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 1.11 2003/08/30 18:29:23 dillon Exp $
41 #include "opt_compat.h"
42 #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>
56 #include <sys/protosw.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/signalvar.h>
61 #include <sys/vnode.h>
63 #include <sys/mount.h>
65 #include <sys/ktrace.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_pageout.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_extern.h>
73 #include <sys/file2.h>
75 static void sf_buf_init(void *arg);
76 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL)
78 static int sendit(int s, struct msghdr *mp, int flags, int *res);
79 static int recvit(int s, struct msghdr *mp, caddr_t namelenp, int *res);
81 static int do_sendfile(struct sendfile_args *uap, int compat);
82 static int getsockname1(struct getsockname_args *uap, int compat);
83 static int getpeername1(struct getpeername_args *uap, int compat);
85 static SLIST_HEAD(, sf_buf) sf_freelist;
86 static vm_offset_t sf_base;
87 static struct sf_buf *sf_bufs;
88 static int sf_buf_alloc_want;
91 * System call interface to the socket abstraction.
93 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
94 #define COMPAT_OLDSOCK
97 extern struct fileops socketops;
100 * socket_args(int domain, int type, int protocol)
103 socket(struct socket_args *uap)
105 struct thread *td = curthread;
106 struct proc *p = td->td_proc;
107 struct filedesc *fdp;
115 error = falloc(p, &fp, &fd);
119 error = socreate(uap->domain, &so, uap->type, uap->protocol, td);
121 if (fdp->fd_ofiles[fd] == fp) {
122 fdp->fd_ofiles[fd] = NULL;
126 fp->f_data = (caddr_t)so;
127 fp->f_flag = FREAD|FWRITE;
128 fp->f_ops = &socketops;
129 fp->f_type = DTYPE_SOCKET;
130 uap->sysmsg_result = fd;
137 bind1(int s, struct sockaddr *sa)
139 struct thread *td = curthread;
140 struct proc *p = td->td_proc;
145 error = holdsock(p->p_fd, s, &fp);
148 error = sobind((struct socket *)fp->f_data, sa, td);
154 * bind_args(int s, caddr_t name, int namelen)
157 bind(struct bind_args *uap)
162 error = getsockaddr(&sa, uap->name, uap->namelen);
165 error = bind1(uap->s, sa);
172 * listen_args(int s, int backlog)
176 listen(struct listen_args *uap)
178 struct thread *td = curthread;
179 struct proc *p = td->td_proc;
184 error = holdsock(p->p_fd, uap->s, &fp);
187 error = solisten((struct socket *)fp->f_data, uap->backlog, td);
193 * The second argument to accept1() is a handle to a struct sockaddr.
194 * This allows accept1() to return a pointer to an allocated struct
195 * sockaddr which must be freed later with FREE(). The caller must
196 * initialize *name to NULL.
199 accept1(int s, struct sockaddr **name, int *namelen, int *res)
201 struct thread *td = curthread;
202 struct proc *p = td->td_proc;
203 struct filedesc *fdp = p->p_fd;
204 struct file *lfp = NULL;
205 struct file *nfp = NULL;
208 struct socket *head, *so;
210 u_int fflag; /* type must match fp->f_flag */
213 if (name && namelen && *namelen < 0)
216 error = holdsock(fdp, s, &lfp);
220 head = (struct socket *)lfp->f_data;
221 if ((head->so_options & SO_ACCEPTCONN) == 0) {
226 while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) {
227 if (head->so_state & SS_CANTRCVMORE) {
228 head->so_error = ECONNABORTED;
231 if ((head->so_state & SS_NBIO) != 0) {
232 head->so_error = EWOULDBLOCK;
235 error = tsleep((caddr_t)&head->so_timeo, PCATCH, "accept", 0);
241 if (head->so_error) {
242 error = head->so_error;
249 * At this point we know that there is at least one connection
250 * ready to be accepted. Remove it from the queue prior to
251 * allocating the file descriptor for it since falloc() may
252 * block allowing another process to accept the connection
255 so = TAILQ_FIRST(&head->so_comp);
256 TAILQ_REMOVE(&head->so_comp, so, so_list);
260 error = falloc(p, &nfp, &fd);
263 * Probably ran out of file descriptors. Put the
264 * unaccepted connection back onto the queue and
265 * do another wakeup so some other process might
266 * have a chance at it.
