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");
203 if (nrp->nr_sotype == SOCK_DGRAM)
204 nconf = getnetconfigent("udp6");
206 nconf = getnetconfigent("tcp6");
208 pktscale = nfs_bufpackets;
213 pktscalesav = pktscale;
215 * soreserve() can fail if sb_max is too small, so shrink pktscale
216 * and try again if there is an error.
217 * Print a log message suggesting increasing sb_max.
218 * Creating a socket and doing this is necessary since, if the
219 * reservation sizes are too large and will make soreserve() fail,
220 * the connection will work until a large send is attempted and
221 * then it will loop in the krpc code.
224 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
225 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
226 nrp->nr_soproto, td->td_ucred, td);
228 td->td_ucred = origcred;
232 if (error != 0 && pktscale > 2) {
233 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
234 pktscale == pktscalesav)
235 printf("Consider increasing kern.ipc.maxsockbuf\n");
238 if (nrp->nr_sotype == SOCK_DGRAM) {
240 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
242 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
245 sndreserve = rcvreserve = 1024 * pktscale;
248 if (nrp->nr_sotype != SOCK_STREAM)
249 panic("nfscon sotype");
251 sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
252 sizeof (u_int32_t)) * pktscale;
253 rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
254 sizeof (u_int32_t)) * pktscale;
256 sndreserve = rcvreserve = 1024 * pktscale;
259 error = soreserve(so, sndreserve, rcvreserve);
260 if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
262 printf("Must increase kern.ipc.maxsockbuf or reduce"
264 } while (error != 0 && pktscale > 2);
267 td->td_ucred = origcred;
271 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
272 nrp->nr_vers, sndreserve, rcvreserve);
273 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
275 if ((nmp->nm_flag & NFSMNT_INT))
276 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
277 if ((nmp->nm_flag & NFSMNT_RESVPORT))
278 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
279 if (NFSHASSOFT(nmp)) {
280 if (nmp->nm_sotype == SOCK_DGRAM)
282 * For UDP, the large timeout for a reconnect
283 * will be set to "nm_retry * nm_timeo / 2", so
284 * we only want to do 2 reconnect timeout
289 retries = nmp->nm_retry;
292 /* cred == NULL for DS connects. */
293 if (NFSHASNFSV4N(nmp) && cred != NULL) {
295 * Make sure the nfscbd_pool doesn't get destroyed
299 if (nfs_numnfscbd > 0) {
302 xprt = svc_vc_create_backchannel(nfscbd_pool);
303 CLNT_CONTROL(client, CLSET_BACKCHANNEL, xprt);
306 if (nfs_numnfscbd == 0)
307 wakeup(&nfs_numnfscbd);
314 * - Null RPC callback to client
315 * - Non-Null RPC callback to client, wait a little longer
316 * - upcalls to nfsuserd and gssd (clp == NULL)
318 if (callback_retry_mult == 0) {
319 retries = NFSV4_UPCALLRETRY;
320 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
322 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
325 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
329 * For UDP, there are 2 timeouts:
330 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
331 * that does a retransmit of an RPC request using the same
332 * socket and xid. This is what you normally want to do,
333 * since NFS servers depend on "same xid" for their
334 * Duplicate Request Cache.
335 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
336 * retransmits on the same socket should fail and a fresh
337 * socket created. Each of these timeouts counts as one
338 * CLSET_RETRIES as set above.
339 * Set the initial retransmit timeout for UDP. This timeout
340 * doesn't exist for TCP and the following call just fails,
343 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
344 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
345 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
348 mtx_lock(&nrp->nr_mtx);
349 if (nrp->nr_client != NULL) {
350 mtx_unlock(&nrp->nr_mtx);
352 * Someone else already connected.
354 CLNT_RELEASE(client);
356 nrp->nr_client = client;
358 * Protocols that do not require connections may be optionally
359 * left unconnected for servers that reply from a port other
362 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
363 mtx_unlock(&nrp->nr_mtx);
364 CLNT_CONTROL(client, CLSET_CONNECT, &one);
366 mtx_unlock(&nrp->nr_mtx);
370 /* Restore current thread's credentials. */
371 td->td_ucred = origcred;
379 * NFS disconnect. Clean up and unlink.
