1/*
2 * Copyright (c) 2000-2015 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28/* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
29/*
30 * Copyright (c) 1989, 1991, 1993, 1995
31 * The Regents of the University of California. All rights reserved.
32 *
33 * This code is derived from software contributed to Berkeley by
34 * Rick Macklem at The University of Guelph.
35 *
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
38 * are met:
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
44 * 3. All advertising materials mentioning features or use of this software
45 * must display the following acknowledgement:
46 * This product includes software developed by the University of
47 * California, Berkeley and its contributors.
48 * 4. Neither the name of the University nor the names of its contributors
49 * may be used to endorse or promote products derived from this software
50 * without specific prior written permission.
51 *
52 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
53 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
56 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62 * SUCH DAMAGE.
63 *
64 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95
65 * FreeBSD-Id: nfs_socket.c,v 1.30 1997/10/28 15:59:07 bde Exp $
66 */
67
68/*
69 * Socket operations for use by nfs
70 */
71
72#include <sys/param.h>
73#include <sys/systm.h>
74#include <sys/proc.h>
75#include <sys/signalvar.h>
76#include <sys/kauth.h>
77#include <sys/mount_internal.h>
78#include <sys/kernel.h>
79#include <sys/kpi_mbuf.h>
80#include <sys/malloc.h>
81#include <sys/vnode.h>
82#include <sys/domain.h>
83#include <sys/protosw.h>
84#include <sys/socket.h>
85#include <sys/syslog.h>
86#include <sys/tprintf.h>
87#include <libkern/OSAtomic.h>
88
89#include <sys/time.h>
90#include <kern/clock.h>
91#include <kern/task.h>
92#include <kern/thread.h>
93#include <kern/thread_call.h>
94#include <sys/user.h>
95#include <sys/acct.h>
96
97#include <netinet/in.h>
98#include <netinet/tcp.h>
99
100#include <nfs/rpcv2.h>
101#include <nfs/krpc.h>
102#include <nfs/nfsproto.h>
103#include <nfs/nfs.h>
104#include <nfs/xdr_subs.h>
105#include <nfs/nfsm_subs.h>
106#include <nfs/nfs_gss.h>
107#include <nfs/nfsmount.h>
108#include <nfs/nfsnode.h>
109
110#define NFS_SOCK_DBG(...) NFS_DBG(NFS_FAC_SOCK, 7, ## __VA_ARGS__)
111
112/* XXX */
113boolean_t current_thread_aborted(void);
114kern_return_t thread_terminate(thread_t);
115
116
117#if NFSSERVER
118int nfsrv_sock_max_rec_queue_length = 128; /* max # RPC records queued on (UDP) socket */
119
120int nfsrv_getstream(struct nfsrv_sock *,int);
121int nfsrv_getreq(struct nfsrv_descript *);
122extern int nfsv3_procid[NFS_NPROCS];
123#endif /* NFSSERVER */
124
125/*
126 * compare two sockaddr structures
127 */
128int
129nfs_sockaddr_cmp(struct sockaddr *sa1, struct sockaddr *sa2)
130{
131 if (!sa1)
132 return (-1);
133 if (!sa2)
134 return (1);
135 if (sa1->sa_family != sa2->sa_family)
136 return ((sa1->sa_family < sa2->sa_family) ? -1 : 1);
137 if (sa1->sa_len != sa2->sa_len)
138 return ((sa1->sa_len < sa2->sa_len) ? -1 : 1);
139 if (sa1->sa_family == AF_INET)
140 return (bcmp(&((struct sockaddr_in*)sa1)->sin_addr,
141 &((struct sockaddr_in*)sa2)->sin_addr, sizeof(((struct sockaddr_in*)sa1)->sin_addr)));
142 if (sa1->sa_family == AF_INET6)
143 return (bcmp(&((struct sockaddr_in6*)sa1)->sin6_addr,
144 &((struct sockaddr_in6*)sa2)->sin6_addr, sizeof(((struct sockaddr_in6*)sa1)->sin6_addr)));
145 return (-1);
146}
147
148#if NFSCLIENT
149
150int nfs_connect_search_new_socket(struct nfsmount *, struct nfs_socket_search *, struct timeval *);
151int nfs_connect_search_socket_connect(struct nfsmount *, struct nfs_socket *, int);
152int nfs_connect_search_ping(struct nfsmount *, struct nfs_socket *, struct timeval *);
153void nfs_connect_search_socket_found(struct nfsmount *, struct nfs_socket_search *, struct nfs_socket *);
154void nfs_connect_search_socket_reap(struct nfsmount *, struct nfs_socket_search *, struct timeval *);
155int nfs_connect_search_check(struct nfsmount *, struct nfs_socket_search *, struct timeval *);
156int nfs_reconnect(struct nfsmount *);
157int nfs_connect_setup(struct nfsmount *);
158void nfs_mount_sock_thread(void *, wait_result_t);
159void nfs_udp_rcv(socket_t, void*, int);
160void nfs_tcp_rcv(socket_t, void*, int);
161void nfs_sock_poke(struct nfsmount *);
162void nfs_request_match_reply(struct nfsmount *, mbuf_t);
163void nfs_reqdequeue(struct nfsreq *);
164void nfs_reqbusy(struct nfsreq *);
165struct nfsreq *nfs_reqnext(struct nfsreq *);
166int nfs_wait_reply(struct nfsreq *);
167void nfs_softterm(struct nfsreq *);
168int nfs_can_squish(struct nfsmount *);
169int nfs_is_squishy(struct nfsmount *);
170int nfs_is_dead(int, struct nfsmount *);
171
172/*
173 * Estimate rto for an nfs rpc sent via. an unreliable datagram.
174 * Use the mean and mean deviation of rtt for the appropriate type of rpc
175 * for the frequent rpcs and a default for the others.
176 * The justification for doing "other" this way is that these rpcs
177 * happen so infrequently that timer est. would probably be stale.
178 * Also, since many of these rpcs are
179 * non-idempotent, a conservative timeout is desired.
180 * getattr, lookup - A+2D
181 * read, write - A+4D
182 * other - nm_timeo
183 */
184#define NFS_RTO(n, t) \
185 ((t) == 0 ? (n)->nm_timeo : \
186 ((t) < 3 ? \
187 (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
188 ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
189#define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
190#define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
191
192/*
193 * Defines which timer to use for the procnum.
194 * 0 - default
195 * 1 - getattr
196 * 2 - lookup
197 * 3 - read
198 * 4 - write
199 */
200static int proct[NFS_NPROCS] = {
201 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0
202};
203
204/*
205 * There is a congestion window for outstanding rpcs maintained per mount
206 * point. The cwnd size is adjusted in roughly the way that:
207 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
208 * SIGCOMM '88". ACM, August 1988.
209 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
210 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
211 * of rpcs is in progress.
212 * (The sent count and cwnd are scaled for integer arith.)
213 * Variants of "slow start" were tried and were found to be too much of a
214 * performance hit (ave. rtt 3 times larger),
215 * I suspect due to the large rtt that nfs rpcs have.
216 */
217#define NFS_CWNDSCALE 256
218#define NFS_MAXCWND (NFS_CWNDSCALE * 32)
219static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
220
221/*
222 * Increment location index to next address/server/location.
223 */
224void
225nfs_location_next(struct nfs_fs_locations *nlp, struct nfs_location_index *nlip)
226{
227 uint8_t loc = nlip->nli_loc;
228 uint8_t serv = nlip->nli_serv;
229 uint8_t addr = nlip->nli_addr;
230
231 /* move to next address */
232 addr++;
233 if (addr >= nlp->nl_locations[loc]->nl_servers[serv]->ns_addrcount) {
234 /* no more addresses on current server, go to first address of next server */
235next_server:
236 addr = 0;
237 serv++;
238 if (serv >= nlp->nl_locations[loc]->nl_servcount) {
239 /* no more servers on current location, go to first server of next location */
240 serv = 0;
241 loc++;
242 if (loc >= nlp->nl_numlocs)
243 loc = 0; /* after last location, wrap back around to first location */
244 }
245 }
246 /*
247 * It's possible for this next server to not have any addresses.
248 * Check for that here and go to the next server.
249 * But bail out if we've managed to come back around to the original
250 * location that was passed in. (That would mean no servers had any
251 * addresses. And we don't want to spin here forever.)
252 */
253 if ((loc == nlip->nli_loc) && (serv == nlip->nli_serv) && (addr == nlip->nli_addr))
254 return;
255 if (addr >= nlp->nl_locations[loc]->nl_servers[serv]->ns_addrcount)
256 goto next_server;
257
258 nlip->nli_loc = loc;
259 nlip->nli_serv = serv;
260 nlip->nli_addr = addr;
261}
262
263/*
264 * Compare two location indices.
265 */
266int
267nfs_location_index_cmp(struct nfs_location_index *nlip1, struct nfs_location_index *nlip2)
268{
269 if (nlip1->nli_loc != nlip2->nli_loc)
270 return (nlip1->nli_loc - nlip2->nli_loc);
271 if (nlip1->nli_serv != nlip2->nli_serv)
272 return (nlip1->nli_serv - nlip2->nli_serv);
273 return (nlip1->nli_addr - nlip2->nli_addr);
274}
275
276/*
277 * Get the mntfromname (or path portion only) for a given location.
278 */
279void
280nfs_location_mntfromname(struct nfs_fs_locations *locs, struct nfs_location_index idx, char *s, int size, int pathonly)
281{
282 struct nfs_fs_location *fsl = locs->nl_locations[idx.nli_loc];
283 char *p;
284 int cnt, i;
285
286 p = s;
287 if (!pathonly) {
288 cnt = snprintf(p, size, "%s:", fsl->nl_servers[idx.nli_serv]->ns_name);
289 p += cnt;
290 size -= cnt;
291 }
292 if (fsl->nl_path.np_compcount == 0) {
293 /* mounting root export on server */
294 if (size > 0) {
295 *p++ = '/';
296 *p++ = '\0';
297 }
298 return;
299 }
300 /* append each server path component */
301 for (i=0; (size > 0) && (i < (int)fsl->nl_path.np_compcount); i++) {
302 cnt = snprintf(p, size, "/%s", fsl->nl_path.np_components[i]);
303 p += cnt;
304 size -= cnt;
305 }
306}
307
308/*
309 * NFS client connect socket upcall.
310 * (Used only during socket connect/search.)
311 */
312void
313nfs_connect_upcall(socket_t so, void *arg, __unused int waitflag)
314{
315 struct nfs_socket *nso = arg;
316 size_t rcvlen;
317 mbuf_t m;
318 int error = 0, recv = 1;
319
320 if (nso->nso_flags & NSO_CONNECTING) {
321 NFS_SOCK_DBG("nfs connect - socket %p upcall - connecting\n", nso);
322 wakeup(nso->nso_wake);
323 return;
324 }
325
326 lck_mtx_lock(&nso->nso_lock);
327 if ((nso->nso_flags & (NSO_UPCALL|NSO_DISCONNECTING|NSO_DEAD)) || !(nso->nso_flags & NSO_PINGING)) {
328 NFS_SOCK_DBG("nfs connect - socket %p upcall - nevermind\n", nso);
329 lck_mtx_unlock(&nso->nso_lock);
330 return;
331 }
332 NFS_SOCK_DBG("nfs connect - socket %p upcall\n", nso);
333 nso->nso_flags |= NSO_UPCALL;
334
335 /* loop while we make error-free progress */
336 while (!error && recv) {
337 /* make sure we're still interested in this socket */
338 if (nso->nso_flags & (NSO_DISCONNECTING|NSO_DEAD))
339 break;
340 lck_mtx_unlock(&nso->nso_lock);
341 m = NULL;
342 if (nso->nso_sotype == SOCK_STREAM) {
343 error = nfs_rpc_record_read(so, &nso->nso_rrs, MSG_DONTWAIT, &recv, &m);
344 } else {
345 rcvlen = 1000000;
346 error = sock_receivembuf(so, NULL, &m, MSG_DONTWAIT, &rcvlen);
347 recv = m ? 1 : 0;
348 }
349 lck_mtx_lock(&nso->nso_lock);
350 if (m) {
351 /* match response with request */
352 struct nfsm_chain nmrep;
353 uint32_t reply = 0, rxid = 0, verf_type, verf_len;
354 uint32_t reply_status, rejected_status, accepted_status;
355
356 nfsm_chain_dissect_init(error, &nmrep, m);
357 nfsm_chain_get_32(error, &nmrep, rxid);
358 nfsm_chain_get_32(error, &nmrep, reply);
359 if (!error && ((reply != RPC_REPLY) || (rxid != nso->nso_pingxid)))
360 error = EBADRPC;
361 nfsm_chain_get_32(error, &nmrep, reply_status);
362 if (!error && (reply_status == RPC_MSGDENIED)) {
363 nfsm_chain_get_32(error, &nmrep, rejected_status);
364 if (!error)
365 error = (rejected_status == RPC_MISMATCH) ? ERPCMISMATCH : EACCES;
366 }
367 nfsm_chain_get_32(error, &nmrep, verf_type); /* verifier flavor */
368 nfsm_chain_get_32(error, &nmrep, verf_len); /* verifier length */
369 nfsmout_if(error);
370 if (verf_len)
371 nfsm_chain_adv(error, &nmrep, nfsm_rndup(verf_len));
372 nfsm_chain_get_32(error, &nmrep, accepted_status);
373 nfsmout_if(error);
374 if ((accepted_status == RPC_PROGMISMATCH) && !nso->nso_version) {
375 uint32_t minvers, maxvers;
376 nfsm_chain_get_32(error, &nmrep, minvers);
377 nfsm_chain_get_32(error, &nmrep, maxvers);
378 nfsmout_if(error);
379 if (nso->nso_protocol == PMAPPROG) {
380 if ((minvers > RPCBVERS4) || (maxvers < PMAPVERS))
381 error = EPROGMISMATCH;
382 else if ((nso->nso_saddr->sa_family == AF_INET) &&
383 (PMAPVERS >= minvers) && (PMAPVERS <= maxvers))
384 nso->nso_version = PMAPVERS;
385 else if (nso->nso_saddr->sa_family == AF_INET6) {
386 if ((RPCBVERS4 >= minvers) && (RPCBVERS4 <= maxvers))
387 nso->nso_version = RPCBVERS4;
388 else if ((RPCBVERS3 >= minvers) && (RPCBVERS3 <= maxvers))
389 nso->nso_version = RPCBVERS3;
390 }
391 } else if (nso->nso_protocol == NFS_PROG) {
392 int vers;
393
394 /*
395 * N.B. Both portmapper and rpcbind V3 are happy to return
396 * addresses for other versions than the one you ask (getport or
397 * getaddr) and thus we may have fallen to this code path. So if
398 * we get a version that we support, use highest supported
399 * version. This assumes that the server supports all versions
400 * between minvers and maxvers. Note for IPv6 we will try and
401 * use rpcbind V4 which has getversaddr and we should not get
402 * here if that was successful.
403 */
404 for (vers = nso->nso_nfs_max_vers; vers >= (int)nso->nso_nfs_min_vers; vers--) {
405 if (vers >= (int)minvers && vers <= (int)maxvers)
406 break;
407 }
408 nso->nso_version = (vers < (int)nso->nso_nfs_min_vers) ? 0 : vers;
409 }
410 if (!error && nso->nso_version)
411 accepted_status = RPC_SUCCESS;
412 }
413 if (!error) {
414 switch (accepted_status) {
415 case RPC_SUCCESS:
416 error = 0;
417 break;
418 case RPC_PROGUNAVAIL:
419 error = EPROGUNAVAIL;
420 break;
421 case RPC_PROGMISMATCH:
422 error = EPROGMISMATCH;
423 break;
424 case RPC_PROCUNAVAIL:
425 error = EPROCUNAVAIL;
426 break;
427 case RPC_GARBAGE:
428 error = EBADRPC;
429 break;
430 case RPC_SYSTEM_ERR:
431 default:
432 error = EIO;
433 break;
434 }
435 }
436nfsmout:
437 nso->nso_flags &= ~NSO_PINGING;
438 if (error) {
439 nso->nso_error = error;
440 nso->nso_flags |= NSO_DEAD;
441 } else {
442 nso->nso_flags |= NSO_VERIFIED;
443 }
444 mbuf_freem(m);
445 /* wake up search thread */
446 wakeup(nso->nso_wake);
447 break;
448 }
449 }
450
451 nso->nso_flags &= ~NSO_UPCALL;
452 if ((error != EWOULDBLOCK) && (error || !recv)) {
453 /* problems with the socket... */
454 nso->nso_error = error ? error : EPIPE;
455 nso->nso_flags |= NSO_DEAD;
456 wakeup(nso->nso_wake);
457 }
458 if (nso->nso_flags & NSO_DISCONNECTING)
459 wakeup(&nso->nso_flags);
460 lck_mtx_unlock(&nso->nso_lock);
461}
462
463/*
464 * Create/initialize an nfs_socket structure.
465 */
466int
467nfs_socket_create(
468 struct nfsmount *nmp,
469 struct sockaddr *sa,
470 int sotype,
471 in_port_t port,
472 uint32_t protocol,
473 uint32_t vers,
474 int resvport,
475 struct nfs_socket **nsop)
476{
477 struct nfs_socket *nso;
478 struct timeval now;
479 int error;
480#ifdef NFS_SOCKET_DEBUGGING
481 char naddr[MAX_IPv6_STR_LEN];
482 void *sinaddr;
483
484 if (sa->sa_family == AF_INET)
485 sinaddr = &((struct sockaddr_in*)sa)->sin_addr;
486 else
487 sinaddr = &((struct sockaddr_in6*)sa)->sin6_addr;
488 if (inet_ntop(sa->sa_family, sinaddr, naddr, sizeof(naddr)) != naddr)
489 strlcpy(naddr, "<unknown>", sizeof(naddr));
490#else
491 char naddr[1] = { 0 };
492#endif
493
494 *nsop = NULL;
495
496 /* Create the socket. */
497 MALLOC(nso, struct nfs_socket *, sizeof(struct nfs_socket), M_TEMP, M_WAITOK|M_ZERO);
498 if (nso)
499 MALLOC(nso->nso_saddr, struct sockaddr *, sa->sa_len, M_SONAME, M_WAITOK|M_ZERO);
500 if (!nso || !nso->nso_saddr) {
501 if (nso)
502 FREE(nso, M_TEMP);
503 return (ENOMEM);
504 }
505 lck_mtx_init(&nso->nso_lock, nfs_request_grp, LCK_ATTR_NULL);
506 nso->nso_sotype = sotype;
507 if (nso->nso_sotype == SOCK_STREAM)
508 nfs_rpc_record_state_init(&nso->nso_rrs);
509 microuptime(&now);
510 nso->nso_timestamp = now.tv_sec;
511 bcopy(sa, nso->nso_saddr, sa->sa_len);
512 if (sa->sa_family == AF_INET)
513 ((struct sockaddr_in*)nso->nso_saddr)->sin_port = htons(port);
514 else if (sa->sa_family == AF_INET6)
515 ((struct sockaddr_in6*)nso->nso_saddr)->sin6_port = htons(port);
516 nso->nso_protocol = protocol;
517 nso->nso_version = vers;
518 nso->nso_nfs_min_vers = PVER2MAJOR(nmp->nm_min_vers);
519 nso->nso_nfs_max_vers = PVER2MAJOR(nmp->nm_max_vers);
520
521 error = sock_socket(sa->sa_family, nso->nso_sotype, 0, NULL, NULL, &nso->nso_so);
522
523 /* Some servers require that the client port be a reserved port number. */
524 if (!error && resvport && ((sa->sa_family == AF_INET) || (sa->sa_family == AF_INET6))) {
525 struct sockaddr_storage ss;
526 int level = (sa->sa_family == AF_INET) ? IPPROTO_IP : IPPROTO_IPV6;
527 int optname = (sa->sa_family == AF_INET) ? IP_PORTRANGE : IPV6_PORTRANGE;
528 int portrange = IP_PORTRANGE_LOW;
529
530 error = sock_setsockopt(nso->nso_so, level, optname, &portrange, sizeof(portrange));
531 if (!error) { /* bind now to check for failure */
532 ss.ss_len = sa->sa_len;
533 ss.ss_family = sa->sa_family;
534 if (ss.ss_family == AF_INET) {
535 ((struct sockaddr_in*)&ss)->sin_addr.s_addr = INADDR_ANY;
536 ((struct sockaddr_in*)&ss)->sin_port = htons(0);
537 } else if (ss.ss_family == AF_INET6) {
538 ((struct sockaddr_in6*)&ss)->sin6_addr = in6addr_any;
539 ((struct sockaddr_in6*)&ss)->sin6_port = htons(0);
540 } else {
541 error = EINVAL;
542 }
543 if (!error)
544 error = sock_bind(nso->nso_so, (struct sockaddr*)&ss);
545 }
546 }
547
548 if (error) {
549 NFS_SOCK_DBG("nfs connect %s error %d creating socket %p %s type %d%s port %d prot %d %d\n",
550 vfs_statfs(nmp->nm_mountp)->f_mntfromname, error, nso, naddr, sotype,
551 resvport ? "r" : "", port, protocol, vers);
552 nfs_socket_destroy(nso);
553 } else {
554 NFS_SOCK_DBG("nfs connect %s created socket %p %s type %d%s port %d prot %d %d\n",
555 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, naddr,
556 sotype, resvport ? "r" : "", port, protocol, vers);
557 *nsop = nso;
558 }
559 return (error);
560}
561
562/*
563 * Destroy an nfs_socket structure.
564 */
565void
566nfs_socket_destroy(struct nfs_socket *nso)
567{
568 struct timespec ts = { 4, 0 };
569
570 lck_mtx_lock(&nso->nso_lock);
571 nso->nso_flags |= NSO_DISCONNECTING;
572 if (nso->nso_flags & NSO_UPCALL) /* give upcall a chance to complete */
573 msleep(&nso->nso_flags, &nso->nso_lock, PZERO-1, "nfswaitupcall", &ts);
574 lck_mtx_unlock(&nso->nso_lock);
575 sock_shutdown(nso->nso_so, SHUT_RDWR);
576 sock_close(nso->nso_so);
577 if (nso->nso_sotype == SOCK_STREAM)
578 nfs_rpc_record_state_cleanup(&nso->nso_rrs);
579 lck_mtx_destroy(&nso->nso_lock, nfs_request_grp);
580 if (nso->nso_saddr)
581 FREE(nso->nso_saddr, M_SONAME);
582 if (nso->nso_saddr2)
583 FREE(nso->nso_saddr2, M_SONAME);
584 NFS_SOCK_DBG("nfs connect - socket %p destroyed\n", nso);
585 FREE(nso, M_TEMP);
586}
587
588/*
589 * Set common socket options on an nfs_socket.
590 */
591void
592nfs_socket_options(struct nfsmount *nmp, struct nfs_socket *nso)
593{
594 /*
595 * Set socket send/receive timeouts
596 * - Receive timeout shouldn't matter because most receives are performed
597 * in the socket upcall non-blocking.
598 * - Send timeout should allow us to react to a blocked socket.
599 * Soft mounts will want to abort sooner.
600 */
601 struct timeval timeo;
602 int on = 1, proto;
603
604 timeo.tv_usec = 0;
605 timeo.tv_sec = (NMFLAG(nmp, SOFT) || nfs_can_squish(nmp)) ? 5 : 60;
606 sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo));
607 sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo));
608 if (nso->nso_sotype == SOCK_STREAM) {
609 /* Assume that SOCK_STREAM always requires a connection */
610 sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_KEEPALIVE, &on, sizeof(on));
611 /* set nodelay for TCP */
612 sock_gettype(nso->nso_so, NULL, NULL, &proto);
613 if (proto == IPPROTO_TCP)
614 sock_setsockopt(nso->nso_so, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on));
615 }
616 if (nso->nso_sotype == SOCK_DGRAM) { /* set socket buffer sizes for UDP */
617 int reserve = NFS_UDPSOCKBUF;
618 sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_SNDBUF, &reserve, sizeof(reserve));
619 sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_RCVBUF, &reserve, sizeof(reserve));
620 }
621 /* set SO_NOADDRERR to detect network changes ASAP */
622 sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on));
623 /* just playin' it safe with upcalls */
624 sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on));
625 /* socket should be interruptible if the mount is */
626 if (!NMFLAG(nmp, INTR))
627 sock_nointerrupt(nso->nso_so, 1);
628}
629
630/*
631 * Release resources held in an nfs_socket_search.
632 */
633void
634nfs_socket_search_cleanup(struct nfs_socket_search *nss)
635{
636 struct nfs_socket *nso, *nsonext;
637
638 TAILQ_FOREACH_SAFE(nso, &nss->nss_socklist, nso_link, nsonext) {
639 TAILQ_REMOVE(&nss->nss_socklist, nso, nso_link);
640 nss->nss_sockcnt--;
641 nfs_socket_destroy(nso);
642 }
643 if (nss->nss_sock) {
644 nfs_socket_destroy(nss->nss_sock);
645 nss->nss_sock = NULL;
646 }
647}
648
649/*
650 * Prefer returning certain errors over others.
651 * This function returns a ranking of the given error.
652 */
653int
654nfs_connect_error_class(int error)
655{
656 switch (error) {
657 case 0:
658 return (0);
659 case ETIMEDOUT:
660 case EAGAIN:
661 return (1);
662 case EPIPE:
663 case EADDRNOTAVAIL:
664 case ENETDOWN:
665 case ENETUNREACH:
666 case ENETRESET:
667 case ECONNABORTED:
668 case ECONNRESET:
669 case EISCONN:
670 case ENOTCONN:
671 case ESHUTDOWN:
672 case ECONNREFUSED:
673 case EHOSTDOWN:
674 case EHOSTUNREACH:
675 return (2);
676 case ERPCMISMATCH:
677 case EPROCUNAVAIL:
678 case EPROGMISMATCH:
679 case EPROGUNAVAIL:
680 return (3);
681 case EBADRPC:
682 return (4);
683 default:
684 return (5);
685 }
686}
687
688/*
689 * Make sure a socket search returns the best error.
690 */
691void
692nfs_socket_search_update_error(struct nfs_socket_search *nss, int error)
693{
694 if (nfs_connect_error_class(error) >= nfs_connect_error_class(nss->nss_error))
695 nss->nss_error = error;
696}
697
698/* nfs_connect_search_new_socket:
699 * Given a socket search structure for an nfs mount try to find a new socket from the set of addresses specified
700 * by nss.
701 *
702 * nss_last is set to -1 at initialization to indicate the first time. Its set to -2 if address was found but
703 * could not be used or if a socket timed out.
704 */
705int
706nfs_connect_search_new_socket(struct nfsmount *nmp, struct nfs_socket_search *nss, struct timeval *now)
707{
708 struct nfs_fs_location *fsl;
709 struct nfs_fs_server *fss;
710 struct sockaddr_storage ss;
711 struct nfs_socket *nso;
712 char *addrstr;
713 int error = 0;
714
715
716 NFS_SOCK_DBG("nfs connect %s nss_addrcnt = %d\n",
717 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nss->nss_addrcnt);
718
719 /*
720 * while there are addresses and:
721 * we have no sockets or
722 * the last address failed and did not produce a socket (nss_last < 0) or
723 * Its been a while (2 seconds) and we have less than the max number of concurrent sockets to search (4)
724 * then attempt to create a socket with the current address.
725 */
726 while (nss->nss_addrcnt > 0 && ((nss->nss_last < 0) || (nss->nss_sockcnt == 0) ||
727 ((nss->nss_sockcnt < 4) && (now->tv_sec >= (nss->nss_last + 2))))) {
728 if (nmp->nm_sockflags & NMSOCK_UNMOUNT)
729 return (EINTR);
730 /* Can we convert the address to a sockaddr? */
731 fsl = nmp->nm_locations.nl_locations[nss->nss_nextloc.nli_loc];
732 fss = fsl->nl_servers[nss->nss_nextloc.nli_serv];
733 addrstr = fss->ns_addresses[nss->nss_nextloc.nli_addr];
734 if (!nfs_uaddr2sockaddr(addrstr, (struct sockaddr*)&ss)) {
735 nfs_location_next(&nmp->nm_locations, &nss->nss_nextloc);
736 nss->nss_addrcnt -= 1;
737 nss->nss_last = -2;
738 continue;
739 }
740 /* Check that socket family is acceptable. */
741 if (nmp->nm_sofamily && (ss.ss_family != nmp->nm_sofamily)) {
742 nfs_location_next(&nmp->nm_locations, &nss->nss_nextloc);
743 nss->nss_addrcnt -= 1;
744 nss->nss_last = -2;
745 continue;
746 }
747
748 /* Create the socket. */
749 error = nfs_socket_create(nmp, (struct sockaddr*)&ss, nss->nss_sotype,
750 nss->nss_port, nss->nss_protocol, nss->nss_version,
751 ((nss->nss_protocol == NFS_PROG) && NMFLAG(nmp, RESVPORT)), &nso);
752 if (error)
753 return (error);
754
755 nso->nso_location = nss->nss_nextloc;
756 nso->nso_wake = nss;
757 error = sock_setupcall(nso->nso_so, nfs_connect_upcall, nso);
758 if (error) {
759 lck_mtx_lock(&nso->nso_lock);
760 nso->nso_error = error;
761 nso->nso_flags |= NSO_DEAD;
762 lck_mtx_unlock(&nso->nso_lock);
763 }
764
765 TAILQ_INSERT_TAIL(&nss->nss_socklist, nso, nso_link);
766 nss->nss_sockcnt++;
767 nfs_location_next(&nmp->nm_locations, &nss->nss_nextloc);
768 nss->nss_addrcnt -= 1;
769
770 nss->nss_last = now->tv_sec;
771 }
772
773 if (nss->nss_addrcnt == 0 && nss->nss_last < 0)
774 nss->nss_last = now->tv_sec;
775
776 return (error);
777}
778
779/*
780 * nfs_connect_search_socket_connect: Connect an nfs socket nso for nfsmount nmp.
781 * If successful set the socket options for the socket as require from the mount.
782 *
783 * Assumes: nso->nso_lock is held on entry and return.
784 */
785int
786nfs_connect_search_socket_connect(struct nfsmount *nmp, struct nfs_socket *nso, int verbose)
787{
788 int error;
789
790 if ((nso->nso_sotype != SOCK_STREAM) && NMFLAG(nmp, NOCONNECT)) {
791 /* no connection needed, just say it's already connected */
792 NFS_SOCK_DBG("nfs connect %s UDP socket %p noconnect\n",
793 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso);
794 nso->nso_flags |= NSO_CONNECTED;
795 nfs_socket_options(nmp, nso);
796 return (1); /* Socket is connected and setup */
797 } else if (!(nso->nso_flags & NSO_CONNECTING)) {
798 /* initiate the connection */
799 nso->nso_flags |= NSO_CONNECTING;
800 lck_mtx_unlock(&nso->nso_lock);
801 NFS_SOCK_DBG("nfs connect %s connecting socket %p\n",
802 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso);
803 error = sock_connect(nso->nso_so, nso->nso_saddr, MSG_DONTWAIT);
804 lck_mtx_lock(&nso->nso_lock);
805 if (error && (error != EINPROGRESS)) {
806 nso->nso_error = error;
807 nso->nso_flags |= NSO_DEAD;
808 return (0);
809 }
810 }
811 if (nso->nso_flags & NSO_CONNECTING) {
812 /* check the connection */
813 if (sock_isconnected(nso->nso_so)) {
814 NFS_SOCK_DBG("nfs connect %s socket %p is connected\n",
815 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso);
816 nso->nso_flags &= ~NSO_CONNECTING;
817 nso->nso_flags |= NSO_CONNECTED;
818 nfs_socket_options(nmp, nso);
819 return (1); /* Socket is connected and setup */
820 } else {
821 int optlen = sizeof(error);
822 error = 0;
823 sock_getsockopt(nso->nso_so, SOL_SOCKET, SO_ERROR, &error, &optlen);
824 if (error) { /* we got an error on the socket */
825 NFS_SOCK_DBG("nfs connect %s socket %p connection error %d\n",
826 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, error);
827 if (verbose)
828 printf("nfs connect socket error %d for %s\n",
829 error, vfs_statfs(nmp->nm_mountp)->f_mntfromname);
830 nso->nso_error = error;
831 nso->nso_flags |= NSO_DEAD;
832 return (0);
833 }
834 }
835 }
836
837 return (0); /* Waiting to be connected */
838}
839
840/*
841 * nfs_connect_search_ping: Send a null proc on the nso socket.
842 */
843int
844nfs_connect_search_ping(struct nfsmount *nmp, struct nfs_socket *nso, struct timeval *now)
845{
846 /* initiate a NULL RPC request */
847 uint64_t xid = nso->nso_pingxid;
848 mbuf_t m, mreq = NULL;
849 struct msghdr msg;
850 size_t reqlen, sentlen;
851 uint32_t vers = nso->nso_version;
852 int error;
853
854 if (!vers) {
855 if (nso->nso_protocol == PMAPPROG)
856 vers = (nso->nso_saddr->sa_family == AF_INET) ? PMAPVERS : RPCBVERS4;
857 else if (nso->nso_protocol == NFS_PROG)
858 vers = PVER2MAJOR(nmp->nm_max_vers);
859 }
860 lck_mtx_unlock(&nso->nso_lock);
861 error = nfsm_rpchead2(nmp, nso->nso_sotype, nso->nso_protocol, vers, 0, RPCAUTH_SYS,
862 vfs_context_ucred(vfs_context_kernel()), NULL, NULL, &xid, &mreq);
863 lck_mtx_lock(&nso->nso_lock);
864 if (!error) {
865 nso->nso_flags |= NSO_PINGING;
866 nso->nso_pingxid = R_XID32(xid);
867 nso->nso_reqtimestamp = now->tv_sec;
868 bzero(&msg, sizeof(msg));
869 if ((nso->nso_sotype != SOCK_STREAM) && !sock_isconnected(nso->nso_so)) {
870 msg.msg_name = nso->nso_saddr;
871 msg.msg_namelen = nso->nso_saddr->sa_len;
872 }
873 for (reqlen=0, m=mreq; m; m = mbuf_next(m))
874 reqlen += mbuf_len(m);
875 lck_mtx_unlock(&nso->nso_lock);
876 error = sock_sendmbuf(nso->nso_so, &msg, mreq, 0, &sentlen);
877 NFS_SOCK_DBG("nfs connect %s verifying socket %p send rv %d\n",
878 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, error);
879 lck_mtx_lock(&nso->nso_lock);
880 if (!error && (sentlen != reqlen))
881 error = ETIMEDOUT;
882 }
883 if (error) {
884 nso->nso_error = error;
885 nso->nso_flags |= NSO_DEAD;
886 return (0);
887 }
888
889 return (1);
890}
891
892/*
893 * nfs_connect_search_socket_found: Take the found socket of the socket search list and assign it to the searched socket.
894 * Set the nfs socket protocol and version if needed.
