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");
211 else if (saddr->sa_family == AF_LOCAL)
212 nconf = getnetconfigent("local");
214 if (nrp->nr_sotype == SOCK_DGRAM)
215 nconf = getnetconfigent("udp6");
217 nconf = getnetconfigent("tcp6");
219 pktscale = nfs_bufpackets;
224 pktscalesav = pktscale;
226 * soreserve() can fail if sb_max is too small, so shrink pktscale
227 * and try again if there is an error.
228 * Print a log message suggesting increasing sb_max.
229 * Creating a socket and doing this is necessary since, if the
230 * reservation sizes are too large and will make soreserve() fail,
231 * the connection will work until a large send is attempted and
232 * then it will loop in the krpc code.
235 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
236 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
237 nrp->nr_soproto, td->td_ucred, td);
239 td->td_ucred = origcred;
243 if (error != 0 && pktscale > 2) {
244 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
245 pktscale == pktscalesav)
246 printf("Consider increasing kern.ipc.maxsockbuf\n");
249 if (nrp->nr_sotype == SOCK_DGRAM) {
251 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
253 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
256 sndreserve = rcvreserve = 1024 * pktscale;
259 if (nrp->nr_sotype != SOCK_STREAM)
260 panic("nfscon sotype");
262 sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
263 sizeof (u_int32_t)) * pktscale;
264 rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
265 sizeof (u_int32_t)) * pktscale;
267 sndreserve = rcvreserve = 1024 * pktscale;
270 error = soreserve(so, sndreserve, rcvreserve);
271 if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
273 printf("Must increase kern.ipc.maxsockbuf or reduce"
275 } while (error != 0 && pktscale > 2);
278 td->td_ucred = origcred;
282 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
283 nrp->nr_vers, sndreserve, rcvreserve);
284 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
286 if ((nmp->nm_flag & NFSMNT_INT))
287 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
288 if ((nmp->nm_flag & NFSMNT_RESVPORT))
289 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
290 if (NFSHASSOFT(nmp)) {
291 if (nmp->nm_sotype == SOCK_DGRAM)
293 * For UDP, the large timeout for a reconnect
294 * will be set to "nm_retry * nm_timeo / 2", so
295 * we only want to do 2 reconnect timeout
300 retries = nmp->nm_retry;
303 if (NFSHASNFSV4N(nmp)) {
305 if (NFSHASSOFT(nmp)) {
307 * This should be a DS mount.
308 * Use CLSET_TIMEOUT to set the timeout
309 * for connections to DSs instead of
310 * specifying a timeout on each RPC.
311 * This is done so that SO_SNDTIMEO
312 * is set on the TCP socket as well
313 * as specifying a time limit when
314 * waiting for an RPC reply. Useful
315 * if the send queue for the TCP
316 * connection has become constipated,
317 * due to a failed DS.
318 * The choice of lease_duration / 4 is
319 * fairly arbitrary, but seems to work
320 * ok, with a lower bound of 10sec.
322 timo.tv_sec = nfsrv_lease / 4;
323 if (timo.tv_sec < 10)
326 CLNT_CONTROL(client, CLSET_TIMEOUT,
330 * Make sure the nfscbd_pool doesn't get
331 * destroyed while doing this.
334 if (nfs_numnfscbd > 0) {
337 xprt = svc_vc_create_backchannel(
339 CLNT_CONTROL(client, CLSET_BACKCHANNEL,
343 if (nfs_numnfscbd == 0)
344 wakeup(&nfs_numnfscbd);
349 * cred == NULL for a DS connect.
350 * For connects to a DS, set a retry limit
351 * so that failed DSs will be detected.
352 * This is ok for NFSv4.1, since a DS does
353 * not maintain open/lock state and is the
354 * only case where using a "soft" mount is
355 * recommended for NFSv4.
356 * For mounts from the MDS to DS, this is done
357 * via mount options, but that is not the case
358 * here. The retry limit here can be adjusted
359 * via the sysctl vfs.nfs.dsretries.
