nfs_commonkrpc.c revision 192616
180709Sjake/*- 280709Sjake * Copyright (c) 1989, 1991, 1993, 1995 380709Sjake * The Regents of the University of California. All rights reserved. 480709Sjake * 580709Sjake * This code is derived from software contributed to Berkeley by 680709Sjake * Rick Macklem at The University of Guelph. 780709Sjake * 880709Sjake * Redistribution and use in source and binary forms, with or without 980709Sjake * modification, are permitted provided that the following conditions 1080709Sjake * are met: 1180709Sjake * 1. Redistributions of source code must retain the above copyright 1280709Sjake * notice, this list of conditions and the following disclaimer. 1380709Sjake * 2. Redistributions in binary form must reproduce the above copyright 1481337Sobrien * notice, this list of conditions and the following disclaimer in the 1580709Sjake * documentation and/or other materials provided with the distribution. 1680709Sjake * 4. Neither the name of the University nor the names of its contributors 1781337Sobrien * may be used to endorse or promote products derived from this software 1880709Sjake * without specific prior written permission. 1980709Sjake * 2080709Sjake * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2180709Sjake * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2280709Sjake * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2380709Sjake * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2480709Sjake * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2580709Sjake * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2680709Sjake * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2780709Sjake * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2880709Sjake * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2980709Sjake * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3080709Sjake * SUCH DAMAGE. 3180709Sjake * 3280709Sjake */ 33131952Smarcel 3480709Sjake#include <sys/cdefs.h> 3580709Sjake__FBSDID("$FreeBSD: head/sys/fs/nfs/nfs_commonkrpc.c 192616 2009-05-23 00:40:17Z rmacklem $"); 3680709Sjake 3780709Sjake/* 3880709Sjake * Socket operations for use by nfs 3980709Sjake */ 4080709Sjake 4180709Sjake#include "opt_inet6.h" 4280709Sjake#include "opt_kgssapi.h" 4380709Sjake#include "opt_nfs.h" 4480709Sjake 4580709Sjake#include <sys/param.h> 4680709Sjake#include <sys/systm.h> 4780709Sjake#include <sys/kernel.h> 4880709Sjake#include <sys/limits.h> 4980709Sjake#include <sys/lock.h> 5080709Sjake#include <sys/malloc.h> 5180709Sjake#include <sys/mbuf.h> 5292205Sjake#include <sys/mount.h> 5384181Sjake#include <sys/mutex.h> 5480709Sjake#include <sys/proc.h> 5580709Sjake#include <sys/signalvar.h> 5680709Sjake#include <sys/syscallsubr.h> 5780709Sjake#include <sys/sysctl.h> 58131952Smarcel#include <sys/syslog.h> 59131952Smarcel#include <sys/vnode.h> 6080709Sjake 61131952Smarcel#include <rpc/rpc.h> 6280709Sjake 63131952Smarcel#include <kgssapi/krb5/kcrypto.h> 64131952Smarcel 65131952Smarcel#include <fs/nfs/nfsport.h> 66131952Smarcel 67131952SmarcelNFSSTATESPINLOCK; 68131952SmarcelNFSREQSPINLOCK; 69131952Smarcelextern struct nfsstats newnfsstats; 70131952Smarcelextern struct nfsreqhead nfsd_reqq; 71131952Smarcelextern int nfscl_ticks; 7280709Sjakeextern void (*ncl_call_invalcaches)(struct vnode *); 7380709Sjake 74131952Smarcelstatic int nfsrv_gsscallbackson = 0; 7580709Sjakestatic int nfs_bufpackets = 4; 7680709Sjakestatic int nfs_reconnects; 77131952Smarcelstatic int