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)
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);
237 td->td_ucred = origcred;
241 if (error != 0 && pktscale > 2) {
242 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
243 pktscale == pktscalesav)
244 printf("Consider increasing kern.ipc.maxsockbuf\n");
247 if (nrp->nr_sotype == SOCK_DGRAM) {
249 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
251 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
254 sndreserve = rcvreserve = 1024 * pktscale;
257 if (nrp->nr_sotype != SOCK_STREAM)
258 panic("nfscon sotype");
260 sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
261 sizeof (u_int32_t)) * pktscale;
262 rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
263 sizeof (u_int32_t)) * pktscale;
265 sndreserve = rcvreserve = 1024 * pktscale;
268 error = soreserve(so, sndreserve, rcvreserve);
269 if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
271 printf("Must increase kern.ipc.maxsockbuf or reduce"
273 } while (error != 0 && pktscale > 2);
276 td->td_ucred = origcred;
280 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
281 nrp->nr_vers, sndreserve, rcvreserve);
282 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
284 if ((nmp->nm_flag & NFSMNT_INT))
285 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
286 if ((nmp->nm_flag & NFSMNT_RESVPORT))
287 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
288 if (NFSHASSOFT(nmp)) {
289 if (nmp->nm_sotype == SOCK_DGRAM)
291 * For UDP, the large timeout for a reconnect
292 * will be set to "nm_retry * nm_timeo / 2", so
293 * we only want to do 2 reconnect timeout
298 retries = nmp->nm_retry;
301 if (NFSHASNFSV4N(nmp)) {
303 if (NFSHASSOFT(nmp)) {
305 * This should be a DS mount.
306 * Use CLSET_TIMEOUT to set the timeout
307 * for connections to DSs instead of
308 * specifying a timeout on each RPC.
309 * This is done so that SO_SNDTIMEO
310 * is set on the TCP socket as well
311 * as specifying a time limit when
312 * waiting for an RPC reply. Useful
313 * if the send queue for the TCP
314 * connection has become constipated,
315 * due to a failed DS.
316 * The choice of lease_duration / 4 is
317 * fairly arbitrary, but seems to work
318 * ok, with a lower bound of 10sec.
320 timo.tv_sec = nfsrv_lease / 4;
321 if (timo.tv_sec < 10)
324 CLNT_CONTROL(client, CLSET_TIMEOUT,
328 * Make sure the nfscbd_pool doesn't get
329 * destroyed while doing this.
332 if (nfs_numnfscbd > 0) {
335 xprt = svc_vc_create_backchannel(
337 CLNT_CONTROL(client, CLSET_BACKCHANNEL,
341 if (nfs_numnfscbd == 0)
342 wakeup(&nfs_numnfscbd);
347 * cred == NULL for a DS connect.
348 * For connects to a DS, set a retry limit
349 * so that failed DSs will be detected.
350 * This is ok for NFSv4.1, since a DS does
351 * not maintain open/lock state and is the
352 * only case where using a "soft" mount is
353 * recommended for NFSv4.
354 * For mounts from the MDS to DS, this is done
355 * via mount options, but that is not the case
356 * here. The retry limit here can be adjusted
357 * via the sysctl vfs.nfs.dsretries.
358 * See the comment above w.r.t. timeout.
360 timo.tv_sec = nfsrv_lease / 4;
361 if (timo.tv_sec < 10)
364 CLNT_CONTROL(client, CLSET_TIMEOUT, &timo);
365 retries = nfs_dsretries;
371 * - Null RPC callback to client
372 * - Non-Null RPC callback to client, wait a little longer
373 * - upcalls to nfsuserd and gssd (clp == NULL)
375 if (callback_retry_mult == 0) {
376 retries = NFSV4_UPCALLRETRY;
377 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
379 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
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;
436 * NFS disconnect. Clean up and unlink.
