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>
64 #include <netinet/tcp.h>
68 #include <rpc/rpc_com.h>
70 #include <security/mac/mac_framework.h>
72 static bool_t svc_vc_rendezvous_recv(SVCXPRT *, struct rpc_msg *,
73 struct sockaddr **, struct mbuf **);
74 static enum xprt_stat svc_vc_rendezvous_stat(SVCXPRT *);
75 static void svc_vc_rendezvous_destroy(SVCXPRT *);
76 static bool_t svc_vc_null(void);
77 static void svc_vc_destroy(SVCXPRT *);
78 static enum xprt_stat svc_vc_stat(SVCXPRT *);
79 static bool_t svc_vc_recv(SVCXPRT *, struct rpc_msg *,
80 struct sockaddr **, struct mbuf **);
81 static bool_t svc_vc_reply(SVCXPRT *, struct rpc_msg *,
82 struct sockaddr *, struct mbuf *);
83 static bool_t svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in);
84 static bool_t svc_vc_rendezvous_control (SVCXPRT *xprt, const u_int rq,
86 static SVCXPRT *svc_vc_create_conn(SVCPOOL *pool, struct socket *so,
87 struct sockaddr *raddr);
88 static int svc_vc_accept(struct socket *head, struct socket **sop);
89 static int svc_vc_soupcall(struct socket *so, void *arg, int waitflag);
91 static struct xp_ops svc_vc_rendezvous_ops = {
92 .xp_recv = svc_vc_rendezvous_recv,
93 .xp_stat = svc_vc_rendezvous_stat,
94 .xp_reply = (bool_t (*)(SVCXPRT *, struct rpc_msg *,
95 struct sockaddr *, struct mbuf *))svc_vc_null,
96 .xp_destroy = svc_vc_rendezvous_destroy,
97 .xp_control = svc_vc_rendezvous_control
100 static struct xp_ops svc_vc_ops = {
101 .xp_recv = svc_vc_recv,
102 .xp_stat = svc_vc_stat,
103 .xp_reply = svc_vc_reply,
104 .xp_destroy = svc_vc_destroy,
105 .xp_control = svc_vc_control
108 struct cf_conn { /* kept in xprt->xp_p1 for actual connection */
109 enum xprt_stat strm_stat;
110 struct mbuf *mpending; /* unparsed data read from the socket */
111 struct mbuf *mreq; /* current record being built from mpending */
112 uint32_t resid; /* number of bytes needed for fragment */
113 bool_t eor; /* reading last fragment of current record */
118 * xprt = svc_vc_create(sock, send_buf_size, recv_buf_size);
120 * Creates, registers, and returns a (rpc) tcp based transporter.
121 * Once *xprt is initialized, it is registered as a transporter
122 * see (svc.h, xprt_register). This routine returns
123 * a NULL if a problem occurred.
125 * The filedescriptor passed in is expected to refer to a bound, but
126 * not yet connected socket.
128 * Since streams do buffered io similar to stdio, the caller can specify
129 * how big the send and receive buffers are via the second and third parms;
130 * 0 => use the system default.
133 svc_vc_create(SVCPOOL *pool, struct socket *so, size_t sendsize,
140 if (so->so_state & SS_ISCONNECTED) {
141 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
144 xprt = svc_vc_create_conn(pool, so, sa);
149 xprt = svc_xprt_alloc();
150 sx_init(&xprt->xp_lock, "xprt->xp_lock");
151 xprt->xp_pool = pool;
152 xprt->xp_socket = so;
155 xprt->xp_ops = &svc_vc_rendezvous_ops;
157 CURVNET_SET(so->so_vnet);
158 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
161 goto cleanup_svc_vc_create;
164 memcpy(&xprt->xp_ltaddr, sa, sa->sa_len);
169 solisten(so, SOMAXCONN, curthread);
172 SOCKBUF_LOCK(&so->so_rcv);
173 xprt->xp_upcallset = 1;
174 soupcall_set(so, SO_RCV, svc_vc_soupcall, xprt);
175 SOCKBUF_UNLOCK(&so->so_rcv);
178 cleanup_svc_vc_create:
185 * Create a new transport for a socket optained via soaccept().
