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 * 4. 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 if (NFSHASNFSV4N(nmp)) {
285 * Make sure the nfscbd_pool doesn't get destroyed
289 if (nfs_numnfscbd > 0) {
292 xprt = svc_vc_create_backchannel(nfscbd_pool);
293 CLNT_CONTROL(client, CLSET_BACKCHANNEL, xprt);
296 if (nfs_numnfscbd == 0)
297 wakeup(&nfs_numnfscbd);
304 * - Null RPC callback to client
305 * - Non-Null RPC callback to client, wait a little longer
306 * - upcalls to nfsuserd and gssd (clp == NULL)
308 if (callback_retry_mult == 0) {
309 retries = NFSV4_UPCALLRETRY;
310 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
312 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
315 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
319 * For UDP, there are 2 timeouts:
320 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
321 * that does a retransmit of an RPC request using the same
322 * socket and xid. This is what you normally want to do,
323 * since NFS servers depend on "same xid" for their
324 * Duplicate Request Cache.
325 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
326 * retransmits on the same socket should fail and a fresh
327 * socket created. Each of these timeouts counts as one
328 * CLSET_RETRIES as set above.
329 * Set the initial retransmit timeout for UDP. This timeout
330 * doesn't exist for TCP and the following call just fails,
333 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
334 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
335 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
338 mtx_lock(&nrp->nr_mtx);
339 if (nrp->nr_client != NULL) {
340 mtx_unlock(&nrp->nr_mtx);
342 * Someone else already connected.
344 CLNT_RELEASE(client);
346 nrp->nr_client = client;
348 * Protocols that do not require connections may be optionally
349 * left unconnected for servers that reply from a port other
352 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
353 mtx_unlock(&nrp->nr_mtx);
354 CLNT_CONTROL(client, CLSET_CONNECT, &one);
356 mtx_unlock(&nrp->nr_mtx);
360 /* Restore current thread's credentials. */
361 td->td_ucred = origcred;
369 * NFS disconnect. Clean up and unlink.
372 newnfs_disconnect(struct nfssockreq *nrp)
376 mtx_lock(&nrp->nr_mtx);
377 if (nrp->nr_client != NULL) {
378 client = nrp->nr_client;
379 nrp->nr_client = NULL;
380 mtx_unlock(&nrp->nr_mtx);
381 rpc_gss_secpurge_call(client);
383 CLNT_RELEASE(client);
385 mtx_unlock(&nrp->nr_mtx);
390 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
391 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
393 rpc_gss_service_t svc;
396 switch (secflavour) {
397 case RPCSEC_GSS_KRB5:
398 case RPCSEC_GSS_KRB5I:
399 case RPCSEC_GSS_KRB5P:
401 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
404 if (secflavour == RPCSEC_GSS_KRB5)
405 svc = rpc_gss_svc_none;
406 else if (secflavour == RPCSEC_GSS_KRB5I)
407 svc = rpc_gss_svc_integrity;
409 svc = rpc_gss_svc_privacy;
411 if (clnt_principal == NULL)
412 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
413 srv_principal, mech_oid, svc);
415 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
416 clnt_principal, srv_principal, "kerberosv5",
417 svc, NULL, NULL, NULL);
425 return (authunix_create(cred));
431 * Callback from the RPC code to generate up/down notifications.
434 struct nfs_feedback_arg {
435 struct nfsmount *nf_mount;
436 int nf_lastmsg; /* last tprintf */
438 struct thread *nf_td;
442 nfs_feedback(int type, int proc, void *arg)
444 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
445 struct nfsmount *nmp = nf->nf_mount;
449 case FEEDBACK_REXMIT2:
450 case FEEDBACK_RECONNECT:
452 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
453 nfs_down(nmp, nf->nf_td,
454 "not responding", 0, NFSSTA_TIMEO);
455 nf->nf_tprintfmsg = TRUE;
456 nf->nf_lastmsg = now;
461 nfs_up(nf->nf_mount, nf->nf_td,
462 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
468 * newnfs_request - goes something like this
469 * - does the rpc by calling the krpc layer
470 * - break down rpc header and return with nfs reply
471 * nb: always frees up nd_mreq mbuf list
474 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
475 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
476 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
477 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
479 uint32_t retseq, retval, slotseq, *tl;
481 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
482 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
483 int freeslot, maxslot, reterr, slotpos, timeo;
485 u_int trylater_delay = 1;
486 struct nfs_feedback_arg nf;
489 struct rpc_callextra ext;
491 struct nfsreq *rep = NULL;
492 char *srv_principal = NULL, *clnt_principal = NULL;
494 struct ucred *authcred;
495 struct nfsclsession *sep;
496 uint8_t sessionid[NFSX_V4SESSIONID];
501 /* Reject requests while attempting a forced unmount. */
502 if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
503 m_freem(nd->nd_mreq);
508 * Set authcred, which is used to acquire RPC credentials to
509 * the cred argument, by default. The crhold() should not be
510 * necessary, but will ensure that some future code change
511 * doesn't result in the credential being free'd prematurely.
