1 /* $NetBSD: svc_vc.c,v 1.7 2000/08/03 00:01:53 fvdl Exp $ */
4 * Sun RPC is a product of Sun Microsystems, Inc. and is provided for
5 * unrestricted use provided that this legend is included on all tape
6 * media and as a part of the software program in whole or part. Users
7 * may copy or modify Sun RPC without charge, but are not authorized
8 * to license or distribute it to anyone else except as part of a product or
9 * program developed by the user.
11 * SUN RPC IS PROVIDED AS IS WITH NO WARRANTIES OF ANY KIND INCLUDING THE
12 * WARRANTIES OF DESIGN, MERCHANTIBILITY AND FITNESS FOR A PARTICULAR
13 * PURPOSE, OR ARISING FROM A COURSE OF DEALING, USAGE OR TRADE PRACTICE.
15 * Sun RPC is provided with no support and without any obligation on the
16 * part of Sun Microsystems, Inc. to assist in its use, correction,
17 * modification or enhancement.
19 * SUN MICROSYSTEMS, INC. SHALL HAVE NO LIABILITY WITH RESPECT TO THE
20 * INFRINGEMENT OF COPYRIGHTS, TRADE SECRETS OR ANY PATENTS BY SUN RPC
21 * OR ANY PART THEREOF.
23 * In no event will Sun Microsystems, Inc. be liable for any lost revenue
24 * or profits or other special, indirect and consequential damages, even if
25 * Sun has been advised of the possibility of such damages.
27 * Sun Microsystems, Inc.
29 * Mountain View, California 94043
32 #if defined(LIBC_SCCS) && !defined(lint)
33 static char *sccsid2 = "@(#)svc_tcp.c 1.21 87/08/11 Copyr 1984 Sun Micro";
34 static char *sccsid = "@(#)svc_tcp.c 2.2 88/08/01 4.0 RPCSRC";
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
40 * svc_vc.c, Server side for Connection Oriented based RPC.
42 * Actually implements two flavors of transporter -
43 * a tcp rendezvouser (a listner and connection establisher)
44 * and a record/tcp stream.
47 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
52 #include <sys/mutex.h>
53 #include <sys/protosw.h>
54 #include <sys/queue.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/systm.h>
59 #include <netinet/tcp.h>
63 #include <rpc/rpc_com.h>
65 static bool_t svc_vc_rendezvous_recv(SVCXPRT *, struct rpc_msg *);
66 static enum xprt_stat svc_vc_rendezvous_stat(SVCXPRT *);
67 static void svc_vc_rendezvous_destroy(SVCXPRT *);
68 static bool_t svc_vc_null(void);
69 static void svc_vc_destroy(SVCXPRT *);
70 static enum xprt_stat svc_vc_stat(SVCXPRT *);
71 static bool_t svc_vc_recv(SVCXPRT *, struct rpc_msg *);
72 static bool_t svc_vc_getargs(SVCXPRT *, xdrproc_t, void *);
73 static bool_t svc_vc_freeargs(SVCXPRT *, xdrproc_t, void *);
74 static bool_t svc_vc_reply(SVCXPRT *, struct rpc_msg *);
75 static bool_t svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in);
76 static bool_t svc_vc_rendezvous_control (SVCXPRT *xprt, const u_int rq,
78 static SVCXPRT *svc_vc_create_conn(SVCPOOL *pool, struct socket *so,
79 struct sockaddr *raddr);
80 static int svc_vc_accept(struct socket *head, struct socket **sop);
81 static void svc_vc_soupcall(struct socket *so, void *arg, int waitflag);
83 static struct xp_ops svc_vc_rendezvous_ops = {
84 .xp_recv = svc_vc_rendezvous_recv,
85 .xp_stat = svc_vc_rendezvous_stat,
86 .xp_getargs = (bool_t (*)(SVCXPRT *, xdrproc_t, void *))svc_vc_null,
87 .xp_reply = (bool_t (*)(SVCXPRT *, struct rpc_msg *))svc_vc_null,
88 .xp_freeargs = (bool_t (*)(SVCXPRT *, xdrproc_t, void *))svc_vc_null,
89 .xp_destroy = svc_vc_rendezvous_destroy,
90 .xp_control = svc_vc_rendezvous_control
93 static struct xp_ops svc_vc_ops = {
94 .xp_recv = svc_vc_recv,
95 .xp_stat = svc_vc_stat,
96 .xp_getargs = svc_vc_getargs,
97 .xp_reply = svc_vc_reply,
98 .xp_freeargs = svc_vc_freeargs,
99 .xp_destroy = svc_vc_destroy,
100 .xp_control = svc_vc_control
103 struct cf_conn { /* kept in xprt->xp_p1 for actual connection */
104 enum xprt_stat strm_stat;
105 struct mbuf *mpending; /* unparsed data read from the socket */
106 struct mbuf *mreq; /* current record being built from mpending */
107 uint32_t resid; /* number of bytes needed for fragment */
108 bool_t eor; /* reading last fragment of current record */
113 * xprt = svc_vc_create(sock, send_buf_size, recv_buf_size);
115 * Creates, registers, and returns a (rpc) tcp based transporter.
