xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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 #ifdef notyet
303 	rpc_gss_options_req_t req_options;
304 #endif
305 #endif
306 
307 	switch (secflavour) {
308 #ifdef KGSSAPI
309 	case RPCSEC_GSS_KRB5:
310 	case RPCSEC_GSS_KRB5I:
311 	case RPCSEC_GSS_KRB5P:
312 		if (!mech_oid) {
313 			if (!rpc_gss_mech_to_oid("kerberosv5", &mech_oid))
314 				return (NULL);
315 		}
316 		if (secflavour == RPCSEC_GSS_KRB5)
317 			svc = rpc_gss_svc_none;
318 		else if (secflavour == RPCSEC_GSS_KRB5I)
319 			svc = rpc_gss_svc_integrity;
320 		else
321 			svc = rpc_gss_svc_privacy;
322 #ifdef notyet
323 		req_options.req_flags = GSS_C_MUTUAL_FLAG;
324 		req_options.time_req = 0;
325 		req_options.my_cred = GSS_C_NO_CREDENTIAL;
326 		req_options.input_channel_bindings = NULL;
327 		req_options.enc_type = nfs_keytab_enctype;
328 
329 		auth = rpc_gss_secfind(nrp->nr_client, cred,
330 		    clnt_principal, srv_principal, mech_oid, svc,
331 		    &req_options);
332 #else
333 		/*
334 		 * Until changes to the rpcsec_gss code are committed,
335 		 * there is no support for host based initiator
336 		 * principals. As such, that case cannot yet be handled.
337 		 */
338 		if (clnt_principal == NULL)
339 			auth = rpc_gss_secfind(nrp->nr_client, cred,
340 			    srv_principal, mech_oid, svc);
341 		else
342 			auth = NULL;
343 #endif
344 		if (auth != NULL)
345 			return (auth);
346 		/* fallthrough */
347 #endif	/* KGSSAPI */
348 	case AUTH_SYS:
349 	default:
350 		return (authunix_create(cred));
351 
352 	}
353 }
354 
355 /*
356  * Callback from the RPC code to generate up/down notifications.
357  */
358 
359 struct nfs_feedback_arg {
360 	struct nfsmount *nf_mount;
361 	int		nf_lastmsg;	/* last tprintf */
362 	int		nf_tprintfmsg;
363 	struct thread	*nf_td;
364 };
365 
366 static void
367 nfs_feedback(int type, int proc, void *arg)
368 {
369 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
370 	struct nfsmount *nmp = nf->nf_mount;
371 	struct timeval now;
372 
373 	getmicrouptime(&now);
374 
375 	switch (type) {
376 	case FEEDBACK_REXMIT2:
377 	case FEEDBACK_RECONNECT:
378 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
379 			nfs_down(nmp, nf->nf_td,
380 			    "not responding", 0, NFSSTA_TIMEO);
381 			nf->nf_tprintfmsg = TRUE;
382 			nf->nf_lastmsg = now.tv_sec;
383 		}
384 		break;
385 
386 	case FEEDBACK_OK:
387 		nfs_up(nf->nf_mount, nf->nf_td,
388 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
389 		break;
390 	}
391 }
392 
393 /*
394  * newnfs_request - goes something like this
395  *	- does the rpc by calling the krpc layer
396  *	- break down rpc header and return with nfs reply
397  * nb: always frees up nd_mreq mbuf list
398  */
399 int
400 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
401     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
402     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
403     u_char *retsum, int toplevel, u_int64_t *xidp)
404 {
405 	u_int32_t *tl;
406 	time_t waituntil;
407 	int i, j, set_uid = 0, set_sigset = 0;
408 	int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
409 	u_int16_t procnum;
410 	u_int trylater_delay = 1;
411 	struct nfs_feedback_arg nf;
412 	struct timeval timo, now;
413 	AUTH *auth;
414 	struct rpc_callextra ext;
415 	enum clnt_stat stat;
416 	struct nfsreq *rep = NULL;
417 	char *srv_principal = NULL;
418 	uid_t saved_uid = (uid_t)-1;
419 	sigset_t oldset;
420 
421 	if (xidp != NULL)
422 		*xidp = 0;
423 	/* Reject requests while attempting a forced unmount. */
424 	if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
425 		m_freem(nd->nd_mreq);
426 		return (ESTALE);
427 	}
428 
429 	/* For client side interruptible mounts, mask off the signals. */
430 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
431 		newnfs_set_sigmask(td, &oldset);
432 		set_sigset = 1;
433 	}
434 
435 	/*
436 	 * XXX if not already connected call nfs_connect now. Longer
437 	 * term, change nfs_mount to call nfs_connect unconditionally
438 	 * and let clnt_reconnect_create handle reconnects.
