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 1.41 2004/08/24 21:53:38 dillon 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>
75 #include <sys/signalvar.h>
77 #include <sys/thread2.h>
78 #include <sys/msgport2.h>
85 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures");
88 * System call interface to the socket abstraction.
91 extern struct fileops socketops;
94 * socket_args(int domain, int type, int protocol)
97 kern_socket(int domain, int type, int protocol, int *res)
99 struct thread *td = curthread;
100 struct proc *p = td->td_proc;
101 struct filedesc *fdp;
109 error = falloc(p, &fp, &fd);
113 error = socreate(domain, &so, type, protocol, td);
115 if (fdp->fd_ofiles[fd] == fp) {
116 fdp->fd_ofiles[fd] = NULL;
120 fp->f_data = (caddr_t)so;
121 fp->f_flag = FREAD|FWRITE;
122 fp->f_ops = &socketops;
123 fp->f_type = DTYPE_SOCKET;
131 socket(struct socket_args *uap)
135 error = kern_socket(uap->domain, uap->type, uap->protocol,
136 &uap->sysmsg_result);
142 kern_bind(int s, struct sockaddr *sa)
144 struct thread *td = curthread;
145 struct proc *p = td->td_proc;
150 error = holdsock(p->p_fd, s, &fp);
153 error = sobind((struct socket *)fp->f_data, sa, td);
159 * bind_args(int s, caddr_t name, int namelen)
162 bind(struct bind_args *uap)
167 error = getsockaddr(&sa, uap->name, uap->namelen);
170 error = kern_bind(uap->s, sa);
177 kern_listen(int s, int backlog)
179 struct thread *td = curthread;
180 struct proc *p = td->td_proc;
185 error = holdsock(p->p_fd, s, &fp);
188 error = solisten((struct socket *)fp->f_data, backlog, td);
194 * listen_args(int s, int backlog)
197 listen(struct listen_args *uap)
201 error = kern_listen(uap->s, uap->backlog);
206 * Returns the accepted socket as well.
209 soaccept_predicate(struct netmsg *msg0)
211 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
212 struct socket *head = msg->nm_so;
214 if (head->so_error != 0) {
215 msg->nm_lmsg.ms_error = head->so_error;
218 if (!TAILQ_EMPTY(&head->so_comp)) {
219 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
220 msg->nm_so = TAILQ_FIRST(&head->so_comp);
221 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
224 msg->nm_lmsg.ms_error = 0;
227 if (head->so_state & SS_CANTRCVMORE) {
228 msg->nm_lmsg.ms_error = ECONNABORTED;
231 if (head->so_state & SS_NBIO) {
232 msg->nm_lmsg.ms_error = EWOULDBLOCK;
240 * The second argument to kern_accept() is a handle to a struct sockaddr.
241 * This allows kern_accept() to return a pointer to an allocated struct
242 * sockaddr which must be freed later with FREE(). The caller must
243 * initialize *name to NULL.
246 kern_accept(int s, struct sockaddr **name, int *namelen, int *res)
248 struct thread *td = curthread;
249 struct proc *p = td->td_proc;
250 struct filedesc *fdp = p->p_fd;
251 struct file *lfp = NULL;
252 struct file *nfp = NULL;
254 struct socket *head, *so;
255 struct netmsg_so_notify msg;
258 u_int fflag; /* type must match fp->f_flag */
261 if (name && namelen && *namelen < 0)
264 error = holdsock(fdp, s, &lfp);
268 error = falloc(p, &nfp, &fd);
269 if (error) { /* Probably ran out of file descriptors. */
277 head = (struct socket *)lfp->f_data;
278 if ((head->so_options & SO_ACCEPTCONN) == 0) {
283 /* optimize for uniprocessor case later XXX JH */
284 port = head->so_proto->pr_mport(head, NULL, PRU_PRED);
285 lwkt_initmsg(&msg.nm_lmsg, &curthread->td_msgport,
286 MSGF_PCATCH | MSGF_ABORTABLE,
287 lwkt_cmd_func(netmsg_so_notify),
288 lwkt_cmd_func(netmsg_so_notify_abort));
289 msg.nm_predicate = soaccept_predicate;
291 msg.nm_etype = NM_REVENT;
292 error = lwkt_domsg(port, &msg.nm_lmsg);
297 * At this point we have the connection that's ready to be accepted.
