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 1.4 2003/06/25 03:55:57 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 __P((int s, struct msghdr *mp, int flags));
79 static int recvit __P((int s, struct msghdr *mp,
82 static int accept1 __P((struct accept_args *uap, int compat));
83 static int do_sendfile __P((struct sendfile_args *uap, int compat));
84 static int getsockname1 __P((struct getsockname_args *uap, int compat));
85 static int getpeername1 __P((struct getpeername_args *uap, int compat));
87 static SLIST_HEAD(, sf_buf) sf_freelist;
88 static vm_offset_t sf_base;
89 static struct sf_buf *sf_bufs;
90 static int sf_buf_alloc_want;
93 * System call interface to the socket abstraction.
95 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
96 #define COMPAT_OLDSOCK
99 extern struct fileops socketops;
102 * socket_args(int domain, int type, int protocol)
105 socket(struct socket_args *uap)
107 struct thread *td = curthread;
108 struct proc *p = td->td_proc;
109 struct filedesc *fdp;
117 error = falloc(p, &fp, &fd);
121 error = socreate(uap->domain, &so, uap->type, uap->protocol, td);
123 if (fdp->fd_ofiles[fd] == fp) {
124 fdp->fd_ofiles[fd] = NULL;
128 fp->f_data = (caddr_t)so;
129 fp->f_flag = FREAD|FWRITE;
130 fp->f_ops = &socketops;
131 fp->f_type = DTYPE_SOCKET;
139 * bind_args(int s, caddr_t name, int namelen)
144 bind(struct bind_args *uap)
146 struct thread *td = curthread;
147 struct proc *p = td->td_proc;
153 error = holdsock(p->p_fd, uap->s, &fp);
156 error = getsockaddr(&sa, uap->name, uap->namelen);
161 error = sobind((struct socket *)fp->f_data, sa, td);
168 * listen_args(int s, int backlog)
172 listen(struct listen_args *uap)
174 struct thread *td = curthread;
175 struct proc *p = td->td_proc;
180 error = holdsock(p->p_fd, uap->s, &fp);
183 error = solisten((struct socket *)fp->f_data, uap->backlog, td);
189 * accept_args(int s, caddr_t name, int *anamelen)
192 accept1(struct accept_args *uap, int compat)
194 struct thread *td = curthread;
195 struct proc *p = td->td_proc;
196 struct filedesc *fdp = p->p_fd;
197 struct file *lfp = NULL;
198 struct file *nfp = NULL;
200 int namelen, error, s;
201 struct socket *head, *so;
203 u_int fflag; /* type must match fp->f_flag */
207 error = copyin((caddr_t)uap->anamelen, (caddr_t)&namelen,
214 error = holdsock(fdp, uap->s, &lfp);
218 head = (struct socket *)lfp->f_data;
219 if ((head->so_options & SO_ACCEPTCONN) == 0) {
224 while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) {
225 if (head->so_state & SS_CANTRCVMORE) {
226 head->so_error = ECONNABORTED;
229 if ((head->so_state & SS_NBIO) != 0) {
230 head->so_error = EWOULDBLOCK;
233 error = tsleep((caddr_t)&head->so_timeo, PSOCK | PCATCH,
240 if (head->so_error) {
241 error = head->so_error;
248 * At this point we know that there is at least one connection
249 * ready to be accepted. Remove it from the queue prior to
250 * allocating the file descriptor for it since falloc() may
251 * block allowing another process to accept the connection
254 so = TAILQ_FIRST(&head->so_comp);
255 TAILQ_REMOVE(&head->so_comp, so, so_list);
259 error = falloc(p, &nfp, &fd);
262 * Probably ran out of file descriptors. Put the
263 * unaccepted connection back onto the queue and
264 * do another wakeup so some other process might
265 * have a chance at it.
267 TAILQ_INSERT_HEAD(&head->so_comp, so, so_list);
269 wakeup_one(&head->so_timeo);
276 /* connection has been removed from the listen queue */
277 KNOTE(&head->so_rcv.sb_sel.si_note, 0);
279 so->so_state &= ~SS_COMP;
281 if (head->so_sigio != NULL)
282 fsetown(fgetown(head->so_sigio), &so->so_sigio);
284 nfp->f_data = (caddr_t)so;
286 nfp->f_ops = &socketops;
287 nfp->f_type = DTYPE_SOCKET;
288 /* Sync socket nonblocking/async state with file flags */
289 tmp = fflag & FNONBLOCK;
290 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td);
291 tmp = fflag & FASYNC;
292 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td);
294 error = soaccept(so, &sa);
297 * return a namelen of zero for older code which might
298 * ignore the return value from accept.