382 newnfs_disconnect(struct nfssockreq *nrp)
386 mtx_lock(&nrp->nr_mtx);
387 if (nrp->nr_client != NULL) {
388 client = nrp->nr_client;
389 nrp->nr_client = NULL;
390 mtx_unlock(&nrp->nr_mtx);
391 rpc_gss_secpurge_call(client);
393 CLNT_RELEASE(client);
395 mtx_unlock(&nrp->nr_mtx);
400 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
401 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
403 rpc_gss_service_t svc;
406 switch (secflavour) {
407 case RPCSEC_GSS_KRB5:
408 case RPCSEC_GSS_KRB5I:
409 case RPCSEC_GSS_KRB5P:
411 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
414 if (secflavour == RPCSEC_GSS_KRB5)
415 svc = rpc_gss_svc_none;
416 else if (secflavour == RPCSEC_GSS_KRB5I)
417 svc = rpc_gss_svc_integrity;
419 svc = rpc_gss_svc_privacy;
421 if (clnt_principal == NULL)
422 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
423 srv_principal, mech_oid, svc);
425 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
426 clnt_principal, srv_principal, "kerberosv5",
427 svc, NULL, NULL, NULL);
435 return (authunix_create(cred));
441 * Callback from the RPC code to generate up/down notifications.
444 struct nfs_feedback_arg {
445 struct nfsmount *nf_mount;
446 int nf_lastmsg; /* last tprintf */
448 struct thread *nf_td;
452 nfs_feedback(int type, int proc, void *arg)
454 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
455 struct nfsmount *nmp = nf->nf_mount;
459 case FEEDBACK_REXMIT2:
460 case FEEDBACK_RECONNECT:
462 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
463 nfs_down(nmp, nf->nf_td,
464 "not responding", 0, NFSSTA_TIMEO);
465 nf->nf_tprintfmsg = TRUE;
466 nf->nf_lastmsg = now;
471 nfs_up(nf->nf_mount, nf->nf_td,
472 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
478 * newnfs_request - goes something like this
479 * - does the rpc by calling the krpc layer
480 * - break down rpc header and return with nfs reply
481 * nb: always frees up nd_mreq mbuf list
484 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
485 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
486 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
487 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
489 uint32_t retseq, retval, slotseq, *tl;
491 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
492 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
493 int freeslot, maxslot, reterr, slotpos, timeo;
495 u_int trylater_delay = 1;
496 struct nfs_feedback_arg nf;
499 struct rpc_callextra ext;
501 struct nfsreq *rep = NULL;
502 char *srv_principal = NULL, *clnt_principal = NULL;
504 struct ucred *authcred;
505 struct nfsclsession *sep;
506 uint8_t sessionid[NFSX_V4SESSIONID];
511 /* Reject requests while attempting a forced unmount. */
512 if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
513 m_freem(nd->nd_mreq);
518 * Set authcred, which is used to acquire RPC credentials to
519 * the cred argument, by default. The crhold() should not be
520 * necessary, but will ensure that some future code change
521 * doesn't result in the credential being free'd prematurely.
523 authcred = crhold(cred);
525 /* For client side interruptible mounts, mask off the signals. */
526 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
527 newnfs_set_sigmask(td, &oldset);
532 * XXX if not already connected call nfs_connect now. Longer
533 * term, change nfs_mount to call nfs_connect unconditionally
534 * and let clnt_reconnect_create handle reconnects.
536 if (nrp->nr_client == NULL)
537 newnfs_connect(nmp, nrp, cred, td, 0);
540 * For a client side mount, nmp is != NULL and clp == NULL. For
541 * server calls (callbacks or upcalls), nmp == NULL.
