1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1989, 1991, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 */
35
36 #include <sys/cdefs.h>
37 /*
38 * Socket operations for use by nfs
39 */
40
41 #include "opt_kgssapi.h"
42 #include "opt_nfs.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/limits.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/signalvar.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysctl.h>
57 #include <sys/syslog.h>
58 #include <sys/vnode.h>
59
60 #include <rpc/rpc.h>
61 #include <rpc/krpc.h>
62
63 #include <kgssapi/krb5/kcrypto.h>
64
65 #include <fs/nfs/nfsport.h>
66
67 #ifdef KDTRACE_HOOKS
68 #include <sys/dtrace_bsd.h>
69
70 dtrace_nfsclient_nfs23_start_probe_func_t
71 dtrace_nfscl_nfs234_start_probe;
72
73 dtrace_nfsclient_nfs23_done_probe_func_t
74 dtrace_nfscl_nfs234_done_probe;
75
76 /*
77 * Registered probes by RPC type.
78 */
79 uint32_t nfscl_nfs2_start_probes[NFSV41_NPROCS + 1];
80 uint32_t nfscl_nfs2_done_probes[NFSV41_NPROCS + 1];
81
82 uint32_t nfscl_nfs3_start_probes[NFSV41_NPROCS + 1];
83 uint32_t nfscl_nfs3_done_probes[NFSV41_NPROCS + 1];
84
85 uint32_t nfscl_nfs4_start_probes[NFSV41_NPROCS + 1];
86 uint32_t nfscl_nfs4_done_probes[NFSV41_NPROCS + 1];
87 #endif
88
89 NFSSTATESPINLOCK;
90 NFSREQSPINLOCK;
91 NFSDLOCKMUTEX;
92 NFSCLSTATEMUTEX;
93 extern struct nfsstatsv1 nfsstatsv1;
94 extern struct nfsreqhead nfsd_reqq;
95 extern int nfscl_ticks;
96 extern void (*ncl_call_invalcaches)(struct vnode *);
97 extern int nfs_numnfscbd;
98 extern int nfscl_debuglevel;
99 extern int nfsrv_lease;
100
101 SVCPOOL *nfscbd_pool;
102 int nfs_bufpackets = 4;
103 static int nfsrv_gsscallbackson = 0;
104 static int nfs_reconnects;
105 static int nfs3_jukebox_delay = 10;
106 static int nfs_skip_wcc_data_onerr = 1;
107 static int nfs_dsretries = 2;
108 static struct timespec nfs_trylater_max = {
109 .tv_sec = NFS_TRYLATERDEL,
110 .tv_nsec = 0,
111 };
112
113 SYSCTL_DECL(_vfs_nfs);
114
115 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
116 "Buffer reservation size 2 < x < 64");
117 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
118 "Number of times the nfs client has had to reconnect");
119 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
120 "Number of seconds to delay a retry after receiving EJUKEBOX");
121 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
122 "Disable weak cache consistency checking when server returns an error");
123 SYSCTL_INT(_vfs_nfs, OID_AUTO, dsretries, CTLFLAG_RW, &nfs_dsretries, 0,
124 "Number of retries for a DS RPC before failure");
125
126 static void nfs_down(struct nfsmount *, struct thread *, const char *,
127 int, int);
128 static void nfs_up(struct nfsmount *, struct thread *, const char *,
129 int, int);
130 static int nfs_msg(struct thread *, const char *, const char *, int);
131
132 struct nfs_cached_auth {
133 int ca_refs; /* refcount, including 1 from the cache */
134 uid_t ca_uid; /* uid that corresponds to this auth */
135 AUTH *ca_auth; /* RPC auth handle */
136 };
137
138 static int nfsv2_procid[NFS_V3NPROCS] = {
139 NFSV2PROC_NULL,
140 NFSV2PROC_GETATTR,
141 NFSV2PROC_SETATTR,
142 NFSV2PROC_LOOKUP,
143 NFSV2PROC_NOOP,
144 NFSV2PROC_READLINK,
145 NFSV2PROC_READ,
146 NFSV2PROC_WRITE,
147 NFSV2PROC_CREATE,
148 NFSV2PROC_MKDIR,
149 NFSV2PROC_SYMLINK,
150 NFSV2PROC_CREATE,
151 NFSV2PROC_REMOVE,
152 NFSV2PROC_RMDIR,
153 NFSV2PROC_RENAME,
154 NFSV2PROC_LINK,
155 NFSV2PROC_READDIR,
156 NFSV2PROC_NOOP,
157 NFSV2PROC_STATFS,
158 NFSV2PROC_NOOP,
159 NFSV2PROC_NOOP,
160 NFSV2PROC_NOOP,
161 };
162
163 /*
164 * This static array indicates that a NFSv4 RPC should use
165 * RPCSEC_GSS, if the mount indicates that via sec=krb5[ip].
166 * System RPCs that do not use file handles will be false
167 * in this array so that they will use AUTH_SYS when the
168 * "syskrb5" mount option is specified, along with
169 * "sec=krb5[ip]".
170 */
171 static bool nfscl_use_gss[NFSV42_NPROCS] = {
172 true,
173 true,
174 true,
175 true,
176 true,
177 true,
178 true,
179 true,
180 true,
181 true,
182 true,
183 true,
184 true,
185 true,
186 true,
187 true,
188 true,
189 true,
190 true,
191 true,
192 true,
193 true,
194 true,
195 false, /* SetClientID */
196 false, /* SetClientIDConfirm */
197 true,
198 true,
199 true,
200 true,
201 true,
202 true,
203 true,
204 false, /* Renew */
205 true,
206 false, /* ReleaseLockOwn */
207 true,
208 true,
209 true,
210 true,
211 true,
212 true,
213 false, /* ExchangeID */
214 false, /* CreateSession */
215 false, /* DestroySession */
216 false, /* DestroyClientID */
217 false, /* FreeStateID */
218 true,
219 true,
220 true,
221 true,
222 false, /* ReclaimComplete */
223 true,
224 true,
225 true,
226 true,
227 true,
228 true,
229 true,
230 true,
231 true,
232 true,
233 true,
234 true,
235 true,
236 true,
237 false, /* BindConnectionToSession */
238 true,
239 true,
240 true,
241 true,
242 };
243
244 /*
245 * Initialize sockets and congestion for a new NFS connection.
246 * We do not free the sockaddr if error.
247 * Which arguments are set to NULL indicate what kind of call it is.
248 * cred == NULL --> a call to connect to a pNFS DS
249 * nmp == NULL --> indicates an upcall to userland or a NFSv4.0 callback
250 */
251 int
newnfs_connect(struct nfsmount * nmp,struct nfssockreq * nrp,struct ucred * cred,NFSPROC_T * p,int callback_retry_mult,bool dotls,struct __rpc_client ** clipp)252 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
253 struct ucred *cred, NFSPROC_T *p, int callback_retry_mult, bool dotls,
254 struct __rpc_client **clipp)
255 {
256 int rcvreserve, sndreserve;
257 int pktscale, pktscalesav;
258 struct sockaddr *saddr;
259 struct ucred *origcred;
260 CLIENT *client;
261 struct netconfig *nconf;
262 struct socket *so;
263 int one = 1, retries, error = 0;
264 struct thread *td = curthread;
265 SVCXPRT *xprt;
266 struct timeval timo;
267 uint64_t tval;
268
269 /*
270 * We need to establish the socket using the credentials of
271 * the mountpoint. Some parts of this process (such as
272 * sobind() and soconnect()) will use the curent thread's
273 * credential instead of the socket credential. To work
274 * around this, temporarily change the current thread's
275 * credential to that of the mountpoint.
