sys/fs/nfsserver/nfs_nfsdport.c

   1 /*-
   2  * Copyright (c) 1989, 1993
   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  * 4. 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 #include <sys/capsicum.h>
  38
  39 /*
  40  * Functions that perform the vfs operations required by the routines in
  41  * nfsd_serv.c. It is hoped that this change will make the server more
  42  * portable.
  43  */
  44
  45 #include <fs/nfs/nfsport.h>
  46 #include <sys/hash.h>
  47 #include <sys/sysctl.h>
  48 #include <nlm/nlm_prot.h>
  49 #include <nlm/nlm.h>
  50
  51 FEATURE(nfsd, "NFSv4 server");
  52
  53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
  54 extern int nfsrv_useacl;
  55 extern int newnfs_numnfsd;
  56 extern struct mount nfsv4root_mnt;
  57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
  58 extern void (*nfsd_call_servertimer)(void);
  59 extern SVCPOOL  *nfsrvd_pool;
  60 extern struct nfsv4lock nfsd_suspend_lock;
  61 extern struct nfsclienthashhead *nfsclienthash;
  62 extern struct nfslockhashhead *nfslockhash;
  63 extern struct nfssessionhash *nfssessionhash;
  64 extern int nfsrv_sessionhashsize;
  65 extern struct nfsstatsv1 nfsstatsv1;
  66 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
  67 NFSDLOCKMUTEX;
  68 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
  69 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
  70 struct mtx nfsrc_udpmtx;
  71 struct mtx nfs_v4root_mutex;
  72 struct nfsrvfh nfs_rootfh, nfs_pubfh;
  73 int nfs_pubfhset = 0, nfs_rootfhset = 0;
  74 struct proc *nfsd_master_proc = NULL;
  75 int nfsd_debuglevel = 0;
  76 static pid_t nfsd_master_pid = (pid_t)-1;
  77 static char nfsd_master_comm[MAXCOMLEN + 1];
  78 static struct timeval nfsd_master_start;
  79 static uint32_t nfsv4_sysid = 0;
  80
  81 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
  82     struct ucred *);
  83
  84 int nfsrv_enable_crossmntpt = 1;
  85 static int nfs_commit_blks;
  86 static int nfs_commit_miss;
  87 extern int nfsrv_issuedelegs;
  88 extern int nfsrv_dolocallocks;
  89 extern int nfsd_enable_stringtouid;
  90
  91 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "NFS server");
  92 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
  93     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
  94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
  95     0, "");
  96 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
  97     0, "");
  98 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
  99     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
 100 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
 101     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
 102 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
 103     0, "Debug level for NFS server");
 104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
 105     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
 106
 107 #define MAX_REORDERED_RPC       16
 108 #define NUM_HEURISTIC           1031
 109 #define NHUSE_INIT              64
 110 #define NHUSE_INC               16
 111 #define NHUSE_MAX               2048
 112
 113 static struct nfsheur {
 114         struct vnode *nh_vp;    /* vp to match (unreferenced pointer) */
 115         off_t nh_nextoff;       /* next offset for sequential detection */
 116         int nh_use;             /* use count for selection */
 117         int nh_seqcount;        /* heuristic */
 118 } nfsheur[NUM_HEURISTIC];
 119
 120
 121 /*
 122  * Heuristic to detect sequential operation.
 123  */
 124 static struct nfsheur *
 125 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
 126 {
 127         struct nfsheur *nh;
 128         int hi, try;
 129
 130         /* Locate best candidate. */
 131         try = 32;
 132         hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
 133         nh = &nfsheur[hi];
 134         while (try--) {
 135                 if (nfsheur[hi].nh_vp == vp) {
 136                         nh = &nfsheur[hi];
 137                         break;
 138                 }
 139                 if (nfsheur[hi].nh_use > 0)
 140                         --nfsheur[hi].nh_use;
 141                 hi = (hi + 1) % NUM_HEURISTIC;
 142                 if (nfsheur[hi].nh_use < nh->nh_use)
 143                         nh = &nfsheur[hi];
 144         }
 145
 146         /* Initialize hint if this is a new file. */
 147         if (nh->nh_vp != vp) {
 148                 nh->nh_vp = vp;
 149                 nh->nh_nextoff = uio->uio_offset;
 150                 nh->nh_use = NHUSE_INIT;
 151                 if (uio->uio_offset == 0)
 152                         nh->nh_seqcount = 4;
 153                 else
 154                         nh->nh_seqcount = 1;
 155         }
 156
 157         /* Calculate heuristic. */
 158         if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
 159             uio->uio_offset == nh->nh_nextoff) {
 160                 /* See comments in vfs_vnops.c:sequential_heuristic(). */
 161                 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
 162                 if (nh->nh_seqcount > IO_SEQMAX)
 163                         nh->nh_seqcount = IO_SEQMAX;
 164         } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
 165             imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
 166                 /* Probably a reordered RPC, leave seqcount alone. */
 167         } else if (nh->nh_seqcount > 1) {
 168                 nh->nh_seqcount /= 2;
 169         } else {
 170                 nh->nh_seqcount = 0;
 171         }
 172         nh->nh_use += NHUSE_INC;
 173         if (nh->nh_use > NHUSE_MAX)
 174                 nh->nh_use = NHUSE_MAX;
 175         return (nh);
 176 }
 177
 178 /*
 179  * Get attributes into nfsvattr structure.
 180  */
 181 int
 182 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 183     struct thread *p, int vpislocked)
 184 {
 185         int error, lockedit = 0;
 186
 187         if (vpislocked == 0) {
 188                 /*
 189                  * When vpislocked == 0, the vnode is either exclusively
 190                  * locked by this thread or not locked by this thread.
 191                  * As such, shared lock it, if not exclusively locked.
 192                  */
 193                 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
 194                         lockedit = 1;
 195                         NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
 196                 }
 197         }
 198         error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
 199         if (lockedit != 0)
 200                 NFSVOPUNLOCK(vp, 0);
 201
 202         NFSEXITCODE(error);
 203         return (error);
 204 }
 205
 206 /*
 207  * Get a file handle for a vnode.
 208  */
 209 int
 210 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
 211 {
 212         int error;
 213
 214         NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
 215         fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
 216         error = VOP_VPTOFH(vp, &fhp->fh_fid);
 217
 218         NFSEXITCODE(error);
 219         return (error);
 220 }
 221
 222 /*
 223  * Perform access checking for vnodes obtained from file handles that would
 224  * refer to files already opened by a Unix client. You cannot just use
 225  * vn_writechk() and VOP_ACCESSX() for two reasons.
 226  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
 227  *     case.
 228  * 2 - The owner is to be given access irrespective of mode bits for some
 229  *     operations, so that processes that chmod after opening a file don't
 230  *     break.
 231  */
 232 int
 233 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
 234     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
 235     u_int32_t *supportedtypep)
 236 {
 237         struct vattr vattr;
 238         int error = 0, getret = 0;
 239
 240         if (vpislocked == 0) {
 241                 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
 242                         error = EPERM;
 243                         goto out;
 244                 }
 245         }
 246         if (accmode & VWRITE) {
 247                 /* Just vn_writechk() changed to check rdonly */
 248                 /*
 249                  * Disallow write attempts on read-only file systems;
 250                  * unless the file is a socket or a block or character
 251                  * device resident on the file system.
 252                  */
 253                 if (NFSVNO_EXRDONLY(exp) ||
 254                     (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 255                         switch (vp->v_type) {
 256                         case VREG:
 257                         case VDIR:
 258                         case VLNK:
 259                                 error = EROFS;
 260                         default:
 261                                 break;
 262                         }
 263                 }
 264                 /*
 265                  * If there's shared text associated with
 266                  * the inode, try to free it up once.  If
 267                  * we fail, we can't allow writing.
 268                  */
 269                 if (VOP_IS_TEXT(vp) && error == 0)
 270                         error = ETXTBSY;
 271         }
 272         if (error != 0) {
 273                 if (vpislocked == 0)
 274                         NFSVOPUNLOCK(vp, 0);
 275                 goto out;
 276         }
 277
 278         /*
 279          * Should the override still be applied when ACLs are enabled?
 280          */
 281         error = VOP_ACCESSX(vp, accmode, cred, p);
 282         if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
 283                 /*
 284                  * Try again with VEXPLICIT_DENY, to see if the test for
 285                  * deletion is supported.
 286                  */
 287                 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
 288                 if (error == 0) {
 289                         if (vp->v_type == VDIR) {
 290                                 accmode &= ~(VDELETE | VDELETE_CHILD);
 291                                 accmode |= VWRITE;
 292                                 error = VOP_ACCESSX(vp, accmode, cred, p);
 293                         } else if (supportedtypep != NULL) {
 294                                 *supportedtypep &= ~NFSACCESS_DELETE;
 295                         }
 296                 }
 297         }
 298
 299         /*
 300          * Allow certain operations for the owner (reads and writes
 301          * on files that are already open).
 302          */
 303         if (override != NFSACCCHK_NOOVERRIDE &&
 304             (error == EPERM || error == EACCES)) {
 305                 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
 306                         error = 0;
 307                 else if (override & NFSACCCHK_ALLOWOWNER) {
 308                         getret = VOP_GETATTR(vp, &vattr, cred);
 309                         if (getret == 0 && cred->cr_uid == vattr.va_uid)
 310                                 error = 0;
 311                 }
 312         }
 313         if (vpislocked == 0)
 314                 NFSVOPUNLOCK(vp, 0);
 315
 316 out:
 317         NFSEXITCODE(error);
 318         return (error);
 319 }
 320
 321 /*
 322  * Set attribute(s) vnop.
 323  */
 324 int
 325 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 326     struct thread *p, struct nfsexstuff *exp)
 327 {
 328         int error;
 329
 330         error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
 331         NFSEXITCODE(error);
 332         return (error);
 333 }
 334
 335 /*
 336  * Set up nameidata for a lookup() call and do it.
 337  */
 338 int
 339 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
 340     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
 341     struct vnode **retdirp)
 342 {
 343         struct componentname *cnp = &ndp->ni_cnd;
 344         int i;
 345         struct iovec aiov;
 346         struct uio auio;
 347         int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
 348         int error = 0, crossmnt;
 349         char *cp;
 350
 351         *retdirp = NULL;
 352         cnp->cn_nameptr = cnp->cn_pnbuf;
 353         ndp->ni_lcf = 0;
 354         /*
 355          * Extract and set starting directory.
 356          */
 357         if (dp->v_type != VDIR) {
 358                 if (islocked)
 359                         vput(dp);
 360                 else
 361                         vrele(dp);
 362                 nfsvno_relpathbuf(ndp);
 363                 error = ENOTDIR;
 364                 goto out1;
 365         }
 366         if (islocked)
 367                 NFSVOPUNLOCK(dp, 0);
 368         VREF(dp);
 369         *retdirp = dp;
 370         if (NFSVNO_EXRDONLY(exp))
 371                 cnp->cn_flags |= RDONLY;
 372         ndp->ni_segflg = UIO_SYSSPACE;
 373         crossmnt = 1;
 374
 375         if (nd->nd_flag & ND_PUBLOOKUP) {
 376                 ndp->ni_loopcnt = 0;
 377                 if (cnp->cn_pnbuf[0] == '/') {
 378                         vrele(dp);
 379                         /*
 380                          * Check for degenerate pathnames here, since lookup()
 381                          * panics on them.
 382                          */
 383                         for (i = 1; i < ndp->ni_pathlen; i++)
 384                                 if (cnp->cn_pnbuf[i] != '/')
 385                                         break;
 386                         if (i == ndp->ni_pathlen) {
 387                                 error = NFSERR_ACCES;
 388                                 goto out;
 389                         }
 390                         dp = rootvnode;
 391                         VREF(dp);
 392                 }
 393         } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
 394             (nd->nd_flag & ND_NFSV4) == 0) {
 395                 /*
 396                  * Only cross mount points for NFSv4 when doing a
 397                  * mount while traversing the file system above
 398                  * the mount point, unless nfsrv_enable_crossmntpt is set.
 399                  */
 400                 cnp->cn_flags |= NOCROSSMOUNT;
 401                 crossmnt = 0;
 402         }
 403
 404         /*
 405          * Initialize for scan, set ni_startdir and bump ref on dp again
 406          * because lookup() will dereference ni_startdir.
 407          */
 408
 409         cnp->cn_thread = p;
 410         ndp->ni_startdir = dp;
 411         ndp->ni_rootdir = rootvnode;
 412         ndp->ni_topdir = NULL;
 413
 414         if (!lockleaf)
 415                 cnp->cn_flags |= LOCKLEAF;
 416         for (;;) {
 417                 cnp->cn_nameptr = cnp->cn_pnbuf;
 418                 /*
 419                  * Call lookup() to do the real work.  If an error occurs,
 420                  * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
 421                  * we do not have to dereference anything before returning.
 422                  * In either case ni_startdir will be dereferenced and NULLed
 423                  * out.
 424                  */
 425                 error = lookup(ndp);
 426                 if (error)
 427                         break;
 428
 429                 /*
 430                  * Check for encountering a symbolic link.  Trivial
 431                  * termination occurs if no symlink encountered.
