xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision c243e4902be8df1e643c76b5f18b68bb77cc5268)
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_kdtrace.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 #ifdef KDTRACE_HOOKS
68 #include <sys/dtrace_bsd.h>
69 
70 dtrace_nfsclient_nfs23_start_probe_func_t
71 		dtrace_nfscl_nfs234_start_probe;
72 
73 dtrace_nfsclient_nfs23_done_probe_func_t
74 		dtrace_nfscl_nfs234_done_probe;
75 
76 /*
77  * Registered probes by RPC type.
78  */
79 uint32_t	nfscl_nfs2_start_probes[NFS_NPROCS + 1];
80 uint32_t	nfscl_nfs2_done_probes[NFS_NPROCS + 1];
81 
82 uint32_t	nfscl_nfs3_start_probes[NFS_NPROCS + 1];
83 uint32_t	nfscl_nfs3_done_probes[NFS_NPROCS + 1];
84 
85 uint32_t	nfscl_nfs4_start_probes[NFS_NPROCS + 1];
86 uint32_t	nfscl_nfs4_done_probes[NFS_NPROCS + 1];
87 #endif
88 
89 NFSSTATESPINLOCK;
90 NFSREQSPINLOCK;
91 extern struct nfsstats newnfsstats;
92 extern struct nfsreqhead nfsd_reqq;
93 extern int nfscl_ticks;
94 extern void (*ncl_call_invalcaches)(struct vnode *);
95 
96 static int	nfsrv_gsscallbackson = 0;
97 static int	nfs_bufpackets = 4;
98 static int	nfs_reconnects;
99 static int	nfs3_jukebox_delay = 10;
100 static int	nfs_skip_wcc_data_onerr = 1;
101 static int	nfs_keytab_enctype = ETYPE_DES_CBC_CRC;
102 
103 SYSCTL_DECL(_vfs_nfs);
104 
105 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
106     "Buffer reservation size 2 < x < 64");
107 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
108     "Number of times the nfs client has had to reconnect");
109 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
110     "Number of seconds to delay a retry after receiving EJUKEBOX");
111 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
112     "Disable weak cache consistency checking when server returns an error");
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, keytab_enctype, CTLFLAG_RW, &nfs_keytab_enctype, 0,
114     "Encryption type for the keytab entry used by nfs");
115 
116 static void	nfs_down(struct nfsmount *, struct thread *, const char *,
117     int, int);
118 static void	nfs_up(struct nfsmount *, struct thread *, const char *,
119     int, int);
120 static int	nfs_msg(struct thread *, const char *, const char *, int);
121 
122 struct nfs_cached_auth {
123 	int		ca_refs; /* refcount, including 1 from the cache */
124 	uid_t		ca_uid;	 /* uid that corresponds to this auth */
125 	AUTH		*ca_auth; /* RPC auth handle */
126 };
127 
128 static int nfsv2_procid[NFS_V3NPROCS] = {
129 	NFSV2PROC_NULL,
130 	NFSV2PROC_GETATTR,
131 	NFSV2PROC_SETATTR,
132 	NFSV2PROC_LOOKUP,
133 	NFSV2PROC_NOOP,
134 	NFSV2PROC_READLINK,
135 	NFSV2PROC_READ,
136 	NFSV2PROC_WRITE,
137 	NFSV2PROC_CREATE,
138 	NFSV2PROC_MKDIR,
139 	NFSV2PROC_SYMLINK,
140 	NFSV2PROC_CREATE,
141 	NFSV2PROC_REMOVE,
142 	NFSV2PROC_RMDIR,
143 	NFSV2PROC_RENAME,
144 	NFSV2PROC_LINK,
145 	NFSV2PROC_READDIR,
146 	NFSV2PROC_NOOP,
147 	NFSV2PROC_STATFS,
148 	NFSV2PROC_NOOP,
149 	NFSV2PROC_NOOP,
150 	NFSV2PROC_NOOP,
151 };
152 
153 /*
154  * Initialize sockets and congestion for a new NFS connection.
155  * We do not free the sockaddr if error.