303 /* connection has been removed from the listen queue */
304 KNOTE(&head->so_rcv.sb_sel.si_note, 0);
306 so->so_state &= ~SS_COMP;
308 if (head->so_sigio != NULL)
309 fsetown(fgetown(head->so_sigio), &so->so_sigio);
311 nfp->f_data = (caddr_t)so;
313 nfp->f_ops = &socketops;
314 nfp->f_type = DTYPE_SOCKET;
315 /* Sync socket nonblocking/async state with file flags */
316 tmp = fflag & FNONBLOCK;
317 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td);
318 tmp = fflag & FASYNC;
319 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td);
322 error = soaccept(so, &sa);
325 * Set the returned name and namelen as applicable. Set the returned
326 * namelen to 0 for older code which might ignore the return value
330 if (sa && name && namelen) {
331 if (*namelen > sa->sa_len)
332 *namelen = sa->sa_len;
342 * close the new descriptor, assuming someone hasn't ripped it
343 * out from under us. Note that *res is normally ignored if an
344 * error is returned but a syscall message will still have access
345 * to the result code.
349 if (fdp->fd_ofiles[fd] == nfp) {
350 fdp->fd_ofiles[fd] = NULL;
356 * Release explicitly held references before returning.
365 * accept_args(int s, caddr_t name, int *anamelen)
368 accept(struct accept_args *uap)
370 struct sockaddr *sa = NULL;
375 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
379 error = kern_accept(uap->s, &sa, &sa_len, &uap->sysmsg_result);
382 error = copyout(sa, uap->name, sa_len);
384 error = copyout(&sa_len, uap->anamelen,
385 sizeof(*uap->anamelen));
390 error = kern_accept(uap->s, NULL, 0, &uap->sysmsg_result);
396 * Returns TRUE if predicate satisfied.
399 soconnected_predicate(struct netmsg *msg0)
401 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
402 struct socket *so = msg->nm_so;
404 /* check predicate */
405 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
406 msg->nm_lmsg.ms_error = so->so_error;
414 kern_connect(int s, struct sockaddr *sa)
416 struct thread *td = curthread;
417 struct proc *p = td->td_proc;
422 error = holdsock(p->p_fd, s, &fp);
425 so = (struct socket *)fp->f_data;
426 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
430 error = soconnect(so, sa, td);
433 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
437 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
438 struct netmsg_so_notify msg;
441 port = so->so_proto->pr_mport(so, sa, PRU_PRED);
442 lwkt_initmsg(&msg.nm_lmsg,
443 &curthread->td_msgport,
444 MSGF_PCATCH | MSGF_ABORTABLE,
445 lwkt_cmd_func(netmsg_so_notify),
446 lwkt_cmd_func(netmsg_so_notify_abort));
447 msg.nm_predicate = soconnected_predicate;
449 msg.nm_etype = NM_REVENT;
450 error = lwkt_domsg(port, &msg.nm_lmsg);
453 error = so->so_error;
457 so->so_state &= ~SS_ISCONNECTING;
458 if (error == ERESTART)
466 * connect_args(int s, caddr_t name, int namelen)
469 connect(struct connect_args *uap)
474 error = getsockaddr(&sa, uap->name, uap->namelen);
477 error = kern_connect(uap->s, sa);
484 kern_socketpair(int domain, int type, int protocol, int *sv)
486 struct thread *td = curthread;
487 struct proc *p = td->td_proc;
488 struct filedesc *fdp;
489 struct file *fp1, *fp2;
490 struct socket *so1, *so2;
495 error = socreate(domain, &so1, type, protocol, td);
498 error = socreate(domain, &so2, type, protocol, td);
501 error = falloc(p, &fp1, &fd);
506 fp1->f_data = (caddr_t)so1;
507 error = falloc(p, &fp2, &fd);
511 fp2->f_data = (caddr_t)so2;
513 error = soconnect2(so1, so2);
516 if (type == SOCK_DGRAM) {
518 * Datagram socket connection is asymmetric.
