2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1989, 1991, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
40 * Socket operations for use by nfs
43 #include "opt_kgssapi.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/limits.h>
51 #include <sys/malloc.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
56 #include <sys/signalvar.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysctl.h>
59 #include <sys/syslog.h>
60 #include <sys/vnode.h>
65 #include <kgssapi/krb5/kcrypto.h>
67 #include <fs/nfs/nfsport.h>
70 #include <sys/dtrace_bsd.h>
72 dtrace_nfsclient_nfs23_start_probe_func_t
73 dtrace_nfscl_nfs234_start_probe;
75 dtrace_nfsclient_nfs23_done_probe_func_t
76 dtrace_nfscl_nfs234_done_probe;
79 * Registered probes by RPC type.
81 uint32_t nfscl_nfs2_start_probes[NFSV41_NPROCS + 1];
82 uint32_t nfscl_nfs2_done_probes[NFSV41_NPROCS + 1];
84 uint32_t nfscl_nfs3_start_probes[NFSV41_NPROCS + 1];
85 uint32_t nfscl_nfs3_done_probes[NFSV41_NPROCS + 1];
87 uint32_t nfscl_nfs4_start_probes[NFSV41_NPROCS + 1];
88 uint32_t nfscl_nfs4_done_probes[NFSV41_NPROCS + 1];
95 extern struct nfsstatsv1 nfsstatsv1;
96 extern struct nfsreqhead nfsd_reqq;
97 extern int nfscl_ticks;
98 extern void (*ncl_call_invalcaches)(struct vnode *);
99 extern int nfs_numnfscbd;
100 extern int nfscl_debuglevel;
101 extern int nfsrv_lease;
103 SVCPOOL *nfscbd_pool;
104 static int nfsrv_gsscallbackson = 0;
105 static int nfs_bufpackets = 4;
106 static int nfs_reconnects;
107 static int nfs3_jukebox_delay = 10;
108 static int nfs_skip_wcc_data_onerr = 1;
109 static int nfs_dsretries = 2;
111 SYSCTL_DECL(_vfs_nfs);
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
114 "Buffer reservation size 2 < x < 64");
115 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
116 "Number of times the nfs client has had to reconnect");
117 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
118 "Number of seconds to delay a retry after receiving EJUKEBOX");
119 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
120 "Disable weak cache consistency checking when server returns an error");
121 SYSCTL_INT(_vfs_nfs, OID_AUTO, dsretries, CTLFLAG_RW, &nfs_dsretries, 0,
122 "Number of retries for a DS RPC before failure");
124 static void nfs_down(struct nfsmount *, struct thread *, const char *,
126 static void nfs_up(struct nfsmount *, struct thread *, const char *,
128 static int nfs_msg(struct thread *, const char *, const char *, int);
130 struct nfs_cached_auth {
131 int ca_refs; /* refcount, including 1 from the cache */
132 uid_t ca_uid; /* uid that corresponds to this auth */
133 AUTH *ca_auth; /* RPC auth handle */
136 static int nfsv2_procid[NFS_V3NPROCS] = {
162 * Initialize sockets and congestion for a new NFS connection.
163 * We do not free the sockaddr if error.
164 * Which arguments are set to NULL indicate what kind of call it is.
165 * cred == NULL --> a call to connect to a pNFS DS
166 * nmp == NULL --> indicates an upcall to userland or a NFSv4.0 callback
169 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
170 struct ucred *cred, NFSPROC_T *p, int callback_retry_mult, bool dotls)
172 int rcvreserve, sndreserve;
173 int pktscale, pktscalesav;
174 struct sockaddr *saddr;
175 struct ucred *origcred;
177 struct netconfig *nconf;
179 int one = 1, retries, error = 0;
180 struct thread *td = curthread;
185 * We need to establish the socket using the credentials of
186 * the mountpoint. Some parts of this process (such as
187 * sobind() and soconnect()) will use the curent thread's
188 * credential instead of the socket credential. To work
189 * around this, temporarily change the current thread's
190 * credential to that of the mountpoint.
192 * XXX: It would be better to explicitly pass the correct
193 * credential to sobind() and soconnect().
195 origcred = td->td_ucred;
198 * Use the credential in nr_cred, if not NULL.
