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 1.45 2004/11/20 20:35:33 dillon Exp $
41 #include "opt_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/sysproto.h>
47 #include <sys/malloc.h>
48 #include <sys/filedesc.h>
49 #include <sys/event.h>
51 #include <sys/fcntl.h>
53 #include <sys/filio.h>
54 #include <sys/kern_syscall.h>
56 #include <sys/protosw.h>
57 #include <sys/sfbuf.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/socketops.h>
62 #include <sys/vnode.h>
64 #include <sys/mount.h>
66 #include <sys/ktrace.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_pageout.h>
72 #include <vm/vm_kern.h>
73 #include <vm/vm_extern.h>
74 #include <sys/file2.h>
75 #include <sys/signalvar.h>
77 #include <sys/thread2.h>
78 #include <sys/msgport2.h>
85 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures");
88 * System call interface to the socket abstraction.
91 extern struct fileops socketops;
94 * socket_args(int domain, int type, int protocol)
97 kern_socket(int domain, int type, int protocol, int *res)
99 struct thread *td = curthread;
100 struct proc *p = td->td_proc;
101 struct filedesc *fdp;
109 error = falloc(p, &fp, &fd);
112 error = socreate(domain, &so, type, protocol, td);
114 if (fdp->fd_ofiles[fd] == fp) {
115 fdp->fd_ofiles[fd] = NULL;
119 fp->f_data = (caddr_t)so;
120 fp->f_flag = FREAD|FWRITE;
121 fp->f_ops = &socketops;
122 fp->f_type = DTYPE_SOCKET;
130 socket(struct socket_args *uap)
134 error = kern_socket(uap->domain, uap->type, uap->protocol,
135 &uap->sysmsg_result);
141 kern_bind(int s, struct sockaddr *sa)
143 struct thread *td = curthread;
144 struct proc *p = td->td_proc;
149 error = holdsock(p->p_fd, s, &fp);
152 error = sobind((struct socket *)fp->f_data, sa, td);
158 * bind_args(int s, caddr_t name, int namelen)
161 bind(struct bind_args *uap)
166 error = getsockaddr(&sa, uap->name, uap->namelen);
169 error = kern_bind(uap->s, sa);
176 kern_listen(int s, int backlog)
178 struct thread *td = curthread;
179 struct proc *p = td->td_proc;
184 error = holdsock(p->p_fd, s, &fp);
187 error = solisten((struct socket *)fp->f_data, backlog, td);
193 * listen_args(int s, int backlog)
196 listen(struct listen_args *uap)
200 error = kern_listen(uap->s, uap->backlog);
205 * Returns the accepted socket as well.
208 soaccept_predicate(struct netmsg *msg0)
210 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
211 struct socket *head = msg->nm_so;
213 if (head->so_error != 0) {
214 msg->nm_lmsg.ms_error = head->so_error;
217 if (!TAILQ_EMPTY(&head->so_comp)) {
218 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
219 msg->nm_so = TAILQ_FIRST(&head->so_comp);
220 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
223 msg->nm_lmsg.ms_error = 0;
226 if (head->so_state & SS_CANTRCVMORE) {
227 msg->nm_lmsg.ms_error = ECONNABORTED;
230 if (head->so_state & SS_NBIO) {
231 msg->nm_lmsg.ms_error = EWOULDBLOCK;
239 * The second argument to kern_accept() is a handle to a struct sockaddr.
240 * This allows kern_accept() to return a pointer to an allocated struct
241 * sockaddr which must be freed later with FREE(). The caller must
242 * initialize *name to NULL.
