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 1.67 2006/05/19 05:15:35 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;
132 socket(struct socket_args *uap)
136 error = kern_socket(uap->domain, uap->type, uap->protocol,
137 &uap->sysmsg_result);
143 kern_bind(int s, struct sockaddr *sa)
145 struct thread *td = curthread;
146 struct proc *p = td->td_proc;
151 error = holdsock(p->p_fd, s, &fp);
154 error = sobind((struct socket *)fp->f_data, sa, td);
160 * bind_args(int s, caddr_t name, int namelen)
163 bind(struct bind_args *uap)
168 error = getsockaddr(&sa, uap->name, uap->namelen);
171 error = kern_bind(uap->s, sa);
178 kern_listen(int s, int backlog)
180 struct thread *td = curthread;
181 struct proc *p = td->td_proc;
186 error = holdsock(p->p_fd, s, &fp);
189 error = solisten((struct socket *)fp->f_data, backlog, td);
195 * listen_args(int s, int backlog)
198 listen(struct listen_args *uap)
202 error = kern_listen(uap->s, uap->backlog);
207 * Returns the accepted socket as well.
210 soaccept_predicate(struct netmsg *msg0)
212 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
213 struct socket *head = msg->nm_so;
215 if (head->so_error != 0) {
216 msg->nm_lmsg.ms_error = head->so_error;
219 if (!TAILQ_EMPTY(&head->so_comp)) {
220 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
221 msg->nm_so = TAILQ_FIRST(&head->so_comp);
222 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
225 msg->nm_lmsg.ms_error = 0;
228 if (head->so_state & SS_CANTRCVMORE) {
229 msg->nm_lmsg.ms_error = ECONNABORTED;
232 if (head->so_state & SS_NBIO) {
233 msg->nm_lmsg.ms_error = EWOULDBLOCK;
241 * The second argument to kern_accept() is a handle to a struct sockaddr.
242 * This allows kern_accept() to return a pointer to an allocated struct
243 * sockaddr which must be freed later with FREE(). The caller must
244 * initialize *name to NULL.
247 kern_accept(int s, struct sockaddr **name, int *namelen, int *res)
249 struct thread *td = curthread;
250 struct proc *p = td->td_proc;
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(p->p_fd, s, &lfp);
268 error = falloc(p, &nfp, &fd);
269 if (error) { /* Probably ran out of file descriptors. */
276 head = (struct socket *)lfp->f_data;
277 if ((head->so_options & SO_ACCEPTCONN) == 0) {
282 /* optimize for uniprocessor case later XXX JH */
283 port = head->so_proto->pr_mport(head, NULL, PRU_PRED);
284 lwkt_initmsg(&msg.nm_lmsg, &curthread->td_msgport,
285 MSGF_PCATCH | MSGF_ABORTABLE,
286 lwkt_cmd_func(netmsg_so_notify),
287 lwkt_cmd_func(netmsg_so_notify_abort));
288 msg.nm_predicate = soaccept_predicate;
290 msg.nm_etype = NM_REVENT;
291 error = lwkt_domsg(port, &msg.nm_lmsg);
296 * At this point we have the connection that's ready to be accepted.
302 /* connection has been removed from the listen queue */
303 KNOTE(&head->so_rcv.sb_sel.si_note, 0);
305 so->so_state &= ~SS_COMP;
307 if (head->so_sigio != NULL)
308 fsetown(fgetown(head->so_sigio), &so->so_sigio);
310 nfp->f_type = DTYPE_SOCKET;
312 nfp->f_ops = &socketops;
314 /* Sync socket nonblocking/async state with file flags */
315 tmp = fflag & FNONBLOCK;
316 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, p->p_ucred);
317 tmp = fflag & FASYNC;
318 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, p->p_ucred);
321 error = soaccept(so, &sa);
324 * Set the returned name and namelen as applicable. Set the returned
325 * namelen to 0 for older code which might ignore the return value
329 if (sa && name && namelen) {
330 if (*namelen > sa->sa_len)
331 *namelen = sa->sa_len;
341 * close the new descriptor, assuming someone hasn't ripped it
342 * out from under us. Note that *res is normally ignored if an
343 * error is returned but a syscall message will still have access
344 * to the result code.
