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 extern int nfsrv_lease;
 102
 103 SVCPOOL         *nfscbd_pool;
 104 static int      nfsrv_gsscallbackson = 0;
 105 static int      nfs_bufpackets = 4;
 106 static int      nfs_reconnects;
 107 static int      nfs3_jukebox_delay = 10;
 108 static int      nfs_skip_wcc_data_onerr = 1;
 109 static int      nfs_dsretries = 2;
 110
 111 SYSCTL_DECL(_vfs_nfs);
 112
 113 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
 114     "Buffer reservation size 2 < x < 64");
 115 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
 116     "Number of times the nfs client has had to reconnect");
 117 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
 118     "Number of seconds to delay a retry after receiving EJUKEBOX");
 119 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
 120     "Disable weak cache consistency checking when server returns an error");
 121 SYSCTL_INT(_vfs_nfs, OID_AUTO, dsretries, CTLFLAG_RW, &nfs_dsretries, 0,
 122     "Number of retries for a DS RPC before failure");
 123
 124 static void     nfs_down(struct nfsmount *, struct thread *, const char *,
 125     int, int);
 126 static void     nfs_up(struct nfsmount *, struct thread *, const char *,
 127     int, int);
 128 static int      nfs_msg(struct thread *, const char *, const char *, int);
 129
 130 struct nfs_cached_auth {
 131         int             ca_refs; /* refcount, including 1 from the cache */
 132         uid_t           ca_uid;  /* uid that corresponds to this auth */
 133         AUTH            *ca_auth; /* RPC auth handle */
 134 };
 135
 136 static int nfsv2_procid[NFS_V3NPROCS] = {
 137         NFSV2PROC_NULL,
 138         NFSV2PROC_GETATTR,
 139         NFSV2PROC_SETATTR,
 140         NFSV2PROC_LOOKUP,
 141         NFSV2PROC_NOOP,
 142         NFSV2PROC_READLINK,
 143         NFSV2PROC_READ,
 144         NFSV2PROC_WRITE,
 145         NFSV2PROC_CREATE,
 146         NFSV2PROC_MKDIR,
 147         NFSV2PROC_SYMLINK,
 148         NFSV2PROC_CREATE,
 149         NFSV2PROC_REMOVE,
 150         NFSV2PROC_RMDIR,
 151         NFSV2PROC_RENAME,
 152         NFSV2PROC_LINK,
 153         NFSV2PROC_READDIR,
 154         NFSV2PROC_NOOP,
 155         NFSV2PROC_STATFS,
 156         NFSV2PROC_NOOP,
 157         NFSV2PROC_NOOP,
 158         NFSV2PROC_NOOP,
 159 };
 160
 161 /*
 162  * Initialize sockets and congestion for a new NFS connection.
 163  * We do not free the sockaddr if error.
 164  * Which arguments are set to NULL indicate what kind of call it is.
 165  * cred == NULL --> a call to connect to a pNFS DS
 166  * nmp == NULL --> indicates an upcall to userland or a NFSv4.0 callback
 167  */
 168 int
 169 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
 170     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
 171 {
 172         int rcvreserve, sndreserve;
 173         int pktscale, pktscalesav;
 174         struct sockaddr *saddr;
 175         struct ucred *origcred;
 176         CLIENT *client;
 177         struct netconfig *nconf;
 178         struct socket *so;
 179         int one = 1, retries, error = 0;
 180         struct thread *td = curthread;
 181         SVCXPRT *xprt;
 182         struct timeval timo;
 183
 184         /*
 185          * We need to establish the socket using the credentials of
 186          * the mountpoint.  Some parts of this process (such as
 187          * sobind() and soconnect()) will use the curent thread's
 188          * credential instead of the socket credential.  To work
 189          * around this, temporarily change the current thread's
 190          * credential to that of the mountpoint.
 191          *
 192          * XXX: It would be better to explicitly pass the correct
 193          * credential to sobind() and soconnect().
 194          */
 195         origcred = td->td_ucred;
 196
 197         /*
 198          * Use the credential in nr_cred, if not NULL.
 199          */
 200         if (nrp->nr_cred != NULL)
 201                 td->td_ucred = nrp->nr_cred;
 202         else
 203                 td->td_ucred = cred;
 204         saddr = nrp->nr_nam;
 205
 206         if (saddr->sa_family == AF_INET)
 207                 if (nrp->nr_sotype == SOCK_DGRAM)
 208                         nconf = getnetconfigent("udp");
 209                 else
 210                         nconf = getnetconfigent("tcp");
 211         else
 212                 if (nrp->nr_sotype == SOCK_DGRAM)
 213                         nconf = getnetconfigent("udp6");
 214                 else
 215                         nconf = getnetconfigent("tcp6");
 216
 217         pktscale = nfs_bufpackets;
 218         if (pktscale < 2)
 219                 pktscale = 2;
 220         if (pktscale > 64)
 221                 pktscale = 64;
 222         pktscalesav = pktscale;
 223         /*
 224          * soreserve() can fail if sb_max is too small, so shrink pktscale
 225          * and try again if there is an error.
 226          * Print a log message suggesting increasing sb_max.
 227          * Creating a socket and doing this is necessary since, if the
 228          * reservation sizes are too large and will make soreserve() fail,
 229          * the connection will work until a large send is attempted and
 230          * then it will loop in the krpc code.
 231          */
 232         so = NULL;
 233         saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
 234         error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
 235             nrp->nr_soproto, td->td_ucred, td);
 236         if (error != 0)
 237                 goto out;
 238         do {
 239             if (error != 0 && pktscale > 2) {
 240                 if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
 241                     pktscale == pktscalesav)
 242                     printf("Consider increasing kern.ipc.maxsockbuf\n");
 243                 pktscale--;
 244             }
 245             if (nrp->nr_sotype == SOCK_DGRAM) {
 246                 if (nmp != NULL) {
 247                         sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
 248                             pktscale;
 249                         rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
 250                             pktscale;
 251                 } else {
 252                         sndreserve = rcvreserve = 1024 * pktscale;
 253                 }
 254             } else {
 255                 if (nrp->nr_sotype != SOCK_STREAM)
 256                         panic("nfscon sotype");
 257                 if (nmp != NULL) {
 258                         sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
 259                             sizeof (u_int32_t)) * pktscale;
 260                         rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
 261                             sizeof (u_int32_t)) * pktscale;
 262                 } else {
 263                         sndreserve = rcvreserve = 1024 * pktscale;
 264                 }
 265             }
 266             error = soreserve(so, sndreserve, rcvreserve);
 267             if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
 268                 pktscale <= 2)
 269                 printf("Must increase kern.ipc.maxsockbuf or reduce"
 270                     " rsize, wsize\n");
 271         } while (error != 0 && pktscale > 2);
 272         soclose(so);
 273         if (error != 0)
 274                 goto out;
 275
 276         client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
 277             nrp->nr_vers, sndreserve, rcvreserve);
 278         CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
 279         if (nmp != NULL) {
 280                 if ((nmp->nm_flag & NFSMNT_INT))
 281                         CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
 282                 if ((nmp->nm_flag & NFSMNT_RESVPORT))
 283                         CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
 284                 if (NFSHASSOFT(nmp)) {
 285                         if (nmp->nm_sotype == SOCK_DGRAM)
 286                                 /*
 287                                  * For UDP, the large timeout for a reconnect
 288                                  * will be set to "nm_retry * nm_timeo / 2", so
 289                                  * we only want to do 2 reconnect timeout
 290                                  * retries.
