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>
64 #include <kgssapi/krb5/kcrypto.h>
66 #include <fs/nfs/nfsport.h>
69 #include <sys/dtrace_bsd.h>
71 dtrace_nfsclient_nfs23_start_probe_func_t
72 dtrace_nfscl_nfs234_start_probe;
74 dtrace_nfsclient_nfs23_done_probe_func_t
75 dtrace_nfscl_nfs234_done_probe;
78 * Registered probes by RPC type.
80 uint32_t nfscl_nfs2_start_probes[NFSV41_NPROCS + 1];
81 uint32_t nfscl_nfs2_done_probes[NFSV41_NPROCS + 1];
83 uint32_t nfscl_nfs3_start_probes[NFSV41_NPROCS + 1];
84 uint32_t nfscl_nfs3_done_probes[NFSV41_NPROCS + 1];
86 uint32_t nfscl_nfs4_start_probes[NFSV41_NPROCS + 1];
87 uint32_t nfscl_nfs4_done_probes[NFSV41_NPROCS + 1];
93 extern struct nfsstats newnfsstats;
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, "newnfsreq");
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 *sep)
479 u_int32_t retseq, retval, *tl;
481 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
482 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
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;
498 /* Reject requests while attempting a forced unmount. */
499 if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
500 m_freem(nd->nd_mreq);
505 * Set authcred, which is used to acquire RPC credentials to
506 * the cred argument, by default. The crhold() should not be
507 * necessary, but will ensure that some future code change
508 * doesn't result in the credential being free'd prematurely.
510 authcred = crhold(cred);
512 /* For client side interruptible mounts, mask off the signals. */
513 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
514 newnfs_set_sigmask(td, &oldset);
519 * XXX if not already connected call nfs_connect now. Longer
520 * term, change nfs_mount to call nfs_connect unconditionally
521 * and let clnt_reconnect_create handle reconnects.
523 if (nrp->nr_client == NULL)
524 newnfs_connect(nmp, nrp, cred, td, 0);
527 * For a client side mount, nmp is != NULL and clp == NULL. For
528 * server calls (callbacks or upcalls), nmp == NULL.
532 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
533 secflavour = RPCSEC_GSS_KRB5;
534 if (nd->nd_procnum != NFSPROC_NULL) {
535 if (clp->lc_flags & LCL_GSSINTEGRITY)
536 secflavour = RPCSEC_GSS_KRB5I;
537 else if (clp->lc_flags & LCL_GSSPRIVACY)
538 secflavour = RPCSEC_GSS_KRB5P;
542 } else if (nmp != NULL && NFSHASKERB(nmp) &&
543 nd->nd_procnum != NFSPROC_NULL) {
544 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
545 nd->nd_flag |= ND_USEGSSNAME;
546 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
548 * If there is a client side host based credential,
549 * use that, otherwise use the system uid, if set.
550 * The system uid is in the nmp->nm_sockreq.nr_cred
553 if (nmp->nm_krbnamelen > 0) {
555 clnt_principal = nmp->nm_krbname;
556 } else if (nmp->nm_uid != (uid_t)-1) {
557 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
558 ("newnfs_request: NULL nr_cred"));
560 authcred = crhold(nmp->nm_sockreq.nr_cred);
562 } else if (nmp->nm_krbnamelen == 0 &&
563 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
565 * If there is no host based principal name and
566 * the system uid is set and this is root, use the
567 * system uid, since root won't have user
568 * credentials in a credentials cache file.
569 * The system uid is in the nmp->nm_sockreq.nr_cred
572 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
573 ("newnfs_request: NULL nr_cred"));
575 authcred = crhold(nmp->nm_sockreq.nr_cred);
577 if (NFSHASINTEGRITY(nmp))
578 secflavour = RPCSEC_GSS_KRB5I;
579 else if (NFSHASPRIVACY(nmp))
580 secflavour = RPCSEC_GSS_KRB5P;
582 secflavour = RPCSEC_GSS_KRB5;
583 srv_principal = NFSMNT_SRVKRBNAME(nmp);
584 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
585 nd->nd_procnum != NFSPROC_NULL &&
586 (nd->nd_flag & ND_USEGSSNAME) != 0) {
588 * Use the uid that did the mount when the RPC is doing
589 * NFSv4 system operations, as indicated by the
590 * ND_USEGSSNAME flag, for the AUTH_SYS case.
591 * The credentials in nm_sockreq.nr_cred were used for the
594 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
595 ("newnfs_request: NULL nr_cred"));
597 authcred = crhold(nmp->nm_sockreq.nr_cred);
601 bzero(&nf, sizeof(struct nfs_feedback_arg));
604 nf.nf_lastmsg = NFSD_MONOSEC -
605 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
608 if (nd->nd_procnum == NFSPROC_NULL)
609 auth = authnone_create();
610 else if (usegssname) {
612 * For this case, the authenticator is held in the
613 * nfssockreq structure, so don't release the reference count
614 * held on it. --> Don't AUTH_DESTROY() it in this function.
