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[NFS_NPROCS + 1];
80 uint32_t nfscl_nfs2_done_probes[NFS_NPROCS + 1];
82 uint32_t nfscl_nfs3_start_probes[NFS_NPROCS + 1];
83 uint32_t nfscl_nfs3_done_probes[NFS_NPROCS + 1];
85 uint32_t nfscl_nfs4_start_probes[NFS_NPROCS + 1];
86 uint32_t nfscl_nfs4_done_probes[NFS_NPROCS + 1];
91 extern struct nfsstats newnfsstats;
92 extern struct nfsreqhead nfsd_reqq;
93 extern int nfscl_ticks;
94 extern void (*ncl_call_invalcaches)(struct vnode *);
96 static int nfsrv_gsscallbackson = 0;
97 static int nfs_bufpackets = 4;
98 static int nfs_reconnects;
99 static int nfs3_jukebox_delay = 10;
100 static int nfs_skip_wcc_data_onerr = 1;
101 static int nfs_keytab_enctype = ETYPE_DES_CBC_CRC;
103 SYSCTL_DECL(_vfs_nfs);
105 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
106 "Buffer reservation size 2 < x < 64");
107 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
108 "Number of times the nfs client has had to reconnect");
109 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
110 "Number of seconds to delay a retry after receiving EJUKEBOX");
111 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
112 "Disable weak cache consistency checking when server returns an error");
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, keytab_enctype, CTLFLAG_RW, &nfs_keytab_enctype, 0,
114 "Encryption type for the keytab entry used by nfs");
116 static void nfs_down(struct nfsmount *, struct thread *, const char *,
118 static void nfs_up(struct nfsmount *, struct thread *, const char *,
120 static int nfs_msg(struct thread *, const char *, const char *, int);
122 struct nfs_cached_auth {
123 int ca_refs; /* refcount, including 1 from the cache */
124 uid_t ca_uid; /* uid that corresponds to this auth */
125 AUTH *ca_auth; /* RPC auth handle */
128 static int nfsv2_procid[NFS_V3NPROCS] = {
154 * Initialize sockets and congestion for a new NFS connection.
155 * We do not free the sockaddr if error.
158 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
159 struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
161 int rcvreserve, sndreserve;
163 struct sockaddr *saddr;
164 struct ucred *origcred;
166 struct netconfig *nconf;
168 int one = 1, retries, error = 0;
169 struct thread *td = curthread;
173 * We need to establish the socket using the credentials of
174 * the mountpoint. Some parts of this process (such as
175 * sobind() and soconnect()) will use the curent thread's
176 * credential instead of the socket credential. To work
177 * around this, temporarily change the current thread's
178 * credential to that of the mountpoint.
180 * XXX: It would be better to explicitly pass the correct
181 * credential to sobind() and soconnect().
183 origcred = td->td_ucred;
186 * Use the credential in nr_cred, if not NULL.
188 if (nrp->nr_cred != NULL)
189 td->td_ucred = nrp->nr_cred;
194 if (saddr->sa_family == AF_INET)
195 if (nrp->nr_sotype == SOCK_DGRAM)
196 nconf = getnetconfigent("udp");
198 nconf = getnetconfigent("tcp");
200 if (nrp->nr_sotype == SOCK_DGRAM)
201 nconf = getnetconfigent("udp6");
203 nconf = getnetconfigent("tcp6");
205 pktscale = nfs_bufpackets;
211 * soreserve() can fail if sb_max is too small, so shrink pktscale
212 * and try again if there is an error.
213 * Print a log message suggesting increasing sb_max.
214 * Creating a socket and doing this is necessary since, if the
215 * reservation sizes are too large and will make soreserve() fail,
216 * the connection will work until a large send is attempted and
217 * then it will loop in the krpc code.
