2 * Copyright (c) 1982, 1986, 1989, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
5 * sendfile(2) and related extensions:
6 * Copyright (c) 1998, David Greenman. All rights reserved.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94
33 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $
36 #include "opt_ktrace.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/sysproto.h>
42 #include <sys/malloc.h>
43 #include <sys/filedesc.h>
44 #include <sys/event.h>
46 #include <sys/fcntl.h>
48 #include <sys/filio.h>
49 #include <sys/kern_syscall.h>
51 #include <sys/protosw.h>
52 #include <sys/sfbuf.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/socketops.h>
57 #include <sys/vnode.h>
59 #include <sys/mount.h>
61 #include <sys/ktrace.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_page.h>
66 #include <vm/vm_pageout.h>
67 #include <vm/vm_kern.h>
68 #include <vm/vm_extern.h>
69 #include <sys/file2.h>
70 #include <sys/signalvar.h>
71 #include <sys/serialize.h>
73 #include <sys/thread2.h>
74 #include <sys/msgport2.h>
75 #include <sys/socketvar2.h>
76 #include <net/netmsg2.h>
77 #include <vm/vm_page2.h>
79 extern int use_soaccept_pred_fast;
80 extern int use_sendfile_async;
81 extern int use_soconnect_async;
84 * System call interface to the socket abstraction.
87 extern struct fileops socketops;
90 * socket_args(int domain, int type, int protocol)
93 kern_socket(int domain, int type, int protocol, int *res)
95 struct thread *td = curthread;
96 struct filedesc *fdp = td->td_proc->p_fd;
101 KKASSERT(td->td_lwp);
103 error = falloc(td->td_lwp, &fp, &fd);
106 error = socreate(domain, &so, type, protocol, td);
108 fsetfd(fdp, NULL, fd);
110 fp->f_type = DTYPE_SOCKET;
111 fp->f_flag = FREAD | FWRITE;
112 fp->f_ops = &socketops;
125 sys_socket(struct socket_args *uap)
129 error = kern_socket(uap->domain, uap->type, uap->protocol,
130 &uap->sysmsg_iresult);
136 kern_bind(int s, struct sockaddr *sa)
138 struct thread *td = curthread;
139 struct proc *p = td->td_proc;
144 error = holdsock(p->p_fd, s, &fp);
147 error = sobind((struct socket *)fp->f_data, sa, td);
153 * bind_args(int s, caddr_t name, int namelen)
158 sys_bind(struct bind_args *uap)
163 error = getsockaddr(&sa, uap->name, uap->namelen);
166 error = kern_bind(uap->s, sa);
173 kern_listen(int s, int backlog)
175 struct thread *td = curthread;
176 struct proc *p = td->td_proc;
181 error = holdsock(p->p_fd, s, &fp);
184 error = solisten((struct socket *)fp->f_data, backlog, td);
190 * listen_args(int s, int backlog)
195 sys_listen(struct listen_args *uap)
199 error = kern_listen(uap->s, uap->backlog);
204 * Returns the accepted socket as well.
206 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the
207 * pool token is absolutely required to avoid a sofree() race,
208 * as well as to avoid tailq handling races.
211 soaccept_predicate(struct netmsg_so_notify *msg)
213 struct socket *head = msg->base.nm_so;
216 if (head->so_error != 0) {
217 msg->base.lmsg.ms_error = head->so_error;
220 lwkt_getpooltoken(head);
221 if (!TAILQ_EMPTY(&head->so_comp)) {
222 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
223 so = TAILQ_FIRST(&head->so_comp);
224 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
225 TAILQ_REMOVE(&head->so_comp, so, so_list);
227 soclrstate(so, SS_COMP);
230 * Keep a reference before clearing the so_head
231 * to avoid racing socket close in netisr.
236 lwkt_relpooltoken(head);
238 msg->base.lmsg.ms_error = 0;
239 msg->base.nm_so = so;
242 lwkt_relpooltoken(head);
243 if (head->so_state & SS_CANTRCVMORE) {
244 msg->base.lmsg.ms_error = ECONNABORTED;
247 if (msg->nm_fflags & FNONBLOCK) {
248 msg->base.lmsg.ms_error = EWOULDBLOCK;
256 * The second argument to kern_accept() is a handle to a struct sockaddr.
