xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision d485c77f203fb0f4cdc08dea5ff81631b51d8809)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1991, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 /*
40  * Socket operations for use by nfs
41  */
42 
43 #include "opt_kgssapi.h"
44 #include "opt_nfs.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/limits.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/signalvar.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysctl.h>
59 #include <sys/syslog.h>
60 #include <sys/vnode.h>
61 
62 #include <rpc/rpc.h>
63 #include <rpc/krpc.h>
64 
65 #include <kgssapi/krb5/kcrypto.h>
66 
67 #include <fs/nfs/nfsport.h>
68 
69 #ifdef KDTRACE_HOOKS
70 #include <sys/dtrace_bsd.h>
71 
72 dtrace_nfsclient_nfs23_start_probe_func_t
73 		dtrace_nfscl_nfs234_start_probe;
74 
75 dtrace_nfsclient_nfs23_done_probe_func_t
76 		dtrace_nfscl_nfs234_done_probe;
77 
78 /*
79  * Registered probes by RPC type.
80  */
81 uint32_t	nfscl_nfs2_start_probes[NFSV41_NPROCS + 1];
82 uint32_t	nfscl_nfs2_done_probes[NFSV41_NPROCS + 1];
83 
84 uint32_t	nfscl_nfs3_start_probes[NFSV41_NPROCS + 1];
85 uint32_t	nfscl_nfs3_done_probes[NFSV41_NPROCS + 1];
86 
87 uint32_t	nfscl_nfs4_start_probes[NFSV41_NPROCS + 1];
88 uint32_t	nfscl_nfs4_done_probes[NFSV41_NPROCS + 1];
89 #endif
90 
91 NFSSTATESPINLOCK;
92 NFSREQSPINLOCK;
93 NFSDLOCKMUTEX;
94 NFSCLSTATEMUTEX;
95 extern struct nfsstatsv1 nfsstatsv1;
96 extern struct nfsreqhead nfsd_reqq;
97 extern int nfscl_ticks;
98 extern void (*ncl_call_invalcaches)(struct vnode *);
99 extern int nfs_numnfscbd;
100 extern int nfscl_debuglevel;
101 extern int nfsrv_lease;
102 
103 SVCPOOL		*nfscbd_pool;
104 static int	nfsrv_gsscallbackson = 0;
105 static int	nfs_bufpackets = 4;
106 static int	nfs_reconnects;
107 static int	nfs3_jukebox_delay = 10;
108 static int	nfs_skip_wcc_data_onerr = 1;
109 static int	nfs_dsretries = 2;
110 
111 SYSCTL_DECL(_vfs_nfs);
112 
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
114     "Buffer reservation size 2 < x < 64");
115 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
116     "Number of times the nfs client has had to reconnect");
117 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
118     "Number of seconds to delay a retry after receiving EJUKEBOX");
119 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
120     "Disable weak cache consistency checking when server returns an error");
121 SYSCTL_INT(_vfs_nfs, OID_AUTO, dsretries, CTLFLAG_RW, &nfs_dsretries, 0,
122     "Number of retries for a DS RPC before failure");
123 
124 static void	nfs_down(struct nfsmount *, struct thread *, const char *,
125     int, int);
126 static void	nfs_up(struct nfsmount *, struct thread *, const char *,
127     int, int);
128 static int	nfs_msg(struct thread *, const char *, const char *, int);
129 
130 struct nfs_cached_auth {
131 	int		ca_refs; /* refcount, including 1 from the cache */
132 	uid_t		ca_uid;	 /* uid that corresponds to this auth */
133 	AUTH		*ca_auth; /* RPC auth handle */
134 };
135 
136 static int nfsv2_procid[NFS_V3NPROCS] = {
137 	NFSV2PROC_NULL,
138 	NFSV2PROC_GETATTR,
139 	NFSV2PROC_SETATTR,
140 	NFSV2PROC_LOOKUP,
141 	NFSV2PROC_NOOP,
142 	NFSV2PROC_READLINK,
143 	NFSV2PROC_READ,
144 	NFSV2PROC_WRITE,
145 	NFSV2PROC_CREATE,
146 	NFSV2PROC_MKDIR,
147 	NFSV2PROC_SYMLINK,
148 	NFSV2PROC_CREATE,
149 	NFSV2PROC_REMOVE,
150 	NFSV2PROC_RMDIR,
151 	NFSV2PROC_RENAME,
152 	NFSV2PROC_LINK,
153 	NFSV2PROC_READDIR,
154 	NFSV2PROC_NOOP,
155 	NFSV2PROC_STATFS,
156 	NFSV2PROC_NOOP,
157 	NFSV2PROC_NOOP,
158 	NFSV2PROC_NOOP,
159 };
160 
161 /*
162  * Initialize sockets and congestion for a new NFS connection.
163  * We do not free the sockaddr if error.
164  * Which arguments are set to NULL indicate what kind of call it is.
165  * cred == NULL --> a call to connect to a pNFS DS
166  * nmp == NULL --> indicates an upcall to userland or a NFSv4.0 callback
167  */
168 int
169 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
170     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult, bool dotls)
171 {
172 	int rcvreserve, sndreserve;
173 	int pktscale, pktscalesav;
174 	struct sockaddr *saddr;
175 	struct ucred *origcred;
176 	CLIENT *client;
177 	struct netconfig *nconf;
178 	struct socket *so;
179 	int one = 1, retries, error = 0;
180 	struct thread *td = curthread;
181 	SVCXPRT *xprt;
182 	struct timeval timo;
183 
184 	/*
185 	 * We need to establish the socket using the credentials of
186 	 * the mountpoint.  Some parts of this process (such as
187 	 * sobind() and soconnect()) will use the curent thread's
188 	 * credential instead of the socket credential.  To work
189 	 * around this, temporarily change the current thread's
190 	 * credential to that of the mountpoint.
191 	 *
192 	 * XXX: It would be better to explicitly pass the correct
193 	 * credential to sobind() and soconnect().
194 	 */
195 	origcred = td->td_ucred;
196 
197 	/*
198 	 * Use the credential in nr_cred, if not NULL.
