xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 1989, 1991, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 /*
38  * Socket operations for use by nfs
39  */
40 
41 #include "opt_inet6.h"
42 #include "opt_kgssapi.h"
43 #include "opt_nfs.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/limits.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.h>
55 #include <sys/signalvar.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysctl.h>
58 #include <sys/syslog.h>
59 #include <sys/vnode.h>
60 
61 #include <rpc/rpc.h>
62 
63 #include <kgssapi/krb5/kcrypto.h>
64 
65 #include <fs/nfs/nfsport.h>
66 
67 NFSSTATESPINLOCK;
68 NFSREQSPINLOCK;
69 extern struct nfsstats newnfsstats;
70 extern struct nfsreqhead nfsd_reqq;
71 extern int nfscl_ticks;
72 extern void (*ncl_call_invalcaches)(struct vnode *);
73 
74 static int	nfsrv_gsscallbackson = 0;
75 static int	nfs_bufpackets = 4;
76 static int	nfs_reconnects;
77 static int	nfs3_jukebox_delay = 10;
78 static int	nfs_skip_wcc_data_onerr = 1;
79 static int	nfs_keytab_enctype = ETYPE_DES_CBC_CRC;
80 
81 SYSCTL_DECL(_vfs_nfs);
82 
83 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
84     "Buffer reservation size 2 < x < 64");
85 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
86     "Number of times the nfs client has had to reconnect");
87 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
88     "Number of seconds to delay a retry after receiving EJUKEBOX");
89 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
90     "Disable weak cache consistency checking when server returns an error");
91 SYSCTL_INT(_vfs_nfs, OID_AUTO, keytab_enctype, CTLFLAG_RW, &nfs_keytab_enctype, 0,
92     "Encryption type for the keytab entry used by nfs");
93 
94 static void	nfs_down(struct nfsmount *, struct thread *, const char *,
95     int, int);
96 static void	nfs_up(struct nfsmount *, struct thread *, const char *,
97     int, int);
98 static int	nfs_msg(struct thread *, const char *, const char *, int);
99 
100 struct nfs_cached_auth {
101 	int		ca_refs; /* refcount, including 1 from the cache */
102 	uid_t		ca_uid;	 /* uid that corresponds to this auth */
103 	AUTH		*ca_auth; /* RPC auth handle */
104 };
105 
106 static int nfsv2_procid[NFS_V3NPROCS] = {
107 	NFSV2PROC_NULL,
108 	NFSV2PROC_GETATTR,
109 	NFSV2PROC_SETATTR,
110 	NFSV2PROC_LOOKUP,
111 	NFSV2PROC_NOOP,
112 	NFSV2PROC_READLINK,
113 	NFSV2PROC_READ,
114 	NFSV2PROC_WRITE,
115 	NFSV2PROC_CREATE,
116 	NFSV2PROC_MKDIR,
117 	NFSV2PROC_SYMLINK,
118 	NFSV2PROC_CREATE,
119 	NFSV2PROC_REMOVE,
120 	NFSV2PROC_RMDIR,
121 	NFSV2PROC_RENAME,
122 	NFSV2PROC_LINK,
123 	NFSV2PROC_READDIR,
124 	NFSV2PROC_NOOP,
125 	NFSV2PROC_STATFS,
126 	NFSV2PROC_NOOP,
127 	NFSV2PROC_NOOP,
128 	NFSV2PROC_NOOP,
129 };
130 
131 /*
132  * Initialize sockets and congestion for a new NFS connection.
133  * We do not free the sockaddr if error.
134  */
135 int
136 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
137     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
138 {
139 	int rcvreserve, sndreserve;
140 	int pktscale;
141 	struct sockaddr *saddr;
142 	struct ucred *origcred;
143 	CLIENT *client;
144 	struct netconfig *nconf;
145 	struct socket *so;
146 	int one = 1, retries, error;
147 	struct thread *td = curthread;
148 
149 	/*
150 	 * We need to establish the socket using the credentials of
151 	 * the mountpoint.  Some parts of this process (such as
152 	 * sobind() and soconnect()) will use the curent thread's
153 	 * credential instead of the socket credential.  To work
154 	 * around this, temporarily change the current thread's
155 	 * credential to that of the mountpoint.
