xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision 3e65b9c6e6b7b2081d54e1dc40983c3c00eaf738)
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 = 0;
170 	struct thread *td = curthread;
171 	struct timeval timo;
172 
173 	/*
174 	 * We need to establish the socket using the credentials of
175 	 * the mountpoint.  Some parts of this process (such as
176 	 * sobind() and soconnect()) will use the curent thread's
177 	 * credential instead of the socket credential.  To work
178 	 * around this, temporarily change the current thread's
179 	 * credential to that of the mountpoint.
180 	 *
181 	 * XXX: It would be better to explicitly pass the correct
182 	 * credential to sobind() and soconnect().
183 	 */
184 	origcred = td->td_ucred;
185 
186 	/*
187 	 * Use the credential in nr_cred, if not NULL.
188 	 */
189 	if (nrp->nr_cred != NULL)
190 		td->td_ucred = nrp->nr_cred;
191 	else
192 		td->td_ucred = cred;
193 	saddr = nrp->nr_nam;
194 
195 	if (saddr->sa_family == AF_INET)
196 		if (nrp->nr_sotype == SOCK_DGRAM)
197 			nconf = getnetconfigent("udp");
198 		else
199 			nconf = getnetconfigent("tcp");
200 	else
201 		if (nrp->nr_sotype == SOCK_DGRAM)
202 			nconf = getnetconfigent("udp6");
203 		else
204 			nconf = getnetconfigent("tcp6");
205 
206 	pktscale = nfs_bufpackets;
207 	if (pktscale < 2)
208 		pktscale = 2;
209 	if (pktscale > 64)
210 		pktscale = 64;
211 	/*
212 	 * soreserve() can fail if sb_max is too small, so shrink pktscale
213 	 * and try again if there is an error.
214 	 * Print a log message suggesting increasing sb_max.
215 	 * Creating a socket and doing this is necessary since, if the
216 	 * reservation sizes are too large and will make soreserve() fail,
217 	 * the connection will work until a large send is attempted and
218 	 * then it will loop in the krpc code.
219 	 */
220 	so = NULL;
221 	saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
222 	error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
223 	    nrp->nr_soproto, td->td_ucred, td);
224 	if (error) {
225 		td->td_ucred = origcred;
226 		goto out;
227 	}
228 	do {
229 	    if (error != 0 && pktscale > 2)
230 		pktscale--;
231 	    if (nrp->nr_sotype == SOCK_DGRAM) {
232 		if (nmp != NULL) {
233 			sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
234 			    pktscale;
235 			rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
236 			    pktscale;
237 		} else {
238 			sndreserve = rcvreserve = 1024 * pktscale;
239 		}
240 	    } else {
241 		if (nrp->nr_sotype != SOCK_STREAM)
242 			panic("nfscon sotype");
243 		if (nmp != NULL) {
244 			sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
245 			    sizeof (u_int32_t)) * pktscale;
246 			rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
247 			    sizeof (u_int32_t)) * pktscale;
248 		} else {
249 			sndreserve = rcvreserve = 1024 * pktscale;
250 		}
251 	    }
252 	    error = soreserve(so, sndreserve, rcvreserve);
253 	} while (error != 0 && pktscale > 2);
254 	soclose(so);
255 	if (error) {
256 		td->td_ucred = origcred;
257 		goto out;
258 	}
259 
260 	client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
261 	    nrp->nr_vers, sndreserve, rcvreserve);
262 	CLNT_CONTROL(client, CLSET_WAITCHAN, "newnfsreq");
263 	if (nmp != NULL) {
264 		if ((nmp->nm_flag & NFSMNT_INT))
265 			CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
266 		if ((nmp->nm_flag & NFSMNT_RESVPORT))
267 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
268 		if (NFSHASSOFT(nmp)) {
269 			if (nmp->nm_sotype == SOCK_DGRAM)
270 				/*
271 				 * For UDP, the large timeout for a reconnect
272 				 * will be set to "nm_retry * nm_timeo / 2", so
273 				 * we only want to do 2 reconnect timeout
274 				 * retries.