 291                                  */
 292                                 retries = 2;
 293                         else
 294                                 retries = nmp->nm_retry;
 295                 } else
 296                         retries = INT_MAX;
 297                 if (NFSHASNFSV4N(nmp)) {
 298                         if (cred != NULL) {
 299                                 if (NFSHASSOFT(nmp)) {
 300                                         /*
 301                                          * This should be a DS mount.
 302                                          * Use CLSET_TIMEOUT to set the timeout
 303                                          * for connections to DSs instead of
 304                                          * specifying a timeout on each RPC.
 305                                          * This is done so that SO_SNDTIMEO
 306                                          * is set on the TCP socket as well
 307                                          * as specifying a time limit when
 308                                          * waiting for an RPC reply.  Useful
 309                                          * if the send queue for the TCP
 310                                          * connection has become constipated,
 311                                          * due to a failed DS.
 312                                          * The choice of lease_duration / 4 is
 313                                          * fairly arbitrary, but seems to work
 314                                          * ok, with a lower bound of 10sec.
 315                                          */
 316                                         timo.tv_sec = nfsrv_lease / 4;
 317                                         if (timo.tv_sec < 10)
 318                                                 timo.tv_sec = 10;
 319                                         timo.tv_usec = 0;
 320                                         CLNT_CONTROL(client, CLSET_TIMEOUT,
 321                                             &timo);
 322                                 }
 323                                 /*
 324                                  * Make sure the nfscbd_pool doesn't get
 325                                  * destroyed while doing this.
 326                                  */
 327                                 NFSD_LOCK();
 328                                 if (nfs_numnfscbd > 0) {
 329                                         nfs_numnfscbd++;
 330                                         NFSD_UNLOCK();
 331                                         xprt = svc_vc_create_backchannel(
 332                                             nfscbd_pool);
 333                                         CLNT_CONTROL(client, CLSET_BACKCHANNEL,
 334                                             xprt);
 335                                         NFSD_LOCK();
 336                                         nfs_numnfscbd--;
 337                                         if (nfs_numnfscbd == 0)
 338                                                 wakeup(&nfs_numnfscbd);
 339                                 }
 340                                 NFSD_UNLOCK();
 341                         } else {
 342                                 /*
 343                                  * cred == NULL for a DS connect.
 344                                  * For connects to a DS, set a retry limit
 345                                  * so that failed DSs will be detected.
 346                                  * This is ok for NFSv4.1, since a DS does
 347                                  * not maintain open/lock state and is the
 348                                  * only case where using a "soft" mount is
 349                                  * recommended for NFSv4.
 350                                  * For mounts from the MDS to DS, this is done
 351                                  * via mount options, but that is not the case
 352                                  * here.  The retry limit here can be adjusted
 353                                  * via the sysctl vfs.nfs.dsretries.
 354                                  * See the comment above w.r.t. timeout.
 355                                  */
 356                                 timo.tv_sec = nfsrv_lease / 4;
 357                                 if (timo.tv_sec < 10)
 358                                         timo.tv_sec = 10;
 359                                 timo.tv_usec = 0;
 360                                 CLNT_CONTROL(client, CLSET_TIMEOUT, &timo);
 361                                 retries = nfs_dsretries;
 362                         }
 363                 }
 364         } else {
 365                 /*
 366                  * Three cases:
 367                  * - Null RPC callback to client
 368                  * - Non-Null RPC callback to client, wait a little longer
 369                  * - upcalls to nfsuserd and gssd (clp == NULL)
 370                  */
 371                 if (callback_retry_mult == 0) {
 372                         retries = NFSV4_UPCALLRETRY;
 373                         CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
 374                 } else {
 375                         retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
 376                 }
 377         }
 378         CLNT_CONTROL(client, CLSET_RETRIES, &retries);
 379
 380         if (nmp != NULL) {
 381                 /*
 382                  * For UDP, there are 2 timeouts:
 383                  * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
 384                  *   that does a retransmit of an RPC request using the same
 385                  *   socket and xid. This is what you normally want to do,
 386                  *   since NFS servers depend on "same xid" for their
 387                  *   Duplicate Request Cache.
 388                  * - timeout specified in CLNT_CALL_MBUF(), which specifies when
 389                  *   retransmits on the same socket should fail and a fresh
 390                  *   socket created. Each of these timeouts counts as one
 391                  *   CLSET_RETRIES as set above.
 392                  * Set the initial retransmit timeout for UDP. This timeout
 393                  * doesn't exist for TCP and the following call just fails,
 394                  * which is ok.
 395                  */
 396                 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
 397                 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
 398                 CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
 399         }
 400
 401         mtx_lock(&nrp->nr_mtx);
 402         if (nrp->nr_client != NULL) {
 403                 mtx_unlock(&nrp->nr_mtx);
 404                 /*
 405                  * Someone else already connected.
 406                  */
 407                 CLNT_RELEASE(client);
 408         } else {
 409                 nrp->nr_client = client;
 410                 /*
 411                  * Protocols that do not require connections may be optionally
 412                  * left unconnected for servers that reply from a port other
 413                  * than NFS_PORT.
 414                  */
 415                 if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
 416                         mtx_unlock(&nrp->nr_mtx);
 417                         CLNT_CONTROL(client, CLSET_CONNECT, &one);
 418                 } else
 419                         mtx_unlock(&nrp->nr_mtx);
 420         }
 421
 422 out:
 423         /* Restore current thread's credentials. */
 424         td->td_ucred = origcred;
 425
 426         NFSEXITCODE(error);
 427         return (error);
 428 }
 429
 430 /*
 431  * NFS disconnect. Clean up and unlink.
