Context Navigation

source: rtems-libbsd/freebsd/sys/fs/nfs/nfs_commonkrpc.c @ 7e282d2

6-freebsd-12

Last change on this file since 7e282d2 was 882425f, checked in by Chris Johns <chrisj@…>, on 07/29/21 at 05:49:52

kern/sys: Add NFSv4 client

Update #4475

Property mode set to 100644

File size: 40.8 KB

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

Note: See TracBrowser for help on using the repository browser.

Download in other formats: