sys/fs/nfsserver/nfs_nfsdport.c

   1 /*-
   2  * Copyright (c) 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Rick Macklem at The University of Guelph.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 4. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  *
  32  */
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 #include <sys/capability.h>
  38
  39 /*
  40  * Functions that perform the vfs operations required by the routines in
  41  * nfsd_serv.c. It is hoped that this change will make the server more
  42  * portable.
  43  */
  44
  45 #include <fs/nfs/nfsport.h>
  46 #include <sys/hash.h>
  47 #include <sys/sysctl.h>
  48 #include <nlm/nlm_prot.h>
  49 #include <nlm/nlm.h>
  50
  51 FEATURE(nfsd, "NFSv4 server");
  52
  53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
  54 extern int nfsrv_useacl;
  55 extern int newnfs_numnfsd;
  56 extern struct mount nfsv4root_mnt;
  57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
  58 extern void (*nfsd_call_servertimer)(void);
  59 extern SVCPOOL  *nfsrvd_pool;
  60 extern struct nfsv4lock nfsd_suspend_lock;
  61 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
  62 NFSDLOCKMUTEX;
  63 struct mtx nfs_cache_mutex;
  64 struct mtx nfs_v4root_mutex;
  65 struct nfsrvfh nfs_rootfh, nfs_pubfh;
  66 int nfs_pubfhset = 0, nfs_rootfhset = 0;
  67 struct proc *nfsd_master_proc = NULL;
  68 static pid_t nfsd_master_pid = (pid_t)-1;
  69 static char nfsd_master_comm[MAXCOMLEN + 1];
  70 static struct timeval nfsd_master_start;
  71 static uint32_t nfsv4_sysid = 0;
  72
  73 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
  74     struct ucred *);
  75
  76 int nfsrv_enable_crossmntpt = 1;
  77 static int nfs_commit_blks;
  78 static int nfs_commit_miss;
  79 extern int nfsrv_issuedelegs;
  80 extern int nfsrv_dolocallocks;
  81
  82 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
  83 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
  84     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
  85 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
  86     0, "");
  87 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
  88     0, "");
  89 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
  90     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
  91 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
  92     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
  93
  94 #define MAX_REORDERED_RPC       16
  95 #define NUM_HEURISTIC           1031
  96 #define NHUSE_INIT              64
  97 #define NHUSE_INC               16
  98 #define NHUSE_MAX               2048
  99
 100 static struct nfsheur {
 101         struct vnode *nh_vp;    /* vp to match (unreferenced pointer) */
 102         off_t nh_nextoff;       /* next offset for sequential detection */
 103         int nh_use;             /* use count for selection */
 104         int nh_seqcount;        /* heuristic */
 105 } nfsheur[NUM_HEURISTIC];
 106
 107
 108 /*
 109  * Heuristic to detect sequential operation.
 110  */
 111 static struct nfsheur *
 112 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
 113 {
 114         struct nfsheur *nh;
 115         int hi, try;
 116
 117         /* Locate best candidate. */
 118         try = 32;
 119         hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
 120         nh = &nfsheur[hi];
 121         while (try--) {
 122                 if (nfsheur[hi].nh_vp == vp) {
 123                         nh = &nfsheur[hi];
 124                         break;
 125                 }
 126                 if (nfsheur[hi].nh_use > 0)
 127                         --nfsheur[hi].nh_use;
 128                 hi = (hi + 1) % NUM_HEURISTIC;
 129                 if (nfsheur[hi].nh_use < nh->nh_use)
 130                         nh = &nfsheur[hi];
 131         }
 132
 133         /* Initialize hint if this is a new file. */
 134         if (nh->nh_vp != vp) {
 135                 nh->nh_vp = vp;
 136                 nh->nh_nextoff = uio->uio_offset;
 137                 nh->nh_use = NHUSE_INIT;
 138                 if (uio->uio_offset == 0)
 139                         nh->nh_seqcount = 4;
 140                 else
 141                         nh->nh_seqcount = 1;
 142         }
 143
 144         /* Calculate heuristic. */
 145         if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
 146             uio->uio_offset == nh->nh_nextoff) {
 147                 /* See comments in vfs_vnops.c:sequential_heuristic(). */
 148                 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
 149                 if (nh->nh_seqcount > IO_SEQMAX)
 150                         nh->nh_seqcount = IO_SEQMAX;
 151         } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
 152             imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
 153                 /* Probably a reordered RPC, leave seqcount alone. */
 154         } else if (nh->nh_seqcount > 1) {
 155                 nh->nh_seqcount /= 2;
 156         } else {
 157                 nh->nh_seqcount = 0;
 158         }
 159         nh->nh_use += NHUSE_INC;
 160         if (nh->nh_use > NHUSE_MAX)
 161                 nh->nh_use = NHUSE_MAX;
 162         return (nh);
 163 }
 164
 165 /*
 166  * Get attributes into nfsvattr structure.
 167  */
 168 int
 169 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 170     struct thread *p, int vpislocked)
 171 {
 172         int error, lockedit = 0;
 173
 174         if (vpislocked == 0) {
 175                 /*
 176                  * When vpislocked == 0, the vnode is either exclusively
 177                  * locked by this thread or not locked by this thread.
 178                  * As such, shared lock it, if not exclusively locked.
 179                  */
 180                 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
 181                         lockedit = 1;
 182                         NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
 183                 }
 184         }
 185         error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
 186         if (lockedit != 0)
 187                 NFSVOPUNLOCK(vp, 0);
 188
 189         NFSEXITCODE(error);
 190         return (error);
 191 }
 192
 193 /*
 194  * Get a file handle for a vnode.
 195  */
 196 int
 197 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
 198 {
 199         int error;
 200
 201         NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
 202         fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
 203         error = VOP_VPTOFH(vp, &fhp->fh_fid);
 204
 205         NFSEXITCODE(error);
 206         return (error);
 207 }
 208
 209 /*
 210  * Perform access checking for vnodes obtained from file handles that would
 211  * refer to files already opened by a Unix client. You cannot just use
 212  * vn_writechk() and VOP_ACCESSX() for two reasons.
 213  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
 214  *     case.
 215  * 2 - The owner is to be given access irrespective of mode bits for some
 216  *     operations, so that processes that chmod after opening a file don't
 217  *     break.
 218  */
 219 int
 220 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
 221     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
 222     u_int32_t *supportedtypep)
 223 {
 224         struct vattr vattr;
 225         int error = 0, getret = 0;
 226
 227         if (vpislocked == 0) {
 228                 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
 229                         error = EPERM;
 230                         goto out;
 231                 }
 232         }
 233         if (accmode & VWRITE) {
 234                 /* Just vn_writechk() changed to check rdonly */
 235                 /*
 236                  * Disallow write attempts on read-only file systems;
 237                  * unless the file is a socket or a block or character
 238                  * device resident on the file system.
 239                  */
 240                 if (NFSVNO_EXRDONLY(exp) ||
 241                     (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 242                         switch (vp->v_type) {
 243                         case VREG:
 244                         case VDIR:
 245                         case VLNK:
 246                                 error = EROFS;
 247                         default:
 248                                 break;
 249                         }
 250                 }
 251                 /*
 252                  * If there's shared text associated with
 253                  * the inode, try to free it up once.  If
 254                  * we fail, we can't allow writing.
 255                  */
 256                 if (VOP_IS_TEXT(vp) && error == 0)
 257                         error = ETXTBSY;
 258         }
 259         if (error != 0) {
 260                 if (vpislocked == 0)
 261                         NFSVOPUNLOCK(vp, 0);
 262                 goto out;
 263         }
 264
 265         /*
 266          * Should the override still be applied when ACLs are enabled?
 267          */
 268         error = VOP_ACCESSX(vp, accmode, cred, p);
 269         if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
 270                 /*
 271                  * Try again with VEXPLICIT_DENY, to see if the test for
 272                  * deletion is supported.
 273                  */
 274                 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
 275                 if (error == 0) {
 276                         if (vp->v_type == VDIR) {
 277                                 accmode &= ~(VDELETE | VDELETE_CHILD);
 278                                 accmode |= VWRITE;
 279                                 error = VOP_ACCESSX(vp, accmode, cred, p);
 280                         } else if (supportedtypep != NULL) {
 281                                 *supportedtypep &= ~NFSACCESS_DELETE;
 282                         }
 283                 }
 284         }
 285
 286         /*
 287          * Allow certain operations for the owner (reads and writes
 288          * on files that are already open).
 289          */
 290         if (override != NFSACCCHK_NOOVERRIDE &&
 291             (error == EPERM || error == EACCES)) {
 292                 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
 293                         error = 0;
 294                 else if (override & NFSACCCHK_ALLOWOWNER) {
 295                         getret = VOP_GETATTR(vp, &vattr, cred);
 296                         if (getret == 0 && cred->cr_uid == vattr.va_uid)
 297                                 error = 0;
 298                 }
 299         }
 300         if (vpislocked == 0)
 301                 NFSVOPUNLOCK(vp, 0);
 302
 303 out:
 304         NFSEXITCODE(error);
 305         return (error);
 306 }
 307
 308 /*
 309  * Set attribute(s) vnop.
 310  */
 311 int
 312 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 313     struct thread *p, struct nfsexstuff *exp)
 314 {
 315         int error;
 316
 317         error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
 318         NFSEXITCODE(error);
 319         return (error);
 320 }
 321
 322 /*
 323  * Set up nameidata for a lookup() call and do it.
 324  */
 325 int
 326 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
 327     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
 328     struct vnode **retdirp)
 329 {
 330         struct componentname *cnp = &ndp->ni_cnd;
 331         int i;
 332         struct iovec aiov;
 333         struct uio auio;
 334         int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
 335         int error = 0, crossmnt;
 336         char *cp;
 337
 338         *retdirp = NULL;
 339         cnp->cn_nameptr = cnp->cn_pnbuf;
 340         ndp->ni_strictrelative = 0;
 341         /*
 342          * Extract and set starting directory.
 343          */
 344         if (dp->v_type != VDIR) {
 345                 if (islocked)
 346                         vput(dp);
 347                 else
 348                         vrele(dp);
 349                 nfsvno_relpathbuf(ndp);
 350                 error = ENOTDIR;
 351                 goto out1;
 352         }
 353         if (islocked)
 354                 NFSVOPUNLOCK(dp, 0);
 355         VREF(dp);
 356         *retdirp = dp;
 357         if (NFSVNO_EXRDONLY(exp))
 358                 cnp->cn_flags |= RDONLY;
 359         ndp->ni_segflg = UIO_SYSSPACE;
 360         crossmnt = 1;
 361
 362         if (nd->nd_flag & ND_PUBLOOKUP) {
 363                 ndp->ni_loopcnt = 0;
 364                 if (cnp->cn_pnbuf[0] == '/') {
 365                         vrele(dp);
 366                         /*
 367                          * Check for degenerate pathnames here, since lookup()
 368                          * panics on them.