156  */
157 int
158 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
159     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
160 {
161 	int rcvreserve, sndreserve;
162 	int pktscale;
163 	struct sockaddr *saddr;
164 	struct ucred *origcred;
165 	CLIENT *client;
166 	struct netconfig *nconf;
167 	struct socket *so;
168 	int one = 1, retries, error = 0;
169 	struct thread *td = curthread;
170 	struct timeval timo;
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 		goto out;
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 		goto out;
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 			if (nmp->nm_sotype == SOCK_DGRAM)
269 				/*
270 				 * For UDP, the large timeout for a reconnect
271 				 * will be set to "nm_retry * nm_timeo / 2", so
272 				 * we only want to do 2 reconnect timeout
273 				 * retries.
274 				 */
275 				retries = 2;
276 			else
277 				retries = nmp->nm_retry;
278 		} else
279 			retries = INT_MAX;
280 	} else {
281 		/*
282 		 * Three cases:
283 		 * - Null RPC callback to client
284 		 * - Non-Null RPC callback to client, wait a little longer
285 		 * - upcalls to nfsuserd and gssd (clp == NULL)
286 		 */
287 		if (callback_retry_mult == 0) {
288 			retries = NFSV4_UPCALLRETRY;
289 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
290 		} else {
291 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
292 		}
293 	}
294 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
295 
296 	if (nmp != NULL) {
297 		/*
298 		 * For UDP, there are 2 timeouts:
299 		 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
300 		 *   that does a retransmit of an RPC request using the same
301 		 *   socket and xid. This is what you normally want to do,
302 		 *   since NFS servers depend on "same xid" for their
303 		 *   Duplicate Request Cache.
304 		 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
305 		 *   retransmits on the same socket should fail and a fresh
306 		 *   socket created. Each of these timeouts counts as one
307 		 *   CLSET_RETRIES as set above.
308 		 * Set the initial retransmit timeout for UDP. This timeout
309 		 * doesn't exist for TCP and the following call just fails,
310 		 * which is ok.
311 		 */
312 		timo.tv_sec = nmp->nm_timeo / NFS_HZ;
313 		timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
314 		CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
315 	}
316 
317 	mtx_lock(&nrp->nr_mtx);
318 	if (nrp->nr_client != NULL) {
319 		/*
320 		 * Someone else already connected.
321 		 */
322 		CLNT_RELEASE(client);
323 	} else {
324 		nrp->nr_client = client;
325 	}
326 
327 	/*
328 	 * Protocols that do not require connections may be optionally left
329 	 * unconnected for servers that reply from a port other than NFS_PORT.
330 	 */
331 	if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
332 		mtx_unlock(&nrp->nr_mtx);
333 		CLNT_CONTROL(client, CLSET_CONNECT, &one);
334 	} else {
335 		mtx_unlock(&nrp->nr_mtx);
336 	}
337 
338 	/* Restore current thread's credentials. */
339 	td->td_ucred = origcred;
340 
341 out:
342 	NFSEXITCODE(error);
343 	return (error);
344 }
345 
346 /*
347  * NFS disconnect. Clean up and unlink.
348  */
349 void
350 newnfs_disconnect(struct nfssockreq *nrp)
351 {
352 	CLIENT *client;
353 
354 	mtx_lock(&nrp->nr_mtx);
355 	if (nrp->nr_client != NULL) {
356 		client = nrp->nr_client;
357 		nrp->nr_client = NULL;
358 		mtx_unlock(&nrp->nr_mtx);
359 		rpc_gss_secpurge_call(client);
360 		CLNT_CLOSE(client);
361 		CLNT_RELEASE(client);
362 	} else {
363 		mtx_unlock(&nrp->nr_mtx);
364 	}
365 }
366 
367 static AUTH *
368 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
369     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
370 {
371 	rpc_gss_service_t svc;
372 	AUTH *auth;
373 #ifdef notyet
374 	rpc_gss_options_req_t req_options;
375 #endif
376 
377 	switch (secflavour) {
378 	case RPCSEC_GSS_KRB5:
379 	case RPCSEC_GSS_KRB5I:
380 	case RPCSEC_GSS_KRB5P:
381 		if (!mech_oid) {
382 			if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
383 				return (NULL);
384 		}
385 		if (secflavour == RPCSEC_GSS_KRB5)
386 			svc = rpc_gss_svc_none;
387 		else if (secflavour == RPCSEC_GSS_KRB5I)
388 			svc = rpc_gss_svc_integrity;
389 		else
390 			svc = rpc_gss_svc_privacy;
391 #ifdef notyet
392 		req_options.req_flags = GSS_C_MUTUAL_FLAG;
393 		req_options.time_req = 0;
394 		req_options.my_cred = GSS_C_NO_CREDENTIAL;
395 		req_options.input_channel_bindings = NULL;
396 		req_options.enc_type = nfs_keytab_enctype;
397 
398 		auth = rpc_gss_secfind_call(nrp->nr_client, cred,
399 		    clnt_principal, srv_principal, mech_oid, svc,
400 		    &req_options);
401 #else
402 		/*
403 		 * Until changes to the rpcsec_gss code are committed,
404 		 * there is no support for host based initiator
405 		 * principals. As such, that case cannot yet be handled.