220 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
221 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
222 nrp->nr_soproto, td->td_ucred, td);
224 td->td_ucred = origcred;
228 if (error != 0 && pktscale > 2)
230 if (nrp->nr_sotype == SOCK_DGRAM) {
232 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
234 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
237 sndreserve = rcvreserve = 1024 * pktscale;
240 if (nrp->nr_sotype != SOCK_STREAM)
241 panic("nfscon sotype");
243 sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
244 sizeof (u_int32_t)) * pktscale;
245 rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
246 sizeof (u_int32_t)) * pktscale;
248 sndreserve = rcvreserve = 1024 * pktscale;
251 error = soreserve(so, sndreserve, rcvreserve);
252 } while (error != 0 && pktscale > 2);
255 td->td_ucred = origcred;
259 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
260 nrp->nr_vers, sndreserve, rcvreserve);
261 CLNT_CONTROL(client, CLSET_WAITCHAN, "newnfsreq");
263 if ((nmp->nm_flag & NFSMNT_INT))
264 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
265 if ((nmp->nm_flag & NFSMNT_RESVPORT))
266 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
267 if (NFSHASSOFT(nmp)) {
268 if (nmp->nm_sotype == SOCK_DGRAM)
270 * For UDP, the large timeout for a reconnect
271 * will be set to "nm_retry * nm_timeo / 2", so
272 * we only want to do 2 reconnect timeout
277 retries = nmp->nm_retry;
283 * - Null RPC callback to client
284 * - Non-Null RPC callback to client, wait a little longer
285 * - upcalls to nfsuserd and gssd (clp == NULL)
287 if (callback_retry_mult == 0) {
288 retries = NFSV4_UPCALLRETRY;
289 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
291 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
294 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
298 * For UDP, there are 2 timeouts:
299 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
300 * that does a retransmit of an RPC request using the same
301 * socket and xid. This is what you normally want to do,
302 * since NFS servers depend on "same xid" for their
303 * Duplicate Request Cache.
304 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
305 * retransmits on the same socket should fail and a fresh
306 * socket created. Each of these timeouts counts as one
307 * CLSET_RETRIES as set above.
308 * Set the initial retransmit timeout for UDP. This timeout
309 * doesn't exist for TCP and the following call just fails,
312 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
313 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
314 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
317 mtx_lock(&nrp->nr_mtx);
318 if (nrp->nr_client != NULL) {
320 * Someone else already connected.
322 CLNT_RELEASE(client);
324 nrp->nr_client = client;
328 * Protocols that do not require connections may be optionally left
329 * unconnected for servers that reply from a port other than NFS_PORT.
331 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
332 mtx_unlock(&nrp->nr_mtx);
333 CLNT_CONTROL(client, CLSET_CONNECT, &one);
335 mtx_unlock(&nrp->nr_mtx);
338 /* Restore current thread's credentials. */
339 td->td_ucred = origcred;
347 * NFS disconnect. Clean up and unlink.
350 newnfs_disconnect(struct nfssockreq *nrp)
354 mtx_lock(&nrp->nr_mtx);
355 if (nrp->nr_client != NULL) {
356 client = nrp->nr_client;
357 nrp->nr_client = NULL;
358 mtx_unlock(&nrp->nr_mtx);
359 rpc_gss_secpurge_call(client);
361 CLNT_RELEASE(client);
363 mtx_unlock(&nrp->nr_mtx);
368 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
369 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
371 rpc_gss_service_t svc;
374 rpc_gss_options_req_t req_options;
377 switch (secflavour) {
378 case RPCSEC_GSS_KRB5:
379 case RPCSEC_GSS_KRB5I:
380 case RPCSEC_GSS_KRB5P:
382 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
385 if (secflavour == RPCSEC_GSS_KRB5)
386 svc = rpc_gss_svc_none;
387 else if (secflavour == RPCSEC_GSS_KRB5I)
388 svc = rpc_gss_svc_integrity;
390 svc = rpc_gss_svc_privacy;
392 req_options.req_flags = GSS_C_MUTUAL_FLAG;
393 req_options.time_req = 0;
394 req_options.my_cred = GSS_C_NO_CREDENTIAL;
395 req_options.input_channel_bindings = NULL;
396 req_options.enc_type = nfs_keytab_enctype;
398 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
399 clnt_principal, srv_principal, mech_oid, svc,
403 * Until changes to the rpcsec_gss code are committed,
404 * there is no support for host based initiator
405 * principals. As such, that case cannot yet be handled.
407 if (clnt_principal == NULL)
408 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
409 srv_principal, mech_oid, svc);
418 return (authunix_create(cred));
424 * Callback from the RPC code to generate up/down notifications.
