 /* Sync socket nonblocking/async state with file flags */
333 tmp = fflag & FNONBLOCK;
334 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL);
335 tmp = fflag & FASYNC;
336 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
339 error = soaccept(so, &sa);
342 * Set the returned name and namelen as applicable. Set the returned
343 * namelen to 0 for older code which might ignore the return value
347 if (sa && name && namelen) {
348 if (*namelen > sa->sa_len)
349 *namelen = sa->sa_len;
359 * If an error occured clear the reserved descriptor, else associate
362 * Note that *res is normally ignored if an error is returned but
363 * a syscall message will still have access to the result code.
366 fsetfd(fdp, NULL, fd);
369 fsetfd(fdp, nfp, fd);
377 * accept(int s, caddr_t name, int *anamelen)
382 sys_accept(struct accept_args *uap)
384 struct sockaddr *sa = NULL;
389 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
394 error = kern_accept(uap->s, 0, &sa, &sa_len,
395 &uap->sysmsg_iresult);
399 error = copyout(sa, uap->name, sa_len);
401 error = copyout(&sa_len, uap->anamelen,
402 sizeof(*uap->anamelen));
408 error = kern_accept(uap->s, 0, NULL, 0,
409 &uap->sysmsg_iresult);
416 * extaccept(int s, int fflags, caddr_t name, int *anamelen)
421 sys_extaccept(struct extaccept_args *uap)
423 struct sockaddr *sa = NULL;
426 int fflags = uap->flags & O_FMASK;
429 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
434 error = kern_accept(uap->s, fflags, &sa, &sa_len,
435 &uap->sysmsg_iresult);
439 error = copyout(sa, uap->name, sa_len);
441 error = copyout(&sa_len, uap->anamelen,
442 sizeof(*uap->anamelen));
448 error = kern_accept(uap->s, fflags, NULL, 0,
449 &uap->sysmsg_iresult);
457 * Returns TRUE if predicate satisfied.
460 soconnected_predicate(struct netmsg *msg0)
462 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
463 struct socket *so = msg->nm_so;
465 /* check predicate */
466 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
467 msg->nm_netmsg.nm_lmsg.ms_error = so->so_error;
475 kern_connect(int s, int fflags, struct sockaddr *sa)
477 struct thread *td = curthread;
478 struct proc *p = td->td_proc;
481 int error, interrupted = 0;
483 error = holdsock(p->p_fd, s, &fp);
486 so = (struct socket *)fp->f_data;
488 if (fflags & O_FBLOCKING)
489 /* fflags &= ~FNONBLOCK; */;
490 else if (fflags & O_FNONBLOCKING)
495 if (so->so_state & SS_ISCONNECTING) {
499 error = soconnect(so, sa, td);
502 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
506 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
507 struct netmsg_so_notify msg;
509 netmsg_init_abortable(&msg.nm_netmsg, so,
510 &curthread->td_msgport,
513 netmsg_so_notify_doabort);
514 msg.nm_predicate = soconnected_predicate;
516 msg.nm_etype = NM_REVENT;
517 error = lwkt_domsg(so->so_port, &msg.nm_netmsg.nm_lmsg, PCATCH);
518 if (error == EINTR || error == ERESTART)
522 error = so->so_error;
527 so->so_state &= ~SS_ISCONNECTING;
528 if (error == ERESTART)
536 * connect_args(int s, caddr_t name, int namelen)
541 sys_connect(struct connect_args *uap)
546 error = getsockaddr(&sa, uap->name, uap->namelen);
550 error = kern_connect(uap->s, 0, sa);
558 * connect_args(int s, int fflags, caddr_t name, int namelen)
563 sys_extconnect(struct extconnect_args *uap)
567 int fflags = uap->flags & O_FMASK;
569 error = getsockaddr(&sa, uap->name, uap->namelen);
573 error = kern_connect(uap->s, fflags, sa);
581 kern_socketpair(int domain, int type, int protocol, int *sv)
583 struct thread *td = curthread;
584 struct filedesc *fdp;
585 struct file *fp1, *fp2;
586 struct socket *so1, *so2;
589 fdp = td->td_proc->p_fd;
590 error = socreate(domain, &so1, type, protocol, td);
593 error = socreate(domain, &so2, type, protocol, td);
596 error = falloc(td->td_lwp, &fp1, &fd1);
601 error = falloc(td->td_lwp, &fp2, &fd2);
606 error = soconnect2(so1, so2);
609 if (type == SOCK_DGRAM) {
611 * Datagram socket connection is asymmetric.