360 * See the comment above w.r.t. timeout.
362 timo.tv_sec = nfsrv_lease / 4;
363 if (timo.tv_sec < 10)
366 CLNT_CONTROL(client, CLSET_TIMEOUT, &timo);
367 retries = nfs_dsretries;
373 * - Null RPC callback to client
374 * - Non-Null RPC callback to client, wait a little longer
375 * - upcalls to nfsuserd and gssd (clp == NULL)
377 if (callback_retry_mult == 0) {
378 retries = NFSV4_UPCALLRETRY;
379 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
381 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
384 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
388 * For UDP, there are 2 timeouts:
389 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
390 * that does a retransmit of an RPC request using the same
391 * socket and xid. This is what you normally want to do,
392 * since NFS servers depend on "same xid" for their
393 * Duplicate Request Cache.
394 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
395 * retransmits on the same socket should fail and a fresh
396 * socket created. Each of these timeouts counts as one
397 * CLSET_RETRIES as set above.
398 * Set the initial retransmit timeout for UDP. This timeout
399 * doesn't exist for TCP and the following call just fails,
402 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
403 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
404 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
407 mtx_lock(&nrp->nr_mtx);
408 if (nrp->nr_client != NULL) {
409 mtx_unlock(&nrp->nr_mtx);
411 * Someone else already connected.
413 CLNT_RELEASE(client);
415 nrp->nr_client = client;
417 * Protocols that do not require connections may be optionally
418 * left unconnected for servers that reply from a port other
421 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
422 mtx_unlock(&nrp->nr_mtx);
423 CLNT_CONTROL(client, CLSET_CONNECT, &one);
425 mtx_unlock(&nrp->nr_mtx);
429 /* Restore current thread's credentials. */
430 td->td_ucred = origcred;
438 * NFS disconnect. Clean up and unlink.
441 newnfs_disconnect(struct nfssockreq *nrp)
445 mtx_lock(&nrp->nr_mtx);
446 if (nrp->nr_client != NULL) {
447 client = nrp->nr_client;
448 nrp->nr_client = NULL;
449 mtx_unlock(&nrp->nr_mtx);
450 rpc_gss_secpurge_call(client);
452 CLNT_RELEASE(client);
454 mtx_unlock(&nrp->nr_mtx);
459 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
460 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
462 rpc_gss_service_t svc;
465 switch (secflavour) {
466 case RPCSEC_GSS_KRB5:
467 case RPCSEC_GSS_KRB5I:
468 case RPCSEC_GSS_KRB5P:
470 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
473 if (secflavour == RPCSEC_GSS_KRB5)
474 svc = rpc_gss_svc_none;
475 else if (secflavour == RPCSEC_GSS_KRB5I)
476 svc = rpc_gss_svc_integrity;
478 svc = rpc_gss_svc_privacy;
480 if (clnt_principal == NULL)
481 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
482 srv_principal, mech_oid, svc);
484 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
485 clnt_principal, srv_principal, "kerberosv5",
486 svc, NULL, NULL, NULL);
494 return (authunix_create(cred));
500 * Callback from the RPC code to generate up/down notifications.
