xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision 8657387683946d0c03e09fe77029edfe309eeb20)
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  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
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_kgssapi.h"
42 #include "opt_nfs.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/limits.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/signalvar.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysctl.h>
57 #include <sys/syslog.h>
58 #include <sys/vnode.h>
59 
60 #include <rpc/rpc.h>
61 #include <rpc/krpc.h>
62 
63 #include <kgssapi/krb5/kcrypto.h>
64 
65 #include <fs/nfs/nfsport.h>
66 
67 #ifdef KDTRACE_HOOKS
68 #include <sys/dtrace_bsd.h>
69 
70 dtrace_nfsclient_nfs23_start_probe_func_t
71 		dtrace_nfscl_nfs234_start_probe;
72 
73 dtrace_nfsclient_nfs23_done_probe_func_t
74 		dtrace_nfscl_nfs234_done_probe;
75 
76 /*
77  * Registered probes by RPC type.
78  */
79 uint32_t	nfscl_nfs2_start_probes[NFSV41_NPROCS + 1];
80 uint32_t	nfscl_nfs2_done_probes[NFSV41_NPROCS + 1];
81 
82 uint32_t	nfscl_nfs3_start_probes[NFSV41_NPROCS + 1];
83 uint32_t	nfscl_nfs3_done_probes[NFSV41_NPROCS + 1];
84 
85 uint32_t	nfscl_nfs4_start_probes[NFSV41_NPROCS + 1];
86 uint32_t	nfscl_nfs4_done_probes[NFSV41_NPROCS + 1];
87 #endif
88 
89 NFSSTATESPINLOCK;
90 NFSREQSPINLOCK;
91 NFSDLOCKMUTEX;
92 NFSCLSTATEMUTEX;
93 extern struct nfsstatsv1 nfsstatsv1;
94 extern struct nfsreqhead nfsd_reqq;
95 extern int nfscl_ticks;
96 extern void (*ncl_call_invalcaches)(struct vnode *);
97 extern int nfs_numnfscbd;
98 extern int nfscl_debuglevel;
99 
100 SVCPOOL		*nfscbd_pool;
101 static int	nfsrv_gsscallbackson = 0;
102 static int	nfs_bufpackets = 4;
103 static int	nfs_reconnects;
104 static int	nfs3_jukebox_delay = 10;
105 static int	nfs_skip_wcc_data_onerr = 1;
106 
107 SYSCTL_DECL(_vfs_nfs);
108 
109 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
110     "Buffer reservation size 2 < x < 64");
111 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
112     "Number of times the nfs client has had to reconnect");
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
114     "Number of seconds to delay a retry after receiving EJUKEBOX");
115 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
116     "Disable weak cache consistency checking when server returns an error");
117 
118 static void	nfs_down(struct nfsmount *, struct thread *, const char *,
119     int, int);
120 static void	nfs_up(struct nfsmount *, struct thread *, const char *,
121     int, int);
122 static int	nfs_msg(struct thread *, const char *, const char *, int);
123 
124 struct nfs_cached_auth {
125 	int		ca_refs; /* refcount, including 1 from the cache */
126 	uid_t		ca_uid;	 /* uid that corresponds to this auth */
127 	AUTH		*ca_auth; /* RPC auth handle */
128 };
129 
130 static int nfsv2_procid[NFS_V3NPROCS] = {
131 	NFSV2PROC_NULL,
132 	NFSV2PROC_GETATTR,
133 	NFSV2PROC_SETATTR,
134 	NFSV2PROC_LOOKUP,
135 	NFSV2PROC_NOOP,
136 	NFSV2PROC_READLINK,
137 	NFSV2PROC_READ,
138 	NFSV2PROC_WRITE,
139 	NFSV2PROC_CREATE,
140 	NFSV2PROC_MKDIR,
141 	NFSV2PROC_SYMLINK,
142 	NFSV2PROC_CREATE,
143 	NFSV2PROC_REMOVE,
144 	NFSV2PROC_RMDIR,
145 	NFSV2PROC_RENAME,
146 	NFSV2PROC_LINK,
147 	NFSV2PROC_READDIR,
148 	NFSV2PROC_NOOP,
149 	NFSV2PROC_STATFS,
150 	NFSV2PROC_NOOP,
151 	NFSV2PROC_NOOP,
152 	NFSV2PROC_NOOP,
153 };
154 
155 /*
156  * Initialize sockets and congestion for a new NFS connection.
157  * We do not free the sockaddr if error.
158  */
159 int
160 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
161     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
162 {
163 	int rcvreserve, sndreserve;
164 	int pktscale, pktscalesav;
165 	struct sockaddr *saddr;
166 	struct ucred *origcred;
167 	CLIENT *client;
168 	struct netconfig *nconf;
169 	struct socket *so;
170 	int one = 1, retries, error = 0;
171 	struct thread *td = curthread;
172 	SVCXPRT *xprt;
173 	struct timeval timo;
174 
175 	/*
176 	 * We need to establish the socket using the credentials of
177 	 * the mountpoint.  Some parts of this process (such as
178 	 * sobind() and soconnect()) will use the curent thread's
179 	 * credential instead of the socket credential.  To work
180 	 * around this, temporarily change the current thread's
181 	 * credential to that of the mountpoint.
182 	 *
183 	 * XXX: It would be better to explicitly pass the correct
184 	 * credential to sobind() and soconnect().
185 	 */
186 	origcred = td->td_ucred;
187 
188 	/*
189 	 * Use the credential in nr_cred, if not NULL.