300 if (uap->name != NULL) {
302 (void) copyout((caddr_t)&namelen,
303 (caddr_t)uap->anamelen, sizeof(*uap->anamelen));
316 /* check sa_len before it is destroyed */
317 if (namelen > sa->sa_len)
318 namelen = sa->sa_len;
319 #ifdef COMPAT_OLDSOCK
321 ((struct osockaddr *)sa)->sa_family =
324 error = copyout(sa, (caddr_t)uap->name, (u_int)namelen);
327 error = copyout((caddr_t)&namelen,
328 (caddr_t)uap->anamelen, sizeof (*uap->anamelen));
335 * close the new descriptor, assuming someone hasn't ripped it
339 if (fdp->fd_ofiles[fd] == nfp) {
340 fdp->fd_ofiles[fd] = NULL;
347 * Release explicitly held references before returning.
357 accept(struct accept_args *uap)
359 return (accept1(uap, 0));
362 #ifdef COMPAT_OLDSOCK
364 oaccept(struct accept_args *uap)
367 return (accept1(uap, 1));
369 #endif /* COMPAT_OLDSOCK */
372 * connect_args(int s, caddr_t name, int namelen)
376 connect(struct connect_args *uap)
378 struct thread *td = curthread;
379 struct proc *p = td->td_proc;
385 error = holdsock(p->p_fd, uap->s, &fp);
388 so = (struct socket *)fp->f_data;
389 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
393 error = getsockaddr(&sa, uap->name, uap->namelen);
396 error = soconnect(so, sa, td);
399 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
405 while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
406 error = tsleep((caddr_t)&so->so_timeo, PSOCK | PCATCH,
412 error = so->so_error;
417 so->so_state &= ~SS_ISCONNECTING;
419 if (error == ERESTART)
427 * socketpair(int domain, int type, int protocol, int *rsv)
430 socketpair(struct socketpair_args *uap)
432 struct thread *td = curthread;
433 struct proc *p = td->td_proc;
434 struct filedesc *fdp;
435 struct file *fp1, *fp2;
436 struct socket *so1, *so2;
437 int fd, error, sv[2];
441 error = socreate(uap->domain, &so1, uap->type, uap->protocol, td);
444 error = socreate(uap->domain, &so2, uap->type, uap->protocol, td);
447 error = falloc(p, &fp1, &fd);
452 fp1->f_data = (caddr_t)so1;
453 error = falloc(p, &fp2, &fd);
457 fp2->f_data = (caddr_t)so2;
459 error = soconnect2(so1, so2);
462 if (uap->type == SOCK_DGRAM) {
464 * Datagram socket connection is asymmetric.
466 error = soconnect2(so2, so1);
470 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
471 fp1->f_ops = fp2->f_ops = &socketops;
472 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
473 error = copyout((caddr_t)sv, (caddr_t)uap->rsv, 2 * sizeof (int));
478 if (fdp->fd_ofiles[sv[1]] == fp2) {
479 fdp->fd_ofiles[sv[1]] = NULL;
484 if (fdp->fd_ofiles[sv[0]] == fp1) {
485 fdp->fd_ofiles[sv[0]] = NULL;
497 sendit(int s, struct msghdr *mp, int flags)
499 struct thread *td = curthread;
500 struct proc *p = td->td_proc;
503 register struct iovec *iov;
505 struct mbuf *control;
510 struct iovec *ktriov = NULL;
514 error = holdsock(p->p_fd, s, &fp);
517 auio.uio_iov = mp->msg_iov;
518 auio.uio_iovcnt = mp->msg_iovlen;
519 auio.uio_segflg = UIO_USERSPACE;
520 auio.uio_rw = UIO_WRITE;
522 auio.uio_offset = 0; /* XXX */
525 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
526 if ((auio.uio_resid += iov->iov_len) < 0) {
532 error = getsockaddr(&to, mp->msg_name, mp->msg_namelen);
540 if (mp->msg_control) {
541 if (mp->msg_controllen < sizeof(struct cmsghdr)
542 #ifdef COMPAT_OLDSOCK
543 && mp->msg_flags != MSG_COMPAT
