2 * Copyright (c) 1989, 1991, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 * Socket operations for use by nfs
41 #include "opt_kgssapi.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/limits.h>
49 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
54 #include <sys/signalvar.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysctl.h>
57 #include <sys/syslog.h>
58 #include <sys/vnode.h>
63 #include <kgssapi/krb5/kcrypto.h>
65 #include <fs/nfs/nfsport.h>
68 #include <sys/dtrace_bsd.h>
70 dtrace_nfsclient_nfs23_start_probe_func_t
71 dtrace_nfscl_nfs234_start_probe;
73 dtrace_nfsclient_nfs23_done_probe_func_t
74 dtrace_nfscl_nfs234_done_probe;
77 * Registered probes by RPC type.
79 uint32_t nfscl_nfs2_start_probes[NFSV41_NPROCS + 1];
80 uint32_t nfscl_nfs2_done_probes[NFSV41_NPROCS + 1];
82 uint32_t nfscl_nfs3_start_probes[NFSV41_NPROCS + 1];
83 uint32_t nfscl_nfs3_done_probes[NFSV41_NPROCS + 1];
85 uint32_t nfscl_nfs4_start_probes[NFSV41_NPROCS + 1];
86 uint32_t nfscl_nfs4_done_probes[NFSV41_NPROCS + 1];
93 extern struct nfsstatsv1 nfsstatsv1;
94 extern struct nfsreqhead nfsd_reqq;
95 extern int nfscl_ticks;
96 extern void (*ncl_call_invalcaches)(struct vnode *);
97 extern int nfs_numnfscbd;
98 extern int nfscl_debuglevel;
100 SVCPOOL *nfscbd_pool;
101 static int nfsrv_gsscallbackson = 0;
102 static int nfs_bufpackets = 4;
103 static int nfs_reconnects;
104 static int nfs3_jukebox_delay = 10;
105 static int nfs_skip_wcc_data_onerr = 1;
107 SYSCTL_DECL(_vfs_nfs);
109 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
110 "Buffer reservation size 2 < x < 64");
111 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
112 "Number of times the nfs client has had to reconnect");
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
114 "Number of seconds to delay a retry after receiving EJUKEBOX");
115 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
116 "Disable weak cache consistency checking when server returns an error");
118 static void nfs_down(struct nfsmount *, struct thread *, const char *,
120 static void nfs_up(struct nfsmount *, struct thread *, const char *,
122 static int nfs_msg(struct thread *, const char *, const char *, int);
124 struct nfs_cached_auth {
125 int ca_refs; /* refcount, including 1 from the cache */
126 uid_t ca_uid; /* uid that corresponds to this auth */
127 AUTH *ca_auth; /* RPC auth handle */
130 static int nfsv2_procid[NFS_V3NPROCS] = {
156 * Initialize sockets and congestion for a new NFS connection.
157 * We do not free the sockaddr if error.
160 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
161 struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
163 int rcvreserve, sndreserve;
165 struct sockaddr *saddr;
166 struct ucred *origcred;
168 struct netconfig *nconf;
170 int one = 1, retries, error = 0;
171 struct thread *td = curthread;
176 * We need to establish the socket using the credentials of
177 * the mountpoint. Some parts of this process (such as
178 * sobind() and soconnect()) will use the curent thread's
179 * credential instead of the socket credential. To work
180 * around this, temporarily change the current thread's
181 * credential to that of the mountpoint.
183 * XXX: It would be better to explicitly pass the correct
184 * credential to sobind() and soconnect().
186 origcred = td->td_ucred;
189 * Use the credential in nr_cred, if not NULL.
191 if (nrp->nr_cred != NULL)
192 td->td_ucred = nrp->nr_cred;
197 if (saddr->sa_family == AF_INET)
198 if (nrp->nr_sotype == SOCK_DGRAM)
199 nconf = getnetconfigent("udp");
201 nconf = getnetconfigent("tcp");
203 if (nrp->nr_sotype == SOCK_DGRAM)
204 nconf = getnetconfigent("udp6");
206 nconf = getnetconfigent("tcp6");
208 pktscale = nfs_bufpackets;
214 * soreserve() can fail if sb_max is too small, so shrink pktscale
215 * and try again if there is an error.