156 	 *
157 	 * XXX: It would be better to explicitly pass the correct
158 	 * credential to sobind() and soconnect().
159 	 */
160 	origcred = td->td_ucred;
161 
162 	/*
163 	 * Use the credential in nr_cred, if not NULL.
164 	 */
165 	if (nrp->nr_cred != NULL)
166 		td->td_ucred = nrp->nr_cred;
167 	else
168 		td->td_ucred = cred;
169 	saddr = nrp->nr_nam;
170 
171 	if (saddr->sa_family == AF_INET)
172 		if (nrp->nr_sotype == SOCK_DGRAM)
173 			nconf = getnetconfigent("udp");
174 		else
175 			nconf = getnetconfigent("tcp");
176 	else
177 		if (nrp->nr_sotype == SOCK_DGRAM)
178 			nconf = getnetconfigent("udp6");
179 		else
180 			nconf = getnetconfigent("tcp6");
181 
182 	pktscale = nfs_bufpackets;
183 	if (pktscale < 2)
184 		pktscale = 2;
185 	if (pktscale > 64)
186 		pktscale = 64;
187 	/*
188 	 * soreserve() can fail if sb_max is too small, so shrink pktscale
189 	 * and try again if there is an error.
190 	 * Print a log message suggesting increasing sb_max.
191 	 * Creating a socket and doing this is necessary since, if the
192 	 * reservation sizes are too large and will make soreserve() fail,
193 	 * the connection will work until a large send is attempted and
194 	 * then it will loop in the krpc code.
195 	 */
196 	so = NULL;
197 	saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
198 	error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
199 	    nrp->nr_soproto, td->td_ucred, td);
200 	if (error) {
201 		td->td_ucred = origcred;
202 		return (error);
203 	}
204 	do {
205 	    if (error != 0 && pktscale > 2)
206 		pktscale--;
207 	    if (nrp->nr_sotype == SOCK_DGRAM) {
208 		if (nmp != NULL) {
209 			sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
210 			    pktscale;
211 			rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
212 			    pktscale;
213 		} else {
214 			sndreserve = rcvreserve = 1024 * pktscale;
215 		}
216 	    } else {
217 		if (nrp->nr_sotype != SOCK_STREAM)
218 			panic("nfscon sotype");
219 		if (nmp != NULL) {
220 			sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
221 			    sizeof (u_int32_t)) * pktscale;
222 			rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
223 			    sizeof (u_int32_t)) * pktscale;
224 		} else {
225 			sndreserve = rcvreserve = 1024 * pktscale;
226 		}
227 	    }
228 	    error = soreserve(so, sndreserve, rcvreserve);
229 	} while (error != 0 && pktscale > 2);
230 	soclose(so);
231 	if (error) {
232 		td->td_ucred = origcred;
233 		return (error);
234 	}
235 
236 	client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
237 	    nrp->nr_vers, sndreserve, rcvreserve);
238 	CLNT_CONTROL(client, CLSET_WAITCHAN, "newnfsreq");
239 	if (nmp != NULL) {
240 		if ((nmp->nm_flag & NFSMNT_INT))
241 			CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
242 		if ((nmp->nm_flag & NFSMNT_RESVPORT))
243 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
244 		if (NFSHASSOFT(nmp))
245 			retries = nmp->nm_retry;
246 		else
247 			retries = INT_MAX;
248 	} else {
249 		/*
250 		 * Three cases:
251 		 * - Null RPC callback to client
252 		 * - Non-Null RPC callback to client, wait a little longer
253 		 * - upcalls to nfsuserd and gssd (clp == NULL)
254 		 */
255 		if (callback_retry_mult == 0) {
256 			retries = NFSV4_UPCALLRETRY;
257 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
258 		} else {
259 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
260 		}
261 	}
262 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
263 
264 	mtx_lock(&nrp->nr_mtx);
265 	if (nrp->nr_client != NULL) {
266 		/*
267 		 * Someone else already connected.