257 * This allows kern_accept() to return a pointer to an allocated struct
258 * sockaddr which must be freed later with FREE(). The caller must
259 * initialize *name to NULL.
262 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res)
264 struct thread *td = curthread;
265 struct filedesc *fdp = td->td_proc->p_fd;
266 struct file *lfp = NULL;
267 struct file *nfp = NULL;
269 struct socket *head, *so;
270 struct netmsg_so_notify msg;
272 u_int fflag; /* type must match fp->f_flag */
276 if (name && namelen && *namelen < 0)
279 error = holdsock(td->td_proc->p_fd, s, &lfp);
283 error = falloc(td->td_lwp, &nfp, &fd);
284 if (error) { /* Probably ran out of file descriptors. */
288 head = (struct socket *)lfp->f_data;
289 if ((head->so_options & SO_ACCEPTCONN) == 0) {
294 if (fflags & O_FBLOCKING)
295 fflags |= lfp->f_flag & ~FNONBLOCK;
296 else if (fflags & O_FNONBLOCKING)
297 fflags |= lfp->f_flag | FNONBLOCK;
299 fflags = lfp->f_flag;
301 if (use_soaccept_pred_fast) {
304 /* Initialize necessary parts for soaccept_predicate() */
305 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL);
306 msg.nm_fflags = fflags;
308 lwkt_getpooltoken(head);
309 pred = soaccept_predicate(&msg);
310 lwkt_relpooltoken(head);
313 error = msg.base.lmsg.ms_error;
321 /* optimize for uniprocessor case later XXX JH */
322 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport,
323 0, netmsg_so_notify, netmsg_so_notify_doabort);
324 msg.nm_predicate = soaccept_predicate;
325 msg.nm_fflags = fflags;
326 msg.nm_etype = NM_REVENT;
327 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH);
333 * At this point we have the connection that's ready to be accepted.
335 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects
336 * to eat the ref and turn it into a descriptor.
342 /* connection has been removed from the listen queue */
343 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0);
345 if (head->so_sigio != NULL)
346 fsetown(fgetown(&head->so_sigio), &so->so_sigio);
348 nfp->f_type = DTYPE_SOCKET;
350 nfp->f_ops = &socketops;
352 /* Sync socket nonblocking/async state with file flags */
353 tmp = fflag & FNONBLOCK;
354 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL);
355 tmp = fflag & FASYNC;
356 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
359 if (so->so_faddr != NULL) {
363 soaccept_generic(so);
366 error = soaccept(so, &sa);
370 * Set the returned name and namelen as applicable. Set the returned
371 * namelen to 0 for older code which might ignore the return value
375 if (sa && name && namelen) {
376 if (*namelen > sa->sa_len)
377 *namelen = sa->sa_len;
387 * If an error occured clear the reserved descriptor, else associate
390 * Note that *res is normally ignored if an error is returned but
391 * a syscall message will still have access to the result code.
394 fsetfd(fdp, NULL, fd);
397 fsetfd(fdp, nfp, fd);
405 * accept(int s, caddr_t name, int *anamelen)
410 sys_accept(struct accept_args *uap)
412 struct sockaddr *sa = NULL;
417 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
421 error = kern_accept(uap->s, 0, &sa, &sa_len,
422 &uap->sysmsg_iresult);
425 error = copyout(sa, uap->name, sa_len);
427 error = copyout(&sa_len, uap->anamelen,
428 sizeof(*uap->anamelen));
433 error = kern_accept(uap->s, 0, NULL, 0,
434 &uap->sysmsg_iresult);
440 * extaccept(int s, int fflags, caddr_t name, int *anamelen)
445 sys_extaccept(struct extaccept_args *uap)
447 struct sockaddr *sa = NULL;
450 int fflags = uap->flags & O_FMASK;
453 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
457 error = kern_accept(uap->s, fflags, &sa, &sa_len,
458 &uap->sysmsg_iresult);
461 error = copyout(sa, uap->name, sa_len);
463 error = copyout(&sa_len, uap->anamelen,
464 sizeof(*uap->anamelen));
469 error = kern_accept(uap->s, fflags, NULL, 0,
470 &uap->sysmsg_iresult);
477 * Returns TRUE if predicate satisfied.