245 kern_accept(int s, struct sockaddr **name, int *namelen, int *res)
247 struct thread *td = curthread;
248 struct proc *p = td->td_proc;
249 struct filedesc *fdp = p->p_fd;
250 struct file *lfp = NULL;
251 struct file *nfp = NULL;
253 struct socket *head, *so;
254 struct netmsg_so_notify msg;
257 u_int fflag; /* type must match fp->f_flag */
260 if (name && namelen && *namelen < 0)
263 error = holdsock(fdp, s, &lfp);
267 error = falloc(p, &nfp, &fd);
268 if (error) { /* Probably ran out of file descriptors. */
275 head = (struct socket *)lfp->f_data;
276 if ((head->so_options & SO_ACCEPTCONN) == 0) {
281 /* optimize for uniprocessor case later XXX JH */
282 port = head->so_proto->pr_mport(head, NULL, PRU_PRED);
283 lwkt_initmsg(&msg.nm_lmsg, &curthread->td_msgport,
284 MSGF_PCATCH | MSGF_ABORTABLE,
285 lwkt_cmd_func(netmsg_so_notify),
286 lwkt_cmd_func(netmsg_so_notify_abort));
287 msg.nm_predicate = soaccept_predicate;
289 msg.nm_etype = NM_REVENT;
290 error = lwkt_domsg(port, &msg.nm_lmsg);
295 * At this point we have the connection that's ready to be accepted.
301 /* connection has been removed from the listen queue */
302 KNOTE(&head->so_rcv.sb_sel.si_note, 0);
304 so->so_state &= ~SS_COMP;
306 if (head->so_sigio != NULL)
307 fsetown(fgetown(head->so_sigio), &so->so_sigio);
309 nfp->f_data = (caddr_t)so;
311 nfp->f_ops = &socketops;
312 nfp->f_type = DTYPE_SOCKET;
313 /* Sync socket nonblocking/async state with file flags */
314 tmp = fflag & FNONBLOCK;
315 (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td);
316 tmp = fflag & FASYNC;
317 (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td);
320 error = soaccept(so, &sa);
323 * Set the returned name and namelen as applicable. Set the returned
324 * namelen to 0 for older code which might ignore the return value
328 if (sa && name && namelen) {
329 if (*namelen > sa->sa_len)
330 *namelen = sa->sa_len;
340 * close the new descriptor, assuming someone hasn't ripped it
341 * out from under us. Note that *res is normally ignored if an
342 * error is returned but a syscall message will still have access
343 * to the result code.
347 if (fdp->fd_ofiles[fd] == nfp) {
348 fdp->fd_ofiles[fd] = NULL;
354 * Release explicitly held references before returning.
363 * accept_args(int s, caddr_t name, int *anamelen)
366 accept(struct accept_args *uap)
368 struct sockaddr *sa = NULL;
373 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
377 error = kern_accept(uap->s, &sa, &sa_len, &uap->sysmsg_result);
380 error = copyout(sa, uap->name, sa_len);
382 error = copyout(&sa_len, uap->anamelen,
383 sizeof(*uap->anamelen));
388 error = kern_accept(uap->s, NULL, 0, &uap->sysmsg_result);
394 * Returns TRUE if predicate satisfied.
397 soconnected_predicate(struct netmsg *msg0)
399 struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
400 struct socket *so = msg->nm_so;
402 /* check predicate */
403 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
404 msg->nm_lmsg.ms_error = so->so_error;
412 kern_connect(int s, struct sockaddr *sa)
414 struct thread *td = curthread;
415 struct proc *p = td->td_proc;
420 error = holdsock(p->p_fd, s, &fp);
423 so = (struct socket *)fp->f_data;
424 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
428 error = soconnect(so, sa, td);
431 if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
435 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
436 struct netmsg_so_notify msg;
439 port = so->so_proto->pr_mport(so, sa, PRU_PRED);
440 lwkt_initmsg(&msg.nm_lmsg,
441 &curthread->td_msgport,
442 MSGF_PCATCH | MSGF_ABORTABLE,
443 lwkt_cmd_func(netmsg_so_notify),
444 lwkt_cmd_func(netmsg_so_notify_abort));
445 msg.nm_predicate = soconnected_predicate;
447 msg.nm_etype = NM_REVENT;
448 error = lwkt_domsg(port, &msg.nm_lmsg);
451 error = so->so_error;
455 so->so_state &= ~SS_ISCONNECTING;
456 if (error == ERESTART)
464 * connect_args(int s, caddr_t name, int namelen)
467 connect(struct connect_args *uap)
472 error = getsockaddr(&sa, uap->name, uap->namelen);
475 error = kern_connect(uap->s, sa);
482 kern_socketpair(int domain, int type, int protocol, int *sv)
484 struct thread *td = curthread;
485 struct proc *p = td->td_proc;
486 struct filedesc *fdp;
487 struct file *fp1, *fp2;
488 struct socket *so1, *so2;
493 error = socreate(domain, &so1, type, protocol, td);
496 error = socreate(domain, &so2, type, protocol, td);
499 error = falloc(p, &fp1, &fd);
503 fp1->f_data = (caddr_t)so1;
504 error = falloc(p, &fp2, &fd);
507 fp2->f_data = (caddr_t)so2;
509 error = soconnect2(so1, so2);
512 if (type == SOCK_DGRAM) {
514 * Datagram socket connection is asymmetric.