348 fdealloc(p, nfp, fd);
352 * Release explicitly held references before returning.
361 * accept_args(int s, caddr_t name, int *anamelen)
364 accept(struct accept_args *uap)
366 struct sockaddr *sa = NULL;
371 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
375 error = kern_accept(uap->s, &sa, &sa_len, &uap->sysmsg_result);
378 error = copyout(sa, uap->name, sa_len);
380 error = copyout(&sa_len, uap->anamelen,
381 sizeof(*uap->anamelen));
386 error = kern_accept(uap->s, NULL, 0, &uap->sysmsg_result);
392 * Returns TRUE if predicate satisfied.
395 soconnected_predicate(struct netmsg *msg0)
397 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
398 struct socket *so = msg->nm_so;
400 /* check predicate */
401 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
402 msg->nm_lmsg.ms_error = so->so_error;
410 kern_connect(int s, struct sockaddr *sa)
412 struct thread *td = curthread;
413 struct proc *p = td->td_proc;
418 error = holdsock(p->p_fd, s, &fp);
421 so = (struct socket *)fp->f_data;
422 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
426 error = soconnect(so, sa, td);
429 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
433 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
434 struct netmsg_so_notify msg;
437 port = so->so_proto->pr_mport(so, sa, PRU_PRED);
438 lwkt_initmsg(&msg.nm_lmsg,
439 &curthread->td_msgport,
440 MSGF_PCATCH | MSGF_ABORTABLE,
441 lwkt_cmd_func(netmsg_so_notify),
442 lwkt_cmd_func(netmsg_so_notify_abort));
443 msg.nm_predicate = soconnected_predicate;
445 msg.nm_etype = NM_REVENT;
446 error = lwkt_domsg(port, &msg.nm_lmsg);
449 error = so->so_error;
453 so->so_state &= ~SS_ISCONNECTING;
454 if (error == ERESTART)
462 * connect_args(int s, caddr_t name, int namelen)
465 connect(struct connect_args *uap)
470 error = getsockaddr(&sa, uap->name, uap->namelen);
473 error = kern_connect(uap->s, sa);
480 kern_socketpair(int domain, int type, int protocol, int *sv)
482 struct thread *td = curthread;
483 struct proc *p = td->td_proc;
484 struct file *fp1, *fp2;
485 struct socket *so1, *so2;
489 error = socreate(domain, &so1, type, protocol, td);
492 error = socreate(domain, &so2, type, protocol, td);
495 error = falloc(p, &fp1, &fd);
500 error = falloc(p, &fp2, &fd);
505 error = soconnect2(so1, so2);
508 if (type == SOCK_DGRAM) {
510 * Datagram socket connection is asymmetric.
512 error = soconnect2(so2, so1);
516 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
517 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
518 fp1->f_ops = fp2->f_ops = &socketops;
523 fdealloc(p, fp2, sv[1]);
526 fdealloc(p, fp1, sv[0]);
536 * socketpair(int domain, int type, int protocol, int *rsv)
539 socketpair(struct socketpair_args *uap)
543 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
546 error = copyout(sockv, uap->rsv, sizeof(sockv));
551 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
552 struct mbuf *control, int flags, int *res)
554 struct thread *td = curthread;
555 struct proc *p = td->td_proc;
560 struct iovec *ktriov = NULL;
564 error = holdsock(p->p_fd, s, &fp);
567 if (auio->uio_resid < 0) {
572 if (KTRPOINT(td, KTR_GENIO)) {
573 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
575 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
576 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
580 len = auio->uio_resid;
581 so = (struct socket *)fp->f_data;
582 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
584 if (auio->uio_resid != len && (error == ERESTART ||
585 error == EINTR || error == EWOULDBLOCK))
591 if (ktriov != NULL) {
593 ktruio.uio_iov = ktriov;
594 ktruio.uio_resid = len - auio->uio_resid;
595 ktrgenio(p, s, UIO_WRITE, &ktruio, error);
597 FREE(ktriov, M_TEMP);
601 *res = len - auio->uio_resid;
608 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
611 sendto(struct sendto_args *uap)
613 struct thread *td = curthread;
616 struct sockaddr *sa = NULL;
620 error = getsockaddr(&sa, uap->to, uap->tolen);
624 aiov.iov_base = uap->buf;
625 aiov.iov_len = uap->len;
626 auio.uio_iov = &aiov;
629 auio.uio_resid = uap->len;
630 auio.uio_segflg = UIO_USERSPACE;
631 auio.uio_rw = UIO_WRITE;
634 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
635 &uap->sysmsg_result);
643 * sendmsg_args(int s, caddr_t msg, int flags)
646 sendmsg(struct sendmsg_args *uap)
648 struct thread *td = curthread;
651 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
652 struct sockaddr *sa = NULL;
653 struct mbuf *control = NULL;
656 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
661 * Conditionally copyin msg.msg_name.