 432  */
 433 void
 434 newnfs_disconnect(struct nfssockreq *nrp)
 435 {
 436         CLIENT *client;
 437
 438         mtx_lock(&nrp->nr_mtx);
 439         if (nrp->nr_client != NULL) {
 440                 client = nrp->nr_client;
 441                 nrp->nr_client = NULL;
 442                 mtx_unlock(&nrp->nr_mtx);
 443                 rpc_gss_secpurge_call(client);
 444                 CLNT_CLOSE(client);
 445                 CLNT_RELEASE(client);
 446         } else {
 447                 mtx_unlock(&nrp->nr_mtx);
 448         }
 449 }
 450
 451 static AUTH *
 452 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
 453     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
 454 {
 455         rpc_gss_service_t svc;
 456         AUTH *auth;
 457
 458         switch (secflavour) {
 459         case RPCSEC_GSS_KRB5:
 460         case RPCSEC_GSS_KRB5I:
 461         case RPCSEC_GSS_KRB5P:
 462                 if (!mech_oid) {
 463                         if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
 464                                 return (NULL);
 465                 }
 466                 if (secflavour == RPCSEC_GSS_KRB5)
 467                         svc = rpc_gss_svc_none;
 468                 else if (secflavour == RPCSEC_GSS_KRB5I)
 469                         svc = rpc_gss_svc_integrity;
 470                 else
 471                         svc = rpc_gss_svc_privacy;
 472
 473                 if (clnt_principal == NULL)
 474                         auth = rpc_gss_secfind_call(nrp->nr_client, cred,
 475                             srv_principal, mech_oid, svc);
 476                 else {
 477                         auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
 478                             clnt_principal, srv_principal, "kerberosv5",
 479                             svc, NULL, NULL, NULL);
 480                         return (auth);
 481                 }
 482                 if (auth != NULL)
 483                         return (auth);
 484                 /* fallthrough */
 485         case AUTH_SYS:
 486         default:
 487                 return (authunix_create(cred));
 488
 489         }
 490 }
 491
 492 /*
 493  * Callback from the RPC code to generate up/down notifications.
 494  */
 495
 496 struct nfs_feedback_arg {
 497         struct nfsmount *nf_mount;
 498         int             nf_lastmsg;     /* last tprintf */
 499         int             nf_tprintfmsg;
 500         struct thread   *nf_td;
 501 };
 502
 503 static void
 504 nfs_feedback(int type, int proc, void *arg)
 505 {
 506         struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
 507         struct nfsmount *nmp = nf->nf_mount;
 508         time_t now;
 509
 510         switch (type) {
 511         case FEEDBACK_REXMIT2:
 512         case FEEDBACK_RECONNECT:
 513                 now = NFSD_MONOSEC;
 514                 if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
 515                         nfs_down(nmp, nf->nf_td,
 516                             "not responding", 0, NFSSTA_TIMEO);
 517                         nf->nf_tprintfmsg = TRUE;
 518                         nf->nf_lastmsg = now;
 519                 }
 520                 break;
 521
 522         case FEEDBACK_OK:
 523                 nfs_up(nf->nf_mount, nf->nf_td,
 524                     "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
 525                 break;
 526         }
 527 }
 528
 529 /*
 530  * newnfs_request - goes something like this
 531  *      - does the rpc by calling the krpc layer
 532  *      - break down rpc header and return with nfs reply
 533  * nb: always frees up nd_mreq mbuf list
 534  */
 535 int
 536 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
 537     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
 538     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
 539     u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
 540 {
 541         uint32_t retseq, retval, slotseq, *tl;
 542         time_t waituntil;
 543         int i = 0, j = 0, opcnt, set_sigset = 0, slot;
 544         int error = 0, usegssname = 0, secflavour = AUTH_SYS;
 545         int freeslot, maxslot, reterr, slotpos, timeo;
 546         u_int16_t procnum;
 547         u_int trylater_delay = 1;
 548         struct nfs_feedback_arg nf;
 549         struct timeval timo;
 550         AUTH *auth;
 551         struct rpc_callextra ext;
 552         enum clnt_stat stat;
 553         struct nfsreq *rep = NULL;
 554         char *srv_principal = NULL, *clnt_principal = NULL;
 555         sigset_t oldset;
 556         struct ucred *authcred;
 557         struct nfsclsession *sep;
 558         uint8_t sessionid[NFSX_V4SESSIONID];
 559
 560         sep = dssep;
 561         if (xidp != NULL)
 562                 *xidp = 0;
 563         /* Reject requests while attempting a forced unmount. */
 564         if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
 565                 m_freem(nd->nd_mreq);
 566                 return (ESTALE);
 567         }
 568
 569         /*
 570          * Set authcred, which is used to acquire RPC credentials to
 571          * the cred argument, by default. The crhold() should not be
 572          * necessary, but will ensure that some future code change
 573          * doesn't result in the credential being free'd prematurely.
 574          */
 575         authcred = crhold(cred);
 576
 577         /* For client side interruptible mounts, mask off the signals. */
 578         if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
 579                 newnfs_set_sigmask(td, &oldset);
 580                 set_sigset = 1;
 581         }
 582
 583         /*
 584          * XXX if not already connected call nfs_connect now. Longer
 585          * term, change nfs_mount to call nfs_connect unconditionally
 586          * and let clnt_reconnect_create handle reconnects.
 587          */
 588         if (nrp->nr_client == NULL)
 589                 newnfs_connect(nmp, nrp, cred, td, 0);
 590
 591         /*
 592          * For a client side mount, nmp is != NULL and clp == NULL. For
 593          * server calls (callbacks or upcalls), nmp == NULL.
 594          */
 595         if (clp != NULL) {
 596                 NFSLOCKSTATE();
 597                 if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
 598                         secflavour = RPCSEC_GSS_KRB5;
 599                         if (nd->nd_procnum != NFSPROC_NULL) {
 600                                 if (clp->lc_flags & LCL_GSSINTEGRITY)
 601                                         secflavour = RPCSEC_GSS_KRB5I;
 602                                 else if (clp->lc_flags & LCL_GSSPRIVACY)
 603                                         secflavour = RPCSEC_GSS_KRB5P;
 604                         }
 605                 }
 606                 NFSUNLOCKSTATE();
 607         } else if (nmp != NULL && NFSHASKERB(nmp) &&
 608              nd->nd_procnum != NFSPROC_NULL) {
 609                 if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
 610                         nd->nd_flag |= ND_USEGSSNAME;
 611                 if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
 612                         /*
 613                          * If there is a client side host based credential,
 614                          * use that, otherwise use the system uid, if set.