nfs3_jukebox_delay = 10; 78131952Smarcelstatic int nfs_skip_wcc_data_onerr = 1; 7980709Sjakestatic int nfs_keytab_enctype = ETYPE_DES_CBC_CRC; 80131952Smarcel 8180709SjakeSYSCTL_DECL(_vfs_newnfs); 82131952Smarcel 83131952SmarcelSYSCTL_INT(_vfs_newnfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, 84131952Smarcel "Buffer reservation size 2 < x < 64"); 85131952SmarcelSYSCTL_INT(_vfs_newnfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0, 86131952Smarcel "Number of times the nfs client has had to reconnect"); 87131952SmarcelSYSCTL_INT(_vfs_newnfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0, 88131952Smarcel "Number of seconds to delay a retry after receiving EJUKEBOX"); 89131952SmarcelSYSCTL_INT(_vfs_newnfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0, 90131952Smarcel "Disable weak cache consistency checking when server returns an error"); 9180709SjakeSYSCTL_INT(_vfs_newnfs, OID_AUTO, keytab_enctype, CTLFLAG_RW, &nfs_keytab_enctype, 0, 9280709Sjake "Encryption type for the keytab entry used by nfs"); 9387702Sjhb 9487702Sjhbstatic void nfs_down(struct nfsmount *, struct thread *, const char *, 9587702Sjhb int, int); 9687702Sjhbstatic void nfs_up(struct nfsmount *, struct thread *, const char *, 9787702Sjhb int, int); 9880709Sjakestatic int nfs_msg(struct thread *, const char *, const char *, int); 9980709Sjake 10080709Sjakeextern int nfsv2_procid[]; 10180709Sjake 10286147Stmmstruct nfs_cached_auth { 10386147Stmm int ca_refs; /* refcount, including 1 from the cache */ 10486147Stmm uid_t ca_uid; /* uid that corresponds to this auth */ 10586147Stmm AUTH *ca_auth; /* RPC auth handle */ 10686147Stmm}; 107 108/* 109 * Initialize sockets and congestion for a new NFS connection. 110 * We do not free the sockaddr if error. 111 */ 112int 113newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp, 114 struct ucred *cred, NFSPROC_T *p, int callback_retry_mult) 115{ 116 int rcvreserve, sndreserve; 117 int pktscale; 118 struct sockaddr *saddr; 119 struct ucred *origcred; 120 CLIENT *client; 121 struct netconfig *nconf; 122 struct socket *so; 123 int one = 1, retries, error, printsbmax = 0; 124 struct thread *td = curthread; 125 126 /* 127 * We need to establish the socket using the credentials of 128 * the mountpoint. Some parts of this process (such as 129 * sobind() and soconnect()) will use the curent thread's 130 * credential instead of the socket credential. To work 131 * around this, temporarily change the current thread's 132 * credential to that of the mountpoint. 133 * 134 * XXX: It would be better to explicitly pass the correct 135 * credential to sobind() and soconnect(). 136 */ 137 origcred = td->td_ucred; 138 139 /* 140 * Use the credential in nr_cred, if not NULL. 141 */ 142 if (nrp->nr_cred != NULL) 143 td->td_ucred = nrp->nr_cred; 144 else 145 td->td_ucred = cred; 146 saddr = nrp->nr_nam; 147 148 if (saddr->sa_family == AF_INET) 149 if (nrp->nr_sotype == SOCK_DGRAM) 150 nconf = getnetconfigent("udp"); 151 else 152 nconf = getnetconfigent("tcp"); 153 else 154 if (nrp->nr_sotype == SOCK_DGRAM) 155 nconf = getnetconfigent("udp6"); 156 else 157 nconf = getnetconfigent("tcp6"); 158 159 pktscale = nfs_bufpackets; 160 if (pktscale < 2) 161 pktscale = 2; 162 if (pktscale > 64) 163 pktscale = 64; 164 /* 165 * soreserve() can fail if sb_max is too small, so shrink pktscale 166 * and try again if there is an error. 