613 error = soconnect2(so2, so1);
617 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
618 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
619 fp1->f_ops = fp2->f_ops = &socketops;
620 fsetfd(fdp, fp1, fd1);
621 fsetfd(fdp, fp2, fd2);
626 fsetfd(fdp, NULL, fd2);
629 fsetfd(fdp, NULL, fd1);
632 (void)soclose(so2, 0);
634 (void)soclose(so1, 0);
639 * socketpair(int domain, int type, int protocol, int *rsv)
644 sys_socketpair(struct socketpair_args *uap)
649 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
653 error = copyout(sockv, uap->rsv, sizeof(sockv));
658 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
659 struct mbuf *control, int flags, size_t *res)
661 struct thread *td = curthread;
662 struct lwp *lp = td->td_lwp;
663 struct proc *p = td->td_proc;
669 struct iovec *ktriov = NULL;
673 error = holdsock(p->p_fd, s, &fp);
677 if (KTRPOINT(td, KTR_GENIO)) {
678 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
680 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
681 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
685 len = auio->uio_resid;
686 so = (struct socket *)fp->f_data;
687 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
688 if (fp->f_flag & FNONBLOCK)
689 flags |= MSG_FNONBLOCKING;
691 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
693 if (auio->uio_resid != len && (error == ERESTART ||
694 error == EINTR || error == EWOULDBLOCK))
697 lwpsignal(p, lp, SIGPIPE);
700 if (ktriov != NULL) {
702 ktruio.uio_iov = ktriov;
703 ktruio.uio_resid = len - auio->uio_resid;
704 ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
706 FREE(ktriov, M_TEMP);
710 *res = len - auio->uio_resid;
716 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
721 sys_sendto(struct sendto_args *uap)
723 struct thread *td = curthread;
726 struct sockaddr *sa = NULL;
730 error = getsockaddr(&sa, uap->to, uap->tolen);
734 aiov.iov_base = uap->buf;
735 aiov.iov_len = uap->len;
736 auio.uio_iov = &aiov;
739 auio.uio_resid = uap->len;
740 auio.uio_segflg = UIO_USERSPACE;
741 auio.uio_rw = UIO_WRITE;
745 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
746 &uap->sysmsg_szresult);
755 * sendmsg_args(int s, caddr_t msg, int flags)
760 sys_sendmsg(struct sendmsg_args *uap)
762 struct thread *td = curthread;
765 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
766 struct sockaddr *sa = NULL;
767 struct mbuf *control = NULL;
770 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
775 * Conditionally copyin msg.msg_name.
778 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
786 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
791 auio.uio_iovcnt = msg.msg_iovlen;
793 auio.uio_segflg = UIO_USERSPACE;
794 auio.uio_rw = UIO_WRITE;
798 * Conditionally copyin msg.msg_control.
800 if (msg.msg_control) {
801 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
802 msg.msg_controllen > MLEN) {
806 control = m_get(MB_WAIT, MT_CONTROL);
807 if (control == NULL) {
811 control->m_len = msg.msg_controllen;
812 error = copyin(msg.msg_control, mtod(control, caddr_t),
821 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
822 &uap->sysmsg_szresult);
826 iovec_free(&iov, aiov);
834 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
835 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
836 * Don't forget to FREE() and m_free() these if they are returned.