 615                          * The system uid is in the nmp->nm_sockreq.nr_cred
 616                          * credentials.
 617                          */
 618                         if (nmp->nm_krbnamelen > 0) {
 619                                 usegssname = 1;
 620                                 clnt_principal = nmp->nm_krbname;
 621                         } else if (nmp->nm_uid != (uid_t)-1) {
 622                                 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
 623                                     ("newnfs_request: NULL nr_cred"));
 624                                 crfree(authcred);
 625                                 authcred = crhold(nmp->nm_sockreq.nr_cred);
 626                         }
 627                 } else if (nmp->nm_krbnamelen == 0 &&
 628                     nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
 629                         /*
 630                          * If there is no host based principal name and
 631                          * the system uid is set and this is root, use the
 632                          * system uid, since root won't have user
 633                          * credentials in a credentials cache file.
 634                          * The system uid is in the nmp->nm_sockreq.nr_cred
 635                          * credentials.
 636                          */
 637                         KASSERT(nmp->nm_sockreq.nr_cred != NULL,
 638                             ("newnfs_request: NULL nr_cred"));
 639                         crfree(authcred);
 640                         authcred = crhold(nmp->nm_sockreq.nr_cred);
 641                 }
 642                 if (NFSHASINTEGRITY(nmp))
 643                         secflavour = RPCSEC_GSS_KRB5I;
 644                 else if (NFSHASPRIVACY(nmp))
 645                         secflavour = RPCSEC_GSS_KRB5P;
 646                 else
 647                         secflavour = RPCSEC_GSS_KRB5;
 648                 srv_principal = NFSMNT_SRVKRBNAME(nmp);
 649         } else if (nmp != NULL && !NFSHASKERB(nmp) &&
 650             nd->nd_procnum != NFSPROC_NULL &&
 651             (nd->nd_flag & ND_USEGSSNAME) != 0) {
 652                 /*
 653                  * Use the uid that did the mount when the RPC is doing
 654                  * NFSv4 system operations, as indicated by the
 655                  * ND_USEGSSNAME flag, for the AUTH_SYS case.
 656                  * The credentials in nm_sockreq.nr_cred were used for the
 657                  * mount.
 658                  */
 659                 KASSERT(nmp->nm_sockreq.nr_cred != NULL,
 660                     ("newnfs_request: NULL nr_cred"));
 661                 crfree(authcred);
 662                 authcred = crhold(nmp->nm_sockreq.nr_cred);
 663         }
 664
 665         if (nmp != NULL) {
 666                 bzero(&nf, sizeof(struct nfs_feedback_arg));
 667                 nf.nf_mount = nmp;
 668                 nf.nf_td = td;
 669                 nf.nf_lastmsg = NFSD_MONOSEC -
 670                     ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
 671         }
 672
 673         if (nd->nd_procnum == NFSPROC_NULL)
 674                 auth = authnone_create();
 675         else if (usegssname) {
 676                 /*
 677                  * For this case, the authenticator is held in the
 678                  * nfssockreq structure, so don't release the reference count
 679                  * held on it. --> Don't AUTH_DESTROY() it in this function.
 680                  */
 681                 if (nrp->nr_auth == NULL)
 682                         nrp->nr_auth = nfs_getauth(nrp, secflavour,
 683                             clnt_principal, srv_principal, NULL, authcred);
 684                 else
 685                         rpc_gss_refresh_auth_call(nrp->nr_auth);
 686                 auth = nrp->nr_auth;
 687         } else
 688                 auth = nfs_getauth(nrp, secflavour, NULL,
 689                     srv_principal, NULL, authcred);
 690         crfree(authcred);
 691         if (auth == NULL) {
 692                 m_freem(nd->nd_mreq);
 693                 if (set_sigset)
 694                         newnfs_restore_sigmask(td, &oldset);
 695                 return (EACCES);
 696         }
 697         bzero(&ext, sizeof(ext));
 698         ext.rc_auth = auth;
 699         if (nmp != NULL) {
 700                 ext.rc_feedback = nfs_feedback;
 701                 ext.rc_feedback_arg = &nf;
 702         }
 703
 704         procnum = nd->nd_procnum;
 705         if ((nd->nd_flag & ND_NFSV4) &&
 706             nd->nd_procnum != NFSPROC_NULL &&
 707             nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
 708                 procnum = NFSV4PROC_COMPOUND;
 709
 710         if (nmp != NULL) {
 711                 NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
 712
 713                 /* Map the procnum to the old NFSv2 one, as required. */
 714                 if ((nd->nd_flag & ND_NFSV2) != 0) {
 715                         if (nd->nd_procnum < NFS_V3NPROCS)
 716                                 procnum = nfsv2_procid[nd->nd_procnum];
 717                         else
 718                                 procnum = NFSV2PROC_NOOP;
 719                 }
 720
 721                 /*
 722                  * Now only used for the R_DONTRECOVER case, but until that is
 723                  * supported within the krpc code, I need to keep a queue of
 724                  * outstanding RPCs for nfsv4 client requests.
 725                  */
 726                 if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
 727                         rep = malloc(sizeof(struct nfsreq),
 728                             M_NFSDREQ, M_WAITOK);
 729 #ifdef KDTRACE_HOOKS
 730                 if (dtrace_nfscl_nfs234_start_probe != NULL) {
 731                         uint32_t probe_id;
 732                         int probe_procnum;
 733
 734                         if (nd->nd_flag & ND_NFSV4) {
 735                                 probe_id =
 736                                     nfscl_nfs4_start_probes[nd->nd_procnum];
 737                                 probe_procnum = nd->nd_procnum;
 738                         } else if (nd->nd_flag & ND_NFSV3) {
 739                                 probe_id = nfscl_nfs3_start_probes[procnum];
 740                                 probe_procnum = procnum;
 741                         } else {
 742                                 probe_id =
 743                                     nfscl_nfs2_start_probes[nd->nd_procnum];
 744                                 probe_procnum = procnum;
 745                         }
 746                         if (probe_id != 0)
 747                                 (dtrace_nfscl_nfs234_start_probe)
 748                                     (probe_id, vp, nd->nd_mreq, cred,
 749                                      probe_procnum);
 750                 }
 751 #endif
 752         }
 753         freeslot = -1;          /* Set to slot that needs to be free'd */
 754 tryagain:
 755         slot = -1;              /* Slot that needs a sequence# increment. */
 756         /*
 757          * This timeout specifies when a new socket should be created,
 758          * along with new xid values. For UDP, this should be done
 759          * infrequently, since retransmits of RPC requests should normally
 760          * use the same xid.
