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 int nfs_bufpackets = 4;
105 static int nfsrv_gsscallbackson = 0;
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;
186 * We need to establish the socket using the credentials of
187 * the mountpoint. Some parts of this process (such as
188 * sobind() and soconnect()) will use the curent thread's
189 * credential instead of the socket credential. To work
190 * around this, temporarily change the current thread's
191 * credential to that of the mountpoint.
193 * XXX: It would be better to explicitly pass the correct
194 * credential to sobind() and soconnect().
196 origcred = td->td_ucred;
199 * Use the credential in nr_cred, if not NULL.
201 if (nrp->nr_cred != NULL)
202 td->td_ucred = nrp->nr_cred;
207 if (saddr->sa_family == AF_INET)
208 if (nrp->nr_sotype == SOCK_DGRAM)
209 nconf = getnetconfigent("udp");
211 nconf = getnetconfigent("tcp");
213 if (nrp->nr_sotype == SOCK_DGRAM)
214 nconf = getnetconfigent("udp6");
216 nconf = getnetconfigent("tcp6");
218 pktscale = nfs_bufpackets;
223 pktscalesav = pktscale;
225 * soreserve() can fail if sb_max is too small, so shrink pktscale
226 * and try again if there is an error.
227 * Print a log message suggesting increasing sb_max.
228 * Creating a socket and doing this is necessary since, if the
229 * reservation sizes are too large and will make soreserve() fail,
230 * the connection will work until a large send is attempted and
231 * then it will loop in the krpc code.
234 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
235 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
236 nrp->nr_soproto, td->td_ucred, td);
240 if (error != 0 && pktscale > 2) {
241 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
242 pktscale == pktscalesav) {
244 * Suggest vfs.nfs.bufpackets * maximum RPC message,
245 * adjusted for the sb_max->sb_max_adj conversion of
246 * MCLBYTES / (MSIZE + MCLBYTES) as the minimum setting
247 * for kern.ipc.maxsockbuf.
249 tval = (NFS_MAXBSIZE + NFS_MAXXDR) * nfs_bufpackets;
250 tval *= MSIZE + MCLBYTES;
251 tval += MCLBYTES - 1; /* Round up divide by MCLBYTES. */
253 printf("Consider increasing kern.ipc.maxsockbuf to a "
254 "minimum of %ju to support %ubyte NFS I/O\n",
255 (uintmax_t)tval, NFS_MAXBSIZE);
259 if (nrp->nr_sotype == SOCK_DGRAM) {
261 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
263 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
266 sndreserve = rcvreserve = 1024 * pktscale;
269 if (nrp->nr_sotype != SOCK_STREAM)
270 panic("nfscon sotype");
272 sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR) *
274 rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR) *
277 sndreserve = rcvreserve = 1024 * pktscale;
280 error = soreserve(so, sndreserve, rcvreserve);
281 if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
283 printf("Must increase kern.ipc.maxsockbuf or reduce"
285 } while (error != 0 && pktscale > 2);
290 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
291 nrp->nr_vers, sndreserve, rcvreserve);
292 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
294 if ((nmp->nm_flag & NFSMNT_INT))
295 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
296 if ((nmp->nm_flag & NFSMNT_RESVPORT))
297 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
298 if (NFSHASTLS(nmp)) {
299 CLNT_CONTROL(client, CLSET_TLS, &one);
300 if (nmp->nm_tlscertname != NULL)
301 CLNT_CONTROL(client, CLSET_TLSCERTNAME,
302 nmp->nm_tlscertname);
304 if (NFSHASSOFT(nmp)) {
305 if (nmp->nm_sotype == SOCK_DGRAM)
307 * For UDP, the large timeout for a reconnect
308 * will be set to "nm_retry * nm_timeo / 2", so
309 * we only want to do 2 reconnect timeout
314 retries = nmp->nm_retry;
317 if (NFSHASNFSV4N(nmp)) {
319 if (NFSHASSOFT(nmp)) {
321 * This should be a DS mount.
322 * Use CLSET_TIMEOUT to set the timeout
323 * for connections to DSs instead of
324 * specifying a timeout on each RPC.