516 error = soconnect2(so2, so1);
520 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
521 fp1->f_ops = fp2->f_ops = &socketops;
522 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
527 if (fdp->fd_ofiles[sv[1]] == fp2) {
528 fdp->fd_ofiles[sv[1]] = NULL;
533 if (fdp->fd_ofiles[sv[0]] == fp1) {
534 fdp->fd_ofiles[sv[0]] = NULL;
546 * socketpair(int domain, int type, int protocol, int *rsv)
549 socketpair(struct socketpair_args *uap)
553 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
556 error = copyout(sockv, uap->rsv, sizeof(sockv));
561 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
562 struct mbuf *control, int flags, int *res)
564 struct thread *td = curthread;
565 struct proc *p = td->td_proc;
570 struct iovec *ktriov = NULL;
574 error = holdsock(p->p_fd, s, &fp);
577 if (auio->uio_resid < 0) {
582 if (KTRPOINT(td, KTR_GENIO)) {
583 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
585 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
586 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
590 len = auio->uio_resid;
591 so = (struct socket *)fp->f_data;
592 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
594 if (auio->uio_resid != len && (error == ERESTART ||
595 error == EINTR || error == EWOULDBLOCK))
601 if (ktriov != NULL) {
603 ktruio.uio_iov = ktriov;
604 ktruio.uio_resid = len - auio->uio_resid;
605 ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error);
607 FREE(ktriov, M_TEMP);
611 *res = len - auio->uio_resid;
618 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
621 sendto(struct sendto_args *uap)
623 struct thread *td = curthread;
626 struct sockaddr *sa = NULL;
630 error = getsockaddr(&sa, uap->to, uap->tolen);
634 aiov.iov_base = uap->buf;
635 aiov.iov_len = uap->len;
636 auio.uio_iov = &aiov;
639 auio.uio_resid = uap->len;
640 auio.uio_segflg = UIO_USERSPACE;
641 auio.uio_rw = UIO_WRITE;
644 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
645 &uap->sysmsg_result);
653 * sendmsg_args(int s, caddr_t msg, int flags)
656 sendmsg(struct sendmsg_args *uap)
658 struct thread *td = curthread;
661 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
662 struct sockaddr *sa = NULL;
663 struct mbuf *control = NULL;
666 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
671 * Conditionally copyin msg.msg_name.
674 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
682 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
687 auio.uio_iovcnt = msg.msg_iovlen;
689 auio.uio_segflg = UIO_USERSPACE;
690 auio.uio_rw = UIO_WRITE;
694 * Conditionally copyin msg.msg_control.
696 if (msg.msg_control) {
697 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
698 msg.msg_controllen > MLEN) {
702 control = m_get(MB_WAIT, MT_CONTROL);
703 if (control == NULL) {
707 control->m_len = msg.msg_controllen;
708 error = copyin(msg.msg_control, mtod(control, caddr_t),
716 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
717 &uap->sysmsg_result);
722 iovec_free(&iov, aiov);
727 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
728 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
729 * Don't forget to FREE() and m_free() these if they are returned.
