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 1.80 2007/04/22 01:13:10 dillon 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>
83 #include <netinet/sctp_peeloff.h>
89 struct lwkt_serialize serializer;
92 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures");
95 * System call interface to the socket abstraction.
98 extern struct fileops socketops;
101 * socket_args(int domain, int type, int protocol)
104 kern_socket(int domain, int type, int protocol, int *res)
106 struct thread *td = curthread;
107 struct proc *p = td->td_proc;
114 error = falloc(p, &fp, &fd);
117 error = socreate(domain, &so, type, protocol, td);
121 fp->f_type = DTYPE_SOCKET;
122 fp->f_flag = FREAD | FWRITE;
123 fp->f_ops = &socketops;
133 sys_socket(struct socket_args *uap)
137 error = kern_socket(uap->domain, uap->type, uap->protocol,
138 &uap->sysmsg_result);
144 kern_bind(int s, struct sockaddr *sa)
146 struct thread *td = curthread;
147 struct proc *p = td->td_proc;
152 error = holdsock(p->p_fd, s, &fp);
155 error = sobind((struct socket *)fp->f_data, sa, td);
161 * bind_args(int s, caddr_t name, int namelen)
164 sys_bind(struct bind_args *uap)
169 error = getsockaddr(&sa, uap->name, uap->namelen);
172 error = kern_bind(uap->s, sa);
179 kern_listen(int s, int backlog)
181 struct thread *td = curthread;
182 struct proc *p = td->td_proc;
187 error = holdsock(p->p_fd, s, &fp);
190 error = solisten((struct socket *)fp->f_data, backlog, td);
196 * listen_args(int s, int backlog)
199 sys_listen(struct listen_args *uap)
203 error = kern_listen(uap->s, uap->backlog);
208 * Returns the accepted socket as well.
211 soaccept_predicate(struct netmsg *msg0)
213 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
214 struct socket *head = msg->nm_so;
216 if (head->so_error != 0) {
217 msg->nm_lmsg.ms_error = head->so_error;
220 if (!TAILQ_EMPTY(&head->so_comp)) {
221 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
222 msg->nm_so = TAILQ_FIRST(&head->so_comp);
223 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
226 msg->nm_lmsg.ms_error = 0;
229 if (head->so_state & SS_CANTRCVMORE) {
230 msg->nm_lmsg.ms_error = ECONNABORTED;
233 if (msg->nm_fflags & FNONBLOCK) {
234 msg->nm_lmsg.ms_error = EWOULDBLOCK;
242 * The second argument to kern_accept() is a handle to a struct sockaddr.
243 * This allows kern_accept() to return a pointer to an allocated struct
244 * sockaddr which must be freed later with FREE(). The caller must
245 * initialize *name to NULL.
248 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res)
250 struct thread *td = curthread;
251 struct proc *p = td->td_proc;
252 struct file *lfp = NULL;
253 struct file *nfp = NULL;
255 struct socket *head, *so;
256 struct netmsg_so_notify msg;
259 u_int fflag; /* type must match fp->f_flag */
263 if (name && namelen && *namelen < 0)
266 error = holdsock(p->p_fd, s, &lfp);
270 error = falloc(p, &nfp, &fd);
271 if (error) { /* Probably ran out of file descriptors. */
275 head = (struct socket *)lfp->f_data;
276 if ((head->so_options & SO_ACCEPTCONN) == 0) {
281 if (fflags & O_FBLOCKING)
282 fflags |= lfp->f_flag & ~FNONBLOCK;
283 else if (fflags & O_FNONBLOCKING)
284 fflags |= lfp->f_flag | FNONBLOCK;
286 fflags = lfp->f_flag;
288 /* optimize for uniprocessor case later XXX JH */
289 port = head->so_proto->pr_mport(head, NULL, PRU_PRED);
290 lwkt_initmsg(&msg.nm_lmsg, &curthread->td_msgport,
291 MSGF_PCATCH | MSGF_ABORTABLE,
292 lwkt_cmd_func(netmsg_so_notify),
293 lwkt_cmd_func(netmsg_so_notify_abort));
294 msg.nm_predicate = soaccept_predicate;
295 msg.nm_fflags = fflags;
297 msg.nm_etype = NM_REVENT;
298 error = lwkt_domsg(port, &msg.nm_lmsg);
303 * At this point we have the connection that's ready to be accepted.
