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_kdtrace.h"
42 #include "opt_kgssapi.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/limits.h>
50 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
55 #include <sys/signalvar.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysctl.h>
58 #include <sys/syslog.h>
59 #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];
92 extern struct nfsstats newnfsstats;
93 extern struct nfsreqhead nfsd_reqq;
94 extern int nfscl_ticks;
95 extern void (*ncl_call_invalcaches)(struct vnode *);
96 extern int nfs_numnfscbd;
97 extern int nfscl_debuglevel;
100 static int nfsrv_gsscallbackson = 0;
101 static int nfs_bufpackets = 4;
102 static int nfs_reconnects;
103 static int nfs3_jukebox_delay = 10;
104 static int nfs_skip_wcc_data_onerr = 1;
105 static int nfs_keytab_enctype = ETYPE_DES_CBC_CRC;
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");
117 SYSCTL_INT(_vfs_nfs, OID_AUTO, keytab_enctype, CTLFLAG_RW, &nfs_keytab_enctype, 0,
118 "Encryption type for the keytab entry used by nfs");
120 static void nfs_down(struct nfsmount *, struct thread *, const char *,
122 static void nfs_up(struct nfsmount *, struct thread *, const char *,
124 static int nfs_msg(struct thread *, const char *, const char *, int);
126 struct nfs_cached_auth {
127 int ca_refs; /* refcount, including 1 from the cache */
128 uid_t ca_uid; /* uid that corresponds to this auth */
129 AUTH *ca_auth; /* RPC auth handle */
132 static int nfsv2_procid[NFS_V3NPROCS] = {
158 * Initialize sockets and congestion for a new NFS connection.
159 * We do not free the sockaddr if error.
162 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
163 struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
165 int rcvreserve, sndreserve;
167 struct sockaddr *saddr;
168 struct ucred *origcred;
170 struct netconfig *nconf;
172 int one = 1, retries, error = 0;
173 struct thread *td = curthread;
178 * We need to establish the socket using the credentials of
179 * the mountpoint. Some parts of this process (such as
180 * sobind() and soconnect()) will use the curent thread's
181 * credential instead of the socket credential. To work
182 * around this, temporarily change the current thread's
183 * credential to that of the mountpoint.
185 * XXX: It would be better to explicitly pass the correct
186 * credential to sobind() and soconnect().
188 origcred = td->td_ucred;
191 * Use the credential in nr_cred, if not NULL.
193 if (nrp->nr_cred != NULL)
194 td->td_ucred = nrp->nr_cred;
199 if (saddr->sa_family == AF_INET)
200 if (nrp->nr_sotype == SOCK_DGRAM)
201 nconf = getnetconfigent("udp");
203 nconf = getnetconfigent("tcp");
205 if (nrp->nr_sotype == SOCK_DGRAM)
206 nconf = getnetconfigent("udp6");
208 nconf = getnetconfigent("tcp6");
210 pktscale = nfs_bufpackets;
216 * soreserve() can fail if sb_max is too small, so shrink pktscale
217 * and try again if there is an error.
218 * Print a log message suggesting increasing sb_max.
219 * Creating a socket and doing this is necessary since, if the
220 * reservation sizes are too large and will make soreserve() fail,
221 * the connection will work until a large send is attempted and
222 * then it will loop in the krpc code.
