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 1.92 2008/11/26 13:10:56 sephe 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>
81 #include <sys/socketvar2.h>
82 #include <net/netmsg2.h>
85 #include <netinet/sctp_peeloff.h>
93 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures");
96 * System call interface to the socket abstraction.
99 extern struct fileops socketops;
102 * socket_args(int domain, int type, int protocol)
105 kern_socket(int domain, int type, int protocol, int *res)
107 struct thread *td = curthread;
108 struct proc *p = td->td_proc;
115 error = falloc(p, &fp, &fd);
118 error = socreate(domain, &so, type, protocol, td);
122 fp->f_type = DTYPE_SOCKET;
123 fp->f_flag = FREAD | FWRITE;
124 fp->f_ops = &socketops;
137 sys_socket(struct socket_args *uap)
142 error = kern_socket(uap->domain, uap->type, uap->protocol,
143 &uap->sysmsg_iresult);
150 kern_bind(int s, struct sockaddr *sa)
152 struct thread *td = curthread;
153 struct proc *p = td->td_proc;
158 error = holdsock(p->p_fd, s, &fp);
161 error = sobind((struct socket *)fp->f_data, sa, td);
167 * bind_args(int s, caddr_t name, int namelen)
172 sys_bind(struct bind_args *uap)
177 error = getsockaddr(&sa, uap->name, uap->namelen);
181 error = kern_bind(uap->s, sa);
189 kern_listen(int s, int backlog)
191 struct thread *td = curthread;
192 struct proc *p = td->td_proc;
197 error = holdsock(p->p_fd, s, &fp);
200 error = solisten((struct socket *)fp->f_data, backlog, td);
206 * listen_args(int s, int backlog)
211 sys_listen(struct listen_args *uap)
216 error = kern_listen(uap->s, uap->backlog);
222 * Returns the accepted socket as well.
225 soaccept_predicate(struct netmsg *msg0)
227 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
228 struct socket *head = msg->nm_so;
230 if (head->so_error != 0) {
231 msg->nm_netmsg.nm_lmsg.ms_error = head->so_error;
234 if (!TAILQ_EMPTY(&head->so_comp)) {
235 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
236 msg->nm_so = TAILQ_FIRST(&head->so_comp);
237 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
240 msg->nm_netmsg.nm_lmsg.ms_error = 0;
243 if (head->so_state & SS_CANTRCVMORE) {
244 msg->nm_netmsg.nm_lmsg.ms_error = ECONNABORTED;
247 if (msg->nm_fflags & FNONBLOCK) {
248 msg->nm_netmsg.nm_lmsg.ms_error = EWOULDBLOCK;
256 * The second argument to kern_accept() is a handle to a struct sockaddr.
257 * This allows kern_accept() to return a pointer to an allocated struct
258 * sockaddr which must be freed later with FREE(). The caller must
259 * initialize *name to NULL.
262 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res)
264 struct thread *td = curthread;
265 struct proc *p = td->td_proc;
266 struct file *lfp = NULL;
267 struct file *nfp = NULL;
269 struct socket *head, *so;
270 struct netmsg_so_notify msg;
272 u_int fflag; /* type must match fp->f_flag */
276 if (name && namelen && *namelen < 0)
279 error = holdsock(p->p_fd, s, &lfp);
283 error = falloc(p, &nfp, &fd);
284 if (error) { /* Probably ran out of file descriptors. */
288 head = (struct socket *)lfp->f_data;
289 if ((head->so_options & SO_ACCEPTCONN) == 0) {
294 if (fflags & O_FBLOCKING)
295 fflags |= lfp->f_flag & ~FNONBLOCK;
296 else if (fflags & O_FNONBLOCKING)
297 fflags |= lfp->f_flag | FNONBLOCK;
299 fflags = lfp->f_flag;
301 /* optimize for uniprocessor case later XXX JH */
302 netmsg_init_abortable(&msg.nm_netmsg, head, &curthread->td_msgport,
303 0, netmsg_so_notify, netmsg_so_notify_doabort);
304 msg.nm_predicate = soaccept_predicate;
305 msg.nm_fflags = fflags;
307 msg.nm_etype = NM_REVENT;
308 error = lwkt_domsg(head->so_port, &msg.nm_netmsg.nm_lmsg, PCATCH);
313 * At this point we have the connection that's ready to be accepted.