549 error = sockargs(&control, mp->msg_control,
550 mp->msg_controllen, MT_CONTROL);
553 #ifdef COMPAT_OLDSOCK
554 if (mp->msg_flags == MSG_COMPAT) {
555 register struct cmsghdr *cm;
557 M_PREPEND(control, sizeof(*cm), M_WAIT);
562 cm = mtod(control, struct cmsghdr *);
563 cm->cmsg_len = control->m_len;
564 cm->cmsg_level = SOL_SOCKET;
565 cm->cmsg_type = SCM_RIGHTS;
573 if (KTRPOINT(td, KTR_GENIO)) {
574 int iovlen = auio.uio_iovcnt * sizeof (struct iovec);
576 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
577 bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen);
581 len = auio.uio_resid;
582 so = (struct socket *)fp->f_data;
583 error = so->so_proto->pr_usrreqs->pru_sosend(so, to, &auio, 0, control,
586 if (auio.uio_resid != len && (error == ERESTART ||
587 error == EINTR || error == EWOULDBLOCK))
593 p->p_retval[0] = len - auio.uio_resid;
595 if (ktriov != NULL) {
597 ktruio.uio_iov = ktriov;
598 ktruio.uio_resid = p->p_retval[0];
599 ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error);
601 FREE(ktriov, M_TEMP);
612 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
615 sendto(struct sendto_args *uap)
620 msg.msg_name = uap->to;
621 msg.msg_namelen = uap->tolen;
625 #ifdef COMPAT_OLDSOCK
628 aiov.iov_base = uap->buf;
629 aiov.iov_len = uap->len;
630 return (sendit(uap->s, &msg, uap->flags));
633 #ifdef COMPAT_OLDSOCK
635 * osend_args(int s, caddr_t buf, int len, int flags)
638 osend(struct osend_args *uap)
647 aiov.iov_base = uap->buf;
648 aiov.iov_len = uap->len;
651 return (sendit(uap->s, &msg, uap->flags));
655 * osendmsg_args(int s, caddr_t msg, int flags)
658 osendmsg(struct osendmsg_args *uap)
661 struct iovec aiov[UIO_SMALLIOV], *iov;
664 error = copyin(uap->msg, (caddr_t)&msg, sizeof (struct omsghdr));
667 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
668 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV)
670 MALLOC(iov, struct iovec *,
671 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
675 error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov,
676 (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
679 msg.msg_flags = MSG_COMPAT;
681 error = sendit(uap->s, &msg, uap->flags);
690 * sendmsg_args(int s, caddr_t msg, int flags)
693 sendmsg(struct sendmsg_args *uap)
696 struct iovec aiov[UIO_SMALLIOV], *iov;
699 error = copyin(uap->msg, (caddr_t)&msg, sizeof (msg));
702 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
703 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV)
705 MALLOC(iov, struct iovec *,
706 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
710 if (msg.msg_iovlen &&
711 (error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov,
712 (unsigned)(msg.msg_iovlen * sizeof (struct iovec)))))
715 #ifdef COMPAT_OLDSOCK
718 error = sendit(uap->s, &msg, uap->flags);
726 recvit(int s, struct msghdr *mp, caddr_t namelenp)
728 struct thread *td = curthread;
729 struct proc *p = td->td_proc;
732 register struct iovec *iov;
735 struct mbuf *m, *control = 0;
738 struct sockaddr *fromsa = 0;
740 struct iovec *ktriov = NULL;
744 error = holdsock(p->p_fd, s, &fp);
747 auio.uio_iov = mp->msg_iov;
748 auio.uio_iovcnt = mp->msg_iovlen;
749 auio.uio_segflg = UIO_USERSPACE;
750 auio.uio_rw = UIO_READ;
752 auio.uio_offset = 0; /* XXX */
755 for (i = 0; i < mp->msg_iovlen; i++, iov++) {