 369                          */
 370                         for (i = 1; i < ndp->ni_pathlen; i++)
 371                                 if (cnp->cn_pnbuf[i] != '/')
 372                                         break;
 373                         if (i == ndp->ni_pathlen) {
 374                                 error = NFSERR_ACCES;
 375                                 goto out;
 376                         }
 377                         dp = rootvnode;
 378                         VREF(dp);
 379                 }
 380         } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
 381             (nd->nd_flag & ND_NFSV4) == 0) {
 382                 /*
 383                  * Only cross mount points for NFSv4 when doing a
 384                  * mount while traversing the file system above
 385                  * the mount point, unless nfsrv_enable_crossmntpt is set.
 386                  */
 387                 cnp->cn_flags |= NOCROSSMOUNT;
 388                 crossmnt = 0;
 389         }
 390
 391         /*
 392          * Initialize for scan, set ni_startdir and bump ref on dp again
 393          * becuase lookup() will dereference ni_startdir.
 394          */
 395
 396         cnp->cn_thread = p;
 397         ndp->ni_startdir = dp;
 398         ndp->ni_rootdir = rootvnode;
 399         ndp->ni_topdir = NULL;
 400
 401         if (!lockleaf)
 402                 cnp->cn_flags |= LOCKLEAF;
 403         for (;;) {
 404                 cnp->cn_nameptr = cnp->cn_pnbuf;
 405                 /*
 406                  * Call lookup() to do the real work.  If an error occurs,
 407                  * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
 408                  * we do not have to dereference anything before returning.
 409                  * In either case ni_startdir will be dereferenced and NULLed
 410                  * out.
 411                  */
 412                 error = lookup(ndp);
 413                 if (error)
 414                         break;
 415
 416                 /*
 417                  * Check for encountering a symbolic link.  Trivial
 418                  * termination occurs if no symlink encountered.
 419                  */
 420                 if ((cnp->cn_flags & ISSYMLINK) == 0) {
 421                         if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
 422                                 nfsvno_relpathbuf(ndp);
 423                         if (ndp->ni_vp && !lockleaf)
 424                                 NFSVOPUNLOCK(ndp->ni_vp, 0);
 425                         break;
 426                 }
 427
 428                 /*
 429                  * Validate symlink
 430                  */
 431                 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
 432                         NFSVOPUNLOCK(ndp->ni_dvp, 0);
 433                 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
 434                         error = EINVAL;
 435                         goto badlink2;
 436                 }
 437
 438                 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
 439                         error = ELOOP;
 440                         goto badlink2;
 441                 }
 442                 if (ndp->ni_pathlen > 1)
 443                         cp = uma_zalloc(namei_zone, M_WAITOK);
 444                 else
 445                         cp = cnp->cn_pnbuf;
 446                 aiov.iov_base = cp;
 447                 aiov.iov_len = MAXPATHLEN;
 448                 auio.uio_iov = &aiov;
 449                 auio.uio_iovcnt = 1;
 450                 auio.uio_offset = 0;
 451                 auio.uio_rw = UIO_READ;
 452                 auio.uio_segflg = UIO_SYSSPACE;
 453                 auio.uio_td = NULL;
 454                 auio.uio_resid = MAXPATHLEN;
 455                 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
 456                 if (error) {
 457                 badlink1:
 458                         if (ndp->ni_pathlen > 1)
 459                                 uma_zfree(namei_zone, cp);
 460                 badlink2:
 461                         vrele(ndp->ni_dvp);
 462                         vput(ndp->ni_vp);
 463                         break;
 464                 }
 465                 linklen = MAXPATHLEN - auio.uio_resid;
 466                 if (linklen == 0) {
 467                         error = ENOENT;
 468                         goto badlink1;
 469                 }
 470                 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
 471                         error = ENAMETOOLONG;
 472                         goto badlink1;
 473                 }
 474
 475                 /*
 476                  * Adjust or replace path
 477                  */
 478                 if (ndp->ni_pathlen > 1) {
 479                         NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
 480                         uma_zfree(namei_zone, cnp->cn_pnbuf);
 481                         cnp->cn_pnbuf = cp;
 482                 } else
 483                         cnp->cn_pnbuf[linklen] = '\0';
 484                 ndp->ni_pathlen += linklen;
 485
 486                 /*
 487                  * Cleanup refs for next loop and check if root directory
 488                  * should replace current directory.  Normally ni_dvp
 489                  * becomes the new base directory and is cleaned up when
 490                  * we loop.  Explicitly null pointers after invalidation
 491                  * to clarify operation.
 492                  */
 493                 vput(ndp->ni_vp);
 494                 ndp->ni_vp = NULL;
 495
 496                 if (cnp->cn_pnbuf[0] == '/') {
 497                         vrele(ndp->ni_dvp);
 498                         ndp->ni_dvp = ndp->ni_rootdir;
 499                         VREF(ndp->ni_dvp);
 500                 }
 501                 ndp->ni_startdir = ndp->ni_dvp;
 502                 ndp->ni_dvp = NULL;
 503         }
 504         if (!lockleaf)
 505                 cnp->cn_flags &= ~LOCKLEAF;
 506
 507 out:
 508         if (error) {
 509                 nfsvno_relpathbuf(ndp);
 510                 ndp->ni_vp = NULL;
 511                 ndp->ni_dvp = NULL;
 512                 ndp->ni_startdir = NULL;
 513         } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
 514                 ndp->ni_dvp = NULL;
 515         }
 516
 517 out1:
 518         NFSEXITCODE2(error, nd);
 519         return (error);
 520 }
 521
 522 /*
 523  * Set up a pathname buffer and return a pointer to it and, optionally
 524  * set a hash pointer.
 525  */
 526 void
 527 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
 528 {
 529         struct componentname *cnp = &ndp->ni_cnd;
 530
 531         cnp->cn_flags |= (NOMACCHECK | HASBUF);
 532         cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
 533         if (hashpp != NULL)
 534                 *hashpp = NULL;
 535         *bufpp = cnp->cn_pnbuf;
 536 }
 537
 538 /*
 539  * Release the above path buffer, if not released by nfsvno_namei().
 540  */
 541 void
 542 nfsvno_relpathbuf(struct nameidata *ndp)
 543 {
 544
 545         if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
 546                 panic("nfsrelpath");
 547         uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
 548         ndp->ni_cnd.cn_flags &= ~HASBUF;
 549 }
 550
 551 /*
 552  * Readlink vnode op into an mbuf list.
 553  */
 554 int
 555 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
 556     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
 557 {
 558         struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
 559         struct iovec *ivp = iv;
 560         struct uio io, *uiop = &io;
 561         struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
 562         int i, len, tlen, error = 0;
 563
 564         len = 0;
 565         i = 0;
 566         while (len < NFS_MAXPATHLEN) {
 567                 NFSMGET(mp);
 568                 MCLGET(mp, M_WAIT);
 569                 mp->m_len = NFSMSIZ(mp);
 570                 if (len == 0) {
 571                         mp3 = mp2 = mp;
 572                 } else {
 573                         mp2->m_next = mp;
 574                         mp2 = mp;
 575                 }
 576                 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
 577                         mp->m_len = NFS_MAXPATHLEN - len;
 578                         len = NFS_MAXPATHLEN;
 579                 } else {
 580                         len += mp->m_len;
 581                 }
 582                 ivp->iov_base = mtod(mp, caddr_t);
 583                 ivp->iov_len = mp->m_len;
 584                 i++;
 585                 ivp++;
 586         }
 587         uiop->uio_iov = iv;
 588         uiop->uio_iovcnt = i;
 589         uiop->uio_offset = 0;
 590         uiop->uio_resid = len;
 591         uiop->uio_rw = UIO_READ;
 592         uiop->uio_segflg = UIO_SYSSPACE;
 593         uiop->uio_td = NULL;
 594         error = VOP_READLINK(vp, uiop, cred);
 595         if (error) {
 596                 m_freem(mp3);
 597                 *lenp = 0;
 598                 goto out;
 599         }
 600         if (uiop->uio_resid > 0) {
 601                 len -= uiop->uio_resid;
 602                 tlen = NFSM_RNDUP(len);
 603                 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
 604         }
 605         *lenp = len;
 606         *mpp = mp3;
 607         *mpendp = mp;
 608
 609 out:
 610         NFSEXITCODE(error);
 611         return (error);
 612 }
 613
 614 /*
 615  * Read vnode op call into mbuf list.
 616  */
 617 int
 618 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
 619     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
 620 {
 621         struct mbuf *m;
 622         int i;
 623         struct iovec *iv;
 624         struct iovec *iv2;
 625         int error = 0, len, left, siz, tlen, ioflag = 0;
 626         struct mbuf *m2 = NULL, *m3;
 627         struct uio io, *uiop = &io;
 628         struct nfsheur *nh;
 629
 630         len = left = NFSM_RNDUP(cnt);
 631         m3 = NULL;
 632         /*
 633          * Generate the mbuf list with the uio_iov ref. to it.
 634          */
 635         i = 0;
 636         while (left > 0) {
 637                 NFSMGET(m);
 638                 MCLGET(m, M_WAIT);
 639                 m->m_len = 0;
 640                 siz = min(M_TRAILINGSPACE(m), left);
 641                 left -= siz;
 642                 i++;
 643                 if (m3)
 644                         m2->m_next = m;
 645                 else
 646                         m3 = m;
 647                 m2 = m;
 648         }
 649         MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
 650             M_TEMP, M_WAITOK);
 651         uiop->uio_iov = iv2 = iv;
 652         m = m3;
 653         left = len;
 654         i = 0;
 655         while (left > 0) {
 656                 if (m == NULL)
 657                         panic("nfsvno_read iov");
 658                 siz = min(M_TRAILINGSPACE(m), left);
 659                 if (siz > 0) {
 660                         iv->iov_base = mtod(m, caddr_t) + m->m_len;
 661                         iv->iov_len = siz;
 662                         m->m_len += siz;
 663                         left -= siz;
 664                         iv++;
 665                         i++;
 666                 }
 667                 m = m->m_next;
 668         }
 669         uiop->uio_iovcnt = i;
 670         uiop->uio_offset = off;
 671         uiop->uio_resid = len;
 672         uiop->uio_rw = UIO_READ;
 673         uiop->uio_segflg = UIO_SYSSPACE;
 674         nh = nfsrv_sequential_heuristic(uiop, vp);
 675         ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
 676         error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
 677         FREE((caddr_t)iv2, M_TEMP);
 678         if (error) {
 679                 m_freem(m3);
 680                 *mpp = NULL;
 681                 goto out;
 682         }
 683         nh->nh_nextoff = uiop->uio_offset;
 684         tlen = len - uiop->uio_resid;
 685         cnt = cnt < tlen ? cnt : tlen;
 686         tlen = NFSM_RNDUP(cnt);
 687         if (tlen == 0) {
 688                 m_freem(m3);
 689                 m3 = NULL;
 690         } else if (len != tlen || tlen != cnt)
 691                 nfsrv_adj(m3, len - tlen, tlen - cnt);
 692         *mpp = m3;
 693         *mpendp = m2;
 694
 695 out:
 696         NFSEXITCODE(error);
 697         return (error);
 698 }
 699
 700 /*
 701  * Write vnode op from an mbuf list.
