1 /*-
2 * Copyright (c) 1989, 1991, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 /*
38 * Socket operations for use by nfs
39 */
40
41 #include "opt_kdtrace.h"
42 #include "opt_kgssapi.h"
43 #include "opt_nfs.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/limits.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.h>
55 #include <sys/signalvar.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysctl.h>
58 #include <sys/syslog.h>
59 #include <sys/vnode.h>
60
61 #include <rpc/rpc.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[NFS_NPROCS + 1];
80 uint32_t nfscl_nfs2_done_probes[NFS_NPROCS + 1];
81
82 uint32_t nfscl_nfs3_start_probes[NFS_NPROCS + 1];
83 uint32_t nfscl_nfs3_done_probes[NFS_NPROCS + 1];
84
85 uint32_t nfscl_nfs4_start_probes[NFS_NPROCS + 1];
86 uint32_t nfscl_nfs4_done_probes[NFS_NPROCS + 1];
87 #endif
88
89 NFSSTATESPINLOCK;
90 NFSREQSPINLOCK;
91 extern struct nfsstats newnfsstats;
92 extern struct nfsreqhead nfsd_reqq;
93 extern int nfscl_ticks;
94 extern void (*ncl_call_invalcaches)(struct vnode *);
95
96 static int nfsrv_gsscallbackson = 0;
97 static int nfs_bufpackets = 4;
98 static int nfs_reconnects;
99 static int nfs3_jukebox_delay = 10;
100 static int nfs_skip_wcc_data_onerr = 1;
101 static int nfs_keytab_enctype = ETYPE_DES_CBC_CRC;
102
103 SYSCTL_DECL(_vfs_nfs);
104
105 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
106 "Buffer reservation size 2 < x < 64");
107 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
108 "Number of times the nfs client has had to reconnect");
109 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
110 "Number of seconds to delay a retry after receiving EJUKEBOX");
111 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
112 "Disable weak cache consistency checking when server returns an error");
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, keytab_enctype, CTLFLAG_RW, &nfs_keytab_enctype, 0,
114 "Encryption type for the keytab entry used by nfs");
115
116 static void nfs_down(struct nfsmount *, struct thread *, const char *,
117 int, int);
118 static void nfs_up(struct nfsmount *, struct thread *, const char *,
119 int, int);
120 static int nfs_msg(struct thread *, const char *, const char *, int);
121
122 struct nfs_cached_auth {
123 int ca_refs; /* refcount, including 1 from the cache */
124 uid_t ca_uid; /* uid that corresponds to this auth */
125 AUTH *ca_auth; /* RPC auth handle */
126 };
127
128 static int nfsv2_procid[NFS_V3NPROCS] = {
129 NFSV2PROC_NULL,
130 NFSV2PROC_GETATTR,
131 NFSV2PROC_SETATTR,
132 NFSV2PROC_LOOKUP,
133 NFSV2PROC_NOOP,
134 NFSV2PROC_READLINK,
135 NFSV2PROC_READ,
136 NFSV2PROC_WRITE,
137 NFSV2PROC_CREATE,
138 NFSV2PROC_MKDIR,
139 NFSV2PROC_SYMLINK,
140 NFSV2PROC_CREATE,
141 NFSV2PROC_REMOVE,
142 NFSV2PROC_RMDIR,
143 NFSV2PROC_RENAME,
144 NFSV2PROC_LINK,
145 NFSV2PROC_READDIR,
146 NFSV2PROC_NOOP,
147 NFSV2PROC_STATFS,
148 NFSV2PROC_NOOP,
149 NFSV2PROC_NOOP,
150 NFSV2PROC_NOOP,
151 };
152
153 /*
154 * Initialize sockets and congestion for a new NFS connection.
155 * We do not free the sockaddr if error.
156 */
157 int
158 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
159 struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
160 {
161 int rcvreserve, sndreserve;
162 int pktscale;
163 struct sockaddr *saddr;
164 struct ucred *origcred;
165 CLIENT *client;
166 struct netconfig *nconf;
167 struct socket *so;
168 int one = 1, retries, error = 0;
169 struct thread *td = curthread;
170 struct timeval timo;
171
172 /*
173 * We need to establish the socket using the credentials of
174 * the mountpoint. Some parts of this process (such as
175 * sobind() and soconnect()) will use the curent thread's
176 * credential instead of the socket credential. To work
177 * around this, temporarily change the current thread's
178 * credential to that of the mountpoint.
179 *
180 * XXX: It would be better to explicitly pass the correct
181 * credential to sobind() and soconnect().
