nfs_krpc.c revision 227690
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 *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: head/sys/nfsclient/nfs_krpc.c 227690 2011-11-19 03:20:15Z rmacklem $");
37
38/*
39 * Socket operations for use by nfs
40 */
41
42#include "opt_inet6.h"
43#include "opt_kdtrace.h"
44#include "opt_kgssapi.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 <nfs/nfsproto.h>
65#include <nfsclient/nfs.h>
66#include <nfs/xdr_subs.h>
67#include <nfsclient/nfsm_subs.h>
68#include <nfsclient/nfsmount.h>
69#include <nfsclient/nfsnode.h>
70
71#ifdef KDTRACE_HOOKS
72#include <sys/dtrace_bsd.h>
73
74dtrace_nfsclient_nfs23_start_probe_func_t
75    dtrace_nfsclient_nfs23_start_probe;
76
77dtrace_nfsclient_nfs23_done_probe_func_t
78    dtrace_nfsclient_nfs23_done_probe;
79
80/*
81 * Registered probes by RPC type.
82 */
83uint32_t	nfsclient_nfs2_start_probes[NFS_NPROCS];
84uint32_t	nfsclient_nfs2_done_probes[NFS_NPROCS];
85
86uint32_t	nfsclient_nfs3_start_probes[NFS_NPROCS];
87uint32_t	nfsclient_nfs3_done_probes[NFS_NPROCS];
88#endif
89
90static int	nfs_bufpackets = 4;
91static int	nfs_reconnects;
92static int	nfs3_jukebox_delay = 10;
93static int	nfs_skip_wcc_data_onerr = 1;
94static int	fake_wchan;
95
96SYSCTL_DECL(_vfs_oldnfs);
97
98SYSCTL_INT(_vfs_oldnfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
99    "Buffer reservation size 2 < x < 64");
100SYSCTL_INT(_vfs_oldnfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
101    "Number of times the nfs client has had to reconnect");
102SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW,
103    &nfs3_jukebox_delay, 0,
104    "Number of seconds to delay a retry after receiving EJUKEBOX");
105SYSCTL_INT(_vfs_oldnfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW,
106    &nfs_skip_wcc_data_onerr, 0,
107    "Disable weak cache consistency checking when server returns an error");
108
109static void	nfs_down(struct nfsmount *, struct thread *, const char *,
110    int, int);
111static void	nfs_up(struct nfsmount *, struct thread *, const char *,
112    int, int);
113static int	nfs_msg(struct thread *, const char *, const char *, int);
114
115extern int nfsv2_procid[];
116
117struct nfs_cached_auth {
118	int		ca_refs; /* refcount, including 1 from the cache */
119	uid_t		ca_uid;	 /* uid that corresponds to this auth */
120	AUTH		*ca_auth; /* RPC auth handle */
121};
122
123/*
124 * RTT estimator
125 */
126
127static enum nfs_rto_timer_t nfs_proct[NFS_NPROCS] = {
128	NFS_DEFAULT_TIMER,	/* NULL */
129	NFS_GETATTR_TIMER,	/* GETATTR */
130	NFS_DEFAULT_TIMER,	/* SETATTR */
131	NFS_LOOKUP_TIMER,	/* LOOKUP */
132	NFS_GETATTR_TIMER,	/* ACCESS */
133	NFS_READ_TIMER,		/* READLINK */
134	NFS_READ_TIMER,		/* READ */
135	NFS_WRITE_TIMER,	/* WRITE */
136	NFS_DEFAULT_TIMER,	/* CREATE */
137	NFS_DEFAULT_TIMER,	/* MKDIR */
138	NFS_DEFAULT_TIMER,	/* SYMLINK */
139	NFS_DEFAULT_TIMER,	/* MKNOD */
140	NFS_DEFAULT_TIMER,	/* REMOVE */
141	NFS_DEFAULT_TIMER,	/* RMDIR */
142	NFS_DEFAULT_TIMER,	/* RENAME */
143	NFS_DEFAULT_TIMER,	/* LINK */
144	NFS_READ_TIMER,		/* READDIR */
145	NFS_READ_TIMER,		/* READDIRPLUS */
146	NFS_DEFAULT_TIMER,	/* FSSTAT */
147	NFS_DEFAULT_TIMER,	/* FSINFO */
148	NFS_DEFAULT_TIMER,	/* PATHCONF */
149	NFS_DEFAULT_TIMER,	/* COMMIT */
150	NFS_DEFAULT_TIMER,	/* NOOP */
151};
152
153/*
154 * Choose the correct RTT timer for this NFS procedure.