 702  */
 703 int
 704 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
 705     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
 706 {
 707         struct iovec *ivp;
 708         int i, len;
 709         struct iovec *iv;
 710         int ioflags, error;
 711         struct uio io, *uiop = &io;
 712         struct nfsheur *nh;
 713
 714         MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
 715             M_WAITOK);
 716         uiop->uio_iov = iv = ivp;
 717         uiop->uio_iovcnt = cnt;
 718         i = mtod(mp, caddr_t) + mp->m_len - cp;
 719         len = retlen;
 720         while (len > 0) {
 721                 if (mp == NULL)
 722                         panic("nfsvno_write");
 723                 if (i > 0) {
 724                         i = min(i, len);
 725                         ivp->iov_base = cp;
 726                         ivp->iov_len = i;
 727                         ivp++;
 728                         len -= i;
 729                 }
 730                 mp = mp->m_next;
 731                 if (mp) {
 732                         i = mp->m_len;
 733                         cp = mtod(mp, caddr_t);
 734                 }
 735         }
 736
 737         if (stable == NFSWRITE_UNSTABLE)
 738                 ioflags = IO_NODELOCKED;
 739         else
 740                 ioflags = (IO_SYNC | IO_NODELOCKED);
 741         uiop->uio_resid = retlen;
 742         uiop->uio_rw = UIO_WRITE;
 743         uiop->uio_segflg = UIO_SYSSPACE;
 744         NFSUIOPROC(uiop, p);
 745         uiop->uio_offset = off;
 746         nh = nfsrv_sequential_heuristic(uiop, vp);
 747         ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
 748         error = VOP_WRITE(vp, uiop, ioflags, cred);
 749         if (error == 0)
 750                 nh->nh_nextoff = uiop->uio_offset;
 751         FREE((caddr_t)iv, M_TEMP);
 752
 753         NFSEXITCODE(error);
 754         return (error);
 755 }
 756
 757 /*
 758  * Common code for creating a regular file (plus special files for V2).
 759  */
 760 int
 761 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
 762     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
 763     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
 764 {
 765         u_quad_t tempsize;
 766         int error;
 767
 768         error = nd->nd_repstat;
 769         if (!error && ndp->ni_vp == NULL) {
 770                 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
 771                         vrele(ndp->ni_startdir);
 772                         error = VOP_CREATE(ndp->ni_dvp,
 773                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
 774                         vput(ndp->ni_dvp);
 775                         nfsvno_relpathbuf(ndp);
 776                         if (!error) {
 777                                 if (*exclusive_flagp) {
 778                                         *exclusive_flagp = 0;
 779                                         NFSVNO_ATTRINIT(nvap);
 780                                         nvap->na_atime.tv_sec = cverf[0];
 781                                         nvap->na_atime.tv_nsec = cverf[1];
 782                                         error = VOP_SETATTR(ndp->ni_vp,
 783                                             &nvap->na_vattr, nd->nd_cred);
 784                                 }
 785                         }
 786                 /*
 787                  * NFS V2 Only. nfsrvd_mknod() does this for V3.
 788                  * (This implies, just get out on an error.)
 789                  */
 790                 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
 791                         nvap->na_type == VFIFO) {
 792                         if (nvap->na_type == VCHR && rdev == 0xffffffff)
 793                                 nvap->na_type = VFIFO;
 794                         if (nvap->na_type != VFIFO &&
 795                             (error = priv_check_cred(nd->nd_cred,
 796                              PRIV_VFS_MKNOD_DEV, 0))) {
 797                                 vrele(ndp->ni_startdir);
 798                                 nfsvno_relpathbuf(ndp);
 799                                 vput(ndp->ni_dvp);
 800                                 goto out;
 801                         }
 802                         nvap->na_rdev = rdev;
 803                         error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 804                             &ndp->ni_cnd, &nvap->na_vattr);
 805                         vput(ndp->ni_dvp);
 806                         nfsvno_relpathbuf(ndp);
 807                         vrele(ndp->ni_startdir);
 808                         if (error)
 809                                 goto out;
 810                 } else {
 811                         vrele(ndp->ni_startdir);
 812                         nfsvno_relpathbuf(ndp);
 813                         vput(ndp->ni_dvp);
 814                         error = ENXIO;
 815                         goto out;
 816                 }
 817                 *vpp = ndp->ni_vp;
 818         } else {
 819                 /*
 820                  * Handle cases where error is already set and/or
 821                  * the file exists.
 822                  * 1 - clean up the lookup
 823                  * 2 - iff !error and na_size set, truncate it
 824                  */
 825                 vrele(ndp->ni_startdir);
 826                 nfsvno_relpathbuf(ndp);
 827                 *vpp = ndp->ni_vp;
 828                 if (ndp->ni_dvp == *vpp)
 829                         vrele(ndp->ni_dvp);
 830                 else
 831                         vput(ndp->ni_dvp);
 832                 if (!error && nvap->na_size != VNOVAL) {
 833                         error = nfsvno_accchk(*vpp, VWRITE,
 834                             nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
 835                             NFSACCCHK_VPISLOCKED, NULL);
 836                         if (!error) {
 837                                 tempsize = nvap->na_size;
 838                                 NFSVNO_ATTRINIT(nvap);
 839                                 nvap->na_size = tempsize;
 840                                 error = VOP_SETATTR(*vpp,
 841                                     &nvap->na_vattr, nd->nd_cred);
 842                         }
 843                 }
 844                 if (error)
 845                         vput(*vpp);
 846         }
 847
 848 out:
 849         NFSEXITCODE(error);
 850         return (error);
 851 }
 852
 853 /*
 854  * Do a mknod vnode op.
 855  */
 856 int
 857 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
 858     struct thread *p)
 859 {
 860         int error = 0;
 861         enum vtype vtyp;
 862
 863         vtyp = nvap->na_type;
 864         /*
 865          * Iff doesn't exist, create it.
 866          */
 867         if (ndp->ni_vp) {
 868                 vrele(ndp->ni_startdir);
 869                 nfsvno_relpathbuf(ndp);
 870                 vput(ndp->ni_dvp);
 871                 vrele(ndp->ni_vp);
 872                 error = EEXIST;
 873                 goto out;
 874         }
 875         if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
 876                 vrele(ndp->ni_startdir);
 877                 nfsvno_relpathbuf(ndp);
 878                 vput(ndp->ni_dvp);
 879                 error = NFSERR_BADTYPE;
 880                 goto out;
 881         }
 882         if (vtyp == VSOCK) {
 883                 vrele(ndp->ni_startdir);
 884                 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
 885                     &ndp->ni_cnd, &nvap->na_vattr);
 886                 vput(ndp->ni_dvp);
 887                 nfsvno_relpathbuf(ndp);
 888         } else {
 889                 if (nvap->na_type != VFIFO &&
 890                     (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
 891                         vrele(ndp->ni_startdir);
 892                         nfsvno_relpathbuf(ndp);
 893                         vput(ndp->ni_dvp);
 894                         goto out;
 895                 }
 896                 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 897                     &ndp->ni_cnd, &nvap->na_vattr);
 898                 vput(ndp->ni_dvp);
 899                 nfsvno_relpathbuf(ndp);
 900                 vrele(ndp->ni_startdir);
 901                 /*
 902                  * Since VOP_MKNOD returns the ni_vp, I can't
 903                  * see any reason to do the lookup.
 904                  */
 905         }
 906
 907 out:
 908         NFSEXITCODE(error);
 909         return (error);
 910 }
 911
 912 /*
 913  * Mkdir vnode op.
 914  */
 915 int
 916 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
 917     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
 918 {
 919         int error = 0;
 920
 921         if (ndp->ni_vp != NULL) {
 922                 if (ndp->ni_dvp == ndp->ni_vp)
 923                         vrele(ndp->ni_dvp);
 924                 else
 925                         vput(ndp->ni_dvp);
 926                 vrele(ndp->ni_vp);
 927                 nfsvno_relpathbuf(ndp);
 928                 error = EEXIST;
 929                 goto out;
 930         }
 931         error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 932             &nvap->na_vattr);
 933         vput(ndp->ni_dvp);
 934         nfsvno_relpathbuf(ndp);
 935
 936 out:
 937         NFSEXITCODE(error);
 938         return (error);
 939 }
 940
 941 /*
 942  * symlink vnode op.
 943  */
 944 int
 945 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
 946     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
 947     struct nfsexstuff *exp)
 948 {
 949         int error = 0;
 950
 951         if (ndp->ni_vp) {
 952                 vrele(ndp->ni_startdir);
 953                 nfsvno_relpathbuf(ndp);
 954                 if (ndp->ni_dvp == ndp->ni_vp)
 955                         vrele(ndp->ni_dvp);
 956                 else
 957                         vput(ndp->ni_dvp);
 958                 vrele(ndp->ni_vp);
 959                 error = EEXIST;
 960                 goto out;
 961         }
 962
 963         error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 964             &nvap->na_vattr, pathcp);
 965         vput(ndp->ni_dvp);
 966         vrele(ndp->ni_startdir);
 967         nfsvno_relpathbuf(ndp);
 968         /*
 969          * Although FreeBSD still had the lookup code in
 970          * it for 7/current, there doesn't seem to be any
 971          * point, since VOP_SYMLINK() returns the ni_vp.
 972          * Just vput it for v2.
 973          */
 974         if (!not_v2 && !error)
 975                 vput(ndp->ni_vp);
 976
 977 out:
 978         NFSEXITCODE(error);
 979         return (error);
 980 }
 981
 982 /*
 983  * Parse symbolic link arguments.