199 	 */
200 	if (nrp->nr_cred != NULL)
201 		td->td_ucred = nrp->nr_cred;
202 	else
203 		td->td_ucred = cred;
204 	saddr = nrp->nr_nam;
205 
206 	if (saddr->sa_family == AF_INET)
207 		if (nrp->nr_sotype == SOCK_DGRAM)
208 			nconf = getnetconfigent("udp");
209 		else
210 			nconf = getnetconfigent("tcp");
211 	else
212 		if (nrp->nr_sotype == SOCK_DGRAM)
213 			nconf = getnetconfigent("udp6");
214 		else
215 			nconf = getnetconfigent("tcp6");
216 
217 	pktscale = nfs_bufpackets;
218 	if (pktscale < 2)
219 		pktscale = 2;
220 	if (pktscale > 64)
221 		pktscale = 64;
222 	pktscalesav = pktscale;
223 	/*
224 	 * soreserve() can fail if sb_max is too small, so shrink pktscale
225 	 * and try again if there is an error.
226 	 * Print a log message suggesting increasing sb_max.
227 	 * Creating a socket and doing this is necessary since, if the
228 	 * reservation sizes are too large and will make soreserve() fail,
229 	 * the connection will work until a large send is attempted and
230 	 * then it will loop in the krpc code.
231 	 */
232 	so = NULL;
233 	saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
234 	error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
235 	    nrp->nr_soproto, td->td_ucred, td);
236 	if (error != 0)
237 		goto out;
238 	do {
239 	    if (error != 0 && pktscale > 2) {
240 		if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
241 		    pktscale == pktscalesav)
242 		    printf("Consider increasing kern.ipc.maxsockbuf\n");
243 		pktscale--;
244 	    }
245 	    if (nrp->nr_sotype == SOCK_DGRAM) {
246 		if (nmp != NULL) {
247 			sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
248 			    pktscale;
249 			rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
250 			    pktscale;
251 		} else {
252 			sndreserve = rcvreserve = 1024 * pktscale;
253 		}
254 	    } else {
255 		if (nrp->nr_sotype != SOCK_STREAM)
256 			panic("nfscon sotype");
257 		if (nmp != NULL) {
258 			sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
259 			    sizeof (u_int32_t)) * pktscale;
260 			rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
261 			    sizeof (u_int32_t)) * pktscale;
262 		} else {
263 			sndreserve = rcvreserve = 1024 * pktscale;
264 		}
265 	    }
266 	    error = soreserve(so, sndreserve, rcvreserve);
267 	    if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
268 		pktscale <= 2)
269 		printf("Must increase kern.ipc.maxsockbuf or reduce"
270 		    " rsize, wsize\n");
271 	} while (error != 0 && pktscale > 2);
272 	soclose(so);
273 	if (error != 0)
274 		goto out;
275 
276 	client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
277 	    nrp->nr_vers, sndreserve, rcvreserve);
278 	CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
279 	if (nmp != NULL) {
280 		if ((nmp->nm_flag & NFSMNT_INT))
281 			CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
282 		if ((nmp->nm_flag & NFSMNT_RESVPORT))
283 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
284 		if (NFSHASTLS(nmp)) {
285 			CLNT_CONTROL(client, CLSET_TLS, &one);
286 			if (nmp->nm_tlscertname != NULL)
287 				CLNT_CONTROL(client, CLSET_TLSCERTNAME,
288 				    nmp->nm_tlscertname);
289 		}
290 		if (NFSHASSOFT(nmp)) {
291 			if (nmp->nm_sotype == SOCK_DGRAM)
292 				/*
293 				 * For UDP, the large timeout for a reconnect
294 				 * will be set to "nm_retry * nm_timeo / 2", so
295 				 * we only want to do 2 reconnect timeout
296 				 * retries.
297 				 */
298 				retries = 2;
299 			else
300 				retries = nmp->nm_retry;
301 		} else
302 			retries = INT_MAX;
303 		if (NFSHASNFSV4N(nmp)) {
304 			if (cred != NULL) {
305 				if (NFSHASSOFT(nmp)) {
306 					/*
307 					 * This should be a DS mount.
308 					 * Use CLSET_TIMEOUT to set the timeout
309 					 * for connections to DSs instead of
310 					 * specifying a timeout on each RPC.
311 					 * This is done so that SO_SNDTIMEO
312 					 * is set on the TCP socket as well
313 					 * as specifying a time limit when
314 					 * waiting for an RPC reply.  Useful
315 					 * if the send queue for the TCP
316 					 * connection has become constipated,
317 					 * due to a failed DS.
318 					 * The choice of lease_duration / 4 is
319 					 * fairly arbitrary, but seems to work
320 					 * ok, with a lower bound of 10sec.
321 					 */
322 					timo.tv_sec = nfsrv_lease / 4;
323 					if (timo.tv_sec < 10)
324 						timo.tv_sec = 10;
325 					timo.tv_usec = 0;
326 					CLNT_CONTROL(client, CLSET_TIMEOUT,
327 					    &timo);
328 				}
329 				/*
330 				 * Make sure the nfscbd_pool doesn't get
331 				 * destroyed while doing this.
332 				 */
333 				NFSD_LOCK();
334 				if (nfs_numnfscbd > 0) {
335 					nfs_numnfscbd++;
336 					NFSD_UNLOCK();
337 					xprt = svc_vc_create_backchannel(
338 					    nfscbd_pool);
339 					CLNT_CONTROL(client, CLSET_BACKCHANNEL,
340 					    xprt);
341 					NFSD_LOCK();
342 					nfs_numnfscbd--;
343 					if (nfs_numnfscbd == 0)
344 						wakeup(&nfs_numnfscbd);
345 				}
346 				NFSD_UNLOCK();
347 			} else {
348 				/*
349 				 * cred == NULL for a DS connect.
350 				 * For connects to a DS, set a retry limit
351 				 * so that failed DSs will be detected.
352 				 * This is ok for NFSv4.1, since a DS does
353 				 * not maintain open/lock state and is the
354 				 * only case where using a "soft" mount is
355 				 * recommended for NFSv4.
356 				 * For mounts from the MDS to DS, this is done
357 				 * via mount options, but that is not the case
358 				 * here.  The retry limit here can be adjusted
359 				 * via the sysctl vfs.nfs.dsretries.
360 				 * See the comment above w.r.t. timeout.
361 				 */
362 				timo.tv_sec = nfsrv_lease / 4;
363 				if (timo.tv_sec < 10)
364 					timo.tv_sec = 10;
365 				timo.tv_usec = 0;
366 				CLNT_CONTROL(client, CLSET_TIMEOUT, &timo);
367 				retries = nfs_dsretries;
368 			}
369 		}
370 	} else {
371 		/*
372 		 * Three cases:
373 		 * - Null RPC callback to client
374 		 * - Non-Null RPC callback to client, wait a little longer
375 		 * - upcalls to nfsuserd and gssd (clp == NULL)
376 		 */
377 		if (callback_retry_mult == 0) {
378 			retries = NFSV4_UPCALLRETRY;
379 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
380 		} else {
381 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
382 		}
383 		if (dotls)
384 			CLNT_CONTROL(client, CLSET_TLS, &one);
385 	}
386 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
387 
388 	if (nmp != NULL) {
389 		/*
390 		 * For UDP, there are 2 timeouts:
391 		 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
392 		 *   that does a retransmit of an RPC request using the same
393 		 *   socket and xid. This is what you normally want to do,
394 		 *   since NFS servers depend on "same xid" for their
395 		 *   Duplicate Request Cache.
