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;
102 SVCPOOL *nfscbd_pool;
103 static int nfsrv_gsscallbackson = 0;
104 static int nfs_bufpackets = 4;
105 static int nfs_reconnects;
106 static int nfs3_jukebox_delay = 10;
107 static int nfs_skip_wcc_data_onerr = 1;
109 SYSCTL_DECL(_vfs_nfs);
111 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
112 "Buffer reservation size 2 < x < 64");
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
114 "Number of times the nfs client has had to reconnect");
115 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
116 "Number of seconds to delay a retry after receiving EJUKEBOX");
117 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
118 "Disable weak cache consistency checking when server returns an error");
120 static void nfs_down(struct nfsmount *, struct thread *, const char *,
122 static void nfs_up(struct nfsmount *, struct thread *, const char *,
124 static int nfs_msg(struct thread *, const char *, const char *, int);
126 struct nfs_cached_auth {
127 int ca_refs; /* refcount, including 1 from the cache */
128 uid_t ca_uid; /* uid that corresponds to this auth */
129 AUTH *ca_auth; /* RPC auth handle */
132 static int nfsv2_procid[NFS_V3NPROCS] = {
158 * Initialize sockets and congestion for a new NFS connection.
159 * We do not free the sockaddr if error.
162 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
163 struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
165 int rcvreserve, sndreserve;
166 int pktscale, pktscalesav;
167 struct sockaddr *saddr;
168 struct ucred *origcred;
170 struct netconfig *nconf;
172 int one = 1, retries, error = 0;
173 struct thread *td = curthread;
178 * We need to establish the socket using the credentials of
179 * the mountpoint. Some parts of this process (such as
180 * sobind() and soconnect()) will use the curent thread's
181 * credential instead of the socket credential. To work
182 * around this, temporarily change the current thread's
183 * credential to that of the mountpoint.
185 * XXX: It would be better to explicitly pass the correct
186 * credential to sobind() and soconnect().
188 origcred = td->td_ucred;
191 * Use the credential in nr_cred, if not NULL.
193 if (nrp->nr_cred != NULL)
194 td->td_ucred = nrp->nr_cred;
199 if (saddr->sa_family == AF_INET)
200 if (nrp->nr_sotype == SOCK_DGRAM)
201 nconf = getnetconfigent("udp");
203 nconf = getnetconfigent("tcp");
204 else if (saddr->sa_family == AF_LOCAL)
205 nconf = getnetconfigent("local");
207 if (nrp->nr_sotype == SOCK_DGRAM)
208 nconf = getnetconfigent("udp6");
210 nconf = getnetconfigent("tcp6");
212 pktscale = nfs_bufpackets;
217 pktscalesav = pktscale;
219 * soreserve() can fail if sb_max is too small, so shrink pktscale
220 * and try again if there is an error.
221 * Print a log message suggesting increasing sb_max.
222 * Creating a socket and doing this is necessary since, if the
223 * reservation sizes are too large and will make soreserve() fail,
224 * the connection will work until a large send is attempted and
225 * then it will loop in the krpc code.
228 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
229 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
230 nrp->nr_soproto, td->td_ucred, td);
232 td->td_ucred = origcred;
236 if (error != 0 && pktscale > 2) {
237 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
238 pktscale == pktscalesav)
239 printf("Consider increasing kern.ipc.maxsockbuf\n");
242 if (nrp->nr_sotype == SOCK_DGRAM) {
244 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
246 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
249 sndreserve = rcvreserve = 1024 * pktscale;
252 if (nrp->nr_sotype != SOCK_STREAM)
253 panic("nfscon sotype");
255 sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
256 sizeof (u_int32_t)) * pktscale;
257 rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
258 sizeof (u_int32_t)) * pktscale;
260 sndreserve = rcvreserve = 1024 * pktscale;
263 error = soreserve(so, sndreserve, rcvreserve);
264 if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
266 printf("Must increase kern.ipc.maxsockbuf or reduce"
268 } while (error != 0 && pktscale > 2);
271 td->td_ucred = origcred;
275 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
276 nrp->nr_vers, sndreserve, rcvreserve);
277 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
279 if ((nmp->nm_flag & NFSMNT_INT))
280 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
281 if ((nmp->nm_flag & NFSMNT_RESVPORT))
282 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
283 if (NFSHASSOFT(nmp)) {
284 if (nmp->nm_sotype == SOCK_DGRAM)
286 * For UDP, the large timeout for a reconnect
287 * will be set to "nm_retry * nm_timeo / 2", so
288 * we only want to do 2 reconnect timeout
293 retries = nmp->nm_retry;
296 /* cred == NULL for DS connects. */
297 if (NFSHASNFSV4N(nmp) && cred != NULL) {
299 * Make sure the nfscbd_pool doesn't get destroyed
303 if (nfs_numnfscbd > 0) {
306 xprt = svc_vc_create_backchannel(nfscbd_pool);
307 CLNT_CONTROL(client, CLSET_BACKCHANNEL, xprt);
310 if (nfs_numnfscbd == 0)
311 wakeup(&nfs_numnfscbd);
318 * - Null RPC callback to client
319 * - Non-Null RPC callback to client, wait a little longer
320 * - upcalls to nfsuserd and gssd (clp == NULL)
322 if (callback_retry_mult == 0) {
323 retries = NFSV4_UPCALLRETRY;
324 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
326 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
329 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
333 * For UDP, there are 2 timeouts:
334 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
335 * that does a retransmit of an RPC request using the same
336 * socket and xid. This is what you normally want to do,
337 * since NFS servers depend on "same xid" for their
338 * Duplicate Request Cache.