216 * Print a log message suggesting increasing sb_max.
217 * Creating a socket and doing this is necessary since, if the
218 * reservation sizes are too large and will make soreserve() fail,
219 * the connection will work until a large send is attempted and
220 * then it will loop in the krpc code.
223 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
224 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
225 nrp->nr_soproto, td->td_ucred, td);
227 td->td_ucred = origcred;
231 if (error != 0 && pktscale > 2)
233 if (nrp->nr_sotype == SOCK_DGRAM) {
235 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
237 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
240 sndreserve = rcvreserve = 1024 * pktscale;
243 if (nrp->nr_sotype != SOCK_STREAM)
244 panic("nfscon sotype");
246 sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
247 sizeof (u_int32_t)) * pktscale;
248 rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
249 sizeof (u_int32_t)) * pktscale;
251 sndreserve = rcvreserve = 1024 * pktscale;
254 error = soreserve(so, sndreserve, rcvreserve);
255 } while (error != 0 && pktscale > 2);
258 td->td_ucred = origcred;
262 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
263 nrp->nr_vers, sndreserve, rcvreserve);
264 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
266 if ((nmp->nm_flag & NFSMNT_INT))
267 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
268 if ((nmp->nm_flag & NFSMNT_RESVPORT))
269 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
270 if (NFSHASSOFT(nmp)) {
271 if (nmp->nm_sotype == SOCK_DGRAM)
273 * For UDP, the large timeout for a reconnect
274 * will be set to "nm_retry * nm_timeo / 2", so
275 * we only want to do 2 reconnect timeout
280 retries = nmp->nm_retry;
283 /* cred == NULL for DS connects. */
284 if (NFSHASNFSV4N(nmp) && cred != NULL) {
286 * Make sure the nfscbd_pool doesn't get destroyed
290 if (nfs_numnfscbd > 0) {
293 xprt = svc_vc_create_backchannel(nfscbd_pool);
294 CLNT_CONTROL(client, CLSET_BACKCHANNEL, xprt);
297 if (nfs_numnfscbd == 0)
298 wakeup(&nfs_numnfscbd);
305 * - Null RPC callback to client
306 * - Non-Null RPC callback to client, wait a little longer
307 * - upcalls to nfsuserd and gssd (clp == NULL)
309 if (callback_retry_mult == 0) {
310 retries = NFSV4_UPCALLRETRY;
311 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
313 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
316 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
320 * For UDP, there are 2 timeouts:
321 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
322 * that does a retransmit of an RPC request using the same
323 * socket and xid. This is what you normally want to do,
324 * since NFS servers depend on "same xid" for their
325 * Duplicate Request Cache.
326 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
327 * retransmits on the same socket should fail and a fresh
328 * socket created. Each of these timeouts counts as one
329 * CLSET_RETRIES as set above.
330 * Set the initial retransmit timeout for UDP. This timeout
331 * doesn't exist for TCP and the following call just fails,
334 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
335 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
336 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
339 mtx_lock(&nrp->nr_mtx);
340 if (nrp->nr_client != NULL) {
341 mtx_unlock(&nrp->nr_mtx);
343 * Someone else already connected.
345 CLNT_RELEASE(client);
347 nrp->nr_client = client;
349 * Protocols that do not require connections may be optionally
350 * left unconnected for servers that reply from a port other
353 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
354 mtx_unlock(&nrp->nr_mtx);
355 CLNT_CONTROL(client, CLSET_CONNECT, &one);
357 mtx_unlock(&nrp->nr_mtx);
361 /* Restore current thread's credentials. */
362 td->td_ucred = origcred;
370 * NFS disconnect. Clean up and unlink.
