2 * Copyright (c) 1989, 1991, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
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 * 4. 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 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
39 * Socket operations for use by nfs
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
53 #include <sys/protosw.h>
54 #include <sys/refcount.h>
55 #include <sys/signalvar.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/sysctl.h>
59 #include <sys/syslog.h>
60 #include <sys/vnode.h>
62 #include <netinet/in.h>
63 #include <netinet/tcp.h>
65 #include <nfs/rpcv2.h>
66 #include <nfs/nfsproto.h>
67 #include <nfsserver/nfs.h>
68 #include <nfs/xdr_subs.h>
69 #include <nfsserver/nfsm_subs.h>
71 #include <security/mac/mac_framework.h>
76 static int nfs_realign_test;
77 static int nfs_realign_count;
79 SYSCTL_DECL(_vfs_nfsrv);
81 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
82 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
86 * There is a congestion window for outstanding rpcs maintained per mount
87 * point. The cwnd size is adjusted in roughly the way that:
88 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
89 * SIGCOMM '88". ACM, August 1988.
90 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
91 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
92 * of rpcs is in progress.
93 * (The sent count and cwnd are scaled for integer arith.)
94 * Variants of "slow start" were tried and were found to be too much of a
95 * performance hit (ave. rtt 3 times larger),
96 * I suspect due to the large rtt that nfs rpcs have.
98 #define NFS_CWNDSCALE 256
99 #define NFS_MAXCWND (NFS_CWNDSCALE * 32)
100 struct callout nfsrv_callout;
102 static void nfs_realign(struct mbuf **pm, int hsiz); /* XXX SHARED */
103 static int nfsrv_getstream(struct nfssvc_sock *, int);
105 int32_t (*nfsrv3_procs[NFS_NPROCS])(struct nfsrv_descript *nd,
106 struct nfssvc_sock *slp,
108 struct mbuf **mreqp) = {
136 * Generate the rpc reply header
137 * siz arg. is used to decide if adding a cluster is worthwhile
140 nfs_rephead(int siz, struct nfsrv_descript *nd, int err,
141 struct mbuf **mbp, caddr_t *bposp)
148 nd->nd_repstat = err;
149 if (err && (nd->nd_flag & ND_NFSV3) == 0) /* XXX recheck */
151 MGETHDR(mreq, M_TRYWAIT, MT_DATA);
154 * If this is a big reply, use a cluster else
155 * try and leave leading space for the lower level headers.
157 mreq->m_len = 6 * NFSX_UNSIGNED;
159 if ((max_hdr + siz) >= MINCLSIZE) {
160 MCLGET(mreq, M_TRYWAIT);
162 mreq->m_data += min(max_hdr, M_TRAILINGSPACE(mreq));
163 tl = mtod(mreq, u_int32_t *);
164 bpos = ((caddr_t)tl) + mreq->m_len;
165 *tl++ = txdr_unsigned(nd->nd_retxid);
166 *tl++ = nfsrv_rpc_reply;
167 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
168 *tl++ = nfsrv_rpc_msgdenied;
169 if (err & NFSERR_AUTHERR) {
170 *tl++ = nfsrv_rpc_autherr;
171 *tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
172 mreq->m_len -= NFSX_UNSIGNED;
173 bpos -= NFSX_UNSIGNED;
175 *tl++ = nfsrv_rpc_mismatch;
176 *tl++ = txdr_unsigned(RPC_VER2);
177 *tl = txdr_unsigned(RPC_VER2);
180 *tl++ = nfsrv_rpc_msgaccepted;
182 * Send a RPCAUTH_NULL verifier - no Kerberos.
188 *tl = txdr_unsigned(RPC_PROGUNAVAIL);
191 *tl = txdr_unsigned(RPC_PROGMISMATCH);
192 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
193 *tl++ = txdr_unsigned(2);
194 *tl = txdr_unsigned(3);
197 *tl = txdr_unsigned(RPC_PROCUNAVAIL);
200 *tl = txdr_unsigned(RPC_GARBAGE);
204 if (err != NFSERR_RETVOID) {
205 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
207 *tl = txdr_unsigned(nfsrv_errmap(nd, err));
216 if (err != 0 && err != NFSERR_RETVOID)
217 nfsrvstats.srvrpc_errs++;
225 * Check for badly aligned mbuf data and realign by copying the unaligned
226 * portion of the data into a new mbuf chain and freeing the portions
227 * of the old chain that were replaced.