339 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
340 * retransmits on the same socket should fail and a fresh
341 * socket created. Each of these timeouts counts as one
342 * CLSET_RETRIES as set above.
343 * Set the initial retransmit timeout for UDP. This timeout
344 * doesn't exist for TCP and the following call just fails,
347 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
348 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
349 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
352 mtx_lock(&nrp->nr_mtx);
353 if (nrp->nr_client != NULL) {
354 mtx_unlock(&nrp->nr_mtx);
356 * Someone else already connected.
358 CLNT_RELEASE(client);
360 nrp->nr_client = client;
362 * Protocols that do not require connections may be optionally
363 * left unconnected for servers that reply from a port other
366 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
367 mtx_unlock(&nrp->nr_mtx);
368 CLNT_CONTROL(client, CLSET_CONNECT, &one);
370 mtx_unlock(&nrp->nr_mtx);
374 /* Restore current thread's credentials. */
375 td->td_ucred = origcred;
383 * NFS disconnect. Clean up and unlink.
386 newnfs_disconnect(struct nfssockreq *nrp)
390 mtx_lock(&nrp->nr_mtx);
391 if (nrp->nr_client != NULL) {
392 client = nrp->nr_client;
393 nrp->nr_client = NULL;
394 mtx_unlock(&nrp->nr_mtx);
395 rpc_gss_secpurge_call(client);
397 CLNT_RELEASE(client);
399 mtx_unlock(&nrp->nr_mtx);
404 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
405 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
407 rpc_gss_service_t svc;
410 switch (secflavour) {
411 case RPCSEC_GSS_KRB5:
412 case RPCSEC_GSS_KRB5I:
413 case RPCSEC_GSS_KRB5P:
415 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
418 if (secflavour == RPCSEC_GSS_KRB5)
419 svc = rpc_gss_svc_none;
420 else if (secflavour == RPCSEC_GSS_KRB5I)
421 svc = rpc_gss_svc_integrity;
423 svc = rpc_gss_svc_privacy;
425 if (clnt_principal == NULL)
426 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
427 srv_principal, mech_oid, svc);
429 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
430 clnt_principal, srv_principal, "kerberosv5",
431 svc, NULL, NULL, NULL);
439 return (authunix_create(cred));
445 * Callback from the RPC code to generate up/down notifications.
