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;
164 int pktscale, pktscalesav;
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");
202 else if (saddr->sa_family == AF_LOCAL)
203 nconf = getnetconfigent("local");
205 if (nrp->nr_sotype == SOCK_DGRAM)
206 nconf = getnetconfigent("udp6");
208 nconf = getnetconfigent("tcp6");
210 pktscale = nfs_bufpackets;
215 pktscalesav = pktscale;
217 * soreserve() can fail if sb_max is too small, so shrink pktscale
218 * and try again if there is an error.
219 * Print a log message suggesting increasing sb_max.
220 * Creating a socket and doing this is necessary since, if the
221 * reservation sizes are too large and will make soreserve() fail,
222 * the connection will work until a large send is attempted and
223 * then it will loop in the krpc code.
226 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
227 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
228 nrp->nr_soproto, td->td_ucred, td);
230 td->td_ucred = origcred;
234 if (error != 0 && pktscale > 2) {
235 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
236 pktscale == pktscalesav)
237 printf("Consider increasing kern.ipc.maxsockbuf\n");
240 if (nrp->nr_sotype == SOCK_DGRAM) {
242 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
244 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
247 sndreserve = rcvreserve = 1024 * pktscale;
250 if (nrp->nr_sotype != SOCK_STREAM)
251 panic("nfscon sotype");
253 sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
254 sizeof (u_int32_t)) * pktscale;
255 rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
256 sizeof (u_int32_t)) * pktscale;
258 sndreserve = rcvreserve = 1024 * pktscale;
261 error = soreserve(so, sndreserve, rcvreserve);
262 if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
264 printf("Must increase kern.ipc.maxsockbuf or reduce"
266 } while (error != 0 && pktscale > 2);
269 td->td_ucred = origcred;
273 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
274 nrp->nr_vers, sndreserve, rcvreserve);
275 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
277 if ((nmp->nm_flag & NFSMNT_INT))
278 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
279 if ((nmp->nm_flag & NFSMNT_RESVPORT))
280 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
281 if (NFSHASSOFT(nmp)) {
282 if (nmp->nm_sotype == SOCK_DGRAM)
284 * For UDP, the large timeout for a reconnect
285 * will be set to "nm_retry * nm_timeo / 2", so
286 * we only want to do 2 reconnect timeout
291 retries = nmp->nm_retry;
294 /* cred == NULL for DS connects. */
295 if (NFSHASNFSV4N(nmp) && cred != NULL) {
297 * Make sure the nfscbd_pool doesn't get destroyed
301 if (nfs_numnfscbd > 0) {
304 xprt = svc_vc_create_backchannel(nfscbd_pool);
305 CLNT_CONTROL(client, CLSET_BACKCHANNEL, xprt);
308 if (nfs_numnfscbd == 0)
309 wakeup(&nfs_numnfscbd);
316 * - Null RPC callback to client
317 * - Non-Null RPC callback to client, wait a little longer
318 * - upcalls to nfsuserd and gssd (clp == NULL)
320 if (callback_retry_mult == 0) {
321 retries = NFSV4_UPCALLRETRY;
322 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
324 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
327 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
331 * For UDP, there are 2 timeouts:
332 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
333 * that does a retransmit of an RPC request using the same
334 * socket and xid. This is what you normally want to do,
335 * since NFS servers depend on "same xid" for their
336 * Duplicate Request Cache.
337 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
338 * retransmits on the same socket should fail and a fresh
339 * socket created. Each of these timeouts counts as one
340 * CLSET_RETRIES as set above.
341 * Set the initial retransmit timeout for UDP. This timeout
342 * doesn't exist for TCP and the following call just fails,
345 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
346 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
347 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
350 mtx_lock(&nrp->nr_mtx);
351 if (nrp->nr_client != NULL) {
352 mtx_unlock(&nrp->nr_mtx);
354 * Someone else already connected.