396 		 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
397 		 *   retransmits on the same socket should fail and a fresh
398 		 *   socket created. Each of these timeouts counts as one
399 		 *   CLSET_RETRIES as set above.
400 		 * Set the initial retransmit timeout for UDP. This timeout
401 		 * doesn't exist for TCP and the following call just fails,
402 		 * which is ok.
403 		 */
404 		timo.tv_sec = nmp->nm_timeo / NFS_HZ;
405 		timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
406 		CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
407 	}
408 
409 	mtx_lock(&nrp->nr_mtx);
410 	if (nrp->nr_client != NULL) {
411 		mtx_unlock(&nrp->nr_mtx);
412 		/*
413 		 * Someone else already connected.
414 		 */
415 		CLNT_RELEASE(client);
416 	} else {
417 		nrp->nr_client = client;
418 		/*
419 		 * Protocols that do not require connections may be optionally
420 		 * left unconnected for servers that reply from a port other
421 		 * than NFS_PORT.
422 		 */
423 		if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
424 			mtx_unlock(&nrp->nr_mtx);
425 			CLNT_CONTROL(client, CLSET_CONNECT, &one);
426 		} else
427 			mtx_unlock(&nrp->nr_mtx);
428 	}
429 
430 out:
431 	/* Restore current thread's credentials. */
432 	td->td_ucred = origcred;
433 
434 	NFSEXITCODE(error);
435 	return (error);
436 }
437 
438 /*
439  * NFS disconnect. Clean up and unlink.
440  */
441 void
442 newnfs_disconnect(struct nfssockreq *nrp)
443 {
444 	CLIENT *client;
445 
446 	mtx_lock(&nrp->nr_mtx);
447 	if (nrp->nr_client != NULL) {
448 		client = nrp->nr_client;
449 		nrp->nr_client = NULL;
450 		mtx_unlock(&nrp->nr_mtx);
451 		rpc_gss_secpurge_call(client);
452 		CLNT_CLOSE(client);
453 		CLNT_RELEASE(client);
454 	} else {
455 		mtx_unlock(&nrp->nr_mtx);
456 	}
457 }
458 
459 static AUTH *
460 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
461     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
462 {
463 	rpc_gss_service_t svc;
464 	AUTH *auth;
465 
466 	switch (secflavour) {
467 	case RPCSEC_GSS_KRB5:
468 	case RPCSEC_GSS_KRB5I:
469 	case RPCSEC_GSS_KRB5P:
470 		if (!mech_oid) {
471 			if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
472 				return (NULL);
473 		}
474 		if (secflavour == RPCSEC_GSS_KRB5)
475 			svc = rpc_gss_svc_none;
476 		else if (secflavour == RPCSEC_GSS_KRB5I)
477 			svc = rpc_gss_svc_integrity;
478 		else
479 			svc = rpc_gss_svc_privacy;
480 
481 		if (clnt_principal == NULL)
482 			auth = rpc_gss_secfind_call(nrp->nr_client, cred,
483 			    srv_principal, mech_oid, svc);
484 		else {
485 			auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
486 			    clnt_principal, srv_principal, "kerberosv5",
487 			    svc, NULL, NULL, NULL);
488 			return (auth);
489 		}
490 		if (auth != NULL)
491 			return (auth);
492 		/* fallthrough */
493 	case AUTH_SYS:
494 	default:
495 		return (authunix_create(cred));
496 	}
497 }
498 
499 /*
500  * Callback from the RPC code to generate up/down notifications.
501  */
502 
503 struct nfs_feedback_arg {
504 	struct nfsmount *nf_mount;
505 	int		nf_lastmsg;	/* last tprintf */
506 	int		nf_tprintfmsg;
507 	struct thread	*nf_td;
508 };
509 
510 static void
511 nfs_feedback(int type, int proc, void *arg)
512 {
513 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
514 	struct nfsmount *nmp = nf->nf_mount;
515 	time_t now;
516 
517 	switch (type) {
518 	case FEEDBACK_REXMIT2:
519 	case FEEDBACK_RECONNECT:
520 		now = NFSD_MONOSEC;
521 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
522 			nfs_down(nmp, nf->nf_td,
523 			    "not responding", 0, NFSSTA_TIMEO);
524 			nf->nf_tprintfmsg = TRUE;
525 			nf->nf_lastmsg = now;
526 		}
527 		break;
528 
529 	case FEEDBACK_OK:
530 		nfs_up(nf->nf_mount, nf->nf_td,
531 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
532 		break;
533 	}
534 }
535 
536 /*
537  * newnfs_request - goes something like this
538  *	- does the rpc by calling the krpc layer
539  *	- break down rpc header and return with nfs reply
540  * nb: always frees up nd_mreq mbuf list
541  */
542 int
543 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
544     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
545     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
546     u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
547 {
548 	uint32_t retseq, retval, slotseq, *tl;
549 	time_t waituntil;
550 	int i = 0, j = 0, opcnt, set_sigset = 0, slot;
551 	int error = 0, usegssname = 0, secflavour = AUTH_SYS;
552 	int freeslot, maxslot, reterr, slotpos, timeo;
553 	u_int16_t procnum;
554 	u_int trylater_delay = 1;
555 	struct nfs_feedback_arg nf;
556 	struct timeval timo;
557 	AUTH *auth;
558 	struct rpc_callextra ext;
559 	enum clnt_stat stat;
560 	struct nfsreq *rep = NULL;
561 	char *srv_principal = NULL, *clnt_principal = NULL;
562 	sigset_t oldset;
563 	struct ucred *authcred;
564 	struct nfsclsession *sep;
565 	uint8_t sessionid[NFSX_V4SESSIONID];
566 
567 	sep = dssep;
568 	if (xidp != NULL)
569 		*xidp = 0;
570 	/* Reject requests while attempting a forced unmount. */
571 	if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
572 		m_freem(nd->nd_mreq);
573 		return (ESTALE);
574 	}
575 
576 	/*
577 	 * Set authcred, which is used to acquire RPC credentials to
578 	 * the cred argument, by default. The crhold() should not be
579 	 * necessary, but will ensure that some future code change
580 	 * doesn't result in the credential being free'd prematurely.
