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 1.62 2006/04/26 17:42:53 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;
108 struct filedesc *fdp;
116 error = falloc(p, &fp, &fd);
119 error = socreate(domain, &so, type, protocol, td);
121 if (fdp->fd_files[fd].fp == fp) {
126 fp->f_type = DTYPE_SOCKET;
127 fp->f_flag = FREAD | FWRITE;
128 fp->f_ops = &socketops;
137 socket(struct socket_args *uap)
141 error = kern_socket(uap->domain, uap->type, uap->protocol,
142 &uap->sysmsg_result);
148 kern_bind(int s, struct sockaddr *sa)
150 struct thread *td = curthread;
151 struct proc *p = td->td_proc;
156 error = holdsock(p->p_fd, s, &fp);
159 error = sobind((struct socket *)fp->f_data, sa, td);
165 * bind_args(int s, caddr_t name, int namelen)
168 bind(struct bind_args *uap)
173 error = getsockaddr(&sa, uap->name, uap->namelen);
176 error = kern_bind(uap->s, sa);
183 kern_listen(int s, int backlog)
185 struct thread *td = curthread;
186 struct proc *p = td->td_proc;
191 error = holdsock(p->p_fd, s, &fp);
194 error = solisten((struct socket *)fp->f_data, backlog, td);
200 * listen_args(int s, int backlog)
203 listen(struct listen_args *uap)
207 error = kern_listen(uap->s, uap->backlog);
212 * Returns the accepted socket as well.
215 soaccept_predicate(struct netmsg *msg0)
217 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
218 struct socket *head = msg->nm_so;
220 if (head->so_error != 0) {
221 msg->nm_lmsg.ms_error = head->so_error;
224 if (!TAILQ_EMPTY(&head->so_comp)) {
225 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
226 msg->nm_so = TAILQ_FIRST(&head->so_comp);
227 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
230 msg->nm_lmsg.ms_error = 0;
233 if (head->so_state & SS_CANTRCVMORE) {
234 msg->nm_lmsg.ms_error = ECONNABORTED;
237 if (head->so_state & SS_NBIO) {
238 msg->nm_lmsg.ms_error = EWOULDBLOCK;
246 * The second argument to kern_accept() is a handle to a struct sockaddr.
247 * This allows kern_accept() to return a pointer to an allocated struct
248 * sockaddr which must be freed later with FREE(). The caller must
249 * initialize *name to NULL.
252 kern_accept(int s, struct sockaddr **name, int *namelen, int *res)
254 struct thread *td = curthread;
255 struct proc *p = td->td_proc;
256 struct filedesc *fdp = p->p_fd;
257 struct file *lfp = NULL;
258 struct file *nfp = NULL;
260 struct socket *head, *so;
261 struct netmsg_so_notify msg;
264 u_int fflag; /* type must match fp->f_flag */
267 if (name && namelen && *namelen < 0)
270 error = holdsock(fdp, s, &lfp);
274 error = falloc(p, &nfp, &fd);
275 if (error) { /* Probably ran out of file descriptors. */
282 head = (struct socket *)lfp->f_data;
283 if ((head->so_options & SO_ACCEPTCONN) == 0) {
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;
296 msg.nm_etype = NM_REVENT;
297 error = lwkt_domsg(port, &msg.nm_lmsg);
302 * At this point we have the connection that's ready to be accepted.
308 /* connection has been removed from the listen queue */
309 KNOTE(&head->so_rcv.sb_sel.si_note, 0);
311 so->so_state &= ~SS_COMP;
313 if (head->so_sigio != NULL)
314 fsetown(fgetown(head->so_sigio), &so->so_sigio);
316 nfp->f_type = DTYPE_SOCKET;
318 nfp->f_ops = &socketops;
320 /* Sync socket nonblocking/async state with file flags */
321 tmp = fflag & FNONBLOCK;
322 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td);
323 tmp = fflag & FASYNC;
324 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td);
327 error = soaccept(so, &sa);
330 * Set the returned name and namelen as applicable. Set the returned
331 * namelen to 0 for older code which might ignore the return value
335 if (sa && name && namelen) {
336 if (*namelen > sa->sa_len)
337 *namelen = sa->sa_len;
347 * close the new descriptor, assuming someone hasn't ripped it
348 * out from under us. Note that *res is normally ignored if an
349 * error is returned but a syscall message will still have access
350 * to the result code.
