xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision de138ec703c3f7d8ed73ba94068d571b2ee7a87c)
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_kdtrace.h"
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 
64 #include <kgssapi/krb5/kcrypto.h>
65 
66 #include <fs/nfs/nfsport.h>
67 
68 #ifdef KDTRACE_HOOKS
69 #include <sys/dtrace_bsd.h>
70 
71 dtrace_nfsclient_nfs23_start_probe_func_t
72 		dtrace_nfscl_nfs234_start_probe;
73 
74 dtrace_nfsclient_nfs23_done_probe_func_t
75 		dtrace_nfscl_nfs234_done_probe;
76 
77 /*
78  * Registered probes by RPC type.
79  */
80 uint32_t	nfscl_nfs2_start_probes[NFS_NPROCS + 1];
81 uint32_t	nfscl_nfs2_done_probes[NFS_NPROCS + 1];
82 
83 uint32_t	nfscl_nfs3_start_probes[NFS_NPROCS + 1];
84 uint32_t	nfscl_nfs3_done_probes[NFS_NPROCS + 1];
85 
86 uint32_t	nfscl_nfs4_start_probes[NFS_NPROCS + 1];
87 uint32_t	nfscl_nfs4_done_probes[NFS_NPROCS + 1];
88 #endif
89 
90 NFSSTATESPINLOCK;
91 NFSREQSPINLOCK;
92 extern struct nfsstats newnfsstats;
93 extern struct nfsreqhead nfsd_reqq;
94 extern int nfscl_ticks;
95 extern void (*ncl_call_invalcaches)(struct vnode *);
96 
97 static int	nfsrv_gsscallbackson = 0;
98 static int	nfs_bufpackets = 4;
99 static int	nfs_reconnects;
100 static int	nfs3_jukebox_delay = 10;
101 static int	nfs_skip_wcc_data_onerr = 1;
102 static int	nfs_keytab_enctype = ETYPE_DES_CBC_CRC;
103 
104 SYSCTL_DECL(_vfs_nfs);
105 
106 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
107     "Buffer reservation size 2 < x < 64");
108 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
109     "Number of times the nfs client has had to reconnect");
110 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
111     "Number of seconds to delay a retry after receiving EJUKEBOX");
112 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
113     "Disable weak cache consistency checking when server returns an error");
114 SYSCTL_INT(_vfs_nfs, OID_AUTO, keytab_enctype, CTLFLAG_RW, &nfs_keytab_enctype, 0,
115     "Encryption type for the keytab entry used by nfs");
116 
117 static void	nfs_down(struct nfsmount *, struct thread *, const char *,
118     int, int);
119 static void	nfs_up(struct nfsmount *, struct thread *, const char *,
120     int, int);
121 static int	nfs_msg(struct thread *, const char *, const char *, int);
122 
123 struct nfs_cached_auth {
124 	int		ca_refs; /* refcount, including 1 from the cache */
125 	uid_t		ca_uid;	 /* uid that corresponds to this auth */
126 	AUTH		*ca_auth; /* RPC auth handle */
127 };
128 
129 static int nfsv2_procid[NFS_V3NPROCS] = {
130 	NFSV2PROC_NULL,
131 	NFSV2PROC_GETATTR,
132 	NFSV2PROC_SETATTR,
133 	NFSV2PROC_LOOKUP,
134 	NFSV2PROC_NOOP,
135 	NFSV2PROC_READLINK,
136 	NFSV2PROC_READ,
137 	NFSV2PROC_WRITE,
138 	NFSV2PROC_CREATE,
139 	NFSV2PROC_MKDIR,
140 	NFSV2PROC_SYMLINK,
141 	NFSV2PROC_CREATE,
142 	NFSV2PROC_REMOVE,
143 	NFSV2PROC_RMDIR,
144 	NFSV2PROC_RENAME,
145 	NFSV2PROC_LINK,
146 	NFSV2PROC_READDIR,
147 	NFSV2PROC_NOOP,
148 	NFSV2PROC_STATFS,
149 	NFSV2PROC_NOOP,
150 	NFSV2PROC_NOOP,
151 	NFSV2PROC_NOOP,
152 };
153 
154 /*
155  * Initialize sockets and congestion for a new NFS connection.
156  * We do not free the sockaddr if error.