309 /* connection has been removed from the listen queue */
310 KNOTE(&head->so_rcv.ssb_sel.si_note, 0);
312 so->so_state &= ~SS_COMP;
314 if (head->so_sigio != NULL)
315 fsetown(fgetown(head->so_sigio), &so->so_sigio);
317 nfp->f_type = DTYPE_SOCKET;
319 nfp->f_ops = &socketops;
321 /* Sync socket nonblocking/async state with file flags */
322 tmp = fflag & FNONBLOCK;
323 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, p->p_ucred);
324 tmp = fflag & FASYNC;
325 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, p->p_ucred);
328 error = soaccept(so, &sa);
331 * Set the returned name and namelen as applicable. Set the returned
332 * namelen to 0 for older code which might ignore the return value
336 if (sa && name && namelen) {
337 if (*namelen > sa->sa_len)
338 *namelen = sa->sa_len;
348 * If an error occured clear the reserved descriptor, else associate
351 * Note that *res is normally ignored if an error is returned but
352 * a syscall message will still have access to the result code.
366 * accept(int s, caddr_t name, int *anamelen)
369 sys_accept(struct accept_args *uap)
371 struct sockaddr *sa = NULL;
376 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
380 error = kern_accept(uap->s, 0, &sa, &sa_len, &uap->sysmsg_result);
383 error = copyout(sa, uap->name, sa_len);
385 error = copyout(&sa_len, uap->anamelen,
386 sizeof(*uap->anamelen));
391 error = kern_accept(uap->s, 0, NULL, 0, &uap->sysmsg_result);
397 * extaccept(int s, int fflags, caddr_t name, int *anamelen)
400 sys_extaccept(struct extaccept_args *uap)
402 struct sockaddr *sa = NULL;
405 int fflags = uap->flags & O_FMASK;
408 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
412 error = kern_accept(uap->s, fflags, &sa, &sa_len, &uap->sysmsg_result);
415 error = copyout(sa, uap->name, sa_len);
417 error = copyout(&sa_len, uap->anamelen,
418 sizeof(*uap->anamelen));
423 error = kern_accept(uap->s, fflags, NULL, 0, &uap->sysmsg_result);
430 * Returns TRUE if predicate satisfied.
433 soconnected_predicate(struct netmsg *msg0)
435 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
436 struct socket *so = msg->nm_so;
438 /* check predicate */
439 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
440 msg->nm_lmsg.ms_error = so->so_error;
448 kern_connect(int s, int fflags, struct sockaddr *sa)
450 struct thread *td = curthread;
451 struct proc *p = td->td_proc;
456 error = holdsock(p->p_fd, s, &fp);
459 so = (struct socket *)fp->f_data;
461 if (fflags & O_FBLOCKING)
462 /* fflags &= ~FNONBLOCK; */;
463 else if (fflags & O_FNONBLOCKING)
468 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
472 error = soconnect(so, sa, td);
475 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
479 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
480 struct netmsg_so_notify msg;
483 port = so->so_proto->pr_mport(so, sa, PRU_PRED);
484 lwkt_initmsg(&msg.nm_lmsg,
485 &curthread->td_msgport,
486 MSGF_PCATCH | MSGF_ABORTABLE,
487 lwkt_cmd_func(netmsg_so_notify),
488 lwkt_cmd_func(netmsg_so_notify_abort));
489 msg.nm_predicate = soconnected_predicate;
491 msg.nm_etype = NM_REVENT;
492 error = lwkt_domsg(port, &msg.nm_lmsg);
495 error = so->so_error;
499 so->so_state &= ~SS_ISCONNECTING;
500 if (error == ERESTART)
508 * connect_args(int s, caddr_t name, int namelen)
511 sys_connect(struct connect_args *uap)
516 error = getsockaddr(&sa, uap->name, uap->namelen);
519 error = kern_connect(uap->s, 0, sa);