520 error = soconnect2(so2, so1);
524 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
525 fp1->f_ops = fp2->f_ops = &socketops;
526 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
531 if (fdp->fd_ofiles[sv[1]] == fp2) {
532 fdp->fd_ofiles[sv[1]] = NULL;
537 if (fdp->fd_ofiles[sv[0]] == fp1) {
538 fdp->fd_ofiles[sv[0]] = NULL;
550 * socketpair(int domain, int type, int protocol, int *rsv)
553 socketpair(struct socketpair_args *uap)
557 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
560 error = copyout(sockv, uap->rsv, sizeof(sockv));
565 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
566 struct mbuf *control, int flags, int *res)
568 struct thread *td = curthread;
569 struct proc *p = td->td_proc;
574 struct iovec *ktriov = NULL;
578 error = holdsock(p->p_fd, s, &fp);
581 if (auio->uio_resid < 0) {
586 if (KTRPOINT(td, KTR_GENIO)) {
587 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
589 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
590 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
594 len = auio->uio_resid;
595 so = (struct socket *)fp->f_data;
596 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
598 if (auio->uio_resid != len && (error == ERESTART ||
599 error == EINTR || error == EWOULDBLOCK))
605 if (ktriov != NULL) {
607 ktruio.uio_iov = ktriov;
608 ktruio.uio_resid = len - auio->uio_resid;
609 ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error);
611 FREE(ktriov, M_TEMP);
615 *res = len - auio->uio_resid;
622 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
625 sendto(struct sendto_args *uap)
627 struct thread *td = curthread;
630 struct sockaddr *sa = NULL;
634 error = getsockaddr(&sa, uap->to, uap->tolen);
638 aiov.iov_base = uap->buf;
639 aiov.iov_len = uap->len;
640 auio.uio_iov = &aiov;
643 auio.uio_resid = uap->len;
644 auio.uio_segflg = UIO_USERSPACE;
645 auio.uio_rw = UIO_WRITE;
648 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
649 &uap->sysmsg_result);
657 * sendmsg_args(int s, caddr_t msg, int flags)
660 sendmsg(struct sendmsg_args *uap)
662 struct thread *td = curthread;
665 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
666 struct sockaddr *sa = NULL;
667 struct mbuf *control = NULL;
670 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
675 * Conditionally copyin msg.msg_name.
678 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
686 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
691 auio.uio_iovcnt = msg.msg_iovlen;
693 auio.uio_segflg = UIO_USERSPACE;
694 auio.uio_rw = UIO_WRITE;
698 * Conditionally copyin msg.msg_control.
700 if (msg.msg_control) {
701 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
702 msg.msg_controllen > MLEN) {
706 control = m_get(MB_WAIT, MT_CONTROL);
707 if (control == NULL) {
711 control->m_len = msg.msg_controllen;
712 error = copyin(msg.msg_control, mtod(control, caddr_t),
720 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
721 &uap->sysmsg_result);
726 iovec_free(&iov, aiov);
731 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
732 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
733 * Don't forget to FREE() and m_free() these if they are returned.
736 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
737 struct mbuf **control, int *flags, int *res)
739 struct thread *td = curthread;
740 struct proc *p = td->td_proc;
745 struct iovec *ktriov = NULL;
749 error = holdsock(p->p_fd, s, &fp);
752 if (auio->uio_resid < 0) {
757 if (KTRPOINT(td, KTR_GENIO)) {
758 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
760 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
761 bcopy(auio->uio_iov, ktriov, iovlen);
765 len = auio->uio_resid;
766 so = (struct socket *)fp->f_data;
767 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
769 if (auio->uio_resid != len && (error == ERESTART ||
770 error == EINTR || error == EWOULDBLOCK))
774 if (ktriov != NULL) {
776 ktruio.uio_iov = ktriov;
777 ktruio.uio_resid = len - auio->uio_resid;
778 ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error);
780 FREE(ktriov, M_TEMP);
784 *res = len - auio->uio_resid;
791 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
792 * caddr_t from, int *fromlenaddr)
795 recvfrom(struct recvfrom_args *uap)
797 struct thread *td = curthread;
800 struct sockaddr *sa = NULL;
803 if (uap->from && uap->fromlenaddr) {
804 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
812 aiov.iov_base = uap->buf;
813 aiov.iov_len = uap->len;
814 auio.uio_iov = &aiov;
817 auio.uio_resid = uap->len;
818 auio.uio_segflg = UIO_USERSPACE;
819 auio.uio_rw = UIO_READ;
822 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
823 &uap->flags, &uap->sysmsg_result);
825 if (error == 0 && uap->from) {
826 /* note: sa may still be NULL */
828 fromlen = MIN(fromlen, sa->sa_len);
829 error = copyout(sa, uap->from, fromlen);
834 error = copyout(&fromlen, uap->fromlenaddr,
845 * recvmsg_args(int s, struct msghdr *msg, int flags)
848 recvmsg(struct recvmsg_args *uap)
850 struct thread *td = curthread;
853 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
854 struct mbuf *m, *control = NULL;
855 struct sockaddr *sa = NULL;
857 socklen_t *ufromlenp, *ucontrollenp;
858 int error, fromlen, controllen, len, flags, *uflagsp;
861 * This copyin handles everything except the iovec.