545 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
546 secflavour = RPCSEC_GSS_KRB5;
547 if (nd->nd_procnum != NFSPROC_NULL) {
548 if (clp->lc_flags & LCL_GSSINTEGRITY)
549 secflavour = RPCSEC_GSS_KRB5I;
550 else if (clp->lc_flags & LCL_GSSPRIVACY)
551 secflavour = RPCSEC_GSS_KRB5P;
555 } else if (nmp != NULL && NFSHASKERB(nmp) &&
556 nd->nd_procnum != NFSPROC_NULL) {
557 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
558 nd->nd_flag |= ND_USEGSSNAME;
559 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
561 * If there is a client side host based credential,
562 * use that, otherwise use the system uid, if set.
563 * The system uid is in the nmp->nm_sockreq.nr_cred
566 if (nmp->nm_krbnamelen > 0) {
568 clnt_principal = nmp->nm_krbname;
569 } else if (nmp->nm_uid != (uid_t)-1) {
570 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
571 ("newnfs_request: NULL nr_cred"));
573 authcred = crhold(nmp->nm_sockreq.nr_cred);
575 } else if (nmp->nm_krbnamelen == 0 &&
576 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
578 * If there is no host based principal name and
579 * the system uid is set and this is root, use the
580 * system uid, since root won't have user
581 * credentials in a credentials cache file.
582 * The system uid is in the nmp->nm_sockreq.nr_cred
585 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
586 ("newnfs_request: NULL nr_cred"));
588 authcred = crhold(nmp->nm_sockreq.nr_cred);
590 if (NFSHASINTEGRITY(nmp))
591 secflavour = RPCSEC_GSS_KRB5I;
592 else if (NFSHASPRIVACY(nmp))
593 secflavour = RPCSEC_GSS_KRB5P;
595 secflavour = RPCSEC_GSS_KRB5;
596 srv_principal = NFSMNT_SRVKRBNAME(nmp);
597 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
598 nd->nd_procnum != NFSPROC_NULL &&
599 (nd->nd_flag & ND_USEGSSNAME) != 0) {
601 * Use the uid that did the mount when the RPC is doing
602 * NFSv4 system operations, as indicated by the
603 * ND_USEGSSNAME flag, for the AUTH_SYS case.
604 * The credentials in nm_sockreq.nr_cred were used for the
607 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
608 ("newnfs_request: NULL nr_cred"));
610 authcred = crhold(nmp->nm_sockreq.nr_cred);
614 bzero(&nf, sizeof(struct nfs_feedback_arg));
617 nf.nf_lastmsg = NFSD_MONOSEC -
618 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
621 if (nd->nd_procnum == NFSPROC_NULL)
622 auth = authnone_create();
623 else if (usegssname) {
625 * For this case, the authenticator is held in the
626 * nfssockreq structure, so don't release the reference count
627 * held on it. --> Don't AUTH_DESTROY() it in this function.
629 if (nrp->nr_auth == NULL)
630 nrp->nr_auth = nfs_getauth(nrp, secflavour,
631 clnt_principal, srv_principal, NULL, authcred);
633 rpc_gss_refresh_auth_call(nrp->nr_auth);
636 auth = nfs_getauth(nrp, secflavour, NULL,
637 srv_principal, NULL, authcred);
640 m_freem(nd->nd_mreq);
642 newnfs_restore_sigmask(td, &oldset);
645 bzero(&ext, sizeof(ext));
648 ext.rc_feedback = nfs_feedback;
649 ext.rc_feedback_arg = &nf;
652 procnum = nd->nd_procnum;
653 if ((nd->nd_flag & ND_NFSV4) &&
654 nd->nd_procnum != NFSPROC_NULL &&
655 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
656 procnum = NFSV4PROC_COMPOUND;
659 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
661 /* Map the procnum to the old NFSv2 one, as required. */
662 if ((nd->nd_flag & ND_NFSV2) != 0) {
663 if (nd->nd_procnum < NFS_V3NPROCS)
664 procnum = nfsv2_procid[nd->nd_procnum];
666 procnum = NFSV2PROC_NOOP;
670 * Now only used for the R_DONTRECOVER case, but until that is
671 * supported within the krpc code, I need to keep a queue of
672 * outstanding RPCs for nfsv4 client requests.