275 				 */
276 				retries = 2;
277 			else
278 				retries = nmp->nm_retry;
279 		} else
280 			retries = INT_MAX;
281 	} else {
282 		/*
283 		 * Three cases:
284 		 * - Null RPC callback to client
285 		 * - Non-Null RPC callback to client, wait a little longer
286 		 * - upcalls to nfsuserd and gssd (clp == NULL)
287 		 */
288 		if (callback_retry_mult == 0) {
289 			retries = NFSV4_UPCALLRETRY;
290 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
291 		} else {
292 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
293 		}
294 	}
295 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
296 
297 	if (nmp != NULL) {
298 		/*
299 		 * For UDP, there are 2 timeouts:
300 		 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
301 		 *   that does a retransmit of an RPC request using the same
302 		 *   socket and xid. This is what you normally want to do,
303 		 *   since NFS servers depend on "same xid" for their
304 		 *   Duplicate Request Cache.
305 		 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
306 		 *   retransmits on the same socket should fail and a fresh
307 		 *   socket created. Each of these timeouts counts as one
308 		 *   CLSET_RETRIES as set above.
309 		 * Set the initial retransmit timeout for UDP. This timeout
310 		 * doesn't exist for TCP and the following call just fails,
311 		 * which is ok.
312 		 */
313 		timo.tv_sec = nmp->nm_timeo / NFS_HZ;
314 		timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
315 		CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
316 	}
317 
318 	mtx_lock(&nrp->nr_mtx);
319 	if (nrp->nr_client != NULL) {
320 		/*
321 		 * Someone else already connected.
322 		 */
323 		CLNT_RELEASE(client);
324 	} else {
325 		nrp->nr_client = client;
326 	}
327 
328 	/*
329 	 * Protocols that do not require connections may be optionally left
330 	 * unconnected for servers that reply from a port other than NFS_PORT.
331 	 */
332 	if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
333 		mtx_unlock(&nrp->nr_mtx);
334 		CLNT_CONTROL(client, CLSET_CONNECT, &one);
335 	} else {
336 		mtx_unlock(&nrp->nr_mtx);
337 	}
338 
339 	/* Restore current thread's credentials. */
340 	td->td_ucred = origcred;
341 
342 out:
343 	NFSEXITCODE(error);
344 	return (error);
345 }
346 
347 /*
348  * NFS disconnect. Clean up and unlink.
349  */
350 void
351 newnfs_disconnect(struct nfssockreq *nrp)
352 {
353 	CLIENT *client;
354 
355 	mtx_lock(&nrp->nr_mtx);
356 	if (nrp->nr_client != NULL) {
357 		client = nrp->nr_client;
358 		nrp->nr_client = NULL;
359 		mtx_unlock(&nrp->nr_mtx);
360 		rpc_gss_secpurge_call(client);
361 		CLNT_CLOSE(client);
362 		CLNT_RELEASE(client);
363 	} else {
364 		mtx_unlock(&nrp->nr_mtx);
365 	}
366 }
367 
368 static AUTH *
369 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
370     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
371 {
372 	rpc_gss_service_t svc;
373 	AUTH *auth;
374 #ifdef notyet
375 	rpc_gss_options_req_t req_options;
376 #endif
377 
378 	switch (secflavour) {
379 	case RPCSEC_GSS_KRB5:
380 	case RPCSEC_GSS_KRB5I:
381 	case RPCSEC_GSS_KRB5P:
382 		if (!mech_oid) {
383 			if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
384 				return (NULL);
385 		}
386 		if (secflavour == RPCSEC_GSS_KRB5)
387 			svc = rpc_gss_svc_none;
388 		else if (secflavour == RPCSEC_GSS_KRB5I)
389 			svc = rpc_gss_svc_integrity;
390 		else
391 			svc = rpc_gss_svc_privacy;
392 #ifdef notyet
393 		req_options.req_flags = GSS_C_MUTUAL_FLAG;
394 		req_options.time_req = 0;
395 		req_options.my_cred = GSS_C_NO_CREDENTIAL;
396 		req_options.input_channel_bindings = NULL;
397 		req_options.enc_type = nfs_keytab_enctype;
398 
399 		auth = rpc_gss_secfind_call(nrp->nr_client, cred,
400 		    clnt_principal, srv_principal, mech_oid, svc,
401 		    &req_options);
402 #else
403 		/*
404 		 * Until changes to the rpcsec_gss code are committed,
405 		 * there is no support for host based initiator
406 		 * principals. As such, that case cannot yet be handled.