839 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
840 struct mbuf **control, int *flags, size_t *res)
842 struct thread *td = curthread;
843 struct proc *p = td->td_proc;
850 struct iovec *ktriov = NULL;
854 error = holdsock(p->p_fd, s, &fp);
858 if (KTRPOINT(td, KTR_GENIO)) {
859 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
861 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
862 bcopy(auio->uio_iov, ktriov, iovlen);
866 len = auio->uio_resid;
867 so = (struct socket *)fp->f_data;
869 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
870 if (fp->f_flag & FNONBLOCK) {
872 *flags |= MSG_FNONBLOCKING;
874 lflags = MSG_FNONBLOCKING;
880 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
882 if (auio->uio_resid != len && (error == ERESTART ||
883 error == EINTR || error == EWOULDBLOCK))
887 if (ktriov != NULL) {
889 ktruio.uio_iov = ktriov;
890 ktruio.uio_resid = len - auio->uio_resid;
891 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
893 FREE(ktriov, M_TEMP);
897 *res = len - auio->uio_resid;
903 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
904 * caddr_t from, int *fromlenaddr)
909 sys_recvfrom(struct recvfrom_args *uap)
911 struct thread *td = curthread;
914 struct sockaddr *sa = NULL;
917 if (uap->from && uap->fromlenaddr) {
918 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
926 aiov.iov_base = uap->buf;
927 aiov.iov_len = uap->len;
928 auio.uio_iov = &aiov;
931 auio.uio_resid = uap->len;
932 auio.uio_segflg = UIO_USERSPACE;
933 auio.uio_rw = UIO_READ;
937 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
938 &uap->flags, &uap->sysmsg_szresult);
941 if (error == 0 && uap->from) {
942 /* note: sa may still be NULL */
944 fromlen = MIN(fromlen, sa->sa_len);
945 error = copyout(sa, uap->from, fromlen);
950 error = copyout(&fromlen, uap->fromlenaddr,
961 * recvmsg_args(int s, struct msghdr *msg, int flags)
966 sys_recvmsg(struct recvmsg_args *uap)
968 struct thread *td = curthread;
971 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
972 struct mbuf *m, *control = NULL;
973 struct sockaddr *sa = NULL;
975 socklen_t *ufromlenp, *ucontrollenp;
976 int error, fromlen, controllen, len, flags, *uflagsp;
979 * This copyin handles everything except the iovec.
981 error = copyin(uap->msg, &msg, sizeof(msg));
985 if (msg.msg_name && msg.msg_namelen < 0)
987 if (msg.msg_control && msg.msg_controllen < 0)
990 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
992 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
994 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
1000 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1005 auio.uio_iovcnt = msg.msg_iovlen;
1006 auio.uio_offset = 0;
1007 auio.uio_segflg = UIO_USERSPACE;
1008 auio.uio_rw = UIO_READ;
1014 error = kern_recvmsg(uap->s,
1015 (msg.msg_name ? &sa : NULL), &auio,
1016 (msg.msg_control ? &control : NULL), &flags,
1017 &uap->sysmsg_szresult);
1021 * Conditionally copyout the name and populate the namelen field.
1023 if (error == 0 && msg.msg_name) {
1024 /* note: sa may still be NULL */
1026 fromlen = MIN(msg.msg_namelen, sa->sa_len);
1027 error = copyout(sa, msg.msg_name, fromlen);
1032 error = copyout(&fromlen, ufromlenp,
1033 sizeof(*ufromlenp));
1037 * Copyout msg.msg_control and msg.msg_controllen.
1039 if (error == 0 && msg.msg_control) {
1040 len = msg.msg_controllen;
1042 ctlbuf = (caddr_t)msg.msg_control;
1044 while(m && len > 0) {
1045 unsigned int tocopy;
1047 if (len >= m->m_len) {
1050 msg.msg_flags |= MSG_CTRUNC;
1054 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1062 controllen = ctlbuf - (caddr_t)msg.msg_control;
1063 error = copyout(&controllen, ucontrollenp,
1064 sizeof(*ucontrollenp));
1068 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1073 iovec_free(&iov, aiov);
1080 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1081 * in kernel pointer instead of a userland pointer. This allows us
1082 * to manipulate socket options in the emulation code.
1085 kern_setsockopt(int s, struct sockopt *sopt)
1087 struct thread *td = curthread;
1088 struct proc *p = td->td_proc;
1092 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1094 if (sopt->sopt_valsize < 0)
1097 error = holdsock(p->p_fd, s, &fp);
1101 error = sosetopt((struct socket *)fp->f_data, sopt);
1107 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1112 sys_setsockopt(struct setsockopt_args *uap)
1114 struct thread *td = curthread;
1115 struct sockopt sopt;
1118 sopt.sopt_level = uap->level;
1119 sopt.sopt_name = uap->name;
1120 sopt.sopt_valsize = uap->valsize;
1122 sopt.sopt_val = NULL;
1124 if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1127 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1128 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1134 error = kern_setsockopt(uap->s, &sopt);
1138 kfree(sopt.sopt_val, M_TEMP);
1143 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1144 * in kernel pointer instead of a userland pointer. This allows us
1145 * to manipulate socket options in the emulation code.
