sys/fs/nfs/nfs_commonkrpc.c

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