229 * We cannot simply realign the data within the existing mbuf chain
230 * because the underlying buffers may contain other rpc commands and
231 * we cannot afford to overwrite them.
233 * We would prefer to avoid this situation entirely. The situation does
234 * not occur with NFS/UDP and is supposed to only occassionally occur
235 * with TCP. Use vfs.nfs.realign_count and realign_test to check this.
238 nfs_realign(struct mbuf **pm, int hsiz) /* XXX COMMON */
241 struct mbuf *n = NULL;
245 while ((m = *pm) != NULL) {
246 if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
247 MGET(n, M_TRYWAIT, MT_DATA);
248 if (m->m_len >= MINCLSIZE) {
249 MCLGET(n, M_TRYWAIT);
258 * If n is non-NULL, loop on m copying data, then replace the
259 * portion of the chain that had to be realigned.
264 m_copyback(n, off, m->m_len, mtod(m, caddr_t));
275 * Parse an RPC request
277 * - fill in the cred struct.
280 nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header)
285 u_int32_t nfsvers, auth_type;
287 struct mbuf *mrep, *md;
295 tl = nfsm_dissect_nonblock(u_int32_t *, 10 * NFSX_UNSIGNED);
296 nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
297 if (*tl++ != nfsrv_rpc_call) {
302 tl = nfsm_dissect_nonblock(u_int32_t *, 8 * NFSX_UNSIGNED);
305 if (*tl++ != nfsrv_rpc_vers) {
306 nd->nd_repstat = ERPCMISMATCH;
307 nd->nd_procnum = NFSPROC_NOOP;
310 if (*tl != nfsrv_nfs_prog) {
311 nd->nd_repstat = EPROGUNAVAIL;
312 nd->nd_procnum = NFSPROC_NOOP;
316 nfsvers = fxdr_unsigned(u_int32_t, *tl++);
317 if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) {
318 nd->nd_repstat = EPROGMISMATCH;
319 nd->nd_procnum = NFSPROC_NOOP;
322 nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
323 if (nd->nd_procnum == NFSPROC_NULL)
325 if (nfsvers == NFS_VER3) {
326 nd->nd_flag = ND_NFSV3;
327 if (nd->nd_procnum >= NFS_NPROCS) {
328 nd->nd_repstat = EPROCUNAVAIL;
329 nd->nd_procnum = NFSPROC_NOOP;
333 if (nd->nd_procnum > NFSV2PROC_STATFS) {
334 nd->nd_repstat = EPROCUNAVAIL;
335 nd->nd_procnum = NFSPROC_NOOP;
338 /* Map the v2 procedure numbers into v3 ones */
339 nd->nd_procnum = nfsrv_nfsv3_procid[nd->nd_procnum];
342 len = fxdr_unsigned(int, *tl++);
343 if (len < 0 || len > RPCAUTH_MAXSIZ) {
351 if (auth_type == nfsrv_rpc_auth_unix) {
352 len = fxdr_unsigned(int, *++tl);
353 if (len < 0 || len > NFS_MAXNAMLEN) {
357 nfsm_adv(nfsm_rndup(len));
358 tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED);
359 nd->nd_cr->cr_uid = nd->nd_cr->cr_ruid =
360 nd->nd_cr->cr_svuid = fxdr_unsigned(uid_t, *tl++);
361 nd->nd_cr->cr_groups[0] = nd->nd_cr->cr_rgid =
362 nd->nd_cr->cr_svgid = fxdr_unsigned(gid_t, *tl++);
364 mac_associate_nfsd_label(nd->nd_cr);
366 len = fxdr_unsigned(int, *tl);
367 if (len < 0 || len > RPCAUTH_UNIXGIDS) {
371 tl = nfsm_dissect_nonblock(u_int32_t *, (len + 2) * NFSX_UNSIGNED);
372 for (i = 1; i <= len; i++)
374 nd->nd_cr->cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
377 nd->nd_cr->cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
378 if (nd->nd_cr->cr_ngroups > 1)
379 nfsrvw_sort(nd->nd_cr->cr_groups, nd->nd_cr->cr_ngroups);
380 len = fxdr_unsigned(int, *++tl);
381 if (len < 0 || len > RPCAUTH_MAXSIZ) {
386 nfsm_adv(nfsm_rndup(len));
388 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
389 nd->nd_procnum = NFSPROC_NOOP;
401 * Socket upcall routine for the nfsd sockets.