732 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
733 struct mbuf **control, int *flags, int *res)
735 struct thread *td = curthread;
736 struct proc *p = td->td_proc;
741 struct iovec *ktriov = NULL;
745 error = holdsock(p->p_fd, s, &fp);
748 if (auio->uio_resid < 0) {
753 if (KTRPOINT(td, KTR_GENIO)) {
754 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
756 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
757 bcopy(auio->uio_iov, ktriov, iovlen);
761 len = auio->uio_resid;
762 so = (struct socket *)fp->f_data;
763 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
765 if (auio->uio_resid != len && (error == ERESTART ||
766 error == EINTR || error == EWOULDBLOCK))
770 if (ktriov != NULL) {
772 ktruio.uio_iov = ktriov;
773 ktruio.uio_resid = len - auio->uio_resid;
774 ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error);
776 FREE(ktriov, M_TEMP);
780 *res = len - auio->uio_resid;
787 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
788 * caddr_t from, int *fromlenaddr)
791 recvfrom(struct recvfrom_args *uap)
793 struct thread *td = curthread;
796 struct sockaddr *sa = NULL;
799 if (uap->from && uap->fromlenaddr) {
800 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
808 aiov.iov_base = uap->buf;
809 aiov.iov_len = uap->len;
810 auio.uio_iov = &aiov;
813 auio.uio_resid = uap->len;
814 auio.uio_segflg = UIO_USERSPACE;
815 auio.uio_rw = UIO_READ;
818 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
819 &uap->flags, &uap->sysmsg_result);
821 if (error == 0 && uap->from) {
822 /* note: sa may still be NULL */
824 fromlen = MIN(fromlen, sa->sa_len);
825 error = copyout(sa, uap->from, fromlen);
830 error = copyout(&fromlen, uap->fromlenaddr,
841 * recvmsg_args(int s, struct msghdr *msg, int flags)
844 recvmsg(struct recvmsg_args *uap)
846 struct thread *td = curthread;
849 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
850 struct mbuf *m, *control = NULL;
851 struct sockaddr *sa = NULL;
853 socklen_t *ufromlenp, *ucontrollenp;
854 int error, fromlen, controllen, len, flags, *uflagsp;
857 * This copyin handles everything except the iovec.
859 error = copyin(uap->msg, &msg, sizeof(msg));
863 if (msg.msg_name && msg.msg_namelen < 0)
865 if (msg.msg_control && msg.msg_controllen < 0)
868 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
870 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
872 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
878 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
883 auio.uio_iovcnt = msg.msg_iovlen;
885 auio.uio_segflg = UIO_USERSPACE;
886 auio.uio_rw = UIO_READ;
891 error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
892 msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
895 * Conditionally copyout the name and populate the namelen field.
897 if (error == 0 && msg.msg_name) {
898 fromlen = MIN(msg.msg_namelen, sa->sa_len);
899 error = copyout(sa, msg.msg_name, fromlen);
901 error = copyout(&fromlen, ufromlenp,
906 * Copyout msg.msg_control and msg.msg_controllen.
908 if (error == 0 && msg.msg_control) {
909 len = msg.msg_controllen;
911 ctlbuf = (caddr_t)msg.msg_control;
913 while(m && len > 0) {
916 if (len >= m->m_len) {
919 msg.msg_flags |= MSG_CTRUNC;
923 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
931 controllen = ctlbuf - (caddr_t)msg.msg_control;
932 error = copyout(&controllen, ucontrollenp,
933 sizeof(*ucontrollenp));
937 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
942 iovec_free(&iov, aiov);
949 * shutdown_args(int s, int how)
952 kern_shutdown(int s, int how)
954 struct thread *td = curthread;
955 struct proc *p = td->td_proc;
960 error = holdsock(p->p_fd, s, &fp);
963 error = soshutdown((struct socket *)fp->f_data, how);
969 shutdown(struct shutdown_args *uap)
973 error = kern_shutdown(uap->s, uap->how);
979 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
980 * in kernel pointer instead of a userland pointer. This allows us
981 * to manipulate socket options in the emulation code.
984 kern_setsockopt(int s, struct sockopt *sopt)
986 struct thread *td = curthread;
987 struct proc *p = td->td_proc;
991 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
993 if (sopt->sopt_valsize < 0)
996 error = holdsock(p->p_fd, s, &fp);
1000 error = sosetopt((struct socket *)fp->f_data, sopt);
1006 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1009 setsockopt(struct setsockopt_args *uap)
1011 struct thread *td = curthread;
1012 struct sockopt sopt;
1015 sopt.sopt_dir = SOPT_SET;
1016 sopt.sopt_level = uap->level;
1017 sopt.sopt_name = uap->name;
1018 sopt.sopt_val = uap->val;
1019 sopt.sopt_valsize = uap->valsize;
1022 error = kern_setsockopt(uap->s, &sopt);
1027 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1028 * in kernel pointer instead of a userland pointer. This allows us
1029 * to manipulate socket options in the emulation code.