182 */
183 origcred = td->td_ucred;
184
185 /*
186 * Use the credential in nr_cred, if not NULL.
187 */
188 if (nrp->nr_cred != NULL)
189 td->td_ucred = nrp->nr_cred;
190 else
191 td->td_ucred = cred;
192 saddr = nrp->nr_nam;
193
194 if (saddr->sa_family == AF_INET)
195 if (nrp->nr_sotype == SOCK_DGRAM)
196 nconf = getnetconfigent("udp");
197 else
198 nconf = getnetconfigent("tcp");
199 else
200 if (nrp->nr_sotype == SOCK_DGRAM)
201 nconf = getnetconfigent("udp6");
202 else
203 nconf = getnetconfigent("tcp6");
204
205 pktscale = nfs_bufpackets;
206 if (pktscale < 2)
207 pktscale = 2;
208 if (pktscale > 64)
209 pktscale = 64;
210 /*
211 * soreserve() can fail if sb_max is too small, so shrink pktscale
212 * and try again if there is an error.
213 * Print a log message suggesting increasing sb_max.
214 * Creating a socket and doing this is necessary since, if the
215 * reservation sizes are too large and will make soreserve() fail,
216 * the connection will work until a large send is attempted and
217 * then it will loop in the krpc code.
218 */
219 so = NULL;
220 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
221 error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
222 nrp->nr_soproto, td->td_ucred, td);
223 if (error) {
224 td->td_ucred = origcred;
225 goto out;
226 }
227 do {
228 if (error != 0 && pktscale > 2)
229 pktscale--;
230 if (nrp->nr_sotype == SOCK_DGRAM) {
231 if (nmp != NULL) {
232 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
233 pktscale;
234 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
235 pktscale;
236 } else {
237 sndreserve = rcvreserve = 1024 * pktscale;
238 }
239 } else {
240 if (nrp->nr_sotype != SOCK_STREAM)
241 panic("nfscon sotype");
242 if (nmp != NULL) {
243 sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
244 sizeof (u_int32_t)) * pktscale;
245 rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR +
246 sizeof (u_int32_t)) * pktscale;
247 } else {
248 sndreserve = rcvreserve = 1024 * pktscale;
249 }
250 }
251 error = soreserve(so, sndreserve, rcvreserve);
252 } while (error != 0 && pktscale > 2);
253 soclose(so);
254 if (error) {
255 td->td_ucred = origcred;
256 goto out;
257 }
258
259 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
260 nrp->nr_vers, sndreserve, rcvreserve);
261 CLNT_CONTROL(client, CLSET_WAITCHAN, "newnfsreq");
262 if (nmp != NULL) {
263 if ((nmp->nm_flag & NFSMNT_INT))
264 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
265 if ((nmp->nm_flag & NFSMNT_RESVPORT))
266 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
267 if (NFSHASSOFT(nmp)) {
268 if (nmp->nm_sotype == SOCK_DGRAM)
269 /*
270 * For UDP, the large timeout for a reconnect
271 * will be set to "nm_retry * nm_timeo / 2", so
272 * we only want to do 2 reconnect timeout
273 * retries.
274 */
275 retries = 2;
276 else
277 retries = nmp->nm_retry;
278 } else
279 retries = INT_MAX;
280 } else {
281 /*
282 * Three cases:
283 * - Null RPC callback to client
284 * - Non-Null RPC callback to client, wait a little longer
285 * - upcalls to nfsuserd and gssd (clp == NULL)
286 */
287 if (callback_retry_mult == 0) {
288 retries = NFSV4_UPCALLRETRY;
289 CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
290 } else {
291 retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
292 }
293 }
294 CLNT_CONTROL(client, CLSET_RETRIES, &retries);
295
296 if (nmp != NULL) {
297 /*
298 * For UDP, there are 2 timeouts:
299 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
300 * that does a retransmit of an RPC request using the same
301 * socket and xid. This is what you normally want to do,
302 * since NFS servers depend on "same xid" for their
303 * Duplicate Request Cache.
304 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
305 * retransmits on the same socket should fail and a fresh
306 * socket created. Each of these timeouts counts as one
307 * CLSET_RETRIES as set above.
308 * Set the initial retransmit timeout for UDP. This timeout
309 * doesn't exist for TCP and the following call just fails,
310 * which is ok.
311 */
312 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
313 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
314 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
315 }
316
317 mtx_lock(&nrp->nr_mtx);
318 if (nrp->nr_client != NULL) {
319 mtx_unlock(&nrp->nr_mtx);
320 /*
321 * Someone else already connected.