268 		 */
269 		CLNT_RELEASE(client);
270 	} else {
271 		nrp->nr_client = client;
272 	}
273 
274 	/*
275 	 * Protocols that do not require connections may be optionally left
276 	 * unconnected for servers that reply from a port other than NFS_PORT.
277 	 */
278 	if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
279 		mtx_unlock(&nrp->nr_mtx);
280 		CLNT_CONTROL(client, CLSET_CONNECT, &one);
281 	} else {
282 		mtx_unlock(&nrp->nr_mtx);
283 	}
284 
285 	/* Restore current thread's credentials. */
286 	td->td_ucred = origcred;
287 	return (0);
288 }
289 
290 /*
291  * NFS disconnect. Clean up and unlink.
292  */
293 void
294 newnfs_disconnect(struct nfssockreq *nrp)
295 {
296 	CLIENT *client;
297 
298 	mtx_lock(&nrp->nr_mtx);
299 	if (nrp->nr_client != NULL) {
300 		client = nrp->nr_client;
301 		nrp->nr_client = NULL;
302 		mtx_unlock(&nrp->nr_mtx);
303 #ifdef KGSSAPI
304 		rpc_gss_secpurge(client);
305 #endif
306 		CLNT_CLOSE(client);
307 		CLNT_RELEASE(client);
308 	} else {
309 		mtx_unlock(&nrp->nr_mtx);
310 	}
311 }
312 
313 static AUTH *
314 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
315     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
316 {
317 #ifdef KGSSAPI
318 	rpc_gss_service_t svc;
319 	AUTH *auth;
320 #ifdef notyet
321 	rpc_gss_options_req_t req_options;
322 #endif
323 #endif
324 
325 	switch (secflavour) {
326 #ifdef KGSSAPI
327 	case RPCSEC_GSS_KRB5:
328 	case RPCSEC_GSS_KRB5I:
329 	case RPCSEC_GSS_KRB5P:
330 		if (!mech_oid) {
331 			if (!rpc_gss_mech_to_oid("kerberosv5", &mech_oid))
332 				return (NULL);
333 		}
334 		if (secflavour == RPCSEC_GSS_KRB5)
335 			svc = rpc_gss_svc_none;
336 		else if (secflavour == RPCSEC_GSS_KRB5I)
337 			svc = rpc_gss_svc_integrity;
338 		else
339 			svc = rpc_gss_svc_privacy;
340 #ifdef notyet
341 		req_options.req_flags = GSS_C_MUTUAL_FLAG;
342 		req_options.time_req = 0;
343 		req_options.my_cred = GSS_C_NO_CREDENTIAL;
344 		req_options.input_channel_bindings = NULL;
345 		req_options.enc_type = nfs_keytab_enctype;
346 
347 		auth = rpc_gss_secfind(nrp->nr_client, cred,
348 		    clnt_principal, srv_principal, mech_oid, svc,
349 		    &req_options);
350 #else
351 		/*
352 		 * Until changes to the rpcsec_gss code are committed,
353 		 * there is no support for host based initiator
354 		 * principals. As such, that case cannot yet be handled.
355 		 */
356 		if (clnt_principal == NULL)
357 			auth = rpc_gss_secfind(nrp->nr_client, cred,
358 			    srv_principal, mech_oid, svc);
359 		else
360 			auth = NULL;
361 #endif
362 		if (auth != NULL)
363 			return (auth);
364 		/* fallthrough */
365 #endif	/* KGSSAPI */
366 	case AUTH_SYS:
367 	default:
368 		return (authunix_create(cred));
369 
370 	}
371 }
372 
373 /*
374  * Callback from the RPC code to generate up/down notifications.