503 struct nfs_feedback_arg {
504 struct nfsmount *nf_mount;
505 int nf_lastmsg; /* last tprintf */
507 struct thread *nf_td;
511 nfs_feedback(int type, int proc, void *arg)
513 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
514 struct nfsmount *nmp = nf->nf_mount;
518 case FEEDBACK_REXMIT2:
519 case FEEDBACK_RECONNECT:
521 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
522 nfs_down(nmp, nf->nf_td,
523 "not responding", 0, NFSSTA_TIMEO);
524 nf->nf_tprintfmsg = TRUE;
525 nf->nf_lastmsg = now;
530 nfs_up(nf->nf_mount, nf->nf_td,
531 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
537 * newnfs_request - goes something like this
538 * - does the rpc by calling the krpc layer
539 * - break down rpc header and return with nfs reply
540 * nb: always frees up nd_mreq mbuf list
543 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
544 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
545 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
546 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
548 uint32_t retseq, retval, slotseq, *tl;
550 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
551 int error = 0, usegssname = 0, secflavour = AUTH_SYS;
552 int freeslot, maxslot, reterr, slotpos, timeo;
554 u_int trylater_delay = 1;
555 struct nfs_feedback_arg nf;
558 struct rpc_callextra ext;
560 struct nfsreq *rep = NULL;
561 char *srv_principal = NULL, *clnt_principal = NULL;
563 struct ucred *authcred;
564 struct nfsclsession *sep;
565 uint8_t sessionid[NFSX_V4SESSIONID];
570 /* Reject requests while attempting a forced unmount. */
571 if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
572 m_freem(nd->nd_mreq);
577 * Set authcred, which is used to acquire RPC credentials to
578 * the cred argument, by default. The crhold() should not be
579 * necessary, but will ensure that some future code change
580 * doesn't result in the credential being free'd prematurely.
582 authcred = crhold(cred);
584 /* For client side interruptible mounts, mask off the signals. */
585 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
586 newnfs_set_sigmask(td, &oldset);
591 * XXX if not already connected call nfs_connect now. Longer
592 * term, change nfs_mount to call nfs_connect unconditionally
593 * and let clnt_reconnect_create handle reconnects.
595 if (nrp->nr_client == NULL)
596 newnfs_connect(nmp, nrp, cred, td, 0);
599 * For a client side mount, nmp is != NULL and clp == NULL. For
600 * server calls (callbacks or upcalls), nmp == NULL.
604 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
605 secflavour = RPCSEC_GSS_KRB5;
606 if (nd->nd_procnum != NFSPROC_NULL) {
607 if (clp->lc_flags & LCL_GSSINTEGRITY)
608 secflavour = RPCSEC_GSS_KRB5I;
609 else if (clp->lc_flags & LCL_GSSPRIVACY)
610 secflavour = RPCSEC_GSS_KRB5P;
614 } else if (nmp != NULL && NFSHASKERB(nmp) &&
615 nd->nd_procnum != NFSPROC_NULL) {
616 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
617 nd->nd_flag |= ND_USEGSSNAME;
618 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
620 * If there is a client side host based credential,
621 * use that, otherwise use the system uid, if set.
622 * The system uid is in the nmp->nm_sockreq.nr_cred
625 if (nmp->nm_krbnamelen > 0) {
627 clnt_principal = nmp->nm_krbname;
628 } else if (nmp->nm_uid != (uid_t)-1) {
629 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
630 ("newnfs_request: NULL nr_cred"));
632 authcred = crhold(nmp->nm_sockreq.nr_cred);
634 } else if (nmp->nm_krbnamelen == 0 &&
635 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
637 * If there is no host based principal name and
638 * the system uid is set and this is root, use the
639 * system uid, since root won't have user
640 * credentials in a credentials cache file.
641 * The system uid is in the nmp->nm_sockreq.nr_cred
644 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
645 ("newnfs_request: NULL nr_cred"));
647 authcred = crhold(nmp->nm_sockreq.nr_cred);
649 if (NFSHASINTEGRITY(nmp))
650 secflavour = RPCSEC_GSS_KRB5I;
651 else if (NFSHASPRIVACY(nmp))
652 secflavour = RPCSEC_GSS_KRB5P;
654 secflavour = RPCSEC_GSS_KRB5;
655 srv_principal = NFSMNT_SRVKRBNAME(nmp);
656 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
657 nd->nd_procnum != NFSPROC_NULL &&
658 (nd->nd_flag & ND_USEGSSNAME) != 0) {
660 * Use the uid that did the mount when the RPC is doing
661 * NFSv4 system operations, as indicated by the
662 * ND_USEGSSNAME flag, for the AUTH_SYS case.