448 struct nfs_feedback_arg {
449 struct nfsmount *nf_mount;
450 int nf_lastmsg; /* last tprintf */
452 struct thread *nf_td;
456 nfs_feedback(int type, int proc, void *arg)
458 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
459 struct nfsmount *nmp = nf->nf_mount;
463 case FEEDBACK_REXMIT2:
464 case FEEDBACK_RECONNECT:
466 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
467 nfs_down(nmp, nf->nf_td,
468 "not responding", 0, NFSSTA_TIMEO);
469 nf->nf_tprintfmsg = TRUE;
470 nf->nf_lastmsg = now;
475 nfs_up(nf->nf_mount, nf->nf_td,
476 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
482 * newnfs_request - goes something like this
483 * - does the rpc by calling the krpc layer
484 * - break down rpc header and return with nfs reply
485 * nb: always frees up nd_mreq mbuf list
488 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
489 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
490 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
491 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
493 uint32_t retseq, retval, slotseq, *tl;
495 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
496 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
497 int freeslot, maxslot, reterr, slotpos, timeo;
499 u_int trylater_delay = 1;
500 struct nfs_feedback_arg nf;
503 struct rpc_callextra ext;
505 struct nfsreq *rep = NULL;
506 char *srv_principal = NULL, *clnt_principal = NULL;
508 struct ucred *authcred;
509 struct nfsclsession *sep;
510 uint8_t sessionid[NFSX_V4SESSIONID];
515 /* Reject requests while attempting a forced unmount. */
516 if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
517 m_freem(nd->nd_mreq);
522 * Set authcred, which is used to acquire RPC credentials to
523 * the cred argument, by default. The crhold() should not be
524 * necessary, but will ensure that some future code change
525 * doesn't result in the credential being free'd prematurely.
527 authcred = crhold(cred);
529 /* For client side interruptible mounts, mask off the signals. */
530 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
531 newnfs_set_sigmask(td, &oldset);
536 * XXX if not already connected call nfs_connect now. Longer
537 * term, change nfs_mount to call nfs_connect unconditionally
538 * and let clnt_reconnect_create handle reconnects.
540 if (nrp->nr_client == NULL)
541 newnfs_connect(nmp, nrp, cred, td, 0);
544 * For a client side mount, nmp is != NULL and clp == NULL. For
545 * server calls (callbacks or upcalls), nmp == NULL.
549 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
550 secflavour = RPCSEC_GSS_KRB5;
551 if (nd->nd_procnum != NFSPROC_NULL) {
552 if (clp->lc_flags & LCL_GSSINTEGRITY)
553 secflavour = RPCSEC_GSS_KRB5I;
554 else if (clp->lc_flags & LCL_GSSPRIVACY)
555 secflavour = RPCSEC_GSS_KRB5P;
559 } else if (nmp != NULL && NFSHASKERB(nmp) &&
560 nd->nd_procnum != NFSPROC_NULL) {
561 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
562 nd->nd_flag |= ND_USEGSSNAME;
563 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
565 * If there is a client side host based credential,
566 * use that, otherwise use the system uid, if set.
567 * The system uid is in the nmp->nm_sockreq.nr_cred
570 if (nmp->nm_krbnamelen > 0) {
572 clnt_principal = nmp->nm_krbname;
573 } else if (nmp->nm_uid != (uid_t)-1) {
574 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
575 ("newnfs_request: NULL nr_cred"));
577 authcred = crhold(nmp->nm_sockreq.nr_cred);
579 } else if (nmp->nm_krbnamelen == 0 &&
580 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
582 * If there is no host based principal name and
583 * the system uid is set and this is root, use the
584 * system uid, since root won't have user
585 * credentials in a credentials cache file.
586 * The system uid is in the nmp->nm_sockreq.nr_cred
589 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
590 ("newnfs_request: NULL nr_cred"));
592 authcred = crhold(nmp->nm_sockreq.nr_cred);
594 if (NFSHASINTEGRITY(nmp))
595 secflavour = RPCSEC_GSS_KRB5I;
596 else if (NFSHASPRIVACY(nmp))
597 secflavour = RPCSEC_GSS_KRB5P;
599 secflavour = RPCSEC_GSS_KRB5;
600 srv_principal = NFSMNT_SRVKRBNAME(nmp);
601 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
602 nd->nd_procnum != NFSPROC_NULL &&
603 (nd->nd_flag & ND_USEGSSNAME) != 0) {
605 * Use the uid that did the mount when the RPC is doing
606 * NFSv4 system operations, as indicated by the
607 * ND_USEGSSNAME flag, for the AUTH_SYS case.
608 * The credentials in nm_sockreq.nr_cred were used for the
611 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
612 ("newnfs_request: NULL nr_cred"));
614 authcred = crhold(nmp->nm_sockreq.nr_cred);
618 bzero(&nf, sizeof(struct nfs_feedback_arg));
621 nf.nf_lastmsg = NFSD_MONOSEC -
622 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
625 if (nd->nd_procnum == NFSPROC_NULL)
626 auth = authnone_create();
627 else if (usegssname) {
629 * For this case, the authenticator is held in the
630 * nfssockreq structure, so don't release the reference count
631 * held on it. --> Don't AUTH_DESTROY() it in this function.