325 * This is done so that SO_SNDTIMEO
326 * is set on the TCP socket as well
327 * as specifying a time limit when
328 * waiting for an RPC reply. Useful
329 * if the send queue for the TCP
330 * connection has become constipated,
331 * due to a failed DS.
332 * The choice of lease_duration / 4 is
333 * fairly arbitrary, but seems to work
334 * ok, with a lower bound of 10sec.
336 timo.tv_sec = nfsrv_lease / 4;
337 if (timo.tv_sec < 10)
340 CLNT_CONTROL(client, CLSET_TIMEOUT,
344 * Make sure the nfscbd_pool doesn't get
345 * destroyed while doing this.
348 if (nfs_numnfscbd > 0) {
351 xprt = svc_vc_create_backchannel(
353 CLNT_CONTROL(client, CLSET_BACKCHANNEL,
357 if (nfs_numnfscbd == 0)
358 wakeup(&nfs_numnfscbd);
363 * cred == NULL for a DS connect.
364 * For connects to a DS, set a retry limit
365 * so that failed DSs will be detected.
366 * This is ok for NFSv4.1, since a DS does
367 * not maintain open/lock state and is the
368 * only case where using a "soft" mount is
369 * recommended for NFSv4.
370 * For mounts from the MDS to DS, this is done
371 * via mount options, but that is not the case
372 * here. The retry limit here can be adjusted
373 * via the sysctl vfs.nfs.dsretries.
374 * See the comment above w.r.t. timeout.
376 timo.tv_sec = nfsrv_lease / 4;
377 if (timo.tv_sec < 10)
380 CLNT_CONTROL(client, CLSET_TIMEOUT, &timo);
381 retries = nfs_dsretries;
387 * - Null RPC callback to client
388 * - Non-Null RPC callback to client, wait a little longer
389 * - upcalls to nfsuserd and gssd (clp == NULL)
391 if (callback_retry_mult == 0) {
392 retries = NFSV4_UPCALLRETRY;
393 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
395 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
398 CLNT_CONTROL(client, CLSET_TLS, &one);
400 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
404 * For UDP, there are 2 timeouts:
405 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
406 * that does a retransmit of an RPC request using the same
407 * socket and xid. This is what you normally want to do,
408 * since NFS servers depend on "same xid" for their
409 * Duplicate Request Cache.
410 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
411 * retransmits on the same socket should fail and a fresh
412 * socket created. Each of these timeouts counts as one
413 * CLSET_RETRIES as set above.
414 * Set the initial retransmit timeout for UDP. This timeout
415 * doesn't exist for TCP and the following call just fails,
418 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
419 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
420 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
423 mtx_lock(&nrp->nr_mtx);
424 if (nrp->nr_client != NULL) {
425 mtx_unlock(&nrp->nr_mtx);
427 * Someone else already connected.
429 CLNT_RELEASE(client);
431 nrp->nr_client = client;
433 * Protocols that do not require connections may be optionally
434 * left unconnected for servers that reply from a port other
437 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
438 mtx_unlock(&nrp->nr_mtx);
439 CLNT_CONTROL(client, CLSET_CONNECT, &one);
441 mtx_unlock(&nrp->nr_mtx);
445 /* Restore current thread's credentials. */
446 td->td_ucred = origcred;
453 * NFS disconnect. Clean up and unlink.
456 newnfs_disconnect(struct nfssockreq *nrp)
460 mtx_lock(&nrp->nr_mtx);
461 if (nrp->nr_client != NULL) {
462 client = nrp->nr_client;
463 nrp->nr_client = NULL;
464 mtx_unlock(&nrp->nr_mtx);
465 rpc_gss_secpurge_call(client);
467 CLNT_RELEASE(client);
469 mtx_unlock(&nrp->nr_mtx);
474 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
475 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
477 rpc_gss_service_t svc;
480 switch (secflavour) {
481 case RPCSEC_GSS_KRB5:
482 case RPCSEC_GSS_KRB5I:
483 case RPCSEC_GSS_KRB5P:
485 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
488 if (secflavour == RPCSEC_GSS_KRB5)
489 svc = rpc_gss_svc_none;
490 else if (secflavour == RPCSEC_GSS_KRB5I)
491 svc = rpc_gss_svc_integrity;
493 svc = rpc_gss_svc_privacy;
495 if (clnt_principal == NULL)
496 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
497 srv_principal, mech_oid, svc);
499 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
500 clnt_principal, srv_principal, "kerberosv5",
501 svc, NULL, NULL, NULL);
509 return (authunix_create(cred));
514 * Callback from the RPC code to generate up/down notifications.