 432                  */
 433                 if ((cnp->cn_flags & ISSYMLINK) == 0) {
 434                         if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
 435                                 nfsvno_relpathbuf(ndp);
 436                         if (ndp->ni_vp && !lockleaf)
 437                                 NFSVOPUNLOCK(ndp->ni_vp, 0);
 438                         break;
 439                 }
 440
 441                 /*
 442                  * Validate symlink
 443                  */
 444                 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
 445                         NFSVOPUNLOCK(ndp->ni_dvp, 0);
 446                 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
 447                         error = EINVAL;
 448                         goto badlink2;
 449                 }
 450
 451                 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
 452                         error = ELOOP;
 453                         goto badlink2;
 454                 }
 455                 if (ndp->ni_pathlen > 1)
 456                         cp = uma_zalloc(namei_zone, M_WAITOK);
 457                 else
 458                         cp = cnp->cn_pnbuf;
 459                 aiov.iov_base = cp;
 460                 aiov.iov_len = MAXPATHLEN;
 461                 auio.uio_iov = &aiov;
 462                 auio.uio_iovcnt = 1;
 463                 auio.uio_offset = 0;
 464                 auio.uio_rw = UIO_READ;
 465                 auio.uio_segflg = UIO_SYSSPACE;
 466                 auio.uio_td = NULL;
 467                 auio.uio_resid = MAXPATHLEN;
 468                 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
 469                 if (error) {
 470                 badlink1:
 471                         if (ndp->ni_pathlen > 1)
 472                                 uma_zfree(namei_zone, cp);
 473                 badlink2:
 474                         vrele(ndp->ni_dvp);
 475                         vput(ndp->ni_vp);
 476                         break;
 477                 }
 478                 linklen = MAXPATHLEN - auio.uio_resid;
 479                 if (linklen == 0) {
 480                         error = ENOENT;
 481                         goto badlink1;
 482                 }
 483                 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
 484                         error = ENAMETOOLONG;
 485                         goto badlink1;
 486                 }
 487
 488                 /*
 489                  * Adjust or replace path
 490                  */
 491                 if (ndp->ni_pathlen > 1) {
 492                         NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
 493                         uma_zfree(namei_zone, cnp->cn_pnbuf);
 494                         cnp->cn_pnbuf = cp;
 495                 } else
 496                         cnp->cn_pnbuf[linklen] = '\0';
 497                 ndp->ni_pathlen += linklen;
 498
 499                 /*
 500                  * Cleanup refs for next loop and check if root directory
 501                  * should replace current directory.  Normally ni_dvp
 502                  * becomes the new base directory and is cleaned up when
 503                  * we loop.  Explicitly null pointers after invalidation
 504                  * to clarify operation.
 505                  */
 506                 vput(ndp->ni_vp);
 507                 ndp->ni_vp = NULL;
 508
 509                 if (cnp->cn_pnbuf[0] == '/') {
 510                         vrele(ndp->ni_dvp);
 511                         ndp->ni_dvp = ndp->ni_rootdir;
 512                         VREF(ndp->ni_dvp);
 513                 }
 514                 ndp->ni_startdir = ndp->ni_dvp;
 515                 ndp->ni_dvp = NULL;
 516         }
 517         if (!lockleaf)
 518                 cnp->cn_flags &= ~LOCKLEAF;
 519
 520 out:
 521         if (error) {
 522                 nfsvno_relpathbuf(ndp);
 523                 ndp->ni_vp = NULL;
 524                 ndp->ni_dvp = NULL;
 525                 ndp->ni_startdir = NULL;
 526         } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
 527                 ndp->ni_dvp = NULL;
 528         }
 529
 530 out1:
 531         NFSEXITCODE2(error, nd);
 532         return (error);
 533 }
 534
 535 /*
 536  * Set up a pathname buffer and return a pointer to it and, optionally
 537  * set a hash pointer.
 538  */
 539 void
 540 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
 541 {
 542         struct componentname *cnp = &ndp->ni_cnd;
 543
 544         cnp->cn_flags |= (NOMACCHECK | HASBUF);
 545         cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
 546         if (hashpp != NULL)
 547                 *hashpp = NULL;
 548         *bufpp = cnp->cn_pnbuf;
 549 }
 550
 551 /*
 552  * Release the above path buffer, if not released by nfsvno_namei().
 553  */
 554 void
 555 nfsvno_relpathbuf(struct nameidata *ndp)
 556 {
 557
 558         if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
 559                 panic("nfsrelpath");
 560         uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
 561         ndp->ni_cnd.cn_flags &= ~HASBUF;
 562 }
 563
 564 /*
 565  * Readlink vnode op into an mbuf list.
 566  */
 567 int
 568 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
 569     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
 570 {
 571         struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
 572         struct iovec *ivp = iv;
 573         struct uio io, *uiop = &io;
 574         struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
 575         int i, len, tlen, error = 0;
 576
 577         len = 0;
 578         i = 0;
 579         while (len < NFS_MAXPATHLEN) {
 580                 NFSMGET(mp);
 581                 MCLGET(mp, M_WAITOK);
 582                 mp->m_len = M_SIZE(mp);
 583                 if (len == 0) {
 584                         mp3 = mp2 = mp;
 585                 } else {
 586                         mp2->m_next = mp;
 587                         mp2 = mp;
 588                 }
 589                 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
 590                         mp->m_len = NFS_MAXPATHLEN - len;
 591                         len = NFS_MAXPATHLEN;
 592                 } else {
 593                         len += mp->m_len;
 594                 }
 595                 ivp->iov_base = mtod(mp, caddr_t);
 596                 ivp->iov_len = mp->m_len;
 597                 i++;
 598                 ivp++;
 599         }
 600         uiop->uio_iov = iv;
 601         uiop->uio_iovcnt = i;
 602         uiop->uio_offset = 0;
 603         uiop->uio_resid = len;
 604         uiop->uio_rw = UIO_READ;
 605         uiop->uio_segflg = UIO_SYSSPACE;
 606         uiop->uio_td = NULL;
 607         error = VOP_READLINK(vp, uiop, cred);
 608         if (error) {
 609                 m_freem(mp3);
 610                 *lenp = 0;
 611                 goto out;
 612         }
 613         if (uiop->uio_resid > 0) {
 614                 len -= uiop->uio_resid;
 615                 tlen = NFSM_RNDUP(len);
 616                 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
 617         }
 618         *lenp = len;
 619         *mpp = mp3;
 620         *mpendp = mp;
 621
 622 out:
 623         NFSEXITCODE(error);
 624         return (error);
 625 }
 626
 627 /*
 628  * Read vnode op call into mbuf list.
 629  */
 630 int
 631 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
 632     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
 633 {
 634         struct mbuf *m;
 635         int i;
 636         struct iovec *iv;
 637         struct iovec *iv2;
 638         int error = 0, len, left, siz, tlen, ioflag = 0;
 639         struct mbuf *m2 = NULL, *m3;
 640         struct uio io, *uiop = &io;
 641         struct nfsheur *nh;
 642
 643         len = left = NFSM_RNDUP(cnt);
 644         m3 = NULL;
 645         /*
 646          * Generate the mbuf list with the uio_iov ref. to it.
 647          */
 648         i = 0;
 649         while (left > 0) {
 650                 NFSMGET(m);
 651                 MCLGET(m, M_WAITOK);
 652                 m->m_len = 0;
 653                 siz = min(M_TRAILINGSPACE(m), left);
 654                 left -= siz;
 655                 i++;
 656                 if (m3)
 657                         m2->m_next = m;
 658                 else
 659                         m3 = m;
 660                 m2 = m;
 661         }
 662         MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
 663             M_TEMP, M_WAITOK);
 664         uiop->uio_iov = iv2 = iv;
 665         m = m3;
 666         left = len;
 667         i = 0;
 668         while (left > 0) {
 669                 if (m == NULL)
 670                         panic("nfsvno_read iov");
 671                 siz = min(M_TRAILINGSPACE(m), left);
 672                 if (siz > 0) {
 673                         iv->iov_base = mtod(m, caddr_t) + m->m_len;
 674                         iv->iov_len = siz;
 675                         m->m_len += siz;
 676                         left -= siz;
 677                         iv++;
 678                         i++;
 679                 }
 680                 m = m->m_next;
 681         }
 682         uiop->uio_iovcnt = i;
 683         uiop->uio_offset = off;
 684         uiop->uio_resid = len;
 685         uiop->uio_rw = UIO_READ;
 686         uiop->uio_segflg = UIO_SYSSPACE;
 687         uiop->uio_td = NULL;
 688         nh = nfsrv_sequential_heuristic(uiop, vp);
 689         ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
 690         /* XXX KDM make this more systematic? */
 691         nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
 692         error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
 693         FREE((caddr_t)iv2, M_TEMP);
 694         if (error) {
 695                 m_freem(m3);
 696                 *mpp = NULL;
 697                 goto out;
 698         }
 699         nh->nh_nextoff = uiop->uio_offset;
 700         tlen = len - uiop->uio_resid;
 701         cnt = cnt < tlen ? cnt : tlen;
 702         tlen = NFSM_RNDUP(cnt);
 703         if (tlen == 0) {
 704                 m_freem(m3);
 705                 m3 = NULL;
 706         } else if (len != tlen || tlen != cnt)
 707                 nfsrv_adj(m3, len - tlen, tlen - cnt);
 708         *mpp = m3;
 709         *mpendp = m2;
 710
 711 out:
 712         NFSEXITCODE(error);
 713         return (error);
 714 }
 715
 716 /*
 717  * Write vnode op from an mbuf list.
 718  */
 719 int
 720 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
 721     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
 722 {
 723         struct iovec *ivp;
 724         int i, len;
 725         struct iovec *iv;
 726         int ioflags, error;
 727         struct uio io, *uiop = &io;
 728         struct nfsheur *nh;
 729
 730         MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
 731             M_WAITOK);
 732         uiop->uio_iov = iv = ivp;
 733         uiop->uio_iovcnt = cnt;
 734         i = mtod(mp, caddr_t) + mp->m_len - cp;
 735         len = retlen;
 736         while (len > 0) {
 737                 if (mp == NULL)
 738                         panic("nfsvno_write");
 739                 if (i > 0) {
 740                         i = min(i, len);
 741                         ivp->iov_base = cp;
 742                         ivp->iov_len = i;
 743                         ivp++;
 744                         len -= i;
 745                 }
 746                 mp = mp->m_next;
 747                 if (mp) {
 748                         i = mp->m_len;
 749                         cp = mtod(mp, caddr_t);
 750                 }
 751         }
 752
 753         if (stable == NFSWRITE_UNSTABLE)
 754                 ioflags = IO_NODELOCKED;
 755         else
 756                 ioflags = (IO_SYNC | IO_NODELOCKED);
 757         uiop->uio_resid = retlen;
 758         uiop->uio_rw = UIO_WRITE;
 759         uiop->uio_segflg = UIO_SYSSPACE;
 760         NFSUIOPROC(uiop, p);
 761         uiop->uio_offset = off;
 762         nh = nfsrv_sequential_heuristic(uiop, vp);
 763         ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
 764         /* XXX KDM make this more systematic? */
 765         nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
 766         error = VOP_WRITE(vp, uiop, ioflags, cred);
 767         if (error == 0)
 768                 nh->nh_nextoff = uiop->uio_offset;
 769         FREE((caddr_t)iv, M_TEMP);
 770
 771         NFSEXITCODE(error);
 772         return (error);
 773 }
 774
 775 /*
 776  * Common code for creating a regular file (plus special files for V2).
 777  */
 778 int
 779 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
 780     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
 781     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
 782 {
 783         u_quad_t tempsize;
 784         int error;
 785
 786         error = nd->nd_repstat;
 787         if (!error && ndp->ni_vp == NULL) {
 788                 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
 789                         vrele(ndp->ni_startdir);
 790                         error = VOP_CREATE(ndp->ni_dvp,
 791                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
 792                         vput(ndp->ni_dvp);
 793                         nfsvno_relpathbuf(ndp);
 794                         if (!error) {
 795                                 if (*exclusive_flagp) {
 796                                         *exclusive_flagp = 0;
 797                                         NFSVNO_ATTRINIT(nvap);
 798                                         nvap->na_atime.tv_sec = cverf[0];
 799                                         nvap->na_atime.tv_nsec = cverf[1];
 800                                         error = VOP_SETATTR(ndp->ni_vp,
 801                                             &nvap->na_vattr, nd->nd_cred);
 802                                         if (error != 0) {
 803                                                 vput(ndp->ni_vp);
 804                                                 ndp->ni_vp = NULL;
 805                                                 error = NFSERR_NOTSUPP;
 806                                         }
 807                                 }
 808                         }
 809                 /*
 810                  * NFS V2 Only. nfsrvd_mknod() does this for V3.
 811                  * (This implies, just get out on an error.)
 812                  */
 813                 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
 814                         nvap->na_type == VFIFO) {
 815                         if (nvap->na_type == VCHR && rdev == 0xffffffff)
 816                                 nvap->na_type = VFIFO;
 817                         if (nvap->na_type != VFIFO &&
 818                             (error = priv_check_cred(nd->nd_cred,
 819                              PRIV_VFS_MKNOD_DEV, 0))) {
 820                                 vrele(ndp->ni_startdir);
 821                                 nfsvno_relpathbuf(ndp);
 822                                 vput(ndp->ni_dvp);
 823                                 goto out;
 824                         }
 825                         nvap->na_rdev = rdev;
 826                         error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 827                             &ndp->ni_cnd, &nvap->na_vattr);
 828                         vput(ndp->ni_dvp);
 829                         nfsvno_relpathbuf(ndp);
 830                         vrele(ndp->ni_startdir);
 831                         if (error)
 832                                 goto out;
 833                 } else {
 834                         vrele(ndp->ni_startdir);
 835                         nfsvno_relpathbuf(ndp);
 836                         vput(ndp->ni_dvp);
 837                         error = ENXIO;
 838                         goto out;
 839                 }
 840                 *vpp = ndp->ni_vp;
 841         } else {
 842                 /*
 843                  * Handle cases where error is already set and/or
 844                  * the file exists.
 845                  * 1 - clean up the lookup
 846                  * 2 - iff !error and na_size set, truncate it
 847                  */
 848                 vrele(ndp->ni_startdir);
 849                 nfsvno_relpathbuf(ndp);
 850                 *vpp = ndp->ni_vp;
 851                 if (ndp->ni_dvp == *vpp)
 852                         vrele(ndp->ni_dvp);
 853                 else
 854                         vput(ndp->ni_dvp);
 855                 if (!error && nvap->na_size != VNOVAL) {
 856                         error = nfsvno_accchk(*vpp, VWRITE,
 857                             nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
 858                             NFSACCCHK_VPISLOCKED, NULL);
 859                         if (!error) {
 860                                 tempsize = nvap->na_size;
 861                                 NFSVNO_ATTRINIT(nvap);
 862                                 nvap->na_size = tempsize;
 863                                 error = VOP_SETATTR(*vpp,
 864                                     &nvap->na_vattr, nd->nd_cred);
 865                         }
 866                 }
 867                 if (error)
 868                         vput(*vpp);
 869         }
 870
 871 out:
 872         NFSEXITCODE(error);
 873         return (error);
 874 }
 875
 876 /*
 877  * Do a mknod vnode op.