526 * connect_args(int s, int fflags, caddr_t name, int namelen)
529 sys_extconnect(struct extconnect_args *uap)
533 int fflags = uap->flags & O_FMASK;
535 error = getsockaddr(&sa, uap->name, uap->namelen);
538 error = kern_connect(uap->s, fflags, sa);
545 kern_socketpair(int domain, int type, int protocol, int *sv)
547 struct thread *td = curthread;
548 struct proc *p = td->td_proc;
549 struct file *fp1, *fp2;
550 struct socket *so1, *so2;
554 error = socreate(domain, &so1, type, protocol, td);
557 error = socreate(domain, &so2, type, protocol, td);
560 error = falloc(p, &fp1, &fd1);
565 error = falloc(p, &fp2, &fd2);
570 error = soconnect2(so1, so2);
573 if (type == SOCK_DGRAM) {
575 * Datagram socket connection is asymmetric.
577 error = soconnect2(so2, so1);
581 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
582 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
583 fp1->f_ops = fp2->f_ops = &socketops;
590 fsetfd(p, NULL, fd2);
593 fsetfd(p, NULL, fd1);
596 (void)soclose(so2, 0);
598 (void)soclose(so1, 0);
603 * socketpair(int domain, int type, int protocol, int *rsv)
606 sys_socketpair(struct socketpair_args *uap)
610 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
613 error = copyout(sockv, uap->rsv, sizeof(sockv));
618 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
619 struct mbuf *control, int flags, int *res)
621 struct thread *td = curthread;
622 struct lwp *lp = td->td_lwp;
623 struct proc *p = td->td_proc;
628 struct iovec *ktriov = NULL;
632 error = holdsock(p->p_fd, s, &fp);
635 if (auio->uio_resid < 0) {
640 if (KTRPOINT(td, KTR_GENIO)) {
641 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
643 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
644 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
648 len = auio->uio_resid;
649 so = (struct socket *)fp->f_data;
650 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
651 if (fp->f_flag & FNONBLOCK)
652 flags |= MSG_FNONBLOCKING;
654 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
656 if (auio->uio_resid != len && (error == ERESTART ||
657 error == EINTR || error == EWOULDBLOCK))
660 lwpsignal(p, lp, SIGPIPE);
663 if (ktriov != NULL) {
665 ktruio.uio_iov = ktriov;
666 ktruio.uio_resid = len - auio->uio_resid;
667 ktrgenio(p, s, UIO_WRITE, &ktruio, error);
669 FREE(ktriov, M_TEMP);
673 *res = len - auio->uio_resid;
680 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
683 sys_sendto(struct sendto_args *uap)
685 struct thread *td = curthread;
688 struct sockaddr *sa = NULL;
692 error = getsockaddr(&sa, uap->to, uap->tolen);
696 aiov.iov_base = uap->buf;
697 aiov.iov_len = uap->len;
698 auio.uio_iov = &aiov;
701 auio.uio_resid = uap->len;
702 auio.uio_segflg = UIO_USERSPACE;
703 auio.uio_rw = UIO_WRITE;
706 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
707 &uap->sysmsg_result);
715 * sendmsg_args(int s, caddr_t msg, int flags)
718 sys_sendmsg(struct sendmsg_args *uap)
720 struct thread *td = curthread;
723 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
724 struct sockaddr *sa = NULL;
725 struct mbuf *control = NULL;
728 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
733 * Conditionally copyin msg.msg_name.
736 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
744 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
749 auio.uio_iovcnt = msg.msg_iovlen;
751 auio.uio_segflg = UIO_USERSPACE;
752 auio.uio_rw = UIO_WRITE;
756 * Conditionally copyin msg.msg_control.