581 	 */
582 	authcred = crhold(cred);
583 
584 	/* For client side interruptible mounts, mask off the signals. */
585 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
586 		newnfs_set_sigmask(td, &oldset);
587 		set_sigset = 1;
588 	}
589 
590 	/*
591 	 * XXX if not already connected call nfs_connect now. Longer
592 	 * term, change nfs_mount to call nfs_connect unconditionally
593 	 * and let clnt_reconnect_create handle reconnects.
594 	 */
595 	if (nrp->nr_client == NULL)
596 		newnfs_connect(nmp, nrp, cred, td, 0, false);
597 
598 	/*
599 	 * For a client side mount, nmp is != NULL and clp == NULL. For
600 	 * server calls (callbacks or upcalls), nmp == NULL.
601 	 */
602 	if (clp != NULL) {
603 		NFSLOCKSTATE();
604 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
605 			secflavour = RPCSEC_GSS_KRB5;
606 			if (nd->nd_procnum != NFSPROC_NULL) {
607 				if (clp->lc_flags & LCL_GSSINTEGRITY)
608 					secflavour = RPCSEC_GSS_KRB5I;
609 				else if (clp->lc_flags & LCL_GSSPRIVACY)
610 					secflavour = RPCSEC_GSS_KRB5P;
611 			}
612 		}
613 		NFSUNLOCKSTATE();
614 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
615 	     nd->nd_procnum != NFSPROC_NULL) {
616 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
617 			nd->nd_flag |= ND_USEGSSNAME;
618 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
619 			/*
620 			 * If there is a client side host based credential,
621 			 * use that, otherwise use the system uid, if set.
622 			 * The system uid is in the nmp->nm_sockreq.nr_cred
623 			 * credentials.
624 			 */
625 			if (nmp->nm_krbnamelen > 0) {
626 				usegssname = 1;
627 				clnt_principal = nmp->nm_krbname;
628 			} else if (nmp->nm_uid != (uid_t)-1) {
629 				KASSERT(nmp->nm_sockreq.nr_cred != NULL,
630 				    ("newnfs_request: NULL nr_cred"));
631 				crfree(authcred);
632 				authcred = crhold(nmp->nm_sockreq.nr_cred);
633 			}
634 		} else if (nmp->nm_krbnamelen == 0 &&
635 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
636 			/*
637 			 * If there is no host based principal name and
638 			 * the system uid is set and this is root, use the
639 			 * system uid, since root won't have user
640 			 * credentials in a credentials cache file.
641 			 * The system uid is in the nmp->nm_sockreq.nr_cred
642 			 * credentials.
643 			 */
644 			KASSERT(nmp->nm_sockreq.nr_cred != NULL,
645 			    ("newnfs_request: NULL nr_cred"));
646 			crfree(authcred);
647 			authcred = crhold(nmp->nm_sockreq.nr_cred);
648 		}
649 		if (NFSHASINTEGRITY(nmp))
650 			secflavour = RPCSEC_GSS_KRB5I;
651 		else if (NFSHASPRIVACY(nmp))
652 			secflavour = RPCSEC_GSS_KRB5P;
653 		else
654 			secflavour = RPCSEC_GSS_KRB5;
655 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
656 	} else if (nmp != NULL && !NFSHASKERB(nmp) &&
657 	    nd->nd_procnum != NFSPROC_NULL &&
658 	    (nd->nd_flag & ND_USEGSSNAME) != 0) {
659 		/*
660 		 * Use the uid that did the mount when the RPC is doing
661 		 * NFSv4 system operations, as indicated by the
662 		 * ND_USEGSSNAME flag, for the AUTH_SYS case.
663 		 * The credentials in nm_sockreq.nr_cred were used for the
664 		 * mount.
665 		 */
666 		KASSERT(nmp->nm_sockreq.nr_cred != NULL,
667 		    ("newnfs_request: NULL nr_cred"));
668 		crfree(authcred);
669 		authcred = crhold(nmp->nm_sockreq.nr_cred);
670 	}
671 
672 	if (nmp != NULL) {
673 		bzero(&nf, sizeof(struct nfs_feedback_arg));
674 		nf.nf_mount = nmp;
675 		nf.nf_td = td;
676 		nf.nf_lastmsg = NFSD_MONOSEC -
677 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
678 	}
679 
680 	if (nd->nd_procnum == NFSPROC_NULL)
681 		auth = authnone_create();
682 	else if (usegssname) {
683 		/*
684 		 * For this case, the authenticator is held in the
685 		 * nfssockreq structure, so don't release the reference count
686 		 * held on it. --> Don't AUTH_DESTROY() it in this function.
687 		 */
688 		if (nrp->nr_auth == NULL)
689 			nrp->nr_auth = nfs_getauth(nrp, secflavour,
690 			    clnt_principal, srv_principal, NULL, authcred);
691 		else
692 			rpc_gss_refresh_auth_call(nrp->nr_auth);
693 		auth = nrp->nr_auth;
694 	} else
695 		auth = nfs_getauth(nrp, secflavour, NULL,
696 		    srv_principal, NULL, authcred);
697 	crfree(authcred);
698 	if (auth == NULL) {
699 		m_freem(nd->nd_mreq);
700 		if (set_sigset)
701 			newnfs_restore_sigmask(td, &oldset);
702 		return (EACCES);
703 	}
704 	bzero(&ext, sizeof(ext));
705 	ext.rc_auth = auth;
706 	if (nmp != NULL) {
707 		ext.rc_feedback = nfs_feedback;
708 		ext.rc_feedback_arg = &nf;
709 	}
710 
711 	procnum = nd->nd_procnum;
712 	if ((nd->nd_flag & ND_NFSV4) &&
713 	    nd->nd_procnum != NFSPROC_NULL &&
714 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
715 		procnum = NFSV4PROC_COMPOUND;
716 
717 	if (nmp != NULL) {
718 		NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
719 
720 		/* Map the procnum to the old NFSv2 one, as required. */
721 		if ((nd->nd_flag & ND_NFSV2) != 0) {
722 			if (nd->nd_procnum < NFS_V3NPROCS)
723 				procnum = nfsv2_procid[nd->nd_procnum];
724 			else
725 				procnum = NFSV2PROC_NOOP;
726 		}
727 
728 		/*
729 		 * Now only used for the R_DONTRECOVER case, but until that is
730 		 * supported within the krpc code, I need to keep a queue of
731 		 * outstanding RPCs for nfsv4 client requests.