200 if (nrp->nr_cred != NULL)
201 td->td_ucred = nrp->nr_cred;
206 if (saddr->sa_family == AF_INET)
207 if (nrp->nr_sotype == SOCK_DGRAM)
208 nconf = getnetconfigent("udp");
210 nconf = getnetconfigent("tcp");
212 if (nrp->nr_sotype == SOCK_DGRAM)
213 nconf = getnetconfigent("udp6");
215 nconf = getnetconfigent("tcp6");
217 pktscale = nfs_bufpackets;
222 pktscalesav = pktscale;
224 * soreserve() can fail if sb_max is too small, so shrink pktscale
225 * and try again if there is an error.
226 * Print a log message suggesting increasing sb_max.
227 * Creating a socket and doing this is necessary since, if the
228 * reservation sizes are too large and will make soreserve() fail,
229 * the connection will work until a large send is attempted and
230 * then it will loop in the krpc code.
233 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
234 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
235 nrp->nr_soproto, td->td_ucred, td);
239 if (error != 0 && pktscale > 2) {
240 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
241 pktscale == pktscalesav)
242 printf("Consider increasing kern.ipc.maxsockbuf\n");
245 if (nrp->nr_sotype == SOCK_DGRAM) {
247 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
249 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
252 sndreserve = rcvreserve = 1024 * pktscale;
255 if (nrp->nr_sotype != SOCK_STREAM)
256 panic("nfscon sotype");
258 sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
259 sizeof (u_int32_t)) * pktscale;
260 rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
261 sizeof (u_int32_t)) * pktscale;
263 sndreserve = rcvreserve = 1024 * pktscale;
266 error = soreserve(so, sndreserve, rcvreserve);
267 if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
269 printf("Must increase kern.ipc.maxsockbuf or reduce"
271 } while (error != 0 && pktscale > 2);
276 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
277 nrp->nr_vers, sndreserve, rcvreserve);
278 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
280 if ((nmp->nm_flag & NFSMNT_INT))
281 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
282 if ((nmp->nm_flag & NFSMNT_RESVPORT))
283 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
285 CLNT_CONTROL(client, CLSET_TLS, &one);
286 if (NFSHASSOFT(nmp)) {
287 if (nmp->nm_sotype == SOCK_DGRAM)
289 * For UDP, the large timeout for a reconnect
290 * will be set to "nm_retry * nm_timeo / 2", so
291 * we only want to do 2 reconnect timeout
296 retries = nmp->nm_retry;
299 if (NFSHASNFSV4N(nmp)) {
301 if (NFSHASSOFT(nmp)) {
303 * This should be a DS mount.
304 * Use CLSET_TIMEOUT to set the timeout
305 * for connections to DSs instead of
306 * specifying a timeout on each RPC.
307 * This is done so that SO_SNDTIMEO
308 * is set on the TCP socket as well
309 * as specifying a time limit when
310 * waiting for an RPC reply. Useful
311 * if the send queue for the TCP
312 * connection has become constipated,
313 * due to a failed DS.
314 * The choice of lease_duration / 4 is
315 * fairly arbitrary, but seems to work
316 * ok, with a lower bound of 10sec.
318 timo.tv_sec = nfsrv_lease / 4;
319 if (timo.tv_sec < 10)
322 CLNT_CONTROL(client, CLSET_TIMEOUT,
326 * Make sure the nfscbd_pool doesn't get
327 * destroyed while doing this.
330 if (nfs_numnfscbd > 0) {
333 xprt = svc_vc_create_backchannel(
335 CLNT_CONTROL(client, CLSET_BACKCHANNEL,
339 if (nfs_numnfscbd == 0)
340 wakeup(&nfs_numnfscbd);
345 * cred == NULL for a DS connect.
346 * For connects to a DS, set a retry limit
347 * so that failed DSs will be detected.
348 * This is ok for NFSv4.1, since a DS does
349 * not maintain open/lock state and is the
350 * only case where using a "soft" mount is
351 * recommended for NFSv4.
352 * For mounts from the MDS to DS, this is done
353 * via mount options, but that is not the case
354 * here. The retry limit here can be adjusted
355 * via the sysctl vfs.nfs.dsretries.
356 * See the comment above w.r.t. timeout.
358 timo.tv_sec = nfsrv_lease / 4;
359 if (timo.tv_sec < 10)
362 CLNT_CONTROL(client, CLSET_TIMEOUT, &timo);
363 retries = nfs_dsretries;
369 * - Null RPC callback to client
370 * - Non-Null RPC callback to client, wait a little longer
371 * - upcalls to nfsuserd and gssd (clp == NULL)
373 if (callback_retry_mult == 0) {
374 retries = NFSV4_UPCALLRETRY;
375 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
377 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
380 CLNT_CONTROL(client, CLSET_TLS, &one);
382 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
386 * For UDP, there are 2 timeouts:
387 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
388 * that does a retransmit of an RPC request using the same
389 * socket and xid. This is what you normally want to do,
390 * since NFS servers depend on "same xid" for their
391 * Duplicate Request Cache.