190 	 */
191 	if (nrp->nr_cred != NULL)
192 		td->td_ucred = nrp->nr_cred;
193 	else
194 		td->td_ucred = cred;
195 	saddr = nrp->nr_nam;
196 
197 	if (saddr->sa_family == AF_INET)
198 		if (nrp->nr_sotype == SOCK_DGRAM)
199 			nconf = getnetconfigent("udp");
200 		else
201 			nconf = getnetconfigent("tcp");
202 	else if (saddr->sa_family == AF_LOCAL)
203 		nconf = getnetconfigent("local");
204 	else
205 		if (nrp->nr_sotype == SOCK_DGRAM)
206 			nconf = getnetconfigent("udp6");
207 		else
208 			nconf = getnetconfigent("tcp6");
209 
210 	pktscale = nfs_bufpackets;
211 	if (pktscale < 2)
212 		pktscale = 2;
213 	if (pktscale > 64)
214 		pktscale = 64;
215 	pktscalesav = pktscale;
216 	/*
217 	 * soreserve() can fail if sb_max is too small, so shrink pktscale
218 	 * and try again if there is an error.
219 	 * Print a log message suggesting increasing sb_max.
220 	 * Creating a socket and doing this is necessary since, if the
221 	 * reservation sizes are too large and will make soreserve() fail,
222 	 * the connection will work until a large send is attempted and
223 	 * then it will loop in the krpc code.
224 	 */
225 	so = NULL;
226 	saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
227 	error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
228 	    nrp->nr_soproto, td->td_ucred, td);
229 	if (error) {
230 		td->td_ucred = origcred;
231 		goto out;
232 	}
233 	do {
234 	    if (error != 0 && pktscale > 2) {
235 		if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
236 		    pktscale == pktscalesav)
237 		    printf("Consider increasing kern.ipc.maxsockbuf\n");
238 		pktscale--;
239 	    }
240 	    if (nrp->nr_sotype == SOCK_DGRAM) {
241 		if (nmp != NULL) {
242 			sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
243 			    pktscale;
244 			rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
245 			    pktscale;
246 		} else {
247 			sndreserve = rcvreserve = 1024 * pktscale;
248 		}
249 	    } else {
250 		if (nrp->nr_sotype != SOCK_STREAM)
251 			panic("nfscon sotype");
252 		if (nmp != NULL) {
253 			sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
254 			    sizeof (u_int32_t)) * pktscale;
255 			rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
256 			    sizeof (u_int32_t)) * pktscale;
257 		} else {
258 			sndreserve = rcvreserve = 1024 * pktscale;
259 		}
260 	    }
261 	    error = soreserve(so, sndreserve, rcvreserve);
262 	    if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
263 		pktscale <= 2)
264 		printf("Must increase kern.ipc.maxsockbuf or reduce"
265 		    " rsize, wsize\n");
266 	} while (error != 0 && pktscale > 2);
267 	soclose(so);
268 	if (error) {
269 		td->td_ucred = origcred;
270 		goto out;
271 	}
272 
273 	client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
274 	    nrp->nr_vers, sndreserve, rcvreserve);
275 	CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
276 	if (nmp != NULL) {
277 		if ((nmp->nm_flag & NFSMNT_INT))
278 			CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
279 		if ((nmp->nm_flag & NFSMNT_RESVPORT))
280 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
281 		if (NFSHASSOFT(nmp)) {
282 			if (nmp->nm_sotype == SOCK_DGRAM)
283 				/*
284 				 * For UDP, the large timeout for a reconnect
285 				 * will be set to "nm_retry * nm_timeo / 2", so
286 				 * we only want to do 2 reconnect timeout
287 				 * retries.
288 				 */
289 				retries = 2;
290 			else
291 				retries = nmp->nm_retry;
292 		} else
293 			retries = INT_MAX;
294 		/* cred == NULL for DS connects. */
295 		if (NFSHASNFSV4N(nmp) && cred != NULL) {
296 			/*
297 			 * Make sure the nfscbd_pool doesn't get destroyed
298 			 * while doing this.
299 			 */
300 			NFSD_LOCK();
301 			if (nfs_numnfscbd > 0) {
302 				nfs_numnfscbd++;
303 				NFSD_UNLOCK();
304 				xprt = svc_vc_create_backchannel(nfscbd_pool);
305 				CLNT_CONTROL(client, CLSET_BACKCHANNEL, xprt);
306 				NFSD_LOCK();
307 				nfs_numnfscbd--;
308 				if (nfs_numnfscbd == 0)
309 					wakeup(&nfs_numnfscbd);
310 			}
311 			NFSD_UNLOCK();
312 		}
313 	} else {
314 		/*
315 		 * Three cases:
316 		 * - Null RPC callback to client
317 		 * - Non-Null RPC callback to client, wait a little longer
318 		 * - upcalls to nfsuserd and gssd (clp == NULL)
319 		 */
320 		if (callback_retry_mult == 0) {
321 			retries = NFSV4_UPCALLRETRY;
322 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
323 		} else {
324 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
325 		}
326 	}
327 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
328 
329 	if (nmp != NULL) {
330 		/*
331 		 * For UDP, there are 2 timeouts:
332 		 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
333 		 *   that does a retransmit of an RPC request using the same
334 		 *   socket and xid. This is what you normally want to do,
335 		 *   since NFS servers depend on "same xid" for their
336 		 *   Duplicate Request Cache.
337 		 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
338 		 *   retransmits on the same socket should fail and a fresh
339 		 *   socket created. Each of these timeouts counts as one
340 		 *   CLSET_RETRIES as set above.
341 		 * Set the initial retransmit timeout for UDP. This timeout
342 		 * doesn't exist for TCP and the following call just fails,
343 		 * which is ok.
344 		 */
345 		timo.tv_sec = nmp->nm_timeo / NFS_HZ;
346 		timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
347 		CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
348 	}
349 
350 	mtx_lock(&nrp->nr_mtx);
351 	if (nrp->nr_client != NULL) {
352 		mtx_unlock(&nrp->nr_mtx);
353 		/*
354 		 * Someone else already connected.
355 		 */
356 		CLNT_RELEASE(client);
357 	} else {
358 		nrp->nr_client = client;
359 		/*
360 		 * Protocols that do not require connections may be optionally
361 		 * left unconnected for servers that reply from a port other
362 		 * than NFS_PORT.