758 if (msg.msg_control) {
759 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
760 msg.msg_controllen > MLEN) {
764 control = m_get(MB_WAIT, MT_CONTROL);
765 if (control == NULL) {
769 control->m_len = msg.msg_controllen;
770 error = copyin(msg.msg_control, mtod(control, caddr_t),
778 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
779 &uap->sysmsg_result);
782 iovec_free(&iov, aiov);
790 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
791 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
792 * Don't forget to FREE() and m_free() these if they are returned.
795 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
796 struct mbuf **control, int *flags, int *res)
798 struct thread *td = curthread;
799 struct proc *p = td->td_proc;
805 struct iovec *ktriov = NULL;
809 error = holdsock(p->p_fd, s, &fp);
812 if (auio->uio_resid < 0) {
817 if (KTRPOINT(td, KTR_GENIO)) {
818 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
820 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
821 bcopy(auio->uio_iov, ktriov, iovlen);
825 len = auio->uio_resid;
826 so = (struct socket *)fp->f_data;
828 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
829 if (fp->f_flag & FNONBLOCK) {
831 *flags |= MSG_FNONBLOCKING;
833 lflags = MSG_FNONBLOCKING;
839 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
841 if (auio->uio_resid != len && (error == ERESTART ||
842 error == EINTR || error == EWOULDBLOCK))
846 if (ktriov != NULL) {
848 ktruio.uio_iov = ktriov;
849 ktruio.uio_resid = len - auio->uio_resid;
850 ktrgenio(p, s, UIO_READ, &ktruio, error);
852 FREE(ktriov, M_TEMP);
856 *res = len - auio->uio_resid;
863 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
864 * caddr_t from, int *fromlenaddr)
867 sys_recvfrom(struct recvfrom_args *uap)
869 struct thread *td = curthread;
872 struct sockaddr *sa = NULL;
875 if (uap->from && uap->fromlenaddr) {
876 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
884 aiov.iov_base = uap->buf;
885 aiov.iov_len = uap->len;
886 auio.uio_iov = &aiov;
889 auio.uio_resid = uap->len;
890 auio.uio_segflg = UIO_USERSPACE;
891 auio.uio_rw = UIO_READ;
894 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
895 &uap->flags, &uap->sysmsg_result);
897 if (error == 0 && uap->from) {
898 /* note: sa may still be NULL */
900 fromlen = MIN(fromlen, sa->sa_len);
901 error = copyout(sa, uap->from, fromlen);
906 error = copyout(&fromlen, uap->fromlenaddr,
917 * recvmsg_args(int s, struct msghdr *msg, int flags)
920 sys_recvmsg(struct recvmsg_args *uap)
922 struct thread *td = curthread;
925 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
926 struct mbuf *m, *control = NULL;
927 struct sockaddr *sa = NULL;
929 socklen_t *ufromlenp, *ucontrollenp;
930 int error, fromlen, controllen, len, flags, *uflagsp;
933 * This copyin handles everything except the iovec.
935 error = copyin(uap->msg, &msg, sizeof(msg));
939 if (msg.msg_name && msg.msg_namelen < 0)
941 if (msg.msg_control && msg.msg_controllen < 0)
944 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
946 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
948 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
954 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
959 auio.uio_iovcnt = msg.msg_iovlen;
961 auio.uio_segflg = UIO_USERSPACE;
962 auio.uio_rw = UIO_READ;
967 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
968 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
971 * Conditionally copyout the name and populate the namelen field.
973 if (error == 0 && msg.msg_name) {
974 /* note: sa may still be NULL */
976 fromlen = MIN(msg.msg_namelen, sa->sa_len);
977 error = copyout(sa, msg.msg_name, fromlen);
981 error = copyout(&fromlen, ufromlenp,
986 * Copyout msg.msg_control and msg.msg_controllen.