167 * Print a log message suggesting increasing sb_max. 168 * Creating a socket and doing this is necessary since, if the 169 * reservation sizes are too large and will make soreserve() fail, 170 * the connection will work until a large send is attempted and 171 * then it will loop in the krpc code. 172 */ 173 so = NULL; 174 saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *); 175 error = socreate(saddr->sa_family, &so, nrp->nr_sotype, 176 nrp->nr_soproto, td->td_ucred, td); 177 if (error) { 178 td->td_ucred = origcred; 179 return (error); 180 } 181 do { 182 if (error != 0 && pktscale > 2) { 183 pktscale--; 184 if (printsbmax == 0) { 185 printf("nfscl: consider increasing kern.ipc.maxsockbuf\n"); 186 printsbmax = 1; 187 } 188 } 189 if (nrp->nr_sotype == SOCK_DGRAM) { 190 if (nmp != NULL) { 191 sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) * 192 pktscale; 193 rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) * 194 pktscale; 195 } else { 196 sndreserve = rcvreserve = 1024 * pktscale; 197 } 198 } else { 199 if (nrp->nr_sotype != SOCK_STREAM) 200 panic("nfscon sotype"); 201 if (nmp != NULL) { 202 sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR + 203 sizeof (u_int32_t)) * pktscale; 204 rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR + 205 sizeof (u_int32_t)) * pktscale; 206 } else { 207 sndreserve = rcvreserve = 1024 * pktscale; 208 } 209 } 210 error = soreserve(so, sndreserve, rcvreserve); 211 } while (error != 0 && pktscale > 2); 212 soclose(so); 213 if (error) { 214 td->td_ucred = origcred; 215 return (error); 216 } 217 218 client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog, 219 nrp->nr_vers, sndreserve, rcvreserve); 220 CLNT_CONTROL(client, CLSET_WAITCHAN, "newnfsreq"); 221 if (nmp != NULL) { 222 if ((nmp->nm_flag & NFSMNT_INT)) 223 CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one); 224 if ((nmp->nm_flag & NFSMNT_RESVPORT)) 225 CLNT_CONTROL(client, CLSET_PRIVPORT, &one); 226 if (NFSHASSOFT(nmp)) 227 retries = nmp->nm_retry; 228 else 229 retries = INT_MAX; 230 } else { 231 /* 232 * Three cases: 233 * - Null RPC callback to client 234 * - Non-Null RPC callback to client, wait a little longer 235 * - upcalls to nfsuserd and gssd (clp == NULL) 236 */ 237 if (callback_retry_mult == 0) { 238 retries = NFSV4_UPCALLRETRY; 239 CLNT_CONTROL(client, CLSET_PRIVPORT, &one); 240 } else { 241 retries = NFSV4_CALLBACKRETRY * callback_retry_mult; 242 } 243 } 244 CLNT_CONTROL(client, CLSET_RETRIES, &retries); 245 246 mtx_lock(&nrp->nr_mtx); 247 if (nrp->nr_client != NULL) { 248 /* 249 * Someone else already connected. 250 */ 251 CLNT_RELEASE(client); 252 } else { 253 nrp->nr_client = client; 254 } 255 256 /* 257 * Protocols that do not require connections may be optionally left 258 * unconnected for servers that reply from a port other than NFS_PORT. 259 */ 260 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) { 261 mtx_unlock(&nrp->nr_mtx); 262 CLNT_CONTROL(client, CLSET_CONNECT, &one); 263 } else { 264 mtx_unlock(&nrp->nr_mtx); 265 } 266 267 /* Restore current thread's credentials. */ 268 td->td_ucred = origcred; 269 return (0); 270} 271 272/* 273 * NFS disconnect. Clean up and unlink. 