276 *
277 * XXX: It would be better to explicitly pass the correct
278 * credential to sobind() and soconnect().
279 */
280 origcred = td->td_ucred;
281
282 /*
283 * Use the credential in nr_cred, if not NULL.
284 */
285 if (nrp->nr_cred != NULL)
286 td->td_ucred = nrp->nr_cred;
287 else
288 td->td_ucred = cred;
289 saddr = nrp->nr_nam;
290
291 if (saddr->sa_family == AF_INET)
292 if (nrp->nr_sotype == SOCK_DGRAM)
293 nconf = getnetconfigent("udp");
294 else
295 nconf = getnetconfigent("tcp");
296 else
297 if (nrp->nr_sotype == SOCK_DGRAM)
298 nconf = getnetconfigent("udp6");
299 else
300 nconf = getnetconfigent("tcp6");
301
302 pktscale = nfs_bufpackets;
303 if (pktscale < 2)
304 pktscale = 2;
305 if (pktscale > 64)
306 pktscale = 64;
307 pktscalesav = pktscale;
308 /*
309 * soreserve() can fail if sb_max is too small, so shrink pktscale
310 * and try again if there is an error.
311 * Print a log message suggesting increasing sb_max.
312 * Creating a socket and doing this is necessary since, if the
313 * reservation sizes are too large and will make soreserve() fail,
314 * the connection will work until a large send is attempted and
315 * then it will loop in the krpc code.
316 */
317 so = NULL;
318 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
319 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
320 nrp->nr_soproto, td->td_ucred, td);
321 if (error != 0)
322 goto out;
323 do {
324 if (error != 0 && pktscale > 2) {
325 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
326 pktscale == pktscalesav) {
327 /*
328 * Suggest vfs.nfs.bufpackets * maximum RPC message,
329 * adjusted for the sb_max->sb_max_adj conversion of
330 * MCLBYTES / (MSIZE + MCLBYTES) as the minimum setting
331 * for kern.ipc.maxsockbuf.
332 */
333 tval = (NFS_MAXBSIZE + NFS_MAXXDR) * nfs_bufpackets;
334 tval *= MSIZE + MCLBYTES;
335 tval += MCLBYTES - 1; /* Round up divide by MCLBYTES. */
336 tval /= MCLBYTES;
337 printf("Consider increasing kern.ipc.maxsockbuf to a "
338 "minimum of %ju to support %ubyte NFS I/O\n",
339 (uintmax_t)tval, NFS_MAXBSIZE);
340 }
341 pktscale--;
342 }
343 if (nrp->nr_sotype == SOCK_DGRAM) {
344 if (nmp != NULL) {
345 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
346 pktscale;
347 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
348 pktscale;
349 } else {
350 sndreserve = rcvreserve = 1024 * pktscale;
351 }
352 } else {
353 if (nrp->nr_sotype != SOCK_STREAM)
354 panic("nfscon sotype");
355 if (nmp != NULL) {
356 sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR) *
357 pktscale;
358 rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR) *
359 pktscale;
360 } else {
361 sndreserve = rcvreserve = 1024 * pktscale;
362 }
363 }
364 error = soreserve(so, sndreserve, rcvreserve);
365 if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
366 pktscale <= 2)
367 printf("Must increase kern.ipc.maxsockbuf or reduce"
368 " rsize, wsize\n");
369 } while (error != 0 && pktscale > 2);
370 soclose(so);
371 if (error != 0)
372 goto out;
373
374 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
375 nrp->nr_vers, sndreserve, rcvreserve);
376 CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
377 if (nmp != NULL) {
378 if ((nmp->nm_flag & NFSMNT_INT))
379 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
380 if ((nmp->nm_flag & NFSMNT_RESVPORT))
381 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
382 if (NFSHASTLS(nmp)) {
383 CLNT_CONTROL(client, CLSET_TLS, &one);
384 if (nmp->nm_tlscertname != NULL)
385 CLNT_CONTROL(client, CLSET_TLSCERTNAME,
386 nmp->nm_tlscertname);
387 }
388 if (NFSHASSOFT(nmp)) {
389 if (nmp->nm_sotype == SOCK_DGRAM)
390 /*
391 * For UDP, the large timeout for a reconnect
392 * will be set to "nm_retry * nm_timeo / 2", so
393 * we only want to do 2 reconnect timeout
394 * retries.
395 */
396 retries = 2;
397 else
398 retries = nmp->nm_retry;
399 } else
400 retries = INT_MAX;
401 if (NFSHASNFSV4N(nmp)) {
402 if (cred != NULL) {
403 if (NFSHASSOFT(nmp)) {
404 /*
405 * This should be a DS mount.
406 * Use CLSET_TIMEOUT to set the timeout
407 * for connections to DSs instead of
408 * specifying a timeout on each RPC.
409 * This is done so that SO_SNDTIMEO
410 * is set on the TCP socket as well
411 * as specifying a time limit when
412 * waiting for an RPC reply. Useful
413 * if the send queue for the TCP
414 * connection has become constipated,
415 * due to a failed DS.
416 * The choice of lease_duration / 4 is
417 * fairly arbitrary, but seems to work
418 * ok, with a lower bound of 10sec.
419 */
420 timo.tv_sec = nfsrv_lease / 4;
421 if (timo.tv_sec < 10)
422 timo.tv_sec = 10;
423 timo.tv_usec = 0;
424 CLNT_CONTROL(client, CLSET_TIMEOUT,
425 &timo);
426 }
427 /*
428 * Make sure the nfscbd_pool doesn't get
429 * destroyed while doing this.
430 */
431 NFSD_LOCK();
432 if (nfs_numnfscbd > 0) {
433 nfs_numnfscbd++;
434 NFSD_UNLOCK();
435 xprt = svc_vc_create_backchannel(
436 nfscbd_pool);
437 CLNT_CONTROL(client, CLSET_BACKCHANNEL,
438 xprt);
439 NFSD_LOCK();
440 nfs_numnfscbd--;
441 if (nfs_numnfscbd == 0)
442 wakeup(&nfs_numnfscbd);
443 }
444 NFSD_UNLOCK();
445 } else {
446 /*
447 * cred == NULL for a DS connect.
448 * For connects to a DS, set a retry limit
449 * so that failed DSs will be detected.
450 * This is ok for NFSv4.1, since a DS does
451 * not maintain open/lock state and is the
452 * only case where using a "soft" mount is
453 * recommended for NFSv4.
454 * For mounts from the MDS to DS, this is done
455 * via mount options, but that is not the case
456 * here. The retry limit here can be adjusted
457 * via the sysctl vfs.nfs.dsretries.
458 * See the comment above w.r.t. timeout.
459 */
460 timo.tv_sec = nfsrv_lease / 4;
461 if (timo.tv_sec < 10)
462 timo.tv_sec = 10;
463 timo.tv_usec = 0;
464 CLNT_CONTROL(client, CLSET_TIMEOUT, &timo);
465 retries = nfs_dsretries;
466 }
467 }
468 } else {
469 /*
470 * Three cases:
471 * - Null RPC callback to client
472 * - Non-Null RPC callback to client, wait a little longer
473 * - upcalls to nfsuserd and gssd (clp == NULL)
474 */
475 if (callback_retry_mult == 0) {
476 retries = NFSV4_UPCALLRETRY;
477 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
478 } else {
479 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
480 }
481 if (dotls)
482 CLNT_CONTROL(client, CLSET_TLS, &one);
483 }
484 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
485
486 if (nmp != NULL) {
487 /*
488 * For UDP, there are 2 timeouts:
489 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
490 * that does a retransmit of an RPC request using the same
491 * socket and xid. This is what you normally want to do,
492 * since NFS servers depend on "same xid" for their
493 * Duplicate Request Cache.