407 		 */
408 		if (clnt_principal == NULL)
409 			auth = rpc_gss_secfind_call(nrp->nr_client, cred,
410 			    srv_principal, mech_oid, svc);
411 		else
412 			auth = NULL;
413 #endif
414 		if (auth != NULL)
415 			return (auth);
416 		/* fallthrough */
417 	case AUTH_SYS:
418 	default:
419 		return (authunix_create(cred));
420 
421 	}
422 }
423 
424 /*
425  * Callback from the RPC code to generate up/down notifications.
426  */
427 
428 struct nfs_feedback_arg {
429 	struct nfsmount *nf_mount;
430 	int		nf_lastmsg;	/* last tprintf */
431 	int		nf_tprintfmsg;
432 	struct thread	*nf_td;
433 };
434 
435 static void
436 nfs_feedback(int type, int proc, void *arg)
437 {
438 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
439 	struct nfsmount *nmp = nf->nf_mount;
440 	struct timeval now;
441 
442 	getmicrouptime(&now);
443 
444 	switch (type) {
445 	case FEEDBACK_REXMIT2:
446 	case FEEDBACK_RECONNECT:
447 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
448 			nfs_down(nmp, nf->nf_td,
449 			    "not responding", 0, NFSSTA_TIMEO);
450 			nf->nf_tprintfmsg = TRUE;
451 			nf->nf_lastmsg = now.tv_sec;
452 		}
453 		break;
454 
455 	case FEEDBACK_OK:
456 		nfs_up(nf->nf_mount, nf->nf_td,
457 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
458 		break;
459 	}
460 }
461 
462 /*
463  * newnfs_request - goes something like this
464  *	- does the rpc by calling the krpc layer
465  *	- break down rpc header and return with nfs reply
466  * nb: always frees up nd_mreq mbuf list
467  */
468 int
469 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
470     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
471     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
472     u_char *retsum, int toplevel, u_int64_t *xidp)
473 {
474 	u_int32_t *tl;
475 	time_t waituntil;
476 	int i, j, set_uid = 0, set_sigset = 0, timeo;
477 	int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
478 	u_int16_t procnum;
479 	u_int trylater_delay = 1;
480 	struct nfs_feedback_arg nf;
481 	struct timeval timo, now;
482 	AUTH *auth;
483 	struct rpc_callextra ext;
484 	enum clnt_stat stat;
485 	struct nfsreq *rep = NULL;
486 	char *srv_principal = NULL;
487 	uid_t saved_uid = (uid_t)-1;
488 	sigset_t oldset;
489 
490 	if (xidp != NULL)
491 		*xidp = 0;
492 	/* Reject requests while attempting a forced unmount. */
493 	if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
494 		m_freem(nd->nd_mreq);
495 		return (ESTALE);
496 	}
497 
498 	/* For client side interruptible mounts, mask off the signals. */
499 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
500 		newnfs_set_sigmask(td, &oldset);
501 		set_sigset = 1;
502 	}
503 
504 	/*
505 	 * XXX if not already connected call nfs_connect now. Longer
506 	 * term, change nfs_mount to call nfs_connect unconditionally
507 	 * and let clnt_reconnect_create handle reconnects.
508 	 */
509 	if (nrp->nr_client == NULL)
510 		newnfs_connect(nmp, nrp, cred, td, 0);
511 
512 	/*
513 	 * For a client side mount, nmp is != NULL and clp == NULL. For
514 	 * server calls (callbacks or upcalls), nmp == NULL.
515 	 */
516 	if (clp != NULL) {
517 		NFSLOCKSTATE();
518 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
519 			secflavour = RPCSEC_GSS_KRB5;
520 			if (nd->nd_procnum != NFSPROC_NULL) {
521 				if (clp->lc_flags & LCL_GSSINTEGRITY)
522 					secflavour = RPCSEC_GSS_KRB5I;
523 				else if (clp->lc_flags & LCL_GSSPRIVACY)
524 					secflavour = RPCSEC_GSS_KRB5P;
525 			}
526 		}
527 		NFSUNLOCKSTATE();
528 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
529 	     nd->nd_procnum != NFSPROC_NULL) {
530 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
531 			nd->nd_flag |= ND_USEGSSNAME;
532 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
533 			/*
534 			 * If there is a client side host based credential,
535 			 * use that, otherwise use the system uid, if set.