373 newnfs_disconnect(struct nfssockreq *nrp)
377 mtx_lock(&nrp->nr_mtx);
378 if (nrp->nr_client != NULL) {
379 client = nrp->nr_client;
380 nrp->nr_client = NULL;
381 mtx_unlock(&nrp->nr_mtx);
382 rpc_gss_secpurge_call(client);
384 CLNT_RELEASE(client);
386 mtx_unlock(&nrp->nr_mtx);
391 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
392 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
394 rpc_gss_service_t svc;
397 switch (secflavour) {
398 case RPCSEC_GSS_KRB5:
399 case RPCSEC_GSS_KRB5I:
400 case RPCSEC_GSS_KRB5P:
402 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
405 if (secflavour == RPCSEC_GSS_KRB5)
406 svc = rpc_gss_svc_none;
407 else if (secflavour == RPCSEC_GSS_KRB5I)
408 svc = rpc_gss_svc_integrity;
410 svc = rpc_gss_svc_privacy;
412 if (clnt_principal == NULL)
413 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
414 srv_principal, mech_oid, svc);
416 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
417 clnt_principal, srv_principal, "kerberosv5",
418 svc, NULL, NULL, NULL);
426 return (authunix_create(cred));
432 * Callback from the RPC code to generate up/down notifications.
435 struct nfs_feedback_arg {
436 struct nfsmount *nf_mount;
437 int nf_lastmsg; /* last tprintf */
439 struct thread *nf_td;
443 nfs_feedback(int type, int proc, void *arg)
445 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
446 struct nfsmount *nmp = nf->nf_mount;
450 case FEEDBACK_REXMIT2:
451 case FEEDBACK_RECONNECT:
453 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
454 nfs_down(nmp, nf->nf_td,
455 "not responding", 0, NFSSTA_TIMEO);
456 nf->nf_tprintfmsg = TRUE;
457 nf->nf_lastmsg = now;
462 nfs_up(nf->nf_mount, nf->nf_td,
463 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
469 * newnfs_request - goes something like this
470 * - does the rpc by calling the krpc layer
471 * - break down rpc header and return with nfs reply
472 * nb: always frees up nd_mreq mbuf list
475 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
476 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
477 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
478 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
480 uint32_t retseq, retval, slotseq, *tl;
482 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
483 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
484 int freeslot, maxslot, reterr, slotpos, timeo;
486 u_int trylater_delay = 1;
487 struct nfs_feedback_arg nf;
490 struct rpc_callextra ext;
492 struct nfsreq *rep = NULL;
493 char *srv_principal = NULL, *clnt_principal = NULL;
495 struct ucred *authcred;
496 struct nfsclsession *sep;
497 uint8_t sessionid[NFSX_V4SESSIONID];
502 /* Reject requests while attempting a forced unmount. */
503 if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
504 m_freem(nd->nd_mreq);
509 * Set authcred, which is used to acquire RPC credentials to
510 * the cred argument, by default. The crhold() should not be
511 * necessary, but will ensure that some future code change
512 * doesn't result in the credential being free'd prematurely.
514 authcred = crhold(cred);
516 /* For client side interruptible mounts, mask off the signals. */
517 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
518 newnfs_set_sigmask(td, &oldset);
523 * XXX if not already connected call nfs_connect now. Longer
524 * term, change nfs_mount to call nfs_connect unconditionally
525 * and let clnt_reconnect_create handle reconnects.
527 if (nrp->nr_client == NULL)
528 newnfs_connect(nmp, nrp, cred, td, 0);
531 * For a client side mount, nmp is != NULL and clp == NULL. For
532 * server calls (callbacks or upcalls), nmp == NULL.
