xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision 942815c54820783d3d4f7f6faa71ab7919b5f0e5)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1991, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 /*
40  * Socket operations for use by nfs
41  */
42 
43 #include "opt_kgssapi.h"
44 #include "opt_nfs.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/limits.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/signalvar.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysctl.h>
59 #include <sys/syslog.h>
60 #include <sys/vnode.h>
61 
62 #include <rpc/rpc.h>
63 #include <rpc/krpc.h>
64 
65 #include <kgssapi/krb5/kcrypto.h>
66 
67 #include <fs/nfs/nfsport.h>
68 
69 #ifdef KDTRACE_HOOKS
70 #include <sys/dtrace_bsd.h>
71 
72 dtrace_nfsclient_nfs23_start_probe_func_t
73 		dtrace_nfscl_nfs234_start_probe;
74 
75 dtrace_nfsclient_nfs23_done_probe_func_t
76 		dtrace_nfscl_nfs234_done_probe;
77 
78 /*
79  * Registered probes by RPC type.
80  */
81 uint32_t	nfscl_nfs2_start_probes[NFSV41_NPROCS + 1];
82 uint32_t	nfscl_nfs2_done_probes[NFSV41_NPROCS + 1];
83 
84 uint32_t	nfscl_nfs3_start_probes[NFSV41_NPROCS + 1];
85 uint32_t	nfscl_nfs3_done_probes[NFSV41_NPROCS + 1];
86 
87 uint32_t	nfscl_nfs4_start_probes[NFSV41_NPROCS + 1];
88 uint32_t	nfscl_nfs4_done_probes[NFSV41_NPROCS + 1];
89 #endif
90 
91 NFSSTATESPINLOCK;
92 NFSREQSPINLOCK;
93 NFSDLOCKMUTEX;
94 NFSCLSTATEMUTEX;
95 extern struct nfsreqhead nfsd_reqq;
96 extern int nfscl_ticks;
97 extern void (*ncl_call_invalcaches)(struct vnode *);
98 extern int nfs_numnfscbd;
99 extern int nfscl_debuglevel;
100 extern int nfsrv_lease;
101 
102 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
103 
104 SVCPOOL		*nfscbd_pool;
105 int		nfs_bufpackets = 4;
106 static int	nfsrv_gsscallbackson = 0;
107 static int	nfs_reconnects;
108 static int	nfs3_jukebox_delay = 10;
109 static int	nfs_skip_wcc_data_onerr = 1;
110 static int	nfs_dsretries = 2;
111 static struct timespec	nfs_trylater_max = {
112 	.tv_sec		= NFS_TRYLATERDEL,
113 	.tv_nsec	= 0,
114 };
115 
116 SYSCTL_DECL(_vfs_nfs);
117 
118 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
119     "Buffer reservation size 2 < x < 64");
120 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
121     "Number of times the nfs client has had to reconnect");
122 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
123     "Number of seconds to delay a retry after receiving EJUKEBOX");
124 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
125     "Disable weak cache consistency checking when server returns an error");
126 SYSCTL_INT(_vfs_nfs, OID_AUTO, dsretries, CTLFLAG_RW, &nfs_dsretries, 0,
127     "Number of retries for a DS RPC before failure");
128 
129 static void	nfs_down(struct nfsmount *, struct thread *, const char *,
130     int, int);
131 static void	nfs_up(struct nfsmount *, struct thread *, const char *,
132     int, int);
133 static int	nfs_msg(struct thread *, const char *, const char *, int);
134 
135 struct nfs_cached_auth {
136 	int		ca_refs; /* refcount, including 1 from the cache */
137 	uid_t		ca_uid;	 /* uid that corresponds to this auth */
138 	AUTH		*ca_auth; /* RPC auth handle */
139 };
140 
141 static int nfsv2_procid[NFS_V3NPROCS] = {
142 	NFSV2PROC_NULL,
143 	NFSV2PROC_GETATTR,
144 	NFSV2PROC_SETATTR,
145 	NFSV2PROC_LOOKUP,
146 	NFSV2PROC_NOOP,
147 	NFSV2PROC_READLINK,
148 	NFSV2PROC_READ,
149 	NFSV2PROC_WRITE,
150 	NFSV2PROC_CREATE,
151 	NFSV2PROC_MKDIR,
152 	NFSV2PROC_SYMLINK,
153 	NFSV2PROC_CREATE,
154 	NFSV2PROC_REMOVE,
155 	NFSV2PROC_RMDIR,
156 	NFSV2PROC_RENAME,
157 	NFSV2PROC_LINK,
158 	NFSV2PROC_READDIR,
159 	NFSV2PROC_NOOP,
160 	NFSV2PROC_STATFS,
161 	NFSV2PROC_NOOP,
162 	NFSV2PROC_NOOP,
163 	NFSV2PROC_NOOP,
164 };
165 
166 /*
167  * Initialize sockets and congestion for a new NFS connection.
168  * We do not free the sockaddr if error.
169  * Which arguments are set to NULL indicate what kind of call it is.
170  * cred == NULL --> a call to connect to a pNFS DS
171  * nmp == NULL --> indicates an upcall to userland or a NFSv4.0 callback
172  */
173 int
174 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
175     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult, bool dotls,
176     struct __rpc_client **clipp)
177 {
178 	int rcvreserve, sndreserve;
179 	int pktscale, pktscalesav;
180 	struct sockaddr *saddr;
181 	struct ucred *origcred;
182 	CLIENT *client;
183 	struct netconfig *nconf;
184 	struct socket *so;
185 	int one = 1, retries, error = 0;
186 	struct thread *td = curthread;
187 	SVCXPRT *xprt;
188 	struct timeval timo;
189 	uint64_t tval;
190 
191 	/*
192 	 * We need to establish the socket using the credentials of
193 	 * the mountpoint.  Some parts of this process (such as
194 	 * sobind() and soconnect()) will use the curent thread's
195 	 * credential instead of the socket credential.  To work
196 	 * around this, temporarily change the current thread's
197 	 * credential to that of the mountpoint.
198 	 *
199 	 * XXX: It would be better to explicitly pass the correct
200 	 * credential to sobind() and soconnect().
201 	 */
202 	origcred = td->td_ucred;
203 
204 	/*
205 	 * Use the credential in nr_cred, if not NULL.
206 	 */
207 	if (nrp->nr_cred != NULL)
208 		td->td_ucred = nrp->nr_cred;
209 	else
210 		td->td_ucred = cred;
211 	saddr = nrp->nr_nam;
212 
213 	if (saddr->sa_family == AF_INET)
214 		if (nrp->nr_sotype == SOCK_DGRAM)
215 			nconf = getnetconfigent("udp");
216 		else
217 			nconf = getnetconfigent("tcp");
218 	else
219 		if (nrp->nr_sotype == SOCK_DGRAM)
220 			nconf = getnetconfigent("udp6");
221 		else
222 			nconf = getnetconfigent("tcp6");
223 
224 	pktscale = nfs_bufpackets;
225 	if (pktscale < 2)
226 		pktscale = 2;
227 	if (pktscale > 64)
228 		pktscale = 64;
229 	pktscalesav = pktscale;
230 	/*
231 	 * soreserve() can fail if sb_max is too small, so shrink pktscale
232 	 * and try again if there is an error.
233 	 * Print a log message suggesting increasing sb_max.
234 	 * Creating a socket and doing this is necessary since, if the
235 	 * reservation sizes are too large and will make soreserve() fail,
236 	 * the connection will work until a large send is attempted and
237 	 * then it will loop in the krpc code.
