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>
54 #include <sys/protosw.h>
55 #include <sys/queue.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
59 #include <sys/systm.h>
61 #include <netinet/tcp.h>
65 #include <rpc/rpc_com.h>
67 #include <security/mac/mac_framework.h>
69 static bool_t svc_vc_rendezvous_recv(SVCXPRT *, struct rpc_msg *,
70 struct sockaddr **, struct mbuf **);
71 static enum xprt_stat svc_vc_rendezvous_stat(SVCXPRT *);
72 static void svc_vc_rendezvous_destroy(SVCXPRT *);
73 static bool_t svc_vc_null(void);
74 static void svc_vc_destroy(SVCXPRT *);
75 static enum xprt_stat svc_vc_stat(SVCXPRT *);
76 static bool_t svc_vc_recv(SVCXPRT *, struct rpc_msg *,
77 struct sockaddr **, struct mbuf **);
78 static bool_t svc_vc_reply(SVCXPRT *, struct rpc_msg *,
79 struct sockaddr *, struct mbuf *);
80 static bool_t svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in);
81 static bool_t svc_vc_rendezvous_control (SVCXPRT *xprt, const u_int rq,
83 static SVCXPRT *svc_vc_create_conn(SVCPOOL *pool, struct socket *so,
84 struct sockaddr *raddr);
85 static int svc_vc_accept(struct socket *head, struct socket **sop);
86 static int svc_vc_soupcall(struct socket *so, void *arg, int waitflag);
88 static struct xp_ops svc_vc_rendezvous_ops = {
89 .xp_recv = svc_vc_rendezvous_recv,
90 .xp_stat = svc_vc_rendezvous_stat,
91 .xp_reply = (bool_t (*)(SVCXPRT *, struct rpc_msg *,
92 struct sockaddr *, struct mbuf *))svc_vc_null,
93 .xp_destroy = svc_vc_rendezvous_destroy,
94 .xp_control = svc_vc_rendezvous_control
97 static struct xp_ops svc_vc_ops = {
98 .xp_recv = svc_vc_recv,
99 .xp_stat = svc_vc_stat,
100 .xp_reply = svc_vc_reply,
101 .xp_destroy = svc_vc_destroy,
102 .xp_control = svc_vc_control
105 struct cf_conn { /* kept in xprt->xp_p1 for actual connection */
106 enum xprt_stat strm_stat;
107 struct mbuf *mpending; /* unparsed data read from the socket */
108 struct mbuf *mreq; /* current record being built from mpending */
109 uint32_t resid; /* number of bytes needed for fragment */
110 bool_t eor; /* reading last fragment of current record */
115 * xprt = svc_vc_create(sock, send_buf_size, recv_buf_size);
117 * Creates, registers, and returns a (rpc) tcp based transporter.
118 * Once *xprt is initialized, it is registered as a transporter
119 * see (svc.h, xprt_register). This routine returns
120 * a NULL if a problem occurred.
122 * The filedescriptor passed in is expected to refer to a bound, but
123 * not yet connected socket.
125 * Since streams do buffered io similar to stdio, the caller can specify
126 * how big the send and receive buffers are via the second and third parms;
127 * 0 => use the system default.
130 svc_vc_create(SVCPOOL *pool, struct socket *so, size_t sendsize,
137 if (so->so_state & SS_ISCONNECTED) {
138 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
141 xprt = svc_vc_create_conn(pool, so, sa);
146 xprt = svc_xprt_alloc();
147 sx_init(&xprt->xp_lock, "xprt->xp_lock");
148 xprt->xp_pool = pool;
149 xprt->xp_socket = so;
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 memcpy(&xprt->xp_ltaddr, sa, sa->sa_len);
163 solisten(so, SOMAXCONN, curthread);
165 SOCKBUF_LOCK(&so->so_rcv);
166 xprt->xp_upcallset = 1;
167 soupcall_set(so, SO_RCV, svc_vc_soupcall, xprt);
168 SOCKBUF_UNLOCK(&so->so_rcv);
171 cleanup_svc_vc_create:
178 * Create a new transport for a socket optained via soaccept().