 984  * This function has an ugly side effect. It will MALLOC() an area for
 985  * the symlink and set iov_base to point to it, only if it succeeds.
 986  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
 987  * be FREE'd later.
 988  */
 989 int
 990 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
 991     struct thread *p, char **pathcpp, int *lenp)
 992 {
 993         u_int32_t *tl;
 994         char *pathcp = NULL;
 995         int error = 0, len;
 996         struct nfsv2_sattr *sp;
 997
 998         *pathcpp = NULL;
 999         *lenp = 0;
1000         if ((nd->nd_flag & ND_NFSV3) &&
1001             (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1002                 goto nfsmout;
1003         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1004         len = fxdr_unsigned(int, *tl);
1005         if (len > NFS_MAXPATHLEN || len <= 0) {
1006                 error = EBADRPC;
1007                 goto nfsmout;
1008         }
1009         MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1010         error = nfsrv_mtostr(nd, pathcp, len);
1011         if (error)
1012                 goto nfsmout;
1013         if (nd->nd_flag & ND_NFSV2) {
1014                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1015                 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1016         }
1017         *pathcpp = pathcp;
1018         *lenp = len;
1019         NFSEXITCODE2(0, nd);
1020         return (0);
1021 nfsmout:
1022         if (pathcp)
1023                 free(pathcp, M_TEMP);
1024         NFSEXITCODE2(error, nd);
1025         return (error);
1026 }
1027
1028 /*
1029  * Remove a non-directory object.
1030  */
1031 int
1032 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1033     struct thread *p, struct nfsexstuff *exp)
1034 {
1035         struct vnode *vp;
1036         int error = 0;
1037
1038         vp = ndp->ni_vp;
1039         if (vp->v_type == VDIR)
1040                 error = NFSERR_ISDIR;
1041         else if (is_v4)
1042                 error = nfsrv_checkremove(vp, 1, p);
1043         if (!error)
1044                 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1045         if (ndp->ni_dvp == vp)
1046                 vrele(ndp->ni_dvp);
1047         else
1048                 vput(ndp->ni_dvp);
1049         vput(vp);
1050         if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1051                 nfsvno_relpathbuf(ndp);
1052         NFSEXITCODE(error);
1053         return (error);
1054 }
1055
1056 /*
1057  * Remove a directory.
1058  */
1059 int
1060 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1061     struct thread *p, struct nfsexstuff *exp)
1062 {
1063         struct vnode *vp;
1064         int error = 0;
1065
1066         vp = ndp->ni_vp;
1067         if (vp->v_type != VDIR) {
1068                 error = ENOTDIR;
1069                 goto out;
1070         }
1071         /*
1072          * No rmdir "." please.
1073          */
1074         if (ndp->ni_dvp == vp) {
1075                 error = EINVAL;
1076                 goto out;
1077         }
1078         /*
1079          * The root of a mounted filesystem cannot be deleted.
1080          */
1081         if (vp->v_vflag & VV_ROOT)
1082                 error = EBUSY;
1083 out:
1084         if (!error)
1085                 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1086         if (ndp->ni_dvp == vp)
1087                 vrele(ndp->ni_dvp);
1088         else
1089                 vput(ndp->ni_dvp);
1090         vput(vp);
1091         if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1092                 nfsvno_relpathbuf(ndp);
1093         NFSEXITCODE(error);
1094         return (error);
1095 }
1096
1097 /*
1098  * Rename vnode op.
1099  */
1100 int
1101 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1102     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1103 {
1104         struct vnode *fvp, *tvp, *tdvp;
1105         int error = 0;
1106
1107         fvp = fromndp->ni_vp;
1108         if (ndstat) {
1109                 vrele(fromndp->ni_dvp);
1110                 vrele(fvp);
1111                 error = ndstat;
1112                 goto out1;
1113         }
1114         tdvp = tondp->ni_dvp;
1115         tvp = tondp->ni_vp;
1116         if (tvp != NULL) {
1117                 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1118                         error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1119                         goto out;
1120                 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1121                         error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1122                         goto out;
1123                 }
1124                 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1125                         error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1126                         goto out;
1127                 }
1128
1129                 /*
1130                  * A rename to '.' or '..' results in a prematurely
1131                  * unlocked vnode on FreeBSD5, so I'm just going to fail that
1132                  * here.
1133                  */
1134                 if ((tondp->ni_cnd.cn_namelen == 1 &&
1135                      tondp->ni_cnd.cn_nameptr[0] == '.') ||
1136                     (tondp->ni_cnd.cn_namelen == 2 &&
1137                      tondp->ni_cnd.cn_nameptr[0] == '.' &&
1138                      tondp->ni_cnd.cn_nameptr[1] == '.')) {
1139                         error = EINVAL;
1140                         goto out;
1141                 }
1142         }
1143         if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1144                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1145                 goto out;
1146         }
1147         if (fvp->v_mount != tdvp->v_mount) {
1148                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1149                 goto out;
1150         }
1151         if (fvp == tdvp) {
1152                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1153                 goto out;
1154         }
1155         if (fvp == tvp) {
1156                 /*
1157                  * If source and destination are the same, there is nothing to
1158                  * do. Set error to -1 to indicate this.
1159                  */
1160                 error = -1;
1161                 goto out;
1162         }
1163         if (ndflag & ND_NFSV4) {
1164                 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1165                         error = nfsrv_checkremove(fvp, 0, p);
1166                         NFSVOPUNLOCK(fvp, 0);
1167                 } else
1168                         error = EPERM;
1169                 if (tvp && !error)
1170                         error = nfsrv_checkremove(tvp, 1, p);
1171         } else {
1172                 /*
1173                  * For NFSv2 and NFSv3, try to get rid of the delegation, so
1174                  * that the NFSv4 client won't be confused by the rename.
1175                  * Since nfsd_recalldelegation() can only be called on an
1176                  * unlocked vnode at this point and fvp is the file that will
1177                  * still exist after the rename, just do fvp.
1178                  */
1179                 nfsd_recalldelegation(fvp, p);
1180         }
1181 out:
1182         if (!error) {
1183                 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1184                     &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1185                     &tondp->ni_cnd);
1186         } else {
1187                 if (tdvp == tvp)
1188                         vrele(tdvp);
1189                 else
1190                         vput(tdvp);
1191                 if (tvp)
1192                         vput(tvp);
1193                 vrele(fromndp->ni_dvp);
1194                 vrele(fvp);
1195                 if (error == -1)
1196                         error = 0;
1197         }
1198         vrele(tondp->ni_startdir);
1199         nfsvno_relpathbuf(tondp);
1200 out1:
1201         vrele(fromndp->ni_startdir);
1202         nfsvno_relpathbuf(fromndp);
1203         NFSEXITCODE(error);
1204         return (error);
1205 }
1206
1207 /*
1208  * Link vnode op.
1209  */
1210 int
1211 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1212     struct thread *p, struct nfsexstuff *exp)
1213 {
1214         struct vnode *xp;
1215         int error = 0;
1216
1217         xp = ndp->ni_vp;
1218         if (xp != NULL) {
1219                 error = EEXIST;
1220         } else {
1221                 xp = ndp->ni_dvp;
1222                 if (vp->v_mount != xp->v_mount)
1223                         error = EXDEV;
1224         }
1225         if (!error) {
1226                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1227                 if ((vp->v_iflag & VI_DOOMED) == 0)
1228                         error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1229                 else
1230                         error = EPERM;
1231                 if (ndp->ni_dvp == vp)
1232                         vrele(ndp->ni_dvp);
1233                 else
1234                         vput(ndp->ni_dvp);
1235                 NFSVOPUNLOCK(vp, 0);
1236         } else {
1237                 if (ndp->ni_dvp == ndp->ni_vp)
1238                         vrele(ndp->ni_dvp);
1239                 else
1240                         vput(ndp->ni_dvp);
1241                 if (ndp->ni_vp)
1242                         vrele(ndp->ni_vp);
1243         }
1244         nfsvno_relpathbuf(ndp);
1245         NFSEXITCODE(error);
1246         return (error);
1247 }
1248
1249 /*
1250  * Do the fsync() appropriate for the commit.
1251  */
1252 int
1253 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1254     struct thread *td)
1255 {
1256         int error = 0;
1257
1258         /*
1259          * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1260          * file is done.  At this time VOP_FSYNC does not accept offset and
1261          * byte count parameters so call VOP_FSYNC the whole file for now.
1262          * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1263          */
1264         if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1265                 /*
1266                  * Give up and do the whole thing
1267                  */
1268                 if (vp->v_object &&
1269                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1270                         VM_OBJECT_LOCK(vp->v_object);
1271                         vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1272                         VM_OBJECT_UNLOCK(vp->v_object);
1273                 }
1274                 error = VOP_FSYNC(vp, MNT_WAIT, td);
1275         } else {
1276                 /*
1277                  * Locate and synchronously write any buffers that fall
1278                  * into the requested range.  Note:  we are assuming that
1279                  * f_iosize is a power of 2.
1280                  */
1281                 int iosize = vp->v_mount->mnt_stat.f_iosize;
1282                 int iomask = iosize - 1;
1283                 struct bufobj *bo;
1284                 daddr_t lblkno;
1285
1286                 /*
1287                  * Align to iosize boundry, super-align to page boundry.
1288                  */
1289                 if (off & iomask) {
1290                         cnt += off & iomask;
1291                         off &= ~(u_quad_t)iomask;
1292                 }
1293                 if (off & PAGE_MASK) {
1294                         cnt += off & PAGE_MASK;
1295                         off &= ~(u_quad_t)PAGE_MASK;
1296                 }
1297                 lblkno = off / iosize;
1298
1299                 if (vp->v_object &&
1300                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1301                         VM_OBJECT_LOCK(vp->v_object);
1302                         vm_object_page_clean(vp->v_object, off, off + cnt,
1303                             OBJPC_SYNC);
1304                         VM_OBJECT_UNLOCK(vp->v_object);
1305                 }
1306
1307                 bo = &vp->v_bufobj;
1308                 BO_LOCK(bo);
1309                 while (cnt > 0) {
1310                         struct buf *bp;
1311
1312                         /*
1313                          * If we have a buffer and it is marked B_DELWRI we
1314                          * have to lock and write it.  Otherwise the prior
1315                          * write is assumed to have already been committed.
1316                          *
1317                          * gbincore() can return invalid buffers now so we
1318                          * have to check that bit as well (though B_DELWRI
1319                          * should not be set if B_INVAL is set there could be
1320                          * a race here since we haven't locked the buffer).