663 * The credentials in nm_sockreq.nr_cred were used for the
666 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
667 ("newnfs_request: NULL nr_cred"));
669 authcred = crhold(nmp->nm_sockreq.nr_cred);
673 bzero(&nf, sizeof(struct nfs_feedback_arg));
676 nf.nf_lastmsg = NFSD_MONOSEC -
677 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
680 if (nd->nd_procnum == NFSPROC_NULL)
681 auth = authnone_create();
682 else if (usegssname) {
684 * For this case, the authenticator is held in the
685 * nfssockreq structure, so don't release the reference count
686 * held on it. --> Don't AUTH_DESTROY() it in this function.
688 if (nrp->nr_auth == NULL)
689 nrp->nr_auth = nfs_getauth(nrp, secflavour,
690 clnt_principal, srv_principal, NULL, authcred);
692 rpc_gss_refresh_auth_call(nrp->nr_auth);
695 auth = nfs_getauth(nrp, secflavour, NULL,
696 srv_principal, NULL, authcred);
699 m_freem(nd->nd_mreq);
701 newnfs_restore_sigmask(td, &oldset);
704 bzero(&ext, sizeof(ext));
707 ext.rc_feedback = nfs_feedback;
708 ext.rc_feedback_arg = &nf;
711 procnum = nd->nd_procnum;
712 if ((nd->nd_flag & ND_NFSV4) &&
713 nd->nd_procnum != NFSPROC_NULL &&
714 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
715 procnum = NFSV4PROC_COMPOUND;
718 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
720 /* Map the procnum to the old NFSv2 one, as required. */
721 if ((nd->nd_flag & ND_NFSV2) != 0) {
722 if (nd->nd_procnum < NFS_V3NPROCS)
723 procnum = nfsv2_procid[nd->nd_procnum];
725 procnum = NFSV2PROC_NOOP;
729 * Now only used for the R_DONTRECOVER case, but until that is
730 * supported within the krpc code, I need to keep a queue of
731 * outstanding RPCs for nfsv4 client requests.
733 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
734 rep = malloc(sizeof(struct nfsreq),
735 M_NFSDREQ, M_WAITOK);
737 if (dtrace_nfscl_nfs234_start_probe != NULL) {
741 if (nd->nd_flag & ND_NFSV4) {
743 nfscl_nfs4_start_probes[nd->nd_procnum];
744 probe_procnum = nd->nd_procnum;
745 } else if (nd->nd_flag & ND_NFSV3) {
746 probe_id = nfscl_nfs3_start_probes[procnum];
747 probe_procnum = procnum;
750 nfscl_nfs2_start_probes[nd->nd_procnum];
751 probe_procnum = procnum;
754 (dtrace_nfscl_nfs234_start_probe)
755 (probe_id, vp, nd->nd_mreq, cred,
760 freeslot = -1; /* Set to slot that needs to be free'd */
762 slot = -1; /* Slot that needs a sequence# increment. */
764 * This timeout specifies when a new socket should be created,
765 * along with new xid values. For UDP, this should be done
766 * infrequently, since retransmits of RPC requests should normally
772 timo.tv_sec = NFSV4_UPCALLTIMEO;
774 timo.tv_sec = NFSV4_CALLBACKTIMEO;
776 if (nrp->nr_sotype != SOCK_DGRAM) {
778 if ((nmp->nm_flag & NFSMNT_NFSV4))
779 timo.tv_sec = INT_MAX;
781 timo.tv_sec = NFS_TCPTIMEO;
783 if (NFSHASSOFT(nmp)) {
785 * CLSET_RETRIES is set to 2, so this should be
786 * half of the total timeout required.
788 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
791 timo.tv_sec = timeo / NFS_HZ;
792 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
795 /* For UDP hard mounts, use a large value. */
796 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
805 * Chain request into list of outstanding requests.
808 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
814 if (clp != NULL && sep != NULL)
815 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
816 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
818 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
819 nd->nd_mreq, &nd->nd_mrep, timo);
820 NFSCL_DEBUG(2, "clnt call=%d\n", stat);
824 * RPC done, unlink the request.