756 if ((auio.uio_resid += iov->iov_len) < 0) {
762 if (KTRPOINT(td, KTR_GENIO)) {
763 int iovlen = auio.uio_iovcnt * sizeof (struct iovec);
765 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
766 bcopy((caddr_t)auio.uio_iov, (caddr_t)ktriov, iovlen);
770 len = auio.uio_resid;
771 so = (struct socket *)fp->f_data;
772 error = so->so_proto->pr_usrreqs->pru_soreceive(so, &fromsa, &auio,
773 (struct mbuf **)0, mp->msg_control ? &control : (struct mbuf **)0,
776 if (auio.uio_resid != len && (error == ERESTART ||
777 error == EINTR || error == EWOULDBLOCK))
781 if (ktriov != NULL) {
783 ktruio.uio_iov = ktriov;
784 ktruio.uio_resid = len - auio.uio_resid;
785 ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error);
787 FREE(ktriov, M_TEMP);
792 p->p_retval[0] = len - auio.uio_resid;
794 len = mp->msg_namelen;
795 if (len <= 0 || fromsa == 0)
799 #define MIN(a,b) ((a)>(b)?(b):(a))
801 /* save sa_len before it is destroyed by MSG_COMPAT */
802 len = MIN(len, fromsa->sa_len);
803 #ifdef COMPAT_OLDSOCK
804 if (mp->msg_flags & MSG_COMPAT)
805 ((struct osockaddr *)fromsa)->sa_family =
808 error = copyout(fromsa,
809 (caddr_t)mp->msg_name, (unsigned)len);
813 mp->msg_namelen = len;
815 (error = copyout((caddr_t)&len, namelenp, sizeof (int)))) {
816 #ifdef COMPAT_OLDSOCK
817 if (mp->msg_flags & MSG_COMPAT)
818 error = 0; /* old recvfrom didn't check */
824 if (mp->msg_control) {
825 #ifdef COMPAT_OLDSOCK
827 * We assume that old recvmsg calls won't receive access
828 * rights and other control info, esp. as control info
829 * is always optional and those options didn't exist in 4.3.
830 * If we receive rights, trim the cmsghdr; anything else
833 if (control && mp->msg_flags & MSG_COMPAT) {
834 if (mtod(control, struct cmsghdr *)->cmsg_level !=
836 mtod(control, struct cmsghdr *)->cmsg_type !=
838 mp->msg_controllen = 0;
841 control->m_len -= sizeof (struct cmsghdr);
842 control->m_data += sizeof (struct cmsghdr);
845 len = mp->msg_controllen;
847 mp->msg_controllen = 0;
848 ctlbuf = (caddr_t) mp->msg_control;
850 while (m && len > 0) {
856 mp->msg_flags |= MSG_CTRUNC;
860 if ((error = copyout((caddr_t)mtod(m, caddr_t),
861 ctlbuf, tocopy)) != 0)
868 mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control;
873 FREE(fromsa, M_SONAME);
880 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
881 * caddr_t from, int *fromlenaddr)
884 recvfrom(struct recvfrom_args *uap)
890 if (uap->fromlenaddr) {
891 error = copyin((caddr_t)uap->fromlenaddr,
892 (caddr_t)&msg.msg_namelen, sizeof (msg.msg_namelen));
897 msg.msg_name = uap->from;
900 aiov.iov_base = uap->buf;
901 aiov.iov_len = uap->len;
903 msg.msg_flags = uap->flags;
904 return (recvit(uap->s, &msg, (caddr_t)uap->fromlenaddr));
907 #ifdef COMPAT_OLDSOCK
909 orecvfrom(struct recvfrom_args *uap)
911 uap->flags |= MSG_COMPAT;
912 return (recvfrom(uap));
916 #ifdef COMPAT_OLDSOCK
918 * struct orecv_args(int s, caddr_t buf, int len, int flags)
921 orecv(struct orecv_args *uap)
930 aiov.iov_base = uap->buf;
931 aiov.iov_len = uap->len;
933 msg.msg_flags = uap->flags;
934 return (recvit(uap->s, &msg, (caddr_t)0));
938 * Old recvmsg. This code takes advantage of the fact that the old msghdr
939 * overlays the new one, missing only the flags, and with the (old) access
940 * rights where the control fields are now.