238 	 */
239 	so = NULL;
240 	saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
241 	error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
242 	    nrp->nr_soproto, td->td_ucred, td);
243 	if (error != 0)
244 		goto out;
245 	do {
246 	    if (error != 0 && pktscale > 2) {
247 		if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
248 		    pktscale == pktscalesav) {
249 		    /*
250 		     * Suggest vfs.nfs.bufpackets * maximum RPC message,
251 		     * adjusted for the sb_max->sb_max_adj conversion of
252 		     * MCLBYTES / (MSIZE + MCLBYTES) as the minimum setting
253 		     * for kern.ipc.maxsockbuf.
254 		     */
255 		    tval = (NFS_MAXBSIZE + NFS_MAXXDR) * nfs_bufpackets;
256 		    tval *= MSIZE + MCLBYTES;
257 		    tval += MCLBYTES - 1; /* Round up divide by MCLBYTES. */
258 		    tval /= MCLBYTES;
259 		    printf("Consider increasing kern.ipc.maxsockbuf to a "
260 			"minimum of %ju to support %ubyte NFS I/O\n",
261 			(uintmax_t)tval, NFS_MAXBSIZE);
262 		}
263 		pktscale--;
264 	    }
265 	    if (nrp->nr_sotype == SOCK_DGRAM) {
266 		if (nmp != NULL) {
267 			sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
268 			    pktscale;
269 			rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
270 			    pktscale;
271 		} else {
272 			sndreserve = rcvreserve = 1024 * pktscale;
273 		}
274 	    } else {
275 		if (nrp->nr_sotype != SOCK_STREAM)
276 			panic("nfscon sotype");
277 		if (nmp != NULL) {
278 			sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR) *
279 			    pktscale;
280 			rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR) *
281 			    pktscale;
282 		} else {
283 			sndreserve = rcvreserve = 1024 * pktscale;
284 		}
285 	    }
286 	    error = soreserve(so, sndreserve, rcvreserve);
287 	    if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
288 		pktscale <= 2)
289 		printf("Must increase kern.ipc.maxsockbuf or reduce"
290 		    " rsize, wsize\n");
291 	} while (error != 0 && pktscale > 2);
292 	soclose(so);
293 	if (error != 0)
294 		goto out;
295 
296 	client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
297 	    nrp->nr_vers, sndreserve, rcvreserve);
298 	CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
299 	if (nmp != NULL) {
300 		if ((nmp->nm_flag & NFSMNT_INT))
301 			CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
302 		if ((nmp->nm_flag & NFSMNT_RESVPORT))
303 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
304 		if (NFSHASTLS(nmp)) {
305 			CLNT_CONTROL(client, CLSET_TLS, &one);
306 			if (nmp->nm_tlscertname != NULL)
307 				CLNT_CONTROL(client, CLSET_TLSCERTNAME,
308 				    nmp->nm_tlscertname);
309 		}
310 		if (NFSHASSOFT(nmp)) {
311 			if (nmp->nm_sotype == SOCK_DGRAM)
312 				/*
313 				 * For UDP, the large timeout for a reconnect
314 				 * will be set to "nm_retry * nm_timeo / 2", so
315 				 * we only want to do 2 reconnect timeout
316 				 * retries.
317 				 */
318 				retries = 2;
319 			else
320 				retries = nmp->nm_retry;
321 		} else
322 			retries = INT_MAX;
323 		if (NFSHASNFSV4N(nmp)) {
324 			if (cred != NULL) {
325 				if (NFSHASSOFT(nmp)) {
326 					/*
327 					 * This should be a DS mount.
328 					 * Use CLSET_TIMEOUT to set the timeout
329 					 * for connections to DSs instead of
330 					 * specifying a timeout on each RPC.
331 					 * This is done so that SO_SNDTIMEO
332 					 * is set on the TCP socket as well
333 					 * as specifying a time limit when
334 					 * waiting for an RPC reply.  Useful
335 					 * if the send queue for the TCP
336 					 * connection has become constipated,
337 					 * due to a failed DS.
338 					 * The choice of lease_duration / 4 is
339 					 * fairly arbitrary, but seems to work
340 					 * ok, with a lower bound of 10sec.
341 					 */
342 					timo.tv_sec = nfsrv_lease / 4;
343 					if (timo.tv_sec < 10)
344 						timo.tv_sec = 10;
345 					timo.tv_usec = 0;
346 					CLNT_CONTROL(client, CLSET_TIMEOUT,
347 					    &timo);
348 				}
349 				/*
350 				 * Make sure the nfscbd_pool doesn't get
351 				 * destroyed while doing this.
352 				 */
353 				NFSD_LOCK();
354 				if (nfs_numnfscbd > 0) {
355 					nfs_numnfscbd++;
356 					NFSD_UNLOCK();
357 					xprt = svc_vc_create_backchannel(
358 					    nfscbd_pool);
359 					CLNT_CONTROL(client, CLSET_BACKCHANNEL,
360 					    xprt);
361 					NFSD_LOCK();
362 					nfs_numnfscbd--;
363 					if (nfs_numnfscbd == 0)
364 						wakeup(&nfs_numnfscbd);
365 				}
366 				NFSD_UNLOCK();
367 			} else {
368 				/*
369 				 * cred == NULL for a DS connect.
370 				 * For connects to a DS, set a retry limit
371 				 * so that failed DSs will be detected.
372 				 * This is ok for NFSv4.1, since a DS does
373 				 * not maintain open/lock state and is the
374 				 * only case where using a "soft" mount is
375 				 * recommended for NFSv4.
376 				 * For mounts from the MDS to DS, this is done
377 				 * via mount options, but that is not the case
378 				 * here.  The retry limit here can be adjusted
379 				 * via the sysctl vfs.nfs.dsretries.
380 				 * See the comment above w.r.t. timeout.
381 				 */
382 				timo.tv_sec = nfsrv_lease / 4;
383 				if (timo.tv_sec < 10)
384 					timo.tv_sec = 10;
385 				timo.tv_usec = 0;
386 				CLNT_CONTROL(client, CLSET_TIMEOUT, &timo);
387 				retries = nfs_dsretries;
388 			}
389 		}
390 	} else {
391 		/*
392 		 * Three cases:
393 		 * - Null RPC callback to client
394 		 * - Non-Null RPC callback to client, wait a little longer
395 		 * - upcalls to nfsuserd and gssd (clp == NULL)
396 		 */
397 		if (callback_retry_mult == 0) {
398 			retries = NFSV4_UPCALLRETRY;
399 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
400 		} else {
401 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
402 		}
403 		if (dotls)
404 			CLNT_CONTROL(client, CLSET_TLS, &one);
405 	}
406 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
407 
408 	if (nmp != NULL) {
409 		/*
410 		 * For UDP, there are 2 timeouts:
411 		 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
412 		 *   that does a retransmit of an RPC request using the same
413 		 *   socket and xid. This is what you normally want to do,
414 		 *   since NFS servers depend on "same xid" for their
415 		 *   Duplicate Request Cache.
416 		 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
417 		 *   retransmits on the same socket should fail and a fresh
418 		 *   socket created. Each of these timeouts counts as one
419 		 *   CLSET_RETRIES as set above.
420 		 * Set the initial retransmit timeout for UDP. This timeout
421 		 * doesn't exist for TCP and the following call just fails,
422 		 * which is ok.
423 		 */
424 		timo.tv_sec = nmp->nm_timeo / NFS_HZ;
425 		timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
426 		CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
427 	}
428 
429 	/*
430 	 * *clipp is &nrp->nr_client or &nm_aconn[nmp->nm_nextaconn].