392 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
393 * retransmits on the same socket should fail and a fresh
394 * socket created. Each of these timeouts counts as one
395 * CLSET_RETRIES as set above.
396 * Set the initial retransmit timeout for UDP. This timeout
397 * doesn't exist for TCP and the following call just fails,
400 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
401 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
402 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
405 mtx_lock(&nrp->nr_mtx);
406 if (nrp->nr_client != NULL) {
407 mtx_unlock(&nrp->nr_mtx);
409 * Someone else already connected.
411 CLNT_RELEASE(client);
413 nrp->nr_client = client;
415 * Protocols that do not require connections may be optionally
416 * left unconnected for servers that reply from a port other
419 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
420 mtx_unlock(&nrp->nr_mtx);
421 CLNT_CONTROL(client, CLSET_CONNECT, &one);
423 mtx_unlock(&nrp->nr_mtx);
427 /* Restore current thread's credentials. */
428 td->td_ucred = origcred;
435 * NFS disconnect. Clean up and unlink.
438 newnfs_disconnect(struct nfssockreq *nrp)
442 mtx_lock(&nrp->nr_mtx);
443 if (nrp->nr_client != NULL) {
444 client = nrp->nr_client;
445 nrp->nr_client = NULL;
446 mtx_unlock(&nrp->nr_mtx);
447 rpc_gss_secpurge_call(client);
449 CLNT_RELEASE(client);
451 mtx_unlock(&nrp->nr_mtx);
456 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
457 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
459 rpc_gss_service_t svc;
462 switch (secflavour) {
463 case RPCSEC_GSS_KRB5:
464 case RPCSEC_GSS_KRB5I:
465 case RPCSEC_GSS_KRB5P:
467 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
470 if (secflavour == RPCSEC_GSS_KRB5)
471 svc = rpc_gss_svc_none;
472 else if (secflavour == RPCSEC_GSS_KRB5I)
473 svc = rpc_gss_svc_integrity;
475 svc = rpc_gss_svc_privacy;
477 if (clnt_principal == NULL)
478 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
479 srv_principal, mech_oid, svc);
481 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
482 clnt_principal, srv_principal, "kerberosv5",
483 svc, NULL, NULL, NULL);
491 return (authunix_create(cred));
496 * Callback from the RPC code to generate up/down notifications.
499 struct nfs_feedback_arg {
500 struct nfsmount *nf_mount;
501 int nf_lastmsg; /* last tprintf */
503 struct thread *nf_td;
507 nfs_feedback(int type, int proc, void *arg)
509 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
510 struct nfsmount *nmp = nf->nf_mount;
514 case FEEDBACK_REXMIT2:
515 case FEEDBACK_RECONNECT:
517 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
518 nfs_down(nmp, nf->nf_td,
519 "not responding", 0, NFSSTA_TIMEO);
520 nf->nf_tprintfmsg = TRUE;
521 nf->nf_lastmsg = now;
526 nfs_up(nf->nf_mount, nf->nf_td,
527 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
533 * newnfs_request - goes something like this
534 * - does the rpc by calling the krpc layer
535 * - break down rpc header and return with nfs reply
536 * nb: always frees up nd_mreq mbuf list
539 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
540 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
541 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
542 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
544 uint32_t retseq, retval, slotseq, *tl;
546 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
547 int error = 0, usegssname = 0, secflavour = AUTH_SYS;
548 int freeslot, maxslot, reterr, slotpos, timeo;
550 u_int trylater_delay = 1;
551 struct nfs_feedback_arg nf;
554 struct rpc_callextra ext;
556 struct nfsreq *rep = NULL;
557 char *srv_principal = NULL, *clnt_principal = NULL;
559 struct ucred *authcred;
560 struct nfsclsession *sep;
561 uint8_t sessionid[NFSX_V4SESSIONID];
566 /* Reject requests while attempting a forced unmount. */
567 if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
568 m_freem(nd->nd_mreq);
573 * Set authcred, which is used to acquire RPC credentials to
574 * the cred argument, by default. The crhold() should not be
575 * necessary, but will ensure that some future code change
576 * doesn't result in the credential being free'd prematurely.