895 */
896void
897nfs_connect_search_socket_found(struct nfsmount *nmp, struct nfs_socket_search *nss, struct nfs_socket *nso)
898{
899 NFS_SOCK_DBG("nfs connect %s socket %p verified\n",
900 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso);
901 if (!nso->nso_version) {
902 /* If the version isn't set, the default must have worked. */
903 if (nso->nso_protocol == PMAPPROG)
904 nso->nso_version = (nso->nso_saddr->sa_family == AF_INET) ? PMAPVERS : RPCBVERS4;
905 if (nso->nso_protocol == NFS_PROG)
906 nso->nso_version = PVER2MAJOR(nmp->nm_max_vers);
907 }
908 TAILQ_REMOVE(&nss->nss_socklist, nso, nso_link);
909 nss->nss_sockcnt--;
910 nss->nss_sock = nso;
911}
912
913/*
914 * nfs_connect_search_socket_reap: For each socket in the search list mark any timed out socket as dead and remove from
915 * the list. Dead socket are then destroyed.
916 */
917void
918nfs_connect_search_socket_reap(struct nfsmount *nmp __unused, struct nfs_socket_search *nss, struct timeval *now)
919{
920 struct nfs_socket *nso, *nsonext;
921
922 TAILQ_FOREACH_SAFE(nso, &nss->nss_socklist, nso_link, nsonext) {
923 lck_mtx_lock(&nso->nso_lock);
924 if (now->tv_sec >= (nso->nso_timestamp + nss->nss_timeo)) {
925 /* took too long */
926 NFS_SOCK_DBG("nfs connect %s socket %p timed out\n",
927 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso);
928 nso->nso_error = ETIMEDOUT;
929 nso->nso_flags |= NSO_DEAD;
930 }
931 if (!(nso->nso_flags & NSO_DEAD)) {
932 lck_mtx_unlock(&nso->nso_lock);
933 continue;
934 }
935 lck_mtx_unlock(&nso->nso_lock);
936 NFS_SOCK_DBG("nfs connect %s reaping socket %p %d\n",
937 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, nso->nso_error);
938 nfs_socket_search_update_error(nss, nso->nso_error);
939 TAILQ_REMOVE(&nss->nss_socklist, nso, nso_link);
940 nss->nss_sockcnt--;
941 nfs_socket_destroy(nso);
942 /* If there are more sockets to try, force the starting of another socket */
943 if (nss->nss_addrcnt > 0)
944 nss->nss_last = -2;
945 }
946}
947
948/*
949 * nfs_connect_search_check: Check on the status of search and wait for replies if needed.
950 */
951int
952nfs_connect_search_check(struct nfsmount *nmp, struct nfs_socket_search *nss, struct timeval *now)
953{
954 int error;
955
956 /* log a warning if connect is taking a while */
957 if (((now->tv_sec - nss->nss_timestamp) >= 8) && ((nss->nss_flags & (NSS_VERBOSE|NSS_WARNED)) == NSS_VERBOSE)) {
958 printf("nfs_connect: socket connect taking a while for %s\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname);
959 nss->nss_flags |= NSS_WARNED;
960 }
961 if (nmp->nm_sockflags & NMSOCK_UNMOUNT)
962 return (EINTR);
963 if ((error = nfs_sigintr(nmp, NULL, current_thread(), 0)))
964 return (error);
965
966 /* If we were succesfull at sending a ping, wait up to a second for a reply */
967 if (nss->nss_last >= 0)
968 tsleep(nss, PSOCK, "nfs_connect_search_wait", hz);
969
970 return (0);
971}
972
973
974/*
975 * Continue the socket search until we have something to report.
976 */
977int
978nfs_connect_search_loop(struct nfsmount *nmp, struct nfs_socket_search *nss)
979{
980 struct nfs_socket *nso;
981 struct timeval now;
982 int error;
983 int verbose = (nss->nss_flags & NSS_VERBOSE);
984
985loop:
986 microuptime(&now);
987 NFS_SOCK_DBG("nfs connect %s search %ld\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname, now.tv_sec);
988
989 /* add a new socket to the socket list if needed and available */
990 error = nfs_connect_search_new_socket(nmp, nss, &now);
991 if (error) {
992 NFS_SOCK_DBG("nfs connect returned %d\n", error);
993 return (error);
994 }
995
996 /* check each active socket on the list and try to push it along */
997 TAILQ_FOREACH(nso, &nss->nss_socklist, nso_link) {
998 lck_mtx_lock(&nso->nso_lock);
999
1000 /* If not connected connect it */
1001 if (!(nso->nso_flags & NSO_CONNECTED)) {
1002 if (!nfs_connect_search_socket_connect(nmp, nso, verbose)) {
1003 lck_mtx_unlock(&nso->nso_lock);
1004 continue;
1005 }
1006 }
1007
1008 /* If the socket hasn't been verified or in a ping, ping it. We also handle UDP retransmits */
1009 if (!(nso->nso_flags & (NSO_PINGING|NSO_VERIFIED)) ||
1010 ((nso->nso_sotype == SOCK_DGRAM) && (now.tv_sec >= nso->nso_reqtimestamp+2))) {
1011 if (!nfs_connect_search_ping(nmp, nso, &now)) {
1012 lck_mtx_unlock(&nso->nso_lock);
1013 continue;
1014 }
1015 }
1016
1017 /* Has the socket been verified by the up call routine? */
1018 if (nso->nso_flags & NSO_VERIFIED) {
1019 /* WOOHOO!! This socket looks good! */
1020 nfs_connect_search_socket_found(nmp, nss, nso);
1021 lck_mtx_unlock(&nso->nso_lock);
1022 break;
1023 }
1024 lck_mtx_unlock(&nso->nso_lock);
1025 }
1026
1027 /* Check for timed out sockets and mark as dead and then remove all dead sockets. */
1028 nfs_connect_search_socket_reap(nmp, nss, &now);
1029
1030 /*
1031 * Keep looping if we haven't found a socket yet and we have more
1032 * sockets to (continue to) try.
1033 */
1034 error = 0;
1035 if (!nss->nss_sock && (!TAILQ_EMPTY(&nss->nss_socklist) || nss->nss_addrcnt)) {
1036 error = nfs_connect_search_check(nmp, nss, &now);
1037 if (!error)
1038 goto loop;
1039 }
1040
1041 NFS_SOCK_DBG("nfs connect %s returning %d\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname, error);
1042 return (error);
1043}
1044
1045/*
1046 * Initialize a new NFS connection.
1047 *
1048 * Search for a location to connect a socket to and initialize the connection.
1049 *
1050 * An NFS mount may have multiple locations/servers/addresses available.
1051 * We attempt to connect to each one asynchronously and will start
1052 * several sockets in parallel if other locations are slow to answer.
1053 * We'll use the first NFS socket we can successfully set up.
1054 *
1055 * The search may involve contacting the portmapper service first.
1056 *
1057 * A mount's initial connection may require negotiating some parameters such
1058 * as socket type and NFS version.
1059 */
1060
1061int
1062nfs_connect(struct nfsmount *nmp, int verbose, int timeo)
1063{
1064 struct nfs_socket_search nss;
1065 struct nfs_socket *nso, *nsonfs;
1066 struct sockaddr_storage ss;
1067 struct sockaddr *saddr, *oldsaddr;
1068 sock_upcall upcall;
1069 struct timeval now, start;
1070 int error, savederror, nfsvers;
1071 int tryv4 = 1;
1072 uint8_t sotype = nmp->nm_sotype ? nmp->nm_sotype : SOCK_STREAM;
1073 fhandle_t *fh = NULL;
1074 char *path = NULL;
1075 in_port_t port;
1076 int addrtotal = 0;
1077
1078 /* paranoia... check that we have at least one address in the locations */
1079 uint32_t loc, serv;
1080 for (loc=0; loc < nmp->nm_locations.nl_numlocs; loc++) {
1081 for (serv=0; serv < nmp->nm_locations.nl_locations[loc]->nl_servcount; serv++) {
1082 addrtotal += nmp->nm_locations.nl_locations[loc]->nl_servers[serv]->ns_addrcount;
1083 if (nmp->nm_locations.nl_locations[loc]->nl_servers[serv]->ns_addrcount == 0)
1084 NFS_SOCK_DBG("nfs connect %s search, server %s has no addresses\n",
1085 vfs_statfs(nmp->nm_mountp)->f_mntfromname,
1086 nmp->nm_locations.nl_locations[loc]->nl_servers[serv]->ns_name);
1087 }
1088 }
1089
1090 if (addrtotal == 0) {
1091 NFS_SOCK_DBG("nfs connect %s search failed, no addresses\n",
1092 vfs_statfs(nmp->nm_mountp)->f_mntfromname);
1093 return (EINVAL);
1094 } else
1095 NFS_SOCK_DBG("nfs connect %s has %d addresses\n",
1096 vfs_statfs(nmp->nm_mountp)->f_mntfromname, addrtotal);
1097
1098 lck_mtx_lock(&nmp->nm_lock);
1099 nmp->nm_sockflags |= NMSOCK_CONNECTING;
1100 nmp->nm_nss = &nss;
1101 lck_mtx_unlock(&nmp->nm_lock);
1102 microuptime(&start);
1103 savederror = error = 0;
1104
1105tryagain:
1106 /* initialize socket search state */
1107 bzero(&nss, sizeof(nss));
1108 nss.nss_addrcnt = addrtotal;
1109 nss.nss_error = savederror;
1110 TAILQ_INIT(&nss.nss_socklist);
1111 nss.nss_sotype = sotype;
1112 nss.nss_startloc = nmp->nm_locations.nl_current;
1113 nss.nss_timestamp = start.tv_sec;
1114 nss.nss_timeo = timeo;
1115 if (verbose)
1116 nss.nss_flags |= NSS_VERBOSE;
1117
1118 /* First time connecting, we may need to negotiate some things */
1119 if (!(nmp->nm_sockflags & NMSOCK_HASCONNECTED)) {
1120 if (!nmp->nm_vers) {
1121 /* No NFS version specified... */
1122 if (!nmp->nm_nfsport || (!NM_OMATTR_GIVEN(nmp, FH) && !nmp->nm_mountport)) {
1123 if (PVER2MAJOR(nmp->nm_max_vers) >= NFS_VER4 && tryv4) {
1124 nss.nss_port = NFS_PORT;
1125 nss.nss_protocol = NFS_PROG;
1126 nss.nss_version = 4;
1127 nss.nss_flags |= NSS_FALLBACK2PMAP;
1128 } else {
1129 /* ...connect to portmapper first if we (may) need any ports. */
1130 nss.nss_port = PMAPPORT;
1131 nss.nss_protocol = PMAPPROG;
1132 nss.nss_version = 0;
1133 }
1134 } else {
1135 /* ...connect to NFS port first. */
1136 nss.nss_port = nmp->nm_nfsport;
1137 nss.nss_protocol = NFS_PROG;
1138 nss.nss_version = 0;
1139 }
1140 } else if (nmp->nm_vers >= NFS_VER4) {
1141 if (tryv4) {
1142 /* For NFSv4, we use the given (or default) port. */
1143 nss.nss_port = nmp->nm_nfsport ? nmp->nm_nfsport : NFS_PORT;
1144 nss.nss_protocol = NFS_PROG;
1145 nss.nss_version = 4;
1146 /*
1147 * set NSS_FALLBACK2PMAP here to pick up any non standard port
1148 * if no port is specified on the mount;
1149 * Note nm_vers is set so we will only try NFS_VER4.
1150 */
1151 if (!nmp->nm_nfsport)
1152 nss.nss_flags |= NSS_FALLBACK2PMAP;
1153 } else {
1154 nss.nss_port = PMAPPORT;
1155 nss.nss_protocol = PMAPPROG;
1156 nss.nss_version = 0;
1157 }
1158 } else {
1159 /* For NFSv3/v2... */
1160 if (!nmp->nm_nfsport || (!NM_OMATTR_GIVEN(nmp, FH) && !nmp->nm_mountport)) {
1161 /* ...connect to portmapper first if we need any ports. */
1162 nss.nss_port = PMAPPORT;
1163 nss.nss_protocol = PMAPPROG;
1164 nss.nss_version = 0;
1165 } else {
1166 /* ...connect to NFS port first. */
1167 nss.nss_port = nmp->nm_nfsport;
1168 nss.nss_protocol = NFS_PROG;
1169 nss.nss_version = nmp->nm_vers;
1170 }
1171 }
1172 NFS_SOCK_DBG("nfs connect first %s, so type %d port %d prot %d %d\n",
1173 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nss.nss_sotype, nss.nss_port,
1174 nss.nss_protocol, nss.nss_version);
1175 } else {
1176 /* we've connected before, just connect to NFS port */
1177 if (!nmp->nm_nfsport) {
1178 /* need to ask portmapper which port that would be */
1179 nss.nss_port = PMAPPORT;
1180 nss.nss_protocol = PMAPPROG;
1181 nss.nss_version = 0;
1182 } else {
1183 nss.nss_port = nmp->nm_nfsport;
1184 nss.nss_protocol = NFS_PROG;
1185 nss.nss_version = nmp->nm_vers;
1186 }
1187 NFS_SOCK_DBG("nfs connect %s, so type %d port %d prot %d %d\n",
1188 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nss.nss_sotype, nss.nss_port,
1189 nss.nss_protocol, nss.nss_version);
1190 }
1191
1192 /* Set next location to first valid location. */
1193 /* If start location is invalid, find next location. */
1194 nss.nss_nextloc = nss.nss_startloc;
1195 if ((nss.nss_nextloc.nli_serv >= nmp->nm_locations.nl_locations[nss.nss_nextloc.nli_loc]->nl_servcount) ||
1196 (nss.nss_nextloc.nli_addr >= nmp->nm_locations.nl_locations[nss.nss_nextloc.nli_loc]->nl_servers[nss.nss_nextloc.nli_serv]->ns_addrcount)) {
1197 nfs_location_next(&nmp->nm_locations, &nss.nss_nextloc);
1198 if (!nfs_location_index_cmp(&nss.nss_nextloc, &nss.nss_startloc)) {
1199 NFS_SOCK_DBG("nfs connect %s search failed, couldn't find a valid location index\n",
1200 vfs_statfs(nmp->nm_mountp)->f_mntfromname);
1201 return (ENOENT);
1202 }
1203 }
1204 nss.nss_last = -1;
1205
1206keepsearching:
1207
1208 error = nfs_connect_search_loop(nmp, &nss);
1209 if (error || !nss.nss_sock) {
1210 /* search failed */
1211 nfs_socket_search_cleanup(&nss);
1212 if (nss.nss_flags & NSS_FALLBACK2PMAP) {
1213 tryv4 = 0;
1214 NFS_SOCK_DBG("nfs connect %s TCP failed for V4 %d %d, trying PORTMAP\n",
1215 vfs_statfs(nmp->nm_mountp)->f_mntfromname, error, nss.nss_error);
1216 goto tryagain;
1217 }
1218
1219 if (!error && (nss.nss_sotype == SOCK_STREAM) && !nmp->nm_sotype && (nmp->nm_vers < NFS_VER4)) {
1220 /* Try using UDP */
1221 sotype = SOCK_DGRAM;
1222 savederror = nss.nss_error;
1223 NFS_SOCK_DBG("nfs connect %s TCP failed %d %d, trying UDP\n",
1224 vfs_statfs(nmp->nm_mountp)->f_mntfromname, error, nss.nss_error);
1225 goto tryagain;
1226 }
1227 if (!error)
1228 error = nss.nss_error ? nss.nss_error : ETIMEDOUT;
1229 lck_mtx_lock(&nmp->nm_lock);
1230 nmp->nm_sockflags &= ~NMSOCK_CONNECTING;
1231 nmp->nm_nss = NULL;
1232 lck_mtx_unlock(&nmp->nm_lock);
1233 if (nss.nss_flags & NSS_WARNED)
1234 log(LOG_INFO, "nfs_connect: socket connect aborted for %s\n",
1235 vfs_statfs(nmp->nm_mountp)->f_mntfromname);
1236 if (fh)
1237 FREE(fh, M_TEMP);
1238 if (path)
1239 FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
1240 NFS_SOCK_DBG("nfs connect %s search failed, returning %d\n",
1241 vfs_statfs(nmp->nm_mountp)->f_mntfromname, error);
1242 return (error);
1243 }
1244
1245 /* try to use nss_sock */
1246 nso = nss.nss_sock;
1247 nss.nss_sock = NULL;
1248
1249 /* We may be speaking to portmap first... to determine port(s). */
1250 if (nso->nso_saddr->sa_family == AF_INET)
1251 port = ntohs(((struct sockaddr_in*)nso->nso_saddr)->sin_port);
1252 else
1253 port = ntohs(((struct sockaddr_in6*)nso->nso_saddr)->sin6_port);
1254 if (port == PMAPPORT) {
1255 /* Use this portmapper port to get the port #s we need. */
1256 NFS_SOCK_DBG("nfs connect %s got portmapper socket %p\n",
1257 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso);
1258
1259 /* remove the connect upcall so nfs_portmap_lookup() can use this socket */
1260 sock_setupcall(nso->nso_so, NULL, NULL);
1261
1262 /* Set up socket address and port for NFS socket. */
1263 bcopy(nso->nso_saddr, &ss, nso->nso_saddr->sa_len);
1264
1265 /* If NFS version not set, try nm_max_vers down to nm_min_vers */
1266 nfsvers = nmp->nm_vers ? nmp->nm_vers : PVER2MAJOR(nmp->nm_max_vers);
1267 if (!(port = nmp->nm_nfsport)) {
1268 if (ss.ss_family == AF_INET)
1269 ((struct sockaddr_in*)&ss)->sin_port = htons(0);
1270 else if (ss.ss_family == AF_INET6)
1271 ((struct sockaddr_in6*)&ss)->sin6_port = htons(0);
1272 for (; nfsvers >= (int)PVER2MAJOR(nmp->nm_min_vers); nfsvers--) {
1273 if (nmp->nm_vers && nmp->nm_vers != nfsvers)
1274 continue; /* Wrong version */
1275 if (nfsvers == NFS_VER4 && nso->nso_sotype == SOCK_DGRAM)
1276 continue; /* NFSv4 does not do UDP */
1277 error = nfs_portmap_lookup(nmp, vfs_context_current(), (struct sockaddr*)&ss,
1278 nso->nso_so, NFS_PROG, nfsvers,
1279 (nso->nso_sotype == SOCK_DGRAM) ? IPPROTO_UDP : IPPROTO_TCP, timeo);
1280 if (!error) {
1281 if (ss.ss_family == AF_INET)
1282 port = ntohs(((struct sockaddr_in*)&ss)->sin_port);
1283 else if (ss.ss_family == AF_INET6)
1284 port = ntohs(((struct sockaddr_in6*)&ss)->sin6_port);
1285 if (!port)
1286 error = EPROGUNAVAIL;
1287 if (port == NFS_PORT && nfsvers == NFS_VER4 && tryv4 == 0)
1288 continue; /* We already tried this */
1289 }
1290 if (!error)
1291 break;
1292 }
1293 if (nfsvers < (int)PVER2MAJOR(nmp->nm_min_vers) && error == 0)
1294 error = EPROGUNAVAIL;
1295 if (error) {
1296 nfs_socket_search_update_error(&nss, error);
1297 nfs_socket_destroy(nso);
1298 goto keepsearching;
1299 }
1300 }
1301 /* Create NFS protocol socket and add it to the list of sockets. */
1302 /* N.B. If nfsvers is NFS_VER4 at this point then we're on a non standard port */
1303 error = nfs_socket_create(nmp, (struct sockaddr*)&ss, nso->nso_sotype, port,
1304 NFS_PROG, nfsvers, NMFLAG(nmp, RESVPORT), &nsonfs);
1305 if (error) {
1306 nfs_socket_search_update_error(&nss, error);
1307 nfs_socket_destroy(nso);
1308 goto keepsearching;
1309 }
1310 nsonfs->nso_location = nso->nso_location;
1311 nsonfs->nso_wake = &nss;
1312 error = sock_setupcall(nsonfs->nso_so, nfs_connect_upcall, nsonfs);
1313 if (error) {
1314 nfs_socket_search_update_error(&nss, error);
1315 nfs_socket_destroy(nsonfs);
1316 nfs_socket_destroy(nso);
1317 goto keepsearching;
1318 }
1319 TAILQ_INSERT_TAIL(&nss.nss_socklist, nsonfs, nso_link);
1320 nss.nss_sockcnt++;
1321 if ((nfsvers < NFS_VER4) && !(nmp->nm_sockflags & NMSOCK_HASCONNECTED) && !NM_OMATTR_GIVEN(nmp, FH)) {
1322 /* Set up socket address and port for MOUNT socket. */
1323 error = 0;
1324 bcopy(nso->nso_saddr, &ss, nso->nso_saddr->sa_len);
1325 port = nmp->nm_mountport;
1326 if (ss.ss_family == AF_INET)
1327 ((struct sockaddr_in*)&ss)->sin_port = htons(port);
1328 else if (ss.ss_family == AF_INET6)
1329 ((struct sockaddr_in6*)&ss)->sin6_port = htons(port);
1330 if (!port) {
1331 /* Get port/sockaddr for MOUNT version corresponding to NFS version. */
1332 /* If NFS version is unknown, optimistically choose for NFSv3. */
1333 int mntvers = (nfsvers == NFS_VER2) ? RPCMNT_VER1 : RPCMNT_VER3;
1334 int mntproto = (NM_OMFLAG(nmp, MNTUDP) || (nso->nso_sotype == SOCK_DGRAM)) ? IPPROTO_UDP : IPPROTO_TCP;
1335 error = nfs_portmap_lookup(nmp, vfs_context_current(), (struct sockaddr*)&ss,
1336 nso->nso_so, RPCPROG_MNT, mntvers, mntproto, timeo);
1337 }
1338 if (!error) {
1339 if (ss.ss_family == AF_INET)
1340 port = ntohs(((struct sockaddr_in*)&ss)->sin_port);
1341 else if (ss.ss_family == AF_INET6)
1342 port = ntohs(((struct sockaddr_in6*)&ss)->sin6_port);
1343 if (!port)
1344 error = EPROGUNAVAIL;
1345 }
1346 /* create sockaddr for MOUNT */
1347 if (!error)
1348 MALLOC(nsonfs->nso_saddr2, struct sockaddr *, ss.ss_len, M_SONAME, M_WAITOK|M_ZERO);
1349 if (!error && !nsonfs->nso_saddr2)
1350 error = ENOMEM;
1351 if (!error)
1352 bcopy(&ss, nsonfs->nso_saddr2, ss.ss_len);
1353 if (error) {
1354 lck_mtx_lock(&nsonfs->nso_lock);
1355 nsonfs->nso_error = error;
1356 nsonfs->nso_flags |= NSO_DEAD;
1357 lck_mtx_unlock(&nsonfs->nso_lock);
1358 }
1359 }
1360 nfs_socket_destroy(nso);
1361 goto keepsearching;
1362 }
1363
1364 /* nso is an NFS socket */
1365 NFS_SOCK_DBG("nfs connect %s got NFS socket %p\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso);
1366
1367 /* If NFS version wasn't specified, it was determined during the connect. */
1368 nfsvers = nmp->nm_vers ? nmp->nm_vers : (int)nso->nso_version;
1369
1370 /* Perform MOUNT call for initial NFSv2/v3 connection/mount. */
1371 if ((nfsvers < NFS_VER4) && !(nmp->nm_sockflags & NMSOCK_HASCONNECTED) && !NM_OMATTR_GIVEN(nmp, FH)) {
1372 error = 0;
1373 saddr = nso->nso_saddr2;
1374 if (!saddr) {
1375 /* Need sockaddr for MOUNT port */
1376 bcopy(nso->nso_saddr, &ss, nso->nso_saddr->sa_len);
1377 port = nmp->nm_mountport;
1378 if (ss.ss_family == AF_INET)
1379 ((struct sockaddr_in*)&ss)->sin_port = htons(port);
1380 else if (ss.ss_family == AF_INET6)
1381 ((struct sockaddr_in6*)&ss)->sin6_port = htons(port);
1382 if (!port) {
1383 /* Get port/sockaddr for MOUNT version corresponding to NFS version. */
1384 int mntvers = (nfsvers == NFS_VER2) ? RPCMNT_VER1 : RPCMNT_VER3;
1385 int mntproto = (NM_OMFLAG(nmp, MNTUDP) || (nso->nso_sotype == SOCK_DGRAM)) ? IPPROTO_UDP : IPPROTO_TCP;
1386 error = nfs_portmap_lookup(nmp, vfs_context_current(), (struct sockaddr*)&ss,
1387 NULL, RPCPROG_MNT, mntvers, mntproto, timeo);
1388 if (ss.ss_family == AF_INET)
1389 port = ntohs(((struct sockaddr_in*)&ss)->sin_port);
1390 else if (ss.ss_family == AF_INET6)
1391 port = ntohs(((struct sockaddr_in6*)&ss)->sin6_port);
1392 }
1393 if (!error) {
1394 if (port)
1395 saddr = (struct sockaddr*)&ss;
1396 else
1397 error = EPROGUNAVAIL;
1398 }
1399 }
1400 if (saddr)
1401 MALLOC(fh, fhandle_t *, sizeof(fhandle_t), M_TEMP, M_WAITOK|M_ZERO);
1402 if (saddr && fh)
1403 MALLOC_ZONE(path, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
1404 if (!saddr || !fh || !path) {
1405 if (!error)
1406 error = ENOMEM;
1407 if (fh)
1408 FREE(fh, M_TEMP);
1409 if (path)
1410 FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
1411 fh = NULL;
1412 path = NULL;
1413 nfs_socket_search_update_error(&nss, error);
1414 nfs_socket_destroy(nso);
1415 goto keepsearching;
1416 }
1417 nfs_location_mntfromname(&nmp->nm_locations, nso->nso_location, path, MAXPATHLEN, 1);
1418 error = nfs3_mount_rpc(nmp, saddr, nso->nso_sotype, nfsvers,
1419 path, vfs_context_current(), timeo, fh, &nmp->nm_servsec);
1420 NFS_SOCK_DBG("nfs connect %s socket %p mount %d\n",
1421 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, error);
1422 if (!error) {
1423 /* Make sure we can agree on a security flavor. */
1424 int o, s; /* indices into mount option and server security flavor lists */
1425 int found = 0;
1426
1427 if ((nfsvers == NFS_VER3) && !nmp->nm_servsec.count) {
1428 /* Some servers return an empty list to indicate RPCAUTH_SYS? */
1429 nmp->nm_servsec.count = 1;
1430 nmp->nm_servsec.flavors[0] = RPCAUTH_SYS;
1431 }
1432 if (nmp->nm_sec.count) {
1433 /* Choose the first flavor in our list that the server supports. */
1434 if (!nmp->nm_servsec.count) {
1435 /* we don't know what the server supports, just use our first choice */
1436 nmp->nm_auth = nmp->nm_sec.flavors[0];
1437 found = 1;
1438 }
1439 for (o=0; !found && (o < nmp->nm_sec.count); o++)
1440 for (s=0; !found && (s < nmp->nm_servsec.count); s++)
1441 if (nmp->nm_sec.flavors[o] == nmp->nm_servsec.flavors[s]) {
1442 nmp->nm_auth = nmp->nm_sec.flavors[o];
1443 found = 1;
1444 }
1445 } else {
1446 /* Choose the first one we support from the server's list. */
1447 if (!nmp->nm_servsec.count) {
1448 nmp->nm_auth = RPCAUTH_SYS;
1449 found = 1;
1450 }
1451 for (s=0; s < nmp->nm_servsec.count; s++)
1452 switch (nmp->nm_servsec.flavors[s]) {
1453 case RPCAUTH_SYS:
1454 /* prefer RPCAUTH_SYS to RPCAUTH_NONE */
1455 if (found && (nmp->nm_auth == RPCAUTH_NONE))
1456 found = 0;
1457 case RPCAUTH_NONE:
1458 case RPCAUTH_KRB5:
1459 case RPCAUTH_KRB5I:
1460 case RPCAUTH_KRB5P:
1461 if (!found) {
1462 nmp->nm_auth = nmp->nm_servsec.flavors[s];
1463 found = 1;
1464 }
1465 break;
1466 }
1467 }
1468 error = !found ? EAUTH : 0;
1469 }
1470 FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
1471 path = NULL;
1472 if (error) {
1473 nfs_socket_search_update_error(&nss, error);
1474 FREE(fh, M_TEMP);
1475 fh = NULL;
1476 nfs_socket_destroy(nso);
1477 goto keepsearching;
1478 }
1479 if (nmp->nm_fh)
1480 FREE(nmp->nm_fh, M_TEMP);
1481 nmp->nm_fh = fh;
1482 fh = NULL;
1483 NFS_BITMAP_SET(nmp->nm_flags, NFS_MFLAG_CALLUMNT);
1484 }
1485
1486 /* put the real upcall in place */
1487 upcall = (nso->nso_sotype == SOCK_STREAM) ? nfs_tcp_rcv : nfs_udp_rcv;
1488 error = sock_setupcall(nso->nso_so, upcall, nmp);
1489 if (error) {
1490 nfs_socket_search_update_error(&nss, error);
1491 nfs_socket_destroy(nso);
1492 goto keepsearching;
1493 }
1494
1495 if (!(nmp->nm_sockflags & NMSOCK_HASCONNECTED)) {
1496 /* set mntfromname to this location */
1497 if (!NM_OMATTR_GIVEN(nmp, MNTFROM))
1498 nfs_location_mntfromname(&nmp->nm_locations, nso->nso_location,
1499 vfs_statfs(nmp->nm_mountp)->f_mntfromname,
1500 sizeof(vfs_statfs(nmp->nm_mountp)->f_mntfromname), 0);
1501 /* some negotiated values need to remain unchanged for the life of the mount */
1502 if (!nmp->nm_sotype)
1503 nmp->nm_sotype = nso->nso_sotype;
1504 if (!nmp->nm_vers) {
1505 nmp->nm_vers = nfsvers;
1506 /* If we negotiated NFSv4, set nm_nfsport if we ended up on the standard NFS port */
1507 if ((nfsvers >= NFS_VER4) && !NFS_BITMAP_ISSET(nmp->nm_mattrs, NFS_MATTR_NFS_PORT)) {
1508 if (nso->nso_saddr->sa_family == AF_INET)
1509 port = ((struct sockaddr_in*)nso->nso_saddr)->sin_port = htons(port);
1510 else if (nso->nso_saddr->sa_family == AF_INET6)
1511 port = ((struct sockaddr_in6*)nso->nso_saddr)->sin6_port = htons(port);
1512 else
1513 port = 0;
1514 if (port == NFS_PORT)
1515 nmp->nm_nfsport = NFS_PORT;
1516 }
1517 }
1518 /* do some version-specific pre-mount set up */
1519 if (nmp->nm_vers >= NFS_VER4) {
1520 microtime(&now);
1521 nmp->nm_mounttime = ((uint64_t)now.tv_sec << 32) | now.tv_usec;
1522 if (!NMFLAG(nmp, NOCALLBACK))
1523 nfs4_mount_callback_setup(nmp);
1524 }
1525 }
1526
1527 /* Initialize NFS socket state variables */
1528 lck_mtx_lock(&nmp->nm_lock);
1529 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
1530 nmp->nm_srtt[3] = (NFS_TIMEO << 3);
1531 nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
1532 nmp->nm_sdrtt[3] = 0;
1533 if (nso->nso_sotype == SOCK_DGRAM) {
1534 nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */
1535 nmp->nm_sent = 0;
1536 } else if (nso->nso_sotype == SOCK_STREAM) {
1537 nmp->nm_timeouts = 0;
1538 }
1539 nmp->nm_sockflags &= ~NMSOCK_CONNECTING;
1540 nmp->nm_sockflags |= NMSOCK_SETUP;
1541 /* move the socket to the mount structure */
1542 nmp->nm_nso = nso;
1543 oldsaddr = nmp->nm_saddr;
1544 nmp->nm_saddr = nso->nso_saddr;
1545 lck_mtx_unlock(&nmp->nm_lock);
1546 error = nfs_connect_setup(nmp);
1547 lck_mtx_lock(&nmp->nm_lock);
1548 nmp->nm_sockflags &= ~NMSOCK_SETUP;
1549 if (!error) {
1550 nmp->nm_sockflags |= NMSOCK_READY;
1551 wakeup(&nmp->nm_sockflags);
1552 }
1553 if (error) {
1554 NFS_SOCK_DBG("nfs connect %s socket %p setup failed %d\n",
1555 vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, error);
1556 nfs_socket_search_update_error(&nss, error);
1557 nmp->nm_saddr = oldsaddr;
1558 if (!(nmp->nm_sockflags & NMSOCK_HASCONNECTED)) {
1559 /* undo settings made prior to setup */
1560 if (!NFS_BITMAP_ISSET(nmp->nm_mattrs, NFS_MATTR_SOCKET_TYPE))
1561 nmp->nm_sotype = 0;
1562 if (!NFS_BITMAP_ISSET(nmp->nm_mattrs, NFS_MATTR_NFS_VERSION)) {
1563 if (nmp->nm_vers >= NFS_VER4) {
1564 if (!NFS_BITMAP_ISSET(nmp->nm_mattrs, NFS_MATTR_NFS_PORT))
1565 nmp->nm_nfsport = 0;
1566 if (nmp->nm_cbid)
1567 nfs4_mount_callback_shutdown(nmp);
1568 if (IS_VALID_CRED(nmp->nm_mcred))
1569 kauth_cred_unref(&nmp->nm_mcred);
1570 bzero(&nmp->nm_un, sizeof(nmp->nm_un));
1571 }
1572 nmp->nm_vers = 0;
1573 }
1574 }
1575 lck_mtx_unlock(&nmp->nm_lock);
1576 nmp->nm_nso = NULL;
1577 nfs_socket_destroy(nso);
1578 goto keepsearching;
1579 }
1580
1581 /* update current location */
1582 if ((nmp->nm_locations.nl_current.nli_flags & NLI_VALID) &&
1583 (nmp->nm_locations.nl_current.nli_serv != nso->nso_location.nli_serv)) {
1584 /* server has changed, we should initiate failover/recovery */
1585 // XXX
1586 }
1587 nmp->nm_locations.nl_current = nso->nso_location;
1588 nmp->nm_locations.nl_current.nli_flags |= NLI_VALID;
1589
1590 if (!(nmp->nm_sockflags & NMSOCK_HASCONNECTED)) {
1591 /* We have now successfully connected... make a note of it. */
1592 nmp->nm_sockflags |= NMSOCK_HASCONNECTED;
1593 }
1594
1595 lck_mtx_unlock(&nmp->nm_lock);
1596 if (oldsaddr)
1597 FREE(oldsaddr, M_SONAME);
1598
1599 if (nss.nss_flags & NSS_WARNED)
1600 log(LOG_INFO, "nfs_connect: socket connect completed for %s\n",
1601 vfs_statfs(nmp->nm_mountp)->f_mntfromname);
1602
1603 nmp->nm_nss = NULL;
1604 nfs_socket_search_cleanup(&nss);
1605 if (fh)
1606 FREE(fh, M_TEMP);
1607 if (path)
1608 FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
1609 NFS_SOCK_DBG("nfs connect %s success\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname);
1610 return (0);
1611}
1612
1613
1614/* setup & confirm socket connection is functional */
1615int
1616nfs_connect_setup(struct nfsmount *nmp)
1617{
1618 int error = 0;
1619
1620 if (nmp->nm_vers >= NFS_VER4) {
1621 if (nmp->nm_state & NFSSTA_CLIENTID) {
1622 /* first, try to renew our current state */
1623 error = nfs4_renew(nmp, R_SETUP);
1624 if ((error == NFSERR_ADMIN_REVOKED) ||
1625 (error == NFSERR_CB_PATH_DOWN) ||
1626 (error == NFSERR_EXPIRED) ||
1627 (error == NFSERR_LEASE_MOVED) ||
1628 (error == NFSERR_STALE_CLIENTID)) {
1629 lck_mtx_lock(&nmp->nm_lock);
1630 nfs_need_recover(nmp, error);
1631 lck_mtx_unlock(&nmp->nm_lock);
1632 }
1633 }
1634 error = nfs4_setclientid(nmp);
1635 }
1636 return (error);
1637}
1638
1639/*
1640 * NFS socket reconnect routine:
1641 * Called when a connection is broken.