988 if (error == 0 && msg.msg_control) {
989 len = msg.msg_controllen;
991 ctlbuf = (caddr_t)msg.msg_control;
993 while(m && len > 0) {
996 if (len >= m->m_len) {
999 msg.msg_flags |= MSG_CTRUNC;
1003 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1011 controllen = ctlbuf - (caddr_t)msg.msg_control;
1012 error = copyout(&controllen, ucontrollenp,
1013 sizeof(*ucontrollenp));
1017 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1022 iovec_free(&iov, aiov);
1029 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1030 * in kernel pointer instead of a userland pointer. This allows us
1031 * to manipulate socket options in the emulation code.
1034 kern_setsockopt(int s, struct sockopt *sopt)
1036 struct thread *td = curthread;
1037 struct proc *p = td->td_proc;
1041 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1043 if (sopt->sopt_valsize < 0)
1046 error = holdsock(p->p_fd, s, &fp);
1050 error = sosetopt((struct socket *)fp->f_data, sopt);
1056 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1059 sys_setsockopt(struct setsockopt_args *uap)
1061 struct thread *td = curthread;
1062 struct sockopt sopt;
1065 sopt.sopt_level = uap->level;
1066 sopt.sopt_name = uap->name;
1067 sopt.sopt_val = uap->val;
1068 sopt.sopt_valsize = uap->valsize;
1071 error = kern_setsockopt(uap->s, &sopt);
1076 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1077 * in kernel pointer instead of a userland pointer. This allows us
1078 * to manipulate socket options in the emulation code.
1081 kern_getsockopt(int s, struct sockopt *sopt)
1083 struct thread *td = curthread;
1084 struct proc *p = td->td_proc;
1088 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1090 if (sopt->sopt_valsize < 0)
1093 error = holdsock(p->p_fd, s, &fp);
1097 error = sogetopt((struct socket *)fp->f_data, sopt);
1103 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1106 sys_getsockopt(struct getsockopt_args *uap)
1108 struct thread *td = curthread;
1109 struct sockopt sopt;
1113 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1122 sopt.sopt_level = uap->level;
1123 sopt.sopt_name = uap->name;
1124 sopt.sopt_val = uap->val;
1125 sopt.sopt_valsize = valsize;
1128 error = kern_getsockopt(uap->s, &sopt);
1130 valsize = sopt.sopt_valsize;
1131 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1137 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1138 * This allows kern_getsockname() to return a pointer to an allocated struct
1139 * sockaddr which must be freed later with FREE(). The caller must
1140 * initialize *name to NULL.
1143 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1145 struct thread *td = curthread;
1146 struct proc *p = td->td_proc;
1149 struct sockaddr *sa = NULL;
1152 error = holdsock(p->p_fd, s, &fp);
1159 so = (struct socket *)fp->f_data;
1160 error = so_pru_sockaddr(so, &sa);
1165 *namelen = MIN(*namelen, sa->sa_len);
1175 * getsockname_args(int fdes, caddr_t asa, int *alen)
1180 sys_getsockname(struct getsockname_args *uap)
1182 struct sockaddr *sa = NULL;
1185 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1189 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1192 error = copyout(sa, uap->asa, sa_len);
1194 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1201 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1202 * This allows kern_getpeername() to return a pointer to an allocated struct
1203 * sockaddr which must be freed later with FREE(). The caller must
1204 * initialize *name to NULL.
1207 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1209 struct thread *td = curthread;
1210 struct proc *p = td->td_proc;
1213 struct sockaddr *sa = NULL;
1216 error = holdsock(p->p_fd, s, &fp);
1223 so = (struct socket *)fp->f_data;
1224 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1228 error = so_pru_peeraddr(so, &sa);
1233 *namelen = MIN(*namelen, sa->sa_len);
1243 * getpeername_args(int fdes, caddr_t asa, int *alen)
1245 * Get name of peer for connected socket.
