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 xprt = mem_alloc(sizeof(SVCXPRT));
136 mtx_init(&xprt->xp_lock, "xprt->xp_lock", NULL, MTX_DEF);
137 xprt->xp_pool = pool;
138 xprt->xp_socket = so;
142 xprt->xp_verf = _null_auth;
143 xprt->xp_ops = &svc_vc_rendezvous_ops;
145 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
147 goto cleanup_svc_vc_create;
149 xprt->xp_ltaddr.buf = mem_alloc(sizeof (struct sockaddr_storage));
150 xprt->xp_ltaddr.maxlen = sizeof (struct sockaddr_storage);
151 xprt->xp_ltaddr.len = sa->sa_len;
152 memcpy(xprt->xp_ltaddr.buf, sa, sa->sa_len);
155 xprt->xp_rtaddr.maxlen = 0;
159 solisten(so, SOMAXCONN, curthread);
161 SOCKBUF_LOCK(&so->so_rcv);
162 so->so_upcallarg = xprt;
163 so->so_upcall = svc_vc_soupcall;
164 so->so_rcv.sb_flags |= SB_UPCALL;
165 SOCKBUF_UNLOCK(&so->so_rcv);
168 cleanup_svc_vc_create:
170 mem_free(xprt, sizeof(*xprt));
175 * Create a new transport for a socket optained via soaccept().
178 svc_vc_create_conn(SVCPOOL *pool, struct socket *so, struct sockaddr *raddr)
180 SVCXPRT *xprt = NULL;
181 struct cf_conn *cd = NULL;
182 struct sockaddr* sa = NULL;
185 cd = mem_alloc(sizeof(*cd));
186 cd->strm_stat = XPRT_IDLE;
188 xprt = mem_alloc(sizeof(SVCXPRT));
189 mtx_init(&xprt->xp_lock, "xprt->xp_lock", NULL, MTX_DEF);
190 xprt->xp_pool = pool;
191 xprt->xp_socket = so;
195 xprt->xp_verf = _null_auth;
196 xprt->xp_ops = &svc_vc_ops;
198 xprt->xp_rtaddr.buf = mem_alloc(sizeof (struct sockaddr_storage));
199 xprt->xp_rtaddr.maxlen = sizeof (struct sockaddr_storage);
200 xprt->xp_rtaddr.len = raddr->sa_len;
201 memcpy(xprt->xp_rtaddr.buf, raddr, raddr->sa_len);
203 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
205 goto cleanup_svc_vc_create;
207 xprt->xp_ltaddr.buf = mem_alloc(sizeof (struct sockaddr_storage));
208 xprt->xp_ltaddr.maxlen = sizeof (struct sockaddr_storage);
209 xprt->xp_ltaddr.len = sa->sa_len;
210 memcpy(xprt->xp_ltaddr.buf, sa, sa->sa_len);
215 SOCKBUF_LOCK(&so->so_rcv);
216 so->so_upcallarg = xprt;
217 so->so_upcall = svc_vc_soupcall;
218 so->so_rcv.sb_flags |= SB_UPCALL;
219 SOCKBUF_UNLOCK(&so->so_rcv);
222 * Throw the transport into the active list in case it already
223 * has some data buffered.
