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: releng/10.0/sys/nfsclient/nfs_krpc.c 247116 2013-02-21 19:02:50Z jhb $");
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
74 dtrace_nfsclient_nfs23_start_probe_func_t
75 dtrace_nfsclient_nfs23_start_probe;
76
77 dtrace_nfsclient_nfs23_done_probe_func_t
78 dtrace_nfsclient_nfs23_done_probe;
79
80 /*
81 * Registered probes by RPC type.
82 */
83 uint32_t nfsclient_nfs2_start_probes[NFS_NPROCS];
84 uint32_t nfsclient_nfs2_done_probes[NFS_NPROCS];
85
86 uint32_t nfsclient_nfs3_start_probes[NFS_NPROCS];
87 uint32_t nfsclient_nfs3_done_probes[NFS_NPROCS];
88 #endif
89
90 static int nfs_bufpackets = 4;
91 static int nfs_reconnects;
92 static int nfs3_jukebox_delay = 10;
93 static int nfs_skip_wcc_data_onerr = 1;
94 static int fake_wchan;
95
96 SYSCTL_DECL(_vfs_oldnfs);
97
98 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
99 "Buffer reservation size 2 < x < 64");
100 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
101 "Number of times the nfs client has had to reconnect");
102 SYSCTL_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");
105 SYSCTL_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
109 static void nfs_down(struct nfsmount *, struct thread *, const char *,
110 int, int);
111 static void nfs_up(struct nfsmount *, struct thread *, const char *,
112 int, int);
113 static int nfs_msg(struct thread *, const char *, const char *, int);
114
115 extern int nfsv2_procid[];
116
117 struct 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
127 static 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 */
156 static inline enum nfs_rto_timer_t
157 nfs_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 */
166 static void
167 nfs_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 */
182 int
183 nfs_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 struct timeval timo;
195
196 /*
197 * We need to establish the socket using the credentials of
198 * the mountpoint. Some parts of this process (such as
199 * sobind() and soconnect()) will use the curent thread's
200 * credential instead of the socket credential. To work
201 * around this, temporarily change the current thread's
202 * credential to that of the mountpoint.
203 *
204 * XXX: It would be better to explicitly pass the correct
205 * credential to sobind() and soconnect().
206 */
207 origcred = td->td_ucred;
208 td->td_ucred = nmp->nm_mountp->mnt_cred;
209 saddr = nmp->nm_nam;
210
211 vers = NFS_VER2;
212 if (nmp->nm_flag & NFSMNT_NFSV3)
213 vers = NFS_VER3;
214 else if (nmp->nm_flag & NFSMNT_NFSV4)
215 vers = NFS_VER4;
216 if (saddr->sa_family == AF_INET)
217 if (nmp->nm_sotype == SOCK_DGRAM)
218 nconf = getnetconfigent("udp");
219 else
220 nconf = getnetconfigent("tcp");
221 else
222 if (nmp->nm_sotype == SOCK_DGRAM)
223 nconf = getnetconfigent("udp6");
224 else
225 nconf = getnetconfigent("tcp6");
226
227 /*
228 * Get buffer reservation size from sysctl, but impose reasonable
229 * limits.
230 */
231 pktscale = nfs_bufpackets;
232 if (pktscale < 2)
233 pktscale = 2;
234 if (pktscale > 64)
235 pktscale = 64;
236 mtx_lock(&nmp->nm_mtx);
237 if (nmp->nm_sotype == SOCK_DGRAM) {
238 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
239 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
240 NFS_MAXPKTHDR) * pktscale;
241 } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
242 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
243 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
244 NFS_MAXPKTHDR) * pktscale;
245 } else {
246 if (nmp->nm_sotype != SOCK_STREAM)
247 panic("nfscon sotype");
248 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
249 sizeof (u_int32_t)) * pktscale;
250 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
251 sizeof (u_int32_t)) * pktscale;
252 }
253 mtx_unlock(&nmp->nm_mtx);
254
255 client = clnt_reconnect_create(nconf, saddr, NFS_PROG, vers,
256 sndreserve, rcvreserve);
257 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
258 if (nmp->nm_flag & NFSMNT_INT)
259 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
260 if (nmp->nm_flag & NFSMNT_RESVPORT)
261 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
262 if ((nmp->nm_flag & NFSMNT_SOFT) != 0) {
263 if (nmp->nm_sotype == SOCK_DGRAM)
264 /*
265 * For UDP, the large timeout for a reconnect will
266 * be set to "nm_retry * nm_timeo / 2", so we only
267 * want to do 2 reconnect timeout retries.
