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 1.72 2006/08/08 03:52:40 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.sb_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 * __accept(int s, int fflags, caddr_t name, int *anamelen)
400 sys___accept(struct __accept_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___connect(struct __connect_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 proc *p = td->td_proc;
627 struct iovec *ktriov = NULL;
631 error = holdsock(p->p_fd, s, &fp);
634 if (auio->uio_resid < 0) {
639 if (KTRPOINT(td, KTR_GENIO)) {
640 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
642 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
643 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
647 len = auio->uio_resid;
648 so = (struct socket *)fp->f_data;
649 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
650 if (fp->f_flag & FNONBLOCK)
651 flags |= MSG_FNONBLOCKING;
653 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
655 if (auio->uio_resid != len && (error == ERESTART ||
656 error == EINTR || error == EWOULDBLOCK))
662 if (ktriov != NULL) {
664 ktruio.uio_iov = ktriov;
665 ktruio.uio_resid = len - auio->uio_resid;
666 ktrgenio(p, s, UIO_WRITE, &ktruio, error);
668 FREE(ktriov, M_TEMP);
672 *res = len - auio->uio_resid;
679 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
682 sys_sendto(struct sendto_args *uap)
684 struct thread *td = curthread;
687 struct sockaddr *sa = NULL;
691 error = getsockaddr(&sa, uap->to, uap->tolen);
695 aiov.iov_base = uap->buf;
696 aiov.iov_len = uap->len;
697 auio.uio_iov = &aiov;
700 auio.uio_resid = uap->len;
701 auio.uio_segflg = UIO_USERSPACE;
702 auio.uio_rw = UIO_WRITE;
705 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
706 &uap->sysmsg_result);
714 * sendmsg_args(int s, caddr_t msg, int flags)
717 sys_sendmsg(struct sendmsg_args *uap)
719 struct thread *td = curthread;
722 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
723 struct sockaddr *sa = NULL;
724 struct mbuf *control = NULL;
727 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
732 * Conditionally copyin msg.msg_name.
735 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
743 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
748 auio.uio_iovcnt = msg.msg_iovlen;
750 auio.uio_segflg = UIO_USERSPACE;
751 auio.uio_rw = UIO_WRITE;
755 * Conditionally copyin msg.msg_control.
757 if (msg.msg_control) {
758 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
759 msg.msg_controllen > MLEN) {
763 control = m_get(MB_WAIT, MT_CONTROL);
764 if (control == NULL) {
768 control->m_len = msg.msg_controllen;
769 error = copyin(msg.msg_control, mtod(control, caddr_t),
777 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
778 &uap->sysmsg_result);
783 iovec_free(&iov, aiov);
788 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
789 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
790 * Don't forget to FREE() and m_free() these if they are returned.
793 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
794 struct mbuf **control, int *flags, int *res)
796 struct thread *td = curthread;
797 struct proc *p = td->td_proc;
803 struct iovec *ktriov = NULL;
807 error = holdsock(p->p_fd, s, &fp);
810 if (auio->uio_resid < 0) {
815 if (KTRPOINT(td, KTR_GENIO)) {
816 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
818 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
819 bcopy(auio->uio_iov, ktriov, iovlen);
823 len = auio->uio_resid;
824 so = (struct socket *)fp->f_data;
826 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
827 if (fp->f_flag & FNONBLOCK) {
829 *flags |= MSG_FNONBLOCKING;
831 lflags = MSG_FNONBLOCKING;
837 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
839 if (auio->uio_resid != len && (error == ERESTART ||
840 error == EINTR || error == EWOULDBLOCK))
844 if (ktriov != NULL) {
846 ktruio.uio_iov = ktriov;
847 ktruio.uio_resid = len - auio->uio_resid;
848 ktrgenio(p, s, UIO_READ, &ktruio, error);
850 FREE(ktriov, M_TEMP);
854 *res = len - auio->uio_resid;
861 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
862 * caddr_t from, int *fromlenaddr)
865 sys_recvfrom(struct recvfrom_args *uap)
867 struct thread *td = curthread;
870 struct sockaddr *sa = NULL;
873 if (uap->from && uap->fromlenaddr) {
874 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
882 aiov.iov_base = uap->buf;
883 aiov.iov_len = uap->len;
884 auio.uio_iov = &aiov;
887 auio.uio_resid = uap->len;
888 auio.uio_segflg = UIO_USERSPACE;
889 auio.uio_rw = UIO_READ;
892 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
893 &uap->flags, &uap->sysmsg_result);
895 if (error == 0 && uap->from) {
896 /* note: sa may still be NULL */
898 fromlen = MIN(fromlen, sa->sa_len);
899 error = copyout(sa, uap->from, fromlen);
904 error = copyout(&fromlen, uap->fromlenaddr,
915 * recvmsg_args(int s, struct msghdr *msg, int flags)
918 sys_recvmsg(struct recvmsg_args *uap)
920 struct thread *td = curthread;
923 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
924 struct mbuf *m, *control = NULL;
925 struct sockaddr *sa = NULL;
927 socklen_t *ufromlenp, *ucontrollenp;
928 int error, fromlen, controllen, len, flags, *uflagsp;
931 * This copyin handles everything except the iovec.