268 TAILQ_INSERT_HEAD(&head->so_comp, so, so_list);
270 wakeup_one(&head->so_timeo);
277 /* connection has been removed from the listen queue */
278 KNOTE(&head->so_rcv.sb_sel.si_note, 0);
280 so->so_state &= ~SS_COMP;
282 if (head->so_sigio != NULL)
283 fsetown(fgetown(head->so_sigio), &so->so_sigio);
285 nfp->f_data = (caddr_t)so;
287 nfp->f_ops = &socketops;
288 nfp->f_type = DTYPE_SOCKET;
289 /* Sync socket nonblocking/async state with file flags */
290 tmp = fflag & FNONBLOCK;
291 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td);
292 tmp = fflag & FASYNC;
293 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td);
296 error = soaccept(so, &sa);
299 * Set the returned name and namelen as applicable. Set the returned
300 * namelen to 0 for older code which might ignore the return value
304 if (sa && name && namelen) {
305 if (*namelen > sa->sa_len)
306 *namelen = sa->sa_len;
315 * close the new descriptor, assuming someone hasn't ripped it
316 * out from under us. Note that *res is normally ignored if an
317 * error is returned but a syscall message will still have access
318 * to the result code.
322 if (fdp->fd_ofiles[fd] == nfp) {
323 fdp->fd_ofiles[fd] = NULL;
330 * Release explicitly held references before returning.
340 * accept_args(int s, caddr_t name, int *anamelen)
343 accept(struct accept_args *uap)
345 struct sockaddr *sa = NULL;
350 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
354 error = accept1(uap->s, &sa, &sa_len, &uap->sysmsg_result);
357 error = copyout(sa, uap->name, sa_len);
359 error = copyout(&sa_len, uap->anamelen,
360 sizeof(*uap->anamelen));
365 error = accept1(uap->s, NULL, 0, &uap->sysmsg_result);
370 #ifdef COMPAT_OLDSOCK
372 oaccept(struct accept_args *uap)
374 struct sockaddr *sa = NULL;
379 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
383 error = accept1(uap->s, &sa, &sa_len, &uap->sysmsg_result);
387 * return a namelen of zero for older code which
388 * might ignore the return value from accept.
391 copyout(&sa_len, uap->anamelen, sizeof(*uap->anamelen));
394 * Convert sa to the 4.3BSD sockaddr structure.
396 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
397 error = copyout(sa, uap->name, sa_len);
399 error = copyout(&sa_len, uap->anamelen,
400 sizeof(*uap->anamelen));
406 error = accept1(uap->s, NULL, 0, &uap->sysmsg_result);
410 #endif /* COMPAT_OLDSOCK */
413 connect1(int s, struct sockaddr *sa)
415 struct thread *td = curthread;
416 struct proc *p = td->td_proc;
421 error = holdsock(p->p_fd, s, &fp);
424 so = (struct socket *)fp->f_data;
425 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
429 error = soconnect(so, sa, td);
432 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
437 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
438 error = tsleep((caddr_t)&so->so_timeo, PCATCH, "connec", 0);
443 error = so->so_error;
448 so->so_state &= ~SS_ISCONNECTING;
449 if (error == ERESTART)
457 * connect_args(int s, caddr_t name, int namelen)
460 connect(struct connect_args *uap)
465 error = getsockaddr(&sa, uap->name, uap->namelen);
468 error = connect1(uap->s, sa);
475 * socketpair(int domain, int type, int protocol, int *rsv)
478 socketpair(struct socketpair_args *uap)
480 struct thread *td = curthread;
481 struct proc *p = td->td_proc;
482 struct filedesc *fdp;
483 struct file *fp1, *fp2;
484 struct socket *so1, *so2;
485 int fd, error, sv[2];
489 error = socreate(uap->domain, &so1, uap->type, uap->protocol, td);
492 error = socreate(uap->domain, &so2, uap->type, uap->protocol, td);
495 error = falloc(p, &fp1, &fd);
500 fp1->f_data = (caddr_t)so1;
501 error = falloc(p, &fp2, &fd);
505 fp2->f_data = (caddr_t)so2;
507 error = soconnect2(so1, so2);
510 if (uap->type == SOCK_DGRAM) {
512 * Datagram socket connection is asymmetric.