319 /* connection has been removed from the listen queue */
320 KNOTE(&head->so_rcv.ssb_sel.si_note, 0);
322 so->so_state &= ~SS_COMP;
324 if (head->so_sigio != NULL)
325 fsetown(fgetown(head->so_sigio), &so->so_sigio);
327 nfp->f_type = DTYPE_SOCKET;
329 nfp->f_ops = &socketops;
331 /* Sync socket nonblocking/async state with file flags */
332 tmp = fflag & FNONBLOCK;
333 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL);
334 tmp = fflag & FASYNC;
335 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
338 error = soaccept(so, &sa);
341 * Set the returned name and namelen as applicable. Set the returned
342 * namelen to 0 for older code which might ignore the return value
346 if (sa && name && namelen) {
347 if (*namelen > sa->sa_len)
348 *namelen = sa->sa_len;
358 * If an error occured clear the reserved descriptor, else associate
361 * Note that *res is normally ignored if an error is returned but
362 * a syscall message will still have access to the result code.
376 * accept(int s, caddr_t name, int *anamelen)
381 sys_accept(struct accept_args *uap)
383 struct sockaddr *sa = NULL;
388 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
393 error = kern_accept(uap->s, 0, &sa, &sa_len,
394 &uap->sysmsg_iresult);
398 error = copyout(sa, uap->name, sa_len);
400 error = copyout(&sa_len, uap->anamelen,
401 sizeof(*uap->anamelen));
407 error = kern_accept(uap->s, 0, NULL, 0,
408 &uap->sysmsg_iresult);
415 * extaccept(int s, int fflags, caddr_t name, int *anamelen)
420 sys_extaccept(struct extaccept_args *uap)
422 struct sockaddr *sa = NULL;
425 int fflags = uap->flags & O_FMASK;
428 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
433 error = kern_accept(uap->s, fflags, &sa, &sa_len,
434 &uap->sysmsg_iresult);
438 error = copyout(sa, uap->name, sa_len);
440 error = copyout(&sa_len, uap->anamelen,
441 sizeof(*uap->anamelen));
447 error = kern_accept(uap->s, fflags, NULL, 0,
448 &uap->sysmsg_iresult);
456 * Returns TRUE if predicate satisfied.
459 soconnected_predicate(struct netmsg *msg0)
461 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
462 struct socket *so = msg->nm_so;
464 /* check predicate */
465 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
466 msg->nm_netmsg.nm_lmsg.ms_error = so->so_error;
474 kern_connect(int s, int fflags, struct sockaddr *sa)
476 struct thread *td = curthread;
477 struct proc *p = td->td_proc;
480 int error, interrupted = 0;
482 error = holdsock(p->p_fd, s, &fp);
485 so = (struct socket *)fp->f_data;
487 if (fflags & O_FBLOCKING)
488 /* fflags &= ~FNONBLOCK; */;
489 else if (fflags & O_FNONBLOCKING)
494 if (so->so_state & SS_ISCONNECTING) {
498 error = soconnect(so, sa, td);
501 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
505 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
506 struct netmsg_so_notify msg;
508 netmsg_init_abortable(&msg.nm_netmsg, so,
509 &curthread->td_msgport,
512 netmsg_so_notify_doabort);
513 msg.nm_predicate = soconnected_predicate;
515 msg.nm_etype = NM_REVENT;
516 error = lwkt_domsg(so->so_port, &msg.nm_netmsg.nm_lmsg, PCATCH);
517 if (error == EINTR || error == ERESTART)
521 error = so->so_error;
526 so->so_state &= ~SS_ISCONNECTING;
527 if (error == ERESTART)
535 * connect_args(int s, caddr_t name, int namelen)
540 sys_connect(struct connect_args *uap)
545 error = getsockaddr(&sa, uap->name, uap->namelen);
549 error = kern_connect(uap->s, 0, sa);