1642 * - disconnect the old socket
1643 * - nfs_connect() again
1644 * - set R_MUSTRESEND for all outstanding requests on mount point
1645 * If this fails the mount point is DEAD!
1646 */
1647int
1648nfs_reconnect(struct nfsmount *nmp)
1649{
1650 struct nfsreq *rq;
1651 struct timeval now;
1652 thread_t thd = current_thread();
1653 int error, wentdown = 0, verbose = 1;
1654 time_t lastmsg;
1655 int timeo;
1656
1657 microuptime(&now);
1658 lastmsg = now.tv_sec - (nmp->nm_tprintf_delay - nmp->nm_tprintf_initial_delay);
1659
1660 nfs_disconnect(nmp);
1661
1662
1663 lck_mtx_lock(&nmp->nm_lock);
1664 timeo = nfs_is_squishy(nmp) ? 8 : 30;
1665 lck_mtx_unlock(&nmp->nm_lock);
1666
1667 while ((error = nfs_connect(nmp, verbose, timeo))) {
1668 verbose = 0;
1669 nfs_disconnect(nmp);
1670 if ((error == EINTR) || (error == ERESTART))
1671 return (EINTR);
1672 if (error == EIO)
1673 return (EIO);
1674 microuptime(&now);
1675 if ((lastmsg + nmp->nm_tprintf_delay) < now.tv_sec) {
1676 lastmsg = now.tv_sec;
1677 nfs_down(nmp, thd, error, NFSSTA_TIMEO, "can not connect", 0);
1678 wentdown = 1;
1679 }
1680 lck_mtx_lock(&nmp->nm_lock);
1681 if (!(nmp->nm_state & NFSSTA_MOUNTED)) {
1682 /* we're not yet completely mounted and */
1683 /* we can't reconnect, so we fail */
1684 lck_mtx_unlock(&nmp->nm_lock);
1685 NFS_SOCK_DBG("Not mounted returning %d\n", error);
1686 return (error);
1687 }
1688
1689 if (nfs_mount_check_dead_timeout(nmp)) {
1690 nfs_mount_make_zombie(nmp);
1691 lck_mtx_unlock(&nmp->nm_lock);
1692 return (ENXIO);
1693 }
1694
1695 if ((error = nfs_sigintr(nmp, NULL, thd, 1))) {
1696 lck_mtx_unlock(&nmp->nm_lock);
1697 return (error);
1698 }
1699 lck_mtx_unlock(&nmp->nm_lock);
1700 tsleep(nfs_reconnect, PSOCK, "nfs_reconnect_delay", 2*hz);
1701 if ((error = nfs_sigintr(nmp, NULL, thd, 0)))
1702 return (error);
1703 }
1704
1705 if (wentdown)
1706 nfs_up(nmp, thd, NFSSTA_TIMEO, "connected");
1707
1708 /*
1709 * Loop through outstanding request list and mark all requests
1710 * as needing a resend. (Though nfs_need_reconnect() probably
1711 * marked them all already.)
1712 */
1713 lck_mtx_lock(nfs_request_mutex);
1714 TAILQ_FOREACH(rq, &nfs_reqq, r_chain) {
1715 if (rq->r_nmp == nmp) {
1716 lck_mtx_lock(&rq->r_mtx);
1717 if (!rq->r_error && !rq->r_nmrep.nmc_mhead && !(rq->r_flags & R_MUSTRESEND)) {
1718 rq->r_flags |= R_MUSTRESEND;
1719 rq->r_rtt = -1;
1720 wakeup(rq);
1721 if ((rq->r_flags & (R_IOD|R_ASYNC|R_ASYNCWAIT|R_SENDING)) == R_ASYNC)
1722 nfs_asyncio_resend(rq);
1723 }
1724 lck_mtx_unlock(&rq->r_mtx);
1725 }
1726 }
1727 lck_mtx_unlock(nfs_request_mutex);
1728 return (0);
1729}
1730
1731/*
1732 * NFS disconnect. Clean up and unlink.
1733 */
1734void
1735nfs_disconnect(struct nfsmount *nmp)
1736{
1737 struct nfs_socket *nso;
1738
1739 lck_mtx_lock(&nmp->nm_lock);
1740tryagain:
1741 if (nmp->nm_nso) {
1742 struct timespec ts = { 1, 0 };
1743 if (nmp->nm_state & NFSSTA_SENDING) { /* wait for sending to complete */
1744 nmp->nm_state |= NFSSTA_WANTSND;
1745 msleep(&nmp->nm_state, &nmp->nm_lock, PZERO-1, "nfswaitsending", &ts);
1746 goto tryagain;
1747 }
1748 if (nmp->nm_sockflags & NMSOCK_POKE) { /* wait for poking to complete */
1749 msleep(&nmp->nm_sockflags, &nmp->nm_lock, PZERO-1, "nfswaitpoke", &ts);
1750 goto tryagain;
1751 }
1752 nmp->nm_sockflags |= NMSOCK_DISCONNECTING;
1753 nmp->nm_sockflags &= ~NMSOCK_READY;
1754 nso = nmp->nm_nso;
1755 nmp->nm_nso = NULL;
1756 if (nso->nso_saddr == nmp->nm_saddr)
1757 nso->nso_saddr = NULL;
1758 lck_mtx_unlock(&nmp->nm_lock);
1759 nfs_socket_destroy(nso);
1760 lck_mtx_lock(&nmp->nm_lock);
1761 nmp->nm_sockflags &= ~NMSOCK_DISCONNECTING;
1762 lck_mtx_unlock(&nmp->nm_lock);
1763 } else {
1764 lck_mtx_unlock(&nmp->nm_lock);
1765 }
1766}
1767
1768/*
1769 * mark an NFS mount as needing a reconnect/resends.
1770 */
1771void
1772nfs_need_reconnect(struct nfsmount *nmp)
1773{
1774 struct nfsreq *rq;
1775
1776 lck_mtx_lock(&nmp->nm_lock);
1777 nmp->nm_sockflags &= ~(NMSOCK_READY|NMSOCK_SETUP);
1778 lck_mtx_unlock(&nmp->nm_lock);
1779
1780 /*
1781 * Loop through outstanding request list and
1782 * mark all requests as needing a resend.
1783 */
1784 lck_mtx_lock(nfs_request_mutex);
1785 TAILQ_FOREACH(rq, &nfs_reqq, r_chain) {
1786 if (rq->r_nmp == nmp) {
1787 lck_mtx_lock(&rq->r_mtx);
1788 if (!rq->r_error && !rq->r_nmrep.nmc_mhead && !(rq->r_flags & R_MUSTRESEND)) {
1789 rq->r_flags |= R_MUSTRESEND;
1790 rq->r_rtt = -1;
1791 wakeup(rq);
1792 if ((rq->r_flags & (R_IOD|R_ASYNC|R_ASYNCWAIT|R_SENDING)) == R_ASYNC)
1793 nfs_asyncio_resend(rq);
1794 }
1795 lck_mtx_unlock(&rq->r_mtx);
1796 }
1797 }
1798 lck_mtx_unlock(nfs_request_mutex);
1799}
1800
1801
1802/*
1803 * thread to handle miscellaneous async NFS socket work (reconnects/resends)
1804 */
1805void
1806nfs_mount_sock_thread(void *arg, __unused wait_result_t wr)
1807{
1808 struct nfsmount *nmp = arg;
1809 struct timespec ts = { 30, 0 };
1810 thread_t thd = current_thread();
1811 struct nfsreq *req;
1812 struct timeval now;
1813 int error, dofinish;
1814 nfsnode_t np;
1815 int do_reconnect_sleep = 0;
1816
1817 lck_mtx_lock(&nmp->nm_lock);
1818 while (!(nmp->nm_sockflags & NMSOCK_READY) ||
1819 !TAILQ_EMPTY(&nmp->nm_resendq) ||
1820 !LIST_EMPTY(&nmp->nm_monlist) ||
1821 nmp->nm_deadto_start ||
1822 (nmp->nm_state & NFSSTA_RECOVER) ||
1823 ((nmp->nm_vers >= NFS_VER4) && !TAILQ_EMPTY(&nmp->nm_dreturnq)))
1824 {
1825 if (nmp->nm_sockflags & NMSOCK_UNMOUNT)
1826 break;
1827 /* do reconnect, if necessary */
1828 if (!(nmp->nm_sockflags & NMSOCK_READY) && !(nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD))) {
1829 if (nmp->nm_reconnect_start <= 0) {
1830 microuptime(&now);
1831 nmp->nm_reconnect_start = now.tv_sec;
1832 }
1833 lck_mtx_unlock(&nmp->nm_lock);
1834 NFS_SOCK_DBG("nfs reconnect %s\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname);
1835 /*
1836 * XXX We don't want to call reconnect again right away if returned errors
1837 * before that may not have blocked. This has caused spamming null procs
1838 * from machines in the pass.
1839 */
1840 if (do_reconnect_sleep)
1841 tsleep(nfs_mount_sock_thread, PSOCK, "nfs_reconnect_sock_thread_delay", hz);
1842 error = nfs_reconnect(nmp);
1843 if (error) {
1844 int lvl = 7;
1845 if (error == EIO || error == EINTR) {
1846 lvl = (do_reconnect_sleep++ % 600) ? 7 : 0;
1847 }
1848 nfs_printf(NFS_FAC_SOCK, lvl, "nfs reconnect %s: returned %d\n",
1849 vfs_statfs(nmp->nm_mountp)->f_mntfromname, error);
1850 } else {
1851 nmp->nm_reconnect_start = 0;
1852 do_reconnect_sleep = 0;
1853 }
1854 lck_mtx_lock(&nmp->nm_lock);
1855 }
1856 if ((nmp->nm_sockflags & NMSOCK_READY) &&
1857 (nmp->nm_state & NFSSTA_RECOVER) &&
1858 !(nmp->nm_sockflags & NMSOCK_UNMOUNT) &&
1859 !(nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD))) {
1860 /* perform state recovery */
1861 lck_mtx_unlock(&nmp->nm_lock);
1862 nfs_recover(nmp);
1863 lck_mtx_lock(&nmp->nm_lock);
1864 }
1865 /* handle NFSv4 delegation returns */
1866 while ((nmp->nm_vers >= NFS_VER4) && !(nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD)) &&
1867 (nmp->nm_sockflags & NMSOCK_READY) && !(nmp->nm_state & NFSSTA_RECOVER) &&
1868 ((np = TAILQ_FIRST(&nmp->nm_dreturnq)))) {
1869 lck_mtx_unlock(&nmp->nm_lock);
1870 nfs4_delegation_return(np, R_RECOVER, thd, nmp->nm_mcred);
1871 lck_mtx_lock(&nmp->nm_lock);
1872 }
1873 /* do resends, if necessary/possible */
1874 while ((((nmp->nm_sockflags & NMSOCK_READY) && !(nmp->nm_state & NFSSTA_RECOVER)) ||
1875 (nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD))) &&
1876 ((req = TAILQ_FIRST(&nmp->nm_resendq)))) {
1877 if (req->r_resendtime)
1878 microuptime(&now);
1879 while (req && !(nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD)) && req->r_resendtime && (now.tv_sec < req->r_resendtime))
1880 req = TAILQ_NEXT(req, r_rchain);
1881 if (!req)
1882 break;
1883 TAILQ_REMOVE(&nmp->nm_resendq, req, r_rchain);
1884 req->r_rchain.tqe_next = NFSREQNOLIST;
1885 lck_mtx_unlock(&nmp->nm_lock);
1886 lck_mtx_lock(&req->r_mtx);
1887 /* Note that we have a reference on the request that was taken nfs_asyncio_resend */
1888 if (req->r_error || req->r_nmrep.nmc_mhead) {
1889 dofinish = req->r_callback.rcb_func && !(req->r_flags & R_WAITSENT);
1890 req->r_flags &= ~R_RESENDQ;
1891 wakeup(req);
1892 lck_mtx_unlock(&req->r_mtx);
1893 if (dofinish)
1894 nfs_asyncio_finish(req);
1895 nfs_request_rele(req);
1896 lck_mtx_lock(&nmp->nm_lock);
1897 continue;
1898 }
1899 if ((req->r_flags & R_RESTART) || nfs_request_using_gss(req)) {
1900 req->r_flags &= ~R_RESTART;
1901 req->r_resendtime = 0;
1902 lck_mtx_unlock(&req->r_mtx);
1903 /* async RPCs on GSS mounts need to be rebuilt and resent. */
1904 nfs_reqdequeue(req);
1905 if (nfs_request_using_gss(req)) {
1906 nfs_gss_clnt_rpcdone(req);
1907 error = nfs_gss_clnt_args_restore(req);
1908 if (error == ENEEDAUTH)
1909 req->r_xid = 0;
1910 }
1911 NFS_SOCK_DBG("nfs async%s restart: p %d x 0x%llx f 0x%x rtt %d\n",
1912 nfs_request_using_gss(req) ? " gss" : "", req->r_procnum, req->r_xid,
1913 req->r_flags, req->r_rtt);
1914 error = nfs_sigintr(nmp, req, req->r_thread, 0);
1915 if (!error)
1916 error = nfs_request_add_header(req);
1917 if (!error)
1918 error = nfs_request_send(req, 0);
1919 lck_mtx_lock(&req->r_mtx);
1920 if (req->r_flags & R_RESENDQ)
1921 req->r_flags &= ~R_RESENDQ;
1922 if (error)
1923 req->r_error = error;
1924 wakeup(req);
1925 dofinish = error && req->r_callback.rcb_func && !(req->r_flags & R_WAITSENT);
1926 lck_mtx_unlock(&req->r_mtx);
1927 if (dofinish)
1928 nfs_asyncio_finish(req);
1929 nfs_request_rele(req);
1930 lck_mtx_lock(&nmp->nm_lock);
1931 error = 0;
1932 continue;
1933 }
1934 NFS_SOCK_DBG("nfs async resend: p %d x 0x%llx f 0x%x rtt %d\n",
1935 req->r_procnum, req->r_xid, req->r_flags, req->r_rtt);
1936 error = nfs_sigintr(nmp, req, req->r_thread, 0);
1937 if (!error) {
1938 req->r_flags |= R_SENDING;
1939 lck_mtx_unlock(&req->r_mtx);
1940 error = nfs_send(req, 0);
1941 lck_mtx_lock(&req->r_mtx);
1942 if (!error) {
1943 if (req->r_flags & R_RESENDQ)
1944 req->r_flags &= ~R_RESENDQ;
1945 wakeup(req);
1946 lck_mtx_unlock(&req->r_mtx);
1947 nfs_request_rele(req);
1948 lck_mtx_lock(&nmp->nm_lock);
1949 continue;
1950 }
1951 }
1952 req->r_error = error;
1953 if (req->r_flags & R_RESENDQ)
1954 req->r_flags &= ~R_RESENDQ;
1955 wakeup(req);
1956 dofinish = req->r_callback.rcb_func && !(req->r_flags & R_WAITSENT);
1957 lck_mtx_unlock(&req->r_mtx);
1958 if (dofinish)
1959 nfs_asyncio_finish(req);
1960 nfs_request_rele(req);
1961 lck_mtx_lock(&nmp->nm_lock);
1962 }
1963 if (nfs_mount_check_dead_timeout(nmp)) {
1964 nfs_mount_make_zombie(nmp);
1965 break;
1966 }
1967
1968 if (nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD))
1969 break;
1970 /* check monitored nodes, if necessary/possible */
1971 if (!LIST_EMPTY(&nmp->nm_monlist)) {
1972 nmp->nm_state |= NFSSTA_MONITOR_SCAN;
1973 LIST_FOREACH(np, &nmp->nm_monlist, n_monlink) {
1974 if (!(nmp->nm_sockflags & NMSOCK_READY) ||
1975 (nmp->nm_state & (NFSSTA_RECOVER|NFSSTA_UNMOUNTING|NFSSTA_FORCE|NFSSTA_DEAD)))
1976 break;
1977 np->n_mflag |= NMMONSCANINPROG;
1978 lck_mtx_unlock(&nmp->nm_lock);
1979 error = nfs_getattr(np, NULL, vfs_context_kernel(), (NGA_UNCACHED|NGA_MONITOR));
1980 if (!error && ISSET(np->n_flag, NUPDATESIZE)) /* update quickly to avoid multiple events */
1981 nfs_data_update_size(np, 0);
1982 lck_mtx_lock(&nmp->nm_lock);
1983 np->n_mflag &= ~NMMONSCANINPROG;
1984 if (np->n_mflag & NMMONSCANWANT) {
1985 np->n_mflag &= ~NMMONSCANWANT;
1986 wakeup(&np->n_mflag);
1987 }
1988 if (error || !(nmp->nm_sockflags & NMSOCK_READY) ||
1989 (nmp->nm_state & (NFSSTA_RECOVER|NFSSTA_UNMOUNTING|NFSSTA_FORCE|NFSSTA_DEAD)))
1990 break;
1991 }
1992 nmp->nm_state &= ~NFSSTA_MONITOR_SCAN;
1993 if (nmp->nm_state & NFSSTA_UNMOUNTING)
1994 wakeup(&nmp->nm_state); /* let unmounting thread know scan is done */
1995 }
1996 if ((nmp->nm_sockflags & NMSOCK_READY) || (nmp->nm_state & (NFSSTA_RECOVER|NFSSTA_UNMOUNTING))) {
1997 if (nmp->nm_deadto_start || !TAILQ_EMPTY(&nmp->nm_resendq) ||
1998 (nmp->nm_state & NFSSTA_RECOVER))
1999 ts.tv_sec = 1;
2000 else
2001 ts.tv_sec = 5;
2002 msleep(&nmp->nm_sockthd, &nmp->nm_lock, PSOCK, "nfssockthread", &ts);
2003 }
2004 }
2005
2006 /* If we're unmounting, send the unmount RPC, if requested/appropriate. */
2007 if ((nmp->nm_sockflags & NMSOCK_UNMOUNT) &&
2008 (nmp->nm_state & NFSSTA_MOUNTED) && NMFLAG(nmp, CALLUMNT) &&
2009 (nmp->nm_vers < NFS_VER4) && !(nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD))) {
2010 lck_mtx_unlock(&nmp->nm_lock);
2011 nfs3_umount_rpc(nmp, vfs_context_kernel(),
2012 (nmp->nm_sockflags & NMSOCK_READY) ? 6 : 2);
2013 lck_mtx_lock(&nmp->nm_lock);
2014 }
2015
2016 if (nmp->nm_sockthd == thd)
2017 nmp->nm_sockthd = NULL;
2018 lck_mtx_unlock(&nmp->nm_lock);
2019 wakeup(&nmp->nm_sockthd);
2020 thread_terminate(thd);
2021}
2022
2023/* start or wake a mount's socket thread */
2024void
2025nfs_mount_sock_thread_wake(struct nfsmount *nmp)
2026{
2027 if (nmp->nm_sockthd)
2028 wakeup(&nmp->nm_sockthd);
2029 else if (kernel_thread_start(nfs_mount_sock_thread, nmp, &nmp->nm_sockthd) == KERN_SUCCESS)
2030 thread_deallocate(nmp->nm_sockthd);
2031}
2032
2033/*
2034 * Check if we should mark the mount dead because the
2035 * unresponsive mount has reached the dead timeout.
2036 * (must be called with nmp locked)
2037 */
2038int
2039nfs_mount_check_dead_timeout(struct nfsmount *nmp)
2040{
2041 struct timeval now;
2042
2043 if (nmp->nm_state & NFSSTA_DEAD)
2044 return 1;
2045 if (nmp->nm_deadto_start == 0)
2046 return 0;
2047 nfs_is_squishy(nmp);
2048 if (nmp->nm_curdeadtimeout <= 0)
2049 return 0;
2050 microuptime(&now);
2051 if ((now.tv_sec - nmp->nm_deadto_start) < nmp->nm_curdeadtimeout)
2052 return 0;
2053 return 1;
2054}
2055
2056/*
2057 * Call nfs_mount_zombie to remove most of the
2058 * nfs state for the mount, and then ask to be forcibly unmounted.
2059 *
2060 * Assumes the nfs mount structure lock nm_lock is held.
2061 */
2062
2063void
2064nfs_mount_make_zombie(struct nfsmount *nmp)
2065{
2066 fsid_t fsid;
2067
2068 if (!nmp)
2069 return;
2070
2071 if (nmp->nm_state & NFSSTA_DEAD)
2072 return;
2073
2074 printf("nfs server %s: %sdead\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname,
2075 (nmp->nm_curdeadtimeout != nmp->nm_deadtimeout) ? "squished " : "");
2076 fsid = vfs_statfs(nmp->nm_mountp)->f_fsid;
2077 lck_mtx_unlock(&nmp->nm_lock);
2078 nfs_mount_zombie(nmp, NFSSTA_DEAD);
2079 vfs_event_signal(&fsid, VQ_DEAD, 0);
2080 lck_mtx_lock(&nmp->nm_lock);
2081}
2082
2083
2084/*
2085 * NFS callback channel socket state
2086 */
2087struct nfs_callback_socket
2088{
2089 TAILQ_ENTRY(nfs_callback_socket) ncbs_link;
2090 socket_t ncbs_so; /* the socket */
2091 struct sockaddr_storage ncbs_saddr; /* socket address */
2092 struct nfs_rpc_record_state ncbs_rrs; /* RPC record parsing state */
2093 time_t ncbs_stamp; /* last accessed at */
2094 uint32_t ncbs_flags; /* see below */
2095};
2096#define NCBSOCK_UPCALL 0x0001
2097#define NCBSOCK_UPCALLWANT 0x0002
2098#define NCBSOCK_DEAD 0x0004
2099
2100/*
2101 * NFS callback channel state
2102 *
2103 * One listening socket for accepting socket connections from servers and
2104 * a list of connected sockets to handle callback requests on.
2105 * Mounts registered with the callback channel are assigned IDs and
2106 * put on a list so that the callback request handling code can match
2107 * the requests up with mounts.
2108 */
2109socket_t nfs4_cb_so = NULL;
2110socket_t nfs4_cb_so6 = NULL;
2111in_port_t nfs4_cb_port = 0;
2112in_port_t nfs4_cb_port6 = 0;
2113uint32_t nfs4_cb_id = 0;
2114uint32_t nfs4_cb_so_usecount = 0;
2115TAILQ_HEAD(nfs4_cb_sock_list,nfs_callback_socket) nfs4_cb_socks;
2116TAILQ_HEAD(nfs4_cb_mount_list,nfsmount) nfs4_cb_mounts;
2117
2118int nfs4_cb_handler(struct nfs_callback_socket *, mbuf_t);
2119
2120/*
2121 * Set up the callback channel for the NFS mount.
2122 *
2123 * Initializes the callback channel socket state and
2124 * assigns a callback ID to the mount.
2125 */
2126void
2127nfs4_mount_callback_setup(struct nfsmount *nmp)
2128{
2129 struct sockaddr_in sin;
2130 struct sockaddr_in6 sin6;
2131 socket_t so = NULL;
2132 socket_t so6 = NULL;
2133 struct timeval timeo;
2134 int error, on = 1;
2135 in_port_t port;
2136
2137 lck_mtx_lock(nfs_global_mutex);
2138 if (nfs4_cb_id == 0) {
2139 TAILQ_INIT(&nfs4_cb_mounts);
2140 TAILQ_INIT(&nfs4_cb_socks);
2141 nfs4_cb_id++;
2142 }
2143 nmp->nm_cbid = nfs4_cb_id++;
2144 if (nmp->nm_cbid == 0)
2145 nmp->nm_cbid = nfs4_cb_id++;
2146 nfs4_cb_so_usecount++;
2147 TAILQ_INSERT_HEAD(&nfs4_cb_mounts, nmp, nm_cblink);
2148
2149 if (nfs4_cb_so) {
2150 lck_mtx_unlock(nfs_global_mutex);
2151 return;
2152 }
2153
2154 /* IPv4 */
2155 error = sock_socket(AF_INET, SOCK_STREAM, IPPROTO_TCP, nfs4_cb_accept, NULL, &nfs4_cb_so);
2156 if (error) {
2157 log(LOG_INFO, "nfs callback setup: error %d creating listening IPv4 socket\n", error);
2158 goto fail;
2159 }
2160 so = nfs4_cb_so;
2161
2162 sock_setsockopt(so, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
2163 sin.sin_len = sizeof(struct sockaddr_in);
2164 sin.sin_family = AF_INET;
2165 sin.sin_addr.s_addr = htonl(INADDR_ANY);
2166 sin.sin_port = htons(nfs_callback_port); /* try to use specified port */
2167 error = sock_bind(so, (struct sockaddr *)&sin);
2168 if (error) {
2169 log(LOG_INFO, "nfs callback setup: error %d binding listening IPv4 socket\n", error);
2170 goto fail;
2171 }
2172 error = sock_getsockname(so, (struct sockaddr *)&sin, sin.sin_len);
2173 if (error) {
2174 log(LOG_INFO, "nfs callback setup: error %d getting listening IPv4 socket port\n", error);
2175 goto fail;
2176 }
2177 nfs4_cb_port = ntohs(sin.sin_port);
2178
2179 error = sock_listen(so, 32);
2180 if (error) {
2181 log(LOG_INFO, "nfs callback setup: error %d on IPv4 listen\n", error);
2182 goto fail;
2183 }
2184
2185 /* receive timeout shouldn't matter. If timeout on send, we'll want to drop the socket */
2186 timeo.tv_usec = 0;
2187 timeo.tv_sec = 60;
2188 error = sock_setsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo));
2189 if (error)
2190 log(LOG_INFO, "nfs callback setup: error %d setting IPv4 socket rx timeout\n", error);
2191 error = sock_setsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo));
2192 if (error)
2193 log(LOG_INFO, "nfs callback setup: error %d setting IPv4 socket tx timeout\n", error);
2194 sock_setsockopt(so, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on));
2195 sock_setsockopt(so, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on));
2196 sock_setsockopt(so, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on));
2197 error = 0;
2198
2199 /* IPv6 */
2200 error = sock_socket(AF_INET6, SOCK_STREAM, IPPROTO_TCP, nfs4_cb_accept, NULL, &nfs4_cb_so6);
2201 if (error) {
2202 log(LOG_INFO, "nfs callback setup: error %d creating listening IPv6 socket\n", error);
2203 goto fail;
2204 }
2205 so6 = nfs4_cb_so6;
2206
2207 sock_setsockopt(so6, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
2208 sock_setsockopt(so6, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof(on));
2209 /* try to use specified port or same port as IPv4 */
2210 port = nfs_callback_port ? nfs_callback_port : nfs4_cb_port;
2211ipv6_bind_again:
2212 sin6.sin6_len = sizeof(struct sockaddr_in6);
2213 sin6.sin6_family = AF_INET6;
2214 sin6.sin6_addr = in6addr_any;
2215 sin6.sin6_port = htons(port);
2216 error = sock_bind(so6, (struct sockaddr *)&sin6);
2217 if (error) {
2218 if (port != nfs_callback_port) {
2219 /* if we simply tried to match the IPv4 port, then try any port */
2220 port = 0;
2221 goto ipv6_bind_again;
2222 }
2223 log(LOG_INFO, "nfs callback setup: error %d binding listening IPv6 socket\n", error);
2224 goto fail;
2225 }
2226 error = sock_getsockname(so6, (struct sockaddr *)&sin6, sin6.sin6_len);
2227 if (error) {
2228 log(LOG_INFO, "nfs callback setup: error %d getting listening IPv6 socket port\n", error);
2229 goto fail;
2230 }
2231 nfs4_cb_port6 = ntohs(sin6.sin6_port);
2232
2233 error = sock_listen(so6, 32);
2234 if (error) {
2235 log(LOG_INFO, "nfs callback setup: error %d on IPv6 listen\n", error);
2236 goto fail;
2237 }
2238
2239 /* receive timeout shouldn't matter. If timeout on send, we'll want to drop the socket */
2240 timeo.tv_usec = 0;
2241 timeo.tv_sec = 60;
2242 error = sock_setsockopt(so6, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo));
2243 if (error)
2244 log(LOG_INFO, "nfs callback setup: error %d setting IPv6 socket rx timeout\n", error);
2245 error = sock_setsockopt(so6, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo));
2246 if (error)
2247 log(LOG_INFO, "nfs callback setup: error %d setting IPv6 socket tx timeout\n", error);
2248 sock_setsockopt(so6, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on));
2249 sock_setsockopt(so6, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on));
2250 sock_setsockopt(so6, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on));
2251 error = 0;
2252
2253fail:
2254 if (error) {
2255 nfs4_cb_so = nfs4_cb_so6 = NULL;
2256 lck_mtx_unlock(nfs_global_mutex);
2257 if (so) {
2258 sock_shutdown(so, SHUT_RDWR);
2259 sock_close(so);
2260 }
2261 if (so6) {
2262 sock_shutdown(so6, SHUT_RDWR);
2263 sock_close(so6);
2264 }
2265 } else {
2266 lck_mtx_unlock(nfs_global_mutex);
2267 }
2268}
2269
2270/*
2271 * Shut down the callback channel for the NFS mount.
2272 *
2273 * Clears the mount's callback ID and releases the mounts
2274 * reference on the callback socket. Last reference dropped
2275 * will also shut down the callback socket(s).
2276 */
2277void
2278nfs4_mount_callback_shutdown(struct nfsmount *nmp)
2279{
2280 struct nfs_callback_socket *ncbsp;
2281 socket_t so, so6;
2282 struct nfs4_cb_sock_list cb_socks;
2283 struct timespec ts = {1,0};
2284
2285 lck_mtx_lock(nfs_global_mutex);
2286 TAILQ_REMOVE(&nfs4_cb_mounts, nmp, nm_cblink);
2287 /* wait for any callbacks in progress to complete */
2288 while (nmp->nm_cbrefs)
2289 msleep(&nmp->nm_cbrefs, nfs_global_mutex, PSOCK, "cbshutwait", &ts);
2290 nmp->nm_cbid = 0;
2291 if (--nfs4_cb_so_usecount) {
2292 lck_mtx_unlock(nfs_global_mutex);
2293 return;
2294 }
2295 so = nfs4_cb_so;
2296 so6 = nfs4_cb_so6;
2297 nfs4_cb_so = nfs4_cb_so6 = NULL;
2298 TAILQ_INIT(&cb_socks);
2299 TAILQ_CONCAT(&cb_socks, &nfs4_cb_socks, ncbs_link);
2300 lck_mtx_unlock(nfs_global_mutex);
2301 if (so) {
2302 sock_shutdown(so, SHUT_RDWR);
2303 sock_close(so);
2304 }
2305 if (so6) {
2306 sock_shutdown(so6, SHUT_RDWR);
2307 sock_close(so6);
2308 }
2309 while ((ncbsp = TAILQ_FIRST(&cb_socks))) {
2310 TAILQ_REMOVE(&cb_socks, ncbsp, ncbs_link);
2311 sock_shutdown(ncbsp->ncbs_so, SHUT_RDWR);
2312 sock_close(ncbsp->ncbs_so);
2313 nfs_rpc_record_state_cleanup(&ncbsp->ncbs_rrs);
2314 FREE(ncbsp, M_TEMP);
2315 }
2316}
2317
2318/*
2319 * Check periodically for stale/unused nfs callback sockets
2320 */
2321#define NFS4_CB_TIMER_PERIOD 30
2322#define NFS4_CB_IDLE_MAX 300
2323void
2324nfs4_callback_timer(__unused void *param0, __unused void *param1)
2325{
2326 struct nfs_callback_socket *ncbsp, *nextncbsp;
2327 struct timeval now;
2328
2329loop:
2330 lck_mtx_lock(nfs_global_mutex);
2331 if (TAILQ_EMPTY(&nfs4_cb_socks)) {
2332 nfs4_callback_timer_on = 0;
2333 lck_mtx_unlock(nfs_global_mutex);
2334 return;
2335 }
2336 microuptime(&now);
2337 TAILQ_FOREACH_SAFE(ncbsp, &nfs4_cb_socks, ncbs_link, nextncbsp) {
2338 if (!(ncbsp->ncbs_flags & NCBSOCK_DEAD) &&
2339 (now.tv_sec < (ncbsp->ncbs_stamp + NFS4_CB_IDLE_MAX)))
2340 continue;
2341 TAILQ_REMOVE(&nfs4_cb_socks, ncbsp, ncbs_link);
2342 lck_mtx_unlock(nfs_global_mutex);
2343 sock_shutdown(ncbsp->ncbs_so, SHUT_RDWR);
2344 sock_close(ncbsp->ncbs_so);
2345 nfs_rpc_record_state_cleanup(&ncbsp->ncbs_rrs);
2346 FREE(ncbsp, M_TEMP);
2347 goto loop;
2348 }
2349 nfs4_callback_timer_on = 1;
2350 nfs_interval_timer_start(nfs4_callback_timer_call,
2351 NFS4_CB_TIMER_PERIOD * 1000);
2352 lck_mtx_unlock(nfs_global_mutex);
2353}
2354
2355/*
2356 * Accept a new callback socket.