354 if (fdp->fd_files[fd].fp == nfp) {
361 * Release explicitly held references before returning.
370 * accept_args(int s, caddr_t name, int *anamelen)
373 accept(struct accept_args *uap)
375 struct sockaddr *sa = NULL;
380 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
384 error = kern_accept(uap->s, &sa, &sa_len, &uap->sysmsg_result);
387 error = copyout(sa, uap->name, sa_len);
389 error = copyout(&sa_len, uap->anamelen,
390 sizeof(*uap->anamelen));
395 error = kern_accept(uap->s, NULL, 0, &uap->sysmsg_result);
401 * Returns TRUE if predicate satisfied.
404 soconnected_predicate(struct netmsg *msg0)
406 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
407 struct socket *so = msg->nm_so;
409 /* check predicate */
410 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
411 msg->nm_lmsg.ms_error = so->so_error;
419 kern_connect(int s, struct sockaddr *sa)
421 struct thread *td = curthread;
422 struct proc *p = td->td_proc;
427 error = holdsock(p->p_fd, s, &fp);
430 so = (struct socket *)fp->f_data;
431 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
435 error = soconnect(so, sa, td);
438 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
442 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
443 struct netmsg_so_notify msg;
446 port = so->so_proto->pr_mport(so, sa, PRU_PRED);
447 lwkt_initmsg(&msg.nm_lmsg,
448 &curthread->td_msgport,
449 MSGF_PCATCH | MSGF_ABORTABLE,
450 lwkt_cmd_func(netmsg_so_notify),
451 lwkt_cmd_func(netmsg_so_notify_abort));
452 msg.nm_predicate = soconnected_predicate;
454 msg.nm_etype = NM_REVENT;
455 error = lwkt_domsg(port, &msg.nm_lmsg);
458 error = so->so_error;
462 so->so_state &= ~SS_ISCONNECTING;
463 if (error == ERESTART)
471 * connect_args(int s, caddr_t name, int namelen)
474 connect(struct connect_args *uap)
479 error = getsockaddr(&sa, uap->name, uap->namelen);
482 error = kern_connect(uap->s, sa);
489 kern_socketpair(int domain, int type, int protocol, int *sv)
491 struct thread *td = curthread;
492 struct proc *p = td->td_proc;
493 struct filedesc *fdp;
494 struct file *fp1, *fp2;
495 struct socket *so1, *so2;
500 error = socreate(domain, &so1, type, protocol, td);
503 error = socreate(domain, &so2, type, protocol, td);
506 error = falloc(p, &fp1, &fd);
511 error = falloc(p, &fp2, &fd);
516 error = soconnect2(so1, so2);
519 if (type == SOCK_DGRAM) {
521 * Datagram socket connection is asymmetric.