322 */
323 CLNT_RELEASE(client);
324 } else {
325 nrp->nr_client = client;
326 /*
327 * Protocols that do not require connections may be optionally
328 * left unconnected for servers that reply from a port other
329 * than NFS_PORT.
330 */
331 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
332 mtx_unlock(&nrp->nr_mtx);
333 CLNT_CONTROL(client, CLSET_CONNECT, &one);
334 } else
335 mtx_unlock(&nrp->nr_mtx);
336 }
337
338
339 /* Restore current thread's credentials. */
340 td->td_ucred = origcred;
341
342 out:
343 NFSEXITCODE(error);
344 return (error);
345 }
346
347 /*
348 * NFS disconnect. Clean up and unlink.
349 */
350 void
351 newnfs_disconnect(struct nfssockreq *nrp)
352 {
353 CLIENT *client;
354
355 mtx_lock(&nrp->nr_mtx);
356 if (nrp->nr_client != NULL) {
357 client = nrp->nr_client;
358 nrp->nr_client = NULL;
359 mtx_unlock(&nrp->nr_mtx);
360 rpc_gss_secpurge_call(client);
361 CLNT_CLOSE(client);
362 CLNT_RELEASE(client);
363 } else {
364 mtx_unlock(&nrp->nr_mtx);
365 }
366 }
367
368 static AUTH *
369 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
370 char *srv_principal, gss_OID mech_oid, struct ucred *cred)
371 {
372 rpc_gss_service_t svc;
373 AUTH *auth;
374 #ifdef notyet
375 rpc_gss_options_req_t req_options;
376 #endif
377
378 switch (secflavour) {
379 case RPCSEC_GSS_KRB5:
380 case RPCSEC_GSS_KRB5I:
381 case RPCSEC_GSS_KRB5P:
382 if (!mech_oid) {
383 if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
384 return (NULL);
385 }
386 if (secflavour == RPCSEC_GSS_KRB5)
387 svc = rpc_gss_svc_none;
388 else if (secflavour == RPCSEC_GSS_KRB5I)
389 svc = rpc_gss_svc_integrity;
390 else
391 svc = rpc_gss_svc_privacy;
392 #ifdef notyet
393 req_options.req_flags = GSS_C_MUTUAL_FLAG;
394 req_options.time_req = 0;
395 req_options.my_cred = GSS_C_NO_CREDENTIAL;
396 req_options.input_channel_bindings = NULL;
397 req_options.enc_type = nfs_keytab_enctype;
398
399 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
400 clnt_principal, srv_principal, mech_oid, svc,
401 &req_options);
402 #else
403 /*
404 * Until changes to the rpcsec_gss code are committed,
405 * there is no support for host based initiator
406 * principals. As such, that case cannot yet be handled.
407 */
408 if (clnt_principal == NULL)
409 auth = rpc_gss_secfind_call(nrp->nr_client, cred,
410 srv_principal, mech_oid, svc);
411 else
412 auth = NULL;
413 #endif
414 if (auth != NULL)
415 return (auth);
416 /* fallthrough */
417 case AUTH_SYS:
418 default:
419 return (authunix_create(cred));
420
421 }
422 }
423
424 /*
425 * Callback from the RPC code to generate up/down notifications.
426 */
427
428 struct nfs_feedback_arg {
429 struct nfsmount *nf_mount;
430 int nf_lastmsg; /* last tprintf */
431 int nf_tprintfmsg;
432 struct thread *nf_td;
433 };
434
435 static void
436 nfs_feedback(int type, int proc, void *arg)
437 {
438 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
439 struct nfsmount *nmp = nf->nf_mount;
440 time_t now;
441
442 switch (type) {
443 case FEEDBACK_REXMIT2:
444 case FEEDBACK_RECONNECT:
445 now = NFSD_MONOSEC;
446 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
447 nfs_down(nmp, nf->nf_td,
448 "not responding", 0, NFSSTA_TIMEO);
449 nf->nf_tprintfmsg = TRUE;
450 nf->nf_lastmsg = now;
451 }
452 break;
453
454 case FEEDBACK_OK:
455 nfs_up(nf->nf_mount, nf->nf_td,
456 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
457 break;
458 }
459 }
460
461 /*
462 * newnfs_request - goes something like this
463 * - does the rpc by calling the krpc layer
464 * - break down rpc header and return with nfs reply
465 * nb: always frees up nd_mreq mbuf list
466 */
467 int
468 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
469 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
470 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
471 u_char *retsum, int toplevel, u_int64_t *xidp)
472 {
473 u_int32_t *tl;
474 time_t waituntil;
475 int i, j, set_sigset = 0, timeo;
476 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS;
477 u_int16_t procnum;
478 u_int trylater_delay = 1;
479 struct nfs_feedback_arg nf;
480 struct timeval timo;
481 AUTH *auth;
482 struct rpc_callextra ext;
483 enum clnt_stat stat;
484 struct nfsreq *rep = NULL;
485 char *srv_principal = NULL;
486 sigset_t oldset;
487 struct ucred *authcred;
488
489 if (xidp != NULL)
490 *xidp = 0;
491 /* Reject requests while attempting a forced unmount. */
492 if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) {
493 m_freem(nd->nd_mreq);
494 return (ESTALE);
495 }
496
497 /*
498 * Set authcred, which is used to acquire RPC credentials to
499 * the cred argument, by default. The crhold() should not be
500 * necessary, but will ensure that some future code change
501 * doesn't result in the credential being free'd prematurely.