616 if (nrp->nr_auth == NULL)
617 nrp->nr_auth = nfs_getauth(nrp, secflavour,
618 clnt_principal, srv_principal, NULL, authcred);
620 rpc_gss_refresh_auth_call(nrp->nr_auth);
623 auth = nfs_getauth(nrp, secflavour, NULL,
624 srv_principal, NULL, authcred);
627 m_freem(nd->nd_mreq);
629 newnfs_restore_sigmask(td, &oldset);
632 bzero(&ext, sizeof(ext));
635 ext.rc_feedback = nfs_feedback;
636 ext.rc_feedback_arg = &nf;
639 procnum = nd->nd_procnum;
640 if ((nd->nd_flag & ND_NFSV4) &&
641 nd->nd_procnum != NFSPROC_NULL &&
642 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
643 procnum = NFSV4PROC_COMPOUND;
646 NFSINCRGLOBAL(newnfsstats.rpcrequests);
648 /* Map the procnum to the old NFSv2 one, as required. */
649 if ((nd->nd_flag & ND_NFSV2) != 0) {
650 if (nd->nd_procnum < NFS_V3NPROCS)
651 procnum = nfsv2_procid[nd->nd_procnum];
653 procnum = NFSV2PROC_NOOP;
657 * Now only used for the R_DONTRECOVER case, but until that is
658 * supported within the krpc code, I need to keep a queue of
659 * outstanding RPCs for nfsv4 client requests.
661 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
662 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
663 M_NFSDREQ, M_WAITOK);
665 if (dtrace_nfscl_nfs234_start_probe != NULL) {
669 if (nd->nd_flag & ND_NFSV4) {
671 nfscl_nfs4_start_probes[nd->nd_procnum];
672 probe_procnum = nd->nd_procnum;
673 } else if (nd->nd_flag & ND_NFSV3) {
674 probe_id = nfscl_nfs3_start_probes[procnum];
675 probe_procnum = procnum;
678 nfscl_nfs2_start_probes[nd->nd_procnum];
679 probe_procnum = procnum;
682 (dtrace_nfscl_nfs234_start_probe)
683 (probe_id, vp, nd->nd_mreq, cred,
689 freeslot = -1; /* Set to slot that needs to be free'd */
691 slot = -1; /* Slot that needs a sequence# increment. */
693 * This timeout specifies when a new socket should be created,
694 * along with new xid values. For UDP, this should be done
695 * infrequently, since retransmits of RPC requests should normally
701 timo.tv_sec = NFSV4_UPCALLTIMEO;
703 timo.tv_sec = NFSV4_CALLBACKTIMEO;
705 if (nrp->nr_sotype != SOCK_DGRAM) {
707 if ((nmp->nm_flag & NFSMNT_NFSV4))
708 timo.tv_sec = INT_MAX;
710 timo.tv_sec = NFS_TCPTIMEO;
712 if (NFSHASSOFT(nmp)) {
714 * CLSET_RETRIES is set to 2, so this should be
715 * half of the total timeout required.
717 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
720 timo.tv_sec = timeo / NFS_HZ;
721 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
724 /* For UDP hard mounts, use a large value. */
725 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
734 * Chain request into list of outstanding requests.
737 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
743 if (clp != NULL && sep != NULL)
744 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
745 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
747 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
748 nd->nd_mreq, &nd->nd_mrep, timo);
752 * RPC done, unlink the request.
755 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
760 * If there was a successful reply and a tprintf msg.
761 * tprintf a response.
763 if (stat == RPC_SUCCESS) {
765 } else if (stat == RPC_TIMEDOUT) {
766 NFSINCRGLOBAL(newnfsstats.rpctimeouts);
768 } else if (stat == RPC_VERSMISMATCH) {
769 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
771 } else if (stat == RPC_PROGVERSMISMATCH) {
772 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
773 error = EPROTONOSUPPORT;
774 } else if (stat == RPC_INTR) {
777 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
781 m_freem(nd->nd_mreq);
785 FREE((caddr_t)rep, M_NFSDREQ);
787 newnfs_restore_sigmask(td, &oldset);
791 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
794 * Search for any mbufs that are not a multiple of 4 bytes long
795 * or with m_data not longword aligned.
796 * These could cause pointer alignment problems, so copy them to
797 * well aligned mbufs.