439 	 */
440 	if (nrp->nr_client == NULL)
441 		newnfs_connect(nmp, nrp, cred, td, 0);
442 
443 	/*
444 	 * For a client side mount, nmp is != NULL and clp == NULL. For
445 	 * server calls (callbacks or upcalls), nmp == NULL.
446 	 */
447 	if (clp != NULL) {
448 		NFSLOCKSTATE();
449 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
450 			secflavour = RPCSEC_GSS_KRB5;
451 			if (nd->nd_procnum != NFSPROC_NULL) {
452 				if (clp->lc_flags & LCL_GSSINTEGRITY)
453 					secflavour = RPCSEC_GSS_KRB5I;
454 				else if (clp->lc_flags & LCL_GSSPRIVACY)
455 					secflavour = RPCSEC_GSS_KRB5P;
456 			}
457 		}
458 		NFSUNLOCKSTATE();
459 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
460 	     nd->nd_procnum != NFSPROC_NULL) {
461 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
462 			nd->nd_flag |= ND_USEGSSNAME;
463 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
464 			/*
465 			 * If there is a client side host based credential,
466 			 * use that, otherwise use the system uid, if set.
467 			 */
468 			if (nmp->nm_krbnamelen > 0) {
469 				usegssname = 1;
470 			} else if (nmp->nm_uid != (uid_t)-1) {
471 				saved_uid = cred->cr_uid;
472 				cred->cr_uid = nmp->nm_uid;
473 				set_uid = 1;
474 			}
475 		} else if (nmp->nm_krbnamelen == 0 &&
476 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
477 			/*
478 			 * If there is no host based principal name and
479 			 * the system uid is set and this is root, use the
480 			 * system uid, since root won't have user
481 			 * credentials in a credentials cache file.
482 			 */
483 			saved_uid = cred->cr_uid;
484 			cred->cr_uid = nmp->nm_uid;
485 			set_uid = 1;
486 		}
487 		if (NFSHASINTEGRITY(nmp))
488 			secflavour = RPCSEC_GSS_KRB5I;
489 		else if (NFSHASPRIVACY(nmp))
490 			secflavour = RPCSEC_GSS_KRB5P;
491 		else
492 			secflavour = RPCSEC_GSS_KRB5;
493 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
494 	}
495 
496 	if (nmp != NULL) {
497 		bzero(&nf, sizeof(struct nfs_feedback_arg));
498 		nf.nf_mount = nmp;
499 		nf.nf_td = td;
500 		getmicrouptime(&now);
501 		nf.nf_lastmsg = now.tv_sec -
502 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
503 	}
504 
505 	if (nd->nd_procnum == NFSPROC_NULL)
506 		auth = authnone_create();
507 	else if (usegssname)
508 		auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
509 		    srv_principal, NULL, cred);
510 	else
511 		auth = nfs_getauth(nrp, secflavour, NULL,
512 		    srv_principal, NULL, cred);
513 	if (set_uid)
514 		cred->cr_uid = saved_uid;
515 	if (auth == NULL) {
516 		m_freem(nd->nd_mreq);
517 		if (set_sigset)
518 			newnfs_restore_sigmask(td, &oldset);
519 		return (EACCES);
520 	}
521 	bzero(&ext, sizeof(ext));
522 	ext.rc_auth = auth;
523 	if (nmp != NULL) {
524 		ext.rc_feedback = nfs_feedback;
525 		ext.rc_feedback_arg = &nf;
526 	}
527 
528 	procnum = nd->nd_procnum;
529 	if ((nd->nd_flag & ND_NFSV4) &&
530 	    nd->nd_procnum != NFSPROC_NULL &&
531 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
532 		procnum = NFSV4PROC_COMPOUND;
533 
534 	if (nmp != NULL) {
535 		NFSINCRGLOBAL(newnfsstats.rpcrequests);
536 		/*
537 		 * Now only used for the R_DONTRECOVER case, but until that is
538 		 * supported within the krpc code, I need to keep a queue of
539 		 * outstanding RPCs for nfsv4 client requests.