439 newnfs_disconnect(struct nfssockreq *nrp)
443 mtx_lock(&nrp->nr_mtx);
444 if (nrp->nr_client != NULL) {
445 client = nrp->nr_client;
446 nrp->nr_client = NULL;
447 mtx_unlock(&nrp->nr_mtx);
448 rpc_gss_secpurge_call(client);
450 CLNT_RELEASE(client);
452 mtx_unlock(&nrp->nr_mtx);
457 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
458 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
460 rpc_gss_service_t svc;
463 switch (secflavour) {
464 case RPCSEC_GSS_KRB5:
465 case RPCSEC_GSS_KRB5I:
466 case RPCSEC_GSS_KRB5P:
468 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
471 if (secflavour == RPCSEC_GSS_KRB5)
472 svc = rpc_gss_svc_none;
473 else if (secflavour == RPCSEC_GSS_KRB5I)
474 svc = rpc_gss_svc_integrity;
476 svc = rpc_gss_svc_privacy;
478 if (clnt_principal == NULL)
479 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
480 srv_principal, mech_oid, svc);
482 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
483 clnt_principal, srv_principal, "kerberosv5",
484 svc, NULL, NULL, NULL);
492 return (authunix_create(cred));
498 * Callback from the RPC code to generate up/down notifications.
501 struct nfs_feedback_arg {
502 struct nfsmount *nf_mount;
503 int nf_lastmsg; /* last tprintf */
505 struct thread *nf_td;
509 nfs_feedback(int type, int proc, void *arg)
511 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
512 struct nfsmount *nmp = nf->nf_mount;
516 case FEEDBACK_REXMIT2:
517 case FEEDBACK_RECONNECT:
519 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
520 nfs_down(nmp, nf->nf_td,
521 "not responding", 0, NFSSTA_TIMEO);
522 nf->nf_tprintfmsg = TRUE;
523 nf->nf_lastmsg = now;
528 nfs_up(nf->nf_mount, nf->nf_td,
529 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
535 * newnfs_request - goes something like this
536 * - does the rpc by calling the krpc layer
537 * - break down rpc header and return with nfs reply
538 * nb: always frees up nd_mreq mbuf list
541 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
542 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
543 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
544 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
546 uint32_t retseq, retval, slotseq, *tl;
548 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
549 int error = 0, usegssname = 0, secflavour = AUTH_SYS;
550 int freeslot, maxslot, reterr, slotpos, timeo;
552 u_int trylater_delay = 1;
553 struct nfs_feedback_arg nf;
556 struct rpc_callextra ext;
558 struct nfsreq *rep = NULL;
559 char *srv_principal = NULL, *clnt_principal = NULL;
561 struct ucred *authcred;
562 struct nfsclsession *sep;
563 uint8_t sessionid[NFSX_V4SESSIONID];
568 /* Reject requests while attempting a forced unmount. */
569 if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
570 m_freem(nd->nd_mreq);
575 * Set authcred, which is used to acquire RPC credentials to
576 * the cred argument, by default. The crhold() should not be
577 * necessary, but will ensure that some future code change
578 * doesn't result in the credential being free'd prematurely.
580 authcred = crhold(cred);
582 /* For client side interruptible mounts, mask off the signals. */
583 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
584 newnfs_set_sigmask(td, &oldset);
589 * XXX if not already connected call nfs_connect now. Longer
590 * term, change nfs_mount to call nfs_connect unconditionally
591 * and let clnt_reconnect_create handle reconnects.
593 if (nrp->nr_client == NULL)
594 newnfs_connect(nmp, nrp, cred, td, 0);
597 * For a client side mount, nmp is != NULL and clp == NULL. For
598 * server calls (callbacks or upcalls), nmp == NULL.
602 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
603 secflavour = RPCSEC_GSS_KRB5;
604 if (nd->nd_procnum != NFSPROC_NULL) {
605 if (clp->lc_flags & LCL_GSSINTEGRITY)
606 secflavour = RPCSEC_GSS_KRB5I;
607 else if (clp->lc_flags & LCL_GSSPRIVACY)
608 secflavour = RPCSEC_GSS_KRB5P;
612 } else if (nmp != NULL && NFSHASKERB(nmp) &&
613 nd->nd_procnum != NFSPROC_NULL) {
614 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
615 nd->nd_flag |= ND_USEGSSNAME;
616 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
618 * If there is a client side host based credential,
619 * use that, otherwise use the system uid, if set.