268 */
269 retries = 2;
270 else
271 retries = nmp->nm_retry;
272 } else
273 retries = INT_MAX;
274 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
275
276 /*
277 * For UDP, there are 2 timeouts:
278 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
279 * that does a retransmit of an RPC request using the same socket
280 * and xid. This is what you normally want to do, since NFS
281 * servers depend on "same xid" for their Duplicate Request Cache.
282 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
283 * retransmits on the same socket should fail and a fresh socket
284 * created. Each of these timeouts counts as one CLSET_RETRIES,
285 * as set above.
286 * Set the initial retransmit timeout for UDP. This timeout doesn't
287 * exist for TCP and the following call just fails, which is ok.
288 */
289 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
290 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
291 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
292
293 mtx_lock(&nmp->nm_mtx);
294 if (nmp->nm_client) {
295 /*
296 * Someone else already connected.
297 */
298 CLNT_RELEASE(client);
299 } else
300 nmp->nm_client = client;
301
302 /*
303 * Protocols that do not require connections may be optionally left
304 * unconnected for servers that reply from a port other than NFS_PORT.
305 */
306 if (!(nmp->nm_flag & NFSMNT_NOCONN)) {
307 mtx_unlock(&nmp->nm_mtx);
308 CLNT_CONTROL(client, CLSET_CONNECT, &one);
309 } else
310 mtx_unlock(&nmp->nm_mtx);
311
312 /* Restore current thread's credentials. */
313 td->td_ucred = origcred;
314
315 mtx_lock(&nmp->nm_mtx);
316 /* Initialize other non-zero congestion variables. */
317 nfs_init_rtt(nmp);
318 mtx_unlock(&nmp->nm_mtx);
319 return (0);
320 }
321
322 /*
323 * NFS disconnect. Clean up and unlink.
324 */
325 void
326 nfs_disconnect(struct nfsmount *nmp)
327 {
328 CLIENT *client;
329
330 mtx_lock(&nmp->nm_mtx);
331 if (nmp->nm_client) {
332 client = nmp->nm_client;
333 nmp->nm_client = NULL;
334 mtx_unlock(&nmp->nm_mtx);
335 rpc_gss_secpurge_call(client);
336 CLNT_CLOSE(client);
337 CLNT_RELEASE(client);
338 } else
339 mtx_unlock(&nmp->nm_mtx);
340 }
341
342 void
343 nfs_safedisconnect(struct nfsmount *nmp)
344 {
345
346 nfs_disconnect(nmp);
347 }
348
349 static AUTH *
350 nfs_getauth(struct nfsmount *nmp, struct ucred *cred)
351 {
352 rpc_gss_service_t svc;
353 AUTH *auth;
354
355 switch (nmp->nm_secflavor) {
356 case RPCSEC_GSS_KRB5:
357 case RPCSEC_GSS_KRB5I:
358 case RPCSEC_GSS_KRB5P:
359 if (!nmp->nm_mech_oid)
360 if (!rpc_gss_mech_to_oid_call("kerberosv5",
361 &nmp->nm_mech_oid))
362 return (NULL);
363 if (nmp->nm_secflavor == RPCSEC_GSS_KRB5)
364 svc = rpc_gss_svc_none;
365 else if (nmp->nm_secflavor == RPCSEC_GSS_KRB5I)
366 svc = rpc_gss_svc_integrity;
367 else
368 svc = rpc_gss_svc_privacy;
369 auth = rpc_gss_secfind_call(nmp->nm_client, cred,
370 nmp->nm_principal, nmp->nm_mech_oid, svc);
371 if (auth)
372 return (auth);
373 /* fallthrough */
374 case AUTH_SYS:
375 default:
376 return (authunix_create(cred));
377
378 }
379 }
380
381 /*
382 * Callback from the RPC code to generate up/down notifications.