517 struct nfs_feedback_arg {
518 struct nfsmount *nf_mount;
519 int nf_lastmsg; /* last tprintf */
521 struct thread *nf_td;
525 nfs_feedback(int type, int proc, void *arg)
527 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
528 struct nfsmount *nmp = nf->nf_mount;
532 case FEEDBACK_REXMIT2:
533 case FEEDBACK_RECONNECT:
535 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
536 nfs_down(nmp, nf->nf_td,
537 "not responding", 0, NFSSTA_TIMEO);
538 nf->nf_tprintfmsg = TRUE;
539 nf->nf_lastmsg = now;
544 nfs_up(nf->nf_mount, nf->nf_td,
545 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
551 * newnfs_request - goes something like this
552 * - does the rpc by calling the krpc layer
553 * - break down rpc header and return with nfs reply
554 * nb: always frees up nd_mreq mbuf list
557 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
558 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
559 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
560 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
562 uint32_t retseq, retval, slotseq, *tl;
564 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
565 int error = 0, usegssname = 0, secflavour = AUTH_SYS;
566 int freeslot, maxslot, reterr, slotpos, timeo;
568 u_int trylater_delay = 1;
569 struct nfs_feedback_arg nf;
572 struct rpc_callextra ext;
574 struct nfsreq *rep = NULL;
575 char *srv_principal = NULL, *clnt_principal = NULL;
577 struct ucred *authcred;
578 struct nfsclsession *sep;
579 uint8_t sessionid[NFSX_V4SESSIONID];
584 /* Reject requests while attempting a forced unmount. */
585 if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
586 m_freem(nd->nd_mreq);
591 * Set authcred, which is used to acquire RPC credentials to
592 * the cred argument, by default. The crhold() should not be
593 * necessary, but will ensure that some future code change
594 * doesn't result in the credential being free'd prematurely.
596 authcred = crhold(cred);
598 /* For client side interruptible mounts, mask off the signals. */
599 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
600 newnfs_set_sigmask(td, &oldset);
605 * XXX if not already connected call nfs_connect now. Longer
606 * term, change nfs_mount to call nfs_connect unconditionally
607 * and let clnt_reconnect_create handle reconnects.
609 if (nrp->nr_client == NULL)
610 newnfs_connect(nmp, nrp, cred, td, 0, false);
613 * For a client side mount, nmp is != NULL and clp == NULL. For
614 * server calls (callbacks or upcalls), nmp == NULL.
618 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
619 secflavour = RPCSEC_GSS_KRB5;
620 if (nd->nd_procnum != NFSPROC_NULL) {
621 if (clp->lc_flags & LCL_GSSINTEGRITY)
622 secflavour = RPCSEC_GSS_KRB5I;
623 else if (clp->lc_flags & LCL_GSSPRIVACY)
624 secflavour = RPCSEC_GSS_KRB5P;
628 } else if (nmp != NULL && NFSHASKERB(nmp) &&
629 nd->nd_procnum != NFSPROC_NULL) {
630 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
631 nd->nd_flag |= ND_USEGSSNAME;
632 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
634 * If there is a client side host based credential,
635 * use that, otherwise use the system uid, if set.
636 * The system uid is in the nmp->nm_sockreq.nr_cred
639 if (nmp->nm_krbnamelen > 0) {
641 clnt_principal = nmp->nm_krbname;
642 } else if (nmp->nm_uid != (uid_t)-1) {
643 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
644 ("newnfs_request: NULL nr_cred"));
646 authcred = crhold(nmp->nm_sockreq.nr_cred);
648 } else if (nmp->nm_krbnamelen == 0 &&
649 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
651 * If there is no host based principal name and
652 * the system uid is set and this is root, use the
653 * system uid, since root won't have user
654 * credentials in a credentials cache file.