620 * The system uid is in the nmp->nm_sockreq.nr_cred
623 if (nmp->nm_krbnamelen > 0) {
625 clnt_principal = nmp->nm_krbname;
626 } else if (nmp->nm_uid != (uid_t)-1) {
627 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
628 ("newnfs_request: NULL nr_cred"));
630 authcred = crhold(nmp->nm_sockreq.nr_cred);
632 } else if (nmp->nm_krbnamelen == 0 &&
633 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
635 * If there is no host based principal name and
636 * the system uid is set and this is root, use the
637 * system uid, since root won't have user
638 * credentials in a credentials cache file.
639 * The system uid is in the nmp->nm_sockreq.nr_cred
642 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
643 ("newnfs_request: NULL nr_cred"));
645 authcred = crhold(nmp->nm_sockreq.nr_cred);
647 if (NFSHASINTEGRITY(nmp))
648 secflavour = RPCSEC_GSS_KRB5I;
649 else if (NFSHASPRIVACY(nmp))
650 secflavour = RPCSEC_GSS_KRB5P;
652 secflavour = RPCSEC_GSS_KRB5;
653 srv_principal = NFSMNT_SRVKRBNAME(nmp);
654 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
655 nd->nd_procnum != NFSPROC_NULL &&
656 (nd->nd_flag & ND_USEGSSNAME) != 0) {
658 * Use the uid that did the mount when the RPC is doing
659 * NFSv4 system operations, as indicated by the
660 * ND_USEGSSNAME flag, for the AUTH_SYS case.
661 * The credentials in nm_sockreq.nr_cred were used for the
664 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
665 ("newnfs_request: NULL nr_cred"));
667 authcred = crhold(nmp->nm_sockreq.nr_cred);
671 bzero(&nf, sizeof(struct nfs_feedback_arg));
674 nf.nf_lastmsg = NFSD_MONOSEC -
675 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
678 if (nd->nd_procnum == NFSPROC_NULL)
679 auth = authnone_create();
680 else if (usegssname) {
682 * For this case, the authenticator is held in the
683 * nfssockreq structure, so don't release the reference count
684 * held on it. --> Don't AUTH_DESTROY() it in this function.
686 if (nrp->nr_auth == NULL)
687 nrp->nr_auth = nfs_getauth(nrp, secflavour,
688 clnt_principal, srv_principal, NULL, authcred);
690 rpc_gss_refresh_auth_call(nrp->nr_auth);
693 auth = nfs_getauth(nrp, secflavour, NULL,
694 srv_principal, NULL, authcred);
697 m_freem(nd->nd_mreq);
699 newnfs_restore_sigmask(td, &oldset);
702 bzero(&ext, sizeof(ext));
705 ext.rc_feedback = nfs_feedback;
706 ext.rc_feedback_arg = &nf;
709 procnum = nd->nd_procnum;
710 if ((nd->nd_flag & ND_NFSV4) &&
711 nd->nd_procnum != NFSPROC_NULL &&
712 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
713 procnum = NFSV4PROC_COMPOUND;
716 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
718 /* Map the procnum to the old NFSv2 one, as required. */
719 if ((nd->nd_flag & ND_NFSV2) != 0) {
720 if (nd->nd_procnum < NFS_V3NPROCS)
721 procnum = nfsv2_procid[nd->nd_procnum];
723 procnum = NFSV2PROC_NOOP;
727 * Now only used for the R_DONTRECOVER case, but until that is
728 * supported within the krpc code, I need to keep a queue of
729 * outstanding RPCs for nfsv4 client requests.
731 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
732 rep = malloc(sizeof(struct nfsreq),
733 M_NFSDREQ, M_WAITOK);
735 if (dtrace_nfscl_nfs234_start_probe != NULL) {
739 if (nd->nd_flag & ND_NFSV4) {
741 nfscl_nfs4_start_probes[nd->nd_procnum];
742 probe_procnum = nd->nd_procnum;
743 } else if (nd->nd_flag & ND_NFSV3) {
744 probe_id = nfscl_nfs3_start_probes[procnum];
745 probe_procnum = procnum;
748 nfscl_nfs2_start_probes[nd->nd_procnum];
749 probe_procnum = procnum;
752 (dtrace_nfscl_nfs234_start_probe)
753 (probe_id, vp, nd->nd_mreq, cred,
758 freeslot = -1; /* Set to slot that needs to be free'd */
760 slot = -1; /* Slot that needs a sequence# increment. */
762 * This timeout specifies when a new socket should be created,
763 * along with new xid values. For UDP, this should be done
764 * infrequently, since retransmits of RPC requests should normally
770 timo.tv_sec = NFSV4_UPCALLTIMEO;
772 timo.tv_sec = NFSV4_CALLBACKTIMEO;
774 if (nrp->nr_sotype != SOCK_DGRAM) {
776 if ((nmp->nm_flag & NFSMNT_NFSV4))
777 timo.tv_sec = INT_MAX;
779 timo.tv_sec = NFS_TCPTIMEO;
781 if (NFSHASSOFT(nmp)) {
783 * CLSET_RETRIES is set to 2, so this should be
784 * half of the total timeout required.