 761          */
 762         if (nmp == NULL) {
 763                 timo.tv_usec = 0;
 764                 if (clp == NULL)
 765                         timo.tv_sec = NFSV4_UPCALLTIMEO;
 766                 else
 767                         timo.tv_sec = NFSV4_CALLBACKTIMEO;
 768         } else {
 769                 if (nrp->nr_sotype != SOCK_DGRAM) {
 770                         timo.tv_usec = 0;
 771                         if ((nmp->nm_flag & NFSMNT_NFSV4))
 772                                 timo.tv_sec = INT_MAX;
 773                         else
 774                                 timo.tv_sec = NFS_TCPTIMEO;
 775                 } else {
 776                         if (NFSHASSOFT(nmp)) {
 777                                 /*
 778                                  * CLSET_RETRIES is set to 2, so this should be
 779                                  * half of the total timeout required.
 780                                  */
 781                                 timeo = nmp->nm_retry * nmp->nm_timeo / 2;
 782                                 if (timeo < 1)
 783                                         timeo = 1;
 784                                 timo.tv_sec = timeo / NFS_HZ;
 785                                 timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
 786                                     NFS_HZ;
 787                         } else {
 788                                 /* For UDP hard mounts, use a large value. */
 789                                 timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
 790                                 timo.tv_usec = 0;
 791                         }
 792                 }
 793
 794                 if (rep != NULL) {
 795                         rep->r_flags = 0;
 796                         rep->r_nmp = nmp;
 797                         /*
 798                          * Chain request into list of outstanding requests.
 799                          */
 800                         NFSLOCKREQ();
 801                         TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
 802                         NFSUNLOCKREQ();
 803                 }
 804         }
 805
 806         nd->nd_mrep = NULL;
 807         if (clp != NULL && sep != NULL)
 808                 stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
 809                     nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
 810         else
 811                 stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
 812                     nd->nd_mreq, &nd->nd_mrep, timo);
 813         NFSCL_DEBUG(2, "clnt call=%d\n", stat);
 814
 815         if (rep != NULL) {
 816                 /*
 817                  * RPC done, unlink the request.
 818                  */
 819                 NFSLOCKREQ();
 820                 TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
 821                 NFSUNLOCKREQ();
 822         }
 823
 824         /*
 825          * If there was a successful reply and a tprintf msg.
 826          * tprintf a response.
 827          */
 828         if (stat == RPC_SUCCESS) {
 829                 error = 0;
 830         } else if (stat == RPC_TIMEDOUT) {
 831                 NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
 832                 error = ETIMEDOUT;
 833         } else if (stat == RPC_VERSMISMATCH) {
 834                 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
 835                 error = EOPNOTSUPP;
 836         } else if (stat == RPC_PROGVERSMISMATCH) {
 837                 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
 838                 error = EPROTONOSUPPORT;
 839         } else if (stat == RPC_INTR) {
 840                 error = EINTR;
 841         } else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
 842              stat == RPC_SYSTEMERROR) {
 843                 /* Check for a session slot that needs to be free'd. */
 844                 if ((nd->nd_flag & (ND_NFSV41 | ND_HASSLOTID)) ==
 845                     (ND_NFSV41 | ND_HASSLOTID) && nmp != NULL &&
 846                     nd->nd_procnum != NFSPROC_NULL) {
 847                         /*
 848                          * This should only occur when either the MDS or
 849                          * a client has an RPC against a DS fail.
 850                          * This happens because these cases use "soft"
 851                          * connections that can time out and fail.
 852                          * The slot used for this RPC is now in a
 853                          * non-deterministic state, but if the slot isn't
 854                          * free'd, threads can get stuck waiting for a slot.
 855                          */
 856                         if (sep == NULL)
 857                                 sep = nfsmnt_mdssession(nmp);
 858                         /*
 859                          * Bump the sequence# out of range, so that reuse of
 860                          * this slot will result in an NFSERR_SEQMISORDERED
 861                          * error and not a bogus cached RPC reply.
 862                          */
 863                         mtx_lock(&sep->nfsess_mtx);
 864                         sep->nfsess_slotseq[nd->nd_slotid] += 10;
 865                         mtx_unlock(&sep->nfsess_mtx);
 866                         /* And free the slot. */
 867                         nfsv4_freeslot(sep, nd->nd_slotid);
 868                 }
 869                 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
 870                 error = ENXIO;
 871         } else {
 872                 NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
 873                 error = EACCES;
 874         }
 875         if (error) {
 876                 m_freem(nd->nd_mreq);
 877                 if (usegssname == 0)
 878                         AUTH_DESTROY(auth);
 879                 if (rep != NULL)
 880                         free(rep, M_NFSDREQ);
 881                 if (set_sigset)
 882                         newnfs_restore_sigmask(td, &oldset);
 883                 return (error);
 884         }
 885
 886         KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
 887
 888         /*
 889          * Search for any mbufs that are not a multiple of 4 bytes long
 890          * or with m_data not longword aligned.
 891          * These could cause pointer alignment problems, so copy them to
 892          * well aligned mbufs.
 893          */
 894         newnfs_realign(&nd->nd_mrep, M_WAITOK);
 895         nd->nd_md = nd->nd_mrep;
 896         nd->nd_dpos = mtod(nd->nd_md, caddr_t);
 897         nd->nd_repstat = 0;
 898         if (nd->nd_procnum != NFSPROC_NULL &&
 899             nd->nd_procnum != NFSV4PROC_CBNULL) {
 900                 /* If sep == NULL, set it to the default in nmp. */
 901                 if (sep == NULL && nmp != NULL)
 902                         sep = nfsmnt_mdssession(nmp);
 903                 /*
 904                  * and now the actual NFS xdr.
 905                  */
 906                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 907                 nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
 908                 if (nd->nd_repstat >= 10000)
 909                         NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
 910                             (int)nd->nd_repstat);
 911
 912                 /*
 913                  * Get rid of the tag, return count and SEQUENCE result for
 914                  * NFSv4.