633 if (nrp->nr_auth == NULL)
634 nrp->nr_auth = nfs_getauth(nrp, secflavour,
635 clnt_principal, srv_principal, NULL, authcred);
637 rpc_gss_refresh_auth_call(nrp->nr_auth);
640 auth = nfs_getauth(nrp, secflavour, NULL,
641 srv_principal, NULL, authcred);
644 m_freem(nd->nd_mreq);
646 newnfs_restore_sigmask(td, &oldset);
649 bzero(&ext, sizeof(ext));
652 ext.rc_feedback = nfs_feedback;
653 ext.rc_feedback_arg = &nf;
656 procnum = nd->nd_procnum;
657 if ((nd->nd_flag & ND_NFSV4) &&
658 nd->nd_procnum != NFSPROC_NULL &&
659 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
660 procnum = NFSV4PROC_COMPOUND;
663 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
665 /* Map the procnum to the old NFSv2 one, as required. */
666 if ((nd->nd_flag & ND_NFSV2) != 0) {
667 if (nd->nd_procnum < NFS_V3NPROCS)
668 procnum = nfsv2_procid[nd->nd_procnum];
670 procnum = NFSV2PROC_NOOP;
674 * Now only used for the R_DONTRECOVER case, but until that is
675 * supported within the krpc code, I need to keep a queue of
676 * outstanding RPCs for nfsv4 client requests.
678 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
679 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
680 M_NFSDREQ, M_WAITOK);
682 if (dtrace_nfscl_nfs234_start_probe != NULL) {
686 if (nd->nd_flag & ND_NFSV4) {
688 nfscl_nfs4_start_probes[nd->nd_procnum];
689 probe_procnum = nd->nd_procnum;
690 } else if (nd->nd_flag & ND_NFSV3) {
691 probe_id = nfscl_nfs3_start_probes[procnum];
692 probe_procnum = procnum;
695 nfscl_nfs2_start_probes[nd->nd_procnum];
696 probe_procnum = procnum;
699 (dtrace_nfscl_nfs234_start_probe)
700 (probe_id, vp, nd->nd_mreq, cred,
706 freeslot = -1; /* Set to slot that needs to be free'd */
708 slot = -1; /* Slot that needs a sequence# increment. */
710 * This timeout specifies when a new socket should be created,
711 * along with new xid values. For UDP, this should be done
712 * infrequently, since retransmits of RPC requests should normally
718 timo.tv_sec = NFSV4_UPCALLTIMEO;
720 timo.tv_sec = NFSV4_CALLBACKTIMEO;
722 if (nrp->nr_sotype != SOCK_DGRAM) {
724 if ((nmp->nm_flag & NFSMNT_NFSV4))
725 timo.tv_sec = INT_MAX;
727 timo.tv_sec = NFS_TCPTIMEO;
729 if (NFSHASSOFT(nmp)) {
731 * CLSET_RETRIES is set to 2, so this should be
732 * half of the total timeout required.
734 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
737 timo.tv_sec = timeo / NFS_HZ;
738 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
741 /* For UDP hard mounts, use a large value. */
742 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
751 * Chain request into list of outstanding requests.
754 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
760 if (clp != NULL && sep != NULL)
761 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
762 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
764 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
765 nd->nd_mreq, &nd->nd_mrep, timo);
769 * RPC done, unlink the request.
772 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
777 * If there was a successful reply and a tprintf msg.
778 * tprintf a response.
780 if (stat == RPC_SUCCESS) {
782 } else if (stat == RPC_TIMEDOUT) {
783 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
785 } else if (stat == RPC_VERSMISMATCH) {
786 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
788 } else if (stat == RPC_PROGVERSMISMATCH) {
789 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
790 error = EPROTONOSUPPORT;
791 } else if (stat == RPC_INTR) {
794 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
798 m_freem(nd->nd_mreq);
802 FREE((caddr_t)rep, M_NFSDREQ);
804 newnfs_restore_sigmask(td, &oldset);
808 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
811 * Search for any mbufs that are not a multiple of 4 bytes long
812 * or with m_data not longword aligned.
813 * These could cause pointer alignment problems, so copy them to
814 * well aligned mbufs.