116 * Once *xprt is initialized, it is registered as a transporter
117 * see (svc.h, xprt_register). This routine returns
118 * a NULL if a problem occurred.
120 * The filedescriptor passed in is expected to refer to a bound, but
121 * not yet connected socket.
123 * Since streams do buffered io similar to stdio, the caller can specify
124 * how big the send and receive buffers are via the second and third parms;
125 * 0 => use the system default.
128 svc_vc_create(SVCPOOL *pool, struct socket *so, size_t sendsize,
135 if (so->so_state & SS_ISCONNECTED) {
136 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
139 xprt = svc_vc_create_conn(pool, so, sa);
144 xprt = mem_alloc(sizeof(SVCXPRT));
145 mtx_init(&xprt->xp_lock, "xprt->xp_lock", NULL, MTX_DEF);
146 xprt->xp_pool = pool;
147 xprt->xp_socket = so;
151 xprt->xp_verf = _null_auth;
152 xprt->xp_ops = &svc_vc_rendezvous_ops;
154 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
156 goto cleanup_svc_vc_create;
158 xprt->xp_ltaddr.buf = mem_alloc(sizeof (struct sockaddr_storage));
159 xprt->xp_ltaddr.maxlen = sizeof (struct sockaddr_storage);
160 xprt->xp_ltaddr.len = sa->sa_len;
161 memcpy(xprt->xp_ltaddr.buf, sa, sa->sa_len);
164 xprt->xp_rtaddr.maxlen = 0;
168 solisten(so, SOMAXCONN, curthread);
170 SOCKBUF_LOCK(&so->so_rcv);
171 so->so_upcallarg = xprt;
172 so->so_upcall = svc_vc_soupcall;
173 so->so_rcv.sb_flags |= SB_UPCALL;
174 SOCKBUF_UNLOCK(&so->so_rcv);
177 cleanup_svc_vc_create:
179 mem_free(xprt, sizeof(*xprt));
184 * Create a new transport for a socket optained via soaccept().
187 svc_vc_create_conn(SVCPOOL *pool, struct socket *so, struct sockaddr *raddr)
189 SVCXPRT *xprt = NULL;
190 struct cf_conn *cd = NULL;
191 struct sockaddr* sa = NULL;
196 bzero(&opt, sizeof(struct sockopt));
197 opt.sopt_dir = SOPT_SET;
198 opt.sopt_level = SOL_SOCKET;
199 opt.sopt_name = SO_KEEPALIVE;
201 opt.sopt_valsize = sizeof(one);
202 error = sosetopt(so, &opt);
206 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
207 bzero(&opt, sizeof(struct sockopt));
208 opt.sopt_dir = SOPT_SET;
209 opt.sopt_level = IPPROTO_TCP;
210 opt.sopt_name = TCP_NODELAY;
212 opt.sopt_valsize = sizeof(one);
213 error = sosetopt(so, &opt);
218 cd = mem_alloc(sizeof(*cd));
219 cd->strm_stat = XPRT_IDLE;
221 xprt = mem_alloc(sizeof(SVCXPRT));
222 mtx_init(&xprt->xp_lock, "xprt->xp_lock", NULL, MTX_DEF);
223 xprt->xp_pool = pool;
224 xprt->xp_socket = so;
228 xprt->xp_verf = _null_auth;
229 xprt->xp_ops = &svc_vc_ops;
231 xprt->xp_rtaddr.buf = mem_alloc(sizeof (struct sockaddr_storage));
232 xprt->xp_rtaddr.maxlen = sizeof (struct sockaddr_storage);
233 xprt->xp_rtaddr.len = raddr->sa_len;
234 memcpy(xprt->xp_rtaddr.buf, raddr, raddr->sa_len);
236 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
238 goto cleanup_svc_vc_create;
240 xprt->xp_ltaddr.buf = mem_alloc(sizeof (struct sockaddr_storage));
241 xprt->xp_ltaddr.maxlen = sizeof (struct sockaddr_storage);
242 xprt->xp_ltaddr.len = sa->sa_len;
243 memcpy(xprt->xp_ltaddr.buf, sa, sa->sa_len);
248 SOCKBUF_LOCK(&so->so_rcv);
249 so->so_upcallarg = xprt;
250 so->so_upcall = svc_vc_soupcall;
251 so->so_rcv.sb_flags |= SB_UPCALL;
252 SOCKBUF_UNLOCK(&so->so_rcv);
255 * Throw the transport into the active list in case it already
256 * has some data buffered.