732 		 */
733 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
734 			rep = malloc(sizeof(struct nfsreq),
735 			    M_NFSDREQ, M_WAITOK);
736 #ifdef KDTRACE_HOOKS
737 		if (dtrace_nfscl_nfs234_start_probe != NULL) {
738 			uint32_t probe_id;
739 			int probe_procnum;
740 
741 			if (nd->nd_flag & ND_NFSV4) {
742 				probe_id =
743 				    nfscl_nfs4_start_probes[nd->nd_procnum];
744 				probe_procnum = nd->nd_procnum;
745 			} else if (nd->nd_flag & ND_NFSV3) {
746 				probe_id = nfscl_nfs3_start_probes[procnum];
747 				probe_procnum = procnum;
748 			} else {
749 				probe_id =
750 				    nfscl_nfs2_start_probes[nd->nd_procnum];
751 				probe_procnum = procnum;
752 			}
753 			if (probe_id != 0)
754 				(dtrace_nfscl_nfs234_start_probe)
755 				    (probe_id, vp, nd->nd_mreq, cred,
756 				     probe_procnum);
757 		}
758 #endif
759 	}
760 	freeslot = -1;		/* Set to slot that needs to be free'd */
761 tryagain:
762 	slot = -1;		/* Slot that needs a sequence# increment. */
763 	/*
764 	 * This timeout specifies when a new socket should be created,
765 	 * along with new xid values. For UDP, this should be done
766 	 * infrequently, since retransmits of RPC requests should normally
767 	 * use the same xid.
768 	 */
769 	if (nmp == NULL) {
770 		timo.tv_usec = 0;
771 		if (clp == NULL)
772 			timo.tv_sec = NFSV4_UPCALLTIMEO;
773 		else
774 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
775 	} else {
776 		if (nrp->nr_sotype != SOCK_DGRAM) {
777 			timo.tv_usec = 0;
778 			if ((nmp->nm_flag & NFSMNT_NFSV4))
779 				timo.tv_sec = INT_MAX;
780 			else
781 				timo.tv_sec = NFS_TCPTIMEO;
782 		} else {
783 			if (NFSHASSOFT(nmp)) {
784 				/*
785 				 * CLSET_RETRIES is set to 2, so this should be
786 				 * half of the total timeout required.
787 				 */
788 				timeo = nmp->nm_retry * nmp->nm_timeo / 2;
789 				if (timeo < 1)
790 					timeo = 1;
791 				timo.tv_sec = timeo / NFS_HZ;
792 				timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
793 				    NFS_HZ;
794 			} else {
795 				/* For UDP hard mounts, use a large value. */
796 				timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
797 				timo.tv_usec = 0;
798 			}
799 		}
800 
801 		if (rep != NULL) {
802 			rep->r_flags = 0;
803 			rep->r_nmp = nmp;
804 			/*
805 			 * Chain request into list of outstanding requests.
806 			 */
807 			NFSLOCKREQ();
808 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
809 			NFSUNLOCKREQ();
810 		}
811 	}
812 
813 	nd->nd_mrep = NULL;
814 	if (clp != NULL && sep != NULL)
815 		stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
816 		    nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
817 	else
818 		stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
819 		    nd->nd_mreq, &nd->nd_mrep, timo);
820 	NFSCL_DEBUG(2, "clnt call=%d\n", stat);
821 
822 	if (rep != NULL) {
823 		/*
824 		 * RPC done, unlink the request.
825 		 */
826 		NFSLOCKREQ();
827 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
828 		NFSUNLOCKREQ();
829 	}
830 
831 	/*
832 	 * If there was a successful reply and a tprintf msg.
833 	 * tprintf a response.
834 	 */
835 	if (stat == RPC_SUCCESS) {
836 		error = 0;
837 	} else if (stat == RPC_TIMEDOUT) {
838 		NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
839 		error = ETIMEDOUT;
840 	} else if (stat == RPC_VERSMISMATCH) {
841 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
842 		error = EOPNOTSUPP;
843 	} else if (stat == RPC_PROGVERSMISMATCH) {
844 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
845 		error = EPROTONOSUPPORT;
846 	} else if (stat == RPC_INTR) {
847 		error = EINTR;
848 	} else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
849 	     stat == RPC_SYSTEMERROR) {
850 		/* Check for a session slot that needs to be free'd. */
851 		if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
852 		    (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
853 		    nd->nd_procnum != NFSPROC_NULL) {
854 			/*
855 			 * This should only occur when either the MDS or
856 			 * a client has an RPC against a DS fail.
857 			 * This happens because these cases use "soft"
858 			 * connections that can time out and fail.
859 			 * The slot used for this RPC is now in a
860 			 * non-deterministic state, but if the slot isn't
861 			 * free'd, threads can get stuck waiting for a slot.
862 			 */
863 			if (sep == NULL)
864 				sep = nfsmnt_mdssession(nmp);
865 			/*
866 			 * Bump the sequence# out of range, so that reuse of
867 			 * this slot will result in an NFSERR_SEQMISORDERED
868 			 * error and not a bogus cached RPC reply.
869 			 */
870 			mtx_lock(&sep->nfsess_mtx);
871 			sep->nfsess_slotseq[nd->nd_slotid] += 10;
872 			mtx_unlock(&sep->nfsess_mtx);
873 			/* And free the slot. */
874 			nfsv4_freeslot(sep, nd->nd_slotid);
875 		}
876 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
877 		error = ENXIO;
878 	} else {
879 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
880 		error = EACCES;
881 	}
882 	if (error) {
883 		m_freem(nd->nd_mreq);
884 		if (usegssname == 0)
885 			AUTH_DESTROY(auth);
886 		if (rep != NULL)
887 			free(rep, M_NFSDREQ);
888 		if (set_sigset)
889 			newnfs_restore_sigmask(td, &oldset);
890 		return (error);
891 	}
892 
893 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
894 
895 	/*
896 	 * Search for any mbufs that are not a multiple of 4 bytes long
897 	 * or with m_data not longword aligned.
898 	 * These could cause pointer alignment problems, so copy them to
899 	 * well aligned mbufs.
900 	 */
901 	newnfs_realign(&nd->nd_mrep, M_WAITOK);
902 	nd->nd_md = nd->nd_mrep;
903 	nd->nd_dpos = mtod(nd->nd_md, caddr_t);
904 	nd->nd_repstat = 0;
905 	if (nd->nd_procnum != NFSPROC_NULL &&
906 	    nd->nd_procnum != NFSV4PROC_CBNULL) {
907 		/* If sep == NULL, set it to the default in nmp. */
908 		if (sep == NULL && nmp != NULL)
909 			sep = nfsmnt_mdssession(nmp);
910 		/*
911 		 * and now the actual NFS xdr.