514 error = soconnect2(so2, so1);
518 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
519 fp1->f_ops = fp2->f_ops = &socketops;
520 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
521 error = copyout((caddr_t)sv, (caddr_t)uap->rsv, 2 * sizeof (int));
526 if (fdp->fd_ofiles[sv[1]] == fp2) {
527 fdp->fd_ofiles[sv[1]] = NULL;
532 if (fdp->fd_ofiles[sv[0]] == fp1) {
533 fdp->fd_ofiles[sv[0]] = NULL;
545 sendit(int s, struct msghdr *mp, int flags, int *res)
547 struct thread *td = curthread;
548 struct proc *p = td->td_proc;
553 struct mbuf *control;
558 struct iovec *ktriov = NULL;
562 error = holdsock(p->p_fd, s, &fp);
565 auio.uio_iov = mp->msg_iov;
566 auio.uio_iovcnt = mp->msg_iovlen;
567 auio.uio_segflg = UIO_USERSPACE;
568 auio.uio_rw = UIO_WRITE;
570 auio.uio_offset = 0; /* XXX */
573 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
574 if ((auio.uio_resid += iov->iov_len) < 0) {
580 error = getsockaddr(&to, mp->msg_name, mp->msg_namelen);
588 if (mp->msg_control) {
589 if (mp->msg_controllen < sizeof(struct cmsghdr)
590 #ifdef COMPAT_OLDSOCK
591 && mp->msg_flags != MSG_COMPAT
597 error = sockargs(&control, mp->msg_control,
598 mp->msg_controllen, MT_CONTROL);
601 #ifdef COMPAT_OLDSOCK
602 if (mp->msg_flags == MSG_COMPAT) {
605 M_PREPEND(control, sizeof(*cm), M_WAIT);
610 cm = mtod(control, struct cmsghdr *);
611 cm->cmsg_len = control->m_len;
612 cm->cmsg_level = SOL_SOCKET;
613 cm->cmsg_type = SCM_RIGHTS;
621 if (KTRPOINT(td, KTR_GENIO)) {
622 int iovlen = auio.uio_iovcnt * sizeof (struct iovec);
624 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
625 bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen);
629 len = auio.uio_resid;
630 so = (struct socket *)fp->f_data;
631 error = so->so_proto->pr_usrreqs->pru_sosend(so, to, &auio, 0, control,
634 if (auio.uio_resid != len && (error == ERESTART ||
635 error == EINTR || error == EWOULDBLOCK))
641 *res = len - auio.uio_resid;
643 if (ktriov != NULL) {
645 ktruio.uio_iov = ktriov;
646 ktruio.uio_resid = *res;
647 ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error);
649 FREE(ktriov, M_TEMP);
660 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
663 sendto(struct sendto_args *uap)
668 msg.msg_name = uap->to;
669 msg.msg_namelen = uap->tolen;
673 #ifdef COMPAT_OLDSOCK
676 aiov.iov_base = uap->buf;
677 aiov.iov_len = uap->len;
678 return (sendit(uap->s, &msg, uap->flags, &uap->sysmsg_result));
681 #ifdef COMPAT_OLDSOCK
683 * osend_args(int s, caddr_t buf, int len, int flags)
686 osend(struct osend_args *uap)
695 aiov.iov_base = uap->buf;
696 aiov.iov_len = uap->len;
699 return (sendit(uap->s, &msg, uap->flags, &uap->sysmsg_result));
703 * osendmsg_args(int s, caddr_t msg, int flags)
706 osendmsg(struct osendmsg_args *uap)
709 struct iovec aiov[UIO_SMALLIOV], *iov;
712 error = copyin(uap->msg, (caddr_t)&msg, sizeof (struct omsghdr));
715 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
716 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV)
718 MALLOC(iov, struct iovec *,
719 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
723 error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov,
724 (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
727 msg.msg_flags = MSG_COMPAT;
729 error = sendit(uap->s, &msg, uap->flags, &uap->sysmsg_result);
738 * sendmsg_args(int s, caddr_t msg, int flags)
741 sendmsg(struct sendmsg_args *uap)
744 struct iovec aiov[UIO_SMALLIOV], *iov;
747 error = copyin(uap->msg, (caddr_t)&msg, sizeof (msg));
750 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
751 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV)
753 MALLOC(iov, struct iovec *,
754 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
758 if (msg.msg_iovlen &&
759 (error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov,
760 (unsigned)(msg.msg_iovlen * sizeof (struct iovec)))))
763 #ifdef COMPAT_OLDSOCK