274 */ 275void 276newnfs_disconnect(struct nfssockreq *nrp) 277{ 278 CLIENT *client; 279 280 mtx_lock(&nrp->nr_mtx); 281 if (nrp->nr_client != NULL) { 282 client = nrp->nr_client; 283 nrp->nr_client = NULL; 284 mtx_unlock(&nrp->nr_mtx); 285#ifdef KGSSAPI 286 rpc_gss_secpurge(client); 287#endif 288 CLNT_CLOSE(client); 289 CLNT_RELEASE(client); 290 } else { 291 mtx_unlock(&nrp->nr_mtx); 292 } 293} 294 295static AUTH * 296nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal, 297 char *srv_principal, gss_OID mech_oid, struct ucred *cred) 298{ 299#ifdef KGSSAPI 300 rpc_gss_service_t svc; 301 AUTH *auth; 302#ifdef notyet 303 rpc_gss_options_req_t req_options; 304#endif 305#endif 306 307 switch (secflavour) { 308#ifdef KGSSAPI 309 case RPCSEC_GSS_KRB5: 310 case RPCSEC_GSS_KRB5I: 311 case RPCSEC_GSS_KRB5P: 312 if (!mech_oid) { 313 if (!rpc_gss_mech_to_oid("kerberosv5", &mech_oid)) 314 return (NULL); 315 } 316 if (secflavour == RPCSEC_GSS_KRB5) 317 svc = rpc_gss_svc_none; 318 else if (secflavour == RPCSEC_GSS_KRB5I) 319 svc = rpc_gss_svc_integrity; 320 else 321 svc = rpc_gss_svc_privacy; 322#ifdef notyet 323 req_options.req_flags = GSS_C_MUTUAL_FLAG; 324 req_options.time_req = 0; 325 req_options.my_cred = GSS_C_NO_CREDENTIAL; 326 req_options.input_channel_bindings = NULL; 327 req_options.enc_type = nfs_keytab_enctype; 328 329 auth = rpc_gss_secfind(nrp->nr_client, cred, 330 clnt_principal, srv_principal, mech_oid, svc, 331 &req_options); 332#else 333 /* 334 * Until changes to the rpcsec_gss code are committed, 335 * there is no support for host based initiator 336 * principals. As such, that case cannot yet be handled. 337 */ 338 if (clnt_principal == NULL) 339 auth = rpc_gss_secfind(nrp->nr_client, cred, 340 srv_principal, mech_oid, svc); 341 else 342 auth = NULL; 343#endif 344 return (auth); 345#endif /* KGSSAPI */ 346 case AUTH_SYS: 347 default: 348 return (authunix_create(cred)); 349 350 } 351} 352 353/* 354 * Callback from the RPC code to generate up/down notifications. 355 */ 356 357struct nfs_feedback_arg { 358 struct nfsmount *nf_mount; 359 int nf_lastmsg; /* last tprintf */ 360 int nf_tprintfmsg; 361 struct thread *nf_td; 362}; 363 364static void 365nfs_feedback(int type, int proc, void *arg) 366{ 367 struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg; 368 struct nfsmount *nmp = nf->nf_mount; 369 struct timeval now; 370 371 getmicrouptime(&now); 372 373 switch (type) { 374 case FEEDBACK_REXMIT2: 375 case FEEDBACK_RECONNECT: 376 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) { 377 nfs_down(nmp, nf->nf_td, 378 "not responding", 0, NFSSTA_TIMEO); 379 nf->nf_tprintfmsg = TRUE; 380 nf->nf_lastmsg = now.tv_sec; 381 } 382 break; 383 384 case FEEDBACK_OK: 385 nfs_up(nf->nf_mount, nf->nf_td, 386 "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg); 387 break; 388 } 389} 390 391/* 392 * newnfs_request - goes something like this 393 * - does the rpc by calling the krpc layer 394 * - break down rpc header and return with nfs reply 395 * nb: always frees up nd_mreq mbuf list 396 */ 397int 398newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp, 399 struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp, 400 struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers, 401 u_char *retsum, int toplevel, u_int64_t *xidp) 402{ 403 u_int32_t *tl; 404 time_t waituntil; 405 int i, j; 406 int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS; 407 u_int16_t procnum; 408 u_int trylater_delay = 1; 409 struct nfs_feedback_arg nf; 410 struct timeval timo, now; 411 AUTH *auth; 412 struct rpc_callextra ext; 413 enum clnt_stat stat; 414 struct nfsreq *rep = NULL; 415 char *srv_principal = NULL; 416 417 if (xidp != NULL) 418 *xidp = 0; 419 /* Reject requests while attempting a forced unmount. */ 420 if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) { 421 