 878  */
 879 int
 880 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
 881     struct thread *p)
 882 {
 883         int error = 0;
 884         enum vtype vtyp;
 885
 886         vtyp = nvap->na_type;
 887         /*
 888          * Iff doesn't exist, create it.
 889          */
 890         if (ndp->ni_vp) {
 891                 vrele(ndp->ni_startdir);
 892                 nfsvno_relpathbuf(ndp);
 893                 vput(ndp->ni_dvp);
 894                 vrele(ndp->ni_vp);
 895                 error = EEXIST;
 896                 goto out;
 897         }
 898         if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
 899                 vrele(ndp->ni_startdir);
 900                 nfsvno_relpathbuf(ndp);
 901                 vput(ndp->ni_dvp);
 902                 error = NFSERR_BADTYPE;
 903                 goto out;
 904         }
 905         if (vtyp == VSOCK) {
 906                 vrele(ndp->ni_startdir);
 907                 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
 908                     &ndp->ni_cnd, &nvap->na_vattr);
 909                 vput(ndp->ni_dvp);
 910                 nfsvno_relpathbuf(ndp);
 911         } else {
 912                 if (nvap->na_type != VFIFO &&
 913                     (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
 914                         vrele(ndp->ni_startdir);
 915                         nfsvno_relpathbuf(ndp);
 916                         vput(ndp->ni_dvp);
 917                         goto out;
 918                 }
 919                 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 920                     &ndp->ni_cnd, &nvap->na_vattr);
 921                 vput(ndp->ni_dvp);
 922                 nfsvno_relpathbuf(ndp);
 923                 vrele(ndp->ni_startdir);
 924                 /*
 925                  * Since VOP_MKNOD returns the ni_vp, I can't
 926                  * see any reason to do the lookup.
 927                  */
 928         }
 929
 930 out:
 931         NFSEXITCODE(error);
 932         return (error);
 933 }
 934
 935 /*
 936  * Mkdir vnode op.
 937  */
 938 int
 939 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
 940     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
 941 {
 942         int error = 0;
 943
 944         if (ndp->ni_vp != NULL) {
 945                 if (ndp->ni_dvp == ndp->ni_vp)
 946                         vrele(ndp->ni_dvp);
 947                 else
 948                         vput(ndp->ni_dvp);
 949                 vrele(ndp->ni_vp);
 950                 nfsvno_relpathbuf(ndp);
 951                 error = EEXIST;
 952                 goto out;
 953         }
 954         error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 955             &nvap->na_vattr);
 956         vput(ndp->ni_dvp);
 957         nfsvno_relpathbuf(ndp);
 958
 959 out:
 960         NFSEXITCODE(error);
 961         return (error);
 962 }
 963
 964 /*
 965  * symlink vnode op.
 966  */
 967 int
 968 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
 969     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
 970     struct nfsexstuff *exp)
 971 {
 972         int error = 0;
 973
 974         if (ndp->ni_vp) {
 975                 vrele(ndp->ni_startdir);
 976                 nfsvno_relpathbuf(ndp);
 977                 if (ndp->ni_dvp == ndp->ni_vp)
 978                         vrele(ndp->ni_dvp);
 979                 else
 980                         vput(ndp->ni_dvp);
 981                 vrele(ndp->ni_vp);
 982                 error = EEXIST;
 983                 goto out;
 984         }
 985
 986         error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 987             &nvap->na_vattr, pathcp);
 988         vput(ndp->ni_dvp);
 989         vrele(ndp->ni_startdir);
 990         nfsvno_relpathbuf(ndp);
 991         /*
 992          * Although FreeBSD still had the lookup code in
 993          * it for 7/current, there doesn't seem to be any
 994          * point, since VOP_SYMLINK() returns the ni_vp.
 995          * Just vput it for v2.
 996          */
 997         if (!not_v2 && !error)
 998                 vput(ndp->ni_vp);
 999
1000 out:
1001         NFSEXITCODE(error);
1002         return (error);
1003 }
1004
1005 /*
1006  * Parse symbolic link arguments.
1007  * This function has an ugly side effect. It will MALLOC() an area for
1008  * the symlink and set iov_base to point to it, only if it succeeds.
1009  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1010  * be FREE'd later.
1011  */
1012 int
1013 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1014     struct thread *p, char **pathcpp, int *lenp)
1015 {
1016         u_int32_t *tl;
1017         char *pathcp = NULL;
1018         int error = 0, len;
1019         struct nfsv2_sattr *sp;
1020
1021         *pathcpp = NULL;
1022         *lenp = 0;
1023         if ((nd->nd_flag & ND_NFSV3) &&
1024             (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1025                 goto nfsmout;
1026         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1027         len = fxdr_unsigned(int, *tl);
1028         if (len > NFS_MAXPATHLEN || len <= 0) {
1029                 error = EBADRPC;
1030                 goto nfsmout;
1031         }
1032         MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1033         error = nfsrv_mtostr(nd, pathcp, len);
1034         if (error)
1035                 goto nfsmout;
1036         if (nd->nd_flag & ND_NFSV2) {
1037                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1038                 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1039         }
1040         *pathcpp = pathcp;
1041         *lenp = len;
1042         NFSEXITCODE2(0, nd);
1043         return (0);
1044 nfsmout:
1045         if (pathcp)
1046                 free(pathcp, M_TEMP);
1047         NFSEXITCODE2(error, nd);
1048         return (error);
1049 }
1050
1051 /*
1052  * Remove a non-directory object.
1053  */
1054 int
1055 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1056     struct thread *p, struct nfsexstuff *exp)
1057 {
1058         struct vnode *vp;
1059         int error = 0;
1060
1061         vp = ndp->ni_vp;
1062         if (vp->v_type == VDIR)
1063                 error = NFSERR_ISDIR;
1064         else if (is_v4)
1065                 error = nfsrv_checkremove(vp, 1, p);
1066         if (!error)
1067                 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1068         if (ndp->ni_dvp == vp)
1069                 vrele(ndp->ni_dvp);
1070         else
1071                 vput(ndp->ni_dvp);
1072         vput(vp);
1073         if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1074                 nfsvno_relpathbuf(ndp);
1075         NFSEXITCODE(error);
1076         return (error);
1077 }
1078
1079 /*
1080  * Remove a directory.
1081  */
1082 int
1083 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1084     struct thread *p, struct nfsexstuff *exp)
1085 {
1086         struct vnode *vp;
1087         int error = 0;
1088
1089         vp = ndp->ni_vp;
1090         if (vp->v_type != VDIR) {
1091                 error = ENOTDIR;
1092                 goto out;
1093         }
1094         /*
1095          * No rmdir "." please.
1096          */
1097         if (ndp->ni_dvp == vp) {
1098                 error = EINVAL;
1099                 goto out;
1100         }
1101         /*
1102          * The root of a mounted filesystem cannot be deleted.
1103          */
1104         if (vp->v_vflag & VV_ROOT)
1105                 error = EBUSY;
1106 out:
1107         if (!error)
1108                 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1109         if (ndp->ni_dvp == vp)
1110                 vrele(ndp->ni_dvp);
1111         else
1112                 vput(ndp->ni_dvp);
1113         vput(vp);
1114         if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1115                 nfsvno_relpathbuf(ndp);
1116         NFSEXITCODE(error);
1117         return (error);
1118 }
1119
1120 /*
1121  * Rename vnode op.
1122  */
1123 int
1124 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1125     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1126 {
1127         struct vnode *fvp, *tvp, *tdvp;
1128         int error = 0;
1129
1130         fvp = fromndp->ni_vp;
1131         if (ndstat) {
1132                 vrele(fromndp->ni_dvp);
1133                 vrele(fvp);
1134                 error = ndstat;
1135                 goto out1;
1136         }
1137         tdvp = tondp->ni_dvp;
1138         tvp = tondp->ni_vp;
1139         if (tvp != NULL) {
1140                 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1141                         error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1142                         goto out;
1143                 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1144                         error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1145                         goto out;
1146                 }
1147                 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1148                         error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1149                         goto out;
1150                 }
1151
1152                 /*
1153                  * A rename to '.' or '..' results in a prematurely
1154                  * unlocked vnode on FreeBSD5, so I'm just going to fail that
1155                  * here.
1156                  */
1157                 if ((tondp->ni_cnd.cn_namelen == 1 &&
1158                      tondp->ni_cnd.cn_nameptr[0] == '.') ||
1159                     (tondp->ni_cnd.cn_namelen == 2 &&
1160                      tondp->ni_cnd.cn_nameptr[0] == '.' &&
1161                      tondp->ni_cnd.cn_nameptr[1] == '.')) {
1162                         error = EINVAL;
1163                         goto out;
1164                 }
1165         }
1166         if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1167                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1168                 goto out;
1169         }
1170         if (fvp->v_mount != tdvp->v_mount) {
1171                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1172                 goto out;
1173         }
1174         if (fvp == tdvp) {
1175                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1176                 goto out;
1177         }
1178         if (fvp == tvp) {
1179                 /*
1180                  * If source and destination are the same, there is nothing to
1181                  * do. Set error to -1 to indicate this.
1182                  */
1183                 error = -1;
1184                 goto out;
1185         }
1186         if (ndflag & ND_NFSV4) {
1187                 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1188                         error = nfsrv_checkremove(fvp, 0, p);
1189                         NFSVOPUNLOCK(fvp, 0);
1190                 } else
1191                         error = EPERM;
1192                 if (tvp && !error)
1193                         error = nfsrv_checkremove(tvp, 1, p);
1194         } else {
1195                 /*
1196                  * For NFSv2 and NFSv3, try to get rid of the delegation, so
1197                  * that the NFSv4 client won't be confused by the rename.
1198                  * Since nfsd_recalldelegation() can only be called on an
1199                  * unlocked vnode at this point and fvp is the file that will
1200                  * still exist after the rename, just do fvp.
1201                  */
1202                 nfsd_recalldelegation(fvp, p);
1203         }
1204 out:
1205         if (!error) {
1206                 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1207                     &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1208                     &tondp->ni_cnd);
1209         } else {
1210                 if (tdvp == tvp)
1211                         vrele(tdvp);
1212                 else
1213                         vput(tdvp);
1214                 if (tvp)
1215                         vput(tvp);
1216                 vrele(fromndp->ni_dvp);
1217                 vrele(fvp);
1218                 if (error == -1)
1219                         error = 0;
1220         }
1221         vrele(tondp->ni_startdir);
1222         nfsvno_relpathbuf(tondp);
1223 out1:
1224         vrele(fromndp->ni_startdir);
1225         nfsvno_relpathbuf(fromndp);
1226         NFSEXITCODE(error);
1227         return (error);
1228 }
1229
1230 /*
1231  * Link vnode op.
1232  */
1233 int
1234 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1235     struct thread *p, struct nfsexstuff *exp)
1236 {
1237         struct vnode *xp;
1238         int error = 0;
1239
1240         xp = ndp->ni_vp;
1241         if (xp != NULL) {
1242                 error = EEXIST;
1243         } else {
1244                 xp = ndp->ni_dvp;
1245                 if (vp->v_mount != xp->v_mount)
1246                         error = EXDEV;
1247         }
1248         if (!error) {
1249                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1250                 if ((vp->v_iflag & VI_DOOMED) == 0)
1251                         error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1252                 else
1253                         error = EPERM;
1254                 if (ndp->ni_dvp == vp)
1255                         vrele(ndp->ni_dvp);
1256                 else
1257                         vput(ndp->ni_dvp);
1258                 NFSVOPUNLOCK(vp, 0);
1259         } else {
1260                 if (ndp->ni_dvp == ndp->ni_vp)
1261                         vrele(ndp->ni_dvp);
1262                 else
1263                         vput(ndp->ni_dvp);
1264                 if (ndp->ni_vp)
1265                         vrele(ndp->ni_vp);
1266         }
1267         nfsvno_relpathbuf(ndp);
1268         NFSEXITCODE(error);
1269         return (error);
1270 }
1271
1272 /*
1273  * Do the fsync() appropriate for the commit.
1274  */
1275 int
1276 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1277     struct thread *td)
1278 {
1279         int error = 0;
1280
1281         /*
1282          * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1283          * file is done.  At this time VOP_FSYNC does not accept offset and
1284          * byte count parameters so call VOP_FSYNC the whole file for now.
1285          * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1286          * File systems that do not use the buffer cache (as indicated
1287          * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1288          */
1289         if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1290             (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1291                 /*
1292                  * Give up and do the whole thing
1293                  */
1294                 if (vp->v_object &&
1295                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1296                         VM_OBJECT_WLOCK(vp->v_object);
1297                         vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1298                         VM_OBJECT_WUNLOCK(vp->v_object);
1299                 }
1300                 error = VOP_FSYNC(vp, MNT_WAIT, td);
1301         } else {
1302                 /*
1303                  * Locate and synchronously write any buffers that fall
1304                  * into the requested range.  Note:  we are assuming that
1305                  * f_iosize is a power of 2.
1306                  */
1307                 int iosize = vp->v_mount->mnt_stat.f_iosize;
1308                 int iomask = iosize - 1;
1309                 struct bufobj *bo;
1310                 daddr_t lblkno;
1311
1312                 /*
1313                  * Align to iosize boundary, super-align to page boundary.