674 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
675 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
676 M_NFSDREQ, M_WAITOK);
678 if (dtrace_nfscl_nfs234_start_probe != NULL) {
682 if (nd->nd_flag & ND_NFSV4) {
684 nfscl_nfs4_start_probes[nd->nd_procnum];
685 probe_procnum = nd->nd_procnum;
686 } else if (nd->nd_flag & ND_NFSV3) {
687 probe_id = nfscl_nfs3_start_probes[procnum];
688 probe_procnum = procnum;
691 nfscl_nfs2_start_probes[nd->nd_procnum];
692 probe_procnum = procnum;
695 (dtrace_nfscl_nfs234_start_probe)
696 (probe_id, vp, nd->nd_mreq, cred,
702 freeslot = -1; /* Set to slot that needs to be free'd */
704 slot = -1; /* Slot that needs a sequence# increment. */
706 * This timeout specifies when a new socket should be created,
707 * along with new xid values. For UDP, this should be done
708 * infrequently, since retransmits of RPC requests should normally
714 timo.tv_sec = NFSV4_UPCALLTIMEO;
716 timo.tv_sec = NFSV4_CALLBACKTIMEO;
718 if (nrp->nr_sotype != SOCK_DGRAM) {
720 if ((nmp->nm_flag & NFSMNT_NFSV4))
721 timo.tv_sec = INT_MAX;
723 timo.tv_sec = NFS_TCPTIMEO;
725 if (NFSHASSOFT(nmp)) {
727 * CLSET_RETRIES is set to 2, so this should be
728 * half of the total timeout required.
730 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
733 timo.tv_sec = timeo / NFS_HZ;
734 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
737 /* For UDP hard mounts, use a large value. */
738 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
747 * Chain request into list of outstanding requests.
750 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
756 if (clp != NULL && sep != NULL)
757 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
758 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
760 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
761 nd->nd_mreq, &nd->nd_mrep, timo);
765 * RPC done, unlink the request.
768 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
773 * If there was a successful reply and a tprintf msg.
774 * tprintf a response.
776 if (stat == RPC_SUCCESS) {
778 } else if (stat == RPC_TIMEDOUT) {
779 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
781 } else if (stat == RPC_VERSMISMATCH) {
782 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
784 } else if (stat == RPC_PROGVERSMISMATCH) {
785 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
786 error = EPROTONOSUPPORT;
787 } else if (stat == RPC_INTR) {
790 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
794 m_freem(nd->nd_mreq);
798 FREE((caddr_t)rep, M_NFSDREQ);
800 newnfs_restore_sigmask(td, &oldset);
804 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
807 * Search for any mbufs that are not a multiple of 4 bytes long
808 * or with m_data not longword aligned.
809 * These could cause pointer alignment problems, so copy them to
810 * well aligned mbufs.
812 newnfs_realign(&nd->nd_mrep, M_WAITOK);
813 nd->nd_md = nd->nd_mrep;
814 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
816 if (nd->nd_procnum != NFSPROC_NULL &&
817 nd->nd_procnum != NFSV4PROC_CBNULL) {
818 /* If sep == NULL, set it to the default in nmp. */
819 if (sep == NULL && nmp != NULL)
820 sep = nfsmnt_mdssession(nmp);
822 * and now the actual NFS xdr.
824 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
825 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
826 if (nd->nd_repstat >= 10000)
827 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
828 (int)nd->nd_repstat);
831 * Get rid of the tag, return count and SEQUENCE result for
834 if ((nd->nd_flag & ND_NFSV4) != 0) {
835 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
836 i = fxdr_unsigned(int, *tl);
837 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
840 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
841 opcnt = fxdr_unsigned(int, *tl++);
842 i = fxdr_unsigned(int, *tl++);
843 j = fxdr_unsigned(int, *tl);
845 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
847 * If the first op is Sequence, free up the slot.