494 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
495 * retransmits on the same socket should fail and a fresh
496 * socket created. Each of these timeouts counts as one
497 * CLSET_RETRIES as set above.
498 * Set the initial retransmit timeout for UDP. This timeout
499 * doesn't exist for TCP and the following call just fails,
500 * which is ok.
501 */
502 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
503 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
504 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
505 }
506
507 /*
508 * *clipp is &nrp->nr_client or &nm_aconn[nmp->nm_nextaconn].
509 * The latter case is for additional connections specified by the
510 * "nconnect" mount option. nr_mtx etc is used for these additional
511 * connections, as well as nr_client in the nfssockreq
512 * structure for the mount.
513 */
514 mtx_lock(&nrp->nr_mtx);
515 if (*clipp != NULL) {
516 mtx_unlock(&nrp->nr_mtx);
517 /*
518 * Someone else already connected.
519 */
520 CLNT_RELEASE(client);
521 } else {
522 *clipp = client;
523 /*
524 * Protocols that do not require connections may be optionally
525 * left unconnected for servers that reply from a port other
526 * than NFS_PORT.
527 */
528 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
529 mtx_unlock(&nrp->nr_mtx);
530 CLNT_CONTROL(client, CLSET_CONNECT, &one);
531 } else
532 mtx_unlock(&nrp->nr_mtx);
533 }
534
535 out:
536 /* Restore current thread's credentials. */
537 td->td_ucred = origcred;
538
539 NFSEXITCODE(error);
540 return (error);
541 }
542
543 /*
544 * NFS disconnect. Clean up and unlink.
545 */
546 void
newnfs_disconnect(struct nfsmount * nmp,struct nfssockreq * nrp)547 newnfs_disconnect(struct nfsmount *nmp, struct nfssockreq *nrp)
548 {
549 CLIENT *client, *aconn[NFS_MAXNCONN - 1];
550 int i;
551
552 mtx_lock(&nrp->nr_mtx);
553 if (nrp->nr_client != NULL) {
554 client = nrp->nr_client;
555 nrp->nr_client = NULL;
556 if (nmp != NULL && nmp->nm_aconnect > 0) {
557 for (i = 0; i < nmp->nm_aconnect; i++) {
558 aconn[i] = nmp->nm_aconn[i];
559 nmp->nm_aconn[i] = NULL;
560 }
561 }
562 mtx_unlock(&nrp->nr_mtx);
563 rpc_gss_secpurge_call(client);
564 CLNT_CLOSE(client);
565 CLNT_RELEASE(client);
566 if (nmp != NULL && nmp->nm_aconnect > 0) {
567 for (i = 0; i < nmp->nm_aconnect; i++) {
568 if (aconn[i] != NULL) {
569 rpc_gss_secpurge_call(aconn[i]);
570 CLNT_CLOSE(aconn[i]);
571 CLNT_RELEASE(aconn[i]);
572 }
573 }
574 }
575 } else {
576 mtx_unlock(&nrp->nr_mtx);
577 }
578 }
579
580 static AUTH *
nfs_getauth(struct nfssockreq * nrp,int secflavour,char * clnt_principal,char * srv_principal,gss_OID mech_oid,struct ucred * cred)581 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
582 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
583 {
584 rpc_gss_service_t svc;
585 AUTH *auth;
586
587 switch (secflavour) {
588 case RPCSEC_GSS_KRB5:
589 case RPCSEC_GSS_KRB5I:
590 case RPCSEC_GSS_KRB5P:
591 if (!mech_oid) {
592 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
593 return (NULL);
594 }
595 if (secflavour == RPCSEC_GSS_KRB5)
596 svc = rpc_gss_svc_none;
597 else if (secflavour == RPCSEC_GSS_KRB5I)
598 svc = rpc_gss_svc_integrity;
599 else
600 svc = rpc_gss_svc_privacy;
601
602 if (clnt_principal == NULL) {
603 NFSCL_DEBUG(1, "nfs_getauth: clnt princ=NULL, "
604 "srv princ=%s\n", srv_principal);
605 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
606 srv_principal, mech_oid, svc);
607 } else {
608 NFSCL_DEBUG(1, "nfs_getauth: clnt princ=%s "
609 "srv princ=%s\n", clnt_principal, srv_principal);
610 auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
611 clnt_principal, srv_principal, "kerberosv5",
612 svc, NULL, NULL, NULL);
613 return (auth);
614 }
615 if (auth != NULL)
616 return (auth);
617 /* fallthrough */
618 case AUTH_SYS:
619 default:
620 return (authunix_create(cred));
621 }
622 }
623
624 /*
625 * Callback from the RPC code to generate up/down notifications.
626 */
627
628 struct nfs_feedback_arg {
629 struct nfsmount *nf_mount;
630 int nf_lastmsg; /* last tprintf */
631 int nf_tprintfmsg;
632 struct thread *nf_td;
633 };
634
635 static void
nfs_feedback(int type,int proc,void * arg)636 nfs_feedback(int type, int proc, void *arg)
637 {
638 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
639 struct nfsmount *nmp = nf->nf_mount;
640 time_t now;
641
642 switch (type) {
643 case FEEDBACK_REXMIT2:
644 case FEEDBACK_RECONNECT:
645 now = NFSD_MONOSEC;
646 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
647 nfs_down(nmp, nf->nf_td,
648 "not responding", 0, NFSSTA_TIMEO);
649 nf->nf_tprintfmsg = TRUE;
650 nf->nf_lastmsg = now;
651 }
652 break;
653
654 case FEEDBACK_OK:
655 nfs_up(nf->nf_mount, nf->nf_td,
656 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
657 break;
658 }
659 }
660
661 /*
662 * newnfs_request - goes something like this
663 * - does the rpc by calling the krpc layer
664 * - break down rpc header and return with nfs reply
665 * nb: always frees up nd_mreq mbuf list
666 */
667 int
newnfs_request(struct nfsrv_descript * nd,struct nfsmount * nmp,struct nfsclient * clp,struct nfssockreq * nrp,vnode_t vp,struct thread * td,struct ucred * cred,u_int32_t prog,u_int32_t vers,u_char * retsum,int toplevel,u_int64_t * xidp,struct nfsclsession * dssep)668 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
669 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
670 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
671 u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
672 {
673 uint32_t retseq, retval, slotseq, *tl;
674 int i = 0, j = 0, opcnt, set_sigset = 0, slot;
675 int error = 0, usegssname = 0, secflavour = AUTH_SYS;
676 int freeslot, maxslot, reterr, slotpos, timeo;
677 u_int16_t procnum;
678 u_int nextconn;
679 struct nfs_feedback_arg nf;
680 struct timeval timo;
681 AUTH *auth;
682 struct rpc_callextra ext;
683 enum clnt_stat stat;
684 struct nfsreq *rep = NULL;
685 char *srv_principal = NULL, *clnt_principal = NULL;
686 sigset_t oldset;
687 struct ucred *authcred;
688 struct nfsclsession *sep;
689 uint8_t sessionid[NFSX_V4SESSIONID];
690 bool nextconn_set;
691 struct timespec trylater_delay, ts, waituntil;
692
693 /* Initially 1msec. */
694 trylater_delay.tv_sec = 0;
695 trylater_delay.tv_nsec = 1000000;
696 sep = dssep;
697 if (xidp != NULL)
698 *xidp = 0;
699 /* Reject requests while attempting a forced unmount. */
700 if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
701 m_freem(nd->nd_mreq);
702 return (ESTALE);
703 }
704
705 /*
706 * Set authcred, which is used to acquire RPC credentials to
707 * the cred argument, by default. The crhold() should not be
708 * necessary, but will ensure that some future code change
709 * doesn't result in the credential being free'd prematurely.
