xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision 86ce6a83d16fdedede88990ffe1d85ad83d8e7c8)
1 /*-
2  * Copyright (c) 1989, 1991, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
19  *
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
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 /*
38  * Socket operations for use by nfs
39  */
40 
41 #include "opt_inet6.h"
42 #include "opt_kgssapi.h"
43 #include "opt_nfs.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/limits.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.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>
60 
61 #include <rpc/rpc.h>
62 
63 #include <kgssapi/krb5/kcrypto.h>
64 
65 #include <fs/nfs/nfsport.h>
66 
67 NFSSTATESPINLOCK;
68 NFSREQSPINLOCK;
69 extern struct nfsstats newnfsstats;
70 extern struct nfsreqhead nfsd_reqq;
71 extern int nfscl_ticks;
72 extern void (*ncl_call_invalcaches)(struct vnode *);
73 
74 static int	nfsrv_gsscallbackson = 0;
75 static int	nfs_bufpackets = 4;
76 static int	nfs_reconnects;
77 static int	nfs3_jukebox_delay = 10;
78 static int	nfs_skip_wcc_data_onerr = 1;
79 static int	nfs_keytab_enctype = ETYPE_DES_CBC_CRC;
80 
81 SYSCTL_DECL(_vfs_newnfs);
82 
83 SYSCTL_INT(_vfs_newnfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
84     "Buffer reservation size 2 < x < 64");
85 SYSCTL_INT(_vfs_newnfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
86     "Number of times the nfs client has had to reconnect");
87 SYSCTL_INT(_vfs_newnfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
88     "Number of seconds to delay a retry after receiving EJUKEBOX");
89 SYSCTL_INT(_vfs_newnfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
90     "Disable weak cache consistency checking when server returns an error");
91 SYSCTL_INT(_vfs_newnfs, OID_AUTO, keytab_enctype, CTLFLAG_RW, &nfs_keytab_enctype, 0,
92     "Encryption type for the keytab entry used by nfs");
93 
94 static void	nfs_down(struct nfsmount *, struct thread *, const char *,
95     int, int);
96 static void	nfs_up(struct nfsmount *, struct thread *, const char *,
97     int, int);
98 static int	nfs_msg(struct thread *, const char *, const char *, int);
99 
100 extern int nfsv2_procid[];
101 
102 struct nfs_cached_auth {
103 	int		ca_refs; /* refcount, including 1 from the cache */
104 	uid_t		ca_uid;	 /* uid that corresponds to this auth */
105 	AUTH		*ca_auth; /* RPC auth handle */
106 };
107 
108 /*
109  * Initialize sockets and congestion for a new NFS connection.
110  * We do not free the sockaddr if error.
111  */
112 int
113 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
114     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
115 {
116 	int rcvreserve, sndreserve;
117 	int pktscale;
118 	struct sockaddr *saddr;
119 	struct ucred *origcred;
120 	CLIENT *client;
121 	struct netconfig *nconf;
122 	struct socket *so;
123 	int one = 1, retries, error, printsbmax = 0;
124 	struct thread *td = curthread;
125 
126 	/*
127 	 * We need to establish the socket using the credentials of
128 	 * the mountpoint.  Some parts of this process (such as
129 	 * sobind() and soconnect()) will use the curent thread's
130 	 * credential instead of the socket credential.  To work
131 	 * around this, temporarily change the current thread's
132 	 * credential to that of the mountpoint.
133 	 *
134 	 * XXX: It would be better to explicitly pass the correct
135 	 * credential to sobind() and soconnect().
136 	 */
137 	origcred = td->td_ucred;
138 
139 	/*
140 	 * Use the credential in nr_cred, if not NULL.
141 	 */
142 	if (nrp->nr_cred != NULL)
143 		td->td_ucred = nrp->nr_cred;
144 	else
145 		td->td_ucred = cred;
146 	saddr = nrp->nr_nam;
147 
148 	if (saddr->sa_family == AF_INET)
149 		if (nrp->nr_sotype == SOCK_DGRAM)
150 			nconf = getnetconfigent("udp");
151 		else
152 			nconf = getnetconfigent("tcp");
153 	else
154 		if (nrp->nr_sotype == SOCK_DGRAM)
155 			nconf = getnetconfigent("udp6");
156 		else
157 			nconf = getnetconfigent("tcp6");
158 
159 	pktscale = nfs_bufpackets;
160 	if (pktscale < 2)
161 		pktscale = 2;
162 	if (pktscale > 64)
163 		pktscale = 64;
164 	/*
165 	 * soreserve() can fail if sb_max is too small, so shrink pktscale
166 	 * and try again if there is an error.