363 		 */
364 		if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
365 			mtx_unlock(&nrp->nr_mtx);
366 			CLNT_CONTROL(client, CLSET_CONNECT, &one);
367 		} else
368 			mtx_unlock(&nrp->nr_mtx);
369 	}
370 
371 
372 	/* Restore current thread's credentials. */
373 	td->td_ucred = origcred;
374 
375 out:
376 	NFSEXITCODE(error);
377 	return (error);
378 }
379 
380 /*
381  * NFS disconnect. Clean up and unlink.
382  */
383 void
384 newnfs_disconnect(struct nfssockreq *nrp)
385 {
386 	CLIENT *client;
387 
388 	mtx_lock(&nrp->nr_mtx);
389 	if (nrp->nr_client != NULL) {
390 		client = nrp->nr_client;
391 		nrp->nr_client = NULL;
392 		mtx_unlock(&nrp->nr_mtx);
393 		rpc_gss_secpurge_call(client);
394 		CLNT_CLOSE(client);
395 		CLNT_RELEASE(client);
396 	} else {
397 		mtx_unlock(&nrp->nr_mtx);
398 	}
399 }
400 
401 static AUTH *
402 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
403     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
404 {
405 	rpc_gss_service_t svc;
406 	AUTH *auth;
407 
408 	switch (secflavour) {
409 	case RPCSEC_GSS_KRB5:
410 	case RPCSEC_GSS_KRB5I:
411 	case RPCSEC_GSS_KRB5P:
412 		if (!mech_oid) {
413 			if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
414 				return (NULL);
415 		}
416 		if (secflavour == RPCSEC_GSS_KRB5)
417 			svc = rpc_gss_svc_none;
418 		else if (secflavour == RPCSEC_GSS_KRB5I)
419 			svc = rpc_gss_svc_integrity;
420 		else
421 			svc = rpc_gss_svc_privacy;
422 
423 		if (clnt_principal == NULL)
424 			auth = rpc_gss_secfind_call(nrp->nr_client, cred,
425 			    srv_principal, mech_oid, svc);
426 		else {
427 			auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
428 			    clnt_principal, srv_principal, "kerberosv5",
429 			    svc, NULL, NULL, NULL);
430 			return (auth);
431 		}
432 		if (auth != NULL)
433 			return (auth);
434 		/* fallthrough */
435 	case AUTH_SYS:
436 	default:
437 		return (authunix_create(cred));
438 
439 	}
440 }
441 
442 /*
443  * Callback from the RPC code to generate up/down notifications.
444  */
445 
446 struct nfs_feedback_arg {
447 	struct nfsmount *nf_mount;
448 	int		nf_lastmsg;	/* last tprintf */
449 	int		nf_tprintfmsg;
450 	struct thread	*nf_td;
451 };
452 
453 static void
454 nfs_feedback(int type, int proc, void *arg)
455 {
456 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
457 	struct nfsmount *nmp = nf->nf_mount;
458 	time_t now;
459 
460 	switch (type) {
461 	case FEEDBACK_REXMIT2:
462 	case FEEDBACK_RECONNECT:
463 		now = NFSD_MONOSEC;
464 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
465 			nfs_down(nmp, nf->nf_td,
466 			    "not responding", 0, NFSSTA_TIMEO);
467 			nf->nf_tprintfmsg = TRUE;
468 			nf->nf_lastmsg = now;
469 		}
470 		break;
471 
472 	case FEEDBACK_OK:
473 		nfs_up(nf->nf_mount, nf->nf_td,
474 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
475 		break;
476 	}
477 }
478 
479 /*
480  * newnfs_request - goes something like this
481  *	- does the rpc by calling the krpc layer
482  *	- break down rpc header and return with nfs reply
483  * nb: always frees up nd_mreq mbuf list
484  */
485 int
486 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
487     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
488     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
489     u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
490 {
491 	uint32_t retseq, retval, slotseq, *tl;
492 	time_t waituntil;
493 	int i = 0, j = 0, opcnt, set_sigset = 0, slot;
494 	int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
495 	int freeslot, maxslot, reterr, slotpos, timeo;
496 	u_int16_t procnum;
497 	u_int trylater_delay = 1;
498 	struct nfs_feedback_arg nf;
499 	struct timeval timo;
500 	AUTH *auth;
501 	struct rpc_callextra ext;
502 	enum clnt_stat stat;
503 	struct nfsreq *rep = NULL;
504 	char *srv_principal = NULL, *clnt_principal = NULL;
505 	sigset_t oldset;
506 	struct ucred *authcred;
507 	struct nfsclsession *sep;
508 	uint8_t sessionid[NFSX_V4SESSIONID];
509 
510 	sep = dssep;
511 	if (xidp != NULL)
512 		*xidp = 0;
513 	/* Reject requests while attempting a forced unmount. */
514 	if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
515 		m_freem(nd->nd_mreq);
516 		return (ESTALE);
517 	}
518 
519 	/*
520 	 * Set authcred, which is used to acquire RPC credentials to
521 	 * the cred argument, by default. The crhold() should not be
522 	 * necessary, but will ensure that some future code change
523 	 * doesn't result in the credential being free'd prematurely.
524 	 */
525 	authcred = crhold(cred);
526 
527 	/* For client side interruptible mounts, mask off the signals. */
528 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
529 		newnfs_set_sigmask(td, &oldset);
530 		set_sigset = 1;
531 	}
532 
533 	/*
534 	 * XXX if not already connected call nfs_connect now. Longer
535 	 * term, change nfs_mount to call nfs_connect unconditionally
536 	 * and let clnt_reconnect_create handle reconnects.
537 	 */
538 	if (nrp->nr_client == NULL)
539 		newnfs_connect(nmp, nrp, cred, td, 0);
540 
541 	/*
542 	 * For a client side mount, nmp is != NULL and clp == NULL. For
543 	 * server calls (callbacks or upcalls), nmp == NULL.