766 error = sendit(uap->s, &msg, uap->flags, &uap->sysmsg_result);
774 recvit(int s, struct msghdr *mp, caddr_t namelenp, int *res)
776 struct thread *td = curthread;
777 struct proc *p = td->td_proc;
783 struct mbuf *m, *control = 0;
786 struct sockaddr *fromsa = 0;
788 struct iovec *ktriov = NULL;
792 error = holdsock(p->p_fd, s, &fp);
795 auio.uio_iov = mp->msg_iov;
796 auio.uio_iovcnt = mp->msg_iovlen;
797 auio.uio_segflg = UIO_USERSPACE;
798 auio.uio_rw = UIO_READ;
800 auio.uio_offset = 0; /* XXX */
803 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
804 if ((auio.uio_resid += iov->iov_len) < 0) {
810 if (KTRPOINT(td, KTR_GENIO)) {
811 int iovlen = auio.uio_iovcnt * sizeof (struct iovec);
813 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
814 bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen);
818 len = auio.uio_resid;
819 so = (struct socket *)fp->f_data;
820 error = so->so_proto->pr_usrreqs->pru_soreceive(so, &fromsa, &auio,
821 (struct mbuf **)0, mp->msg_control ? &control : (struct mbuf **)0,
824 if (auio.uio_resid != len && (error == ERESTART ||
825 error == EINTR || error == EWOULDBLOCK))
829 if (ktriov != NULL) {
831 ktruio.uio_iov = ktriov;
832 ktruio.uio_resid = len - auio.uio_resid;
833 ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error);
835 FREE(ktriov, M_TEMP);
840 *res = len - auio.uio_resid;
842 len = mp->msg_namelen;
843 if (len <= 0 || fromsa == 0)
846 /* save sa_len before it is destroyed by MSG_COMPAT */
847 len = MIN(len, fromsa->sa_len);
848 #ifdef COMPAT_OLDSOCK
849 if (mp->msg_flags & MSG_COMPAT)
850 ((struct osockaddr *)fromsa)->sa_family =
853 error = copyout(fromsa,
854 (caddr_t)mp->msg_name, (unsigned)len);
858 mp->msg_namelen = len;
860 (error = copyout((caddr_t)&len, namelenp, sizeof (int)))) {
861 #ifdef COMPAT_OLDSOCK
862 if (mp->msg_flags & MSG_COMPAT)
863 error = 0; /* old recvfrom didn't check */
869 if (mp->msg_control) {
870 #ifdef COMPAT_OLDSOCK
872 * We assume that old recvmsg calls won't receive access
873 * rights and other control info, esp. as control info
874 * is always optional and those options didn't exist in 4.3.
875 * If we receive rights, trim the cmsghdr; anything else
878 if (control && mp->msg_flags & MSG_COMPAT) {
879 if (mtod(control, struct cmsghdr *)->cmsg_level !=
881 mtod(control, struct cmsghdr *)->cmsg_type !=
883 mp->msg_controllen = 0;
886 control->m_len -= sizeof (struct cmsghdr);
887 control->m_data += sizeof (struct cmsghdr);
890 len = mp->msg_controllen;
892 mp->msg_controllen = 0;
893 ctlbuf = (caddr_t) mp->msg_control;
895 while (m && len > 0) {
901 mp->msg_flags |= MSG_CTRUNC;
905 if ((error = copyout((caddr_t)mtod(m, caddr_t),
906 ctlbuf, tocopy)) != 0)
913 mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control;
918 FREE(fromsa, M_SONAME);
925 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
926 * caddr_t from, int *fromlenaddr)
929 recvfrom(struct recvfrom_args *uap)
935 if (uap->fromlenaddr) {
936 error = copyin((caddr_t)uap->fromlenaddr,
937 (caddr_t)&msg.msg_namelen, sizeof (msg.msg_namelen));
942 msg.msg_name = uap->from;
945 aiov.iov_base = uap->buf;
946 aiov.iov_len = uap->len;
948 msg.msg_flags = uap->flags;
949 return (recvit(uap->s, &msg, (caddr_t)uap->fromlenaddr, &uap->sysmsg_result));
952 #ifdef COMPAT_OLDSOCK
954 orecvfrom(struct recvfrom_args *uap)
956 uap->flags |= MSG_COMPAT;
957 return (recvfrom(uap));
961 #ifdef COMPAT_OLDSOCK
963 * struct orecv_args(int s, caddr_t buf, int len, int flags)
966 orecv(struct orecv_args *uap)
975 aiov.iov_base = uap->buf;
976 aiov.iov_len = uap->len;
978 msg.msg_flags = uap->flags;
979 return (recvit(uap->s, &msg, (caddr_t)0, &uap->sysmsg_result));
983 * Old recvmsg. This code takes advantage of the fact that the old msghdr
984 * overlays the new one, missing only the flags, and with the (old) access
985 * rights where the control fields are now.