536 			 */
537 			if (nmp->nm_krbnamelen > 0) {
538 				usegssname = 1;
539 			} else if (nmp->nm_uid != (uid_t)-1) {
540 				saved_uid = cred->cr_uid;
541 				cred->cr_uid = nmp->nm_uid;
542 				set_uid = 1;
543 			}
544 		} else if (nmp->nm_krbnamelen == 0 &&
545 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
546 			/*
547 			 * If there is no host based principal name and
548 			 * the system uid is set and this is root, use the
549 			 * system uid, since root won't have user
550 			 * credentials in a credentials cache file.
551 			 */
552 			saved_uid = cred->cr_uid;
553 			cred->cr_uid = nmp->nm_uid;
554 			set_uid = 1;
555 		}
556 		if (NFSHASINTEGRITY(nmp))
557 			secflavour = RPCSEC_GSS_KRB5I;
558 		else if (NFSHASPRIVACY(nmp))
559 			secflavour = RPCSEC_GSS_KRB5P;
560 		else
561 			secflavour = RPCSEC_GSS_KRB5;
562 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
563 	} else if (nmp != NULL && !NFSHASKERB(nmp) &&
564 	    nd->nd_procnum != NFSPROC_NULL &&
565 	    (nd->nd_flag & ND_USEGSSNAME) != 0) {
566 		/*
567 		 * Use the uid that did the mount when the RPC is doing
568 		 * NFSv4 system operations, as indicated by the
569 		 * ND_USEGSSNAME flag, for the AUTH_SYS case.
570 		 */
571 		saved_uid = cred->cr_uid;
572 		if (nmp->nm_uid != (uid_t)-1)
573 			cred->cr_uid = nmp->nm_uid;
574 		else
575 			cred->cr_uid = 0;
576 		set_uid = 1;
577 	}
578 
579 	if (nmp != NULL) {
580 		bzero(&nf, sizeof(struct nfs_feedback_arg));
581 		nf.nf_mount = nmp;
582 		nf.nf_td = td;
583 		getmicrouptime(&now);
584 		nf.nf_lastmsg = now.tv_sec -
585 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
586 	}
587 
588 	if (nd->nd_procnum == NFSPROC_NULL)
589 		auth = authnone_create();
590 	else if (usegssname)
591 		auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
592 		    srv_principal, NULL, cred);
593 	else
594 		auth = nfs_getauth(nrp, secflavour, NULL,
595 		    srv_principal, NULL, cred);
596 	if (set_uid)
597 		cred->cr_uid = saved_uid;
598 	if (auth == NULL) {
599 		m_freem(nd->nd_mreq);
600 		if (set_sigset)
601 			newnfs_restore_sigmask(td, &oldset);
602 		return (EACCES);
603 	}
604 	bzero(&ext, sizeof(ext));
605 	ext.rc_auth = auth;
606 	if (nmp != NULL) {
607 		ext.rc_feedback = nfs_feedback;
608 		ext.rc_feedback_arg = &nf;
609 	}
610 
611 	procnum = nd->nd_procnum;
612 	if ((nd->nd_flag & ND_NFSV4) &&
613 	    nd->nd_procnum != NFSPROC_NULL &&
614 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
615 		procnum = NFSV4PROC_COMPOUND;
616 
617 	if (nmp != NULL) {
618 		NFSINCRGLOBAL(newnfsstats.rpcrequests);
619 
620 		/* Map the procnum to the old NFSv2 one, as required. */
621 		if ((nd->nd_flag & ND_NFSV2) != 0) {
622 			if (nd->nd_procnum < NFS_V3NPROCS)
623 				procnum = nfsv2_procid[nd->nd_procnum];
624 			else
625 				procnum = NFSV2PROC_NOOP;
626 		}
627 
628 		/*
629 		 * Now only used for the R_DONTRECOVER case, but until that is
630 		 * supported within the krpc code, I need to keep a queue of
631 		 * outstanding RPCs for nfsv4 client requests.