258 mtx_lock(&xprt->xp_lock);
260 mtx_unlock(&xprt->xp_lock);
263 cleanup_svc_vc_create:
265 if (xprt->xp_ltaddr.buf)
266 mem_free(xprt->xp_ltaddr.buf,
267 sizeof(struct sockaddr_storage));
268 if (xprt->xp_rtaddr.buf)
269 mem_free(xprt->xp_rtaddr.buf,
270 sizeof(struct sockaddr_storage));
271 mem_free(xprt, sizeof(*xprt));
274 mem_free(cd, sizeof(*cd));
279 * This does all of the accept except the final call to soaccept. The
280 * caller will call soaccept after dropping its locks (soaccept may
284 svc_vc_accept(struct socket *head, struct socket **sop)
289 if ((head->so_options & SO_ACCEPTCONN) == 0) {
295 error = mac_socket_check_accept(td->td_ucred, head);
301 if (TAILQ_EMPTY(&head->so_comp)) {
306 so = TAILQ_FIRST(&head->so_comp);
307 KASSERT(!(so->so_qstate & SQ_INCOMP), ("svc_vc_accept: so SQ_INCOMP"));
308 KASSERT(so->so_qstate & SQ_COMP, ("svc_vc_accept: so not SQ_COMP"));
311 * Before changing the flags on the socket, we have to bump the
312 * reference count. Otherwise, if the protocol calls sofree(),
313 * the socket will be released due to a zero refcount.
314 * XXX might not need soref() since this is simpler than kern_accept.
316 SOCK_LOCK(so); /* soref() and so_state update */
317 soref(so); /* file descriptor reference */
319 TAILQ_REMOVE(&head->so_comp, so, so_list);
321 so->so_state |= (head->so_state & SS_NBIO);
322 so->so_qstate &= ~SQ_COMP;
330 /* connection has been removed from the listen queue */
331 KNOTE_UNLOCKED(&head->so_rcv.sb_sel.si_note, 0);
338 svc_vc_rendezvous_recv(SVCXPRT *xprt, struct rpc_msg *msg)
340 struct socket *so = NULL;
341 struct sockaddr *sa = NULL;
345 * The socket upcall calls xprt_active() which will eventually
346 * cause the server to call us here. We attempt to accept a
347 * connection from the socket and turn it into a new
348 * transport. If the accept fails, we have drained all pending
349 * connections so we call xprt_inactive().
351 * The lock protects us in the case where a new connection arrives
352 * on the socket after our call to accept fails with
353 * EWOULDBLOCK - the call to xprt_active() in the upcall will
354 * happen only after our call to xprt_inactive() which ensures
355 * that we will remain active. It might be possible to use
356 * SOCKBUF_LOCK for this - its not clear to me what locks are
357 * held during the upcall.
359 mtx_lock(&xprt->xp_lock);
361 error = svc_vc_accept(xprt->xp_socket, &so);
363 if (error == EWOULDBLOCK) {
365 mtx_unlock(&xprt->xp_lock);
370 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
371 xprt->xp_socket->so_upcallarg = NULL;
372 xprt->xp_socket->so_upcall = NULL;
373 xprt->xp_socket->so_rcv.sb_flags &= ~SB_UPCALL;
374 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
376 mtx_unlock(&xprt->xp_lock);
380 mtx_unlock(&xprt->xp_lock);
383 error = soaccept(so, &sa);
387 * XXX not sure if I need to call sofree or soclose here.