427 struct nfs_feedback_arg {
428 struct nfsmount *nf_mount;
429 int nf_lastmsg; /* last tprintf */
431 struct thread *nf_td;
435 nfs_feedback(int type, int proc, void *arg)
437 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
438 struct nfsmount *nmp = nf->nf_mount;
442 case FEEDBACK_REXMIT2:
443 case FEEDBACK_RECONNECT:
445 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
446 nfs_down(nmp, nf->nf_td,
447 "not responding", 0, NFSSTA_TIMEO);
448 nf->nf_tprintfmsg = TRUE;
449 nf->nf_lastmsg = now;
454 nfs_up(nf->nf_mount, nf->nf_td,
455 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
461 * newnfs_request - goes something like this
462 * - does the rpc by calling the krpc layer
463 * - break down rpc header and return with nfs reply
464 * nb: always frees up nd_mreq mbuf list
467 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
468 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
469 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
470 u_char *retsum, int toplevel, u_int64_t *xidp)
474 int i, j, set_sigset = 0, timeo;
475 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
477 u_int trylater_delay = 1;
478 struct nfs_feedback_arg nf;
481 struct rpc_callextra ext;
483 struct nfsreq *rep = NULL;
484 char *srv_principal = NULL;
486 struct ucred *authcred;
490 /* Reject requests while attempting a forced unmount. */
491 if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
492 m_freem(nd->nd_mreq);
497 * Set authcred, which is used to acquire RPC credentials to
498 * the cred argument, by default. The crhold() should not be
499 * necessary, but will ensure that some future code change
500 * doesn't result in the credential being free'd prematurely.
502 authcred = crhold(cred);
504 /* For client side interruptible mounts, mask off the signals. */
505 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
506 newnfs_set_sigmask(td, &oldset);
511 * XXX if not already connected call nfs_connect now. Longer
512 * term, change nfs_mount to call nfs_connect unconditionally
513 * and let clnt_reconnect_create handle reconnects.
515 if (nrp->nr_client == NULL)
516 newnfs_connect(nmp, nrp, cred, td, 0);
519 * For a client side mount, nmp is != NULL and clp == NULL. For
520 * server calls (callbacks or upcalls), nmp == NULL.
524 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
525 secflavour = RPCSEC_GSS_KRB5;
526 if (nd->nd_procnum != NFSPROC_NULL) {
527 if (clp->lc_flags & LCL_GSSINTEGRITY)
528 secflavour = RPCSEC_GSS_KRB5I;
529 else if (clp->lc_flags & LCL_GSSPRIVACY)
530 secflavour = RPCSEC_GSS_KRB5P;
534 } else if (nmp != NULL && NFSHASKERB(nmp) &&
535 nd->nd_procnum != NFSPROC_NULL) {
536 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
537 nd->nd_flag |= ND_USEGSSNAME;
538 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
540 * If there is a client side host based credential,
541 * use that, otherwise use the system uid, if set.
542 * The system uid is in the nmp->nm_sockreq.nr_cred
545 if (nmp->nm_krbnamelen > 0) {
547 } else if (nmp->nm_uid != (uid_t)-1) {
548 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
549 ("newnfs_request: NULL nr_cred"));
551 authcred = crhold(nmp->nm_sockreq.nr_cred);
553 } else if (nmp->nm_krbnamelen == 0 &&
554 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
556 * If there is no host based principal name and
557 * the system uid is set and this is root, use the
558 * system uid, since root won't have user
559 * credentials in a credentials cache file.
560 * The system uid is in the nmp->nm_sockreq.nr_cred
563 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
564 ("newnfs_request: NULL nr_cred"));
566 authcred = crhold(nmp->nm_sockreq.nr_cred);
568 if (NFSHASINTEGRITY(nmp))
569 secflavour = RPCSEC_GSS_KRB5I;
570 else if (NFSHASPRIVACY(nmp))
571 secflavour = RPCSEC_GSS_KRB5P;
573 secflavour = RPCSEC_GSS_KRB5;
574 srv_principal = NFSMNT_SRVKRBNAME(nmp);
575 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
576 nd->nd_procnum != NFSPROC_NULL &&
577 (nd->nd_flag & ND_USEGSSNAME) != 0) {
579 * Use the uid that did the mount when the RPC is doing
580 * NFSv4 system operations, as indicated by the
581 * ND_USEGSSNAME flag, for the AUTH_SYS case.