664 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
672 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
677 auio.uio_iovcnt = msg.msg_iovlen;
679 auio.uio_segflg = UIO_USERSPACE;
680 auio.uio_rw = UIO_WRITE;
684 * Conditionally copyin msg.msg_control.
686 if (msg.msg_control) {
687 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
688 msg.msg_controllen > MLEN) {
692 control = m_get(MB_WAIT, MT_CONTROL);
693 if (control == NULL) {
697 control->m_len = msg.msg_controllen;
698 error = copyin(msg.msg_control, mtod(control, caddr_t),
706 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
707 &uap->sysmsg_result);
712 iovec_free(&iov, aiov);
717 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
718 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
719 * Don't forget to FREE() and m_free() these if they are returned.
722 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
723 struct mbuf **control, int *flags, int *res)
725 struct thread *td = curthread;
726 struct proc *p = td->td_proc;
731 struct iovec *ktriov = NULL;
735 error = holdsock(p->p_fd, s, &fp);
738 if (auio->uio_resid < 0) {
743 if (KTRPOINT(td, KTR_GENIO)) {
744 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
746 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
747 bcopy(auio->uio_iov, ktriov, iovlen);
751 len = auio->uio_resid;
752 so = (struct socket *)fp->f_data;
753 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
755 if (auio->uio_resid != len && (error == ERESTART ||
756 error == EINTR || error == EWOULDBLOCK))
760 if (ktriov != NULL) {
762 ktruio.uio_iov = ktriov;
763 ktruio.uio_resid = len - auio->uio_resid;
764 ktrgenio(p, s, UIO_READ, &ktruio, error);
766 FREE(ktriov, M_TEMP);
770 *res = len - auio->uio_resid;
777 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
778 * caddr_t from, int *fromlenaddr)
781 recvfrom(struct recvfrom_args *uap)
783 struct thread *td = curthread;
786 struct sockaddr *sa = NULL;
789 if (uap->from && uap->fromlenaddr) {
790 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
798 aiov.iov_base = uap->buf;
799 aiov.iov_len = uap->len;
800 auio.uio_iov = &aiov;
803 auio.uio_resid = uap->len;
804 auio.uio_segflg = UIO_USERSPACE;
805 auio.uio_rw = UIO_READ;
808 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
809 &uap->flags, &uap->sysmsg_result);
811 if (error == 0 && uap->from) {
812 /* note: sa may still be NULL */
814 fromlen = MIN(fromlen, sa->sa_len);
815 error = copyout(sa, uap->from, fromlen);
820 error = copyout(&fromlen, uap->fromlenaddr,
831 * recvmsg_args(int s, struct msghdr *msg, int flags)
834 recvmsg(struct recvmsg_args *uap)
836 struct thread *td = curthread;
839 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
840 struct mbuf *m, *control = NULL;
841 struct sockaddr *sa = NULL;
843 socklen_t *ufromlenp, *ucontrollenp;
844 int error, fromlen, controllen, len, flags, *uflagsp;
847 * This copyin handles everything except the iovec.