406 		 */
407 		if (clnt_principal == NULL)
408 			auth = rpc_gss_secfind_call(nrp->nr_client, cred,
409 			    srv_principal, mech_oid, svc);
410 		else
411 			auth = NULL;
412 #endif
413 		if (auth != NULL)
414 			return (auth);
415 		/* fallthrough */
416 	case AUTH_SYS:
417 	default:
418 		return (authunix_create(cred));
419 
420 	}
421 }
422 
423 /*
424  * Callback from the RPC code to generate up/down notifications.
425  */
426 
427 struct nfs_feedback_arg {
428 	struct nfsmount *nf_mount;
429 	int		nf_lastmsg;	/* last tprintf */
430 	int		nf_tprintfmsg;
431 	struct thread	*nf_td;
432 };
433 
434 static void
435 nfs_feedback(int type, int proc, void *arg)
436 {
437 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
438 	struct nfsmount *nmp = nf->nf_mount;
439 	struct timeval now;
440 
441 	getmicrouptime(&now);
442 
443 	switch (type) {
444 	case FEEDBACK_REXMIT2:
445 	case FEEDBACK_RECONNECT:
446 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
447 			nfs_down(nmp, nf->nf_td,
448 			    "not responding", 0, NFSSTA_TIMEO);
449 			nf->nf_tprintfmsg = TRUE;
450 			nf->nf_lastmsg = now.tv_sec;
451 		}
452 		break;
453 
454 	case FEEDBACK_OK:
455 		nfs_up(nf->nf_mount, nf->nf_td,
456 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
457 		break;
458 	}
459 }
460 
461 /*
462  * newnfs_request - goes something like this
463  *	- does the rpc by calling the krpc layer
464  *	- break down rpc header and return with nfs reply
465  * nb: always frees up nd_mreq mbuf list
466  */
467 int
468 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
469     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
470     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
471     u_char *retsum, int toplevel, u_int64_t *xidp)
472 {
473 	u_int32_t *tl;
474 	time_t waituntil;
475 	int i, j, set_sigset = 0, timeo;
476 	int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
477 	u_int16_t procnum;
478 	u_int trylater_delay = 1;
479 	struct nfs_feedback_arg nf;
480 	struct timeval timo, now;
481 	AUTH *auth;
482 	struct rpc_callextra ext;
483 	enum clnt_stat stat;
484 	struct nfsreq *rep = NULL;
485 	char *srv_principal = NULL;
486 	sigset_t oldset;
487 	struct ucred *authcred;
488 
489 	if (xidp != NULL)
490 		*xidp = 0;
491 	/* Reject requests while attempting a forced unmount. */
492 	if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
493 		m_freem(nd->nd_mreq);
494 		return (ESTALE);
495 	}
496 
497 	/*
498 	 * Set authcred, which is used to acquire RPC credentials to
499 	 * the cred argument, by default. The crhold() should not be
500 	 * necessary, but will ensure that some future code change
501 	 * doesn't result in the credential being free'd prematurely.
502 	 */
503 	authcred = crhold(cred);
504 
505 	/* For client side interruptible mounts, mask off the signals. */
506 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
507 		newnfs_set_sigmask(td, &oldset);
508 		set_sigset = 1;
509 	}
510 
511 	/*
512 	 * XXX if not already connected call nfs_connect now. Longer
513 	 * term, change nfs_mount to call nfs_connect unconditionally
514 	 * and let clnt_reconnect_create handle reconnects.
515 	 */
516 	if (nrp->nr_client == NULL)
517 		newnfs_connect(nmp, nrp, cred, td, 0);
518 
519 	/*
520 	 * For a client side mount, nmp is != NULL and clp == NULL. For
521 	 * server calls (callbacks or upcalls), nmp == NULL.