536 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
537 secflavour = RPCSEC_GSS_KRB5;
538 if (nd->nd_procnum != NFSPROC_NULL) {
539 if (clp->lc_flags & LCL_GSSINTEGRITY)
540 secflavour = RPCSEC_GSS_KRB5I;
541 else if (clp->lc_flags & LCL_GSSPRIVACY)
542 secflavour = RPCSEC_GSS_KRB5P;
546 } else if (nmp != NULL && NFSHASKERB(nmp) &&
547 nd->nd_procnum != NFSPROC_NULL) {
548 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
549 nd->nd_flag |= ND_USEGSSNAME;
550 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
552 * If there is a client side host based credential,
553 * use that, otherwise use the system uid, if set.
554 * The system uid is in the nmp->nm_sockreq.nr_cred
557 if (nmp->nm_krbnamelen > 0) {
559 clnt_principal = nmp->nm_krbname;
560 } else if (nmp->nm_uid != (uid_t)-1) {
561 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
562 ("newnfs_request: NULL nr_cred"));
564 authcred = crhold(nmp->nm_sockreq.nr_cred);
566 } else if (nmp->nm_krbnamelen == 0 &&
567 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
569 * If there is no host based principal name and
570 * the system uid is set and this is root, use the
571 * system uid, since root won't have user
572 * credentials in a credentials cache file.
573 * The system uid is in the nmp->nm_sockreq.nr_cred
576 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
577 ("newnfs_request: NULL nr_cred"));
579 authcred = crhold(nmp->nm_sockreq.nr_cred);
581 if (NFSHASINTEGRITY(nmp))
582 secflavour = RPCSEC_GSS_KRB5I;
583 else if (NFSHASPRIVACY(nmp))
584 secflavour = RPCSEC_GSS_KRB5P;
586 secflavour = RPCSEC_GSS_KRB5;
587 srv_principal = NFSMNT_SRVKRBNAME(nmp);
588 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
589 nd->nd_procnum != NFSPROC_NULL &&
590 (nd->nd_flag & ND_USEGSSNAME) != 0) {
592 * Use the uid that did the mount when the RPC is doing
593 * NFSv4 system operations, as indicated by the
594 * ND_USEGSSNAME flag, for the AUTH_SYS case.
595 * The credentials in nm_sockreq.nr_cred were used for the
598 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
599 ("newnfs_request: NULL nr_cred"));
601 authcred = crhold(nmp->nm_sockreq.nr_cred);
605 bzero(&nf, sizeof(struct nfs_feedback_arg));
608 nf.nf_lastmsg = NFSD_MONOSEC -
609 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
612 if (nd->nd_procnum == NFSPROC_NULL)
613 auth = authnone_create();
614 else if (usegssname) {
616 * For this case, the authenticator is held in the
617 * nfssockreq structure, so don't release the reference count
618 * held on it. --> Don't AUTH_DESTROY() it in this function.
620 if (nrp->nr_auth == NULL)
621 nrp->nr_auth = nfs_getauth(nrp, secflavour,
622 clnt_principal, srv_principal, NULL, authcred);
624 rpc_gss_refresh_auth_call(nrp->nr_auth);
627 auth = nfs_getauth(nrp, secflavour, NULL,
628 srv_principal, NULL, authcred);
631 m_freem(nd->nd_mreq);
633 newnfs_restore_sigmask(td, &oldset);
636 bzero(&ext, sizeof(ext));
639 ext.rc_feedback = nfs_feedback;
640 ext.rc_feedback_arg = &nf;
643 procnum = nd->nd_procnum;
644 if ((nd->nd_flag & ND_NFSV4) &&
645 nd->nd_procnum != NFSPROC_NULL &&
646 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
647 procnum = NFSV4PROC_COMPOUND;
650 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
652 /* Map the procnum to the old NFSv2 one, as required. */
653 if ((nd->nd_flag & ND_NFSV2) != 0) {
654 if (nd->nd_procnum < NFS_V3NPROCS)
655 procnum = nfsv2_procid[nd->nd_procnum];
657 procnum = NFSV2PROC_NOOP;
661 * Now only used for the R_DONTRECOVER case, but until that is
662 * supported within the krpc code, I need to keep a queue of
663 * outstanding RPCs for nfsv4 client requests.