188 svc_vc_create_conn(SVCPOOL *pool, struct socket *so, struct sockaddr *raddr)
190 SVCXPRT *xprt = NULL;
191 struct cf_conn *cd = NULL;
192 struct sockaddr* sa = NULL;
197 bzero(&opt, sizeof(struct sockopt));
198 opt.sopt_dir = SOPT_SET;
199 opt.sopt_level = SOL_SOCKET;
200 opt.sopt_name = SO_KEEPALIVE;
202 opt.sopt_valsize = sizeof(one);
203 CURVNET_SET(so->so_vnet);
204 error = sosetopt(so, &opt);
210 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
211 bzero(&opt, sizeof(struct sockopt));
212 opt.sopt_dir = SOPT_SET;
213 opt.sopt_level = IPPROTO_TCP;
214 opt.sopt_name = TCP_NODELAY;
216 opt.sopt_valsize = sizeof(one);
217 error = sosetopt(so, &opt);
225 cd = mem_alloc(sizeof(*cd));
226 cd->strm_stat = XPRT_IDLE;
228 xprt = svc_xprt_alloc();
229 sx_init(&xprt->xp_lock, "xprt->xp_lock");
230 xprt->xp_pool = pool;
231 xprt->xp_socket = so;
234 xprt->xp_ops = &svc_vc_ops;
237 * See http://www.connectathon.org/talks96/nfstcp.pdf - client
238 * has a 5 minute timer, server has a 6 minute timer.
240 xprt->xp_idletimeout = 6 * 60;
242 memcpy(&xprt->xp_rtaddr, raddr, raddr->sa_len);
244 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
246 goto cleanup_svc_vc_create;
248 memcpy(&xprt->xp_ltaddr, sa, sa->sa_len);
253 SOCKBUF_LOCK(&so->so_rcv);
254 xprt->xp_upcallset = 1;
255 soupcall_set(so, SO_RCV, svc_vc_soupcall, xprt);
256 SOCKBUF_UNLOCK(&so->so_rcv);
259 * Throw the transport into the active list in case it already
260 * has some data buffered.
262 sx_xlock(&xprt->xp_lock);
264 sx_xunlock(&xprt->xp_lock);
267 cleanup_svc_vc_create:
269 mem_free(xprt, sizeof(*xprt));
272 mem_free(cd, sizeof(*cd));
277 * This does all of the accept except the final call to soaccept. The
278 * caller will call soaccept after dropping its locks (soaccept may
282 svc_vc_accept(struct socket *head, struct socket **sop)
287 if ((head->so_options & SO_ACCEPTCONN) == 0) {
292 error = mac_socket_check_accept(curthread->td_ucred, head);
297 if (TAILQ_EMPTY(&head->so_comp)) {
302 so = TAILQ_FIRST(&head->so_comp);
303 KASSERT(!(so->so_qstate & SQ_INCOMP), ("svc_vc_accept: so SQ_INCOMP"));
304 KASSERT(so->so_qstate & SQ_COMP, ("svc_vc_accept: so not SQ_COMP"));
307 * Before changing the flags on the socket, we have to bump the
308 * reference count. Otherwise, if the protocol calls sofree(),
309 * the socket will be released due to a zero refcount.
310 * XXX might not need soref() since this is simpler than kern_accept.
312 SOCK_LOCK(so); /* soref() and so_state update */
313 soref(so); /* file descriptor reference */
315 TAILQ_REMOVE(&head->so_comp, so, so_list);
317 so->so_state |= (head->so_state & SS_NBIO);
318 so->so_qstate &= ~SQ_COMP;
326 /* connection has been removed from the listen queue */
327 KNOTE_UNLOCKED(&head->so_rcv.sb_sel.si_note, 0);
334 svc_vc_rendezvous_recv(SVCXPRT *xprt, struct rpc_msg *msg,
335 struct sockaddr **addrp, struct mbuf **mp)
337 struct socket *so = NULL;
338 struct sockaddr *sa = NULL;
343 * The socket upcall calls xprt_active() which will eventually
344 * cause the server to call us here. We attempt to accept a
345 * connection from the socket and turn it into a new
346 * transport. If the accept fails, we have drained all pending
347 * connections so we call xprt_inactive().