383 */
384
385 struct nfs_feedback_arg {
386 struct nfsmount *nf_mount;
387 int nf_lastmsg; /* last tprintf */
388 int nf_tprintfmsg;
389 struct thread *nf_td;
390 };
391
392 static void
393 nfs_feedback(int type, int proc, void *arg)
394 {
395 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
396 struct nfsmount *nmp = nf->nf_mount;
397 time_t now;
398
399 switch (type) {
400 case FEEDBACK_REXMIT2:
401 case FEEDBACK_RECONNECT:
402 now = time_uptime;
403 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
404 nfs_down(nmp, nf->nf_td,
405 "not responding", 0, NFSSTA_TIMEO);
406 nf->nf_tprintfmsg = TRUE;
407 nf->nf_lastmsg = now;
408 }
409 break;
410
411 case FEEDBACK_OK:
412 nfs_up(nf->nf_mount, nf->nf_td,
413 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
414 break;
415 }
416 }
417
418 /*
419 * nfs_request - goes something like this
420 * - fill in request struct
421 * - links it into list
422 * - calls nfs_send() for first transmit
423 * - calls nfs_receive() to get reply
424 * - break down rpc header and return with nfs reply pointed to
425 * by mrep or error
426 * nb: always frees up mreq mbuf list
427 */
428 int
429 nfs_request(struct vnode *vp, struct mbuf *mreq, int procnum,
430 struct thread *td, struct ucred *cred, struct mbuf **mrp,
431 struct mbuf **mdp, caddr_t *dposp)
432 {
433 struct mbuf *mrep;
434 u_int32_t *tl;
435 struct nfsmount *nmp;
436 struct mbuf *md;
437 time_t waituntil;
438 caddr_t dpos;
439 int error = 0, timeo;
440 AUTH *auth = NULL;
441 enum nfs_rto_timer_t timer;
442 struct nfs_feedback_arg nf;
443 struct rpc_callextra ext;
444 enum clnt_stat stat;
445 struct timeval timo;
446
447 /* Reject requests while attempting a forced unmount. */
448 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
449 m_freem(mreq);
450 return (ESTALE);
451 }
452 nmp = VFSTONFS(vp->v_mount);
453 bzero(&nf, sizeof(struct nfs_feedback_arg));
454 nf.nf_mount = nmp;
455 nf.nf_td = td;
456 nf.nf_lastmsg = time_uptime -
457 ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay));
458
459 /*
460 * XXX if not already connected call nfs_connect now. Longer
461 * term, change nfs_mount to call nfs_connect unconditionally
462 * and let clnt_reconnect_create handle reconnects.
463 */
464 if (!nmp->nm_client)
465 nfs_connect(nmp);
466
467 auth = nfs_getauth(nmp, cred);
468 if (!auth) {
469 m_freem(mreq);
470 return (EACCES);
471 }
472 bzero(&ext, sizeof(ext));
473 ext.rc_auth = auth;
474
475 ext.rc_feedback = nfs_feedback;
476 ext.rc_feedback_arg = &nf;
477
478 /*
479 * Use a conservative timeout for RPCs other than getattr,
480 * lookup, read or write. The justification for doing "other"
481 * this way is that these RPCs happen so infrequently that
482 * timer est. would probably be stale. Also, since many of
483 * these RPCs are non-idempotent, a conservative timeout is
484 * desired.
485 */
486 timer = nfs_rto_timer(procnum);
487 if (timer != NFS_DEFAULT_TIMER)
488 ext.rc_timers = &nmp->nm_timers[timer - 1];
489 else
490 ext.rc_timers = NULL;
491
492 #ifdef KDTRACE_HOOKS
493 if (dtrace_nfsclient_nfs23_start_probe != NULL) {
494 uint32_t probe_id;
495 int probe_procnum;
496
497 if (nmp->nm_flag & NFSMNT_NFSV3) {
498 probe_id = nfsclient_nfs3_start_probes[procnum];
499 probe_procnum = procnum;
500 } else {
501 probe_id = nfsclient_nfs2_start_probes[procnum];
502 probe_procnum = nfsv2_procid[procnum];
503 }
504 if (probe_id != 0)
505 (dtrace_nfsclient_nfs23_start_probe)(probe_id, vp,
506 mreq, cred, probe_procnum);
507 }
508 #endif
509
510 nfsstats.rpcrequests++;
511 tryagain:
512 /*
513 * This timeout specifies when a new socket should be created,
514 * along with new xid values. For UDP, this should be done
515 * infrequently, since retransmits of RPC requests should normally
516 * use the same xid.