816 newnfs_realign(&nd->nd_mrep, M_WAITOK);
817 nd->nd_md = nd->nd_mrep;
818 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
820 if (nd->nd_procnum != NFSPROC_NULL &&
821 nd->nd_procnum != NFSV4PROC_CBNULL) {
822 /* If sep == NULL, set it to the default in nmp. */
823 if (sep == NULL && nmp != NULL)
824 sep = nfsmnt_mdssession(nmp);
826 * and now the actual NFS xdr.
828 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
829 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
830 if (nd->nd_repstat >= 10000)
831 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
832 (int)nd->nd_repstat);
835 * Get rid of the tag, return count and SEQUENCE result for
838 if ((nd->nd_flag & ND_NFSV4) != 0) {
839 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
840 i = fxdr_unsigned(int, *tl);
841 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
844 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
845 opcnt = fxdr_unsigned(int, *tl++);
846 i = fxdr_unsigned(int, *tl++);
847 j = fxdr_unsigned(int, *tl);
849 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
851 * If the first op is Sequence, free up the slot.
853 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
854 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
855 NFSCL_DEBUG(1, "failed seq=%d\n", j);
856 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
857 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j == 0)
859 if (i == NFSV4OP_SEQUENCE)
860 NFSM_DISSECT(tl, uint32_t *,
864 NFSM_DISSECT(tl, uint32_t *,
867 mtx_lock(&sep->nfsess_mtx);
868 if (bcmp(tl, sep->nfsess_sessionid,
869 NFSX_V4SESSIONID) == 0) {
870 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
871 retseq = fxdr_unsigned(uint32_t, *tl++);
872 slot = fxdr_unsigned(int, *tl++);
874 if (retseq != sep->nfsess_slotseq[slot])
875 printf("retseq diff 0x%x\n",
877 retval = fxdr_unsigned(uint32_t, *++tl);
878 if ((retval + 1) < sep->nfsess_foreslots
880 sep->nfsess_foreslots = (retval
882 else if ((retval + 1) >
883 sep->nfsess_foreslots)
884 sep->nfsess_foreslots = (retval
885 < 64) ? (retval + 1) : 64;
887 mtx_unlock(&sep->nfsess_mtx);
889 /* Grab the op and status for the next one. */
891 NFSM_DISSECT(tl, uint32_t *,
893 i = fxdr_unsigned(int, *tl++);
894 j = fxdr_unsigned(int, *tl);
898 if (nd->nd_repstat != 0) {
899 if (nd->nd_repstat == NFSERR_BADSESSION &&
900 nmp != NULL && dssep == NULL) {
902 * If this is a client side MDS RPC, mark
903 * the MDS session defunct and initiate
904 * recovery, as required.
905 * The nfsess_defunct field is protected by
906 * the NFSLOCKMNT()/nm_mtx lock and not the
907 * nfsess_mtx lock to simplify its handling,
908 * for the MDS session. This lock is also
909 * sufficient for nfsess_sessionid, since it
910 * never changes in the structure.
912 NFSCL_DEBUG(1, "Got badsession\n");
915 sep = NFSMNT_MDSSESSION(nmp);
916 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
917 NFSX_V4SESSIONID) == 0) {
918 /* Initiate recovery. */
919 sep->nfsess_defunct = 1;
920 NFSCL_DEBUG(1, "Marked defunct\n");
921 if (nmp->nm_clp != NULL) {
922 nmp->nm_clp->nfsc_flags |=
929 * Sleep for up to 1sec waiting for a new
932 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
935 * Get the session again, in case a new one
936 * has been created during the sleep.