665 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
666 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
667 M_NFSDREQ, M_WAITOK);
669 if (dtrace_nfscl_nfs234_start_probe != NULL) {
673 if (nd->nd_flag & ND_NFSV4) {
675 nfscl_nfs4_start_probes[nd->nd_procnum];
676 probe_procnum = nd->nd_procnum;
677 } else if (nd->nd_flag & ND_NFSV3) {
678 probe_id = nfscl_nfs3_start_probes[procnum];
679 probe_procnum = procnum;
682 nfscl_nfs2_start_probes[nd->nd_procnum];
683 probe_procnum = procnum;
686 (dtrace_nfscl_nfs234_start_probe)
687 (probe_id, vp, nd->nd_mreq, cred,
693 freeslot = -1; /* Set to slot that needs to be free'd */
695 slot = -1; /* Slot that needs a sequence# increment. */
697 * This timeout specifies when a new socket should be created,
698 * along with new xid values. For UDP, this should be done
699 * infrequently, since retransmits of RPC requests should normally
705 timo.tv_sec = NFSV4_UPCALLTIMEO;
707 timo.tv_sec = NFSV4_CALLBACKTIMEO;
709 if (nrp->nr_sotype != SOCK_DGRAM) {
711 if ((nmp->nm_flag & NFSMNT_NFSV4))
712 timo.tv_sec = INT_MAX;
714 timo.tv_sec = NFS_TCPTIMEO;
716 if (NFSHASSOFT(nmp)) {
718 * CLSET_RETRIES is set to 2, so this should be
719 * half of the total timeout required.
721 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
724 timo.tv_sec = timeo / NFS_HZ;
725 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
728 /* For UDP hard mounts, use a large value. */
729 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
738 * Chain request into list of outstanding requests.
741 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
747 if (clp != NULL && sep != NULL)
748 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
749 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
751 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
752 nd->nd_mreq, &nd->nd_mrep, timo);
756 * RPC done, unlink the request.
759 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
764 * If there was a successful reply and a tprintf msg.
765 * tprintf a response.
767 if (stat == RPC_SUCCESS) {
769 } else if (stat == RPC_TIMEDOUT) {
770 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
772 } else if (stat == RPC_VERSMISMATCH) {
773 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
775 } else if (stat == RPC_PROGVERSMISMATCH) {
776 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
777 error = EPROTONOSUPPORT;
778 } else if (stat == RPC_INTR) {
781 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
785 m_freem(nd->nd_mreq);
789 FREE((caddr_t)rep, M_NFSDREQ);
791 newnfs_restore_sigmask(td, &oldset);
795 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
798 * Search for any mbufs that are not a multiple of 4 bytes long
799 * or with m_data not longword aligned.
800 * These could cause pointer alignment problems, so copy them to
801 * well aligned mbufs.
803 newnfs_realign(&nd->nd_mrep, M_WAITOK);
804 nd->nd_md = nd->nd_mrep;
805 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
807 if (nd->nd_procnum != NFSPROC_NULL &&
808 nd->nd_procnum != NFSV4PROC_CBNULL) {
809 /* If sep == NULL, set it to the default in nmp. */
810 if (sep == NULL && nmp != NULL)
811 sep = nfsmnt_mdssession(nmp);
813 * and now the actual NFS xdr.
815 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
816 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
817 if (nd->nd_repstat >= 10000)
818 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
819 (int)nd->nd_repstat);
822 * Get rid of the tag, return count and SEQUENCE result for
825 if ((nd->nd_flag & ND_NFSV4) != 0) {
826 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
827 i = fxdr_unsigned(int, *tl);
828 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
831 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
832 opcnt = fxdr_unsigned(int, *tl++);
833 i = fxdr_unsigned(int, *tl++);
834 j = fxdr_unsigned(int, *tl);
836 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
838 * If the first op is Sequence, free up the slot.