m_freem(nd->nd_mreq); 422 return (ESTALE); 423 } 424 425 /* 426 * For a client side mount, nmp is != NULL and clp == NULL. For 427 * server calls (callbacks or upcalls), nmp == NULL. 428 */ 429 if (clp != NULL) { 430 NFSLOCKSTATE(); 431 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) { 432 secflavour = RPCSEC_GSS_KRB5; 433 if (nd->nd_procnum != NFSPROC_NULL) { 434 if (clp->lc_flags & LCL_GSSINTEGRITY) 435 secflavour = RPCSEC_GSS_KRB5I; 436 else if (clp->lc_flags & LCL_GSSPRIVACY) 437 secflavour = RPCSEC_GSS_KRB5P; 438 } 439 } 440 NFSUNLOCKSTATE(); 441 } else if (nmp != NULL && NFSHASKERB(nmp) && 442 nd->nd_procnum != NFSPROC_NULL) { 443 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0) 444 nd->nd_flag |= ND_USEGSSNAME; 445 if ((nd->nd_flag & ND_USEGSSNAME) && nmp->nm_krbnamelen > 0) 446 usegssname = 1; 447 if (NFSHASINTEGRITY(nmp)) 448 secflavour = RPCSEC_GSS_KRB5I; 449 else if (NFSHASPRIVACY(nmp)) 450 secflavour = RPCSEC_GSS_KRB5P; 451 else 452 secflavour = RPCSEC_GSS_KRB5; 453 srv_principal = NFSMNT_SRVKRBNAME(nmp); 454 } 455 456 if (nmp != NULL) { 457 bzero(&nf, sizeof(struct nfs_feedback_arg)); 458 nf.nf_mount = nmp; 459 nf.nf_td = td; 460 getmicrouptime(&now); 461 nf.nf_lastmsg = now.tv_sec - 462 ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay)); 463 } 464 465 /* 466 * XXX if not already connected call nfs_connect now. Longer 467 * term, change nfs_mount to call nfs_connect unconditionally 468 * and let clnt_reconnect_create handle reconnects. 469 */ 470 if (nrp->nr_client == NULL) 471 newnfs_connect(nmp, nrp, cred, td, 0); 472 473 if (nd->nd_procnum == NFSPROC_NULL) 474 auth = authnone_create(); 475 else if (usegssname) 476 auth = nfs_getauth(nrp, secflavour, nmp->nm_krbname, 477 srv_principal, NULL, cred); 478 else 479 auth = nfs_getauth(nrp, secflavour, NULL, 480 srv_principal, NULL, cred); 481 if (auth == NULL) { 482 m_freem(nd->nd_mreq); 483 return (EACCES); 484 } 485 bzero(&ext, sizeof(ext)); 486 ext.rc_auth = auth; 487 if (nmp != NULL) { 488 ext.rc_feedback = nfs_feedback; 489 ext.rc_feedback_arg = &nf; 490 } 491 492 procnum = nd->nd_procnum; 493 if ((nd->nd_flag & ND_NFSV4) && 494 nd->nd_procnum != NFSPROC_NULL && 495 nd->nd_procnum != NFSV4PROC_CBCOMPOUND) 496 procnum = NFSV4PROC_COMPOUND; 497 498 if (nmp != NULL) { 499 NFSINCRGLOBAL(newnfsstats.rpcrequests); 500 /* 501 * Now only used for the R_DONTRECOVER case, but until that is 502 * supported within the krpc code, I need to keep a queue of 503 * outstanding RPCs for nfsv4 client requests. 504 */ 505 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND) 506 MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), 507 M_NFSDREQ, M_WAITOK); 508 } 509 trycnt = 0; 510tryagain: 511 if (nmp == NULL) { 512 timo.tv_usec = 0; 513 if (clp == NULL) 514 timo.tv_sec = NFSV4_UPCALLTIMEO; 515 else 516 timo.tv_sec = NFSV4_CALLBACKTIMEO; 517 } else { 518 if (nrp->nr_sotype != SOCK_DGRAM) { 519 timo.tv_usec = 0; 520 if ((nmp->nm_flag & NFSMNT_NFSV4)) 521 timo.tv_sec = INT_MAX; 522 else 523 timo.tv_sec = NFS_TCPTIMEO; 524 } else { 525 timo.tv_sec = nmp->nm_timeo / NFS_HZ; 526 timo.tv_usec = (nmp->nm_timeo * 1000000) / NFS_HZ; 527 } 528 529 if (rep != NULL) { 530 rep->r_flags = 0; 531 rep->r_nmp = nmp; 532 /* 533 * Chain request into list of outstanding requests. 534 */ 535 NFSLOCKREQ(); 536 TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain); 537 NFSUNLOCKREQ(); 538 } 539 } 540 541 nd->nd_mrep = NULL; 542 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq, 543 &nd->nd_mrep, timo); 544 545 if (rep != NULL) { 546 /* 547 * RPC done, unlink the request. 