987 * orecvmsg_args(int s, struct omsghdr *msg, int flags)
990 orecvmsg(struct orecvmsg_args *uap)
993 struct iovec aiov[UIO_SMALLIOV], *iov;
996 error = copyin((caddr_t)uap->msg, (caddr_t)&msg,
997 sizeof (struct omsghdr));
1000 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
1001 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV)
1003 MALLOC(iov, struct iovec *,
1004 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
1008 msg.msg_flags = uap->flags | MSG_COMPAT;
1009 error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov,
1010 (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
1014 error = recvit(uap->s, &msg, (caddr_t)&uap->msg->msg_namelen, &uap->sysmsg_result);
1016 if (msg.msg_controllen && error == 0)
1017 error = copyout((caddr_t)&msg.msg_controllen,
1018 (caddr_t)&uap->msg->msg_accrightslen, sizeof (int));
1027 * recvmsg_args(int s, struct msghdr *msg, int flags)
1030 recvmsg(struct recvmsg_args *uap)
1033 struct iovec aiov[UIO_SMALLIOV], *uiov, *iov;
1036 error = copyin((caddr_t)uap->msg, (caddr_t)&msg, sizeof (msg));
1039 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
1040 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV)
1042 MALLOC(iov, struct iovec *,
1043 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
1047 #ifdef COMPAT_OLDSOCK
1048 msg.msg_flags = uap->flags &~ MSG_COMPAT;
1050 msg.msg_flags = uap->flags;
1054 error = copyin((caddr_t)uiov, (caddr_t)iov,
1055 (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
1058 error = recvit(uap->s, &msg, (caddr_t)0, &uap->sysmsg_result);
1061 error = copyout((caddr_t)&msg, (caddr_t)uap->msg, sizeof(msg));
1070 * shutdown_args(int s, int how)
1074 shutdown(struct shutdown_args *uap)
1076 struct thread *td = curthread;
1077 struct proc *p = td->td_proc;
1082 error = holdsock(p->p_fd, uap->s, &fp);
1085 error = soshutdown((struct socket *)fp->f_data, uap->how);
1091 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1095 setsockopt(struct setsockopt_args *uap)
1097 struct thread *td = curthread;
1098 struct proc *p = td->td_proc;
1100 struct sockopt sopt;
1103 if (uap->val == 0 && uap->valsize != 0)
1105 if (uap->valsize < 0)
1108 error = holdsock(p->p_fd, uap->s, &fp);
1112 sopt.sopt_dir = SOPT_SET;
1113 sopt.sopt_level = uap->level;
1114 sopt.sopt_name = uap->name;
1115 sopt.sopt_val = uap->val;
1116 sopt.sopt_valsize = uap->valsize;
1118 error = sosetopt((struct socket *)fp->f_data, &sopt);
1124 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1128 getsockopt(struct getsockopt_args *uap)
1130 struct thread *td = curthread;
1131 struct proc *p = td->td_proc;
1134 struct sockopt sopt;
1136 error = holdsock(p->p_fd, uap->s, &fp);
1140 error = copyin((caddr_t)uap->avalsize, (caddr_t)&valsize,
1154 sopt.sopt_dir = SOPT_GET;
1155 sopt.sopt_level = uap->level;
1156 sopt.sopt_name = uap->name;
1157 sopt.sopt_val = uap->val;
1158 sopt.sopt_valsize = (size_t)valsize; /* checked non-negative above */
1161 error = sogetopt((struct socket *)fp->f_data, &sopt);
1163 valsize = sopt.sopt_valsize;
1164 error = copyout((caddr_t)&valsize,
1165 (caddr_t)uap->avalsize, sizeof (valsize));
1172 * getsockname_args(int fdes, caddr_t asa, int *alen)
1178 getsockname1(struct getsockname_args *uap, int compat)
1180 struct thread *td = curthread;
1181 struct proc *p = td->td_proc;
1184 struct sockaddr *sa;
1187 error = holdsock(p->p_fd, uap->fdes, &fp);
1190 error = copyin((caddr_t)uap->alen, (caddr_t)&len, sizeof (len));
1199 so = (struct socket *)fp->f_data;
1201 error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, &sa);
1209 len = MIN(len, sa->sa_len);
1210 #ifdef COMPAT_OLDSOCK
1212 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
1214 error = copyout(sa, (caddr_t)uap->asa, (u_int)len);
1217 error = copyout((caddr_t)&len, (caddr_t)uap->alen,
1227 getsockname(struct getsockname_args *uap)
1230 return (getsockname1(uap, 0));
1233 #ifdef COMPAT_OLDSOCK
1235 ogetsockname(struct getsockname_args *uap)
1238 return (getsockname1(uap, 1));
1240 #endif /* COMPAT_OLDSOCK */
1243 * getpeername_args(int fdes, caddr_t asa, int *alen)
1245 * Get name of peer for connected socket.