540 		 */
541 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
542 			MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
543 			    M_NFSDREQ, M_WAITOK);
544 	}
545 	trycnt = 0;
546 tryagain:
547 	if (nmp == NULL) {
548 		timo.tv_usec = 0;
549 		if (clp == NULL)
550 			timo.tv_sec = NFSV4_UPCALLTIMEO;
551 		else
552 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
553 	} else {
554 		if (nrp->nr_sotype != SOCK_DGRAM) {
555 			timo.tv_usec = 0;
556 			if ((nmp->nm_flag & NFSMNT_NFSV4))
557 				timo.tv_sec = INT_MAX;
558 			else
559 				timo.tv_sec = NFS_TCPTIMEO;
560 		} else {
561 			timo.tv_sec = nmp->nm_timeo / NFS_HZ;
562 			timo.tv_usec = (nmp->nm_timeo * 1000000) / NFS_HZ;
563 		}
564 
565 		if (rep != NULL) {
566 			rep->r_flags = 0;
567 			rep->r_nmp = nmp;
568 			/*
569 			 * Chain request into list of outstanding requests.
570 			 */
571 			NFSLOCKREQ();
572 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
573 			NFSUNLOCKREQ();
574 		}
575 	}
576 
577 	nd->nd_mrep = NULL;
578 	stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
579 	    &nd->nd_mrep, timo);
580 
581 	if (rep != NULL) {
582 		/*
583 		 * RPC done, unlink the request.
584 		 */
585 		NFSLOCKREQ();
586 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
587 		NFSUNLOCKREQ();
588 	}
589 
590 	/*
591 	 * If there was a successful reply and a tprintf msg.
592 	 * tprintf a response.
593 	 */
594 	if (stat == RPC_SUCCESS) {
595 		error = 0;
596 	} else if (stat == RPC_TIMEDOUT) {
597 		error = ETIMEDOUT;
598 	} else if (stat == RPC_VERSMISMATCH) {
599 		error = EOPNOTSUPP;
600 	} else if (stat == RPC_PROGVERSMISMATCH) {
601 		error = EPROTONOSUPPORT;
602 	} else {
603 		error = EACCES;
604 	}
605 	if (error) {
606 		m_freem(nd->nd_mreq);
607 		AUTH_DESTROY(auth);
608 		if (rep != NULL)
609 			FREE((caddr_t)rep, M_NFSDREQ);
610 		if (set_sigset)
611 			newnfs_restore_sigmask(td, &oldset);
612 		return (error);
613 	}
614 
615 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
616 
617 	/*
618 	 * Search for any mbufs that are not a multiple of 4 bytes long
619 	 * or with m_data not longword aligned.
620 	 * These could cause pointer alignment problems, so copy them to
621 	 * well aligned mbufs.
622 	 */
623 	newnfs_realign(&nd->nd_mrep);
624 	nd->nd_md = nd->nd_mrep;
625 	nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
626 	nd->nd_repstat = 0;
627 	if (nd->nd_procnum != NFSPROC_NULL) {
628 		/*
629 		 * and now the actual NFS xdr.
630 		 */
631 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
632 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
633 		if (nd->nd_repstat != 0) {
634 			if ((nd->nd_repstat == NFSERR_DELAY &&
635 			     (nd->nd_flag & ND_NFSV4) &&
636 			     nd->nd_procnum != NFSPROC_SETATTR &&
637 			     nd->nd_procnum != NFSPROC_READ &&
638 			     nd->nd_procnum != NFSPROC_WRITE &&
639 			     nd->nd_procnum != NFSPROC_OPEN &&
640 			     nd->nd_procnum != NFSPROC_CREATE &&
641 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
642 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
643 			     nd->nd_procnum != NFSPROC_CLOSE &&
644 			     nd->nd_procnum != NFSPROC_LOCK &&
645 			     nd->nd_procnum != NFSPROC_LOCKU) ||
646 			    (nd->nd_repstat == NFSERR_DELAY &&
647 			     (nd->nd_flag & ND_NFSV4) == 0) ||
648 			    nd->nd_repstat == NFSERR_RESOURCE) {
649 				if (trylater_delay > NFS_TRYLATERDEL)
650 					trylater_delay = NFS_TRYLATERDEL;
651 				waituntil = NFSD_MONOSEC + trylater_delay;
652 				while (NFSD_MONOSEC < waituntil)
653 					(void) nfs_catnap(PZERO, "nfstry");
654 				trylater_delay *= 2;
655 				goto tryagain;
656 			}
657 
658 			/*
659 			 * If the File Handle was stale, invalidate the
660 			 * lookup cache, just in case.