225 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
226 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
227 nrp->nr_soproto, td->td_ucred, td);
229 td->td_ucred = origcred;
233 if (error != 0 && pktscale > 2)
235 if (nrp->nr_sotype == SOCK_DGRAM) {
237 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
239 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
242 sndreserve = rcvreserve = 1024 * pktscale;
245 if (nrp->nr_sotype != SOCK_STREAM)
246 panic("nfscon sotype");
248 sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
249 sizeof (u_int32_t)) * pktscale;
250 rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
251 sizeof (u_int32_t)) * pktscale;
253 sndreserve = rcvreserve = 1024 * pktscale;
256 error = soreserve(so, sndreserve, rcvreserve);
257 } while (error != 0 && pktscale > 2);
260 td->td_ucred = origcred;
264 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
265 nrp->nr_vers, sndreserve, rcvreserve);
266 CLNT_CONTROL(client, CLSET_WAITCHAN, "newnfsreq");
268 if ((nmp->nm_flag & NFSMNT_INT))
269 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
270 if ((nmp->nm_flag & NFSMNT_RESVPORT))
271 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
272 if (NFSHASSOFT(nmp)) {
273 if (nmp->nm_sotype == SOCK_DGRAM)
275 * For UDP, the large timeout for a reconnect
276 * will be set to "nm_retry * nm_timeo / 2", so
277 * we only want to do 2 reconnect timeout
282 retries = nmp->nm_retry;
285 if (NFSHASNFSV4N(nmp)) {
287 * Make sure the nfscbd_pool doesn't get destroyed
291 if (nfs_numnfscbd > 0) {
294 xprt = svc_vc_create_backchannel(nfscbd_pool);
295 CLNT_CONTROL(client, CLSET_BACKCHANNEL, xprt);
298 if (nfs_numnfscbd == 0)
299 wakeup(&nfs_numnfscbd);
306 * - Null RPC callback to client
307 * - Non-Null RPC callback to client, wait a little longer
308 * - upcalls to nfsuserd and gssd (clp == NULL)
310 if (callback_retry_mult == 0) {
311 retries = NFSV4_UPCALLRETRY;
312 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
314 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
317 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
321 * For UDP, there are 2 timeouts:
322 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
323 * that does a retransmit of an RPC request using the same
324 * socket and xid. This is what you normally want to do,
325 * since NFS servers depend on "same xid" for their
326 * Duplicate Request Cache.
327 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
328 * retransmits on the same socket should fail and a fresh
329 * socket created. Each of these timeouts counts as one
330 * CLSET_RETRIES as set above.
331 * Set the initial retransmit timeout for UDP. This timeout
332 * doesn't exist for TCP and the following call just fails,
335 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
336 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
337 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
340 mtx_lock(&nrp->nr_mtx);
341 if (nrp->nr_client != NULL) {
343 * Someone else already connected.
345 CLNT_RELEASE(client);
347 nrp->nr_client = client;
351 * Protocols that do not require connections may be optionally left
352 * unconnected for servers that reply from a port other than NFS_PORT.
354 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
355 mtx_unlock(&nrp->nr_mtx);
356 CLNT_CONTROL(client, CLSET_CONNECT, &one);
358 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 rpc_gss_options_req_t req_options;
400 switch (secflavour) {
401 case RPCSEC_GSS_KRB5:
402 case RPCSEC_GSS_KRB5I:
403 case RPCSEC_GSS_KRB5P:
405 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
408 if (secflavour == RPCSEC_GSS_KRB5)
409 svc = rpc_gss_svc_none;
410 else if (secflavour == RPCSEC_GSS_KRB5I)
411 svc = rpc_gss_svc_integrity;
413 svc = rpc_gss_svc_privacy;
415 req_options.req_flags = GSS_C_MUTUAL_FLAG;
416 req_options.time_req = 0;
417 req_options.my_cred = GSS_C_NO_CREDENTIAL;
418 req_options.input_channel_bindings = NULL;
419 req_options.enc_type = nfs_keytab_enctype;
421 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
422 clnt_principal, srv_principal, mech_oid, svc,
426 * Until changes to the rpcsec_gss code are committed,
427 * there is no support for host based initiator
428 * principals. As such, that case cannot yet be handled.
430 if (clnt_principal == NULL)
431 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
432 srv_principal, mech_oid, svc);
441 return (authunix_create(cred));
447 * Callback from the RPC code to generate up/down notifications.