 915                  */
 916                 if ((nd->nd_flag & ND_NFSV4) != 0 && nd->nd_repstat !=
 917                     NFSERR_MINORVERMISMATCH) {
 918                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 919                         i = fxdr_unsigned(int, *tl);
 920                         error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
 921                         if (error)
 922                                 goto nfsmout;
 923                         NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
 924                         opcnt = fxdr_unsigned(int, *tl++);
 925                         i = fxdr_unsigned(int, *tl++);
 926                         j = fxdr_unsigned(int, *tl);
 927                         if (j >= 10000)
 928                                 NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
 929                         /*
 930                          * If the first op is Sequence, free up the slot.
 931                          */
 932                         if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
 933                             (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
 934                                 NFSCL_DEBUG(1, "failed seq=%d\n", j);
 935                         if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
 936                             (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
 937                             j == 0)) && sep != NULL) {
 938                                 if (i == NFSV4OP_SEQUENCE)
 939                                         NFSM_DISSECT(tl, uint32_t *,
 940                                             NFSX_V4SESSIONID +
 941                                             5 * NFSX_UNSIGNED);
 942                                 else
 943                                         NFSM_DISSECT(tl, uint32_t *,
 944                                             NFSX_V4SESSIONID +
 945                                             4 * NFSX_UNSIGNED);
 946                                 mtx_lock(&sep->nfsess_mtx);
 947                                 if (bcmp(tl, sep->nfsess_sessionid,
 948                                     NFSX_V4SESSIONID) == 0) {
 949                                         tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
 950                                         retseq = fxdr_unsigned(uint32_t, *tl++);
 951                                         slot = fxdr_unsigned(int, *tl++);
 952                                         freeslot = slot;
 953                                         if (retseq != sep->nfsess_slotseq[slot])
 954                                                 printf("retseq diff 0x%x\n",
 955                                                     retseq);
 956                                         retval = fxdr_unsigned(uint32_t, *++tl);
 957                                         if ((retval + 1) < sep->nfsess_foreslots
 958                                             )
 959                                                 sep->nfsess_foreslots = (retval
 960                                                     + 1);
 961                                         else if ((retval + 1) >
 962                                             sep->nfsess_foreslots)
 963                                                 sep->nfsess_foreslots = (retval
 964                                                     < 64) ? (retval + 1) : 64;
 965                                 }
 966                                 mtx_unlock(&sep->nfsess_mtx);
 967
 968                                 /* Grab the op and status for the next one. */
 969                                 if (opcnt > 1) {
 970                                         NFSM_DISSECT(tl, uint32_t *,
 971                                             2 * NFSX_UNSIGNED);
 972                                         i = fxdr_unsigned(int, *tl++);
 973                                         j = fxdr_unsigned(int, *tl);
 974                                 }
 975                         }
 976                 }
 977                 if (nd->nd_repstat != 0) {
 978                         if (nd->nd_repstat == NFSERR_BADSESSION &&
 979                             nmp != NULL && dssep == NULL &&
 980                             (nd->nd_flag & ND_NFSV41) != 0) {
 981                                 /*
 982                                  * If this is a client side MDS RPC, mark
 983                                  * the MDS session defunct and initiate
 984                                  * recovery, as required.
 985                                  * The nfsess_defunct field is protected by
 986                                  * the NFSLOCKMNT()/nm_mtx lock and not the
 987                                  * nfsess_mtx lock to simplify its handling,
 988                                  * for the MDS session. This lock is also
 989                                  * sufficient for nfsess_sessionid, since it
 990                                  * never changes in the structure.
 991                                  */
 992                                 NFSCL_DEBUG(1, "Got badsession\n");
 993                                 NFSLOCKCLSTATE();
 994                                 NFSLOCKMNT(nmp);
 995                                 sep = NFSMNT_MDSSESSION(nmp);
 996                                 if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
 997                                     NFSX_V4SESSIONID) == 0) {
 998                                         /* Initiate recovery. */
 999                                         sep->nfsess_defunct = 1;
1000                                         NFSCL_DEBUG(1, "Marked defunct\n");
1001                                         if (nmp->nm_clp != NULL) {
1002                                                 nmp->nm_clp->nfsc_flags |=
1003                                                     NFSCLFLAGS_RECOVER;
1004                                                 wakeup(nmp->nm_clp);
1005                                         }
1006                                 }
1007                                 NFSUNLOCKCLSTATE();
1008                                 /*
1009                                  * Sleep for up to 1sec waiting for a new
1010                                  * session.
1011                                  */
1012                                 mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1013                                     "nfsbadsess", hz);
1014                                 /*
1015                                  * Get the session again, in case a new one
1016                                  * has been created during the sleep.
1017                                  */
1018                                 sep = NFSMNT_MDSSESSION(nmp);
1019                                 NFSUNLOCKMNT(nmp);
1020                                 if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1021                                         reterr = nfsv4_sequencelookup(nmp, sep,
1022                                             &slotpos, &maxslot, &slotseq,
1023                                             sessionid);
1024                                         if (reterr == 0) {
1025                                                 /* Fill in new session info. */
1026                                                 NFSCL_DEBUG(1,
1027                                                   "Filling in new sequence\n");
1028                                                 tl = nd->nd_sequence;
1029                                                 bcopy(sessionid, tl,
1030                                                     NFSX_V4SESSIONID);
1031                                                 tl += NFSX_V4SESSIONID /
1032                                                     NFSX_UNSIGNED;
1033                                                 *tl++ = txdr_unsigned(slotseq);
1034                                                 *tl++ = txdr_unsigned(slotpos);
1035                                                 *tl = txdr_unsigned(maxslot);
1036                                         }
1037                                         if (reterr == NFSERR_BADSESSION ||
1038                                             reterr == 0) {
1039                                                 NFSCL_DEBUG(1,
1040                                                     "Badsession looping\n");
1041                                                 m_freem(nd->nd_mrep);
1042                                                 nd->nd_mrep = NULL;
1043                                                 goto tryagain;
1044                                         }
1045                                         nd->nd_repstat = reterr;
1046                                         NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1047                                 }
1048                         }
1049                         /*
1050                          * When clp != NULL, it is a callback and all
1051                          * callback operations can be retried for NFSERR_DELAY.