225 mtx_lock(&xprt->xp_lock);
227 mtx_unlock(&xprt->xp_lock);
230 cleanup_svc_vc_create:
232 if (xprt->xp_ltaddr.buf)
233 mem_free(xprt->xp_ltaddr.buf,
234 sizeof(struct sockaddr_storage));
235 if (xprt->xp_rtaddr.buf)
236 mem_free(xprt->xp_rtaddr.buf,
237 sizeof(struct sockaddr_storage));
238 mem_free(xprt, sizeof(*xprt));
241 mem_free(cd, sizeof(*cd));
246 * This does all of the accept except the final call to soaccept. The
247 * caller will call soaccept after dropping its locks (soaccept may
251 svc_vc_accept(struct socket *head, struct socket **sop)
256 if ((head->so_options & SO_ACCEPTCONN) == 0) {
262 error = mac_socket_check_accept(td->td_ucred, head);
268 if (TAILQ_EMPTY(&head->so_comp)) {
273 so = TAILQ_FIRST(&head->so_comp);
274 KASSERT(!(so->so_qstate & SQ_INCOMP), ("svc_vc_accept: so SQ_INCOMP"));
275 KASSERT(so->so_qstate & SQ_COMP, ("svc_vc_accept: so not SQ_COMP"));
278 * Before changing the flags on the socket, we have to bump the
279 * reference count. Otherwise, if the protocol calls sofree(),
280 * the socket will be released due to a zero refcount.
281 * XXX might not need soref() since this is simpler than kern_accept.
283 SOCK_LOCK(so); /* soref() and so_state update */
284 soref(so); /* file descriptor reference */
286 TAILQ_REMOVE(&head->so_comp, so, so_list);
288 so->so_state |= (head->so_state & SS_NBIO);
289 so->so_qstate &= ~SQ_COMP;
297 /* connection has been removed from the listen queue */
298 KNOTE_UNLOCKED(&head->so_rcv.sb_sel.si_note, 0);
305 svc_vc_rendezvous_recv(SVCXPRT *xprt, struct rpc_msg *msg)
307 struct socket *so = NULL;
308 struct sockaddr *sa = NULL;
314 * The socket upcall calls xprt_active() which will eventually
315 * cause the server to call us here. We attempt to accept a
316 * connection from the socket and turn it into a new
317 * transport. If the accept fails, we have drained all pending
318 * connections so we call xprt_inactive().
320 * The lock protects us in the case where a new connection arrives
321 * on the socket after our call to accept fails with
322 * EWOULDBLOCK - the call to xprt_active() in the upcall will
323 * happen only after our call to xprt_inactive() which ensures
324 * that we will remain active. It might be possible to use
325 * SOCKBUF_LOCK for this - its not clear to me what locks are
326 * held during the upcall.
328 mtx_lock(&xprt->xp_lock);
330 error = svc_vc_accept(xprt->xp_socket, &so);
332 if (error == EWOULDBLOCK) {
334 mtx_unlock(&xprt->xp_lock);
339 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
340 xprt->xp_socket->so_upcallarg = NULL;
341 xprt->xp_socket->so_upcall = NULL;
342 xprt->xp_socket->so_rcv.sb_flags &= ~SB_UPCALL;
343 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
345 mtx_unlock(&xprt->xp_lock);
349 mtx_unlock(&xprt->xp_lock);
352 error = soaccept(so, &sa);
355 bzero(&opt, sizeof(struct sockopt));
356 opt.sopt_dir = SOPT_SET;
357 opt.sopt_level = IPPROTO_TCP;
358 opt.sopt_name = TCP_NODELAY;
360 opt.sopt_valsize = sizeof(one);
361 error = sosetopt(so, &opt);
366 * XXX not sure if I need to call sofree or soclose here.
374 * svc_vc_create_conn will call xprt_register - we don't need
375 * to do anything with the new connection.
377 svc_vc_create_conn(xprt->xp_pool, so, sa);
380 return (FALSE); /* there is never an rpc msg to be processed */
384 static enum xprt_stat
385 svc_vc_rendezvous_stat(SVCXPRT *xprt)
392 svc_vc_destroy_common(SVCXPRT *xprt)
394 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
395 xprt->xp_socket->so_upcallarg = NULL;
396 xprt->xp_socket->so_upcall = NULL;
397 xprt->xp_socket->so_rcv.sb_flags &= ~SB_UPCALL;
398 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
400 xprt_unregister(xprt);
402 mtx_destroy(&xprt->xp_lock);
404 (void)soclose(xprt->xp_socket);
406 if (xprt->xp_rtaddr.buf)
407 (void) mem_free(xprt->xp_rtaddr.buf, xprt->xp_rtaddr.maxlen);
408 if (xprt->xp_ltaddr.buf)
409 (void) mem_free(xprt->xp_ltaddr.buf, xprt->xp_ltaddr.maxlen);
410 (void) mem_free(xprt, sizeof (SVCXPRT));
415 svc_vc_rendezvous_destroy(SVCXPRT *xprt)
418 svc_vc_destroy_common(xprt);
422 svc_vc_destroy(SVCXPRT *xprt)
424 struct cf_conn *cd = (struct cf_conn *)xprt->xp_p1;
426 svc_vc_destroy_common(xprt);
431 m_freem(cd->mpending);
432 mem_free(cd, sizeof(*cd));
437 svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in)
443 svc_vc_rendezvous_control(SVCXPRT *xprt, const u_int rq, void *in)
449 static enum xprt_stat
450 svc_vc_stat(SVCXPRT *xprt)
456 cd = (struct cf_conn *)(xprt->xp_p1);
458 if (cd->strm_stat == XPRT_DIED)
462 * Return XPRT_MOREREQS if we have buffered data and we are
463 * mid-record or if we have enough data for a record marker.
467 return (XPRT_MOREREQS);
470 while (m && n < sizeof(uint32_t)) {
474 if (n >= sizeof(uint32_t))
475 return (XPRT_MOREREQS);
482 svc_vc_recv(SVCXPRT *xprt, struct rpc_msg *msg)
484 struct cf_conn *cd = (struct cf_conn *) xprt->xp_p1;
491 * If we have an mbuf chain in cd->mpending, try to parse a
492 * record from it, leaving the result in cd->mreq. If we don't
493 * have a complete record, leave the partial result in
494 * cd->mreq and try to read more from the socket.
498 * If cd->resid is non-zero, we have part of the
499 * record already, otherwise we are expecting a record
504 * See if there is enough data buffered to
505 * make up a record marker. Make sure we can
506 * handle the case where the record marker is
507 * split across more than one mbuf.
513 while (n < sizeof(uint32_t) && m) {
517 if (n < sizeof(uint32_t))
519 cd->mpending = m_pullup(cd->mpending, sizeof(uint32_t));
520 memcpy(&header, mtod(cd->mpending, uint32_t *),
522 header = ntohl(header);
523 cd->eor = (header & 0x80000000) != 0;
524 cd->resid = header & 0x7fffffff;
525 m_adj(cd->mpending, sizeof(uint32_t));
529 * Start pulling off mbufs from cd->mpending
530 * until we either have a complete record or
531 * we run out of data. We use m_split to pull
532 * data - it will pull as much as possible and
533 * split the last mbuf if necessary.
535 while (cd->mpending && cd->resid) {
537 cd->mpending = m_split(cd->mpending, cd->resid,
540 m_last(cd->mreq)->m_next = m;
544 cd->resid -= m->m_len;
550 * If cd->resid is zero now, we have managed to
551 * receive a record fragment from the stream. Check
552 * for the end-of-record mark to see if we need more.
554 if (cd->resid == 0) {
559 * Success - we have a complete record in
562 xdrmbuf_create(&xprt->xp_xdrreq, cd->mreq, XDR_DECODE);
564 if (! xdr_callmsg(&xprt->xp_xdrreq, msg)) {
565 XDR_DESTROY(&xprt->xp_xdrreq);
568 xprt->xp_xid = msg->rm_xid;
576 * The socket upcall calls xprt_active() which will eventually
577 * cause the server to call us here. We attempt to
578 * read as much as possible from the socket and put
579 * the result in cd->mpending. If the read fails,
580 * we have drained both cd->mpending and the socket so
581 * we can call xprt_inactive().