863 error = copyin(uap->msg, &msg, sizeof(msg));
867 if (msg.msg_name && msg.msg_namelen < 0)
869 if (msg.msg_control && msg.msg_controllen < 0)
872 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
874 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
876 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
882 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
887 auio.uio_iovcnt = msg.msg_iovlen;
889 auio.uio_segflg = UIO_USERSPACE;
890 auio.uio_rw = UIO_READ;
893 flags = msg.msg_flags;
895 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
896 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
899 * Conditionally copyout the name and populate the namelen field.
901 if (error == 0 && msg.msg_name) {
902 fromlen = MIN(msg.msg_namelen, sa->sa_len);
903 error = copyout(sa, msg.msg_name, fromlen);
905 error = copyout(&fromlen, ufromlenp,
910 * Copyout msg.msg_control and msg.msg_controllen.
912 if (error == 0 && msg.msg_control) {
913 len = msg.msg_controllen;
915 ctlbuf = (caddr_t)msg.msg_control;
917 while(m && len > 0) {
920 if (len >= m->m_len) {
923 msg.msg_flags |= MSG_CTRUNC;
927 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
935 controllen = ctlbuf - (caddr_t)msg.msg_control;
936 error = copyout(&controllen, ucontrollenp,
937 sizeof(*ucontrollenp));
941 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
946 iovec_free(&iov, aiov);
953 * shutdown_args(int s, int how)
956 kern_shutdown(int s, int how)
958 struct thread *td = curthread;
959 struct proc *p = td->td_proc;
964 error = holdsock(p->p_fd, s, &fp);
967 error = soshutdown((struct socket *)fp->f_data, how);
973 shutdown(struct shutdown_args *uap)
977 error = kern_shutdown(uap->s, uap->how);
983 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
984 * in kernel pointer instead of a userland pointer. This allows us
985 * to manipulate socket options in the emulation code.
988 kern_setsockopt(int s, struct sockopt *sopt)
990 struct thread *td = curthread;
991 struct proc *p = td->td_proc;
995 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
997 if (sopt->sopt_valsize < 0)
1000 error = holdsock(p->p_fd, s, &fp);
1004 error = sosetopt((struct socket *)fp->f_data, sopt);
1010 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1013 setsockopt(struct setsockopt_args *uap)
1015 struct thread *td = curthread;
1016 struct sockopt sopt;
1019 sopt.sopt_dir = SOPT_SET;
1020 sopt.sopt_level = uap->level;
1021 sopt.sopt_name = uap->name;
1022 sopt.sopt_val = uap->val;
1023 sopt.sopt_valsize = uap->valsize;
1026 error = kern_setsockopt(uap->s, &sopt);
1031 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1032 * in kernel pointer instead of a userland pointer. This allows us
1033 * to manipulate socket options in the emulation code.