938 sep = NFSMNT_MDSSESSION(nmp);
940 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
941 reterr = nfsv4_sequencelookup(nmp, sep,
942 &slotpos, &maxslot, &slotseq,
945 /* Fill in new session info. */
947 "Filling in new sequence\n");
948 tl = nd->nd_sequence;
951 tl += NFSX_V4SESSIONID /
953 *tl++ = txdr_unsigned(slotseq);
954 *tl++ = txdr_unsigned(slotpos);
955 *tl = txdr_unsigned(maxslot);
957 if (reterr == NFSERR_BADSESSION ||
960 "Badsession looping\n");
961 m_freem(nd->nd_mrep);
965 nd->nd_repstat = reterr;
966 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
969 if (((nd->nd_repstat == NFSERR_DELAY ||
970 nd->nd_repstat == NFSERR_GRACE) &&
971 (nd->nd_flag & ND_NFSV4) &&
972 nd->nd_procnum != NFSPROC_DELEGRETURN &&
973 nd->nd_procnum != NFSPROC_SETATTR &&
974 nd->nd_procnum != NFSPROC_READ &&
975 nd->nd_procnum != NFSPROC_READDS &&
976 nd->nd_procnum != NFSPROC_WRITE &&
977 nd->nd_procnum != NFSPROC_WRITEDS &&
978 nd->nd_procnum != NFSPROC_OPEN &&
979 nd->nd_procnum != NFSPROC_CREATE &&
980 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
981 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
982 nd->nd_procnum != NFSPROC_CLOSE &&
983 nd->nd_procnum != NFSPROC_LOCK &&
984 nd->nd_procnum != NFSPROC_LOCKU) ||
985 (nd->nd_repstat == NFSERR_DELAY &&
986 (nd->nd_flag & ND_NFSV4) == 0) ||
987 nd->nd_repstat == NFSERR_RESOURCE) {
988 if (trylater_delay > NFS_TRYLATERDEL)
989 trylater_delay = NFS_TRYLATERDEL;
990 waituntil = NFSD_MONOSEC + trylater_delay;
991 while (NFSD_MONOSEC < waituntil)
992 (void) nfs_catnap(PZERO, 0, "nfstry");
995 mtx_lock(&sep->nfsess_mtx);
996 sep->nfsess_slotseq[slot]++;
997 *nd->nd_slotseq = txdr_unsigned(
998 sep->nfsess_slotseq[slot]);
999 mtx_unlock(&sep->nfsess_mtx);
1001 m_freem(nd->nd_mrep);
1007 * If the File Handle was stale, invalidate the
1008 * lookup cache, just in case.
1009 * (vp != NULL implies a client side call)
1011 if (nd->nd_repstat == ESTALE && vp != NULL) {
1013 if (ncl_call_invalcaches != NULL)
1014 (*ncl_call_invalcaches)(vp);
1017 if ((nd->nd_flag & ND_NFSV4) != 0) {
1018 /* Free the slot, as required. */
1020 nfsv4_freeslot(sep, freeslot);
1022 * If this op is Putfh, throw its results away.
1025 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1026 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1027 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1028 i = fxdr_unsigned(int, *tl++);
1029 j = fxdr_unsigned(int, *tl);
1031 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1034 * All Compounds that do an Op that must
1035 * be in sequence consist of NFSV4OP_PUTFH
1036 * followed by one of these. As such, we
1037 * can determine if the seqid# should be
1038 * incremented, here.
1040 if ((i == NFSV4OP_OPEN ||
1041 i == NFSV4OP_OPENCONFIRM ||
1042 i == NFSV4OP_OPENDOWNGRADE ||
1043 i == NFSV4OP_CLOSE ||
1044 i == NFSV4OP_LOCK ||
1045 i == NFSV4OP_LOCKU) &&
1047 (j != NFSERR_STALECLIENTID &&
1048 j != NFSERR_STALESTATEID &&
1049 j != NFSERR_BADSTATEID &&
1050 j != NFSERR_BADSEQID &&
1051 j != NFSERR_BADXDR &&
1052 j != NFSERR_RESOURCE &&
1053 j != NFSERR_NOFILEHANDLE)))
1054 nd->nd_flag |= ND_INCRSEQID;
1057 * If this op's status is non-zero, mark
1058 * that there is no more data to process.
1059 * The exception is Setattr, which always has xdr
1060 * when it has failed.
1062 if (j != 0 && i != NFSV4OP_SETATTR)
1063 nd->nd_flag |= ND_NOMOREDATA;
1066 * If R_DONTRECOVER is set, replace the stale error
1067 * reply, so that recovery isn't initiated.
1069 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1070 nd->nd_repstat == NFSERR_BADSESSION ||
1071 nd->nd_repstat == NFSERR_STALESTATEID) &&
1072 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1073 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1077 #ifdef KDTRACE_HOOKS
1078 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1082 if (nd->nd_flag & ND_NFSV4) {
1083 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1084 probe_procnum = nd->nd_procnum;
1085 } else if (nd->nd_flag & ND_NFSV3) {
1086 probe_id = nfscl_nfs3_done_probes[procnum];
1087 probe_procnum = procnum;
1089 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1090 probe_procnum = procnum;
1093 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1094 nd->nd_mreq, cred, probe_procnum, 0);
1098 m_freem(nd->nd_mreq);
1099 if (usegssname == 0)
1102 FREE((caddr_t)rep, M_NFSDREQ);
1104 newnfs_restore_sigmask(td, &oldset);
1107 mbuf_freem(nd->nd_mrep);
1108 mbuf_freem(nd->nd_mreq);
1109 if (usegssname == 0)
1112 FREE((caddr_t)rep, M_NFSDREQ);
1114 newnfs_restore_sigmask(td, &oldset);
1119 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1120 * wait for all requests to complete. This is used by forced unmounts
1121 * to terminate any outstanding RPCs.