450 struct nfs_feedback_arg {
451 struct nfsmount *nf_mount;
452 int nf_lastmsg; /* last tprintf */
454 struct thread *nf_td;
458 nfs_feedback(int type, int proc, void *arg)
460 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
461 struct nfsmount *nmp = nf->nf_mount;
465 case FEEDBACK_REXMIT2:
466 case FEEDBACK_RECONNECT:
468 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
469 nfs_down(nmp, nf->nf_td,
470 "not responding", 0, NFSSTA_TIMEO);
471 nf->nf_tprintfmsg = TRUE;
472 nf->nf_lastmsg = now;
477 nfs_up(nf->nf_mount, nf->nf_td,
478 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
484 * newnfs_request - goes something like this
485 * - does the rpc by calling the krpc layer
486 * - break down rpc header and return with nfs reply
487 * nb: always frees up nd_mreq mbuf list
490 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
491 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
492 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
493 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *sep)
495 u_int32_t retseq, retval, *tl;
497 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
498 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
501 u_int trylater_delay = 1;
502 struct nfs_feedback_arg nf;
505 struct rpc_callextra ext;
507 struct nfsreq *rep = NULL;
508 char *srv_principal = NULL;
510 struct ucred *authcred;
514 /* Reject requests while attempting a forced unmount. */
515 if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
516 m_freem(nd->nd_mreq);
521 * Set authcred, which is used to acquire RPC credentials to
522 * the cred argument, by default. The crhold() should not be
523 * necessary, but will ensure that some future code change
524 * doesn't result in the credential being free'd prematurely.
526 authcred = crhold(cred);
528 /* For client side interruptible mounts, mask off the signals. */
529 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
530 newnfs_set_sigmask(td, &oldset);
535 * XXX if not already connected call nfs_connect now. Longer
536 * term, change nfs_mount to call nfs_connect unconditionally
537 * and let clnt_reconnect_create handle reconnects.
539 if (nrp->nr_client == NULL)
540 newnfs_connect(nmp, nrp, cred, td, 0);
543 * For a client side mount, nmp is != NULL and clp == NULL. For
544 * server calls (callbacks or upcalls), nmp == NULL.
548 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
549 secflavour = RPCSEC_GSS_KRB5;
550 if (nd->nd_procnum != NFSPROC_NULL) {
551 if (clp->lc_flags & LCL_GSSINTEGRITY)
552 secflavour = RPCSEC_GSS_KRB5I;
553 else if (clp->lc_flags & LCL_GSSPRIVACY)
554 secflavour = RPCSEC_GSS_KRB5P;
558 } else if (nmp != NULL && NFSHASKERB(nmp) &&
559 nd->nd_procnum != NFSPROC_NULL) {
560 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
561 nd->nd_flag |= ND_USEGSSNAME;
562 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
564 * If there is a client side host based credential,
565 * use that, otherwise use the system uid, if set.
566 * The system uid is in the nmp->nm_sockreq.nr_cred
569 if (nmp->nm_krbnamelen > 0) {
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();
626 auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
627 srv_principal, NULL, authcred);
629 auth = nfs_getauth(nrp, secflavour, NULL,
630 srv_principal, NULL, authcred);
633 m_freem(nd->nd_mreq);
635 newnfs_restore_sigmask(td, &oldset);
638 bzero(&ext, sizeof(ext));
641 ext.rc_feedback = nfs_feedback;
642 ext.rc_feedback_arg = &nf;
645 procnum = nd->nd_procnum;
646 if ((nd->nd_flag & ND_NFSV4) &&
647 nd->nd_procnum != NFSPROC_NULL &&
648 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
649 procnum = NFSV4PROC_COMPOUND;
652 NFSINCRGLOBAL(newnfsstats.rpcrequests);
654 /* Map the procnum to the old NFSv2 one, as required. */
655 if ((nd->nd_flag & ND_NFSV2) != 0) {
656 if (nd->nd_procnum < NFS_V3NPROCS)
657 procnum = nfsv2_procid[nd->nd_procnum];
659 procnum = NFSV2PROC_NOOP;
663 * Now only used for the R_DONTRECOVER case, but until that is
664 * supported within the krpc code, I need to keep a queue of
665 * outstanding RPCs for nfsv4 client requests.