933 error = copyin(uap->msg, &msg, sizeof(msg));
937 if (msg.msg_name && msg.msg_namelen < 0)
939 if (msg.msg_control && msg.msg_controllen < 0)
942 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
944 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
946 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
952 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
957 auio.uio_iovcnt = msg.msg_iovlen;
959 auio.uio_segflg = UIO_USERSPACE;
960 auio.uio_rw = UIO_READ;
965 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
966 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
969 * Conditionally copyout the name and populate the namelen field.
971 if (error == 0 && msg.msg_name) {
972 fromlen = MIN(msg.msg_namelen, sa->sa_len);
973 error = copyout(sa, msg.msg_name, fromlen);
975 error = copyout(&fromlen, ufromlenp,
980 * Copyout msg.msg_control and msg.msg_controllen.
982 if (error == 0 && msg.msg_control) {
983 len = msg.msg_controllen;
985 ctlbuf = (caddr_t)msg.msg_control;
987 while(m && len > 0) {
990 if (len >= m->m_len) {
993 msg.msg_flags |= MSG_CTRUNC;
997 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1005 controllen = ctlbuf - (caddr_t)msg.msg_control;
1006 error = copyout(&controllen, ucontrollenp,
1007 sizeof(*ucontrollenp));
1011 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1016 iovec_free(&iov, aiov);
1023 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1024 * in kernel pointer instead of a userland pointer. This allows us
1025 * to manipulate socket options in the emulation code.
1028 kern_setsockopt(int s, struct sockopt *sopt)
1030 struct thread *td = curthread;
1031 struct proc *p = td->td_proc;
1035 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1037 if (sopt->sopt_valsize < 0)
1040 error = holdsock(p->p_fd, s, &fp);
1044 error = sosetopt((struct socket *)fp->f_data, sopt);
1050 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1053 sys_setsockopt(struct setsockopt_args *uap)
1055 struct thread *td = curthread;
1056 struct sockopt sopt;
1059 sopt.sopt_level = uap->level;
1060 sopt.sopt_name = uap->name;
1061 sopt.sopt_val = uap->val;
1062 sopt.sopt_valsize = uap->valsize;
1065 error = kern_setsockopt(uap->s, &sopt);
1070 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1071 * in kernel pointer instead of a userland pointer. This allows us
1072 * to manipulate socket options in the emulation code.
1075 kern_getsockopt(int s, struct sockopt *sopt)
1077 struct thread *td = curthread;
1078 struct proc *p = td->td_proc;
1082 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1084 if (sopt->sopt_valsize < 0)
1087 error = holdsock(p->p_fd, s, &fp);
1091 error = sogetopt((struct socket *)fp->f_data, sopt);
1097 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1100 sys_getsockopt(struct getsockopt_args *uap)
1102 struct thread *td = curthread;
1103 struct sockopt sopt;
1107 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1116 sopt.sopt_level = uap->level;
1117 sopt.sopt_name = uap->name;
1118 sopt.sopt_val = uap->val;
1119 sopt.sopt_valsize = valsize;
1122 error = kern_getsockopt(uap->s, &sopt);
1124 valsize = sopt.sopt_valsize;
1125 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1131 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1132 * This allows kern_getsockname() to return a pointer to an allocated struct
1133 * sockaddr which must be freed later with FREE(). The caller must
1134 * initialize *name to NULL.