402 * The caddr_t arg is a pointer to the "struct nfssvc_sock".
403 * Essentially do as much as possible non-blocking, else punt and it will
404 * be called with M_TRYWAIT from an nfsd.
407 nfsrv_rcv(struct socket *so, void *arg, int waitflag)
409 struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
412 struct sockaddr *nam;
416 NFSD_UNLOCK_ASSERT();
418 /* XXXRW: Unlocked read. */
419 if ((slp->ns_flag & SLP_VALID) == 0)
423 * We can't do this in the context of a socket callback
424 * because we're called with locks held.
427 if (waitflag == M_DONTWAIT) {
429 slp->ns_flag |= SLP_NEEDQ;
436 if (so->so_type == SOCK_STREAM) {
438 * If there are already records on the queue, defer soreceive()
439 * to an nfsd so that there is feedback to the TCP layer that
440 * the nfs servers are heavily loaded.
442 if (STAILQ_FIRST(&slp->ns_rec) != NULL &&
443 waitflag == M_DONTWAIT) {
444 slp->ns_flag |= SLP_NEEDQ;
451 auio.uio_resid = 1000000000;
452 flags = MSG_DONTWAIT;
454 error = soreceive(so, &nam, &auio, &mp, NULL, &flags);
456 if (error || mp == NULL) {
457 if (error == EWOULDBLOCK)
458 slp->ns_flag |= SLP_NEEDQ;
460 slp->ns_flag |= SLP_DISCONN;
464 if (slp->ns_rawend) {
465 slp->ns_rawend->m_next = m;
466 slp->ns_cc += 1000000000 - auio.uio_resid;
469 slp->ns_cc = 1000000000 - auio.uio_resid;
476 * Now try and parse record(s) out of the raw stream data.
478 error = nfsrv_getstream(slp, waitflag);
481 slp->ns_flag |= SLP_DISCONN;
483 slp->ns_flag |= SLP_NEEDQ;
487 auio.uio_resid = 1000000000;
488 flags = MSG_DONTWAIT;
490 error = soreceive(so, &nam, &auio, &mp, NULL, &flags);
492 struct nfsrv_rec *rec;
493 rec = malloc(sizeof(struct nfsrv_rec),
495 waitflag == M_DONTWAIT ? M_NOWAIT : M_WAITOK);
503 nfs_realign(&mp, 10 * NFSX_UNSIGNED);
505 rec->nr_address = nam;
507 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
511 if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
512 && error != EWOULDBLOCK) {
513 slp->ns_flag |= SLP_DISCONN;
521 * Now try and process the request records, non-blocking.
524 if (waitflag == M_DONTWAIT &&
525 (STAILQ_FIRST(&slp->ns_rec) != NULL ||
526 (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
532 * Try and extract an RPC request from the mbuf data list received on a
533 * stream socket. The "waitflag" argument indicates whether or not it
537 nfsrv_getstream(struct nfssvc_sock *slp, int waitflag)
539 struct mbuf *m, **mpp;
542 struct mbuf *om, *m2, *recm;
547 if (slp->ns_flag & SLP_GETSTREAM)
548 panic("nfs getstream");
549 slp->ns_flag |= SLP_GETSTREAM;
551 if (slp->ns_reclen == 0) {
552 if (slp->ns_cc < NFSX_UNSIGNED) {
553 slp->ns_flag &= ~SLP_GETSTREAM;
557 if (m->m_len >= NFSX_UNSIGNED) {
558 bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
559 m->m_data += NFSX_UNSIGNED;
560 m->m_len -= NFSX_UNSIGNED;
562 cp1 = (caddr_t)&recmark;
563 cp2 = mtod(m, caddr_t);
564 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
565 while (m->m_len == 0) {
567 cp2 = mtod(m, caddr_t);
574 slp->ns_cc -= NFSX_UNSIGNED;
575 recmark = ntohl(recmark);
576 slp->ns_reclen = recmark & ~0x80000000;
577 if (recmark & 0x80000000)
578 slp->ns_flag |= SLP_LASTFRAG;
580 slp->ns_flag &= ~SLP_LASTFRAG;
581 if (slp->ns_reclen > NFS_MAXPACKET || slp->ns_reclen <= 0) {
582 slp->ns_flag &= ~SLP_GETSTREAM;
588 * Now get the record part.
590 * Note that slp->ns_reclen may be 0. Linux sometimes
591 * generates 0-length RPCs.
