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. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94
33 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $
36 #include "opt_ktrace.h"
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/filedesc.h>
45 #include <sys/event.h>
47 #include <sys/fcntl.h>
49 #include <sys/filio.h>
50 #include <sys/kern_syscall.h>
52 #include <sys/protosw.h>
53 #include <sys/sfbuf.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/socketops.h>
58 #include <sys/vnode.h>
60 #include <sys/mount.h>
62 #include <sys/ktrace.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_pageout.h>
68 #include <vm/vm_kern.h>
69 #include <vm/vm_extern.h>
70 #include <sys/file2.h>
71 #include <sys/signalvar.h>
72 #include <sys/serialize.h>
74 #include <sys/thread2.h>
75 #include <sys/msgport2.h>
76 #include <sys/socketvar2.h>
77 #include <net/netmsg2.h>
78 #include <vm/vm_page2.h>
81 #include <netinet/sctp_peeloff.h>
84 extern int use_soaccept_pred_fast;
85 extern int use_sendfile_async;
86 extern int use_soconnect_async;
89 * System call interface to the socket abstraction.
92 extern struct fileops socketops;
95 * socket_args(int domain, int type, int protocol)
98 kern_socket(int domain, int type, int protocol, int *res)
100 struct thread *td = curthread;
101 struct filedesc *fdp = td->td_proc->p_fd;
106 KKASSERT(td->td_lwp);
108 error = falloc(td->td_lwp, &fp, &fd);
111 error = socreate(domain, &so, type, protocol, td);
113 fsetfd(fdp, NULL, fd);
115 fp->f_type = DTYPE_SOCKET;
116 fp->f_flag = FREAD | FWRITE;
117 fp->f_ops = &socketops;
130 sys_socket(struct socket_args *uap)
134 error = kern_socket(uap->domain, uap->type, uap->protocol,
135 &uap->sysmsg_iresult);
141 kern_bind(int s, struct sockaddr *sa)
143 struct thread *td = curthread;
144 struct proc *p = td->td_proc;
149 error = holdsock(p->p_fd, s, &fp);
152 error = sobind((struct socket *)fp->f_data, sa, td);
158 * bind_args(int s, caddr_t name, int namelen)
163 sys_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)
200 sys_listen(struct listen_args *uap)
204 error = kern_listen(uap->s, uap->backlog);
209 * Returns the accepted socket as well.
211 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the
212 * pool token is absolutely required to avoid a sofree() race,
213 * as well as to avoid tailq handling races.
216 soaccept_predicate(struct netmsg_so_notify *msg)
218 struct socket *head = msg->base.nm_so;
221 if (head->so_error != 0) {
222 msg->base.lmsg.ms_error = head->so_error;
225 lwkt_getpooltoken(head);
226 if (!TAILQ_EMPTY(&head->so_comp)) {
227 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
228 so = TAILQ_FIRST(&head->so_comp);
229 TAILQ_REMOVE(&head->so_comp, so, so_list);
231 soclrstate(so, SS_COMP);
235 lwkt_relpooltoken(head);
237 msg->base.lmsg.ms_error = 0;
238 msg->base.nm_so = so;
241 lwkt_relpooltoken(head);
242 if (head->so_state & SS_CANTRCVMORE) {
243 msg->base.lmsg.ms_error = ECONNABORTED;
246 if (msg->nm_fflags & FNONBLOCK) {
247 msg->base.lmsg.ms_error = EWOULDBLOCK;
255 * The second argument to kern_accept() is a handle to a struct sockaddr.
256 * This allows kern_accept() to return a pointer to an allocated struct
257 * sockaddr which must be freed later with FREE(). The caller must
258 * initialize *name to NULL.
261 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res)
263 struct thread *td = curthread;
264 struct filedesc *fdp = td->td_proc->p_fd;
265 struct file *lfp = NULL;
266 struct file *nfp = NULL;
268 struct socket *head, *so;
269 struct netmsg_so_notify msg;
271 u_int fflag; /* type must match fp->f_flag */
275 if (name && namelen && *namelen < 0)
278 error = holdsock(td->td_proc->p_fd, s, &lfp);
282 error = falloc(td->td_lwp, &nfp, &fd);
283 if (error) { /* Probably ran out of file descriptors. */
287 head = (struct socket *)lfp->f_data;
288 if ((head->so_options & SO_ACCEPTCONN) == 0) {
293 if (fflags & O_FBLOCKING)
294 fflags |= lfp->f_flag & ~FNONBLOCK;
295 else if (fflags & O_FNONBLOCKING)
296 fflags |= lfp->f_flag | FNONBLOCK;
298 fflags = lfp->f_flag;
300 if (use_soaccept_pred_fast) {
303 /* Initialize necessary parts for soaccept_predicate() */
304 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL);
305 msg.nm_fflags = fflags;
307 lwkt_getpooltoken(head);
308 pred = soaccept_predicate(&msg);
309 lwkt_relpooltoken(head);
312 error = msg.base.lmsg.ms_error;
320 /* optimize for uniprocessor case later XXX JH */
321 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport,
322 0, netmsg_so_notify, netmsg_so_notify_doabort);
323 msg.nm_predicate = soaccept_predicate;
324 msg.nm_fflags = fflags;
325 msg.nm_etype = NM_REVENT;
326 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH);
332 * At this point we have the connection that's ready to be accepted.
334 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects
335 * to eat the ref and turn it into a descriptor.
341 /* connection has been removed from the listen queue */
342 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0);
344 if (head->so_sigio != NULL)
345 fsetown(fgetown(&head->so_sigio), &so->so_sigio);
347 nfp->f_type = DTYPE_SOCKET;
349 nfp->f_ops = &socketops;
351 /* Sync socket nonblocking/async state with file flags */
352 tmp = fflag & FNONBLOCK;
353 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL);
354 tmp = fflag & FASYNC;
355 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
358 if (so->so_faddr != NULL) {
362 soaccept_generic(so);
365 error = soaccept(so, &sa);
369 * Set the returned name and namelen as applicable. Set the returned
370 * namelen to 0 for older code which might ignore the return value
374 if (sa && name && namelen) {
375 if (*namelen > sa->sa_len)
376 *namelen = sa->sa_len;
386 * If an error occured clear the reserved descriptor, else associate
389 * Note that *res is normally ignored if an error is returned but
390 * a syscall message will still have access to the result code.
393 fsetfd(fdp, NULL, fd);
396 fsetfd(fdp, nfp, fd);
404 * accept(int s, caddr_t name, int *anamelen)
409 sys_accept(struct accept_args *uap)
411 struct sockaddr *sa = NULL;
416 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
420 error = kern_accept(uap->s, 0, &sa, &sa_len,
421 &uap->sysmsg_iresult);
424 error = copyout(sa, uap->name, sa_len);
426 error = copyout(&sa_len, uap->anamelen,
427 sizeof(*uap->anamelen));
432 error = kern_accept(uap->s, 0, NULL, 0,
433 &uap->sysmsg_iresult);
439 * extaccept(int s, int fflags, caddr_t name, int *anamelen)
444 sys_extaccept(struct extaccept_args *uap)
446 struct sockaddr *sa = NULL;
449 int fflags = uap->flags & O_FMASK;
452 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
456 error = kern_accept(uap->s, fflags, &sa, &sa_len,
457 &uap->sysmsg_iresult);
460 error = copyout(sa, uap->name, sa_len);
462 error = copyout(&sa_len, uap->anamelen,
463 sizeof(*uap->anamelen));
468 error = kern_accept(uap->s, fflags, NULL, 0,
469 &uap->sysmsg_iresult);
476 * Returns TRUE if predicate satisfied.
479 soconnected_predicate(struct netmsg_so_notify *msg)
481 struct socket *so = msg->base.nm_so;
483 /* check predicate */
484 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
485 msg->base.lmsg.ms_error = so->so_error;
493 kern_connect(int s, int fflags, struct sockaddr *sa)
495 struct thread *td = curthread;
496 struct proc *p = td->td_proc;
499 int error, interrupted = 0;
501 error = holdsock(p->p_fd, s, &fp);
504 so = (struct socket *)fp->f_data;
506 if (fflags & O_FBLOCKING)
507 /* fflags &= ~FNONBLOCK; */;
508 else if (fflags & O_FNONBLOCKING)
513 if (so->so_state & SS_ISCONNECTING) {
517 error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE);
520 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
524 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
525 struct netmsg_so_notify msg;
527 netmsg_init_abortable(&msg.base, so,
528 &curthread->td_msgport,
531 netmsg_so_notify_doabort);
532 msg.nm_predicate = soconnected_predicate;
533 msg.nm_etype = NM_REVENT;
534 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH);
535 if (error == EINTR || error == ERESTART)
539 error = so->so_error;
544 soclrstate(so, SS_ISCONNECTING);
545 if (error == ERESTART)
553 * connect_args(int s, caddr_t name, int namelen)
558 sys_connect(struct connect_args *uap)
563 error = getsockaddr(&sa, uap->name, uap->namelen);
566 error = kern_connect(uap->s, 0, sa);
573 * connect_args(int s, int fflags, caddr_t name, int namelen)
578 sys_extconnect(struct extconnect_args *uap)
582 int fflags = uap->flags & O_FMASK;
584 error = getsockaddr(&sa, uap->name, uap->namelen);
587 error = kern_connect(uap->s, fflags, sa);
594 kern_socketpair(int domain, int type, int protocol, int *sv)
596 struct thread *td = curthread;
597 struct filedesc *fdp;
598 struct file *fp1, *fp2;
599 struct socket *so1, *so2;
602 fdp = td->td_proc->p_fd;
603 error = socreate(domain, &so1, type, protocol, td);
606 error = socreate(domain, &so2, type, protocol, td);
609 error = falloc(td->td_lwp, &fp1, &fd1);
614 error = falloc(td->td_lwp, &fp2, &fd2);
619 error = soconnect2(so1, so2);
622 if (type == SOCK_DGRAM) {
624 * Datagram socket connection is asymmetric.
626 error = soconnect2(so2, so1);
630 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
631 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
632 fp1->f_ops = fp2->f_ops = &socketops;
633 fsetfd(fdp, fp1, fd1);
634 fsetfd(fdp, fp2, fd2);
639 fsetfd(fdp, NULL, fd2);
642 fsetfd(fdp, NULL, fd1);
645 (void)soclose(so2, 0);
647 (void)soclose(so1, 0);
652 * socketpair(int domain, int type, int protocol, int *rsv)
655 sys_socketpair(struct socketpair_args *uap)
659 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
662 error = copyout(sockv, uap->rsv, sizeof(sockv));
665 kern_close(sockv[0]);
666 kern_close(sockv[1]);
674 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
675 struct mbuf *control, int flags, size_t *res)
677 struct thread *td = curthread;
678 struct lwp *lp = td->td_lwp;
679 struct proc *p = td->td_proc;
685 struct iovec *ktriov = NULL;
689 error = holdsock(p->p_fd, s, &fp);
693 if (KTRPOINT(td, KTR_GENIO)) {
694 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
696 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
697 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
701 len = auio->uio_resid;
702 so = (struct socket *)fp->f_data;
703 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
704 if (fp->f_flag & FNONBLOCK)
705 flags |= MSG_FNONBLOCKING;
707 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
709 if (auio->uio_resid != len && (error == ERESTART ||
710 error == EINTR || error == EWOULDBLOCK))
712 if (error == EPIPE && !(flags & MSG_NOSIGNAL) &&
713 !(so->so_options & SO_NOSIGPIPE))
714 lwpsignal(p, lp, SIGPIPE);
717 if (ktriov != NULL) {
719 ktruio.uio_iov = ktriov;
720 ktruio.uio_resid = len - auio->uio_resid;
721 ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
723 kfree(ktriov, M_TEMP);
727 *res = len - auio->uio_resid;
733 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
738 sys_sendto(struct sendto_args *uap)
740 struct thread *td = curthread;
743 struct sockaddr *sa = NULL;
747 error = getsockaddr(&sa, uap->to, uap->tolen);
751 aiov.iov_base = uap->buf;
752 aiov.iov_len = uap->len;
753 auio.uio_iov = &aiov;
756 auio.uio_resid = uap->len;
757 auio.uio_segflg = UIO_USERSPACE;
758 auio.uio_rw = UIO_WRITE;
761 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
762 &uap->sysmsg_szresult);
770 * sendmsg_args(int s, caddr_t msg, int flags)
775 sys_sendmsg(struct sendmsg_args *uap)
777 struct thread *td = curthread;
780 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
781 struct sockaddr *sa = NULL;
782 struct mbuf *control = NULL;
785 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
790 * Conditionally copyin msg.msg_name.
793 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
801 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
806 auio.uio_iovcnt = msg.msg_iovlen;
808 auio.uio_segflg = UIO_USERSPACE;
809 auio.uio_rw = UIO_WRITE;
813 * Conditionally copyin msg.msg_control.
815 if (msg.msg_control) {
816 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
817 msg.msg_controllen > MLEN) {
821 control = m_get(MB_WAIT, MT_CONTROL);
822 if (control == NULL) {
826 control->m_len = msg.msg_controllen;
827 error = copyin(msg.msg_control, mtod(control, caddr_t),
835 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
836 &uap->sysmsg_szresult);
839 iovec_free(&iov, aiov);
847 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
848 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
849 * Don't forget to FREE() and m_free() these if they are returned.
852 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
853 struct mbuf **control, int *flags, size_t *res)
855 struct thread *td = curthread;
856 struct proc *p = td->td_proc;
863 struct iovec *ktriov = NULL;
867 error = holdsock(p->p_fd, s, &fp);
871 if (KTRPOINT(td, KTR_GENIO)) {
872 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
874 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
875 bcopy(auio->uio_iov, ktriov, iovlen);
879 len = auio->uio_resid;
880 so = (struct socket *)fp->f_data;
882 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
883 if (fp->f_flag & FNONBLOCK) {
885 *flags |= MSG_FNONBLOCKING;
887 lflags = MSG_FNONBLOCKING;
893 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
895 if (auio->uio_resid != len && (error == ERESTART ||
896 error == EINTR || error == EWOULDBLOCK))
900 if (ktriov != NULL) {
902 ktruio.uio_iov = ktriov;
903 ktruio.uio_resid = len - auio->uio_resid;
904 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
906 kfree(ktriov, M_TEMP);
910 *res = len - auio->uio_resid;
916 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
917 * caddr_t from, int *fromlenaddr)
922 sys_recvfrom(struct recvfrom_args *uap)
924 struct thread *td = curthread;
927 struct sockaddr *sa = NULL;
930 if (uap->from && uap->fromlenaddr) {
931 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
939 aiov.iov_base = uap->buf;
940 aiov.iov_len = uap->len;
941 auio.uio_iov = &aiov;
944 auio.uio_resid = uap->len;
945 auio.uio_segflg = UIO_USERSPACE;
946 auio.uio_rw = UIO_READ;
949 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
950 &uap->flags, &uap->sysmsg_szresult);
952 if (error == 0 && uap->from) {
953 /* note: sa may still be NULL */
955 fromlen = MIN(fromlen, sa->sa_len);
956 error = copyout(sa, uap->from, fromlen);
961 error = copyout(&fromlen, uap->fromlenaddr,
972 * recvmsg_args(int s, struct msghdr *msg, int flags)
977 sys_recvmsg(struct recvmsg_args *uap)
979 struct thread *td = curthread;
982 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
983 struct mbuf *m, *control = NULL;
984 struct sockaddr *sa = NULL;
986 socklen_t *ufromlenp, *ucontrollenp;
987 int error, fromlen, controllen, len, flags, *uflagsp;
990 * This copyin handles everything except the iovec.
992 error = copyin(uap->msg, &msg, sizeof(msg));
996 if (msg.msg_name && msg.msg_namelen < 0)
998 if (msg.msg_control && msg.msg_controllen < 0)
1001 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1003 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1005 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
1011 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1016 auio.uio_iovcnt = msg.msg_iovlen;
1017 auio.uio_offset = 0;
1018 auio.uio_segflg = UIO_USERSPACE;
1019 auio.uio_rw = UIO_READ;
1024 error = kern_recvmsg(uap->s,
1025 (msg.msg_name ? &sa : NULL), &auio,
1026 (msg.msg_control ? &control : NULL), &flags,
1027 &uap->sysmsg_szresult);
1030 * Conditionally copyout the name and populate the namelen field.
1032 if (error == 0 && msg.msg_name) {
1033 /* note: sa may still be NULL */
1035 fromlen = MIN(msg.msg_namelen, sa->sa_len);
1036 error = copyout(sa, msg.msg_name, fromlen);
1041 error = copyout(&fromlen, ufromlenp,
1042 sizeof(*ufromlenp));
1046 * Copyout msg.msg_control and msg.msg_controllen.
1048 if (error == 0 && msg.msg_control) {
1049 len = msg.msg_controllen;
1051 ctlbuf = (caddr_t)msg.msg_control;
1053 while(m && len > 0) {
1054 unsigned int tocopy;
1056 if (len >= m->m_len) {
1059 msg.msg_flags |= MSG_CTRUNC;
1063 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1071 controllen = ctlbuf - (caddr_t)msg.msg_control;
1072 error = copyout(&controllen, ucontrollenp,
1073 sizeof(*ucontrollenp));
1077 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1081 kfree(sa, M_SONAME);
1082 iovec_free(&iov, aiov);
1089 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1090 * in kernel pointer instead of a userland pointer. This allows us
1091 * to manipulate socket options in the emulation code.
1094 kern_setsockopt(int s, struct sockopt *sopt)
1096 struct thread *td = curthread;
1097 struct proc *p = td->td_proc;
1101 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1103 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1105 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1108 error = holdsock(p->p_fd, s, &fp);
1112 error = sosetopt((struct socket *)fp->f_data, sopt);
1118 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1123 sys_setsockopt(struct setsockopt_args *uap)
1125 struct thread *td = curthread;
1126 struct sockopt sopt;
1129 sopt.sopt_level = uap->level;
1130 sopt.sopt_name = uap->name;
1131 sopt.sopt_valsize = uap->valsize;
1133 sopt.sopt_val = NULL;
1135 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1138 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1139 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1144 error = kern_setsockopt(uap->s, &sopt);
1147 kfree(sopt.sopt_val, M_TEMP);
1152 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1153 * in kernel pointer instead of a userland pointer. This allows us
1154 * to manipulate socket options in the emulation code.
1157 kern_getsockopt(int s, struct sockopt *sopt)
1159 struct thread *td = curthread;
1160 struct proc *p = td->td_proc;
1164 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1166 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1168 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1171 error = holdsock(p->p_fd, s, &fp);
1175 error = sogetopt((struct socket *)fp->f_data, sopt);
1181 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1186 sys_getsockopt(struct getsockopt_args *uap)
1188 struct thread *td = curthread;
1189 struct sockopt sopt;
1193 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1200 sopt.sopt_level = uap->level;
1201 sopt.sopt_name = uap->name;
1202 sopt.sopt_valsize = valsize;
1204 sopt.sopt_val = NULL;
1206 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1209 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1210 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1215 error = kern_getsockopt(uap->s, &sopt);
1218 valsize = sopt.sopt_valsize;
1219 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1223 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1226 kfree(sopt.sopt_val, M_TEMP);
1231 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1232 * This allows kern_getsockname() to return a pointer to an allocated struct
1233 * sockaddr which must be freed later with FREE(). The caller must
1234 * initialize *name to NULL.
1237 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1239 struct thread *td = curthread;
1240 struct proc *p = td->td_proc;
1243 struct sockaddr *sa = NULL;
1246 error = holdsock(p->p_fd, s, &fp);
1253 so = (struct socket *)fp->f_data;
1254 error = so_pru_sockaddr(so, &sa);
1259 *namelen = MIN(*namelen, sa->sa_len);
1269 * getsockname_args(int fdes, caddr_t asa, int *alen)
1276 sys_getsockname(struct getsockname_args *uap)
1278 struct sockaddr *sa = NULL;
1281 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1285 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1288 error = copyout(sa, uap->asa, sa_len);
1290 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1292 kfree(sa, M_SONAME);
1297 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1298 * This allows kern_getpeername() to return a pointer to an allocated struct
1299 * sockaddr which must be freed later with FREE(). The caller must
1300 * initialize *name to NULL.
1303 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1305 struct thread *td = curthread;
1306 struct proc *p = td->td_proc;
1309 struct sockaddr *sa = NULL;
1312 error = holdsock(p->p_fd, s, &fp);
1319 so = (struct socket *)fp->f_data;
1320 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1324 error = so_pru_peeraddr(so, &sa);
1329 *namelen = MIN(*namelen, sa->sa_len);
1339 * getpeername_args(int fdes, caddr_t asa, int *alen)
1341 * Get name of peer for connected socket.
1346 sys_getpeername(struct getpeername_args *uap)
1348 struct sockaddr *sa = NULL;
1351 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1355 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1358 error = copyout(sa, uap->asa, sa_len);
1360 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1362 kfree(sa, M_SONAME);
1367 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1369 struct sockaddr *sa;
1373 if (len > SOCK_MAXADDRLEN)
1374 return ENAMETOOLONG;
1375 if (len < offsetof(struct sockaddr, sa_data[0]))
1377 sa = kmalloc(len, M_SONAME, M_WAITOK);
1378 error = copyin(uaddr, sa, len);
1380 kfree(sa, M_SONAME);
1382 #if BYTE_ORDER != BIG_ENDIAN
1384 * The bind(), connect(), and sendto() syscalls were not
1385 * versioned for COMPAT_43. Thus, this check must stay.
1387 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1388 sa->sa_family = sa->sa_len;
1397 * Detach a mapped page and release resources back to the system.
1398 * We must release our wiring and if the object is ripped out
1399 * from under the vm_page we become responsible for freeing the
1405 sf_buf_mfree(void *arg)
1407 struct sf_buf *sf = arg;
1410 m = sf_buf_page(sf);
1411 if (sf_buf_free(sf)) {
1412 /* sf invalid now */
1413 vm_page_busy_wait(m, FALSE, "sockpgf");
1414 vm_page_unwire(m, 0);
1415 if (m->object == NULL &&
1416 m->wire_count == 0 &&
1417 (m->flags & PG_NEED_COMMIT) == 0) {
1427 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1428 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1430 * Send a file specified by 'fd' and starting at 'offset' to a socket
1431 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1432 * nbytes == 0. Optionally add a header and/or trailer to the socket
1433 * output. If specified, write the total number of bytes sent into *sbytes.
1435 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1436 * the headers to count against the remaining bytes to be sent from
1437 * the file descriptor. We may wish to implement a compatibility syscall
1443 sys_sendfile(struct sendfile_args *uap)
1445 struct thread *td = curthread;
1446 struct proc *p = td->td_proc;
1448 struct vnode *vp = NULL;
1449 struct sf_hdtr hdtr;
1450 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1452 struct mbuf *mheader = NULL;
1455 off_t hdtr_size = 0;
1462 * Do argument checking. Must be a regular file in, stream
1463 * type and connected socket out, positive offset.
1465 fp = holdfp(p->p_fd, uap->fd, FREAD);
1469 if (fp->f_type != DTYPE_VNODE) {
1473 vp = (struct vnode *)fp->f_data;
1478 * If specified, get the pointer to the sf_hdtr struct for
1479 * any headers/trailers.
1482 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1489 error = iovec_copyin(hdtr.headers, &iov, aiov,
1490 hdtr.hdr_cnt, &hbytes);
1494 auio.uio_iovcnt = hdtr.hdr_cnt;
1495 auio.uio_offset = 0;
1496 auio.uio_segflg = UIO_USERSPACE;
1497 auio.uio_rw = UIO_WRITE;
1499 auio.uio_resid = hbytes;
1501 mheader = m_uiomove(&auio);
1503 iovec_free(&iov, aiov);
1504 if (mheader == NULL)
1509 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1510 &sbytes, uap->flags);
1515 * Send trailers. Wimp out and use writev(2).
1517 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1518 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1519 hdtr.trl_cnt, &auio.uio_resid);
1523 auio.uio_iovcnt = hdtr.trl_cnt;
1524 auio.uio_offset = 0;
1525 auio.uio_segflg = UIO_USERSPACE;
1526 auio.uio_rw = UIO_WRITE;
1529 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1531 iovec_free(&iov, aiov);
1534 hdtr_size += tbytes; /* trailer bytes successfully sent */
1540 if (uap->sbytes != NULL) {
1541 sbytes += hdtr_size;
1542 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1548 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1549 struct mbuf *mheader, off_t *sbytes, int flags)
1551 struct thread *td = curthread;
1552 struct proc *p = td->td_proc;
1553 struct vm_object *obj;
1556 struct mbuf *m, *mp;
1563 if (vp->v_type != VREG) {
1567 if ((obj = vp->v_object) == NULL) {
1571 error = holdsock(p->p_fd, sfd, &fp);
1574 so = (struct socket *)fp->f_data;
1575 if (so->so_type != SOCK_STREAM) {
1579 if ((so->so_state & SS_ISCONNECTED) == 0) {
1590 * Protect against multiple writers to the socket.
1592 ssb_lock(&so->so_snd, M_WAITOK);
1595 * Loop through the pages in the file, starting with the requested
1596 * offset. Get a file page (do I/O if necessary), map the file page
1597 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1600 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1605 pindex = OFF_TO_IDX(off);
1608 * Calculate the amount to transfer. Not to exceed a page,
1609 * the EOF, or the passed in nbytes.
1611 xfsize = vp->v_filesize - off;
1612 if (xfsize > PAGE_SIZE)
1614 pgoff = (vm_offset_t)(off & PAGE_MASK);
1615 if (PAGE_SIZE - pgoff < xfsize)
1616 xfsize = PAGE_SIZE - pgoff;
1617 if (nbytes && xfsize > (nbytes - *sbytes))
1618 xfsize = nbytes - *sbytes;
1622 * Optimize the non-blocking case by looking at the socket space
1623 * before going to the extra work of constituting the sf_buf.
1625 if ((fp->f_flag & FNONBLOCK) &&
1626 ssb_space_prealloc(&so->so_snd) <= 0) {
1627 if (so->so_state & SS_CANTSENDMORE)
1631 ssb_unlock(&so->so_snd);
1635 * Attempt to look up the page.
1637 * Allocate if not found, wait and loop if busy, then
1638 * wire the page. critical section protection is
1639 * required to maintain the object association (an
1640 * interrupt can free the page) through to the
1641 * vm_page_wire() call.
1643 vm_object_hold(obj);
1644 pg = vm_page_lookup_busy_try(obj, pindex, TRUE, &error);
1646 vm_page_sleep_busy(pg, TRUE, "sfpbsy");
1647 vm_object_drop(obj);
1651 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL |
1655 vm_object_drop(obj);
1660 vm_object_drop(obj);
1663 * If page is not valid for what we need, initiate I/O
1666 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1672 * Ensure that our page is still around when the I/O
1675 * Ensure that our page is not modified while part of
1676 * a mbuf as this could mess up tcp checksums, DMA,
1677 * etc (XXX NEEDS WORK). The softbusy is supposed to
1678 * help here but it actually doesn't.
1680 * XXX THIS HAS MULTIPLE PROBLEMS. The underlying
1681 * VM pages are not protected by the soft-busy
1682 * unless we vm_page_protect... READ them, and
1683 * they STILL aren't protected against
1684 * modification via the buffer cache (VOP_WRITE).
1686 * Fixing the second issue is particularly
1689 * XXX We also can't soft-busy anyway because it can
1690 * deadlock against the syncer doing a vfs_msync(),
1691 * vfs_msync->vmntvnodesca->vfs_msync_scan2->
1692 * vm_object_page_clean->(scan)-> ... page
1695 /*vm_page_io_start(pg);*/
1699 * Get the page from backing store.
1701 bsize = vp->v_mount->mnt_stat.f_iosize;
1702 auio.uio_iov = &aiov;
1703 auio.uio_iovcnt = 1;
1705 aiov.iov_len = MAXBSIZE;
1706 auio.uio_resid = MAXBSIZE;
1707 auio.uio_offset = trunc_page(off);
1708 auio.uio_segflg = UIO_NOCOPY;
1709 auio.uio_rw = UIO_READ;
1711 vn_lock(vp, LK_SHARED | LK_RETRY);
1712 error = VOP_READ(vp, &auio,
1713 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1716 vm_page_flag_clear(pg, PG_ZERO);
1717 vm_page_busy_wait(pg, FALSE, "sockpg");
1718 /*vm_page_io_finish(pg);*/
1720 vm_page_unwire(pg, 0);
1722 vm_page_try_to_free(pg);
1723 ssb_unlock(&so->so_snd);
1730 * Get a sendfile buf. We usually wait as long as necessary,
1731 * but this wait can be interrupted.
1733 if ((sf = sf_buf_alloc(pg)) == NULL) {
1734 vm_page_unwire(pg, 0);
1736 vm_page_try_to_free(pg);
1737 ssb_unlock(&so->so_snd);
1744 * Get an mbuf header and set it up as having external storage.