632 		 */
633 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
634 			MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
635 			    M_NFSDREQ, M_WAITOK);
636 #ifdef KDTRACE_HOOKS
637 		if (dtrace_nfscl_nfs234_start_probe != NULL) {
638 			uint32_t probe_id;
639 			int probe_procnum;
640 
641 			if (nd->nd_flag & ND_NFSV4) {
642 				probe_id =
643 				    nfscl_nfs4_start_probes[nd->nd_procnum];
644 				probe_procnum = nd->nd_procnum;
645 			} else if (nd->nd_flag & ND_NFSV3) {
646 				probe_id = nfscl_nfs3_start_probes[procnum];
647 				probe_procnum = procnum;
648 			} else {
649 				probe_id =
650 				    nfscl_nfs2_start_probes[nd->nd_procnum];
651 				probe_procnum = procnum;
652 			}
653 			if (probe_id != 0)
654 				(dtrace_nfscl_nfs234_start_probe)
655 				    (probe_id, vp, nd->nd_mreq, cred,
656 				     probe_procnum);
657 		}
658 #endif
659 	}
660 	trycnt = 0;
661 tryagain:
662 	/*
663 	 * This timeout specifies when a new socket should be created,
664 	 * along with new xid values. For UDP, this should be done
665 	 * infrequently, since retransmits of RPC requests should normally
666 	 * use the same xid.
667 	 */
668 	if (nmp == NULL) {
669 		timo.tv_usec = 0;
670 		if (clp == NULL)
671 			timo.tv_sec = NFSV4_UPCALLTIMEO;
672 		else
673 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
674 	} else {
675 		if (nrp->nr_sotype != SOCK_DGRAM) {
676 			timo.tv_usec = 0;
677 			if ((nmp->nm_flag & NFSMNT_NFSV4))
678 				timo.tv_sec = INT_MAX;
679 			else
680 				timo.tv_sec = NFS_TCPTIMEO;
681 		} else {
682 			if (NFSHASSOFT(nmp)) {
683 				/*
684 				 * CLSET_RETRIES is set to 2, so this should be
685 				 * half of the total timeout required.
686 				 */
687 				timeo = nmp->nm_retry * nmp->nm_timeo / 2;
688 				if (timeo < 1)
689 					timeo = 1;
690 				timo.tv_sec = timeo / NFS_HZ;
691 				timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
692 				    NFS_HZ;
693 			} else {
694 				/* For UDP hard mounts, use a large value. */
695 				timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
696 				timo.tv_usec = 0;
697 			}
698 		}
699 
700 		if (rep != NULL) {
701 			rep->r_flags = 0;
702 			rep->r_nmp = nmp;
703 			/*
704 			 * Chain request into list of outstanding requests.
705 			 */
706 			NFSLOCKREQ();
707 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
708 			NFSUNLOCKREQ();
709 		}
710 	}
711 
712 	nd->nd_mrep = NULL;
713 	stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
714 	    &nd->nd_mrep, timo);
715 
716 	if (rep != NULL) {
717 		/*
718 		 * RPC done, unlink the request.
719 		 */
720 		NFSLOCKREQ();
721 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
722 		NFSUNLOCKREQ();
723 	}
724 
725 	/*
726 	 * If there was a successful reply and a tprintf msg.
727 	 * tprintf a response.
728 	 */
729 	if (stat == RPC_SUCCESS) {
730 		error = 0;
731 	} else if (stat == RPC_TIMEDOUT) {
732 		error = ETIMEDOUT;
733 	} else if (stat == RPC_VERSMISMATCH) {
734 		error = EOPNOTSUPP;
735 	} else if (stat == RPC_PROGVERSMISMATCH) {
736 		error = EPROTONOSUPPORT;
737 	} else {
738 		error = EACCES;
739 	}
740 	if (error) {
741 		m_freem(nd->nd_mreq);
742 		AUTH_DESTROY(auth);
743 		if (rep != NULL)
744 			FREE((caddr_t)rep, M_NFSDREQ);
745 		if (set_sigset)
746 			newnfs_restore_sigmask(td, &oldset);
747 		return (error);
748 	}
749 
750 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
751 
752 	/*
753 	 * Search for any mbufs that are not a multiple of 4 bytes long
754 	 * or with m_data not longword aligned.