155 */
156static inline enum nfs_rto_timer_t
157nfs_rto_timer(u_int32_t procnum)
158{
159
160	return (nfs_proct[procnum]);
161}
162
163/*
164 * Initialize the RTT estimator state for a new mount point.
165 */
166static void
167nfs_init_rtt(struct nfsmount *nmp)
168{
169	int i;
170
171	for (i = 0; i < NFS_MAX_TIMER; i++) {
172		nmp->nm_timers[i].rt_srtt = hz;
173		nmp->nm_timers[i].rt_deviate = 0;
174		nmp->nm_timers[i].rt_rtxcur = hz;
175	}
176}
177
178/*
179 * Initialize sockets and congestion for a new NFS connection.
180 * We do not free the sockaddr if error.
181 */
182int
183nfs_connect(struct nfsmount *nmp)
184{
185	int rcvreserve, sndreserve;
186	int pktscale;
187	struct sockaddr *saddr;
188	struct ucred *origcred;
189	struct thread *td = curthread;
190	CLIENT *client;
191	struct netconfig *nconf;
192	rpcvers_t vers;
193	int one = 1, retries;
194
195	/*
196	 * We need to establish the socket using the credentials of
197	 * the mountpoint.  Some parts of this process (such as
198	 * sobind() and soconnect()) will use the curent thread's
199	 * credential instead of the socket credential.  To work
200	 * around this, temporarily change the current thread's
201	 * credential to that of the mountpoint.
202	 *
203	 * XXX: It would be better to explicitly pass the correct
204	 * credential to sobind() and soconnect().
205	 */
206	origcred = td->td_ucred;
207	td->td_ucred = nmp->nm_mountp->mnt_cred;
208	saddr = nmp->nm_nam;
209
210	vers = NFS_VER2;
211	if (nmp->nm_flag & NFSMNT_NFSV3)
212		vers = NFS_VER3;
213	else if (nmp->nm_flag & NFSMNT_NFSV4)
214		vers = NFS_VER4;
215	if (saddr->sa_family == AF_INET)
216		if (nmp->nm_sotype == SOCK_DGRAM)
217			nconf = getnetconfigent("udp");
218		else
219			nconf = getnetconfigent("tcp");
220	else
221		if (nmp->nm_sotype == SOCK_DGRAM)
222			nconf = getnetconfigent("udp6");
223		else
224			nconf = getnetconfigent("tcp6");
225
226	/*
227	 * Get buffer reservation size from sysctl, but impose reasonable
228	 * limits.
229	 */
230	pktscale = nfs_bufpackets;
231	if (pktscale < 2)
232		pktscale = 2;
233	if (pktscale > 64)
234		pktscale = 64;
235	mtx_lock(&nmp->nm_mtx);
236	if (nmp->nm_sotype == SOCK_DGRAM) {
237		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
238		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
239		    NFS_MAXPKTHDR) * pktscale;
240	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
241		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
242		rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
243		    NFS_MAXPKTHDR) * pktscale;
244	} else {
245		if (nmp->nm_sotype != SOCK_STREAM)
246			panic("nfscon sotype");
247		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
248		    sizeof (u_int32_t)) * pktscale;
249		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
250		    sizeof (u_int32_t)) * pktscale;
251	}
252	mtx_unlock(&nmp->nm_mtx);
253
254	client = clnt_reconnect_create(nconf, saddr, NFS_PROG, vers,
255	    sndreserve, rcvreserve);
256	CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
257	if (nmp->nm_flag & NFSMNT_INT)
258		CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
259	if (nmp->nm_flag & NFSMNT_RESVPORT)
260		CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
261	if (nmp->nm_flag & NFSMNT_SOFT)
262		retries = nmp->nm_retry;
263	else
264		retries = INT_MAX;
265	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
266
267	mtx_lock(&nmp->nm_mtx);
268	if (nmp->nm_client) {
269		/*
270		 * Someone else already connected.