480 soconnected_predicate(struct netmsg_so_notify *msg)
482 struct socket *so = msg->base.nm_so;
484 /* check predicate */
485 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
486 msg->base.lmsg.ms_error = so->so_error;
494 kern_connect(int s, int fflags, struct sockaddr *sa)
496 struct thread *td = curthread;
497 struct proc *p = td->td_proc;
500 int error, interrupted = 0;
502 error = holdsock(p->p_fd, s, &fp);
505 so = (struct socket *)fp->f_data;
507 if (fflags & O_FBLOCKING)
508 /* fflags &= ~FNONBLOCK; */;
509 else if (fflags & O_FNONBLOCKING)
514 if (so->so_state & SS_ISCONNECTING) {
518 error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE);
521 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
525 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
526 struct netmsg_so_notify msg;
528 netmsg_init_abortable(&msg.base, so,
529 &curthread->td_msgport,
532 netmsg_so_notify_doabort);
533 msg.nm_predicate = soconnected_predicate;
534 msg.nm_etype = NM_REVENT;
535 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH);
536 if (error == EINTR || error == ERESTART)
540 error = so->so_error;
545 soclrstate(so, SS_ISCONNECTING);
546 if (error == ERESTART)
554 * connect_args(int s, caddr_t name, int namelen)
559 sys_connect(struct connect_args *uap)
564 error = getsockaddr(&sa, uap->name, uap->namelen);
567 error = kern_connect(uap->s, 0, sa);
574 * connect_args(int s, int fflags, caddr_t name, int namelen)
579 sys_extconnect(struct extconnect_args *uap)
583 int fflags = uap->flags & O_FMASK;
585 error = getsockaddr(&sa, uap->name, uap->namelen);
588 error = kern_connect(uap->s, fflags, sa);
595 kern_socketpair(int domain, int type, int protocol, int *sv)
597 struct thread *td = curthread;
598 struct filedesc *fdp;
599 struct file *fp1, *fp2;
600 struct socket *so1, *so2;
603 fdp = td->td_proc->p_fd;
604 error = socreate(domain, &so1, type, protocol, td);
607 error = socreate(domain, &so2, type, protocol, td);
610 error = falloc(td->td_lwp, &fp1, &fd1);
615 error = falloc(td->td_lwp, &fp2, &fd2);
620 error = soconnect2(so1, so2);
623 if (type == SOCK_DGRAM) {
625 * Datagram socket connection is asymmetric.
627 error = soconnect2(so2, so1);
631 fp1->f_type = fp2->f_type = DTYPE_SOCKET;
632 fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
633 fp1->f_ops = fp2->f_ops = &socketops;
634 fsetfd(fdp, fp1, fd1);
635 fsetfd(fdp, fp2, fd2);
640 fsetfd(fdp, NULL, fd2);
643 fsetfd(fdp, NULL, fd1);
646 (void)soclose(so2, 0);
648 (void)soclose(so1, 0);
653 * socketpair(int domain, int type, int protocol, int *rsv)
656 sys_socketpair(struct socketpair_args *uap)
660 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
663 error = copyout(sockv, uap->rsv, sizeof(sockv));
666 kern_close(sockv[0]);
667 kern_close(sockv[1]);
675 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
676 struct mbuf *control, int flags, size_t *res)
678 struct thread *td = curthread;
679 struct lwp *lp = td->td_lwp;
680 struct proc *p = td->td_proc;
686 struct iovec *ktriov = NULL;
690 error = holdsock(p->p_fd, s, &fp);
694 if (KTRPOINT(td, KTR_GENIO)) {
695 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
697 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
698 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
702 len = auio->uio_resid;
703 so = (struct socket *)fp->f_data;
704 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
705 if (fp->f_flag & FNONBLOCK)
706 flags |= MSG_FNONBLOCKING;
708 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
710 if (auio->uio_resid != len && (error == ERESTART ||
711 error == EINTR || error == EWOULDBLOCK))
713 if (error == EPIPE && !(flags & MSG_NOSIGNAL) &&
714 !(so->so_options & SO_NOSIGPIPE))
715 lwpsignal(p, lp, SIGPIPE);
718 if (ktriov != NULL) {
720 ktruio.uio_iov = ktriov;
721 ktruio.uio_resid = len - auio->uio_resid;
722 ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
724 kfree(ktriov, M_TEMP);
728 *res = len - auio->uio_resid;
734 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
739 sys_sendto(struct sendto_args *uap)
741 struct thread *td = curthread;
744 struct sockaddr *sa = NULL;
748 error = getsockaddr(&sa, uap->to, uap->tolen);
752 aiov.iov_base = uap->buf;
753 aiov.iov_len = uap->len;
754 auio.uio_iov = &aiov;
757 auio.uio_resid = uap->len;
758 auio.uio_segflg = UIO_USERSPACE;
759 auio.uio_rw = UIO_WRITE;
762 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
763 &uap->sysmsg_szresult);
771 * sendmsg_args(int s, caddr_t msg, int flags)
776 sys_sendmsg(struct sendmsg_args *uap)
778 struct thread *td = curthread;
781 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
782 struct sockaddr *sa = NULL;
783 struct mbuf *control = NULL;
786 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
791 * Conditionally copyin msg.msg_name.