1124 newnfs_nmcancelreqs(struct nfsmount *nmp)
1126 struct nfsclds *dsp;
1127 struct __rpc_client *cl;
1129 if (nmp->nm_sockreq.nr_client != NULL)
1130 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1133 TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1135 if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1136 (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1137 dsp->nfsclds_sockp != NULL &&
1138 dsp->nfsclds_sockp->nr_client != NULL) {
1139 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1140 cl = dsp->nfsclds_sockp->nr_client;
1153 * Any signal that can interrupt an NFS operation in an intr mount
1154 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1156 int newnfs_sig_set[] = {
1165 * Check to see if one of the signals in our subset is pending on
1166 * the process (in an intr mount).
1169 nfs_sig_pending(sigset_t set)
1173 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1174 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1180 * The set/restore sigmask functions are used to (temporarily) overwrite
1181 * the thread td_sigmask during an RPC call (for example). These are also
1182 * used in other places in the NFS client that might tsleep().
1185 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1193 td = curthread; /* XXX */
1195 /* Remove the NFS set of signals from newset */
1197 mtx_lock(&p->p_sigacts->ps_mtx);
1198 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1200 * But make sure we leave the ones already masked
1201 * by the process, ie. remove the signal from the
1202 * temporary signalmask only if it wasn't already
1205 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1206 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1207 SIGDELSET(newset, newnfs_sig_set[i]);
1209 mtx_unlock(&p->p_sigacts->ps_mtx);
1210 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1211 SIGPROCMASK_PROC_LOCKED);
1216 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1219 td = curthread; /* XXX */
1220 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1224 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1225 * old one after msleep() returns.
1228 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1234 if ((priority & PCATCH) == 0)
1235 return msleep(ident, mtx, priority, wmesg, timo);
1237 td = curthread; /* XXX */
1238 newnfs_set_sigmask(td, &oldset);
1239 error = msleep(ident, mtx, priority, wmesg, timo);
1240 newnfs_restore_sigmask(td, &oldset);
1246 * Test for a termination condition pending on the process.
1247 * This is used for NFSMNT_INT mounts.
1250 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1255 /* Terminate all requests while attempting a forced unmount. */
1256 if (NFSCL_FORCEDISM(nmp->nm_mountp))
1258 if (!(nmp->nm_flag & NFSMNT_INT))
1264 tmpset = p->p_siglist;
1265 SIGSETOR(tmpset, td->td_siglist);
1266 SIGSETNAND(tmpset, td->td_sigmask);
1267 mtx_lock(&p->p_sigacts->ps_mtx);
1268 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1269 mtx_unlock(&p->p_sigacts->ps_mtx);
1270 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1271 && nfs_sig_pending(tmpset)) {
1280 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1284 p = td ? td->td_proc : NULL;
1286 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1287 server, msg, error);
1289 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1295 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1296 int error, int flags)
1300 mtx_lock(&nmp->nm_mtx);
1301 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1302 nmp->nm_state |= NFSSTA_TIMEO;
1303 mtx_unlock(&nmp->nm_mtx);
1304 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1307 mtx_unlock(&nmp->nm_mtx);
1308 mtx_lock(&nmp->nm_mtx);
1309 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1310 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1311 mtx_unlock(&nmp->nm_mtx);
1312 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1315 mtx_unlock(&nmp->nm_mtx);
1316 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1320 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1321 int flags, int tprintfmsg)
1326 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1329 mtx_lock(&nmp->nm_mtx);
1330 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1331 nmp->nm_state &= ~NFSSTA_TIMEO;
1332 mtx_unlock(&nmp->nm_mtx);
1333 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1336 mtx_unlock(&nmp->nm_mtx);
1338 mtx_lock(&nmp->nm_mtx);
1339 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1340 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1341 mtx_unlock(&nmp->nm_mtx);
1342 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1345 mtx_unlock(&nmp->nm_mtx);