548 */ 549 NFSLOCKREQ(); 550 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain); 551 NFSUNLOCKREQ(); 552 } 553 554 /* 555 * If there was a successful reply and a tprintf msg. 556 * tprintf a response. 557 */ 558 if (stat == RPC_SUCCESS) { 559 error = 0; 560 } else if (stat == RPC_TIMEDOUT) { 561 error = ETIMEDOUT; 562 } else if (stat == RPC_VERSMISMATCH) { 563 error = EOPNOTSUPP; 564 } else if (stat == RPC_PROGVERSMISMATCH) { 565 error = EPROTONOSUPPORT; 566 } else { 567 error = EACCES; 568 } 569 if (error) { 570 m_freem(nd->nd_mreq); 571 AUTH_DESTROY(auth); 572 if (rep != NULL) 573 FREE((caddr_t)rep, M_NFSDREQ); 574 return (error); 575 } 576 577 KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n")); 578 579 nd->nd_md = nd->nd_mrep; 580 nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t); 581 nd->nd_repstat = 0; 582 if (nd->nd_procnum != NFSPROC_NULL) { 583 /* 584 * and now the actual NFS xdr. 585 */ 586 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 587 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl); 588 if (nd->nd_repstat != 0) { 589 if ((nd->nd_repstat == NFSERR_DELAY && 590 (nd->nd_flag & ND_NFSV4) && 591 nd->nd_procnum != NFSPROC_SETATTR && 592 nd->nd_procnum != NFSPROC_READ && 593 nd->nd_procnum != NFSPROC_WRITE && 594 nd->nd_procnum != NFSPROC_OPEN && 595 nd->nd_procnum != NFSPROC_CREATE && 596 nd->nd_procnum != NFSPROC_OPENCONFIRM && 597 nd->nd_procnum != NFSPROC_OPENDOWNGRADE && 598 nd->nd_procnum != NFSPROC_CLOSE && 599 nd->nd_procnum != NFSPROC_LOCK && 600 nd->nd_procnum != NFSPROC_LOCKU) || 601 (nd->nd_repstat == NFSERR_DELAY && 602 (nd->nd_flag & ND_NFSV4) == 0) || 603 nd->nd_repstat == NFSERR_RESOURCE) { 604 if (trylater_delay > NFS_TRYLATERDEL) 605 trylater_delay = NFS_TRYLATERDEL; 606 waituntil = NFSD_MONOSEC + trylater_delay; 607 while (NFSD_MONOSEC < waituntil) 608 (void) nfs_catnap(PZERO, "nfstry"); 609 trylater_delay *= 2; 610 goto tryagain; 611 } 612 613 /* 614 * If the File Handle was stale, invalidate the 615 * lookup cache, just in case. 616 * (vp != NULL implies a client side call) 617 */ 618 if (nd->nd_repstat == ESTALE && vp != NULL) { 619 cache_purge(vp); 620 if (ncl_call_invalcaches != NULL) 621 (*ncl_call_invalcaches)(vp); 622 } 623 } 624 625 /* 626 * Get rid of the tag, return count, and PUTFH result for V4. 627 */ 628 if (nd->nd_flag & ND_NFSV4) { 629 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 630 i = fxdr_unsigned(int, *tl); 631 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 632 if (error) 633 goto nfsmout; 634 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 635 i = fxdr_unsigned(int, *++tl); 636 637 /* 638 * If the first op's status is non-zero, mark that 639 * there is no more data to process. 640 */ 641 if (*++tl) 642 nd->nd_flag |= ND_NOMOREDATA; 643 644 /* 645 * If the first op is Putfh, throw its results away 646 * and toss the op# and status for the first op. 647 */ 648 if (nmp != NULL && i == NFSV4OP_PUTFH && *tl == 0) { 649 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED); 650 i = fxdr_unsigned(int, *tl++); 651 j = fxdr_unsigned(int, *tl); 652 /* 653 * All Compounds that do an Op that must 654 * be in sequence consist of NFSV4OP_PUTFH 655 * followed by one of these. As such, we 656 * can determine if the seqid# should be 657 * incremented, here. 658 */ 659 if ((i == NFSV4OP_OPEN || 660 i == NFSV4OP_OPENCONFIRM || 661 i == NFSV4OP_OPENDOWNGRADE || 662 i == NFSV4OP_CLOSE || 663 i == NFSV4OP_LOCK || 664 i == NFSV4OP_LOCKU) && 665 (j == 0 || 666 (j != NFSERR_STALECLIENTID && 667 j != NFSERR_STALESTATEID && 668 j != NFSERR_BADSTATEID && 669 j != NFSERR_BADSEQID && 670 j != NFSERR_BADXDR && 671 j != NFSERR_RESOURCE && 672 j != NFSERR_NOFILEHANDLE))) 673 nd->nd_flag |= ND_INCRSEQID; 674 /* 675 * If the first op's status is non-zero, mark 676 * that there is no more data to process. 