1248 sys_getpeername(struct getpeername_args *uap)
1250 struct sockaddr *sa = NULL;
1253 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1257 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1260 error = copyout(sa, uap->asa, sa_len);
1262 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1269 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1271 struct sockaddr *sa;
1275 if (len > SOCK_MAXADDRLEN)
1276 return ENAMETOOLONG;
1277 if (len < offsetof(struct sockaddr, sa_data[0]))
1279 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1280 error = copyin(uaddr, sa, len);
1284 #if BYTE_ORDER != BIG_ENDIAN
1286 * The bind(), connect(), and sendto() syscalls were not
1287 * versioned for COMPAT_43. Thus, this check must stay.
1289 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1290 sa->sa_family = sa->sa_len;
1299 * Detach a mapped page and release resources back to the system.
1300 * We must release our wiring and if the object is ripped out
1301 * from under the vm_page we become responsible for freeing the
1302 * page. These routines must be MPSAFE.
1304 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1306 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1309 sf_buf_mref(void *arg)
1311 struct sfbuf_mref *sfm = arg;
1314 * We must already hold a ref so there is no race to 0, just
1315 * atomically increment the count.
1317 atomic_add_int(&sfm->mref_count, 1);
1321 sf_buf_mfree(void *arg)
1323 struct sfbuf_mref *sfm = arg;
1326 KKASSERT(sfm->mref_count > 0);
1327 if (sfm->mref_count == 1) {
1329 * We are the only holder so no further locking is required,
1330 * the sfbuf can simply be freed.
1332 sfm->mref_count = 0;
1336 * There may be other holders, we must obtain the serializer
1337 * to protect against a sf_buf_mfree() race to 0. An atomic
1338 * operation is still required for races against
1341 * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1343 lwkt_serialize_enter(&sfm->serializer);
1344 atomic_subtract_int(&sfm->mref_count, 1);
1345 if (sfm->mref_count == 0) {
1346 lwkt_serialize_exit(&sfm->serializer);
1350 m = sf_buf_page(sfm->sf);
1351 sf_buf_free(sfm->sf);
1352 vm_page_unwire(m, 0);
1353 if (m->wire_count == 0 && m->object == NULL)
1354 vm_page_try_to_free(m);
1357 kfree(sfm, M_SENDFILE);
1359 lwkt_serialize_exit(&sfm->serializer);
1366 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1367 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1369 * Send a file specified by 'fd' and starting at 'offset' to a socket
1370 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1371 * nbytes == 0. Optionally add a header and/or trailer to the socket
1372 * output. If specified, write the total number of bytes sent into *sbytes.
1374 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1375 * the headers to count against the remaining bytes to be sent from
1376 * the file descriptor. We may wish to implement a compatibility syscall
1380 sys_sendfile(struct sendfile_args *uap)
1382 struct thread *td = curthread;
1383 struct proc *p = td->td_proc;
1385 struct vnode *vp = NULL;
1386 struct sf_hdtr hdtr;
1387 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1389 struct mbuf *mheader = NULL;
1390 off_t hdtr_size = 0, sbytes;
1391 int error, hbytes = 0, tbytes;
1396 * Do argument checking. Must be a regular file in, stream
1397 * type and connected socket out, positive offset.
1399 fp = holdfp(p->p_fd, uap->fd, FREAD);
1403 if (fp->f_type != DTYPE_VNODE) {
1407 vp = (struct vnode *)fp->f_data;
1412 * If specified, get the pointer to the sf_hdtr struct for
1413 * any headers/trailers.
1416 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1423 error = iovec_copyin(hdtr.headers, &iov, aiov,
1424 hdtr.hdr_cnt, &hbytes);
1428 auio.uio_iovcnt = hdtr.hdr_cnt;
1429 auio.uio_offset = 0;
1430 auio.uio_segflg = UIO_USERSPACE;
1431 auio.uio_rw = UIO_WRITE;
1433 auio.uio_resid = hbytes;
1435 mheader = m_uiomove(&auio);
1437 iovec_free(&iov, aiov);
1438 if (mheader == NULL)
1443 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1444 &sbytes, uap->flags);
1449 * Send trailers. Wimp out and use writev(2).