356 CLNT_RELEASE(client);
358 nrp->nr_client = client;
360 * Protocols that do not require connections may be optionally
361 * left unconnected for servers that reply from a port other
364 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
365 mtx_unlock(&nrp->nr_mtx);
366 CLNT_CONTROL(client, CLSET_CONNECT, &one);
368 mtx_unlock(&nrp->nr_mtx);
372 /* Restore current thread's credentials. */
373 td->td_ucred = origcred;
381 * NFS disconnect. Clean up and unlink.
384 newnfs_disconnect(struct nfssockreq *nrp)
388 mtx_lock(&nrp->nr_mtx);
389 if (nrp->nr_client != NULL) {
390 client = nrp->nr_client;
391 nrp->nr_client = NULL;
392 mtx_unlock(&nrp->nr_mtx);
393 rpc_gss_secpurge_call(client);
395 CLNT_RELEASE(client);
397 mtx_unlock(&nrp->nr_mtx);
402 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
403 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
405 rpc_gss_service_t svc;
408 switch (secflavour) {
409 case RPCSEC_GSS_KRB5:
410 case RPCSEC_GSS_KRB5I:
411 case RPCSEC_GSS_KRB5P:
413 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
416 if (secflavour == RPCSEC_GSS_KRB5)
417 svc = rpc_gss_svc_none;
418 else if (secflavour == RPCSEC_GSS_KRB5I)
419 svc = rpc_gss_svc_integrity;
421 svc = rpc_gss_svc_privacy;
423 if (clnt_principal == NULL)
424 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
425 srv_principal, mech_oid, svc);
427 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
428 clnt_principal, srv_principal, "kerberosv5",
429 svc, NULL, NULL, NULL);
437 return (authunix_create(cred));
443 * Callback from the RPC code to generate up/down notifications.
446 struct nfs_feedback_arg {
447 struct nfsmount *nf_mount;
448 int nf_lastmsg; /* last tprintf */
450 struct thread *nf_td;
454 nfs_feedback(int type, int proc, void *arg)
456 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
457 struct nfsmount *nmp = nf->nf_mount;
461 case FEEDBACK_REXMIT2:
462 case FEEDBACK_RECONNECT:
464 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
465 nfs_down(nmp, nf->nf_td,
466 "not responding", 0, NFSSTA_TIMEO);
467 nf->nf_tprintfmsg = TRUE;
468 nf->nf_lastmsg = now;
473 nfs_up(nf->nf_mount, nf->nf_td,
474 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
480 * newnfs_request - goes something like this
481 * - does the rpc by calling the krpc layer
482 * - break down rpc header and return with nfs reply
483 * nb: always frees up nd_mreq mbuf list
486 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
487 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
488 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
489 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
491 uint32_t retseq, retval, slotseq, *tl;
493 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
494 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
495 int freeslot, maxslot, reterr, slotpos, timeo;
497 u_int trylater_delay = 1;
498 struct nfs_feedback_arg nf;
501 struct rpc_callextra ext;
503 struct nfsreq *rep = NULL;
504 char *srv_principal = NULL, *clnt_principal = NULL;
506 struct ucred *authcred;
507 struct nfsclsession *sep;
508 uint8_t sessionid[NFSX_V4SESSIONID];
513 /* Reject requests while attempting a forced unmount. */
514 if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
515 m_freem(nd->nd_mreq);
520 * Set authcred, which is used to acquire RPC credentials to
521 * the cred argument, by default. The crhold() should not be
522 * necessary, but will ensure that some future code change
523 * doesn't result in the credential being free'd prematurely.
525 authcred = crhold(cred);
527 /* For client side interruptible mounts, mask off the signals. */
528 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
529 newnfs_set_sigmask(td, &oldset);
534 * XXX if not already connected call nfs_connect now. Longer
535 * term, change nfs_mount to call nfs_connect unconditionally
536 * and let clnt_reconnect_create handle reconnects.
538 if (nrp->nr_client == NULL)
539 newnfs_connect(nmp, nrp, cred, td, 0);
542 * For a client side mount, nmp is != NULL and clp == NULL. For
543 * server calls (callbacks or upcalls), nmp == NULL.