523 error = soconnect2(so2, so1);
527 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
528 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
529 fp1->f_ops = fp2->f_ops = &socketops;
534 if (fdp->fd_files[sv[1]].fp == fp2) {
535 funsetfd(fdp, sv[1]);
540 if (fdp->fd_files[sv[0]].fp == fp1) {
541 funsetfd(fdp, sv[0]);
553 * socketpair(int domain, int type, int protocol, int *rsv)
556 socketpair(struct socketpair_args *uap)
560 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
563 error = copyout(sockv, uap->rsv, sizeof(sockv));
568 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
569 struct mbuf *control, int flags, int *res)
571 struct thread *td = curthread;
572 struct proc *p = td->td_proc;
577 struct iovec *ktriov = NULL;
581 error = holdsock(p->p_fd, s, &fp);
584 if (auio->uio_resid < 0) {
589 if (KTRPOINT(td, KTR_GENIO)) {
590 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
592 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
593 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
597 len = auio->uio_resid;
598 so = (struct socket *)fp->f_data;
599 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
601 if (auio->uio_resid != len && (error == ERESTART ||
602 error == EINTR || error == EWOULDBLOCK))
608 if (ktriov != NULL) {
610 ktruio.uio_iov = ktriov;
611 ktruio.uio_resid = len - auio->uio_resid;
612 ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error);
614 FREE(ktriov, M_TEMP);
618 *res = len - auio->uio_resid;
625 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
628 sendto(struct sendto_args *uap)
630 struct thread *td = curthread;
633 struct sockaddr *sa = NULL;
637 error = getsockaddr(&sa, uap->to, uap->tolen);
641 aiov.iov_base = uap->buf;
642 aiov.iov_len = uap->len;
643 auio.uio_iov = &aiov;
646 auio.uio_resid = uap->len;
647 auio.uio_segflg = UIO_USERSPACE;
648 auio.uio_rw = UIO_WRITE;
651 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
652 &uap->sysmsg_result);
660 * sendmsg_args(int s, caddr_t msg, int flags)
663 sendmsg(struct sendmsg_args *uap)
665 struct thread *td = curthread;
668 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
669 struct sockaddr *sa = NULL;
670 struct mbuf *control = NULL;
673 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
678 * Conditionally copyin msg.msg_name.
681 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
689 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
694 auio.uio_iovcnt = msg.msg_iovlen;
696 auio.uio_segflg = UIO_USERSPACE;
697 auio.uio_rw = UIO_WRITE;
701 * Conditionally copyin msg.msg_control.
703 if (msg.msg_control) {
704 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
705 msg.msg_controllen > MLEN) {
709 control = m_get(MB_WAIT, MT_CONTROL);
710 if (control == NULL) {
714 control->m_len = msg.msg_controllen;
715 error = copyin(msg.msg_control, mtod(control, caddr_t),
723 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
724 &uap->sysmsg_result);
729 iovec_free(&iov, aiov);
734 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
735 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
736 * Don't forget to FREE() and m_free() these if they are returned.
739 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
740 struct mbuf **control, int *flags, int *res)
742 struct thread *td = curthread;
743 struct proc *p = td->td_proc;
748 struct iovec *ktriov = NULL;
752 error = holdsock(p->p_fd, s, &fp);
755 if (auio->uio_resid < 0) {
760 if (KTRPOINT(td, KTR_GENIO)) {
761 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
763 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
764 bcopy(auio->uio_iov, ktriov, iovlen);
768 len = auio->uio_resid;
769 so = (struct socket *)fp->f_data;
770 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
772 if (auio->uio_resid != len && (error == ERESTART ||
773 error == EINTR || error == EWOULDBLOCK))
777 if (ktriov != NULL) {
779 ktruio.uio_iov = ktriov;
780 ktruio.uio_resid = len - auio->uio_resid;
781 ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error);
783 FREE(ktriov, M_TEMP);
787 *res = len - auio->uio_resid;
794 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
795 * caddr_t from, int *fromlenaddr)
798 recvfrom(struct recvfrom_args *uap)
800 struct thread *td = curthread;
803 struct sockaddr *sa = NULL;
806 if (uap->from && uap->fromlenaddr) {
807 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
815 aiov.iov_base = uap->buf;
816 aiov.iov_len = uap->len;
817 auio.uio_iov = &aiov;
820 auio.uio_resid = uap->len;
821 auio.uio_segflg = UIO_USERSPACE;
822 auio.uio_rw = UIO_READ;
825 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
826 &uap->flags, &uap->sysmsg_result);
828 if (error == 0 && uap->from) {
829 /* note: sa may still be NULL */
831 fromlen = MIN(fromlen, sa->sa_len);
832 error = copyout(sa, uap->from, fromlen);
837 error = copyout(&fromlen, uap->fromlenaddr,
848 * recvmsg_args(int s, struct msghdr *msg, int flags)
851 recvmsg(struct recvmsg_args *uap)
853 struct thread *td = curthread;
856 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
857 struct mbuf *m, *control = NULL;
858 struct sockaddr *sa = NULL;
860 socklen_t *ufromlenp, *ucontrollenp;
861 int error, fromlen, controllen, len, flags, *uflagsp;
864 * This copyin handles everything except the iovec.