786 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
789 timo.tv_sec = timeo / NFS_HZ;
790 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
793 /* For UDP hard mounts, use a large value. */
794 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
803 * Chain request into list of outstanding requests.
806 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
812 if (clp != NULL && sep != NULL)
813 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
814 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
816 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
817 nd->nd_mreq, &nd->nd_mrep, timo);
818 NFSCL_DEBUG(2, "clnt call=%d\n", stat);
822 * RPC done, unlink the request.
825 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
830 * If there was a successful reply and a tprintf msg.
831 * tprintf a response.
833 if (stat == RPC_SUCCESS) {
835 } else if (stat == RPC_TIMEDOUT) {
836 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
838 } else if (stat == RPC_VERSMISMATCH) {
839 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
841 } else if (stat == RPC_PROGVERSMISMATCH) {
842 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
843 error = EPROTONOSUPPORT;
844 } else if (stat == RPC_INTR) {
846 } else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
847 stat == RPC_SYSTEMERROR) {
848 /* Check for a session slot that needs to be free'd. */
849 if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
850 (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
851 nd->nd_procnum != NFSPROC_NULL) {
853 * This should only occur when either the MDS or
854 * a client has an RPC against a DS fail.
855 * This happens because these cases use "soft"
856 * connections that can time out and fail.
857 * The slot used for this RPC is now in a
858 * non-deterministic state, but if the slot isn't
859 * free'd, threads can get stuck waiting for a slot.
862 sep = nfsmnt_mdssession(nmp);
864 * Bump the sequence# out of range, so that reuse of
865 * this slot will result in an NFSERR_SEQMISORDERED
866 * error and not a bogus cached RPC reply.
868 mtx_lock(&sep->nfsess_mtx);
869 sep->nfsess_slotseq[nd->nd_slotid] += 10;
870 mtx_unlock(&sep->nfsess_mtx);
871 /* And free the slot. */
872 nfsv4_freeslot(sep, nd->nd_slotid);
874 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
877 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
881 m_freem(nd->nd_mreq);
885 free(rep, M_NFSDREQ);
887 newnfs_restore_sigmask(td, &oldset);
891 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
894 * Search for any mbufs that are not a multiple of 4 bytes long
895 * or with m_data not longword aligned.
896 * These could cause pointer alignment problems, so copy them to
897 * well aligned mbufs.
899 newnfs_realign(&nd->nd_mrep, M_WAITOK);
900 nd->nd_md = nd->nd_mrep;
901 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
903 if (nd->nd_procnum != NFSPROC_NULL &&
904 nd->nd_procnum != NFSV4PROC_CBNULL) {
905 /* If sep == NULL, set it to the default in nmp. */
906 if (sep == NULL && nmp != NULL)
907 sep = nfsmnt_mdssession(nmp);
909 * and now the actual NFS xdr.
911 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
912 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
913 if (nd->nd_repstat >= 10000)
914 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
915 (int)nd->nd_repstat);
918 * Get rid of the tag, return count and SEQUENCE result for
921 if ((nd->nd_flag & ND_NFSV4) != 0) {
922 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
923 i = fxdr_unsigned(int, *tl);
924 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
927 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
928 opcnt = fxdr_unsigned(int, *tl++);
929 i = fxdr_unsigned(int, *tl++);
930 j = fxdr_unsigned(int, *tl);
932 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
934 * If the first op is Sequence, free up the slot.