655 * The system uid is in the nmp->nm_sockreq.nr_cred
658 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
659 ("newnfs_request: NULL nr_cred"));
661 authcred = crhold(nmp->nm_sockreq.nr_cred);
663 if (NFSHASINTEGRITY(nmp))
664 secflavour = RPCSEC_GSS_KRB5I;
665 else if (NFSHASPRIVACY(nmp))
666 secflavour = RPCSEC_GSS_KRB5P;
668 secflavour = RPCSEC_GSS_KRB5;
669 srv_principal = NFSMNT_SRVKRBNAME(nmp);
670 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
671 nd->nd_procnum != NFSPROC_NULL &&
672 (nd->nd_flag & ND_USEGSSNAME) != 0) {
674 * Use the uid that did the mount when the RPC is doing
675 * NFSv4 system operations, as indicated by the
676 * ND_USEGSSNAME flag, for the AUTH_SYS case.
677 * The credentials in nm_sockreq.nr_cred were used for the
680 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
681 ("newnfs_request: NULL nr_cred"));
683 authcred = crhold(nmp->nm_sockreq.nr_cred);
687 bzero(&nf, sizeof(struct nfs_feedback_arg));
690 nf.nf_lastmsg = NFSD_MONOSEC -
691 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
694 if (nd->nd_procnum == NFSPROC_NULL)
695 auth = authnone_create();
696 else if (usegssname) {
698 * For this case, the authenticator is held in the
699 * nfssockreq structure, so don't release the reference count
700 * held on it. --> Don't AUTH_DESTROY() it in this function.
702 if (nrp->nr_auth == NULL)
703 nrp->nr_auth = nfs_getauth(nrp, secflavour,
704 clnt_principal, srv_principal, NULL, authcred);
706 rpc_gss_refresh_auth_call(nrp->nr_auth);
709 auth = nfs_getauth(nrp, secflavour, NULL,
710 srv_principal, NULL, authcred);
713 m_freem(nd->nd_mreq);
715 newnfs_restore_sigmask(td, &oldset);
718 bzero(&ext, sizeof(ext));
721 ext.rc_feedback = nfs_feedback;
722 ext.rc_feedback_arg = &nf;
725 procnum = nd->nd_procnum;
726 if ((nd->nd_flag & ND_NFSV4) &&
727 nd->nd_procnum != NFSPROC_NULL &&
728 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
729 procnum = NFSV4PROC_COMPOUND;
732 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
734 /* Map the procnum to the old NFSv2 one, as required. */
735 if ((nd->nd_flag & ND_NFSV2) != 0) {
736 if (nd->nd_procnum < NFS_V3NPROCS)
737 procnum = nfsv2_procid[nd->nd_procnum];
739 procnum = NFSV2PROC_NOOP;
743 * Now only used for the R_DONTRECOVER case, but until that is
744 * supported within the krpc code, I need to keep a queue of
745 * outstanding RPCs for nfsv4 client requests.
747 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
748 rep = malloc(sizeof(struct nfsreq),
749 M_NFSDREQ, M_WAITOK);
751 if (dtrace_nfscl_nfs234_start_probe != NULL) {
755 if (nd->nd_flag & ND_NFSV4) {
757 nfscl_nfs4_start_probes[nd->nd_procnum];
758 probe_procnum = nd->nd_procnum;
759 } else if (nd->nd_flag & ND_NFSV3) {
760 probe_id = nfscl_nfs3_start_probes[procnum];
761 probe_procnum = procnum;
764 nfscl_nfs2_start_probes[nd->nd_procnum];
765 probe_procnum = procnum;
768 (dtrace_nfscl_nfs234_start_probe)
769 (probe_id, vp, nd->nd_mreq, cred,
774 freeslot = -1; /* Set to slot that needs to be free'd */
776 slot = -1; /* Slot that needs a sequence# increment. */
778 * This timeout specifies when a new socket should be created,
779 * along with new xid values. For UDP, this should be done
780 * infrequently, since retransmits of RPC requests should normally
785 timo.tv_sec = NFSV4_UPCALLTIMEO;
788 timo.tv_sec = NFSV4_CALLBACKTIMEO / 1000;
789 timo.tv_usec = NFSV4_CALLBACKTIMEO * 1000;
792 if (nrp->nr_sotype != SOCK_DGRAM) {
794 if ((nmp->nm_flag & NFSMNT_NFSV4))
795 timo.tv_sec = INT_MAX;
797 timo.tv_sec = NFS_TCPTIMEO;
799 if (NFSHASSOFT(nmp)) {
801 * CLSET_RETRIES is set to 2, so this should be
802 * half of the total timeout required.
