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 1.25 2004/03/04 10:29:23 hsu Exp $
41 #include "opt_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/sysproto.h>
47 #include <sys/malloc.h>
48 #include <sys/filedesc.h>
49 #include <sys/event.h>
51 #include <sys/fcntl.h>
53 #include <sys/filio.h>
54 #include <sys/kern_syscall.h>
56 #include <sys/protosw.h>
57 #include <sys/sfbuf.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/socketops.h>
62 #include <sys/vnode.h>
64 #include <sys/mount.h>
66 #include <sys/ktrace.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_pageout.h>
72 #include <vm/vm_kern.h>
73 #include <vm/vm_extern.h>
74 #include <sys/file2.h>
77 * System call interface to the socket abstraction.
80 extern struct fileops socketops;
83 * socket_args(int domain, int type, int protocol)
86 kern_socket(int domain, int type, int protocol, int *res)
88 struct thread *td = curthread;
89 struct proc *p = td->td_proc;
98 error = falloc(p, &fp, &fd);
102 error = socreate(domain, &so, type, protocol, td);
104 if (fdp->fd_ofiles[fd] == fp) {
105 fdp->fd_ofiles[fd] = NULL;
109 fp->f_data = (caddr_t)so;
110 fp->f_flag = FREAD|FWRITE;
111 fp->f_ops = &socketops;
112 fp->f_type = DTYPE_SOCKET;
120 socket(struct socket_args *uap)
124 error = kern_socket(uap->domain, uap->type, uap->protocol,
125 &uap->sysmsg_result);
131 kern_bind(int s, struct sockaddr *sa)
133 struct thread *td = curthread;
134 struct proc *p = td->td_proc;
139 error = holdsock(p->p_fd, s, &fp);
142 error = sobind((struct socket *)fp->f_data, sa, td);
148 * bind_args(int s, caddr_t name, int namelen)
151 bind(struct bind_args *uap)
156 error = getsockaddr(&sa, uap->name, uap->namelen);
159 error = kern_bind(uap->s, sa);
166 kern_listen(int s, int backlog)
168 struct thread *td = curthread;
169 struct proc *p = td->td_proc;
174 error = holdsock(p->p_fd, s, &fp);
177 error = solisten((struct socket *)fp->f_data, backlog, td);
183 * listen_args(int s, int backlog)
186 listen(struct listen_args *uap)
190 error = kern_listen(uap->s, uap->backlog);
195 * The second argument to kern_accept() is a handle to a struct sockaddr.
196 * This allows kern_accept() to return a pointer to an allocated struct
197 * sockaddr which must be freed later with FREE(). The caller must
198 * initialize *name to NULL.
201 kern_accept(int s, struct sockaddr **name, int *namelen, int *res)
203 struct thread *td = curthread;
204 struct proc *p = td->td_proc;
205 struct filedesc *fdp = p->p_fd;
206 struct file *lfp = NULL;
207 struct file *nfp = NULL;
210 struct socket *head, *so;
212 u_int fflag; /* type must match fp->f_flag */
215 if (name && namelen && *namelen < 0)
218 error = holdsock(fdp, s, &lfp);
222 head = (struct socket *)lfp->f_data;
223 if ((head->so_options & SO_ACCEPTCONN) == 0) {
228 while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) {
229 if (head->so_state & SS_CANTRCVMORE) {
230 head->so_error = ECONNABORTED;
233 if ((head->so_state & SS_NBIO) != 0) {
234 head->so_error = EWOULDBLOCK;
237 error = tsleep((caddr_t)&head->so_timeo, PCATCH, "accept", 0);
243 if (head->so_error) {
244 error = head->so_error;
251 * At this point we know that there is at least one connection
252 * ready to be accepted. Remove it from the queue prior to
253 * allocating the file descriptor for it since falloc() may
254 * block allowing another process to accept the connection
257 so = TAILQ_FIRST(&head->so_comp);
258 TAILQ_REMOVE(&head->so_comp, so, so_list);
262 error = falloc(p, &nfp, &fd);
265 * Probably ran out of file descriptors. Put the
266 * unaccepted connection back onto the queue and
267 * do another wakeup so some other process might
268 * have a chance at it.