2357 */
2358void
2359nfs4_cb_accept(socket_t so, __unused void *arg, __unused int waitflag)
2360{
2361 socket_t newso = NULL;
2362 struct nfs_callback_socket *ncbsp;
2363 struct nfsmount *nmp;
2364 struct timeval timeo, now;
2365 int error, on = 1, ip;
2366
2367 if (so == nfs4_cb_so)
2368 ip = 4;
2369 else if (so == nfs4_cb_so6)
2370 ip = 6;
2371 else
2372 return;
2373
2374 /* allocate/initialize a new nfs_callback_socket */
2375 MALLOC(ncbsp, struct nfs_callback_socket *, sizeof(struct nfs_callback_socket), M_TEMP, M_WAITOK);
2376 if (!ncbsp) {
2377 log(LOG_ERR, "nfs callback accept: no memory for new socket\n");
2378 return;
2379 }
2380 bzero(ncbsp, sizeof(*ncbsp));
2381 ncbsp->ncbs_saddr.ss_len = (ip == 4) ? sizeof(struct sockaddr_in) : sizeof(struct sockaddr_in6);
2382 nfs_rpc_record_state_init(&ncbsp->ncbs_rrs);
2383
2384 /* accept a new socket */
2385 error = sock_accept(so, (struct sockaddr*)&ncbsp->ncbs_saddr,
2386 ncbsp->ncbs_saddr.ss_len, MSG_DONTWAIT,
2387 nfs4_cb_rcv, ncbsp, &newso);
2388 if (error) {
2389 log(LOG_INFO, "nfs callback accept: error %d accepting IPv%d socket\n", error, ip);
2390 FREE(ncbsp, M_TEMP);
2391 return;
2392 }
2393
2394 /* set up the new socket */
2395 /* receive timeout shouldn't matter. If timeout on send, we'll want to drop the socket */
2396 timeo.tv_usec = 0;
2397 timeo.tv_sec = 60;
2398 error = sock_setsockopt(newso, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo));
2399 if (error)
2400 log(LOG_INFO, "nfs callback socket: error %d setting IPv%d socket rx timeout\n", error, ip);
2401 error = sock_setsockopt(newso, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo));
2402 if (error)
2403 log(LOG_INFO, "nfs callback socket: error %d setting IPv%d socket tx timeout\n", error, ip);
2404 sock_setsockopt(newso, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on));
2405 sock_setsockopt(newso, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
2406 sock_setsockopt(newso, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on));
2407 sock_setsockopt(newso, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on));
2408
2409 ncbsp->ncbs_so = newso;
2410 microuptime(&now);
2411 ncbsp->ncbs_stamp = now.tv_sec;
2412
2413 lck_mtx_lock(nfs_global_mutex);
2414
2415 /* add it to the list */
2416 TAILQ_INSERT_HEAD(&nfs4_cb_socks, ncbsp, ncbs_link);
2417
2418 /* verify it's from a host we have mounted */
2419 TAILQ_FOREACH(nmp, &nfs4_cb_mounts, nm_cblink) {
2420 /* check if socket's source address matches this mount's server address */
2421 if (!nmp->nm_saddr)
2422 continue;
2423 if (nfs_sockaddr_cmp((struct sockaddr*)&ncbsp->ncbs_saddr, nmp->nm_saddr) == 0)
2424 break;
2425 }
2426 if (!nmp) /* we don't want this socket, mark it dead */
2427 ncbsp->ncbs_flags |= NCBSOCK_DEAD;
2428
2429 /* make sure the callback socket cleanup timer is running */
2430 /* (shorten the timer if we've got a socket we don't want) */
2431 if (!nfs4_callback_timer_on) {
2432 nfs4_callback_timer_on = 1;
2433 nfs_interval_timer_start(nfs4_callback_timer_call,
2434 !nmp ? 500 : (NFS4_CB_TIMER_PERIOD * 1000));
2435 } else if (!nmp && (nfs4_callback_timer_on < 2)) {
2436 nfs4_callback_timer_on = 2;
2437 thread_call_cancel(nfs4_callback_timer_call);
2438 nfs_interval_timer_start(nfs4_callback_timer_call, 500);
2439 }
2440
2441 lck_mtx_unlock(nfs_global_mutex);
2442}
2443
2444/*
2445 * Receive mbufs from callback sockets into RPC records and process each record.
2446 * Detect connection has been closed and shut down.
2447 */
2448void
2449nfs4_cb_rcv(socket_t so, void *arg, __unused int waitflag)
2450{
2451 struct nfs_callback_socket *ncbsp = arg;
2452 struct timespec ts = {1,0};
2453 struct timeval now;
2454 mbuf_t m;
2455 int error = 0, recv = 1;
2456
2457 lck_mtx_lock(nfs_global_mutex);
2458 while (ncbsp->ncbs_flags & NCBSOCK_UPCALL) {
2459 /* wait if upcall is already in progress */
2460 ncbsp->ncbs_flags |= NCBSOCK_UPCALLWANT;
2461 msleep(ncbsp, nfs_global_mutex, PSOCK, "cbupcall", &ts);
2462 }
2463 ncbsp->ncbs_flags |= NCBSOCK_UPCALL;
2464 lck_mtx_unlock(nfs_global_mutex);
2465
2466 /* loop while we make error-free progress */
2467 while (!error && recv) {
2468 error = nfs_rpc_record_read(so, &ncbsp->ncbs_rrs, MSG_DONTWAIT, &recv, &m);
2469 if (m) /* handle the request */
2470 error = nfs4_cb_handler(ncbsp, m);
2471 }
2472
2473 /* note: no error and no data indicates server closed its end */
2474 if ((error != EWOULDBLOCK) && (error || !recv)) {
2475 /*
2476 * Socket is either being closed or should be.
2477 * We can't close the socket in the context of the upcall.
2478 * So we mark it as dead and leave it for the cleanup timer to reap.
2479 */
2480 ncbsp->ncbs_stamp = 0;
2481 ncbsp->ncbs_flags |= NCBSOCK_DEAD;
2482 } else {
2483 microuptime(&now);
2484 ncbsp->ncbs_stamp = now.tv_sec;
2485 }
2486
2487 lck_mtx_lock(nfs_global_mutex);
2488 ncbsp->ncbs_flags &= ~NCBSOCK_UPCALL;
2489 lck_mtx_unlock(nfs_global_mutex);
2490 wakeup(ncbsp);
2491}
2492
2493/*
2494 * Handle an NFS callback channel request.
2495 */
2496int
2497nfs4_cb_handler(struct nfs_callback_socket *ncbsp, mbuf_t mreq)
2498{
2499 socket_t so = ncbsp->ncbs_so;
2500 struct nfsm_chain nmreq, nmrep;
2501 mbuf_t mhead = NULL, mrest = NULL, m;
2502 struct msghdr msg;
2503 struct nfsmount *nmp;
2504 fhandle_t fh;
2505 nfsnode_t np;
2506 nfs_stateid stateid;
2507 uint32_t bitmap[NFS_ATTR_BITMAP_LEN], rbitmap[NFS_ATTR_BITMAP_LEN], bmlen, truncate, attrbytes;
2508 uint32_t val, xid, procnum, taglen, cbid, numops, op, status;
2509 uint32_t auth_type, auth_len;
2510 uint32_t numres, *pnumres;
2511 int error = 0, replen, len;
2512 size_t sentlen = 0;
2513
2514 xid = numops = op = status = procnum = taglen = cbid = 0;
2515
2516 nfsm_chain_dissect_init(error, &nmreq, mreq);
2517 nfsm_chain_get_32(error, &nmreq, xid); // RPC XID
2518 nfsm_chain_get_32(error, &nmreq, val); // RPC Call
2519 nfsm_assert(error, (val == RPC_CALL), EBADRPC);
2520 nfsm_chain_get_32(error, &nmreq, val); // RPC Version
2521 nfsm_assert(error, (val == RPC_VER2), ERPCMISMATCH);
2522 nfsm_chain_get_32(error, &nmreq, val); // RPC Program Number
2523 nfsm_assert(error, (val == NFS4_CALLBACK_PROG), EPROGUNAVAIL);
2524 nfsm_chain_get_32(error, &nmreq, val); // NFS Callback Program Version Number
2525 nfsm_assert(error, (val == NFS4_CALLBACK_PROG_VERSION), EPROGMISMATCH);
2526 nfsm_chain_get_32(error, &nmreq, procnum); // NFS Callback Procedure Number
2527 nfsm_assert(error, (procnum <= NFSPROC4_CB_COMPOUND), EPROCUNAVAIL);
2528
2529 /* Handle authentication */
2530 /* XXX just ignore auth for now - handling kerberos may be tricky */
2531 nfsm_chain_get_32(error, &nmreq, auth_type); // RPC Auth Flavor
2532 nfsm_chain_get_32(error, &nmreq, auth_len); // RPC Auth Length
2533 nfsm_assert(error, (auth_len <= RPCAUTH_MAXSIZ), EBADRPC);
2534 if (!error && (auth_len > 0))
2535 nfsm_chain_adv(error, &nmreq, nfsm_rndup(auth_len));
2536 nfsm_chain_adv(error, &nmreq, NFSX_UNSIGNED); // verifier flavor (should be AUTH_NONE)
2537 nfsm_chain_get_32(error, &nmreq, auth_len); // verifier length
2538 nfsm_assert(error, (auth_len <= RPCAUTH_MAXSIZ), EBADRPC);
2539 if (!error && (auth_len > 0))
2540 nfsm_chain_adv(error, &nmreq, nfsm_rndup(auth_len));
2541 if (error) {
2542 status = error;
2543 error = 0;
2544 goto nfsmout;
2545 }
2546
2547 switch (procnum) {
2548 case NFSPROC4_CB_NULL:
2549 status = NFSERR_RETVOID;
2550 break;
2551 case NFSPROC4_CB_COMPOUND:
2552 /* tag, minorversion, cb ident, numops, op array */
2553 nfsm_chain_get_32(error, &nmreq, taglen); /* tag length */
2554 nfsm_assert(error, (val <= NFS4_OPAQUE_LIMIT), EBADRPC);
2555
2556 /* start building the body of the response */
2557 nfsm_mbuf_get(error, &mrest, nfsm_rndup(taglen) + 5*NFSX_UNSIGNED);
2558 nfsm_chain_init(&nmrep, mrest);
2559
2560 /* copy tag from request to response */
2561 nfsm_chain_add_32(error, &nmrep, taglen); /* tag length */
2562 for (len = (int)taglen; !error && (len > 0); len -= NFSX_UNSIGNED) {
2563 nfsm_chain_get_32(error, &nmreq, val);
2564 nfsm_chain_add_32(error, &nmrep, val);
2565 }
2566
2567 /* insert number of results placeholder */
2568 numres = 0;
2569 nfsm_chain_add_32(error, &nmrep, numres);
2570 pnumres = (uint32_t*)(nmrep.nmc_ptr - NFSX_UNSIGNED);
2571
2572 nfsm_chain_get_32(error, &nmreq, val); /* minorversion */
2573 nfsm_assert(error, (val == 0), NFSERR_MINOR_VERS_MISMATCH);
2574 nfsm_chain_get_32(error, &nmreq, cbid); /* callback ID */
2575 nfsm_chain_get_32(error, &nmreq, numops); /* number of operations */
2576 if (error) {
2577 if ((error == EBADRPC) || (error == NFSERR_MINOR_VERS_MISMATCH))
2578 status = error;
2579 else if ((error == ENOBUFS) || (error == ENOMEM))
2580 status = NFSERR_RESOURCE;
2581 else
2582 status = NFSERR_SERVERFAULT;
2583 error = 0;
2584 nfsm_chain_null(&nmrep);
2585 goto nfsmout;
2586 }
2587 /* match the callback ID to a registered mount */
2588 lck_mtx_lock(nfs_global_mutex);
2589 TAILQ_FOREACH(nmp, &nfs4_cb_mounts, nm_cblink) {
2590 if (nmp->nm_cbid != cbid)
2591 continue;
2592 /* verify socket's source address matches this mount's server address */
2593 if (!nmp->nm_saddr)
2594 continue;
2595 if (nfs_sockaddr_cmp((struct sockaddr*)&ncbsp->ncbs_saddr, nmp->nm_saddr) == 0)
2596 break;
2597 }
2598 /* mark the NFS mount as busy */
2599 if (nmp)
2600 nmp->nm_cbrefs++;
2601 lck_mtx_unlock(nfs_global_mutex);
2602 if (!nmp) {
2603 /* if no mount match, just drop socket. */
2604 error = EPERM;
2605 nfsm_chain_null(&nmrep);
2606 goto out;
2607 }
2608
2609 /* process ops, adding results to mrest */
2610 while (numops > 0) {
2611 numops--;
2612 nfsm_chain_get_32(error, &nmreq, op);
2613 if (error)
2614 break;
2615 switch (op) {
2616 case NFS_OP_CB_GETATTR:
2617 // (FH, BITMAP) -> (STATUS, BITMAP, ATTRS)
2618 np = NULL;
2619 nfsm_chain_get_fh(error, &nmreq, NFS_VER4, &fh);
2620 bmlen = NFS_ATTR_BITMAP_LEN;
2621 nfsm_chain_get_bitmap(error, &nmreq, bitmap, bmlen);
2622 if (error) {
2623 status = error;
2624 error = 0;
2625 numops = 0; /* don't process any more ops */
2626 } else {
2627 /* find the node for the file handle */
2628 error = nfs_nget(nmp->nm_mountp, NULL, NULL, fh.fh_data, fh.fh_len, NULL, NULL, RPCAUTH_UNKNOWN, NG_NOCREATE, &np);
2629 if (error || !np) {
2630 status = NFSERR_BADHANDLE;
2631 error = 0;
2632 np = NULL;
2633 numops = 0; /* don't process any more ops */
2634 }
2635 }
2636 nfsm_chain_add_32(error, &nmrep, op);
2637 nfsm_chain_add_32(error, &nmrep, status);
2638 if (!error && (status == EBADRPC))
2639 error = status;
2640 if (np) {
2641 /* only allow returning size, change, and mtime attrs */
2642 NFS_CLEAR_ATTRIBUTES(&rbitmap);
2643 attrbytes = 0;
2644 if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_CHANGE)) {
2645 NFS_BITMAP_SET(&rbitmap, NFS_FATTR_CHANGE);
2646 attrbytes += 2 * NFSX_UNSIGNED;
2647 }
2648 if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_SIZE)) {
2649 NFS_BITMAP_SET(&rbitmap, NFS_FATTR_SIZE);
2650 attrbytes += 2 * NFSX_UNSIGNED;
2651 }
2652 if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_TIME_MODIFY)) {
2653 NFS_BITMAP_SET(&rbitmap, NFS_FATTR_TIME_MODIFY);
2654 attrbytes += 3 * NFSX_UNSIGNED;
2655 }
2656 nfsm_chain_add_bitmap(error, &nmrep, rbitmap, NFS_ATTR_BITMAP_LEN);
2657 nfsm_chain_add_32(error, &nmrep, attrbytes);
2658 if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_CHANGE))
2659 nfsm_chain_add_64(error, &nmrep,
2660 np->n_vattr.nva_change + ((np->n_flag & NMODIFIED) ? 1 : 0));
2661 if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_SIZE))
2662 nfsm_chain_add_64(error, &nmrep, np->n_size);
2663 if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_TIME_MODIFY)) {
2664 nfsm_chain_add_64(error, &nmrep, np->n_vattr.nva_timesec[NFSTIME_MODIFY]);
2665 nfsm_chain_add_32(error, &nmrep, np->n_vattr.nva_timensec[NFSTIME_MODIFY]);
2666 }
2667 nfs_node_unlock(np);
2668 vnode_put(NFSTOV(np));
2669 np = NULL;
2670 }
2671 /*
2672 * If we hit an error building the reply, we can't easily back up.
2673 * So we'll just update the status and hope the server ignores the
2674 * extra garbage.
2675 */
2676 break;
2677 case NFS_OP_CB_RECALL:
2678 // (STATEID, TRUNCATE, FH) -> (STATUS)
2679 np = NULL;
2680 nfsm_chain_get_stateid(error, &nmreq, &stateid);
2681 nfsm_chain_get_32(error, &nmreq, truncate);
2682 nfsm_chain_get_fh(error, &nmreq, NFS_VER4, &fh);
2683 if (error) {
2684 status = error;
2685 error = 0;
2686 numops = 0; /* don't process any more ops */
2687 } else {
2688 /* find the node for the file handle */
2689 error = nfs_nget(nmp->nm_mountp, NULL, NULL, fh.fh_data, fh.fh_len, NULL, NULL, RPCAUTH_UNKNOWN, NG_NOCREATE, &np);
2690 if (error || !np) {
2691 status = NFSERR_BADHANDLE;
2692 error = 0;
2693 np = NULL;
2694 numops = 0; /* don't process any more ops */
2695 } else if (!(np->n_openflags & N_DELEG_MASK) ||
2696 bcmp(&np->n_dstateid, &stateid, sizeof(stateid))) {
2697 /* delegation stateid state doesn't match */
2698 status = NFSERR_BAD_STATEID;
2699 numops = 0; /* don't process any more ops */
2700 }
2701 if (!status) /* add node to recall queue, and wake socket thread */
2702 nfs4_delegation_return_enqueue(np);
2703 if (np) {
2704 nfs_node_unlock(np);
2705 vnode_put(NFSTOV(np));
2706 }
2707 }
2708 nfsm_chain_add_32(error, &nmrep, op);
2709 nfsm_chain_add_32(error, &nmrep, status);
2710 if (!error && (status == EBADRPC))
2711 error = status;
2712 break;
2713 case NFS_OP_CB_ILLEGAL:
2714 default:
2715 nfsm_chain_add_32(error, &nmrep, NFS_OP_CB_ILLEGAL);
2716 status = NFSERR_OP_ILLEGAL;
2717 nfsm_chain_add_32(error, &nmrep, status);
2718 numops = 0; /* don't process any more ops */
2719 break;
2720 }
2721 numres++;
2722 }
2723
2724 if (!status && error) {
2725 if (error == EBADRPC)
2726 status = error;
2727 else if ((error == ENOBUFS) || (error == ENOMEM))
2728 status = NFSERR_RESOURCE;
2729 else
2730 status = NFSERR_SERVERFAULT;
2731 error = 0;
2732 }
2733
2734 /* Now, set the numres field */
2735 *pnumres = txdr_unsigned(numres);
2736 nfsm_chain_build_done(error, &nmrep);
2737 nfsm_chain_null(&nmrep);
2738
2739 /* drop the callback reference on the mount */
2740 lck_mtx_lock(nfs_global_mutex);
2741 nmp->nm_cbrefs--;
2742 if (!nmp->nm_cbid)
2743 wakeup(&nmp->nm_cbrefs);
2744 lck_mtx_unlock(nfs_global_mutex);
2745 break;
2746 }
2747
2748nfsmout:
2749 if (status == EBADRPC)
2750 OSAddAtomic64(1, &nfsstats.rpcinvalid);
2751
2752 /* build reply header */
2753 error = mbuf_gethdr(MBUF_WAITOK, MBUF_TYPE_DATA, &mhead);
2754 nfsm_chain_init(&nmrep, mhead);
2755 nfsm_chain_add_32(error, &nmrep, 0); /* insert space for an RPC record mark */
2756 nfsm_chain_add_32(error, &nmrep, xid);
2757 nfsm_chain_add_32(error, &nmrep, RPC_REPLY);
2758 if ((status == ERPCMISMATCH) || (status & NFSERR_AUTHERR)) {
2759 nfsm_chain_add_32(error, &nmrep, RPC_MSGDENIED);
2760 if (status & NFSERR_AUTHERR) {
2761 nfsm_chain_add_32(error, &nmrep, RPC_AUTHERR);
2762 nfsm_chain_add_32(error, &nmrep, (status & ~NFSERR_AUTHERR));
2763 } else {
2764 nfsm_chain_add_32(error, &nmrep, RPC_MISMATCH);
2765 nfsm_chain_add_32(error, &nmrep, RPC_VER2);
2766 nfsm_chain_add_32(error, &nmrep, RPC_VER2);
2767 }
2768 } else {
2769 /* reply status */
2770 nfsm_chain_add_32(error, &nmrep, RPC_MSGACCEPTED);
2771 /* XXX RPCAUTH_NULL verifier */
2772 nfsm_chain_add_32(error, &nmrep, RPCAUTH_NULL);
2773 nfsm_chain_add_32(error, &nmrep, 0);
2774 /* accepted status */
2775 switch (status) {
2776 case EPROGUNAVAIL:
2777 nfsm_chain_add_32(error, &nmrep, RPC_PROGUNAVAIL);
2778 break;
2779 case EPROGMISMATCH:
2780 nfsm_chain_add_32(error, &nmrep, RPC_PROGMISMATCH);
2781 nfsm_chain_add_32(error, &nmrep, NFS4_CALLBACK_PROG_VERSION);
2782 nfsm_chain_add_32(error, &nmrep, NFS4_CALLBACK_PROG_VERSION);
2783 break;
2784 case EPROCUNAVAIL:
2785 nfsm_chain_add_32(error, &nmrep, RPC_PROCUNAVAIL);
2786 break;
2787 case EBADRPC:
2788 nfsm_chain_add_32(error, &nmrep, RPC_GARBAGE);
2789 break;
2790 default:
2791 nfsm_chain_add_32(error, &nmrep, RPC_SUCCESS);
2792 if (status != NFSERR_RETVOID)
2793 nfsm_chain_add_32(error, &nmrep, status);
2794 break;
2795 }
2796 }
2797 nfsm_chain_build_done(error, &nmrep);
2798 if (error) {
2799 nfsm_chain_null(&nmrep);
2800 goto out;
2801 }
2802 error = mbuf_setnext(nmrep.nmc_mcur, mrest);
2803 if (error) {
2804 printf("nfs cb: mbuf_setnext failed %d\n", error);
2805 goto out;
2806 }
2807 mrest = NULL;
2808 /* Calculate the size of the reply */
2809 replen = 0;
2810 for (m = nmrep.nmc_mhead; m; m = mbuf_next(m))
2811 replen += mbuf_len(m);
2812 mbuf_pkthdr_setlen(mhead, replen);
2813 error = mbuf_pkthdr_setrcvif(mhead, NULL);
2814 nfsm_chain_set_recmark(error, &nmrep, (replen - NFSX_UNSIGNED) | 0x80000000);
2815 nfsm_chain_null(&nmrep);
2816
2817 /* send the reply */
2818 bzero(&msg, sizeof(msg));
2819 error = sock_sendmbuf(so, &msg, mhead, 0, &sentlen);
2820 mhead = NULL;
2821 if (!error && ((int)sentlen != replen))
2822 error = EWOULDBLOCK;
2823 if (error == EWOULDBLOCK) /* inability to send response is considered fatal */
2824 error = ETIMEDOUT;
2825out:
2826 if (error)
2827 nfsm_chain_cleanup(&nmrep);
2828 if (mhead)
2829 mbuf_freem(mhead);
2830 if (mrest)
2831 mbuf_freem(mrest);
2832 if (mreq)
2833 mbuf_freem(mreq);
2834 return (error);
2835}
2836
2837
2838/*
2839 * Initialize an nfs_rpc_record_state structure.
2840 */
2841void
2842nfs_rpc_record_state_init(struct nfs_rpc_record_state *nrrsp)
2843{
2844 bzero(nrrsp, sizeof(*nrrsp));
2845 nrrsp->nrrs_markerleft = sizeof(nrrsp->nrrs_fragleft);
2846}
2847
2848/*
2849 * Clean up an nfs_rpc_record_state structure.
2850 */
2851void
2852nfs_rpc_record_state_cleanup(struct nfs_rpc_record_state *nrrsp)
2853{
2854 if (nrrsp->nrrs_m) {
2855 mbuf_freem(nrrsp->nrrs_m);
2856 nrrsp->nrrs_m = nrrsp->nrrs_mlast = NULL;
2857 }
2858}
2859
2860/*
2861 * Read the next (marked) RPC record from the socket.
2862 *
2863 * *recvp returns if any data was received.
2864 * *mp returns the next complete RPC record
2865 */
2866int
2867nfs_rpc_record_read(socket_t so, struct nfs_rpc_record_state *nrrsp, int flags, int *recvp, mbuf_t *mp)
2868{
2869 struct iovec aio;
2870 struct msghdr msg;
2871 size_t rcvlen;
2872 int error = 0;
2873 mbuf_t m;
2874
2875 *recvp = 0;
2876 *mp = NULL;
2877
2878 /* read the TCP RPC record marker */
2879 while (!error && nrrsp->nrrs_markerleft) {
2880 aio.iov_base = ((char*)&nrrsp->nrrs_fragleft +
2881 sizeof(nrrsp->nrrs_fragleft) - nrrsp->nrrs_markerleft);
2882 aio.iov_len = nrrsp->nrrs_markerleft;
2883 bzero(&msg, sizeof(msg));
2884 msg.msg_iov = &aio;
2885 msg.msg_iovlen = 1;
2886 error = sock_receive(so, &msg, flags, &rcvlen);
2887 if (error || !rcvlen)
2888 break;
2889 *recvp = 1;
2890 nrrsp->nrrs_markerleft -= rcvlen;
2891 if (nrrsp->nrrs_markerleft)
2892 continue;
2893 /* record marker complete */
2894 nrrsp->nrrs_fragleft = ntohl(nrrsp->nrrs_fragleft);
2895 if (nrrsp->nrrs_fragleft & 0x80000000) {
2896 nrrsp->nrrs_lastfrag = 1;
2897 nrrsp->nrrs_fragleft &= ~0x80000000;
2898 }
2899 nrrsp->nrrs_reclen += nrrsp->nrrs_fragleft;
2900 if (nrrsp->nrrs_reclen > NFS_MAXPACKET) {
2901 /* This is SERIOUS! We are out of sync with the sender. */
2902 log(LOG_ERR, "impossible RPC record length (%d) on callback", nrrsp->nrrs_reclen);
2903 error = EFBIG;
2904 }
2905 }
2906
2907 /* read the TCP RPC record fragment */
2908 while (!error && !nrrsp->nrrs_markerleft && nrrsp->nrrs_fragleft) {
2909 m = NULL;
2910 rcvlen = nrrsp->nrrs_fragleft;
2911 error = sock_receivembuf(so, NULL, &m, flags, &rcvlen);
2912 if (error || !rcvlen || !m)
2913 break;
2914 *recvp = 1;
2915 /* append mbufs to list */
2916 nrrsp->nrrs_fragleft -= rcvlen;
2917 if (!nrrsp->nrrs_m) {
2918 nrrsp->nrrs_m = m;
2919 } else {
2920 error = mbuf_setnext(nrrsp->nrrs_mlast, m);
2921 if (error) {
2922 printf("nfs tcp rcv: mbuf_setnext failed %d\n", error);
2923 mbuf_freem(m);
2924 break;
2925 }
2926 }
2927 while (mbuf_next(m))
2928 m = mbuf_next(m);
2929 nrrsp->nrrs_mlast = m;
2930 }
2931
2932 /* done reading fragment? */
2933 if (!error && !nrrsp->nrrs_markerleft && !nrrsp->nrrs_fragleft) {
2934 /* reset socket fragment parsing state */
2935 nrrsp->nrrs_markerleft = sizeof(nrrsp->nrrs_fragleft);
2936 if (nrrsp->nrrs_lastfrag) {
2937 /* RPC record complete */
2938 *mp = nrrsp->nrrs_m;
2939 /* reset socket record parsing state */
2940 nrrsp->nrrs_reclen = 0;
2941 nrrsp->nrrs_m = nrrsp->nrrs_mlast = NULL;
2942 nrrsp->nrrs_lastfrag = 0;
2943 }
2944 }
2945
2946 return (error);
2947}
2948
2949
2950
2951/*
2952 * The NFS client send routine.
2953 *
2954 * Send the given NFS request out the mount's socket.
2955 * Holds nfs_sndlock() for the duration of this call.
2956 *
2957 * - check for request termination (sigintr)
2958 * - wait for reconnect, if necessary
2959 * - UDP: check the congestion window
2960 * - make a copy of the request to send
2961 * - UDP: update the congestion window
2962 * - send the request
2963 *
2964 * If sent successfully, R_MUSTRESEND and R_RESENDERR are cleared.
2965 * rexmit count is also updated if this isn't the first send.
2966 *
2967 * If the send is not successful, make sure R_MUSTRESEND is set.
2968 * If this wasn't the first transmit, set R_RESENDERR.
2969 * Also, undo any UDP congestion window changes made.
2970 *
2971 * If the error appears to indicate that the socket should
2972 * be reconnected, mark the socket for reconnection.
2973 *
2974 * Only return errors when the request should be aborted.
2975 */
2976int
2977nfs_send(struct nfsreq *req, int wait)
2978{
2979 struct nfsmount *nmp;
2980 struct nfs_socket *nso;
2981 int error, error2, sotype, rexmit, slpflag = 0, needrecon;
2982 struct msghdr msg;
2983 struct sockaddr *sendnam;
2984 mbuf_t mreqcopy;
2985 size_t sentlen = 0;
2986 struct timespec ts = { 2, 0 };
2987
2988again:
2989 error = nfs_sndlock(req);
2990 if (error) {
2991 lck_mtx_lock(&req->r_mtx);
2992 req->r_error = error;
2993 req->r_flags &= ~R_SENDING;
2994 lck_mtx_unlock(&req->r_mtx);
2995 return (error);
2996 }
2997
2998 error = nfs_sigintr(req->r_nmp, req, NULL, 0);
2999 if (error) {
3000 nfs_sndunlock(req);
3001 lck_mtx_lock(&req->r_mtx);
3002 req->r_error = error;
3003 req->r_flags &= ~R_SENDING;
3004 lck_mtx_unlock(&req->r_mtx);
3005 return (error);
3006 }
3007 nmp = req->r_nmp;
3008 sotype = nmp->nm_sotype;
3009
3010 /*
3011 * If it's a setup RPC but we're not in SETUP... must need reconnect.
3012 * If it's a recovery RPC but the socket's not ready... must need reconnect.
3013 */
3014 if (((req->r_flags & R_SETUP) && !(nmp->nm_sockflags & NMSOCK_SETUP)) ||
3015 ((req->r_flags & R_RECOVER) && !(nmp->nm_sockflags & NMSOCK_READY))) {
3016 error = ETIMEDOUT;
3017 nfs_sndunlock(req);
3018 lck_mtx_lock(&req->r_mtx);
3019 req->r_error = error;
3020 req->r_flags &= ~R_SENDING;
3021 lck_mtx_unlock(&req->r_mtx);
3022 return (error);
3023 }
3024
3025 /* If the socket needs reconnection, do that now. */
3026 /* wait until socket is ready - unless this request is part of setup */
3027 lck_mtx_lock(&nmp->nm_lock);
3028 if (!(nmp->nm_sockflags & NMSOCK_READY) &&
3029 !((nmp->nm_sockflags & NMSOCK_SETUP) && (req->r_flags & R_SETUP))) {
3030 if (NMFLAG(nmp, INTR) && !(req->r_flags & R_NOINTR))
3031 slpflag |= PCATCH;
3032 lck_mtx_unlock(&nmp->nm_lock);
3033 nfs_sndunlock(req);
3034 if (!wait) {
3035 lck_mtx_lock(&req->r_mtx);
3036 req->r_flags &= ~R_SENDING;
3037 req->r_flags |= R_MUSTRESEND;
3038 req->r_rtt = 0;
3039 lck_mtx_unlock(&req->r_mtx);
3040 return (0);
3041 }
3042 NFS_SOCK_DBG("nfs_send: 0x%llx wait reconnect\n", req->r_xid);
3043 lck_mtx_lock(&req->r_mtx);
3044 req->r_flags &= ~R_MUSTRESEND;
3045 req->r_rtt = 0;
3046 lck_mtx_unlock(&req->r_mtx);
3047 lck_mtx_lock(&nmp->nm_lock);
3048 while (!(nmp->nm_sockflags & NMSOCK_READY)) {
3049 /* don't bother waiting if the socket thread won't be reconnecting it */
3050 if (nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD)) {
3051 error = EIO;
3052 break;
3053 }
3054 if ((NMFLAG(nmp, SOFT) || (req->r_flags & R_SOFT)) && (nmp->nm_reconnect_start > 0)) {
3055 struct timeval now;
3056 microuptime(&now);
3057 if ((now.tv_sec - nmp->nm_reconnect_start) >= 8) {
3058 /* soft mount in reconnect for a while... terminate ASAP */
3059 OSAddAtomic64(1, &nfsstats.rpctimeouts);
3060 req->r_flags |= R_SOFTTERM;
3061 req->r_error = error = ETIMEDOUT;
3062 break;
3063 }
3064 }
3065 /* make sure socket thread is running, then wait */
3066 nfs_mount_sock_thread_wake(nmp);
3067 if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 1)))
3068 break;
3069 msleep(req, &nmp->nm_lock, slpflag|PSOCK, "nfsconnectwait", &ts);
3070 slpflag = 0;
3071 }
3072 lck_mtx_unlock(&nmp->nm_lock);
3073 if (error) {
3074 lck_mtx_lock(&req->r_mtx);
3075 req->r_error = error;
3076 req->r_flags &= ~R_SENDING;
3077 lck_mtx_unlock(&req->r_mtx);
3078 return (error);
3079 }
3080 goto again;
3081 }
3082 nso = nmp->nm_nso;
3083 /* note that we're using the mount's socket to do the send */
3084 nmp->nm_state |= NFSSTA_SENDING; /* will be cleared by nfs_sndunlock() */
3085 lck_mtx_unlock(&nmp->nm_lock);
3086 if (!nso) {
3087 nfs_sndunlock(req);
3088 lck_mtx_lock(&req->r_mtx);
3089 req->r_flags &= ~R_SENDING;
3090 req->r_flags |= R_MUSTRESEND;
3091 req->r_rtt = 0;
3092 lck_mtx_unlock(&req->r_mtx);
3093 return (0);
3094 }
3095
3096 lck_mtx_lock(&req->r_mtx);
3097 rexmit = (req->r_flags & R_SENT);
3098
3099 if (sotype == SOCK_DGRAM) {
3100 lck_mtx_lock(&nmp->nm_lock);
3101 if (!(req->r_flags & R_CWND) && (nmp->nm_sent >= nmp->nm_cwnd)) {
3102 /* if we can't send this out yet, wait on the cwnd queue */
3103 slpflag = (NMFLAG(nmp, INTR) && req->r_thread) ? PCATCH : 0;
3104 lck_mtx_unlock(&nmp->nm_lock);
3105 nfs_sndunlock(req);
3106 req->r_flags &= ~R_SENDING;
3107 req->r_flags |= R_MUSTRESEND;
3108 lck_mtx_unlock(&req->r_mtx);
3109 if (!wait) {
3110 req->r_rtt = 0;
3111 return (0);
3112 }
3113 lck_mtx_lock(&nmp->nm_lock);
3114 while (nmp->nm_sent >= nmp->nm_cwnd) {
3115 if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 1)))
3116 break;
3117 TAILQ_INSERT_TAIL(&nmp->nm_cwndq, req, r_cchain);
3118 msleep(req, &nmp->nm_lock, slpflag | (PZERO - 1), "nfswaitcwnd", &ts);
3119 slpflag = 0;
3120 if ((req->r_cchain.tqe_next != NFSREQNOLIST)) {
3121 TAILQ_REMOVE(&nmp->nm_cwndq, req, r_cchain);
3122 req->r_cchain.tqe_next = NFSREQNOLIST;
3123 }
3124 }
3125 lck_mtx_unlock(&nmp->nm_lock);
3126 goto again;
3127 }
3128 /*
3129 * We update these *before* the send to avoid racing
3130 * against others who may be looking to send requests.