395 * svc_vc_create_conn will call xprt_register - we don't need
396 * to do anything with the new connection.
398 if (!svc_vc_create_conn(xprt->xp_pool, so, sa))
403 return (FALSE); /* there is never an rpc msg to be processed */
407 static enum xprt_stat
408 svc_vc_rendezvous_stat(SVCXPRT *xprt)
415 svc_vc_destroy_common(SVCXPRT *xprt)
417 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
418 xprt->xp_socket->so_upcallarg = NULL;
419 xprt->xp_socket->so_upcall = NULL;
420 xprt->xp_socket->so_rcv.sb_flags &= ~SB_UPCALL;
421 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
423 xprt_unregister(xprt);
425 mtx_destroy(&xprt->xp_lock);
427 (void)soclose(xprt->xp_socket);
429 if (xprt->xp_rtaddr.buf)
430 (void) mem_free(xprt->xp_rtaddr.buf, xprt->xp_rtaddr.maxlen);
431 if (xprt->xp_ltaddr.buf)
432 (void) mem_free(xprt->xp_ltaddr.buf, xprt->xp_ltaddr.maxlen);
433 (void) mem_free(xprt, sizeof (SVCXPRT));
438 svc_vc_rendezvous_destroy(SVCXPRT *xprt)
441 svc_vc_destroy_common(xprt);
445 svc_vc_destroy(SVCXPRT *xprt)
447 struct cf_conn *cd = (struct cf_conn *)xprt->xp_p1;
449 svc_vc_destroy_common(xprt);
454 m_freem(cd->mpending);
455 mem_free(cd, sizeof(*cd));
460 svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in)
466 svc_vc_rendezvous_control(SVCXPRT *xprt, const u_int rq, void *in)
472 static enum xprt_stat
473 svc_vc_stat(SVCXPRT *xprt)
479 cd = (struct cf_conn *)(xprt->xp_p1);
481 if (cd->strm_stat == XPRT_DIED)
485 * Return XPRT_MOREREQS if we have buffered data and we are
486 * mid-record or if we have enough data for a record marker.
490 return (XPRT_MOREREQS);
493 while (m && n < sizeof(uint32_t)) {
497 if (n >= sizeof(uint32_t))
498 return (XPRT_MOREREQS);
505 svc_vc_recv(SVCXPRT *xprt, struct rpc_msg *msg)
507 struct cf_conn *cd = (struct cf_conn *) xprt->xp_p1;
514 * If we have an mbuf chain in cd->mpending, try to parse a
515 * record from it, leaving the result in cd->mreq. If we don't
516 * have a complete record, leave the partial result in
517 * cd->mreq and try to read more from the socket.
521 * If cd->resid is non-zero, we have part of the
522 * record already, otherwise we are expecting a record
527 * See if there is enough data buffered to
528 * make up a record marker. Make sure we can
529 * handle the case where the record marker is
530 * split across more than one mbuf.
536 while (n < sizeof(uint32_t) && m) {
540 if (n < sizeof(uint32_t))
542 cd->mpending = m_pullup(cd->mpending, sizeof(uint32_t));
543 memcpy(&header, mtod(cd->mpending, uint32_t *),
545 header = ntohl(header);
546 cd->eor = (header & 0x80000000) != 0;
547 cd->resid = header & 0x7fffffff;
548 m_adj(cd->mpending, sizeof(uint32_t));
552 * Start pulling off mbufs from cd->mpending
553 * until we either have a complete record or
554 * we run out of data. We use m_split to pull
555 * data - it will pull as much as possible and
556 * split the last mbuf if necessary.
558 while (cd->mpending && cd->resid) {
560 cd->mpending = m_split(cd->mpending, cd->resid,
563 m_last(cd->mreq)->m_next = m;
567 cd->resid -= m->m_len;
573 * If cd->resid is zero now, we have managed to
574 * receive a record fragment from the stream. Check
575 * for the end-of-record mark to see if we need more.