755 	 * These could cause pointer alignment problems, so copy them to
756 	 * well aligned mbufs.
757 	 */
758 	newnfs_realign(&nd->nd_mrep);
759 	nd->nd_md = nd->nd_mrep;
760 	nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
761 	nd->nd_repstat = 0;
762 	if (nd->nd_procnum != NFSPROC_NULL) {
763 		/*
764 		 * and now the actual NFS xdr.
765 		 */
766 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
767 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
768 		if (nd->nd_repstat != 0) {
769 			if (((nd->nd_repstat == NFSERR_DELAY ||
770 			      nd->nd_repstat == NFSERR_GRACE) &&
771 			     (nd->nd_flag & ND_NFSV4) &&
772 			     nd->nd_procnum != NFSPROC_DELEGRETURN &&
773 			     nd->nd_procnum != NFSPROC_SETATTR &&
774 			     nd->nd_procnum != NFSPROC_READ &&
775 			     nd->nd_procnum != NFSPROC_WRITE &&
776 			     nd->nd_procnum != NFSPROC_OPEN &&
777 			     nd->nd_procnum != NFSPROC_CREATE &&
778 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
779 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
780 			     nd->nd_procnum != NFSPROC_CLOSE &&
781 			     nd->nd_procnum != NFSPROC_LOCK &&
782 			     nd->nd_procnum != NFSPROC_LOCKU) ||
783 			    (nd->nd_repstat == NFSERR_DELAY &&
784 			     (nd->nd_flag & ND_NFSV4) == 0) ||
785 			    nd->nd_repstat == NFSERR_RESOURCE) {
786 				if (trylater_delay > NFS_TRYLATERDEL)
787 					trylater_delay = NFS_TRYLATERDEL;
788 				waituntil = NFSD_MONOSEC + trylater_delay;
789 				while (NFSD_MONOSEC < waituntil)
790 					(void) nfs_catnap(PZERO, 0, "nfstry");
791 				trylater_delay *= 2;
792 				m_freem(nd->nd_mrep);
793 				nd->nd_mrep = NULL;
794 				goto tryagain;
795 			}
796 
797 			/*
798 			 * If the File Handle was stale, invalidate the
799 			 * lookup cache, just in case.
800 			 * (vp != NULL implies a client side call)
801 			 */
802 			if (nd->nd_repstat == ESTALE && vp != NULL) {
803 				cache_purge(vp);
804 				if (ncl_call_invalcaches != NULL)
805 					(*ncl_call_invalcaches)(vp);
806 			}
807 		}
808 
809 		/*
810 		 * Get rid of the tag, return count, and PUTFH result for V4.
811 		 */
812 		if (nd->nd_flag & ND_NFSV4) {
813 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
814 			i = fxdr_unsigned(int, *tl);
815 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
816 			if (error)
817 				goto nfsmout;
818 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
819 			i = fxdr_unsigned(int, *++tl);
820 
821 			/*
822 			 * If the first op's status is non-zero, mark that
823 			 * there is no more data to process.
824 			 */
825 			if (*++tl)
826 				nd->nd_flag |= ND_NOMOREDATA;
827 
828 			/*
829 			 * If the first op is Putfh, throw its results away
830 			 * and toss the op# and status for the first op.
831 			 */
832 			if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) {
833 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
834 				i = fxdr_unsigned(int, *tl++);
835 				j = fxdr_unsigned(int, *tl);
836 				/*
837 				 * All Compounds that do an Op that must
838 				 * be in sequence consist of NFSV4OP_PUTFH
839 				 * followed by one of these. As such, we
840 				 * can determine if the seqid# should be
841 				 * incremented, here.