270 TAILQ_INSERT_HEAD(&head->so_comp, so, so_list);
272 wakeup_one(&head->so_timeo);
279 /* connection has been removed from the listen queue */
280 KNOTE(&head->so_rcv.sb_sel.si_note, 0);
282 so->so_state &= ~SS_COMP;
284 if (head->so_sigio != NULL)
285 fsetown(fgetown(head->so_sigio), &so->so_sigio);
287 nfp->f_data = (caddr_t)so;
289 nfp->f_ops = &socketops;
290 nfp->f_type = DTYPE_SOCKET;
291 /* Sync socket nonblocking/async state with file flags */
292 tmp = fflag & FNONBLOCK;
293 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td);
294 tmp = fflag & FASYNC;
295 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td);
298 error = soaccept(so, &sa);
301 * Set the returned name and namelen as applicable. Set the returned
302 * namelen to 0 for older code which might ignore the return value
306 if (sa && name && namelen) {
307 if (*namelen > sa->sa_len)
308 *namelen = sa->sa_len;
317 * close the new descriptor, assuming someone hasn't ripped it
318 * out from under us. Note that *res is normally ignored if an
319 * error is returned but a syscall message will still have access
320 * to the result code.
324 if (fdp->fd_ofiles[fd] == nfp) {
325 fdp->fd_ofiles[fd] = NULL;
332 * Release explicitly held references before returning.
342 * accept_args(int s, caddr_t name, int *anamelen)
345 accept(struct accept_args *uap)
347 struct sockaddr *sa = NULL;
352 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
356 error = kern_accept(uap->s, &sa, &sa_len, &uap->sysmsg_result);
359 error = copyout(sa, uap->name, sa_len);
361 error = copyout(&sa_len, uap->anamelen,
362 sizeof(*uap->anamelen));
367 error = kern_accept(uap->s, NULL, 0, &uap->sysmsg_result);
373 kern_connect(int s, struct sockaddr *sa)
375 struct thread *td = curthread;
376 struct proc *p = td->td_proc;
381 error = holdsock(p->p_fd, s, &fp);
384 so = (struct socket *)fp->f_data;
385 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
389 error = soconnect(so, sa, td);
392 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
397 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
398 error = tsleep((caddr_t)&so->so_timeo, PCATCH, "connec", 0);
403 error = so->so_error;
408 so->so_state &= ~SS_ISCONNECTING;
409 if (error == ERESTART)
417 * connect_args(int s, caddr_t name, int namelen)
420 connect(struct connect_args *uap)
425 error = getsockaddr(&sa, uap->name, uap->namelen);
428 error = kern_connect(uap->s, sa);
435 kern_socketpair(int domain, int type, int protocol, int *sv)
437 struct thread *td = curthread;
438 struct proc *p = td->td_proc;
439 struct filedesc *fdp;
440 struct file *fp1, *fp2;
441 struct socket *so1, *so2;
446 error = socreate(domain, &so1, type, protocol, td);
449 error = socreate(domain, &so2, type, protocol, td);
452 error = falloc(p, &fp1, &fd);
457 fp1->f_data = (caddr_t)so1;
458 error = falloc(p, &fp2, &fd);
462 fp2->f_data = (caddr_t)so2;
464 error = soconnect2(so1, so2);
467 if (type == SOCK_DGRAM) {
469 * Datagram socket connection is asymmetric.
471 error = soconnect2(so2, so1);
475 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
476 fp1->f_ops = fp2->f_ops = &socketops;
477 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
482 if (fdp->fd_ofiles[sv[1]] == fp2) {
483 fdp->fd_ofiles[sv[1]] = NULL;
488 if (fdp->fd_ofiles[sv[0]] == fp1) {
489 fdp->fd_ofiles[sv[0]] = NULL;
501 * socketpair(int domain, int type, int protocol, int *rsv)
504 socketpair(struct socketpair_args *uap)
508 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
511 error = copyout(sockv, uap->rsv, sizeof(sockv));
516 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
517 struct mbuf *control, int flags, int *res)
519 struct thread *td = curthread;
520 struct proc *p = td->td_proc;
525 struct iovec *ktriov = NULL;
529 error = holdsock(p->p_fd, s, &fp);
532 if (auio->uio_resid < 0) {
537 if (KTRPOINT(td, KTR_GENIO)) {
538 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
540 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
541 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
545 len = auio->uio_resid;
546 so = (struct socket *)fp->f_data;
547 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
549 if (auio->uio_resid != len && (error == ERESTART ||
550 error == EINTR || error == EWOULDBLOCK))
556 if (ktriov != NULL) {
558 ktruio.uio_iov = ktriov;
559 ktruio.uio_resid = len - auio->uio_resid;
560 ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error);
562 FREE(ktriov, M_TEMP);
566 *res = len - auio->uio_resid;
573 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
576 sendto(struct sendto_args *uap)
578 struct thread *td = curthread;
581 struct sockaddr *sa = NULL;
585 error = getsockaddr(&sa, uap->to, uap->tolen);
589 aiov.iov_base = uap->buf;
590 aiov.iov_len = uap->len;
591 auio.uio_iov = &aiov;
594 auio.uio_resid = uap->len;
595 auio.uio_segflg = UIO_USERSPACE;
596 auio.uio_rw = UIO_WRITE;
599 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
600 &uap->sysmsg_result);
608 * sendmsg_args(int s, caddr_t msg, int flags)
611 sendmsg(struct sendmsg_args *uap)
613 struct thread *td = curthread;
616 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
617 struct sockaddr *sa = NULL;
618 struct mbuf *control = NULL;
621 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
626 * Conditionally copyin msg.msg_name.