3131 */
3132 if (!rexmit) {
3133 /* first transmit */
3134 req->r_flags |= R_CWND;
3135 nmp->nm_sent += NFS_CWNDSCALE;
3136 } else {
3137 /*
3138 * When retransmitting, turn timing off
3139 * and divide congestion window by 2.
3140 */
3141 req->r_flags &= ~R_TIMING;
3142 nmp->nm_cwnd >>= 1;
3143 if (nmp->nm_cwnd < NFS_CWNDSCALE)
3144 nmp->nm_cwnd = NFS_CWNDSCALE;
3145 }
3146 lck_mtx_unlock(&nmp->nm_lock);
3147 }
3148
3149 req->r_flags &= ~R_MUSTRESEND;
3150 lck_mtx_unlock(&req->r_mtx);
3151
3152 error = mbuf_copym(req->r_mhead, 0, MBUF_COPYALL,
3153 wait ? MBUF_WAITOK : MBUF_DONTWAIT, &mreqcopy);
3154 if (error) {
3155 if (wait)
3156 log(LOG_INFO, "nfs_send: mbuf copy failed %d\n", error);
3157 nfs_sndunlock(req);
3158 lck_mtx_lock(&req->r_mtx);
3159 req->r_flags &= ~R_SENDING;
3160 req->r_flags |= R_MUSTRESEND;
3161 req->r_rtt = 0;
3162 lck_mtx_unlock(&req->r_mtx);
3163 return (0);
3164 }
3165
3166 bzero(&msg, sizeof(msg));
3167 if ((sotype != SOCK_STREAM) && !sock_isconnected(nso->nso_so) && ((sendnam = nmp->nm_saddr))) {
3168 msg.msg_name = (caddr_t)sendnam;
3169 msg.msg_namelen = sendnam->sa_len;
3170 }
3171 error = sock_sendmbuf(nso->nso_so, &msg, mreqcopy, 0, &sentlen);
3172 if (error || (sentlen != req->r_mreqlen)) {
3173 NFS_SOCK_DBG("nfs_send: 0x%llx sent %d/%d error %d\n",
3174 req->r_xid, (int)sentlen, (int)req->r_mreqlen, error);
3175 }
3176
3177 if (!error && (sentlen != req->r_mreqlen))
3178 error = EWOULDBLOCK;
3179 needrecon = ((sotype == SOCK_STREAM) && sentlen && (sentlen != req->r_mreqlen));
3180
3181 lck_mtx_lock(&req->r_mtx);
3182 req->r_flags &= ~R_SENDING;
3183 req->r_rtt = 0;
3184 if (rexmit && (++req->r_rexmit > NFS_MAXREXMIT))
3185 req->r_rexmit = NFS_MAXREXMIT;
3186
3187 if (!error) {
3188 /* SUCCESS */
3189 req->r_flags &= ~R_RESENDERR;
3190 if (rexmit)
3191 OSAddAtomic64(1, &nfsstats.rpcretries);
3192 req->r_flags |= R_SENT;
3193 if (req->r_flags & R_WAITSENT) {
3194 req->r_flags &= ~R_WAITSENT;
3195 wakeup(req);
3196 }
3197 nfs_sndunlock(req);
3198 lck_mtx_unlock(&req->r_mtx);
3199 return (0);
3200 }
3201
3202 /* send failed */
3203 req->r_flags |= R_MUSTRESEND;
3204 if (rexmit)
3205 req->r_flags |= R_RESENDERR;
3206 if ((error == EINTR) || (error == ERESTART))
3207 req->r_error = error;
3208 lck_mtx_unlock(&req->r_mtx);
3209
3210 if (sotype == SOCK_DGRAM) {
3211 /*
3212 * Note: even though a first send may fail, we consider
3213 * the request sent for congestion window purposes.
3214 * So we don't need to undo any of the changes made above.
3215 */
3216 /*
3217 * Socket errors ignored for connectionless sockets??
3218 * For now, ignore them all
3219 */
3220 if ((error != EINTR) && (error != ERESTART) &&
3221 (error != EWOULDBLOCK) && (error != EIO) && (nso == nmp->nm_nso)) {
3222 int clearerror = 0, optlen = sizeof(clearerror);
3223 sock_getsockopt(nso->nso_so, SOL_SOCKET, SO_ERROR, &clearerror, &optlen);
3224#ifdef NFS_SOCKET_DEBUGGING
3225 if (clearerror)
3226 NFS_SOCK_DBG("nfs_send: ignoring UDP socket error %d so %d\n",
3227 error, clearerror);
3228#endif
3229 }
3230 }
3231
3232 /* check if it appears we should reconnect the socket */
3233 switch (error) {
3234 case EWOULDBLOCK:
3235 /* if send timed out, reconnect if on TCP */
3236 if (sotype != SOCK_STREAM)
3237 break;
3238 case EPIPE:
3239 case EADDRNOTAVAIL:
3240 case ENETDOWN:
3241 case ENETUNREACH:
3242 case ENETRESET:
3243 case ECONNABORTED:
3244 case ECONNRESET:
3245 case ENOTCONN:
3246 case ESHUTDOWN:
3247 case ECONNREFUSED:
3248 case EHOSTDOWN:
3249 case EHOSTUNREACH:
3250 /* case ECANCELED??? */
3251 needrecon = 1;
3252 break;
3253 }
3254 if (needrecon && (nso == nmp->nm_nso)) { /* mark socket as needing reconnect */
3255 NFS_SOCK_DBG("nfs_send: 0x%llx need reconnect %d\n", req->r_xid, error);
3256 nfs_need_reconnect(nmp);
3257 }
3258
3259 nfs_sndunlock(req);
3260
3261 if (nfs_is_dead(error, nmp))
3262 error = EIO;
3263
3264 /*
3265 * Don't log some errors:
3266 * EPIPE errors may be common with servers that drop idle connections.
3267 * EADDRNOTAVAIL may occur on network transitions.
3268 * ENOTCONN may occur under some network conditions.
3269 */
3270 if ((error == EPIPE) || (error == EADDRNOTAVAIL) || (error == ENOTCONN))
3271 error = 0;
3272 if (error && (error != EINTR) && (error != ERESTART))
3273 log(LOG_INFO, "nfs send error %d for server %s\n", error,
3274 !req->r_nmp ? "<unmounted>" :
3275 vfs_statfs(req->r_nmp->nm_mountp)->f_mntfromname);
3276
3277 /* prefer request termination error over other errors */
3278 error2 = nfs_sigintr(req->r_nmp, req, req->r_thread, 0);
3279 if (error2)
3280 error = error2;
3281
3282 /* only allow the following errors to be returned */
3283 if ((error != EINTR) && (error != ERESTART) && (error != EIO) &&
3284 (error != ENXIO) && (error != ETIMEDOUT))
3285 /*
3286 * We got some error we don't know what do do with,
3287 * i.e., we're not reconnecting, we map it to
3288 * EIO. Presumably our send failed and we better tell
3289 * the caller so they don't wait for a reply that is
3290 * never going to come. If we are reconnecting we
3291 * return 0 and the request will be resent.
3292 */
3293 error = needrecon ? 0 : EIO;
3294 return (error);
3295}
3296
3297/*
3298 * NFS client socket upcalls
3299 *
3300 * Pull RPC replies out of an NFS mount's socket and match them
3301 * up with the pending request.
3302 *
3303 * The datagram code is simple because we always get whole
3304 * messages out of the socket.
3305 *
3306 * The stream code is more involved because we have to parse
3307 * the RPC records out of the stream.
3308 */
3309
3310/* NFS client UDP socket upcall */
3311void
3312nfs_udp_rcv(socket_t so, void *arg, __unused int waitflag)
3313{
3314 struct nfsmount *nmp = arg;
3315 struct nfs_socket *nso = nmp->nm_nso;
3316 size_t rcvlen;
3317 mbuf_t m;
3318 int error = 0;
3319
3320 if (nmp->nm_sockflags & NMSOCK_CONNECTING)
3321 return;
3322
3323 do {
3324 /* make sure we're on the current socket */
3325 if (!nso || (nso->nso_so != so))
3326 return;
3327
3328 m = NULL;
3329 rcvlen = 1000000;
3330 error = sock_receivembuf(so, NULL, &m, MSG_DONTWAIT, &rcvlen);
3331 if (m)
3332 nfs_request_match_reply(nmp, m);
3333 } while (m && !error);
3334
3335 if (error && (error != EWOULDBLOCK)) {
3336 /* problems with the socket... mark for reconnection */
3337 NFS_SOCK_DBG("nfs_udp_rcv: need reconnect %d\n", error);
3338 nfs_need_reconnect(nmp);
3339 }
3340}
3341
3342/* NFS client TCP socket upcall */
3343void
3344nfs_tcp_rcv(socket_t so, void *arg, __unused int waitflag)
3345{
3346 struct nfsmount *nmp = arg;
3347 struct nfs_socket *nso = nmp->nm_nso;
3348 struct nfs_rpc_record_state nrrs;
3349 mbuf_t m;
3350 int error = 0;
3351 int recv = 1;
3352 int wup = 0;
3353
3354 if (nmp->nm_sockflags & NMSOCK_CONNECTING)
3355 return;
3356
3357 /* make sure we're on the current socket */
3358 lck_mtx_lock(&nmp->nm_lock);
3359 nso = nmp->nm_nso;
3360 if (!nso || (nso->nso_so != so) || (nmp->nm_sockflags & (NMSOCK_DISCONNECTING))) {
3361 lck_mtx_unlock(&nmp->nm_lock);
3362 return;
3363 }
3364 lck_mtx_unlock(&nmp->nm_lock);
3365
3366 /* make sure this upcall should be trying to do work */
3367 lck_mtx_lock(&nso->nso_lock);
3368 if (nso->nso_flags & (NSO_UPCALL|NSO_DISCONNECTING|NSO_DEAD)) {
3369 lck_mtx_unlock(&nso->nso_lock);
3370 return;
3371 }
3372 nso->nso_flags |= NSO_UPCALL;
3373 nrrs = nso->nso_rrs;
3374 lck_mtx_unlock(&nso->nso_lock);
3375
3376 /* loop while we make error-free progress */
3377 while (!error && recv) {
3378 error = nfs_rpc_record_read(so, &nrrs, MSG_DONTWAIT, &recv, &m);
3379 if (m) /* match completed response with request */
3380 nfs_request_match_reply(nmp, m);
3381 }
3382
3383 /* Update the sockets's rpc parsing state */
3384 lck_mtx_lock(&nso->nso_lock);
3385 nso->nso_rrs = nrrs;
3386 if (nso->nso_flags & NSO_DISCONNECTING)
3387 wup = 1;
3388 nso->nso_flags &= ~NSO_UPCALL;
3389 lck_mtx_unlock(&nso->nso_lock);
3390 if (wup)
3391 wakeup(&nso->nso_flags);
3392
3393#ifdef NFS_SOCKET_DEBUGGING
3394 if (!recv && (error != EWOULDBLOCK))
3395 NFS_SOCK_DBG("nfs_tcp_rcv: got nothing, error %d, got FIN?\n", error);
3396#endif
3397 /* note: no error and no data indicates server closed its end */
3398 if ((error != EWOULDBLOCK) && (error || !recv)) {
3399 /* problems with the socket... mark for reconnection */
3400 NFS_SOCK_DBG("nfs_tcp_rcv: need reconnect %d\n", error);
3401 nfs_need_reconnect(nmp);
3402 }
3403}
3404
3405/*
3406 * "poke" a socket to try to provoke any pending errors
3407 */
3408void
3409nfs_sock_poke(struct nfsmount *nmp)
3410{
3411 struct iovec aio;
3412 struct msghdr msg;
3413 size_t len;
3414 int error = 0;
3415 int dummy;
3416
3417 lck_mtx_lock(&nmp->nm_lock);
3418 if ((nmp->nm_sockflags & NMSOCK_UNMOUNT) ||
3419 !(nmp->nm_sockflags & NMSOCK_READY) || !nmp->nm_nso || !nmp->nm_nso->nso_so) {
3420 /* Nothing to poke */
3421 nmp->nm_sockflags &= ~NMSOCK_POKE;
3422 wakeup(&nmp->nm_sockflags);
3423 lck_mtx_unlock(&nmp->nm_lock);
3424 return;
3425 }
3426 lck_mtx_unlock(&nmp->nm_lock);
3427 aio.iov_base = &dummy;
3428 aio.iov_len = 0;
3429 len = 0;
3430 bzero(&msg, sizeof(msg));
3431 msg.msg_iov = &aio;
3432 msg.msg_iovlen = 1;
3433 error = sock_send(nmp->nm_nso->nso_so, &msg, MSG_DONTWAIT, &len);
3434 NFS_SOCK_DBG("nfs_sock_poke: error %d\n", error);
3435 lck_mtx_lock(&nmp->nm_lock);
3436 nmp->nm_sockflags &= ~NMSOCK_POKE;
3437 wakeup(&nmp->nm_sockflags);
3438 lck_mtx_unlock(&nmp->nm_lock);
3439 nfs_is_dead(error, nmp);
3440}
3441
3442/*
3443 * Match an RPC reply with the corresponding request
3444 */
3445void
3446nfs_request_match_reply(struct nfsmount *nmp, mbuf_t mrep)
3447{
3448 struct nfsreq *req;
3449 struct nfsm_chain nmrep;
3450 u_int32_t reply = 0, rxid = 0;
3451 int error = 0, asyncioq, t1;
3452
3453 /* Get the xid and check that it is an rpc reply */
3454 nfsm_chain_dissect_init(error, &nmrep, mrep);
3455 nfsm_chain_get_32(error, &nmrep, rxid);
3456 nfsm_chain_get_32(error, &nmrep, reply);
3457 if (error || (reply != RPC_REPLY)) {
3458 OSAddAtomic64(1, &nfsstats.rpcinvalid);
3459 mbuf_freem(mrep);
3460 return;
3461 }
3462
3463 /*
3464 * Loop through the request list to match up the reply
3465 * Iff no match, just drop it.
3466 */
3467 lck_mtx_lock(nfs_request_mutex);
3468 TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
3469 if (req->r_nmrep.nmc_mhead || (rxid != R_XID32(req->r_xid)))
3470 continue;
3471 /* looks like we have it, grab lock and double check */
3472 lck_mtx_lock(&req->r_mtx);
3473 if (req->r_nmrep.nmc_mhead || (rxid != R_XID32(req->r_xid))) {
3474 lck_mtx_unlock(&req->r_mtx);
3475 continue;
3476 }
3477 /* Found it.. */
3478 req->r_nmrep = nmrep;
3479 lck_mtx_lock(&nmp->nm_lock);
3480 if (nmp->nm_sotype == SOCK_DGRAM) {
3481 /*
3482 * Update congestion window.
3483 * Do the additive increase of one rpc/rtt.
3484 */
3485 FSDBG(530, R_XID32(req->r_xid), req, nmp->nm_sent, nmp->nm_cwnd);
3486 if (nmp->nm_cwnd <= nmp->nm_sent) {
3487 nmp->nm_cwnd +=
3488 ((NFS_CWNDSCALE * NFS_CWNDSCALE) +
3489 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd;
3490 if (nmp->nm_cwnd > NFS_MAXCWND)
3491 nmp->nm_cwnd = NFS_MAXCWND;
3492 }
3493 if (req->r_flags & R_CWND) {
3494 nmp->nm_sent -= NFS_CWNDSCALE;
3495 req->r_flags &= ~R_CWND;
3496 }
3497 if ((nmp->nm_sent < nmp->nm_cwnd) && !TAILQ_EMPTY(&nmp->nm_cwndq)) {
3498 /* congestion window is open, poke the cwnd queue */
3499 struct nfsreq *req2 = TAILQ_FIRST(&nmp->nm_cwndq);
3500 TAILQ_REMOVE(&nmp->nm_cwndq, req2, r_cchain);
3501 req2->r_cchain.tqe_next = NFSREQNOLIST;
3502 wakeup(req2);
3503 }
3504 }
3505 /*
3506 * Update rtt using a gain of 0.125 on the mean
3507 * and a gain of 0.25 on the deviation.
3508 */
3509 if (req->r_flags & R_TIMING) {
3510 /*
3511 * Since the timer resolution of
3512 * NFS_HZ is so course, it can often
3513 * result in r_rtt == 0. Since
3514 * r_rtt == N means that the actual
3515 * rtt is between N+dt and N+2-dt ticks,
3516 * add 1.
3517 */
3518 if (proct[req->r_procnum] == 0)
3519 panic("nfs_request_match_reply: proct[%d] is zero", req->r_procnum);
3520 t1 = req->r_rtt + 1;
3521 t1 -= (NFS_SRTT(req) >> 3);
3522 NFS_SRTT(req) += t1;
3523 if (t1 < 0)
3524 t1 = -t1;
3525 t1 -= (NFS_SDRTT(req) >> 2);
3526 NFS_SDRTT(req) += t1;
3527 }
3528 nmp->nm_timeouts = 0;
3529 lck_mtx_unlock(&nmp->nm_lock);
3530 /* signal anyone waiting on this request */
3531 wakeup(req);
3532 asyncioq = (req->r_callback.rcb_func != NULL);
3533 if (nfs_request_using_gss(req))
3534 nfs_gss_clnt_rpcdone(req);
3535 lck_mtx_unlock(&req->r_mtx);
3536 lck_mtx_unlock(nfs_request_mutex);
3537 /* if it's an async RPC with a callback, queue it up */
3538 if (asyncioq)
3539 nfs_asyncio_finish(req);
3540 break;
3541 }
3542
3543 if (!req) {
3544 /* not matched to a request, so drop it. */
3545 lck_mtx_unlock(nfs_request_mutex);
3546 OSAddAtomic64(1, &nfsstats.rpcunexpected);
3547 mbuf_freem(mrep);
3548 }
3549}
3550
3551/*
3552 * Wait for the reply for a given request...
3553 * ...potentially resending the request if necessary.
3554 */
3555int
3556nfs_wait_reply(struct nfsreq *req)
3557{
3558 struct timespec ts = { 2, 0 };
3559 int error = 0, slpflag, first = 1;
3560
3561 if (req->r_nmp && NMFLAG(req->r_nmp, INTR) && req->r_thread && !(req->r_flags & R_NOINTR))
3562 slpflag = PCATCH;
3563 else
3564 slpflag = 0;
3565
3566 lck_mtx_lock(&req->r_mtx);
3567 while (!req->r_nmrep.nmc_mhead) {
3568 if ((error = nfs_sigintr(req->r_nmp, req, first ? NULL : req->r_thread, 0)))
3569 break;
3570 if (((error = req->r_error)) || req->r_nmrep.nmc_mhead)
3571 break;
3572 /* check if we need to resend */
3573 if (req->r_flags & R_MUSTRESEND) {
3574 NFS_SOCK_DBG("nfs wait resend: p %d x 0x%llx f 0x%x rtt %d\n",
3575 req->r_procnum, req->r_xid, req->r_flags, req->r_rtt);
3576 req->r_flags |= R_SENDING;
3577 lck_mtx_unlock(&req->r_mtx);
3578 if (nfs_request_using_gss(req)) {
3579 /*
3580 * It's an RPCSEC_GSS request.
3581 * Can't just resend the original request
3582 * without bumping the cred sequence number.
3583 * Go back and re-build the request.
3584 */
3585 lck_mtx_lock(&req->r_mtx);
3586 req->r_flags &= ~R_SENDING;
3587 lck_mtx_unlock(&req->r_mtx);
3588 return (EAGAIN);
3589 }
3590 error = nfs_send(req, 1);
3591 lck_mtx_lock(&req->r_mtx);
3592 NFS_SOCK_DBG("nfs wait resend: p %d x 0x%llx f 0x%x rtt %d err %d\n",
3593 req->r_procnum, req->r_xid, req->r_flags, req->r_rtt, error);
3594 if (error)
3595 break;
3596 if (((error = req->r_error)) || req->r_nmrep.nmc_mhead)
3597 break;
3598 }
3599 /* need to poll if we're P_NOREMOTEHANG */
3600 if (nfs_noremotehang(req->r_thread))
3601 ts.tv_sec = 1;
3602 msleep(req, &req->r_mtx, slpflag | (PZERO - 1), "nfswaitreply", &ts);
3603 first = slpflag = 0;
3604 }
3605 lck_mtx_unlock(&req->r_mtx);
3606
3607 return (error);
3608}
3609
3610/*
3611 * An NFS request goes something like this:
3612 * (nb: always frees up mreq mbuf list)
3613 * nfs_request_create()
3614 * - allocates a request struct if one is not provided
3615 * - initial fill-in of the request struct
3616 * nfs_request_add_header()
3617 * - add the RPC header
3618 * nfs_request_send()
3619 * - link it into list
3620 * - call nfs_send() for first transmit
3621 * nfs_request_wait()
3622 * - call nfs_wait_reply() to wait for the reply
3623 * nfs_request_finish()
3624 * - break down rpc header and return with error or nfs reply
3625 * pointed to by nmrep.
3626 * nfs_request_rele()
3627 * nfs_request_destroy()
3628 * - clean up the request struct
3629 * - free the request struct if it was allocated by nfs_request_create()
3630 */
3631
3632/*
3633 * Set up an NFS request struct (allocating if no request passed in).
3634 */
3635int
3636nfs_request_create(
3637 nfsnode_t np,
3638 mount_t mp, /* used only if !np */
3639 struct nfsm_chain *nmrest,
3640 int procnum,
3641 thread_t thd,
3642 kauth_cred_t cred,
3643 struct nfsreq **reqp)
3644{
3645 struct nfsreq *req, *newreq = NULL;
3646 struct nfsmount *nmp;
3647
3648 req = *reqp;
3649 if (!req) {
3650 /* allocate a new NFS request structure */
3651 MALLOC_ZONE(newreq, struct nfsreq*, sizeof(*newreq), M_NFSREQ, M_WAITOK);
3652 if (!newreq) {
3653 mbuf_freem(nmrest->nmc_mhead);
3654 nmrest->nmc_mhead = NULL;
3655 return (ENOMEM);
3656 }
3657 req = newreq;
3658 }
3659
3660 bzero(req, sizeof(*req));
3661 if (req == newreq)
3662 req->r_flags = R_ALLOCATED;
3663
3664 nmp = VFSTONFS(np ? NFSTOMP(np) : mp);
3665 if (nfs_mount_gone(nmp)) {
3666 if (newreq)
3667 FREE_ZONE(newreq, sizeof(*newreq), M_NFSREQ);
3668 return (ENXIO);
3669 }
3670 lck_mtx_lock(&nmp->nm_lock);
3671 if ((nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD)) &&
3672 (nmp->nm_state & NFSSTA_TIMEO)) {
3673 lck_mtx_unlock(&nmp->nm_lock);
3674 mbuf_freem(nmrest->nmc_mhead);
3675 nmrest->nmc_mhead = NULL;
3676 if (newreq)
3677 FREE_ZONE(newreq, sizeof(*newreq), M_NFSREQ);
3678 return (ENXIO);
3679 }
3680
3681 if ((nmp->nm_vers != NFS_VER4) && (procnum >= 0) && (procnum < NFS_NPROCS))
3682 OSAddAtomic64(1, &nfsstats.rpccnt[procnum]);
3683 if ((nmp->nm_vers == NFS_VER4) && (procnum != NFSPROC4_COMPOUND) && (procnum != NFSPROC4_NULL))
3684 panic("nfs_request: invalid NFSv4 RPC request %d\n", procnum);
3685
3686 lck_mtx_init(&req->r_mtx, nfs_request_grp, LCK_ATTR_NULL);
3687 req->r_nmp = nmp;
3688 nmp->nm_ref++;
3689 req->r_np = np;
3690 req->r_thread = thd;
3691 if (!thd)
3692 req->r_flags |= R_NOINTR;
3693 if (IS_VALID_CRED(cred)) {
3694 kauth_cred_ref(cred);
3695 req->r_cred = cred;
3696 }
3697 req->r_procnum = procnum;
3698 if (proct[procnum] > 0)
3699 req->r_flags |= R_TIMING;
3700 req->r_nmrep.nmc_mhead = NULL;
3701 SLIST_INIT(&req->r_gss_seqlist);
3702 req->r_achain.tqe_next = NFSREQNOLIST;
3703 req->r_rchain.tqe_next = NFSREQNOLIST;
3704 req->r_cchain.tqe_next = NFSREQNOLIST;
3705
3706 /* set auth flavor to use for request */
3707 if (!req->r_cred)
3708 req->r_auth = RPCAUTH_NONE;
3709 else if (req->r_np && (req->r_np->n_auth != RPCAUTH_INVALID))
3710 req->r_auth = req->r_np->n_auth;
3711 else
3712 req->r_auth = nmp->nm_auth;
3713
3714 lck_mtx_unlock(&nmp->nm_lock);
3715
3716 /* move the request mbuf chain to the nfsreq */
3717 req->r_mrest = nmrest->nmc_mhead;
3718 nmrest->nmc_mhead = NULL;
3719
3720 req->r_flags |= R_INITTED;
3721 req->r_refs = 1;
3722 if (newreq)
3723 *reqp = req;
3724 return (0);
3725}
3726
3727/*
3728 * Clean up and free an NFS request structure.
3729 */
3730void
3731nfs_request_destroy(struct nfsreq *req)
3732{
3733 struct nfsmount *nmp;
3734 struct gss_seq *gsp, *ngsp;
3735 int clearjbtimeo = 0;
3736
3737 if (!req || !(req->r_flags & R_INITTED))
3738 return;
3739 nmp = req->r_nmp;
3740 req->r_flags &= ~R_INITTED;
3741 if (req->r_lflags & RL_QUEUED)
3742 nfs_reqdequeue(req);
3743
3744 if (req->r_achain.tqe_next != NFSREQNOLIST) {
3745 /*
3746 * Still on an async I/O queue?
3747 * %%% But which one, we may be on a local iod.
3748 */
3749 lck_mtx_lock(nfsiod_mutex);
3750 if (nmp && req->r_achain.tqe_next != NFSREQNOLIST) {
3751 TAILQ_REMOVE(&nmp->nm_iodq, req, r_achain);
3752 req->r_achain.tqe_next = NFSREQNOLIST;
3753 }
3754 lck_mtx_unlock(nfsiod_mutex);
3755 }
3756
3757 lck_mtx_lock(&req->r_mtx);
3758 if (nmp) {
3759 lck_mtx_lock(&nmp->nm_lock);
3760 if (req->r_flags & R_CWND) {
3761 /* Decrement the outstanding request count. */
3762 req->r_flags &= ~R_CWND;
3763 nmp->nm_sent -= NFS_CWNDSCALE;
3764 if ((nmp->nm_sent < nmp->nm_cwnd) && !TAILQ_EMPTY(&nmp->nm_cwndq)) {
3765 /* congestion window is open, poke the cwnd queue */
3766 struct nfsreq *req2 = TAILQ_FIRST(&nmp->nm_cwndq);
3767 TAILQ_REMOVE(&nmp->nm_cwndq, req2, r_cchain);
3768 req2->r_cchain.tqe_next = NFSREQNOLIST;
3769 wakeup(req2);
3770 }
3771 }
3772 assert((req->r_flags & R_RESENDQ) == 0);
3773 /* XXX should we just remove this conditional, we should have a reference if we're resending */
3774 if (req->r_rchain.tqe_next != NFSREQNOLIST) {
3775 TAILQ_REMOVE(&nmp->nm_resendq, req, r_rchain);
3776 req->r_rchain.tqe_next = NFSREQNOLIST;
3777 if (req->r_flags & R_RESENDQ)
3778 req->r_flags &= ~R_RESENDQ;
3779 }
3780 if (req->r_cchain.tqe_next != NFSREQNOLIST) {
3781 TAILQ_REMOVE(&nmp->nm_cwndq, req, r_cchain);
3782 req->r_cchain.tqe_next = NFSREQNOLIST;
3783 }
3784 if (req->r_flags & R_JBTPRINTFMSG) {
3785 req->r_flags &= ~R_JBTPRINTFMSG;
3786 nmp->nm_jbreqs--;
3787 clearjbtimeo = (nmp->nm_jbreqs == 0) ? NFSSTA_JUKEBOXTIMEO : 0;
3788 }
3789 lck_mtx_unlock(&nmp->nm_lock);
3790 }
3791 lck_mtx_unlock(&req->r_mtx);
3792
3793 if (clearjbtimeo)
3794 nfs_up(nmp, req->r_thread, clearjbtimeo, NULL);
3795 if (req->r_mhead)
3796 mbuf_freem(req->r_mhead);
3797 else if (req->r_mrest)
3798 mbuf_freem(req->r_mrest);
3799 if (req->r_nmrep.nmc_mhead)
3800 mbuf_freem(req->r_nmrep.nmc_mhead);
3801 if (IS_VALID_CRED(req->r_cred))
3802 kauth_cred_unref(&req->r_cred);
3803 if (nfs_request_using_gss(req))
3804 nfs_gss_clnt_rpcdone(req);
3805 SLIST_FOREACH_SAFE(gsp, &req->r_gss_seqlist, gss_seqnext, ngsp)
3806 FREE(gsp, M_TEMP);
3807 if (req->r_gss_ctx)
3808 nfs_gss_clnt_ctx_unref(req);
3809 if (req->r_wrongsec)
3810 FREE(req->r_wrongsec, M_TEMP);
3811 if (nmp)
3812 nfs_mount_rele(nmp);
3813 lck_mtx_destroy(&req->r_mtx, nfs_request_grp);
3814 if (req->r_flags & R_ALLOCATED)
3815 FREE_ZONE(req, sizeof(*req), M_NFSREQ);
3816}
3817
3818void
3819nfs_request_ref(struct nfsreq *req, int locked)
3820{
3821 if (!locked)
3822 lck_mtx_lock(&req->r_mtx);
3823 if (req->r_refs <= 0)
3824 panic("nfsreq reference error");
3825 req->r_refs++;
3826 if (!locked)
3827 lck_mtx_unlock(&req->r_mtx);
3828}
3829
3830void
3831nfs_request_rele(struct nfsreq *req)
3832{
3833 int destroy;
3834
3835 lck_mtx_lock(&req->r_mtx);
3836 if (req->r_refs <= 0)
3837 panic("nfsreq reference underflow");
3838 req->r_refs--;
3839 destroy = (req->r_refs == 0);
3840 lck_mtx_unlock(&req->r_mtx);
3841 if (destroy)
3842 nfs_request_destroy(req);
3843}
3844
3845
3846/*
3847 * Add an (updated) RPC header with authorization to an NFS request.
3848 */
3849int
3850nfs_request_add_header(struct nfsreq *req)
3851{
3852 struct nfsmount *nmp;
3853 int error = 0;
3854 mbuf_t m;
3855
3856 /* free up any previous header */
3857 if ((m = req->r_mhead)) {
3858 while (m && (m != req->r_mrest))
3859 m = mbuf_free(m);
3860 req->r_mhead = NULL;
3861 }
3862
3863 nmp = req->r_nmp;
3864 if (nfs_mount_gone(nmp))
3865 return (ENXIO);
3866
3867 error = nfsm_rpchead(req, req->r_mrest, &req->r_xid, &req->r_mhead);
3868 if (error)
3869 return (error);
3870
3871 req->r_mreqlen = mbuf_pkthdr_len(req->r_mhead);
3872 nmp = req->r_nmp;
3873 if (nfs_mount_gone(nmp))
3874 return (ENXIO);
3875 lck_mtx_lock(&nmp->nm_lock);
3876 if (NMFLAG(nmp, SOFT) || (req->r_flags & R_SOFT))
3877 req->r_retry = nmp->nm_retry;
3878 else
3879 req->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */
3880 lck_mtx_unlock(&nmp->nm_lock);
3881
3882 return (error);
3883}
3884
3885
3886/*
3887 * Queue an NFS request up and send it out.
3888 */
3889int
3890nfs_request_send(struct nfsreq *req, int wait)
3891{
3892 struct nfsmount *nmp;
3893 struct timeval now;
3894
3895 lck_mtx_lock(&req->r_mtx);
3896 req->r_flags |= R_SENDING;
3897 lck_mtx_unlock(&req->r_mtx);
3898
3899 lck_mtx_lock(nfs_request_mutex);
3900
3901 nmp = req->r_nmp;
3902 if (nfs_mount_gone(nmp)) {
3903 lck_mtx_unlock(nfs_request_mutex);
3904 return (ENXIO);
3905 }
3906
3907 microuptime(&now);
3908 if (!req->r_start) {
3909 req->r_start = now.tv_sec;
3910 req->r_lastmsg = now.tv_sec -
3911 ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay));
3912 }
3913
3914 OSAddAtomic64(1, &nfsstats.rpcrequests);
3915
3916 /*
3917 * Chain request into list of outstanding requests. Be sure
3918 * to put it LAST so timer finds oldest requests first.
3919 * Make sure that the request queue timer is running
3920 * to check for possible request timeout.
3921 */
3922 TAILQ_INSERT_TAIL(&nfs_reqq, req, r_chain);
3923 req->r_lflags |= RL_QUEUED;
3924 if (!nfs_request_timer_on) {
3925 nfs_request_timer_on = 1;
3926 nfs_interval_timer_start(nfs_request_timer_call,
3927 NFS_REQUESTDELAY);
3928 }
3929 lck_mtx_unlock(nfs_request_mutex);
3930
3931 /* Send the request... */
3932 return (nfs_send(req, wait));
3933}
3934
3935/*
3936 * Call nfs_wait_reply() to wait for the reply.
3937 */
3938void
3939nfs_request_wait(struct nfsreq *req)
3940{
3941 req->r_error = nfs_wait_reply(req);
3942}
3943
3944/*
3945 * Finish up an NFS request by dequeueing it and
3946 * doing the initial NFS request reply processing.
3947 */
3948int
3949nfs_request_finish(
3950 struct nfsreq *req,
3951 struct nfsm_chain *nmrepp,
3952 int *status)
3953{
3954 struct nfsmount *nmp;
3955 mbuf_t mrep;
3956 int verf_type = 0;
3957 uint32_t verf_len = 0;
3958 uint32_t reply_status = 0;
3959 uint32_t rejected_status = 0;
3960 uint32_t auth_status = 0;
3961 uint32_t accepted_status = 0;
3962 struct nfsm_chain nmrep;
3963 int error, clearjbtimeo;
3964
3965 error = req->r_error;
3966
3967 if (nmrepp)
3968 nmrepp->nmc_mhead = NULL;
3969
3970 /* RPC done, unlink the request. */
3971 nfs_reqdequeue(req);
3972
3973 mrep = req->r_nmrep.nmc_mhead;
3974
3975 nmp = req->r_nmp;
3976
3977 if ((req->r_flags & R_CWND) && nmp) {
3978 /*
3979 * Decrement the outstanding request count.