167 	 * Print a log message suggesting increasing sb_max.
168 	 * Creating a socket and doing this is necessary since, if the
169 	 * reservation sizes are too large and will make soreserve() fail,
170 	 * the connection will work until a large send is attempted and
171 	 * then it will loop in the krpc code.
172 	 */
173 	so = NULL;
174 	saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
175 	error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
176 	    nrp->nr_soproto, td->td_ucred, td);
177 	if (error) {
178 		td->td_ucred = origcred;
179 		return (error);
180 	}
181 	do {
182 	    if (error != 0 && pktscale > 2) {
183 		pktscale--;
184 		if (printsbmax == 0) {
185 		    printf("nfscl: consider increasing kern.ipc.maxsockbuf\n");
186 		    printsbmax = 1;
187 		}
188 	    }
189 	    if (nrp->nr_sotype == SOCK_DGRAM) {
190 		if (nmp != NULL) {
191 			sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
192 			    pktscale;
193 			rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
194 			    pktscale;
195 		} else {
196 			sndreserve = rcvreserve = 1024 * pktscale;
197 		}
198 	    } else {
199 		if (nrp->nr_sotype != SOCK_STREAM)
200 			panic("nfscon sotype");
201 		if (nmp != NULL) {
202 			sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
203 			    sizeof (u_int32_t)) * pktscale;
204 			rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
205 			    sizeof (u_int32_t)) * pktscale;
206 		} else {
207 			sndreserve = rcvreserve = 1024 * pktscale;
208 		}
209 	    }
210 	    error = soreserve(so, sndreserve, rcvreserve);
211 	} while (error != 0 && pktscale > 2);
212 	soclose(so);
213 	if (error) {
214 		td->td_ucred = origcred;
215 		return (error);
216 	}
217 
218 	client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
219 	    nrp->nr_vers, sndreserve, rcvreserve);
220 	CLNT_CONTROL(client, CLSET_WAITCHAN, "newnfsreq");
221 	if (nmp != NULL) {
222 		if ((nmp->nm_flag & NFSMNT_INT))
223 			CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
224 		if ((nmp->nm_flag & NFSMNT_RESVPORT))
225 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
226 		if (NFSHASSOFT(nmp))
227 			retries = nmp->nm_retry;
228 		else
229 			retries = INT_MAX;
230 	} else {
231 		/*
232 		 * Three cases:
233 		 * - Null RPC callback to client
234 		 * - Non-Null RPC callback to client, wait a little longer
235 		 * - upcalls to nfsuserd and gssd (clp == NULL)
236 		 */
237 		if (callback_retry_mult == 0) {
238 			retries = NFSV4_UPCALLRETRY;
239 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
240 		} else {
241 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
242 		}
243 	}
244 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
245 
246 	mtx_lock(&nrp->nr_mtx);
247 	if (nrp->nr_client != NULL) {
248 		/*
249 		 * Someone else already connected.
250 		 */
251 		CLNT_RELEASE(client);
252 	} else {
253 		nrp->nr_client = client;
254 	}
255 
256 	/*
257 	 * Protocols that do not require connections may be optionally left
258 	 * unconnected for servers that reply from a port other than NFS_PORT.
259 	 */
260 	if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
261 		mtx_unlock(&nrp->nr_mtx);
262 		CLNT_CONTROL(client, CLSET_CONNECT, &one);
263 	} else {
264 		mtx_unlock(&nrp->nr_mtx);
265 	}
266 
267 	/* Restore current thread's credentials. */
268 	td->td_ucred = origcred;
269 	return (0);
270 }
271 
272 /*
273  * NFS disconnect. Clean up and unlink.