1321                          */
1322                         if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1323                                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1324                                     LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1325                                         BO_LOCK(bo);
1326                                         continue; /* retry */
1327                                 }
1328                                 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1329                                     B_DELWRI) {
1330                                         bremfree(bp);
1331                                         bp->b_flags &= ~B_ASYNC;
1332                                         bwrite(bp);
1333                                         ++nfs_commit_miss;
1334                                 } else
1335                                         BUF_UNLOCK(bp);
1336                                 BO_LOCK(bo);
1337                         }
1338                         ++nfs_commit_blks;
1339                         if (cnt < iosize)
1340                                 break;
1341                         cnt -= iosize;
1342                         ++lblkno;
1343                 }
1344                 BO_UNLOCK(bo);
1345         }
1346         NFSEXITCODE(error);
1347         return (error);
1348 }
1349
1350 /*
1351  * Statfs vnode op.
1352  */
1353 int
1354 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1355 {
1356         int error;
1357
1358         error = VFS_STATFS(vp->v_mount, sf);
1359         if (error == 0) {
1360                 /*
1361                  * Since NFS handles these values as unsigned on the
1362                  * wire, there is no way to represent negative values,
1363                  * so set them to 0. Without this, they will appear
1364                  * to be very large positive values for clients like
1365                  * Solaris10.
1366                  */
1367                 if (sf->f_bavail < 0)
1368                         sf->f_bavail = 0;
1369                 if (sf->f_ffree < 0)
1370                         sf->f_ffree = 0;
1371         }
1372         NFSEXITCODE(error);
1373         return (error);
1374 }
1375
1376 /*
1377  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1378  * must handle nfsrv_opencheck() calls after any other access checks.
1379  */
1380 void
1381 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1382     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1383     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1384     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1385     struct nfsexstuff *exp, struct vnode **vpp)
1386 {
1387         struct vnode *vp = NULL;
1388         u_quad_t tempsize;
1389         struct nfsexstuff nes;
1390
1391         if (ndp->ni_vp == NULL)
1392                 nd->nd_repstat = nfsrv_opencheck(clientid,
1393                     stateidp, stp, NULL, nd, p, nd->nd_repstat);
1394         if (!nd->nd_repstat) {
1395                 if (ndp->ni_vp == NULL) {
1396                         vrele(ndp->ni_startdir);
1397                         nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1398                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1399                         vput(ndp->ni_dvp);
1400                         nfsvno_relpathbuf(ndp);
1401                         if (!nd->nd_repstat) {
1402                                 if (*exclusive_flagp) {
1403                                         *exclusive_flagp = 0;
1404                                         NFSVNO_ATTRINIT(nvap);
1405                                         nvap->na_atime.tv_sec = cverf[0];
1406                                         nvap->na_atime.tv_nsec = cverf[1];
1407                                         nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1408                                             &nvap->na_vattr, cred);
1409                                 } else {
1410                                         nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1411                                             aclp, p, attrbitp, exp);
1412                                 }
1413                         }
1414                         vp = ndp->ni_vp;
1415                 } else {
1416                         if (ndp->ni_startdir)
1417                                 vrele(ndp->ni_startdir);
1418                         nfsvno_relpathbuf(ndp);
1419                         vp = ndp->ni_vp;
1420                         if (create == NFSV4OPEN_CREATE) {
1421                                 if (ndp->ni_dvp == vp)
1422                                         vrele(ndp->ni_dvp);
1423                                 else
1424                                         vput(ndp->ni_dvp);
1425                         }
1426                         if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1427                                 if (ndp->ni_cnd.cn_flags & RDONLY)
1428                                         NFSVNO_SETEXRDONLY(&nes);
1429                                 else
1430                                         NFSVNO_EXINIT(&nes);
1431                                 nd->nd_repstat = nfsvno_accchk(vp,
1432                                     VWRITE, cred, &nes, p,
1433                                     NFSACCCHK_NOOVERRIDE,
1434                                     NFSACCCHK_VPISLOCKED, NULL);
1435                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1436                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1437                                 if (!nd->nd_repstat) {
1438                                         tempsize = nvap->na_size;
1439                                         NFSVNO_ATTRINIT(nvap);
1440                                         nvap->na_size = tempsize;
1441                                         nd->nd_repstat = VOP_SETATTR(vp,
1442                                             &nvap->na_vattr, cred);
1443                                 }
1444                         } else if (vp->v_type == VREG) {
1445                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1446                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1447                         }
1448                 }
1449         } else {
1450                 if (ndp->ni_cnd.cn_flags & HASBUF)
1451                         nfsvno_relpathbuf(ndp);
1452                 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1453                         vrele(ndp->ni_startdir);
1454                         if (ndp->ni_dvp == ndp->ni_vp)
1455                                 vrele(ndp->ni_dvp);
1456                         else
1457                                 vput(ndp->ni_dvp);
1458                         if (ndp->ni_vp)
1459                                 vput(ndp->ni_vp);
1460                 }
1461         }
1462         *vpp = vp;
1463
1464         NFSEXITCODE2(0, nd);
1465 }
1466
1467 /*
1468  * Updates the file rev and sets the mtime and ctime
1469  * to the current clock time, returning the va_filerev and va_Xtime
1470  * values.
1471  */
1472 void
1473 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1474     struct ucred *cred, struct thread *p)
1475 {
1476         struct vattr va;
1477
1478         VATTR_NULL(&va);
1479         getnanotime(&va.va_mtime);
1480         (void) VOP_SETATTR(vp, &va, cred);
1481         (void) nfsvno_getattr(vp, nvap, cred, p, 1);
1482 }
1483
1484 /*
1485  * Glue routine to nfsv4_fillattr().
1486  */
1487 int
1488 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1489     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1490     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1491     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1492 {
1493         int error;
1494
1495         error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1496             attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1497             mounted_on_fileno);
1498         NFSEXITCODE2(0, nd);
1499         return (error);
1500 }
1501
1502 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1503 /*
1504  * nfs readdir service
1505  * - mallocs what it thinks is enough to read
1506  *      count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1507  * - calls VOP_READDIR()
1508  * - loops around building the reply
1509  *      if the output generated exceeds count break out of loop
1510  *      The NFSM_CLGET macro is used here so that the reply will be packed
1511  *      tightly in mbuf clusters.
1512  * - it trims out records with d_fileno == 0
1513  *      this doesn't matter for Unix clients, but they might confuse clients
1514  *      for other os'.
1515  * - it trims out records with d_type == DT_WHT
1516  *      these cannot be seen through NFS (unless we extend the protocol)
1517  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1518  * PS: The NFS protocol spec. does not clarify what the "count" byte
1519  *      argument is a count of.. just name strings and file id's or the
1520  *      entire reply rpc or ...
1521  *      I tried just file name and id sizes and it confused the Sun client,
1522  *      so I am using the full rpc size now. The "paranoia.." comment refers
1523  *      to including the status longwords that are not a part of the dir.
1524  *      "entry" structures, but are in the rpc.
1525  */
1526 int
1527 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1528     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1529 {
1530         struct dirent *dp;
1531         u_int32_t *tl;
1532         int dirlen;
1533         char *cpos, *cend, *rbuf;
1534         struct nfsvattr at;
1535         int nlen, error = 0, getret = 1;
1536         int siz, cnt, fullsiz, eofflag, ncookies;
1537         u_int64_t off, toff, verf;
1538         u_long *cookies = NULL, *cookiep;
1539         struct uio io;
1540         struct iovec iv;
1541         int not_zfs;
1542
1543         if (nd->nd_repstat) {
1544                 nfsrv_postopattr(nd, getret, &at);
1545                 goto out;
1546         }
1547         if (nd->nd_flag & ND_NFSV2) {
1548                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1549                 off = fxdr_unsigned(u_quad_t, *tl++);
1550         } else {
1551                 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1552                 off = fxdr_hyper(tl);
1553                 tl += 2;
1554                 verf = fxdr_hyper(tl);
1555                 tl += 2;
1556         }
1557         toff = off;
1558         cnt = fxdr_unsigned(int, *tl);
1559         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1560                 cnt = NFS_SRVMAXDATA(nd);
1561         siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1562         fullsiz = siz;
1563         if (nd->nd_flag & ND_NFSV3) {
1564                 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1565                     p, 1);
1566 #if 0
1567                 /*
1568                  * va_filerev is not sufficient as a cookie verifier,
1569                  * since it is not supposed to change when entries are
1570                  * removed/added unless that offset cookies returned to
1571                  * the client are no longer valid.
1572                  */
1573                 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1574                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1575 #endif
1576         }
1577         if (nd->nd_repstat == 0 && cnt == 0) {
1578                 if (nd->nd_flag & ND_NFSV2)
1579                         /* NFSv2 does not have NFSERR_TOOSMALL */
1580                         nd->nd_repstat = EPERM;
1581                 else
1582                         nd->nd_repstat = NFSERR_TOOSMALL;
1583         }
1584         if (!nd->nd_repstat)
1585                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1586                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1587                     NFSACCCHK_VPISLOCKED, NULL);
1588         if (nd->nd_repstat) {
1589                 vput(vp);
1590                 if (nd->nd_flag & ND_NFSV3)
1591                         nfsrv_postopattr(nd, getret, &at);
1592                 goto out;
1593         }
1594         not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1595         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1596 again:
1597         eofflag = 0;
1598         if (cookies) {
1599                 free((caddr_t)cookies, M_TEMP);
1600                 cookies = NULL;
1601         }
1602
1603         iv.iov_base = rbuf;
1604         iv.iov_len = siz;
1605         io.uio_iov = &iv;
1606         io.uio_iovcnt = 1;
1607         io.uio_offset = (off_t)off;
1608         io.uio_resid = siz;
1609         io.uio_segflg = UIO_SYSSPACE;
1610         io.uio_rw = UIO_READ;
1611         io.uio_td = NULL;
1612         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1613             &cookies);
1614         off = (u_int64_t)io.uio_offset;
1615         if (io.uio_resid)
1616                 siz -= io.uio_resid;
1617
1618         if (!cookies && !nd->nd_repstat)
1619                 nd->nd_repstat = NFSERR_PERM;
1620         if (nd->nd_flag & ND_NFSV3) {
1621                 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1622                 if (!nd->nd_repstat)
1623                         nd->nd_repstat = getret;
1624         }
1625
1626         /*
1627          * Handles the failed cases. nd->nd_repstat == 0 past here.