849 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
850 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
851 NFSCL_DEBUG(1, "failed seq=%d\n", j);
852 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
853 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j == 0)
855 if (i == NFSV4OP_SEQUENCE)
856 NFSM_DISSECT(tl, uint32_t *,
860 NFSM_DISSECT(tl, uint32_t *,
863 mtx_lock(&sep->nfsess_mtx);
864 if (bcmp(tl, sep->nfsess_sessionid,
865 NFSX_V4SESSIONID) == 0) {
866 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
867 retseq = fxdr_unsigned(uint32_t, *tl++);
868 slot = fxdr_unsigned(int, *tl++);
870 if (retseq != sep->nfsess_slotseq[slot])
871 printf("retseq diff 0x%x\n",
873 retval = fxdr_unsigned(uint32_t, *++tl);
874 if ((retval + 1) < sep->nfsess_foreslots
876 sep->nfsess_foreslots = (retval
878 else if ((retval + 1) >
879 sep->nfsess_foreslots)
880 sep->nfsess_foreslots = (retval
881 < 64) ? (retval + 1) : 64;
883 mtx_unlock(&sep->nfsess_mtx);
885 /* Grab the op and status for the next one. */
887 NFSM_DISSECT(tl, uint32_t *,
889 i = fxdr_unsigned(int, *tl++);
890 j = fxdr_unsigned(int, *tl);
894 if (nd->nd_repstat != 0) {
895 if (nd->nd_repstat == NFSERR_BADSESSION &&
896 nmp != NULL && dssep == NULL) {
898 * If this is a client side MDS RPC, mark
899 * the MDS session defunct and initiate
900 * recovery, as required.
901 * The nfsess_defunct field is protected by
902 * the NFSLOCKMNT()/nm_mtx lock and not the
903 * nfsess_mtx lock to simplify its handling,
904 * for the MDS session. This lock is also
905 * sufficient for nfsess_sessionid, since it
906 * never changes in the structure.
908 NFSCL_DEBUG(1, "Got badsession\n");
911 sep = NFSMNT_MDSSESSION(nmp);
912 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
913 NFSX_V4SESSIONID) == 0) {
914 /* Initiate recovery. */
915 sep->nfsess_defunct = 1;
916 NFSCL_DEBUG(1, "Marked defunct\n");
917 if (nmp->nm_clp != NULL) {
918 nmp->nm_clp->nfsc_flags |=
925 * Sleep for up to 1sec waiting for a new
928 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
931 * Get the session again, in case a new one
932 * has been created during the sleep.
934 sep = NFSMNT_MDSSESSION(nmp);
936 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
937 reterr = nfsv4_sequencelookup(nmp, sep,
938 &slotpos, &maxslot, &slotseq,
941 /* Fill in new session info. */
943 "Filling in new sequence\n");
944 tl = nd->nd_sequence;
947 tl += NFSX_V4SESSIONID /
949 *tl++ = txdr_unsigned(slotseq);
950 *tl++ = txdr_unsigned(slotpos);
951 *tl = txdr_unsigned(maxslot);
953 if (reterr == NFSERR_BADSESSION ||
956 "Badsession looping\n");
957 m_freem(nd->nd_mrep);
961 nd->nd_repstat = reterr;
962 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
965 if (((nd->nd_repstat == NFSERR_DELAY ||
966 nd->nd_repstat == NFSERR_GRACE) &&
967 (nd->nd_flag & ND_NFSV4) &&
968 nd->nd_procnum != NFSPROC_DELEGRETURN &&
969 nd->nd_procnum != NFSPROC_SETATTR &&
970 nd->nd_procnum != NFSPROC_READ &&
971 nd->nd_procnum != NFSPROC_READDS &&
972 nd->nd_procnum != NFSPROC_WRITE &&
973 nd->nd_procnum != NFSPROC_WRITEDS &&
974 nd->nd_procnum != NFSPROC_OPEN &&
975 nd->nd_procnum != NFSPROC_CREATE &&
976 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
977 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
978 nd->nd_procnum != NFSPROC_CLOSE &&
979 nd->nd_procnum != NFSPROC_LOCK &&
980 nd->nd_procnum != NFSPROC_LOCKU) ||
981 (nd->nd_repstat == NFSERR_DELAY &&
982 (nd->nd_flag & ND_NFSV4) == 0) ||
983 nd->nd_repstat == NFSERR_RESOURCE) {
984 if (trylater_delay > NFS_TRYLATERDEL)
985 trylater_delay = NFS_TRYLATERDEL;
986 waituntil = NFSD_MONOSEC + trylater_delay;
987 while (NFSD_MONOSEC < waituntil)
988 (void) nfs_catnap(PZERO, 0, "nfstry");
991 mtx_lock(&sep->nfsess_mtx);
992 sep->nfsess_slotseq[slot]++;
993 *nd->nd_slotseq = txdr_unsigned(
994 sep->nfsess_slotseq[slot]);
995 mtx_unlock(&sep->nfsess_mtx);
997 m_freem(nd->nd_mrep);
1003 * If the File Handle was stale, invalidate the
1004 * lookup cache, just in case.