942 * orecvmsg_args(int s, struct omsghdr *msg, int flags)
945 orecvmsg(struct orecvmsg_args *uap)
948 struct iovec aiov[UIO_SMALLIOV], *iov;
951 error = copyin((caddr_t)uap->msg, (caddr_t)&msg,
952 sizeof (struct omsghdr));
955 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
956 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV)
958 MALLOC(iov, struct iovec *,
959 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
963 msg.msg_flags = uap->flags | MSG_COMPAT;
964 error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov,
965 (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
969 error = recvit(uap->s, &msg, (caddr_t)&uap->msg->msg_namelen);
971 if (msg.msg_controllen && error == 0)
972 error = copyout((caddr_t)&msg.msg_controllen,
973 (caddr_t)&uap->msg->msg_accrightslen, sizeof (int));
982 * recvmsg_args(int s, struct msghdr *msg, int flags)
985 recvmsg(struct recvmsg_args *uap)
988 struct iovec aiov[UIO_SMALLIOV], *uiov, *iov;
991 error = copyin((caddr_t)uap->msg, (caddr_t)&msg, sizeof (msg));
994 if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
995 if ((u_int)msg.msg_iovlen >= UIO_MAXIOV)
997 MALLOC(iov, struct iovec *,
998 sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
1002 #ifdef COMPAT_OLDSOCK
1003 msg.msg_flags = uap->flags &~ MSG_COMPAT;
1005 msg.msg_flags = uap->flags;
1009 error = copyin((caddr_t)uiov, (caddr_t)iov,
1010 (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
1013 error = recvit(uap->s, &msg, (caddr_t)0);
1016 error = copyout((caddr_t)&msg, (caddr_t)uap->msg, sizeof(msg));
1025 * shutdown_args(int s, int how)
1029 shutdown(struct shutdown_args *uap)
1031 struct thread *td = curthread;
1032 struct proc *p = td->td_proc;
1037 error = holdsock(p->p_fd, uap->s, &fp);
1040 error = soshutdown((struct socket *)fp->f_data, uap->how);
1046 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1050 setsockopt(struct setsockopt_args *uap)
1052 struct thread *td = curthread;
1053 struct proc *p = td->td_proc;
1055 struct sockopt sopt;
1058 if (uap->val == 0 && uap->valsize != 0)
1060 if (uap->valsize < 0)
1063 error = holdsock(p->p_fd, uap->s, &fp);
1067 sopt.sopt_dir = SOPT_SET;
1068 sopt.sopt_level = uap->level;
1069 sopt.sopt_name = uap->name;
1070 sopt.sopt_val = uap->val;
1071 sopt.sopt_valsize = uap->valsize;
1073 error = sosetopt((struct socket *)fp->f_data, &sopt);
1079 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1083 getsockopt(struct getsockopt_args *uap)
1085 struct thread *td = curthread;
1086 struct proc *p = td->td_proc;
1089 struct sockopt sopt;
1091 error = holdsock(p->p_fd, uap->s, &fp);
1095 error = copyin((caddr_t)uap->avalsize, (caddr_t)&valsize,
1109 sopt.sopt_dir = SOPT_GET;
1110 sopt.sopt_level = uap->level;
1111 sopt.sopt_name = uap->name;
1112 sopt.sopt_val = uap->val;
1113 sopt.sopt_valsize = (size_t)valsize; /* checked non-negative above */
1116 error = sogetopt((struct socket *)fp->f_data, &sopt);
1118 valsize = sopt.sopt_valsize;
1119 error = copyout((caddr_t)&valsize,
1120 (caddr_t)uap->avalsize, sizeof (valsize));
1127 * getsockname_args(int fdes, caddr_t asa, int *alen)
1133 getsockname1(struct getsockname_args *uap, int compat)
1135 struct thread *td = curthread;
1136 struct proc *p = td->td_proc;
1139 struct sockaddr *sa;
1142 error = holdsock(p->p_fd, uap->fdes, &fp);
1145 error = copyin((caddr_t)uap->alen, (caddr_t)&len, sizeof (len));
1154 so = (struct socket *)fp->f_data;
1156 error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, &sa);
1164 len = MIN(len, sa->sa_len);
1165 #ifdef COMPAT_OLDSOCK
1167 ((struct osockaddr *)sa)->sa_family = sa->sa_family;
1169 error = copyout(sa, (caddr_t)uap->asa, (u_int)len);
1172 error = copyout((caddr_t)&len, (caddr_t)uap->alen,
1182 getsockname(struct getsockname_args *uap)
1185 return (getsockname1(uap, 0));
1188 #ifdef COMPAT_OLDSOCK
1190 ogetsockname(struct getsockname_args *uap)
1193 return (getsockname1(uap, 1));
1195 #endif /* COMPAT_OLDSOCK */
1198 * getpeername_args(int fdes, caddr_t asa, int *alen)
1200 * Get name of peer for connected socket.