547 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
548 secflavour = RPCSEC_GSS_KRB5;
549 if (nd->nd_procnum != NFSPROC_NULL) {
550 if (clp->lc_flags & LCL_GSSINTEGRITY)
551 secflavour = RPCSEC_GSS_KRB5I;
552 else if (clp->lc_flags & LCL_GSSPRIVACY)
553 secflavour = RPCSEC_GSS_KRB5P;
557 } else if (nmp != NULL && NFSHASKERB(nmp) &&
558 nd->nd_procnum != NFSPROC_NULL) {
559 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
560 nd->nd_flag |= ND_USEGSSNAME;
561 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
563 * If there is a client side host based credential,
564 * use that, otherwise use the system uid, if set.
565 * The system uid is in the nmp->nm_sockreq.nr_cred
568 if (nmp->nm_krbnamelen > 0) {
570 clnt_principal = nmp->nm_krbname;
571 } else if (nmp->nm_uid != (uid_t)-1) {
572 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
573 ("newnfs_request: NULL nr_cred"));
575 authcred = crhold(nmp->nm_sockreq.nr_cred);
577 } else if (nmp->nm_krbnamelen == 0 &&
578 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
580 * If there is no host based principal name and
581 * the system uid is set and this is root, use the
582 * system uid, since root won't have user
583 * credentials in a credentials cache file.
584 * The system uid is in the nmp->nm_sockreq.nr_cred
587 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
588 ("newnfs_request: NULL nr_cred"));
590 authcred = crhold(nmp->nm_sockreq.nr_cred);
592 if (NFSHASINTEGRITY(nmp))
593 secflavour = RPCSEC_GSS_KRB5I;
594 else if (NFSHASPRIVACY(nmp))
595 secflavour = RPCSEC_GSS_KRB5P;
597 secflavour = RPCSEC_GSS_KRB5;
598 srv_principal = NFSMNT_SRVKRBNAME(nmp);
599 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
600 nd->nd_procnum != NFSPROC_NULL &&
601 (nd->nd_flag & ND_USEGSSNAME) != 0) {
603 * Use the uid that did the mount when the RPC is doing
604 * NFSv4 system operations, as indicated by the
605 * ND_USEGSSNAME flag, for the AUTH_SYS case.
606 * The credentials in nm_sockreq.nr_cred were used for the
609 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
610 ("newnfs_request: NULL nr_cred"));
612 authcred = crhold(nmp->nm_sockreq.nr_cred);
616 bzero(&nf, sizeof(struct nfs_feedback_arg));
619 nf.nf_lastmsg = NFSD_MONOSEC -
620 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
623 if (nd->nd_procnum == NFSPROC_NULL)
624 auth = authnone_create();
625 else if (usegssname) {
627 * For this case, the authenticator is held in the
628 * nfssockreq structure, so don't release the reference count
629 * held on it. --> Don't AUTH_DESTROY() it in this function.
631 if (nrp->nr_auth == NULL)
632 nrp->nr_auth = nfs_getauth(nrp, secflavour,
633 clnt_principal, srv_principal, NULL, authcred);
635 rpc_gss_refresh_auth_call(nrp->nr_auth);
638 auth = nfs_getauth(nrp, secflavour, NULL,
639 srv_principal, NULL, authcred);
642 m_freem(nd->nd_mreq);
644 newnfs_restore_sigmask(td, &oldset);
647 bzero(&ext, sizeof(ext));
650 ext.rc_feedback = nfs_feedback;
651 ext.rc_feedback_arg = &nf;
654 procnum = nd->nd_procnum;
655 if ((nd->nd_flag & ND_NFSV4) &&
656 nd->nd_procnum != NFSPROC_NULL &&
657 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
658 procnum = NFSV4PROC_COMPOUND;
661 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
663 /* Map the procnum to the old NFSv2 one, as required. */
664 if ((nd->nd_flag & ND_NFSV2) != 0) {
665 if (nd->nd_procnum < NFS_V3NPROCS)
666 procnum = nfsv2_procid[nd->nd_procnum];
668 procnum = NFSV2PROC_NOOP;
672 * Now only used for the R_DONTRECOVER case, but until that is
673 * supported within the krpc code, I need to keep a queue of
674 * outstanding RPCs for nfsv4 client requests.