557 * connect_args(int s, int fflags, caddr_t name, int namelen)
562 sys_extconnect(struct extconnect_args *uap)
566 int fflags = uap->flags & O_FMASK;
568 error = getsockaddr(&sa, uap->name, uap->namelen);
572 error = kern_connect(uap->s, fflags, sa);
580 kern_socketpair(int domain, int type, int protocol, int *sv)
582 struct thread *td = curthread;
583 struct proc *p = td->td_proc;
584 struct file *fp1, *fp2;
585 struct socket *so1, *so2;
589 error = socreate(domain, &so1, type, protocol, td);
592 error = socreate(domain, &so2, type, protocol, td);
595 error = falloc(p, &fp1, &fd1);
600 error = falloc(p, &fp2, &fd2);
605 error = soconnect2(so1, so2);
608 if (type == SOCK_DGRAM) {
610 * Datagram socket connection is asymmetric.
612 error = soconnect2(so2, so1);
616 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
617 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
618 fp1->f_ops = fp2->f_ops = &socketops;
625 fsetfd(p, NULL, fd2);
628 fsetfd(p, NULL, fd1);
631 (void)soclose(so2, 0);
633 (void)soclose(so1, 0);
638 * socketpair(int domain, int type, int protocol, int *rsv)
643 sys_socketpair(struct socketpair_args *uap)
648 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
652 error = copyout(sockv, uap->rsv, sizeof(sockv));
657 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
658 struct mbuf *control, int flags, size_t *res)
660 struct thread *td = curthread;
661 struct lwp *lp = td->td_lwp;
662 struct proc *p = td->td_proc;
668 struct iovec *ktriov = NULL;
672 error = holdsock(p->p_fd, s, &fp);
676 if (KTRPOINT(td, KTR_GENIO)) {
677 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
679 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
680 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
684 len = auio->uio_resid;
685 so = (struct socket *)fp->f_data;
686 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
687 if (fp->f_flag & FNONBLOCK)
688 flags |= MSG_FNONBLOCKING;
690 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
692 if (auio->uio_resid != len && (error == ERESTART ||
693 error == EINTR || error == EWOULDBLOCK))
696 lwpsignal(p, lp, SIGPIPE);
699 if (ktriov != NULL) {
701 ktruio.uio_iov = ktriov;
702 ktruio.uio_resid = len - auio->uio_resid;
703 ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
705 FREE(ktriov, M_TEMP);
709 *res = len - auio->uio_resid;
715 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
720 sys_sendto(struct sendto_args *uap)
722 struct thread *td = curthread;
725 struct sockaddr *sa = NULL;
729 error = getsockaddr(&sa, uap->to, uap->tolen);
733 aiov.iov_base = uap->buf;
734 aiov.iov_len = uap->len;
735 auio.uio_iov = &aiov;
738 auio.uio_resid = uap->len;
739 auio.uio_segflg = UIO_USERSPACE;
740 auio.uio_rw = UIO_WRITE;
744 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
745 &uap->sysmsg_szresult);
754 * sendmsg_args(int s, caddr_t msg, int flags)
759 sys_sendmsg(struct sendmsg_args *uap)
761 struct thread *td = curthread;
764 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
765 struct sockaddr *sa = NULL;
766 struct mbuf *control = NULL;
769 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
774 * Conditionally copyin msg.msg_name.
777 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
785 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
790 auio.uio_iovcnt = msg.msg_iovlen;
792 auio.uio_segflg = UIO_USERSPACE;
793 auio.uio_rw = UIO_WRITE;
797 * Conditionally copyin msg.msg_control.