578 authcred = crhold(cred);
580 /* For client side interruptible mounts, mask off the signals. */
581 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
582 newnfs_set_sigmask(td, &oldset);
587 * XXX if not already connected call nfs_connect now. Longer
588 * term, change nfs_mount to call nfs_connect unconditionally
589 * and let clnt_reconnect_create handle reconnects.
591 if (nrp->nr_client == NULL)
592 newnfs_connect(nmp, nrp, cred, td, 0, false);
595 * For a client side mount, nmp is != NULL and clp == NULL. For
596 * server calls (callbacks or upcalls), nmp == NULL.
600 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
601 secflavour = RPCSEC_GSS_KRB5;
602 if (nd->nd_procnum != NFSPROC_NULL) {
603 if (clp->lc_flags & LCL_GSSINTEGRITY)
604 secflavour = RPCSEC_GSS_KRB5I;
605 else if (clp->lc_flags & LCL_GSSPRIVACY)
606 secflavour = RPCSEC_GSS_KRB5P;
610 } else if (nmp != NULL && NFSHASKERB(nmp) &&
611 nd->nd_procnum != NFSPROC_NULL) {
612 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
613 nd->nd_flag |= ND_USEGSSNAME;
614 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
616 * If there is a client side host based credential,
617 * use that, otherwise use the system uid, if set.
618 * The system uid is in the nmp->nm_sockreq.nr_cred
621 if (nmp->nm_krbnamelen > 0) {
623 clnt_principal = nmp->nm_krbname;
624 } else if (nmp->nm_uid != (uid_t)-1) {
625 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
626 ("newnfs_request: NULL nr_cred"));
628 authcred = crhold(nmp->nm_sockreq.nr_cred);
630 } else if (nmp->nm_krbnamelen == 0 &&
631 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
633 * If there is no host based principal name and
634 * the system uid is set and this is root, use the
635 * system uid, since root won't have user
636 * credentials in a credentials cache file.
637 * The system uid is in the nmp->nm_sockreq.nr_cred
640 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
641 ("newnfs_request: NULL nr_cred"));
643 authcred = crhold(nmp->nm_sockreq.nr_cred);
645 if (NFSHASINTEGRITY(nmp))
646 secflavour = RPCSEC_GSS_KRB5I;
647 else if (NFSHASPRIVACY(nmp))
648 secflavour = RPCSEC_GSS_KRB5P;
650 secflavour = RPCSEC_GSS_KRB5;
651 srv_principal = NFSMNT_SRVKRBNAME(nmp);
652 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
653 nd->nd_procnum != NFSPROC_NULL &&
654 (nd->nd_flag & ND_USEGSSNAME) != 0) {
656 * Use the uid that did the mount when the RPC is doing
657 * NFSv4 system operations, as indicated by the
658 * ND_USEGSSNAME flag, for the AUTH_SYS case.
659 * The credentials in nm_sockreq.nr_cred were used for the
662 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
663 ("newnfs_request: NULL nr_cred"));
665 authcred = crhold(nmp->nm_sockreq.nr_cred);
669 bzero(&nf, sizeof(struct nfs_feedback_arg));
672 nf.nf_lastmsg = NFSD_MONOSEC -
673 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
676 if (nd->nd_procnum == NFSPROC_NULL)
677 auth = authnone_create();
678 else if (usegssname) {
680 * For this case, the authenticator is held in the
681 * nfssockreq structure, so don't release the reference count
682 * held on it. --> Don't AUTH_DESTROY() it in this function.
684 if (nrp->nr_auth == NULL)
685 nrp->nr_auth = nfs_getauth(nrp, secflavour,
686 clnt_principal, srv_principal, NULL, authcred);
688 rpc_gss_refresh_auth_call(nrp->nr_auth);
691 auth = nfs_getauth(nrp, secflavour, NULL,
692 srv_principal, NULL, authcred);
695 m_freem(nd->nd_mreq);
697 newnfs_restore_sigmask(td, &oldset);
700 bzero(&ext, sizeof(ext));
703 ext.rc_feedback = nfs_feedback;
704 ext.rc_feedback_arg = &nf;
707 procnum = nd->nd_procnum;
708 if ((nd->nd_flag & ND_NFSV4) &&
709 nd->nd_procnum != NFSPROC_NULL &&
710 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
711 procnum = NFSV4PROC_COMPOUND;
714 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
716 /* Map the procnum to the old NFSv2 one, as required. */
717 if ((nd->nd_flag & ND_NFSV2) != 0) {
718 if (nd->nd_procnum < NFS_V3NPROCS)
719 procnum = nfsv2_procid[nd->nd_procnum];
721 procnum = NFSV2PROC_NOOP;
725 * Now only used for the R_DONTRECOVER case, but until that is
726 * supported within the krpc code, I need to keep a queue of
727 * outstanding RPCs for nfsv4 client requests.