710 */
711 authcred = crhold(cred);
712
713 /* For client side interruptible mounts, mask off the signals. */
714 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
715 newnfs_set_sigmask(td, &oldset);
716 set_sigset = 1;
717 }
718
719 /*
720 * If not already connected call newnfs_connect now.
721 */
722 if (nrp->nr_client == NULL)
723 newnfs_connect(nmp, nrp, cred, td, 0, false, &nrp->nr_client);
724
725 /*
726 * If the "nconnect" mount option was specified and this RPC is
727 * one that can have a large RPC message and is being done through
728 * the NFS/MDS server, use an additional connection. (When the RPC is
729 * being done through the server/MDS, nrp == &nmp->nm_sockreq.)
730 * The "nconnect" mount option normally has minimal effect when the
731 * "pnfs" mount option is specified, since only Readdir RPCs are
732 * normally done through the NFS/MDS server.
733 */
734 nextconn_set = false;
735 if (nmp != NULL && nmp->nm_aconnect > 0 && nrp == &nmp->nm_sockreq &&
736 (nd->nd_procnum == NFSPROC_READ ||
737 nd->nd_procnum == NFSPROC_READDIR ||
738 nd->nd_procnum == NFSPROC_READDIRPLUS ||
739 nd->nd_procnum == NFSPROC_WRITE)) {
740 nextconn = atomic_fetchadd_int(&nmp->nm_nextaconn, 1);
741 nextconn %= nmp->nm_aconnect;
742 nextconn_set = true;
743 if (nmp->nm_aconn[nextconn] == NULL)
744 newnfs_connect(nmp, nrp, cred, td, 0, false,
745 &nmp->nm_aconn[nextconn]);
746 }
747
748 /*
749 * For a client side mount, nmp is != NULL and clp == NULL. For
750 * server calls (callbacks or upcalls), nmp == NULL.
751 */
752 if (clp != NULL) {
753 NFSLOCKSTATE();
754 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
755 secflavour = RPCSEC_GSS_KRB5;
756 if (nd->nd_procnum != NFSPROC_NULL) {
757 if (clp->lc_flags & LCL_GSSINTEGRITY)
758 secflavour = RPCSEC_GSS_KRB5I;
759 else if (clp->lc_flags & LCL_GSSPRIVACY)
760 secflavour = RPCSEC_GSS_KRB5P;
761 }
762 }
763 NFSUNLOCKSTATE();
764 } else if (nmp != NULL && NFSHASKERB(nmp) &&
765 nd->nd_procnum != NFSPROC_NULL && (!NFSHASSYSKRB5(nmp) ||
766 nfscl_use_gss[nd->nd_procnum])) {
767 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
768 nd->nd_flag |= ND_USEGSSNAME;
769 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
770 /*
771 * If there is a client side host based credential,
772 * use that, otherwise use the system uid, if set.
773 * The system uid is in the nmp->nm_sockreq.nr_cred
774 * credentials.
775 */
776 if (nmp->nm_krbnamelen > 0) {
777 usegssname = 1;
778 clnt_principal = nmp->nm_krbname;
779 } else if (nmp->nm_uid != (uid_t)-1) {
780 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
781 ("newnfs_request: NULL nr_cred"));
782 crfree(authcred);
783 authcred = crhold(nmp->nm_sockreq.nr_cred);
784 }
785 } else if (nmp->nm_krbnamelen == 0 &&
786 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
787 /*
788 * If there is no host based principal name and
789 * the system uid is set and this is root, use the
790 * system uid, since root won't have user
791 * credentials in a credentials cache file.
792 * The system uid is in the nmp->nm_sockreq.nr_cred
793 * credentials.
794 */
795 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
796 ("newnfs_request: NULL nr_cred"));
797 crfree(authcred);
798 authcred = crhold(nmp->nm_sockreq.nr_cred);
799 }
800 if (NFSHASINTEGRITY(nmp))
801 secflavour = RPCSEC_GSS_KRB5I;
802 else if (NFSHASPRIVACY(nmp))
803 secflavour = RPCSEC_GSS_KRB5P;
804 else
805 secflavour = RPCSEC_GSS_KRB5;
806 if (nrp->nr_srvprinc[0] == '\0')
807 srv_principal = NFSMNT_SRVKRBNAME(nmp);
808 else
809 srv_principal = nrp->nr_srvprinc;
810 } else if (nmp != NULL && (!NFSHASKERB(nmp) || NFSHASSYSKRB5(nmp)) &&
811 nd->nd_procnum != NFSPROC_NULL &&
812 (nd->nd_flag & ND_USEGSSNAME) != 0) {
813 /*
814 * Use the uid that did the mount when the RPC is doing
815 * NFSv4 system operations, as indicated by the
816 * ND_USEGSSNAME flag, for the AUTH_SYS case.
817 * The credentials in nm_sockreq.nr_cred were used for the
818 * mount.
819 */
820 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
821 ("newnfs_request: NULL nr_cred"));
822 crfree(authcred);
823 authcred = crhold(nmp->nm_sockreq.nr_cred);
824 }
825
826 if (nmp != NULL) {
827 bzero(&nf, sizeof(struct nfs_feedback_arg));
828 nf.nf_mount = nmp;
829 nf.nf_td = td;
830 nf.nf_lastmsg = NFSD_MONOSEC -
831 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
832 }
833
834 if (nd->nd_procnum == NFSPROC_NULL)
835 auth = authnone_create();
836 else if (usegssname) {
837 /*
838 * For this case, the authenticator is held in the
839 * nfssockreq structure, so don't release the reference count
840 * held on it. --> Don't AUTH_DESTROY() it in this function.
841 */
842 if (nrp->nr_auth == NULL)
843 nrp->nr_auth = nfs_getauth(nrp, secflavour,
844 clnt_principal, srv_principal, NULL, authcred);
845 else
846 rpc_gss_refresh_auth_call(nrp->nr_auth);
847 auth = nrp->nr_auth;
848 } else
849 auth = nfs_getauth(nrp, secflavour, NULL,
850 srv_principal, NULL, authcred);
851 crfree(authcred);
852 if (auth == NULL) {
853 m_freem(nd->nd_mreq);
854 if (set_sigset)
855 newnfs_restore_sigmask(td, &oldset);
856 return (EACCES);
857 }
858 bzero(&ext, sizeof(ext));
859 ext.rc_auth = auth;
860 if (nmp != NULL) {
861 ext.rc_feedback = nfs_feedback;
862 ext.rc_feedback_arg = &nf;
863 }
864
865 procnum = nd->nd_procnum;
866 if ((nd->nd_flag & ND_NFSV4) &&
867 nd->nd_procnum != NFSPROC_NULL &&
868 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
869 procnum = NFSV4PROC_COMPOUND;
870
871 if (nmp != NULL) {
872 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
873
874 /* Map the procnum to the old NFSv2 one, as required. */
875 if ((nd->nd_flag & ND_NFSV2) != 0) {
876 if (nd->nd_procnum < NFS_V3NPROCS)
877 procnum = nfsv2_procid[nd->nd_procnum];
878 else
879 procnum = NFSV2PROC_NOOP;
880 }
881
882 /*
883 * Now only used for the R_DONTRECOVER case, but until that is
884 * supported within the krpc code, I need to keep a queue of
885 * outstanding RPCs for nfsv4 client requests.