502 */
503 authcred = crhold(cred);
504
505 /* For client side interruptible mounts, mask off the signals. */
506 if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
507 newnfs_set_sigmask(td, &oldset);
508 set_sigset = 1;
509 }
510
511 /*
512 * XXX if not already connected call nfs_connect now. Longer
513 * term, change nfs_mount to call nfs_connect unconditionally
514 * and let clnt_reconnect_create handle reconnects.
515 */
516 if (nrp->nr_client == NULL)
517 newnfs_connect(nmp, nrp, cred, td, 0);
518
519 /*
520 * For a client side mount, nmp is != NULL and clp == NULL. For
521 * server calls (callbacks or upcalls), nmp == NULL.
522 */
523 if (clp != NULL) {
524 NFSLOCKSTATE();
525 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
526 secflavour = RPCSEC_GSS_KRB5;
527 if (nd->nd_procnum != NFSPROC_NULL) {
528 if (clp->lc_flags & LCL_GSSINTEGRITY)
529 secflavour = RPCSEC_GSS_KRB5I;
530 else if (clp->lc_flags & LCL_GSSPRIVACY)
531 secflavour = RPCSEC_GSS_KRB5P;
532 }
533 }
534 NFSUNLOCKSTATE();
535 } else if (nmp != NULL && NFSHASKERB(nmp) &&
536 nd->nd_procnum != NFSPROC_NULL) {
537 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
538 nd->nd_flag |= ND_USEGSSNAME;
539 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
540 /*
541 * If there is a client side host based credential,
542 * use that, otherwise use the system uid, if set.
543 * The system uid is in the nmp->nm_sockreq.nr_cred
544 * credentials.
545 */
546 if (nmp->nm_krbnamelen > 0) {
547 usegssname = 1;
548 } else if (nmp->nm_uid != (uid_t)-1) {
549 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
550 ("newnfs_request: NULL nr_cred"));
551 crfree(authcred);
552 authcred = crhold(nmp->nm_sockreq.nr_cred);
553 }
554 } else if (nmp->nm_krbnamelen == 0 &&
555 nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
556 /*
557 * If there is no host based principal name and
558 * the system uid is set and this is root, use the
559 * system uid, since root won't have user
560 * credentials in a credentials cache file.
561 * The system uid is in the nmp->nm_sockreq.nr_cred
562 * credentials.
563 */
564 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
565 ("newnfs_request: NULL nr_cred"));
566 crfree(authcred);
567 authcred = crhold(nmp->nm_sockreq.nr_cred);
568 }
569 if (NFSHASINTEGRITY(nmp))
570 secflavour = RPCSEC_GSS_KRB5I;
571 else if (NFSHASPRIVACY(nmp))
572 secflavour = RPCSEC_GSS_KRB5P;
573 else
574 secflavour = RPCSEC_GSS_KRB5;
575 srv_principal = NFSMNT_SRVKRBNAME(nmp);
576 } else if (nmp != NULL && !NFSHASKERB(nmp) &&
577 nd->nd_procnum != NFSPROC_NULL &&
578 (nd->nd_flag & ND_USEGSSNAME) != 0) {
579 /*
580 * Use the uid that did the mount when the RPC is doing
581 * NFSv4 system operations, as indicated by the
582 * ND_USEGSSNAME flag, for the AUTH_SYS case.
583 * The credentials in nm_sockreq.nr_cred were used for the
584 * mount.