1314                  */
1315                 if (off & iomask) {
1316                         cnt += off & iomask;
1317                         off &= ~(u_quad_t)iomask;
1318                 }
1319                 if (off & PAGE_MASK) {
1320                         cnt += off & PAGE_MASK;
1321                         off &= ~(u_quad_t)PAGE_MASK;
1322                 }
1323                 lblkno = off / iosize;
1324
1325                 if (vp->v_object &&
1326                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1327                         VM_OBJECT_WLOCK(vp->v_object);
1328                         vm_object_page_clean(vp->v_object, off, off + cnt,
1329                             OBJPC_SYNC);
1330                         VM_OBJECT_WUNLOCK(vp->v_object);
1331                 }
1332
1333                 bo = &vp->v_bufobj;
1334                 BO_LOCK(bo);
1335                 while (cnt > 0) {
1336                         struct buf *bp;
1337
1338                         /*
1339                          * If we have a buffer and it is marked B_DELWRI we
1340                          * have to lock and write it.  Otherwise the prior
1341                          * write is assumed to have already been committed.
1342                          *
1343                          * gbincore() can return invalid buffers now so we
1344                          * have to check that bit as well (though B_DELWRI
1345                          * should not be set if B_INVAL is set there could be
1346                          * a race here since we haven't locked the buffer).
1347                          */
1348                         if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1349                                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1350                                     LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1351                                         BO_LOCK(bo);
1352                                         continue; /* retry */
1353                                 }
1354                                 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1355                                     B_DELWRI) {
1356                                         bremfree(bp);
1357                                         bp->b_flags &= ~B_ASYNC;
1358                                         bwrite(bp);
1359                                         ++nfs_commit_miss;
1360                                 } else
1361                                         BUF_UNLOCK(bp);
1362                                 BO_LOCK(bo);
1363                         }
1364                         ++nfs_commit_blks;
1365                         if (cnt < iosize)
1366                                 break;
1367                         cnt -= iosize;
1368                         ++lblkno;
1369                 }
1370                 BO_UNLOCK(bo);
1371         }
1372         NFSEXITCODE(error);
1373         return (error);
1374 }
1375
1376 /*
1377  * Statfs vnode op.
1378  */
1379 int
1380 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1381 {
1382         int error;
1383
1384         error = VFS_STATFS(vp->v_mount, sf);
1385         if (error == 0) {
1386                 /*
1387                  * Since NFS handles these values as unsigned on the
1388                  * wire, there is no way to represent negative values,
1389                  * so set them to 0. Without this, they will appear
1390                  * to be very large positive values for clients like
1391                  * Solaris10.
1392                  */
1393                 if (sf->f_bavail < 0)
1394                         sf->f_bavail = 0;
1395                 if (sf->f_ffree < 0)
1396                         sf->f_ffree = 0;
1397         }
1398         NFSEXITCODE(error);
1399         return (error);
1400 }
1401
1402 /*
1403  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1404  * must handle nfsrv_opencheck() calls after any other access checks.
1405  */
1406 void
1407 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1408     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1409     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1410     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1411     struct nfsexstuff *exp, struct vnode **vpp)
1412 {
1413         struct vnode *vp = NULL;
1414         u_quad_t tempsize;
1415         struct nfsexstuff nes;
1416
1417         if (ndp->ni_vp == NULL)
1418                 nd->nd_repstat = nfsrv_opencheck(clientid,
1419                     stateidp, stp, NULL, nd, p, nd->nd_repstat);
1420         if (!nd->nd_repstat) {
1421                 if (ndp->ni_vp == NULL) {
1422                         vrele(ndp->ni_startdir);
1423                         nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1424                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1425                         vput(ndp->ni_dvp);
1426                         nfsvno_relpathbuf(ndp);
1427                         if (!nd->nd_repstat) {
1428                                 if (*exclusive_flagp) {
1429                                         *exclusive_flagp = 0;
1430                                         NFSVNO_ATTRINIT(nvap);
1431                                         nvap->na_atime.tv_sec = cverf[0];
1432                                         nvap->na_atime.tv_nsec = cverf[1];
1433                                         nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1434                                             &nvap->na_vattr, cred);
1435                                         if (nd->nd_repstat != 0) {
1436                                                 vput(ndp->ni_vp);
1437                                                 ndp->ni_vp = NULL;
1438                                                 nd->nd_repstat = NFSERR_NOTSUPP;
1439                                         }
1440                                 } else {
1441                                         nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1442                                             aclp, p, attrbitp, exp);
1443                                 }
1444                         }
1445                         vp = ndp->ni_vp;
1446                 } else {
1447                         if (ndp->ni_startdir)
1448                                 vrele(ndp->ni_startdir);
1449                         nfsvno_relpathbuf(ndp);
1450                         vp = ndp->ni_vp;
1451                         if (create == NFSV4OPEN_CREATE) {
1452                                 if (ndp->ni_dvp == vp)
1453                                         vrele(ndp->ni_dvp);
1454                                 else
1455                                         vput(ndp->ni_dvp);
1456                         }
1457                         if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1458                                 if (ndp->ni_cnd.cn_flags & RDONLY)
1459                                         NFSVNO_SETEXRDONLY(&nes);
1460                                 else
1461                                         NFSVNO_EXINIT(&nes);
1462                                 nd->nd_repstat = nfsvno_accchk(vp,
1463                                     VWRITE, cred, &nes, p,
1464                                     NFSACCCHK_NOOVERRIDE,
1465                                     NFSACCCHK_VPISLOCKED, NULL);
1466                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1467                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1468                                 if (!nd->nd_repstat) {
1469                                         tempsize = nvap->na_size;
1470                                         NFSVNO_ATTRINIT(nvap);
1471                                         nvap->na_size = tempsize;
1472                                         nd->nd_repstat = VOP_SETATTR(vp,
1473                                             &nvap->na_vattr, cred);
1474                                 }
1475                         } else if (vp->v_type == VREG) {
1476                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1477                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1478                         }
1479                 }
1480         } else {
1481                 if (ndp->ni_cnd.cn_flags & HASBUF)
1482                         nfsvno_relpathbuf(ndp);
1483                 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1484                         vrele(ndp->ni_startdir);
1485                         if (ndp->ni_dvp == ndp->ni_vp)
1486                                 vrele(ndp->ni_dvp);
1487                         else
1488                                 vput(ndp->ni_dvp);
1489                         if (ndp->ni_vp)
1490                                 vput(ndp->ni_vp);
1491                 }
1492         }
1493         *vpp = vp;
1494
1495         NFSEXITCODE2(0, nd);
1496 }
1497
1498 /*
1499  * Updates the file rev and sets the mtime and ctime
1500  * to the current clock time, returning the va_filerev and va_Xtime
1501  * values.
1502  * Return ESTALE to indicate the vnode is VI_DOOMED.
1503  */
1504 int
1505 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1506     struct ucred *cred, struct thread *p)
1507 {
1508         struct vattr va;
1509
1510         VATTR_NULL(&va);
1511         vfs_timestamp(&va.va_mtime);
1512         if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1513                 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1514                 if ((vp->v_iflag & VI_DOOMED) != 0)
1515                         return (ESTALE);
1516         }
1517         (void) VOP_SETATTR(vp, &va, cred);
1518         (void) nfsvno_getattr(vp, nvap, cred, p, 1);
1519         return (0);
1520 }
1521
1522 /*
1523  * Glue routine to nfsv4_fillattr().
1524  */
1525 int
1526 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1527     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1528     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1529     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1530 {
1531         int error;
1532
1533         error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1534             attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1535             mounted_on_fileno);
1536         NFSEXITCODE2(0, nd);
1537         return (error);
1538 }
1539
1540 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1541 /*
1542  * nfs readdir service
1543  * - mallocs what it thinks is enough to read
1544  *      count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1545  * - calls VOP_READDIR()
1546  * - loops around building the reply
1547  *      if the output generated exceeds count break out of loop
1548  *      The NFSM_CLGET macro is used here so that the reply will be packed
1549  *      tightly in mbuf clusters.
1550  * - it trims out records with d_fileno == 0
1551  *      this doesn't matter for Unix clients, but they might confuse clients
1552  *      for other os'.
1553  * - it trims out records with d_type == DT_WHT
1554  *      these cannot be seen through NFS (unless we extend the protocol)
1555  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1556  * PS: The NFS protocol spec. does not clarify what the "count" byte
1557  *      argument is a count of.. just name strings and file id's or the
1558  *      entire reply rpc or ...
1559  *      I tried just file name and id sizes and it confused the Sun client,
1560  *      so I am using the full rpc size now. The "paranoia.." comment refers
1561  *      to including the status longwords that are not a part of the dir.
1562  *      "entry" structures, but are in the rpc.
1563  */
1564 int
1565 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1566     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1567 {
1568         struct dirent *dp;
1569         u_int32_t *tl;
1570         int dirlen;
1571         char *cpos, *cend, *rbuf;
1572         struct nfsvattr at;
1573         int nlen, error = 0, getret = 1;
1574         int siz, cnt, fullsiz, eofflag, ncookies;
1575         u_int64_t off, toff, verf;
1576         u_long *cookies = NULL, *cookiep;
1577         struct uio io;
1578         struct iovec iv;
1579         int is_ufs;
1580
1581         if (nd->nd_repstat) {
1582                 nfsrv_postopattr(nd, getret, &at);
1583                 goto out;
1584         }
1585         if (nd->nd_flag & ND_NFSV2) {
1586                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1587                 off = fxdr_unsigned(u_quad_t, *tl++);
1588         } else {
1589                 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1590                 off = fxdr_hyper(tl);
1591                 tl += 2;
1592                 verf = fxdr_hyper(tl);
1593                 tl += 2;
1594         }
1595         toff = off;
1596         cnt = fxdr_unsigned(int, *tl);
1597         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1598                 cnt = NFS_SRVMAXDATA(nd);
1599         siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1600         fullsiz = siz;
1601         if (nd->nd_flag & ND_NFSV3) {
1602                 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1603                     p, 1);
1604 #if 0
1605                 /*
1606                  * va_filerev is not sufficient as a cookie verifier,
1607                  * since it is not supposed to change when entries are
1608                  * removed/added unless that offset cookies returned to
1609                  * the client are no longer valid.
1610                  */
1611                 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1612                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1613 #endif
1614         }
1615         if (!nd->nd_repstat && vp->v_type != VDIR)
1616                 nd->nd_repstat = NFSERR_NOTDIR;
1617         if (nd->nd_repstat == 0 && cnt == 0) {
1618                 if (nd->nd_flag & ND_NFSV2)
1619                         /* NFSv2 does not have NFSERR_TOOSMALL */
1620                         nd->nd_repstat = EPERM;
1621                 else
1622                         nd->nd_repstat = NFSERR_TOOSMALL;
1623         }
1624         if (!nd->nd_repstat)
1625                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1626                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1627                     NFSACCCHK_VPISLOCKED, NULL);
1628         if (nd->nd_repstat) {
1629                 vput(vp);
1630                 if (nd->nd_flag & ND_NFSV3)
1631                         nfsrv_postopattr(nd, getret, &at);
1632                 goto out;
1633         }
1634         is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1635         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1636 again:
1637         eofflag = 0;
1638         if (cookies) {
1639                 free((caddr_t)cookies, M_TEMP);
1640                 cookies = NULL;
1641         }
1642
1643         iv.iov_base = rbuf;
1644         iv.iov_len = siz;
1645         io.uio_iov = &iv;
1646         io.uio_iovcnt = 1;
1647         io.uio_offset = (off_t)off;
1648         io.uio_resid = siz;
1649         io.uio_segflg = UIO_SYSSPACE;
1650         io.uio_rw = UIO_READ;
1651         io.uio_td = NULL;
1652         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1653             &cookies);
1654         off = (u_int64_t)io.uio_offset;
1655         if (io.uio_resid)
1656                 siz -= io.uio_resid;
1657
1658         if (!cookies && !nd->nd_repstat)
1659                 nd->nd_repstat = NFSERR_PERM;
1660         if (nd->nd_flag & ND_NFSV3) {
1661                 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1662                 if (!nd->nd_repstat)
1663                         nd->nd_repstat = getret;
1664         }
1665
1666         /*
1667          * Handles the failed cases. nd->nd_repstat == 0 past here.
1668          */
1669         if (nd->nd_repstat) {
1670                 vput(vp);
1671                 free((caddr_t)rbuf, M_TEMP);
1672                 if (cookies)
1673                         free((caddr_t)cookies, M_TEMP);
1674                 if (nd->nd_flag & ND_NFSV3)
1675                         nfsrv_postopattr(nd, getret, &at);
1676                 goto out;
1677         }
1678         /*
1679          * If nothing read, return eof
1680          * rpc reply
1681          */
1682         if (siz == 0) {
1683                 vput(vp);
1684                 if (nd->nd_flag & ND_NFSV2) {
1685                         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1686                 } else {
1687                         nfsrv_postopattr(nd, getret, &at);
1688                         NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1689                         txdr_hyper(at.na_filerev, tl);
1690                         tl += 2;
1691                 }
1692                 *tl++ = newnfs_false;
1693                 *tl = newnfs_true;
1694                 FREE((caddr_t)rbuf, M_TEMP);
1695                 FREE((caddr_t)cookies, M_TEMP);
1696                 goto out;
1697         }
1698
1699         /*
1700          * Check for degenerate cases of nothing useful read.
1701          * If so go try again
1702          */
1703         cpos = rbuf;
1704         cend = rbuf + siz;
1705         dp = (struct dirent *)cpos;
1706         cookiep = cookies;
1707
1708         /*
1709          * For some reason FreeBSD's ufs_readdir() chooses to back the
1710          * directory offset up to a block boundary, so it is necessary to
1711          * skip over the records that precede the requested offset. This
1712          * requires the assumption that file offset cookies monotonically
1713          * increase.
1714          */
1715         while (cpos < cend && ncookies > 0 &&
1716             (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1717              (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1718                 cpos += dp->d_reclen;
1719                 dp = (struct dirent *)cpos;
1720                 cookiep++;
1721                 ncookies--;
1722         }
1723         if (cpos >= cend || ncookies == 0) {
1724                 siz = fullsiz;
1725                 toff = off;
1726                 goto again;
1727         }
1728         vput(vp);
1729
1730         /*
1731          * dirlen is the size of the reply, including all XDR and must
1732          * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1733          * if the XDR should be included in "count", but to be safe, we do.