522 	 */
523 	if (clp != NULL) {
524 		NFSLOCKSTATE();
525 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
526 			secflavour = RPCSEC_GSS_KRB5;
527 			if (nd->nd_procnum != NFSPROC_NULL) {
528 				if (clp->lc_flags & LCL_GSSINTEGRITY)
529 					secflavour = RPCSEC_GSS_KRB5I;
530 				else if (clp->lc_flags & LCL_GSSPRIVACY)
531 					secflavour = RPCSEC_GSS_KRB5P;
532 			}
533 		}
534 		NFSUNLOCKSTATE();
535 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
536 	     nd->nd_procnum != NFSPROC_NULL) {
537 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
538 			nd->nd_flag |= ND_USEGSSNAME;
539 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
540 			/*
541 			 * If there is a client side host based credential,
542 			 * use that, otherwise use the system uid, if set.
543 			 * The system uid is in the nmp->nm_sockreq.nr_cred
544 			 * credentials.
545 			 */
546 			if (nmp->nm_krbnamelen > 0) {
547 				usegssname = 1;
548 			} else if (nmp->nm_uid != (uid_t)-1) {
549 				KASSERT(nmp->nm_sockreq.nr_cred != NULL,
550 				    ("newnfs_request: NULL nr_cred"));
551 				crfree(authcred);
552 				authcred = crhold(nmp->nm_sockreq.nr_cred);
553 			}
554 		} else if (nmp->nm_krbnamelen == 0 &&
555 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
556 			/*
557 			 * If there is no host based principal name and
558 			 * the system uid is set and this is root, use the
559 			 * system uid, since root won't have user
560 			 * credentials in a credentials cache file.
561 			 * The system uid is in the nmp->nm_sockreq.nr_cred
562 			 * credentials.
563 			 */
564 			KASSERT(nmp->nm_sockreq.nr_cred != NULL,
565 			    ("newnfs_request: NULL nr_cred"));
566 			crfree(authcred);
567 			authcred = crhold(nmp->nm_sockreq.nr_cred);
568 		}
569 		if (NFSHASINTEGRITY(nmp))
570 			secflavour = RPCSEC_GSS_KRB5I;
571 		else if (NFSHASPRIVACY(nmp))
572 			secflavour = RPCSEC_GSS_KRB5P;
573 		else
574 			secflavour = RPCSEC_GSS_KRB5;
575 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
576 	} else if (nmp != NULL && !NFSHASKERB(nmp) &&
577 	    nd->nd_procnum != NFSPROC_NULL &&
578 	    (nd->nd_flag & ND_USEGSSNAME) != 0) {
579 		/*
580 		 * Use the uid that did the mount when the RPC is doing
581 		 * NFSv4 system operations, as indicated by the
582 		 * ND_USEGSSNAME flag, for the AUTH_SYS case.
583 		 * The credentials in nm_sockreq.nr_cred were used for the
584 		 * mount.
585 		 */
586 		KASSERT(nmp->nm_sockreq.nr_cred != NULL,
587 		    ("newnfs_request: NULL nr_cred"));
588 		crfree(authcred);
589 		authcred = crhold(nmp->nm_sockreq.nr_cred);
590 	}
591 
592 	if (nmp != NULL) {
593 		bzero(&nf, sizeof(struct nfs_feedback_arg));
594 		nf.nf_mount = nmp;
595 		nf.nf_td = td;
596 		getmicrouptime(&now);
597 		nf.nf_lastmsg = now.tv_sec -
598 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
599 	}
600 
601 	if (nd->nd_procnum == NFSPROC_NULL)
602 		auth = authnone_create();
603 	else if (usegssname)
604 		auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
605 		    srv_principal, NULL, authcred);
606 	else
607 		auth = nfs_getauth(nrp, secflavour, NULL,
608 		    srv_principal, NULL, authcred);
609 	crfree(authcred);
610 	if (auth == NULL) {
611 		m_freem(nd->nd_mreq);
612 		if (set_sigset)
613 			newnfs_restore_sigmask(td, &oldset);
614 		return (EACCES);
615 	}
616 	bzero(&ext, sizeof(ext));
617 	ext.rc_auth = auth;
618 	if (nmp != NULL) {
619 		ext.rc_feedback = nfs_feedback;
620 		ext.rc_feedback_arg = &nf;
621 	}
622 
623 	procnum = nd->nd_procnum;
624 	if ((nd->nd_flag & ND_NFSV4) &&
625 	    nd->nd_procnum != NFSPROC_NULL &&
626 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
627 		procnum = NFSV4PROC_COMPOUND;
628 
629 	if (nmp != NULL) {
630 		NFSINCRGLOBAL(newnfsstats.rpcrequests);
631 
632 		/* Map the procnum to the old NFSv2 one, as required. */
633 		if ((nd->nd_flag & ND_NFSV2) != 0) {
634 			if (nd->nd_procnum < NFS_V3NPROCS)
635 				procnum = nfsv2_procid[nd->nd_procnum];
636 			else
637 				procnum = NFSV2PROC_NOOP;
638 		}
639 
640 		/*
641 		 * Now only used for the R_DONTRECOVER case, but until that is
642 		 * supported within the krpc code, I need to keep a queue of
643 		 * outstanding RPCs for nfsv4 client requests.