517 */
518 if (nmp->nm_sotype == SOCK_DGRAM) {
519 if ((nmp->nm_flag & NFSMNT_SOFT) != 0) {
520 /*
521 * CLSET_RETRIES is set to 2, so this should be half
522 * of the total timeout required.
523 */
524 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
525 if (timeo < 1)
526 timeo = 1;
527 timo.tv_sec = timeo / NFS_HZ;
528 timo.tv_usec = (timeo % NFS_HZ) * 1000000 / NFS_HZ;
529 } else {
530 /* For UDP hard mounts, use a large value. */
531 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
532 timo.tv_usec = 0;
533 }
534 } else {
535 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
536 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
537 }
538 mrep = NULL;
539 stat = CLNT_CALL_MBUF(nmp->nm_client, &ext,
540 (nmp->nm_flag & NFSMNT_NFSV3) ? procnum : nfsv2_procid[procnum],
541 mreq, &mrep, timo);
542
543 /*
544 * If there was a successful reply and a tprintf msg.
545 * tprintf a response.
546 */
547 if (stat == RPC_SUCCESS)
548 error = 0;
549 else if (stat == RPC_TIMEDOUT) {
550 nfsstats.rpctimeouts++;
551 error = ETIMEDOUT;
552 } else if (stat == RPC_VERSMISMATCH) {
553 nfsstats.rpcinvalid++;
554 error = EOPNOTSUPP;
555 } else if (stat == RPC_PROGVERSMISMATCH) {
556 nfsstats.rpcinvalid++;
557 error = EPROTONOSUPPORT;
558 } else if (stat == RPC_INTR) {
559 error = EINTR;
560 } else {
561 nfsstats.rpcinvalid++;
562 error = EACCES;
563 }
564 if (error)
565 goto nfsmout;
566
567 KASSERT(mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
568
569 /*
570 * Search for any mbufs that are not a multiple of 4 bytes long
571 * or with m_data not longword aligned.
572 * These could cause pointer alignment problems, so copy them to
573 * well aligned mbufs.
574 */
575 error = nfs_realign(&mrep, M_NOWAIT);
576 if (error == ENOMEM) {
577 m_freem(mrep);
578 AUTH_DESTROY(auth);
579 nfsstats.rpcinvalid++;
580 return (error);
581 }
582
583 md = mrep;
584 dpos = mtod(mrep, caddr_t);
585 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
586 if (*tl != 0) {
587 error = fxdr_unsigned(int, *tl);
588 if ((nmp->nm_flag & NFSMNT_NFSV3) &&
589 error == NFSERR_TRYLATER) {
590 m_freem(mrep);
591 error = 0;
592 waituntil = time_second + nfs3_jukebox_delay;
593 while (time_second < waituntil)
594 (void)tsleep(&fake_wchan, PSOCK, "nqnfstry",
595 hz);
596 goto tryagain;
597 }
598 /*
599 * Make sure NFSERR_RETERR isn't bogusly set by a server
600 * such as amd. (No actual NFS error has bit 31 set.)
601 */
602 error &= ~NFSERR_RETERR;
603
604 /*
605 * If the File Handle was stale, invalidate the lookup
606 * cache, just in case.
607 */
608 if (error == ESTALE)
609 nfs_purgecache(vp);
610 /*
611 * Skip wcc data on non-ENOENT NFS errors for now.
612 * NetApp filers return corrupt postop attrs in the
613 * wcc data for NFS err EROFS. Not sure if they could
614 * return corrupt postop attrs for others errors.
615 * Blocking ENOENT post-op attributes breaks negative
616 * name caching, so always allow it through.