375  */
376 
377 struct nfs_feedback_arg {
378 	struct nfsmount *nf_mount;
379 	int		nf_lastmsg;	/* last tprintf */
380 	int		nf_tprintfmsg;
381 	struct thread	*nf_td;
382 };
383 
384 static void
385 nfs_feedback(int type, int proc, void *arg)
386 {
387 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
388 	struct nfsmount *nmp = nf->nf_mount;
389 	struct timeval now;
390 
391 	getmicrouptime(&now);
392 
393 	switch (type) {
394 	case FEEDBACK_REXMIT2:
395 	case FEEDBACK_RECONNECT:
396 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
397 			nfs_down(nmp, nf->nf_td,
398 			    "not responding", 0, NFSSTA_TIMEO);
399 			nf->nf_tprintfmsg = TRUE;
400 			nf->nf_lastmsg = now.tv_sec;
401 		}
402 		break;
403 
404 	case FEEDBACK_OK:
405 		nfs_up(nf->nf_mount, nf->nf_td,
406 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
407 		break;
408 	}
409 }
410 
411 /*
412  * newnfs_request - goes something like this
413  *	- does the rpc by calling the krpc layer
414  *	- break down rpc header and return with nfs reply
415  * nb: always frees up nd_mreq mbuf list
416  */
417 int
418 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
419     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
420     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
421     u_char *retsum, int toplevel, u_int64_t *xidp)
422 {
423 	u_int32_t *tl;
424 	time_t waituntil;
425 	int i, j, set_uid = 0, set_sigset = 0;
426 	int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
427 	u_int16_t procnum;
428 	u_int trylater_delay = 1;
429 	struct nfs_feedback_arg nf;
430 	struct timeval timo, now;
431 	AUTH *auth;
432 	struct rpc_callextra ext;
433 	enum clnt_stat stat;
434 	struct nfsreq *rep = NULL;
435 	char *srv_principal = NULL;
436 	uid_t saved_uid = (uid_t)-1;
437 	sigset_t oldset;
438 
439 	if (xidp != NULL)
440 		*xidp = 0;
441 	/* Reject requests while attempting a forced unmount. */
442 	if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
443 		m_freem(nd->nd_mreq);
444 		return (ESTALE);
445 	}
446 
447 	/* For client side interruptible mounts, mask off the signals. */
448 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
449 		newnfs_set_sigmask(td, &oldset);
450 		set_sigset = 1;
451 	}
452 
453 	/*
454 	 * XXX if not already connected call nfs_connect now. Longer
455 	 * term, change nfs_mount to call nfs_connect unconditionally
456 	 * and let clnt_reconnect_create handle reconnects.
457 	 */
458 	if (nrp->nr_client == NULL)
459 		newnfs_connect(nmp, nrp, cred, td, 0);
460 
461 	/*
462 	 * For a client side mount, nmp is != NULL and clp == NULL. For
463 	 * server calls (callbacks or upcalls), nmp == NULL.
464 	 */
465 	if (clp != NULL) {
466 		NFSLOCKSTATE();
467 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
468 			secflavour = RPCSEC_GSS_KRB5;
469 			if (nd->nd_procnum != NFSPROC_NULL) {
470 				if (clp->lc_flags & LCL_GSSINTEGRITY)
471 					secflavour = RPCSEC_GSS_KRB5I;
472 				else if (clp->lc_flags & LCL_GSSPRIVACY)
473 					secflavour = RPCSEC_GSS_KRB5P;
474 			}
475 		}
476 		NFSUNLOCKSTATE();
477 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
478 	     nd->nd_procnum != NFSPROC_NULL) {
479 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
480 			nd->nd_flag |= ND_USEGSSNAME;
481 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
482 			/*
483 			 * If there is a client side host based credential,
484 			 * use that, otherwise use the system uid, if set.
485 			 */
486 			if (nmp->nm_krbnamelen > 0) {
487 				usegssname = 1;
488 			} else if (nmp->nm_uid != (uid_t)-1) {
489 				saved_uid = cred->cr_uid;
490 				cred->cr_uid = nmp->nm_uid;
491 				set_uid = 1;
492 			}
493 		} else if (nmp->nm_krbnamelen == 0 &&
494 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
495 			/*
496 			 * If there is no host based principal name and
497 			 * the system uid is set and this is root, use the
498 			 * system uid, since root won't have user
499 			 * credentials in a credentials cache file.