513 authcred = crhold(cred);
515 /* For client side interruptible mounts, mask off the signals. */
516 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
517 newnfs_set_sigmask(td, &oldset);
522 * XXX if not already connected call nfs_connect now. Longer
523 * term, change nfs_mount to call nfs_connect unconditionally
524 * and let clnt_reconnect_create handle reconnects.
526 if (nrp->nr_client == NULL)
527 newnfs_connect(nmp, nrp, cred, td, 0);
530 * For a client side mount, nmp is != NULL and clp == NULL. For
531 * server calls (callbacks or upcalls), nmp == NULL.
535 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
536 secflavour = RPCSEC_GSS_KRB5;
537 if (nd->nd_procnum != NFSPROC_NULL) {
538 if (clp->lc_flags & LCL_GSSINTEGRITY)
539 secflavour = RPCSEC_GSS_KRB5I;
540 else if (clp->lc_flags & LCL_GSSPRIVACY)
541 secflavour = RPCSEC_GSS_KRB5P;
545 } else if (nmp != NULL && NFSHASKERB(nmp) &&
546 nd->nd_procnum != NFSPROC_NULL) {
547 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
548 nd->nd_flag |= ND_USEGSSNAME;
549 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
551 * If there is a client side host based credential,
552 * use that, otherwise use the system uid, if set.
553 * The system uid is in the nmp->nm_sockreq.nr_cred
556 if (nmp->nm_krbnamelen > 0) {
558 clnt_principal = nmp->nm_krbname;
559 } else if (nmp->nm_uid != (uid_t)-1) {
560 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
561 ("newnfs_request: NULL nr_cred"));
563 authcred = crhold(nmp->nm_sockreq.nr_cred);
565 } else if (nmp->nm_krbnamelen == 0 &&
566 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
568 * If there is no host based principal name and
569 * the system uid is set and this is root, use the
570 * system uid, since root won't have user
571 * credentials in a credentials cache file.
572 * The system uid is in the nmp->nm_sockreq.nr_cred
575 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
576 ("newnfs_request: NULL nr_cred"));
578 authcred = crhold(nmp->nm_sockreq.nr_cred);
580 if (NFSHASINTEGRITY(nmp))
581 secflavour = RPCSEC_GSS_KRB5I;
582 else if (NFSHASPRIVACY(nmp))
583 secflavour = RPCSEC_GSS_KRB5P;
585 secflavour = RPCSEC_GSS_KRB5;
586 srv_principal = NFSMNT_SRVKRBNAME(nmp);
587 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
588 nd->nd_procnum != NFSPROC_NULL &&
589 (nd->nd_flag & ND_USEGSSNAME) != 0) {
591 * Use the uid that did the mount when the RPC is doing
592 * NFSv4 system operations, as indicated by the
593 * ND_USEGSSNAME flag, for the AUTH_SYS case.
594 * The credentials in nm_sockreq.nr_cred were used for the
597 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
598 ("newnfs_request: NULL nr_cred"));
600 authcred = crhold(nmp->nm_sockreq.nr_cred);
604 bzero(&nf, sizeof(struct nfs_feedback_arg));
607 nf.nf_lastmsg = NFSD_MONOSEC -
608 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
611 if (nd->nd_procnum == NFSPROC_NULL)
612 auth = authnone_create();
613 else if (usegssname) {
615 * For this case, the authenticator is held in the
616 * nfssockreq structure, so don't release the reference count
617 * held on it. --> Don't AUTH_DESTROY() it in this function.