1204 getpeername1(struct getpeername_args *uap, int compat)
1206 struct thread *td = curthread;
1207 struct proc *p = td->td_proc;
1210 struct sockaddr *sa;
1213 error = holdsock(p->p_fd, uap->fdes, &fp);
1216 so = (struct socket *)fp->f_data;
1217 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1221 error = copyin((caddr_t)uap->alen, (caddr_t)&len, sizeof (len));
1231 error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, &sa);
1238 len = MIN(len, sa->sa_len);
1239 #ifdef COMPAT_OLDSOCK
1241 ((struct osockaddr *)sa)->sa_family =
1244 error = copyout(sa, (caddr_t)uap->asa, (u_int)len);
1248 error = copyout((caddr_t)&len, (caddr_t)uap->alen, sizeof (len));
1257 getpeername(struct getpeername_args *uap)
1259 return (getpeername1(uap, 0));
1262 #ifdef COMPAT_OLDSOCK
1264 ogetpeername(struct ogetpeername_args *uap)
1266 /* XXX uap should have type `getpeername_args *' to begin with. */
1267 return (getpeername1((struct getpeername_args *)uap, 1));
1269 #endif /* COMPAT_OLDSOCK */
1272 sockargs(mp, buf, buflen, type)
1277 register struct sockaddr *sa;
1278 register struct mbuf *m;
1281 if ((u_int)buflen > MLEN) {
1282 #ifdef COMPAT_OLDSOCK
1283 if (type == MT_SONAME && (u_int)buflen <= 112)
1284 buflen = MLEN; /* unix domain compat. hack */
1289 m = m_get(M_WAIT, type);
1293 error = copyin(buf, mtod(m, caddr_t), (u_int)buflen);
1298 if (type == MT_SONAME) {
1299 sa = mtod(m, struct sockaddr *);
1301 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1302 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1303 sa->sa_family = sa->sa_len;
1305 sa->sa_len = buflen;
1312 getsockaddr(namp, uaddr, len)
1313 struct sockaddr **namp;
1317 struct sockaddr *sa;
1320 if (len > SOCK_MAXADDRLEN)
1321 return ENAMETOOLONG;
1322 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1323 error = copyin(uaddr, sa, len);
1327 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1328 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1329 sa->sa_family = sa->sa_len;
1338 * holdsock() - load the struct file pointer associated
1339 * with a socket into *fpp. If an error occurs, non-zero
1340 * will be returned and *fpp will be set to NULL.
1343 holdsock(fdp, fdes, fpp)
1344 struct filedesc *fdp;
1348 register struct file *fp = NULL;
1351 if ((unsigned)fdes >= fdp->fd_nfiles ||
1352 (fp = fdp->fd_ofiles[fdes]) == NULL) {
1354 } else if (fp->f_type != DTYPE_SOCKET) {
1365 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
1368 sf_buf_init(void *arg)
1372 SLIST_INIT(&sf_freelist);
1373 sf_base = kmem_alloc_pageable(kernel_map, nsfbufs * PAGE_SIZE);
1374 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP, M_NOWAIT);
1375 bzero(sf_bufs, nsfbufs * sizeof(struct sf_buf));
1376 for (i = 0; i < nsfbufs; i++) {
1377 sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
1378 SLIST_INSERT_HEAD(&sf_freelist, &sf_bufs[i], free_list);
1383 * Get an sf_buf from the freelist. Will block if none are available.
1393 while ((sf = SLIST_FIRST(&sf_freelist)) == NULL) {
1394 sf_buf_alloc_want = 1;
1395 error = tsleep(&sf_freelist, PVM|PCATCH, "sfbufa", 0);
1400 SLIST_REMOVE_HEAD(&sf_freelist, free_list);
1407 #define dtosf(x) (&sf_bufs[((uintptr_t)(x) - (uintptr_t)sf_base) >> PAGE_SHIFT])
1409 sf_buf_ref(caddr_t addr, u_int size)
1414 if (sf->refcnt == 0)
1415 panic("sf_buf_ref: referencing a free sf_buf");
1420 * Lose a reference to an sf_buf. When none left, detach mapped page
1421 * and release resources back to the system.