1148 kern_getsockopt(int s, struct sockopt *sopt)
1150 struct thread *td = curthread;
1151 struct proc *p = td->td_proc;
1155 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1157 if (sopt->sopt_valsize < 0 || sopt->sopt_valsize > SOMAXOPT_SIZE)
1160 error = holdsock(p->p_fd, s, &fp);
1164 error = sogetopt((struct socket *)fp->f_data, sopt);
1170 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1175 sys_getsockopt(struct getsockopt_args *uap)
1177 struct thread *td = curthread;
1178 struct sockopt sopt;
1182 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1189 sopt.sopt_level = uap->level;
1190 sopt.sopt_name = uap->name;
1191 sopt.sopt_valsize = valsize;
1193 sopt.sopt_val = NULL;
1195 if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1198 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1199 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1205 error = kern_getsockopt(uap->s, &sopt);
1209 valsize = sopt.sopt_valsize;
1210 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1214 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1217 kfree(sopt.sopt_val, M_TEMP);
1222 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1223 * This allows kern_getsockname() to return a pointer to an allocated struct
1224 * sockaddr which must be freed later with FREE(). The caller must
1225 * initialize *name to NULL.
1228 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1230 struct thread *td = curthread;
1231 struct proc *p = td->td_proc;
1234 struct sockaddr *sa = NULL;
1237 error = holdsock(p->p_fd, s, &fp);
1244 so = (struct socket *)fp->f_data;
1245 error = so_pru_sockaddr(so, &sa);
1250 *namelen = MIN(*namelen, sa->sa_len);
1260 * getsockname_args(int fdes, caddr_t asa, int *alen)
1267 sys_getsockname(struct getsockname_args *uap)
1269 struct sockaddr *sa = NULL;
1272 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1277 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1281 error = copyout(sa, uap->asa, sa_len);
1283 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1290 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1291 * This allows kern_getpeername() to return a pointer to an allocated struct
1292 * sockaddr which must be freed later with FREE(). The caller must
1293 * initialize *name to NULL.
1296 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1298 struct thread *td = curthread;
1299 struct proc *p = td->td_proc;
1302 struct sockaddr *sa = NULL;
1305 error = holdsock(p->p_fd, s, &fp);
1312 so = (struct socket *)fp->f_data;
1313 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1317 error = so_pru_peeraddr(so, &sa);
1322 *namelen = MIN(*namelen, sa->sa_len);
1332 * getpeername_args(int fdes, caddr_t asa, int *alen)
1334 * Get name of peer for connected socket.
1339 sys_getpeername(struct getpeername_args *uap)
1341 struct sockaddr *sa = NULL;
1344 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1349 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1353 error = copyout(sa, uap->asa, sa_len);
1355 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1362 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1364 struct sockaddr *sa;
1368 if (len > SOCK_MAXADDRLEN)
1369 return ENAMETOOLONG;
1370 if (len < offsetof(struct sockaddr, sa_data[0]))
1372 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1373 error = copyin(uaddr, sa, len);
1377 #if BYTE_ORDER != BIG_ENDIAN
1379 * The bind(), connect(), and sendto() syscalls were not
1380 * versioned for COMPAT_43. Thus, this check must stay.
1382 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1383 sa->sa_family = sa->sa_len;
1392 * Detach a mapped page and release resources back to the system.
1393 * We must release our wiring and if the object is ripped out
1394 * from under the vm_page we become responsible for freeing the
1395 * page. These routines must be MPSAFE.
1397 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1399 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1402 sf_buf_mref(void *arg)
1404 struct sfbuf_mref *sfm = arg;
1407 * We must already hold a ref so there is no race to 0, just
1408 * atomically increment the count.
1410 atomic_add_int(&sfm->mref_count, 1);
1414 sf_buf_mfree(void *arg)
1416 struct sfbuf_mref *sfm = arg;
1419 KKASSERT(sfm->mref_count > 0);
1420 if (atomic_fetchadd_int(&sfm->mref_count, -1) == 1) {
1422 * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1426 m = sf_buf_page(sfm->sf);
1427 sf_buf_free(sfm->sf);
1428 vm_page_unwire(m, 0);
1429 if (m->wire_count == 0 && m->object == NULL)
1430 vm_page_try_to_free(m);
1433 kfree(sfm, M_SENDFILE);
1439 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1440 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1442 * Send a file specified by 'fd' and starting at 'offset' to a socket
1443 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1444 * nbytes == 0. Optionally add a header and/or trailer to the socket
1445 * output. If specified, write the total number of bytes sent into *sbytes.