1005 * (vp != NULL implies a client side call)
1007 if (nd->nd_repstat == ESTALE && vp != NULL) {
1009 if (ncl_call_invalcaches != NULL)
1010 (*ncl_call_invalcaches)(vp);
1013 if ((nd->nd_flag & ND_NFSV4) != 0) {
1014 /* Free the slot, as required. */
1016 nfsv4_freeslot(sep, freeslot);
1018 * If this op is Putfh, throw its results away.
1021 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1022 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1023 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1024 i = fxdr_unsigned(int, *tl++);
1025 j = fxdr_unsigned(int, *tl);
1027 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1030 * All Compounds that do an Op that must
1031 * be in sequence consist of NFSV4OP_PUTFH
1032 * followed by one of these. As such, we
1033 * can determine if the seqid# should be
1034 * incremented, here.
1036 if ((i == NFSV4OP_OPEN ||
1037 i == NFSV4OP_OPENCONFIRM ||
1038 i == NFSV4OP_OPENDOWNGRADE ||
1039 i == NFSV4OP_CLOSE ||
1040 i == NFSV4OP_LOCK ||
1041 i == NFSV4OP_LOCKU) &&
1043 (j != NFSERR_STALECLIENTID &&
1044 j != NFSERR_STALESTATEID &&
1045 j != NFSERR_BADSTATEID &&
1046 j != NFSERR_BADSEQID &&
1047 j != NFSERR_BADXDR &&
1048 j != NFSERR_RESOURCE &&
1049 j != NFSERR_NOFILEHANDLE)))
1050 nd->nd_flag |= ND_INCRSEQID;
1053 * If this op's status is non-zero, mark
1054 * that there is no more data to process.
1055 * The exception is Setattr, which always has xdr
1056 * when it has failed.
1058 if (j != 0 && i != NFSV4OP_SETATTR)
1059 nd->nd_flag |= ND_NOMOREDATA;
1062 * If R_DONTRECOVER is set, replace the stale error
1063 * reply, so that recovery isn't initiated.
1065 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1066 nd->nd_repstat == NFSERR_BADSESSION ||
1067 nd->nd_repstat == NFSERR_STALESTATEID) &&
1068 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1069 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1073 #ifdef KDTRACE_HOOKS
1074 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1078 if (nd->nd_flag & ND_NFSV4) {
1079 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1080 probe_procnum = nd->nd_procnum;
1081 } else if (nd->nd_flag & ND_NFSV3) {
1082 probe_id = nfscl_nfs3_done_probes[procnum];
1083 probe_procnum = procnum;
1085 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1086 probe_procnum = procnum;
1089 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1090 nd->nd_mreq, cred, probe_procnum, 0);
1094 m_freem(nd->nd_mreq);
1095 if (usegssname == 0)
1098 FREE((caddr_t)rep, M_NFSDREQ);
1100 newnfs_restore_sigmask(td, &oldset);
1103 mbuf_freem(nd->nd_mrep);
1104 mbuf_freem(nd->nd_mreq);
1105 if (usegssname == 0)
1108 FREE((caddr_t)rep, M_NFSDREQ);
1110 newnfs_restore_sigmask(td, &oldset);
1115 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1116 * wait for all requests to complete. This is used by forced unmounts
1117 * to terminate any outstanding RPCs.