912 		 */
913 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
914 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
915 		if (nd->nd_repstat >= 10000)
916 			NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
917 			    (int)nd->nd_repstat);
918 
919 		/*
920 		 * Get rid of the tag, return count and SEQUENCE result for
921 		 * NFSv4.
922 		 */
923 		if ((nd->nd_flag & ND_NFSV4) != 0 && nd->nd_repstat !=
924 		    NFSERR_MINORVERMISMATCH) {
925 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
926 			i = fxdr_unsigned(int, *tl);
927 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
928 			if (error)
929 				goto nfsmout;
930 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
931 			opcnt = fxdr_unsigned(int, *tl++);
932 			i = fxdr_unsigned(int, *tl++);
933 			j = fxdr_unsigned(int, *tl);
934 			if (j >= 10000)
935 				NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
936 			/*
937 			 * If the first op is Sequence, free up the slot.
938 			 */
939 			if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
940 			    (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
941 				NFSCL_DEBUG(1, "failed seq=%d\n", j);
942 			if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
943 			    (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
944 			    j == 0)) && sep != NULL) {
945 				if (i == NFSV4OP_SEQUENCE)
946 					NFSM_DISSECT(tl, uint32_t *,
947 					    NFSX_V4SESSIONID +
948 					    5 * NFSX_UNSIGNED);
949 				else
950 					NFSM_DISSECT(tl, uint32_t *,
951 					    NFSX_V4SESSIONID +
952 					    4 * NFSX_UNSIGNED);
953 				mtx_lock(&sep->nfsess_mtx);
954 				if (bcmp(tl, sep->nfsess_sessionid,
955 				    NFSX_V4SESSIONID) == 0) {
956 					tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
957 					retseq = fxdr_unsigned(uint32_t, *tl++);
958 					slot = fxdr_unsigned(int, *tl++);
959 					freeslot = slot;
960 					if (retseq != sep->nfsess_slotseq[slot])
961 						printf("retseq diff 0x%x\n",
962 						    retseq);
963 					retval = fxdr_unsigned(uint32_t, *++tl);
964 					if ((retval + 1) < sep->nfsess_foreslots
965 					    )
966 						sep->nfsess_foreslots = (retval
967 						    + 1);
968 					else if ((retval + 1) >
969 					    sep->nfsess_foreslots)
970 						sep->nfsess_foreslots = (retval
971 						    < 64) ? (retval + 1) : 64;
972 				}
973 				mtx_unlock(&sep->nfsess_mtx);
974 
975 				/* Grab the op and status for the next one. */
976 				if (opcnt > 1) {
977 					NFSM_DISSECT(tl, uint32_t *,
978 					    2 * NFSX_UNSIGNED);
979 					i = fxdr_unsigned(int, *tl++);
980 					j = fxdr_unsigned(int, *tl);
981 				}
982 			}
983 		}
984 		if (nd->nd_repstat != 0) {
985 			if (nd->nd_repstat == NFSERR_BADSESSION &&
986 			    nmp != NULL && dssep == NULL &&
987 			    (nd->nd_flag & ND_NFSV41) != 0) {
988 				/*
989 				 * If this is a client side MDS RPC, mark
990 				 * the MDS session defunct and initiate
991 				 * recovery, as required.
992 				 * The nfsess_defunct field is protected by
993 				 * the NFSLOCKMNT()/nm_mtx lock and not the
994 				 * nfsess_mtx lock to simplify its handling,
995 				 * for the MDS session. This lock is also
996 				 * sufficient for nfsess_sessionid, since it
997 				 * never changes in the structure.
998 				 */
999 				NFSCL_DEBUG(1, "Got badsession\n");
1000 				NFSLOCKCLSTATE();
1001 				NFSLOCKMNT(nmp);
1002 				sep = NFSMNT_MDSSESSION(nmp);
1003 				if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
1004 				    NFSX_V4SESSIONID) == 0) {
1005 					/* Initiate recovery. */
1006 					sep->nfsess_defunct = 1;
1007 					NFSCL_DEBUG(1, "Marked defunct\n");
1008 					if (nmp->nm_clp != NULL) {
1009 						nmp->nm_clp->nfsc_flags |=
1010 						    NFSCLFLAGS_RECOVER;
1011 						wakeup(nmp->nm_clp);
1012 					}
1013 				}
1014 				NFSUNLOCKCLSTATE();
1015 				/*
1016 				 * Sleep for up to 1sec waiting for a new
1017 				 * session.
1018 				 */
1019 				mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1020 				    "nfsbadsess", hz);
1021 				/*
1022 				 * Get the session again, in case a new one
1023 				 * has been created during the sleep.
1024 				 */
1025 				sep = NFSMNT_MDSSESSION(nmp);
1026 				NFSUNLOCKMNT(nmp);
1027 				if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1028 					reterr = nfsv4_sequencelookup(nmp, sep,
1029 					    &slotpos, &maxslot, &slotseq,
1030 					    sessionid);
1031 					if (reterr == 0) {
1032 						/* Fill in new session info. */
1033 						NFSCL_DEBUG(1,
1034 						  "Filling in new sequence\n");
1035 						tl = nd->nd_sequence;
1036 						bcopy(sessionid, tl,
1037 						    NFSX_V4SESSIONID);
1038 						tl += NFSX_V4SESSIONID /
1039 						    NFSX_UNSIGNED;
1040 						*tl++ = txdr_unsigned(slotseq);
1041 						*tl++ = txdr_unsigned(slotpos);
1042 						*tl = txdr_unsigned(maxslot);
1043 					}
1044 					if (reterr == NFSERR_BADSESSION ||
1045 					    reterr == 0) {
1046 						NFSCL_DEBUG(1,
1047 						    "Badsession looping\n");
1048 						m_freem(nd->nd_mrep);
1049 						nd->nd_mrep = NULL;
1050 						goto tryagain;
1051 					}
1052 					nd->nd_repstat = reterr;
1053 					NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1054 				}
1055 			}
1056 			/*
1057 			 * When clp != NULL, it is a callback and all
1058 			 * callback operations can be retried for NFSERR_DELAY.
1059 			 */
1060 			if (((nd->nd_repstat == NFSERR_DELAY ||
1061 			      nd->nd_repstat == NFSERR_GRACE) &&
1062 			     (nd->nd_flag & ND_NFSV4) && (clp != NULL ||
1063 			     (nd->nd_procnum != NFSPROC_DELEGRETURN &&
1064 			     nd->nd_procnum != NFSPROC_SETATTR &&
1065 			     nd->nd_procnum != NFSPROC_READ &&
1066 			     nd->nd_procnum != NFSPROC_READDS &&
1067 			     nd->nd_procnum != NFSPROC_WRITE &&
1068 			     nd->nd_procnum != NFSPROC_WRITEDS &&
1069 			     nd->nd_procnum != NFSPROC_OPEN &&
1070 			     nd->nd_procnum != NFSPROC_CREATE &&
1071 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1072 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1073 			     nd->nd_procnum != NFSPROC_CLOSE &&
1074 			     nd->nd_procnum != NFSPROC_LOCK &&
1075 			     nd->nd_procnum != NFSPROC_LOCKU))) ||
1076 			    (nd->nd_repstat == NFSERR_DELAY &&
1077 			     (nd->nd_flag & ND_NFSV4) == 0) ||
1078 			    nd->nd_repstat == NFSERR_RESOURCE) {
1079 				if (trylater_delay > NFS_TRYLATERDEL)
1080 					trylater_delay = NFS_TRYLATERDEL;
1081 				waituntil = NFSD_MONOSEC + trylater_delay;
1082 				while (NFSD_MONOSEC < waituntil)
1083 					(void) nfs_catnap(PZERO, 0, "nfstry");
1084 				trylater_delay *= 2;
1085 				if (slot != -1) {
1086 					mtx_lock(&sep->nfsess_mtx);
1087 					sep->nfsess_slotseq[slot]++;
1088 					*nd->nd_slotseq = txdr_unsigned(
1089 					    sep->nfsess_slotseq[slot]);
1090 					mtx_unlock(&sep->nfsess_mtx);
1091 				}
1092 				m_freem(nd->nd_mrep);
1093 				nd->nd_mrep = NULL;
1094 				goto tryagain;
1095 			}
1096 
1097 			/*
1098 			 * If the File Handle was stale, invalidate the
1099 			 * lookup cache, just in case.
1100 			 * (vp != NULL implies a client side call)
1101 			 */
1102 			if (nd->nd_repstat == ESTALE && vp != NULL) {
1103 				cache_purge(vp);
1104 				if (ncl_call_invalcaches != NULL)
1105 					(*ncl_call_invalcaches)(vp);
1106 			}
1107 		}
1108 		if ((nd->nd_flag & ND_NFSV4) != 0) {
1109 			/* Free the slot, as required. */
1110 			if (freeslot != -1)
1111 				nfsv4_freeslot(sep, freeslot);
1112 			/*
1113 			 * If this op is Putfh, throw its results away.
1114 			 */
1115 			if (j >= 10000)
1116 				NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1117 			if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1118 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1119 				i = fxdr_unsigned(int, *tl++);
1120 				j = fxdr_unsigned(int, *tl);
1121 				if (j >= 10000)
1122 					NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1123 					    j);
1124 				/*
1125 				 * All Compounds that do an Op that must
1126 				 * be in sequence consist of NFSV4OP_PUTFH
1127 				 * followed by one of these. As such, we
1128 				 * can determine if the seqid# should be
1129 				 * incremented, here.
1130 				 */
1131 				if ((i == NFSV4OP_OPEN ||
1132 				     i == NFSV4OP_OPENCONFIRM ||
1133 				     i == NFSV4OP_OPENDOWNGRADE ||
1134 				     i == NFSV4OP_CLOSE ||
1135 				     i == NFSV4OP_LOCK ||
1136 				     i == NFSV4OP_LOCKU) &&
1137 				    (j == 0 ||
1138 				     (j != NFSERR_STALECLIENTID &&
1139 				      j != NFSERR_STALESTATEID &&
1140 				      j != NFSERR_BADSTATEID &&
1141 				      j != NFSERR_BADSEQID &&
1142 				      j != NFSERR_BADXDR &&
1143 				      j != NFSERR_RESOURCE &&
1144 				      j != NFSERR_NOFILEHANDLE)))
1145 					nd->nd_flag |= ND_INCRSEQID;
1146 			}
1147 			/*
1148 			 * If this op's status is non-zero, mark
1149 			 * that there is no more data to process.
1150 			 * The exception is Setattr, which always has xdr
1151 			 * when it has failed.
1152 			 */
1153 			if (j != 0 && i != NFSV4OP_SETATTR)
1154 				nd->nd_flag |= ND_NOMOREDATA;
1155 
1156 			/*
1157 			 * If R_DONTRECOVER is set, replace the stale error
1158 			 * reply, so that recovery isn't initiated.
1159 			 */
1160 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1161 			     nd->nd_repstat == NFSERR_BADSESSION ||
1162 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
1163 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
1164 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1165 		}
1166 	}
1167 
1168 #ifdef KDTRACE_HOOKS
1169 	if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1170 		uint32_t probe_id;
1171 		int probe_procnum;
1172 
1173 		if (nd->nd_flag & ND_NFSV4) {
1174 			probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1175 			probe_procnum = nd->nd_procnum;
1176 		} else if (nd->nd_flag & ND_NFSV3) {
1177 			probe_id = nfscl_nfs3_done_probes[procnum];
1178 			probe_procnum = procnum;
1179 		} else {
1180 			probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1181 			probe_procnum = procnum;
1182 		}
1183 		if (probe_id != 0)
1184 			(dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1185 			    nd->nd_mreq, cred, probe_procnum, 0);
1186 	}
1187 #endif
1188 
1189 	m_freem(nd->nd_mreq);
1190 	if (usegssname == 0)
1191 		AUTH_DESTROY(auth);
1192 	if (rep != NULL)
1193 		free(rep, M_NFSDREQ);
1194 	if (set_sigset)
1195 		newnfs_restore_sigmask(td, &oldset);
1196 	return (0);
1197 nfsmout:
1198 	m_freem(nd->nd_mrep);
1199 	m_freem(nd->nd_mreq);
1200 	if (usegssname == 0)
1201 		AUTH_DESTROY(auth);
1202 	if (rep != NULL)
1203 		free(rep, M_NFSDREQ);
1204 	if (set_sigset)
1205 		newnfs_restore_sigmask(td, &oldset);
1206 	return (error);
1207 }
1208 
1209 /*
1210  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1211  * wait for all requests to complete. This is used by forced unmounts
1212  * to terminate any outstanding RPCs.