274  */
275 void
276 newnfs_disconnect(struct nfssockreq *nrp)
277 {
278 	CLIENT *client;
279 
280 	mtx_lock(&nrp->nr_mtx);
281 	if (nrp->nr_client != NULL) {
282 		client = nrp->nr_client;
283 		nrp->nr_client = NULL;
284 		mtx_unlock(&nrp->nr_mtx);
285 #ifdef KGSSAPI
286 		rpc_gss_secpurge(client);
287 #endif
288 		CLNT_CLOSE(client);
289 		CLNT_RELEASE(client);
290 	} else {
291 		mtx_unlock(&nrp->nr_mtx);
292 	}
293 }
294 
295 static AUTH *
296 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
297     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
298 {
299 #ifdef KGSSAPI
300 	rpc_gss_service_t svc;
301 	AUTH *auth;
302 	rpc_gss_options_req_t req_options;
303 #endif
304 
305 	switch (secflavour) {
306 #ifdef KGSSAPI
307 	case RPCSEC_GSS_KRB5:
308 	case RPCSEC_GSS_KRB5I:
309 	case RPCSEC_GSS_KRB5P:
310 		if (!mech_oid) {
311 			if (!rpc_gss_mech_to_oid("kerberosv5", &mech_oid))
312 				return (NULL);
313 		}
314 		if (secflavour == RPCSEC_GSS_KRB5)
315 			svc = rpc_gss_svc_none;
316 		else if (secflavour == RPCSEC_GSS_KRB5I)
317 			svc = rpc_gss_svc_integrity;
318 		else
319 			svc = rpc_gss_svc_privacy;
320 		req_options.req_flags = GSS_C_MUTUAL_FLAG;
321 		req_options.time_req = 0;
322 		req_options.my_cred = GSS_C_NO_CREDENTIAL;
323 		req_options.input_channel_bindings = NULL;
324 		req_options.enc_type = nfs_keytab_enctype;
325 
326 		auth = rpc_gss_secfind(nrp->nr_client, cred,
327 		    clnt_principal, srv_principal, mech_oid, svc,
328 		    &req_options);
329 		return (auth);
330 #endif
331 	case AUTH_SYS:
332 	default:
333 		return (authunix_create(cred));
334 
335 	}
336 }
337 
338 /*
339  * Callback from the RPC code to generate up/down notifications.
340  */
341 
342 struct nfs_feedback_arg {
343 	struct nfsmount *nf_mount;
344 	int		nf_lastmsg;	/* last tprintf */
345 	int		nf_tprintfmsg;
346 	struct thread	*nf_td;
347 };
348 
349 static void
350 nfs_feedback(int type, int proc, void *arg)
351 {
352 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
353 	struct nfsmount *nmp = nf->nf_mount;
354 	struct timeval now;
355 
356 	getmicrouptime(&now);
357 
358 	switch (type) {
359 	case FEEDBACK_REXMIT2:
360 	case FEEDBACK_RECONNECT:
361 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
362 			nfs_down(nmp, nf->nf_td,
363 			    "not responding", 0, NFSSTA_TIMEO);
364 			nf->nf_tprintfmsg = TRUE;
365 			nf->nf_lastmsg = now.tv_sec;
366 		}
367 		break;
368 
369 	case FEEDBACK_OK:
370 		nfs_up(nf->nf_mount, nf->nf_td,
371 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
372 		break;
373 	}
374 }
375 
376 /*
377  * newnfs_request - goes something like this
378  *	- does the rpc by calling the krpc layer
379  *	- break down rpc header and return with nfs reply
380  * nb: always frees up nd_mreq mbuf list
381  */
382 int
383 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
384     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
385     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
386     u_char *retsum, int toplevel, u_int64_t *xidp)
387 {
388 	u_int32_t *tl;
389 	time_t waituntil;
390 	int i, j;
391 	int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
392 	u_int16_t procnum;
393 	u_int trylater_delay = 1;
394 	struct nfs_feedback_arg nf;
395 	struct timeval timo, now;
396 	AUTH *auth;
397 	struct rpc_callextra ext;
398 	enum clnt_stat stat;
399 	struct nfsreq *rep = NULL;
400 	char *srv_principal = NULL;
401 
402 	if (xidp != NULL)
403 		*xidp = 0;
404 	/* Reject requests while attempting a forced unmount. */
405 	if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
406 		m_freem(nd->nd_mreq);
407 		return (ESTALE);
408 	}
409 
410 	/*
411 	 * For a client side mount, nmp is != NULL and clp == NULL. For
412 	 * server calls (callbacks or upcalls), nmp == NULL.