804 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
807 timo.tv_sec = timeo / NFS_HZ;
808 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
811 /* For UDP hard mounts, use a large value. */
812 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
821 * Chain request into list of outstanding requests.
824 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
830 if (clp != NULL && sep != NULL)
831 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
832 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
834 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
835 nd->nd_mreq, &nd->nd_mrep, timo);
836 NFSCL_DEBUG(2, "clnt call=%d\n", stat);
840 * RPC done, unlink the request.
843 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
848 * If there was a successful reply and a tprintf msg.
849 * tprintf a response.
851 if (stat == RPC_SUCCESS) {
853 } else if (stat == RPC_TIMEDOUT) {
854 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
856 } else if (stat == RPC_VERSMISMATCH) {
857 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
859 } else if (stat == RPC_PROGVERSMISMATCH) {
860 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
861 error = EPROTONOSUPPORT;
862 } else if (stat == RPC_INTR) {
864 } else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
865 stat == RPC_SYSTEMERROR) {
866 /* Check for a session slot that needs to be free'd. */
867 if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
868 (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
869 nd->nd_procnum != NFSPROC_NULL) {
871 * This should only occur when either the MDS or
872 * a client has an RPC against a DS fail.
873 * This happens because these cases use "soft"
874 * connections that can time out and fail.
875 * The slot used for this RPC is now in a
876 * non-deterministic state, but if the slot isn't
877 * free'd, threads can get stuck waiting for a slot.
880 sep = nfsmnt_mdssession(nmp);
882 * Bump the sequence# out of range, so that reuse of
883 * this slot will result in an NFSERR_SEQMISORDERED
884 * error and not a bogus cached RPC reply.
886 mtx_lock(&sep->nfsess_mtx);
887 sep->nfsess_slotseq[nd->nd_slotid] += 10;
888 mtx_unlock(&sep->nfsess_mtx);
889 /* And free the slot. */
890 nfsv4_freeslot(sep, nd->nd_slotid, false);
892 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
895 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
899 m_freem(nd->nd_mreq);
903 free(rep, M_NFSDREQ);
905 newnfs_restore_sigmask(td, &oldset);
909 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
912 * Search for any mbufs that are not a multiple of 4 bytes long
913 * or with m_data not longword aligned.
914 * These could cause pointer alignment problems, so copy them to
915 * well aligned mbufs.
917 newnfs_realign(&nd->nd_mrep, M_WAITOK);
918 nd->nd_md = nd->nd_mrep;
919 nd->nd_dpos = mtod(nd->nd_md, caddr_t);
921 if (nd->nd_procnum != NFSPROC_NULL &&
922 nd->nd_procnum != NFSV4PROC_CBNULL) {
923 /* If sep == NULL, set it to the default in nmp. */
924 if (sep == NULL && nmp != NULL)
925 sep = nfsmnt_mdssession(nmp);
927 * and now the actual NFS xdr.
929 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
930 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
931 if (nd->nd_repstat >= 10000)
932 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
933 (int)nd->nd_repstat);
936 * Get rid of the tag, return count and SEQUENCE result for
939 if ((nd->nd_flag & ND_NFSV4) != 0 && nd->nd_repstat !=
940 NFSERR_MINORVERMISMATCH) {
941 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
942 i = fxdr_unsigned(int, *tl);
943 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
946 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
947 opcnt = fxdr_unsigned(int, *tl++);
948 i = fxdr_unsigned(int, *tl++);
949 j = fxdr_unsigned(int, *tl);
951 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
953 * If the first op is Sequence, free up the slot.