886 */
887 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
888 rep = malloc(sizeof(struct nfsreq),
889 M_NFSDREQ, M_WAITOK);
890 #ifdef KDTRACE_HOOKS
891 if (dtrace_nfscl_nfs234_start_probe != NULL) {
892 uint32_t probe_id;
893 int probe_procnum;
894
895 if (nd->nd_flag & ND_NFSV4) {
896 probe_id =
897 nfscl_nfs4_start_probes[nd->nd_procnum];
898 probe_procnum = nd->nd_procnum;
899 } else if (nd->nd_flag & ND_NFSV3) {
900 probe_id = nfscl_nfs3_start_probes[procnum];
901 probe_procnum = procnum;
902 } else {
903 probe_id =
904 nfscl_nfs2_start_probes[nd->nd_procnum];
905 probe_procnum = procnum;
906 }
907 if (probe_id != 0)
908 (dtrace_nfscl_nfs234_start_probe)
909 (probe_id, vp, nd->nd_mreq, cred,
910 probe_procnum);
911 }
912 #endif
913 }
914 freeslot = -1; /* Set to slot that needs to be free'd */
915 tryagain:
916 slot = -1; /* Slot that needs a sequence# increment. */
917 /*
918 * This timeout specifies when a new socket should be created,
919 * along with new xid values. For UDP, this should be done
920 * infrequently, since retransmits of RPC requests should normally
921 * use the same xid.
922 */
923 if (nmp == NULL) {
924 if (clp == NULL) {
925 timo.tv_sec = NFSV4_UPCALLTIMEO;
926 timo.tv_usec = 0;
927 } else {
928 timo.tv_sec = NFSV4_CALLBACKTIMEO / 1000;
929 timo.tv_usec = NFSV4_CALLBACKTIMEO * 1000;
930 }
931 } else {
932 if (nrp->nr_sotype != SOCK_DGRAM) {
933 timo.tv_usec = 0;
934 if ((nmp->nm_flag & NFSMNT_NFSV4))
935 timo.tv_sec = INT_MAX;
936 else
937 timo.tv_sec = NFS_TCPTIMEO;
938 } else {
939 if (NFSHASSOFT(nmp)) {
940 /*
941 * CLSET_RETRIES is set to 2, so this should be
942 * half of the total timeout required.
943 */
944 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
945 if (timeo < 1)
946 timeo = 1;
947 timo.tv_sec = timeo / NFS_HZ;
948 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
949 NFS_HZ;
950 } else {
951 /* For UDP hard mounts, use a large value. */
952 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
953 timo.tv_usec = 0;
954 }
955 }
956
957 if (rep != NULL) {
958 rep->r_flags = 0;
959 rep->r_nmp = nmp;
960 /*
961 * Chain request into list of outstanding requests.
962 */
963 NFSLOCKREQ();
964 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
965 NFSUNLOCKREQ();
966 }
967 }
968
969 nd->nd_mrep = NULL;
970 if (clp != NULL && sep != NULL)
971 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
972 nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
973 else if (nextconn_set)
974 /*
975 * When there are multiple TCP connections, send the
976 * RPCs with large messages on the alternate TCP
977 * connection(s) in a round robin fashion.
978 * The small RPC messages are sent on the default
979 * TCP connection because they do not require much
980 * network bandwidth and separating them from the
981 * large RPC messages avoids them getting "log jammed"
982 * behind several large RPC messages.
983 */
984 stat = CLNT_CALL_MBUF(nmp->nm_aconn[nextconn],
985 &ext, procnum, nd->nd_mreq, &nd->nd_mrep, timo);
986 else
987 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
988 nd->nd_mreq, &nd->nd_mrep, timo);
989 NFSCL_DEBUG(2, "clnt call=%d\n", stat);
990
991 if (rep != NULL) {
992 /*
993 * RPC done, unlink the request.
994 */
995 NFSLOCKREQ();
996 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
997 NFSUNLOCKREQ();
998 }
999
1000 /*
1001 * If there was a successful reply and a tprintf msg.
1002 * tprintf a response.
1003 */
1004 if (stat == RPC_SUCCESS) {
1005 error = 0;
1006 } else if (stat == RPC_TIMEDOUT) {
1007 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
1008 error = ETIMEDOUT;
1009 } else if (stat == RPC_VERSMISMATCH) {
1010 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
1011 error = EOPNOTSUPP;
1012 } else if (stat == RPC_PROGVERSMISMATCH) {
1013 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
1014 error = EPROTONOSUPPORT;
1015 } else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
1016 stat == RPC_SYSTEMERROR || stat == RPC_INTR) {
1017 /* Check for a session slot that needs to be free'd. */
1018 if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
1019 (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
1020 nd->nd_procnum != NFSPROC_NULL) {
1021 /*
1022 * This should only occur when either the MDS or
1023 * a client has an RPC against a DS fail.
1024 * This happens because these cases use "soft"
1025 * connections that can time out and fail.
1026 * The slot used for this RPC is now in a
1027 * non-deterministic state, but if the slot isn't
1028 * free'd, threads can get stuck waiting for a slot.
1029 */
1030 if (sep == NULL)
1031 sep = nfsmnt_mdssession(nmp);
1032 /*
1033 * Bump the sequence# out of range, so that reuse of
1034 * this slot will result in an NFSERR_SEQMISORDERED
1035 * error and not a bogus cached RPC reply.
1036 */
1037 mtx_lock(&sep->nfsess_mtx);
1038 sep->nfsess_slotseq[nd->nd_slotid] += 10;
1039 sep->nfsess_badslots |= (0x1ULL << nd->nd_slotid);
1040 mtx_unlock(&sep->nfsess_mtx);
1041 /* And free the slot. */
1042 nfsv4_freeslot(sep, nd->nd_slotid, false);
1043 }
1044 if (stat == RPC_INTR)
1045 error = EINTR;
1046 else {
1047 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
1048 error = ENXIO;
1049 }
1050 } else if (stat == RPC_AUTHERROR) {
1051 /* Check for a session slot that needs to be free'd. */
1052 if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
1053 (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
1054 nd->nd_procnum != NFSPROC_NULL) {
1055 /*
1056 * This can occur when a Kerberos/RPCSEC_GSS session
1057 * expires, due to TGT expiration.
1058 * Free the slot, resetting the slot's sequence#.
1059 */
1060 if (sep == NULL)
1061 sep = nfsmnt_mdssession(nmp);
1062 nfsv4_freeslot(sep, nd->nd_slotid, true);
1063 }
1064 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
1065 error = EACCES;
1066 } else {
1067 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
1068 error = EACCES;
1069 }
1070 if (error) {
1071 m_freem(nd->nd_mreq);
1072 if (usegssname == 0)
1073 AUTH_DESTROY(auth);
1074 if (rep != NULL)
1075 free(rep, M_NFSDREQ);
1076 if (set_sigset)
1077 newnfs_restore_sigmask(td, &oldset);
1078 return (error);
1079 }
1080
1081 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
1082
1083 /*
1084 * Search for any mbufs that are not a multiple of 4 bytes long
1085 * or with m_data not longword aligned.
1086 * These could cause pointer alignment problems, so copy them to
1087 * well aligned mbufs.