629 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
637 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
642 auio.uio_iovcnt = msg.msg_iovlen;
644 auio.uio_segflg = UIO_USERSPACE;
645 auio.uio_rw = UIO_WRITE;
649 * Conditionally copyin msg.msg_control.
651 if (msg.msg_control) {
652 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
653 msg.msg_controllen > MLEN) {
657 control = m_get(M_WAIT, MT_CONTROL);
658 if (control == NULL) {
662 control->m_len = msg.msg_controllen;
663 error = copyin(msg.msg_control, mtod(control, caddr_t),
671 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
672 &uap->sysmsg_result);
677 iovec_free(&iov, aiov);
682 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
683 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
684 * Don't forget to FREE() and m_free() these if they are returned.
687 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
688 struct mbuf **control, int *flags, int *res)
690 struct thread *td = curthread;
691 struct proc *p = td->td_proc;
696 struct iovec *ktriov = NULL;
700 error = holdsock(p->p_fd, s, &fp);
703 if (auio->uio_resid < 0) {
708 if (KTRPOINT(td, KTR_GENIO)) {
709 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
711 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
712 bcopy(auio->uio_iov, ktriov, iovlen);
716 len = auio->uio_resid;
717 so = (struct socket *)fp->f_data;
718 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
720 if (auio->uio_resid != len && (error == ERESTART ||
721 error == EINTR || error == EWOULDBLOCK))
725 if (ktriov != NULL) {
727 ktruio.uio_iov = ktriov;
728 ktruio.uio_resid = len - auio->uio_resid;
729 ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error);
731 FREE(ktriov, M_TEMP);
735 *res = len - auio->uio_resid;
742 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
743 * caddr_t from, int *fromlenaddr)
746 recvfrom(struct recvfrom_args *uap)
748 struct thread *td = curthread;
751 struct sockaddr *sa = NULL;
754 if (uap->from && uap->fromlenaddr) {
755 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
763 aiov.iov_base = uap->buf;
764 aiov.iov_len = uap->len;
765 auio.uio_iov = &aiov;
768 auio.uio_resid = uap->len;
769 auio.uio_segflg = UIO_USERSPACE;
770 auio.uio_rw = UIO_READ;
773 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
774 &uap->flags, &uap->sysmsg_result);
776 if (error == 0 && uap->from) {
777 /* note: sa may still be NULL */
779 fromlen = MIN(fromlen, sa->sa_len);
780 error = copyout(sa, uap->from, fromlen);
785 error = copyout(&fromlen, uap->fromlenaddr,
796 * recvmsg_args(int s, struct msghdr *msg, int flags)
799 recvmsg(struct recvmsg_args *uap)
801 struct thread *td = curthread;
804 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
805 struct mbuf *m, *control = NULL;
806 struct sockaddr *sa = NULL;
808 socklen_t *ufromlenp, *ucontrollenp;
809 int error, fromlen, controllen, len, flags, *uflagsp;
812 * This copyin handles everything except the iovec.
814 error = copyin(uap->msg, &msg, sizeof(msg));
818 if (msg.msg_name && msg.msg_namelen < 0)
820 if (msg.msg_control && msg.msg_controllen < 0)
823 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
825 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
827 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
833 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
838 auio.uio_iovcnt = msg.msg_iovlen;
840 auio.uio_segflg = UIO_USERSPACE;
841 auio.uio_rw = UIO_READ;
844 flags = msg.msg_flags;
846 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
847 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
850 * Conditionally copyout the name and populate the namelen field.