157  */
158 int
159 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
160     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
161 {
162 	int rcvreserve, sndreserve;
163 	int pktscale;
164 	struct sockaddr *saddr;
165 	struct ucred *origcred;
166 	CLIENT *client;
167 	struct netconfig *nconf;
168 	struct socket *so;
169 	int one = 1, retries, error;
170 	struct thread *td = curthread;
171 
172 	/*
173 	 * We need to establish the socket using the credentials of
174 	 * the mountpoint.  Some parts of this process (such as
175 	 * sobind() and soconnect()) will use the curent thread's
176 	 * credential instead of the socket credential.  To work
177 	 * around this, temporarily change the current thread's
178 	 * credential to that of the mountpoint.
179 	 *
180 	 * XXX: It would be better to explicitly pass the correct
181 	 * credential to sobind() and soconnect().
182 	 */
183 	origcred = td->td_ucred;
184 
185 	/*
186 	 * Use the credential in nr_cred, if not NULL.
187 	 */
188 	if (nrp->nr_cred != NULL)
189 		td->td_ucred = nrp->nr_cred;
190 	else
191 		td->td_ucred = cred;
192 	saddr = nrp->nr_nam;
193 
194 	if (saddr->sa_family == AF_INET)
195 		if (nrp->nr_sotype == SOCK_DGRAM)
196 			nconf = getnetconfigent("udp");
197 		else
198 			nconf = getnetconfigent("tcp");
199 	else
200 		if (nrp->nr_sotype == SOCK_DGRAM)
201 			nconf = getnetconfigent("udp6");
202 		else
203 			nconf = getnetconfigent("tcp6");
204 
205 	pktscale = nfs_bufpackets;
206 	if (pktscale < 2)
207 		pktscale = 2;
208 	if (pktscale > 64)
209 		pktscale = 64;
210 	/*
211 	 * soreserve() can fail if sb_max is too small, so shrink pktscale
212 	 * and try again if there is an error.
213 	 * Print a log message suggesting increasing sb_max.
214 	 * Creating a socket and doing this is necessary since, if the
215 	 * reservation sizes are too large and will make soreserve() fail,
216 	 * the connection will work until a large send is attempted and
217 	 * then it will loop in the krpc code.
218 	 */
219 	so = NULL;
220 	saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
221 	error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
222 	    nrp->nr_soproto, td->td_ucred, td);
223 	if (error) {
224 		td->td_ucred = origcred;
225 		return (error);
226 	}
227 	do {
228 	    if (error != 0 && pktscale > 2)
229 		pktscale--;
230 	    if (nrp->nr_sotype == SOCK_DGRAM) {
231 		if (nmp != NULL) {
232 			sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
233 			    pktscale;
234 			rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
235 			    pktscale;
236 		} else {
237 			sndreserve = rcvreserve = 1024 * pktscale;
238 		}
239 	    } else {
240 		if (nrp->nr_sotype != SOCK_STREAM)
241 			panic("nfscon sotype");
242 		if (nmp != NULL) {
243 			sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
244 			    sizeof (u_int32_t)) * pktscale;
245 			rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
246 			    sizeof (u_int32_t)) * pktscale;
247 		} else {
248 			sndreserve = rcvreserve = 1024 * pktscale;
249 		}
250 	    }
251 	    error = soreserve(so, sndreserve, rcvreserve);
252 	} while (error != 0 && pktscale > 2);
253 	soclose(so);
254 	if (error) {
255 		td->td_ucred = origcred;
256 		return (error);
257 	}
258 
259 	client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
260 	    nrp->nr_vers, sndreserve, rcvreserve);
261 	CLNT_CONTROL(client, CLSET_WAITCHAN, "newnfsreq");
262 	if (nmp != NULL) {
263 		if ((nmp->nm_flag & NFSMNT_INT))
264 			CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
265 		if ((nmp->nm_flag & NFSMNT_RESVPORT))
266 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
267 		if (NFSHASSOFT(nmp))
268 			retries = nmp->nm_retry;
269 		else
270 			retries = INT_MAX;
271 	} else {
272 		/*
273 		 * Three cases:
274 		 * - Null RPC callback to client
275 		 * - Non-Null RPC callback to client, wait a little longer
276 		 * - upcalls to nfsuserd and gssd (clp == NULL)
277 		 */
278 		if (callback_retry_mult == 0) {
279 			retries = NFSV4_UPCALLRETRY;
280 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
281 		} else {
282 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
283 		}
284 	}
285 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
286 
287 	mtx_lock(&nrp->nr_mtx);
288 	if (nrp->nr_client != NULL) {
289 		/*
290 		 * Someone else already connected.