1120 newnfs_nmcancelreqs(struct nfsmount *nmp)
1123 if (nmp->nm_sockreq.nr_client != NULL)
1124 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1129 * Any signal that can interrupt an NFS operation in an intr mount
1130 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1132 int newnfs_sig_set[] = {
1141 * Check to see if one of the signals in our subset is pending on
1142 * the process (in an intr mount).
1145 nfs_sig_pending(sigset_t set)
1149 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1150 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1156 * The set/restore sigmask functions are used to (temporarily) overwrite
1157 * the thread td_sigmask during an RPC call (for example). These are also
1158 * used in other places in the NFS client that might tsleep().
1161 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1169 td = curthread; /* XXX */
1171 /* Remove the NFS set of signals from newset */
1173 mtx_lock(&p->p_sigacts->ps_mtx);
1174 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1176 * But make sure we leave the ones already masked
1177 * by the process, ie. remove the signal from the
1178 * temporary signalmask only if it wasn't already
1181 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1182 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1183 SIGDELSET(newset, newnfs_sig_set[i]);
1185 mtx_unlock(&p->p_sigacts->ps_mtx);
1186 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1187 SIGPROCMASK_PROC_LOCKED);
1192 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1195 td = curthread; /* XXX */
1196 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1200 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1201 * old one after msleep() returns.
1204 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1210 if ((priority & PCATCH) == 0)
1211 return msleep(ident, mtx, priority, wmesg, timo);
1213 td = curthread; /* XXX */
1214 newnfs_set_sigmask(td, &oldset);
1215 error = msleep(ident, mtx, priority, wmesg, timo);
1216 newnfs_restore_sigmask(td, &oldset);
1222 * Test for a termination condition pending on the process.
1223 * This is used for NFSMNT_INT mounts.
1226 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1231 /* Terminate all requests while attempting a forced unmount. */
1232 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1234 if (!(nmp->nm_flag & NFSMNT_INT))
1240 tmpset = p->p_siglist;
1241 SIGSETOR(tmpset, td->td_siglist);
1242 SIGSETNAND(tmpset, td->td_sigmask);
1243 mtx_lock(&p->p_sigacts->ps_mtx);
1244 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1245 mtx_unlock(&p->p_sigacts->ps_mtx);
1246 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1247 && nfs_sig_pending(tmpset)) {
1256 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1260 p = td ? td->td_proc : NULL;
1262 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1263 server, msg, error);
1265 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1271 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1272 int error, int flags)
1276 mtx_lock(&nmp->nm_mtx);
1277 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1278 nmp->nm_state |= NFSSTA_TIMEO;
1279 mtx_unlock(&nmp->nm_mtx);
1280 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1283 mtx_unlock(&nmp->nm_mtx);
1284 mtx_lock(&nmp->nm_mtx);
1285 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1286 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1287 mtx_unlock(&nmp->nm_mtx);
1288 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1291 mtx_unlock(&nmp->nm_mtx);
1292 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1296 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1297 int flags, int tprintfmsg)
1302 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1305 mtx_lock(&nmp->nm_mtx);
1306 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1307 nmp->nm_state &= ~NFSSTA_TIMEO;
1308 mtx_unlock(&nmp->nm_mtx);
1309 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1312 mtx_unlock(&nmp->nm_mtx);
1314 mtx_lock(&nmp->nm_mtx);
1315 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1316 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1317 mtx_unlock(&nmp->nm_mtx);
1318 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1321 mtx_unlock(&nmp->nm_mtx);