842 				 */
843 				if ((i == NFSV4OP_OPEN ||
844 				     i == NFSV4OP_OPENCONFIRM ||
845 				     i == NFSV4OP_OPENDOWNGRADE ||
846 				     i == NFSV4OP_CLOSE ||
847 				     i == NFSV4OP_LOCK ||
848 				     i == NFSV4OP_LOCKU) &&
849 				    (j == 0 ||
850 				     (j != NFSERR_STALECLIENTID &&
851 				      j != NFSERR_STALESTATEID &&
852 				      j != NFSERR_BADSTATEID &&
853 				      j != NFSERR_BADSEQID &&
854 				      j != NFSERR_BADXDR &&
855 				      j != NFSERR_RESOURCE &&
856 				      j != NFSERR_NOFILEHANDLE)))
857 					nd->nd_flag |= ND_INCRSEQID;
858 				/*
859 				 * If the first op's status is non-zero, mark
860 				 * that there is no more data to process.
861 				 */
862 				if (j)
863 					nd->nd_flag |= ND_NOMOREDATA;
864 			}
865 
866 			/*
867 			 * If R_DONTRECOVER is set, replace the stale error
868 			 * reply, so that recovery isn't initiated.
869 			 */
870 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
871 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
872 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
873 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
874 		}
875 	}
876 
877 #ifdef KDTRACE_HOOKS
878 	if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
879 		uint32_t probe_id;
880 		int probe_procnum;
881 
882 		if (nd->nd_flag & ND_NFSV4) {
883 			probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
884 			probe_procnum = nd->nd_procnum;
885 		} else if (nd->nd_flag & ND_NFSV3) {
886 			probe_id = nfscl_nfs3_done_probes[procnum];
887 			probe_procnum = procnum;
888 		} else {
889 			probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
890 			probe_procnum = procnum;
891 		}
892 		if (probe_id != 0)
893 			(dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
894 			    nd->nd_mreq, cred, probe_procnum, 0);
895 	}
896 #endif
897 
898 	m_freem(nd->nd_mreq);
899 	AUTH_DESTROY(auth);
900 	if (rep != NULL)
901 		FREE((caddr_t)rep, M_NFSDREQ);
902 	if (set_sigset)
903 		newnfs_restore_sigmask(td, &oldset);
904 	return (0);
905 nfsmout:
906 	mbuf_freem(nd->nd_mrep);
907 	mbuf_freem(nd->nd_mreq);
908 	AUTH_DESTROY(auth);
909 	if (rep != NULL)
910 		FREE((caddr_t)rep, M_NFSDREQ);
911 	if (set_sigset)
912 		newnfs_restore_sigmask(td, &oldset);
913 	return (error);
914 }
915 
916 /*
917  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
918  * wait for all requests to complete. This is used by forced unmounts
919  * to terminate any outstanding RPCs.
920  */
921 int
922 newnfs_nmcancelreqs(struct nfsmount *nmp)
923 {
924 
925 	if (nmp->nm_sockreq.nr_client != NULL)
926 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
927 	return (0);
928 }
929 
930 /*
931  * Any signal that can interrupt an NFS operation in an intr mount
932  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
933  */
934 int newnfs_sig_set[] = {
935 	SIGINT,
936 	SIGTERM,
937 	SIGHUP,
938 	SIGKILL,
939 	SIGSTOP,
940 	SIGQUIT
941 };
942 
943 /*
944  * Check to see if one of the signals in our subset is pending on
945  * the process (in an intr mount).
946  */
947 static int
948 nfs_sig_pending(sigset_t set)
949 {
950 	int i;
951 
952 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
953 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
954 			return (1);
955 	return (0);
956 }
957 
958 /*
959  * The set/restore sigmask functions are used to (temporarily) overwrite
960  * the process p_sigmask during an RPC call (for example). These are also
961  * used in other places in the NFS client that might tsleep().
962  */
963 void
964 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
965 {
966 	sigset_t newset;
967 	int i;
968 	struct proc *p;
969 
970 	SIGFILLSET(newset);
971 	if (td == NULL)
972 		td = curthread; /* XXX */
973 	p = td->td_proc;
974 	/* Remove the NFS set of signals from newset */
975 	PROC_LOCK(p);
976 	mtx_lock(&p->p_sigacts->ps_mtx);
977 	for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
978 		/*
979 		 * But make sure we leave the ones already masked
980 		 * by the process, ie. remove the signal from the
981 		 * temporary signalmask only if it wasn't already
982 		 * in p_sigmask.