349 sx_xlock(&xprt->xp_lock);
351 error = svc_vc_accept(xprt->xp_socket, &so);
353 if (error == EWOULDBLOCK) {
355 * We must re-test for new connections after taking
356 * the lock to protect us in the case where a new
357 * connection arrives after our call to accept fails
358 * with EWOULDBLOCK. The pool lock protects us from
359 * racing the upcall after our TAILQ_EMPTY() call
363 mtx_lock(&xprt->xp_pool->sp_lock);
364 if (TAILQ_EMPTY(&xprt->xp_socket->so_comp))
365 xprt_inactive_locked(xprt);
366 mtx_unlock(&xprt->xp_pool->sp_lock);
368 sx_xunlock(&xprt->xp_lock);
373 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
374 if (xprt->xp_upcallset) {
375 xprt->xp_upcallset = 0;
376 soupcall_clear(xprt->xp_socket, SO_RCV);
378 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
380 sx_xunlock(&xprt->xp_lock);
384 sx_xunlock(&xprt->xp_lock);
387 error = soaccept(so, &sa);
391 * XXX not sure if I need to call sofree or soclose here.
399 * svc_vc_create_conn will call xprt_register - we don't need
400 * to do anything with the new connection except derefence it.
402 new_xprt = svc_vc_create_conn(xprt->xp_pool, so, sa);
406 SVC_RELEASE(new_xprt);
411 return (FALSE); /* there is never an rpc msg to be processed */
415 static enum xprt_stat
416 svc_vc_rendezvous_stat(SVCXPRT *xprt)
423 svc_vc_destroy_common(SVCXPRT *xprt)
425 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
426 if (xprt->xp_upcallset) {
427 xprt->xp_upcallset = 0;
428 soupcall_clear(xprt->xp_socket, SO_RCV);
430 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
432 sx_destroy(&xprt->xp_lock);
434 (void)soclose(xprt->xp_socket);
437 (void) mem_free(xprt->xp_netid, strlen(xprt->xp_netid) + 1);
442 svc_vc_rendezvous_destroy(SVCXPRT *xprt)
445 svc_vc_destroy_common(xprt);
449 svc_vc_destroy(SVCXPRT *xprt)
451 struct cf_conn *cd = (struct cf_conn *)xprt->xp_p1;
453 svc_vc_destroy_common(xprt);
458 m_freem(cd->mpending);
459 mem_free(cd, sizeof(*cd));
464 svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in)
470 svc_vc_rendezvous_control(SVCXPRT *xprt, const u_int rq, void *in)
476 static enum xprt_stat
477 svc_vc_stat(SVCXPRT *xprt)
483 cd = (struct cf_conn *)(xprt->xp_p1);
485 if (cd->strm_stat == XPRT_DIED)
489 * Return XPRT_MOREREQS if we have buffered data and we are
490 * mid-record or if we have enough data for a record
491 * marker. Since this is only a hint, we read mpending and
492 * resid outside the lock. We do need to take the lock if we
493 * have to traverse the mbuf chain.
497 return (XPRT_MOREREQS);
499 sx_xlock(&xprt->xp_lock);
501 while (m && n < sizeof(uint32_t)) {
505 sx_xunlock(&xprt->xp_lock);
506 if (n >= sizeof(uint32_t))
507 return (XPRT_MOREREQS);
510 if (soreadable(xprt->xp_socket))
511 return (XPRT_MOREREQS);
517 svc_vc_recv(SVCXPRT *xprt, struct rpc_msg *msg,
518 struct sockaddr **addrp, struct mbuf **mp)
520 struct cf_conn *cd = (struct cf_conn *) xprt->xp_p1;
527 * Serialise access to the socket and our own record parsing
530 sx_xlock(&xprt->xp_lock);
534 * If we have an mbuf chain in cd->mpending, try to parse a
535 * record from it, leaving the result in cd->mreq. If we don't
536 * have a complete record, leave the partial result in
537 * cd->mreq and try to read more from the socket.
541 * If cd->resid is non-zero, we have part of the
542 * record already, otherwise we are expecting a record
547 * See if there is enough data buffered to
548 * make up a record marker. Make sure we can
549 * handle the case where the record marker is
550 * split across more than one mbuf.
556 while (n < sizeof(uint32_t) && m) {
560 if (n < sizeof(uint32_t))
562 if (cd->mpending->m_len < sizeof(uint32_t))
563 cd->mpending = m_pullup(cd->mpending,
565 memcpy(&header, mtod(cd->mpending, uint32_t *),
567 header = ntohl(header);
568 cd->eor = (header & 0x80000000) != 0;
569 cd->resid = header & 0x7fffffff;
570 m_adj(cd->mpending, sizeof(uint32_t));
574 * Start pulling off mbufs from cd->mpending
575 * until we either have a complete record or
576 * we run out of data. We use m_split to pull
577 * data - it will pull as much as possible and
578 * split the last mbuf if necessary.