431 	 * The latter case is for additional connections specified by the
432 	 * "nconnect" mount option.  nr_mtx etc is used for these additional
433 	 * connections, as well as nr_client in the nfssockreq
434 	 * structure for the mount.
435 	 */
436 	mtx_lock(&nrp->nr_mtx);
437 	if (*clipp != NULL) {
438 		mtx_unlock(&nrp->nr_mtx);
439 		/*
440 		 * Someone else already connected.
441 		 */
442 		CLNT_RELEASE(client);
443 	} else {
444 		*clipp = client;
445 		/*
446 		 * Protocols that do not require connections may be optionally
447 		 * left unconnected for servers that reply from a port other
448 		 * than NFS_PORT.
449 		 */
450 		if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
451 			mtx_unlock(&nrp->nr_mtx);
452 			CLNT_CONTROL(client, CLSET_CONNECT, &one);
453 		} else
454 			mtx_unlock(&nrp->nr_mtx);
455 	}
456 
457 out:
458 	/* Restore current thread's credentials. */
459 	td->td_ucred = origcred;
460 
461 	NFSEXITCODE(error);
462 	return (error);
463 }
464 
465 /*
466  * NFS disconnect. Clean up and unlink.
467  */
468 void
469 newnfs_disconnect(struct nfsmount *nmp, struct nfssockreq *nrp)
470 {
471 	CLIENT *client, *aconn[NFS_MAXNCONN - 1];
472 	int i;
473 
474 	mtx_lock(&nrp->nr_mtx);
475 	if (nrp->nr_client != NULL) {
476 		client = nrp->nr_client;
477 		nrp->nr_client = NULL;
478 		if (nmp != NULL && nmp->nm_aconnect > 0) {
479 			for (i = 0; i < nmp->nm_aconnect; i++) {
480 				aconn[i] = nmp->nm_aconn[i];
481 				nmp->nm_aconn[i] = NULL;
482 			}
483 		}
484 		mtx_unlock(&nrp->nr_mtx);
485 		rpc_gss_secpurge_call(client);
486 		CLNT_CLOSE(client);
487 		CLNT_RELEASE(client);
488 		if (nmp != NULL && nmp->nm_aconnect > 0) {
489 			for (i = 0; i < nmp->nm_aconnect; i++) {
490 				if (aconn[i] != NULL) {
491 					rpc_gss_secpurge_call(aconn[i]);
492 					CLNT_CLOSE(aconn[i]);
493 					CLNT_RELEASE(aconn[i]);
494 				}
495 			}
496 		}
497 	} else {
498 		mtx_unlock(&nrp->nr_mtx);
499 	}
500 }
501 
502 static AUTH *
503 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
504     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
505 {
506 	rpc_gss_service_t svc;
507 	AUTH *auth;
508 
509 	switch (secflavour) {
510 	case RPCSEC_GSS_KRB5:
511 	case RPCSEC_GSS_KRB5I:
512 	case RPCSEC_GSS_KRB5P:
513 		if (!mech_oid) {
514 			if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
515 				return (NULL);
516 		}
517 		if (secflavour == RPCSEC_GSS_KRB5)
518 			svc = rpc_gss_svc_none;
519 		else if (secflavour == RPCSEC_GSS_KRB5I)
520 			svc = rpc_gss_svc_integrity;
521 		else
522 			svc = rpc_gss_svc_privacy;
523 
524 		if (clnt_principal == NULL)
525 			auth = rpc_gss_secfind_call(nrp->nr_client, cred,
526 			    srv_principal, mech_oid, svc);
527 		else {
528 			auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
529 			    clnt_principal, srv_principal, "kerberosv5",
530 			    svc, NULL, NULL, NULL);
531 			return (auth);
532 		}
533 		if (auth != NULL)
534 			return (auth);
535 		/* fallthrough */
536 	case AUTH_SYS:
537 	default:
538 		return (authunix_create(cred));
539 	}
540 }
541 
542 /*
543  * Callback from the RPC code to generate up/down notifications.
544  */
545 
546 struct nfs_feedback_arg {
547 	struct nfsmount *nf_mount;
548 	int		nf_lastmsg;	/* last tprintf */
549 	int		nf_tprintfmsg;
550 	struct thread	*nf_td;
551 };
552 
553 static void
554 nfs_feedback(int type, int proc, void *arg)
555 {
556 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
557 	struct nfsmount *nmp = nf->nf_mount;
558 	time_t now;
559 
560 	switch (type) {
561 	case FEEDBACK_REXMIT2:
562 	case FEEDBACK_RECONNECT:
563 		now = NFSD_MONOSEC;
564 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
565 			nfs_down(nmp, nf->nf_td,
566 			    "not responding", 0, NFSSTA_TIMEO);
567 			nf->nf_tprintfmsg = TRUE;
568 			nf->nf_lastmsg = now;
569 		}
570 		break;
571 
572 	case FEEDBACK_OK:
573 		nfs_up(nf->nf_mount, nf->nf_td,
574 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
575 		break;
576 	}
577 }
578 
579 /*
580  * newnfs_request - goes something like this
581  *	- does the rpc by calling the krpc layer
582  *	- break down rpc header and return with nfs reply
583  * nb: always frees up nd_mreq mbuf list
584  */
585 int
586 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
587     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
588     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
589     u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
590 {
591 	uint32_t retseq, retval, slotseq, *tl;
592 	int i = 0, j = 0, opcnt, set_sigset = 0, slot;
593 	int error = 0, usegssname = 0, secflavour = AUTH_SYS;
594 	int freeslot, maxslot, reterr, slotpos, timeo;
595 	u_int16_t procnum;
596 	u_int nextconn;
597 	struct nfs_feedback_arg nf;
598 	struct timeval timo;
599 	AUTH *auth;
600 	struct rpc_callextra ext;
601 	enum clnt_stat stat;
602 	struct nfsreq *rep = NULL;
603 	char *srv_principal = NULL, *clnt_principal = NULL;
604 	sigset_t oldset;
605 	struct ucred *authcred;
606 	struct nfsclsession *sep;
607 	uint8_t sessionid[NFSX_V4SESSIONID];
608 	bool nextconn_set;
609 	struct timespec trylater_delay, ts, waituntil;
610 
611 	/* Initially 1msec. */
612 	trylater_delay.tv_sec = 0;
613 	trylater_delay.tv_nsec = 1000000;
614 	sep = dssep;
615 	if (xidp != NULL)
616 		*xidp = 0;
617 	/* Reject requests while attempting a forced unmount. */
618 	if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
619 		m_freem(nd->nd_mreq);
620 		return (ESTALE);
621 	}
622 
623 	/*
624 	 * Set authcred, which is used to acquire RPC credentials to
625 	 * the cred argument, by default. The crhold() should not be
626 	 * necessary, but will ensure that some future code change
627 	 * doesn't result in the credential being free'd prematurely.
628 	 */
629 	authcred = crhold(cred);
630 
631 	/* For client side interruptible mounts, mask off the signals. */
632 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
633 		newnfs_set_sigmask(td, &oldset);
634 		set_sigset = 1;
635 	}
636 
637 	/*
638 	 * If not already connected call newnfs_connect now.
639 	 */
640 	if (nrp->nr_client == NULL)
641 		newnfs_connect(nmp, nrp, cred, td, 0, false, &nrp->nr_client);
642 
643 	/*
644 	 * If the "nconnect" mount option was specified and this RPC is
645 	 * one that can have a large RPC message and is being done through
646 	 * the NFS/MDS server, use an additional connection. (When the RPC is
647 	 * being done through the server/MDS, nrp == &nmp->nm_sockreq.)