799 newnfs_realign(&nd->nd_mrep, M_WAITOK);
800 nd->nd_md = nd->nd_mrep;
801 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
803 if (nd->nd_procnum != NFSPROC_NULL &&
804 nd->nd_procnum != NFSV4PROC_CBNULL) {
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 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
838 NFSCL_DEBUG(1, "failed seq=%d\n", j);
839 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
840 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j == 0)
842 if (i == NFSV4OP_SEQUENCE)
843 NFSM_DISSECT(tl, uint32_t *,
847 NFSM_DISSECT(tl, uint32_t *,
850 mtx_lock(&sep->nfsess_mtx);
851 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
852 retseq = fxdr_unsigned(uint32_t, *tl++);
853 slot = fxdr_unsigned(int, *tl++);
855 if (retseq != sep->nfsess_slotseq[slot])
856 printf("retseq diff 0x%x\n", retseq);
857 retval = fxdr_unsigned(uint32_t, *++tl);
858 if ((retval + 1) < sep->nfsess_foreslots)
859 sep->nfsess_foreslots = (retval + 1);
860 else if ((retval + 1) > sep->nfsess_foreslots)
861 sep->nfsess_foreslots = (retval < 64) ?
863 mtx_unlock(&sep->nfsess_mtx);
865 /* Grab the op and status for the next one. */
867 NFSM_DISSECT(tl, uint32_t *,
869 i = fxdr_unsigned(int, *tl++);
870 j = fxdr_unsigned(int, *tl);
874 if (nd->nd_repstat != 0) {
875 if (((nd->nd_repstat == NFSERR_DELAY ||
876 nd->nd_repstat == NFSERR_GRACE) &&
877 (nd->nd_flag & ND_NFSV4) &&
878 nd->nd_procnum != NFSPROC_DELEGRETURN &&
879 nd->nd_procnum != NFSPROC_SETATTR &&
880 nd->nd_procnum != NFSPROC_READ &&
881 nd->nd_procnum != NFSPROC_READDS &&
882 nd->nd_procnum != NFSPROC_WRITE &&
883 nd->nd_procnum != NFSPROC_WRITEDS &&
884 nd->nd_procnum != NFSPROC_OPEN &&
885 nd->nd_procnum != NFSPROC_CREATE &&
886 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
887 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
888 nd->nd_procnum != NFSPROC_CLOSE &&
889 nd->nd_procnum != NFSPROC_LOCK &&
890 nd->nd_procnum != NFSPROC_LOCKU) ||
891 (nd->nd_repstat == NFSERR_DELAY &&
892 (nd->nd_flag & ND_NFSV4) == 0) ||
893 nd->nd_repstat == NFSERR_RESOURCE) {
894 if (trylater_delay > NFS_TRYLATERDEL)
895 trylater_delay = NFS_TRYLATERDEL;
896 waituntil = NFSD_MONOSEC + trylater_delay;
897 while (NFSD_MONOSEC < waituntil)
898 (void) nfs_catnap(PZERO, 0, "nfstry");
901 mtx_lock(&sep->nfsess_mtx);
902 sep->nfsess_slotseq[slot]++;
903 *nd->nd_slotseq = txdr_unsigned(
904 sep->nfsess_slotseq[slot]);
905 mtx_unlock(&sep->nfsess_mtx);
907 m_freem(nd->nd_mrep);
913 * If the File Handle was stale, invalidate the
914 * lookup cache, just in case.
915 * (vp != NULL implies a client side call)
917 if (nd->nd_repstat == ESTALE && vp != NULL) {
919 if (ncl_call_invalcaches != NULL)
920 (*ncl_call_invalcaches)(vp);
923 if ((nd->nd_flag & ND_NFSV4) != 0) {
924 /* Free the slot, as required. */
926 nfsv4_freeslot(sep, freeslot);
928 * If this op is Putfh, throw its results away.
931 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
932 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
933 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
934 i = fxdr_unsigned(int, *tl++);
935 j = fxdr_unsigned(int, *tl);
937 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
940 * All Compounds that do an Op that must
941 * be in sequence consist of NFSV4OP_PUTFH
942 * followed by one of these. As such, we
943 * can determine if the seqid# should be
946 if ((i == NFSV4OP_OPEN ||
947 i == NFSV4OP_OPENCONFIRM ||
948 i == NFSV4OP_OPENDOWNGRADE ||
949 i == NFSV4OP_CLOSE ||
951 i == NFSV4OP_LOCKU) &&
953 (j != NFSERR_STALECLIENTID &&
954 j != NFSERR_STALESTATEID &&
955 j != NFSERR_BADSTATEID &&
956 j != NFSERR_BADSEQID &&
957 j != NFSERR_BADXDR &&
958 j != NFSERR_RESOURCE &&
959 j != NFSERR_NOFILEHANDLE)))
960 nd->nd_flag |= ND_INCRSEQID;
963 * If this op's status is non-zero, mark
964 * that there is no more data to process.
967 nd->nd_flag |= ND_NOMOREDATA;
970 * If R_DONTRECOVER is set, replace the stale error
971 * reply, so that recovery isn't initiated.