1213  */
1214 int
1215 newnfs_nmcancelreqs(struct nfsmount *nmp)
1216 {
1217 	struct nfsclds *dsp;
1218 	struct __rpc_client *cl;
1219 
1220 	if (nmp->nm_sockreq.nr_client != NULL)
1221 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1222 lookformore:
1223 	NFSLOCKMNT(nmp);
1224 	TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1225 		NFSLOCKDS(dsp);
1226 		if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1227 		    (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1228 		    dsp->nfsclds_sockp != NULL &&
1229 		    dsp->nfsclds_sockp->nr_client != NULL) {
1230 			dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1231 			cl = dsp->nfsclds_sockp->nr_client;
1232 			NFSUNLOCKDS(dsp);
1233 			NFSUNLOCKMNT(nmp);
1234 			CLNT_CLOSE(cl);
1235 			goto lookformore;
1236 		}
1237 		NFSUNLOCKDS(dsp);
1238 	}
1239 	NFSUNLOCKMNT(nmp);
1240 	return (0);
1241 }
1242 
1243 /*
1244  * Any signal that can interrupt an NFS operation in an intr mount
1245  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1246  */
1247 int newnfs_sig_set[] = {
1248 	SIGINT,
1249 	SIGTERM,
1250 	SIGHUP,
1251 	SIGKILL,
1252 	SIGQUIT
1253 };
1254 
1255 /*
1256  * Check to see if one of the signals in our subset is pending on
1257  * the process (in an intr mount).
1258  */
1259 static int
1260 nfs_sig_pending(sigset_t set)
1261 {
1262 	int i;
1263 
1264 	for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1265 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
1266 			return (1);
1267 	return (0);
1268 }
1269 
1270 /*
1271  * The set/restore sigmask functions are used to (temporarily) overwrite
1272  * the thread td_sigmask during an RPC call (for example). These are also
1273  * used in other places in the NFS client that might tsleep().
1274  */
1275 void
1276 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1277 {
1278 	sigset_t newset;
1279 	int i;
1280 	struct proc *p;
1281 
1282 	SIGFILLSET(newset);
1283 	if (td == NULL)
1284 		td = curthread; /* XXX */
1285 	p = td->td_proc;
1286 	/* Remove the NFS set of signals from newset */
1287 	PROC_LOCK(p);
1288 	mtx_lock(&p->p_sigacts->ps_mtx);
1289 	for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1290 		/*
1291 		 * But make sure we leave the ones already masked
1292 		 * by the process, ie. remove the signal from the
1293 		 * temporary signalmask only if it wasn't already
1294 		 * in p_sigmask.
1295 		 */
1296 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1297 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1298 			SIGDELSET(newset, newnfs_sig_set[i]);
1299 	}
1300 	mtx_unlock(&p->p_sigacts->ps_mtx);
1301 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1302 	    SIGPROCMASK_PROC_LOCKED);
1303 	PROC_UNLOCK(p);
1304 }
1305 
1306 void
1307 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1308 {
1309 	if (td == NULL)
1310 		td = curthread; /* XXX */
1311 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1312 }
1313 
1314 /*
1315  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1316  * old one after msleep() returns.
1317  */
1318 int
1319 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1320 {
1321 	sigset_t oldset;
1322 	int error;
1323 
1324 	if ((priority & PCATCH) == 0)
1325 		return msleep(ident, mtx, priority, wmesg, timo);
1326 	if (td == NULL)
1327 		td = curthread; /* XXX */
1328 	newnfs_set_sigmask(td, &oldset);
1329 	error = msleep(ident, mtx, priority, wmesg, timo);
1330 	newnfs_restore_sigmask(td, &oldset);
1331 	return (error);
1332 }
1333 
1334 /*
1335  * Test for a termination condition pending on the process.
1336  * This is used for NFSMNT_INT mounts.
1337  */
1338 int
1339 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1340 {
1341 	struct proc *p;
1342 	sigset_t tmpset;
1343 
1344 	/* Terminate all requests while attempting a forced unmount. */
1345 	if (NFSCL_FORCEDISM(nmp->nm_mountp))
1346 		return (EIO);
1347 	if (!(nmp->nm_flag & NFSMNT_INT))
1348 		return (0);
1349 	if (td == NULL)
1350 		return (0);
1351 	p = td->td_proc;
1352 	PROC_LOCK(p);
1353 	tmpset = p->p_siglist;
1354 	SIGSETOR(tmpset, td->td_siglist);
1355 	SIGSETNAND(tmpset, td->td_sigmask);
1356 	mtx_lock(&p->p_sigacts->ps_mtx);
1357 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1358 	mtx_unlock(&p->p_sigacts->ps_mtx);
1359 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1360 	    && nfs_sig_pending(tmpset)) {
1361 		PROC_UNLOCK(p);
1362 		return (EINTR);
1363 	}
1364 	PROC_UNLOCK(p);
1365 	return (0);
1366 }
1367 
1368 static int
1369 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1370 {
1371 	struct proc *p;
1372 
1373 	p = td ? td->td_proc : NULL;
1374 	if (error) {
1375 		tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1376 		    server, msg, error);
1377 	} else {
1378 		tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1379 	}
1380 	return (0);
1381 }
1382 
1383 static void
1384 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1385     int error, int flags)
1386 {
1387 	if (nmp == NULL)
1388 		return;
1389 	mtx_lock(&nmp->nm_mtx);
1390 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1391 		nmp->nm_state |= NFSSTA_TIMEO;
1392 		mtx_unlock(&nmp->nm_mtx);
1393 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1394 		    VQ_NOTRESP, 0);
1395 	} else
1396 		mtx_unlock(&nmp->nm_mtx);
1397 	mtx_lock(&nmp->nm_mtx);
1398 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1399 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
1400 		mtx_unlock(&nmp->nm_mtx);
1401 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1402 		    VQ_NOTRESPLOCK, 0);
1403 	} else
1404 		mtx_unlock(&nmp->nm_mtx);
1405 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1406 }
1407 
1408 static void
1409 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1410     int flags, int tprintfmsg)
1411 {
1412 	if (nmp == NULL)
1413 		return;
1414 	if (tprintfmsg) {
1415 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1416 	}
1417 
1418 	mtx_lock(&nmp->nm_mtx);
1419 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1420 		nmp->nm_state &= ~NFSSTA_TIMEO;
1421 		mtx_unlock(&nmp->nm_mtx);
1422 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1423 		    VQ_NOTRESP, 1);
1424 	} else
1425 		mtx_unlock(&nmp->nm_mtx);
1426 
1427 	mtx_lock(&nmp->nm_mtx);
1428 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1429 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1430 		mtx_unlock(&nmp->nm_mtx);
1431 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1432 		    VQ_NOTRESPLOCK, 1);
1433 	} else
1434 		mtx_unlock(&nmp->nm_mtx);
1435 }
1436