1734          * (Include the two booleans at the end of the reply in dirlen now.)
1735          */
1736         if (nd->nd_flag & ND_NFSV3) {
1737                 nfsrv_postopattr(nd, getret, &at);
1738                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1739                 txdr_hyper(at.na_filerev, tl);
1740                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1741         } else {
1742                 dirlen = 2 * NFSX_UNSIGNED;
1743         }
1744
1745         /* Loop through the records and build reply */
1746         while (cpos < cend && ncookies > 0) {
1747                 nlen = dp->d_namlen;
1748                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1749                         nlen <= NFS_MAXNAMLEN) {
1750                         if (nd->nd_flag & ND_NFSV3)
1751                                 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1752                         else
1753                                 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1754                         if (dirlen > cnt) {
1755                                 eofflag = 0;
1756                                 break;
1757                         }
1758
1759                         /*
1760                          * Build the directory record xdr from
1761                          * the dirent entry.
1762                          */
1763                         if (nd->nd_flag & ND_NFSV3) {
1764                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1765                                 *tl++ = newnfs_true;
1766                                 *tl++ = 0;
1767                         } else {
1768                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1769                                 *tl++ = newnfs_true;
1770                         }
1771                         *tl = txdr_unsigned(dp->d_fileno);
1772                         (void) nfsm_strtom(nd, dp->d_name, nlen);
1773                         if (nd->nd_flag & ND_NFSV3) {
1774                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1775                                 *tl++ = 0;
1776                         } else
1777                                 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1778                         *tl = txdr_unsigned(*cookiep);
1779                 }
1780                 cpos += dp->d_reclen;
1781                 dp = (struct dirent *)cpos;
1782                 cookiep++;
1783                 ncookies--;
1784         }
1785         if (cpos < cend)
1786                 eofflag = 0;
1787         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1788         *tl++ = newnfs_false;
1789         if (eofflag)
1790                 *tl = newnfs_true;
1791         else
1792                 *tl = newnfs_false;
1793         FREE((caddr_t)rbuf, M_TEMP);
1794         FREE((caddr_t)cookies, M_TEMP);
1795
1796 out:
1797         NFSEXITCODE2(0, nd);
1798         return (0);
1799 nfsmout:
1800         vput(vp);
1801         NFSEXITCODE2(error, nd);
1802         return (error);
1803 }
1804
1805 /*
1806  * Readdirplus for V3 and Readdir for V4.
1807  */
1808 int
1809 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1810     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1811 {
1812         struct dirent *dp;
1813         u_int32_t *tl;
1814         int dirlen;
1815         char *cpos, *cend, *rbuf;
1816         struct vnode *nvp;
1817         fhandle_t nfh;
1818         struct nfsvattr nva, at, *nvap = &nva;
1819         struct mbuf *mb0, *mb1;
1820         struct nfsreferral *refp;
1821         int nlen, r, error = 0, getret = 1, usevget = 1;
1822         int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1823         caddr_t bpos0, bpos1;
1824         u_int64_t off, toff, verf;
1825         u_long *cookies = NULL, *cookiep;
1826         nfsattrbit_t attrbits, rderrbits, savbits;
1827         struct uio io;
1828         struct iovec iv;
1829         struct componentname cn;
1830         int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
1831         struct mount *mp, *new_mp;
1832         uint64_t mounted_on_fileno;
1833
1834         if (nd->nd_repstat) {
1835                 nfsrv_postopattr(nd, getret, &at);
1836                 goto out;
1837         }
1838         NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1839         off = fxdr_hyper(tl);
1840         toff = off;
1841         tl += 2;
1842         verf = fxdr_hyper(tl);
1843         tl += 2;
1844         siz = fxdr_unsigned(int, *tl++);
1845         cnt = fxdr_unsigned(int, *tl);
1846
1847         /*
1848          * Use the server's maximum data transfer size as the upper bound
1849          * on reply datalen.
1850          */
1851         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1852                 cnt = NFS_SRVMAXDATA(nd);
1853
1854         /*
1855          * siz is a "hint" of how much directory information (name, fileid,
1856          * cookie) should be in the reply. At least one client "hints" 0,
1857          * so I set it to cnt for that case. I also round it up to the
1858          * next multiple of DIRBLKSIZ.
1859          */
1860         if (siz <= 0)
1861                 siz = cnt;
1862         siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1863
1864         if (nd->nd_flag & ND_NFSV4) {
1865                 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1866                 if (error)
1867                         goto nfsmout;
1868                 NFSSET_ATTRBIT(&savbits, &attrbits);
1869                 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1870                 NFSZERO_ATTRBIT(&rderrbits);
1871                 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1872         } else {
1873                 NFSZERO_ATTRBIT(&attrbits);
1874         }
1875         fullsiz = siz;
1876         nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1877         if (!nd->nd_repstat) {
1878             if (off && verf != at.na_filerev) {
1879                 /*
1880                  * va_filerev is not sufficient as a cookie verifier,
1881                  * since it is not supposed to change when entries are
1882                  * removed/added unless that offset cookies returned to
1883                  * the client are no longer valid.
1884                  */
1885 #if 0
1886                 if (nd->nd_flag & ND_NFSV4) {
1887                         nd->nd_repstat = NFSERR_NOTSAME;
1888                 } else {
1889                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1890                 }
1891 #endif
1892             } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1893                 nd->nd_repstat = NFSERR_BAD_COOKIE;
1894             }
1895         }
1896         if (!nd->nd_repstat && vp->v_type != VDIR)
1897                 nd->nd_repstat = NFSERR_NOTDIR;
1898         if (!nd->nd_repstat && cnt == 0)
1899                 nd->nd_repstat = NFSERR_TOOSMALL;
1900         if (!nd->nd_repstat)
1901                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1902                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1903                     NFSACCCHK_VPISLOCKED, NULL);
1904         if (nd->nd_repstat) {
1905                 vput(vp);
1906                 if (nd->nd_flag & ND_NFSV3)
1907                         nfsrv_postopattr(nd, getret, &at);
1908                 goto out;
1909         }
1910         is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1911         is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
1912
1913         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1914 again:
1915         eofflag = 0;
1916         if (cookies) {
1917                 free((caddr_t)cookies, M_TEMP);
1918                 cookies = NULL;
1919         }
1920
1921         iv.iov_base = rbuf;
1922         iv.iov_len = siz;
1923         io.uio_iov = &iv;
1924         io.uio_iovcnt = 1;
1925         io.uio_offset = (off_t)off;
1926         io.uio_resid = siz;
1927         io.uio_segflg = UIO_SYSSPACE;
1928         io.uio_rw = UIO_READ;
1929         io.uio_td = NULL;
1930         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1931             &cookies);
1932         off = (u_int64_t)io.uio_offset;
1933         if (io.uio_resid)
1934                 siz -= io.uio_resid;
1935
1936         getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1937
1938         if (!cookies && !nd->nd_repstat)
1939                 nd->nd_repstat = NFSERR_PERM;
1940         if (!nd->nd_repstat)
1941                 nd->nd_repstat = getret;
1942         if (nd->nd_repstat) {
1943                 vput(vp);
1944                 if (cookies)
1945                         free((caddr_t)cookies, M_TEMP);
1946                 free((caddr_t)rbuf, M_TEMP);
1947                 if (nd->nd_flag & ND_NFSV3)
1948                         nfsrv_postopattr(nd, getret, &at);
1949                 goto out;
1950         }
1951         /*
1952          * If nothing read, return eof
1953          * rpc reply
1954          */
1955         if (siz == 0) {
1956                 vput(vp);
1957                 if (nd->nd_flag & ND_NFSV3)
1958                         nfsrv_postopattr(nd, getret, &at);
1959                 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1960                 txdr_hyper(at.na_filerev, tl);
1961                 tl += 2;
1962                 *tl++ = newnfs_false;
1963                 *tl = newnfs_true;
1964                 free((caddr_t)cookies, M_TEMP);
1965                 free((caddr_t)rbuf, M_TEMP);
1966                 goto out;
1967         }
1968
1969         /*
1970          * Check for degenerate cases of nothing useful read.
1971          * If so go try again
1972          */
1973         cpos = rbuf;
1974         cend = rbuf + siz;
1975         dp = (struct dirent *)cpos;
1976         cookiep = cookies;
1977
1978         /*
1979          * For some reason FreeBSD's ufs_readdir() chooses to back the
1980          * directory offset up to a block boundary, so it is necessary to
1981          * skip over the records that precede the requested offset. This
1982          * requires the assumption that file offset cookies monotonically
1983          * increase.
1984          */
1985         while (cpos < cend && ncookies > 0 &&
1986           (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1987            (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
1988            ((nd->nd_flag & ND_NFSV4) &&
1989             ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1990              (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1991                 cpos += dp->d_reclen;
1992                 dp = (struct dirent *)cpos;
1993                 cookiep++;
1994                 ncookies--;
1995         }
1996         if (cpos >= cend || ncookies == 0) {
1997                 siz = fullsiz;
1998                 toff = off;
1999                 goto again;
2000         }
2001
2002         /*
2003          * Busy the file system so that the mount point won't go away
2004          * and, as such, VFS_VGET() can be used safely.
2005          */
2006         mp = vp->v_mount;
2007         vfs_ref(mp);
2008         NFSVOPUNLOCK(vp, 0);
2009         nd->nd_repstat = vfs_busy(mp, 0);
2010         vfs_rel(mp);
2011         if (nd->nd_repstat != 0) {
2012                 vrele(vp);
2013                 free(cookies, M_TEMP);
2014                 free(rbuf, M_TEMP);
2015                 if (nd->nd_flag & ND_NFSV3)
2016                         nfsrv_postopattr(nd, getret, &at);
2017                 goto out;
2018         }
2019
2020         /*
2021          * For now ZFS requires VOP_LOOKUP as a workaround.  Until ino_t is changed
2022          * to 64 bit type a ZFS filesystem with over 1 billion files in it
2023          * will suffer from 64bit -> 32bit truncation.
2024          */
2025         if (is_zfs == 1)
2026                 usevget = 0;
2027
2028         cn.cn_nameiop = LOOKUP;
2029         cn.cn_lkflags = LK_SHARED | LK_RETRY;
2030         cn.cn_cred = nd->nd_cred;
2031         cn.cn_thread = p;
2032
2033         /*
2034          * Save this position, in case there is an error before one entry
2035          * is created.
2036          */
2037         mb0 = nd->nd_mb;
2038         bpos0 = nd->nd_bpos;
2039
2040         /*
2041          * Fill in the first part of the reply.
2042          * dirlen is the reply length in bytes and cannot exceed cnt.
2043          * (Include the two booleans at the end of the reply in dirlen now,
2044          *  so we recognize when we have exceeded cnt.)
2045          */
2046         if (nd->nd_flag & ND_NFSV3) {
2047                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2048                 nfsrv_postopattr(nd, getret, &at);
2049         } else {
2050                 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2051         }
2052         NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2053         txdr_hyper(at.na_filerev, tl);
2054
2055         /*
2056          * Save this position, in case there is an empty reply needed.
2057          */
2058         mb1 = nd->nd_mb;
2059         bpos1 = nd->nd_bpos;
2060
2061         /* Loop through the records and build reply */
2062         entrycnt = 0;
2063         while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2064                 nlen = dp->d_namlen;
2065                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2066                     nlen <= NFS_MAXNAMLEN &&
2067                     ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2068                      (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2069                       || (nlen == 1 && dp->d_name[0] != '.'))) {
2070                         /*
2071                          * Save the current position in the reply, in case
2072                          * this entry exceeds cnt.
2073                          */
2074                         mb1 = nd->nd_mb;
2075                         bpos1 = nd->nd_bpos;
2076
2077                         /*
2078                          * For readdir_and_lookup get the vnode using
2079                          * the file number.
2080                          */
2081                         nvp = NULL;
2082                         refp = NULL;
2083                         r = 0;
2084                         at_root = 0;
2085                         needs_unbusy = 0;
2086                         new_mp = mp;
2087                         mounted_on_fileno = (uint64_t)dp->d_fileno;
2088                         if ((nd->nd_flag & ND_NFSV3) ||
2089                             NFSNONZERO_ATTRBIT(&savbits)) {
2090                                 if (nd->nd_flag & ND_NFSV4)
2091                                         refp = nfsv4root_getreferral(NULL,
2092                                             vp, dp->d_fileno);
2093                                 if (refp == NULL) {
2094                                         if (usevget)
2095                                                 r = VFS_VGET(mp, dp->d_fileno,
2096                                                     LK_SHARED, &nvp);
2097                                         else
2098                                                 r = EOPNOTSUPP;
2099                                         if (r == EOPNOTSUPP) {
2100                                                 usevget = 0;
2101                                                 cn.cn_nameptr = dp->d_name;
2102                                                 cn.cn_namelen = nlen;
2103                                                 cn.cn_flags = ISLASTCN |
2104                                                     NOFOLLOW | LOCKLEAF;
2105                                                 if (nlen == 2 &&
2106                                                     dp->d_name[0] == '.' &&
2107                                                     dp->d_name[1] == '.')
2108                                                         cn.cn_flags |=
2109                                                             ISDOTDOT;
2110                                                 if (NFSVOPLOCK(vp, LK_SHARED)
2111                                                     != 0) {
2112                                                         nd->nd_repstat = EPERM;
2113                                                         break;
2114                                                 }
2115                                                 if ((vp->v_vflag & VV_ROOT) != 0
2116                                                     && (cn.cn_flags & ISDOTDOT)
2117                                                     != 0) {
2118                                                         vref(vp);
2119                                                         nvp = vp;
2120                                                         r = 0;
2121                                                 } else {
2122                                                         r = VOP_LOOKUP(vp, &nvp,
2123                                                             &cn);
2124                                                         if (vp != nvp)
2125                                                                 NFSVOPUNLOCK(vp,
2126                                                                     0);
2127                                                 }
2128                                         }
2129
2130                                         /*
2131                                          * For NFSv4, check to see if nvp is
2132                                          * a mount point and get the mount
2133                                          * point vnode, as required.