648 	 * The "nconnect" mount option normally has minimal effect when the
649 	 * "pnfs" mount option is specified, since only Readdir RPCs are
650 	 * normally done through the NFS/MDS server.
651 	 */
652 	nextconn_set = false;
653 	if (nmp != NULL && nmp->nm_aconnect > 0 && nrp == &nmp->nm_sockreq &&
654 	    (nd->nd_procnum == NFSPROC_READ ||
655 	     nd->nd_procnum == NFSPROC_READDIR ||
656 	     nd->nd_procnum == NFSPROC_READDIRPLUS ||
657 	     nd->nd_procnum == NFSPROC_WRITE)) {
658 		nextconn = atomic_fetchadd_int(&nmp->nm_nextaconn, 1);
659 		nextconn %= nmp->nm_aconnect;
660 		nextconn_set = true;
661 		if (nmp->nm_aconn[nextconn] == NULL)
662 			newnfs_connect(nmp, nrp, cred, td, 0, false,
663 			    &nmp->nm_aconn[nextconn]);
664 	}
665 
666 	/*
667 	 * For a client side mount, nmp is != NULL and clp == NULL. For
668 	 * server calls (callbacks or upcalls), nmp == NULL.
669 	 */
670 	if (clp != NULL) {
671 		NFSLOCKSTATE();
672 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
673 			secflavour = RPCSEC_GSS_KRB5;
674 			if (nd->nd_procnum != NFSPROC_NULL) {
675 				if (clp->lc_flags & LCL_GSSINTEGRITY)
676 					secflavour = RPCSEC_GSS_KRB5I;
677 				else if (clp->lc_flags & LCL_GSSPRIVACY)
678 					secflavour = RPCSEC_GSS_KRB5P;
679 			}
680 		}
681 		NFSUNLOCKSTATE();
682 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
683 	     nd->nd_procnum != NFSPROC_NULL) {
684 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
685 			nd->nd_flag |= ND_USEGSSNAME;
686 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
687 			/*
688 			 * If there is a client side host based credential,
689 			 * use that, otherwise use the system uid, if set.
690 			 * The system uid is in the nmp->nm_sockreq.nr_cred
691 			 * credentials.
692 			 */
693 			if (nmp->nm_krbnamelen > 0) {
694 				usegssname = 1;
695 				clnt_principal = nmp->nm_krbname;
696 			} else if (nmp->nm_uid != (uid_t)-1) {
697 				KASSERT(nmp->nm_sockreq.nr_cred != NULL,
698 				    ("newnfs_request: NULL nr_cred"));
699 				crfree(authcred);
700 				authcred = crhold(nmp->nm_sockreq.nr_cred);
701 			}
702 		} else if (nmp->nm_krbnamelen == 0 &&
703 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
704 			/*
705 			 * If there is no host based principal name and
706 			 * the system uid is set and this is root, use the
707 			 * system uid, since root won't have user
708 			 * credentials in a credentials cache file.
709 			 * The system uid is in the nmp->nm_sockreq.nr_cred
710 			 * credentials.
711 			 */
712 			KASSERT(nmp->nm_sockreq.nr_cred != NULL,
713 			    ("newnfs_request: NULL nr_cred"));
714 			crfree(authcred);
715 			authcred = crhold(nmp->nm_sockreq.nr_cred);
716 		}
717 		if (NFSHASINTEGRITY(nmp))
718 			secflavour = RPCSEC_GSS_KRB5I;
719 		else if (NFSHASPRIVACY(nmp))
720 			secflavour = RPCSEC_GSS_KRB5P;
721 		else
722 			secflavour = RPCSEC_GSS_KRB5;
723 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
724 	} else if (nmp != NULL && !NFSHASKERB(nmp) &&
725 	    nd->nd_procnum != NFSPROC_NULL &&
726 	    (nd->nd_flag & ND_USEGSSNAME) != 0) {
727 		/*
728 		 * Use the uid that did the mount when the RPC is doing
729 		 * NFSv4 system operations, as indicated by the
730 		 * ND_USEGSSNAME flag, for the AUTH_SYS case.
731 		 * The credentials in nm_sockreq.nr_cred were used for the
732 		 * mount.
733 		 */
734 		KASSERT(nmp->nm_sockreq.nr_cred != NULL,
735 		    ("newnfs_request: NULL nr_cred"));
736 		crfree(authcred);
737 		authcred = crhold(nmp->nm_sockreq.nr_cred);
738 	}
739 
740 	if (nmp != NULL) {
741 		bzero(&nf, sizeof(struct nfs_feedback_arg));
742 		nf.nf_mount = nmp;
743 		nf.nf_td = td;
744 		nf.nf_lastmsg = NFSD_MONOSEC -
745 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
746 	}
747 
748 	if (nd->nd_procnum == NFSPROC_NULL)
749 		auth = authnone_create();
750 	else if (usegssname) {
751 		/*
752 		 * For this case, the authenticator is held in the
753 		 * nfssockreq structure, so don't release the reference count
754 		 * held on it. --> Don't AUTH_DESTROY() it in this function.
755 		 */
756 		if (nrp->nr_auth == NULL)
757 			nrp->nr_auth = nfs_getauth(nrp, secflavour,
758 			    clnt_principal, srv_principal, NULL, authcred);
759 		else
760 			rpc_gss_refresh_auth_call(nrp->nr_auth);
761 		auth = nrp->nr_auth;
762 	} else
763 		auth = nfs_getauth(nrp, secflavour, NULL,
764 		    srv_principal, NULL, authcred);
765 	crfree(authcred);
766 	if (auth == NULL) {
767 		m_freem(nd->nd_mreq);
768 		if (set_sigset)
769 			newnfs_restore_sigmask(td, &oldset);
770 		return (EACCES);
771 	}
772 	bzero(&ext, sizeof(ext));
773 	ext.rc_auth = auth;
774 	if (nmp != NULL) {
775 		ext.rc_feedback = nfs_feedback;
776 		ext.rc_feedback_arg = &nf;
777 	}
778 
779 	procnum = nd->nd_procnum;
780 	if ((nd->nd_flag & ND_NFSV4) &&
781 	    nd->nd_procnum != NFSPROC_NULL &&
782 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
783 		procnum = NFSV4PROC_COMPOUND;
784 
785 	if (nmp != NULL) {
786 		NFSINCRGLOBAL(nfsstatsv1_p->rpcrequests);
787 
788 		/* Map the procnum to the old NFSv2 one, as required. */
789 		if ((nd->nd_flag & ND_NFSV2) != 0) {
790 			if (nd->nd_procnum < NFS_V3NPROCS)
791 				procnum = nfsv2_procid[nd->nd_procnum];
792 			else
793 				procnum = NFSV2PROC_NOOP;
794 		}
795 
796 		/*
797 		 * Now only used for the R_DONTRECOVER case, but until that is
798 		 * supported within the krpc code, I need to keep a queue of
799 		 * outstanding RPCs for nfsv4 client requests.