729 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
730 rep = malloc(sizeof(struct nfsreq),
731 M_NFSDREQ, M_WAITOK);
733 if (dtrace_nfscl_nfs234_start_probe != NULL) {
737 if (nd->nd_flag & ND_NFSV4) {
739 nfscl_nfs4_start_probes[nd->nd_procnum];
740 probe_procnum = nd->nd_procnum;
741 } else if (nd->nd_flag & ND_NFSV3) {
742 probe_id = nfscl_nfs3_start_probes[procnum];
743 probe_procnum = procnum;
746 nfscl_nfs2_start_probes[nd->nd_procnum];
747 probe_procnum = procnum;
750 (dtrace_nfscl_nfs234_start_probe)
751 (probe_id, vp, nd->nd_mreq, cred,
756 freeslot = -1; /* Set to slot that needs to be free'd */
758 slot = -1; /* Slot that needs a sequence# increment. */
760 * This timeout specifies when a new socket should be created,
761 * along with new xid values. For UDP, this should be done
762 * infrequently, since retransmits of RPC requests should normally
768 timo.tv_sec = NFSV4_UPCALLTIMEO;
770 timo.tv_sec = NFSV4_CALLBACKTIMEO;
772 if (nrp->nr_sotype != SOCK_DGRAM) {
774 if ((nmp->nm_flag & NFSMNT_NFSV4))
775 timo.tv_sec = INT_MAX;
777 timo.tv_sec = NFS_TCPTIMEO;
779 if (NFSHASSOFT(nmp)) {
781 * CLSET_RETRIES is set to 2, so this should be
782 * half of the total timeout required.
784 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
787 timo.tv_sec = timeo / NFS_HZ;
788 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
791 /* For UDP hard mounts, use a large value. */
792 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
801 * Chain request into list of outstanding requests.
804 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
810 if (clp != NULL && sep != NULL)
811 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
812 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
814 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
815 nd->nd_mreq, &nd->nd_mrep, timo);
816 NFSCL_DEBUG(2, "clnt call=%d\n", stat);
820 * RPC done, unlink the request.
823 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
828 * If there was a successful reply and a tprintf msg.
829 * tprintf a response.
831 if (stat == RPC_SUCCESS) {
833 } else if (stat == RPC_TIMEDOUT) {
834 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
836 } else if (stat == RPC_VERSMISMATCH) {
837 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
839 } else if (stat == RPC_PROGVERSMISMATCH) {
840 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
841 error = EPROTONOSUPPORT;
842 } else if (stat == RPC_INTR) {
844 } else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
845 stat == RPC_SYSTEMERROR) {
846 /* Check for a session slot that needs to be free'd. */
847 if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
848 (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
849 nd->nd_procnum != NFSPROC_NULL) {
851 * This should only occur when either the MDS or
852 * a client has an RPC against a DS fail.
853 * This happens because these cases use "soft"
854 * connections that can time out and fail.
855 * The slot used for this RPC is now in a
856 * non-deterministic state, but if the slot isn't
857 * free'd, threads can get stuck waiting for a slot.
860 sep = nfsmnt_mdssession(nmp);
862 * Bump the sequence# out of range, so that reuse of
863 * this slot will result in an NFSERR_SEQMISORDERED
864 * error and not a bogus cached RPC reply.
866 mtx_lock(&sep->nfsess_mtx);
867 sep->nfsess_slotseq[nd->nd_slotid] += 10;
868 mtx_unlock(&sep->nfsess_mtx);
869 /* And free the slot. */
870 nfsv4_freeslot(sep, nd->nd_slotid);
872 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
875 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
879 m_freem(nd->nd_mreq);
883 free(rep, M_NFSDREQ);
885 newnfs_restore_sigmask(td, &oldset);
889 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
892 * Search for any mbufs that are not a multiple of 4 bytes long
893 * or with m_data not longword aligned.
894 * These could cause pointer alignment problems, so copy them to
895 * well aligned mbufs.
897 newnfs_realign(&nd->nd_mrep, M_WAITOK);
898 nd->nd_md = nd->nd_mrep;
899 nd->nd_dpos = mtod(nd->nd_md, caddr_t);
901 if (nd->nd_procnum != NFSPROC_NULL &&
902 nd->nd_procnum != NFSV4PROC_CBNULL) {
903 /* If sep == NULL, set it to the default in nmp. */
904 if (sep == NULL && nmp != NULL)
905 sep = nfsmnt_mdssession(nmp);
907 * and now the actual NFS xdr.