585 */
586 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
587 ("newnfs_request: NULL nr_cred"));
588 crfree(authcred);
589 authcred = crhold(nmp->nm_sockreq.nr_cred);
590 }
591
592 if (nmp != NULL) {
593 bzero(&nf, sizeof(struct nfs_feedback_arg));
594 nf.nf_mount = nmp;
595 nf.nf_td = td;
596 nf.nf_lastmsg = NFSD_MONOSEC -
597 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
598 }
599
600 if (nd->nd_procnum == NFSPROC_NULL)
601 auth = authnone_create();
602 else if (usegssname)
603 auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname,
604 srv_principal, NULL, authcred);
605 else
606 auth = nfs_getauth(nrp, secflavour, NULL,
607 srv_principal, NULL, authcred);
608 crfree(authcred);
609 if (auth == NULL) {
610 m_freem(nd->nd_mreq);
611 if (set_sigset)
612 newnfs_restore_sigmask(td, &oldset);
613 return (EACCES);
614 }
615 bzero(&ext, sizeof(ext));
616 ext.rc_auth = auth;
617 if (nmp != NULL) {
618 ext.rc_feedback = nfs_feedback;
619 ext.rc_feedback_arg = &nf;
620 }
621
622 procnum = nd->nd_procnum;
623 if ((nd->nd_flag & ND_NFSV4) &&
624 nd->nd_procnum != NFSPROC_NULL &&
625 nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
626 procnum = NFSV4PROC_COMPOUND;
627
628 if (nmp != NULL) {
629 NFSINCRGLOBAL(newnfsstats.rpcrequests);
630
631 /* Map the procnum to the old NFSv2 one, as required. */
632 if ((nd->nd_flag & ND_NFSV2) != 0) {
633 if (nd->nd_procnum < NFS_V3NPROCS)
634 procnum = nfsv2_procid[nd->nd_procnum];
635 else
636 procnum = NFSV2PROC_NOOP;
637 }
638
639 /*
640 * Now only used for the R_DONTRECOVER case, but until that is
641 * supported within the krpc code, I need to keep a queue of
642 * outstanding RPCs for nfsv4 client requests.
643 */
644 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
645 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq),
646 M_NFSDREQ, M_WAITOK);
647 #ifdef KDTRACE_HOOKS
648 if (dtrace_nfscl_nfs234_start_probe != NULL) {
649 uint32_t probe_id;
650 int probe_procnum;
651
652 if (nd->nd_flag & ND_NFSV4) {
653 probe_id =
654 nfscl_nfs4_start_probes[nd->nd_procnum];
655 probe_procnum = nd->nd_procnum;
656 } else if (nd->nd_flag & ND_NFSV3) {
657 probe_id = nfscl_nfs3_start_probes[procnum];
658 probe_procnum = procnum;
659 } else {
660 probe_id =
661 nfscl_nfs2_start_probes[nd->nd_procnum];
662 probe_procnum = procnum;
663 }
664 if (probe_id != 0)
665 (dtrace_nfscl_nfs234_start_probe)
666 (probe_id, vp, nd->nd_mreq, cred,
667 probe_procnum);
668 }
669 #endif
670 }
671 trycnt = 0;
672 tryagain:
673 /*
674 * This timeout specifies when a new socket should be created,
675 * along with new xid values. For UDP, this should be done
676 * infrequently, since retransmits of RPC requests should normally
677 * use the same xid.
678 */
679 if (nmp == NULL) {
680 timo.tv_usec = 0;
681 if (clp == NULL)
682 timo.tv_sec = NFSV4_UPCALLTIMEO;
683 else
684 timo.tv_sec = NFSV4_CALLBACKTIMEO;
685 } else {
686 if (nrp->nr_sotype != SOCK_DGRAM) {
687 timo.tv_usec = 0;
688 if ((nmp->nm_flag & NFSMNT_NFSV4))
689 timo.tv_sec = INT_MAX;
690 else
691 timo.tv_sec = NFS_TCPTIMEO;
692 } else {
693 if (NFSHASSOFT(nmp)) {
694 /*
695 * CLSET_RETRIES is set to 2, so this should be
696 * half of the total timeout required.
697 */
698 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
699 if (timeo < 1)
700 timeo = 1;
701 timo.tv_sec = timeo / NFS_HZ;
702 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
703 NFS_HZ;
704 } else {
705 /* For UDP hard mounts, use a large value. */
706 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
707 timo.tv_usec = 0;
708 }
709 }
710
711 if (rep != NULL) {
712 rep->r_flags = 0;
713 rep->r_nmp = nmp;
714 /*
715 * Chain request into list of outstanding requests.
716 */
717 NFSLOCKREQ();
718 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
719 NFSUNLOCKREQ();
720 }
721 }
722
723 nd->nd_mrep = NULL;
724 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq,
725 &nd->nd_mrep, timo);
726
727 if (rep != NULL) {
728 /*
729 * RPC done, unlink the request.
730 */
731 NFSLOCKREQ();
732 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
733 NFSUNLOCKREQ();
734 }
735
736 /*
737 * If there was a successful reply and a tprintf msg.
738 * tprintf a response.
739 */
740 if (stat == RPC_SUCCESS) {
741 error = 0;
742 } else if (stat == RPC_TIMEDOUT) {
743 NFSINCRGLOBAL(newnfsstats.rpctimeouts);
744 error = ETIMEDOUT;
745 } else if (stat == RPC_VERSMISMATCH) {
746 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
747 error = EOPNOTSUPP;
748 } else if (stat == RPC_PROGVERSMISMATCH) {
749 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
750 error = EPROTONOSUPPORT;
751 } else if (stat == RPC_INTR) {
752 error = EINTR;
753 } else {
754 NFSINCRGLOBAL(newnfsstats.rpcinvalid);
755 error = EACCES;
756 }
757 if (error) {
758 m_freem(nd->nd_mreq);
759 AUTH_DESTROY(auth);
760 if (rep != NULL)
761 FREE((caddr_t)rep, M_NFSDREQ);
762 if (set_sigset)
763 newnfs_restore_sigmask(td, &oldset);
764 return (error);
765 }
766
767 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
768
769 /*
770 * Search for any mbufs that are not a multiple of 4 bytes long
771 * or with m_data not longword aligned.
772 * These could cause pointer alignment problems, so copy them to
773 * well aligned mbufs.
774 */
775 newnfs_realign(&nd->nd_mrep, M_WAITOK);
776 nd->nd_md = nd->nd_mrep;
777 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
778 nd->nd_repstat = 0;
779 if (nd->nd_procnum != NFSPROC_NULL) {
780 /*
781 * and now the actual NFS xdr.
782 */
783 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
784 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
785 if (nd->nd_repstat != 0) {
786 if (((nd->nd_repstat == NFSERR_DELAY ||
787 nd->nd_repstat == NFSERR_GRACE) &&
788 (nd->nd_flag & ND_NFSV4) &&
789 nd->nd_procnum != NFSPROC_DELEGRETURN &&
790 nd->nd_procnum != NFSPROC_SETATTR &&
791 nd->nd_procnum != NFSPROC_READ &&
792 nd->nd_procnum != NFSPROC_WRITE &&
793 nd->nd_procnum != NFSPROC_OPEN &&
794 nd->nd_procnum != NFSPROC_CREATE &&
795 nd->nd_procnum != NFSPROC_OPENCONFIRM &&
796 nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
797 nd->nd_procnum != NFSPROC_CLOSE &&
798 nd->nd_procnum != NFSPROC_LOCK &&
799 nd->nd_procnum != NFSPROC_LOCKU) ||
800 (nd->nd_repstat == NFSERR_DELAY &&
801 (nd->nd_flag & ND_NFSV4) == 0) ||
802 nd->nd_repstat == NFSERR_RESOURCE) {
803 if (trylater_delay > NFS_TRYLATERDEL)
804 trylater_delay = NFS_TRYLATERDEL;
805 waituntil = NFSD_MONOSEC + trylater_delay;
806 while (NFSD_MONOSEC < waituntil)
807 (void) nfs_catnap(PZERO, 0, "nfstry");
808 trylater_delay *= 2;
809 m_freem(nd->nd_mrep);
810 nd->nd_mrep = NULL;
811 goto tryagain;
812 }
813
814 /*
815 * If the File Handle was stale, invalidate the
816 * lookup cache, just in case.
817 * (vp != NULL implies a client side call)
818 */
819 if (nd->nd_repstat == ESTALE && vp != NULL) {
820 cache_purge(vp);
821 if (ncl_call_invalcaches != NULL)
822 (*ncl_call_invalcaches)(vp);
823 }
824 }
825
826 /*
827 * Get rid of the tag, return count, and PUTFH result for V4.
828 */
829 if (nd->nd_flag & ND_NFSV4) {
830 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
831 i = fxdr_unsigned(int, *tl);
832 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
833 if (error)
834 goto nfsmout;
835 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
836 i = fxdr_unsigned(int, *++tl);
837
838 /*
839 * If the first op's status is non-zero, mark that
840 * there is no more data to process.