799 if (msg.msg_control) {
800 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
801 msg.msg_controllen > MLEN) {
805 control = m_get(MB_WAIT, MT_CONTROL);
806 if (control == NULL) {
810 control->m_len = msg.msg_controllen;
811 error = copyin(msg.msg_control, mtod(control, caddr_t),
820 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
821 &uap->sysmsg_szresult);
825 iovec_free(&iov, aiov);
833 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
834 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
835 * Don't forget to FREE() and m_free() these if they are returned.
838 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
839 struct mbuf **control, int *flags, size_t *res)
841 struct thread *td = curthread;
842 struct proc *p = td->td_proc;
849 struct iovec *ktriov = NULL;
853 error = holdsock(p->p_fd, s, &fp);
857 if (KTRPOINT(td, KTR_GENIO)) {
858 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
860 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
861 bcopy(auio->uio_iov, ktriov, iovlen);
865 len = auio->uio_resid;
866 so = (struct socket *)fp->f_data;
868 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
869 if (fp->f_flag & FNONBLOCK) {
871 *flags |= MSG_FNONBLOCKING;
873 lflags = MSG_FNONBLOCKING;
879 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
881 if (auio->uio_resid != len && (error == ERESTART ||
882 error == EINTR || error == EWOULDBLOCK))
886 if (ktriov != NULL) {
888 ktruio.uio_iov = ktriov;
889 ktruio.uio_resid = len - auio->uio_resid;
890 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
892 FREE(ktriov, M_TEMP);
896 *res = len - auio->uio_resid;
902 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
903 * caddr_t from, int *fromlenaddr)
908 sys_recvfrom(struct recvfrom_args *uap)
910 struct thread *td = curthread;
913 struct sockaddr *sa = NULL;
916 if (uap->from && uap->fromlenaddr) {
917 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
925 aiov.iov_base = uap->buf;
926 aiov.iov_len = uap->len;
927 auio.uio_iov = &aiov;
930 auio.uio_resid = uap->len;
931 auio.uio_segflg = UIO_USERSPACE;
932 auio.uio_rw = UIO_READ;
936 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
937 &uap->flags, &uap->sysmsg_szresult);
940 if (error == 0 && uap->from) {
941 /* note: sa may still be NULL */
943 fromlen = MIN(fromlen, sa->sa_len);
944 error = copyout(sa, uap->from, fromlen);
949 error = copyout(&fromlen, uap->fromlenaddr,
960 * recvmsg_args(int s, struct msghdr *msg, int flags)
965 sys_recvmsg(struct recvmsg_args *uap)
967 struct thread *td = curthread;
970 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
971 struct mbuf *m, *control = NULL;
972 struct sockaddr *sa = NULL;
974 socklen_t *ufromlenp, *ucontrollenp;
975 int error, fromlen, controllen, len, flags, *uflagsp;
978 * This copyin handles everything except the iovec.
980 error = copyin(uap->msg, &msg, sizeof(msg));
984 if (msg.msg_name && msg.msg_namelen < 0)
986 if (msg.msg_control && msg.msg_controllen < 0)
989 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
991 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
993 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
999 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1004 auio.uio_iovcnt = msg.msg_iovlen;
1005 auio.uio_offset = 0;
1006 auio.uio_segflg = UIO_USERSPACE;
1007 auio.uio_rw = UIO_READ;
1013 error = kern_recvmsg(uap->s,
1014 (msg.msg_name ? &sa : NULL), &auio,
1015 (msg.msg_control ? &control : NULL), &flags,
1016 &uap->sysmsg_szresult);
1020 * Conditionally copyout the name and populate the namelen field.
1022 if (error == 0 && msg.msg_name) {
1023 /* note: sa may still be NULL */
1025 fromlen = MIN(msg.msg_namelen, sa->sa_len);
1026 error = copyout(sa, msg.msg_name, fromlen);
1031 error = copyout(&fromlen, ufromlenp,
1032 sizeof(*ufromlenp));
1036 * Copyout msg.msg_control and msg.msg_controllen.