866 error = copyin(uap->msg, &msg, sizeof(msg));
870 if (msg.msg_name && msg.msg_namelen < 0)
872 if (msg.msg_control && msg.msg_controllen < 0)
875 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
877 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
879 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
885 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
890 auio.uio_iovcnt = msg.msg_iovlen;
892 auio.uio_segflg = UIO_USERSPACE;
893 auio.uio_rw = UIO_READ;
898 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
899 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
902 * Conditionally copyout the name and populate the namelen field.
904 if (error == 0 && msg.msg_name) {
905 fromlen = MIN(msg.msg_namelen, sa->sa_len);
906 error = copyout(sa, msg.msg_name, fromlen);
908 error = copyout(&fromlen, ufromlenp,
913 * Copyout msg.msg_control and msg.msg_controllen.
915 if (error == 0 && msg.msg_control) {
916 len = msg.msg_controllen;
918 ctlbuf = (caddr_t)msg.msg_control;
920 while(m && len > 0) {
923 if (len >= m->m_len) {
926 msg.msg_flags |= MSG_CTRUNC;
930 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
938 controllen = ctlbuf - (caddr_t)msg.msg_control;
939 error = copyout(&controllen, ucontrollenp,
940 sizeof(*ucontrollenp));
944 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
949 iovec_free(&iov, aiov);
956 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
957 * in kernel pointer instead of a userland pointer. This allows us
958 * to manipulate socket options in the emulation code.
961 kern_setsockopt(int s, struct sockopt *sopt)
963 struct thread *td = curthread;
964 struct proc *p = td->td_proc;
968 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
970 if (sopt->sopt_valsize < 0)
973 error = holdsock(p->p_fd, s, &fp);
977 error = sosetopt((struct socket *)fp->f_data, sopt);
983 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
986 setsockopt(struct setsockopt_args *uap)
988 struct thread *td = curthread;
992 sopt.sopt_level = uap->level;
993 sopt.sopt_name = uap->name;
994 sopt.sopt_val = uap->val;
995 sopt.sopt_valsize = uap->valsize;
998 error = kern_setsockopt(uap->s, &sopt);
1003 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1004 * in kernel pointer instead of a userland pointer. This allows us
1005 * to manipulate socket options in the emulation code.
1008 kern_getsockopt(int s, struct sockopt *sopt)
1010 struct thread *td = curthread;
1011 struct proc *p = td->td_proc;
1015 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1017 if (sopt->sopt_valsize < 0)
1020 error = holdsock(p->p_fd, s, &fp);
1024 error = sogetopt((struct socket *)fp->f_data, sopt);
1030 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1033 getsockopt(struct getsockopt_args *uap)
1035 struct thread *td = curthread;
1036 struct sockopt sopt;
1040 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1049 sopt.sopt_level = uap->level;
1050 sopt.sopt_name = uap->name;
1051 sopt.sopt_val = uap->val;
1052 sopt.sopt_valsize = valsize;
1055 error = kern_getsockopt(uap->s, &sopt);
1057 valsize = sopt.sopt_valsize;
1058 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1064 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1065 * This allows kern_getsockname() to return a pointer to an allocated struct
1066 * sockaddr which must be freed later with FREE(). The caller must
1067 * initialize *name to NULL.
1070 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1072 struct thread *td = curthread;
1073 struct proc *p = td->td_proc;
1076 struct sockaddr *sa = NULL;
1079 error = holdsock(p->p_fd, s, &fp);
1086 so = (struct socket *)fp->f_data;
1087 error = so_pru_sockaddr(so, &sa);
1092 *namelen = MIN(*namelen, sa->sa_len);
1102 * getsockname_args(int fdes, caddr_t asa, int *alen)
1107 getsockname(struct getsockname_args *uap)
1109 struct sockaddr *sa = NULL;
1112 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1116 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1119 error = copyout(sa, uap->asa, sa_len);
1121 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1128 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1129 * This allows kern_getpeername() to return a pointer to an allocated struct
1130 * sockaddr which must be freed later with FREE(). The caller must
1131 * initialize *name to NULL.