271		 */
272		CLNT_RELEASE(client);
273	} else
274		nmp->nm_client = client;
275
276	/*
277	 * Protocols that do not require connections may be optionally left
278	 * unconnected for servers that reply from a port other than NFS_PORT.
279	 */
280	if (!(nmp->nm_flag & NFSMNT_NOCONN)) {
281		mtx_unlock(&nmp->nm_mtx);
282		CLNT_CONTROL(client, CLSET_CONNECT, &one);
283	} else
284		mtx_unlock(&nmp->nm_mtx);
285
286	/* Restore current thread's credentials. */
287	td->td_ucred = origcred;
288
289	mtx_lock(&nmp->nm_mtx);
290	/* Initialize other non-zero congestion variables. */
291	nfs_init_rtt(nmp);
292	mtx_unlock(&nmp->nm_mtx);
293	return (0);
294}
295
296/*
297 * NFS disconnect.  Clean up and unlink.
298 */
299void
300nfs_disconnect(struct nfsmount *nmp)
301{
302	CLIENT *client;
303
304	mtx_lock(&nmp->nm_mtx);
305	if (nmp->nm_client) {
306		client = nmp->nm_client;
307		nmp->nm_client = NULL;
308		mtx_unlock(&nmp->nm_mtx);
309		rpc_gss_secpurge_call(client);
310		CLNT_CLOSE(client);
311		CLNT_RELEASE(client);
312	} else
313		mtx_unlock(&nmp->nm_mtx);
314}
315
316void
317nfs_safedisconnect(struct nfsmount *nmp)
318{
319
320	nfs_disconnect(nmp);
321}
322
323static AUTH *
324nfs_getauth(struct nfsmount *nmp, struct ucred *cred)
325{
326	rpc_gss_service_t svc;
327	AUTH *auth;
328
329	switch (nmp->nm_secflavor) {
330	case RPCSEC_GSS_KRB5:
331	case RPCSEC_GSS_KRB5I:
332	case RPCSEC_GSS_KRB5P:
333		if (!nmp->nm_mech_oid)
334			if (!rpc_gss_mech_to_oid_call("kerberosv5",
335			    &nmp->nm_mech_oid))
336				return (NULL);
337		if (nmp->nm_secflavor == RPCSEC_GSS_KRB5)
338			svc = rpc_gss_svc_none;
339		else if (nmp->nm_secflavor == RPCSEC_GSS_KRB5I)
340			svc = rpc_gss_svc_integrity;
341		else
342			svc = rpc_gss_svc_privacy;
343		auth = rpc_gss_secfind_call(nmp->nm_client, cred,
344		    nmp->nm_principal, nmp->nm_mech_oid, svc);
345		if (auth)
346			return (auth);
347		/* fallthrough */
348	case AUTH_SYS:
349	default:
350		return (authunix_create(cred));
351
352	}
353}
354
355/*
356 * Callback from the RPC code to generate up/down notifications.