827 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
832 * If there was a successful reply and a tprintf msg.
833 * tprintf a response.
835 if (stat == RPC_SUCCESS) {
837 } else if (stat == RPC_TIMEDOUT) {
838 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
840 } else if (stat == RPC_VERSMISMATCH) {
841 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
843 } else if (stat == RPC_PROGVERSMISMATCH) {
844 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
845 error = EPROTONOSUPPORT;
846 } else if (stat == RPC_INTR) {
848 } else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
849 stat == RPC_SYSTEMERROR) {
850 /* Check for a session slot that needs to be free'd. */
851 if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
852 (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
853 nd->nd_procnum != NFSPROC_NULL) {
855 * This should only occur when either the MDS or
856 * a client has an RPC against a DS fail.
857 * This happens because these cases use "soft"
858 * connections that can time out and fail.
859 * The slot used for this RPC is now in a
860 * non-deterministic state, but if the slot isn't
861 * free'd, threads can get stuck waiting for a slot.
864 sep = nfsmnt_mdssession(nmp);
866 * Bump the sequence# out of range, so that reuse of
867 * this slot will result in an NFSERR_SEQMISORDERED
868 * error and not a bogus cached RPC reply.
870 mtx_lock(&sep->nfsess_mtx);
871 sep->nfsess_slotseq[nd->nd_slotid] += 10;
872 mtx_unlock(&sep->nfsess_mtx);
873 /* And free the slot. */
874 nfsv4_freeslot(sep, nd->nd_slotid);
876 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
879 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
883 m_freem(nd->nd_mreq);
887 free(rep, M_NFSDREQ);
889 newnfs_restore_sigmask(td, &oldset);
893 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
896 * Search for any mbufs that are not a multiple of 4 bytes long
897 * or with m_data not longword aligned.
898 * These could cause pointer alignment problems, so copy them to
899 * well aligned mbufs.
901 newnfs_realign(&nd->nd_mrep, M_WAITOK);
902 nd->nd_md = nd->nd_mrep;
903 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
905 if (nd->nd_procnum != NFSPROC_NULL &&
906 nd->nd_procnum != NFSV4PROC_CBNULL) {
907 /* If sep == NULL, set it to the default in nmp. */
908 if (sep == NULL && nmp != NULL)
909 sep = nfsmnt_mdssession(nmp);
911 * and now the actual NFS xdr.
913 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
914 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
915 if (nd->nd_repstat >= 10000)
916 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
917 (int)nd->nd_repstat);
920 * Get rid of the tag, return count and SEQUENCE result for
923 if ((nd->nd_flag & ND_NFSV4) != 0) {
924 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
925 i = fxdr_unsigned(int, *tl);
926 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
929 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
930 opcnt = fxdr_unsigned(int, *tl++);
931 i = fxdr_unsigned(int, *tl++);
932 j = fxdr_unsigned(int, *tl);
934 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
936 * If the first op is Sequence, free up the slot.
938 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
939 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
940 NFSCL_DEBUG(1, "failed seq=%d\n", j);
941 if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
942 (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
943 j == 0)) && sep != NULL) {
944 if (i == NFSV4OP_SEQUENCE)
945 NFSM_DISSECT(tl, uint32_t *,
949 NFSM_DISSECT(tl, uint32_t *,
952 mtx_lock(&sep->nfsess_mtx);
953 if (bcmp(tl, sep->nfsess_sessionid,
954 NFSX_V4SESSIONID) == 0) {
955 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
956 retseq = fxdr_unsigned(uint32_t, *tl++);
957 slot = fxdr_unsigned(int, *tl++);
959 if (retseq != sep->nfsess_slotseq[slot])
960 printf("retseq diff 0x%x\n",
962 retval = fxdr_unsigned(uint32_t, *++tl);
963 if ((retval + 1) < sep->nfsess_foreslots
965 sep->nfsess_foreslots = (retval
967 else if ((retval + 1) >
968 sep->nfsess_foreslots)
969 sep->nfsess_foreslots = (retval
970 < 64) ? (retval + 1) : 64;
972 mtx_unlock(&sep->nfsess_mtx);
974 /* Grab the op and status for the next one. */
976 NFSM_DISSECT(tl, uint32_t *,
978 i = fxdr_unsigned(int, *tl++);
979 j = fxdr_unsigned(int, *tl);
983 if (nd->nd_repstat != 0) {
984 if (nd->nd_repstat == NFSERR_BADSESSION &&
985 nmp != NULL && dssep == NULL &&
986 (nd->nd_flag & ND_NFSV41) != 0) {
988 * If this is a client side MDS RPC, mark
989 * the MDS session defunct and initiate
990 * recovery, as required.