2134                                          */
2135                                         if (r == 0 &&
2136                                             nfsrv_enable_crossmntpt != 0 &&
2137                                             (nd->nd_flag & ND_NFSV4) != 0 &&
2138                                             nvp->v_type == VDIR &&
2139                                             nvp->v_mountedhere != NULL) {
2140                                                 new_mp = nvp->v_mountedhere;
2141                                                 r = vfs_busy(new_mp, 0);
2142                                                 vput(nvp);
2143                                                 nvp = NULL;
2144                                                 if (r == 0) {
2145                                                         r = VFS_ROOT(new_mp,
2146                                                             LK_SHARED, &nvp);
2147                                                         needs_unbusy = 1;
2148                                                         if (r == 0)
2149                                                                 at_root = 1;
2150                                                 }
2151                                         }
2152                                 }
2153                                 if (!r) {
2154                                     if (refp == NULL &&
2155                                         ((nd->nd_flag & ND_NFSV3) ||
2156                                          NFSNONZERO_ATTRBIT(&attrbits))) {
2157                                         r = nfsvno_getfh(nvp, &nfh, p);
2158                                         if (!r)
2159                                             r = nfsvno_getattr(nvp, nvap,
2160                                                 nd->nd_cred, p, 1);
2161                                         if (r == 0 && is_zfs == 1 &&
2162                                             nfsrv_enable_crossmntpt != 0 &&
2163                                             (nd->nd_flag & ND_NFSV4) != 0 &&
2164                                             nvp->v_type == VDIR &&
2165                                             vp->v_mount != nvp->v_mount) {
2166                                             /*
2167                                              * For a ZFS snapshot, there is a
2168                                              * pseudo mount that does not set
2169                                              * v_mountedhere, so it needs to
2170                                              * be detected via a different
2171                                              * mount structure.
2172                                              */
2173                                             at_root = 1;
2174                                             if (new_mp == mp)
2175                                                 new_mp = nvp->v_mount;
2176                                         }
2177                                     }
2178                                 } else {
2179                                     nvp = NULL;
2180                                 }
2181                                 if (r) {
2182                                         if (!NFSISSET_ATTRBIT(&attrbits,
2183                                             NFSATTRBIT_RDATTRERROR)) {
2184                                                 if (nvp != NULL)
2185                                                         vput(nvp);
2186                                                 if (needs_unbusy != 0)
2187                                                         vfs_unbusy(new_mp);
2188                                                 nd->nd_repstat = r;
2189                                                 break;
2190                                         }
2191                                 }
2192                         }
2193
2194                         /*
2195                          * Build the directory record xdr
2196                          */
2197                         if (nd->nd_flag & ND_NFSV3) {
2198                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2199                                 *tl++ = newnfs_true;
2200                                 *tl++ = 0;
2201                                 *tl = txdr_unsigned(dp->d_fileno);
2202                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2203                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2204                                 *tl++ = 0;
2205                                 *tl = txdr_unsigned(*cookiep);
2206                                 nfsrv_postopattr(nd, 0, nvap);
2207                                 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2208                                 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2209                                 if (nvp != NULL)
2210                                         vput(nvp);
2211                         } else {
2212                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2213                                 *tl++ = newnfs_true;
2214                                 *tl++ = 0;
2215                                 *tl = txdr_unsigned(*cookiep);
2216                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2217                                 if (nvp != NULL) {
2218                                         supports_nfsv4acls =
2219                                             nfs_supportsnfsv4acls(nvp);
2220                                         NFSVOPUNLOCK(nvp, 0);
2221                                 } else
2222                                         supports_nfsv4acls = 0;
2223                                 if (refp != NULL) {
2224                                         dirlen += nfsrv_putreferralattr(nd,
2225                                             &savbits, refp, 0,
2226                                             &nd->nd_repstat);
2227                                         if (nd->nd_repstat) {
2228                                                 if (nvp != NULL)
2229                                                         vrele(nvp);
2230                                                 if (needs_unbusy != 0)
2231                                                         vfs_unbusy(new_mp);
2232                                                 break;
2233                                         }
2234                                 } else if (r) {
2235                                         dirlen += nfsvno_fillattr(nd, new_mp,
2236                                             nvp, nvap, &nfh, r, &rderrbits,
2237                                             nd->nd_cred, p, isdgram, 0,
2238                                             supports_nfsv4acls, at_root,
2239                                             mounted_on_fileno);
2240                                 } else {
2241                                         dirlen += nfsvno_fillattr(nd, new_mp,
2242                                             nvp, nvap, &nfh, r, &attrbits,
2243                                             nd->nd_cred, p, isdgram, 0,
2244                                             supports_nfsv4acls, at_root,
2245                                             mounted_on_fileno);
2246                                 }
2247                                 if (nvp != NULL)
2248                                         vrele(nvp);
2249                                 dirlen += (3 * NFSX_UNSIGNED);
2250                         }
2251                         if (needs_unbusy != 0)
2252                                 vfs_unbusy(new_mp);
2253                         if (dirlen <= cnt)
2254                                 entrycnt++;
2255                 }
2256                 cpos += dp->d_reclen;
2257                 dp = (struct dirent *)cpos;
2258                 cookiep++;
2259                 ncookies--;
2260         }
2261         vrele(vp);
2262         vfs_unbusy(mp);
2263
2264         /*
2265          * If dirlen > cnt, we must strip off the last entry. If that
2266          * results in an empty reply, report NFSERR_TOOSMALL.
2267          */
2268         if (dirlen > cnt || nd->nd_repstat) {
2269                 if (!nd->nd_repstat && entrycnt == 0)
2270                         nd->nd_repstat = NFSERR_TOOSMALL;
2271                 if (nd->nd_repstat) {
2272                         newnfs_trimtrailing(nd, mb0, bpos0);
2273                         if (nd->nd_flag & ND_NFSV3)
2274                                 nfsrv_postopattr(nd, getret, &at);
2275                 } else
2276                         newnfs_trimtrailing(nd, mb1, bpos1);
2277                 eofflag = 0;
2278         } else if (cpos < cend)
2279                 eofflag = 0;
2280         if (!nd->nd_repstat) {
2281                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2282                 *tl++ = newnfs_false;
2283                 if (eofflag)
2284                         *tl = newnfs_true;
2285                 else
2286                         *tl = newnfs_false;
2287         }
2288         FREE((caddr_t)cookies, M_TEMP);
2289         FREE((caddr_t)rbuf, M_TEMP);
2290
2291 out:
2292         NFSEXITCODE2(0, nd);
2293         return (0);
2294 nfsmout:
2295         vput(vp);
2296         NFSEXITCODE2(error, nd);
2297         return (error);
2298 }
2299
2300 /*
2301  * Get the settable attributes out of the mbuf list.
2302  * (Return 0 or EBADRPC)
2303  */
2304 int
2305 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2306     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2307 {
2308         u_int32_t *tl;
2309         struct nfsv2_sattr *sp;
2310         int error = 0, toclient = 0;
2311
2312         switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2313         case ND_NFSV2:
2314                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2315                 /*
2316                  * Some old clients didn't fill in the high order 16bits.
2317                  * --> check the low order 2 bytes for 0xffff
2318                  */
2319                 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2320                         nvap->na_mode = nfstov_mode(sp->sa_mode);
2321                 if (sp->sa_uid != newnfs_xdrneg1)
2322                         nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2323                 if (sp->sa_gid != newnfs_xdrneg1)
2324                         nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2325                 if (sp->sa_size != newnfs_xdrneg1)
2326                         nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2327                 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2328 #ifdef notyet
2329                         fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2330 #else
2331                         nvap->na_atime.tv_sec =
2332                                 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2333                         nvap->na_atime.tv_nsec = 0;
2334 #endif
2335                 }
2336                 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2337                         fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2338                 break;
2339         case ND_NFSV3:
2340                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2341                 if (*tl == newnfs_true) {
2342                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2343                         nvap->na_mode = nfstov_mode(*tl);
2344                 }
2345                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2346                 if (*tl == newnfs_true) {
2347                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2348                         nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2349                 }
2350                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2351                 if (*tl == newnfs_true) {
2352                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2353                         nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2354                 }
2355                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2356                 if (*tl == newnfs_true) {
2357                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2358                         nvap->na_size = fxdr_hyper(tl);
2359                 }
2360                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2361                 switch (fxdr_unsigned(int, *tl)) {
2362                 case NFSV3SATTRTIME_TOCLIENT:
2363                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2364                         fxdr_nfsv3time(tl, &nvap->na_atime);
2365                         toclient = 1;
2366                         break;
2367                 case NFSV3SATTRTIME_TOSERVER:
2368                         vfs_timestamp(&nvap->na_atime);
2369                         nvap->na_vaflags |= VA_UTIMES_NULL;
2370                         break;
2371                 }
2372                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2373                 switch (fxdr_unsigned(int, *tl)) {
2374                 case NFSV3SATTRTIME_TOCLIENT:
2375                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2376                         fxdr_nfsv3time(tl, &nvap->na_mtime);
2377                         nvap->na_vaflags &= ~VA_UTIMES_NULL;
2378                         break;
2379                 case NFSV3SATTRTIME_TOSERVER:
2380                         vfs_timestamp(&nvap->na_mtime);
2381                         if (!toclient)
2382                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2383                         break;
2384                 }
2385                 break;
2386         case ND_NFSV4:
2387                 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2388         }
2389 nfsmout:
2390         NFSEXITCODE2(error, nd);
2391         return (error);
2392 }
2393
2394 /*
2395  * Handle the setable attributes for V4.
2396  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2397  */
2398 int
2399 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2400     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2401 {
2402         u_int32_t *tl;
2403         int attrsum = 0;
2404         int i, j;
2405         int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2406         int toclient = 0;
2407         u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2408         uid_t uid;
2409         gid_t gid;
2410
2411         error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2412         if (error)
2413                 goto nfsmout;
2414         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2415         attrsize = fxdr_unsigned(int, *tl);
2416
2417         /*
2418          * Loop around getting the setable attributes. If an unsupported
2419          * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2420          */
2421         if (retnotsup) {
2422                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2423                 bitpos = NFSATTRBIT_MAX;
2424         } else {
2425                 bitpos = 0;
2426         }
2427         for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2428             if (attrsum > attrsize) {
2429                 error = NFSERR_BADXDR;
2430                 goto nfsmout;
2431             }
2432             if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2433                 switch (bitpos) {
2434                 case NFSATTRBIT_SIZE:
2435                         NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2436                      if (vp != NULL && vp->v_type != VREG) {
2437                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2438                                 NFSERR_INVAL;
2439                             goto nfsmout;
2440                         }
2441                         nvap->na_size = fxdr_hyper(tl);
2442                         attrsum += NFSX_HYPER;
2443                         break;
2444                 case NFSATTRBIT_ACL:
2445                         error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2446                             p);
2447                         if (error)
2448                                 goto nfsmout;
2449                         if (aceerr && !nd->nd_repstat)
2450                                 nd->nd_repstat = aceerr;
2451                         attrsum += aclsize;
2452                         break;
2453                 case NFSATTRBIT_ARCHIVE:
2454                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2455                         if (!nd->nd_repstat)
2456                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2457                         attrsum += NFSX_UNSIGNED;
2458                         break;
2459                 case NFSATTRBIT_HIDDEN:
2460                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2461                         if (!nd->nd_repstat)
2462                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2463                         attrsum += NFSX_UNSIGNED;
2464                         break;
2465                 case NFSATTRBIT_MIMETYPE:
2466                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2467                         i = fxdr_unsigned(int, *tl);
2468                         error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2469                         if (error)
2470                                 goto nfsmout;
2471                         if (!nd->nd_repstat)
2472                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2473                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2474                         break;
2475                 case NFSATTRBIT_MODE:
2476                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2477                         nvap->na_mode = nfstov_mode(*tl);
2478                         attrsum += NFSX_UNSIGNED;
2479                         break;
2480                 case NFSATTRBIT_OWNER:
2481                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2482                         j = fxdr_unsigned(int, *tl);
2483                         if (j < 0) {
2484                                 error = NFSERR_BADXDR;
2485                                 goto nfsmout;
2486                         }
2487                         if (j > NFSV4_SMALLSTR)
2488                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2489                         else
2490                                 cp = namestr;
2491                         error = nfsrv_mtostr(nd, cp, j);
2492                         if (error) {
2493                                 if (j > NFSV4_SMALLSTR)
2494                                         free(cp, M_NFSSTRING);
2495                                 goto nfsmout;
2496                         }
2497                         if (!nd->nd_repstat) {
2498                                 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2499                                     p);
2500                                 if (!nd->nd_repstat)
2501                                         nvap->na_uid = uid;
2502                         }
2503                         if (j > NFSV4_SMALLSTR)
2504                                 free(cp, M_NFSSTRING);
2505                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2506                         break;
2507                 case NFSATTRBIT_OWNERGROUP:
2508                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2509                         j = fxdr_unsigned(int, *tl);
2510                         if (j < 0) {
2511                                 error = NFSERR_BADXDR;
2512                                 goto nfsmout;
2513                         }
2514                         if (j > NFSV4_SMALLSTR)
2515                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2516                         else
2517                                 cp = namestr;
2518                         error = nfsrv_mtostr(nd, cp, j);
2519                         if (error) {
2520                                 if (j > NFSV4_SMALLSTR)
2521                                         free(cp, M_NFSSTRING);
2522                                 goto nfsmout;
2523                         }
2524                         if (!nd->nd_repstat) {
2525                                 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2526                                     p);
2527                                 if (!nd->nd_repstat)
2528                                         nvap->na_gid = gid;
2529                         }
2530                         if (j > NFSV4_SMALLSTR)
2531                                 free(cp, M_NFSSTRING);
2532                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2533                         break;
2534                 case NFSATTRBIT_SYSTEM:
2535                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2536                         if (!nd->nd_repstat)
2537                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2538                         attrsum += NFSX_UNSIGNED;
2539                         break;
2540                 case NFSATTRBIT_TIMEACCESSSET:
2541                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2542                         attrsum += NFSX_UNSIGNED;
2543                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2544                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2545                             fxdr_nfsv4time(tl, &nvap->na_atime);
2546                             toclient = 1;
2547                             attrsum += NFSX_V4TIME;
2548                         } else {
2549                             vfs_timestamp(&nvap->na_atime);
2550                             nvap->na_vaflags |= VA_UTIMES_NULL;
2551                         }
2552                         break;
2553                 case NFSATTRBIT_TIMEBACKUP:
2554                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2555                         if (!nd->nd_repstat)
2556                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2557                         attrsum += NFSX_V4TIME;
2558                         break;
2559                 case NFSATTRBIT_TIMECREATE:
2560                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2561                         if (!nd->nd_repstat)
2562                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2563                         attrsum += NFSX_V4TIME;
2564                         break;
2565                 case NFSATTRBIT_TIMEMODIFYSET:
2566                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2567                         attrsum += NFSX_UNSIGNED;
2568                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2569                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2570                             fxdr_nfsv4time(tl, &nvap->na_mtime);
2571                             nvap->na_vaflags &= ~VA_UTIMES_NULL;
2572                             attrsum += NFSX_V4TIME;
2573                         } else {
2574                             vfs_timestamp(&nvap->na_mtime);
2575                             if (!toclient)
2576                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2577                         }
2578                         break;
2579                 default:
2580                         nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2581                         /*
2582                          * set bitpos so we drop out of the loop.