1032 kern_getsockopt(int s, struct sockopt *sopt)
1034 struct thread *td = curthread;
1035 struct proc *p = td->td_proc;
1039 if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1041 if (sopt->sopt_valsize < 0)
1044 error = holdsock(p->p_fd, s, &fp);
1048 error = sogetopt((struct socket *)fp->f_data, sopt);
1054 * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1057 getsockopt(struct getsockopt_args *uap)
1059 struct thread *td = curthread;
1060 struct sockopt sopt;
1064 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1073 sopt.sopt_dir = SOPT_GET;
1074 sopt.sopt_level = uap->level;
1075 sopt.sopt_name = uap->name;
1076 sopt.sopt_val = uap->val;
1077 sopt.sopt_valsize = valsize;
1080 error = kern_getsockopt(uap->s, &sopt);
1082 valsize = sopt.sopt_valsize;
1083 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1089 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1090 * This allows kern_getsockname() to return a pointer to an allocated struct
1091 * sockaddr which must be freed later with FREE(). The caller must
1092 * initialize *name to NULL.
1095 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1097 struct thread *td = curthread;
1098 struct proc *p = td->td_proc;
1101 struct sockaddr *sa = NULL;
1104 error = holdsock(p->p_fd, s, &fp);
1111 so = (struct socket *)fp->f_data;
1112 error = so_pru_sockaddr(so, &sa);
1117 *namelen = MIN(*namelen, sa->sa_len);
1127 * getsockname_args(int fdes, caddr_t asa, int *alen)
1132 getsockname(struct getsockname_args *uap)
1134 struct sockaddr *sa = NULL;
1137 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1141 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1144 error = copyout(sa, uap->asa, sa_len);
1146 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1153 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1154 * This allows kern_getpeername() to return a pointer to an allocated struct
1155 * sockaddr which must be freed later with FREE(). The caller must
1156 * initialize *name to NULL.
1159 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1161 struct thread *td = curthread;
1162 struct proc *p = td->td_proc;
1165 struct sockaddr *sa = NULL;
1168 error = holdsock(p->p_fd, s, &fp);
1175 so = (struct socket *)fp->f_data;
1176 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1180 error = so_pru_peeraddr(so, &sa);
1185 *namelen = MIN(*namelen, sa->sa_len);
1195 * getpeername_args(int fdes, caddr_t asa, int *alen)
1197 * Get name of peer for connected socket.
1200 getpeername(struct getpeername_args *uap)
1202 struct sockaddr *sa = NULL;
1205 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1209 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1212 error = copyout(sa, uap->asa, sa_len);
1214 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1221 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1223 struct sockaddr *sa;
1227 if (len > SOCK_MAXADDRLEN)
1228 return ENAMETOOLONG;
1229 if (len < offsetof(struct sockaddr, sa_data[0]))
1231 MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1232 error = copyin(uaddr, sa, len);
1236 #if BYTE_ORDER != BIG_ENDIAN
1238 * The bind(), connect(), and sendto() syscalls were not
1239 * versioned for COMPAT_43. Thus, this check must stay.
1241 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1242 sa->sa_family = sa->sa_len;
1251 * holdsock() - load the struct file pointer associated
1252 * with a socket into *fpp. If an error occurs, non-zero
1253 * will be returned and *fpp will be set to NULL.
1256 holdsock(fdp, fdes, fpp)
1257 struct filedesc *fdp;
1261 struct file *fp = NULL;
1264 if ((unsigned)fdes >= fdp->fd_nfiles ||
1265 (fp = fdp->fd_ofiles[fdes]) == NULL) {
1267 } else if (fp->f_type != DTYPE_SOCKET) {
1278 * Detach a mapped page and release resources back to the system.
1279 * We must release our wiring and if the object is ripped out
1280 * from under the vm_page we become responsible for freeing the
1283 * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1286 sf_buf_mref(void *arg)
1288 struct sfbuf_mref *sfm = arg;
1294 sf_buf_mfree(void *arg)
1296 struct sfbuf_mref *sfm = arg;
1299 KKASSERT(sfm->mref_count > 0);
1300 if (--sfm->mref_count == 0) {
1301 m = sf_buf_page(sfm->sf);
1302 sf_buf_free(sfm->sf);
1304 vm_page_unwire(m, 0);
1305 if (m->wire_count == 0 && m->object == NULL)
1306 vm_page_try_to_free(m);
1308 free(sfm, M_SENDFILE);
1314 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1315 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1317 * Send a file specified by 'fd' and starting at 'offset' to a socket
1318 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1319 * nbytes == 0. Optionally add a header and/or trailer to the socket
1320 * output. If specified, write the total number of bytes sent into *sbytes.