800 		 */
801 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
802 			rep = malloc(sizeof(struct nfsreq),
803 			    M_NFSDREQ, M_WAITOK);
804 #ifdef KDTRACE_HOOKS
805 		if (dtrace_nfscl_nfs234_start_probe != NULL) {
806 			uint32_t probe_id;
807 			int probe_procnum;
808 
809 			if (nd->nd_flag & ND_NFSV4) {
810 				probe_id =
811 				    nfscl_nfs4_start_probes[nd->nd_procnum];
812 				probe_procnum = nd->nd_procnum;
813 			} else if (nd->nd_flag & ND_NFSV3) {
814 				probe_id = nfscl_nfs3_start_probes[procnum];
815 				probe_procnum = procnum;
816 			} else {
817 				probe_id =
818 				    nfscl_nfs2_start_probes[nd->nd_procnum];
819 				probe_procnum = procnum;
820 			}
821 			if (probe_id != 0)
822 				(dtrace_nfscl_nfs234_start_probe)
823 				    (probe_id, vp, nd->nd_mreq, cred,
824 				     probe_procnum);
825 		}
826 #endif
827 	}
828 	freeslot = -1;		/* Set to slot that needs to be free'd */
829 tryagain:
830 	slot = -1;		/* Slot that needs a sequence# increment. */
831 	/*
832 	 * This timeout specifies when a new socket should be created,
833 	 * along with new xid values. For UDP, this should be done
834 	 * infrequently, since retransmits of RPC requests should normally
835 	 * use the same xid.
836 	 */
837 	if (nmp == NULL) {
838 		if (clp == NULL) {
839 			timo.tv_sec = NFSV4_UPCALLTIMEO;
840 			timo.tv_usec = 0;
841 		} else {
842 			timo.tv_sec = NFSV4_CALLBACKTIMEO / 1000;
843 			timo.tv_usec = NFSV4_CALLBACKTIMEO * 1000;
844 		}
845 	} else {
846 		if (nrp->nr_sotype != SOCK_DGRAM) {
847 			timo.tv_usec = 0;
848 			if ((nmp->nm_flag & NFSMNT_NFSV4))
849 				timo.tv_sec = INT_MAX;
850 			else
851 				timo.tv_sec = NFS_TCPTIMEO;
852 		} else {
853 			if (NFSHASSOFT(nmp)) {
854 				/*
855 				 * CLSET_RETRIES is set to 2, so this should be
856 				 * half of the total timeout required.
857 				 */
858 				timeo = nmp->nm_retry * nmp->nm_timeo / 2;
859 				if (timeo < 1)
860 					timeo = 1;
861 				timo.tv_sec = timeo / NFS_HZ;
862 				timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
863 				    NFS_HZ;
864 			} else {
865 				/* For UDP hard mounts, use a large value. */
866 				timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
867 				timo.tv_usec = 0;
868 			}
869 		}
870 
871 		if (rep != NULL) {
872 			rep->r_flags = 0;
873 			rep->r_nmp = nmp;
874 			/*
875 			 * Chain request into list of outstanding requests.
876 			 */
877 			NFSLOCKREQ();
878 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
879 			NFSUNLOCKREQ();
880 		}
881 	}
882 
883 	nd->nd_mrep = NULL;
884 	if (clp != NULL && sep != NULL)
885 		stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
886 		    nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
887 	else if (nextconn_set)
888 		/*
889 		 * When there are multiple TCP connections, send the
890 		 * RPCs with large messages on the alternate TCP
891 		 * connection(s) in a round robin fashion.
892 		 * The small RPC messages are sent on the default
893 		 * TCP connection because they do not require much
894 		 * network bandwidth and separating them from the
895 		 * large RPC messages avoids them getting "log jammed"
896 		 * behind several large RPC messages.
897 		 */
898 		stat = CLNT_CALL_MBUF(nmp->nm_aconn[nextconn],
899 		    &ext, procnum, nd->nd_mreq, &nd->nd_mrep, timo);
900 	else
901 		stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
902 		    nd->nd_mreq, &nd->nd_mrep, timo);
903 	NFSCL_DEBUG(2, "clnt call=%d\n", stat);
904 
905 	if (rep != NULL) {
906 		/*
907 		 * RPC done, unlink the request.
908 		 */
909 		NFSLOCKREQ();
910 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
911 		NFSUNLOCKREQ();
912 	}
913 
914 	/*
915 	 * If there was a successful reply and a tprintf msg.
916 	 * tprintf a response.
917 	 */
918 	if (stat == RPC_SUCCESS) {
919 		error = 0;
920 	} else if (stat == RPC_TIMEDOUT) {
921 		NFSINCRGLOBAL(nfsstatsv1_p->rpctimeouts);
922 		error = ETIMEDOUT;
923 	} else if (stat == RPC_VERSMISMATCH) {
924 		NFSINCRGLOBAL(nfsstatsv1_p->rpcinvalid);
925 		error = EOPNOTSUPP;
926 	} else if (stat == RPC_PROGVERSMISMATCH) {
927 		NFSINCRGLOBAL(nfsstatsv1_p->rpcinvalid);
928 		error = EPROTONOSUPPORT;
929 	} else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
930 	     stat == RPC_SYSTEMERROR || stat == RPC_INTR) {
931 		/* Check for a session slot that needs to be free'd. */
932 		if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
933 		    (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
934 		    nd->nd_procnum != NFSPROC_NULL) {
935 			/*
936 			 * This should only occur when either the MDS or
937 			 * a client has an RPC against a DS fail.
938 			 * This happens because these cases use "soft"
939 			 * connections that can time out and fail.
940 			 * The slot used for this RPC is now in a
941 			 * non-deterministic state, but if the slot isn't
942 			 * free'd, threads can get stuck waiting for a slot.
943 			 */
944 			if (sep == NULL)
945 				sep = nfsmnt_mdssession(nmp);
946 			/*
947 			 * Bump the sequence# out of range, so that reuse of
948 			 * this slot will result in an NFSERR_SEQMISORDERED
949 			 * error and not a bogus cached RPC reply.
950 			 */
951 			mtx_lock(&sep->nfsess_mtx);
952 			sep->nfsess_slotseq[nd->nd_slotid] += 10;
953 			sep->nfsess_badslots |= (0x1ULL << nd->nd_slotid);
954 			mtx_unlock(&sep->nfsess_mtx);
955 			/* And free the slot. */
956 			nfsv4_freeslot(sep, nd->nd_slotid, false);
957 		}
958 		if (stat == RPC_INTR)
959 			error = EINTR;
960 		else {
961 			NFSINCRGLOBAL(nfsstatsv1_p->rpcinvalid);
962 			error = ENXIO;
963 		}
964 	} else {
965 		NFSINCRGLOBAL(nfsstatsv1_p->rpcinvalid);
966 		error = EACCES;
967 	}
968 	if (error) {
969 		m_freem(nd->nd_mreq);
970 		if (usegssname == 0)
971 			AUTH_DESTROY(auth);
972 		if (rep != NULL)
973 			free(rep, M_NFSDREQ);
974 		if (set_sigset)
975 			newnfs_restore_sigmask(td, &oldset);
976 		return (error);
977 	}
978 
979 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
980 
981 	/*
982 	 * Search for any mbufs that are not a multiple of 4 bytes long
983 	 * or with m_data not longword aligned.
984 	 * These could cause pointer alignment problems, so copy them to
985 	 * well aligned mbufs.
986 	 */
987 	newnfs_realign(&nd->nd_mrep, M_WAITOK);
988 	nd->nd_md = nd->nd_mrep;
989 	nd->nd_dpos = mtod(nd->nd_md, caddr_t);
990 	nd->nd_repstat = 0;
991 	if (nd->nd_procnum != NFSPROC_NULL &&
992 	    nd->nd_procnum != NFSV4PROC_CBNULL) {
993 		/* If sep == NULL, set it to the default in nmp. */
994 		if (sep == NULL && nmp != NULL)
995 			sep = nfsmnt_mdssession(nmp);
996 		/*
997 		 * and now the actual NFS xdr.