909 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
910 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
911 if (nd->nd_repstat >= 10000)
912 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
913 (int)nd->nd_repstat);
916 * Get rid of the tag, return count and SEQUENCE result for
919 if ((nd->nd_flag & ND_NFSV4) != 0 && nd->nd_repstat !=
920 NFSERR_MINORVERMISMATCH) {
921 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
922 i = fxdr_unsigned(int, *tl);
923 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
926 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
927 opcnt = fxdr_unsigned(int, *tl++);
928 i = fxdr_unsigned(int, *tl++);
929 j = fxdr_unsigned(int, *tl);
931 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
933 * If the first op is Sequence, free up the slot.
935 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
936 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
937 NFSCL_DEBUG(1, "failed seq=%d\n", j);
938 if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
939 (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
940 j == 0)) && sep != NULL) {
941 if (i == NFSV4OP_SEQUENCE)
942 NFSM_DISSECT(tl, uint32_t *,
946 NFSM_DISSECT(tl, uint32_t *,
949 mtx_lock(&sep->nfsess_mtx);
950 if (bcmp(tl, sep->nfsess_sessionid,
951 NFSX_V4SESSIONID) == 0) {
952 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
953 retseq = fxdr_unsigned(uint32_t, *tl++);
954 slot = fxdr_unsigned(int, *tl++);
956 if (retseq != sep->nfsess_slotseq[slot])
957 printf("retseq diff 0x%x\n",
959 retval = fxdr_unsigned(uint32_t, *++tl);
960 if ((retval + 1) < sep->nfsess_foreslots
962 sep->nfsess_foreslots = (retval
964 else if ((retval + 1) >
965 sep->nfsess_foreslots)
966 sep->nfsess_foreslots = (retval
967 < 64) ? (retval + 1) : 64;
969 mtx_unlock(&sep->nfsess_mtx);
971 /* Grab the op and status for the next one. */
973 NFSM_DISSECT(tl, uint32_t *,
975 i = fxdr_unsigned(int, *tl++);
976 j = fxdr_unsigned(int, *tl);
980 if (nd->nd_repstat != 0) {
981 if (nd->nd_repstat == NFSERR_BADSESSION &&
982 nmp != NULL && dssep == NULL &&
983 (nd->nd_flag & ND_NFSV41) != 0) {
985 * If this is a client side MDS RPC, mark
986 * the MDS session defunct and initiate
987 * recovery, as required.
988 * The nfsess_defunct field is protected by
989 * the NFSLOCKMNT()/nm_mtx lock and not the
990 * nfsess_mtx lock to simplify its handling,
991 * for the MDS session. This lock is also
992 * sufficient for nfsess_sessionid, since it
993 * never changes in the structure.
995 NFSCL_DEBUG(1, "Got badsession\n");
998 sep = NFSMNT_MDSSESSION(nmp);
999 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
1000 NFSX_V4SESSIONID) == 0) {
1001 /* Initiate recovery. */
1002 sep->nfsess_defunct = 1;
1003 NFSCL_DEBUG(1, "Marked defunct\n");
1004 if (nmp->nm_clp != NULL) {
1005 nmp->nm_clp->nfsc_flags |=
1007 wakeup(nmp->nm_clp);
1012 * Sleep for up to 1sec waiting for a new
1015 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1018 * Get the session again, in case a new one
1019 * has been created during the sleep.
1021 sep = NFSMNT_MDSSESSION(nmp);
1023 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1024 reterr = nfsv4_sequencelookup(nmp, sep,
1025 &slotpos, &maxslot, &slotseq,
1028 /* Fill in new session info. */
1030 "Filling in new sequence\n");
1031 tl = nd->nd_sequence;
1032 bcopy(sessionid, tl,
1034 tl += NFSX_V4SESSIONID /
1036 *tl++ = txdr_unsigned(slotseq);
1037 *tl++ = txdr_unsigned(slotpos);
1038 *tl = txdr_unsigned(maxslot);
1040 if (reterr == NFSERR_BADSESSION ||
1043 "Badsession looping\n");
1044 m_freem(nd->nd_mrep);
1048 nd->nd_repstat = reterr;
1049 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1053 * When clp != NULL, it is a callback and all
1054 * callback operations can be retried for NFSERR_DELAY.