1134 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1136 struct thread *td = curthread;
1137 struct proc *p = td->td_proc;
1140 struct sockaddr *sa = NULL;
1143 error = holdsock(p->p_fd, s, &fp);
1150 so = (struct socket *)fp->f_data;
1151 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1155 error = so_pru_peeraddr(so, &sa);
1160 *namelen = MIN(*namelen, sa->sa_len);
1170 * getpeername_args(int fdes, caddr_t asa, int *alen)
1172 * Get name of peer for connected socket.
1175 getpeername(struct getpeername_args *uap)
1177 struct sockaddr *sa = NULL;
1180 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1184 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1187 error = copyout(sa, uap->asa, sa_len);
1189 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1196 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1198 struct sockaddr *sa;
1202 if (len > SOCK_MAXADDRLEN)
1203 return ENAMETOOLONG;
1204 if (len < offsetof(struct sockaddr, sa_data[0]))
1206 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1207 error = copyin(uaddr, sa, len);
1211 #if BYTE_ORDER != BIG_ENDIAN
1213 * The bind(), connect(), and sendto() syscalls were not
1214 * versioned for COMPAT_43. Thus, this check must stay.
1216 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1217 sa->sa_family = sa->sa_len;
1226 * holdsock() - load the struct file pointer associated
1227 * with a socket into *fpp. If an error occurs, non-zero
1228 * will be returned and *fpp will be set to NULL.
1231 holdsock(struct filedesc *fdp, int fdes, struct file **fpp)
1237 if ((unsigned)fdes >= fdp->fd_nfiles)
1239 if ((fp = fdp->fd_files[fdes].fp) == NULL)
1241 if (fp->f_type != DTYPE_SOCKET)
1249 * Detach a mapped page and release resources back to the system.
1250 * We must release our wiring and if the object is ripped out
1251 * from under the vm_page we become responsible for freeing the
1252 * page. These routines must be MPSAFE.
1254 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1256 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1259 sf_buf_mref(void *arg)
1261 struct sfbuf_mref *sfm = arg;
1264 * We must already hold a ref so there is no race to 0, just
1265 * atomically increment the count.
1267 atomic_add_int(&sfm->mref_count, 1);
1271 sf_buf_mfree(void *arg)
1273 struct sfbuf_mref *sfm = arg;
1276 KKASSERT(sfm->mref_count > 0);
1277 if (sfm->mref_count == 1) {
1279 * We are the only holder so no further locking is required,
1280 * the sfbuf can simply be freed.
1282 sfm->mref_count = 0;
1286 * There may be other holders, we must obtain the serializer
1287 * to protect against a sf_buf_mfree() race to 0. An atomic
1288 * operation is still required for races against
1291 * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1293 lwkt_serialize_enter(&sfm->serializer);
1294 atomic_subtract_int(&sfm->mref_count, 1);
1295 if (sfm->mref_count == 0) {
1296 lwkt_serialize_exit(&sfm->serializer);
1300 m = sf_buf_page(sfm->sf);
1301 sf_buf_free(sfm->sf);
1302 vm_page_unwire(m, 0);
1303 if (m->wire_count == 0 && m->object == NULL)
1304 vm_page_try_to_free(m);
1307 free(sfm, M_SENDFILE);
1309 lwkt_serialize_exit(&sfm->serializer);
1316 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1317 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1319 * Send a file specified by 'fd' and starting at 'offset' to a socket
1320 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1321 * nbytes == 0. Optionally add a header and/or trailer to the socket
1322 * output. If specified, write the total number of bytes sent into *sbytes.