998 		 */
999 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1000 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
1001 		if (nd->nd_repstat >= 10000)
1002 			NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
1003 			    (int)nd->nd_repstat);
1004 
1005 		/*
1006 		 * Get rid of the tag, return count and SEQUENCE result for
1007 		 * NFSv4.
1008 		 */
1009 		if ((nd->nd_flag & ND_NFSV4) != 0 && nd->nd_repstat !=
1010 		    NFSERR_MINORVERMISMATCH) {
1011 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1012 			i = fxdr_unsigned(int, *tl);
1013 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
1014 			if (error)
1015 				goto nfsmout;
1016 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1017 			opcnt = fxdr_unsigned(int, *tl++);
1018 			i = fxdr_unsigned(int, *tl++);
1019 			j = fxdr_unsigned(int, *tl);
1020 			if (j >= 10000)
1021 				NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
1022 			/*
1023 			 * If the first op is Sequence, free up the slot.
1024 			 */
1025 			if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
1026 			   (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0)) {
1027 				NFSCL_DEBUG(1, "failed seq=%d\n", j);
1028 				if (sep != NULL && i == NFSV4OP_SEQUENCE &&
1029 				    j == NFSERR_SEQMISORDERED) {
1030 					mtx_lock(&sep->nfsess_mtx);
1031 					sep->nfsess_badslots |=
1032 					    (0x1ULL << nd->nd_slotid);
1033 					mtx_unlock(&sep->nfsess_mtx);
1034 				}
1035 			}
1036 			if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
1037 			    (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
1038 			    j == 0)) && sep != NULL) {
1039 				if (i == NFSV4OP_SEQUENCE)
1040 					NFSM_DISSECT(tl, uint32_t *,
1041 					    NFSX_V4SESSIONID +
1042 					    5 * NFSX_UNSIGNED);
1043 				else
1044 					NFSM_DISSECT(tl, uint32_t *,
1045 					    NFSX_V4SESSIONID +
1046 					    4 * NFSX_UNSIGNED);
1047 				mtx_lock(&sep->nfsess_mtx);
1048 				if (bcmp(tl, sep->nfsess_sessionid,
1049 				    NFSX_V4SESSIONID) == 0) {
1050 					tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
1051 					retseq = fxdr_unsigned(uint32_t, *tl++);
1052 					slot = fxdr_unsigned(int, *tl++);
1053 					if ((nd->nd_flag & ND_HASSLOTID) != 0) {
1054 						if (slot >= NFSV4_SLOTS ||
1055 						    (i == NFSV4OP_CBSEQUENCE &&
1056 						     slot >= NFSV4_CBSLOTS)) {
1057 							printf("newnfs_request:"
1058 							    " Bogus slot\n");
1059 							slot = nd->nd_slotid;
1060 						} else if (slot !=
1061 						    nd->nd_slotid) {
1062 						    printf("newnfs_request:"
1063 							" Wrong session "
1064 							"srvslot=%d "
1065 							"slot=%d\n", slot,
1066 							nd->nd_slotid);
1067 						    if (i == NFSV4OP_SEQUENCE) {
1068 							/*
1069 							 * Mark both slots as
1070 							 * bad, because we do
1071 							 * not know if the
1072 							 * server has advanced
1073 							 * the sequence# for
1074 							 * either of them.
1075 							 */
1076 							sep->nfsess_badslots |=
1077 							    (0x1ULL << slot);
1078 							sep->nfsess_badslots |=
1079 							    (0x1ULL <<
1080 							     nd->nd_slotid);
1081 						    }
1082 						    slot = nd->nd_slotid;
1083 						}
1084 						freeslot = slot;
1085 					} else if (slot != 0) {
1086 						printf("newnfs_request: Bad "
1087 						    "session slot=%d\n", slot);
1088 						slot = 0;
1089 					}
1090 					if (retseq != sep->nfsess_slotseq[slot])
1091 						printf("retseq diff 0x%x\n",
1092 						    retseq);
1093 					retval = fxdr_unsigned(uint32_t, *++tl);
1094 					if ((retval + 1) < sep->nfsess_foreslots
1095 					    )
1096 						sep->nfsess_foreslots = (retval
1097 						    + 1);
1098 					else if ((retval + 1) >
1099 					    sep->nfsess_foreslots)
1100 						sep->nfsess_foreslots = (retval
1101 						    < 64) ? (retval + 1) : 64;
1102 				}
1103 				mtx_unlock(&sep->nfsess_mtx);
1104 
1105 				/* Grab the op and status for the next one. */
1106 				if (opcnt > 1) {
1107 					NFSM_DISSECT(tl, uint32_t *,
1108 					    2 * NFSX_UNSIGNED);
1109 					i = fxdr_unsigned(int, *tl++);
1110 					j = fxdr_unsigned(int, *tl);
1111 				}
1112 			}
1113 		}
1114 		if (nd->nd_repstat != 0) {
1115 			if (nd->nd_repstat == NFSERR_BADSESSION &&
1116 			    nmp != NULL && dssep == NULL &&
1117 			    (nd->nd_flag & ND_NFSV41) != 0) {
1118 				/*
1119 				 * If this is a client side MDS RPC, mark
1120 				 * the MDS session defunct and initiate
1121 				 * recovery, as required.
1122 				 * The nfsess_defunct field is protected by
1123 				 * the NFSLOCKMNT()/nm_mtx lock and not the
1124 				 * nfsess_mtx lock to simplify its handling,
1125 				 * for the MDS session. This lock is also
1126 				 * sufficient for nfsess_sessionid, since it
1127 				 * never changes in the structure.
1128 				 */
1129 				NFSCL_DEBUG(1, "Got badsession\n");
1130 				NFSLOCKCLSTATE();
1131 				NFSLOCKMNT(nmp);
1132 				sep = NFSMNT_MDSSESSION(nmp);
1133 				if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
1134 				    NFSX_V4SESSIONID) == 0) {
1135 					printf("Initiate recovery. If server "
1136 					    "has not rebooted, "
1137 					    "check NFS clients for unique "
1138 					    "/etc/hostid's\n");
1139 					/* Initiate recovery. */
1140 					sep->nfsess_defunct = 1;
1141 					NFSCL_DEBUG(1, "Marked defunct\n");
1142 					if (nmp->nm_clp != NULL) {
1143 						nmp->nm_clp->nfsc_flags |=
1144 						    NFSCLFLAGS_RECOVER;
1145 						wakeup(nmp->nm_clp);
1146 					}
1147 				}
1148 				NFSUNLOCKCLSTATE();
1149 				/*
1150 				 * Sleep for up to 1sec waiting for a new
1151 				 * session.
1152 				 */
1153 				mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1154 				    "nfsbadsess", hz);
1155 				/*
1156 				 * Get the session again, in case a new one
1157 				 * has been created during the sleep.