583 * The lock protects us in the case where a new packet arrives
584 * on the socket after our call to soreceive fails with
585 * EWOULDBLOCK - the call to xprt_active() in the upcall will
586 * happen only after our call to xprt_inactive() which ensures
587 * that we will remain active. It might be possible to use
588 * SOCKBUF_LOCK for this - its not clear to me what locks are
589 * held during the upcall.
591 mtx_lock(&xprt->xp_lock);
593 uio.uio_resid = 1000000000;
594 uio.uio_td = curthread;
596 rcvflag = MSG_DONTWAIT;
597 error = soreceive(xprt->xp_socket, NULL, &uio, &m, NULL,
600 if (error == EWOULDBLOCK) {
602 mtx_unlock(&xprt->xp_lock);
607 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
608 xprt->xp_socket->so_upcallarg = NULL;
609 xprt->xp_socket->so_upcall = NULL;
610 xprt->xp_socket->so_rcv.sb_flags &= ~SB_UPCALL;
611 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
613 cd->strm_stat = XPRT_DIED;
614 mtx_unlock(&xprt->xp_lock);
620 * EOF - the other end has closed the socket.
622 cd->strm_stat = XPRT_DIED;
623 mtx_unlock(&xprt->xp_lock);
628 m_last(cd->mpending)->m_next = m;
632 mtx_unlock(&xprt->xp_lock);
637 svc_vc_getargs(SVCXPRT *xprt, xdrproc_t xdr_args, void *args_ptr)
640 return (xdr_args(&xprt->xp_xdrreq, args_ptr));
644 svc_vc_freeargs(SVCXPRT *xprt, xdrproc_t xdr_args, void *args_ptr)
649 * Free the request mbuf here - this allows us to handle
650 * protocols where not all requests have replies
651 * (i.e. NLM). Note that xdrmbuf_destroy handles being called
652 * twice correctly - the mbuf will only be freed once.
654 XDR_DESTROY(&xprt->xp_xdrreq);
656 xdrs.x_op = XDR_FREE;
657 return (xdr_args(&xdrs, args_ptr));
661 svc_vc_reply(SVCXPRT *xprt, struct rpc_msg *msg)
668 * Leave space for record mark.
670 MGETHDR(mrep, M_WAIT, MT_DATA);
671 MCLGET(mrep, M_WAIT);
673 mrep->m_data += sizeof(uint32_t);
675 xdrmbuf_create(&xprt->xp_xdrrep, mrep, XDR_ENCODE);
676 msg->rm_xid = xprt->xp_xid;
677 if (xdr_replymsg(&xprt->xp_xdrrep, msg)) {
681 * Prepend a record marker containing the reply length.
683 M_PREPEND(mrep, sizeof(uint32_t), M_WAIT);
684 *mtod(mrep, uint32_t *) =
685 htonl(0x80000000 | (mrep->m_pkthdr.len
686 - sizeof(uint32_t)));
687 error = sosend(xprt->xp_socket, NULL, NULL, mrep, NULL,
697 * This frees the request mbuf chain as well. The reply mbuf
698 * chain was consumed by sosend.
700 XDR_DESTROY(&xprt->xp_xdrreq);
701 XDR_DESTROY(&xprt->xp_xdrrep);
715 svc_vc_soupcall(struct socket *so, void *arg, int waitflag)
717 SVCXPRT *xprt = (SVCXPRT *) arg;
719 mtx_lock(&xprt->xp_lock);
721 mtx_unlock(&xprt->xp_lock);
726 * Get the effective UID of the sending process. Used by rpcbind, keyserv
727 * and rpc.yppasswdd on AF_LOCAL.
730 __rpc_get_local_uid(SVCXPRT *transp, uid_t *uid) {
736 sock = transp->xp_fd;
737 sa = (struct sockaddr *)transp->xp_rtaddr.buf;
738 if (sa->sa_family == AF_LOCAL) {
739 ret = getpeereid(sock, &euid, &egid);