1038 if (error == 0 && msg.msg_control) {
1039 len = msg.msg_controllen;
1041 ctlbuf = (caddr_t)msg.msg_control;
1043 while(m && len > 0) {
1044 unsigned int tocopy;
1046 if (len >= m->m_len) {
1049 msg.msg_flags |= MSG_CTRUNC;
1053 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1061 controllen = ctlbuf - (caddr_t)msg.msg_control;
1062 error = copyout(&controllen, ucontrollenp,
1063 sizeof(*ucontrollenp));
1067 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1072 iovec_free(&iov, aiov);
1079 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1080 * in kernel pointer instead of a userland pointer. This allows us
1081 * to manipulate socket options in the emulation code.
1084 kern_setsockopt(int s, struct sockopt *sopt)
1086 struct thread *td = curthread;
1087 struct proc *p = td->td_proc;
1091 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1093 if (sopt->sopt_valsize < 0)
1096 error = holdsock(p->p_fd, s, &fp);
1100 error = sosetopt((struct socket *)fp->f_data, sopt);
1106 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1111 sys_setsockopt(struct setsockopt_args *uap)
1113 struct thread *td = curthread;
1114 struct sockopt sopt;
1117 sopt.sopt_level = uap->level;
1118 sopt.sopt_name = uap->name;
1119 sopt.sopt_valsize = uap->valsize;
1121 sopt.sopt_val = NULL;
1123 if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1126 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1127 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1133 error = kern_setsockopt(uap->s, &sopt);
1137 kfree(sopt.sopt_val, M_TEMP);
1142 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1143 * in kernel pointer instead of a userland pointer. This allows us
1144 * to manipulate socket options in the emulation code.
1147 kern_getsockopt(int s, struct sockopt *sopt)
1149 struct thread *td = curthread;
1150 struct proc *p = td->td_proc;
1154 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1156 if (sopt->sopt_valsize < 0 || sopt->sopt_valsize > SOMAXOPT_SIZE)
1159 error = holdsock(p->p_fd, s, &fp);
1163 error = sogetopt((struct socket *)fp->f_data, sopt);
1169 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1174 sys_getsockopt(struct getsockopt_args *uap)
1176 struct thread *td = curthread;
1177 struct sockopt sopt;
1181 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1188 sopt.sopt_level = uap->level;
1189 sopt.sopt_name = uap->name;
1190 sopt.sopt_valsize = valsize;
1192 sopt.sopt_val = NULL;
1194 if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1197 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1198 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1204 error = kern_getsockopt(uap->s, &sopt);
1208 valsize = sopt.sopt_valsize;
1209 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1213 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1216 kfree(sopt.sopt_val, M_TEMP);
1221 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1222 * This allows kern_getsockname() to return a pointer to an allocated struct
1223 * sockaddr which must be freed later with FREE(). The caller must
1224 * initialize *name to NULL.
1227 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1229 struct thread *td = curthread;
1230 struct proc *p = td->td_proc;
1233 struct sockaddr *sa = NULL;
1236 error = holdsock(p->p_fd, s, &fp);
1243 so = (struct socket *)fp->f_data;
1244 error = so_pru_sockaddr(so, &sa);
1249 *namelen = MIN(*namelen, sa->sa_len);
1259 * getsockname_args(int fdes, caddr_t asa, int *alen)
1266 sys_getsockname(struct getsockname_args *uap)
1268 struct sockaddr *sa = NULL;
1271 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1276 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1280 error = copyout(sa, uap->asa, sa_len);
1282 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1289 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1290 * This allows kern_getpeername() to return a pointer to an allocated struct
1291 * sockaddr which must be freed later with FREE(). The caller must
1292 * initialize *name to NULL.