617 */
618 if ((nmp->nm_flag & NFSMNT_NFSV3) &&
619 (!nfs_skip_wcc_data_onerr || error == ENOENT)) {
620 *mrp = mrep;
621 *mdp = md;
622 *dposp = dpos;
623 error |= NFSERR_RETERR;
624 } else
625 m_freem(mrep);
626 goto nfsmout;
627 }
628
629 #ifdef KDTRACE_HOOKS
630 if (dtrace_nfsclient_nfs23_done_probe != NULL) {
631 uint32_t probe_id;
632 int probe_procnum;
633
634 if (nmp->nm_flag & NFSMNT_NFSV3) {
635 probe_id = nfsclient_nfs3_done_probes[procnum];
636 probe_procnum = procnum;
637 } else {
638 probe_id = nfsclient_nfs2_done_probes[procnum];
639 probe_procnum = (nmp->nm_flag & NFSMNT_NFSV3) ?
640 procnum : nfsv2_procid[procnum];
641 }
642 if (probe_id != 0)
643 (dtrace_nfsclient_nfs23_done_probe)(probe_id, vp,
644 mreq, cred, probe_procnum, 0);
645 }
646 #endif
647 m_freem(mreq);
648 *mrp = mrep;
649 *mdp = md;
650 *dposp = dpos;
651 AUTH_DESTROY(auth);
652 return (0);
653
654 nfsmout:
655 #ifdef KDTRACE_HOOKS
656 if (dtrace_nfsclient_nfs23_done_probe != NULL) {
657 uint32_t probe_id;
658 int probe_procnum;
659
660 if (nmp->nm_flag & NFSMNT_NFSV3) {
661 probe_id = nfsclient_nfs3_done_probes[procnum];
662 probe_procnum = procnum;
663 } else {
664 probe_id = nfsclient_nfs2_done_probes[procnum];
665 probe_procnum = (nmp->nm_flag & NFSMNT_NFSV3) ?
666 procnum : nfsv2_procid[procnum];
667 }
668 if (probe_id != 0)
669 (dtrace_nfsclient_nfs23_done_probe)(probe_id, vp,
670 mreq, cred, probe_procnum, error);
671 }
672 #endif
673 m_freem(mreq);
674 if (auth)
675 AUTH_DESTROY(auth);
676 return (error);
677 }
678
679 /*
680 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
681 * wait for all requests to complete. This is used by forced unmounts
682 * to terminate any outstanding RPCs.
683 */
684 int
685 nfs_nmcancelreqs(struct nfsmount *nmp)
686 {
687
688 if (nmp->nm_client)
689 CLNT_CLOSE(nmp->nm_client);
690 return (0);
691 }
692
693 /*
694 * Any signal that can interrupt an NFS operation in an intr mount
695 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
696 */
697 int nfs_sig_set[] = {
698 SIGINT,
699 SIGTERM,
700 SIGHUP,
701 SIGKILL,
702 SIGQUIT
703 };
704
705 /*
706 * Check to see if one of the signals in our subset is pending on
707 * the process (in an intr mount).
708 */
709 static int
710 nfs_sig_pending(sigset_t set)
711 {
712 int i;
713
714 for (i = 0 ; i < sizeof(nfs_sig_set)/sizeof(int) ; i++)
715 if (SIGISMEMBER(set, nfs_sig_set[i]))
716 return (1);
717 return (0);
718 }
719
720 /*
721 * The set/restore sigmask functions are used to (temporarily) overwrite
722 * the thread td_sigmask during an RPC call (for example). These are also
723 * used in other places in the NFS client that might tsleep().
724 */
725 void
726 nfs_set_sigmask(struct thread *td, sigset_t *oldset)
727 {
728 sigset_t newset;
729 int i;
730 struct proc *p;
731
732 SIGFILLSET(newset);
733 if (td == NULL)
734 td = curthread; /* XXX */
735 p = td->td_proc;
736 /* Remove the NFS set of signals from newset. */
737 PROC_LOCK(p);
738 mtx_lock(&p->p_sigacts->ps_mtx);
739 for (i = 0 ; i < sizeof(nfs_sig_set)/sizeof(int) ; i++) {
740 /*
741 * But make sure we leave the ones already masked
742 * by the process, i.e. remove the signal from the
743 * temporary signalmask only if it wasn't already
744 * in p_sigmask.