644 		 */
645 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
646 			MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
647 			    M_NFSDREQ, M_WAITOK);
648 #ifdef KDTRACE_HOOKS
649 		if (dtrace_nfscl_nfs234_start_probe != NULL) {
650 			uint32_t probe_id;
651 			int probe_procnum;
652 
653 			if (nd->nd_flag & ND_NFSV4) {
654 				probe_id =
655 				    nfscl_nfs4_start_probes[nd->nd_procnum];
656 				probe_procnum = nd->nd_procnum;
657 			} else if (nd->nd_flag & ND_NFSV3) {
658 				probe_id = nfscl_nfs3_start_probes[procnum];
659 				probe_procnum = procnum;
660 			} else {
661 				probe_id =
662 				    nfscl_nfs2_start_probes[nd->nd_procnum];
663 				probe_procnum = procnum;
664 			}
665 			if (probe_id != 0)
666 				(dtrace_nfscl_nfs234_start_probe)
667 				    (probe_id, vp, nd->nd_mreq, cred,
668 				     probe_procnum);
669 		}
670 #endif
671 	}
672 	trycnt = 0;
673 tryagain:
674 	/*
675 	 * This timeout specifies when a new socket should be created,
676 	 * along with new xid values. For UDP, this should be done
677 	 * infrequently, since retransmits of RPC requests should normally
678 	 * use the same xid.
679 	 */
680 	if (nmp == NULL) {
681 		timo.tv_usec = 0;
682 		if (clp == NULL)
683 			timo.tv_sec = NFSV4_UPCALLTIMEO;
684 		else
685 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
686 	} else {
687 		if (nrp->nr_sotype != SOCK_DGRAM) {
688 			timo.tv_usec = 0;
689 			if ((nmp->nm_flag & NFSMNT_NFSV4))
690 				timo.tv_sec = INT_MAX;
691 			else
692 				timo.tv_sec = NFS_TCPTIMEO;
693 		} else {
694 			if (NFSHASSOFT(nmp)) {
695 				/*
696 				 * CLSET_RETRIES is set to 2, so this should be
697 				 * half of the total timeout required.
698 				 */
699 				timeo = nmp->nm_retry * nmp->nm_timeo / 2;
700 				if (timeo < 1)
701 					timeo = 1;
702 				timo.tv_sec = timeo / NFS_HZ;
703 				timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
704 				    NFS_HZ;
705 			} else {
706 				/* For UDP hard mounts, use a large value. */
707 				timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
708 				timo.tv_usec = 0;
709 			}
710 		}
711 
712 		if (rep != NULL) {
713 			rep->r_flags = 0;
714 			rep->r_nmp = nmp;
715 			/*
716 			 * Chain request into list of outstanding requests.
717 			 */
718 			NFSLOCKREQ();
719 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
720 			NFSUNLOCKREQ();
721 		}
722 	}
723 
724 	nd->nd_mrep = NULL;
725 	stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
726 	    &nd->nd_mrep, timo);
727 
728 	if (rep != NULL) {
729 		/*
730 		 * RPC done, unlink the request.
731 		 */
732 		NFSLOCKREQ();
733 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
734 		NFSUNLOCKREQ();
735 	}
736 
737 	/*
738 	 * If there was a successful reply and a tprintf msg.
739 	 * tprintf a response.