500 			 */
501 			saved_uid = cred->cr_uid;
502 			cred->cr_uid = nmp->nm_uid;
503 			set_uid = 1;
504 		}
505 		if (NFSHASINTEGRITY(nmp))
506 			secflavour = RPCSEC_GSS_KRB5I;
507 		else if (NFSHASPRIVACY(nmp))
508 			secflavour = RPCSEC_GSS_KRB5P;
509 		else
510 			secflavour = RPCSEC_GSS_KRB5;
511 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
512 	}
513 
514 	if (nmp != NULL) {
515 		bzero(&nf, sizeof(struct nfs_feedback_arg));
516 		nf.nf_mount = nmp;
517 		nf.nf_td = td;
518 		getmicrouptime(&now);
519 		nf.nf_lastmsg = now.tv_sec -
520 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
521 	}
522 
523 	if (nd->nd_procnum == NFSPROC_NULL)
524 		auth = authnone_create();
525 	else if (usegssname)
526 		auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
527 		    srv_principal, NULL, cred);
528 	else
529 		auth = nfs_getauth(nrp, secflavour, NULL,
530 		    srv_principal, NULL, cred);
531 	if (set_uid)
532 		cred->cr_uid = saved_uid;
533 	if (auth == NULL) {
534 		m_freem(nd->nd_mreq);
535 		if (set_sigset)
536 			newnfs_restore_sigmask(td, &oldset);
537 		return (EACCES);
538 	}
539 	bzero(&ext, sizeof(ext));
540 	ext.rc_auth = auth;
541 	if (nmp != NULL) {
542 		ext.rc_feedback = nfs_feedback;
543 		ext.rc_feedback_arg = &nf;
544 	}
545 
546 	procnum = nd->nd_procnum;
547 	if ((nd->nd_flag & ND_NFSV4) &&
548 	    nd->nd_procnum != NFSPROC_NULL &&
549 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
550 		procnum = NFSV4PROC_COMPOUND;
551 
552 	if (nmp != NULL) {
553 		NFSINCRGLOBAL(newnfsstats.rpcrequests);
554 
555 		/* Map the procnum to the old NFSv2 one, as required. */
556 		if ((nd->nd_flag & ND_NFSV2) != 0) {
557 			if (nd->nd_procnum < NFS_V3NPROCS)
558 				procnum = nfsv2_procid[nd->nd_procnum];
559 			else
560 				procnum = NFSV2PROC_NOOP;
561 		}
562 
563 		/*
564 		 * Now only used for the R_DONTRECOVER case, but until that is
565 		 * supported within the krpc code, I need to keep a queue of
566 		 * outstanding RPCs for nfsv4 client requests.
567 		 */
568 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
569 			MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
570 			    M_NFSDREQ, M_WAITOK);
571 	}
572 	trycnt = 0;
573 tryagain:
574 	if (nmp == NULL) {
575 		timo.tv_usec = 0;
576 		if (clp == NULL)
577 			timo.tv_sec = NFSV4_UPCALLTIMEO;
578 		else
579 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
580 	} else {
581 		if (nrp->nr_sotype != SOCK_DGRAM) {
582 			timo.tv_usec = 0;
583 			if ((nmp->nm_flag & NFSMNT_NFSV4))
584 				timo.tv_sec = INT_MAX;
585 			else
586 				timo.tv_sec = NFS_TCPTIMEO;
587 		} else {
588 			timo.tv_sec = nmp->nm_timeo / NFS_HZ;
589 			timo.tv_usec = (nmp->nm_timeo * 1000000) / NFS_HZ;
590 		}
591 
592 		if (rep != NULL) {
593 			rep->r_flags = 0;
594 			rep->r_nmp = nmp;
595 			/*
596 			 * Chain request into list of outstanding requests.
597 			 */
598 			NFSLOCKREQ();
599 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
600 			NFSUNLOCKREQ();
601 		}
602 	}
603 
604 	nd->nd_mrep = NULL;
605 	stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
606 	    &nd->nd_mrep, timo);
607 
608 	if (rep != NULL) {
609 		/*
610 		 * RPC done, unlink the request.
611 		 */
612 		NFSLOCKREQ();
613 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
614 		NFSUNLOCKREQ();
615 	}
616 
617 	/*
618 	 * If there was a successful reply and a tprintf msg.