619 if (nrp->nr_auth == NULL)
620 nrp->nr_auth = nfs_getauth(nrp, secflavour,
621 clnt_principal, srv_principal, NULL, authcred);
623 rpc_gss_refresh_auth_call(nrp->nr_auth);
626 auth = nfs_getauth(nrp, secflavour, NULL,
627 srv_principal, NULL, authcred);
630 m_freem(nd->nd_mreq);
632 newnfs_restore_sigmask(td, &oldset);
635 bzero(&ext, sizeof(ext));
638 ext.rc_feedback = nfs_feedback;
639 ext.rc_feedback_arg = &nf;
642 procnum = nd->nd_procnum;
643 if ((nd->nd_flag & ND_NFSV4) &&
644 nd->nd_procnum != NFSPROC_NULL &&
645 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
646 procnum = NFSV4PROC_COMPOUND;
649 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
651 /* Map the procnum to the old NFSv2 one, as required. */
652 if ((nd->nd_flag & ND_NFSV2) != 0) {
653 if (nd->nd_procnum < NFS_V3NPROCS)
654 procnum = nfsv2_procid[nd->nd_procnum];
656 procnum = NFSV2PROC_NOOP;
660 * Now only used for the R_DONTRECOVER case, but until that is
661 * supported within the krpc code, I need to keep a queue of
662 * outstanding RPCs for nfsv4 client requests.
664 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
665 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
666 M_NFSDREQ, M_WAITOK);
668 if (dtrace_nfscl_nfs234_start_probe != NULL) {
672 if (nd->nd_flag & ND_NFSV4) {
674 nfscl_nfs4_start_probes[nd->nd_procnum];
675 probe_procnum = nd->nd_procnum;
676 } else if (nd->nd_flag & ND_NFSV3) {
677 probe_id = nfscl_nfs3_start_probes[procnum];
678 probe_procnum = procnum;
681 nfscl_nfs2_start_probes[nd->nd_procnum];
682 probe_procnum = procnum;
685 (dtrace_nfscl_nfs234_start_probe)
686 (probe_id, vp, nd->nd_mreq, cred,
692 freeslot = -1; /* Set to slot that needs to be free'd */
694 slot = -1; /* Slot that needs a sequence# increment. */
696 * This timeout specifies when a new socket should be created,
697 * along with new xid values. For UDP, this should be done
698 * infrequently, since retransmits of RPC requests should normally
704 timo.tv_sec = NFSV4_UPCALLTIMEO;
706 timo.tv_sec = NFSV4_CALLBACKTIMEO;
708 if (nrp->nr_sotype != SOCK_DGRAM) {
710 if ((nmp->nm_flag & NFSMNT_NFSV4))
711 timo.tv_sec = INT_MAX;
713 timo.tv_sec = NFS_TCPTIMEO;
715 if (NFSHASSOFT(nmp)) {
717 * CLSET_RETRIES is set to 2, so this should be
718 * half of the total timeout required.
720 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
723 timo.tv_sec = timeo / NFS_HZ;
724 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
727 /* For UDP hard mounts, use a large value. */
728 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
737 * Chain request into list of outstanding requests.
740 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
746 if (clp != NULL && sep != NULL)
747 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
748 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
750 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
751 nd->nd_mreq, &nd->nd_mrep, timo);
755 * RPC done, unlink the request.
758 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
763 * If there was a successful reply and a tprintf msg.
764 * tprintf a response.
766 if (stat == RPC_SUCCESS) {
768 } else if (stat == RPC_TIMEDOUT) {
769 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
771 } else if (stat == RPC_VERSMISMATCH) {
772 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
774 } else if (stat == RPC_PROGVERSMISMATCH) {
775 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
776 error = EPROTONOSUPPORT;
777 } else if (stat == RPC_INTR) {
780 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
784 m_freem(nd->nd_mreq);
788 FREE((caddr_t)rep, M_NFSDREQ);
790 newnfs_restore_sigmask(td, &oldset);
794 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
797 * Search for any mbufs that are not a multiple of 4 bytes long
798 * or with m_data not longword aligned.
799 * These could cause pointer alignment problems, so copy them to
800 * well aligned mbufs.