1628          */
1629         if (nd->nd_repstat) {
1630                 vput(vp);
1631                 free((caddr_t)rbuf, M_TEMP);
1632                 if (cookies)
1633                         free((caddr_t)cookies, M_TEMP);
1634                 if (nd->nd_flag & ND_NFSV3)
1635                         nfsrv_postopattr(nd, getret, &at);
1636                 goto out;
1637         }
1638         /*
1639          * If nothing read, return eof
1640          * rpc reply
1641          */
1642         if (siz == 0) {
1643                 vput(vp);
1644                 if (nd->nd_flag & ND_NFSV2) {
1645                         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1646                 } else {
1647                         nfsrv_postopattr(nd, getret, &at);
1648                         NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1649                         txdr_hyper(at.na_filerev, tl);
1650                         tl += 2;
1651                 }
1652                 *tl++ = newnfs_false;
1653                 *tl = newnfs_true;
1654                 FREE((caddr_t)rbuf, M_TEMP);
1655                 FREE((caddr_t)cookies, M_TEMP);
1656                 goto out;
1657         }
1658
1659         /*
1660          * Check for degenerate cases of nothing useful read.
1661          * If so go try again
1662          */
1663         cpos = rbuf;
1664         cend = rbuf + siz;
1665         dp = (struct dirent *)cpos;
1666         cookiep = cookies;
1667
1668         /*
1669          * For some reason FreeBSD's ufs_readdir() chooses to back the
1670          * directory offset up to a block boundary, so it is necessary to
1671          * skip over the records that precede the requested offset. This
1672          * requires the assumption that file offset cookies monotonically
1673          * increase.
1674          * Since the offset cookies don't monotonically increase for ZFS,
1675          * this is not done when ZFS is the file system.
1676          */
1677         while (cpos < cend && ncookies > 0 &&
1678             (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1679              (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1680                 cpos += dp->d_reclen;
1681                 dp = (struct dirent *)cpos;
1682                 cookiep++;
1683                 ncookies--;
1684         }
1685         if (cpos >= cend || ncookies == 0) {
1686                 siz = fullsiz;
1687                 toff = off;
1688                 goto again;
1689         }
1690         vput(vp);
1691
1692         /*
1693          * dirlen is the size of the reply, including all XDR and must
1694          * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1695          * if the XDR should be included in "count", but to be safe, we do.
1696          * (Include the two booleans at the end of the reply in dirlen now.)
1697          */
1698         if (nd->nd_flag & ND_NFSV3) {
1699                 nfsrv_postopattr(nd, getret, &at);
1700                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1701                 txdr_hyper(at.na_filerev, tl);
1702                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1703         } else {
1704                 dirlen = 2 * NFSX_UNSIGNED;
1705         }
1706
1707         /* Loop through the records and build reply */
1708         while (cpos < cend && ncookies > 0) {
1709                 nlen = dp->d_namlen;
1710                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1711                         nlen <= NFS_MAXNAMLEN) {
1712                         if (nd->nd_flag & ND_NFSV3)
1713                                 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1714                         else
1715                                 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1716                         if (dirlen > cnt) {
1717                                 eofflag = 0;
1718                                 break;
1719                         }
1720
1721                         /*
1722                          * Build the directory record xdr from
1723                          * the dirent entry.
1724                          */
1725                         if (nd->nd_flag & ND_NFSV3) {
1726                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1727                                 *tl++ = newnfs_true;
1728                                 *tl++ = 0;
1729                         } else {
1730                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1731                                 *tl++ = newnfs_true;
1732                         }
1733                         *tl = txdr_unsigned(dp->d_fileno);
1734                         (void) nfsm_strtom(nd, dp->d_name, nlen);
1735                         if (nd->nd_flag & ND_NFSV3) {
1736                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1737                                 *tl++ = 0;
1738                         } else
1739                                 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1740                         *tl = txdr_unsigned(*cookiep);
1741                 }
1742                 cpos += dp->d_reclen;
1743                 dp = (struct dirent *)cpos;
1744                 cookiep++;
1745                 ncookies--;
1746         }
1747         if (cpos < cend)
1748                 eofflag = 0;
1749         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1750         *tl++ = newnfs_false;
1751         if (eofflag)
1752                 *tl = newnfs_true;
1753         else
1754                 *tl = newnfs_false;
1755         FREE((caddr_t)rbuf, M_TEMP);
1756         FREE((caddr_t)cookies, M_TEMP);
1757
1758 out:
1759         NFSEXITCODE2(0, nd);
1760         return (0);
1761 nfsmout:
1762         vput(vp);
1763         NFSEXITCODE2(error, nd);
1764         return (error);
1765 }
1766
1767 /*
1768  * Readdirplus for V3 and Readdir for V4.
1769  */
1770 int
1771 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1772     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1773 {
1774         struct dirent *dp;
1775         u_int32_t *tl;
1776         int dirlen;
1777         char *cpos, *cend, *rbuf;
1778         struct vnode *nvp;
1779         fhandle_t nfh;
1780         struct nfsvattr nva, at, *nvap = &nva;
1781         struct mbuf *mb0, *mb1;
1782         struct nfsreferral *refp;
1783         int nlen, r, error = 0, getret = 1, usevget = 1;
1784         int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1785         caddr_t bpos0, bpos1;
1786         u_int64_t off, toff, verf;
1787         u_long *cookies = NULL, *cookiep;
1788         nfsattrbit_t attrbits, rderrbits, savbits;
1789         struct uio io;
1790         struct iovec iv;
1791         struct componentname cn;
1792         int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1793         struct mount *mp, *new_mp;
1794         uint64_t mounted_on_fileno;
1795
1796         if (nd->nd_repstat) {
1797                 nfsrv_postopattr(nd, getret, &at);
1798                 goto out;
1799         }
1800         NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1801         off = fxdr_hyper(tl);
1802         toff = off;
1803         tl += 2;
1804         verf = fxdr_hyper(tl);
1805         tl += 2;
1806         siz = fxdr_unsigned(int, *tl++);
1807         cnt = fxdr_unsigned(int, *tl);
1808
1809         /*
1810          * Use the server's maximum data transfer size as the upper bound
1811          * on reply datalen.
1812          */
1813         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1814                 cnt = NFS_SRVMAXDATA(nd);
1815
1816         /*
1817          * siz is a "hint" of how much directory information (name, fileid,
1818          * cookie) should be in the reply. At least one client "hints" 0,
1819          * so I set it to cnt for that case. I also round it up to the
1820          * next multiple of DIRBLKSIZ.
1821          */
1822         if (siz <= 0)
1823                 siz = cnt;
1824         siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1825
1826         if (nd->nd_flag & ND_NFSV4) {
1827                 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1828                 if (error)
1829                         goto nfsmout;
1830                 NFSSET_ATTRBIT(&savbits, &attrbits);
1831                 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1832                 NFSZERO_ATTRBIT(&rderrbits);
1833                 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1834         } else {
1835                 NFSZERO_ATTRBIT(&attrbits);
1836         }
1837         fullsiz = siz;
1838         nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1839         if (!nd->nd_repstat) {
1840             if (off && verf != at.na_filerev) {
1841                 /*
1842                  * va_filerev is not sufficient as a cookie verifier,
1843                  * since it is not supposed to change when entries are
1844                  * removed/added unless that offset cookies returned to
1845                  * the client are no longer valid.
1846                  */
1847 #if 0
1848                 if (nd->nd_flag & ND_NFSV4) {
1849                         nd->nd_repstat = NFSERR_NOTSAME;
1850                 } else {
1851                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1852                 }
1853 #endif
1854             } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1855                 nd->nd_repstat = NFSERR_BAD_COOKIE;
1856             }
1857         }
1858         if (!nd->nd_repstat && vp->v_type != VDIR)
1859                 nd->nd_repstat = NFSERR_NOTDIR;
1860         if (!nd->nd_repstat && cnt == 0)
1861                 nd->nd_repstat = NFSERR_TOOSMALL;
1862         if (!nd->nd_repstat)
1863                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1864                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1865                     NFSACCCHK_VPISLOCKED, NULL);
1866         if (nd->nd_repstat) {
1867                 vput(vp);
1868                 if (nd->nd_flag & ND_NFSV3)
1869                         nfsrv_postopattr(nd, getret, &at);
1870                 goto out;
1871         }
1872         not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1873
1874         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1875 again:
1876         eofflag = 0;
1877         if (cookies) {
1878                 free((caddr_t)cookies, M_TEMP);
1879                 cookies = NULL;
1880         }
1881
1882         iv.iov_base = rbuf;
1883         iv.iov_len = siz;
1884         io.uio_iov = &iv;
1885         io.uio_iovcnt = 1;
1886         io.uio_offset = (off_t)off;
1887         io.uio_resid = siz;
1888         io.uio_segflg = UIO_SYSSPACE;
1889         io.uio_rw = UIO_READ;
1890         io.uio_td = NULL;
1891         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1892             &cookies);
1893         off = (u_int64_t)io.uio_offset;
1894         if (io.uio_resid)
1895                 siz -= io.uio_resid;
1896
1897         getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1898
1899         if (!cookies && !nd->nd_repstat)
1900                 nd->nd_repstat = NFSERR_PERM;
1901         if (!nd->nd_repstat)
1902                 nd->nd_repstat = getret;
1903         if (nd->nd_repstat) {
1904                 vput(vp);
1905                 if (cookies)
1906                         free((caddr_t)cookies, M_TEMP);
1907                 free((caddr_t)rbuf, M_TEMP);
1908                 if (nd->nd_flag & ND_NFSV3)
1909                         nfsrv_postopattr(nd, getret, &at);
1910                 goto out;
1911         }
1912         /*
1913          * If nothing read, return eof
1914          * rpc reply
1915          */
1916         if (siz == 0) {
1917                 vput(vp);
1918                 if (nd->nd_flag & ND_NFSV3)
1919                         nfsrv_postopattr(nd, getret, &at);
1920                 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1921                 txdr_hyper(at.na_filerev, tl);
1922                 tl += 2;
1923                 *tl++ = newnfs_false;
1924                 *tl = newnfs_true;
1925                 free((caddr_t)cookies, M_TEMP);
1926                 free((caddr_t)rbuf, M_TEMP);
1927                 goto out;
1928         }
1929
1930         /*
1931          * Check for degenerate cases of nothing useful read.
1932          * If so go try again
1933          */
1934         cpos = rbuf;
1935         cend = rbuf + siz;
1936         dp = (struct dirent *)cpos;
1937         cookiep = cookies;
1938
1939         /*
1940          * For some reason FreeBSD's ufs_readdir() chooses to back the
1941          * directory offset up to a block boundary, so it is necessary to
1942          * skip over the records that precede the requested offset. This
1943          * requires the assumption that file offset cookies monotonically
1944          * increase.
1945          * Since the offset cookies don't monotonically increase for ZFS,
1946          * this is not done when ZFS is the file system.