991 * The nfsess_defunct field is protected by
992 * the NFSLOCKMNT()/nm_mtx lock and not the
993 * nfsess_mtx lock to simplify its handling,
994 * for the MDS session. This lock is also
995 * sufficient for nfsess_sessionid, since it
996 * never changes in the structure.
998 NFSCL_DEBUG(1, "Got badsession\n");
1001 sep = NFSMNT_MDSSESSION(nmp);
1002 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
1003 NFSX_V4SESSIONID) == 0) {
1004 /* Initiate recovery. */
1005 sep->nfsess_defunct = 1;
1006 NFSCL_DEBUG(1, "Marked defunct\n");
1007 if (nmp->nm_clp != NULL) {
1008 nmp->nm_clp->nfsc_flags |=
1010 wakeup(nmp->nm_clp);
1015 * Sleep for up to 1sec waiting for a new
1018 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1021 * Get the session again, in case a new one
1022 * has been created during the sleep.
1024 sep = NFSMNT_MDSSESSION(nmp);
1026 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1027 reterr = nfsv4_sequencelookup(nmp, sep,
1028 &slotpos, &maxslot, &slotseq,
1031 /* Fill in new session info. */
1033 "Filling in new sequence\n");
1034 tl = nd->nd_sequence;
1035 bcopy(sessionid, tl,
1037 tl += NFSX_V4SESSIONID /
1039 *tl++ = txdr_unsigned(slotseq);
1040 *tl++ = txdr_unsigned(slotpos);
1041 *tl = txdr_unsigned(maxslot);
1043 if (reterr == NFSERR_BADSESSION ||
1046 "Badsession looping\n");
1047 m_freem(nd->nd_mrep);
1051 nd->nd_repstat = reterr;
1052 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1056 * When clp != NULL, it is a callback and all
1057 * callback operations can be retried for NFSERR_DELAY.
1059 if (((nd->nd_repstat == NFSERR_DELAY ||
1060 nd->nd_repstat == NFSERR_GRACE) &&
1061 (nd->nd_flag & ND_NFSV4) && (clp != NULL ||
1062 (nd->nd_procnum != NFSPROC_DELEGRETURN &&
1063 nd->nd_procnum != NFSPROC_SETATTR &&
1064 nd->nd_procnum != NFSPROC_READ &&
1065 nd->nd_procnum != NFSPROC_READDS &&
1066 nd->nd_procnum != NFSPROC_WRITE &&
1067 nd->nd_procnum != NFSPROC_WRITEDS &&
1068 nd->nd_procnum != NFSPROC_OPEN &&
1069 nd->nd_procnum != NFSPROC_CREATE &&
1070 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1071 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1072 nd->nd_procnum != NFSPROC_CLOSE &&
1073 nd->nd_procnum != NFSPROC_LOCK &&
1074 nd->nd_procnum != NFSPROC_LOCKU))) ||
1075 (nd->nd_repstat == NFSERR_DELAY &&
1076 (nd->nd_flag & ND_NFSV4) == 0) ||
1077 nd->nd_repstat == NFSERR_RESOURCE) {
1078 if (trylater_delay > NFS_TRYLATERDEL)
1079 trylater_delay = NFS_TRYLATERDEL;
1080 waituntil = NFSD_MONOSEC + trylater_delay;
1081 while (NFSD_MONOSEC < waituntil)
1082 (void) nfs_catnap(PZERO, 0, "nfstry");
1083 trylater_delay *= 2;
1085 mtx_lock(&sep->nfsess_mtx);
1086 sep->nfsess_slotseq[slot]++;
1087 *nd->nd_slotseq = txdr_unsigned(
1088 sep->nfsess_slotseq[slot]);
1089 mtx_unlock(&sep->nfsess_mtx);
1091 m_freem(nd->nd_mrep);
1097 * If the File Handle was stale, invalidate the
1098 * lookup cache, just in case.