1158 				 */
1159 				sep = NFSMNT_MDSSESSION(nmp);
1160 				NFSUNLOCKMNT(nmp);
1161 				if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1162 					reterr = nfsv4_sequencelookup(nmp, sep,
1163 					    &slotpos, &maxslot, &slotseq,
1164 					    sessionid, true);
1165 					if (reterr == 0) {
1166 						/* Fill in new session info. */
1167 						NFSCL_DEBUG(1,
1168 						  "Filling in new sequence\n");
1169 						tl = nd->nd_sequence;
1170 						bcopy(sessionid, tl,
1171 						    NFSX_V4SESSIONID);
1172 						tl += NFSX_V4SESSIONID /
1173 						    NFSX_UNSIGNED;
1174 						*tl++ = txdr_unsigned(slotseq);
1175 						*tl++ = txdr_unsigned(slotpos);
1176 						*tl = txdr_unsigned(maxslot);
1177 						nd->nd_slotid = slotpos;
1178 						nd->nd_flag |= ND_HASSLOTID;
1179 					}
1180 					if (reterr == NFSERR_BADSESSION ||
1181 					    reterr == 0) {
1182 						NFSCL_DEBUG(1,
1183 						    "Badsession looping\n");
1184 						m_freem(nd->nd_mrep);
1185 						nd->nd_mrep = NULL;
1186 						goto tryagain;
1187 					}
1188 					nd->nd_repstat = reterr;
1189 					NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1190 				}
1191 			}
1192 			/*
1193 			 * When clp != NULL, it is a callback and all
1194 			 * callback operations can be retried for NFSERR_DELAY.
1195 			 */
1196 			if (((nd->nd_repstat == NFSERR_DELAY ||
1197 			      nd->nd_repstat == NFSERR_GRACE) &&
1198 			     (nd->nd_flag & ND_NFSV4) && (clp != NULL ||
1199 			     (nd->nd_procnum != NFSPROC_DELEGRETURN &&
1200 			     nd->nd_procnum != NFSPROC_SETATTR &&
1201 			     nd->nd_procnum != NFSPROC_READ &&
1202 			     nd->nd_procnum != NFSPROC_READDS &&
1203 			     nd->nd_procnum != NFSPROC_WRITE &&
1204 			     nd->nd_procnum != NFSPROC_WRITEDS &&
1205 			     nd->nd_procnum != NFSPROC_OPEN &&
1206 			     nd->nd_procnum != NFSPROC_OPENLAYGET &&
1207 			     nd->nd_procnum != NFSPROC_CREATE &&
1208 			     nd->nd_procnum != NFSPROC_CREATELAYGET &&
1209 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1210 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1211 			     nd->nd_procnum != NFSPROC_CLOSE &&
1212 			     nd->nd_procnum != NFSPROC_LOCK &&
1213 			     nd->nd_procnum != NFSPROC_LOCKU))) ||
1214 			    (nd->nd_repstat == NFSERR_DELAY &&
1215 			     (nd->nd_flag & ND_NFSV4) == 0) ||
1216 			    nd->nd_repstat == NFSERR_RESOURCE) {
1217 				/* Clip at NFS_TRYLATERDEL. */
1218 				if (timespeccmp(&trylater_delay,
1219 				    &nfs_trylater_max, >))
1220 					trylater_delay = nfs_trylater_max;
1221 				getnanouptime(&waituntil);
1222 				timespecadd(&waituntil, &trylater_delay,
1223 				    &waituntil);
1224 				do {
1225 					nfs_catnap(PZERO, 0, "nfstry");
1226 					getnanouptime(&ts);
1227 				} while (timespeccmp(&ts, &waituntil, <));
1228 				timespecadd(&trylater_delay, &trylater_delay,
1229 				    &trylater_delay);	/* Double each time. */
1230 				if (slot != -1) {
1231 					mtx_lock(&sep->nfsess_mtx);
1232 					sep->nfsess_slotseq[slot]++;
1233 					*nd->nd_slotseq = txdr_unsigned(
1234 					    sep->nfsess_slotseq[slot]);
1235 					mtx_unlock(&sep->nfsess_mtx);
1236 				}
1237 				m_freem(nd->nd_mrep);
1238 				nd->nd_mrep = NULL;
1239 				goto tryagain;
1240 			}
1241 
1242 			/*
1243 			 * If the File Handle was stale, invalidate the
1244 			 * lookup cache, just in case.
1245 			 * (vp != NULL implies a client side call)
1246 			 */
1247 			if (nd->nd_repstat == ESTALE && vp != NULL) {
1248 				cache_purge(vp);
1249 				if (ncl_call_invalcaches != NULL)
1250 					(*ncl_call_invalcaches)(vp);
1251 			}
1252 		}
1253 		if ((nd->nd_flag & ND_NFSV4) != 0) {
1254 			/* Free the slot, as required. */
1255 			if (freeslot != -1)
1256 				nfsv4_freeslot(sep, freeslot, false);
1257 			/*
1258 			 * If this op is Putfh, throw its results away.
1259 			 */
1260 			if (j >= 10000)
1261 				NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1262 			if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1263 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1264 				i = fxdr_unsigned(int, *tl++);
1265 				j = fxdr_unsigned(int, *tl);
1266 				if (j >= 10000)
1267 					NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1268 					    j);
1269 				/*
1270 				 * All Compounds that do an Op that must
1271 				 * be in sequence consist of NFSV4OP_PUTFH
1272 				 * followed by one of these. As such, we
1273 				 * can determine if the seqid# should be
1274 				 * incremented, here.
1275 				 */
1276 				if ((i == NFSV4OP_OPEN ||
1277 				     i == NFSV4OP_OPENCONFIRM ||
1278 				     i == NFSV4OP_OPENDOWNGRADE ||
1279 				     i == NFSV4OP_CLOSE ||
1280 				     i == NFSV4OP_LOCK ||
1281 				     i == NFSV4OP_LOCKU) &&
1282 				    (j == 0 ||
1283 				     (j != NFSERR_STALECLIENTID &&
1284 				      j != NFSERR_STALESTATEID &&
1285 				      j != NFSERR_BADSTATEID &&
1286 				      j != NFSERR_BADSEQID &&
1287 				      j != NFSERR_BADXDR &&
1288 				      j != NFSERR_RESOURCE &&
1289 				      j != NFSERR_NOFILEHANDLE)))
1290 					nd->nd_flag |= ND_INCRSEQID;
1291 			}
1292 			/*
1293 			 * If this op's status is non-zero, mark
1294 			 * that there is no more data to process.
1295 			 * The exception is Setattr, which always has xdr
1296 			 * when it has failed.
1297 			 */
1298 			if (j != 0 && i != NFSV4OP_SETATTR)
1299 				nd->nd_flag |= ND_NOMOREDATA;
1300 
1301 			/*
1302 			 * If R_DONTRECOVER is set, replace the stale error
1303 			 * reply, so that recovery isn't initiated.
1304 			 */
1305 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1306 			     nd->nd_repstat == NFSERR_BADSESSION ||
1307 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
1308 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
1309 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1310 		}
1311 	}
1312 
1313 #ifdef KDTRACE_HOOKS
1314 	if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1315 		uint32_t probe_id;
1316 		int probe_procnum;
1317 
1318 		if (nd->nd_flag & ND_NFSV4) {
1319 			probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1320 			probe_procnum = nd->nd_procnum;
1321 		} else if (nd->nd_flag & ND_NFSV3) {
1322 			probe_id = nfscl_nfs3_done_probes[procnum];
1323 			probe_procnum = procnum;
1324 		} else {
1325 			probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1326 			probe_procnum = procnum;
1327 		}
1328 		if (probe_id != 0)
1329 			(dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1330 			    nd->nd_mreq, cred, probe_procnum, 0);
1331 	}
1332 #endif
1333 
1334 	m_freem(nd->nd_mreq);
1335 	if (usegssname == 0)
1336 		AUTH_DESTROY(auth);
1337 	if (rep != NULL)
1338 		free(rep, M_NFSDREQ);
1339 	if (set_sigset)
1340 		newnfs_restore_sigmask(td, &oldset);
1341 	return (0);
1342 nfsmout:
1343 	m_freem(nd->nd_mrep);
1344 	m_freem(nd->nd_mreq);
1345 	if (usegssname == 0)
1346 		AUTH_DESTROY(auth);
1347 	if (rep != NULL)
1348 		free(rep, M_NFSDREQ);
1349 	if (set_sigset)
1350 		newnfs_restore_sigmask(td, &oldset);
1351 	return (error);
1352 }
1353 
1354 /*
1355  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1356  * wait for all requests to complete. This is used by forced unmounts
1357  * to terminate any outstanding RPCs.