1947          */
1948         while (cpos < cend && ncookies > 0 &&
1949           (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1950            (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1951            ((nd->nd_flag & ND_NFSV4) &&
1952             ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1953              (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1954                 cpos += dp->d_reclen;
1955                 dp = (struct dirent *)cpos;
1956                 cookiep++;
1957                 ncookies--;
1958         }
1959         if (cpos >= cend || ncookies == 0) {
1960                 siz = fullsiz;
1961                 toff = off;
1962                 goto again;
1963         }
1964
1965         /*
1966          * Busy the file system so that the mount point won't go away
1967          * and, as such, VFS_VGET() can be used safely.
1968          */
1969         mp = vp->v_mount;
1970         vfs_ref(mp);
1971         NFSVOPUNLOCK(vp, 0);
1972         nd->nd_repstat = vfs_busy(mp, 0);
1973         vfs_rel(mp);
1974         if (nd->nd_repstat != 0) {
1975                 vrele(vp);
1976                 free(cookies, M_TEMP);
1977                 free(rbuf, M_TEMP);
1978                 if (nd->nd_flag & ND_NFSV3)
1979                         nfsrv_postopattr(nd, getret, &at);
1980                 goto out;
1981         }
1982
1983         /*
1984          * Save this position, in case there is an error before one entry
1985          * is created.
1986          */
1987         mb0 = nd->nd_mb;
1988         bpos0 = nd->nd_bpos;
1989
1990         /*
1991          * Fill in the first part of the reply.
1992          * dirlen is the reply length in bytes and cannot exceed cnt.
1993          * (Include the two booleans at the end of the reply in dirlen now,
1994          *  so we recognize when we have exceeded cnt.)
1995          */
1996         if (nd->nd_flag & ND_NFSV3) {
1997                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1998                 nfsrv_postopattr(nd, getret, &at);
1999         } else {
2000                 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2001         }
2002         NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2003         txdr_hyper(at.na_filerev, tl);
2004
2005         /*
2006          * Save this position, in case there is an empty reply needed.
2007          */
2008         mb1 = nd->nd_mb;
2009         bpos1 = nd->nd_bpos;
2010
2011         /* Loop through the records and build reply */
2012         entrycnt = 0;
2013         while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2014                 nlen = dp->d_namlen;
2015                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2016                     nlen <= NFS_MAXNAMLEN &&
2017                     ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2018                      (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2019                       || (nlen == 1 && dp->d_name[0] != '.'))) {
2020                         /*
2021                          * Save the current position in the reply, in case
2022                          * this entry exceeds cnt.
2023                          */
2024                         mb1 = nd->nd_mb;
2025                         bpos1 = nd->nd_bpos;
2026
2027                         /*
2028                          * For readdir_and_lookup get the vnode using
2029                          * the file number.
2030                          */
2031                         nvp = NULL;
2032                         refp = NULL;
2033                         r = 0;
2034                         at_root = 0;
2035                         needs_unbusy = 0;
2036                         new_mp = mp;
2037                         mounted_on_fileno = (uint64_t)dp->d_fileno;
2038                         if ((nd->nd_flag & ND_NFSV3) ||
2039                             NFSNONZERO_ATTRBIT(&savbits)) {
2040                                 if (nd->nd_flag & ND_NFSV4)
2041                                         refp = nfsv4root_getreferral(NULL,
2042                                             vp, dp->d_fileno);
2043                                 if (refp == NULL) {
2044                                         if (usevget)
2045                                                 r = VFS_VGET(mp, dp->d_fileno,
2046                                                     LK_SHARED, &nvp);
2047                                         else
2048                                                 r = EOPNOTSUPP;
2049                                         if (r == EOPNOTSUPP) {
2050                                                 if (usevget) {
2051                                                         usevget = 0;
2052                                                         cn.cn_nameiop = LOOKUP;
2053                                                         cn.cn_lkflags =
2054                                                             LK_SHARED |
2055                                                             LK_RETRY;
2056                                                         cn.cn_cred =
2057                                                             nd->nd_cred;
2058                                                         cn.cn_thread = p;
2059                                                 }
2060                                                 cn.cn_nameptr = dp->d_name;
2061                                                 cn.cn_namelen = nlen;
2062                                                 cn.cn_flags = ISLASTCN |
2063                                                     NOFOLLOW | LOCKLEAF |
2064                                                     MPSAFE;
2065                                                 if (nlen == 2 &&
2066                                                     dp->d_name[0] == '.' &&
2067                                                     dp->d_name[1] == '.')
2068                                                         cn.cn_flags |=
2069                                                             ISDOTDOT;
2070                                                 if (NFSVOPLOCK(vp, LK_SHARED)
2071                                                     != 0) {
2072                                                         nd->nd_repstat = EPERM;
2073                                                         break;
2074                                                 }
2075                                                 if ((vp->v_vflag & VV_ROOT) != 0
2076                                                     && (cn.cn_flags & ISDOTDOT)
2077                                                     != 0) {
2078                                                         vref(vp);
2079                                                         nvp = vp;
2080                                                         r = 0;
2081                                                 } else {
2082                                                         r = VOP_LOOKUP(vp, &nvp,
2083                                                             &cn);
2084                                                         if (vp != nvp)
2085                                                                 NFSVOPUNLOCK(vp,
2086                                                                     0);
2087                                                 }
2088                                         }
2089
2090                                         /*
2091                                          * For NFSv4, check to see if nvp is
2092                                          * a mount point and get the mount
2093                                          * point vnode, as required.
2094                                          */
2095                                         if (r == 0 &&
2096                                             nfsrv_enable_crossmntpt != 0 &&
2097                                             (nd->nd_flag & ND_NFSV4) != 0 &&
2098                                             nvp->v_type == VDIR &&
2099                                             nvp->v_mountedhere != NULL) {
2100                                                 new_mp = nvp->v_mountedhere;
2101                                                 r = vfs_busy(new_mp, 0);
2102                                                 vput(nvp);
2103                                                 nvp = NULL;
2104                                                 if (r == 0) {
2105                                                         r = VFS_ROOT(new_mp,
2106                                                             LK_SHARED, &nvp);
2107                                                         needs_unbusy = 1;
2108                                                         if (r == 0)
2109                                                                 at_root = 1;
2110                                                 }
2111                                         }
2112                                 }
2113                                 if (!r) {
2114                                     if (refp == NULL &&
2115                                         ((nd->nd_flag & ND_NFSV3) ||
2116                                          NFSNONZERO_ATTRBIT(&attrbits))) {
2117                                         r = nfsvno_getfh(nvp, &nfh, p);
2118                                         if (!r)
2119                                             r = nfsvno_getattr(nvp, nvap,
2120                                                 nd->nd_cred, p, 1);
2121                                     }
2122                                 } else {
2123                                     nvp = NULL;
2124                                 }
2125                                 if (r) {
2126                                         if (!NFSISSET_ATTRBIT(&attrbits,
2127                                             NFSATTRBIT_RDATTRERROR)) {
2128                                                 if (nvp != NULL)
2129                                                         vput(nvp);
2130                                                 if (needs_unbusy != 0)
2131                                                         vfs_unbusy(new_mp);
2132                                                 nd->nd_repstat = r;
2133                                                 break;
2134                                         }
2135                                 }
2136                         }
2137
2138                         /*
2139                          * Build the directory record xdr
2140                          */
2141                         if (nd->nd_flag & ND_NFSV3) {
2142                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2143                                 *tl++ = newnfs_true;
2144                                 *tl++ = 0;
2145                                 *tl = txdr_unsigned(dp->d_fileno);
2146                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2147                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2148                                 *tl++ = 0;
2149                                 *tl = txdr_unsigned(*cookiep);
2150                                 nfsrv_postopattr(nd, 0, nvap);
2151                                 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2152                                 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2153                                 if (nvp != NULL)
2154                                         vput(nvp);
2155                         } else {
2156                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2157                                 *tl++ = newnfs_true;
2158                                 *tl++ = 0;
2159                                 *tl = txdr_unsigned(*cookiep);
2160                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2161                                 if (nvp != NULL) {
2162                                         supports_nfsv4acls =
2163                                             nfs_supportsnfsv4acls(nvp);
2164                                         NFSVOPUNLOCK(nvp, 0);
2165                                 } else
2166                                         supports_nfsv4acls = 0;
2167                                 if (refp != NULL) {
2168                                         dirlen += nfsrv_putreferralattr(nd,
2169                                             &savbits, refp, 0,
2170                                             &nd->nd_repstat);
2171                                         if (nd->nd_repstat) {
2172                                                 if (nvp != NULL)
2173                                                         vrele(nvp);
2174                                                 if (needs_unbusy != 0)
2175                                                         vfs_unbusy(new_mp);
2176                                                 break;
2177                                         }
2178                                 } else if (r) {
2179                                         dirlen += nfsvno_fillattr(nd, new_mp,
2180                                             nvp, nvap, &nfh, r, &rderrbits,
2181                                             nd->nd_cred, p, isdgram, 0,
2182                                             supports_nfsv4acls, at_root,
2183                                             mounted_on_fileno);
2184                                 } else {
2185                                         dirlen += nfsvno_fillattr(nd, new_mp,
2186                                             nvp, nvap, &nfh, r, &attrbits,
2187                                             nd->nd_cred, p, isdgram, 0,
2188                                             supports_nfsv4acls, at_root,
2189                                             mounted_on_fileno);
2190                                 }
2191                                 if (nvp != NULL)
2192                                         vrele(nvp);
2193                                 dirlen += (3 * NFSX_UNSIGNED);
2194                         }
2195                         if (needs_unbusy != 0)
2196                                 vfs_unbusy(new_mp);
2197                         if (dirlen <= cnt)
2198                                 entrycnt++;
2199                 }
2200                 cpos += dp->d_reclen;
2201                 dp = (struct dirent *)cpos;
2202                 cookiep++;
2203                 ncookies--;
2204         }
2205         vrele(vp);
2206         vfs_unbusy(mp);
2207
2208         /*
2209          * If dirlen > cnt, we must strip off the last entry. If that
2210          * results in an empty reply, report NFSERR_TOOSMALL.