1137 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1139 struct thread *td = curthread;
1140 struct proc *p = td->td_proc;
1143 struct sockaddr *sa = NULL;
1146 error = holdsock(p->p_fd, s, &fp);
1153 so = (struct socket *)fp->f_data;
1154 error = so_pru_sockaddr(so, &sa);
1159 *namelen = MIN(*namelen, sa->sa_len);
1169 * getsockname_args(int fdes, caddr_t asa, int *alen)
1174 sys_getsockname(struct getsockname_args *uap)
1176 struct sockaddr *sa = NULL;
1179 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1183 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1186 error = copyout(sa, uap->asa, sa_len);
1188 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1195 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1196 * This allows kern_getpeername() to return a pointer to an allocated struct
1197 * sockaddr which must be freed later with FREE(). The caller must
1198 * initialize *name to NULL.
1201 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1203 struct thread *td = curthread;
1204 struct proc *p = td->td_proc;
1207 struct sockaddr *sa = NULL;
1210 error = holdsock(p->p_fd, s, &fp);
1217 so = (struct socket *)fp->f_data;
1218 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1222 error = so_pru_peeraddr(so, &sa);
1227 *namelen = MIN(*namelen, sa->sa_len);
1237 * getpeername_args(int fdes, caddr_t asa, int *alen)
1239 * Get name of peer for connected socket.
1242 sys_getpeername(struct getpeername_args *uap)
1244 struct sockaddr *sa = NULL;
1247 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1251 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1254 error = copyout(sa, uap->asa, sa_len);
1256 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1263 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1265 struct sockaddr *sa;
1269 if (len > SOCK_MAXADDRLEN)
1270 return ENAMETOOLONG;
1271 if (len < offsetof(struct sockaddr, sa_data[0]))
1273 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1274 error = copyin(uaddr, sa, len);
1278 #if BYTE_ORDER != BIG_ENDIAN
1280 * The bind(), connect(), and sendto() syscalls were not
1281 * versioned for COMPAT_43. Thus, this check must stay.
1283 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1284 sa->sa_family = sa->sa_len;
1293 * Detach a mapped page and release resources back to the system.
1294 * We must release our wiring and if the object is ripped out
1295 * from under the vm_page we become responsible for freeing the
1296 * page. These routines must be MPSAFE.
1298 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1300 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1303 sf_buf_mref(void *arg)
1305 struct sfbuf_mref *sfm = arg;
1308 * We must already hold a ref so there is no race to 0, just
1309 * atomically increment the count.
1311 atomic_add_int(&sfm->mref_count, 1);
1315 sf_buf_mfree(void *arg)
1317 struct sfbuf_mref *sfm = arg;
1320 KKASSERT(sfm->mref_count > 0);
1321 if (sfm->mref_count == 1) {
1323 * We are the only holder so no further locking is required,
1324 * the sfbuf can simply be freed.
1326 sfm->mref_count = 0;
1330 * There may be other holders, we must obtain the serializer
1331 * to protect against a sf_buf_mfree() race to 0. An atomic
1332 * operation is still required for races against
1335 * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1337 lwkt_serialize_enter(&sfm->serializer);
1338 atomic_subtract_int(&sfm->mref_count, 1);
1339 if (sfm->mref_count == 0) {
1340 lwkt_serialize_exit(&sfm->serializer);
1344 m = sf_buf_page(sfm->sf);
1345 sf_buf_free(sfm->sf);
1346 vm_page_unwire(m, 0);
1347 if (m->wire_count == 0 && m->object == NULL)
1348 vm_page_try_to_free(m);
1351 free(sfm, M_SENDFILE);
1353 lwkt_serialize_exit(&sfm->serializer);
1360 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1361 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1363 * Send a file specified by 'fd' and starting at 'offset' to a socket
1364 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1365 * nbytes == 0. Optionally add a header and/or trailer to the socket
1366 * output. If specified, write the total number of bytes sent into *sbytes.
1368 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1369 * the headers to count against the remaining bytes to be sent from
1370 * the file descriptor. We may wish to implement a compatibility syscall
1374 sys_sendfile(struct sendfile_args *uap)
1376 struct thread *td = curthread;
1377 struct proc *p = td->td_proc;
1379 struct vnode *vp = NULL;
1380 struct sf_hdtr hdtr;
1381 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1383 struct mbuf *mheader = NULL;
1384 off_t hdtr_size = 0, sbytes;
1385 int error, hbytes = 0, tbytes;
1390 * Do argument checking. Must be a regular file in, stream
1391 * type and connected socket out, positive offset.