357 */
358
359struct nfs_feedback_arg {
360	struct nfsmount *nf_mount;
361	int		nf_lastmsg;	/* last tprintf */
362	int		nf_tprintfmsg;
363	struct thread	*nf_td;
364};
365
366static void
367nfs_feedback(int type, int proc, void *arg)
368{
369	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
370	struct nfsmount *nmp = nf->nf_mount;
371	struct timeval now;
372
373	getmicrouptime(&now);
374
375	switch (type) {
376	case FEEDBACK_REXMIT2:
377	case FEEDBACK_RECONNECT:
378		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
379			nfs_down(nmp, nf->nf_td,
380			    "not responding", 0, NFSSTA_TIMEO);
381			nf->nf_tprintfmsg = TRUE;
382			nf->nf_lastmsg = now.tv_sec;
383		}
384		break;
385
386	case FEEDBACK_OK:
387		nfs_up(nf->nf_mount, nf->nf_td,
388		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
389		break;
390	}
391}
392
393/*
394 * nfs_request - goes something like this
395 *	- fill in request struct
396 *	- links it into list
397 *	- calls nfs_send() for first transmit
398 *	- calls nfs_receive() to get reply
399 *	- break down rpc header and return with nfs reply pointed to
400 *	  by mrep or error
401 * nb: always frees up mreq mbuf list
402 */
403int
404nfs_request(struct vnode *vp, struct mbuf *mreq, int procnum,
405    struct thread *td, struct ucred *cred, struct mbuf **mrp,
406    struct mbuf **mdp, caddr_t *dposp)
407{
408	struct mbuf *mrep;
409	u_int32_t *tl;
410	struct nfsmount *nmp;
411	struct mbuf *md;
412	time_t waituntil;
413	caddr_t dpos;
414	int error = 0;
415	struct timeval now;
416	AUTH *auth = NULL;
417	enum nfs_rto_timer_t timer;
418	struct nfs_feedback_arg nf;
419	struct rpc_callextra ext;
420	enum clnt_stat stat;
421	struct timeval timo;
422
423	/* Reject requests while attempting a forced unmount. */
424	if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
425		m_freem(mreq);
426		return (ESTALE);
427	}
428	nmp = VFSTONFS(vp->v_mount);
429	bzero(&nf, sizeof(struct nfs_feedback_arg));
430	nf.nf_mount = nmp;
431	nf.nf_td = td;
432	getmicrouptime(&now);
433	nf.nf_lastmsg = now.tv_sec -
434	    ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay));
435
436	/*
437	 * XXX if not already connected call nfs_connect now.  Longer
438	 * term, change nfs_mount to call nfs_connect unconditionally
439	 * and let clnt_reconnect_create handle reconnects.
440	 */
441	if (!nmp->nm_client)
442		nfs_connect(nmp);
443
444	auth = nfs_getauth(nmp, cred);
445	if (!auth) {
446		m_freem(mreq);
447		return (EACCES);
448	}
449	bzero(&ext, sizeof(ext));
450	ext.rc_auth = auth;
451
452	ext.rc_feedback = nfs_feedback;
453	ext.rc_feedback_arg = &nf;
454
455	/*
456	 * Use a conservative timeout for RPCs other than getattr,
457	 * lookup, read or write.  The justification for doing "other"
458	 * this way is that these RPCs happen so infrequently that
459	 * timer est. would probably be stale.  Also, since many of
460	 * these RPCs are non-idempotent, a conservative timeout is
461	 * desired.