661 			 * (vp != NULL implies a client side call)
662 			 */
663 			if (nd->nd_repstat == ESTALE && vp != NULL) {
664 				cache_purge(vp);
665 				if (ncl_call_invalcaches != NULL)
666 					(*ncl_call_invalcaches)(vp);
667 			}
668 		}
669 
670 		/*
671 		 * Get rid of the tag, return count, and PUTFH result for V4.
672 		 */
673 		if (nd->nd_flag & ND_NFSV4) {
674 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
675 			i = fxdr_unsigned(int, *tl);
676 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
677 			if (error)
678 				goto nfsmout;
679 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
680 			i = fxdr_unsigned(int, *++tl);
681 
682 			/*
683 			 * If the first op's status is non-zero, mark that
684 			 * there is no more data to process.
685 			 */
686 			if (*++tl)
687 				nd->nd_flag |= ND_NOMOREDATA;
688 
689 			/*
690 			 * If the first op is Putfh, throw its results away
691 			 * and toss the op# and status for the first op.
692 			 */
693 			if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) {
694 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
695 				i = fxdr_unsigned(int, *tl++);
696 				j = fxdr_unsigned(int, *tl);
697 				/*
698 				 * All Compounds that do an Op that must
699 				 * be in sequence consist of NFSV4OP_PUTFH
700 				 * followed by one of these. As such, we
701 				 * can determine if the seqid# should be
702 				 * incremented, here.
703 				 */
704 				if ((i == NFSV4OP_OPEN ||
705 				     i == NFSV4OP_OPENCONFIRM ||
706 				     i == NFSV4OP_OPENDOWNGRADE ||
707 				     i == NFSV4OP_CLOSE ||
708 				     i == NFSV4OP_LOCK ||
709 				     i == NFSV4OP_LOCKU) &&
710 				    (j == 0 ||
711 				     (j != NFSERR_STALECLIENTID &&
712 				      j != NFSERR_STALESTATEID &&
713 				      j != NFSERR_BADSTATEID &&
714 				      j != NFSERR_BADSEQID &&
715 				      j != NFSERR_BADXDR &&
716 				      j != NFSERR_RESOURCE &&
717 				      j != NFSERR_NOFILEHANDLE)))
718 					nd->nd_flag |= ND_INCRSEQID;
719 				/*
720 				 * If the first op's status is non-zero, mark
721 				 * that there is no more data to process.
722 				 */
723 				if (j)
724 					nd->nd_flag |= ND_NOMOREDATA;
725 			}
726 
727 			/*
728 			 * If R_DONTRECOVER is set, replace the stale error
729 			 * reply, so that recovery isn't initiated.
730 			 */
731 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
732 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
733 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
734 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
735 		}
736 	}
737 
738 	m_freem(nd->nd_mreq);
739 	AUTH_DESTROY(auth);
740 	if (rep != NULL)
741 		FREE((caddr_t)rep, M_NFSDREQ);
742 	if (set_sigset)
743 		newnfs_restore_sigmask(td, &oldset);
744 	return (0);
745 nfsmout:
746 	mbuf_freem(nd->nd_mrep);
747 	mbuf_freem(nd->nd_mreq);
748 	AUTH_DESTROY(auth);
749 	if (rep != NULL)
750 		FREE((caddr_t)rep, M_NFSDREQ);
751 	if (set_sigset)
752 		newnfs_restore_sigmask(td, &oldset);
753 	return (error);
754 }
755 
756 /*
757  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
758  * wait for all requests to complete. This is used by forced unmounts
759  * to terminate any outstanding RPCs.
760  */
761 int
762 newnfs_nmcancelreqs(struct nfsmount *nmp)
763 {
764 
765 	if (nmp->nm_sockreq.nr_client != NULL)
766 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
767 	return (0);
768 }
769 
770 /*
771  * Any signal that can interrupt an NFS operation in an intr mount
772  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
773  */
774 int newnfs_sig_set[] = {
775 	SIGINT,
776 	SIGTERM,
777 	SIGHUP,
778 	SIGKILL,
779 	SIGSTOP,
780 	SIGQUIT
781 };
782 
783 /*
784  * Check to see if one of the signals in our subset is pending on
785  * the process (in an intr mount).
786  */
787 static int
788 nfs_sig_pending(sigset_t set)
789 {
790 	int i;
791 
792 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
793 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
794 			return (1);
795 	return (0);
796 }
797 
798 /*
799  * The set/restore sigmask functions are used to (temporarily) overwrite
800  * the process p_sigmask during an RPC call (for example). These are also
801  * used in other places in the NFS client that might tsleep().