794 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
802 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
807 auio.uio_iovcnt = msg.msg_iovlen;
809 auio.uio_segflg = UIO_USERSPACE;
810 auio.uio_rw = UIO_WRITE;
814 * Conditionally copyin msg.msg_control.
816 if (msg.msg_control) {
817 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
818 msg.msg_controllen > MLEN) {
822 control = m_get(M_WAITOK, MT_CONTROL);
823 if (control == NULL) {
827 control->m_len = msg.msg_controllen;
828 error = copyin(msg.msg_control, mtod(control, caddr_t),
836 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
837 &uap->sysmsg_szresult);
840 iovec_free(&iov, aiov);
848 * kern_recvmsg() takes a handle to sa and control. If the handle is non-
849 * null, it returns a dynamically allocated struct sockaddr and an mbuf.
850 * Don't forget to FREE() and m_free() these if they are returned.
853 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
854 struct mbuf **control, int *flags, size_t *res)
856 struct thread *td = curthread;
857 struct proc *p = td->td_proc;
864 struct iovec *ktriov = NULL;
868 error = holdsock(p->p_fd, s, &fp);
872 if (KTRPOINT(td, KTR_GENIO)) {
873 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
875 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
876 bcopy(auio->uio_iov, ktriov, iovlen);
880 len = auio->uio_resid;
881 so = (struct socket *)fp->f_data;
883 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
884 if (fp->f_flag & FNONBLOCK) {
886 *flags |= MSG_FNONBLOCKING;
888 lflags = MSG_FNONBLOCKING;
894 error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
896 if (auio->uio_resid != len && (error == ERESTART ||
897 error == EINTR || error == EWOULDBLOCK))
901 if (ktriov != NULL) {
903 ktruio.uio_iov = ktriov;
904 ktruio.uio_resid = len - auio->uio_resid;
905 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
907 kfree(ktriov, M_TEMP);
911 *res = len - auio->uio_resid;
917 * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
918 * caddr_t from, int *fromlenaddr)
923 sys_recvfrom(struct recvfrom_args *uap)
925 struct thread *td = curthread;
928 struct sockaddr *sa = NULL;
931 if (uap->from && uap->fromlenaddr) {
932 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
940 aiov.iov_base = uap->buf;
941 aiov.iov_len = uap->len;
942 auio.uio_iov = &aiov;
945 auio.uio_resid = uap->len;
946 auio.uio_segflg = UIO_USERSPACE;
947 auio.uio_rw = UIO_READ;
950 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
951 &uap->flags, &uap->sysmsg_szresult);
953 if (error == 0 && uap->from) {
954 /* note: sa may still be NULL */
956 fromlen = MIN(fromlen, sa->sa_len);
957 error = copyout(sa, uap->from, fromlen);
962 error = copyout(&fromlen, uap->fromlenaddr,
973 * recvmsg_args(int s, struct msghdr *msg, int flags)
978 sys_recvmsg(struct recvmsg_args *uap)
980 struct thread *td = curthread;
983 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
984 struct mbuf *m, *control = NULL;
985 struct sockaddr *sa = NULL;
987 socklen_t *ufromlenp, *ucontrollenp;
988 int error, fromlen, controllen, len, flags, *uflagsp;
991 * This copyin handles everything except the iovec.