1088 */
1089 newnfs_realign(&nd->nd_mrep, M_WAITOK);
1090 nd->nd_md = nd->nd_mrep;
1091 nd->nd_dpos = mtod(nd->nd_md, caddr_t);
1092 nd->nd_repstat = 0;
1093 if (nd->nd_procnum != NFSPROC_NULL &&
1094 nd->nd_procnum != NFSV4PROC_CBNULL) {
1095 /* If sep == NULL, set it to the default in nmp. */
1096 if (sep == NULL && nmp != NULL)
1097 sep = nfsmnt_mdssession(nmp);
1098 /*
1099 * and now the actual NFS xdr.
1100 */
1101 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1102 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
1103 if (nd->nd_repstat >= 10000)
1104 NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
1105 (int)nd->nd_repstat);
1106
1107 /*
1108 * Get rid of the tag, return count and SEQUENCE result for
1109 * NFSv4.
1110 */
1111 if ((nd->nd_flag & ND_NFSV4) != 0 && nd->nd_repstat !=
1112 NFSERR_MINORVERMISMATCH) {
1113 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1114 i = fxdr_unsigned(int, *tl);
1115 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
1116 if (error)
1117 goto nfsmout;
1118 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1119 opcnt = fxdr_unsigned(int, *tl++);
1120 i = fxdr_unsigned(int, *tl++);
1121 j = fxdr_unsigned(int, *tl);
1122 if (j >= 10000)
1123 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
1124 /*
1125 * If the first op is Sequence, free up the slot.
1126 */
1127 if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
1128 (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0)) {
1129 NFSCL_DEBUG(1, "failed seq=%d\n", j);
1130 if (sep != NULL && i == NFSV4OP_SEQUENCE &&
1131 j == NFSERR_SEQMISORDERED) {
1132 mtx_lock(&sep->nfsess_mtx);
1133 sep->nfsess_badslots |=
1134 (0x1ULL << nd->nd_slotid);
1135 mtx_unlock(&sep->nfsess_mtx);
1136 }
1137 }
1138 if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
1139 (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
1140 j == 0)) && sep != NULL) {
1141 if (i == NFSV4OP_SEQUENCE)
1142 NFSM_DISSECT(tl, uint32_t *,
1143 NFSX_V4SESSIONID +
1144 5 * NFSX_UNSIGNED);
1145 else
1146 NFSM_DISSECT(tl, uint32_t *,
1147 NFSX_V4SESSIONID +
1148 4 * NFSX_UNSIGNED);
1149 mtx_lock(&sep->nfsess_mtx);
1150 if (bcmp(tl, sep->nfsess_sessionid,
1151 NFSX_V4SESSIONID) == 0) {
1152 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
1153 retseq = fxdr_unsigned(uint32_t, *tl++);
1154 slot = fxdr_unsigned(int, *tl++);
1155 if ((nd->nd_flag & ND_HASSLOTID) != 0) {
1156 if (slot >= NFSV4_SLOTS ||
1157 (i == NFSV4OP_CBSEQUENCE &&
1158 slot >= NFSV4_CBSLOTS)) {
1159 printf("newnfs_request:"
1160 " Bogus slot\n");
1161 slot = nd->nd_slotid;
1162 } else if (slot !=
1163 nd->nd_slotid) {
1164 printf("newnfs_request:"
1165 " Wrong session "
1166 "srvslot=%d "
1167 "slot=%d\n", slot,
1168 nd->nd_slotid);
1169 if (i == NFSV4OP_SEQUENCE) {
1170 /*
1171 * Mark both slots as
1172 * bad, because we do
1173 * not know if the
1174 * server has advanced
1175 * the sequence# for
1176 * either of them.
1177 */
1178 sep->nfsess_badslots |=
1179 (0x1ULL << slot);
1180 sep->nfsess_badslots |=
1181 (0x1ULL <<
1182 nd->nd_slotid);
1183 }
1184 slot = nd->nd_slotid;
1185 }
1186 freeslot = slot;
1187 } else if (slot != 0) {
1188 printf("newnfs_request: Bad "
1189 "session slot=%d\n", slot);
1190 slot = 0;
1191 }
1192 if (retseq != sep->nfsess_slotseq[slot])
1193 printf("retseq diff 0x%x\n",
1194 retseq);
1195 retval = fxdr_unsigned(uint32_t, *++tl);
1196 if ((retval + 1) < sep->nfsess_foreslots
1197 )
1198 sep->nfsess_foreslots = (retval
1199 + 1);
1200 else if ((retval + 1) >
1201 sep->nfsess_foreslots)
1202 sep->nfsess_foreslots = (retval
1203 < 64) ? (retval + 1) : 64;
1204 }
1205 mtx_unlock(&sep->nfsess_mtx);
1206
1207 /* Grab the op and status for the next one. */
1208 if (opcnt > 1) {
1209 NFSM_DISSECT(tl, uint32_t *,
1210 2 * NFSX_UNSIGNED);
1211 i = fxdr_unsigned(int, *tl++);
1212 j = fxdr_unsigned(int, *tl);
1213 }
1214 }
1215 }
1216 if (nd->nd_repstat != 0) {
1217 if (nd->nd_repstat == NFSERR_BADSESSION &&
1218 nmp != NULL && dssep == NULL &&
1219 (nd->nd_flag & ND_NFSV41) != 0) {
1220 /*
1221 * If this is a client side MDS RPC, mark
1222 * the MDS session defunct and initiate
1223 * recovery, as required.
1224 * The nfsess_defunct field is protected by
1225 * the NFSLOCKMNT()/nm_mtx lock and not the
1226 * nfsess_mtx lock to simplify its handling,
1227 * for the MDS session. This lock is also
1228 * sufficient for nfsess_sessionid, since it
1229 * never changes in the structure.
1230 */
1231 NFSCL_DEBUG(1, "Got badsession\n");
1232 NFSLOCKCLSTATE();
1233 NFSLOCKMNT(nmp);
1234 if (TAILQ_EMPTY(&nmp->nm_sess)) {
1235 NFSUNLOCKMNT(nmp);
1236 NFSUNLOCKCLSTATE();
1237 printf("If server has not rebooted, "
1238 "check NFS clients for unique "
1239 "/etc/hostid's\n");
1240 goto out;
1241 }
1242 sep = NFSMNT_MDSSESSION(nmp);
1243 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
1244 NFSX_V4SESSIONID) == 0) {
1245 printf("Initiate recovery. If server "
1246 "has not rebooted, "
1247 "check NFS clients for unique "
1248 "/etc/hostid's\n");
1249 /* Initiate recovery. */
1250 sep->nfsess_defunct = 1;
1251 NFSCL_DEBUG(1, "Marked defunct\n");
1252 if (nmp->nm_clp != NULL) {
1253 nmp->nm_clp->nfsc_flags |=
1254 NFSCLFLAGS_RECOVER;
1255 wakeup(nmp->nm_clp);
1256 }
1257 }
1258 NFSUNLOCKCLSTATE();
1259 /*
1260 * Sleep for up to 1sec waiting for a new
1261 * session.
1262 */
1263 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1264 "nfsbadsess", hz);
1265 /*
1266 * Get the session again, in case a new one
1267 * has been created during the sleep.