594 if (slp->ns_cc == slp->ns_reclen) {
596 slp->ns_raw = slp->ns_rawend = NULL;
597 slp->ns_cc = slp->ns_reclen = 0;
598 } else if (slp->ns_cc > slp->ns_reclen) {
603 while (len < slp->ns_reclen) {
604 if ((len + m->m_len) > slp->ns_reclen) {
606 m2 = m_copym(m, 0, slp->ns_reclen - len,
615 m->m_data += slp->ns_reclen - len;
616 m->m_len -= slp->ns_reclen - len;
617 len = slp->ns_reclen;
619 slp->ns_flag &= ~SLP_GETSTREAM;
620 return (EWOULDBLOCK);
622 } else if ((len + m->m_len) == slp->ns_reclen) {
638 slp->ns_flag &= ~SLP_GETSTREAM;
643 * Accumulate the fragments into a record.
647 mpp = &((*mpp)->m_next);
649 if (slp->ns_flag & SLP_LASTFRAG) {
650 struct nfsrv_rec *rec;
652 rec = malloc(sizeof(struct nfsrv_rec), M_NFSRVDESC,
653 waitflag == M_DONTWAIT ? M_NOWAIT : M_WAITOK);
655 nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED);
656 rec->nr_address = NULL;
657 rec->nr_packet = slp->ns_frag;
659 STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
664 m_freem(slp->ns_frag);
672 * Parse an RPC header.
675 nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd,
676 struct nfsrv_descript **ndp)
678 struct nfsrv_rec *rec;
680 struct sockaddr *nam;
681 struct nfsrv_descript *nd;
687 if ((slp->ns_flag & SLP_VALID) == 0 ||
688 STAILQ_FIRST(&slp->ns_rec) == NULL)
690 rec = STAILQ_FIRST(&slp->ns_rec);
691 KASSERT(rec->nr_packet != NULL, ("nfsrv_dorec: missing mbuf"));
692 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
693 nam = rec->nr_address;
695 free(rec, M_NFSRVDESC);
697 MALLOC(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript),
698 M_NFSRVDESC, M_WAITOK);
701 nd->nd_md = nd->nd_mrep = m;
703 nd->nd_dpos = mtod(m, caddr_t);
704 error = nfs_getreq(nd, nfsd, TRUE);
709 if (nd->nd_cr != NULL)
711 free((caddr_t)nd, M_NFSRVDESC);
720 * Search for a sleeping nfsd and wake it up.
721 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
722 * running nfsds will go look for the work in the nfssvc_sock list.
725 nfsrv_wakenfsd(struct nfssvc_sock *slp)
731 if ((slp->ns_flag & SLP_VALID) == 0)
733 TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) {
734 if (nd->nfsd_flag & NFSD_WAITING) {
735 nd->nfsd_flag &= ~NFSD_WAITING;
737 panic("nfsd wakeup");
744 slp->ns_flag |= SLP_DOREC;
745 nfsd_head_flag |= NFSD_CHECKSLP;
749 * This is the nfs send routine.
750 * For the server side:
751 * - return EINTR or ERESTART if interrupted by a signal
752 * - return EPIPE if a connection is lost for connection based sockets (TCP...)
753 * - do any cleanup required by recoverable socket errors (?)
756 nfsrv_send(struct socket *so, struct sockaddr *nam, struct mbuf *top)
758 struct sockaddr *sendnam;
759 int error, soflags, flags;
761 NFSD_UNLOCK_ASSERT();
763 soflags = so->so_proto->pr_flags;
764 if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
768 if (so->so_type == SOCK_SEQPACKET)
773 error = sosend(so, sendnam, 0, top, 0, flags, curthread/*XXX*/);
774 if (error == ENOBUFS && so->so_type == SOCK_DGRAM)
778 log(LOG_INFO, "nfsd send error %d\n", error);
781 * Handle any recoverable (soft) socket errors here. (?)
783 if (error != EINTR && error != ERESTART &&
784 error != EWOULDBLOCK && error != EPIPE)
791 * NFS server timer routine.
794 nfsrv_timer(void *arg)
796 struct nfssvc_sock *slp;
801 * Scan the write gathering queues for writes that need to be
804 cur_usec = nfs_curusec();
805 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
806 if (LIST_FIRST(&slp->ns_tq) &&
807 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
811 callout_reset(&nfsrv_callout, nfsrv_ticks, nfsrv_timer, NULL);