582 * The credentials in nm_sockreq.nr_cred were used for the
585 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
586 ("newnfs_request: NULL nr_cred"));
588 authcred = crhold(nmp->nm_sockreq.nr_cred);
592 bzero(&nf, sizeof(struct nfs_feedback_arg));
595 nf.nf_lastmsg = NFSD_MONOSEC -
596 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
599 if (nd->nd_procnum == NFSPROC_NULL)
600 auth = authnone_create();
602 auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
603 srv_principal, NULL, authcred);
605 auth = nfs_getauth(nrp, secflavour, NULL,
606 srv_principal, NULL, authcred);
609 m_freem(nd->nd_mreq);
611 newnfs_restore_sigmask(td, &oldset);
614 bzero(&ext, sizeof(ext));
617 ext.rc_feedback = nfs_feedback;
618 ext.rc_feedback_arg = &nf;
621 procnum = nd->nd_procnum;
622 if ((nd->nd_flag & ND_NFSV4) &&
623 nd->nd_procnum != NFSPROC_NULL &&
624 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
625 procnum = NFSV4PROC_COMPOUND;
628 NFSINCRGLOBAL(newnfsstats.rpcrequests);
630 /* Map the procnum to the old NFSv2 one, as required. */
631 if ((nd->nd_flag & ND_NFSV2) != 0) {
632 if (nd->nd_procnum < NFS_V3NPROCS)
633 procnum = nfsv2_procid[nd->nd_procnum];
635 procnum = NFSV2PROC_NOOP;
639 * Now only used for the R_DONTRECOVER case, but until that is
640 * supported within the krpc code, I need to keep a queue of
641 * outstanding RPCs for nfsv4 client requests.
643 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
644 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
645 M_NFSDREQ, M_WAITOK);
647 if (dtrace_nfscl_nfs234_start_probe != NULL) {
651 if (nd->nd_flag & ND_NFSV4) {
653 nfscl_nfs4_start_probes[nd->nd_procnum];
654 probe_procnum = nd->nd_procnum;
655 } else if (nd->nd_flag & ND_NFSV3) {
656 probe_id = nfscl_nfs3_start_probes[procnum];
657 probe_procnum = procnum;
660 nfscl_nfs2_start_probes[nd->nd_procnum];
661 probe_procnum = procnum;
664 (dtrace_nfscl_nfs234_start_probe)
665 (probe_id, vp, nd->nd_mreq, cred,
673 * This timeout specifies when a new socket should be created,
674 * along with new xid values. For UDP, this should be done
675 * infrequently, since retransmits of RPC requests should normally
681 timo.tv_sec = NFSV4_UPCALLTIMEO;
683 timo.tv_sec = NFSV4_CALLBACKTIMEO;
685 if (nrp->nr_sotype != SOCK_DGRAM) {
687 if ((nmp->nm_flag & NFSMNT_NFSV4))
688 timo.tv_sec = INT_MAX;
690 timo.tv_sec = NFS_TCPTIMEO;
692 if (NFSHASSOFT(nmp)) {
694 * CLSET_RETRIES is set to 2, so this should be
695 * half of the total timeout required.
697 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
700 timo.tv_sec = timeo / NFS_HZ;
701 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
704 /* For UDP hard mounts, use a large value. */
705 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
714 * Chain request into list of outstanding requests.
717 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
723 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
728 * RPC done, unlink the request.
731 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
736 * If there was a successful reply and a tprintf msg.
737 * tprintf a response.
739 if (stat == RPC_SUCCESS) {
741 } else if (stat == RPC_TIMEDOUT) {
742 NFSINCRGLOBAL(newnfsstats.rpctimeouts);
744 } else if (stat == RPC_VERSMISMATCH) {
745 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
747 } else if (stat == RPC_PROGVERSMISMATCH) {
748 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
749 error = EPROTONOSUPPORT;
750 } else if (stat == RPC_INTR) {
753 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
757 m_freem(nd->nd_mreq);
760 FREE((caddr_t)rep, M_NFSDREQ);
762 newnfs_restore_sigmask(td, &oldset);
766 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
769 * Search for any mbufs that are not a multiple of 4 bytes long
770 * or with m_data not longword aligned.