291 		 */
292 		CLNT_RELEASE(client);
293 	} else {
294 		nrp->nr_client = client;
295 	}
296 
297 	/*
298 	 * Protocols that do not require connections may be optionally left
299 	 * unconnected for servers that reply from a port other than NFS_PORT.
300 	 */
301 	if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
302 		mtx_unlock(&nrp->nr_mtx);
303 		CLNT_CONTROL(client, CLSET_CONNECT, &one);
304 	} else {
305 		mtx_unlock(&nrp->nr_mtx);
306 	}
307 
308 	/* Restore current thread's credentials. */
309 	td->td_ucred = origcred;
310 	return (0);
311 }
312 
313 /*
314  * NFS disconnect. Clean up and unlink.
315  */
316 void
317 newnfs_disconnect(struct nfssockreq *nrp)
318 {
319 	CLIENT *client;
320 
321 	mtx_lock(&nrp->nr_mtx);
322 	if (nrp->nr_client != NULL) {
323 		client = nrp->nr_client;
324 		nrp->nr_client = NULL;
325 		mtx_unlock(&nrp->nr_mtx);
326 		rpc_gss_secpurge_call(client);
327 		CLNT_CLOSE(client);
328 		CLNT_RELEASE(client);
329 	} else {
330 		mtx_unlock(&nrp->nr_mtx);
331 	}
332 }
333 
334 static AUTH *
335 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
336     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
337 {
338 	rpc_gss_service_t svc;
339 	AUTH *auth;
340 #ifdef notyet
341 	rpc_gss_options_req_t req_options;
342 #endif
343 
344 	switch (secflavour) {
345 	case RPCSEC_GSS_KRB5:
346 	case RPCSEC_GSS_KRB5I:
347 	case RPCSEC_GSS_KRB5P:
348 		if (!mech_oid) {
349 			if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
350 				return (NULL);
351 		}
352 		if (secflavour == RPCSEC_GSS_KRB5)
353 			svc = rpc_gss_svc_none;
354 		else if (secflavour == RPCSEC_GSS_KRB5I)
355 			svc = rpc_gss_svc_integrity;
356 		else
357 			svc = rpc_gss_svc_privacy;
358 #ifdef notyet
359 		req_options.req_flags = GSS_C_MUTUAL_FLAG;
360 		req_options.time_req = 0;
361 		req_options.my_cred = GSS_C_NO_CREDENTIAL;
362 		req_options.input_channel_bindings = NULL;
363 		req_options.enc_type = nfs_keytab_enctype;
364 
365 		auth = rpc_gss_secfind_call(nrp->nr_client, cred,
366 		    clnt_principal, srv_principal, mech_oid, svc,
367 		    &req_options);
368 #else
369 		/*
370 		 * Until changes to the rpcsec_gss code are committed,
371 		 * there is no support for host based initiator
372 		 * principals. As such, that case cannot yet be handled.
373 		 */
374 		if (clnt_principal == NULL)
375 			auth = rpc_gss_secfind_call(nrp->nr_client, cred,
376 			    srv_principal, mech_oid, svc);
377 		else
378 			auth = NULL;
379 #endif
380 		if (auth != NULL)
381 			return (auth);
382 		/* fallthrough */
383 	case AUTH_SYS:
384 	default:
385 		return (authunix_create(cred));
386 
387 	}
388 }
389 
390 /*
391  * Callback from the RPC code to generate up/down notifications.