849 error = copyin(uap->msg, &msg, sizeof(msg));
853 if (msg.msg_name && msg.msg_namelen < 0)
855 if (msg.msg_control && msg.msg_controllen < 0)
858 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
860 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
862 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
868 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
873 auio.uio_iovcnt = msg.msg_iovlen;
875 auio.uio_segflg = UIO_USERSPACE;
876 auio.uio_rw = UIO_READ;
881 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
882 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
885 * Conditionally copyout the name and populate the namelen field.
887 if (error == 0 && msg.msg_name) {
888 fromlen = MIN(msg.msg_namelen, sa->sa_len);
889 error = copyout(sa, msg.msg_name, fromlen);
891 error = copyout(&fromlen, ufromlenp,
896 * Copyout msg.msg_control and msg.msg_controllen.
898 if (error == 0 && msg.msg_control) {
899 len = msg.msg_controllen;
901 ctlbuf = (caddr_t)msg.msg_control;
903 while(m && len > 0) {
906 if (len >= m->m_len) {
909 msg.msg_flags |= MSG_CTRUNC;
913 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
921 controllen = ctlbuf - (caddr_t)msg.msg_control;
922 error = copyout(&controllen, ucontrollenp,
923 sizeof(*ucontrollenp));
927 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
932 iovec_free(&iov, aiov);
939 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
940 * in kernel pointer instead of a userland pointer. This allows us
941 * to manipulate socket options in the emulation code.
944 kern_setsockopt(int s, struct sockopt *sopt)
946 struct thread *td = curthread;
947 struct proc *p = td->td_proc;
951 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
953 if (sopt->sopt_valsize < 0)
956 error = holdsock(p->p_fd, s, &fp);
960 error = sosetopt((struct socket *)fp->f_data, sopt);
966 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
969 setsockopt(struct setsockopt_args *uap)
971 struct thread *td = curthread;
975 sopt.sopt_level = uap->level;
976 sopt.sopt_name = uap->name;
977 sopt.sopt_val = uap->val;
978 sopt.sopt_valsize = uap->valsize;
981 error = kern_setsockopt(uap->s, &sopt);
986 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
987 * in kernel pointer instead of a userland pointer. This allows us
988 * to manipulate socket options in the emulation code.
991 kern_getsockopt(int s, struct sockopt *sopt)
993 struct thread *td = curthread;
994 struct proc *p = td->td_proc;
998 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1000 if (sopt->sopt_valsize < 0)
1003 error = holdsock(p->p_fd, s, &fp);
1007 error = sogetopt((struct socket *)fp->f_data, sopt);
1013 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1016 getsockopt(struct getsockopt_args *uap)
1018 struct thread *td = curthread;
1019 struct sockopt sopt;
1023 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1032 sopt.sopt_level = uap->level;
1033 sopt.sopt_name = uap->name;
1034 sopt.sopt_val = uap->val;
1035 sopt.sopt_valsize = valsize;
1038 error = kern_getsockopt(uap->s, &sopt);
1040 valsize = sopt.sopt_valsize;
1041 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1047 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1048 * This allows kern_getsockname() to return a pointer to an allocated struct
1049 * sockaddr which must be freed later with FREE(). The caller must
1050 * initialize *name to NULL.
1053 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1055 struct thread *td = curthread;
1056 struct proc *p = td->td_proc;
1059 struct sockaddr *sa = NULL;
1062 error = holdsock(p->p_fd, s, &fp);
1069 so = (struct socket *)fp->f_data;
1070 error = so_pru_sockaddr(so, &sa);
1075 *namelen = MIN(*namelen, sa->sa_len);
1085 * getsockname_args(int fdes, caddr_t asa, int *alen)
1090 getsockname(struct getsockname_args *uap)
1092 struct sockaddr *sa = NULL;
1095 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1099 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1102 error = copyout(sa, uap->asa, sa_len);
1104 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1111 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1112 * This allows kern_getpeername() to return a pointer to an allocated struct
1113 * sockaddr which must be freed later with FREE(). The caller must
1114 * initialize *name to NULL.