802 newnfs_realign(&nd->nd_mrep, M_WAITOK);
803 nd->nd_md = nd->nd_mrep;
804 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
806 if (nd->nd_procnum != NFSPROC_NULL &&
807 nd->nd_procnum != NFSV4PROC_CBNULL) {
808 /* If sep == NULL, set it to the default in nmp. */
809 if (sep == NULL && nmp != NULL)
810 sep = nfsmnt_mdssession(nmp);
812 * and now the actual NFS xdr.
814 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
815 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
816 if (nd->nd_repstat >= 10000)
817 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
818 (int)nd->nd_repstat);
821 * Get rid of the tag, return count and SEQUENCE result for
824 if ((nd->nd_flag & ND_NFSV4) != 0) {
825 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
826 i = fxdr_unsigned(int, *tl);
827 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
830 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
831 opcnt = fxdr_unsigned(int, *tl++);
832 i = fxdr_unsigned(int, *tl++);
833 j = fxdr_unsigned(int, *tl);
835 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
837 * If the first op is Sequence, free up the slot.
839 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
840 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
841 NFSCL_DEBUG(1, "failed seq=%d\n", j);
842 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
843 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j == 0)
845 if (i == NFSV4OP_SEQUENCE)
846 NFSM_DISSECT(tl, uint32_t *,
850 NFSM_DISSECT(tl, uint32_t *,
853 mtx_lock(&sep->nfsess_mtx);
854 if (bcmp(tl, sep->nfsess_sessionid,
855 NFSX_V4SESSIONID) == 0) {
856 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
857 retseq = fxdr_unsigned(uint32_t, *tl++);
858 slot = fxdr_unsigned(int, *tl++);
860 if (retseq != sep->nfsess_slotseq[slot])
861 printf("retseq diff 0x%x\n",
863 retval = fxdr_unsigned(uint32_t, *++tl);
864 if ((retval + 1) < sep->nfsess_foreslots
866 sep->nfsess_foreslots = (retval
868 else if ((retval + 1) >
869 sep->nfsess_foreslots)
870 sep->nfsess_foreslots = (retval
871 < 64) ? (retval + 1) : 64;
873 mtx_unlock(&sep->nfsess_mtx);
875 /* Grab the op and status for the next one. */
877 NFSM_DISSECT(tl, uint32_t *,
879 i = fxdr_unsigned(int, *tl++);
880 j = fxdr_unsigned(int, *tl);
884 if (nd->nd_repstat != 0) {
885 if (nd->nd_repstat == NFSERR_BADSESSION &&
886 nmp != NULL && dssep == NULL) {
888 * If this is a client side MDS RPC, mark
889 * the MDS session defunct and initiate
890 * recovery, as required.
891 * The nfsess_defunct field is protected by
892 * the NFSLOCKMNT()/nm_mtx lock and not the
893 * nfsess_mtx lock to simplify its handling,
894 * for the MDS session. This lock is also
895 * sufficient for nfsess_sessionid, since it
896 * never changes in the structure.
898 NFSCL_DEBUG(1, "Got badsession\n");
901 sep = NFSMNT_MDSSESSION(nmp);
902 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
903 NFSX_V4SESSIONID) == 0) {
904 /* Initiate recovery. */
905 sep->nfsess_defunct = 1;
906 NFSCL_DEBUG(1, "Marked defunct\n");
907 if (nmp->nm_clp != NULL) {
908 nmp->nm_clp->nfsc_flags |=
915 * Sleep for up to 1sec waiting for a new
918 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
921 * Get the session again, in case a new one
922 * has been created during the sleep.