462	 */
463	timer = nfs_rto_timer(procnum);
464	if (timer != NFS_DEFAULT_TIMER)
465		ext.rc_timers = &nmp->nm_timers[timer - 1];
466	else
467		ext.rc_timers = NULL;
468
469#ifdef KDTRACE_HOOKS
470	if (dtrace_nfsclient_nfs23_start_probe != NULL) {
471		uint32_t probe_id;
472		int probe_procnum;
473
474		if (nmp->nm_flag & NFSMNT_NFSV3) {
475			probe_id = nfsclient_nfs3_start_probes[procnum];
476			probe_procnum = procnum;
477		} else {
478			probe_id = nfsclient_nfs2_start_probes[procnum];
479			probe_procnum = nfsv2_procid[procnum];
480		}
481		if (probe_id != 0)
482			(dtrace_nfsclient_nfs23_start_probe)(probe_id, vp,
483			    mreq, cred, probe_procnum);
484	}
485#endif
486
487	nfsstats.rpcrequests++;
488tryagain:
489	timo.tv_sec = nmp->nm_timeo / NFS_HZ;
490	timo.tv_usec = (nmp->nm_timeo * 1000000) / NFS_HZ;
491	mrep = NULL;
492	stat = CLNT_CALL_MBUF(nmp->nm_client, &ext,
493	    (nmp->nm_flag & NFSMNT_NFSV3) ? procnum : nfsv2_procid[procnum],
494	    mreq, &mrep, timo);
495
496	/*
497	 * If there was a successful reply and a tprintf msg.
498	 * tprintf a response.
499	 */
500	if (stat == RPC_SUCCESS)
501		error = 0;
502	else if (stat == RPC_TIMEDOUT)
503		error = ETIMEDOUT;
504	else if (stat == RPC_VERSMISMATCH)
505		error = EOPNOTSUPP;
506	else if (stat == RPC_PROGVERSMISMATCH)
507		error = EPROTONOSUPPORT;
508	else
509		error = EACCES;
510	if (error)
511		goto nfsmout;
512
513	KASSERT(mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
514
515	/*
516	 * Search for any mbufs that are not a multiple of 4 bytes long
517	 * or with m_data not longword aligned.
518	 * These could cause pointer alignment problems, so copy them to
519	 * well aligned mbufs.
520	 */
521	error = nfs_realign(&mrep, M_DONTWAIT);
522	if (error == ENOMEM) {
523		m_freem(mrep);
524		AUTH_DESTROY(auth);
525		return (error);
526	}
527
528	md = mrep;
529	dpos = mtod(mrep, caddr_t);
530	tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
531	if (*tl != 0) {
532		error = fxdr_unsigned(int, *tl);
533		if ((nmp->nm_flag & NFSMNT_NFSV3) &&
534		    error == NFSERR_TRYLATER) {
535			m_freem(mrep);
536			error = 0;
537			waituntil = time_second + nfs3_jukebox_delay;
538			while (time_second < waituntil)
539				(void)tsleep(&fake_wchan, PSOCK, "nqnfstry",
540				    hz);
541			goto tryagain;
542		}
543		/*
544		 * Make sure NFSERR_RETERR isn't bogusly set by a server
545		 * such as amd. (No actual NFS error has bit 31 set.)
546		 */
547		error &= ~NFSERR_RETERR;
548
549		/*
550		 * If the File Handle was stale, invalidate the lookup
551		 * cache, just in case.
552		 */
553		if (error == ESTALE)
554			nfs_purgecache(vp);
555		/*
556		 * Skip wcc data on NFS errors for now.  NetApp filers
557		 * return corrupt postop attrs in the wcc data for NFS
558		 * err EROFS.  Not sure if they could return corrupt
559		 * postop attrs for others errors.
560		 */
561		if ((nmp->nm_flag & NFSMNT_NFSV3) &&
562		    !nfs_skip_wcc_data_onerr) {
563			*mrp = mrep;
564			*mdp = md;
565			*dposp = dpos;
566			error |= NFSERR_RETERR;
567		} else
568			m_freem(mrep);
569		goto nfsmout;
570	}
571
572#ifdef KDTRACE_HOOKS
573	if (dtrace_nfsclient_nfs23_done_probe != NULL) {
574		uint32_t probe_id;
575		int probe_procnum;
576
577		if (nmp->nm_flag & NFSMNT_NFSV3) {
578			probe_id = nfsclient_nfs3_done_probes[procnum];
579			probe_procnum = procnum;
580		} else {
581			probe_id = nfsclient_nfs2_done_probes[procnum];
582			probe_procnum = (nmp->nm_flag & NFSMNT_NFSV3) ?