577 if (cd->resid == 0) {
582 * Success - we have a complete record in
585 xdrmbuf_create(&xprt->xp_xdrreq, cd->mreq, XDR_DECODE);
587 if (! xdr_callmsg(&xprt->xp_xdrreq, msg)) {
588 XDR_DESTROY(&xprt->xp_xdrreq);
591 xprt->xp_xid = msg->rm_xid;
599 * The socket upcall calls xprt_active() which will eventually
600 * cause the server to call us here. We attempt to
601 * read as much as possible from the socket and put
602 * the result in cd->mpending. If the read fails,
603 * we have drained both cd->mpending and the socket so
604 * we can call xprt_inactive().
606 * The lock protects us in the case where a new packet arrives
607 * on the socket after our call to soreceive fails with
608 * EWOULDBLOCK - the call to xprt_active() in the upcall will
609 * happen only after our call to xprt_inactive() which ensures
610 * that we will remain active. It might be possible to use
611 * SOCKBUF_LOCK for this - its not clear to me what locks are
612 * held during the upcall.
614 mtx_lock(&xprt->xp_lock);
616 uio.uio_resid = 1000000000;
617 uio.uio_td = curthread;
619 rcvflag = MSG_DONTWAIT;
620 error = soreceive(xprt->xp_socket, NULL, &uio, &m, NULL,
623 if (error == EWOULDBLOCK) {
625 mtx_unlock(&xprt->xp_lock);
630 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
631 xprt->xp_socket->so_upcallarg = NULL;
632 xprt->xp_socket->so_upcall = NULL;
633 xprt->xp_socket->so_rcv.sb_flags &= ~SB_UPCALL;
634 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
636 cd->strm_stat = XPRT_DIED;
637 mtx_unlock(&xprt->xp_lock);
643 * EOF - the other end has closed the socket.
645 cd->strm_stat = XPRT_DIED;
646 mtx_unlock(&xprt->xp_lock);
651 m_last(cd->mpending)->m_next = m;
655 mtx_unlock(&xprt->xp_lock);
660 svc_vc_getargs(SVCXPRT *xprt, xdrproc_t xdr_args, void *args_ptr)
663 return (xdr_args(&xprt->xp_xdrreq, args_ptr));
667 svc_vc_freeargs(SVCXPRT *xprt, xdrproc_t xdr_args, void *args_ptr)
672 * Free the request mbuf here - this allows us to handle
673 * protocols where not all requests have replies
674 * (i.e. NLM). Note that xdrmbuf_destroy handles being called
675 * twice correctly - the mbuf will only be freed once.
677 XDR_DESTROY(&xprt->xp_xdrreq);
679 xdrs.x_op = XDR_FREE;
680 return (xdr_args(&xdrs, args_ptr));
684 svc_vc_reply(SVCXPRT *xprt, struct rpc_msg *msg)
691 * Leave space for record mark.
693 MGETHDR(mrep, M_WAIT, MT_DATA);
694 MCLGET(mrep, M_WAIT);
696 mrep->m_data += sizeof(uint32_t);
698 xdrmbuf_create(&xprt->xp_xdrrep, mrep, XDR_ENCODE);
699 msg->rm_xid = xprt->xp_xid;
700 if (xdr_replymsg(&xprt->xp_xdrrep, msg)) {
704 * Prepend a record marker containing the reply length.
706 M_PREPEND(mrep, sizeof(uint32_t), M_WAIT);
707 *mtod(mrep, uint32_t *) =
708 htonl(0x80000000 | (mrep->m_pkthdr.len
709 - sizeof(uint32_t)));
710 error = sosend(xprt->xp_socket, NULL, NULL, mrep, NULL,
720 * This frees the request mbuf chain as well. The reply mbuf
721 * chain was consumed by sosend.
723 XDR_DESTROY(&xprt->xp_xdrreq);
724 XDR_DESTROY(&xprt->xp_xdrrep);
738 svc_vc_soupcall(struct socket *so, void *arg, int waitflag)
740 SVCXPRT *xprt = (SVCXPRT *) arg;
742 mtx_lock(&xprt->xp_lock);
744 mtx_unlock(&xprt->xp_lock);
749 * Get the effective UID of the sending process. Used by rpcbind, keyserv
750 * and rpc.yppasswdd on AF_LOCAL.
753 __rpc_get_local_uid(SVCXPRT *transp, uid_t *uid) {
759 sock = transp->xp_fd;
760 sa = (struct sockaddr *)transp->xp_rtaddr.buf;
761 if (sa->sa_family == AF_LOCAL) {
762 ret = getpeereid(sock, &euid, &egid);