1117 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1119 struct thread *td = curthread;
1120 struct proc *p = td->td_proc;
1123 struct sockaddr *sa = NULL;
1126 error = holdsock(p->p_fd, s, &fp);
1133 so = (struct socket *)fp->f_data;
1134 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1138 error = so_pru_peeraddr(so, &sa);
1143 *namelen = MIN(*namelen, sa->sa_len);
1153 * getpeername_args(int fdes, caddr_t asa, int *alen)
1155 * Get name of peer for connected socket.
1158 getpeername(struct getpeername_args *uap)
1160 struct sockaddr *sa = NULL;
1163 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1167 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1170 error = copyout(sa, uap->asa, sa_len);
1172 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1179 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1181 struct sockaddr *sa;
1185 if (len > SOCK_MAXADDRLEN)
1186 return ENAMETOOLONG;
1187 if (len < offsetof(struct sockaddr, sa_data[0]))
1189 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1190 error = copyin(uaddr, sa, len);
1194 #if BYTE_ORDER != BIG_ENDIAN
1196 * The bind(), connect(), and sendto() syscalls were not
1197 * versioned for COMPAT_43. Thus, this check must stay.
1199 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1200 sa->sa_family = sa->sa_len;
1209 * Detach a mapped page and release resources back to the system.
1210 * We must release our wiring and if the object is ripped out
1211 * from under the vm_page we become responsible for freeing the
1212 * page. These routines must be MPSAFE.
1214 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1216 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1219 sf_buf_mref(void *arg)
1221 struct sfbuf_mref *sfm = arg;
1224 * We must already hold a ref so there is no race to 0, just
1225 * atomically increment the count.
1227 atomic_add_int(&sfm->mref_count, 1);
1231 sf_buf_mfree(void *arg)
1233 struct sfbuf_mref *sfm = arg;
1236 KKASSERT(sfm->mref_count > 0);
1237 if (sfm->mref_count == 1) {
1239 * We are the only holder so no further locking is required,
1240 * the sfbuf can simply be freed.
1242 sfm->mref_count = 0;
1246 * There may be other holders, we must obtain the serializer
1247 * to protect against a sf_buf_mfree() race to 0. An atomic
1248 * operation is still required for races against
1251 * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1253 lwkt_serialize_enter(&sfm->serializer);
1254 atomic_subtract_int(&sfm->mref_count, 1);
1255 if (sfm->mref_count == 0) {
1256 lwkt_serialize_exit(&sfm->serializer);
1260 m = sf_buf_page(sfm->sf);
1261 sf_buf_free(sfm->sf);
1262 vm_page_unwire(m, 0);
1263 if (m->wire_count == 0 && m->object == NULL)
1264 vm_page_try_to_free(m);
1267 free(sfm, M_SENDFILE);
1269 lwkt_serialize_exit(&sfm->serializer);
1276 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1277 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1279 * Send a file specified by 'fd' and starting at 'offset' to a socket
1280 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1281 * nbytes == 0. Optionally add a header and/or trailer to the socket
1282 * output. If specified, write the total number of bytes sent into *sbytes.
1284 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1285 * the headers to count against the remaining bytes to be sent from
1286 * the file descriptor. We may wish to implement a compatibility syscall
1290 sendfile(struct sendfile_args *uap)
1292 struct thread *td = curthread;
1293 struct proc *p = td->td_proc;
1295 struct vnode *vp = NULL;
1296 struct sf_hdtr hdtr;
1297 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1299 struct mbuf *mheader = NULL;
1300 off_t hdtr_size = 0, sbytes;
1301 int error, hbytes = 0, tbytes;
1306 * Do argument checking. Must be a regular file in, stream
1307 * type and connected socket out, positive offset.
1309 fp = holdfp(p->p_fd, uap->fd, FREAD);
1313 if (fp->f_type != DTYPE_VNODE) {
1317 vp = (struct vnode *)fp->f_data;
1322 * If specified, get the pointer to the sf_hdtr struct for
1323 * any headers/trailers.