1324 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1325 * the headers to count against the remaining bytes to be sent from
1326 * the file descriptor. We may wish to implement a compatibility syscall
1330 sendfile(struct sendfile_args *uap)
1332 struct thread *td = curthread;
1333 struct proc *p = td->td_proc;
1335 struct filedesc *fdp;
1336 struct vnode *vp = NULL;
1337 struct sf_hdtr hdtr;
1338 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1340 struct mbuf *mheader = NULL;
1341 off_t hdtr_size = 0, sbytes;
1342 int error, hbytes = 0, tbytes;
1348 * Do argument checking. Must be a regular file in, stream
1349 * type and connected socket out, positive offset.
1351 fp = holdfp(fdp, uap->fd, FREAD);
1355 if (fp->f_type != DTYPE_VNODE) {
1359 vp = (struct vnode *)fp->f_data;
1364 * If specified, get the pointer to the sf_hdtr struct for
1365 * any headers/trailers.
1368 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1375 error = iovec_copyin(hdtr.headers, &iov, aiov,
1376 hdtr.hdr_cnt, &hbytes);
1380 auio.uio_iovcnt = hdtr.hdr_cnt;
1381 auio.uio_offset = 0;
1382 auio.uio_segflg = UIO_USERSPACE;
1383 auio.uio_rw = UIO_WRITE;
1385 auio.uio_resid = hbytes;
1387 mheader = m_uiomove(&auio);
1389 iovec_free(&iov, aiov);
1390 if (mheader == NULL)
1395 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1396 &sbytes, uap->flags);
1401 * Send trailers. Wimp out and use writev(2).
1403 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1404 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1405 hdtr.trl_cnt, &auio.uio_resid);
1409 auio.uio_iovcnt = hdtr.trl_cnt;
1410 auio.uio_offset = 0;
1411 auio.uio_segflg = UIO_USERSPACE;
1412 auio.uio_rw = UIO_WRITE;
1415 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1417 iovec_free(&iov, aiov);
1420 hdtr_size += tbytes; /* trailer bytes successfully sent */
1424 if (uap->sbytes != NULL) {
1425 sbytes += hdtr_size;
1426 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1434 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1435 struct mbuf *mheader, off_t *sbytes, int flags)
1437 struct thread *td = curthread;
1438 struct proc *p = td->td_proc;
1439 struct vm_object *obj;
1444 struct sfbuf_mref *sfm;
1450 if (vp->v_type != VREG) {
1454 if ((obj = vp->v_object) == NULL) {
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 = vp->v_filesize - 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. critical section protection is
1524 * required to maintain the object association (an
1525 * interrupt can free the page) through to the
1526 * vm_page_wire() call.
1529 pg = vm_page_lookup(obj, pindex);
1531 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1538 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1546 * If page is not valid for what we need, initiate I/O
1549 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1555 * Ensure that our page is still around when the I/O
1558 vm_page_io_start(pg);
1561 * Get the page from backing store.
1563 bsize = vp->v_mount->mnt_stat.f_iosize;
1564 auio.uio_iov = &aiov;
1565 auio.uio_iovcnt = 1;
1567 aiov.iov_len = MAXBSIZE;
1568 auio.uio_resid = MAXBSIZE;
1569 auio.uio_offset = trunc_page(off);
1570 auio.uio_segflg = UIO_NOCOPY;
1571 auio.uio_rw = UIO_READ;
1573 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
1574 error = VOP_READ(vp, &auio,
1575 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1577 VOP_UNLOCK(vp, 0, td);
1578 vm_page_flag_clear(pg, PG_ZERO);
1579 vm_page_io_finish(pg);
1582 vm_page_unwire(pg, 0);
1583 vm_page_try_to_free(pg);
1585 sbunlock(&so->so_snd);
1592 * Get a sendfile buf. We usually wait as long as necessary,
1593 * but this wait can be interrupted.
1595 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1597 vm_page_unwire(pg, 0);
1598 vm_page_try_to_free(pg);
1600 sbunlock(&so->so_snd);
1606 * Get an mbuf header and set it up as having external storage.