181 svc_vc_create_conn(SVCPOOL *pool, struct socket *so, struct sockaddr *raddr)
183 SVCXPRT *xprt = NULL;
184 struct cf_conn *cd = NULL;
185 struct sockaddr* sa = NULL;
190 bzero(&opt, sizeof(struct sockopt));
191 opt.sopt_dir = SOPT_SET;
192 opt.sopt_level = SOL_SOCKET;
193 opt.sopt_name = SO_KEEPALIVE;
195 opt.sopt_valsize = sizeof(one);
196 error = sosetopt(so, &opt);
200 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
201 bzero(&opt, sizeof(struct sockopt));
202 opt.sopt_dir = SOPT_SET;
203 opt.sopt_level = IPPROTO_TCP;
204 opt.sopt_name = TCP_NODELAY;
206 opt.sopt_valsize = sizeof(one);
207 error = sosetopt(so, &opt);
212 cd = mem_alloc(sizeof(*cd));
213 cd->strm_stat = XPRT_IDLE;
215 xprt = svc_xprt_alloc();
216 sx_init(&xprt->xp_lock, "xprt->xp_lock");
217 xprt->xp_pool = pool;
218 xprt->xp_socket = so;
221 xprt->xp_ops = &svc_vc_ops;
224 * See http://www.connectathon.org/talks96/nfstcp.pdf - client
225 * has a 5 minute timer, server has a 6 minute timer.
227 xprt->xp_idletimeout = 6 * 60;
229 memcpy(&xprt->xp_rtaddr, raddr, raddr->sa_len);
231 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
233 goto cleanup_svc_vc_create;
235 memcpy(&xprt->xp_ltaddr, sa, sa->sa_len);
240 SOCKBUF_LOCK(&so->so_rcv);
241 xprt->xp_upcallset = 1;
242 soupcall_set(so, SO_RCV, svc_vc_soupcall, xprt);
243 SOCKBUF_UNLOCK(&so->so_rcv);
246 * Throw the transport into the active list in case it already
247 * has some data buffered.
249 sx_xlock(&xprt->xp_lock);
251 sx_xunlock(&xprt->xp_lock);
254 cleanup_svc_vc_create:
256 mem_free(xprt, sizeof(*xprt));
259 mem_free(cd, sizeof(*cd));
264 * This does all of the accept except the final call to soaccept. The
265 * caller will call soaccept after dropping its locks (soaccept may
269 svc_vc_accept(struct socket *head, struct socket **sop)
274 if ((head->so_options & SO_ACCEPTCONN) == 0) {
279 error = mac_socket_check_accept(curthread->td_ucred, head);
284 if (TAILQ_EMPTY(&head->so_comp)) {
289 so = TAILQ_FIRST(&head->so_comp);
290 KASSERT(!(so->so_qstate & SQ_INCOMP), ("svc_vc_accept: so SQ_INCOMP"));
291 KASSERT(so->so_qstate & SQ_COMP, ("svc_vc_accept: so not SQ_COMP"));
294 * Before changing the flags on the socket, we have to bump the
295 * reference count. Otherwise, if the protocol calls sofree(),
296 * the socket will be released due to a zero refcount.
297 * XXX might not need soref() since this is simpler than kern_accept.
299 SOCK_LOCK(so); /* soref() and so_state update */
300 soref(so); /* file descriptor reference */
302 TAILQ_REMOVE(&head->so_comp, so, so_list);
304 so->so_state |= (head->so_state & SS_NBIO);
305 so->so_qstate &= ~SQ_COMP;
313 /* connection has been removed from the listen queue */
314 KNOTE_UNLOCKED(&head->so_rcv.sb_sel.si_note, 0);
321 svc_vc_rendezvous_recv(SVCXPRT *xprt, struct rpc_msg *msg,
322 struct sockaddr **addrp, struct mbuf **mp)
324 struct socket *so = NULL;
325 struct sockaddr *sa = NULL;
330 * The socket upcall calls xprt_active() which will eventually
331 * cause the server to call us here. We attempt to accept a
332 * connection from the socket and turn it into a new
333 * transport. If the accept fails, we have drained all pending
334 * connections so we call xprt_inactive().