936 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
937 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
938 NFSCL_DEBUG(1, "failed seq=%d\n", j);
939 if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
940 (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
941 j == 0)) && sep != NULL) {
942 if (i == NFSV4OP_SEQUENCE)
943 NFSM_DISSECT(tl, uint32_t *,
947 NFSM_DISSECT(tl, uint32_t *,
950 mtx_lock(&sep->nfsess_mtx);
951 if (bcmp(tl, sep->nfsess_sessionid,
952 NFSX_V4SESSIONID) == 0) {
953 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
954 retseq = fxdr_unsigned(uint32_t, *tl++);
955 slot = fxdr_unsigned(int, *tl++);
957 if (retseq != sep->nfsess_slotseq[slot])
958 printf("retseq diff 0x%x\n",
960 retval = fxdr_unsigned(uint32_t, *++tl);
961 if ((retval + 1) < sep->nfsess_foreslots
963 sep->nfsess_foreslots = (retval
965 else if ((retval + 1) >
966 sep->nfsess_foreslots)
967 sep->nfsess_foreslots = (retval
968 < 64) ? (retval + 1) : 64;
970 mtx_unlock(&sep->nfsess_mtx);
972 /* Grab the op and status for the next one. */
974 NFSM_DISSECT(tl, uint32_t *,
976 i = fxdr_unsigned(int, *tl++);
977 j = fxdr_unsigned(int, *tl);
981 if (nd->nd_repstat != 0) {
982 if (nd->nd_repstat == NFSERR_BADSESSION &&
983 nmp != NULL && dssep == NULL &&
984 (nd->nd_flag & ND_NFSV41) != 0) {
986 * If this is a client side MDS RPC, mark
987 * the MDS session defunct and initiate
988 * recovery, as required.
989 * The nfsess_defunct field is protected by
990 * the NFSLOCKMNT()/nm_mtx lock and not the
991 * nfsess_mtx lock to simplify its handling,
992 * for the MDS session. This lock is also
993 * sufficient for nfsess_sessionid, since it
994 * never changes in the structure.
996 NFSCL_DEBUG(1, "Got badsession\n");
999 sep = NFSMNT_MDSSESSION(nmp);
1000 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
1001 NFSX_V4SESSIONID) == 0) {
1002 /* Initiate recovery. */
1003 sep->nfsess_defunct = 1;
1004 NFSCL_DEBUG(1, "Marked defunct\n");
1005 if (nmp->nm_clp != NULL) {
1006 nmp->nm_clp->nfsc_flags |=
1008 wakeup(nmp->nm_clp);
1013 * Sleep for up to 1sec waiting for a new
1016 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1019 * Get the session again, in case a new one
1020 * has been created during the sleep.
1022 sep = NFSMNT_MDSSESSION(nmp);
1024 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1025 reterr = nfsv4_sequencelookup(nmp, sep,
1026 &slotpos, &maxslot, &slotseq,
1029 /* Fill in new session info. */
1031 "Filling in new sequence\n");
1032 tl = nd->nd_sequence;
1033 bcopy(sessionid, tl,
1035 tl += NFSX_V4SESSIONID /
1037 *tl++ = txdr_unsigned(slotseq);
1038 *tl++ = txdr_unsigned(slotpos);
1039 *tl = txdr_unsigned(maxslot);
1041 if (reterr == NFSERR_BADSESSION ||
1044 "Badsession looping\n");
1045 m_freem(nd->nd_mrep);
1049 nd->nd_repstat = reterr;
1050 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1054 * When clp != NULL, it is a callback and all
1055 * callback operations can be retried for NFSERR_DELAY.