392  */
393 
394 struct nfs_feedback_arg {
395 	struct nfsmount *nf_mount;
396 	int		nf_lastmsg;	/* last tprintf */
397 	int		nf_tprintfmsg;
398 	struct thread	*nf_td;
399 };
400 
401 static void
402 nfs_feedback(int type, int proc, void *arg)
403 {
404 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
405 	struct nfsmount *nmp = nf->nf_mount;
406 	struct timeval now;
407 
408 	getmicrouptime(&now);
409 
410 	switch (type) {
411 	case FEEDBACK_REXMIT2:
412 	case FEEDBACK_RECONNECT:
413 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
414 			nfs_down(nmp, nf->nf_td,
415 			    "not responding", 0, NFSSTA_TIMEO);
416 			nf->nf_tprintfmsg = TRUE;
417 			nf->nf_lastmsg = now.tv_sec;
418 		}
419 		break;
420 
421 	case FEEDBACK_OK:
422 		nfs_up(nf->nf_mount, nf->nf_td,
423 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
424 		break;
425 	}
426 }
427 
428 /*
429  * newnfs_request - goes something like this
430  *	- does the rpc by calling the krpc layer
431  *	- break down rpc header and return with nfs reply
432  * nb: always frees up nd_mreq mbuf list
433  */
434 int
435 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
436     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
437     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
438     u_char *retsum, int toplevel, u_int64_t *xidp)
439 {
440 	u_int32_t *tl;
441 	time_t waituntil;
442 	int i, j, set_uid = 0, set_sigset = 0;
443 	int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
444 	u_int16_t procnum;
445 	u_int trylater_delay = 1;
446 	struct nfs_feedback_arg nf;
447 	struct timeval timo, now;
448 	AUTH *auth;
449 	struct rpc_callextra ext;
450 	enum clnt_stat stat;
451 	struct nfsreq *rep = NULL;
452 	char *srv_principal = NULL;
453 	uid_t saved_uid = (uid_t)-1;
454 	sigset_t oldset;
455 
456 	if (xidp != NULL)
457 		*xidp = 0;
458 	/* Reject requests while attempting a forced unmount. */
459 	if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
460 		m_freem(nd->nd_mreq);
461 		return (ESTALE);
462 	}
463 
464 	/* For client side interruptible mounts, mask off the signals. */
465 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
466 		newnfs_set_sigmask(td, &oldset);
467 		set_sigset = 1;
468 	}
469 
470 	/*
471 	 * XXX if not already connected call nfs_connect now. Longer
472 	 * term, change nfs_mount to call nfs_connect unconditionally
473 	 * and let clnt_reconnect_create handle reconnects.
474 	 */
475 	if (nrp->nr_client == NULL)
476 		newnfs_connect(nmp, nrp, cred, td, 0);
477 
478 	/*
479 	 * For a client side mount, nmp is != NULL and clp == NULL. For
480 	 * server calls (callbacks or upcalls), nmp == NULL.
481 	 */
482 	if (clp != NULL) {
483 		NFSLOCKSTATE();
484 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
485 			secflavour = RPCSEC_GSS_KRB5;
486 			if (nd->nd_procnum != NFSPROC_NULL) {
487 				if (clp->lc_flags & LCL_GSSINTEGRITY)
488 					secflavour = RPCSEC_GSS_KRB5I;
489 				else if (clp->lc_flags & LCL_GSSPRIVACY)
490 					secflavour = RPCSEC_GSS_KRB5P;
491 			}
492 		}
493 		NFSUNLOCKSTATE();
494 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
495 	     nd->nd_procnum != NFSPROC_NULL) {
496 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
497 			nd->nd_flag |= ND_USEGSSNAME;
498 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
499 			/*
500 			 * If there is a client side host based credential,
501 			 * use that, otherwise use the system uid, if set.
502 			 */
503 			if (nmp->nm_krbnamelen > 0) {
504 				usegssname = 1;
505 			} else if (nmp->nm_uid != (uid_t)-1) {
506 				saved_uid = cred->cr_uid;
507 				cred->cr_uid = nmp->nm_uid;
508 				set_uid = 1;
509 			}
510 		} else if (nmp->nm_krbnamelen == 0 &&
511 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
512 			/*
513 			 * If there is no host based principal name and
514 			 * the system uid is set and this is root, use the
515 			 * system uid, since root won't have user
516 			 * credentials in a credentials cache file.
517 			 */
518 			saved_uid = cred->cr_uid;
519 			cred->cr_uid = nmp->nm_uid;
520 			set_uid = 1;
521 		}
522 		if (NFSHASINTEGRITY(nmp))
523 			secflavour = RPCSEC_GSS_KRB5I;
524 		else if (NFSHASPRIVACY(nmp))
525 			secflavour = RPCSEC_GSS_KRB5P;
526 		else
527 			secflavour = RPCSEC_GSS_KRB5;
528 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
529 	} else if (nmp != NULL && !NFSHASKERB(nmp) &&
530 	    nd->nd_procnum != NFSPROC_NULL &&
531 	    (nd->nd_flag & ND_USEGSSNAME) != 0) {
532 		/*
533 		 * Use the uid that did the mount when the RPC is doing
534 		 * NFSv4 system operations, as indicated by the
535 		 * ND_USEGSSNAME flag, for the AUTH_SYS case.