1322 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1323 * the headers to count against the remaining bytes to be sent from
1324 * the file descriptor. We may wish to implement a compatibility syscall
1328 sendfile(struct sendfile_args *uap)
1330 struct thread *td = curthread;
1331 struct proc *p = td->td_proc;
1333 struct filedesc *fdp;
1334 struct vnode *vp = NULL;
1335 struct sf_hdtr hdtr;
1336 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1338 struct mbuf *mheader = NULL;
1339 off_t hdtr_size = 0, sbytes;
1340 int error, hbytes = 0, tbytes;
1346 * Do argument checking. Must be a regular file in, stream
1347 * type and connected socket out, positive offset.
1349 fp = holdfp(fdp, uap->fd, FREAD);
1353 if (fp->f_type != DTYPE_VNODE) {
1357 vp = (struct vnode *)fp->f_data;
1362 * If specified, get the pointer to the sf_hdtr struct for
1363 * any headers/trailers.
1366 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1373 error = iovec_copyin(hdtr.headers, &iov, aiov,
1374 hdtr.hdr_cnt, &hbytes);
1378 auio.uio_iovcnt = hdtr.hdr_cnt;
1379 auio.uio_offset = 0;
1380 auio.uio_segflg = UIO_USERSPACE;
1381 auio.uio_rw = UIO_WRITE;
1383 auio.uio_resid = hbytes;
1385 mheader = m_uiomove(&auio, MB_WAIT, 0);
1387 iovec_free(&iov, aiov);
1388 if (mheader == NULL)
1393 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1394 &sbytes, uap->flags);
1399 * Send trailers. Wimp out and use writev(2).
1401 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1402 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1403 hdtr.trl_cnt, &auio.uio_resid);
1407 auio.uio_iovcnt = hdtr.trl_cnt;
1408 auio.uio_offset = 0;
1409 auio.uio_segflg = UIO_USERSPACE;
1410 auio.uio_rw = UIO_WRITE;
1413 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1415 iovec_free(&iov, aiov);
1418 hdtr_size += tbytes; /* trailer bytes successfully sent */
1422 if (uap->sbytes != NULL) {
1423 sbytes += hdtr_size;
1424 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1432 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1433 struct mbuf *mheader, off_t *sbytes, int flags)
1435 struct thread *td = curthread;
1436 struct proc *p = td->td_proc;
1437 struct vm_object *obj;
1442 struct sfbuf_mref *sfm;
1448 if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) {
1452 error = holdsock(p->p_fd, sfd, &fp);
1455 so = (struct socket *)fp->f_data;
1456 if (so->so_type != SOCK_STREAM) {
1460 if ((so->so_state & SS_ISCONNECTED) == 0) {
1471 * Protect against multiple writers to the socket.
1473 (void) sblock(&so->so_snd, M_WAITOK);
1476 * Loop through the pages in the file, starting with the requested
1477 * offset. Get a file page (do I/O if necessary), map the file page
1478 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1481 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1485 pindex = OFF_TO_IDX(off);
1488 * Calculate the amount to transfer. Not to exceed a page,
1489 * the EOF, or the passed in nbytes.
1491 xfsize = obj->un_pager.vnp.vnp_size - off;
1492 if (xfsize > PAGE_SIZE)
1494 pgoff = (vm_offset_t)(off & PAGE_MASK);
1495 if (PAGE_SIZE - pgoff < xfsize)
1496 xfsize = PAGE_SIZE - pgoff;
1497 if (nbytes && xfsize > (nbytes - *sbytes))
1498 xfsize = nbytes - *sbytes;
1502 * Optimize the non-blocking case by looking at the socket space
1503 * before going to the extra work of constituting the sf_buf.
1505 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) {
1506 if (so->so_state & SS_CANTSENDMORE)
1510 sbunlock(&so->so_snd);
1514 * Attempt to look up the page.