1451 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1452 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1453 hdtr.trl_cnt, &auio.uio_resid);
1457 auio.uio_iovcnt = hdtr.trl_cnt;
1458 auio.uio_offset = 0;
1459 auio.uio_segflg = UIO_USERSPACE;
1460 auio.uio_rw = UIO_WRITE;
1463 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1465 iovec_free(&iov, aiov);
1468 hdtr_size += tbytes; /* trailer bytes successfully sent */
1472 if (uap->sbytes != NULL) {
1473 sbytes += hdtr_size;
1474 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1482 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1483 struct mbuf *mheader, off_t *sbytes, int flags)
1485 struct thread *td = curthread;
1486 struct proc *p = td->td_proc;
1487 struct vm_object *obj;
1492 struct sfbuf_mref *sfm;
1498 if (vp->v_type != VREG) {
1502 if ((obj = vp->v_object) == NULL) {
1506 error = holdsock(p->p_fd, sfd, &fp);
1509 so = (struct socket *)fp->f_data;
1510 if (so->so_type != SOCK_STREAM) {
1514 if ((so->so_state & SS_ISCONNECTED) == 0) {
1525 * Protect against multiple writers to the socket.
1527 ssb_lock(&so->so_snd, M_WAITOK);
1530 * Loop through the pages in the file, starting with the requested
1531 * offset. Get a file page (do I/O if necessary), map the file page
1532 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1535 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1539 pindex = OFF_TO_IDX(off);
1542 * Calculate the amount to transfer. Not to exceed a page,
1543 * the EOF, or the passed in nbytes.
1545 xfsize = vp->v_filesize - off;
1546 if (xfsize > PAGE_SIZE)
1548 pgoff = (vm_offset_t)(off & PAGE_MASK);
1549 if (PAGE_SIZE - pgoff < xfsize)
1550 xfsize = PAGE_SIZE - pgoff;
1551 if (nbytes && xfsize > (nbytes - *sbytes))
1552 xfsize = nbytes - *sbytes;
1556 * Optimize the non-blocking case by looking at the socket space
1557 * before going to the extra work of constituting the sf_buf.
1559 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) {
1560 if (so->so_state & SS_CANTSENDMORE)
1564 ssb_unlock(&so->so_snd);
1568 * Attempt to look up the page.
1570 * Allocate if not found, wait and loop if busy, then
1571 * wire the page. critical section protection is
1572 * required to maintain the object association (an
1573 * interrupt can free the page) through to the
1574 * vm_page_wire() call.
1577 pg = vm_page_lookup(obj, pindex);
1579 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1586 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1594 * If page is not valid for what we need, initiate I/O
1597 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1603 * Ensure that our page is still around when the I/O
1606 vm_page_io_start(pg);
1609 * Get the page from backing store.
1611 bsize = vp->v_mount->mnt_stat.f_iosize;
1612 auio.uio_iov = &aiov;
1613 auio.uio_iovcnt = 1;
1615 aiov.iov_len = MAXBSIZE;
1616 auio.uio_resid = MAXBSIZE;
1617 auio.uio_offset = trunc_page(off);
1618 auio.uio_segflg = UIO_NOCOPY;
1619 auio.uio_rw = UIO_READ;
1621 vn_lock(vp, LK_SHARED | LK_RETRY);
1622 error = VOP_READ(vp, &auio,
1623 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1626 vm_page_flag_clear(pg, PG_ZERO);
1627 vm_page_io_finish(pg);
1630 vm_page_unwire(pg, 0);
1631 vm_page_try_to_free(pg);
1633 ssb_unlock(&so->so_snd);
1640 * Get a sendfile buf. We usually wait as long as necessary,
1641 * but this wait can be interrupted.
1643 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1645 vm_page_unwire(pg, 0);
1646 vm_page_try_to_free(pg);
1648 ssb_unlock(&so->so_snd);
1654 * Get an mbuf header and set it up as having external storage.