1036 kern_getsockopt(int s, struct sockopt *sopt)
1038 struct thread *td = curthread;
1039 struct proc *p = td->td_proc;
1043 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1045 if (sopt->sopt_valsize < 0)
1048 error = holdsock(p->p_fd, s, &fp);
1052 error = sogetopt((struct socket *)fp->f_data, sopt);
1058 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1061 getsockopt(struct getsockopt_args *uap)
1063 struct thread *td = curthread;
1064 struct sockopt sopt;
1068 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1077 sopt.sopt_dir = SOPT_GET;
1078 sopt.sopt_level = uap->level;
1079 sopt.sopt_name = uap->name;
1080 sopt.sopt_val = uap->val;
1081 sopt.sopt_valsize = valsize;
1084 error = kern_getsockopt(uap->s, &sopt);
1086 valsize = sopt.sopt_valsize;
1087 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1093 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1094 * This allows kern_getsockname() to return a pointer to an allocated struct
1095 * sockaddr which must be freed later with FREE(). The caller must
1096 * initialize *name to NULL.
1099 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1101 struct thread *td = curthread;
1102 struct proc *p = td->td_proc;
1105 struct sockaddr *sa = NULL;
1108 error = holdsock(p->p_fd, s, &fp);
1115 so = (struct socket *)fp->f_data;
1116 error = so_pru_sockaddr(so, &sa);
1121 *namelen = MIN(*namelen, sa->sa_len);
1131 * getsockname_args(int fdes, caddr_t asa, int *alen)
1136 getsockname(struct getsockname_args *uap)
1138 struct sockaddr *sa = NULL;
1141 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1145 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1148 error = copyout(sa, uap->asa, sa_len);
1150 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1157 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1158 * This allows kern_getpeername() to return a pointer to an allocated struct
1159 * sockaddr which must be freed later with FREE(). The caller must
1160 * initialize *name to NULL.
1163 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1165 struct thread *td = curthread;
1166 struct proc *p = td->td_proc;
1169 struct sockaddr *sa = NULL;
1172 error = holdsock(p->p_fd, s, &fp);
1179 so = (struct socket *)fp->f_data;
1180 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1184 error = so_pru_peeraddr(so, &sa);
1189 *namelen = MIN(*namelen, sa->sa_len);
1199 * getpeername_args(int fdes, caddr_t asa, int *alen)
1201 * Get name of peer for connected socket.
1204 getpeername(struct getpeername_args *uap)
1206 struct sockaddr *sa = NULL;
1209 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1213 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1216 error = copyout(sa, uap->asa, sa_len);
1218 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1225 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1227 struct sockaddr *sa;
1231 if (len > SOCK_MAXADDRLEN)
1232 return ENAMETOOLONG;
1233 if (len < offsetof(struct sockaddr, sa_data[0]))
1235 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1236 error = copyin(uaddr, sa, len);
1240 #if BYTE_ORDER != BIG_ENDIAN
1242 * The bind(), connect(), and sendto() syscalls were not
1243 * versioned for COMPAT_43. Thus, this check must stay.
1245 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1246 sa->sa_family = sa->sa_len;
1255 * holdsock() - load the struct file pointer associated
1256 * with a socket into *fpp. If an error occurs, non-zero
1257 * will be returned and *fpp will be set to NULL.
1260 holdsock(fdp, fdes, fpp)
1261 struct filedesc *fdp;
1265 struct file *fp = NULL;
1268 if ((unsigned)fdes >= fdp->fd_nfiles ||
1269 (fp = fdp->fd_ofiles[fdes]) == NULL) {
1271 } else if (fp->f_type != DTYPE_SOCKET) {
1282 * Detach a mapped page and release resources back to the system.
1283 * We must release our wiring and if the object is ripped out
1284 * from under the vm_page we become responsible for freeing the
1287 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1290 sf_buf_mref(void *arg)
1292 struct sfbuf_mref *sfm = arg;
1298 sf_buf_mfree(void *arg)
1300 struct sfbuf_mref *sfm = arg;
1304 KKASSERT(sfm->mref_count > 0);
1305 if (--sfm->mref_count == 0) {
1306 m = sf_buf_page(sfm->sf);
1307 sf_buf_free(sfm->sf);
1309 vm_page_unwire(m, 0);
1310 if (m->wire_count == 0 && m->object == NULL)
1313 free(sfm, M_SENDFILE);
1319 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1320 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1322 * Send a file specified by 'fd' and starting at 'offset' to a socket
1323 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1324 * nbytes == 0. Optionally add a header and/or trailer to the socket
1325 * output. If specified, write the total number of bytes sent into *sbytes.