676 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
677 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
678 M_NFSDREQ, M_WAITOK);
680 if (dtrace_nfscl_nfs234_start_probe != NULL) {
684 if (nd->nd_flag & ND_NFSV4) {
686 nfscl_nfs4_start_probes[nd->nd_procnum];
687 probe_procnum = nd->nd_procnum;
688 } else if (nd->nd_flag & ND_NFSV3) {
689 probe_id = nfscl_nfs3_start_probes[procnum];
690 probe_procnum = procnum;
693 nfscl_nfs2_start_probes[nd->nd_procnum];
694 probe_procnum = procnum;
697 (dtrace_nfscl_nfs234_start_probe)
698 (probe_id, vp, nd->nd_mreq, cred,
704 freeslot = -1; /* Set to slot that needs to be free'd */
706 slot = -1; /* Slot that needs a sequence# increment. */
708 * This timeout specifies when a new socket should be created,
709 * along with new xid values. For UDP, this should be done
710 * infrequently, since retransmits of RPC requests should normally
716 timo.tv_sec = NFSV4_UPCALLTIMEO;
718 timo.tv_sec = NFSV4_CALLBACKTIMEO;
720 if (nrp->nr_sotype != SOCK_DGRAM) {
722 if ((nmp->nm_flag & NFSMNT_NFSV4))
723 timo.tv_sec = INT_MAX;
725 timo.tv_sec = NFS_TCPTIMEO;
727 if (NFSHASSOFT(nmp)) {
729 * CLSET_RETRIES is set to 2, so this should be
730 * half of the total timeout required.
732 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
735 timo.tv_sec = timeo / NFS_HZ;
736 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
739 /* For UDP hard mounts, use a large value. */
740 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
749 * Chain request into list of outstanding requests.
752 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
758 if (clp != NULL && sep != NULL)
759 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
760 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
762 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
763 nd->nd_mreq, &nd->nd_mrep, timo);
767 * RPC done, unlink the request.
770 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
775 * If there was a successful reply and a tprintf msg.
776 * tprintf a response.
778 if (stat == RPC_SUCCESS) {
780 } else if (stat == RPC_TIMEDOUT) {
781 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
783 } else if (stat == RPC_VERSMISMATCH) {
784 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
786 } else if (stat == RPC_PROGVERSMISMATCH) {
787 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
788 error = EPROTONOSUPPORT;
789 } else if (stat == RPC_INTR) {
792 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
796 m_freem(nd->nd_mreq);
800 FREE((caddr_t)rep, M_NFSDREQ);
802 newnfs_restore_sigmask(td, &oldset);
806 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
809 * Search for any mbufs that are not a multiple of 4 bytes long
810 * or with m_data not longword aligned.
811 * These could cause pointer alignment problems, so copy them to
812 * well aligned mbufs.
814 newnfs_realign(&nd->nd_mrep, M_WAITOK);
815 nd->nd_md = nd->nd_mrep;
816 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
818 if (nd->nd_procnum != NFSPROC_NULL &&
819 nd->nd_procnum != NFSV4PROC_CBNULL) {
820 /* If sep == NULL, set it to the default in nmp. */
821 if (sep == NULL && nmp != NULL)
822 sep = nfsmnt_mdssession(nmp);
824 * and now the actual NFS xdr.
826 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
827 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
828 if (nd->nd_repstat >= 10000)
829 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
830 (int)nd->nd_repstat);
833 * Get rid of the tag, return count and SEQUENCE result for
836 if ((nd->nd_flag & ND_NFSV4) != 0) {
837 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
838 i = fxdr_unsigned(int, *tl);
839 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
842 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
843 opcnt = fxdr_unsigned(int, *tl++);
844 i = fxdr_unsigned(int, *tl++);
845 j = fxdr_unsigned(int, *tl);
847 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
849 * If the first op is Sequence, free up the slot.