1326 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1333 error = iovec_copyin(hdtr.headers, &iov, aiov,
1334 hdtr.hdr_cnt, &hbytes);
1338 auio.uio_iovcnt = hdtr.hdr_cnt;
1339 auio.uio_offset = 0;
1340 auio.uio_segflg = UIO_USERSPACE;
1341 auio.uio_rw = UIO_WRITE;
1343 auio.uio_resid = hbytes;
1345 mheader = m_uiomove(&auio);
1347 iovec_free(&iov, aiov);
1348 if (mheader == NULL)
1353 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1354 &sbytes, uap->flags);
1359 * Send trailers. Wimp out and use writev(2).
1361 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1362 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1363 hdtr.trl_cnt, &auio.uio_resid);
1367 auio.uio_iovcnt = hdtr.trl_cnt;
1368 auio.uio_offset = 0;
1369 auio.uio_segflg = UIO_USERSPACE;
1370 auio.uio_rw = UIO_WRITE;
1373 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1375 iovec_free(&iov, aiov);
1378 hdtr_size += tbytes; /* trailer bytes successfully sent */
1382 if (uap->sbytes != NULL) {
1383 sbytes += hdtr_size;
1384 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1392 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1393 struct mbuf *mheader, off_t *sbytes, int flags)
1395 struct thread *td = curthread;
1396 struct proc *p = td->td_proc;
1397 struct vm_object *obj;
1402 struct sfbuf_mref *sfm;
1408 if (vp->v_type != VREG) {
1412 if ((obj = vp->v_object) == NULL) {
1416 error = holdsock(p->p_fd, sfd, &fp);
1419 so = (struct socket *)fp->f_data;
1420 if (so->so_type != SOCK_STREAM) {
1424 if ((so->so_state & SS_ISCONNECTED) == 0) {
1435 * Protect against multiple writers to the socket.
1437 (void) sblock(&so->so_snd, M_WAITOK);
1440 * Loop through the pages in the file, starting with the requested
1441 * offset. Get a file page (do I/O if necessary), map the file page
1442 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1445 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1449 pindex = OFF_TO_IDX(off);
1452 * Calculate the amount to transfer. Not to exceed a page,
1453 * the EOF, or the passed in nbytes.
1455 xfsize = vp->v_filesize - off;
1456 if (xfsize > PAGE_SIZE)
1458 pgoff = (vm_offset_t)(off & PAGE_MASK);
1459 if (PAGE_SIZE - pgoff < xfsize)
1460 xfsize = PAGE_SIZE - pgoff;
1461 if (nbytes && xfsize > (nbytes - *sbytes))
1462 xfsize = nbytes - *sbytes;
1466 * Optimize the non-blocking case by looking at the socket space
1467 * before going to the extra work of constituting the sf_buf.
1469 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) {
1470 if (so->so_state & SS_CANTSENDMORE)
1474 sbunlock(&so->so_snd);
1478 * Attempt to look up the page.
1480 * Allocate if not found, wait and loop if busy, then
1481 * wire the page. critical section protection is
1482 * required to maintain the object association (an
1483 * interrupt can free the page) through to the
1484 * vm_page_wire() call.
1487 pg = vm_page_lookup(obj, pindex);
1489 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1496 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1504 * If page is not valid for what we need, initiate I/O
1507 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1513 * Ensure that our page is still around when the I/O
1516 vm_page_io_start(pg);
1519 * Get the page from backing store.
1521 bsize = vp->v_mount->mnt_stat.f_iosize;
1522 auio.uio_iov = &aiov;
1523 auio.uio_iovcnt = 1;
1525 aiov.iov_len = MAXBSIZE;
1526 auio.uio_resid = MAXBSIZE;
1527 auio.uio_offset = trunc_page(off);
1528 auio.uio_segflg = UIO_NOCOPY;
1529 auio.uio_rw = UIO_READ;
1531 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY);
1532 error = VOP_READ(vp, &auio,
1533 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1536 vm_page_flag_clear(pg, PG_ZERO);
1537 vm_page_io_finish(pg);
1540 vm_page_unwire(pg, 0);
1541 vm_page_try_to_free(pg);
1543 sbunlock(&so->so_snd);
1550 * Get a sendfile buf. We usually wait as long as necessary,
1551 * but this wait can be interrupted.
1553 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1555 vm_page_unwire(pg, 0);
1556 vm_page_try_to_free(pg);
1558 sbunlock(&so->so_snd);
1564 * Get an mbuf header and set it up as having external storage.