536 		 */
537 		saved_uid = cred->cr_uid;
538 		if (nmp->nm_uid != (uid_t)-1)
539 			cred->cr_uid = nmp->nm_uid;
540 		else
541 			cred->cr_uid = 0;
542 		set_uid = 1;
543 	}
544 
545 	if (nmp != NULL) {
546 		bzero(&nf, sizeof(struct nfs_feedback_arg));
547 		nf.nf_mount = nmp;
548 		nf.nf_td = td;
549 		getmicrouptime(&now);
550 		nf.nf_lastmsg = now.tv_sec -
551 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
552 	}
553 
554 	if (nd->nd_procnum == NFSPROC_NULL)
555 		auth = authnone_create();
556 	else if (usegssname)
557 		auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
558 		    srv_principal, NULL, cred);
559 	else
560 		auth = nfs_getauth(nrp, secflavour, NULL,
561 		    srv_principal, NULL, cred);
562 	if (set_uid)
563 		cred->cr_uid = saved_uid;
564 	if (auth == NULL) {
565 		m_freem(nd->nd_mreq);
566 		if (set_sigset)
567 			newnfs_restore_sigmask(td, &oldset);
568 		return (EACCES);
569 	}
570 	bzero(&ext, sizeof(ext));
571 	ext.rc_auth = auth;
572 	if (nmp != NULL) {
573 		ext.rc_feedback = nfs_feedback;
574 		ext.rc_feedback_arg = &nf;
575 	}
576 
577 	procnum = nd->nd_procnum;
578 	if ((nd->nd_flag & ND_NFSV4) &&
579 	    nd->nd_procnum != NFSPROC_NULL &&
580 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
581 		procnum = NFSV4PROC_COMPOUND;
582 
583 	if (nmp != NULL) {
584 		NFSINCRGLOBAL(newnfsstats.rpcrequests);
585 
586 		/* Map the procnum to the old NFSv2 one, as required. */
587 		if ((nd->nd_flag & ND_NFSV2) != 0) {
588 			if (nd->nd_procnum < NFS_V3NPROCS)
589 				procnum = nfsv2_procid[nd->nd_procnum];
590 			else
591 				procnum = NFSV2PROC_NOOP;
592 		}
593 
594 		/*
595 		 * Now only used for the R_DONTRECOVER case, but until that is
596 		 * supported within the krpc code, I need to keep a queue of
597 		 * outstanding RPCs for nfsv4 client requests.
598 		 */
599 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
600 			MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
601 			    M_NFSDREQ, M_WAITOK);
602 #ifdef KDTRACE_HOOKS
603 		if (dtrace_nfscl_nfs234_start_probe != NULL) {
604 			uint32_t probe_id;
605 			int probe_procnum;
606 
607 			if (nd->nd_flag & ND_NFSV4) {
608 				probe_id =
609 				    nfscl_nfs4_start_probes[nd->nd_procnum];
610 				probe_procnum = nd->nd_procnum;
611 			} else if (nd->nd_flag & ND_NFSV3) {
612 				probe_id = nfscl_nfs3_start_probes[procnum];
613 				probe_procnum = procnum;
614 			} else {
615 				probe_id =
616 				    nfscl_nfs2_start_probes[nd->nd_procnum];
617 				probe_procnum = procnum;
618 			}
619 			if (probe_id != 0)
620 				(dtrace_nfscl_nfs234_start_probe)
621 				    (probe_id, vp, nd->nd_mreq, cred,
622 				     probe_procnum);
623 		}
624 #endif
625 	}
626 	trycnt = 0;
627 tryagain:
628 	if (nmp == NULL) {
629 		timo.tv_usec = 0;
630 		if (clp == NULL)
631 			timo.tv_sec = NFSV4_UPCALLTIMEO;
632 		else
633 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
634 	} else {
635 		if (nrp->nr_sotype != SOCK_DGRAM) {
636 			timo.tv_usec = 0;
637 			if ((nmp->nm_flag & NFSMNT_NFSV4))
638 				timo.tv_sec = INT_MAX;
639 			else
640 				timo.tv_sec = NFS_TCPTIMEO;
641 		} else {
642 			timo.tv_sec = nmp->nm_timeo / NFS_HZ;
643 			timo.tv_usec = (nmp->nm_timeo * 1000000) / NFS_HZ;
644 		}
645 
646 		if (rep != NULL) {
647 			rep->r_flags = 0;
648 			rep->r_nmp = nmp;
649 			/*
650 			 * Chain request into list of outstanding requests.
651 			 */
652 			NFSLOCKREQ();
653 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
654 			NFSUNLOCKREQ();
655 		}
656 	}
657 
658 	nd->nd_mrep = NULL;
659 	stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
660 	    &nd->nd_mrep, timo);
661 
662 	if (rep != NULL) {
663 		/*
664 		 * RPC done, unlink the request.