1327 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1328 * the headers to count against the remaining bytes to be sent from
1329 * the file descriptor. We may wish to implement a compatibility syscall
1333 sendfile(struct sendfile_args *uap)
1335 struct thread *td = curthread;
1336 struct proc *p = td->td_proc;
1338 struct filedesc *fdp;
1339 struct vnode *vp = NULL;
1340 struct sf_hdtr hdtr;
1341 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1343 struct mbuf *mheader = NULL;
1344 off_t hdtr_size = 0, sbytes;
1345 int error, hbytes = 0, tbytes;
1351 * Do argument checking. Must be a regular file in, stream
1352 * type and connected socket out, positive offset.
1354 fp = holdfp(fdp, uap->fd, FREAD);
1358 if (fp->f_type != DTYPE_VNODE) {
1362 vp = (struct vnode *)fp->f_data;
1367 * If specified, get the pointer to the sf_hdtr struct for
1368 * any headers/trailers.
1371 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1378 error = iovec_copyin(hdtr.headers, &iov, aiov,
1379 hdtr.hdr_cnt, &hbytes);
1383 auio.uio_iovcnt = hdtr.hdr_cnt;
1384 auio.uio_offset = 0;
1385 auio.uio_segflg = UIO_USERSPACE;
1386 auio.uio_rw = UIO_WRITE;
1388 auio.uio_resid = hbytes;
1390 mheader = m_uiomove(&auio, MB_WAIT, 0);
1392 iovec_free(&iov, aiov);
1393 if (mheader == NULL)
1398 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1399 &sbytes, uap->flags);
1404 * Send trailers. Wimp out and use writev(2).
1406 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1407 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1408 hdtr.trl_cnt, &auio.uio_resid);
1412 auio.uio_iovcnt = hdtr.trl_cnt;
1413 auio.uio_offset = 0;
1414 auio.uio_segflg = UIO_USERSPACE;
1415 auio.uio_rw = UIO_WRITE;
1418 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1420 iovec_free(&iov, aiov);
1423 hdtr_size += tbytes; /* trailer bytes successfully sent */
1427 if (uap->sbytes != NULL) {
1428 sbytes += hdtr_size;
1429 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1437 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1438 struct mbuf *mheader, off_t *sbytes, int flags)
1440 struct thread *td = curthread;
1441 struct proc *p = td->td_proc;
1442 struct vm_object *obj;
1447 struct sfbuf_mref *sfm;
1454 if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) {
1458 error = holdsock(p->p_fd, sfd, &fp);
1461 so = (struct socket *)fp->f_data;
1462 if (so->so_type != SOCK_STREAM) {
1466 if ((so->so_state & SS_ISCONNECTED) == 0) {
1477 * Protect against multiple writers to the socket.
1479 (void) sblock(&so->so_snd, M_WAITOK);
1482 * Loop through the pages in the file, starting with the requested
1483 * offset. Get a file page (do I/O if necessary), map the file page
1484 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1487 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1491 pindex = OFF_TO_IDX(off);
1494 * Calculate the amount to transfer. Not to exceed a page,
1495 * the EOF, or the passed in nbytes.
1497 xfsize = obj->un_pager.vnp.vnp_size - off;
1498 if (xfsize > PAGE_SIZE)
1500 pgoff = (vm_offset_t)(off & PAGE_MASK);
1501 if (PAGE_SIZE - pgoff < xfsize)
1502 xfsize = PAGE_SIZE - pgoff;
1503 if (nbytes && xfsize > (nbytes - *sbytes))
1504 xfsize = nbytes - *sbytes;
1508 * Optimize the non-blocking case by looking at the socket space
1509 * before going to the extra work of constituting the sf_buf.
1511 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) {
1512 if (so->so_state & SS_CANTSENDMORE)
1516 sbunlock(&so->so_snd);
1520 * Attempt to look up the page.
1522 * Allocate if not found, wait and loop if busy, then
1523 * wire the page. splvm() protection is required to
1524 * maintain the object association (an interrupt can
1525 * free the page) through to the vm_page_wire() call.