924 sep = NFSMNT_MDSSESSION(nmp);
926 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
927 reterr = nfsv4_sequencelookup(nmp, sep,
928 &slotpos, &maxslot, &slotseq,
931 /* Fill in new session info. */
933 "Filling in new sequence\n");
934 tl = nd->nd_sequence;
937 tl += NFSX_V4SESSIONID /
939 *tl++ = txdr_unsigned(slotseq);
940 *tl++ = txdr_unsigned(slotpos);
941 *tl = txdr_unsigned(maxslot);
943 if (reterr == NFSERR_BADSESSION ||
946 "Badsession looping\n");
947 m_freem(nd->nd_mrep);
951 nd->nd_repstat = reterr;
952 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
955 if (((nd->nd_repstat == NFSERR_DELAY ||
956 nd->nd_repstat == NFSERR_GRACE) &&
957 (nd->nd_flag & ND_NFSV4) &&
958 nd->nd_procnum != NFSPROC_DELEGRETURN &&
959 nd->nd_procnum != NFSPROC_SETATTR &&
960 nd->nd_procnum != NFSPROC_READ &&
961 nd->nd_procnum != NFSPROC_READDS &&
962 nd->nd_procnum != NFSPROC_WRITE &&
963 nd->nd_procnum != NFSPROC_WRITEDS &&
964 nd->nd_procnum != NFSPROC_OPEN &&
965 nd->nd_procnum != NFSPROC_CREATE &&
966 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
967 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
968 nd->nd_procnum != NFSPROC_CLOSE &&
969 nd->nd_procnum != NFSPROC_LOCK &&
970 nd->nd_procnum != NFSPROC_LOCKU) ||
971 (nd->nd_repstat == NFSERR_DELAY &&
972 (nd->nd_flag & ND_NFSV4) == 0) ||
973 nd->nd_repstat == NFSERR_RESOURCE) {
974 if (trylater_delay > NFS_TRYLATERDEL)
975 trylater_delay = NFS_TRYLATERDEL;
976 waituntil = NFSD_MONOSEC + trylater_delay;
977 while (NFSD_MONOSEC < waituntil)
978 (void) nfs_catnap(PZERO, 0, "nfstry");
981 mtx_lock(&sep->nfsess_mtx);
982 sep->nfsess_slotseq[slot]++;
983 *nd->nd_slotseq = txdr_unsigned(
984 sep->nfsess_slotseq[slot]);
985 mtx_unlock(&sep->nfsess_mtx);
987 m_freem(nd->nd_mrep);
993 * If the File Handle was stale, invalidate the
994 * lookup cache, just in case.
995 * (vp != NULL implies a client side call)
997 if (nd->nd_repstat == ESTALE && vp != NULL) {
999 if (ncl_call_invalcaches != NULL)
1000 (*ncl_call_invalcaches)(vp);
1003 if ((nd->nd_flag & ND_NFSV4) != 0) {
1004 /* Free the slot, as required. */
1006 nfsv4_freeslot(sep, freeslot);
1008 * If this op is Putfh, throw its results away.
1011 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1012 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1013 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1014 i = fxdr_unsigned(int, *tl++);
1015 j = fxdr_unsigned(int, *tl);
1017 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1020 * All Compounds that do an Op that must
1021 * be in sequence consist of NFSV4OP_PUTFH
1022 * followed by one of these. As such, we
1023 * can determine if the seqid# should be
1024 * incremented, here.
1026 if ((i == NFSV4OP_OPEN ||
1027 i == NFSV4OP_OPENCONFIRM ||
1028 i == NFSV4OP_OPENDOWNGRADE ||
1029 i == NFSV4OP_CLOSE ||
1030 i == NFSV4OP_LOCK ||
1031 i == NFSV4OP_LOCKU) &&
1033 (j != NFSERR_STALECLIENTID &&
1034 j != NFSERR_STALESTATEID &&
1035 j != NFSERR_BADSTATEID &&
1036 j != NFSERR_BADSEQID &&
1037 j != NFSERR_BADXDR &&
1038 j != NFSERR_RESOURCE &&
1039 j != NFSERR_NOFILEHANDLE)))
1040 nd->nd_flag |= ND_INCRSEQID;
1043 * If this op's status is non-zero, mark
1044 * that there is no more data to process.
1047 nd->nd_flag |= ND_NOMOREDATA;
1050 * If R_DONTRECOVER is set, replace the stale error
1051 * reply, so that recovery isn't initiated.