1099 * (vp != NULL implies a client side call)
1101 if (nd->nd_repstat == ESTALE && vp != NULL) {
1103 if (ncl_call_invalcaches != NULL)
1104 (*ncl_call_invalcaches)(vp);
1107 if ((nd->nd_flag & ND_NFSV4) != 0) {
1108 /* Free the slot, as required. */
1110 nfsv4_freeslot(sep, freeslot);
1112 * If this op is Putfh, throw its results away.
1115 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1116 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1117 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1118 i = fxdr_unsigned(int, *tl++);
1119 j = fxdr_unsigned(int, *tl);
1121 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1124 * All Compounds that do an Op that must
1125 * be in sequence consist of NFSV4OP_PUTFH
1126 * followed by one of these. As such, we
1127 * can determine if the seqid# should be
1128 * incremented, here.
1130 if ((i == NFSV4OP_OPEN ||
1131 i == NFSV4OP_OPENCONFIRM ||
1132 i == NFSV4OP_OPENDOWNGRADE ||
1133 i == NFSV4OP_CLOSE ||
1134 i == NFSV4OP_LOCK ||
1135 i == NFSV4OP_LOCKU) &&
1137 (j != NFSERR_STALECLIENTID &&
1138 j != NFSERR_STALESTATEID &&
1139 j != NFSERR_BADSTATEID &&
1140 j != NFSERR_BADSEQID &&
1141 j != NFSERR_BADXDR &&
1142 j != NFSERR_RESOURCE &&
1143 j != NFSERR_NOFILEHANDLE)))
1144 nd->nd_flag |= ND_INCRSEQID;
1147 * If this op's status is non-zero, mark
1148 * that there is no more data to process.
1149 * The exception is Setattr, which always has xdr
1150 * when it has failed.
1152 if (j != 0 && i != NFSV4OP_SETATTR)
1153 nd->nd_flag |= ND_NOMOREDATA;
1156 * If R_DONTRECOVER is set, replace the stale error
1157 * reply, so that recovery isn't initiated.
1159 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1160 nd->nd_repstat == NFSERR_BADSESSION ||
1161 nd->nd_repstat == NFSERR_STALESTATEID) &&
1162 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1163 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1167 #ifdef KDTRACE_HOOKS
1168 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1172 if (nd->nd_flag & ND_NFSV4) {
1173 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1174 probe_procnum = nd->nd_procnum;
1175 } else if (nd->nd_flag & ND_NFSV3) {
1176 probe_id = nfscl_nfs3_done_probes[procnum];
1177 probe_procnum = procnum;
1179 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1180 probe_procnum = procnum;
1183 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1184 nd->nd_mreq, cred, probe_procnum, 0);
1188 m_freem(nd->nd_mreq);
1189 if (usegssname == 0)
1192 free(rep, M_NFSDREQ);
1194 newnfs_restore_sigmask(td, &oldset);
1197 mbuf_freem(nd->nd_mrep);
1198 mbuf_freem(nd->nd_mreq);
1199 if (usegssname == 0)
1202 free(rep, M_NFSDREQ);
1204 newnfs_restore_sigmask(td, &oldset);
1209 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1210 * wait for all requests to complete. This is used by forced unmounts
1211 * to terminate any outstanding RPCs.