544 	 */
545 	if (clp != NULL) {
546 		NFSLOCKSTATE();
547 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
548 			secflavour = RPCSEC_GSS_KRB5;
549 			if (nd->nd_procnum != NFSPROC_NULL) {
550 				if (clp->lc_flags & LCL_GSSINTEGRITY)
551 					secflavour = RPCSEC_GSS_KRB5I;
552 				else if (clp->lc_flags & LCL_GSSPRIVACY)
553 					secflavour = RPCSEC_GSS_KRB5P;
554 			}
555 		}
556 		NFSUNLOCKSTATE();
557 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
558 	     nd->nd_procnum != NFSPROC_NULL) {
559 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
560 			nd->nd_flag |= ND_USEGSSNAME;
561 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
562 			/*
563 			 * If there is a client side host based credential,
564 			 * use that, otherwise use the system uid, if set.
565 			 * The system uid is in the nmp->nm_sockreq.nr_cred
566 			 * credentials.
567 			 */
568 			if (nmp->nm_krbnamelen > 0) {
569 				usegssname = 1;
570 				clnt_principal = nmp->nm_krbname;
571 			} else if (nmp->nm_uid != (uid_t)-1) {
572 				KASSERT(nmp->nm_sockreq.nr_cred != NULL,
573 				    ("newnfs_request: NULL nr_cred"));
574 				crfree(authcred);
575 				authcred = crhold(nmp->nm_sockreq.nr_cred);
576 			}
577 		} else if (nmp->nm_krbnamelen == 0 &&
578 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
579 			/*
580 			 * If there is no host based principal name and
581 			 * the system uid is set and this is root, use the
582 			 * system uid, since root won't have user
583 			 * credentials in a credentials cache file.
584 			 * The system uid is in the nmp->nm_sockreq.nr_cred
585 			 * credentials.
586 			 */
587 			KASSERT(nmp->nm_sockreq.nr_cred != NULL,
588 			    ("newnfs_request: NULL nr_cred"));
589 			crfree(authcred);
590 			authcred = crhold(nmp->nm_sockreq.nr_cred);
591 		}
592 		if (NFSHASINTEGRITY(nmp))
593 			secflavour = RPCSEC_GSS_KRB5I;
594 		else if (NFSHASPRIVACY(nmp))
595 			secflavour = RPCSEC_GSS_KRB5P;
596 		else
597 			secflavour = RPCSEC_GSS_KRB5;
598 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
599 	} else if (nmp != NULL && !NFSHASKERB(nmp) &&
600 	    nd->nd_procnum != NFSPROC_NULL &&
601 	    (nd->nd_flag & ND_USEGSSNAME) != 0) {
602 		/*
603 		 * Use the uid that did the mount when the RPC is doing
604 		 * NFSv4 system operations, as indicated by the
605 		 * ND_USEGSSNAME flag, for the AUTH_SYS case.
606 		 * The credentials in nm_sockreq.nr_cred were used for the
607 		 * mount.
608 		 */
609 		KASSERT(nmp->nm_sockreq.nr_cred != NULL,
610 		    ("newnfs_request: NULL nr_cred"));
611 		crfree(authcred);
612 		authcred = crhold(nmp->nm_sockreq.nr_cred);
613 	}
614 
615 	if (nmp != NULL) {
616 		bzero(&nf, sizeof(struct nfs_feedback_arg));
617 		nf.nf_mount = nmp;
618 		nf.nf_td = td;
619 		nf.nf_lastmsg = NFSD_MONOSEC -
620 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
621 	}
622 
623 	if (nd->nd_procnum == NFSPROC_NULL)
624 		auth = authnone_create();
625 	else if (usegssname) {
626 		/*
627 		 * For this case, the authenticator is held in the
628 		 * nfssockreq structure, so don't release the reference count
629 		 * held on it. --> Don't AUTH_DESTROY() it in this function.
630 		 */
631 		if (nrp->nr_auth == NULL)
632 			nrp->nr_auth = nfs_getauth(nrp, secflavour,
633 			    clnt_principal, srv_principal, NULL, authcred);
634 		else
635 			rpc_gss_refresh_auth_call(nrp->nr_auth);
636 		auth = nrp->nr_auth;
637 	} else
638 		auth = nfs_getauth(nrp, secflavour, NULL,
639 		    srv_principal, NULL, authcred);
640 	crfree(authcred);
641 	if (auth == NULL) {
642 		m_freem(nd->nd_mreq);
643 		if (set_sigset)
644 			newnfs_restore_sigmask(td, &oldset);
645 		return (EACCES);
646 	}
647 	bzero(&ext, sizeof(ext));
648 	ext.rc_auth = auth;
649 	if (nmp != NULL) {
650 		ext.rc_feedback = nfs_feedback;
651 		ext.rc_feedback_arg = &nf;
652 	}
653 
654 	procnum = nd->nd_procnum;
655 	if ((nd->nd_flag & ND_NFSV4) &&
656 	    nd->nd_procnum != NFSPROC_NULL &&
657 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
658 		procnum = NFSV4PROC_COMPOUND;
659 
660 	if (nmp != NULL) {
661 		NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
662 
663 		/* Map the procnum to the old NFSv2 one, as required. */
664 		if ((nd->nd_flag & ND_NFSV2) != 0) {
665 			if (nd->nd_procnum < NFS_V3NPROCS)
666 				procnum = nfsv2_procid[nd->nd_procnum];
667 			else
668 				procnum = NFSV2PROC_NOOP;
669 		}
670 
671 		/*
672 		 * Now only used for the R_DONTRECOVER case, but until that is
673 		 * supported within the krpc code, I need to keep a queue of
674 		 * outstanding RPCs for nfsv4 client requests.