1249 getpeername1(struct getpeername_args *uap, int compat)
1251 struct thread *td = curthread;
1252 struct proc *p = td->td_proc;
1255 struct sockaddr *sa;
1258 error = holdsock(p->p_fd, uap->fdes, &fp);
1261 so = (struct socket *)fp->f_data;
1262 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1266 error = copyin((caddr_t)uap->alen, (caddr_t)&len, sizeof (len));
1276 error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, &sa);
1283 len = MIN(len, sa->sa_len);
1284 #ifdef COMPAT_OLDSOCK
1286 ((struct osockaddr *)sa)->sa_family =
1289 error = copyout(sa, (caddr_t)uap->asa, (u_int)len);
1293 error = copyout((caddr_t)&len, (caddr_t)uap->alen, sizeof (len));
1302 getpeername(struct getpeername_args *uap)
1304 return (getpeername1(uap, 0));
1307 #ifdef COMPAT_OLDSOCK
1309 ogetpeername(struct ogetpeername_args *uap)
1311 /* XXX uap should have type `getpeername_args *' to begin with. */
1312 return (getpeername1((struct getpeername_args *)uap, 1));
1314 #endif /* COMPAT_OLDSOCK */
1317 sockargs(mp, buf, buflen, type)
1322 struct sockaddr *sa;
1326 if ((u_int)buflen > MLEN) {
1327 #ifdef COMPAT_OLDSOCK
1328 if (type == MT_SONAME && (u_int)buflen <= 112)
1329 buflen = MLEN; /* unix domain compat. hack */
1334 m = m_get(M_WAIT, type);
1338 error = copyin(buf, mtod(m, caddr_t), (u_int)buflen);
1343 if (type == MT_SONAME) {
1344 sa = mtod(m, struct sockaddr *);
1346 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1347 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1348 sa->sa_family = sa->sa_len;
1350 sa->sa_len = buflen;
1357 getsockaddr(namp, uaddr, len)
1358 struct sockaddr **namp;
1362 struct sockaddr *sa;
1365 if (len > SOCK_MAXADDRLEN)
1366 return ENAMETOOLONG;
1367 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1368 error = copyin(uaddr, sa, len);
1372 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1373 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1374 sa->sa_family = sa->sa_len;
1383 * holdsock() - load the struct file pointer associated
1384 * with a socket into *fpp. If an error occurs, non-zero
1385 * will be returned and *fpp will be set to NULL.
1388 holdsock(fdp, fdes, fpp)
1389 struct filedesc *fdp;
1393 struct file *fp = NULL;
1396 if ((unsigned)fdes >= fdp->fd_nfiles ||
1397 (fp = fdp->fd_ofiles[fdes]) == NULL) {
1399 } else if (fp->f_type != DTYPE_SOCKET) {
1410 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
1413 sf_buf_init(void *arg)
1417 SLIST_INIT(&sf_freelist);
1418 sf_base = kmem_alloc_pageable(kernel_map, nsfbufs * PAGE_SIZE);
1419 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP, M_NOWAIT);
1420 bzero(sf_bufs, nsfbufs * sizeof(struct sf_buf));
1421 for (i = 0; i < nsfbufs; i++) {
1422 sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
1423 SLIST_INSERT_HEAD(&sf_freelist, &sf_bufs[i], free_list);
1428 * Get an sf_buf from the freelist. Will block if none are available.
1438 while ((sf = SLIST_FIRST(&sf_freelist)) == NULL) {
1439 sf_buf_alloc_want = 1;
1440 error = tsleep(&sf_freelist, PCATCH, "sfbufa", 0);
1445 SLIST_REMOVE_HEAD(&sf_freelist, free_list);
1452 #define dtosf(x) (&sf_bufs[((uintptr_t)(x) - (uintptr_t)sf_base) >> PAGE_SHIFT])
1454 sf_buf_ref(caddr_t addr, u_int size)
1459 if (sf->refcnt == 0)
1460 panic("sf_buf_ref: referencing a free sf_buf");
1465 * Lose a reference to an sf_buf. When none left, detach mapped page
1466 * and release resources back to the system.