619 	 * tprintf a response.
620 	 */
621 	if (stat == RPC_SUCCESS) {
622 		error = 0;
623 	} else if (stat == RPC_TIMEDOUT) {
624 		error = ETIMEDOUT;
625 	} else if (stat == RPC_VERSMISMATCH) {
626 		error = EOPNOTSUPP;
627 	} else if (stat == RPC_PROGVERSMISMATCH) {
628 		error = EPROTONOSUPPORT;
629 	} else {
630 		error = EACCES;
631 	}
632 	if (error) {
633 		m_freem(nd->nd_mreq);
634 		AUTH_DESTROY(auth);
635 		if (rep != NULL)
636 			FREE((caddr_t)rep, M_NFSDREQ);
637 		if (set_sigset)
638 			newnfs_restore_sigmask(td, &oldset);
639 		return (error);
640 	}
641 
642 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
643 
644 	/*
645 	 * Search for any mbufs that are not a multiple of 4 bytes long
646 	 * or with m_data not longword aligned.
647 	 * These could cause pointer alignment problems, so copy them to
648 	 * well aligned mbufs.
649 	 */
650 	newnfs_realign(&nd->nd_mrep);
651 	nd->nd_md = nd->nd_mrep;
652 	nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
653 	nd->nd_repstat = 0;
654 	if (nd->nd_procnum != NFSPROC_NULL) {
655 		/*
656 		 * and now the actual NFS xdr.
657 		 */
658 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
659 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
660 		if (nd->nd_repstat != 0) {
661 			if ((nd->nd_repstat == NFSERR_DELAY &&
662 			     (nd->nd_flag & ND_NFSV4) &&
663 			     nd->nd_procnum != NFSPROC_SETATTR &&
664 			     nd->nd_procnum != NFSPROC_READ &&
665 			     nd->nd_procnum != NFSPROC_WRITE &&
666 			     nd->nd_procnum != NFSPROC_OPEN &&
667 			     nd->nd_procnum != NFSPROC_CREATE &&
668 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
669 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
670 			     nd->nd_procnum != NFSPROC_CLOSE &&
671 			     nd->nd_procnum != NFSPROC_LOCK &&
672 			     nd->nd_procnum != NFSPROC_LOCKU) ||
673 			    (nd->nd_repstat == NFSERR_DELAY &&
674 			     (nd->nd_flag & ND_NFSV4) == 0) ||
675 			    nd->nd_repstat == NFSERR_RESOURCE) {
676 				if (trylater_delay > NFS_TRYLATERDEL)
677 					trylater_delay = NFS_TRYLATERDEL;
678 				waituntil = NFSD_MONOSEC + trylater_delay;
679 				while (NFSD_MONOSEC < waituntil)
680 					(void) nfs_catnap(PZERO, 0, "nfstry");
681 				trylater_delay *= 2;
682 				goto tryagain;
683 			}
684 
685 			/*
686 			 * If the File Handle was stale, invalidate the
687 			 * lookup cache, just in case.
688 			 * (vp != NULL implies a client side call)
689 			 */
690 			if (nd->nd_repstat == ESTALE && vp != NULL) {
691 				cache_purge(vp);
692 				if (ncl_call_invalcaches != NULL)
693 					(*ncl_call_invalcaches)(vp);
694 			}
695 		}
696 
697 		/*
698 		 * Get rid of the tag, return count, and PUTFH result for V4.
699 		 */
700 		if (nd->nd_flag & ND_NFSV4) {
701 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
702 			i = fxdr_unsigned(int, *tl);
703 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
704 			if (error)
705 				goto nfsmout;
706 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
707 			i = fxdr_unsigned(int, *++tl);
708 
709 			/*
710 			 * If the first op's status is non-zero, mark that
711 			 * there is no more data to process.
712 			 */
713 			if (*++tl)
714 				nd->nd_flag |= ND_NOMOREDATA;
715 
716 			/*
717 			 * If the first op is Putfh, throw its results away
718 			 * and toss the op# and status for the first op.