675 		 */
676 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
677 			MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
678 			    M_NFSDREQ, M_WAITOK);
679 #ifdef KDTRACE_HOOKS
680 		if (dtrace_nfscl_nfs234_start_probe != NULL) {
681 			uint32_t probe_id;
682 			int probe_procnum;
683 
684 			if (nd->nd_flag & ND_NFSV4) {
685 				probe_id =
686 				    nfscl_nfs4_start_probes[nd->nd_procnum];
687 				probe_procnum = nd->nd_procnum;
688 			} else if (nd->nd_flag & ND_NFSV3) {
689 				probe_id = nfscl_nfs3_start_probes[procnum];
690 				probe_procnum = procnum;
691 			} else {
692 				probe_id =
693 				    nfscl_nfs2_start_probes[nd->nd_procnum];
694 				probe_procnum = procnum;
695 			}
696 			if (probe_id != 0)
697 				(dtrace_nfscl_nfs234_start_probe)
698 				    (probe_id, vp, nd->nd_mreq, cred,
699 				     probe_procnum);
700 		}
701 #endif
702 	}
703 	trycnt = 0;
704 	freeslot = -1;		/* Set to slot that needs to be free'd */
705 tryagain:
706 	slot = -1;		/* Slot that needs a sequence# increment. */
707 	/*
708 	 * This timeout specifies when a new socket should be created,
709 	 * along with new xid values. For UDP, this should be done
710 	 * infrequently, since retransmits of RPC requests should normally
711 	 * use the same xid.
712 	 */
713 	if (nmp == NULL) {
714 		timo.tv_usec = 0;
715 		if (clp == NULL)
716 			timo.tv_sec = NFSV4_UPCALLTIMEO;
717 		else
718 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
719 	} else {
720 		if (nrp->nr_sotype != SOCK_DGRAM) {
721 			timo.tv_usec = 0;
722 			if ((nmp->nm_flag & NFSMNT_NFSV4))
723 				timo.tv_sec = INT_MAX;
724 			else
725 				timo.tv_sec = NFS_TCPTIMEO;
726 		} else {
727 			if (NFSHASSOFT(nmp)) {
728 				/*
729 				 * CLSET_RETRIES is set to 2, so this should be
730 				 * half of the total timeout required.
731 				 */
732 				timeo = nmp->nm_retry * nmp->nm_timeo / 2;
733 				if (timeo < 1)
734 					timeo = 1;
735 				timo.tv_sec = timeo / NFS_HZ;
736 				timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
737 				    NFS_HZ;
738 			} else {
739 				/* For UDP hard mounts, use a large value. */
740 				timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
741 				timo.tv_usec = 0;
742 			}
743 		}
744 
745 		if (rep != NULL) {
746 			rep->r_flags = 0;
747 			rep->r_nmp = nmp;
748 			/*
749 			 * Chain request into list of outstanding requests.
750 			 */
751 			NFSLOCKREQ();
752 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
753 			NFSUNLOCKREQ();
754 		}
755 	}
756 
757 	nd->nd_mrep = NULL;
758 	if (clp != NULL && sep != NULL)
759 		stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
760 		    nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
761 	else
762 		stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
763 		    nd->nd_mreq, &nd->nd_mrep, timo);
764 
765 	if (rep != NULL) {
766 		/*
767 		 * RPC done, unlink the request.
768 		 */
769 		NFSLOCKREQ();
770 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
771 		NFSUNLOCKREQ();
772 	}
773 
774 	/*
775 	 * If there was a successful reply and a tprintf msg.
776 	 * tprintf a response.
777 	 */
778 	if (stat == RPC_SUCCESS) {
779 		error = 0;
780 	} else if (stat == RPC_TIMEDOUT) {
781 		NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
782 		error = ETIMEDOUT;
783 	} else if (stat == RPC_VERSMISMATCH) {
784 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
785 		error = EOPNOTSUPP;
786 	} else if (stat == RPC_PROGVERSMISMATCH) {
787 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
788 		error = EPROTONOSUPPORT;
789 	} else if (stat == RPC_INTR) {
790 		error = EINTR;
791 	} else {
792 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
793 		error = EACCES;
794 	}
795 	if (error) {
796 		m_freem(nd->nd_mreq);
797 		if (usegssname == 0)
798 			AUTH_DESTROY(auth);
799 		if (rep != NULL)
800 			FREE((caddr_t)rep, M_NFSDREQ);
801 		if (set_sigset)
802 			newnfs_restore_sigmask(td, &oldset);
803 		return (error);
804 	}
805 
806 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
807 
808 	/*
809 	 * Search for any mbufs that are not a multiple of 4 bytes long
810 	 * or with m_data not longword aligned.
811 	 * These could cause pointer alignment problems, so copy them to
812 	 * well aligned mbufs.
813 	 */
814 	newnfs_realign(&nd->nd_mrep, M_WAITOK);
815 	nd->nd_md = nd->nd_mrep;
816 	nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
817 	nd->nd_repstat = 0;
818 	if (nd->nd_procnum != NFSPROC_NULL &&
819 	    nd->nd_procnum != NFSV4PROC_CBNULL) {
820 		/* If sep == NULL, set it to the default in nmp. */
821 		if (sep == NULL && nmp != NULL)
822 			sep = nfsmnt_mdssession(nmp);
823 		/*
824 		 * and now the actual NFS xdr.
825 		 */
826 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
827 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
828 		if (nd->nd_repstat >= 10000)
829 			NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
830 			    (int)nd->nd_repstat);
831 
832 		/*
833 		 * Get rid of the tag, return count and SEQUENCE result for
834 		 * NFSv4.
835 		 */
836 		if ((nd->nd_flag & ND_NFSV4) != 0) {
837 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
838 			i = fxdr_unsigned(int, *tl);
839 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
840 			if (error)
841 				goto nfsmout;
842 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
843 			opcnt = fxdr_unsigned(int, *tl++);
844 			i = fxdr_unsigned(int, *tl++);
845 			j = fxdr_unsigned(int, *tl);
846 			if (j >= 10000)
847 				NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
848 			/*
849 			 * If the first op is Sequence, free up the slot.