413 	 */
414 	if (clp != NULL) {
415 		NFSLOCKSTATE();
416 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
417 			secflavour = RPCSEC_GSS_KRB5;
418 			if (nd->nd_procnum != NFSPROC_NULL) {
419 				if (clp->lc_flags & LCL_GSSINTEGRITY)
420 					secflavour = RPCSEC_GSS_KRB5I;
421 				else if (clp->lc_flags & LCL_GSSPRIVACY)
422 					secflavour = RPCSEC_GSS_KRB5P;
423 			}
424 		}
425 		NFSUNLOCKSTATE();
426 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
427 	     nd->nd_procnum != NFSPROC_NULL) {
428 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
429 			nd->nd_flag |= ND_USEGSSNAME;
430 		if ((nd->nd_flag & ND_USEGSSNAME) && nmp->nm_krbnamelen > 0)
431 			usegssname = 1;
432 		if (NFSHASINTEGRITY(nmp))
433 			secflavour = RPCSEC_GSS_KRB5I;
434 		else if (NFSHASPRIVACY(nmp))
435 			secflavour = RPCSEC_GSS_KRB5P;
436 		else
437 			secflavour = RPCSEC_GSS_KRB5;
438 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
439 	}
440 
441 	if (nmp != NULL) {
442 		bzero(&nf, sizeof(struct nfs_feedback_arg));
443 		nf.nf_mount = nmp;
444 		nf.nf_td = td;
445 		getmicrouptime(&now);
446 		nf.nf_lastmsg = now.tv_sec -
447 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
448 	}
449 
450 	/*
451 	 * XXX if not already connected call nfs_connect now. Longer
452 	 * term, change nfs_mount to call nfs_connect unconditionally
453 	 * and let clnt_reconnect_create handle reconnects.
454 	 */
455 	if (nrp->nr_client == NULL)
456 		newnfs_connect(nmp, nrp, cred, td, 0);
457 
458 	if (nd->nd_procnum == NFSPROC_NULL)
459 		auth = authnone_create();
460 	else if (usegssname)
461 		auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
462 		    srv_principal, NULL, cred);
463 	else
464 		auth = nfs_getauth(nrp, secflavour, NULL,
465 		    srv_principal, NULL, cred);
466 	if (auth == NULL) {
467 		m_freem(nd->nd_mreq);
468 		return (EACCES);
469 	}
470 	bzero(&ext, sizeof(ext));
471 	ext.rc_auth = auth;
472 	if (nmp != NULL) {
473 		ext.rc_feedback = nfs_feedback;
474 		ext.rc_feedback_arg = &nf;
475 	}
476 
477 	procnum = nd->nd_procnum;
478 	if ((nd->nd_flag & ND_NFSV4) &&
479 	    nd->nd_procnum != NFSPROC_NULL &&
480 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
481 		procnum = NFSV4PROC_COMPOUND;
482 
483 	if (nmp != NULL) {
484 		NFSINCRGLOBAL(newnfsstats.rpcrequests);
485 		/*
486 		 * Now only used for the R_DONTRECOVER case, but until that is
487 		 * supported within the krpc code, I need to keep a queue of
488 		 * outstanding RPCs for nfsv4 client requests.
489 		 */
490 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
491 			MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
492 			    M_NFSDREQ, M_WAITOK);
493 	}
494 	trycnt = 0;
495 tryagain:
496 	if (nmp == NULL) {
497 		timo.tv_usec = 0;
498 		if (clp == NULL)
499 			timo.tv_sec = NFSV4_UPCALLTIMEO;
500 		else
501 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
502 	} else {
503 		if (nrp->nr_sotype != SOCK_DGRAM) {
504 			timo.tv_usec = 0;
505 			if ((nmp->nm_flag & NFSMNT_NFSV4))
506 				timo.tv_sec = INT_MAX;
507 			else
508 				timo.tv_sec = NFS_TCPTIMEO;
509 		} else {
510 			timo.tv_sec = nmp->nm_timeo / NFS_HZ;
511 			timo.tv_usec = (nmp->nm_timeo * 1000000) / NFS_HZ;
512 		}
513 
514 		if (rep != NULL) {
515 			rep->r_flags = 0;
516 			rep->r_nmp = nmp;
517 			/*
518 			 * Chain request into list of outstanding requests.