1656 MGETHDR(m, MB_WAIT, MT_DATA);
1660 ssb_unlock(&so->so_snd);
1665 * sfm is a temporary hack, use a per-cpu cache for this.
1667 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1669 sfm->mref_count = 1;
1670 lwkt_serialize_init(&sfm->serializer);
1672 m->m_ext.ext_free = sf_buf_mfree;
1673 m->m_ext.ext_ref = sf_buf_mref;
1674 m->m_ext.ext_arg = sfm;
1675 m->m_ext.ext_buf = (void *)sf->kva;
1676 m->m_ext.ext_size = PAGE_SIZE;
1677 m->m_data = (char *) sf->kva + pgoff;
1678 m->m_flags |= M_EXT;
1679 m->m_pkthdr.len = m->m_len = xfsize;
1680 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1682 if (mheader != NULL) {
1683 hbytes = mheader->m_pkthdr.len;
1684 mheader->m_pkthdr.len += m->m_pkthdr.len;
1692 * Add the buffer to the socket buffer chain.
1697 * Make sure that the socket is still able to take more data.
1698 * CANTSENDMORE being true usually means that the connection
1699 * was closed. so_error is true when an error was sensed after
1701 * The state is checked after the page mapping and buffer
1702 * allocation above since those operations may block and make
1703 * any socket checks stale. From this point forward, nothing
1704 * blocks before the pru_send (or more accurately, any blocking
1705 * results in a loop back to here to re-check).
1707 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1708 if (so->so_state & SS_CANTSENDMORE) {
1711 error = so->so_error;
1715 ssb_unlock(&so->so_snd);
1720 * Wait for socket space to become available. We do this just
1721 * after checking the connection state above in order to avoid
1722 * a race condition with ssb_wait().
1724 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) {
1725 if (fp->f_flag & FNONBLOCK) {
1727 ssb_unlock(&so->so_snd);
1732 error = ssb_wait(&so->so_snd);
1734 * An error from ssb_wait usually indicates that we've
1735 * been interrupted by a signal. If we've sent anything
1736 * then return bytes sent, otherwise return the error.
1740 ssb_unlock(&so->so_snd);
1746 error = so_pru_send(so, 0, m, NULL, NULL, td);
1749 ssb_unlock(&so->so_snd);
1753 if (mheader != NULL) {
1754 *sbytes += mheader->m_pkthdr.len;
1755 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1758 ssb_unlock(&so->so_snd);
1763 if (mheader != NULL)
1769 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1772 struct thread *td = curthread;
1773 struct proc *p = td->td_proc;
1774 struct file *lfp = NULL;
1775 struct file *nfp = NULL;
1777 struct socket *head, *so;
1780 short fflag; /* type must match fp->f_flag */
1782 assoc_id = uap->name;
1783 error = holdsock(p->p_fd, uap->sd, &lfp);
1788 head = (struct socket *)lfp->f_data;
1789 error = sctp_can_peel_off(head, assoc_id);
1795 * At this point we know we do have a assoc to pull
1796 * we proceed to get the fd setup. This may block
1800 fflag = lfp->f_flag;
1801 error = falloc(p, &nfp, &fd);
1804 * Probably ran out of file descriptors. Put the
1805 * unaccepted connection back onto the queue and
1806 * do another wakeup so some other process might
1807 * have a chance at it.
1812 uap->sysmsg_result = fd;
1814 so = sctp_get_peeloff(head, assoc_id, &error);
1817 * Either someone else peeled it off OR
1818 * we can't get a socket.
1822 so->so_state &= ~SS_COMP;
1823 so->so_state &= ~SS_NOFDREF;
1825 if (head->so_sigio != NULL)
1826 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1828 nfp->f_type = DTYPE_SOCKET;
1829 nfp->f_flag = fflag;
1830 nfp->f_ops = &socketops;
1835 * Assign the file pointer to the reserved descriptor, or clear
1836 * the reserved descriptor if an error occured.
1839 fsetfd(p, NULL, fd);
1844 * Release explicitly held references before returning.