1423 * Must be called at splimp.
1426 sf_buf_free(caddr_t addr, u_int size)
1433 if (sf->refcnt == 0)
1434 panic("sf_buf_free: freeing free sf_buf");
1436 if (sf->refcnt == 0) {
1437 pmap_qremove((vm_offset_t)addr, 1);
1440 vm_page_unwire(m, 0);
1442 * Check for the object going away on us. This can
1443 * happen since we don't hold a reference to it.
1444 * If so, we're responsible for freeing the page.
1446 if (m->wire_count == 0 && m->object == NULL)
1450 SLIST_INSERT_HEAD(&sf_freelist, sf, free_list);
1451 if (sf_buf_alloc_want) {
1452 sf_buf_alloc_want = 0;
1453 wakeup(&sf_freelist);
1460 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1461 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1463 * Send a file specified by 'fd' and starting at 'offset' to a socket
1464 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1465 * nbytes == 0. Optionally add a header and/or trailer to the socket
1466 * output. If specified, write the total number of bytes sent into *sbytes.
1469 sendfile(struct sendfile_args *uap)
1471 return (do_sendfile(uap, 0));
1476 osendfile(struct osendfile_args *uap)
1478 struct sendfile_args args;
1482 args.offset = uap->offset;
1483 args.nbytes = uap->nbytes;
1484 args.hdtr = uap->hdtr;
1485 args.sbytes = uap->sbytes;
1486 args.flags = uap->flags;
1488 return (do_sendfile(&args, 1));
1493 do_sendfile(struct sendfile_args *uap, int compat)
1495 struct thread *td = curthread;
1496 struct proc *p = td->td_proc;
1498 struct filedesc *fdp;
1500 struct vm_object *obj;
1505 struct writev_args nuap;
1506 struct sf_hdtr hdtr;
1507 off_t off, xfsize, hdtr_size, sbytes = 0;
1516 * Do argument checking. Must be a regular file in, stream
1517 * type and connected socket out, positive offset.
1519 fp = holdfp(fdp, uap->fd, FREAD);
1524 if (fp->f_type != DTYPE_VNODE) {
1528 vp = (struct vnode *)fp->f_data;
1530 if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) {
1535 error = holdsock(p->p_fd, uap->s, &fp);
1538 so = (struct socket *)fp->f_data;
1539 if (so->so_type != SOCK_STREAM) {
1543 if ((so->so_state & SS_ISCONNECTED) == 0) {
1547 if (uap->offset < 0) {
1553 * If specified, get the pointer to the sf_hdtr struct for
1554 * any headers/trailers.
1556 if (uap->hdtr != NULL) {
1557 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1561 * Send any headers. Wimp out and use writev(2).
1563 if (hdtr.headers != NULL) {
1565 nuap.iovp = hdtr.headers;
1566 nuap.iovcnt = hdtr.hdr_cnt;
1567 error = writev(&nuap);
1571 sbytes += p->p_retval[0];
1573 hdtr_size += p->p_retval[0];
1578 * Protect against multiple writers to the socket.
1580 (void) sblock(&so->so_snd, M_WAITOK);
1583 * Loop through the pages in the file, starting with the requested
1584 * offset. Get a file page (do I/O if necessary), map the file page
1585 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1588 for (off = uap->offset; ; off += xfsize, sbytes += xfsize) {
1592 pindex = OFF_TO_IDX(off);
1595 * Calculate the amount to transfer. Not to exceed a page,
1596 * the EOF, or the passed in nbytes.
1598 xfsize = obj->un_pager.vnp.vnp_size - off;
1599 if (xfsize > PAGE_SIZE)
1601 pgoff = (vm_offset_t)(off & PAGE_MASK);
1602 if (PAGE_SIZE - pgoff < xfsize)
1603 xfsize = PAGE_SIZE - pgoff;
1604 if (uap->nbytes && xfsize > (uap->nbytes - sbytes))
1605 xfsize = uap->nbytes - sbytes;
1609 * Optimize the non-blocking case by looking at the socket space
1610 * before going to the extra work of constituting the sf_buf.
1612 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) {
1613 if (so->so_state & SS_CANTSENDMORE)
1617 sbunlock(&so->so_snd);
1621 * Attempt to look up the page.
1623 * Allocate if not found
1625 * Wait and loop if busy.