1056 if (((nd->nd_repstat == NFSERR_DELAY ||
1057 nd->nd_repstat == NFSERR_GRACE) &&
1058 (nd->nd_flag & ND_NFSV4) && (clp != NULL ||
1059 (nd->nd_procnum != NFSPROC_DELEGRETURN &&
1060 nd->nd_procnum != NFSPROC_SETATTR &&
1061 nd->nd_procnum != NFSPROC_READ &&
1062 nd->nd_procnum != NFSPROC_READDS &&
1063 nd->nd_procnum != NFSPROC_WRITE &&
1064 nd->nd_procnum != NFSPROC_WRITEDS &&
1065 nd->nd_procnum != NFSPROC_OPEN &&
1066 nd->nd_procnum != NFSPROC_CREATE &&
1067 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1068 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1069 nd->nd_procnum != NFSPROC_CLOSE &&
1070 nd->nd_procnum != NFSPROC_LOCK &&
1071 nd->nd_procnum != NFSPROC_LOCKU))) ||
1072 (nd->nd_repstat == NFSERR_DELAY &&
1073 (nd->nd_flag & ND_NFSV4) == 0) ||
1074 nd->nd_repstat == NFSERR_RESOURCE) {
1075 if (trylater_delay > NFS_TRYLATERDEL)
1076 trylater_delay = NFS_TRYLATERDEL;
1077 waituntil = NFSD_MONOSEC + trylater_delay;
1078 while (NFSD_MONOSEC < waituntil)
1079 (void) nfs_catnap(PZERO, 0, "nfstry");
1080 trylater_delay *= 2;
1082 mtx_lock(&sep->nfsess_mtx);
1083 sep->nfsess_slotseq[slot]++;
1084 *nd->nd_slotseq = txdr_unsigned(
1085 sep->nfsess_slotseq[slot]);
1086 mtx_unlock(&sep->nfsess_mtx);
1088 m_freem(nd->nd_mrep);
1094 * If the File Handle was stale, invalidate the
1095 * lookup cache, just in case.
1096 * (vp != NULL implies a client side call)
1098 if (nd->nd_repstat == ESTALE && vp != NULL) {
1100 if (ncl_call_invalcaches != NULL)
1101 (*ncl_call_invalcaches)(vp);
1104 if ((nd->nd_flag & ND_NFSV4) != 0) {
1105 /* Free the slot, as required. */
1107 nfsv4_freeslot(sep, freeslot);
1109 * If this op is Putfh, throw its results away.
1112 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1113 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1114 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1115 i = fxdr_unsigned(int, *tl++);
1116 j = fxdr_unsigned(int, *tl);
1118 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1121 * All Compounds that do an Op that must
1122 * be in sequence consist of NFSV4OP_PUTFH
1123 * followed by one of these. As such, we
1124 * can determine if the seqid# should be
1125 * incremented, here.
1127 if ((i == NFSV4OP_OPEN ||
1128 i == NFSV4OP_OPENCONFIRM ||
1129 i == NFSV4OP_OPENDOWNGRADE ||
1130 i == NFSV4OP_CLOSE ||
1131 i == NFSV4OP_LOCK ||
1132 i == NFSV4OP_LOCKU) &&
1134 (j != NFSERR_STALECLIENTID &&
1135 j != NFSERR_STALESTATEID &&
1136 j != NFSERR_BADSTATEID &&
1137 j != NFSERR_BADSEQID &&
1138 j != NFSERR_BADXDR &&
1139 j != NFSERR_RESOURCE &&
1140 j != NFSERR_NOFILEHANDLE)))
1141 nd->nd_flag |= ND_INCRSEQID;
1144 * If this op's status is non-zero, mark
1145 * that there is no more data to process.
1146 * The exception is Setattr, which always has xdr
1147 * when it has failed.
1149 if (j != 0 && i != NFSV4OP_SETATTR)
1150 nd->nd_flag |= ND_NOMOREDATA;
1153 * If R_DONTRECOVER is set, replace the stale error
1154 * reply, so that recovery isn't initiated.