840 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
841 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
842 NFSCL_DEBUG(1, "failed seq=%d\n", j);
843 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
844 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j == 0)
846 if (i == NFSV4OP_SEQUENCE)
847 NFSM_DISSECT(tl, uint32_t *,
851 NFSM_DISSECT(tl, uint32_t *,
854 mtx_lock(&sep->nfsess_mtx);
855 if (bcmp(tl, sep->nfsess_sessionid,
856 NFSX_V4SESSIONID) == 0) {
857 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
858 retseq = fxdr_unsigned(uint32_t, *tl++);
859 slot = fxdr_unsigned(int, *tl++);
861 if (retseq != sep->nfsess_slotseq[slot])
862 printf("retseq diff 0x%x\n",
864 retval = fxdr_unsigned(uint32_t, *++tl);
865 if ((retval + 1) < sep->nfsess_foreslots
867 sep->nfsess_foreslots = (retval
869 else if ((retval + 1) >
870 sep->nfsess_foreslots)
871 sep->nfsess_foreslots = (retval
872 < 64) ? (retval + 1) : 64;
874 mtx_unlock(&sep->nfsess_mtx);
876 /* Grab the op and status for the next one. */
878 NFSM_DISSECT(tl, uint32_t *,
880 i = fxdr_unsigned(int, *tl++);
881 j = fxdr_unsigned(int, *tl);
885 if (nd->nd_repstat != 0) {
886 if (nd->nd_repstat == NFSERR_BADSESSION &&
887 nmp != NULL && dssep == NULL) {
889 * If this is a client side MDS RPC, mark
890 * the MDS session defunct and initiate
891 * recovery, as required.
892 * The nfsess_defunct field is protected by
893 * the NFSLOCKMNT()/nm_mtx lock and not the
894 * nfsess_mtx lock to simplify its handling,
895 * for the MDS session. This lock is also
896 * sufficient for nfsess_sessionid, since it
897 * never changes in the structure.
899 NFSCL_DEBUG(1, "Got badsession\n");
902 sep = NFSMNT_MDSSESSION(nmp);
903 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
904 NFSX_V4SESSIONID) == 0) {
905 /* Initiate recovery. */
906 sep->nfsess_defunct = 1;
907 NFSCL_DEBUG(1, "Marked defunct\n");
908 if (nmp->nm_clp != NULL) {
909 nmp->nm_clp->nfsc_flags |=
916 * Sleep for up to 1sec waiting for a new
919 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
922 * Get the session again, in case a new one
923 * has been created during the sleep.
925 sep = NFSMNT_MDSSESSION(nmp);
927 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
928 reterr = nfsv4_sequencelookup(nmp, sep,
929 &slotpos, &maxslot, &slotseq,
932 /* Fill in new session info. */
934 "Filling in new sequence\n");
935 tl = nd->nd_sequence;
938 tl += NFSX_V4SESSIONID /
940 *tl++ = txdr_unsigned(slotseq);
941 *tl++ = txdr_unsigned(slotpos);
942 *tl = txdr_unsigned(maxslot);
944 if (reterr == NFSERR_BADSESSION ||
947 "Badsession looping\n");
948 m_freem(nd->nd_mrep);
952 nd->nd_repstat = reterr;
953 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
956 if (((nd->nd_repstat == NFSERR_DELAY ||
957 nd->nd_repstat == NFSERR_GRACE) &&
958 (nd->nd_flag & ND_NFSV4) &&
959 nd->nd_procnum != NFSPROC_DELEGRETURN &&
960 nd->nd_procnum != NFSPROC_SETATTR &&
961 nd->nd_procnum != NFSPROC_READ &&
962 nd->nd_procnum != NFSPROC_READDS &&
963 nd->nd_procnum != NFSPROC_WRITE &&
964 nd->nd_procnum != NFSPROC_WRITEDS &&
965 nd->nd_procnum != NFSPROC_OPEN &&
966 nd->nd_procnum != NFSPROC_CREATE &&
967 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
968 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
969 nd->nd_procnum != NFSPROC_CLOSE &&
970 nd->nd_procnum != NFSPROC_LOCK &&
971 nd->nd_procnum != NFSPROC_LOCKU) ||
972 (nd->nd_repstat == NFSERR_DELAY &&
973 (nd->nd_flag & ND_NFSV4) == 0) ||
974 nd->nd_repstat == NFSERR_RESOURCE) {
975 if (trylater_delay > NFS_TRYLATERDEL)
976 trylater_delay = NFS_TRYLATERDEL;
977 waituntil = NFSD_MONOSEC + trylater_delay;
978 while (NFSD_MONOSEC < waituntil)
979 (void) nfs_catnap(PZERO, 0, "nfstry");
982 mtx_lock(&sep->nfsess_mtx);
983 sep->nfsess_slotseq[slot]++;
984 *nd->nd_slotseq = txdr_unsigned(
985 sep->nfsess_slotseq[slot]);
986 mtx_unlock(&sep->nfsess_mtx);
988 m_freem(nd->nd_mrep);
994 * If the File Handle was stale, invalidate the
995 * lookup cache, just in case.