993 error = copyin(uap->msg, &msg, sizeof(msg));
997 if (msg.msg_name && msg.msg_namelen < 0)
999 if (msg.msg_control && msg.msg_controllen < 0)
1002 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1004 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1006 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
1012 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1017 auio.uio_iovcnt = msg.msg_iovlen;
1018 auio.uio_offset = 0;
1019 auio.uio_segflg = UIO_USERSPACE;
1020 auio.uio_rw = UIO_READ;
1025 error = kern_recvmsg(uap->s,
1026 (msg.msg_name ? &sa : NULL), &auio,
1027 (msg.msg_control ? &control : NULL), &flags,
1028 &uap->sysmsg_szresult);
1031 * Conditionally copyout the name and populate the namelen field.
1033 if (error == 0 && msg.msg_name) {
1034 /* note: sa may still be NULL */
1036 fromlen = MIN(msg.msg_namelen, sa->sa_len);
1037 error = copyout(sa, msg.msg_name, fromlen);
1042 error = copyout(&fromlen, ufromlenp,
1043 sizeof(*ufromlenp));
1047 * Copyout msg.msg_control and msg.msg_controllen.
1049 if (error == 0 && msg.msg_control) {
1050 len = msg.msg_controllen;
1052 ctlbuf = (caddr_t)msg.msg_control;
1054 while(m && len > 0) {
1055 unsigned int tocopy;
1057 if (len >= m->m_len) {
1060 msg.msg_flags |= MSG_CTRUNC;
1064 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1072 controllen = ctlbuf - (caddr_t)msg.msg_control;
1073 error = copyout(&controllen, ucontrollenp,
1074 sizeof(*ucontrollenp));
1078 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1082 kfree(sa, M_SONAME);
1083 iovec_free(&iov, aiov);
1090 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1091 * in kernel pointer instead of a userland pointer. This allows us
1092 * to manipulate socket options in the emulation code.
1095 kern_setsockopt(int s, struct sockopt *sopt)
1097 struct thread *td = curthread;
1098 struct proc *p = td->td_proc;
1102 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1104 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1106 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1109 error = holdsock(p->p_fd, s, &fp);
1113 error = sosetopt((struct socket *)fp->f_data, sopt);
1119 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1124 sys_setsockopt(struct setsockopt_args *uap)
1126 struct thread *td = curthread;
1127 struct sockopt sopt;
1130 sopt.sopt_level = uap->level;
1131 sopt.sopt_name = uap->name;
1132 sopt.sopt_valsize = uap->valsize;
1134 sopt.sopt_val = NULL;
1136 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1139 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1140 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1145 error = kern_setsockopt(uap->s, &sopt);
1148 kfree(sopt.sopt_val, M_TEMP);
1153 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1154 * in kernel pointer instead of a userland pointer. This allows us
1155 * to manipulate socket options in the emulation code.
1158 kern_getsockopt(int s, struct sockopt *sopt)
1160 struct thread *td = curthread;
1161 struct proc *p = td->td_proc;
1165 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1167 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1169 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1172 error = holdsock(p->p_fd, s, &fp);
1176 error = sogetopt((struct socket *)fp->f_data, sopt);
1182 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1187 sys_getsockopt(struct getsockopt_args *uap)
1189 struct thread *td = curthread;
1190 struct sockopt sopt;
1194 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1201 sopt.sopt_level = uap->level;
1202 sopt.sopt_name = uap->name;
1203 sopt.sopt_valsize = valsize;
1205 sopt.sopt_val = NULL;
1207 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1210 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1211 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1216 error = kern_getsockopt(uap->s, &sopt);
1219 valsize = sopt.sopt_valsize;
1220 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1224 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1227 kfree(sopt.sopt_val, M_TEMP);
1232 * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1233 * This allows kern_getsockname() to return a pointer to an allocated struct
1234 * sockaddr which must be freed later with FREE(). The caller must
1235 * initialize *name to NULL.
1238 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1240 struct thread *td = curthread;
1241 struct proc *p = td->td_proc;
1244 struct sockaddr *sa = NULL;
1247 error = holdsock(p->p_fd, s, &fp);
1254 so = (struct socket *)fp->f_data;
1255 error = so_pru_sockaddr(so, &sa);
1260 *namelen = MIN(*namelen, sa->sa_len);
1270 * getsockname_args(int fdes, caddr_t asa, int *alen)
1277 sys_getsockname(struct getsockname_args *uap)
1279 struct sockaddr *sa = NULL;
1282 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1286 error = kern_getsockname(uap->fdes, &sa, &sa_len);
1289 error = copyout(sa, uap->asa, sa_len);
1291 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1293 kfree(sa, M_SONAME);
1298 * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1299 * This allows kern_getpeername() to return a pointer to an allocated struct
1300 * sockaddr which must be freed later with FREE(). The caller must
1301 * initialize *name to NULL.