1608 MGETHDR(m, MB_WAIT, MT_DATA);
1612 sbunlock(&so->so_snd);
1617 * sfm is a temporary hack, use a per-cpu cache for this.
1619 sfm = malloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1621 sfm->mref_count = 1;
1622 lwkt_serialize_init(&sfm->serializer);
1624 m->m_ext.ext_free = sf_buf_mfree;
1625 m->m_ext.ext_ref = sf_buf_mref;
1626 m->m_ext.ext_arg = sfm;
1627 m->m_ext.ext_buf = (void *)sf->kva;
1628 m->m_ext.ext_size = PAGE_SIZE;
1629 m->m_data = (char *) sf->kva + pgoff;
1630 m->m_flags |= M_EXT;
1631 m->m_pkthdr.len = m->m_len = xfsize;
1632 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1634 if (mheader != NULL) {
1635 hbytes = mheader->m_pkthdr.len;
1636 mheader->m_pkthdr.len += m->m_pkthdr.len;
1644 * Add the buffer to the socket buffer chain.
1649 * Make sure that the socket is still able to take more data.
1650 * CANTSENDMORE being true usually means that the connection
1651 * was closed. so_error is true when an error was sensed after
1653 * The state is checked after the page mapping and buffer
1654 * allocation above since those operations may block and make
1655 * any socket checks stale. From this point forward, nothing
1656 * blocks before the pru_send (or more accurately, any blocking
1657 * results in a loop back to here to re-check).
1659 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1660 if (so->so_state & SS_CANTSENDMORE) {
1663 error = so->so_error;
1667 sbunlock(&so->so_snd);
1672 * Wait for socket space to become available. We do this just
1673 * after checking the connection state above in order to avoid
1674 * a race condition with sbwait().
1676 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1677 if (so->so_state & SS_NBIO) {
1679 sbunlock(&so->so_snd);
1684 error = sbwait(&so->so_snd);
1686 * An error from sbwait usually indicates that we've
1687 * been interrupted by a signal. If we've sent anything
1688 * then return bytes sent, otherwise return the error.
1692 sbunlock(&so->so_snd);
1698 error = so_pru_send(so, 0, m, NULL, NULL, td);
1701 sbunlock(&so->so_snd);
1705 if (mheader != NULL) {
1706 *sbytes += mheader->m_pkthdr.len;
1707 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1710 sbunlock(&so->so_snd);
1715 if (mheader != NULL)
1721 sctp_peeloff(struct sctp_peeloff_args *uap)
1724 struct thread *td = curthread;
1725 struct proc *p = td->td_proc;
1726 struct filedesc *fdp = p->p_fd;
1727 struct file *lfp = NULL;
1728 struct file *nfp = NULL;
1730 struct socket *head, *so;
1733 short fflag; /* type must match fp->f_flag */
1735 assoc_id = uap->name;
1736 error = holdsock(fdp, uap->sd, &lfp);
1741 head = (struct socket *)lfp->f_data;
1742 error = sctp_can_peel_off(head, assoc_id);
1748 * At this point we know we do have a assoc to pull
1749 * we proceed to get the fd setup. This may block
1753 fflag = lfp->f_flag;
1754 error = falloc(p, &nfp, &fd);
1757 * Probably ran out of file descriptors. Put the
1758 * unaccepted connection back onto the queue and
1759 * do another wakeup so some other process might
1760 * have a chance at it.
1766 uap->sysmsg_result = fd;
1768 so = sctp_get_peeloff(head, assoc_id, &error);
1771 * Either someone else peeled it off OR
1772 * we can't get a socket.
1776 so->so_state &= ~SS_COMP;
1777 so->so_state &= ~SS_NOFDREF;
1779 if (head->so_sigio != NULL)
1780 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1782 nfp->f_type = DTYPE_SOCKET;
1783 nfp->f_flag = fflag;
1784 nfp->f_ops = &socketops;
1789 * close the new descriptor, assuming someone hasn't ripped it
1790 * out from under us.
1793 if (fdp->fd_files[fd].fp == nfp) {
1800 * Release explicitly held references before returning.