1156 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1157 nd->nd_repstat == NFSERR_BADSESSION ||
1158 nd->nd_repstat == NFSERR_STALESTATEID) &&
1159 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1160 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1164 #ifdef KDTRACE_HOOKS
1165 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1169 if (nd->nd_flag & ND_NFSV4) {
1170 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1171 probe_procnum = nd->nd_procnum;
1172 } else if (nd->nd_flag & ND_NFSV3) {
1173 probe_id = nfscl_nfs3_done_probes[procnum];
1174 probe_procnum = procnum;
1176 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1177 probe_procnum = procnum;
1180 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1181 nd->nd_mreq, cred, probe_procnum, 0);
1185 m_freem(nd->nd_mreq);
1186 if (usegssname == 0)
1189 free(rep, M_NFSDREQ);
1191 newnfs_restore_sigmask(td, &oldset);
1194 m_freem(nd->nd_mrep);
1195 m_freem(nd->nd_mreq);
1196 if (usegssname == 0)
1199 free(rep, M_NFSDREQ);
1201 newnfs_restore_sigmask(td, &oldset);
1206 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1207 * wait for all requests to complete. This is used by forced unmounts
1208 * to terminate any outstanding RPCs.
1211 newnfs_nmcancelreqs(struct nfsmount *nmp)
1213 struct nfsclds *dsp;
1214 struct __rpc_client *cl;
1216 if (nmp->nm_sockreq.nr_client != NULL)
1217 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1220 TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1222 if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1223 (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1224 dsp->nfsclds_sockp != NULL &&
1225 dsp->nfsclds_sockp->nr_client != NULL) {
1226 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1227 cl = dsp->nfsclds_sockp->nr_client;
1240 * Any signal that can interrupt an NFS operation in an intr mount
1241 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1243 int newnfs_sig_set[] = {
1252 * Check to see if one of the signals in our subset is pending on
1253 * the process (in an intr mount).
1256 nfs_sig_pending(sigset_t set)
1260 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1261 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1267 * The set/restore sigmask functions are used to (temporarily) overwrite
1268 * the thread td_sigmask during an RPC call (for example). These are also
1269 * used in other places in the NFS client that might tsleep().
1272 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1280 td = curthread; /* XXX */
1282 /* Remove the NFS set of signals from newset */
1284 mtx_lock(&p->p_sigacts->ps_mtx);
1285 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1287 * But make sure we leave the ones already masked
1288 * by the process, ie. remove the signal from the
1289 * temporary signalmask only if it wasn't already
1292 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1293 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1294 SIGDELSET(newset, newnfs_sig_set[i]);
1296 mtx_unlock(&p->p_sigacts->ps_mtx);
1297 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1298 SIGPROCMASK_PROC_LOCKED);
1303 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1306 td = curthread; /* XXX */
1307 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1311 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1312 * old one after msleep() returns.
1315 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1320 if ((priority & PCATCH) == 0)
1321 return msleep(ident, mtx, priority, wmesg, timo);
1323 td = curthread; /* XXX */
1324 newnfs_set_sigmask(td, &oldset);
1325 error = msleep(ident, mtx, priority, wmesg, timo);
1326 newnfs_restore_sigmask(td, &oldset);
1331 * Test for a termination condition pending on the process.
1332 * This is used for NFSMNT_INT mounts.
1335 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1340 /* Terminate all requests while attempting a forced unmount. */
1341 if (NFSCL_FORCEDISM(nmp->nm_mountp))
1343 if (!(nmp->nm_flag & NFSMNT_INT))
1349 tmpset = p->p_siglist;
1350 SIGSETOR(tmpset, td->td_siglist);
1351 SIGSETNAND(tmpset, td->td_sigmask);
1352 mtx_lock(&p->p_sigacts->ps_mtx);
1353 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1354 mtx_unlock(&p->p_sigacts->ps_mtx);
1355 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1356 && nfs_sig_pending(tmpset)) {
1365 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1369 p = td ? td->td_proc : NULL;
1371 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1372 server, msg, error);
1374 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1380 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1381 int error, int flags)
1385 mtx_lock(&nmp->nm_mtx);
1386 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1387 nmp->nm_state |= NFSSTA_TIMEO;
1388 mtx_unlock(&nmp->nm_mtx);
1389 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1392 mtx_unlock(&nmp->nm_mtx);
1393 mtx_lock(&nmp->nm_mtx);
1394 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1395 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1396 mtx_unlock(&nmp->nm_mtx);
1397 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1400 mtx_unlock(&nmp->nm_mtx);
1401 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1405 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1406 int flags, int tprintfmsg)
1411 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1414 mtx_lock(&nmp->nm_mtx);
1415 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1416 nmp->nm_state &= ~NFSSTA_TIMEO;
1417 mtx_unlock(&nmp->nm_mtx);
1418 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1421 mtx_unlock(&nmp->nm_mtx);
1423 mtx_lock(&nmp->nm_mtx);
1424 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1425 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1426 mtx_unlock(&nmp->nm_mtx);
1427 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1430 mtx_unlock(&nmp->nm_mtx);