1268 */
1269 sep = NFSMNT_MDSSESSION(nmp);
1270 NFSUNLOCKMNT(nmp);
1271 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1272 reterr = nfsv4_sequencelookup(nmp, sep,
1273 &slotpos, &maxslot, &slotseq,
1274 sessionid, true);
1275 if (reterr == 0) {
1276 /* Fill in new session info. */
1277 NFSCL_DEBUG(1,
1278 "Filling in new sequence\n");
1279 tl = nd->nd_sequence;
1280 bcopy(sessionid, tl,
1281 NFSX_V4SESSIONID);
1282 tl += NFSX_V4SESSIONID /
1283 NFSX_UNSIGNED;
1284 *tl++ = txdr_unsigned(slotseq);
1285 *tl++ = txdr_unsigned(slotpos);
1286 *tl = txdr_unsigned(maxslot);
1287 nd->nd_slotid = slotpos;
1288 nd->nd_flag |= ND_HASSLOTID;
1289 }
1290 if (reterr == NFSERR_BADSESSION ||
1291 reterr == 0) {
1292 NFSCL_DEBUG(1,
1293 "Badsession looping\n");
1294 m_freem(nd->nd_mrep);
1295 nd->nd_mrep = NULL;
1296 goto tryagain;
1297 }
1298 nd->nd_repstat = reterr;
1299 NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1300 }
1301 }
1302 /*
1303 * When clp != NULL, it is a callback and all
1304 * callback operations can be retried for NFSERR_DELAY.
1305 */
1306 if (((nd->nd_repstat == NFSERR_DELAY ||
1307 nd->nd_repstat == NFSERR_GRACE) &&
1308 (nd->nd_flag & ND_NFSV4) && (clp != NULL ||
1309 (nd->nd_procnum != NFSPROC_DELEGRETURN &&
1310 nd->nd_procnum != NFSPROC_SETATTR &&
1311 nd->nd_procnum != NFSPROC_READ &&
1312 nd->nd_procnum != NFSPROC_READDS &&
1313 nd->nd_procnum != NFSPROC_WRITE &&
1314 nd->nd_procnum != NFSPROC_WRITEDS &&
1315 nd->nd_procnum != NFSPROC_OPEN &&
1316 nd->nd_procnum != NFSPROC_OPENLAYGET &&
1317 nd->nd_procnum != NFSPROC_CREATE &&
1318 nd->nd_procnum != NFSPROC_CREATELAYGET &&
1319 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1320 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1321 nd->nd_procnum != NFSPROC_CLOSE &&
1322 nd->nd_procnum != NFSPROC_LOCK &&
1323 nd->nd_procnum != NFSPROC_LOCKU))) ||
1324 (nd->nd_repstat == NFSERR_DELAY &&
1325 (nd->nd_flag & ND_NFSV4) == 0) ||
1326 nd->nd_repstat == NFSERR_RESOURCE ||
1327 nd->nd_repstat == NFSERR_RETRYUNCACHEDREP) {
1328 /* Clip at NFS_TRYLATERDEL. */
1329 if (timespeccmp(&trylater_delay,
1330 &nfs_trylater_max, >))
1331 trylater_delay = nfs_trylater_max;
1332 getnanouptime(&waituntil);
1333 timespecadd(&waituntil, &trylater_delay,
1334 &waituntil);
1335 do {
1336 nfs_catnap(PZERO, 0, "nfstry");
1337 getnanouptime(&ts);
1338 } while (timespeccmp(&ts, &waituntil, <));
1339 timespecadd(&trylater_delay, &trylater_delay,
1340 &trylater_delay); /* Double each time. */
1341 if (slot != -1) {
1342 mtx_lock(&sep->nfsess_mtx);
1343 sep->nfsess_slotseq[slot]++;
1344 *nd->nd_slotseq = txdr_unsigned(
1345 sep->nfsess_slotseq[slot]);
1346 mtx_unlock(&sep->nfsess_mtx);
1347 }
1348 m_freem(nd->nd_mrep);
1349 nd->nd_mrep = NULL;
1350 goto tryagain;
1351 }
1352
1353 /*
1354 * If the File Handle was stale, invalidate the
1355 * lookup cache, just in case.
1356 * (vp != NULL implies a client side call)
1357 */
1358 if (nd->nd_repstat == ESTALE && vp != NULL) {
1359 cache_purge(vp);
1360 if (ncl_call_invalcaches != NULL)
1361 (*ncl_call_invalcaches)(vp);
1362 }
1363 }
1364 if ((nd->nd_flag & ND_NFSV4) != 0) {
1365 /* Free the slot, as required. */
1366 if (freeslot != -1)
1367 nfsv4_freeslot(sep, freeslot, false);
1368 /*
1369 * If this op is Putfh, throw its results away.
1370 */
1371 if (j >= 10000)
1372 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1373 if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1374 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1375 i = fxdr_unsigned(int, *tl++);
1376 j = fxdr_unsigned(int, *tl);
1377 if (j >= 10000)
1378 NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1379 j);
1380 /*
1381 * All Compounds that do an Op that must
1382 * be in sequence consist of NFSV4OP_PUTFH
1383 * followed by one of these. As such, we
1384 * can determine if the seqid# should be
1385 * incremented, here.
1386 */
1387 if ((i == NFSV4OP_OPEN ||
1388 i == NFSV4OP_OPENCONFIRM ||
1389 i == NFSV4OP_OPENDOWNGRADE ||
1390 i == NFSV4OP_CLOSE ||
1391 i == NFSV4OP_LOCK ||
1392 i == NFSV4OP_LOCKU) &&
1393 (j == 0 ||
1394 (j != NFSERR_STALECLIENTID &&
1395 j != NFSERR_STALESTATEID &&
1396 j != NFSERR_BADSTATEID &&
1397 j != NFSERR_BADSEQID &&
1398 j != NFSERR_BADXDR &&
1399 j != NFSERR_RESOURCE &&
1400 j != NFSERR_NOFILEHANDLE)))
1401 nd->nd_flag |= ND_INCRSEQID;
1402 }
1403 /*
1404 * If this op's status is non-zero, mark
1405 * that there is no more data to process.
1406 * The exception is Setattr, which always has xdr
1407 * when it has failed.
1408 */
1409 if (j != 0 && i != NFSV4OP_SETATTR)
1410 nd->nd_flag |= ND_NOMOREDATA;
1411
1412 /*
1413 * If R_DONTRECOVER is set, replace the stale error
1414 * reply, so that recovery isn't initiated.
1415 */
1416 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1417 nd->nd_repstat == NFSERR_BADSESSION ||
1418 nd->nd_repstat == NFSERR_STALESTATEID) &&
1419 rep != NULL && (rep->r_flags & R_DONTRECOVER))
1420 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1421 }
1422 }
1423 out:
1424
1425 #ifdef KDTRACE_HOOKS
1426 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1427 uint32_t probe_id;
1428 int probe_procnum;
1429
1430 if (nd->nd_flag & ND_NFSV4) {
1431 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1432 probe_procnum = nd->nd_procnum;
1433 } else if (nd->nd_flag & ND_NFSV3) {
1434 probe_id = nfscl_nfs3_done_probes[procnum];
1435 probe_procnum = procnum;
1436 } else {
1437 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1438 probe_procnum = procnum;
1439 }
1440 if (probe_id != 0)
1441 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1442 nd->nd_mreq, cred, probe_procnum, 0);
1443 }
1444 #endif
1445
1446 m_freem(nd->nd_mreq);
1447 if (usegssname == 0)
1448 AUTH_DESTROY(auth);
1449 if (rep != NULL)
1450 free(rep, M_NFSDREQ);
1451 if (set_sigset)
1452 newnfs_restore_sigmask(td, &oldset);
1453 return (0);
1454 nfsmout:
1455 m_freem(nd->nd_mrep);
1456 m_freem(nd->nd_mreq);
1457 if (usegssname == 0)
1458 AUTH_DESTROY(auth);
1459 if (rep != NULL)
1460 free(rep, M_NFSDREQ);
1461 if (set_sigset)
1462 newnfs_restore_sigmask(td, &oldset);
1463 return (error);
1464 }
1465
1466 /*
1467 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1468 * wait for all requests to complete. This is used by forced unmounts
1469 * to terminate any outstanding RPCs.