850 			 */
851 			if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
852 			    (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
853 				NFSCL_DEBUG(1, "failed seq=%d\n", j);
854 			if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
855 			    (clp != NULL && i == NFSV4OP_CBSEQUENCE && j == 0)
856 			    ) {
857 				if (i == NFSV4OP_SEQUENCE)
858 					NFSM_DISSECT(tl, uint32_t *,
859 					    NFSX_V4SESSIONID +
860 					    5 * NFSX_UNSIGNED);
861 				else
862 					NFSM_DISSECT(tl, uint32_t *,
863 					    NFSX_V4SESSIONID +
864 					    4 * NFSX_UNSIGNED);
865 				mtx_lock(&sep->nfsess_mtx);
866 				if (bcmp(tl, sep->nfsess_sessionid,
867 				    NFSX_V4SESSIONID) == 0) {
868 					tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
869 					retseq = fxdr_unsigned(uint32_t, *tl++);
870 					slot = fxdr_unsigned(int, *tl++);
871 					freeslot = slot;
872 					if (retseq != sep->nfsess_slotseq[slot])
873 						printf("retseq diff 0x%x\n",
874 						    retseq);
875 					retval = fxdr_unsigned(uint32_t, *++tl);
876 					if ((retval + 1) < sep->nfsess_foreslots
877 					    )
878 						sep->nfsess_foreslots = (retval
879 						    + 1);
880 					else if ((retval + 1) >
881 					    sep->nfsess_foreslots)
882 						sep->nfsess_foreslots = (retval
883 						    < 64) ? (retval + 1) : 64;
884 				}
885 				mtx_unlock(&sep->nfsess_mtx);
886 
887 				/* Grab the op and status for the next one. */
888 				if (opcnt > 1) {
889 					NFSM_DISSECT(tl, uint32_t *,
890 					    2 * NFSX_UNSIGNED);
891 					i = fxdr_unsigned(int, *tl++);
892 					j = fxdr_unsigned(int, *tl);
893 				}
894 			}
895 		}
896 		if (nd->nd_repstat != 0) {
897 			if (nd->nd_repstat == NFSERR_BADSESSION &&
898 			    nmp != NULL && dssep == NULL) {
899 				/*
900 				 * If this is a client side MDS RPC, mark
901 				 * the MDS session defunct and initiate
902 				 * recovery, as required.
903 				 * The nfsess_defunct field is protected by
904 				 * the NFSLOCKMNT()/nm_mtx lock and not the
905 				 * nfsess_mtx lock to simplify its handling,
906 				 * for the MDS session. This lock is also
907 				 * sufficient for nfsess_sessionid, since it
908 				 * never changes in the structure.
909 				 */
910 				NFSCL_DEBUG(1, "Got badsession\n");
911 				NFSLOCKCLSTATE();
912 				NFSLOCKMNT(nmp);
913 				sep = NFSMNT_MDSSESSION(nmp);
914 				if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
915 				    NFSX_V4SESSIONID) == 0) {
916 					/* Initiate recovery. */
917 					sep->nfsess_defunct = 1;
918 					NFSCL_DEBUG(1, "Marked defunct\n");
919 					if (nmp->nm_clp != NULL) {
920 						nmp->nm_clp->nfsc_flags |=
921 						    NFSCLFLAGS_RECOVER;
922 						wakeup(nmp->nm_clp);
923 					}
924 				}
925 				NFSUNLOCKCLSTATE();
926 				/*
927 				 * Sleep for up to 1sec waiting for a new
928 				 * session.
929 				 */
930 				mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
931 				    "nfsbadsess", hz);
932 				/*
933 				 * Get the session again, in case a new one
934 				 * has been created during the sleep.
935 				 */
936 				sep = NFSMNT_MDSSESSION(nmp);
937 				NFSUNLOCKMNT(nmp);
938 				if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
939 					reterr = nfsv4_sequencelookup(nmp, sep,
940 					    &slotpos, &maxslot, &slotseq,
941 					    sessionid);
942 					if (reterr == 0) {
943 						/* Fill in new session info. */
944 						NFSCL_DEBUG(1,
945 						  "Filling in new sequence\n");
946 						tl = nd->nd_sequence;
947 						bcopy(sessionid, tl,
948 						    NFSX_V4SESSIONID);
949 						tl += NFSX_V4SESSIONID /
950 						    NFSX_UNSIGNED;
951 						*tl++ = txdr_unsigned(slotseq);
952 						*tl++ = txdr_unsigned(slotpos);
953 						*tl = txdr_unsigned(maxslot);
954 					}
955 					if (reterr == NFSERR_BADSESSION ||
956 					    reterr == 0) {
957 						NFSCL_DEBUG(1,
958 						    "Badsession looping\n");
959 						m_freem(nd->nd_mrep);
960 						nd->nd_mrep = NULL;
961 						goto tryagain;
962 					}
963 					nd->nd_repstat = reterr;
964 					NFSCL_DEBUG(1, "Got err=%d\n", reterr);
965 				}
966 			}
967 			if (((nd->nd_repstat == NFSERR_DELAY ||
968 			      nd->nd_repstat == NFSERR_GRACE) &&
969 			     (nd->nd_flag & ND_NFSV4) &&
970 			     nd->nd_procnum != NFSPROC_DELEGRETURN &&
971 			     nd->nd_procnum != NFSPROC_SETATTR &&
972 			     nd->nd_procnum != NFSPROC_READ &&
973 			     nd->nd_procnum != NFSPROC_READDS &&
974 			     nd->nd_procnum != NFSPROC_WRITE &&
975 			     nd->nd_procnum != NFSPROC_WRITEDS &&
976 			     nd->nd_procnum != NFSPROC_OPEN &&
977 			     nd->nd_procnum != NFSPROC_CREATE &&
978 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
979 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
980 			     nd->nd_procnum != NFSPROC_CLOSE &&
981 			     nd->nd_procnum != NFSPROC_LOCK &&
982 			     nd->nd_procnum != NFSPROC_LOCKU) ||
983 			    (nd->nd_repstat == NFSERR_DELAY &&
984 			     (nd->nd_flag & ND_NFSV4) == 0) ||
985 			    nd->nd_repstat == NFSERR_RESOURCE) {
986 				if (trylater_delay > NFS_TRYLATERDEL)
987 					trylater_delay = NFS_TRYLATERDEL;
988 				waituntil = NFSD_MONOSEC + trylater_delay;
989 				while (NFSD_MONOSEC < waituntil)
990 					(void) nfs_catnap(PZERO, 0, "nfstry");
991 				trylater_delay *= 2;
992 				if (slot != -1) {
993 					mtx_lock(&sep->nfsess_mtx);
994 					sep->nfsess_slotseq[slot]++;
995 					*nd->nd_slotseq = txdr_unsigned(
996 					    sep->nfsess_slotseq[slot]);
997 					mtx_unlock(&sep->nfsess_mtx);
998 				}
999 				m_freem(nd->nd_mrep);
1000 				nd->nd_mrep = NULL;
1001 				goto tryagain;
1002 			}
1003 
1004 			/*
1005 			 * If the File Handle was stale, invalidate the
1006 			 * lookup cache, just in case.