1295 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1297 struct thread *td = curthread;
1298 struct proc *p = td->td_proc;
1301 struct sockaddr *sa = NULL;
1304 error = holdsock(p->p_fd, s, &fp);
1311 so = (struct socket *)fp->f_data;
1312 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1316 error = so_pru_peeraddr(so, &sa);
1321 *namelen = MIN(*namelen, sa->sa_len);
1331 * getpeername_args(int fdes, caddr_t asa, int *alen)
1333 * Get name of peer for connected socket.
1338 sys_getpeername(struct getpeername_args *uap)
1340 struct sockaddr *sa = NULL;
1343 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1348 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1352 error = copyout(sa, uap->asa, sa_len);
1354 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1361 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1363 struct sockaddr *sa;
1367 if (len > SOCK_MAXADDRLEN)
1368 return ENAMETOOLONG;
1369 if (len < offsetof(struct sockaddr, sa_data[0]))
1371 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1372 error = copyin(uaddr, sa, len);
1376 #if BYTE_ORDER != BIG_ENDIAN
1378 * The bind(), connect(), and sendto() syscalls were not
1379 * versioned for COMPAT_43. Thus, this check must stay.
1381 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1382 sa->sa_family = sa->sa_len;
1391 * Detach a mapped page and release resources back to the system.
1392 * We must release our wiring and if the object is ripped out
1393 * from under the vm_page we become responsible for freeing the
1394 * page. These routines must be MPSAFE.
1396 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1398 * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1401 sf_buf_mref(void *arg)
1403 struct sfbuf_mref *sfm = arg;
1406 * We must already hold a ref so there is no race to 0, just
1407 * atomically increment the count.
1409 atomic_add_int(&sfm->mref_count, 1);
1413 sf_buf_mfree(void *arg)
1415 struct sfbuf_mref *sfm = arg;
1418 KKASSERT(sfm->mref_count > 0);
1419 if (atomic_fetchadd_int(&sfm->mref_count, -1) == 1) {
1421 * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1425 m = sf_buf_page(sfm->sf);
1426 sf_buf_free(sfm->sf);
1427 vm_page_unwire(m, 0);
1428 if (m->wire_count == 0 && m->object == NULL)
1429 vm_page_try_to_free(m);
1432 kfree(sfm, M_SENDFILE);
1438 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1439 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1441 * Send a file specified by 'fd' and starting at 'offset' to a socket
1442 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1443 * nbytes == 0. Optionally add a header and/or trailer to the socket
1444 * output. If specified, write the total number of bytes sent into *sbytes.
1446 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1447 * the headers to count against the remaining bytes to be sent from
1448 * the file descriptor. We may wish to implement a compatibility syscall
1454 sys_sendfile(struct sendfile_args *uap)
1456 struct thread *td = curthread;
1457 struct proc *p = td->td_proc;
1459 struct vnode *vp = NULL;
1460 struct sf_hdtr hdtr;
1461 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1463 struct mbuf *mheader = NULL;
1466 off_t hdtr_size = 0;
1473 * Do argument checking. Must be a regular file in, stream
1474 * type and connected socket out, positive offset.
1476 fp = holdfp(p->p_fd, uap->fd, FREAD);
1480 if (fp->f_type != DTYPE_VNODE) {
1485 vp = (struct vnode *)fp->f_data;
1490 * If specified, get the pointer to the sf_hdtr struct for
1491 * any headers/trailers.
1494 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1501 error = iovec_copyin(hdtr.headers, &iov, aiov,
1502 hdtr.hdr_cnt, &hbytes);
1506 auio.uio_iovcnt = hdtr.hdr_cnt;
1507 auio.uio_offset = 0;
1508 auio.uio_segflg = UIO_USERSPACE;
1509 auio.uio_rw = UIO_WRITE;
1511 auio.uio_resid = hbytes;
1513 mheader = m_uiomove(&auio);
1515 iovec_free(&iov, aiov);
1516 if (mheader == NULL)
1521 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1522 &sbytes, uap->flags);
1527 * Send trailers. Wimp out and use writev(2).