1052                          */
1053                         if (((nd->nd_repstat == NFSERR_DELAY ||
1054                               nd->nd_repstat == NFSERR_GRACE) &&
1055                              (nd->nd_flag & ND_NFSV4) && (clp != NULL ||
1056                              (nd->nd_procnum != NFSPROC_DELEGRETURN &&
1057                              nd->nd_procnum != NFSPROC_SETATTR &&
1058                              nd->nd_procnum != NFSPROC_READ &&
1059                              nd->nd_procnum != NFSPROC_READDS &&
1060                              nd->nd_procnum != NFSPROC_WRITE &&
1061                              nd->nd_procnum != NFSPROC_WRITEDS &&
1062                              nd->nd_procnum != NFSPROC_OPEN &&
1063                              nd->nd_procnum != NFSPROC_CREATE &&
1064                              nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1065                              nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1066                              nd->nd_procnum != NFSPROC_CLOSE &&
1067                              nd->nd_procnum != NFSPROC_LOCK &&
1068                              nd->nd_procnum != NFSPROC_LOCKU))) ||
1069                             (nd->nd_repstat == NFSERR_DELAY &&
1070                              (nd->nd_flag & ND_NFSV4) == 0) ||
1071                             nd->nd_repstat == NFSERR_RESOURCE) {
1072                                 if (trylater_delay > NFS_TRYLATERDEL)
1073                                         trylater_delay = NFS_TRYLATERDEL;
1074                                 waituntil = NFSD_MONOSEC + trylater_delay;
1075                                 while (NFSD_MONOSEC < waituntil)
1076                                         (void) nfs_catnap(PZERO, 0, "nfstry");
1077                                 trylater_delay *= 2;
1078                                 if (slot != -1) {
1079                                         mtx_lock(&sep->nfsess_mtx);
1080                                         sep->nfsess_slotseq[slot]++;
1081                                         *nd->nd_slotseq = txdr_unsigned(
1082                                             sep->nfsess_slotseq[slot]);
1083                                         mtx_unlock(&sep->nfsess_mtx);
1084                                 }
1085                                 m_freem(nd->nd_mrep);
1086                                 nd->nd_mrep = NULL;
1087                                 goto tryagain;
1088                         }
1089
1090                         /*
1091                          * If the File Handle was stale, invalidate the
1092                          * lookup cache, just in case.
1093                          * (vp != NULL implies a client side call)
1094                          */
1095                         if (nd->nd_repstat == ESTALE && vp != NULL) {
1096                                 cache_purge(vp);
1097                                 if (ncl_call_invalcaches != NULL)
1098                                         (*ncl_call_invalcaches)(vp);
1099                         }
1100                 }
1101                 if ((nd->nd_flag & ND_NFSV4) != 0) {
1102                         /* Free the slot, as required. */
1103                         if (freeslot != -1)
1104                                 nfsv4_freeslot(sep, freeslot);
1105                         /*
1106                          * If this op is Putfh, throw its results away.
1107                          */
1108                         if (j >= 10000)
1109                                 NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1110                         if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1111                                 NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1112                                 i = fxdr_unsigned(int, *tl++);
1113                                 j = fxdr_unsigned(int, *tl);
1114                                 if (j >= 10000)
1115                                         NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1116                                             j);
1117                                 /*
1118                                  * All Compounds that do an Op that must
1119                                  * be in sequence consist of NFSV4OP_PUTFH
1120                                  * followed by one of these. As such, we
1121                                  * can determine if the seqid# should be
1122                                  * incremented, here.
1123                                  */
1124                                 if ((i == NFSV4OP_OPEN ||
1125                                      i == NFSV4OP_OPENCONFIRM ||
1126                                      i == NFSV4OP_OPENDOWNGRADE ||
1127                                      i == NFSV4OP_CLOSE ||
1128                                      i == NFSV4OP_LOCK ||
1129                                      i == NFSV4OP_LOCKU) &&
1130                                     (j == 0 ||
1131                                      (j != NFSERR_STALECLIENTID &&
1132                                       j != NFSERR_STALESTATEID &&
1133                                       j != NFSERR_BADSTATEID &&
1134                                       j != NFSERR_BADSEQID &&
1135                                       j != NFSERR_BADXDR &&
1136                                       j != NFSERR_RESOURCE &&
1137                                       j != NFSERR_NOFILEHANDLE)))
1138                                         nd->nd_flag |= ND_INCRSEQID;
1139                         }
1140                         /*
1141                          * If this op's status is non-zero, mark
1142                          * that there is no more data to process.
1143                          * The exception is Setattr, which always has xdr
1144                          * when it has failed.
1145                          */
1146                         if (j != 0 && i != NFSV4OP_SETATTR)
1147                                 nd->nd_flag |= ND_NOMOREDATA;
1148
1149                         /*
1150                          * If R_DONTRECOVER is set, replace the stale error
1151                          * reply, so that recovery isn't initiated.
1152                          */
1153                         if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1154                              nd->nd_repstat == NFSERR_BADSESSION ||
1155                              nd->nd_repstat == NFSERR_STALESTATEID) &&
1156                             rep != NULL && (rep->r_flags & R_DONTRECOVER))
1157                                 nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1158                 }
1159         }
1160
1161 #ifdef KDTRACE_HOOKS
1162         if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1163                 uint32_t probe_id;
1164                 int probe_procnum;
1165
1166                 if (nd->nd_flag & ND_NFSV4) {
1167                         probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1168                         probe_procnum = nd->nd_procnum;
1169                 } else if (nd->nd_flag & ND_NFSV3) {
1170                         probe_id = nfscl_nfs3_done_probes[procnum];
1171                         probe_procnum = procnum;
1172                 } else {
1173                         probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1174                         probe_procnum = procnum;
1175                 }
1176                 if (probe_id != 0)
1177                         (dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1178                             nd->nd_mreq, cred, probe_procnum, 0);
1179         }
1180 #endif
1181
1182         m_freem(nd->nd_mreq);
1183         if (usegssname == 0)
1184                 AUTH_DESTROY(auth);
1185         if (rep != NULL)
1186                 free(rep, M_NFSDREQ);
1187         if (set_sigset)
1188                 newnfs_restore_sigmask(td, &oldset);
1189         return (0);
1190 nfsmout:
1191         m_freem(nd->nd_mrep);
1192         m_freem(nd->nd_mreq);
1193         if (usegssname == 0)
1194                 AUTH_DESTROY(auth);
1195         if (rep != NULL)
1196                 free(rep, M_NFSDREQ);
1197         if (set_sigset)
1198                 newnfs_restore_sigmask(td, &oldset);
1199         return (error);
1200 }
1201
1202 /*
1203  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1204  * wait for all requests to complete. This is used by forced unmounts
1205  * to terminate any outstanding RPCs.
