sys/fs/nfs/nfs_commonkrpc.c

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