983 		 */
984 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
985 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
986 			SIGDELSET(newset, newnfs_sig_set[i]);
987 	}
988 	mtx_unlock(&p->p_sigacts->ps_mtx);
989 	PROC_UNLOCK(p);
990 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0);
991 }
992 
993 void
994 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
995 {
996 	if (td == NULL)
997 		td = curthread; /* XXX */
998 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
999 }
1000 
1001 /*
1002  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1003  * old one after msleep() returns.
1004  */
1005 int
1006 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1007 {
1008 	sigset_t oldset;
1009 	int error;
1010 	struct proc *p;
1011 
1012 	if ((priority & PCATCH) == 0)
1013 		return msleep(ident, mtx, priority, wmesg, timo);
1014 	if (td == NULL)
1015 		td = curthread; /* XXX */
1016 	newnfs_set_sigmask(td, &oldset);
1017 	error = msleep(ident, mtx, priority, wmesg, timo);
1018 	newnfs_restore_sigmask(td, &oldset);
1019 	p = td->td_proc;
1020 	return (error);
1021 }
1022 
1023 /*
1024  * Test for a termination condition pending on the process.
1025  * This is used for NFSMNT_INT mounts.
1026  */
1027 int
1028 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1029 {
1030 	struct proc *p;
1031 	sigset_t tmpset;
1032 
1033 	/* Terminate all requests while attempting a forced unmount. */
1034 	if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1035 		return (EIO);
1036 	if (!(nmp->nm_flag & NFSMNT_INT))
1037 		return (0);
1038 	if (td == NULL)
1039 		return (0);
1040 	p = td->td_proc;
1041 	PROC_LOCK(p);
1042 	tmpset = p->p_siglist;
1043 	SIGSETOR(tmpset, td->td_siglist);
1044 	SIGSETNAND(tmpset, td->td_sigmask);
1045 	mtx_lock(&p->p_sigacts->ps_mtx);
1046 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1047 	mtx_unlock(&p->p_sigacts->ps_mtx);
1048 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1049 	    && nfs_sig_pending(tmpset)) {
1050 		PROC_UNLOCK(p);
1051 		return (EINTR);
1052 	}
1053 	PROC_UNLOCK(p);
1054 	return (0);
1055 }
1056 
1057 static int
1058 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1059 {
1060 	struct proc *p;
1061 
1062 	p = td ? td->td_proc : NULL;
1063 	if (error) {
1064 		tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
1065 		    server, msg, error);
1066 	} else {
1067 		tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
1068 	}
1069 	return (0);
1070 }
1071 
1072 static void
1073 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1074     int error, int flags)
1075 {
1076 	if (nmp == NULL)
1077 		return;
1078 	mtx_lock(&nmp->nm_mtx);
1079 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1080 		nmp->nm_state |= NFSSTA_TIMEO;
1081 		mtx_unlock(&nmp->nm_mtx);
1082 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1083 		    VQ_NOTRESP, 0);
1084 	} else
1085 		mtx_unlock(&nmp->nm_mtx);
1086 	mtx_lock(&nmp->nm_mtx);
1087 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1088 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
1089 		mtx_unlock(&nmp->nm_mtx);
1090 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1091 		    VQ_NOTRESPLOCK, 0);
1092 	} else
1093 		mtx_unlock(&nmp->nm_mtx);
1094 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1095 }
1096 
1097 static void
1098 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1099     int flags, int tprintfmsg)
1100 {
1101 	if (nmp == NULL)
1102 		return;
1103 	if (tprintfmsg) {
1104 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1105 	}
1106 
1107 	mtx_lock(&nmp->nm_mtx);
1108 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1109 		nmp->nm_state &= ~NFSSTA_TIMEO;
1110 		mtx_unlock(&nmp->nm_mtx);
1111 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1112 		    VQ_NOTRESP, 1);
1113 	} else
1114 		mtx_unlock(&nmp->nm_mtx);
1115 
1116 	mtx_lock(&nmp->nm_mtx);
1117 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1118 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1119 		mtx_unlock(&nmp->nm_mtx);
1120 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1121 		    VQ_NOTRESPLOCK, 1);
1122 	} else
1123 		mtx_unlock(&nmp->nm_mtx);
1124 }
1125 
1126