580 while (cd->mpending && cd->resid) {
582 if (cd->mpending->m_next
583 || cd->mpending->m_len > cd->resid)
584 cd->mpending = m_split(cd->mpending,
589 m_last(cd->mreq)->m_next = m;
593 cd->resid -= m->m_len;
599 * If cd->resid is zero now, we have managed to
600 * receive a record fragment from the stream. Check
601 * for the end-of-record mark to see if we need more.
603 if (cd->resid == 0) {
608 * Success - we have a complete record in
611 xdrmbuf_create(&xdrs, cd->mreq, XDR_DECODE);
613 sx_xunlock(&xprt->xp_lock);
615 if (! xdr_callmsg(&xdrs, msg)) {
621 *mp = xdrmbuf_getall(&xdrs);
630 * The socket upcall calls xprt_active() which will eventually
631 * cause the server to call us here. We attempt to
632 * read as much as possible from the socket and put
633 * the result in cd->mpending. If the read fails,
634 * we have drained both cd->mpending and the socket so
635 * we can call xprt_inactive().
637 uio.uio_resid = 1000000000;
638 uio.uio_td = curthread;
640 rcvflag = MSG_DONTWAIT;
641 CURVNET_SET(xprt->xp_socket->so_vnet);
642 error = soreceive(xprt->xp_socket, NULL, &uio, &m, NULL,
646 if (error == EWOULDBLOCK) {
648 * We must re-test for readability after
649 * taking the lock to protect us in the case
650 * where a new packet arrives on the socket
651 * after our call to soreceive fails with
652 * EWOULDBLOCK. The pool lock protects us from
653 * racing the upcall after our soreadable()
654 * call returns false.
656 mtx_lock(&xprt->xp_pool->sp_lock);
657 if (!soreadable(xprt->xp_socket))
658 xprt_inactive_locked(xprt);
659 mtx_unlock(&xprt->xp_pool->sp_lock);
660 sx_xunlock(&xprt->xp_lock);
665 SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
666 if (xprt->xp_upcallset) {
667 xprt->xp_upcallset = 0;
668 soupcall_clear(xprt->xp_socket, SO_RCV);
670 SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
672 cd->strm_stat = XPRT_DIED;
673 sx_xunlock(&xprt->xp_lock);
679 * EOF - the other end has closed the socket.
682 cd->strm_stat = XPRT_DIED;
683 sx_xunlock(&xprt->xp_lock);
688 m_last(cd->mpending)->m_next = m;
695 svc_vc_reply(SVCXPRT *xprt, struct rpc_msg *msg,
696 struct sockaddr *addr, struct mbuf *m)
704 * Leave space for record mark.
706 MGETHDR(mrep, M_WAIT, MT_DATA);
708 mrep->m_data += sizeof(uint32_t);
710 xdrmbuf_create(&xdrs, mrep, XDR_ENCODE);
712 if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
713 msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
714 if (!xdr_replymsg(&xdrs, msg))
717 xdrmbuf_append(&xdrs, m);
719 stat = xdr_replymsg(&xdrs, msg);
726 * Prepend a record marker containing the reply length.
728 M_PREPEND(mrep, sizeof(uint32_t), M_WAIT);
729 *mtod(mrep, uint32_t *) =
730 htonl(0x80000000 | (mrep->m_pkthdr.len
731 - sizeof(uint32_t)));
732 error = sosend(xprt->xp_socket, NULL, NULL, mrep, NULL,
755 svc_vc_soupcall(struct socket *so, void *arg, int waitflag)
757 SVCXPRT *xprt = (SVCXPRT *) arg;
765 * Get the effective UID of the sending process. Used by rpcbind, keyserv
766 * and rpc.yppasswdd on AF_LOCAL.
769 __rpc_get_local_uid(SVCXPRT *transp, uid_t *uid) {
775 sock = transp->xp_fd;
776 sa = (struct sockaddr *)transp->xp_rtaddr;
777 if (sa->sa_family == AF_LOCAL) {
778 ret = getpeereid(sock, &euid, &egid);