841 */
842 if (*++tl)
843 nd->nd_flag |= ND_NOMOREDATA;
844
845 /*
846 * If the first op is Putfh, throw its results away
847 * and toss the op# and status for the first op.
848 */
849 if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) {
850 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
851 i = fxdr_unsigned(int, *tl++);
852 j = fxdr_unsigned(int, *tl);
853 /*
854 * All Compounds that do an Op that must
855 * be in sequence consist of NFSV4OP_PUTFH
856 * followed by one of these. As such, we
857 * can determine if the seqid# should be
858 * incremented, here.
859 */
860 if ((i == NFSV4OP_OPEN ||
861 i == NFSV4OP_OPENCONFIRM ||
862 i == NFSV4OP_OPENDOWNGRADE ||
863 i == NFSV4OP_CLOSE ||
864 i == NFSV4OP_LOCK ||
865 i == NFSV4OP_LOCKU) &&
866 (j == 0 ||
867 (j != NFSERR_STALECLIENTID &&
868 j != NFSERR_STALESTATEID &&
869 j != NFSERR_BADSTATEID &&
870 j != NFSERR_BADSEQID &&
871 j != NFSERR_BADXDR &&
872 j != NFSERR_RESOURCE &&
873 j != NFSERR_NOFILEHANDLE)))
874 nd->nd_flag |= ND_INCRSEQID;
875 /*
876 * If the first op's status is non-zero, mark
877 * that there is no more data to process.
878 */
879 if (j)
880 nd->nd_flag |= ND_NOMOREDATA;
881 }
882
883 /*
884 * If R_DONTRECOVER is set, replace the stale error
885 * reply, so that recovery isn't initiated.
886 */
887 if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
888 nd->nd_repstat == NFSERR_STALESTATEID) &&
889 rep != NULL && (rep->r_flags & R_DONTRECOVER))
890 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
891 }
892 }
893
894 #ifdef KDTRACE_HOOKS
895 if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
896 uint32_t probe_id;
897 int probe_procnum;
898
899 if (nd->nd_flag & ND_NFSV4) {
900 probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
901 probe_procnum = nd->nd_procnum;
902 } else if (nd->nd_flag & ND_NFSV3) {
903 probe_id = nfscl_nfs3_done_probes[procnum];
904 probe_procnum = procnum;
905 } else {
906 probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
907 probe_procnum = procnum;
908 }
909 if (probe_id != 0)
910 (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
911 nd->nd_mreq, cred, probe_procnum, 0);
912 }
913 #endif
914
915 m_freem(nd->nd_mreq);
916 AUTH_DESTROY(auth);
917 if (rep != NULL)
918 FREE((caddr_t)rep, M_NFSDREQ);
919 if (set_sigset)
920 newnfs_restore_sigmask(td, &oldset);
921 return (0);
922 nfsmout:
923 mbuf_freem(nd->nd_mrep);
924 mbuf_freem(nd->nd_mreq);
925 AUTH_DESTROY(auth);
926 if (rep != NULL)
927 FREE((caddr_t)rep, M_NFSDREQ);
928 if (set_sigset)
929 newnfs_restore_sigmask(td, &oldset);
930 return (error);
931 }
932
933 /*
934 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
935 * wait for all requests to complete. This is used by forced unmounts
936 * to terminate any outstanding RPCs.
937 */
938 int
939 newnfs_nmcancelreqs(struct nfsmount *nmp)
940 {
941
942 if (nmp->nm_sockreq.nr_client != NULL)
943 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
944 return (0);
945 }
946
947 /*
948 * Any signal that can interrupt an NFS operation in an intr mount
949 * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
950 */
951 int newnfs_sig_set[] = {
952 SIGINT,
953 SIGTERM,
954 SIGHUP,
955 SIGKILL,
956 SIGQUIT
957 };
958
959 /*
960 * Check to see if one of the signals in our subset is pending on
961 * the process (in an intr mount).
962 */
963 static int
964 nfs_sig_pending(sigset_t set)
965 {
966 int i;
967
968 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++)
969 if (SIGISMEMBER(set, newnfs_sig_set[i]))
970 return (1);
971 return (0);
972 }
973
974 /*
975 * The set/restore sigmask functions are used to (temporarily) overwrite
976 * the thread td_sigmask during an RPC call (for example). These are also
977 * used in other places in the NFS client that might tsleep().