1746 MGETHDR(m, MB_WAIT, MT_DATA);
1750 ssb_unlock(&so->so_snd);
1754 m->m_ext.ext_free = sf_buf_mfree;
1755 m->m_ext.ext_ref = sf_buf_ref;
1756 m->m_ext.ext_arg = sf;
1757 m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1758 m->m_ext.ext_size = PAGE_SIZE;
1759 m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1760 m->m_flags |= M_EXT;
1761 m->m_pkthdr.len = m->m_len = xfsize;
1762 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1764 if (mheader != NULL) {
1765 hbytes = mheader->m_pkthdr.len;
1766 mheader->m_pkthdr.len += m->m_pkthdr.len;
1774 * Add the buffer to the socket buffer chain.
1779 * Make sure that the socket is still able to take more data.
1780 * CANTSENDMORE being true usually means that the connection
1781 * was closed. so_error is true when an error was sensed after
1783 * The state is checked after the page mapping and buffer
1784 * allocation above since those operations may block and make
1785 * any socket checks stale. From this point forward, nothing
1786 * blocks before the pru_send (or more accurately, any blocking
1787 * results in a loop back to here to re-check).
1789 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1790 if (so->so_state & SS_CANTSENDMORE) {
1793 error = so->so_error;
1797 ssb_unlock(&so->so_snd);
1802 * Wait for socket space to become available. We do this just
1803 * after checking the connection state above in order to avoid
1804 * a race condition with ssb_wait().
1806 space = ssb_space_prealloc(&so->so_snd);
1807 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) {
1808 if (fp->f_flag & FNONBLOCK) {
1810 ssb_unlock(&so->so_snd);
1815 error = ssb_wait(&so->so_snd);
1817 * An error from ssb_wait usually indicates that we've
1818 * been interrupted by a signal. If we've sent anything
1819 * then return bytes sent, otherwise return the error.
1823 ssb_unlock(&so->so_snd);
1830 for (mp = m; mp != NULL; mp = mp->m_next)
1831 ssb_preallocstream(&so->so_snd, mp);
1832 if (use_sendfile_async)
1833 error = so_pru_senda(so, 0, m, NULL, NULL, td);
1835 error = so_pru_send(so, 0, m, NULL, NULL, td);
1839 ssb_unlock(&so->so_snd);
1843 if (mheader != NULL) {
1844 *sbytes += mheader->m_pkthdr.len;
1846 for (mp = mheader; mp != NULL; mp = mp->m_next)
1847 ssb_preallocstream(&so->so_snd, mp);
1848 if (use_sendfile_async)
1849 error = so_pru_senda(so, 0, mheader, NULL, NULL, td);
1851 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1855 ssb_unlock(&so->so_snd);
1860 if (mheader != NULL)
1869 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1872 struct thread *td = curthread;
1873 struct filedesc *fdp = td->td_proc->p_fd;
1874 struct file *lfp = NULL;
1875 struct file *nfp = NULL;
1877 struct socket *head, *so;
1880 short fflag; /* type must match fp->f_flag */
1882 assoc_id = uap->name;
1883 error = holdsock(td->td_proc->p_fd, uap->sd, &lfp);
1888 head = (struct socket *)lfp->f_data;
1889 error = sctp_can_peel_off(head, assoc_id);
1895 * At this point we know we do have a assoc to pull
1896 * we proceed to get the fd setup. This may block
1900 fflag = lfp->f_flag;
1901 error = falloc(td->td_lwp, &nfp, &fd);
1904 * Probably ran out of file descriptors. Put the
1905 * unaccepted connection back onto the queue and
1906 * do another wakeup so some other process might
1907 * have a chance at it.
1912 uap->sysmsg_iresult = fd;
1914 so = sctp_get_peeloff(head, assoc_id, &error);
1917 * Either someone else peeled it off OR
1918 * we can't get a socket.
1922 soreference(so); /* reference needed */
1923 soclrstate(so, SS_NOFDREF | SS_COMP); /* when clearing NOFDREF */
1925 if (head->so_sigio != NULL)
1926 fsetown(fgetown(&head->so_sigio), &so->so_sigio);
1928 nfp->f_type = DTYPE_SOCKET;
1929 nfp->f_flag = fflag;
1930 nfp->f_ops = &socketops;
1935 * Assign the file pointer to the reserved descriptor, or clear
1936 * the reserved descriptor if an error occured.
1939 fsetfd(fdp, NULL, fd);
1941 fsetfd(fdp, nfp, fd);
1944 * Release explicitly held references before returning.