740 	 */
741 	if (stat == RPC_SUCCESS) {
742 		error = 0;
743 	} else if (stat == RPC_TIMEDOUT) {
744 		error = ETIMEDOUT;
745 	} else if (stat == RPC_VERSMISMATCH) {
746 		error = EOPNOTSUPP;
747 	} else if (stat == RPC_PROGVERSMISMATCH) {
748 		error = EPROTONOSUPPORT;
749 	} else {
750 		error = EACCES;
751 	}
752 	if (error) {
753 		m_freem(nd->nd_mreq);
754 		AUTH_DESTROY(auth);
755 		if (rep != NULL)
756 			FREE((caddr_t)rep, M_NFSDREQ);
757 		if (set_sigset)
758 			newnfs_restore_sigmask(td, &oldset);
759 		return (error);
760 	}
761 
762 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
763 
764 	/*
765 	 * Search for any mbufs that are not a multiple of 4 bytes long
766 	 * or with m_data not longword aligned.
767 	 * These could cause pointer alignment problems, so copy them to
768 	 * well aligned mbufs.
769 	 */
770 	newnfs_realign(&nd->nd_mrep);
771 	nd->nd_md = nd->nd_mrep;
772 	nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
773 	nd->nd_repstat = 0;
774 	if (nd->nd_procnum != NFSPROC_NULL) {
775 		/*
776 		 * and now the actual NFS xdr.
777 		 */
778 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
779 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
780 		if (nd->nd_repstat != 0) {
781 			if (((nd->nd_repstat == NFSERR_DELAY ||
782 			      nd->nd_repstat == NFSERR_GRACE) &&
783 			     (nd->nd_flag & ND_NFSV4) &&
784 			     nd->nd_procnum != NFSPROC_DELEGRETURN &&
785 			     nd->nd_procnum != NFSPROC_SETATTR &&
786 			     nd->nd_procnum != NFSPROC_READ &&
787 			     nd->nd_procnum != NFSPROC_WRITE &&
788 			     nd->nd_procnum != NFSPROC_OPEN &&
789 			     nd->nd_procnum != NFSPROC_CREATE &&
790 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
791 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
792 			     nd->nd_procnum != NFSPROC_CLOSE &&
793 			     nd->nd_procnum != NFSPROC_LOCK &&
794 			     nd->nd_procnum != NFSPROC_LOCKU) ||
795 			    (nd->nd_repstat == NFSERR_DELAY &&
796 			     (nd->nd_flag & ND_NFSV4) == 0) ||
797 			    nd->nd_repstat == NFSERR_RESOURCE) {
798 				if (trylater_delay > NFS_TRYLATERDEL)
799 					trylater_delay = NFS_TRYLATERDEL;
800 				waituntil = NFSD_MONOSEC + trylater_delay;
801 				while (NFSD_MONOSEC < waituntil)
802 					(void) nfs_catnap(PZERO, 0, "nfstry");
803 				trylater_delay *= 2;
804 				m_freem(nd->nd_mrep);
805 				nd->nd_mrep = NULL;
806 				goto tryagain;
807 			}
808 
809 			/*
810 			 * If the File Handle was stale, invalidate the
811 			 * lookup cache, just in case.
812 			 * (vp != NULL implies a client side call)
813 			 */
814 			if (nd->nd_repstat == ESTALE && vp != NULL) {
815 				cache_purge(vp);
816 				if (ncl_call_invalcaches != NULL)
817 					(*ncl_call_invalcaches)(vp);
818 			}
819 		}
820 
821 		/*
822 		 * Get rid of the tag, return count, and PUTFH result for V4.
823 		 */
824 		if (nd->nd_flag & ND_NFSV4) {
825 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
826 			i = fxdr_unsigned(int, *tl);
827 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
828 			if (error)
829 				goto nfsmout;
830 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
831 			i = fxdr_unsigned(int, *++tl);
832 
833 			/*
834 			 * If the first op's status is non-zero, mark that
835 			 * there is no more data to process.
836 			 */
837 			if (*++tl)
838 				nd->nd_flag |= ND_NOMOREDATA;
839 
840 			/*
841 			 * If the first op is Putfh, throw its results away
842 			 * and toss the op# and status for the first op.
843 			 */
844 			if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) {
845 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
846 				i = fxdr_unsigned(int, *tl++);
847 				j = fxdr_unsigned(int, *tl);
848 				/*
849 				 * All Compounds that do an Op that must
850 				 * be in sequence consist of NFSV4OP_PUTFH
851 				 * followed by one of these. As such, we
852 				 * can determine if the seqid# should be
853 				 * incremented, here.