583			    procnum : nfsv2_procid[procnum];
584		}
585		if (probe_id != 0)
586			(dtrace_nfsclient_nfs23_done_probe)(probe_id, vp,
587			    mreq, cred, probe_procnum, 0);
588	}
589#endif
590	m_freem(mreq);
591	*mrp = mrep;
592	*mdp = md;
593	*dposp = dpos;
594	AUTH_DESTROY(auth);
595	return (0);
596
597nfsmout:
598#ifdef KDTRACE_HOOKS
599	if (dtrace_nfsclient_nfs23_done_probe != NULL) {
600		uint32_t probe_id;
601		int probe_procnum;
602
603		if (nmp->nm_flag & NFSMNT_NFSV3) {
604			probe_id = nfsclient_nfs3_done_probes[procnum];
605			probe_procnum = procnum;
606		} else {
607			probe_id = nfsclient_nfs2_done_probes[procnum];
608			probe_procnum = (nmp->nm_flag & NFSMNT_NFSV3) ?
609			    procnum : nfsv2_procid[procnum];
610		}
611		if (probe_id != 0)
612			(dtrace_nfsclient_nfs23_done_probe)(probe_id, vp,
613			    mreq, cred, probe_procnum, error);
614	}
615#endif
616	m_freem(mreq);
617	if (auth)
618		AUTH_DESTROY(auth);
619	return (error);
620}
621
622/*
623 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
624 * wait for all requests to complete.  This is used by forced unmounts
625 * to terminate any outstanding RPCs.
626 */
627int
628nfs_nmcancelreqs(struct nfsmount *nmp)
629{
630
631	if (nmp->nm_client)
632		CLNT_CLOSE(nmp->nm_client);
633	return (0);
634}
635
636/*
637 * Any signal that can interrupt an NFS operation in an intr mount
638 * should be added to this set.  SIGSTOP and SIGKILL cannot be masked.
639 */
640int nfs_sig_set[] = {
641	SIGINT,
642	SIGTERM,
643	SIGHUP,
644	SIGKILL,
645	SIGSTOP,
646	SIGQUIT
647};
648
649/*
650 * Check to see if one of the signals in our subset is pending on
651 * the process (in an intr mount).
652 */
653static int
654nfs_sig_pending(sigset_t set)
655{
656	int i;
657
658	for (i = 0 ; i < sizeof(nfs_sig_set)/sizeof(int) ; i++)
659		if (SIGISMEMBER(set, nfs_sig_set[i]))
660			return (1);
661	return (0);
662}
663
664/*
665 * The set/restore sigmask functions are used to (temporarily) overwrite
666 * the process p_sigmask during an RPC call (for example).  These are also
667 * used in other places in the NFS client that might tsleep().
668 */
669void
670nfs_set_sigmask(struct thread *td, sigset_t *oldset)
671{
672	sigset_t newset;
673	int i;
674	struct proc *p;
675
676	SIGFILLSET(newset);
677	if (td == NULL)
678		td = curthread; /* XXX */
679	p = td->td_proc;
680	/* Remove the NFS set of signals from newset. */
681	PROC_LOCK(p);
682	mtx_lock(&p->p_sigacts->ps_mtx);
683	for (i = 0 ; i < sizeof(nfs_sig_set)/sizeof(int) ; i++) {
684		/*
685		 * But make sure we leave the ones already masked
686		 * by the process, i.e. remove the signal from the
687		 * temporary signalmask only if it wasn't already
688		 * in p_sigmask.
689		 */
690		if (!SIGISMEMBER(td->td_sigmask, nfs_sig_set[i]) &&
691		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, nfs_sig_set[i]))
692			SIGDELSET(newset, nfs_sig_set[i]);
693	}
694	mtx_unlock(&p->p_sigacts->ps_mtx);
695	PROC_UNLOCK(p);
696	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0);
697}
698
699void
700nfs_restore_sigmask(struct thread *td, sigset_t *set)
701{
702	if (td == NULL)
703		td = curthread; /* XXX */
704	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
705}
706
707/*
708 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
709 * old one after msleep() returns.