519 			 */
520 			NFSLOCKREQ();
521 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
522 			NFSUNLOCKREQ();
523 		}
524 	}
525 
526 	nd->nd_mrep = NULL;
527 	stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
528 	    &nd->nd_mrep, timo);
529 
530 	if (rep != NULL) {
531 		/*
532 		 * RPC done, unlink the request.
533 		 */
534 		NFSLOCKREQ();
535 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
536 		NFSUNLOCKREQ();
537 	}
538 
539 	/*
540 	 * If there was a successful reply and a tprintf msg.
541 	 * tprintf a response.
542 	 */
543 	if (stat == RPC_SUCCESS) {
544 		error = 0;
545 	} else if (stat == RPC_TIMEDOUT) {
546 		error = ETIMEDOUT;
547 	} else if (stat == RPC_VERSMISMATCH) {
548 		error = EOPNOTSUPP;
549 	} else if (stat == RPC_PROGVERSMISMATCH) {
550 		error = EPROTONOSUPPORT;
551 	} else {
552 		error = EACCES;
553 	}
554 	if (error) {
555 		m_freem(nd->nd_mreq);
556 		AUTH_DESTROY(auth);
557 		if (rep != NULL)
558 			FREE((caddr_t)rep, M_NFSDREQ);
559 		return (error);
560 	}
561 
562 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
563 
564 	nd->nd_md = nd->nd_mrep;
565 	nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
566 	nd->nd_repstat = 0;
567 	if (nd->nd_procnum != NFSPROC_NULL) {
568 		/*
569 		 * and now the actual NFS xdr.
570 		 */
571 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
572 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
573 		if (nd->nd_repstat != 0) {
574 			if ((nd->nd_repstat == NFSERR_DELAY &&
575 			     (nd->nd_flag & ND_NFSV4) &&
576 			     nd->nd_procnum != NFSPROC_SETATTR &&
577 			     nd->nd_procnum != NFSPROC_READ &&
578 			     nd->nd_procnum != NFSPROC_WRITE &&
579 			     nd->nd_procnum != NFSPROC_OPEN &&
580 			     nd->nd_procnum != NFSPROC_CREATE &&
581 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
582 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
583 			     nd->nd_procnum != NFSPROC_CLOSE &&
584 			     nd->nd_procnum != NFSPROC_LOCK &&
585 			     nd->nd_procnum != NFSPROC_LOCKU) ||
586 			    (nd->nd_repstat == NFSERR_DELAY &&
587 			     (nd->nd_flag & ND_NFSV4) == 0) ||
588 			    nd->nd_repstat == NFSERR_RESOURCE) {
589 				if (trylater_delay > NFS_TRYLATERDEL)
590 					trylater_delay = NFS_TRYLATERDEL;
591 				waituntil = NFSD_MONOSEC + trylater_delay;
592 				while (NFSD_MONOSEC < waituntil)
593 					(void) nfs_catnap(PZERO, "nfstry");
594 				trylater_delay *= 2;
595 				goto tryagain;
596 			}
597 
598 			/*
599 			 * If the File Handle was stale, invalidate the
600 			 * lookup cache, just in case.
601 			 * (vp != NULL implies a client side call)
602 			 */
603 			if (nd->nd_repstat == ESTALE && vp != NULL) {
604 				cache_purge(vp);
605 				if (ncl_call_invalcaches != NULL)
606 					(*ncl_call_invalcaches)(vp);
607 			}
608 		}
609 
610 		/*
611 		 * Get rid of the tag, return count, and PUTFH result for V4.
612 		 */
613 		if (nd->nd_flag & ND_NFSV4) {
614 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
615 			i = fxdr_unsigned(int, *tl);
616 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
617 			if (error)
618 				goto nfsmout;
619 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
620 			i = fxdr_unsigned(int, *++tl);
621 
622 			/*
623 			 * If the first op's status is non-zero, mark that
624 			 * there is no more data to process.