955 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
956 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
957 NFSCL_DEBUG(1, "failed seq=%d\n", j);
958 if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
959 (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
960 j == 0)) && sep != NULL) {
961 if (i == NFSV4OP_SEQUENCE)
962 NFSM_DISSECT(tl, uint32_t *,
966 NFSM_DISSECT(tl, uint32_t *,
969 mtx_lock(&sep->nfsess_mtx);
970 if (bcmp(tl, sep->nfsess_sessionid,
971 NFSX_V4SESSIONID) == 0) {
972 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
973 retseq = fxdr_unsigned(uint32_t, *tl++);
974 slot = fxdr_unsigned(int, *tl++);
976 if (retseq != sep->nfsess_slotseq[slot])
977 printf("retseq diff 0x%x\n",
979 retval = fxdr_unsigned(uint32_t, *++tl);
980 if ((retval + 1) < sep->nfsess_foreslots
982 sep->nfsess_foreslots = (retval
984 else if ((retval + 1) >
985 sep->nfsess_foreslots)
986 sep->nfsess_foreslots = (retval
987 < 64) ? (retval + 1) : 64;
989 mtx_unlock(&sep->nfsess_mtx);
991 /* Grab the op and status for the next one. */
993 NFSM_DISSECT(tl, uint32_t *,
995 i = fxdr_unsigned(int, *tl++);
996 j = fxdr_unsigned(int, *tl);
1000 if (nd->nd_repstat != 0) {
1001 if (nd->nd_repstat == NFSERR_BADSESSION &&
1002 nmp != NULL && dssep == NULL &&
1003 (nd->nd_flag & ND_NFSV41) != 0) {
1005 * If this is a client side MDS RPC, mark
1006 * the MDS session defunct and initiate
1007 * recovery, as required.
1008 * The nfsess_defunct field is protected by
1009 * the NFSLOCKMNT()/nm_mtx lock and not the
1010 * nfsess_mtx lock to simplify its handling,
1011 * for the MDS session. This lock is also
1012 * sufficient for nfsess_sessionid, since it
1013 * never changes in the structure.
1015 NFSCL_DEBUG(1, "Got badsession\n");
1018 sep = NFSMNT_MDSSESSION(nmp);
1019 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
1020 NFSX_V4SESSIONID) == 0) {
1021 /* Initiate recovery. */
1022 sep->nfsess_defunct = 1;
1023 NFSCL_DEBUG(1, "Marked defunct\n");
1024 if (nmp->nm_clp != NULL) {
1025 nmp->nm_clp->nfsc_flags |=
1027 wakeup(nmp->nm_clp);
1032 * Sleep for up to 1sec waiting for a new
1035 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1038 * Get the session again, in case a new one
1039 * has been created during the sleep.
1041 sep = NFSMNT_MDSSESSION(nmp);
1043 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1044 reterr = nfsv4_sequencelookup(nmp, sep,
1045 &slotpos, &maxslot, &slotseq,
1048 /* Fill in new session info. */
1050 "Filling in new sequence\n");
1051 tl = nd->nd_sequence;
1052 bcopy(sessionid, tl,
1054 tl += NFSX_V4SESSIONID /
1056 *tl++ = txdr_unsigned(slotseq);
1057 *tl++ = txdr_unsigned(slotpos);
1058 *tl = txdr_unsigned(maxslot);
1060 if (reterr == NFSERR_BADSESSION ||
1063 "Badsession looping\n");
1064 m_freem(nd->nd_mrep);
1068 nd->nd_repstat = reterr;
1069 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1073 * When clp != NULL, it is a callback and all
1074 * callback operations can be retried for NFSERR_DELAY.