1057 if (((nd->nd_repstat == NFSERR_DELAY ||
1058 nd->nd_repstat == NFSERR_GRACE) &&
1059 (nd->nd_flag & ND_NFSV4) && (clp != NULL ||
1060 (nd->nd_procnum != NFSPROC_DELEGRETURN &&
1061 nd->nd_procnum != NFSPROC_SETATTR &&
1062 nd->nd_procnum != NFSPROC_READ &&
1063 nd->nd_procnum != NFSPROC_READDS &&
1064 nd->nd_procnum != NFSPROC_WRITE &&
1065 nd->nd_procnum != NFSPROC_WRITEDS &&
1066 nd->nd_procnum != NFSPROC_OPEN &&
1067 nd->nd_procnum != NFSPROC_CREATE &&
1068 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1069 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1070 nd->nd_procnum != NFSPROC_CLOSE &&
1071 nd->nd_procnum != NFSPROC_LOCK &&
1072 nd->nd_procnum != NFSPROC_LOCKU))) ||
1073 (nd->nd_repstat == NFSERR_DELAY &&
1074 (nd->nd_flag & ND_NFSV4) == 0) ||
1075 nd->nd_repstat == NFSERR_RESOURCE) {
1076 if (trylater_delay > NFS_TRYLATERDEL)
1077 trylater_delay = NFS_TRYLATERDEL;
1078 waituntil = NFSD_MONOSEC + trylater_delay;
1079 while (NFSD_MONOSEC < waituntil)
1080 (void) nfs_catnap(PZERO, 0, "nfstry");
1081 trylater_delay *= 2;
1083 mtx_lock(&sep->nfsess_mtx);
1084 sep->nfsess_slotseq[slot]++;
1085 *nd->nd_slotseq = txdr_unsigned(
1086 sep->nfsess_slotseq[slot]);
1087 mtx_unlock(&sep->nfsess_mtx);
1089 m_freem(nd->nd_mrep);
1095 * If the File Handle was stale, invalidate the
1096 * lookup cache, just in case.
1097 * (vp != NULL implies a client side call)
1099 if (nd->nd_repstat == ESTALE && vp != NULL) {
1101 if (ncl_call_invalcaches != NULL)
1102 (*ncl_call_invalcaches)(vp);
1105 if ((nd->nd_flag & ND_NFSV4) != 0) {
1106 /* Free the slot, as required. */
1108 nfsv4_freeslot(sep, freeslot);
1110 * If this op is Putfh, throw its results away.
1113 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1114 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1115 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1116 i = fxdr_unsigned(int, *tl++);
1117 j = fxdr_unsigned(int, *tl);
1119 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1122 * All Compounds that do an Op that must
1123 * be in sequence consist of NFSV4OP_PUTFH
1124 * followed by one of these. As such, we
1125 * can determine if the seqid# should be
1126 * incremented, here.
1128 if ((i == NFSV4OP_OPEN ||
1129 i == NFSV4OP_OPENCONFIRM ||
1130 i == NFSV4OP_OPENDOWNGRADE ||
1131 i == NFSV4OP_CLOSE ||
1132 i == NFSV4OP_LOCK ||
1133 i == NFSV4OP_LOCKU) &&
1135 (j != NFSERR_STALECLIENTID &&
1136 j != NFSERR_STALESTATEID &&
1137 j != NFSERR_BADSTATEID &&
1138 j != NFSERR_BADSEQID &&
1139 j != NFSERR_BADXDR &&
1140 j != NFSERR_RESOURCE &&
1141 j != NFSERR_NOFILEHANDLE)))
1142 nd->nd_flag |= ND_INCRSEQID;
1145 * If this op's status is non-zero, mark
1146 * that there is no more data to process.
1147 * The exception is Setattr, which always has xdr
1148 * when it has failed.
1150 if (j != 0 && i != NFSV4OP_SETATTR)
1151 nd->nd_flag |= ND_NOMOREDATA;
1154 * If R_DONTRECOVER is set, replace the stale error
1155 * reply, so that recovery isn't initiated.
1157 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1158 nd->nd_repstat == NFSERR_BADSESSION ||
1159 nd->nd_repstat == NFSERR_STALESTATEID) &&
1160 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1161 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1165 #ifdef KDTRACE_HOOKS
1166 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1170 if (nd->nd_flag & ND_NFSV4) {
1171 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1172 probe_procnum = nd->nd_procnum;
1173 } else if (nd->nd_flag & ND_NFSV3) {
1174 probe_id = nfscl_nfs3_done_probes[procnum];
1175 probe_procnum = procnum;
1177 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1178 probe_procnum = procnum;
1181 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1182 nd->nd_mreq, cred, probe_procnum, 0);
1186 m_freem(nd->nd_mreq);
1187 if (usegssname == 0)
1190 free(rep, M_NFSDREQ);
1192 newnfs_restore_sigmask(td, &oldset);
1195 mbuf_freem(nd->nd_mrep);
1196 mbuf_freem(nd->nd_mreq);
1197 if (usegssname == 0)
1200 free(rep, M_NFSDREQ);
1202 newnfs_restore_sigmask(td, &oldset);
1207 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1208 * wait for all requests to complete. This is used by forced unmounts
1209 * to terminate any outstanding RPCs.