710 */
711int
712nfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority,
713    char *wmesg, int timo)
714{
715	sigset_t oldset;
716	int error;
717	struct proc *p;
718
719	if ((priority & PCATCH) == 0)
720		return msleep(ident, mtx, priority, wmesg, timo);
721	if (td == NULL)
722		td = curthread; /* XXX */
723	nfs_set_sigmask(td, &oldset);
724	error = msleep(ident, mtx, priority, wmesg, timo);
725	nfs_restore_sigmask(td, &oldset);
726	p = td->td_proc;
727	return (error);
728}
729
730/*
731 * Test for a termination condition pending on the process.
732 * This is used for NFSMNT_INT mounts.
733 */
734int
735nfs_sigintr(struct nfsmount *nmp, struct thread *td)
736{
737	struct proc *p;
738	sigset_t tmpset;
739
740	/* Terminate all requests while attempting a forced unmount. */
741	if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
742		return (EIO);
743	if (!(nmp->nm_flag & NFSMNT_INT))
744		return (0);
745	if (td == NULL)
746		return (0);
747	p = td->td_proc;
748	PROC_LOCK(p);
749	tmpset = p->p_siglist;
750	SIGSETOR(tmpset, td->td_siglist);
751	SIGSETNAND(tmpset, td->td_sigmask);
752	mtx_lock(&p->p_sigacts->ps_mtx);
753	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
754	mtx_unlock(&p->p_sigacts->ps_mtx);
755	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
756	    && nfs_sig_pending(tmpset)) {
757		PROC_UNLOCK(p);
758		return (EINTR);
759	}
760	PROC_UNLOCK(p);
761	return (0);
762}
763
764static int
765nfs_msg(struct thread *td, const char *server, const char *msg, int error)
766{
767	struct proc *p;
768
769	p = td ? td->td_proc : NULL;
770	if (error)
771		tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n", server,
772		    msg, error);
773	else
774		tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
775	return (0);
776}
777
778static void
779nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
780    int error, int flags)
781{
782	if (nmp == NULL)
783		return;
784	mtx_lock(&nmp->nm_mtx);
785	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
786		nmp->nm_state |= NFSSTA_TIMEO;
787		mtx_unlock(&nmp->nm_mtx);
788		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
789		    VQ_NOTRESP, 0);
790	} else
791		mtx_unlock(&nmp->nm_mtx);
792	mtx_lock(&nmp->nm_mtx);
793	if ((flags & NFSSTA_LOCKTIMEO) &&
794	    !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
795		nmp->nm_state |= NFSSTA_LOCKTIMEO;
796		mtx_unlock(&nmp->nm_mtx);
797		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
798		    VQ_NOTRESPLOCK, 0);
799	} else
800		mtx_unlock(&nmp->nm_mtx);
801	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
802}
803
804static void
805nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
806    int flags, int tprintfmsg)
807{
808	if (nmp == NULL)
809		return;
810	if (tprintfmsg)
811		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
812
813	mtx_lock(&nmp->nm_mtx);
814	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
815		nmp->nm_state &= ~NFSSTA_TIMEO;
816		mtx_unlock(&nmp->nm_mtx);
817		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
818		    VQ_NOTRESP, 1);
819	} else
820		mtx_unlock(&nmp->nm_mtx);
821
822	mtx_lock(&nmp->nm_mtx);
823	if ((flags & NFSSTA_LOCKTIMEO) &&
824	    (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
825		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
826		mtx_unlock(&nmp->nm_mtx);
827		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
828		    VQ_NOTRESPLOCK, 1);
829	} else
830		mtx_unlock(&nmp->nm_mtx);
831}
832