667 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
668 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
669 M_NFSDREQ, M_WAITOK);
671 if (dtrace_nfscl_nfs234_start_probe != NULL) {
675 if (nd->nd_flag & ND_NFSV4) {
677 nfscl_nfs4_start_probes[nd->nd_procnum];
678 probe_procnum = nd->nd_procnum;
679 } else if (nd->nd_flag & ND_NFSV3) {
680 probe_id = nfscl_nfs3_start_probes[procnum];
681 probe_procnum = procnum;
684 nfscl_nfs2_start_probes[nd->nd_procnum];
685 probe_procnum = procnum;
688 (dtrace_nfscl_nfs234_start_probe)
689 (probe_id, vp, nd->nd_mreq, cred,
695 freeslot = -1; /* Set to slot that needs to be free'd */
697 slot = -1; /* Slot that needs a sequence# increment. */
699 * This timeout specifies when a new socket should be created,
700 * along with new xid values. For UDP, this should be done
701 * infrequently, since retransmits of RPC requests should normally
707 timo.tv_sec = NFSV4_UPCALLTIMEO;
709 timo.tv_sec = NFSV4_CALLBACKTIMEO;
711 if (nrp->nr_sotype != SOCK_DGRAM) {
713 if ((nmp->nm_flag & NFSMNT_NFSV4))
714 timo.tv_sec = INT_MAX;
716 timo.tv_sec = NFS_TCPTIMEO;
718 if (NFSHASSOFT(nmp)) {
720 * CLSET_RETRIES is set to 2, so this should be
721 * half of the total timeout required.
723 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
726 timo.tv_sec = timeo / NFS_HZ;
727 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
730 /* For UDP hard mounts, use a large value. */
731 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
740 * Chain request into list of outstanding requests.
743 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
749 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
754 * RPC done, unlink the request.
757 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
762 * If there was a successful reply and a tprintf msg.
763 * tprintf a response.
765 if (stat == RPC_SUCCESS) {
767 } else if (stat == RPC_TIMEDOUT) {
768 NFSINCRGLOBAL(newnfsstats.rpctimeouts);
770 } else if (stat == RPC_VERSMISMATCH) {
771 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
773 } else if (stat == RPC_PROGVERSMISMATCH) {
774 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
775 error = EPROTONOSUPPORT;
776 } else if (stat == RPC_INTR) {
779 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
783 m_freem(nd->nd_mreq);
786 FREE((caddr_t)rep, M_NFSDREQ);
788 newnfs_restore_sigmask(td, &oldset);
792 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
795 * Search for any mbufs that are not a multiple of 4 bytes long
796 * or with m_data not longword aligned.
797 * These could cause pointer alignment problems, so copy them to
798 * well aligned mbufs.
800 newnfs_realign(&nd->nd_mrep);
801 nd->nd_md = nd->nd_mrep;
802 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
804 if (nd->nd_procnum != NFSPROC_NULL) {
805 /* If sep == NULL, set it to the default in nmp. */
806 if (sep == NULL && nmp != NULL)
807 sep = NFSMNT_MDSSESSION(nmp);
809 * and now the actual NFS xdr.
811 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
812 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
813 if (nd->nd_repstat >= 10000)
814 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
815 (int)nd->nd_repstat);
818 * Get rid of the tag, return count and SEQUENCE result for
821 if ((nd->nd_flag & ND_NFSV4) != 0) {
822 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
823 i = fxdr_unsigned(int, *tl);
824 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
827 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
828 opcnt = fxdr_unsigned(int, *tl++);
829 i = fxdr_unsigned(int, *tl++);
830 j = fxdr_unsigned(int, *tl);
832 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
834 * If the first op is Sequence, free up the slot.
836 if (nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0)
837 NFSCL_DEBUG(1, "failed seq=%d\n", j);
838 if (nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) {
839 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
841 mtx_lock(&sep->nfsess_mtx);
842 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
843 retseq = fxdr_unsigned(uint32_t, *tl++);
844 slot = fxdr_unsigned(int, *tl++);
846 if (retseq != sep->nfsess_slotseq[slot])
847 printf("retseq diff 0x%x\n", retseq);
848 retval = fxdr_unsigned(uint32_t, *++tl);
849 if ((retval + 1) < sep->nfsess_foreslots)
850 sep->nfsess_foreslots = (retval + 1);
851 else if ((retval + 1) > sep->nfsess_foreslots)
852 sep->nfsess_foreslots = (retval < 64) ?
854 mtx_unlock(&sep->nfsess_mtx);
856 /* Grab the op and status for the next one. */
858 NFSM_DISSECT(tl, uint32_t *,
860 i = fxdr_unsigned(int, *tl++);
861 j = fxdr_unsigned(int, *tl);
865 if (nd->nd_repstat != 0) {
866 if (((nd->nd_repstat == NFSERR_DELAY ||
867 nd->nd_repstat == NFSERR_GRACE) &&
868 (nd->nd_flag & ND_NFSV4) &&
869 nd->nd_procnum != NFSPROC_DELEGRETURN &&
870 nd->nd_procnum != NFSPROC_SETATTR &&
871 nd->nd_procnum != NFSPROC_READ &&
872 nd->nd_procnum != NFSPROC_READDS &&
873 nd->nd_procnum != NFSPROC_WRITE &&
874 nd->nd_procnum != NFSPROC_WRITEDS &&
875 nd->nd_procnum != NFSPROC_OPEN &&
876 nd->nd_procnum != NFSPROC_CREATE &&
877 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
878 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
879 nd->nd_procnum != NFSPROC_CLOSE &&
880 nd->nd_procnum != NFSPROC_LOCK &&
881 nd->nd_procnum != NFSPROC_LOCKU) ||
882 (nd->nd_repstat == NFSERR_DELAY &&
883 (nd->nd_flag & ND_NFSV4) == 0) ||
884 nd->nd_repstat == NFSERR_RESOURCE) {
885 if (trylater_delay > NFS_TRYLATERDEL)
886 trylater_delay = NFS_TRYLATERDEL;
887 waituntil = NFSD_MONOSEC + trylater_delay;
888 while (NFSD_MONOSEC < waituntil)
889 (void) nfs_catnap(PZERO, 0, "nfstry");
892 mtx_lock(&sep->nfsess_mtx);
893 sep->nfsess_slotseq[slot]++;
894 *nd->nd_slotseq = txdr_unsigned(
895 sep->nfsess_slotseq[slot]);
896 mtx_unlock(&sep->nfsess_mtx);
898 m_freem(nd->nd_mrep);
904 * If the File Handle was stale, invalidate the
905 * lookup cache, just in case.
906 * (vp != NULL implies a client side call)
908 if (nd->nd_repstat == ESTALE && vp != NULL) {
910 if (ncl_call_invalcaches != NULL)
911 (*ncl_call_invalcaches)(vp);
914 if ((nd->nd_flag & ND_NFSV4) != 0) {
915 /* Free the slot, as required. */
917 nfsv4_freeslot(sep, freeslot);
919 * If this op is Putfh, throw its results away.
922 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
923 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
924 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
925 i = fxdr_unsigned(int, *tl++);
926 j = fxdr_unsigned(int, *tl);
928 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
931 * All Compounds that do an Op that must
932 * be in sequence consist of NFSV4OP_PUTFH
933 * followed by one of these. As such, we
934 * can determine if the seqid# should be
937 if ((i == NFSV4OP_OPEN ||
938 i == NFSV4OP_OPENCONFIRM ||
939 i == NFSV4OP_OPENDOWNGRADE ||
940 i == NFSV4OP_CLOSE ||
942 i == NFSV4OP_LOCKU) &&
944 (j != NFSERR_STALECLIENTID &&
945 j != NFSERR_STALESTATEID &&
946 j != NFSERR_BADSTATEID &&
947 j != NFSERR_BADSEQID &&
948 j != NFSERR_BADXDR &&
949 j != NFSERR_RESOURCE &&
950 j != NFSERR_NOFILEHANDLE)))
951 nd->nd_flag |= ND_INCRSEQID;
954 * If this op's status is non-zero, mark
955 * that there is no more data to process.
958 nd->nd_flag |= ND_NOMOREDATA;
961 * If R_DONTRECOVER is set, replace the stale error
962 * reply, so that recovery isn't initiated.