854 				 */
855 				if ((i == NFSV4OP_OPEN ||
856 				     i == NFSV4OP_OPENCONFIRM ||
857 				     i == NFSV4OP_OPENDOWNGRADE ||
858 				     i == NFSV4OP_CLOSE ||
859 				     i == NFSV4OP_LOCK ||
860 				     i == NFSV4OP_LOCKU) &&
861 				    (j == 0 ||
862 				     (j != NFSERR_STALECLIENTID &&
863 				      j != NFSERR_STALESTATEID &&
864 				      j != NFSERR_BADSTATEID &&
865 				      j != NFSERR_BADSEQID &&
866 				      j != NFSERR_BADXDR &&
867 				      j != NFSERR_RESOURCE &&
868 				      j != NFSERR_NOFILEHANDLE)))
869 					nd->nd_flag |= ND_INCRSEQID;
870 				/*
871 				 * If the first op's status is non-zero, mark
872 				 * that there is no more data to process.
873 				 */
874 				if (j)
875 					nd->nd_flag |= ND_NOMOREDATA;
876 			}
877 
878 			/*
879 			 * If R_DONTRECOVER is set, replace the stale error
880 			 * reply, so that recovery isn't initiated.
881 			 */
882 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
883 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
884 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
885 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
886 		}
887 	}
888 
889 #ifdef KDTRACE_HOOKS
890 	if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
891 		uint32_t probe_id;
892 		int probe_procnum;
893 
894 		if (nd->nd_flag & ND_NFSV4) {
895 			probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
896 			probe_procnum = nd->nd_procnum;
897 		} else if (nd->nd_flag & ND_NFSV3) {
898 			probe_id = nfscl_nfs3_done_probes[procnum];
899 			probe_procnum = procnum;
900 		} else {
901 			probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
902 			probe_procnum = procnum;
903 		}
904 		if (probe_id != 0)
905 			(dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
906 			    nd->nd_mreq, cred, probe_procnum, 0);
907 	}
908 #endif
909 
910 	m_freem(nd->nd_mreq);
911 	AUTH_DESTROY(auth);
912 	if (rep != NULL)
913 		FREE((caddr_t)rep, M_NFSDREQ);
914 	if (set_sigset)
915 		newnfs_restore_sigmask(td, &oldset);
916 	return (0);
917 nfsmout:
918 	mbuf_freem(nd->nd_mrep);
919 	mbuf_freem(nd->nd_mreq);
920 	AUTH_DESTROY(auth);
921 	if (rep != NULL)
922 		FREE((caddr_t)rep, M_NFSDREQ);
923 	if (set_sigset)
924 		newnfs_restore_sigmask(td, &oldset);
925 	return (error);
926 }
927 
928 /*
929  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
930  * wait for all requests to complete. This is used by forced unmounts
931  * to terminate any outstanding RPCs.
932  */
933 int
934 newnfs_nmcancelreqs(struct nfsmount *nmp)
935 {
936 
937 	if (nmp->nm_sockreq.nr_client != NULL)
938 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
939 	return (0);
940 }
941 
942 /*
943  * Any signal that can interrupt an NFS operation in an intr mount
944  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
945  */
946 int newnfs_sig_set[] = {
947 	SIGINT,
948 	SIGTERM,
949 	SIGHUP,
950 	SIGKILL,
951 	SIGSTOP,
952 	SIGQUIT
953 };
954 
955 /*
956  * Check to see if one of the signals in our subset is pending on
957  * the process (in an intr mount).
958  */
959 static int
960 nfs_sig_pending(sigset_t set)
961 {
962 	int i;
963 
964 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
965 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
966 			return (1);
967 	return (0);
968 }
969 
970 /*
971  * The set/restore sigmask functions are used to (temporarily) overwrite
972  * the process p_sigmask during an RPC call (for example). These are also
973  * used in other places in the NFS client that might tsleep().
974  */
975 void
976 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
977 {
978 	sigset_t newset;
979 	int i;
980 	struct proc *p;
981 
982 	SIGFILLSET(newset);
983 	if (td == NULL)
984 		td = curthread; /* XXX */
985 	p = td->td_proc;
986 	/* Remove the NFS set of signals from newset */
987 	PROC_LOCK(p);
988 	mtx_lock(&p->p_sigacts->ps_mtx);
989 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
990 		/*
991 		 * But make sure we leave the ones already masked
992 		 * by the process, ie. remove the signal from the
993 		 * temporary signalmask only if it wasn't already
994 		 * in p_sigmask.