802  */
803 void
804 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
805 {
806 	sigset_t newset;
807 	int i;
808 	struct proc *p;
809 
810 	SIGFILLSET(newset);
811 	if (td == NULL)
812 		td = curthread; /* XXX */
813 	p = td->td_proc;
814 	/* Remove the NFS set of signals from newset */
815 	PROC_LOCK(p);
816 	mtx_lock(&p->p_sigacts->ps_mtx);
817 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
818 		/*
819 		 * But make sure we leave the ones already masked
820 		 * by the process, ie. remove the signal from the
821 		 * temporary signalmask only if it wasn't already
822 		 * in p_sigmask.
823 		 */
824 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
825 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
826 			SIGDELSET(newset, newnfs_sig_set[i]);
827 	}
828 	mtx_unlock(&p->p_sigacts->ps_mtx);
829 	PROC_UNLOCK(p);
830 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0);
831 }
832 
833 void
834 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
835 {
836 	if (td == NULL)
837 		td = curthread; /* XXX */
838 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
839 }
840 
841 /*
842  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
843  * old one after msleep() returns.
844  */
845 int
846 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
847 {
848 	sigset_t oldset;
849 	int error;
850 	struct proc *p;
851 
852 	if ((priority & PCATCH) == 0)
853 		return msleep(ident, mtx, priority, wmesg, timo);
854 	if (td == NULL)
855 		td = curthread; /* XXX */
856 	newnfs_set_sigmask(td, &oldset);
857 	error = msleep(ident, mtx, priority, wmesg, timo);
858 	newnfs_restore_sigmask(td, &oldset);
859 	p = td->td_proc;
860 	return (error);
861 }
862 
863 /*
864  * Test for a termination condition pending on the process.
865  * This is used for NFSMNT_INT mounts.
866  */
867 int
868 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
869 {
870 	struct proc *p;
871 	sigset_t tmpset;
872 
873 	/* Terminate all requests while attempting a forced unmount. */
874 	if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
875 		return (EIO);
876 	if (!(nmp->nm_flag & NFSMNT_INT))
877 		return (0);
878 	if (td == NULL)
879 		return (0);
880 	p = td->td_proc;
881 	PROC_LOCK(p);
882 	tmpset = p->p_siglist;
883 	SIGSETOR(tmpset, td->td_siglist);
884 	SIGSETNAND(tmpset, td->td_sigmask);
885 	mtx_lock(&p->p_sigacts->ps_mtx);
886 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
887 	mtx_unlock(&p->p_sigacts->ps_mtx);
888 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
889 	    && nfs_sig_pending(tmpset)) {
890 		PROC_UNLOCK(p);
891 		return (EINTR);
892 	}
893 	PROC_UNLOCK(p);
894 	return (0);
895 }
896 
897 static int
898 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
899 {
900 	struct proc *p;
901 
902 	p = td ? td->td_proc : NULL;
903 	if (error) {
904 		tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
905 		    server, msg, error);
906 	} else {
907 		tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
908 	}
909 	return (0);
910 }
911 
912 static void
913 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
914     int error, int flags)
915 {
916 	if (nmp == NULL)
917 		return;
918 	mtx_lock(&nmp->nm_mtx);
919 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
920 		nmp->nm_state |= NFSSTA_TIMEO;
921 		mtx_unlock(&nmp->nm_mtx);
922 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
923 		    VQ_NOTRESP, 0);
924 	} else
925 		mtx_unlock(&nmp->nm_mtx);
926 	mtx_lock(&nmp->nm_mtx);
927 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
928 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
929 		mtx_unlock(&nmp->nm_mtx);
930 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
931 		    VQ_NOTRESPLOCK, 0);
932 	} else
933 		mtx_unlock(&nmp->nm_mtx);
934 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
935 }
936 
937 static void
938 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
939     int flags, int tprintfmsg)
940 {
941 	if (nmp == NULL)
942 		return;
943 	if (tprintfmsg) {
944 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
945 	}
946 
947 	mtx_lock(&nmp->nm_mtx);
948 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
949 		nmp->nm_state &= ~NFSSTA_TIMEO;
950 		mtx_unlock(&nmp->nm_mtx);
951 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
952 		    VQ_NOTRESP, 1);
953 	} else
954 		mtx_unlock(&nmp->nm_mtx);
955 
956 	mtx_lock(&nmp->nm_mtx);
957 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
958 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
959 		mtx_unlock(&nmp->nm_mtx);
960 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
961 		    VQ_NOTRESPLOCK, 1);
962 	} else
963 		mtx_unlock(&nmp->nm_mtx);
964 }
965 
966