2583                          */
2584                         bitpos = NFSATTRBIT_MAX;
2585                         break;
2586                 }
2587         }
2588
2589         /*
2590          * some clients pad the attrlist, so we need to skip over the
2591          * padding.
2592          */
2593         if (attrsum > attrsize) {
2594                 error = NFSERR_BADXDR;
2595         } else {
2596                 attrsize = NFSM_RNDUP(attrsize);
2597                 if (attrsum < attrsize)
2598                         error = nfsm_advance(nd, attrsize - attrsum, -1);
2599         }
2600 nfsmout:
2601         NFSEXITCODE2(error, nd);
2602         return (error);
2603 }
2604
2605 /*
2606  * Check/setup export credentials.
2607  */
2608 int
2609 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2610     struct ucred *credanon)
2611 {
2612         int error = 0;
2613
2614         /*
2615          * Check/setup credentials.
2616          */
2617         if (nd->nd_flag & ND_GSS)
2618                 exp->nes_exflag &= ~MNT_EXPORTANON;
2619
2620         /*
2621          * Check to see if the operation is allowed for this security flavor.
2622          * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2623          * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2624          * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2625          */
2626         if (nfsvno_testexp(nd, exp) &&
2627             nd->nd_procnum != NFSV4OP_SECINFO &&
2628             nd->nd_procnum != NFSPROC_FSINFO) {
2629                 if (nd->nd_flag & ND_NFSV4)
2630                         error = NFSERR_WRONGSEC;
2631                 else
2632                         error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2633                 goto out;
2634         }
2635
2636         /*
2637          * Check to see if the file system is exported V4 only.
2638          */
2639         if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2640                 error = NFSERR_PROGNOTV4;
2641                 goto out;
2642         }
2643
2644         /*
2645          * Now, map the user credentials.
2646          * (Note that ND_AUTHNONE will only be set for an NFSv3
2647          *  Fsinfo RPC. If set for anything else, this code might need
2648          *  to change.)
2649          */
2650         if (NFSVNO_EXPORTED(exp)) {
2651                 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
2652                      NFSVNO_EXPORTANON(exp) ||
2653                      (nd->nd_flag & ND_AUTHNONE) != 0) {
2654                         nd->nd_cred->cr_uid = credanon->cr_uid;
2655                         nd->nd_cred->cr_gid = credanon->cr_gid;
2656                         crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2657                             credanon->cr_groups);
2658                 } else if ((nd->nd_flag & ND_GSS) == 0) {
2659                         /*
2660                          * If using AUTH_SYS, call nfsrv_getgrpscred() to see
2661                          * if there is a replacement credential with a group
2662                          * list set up by "nfsuserd -manage-gids".
2663                          * If there is no replacement, nfsrv_getgrpscred()
2664                          * simply returns its argument.
2665                          */
2666                         nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
2667                 }
2668         }
2669
2670 out:
2671         NFSEXITCODE2(error, nd);
2672         return (error);
2673 }
2674
2675 /*
2676  * Check exports.
2677  */
2678 int
2679 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2680     struct ucred **credp)
2681 {
2682         int i, error, *secflavors;
2683
2684         error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2685             &exp->nes_numsecflavor, &secflavors);
2686         if (error) {
2687                 if (nfs_rootfhset) {
2688                         exp->nes_exflag = 0;
2689                         exp->nes_numsecflavor = 0;
2690                         error = 0;
2691                 }
2692         } else {
2693                 /* Copy the security flavors. */
2694                 for (i = 0; i < exp->nes_numsecflavor; i++)
2695                         exp->nes_secflavors[i] = secflavors[i];
2696         }
2697         NFSEXITCODE(error);
2698         return (error);
2699 }
2700
2701 /*
2702  * Get a vnode for a file handle and export stuff.
2703  */
2704 int
2705 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2706     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2707     struct ucred **credp)
2708 {
2709         int i, error, *secflavors;
2710
2711         *credp = NULL;
2712         exp->nes_numsecflavor = 0;
2713         error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2714         if (error != 0)
2715                 /* Make sure the server replies ESTALE to the client. */
2716                 error = ESTALE;
2717         if (nam && !error) {
2718                 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2719                     &exp->nes_numsecflavor, &secflavors);
2720                 if (error) {
2721                         if (nfs_rootfhset) {
2722                                 exp->nes_exflag = 0;
2723                                 exp->nes_numsecflavor = 0;
2724                                 error = 0;
2725                         } else {
2726                                 vput(*vpp);
2727                         }
2728                 } else {
2729                         /* Copy the security flavors. */
2730                         for (i = 0; i < exp->nes_numsecflavor; i++)
2731                                 exp->nes_secflavors[i] = secflavors[i];
2732                 }
2733         }
2734         NFSEXITCODE(error);
2735         return (error);
2736 }
2737
2738 /*
2739  * nfsd_fhtovp() - convert a fh to a vnode ptr
2740  *      - look up fsid in mount list (if not found ret error)
2741  *      - get vp and export rights by calling nfsvno_fhtovp()
2742  *      - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2743  *        for AUTH_SYS
2744  *      - if mpp != NULL, return the mount point so that it can
2745  *        be used for vn_finished_write() by the caller
2746  */
2747 void
2748 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2749     struct vnode **vpp, struct nfsexstuff *exp,
2750     struct mount **mpp, int startwrite, struct thread *p)
2751 {
2752         struct mount *mp;
2753         struct ucred *credanon;
2754         fhandle_t *fhp;
2755
2756         fhp = (fhandle_t *)nfp->nfsrvfh_data;
2757         /*
2758          * Check for the special case of the nfsv4root_fh.
2759          */
2760         mp = vfs_busyfs(&fhp->fh_fsid);
2761         if (mpp != NULL)
2762                 *mpp = mp;
2763         if (mp == NULL) {
2764                 *vpp = NULL;
2765                 nd->nd_repstat = ESTALE;
2766                 goto out;
2767         }
2768
2769         if (startwrite) {
2770                 vn_start_write(NULL, mpp, V_WAIT);
2771                 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2772                         lktype = LK_EXCLUSIVE;
2773         }
2774         nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2775             &credanon);
2776         vfs_unbusy(mp);
2777
2778         /*
2779          * For NFSv4 without a pseudo root fs, unexported file handles
2780          * can be returned, so that Lookup works everywhere.
2781          */
2782         if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2783             !(nd->nd_flag & ND_NFSV4)) {
2784                 vput(*vpp);
2785                 nd->nd_repstat = EACCES;
2786         }
2787
2788         /*
2789          * Personally, I've never seen any point in requiring a
2790          * reserved port#, since only in the rare case where the
2791          * clients are all boxes with secure system privileges,
2792          * does it provide any enhanced security, but... some people
2793          * believe it to be useful and keep putting this code back in.
2794          * (There is also some "security checker" out there that
2795          *  complains if the nfs server doesn't enforce this.)
2796          * However, note the following:
2797          * RFC3530 (NFSv4) specifies that a reserved port# not be
2798          *      required.
2799          * RFC2623 recommends that, if a reserved port# is checked for,
2800          *      that there be a way to turn that off--> ifdef'd.
2801          */
2802 #ifdef NFS_REQRSVPORT
2803         if (!nd->nd_repstat) {
2804                 struct sockaddr_in *saddr;
2805                 struct sockaddr_in6 *saddr6;
2806
2807                 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2808                 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2809                 if (!(nd->nd_flag & ND_NFSV4) &&
2810                     ((saddr->sin_family == AF_INET &&
2811                       ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2812                      (saddr6->sin6_family == AF_INET6 &&
2813                       ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2814                         vput(*vpp);
2815                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2816                 }
2817         }
2818 #endif  /* NFS_REQRSVPORT */
2819
2820         /*
2821          * Check/setup credentials.
2822          */
2823         if (!nd->nd_repstat) {
2824                 nd->nd_saveduid = nd->nd_cred->cr_uid;
2825                 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2826                 if (nd->nd_repstat)
2827                         vput(*vpp);
2828         }
2829         if (credanon != NULL)
2830                 crfree(credanon);
2831         if (nd->nd_repstat) {
2832                 if (startwrite)
2833                         vn_finished_write(mp);
2834                 *vpp = NULL;
2835                 if (mpp != NULL)
2836                         *mpp = NULL;
2837         }
2838
2839 out:
2840         NFSEXITCODE2(0, nd);
2841 }
2842
2843 /*
2844  * glue for fp.
2845  */
2846 static int
2847 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2848 {
2849         struct filedesc *fdp;
2850         struct file *fp;
2851         int error = 0;
2852
2853         fdp = p->td_proc->p_fd;
2854         if (fd < 0 || fd >= fdp->fd_nfiles ||
2855             (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2856                 error = EBADF;
2857                 goto out;
2858         }
2859         *fpp = fp;
2860
2861 out:
2862         NFSEXITCODE(error);
2863         return (error);
2864 }
2865
2866 /*
2867  * Called from nfssvc() to update the exports list. Just call
2868  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2869  * in the mount list.
2870  */
2871 int
2872 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2873 {
2874         struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2875         int error = 0;
2876         struct nameidata nd;
2877         fhandle_t fh;
2878
2879         error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2880         if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2881                 nfs_rootfhset = 0;
2882         else if (error == 0) {
2883                 if (nfsexargp->fspec == NULL) {
2884                         error = EPERM;
2885                         goto out;
2886                 }
2887                 /*
2888                  * If fspec != NULL, this is the v4root path.
2889                  */
2890                 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2891                     nfsexargp->fspec, p);
2892                 if ((error = namei(&nd)) != 0)
2893                         goto out;
2894                 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2895                 vrele(nd.ni_vp);
2896                 if (!error) {
2897                         nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2898                         NFSBCOPY((caddr_t)&fh,
2899                             nfs_rootfh.nfsrvfh_data,
2900                             sizeof (fhandle_t));
2901                         nfs_rootfhset = 1;
2902                 }
2903         }
2904
2905 out:
2906         NFSEXITCODE(error);
2907         return (error);
2908 }
2909
2910 /*
2911  * This function needs to test to see if the system is near its limit
2912  * for memory allocation via malloc() or mget() and return True iff
2913  * either of these resources are near their limit.
2914  * XXX (For now, this is just a stub.)
2915  */
2916 int nfsrv_testmalloclimit = 0;
2917 int
2918 nfsrv_mallocmget_limit(void)
2919 {
2920         static int printmesg = 0;
2921         static int testval = 1;
2922
2923         if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2924                 if ((printmesg++ % 100) == 0)
2925                         printf("nfsd: malloc/mget near limit\n");
2926                 return (1);
2927         }
2928         return (0);
2929 }
2930
2931 /*
2932  * BSD specific initialization of a mount point.
2933  */
2934 void
2935 nfsd_mntinit(void)
2936 {
2937         static int inited = 0;
2938
2939         if (inited)
2940                 return;
2941         inited = 1;
2942         nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2943         TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2944         TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2945         nfsv4root_mnt.mnt_export = NULL;
2946         TAILQ_INIT(&nfsv4root_opt);
2947         TAILQ_INIT(&nfsv4root_newopt);
2948         nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2949         nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2950         nfsv4root_mnt.mnt_nvnodelistsize = 0;
2951         nfsv4root_mnt.mnt_activevnodelistsize = 0;
2952 }
2953
2954 /*
2955  * Get a vnode for a file handle, without checking exports, etc.
2956  */
2957 struct vnode *
2958 nfsvno_getvp(fhandle_t *fhp)
2959 {
2960         struct mount *mp;
2961         struct vnode *vp;
2962         int error;
2963
2964         mp = vfs_busyfs(&fhp->fh_fsid);
2965         if (mp == NULL)
2966                 return (NULL);
2967         error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2968         vfs_unbusy(mp);
2969         if (error)
2970                 return (NULL);
2971         return (vp);
2972 }
2973
2974 /*
2975  * Do a local VOP_ADVLOCK().
2976  */
2977 int
2978 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2979     u_int64_t end, struct thread *td)
2980 {
2981         int error = 0;
2982         struct flock fl;
2983         u_int64_t tlen;
2984
2985         if (nfsrv_dolocallocks == 0)
2986                 goto out;
2987         ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
2988
2989         fl.l_whence = SEEK_SET;
2990         fl.l_type = ftype;
2991         fl.l_start = (off_t)first;
2992         if (end == NFS64BITSSET) {
2993                 fl.l_len = 0;
2994         } else {
2995                 tlen = end - first;
2996                 fl.l_len = (off_t)tlen;
2997         }
2998         /*
2999          * For FreeBSD8, the l_pid and l_sysid must be set to the same
3000          * values for all calls, so that all locks will be held by the
3001          * nfsd server. (The nfsd server handles conflicts between the
3002          * various clients.)