1304 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1306 struct thread *td = curthread;
1307 struct proc *p = td->td_proc;
1310 struct sockaddr *sa = NULL;
1313 error = holdsock(p->p_fd, s, &fp);
1320 so = (struct socket *)fp->f_data;
1321 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1325 error = so_pru_peeraddr(so, &sa);
1330 *namelen = MIN(*namelen, sa->sa_len);
1340 * getpeername_args(int fdes, caddr_t asa, int *alen)
1342 * Get name of peer for connected socket.
1347 sys_getpeername(struct getpeername_args *uap)
1349 struct sockaddr *sa = NULL;
1352 error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1356 error = kern_getpeername(uap->fdes, &sa, &sa_len);
1359 error = copyout(sa, uap->asa, sa_len);
1361 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1363 kfree(sa, M_SONAME);
1368 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1370 struct sockaddr *sa;
1374 if (len > SOCK_MAXADDRLEN)
1375 return ENAMETOOLONG;
1376 if (len < offsetof(struct sockaddr, sa_data[0]))
1378 sa = kmalloc(len, M_SONAME, M_WAITOK);
1379 error = copyin(uaddr, sa, len);
1381 kfree(sa, M_SONAME);
1383 #if BYTE_ORDER != BIG_ENDIAN
1385 * The bind(), connect(), and sendto() syscalls were not
1386 * versioned for COMPAT_43. Thus, this check must stay.
1388 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1389 sa->sa_family = sa->sa_len;
1398 * Detach a mapped page and release resources back to the system.
1399 * We must release our wiring and if the object is ripped out
1400 * from under the vm_page we become responsible for freeing the
1406 sf_buf_mfree(void *arg)
1408 struct sf_buf *sf = arg;
1411 m = sf_buf_page(sf);
1412 if (sf_buf_free(sf)) {
1413 /* sf invalid now */
1415 vm_page_busy_wait(m, FALSE, "sockpgf");
1420 if (m->object == NULL &&
1421 m->wire_count == 0 &&
1422 (m->flags & PG_NEED_COMMIT) == 0) {
1433 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1434 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1436 * Send a file specified by 'fd' and starting at 'offset' to a socket
1437 * specified by 's'. Send only 'nbytes' of the file or until EOF if
1438 * nbytes == 0. Optionally add a header and/or trailer to the socket
1439 * output. If specified, write the total number of bytes sent into *sbytes.
1441 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1442 * the headers to count against the remaining bytes to be sent from
1443 * the file descriptor. We may wish to implement a compatibility syscall
1449 sys_sendfile(struct sendfile_args *uap)
1451 struct thread *td = curthread;
1452 struct proc *p = td->td_proc;
1454 struct vnode *vp = NULL;
1455 struct sf_hdtr hdtr;
1456 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1458 struct mbuf *mheader = NULL;
1461 off_t hdtr_size = 0;
1468 * Do argument checking. Must be a regular file in, stream
1469 * type and connected socket out, positive offset.
1471 fp = holdfp(p->p_fd, uap->fd, FREAD);
1475 if (fp->f_type != DTYPE_VNODE) {
1479 vp = (struct vnode *)fp->f_data;
1484 * If specified, get the pointer to the sf_hdtr struct for
1485 * any headers/trailers.
1488 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1495 error = iovec_copyin(hdtr.headers, &iov, aiov,
1496 hdtr.hdr_cnt, &hbytes);
1500 auio.uio_iovcnt = hdtr.hdr_cnt;
1501 auio.uio_offset = 0;
1502 auio.uio_segflg = UIO_USERSPACE;
1503 auio.uio_rw = UIO_WRITE;
1505 auio.uio_resid = hbytes;
1507 mheader = m_uiomove(&auio);
1509 iovec_free(&iov, aiov);
1510 if (mheader == NULL)
1515 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1516 &sbytes, uap->flags);
1521 * Send trailers. Wimp out and use writev(2).