1516 * Allocate if not found, wait and loop if busy, then
1517 * wire the page. critical section protection is
1518 * required to maintain the object association (an
1519 * interrupt can free the page) through to the
1520 * vm_page_wire() call.
1523 pg = vm_page_lookup(obj, pindex);
1525 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1532 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1540 * If page is not valid for what we need, initiate I/O
1543 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1549 * Ensure that our page is still around when the I/O
1552 vm_page_io_start(pg);
1555 * Get the page from backing store.
1557 bsize = vp->v_mount->mnt_stat.f_iosize;
1558 auio.uio_iov = &aiov;
1559 auio.uio_iovcnt = 1;
1561 aiov.iov_len = MAXBSIZE;
1562 auio.uio_resid = MAXBSIZE;
1563 auio.uio_offset = trunc_page(off);
1564 auio.uio_segflg = UIO_NOCOPY;
1565 auio.uio_rw = UIO_READ;
1567 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
1568 error = VOP_READ(vp, &auio,
1569 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1571 VOP_UNLOCK(vp, 0, td);
1572 vm_page_flag_clear(pg, PG_ZERO);
1573 vm_page_io_finish(pg);
1576 vm_page_unwire(pg, 0);
1577 vm_page_try_to_free(pg);
1579 sbunlock(&so->so_snd);
1586 * Get a sendfile buf. We usually wait as long as necessary,
1587 * but this wait can be interrupted.
1589 if ((sf = sf_buf_alloc(pg, SFBA_PCATCH)) == NULL) {
1591 vm_page_unwire(pg, 0);
1592 vm_page_try_to_free(pg);
1594 sbunlock(&so->so_snd);
1600 * Get an mbuf header and set it up as having external storage.
1602 MGETHDR(m, MB_WAIT, MT_DATA);
1606 sbunlock(&so->so_snd);
1611 * sfm is a temporary hack, use a per-cpu cache for this.
1613 sfm = malloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1615 sfm->mref_count = 1;
1617 m->m_ext.ext_nfree.new = sf_buf_mfree;
1618 m->m_ext.ext_nref.new = sf_buf_mref;
1619 m->m_ext.ext_arg = sfm;
1620 m->m_ext.ext_buf = (void *)sf->kva;
1621 m->m_ext.ext_size = PAGE_SIZE;
1622 m->m_data = (char *) sf->kva + pgoff;
1623 m->m_flags |= M_EXT;
1624 m->m_pkthdr.len = m->m_len = xfsize;
1625 KKASSERT((m->m_flags & (M_EXT_OLD|M_EXT_CLUSTER)) == 0);
1627 if (mheader != NULL) {
1628 hbytes = mheader->m_pkthdr.len;
1629 mheader->m_pkthdr.len += m->m_pkthdr.len;
1637 * Add the buffer to the socket buffer chain.
1642 * Make sure that the socket is still able to take more data.
1643 * CANTSENDMORE being true usually means that the connection
1644 * was closed. so_error is true when an error was sensed after
1646 * The state is checked after the page mapping and buffer
1647 * allocation above since those operations may block and make
1648 * any socket checks stale. From this point forward, nothing
1649 * blocks before the pru_send (or more accurately, any blocking
1650 * results in a loop back to here to re-check).
1652 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1653 if (so->so_state & SS_CANTSENDMORE) {
1656 error = so->so_error;
1660 sbunlock(&so->so_snd);
1665 * Wait for socket space to become available. We do this just
1666 * after checking the connection state above in order to avoid
1667 * a race condition with sbwait().
1669 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1670 if (so->so_state & SS_NBIO) {
1672 sbunlock(&so->so_snd);
1677 error = sbwait(&so->so_snd);
1679 * An error from sbwait usually indicates that we've
1680 * been interrupted by a signal. If we've sent anything
1681 * then return bytes sent, otherwise return the error.
1685 sbunlock(&so->so_snd);
1691 error = so_pru_send(so, 0, m, NULL, NULL, td);
1694 sbunlock(&so->so_snd);
1698 if (mheader != NULL) {
1699 *sbytes += mheader->m_pkthdr.len;
1700 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1703 sbunlock(&so->so_snd);
1708 if (mheader != NULL)