771 * These could cause pointer alignment problems, so copy them to
772 * well aligned mbufs.
774 newnfs_realign(&nd->nd_mrep, M_WAITOK);
775 nd->nd_md = nd->nd_mrep;
776 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
778 if (nd->nd_procnum != NFSPROC_NULL) {
780 * and now the actual NFS xdr.
782 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
783 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
784 if (nd->nd_repstat != 0) {
785 if (((nd->nd_repstat == NFSERR_DELAY ||
786 nd->nd_repstat == NFSERR_GRACE) &&
787 (nd->nd_flag & ND_NFSV4) &&
788 nd->nd_procnum != NFSPROC_DELEGRETURN &&
789 nd->nd_procnum != NFSPROC_SETATTR &&
790 nd->nd_procnum != NFSPROC_READ &&
791 nd->nd_procnum != NFSPROC_WRITE &&
792 nd->nd_procnum != NFSPROC_OPEN &&
793 nd->nd_procnum != NFSPROC_CREATE &&
794 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
795 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
796 nd->nd_procnum != NFSPROC_CLOSE &&
797 nd->nd_procnum != NFSPROC_LOCK &&
798 nd->nd_procnum != NFSPROC_LOCKU) ||
799 (nd->nd_repstat == NFSERR_DELAY &&
800 (nd->nd_flag & ND_NFSV4) == 0) ||
801 nd->nd_repstat == NFSERR_RESOURCE) {
802 if (trylater_delay > NFS_TRYLATERDEL)
803 trylater_delay = NFS_TRYLATERDEL;
804 waituntil = NFSD_MONOSEC + trylater_delay;
805 while (NFSD_MONOSEC < waituntil)
806 (void) nfs_catnap(PZERO, 0, "nfstry");
808 m_freem(nd->nd_mrep);
814 * If the File Handle was stale, invalidate the
815 * lookup cache, just in case.
816 * (vp != NULL implies a client side call)
818 if (nd->nd_repstat == ESTALE && vp != NULL) {
820 if (ncl_call_invalcaches != NULL)
821 (*ncl_call_invalcaches)(vp);
826 * Get rid of the tag, return count, and PUTFH result for V4.
828 if (nd->nd_flag & ND_NFSV4) {
829 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
830 i = fxdr_unsigned(int, *tl);
831 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
834 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
835 i = fxdr_unsigned(int, *++tl);
838 * If the first op's status is non-zero, mark that
839 * there is no more data to process.
842 nd->nd_flag |= ND_NOMOREDATA;
845 * If the first op is Putfh, throw its results away
846 * and toss the op# and status for the first op.
848 if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) {
849 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
850 i = fxdr_unsigned(int, *tl++);
851 j = fxdr_unsigned(int, *tl);
853 * All Compounds that do an Op that must
854 * be in sequence consist of NFSV4OP_PUTFH
855 * followed by one of these. As such, we
856 * can determine if the seqid# should be
859 if ((i == NFSV4OP_OPEN ||
860 i == NFSV4OP_OPENCONFIRM ||
861 i == NFSV4OP_OPENDOWNGRADE ||
862 i == NFSV4OP_CLOSE ||
864 i == NFSV4OP_LOCKU) &&
866 (j != NFSERR_STALECLIENTID &&
867 j != NFSERR_STALESTATEID &&
868 j != NFSERR_BADSTATEID &&
869 j != NFSERR_BADSEQID &&
870 j != NFSERR_BADXDR &&
871 j != NFSERR_RESOURCE &&
872 j != NFSERR_NOFILEHANDLE)))
873 nd->nd_flag |= ND_INCRSEQID;
875 * If the first op's status is non-zero, mark
876 * that there is no more data to process.
879 nd->nd_flag |= ND_NOMOREDATA;
883 * If R_DONTRECOVER is set, replace the stale error
884 * reply, so that recovery isn't initiated.