851 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
852 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
853 NFSCL_DEBUG(1, "failed seq=%d\n", j);
854 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
855 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j == 0)
857 if (i == NFSV4OP_SEQUENCE)
858 NFSM_DISSECT(tl, uint32_t *,
862 NFSM_DISSECT(tl, uint32_t *,
865 mtx_lock(&sep->nfsess_mtx);
866 if (bcmp(tl, sep->nfsess_sessionid,
867 NFSX_V4SESSIONID) == 0) {
868 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
869 retseq = fxdr_unsigned(uint32_t, *tl++);
870 slot = fxdr_unsigned(int, *tl++);
872 if (retseq != sep->nfsess_slotseq[slot])
873 printf("retseq diff 0x%x\n",
875 retval = fxdr_unsigned(uint32_t, *++tl);
876 if ((retval + 1) < sep->nfsess_foreslots
878 sep->nfsess_foreslots = (retval
880 else if ((retval + 1) >
881 sep->nfsess_foreslots)
882 sep->nfsess_foreslots = (retval
883 < 64) ? (retval + 1) : 64;
885 mtx_unlock(&sep->nfsess_mtx);
887 /* Grab the op and status for the next one. */
889 NFSM_DISSECT(tl, uint32_t *,
891 i = fxdr_unsigned(int, *tl++);
892 j = fxdr_unsigned(int, *tl);
896 if (nd->nd_repstat != 0) {
897 if (nd->nd_repstat == NFSERR_BADSESSION &&
898 nmp != NULL && dssep == NULL) {
900 * If this is a client side MDS RPC, mark
901 * the MDS session defunct and initiate
902 * recovery, as required.
903 * The nfsess_defunct field is protected by
904 * the NFSLOCKMNT()/nm_mtx lock and not the
905 * nfsess_mtx lock to simplify its handling,
906 * for the MDS session. This lock is also
907 * sufficient for nfsess_sessionid, since it
908 * never changes in the structure.
910 NFSCL_DEBUG(1, "Got badsession\n");
913 sep = NFSMNT_MDSSESSION(nmp);
914 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
915 NFSX_V4SESSIONID) == 0) {
916 /* Initiate recovery. */
917 sep->nfsess_defunct = 1;
918 NFSCL_DEBUG(1, "Marked defunct\n");
919 if (nmp->nm_clp != NULL) {
920 nmp->nm_clp->nfsc_flags |=
927 * Sleep for up to 1sec waiting for a new
930 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
933 * Get the session again, in case a new one
934 * has been created during the sleep.
936 sep = NFSMNT_MDSSESSION(nmp);
938 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
939 reterr = nfsv4_sequencelookup(nmp, sep,
940 &slotpos, &maxslot, &slotseq,
943 /* Fill in new session info. */
945 "Filling in new sequence\n");
946 tl = nd->nd_sequence;
949 tl += NFSX_V4SESSIONID /
951 *tl++ = txdr_unsigned(slotseq);
952 *tl++ = txdr_unsigned(slotpos);
953 *tl = txdr_unsigned(maxslot);
955 if (reterr == NFSERR_BADSESSION ||
958 "Badsession looping\n");
959 m_freem(nd->nd_mrep);
963 nd->nd_repstat = reterr;
964 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
967 if (((nd->nd_repstat == NFSERR_DELAY ||
968 nd->nd_repstat == NFSERR_GRACE) &&
969 (nd->nd_flag & ND_NFSV4) &&
970 nd->nd_procnum != NFSPROC_DELEGRETURN &&
971 nd->nd_procnum != NFSPROC_SETATTR &&
972 nd->nd_procnum != NFSPROC_READ &&
973 nd->nd_procnum != NFSPROC_READDS &&
974 nd->nd_procnum != NFSPROC_WRITE &&
975 nd->nd_procnum != NFSPROC_WRITEDS &&
976 nd->nd_procnum != NFSPROC_OPEN &&
977 nd->nd_procnum != NFSPROC_CREATE &&
978 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
979 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
980 nd->nd_procnum != NFSPROC_CLOSE &&
981 nd->nd_procnum != NFSPROC_LOCK &&
982 nd->nd_procnum != NFSPROC_LOCKU) ||
983 (nd->nd_repstat == NFSERR_DELAY &&
984 (nd->nd_flag & ND_NFSV4) == 0) ||
985 nd->nd_repstat == NFSERR_RESOURCE) {
986 if (trylater_delay > NFS_TRYLATERDEL)
987 trylater_delay = NFS_TRYLATERDEL;
988 waituntil = NFSD_MONOSEC + trylater_delay;
989 while (NFSD_MONOSEC < waituntil)
990 (void) nfs_catnap(PZERO, 0, "nfstry");
993 mtx_lock(&sep->nfsess_mtx);
994 sep->nfsess_slotseq[slot]++;
995 *nd->nd_slotseq = txdr_unsigned(
996 sep->nfsess_slotseq[slot]);
997 mtx_unlock(&sep->nfsess_mtx);
999 m_freem(nd->nd_mrep);
1005 * If the File Handle was stale, invalidate the
1006 * lookup cache, just in case.