1447 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1448 * the headers to count against the remaining bytes to be sent from
1449 * the file descriptor. We may wish to implement a compatibility syscall
1455 sys_sendfile(struct sendfile_args *uap)
1457 struct thread *td = curthread;
1458 struct proc *p = td->td_proc;
1460 struct vnode *vp = NULL;
1461 struct sf_hdtr hdtr;
1462 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1464 struct mbuf *mheader = NULL;
1467 off_t hdtr_size = 0;
1474 * Do argument checking. Must be a regular file in, stream
1475 * type and connected socket out, positive offset.
1477 fp = holdfp(p->p_fd, uap->fd, FREAD);
1481 if (fp->f_type != DTYPE_VNODE) {
1486 vp = (struct vnode *)fp->f_data;
1491 * If specified, get the pointer to the sf_hdtr struct for
1492 * any headers/trailers.
1495 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1502 error = iovec_copyin(hdtr.headers, &iov, aiov,
1503 hdtr.hdr_cnt, &hbytes);
1507 auio.uio_iovcnt = hdtr.hdr_cnt;
1508 auio.uio_offset = 0;
1509 auio.uio_segflg = UIO_USERSPACE;
1510 auio.uio_rw = UIO_WRITE;
1512 auio.uio_resid = hbytes;
1514 mheader = m_uiomove(&auio);
1516 iovec_free(&iov, aiov);
1517 if (mheader == NULL)
1522 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1523 &sbytes, uap->flags);
1528 * Send trailers. Wimp out and use writev(2).
1530 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1531 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1532 hdtr.trl_cnt, &auio.uio_resid);
1536 auio.uio_iovcnt = hdtr.trl_cnt;
1537 auio.uio_offset = 0;
1538 auio.uio_segflg = UIO_USERSPACE;
1539 auio.uio_rw = UIO_WRITE;
1542 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1544 iovec_free(&iov, aiov);
1547 hdtr_size += tbytes; /* trailer bytes successfully sent */
1554 if (uap->sbytes != NULL) {
1555 sbytes += hdtr_size;
1556 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1562 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1563 struct mbuf *mheader, off_t *sbytes, int flags)
1565 struct thread *td = curthread;
1566 struct proc *p = td->td_proc;
1567 struct vm_object *obj;
1572 struct sfbuf_mref *sfm;
1578 if (vp->v_type != VREG) {
1582 if ((obj = vp->v_object) == NULL) {
1586 error = holdsock(p->p_fd, sfd, &fp);
1589 so = (struct socket *)fp->f_data;
1590 if (so->so_type != SOCK_STREAM) {
1594 if ((so->so_state & SS_ISCONNECTED) == 0) {
1605 * Protect against multiple writers to the socket.
1607 ssb_lock(&so->so_snd, M_WAITOK);
1610 * Loop through the pages in the file, starting with the requested
1611 * offset. Get a file page (do I/O if necessary), map the file page
1612 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1615 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1619 pindex = OFF_TO_IDX(off);
1622 * Calculate the amount to transfer. Not to exceed a page,
1623 * the EOF, or the passed in nbytes.
1625 xfsize = vp->v_filesize - off;
1626 if (xfsize > PAGE_SIZE)
1628 pgoff = (vm_offset_t)(off & PAGE_MASK);
1629 if (PAGE_SIZE - pgoff < xfsize)
1630 xfsize = PAGE_SIZE - pgoff;
1631 if (nbytes && xfsize > (nbytes - *sbytes))
1632 xfsize = nbytes - *sbytes;
1636 * Optimize the non-blocking case by looking at the socket space
1637 * before going to the extra work of constituting the sf_buf.
1639 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) {
1640 if (so->so_state & SS_CANTSENDMORE)
1644 ssb_unlock(&so->so_snd);
1648 * Attempt to look up the page.
1650 * Allocate if not found, wait and loop if busy, then
1651 * wire the page. critical section protection is
1652 * required to maintain the object association (an
1653 * interrupt can free the page) through to the
1654 * vm_page_wire() call.
1657 pg = vm_page_lookup(obj, pindex);
1659 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1666 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1674 * If page is not valid for what we need, initiate I/O
1677 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1683 * Ensure that our page is still around when the I/O
1686 vm_page_io_start(pg);
1689 * Get the page from backing store.