625 			 */
626 			if (*++tl)
627 				nd->nd_flag |= ND_NOMOREDATA;
628 
629 			/*
630 			 * If the first op is Putfh, throw its results away
631 			 * and toss the op# and status for the first op.
632 			 */
633 			if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) {
634 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
635 				i = fxdr_unsigned(int, *tl++);
636 				j = fxdr_unsigned(int, *tl);
637 				/*
638 				 * All Compounds that do an Op that must
639 				 * be in sequence consist of NFSV4OP_PUTFH
640 				 * followed by one of these. As such, we
641 				 * can determine if the seqid# should be
642 				 * incremented, here.
643 				 */
644 				if ((i == NFSV4OP_OPEN ||
645 				     i == NFSV4OP_OPENCONFIRM ||
646 				     i == NFSV4OP_OPENDOWNGRADE ||
647 				     i == NFSV4OP_CLOSE ||
648 				     i == NFSV4OP_LOCK ||
649 				     i == NFSV4OP_LOCKU) &&
650 				    (j == 0 ||
651 				     (j != NFSERR_STALECLIENTID &&
652 				      j != NFSERR_STALESTATEID &&
653 				      j != NFSERR_BADSTATEID &&
654 				      j != NFSERR_BADSEQID &&
655 				      j != NFSERR_BADXDR &&
656 				      j != NFSERR_RESOURCE &&
657 				      j != NFSERR_NOFILEHANDLE)))
658 					nd->nd_flag |= ND_INCRSEQID;
659 				/*
660 				 * If the first op's status is non-zero, mark
661 				 * that there is no more data to process.
662 				 */
663 				if (j)
664 					nd->nd_flag |= ND_NOMOREDATA;
665 			}
666 
667 			/*
668 			 * If R_DONTRECOVER is set, replace the stale error
669 			 * reply, so that recovery isn't initiated.
670 			 */
671 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
672 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
673 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
674 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
675 		}
676 	}
677 
678 	m_freem(nd->nd_mreq);
679 	AUTH_DESTROY(auth);
680 	if (rep != NULL)
681 		FREE((caddr_t)rep, M_NFSDREQ);
682 	return (0);
683 nfsmout:
684 	mbuf_freem(nd->nd_mrep);
685 	mbuf_freem(nd->nd_mreq);
686 	AUTH_DESTROY(auth);
687 	if (rep != NULL)
688 		FREE((caddr_t)rep, M_NFSDREQ);
689 	return (error);
690 }
691 
692 /*
693  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
694  * wait for all requests to complete. This is used by forced unmounts
695  * to terminate any outstanding RPCs.
696  */
697 int
698 newnfs_nmcancelreqs(struct nfsmount *nmp)
699 {
700 
701 	if (nmp->nm_sockreq.nr_client != NULL)
702 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
703 	return (0);
704 }
705 
706 /*
707  * Any signal that can interrupt an NFS operation in an intr mount
708  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
709  */
710 int newnfs_sig_set[] = {
711 	SIGINT,
712 	SIGTERM,
713 	SIGHUP,
714 	SIGKILL,
715 	SIGSTOP,
716 	SIGQUIT
717 };
718 
719 /*
720  * Check to see if one of the signals in our subset is pending on
721  * the process (in an intr mount).
722  */
723 static int
724 nfs_sig_pending(sigset_t set)
725 {
726 	int i;
727 
728 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
729 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
730 			return (1);
731 	return (0);
732 }
733 
734 /*
735  * The set/restore sigmask functions are used to (temporarily) overwrite
736  * the process p_sigmask during an RPC call (for example). These are also
737  * used in other places in the NFS client that might tsleep().
738  */
739 void
740 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
741 {
742 	sigset_t newset;
743 	int i;
744 	struct proc *p;
745 
746 	SIGFILLSET(newset);
747 	if (td == NULL)
748 		td = curthread; /* XXX */
749 	p = td->td_proc;
750 	/* Remove the NFS set of signals from newset */
751 	PROC_LOCK(p);
752 	mtx_lock(&p->p_sigacts->ps_mtx);
753 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
754 		/*
755 		 * But make sure we leave the ones already masked
756 		 * by the process, ie. remove the signal from the
757 		 * temporary signalmask only if it wasn't already
758 		 * in p_sigmask.