1076 if (((nd->nd_repstat == NFSERR_DELAY ||
1077 nd->nd_repstat == NFSERR_GRACE) &&
1078 (nd->nd_flag & ND_NFSV4) && (clp != NULL ||
1079 (nd->nd_procnum != NFSPROC_DELEGRETURN &&
1080 nd->nd_procnum != NFSPROC_SETATTR &&
1081 nd->nd_procnum != NFSPROC_READ &&
1082 nd->nd_procnum != NFSPROC_READDS &&
1083 nd->nd_procnum != NFSPROC_WRITE &&
1084 nd->nd_procnum != NFSPROC_WRITEDS &&
1085 nd->nd_procnum != NFSPROC_OPEN &&
1086 nd->nd_procnum != NFSPROC_OPENLAYGET &&
1087 nd->nd_procnum != NFSPROC_CREATE &&
1088 nd->nd_procnum != NFSPROC_CREATELAYGET &&
1089 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1090 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1091 nd->nd_procnum != NFSPROC_CLOSE &&
1092 nd->nd_procnum != NFSPROC_LOCK &&
1093 nd->nd_procnum != NFSPROC_LOCKU))) ||
1094 (nd->nd_repstat == NFSERR_DELAY &&
1095 (nd->nd_flag & ND_NFSV4) == 0) ||
1096 nd->nd_repstat == NFSERR_RESOURCE) {
1097 if (trylater_delay > NFS_TRYLATERDEL)
1098 trylater_delay = NFS_TRYLATERDEL;
1099 waituntil = NFSD_MONOSEC + trylater_delay;
1100 while (NFSD_MONOSEC < waituntil)
1101 (void) nfs_catnap(PZERO, 0, "nfstry");
1102 trylater_delay *= 2;
1104 mtx_lock(&sep->nfsess_mtx);
1105 sep->nfsess_slotseq[slot]++;
1106 *nd->nd_slotseq = txdr_unsigned(
1107 sep->nfsess_slotseq[slot]);
1108 mtx_unlock(&sep->nfsess_mtx);
1110 m_freem(nd->nd_mrep);
1116 * If the File Handle was stale, invalidate the
1117 * lookup cache, just in case.
1118 * (vp != NULL implies a client side call)
1120 if (nd->nd_repstat == ESTALE && vp != NULL) {
1122 if (ncl_call_invalcaches != NULL)
1123 (*ncl_call_invalcaches)(vp);
1126 if ((nd->nd_flag & ND_NFSV4) != 0) {
1127 /* Free the slot, as required. */
1129 nfsv4_freeslot(sep, freeslot, false);
1131 * If this op is Putfh, throw its results away.
1134 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1135 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1136 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1137 i = fxdr_unsigned(int, *tl++);
1138 j = fxdr_unsigned(int, *tl);
1140 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1143 * All Compounds that do an Op that must
1144 * be in sequence consist of NFSV4OP_PUTFH
1145 * followed by one of these. As such, we
1146 * can determine if the seqid# should be
1147 * incremented, here.
1149 if ((i == NFSV4OP_OPEN ||
1150 i == NFSV4OP_OPENCONFIRM ||
1151 i == NFSV4OP_OPENDOWNGRADE ||
1152 i == NFSV4OP_CLOSE ||
1153 i == NFSV4OP_LOCK ||
1154 i == NFSV4OP_LOCKU) &&
1156 (j != NFSERR_STALECLIENTID &&
1157 j != NFSERR_STALESTATEID &&
1158 j != NFSERR_BADSTATEID &&
1159 j != NFSERR_BADSEQID &&
1160 j != NFSERR_BADXDR &&
1161 j != NFSERR_RESOURCE &&
1162 j != NFSERR_NOFILEHANDLE)))
1163 nd->nd_flag |= ND_INCRSEQID;
1166 * If this op's status is non-zero, mark
1167 * that there is no more data to process.
1168 * The exception is Setattr, which always has xdr
1169 * when it has failed.
1171 if (j != 0 && i != NFSV4OP_SETATTR)
1172 nd->nd_flag |= ND_NOMOREDATA;
1175 * If R_DONTRECOVER is set, replace the stale error
1176 * reply, so that recovery isn't initiated.
1178 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1179 nd->nd_repstat == NFSERR_BADSESSION ||
1180 nd->nd_repstat == NFSERR_STALESTATEID) &&
1181 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1182 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1186 #ifdef KDTRACE_HOOKS
1187 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1191 if (nd->nd_flag & ND_NFSV4) {
1192 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1193 probe_procnum = nd->nd_procnum;
1194 } else if (nd->nd_flag & ND_NFSV3) {
1195 probe_id = nfscl_nfs3_done_probes[procnum];
1196 probe_procnum = procnum;
1198 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1199 probe_procnum = procnum;
1202 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1203 nd->nd_mreq, cred, probe_procnum, 0);
1207 m_freem(nd->nd_mreq);
1208 if (usegssname == 0)
1211 free(rep, M_NFSDREQ);
1213 newnfs_restore_sigmask(td, &oldset);
1216 m_freem(nd->nd_mrep);
1217 m_freem(nd->nd_mreq);
1218 if (usegssname == 0)
1221 free(rep, M_NFSDREQ);
1223 newnfs_restore_sigmask(td, &oldset);
1228 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1229 * wait for all requests to complete. This is used by forced unmounts
1230 * to terminate any outstanding RPCs.