1529 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1530 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1531 hdtr.trl_cnt, &auio.uio_resid);
1535 auio.uio_iovcnt = hdtr.trl_cnt;
1536 auio.uio_offset = 0;
1537 auio.uio_segflg = UIO_USERSPACE;
1538 auio.uio_rw = UIO_WRITE;
1541 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1543 iovec_free(&iov, aiov);
1546 hdtr_size += tbytes; /* trailer bytes successfully sent */
1553 if (uap->sbytes != NULL) {
1554 sbytes += hdtr_size;
1555 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1561 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1562 struct mbuf *mheader, off_t *sbytes, int flags)
1564 struct thread *td = curthread;
1565 struct proc *p = td->td_proc;
1566 struct vm_object *obj;
1571 struct sfbuf_mref *sfm;
1577 if (vp->v_type != VREG) {
1581 if ((obj = vp->v_object) == NULL) {
1585 error = holdsock(p->p_fd, sfd, &fp);
1588 so = (struct socket *)fp->f_data;
1589 if (so->so_type != SOCK_STREAM) {
1593 if ((so->so_state & SS_ISCONNECTED) == 0) {
1604 * Protect against multiple writers to the socket.
1606 ssb_lock(&so->so_snd, M_WAITOK);
1609 * Loop through the pages in the file, starting with the requested
1610 * offset. Get a file page (do I/O if necessary), map the file page
1611 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1614 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1618 pindex = OFF_TO_IDX(off);
1621 * Calculate the amount to transfer. Not to exceed a page,
1622 * the EOF, or the passed in nbytes.
1624 xfsize = vp->v_filesize - off;
1625 if (xfsize > PAGE_SIZE)
1627 pgoff = (vm_offset_t)(off & PAGE_MASK);
1628 if (PAGE_SIZE - pgoff < xfsize)
1629 xfsize = PAGE_SIZE - pgoff;
1630 if (nbytes && xfsize > (nbytes - *sbytes))
1631 xfsize = nbytes - *sbytes;
1635 * Optimize the non-blocking case by looking at the socket space
1636 * before going to the extra work of constituting the sf_buf.
1638 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) {
1639 if (so->so_state & SS_CANTSENDMORE)
1643 ssb_unlock(&so->so_snd);
1647 * Attempt to look up the page.
1649 * Allocate if not found, wait and loop if busy, then
1650 * wire the page. critical section protection is
1651 * required to maintain the object association (an
1652 * interrupt can free the page) through to the
1653 * vm_page_wire() call.
1656 pg = vm_page_lookup(obj, pindex);
1658 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1665 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1673 * If page is not valid for what we need, initiate I/O
1676 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1682 * Ensure that our page is still around when the I/O
1685 vm_page_io_start(pg);
1688 * Get the page from backing store.
1690 bsize = vp->v_mount->mnt_stat.f_iosize;
1691 auio.uio_iov = &aiov;
1692 auio.uio_iovcnt = 1;
1694 aiov.iov_len = MAXBSIZE;
1695 auio.uio_resid = MAXBSIZE;
1696 auio.uio_offset = trunc_page(off);
1697 auio.uio_segflg = UIO_NOCOPY;
1698 auio.uio_rw = UIO_READ;
1700 vn_lock(vp, LK_SHARED | LK_RETRY);
1701 error = VOP_READ(vp, &auio,
1702 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1705 vm_page_flag_clear(pg, PG_ZERO);
1706 vm_page_io_finish(pg);
1709 vm_page_unwire(pg, 0);
1710 vm_page_try_to_free(pg);
1712 ssb_unlock(&so->so_snd);
1719 * Get a sendfile buf. We usually wait as long as necessary,
1720 * but this wait can be interrupted.
1722 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1724 vm_page_unwire(pg, 0);
1725 vm_page_try_to_free(pg);
1727 ssb_unlock(&so->so_snd);
1733 * Get an mbuf header and set it up as having external storage.