759 		 */
760 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
761 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
762 			SIGDELSET(newset, newnfs_sig_set[i]);
763 	}
764 	mtx_unlock(&p->p_sigacts->ps_mtx);
765 	PROC_UNLOCK(p);
766 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0);
767 }
768 
769 void
770 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
771 {
772 	if (td == NULL)
773 		td = curthread; /* XXX */
774 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
775 }
776 
777 /*
778  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
779  * old one after msleep() returns.
780  */
781 int
782 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
783 {
784 	sigset_t oldset;
785 	int error;
786 	struct proc *p;
787 
788 	if ((priority & PCATCH) == 0)
789 		return msleep(ident, mtx, priority, wmesg, timo);
790 	if (td == NULL)
791 		td = curthread; /* XXX */
792 	newnfs_set_sigmask(td, &oldset);
793 	error = msleep(ident, mtx, priority, wmesg, timo);
794 	newnfs_restore_sigmask(td, &oldset);
795 	p = td->td_proc;
796 	return (error);
797 }
798 
799 /*
800  * Test for a termination condition pending on the process.
801  * This is used for NFSMNT_INT mounts.
802  */
803 int
804 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
805 {
806 	struct proc *p;
807 	sigset_t tmpset;
808 
809 	/* Terminate all requests while attempting a forced unmount. */
810 	if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
811 		return (EIO);
812 	if (!(nmp->nm_flag & NFSMNT_INT))
813 		return (0);
814 	if (td == NULL)
815 		return (0);
816 	p = td->td_proc;
817 	PROC_LOCK(p);
818 	tmpset = p->p_siglist;
819 	SIGSETOR(tmpset, td->td_siglist);
820 	SIGSETNAND(tmpset, td->td_sigmask);
821 	mtx_lock(&p->p_sigacts->ps_mtx);
822 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
823 	mtx_unlock(&p->p_sigacts->ps_mtx);
824 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
825 	    && nfs_sig_pending(tmpset)) {
826 		PROC_UNLOCK(p);
827 		return (EINTR);
828 	}
829 	PROC_UNLOCK(p);
830 	return (0);
831 }
832 
833 static int
834 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
835 {
836 	struct proc *p;
837 
838 	p = td ? td->td_proc : NULL;
839 	if (error) {
840 		tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
841 		    server, msg, error);
842 	} else {
843 		tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
844 	}
845 	return (0);
846 }
847 
848 static void
849 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
850     int error, int flags)
851 {
852 	if (nmp == NULL)
853 		return;
854 	mtx_lock(&nmp->nm_mtx);
855 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
856 		nmp->nm_state |= NFSSTA_TIMEO;
857 		mtx_unlock(&nmp->nm_mtx);
858 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
859 		    VQ_NOTRESP, 0);
860 	} else
861 		mtx_unlock(&nmp->nm_mtx);
862 	mtx_lock(&nmp->nm_mtx);
863 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
864 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
865 		mtx_unlock(&nmp->nm_mtx);
866 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
867 		    VQ_NOTRESPLOCK, 0);
868 	} else
869 		mtx_unlock(&nmp->nm_mtx);
870 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
871 }
872 
873 static void
874 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
875     int flags, int tprintfmsg)
876 {
877 	if (nmp == NULL)
878 		return;
879 	if (tprintfmsg) {
880 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
881 	}
882 
883 	mtx_lock(&nmp->nm_mtx);
884 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
885 		nmp->nm_state &= ~NFSSTA_TIMEO;
886 		mtx_unlock(&nmp->nm_mtx);
887 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
888 		    VQ_NOTRESP, 1);
889 	} else
890 		mtx_unlock(&nmp->nm_mtx);
891 
892 	mtx_lock(&nmp->nm_mtx);
893 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
894 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
895 		mtx_unlock(&nmp->nm_mtx);
896 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
897 		    VQ_NOTRESPLOCK, 1);
898 	} else
899 		mtx_unlock(&nmp->nm_mtx);
900 }
901 
902