677 */ 678 if (j) 679 nd->nd_flag |= ND_NOMOREDATA; 680 } 681 682 /* 683 * If R_DONTRECOVER is set, replace the stale error 684 * reply, so that recovery isn't initiated. 685 */ 686 if ((nd->nd_repstat == NFSERR_STALECLIENTID || 687 nd->nd_repstat == NFSERR_STALESTATEID) && 688 rep != NULL && (rep->r_flags & R_DONTRECOVER)) 689 nd->nd_repstat = NFSERR_STALEDONTRECOVER; 690 } 691 } 692 693 m_freem(nd->nd_mreq); 694 AUTH_DESTROY(auth); 695 if (rep != NULL) 696 FREE((caddr_t)rep, M_NFSDREQ); 697 return (0); 698nfsmout: 699 mbuf_freem(nd->nd_mrep); 700 mbuf_freem(nd->nd_mreq); 701 AUTH_DESTROY(auth); 702 if (rep != NULL) 703 FREE((caddr_t)rep, M_NFSDREQ); 704 return (error); 705} 706 707/* 708 * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and 709 * wait for all requests to complete. This is used by forced unmounts 710 * to terminate any outstanding RPCs. 711 */ 712int 713newnfs_nmcancelreqs(struct nfsmount *nmp) 714{ 715 716 if (nmp->nm_sockreq.nr_client != NULL) 717 CLNT_CLOSE(nmp->nm_sockreq.nr_client); 718 return (0); 719} 720 721/* 722 * Any signal that can interrupt an NFS operation in an intr mount 723 * should be added to this set. SIGSTOP and SIGKILL cannot be masked. 724 */ 725int newnfs_sig_set[] = { 726 SIGINT, 727 SIGTERM, 728 SIGHUP, 729 SIGKILL, 730 SIGSTOP, 731 SIGQUIT 732}; 733 734/* 735 * Check to see if one of the signals in our subset is pending on 736 * the process (in an intr mount). 737 */ 738static int 739nfs_sig_pending(sigset_t set) 740{ 741 int i; 742 743 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) 744 if (SIGISMEMBER(set, newnfs_sig_set[i])) 745 return (1); 746 return (0); 747} 748 749/* 750 * The set/restore sigmask functions are used to (temporarily) overwrite 751 * the process p_sigmask during an RPC call (for example). These are also 752 * used in other places in the NFS client that might tsleep(). 753 */ 754void 755newnfs_set_sigmask(struct thread *td, sigset_t *oldset) 756{ 757 sigset_t newset; 758 int i; 759 struct proc *p; 760 761 SIGFILLSET(newset); 762 if (td == NULL) 763 td = curthread; /* XXX */ 764 p = td->td_proc; 765 /* Remove the NFS set of signals from newset */ 766 PROC_LOCK(p); 767 mtx_lock(&p->p_sigacts->ps_mtx); 768 for (i = 0 ; i < sizeof(newnfs_sig_set)/sizeof(int) ; i++) { 769 /* 770 * But make sure we leave the ones already masked 771 * by the process, ie. remove the signal from the 772 * temporary signalmask only if it wasn't already 773 * in p_sigmask. 774 */ 775 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) && 776 !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i])) 777 SIGDELSET(newset, newnfs_sig_set[i]); 778 } 779 mtx_unlock(&p->p_sigacts->ps_mtx); 780 PROC_UNLOCK(p); 781 kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0); 782} 783 784void 785newnfs_restore_sigmask(struct thread *td, sigset_t *set) 786{ 787 if (td == NULL) 788 td = curthread; /* XXX */ 789 kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0); 790} 791 792/* 793 * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the 794 * old one after msleep() returns. 