3980 */
3981 req->r_flags &= ~R_CWND;
3982 lck_mtx_lock(&nmp->nm_lock);
3983 FSDBG(273, R_XID32(req->r_xid), req, nmp->nm_sent, nmp->nm_cwnd);
3984 nmp->nm_sent -= NFS_CWNDSCALE;
3985 if ((nmp->nm_sent < nmp->nm_cwnd) && !TAILQ_EMPTY(&nmp->nm_cwndq)) {
3986 /* congestion window is open, poke the cwnd queue */
3987 struct nfsreq *req2 = TAILQ_FIRST(&nmp->nm_cwndq);
3988 TAILQ_REMOVE(&nmp->nm_cwndq, req2, r_cchain);
3989 req2->r_cchain.tqe_next = NFSREQNOLIST;
3990 wakeup(req2);
3991 }
3992 lck_mtx_unlock(&nmp->nm_lock);
3993 }
3994
3995 if (nfs_request_using_gss(req)) {
3996 /*
3997 * If the request used an RPCSEC_GSS credential
3998 * then reset its sequence number bit in the
3999 * request window.
4000 */
4001 nfs_gss_clnt_rpcdone(req);
4002
4003 /*
4004 * If we need to re-send, go back and re-build the
4005 * request based on a new sequence number.
4006 * Note that we're using the original XID.
4007 */
4008 if (error == EAGAIN) {
4009 req->r_error = 0;
4010 if (mrep)
4011 mbuf_freem(mrep);
4012 error = nfs_gss_clnt_args_restore(req); // remove any trailer mbufs
4013 req->r_nmrep.nmc_mhead = NULL;
4014 req->r_flags |= R_RESTART;
4015 if (error == ENEEDAUTH) {
4016 req->r_xid = 0; // get a new XID
4017 error = 0;
4018 }
4019 goto nfsmout;
4020 }
4021 }
4022
4023 /*
4024 * If there was a successful reply, make sure to mark the mount as up.
4025 * If a tprintf message was given (or if this is a timed-out soft mount)
4026 * then post a tprintf message indicating the server is alive again.
4027 */
4028 if (!error) {
4029 if ((req->r_flags & R_TPRINTFMSG) ||
4030 (nmp && (NMFLAG(nmp, SOFT) || (req->r_flags & R_SOFT)) &&
4031 ((nmp->nm_state & (NFSSTA_TIMEO|NFSSTA_FORCE|NFSSTA_DEAD)) == NFSSTA_TIMEO)))
4032 nfs_up(nmp, req->r_thread, NFSSTA_TIMEO, "is alive again");
4033 else
4034 nfs_up(nmp, req->r_thread, NFSSTA_TIMEO, NULL);
4035 }
4036 if (!error && !nmp)
4037 error = ENXIO;
4038 nfsmout_if(error);
4039
4040 /*
4041 * break down the RPC header and check if ok
4042 */
4043 nmrep = req->r_nmrep;
4044 nfsm_chain_get_32(error, &nmrep, reply_status);
4045 nfsmout_if(error);
4046 if (reply_status == RPC_MSGDENIED) {
4047 nfsm_chain_get_32(error, &nmrep, rejected_status);
4048 nfsmout_if(error);
4049 if (rejected_status == RPC_MISMATCH) {
4050 error = ENOTSUP;
4051 goto nfsmout;
4052 }
4053 nfsm_chain_get_32(error, &nmrep, auth_status);
4054 nfsmout_if(error);
4055 switch (auth_status) {
4056 case RPCSEC_GSS_CREDPROBLEM:
4057 case RPCSEC_GSS_CTXPROBLEM:
4058 /*
4059 * An RPCSEC_GSS cred or context problem.
4060 * We can't use it anymore.
4061 * Restore the args, renew the context
4062 * and set up for a resend.
4063 */
4064 error = nfs_gss_clnt_args_restore(req);
4065 if (error && error != ENEEDAUTH)
4066 break;
4067
4068 if (!error) {
4069 error = nfs_gss_clnt_ctx_renew(req);
4070 if (error)
4071 break;
4072 }
4073 mbuf_freem(mrep);
4074 req->r_nmrep.nmc_mhead = NULL;
4075 req->r_xid = 0; // get a new XID
4076 req->r_flags |= R_RESTART;
4077 goto nfsmout;
4078 default:
4079 error = EACCES;
4080 break;
4081 }
4082 goto nfsmout;
4083 }
4084
4085 /* Now check the verifier */
4086 nfsm_chain_get_32(error, &nmrep, verf_type); // verifier flavor
4087 nfsm_chain_get_32(error, &nmrep, verf_len); // verifier length
4088 nfsmout_if(error);
4089
4090 switch (req->r_auth) {
4091 case RPCAUTH_NONE:
4092 case RPCAUTH_SYS:
4093 /* Any AUTH_SYS verifier is ignored */
4094 if (verf_len > 0)
4095 nfsm_chain_adv(error, &nmrep, nfsm_rndup(verf_len));
4096 nfsm_chain_get_32(error, &nmrep, accepted_status);
4097 break;
4098 case RPCAUTH_KRB5:
4099 case RPCAUTH_KRB5I:
4100 case RPCAUTH_KRB5P:
4101 error = nfs_gss_clnt_verf_get(req, &nmrep,
4102 verf_type, verf_len, &accepted_status);
4103 break;
4104 }
4105 nfsmout_if(error);
4106
4107 switch (accepted_status) {
4108 case RPC_SUCCESS:
4109 if (req->r_procnum == NFSPROC_NULL) {
4110 /*
4111 * The NFS null procedure is unique,
4112 * in not returning an NFS status.
4113 */
4114 *status = NFS_OK;
4115 } else {
4116 nfsm_chain_get_32(error, &nmrep, *status);
4117 nfsmout_if(error);
4118 }
4119
4120 if ((nmp->nm_vers != NFS_VER2) && (*status == NFSERR_TRYLATER)) {
4121 /*
4122 * It's a JUKEBOX error - delay and try again
4123 */
4124 int delay, slpflag = (NMFLAG(nmp, INTR) && !(req->r_flags & R_NOINTR)) ? PCATCH : 0;
4125
4126 mbuf_freem(mrep);
4127 req->r_nmrep.nmc_mhead = NULL;
4128 if ((req->r_delay >= 30) && !(nmp->nm_state & NFSSTA_MOUNTED)) {
4129 /* we're not yet completely mounted and */
4130 /* we can't complete an RPC, so we fail */
4131 OSAddAtomic64(1, &nfsstats.rpctimeouts);
4132 nfs_softterm(req);
4133 error = req->r_error;
4134 goto nfsmout;
4135 }
4136 req->r_delay = !req->r_delay ? NFS_TRYLATERDEL : (req->r_delay * 2);
4137 if (req->r_delay > 30)
4138 req->r_delay = 30;
4139 if (nmp->nm_tprintf_initial_delay && (req->r_delay >= nmp->nm_tprintf_initial_delay)) {
4140 if (!(req->r_flags & R_JBTPRINTFMSG)) {
4141 req->r_flags |= R_JBTPRINTFMSG;
4142 lck_mtx_lock(&nmp->nm_lock);
4143 nmp->nm_jbreqs++;
4144 lck_mtx_unlock(&nmp->nm_lock);
4145 }
4146 nfs_down(req->r_nmp, req->r_thread, 0, NFSSTA_JUKEBOXTIMEO,
4147 "resource temporarily unavailable (jukebox)", 0);
4148 }
4149 if ((NMFLAG(nmp, SOFT) || (req->r_flags & R_SOFT)) && (req->r_delay == 30) &&
4150 !(req->r_flags & R_NOINTR)) {
4151 /* for soft mounts, just give up after a short while */
4152 OSAddAtomic64(1, &nfsstats.rpctimeouts);
4153 nfs_softterm(req);
4154 error = req->r_error;
4155 goto nfsmout;
4156 }
4157 delay = req->r_delay;
4158 if (req->r_callback.rcb_func) {
4159 struct timeval now;
4160 microuptime(&now);
4161 req->r_resendtime = now.tv_sec + delay;
4162 } else {
4163 do {
4164 if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 0)))
4165 goto nfsmout;
4166 tsleep(nfs_request_finish, PSOCK|slpflag, "nfs_jukebox_trylater", hz);
4167 slpflag = 0;
4168 } while (--delay > 0);
4169 }
4170 req->r_xid = 0; // get a new XID
4171 req->r_flags |= R_RESTART;
4172 req->r_start = 0;
4173 FSDBG(273, R_XID32(req->r_xid), nmp, req, NFSERR_TRYLATER);
4174 return (0);
4175 }
4176
4177 if (req->r_flags & R_JBTPRINTFMSG) {
4178 req->r_flags &= ~R_JBTPRINTFMSG;
4179 lck_mtx_lock(&nmp->nm_lock);
4180 nmp->nm_jbreqs--;
4181 clearjbtimeo = (nmp->nm_jbreqs == 0) ? NFSSTA_JUKEBOXTIMEO : 0;
4182 lck_mtx_unlock(&nmp->nm_lock);
4183 nfs_up(nmp, req->r_thread, clearjbtimeo, "resource available again");
4184 }
4185
4186 if ((nmp->nm_vers >= NFS_VER4) && (*status == NFSERR_WRONGSEC)) {
4187 /*
4188 * Hmmm... we need to try a different security flavor.
4189 * The first time a request hits this, we will allocate an array
4190 * to track flavors to try. We fill the array with the mount's
4191 * preferred flavors or the server's preferred flavors or just the
4192 * flavors we support.
4193 */
4194 uint32_t srvflavors[NX_MAX_SEC_FLAVORS];
4195 int srvcount, i, j;
4196
4197 /* Call SECINFO to try to get list of flavors from server. */
4198 srvcount = NX_MAX_SEC_FLAVORS;
4199 nfs4_secinfo_rpc(nmp, &req->r_secinfo, req->r_cred, srvflavors, &srvcount);
4200
4201 if (!req->r_wrongsec) {
4202 /* first time... set up flavor array */
4203 MALLOC(req->r_wrongsec, uint32_t*, NX_MAX_SEC_FLAVORS*sizeof(uint32_t), M_TEMP, M_WAITOK);
4204 if (!req->r_wrongsec) {
4205 error = EACCES;
4206 goto nfsmout;
4207 }
4208 i=0;
4209 if (nmp->nm_sec.count) { /* use the mount's preferred list of flavors */
4210 for(; i < nmp->nm_sec.count; i++)
4211 req->r_wrongsec[i] = nmp->nm_sec.flavors[i];
4212 } else if (srvcount) { /* otherwise use the server's list of flavors */
4213 for(; i < srvcount; i++)
4214 req->r_wrongsec[i] = srvflavors[i];
4215 } else { /* otherwise, just try the flavors we support. */
4216 req->r_wrongsec[i++] = RPCAUTH_KRB5P;
4217 req->r_wrongsec[i++] = RPCAUTH_KRB5I;
4218 req->r_wrongsec[i++] = RPCAUTH_KRB5;
4219 req->r_wrongsec[i++] = RPCAUTH_SYS;
4220 req->r_wrongsec[i++] = RPCAUTH_NONE;
4221 }
4222 for(; i < NX_MAX_SEC_FLAVORS; i++) /* invalidate any remaining slots */
4223 req->r_wrongsec[i] = RPCAUTH_INVALID;
4224 }
4225
4226 /* clear the current flavor from the list */
4227 for(i=0; i < NX_MAX_SEC_FLAVORS; i++)
4228 if (req->r_wrongsec[i] == req->r_auth)
4229 req->r_wrongsec[i] = RPCAUTH_INVALID;
4230
4231 /* find the next flavor to try */
4232 for(i=0; i < NX_MAX_SEC_FLAVORS; i++)
4233 if (req->r_wrongsec[i] != RPCAUTH_INVALID) {
4234 if (!srvcount) /* no server list, just try it */
4235 break;
4236 /* check that it's in the server's list */
4237 for(j=0; j < srvcount; j++)
4238 if (req->r_wrongsec[i] == srvflavors[j])
4239 break;
4240 if (j < srvcount) /* found */
4241 break;
4242 /* not found in server list */
4243 req->r_wrongsec[i] = RPCAUTH_INVALID;
4244 }
4245 if (i == NX_MAX_SEC_FLAVORS) {
4246 /* nothing left to try! */
4247 error = EACCES;
4248 goto nfsmout;
4249 }
4250
4251 /* retry with the next auth flavor */
4252 req->r_auth = req->r_wrongsec[i];
4253 req->r_xid = 0; // get a new XID
4254 req->r_flags |= R_RESTART;
4255 req->r_start = 0;
4256 FSDBG(273, R_XID32(req->r_xid), nmp, req, NFSERR_WRONGSEC);
4257 return (0);
4258 }
4259 if ((nmp->nm_vers >= NFS_VER4) && req->r_wrongsec) {
4260 /*
4261 * We renegotiated security for this request; so update the
4262 * default security flavor for the associated node.
4263 */
4264 if (req->r_np)
4265 req->r_np->n_auth = req->r_auth;
4266 }
4267
4268 if (*status == NFS_OK) {
4269 /*
4270 * Successful NFS request
4271 */
4272 *nmrepp = nmrep;
4273 req->r_nmrep.nmc_mhead = NULL;
4274 break;
4275 }
4276 /* Got an NFS error of some kind */
4277
4278 /*
4279 * If the File Handle was stale, invalidate the
4280 * lookup cache, just in case.
4281 */
4282 if ((*status == ESTALE) && req->r_np) {
4283 cache_purge(NFSTOV(req->r_np));
4284 /* if monitored, also send delete event */
4285 if (vnode_ismonitored(NFSTOV(req->r_np)))
4286 nfs_vnode_notify(req->r_np, (VNODE_EVENT_ATTRIB|VNODE_EVENT_DELETE));
4287 }
4288 if (nmp->nm_vers == NFS_VER2)
4289 mbuf_freem(mrep);
4290 else
4291 *nmrepp = nmrep;
4292 req->r_nmrep.nmc_mhead = NULL;
4293 error = 0;
4294 break;
4295 case RPC_PROGUNAVAIL:
4296 error = EPROGUNAVAIL;
4297 break;
4298 case RPC_PROGMISMATCH:
4299 error = ERPCMISMATCH;
4300 break;
4301 case RPC_PROCUNAVAIL:
4302 error = EPROCUNAVAIL;
4303 break;
4304 case RPC_GARBAGE:
4305 error = EBADRPC;
4306 break;
4307 case RPC_SYSTEM_ERR:
4308 default:
4309 error = EIO;
4310 break;
4311 }
4312nfsmout:
4313 if (req->r_flags & R_JBTPRINTFMSG) {
4314 req->r_flags &= ~R_JBTPRINTFMSG;
4315 lck_mtx_lock(&nmp->nm_lock);
4316 nmp->nm_jbreqs--;
4317 clearjbtimeo = (nmp->nm_jbreqs == 0) ? NFSSTA_JUKEBOXTIMEO : 0;
4318 lck_mtx_unlock(&nmp->nm_lock);
4319 if (clearjbtimeo)
4320 nfs_up(nmp, req->r_thread, clearjbtimeo, NULL);
4321 }
4322 FSDBG(273, R_XID32(req->r_xid), nmp, req,
4323 (!error && (*status == NFS_OK)) ? 0xf0f0f0f0 : error);
4324 return (error);
4325}
4326
4327/*
4328 * NFS request using a GSS/Kerberos security flavor?
4329 */
4330int
4331nfs_request_using_gss(struct nfsreq *req)
4332{
4333 if (!req->r_gss_ctx)
4334 return (0);
4335 switch (req->r_auth) {
4336 case RPCAUTH_KRB5:
4337 case RPCAUTH_KRB5I:
4338 case RPCAUTH_KRB5P:
4339 return (1);
4340 }
4341 return (0);
4342}
4343
4344/*
4345 * Perform an NFS request synchronously.
4346 */
4347
4348int
4349nfs_request(
4350 nfsnode_t np,
4351 mount_t mp, /* used only if !np */
4352 struct nfsm_chain *nmrest,
4353 int procnum,
4354 vfs_context_t ctx,
4355 struct nfsreq_secinfo_args *si,
4356 struct nfsm_chain *nmrepp,
4357 u_int64_t *xidp,
4358 int *status)
4359{
4360 return nfs_request2(np, mp, nmrest, procnum,
4361 vfs_context_thread(ctx), vfs_context_ucred(ctx),
4362 si, 0, nmrepp, xidp, status);
4363}
4364
4365int
4366nfs_request2(
4367 nfsnode_t np,
4368 mount_t mp, /* used only if !np */
4369 struct nfsm_chain *nmrest,
4370 int procnum,
4371 thread_t thd,
4372 kauth_cred_t cred,
4373 struct nfsreq_secinfo_args *si,
4374 int flags,
4375 struct nfsm_chain *nmrepp,
4376 u_int64_t *xidp,
4377 int *status)
4378{
4379 struct nfsreq rq, *req = &rq;
4380 int error;
4381
4382 if ((error = nfs_request_create(np, mp, nmrest, procnum, thd, cred, &req)))
4383 return (error);
4384 req->r_flags |= (flags & (R_OPTMASK | R_SOFT));
4385 if (si)
4386 req->r_secinfo = *si;
4387
4388 FSDBG_TOP(273, R_XID32(req->r_xid), np, procnum, 0);
4389 do {
4390 req->r_error = 0;
4391 req->r_flags &= ~R_RESTART;
4392 if ((error = nfs_request_add_header(req)))
4393 break;
4394 if (xidp)
4395 *xidp = req->r_xid;
4396 if ((error = nfs_request_send(req, 1)))
4397 break;
4398 nfs_request_wait(req);
4399 if ((error = nfs_request_finish(req, nmrepp, status)))
4400 break;
4401 } while (req->r_flags & R_RESTART);
4402
4403 FSDBG_BOT(273, R_XID32(req->r_xid), np, procnum, error);
4404 nfs_request_rele(req);
4405 return (error);
4406}
4407
4408
4409/*
4410 * Set up a new null proc request to exchange GSS context tokens with the
4411 * server. Associate the context that we are setting up with the request that we
4412 * are sending.
4413 */
4414
4415int
4416nfs_request_gss(
4417 mount_t mp,
4418 struct nfsm_chain *nmrest,
4419 thread_t thd,
4420 kauth_cred_t cred,
4421 int flags,
4422 struct nfs_gss_clnt_ctx *cp, /* Set to gss context to renew or setup */
4423 struct nfsm_chain *nmrepp,
4424 int *status)
4425{
4426 struct nfsreq rq, *req = &rq;
4427 int error, wait = 1;
4428
4429 if ((error = nfs_request_create(NULL, mp, nmrest, NFSPROC_NULL, thd, cred, &req)))
4430 return (error);
4431 req->r_flags |= (flags & R_OPTMASK);
4432
4433 if (cp == NULL) {
4434 printf("nfs_request_gss request has no context\n");
4435 nfs_request_rele(req);
4436 return (NFSERR_EAUTH);
4437 }
4438 nfs_gss_clnt_ctx_ref(req, cp);
4439
4440 /*
4441 * Don't wait for a reply to a context destroy advisory
4442 * to avoid hanging on a dead server.
4443 */
4444 if (cp->gss_clnt_proc == RPCSEC_GSS_DESTROY)
4445 wait = 0;
4446
4447 FSDBG_TOP(273, R_XID32(req->r_xid), NULL, NFSPROC_NULL, 0);
4448 do {
4449 req->r_error = 0;
4450 req->r_flags &= ~R_RESTART;
4451 if ((error = nfs_request_add_header(req)))
4452 break;
4453
4454 if ((error = nfs_request_send(req, wait)))
4455 break;
4456 if (!wait)
4457 break;
4458
4459 nfs_request_wait(req);
4460 if ((error = nfs_request_finish(req, nmrepp, status)))
4461 break;
4462 } while (req->r_flags & R_RESTART);
4463
4464 FSDBG_BOT(273, R_XID32(req->r_xid), NULL, NFSPROC_NULL, error);
4465
4466 nfs_gss_clnt_ctx_unref(req);
4467 nfs_request_rele(req);
4468
4469 return (error);
4470}
4471
4472/*
4473 * Create and start an asynchronous NFS request.
4474 */
4475int
4476nfs_request_async(
4477 nfsnode_t np,
4478 mount_t mp, /* used only if !np */
4479 struct nfsm_chain *nmrest,
4480 int procnum,
4481 thread_t thd,
4482 kauth_cred_t cred,
4483 struct nfsreq_secinfo_args *si,
4484 int flags,
4485 struct nfsreq_cbinfo *cb,
4486 struct nfsreq **reqp)
4487{
4488 struct nfsreq *req;
4489 struct nfsmount *nmp;
4490 int error, sent;
4491
4492 error = nfs_request_create(np, mp, nmrest, procnum, thd, cred, reqp);
4493 req = *reqp;
4494 FSDBG(274, (req ? R_XID32(req->r_xid) : 0), np, procnum, error);
4495 if (error)
4496 return (error);
4497 req->r_flags |= (flags & R_OPTMASK);
4498 req->r_flags |= R_ASYNC;
4499 if (si)
4500 req->r_secinfo = *si;
4501 if (cb)
4502 req->r_callback = *cb;
4503 error = nfs_request_add_header(req);
4504 if (!error) {
4505 req->r_flags |= R_WAITSENT;
4506 if (req->r_callback.rcb_func)
4507 nfs_request_ref(req, 0);
4508 error = nfs_request_send(req, 1);
4509 lck_mtx_lock(&req->r_mtx);
4510 if (!error && !(req->r_flags & R_SENT) && req->r_callback.rcb_func) {
4511 /* make sure to wait until this async I/O request gets sent */
4512 int slpflag = (req->r_nmp && NMFLAG(req->r_nmp, INTR) && req->r_thread && !(req->r_flags & R_NOINTR)) ? PCATCH : 0;
4513 struct timespec ts = { 2, 0 };
4514 while (!(req->r_flags & R_SENT)) {
4515 nmp = req->r_nmp;
4516 if ((req->r_flags & R_RESENDQ) && !nfs_mount_gone(nmp)) {
4517 lck_mtx_lock(&nmp->nm_lock);
4518 if ((nmp->nm_state & NFSSTA_RECOVER) && (req->r_rchain.tqe_next != NFSREQNOLIST)) {
4519 /*
4520 * It's not going to get off the resend queue if we're in recovery.
4521 * So, just take it off ourselves. We could be holding mount state
4522 * busy and thus holding up the start of recovery.
4523 */
4524 TAILQ_REMOVE(&nmp->nm_resendq, req, r_rchain);
4525 req->r_rchain.tqe_next = NFSREQNOLIST;
4526 if (req->r_flags & R_RESENDQ)
4527 req->r_flags &= ~R_RESENDQ;
4528 lck_mtx_unlock(&nmp->nm_lock);
4529 req->r_flags |= R_SENDING;
4530 lck_mtx_unlock(&req->r_mtx);
4531 error = nfs_send(req, 1);
4532 /* Remove the R_RESENDQ reference */
4533 nfs_request_rele(req);
4534 lck_mtx_lock(&req->r_mtx);
4535 if (error)
4536 break;
4537 continue;
4538 }
4539 lck_mtx_unlock(&nmp->nm_lock);
4540 }
4541 if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 0)))
4542 break;
4543 msleep(req, &req->r_mtx, slpflag | (PZERO - 1), "nfswaitsent", &ts);
4544 slpflag = 0;
4545 }
4546 }
4547 sent = req->r_flags & R_SENT;
4548 lck_mtx_unlock(&req->r_mtx);
4549 if (error && req->r_callback.rcb_func && !sent) {
4550 nfs_request_rele(req);
4551 }
4552 }
4553 FSDBG(274, R_XID32(req->r_xid), np, procnum, error);
4554 if (error || req->r_callback.rcb_func)
4555 nfs_request_rele(req);
4556
4557 return (error);
4558}
4559
4560/*
4561 * Wait for and finish an asynchronous NFS request.
4562 */
4563int
4564nfs_request_async_finish(
4565 struct nfsreq *req,
4566 struct nfsm_chain *nmrepp,
4567 u_int64_t *xidp,
4568 int *status)
4569{
4570 int error = 0, asyncio = req->r_callback.rcb_func ? 1 : 0;
4571 struct nfsmount *nmp;
4572
4573 lck_mtx_lock(&req->r_mtx);
4574 if (!asyncio)
4575 req->r_flags |= R_ASYNCWAIT;
4576 while (req->r_flags & R_RESENDQ) { /* wait until the request is off the resend queue */
4577 struct timespec ts = { 2, 0 };
4578
4579 if ((nmp = req->r_nmp)) {
4580 lck_mtx_lock(&nmp->nm_lock);
4581 if ((nmp->nm_state & NFSSTA_RECOVER) && (req->r_rchain.tqe_next != NFSREQNOLIST)) {
4582 /*
4583 * It's not going to get off the resend queue if we're in recovery.
4584 * So, just take it off ourselves. We could be holding mount state
4585 * busy and thus holding up the start of recovery.
4586 */
4587 TAILQ_REMOVE(&nmp->nm_resendq, req, r_rchain);
4588 req->r_rchain.tqe_next = NFSREQNOLIST;
4589 if (req->r_flags & R_RESENDQ)
4590 req->r_flags &= ~R_RESENDQ;
4591 /* Remove the R_RESENDQ reference */
4592 assert(req->r_refs > 0);
4593 req->r_refs--;
4594 lck_mtx_unlock(&nmp->nm_lock);
4595 break;
4596 }
4597 lck_mtx_unlock(&nmp->nm_lock);
4598 }
4599 if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 0)))
4600 break;
4601 msleep(req, &req->r_mtx, PZERO-1, "nfsresendqwait", &ts);
4602 }
4603 lck_mtx_unlock(&req->r_mtx);
4604
4605 if (!error) {
4606 nfs_request_wait(req);
4607 error = nfs_request_finish(req, nmrepp, status);
4608 }
4609
4610 while (!error && (req->r_flags & R_RESTART)) {
4611 if (asyncio) {
4612 assert(req->r_achain.tqe_next == NFSREQNOLIST);
4613 lck_mtx_lock(&req->r_mtx);
4614 req->r_flags &= ~R_IOD;
4615 if (req->r_resendtime) { /* send later */
4616 nfs_asyncio_resend(req);
4617 lck_mtx_unlock(&req->r_mtx);
4618 return (EINPROGRESS);
4619 }
4620 lck_mtx_unlock(&req->r_mtx);
4621 }
4622 req->r_error = 0;
4623 req->r_flags &= ~R_RESTART;
4624 if ((error = nfs_request_add_header(req)))
4625 break;
4626 if ((error = nfs_request_send(req, !asyncio)))
4627 break;
4628 if (asyncio)
4629 return (EINPROGRESS);
4630 nfs_request_wait(req);
4631 if ((error = nfs_request_finish(req, nmrepp, status)))
4632 break;
4633 }
4634 if (xidp)
4635 *xidp = req->r_xid;
4636
4637 FSDBG(275, R_XID32(req->r_xid), req->r_np, req->r_procnum, error);
4638 nfs_request_rele(req);
4639 return (error);
4640}
4641
4642/*
4643 * Cancel a pending asynchronous NFS request.
4644 */
4645void
4646nfs_request_async_cancel(struct nfsreq *req)
4647{
4648 FSDBG(275, R_XID32(req->r_xid), req->r_np, req->r_procnum, 0xD1ED1E);
4649 nfs_request_rele(req);
4650}
4651
4652/*
4653 * Flag a request as being terminated.
4654 */
4655void
4656nfs_softterm(struct nfsreq *req)
4657{
4658 struct nfsmount *nmp = req->r_nmp;
4659 req->r_flags |= R_SOFTTERM;
4660 req->r_error = ETIMEDOUT;
4661 if (!(req->r_flags & R_CWND) || nfs_mount_gone(nmp))
4662 return;
4663 /* update congestion window */
4664 req->r_flags &= ~R_CWND;
4665 lck_mtx_lock(&nmp->nm_lock);
4666 FSDBG(532, R_XID32(req->r_xid), req, nmp->nm_sent, nmp->nm_cwnd);
4667 nmp->nm_sent -= NFS_CWNDSCALE;
4668 if ((nmp->nm_sent < nmp->nm_cwnd) && !TAILQ_EMPTY(&nmp->nm_cwndq)) {
4669 /* congestion window is open, poke the cwnd queue */
4670 struct nfsreq *req2 = TAILQ_FIRST(&nmp->nm_cwndq);
4671 TAILQ_REMOVE(&nmp->nm_cwndq, req2, r_cchain);
4672 req2->r_cchain.tqe_next = NFSREQNOLIST;
4673 wakeup(req2);
4674 }
4675 lck_mtx_unlock(&nmp->nm_lock);
4676}
4677
4678/*
4679 * Ensure req isn't in use by the timer, then dequeue it.
4680 */
4681void
4682nfs_reqdequeue(struct nfsreq *req)
4683{
4684 lck_mtx_lock(nfs_request_mutex);
4685 while (req->r_lflags & RL_BUSY) {
4686 req->r_lflags |= RL_WAITING;
4687 msleep(&req->r_lflags, nfs_request_mutex, PSOCK, "reqdeq", NULL);
4688 }
4689 if (req->r_lflags & RL_QUEUED) {
4690 TAILQ_REMOVE(&nfs_reqq, req, r_chain);
4691 req->r_lflags &= ~RL_QUEUED;
4692 }
4693 lck_mtx_unlock(nfs_request_mutex);
4694}
4695
4696/*
4697 * Busy (lock) a nfsreq, used by the nfs timer to make sure it's not
4698 * free()'d out from under it.
4699 */
4700void
4701nfs_reqbusy(struct nfsreq *req)
4702{
4703 if (req->r_lflags & RL_BUSY)
4704 panic("req locked");
4705 req->r_lflags |= RL_BUSY;
4706}
4707
4708/*
4709 * Unbusy the nfsreq passed in, return the next nfsreq in the chain busied.
4710 */
4711struct nfsreq *
4712nfs_reqnext(struct nfsreq *req)
4713{
4714 struct nfsreq * nextreq;
4715
4716 if (req == NULL)
4717 return (NULL);
4718 /*
4719 * We need to get and busy the next req before signalling the
4720 * current one, otherwise wakeup() may block us and we'll race to
4721 * grab the next req.
4722 */
4723 nextreq = TAILQ_NEXT(req, r_chain);
4724 if (nextreq != NULL)
4725 nfs_reqbusy(nextreq);
4726 /* unbusy and signal. */
4727 req->r_lflags &= ~RL_BUSY;
4728 if (req->r_lflags & RL_WAITING) {
4729 req->r_lflags &= ~RL_WAITING;
4730 wakeup(&req->r_lflags);
4731 }
4732 return (nextreq);
4733}
4734
4735/*
4736 * NFS request queue timer routine
4737 *
4738 * Scan the NFS request queue for any requests that have timed out.
4739 *
4740 * Alert the system of unresponsive servers.
4741 * Mark expired requests on soft mounts as terminated.
4742 * For UDP, mark/signal requests for retransmission.
4743 */
4744void
4745nfs_request_timer(__unused void *param0, __unused void *param1)
4746{
4747 struct nfsreq *req;
4748 struct nfsmount *nmp;
4749 int timeo, maxtime, finish_asyncio, error;
4750 struct timeval now;
4751 TAILQ_HEAD(nfs_mount_pokeq, nfsmount) nfs_mount_poke_queue;
4752 TAILQ_INIT(&nfs_mount_poke_queue);
4753
4754restart:
4755 lck_mtx_lock(nfs_request_mutex);
4756 req = TAILQ_FIRST(&nfs_reqq);
4757 if (req == NULL) { /* no requests - turn timer off */
4758 nfs_request_timer_on = 0;
4759 lck_mtx_unlock(nfs_request_mutex);
4760 return;
4761 }
4762
4763 nfs_reqbusy(req);
4764
4765 microuptime(&now);
4766 for ( ; req != NULL ; req = nfs_reqnext(req)) {
4767 nmp = req->r_nmp;
4768 if (nmp == NULL) {
4769 NFS_SOCK_DBG("Found a request with out a mount!\n");
4770 continue;
4771 }
4772 if (req->r_error || req->r_nmrep.nmc_mhead)
4773 continue;
4774 if ((error = nfs_sigintr(nmp, req, req->r_thread, 0))) {
4775 if (req->r_callback.rcb_func != NULL) {
4776 /* async I/O RPC needs to be finished */
4777 lck_mtx_lock(&req->r_mtx);
4778 req->r_error = error;
4779 finish_asyncio = !(req->r_flags & R_WAITSENT);
4780 wakeup(req);
4781 lck_mtx_unlock(&req->r_mtx);
4782 if (finish_asyncio)
4783 nfs_asyncio_finish(req);
4784 }
4785 continue;
4786 }
4787
4788 lck_mtx_lock(&req->r_mtx);
4789
4790 if (nmp->nm_tprintf_initial_delay &&
4791 ((req->r_rexmit > 2) || (req->r_flags & R_RESENDERR)) &&
4792 ((req->r_lastmsg + nmp->nm_tprintf_delay) < now.tv_sec)) {
4793 req->r_lastmsg = now.tv_sec;
4794 nfs_down(req->r_nmp, req->r_thread, 0, NFSSTA_TIMEO,
4795 "not responding", 1);
4796 req->r_flags |= R_TPRINTFMSG;
4797 lck_mtx_lock(&nmp->nm_lock);
4798 if (!(nmp->nm_state & NFSSTA_MOUNTED)) {
4799 lck_mtx_unlock(&nmp->nm_lock);
4800 /* we're not yet completely mounted and */
4801 /* we can't complete an RPC, so we fail */
4802 OSAddAtomic64(1, &nfsstats.rpctimeouts);
4803 nfs_softterm(req);
4804 finish_asyncio = ((req->r_callback.rcb_func != NULL) && !(req->r_flags & R_WAITSENT));
4805 wakeup(req);
4806 lck_mtx_unlock(&req->r_mtx);
4807 if (finish_asyncio)
4808 nfs_asyncio_finish(req);
4809 continue;
4810 }
4811 lck_mtx_unlock(&nmp->nm_lock);
4812 }
4813
4814 /*
4815 * Put a reasonable limit on the maximum timeout,
4816 * and reduce that limit when soft mounts get timeouts or are in reconnect.
4817 */
4818 if (!(NMFLAG(nmp, SOFT) || (req->r_flags & R_SOFT)) && !nfs_can_squish(nmp))
4819 maxtime = NFS_MAXTIMEO;
4820 else if ((req->r_flags & (R_SETUP|R_RECOVER)) ||
4821 ((nmp->nm_reconnect_start <= 0) || ((now.tv_sec - nmp->nm_reconnect_start) < 8)))
4822 maxtime = (NFS_MAXTIMEO / (nmp->nm_timeouts+1))/2;
4823 else
4824 maxtime = NFS_MINTIMEO/4;
4825
4826 /*
4827 * Check for request timeout.