1468 * Must be called at splimp.
1471 sf_buf_free(caddr_t addr, u_int size)
1478 if (sf->refcnt == 0)
1479 panic("sf_buf_free: freeing free sf_buf");
1481 if (sf->refcnt == 0) {
1482 pmap_qremove((vm_offset_t)addr, 1);
1485 vm_page_unwire(m, 0);
1487 * Check for the object going away on us. This can
1488 * happen since we don't hold a reference to it.
1489 * If so, we're responsible for freeing the page.
1491 if (m->wire_count == 0 && m->object == NULL)
1495 SLIST_INSERT_HEAD(&sf_freelist, sf, free_list);
1496 if (sf_buf_alloc_want) {
1497 sf_buf_alloc_want = 0;
1498 wakeup(&sf_freelist);
1505 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1506 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1508 * Send a file specified by 'fd' and starting at 'offset' to a socket
1509 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1510 * nbytes == 0. Optionally add a header and/or trailer to the socket
1511 * output. If specified, write the total number of bytes sent into *sbytes.
1514 sendfile(struct sendfile_args *uap)
1516 return (do_sendfile(uap, 0));
1521 osendfile(struct osendfile_args *uap)
1523 struct sendfile_args args;
1527 args.offset = uap->offset;
1528 args.nbytes = uap->nbytes;
1529 args.hdtr = uap->hdtr;
1530 args.sbytes = uap->sbytes;
1531 args.flags = uap->flags;
1533 return (do_sendfile(&args, 1));
1538 do_sendfile(struct sendfile_args *uap, int compat)
1540 struct thread *td = curthread;
1541 struct proc *p = td->td_proc;
1543 struct filedesc *fdp;
1545 struct vm_object *obj;
1550 struct writev_args nuap;
1551 struct sf_hdtr hdtr;
1552 off_t off, xfsize, hdtr_size, sbytes = 0;
1561 * Do argument checking. Must be a regular file in, stream
1562 * type and connected socket out, positive offset.
1564 fp = holdfp(fdp, uap->fd, FREAD);
1569 if (fp->f_type != DTYPE_VNODE) {
1573 vp = (struct vnode *)fp->f_data;
1575 if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) {
1580 error = holdsock(p->p_fd, uap->s, &fp);
1583 so = (struct socket *)fp->f_data;
1584 if (so->so_type != SOCK_STREAM) {
1588 if ((so->so_state & SS_ISCONNECTED) == 0) {
1592 if (uap->offset < 0) {
1598 * If specified, get the pointer to the sf_hdtr struct for
1599 * any headers/trailers.
1601 if (uap->hdtr != NULL) {
1602 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1606 * Send any headers. Wimp out and use writev(2).
1608 if (hdtr.headers != NULL) {
1610 nuap.iovp = hdtr.headers;
1611 nuap.iovcnt = hdtr.hdr_cnt;
1612 error = writev(&nuap);
1616 sbytes += nuap.sysmsg_result;
1618 hdtr_size += nuap.sysmsg_result;
1623 * Protect against multiple writers to the socket.
1625 (void) sblock(&so->so_snd, M_WAITOK);
1628 * Loop through the pages in the file, starting with the requested
1629 * offset. Get a file page (do I/O if necessary), map the file page
1630 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1633 for (off = uap->offset; ; off += xfsize, sbytes += xfsize) {
1637 pindex = OFF_TO_IDX(off);
1640 * Calculate the amount to transfer. Not to exceed a page,
1641 * the EOF, or the passed in nbytes.
1643 xfsize = obj->un_pager.vnp.vnp_size - off;
1644 if (xfsize > PAGE_SIZE)
1646 pgoff = (vm_offset_t)(off & PAGE_MASK);
1647 if (PAGE_SIZE - pgoff < xfsize)
1648 xfsize = PAGE_SIZE - pgoff;
1649 if (uap->nbytes && xfsize > (uap->nbytes - sbytes))
1650 xfsize = uap->nbytes - sbytes;
1654 * Optimize the non-blocking case by looking at the socket space
1655 * before going to the extra work of constituting the sf_buf.
1657 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) {
1658 if (so->so_state & SS_CANTSENDMORE)
1662 sbunlock(&so->so_snd);
1666 * Attempt to look up the page.