852 if (error == 0 && msg.msg_name) {
853 fromlen = MIN(msg.msg_namelen, sa->sa_len);
854 error = copyout(sa, msg.msg_name, fromlen);
856 error = copyout(&fromlen, ufromlenp,
861 * Copyout msg.msg_control and msg.msg_controllen.
863 if (error == 0 && msg.msg_control) {
864 len = msg.msg_controllen;
866 ctlbuf = (caddr_t)msg.msg_control;
868 while(m && len > 0) {
871 if (len >= m->m_len) {
874 msg.msg_flags |= MSG_CTRUNC;
878 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
886 controllen = ctlbuf - (caddr_t)msg.msg_control;
887 error = copyout(&controllen, ucontrollenp,
888 sizeof(*ucontrollenp));
892 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
897 iovec_free(&iov, aiov);
904 * shutdown_args(int s, int how)
907 kern_shutdown(int s, int how)
909 struct thread *td = curthread;
910 struct proc *p = td->td_proc;
915 error = holdsock(p->p_fd, s, &fp);
918 error = soshutdown((struct socket *)fp->f_data, how);
924 shutdown(struct shutdown_args *uap)
928 error = kern_shutdown(uap->s, uap->how);
934 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
935 * in kernel pointer instead of a userland pointer. This allows us
936 * to manipulate socket options in the emulation code.
939 kern_setsockopt(int s, struct sockopt *sopt)
941 struct thread *td = curthread;
942 struct proc *p = td->td_proc;
946 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
948 if (sopt->sopt_valsize < 0)
951 error = holdsock(p->p_fd, s, &fp);
955 error = sosetopt((struct socket *)fp->f_data, sopt);
961 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
964 setsockopt(struct setsockopt_args *uap)
966 struct thread *td = curthread;
970 sopt.sopt_dir = SOPT_SET;
971 sopt.sopt_level = uap->level;
972 sopt.sopt_name = uap->name;
973 sopt.sopt_val = uap->val;
974 sopt.sopt_valsize = uap->valsize;
977 error = kern_setsockopt(uap->s, &sopt);
982 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
983 * in kernel pointer instead of a userland pointer. This allows us
984 * to manipulate socket options in the emulation code.
987 kern_getsockopt(int s, struct sockopt *sopt)
989 struct thread *td = curthread;
990 struct proc *p = td->td_proc;
994 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
996 if (sopt->sopt_valsize < 0)
999 error = holdsock(p->p_fd, s, &fp);
1003 error = sogetopt((struct socket *)fp->f_data, sopt);
1009 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1012 getsockopt(struct getsockopt_args *uap)
1014 struct thread *td = curthread;
1015 struct sockopt sopt;
1019 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1028 sopt.sopt_dir = SOPT_GET;
1029 sopt.sopt_level = uap->level;
1030 sopt.sopt_name = uap->name;
1031 sopt.sopt_val = uap->val;
1032 sopt.sopt_valsize = valsize;
1035 error = kern_getsockopt(uap->s, &sopt);
1037 valsize = sopt.sopt_valsize;
1038 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1044 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1045 * This allows kern_getsockname() to return a pointer to an allocated struct
1046 * sockaddr which must be freed later with FREE(). The caller must
1047 * initialize *name to NULL.
1050 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1052 struct thread *td = curthread;
1053 struct proc *p = td->td_proc;
1056 struct sockaddr *sa = NULL;
1059 error = holdsock(p->p_fd, s, &fp);
1066 so = (struct socket *)fp->f_data;
1067 error = so_pru_sockaddr(so, &sa);
1072 *namelen = MIN(*namelen, sa->sa_len);
1082 * getsockname_args(int fdes, caddr_t asa, int *alen)
1087 getsockname(struct getsockname_args *uap)
1089 struct sockaddr *sa = NULL;
1092 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1096 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1099 error = copyout(sa, uap->asa, sa_len);
1101 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1108 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1109 * This allows kern_getpeername() to return a pointer to an allocated struct
1110 * sockaddr which must be freed later with FREE(). The caller must
1111 * initialize *name to NULL.
1114 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1116 struct thread *td = curthread;
1117 struct proc *p = td->td_proc;
1120 struct sockaddr *sa = NULL;
1123 error = holdsock(p->p_fd, s, &fp);
1130 so = (struct socket *)fp->f_data;
1131 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1135 error = so_pru_peeraddr(so, &sa);
1140 *namelen = MIN(*namelen, sa->sa_len);
1150 * getpeername_args(int fdes, caddr_t asa, int *alen)
1152 * Get name of peer for connected socket.