1523 if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1524 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1525 hdtr.trl_cnt, &auio.uio_resid);
1529 auio.uio_iovcnt = hdtr.trl_cnt;
1530 auio.uio_offset = 0;
1531 auio.uio_segflg = UIO_USERSPACE;
1532 auio.uio_rw = UIO_WRITE;
1535 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1537 iovec_free(&iov, aiov);
1540 hdtr_size += tbytes; /* trailer bytes successfully sent */
1546 if (uap->sbytes != NULL) {
1547 sbytes += hdtr_size;
1548 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1554 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1555 struct mbuf *mheader, off_t *sbytes, int flags)
1557 struct thread *td = curthread;
1558 struct proc *p = td->td_proc;
1559 struct vm_object *obj;
1562 struct mbuf *m, *mp;
1565 off_t off, xfsize, xbytes;
1569 if (vp->v_type != VREG) {
1573 if ((obj = vp->v_object) == NULL) {
1577 error = holdsock(p->p_fd, sfd, &fp);
1580 so = (struct socket *)fp->f_data;
1581 if (so->so_type != SOCK_STREAM) {
1585 if ((so->so_state & SS_ISCONNECTED) == 0) {
1595 * preallocation is required for asynchronous passing of mbufs,
1596 * otherwise we can wind up building up an infinite number of
1597 * mbufs during the asynchronous latency.
1599 if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) {
1607 * Protect against multiple writers to the socket.
1609 ssb_lock(&so->so_snd, M_WAITOK);
1612 * Loop through the pages in the file, starting with the requested
1613 * offset. Get a file page (do I/O if necessary), map the file page
1614 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1617 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) {
1622 pindex = OFF_TO_IDX(off);
1625 * Calculate the amount to transfer. Not to exceed a page,
1626 * the EOF, or the passed in nbytes.
1628 xfsize = vp->v_filesize - off;
1629 if (xfsize > PAGE_SIZE)
1631 pgoff = (vm_offset_t)(off & PAGE_MASK);
1632 if (PAGE_SIZE - pgoff < xfsize)
1633 xfsize = PAGE_SIZE - pgoff;
1634 if (nbytes && xfsize > (nbytes - xbytes))
1635 xfsize = nbytes - xbytes;
1639 * Optimize the non-blocking case by looking at the socket space
1640 * before going to the extra work of constituting the sf_buf.
1642 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1643 space = ssb_space_prealloc(&so->so_snd);
1645 space = ssb_space(&so->so_snd);
1647 if ((fp->f_flag & FNONBLOCK) && space <= 0) {
1648 if (so->so_state & SS_CANTSENDMORE)
1652 ssb_unlock(&so->so_snd);
1656 * Attempt to look up the page.
1658 * Allocate if not found, wait and loop if busy, then hold the page.
1659 * We hold rather than wire the page because we do not want to prevent
1660 * filesystem truncation operations from occuring on the file. This
1661 * can happen even under normal operation if the file being sent is
1662 * remove()d after the sendfile() call completes, because the socket buffer
1663 * may still be draining. tmpfs will crash if we try to use wire.
1665 vm_object_hold(obj);
1666 pg = vm_page_lookup_busy_try(obj, pindex, TRUE, &error);
1668 vm_page_sleep_busy(pg, TRUE, "sfpbsy");
1669 vm_object_drop(obj);
1673 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL |
1677 vm_object_drop(obj);
1682 vm_object_drop(obj);
1685 * If page is not valid for what we need, initiate I/O
1688 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1694 * Ensure that our page is still around when the I/O
1697 * Ensure that our page is not modified while part of
1698 * a mbuf as this could mess up tcp checksums, DMA,
1699 * etc (XXX NEEDS WORK). The softbusy is supposed to
1700 * help here but it actually doesn't.
1702 * XXX THIS HAS MULTIPLE PROBLEMS. The underlying
1703 * VM pages are not protected by the soft-busy
1704 * unless we vm_page_protect... READ them, and
1705 * they STILL aren't protected against
1706 * modification via the buffer cache (VOP_WRITE).
1708 * Fixing the second issue is particularly
1711 * XXX We also can't soft-busy anyway because it can
1712 * deadlock against the syncer doing a vfs_msync(),
1713 * vfs_msync->vmntvnodesca->vfs_msync_scan2->
1714 * vm_object_page_clean->(scan)-> ... page
1717 /*vm_page_io_start(pg);*/
1721 * Get the page from backing store.