1668 * Allocate if not found
1670 * Wait and loop if busy.
1672 pg = vm_page_lookup(obj, pindex);
1675 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1681 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1686 * Wire the page so it does not get ripped out from under
1693 * If page is not valid for what we need, initiate I/O
1696 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1702 * Ensure that our page is still around when the I/O
1705 vm_page_io_start(pg);
1708 * Get the page from backing store.
1710 bsize = vp->v_mount->mnt_stat.f_iosize;
1711 auio.uio_iov = &aiov;
1712 auio.uio_iovcnt = 1;
1714 aiov.iov_len = MAXBSIZE;
1715 auio.uio_resid = MAXBSIZE;
1716 auio.uio_offset = trunc_page(off);
1717 auio.uio_segflg = UIO_NOCOPY;
1718 auio.uio_rw = UIO_READ;
1720 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
1721 error = VOP_READ(vp, &auio,
1722 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1724 VOP_UNLOCK(vp, 0, td);
1725 vm_page_flag_clear(pg, PG_ZERO);
1726 vm_page_io_finish(pg);
1728 vm_page_unwire(pg, 0);
1730 * See if anyone else might know about this page.
1731 * If not and it is not valid, then free it.
1733 if (pg->wire_count == 0 && pg->valid == 0 &&
1734 pg->busy == 0 && !(pg->flags & PG_BUSY) &&
1735 pg->hold_count == 0) {
1739 sbunlock(&so->so_snd);
1746 * Get a sendfile buf. We usually wait as long as necessary,
1747 * but this wait can be interrupted.
1749 if ((sf = sf_buf_alloc()) == NULL) {
1751 vm_page_unwire(pg, 0);
1752 if (pg->wire_count == 0 && pg->object == NULL)
1755 sbunlock(&so->so_snd);
1762 * Allocate a kernel virtual page and insert the physical page
1767 pmap_qenter(sf->kva, &pg, 1);
1769 * Get an mbuf header and set it up as having external storage.
1771 MGETHDR(m, M_WAIT, MT_DATA);
1774 sf_buf_free((void *)sf->kva, PAGE_SIZE);
1775 sbunlock(&so->so_snd);
1778 m->m_ext.ext_free = sf_buf_free;
1779 m->m_ext.ext_ref = sf_buf_ref;
1780 m->m_ext.ext_buf = (void *)sf->kva;
1781 m->m_ext.ext_size = PAGE_SIZE;
1782 m->m_data = (char *) sf->kva + pgoff;
1783 m->m_flags |= M_EXT;
1784 m->m_pkthdr.len = m->m_len = xfsize;
1786 * Add the buffer to the socket buffer chain.
1791 * Make sure that the socket is still able to take more data.
1792 * CANTSENDMORE being true usually means that the connection
1793 * was closed. so_error is true when an error was sensed after
1795 * The state is checked after the page mapping and buffer
1796 * allocation above since those operations may block and make
1797 * any socket checks stale. From this point forward, nothing
1798 * blocks before the pru_send (or more accurately, any blocking
1799 * results in a loop back to here to re-check).
1801 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1802 if (so->so_state & SS_CANTSENDMORE) {
1805 error = so->so_error;
1809 sbunlock(&so->so_snd);
1814 * Wait for socket space to become available. We do this just
1815 * after checking the connection state above in order to avoid
1816 * a race condition with sbwait().
1818 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1819 if (so->so_state & SS_NBIO) {
1821 sbunlock(&so->so_snd);
1826 error = sbwait(&so->so_snd);
1828 * An error from sbwait usually indicates that we've
1829 * been interrupted by a signal. If we've sent anything
1830 * then return bytes sent, otherwise return the error.
1834 sbunlock(&so->so_snd);
1841 (*so->so_proto->pr_usrreqs->pru_send)(so, 0, m, 0, 0, td);
1844 sbunlock(&so->so_snd);
1848 sbunlock(&so->so_snd);
1851 * Send trailers. Wimp out and use writev(2).
1853 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1855 nuap.iovp = hdtr.trailers;
1856 nuap.iovcnt = hdtr.trl_cnt;
1857 error = writev(&nuap);
1861 sbytes += nuap.sysmsg_result;
1863 hdtr_size += nuap.sysmsg_result;
1867 if (uap->sbytes != NULL) {
1869 sbytes += hdtr_size;
1870 copyout(&sbytes, uap->sbytes, sizeof(off_t));