665 		 */
666 		NFSLOCKREQ();
667 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
668 		NFSUNLOCKREQ();
669 	}
670 
671 	/*
672 	 * If there was a successful reply and a tprintf msg.
673 	 * tprintf a response.
674 	 */
675 	if (stat == RPC_SUCCESS) {
676 		error = 0;
677 	} else if (stat == RPC_TIMEDOUT) {
678 		error = ETIMEDOUT;
679 	} else if (stat == RPC_VERSMISMATCH) {
680 		error = EOPNOTSUPP;
681 	} else if (stat == RPC_PROGVERSMISMATCH) {
682 		error = EPROTONOSUPPORT;
683 	} else {
684 		error = EACCES;
685 	}
686 	if (error) {
687 		m_freem(nd->nd_mreq);
688 		AUTH_DESTROY(auth);
689 		if (rep != NULL)
690 			FREE((caddr_t)rep, M_NFSDREQ);
691 		if (set_sigset)
692 			newnfs_restore_sigmask(td, &oldset);
693 		return (error);
694 	}
695 
696 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
697 
698 	/*
699 	 * Search for any mbufs that are not a multiple of 4 bytes long
700 	 * or with m_data not longword aligned.
701 	 * These could cause pointer alignment problems, so copy them to
702 	 * well aligned mbufs.
703 	 */
704 	newnfs_realign(&nd->nd_mrep);
705 	nd->nd_md = nd->nd_mrep;
706 	nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
707 	nd->nd_repstat = 0;
708 	if (nd->nd_procnum != NFSPROC_NULL) {
709 		/*
710 		 * and now the actual NFS xdr.
711 		 */
712 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
713 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
714 		if (nd->nd_repstat != 0) {
715 			if ((nd->nd_repstat == NFSERR_DELAY &&
716 			     (nd->nd_flag & ND_NFSV4) &&
717 			     nd->nd_procnum != NFSPROC_SETATTR &&
718 			     nd->nd_procnum != NFSPROC_READ &&
719 			     nd->nd_procnum != NFSPROC_WRITE &&
720 			     nd->nd_procnum != NFSPROC_OPEN &&
721 			     nd->nd_procnum != NFSPROC_CREATE &&
722 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
723 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
724 			     nd->nd_procnum != NFSPROC_CLOSE &&
725 			     nd->nd_procnum != NFSPROC_LOCK &&
726 			     nd->nd_procnum != NFSPROC_LOCKU) ||
727 			    (nd->nd_repstat == NFSERR_DELAY &&
728 			     (nd->nd_flag & ND_NFSV4) == 0) ||
729 			    nd->nd_repstat == NFSERR_RESOURCE) {
730 				if (trylater_delay > NFS_TRYLATERDEL)
731 					trylater_delay = NFS_TRYLATERDEL;
732 				waituntil = NFSD_MONOSEC + trylater_delay;
733 				while (NFSD_MONOSEC < waituntil)
734 					(void) nfs_catnap(PZERO, 0, "nfstry");
735 				trylater_delay *= 2;
736 				m_freem(nd->nd_mrep);
737 				nd->nd_mrep = NULL;
738 				goto tryagain;
739 			}
740 
741 			/*
742 			 * If the File Handle was stale, invalidate the
743 			 * lookup cache, just in case.
744 			 * (vp != NULL implies a client side call)
745 			 */
746 			if (nd->nd_repstat == ESTALE && vp != NULL) {
747 				cache_purge(vp);
748 				if (ncl_call_invalcaches != NULL)
749 					(*ncl_call_invalcaches)(vp);
750 			}
751 		}
752 
753 		/*
754 		 * Get rid of the tag, return count, and PUTFH result for V4.
755 		 */
756 		if (nd->nd_flag & ND_NFSV4) {
757 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
758 			i = fxdr_unsigned(int, *tl);
759 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
760 			if (error)
761 				goto nfsmout;
762 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
763 			i = fxdr_unsigned(int, *++tl);
764 
765 			/*
766 			 * If the first op's status is non-zero, mark that
767 			 * there is no more data to process.
768 			 */
769 			if (*++tl)
770 				nd->nd_flag |= ND_NOMOREDATA;
771 
772 			/*
773 			 * If the first op is Putfh, throw its results away
774 			 * and toss the op# and status for the first op.