795 */ 796int 797newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo) 798{ 799 sigset_t oldset; 800 int error; 801 struct proc *p; 802 803 if ((priority & PCATCH) == 0) 804 return msleep(ident, mtx, priority, wmesg, timo); 805 if (td == NULL) 806 td = curthread; /* XXX */ 807 newnfs_set_sigmask(td, &oldset); 808 error = msleep(ident, mtx, priority, wmesg, timo); 809 newnfs_restore_sigmask(td, &oldset); 810 p = td->td_proc; 811 return (error); 812} 813 814/* 815 * Test for a termination condition pending on the process. 816 * This is used for NFSMNT_INT mounts. 817 */ 818int 819newnfs_sigintr(struct nfsmount *nmp, struct thread *td) 820{ 821 struct proc *p; 822 sigset_t tmpset; 823 824 /* Terminate all requests while attempting a forced unmount. */ 825 if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) 826 return (EIO); 827 if (!(nmp->nm_flag & NFSMNT_INT)) 828 return (0); 829 if (td == NULL) 830 return (0); 831 p = td->td_proc; 832 PROC_LOCK(p); 833 tmpset = p->p_siglist; 834 SIGSETOR(tmpset, td->td_siglist); 835 SIGSETNAND(tmpset, td->td_sigmask); 836 mtx_lock(&p->p_sigacts->ps_mtx); 837 SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore); 838 mtx_unlock(&p->p_sigacts->ps_mtx); 839 if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist)) 840 && nfs_sig_pending(tmpset)) { 841 PROC_UNLOCK(p); 842 return (EINTR); 843 } 844 PROC_UNLOCK(p); 845 return (0); 846} 847 848static int 849nfs_msg(struct thread *td, const char *server, const char *msg, int error) 850{ 851 struct proc *p; 852 853 p = td ? td->td_proc : NULL; 854 if (error) { 855 tprintf(p, LOG_INFO, "newnfs server %s: %s, error %d\n", 856 server, msg, error); 857 } else { 858 tprintf(p, LOG_INFO, "newnfs server %s: %s\n", server, msg); 859 } 860 return (0); 861} 862 863static void 864nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg, 865 int error, int flags) 866{ 867 if (nmp == NULL) 868 return; 869 mtx_lock(&nmp->nm_mtx); 870 if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) { 871 nmp->nm_state |= NFSSTA_TIMEO; 872 mtx_unlock(&nmp->nm_mtx); 873 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, 874 VQ_NOTRESP, 0); 875 } else 876 mtx_unlock(&nmp->nm_mtx); 877 mtx_lock(&nmp->nm_mtx); 878 if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) { 879 nmp->nm_state |= NFSSTA_LOCKTIMEO; 880 mtx_unlock(&nmp->nm_mtx); 881 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, 882 VQ_NOTRESPLOCK, 0); 883 } else 884 mtx_unlock(&nmp->nm_mtx); 885 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error); 886} 887 888static void 889nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg, 890 int flags, int tprintfmsg) 891{ 892 if (nmp == NULL) 893 return; 894 if (tprintfmsg) { 895 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0); 896 } 897 898 mtx_lock(&nmp->nm_mtx); 899 if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) { 900 nmp->nm_state &= ~NFSSTA_TIMEO; 901 mtx_unlock(&nmp->nm_mtx); 902 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, 903 VQ_NOTRESP, 1); 904 } else 905 mtx_unlock(&nmp->nm_mtx); 906 907 mtx_lock(&nmp->nm_mtx); 908 if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) { 909 nmp->nm_state &= ~NFSSTA_LOCKTIMEO; 910 mtx_unlock(&nmp->nm_mtx); 911 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, 912 VQ_NOTRESPLOCK, 1); 913 } else 914 mtx_unlock(&nmp->nm_mtx); 915} 916 917