996 * (vp != NULL implies a client side call)
998 if (nd->nd_repstat == ESTALE && vp != NULL) {
1000 if (ncl_call_invalcaches != NULL)
1001 (*ncl_call_invalcaches)(vp);
1004 if ((nd->nd_flag & ND_NFSV4) != 0) {
1005 /* Free the slot, as required. */
1007 nfsv4_freeslot(sep, freeslot);
1009 * If this op is Putfh, throw its results away.
1012 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1013 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1014 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1015 i = fxdr_unsigned(int, *tl++);
1016 j = fxdr_unsigned(int, *tl);
1018 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1021 * All Compounds that do an Op that must
1022 * be in sequence consist of NFSV4OP_PUTFH
1023 * followed by one of these. As such, we
1024 * can determine if the seqid# should be
1025 * incremented, here.
1027 if ((i == NFSV4OP_OPEN ||
1028 i == NFSV4OP_OPENCONFIRM ||
1029 i == NFSV4OP_OPENDOWNGRADE ||
1030 i == NFSV4OP_CLOSE ||
1031 i == NFSV4OP_LOCK ||
1032 i == NFSV4OP_LOCKU) &&
1034 (j != NFSERR_STALECLIENTID &&
1035 j != NFSERR_STALESTATEID &&
1036 j != NFSERR_BADSTATEID &&
1037 j != NFSERR_BADSEQID &&
1038 j != NFSERR_BADXDR &&
1039 j != NFSERR_RESOURCE &&
1040 j != NFSERR_NOFILEHANDLE)))
1041 nd->nd_flag |= ND_INCRSEQID;
1044 * If this op's status is non-zero, mark
1045 * that there is no more data to process.
1046 * The exception is Setattr, which always has xdr
1047 * when it has failed.
1049 if (j != 0 && i != NFSV4OP_SETATTR)
1050 nd->nd_flag |= ND_NOMOREDATA;
1053 * If R_DONTRECOVER is set, replace the stale error
1054 * reply, so that recovery isn't initiated.
1056 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1057 nd->nd_repstat == NFSERR_BADSESSION ||
1058 nd->nd_repstat == NFSERR_STALESTATEID) &&
1059 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1060 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1064 #ifdef KDTRACE_HOOKS
1065 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1069 if (nd->nd_flag & ND_NFSV4) {
1070 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1071 probe_procnum = nd->nd_procnum;
1072 } else if (nd->nd_flag & ND_NFSV3) {
1073 probe_id = nfscl_nfs3_done_probes[procnum];
1074 probe_procnum = procnum;
1076 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1077 probe_procnum = procnum;
1080 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1081 nd->nd_mreq, cred, probe_procnum, 0);
1085 m_freem(nd->nd_mreq);
1086 if (usegssname == 0)
1089 FREE((caddr_t)rep, M_NFSDREQ);
1091 newnfs_restore_sigmask(td, &oldset);
1094 mbuf_freem(nd->nd_mrep);
1095 mbuf_freem(nd->nd_mreq);
1096 if (usegssname == 0)
1099 FREE((caddr_t)rep, M_NFSDREQ);
1101 newnfs_restore_sigmask(td, &oldset);
1106 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1107 * wait for all requests to complete. This is used by forced unmounts
1108 * to terminate any outstanding RPCs.