1393 fp = holdfp(p->p_fd, uap->fd, FREAD);
1397 if (fp->f_type != DTYPE_VNODE) {
1401 vp = (struct vnode *)fp->f_data;
1406 * If specified, get the pointer to the sf_hdtr struct for
1407 * any headers/trailers.
1410 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1417 error = iovec_copyin(hdtr.headers, &iov, aiov,
1418 hdtr.hdr_cnt, &hbytes);
1422 auio.uio_iovcnt = hdtr.hdr_cnt;
1423 auio.uio_offset = 0;
1424 auio.uio_segflg = UIO_USERSPACE;
1425 auio.uio_rw = UIO_WRITE;
1427 auio.uio_resid = hbytes;
1429 mheader = m_uiomove(&auio);
1431 iovec_free(&iov, aiov);
1432 if (mheader == NULL)
1437 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1438 &sbytes, uap->flags);
1443 * Send trailers. Wimp out and use writev(2).
1445 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1446 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1447 hdtr.trl_cnt, &auio.uio_resid);
1451 auio.uio_iovcnt = hdtr.trl_cnt;
1452 auio.uio_offset = 0;
1453 auio.uio_segflg = UIO_USERSPACE;
1454 auio.uio_rw = UIO_WRITE;
1457 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1459 iovec_free(&iov, aiov);
1462 hdtr_size += tbytes; /* trailer bytes successfully sent */
1466 if (uap->sbytes != NULL) {
1467 sbytes += hdtr_size;
1468 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1476 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1477 struct mbuf *mheader, off_t *sbytes, int flags)
1479 struct thread *td = curthread;
1480 struct proc *p = td->td_proc;
1481 struct vm_object *obj;
1486 struct sfbuf_mref *sfm;
1492 if (vp->v_type != VREG) {
1496 if ((obj = vp->v_object) == NULL) {
1500 error = holdsock(p->p_fd, sfd, &fp);
1503 so = (struct socket *)fp->f_data;
1504 if (so->so_type != SOCK_STREAM) {
1508 if ((so->so_state & SS_ISCONNECTED) == 0) {
1519 * Protect against multiple writers to the socket.
1521 (void) sblock(&so->so_snd, M_WAITOK);
1524 * Loop through the pages in the file, starting with the requested
1525 * offset. Get a file page (do I/O if necessary), map the file page
1526 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1529 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1533 pindex = OFF_TO_IDX(off);
1536 * Calculate the amount to transfer. Not to exceed a page,
1537 * the EOF, or the passed in nbytes.
1539 xfsize = vp->v_filesize - off;
1540 if (xfsize > PAGE_SIZE)
1542 pgoff = (vm_offset_t)(off & PAGE_MASK);
1543 if (PAGE_SIZE - pgoff < xfsize)
1544 xfsize = PAGE_SIZE - pgoff;
1545 if (nbytes && xfsize > (nbytes - *sbytes))
1546 xfsize = nbytes - *sbytes;
1550 * Optimize the non-blocking case by looking at the socket space
1551 * before going to the extra work of constituting the sf_buf.
1553 if ((fp->f_flag & FNONBLOCK) && sbspace(&so->so_snd) <= 0) {
1554 if (so->so_state & SS_CANTSENDMORE)
1558 sbunlock(&so->so_snd);
1562 * Attempt to look up the page.
1564 * Allocate if not found, wait and loop if busy, then
1565 * wire the page. critical section protection is
1566 * required to maintain the object association (an
1567 * interrupt can free the page) through to the
1568 * vm_page_wire() call.
1571 pg = vm_page_lookup(obj, pindex);
1573 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1580 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1588 * If page is not valid for what we need, initiate I/O
1591 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1597 * Ensure that our page is still around when the I/O
1600 vm_page_io_start(pg);
1603 * Get the page from backing store.
1605 bsize = vp->v_mount->mnt_stat.f_iosize;
1606 auio.uio_iov = &aiov;
1607 auio.uio_iovcnt = 1;
1609 aiov.iov_len = MAXBSIZE;
1610 auio.uio_resid = MAXBSIZE;
1611 auio.uio_offset = trunc_page(off);
1612 auio.uio_segflg = UIO_NOCOPY;
1613 auio.uio_rw = UIO_READ;
1615 vn_lock(vp, LK_SHARED | LK_RETRY);
1616 error = VOP_READ(vp, &auio,
1617 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1620 vm_page_flag_clear(pg, PG_ZERO);
1621 vm_page_io_finish(pg);
1624 vm_page_unwire(pg, 0);
1625 vm_page_try_to_free(pg);
1627 sbunlock(&so->so_snd);
1634 * Get a sendfile buf. We usually wait as long as necessary,
1635 * but this wait can be interrupted.