886 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
887 nd->nd_repstat == NFSERR_STALESTATEID) &&
888 rep != NULL && (rep->r_flags & R_DONTRECOVER))
889 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
894 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
898 if (nd->nd_flag & ND_NFSV4) {
899 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
900 probe_procnum = nd->nd_procnum;
901 } else if (nd->nd_flag & ND_NFSV3) {
902 probe_id = nfscl_nfs3_done_probes[procnum];
903 probe_procnum = procnum;
905 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
906 probe_procnum = procnum;
909 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
910 nd->nd_mreq, cred, probe_procnum, 0);
914 m_freem(nd->nd_mreq);
917 FREE((caddr_t)rep, M_NFSDREQ);
919 newnfs_restore_sigmask(td, &oldset);
922 mbuf_freem(nd->nd_mrep);
923 mbuf_freem(nd->nd_mreq);
926 FREE((caddr_t)rep, M_NFSDREQ);
928 newnfs_restore_sigmask(td, &oldset);
933 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
934 * wait for all requests to complete. This is used by forced unmounts
935 * to terminate any outstanding RPCs.
938 newnfs_nmcancelreqs(struct nfsmount *nmp)
941 if (nmp->nm_sockreq.nr_client != NULL)
942 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
947 * Any signal that can interrupt an NFS operation in an intr mount
948 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
950 int newnfs_sig_set[] = {
959 * Check to see if one of the signals in our subset is pending on
960 * the process (in an intr mount).
963 nfs_sig_pending(sigset_t set)
967 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
968 if (SIGISMEMBER(set, newnfs_sig_set[i]))
974 * The set/restore sigmask functions are used to (temporarily) overwrite
975 * the thread td_sigmask during an RPC call (for example). These are also
976 * used in other places in the NFS client that might tsleep().
979 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
987 td = curthread; /* XXX */
989 /* Remove the NFS set of signals from newset */
991 mtx_lock(&p->p_sigacts->ps_mtx);
992 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
994 * But make sure we leave the ones already masked
995 * by the process, ie. remove the signal from the
996 * temporary signalmask only if it wasn't already
999 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1000 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1001 SIGDELSET(newset, newnfs_sig_set[i]);
1003 mtx_unlock(&p->p_sigacts->ps_mtx);
1004 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1005 SIGPROCMASK_PROC_LOCKED);
1010 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1013 td = curthread; /* XXX */
1014 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1018 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1019 * old one after msleep() returns.
1022 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1028 if ((priority & PCATCH) == 0)
1029 return msleep(ident, mtx, priority, wmesg, timo);
1031 td = curthread; /* XXX */
1032 newnfs_set_sigmask(td, &oldset);
1033 error = msleep(ident, mtx, priority, wmesg, timo);
1034 newnfs_restore_sigmask(td, &oldset);
1040 * Test for a termination condition pending on the process.
1041 * This is used for NFSMNT_INT mounts.
1044 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1049 /* Terminate all requests while attempting a forced unmount. */
1050 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1052 if (!(nmp->nm_flag & NFSMNT_INT))
1058 tmpset = p->p_siglist;
1059 SIGSETOR(tmpset, td->td_siglist);
1060 SIGSETNAND(tmpset, td->td_sigmask);
1061 mtx_lock(&p->p_sigacts->ps_mtx);
1062 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1063 mtx_unlock(&p->p_sigacts->ps_mtx);
1064 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1065 && nfs_sig_pending(tmpset)) {
1074 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1078 p = td ? td->td_proc : NULL;
1080 tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
1081 server, msg, error);
1083 tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
1089 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1090 int error, int flags)
1094 mtx_lock(&nmp->nm_mtx);
1095 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1096 nmp->nm_state |= NFSSTA_TIMEO;
1097 mtx_unlock(&nmp->nm_mtx);
1098 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1101 mtx_unlock(&nmp->nm_mtx);
1102 mtx_lock(&nmp->nm_mtx);
1103 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1104 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1105 mtx_unlock(&nmp->nm_mtx);
1106 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1109 mtx_unlock(&nmp->nm_mtx);
1110 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1114 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1115 int flags, int tprintfmsg)
1120 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1123 mtx_lock(&nmp->nm_mtx);
1124 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1125 nmp->nm_state &= ~NFSSTA_TIMEO;
1126 mtx_unlock(&nmp->nm_mtx);
1127 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1130 mtx_unlock(&nmp->nm_mtx);
1132 mtx_lock(&nmp->nm_mtx);
1133 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1134 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1135 mtx_unlock(&nmp->nm_mtx);
1136 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1139 mtx_unlock(&nmp->nm_mtx);