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