4828 */
4829 if (req->r_rtt >= 0) {
4830 req->r_rtt++;
4831 lck_mtx_lock(&nmp->nm_lock);
4832 if (req->r_flags & R_RESENDERR) {
4833 /* with resend errors, retry every few seconds */
4834 timeo = 4*hz;
4835 } else {
4836 if (req->r_procnum == NFSPROC_NULL && req->r_gss_ctx != NULL)
4837 timeo = NFS_MINIDEMTIMEO; // gss context setup
4838 else if (NMFLAG(nmp, DUMBTIMER))
4839 timeo = nmp->nm_timeo;
4840 else
4841 timeo = NFS_RTO(nmp, proct[req->r_procnum]);
4842
4843 /* ensure 62.5 ms floor */
4844 while (16 * timeo < hz)
4845 timeo *= 2;
4846 if (nmp->nm_timeouts > 0)
4847 timeo *= nfs_backoff[nmp->nm_timeouts - 1];
4848 }
4849 /* limit timeout to max */
4850 if (timeo > maxtime)
4851 timeo = maxtime;
4852 if (req->r_rtt <= timeo) {
4853 NFS_SOCK_DBG("nfs timeout: req time %d and timeo is %d continue\n", req->r_rtt, timeo);
4854 lck_mtx_unlock(&nmp->nm_lock);
4855 lck_mtx_unlock(&req->r_mtx);
4856 continue;
4857 }
4858 /* The request has timed out */
4859 NFS_SOCK_DBG("nfs timeout: proc %d %d xid %llx rtt %d to %d # %d, t %ld/%d\n",
4860 req->r_procnum, proct[req->r_procnum],
4861 req->r_xid, req->r_rtt, timeo, nmp->nm_timeouts,
4862 (now.tv_sec - req->r_start)*NFS_HZ, maxtime);
4863 if (nmp->nm_timeouts < 8)
4864 nmp->nm_timeouts++;
4865 if (nfs_mount_check_dead_timeout(nmp)) {
4866 /* Unbusy this request */
4867 req->r_lflags &= ~RL_BUSY;
4868 if (req->r_lflags & RL_WAITING) {
4869 req->r_lflags &= ~RL_WAITING;
4870 wakeup(&req->r_lflags);
4871 }
4872 lck_mtx_unlock(&req->r_mtx);
4873
4874 /* No need to poke this mount */
4875 if (nmp->nm_sockflags & NMSOCK_POKE) {
4876 nmp->nm_sockflags &= ~NMSOCK_POKE;
4877 TAILQ_REMOVE(&nfs_mount_poke_queue, nmp, nm_pokeq);
4878 }
4879 /* Release our lock state, so we can become a zombie */
4880 lck_mtx_unlock(nfs_request_mutex);
4881
4882 /*
4883 * Note nfs_mount_make zombie(nmp) must be
4884 * called with nm_lock held. After doing some
4885 * work we release nm_lock in
4886 * nfs_make_mount_zombie with out acquiring any
4887 * other locks. (Later, in nfs_mount_zombie we
4888 * will acquire nfs_request_mutex, r_mtx,
4889 * nm_lock in that order). So we should not be
4890 * introducing deadlock here. We take a reference
4891 * on the mount so that its still there when we
4892 * release the lock.
4893 */
4894 nmp->nm_ref++;
4895 nfs_mount_make_zombie(nmp);
4896 lck_mtx_unlock(&nmp->nm_lock);
4897 nfs_mount_rele(nmp);
4898
4899 /*
4900 * All the request for this mount have now been
4901 * removed from the request queue. Restart to
4902 * process the remaining mounts
4903 */
4904 goto restart;
4905 }
4906
4907 /* if it's been a few seconds, try poking the socket */
4908 if ((nmp->nm_sotype == SOCK_STREAM) &&
4909 ((now.tv_sec - req->r_start) >= 3) &&
4910 !(nmp->nm_sockflags & (NMSOCK_POKE|NMSOCK_UNMOUNT)) &&
4911 (nmp->nm_sockflags & NMSOCK_READY)) {
4912 nmp->nm_sockflags |= NMSOCK_POKE;
4913 /*
4914 * We take a ref on the mount so that we know the mount will still be there
4915 * when we process the nfs_mount_poke_queue. An unmount request will block
4916 * in nfs_mount_drain_and_cleanup until after the poke is finished. We release
4917 * the reference after calling nfs_sock_poke below;
4918 */
4919 nmp->nm_ref++;
4920 TAILQ_INSERT_TAIL(&nfs_mount_poke_queue, nmp, nm_pokeq);
4921 }
4922 lck_mtx_unlock(&nmp->nm_lock);
4923 }
4924
4925 /* For soft mounts (& SETUPs/RECOVERs), check for too many retransmits/timeout. */
4926 if ((NMFLAG(nmp, SOFT) || (req->r_flags & (R_SETUP|R_RECOVER|R_SOFT))) &&
4927 ((req->r_rexmit >= req->r_retry) || /* too many */
4928 ((now.tv_sec - req->r_start)*NFS_HZ > maxtime))) { /* too long */
4929 OSAddAtomic64(1, &nfsstats.rpctimeouts);
4930 lck_mtx_lock(&nmp->nm_lock);
4931 if (!(nmp->nm_state & NFSSTA_TIMEO)) {
4932 lck_mtx_unlock(&nmp->nm_lock);
4933 /* make sure we note the unresponsive server */
4934 /* (maxtime may be less than tprintf delay) */
4935 nfs_down(req->r_nmp, req->r_thread, 0, NFSSTA_TIMEO,
4936 "not responding", 1);
4937 req->r_lastmsg = now.tv_sec;
4938 req->r_flags |= R_TPRINTFMSG;
4939 } else {
4940 lck_mtx_unlock(&nmp->nm_lock);
4941 }
4942 if (req->r_flags & R_NOINTR) {
4943 /* don't terminate nointr requests on timeout */
4944 lck_mtx_unlock(&req->r_mtx);
4945 continue;
4946 }
4947 NFS_SOCK_DBG("nfs timer TERMINATE: p %d x 0x%llx f 0x%x rtt %d t %ld\n",
4948 req->r_procnum, req->r_xid, req->r_flags, req->r_rtt,
4949 now.tv_sec - req->r_start);
4950 nfs_softterm(req);
4951 finish_asyncio = ((req->r_callback.rcb_func != NULL) && !(req->r_flags & R_WAITSENT));
4952 wakeup(req);
4953 lck_mtx_unlock(&req->r_mtx);
4954 if (finish_asyncio)
4955 nfs_asyncio_finish(req);
4956 continue;
4957 }
4958
4959 /* for TCP, only resend if explicitly requested */
4960 if ((nmp->nm_sotype == SOCK_STREAM) && !(req->r_flags & R_MUSTRESEND)) {
4961 if (++req->r_rexmit > NFS_MAXREXMIT)
4962 req->r_rexmit = NFS_MAXREXMIT;
4963 req->r_rtt = 0;
4964 lck_mtx_unlock(&req->r_mtx);
4965 continue;
4966 }
4967
4968 /*
4969 * The request needs to be (re)sent. Kick the requester to resend it.
4970 * (unless it's already marked as needing a resend)
4971 */
4972 if ((req->r_flags & R_MUSTRESEND) && (req->r_rtt == -1)) {
4973 lck_mtx_unlock(&req->r_mtx);
4974 continue;
4975 }
4976 NFS_SOCK_DBG("nfs timer mark resend: p %d x 0x%llx f 0x%x rtt %d\n",
4977 req->r_procnum, req->r_xid, req->r_flags, req->r_rtt);
4978 req->r_flags |= R_MUSTRESEND;
4979 req->r_rtt = -1;
4980 wakeup(req);
4981 if ((req->r_flags & (R_IOD|R_ASYNC|R_ASYNCWAIT|R_SENDING)) == R_ASYNC)
4982 nfs_asyncio_resend(req);
4983 lck_mtx_unlock(&req->r_mtx);
4984 }
4985
4986 lck_mtx_unlock(nfs_request_mutex);
4987
4988 /* poke any sockets */
4989 while ((nmp = TAILQ_FIRST(&nfs_mount_poke_queue))) {
4990 TAILQ_REMOVE(&nfs_mount_poke_queue, nmp, nm_pokeq);
4991 nfs_sock_poke(nmp);
4992 nfs_mount_rele(nmp);
4993 }
4994
4995 nfs_interval_timer_start(nfs_request_timer_call, NFS_REQUESTDELAY);
4996}
4997
4998/*
4999 * check a thread's proc for the "noremotehang" flag.
5000 */
5001int
5002nfs_noremotehang(thread_t thd)
5003{
5004 proc_t p = thd ? get_bsdthreadtask_info(thd) : NULL;
5005 return (p && proc_noremotehang(p));
5006}
5007
5008/*
5009 * Test for a termination condition pending on the process.
5010 * This is used to determine if we need to bail on a mount.
5011 * ETIMEDOUT is returned if there has been a soft timeout.
5012 * EINTR is returned if there is a signal pending that is not being ignored
5013 * and the mount is interruptable, or if we are a thread that is in the process
5014 * of cancellation (also SIGKILL posted).
5015 */
5016extern int sigprop[NSIG+1];
5017int
5018nfs_sigintr(struct nfsmount *nmp, struct nfsreq *req, thread_t thd, int nmplocked)
5019{
5020 proc_t p;
5021 int error = 0;
5022
5023 if (!nmp)
5024 return (ENXIO);
5025
5026 if (req && (req->r_flags & R_SOFTTERM))
5027 return (ETIMEDOUT); /* request has been terminated. */
5028 if (req && (req->r_flags & R_NOINTR))
5029 thd = NULL; /* don't check for signal on R_NOINTR */
5030
5031 if (!nmplocked)
5032 lck_mtx_lock(&nmp->nm_lock);
5033 if (nmp->nm_state & NFSSTA_FORCE) {
5034 /* If a force unmount is in progress then fail. */
5035 error = EIO;
5036 } else if (vfs_isforce(nmp->nm_mountp)) {
5037 /* Someone is unmounting us, go soft and mark it. */
5038 NFS_BITMAP_SET(nmp->nm_flags, NFS_MFLAG_SOFT);
5039 nmp->nm_state |= NFSSTA_FORCE;
5040 }
5041
5042 /* Check if the mount is marked dead. */
5043 if (!error && (nmp->nm_state & NFSSTA_DEAD))
5044 error = ENXIO;
5045
5046 /*
5047 * If the mount is hung and we've requested not to hang
5048 * on remote filesystems, then bail now.
5049 */
5050 if (current_proc() != kernproc &&
5051 !error && (nmp->nm_state & NFSSTA_TIMEO) && nfs_noremotehang(thd))
5052 error = EIO;
5053
5054 if (!nmplocked)
5055 lck_mtx_unlock(&nmp->nm_lock);
5056 if (error)
5057 return (error);
5058
5059 /* may not have a thread for async I/O */
5060 if (thd == NULL || current_proc() == kernproc)
5061 return (0);
5062
5063 /*
5064 * Check if the process is aborted, but don't interrupt if we
5065 * were killed by a signal and this is the exiting thread which
5066 * is attempting to dump core.
5067 */
5068 if (((p = current_proc()) != kernproc) && current_thread_aborted() &&
5069 (!(p->p_acflag & AXSIG) || (p->exit_thread != current_thread()) ||
5070 (p->p_sigacts == NULL) ||
5071 (p->p_sigacts->ps_sig < 1) || (p->p_sigacts->ps_sig > NSIG) ||
5072 !(sigprop[p->p_sigacts->ps_sig] & SA_CORE)))
5073 return (EINTR);
5074
5075 /* mask off thread and process blocked signals. */
5076 if (NMFLAG(nmp, INTR) && ((p = get_bsdthreadtask_info(thd))) &&
5077 proc_pendingsignals(p, NFSINT_SIGMASK))
5078 return (EINTR);
5079 return (0);
5080}
5081
5082/*
5083 * Lock a socket against others.
5084 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply
5085 * and also to avoid race conditions between the processes with nfs requests
5086 * in progress when a reconnect is necessary.
5087 */
5088int
5089nfs_sndlock(struct nfsreq *req)
5090{
5091 struct nfsmount *nmp = req->r_nmp;
5092 int *statep;
5093 int error = 0, slpflag = 0;
5094 struct timespec ts = { 0, 0 };
5095
5096 if (nfs_mount_gone(nmp))
5097 return (ENXIO);
5098
5099 lck_mtx_lock(&nmp->nm_lock);
5100 statep = &nmp->nm_state;
5101
5102 if (NMFLAG(nmp, INTR) && req->r_thread && !(req->r_flags & R_NOINTR))
5103 slpflag = PCATCH;
5104 while (*statep & NFSSTA_SNDLOCK) {
5105 if ((error = nfs_sigintr(nmp, req, req->r_thread, 1)))
5106 break;
5107 *statep |= NFSSTA_WANTSND;
5108 if (nfs_noremotehang(req->r_thread))
5109 ts.tv_sec = 1;
5110 msleep(statep, &nmp->nm_lock, slpflag | (PZERO - 1), "nfsndlck", &ts);
5111 if (slpflag == PCATCH) {
5112 slpflag = 0;
5113 ts.tv_sec = 2;
5114 }
5115 }
5116 if (!error)
5117 *statep |= NFSSTA_SNDLOCK;
5118 lck_mtx_unlock(&nmp->nm_lock);
5119 return (error);
5120}
5121
5122/*
5123 * Unlock the stream socket for others.
5124 */
5125void
5126nfs_sndunlock(struct nfsreq *req)
5127{
5128 struct nfsmount *nmp = req->r_nmp;
5129 int *statep, wake = 0;
5130
5131 if (!nmp)
5132 return;
5133 lck_mtx_lock(&nmp->nm_lock);
5134 statep = &nmp->nm_state;
5135 if ((*statep & NFSSTA_SNDLOCK) == 0)
5136 panic("nfs sndunlock");
5137 *statep &= ~(NFSSTA_SNDLOCK|NFSSTA_SENDING);
5138 if (*statep & NFSSTA_WANTSND) {
5139 *statep &= ~NFSSTA_WANTSND;
5140 wake = 1;
5141 }
5142 lck_mtx_unlock(&nmp->nm_lock);
5143 if (wake)
5144 wakeup(statep);
5145}
5146
5147int
5148nfs_aux_request(
5149 struct nfsmount *nmp,
5150 thread_t thd,
5151 struct sockaddr *saddr,
5152 socket_t so,
5153 int sotype,
5154 mbuf_t mreq,
5155 uint32_t xid,
5156 int bindresv,
5157 int timeo,
5158 struct nfsm_chain *nmrep)
5159{
5160 int error = 0, on = 1, try, sendat = 2, soproto, recv, optlen, restoreto = 0;
5161 socket_t newso = NULL;
5162 struct sockaddr_storage ss;
5163 struct timeval orig_rcvto, orig_sndto, tv = { 1, 0 };
5164 mbuf_t m, mrep = NULL;
5165 struct msghdr msg;
5166 uint32_t rxid = 0, reply = 0, reply_status, rejected_status;
5167 uint32_t verf_type, verf_len, accepted_status;
5168 size_t readlen, sentlen;
5169 struct nfs_rpc_record_state nrrs;
5170
5171 if (!so) {
5172 /* create socket and set options */
5173 soproto = (sotype == SOCK_DGRAM) ? IPPROTO_UDP : IPPROTO_TCP;
5174 if ((error = sock_socket(saddr->sa_family, sotype, soproto, NULL, NULL, &newso)))
5175 goto nfsmout;
5176
5177 if (bindresv) {
5178 int level = (saddr->sa_family == AF_INET) ? IPPROTO_IP : IPPROTO_IPV6;
5179 int optname = (saddr->sa_family == AF_INET) ? IP_PORTRANGE : IPV6_PORTRANGE;
5180 int portrange = IP_PORTRANGE_LOW;
5181 error = sock_setsockopt(newso, level, optname, &portrange, sizeof(portrange));
5182 nfsmout_if(error);
5183 ss.ss_len = saddr->sa_len;
5184 ss.ss_family = saddr->sa_family;
5185 if (ss.ss_family == AF_INET) {
5186 ((struct sockaddr_in*)&ss)->sin_addr.s_addr = INADDR_ANY;
5187 ((struct sockaddr_in*)&ss)->sin_port = htons(0);
5188 } else if (ss.ss_family == AF_INET6) {
5189 ((struct sockaddr_in6*)&ss)->sin6_addr = in6addr_any;
5190 ((struct sockaddr_in6*)&ss)->sin6_port = htons(0);
5191 } else {
5192 error = EINVAL;
5193 }
5194 if (!error)
5195 error = sock_bind(newso, (struct sockaddr *)&ss);
5196 nfsmout_if(error);
5197 }
5198
5199 if (sotype == SOCK_STREAM) {
5200# define NFS_AUX_CONNECTION_TIMEOUT 4 /* 4 second timeout for connections */
5201 int count = 0;
5202
5203 error = sock_connect(newso, saddr, MSG_DONTWAIT);
5204 if (error == EINPROGRESS)
5205 error = 0;
5206 nfsmout_if(error);
5207
5208 while ((error = sock_connectwait(newso, &tv)) == EINPROGRESS) {
5209 /* After NFS_AUX_CONNECTION_TIMEOUT bail */
5210 if (++count >= NFS_AUX_CONNECTION_TIMEOUT) {
5211 error = ETIMEDOUT;
5212 break;
5213 }
5214 }
5215 nfsmout_if(error);
5216 }
5217 if (((error = sock_setsockopt(newso, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)))) ||
5218 ((error = sock_setsockopt(newso, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv)))) ||
5219 ((error = sock_setsockopt(newso, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on)))))
5220 goto nfsmout;
5221 so = newso;
5222 } else {
5223 /* make sure socket is using a one second timeout in this function */
5224 optlen = sizeof(orig_rcvto);
5225 error = sock_getsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &orig_rcvto, &optlen);
5226 if (!error) {
5227 optlen = sizeof(orig_sndto);
5228 error = sock_getsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &orig_sndto, &optlen);
5229 }
5230 if (!error) {
5231 sock_setsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv));
5232 sock_setsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv));
5233 restoreto = 1;
5234 }
5235 }
5236
5237 if (sotype == SOCK_STREAM) {
5238 sendat = 0; /* we only resend the request for UDP */
5239 nfs_rpc_record_state_init(&nrrs);
5240 }
5241
5242 for (try=0; try < timeo; try++) {
5243 if ((error = nfs_sigintr(nmp, NULL, !try ? NULL : thd, 0)))
5244 break;
5245 if (!try || (try == sendat)) {
5246 /* send the request (resending periodically for UDP) */
5247 if ((error = mbuf_copym(mreq, 0, MBUF_COPYALL, MBUF_WAITOK, &m)))
5248 goto nfsmout;
5249 bzero(&msg, sizeof(msg));
5250 if ((sotype == SOCK_DGRAM) && !sock_isconnected(so)) {
5251 msg.msg_name = saddr;
5252 msg.msg_namelen = saddr->sa_len;
5253 }
5254 if ((error = sock_sendmbuf(so, &msg, m, 0, &sentlen)))
5255 goto nfsmout;
5256 sendat *= 2;
5257 if (sendat > 30)
5258 sendat = 30;
5259 }
5260 /* wait for the response */
5261 if (sotype == SOCK_STREAM) {
5262 /* try to read (more of) record */
5263 error = nfs_rpc_record_read(so, &nrrs, 0, &recv, &mrep);
5264 /* if we don't have the whole record yet, we'll keep trying */
5265 } else {
5266 readlen = 1<<18;
5267 bzero(&msg, sizeof(msg));
5268 error = sock_receivembuf(so, &msg, &mrep, 0, &readlen);
5269 }
5270 if (error == EWOULDBLOCK)
5271 continue;
5272 nfsmout_if(error);
5273 /* parse the response */
5274 nfsm_chain_dissect_init(error, nmrep, mrep);
5275 nfsm_chain_get_32(error, nmrep, rxid);
5276 nfsm_chain_get_32(error, nmrep, reply);
5277 nfsmout_if(error);
5278 if ((rxid != xid) || (reply != RPC_REPLY))
5279 error = EBADRPC;
5280 nfsm_chain_get_32(error, nmrep, reply_status);
5281 nfsmout_if(error);
5282 if (reply_status == RPC_MSGDENIED) {
5283 nfsm_chain_get_32(error, nmrep, rejected_status);
5284 nfsmout_if(error);
5285 error = (rejected_status == RPC_MISMATCH) ? ERPCMISMATCH : EACCES;
5286 goto nfsmout;
5287 }
5288 nfsm_chain_get_32(error, nmrep, verf_type); /* verifier flavor */
5289 nfsm_chain_get_32(error, nmrep, verf_len); /* verifier length */
5290 nfsmout_if(error);
5291 if (verf_len)
5292 nfsm_chain_adv(error, nmrep, nfsm_rndup(verf_len));
5293 nfsm_chain_get_32(error, nmrep, accepted_status);
5294 nfsmout_if(error);
5295 switch (accepted_status) {
5296 case RPC_SUCCESS:
5297 error = 0;
5298 break;
5299 case RPC_PROGUNAVAIL:
5300 error = EPROGUNAVAIL;
5301 break;
5302 case RPC_PROGMISMATCH:
5303 error = EPROGMISMATCH;
5304 break;
5305 case RPC_PROCUNAVAIL:
5306 error = EPROCUNAVAIL;
5307 break;
5308 case RPC_GARBAGE:
5309 error = EBADRPC;
5310 break;
5311 case RPC_SYSTEM_ERR:
5312 default:
5313 error = EIO;
5314 break;
5315 }
5316 break;
5317 }
5318nfsmout:
5319 if (restoreto) {
5320 sock_setsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &orig_rcvto, sizeof(tv));
5321 sock_setsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &orig_sndto, sizeof(tv));
5322 }
5323 if (newso) {
5324 sock_shutdown(newso, SHUT_RDWR);
5325 sock_close(newso);
5326 }
5327 mbuf_freem(mreq);
5328 return (error);
5329}
5330
5331int
5332nfs_portmap_lookup(
5333 struct nfsmount *nmp,
5334 vfs_context_t ctx,
5335 struct sockaddr *sa,
5336 socket_t so,
5337 uint32_t protocol,
5338 uint32_t vers,
5339 uint32_t ipproto,
5340 int timeo)
5341{
5342 thread_t thd = vfs_context_thread(ctx);
5343 kauth_cred_t cred = vfs_context_ucred(ctx);
5344 struct sockaddr_storage ss;
5345 struct sockaddr *saddr = (struct sockaddr*)&ss;
5346 struct nfsm_chain nmreq, nmrep;
5347 mbuf_t mreq;
5348 int error = 0, ip, pmprog, pmvers, pmproc;
5349 uint32_t ualen = 0;
5350 uint32_t port;
5351 uint64_t xid = 0;
5352 char uaddr[MAX_IPv6_STR_LEN+16];
5353
5354 bcopy(sa, saddr, min(sizeof(ss), sa->sa_len));
5355 if (saddr->sa_family == AF_INET) {
5356 ip = 4;
5357 pmprog = PMAPPROG;
5358 pmvers = PMAPVERS;
5359 pmproc = PMAPPROC_GETPORT;
5360 } else if (saddr->sa_family == AF_INET6) {
5361 ip = 6;
5362 pmprog = RPCBPROG;
5363 pmvers = RPCBVERS4;
5364 pmproc = RPCBPROC_GETVERSADDR;
5365 } else {
5366 return (EINVAL);
5367 }
5368 nfsm_chain_null(&nmreq);
5369 nfsm_chain_null(&nmrep);
5370
5371tryagain:
5372 /* send portmapper request to get port/uaddr */
5373 if (ip == 4)
5374 ((struct sockaddr_in*)saddr)->sin_port = htons(PMAPPORT);
5375 else
5376 ((struct sockaddr_in6*)saddr)->sin6_port = htons(PMAPPORT);
5377 nfsm_chain_build_alloc_init(error, &nmreq, 8*NFSX_UNSIGNED);
5378 nfsm_chain_add_32(error, &nmreq, protocol);
5379 nfsm_chain_add_32(error, &nmreq, vers);
5380 if (ip == 4) {
5381 nfsm_chain_add_32(error, &nmreq, ipproto);
5382 nfsm_chain_add_32(error, &nmreq, 0);
5383 } else {
5384 if (ipproto == IPPROTO_TCP)
5385 nfsm_chain_add_string(error, &nmreq, "tcp6", 4);
5386 else
5387 nfsm_chain_add_string(error, &nmreq, "udp6", 4);
5388 nfsm_chain_add_string(error, &nmreq, "", 0); /* uaddr */
5389 nfsm_chain_add_string(error, &nmreq, "", 0); /* owner */
5390 }
5391 nfsm_chain_build_done(error, &nmreq);
5392 nfsmout_if(error);
5393 error = nfsm_rpchead2(nmp, (ipproto == IPPROTO_UDP) ? SOCK_DGRAM : SOCK_STREAM,
5394 pmprog, pmvers, pmproc, RPCAUTH_SYS, cred, NULL, nmreq.nmc_mhead,
5395 &xid, &mreq);
5396 nfsmout_if(error);
5397 nmreq.nmc_mhead = NULL;
5398 error = nfs_aux_request(nmp, thd, saddr, so, (ipproto == IPPROTO_UDP) ? SOCK_DGRAM : SOCK_STREAM,
5399 mreq, R_XID32(xid), 0, timeo, &nmrep);
5400
5401 /* grab port from portmap response */
5402 if (ip == 4) {
5403 nfsm_chain_get_32(error, &nmrep, port);
5404 if (!error)
5405 ((struct sockaddr_in*)sa)->sin_port = htons(port);
5406 } else {
5407 /* get uaddr string and convert to sockaddr */
5408 nfsm_chain_get_32(error, &nmrep, ualen);
5409 if (!error) {
5410 if (ualen > (sizeof(uaddr)-1))
5411 error = EIO;
5412 if (ualen < 1) {
5413 /* program is not available, just return a zero port */
5414 bcopy(sa, saddr, min(sizeof(ss), sa->sa_len));
5415 ((struct sockaddr_in6*)saddr)->sin6_port = htons(0);
5416 } else {
5417 nfsm_chain_get_opaque(error, &nmrep, ualen, uaddr);
5418 if (!error) {
5419 uaddr[ualen] = '\0';
5420 if (!nfs_uaddr2sockaddr(uaddr, saddr))
5421 error = EIO;
5422 }
5423 }
5424 }
5425 if ((error == EPROGMISMATCH) || (error == EPROCUNAVAIL) || (error == EIO) || (error == EBADRPC)) {
5426 /* remote doesn't support rpcbind version or proc (or we couldn't parse uaddr) */
5427 if (pmvers == RPCBVERS4) {
5428 /* fall back to v3 and GETADDR */
5429 pmvers = RPCBVERS3;
5430 pmproc = RPCBPROC_GETADDR;
5431 nfsm_chain_cleanup(&nmreq);
5432 nfsm_chain_cleanup(&nmrep);
5433 bcopy(sa, saddr, min(sizeof(ss), sa->sa_len));
5434 xid = 0;
5435 error = 0;
5436 goto tryagain;
5437 }
5438 }
5439 if (!error)
5440 bcopy(saddr, sa, min(saddr->sa_len, sa->sa_len));
5441 }
5442nfsmout:
5443 nfsm_chain_cleanup(&nmreq);
5444 nfsm_chain_cleanup(&nmrep);
5445 return (error);
5446}
5447
5448int
5449nfs_msg(thread_t thd,
5450 const char *server,
5451 const char *msg,
5452 int error)
5453{
5454 proc_t p = thd ? get_bsdthreadtask_info(thd) : NULL;
5455 tpr_t tpr;
5456
5457 if (p)
5458 tpr = tprintf_open(p);
5459 else
5460 tpr = NULL;
5461 if (error)
5462 tprintf(tpr, "nfs server %s: %s, error %d\n", server, msg, error);
5463 else
5464 tprintf(tpr, "nfs server %s: %s\n", server, msg);
5465 tprintf_close(tpr);
5466 return (0);
5467}
5468
5469#define NFS_SQUISH_MOBILE_ONLY 0x0001 /* Squish mounts only on mobile machines */
5470#define NFS_SQUISH_AUTOMOUNTED_ONLY 0x0002 /* Squish mounts only if the are automounted */
5471#define NFS_SQUISH_SOFT 0x0004 /* Treat all soft mounts as though they were on a mobile machine */
5472#define NFS_SQUISH_QUICK 0x0008 /* Try to squish mounts more quickly. */
5473#define NFS_SQUISH_SHUTDOWN 0x1000 /* Squish all mounts on shutdown. Currently not implemented */
5474
5475uint32_t nfs_squishy_flags = NFS_SQUISH_MOBILE_ONLY | NFS_SQUISH_AUTOMOUNTED_ONLY | NFS_SQUISH_QUICK;
5476int32_t nfs_is_mobile;
5477
5478#define NFS_SQUISHY_DEADTIMEOUT 8 /* Dead time out for squishy mounts */
5479#define NFS_SQUISHY_QUICKTIMEOUT 4 /* Quicker dead time out when nfs_squish_flags NFS_SQUISH_QUICK bit is set*/
5480
5481/*
5482 * Could this mount be squished?
5483 */
5484int
5485nfs_can_squish(struct nfsmount *nmp)
5486{
5487 uint64_t flags = vfs_flags(nmp->nm_mountp);
5488 int softsquish = ((nfs_squishy_flags & NFS_SQUISH_SOFT) & NMFLAG(nmp, SOFT));
5489
5490 if (!softsquish && (nfs_squishy_flags & NFS_SQUISH_MOBILE_ONLY) && nfs_is_mobile == 0)
5491 return (0);
5492
5493 if ((nfs_squishy_flags & NFS_SQUISH_AUTOMOUNTED_ONLY) && (flags & MNT_AUTOMOUNTED) == 0)
5494 return (0);
5495
5496 return (1);
5497}
5498
5499/*
5500 * NFS mounts default to "rw,hard" - but frequently on mobile clients
5501 * the mount may become "not responding". It's desirable to be able
5502 * to unmount these dead mounts, but only if there is no risk of
5503 * losing data or crashing applications. A "squishy" NFS mount is one
5504 * that can be force unmounted with little risk of harm.
5505 *
5506 * nfs_is_squishy checks if a mount is in a squishy state. A mount is
5507 * in a squishy state iff it is allowed to be squishy and there are no
5508 * dirty pages and there are no mmapped files and there are no files
5509 * open for write. Mounts are allowed to be squishy is controlled by
5510 * the settings of the nfs_squishy_flags and its mobility state. These
5511 * flags can be set by sysctls.
5512 *
5513 * If nfs_is_squishy determines that we are in a squishy state we will
5514 * update the current dead timeout to at least NFS_SQUISHY_DEADTIMEOUT
5515 * (or NFS_SQUISHY_QUICKTIMEOUT if NFS_SQUISH_QUICK is set) (see
5516 * above) or 1/8th of the mount's nm_deadtimeout value, otherwise we just
5517 * update the current dead timeout with the mount's nm_deadtimeout
5518 * value set at mount time.
5519 *
5520 * Assumes that nm_lock is held.
5521 *
5522 * Note this routine is racey, but its effects on setting the
5523 * dead timeout only have effects when we're in trouble and are likely
5524 * to stay that way. Since by default its only for automounted
5525 * volumes on mobile machines; this is a reasonable trade off between
5526 * data integrity and user experience. It can be disabled or set via
5527 * nfs.conf file.
5528 */
5529
5530int
5531nfs_is_squishy(struct nfsmount *nmp)
5532{
5533 mount_t mp = nmp->nm_mountp;
5534 int squishy = 0;
5535 int timeo = (nfs_squishy_flags & NFS_SQUISH_QUICK) ? NFS_SQUISHY_QUICKTIMEOUT : NFS_SQUISHY_DEADTIMEOUT;
5536
5537 NFS_SOCK_DBG("%s: nm_curdeadtimeout = %d, nfs_is_mobile = %d\n",
5538 vfs_statfs(mp)->f_mntfromname, nmp->nm_curdeadtimeout, nfs_is_mobile);
5539
5540 if (!nfs_can_squish(nmp))
5541 goto out;
5542
5543 timeo = (nmp->nm_deadtimeout > timeo) ? max(nmp->nm_deadtimeout/8, timeo) : timeo;
5544 NFS_SOCK_DBG("nm_writers = %d nm_mappers = %d timeo = %d\n", nmp->nm_writers, nmp->nm_mappers, timeo);
5545
5546 if (nmp->nm_writers == 0 && nmp->nm_mappers == 0) {
5547 uint64_t flags = mp ? vfs_flags(mp) : 0;
5548 squishy = 1;
5549
5550 /*
5551 * Walk the nfs nodes and check for dirty buffers it we're not
5552 * RDONLY and we've not already been declared as squishy since
5553 * this can be a bit expensive.
5554 */
5555 if (!(flags & MNT_RDONLY) && !(nmp->nm_state & NFSSTA_SQUISHY))
5556 squishy = !nfs_mount_is_dirty(mp);
5557 }
5558
5559out:
5560 if (squishy)
5561 nmp->nm_state |= NFSSTA_SQUISHY;
5562 else
5563 nmp->nm_state &= ~NFSSTA_SQUISHY;
5564
5565 nmp->nm_curdeadtimeout = squishy ? timeo : nmp->nm_deadtimeout;
5566
5567 NFS_SOCK_DBG("nm_curdeadtimeout = %d\n", nmp->nm_curdeadtimeout);
5568
5569 return (squishy);
5570}
5571
5572/*
5573 * On a send operation, if we can't reach the server and we've got only one server to talk to
5574 * and NFS_SQUISH_QUICK flag is set and we are in a squishy state then mark the mount as dead
5575 * and ask to be forcibly unmounted. Return 1 if we're dead and 0 otherwise.
5576 */
5577int
5578nfs_is_dead(int error, struct nfsmount *nmp)
5579{
5580 fsid_t fsid;
5581
5582 lck_mtx_lock(&nmp->nm_lock);
5583 if (nmp->nm_state & NFSSTA_DEAD) {
5584 lck_mtx_unlock(&nmp->nm_lock);
5585 return (1);
5586 }
5587
5588 if ((error != ENETUNREACH && error != EHOSTUNREACH && error != EADDRNOTAVAIL) ||
5589 !(nmp->nm_locations.nl_numlocs == 1 && nmp->nm_locations.nl_locations[0]->nl_servcount == 1)) {
5590 lck_mtx_unlock(&nmp->nm_lock);
5591 return (0);
5592 }
5593
5594 if ((nfs_squishy_flags & NFS_SQUISH_QUICK) && nfs_is_squishy(nmp)) {
5595 printf("nfs_is_dead: nfs server %s: unreachable. Squished dead\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname);
5596 fsid = vfs_statfs(nmp->nm_mountp)->f_fsid;
5597 lck_mtx_unlock(&nmp->nm_lock);
5598 nfs_mount_zombie(nmp, NFSSTA_DEAD);
5599 vfs_event_signal(&fsid, VQ_DEAD, 0);
5600 return (1);
5601 }
5602 lck_mtx_unlock(&nmp->nm_lock);
5603 return (0);
5604}
5605
5606/*
5607 * If we've experienced timeouts and we're not really a
5608 * classic hard mount, then just return cached data to
5609 * the caller instead of likely hanging on an RPC.