978 */
979 void
980 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
981 {
982 sigset_t newset;
983 int i;
984 struct proc *p;
985
986 SIGFILLSET(newset);
987 if (td == NULL)
988 td = curthread; /* XXX */
989 p = td->td_proc;
990 /* Remove the NFS set of signals from newset */
991 PROC_LOCK(p);
992 mtx_lock(&p->p_sigacts->ps_mtx);
993 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) {
994 /*
995 * But make sure we leave the ones already masked
996 * by the process, ie. remove the signal from the
997 * temporary signalmask only if it wasn't already
998 * in p_sigmask.
999 */
1000 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1001 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1002 SIGDELSET(newset, newnfs_sig_set[i]);
1003 }
1004 mtx_unlock(&p->p_sigacts->ps_mtx);
1005 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1006 SIGPROCMASK_PROC_LOCKED);
1007 PROC_UNLOCK(p);
1008 }
1009
1010 void
1011 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1012 {
1013 if (td == NULL)
1014 td = curthread; /* XXX */
1015 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1016 }
1017
1018 /*
1019 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1020 * old one after msleep() returns.
1021 */
1022 int
1023 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1024 {
1025 sigset_t oldset;
1026 int error;
1027 struct proc *p;
1028
1029 if ((priority & PCATCH) == 0)
1030 return msleep(ident, mtx, priority, wmesg, timo);
1031 if (td == NULL)
1032 td = curthread; /* XXX */
1033 newnfs_set_sigmask(td, &oldset);
1034 error = msleep(ident, mtx, priority, wmesg, timo);
1035 newnfs_restore_sigmask(td, &oldset);
1036 p = td->td_proc;
1037 return (error);
1038 }
1039
1040 /*
1041 * Test for a termination condition pending on the process.
1042 * This is used for NFSMNT_INT mounts.
1043 */
1044 int
1045 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1046 {
1047 struct proc *p;
1048 sigset_t tmpset;
1049
1050 /* Terminate all requests while attempting a forced unmount. */
1051 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
1052 return (EIO);
1053 if (!(nmp->nm_flag & NFSMNT_INT))
1054 return (0);
1055 if (td == NULL)
1056 return (0);
1057 p = td->td_proc;
1058 PROC_LOCK(p);
1059 tmpset = p->p_siglist;
1060 SIGSETOR(tmpset, td->td_siglist);
1061 SIGSETNAND(tmpset, td->td_sigmask);
1062 mtx_lock(&p->p_sigacts->ps_mtx);
1063 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1064 mtx_unlock(&p->p_sigacts->ps_mtx);
1065 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1066 && nfs_sig_pending(tmpset)) {
1067 PROC_UNLOCK(p);
1068 return (EINTR);
1069 }
1070 PROC_UNLOCK(p);
1071 return (0);
1072 }
1073
1074 static int
1075 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1076 {
1077 struct proc *p;
1078
1079 p = td ? td->td_proc : NULL;
1080 if (error) {
1081 tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n",
1082 server, msg, error);
1083 } else {
1084 tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg);
1085 }
1086 return (0);
1087 }
1088
1089 static void
1090 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1091 int error, int flags)
1092 {
1093 if (nmp == NULL)
1094 return;
1095 mtx_lock(&nmp->nm_mtx);
1096 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1097 nmp->nm_state |= NFSSTA_TIMEO;
1098 mtx_unlock(&nmp->nm_mtx);
1099 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1100 VQ_NOTRESP, 0);
1101 } else
1102 mtx_unlock(&nmp->nm_mtx);
1103 mtx_lock(&nmp->nm_mtx);
1104 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1105 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1106 mtx_unlock(&nmp->nm_mtx);
1107 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1108 VQ_NOTRESPLOCK, 0);
1109 } else
1110 mtx_unlock(&nmp->nm_mtx);
1111 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1112 }
1113
1114 static void
1115 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1116 int flags, int tprintfmsg)
1117 {
1118 if (nmp == NULL)
1119 return;
1120 if (tprintfmsg) {
1121 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1122 }
1123
1124 mtx_lock(&nmp->nm_mtx);
1125 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1126 nmp->nm_state &= ~NFSSTA_TIMEO;
1127 mtx_unlock(&nmp->nm_mtx);
1128 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1129 VQ_NOTRESP, 1);
1130 } else
1131 mtx_unlock(&nmp->nm_mtx);
1132
1133 mtx_lock(&nmp->nm_mtx);
1134 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1135 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1136 mtx_unlock(&nmp->nm_mtx);
1137 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1138 VQ_NOTRESPLOCK, 1);
1139 } else
1140 mtx_unlock(&nmp->nm_mtx);
1141 }
1142
Cache object: d7ee28d6fed7422c374b04a4ec1ab89b
|