1007 			 * (vp != NULL implies a client side call)
1008 			 */
1009 			if (nd->nd_repstat == ESTALE && vp != NULL) {
1010 				cache_purge(vp);
1011 				if (ncl_call_invalcaches != NULL)
1012 					(*ncl_call_invalcaches)(vp);
1013 			}
1014 		}
1015 		if ((nd->nd_flag & ND_NFSV4) != 0) {
1016 			/* Free the slot, as required. */
1017 			if (freeslot != -1)
1018 				nfsv4_freeslot(sep, freeslot);
1019 			/*
1020 			 * If this op is Putfh, throw its results away.
1021 			 */
1022 			if (j >= 10000)
1023 				NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1024 			if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1025 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1026 				i = fxdr_unsigned(int, *tl++);
1027 				j = fxdr_unsigned(int, *tl);
1028 				if (j >= 10000)
1029 					NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1030 					    j);
1031 				/*
1032 				 * All Compounds that do an Op that must
1033 				 * be in sequence consist of NFSV4OP_PUTFH
1034 				 * followed by one of these. As such, we
1035 				 * can determine if the seqid# should be
1036 				 * incremented, here.
1037 				 */
1038 				if ((i == NFSV4OP_OPEN ||
1039 				     i == NFSV4OP_OPENCONFIRM ||
1040 				     i == NFSV4OP_OPENDOWNGRADE ||
1041 				     i == NFSV4OP_CLOSE ||
1042 				     i == NFSV4OP_LOCK ||
1043 				     i == NFSV4OP_LOCKU) &&
1044 				    (j == 0 ||
1045 				     (j != NFSERR_STALECLIENTID &&
1046 				      j != NFSERR_STALESTATEID &&
1047 				      j != NFSERR_BADSTATEID &&
1048 				      j != NFSERR_BADSEQID &&
1049 				      j != NFSERR_BADXDR &&
1050 				      j != NFSERR_RESOURCE &&
1051 				      j != NFSERR_NOFILEHANDLE)))
1052 					nd->nd_flag |= ND_INCRSEQID;
1053 			}
1054 			/*
1055 			 * If this op's status is non-zero, mark
1056 			 * that there is no more data to process.
1057 			 * The exception is Setattr, which always has xdr
1058 			 * when it has failed.
1059 			 */
1060 			if (j != 0 && i != NFSV4OP_SETATTR)
1061 				nd->nd_flag |= ND_NOMOREDATA;
1062 
1063 			/*
1064 			 * If R_DONTRECOVER is set, replace the stale error
1065 			 * reply, so that recovery isn't initiated.
1066 			 */
1067 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1068 			     nd->nd_repstat == NFSERR_BADSESSION ||
1069 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
1070 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
1071 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1072 		}
1073 	}
1074 
1075 #ifdef KDTRACE_HOOKS
1076 	if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1077 		uint32_t probe_id;
1078 		int probe_procnum;
1079 
1080 		if (nd->nd_flag & ND_NFSV4) {
1081 			probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1082 			probe_procnum = nd->nd_procnum;
1083 		} else if (nd->nd_flag & ND_NFSV3) {
1084 			probe_id = nfscl_nfs3_done_probes[procnum];
1085 			probe_procnum = procnum;
1086 		} else {
1087 			probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1088 			probe_procnum = procnum;
1089 		}
1090 		if (probe_id != 0)
1091 			(dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1092 			    nd->nd_mreq, cred, probe_procnum, 0);
1093 	}
1094 #endif
1095 
1096 	m_freem(nd->nd_mreq);
1097 	if (usegssname == 0)
1098 		AUTH_DESTROY(auth);
1099 	if (rep != NULL)
1100 		FREE((caddr_t)rep, M_NFSDREQ);
1101 	if (set_sigset)
1102 		newnfs_restore_sigmask(td, &oldset);
1103 	return (0);
1104 nfsmout:
1105 	mbuf_freem(nd->nd_mrep);
1106 	mbuf_freem(nd->nd_mreq);
1107 	if (usegssname == 0)
1108 		AUTH_DESTROY(auth);
1109 	if (rep != NULL)
1110 		FREE((caddr_t)rep, M_NFSDREQ);
1111 	if (set_sigset)
1112 		newnfs_restore_sigmask(td, &oldset);
1113 	return (error);
1114 }
1115 
1116 /*
1117  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1118  * wait for all requests to complete. This is used by forced unmounts
1119  * to terminate any outstanding RPCs.