719 			 */
720 			if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) {
721 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
722 				i = fxdr_unsigned(int, *tl++);
723 				j = fxdr_unsigned(int, *tl);
724 				/*
725 				 * All Compounds that do an Op that must
726 				 * be in sequence consist of NFSV4OP_PUTFH
727 				 * followed by one of these. As such, we
728 				 * can determine if the seqid# should be
729 				 * incremented, here.
730 				 */
731 				if ((i == NFSV4OP_OPEN ||
732 				     i == NFSV4OP_OPENCONFIRM ||
733 				     i == NFSV4OP_OPENDOWNGRADE ||
734 				     i == NFSV4OP_CLOSE ||
735 				     i == NFSV4OP_LOCK ||
736 				     i == NFSV4OP_LOCKU) &&
737 				    (j == 0 ||
738 				     (j != NFSERR_STALECLIENTID &&
739 				      j != NFSERR_STALESTATEID &&
740 				      j != NFSERR_BADSTATEID &&
741 				      j != NFSERR_BADSEQID &&
742 				      j != NFSERR_BADXDR &&
743 				      j != NFSERR_RESOURCE &&
744 				      j != NFSERR_NOFILEHANDLE)))
745 					nd->nd_flag |= ND_INCRSEQID;
746 				/*
747 				 * If the first op's status is non-zero, mark
748 				 * that there is no more data to process.
749 				 */
750 				if (j)
751 					nd->nd_flag |= ND_NOMOREDATA;
752 			}
753 
754 			/*
755 			 * If R_DONTRECOVER is set, replace the stale error
756 			 * reply, so that recovery isn't initiated.
757 			 */
758 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
759 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
760 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
761 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
762 		}
763 	}
764 
765 	m_freem(nd->nd_mreq);
766 	AUTH_DESTROY(auth);
767 	if (rep != NULL)
768 		FREE((caddr_t)rep, M_NFSDREQ);
769 	if (set_sigset)
770 		newnfs_restore_sigmask(td, &oldset);
771 	return (0);
772 nfsmout:
773 	mbuf_freem(nd->nd_mrep);
774 	mbuf_freem(nd->nd_mreq);
775 	AUTH_DESTROY(auth);
776 	if (rep != NULL)
777 		FREE((caddr_t)rep, M_NFSDREQ);
778 	if (set_sigset)
779 		newnfs_restore_sigmask(td, &oldset);
780 	return (error);
781 }
782 
783 /*
784  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
785  * wait for all requests to complete. This is used by forced unmounts
786  * to terminate any outstanding RPCs.
787  */
788 int
789 newnfs_nmcancelreqs(struct nfsmount *nmp)
790 {
791 
792 	if (nmp->nm_sockreq.nr_client != NULL)
793 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
794 	return (0);
795 }
796 
797 /*
798  * Any signal that can interrupt an NFS operation in an intr mount
799  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
800  */
801 int newnfs_sig_set[] = {
802 	SIGINT,
803 	SIGTERM,
804 	SIGHUP,
805 	SIGKILL,
806 	SIGSTOP,
807 	SIGQUIT
808 };
809 
810 /*
811  * Check to see if one of the signals in our subset is pending on
812  * the process (in an intr mount).
813  */
814 static int
815 nfs_sig_pending(sigset_t set)
816 {
817 	int i;
818 
819 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
820 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
821 			return (1);
822 	return (0);
823 }
824 
825 /*
826  * The set/restore sigmask functions are used to (temporarily) overwrite
827  * the process p_sigmask during an RPC call (for example). These are also
828  * used in other places in the NFS client that might tsleep().
829  */
830 void
831 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
832 {
833 	sigset_t newset;
834 	int i;
835 	struct proc *p;
836 
837 	SIGFILLSET(newset);
838 	if (td == NULL)
839 		td = curthread; /* XXX */
840 	p = td->td_proc;
841 	/* Remove the NFS set of signals from newset */
842 	PROC_LOCK(p);
843 	mtx_lock(&p->p_sigacts->ps_mtx);
844 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
845 		/*
846 		 * But make sure we leave the ones already masked
847 		 * by the process, ie. remove the signal from the
848 		 * temporary signalmask only if it wasn't already
849 		 * in p_sigmask.