775 			 */
776 			if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) {
777 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
778 				i = fxdr_unsigned(int, *tl++);
779 				j = fxdr_unsigned(int, *tl);
780 				/*
781 				 * All Compounds that do an Op that must
782 				 * be in sequence consist of NFSV4OP_PUTFH
783 				 * followed by one of these. As such, we
784 				 * can determine if the seqid# should be
785 				 * incremented, here.
786 				 */
787 				if ((i == NFSV4OP_OPEN ||
788 				     i == NFSV4OP_OPENCONFIRM ||
789 				     i == NFSV4OP_OPENDOWNGRADE ||
790 				     i == NFSV4OP_CLOSE ||
791 				     i == NFSV4OP_LOCK ||
792 				     i == NFSV4OP_LOCKU) &&
793 				    (j == 0 ||
794 				     (j != NFSERR_STALECLIENTID &&
795 				      j != NFSERR_STALESTATEID &&
796 				      j != NFSERR_BADSTATEID &&
797 				      j != NFSERR_BADSEQID &&
798 				      j != NFSERR_BADXDR &&
799 				      j != NFSERR_RESOURCE &&
800 				      j != NFSERR_NOFILEHANDLE)))
801 					nd->nd_flag |= ND_INCRSEQID;
802 				/*
803 				 * If the first op's status is non-zero, mark
804 				 * that there is no more data to process.
805 				 */
806 				if (j)
807 					nd->nd_flag |= ND_NOMOREDATA;
808 			}
809 
810 			/*
811 			 * If R_DONTRECOVER is set, replace the stale error
812 			 * reply, so that recovery isn't initiated.
813 			 */
814 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
815 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
816 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
817 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
818 		}
819 	}
820 
821 #ifdef KDTRACE_HOOKS
822 	if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
823 		uint32_t probe_id;
824 		int probe_procnum;
825 
826 		if (nd->nd_flag & ND_NFSV4) {
827 			probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
828 			probe_procnum = nd->nd_procnum;
829 		} else if (nd->nd_flag & ND_NFSV3) {
830 			probe_id = nfscl_nfs3_done_probes[procnum];
831 			probe_procnum = procnum;
832 		} else {
833 			probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
834 			probe_procnum = procnum;
835 		}
836 		if (probe_id != 0)
837 			(dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
838 			    nd->nd_mreq, cred, probe_procnum, 0);
839 	}
840 #endif
841 
842 	m_freem(nd->nd_mreq);
843 	AUTH_DESTROY(auth);
844 	if (rep != NULL)
845 		FREE((caddr_t)rep, M_NFSDREQ);
846 	if (set_sigset)
847 		newnfs_restore_sigmask(td, &oldset);
848 	return (0);
849 nfsmout:
850 	mbuf_freem(nd->nd_mrep);
851 	mbuf_freem(nd->nd_mreq);
852 	AUTH_DESTROY(auth);
853 	if (rep != NULL)
854 		FREE((caddr_t)rep, M_NFSDREQ);
855 	if (set_sigset)
856 		newnfs_restore_sigmask(td, &oldset);
857 	return (error);
858 }
859 
860 /*
861  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
862  * wait for all requests to complete. This is used by forced unmounts
863  * to terminate any outstanding RPCs.
864  */
865 int
866 newnfs_nmcancelreqs(struct nfsmount *nmp)
867 {
868 
869 	if (nmp->nm_sockreq.nr_client != NULL)
870 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
871 	return (0);
872 }
873 
874 /*
875  * Any signal that can interrupt an NFS operation in an intr mount
876  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
877  */
878 int newnfs_sig_set[] = {
879 	SIGINT,
880 	SIGTERM,
881 	SIGHUP,
882 	SIGKILL,
883 	SIGSTOP,
884 	SIGQUIT
885 };
886 
887 /*
888  * Check to see if one of the signals in our subset is pending on
889  * the process (in an intr mount).
890  */
891 static int
892 nfs_sig_pending(sigset_t set)
893 {
894 	int i;
895 
896 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
897 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
898 			return (1);
899 	return (0);
900 }
901 
902 /*
903  * The set/restore sigmask functions are used to (temporarily) overwrite
904  * the process p_sigmask during an RPC call (for example). These are also
905  * used in other places in the NFS client that might tsleep().