336 sx_xlock(&xprt->xp_lock);
338 error = svc_vc_accept(xprt->xp_socket, &so);
340 if (error == EWOULDBLOCK) {
342 * We must re-test for new connections after taking
343 * the lock to protect us in the case where a new
344 * connection arrives after our call to accept fails
345 * with EWOULDBLOCK. The pool lock protects us from
346 * racing the upcall after our TAILQ_EMPTY() call
350 mtx_lock(&xprt->xp_pool->sp_lock);
351 if (TAILQ_EMPTY(&xprt->xp_socket->so_comp))
352 xprt_inactive_locked(xprt);
353 mtx_unlock(&xprt->xp_pool->sp_lock);
355 sx_xunlock(&xprt->xp_lock);
360 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
361 if (xprt->xp_upcallset) {
362 xprt->xp_upcallset = 0;
363 soupcall_clear(xprt->xp_socket, SO_RCV);
365 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
367 sx_xunlock(&xprt->xp_lock);
371 sx_xunlock(&xprt->xp_lock);
374 error = soaccept(so, &sa);
378 * XXX not sure if I need to call sofree or soclose here.
386 * svc_vc_create_conn will call xprt_register - we don't need
387 * to do anything with the new connection except derefence it.
389 new_xprt = svc_vc_create_conn(xprt->xp_pool, so, sa);
393 SVC_RELEASE(new_xprt);
398 return (FALSE); /* there is never an rpc msg to be processed */
402 static enum xprt_stat
403 svc_vc_rendezvous_stat(SVCXPRT *xprt)
410 svc_vc_destroy_common(SVCXPRT *xprt)
412 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
413 if (xprt->xp_upcallset) {
414 xprt->xp_upcallset = 0;
415 soupcall_clear(xprt->xp_socket, SO_RCV);
417 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
419 sx_destroy(&xprt->xp_lock);
421 (void)soclose(xprt->xp_socket);
424 (void) mem_free(xprt->xp_netid, strlen(xprt->xp_netid) + 1);
429 svc_vc_rendezvous_destroy(SVCXPRT *xprt)
432 svc_vc_destroy_common(xprt);
436 svc_vc_destroy(SVCXPRT *xprt)
438 struct cf_conn *cd = (struct cf_conn *)xprt->xp_p1;
440 svc_vc_destroy_common(xprt);
445 m_freem(cd->mpending);
446 mem_free(cd, sizeof(*cd));
451 svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in)
457 svc_vc_rendezvous_control(SVCXPRT *xprt, const u_int rq, void *in)
463 static enum xprt_stat
464 svc_vc_stat(SVCXPRT *xprt)
470 cd = (struct cf_conn *)(xprt->xp_p1);
472 if (cd->strm_stat == XPRT_DIED)
476 * Return XPRT_MOREREQS if we have buffered data and we are
477 * mid-record or if we have enough data for a record
478 * marker. Since this is only a hint, we read mpending and
479 * resid outside the lock. We do need to take the lock if we
480 * have to traverse the mbuf chain.
484 return (XPRT_MOREREQS);
486 sx_xlock(&xprt->xp_lock);
488 while (m && n < sizeof(uint32_t)) {
492 sx_xunlock(&xprt->xp_lock);
493 if (n >= sizeof(uint32_t))
494 return (XPRT_MOREREQS);
497 if (soreadable(xprt->xp_socket))
498 return (XPRT_MOREREQS);
504 svc_vc_recv(SVCXPRT *xprt, struct rpc_msg *msg,
505 struct sockaddr **addrp, struct mbuf **mp)
507 struct cf_conn *cd = (struct cf_conn *) xprt->xp_p1;
514 * Serialise access to the socket and our own record parsing
517 sx_xlock(&xprt->xp_lock);
521 * If we have an mbuf chain in cd->mpending, try to parse a
522 * record from it, leaving the result in cd->mreq. If we don't
523 * have a complete record, leave the partial result in
524 * cd->mreq and try to read more from the socket.
528 * If cd->resid is non-zero, we have part of the
529 * record already, otherwise we are expecting a record
534 * See if there is enough data buffered to
535 * make up a record marker. Make sure we can
536 * handle the case where the record marker is
537 * split across more than one mbuf.
543 while (n < sizeof(uint32_t) && m) {
547 if (n < sizeof(uint32_t))
549 if (cd->mpending->m_len < sizeof(uint32_t))
550 cd->mpending = m_pullup(cd->mpending,
552 memcpy(&header, mtod(cd->mpending, uint32_t *),
554 header = ntohl(header);
555 cd->eor = (header & 0x80000000) != 0;
556 cd->resid = header & 0x7fffffff;
557 m_adj(cd->mpending, sizeof(uint32_t));
561 * Start pulling off mbufs from cd->mpending
562 * until we either have a complete record or
563 * we run out of data. We use m_split to pull
564 * data - it will pull as much as possible and
565 * split the last mbuf if necessary.