964 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
965 nd->nd_repstat == NFSERR_BADSESSION ||
966 nd->nd_repstat == NFSERR_STALESTATEID) &&
967 rep != NULL && (rep->r_flags & R_DONTRECOVER))
968 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
973 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
977 if (nd->nd_flag & ND_NFSV4) {
978 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
979 probe_procnum = nd->nd_procnum;
980 } else if (nd->nd_flag & ND_NFSV3) {
981 probe_id = nfscl_nfs3_done_probes[procnum];
982 probe_procnum = procnum;
984 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
985 probe_procnum = procnum;
988 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
989 nd->nd_mreq, cred, probe_procnum, 0);
993 m_freem(nd->nd_mreq);
996 FREE((caddr_t)rep, M_NFSDREQ);
998 newnfs_restore_sigmask(td, &oldset);
1001 mbuf_freem(nd->nd_mrep);
1002 mbuf_freem(nd->nd_mreq);
1005 FREE((caddr_t)rep, M_NFSDREQ);
1007 newnfs_restore_sigmask(td, &oldset);
1012 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1013 * wait for all requests to complete. This is used by forced unmounts
1014 * to terminate any outstanding RPCs.
1017 newnfs_nmcancelreqs(struct nfsmount *nmp)
1020 if (nmp->nm_sockreq.nr_client != NULL)
1021 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1026 * Any signal that can interrupt an NFS operation in an intr mount
1027 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1029 int newnfs_sig_set[] = {
1039 * Check to see if one of the signals in our subset is pending on
1040 * the process (in an intr mount).
1043 nfs_sig_pending(sigset_t set)
1047 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
1048 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1054 * The set/restore sigmask functions are used to (temporarily) overwrite
1055 * the process p_sigmask during an RPC call (for example). These are also
1056 * used in other places in the NFS client that might tsleep().
1059 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1067 td = curthread; /* XXX */
1069 /* Remove the NFS set of signals from newset */
1071 mtx_lock(&p->p_sigacts->ps_mtx);
1072 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
1074 * But make sure we leave the ones already masked
1075 * by the process, ie. remove the signal from the
1076 * temporary signalmask only if it wasn't already
1079 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1080 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1081 SIGDELSET(newset, newnfs_sig_set[i]);
1083 mtx_unlock(&p->p_sigacts->ps_mtx);
1085 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0);
1089 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1092 td = curthread; /* XXX */
1093 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1097 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1098 * old one after msleep() returns.
1101 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1107 if ((priority & PCATCH) == 0)
1108 return msleep(ident, mtx, priority, wmesg, timo);
1110 td = curthread; /* XXX */
1111 newnfs_set_sigmask(td, &oldset);
1112 error = msleep(ident, mtx, priority, wmesg, timo);
1113 newnfs_restore_sigmask(td, &oldset);
1119 * Test for a termination condition pending on the process.
1120 * This is used for NFSMNT_INT mounts.
1123 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1128 /* Terminate all requests while attempting a forced unmount. */
1129 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1131 if (!(nmp->nm_flag & NFSMNT_INT))
1137 tmpset = p->p_siglist;
1138 SIGSETOR(tmpset, td->td_siglist);
1139 SIGSETNAND(tmpset, td->td_sigmask);
1140 mtx_lock(&p->p_sigacts->ps_mtx);
1141 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1142 mtx_unlock(&p->p_sigacts->ps_mtx);
1143 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1144 && nfs_sig_pending(tmpset)) {
1153 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1157 p = td ? td->td_proc : NULL;
1159 tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
1160 server, msg, error);
1162 tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
1168 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1169 int error, int flags)
1173 mtx_lock(&nmp->nm_mtx);
1174 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1175 nmp->nm_state |= NFSSTA_TIMEO;
1176 mtx_unlock(&nmp->nm_mtx);
1177 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1180 mtx_unlock(&nmp->nm_mtx);
1181 mtx_lock(&nmp->nm_mtx);
1182 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1183 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1184 mtx_unlock(&nmp->nm_mtx);
1185 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1188 mtx_unlock(&nmp->nm_mtx);
1189 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1193 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1194 int flags, int tprintfmsg)
1199 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1202 mtx_lock(&nmp->nm_mtx);
1203 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1204 nmp->nm_state &= ~NFSSTA_TIMEO;
1205 mtx_unlock(&nmp->nm_mtx);
1206 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1209 mtx_unlock(&nmp->nm_mtx);
1211 mtx_lock(&nmp->nm_mtx);
1212 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1213 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1214 mtx_unlock(&nmp->nm_mtx);
1215 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1218 mtx_unlock(&nmp->nm_mtx);