1206  */
1207 int
1208 newnfs_nmcancelreqs(struct nfsmount *nmp)
1209 {
1210         struct nfsclds *dsp;
1211         struct __rpc_client *cl;
1212
1213         if (nmp->nm_sockreq.nr_client != NULL)
1214                 CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1215 lookformore:
1216         NFSLOCKMNT(nmp);
1217         TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1218                 NFSLOCKDS(dsp);
1219                 if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1220                     (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1221                     dsp->nfsclds_sockp != NULL &&
1222                     dsp->nfsclds_sockp->nr_client != NULL) {
1223                         dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1224                         cl = dsp->nfsclds_sockp->nr_client;
1225                         NFSUNLOCKDS(dsp);
1226                         NFSUNLOCKMNT(nmp);
1227                         CLNT_CLOSE(cl);
1228                         goto lookformore;
1229                 }
1230                 NFSUNLOCKDS(dsp);
1231         }
1232         NFSUNLOCKMNT(nmp);
1233         return (0);
1234 }
1235
1236 /*
1237  * Any signal that can interrupt an NFS operation in an intr mount
1238  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1239  */
1240 int newnfs_sig_set[] = {
1241         SIGINT,
1242         SIGTERM,
1243         SIGHUP,
1244         SIGKILL,
1245         SIGQUIT
1246 };
1247
1248 /*
1249  * Check to see if one of the signals in our subset is pending on
1250  * the process (in an intr mount).
1251  */
1252 static int
1253 nfs_sig_pending(sigset_t set)
1254 {
1255         int i;
1256
1257         for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1258                 if (SIGISMEMBER(set, newnfs_sig_set[i]))
1259                         return (1);
1260         return (0);
1261 }
1262
1263 /*
1264  * The set/restore sigmask functions are used to (temporarily) overwrite
1265  * the thread td_sigmask during an RPC call (for example). These are also
1266  * used in other places in the NFS client that might tsleep().
1267  */
1268 void
1269 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1270 {
1271         sigset_t newset;
1272         int i;
1273         struct proc *p;
1274
1275         SIGFILLSET(newset);
1276         if (td == NULL)
1277                 td = curthread; /* XXX */
1278         p = td->td_proc;
1279         /* Remove the NFS set of signals from newset */
1280         PROC_LOCK(p);
1281         mtx_lock(&p->p_sigacts->ps_mtx);
1282         for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1283                 /*
1284                  * But make sure we leave the ones already masked
1285                  * by the process, ie. remove the signal from the
1286                  * temporary signalmask only if it wasn't already
1287                  * in p_sigmask.
1288                  */
1289                 if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1290                     !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1291                         SIGDELSET(newset, newnfs_sig_set[i]);
1292         }
1293         mtx_unlock(&p->p_sigacts->ps_mtx);
1294         kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1295             SIGPROCMASK_PROC_LOCKED);
1296         PROC_UNLOCK(p);
1297 }
1298
1299 void
1300 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1301 {
1302         if (td == NULL)
1303                 td = curthread; /* XXX */
1304         kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1305 }
1306
1307 /*
1308  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1309  * old one after msleep() returns.
1310  */
1311 int
1312 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1313 {
1314         sigset_t oldset;
1315         int error;
1316
1317         if ((priority & PCATCH) == 0)
1318                 return msleep(ident, mtx, priority, wmesg, timo);
1319         if (td == NULL)
1320                 td = curthread; /* XXX */
1321         newnfs_set_sigmask(td, &oldset);
1322         error = msleep(ident, mtx, priority, wmesg, timo);
1323         newnfs_restore_sigmask(td, &oldset);
1324         return (error);
1325 }
1326
1327 /*
1328  * Test for a termination condition pending on the process.
1329  * This is used for NFSMNT_INT mounts.
1330  */
1331 int
1332 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1333 {
1334         struct proc *p;
1335         sigset_t tmpset;
1336
1337         /* Terminate all requests while attempting a forced unmount. */
1338         if (NFSCL_FORCEDISM(nmp->nm_mountp))
1339                 return (EIO);
1340         if (!(nmp->nm_flag & NFSMNT_INT))
1341                 return (0);
1342         if (td == NULL)
1343                 return (0);
1344         p = td->td_proc;
1345         PROC_LOCK(p);
1346         tmpset = p->p_siglist;
1347         SIGSETOR(tmpset, td->td_siglist);
1348         SIGSETNAND(tmpset, td->td_sigmask);
1349         mtx_lock(&p->p_sigacts->ps_mtx);
1350         SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1351         mtx_unlock(&p->p_sigacts->ps_mtx);
1352         if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1353             && nfs_sig_pending(tmpset)) {
1354                 PROC_UNLOCK(p);
1355                 return (EINTR);
1356         }
1357         PROC_UNLOCK(p);
1358         return (0);
1359 }
1360
1361 static int
1362 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1363 {
1364         struct proc *p;
1365
1366         p = td ? td->td_proc : NULL;
1367         if (error) {
1368                 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1369                     server, msg, error);
1370         } else {
1371                 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1372         }
1373         return (0);
1374 }
1375
1376 static void
1377 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1378     int error, int flags)
1379 {
1380         if (nmp == NULL)
1381                 return;
1382         mtx_lock(&nmp->nm_mtx);
1383         if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1384                 nmp->nm_state |= NFSSTA_TIMEO;
1385                 mtx_unlock(&nmp->nm_mtx);
1386                 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1387                     VQ_NOTRESP, 0);
1388         } else
1389                 mtx_unlock(&nmp->nm_mtx);
1390         mtx_lock(&nmp->nm_mtx);
1391         if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1392                 nmp->nm_state |= NFSSTA_LOCKTIMEO;
1393                 mtx_unlock(&nmp->nm_mtx);
1394                 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1395                     VQ_NOTRESPLOCK, 0);
1396         } else
1397                 mtx_unlock(&nmp->nm_mtx);
1398         nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1399 }
1400
1401 static void
1402 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1403     int flags, int tprintfmsg)
1404 {
1405         if (nmp == NULL)
1406                 return;
1407         if (tprintfmsg) {
1408                 nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1409         }
1410
1411         mtx_lock(&nmp->nm_mtx);
1412         if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1413                 nmp->nm_state &= ~NFSSTA_TIMEO;
1414                 mtx_unlock(&nmp->nm_mtx);
1415                 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1416                     VQ_NOTRESP, 1);
1417         } else
1418                 mtx_unlock(&nmp->nm_mtx);
1419
1420         mtx_lock(&nmp->nm_mtx);
1421         if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1422                 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1423                 mtx_unlock(&nmp->nm_mtx);
1424                 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1425                     VQ_NOTRESPLOCK, 1);
1426         } else
1427                 mtx_unlock(&nmp->nm_mtx);
1428 }
1429