1627 pg = vm_page_lookup(obj, pindex);
1630 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1636 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1641 * Wire the page so it does not get ripped out from under
1648 * If page is not valid for what we need, initiate I/O
1651 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1657 * Ensure that our page is still around when the I/O
1660 vm_page_io_start(pg);
1663 * Get the page from backing store.
1665 bsize = vp->v_mount->mnt_stat.f_iosize;
1666 auio.uio_iov = &aiov;
1667 auio.uio_iovcnt = 1;
1669 aiov.iov_len = MAXBSIZE;
1670 auio.uio_resid = MAXBSIZE;
1671 auio.uio_offset = trunc_page(off);
1672 auio.uio_segflg = UIO_NOCOPY;
1673 auio.uio_rw = UIO_READ;
1675 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
1676 error = VOP_READ(vp, &auio,
1677 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1679 VOP_UNLOCK(vp, 0, td);
1680 vm_page_flag_clear(pg, PG_ZERO);
1681 vm_page_io_finish(pg);
1683 vm_page_unwire(pg, 0);
1685 * See if anyone else might know about this page.
1686 * If not and it is not valid, then free it.
1688 if (pg->wire_count == 0 && pg->valid == 0 &&
1689 pg->busy == 0 && !(pg->flags & PG_BUSY) &&
1690 pg->hold_count == 0) {
1694 sbunlock(&so->so_snd);
1701 * Get a sendfile buf. We usually wait as long as necessary,
1702 * but this wait can be interrupted.
1704 if ((sf = sf_buf_alloc()) == NULL) {
1706 vm_page_unwire(pg, 0);
1707 if (pg->wire_count == 0 && pg->object == NULL)
1710 sbunlock(&so->so_snd);
1717 * Allocate a kernel virtual page and insert the physical page
1722 pmap_qenter(sf->kva, &pg, 1);
1724 * Get an mbuf header and set it up as having external storage.
1726 MGETHDR(m, M_WAIT, MT_DATA);
1729 sf_buf_free((void *)sf->kva, PAGE_SIZE);
1730 sbunlock(&so->so_snd);
1733 m->m_ext.ext_free = sf_buf_free;
1734 m->m_ext.ext_ref = sf_buf_ref;
1735 m->m_ext.ext_buf = (void *)sf->kva;
1736 m->m_ext.ext_size = PAGE_SIZE;
1737 m->m_data = (char *) sf->kva + pgoff;
1738 m->m_flags |= M_EXT;
1739 m->m_pkthdr.len = m->m_len = xfsize;
1741 * Add the buffer to the socket buffer chain.
1746 * Make sure that the socket is still able to take more data.
1747 * CANTSENDMORE being true usually means that the connection
1748 * was closed. so_error is true when an error was sensed after
1750 * The state is checked after the page mapping and buffer
1751 * allocation above since those operations may block and make
1752 * any socket checks stale. From this point forward, nothing
1753 * blocks before the pru_send (or more accurately, any blocking
1754 * results in a loop back to here to re-check).
1756 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1757 if (so->so_state & SS_CANTSENDMORE) {
1760 error = so->so_error;
1764 sbunlock(&so->so_snd);
1769 * Wait for socket space to become available. We do this just
1770 * after checking the connection state above in order to avoid
1771 * a race condition with sbwait().
1773 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1774 if (so->so_state & SS_NBIO) {
1776 sbunlock(&so->so_snd);
1781 error = sbwait(&so->so_snd);
1783 * An error from sbwait usually indicates that we've
1784 * been interrupted by a signal. If we've sent anything
1785 * then return bytes sent, otherwise return the error.
1789 sbunlock(&so->so_snd);
1796 (*so->so_proto->pr_usrreqs->pru_send)(so, 0, m, 0, 0, td);
1799 sbunlock(&so->so_snd);
1803 sbunlock(&so->so_snd);
1806 * Send trailers. Wimp out and use writev(2).
1808 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1810 nuap.iovp = hdtr.trailers;
1811 nuap.iovcnt = hdtr.trl_cnt;
1812 error = writev(&nuap);
1816 sbytes += p->p_retval[0];
1818 hdtr_size += p->p_retval[0];
1822 if (uap->sbytes != NULL) {
1824 sbytes += hdtr_size;
1825 copyout(&sbytes, uap->sbytes, sizeof(off_t));