3003          * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3004          * bytes, so it can't be put in l_sysid.
3005          */
3006         if (nfsv4_sysid == 0)
3007                 nfsv4_sysid = nlm_acquire_next_sysid();
3008         fl.l_pid = (pid_t)0;
3009         fl.l_sysid = (int)nfsv4_sysid;
3010
3011         if (ftype == F_UNLCK)
3012                 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3013                     (F_POSIX | F_REMOTE));
3014         else
3015                 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3016                     (F_POSIX | F_REMOTE));
3017
3018 out:
3019         NFSEXITCODE(error);
3020         return (error);
3021 }
3022
3023 /*
3024  * Check the nfsv4 root exports.
3025  */
3026 int
3027 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3028 {
3029         struct ucred *credanon;
3030         int exflags, error = 0, numsecflavor, *secflavors, i;
3031
3032         error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3033             &credanon, &numsecflavor, &secflavors);
3034         if (error) {
3035                 error = NFSERR_PROGUNAVAIL;
3036                 goto out;
3037         }
3038         if (credanon != NULL)
3039                 crfree(credanon);
3040         for (i = 0; i < numsecflavor; i++) {
3041                 if (secflavors[i] == AUTH_SYS)
3042                         nd->nd_flag |= ND_EXAUTHSYS;
3043                 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3044                         nd->nd_flag |= ND_EXGSS;
3045                 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3046                         nd->nd_flag |= ND_EXGSSINTEGRITY;
3047                 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3048                         nd->nd_flag |= ND_EXGSSPRIVACY;
3049         }
3050
3051 out:
3052         NFSEXITCODE(error);
3053         return (error);
3054 }
3055
3056 /*
3057  * Nfs server pseudo system call for the nfsd's
3058  */
3059 /*
3060  * MPSAFE
3061  */
3062 static int
3063 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3064 {
3065         struct file *fp;
3066         struct nfsd_addsock_args sockarg;
3067         struct nfsd_nfsd_args nfsdarg;
3068         cap_rights_t rights;
3069         int error;
3070
3071         if (uap->flag & NFSSVC_NFSDADDSOCK) {
3072                 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3073                 if (error)
3074                         goto out;
3075                 /*
3076                  * Since we don't know what rights might be required,
3077                  * pretend that we need them all. It is better to be too
3078                  * careful than too reckless.
3079                  */
3080                 error = fget(td, sockarg.sock,
3081                     cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3082                 if (error != 0)
3083                         goto out;
3084                 if (fp->f_type != DTYPE_SOCKET) {
3085                         fdrop(fp, td);
3086                         error = EPERM;
3087                         goto out;
3088                 }
3089                 error = nfsrvd_addsock(fp);
3090                 fdrop(fp, td);
3091         } else if (uap->flag & NFSSVC_NFSDNFSD) {
3092                 if (uap->argp == NULL) {
3093                         error = EINVAL;
3094                         goto out;
3095                 }
3096                 error = copyin(uap->argp, (caddr_t)&nfsdarg,
3097                     sizeof (nfsdarg));
3098                 if (error)
3099                         goto out;
3100                 error = nfsrvd_nfsd(td, &nfsdarg);
3101         } else {
3102                 error = nfssvc_srvcall(td, uap, td->td_ucred);
3103         }
3104
3105 out:
3106         NFSEXITCODE(error);
3107         return (error);
3108 }
3109
3110 static int
3111 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3112 {
3113         struct nfsex_args export;
3114         struct file *fp = NULL;
3115         int stablefd, len;
3116         struct nfsd_clid adminrevoke;
3117         struct nfsd_dumplist dumplist;
3118         struct nfsd_dumpclients *dumpclients;
3119         struct nfsd_dumplocklist dumplocklist;
3120         struct nfsd_dumplocks *dumplocks;
3121         struct nameidata nd;
3122         vnode_t vp;
3123         int error = EINVAL, igotlock;
3124         struct proc *procp;
3125         static int suspend_nfsd = 0;
3126
3127         if (uap->flag & NFSSVC_PUBLICFH) {
3128                 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3129                     sizeof (fhandle_t));
3130                 error = copyin(uap->argp,
3131                     &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3132                 if (!error)
3133                         nfs_pubfhset = 1;
3134         } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3135                 error = copyin(uap->argp,(caddr_t)&export,
3136                     sizeof (struct nfsex_args));
3137                 if (!error)
3138                         error = nfsrv_v4rootexport(&export, cred, p);
3139         } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3140                 nfs_pubfhset = 0;
3141                 error = 0;
3142         } else if (uap->flag & NFSSVC_STABLERESTART) {
3143                 error = copyin(uap->argp, (caddr_t)&stablefd,
3144                     sizeof (int));
3145                 if (!error)
3146                         error = fp_getfvp(p, stablefd, &fp, &vp);
3147                 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3148                         error = EBADF;
3149                 if (!error && newnfs_numnfsd != 0)
3150                         error = EPERM;
3151                 if (!error) {
3152                         nfsrv_stablefirst.nsf_fp = fp;
3153                         nfsrv_setupstable(p);
3154                 }
3155         } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3156                 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3157                     sizeof (struct nfsd_clid));
3158                 if (!error)
3159                         error = nfsrv_adminrevoke(&adminrevoke, p);
3160         } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3161                 error = copyin(uap->argp, (caddr_t)&dumplist,
3162                     sizeof (struct nfsd_dumplist));
3163                 if (!error && (dumplist.ndl_size < 1 ||
3164                         dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3165                         error = EPERM;
3166                 if (!error) {
3167                     len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3168                     dumpclients = (struct nfsd_dumpclients *)malloc(len,
3169                         M_TEMP, M_WAITOK);
3170                     nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3171                     error = copyout(dumpclients,
3172                         CAST_USER_ADDR_T(dumplist.ndl_list), len);
3173                     free((caddr_t)dumpclients, M_TEMP);
3174                 }
3175         } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3176                 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3177                     sizeof (struct nfsd_dumplocklist));
3178                 if (!error && (dumplocklist.ndllck_size < 1 ||
3179                         dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3180                         error = EPERM;
3181                 if (!error)
3182                         error = nfsrv_lookupfilename(&nd,
3183                                 dumplocklist.ndllck_fname, p);
3184                 if (!error) {
3185                         len = sizeof (struct nfsd_dumplocks) *
3186                                 dumplocklist.ndllck_size;
3187                         dumplocks = (struct nfsd_dumplocks *)malloc(len,
3188                                 M_TEMP, M_WAITOK);
3189                         nfsrv_dumplocks(nd.ni_vp, dumplocks,
3190                             dumplocklist.ndllck_size, p);
3191                         vput(nd.ni_vp);
3192                         error = copyout(dumplocks,
3193                             CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3194                         free((caddr_t)dumplocks, M_TEMP);
3195                 }
3196         } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3197                 procp = p->td_proc;
3198                 PROC_LOCK(procp);
3199                 nfsd_master_pid = procp->p_pid;
3200                 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3201                 nfsd_master_start = procp->p_stats->p_start;
3202                 nfsd_master_proc = procp;
3203                 PROC_UNLOCK(procp);
3204         } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3205                 NFSLOCKV4ROOTMUTEX();
3206                 if (suspend_nfsd == 0) {
3207                         /* Lock out all nfsd threads */
3208                         do {
3209                                 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3210                                     NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3211                         } while (igotlock == 0 && suspend_nfsd == 0);
3212                         suspend_nfsd = 1;
3213                 }
3214                 NFSUNLOCKV4ROOTMUTEX();
3215                 error = 0;
3216         } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3217                 NFSLOCKV4ROOTMUTEX();
3218                 if (suspend_nfsd != 0) {
3219                         nfsv4_unlock(&nfsd_suspend_lock, 0);
3220                         suspend_nfsd = 0;
3221                 }
3222                 NFSUNLOCKV4ROOTMUTEX();
3223                 error = 0;
3224         }
3225
3226         NFSEXITCODE(error);
3227         return (error);
3228 }
3229
3230 /*
3231  * Check exports.
3232  * Returns 0 if ok, 1 otherwise.
3233  */
3234 int
3235 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3236 {
3237         int i;
3238
3239         /*
3240          * This seems odd, but allow the case where the security flavor
3241          * list is empty. This happens when NFSv4 is traversing non-exported
3242          * file systems. Exported file systems should always have a non-empty
3243          * security flavor list.
3244          */
3245         if (exp->nes_numsecflavor == 0)
3246                 return (0);
3247
3248         for (i = 0; i < exp->nes_numsecflavor; i++) {
3249                 /*
3250                  * The tests for privacy and integrity must be first,
3251                  * since ND_GSS is set for everything but AUTH_SYS.
3252                  */
3253                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3254                     (nd->nd_flag & ND_GSSPRIVACY))
3255                         return (0);
3256                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3257                     (nd->nd_flag & ND_GSSINTEGRITY))
3258                         return (0);
3259                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3260                     (nd->nd_flag & ND_GSS))
3261                         return (0);
3262                 if (exp->nes_secflavors[i] == AUTH_SYS &&
3263                     (nd->nd_flag & ND_GSS) == 0)
3264                         return (0);
3265         }
3266         return (1);
3267 }
3268
3269 /*
3270  * Calculate a hash value for the fid in a file handle.
3271  */
3272 uint32_t
3273 nfsrv_hashfh(fhandle_t *fhp)
3274 {
3275         uint32_t hashval;
3276
3277         hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3278         return (hashval);
3279 }
3280
3281 /*
3282  * Calculate a hash value for the sessionid.
3283  */
3284 uint32_t
3285 nfsrv_hashsessionid(uint8_t *sessionid)
3286 {
3287         uint32_t hashval;
3288
3289         hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3290         return (hashval);
3291 }
3292
3293 /*
3294  * Signal the userland master nfsd to backup the stable restart file.
3295  */
3296 void
3297 nfsrv_backupstable(void)
3298 {
3299         struct proc *procp;
3300
3301         if (nfsd_master_proc != NULL) {
3302                 procp = pfind(nfsd_master_pid);
3303                 /* Try to make sure it is the correct process. */
3304                 if (procp == nfsd_master_proc &&
3305                     procp->p_stats->p_start.tv_sec ==
3306                     nfsd_master_start.tv_sec &&
3307                     procp->p_stats->p_start.tv_usec ==
3308                     nfsd_master_start.tv_usec &&
3309                     strcmp(procp->p_comm, nfsd_master_comm) == 0)
3310                         kern_psignal(procp, SIGUSR2);
3311                 else
3312                         nfsd_master_proc = NULL;
3313
3314                 if (procp != NULL)
3315                         PROC_UNLOCK(procp);
3316         }
3317 }
3318
3319 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3320
3321 /*
3322  * Called once to initialize data structures...
3323  */
3324 static int
3325 nfsd_modevent(module_t mod, int type, void *data)
3326 {
3327         int error = 0, i;
3328         static int loaded = 0;
3329
3330         switch (type) {
3331         case MOD_LOAD:
3332                 if (loaded)
3333                         goto out;
3334                 newnfs_portinit();
3335                 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3336                         mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
3337                             MTX_DEF);
3338                         mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
3339                             MTX_DEF);
3340                 }
3341                 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
3342                 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
3343                 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3344                 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3345                 nfsrvd_initcache();
3346                 nfsd_init();
3347                 NFSD_LOCK();
3348                 nfsrvd_init(0);
3349                 NFSD_UNLOCK();
3350                 nfsd_mntinit();
3351 #ifdef VV_DISABLEDELEG
3352                 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3353                 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3354 #endif
3355                 nfsd_call_servertimer = nfsrv_servertimer;
3356                 nfsd_call_nfsd = nfssvc_nfsd;
3357                 loaded = 1;
3358                 break;
3359
3360         case MOD_UNLOAD:
3361                 if (newnfs_numnfsd != 0) {
3362                         error = EBUSY;
3363                         break;
3364                 }
3365
3366 #ifdef VV_DISABLEDELEG
3367                 vn_deleg_ops.vndeleg_recall = NULL;
3368                 vn_deleg_ops.vndeleg_disable = NULL;
3369 #endif
3370                 nfsd_call_servertimer = NULL;
3371                 nfsd_call_nfsd = NULL;
3372
3373                 /* Clean out all NFSv4 state. */
3374                 nfsrv_throwawayallstate(curthread);
3375
3376                 /* Clean the NFS server reply cache */
3377                 nfsrvd_cleancache();
3378
3379                 /* Free up the krpc server pool. */
3380                 if (nfsrvd_pool != NULL)
3381                         svcpool_destroy(nfsrvd_pool);
3382
3383                 /* and get rid of the locks */
3384                 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3385                         mtx_destroy(&nfsrchash_table[i].mtx);
3386                         mtx_destroy(&nfsrcahash_table[i].mtx);
3387                 }
3388                 mtx_destroy(&nfsrc_udpmtx);
3389                 mtx_destroy(&nfs_v4root_mutex);
3390                 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3391                 for (i = 0; i < nfsrv_sessionhashsize; i++)
3392                         mtx_destroy(&nfssessionhash[i].mtx);
3393                 lockdestroy(&nfsv4root_mnt.mnt_explock);
3394                 free(nfsclienthash, M_NFSDCLIENT);
3395                 free(nfslockhash, M_NFSDLOCKFILE);
3396                 free(nfssessionhash, M_NFSDSESSION);
3397                 loaded = 0;
3398                 break;
3399         default:
3400                 error = EOPNOTSUPP;
3401                 break;
3402         }
3403
3404 out:
3405         NFSEXITCODE(error);
3406         return (error);
3407 }
3408 static moduledata_t nfsd_mod = {
3409         "nfsd",
3410         nfsd_modevent,
3411         NULL,
3412 };
3413 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3414
3415 /* So that loader and kldload(2) can find us, wherever we are.. */
3416 MODULE_VERSION(nfsd, 1);
3417 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3418 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3419 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3420 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3421 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3422