995 		 */
996 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
997 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
998 			SIGDELSET(newset, newnfs_sig_set[i]);
999 	}
1000 	mtx_unlock(&p->p_sigacts->ps_mtx);
1001 	PROC_UNLOCK(p);
1002 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0);
1003 }
1004 
1005 void
1006 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1007 {
1008 	if (td == NULL)
1009 		td = curthread; /* XXX */
1010 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1011 }
1012 
1013 /*
1014  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1015  * old one after msleep() returns.
1016  */
1017 int
1018 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1019 {
1020 	sigset_t oldset;
1021 	int error;
1022 	struct proc *p;
1023 
1024 	if ((priority & PCATCH) == 0)
1025 		return msleep(ident, mtx, priority, wmesg, timo);
1026 	if (td == NULL)
1027 		td = curthread; /* XXX */
1028 	newnfs_set_sigmask(td, &oldset);
1029 	error = msleep(ident, mtx, priority, wmesg, timo);
1030 	newnfs_restore_sigmask(td, &oldset);
1031 	p = td->td_proc;
1032 	return (error);
1033 }
1034 
1035 /*
1036  * Test for a termination condition pending on the process.
1037  * This is used for NFSMNT_INT mounts.
1038  */
1039 int
1040 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1041 {
1042 	struct proc *p;
1043 	sigset_t tmpset;
1044 
1045 	/* Terminate all requests while attempting a forced unmount. */
1046 	if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1047 		return (EIO);
1048 	if (!(nmp->nm_flag & NFSMNT_INT))
1049 		return (0);
1050 	if (td == NULL)
1051 		return (0);
1052 	p = td->td_proc;
1053 	PROC_LOCK(p);
1054 	tmpset = p->p_siglist;
1055 	SIGSETOR(tmpset, td->td_siglist);
1056 	SIGSETNAND(tmpset, td->td_sigmask);
1057 	mtx_lock(&p->p_sigacts->ps_mtx);
1058 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1059 	mtx_unlock(&p->p_sigacts->ps_mtx);
1060 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1061 	    && nfs_sig_pending(tmpset)) {
1062 		PROC_UNLOCK(p);
1063 		return (EINTR);
1064 	}
1065 	PROC_UNLOCK(p);
1066 	return (0);
1067 }
1068 
1069 static int
1070 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1071 {
1072 	struct proc *p;
1073 
1074 	p = td ? td->td_proc : NULL;
1075 	if (error) {
1076 		tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
1077 		    server, msg, error);
1078 	} else {
1079 		tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
1080 	}
1081 	return (0);
1082 }
1083 
1084 static void
1085 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1086     int error, int flags)
1087 {
1088 	if (nmp == NULL)
1089 		return;
1090 	mtx_lock(&nmp->nm_mtx);
1091 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1092 		nmp->nm_state |= NFSSTA_TIMEO;
1093 		mtx_unlock(&nmp->nm_mtx);
1094 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1095 		    VQ_NOTRESP, 0);
1096 	} else
1097 		mtx_unlock(&nmp->nm_mtx);
1098 	mtx_lock(&nmp->nm_mtx);
1099 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1100 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
1101 		mtx_unlock(&nmp->nm_mtx);
1102 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1103 		    VQ_NOTRESPLOCK, 0);
1104 	} else
1105 		mtx_unlock(&nmp->nm_mtx);
1106 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1107 }
1108 
1109 static void
1110 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1111     int flags, int tprintfmsg)
1112 {
1113 	if (nmp == NULL)
1114 		return;
1115 	if (tprintfmsg) {
1116 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1117 	}
1118 
1119 	mtx_lock(&nmp->nm_mtx);
1120 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1121 		nmp->nm_state &= ~NFSSTA_TIMEO;
1122 		mtx_unlock(&nmp->nm_mtx);
1123 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1124 		    VQ_NOTRESP, 1);
1125 	} else
1126 		mtx_unlock(&nmp->nm_mtx);
1127 
1128 	mtx_lock(&nmp->nm_mtx);
1129 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1130 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1131 		mtx_unlock(&nmp->nm_mtx);
1132 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1133 		    VQ_NOTRESPLOCK, 1);
1134 	} else
1135 		mtx_unlock(&nmp->nm_mtx);
1136 }
1137 
1138