745 */
746 if (!SIGISMEMBER(td->td_sigmask, nfs_sig_set[i]) &&
747 !SIGISMEMBER(p->p_sigacts->ps_sigignore, nfs_sig_set[i]))
748 SIGDELSET(newset, nfs_sig_set[i]);
749 }
750 mtx_unlock(&p->p_sigacts->ps_mtx);
751 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
752 SIGPROCMASK_PROC_LOCKED);
753 PROC_UNLOCK(p);
754 }
755
756 void
757 nfs_restore_sigmask(struct thread *td, sigset_t *set)
758 {
759 if (td == NULL)
760 td = curthread; /* XXX */
761 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
762 }
763
764 /*
765 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
766 * old one after msleep() returns.
767 */
768 int
769 nfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority,
770 char *wmesg, int timo)
771 {
772 sigset_t oldset;
773 int error;
774 struct proc *p;
775
776 if ((priority & PCATCH) == 0)
777 return msleep(ident, mtx, priority, wmesg, timo);
778 if (td == NULL)
779 td = curthread; /* XXX */
780 nfs_set_sigmask(td, &oldset);
781 error = msleep(ident, mtx, priority, wmesg, timo);
782 nfs_restore_sigmask(td, &oldset);
783 p = td->td_proc;
784 return (error);
785 }
786
787 /*
788 * Test for a termination condition pending on the process.
789 * This is used for NFSMNT_INT mounts.
790 */
791 int
792 nfs_sigintr(struct nfsmount *nmp, struct thread *td)
793 {
794 struct proc *p;
795 sigset_t tmpset;
796
797 /* Terminate all requests while attempting a forced unmount. */
798 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
799 return (EIO);
800 if (!(nmp->nm_flag & NFSMNT_INT))
801 return (0);
802 if (td == NULL)
803 return (0);
804 p = td->td_proc;
805 PROC_LOCK(p);
806 tmpset = p->p_siglist;
807 SIGSETOR(tmpset, td->td_siglist);
808 SIGSETNAND(tmpset, td->td_sigmask);
809 mtx_lock(&p->p_sigacts->ps_mtx);
810 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
811 mtx_unlock(&p->p_sigacts->ps_mtx);
812 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
813 && nfs_sig_pending(tmpset)) {
814 PROC_UNLOCK(p);
815 return (EINTR);
816 }
817 PROC_UNLOCK(p);
818 return (0);
819 }
820
821 static int
822 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
823 {
824 struct proc *p;
825
826 p = td ? td->td_proc : NULL;
827 if (error)
828 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n", server,
829 msg, error);
830 else
831 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
832 return (0);
833 }
834
835 static void
836 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
837 int error, int flags)
838 {
839 if (nmp == NULL)
840 return;
841 mtx_lock(&nmp->nm_mtx);
842 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
843 nmp->nm_state |= NFSSTA_TIMEO;
844 mtx_unlock(&nmp->nm_mtx);
845 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
846 VQ_NOTRESP, 0);
847 } else
848 mtx_unlock(&nmp->nm_mtx);
849 mtx_lock(&nmp->nm_mtx);
850 if ((flags & NFSSTA_LOCKTIMEO) &&
851 !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
852 nmp->nm_state |= NFSSTA_LOCKTIMEO;
853 mtx_unlock(&nmp->nm_mtx);
854 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
855 VQ_NOTRESPLOCK, 0);
856 } else
857 mtx_unlock(&nmp->nm_mtx);
858 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
859 }
860
861 static void
862 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
863 int flags, int tprintfmsg)
864 {
865 if (nmp == NULL)
866 return;
867 if (tprintfmsg)
868 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
869
870 mtx_lock(&nmp->nm_mtx);
871 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
872 nmp->nm_state &= ~NFSSTA_TIMEO;
873 mtx_unlock(&nmp->nm_mtx);
874 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
875 VQ_NOTRESP, 1);
876 } else
877 mtx_unlock(&nmp->nm_mtx);
878
879 mtx_lock(&nmp->nm_mtx);
880 if ((flags & NFSSTA_LOCKTIMEO) &&
881 (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
882 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
883 mtx_unlock(&nmp->nm_mtx);
884 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
885 VQ_NOTRESPLOCK, 1);
886 } else
887 mtx_unlock(&nmp->nm_mtx);
888 }
Cache object: 9dad5aa0164db414442b69e0247a83aa
|