1155 getpeername(struct getpeername_args *uap)
1157 struct sockaddr *sa = NULL;
1160 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1164 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1167 error = copyout(sa, uap->asa, sa_len);
1169 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1176 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1178 struct sockaddr *sa;
1182 if (len > SOCK_MAXADDRLEN)
1183 return ENAMETOOLONG;
1184 if (len < offsetof(struct sockaddr, sa_data[0]))
1186 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1187 error = copyin(uaddr, sa, len);
1191 #if BYTE_ORDER != BIG_ENDIAN
1193 * The bind(), connect(), and sendto() syscalls were not
1194 * versioned for COMPAT_43. Thus, this check must stay.
1196 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1197 sa->sa_family = sa->sa_len;
1206 * holdsock() - load the struct file pointer associated
1207 * with a socket into *fpp. If an error occurs, non-zero
1208 * will be returned and *fpp will be set to NULL.
1211 holdsock(fdp, fdes, fpp)
1212 struct filedesc *fdp;
1216 struct file *fp = NULL;
1219 if ((unsigned)fdes >= fdp->fd_nfiles ||
1220 (fp = fdp->fd_ofiles[fdes]) == NULL) {
1222 } else if (fp->f_type != DTYPE_SOCKET) {
1234 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1235 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1237 * Send a file specified by 'fd' and starting at 'offset' to a socket
1238 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1239 * nbytes == 0. Optionally add a header and/or trailer to the socket
1240 * output. If specified, write the total number of bytes sent into *sbytes.
1242 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1243 * the headers to count against the remaining bytes to be sent from
1244 * the file descriptor. We may wish to implement a compatibility syscall
1248 sendfile(struct sendfile_args *uap)
1250 struct thread *td = curthread;
1251 struct proc *p = td->td_proc;
1253 struct filedesc *fdp;
1254 struct vnode *vp = NULL;
1255 struct sf_hdtr hdtr;
1256 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1258 off_t hdtr_size = 0, sbytes;
1265 * Do argument checking. Must be a regular file in, stream
1266 * type and connected socket out, positive offset.
1268 fp = holdfp(fdp, uap->fd, FREAD);
1272 if (fp->f_type != DTYPE_VNODE) {
1276 vp = (struct vnode *)fp->f_data;
1281 * If specified, get the pointer to the sf_hdtr struct for
1282 * any headers/trailers.
1285 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1292 error = iovec_copyin(hdtr.headers, &iov, aiov,
1293 hdtr.hdr_cnt, &auio.uio_resid);
1297 auio.uio_iovcnt = hdtr.hdr_cnt;
1298 auio.uio_offset = 0;
1299 auio.uio_segflg = UIO_USERSPACE;
1300 auio.uio_rw = UIO_WRITE;
1303 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0,
1306 iovec_free(&iov, aiov);
1313 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes,
1314 &sbytes, uap->flags);
1319 * Send trailers. Wimp out and use writev(2).
1321 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1322 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1323 hdtr.trl_cnt, &auio.uio_resid);
1327 auio.uio_iovcnt = hdtr.trl_cnt;
1328 auio.uio_offset = 0;
1329 auio.uio_segflg = UIO_USERSPACE;
1330 auio.uio_rw = UIO_WRITE;
1333 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &res);
1335 iovec_free(&iov, aiov);
1342 if (uap->sbytes != NULL) {
1343 sbytes += hdtr_size;
1344 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1352 kern_sendfile(struct vnode *vp, int s, off_t offset, size_t nbytes,
1353 off_t *sbytes, int flags)
1355 struct thread *td = curthread;
1356 struct proc *p = td->td_proc;
1357 struct vm_object *obj;
1366 if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) {
1370 error = holdsock(p->p_fd, s, &fp);
1373 so = (struct socket *)fp->f_data;
1374 if (so->so_type != SOCK_STREAM) {
1378 if ((so->so_state & SS_ISCONNECTED) == 0) {
1389 * Protect against multiple writers to the socket.
1391 (void) sblock(&so->so_snd, M_WAITOK);
1394 * Loop through the pages in the file, starting with the requested
1395 * offset. Get a file page (do I/O if necessary), map the file page
1396 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1399 for (off = offset; ; off += xfsize, *sbytes += xfsize) {
1403 pindex = OFF_TO_IDX(off);
1406 * Calculate the amount to transfer. Not to exceed a page,
1407 * the EOF, or the passed in nbytes.
1409 xfsize = obj->un_pager.vnp.vnp_size - off;
1410 if (xfsize > PAGE_SIZE)
1412 pgoff = (vm_offset_t)(off & PAGE_MASK);
1413 if (PAGE_SIZE - pgoff < xfsize)
1414 xfsize = PAGE_SIZE - pgoff;
1415 if (nbytes && xfsize > (nbytes - *sbytes))
1416 xfsize = nbytes - *sbytes;
1420 * Optimize the non-blocking case by looking at the socket space
1421 * before going to the extra work of constituting the sf_buf.
1423 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) {
1424 if (so->so_state & SS_CANTSENDMORE)
1428 sbunlock(&so->so_snd);
1432 * Attempt to look up the page.
1434 * Allocate if not found
1436 * Wait and loop if busy.
1438 pg = vm_page_lookup(obj, pindex);
1441 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1447 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1452 * Wire the page so it does not get ripped out from under
1459 * If page is not valid for what we need, initiate I/O
1462 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1468 * Ensure that our page is still around when the I/O
1471 vm_page_io_start(pg);
1474 * Get the page from backing store.
1476 bsize = vp->v_mount->mnt_stat.f_iosize;
1477 auio.uio_iov = &aiov;
1478 auio.uio_iovcnt = 1;
1480 aiov.iov_len = MAXBSIZE;
1481 auio.uio_resid = MAXBSIZE;
1482 auio.uio_offset = trunc_page(off);
1483 auio.uio_segflg = UIO_NOCOPY;
1484 auio.uio_rw = UIO_READ;
1486 vn_lock(vp, NULL, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
1487 error = VOP_READ(vp, &auio,
1488 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1490 VOP_UNLOCK(vp, NULL, 0, td);
1491 vm_page_flag_clear(pg, PG_ZERO);
1492 vm_page_io_finish(pg);
1494 vm_page_unwire(pg, 0);
1496 * See if anyone else might know about this page.
1497 * If not and it is not valid, then free it.
1499 if (pg->wire_count == 0 && pg->valid == 0 &&
1500 pg->busy == 0 && !(pg->flags & PG_BUSY) &&
1501 pg->hold_count == 0) {
1505 sbunlock(&so->so_snd);
1512 * Get a sendfile buf. We usually wait as long as necessary,
1513 * but this wait can be interrupted.
1515 if ((sf = sf_buf_alloc(pg)) == NULL) {
1517 vm_page_unwire(pg, 0);
1518 if (pg->wire_count == 0 && pg->object == NULL)
1521 sbunlock(&so->so_snd);
1527 * Get an mbuf header and set it up as having external storage.
1529 MGETHDR(m, M_WAIT, MT_DATA);
1532 sf_buf_free((void *)sf->kva, PAGE_SIZE);
1533 sbunlock(&so->so_snd);
1536 m->m_ext.ext_free = sf_buf_free;
1537 m->m_ext.ext_ref = sf_buf_ref;
1538 m->m_ext.ext_buf = (void *)sf->kva;
1539 m->m_ext.ext_size = PAGE_SIZE;
1540 m->m_data = (char *) sf->kva + pgoff;
1541 m->m_flags |= M_EXT;
1542 m->m_pkthdr.len = m->m_len = xfsize;
1544 * Add the buffer to the socket buffer chain.
1549 * Make sure that the socket is still able to take more data.
1550 * CANTSENDMORE being true usually means that the connection
1551 * was closed. so_error is true when an error was sensed after
1553 * The state is checked after the page mapping and buffer
1554 * allocation above since those operations may block and make
1555 * any socket checks stale. From this point forward, nothing
1556 * blocks before the pru_send (or more accurately, any blocking
1557 * results in a loop back to here to re-check).
1559 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1560 if (so->so_state & SS_CANTSENDMORE) {
1563 error = so->so_error;
1567 sbunlock(&so->so_snd);
1572 * Wait for socket space to become available. We do this just
1573 * after checking the connection state above in order to avoid
1574 * a race condition with sbwait().
1576 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1577 if (so->so_state & SS_NBIO) {
1579 sbunlock(&so->so_snd);
1584 error = sbwait(&so->so_snd);
1586 * An error from sbwait usually indicates that we've
1587 * been interrupted by a signal. If we've sent anything
1588 * then return bytes sent, otherwise return the error.
1592 sbunlock(&so->so_snd);
1598 error = so_pru_send(so, NULL, m, NULL, NULL, td);
1601 sbunlock(&so->so_snd);
1605 sbunlock(&so->so_snd);