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