973 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
974 nd->nd_repstat == NFSERR_BADSESSION ||
975 nd->nd_repstat == NFSERR_STALESTATEID) &&
976 rep != NULL && (rep->r_flags & R_DONTRECOVER))
977 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
982 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
986 if (nd->nd_flag & ND_NFSV4) {
987 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
988 probe_procnum = nd->nd_procnum;
989 } else if (nd->nd_flag & ND_NFSV3) {
990 probe_id = nfscl_nfs3_done_probes[procnum];
991 probe_procnum = procnum;
993 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
994 probe_procnum = procnum;
997 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
998 nd->nd_mreq, cred, probe_procnum, 0);
1002 m_freem(nd->nd_mreq);
1003 if (usegssname == 0)
1006 FREE((caddr_t)rep, M_NFSDREQ);
1008 newnfs_restore_sigmask(td, &oldset);
1011 mbuf_freem(nd->nd_mrep);
1012 mbuf_freem(nd->nd_mreq);
1013 if (usegssname == 0)
1016 FREE((caddr_t)rep, M_NFSDREQ);
1018 newnfs_restore_sigmask(td, &oldset);
1023 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1024 * wait for all requests to complete. This is used by forced unmounts
1025 * to terminate any outstanding RPCs.
1028 newnfs_nmcancelreqs(struct nfsmount *nmp)
1031 if (nmp->nm_sockreq.nr_client != NULL)
1032 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1037 * Any signal that can interrupt an NFS operation in an intr mount
1038 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1040 int newnfs_sig_set[] = {
1049 * Check to see if one of the signals in our subset is pending on
1050 * the process (in an intr mount).
1053 nfs_sig_pending(sigset_t set)
1057 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
1058 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1064 * The set/restore sigmask functions are used to (temporarily) overwrite
1065 * the thread td_sigmask during an RPC call (for example). These are also
1066 * used in other places in the NFS client that might tsleep().
1069 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1077 td = curthread; /* XXX */
1079 /* Remove the NFS set of signals from newset */
1081 mtx_lock(&p->p_sigacts->ps_mtx);
1082 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
1084 * But make sure we leave the ones already masked
1085 * by the process, ie. remove the signal from the
1086 * temporary signalmask only if it wasn't already
1089 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1090 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1091 SIGDELSET(newset, newnfs_sig_set[i]);
1093 mtx_unlock(&p->p_sigacts->ps_mtx);
1094 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1095 SIGPROCMASK_PROC_LOCKED);
1100 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1103 td = curthread; /* XXX */
1104 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1108 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1109 * old one after msleep() returns.
1112 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1118 if ((priority & PCATCH) == 0)
1119 return msleep(ident, mtx, priority, wmesg, timo);
1121 td = curthread; /* XXX */
1122 newnfs_set_sigmask(td, &oldset);
1123 error = msleep(ident, mtx, priority, wmesg, timo);
1124 newnfs_restore_sigmask(td, &oldset);
1130 * Test for a termination condition pending on the process.
1131 * This is used for NFSMNT_INT mounts.
1134 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1139 /* Terminate all requests while attempting a forced unmount. */
1140 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1142 if (!(nmp->nm_flag & NFSMNT_INT))
1148 tmpset = p->p_siglist;
1149 SIGSETOR(tmpset, td->td_siglist);
1150 SIGSETNAND(tmpset, td->td_sigmask);
1151 mtx_lock(&p->p_sigacts->ps_mtx);
1152 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1153 mtx_unlock(&p->p_sigacts->ps_mtx);
1154 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1155 && nfs_sig_pending(tmpset)) {
1164 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1168 p = td ? td->td_proc : NULL;
1170 tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
1171 server, msg, error);
1173 tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
1179 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1180 int error, int flags)
1184 mtx_lock(&nmp->nm_mtx);
1185 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1186 nmp->nm_state |= NFSSTA_TIMEO;
1187 mtx_unlock(&nmp->nm_mtx);
1188 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1191 mtx_unlock(&nmp->nm_mtx);
1192 mtx_lock(&nmp->nm_mtx);
1193 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1194 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1195 mtx_unlock(&nmp->nm_mtx);
1196 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1199 mtx_unlock(&nmp->nm_mtx);
1200 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1204 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1205 int flags, int tprintfmsg)
1210 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1213 mtx_lock(&nmp->nm_mtx);
1214 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1215 nmp->nm_state &= ~NFSSTA_TIMEO;
1216 mtx_unlock(&nmp->nm_mtx);
1217 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1220 mtx_unlock(&nmp->nm_mtx);
1222 mtx_lock(&nmp->nm_mtx);
1223 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1224 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1225 mtx_unlock(&nmp->nm_mtx);
1226 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1229 mtx_unlock(&nmp->nm_mtx);