5610 */
5611int
5612nfs_use_cache(struct nfsmount *nmp)
5613{
5614 /*
5615 *%%% We always let mobile users goto the cache,
5616 * perhaps we should not even require them to have
5617 * a timeout?
5618 */
5619 int cache_ok = (nfs_is_mobile || NMFLAG(nmp, SOFT) ||
5620 nfs_can_squish(nmp) || nmp->nm_deadtimeout);
5621
5622 int timeoutmask = NFSSTA_TIMEO | NFSSTA_LOCKTIMEO | NFSSTA_JUKEBOXTIMEO;
5623
5624 /*
5625 * So if we have a timeout and we're not really a hard hard-mount,
5626 * return 1 to not get things out of the cache.
5627 */
5628
5629 return ((nmp->nm_state & timeoutmask) && cache_ok);
5630}
5631
5632/*
5633 * Log a message that nfs or lockd server is unresponsive. Check if we
5634 * can be squished and if we can, or that our dead timeout has
5635 * expired, and we're not holding state, set our mount as dead, remove
5636 * our mount state and ask to be unmounted. If we are holding state
5637 * we're being called from the nfs_request_timer and will soon detect
5638 * that we need to unmount.
5639 */
5640void
5641nfs_down(struct nfsmount *nmp, thread_t thd, int error, int flags, const char *msg, int holding_state)
5642{
5643 int timeoutmask, wasunresponsive, unresponsive, softnobrowse;
5644 uint32_t do_vfs_signal = 0;
5645 struct timeval now;
5646
5647 if (nfs_mount_gone(nmp))
5648 return;
5649
5650 lck_mtx_lock(&nmp->nm_lock);
5651
5652 timeoutmask = NFSSTA_TIMEO | NFSSTA_LOCKTIMEO | NFSSTA_JUKEBOXTIMEO;
5653 if (NMFLAG(nmp, MUTEJUKEBOX)) /* jukebox timeouts don't count as unresponsive if muted */
5654 timeoutmask &= ~NFSSTA_JUKEBOXTIMEO;
5655 wasunresponsive = (nmp->nm_state & timeoutmask);
5656
5657 /* XXX don't allow users to know about/disconnect unresponsive, soft, nobrowse mounts */
5658 softnobrowse = (NMFLAG(nmp, SOFT) && (vfs_flags(nmp->nm_mountp) & MNT_DONTBROWSE));
5659
5660 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO))
5661 nmp->nm_state |= NFSSTA_TIMEO;
5662 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO))
5663 nmp->nm_state |= NFSSTA_LOCKTIMEO;
5664 if ((flags & NFSSTA_JUKEBOXTIMEO) && !(nmp->nm_state & NFSSTA_JUKEBOXTIMEO))
5665 nmp->nm_state |= NFSSTA_JUKEBOXTIMEO;
5666
5667 unresponsive = (nmp->nm_state & timeoutmask);
5668
5669 nfs_is_squishy(nmp);
5670
5671 if (unresponsive && (nmp->nm_curdeadtimeout > 0)) {
5672 microuptime(&now);
5673 if (!wasunresponsive) {
5674 nmp->nm_deadto_start = now.tv_sec;
5675 nfs_mount_sock_thread_wake(nmp);
5676 } else if ((now.tv_sec - nmp->nm_deadto_start) > nmp->nm_curdeadtimeout && !holding_state) {
5677 if (!(nmp->nm_state & NFSSTA_DEAD))
5678 printf("nfs server %s: %sdead\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname,
5679 (nmp->nm_curdeadtimeout != nmp->nm_deadtimeout) ? "squished " : "");
5680 do_vfs_signal = VQ_DEAD;
5681 }
5682 }
5683 lck_mtx_unlock(&nmp->nm_lock);
5684
5685 if (do_vfs_signal == VQ_DEAD && !(nmp->nm_state & NFSSTA_DEAD))
5686 nfs_mount_zombie(nmp, NFSSTA_DEAD);
5687 else if (softnobrowse || wasunresponsive || !unresponsive)
5688 do_vfs_signal = 0;
5689 else
5690 do_vfs_signal = VQ_NOTRESP;
5691 if (do_vfs_signal)
5692 vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, do_vfs_signal, 0);
5693
5694 nfs_msg(thd, vfs_statfs(nmp->nm_mountp)->f_mntfromname, msg, error);
5695}
5696
5697void
5698nfs_up(struct nfsmount *nmp, thread_t thd, int flags, const char *msg)
5699{
5700 int timeoutmask, wasunresponsive, unresponsive, softnobrowse;
5701 int do_vfs_signal;
5702
5703 if (nfs_mount_gone(nmp))
5704 return;
5705
5706 if (msg)
5707 nfs_msg(thd, vfs_statfs(nmp->nm_mountp)->f_mntfromname, msg, 0);
5708
5709 lck_mtx_lock(&nmp->nm_lock);
5710
5711 timeoutmask = NFSSTA_TIMEO | NFSSTA_LOCKTIMEO | NFSSTA_JUKEBOXTIMEO;
5712 if (NMFLAG(nmp, MUTEJUKEBOX)) /* jukebox timeouts don't count as unresponsive if muted */
5713 timeoutmask &= ~NFSSTA_JUKEBOXTIMEO;
5714 wasunresponsive = (nmp->nm_state & timeoutmask);
5715
5716 /* XXX don't allow users to know about/disconnect unresponsive, soft, nobrowse mounts */
5717 softnobrowse = (NMFLAG(nmp, SOFT) && (vfs_flags(nmp->nm_mountp) & MNT_DONTBROWSE));
5718
5719 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO))
5720 nmp->nm_state &= ~NFSSTA_TIMEO;
5721 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO))
5722 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
5723 if ((flags & NFSSTA_JUKEBOXTIMEO) && (nmp->nm_state & NFSSTA_JUKEBOXTIMEO))
5724 nmp->nm_state &= ~NFSSTA_JUKEBOXTIMEO;
5725
5726 unresponsive = (nmp->nm_state & timeoutmask);
5727
5728 nmp->nm_deadto_start = 0;
5729 nmp->nm_curdeadtimeout = nmp->nm_deadtimeout;
5730 nmp->nm_state &= ~NFSSTA_SQUISHY;
5731 lck_mtx_unlock(&nmp->nm_lock);
5732
5733 if (softnobrowse)
5734 do_vfs_signal = 0;
5735 else
5736 do_vfs_signal = (wasunresponsive && !unresponsive);
5737 if (do_vfs_signal)
5738 vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, VQ_NOTRESP, 1);
5739}
5740
5741
5742#endif /* NFSCLIENT */
5743
5744#if NFSSERVER
5745
5746/*
5747 * Generate the rpc reply header
5748 * siz arg. is used to decide if adding a cluster is worthwhile
5749 */
5750int
5751nfsrv_rephead(
5752 struct nfsrv_descript *nd,
5753 __unused struct nfsrv_sock *slp,
5754 struct nfsm_chain *nmrepp,
5755 size_t siz)
5756{
5757 mbuf_t mrep;
5758 u_int32_t *tl;
5759 struct nfsm_chain nmrep;
5760 int err, error;
5761
5762 err = nd->nd_repstat;
5763 if (err && (nd->nd_vers == NFS_VER2))
5764 siz = 0;
5765
5766 /*
5767 * If this is a big reply, use a cluster else
5768 * try and leave leading space for the lower level headers.
5769 */
5770 siz += RPC_REPLYSIZ;
5771 if (siz >= nfs_mbuf_minclsize) {
5772 error = mbuf_getpacket(MBUF_WAITOK, &mrep);
5773 } else {
5774 error = mbuf_gethdr(MBUF_WAITOK, MBUF_TYPE_DATA, &mrep);
5775 }
5776 if (error) {
5777 /* unable to allocate packet */
5778 /* XXX should we keep statistics for these errors? */
5779 return (error);
5780 }
5781 if (siz < nfs_mbuf_minclsize) {
5782 /* leave space for lower level headers */
5783 tl = mbuf_data(mrep);
5784 tl += 80/sizeof(*tl); /* XXX max_hdr? XXX */
5785 mbuf_setdata(mrep, tl, 6 * NFSX_UNSIGNED);
5786 }
5787 nfsm_chain_init(&nmrep, mrep);
5788 nfsm_chain_add_32(error, &nmrep, nd->nd_retxid);
5789 nfsm_chain_add_32(error, &nmrep, RPC_REPLY);
5790 if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
5791 nfsm_chain_add_32(error, &nmrep, RPC_MSGDENIED);
5792 if (err & NFSERR_AUTHERR) {
5793 nfsm_chain_add_32(error, &nmrep, RPC_AUTHERR);
5794 nfsm_chain_add_32(error, &nmrep, (err & ~NFSERR_AUTHERR));
5795 } else {
5796 nfsm_chain_add_32(error, &nmrep, RPC_MISMATCH);
5797 nfsm_chain_add_32(error, &nmrep, RPC_VER2);
5798 nfsm_chain_add_32(error, &nmrep, RPC_VER2);
5799 }
5800 } else {
5801 /* reply status */
5802 nfsm_chain_add_32(error, &nmrep, RPC_MSGACCEPTED);
5803 if (nd->nd_gss_context != NULL) {
5804 /* RPCSEC_GSS verifier */
5805 error = nfs_gss_svc_verf_put(nd, &nmrep);
5806 if (error) {
5807 nfsm_chain_add_32(error, &nmrep, RPC_SYSTEM_ERR);
5808 goto done;
5809 }
5810 } else {
5811 /* RPCAUTH_NULL verifier */
5812 nfsm_chain_add_32(error, &nmrep, RPCAUTH_NULL);
5813 nfsm_chain_add_32(error, &nmrep, 0);
5814 }
5815 /* accepted status */
5816 switch (err) {
5817 case EPROGUNAVAIL:
5818 nfsm_chain_add_32(error, &nmrep, RPC_PROGUNAVAIL);
5819 break;
5820 case EPROGMISMATCH:
5821 nfsm_chain_add_32(error, &nmrep, RPC_PROGMISMATCH);
5822 /* XXX hard coded versions? */
5823 nfsm_chain_add_32(error, &nmrep, NFS_VER2);
5824 nfsm_chain_add_32(error, &nmrep, NFS_VER3);
5825 break;
5826 case EPROCUNAVAIL:
5827 nfsm_chain_add_32(error, &nmrep, RPC_PROCUNAVAIL);
5828 break;
5829 case EBADRPC:
5830 nfsm_chain_add_32(error, &nmrep, RPC_GARBAGE);
5831 break;
5832 default:
5833 nfsm_chain_add_32(error, &nmrep, RPC_SUCCESS);
5834 if (nd->nd_gss_context != NULL)
5835 error = nfs_gss_svc_prepare_reply(nd, &nmrep);
5836 if (err != NFSERR_RETVOID)
5837 nfsm_chain_add_32(error, &nmrep,
5838 (err ? nfsrv_errmap(nd, err) : 0));
5839 break;
5840 }
5841 }
5842
5843done:
5844 nfsm_chain_build_done(error, &nmrep);
5845 if (error) {
5846 /* error composing reply header */
5847 /* XXX should we keep statistics for these errors? */
5848 mbuf_freem(mrep);
5849 return (error);
5850 }
5851
5852 *nmrepp = nmrep;
5853 if ((err != 0) && (err != NFSERR_RETVOID))
5854 OSAddAtomic64(1, &nfsstats.srvrpc_errs);
5855 return (0);
5856}
5857
5858/*
5859 * The nfs server send routine.
5860 *
5861 * - return EINTR or ERESTART if interrupted by a signal
5862 * - return EPIPE if a connection is lost for connection based sockets (TCP...)
5863 * - do any cleanup required by recoverable socket errors (???)
5864 */
5865int
5866nfsrv_send(struct nfsrv_sock *slp, mbuf_t nam, mbuf_t top)
5867{
5868 int error;
5869 socket_t so = slp->ns_so;
5870 struct sockaddr *sendnam;
5871 struct msghdr msg;
5872
5873 bzero(&msg, sizeof(msg));
5874 if (nam && !sock_isconnected(so) && (slp->ns_sotype != SOCK_STREAM)) {
5875 if ((sendnam = mbuf_data(nam))) {
5876 msg.msg_name = (caddr_t)sendnam;
5877 msg.msg_namelen = sendnam->sa_len;
5878 }
5879 }
5880 error = sock_sendmbuf(so, &msg, top, 0, NULL);
5881 if (!error)
5882 return (0);
5883 log(LOG_INFO, "nfsd send error %d\n", error);
5884
5885 if ((error == EWOULDBLOCK) && (slp->ns_sotype == SOCK_STREAM))
5886 error = EPIPE; /* zap TCP sockets if they time out on send */
5887
5888 /* Handle any recoverable (soft) socket errors here. (???) */
5889 if (error != EINTR && error != ERESTART && error != EIO &&
5890 error != EWOULDBLOCK && error != EPIPE)
5891 error = 0;
5892
5893 return (error);
5894}
5895
5896/*
5897 * Socket upcall routine for the nfsd sockets.
5898 * The caddr_t arg is a pointer to the "struct nfsrv_sock".
5899 * Essentially do as much as possible non-blocking, else punt and it will
5900 * be called with MBUF_WAITOK from an nfsd.
5901 */
5902void
5903nfsrv_rcv(socket_t so, void *arg, int waitflag)
5904{
5905 struct nfsrv_sock *slp = arg;
5906
5907 if (!nfsd_thread_count || !(slp->ns_flag & SLP_VALID))
5908 return;
5909
5910 lck_rw_lock_exclusive(&slp->ns_rwlock);
5911 nfsrv_rcv_locked(so, slp, waitflag);
5912 /* Note: ns_rwlock gets dropped when called with MBUF_DONTWAIT */
5913}
5914void
5915nfsrv_rcv_locked(socket_t so, struct nfsrv_sock *slp, int waitflag)
5916{
5917 mbuf_t m, mp, mhck, m2;
5918 int ns_flag=0, error;
5919 struct msghdr msg;
5920 size_t bytes_read;
5921
5922 if ((slp->ns_flag & SLP_VALID) == 0) {
5923 if (waitflag == MBUF_DONTWAIT)
5924 lck_rw_done(&slp->ns_rwlock);
5925 return;
5926 }
5927
5928#ifdef notdef
5929 /*
5930 * Define this to test for nfsds handling this under heavy load.
5931 */
5932 if (waitflag == MBUF_DONTWAIT) {
5933 ns_flag = SLP_NEEDQ;
5934 goto dorecs;
5935 }
5936#endif
5937 if (slp->ns_sotype == SOCK_STREAM) {
5938 /*
5939 * If there are already records on the queue, defer soreceive()
5940 * to an(other) nfsd so that there is feedback to the TCP layer that
5941 * the nfs servers are heavily loaded.
5942 */
5943 if (slp->ns_rec) {
5944 ns_flag = SLP_NEEDQ;
5945 goto dorecs;
5946 }
5947
5948 /*
5949 * Do soreceive().
5950 */
5951 bytes_read = 1000000000;
5952 error = sock_receivembuf(so, NULL, &mp, MSG_DONTWAIT, &bytes_read);
5953 if (error || mp == NULL) {
5954 if (error == EWOULDBLOCK)
5955 ns_flag = (waitflag == MBUF_DONTWAIT) ? SLP_NEEDQ : 0;
5956 else
5957 ns_flag = SLP_DISCONN;
5958 goto dorecs;
5959 }
5960 m = mp;
5961 if (slp->ns_rawend) {
5962 if ((error = mbuf_setnext(slp->ns_rawend, m)))
5963 panic("nfsrv_rcv: mbuf_setnext failed %d\n", error);
5964 slp->ns_cc += bytes_read;
5965 } else {
5966 slp->ns_raw = m;
5967 slp->ns_cc = bytes_read;
5968 }
5969 while ((m2 = mbuf_next(m)))
5970 m = m2;
5971 slp->ns_rawend = m;
5972
5973 /*
5974 * Now try and parse record(s) out of the raw stream data.
5975 */
5976 error = nfsrv_getstream(slp, waitflag);
5977 if (error) {
5978 if (error == EPERM)
5979 ns_flag = SLP_DISCONN;
5980 else
5981 ns_flag = SLP_NEEDQ;
5982 }
5983 } else {
5984 struct sockaddr_storage nam;
5985
5986 if (slp->ns_reccnt >= nfsrv_sock_max_rec_queue_length) {
5987 /* already have max # RPC records queued on this socket */
5988 ns_flag = SLP_NEEDQ;
5989 goto dorecs;
5990 }
5991
5992 bzero(&msg, sizeof(msg));
5993 msg.msg_name = (caddr_t)&nam;
5994 msg.msg_namelen = sizeof(nam);
5995
5996 do {
5997 bytes_read = 1000000000;
5998 error = sock_receivembuf(so, &msg, &mp, MSG_DONTWAIT | MSG_NEEDSA, &bytes_read);
5999 if (mp) {
6000 if (msg.msg_name && (mbuf_get(MBUF_WAITOK, MBUF_TYPE_SONAME, &mhck) == 0)) {
6001 mbuf_setlen(mhck, nam.ss_len);
6002 bcopy(&nam, mbuf_data(mhck), nam.ss_len);
6003 m = mhck;
6004 if (mbuf_setnext(m, mp)) {
6005 /* trouble... just drop it */
6006 printf("nfsrv_rcv: mbuf_setnext failed\n");
6007 mbuf_free(mhck);
6008 m = mp;
6009 }
6010 } else {
6011 m = mp;
6012 }
6013 if (slp->ns_recend)
6014 mbuf_setnextpkt(slp->ns_recend, m);
6015 else {
6016 slp->ns_rec = m;
6017 slp->ns_flag |= SLP_DOREC;
6018 }
6019 slp->ns_recend = m;
6020 mbuf_setnextpkt(m, NULL);
6021 slp->ns_reccnt++;
6022 }
6023 } while (mp);
6024 }
6025
6026 /*
6027 * Now try and process the request records, non-blocking.
6028 */
6029dorecs:
6030 if (ns_flag)
6031 slp->ns_flag |= ns_flag;
6032 if (waitflag == MBUF_DONTWAIT) {
6033 int wake = (slp->ns_flag & SLP_WORKTODO);
6034 lck_rw_done(&slp->ns_rwlock);
6035 if (wake && nfsd_thread_count) {
6036 lck_mtx_lock(nfsd_mutex);
6037 nfsrv_wakenfsd(slp);
6038 lck_mtx_unlock(nfsd_mutex);
6039 }
6040 }
6041}
6042
6043/*
6044 * Try and extract an RPC request from the mbuf data list received on a
6045 * stream socket. The "waitflag" argument indicates whether or not it
6046 * can sleep.
6047 */
6048int
6049nfsrv_getstream(struct nfsrv_sock *slp, int waitflag)
6050{
6051 mbuf_t m;
6052 char *cp1, *cp2, *mdata;
6053 int len, mlen, error;
6054 mbuf_t om, m2, recm;
6055 u_int32_t recmark;
6056
6057 if (slp->ns_flag & SLP_GETSTREAM)
6058 panic("nfs getstream");
6059 slp->ns_flag |= SLP_GETSTREAM;
6060 for (;;) {
6061 if (slp->ns_reclen == 0) {
6062 if (slp->ns_cc < NFSX_UNSIGNED) {
6063 slp->ns_flag &= ~SLP_GETSTREAM;
6064 return (0);
6065 }
6066 m = slp->ns_raw;
6067 mdata = mbuf_data(m);
6068 mlen = mbuf_len(m);
6069 if (mlen >= NFSX_UNSIGNED) {
6070 bcopy(mdata, (caddr_t)&recmark, NFSX_UNSIGNED);
6071 mdata += NFSX_UNSIGNED;
6072 mlen -= NFSX_UNSIGNED;
6073 mbuf_setdata(m, mdata, mlen);
6074 } else {
6075 cp1 = (caddr_t)&recmark;
6076 cp2 = mdata;
6077 while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
6078 while (mlen == 0) {
6079 m = mbuf_next(m);
6080 cp2 = mbuf_data(m);
6081 mlen = mbuf_len(m);
6082 }
6083 *cp1++ = *cp2++;
6084 mlen--;
6085 mbuf_setdata(m, cp2, mlen);
6086 }
6087 }
6088 slp->ns_cc -= NFSX_UNSIGNED;
6089 recmark = ntohl(recmark);
6090 slp->ns_reclen = recmark & ~0x80000000;
6091 if (recmark & 0x80000000)
6092 slp->ns_flag |= SLP_LASTFRAG;
6093 else
6094 slp->ns_flag &= ~SLP_LASTFRAG;
6095 if (slp->ns_reclen <= 0 || slp->ns_reclen > NFS_MAXPACKET) {
6096 slp->ns_flag &= ~SLP_GETSTREAM;
6097 return (EPERM);
6098 }
6099 }
6100
6101 /*
6102 * Now get the record part.
6103 *
6104 * Note that slp->ns_reclen may be 0. Linux sometimes
6105 * generates 0-length RPCs
6106 */
6107 recm = NULL;
6108 if (slp->ns_cc == slp->ns_reclen) {
6109 recm = slp->ns_raw;
6110 slp->ns_raw = slp->ns_rawend = NULL;
6111 slp->ns_cc = slp->ns_reclen = 0;
6112 } else if (slp->ns_cc > slp->ns_reclen) {
6113 len = 0;
6114 m = slp->ns_raw;
6115 mlen = mbuf_len(m);
6116 mdata = mbuf_data(m);
6117 om = NULL;
6118 while (len < slp->ns_reclen) {
6119 if ((len + mlen) > slp->ns_reclen) {
6120 if (mbuf_copym(m, 0, slp->ns_reclen - len, waitflag, &m2)) {
6121 slp->ns_flag &= ~SLP_GETSTREAM;
6122 return (EWOULDBLOCK);
6123 }
6124 if (om) {
6125 if (mbuf_setnext(om, m2)) {
6126 /* trouble... just drop it */
6127 printf("nfsrv_getstream: mbuf_setnext failed\n");
6128 mbuf_freem(m2);
6129 slp->ns_flag &= ~SLP_GETSTREAM;
6130 return (EWOULDBLOCK);
6131 }
6132 recm = slp->ns_raw;
6133 } else {
6134 recm = m2;
6135 }
6136 mdata += slp->ns_reclen - len;
6137 mlen -= slp->ns_reclen - len;
6138 mbuf_setdata(m, mdata, mlen);
6139 len = slp->ns_reclen;
6140 } else if ((len + mlen) == slp->ns_reclen) {
6141 om = m;
6142 len += mlen;
6143 m = mbuf_next(m);
6144 recm = slp->ns_raw;
6145 if (mbuf_setnext(om, NULL)) {
6146 printf("nfsrv_getstream: mbuf_setnext failed 2\n");
6147 slp->ns_flag &= ~SLP_GETSTREAM;
6148 return (EWOULDBLOCK);
6149 }
6150 mlen = mbuf_len(m);
6151 mdata = mbuf_data(m);
6152 } else {
6153 om = m;
6154 len += mlen;
6155 m = mbuf_next(m);
6156 mlen = mbuf_len(m);
6157 mdata = mbuf_data(m);
6158 }
6159 }
6160 slp->ns_raw = m;
6161 slp->ns_cc -= len;
6162 slp->ns_reclen = 0;
6163 } else {
6164 slp->ns_flag &= ~SLP_GETSTREAM;
6165 return (0);
6166 }
6167
6168 /*
6169 * Accumulate the fragments into a record.
6170 */
6171 if (slp->ns_frag == NULL) {
6172 slp->ns_frag = recm;
6173 } else {
6174 m = slp->ns_frag;
6175 while ((m2 = mbuf_next(m)))
6176 m = m2;
6177 if ((error = mbuf_setnext(m, recm)))
6178 panic("nfsrv_getstream: mbuf_setnext failed 3, %d\n", error);
6179 }
6180 if (slp->ns_flag & SLP_LASTFRAG) {
6181 if (slp->ns_recend)
6182 mbuf_setnextpkt(slp->ns_recend, slp->ns_frag);
6183 else {
6184 slp->ns_rec = slp->ns_frag;
6185 slp->ns_flag |= SLP_DOREC;
6186 }
6187 slp->ns_recend = slp->ns_frag;
6188 slp->ns_frag = NULL;
6189 }
6190 }
6191}
6192
6193/*
6194 * Parse an RPC header.
6195 */
6196int
6197nfsrv_dorec(
6198 struct nfsrv_sock *slp,
6199 struct nfsd *nfsd,
6200 struct nfsrv_descript **ndp)
6201{
6202 mbuf_t m;
6203 mbuf_t nam;
6204 struct nfsrv_descript *nd;
6205 int error = 0;
6206
6207 *ndp = NULL;
6208 if (!(slp->ns_flag & (SLP_VALID|SLP_DOREC)) || (slp->ns_rec == NULL))
6209 return (ENOBUFS);
6210 MALLOC_ZONE(nd, struct nfsrv_descript *,
6211 sizeof (struct nfsrv_descript), M_NFSRVDESC, M_WAITOK);
6212 if (!nd)
6213 return (ENOMEM);
6214 m = slp->ns_rec;
6215 slp->ns_rec = mbuf_nextpkt(m);
6216 if (slp->ns_rec)
6217 mbuf_setnextpkt(m, NULL);
6218 else {
6219 slp->ns_flag &= ~SLP_DOREC;
6220 slp->ns_recend = NULL;
6221 }
6222 slp->ns_reccnt--;
6223 if (mbuf_type(m) == MBUF_TYPE_SONAME) {
6224 nam = m;
6225 m = mbuf_next(m);
6226 if ((error = mbuf_setnext(nam, NULL)))
6227 panic("nfsrv_dorec: mbuf_setnext failed %d\n", error);
6228 } else
6229 nam = NULL;
6230 nd->nd_nam2 = nam;
6231 nfsm_chain_dissect_init(error, &nd->nd_nmreq, m);
6232 if (!error)
6233 error = nfsrv_getreq(nd);
6234 if (error) {
6235 if (nam)
6236 mbuf_freem(nam);
6237 if (nd->nd_gss_context)
6238 nfs_gss_svc_ctx_deref(nd->nd_gss_context);
6239 FREE_ZONE(nd, sizeof(*nd), M_NFSRVDESC);
6240 return (error);
6241 }
6242 nd->nd_mrep = NULL;
6243 *ndp = nd;
6244 nfsd->nfsd_nd = nd;
6245 return (0);
6246}
6247
6248/*
6249 * Parse an RPC request
6250 * - verify it
6251 * - fill in the cred struct.
6252 */
6253int
6254nfsrv_getreq(struct nfsrv_descript *nd)
6255{
6256 struct nfsm_chain *nmreq;
6257 int len, i;
6258 u_int32_t nfsvers, auth_type;
6259 int error = 0;
6260 uid_t user_id;
6261 gid_t group_id;
6262 int ngroups;
6263 uint32_t val;
6264
6265 nd->nd_cr = NULL;
6266 nd->nd_gss_context = NULL;
6267 nd->nd_gss_seqnum = 0;
6268 nd->nd_gss_mb = NULL;
6269
6270 user_id = group_id = -2;
6271 val = auth_type = len = 0;
6272
6273 nmreq = &nd->nd_nmreq;
6274 nfsm_chain_get_32(error, nmreq, nd->nd_retxid); // XID
6275 nfsm_chain_get_32(error, nmreq, val); // RPC Call
6276 if (!error && (val != RPC_CALL))
6277 error = EBADRPC;
6278 nfsmout_if(error);
6279 nd->nd_repstat = 0;
6280 nfsm_chain_get_32(error, nmreq, val); // RPC Version
6281 nfsmout_if(error);
6282 if (val != RPC_VER2) {
6283 nd->nd_repstat = ERPCMISMATCH;
6284 nd->nd_procnum = NFSPROC_NOOP;
6285 return (0);
6286 }
6287 nfsm_chain_get_32(error, nmreq, val); // RPC Program Number
6288 nfsmout_if(error);
6289 if (val != NFS_PROG) {
6290 nd->nd_repstat = EPROGUNAVAIL;
6291 nd->nd_procnum = NFSPROC_NOOP;
6292 return (0);
6293 }
6294 nfsm_chain_get_32(error, nmreq, nfsvers);// NFS Version Number
6295 nfsmout_if(error);
6296 if ((nfsvers < NFS_VER2) || (nfsvers > NFS_VER3)) {
6297 nd->nd_repstat = EPROGMISMATCH;
6298 nd->nd_procnum = NFSPROC_NOOP;
6299 return (0);
6300 }
6301 nd->nd_vers = nfsvers;
6302 nfsm_chain_get_32(error, nmreq, nd->nd_procnum);// NFS Procedure Number
6303 nfsmout_if(error);
6304 if ((nd->nd_procnum >= NFS_NPROCS) ||
6305 ((nd->nd_vers == NFS_VER2) && (nd->nd_procnum > NFSV2PROC_STATFS))) {
6306 nd->nd_repstat = EPROCUNAVAIL;
6307 nd->nd_procnum = NFSPROC_NOOP;
6308 return (0);
6309 }
6310 if (nfsvers != NFS_VER3)
6311 nd->nd_procnum = nfsv3_procid[nd->nd_procnum];
6312 nfsm_chain_get_32(error, nmreq, auth_type); // Auth Flavor
6313 nfsm_chain_get_32(error, nmreq, len); // Auth Length
6314 if (!error && (len < 0 || len > RPCAUTH_MAXSIZ))
6315 error = EBADRPC;
6316 nfsmout_if(error);
6317
6318 /* Handle authentication */
6319 if (auth_type == RPCAUTH_SYS) {
6320 struct posix_cred temp_pcred;
6321 if (nd->nd_procnum == NFSPROC_NULL)
6322 return (0);
6323 nd->nd_sec = RPCAUTH_SYS;
6324 nfsm_chain_adv(error, nmreq, NFSX_UNSIGNED); // skip stamp
6325 nfsm_chain_get_32(error, nmreq, len); // hostname length
6326 if (len < 0 || len > NFS_MAXNAMLEN)
6327 error = EBADRPC;
6328 nfsm_chain_adv(error, nmreq, nfsm_rndup(len)); // skip hostname
6329 nfsmout_if(error);
6330
6331 /* create a temporary credential using the bits from the wire */
6332 bzero(&temp_pcred, sizeof(temp_pcred));
6333 nfsm_chain_get_32(error, nmreq, user_id);
6334 nfsm_chain_get_32(error, nmreq, group_id);
6335 temp_pcred.cr_groups[0] = group_id;
6336 nfsm_chain_get_32(error, nmreq, len); // extra GID count
6337 if ((len < 0) || (len > RPCAUTH_UNIXGIDS))
6338 error = EBADRPC;
6339 nfsmout_if(error);
6340 for (i = 1; i <= len; i++)
6341 if (i < NGROUPS)
6342 nfsm_chain_get_32(error, nmreq, temp_pcred.cr_groups[i]);
6343 else
6344 nfsm_chain_adv(error, nmreq, NFSX_UNSIGNED);
6345 nfsmout_if(error);
6346 ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
6347 if (ngroups > 1)
6348 nfsrv_group_sort(&temp_pcred.cr_groups[0], ngroups);
6349 nfsm_chain_adv(error, nmreq, NFSX_UNSIGNED); // verifier flavor (should be AUTH_NONE)
6350 nfsm_chain_get_32(error, nmreq, len); // verifier length
6351 if (len < 0 || len > RPCAUTH_MAXSIZ)
6352 error = EBADRPC;
6353 if (len > 0)
6354 nfsm_chain_adv(error, nmreq, nfsm_rndup(len));
6355
6356 /* request creation of a real credential */
6357 temp_pcred.cr_uid = user_id;
6358 temp_pcred.cr_ngroups = ngroups;
6359 nd->nd_cr = posix_cred_create(&temp_pcred);
6360 if (nd->nd_cr == NULL) {
6361 nd->nd_repstat = ENOMEM;
6362 nd->nd_procnum = NFSPROC_NOOP;
6363 return (0);
6364 }
6365 } else if (auth_type == RPCSEC_GSS) {
6366 error = nfs_gss_svc_cred_get(nd, nmreq);
6367 if (error) {
6368 if (error == EINVAL)
6369 goto nfsmout; // drop the request
6370 nd->nd_repstat = error;
6371 nd->nd_procnum = NFSPROC_NOOP;
6372 return (0);
6373 }
6374 } else {
6375 if (nd->nd_procnum == NFSPROC_NULL) // assume it's AUTH_NONE
6376 return (0);
6377 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
6378 nd->nd_procnum = NFSPROC_NOOP;
6379 return (0);
6380 }
6381 return (0);
6382nfsmout:
6383 if (IS_VALID_CRED(nd->nd_cr))
6384 kauth_cred_unref(&nd->nd_cr);
6385 nfsm_chain_cleanup(nmreq);
6386 return (error);
6387}
6388
6389/*
6390 * Search for a sleeping nfsd and wake it up.
6391 * SIDE EFFECT: If none found, make sure the socket is queued up so that one
6392 * of the running nfsds will go look for the work in the nfsrv_sockwait list.
6393 * Note: Must be called with nfsd_mutex held.
6394 */
6395void
6396nfsrv_wakenfsd(struct nfsrv_sock *slp)
6397{
6398 struct nfsd *nd;
6399
6400 if ((slp->ns_flag & SLP_VALID) == 0)
6401 return;
6402
6403 lck_rw_lock_exclusive(&slp->ns_rwlock);
6404 /* if there's work to do on this socket, make sure it's queued up */
6405 if ((slp->ns_flag & SLP_WORKTODO) && !(slp->ns_flag & SLP_QUEUED)) {
6406 TAILQ_INSERT_TAIL(&nfsrv_sockwait, slp, ns_svcq);
6407 slp->ns_flag |= SLP_WAITQ;
6408 }
6409 lck_rw_done(&slp->ns_rwlock);
6410
6411 /* wake up a waiting nfsd, if possible */
6412 nd = TAILQ_FIRST(&nfsd_queue);
6413 if (!nd)
6414 return;
6415
6416 TAILQ_REMOVE(&nfsd_queue, nd, nfsd_queue);
6417 nd->nfsd_flag &= ~NFSD_WAITING;
6418 wakeup(nd);
6419}
6420
6421#endif /* NFSSERVER */
6422