1528 pg = vm_page_lookup(obj, pindex);
1530 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1537 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1545 * If page is not valid for what we need, initiate I/O
1548 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1554 * Ensure that our page is still around when the I/O
1557 vm_page_io_start(pg);
1560 * Get the page from backing store.
1562 bsize = vp->v_mount->mnt_stat.f_iosize;
1563 auio.uio_iov = &aiov;
1564 auio.uio_iovcnt = 1;
1566 aiov.iov_len = MAXBSIZE;
1567 auio.uio_resid = MAXBSIZE;
1568 auio.uio_offset = trunc_page(off);
1569 auio.uio_segflg = UIO_NOCOPY;
1570 auio.uio_rw = UIO_READ;
1572 vn_lock(vp, NULL, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
1573 error = VOP_READ(vp, &auio,
1574 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1576 VOP_UNLOCK(vp, NULL, 0, td);
1577 vm_page_flag_clear(pg, PG_ZERO);
1578 vm_page_io_finish(pg);
1580 vm_page_unwire(pg, 0);
1582 * See if anyone else might know about this page.
1583 * If not and it is not valid, then free it.
1585 if (pg->wire_count == 0 && pg->valid == 0 &&
1586 pg->busy == 0 && !(pg->flags & PG_BUSY) &&
1587 pg->hold_count == 0) {
1591 sbunlock(&so->so_snd);
1598 * Get a sendfile buf. We usually wait as long as necessary,
1599 * but this wait can be interrupted.
1601 if ((sf = sf_buf_alloc(pg, SFBA_PCATCH)) == NULL) {
1603 vm_page_unwire(pg, 0);
1604 if (pg->wire_count == 0 && pg->object == NULL)
1607 sbunlock(&so->so_snd);
1613 * Get an mbuf header and set it up as having external storage.
1615 MGETHDR(m, MB_WAIT, MT_DATA);
1619 sbunlock(&so->so_snd);
1624 * sfm is a temporary hack, use a per-cpu cache for this.
1626 sfm = malloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1628 sfm->mref_count = 1;
1630 m->m_ext.ext_nfree.new = sf_buf_mfree;
1631 m->m_ext.ext_nref.new = sf_buf_mref;
1632 m->m_ext.ext_arg = sfm;
1633 m->m_ext.ext_buf = (void *)sf->kva;
1634 m->m_ext.ext_size = PAGE_SIZE;
1635 m->m_data = (char *) sf->kva + pgoff;
1636 m->m_flags |= M_EXT;
1637 m->m_pkthdr.len = m->m_len = xfsize;
1638 KKASSERT((m->m_flags & (M_EXT_OLD|M_EXT_CLUSTER)) == 0);
1640 if (mheader != NULL) {
1641 hbytes = mheader->m_pkthdr.len;
1642 mheader->m_pkthdr.len += m->m_pkthdr.len;
1650 * Add the buffer to the socket buffer chain.
1655 * Make sure that the socket is still able to take more data.
1656 * CANTSENDMORE being true usually means that the connection
1657 * was closed. so_error is true when an error was sensed after
1659 * The state is checked after the page mapping and buffer
1660 * allocation above since those operations may block and make
1661 * any socket checks stale. From this point forward, nothing
1662 * blocks before the pru_send (or more accurately, any blocking
1663 * results in a loop back to here to re-check).
1665 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1666 if (so->so_state & SS_CANTSENDMORE) {
1669 error = so->so_error;
1673 sbunlock(&so->so_snd);
1678 * Wait for socket space to become available. We do this just
1679 * after checking the connection state above in order to avoid
1680 * a race condition with sbwait().
1682 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1683 if (so->so_state & SS_NBIO) {
1685 sbunlock(&so->so_snd);
1690 error = sbwait(&so->so_snd);
1692 * An error from sbwait usually indicates that we've
1693 * been interrupted by a signal. If we've sent anything
1694 * then return bytes sent, otherwise return the error.
1698 sbunlock(&so->so_snd);
1704 error = so_pru_send(so, 0, m, NULL, NULL, td);
1707 sbunlock(&so->so_snd);
1711 if (mheader != NULL) {
1712 *sbytes += mheader->m_pkthdr.len;
1713 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1716 sbunlock(&so->so_snd);
1721 if (mheader != NULL)