1735 MGETHDR(m, MB_WAIT, MT_DATA);
1739 ssb_unlock(&so->so_snd);
1744 * sfm is a temporary hack, use a per-cpu cache for this.
1746 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1748 sfm->mref_count = 1;
1750 m->m_ext.ext_free = sf_buf_mfree;
1751 m->m_ext.ext_ref = sf_buf_mref;
1752 m->m_ext.ext_arg = sfm;
1753 m->m_ext.ext_buf = (void *)sf->kva;
1754 m->m_ext.ext_size = PAGE_SIZE;
1755 m->m_data = (char *) sf->kva + pgoff;
1756 m->m_flags |= M_EXT;
1757 m->m_pkthdr.len = m->m_len = xfsize;
1758 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1760 if (mheader != NULL) {
1761 hbytes = mheader->m_pkthdr.len;
1762 mheader->m_pkthdr.len += m->m_pkthdr.len;
1770 * Add the buffer to the socket buffer chain.
1775 * Make sure that the socket is still able to take more data.
1776 * CANTSENDMORE being true usually means that the connection
1777 * was closed. so_error is true when an error was sensed after
1779 * The state is checked after the page mapping and buffer
1780 * allocation above since those operations may block and make
1781 * any socket checks stale. From this point forward, nothing
1782 * blocks before the pru_send (or more accurately, any blocking
1783 * results in a loop back to here to re-check).
1785 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1786 if (so->so_state & SS_CANTSENDMORE) {
1789 error = so->so_error;
1793 ssb_unlock(&so->so_snd);
1798 * Wait for socket space to become available. We do this just
1799 * after checking the connection state above in order to avoid
1800 * a race condition with ssb_wait().
1802 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) {
1803 if (fp->f_flag & FNONBLOCK) {
1805 ssb_unlock(&so->so_snd);
1810 error = ssb_wait(&so->so_snd);
1812 * An error from ssb_wait usually indicates that we've
1813 * been interrupted by a signal. If we've sent anything
1814 * then return bytes sent, otherwise return the error.
1818 ssb_unlock(&so->so_snd);
1824 error = so_pru_send(so, 0, m, NULL, NULL, td);
1827 ssb_unlock(&so->so_snd);
1831 if (mheader != NULL) {
1832 *sbytes += mheader->m_pkthdr.len;
1833 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1836 ssb_unlock(&so->so_snd);
1841 if (mheader != NULL)
1850 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1853 struct thread *td = curthread;
1854 struct proc *p = td->td_proc;
1855 struct file *lfp = NULL;
1856 struct file *nfp = NULL;
1858 struct socket *head, *so;
1861 short fflag; /* type must match fp->f_flag */
1863 assoc_id = uap->name;
1864 error = holdsock(p->p_fd, uap->sd, &lfp);
1870 head = (struct socket *)lfp->f_data;
1871 error = sctp_can_peel_off(head, assoc_id);
1877 * At this point we know we do have a assoc to pull
1878 * we proceed to get the fd setup. This may block
1882 fflag = lfp->f_flag;
1883 error = falloc(p, &nfp, &fd);
1886 * Probably ran out of file descriptors. Put the
1887 * unaccepted connection back onto the queue and
1888 * do another wakeup so some other process might
1889 * have a chance at it.
1894 uap->sysmsg_iresult = fd;
1896 so = sctp_get_peeloff(head, assoc_id, &error);
1899 * Either someone else peeled it off OR
1900 * we can't get a socket.
1904 so->so_state &= ~SS_COMP;
1905 so->so_state &= ~SS_NOFDREF;
1907 if (head->so_sigio != NULL)
1908 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1910 nfp->f_type = DTYPE_SOCKET;
1911 nfp->f_flag = fflag;
1912 nfp->f_ops = &socketops;
1917 * Assign the file pointer to the reserved descriptor, or clear
1918 * the reserved descriptor if an error occured.
1921 fsetfd(p, NULL, fd);
1926 * Release explicitly held references before returning.