850 		 */
851 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
852 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
853 			SIGDELSET(newset, newnfs_sig_set[i]);
854 	}
855 	mtx_unlock(&p->p_sigacts->ps_mtx);
856 	PROC_UNLOCK(p);
857 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0);
858 }
859 
860 void
861 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
862 {
863 	if (td == NULL)
864 		td = curthread; /* XXX */
865 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
866 }
867 
868 /*
869  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
870  * old one after msleep() returns.
871  */
872 int
873 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
874 {
875 	sigset_t oldset;
876 	int error;
877 	struct proc *p;
878 
879 	if ((priority & PCATCH) == 0)
880 		return msleep(ident, mtx, priority, wmesg, timo);
881 	if (td == NULL)
882 		td = curthread; /* XXX */
883 	newnfs_set_sigmask(td, &oldset);
884 	error = msleep(ident, mtx, priority, wmesg, timo);
885 	newnfs_restore_sigmask(td, &oldset);
886 	p = td->td_proc;
887 	return (error);
888 }
889 
890 /*
891  * Test for a termination condition pending on the process.
892  * This is used for NFSMNT_INT mounts.
893  */
894 int
895 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
896 {
897 	struct proc *p;
898 	sigset_t tmpset;
899 
900 	/* Terminate all requests while attempting a forced unmount. */
901 	if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
902 		return (EIO);
903 	if (!(nmp->nm_flag & NFSMNT_INT))
904 		return (0);
905 	if (td == NULL)
906 		return (0);
907 	p = td->td_proc;
908 	PROC_LOCK(p);
909 	tmpset = p->p_siglist;
910 	SIGSETOR(tmpset, td->td_siglist);
911 	SIGSETNAND(tmpset, td->td_sigmask);
912 	mtx_lock(&p->p_sigacts->ps_mtx);
913 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
914 	mtx_unlock(&p->p_sigacts->ps_mtx);
915 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
916 	    && nfs_sig_pending(tmpset)) {
917 		PROC_UNLOCK(p);
918 		return (EINTR);
919 	}
920 	PROC_UNLOCK(p);
921 	return (0);
922 }
923 
924 static int
925 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
926 {
927 	struct proc *p;
928 
929 	p = td ? td->td_proc : NULL;
930 	if (error) {
931 		tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
932 		    server, msg, error);
933 	} else {
934 		tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
935 	}
936 	return (0);
937 }
938 
939 static void
940 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
941     int error, int flags)
942 {
943 	if (nmp == NULL)
944 		return;
945 	mtx_lock(&nmp->nm_mtx);
946 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
947 		nmp->nm_state |= NFSSTA_TIMEO;
948 		mtx_unlock(&nmp->nm_mtx);
949 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
950 		    VQ_NOTRESP, 0);
951 	} else
952 		mtx_unlock(&nmp->nm_mtx);
953 	mtx_lock(&nmp->nm_mtx);
954 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
955 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
956 		mtx_unlock(&nmp->nm_mtx);
957 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
958 		    VQ_NOTRESPLOCK, 0);
959 	} else
960 		mtx_unlock(&nmp->nm_mtx);
961 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
962 }
963 
964 static void
965 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
966     int flags, int tprintfmsg)
967 {
968 	if (nmp == NULL)
969 		return;
970 	if (tprintfmsg) {
971 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
972 	}
973 
974 	mtx_lock(&nmp->nm_mtx);
975 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
976 		nmp->nm_state &= ~NFSSTA_TIMEO;
977 		mtx_unlock(&nmp->nm_mtx);
978 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
979 		    VQ_NOTRESP, 1);
980 	} else
981 		mtx_unlock(&nmp->nm_mtx);
982 
983 	mtx_lock(&nmp->nm_mtx);
984 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
985 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
986 		mtx_unlock(&nmp->nm_mtx);
987 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
988 		    VQ_NOTRESPLOCK, 1);
989 	} else
990 		mtx_unlock(&nmp->nm_mtx);
991 }
992 
993