1470 */
1471 int
newnfs_nmcancelreqs(struct nfsmount * nmp)1472 newnfs_nmcancelreqs(struct nfsmount *nmp)
1473 {
1474 struct nfsclds *dsp;
1475 struct __rpc_client *cl;
1476 int i;
1477
1478 if (nmp->nm_sockreq.nr_client != NULL)
1479 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1480 for (i = 0; i < nmp->nm_aconnect; i++)
1481 if (nmp->nm_aconn[i] != NULL)
1482 CLNT_CLOSE(nmp->nm_aconn[i]);
1483 lookformore:
1484 NFSLOCKMNT(nmp);
1485 TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1486 NFSLOCKDS(dsp);
1487 if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1488 (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1489 dsp->nfsclds_sockp != NULL &&
1490 dsp->nfsclds_sockp->nr_client != NULL) {
1491 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1492 cl = dsp->nfsclds_sockp->nr_client;
1493 NFSUNLOCKDS(dsp);
1494 NFSUNLOCKMNT(nmp);
1495 CLNT_CLOSE(cl);
1496 goto lookformore;
1497 }
1498 NFSUNLOCKDS(dsp);
1499 }
1500 NFSUNLOCKMNT(nmp);
1501 return (0);
1502 }
1503
1504 /*
1505 * Any signal that can interrupt an NFS operation in an intr mount
1506 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1507 */
1508 int newnfs_sig_set[] = {
1509 SIGINT,
1510 SIGTERM,
1511 SIGHUP,
1512 SIGKILL,
1513 SIGQUIT
1514 };
1515
1516 /*
1517 * Check to see if one of the signals in our subset is pending on
1518 * the process (in an intr mount).
1519 */
1520 static int
nfs_sig_pending(sigset_t set)1521 nfs_sig_pending(sigset_t set)
1522 {
1523 int i;
1524
1525 for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1526 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1527 return (1);
1528 return (0);
1529 }
1530
1531 /*
1532 * The set/restore sigmask functions are used to (temporarily) overwrite
1533 * the thread td_sigmask during an RPC call (for example). These are also
1534 * used in other places in the NFS client that might tsleep().
1535 */
1536 void
newnfs_set_sigmask(struct thread * td,sigset_t * oldset)1537 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1538 {
1539 sigset_t newset;
1540 int i;
1541 struct proc *p;
1542
1543 SIGFILLSET(newset);
1544 if (td == NULL)
1545 td = curthread; /* XXX */
1546 p = td->td_proc;
1547 /* Remove the NFS set of signals from newset */
1548 PROC_LOCK(p);
1549 mtx_lock(&p->p_sigacts->ps_mtx);
1550 for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1551 /*
1552 * But make sure we leave the ones already masked
1553 * by the process, ie. remove the signal from the
1554 * temporary signalmask only if it wasn't already
1555 * in p_sigmask.
1556 */
1557 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1558 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1559 SIGDELSET(newset, newnfs_sig_set[i]);
1560 }
1561 mtx_unlock(&p->p_sigacts->ps_mtx);
1562 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1563 SIGPROCMASK_PROC_LOCKED);
1564 PROC_UNLOCK(p);
1565 }
1566
1567 void
newnfs_restore_sigmask(struct thread * td,sigset_t * set)1568 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1569 {
1570 if (td == NULL)
1571 td = curthread; /* XXX */
1572 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1573 }
1574
1575 /*
1576 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1577 * old one after msleep() returns.
1578 */
1579 int
newnfs_msleep(struct thread * td,void * ident,struct mtx * mtx,int priority,char * wmesg,int timo)1580 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1581 {
1582 sigset_t oldset;
1583 int error;
1584
1585 if ((priority & PCATCH) == 0)
1586 return msleep(ident, mtx, priority, wmesg, timo);
1587 if (td == NULL)
1588 td = curthread; /* XXX */
1589 newnfs_set_sigmask(td, &oldset);
1590 error = msleep(ident, mtx, priority, wmesg, timo);
1591 newnfs_restore_sigmask(td, &oldset);
1592 return (error);
1593 }
1594
1595 /*
1596 * Test for a termination condition pending on the process.
1597 * This is used for NFSMNT_INT mounts.
1598 */
1599 int
newnfs_sigintr(struct nfsmount * nmp,struct thread * td)1600 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1601 {
1602 struct proc *p;
1603 sigset_t tmpset;
1604
1605 /* Terminate all requests while attempting a forced unmount. */
1606 if (NFSCL_FORCEDISM(nmp->nm_mountp))
1607 return (EIO);
1608 if (!(nmp->nm_flag & NFSMNT_INT))
1609 return (0);
1610 if (td == NULL)
1611 return (0);
1612 p = td->td_proc;
1613 PROC_LOCK(p);
1614 tmpset = p->p_siglist;
1615 SIGSETOR(tmpset, td->td_siglist);
1616 SIGSETNAND(tmpset, td->td_sigmask);
1617 mtx_lock(&p->p_sigacts->ps_mtx);
1618 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1619 mtx_unlock(&p->p_sigacts->ps_mtx);
1620 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1621 && nfs_sig_pending(tmpset)) {
1622 PROC_UNLOCK(p);
1623 return (EINTR);
1624 }
1625 PROC_UNLOCK(p);
1626 return (0);
1627 }
1628
1629 static int
nfs_msg(struct thread * td,const char * server,const char * msg,int error)1630 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1631 {
1632 struct proc *p;
1633
1634 p = td ? td->td_proc : NULL;
1635 if (error) {
1636 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1637 server, msg, error);
1638 } else {
1639 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1640 }
1641 return (0);
1642 }
1643
1644 static void
nfs_down(struct nfsmount * nmp,struct thread * td,const char * msg,int error,int flags)1645 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1646 int error, int flags)
1647 {
1648 if (nmp == NULL)
1649 return;
1650 mtx_lock(&nmp->nm_mtx);
1651 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1652 nmp->nm_state |= NFSSTA_TIMEO;
1653 mtx_unlock(&nmp->nm_mtx);
1654 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1655 VQ_NOTRESP, 0);
1656 } else
1657 mtx_unlock(&nmp->nm_mtx);
1658 mtx_lock(&nmp->nm_mtx);
1659 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1660 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1661 mtx_unlock(&nmp->nm_mtx);
1662 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1663 VQ_NOTRESPLOCK, 0);
1664 } else
1665 mtx_unlock(&nmp->nm_mtx);
1666 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1667 }
1668
1669 static void
nfs_up(struct nfsmount * nmp,struct thread * td,const char * msg,int flags,int tprintfmsg)1670 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1671 int flags, int tprintfmsg)
1672 {
1673 if (nmp == NULL)
1674 return;
1675 if (tprintfmsg) {
1676 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1677 }
1678
1679 mtx_lock(&nmp->nm_mtx);
1680 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1681 nmp->nm_state &= ~NFSSTA_TIMEO;
1682 mtx_unlock(&nmp->nm_mtx);
1683 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1684 VQ_NOTRESP, 1);
1685 } else
1686 mtx_unlock(&nmp->nm_mtx);
1687
1688 mtx_lock(&nmp->nm_mtx);
1689 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1690 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1691 mtx_unlock(&nmp->nm_mtx);
1692 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1693 VQ_NOTRESPLOCK, 1);
1694 } else
1695 mtx_unlock(&nmp->nm_mtx);
1696 }
1697