1111 newnfs_nmcancelreqs(struct nfsmount *nmp)
1114 if (nmp->nm_sockreq.nr_client != NULL)
1115 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1120 * Any signal that can interrupt an NFS operation in an intr mount
1121 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1123 int newnfs_sig_set[] = {
1132 * Check to see if one of the signals in our subset is pending on
1133 * the process (in an intr mount).
1136 nfs_sig_pending(sigset_t set)
1140 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1141 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1147 * The set/restore sigmask functions are used to (temporarily) overwrite
1148 * the thread td_sigmask during an RPC call (for example). These are also
1149 * used in other places in the NFS client that might tsleep().
1152 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1160 td = curthread; /* XXX */
1162 /* Remove the NFS set of signals from newset */
1164 mtx_lock(&p->p_sigacts->ps_mtx);
1165 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1167 * But make sure we leave the ones already masked
1168 * by the process, ie. remove the signal from the
1169 * temporary signalmask only if it wasn't already
1172 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1173 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1174 SIGDELSET(newset, newnfs_sig_set[i]);
1176 mtx_unlock(&p->p_sigacts->ps_mtx);
1177 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1178 SIGPROCMASK_PROC_LOCKED);
1183 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1186 td = curthread; /* XXX */
1187 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1191 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1192 * old one after msleep() returns.
1195 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1201 if ((priority & PCATCH) == 0)
1202 return msleep(ident, mtx, priority, wmesg, timo);
1204 td = curthread; /* XXX */
1205 newnfs_set_sigmask(td, &oldset);
1206 error = msleep(ident, mtx, priority, wmesg, timo);
1207 newnfs_restore_sigmask(td, &oldset);
1213 * Test for a termination condition pending on the process.
1214 * This is used for NFSMNT_INT mounts.
1217 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1222 /* Terminate all requests while attempting a forced unmount. */
1223 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1225 if (!(nmp->nm_flag & NFSMNT_INT))
1231 tmpset = p->p_siglist;
1232 SIGSETOR(tmpset, td->td_siglist);
1233 SIGSETNAND(tmpset, td->td_sigmask);
1234 mtx_lock(&p->p_sigacts->ps_mtx);
1235 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1236 mtx_unlock(&p->p_sigacts->ps_mtx);
1237 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1238 && nfs_sig_pending(tmpset)) {
1247 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1251 p = td ? td->td_proc : NULL;
1253 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1254 server, msg, error);
1256 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1262 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1263 int error, int flags)
1267 mtx_lock(&nmp->nm_mtx);
1268 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1269 nmp->nm_state |= NFSSTA_TIMEO;
1270 mtx_unlock(&nmp->nm_mtx);
1271 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1274 mtx_unlock(&nmp->nm_mtx);
1275 mtx_lock(&nmp->nm_mtx);
1276 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1277 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1278 mtx_unlock(&nmp->nm_mtx);
1279 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1282 mtx_unlock(&nmp->nm_mtx);
1283 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1287 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1288 int flags, int tprintfmsg)
1293 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1296 mtx_lock(&nmp->nm_mtx);
1297 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1298 nmp->nm_state &= ~NFSSTA_TIMEO;
1299 mtx_unlock(&nmp->nm_mtx);
1300 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1303 mtx_unlock(&nmp->nm_mtx);
1305 mtx_lock(&nmp->nm_mtx);
1306 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1307 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1308 mtx_unlock(&nmp->nm_mtx);
1309 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1312 mtx_unlock(&nmp->nm_mtx);