1358  */
1359 int
1360 newnfs_nmcancelreqs(struct nfsmount *nmp)
1361 {
1362 	struct nfsclds *dsp;
1363 	struct __rpc_client *cl;
1364 	int i;
1365 
1366 	if (nmp->nm_sockreq.nr_client != NULL)
1367 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1368 	for (i = 0; i < nmp->nm_aconnect; i++)
1369 		if (nmp->nm_aconn[i] != NULL)
1370 			CLNT_CLOSE(nmp->nm_aconn[i]);
1371 lookformore:
1372 	NFSLOCKMNT(nmp);
1373 	TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1374 		NFSLOCKDS(dsp);
1375 		if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1376 		    (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1377 		    dsp->nfsclds_sockp != NULL &&
1378 		    dsp->nfsclds_sockp->nr_client != NULL) {
1379 			dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1380 			cl = dsp->nfsclds_sockp->nr_client;
1381 			NFSUNLOCKDS(dsp);
1382 			NFSUNLOCKMNT(nmp);
1383 			CLNT_CLOSE(cl);
1384 			goto lookformore;
1385 		}
1386 		NFSUNLOCKDS(dsp);
1387 	}
1388 	NFSUNLOCKMNT(nmp);
1389 	return (0);
1390 }
1391 
1392 /*
1393  * Any signal that can interrupt an NFS operation in an intr mount
1394  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1395  */
1396 int newnfs_sig_set[] = {
1397 	SIGINT,
1398 	SIGTERM,
1399 	SIGHUP,
1400 	SIGKILL,
1401 	SIGQUIT
1402 };
1403 
1404 /*
1405  * Check to see if one of the signals in our subset is pending on
1406  * the process (in an intr mount).
1407  */
1408 static int
1409 nfs_sig_pending(sigset_t set)
1410 {
1411 	int i;
1412 
1413 	for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1414 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
1415 			return (1);
1416 	return (0);
1417 }
1418 
1419 /*
1420  * The set/restore sigmask functions are used to (temporarily) overwrite
1421  * the thread td_sigmask during an RPC call (for example). These are also
1422  * used in other places in the NFS client that might tsleep().
1423  */
1424 void
1425 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1426 {
1427 	sigset_t newset;
1428 	int i;
1429 	struct proc *p;
1430 
1431 	SIGFILLSET(newset);
1432 	if (td == NULL)
1433 		td = curthread; /* XXX */
1434 	p = td->td_proc;
1435 	/* Remove the NFS set of signals from newset */
1436 	PROC_LOCK(p);
1437 	mtx_lock(&p->p_sigacts->ps_mtx);
1438 	for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1439 		/*
1440 		 * But make sure we leave the ones already masked
1441 		 * by the process, ie. remove the signal from the
1442 		 * temporary signalmask only if it wasn't already
1443 		 * in p_sigmask.
1444 		 */
1445 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1446 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1447 			SIGDELSET(newset, newnfs_sig_set[i]);
1448 	}
1449 	mtx_unlock(&p->p_sigacts->ps_mtx);
1450 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1451 	    SIGPROCMASK_PROC_LOCKED);
1452 	PROC_UNLOCK(p);
1453 }
1454 
1455 void
1456 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1457 {
1458 	if (td == NULL)
1459 		td = curthread; /* XXX */
1460 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1461 }
1462 
1463 /*
1464  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1465  * old one after msleep() returns.
1466  */
1467 int
1468 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1469 {
1470 	sigset_t oldset;
1471 	int error;
1472 
1473 	if ((priority & PCATCH) == 0)
1474 		return msleep(ident, mtx, priority, wmesg, timo);
1475 	if (td == NULL)
1476 		td = curthread; /* XXX */
1477 	newnfs_set_sigmask(td, &oldset);
1478 	error = msleep(ident, mtx, priority, wmesg, timo);
1479 	newnfs_restore_sigmask(td, &oldset);
1480 	return (error);
1481 }
1482 
1483 /*
1484  * Test for a termination condition pending on the process.
1485  * This is used for NFSMNT_INT mounts.
1486  */
1487 int
1488 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1489 {
1490 	struct proc *p;
1491 	sigset_t tmpset;
1492 
1493 	/* Terminate all requests while attempting a forced unmount. */
1494 	if (NFSCL_FORCEDISM(nmp->nm_mountp))
1495 		return (EIO);
1496 	if (!(nmp->nm_flag & NFSMNT_INT))
1497 		return (0);
1498 	if (td == NULL)
1499 		return (0);
1500 	p = td->td_proc;
1501 	PROC_LOCK(p);
1502 	tmpset = p->p_siglist;
1503 	SIGSETOR(tmpset, td->td_siglist);
1504 	SIGSETNAND(tmpset, td->td_sigmask);
1505 	mtx_lock(&p->p_sigacts->ps_mtx);
1506 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1507 	mtx_unlock(&p->p_sigacts->ps_mtx);
1508 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1509 	    && nfs_sig_pending(tmpset)) {
1510 		PROC_UNLOCK(p);
1511 		return (EINTR);
1512 	}
1513 	PROC_UNLOCK(p);
1514 	return (0);
1515 }
1516 
1517 static int
1518 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1519 {
1520 	struct proc *p;
1521 
1522 	p = td ? td->td_proc : NULL;
1523 	if (error) {
1524 		tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1525 		    server, msg, error);
1526 	} else {
1527 		tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1528 	}
1529 	return (0);
1530 }
1531 
1532 static void
1533 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1534     int error, int flags)
1535 {
1536 	if (nmp == NULL)
1537 		return;
1538 	mtx_lock(&nmp->nm_mtx);
1539 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1540 		nmp->nm_state |= NFSSTA_TIMEO;
1541 		mtx_unlock(&nmp->nm_mtx);
1542 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1543 		    VQ_NOTRESP, 0);
1544 	} else
1545 		mtx_unlock(&nmp->nm_mtx);
1546 	mtx_lock(&nmp->nm_mtx);
1547 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1548 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
1549 		mtx_unlock(&nmp->nm_mtx);
1550 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1551 		    VQ_NOTRESPLOCK, 0);
1552 	} else
1553 		mtx_unlock(&nmp->nm_mtx);
1554 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1555 }
1556 
1557 static void
1558 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1559     int flags, int tprintfmsg)
1560 {
1561 	if (nmp == NULL)
1562 		return;
1563 	if (tprintfmsg) {
1564 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1565 	}
1566 
1567 	mtx_lock(&nmp->nm_mtx);
1568 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1569 		nmp->nm_state &= ~NFSSTA_TIMEO;
1570 		mtx_unlock(&nmp->nm_mtx);
1571 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1572 		    VQ_NOTRESP, 1);
1573 	} else
1574 		mtx_unlock(&nmp->nm_mtx);
1575 
1576 	mtx_lock(&nmp->nm_mtx);
1577 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1578 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1579 		mtx_unlock(&nmp->nm_mtx);
1580 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1581 		    VQ_NOTRESPLOCK, 1);
1582 	} else
1583 		mtx_unlock(&nmp->nm_mtx);
1584 }
1585