1007 * (vp != NULL implies a client side call)
1009 if (nd->nd_repstat == ESTALE && vp != NULL) {
1011 if (ncl_call_invalcaches != NULL)
1012 (*ncl_call_invalcaches)(vp);
1015 if ((nd->nd_flag & ND_NFSV4) != 0) {
1016 /* Free the slot, as required. */
1018 nfsv4_freeslot(sep, freeslot);
1020 * If this op is Putfh, throw its results away.
1023 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1024 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1025 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1026 i = fxdr_unsigned(int, *tl++);
1027 j = fxdr_unsigned(int, *tl);
1029 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1032 * All Compounds that do an Op that must
1033 * be in sequence consist of NFSV4OP_PUTFH
1034 * followed by one of these. As such, we
1035 * can determine if the seqid# should be
1036 * incremented, here.
1038 if ((i == NFSV4OP_OPEN ||
1039 i == NFSV4OP_OPENCONFIRM ||
1040 i == NFSV4OP_OPENDOWNGRADE ||
1041 i == NFSV4OP_CLOSE ||
1042 i == NFSV4OP_LOCK ||
1043 i == NFSV4OP_LOCKU) &&
1045 (j != NFSERR_STALECLIENTID &&
1046 j != NFSERR_STALESTATEID &&
1047 j != NFSERR_BADSTATEID &&
1048 j != NFSERR_BADSEQID &&
1049 j != NFSERR_BADXDR &&
1050 j != NFSERR_RESOURCE &&
1051 j != NFSERR_NOFILEHANDLE)))
1052 nd->nd_flag |= ND_INCRSEQID;
1055 * If this op's status is non-zero, mark
1056 * that there is no more data to process.
1057 * The exception is Setattr, which always has xdr
1058 * when it has failed.
1060 if (j != 0 && i != NFSV4OP_SETATTR)
1061 nd->nd_flag |= ND_NOMOREDATA;
1064 * If R_DONTRECOVER is set, replace the stale error
1065 * reply, so that recovery isn't initiated.
1067 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1068 nd->nd_repstat == NFSERR_BADSESSION ||
1069 nd->nd_repstat == NFSERR_STALESTATEID) &&
1070 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1071 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1075 #ifdef KDTRACE_HOOKS
1076 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1080 if (nd->nd_flag & ND_NFSV4) {
1081 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1082 probe_procnum = nd->nd_procnum;
1083 } else if (nd->nd_flag & ND_NFSV3) {
1084 probe_id = nfscl_nfs3_done_probes[procnum];
1085 probe_procnum = procnum;
1087 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1088 probe_procnum = procnum;
1091 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1092 nd->nd_mreq, cred, probe_procnum, 0);
1096 m_freem(nd->nd_mreq);
1097 if (usegssname == 0)
1100 FREE((caddr_t)rep, M_NFSDREQ);
1102 newnfs_restore_sigmask(td, &oldset);
1105 mbuf_freem(nd->nd_mrep);
1106 mbuf_freem(nd->nd_mreq);
1107 if (usegssname == 0)
1110 FREE((caddr_t)rep, M_NFSDREQ);
1112 newnfs_restore_sigmask(td, &oldset);
1117 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1118 * wait for all requests to complete. This is used by forced unmounts
1119 * to terminate any outstanding RPCs.