1120  */
1121 int
1122 newnfs_nmcancelreqs(struct nfsmount *nmp)
1123 {
1124 
1125 	if (nmp->nm_sockreq.nr_client != NULL)
1126 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1127 	return (0);
1128 }
1129 
1130 /*
1131  * Any signal that can interrupt an NFS operation in an intr mount
1132  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1133  */
1134 int newnfs_sig_set[] = {
1135 	SIGINT,
1136 	SIGTERM,
1137 	SIGHUP,
1138 	SIGKILL,
1139 	SIGQUIT
1140 };
1141 
1142 /*
1143  * Check to see if one of the signals in our subset is pending on
1144  * the process (in an intr mount).
1145  */
1146 static int
1147 nfs_sig_pending(sigset_t set)
1148 {
1149 	int i;
1150 
1151 	for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1152 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
1153 			return (1);
1154 	return (0);
1155 }
1156 
1157 /*
1158  * The set/restore sigmask functions are used to (temporarily) overwrite
1159  * the thread td_sigmask during an RPC call (for example). These are also
1160  * used in other places in the NFS client that might tsleep().
1161  */
1162 void
1163 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1164 {
1165 	sigset_t newset;
1166 	int i;
1167 	struct proc *p;
1168 
1169 	SIGFILLSET(newset);
1170 	if (td == NULL)
1171 		td = curthread; /* XXX */
1172 	p = td->td_proc;
1173 	/* Remove the NFS set of signals from newset */
1174 	PROC_LOCK(p);
1175 	mtx_lock(&p->p_sigacts->ps_mtx);
1176 	for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1177 		/*
1178 		 * But make sure we leave the ones already masked
1179 		 * by the process, ie. remove the signal from the
1180 		 * temporary signalmask only if it wasn't already
1181 		 * in p_sigmask.
1182 		 */
1183 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1184 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1185 			SIGDELSET(newset, newnfs_sig_set[i]);
1186 	}
1187 	mtx_unlock(&p->p_sigacts->ps_mtx);
1188 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1189 	    SIGPROCMASK_PROC_LOCKED);
1190 	PROC_UNLOCK(p);
1191 }
1192 
1193 void
1194 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1195 {
1196 	if (td == NULL)
1197 		td = curthread; /* XXX */
1198 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1199 }
1200 
1201 /*
1202  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1203  * old one after msleep() returns.
1204  */
1205 int
1206 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1207 {
1208 	sigset_t oldset;
1209 	int error;
1210 	struct proc *p;
1211 
1212 	if ((priority & PCATCH) == 0)
1213 		return msleep(ident, mtx, priority, wmesg, timo);
1214 	if (td == NULL)
1215 		td = curthread; /* XXX */
1216 	newnfs_set_sigmask(td, &oldset);
1217 	error = msleep(ident, mtx, priority, wmesg, timo);
1218 	newnfs_restore_sigmask(td, &oldset);
1219 	p = td->td_proc;
1220 	return (error);
1221 }
1222 
1223 /*
1224  * Test for a termination condition pending on the process.
1225  * This is used for NFSMNT_INT mounts.
1226  */
1227 int
1228 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1229 {
1230 	struct proc *p;
1231 	sigset_t tmpset;
1232 
1233 	/* Terminate all requests while attempting a forced unmount. */
1234 	if (NFSCL_FORCEDISM(nmp->nm_mountp))
1235 		return (EIO);
1236 	if (!(nmp->nm_flag & NFSMNT_INT))
1237 		return (0);
1238 	if (td == NULL)
1239 		return (0);
1240 	p = td->td_proc;
1241 	PROC_LOCK(p);
1242 	tmpset = p->p_siglist;
1243 	SIGSETOR(tmpset, td->td_siglist);
1244 	SIGSETNAND(tmpset, td->td_sigmask);
1245 	mtx_lock(&p->p_sigacts->ps_mtx);
1246 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1247 	mtx_unlock(&p->p_sigacts->ps_mtx);
1248 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1249 	    && nfs_sig_pending(tmpset)) {
1250 		PROC_UNLOCK(p);
1251 		return (EINTR);
1252 	}
1253 	PROC_UNLOCK(p);
1254 	return (0);
1255 }
1256 
1257 static int
1258 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1259 {
1260 	struct proc *p;
1261 
1262 	p = td ? td->td_proc : NULL;
1263 	if (error) {
1264 		tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1265 		    server, msg, error);
1266 	} else {
1267 		tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1268 	}
1269 	return (0);
1270 }
1271 
1272 static void
1273 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1274     int error, int flags)
1275 {
1276 	if (nmp == NULL)
1277 		return;
1278 	mtx_lock(&nmp->nm_mtx);
1279 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1280 		nmp->nm_state |= NFSSTA_TIMEO;
1281 		mtx_unlock(&nmp->nm_mtx);
1282 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1283 		    VQ_NOTRESP, 0);
1284 	} else
1285 		mtx_unlock(&nmp->nm_mtx);
1286 	mtx_lock(&nmp->nm_mtx);
1287 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1288 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
1289 		mtx_unlock(&nmp->nm_mtx);
1290 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1291 		    VQ_NOTRESPLOCK, 0);
1292 	} else
1293 		mtx_unlock(&nmp->nm_mtx);
1294 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1295 }
1296 
1297 static void
1298 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1299     int flags, int tprintfmsg)
1300 {
1301 	if (nmp == NULL)
1302 		return;
1303 	if (tprintfmsg) {
1304 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1305 	}
1306 
1307 	mtx_lock(&nmp->nm_mtx);
1308 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1309 		nmp->nm_state &= ~NFSSTA_TIMEO;
1310 		mtx_unlock(&nmp->nm_mtx);
1311 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1312 		    VQ_NOTRESP, 1);
1313 	} else
1314 		mtx_unlock(&nmp->nm_mtx);
1315 
1316 	mtx_lock(&nmp->nm_mtx);
1317 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1318 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1319 		mtx_unlock(&nmp->nm_mtx);
1320 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1321 		    VQ_NOTRESPLOCK, 1);
1322 	} else
1323 		mtx_unlock(&nmp->nm_mtx);
1324 }
1325 
1326