567 while (cd->mpending && cd->resid) {
569 if (cd->mpending->m_next
570 || cd->mpending->m_len > cd->resid)
571 cd->mpending = m_split(cd->mpending,
576 m_last(cd->mreq)->m_next = m;
580 cd->resid -= m->m_len;
586 * If cd->resid is zero now, we have managed to
587 * receive a record fragment from the stream. Check
588 * for the end-of-record mark to see if we need more.
590 if (cd->resid == 0) {
595 * Success - we have a complete record in
598 xdrmbuf_create(&xdrs, cd->mreq, XDR_DECODE);
600 sx_xunlock(&xprt->xp_lock);
602 if (! xdr_callmsg(&xdrs, msg)) {
608 *mp = xdrmbuf_getall(&xdrs);
617 * The socket upcall calls xprt_active() which will eventually
618 * cause the server to call us here. We attempt to
619 * read as much as possible from the socket and put
620 * the result in cd->mpending. If the read fails,
621 * we have drained both cd->mpending and the socket so
622 * we can call xprt_inactive().
624 uio.uio_resid = 1000000000;
625 uio.uio_td = curthread;
627 rcvflag = MSG_DONTWAIT;
628 error = soreceive(xprt->xp_socket, NULL, &uio, &m, NULL,
631 if (error == EWOULDBLOCK) {
633 * We must re-test for readability after
634 * taking the lock to protect us in the case
635 * where a new packet arrives on the socket
636 * after our call to soreceive fails with
637 * EWOULDBLOCK. The pool lock protects us from
638 * racing the upcall after our soreadable()
639 * call returns false.
641 mtx_lock(&xprt->xp_pool->sp_lock);
642 if (!soreadable(xprt->xp_socket))
643 xprt_inactive_locked(xprt);
644 mtx_unlock(&xprt->xp_pool->sp_lock);
645 sx_xunlock(&xprt->xp_lock);
650 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
651 if (xprt->xp_upcallset) {
652 xprt->xp_upcallset = 0;
653 soupcall_clear(xprt->xp_socket, SO_RCV);
655 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
657 cd->strm_stat = XPRT_DIED;
658 sx_xunlock(&xprt->xp_lock);
664 * EOF - the other end has closed the socket.
667 cd->strm_stat = XPRT_DIED;
668 sx_xunlock(&xprt->xp_lock);
673 m_last(cd->mpending)->m_next = m;
680 svc_vc_reply(SVCXPRT *xprt, struct rpc_msg *msg,
681 struct sockaddr *addr, struct mbuf *m)
689 * Leave space for record mark.
691 MGETHDR(mrep, M_WAIT, MT_DATA);
693 mrep->m_data += sizeof(uint32_t);
695 xdrmbuf_create(&xdrs, mrep, XDR_ENCODE);
697 if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
698 msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
699 if (!xdr_replymsg(&xdrs, msg))
702 xdrmbuf_append(&xdrs, m);
704 stat = xdr_replymsg(&xdrs, msg);
711 * Prepend a record marker containing the reply length.
713 M_PREPEND(mrep, sizeof(uint32_t), M_WAIT);
714 *mtod(mrep, uint32_t *) =
715 htonl(0x80000000 | (mrep->m_pkthdr.len
716 - sizeof(uint32_t)));
717 error = sosend(xprt->xp_socket, NULL, NULL, mrep, NULL,
740 svc_vc_soupcall(struct socket *so, void *arg, int waitflag)
742 SVCXPRT *xprt = (SVCXPRT *) arg;
750 * Get the effective UID of the sending process. Used by rpcbind, keyserv
751 * and rpc.yppasswdd on AF_LOCAL.
754 __rpc_get_local_uid(SVCXPRT *transp, uid_t *uid) {
760 sock = transp->xp_fd;
761 sa = (struct sockaddr *)transp->xp_rtaddr;
762 if (sa->sa_family == AF_LOCAL) {
763 ret = getpeereid(sock, &euid, &egid);