906  */
907 void
908 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
909 {
910 	sigset_t newset;
911 	int i;
912 	struct proc *p;
913 
914 	SIGFILLSET(newset);
915 	if (td == NULL)
916 		td = curthread; /* XXX */
917 	p = td->td_proc;
918 	/* Remove the NFS set of signals from newset */
919 	PROC_LOCK(p);
920 	mtx_lock(&p->p_sigacts->ps_mtx);
921 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
922 		/*
923 		 * But make sure we leave the ones already masked
924 		 * by the process, ie. remove the signal from the
925 		 * temporary signalmask only if it wasn't already
926 		 * in p_sigmask.
927 		 */
928 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
929 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
930 			SIGDELSET(newset, newnfs_sig_set[i]);
931 	}
932 	mtx_unlock(&p->p_sigacts->ps_mtx);
933 	PROC_UNLOCK(p);
934 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0);
935 }
936 
937 void
938 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
939 {
940 	if (td == NULL)
941 		td = curthread; /* XXX */
942 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
943 }
944 
945 /*
946  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
947  * old one after msleep() returns.
948  */
949 int
950 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
951 {
952 	sigset_t oldset;
953 	int error;
954 	struct proc *p;
955 
956 	if ((priority & PCATCH) == 0)
957 		return msleep(ident, mtx, priority, wmesg, timo);
958 	if (td == NULL)
959 		td = curthread; /* XXX */
960 	newnfs_set_sigmask(td, &oldset);
961 	error = msleep(ident, mtx, priority, wmesg, timo);
962 	newnfs_restore_sigmask(td, &oldset);
963 	p = td->td_proc;
964 	return (error);
965 }
966 
967 /*
968  * Test for a termination condition pending on the process.
969  * This is used for NFSMNT_INT mounts.
970  */
971 int
972 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
973 {
974 	struct proc *p;
975 	sigset_t tmpset;
976 
977 	/* Terminate all requests while attempting a forced unmount. */
978 	if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
979 		return (EIO);
980 	if (!(nmp->nm_flag & NFSMNT_INT))
981 		return (0);
982 	if (td == NULL)
983 		return (0);
984 	p = td->td_proc;
985 	PROC_LOCK(p);
986 	tmpset = p->p_siglist;
987 	SIGSETOR(tmpset, td->td_siglist);
988 	SIGSETNAND(tmpset, td->td_sigmask);
989 	mtx_lock(&p->p_sigacts->ps_mtx);
990 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
991 	mtx_unlock(&p->p_sigacts->ps_mtx);
992 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
993 	    && nfs_sig_pending(tmpset)) {
994 		PROC_UNLOCK(p);
995 		return (EINTR);
996 	}
997 	PROC_UNLOCK(p);
998 	return (0);
999 }
1000 
1001 static int
1002 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1003 {
1004 	struct proc *p;
1005 
1006 	p = td ? td->td_proc : NULL;
1007 	if (error) {
1008 		tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
1009 		    server, msg, error);
1010 	} else {
1011 		tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
1012 	}
1013 	return (0);
1014 }
1015 
1016 static void
1017 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1018     int error, int flags)
1019 {
1020 	if (nmp == NULL)
1021 		return;
1022 	mtx_lock(&nmp->nm_mtx);
1023 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1024 		nmp->nm_state |= NFSSTA_TIMEO;
1025 		mtx_unlock(&nmp->nm_mtx);
1026 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1027 		    VQ_NOTRESP, 0);
1028 	} else
1029 		mtx_unlock(&nmp->nm_mtx);
1030 	mtx_lock(&nmp->nm_mtx);
1031 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1032 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
1033 		mtx_unlock(&nmp->nm_mtx);
1034 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1035 		    VQ_NOTRESPLOCK, 0);
1036 	} else
1037 		mtx_unlock(&nmp->nm_mtx);
1038 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1039 }
1040 
1041 static void
1042 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1043     int flags, int tprintfmsg)
1044 {
1045 	if (nmp == NULL)
1046 		return;
1047 	if (tprintfmsg) {
1048 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1049 	}
1050 
1051 	mtx_lock(&nmp->nm_mtx);
1052 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1053 		nmp->nm_state &= ~NFSSTA_TIMEO;
1054 		mtx_unlock(&nmp->nm_mtx);
1055 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1056 		    VQ_NOTRESP, 1);
1057 	} else
1058 		mtx_unlock(&nmp->nm_mtx);
1059 
1060 	mtx_lock(&nmp->nm_mtx);
1061 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1062 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1063 		mtx_unlock(&nmp->nm_mtx);
1064 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1065 		    VQ_NOTRESPLOCK, 1);
1066 	} else
1067 		mtx_unlock(&nmp->nm_mtx);
1068 }
1069 
1070