1233 newnfs_nmcancelreqs(struct nfsmount *nmp)
1235 struct nfsclds *dsp;
1236 struct __rpc_client *cl;
1238 if (nmp->nm_sockreq.nr_client != NULL)
1239 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1242 TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1244 if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1245 (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1246 dsp->nfsclds_sockp != NULL &&
1247 dsp->nfsclds_sockp->nr_client != NULL) {
1248 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1249 cl = dsp->nfsclds_sockp->nr_client;
1262 * Any signal that can interrupt an NFS operation in an intr mount
1263 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1265 int newnfs_sig_set[] = {
1274 * Check to see if one of the signals in our subset is pending on
1275 * the process (in an intr mount).
1278 nfs_sig_pending(sigset_t set)
1282 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1283 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1289 * The set/restore sigmask functions are used to (temporarily) overwrite
1290 * the thread td_sigmask during an RPC call (for example). These are also
1291 * used in other places in the NFS client that might tsleep().
1294 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1302 td = curthread; /* XXX */
1304 /* Remove the NFS set of signals from newset */
1306 mtx_lock(&p->p_sigacts->ps_mtx);
1307 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1309 * But make sure we leave the ones already masked
1310 * by the process, ie. remove the signal from the
1311 * temporary signalmask only if it wasn't already
1314 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1315 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1316 SIGDELSET(newset, newnfs_sig_set[i]);
1318 mtx_unlock(&p->p_sigacts->ps_mtx);
1319 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1320 SIGPROCMASK_PROC_LOCKED);
1325 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1328 td = curthread; /* XXX */
1329 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1333 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1334 * old one after msleep() returns.
1337 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1342 if ((priority & PCATCH) == 0)
1343 return msleep(ident, mtx, priority, wmesg, timo);
1345 td = curthread; /* XXX */
1346 newnfs_set_sigmask(td, &oldset);
1347 error = msleep(ident, mtx, priority, wmesg, timo);
1348 newnfs_restore_sigmask(td, &oldset);
1353 * Test for a termination condition pending on the process.
1354 * This is used for NFSMNT_INT mounts.
1357 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1362 /* Terminate all requests while attempting a forced unmount. */
1363 if (NFSCL_FORCEDISM(nmp->nm_mountp))
1365 if (!(nmp->nm_flag & NFSMNT_INT))
1371 tmpset = p->p_siglist;
1372 SIGSETOR(tmpset, td->td_siglist);
1373 SIGSETNAND(tmpset, td->td_sigmask);
1374 mtx_lock(&p->p_sigacts->ps_mtx);
1375 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1376 mtx_unlock(&p->p_sigacts->ps_mtx);
1377 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1378 && nfs_sig_pending(tmpset)) {
1387 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1391 p = td ? td->td_proc : NULL;
1393 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1394 server, msg, error);
1396 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1402 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1403 int error, int flags)
1407 mtx_lock(&nmp->nm_mtx);
1408 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1409 nmp->nm_state |= NFSSTA_TIMEO;
1410 mtx_unlock(&nmp->nm_mtx);
1411 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1414 mtx_unlock(&nmp->nm_mtx);
1415 mtx_lock(&nmp->nm_mtx);
1416 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1417 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1418 mtx_unlock(&nmp->nm_mtx);
1419 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1422 mtx_unlock(&nmp->nm_mtx);
1423 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1427 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1428 int flags, int tprintfmsg)
1433 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1436 mtx_lock(&nmp->nm_mtx);
1437 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1438 nmp->nm_state &= ~NFSSTA_TIMEO;
1439 mtx_unlock(&nmp->nm_mtx);
1440 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1443 mtx_unlock(&nmp->nm_mtx);
1445 mtx_lock(&nmp->nm_mtx);
1446 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1447 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1448 mtx_unlock(&nmp->nm_mtx);
1449 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1452 mtx_unlock(&nmp->nm_mtx);