1723 bsize = vp->v_mount->mnt_stat.f_iosize;
1724 auio.uio_iov = &aiov;
1725 auio.uio_iovcnt = 1;
1727 aiov.iov_len = MAXBSIZE;
1728 auio.uio_resid = MAXBSIZE;
1729 auio.uio_offset = trunc_page(off);
1730 auio.uio_segflg = UIO_NOCOPY;
1731 auio.uio_rw = UIO_READ;
1733 vn_lock(vp, LK_SHARED | LK_RETRY);
1734 error = VOP_READ(vp, &auio,
1735 IO_VMIO | ((MAXBSIZE / bsize) << 16),
1738 vm_page_flag_clear(pg, PG_ZERO);
1739 vm_page_busy_wait(pg, FALSE, "sockpg");
1740 /*vm_page_io_finish(pg);*/
1744 /* vm_page_try_to_free(pg); */
1745 ssb_unlock(&so->so_snd);
1752 * Get a sendfile buf. We usually wait as long as necessary,
1753 * but this wait can be interrupted.
1755 if ((sf = sf_buf_alloc(pg)) == NULL) {
1758 /* vm_page_try_to_free(pg); */
1759 ssb_unlock(&so->so_snd);
1765 * Get an mbuf header and set it up as having external storage.
1767 MGETHDR(m, M_WAITOK, MT_DATA);
1772 /* vm_page_try_to_free(pg); */
1774 ssb_unlock(&so->so_snd);
1780 m->m_ext.ext_free = sf_buf_mfree;
1781 m->m_ext.ext_ref = sf_buf_ref;
1782 m->m_ext.ext_arg = sf;
1783 m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1784 m->m_ext.ext_size = PAGE_SIZE;
1785 m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1786 m->m_flags |= M_EXT;
1787 m->m_pkthdr.len = m->m_len = xfsize;
1788 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1790 if (mheader != NULL) {
1791 hbytes = mheader->m_pkthdr.len;
1792 mheader->m_pkthdr.len += m->m_pkthdr.len;
1800 * Add the buffer to the socket buffer chain.
1805 * Make sure that the socket is still able to take more data.
1806 * CANTSENDMORE being true usually means that the connection
1807 * was closed. so_error is true when an error was sensed after
1809 * The state is checked after the page mapping and buffer
1810 * allocation above since those operations may block and make
1811 * any socket checks stale. From this point forward, nothing
1812 * blocks before the pru_send (or more accurately, any blocking
1813 * results in a loop back to here to re-check).
1815 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1816 if (so->so_state & SS_CANTSENDMORE) {
1819 error = so->so_error;
1823 ssb_unlock(&so->so_snd);
1828 * Wait for socket space to become available. We do this just
1829 * after checking the connection state above in order to avoid
1830 * a race condition with ssb_wait().
1832 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1833 space = ssb_space_prealloc(&so->so_snd);
1835 space = ssb_space(&so->so_snd);
1837 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) {
1838 if (fp->f_flag & FNONBLOCK) {
1840 ssb_unlock(&so->so_snd);
1845 error = ssb_wait(&so->so_snd);
1847 * An error from ssb_wait usually indicates that we've
1848 * been interrupted by a signal. If we've sent anything
1849 * then return bytes sent, otherwise return the error.
1853 ssb_unlock(&so->so_snd);
1860 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1861 for (mp = m; mp != NULL; mp = mp->m_next)
1862 ssb_preallocstream(&so->so_snd, mp);
1864 if (use_sendfile_async)
1865 error = so_pru_senda(so, 0, m, NULL, NULL, td);
1867 error = so_pru_send(so, 0, m, NULL, NULL, td);
1871 ssb_unlock(&so->so_snd);
1875 if (mheader != NULL) {
1876 *sbytes += mheader->m_pkthdr.len;
1878 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1879 for (mp = mheader; mp != NULL; mp = mp->m_next)
1880 ssb_preallocstream(&so->so_snd, mp);
1882 if (use_sendfile_async)
1883 error = so_pru_senda(so, 0, mheader, NULL, NULL, td);
1885 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1889 ssb_unlock(&so->so_snd);
1894 if (mheader != NULL)