1053 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1054 nd->nd_repstat == NFSERR_BADSESSION ||
1055 nd->nd_repstat == NFSERR_STALESTATEID) &&
1056 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1057 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1061 #ifdef KDTRACE_HOOKS
1062 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1066 if (nd->nd_flag & ND_NFSV4) {
1067 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1068 probe_procnum = nd->nd_procnum;
1069 } else if (nd->nd_flag & ND_NFSV3) {
1070 probe_id = nfscl_nfs3_done_probes[procnum];
1071 probe_procnum = procnum;
1073 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1074 probe_procnum = procnum;
1077 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1078 nd->nd_mreq, cred, probe_procnum, 0);
1082 m_freem(nd->nd_mreq);
1083 if (usegssname == 0)
1086 FREE((caddr_t)rep, M_NFSDREQ);
1088 newnfs_restore_sigmask(td, &oldset);
1091 mbuf_freem(nd->nd_mrep);
1092 mbuf_freem(nd->nd_mreq);
1093 if (usegssname == 0)
1096 FREE((caddr_t)rep, M_NFSDREQ);
1098 newnfs_restore_sigmask(td, &oldset);
1103 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1104 * wait for all requests to complete. This is used by forced unmounts
1105 * to terminate any outstanding RPCs.
1108 newnfs_nmcancelreqs(struct nfsmount *nmp)
1111 if (nmp->nm_sockreq.nr_client != NULL)
1112 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1117 * Any signal that can interrupt an NFS operation in an intr mount
1118 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1120 int newnfs_sig_set[] = {
1129 * Check to see if one of the signals in our subset is pending on
1130 * the process (in an intr mount).
1133 nfs_sig_pending(sigset_t set)
1137 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1138 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1144 * The set/restore sigmask functions are used to (temporarily) overwrite
1145 * the thread td_sigmask during an RPC call (for example). These are also
1146 * used in other places in the NFS client that might tsleep().
1149 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1157 td = curthread; /* XXX */
1159 /* Remove the NFS set of signals from newset */
1161 mtx_lock(&p->p_sigacts->ps_mtx);
1162 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1164 * But make sure we leave the ones already masked
1165 * by the process, ie. remove the signal from the
1166 * temporary signalmask only if it wasn't already
1169 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1170 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1171 SIGDELSET(newset, newnfs_sig_set[i]);
1173 mtx_unlock(&p->p_sigacts->ps_mtx);
1174 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1175 SIGPROCMASK_PROC_LOCKED);
1180 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1183 td = curthread; /* XXX */
1184 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1188 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1189 * old one after msleep() returns.
1192 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1198 if ((priority & PCATCH) == 0)
1199 return msleep(ident, mtx, priority, wmesg, timo);
1201 td = curthread; /* XXX */
1202 newnfs_set_sigmask(td, &oldset);
1203 error = msleep(ident, mtx, priority, wmesg, timo);
1204 newnfs_restore_sigmask(td, &oldset);
1210 * Test for a termination condition pending on the process.
1211 * This is used for NFSMNT_INT mounts.
1214 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1219 /* Terminate all requests while attempting a forced unmount. */
1220 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1222 if (!(nmp->nm_flag & NFSMNT_INT))
1228 tmpset = p->p_siglist;
1229 SIGSETOR(tmpset, td->td_siglist);
1230 SIGSETNAND(tmpset, td->td_sigmask);
1231 mtx_lock(&p->p_sigacts->ps_mtx);
1232 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1233 mtx_unlock(&p->p_sigacts->ps_mtx);
1234 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1235 && nfs_sig_pending(tmpset)) {
1244 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1248 p = td ? td->td_proc : NULL;
1250 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1251 server, msg, error);
1253 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1259 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1260 int error, int flags)
1264 mtx_lock(&nmp->nm_mtx);
1265 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1266 nmp->nm_state |= NFSSTA_TIMEO;
1267 mtx_unlock(&nmp->nm_mtx);
1268 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1271 mtx_unlock(&nmp->nm_mtx);
1272 mtx_lock(&nmp->nm_mtx);
1273 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1274 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1275 mtx_unlock(&nmp->nm_mtx);
1276 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1279 mtx_unlock(&nmp->nm_mtx);
1280 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1284 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1285 int flags, int tprintfmsg)
1290 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1293 mtx_lock(&nmp->nm_mtx);
1294 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1295 nmp->nm_state &= ~NFSSTA_TIMEO;
1296 mtx_unlock(&nmp->nm_mtx);
1297 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1300 mtx_unlock(&nmp->nm_mtx);
1302 mtx_lock(&nmp->nm_mtx);
1303 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1304 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1305 mtx_unlock(&nmp->nm_mtx);
1306 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1309 mtx_unlock(&nmp->nm_mtx);