1122 newnfs_nmcancelreqs(struct nfsmount *nmp)
1124 struct nfsclds *dsp;
1125 struct __rpc_client *cl;
1127 if (nmp->nm_sockreq.nr_client != NULL)
1128 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1131 TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1133 if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1134 (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1135 dsp->nfsclds_sockp != NULL &&
1136 dsp->nfsclds_sockp->nr_client != NULL) {
1137 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1138 cl = dsp->nfsclds_sockp->nr_client;
1151 * Any signal that can interrupt an NFS operation in an intr mount
1152 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1154 int newnfs_sig_set[] = {
1163 * Check to see if one of the signals in our subset is pending on
1164 * the process (in an intr mount).
1167 nfs_sig_pending(sigset_t set)
1171 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1172 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1178 * The set/restore sigmask functions are used to (temporarily) overwrite
1179 * the thread td_sigmask during an RPC call (for example). These are also
1180 * used in other places in the NFS client that might tsleep().
1183 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1191 td = curthread; /* XXX */
1193 /* Remove the NFS set of signals from newset */
1195 mtx_lock(&p->p_sigacts->ps_mtx);
1196 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1198 * But make sure we leave the ones already masked
1199 * by the process, ie. remove the signal from the
1200 * temporary signalmask only if it wasn't already
1203 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1204 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1205 SIGDELSET(newset, newnfs_sig_set[i]);
1207 mtx_unlock(&p->p_sigacts->ps_mtx);
1208 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1209 SIGPROCMASK_PROC_LOCKED);
1214 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1217 td = curthread; /* XXX */
1218 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1222 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1223 * old one after msleep() returns.
1226 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1232 if ((priority & PCATCH) == 0)
1233 return msleep(ident, mtx, priority, wmesg, timo);
1235 td = curthread; /* XXX */
1236 newnfs_set_sigmask(td, &oldset);
1237 error = msleep(ident, mtx, priority, wmesg, timo);
1238 newnfs_restore_sigmask(td, &oldset);
1244 * Test for a termination condition pending on the process.
1245 * This is used for NFSMNT_INT mounts.
1248 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1253 /* Terminate all requests while attempting a forced unmount. */
1254 if (NFSCL_FORCEDISM(nmp->nm_mountp))
1256 if (!(nmp->nm_flag & NFSMNT_INT))
1262 tmpset = p->p_siglist;
1263 SIGSETOR(tmpset, td->td_siglist);
1264 SIGSETNAND(tmpset, td->td_sigmask);
1265 mtx_lock(&p->p_sigacts->ps_mtx);
1266 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1267 mtx_unlock(&p->p_sigacts->ps_mtx);
1268 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1269 && nfs_sig_pending(tmpset)) {
1278 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1282 p = td ? td->td_proc : NULL;
1284 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1285 server, msg, error);
1287 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1293 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1294 int error, int flags)
1298 mtx_lock(&nmp->nm_mtx);
1299 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1300 nmp->nm_state |= NFSSTA_TIMEO;
1301 mtx_unlock(&nmp->nm_mtx);
1302 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1305 mtx_unlock(&nmp->nm_mtx);
1306 mtx_lock(&nmp->nm_mtx);
1307 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1308 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1309 mtx_unlock(&nmp->nm_mtx);
1310 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1313 mtx_unlock(&nmp->nm_mtx);
1314 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1318 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1319 int flags, int tprintfmsg)
1324 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1327 mtx_lock(&nmp->nm_mtx);
1328 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1329 nmp->nm_state &= ~NFSSTA_TIMEO;
1330 mtx_unlock(&nmp->nm_mtx);
1331 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1334 mtx_unlock(&nmp->nm_mtx);
1336 mtx_lock(&nmp->nm_mtx);
1337 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1338 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1339 mtx_unlock(&nmp->nm_mtx);
1340 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1343 mtx_unlock(&nmp->nm_mtx);