1637 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1639 vm_page_unwire(pg, 0);
1640 vm_page_try_to_free(pg);
1642 sbunlock(&so->so_snd);
1648 * Get an mbuf header and set it up as having external storage.
1650 MGETHDR(m, MB_WAIT, MT_DATA);
1654 sbunlock(&so->so_snd);
1659 * sfm is a temporary hack, use a per-cpu cache for this.
1661 sfm = malloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1663 sfm->mref_count = 1;
1664 lwkt_serialize_init(&sfm->serializer);
1666 m->m_ext.ext_free = sf_buf_mfree;
1667 m->m_ext.ext_ref = sf_buf_mref;
1668 m->m_ext.ext_arg = sfm;
1669 m->m_ext.ext_buf = (void *)sf->kva;
1670 m->m_ext.ext_size = PAGE_SIZE;
1671 m->m_data = (char *) sf->kva + pgoff;
1672 m->m_flags |= M_EXT;
1673 m->m_pkthdr.len = m->m_len = xfsize;
1674 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1676 if (mheader != NULL) {
1677 hbytes = mheader->m_pkthdr.len;
1678 mheader->m_pkthdr.len += m->m_pkthdr.len;
1686 * Add the buffer to the socket buffer chain.
1691 * Make sure that the socket is still able to take more data.
1692 * CANTSENDMORE being true usually means that the connection
1693 * was closed. so_error is true when an error was sensed after
1695 * The state is checked after the page mapping and buffer
1696 * allocation above since those operations may block and make
1697 * any socket checks stale. From this point forward, nothing
1698 * blocks before the pru_send (or more accurately, any blocking
1699 * results in a loop back to here to re-check).
1701 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1702 if (so->so_state & SS_CANTSENDMORE) {
1705 error = so->so_error;
1709 sbunlock(&so->so_snd);
1714 * Wait for socket space to become available. We do this just
1715 * after checking the connection state above in order to avoid
1716 * a race condition with sbwait().
1718 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1719 if (fp->f_flag & FNONBLOCK) {
1721 sbunlock(&so->so_snd);
1726 error = sbwait(&so->so_snd);
1728 * An error from sbwait usually indicates that we've
1729 * been interrupted by a signal. If we've sent anything
1730 * then return bytes sent, otherwise return the error.
1734 sbunlock(&so->so_snd);
1740 error = so_pru_send(so, 0, m, NULL, NULL, td);
1743 sbunlock(&so->so_snd);
1747 if (mheader != NULL) {
1748 *sbytes += mheader->m_pkthdr.len;
1749 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1752 sbunlock(&so->so_snd);
1757 if (mheader != NULL)
1763 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1766 struct thread *td = curthread;
1767 struct proc *p = td->td_proc;
1768 struct file *lfp = NULL;
1769 struct file *nfp = NULL;
1771 struct socket *head, *so;
1774 short fflag; /* type must match fp->f_flag */
1776 assoc_id = uap->name;
1777 error = holdsock(p->p_fd, uap->sd, &lfp);
1782 head = (struct socket *)lfp->f_data;
1783 error = sctp_can_peel_off(head, assoc_id);
1789 * At this point we know we do have a assoc to pull
1790 * we proceed to get the fd setup. This may block
1794 fflag = lfp->f_flag;
1795 error = falloc(p, &nfp, &fd);
1798 * Probably ran out of file descriptors. Put the
1799 * unaccepted connection back onto the queue and
1800 * do another wakeup so some other process might
1801 * have a chance at it.
1806 uap->sysmsg_result = fd;
1808 so = sctp_get_peeloff(head, assoc_id, &error);
1811 * Either someone else peeled it off OR
1812 * we can't get a socket.
1816 so->so_state &= ~SS_COMP;
1817 so->so_state &= ~SS_NOFDREF;
1819 if (head->so_sigio != NULL)
1820 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1822 nfp->f_type = DTYPE_SOCKET;
1823 nfp->f_flag = fflag;
1824 nfp->f_ops = &socketops;
1829 * Assign the file pointer to the reserved descriptor, or clear
1830 * the reserved descriptor if an error occured.
1833 fsetfd(p, NULL, fd);
1838 * Release explicitly held references before returning.