1214 newnfs_nmcancelreqs(struct nfsmount *nmp)
1216 struct nfsclds *dsp;
1217 struct __rpc_client *cl;
1219 if (nmp->nm_sockreq.nr_client != NULL)
1220 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1223 TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1225 if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1226 (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1227 dsp->nfsclds_sockp != NULL &&
1228 dsp->nfsclds_sockp->nr_client != NULL) {
1229 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1230 cl = dsp->nfsclds_sockp->nr_client;
1243 * Any signal that can interrupt an NFS operation in an intr mount
1244 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1246 int newnfs_sig_set[] = {
1255 * Check to see if one of the signals in our subset is pending on
1256 * the process (in an intr mount).
1259 nfs_sig_pending(sigset_t set)
1263 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1264 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1270 * The set/restore sigmask functions are used to (temporarily) overwrite
1271 * the thread td_sigmask during an RPC call (for example). These are also
1272 * used in other places in the NFS client that might tsleep().
1275 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1283 td = curthread; /* XXX */
1285 /* Remove the NFS set of signals from newset */
1287 mtx_lock(&p->p_sigacts->ps_mtx);
1288 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1290 * But make sure we leave the ones already masked
1291 * by the process, ie. remove the signal from the
1292 * temporary signalmask only if it wasn't already
1295 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1296 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1297 SIGDELSET(newset, newnfs_sig_set[i]);
1299 mtx_unlock(&p->p_sigacts->ps_mtx);
1300 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1301 SIGPROCMASK_PROC_LOCKED);
1306 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1309 td = curthread; /* XXX */
1310 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1314 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1315 * old one after msleep() returns.
1318 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1323 if ((priority & PCATCH) == 0)
1324 return msleep(ident, mtx, priority, wmesg, timo);
1326 td = curthread; /* XXX */
1327 newnfs_set_sigmask(td, &oldset);
1328 error = msleep(ident, mtx, priority, wmesg, timo);
1329 newnfs_restore_sigmask(td, &oldset);
1334 * Test for a termination condition pending on the process.
1335 * This is used for NFSMNT_INT mounts.
1338 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1343 /* Terminate all requests while attempting a forced unmount. */
1344 if (NFSCL_FORCEDISM(nmp->nm_mountp))
1346 if (!(nmp->nm_flag & NFSMNT_INT))
1352 tmpset = p->p_siglist;
1353 SIGSETOR(tmpset, td->td_siglist);
1354 SIGSETNAND(tmpset, td->td_sigmask);
1355 mtx_lock(&p->p_sigacts->ps_mtx);
1356 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1357 mtx_unlock(&p->p_sigacts->ps_mtx);
1358 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1359 && nfs_sig_pending(tmpset)) {
1368 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1372 p = td ? td->td_proc : NULL;
1374 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1375 server, msg, error);
1377 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1383 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1384 int error, int flags)
1388 mtx_lock(&nmp->nm_mtx);
1389 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1390 nmp->nm_state |= NFSSTA_TIMEO;
1391 mtx_unlock(&nmp->nm_mtx);
1392 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1395 mtx_unlock(&nmp->nm_mtx);
1396 mtx_lock(&nmp->nm_mtx);
1397 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1398 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1399 mtx_unlock(&nmp->nm_mtx);
1400 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1403 mtx_unlock(&nmp->nm_mtx);
1404 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1408 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1409 int flags, int tprintfmsg)
1414 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1417 mtx_lock(&nmp->nm_mtx);
1418 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1419 nmp->nm_state &= ~NFSSTA_TIMEO;
1420 mtx_unlock(&nmp->nm_mtx);
1421 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1424 mtx_unlock(&nmp->nm_mtx);
1426 mtx_lock(&nmp->nm_mtx);
1427 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1428 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1429 mtx_unlock(&nmp->nm_mtx);
1430 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1433 mtx_unlock(&nmp->nm_mtx);