1691 bsize = vp->v_mount->mnt_stat.f_iosize;
1692 auio.uio_iov = &aiov;
1693 auio.uio_iovcnt = 1;
1695 aiov.iov_len = MAXBSIZE;
1696 auio.uio_resid = MAXBSIZE;
1697 auio.uio_offset = trunc_page(off);
1698 auio.uio_segflg = UIO_NOCOPY;
1699 auio.uio_rw = UIO_READ;
1701 vn_lock(vp, LK_SHARED | LK_RETRY);
1702 error = VOP_READ(vp, &auio,
1703 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1706 vm_page_flag_clear(pg, PG_ZERO);
1707 vm_page_io_finish(pg);
1710 vm_page_unwire(pg, 0);
1711 vm_page_try_to_free(pg);
1713 ssb_unlock(&so->so_snd);
1720 * Get a sendfile buf. We usually wait as long as necessary,
1721 * but this wait can be interrupted.
1723 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1725 vm_page_unwire(pg, 0);
1726 vm_page_try_to_free(pg);
1728 ssb_unlock(&so->so_snd);
1734 * Get an mbuf header and set it up as having external storage.
1736 MGETHDR(m, MB_WAIT, MT_DATA);
1740 ssb_unlock(&so->so_snd);
1745 * sfm is a temporary hack, use a per-cpu cache for this.
1747 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1749 sfm->mref_count = 1;
1751 m->m_ext.ext_free = sf_buf_mfree;
1752 m->m_ext.ext_ref = sf_buf_mref;
1753 m->m_ext.ext_arg = sfm;
1754 m->m_ext.ext_buf = (void *)sf->kva;
1755 m->m_ext.ext_size = PAGE_SIZE;
1756 m->m_data = (char *) sf->kva + pgoff;
1757 m->m_flags |= M_EXT;
1758 m->m_pkthdr.len = m->m_len = xfsize;
1759 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1761 if (mheader != NULL) {
1762 hbytes = mheader->m_pkthdr.len;
1763 mheader->m_pkthdr.len += m->m_pkthdr.len;
1771 * Add the buffer to the socket buffer chain.
1776 * Make sure that the socket is still able to take more data.
1777 * CANTSENDMORE being true usually means that the connection
1778 * was closed. so_error is true when an error was sensed after
1780 * The state is checked after the page mapping and buffer
1781 * allocation above since those operations may block and make
1782 * any socket checks stale. From this point forward, nothing
1783 * blocks before the pru_send (or more accurately, any blocking
1784 * results in a loop back to here to re-check).
1786 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1787 if (so->so_state & SS_CANTSENDMORE) {
1790 error = so->so_error;
1794 ssb_unlock(&so->so_snd);
1799 * Wait for socket space to become available. We do this just
1800 * after checking the connection state above in order to avoid
1801 * a race condition with ssb_wait().
1803 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) {
1804 if (fp->f_flag & FNONBLOCK) {
1806 ssb_unlock(&so->so_snd);
1811 error = ssb_wait(&so->so_snd);
1813 * An error from ssb_wait usually indicates that we've
1814 * been interrupted by a signal. If we've sent anything
1815 * then return bytes sent, otherwise return the error.
1819 ssb_unlock(&so->so_snd);
1825 error = so_pru_send(so, 0, m, NULL, NULL, td);
1828 ssb_unlock(&so->so_snd);
1832 if (mheader != NULL) {
1833 *sbytes += mheader->m_pkthdr.len;
1834 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);
1871 head = (struct socket *)lfp->f_data;
1872 error = sctp_can_peel_off(head, assoc_id);
1878 * At this point we know we do have a assoc to pull
1879 * we proceed to get the fd setup. This may block
1883 fflag = lfp->f_flag;
1884 error = falloc(td->td_lwp, &nfp, &fd);
1887 * Probably ran out of file descriptors. Put the
1888 * unaccepted connection back onto the queue and
1889 * do another wakeup so some other process might
1890 * have a chance at it.
1895 uap->sysmsg_iresult = fd;
1897 so = sctp_get_peeloff(head, assoc_id, &error);
1900 * Either someone else peeled it off OR
1901 * we can't get a socket.
1905 so->so_state &= ~SS_COMP;
1906 so->so_state &= ~SS_NOFDREF;
1908 if (head->so_sigio != NULL)
1909 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1911 nfp->f_type = DTYPE_SOCKET;
1912 nfp->f_flag = fflag;
1913 nfp->f_ops = &socketops;
1918 * Assign the file pointer to the reserved descriptor, or clear
1919 * the reserved descriptor if an error occured.
1922 fsetfd(fdp, NULL, fd);
1924 fsetfd(fdp, nfp, fd);
1927 * Release explicitly held references before returning.