1212 newnfs_nmcancelreqs(struct nfsmount *nmp)
1214 struct nfsclds *dsp;
1215 struct __rpc_client *cl;
1217 if (nmp->nm_sockreq.nr_client != NULL)
1218 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1221 TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1223 if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1224 (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1225 dsp->nfsclds_sockp != NULL &&
1226 dsp->nfsclds_sockp->nr_client != NULL) {
1227 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1228 cl = dsp->nfsclds_sockp->nr_client;
1241 * Any signal that can interrupt an NFS operation in an intr mount
1242 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1244 int newnfs_sig_set[] = {
1253 * Check to see if one of the signals in our subset is pending on
1254 * the process (in an intr mount).
1257 nfs_sig_pending(sigset_t set)
1261 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1262 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1268 * The set/restore sigmask functions are used to (temporarily) overwrite
1269 * the thread td_sigmask during an RPC call (for example). These are also
1270 * used in other places in the NFS client that might tsleep().
1273 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1281 td = curthread; /* XXX */
1283 /* Remove the NFS set of signals from newset */
1285 mtx_lock(&p->p_sigacts->ps_mtx);
1286 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1288 * But make sure we leave the ones already masked
1289 * by the process, ie. remove the signal from the
1290 * temporary signalmask only if it wasn't already
1293 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1294 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1295 SIGDELSET(newset, newnfs_sig_set[i]);
1297 mtx_unlock(&p->p_sigacts->ps_mtx);
1298 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1299 SIGPROCMASK_PROC_LOCKED);
1304 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1307 td = curthread; /* XXX */
1308 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1312 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1313 * old one after msleep() returns.
1316 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1321 if ((priority & PCATCH) == 0)
1322 return msleep(ident, mtx, priority, wmesg, timo);
1324 td = curthread; /* XXX */
1325 newnfs_set_sigmask(td, &oldset);
1326 error = msleep(ident, mtx, priority, wmesg, timo);
1327 newnfs_restore_sigmask(td, &oldset);
1332 * Test for a termination condition pending on the process.
1333 * This is used for NFSMNT_INT mounts.
1336 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1341 /* Terminate all requests while attempting a forced unmount. */
1342 if (NFSCL_FORCEDISM(nmp->nm_mountp))
1344 if (!(nmp->nm_flag & NFSMNT_INT))
1350 tmpset = p->p_siglist;
1351 SIGSETOR(tmpset, td->td_siglist);
1352 SIGSETNAND(tmpset, td->td_sigmask);
1353 mtx_lock(&p->p_sigacts->ps_mtx);
1354 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1355 mtx_unlock(&p->p_sigacts->ps_mtx);
1356 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1357 && nfs_sig_pending(tmpset)) {
1366 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1370 p = td ? td->td_proc : NULL;
1372 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1373 server, msg, error);
1375 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1381 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1382 int error, int flags)
1386 mtx_lock(&nmp->nm_mtx);
1387 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1388 nmp->nm_state |= NFSSTA_TIMEO;
1389 mtx_unlock(&nmp->nm_mtx);
1390 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1393 mtx_unlock(&nmp->nm_mtx);
1394 mtx_lock(&nmp->nm_mtx);
1395 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1396 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1397 mtx_unlock(&nmp->nm_mtx);
1398 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1401 mtx_unlock(&nmp->nm_mtx);
1402 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1406 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1407 int flags, int tprintfmsg)
1412 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1415 mtx_lock(&nmp->nm_mtx);
1416 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1417 nmp->nm_state &= ~NFSSTA_TIMEO;
1418 mtx_unlock(&nmp->nm_mtx);
1419 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1422 mtx_unlock(&nmp->nm_mtx);
1424 mtx_lock(&nmp->nm_mtx);
1425 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1426 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1427 mtx_unlock(&nmp->nm_mtx);
1428 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1431 mtx_unlock(&nmp->nm_mtx);