1566 MGETHDR(m, MB_WAIT, MT_DATA);
1570 sbunlock(&so->so_snd);
1575 * sfm is a temporary hack, use a per-cpu cache for this.
1577 sfm = malloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1579 sfm->mref_count = 1;
1580 lwkt_serialize_init(&sfm->serializer);
1582 m->m_ext.ext_free = sf_buf_mfree;
1583 m->m_ext.ext_ref = sf_buf_mref;
1584 m->m_ext.ext_arg = sfm;
1585 m->m_ext.ext_buf = (void *)sf->kva;
1586 m->m_ext.ext_size = PAGE_SIZE;
1587 m->m_data = (char *) sf->kva + pgoff;
1588 m->m_flags |= M_EXT;
1589 m->m_pkthdr.len = m->m_len = xfsize;
1590 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1592 if (mheader != NULL) {
1593 hbytes = mheader->m_pkthdr.len;
1594 mheader->m_pkthdr.len += m->m_pkthdr.len;
1602 * Add the buffer to the socket buffer chain.
1607 * Make sure that the socket is still able to take more data.
1608 * CANTSENDMORE being true usually means that the connection
1609 * was closed. so_error is true when an error was sensed after
1611 * The state is checked after the page mapping and buffer
1612 * allocation above since those operations may block and make
1613 * any socket checks stale. From this point forward, nothing
1614 * blocks before the pru_send (or more accurately, any blocking
1615 * results in a loop back to here to re-check).
1617 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1618 if (so->so_state & SS_CANTSENDMORE) {
1621 error = so->so_error;
1625 sbunlock(&so->so_snd);
1630 * Wait for socket space to become available. We do this just
1631 * after checking the connection state above in order to avoid
1632 * a race condition with sbwait().
1634 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1635 if (so->so_state & SS_NBIO) {
1637 sbunlock(&so->so_snd);
1642 error = sbwait(&so->so_snd);
1644 * An error from sbwait usually indicates that we've
1645 * been interrupted by a signal. If we've sent anything
1646 * then return bytes sent, otherwise return the error.
1650 sbunlock(&so->so_snd);
1656 error = so_pru_send(so, 0, m, NULL, NULL, td);
1659 sbunlock(&so->so_snd);
1663 if (mheader != NULL) {
1664 *sbytes += mheader->m_pkthdr.len;
1665 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1668 sbunlock(&so->so_snd);
1673 if (mheader != NULL)
1679 sctp_peeloff(struct sctp_peeloff_args *uap)
1682 struct thread *td = curthread;
1683 struct proc *p = td->td_proc;
1684 struct file *lfp = NULL;
1685 struct file *nfp = NULL;
1687 struct socket *head, *so;
1690 short fflag; /* type must match fp->f_flag */
1692 assoc_id = uap->name;
1693 error = holdsock(p->p_fd, uap->sd, &lfp);
1698 head = (struct socket *)lfp->f_data;
1699 error = sctp_can_peel_off(head, assoc_id);
1705 * At this point we know we do have a assoc to pull
1706 * we proceed to get the fd setup. This may block
1710 fflag = lfp->f_flag;
1711 error = falloc(p, &nfp, &fd);
1714 * Probably ran out of file descriptors. Put the
1715 * unaccepted connection back onto the queue and
1716 * do another wakeup so some other process might
1717 * have a chance at it.
1723 uap->sysmsg_result = fd;
1725 so = sctp_get_peeloff(head, assoc_id, &error);
1728 * Either someone else peeled it off OR
1729 * we can't get a socket.
1733 so->so_state &= ~SS_COMP;
1734 so->so_state &= ~SS_NOFDREF;
1736 if (head->so_sigio != NULL)
1737 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1739 nfp->f_type = DTYPE_SOCKET;
1740 nfp->f_flag = fflag;
1741 nfp->f_ops = &socketops;
1746 * close the new descriptor, assuming someone hasn't ripped it
1747 * out from under us.
1750 fdealloc(p, nfp, fd);
1753 * Release explicitly held references before returning.