2211          */
2212         if (dirlen > cnt || nd->nd_repstat) {
2213                 if (!nd->nd_repstat && entrycnt == 0)
2214                         nd->nd_repstat = NFSERR_TOOSMALL;
2215                 if (nd->nd_repstat)
2216                         newnfs_trimtrailing(nd, mb0, bpos0);
2217                 else
2218                         newnfs_trimtrailing(nd, mb1, bpos1);
2219                 eofflag = 0;
2220         } else if (cpos < cend)
2221                 eofflag = 0;
2222         if (!nd->nd_repstat) {
2223                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2224                 *tl++ = newnfs_false;
2225                 if (eofflag)
2226                         *tl = newnfs_true;
2227                 else
2228                         *tl = newnfs_false;
2229         }
2230         FREE((caddr_t)cookies, M_TEMP);
2231         FREE((caddr_t)rbuf, M_TEMP);
2232
2233 out:
2234         NFSEXITCODE2(0, nd);
2235         return (0);
2236 nfsmout:
2237         vput(vp);
2238         NFSEXITCODE2(error, nd);
2239         return (error);
2240 }
2241
2242 /*
2243  * Get the settable attributes out of the mbuf list.
2244  * (Return 0 or EBADRPC)
2245  */
2246 int
2247 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2248     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2249 {
2250         u_int32_t *tl;
2251         struct nfsv2_sattr *sp;
2252         struct timeval curtime;
2253         int error = 0, toclient = 0;
2254
2255         switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2256         case ND_NFSV2:
2257                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2258                 /*
2259                  * Some old clients didn't fill in the high order 16bits.
2260                  * --> check the low order 2 bytes for 0xffff
2261                  */
2262                 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2263                         nvap->na_mode = nfstov_mode(sp->sa_mode);
2264                 if (sp->sa_uid != newnfs_xdrneg1)
2265                         nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2266                 if (sp->sa_gid != newnfs_xdrneg1)
2267                         nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2268                 if (sp->sa_size != newnfs_xdrneg1)
2269                         nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2270                 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2271 #ifdef notyet
2272                         fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2273 #else
2274                         nvap->na_atime.tv_sec =
2275                                 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2276                         nvap->na_atime.tv_nsec = 0;
2277 #endif
2278                 }
2279                 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2280                         fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2281                 break;
2282         case ND_NFSV3:
2283                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2284                 if (*tl == newnfs_true) {
2285                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2286                         nvap->na_mode = nfstov_mode(*tl);
2287                 }
2288                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2289                 if (*tl == newnfs_true) {
2290                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2291                         nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2292                 }
2293                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2294                 if (*tl == newnfs_true) {
2295                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2296                         nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2297                 }
2298                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2299                 if (*tl == newnfs_true) {
2300                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2301                         nvap->na_size = fxdr_hyper(tl);
2302                 }
2303                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2304                 switch (fxdr_unsigned(int, *tl)) {
2305                 case NFSV3SATTRTIME_TOCLIENT:
2306                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2307                         fxdr_nfsv3time(tl, &nvap->na_atime);
2308                         toclient = 1;
2309                         break;
2310                 case NFSV3SATTRTIME_TOSERVER:
2311                         NFSGETTIME(&curtime);
2312                         nvap->na_atime.tv_sec = curtime.tv_sec;
2313                         nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2314                         nvap->na_vaflags |= VA_UTIMES_NULL;
2315                         break;
2316                 };
2317                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2318                 switch (fxdr_unsigned(int, *tl)) {
2319                 case NFSV3SATTRTIME_TOCLIENT:
2320                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2321                         fxdr_nfsv3time(tl, &nvap->na_mtime);
2322                         nvap->na_vaflags &= ~VA_UTIMES_NULL;
2323                         break;
2324                 case NFSV3SATTRTIME_TOSERVER:
2325                         NFSGETTIME(&curtime);
2326                         nvap->na_mtime.tv_sec = curtime.tv_sec;
2327                         nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2328                         if (!toclient)
2329                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2330                         break;
2331                 };
2332                 break;
2333         case ND_NFSV4:
2334                 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2335         };
2336 nfsmout:
2337         NFSEXITCODE2(error, nd);
2338         return (error);
2339 }
2340
2341 /*
2342  * Handle the setable attributes for V4.
2343  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2344  */
2345 int
2346 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2347     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2348 {
2349         u_int32_t *tl;
2350         int attrsum = 0;
2351         int i, j;
2352         int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2353         int toclient = 0;
2354         u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2355         uid_t uid;
2356         gid_t gid;
2357         struct timeval curtime;
2358
2359         error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2360         if (error)
2361                 goto nfsmout;
2362         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2363         attrsize = fxdr_unsigned(int, *tl);
2364
2365         /*
2366          * Loop around getting the setable attributes. If an unsupported
2367          * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2368          */
2369         if (retnotsup) {
2370                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2371                 bitpos = NFSATTRBIT_MAX;
2372         } else {
2373                 bitpos = 0;
2374         }
2375         for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2376             if (attrsum > attrsize) {
2377                 error = NFSERR_BADXDR;
2378                 goto nfsmout;
2379             }
2380             if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2381                 switch (bitpos) {
2382                 case NFSATTRBIT_SIZE:
2383                         NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2384                         nvap->na_size = fxdr_hyper(tl);
2385                         attrsum += NFSX_HYPER;
2386                         break;
2387                 case NFSATTRBIT_ACL:
2388                         error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2389                             p);
2390                         if (error)
2391                                 goto nfsmout;
2392                         if (aceerr && !nd->nd_repstat)
2393                                 nd->nd_repstat = aceerr;
2394                         attrsum += aclsize;
2395                         break;
2396                 case NFSATTRBIT_ARCHIVE:
2397                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2398                         if (!nd->nd_repstat)
2399                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2400                         attrsum += NFSX_UNSIGNED;
2401                         break;
2402                 case NFSATTRBIT_HIDDEN:
2403                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2404                         if (!nd->nd_repstat)
2405                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2406                         attrsum += NFSX_UNSIGNED;
2407                         break;
2408                 case NFSATTRBIT_MIMETYPE:
2409                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2410                         i = fxdr_unsigned(int, *tl);
2411                         error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2412                         if (error)
2413                                 goto nfsmout;
2414                         if (!nd->nd_repstat)
2415                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2416                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2417                         break;
2418                 case NFSATTRBIT_MODE:
2419                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2420                         nvap->na_mode = nfstov_mode(*tl);
2421                         attrsum += NFSX_UNSIGNED;
2422                         break;
2423                 case NFSATTRBIT_OWNER:
2424                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2425                         j = fxdr_unsigned(int, *tl);
2426                         if (j < 0) {
2427                                 error = NFSERR_BADXDR;
2428                                 goto nfsmout;
2429                         }
2430                         if (j > NFSV4_SMALLSTR)
2431                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2432                         else
2433                                 cp = namestr;
2434                         error = nfsrv_mtostr(nd, cp, j);
2435                         if (error) {
2436                                 if (j > NFSV4_SMALLSTR)
2437                                         free(cp, M_NFSSTRING);
2438                                 goto nfsmout;
2439                         }
2440                         if (!nd->nd_repstat) {
2441                                 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2442                                     p);
2443                                 if (!nd->nd_repstat)
2444                                         nvap->na_uid = uid;
2445                         }
2446                         if (j > NFSV4_SMALLSTR)
2447                                 free(cp, M_NFSSTRING);
2448                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2449                         break;
2450                 case NFSATTRBIT_OWNERGROUP:
2451                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2452                         j = fxdr_unsigned(int, *tl);
2453                         if (j < 0) {
2454                                 error = NFSERR_BADXDR;
2455                                 goto nfsmout;
2456                         }
2457                         if (j > NFSV4_SMALLSTR)
2458                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2459                         else
2460                                 cp = namestr;
2461                         error = nfsrv_mtostr(nd, cp, j);
2462                         if (error) {
2463                                 if (j > NFSV4_SMALLSTR)
2464                                         free(cp, M_NFSSTRING);
2465                                 goto nfsmout;
2466                         }
2467                         if (!nd->nd_repstat) {
2468                                 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2469                                     p);
2470                                 if (!nd->nd_repstat)
2471                                         nvap->na_gid = gid;
2472                         }
2473                         if (j > NFSV4_SMALLSTR)
2474                                 free(cp, M_NFSSTRING);
2475                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2476                         break;
2477                 case NFSATTRBIT_SYSTEM:
2478                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2479                         if (!nd->nd_repstat)
2480                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2481                         attrsum += NFSX_UNSIGNED;
2482                         break;
2483                 case NFSATTRBIT_TIMEACCESSSET:
2484                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2485                         attrsum += NFSX_UNSIGNED;
2486                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2487                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2488                             fxdr_nfsv4time(tl, &nvap->na_atime);
2489                             toclient = 1;
2490                             attrsum += NFSX_V4TIME;
2491                         } else {
2492                             NFSGETTIME(&curtime);
2493                             nvap->na_atime.tv_sec = curtime.tv_sec;
2494                             nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2495                             nvap->na_vaflags |= VA_UTIMES_NULL;
2496                         }
2497                         break;
2498                 case NFSATTRBIT_TIMEBACKUP:
2499                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2500                         if (!nd->nd_repstat)
2501                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2502                         attrsum += NFSX_V4TIME;
2503                         break;
2504                 case NFSATTRBIT_TIMECREATE:
2505                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2506                         if (!nd->nd_repstat)
2507                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2508                         attrsum += NFSX_V4TIME;
2509                         break;
2510                 case NFSATTRBIT_TIMEMODIFYSET:
2511                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2512                         attrsum += NFSX_UNSIGNED;
2513                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2514                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2515                             fxdr_nfsv4time(tl, &nvap->na_mtime);
2516                             nvap->na_vaflags &= ~VA_UTIMES_NULL;
2517                             attrsum += NFSX_V4TIME;
2518                         } else {
2519                             NFSGETTIME(&curtime);
2520                             nvap->na_mtime.tv_sec = curtime.tv_sec;
2521                             nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2522                             if (!toclient)
2523                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2524                         }
2525                         break;
2526                 default:
2527                         nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2528                         /*
2529                          * set bitpos so we drop out of the loop.
2530                          */
2531                         bitpos = NFSATTRBIT_MAX;
2532                         break;
2533                 };
2534         }
2535
2536         /*
2537          * some clients pad the attrlist, so we need to skip over the
2538          * padding.
2539          */
2540         if (attrsum > attrsize) {
2541                 error = NFSERR_BADXDR;
2542         } else {
2543                 attrsize = NFSM_RNDUP(attrsize);
2544                 if (attrsum < attrsize)
2545                         error = nfsm_advance(nd, attrsize - attrsum, -1);
2546         }
2547 nfsmout:
2548         NFSEXITCODE2(error, nd);
2549         return (error);
2550 }
2551
2552 /*
2553  * Check/setup export credentials.
2554  */
2555 int
2556 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2557     struct ucred *credanon)
2558 {
2559         int error = 0;
2560
2561         /*
2562          * Check/setup credentials.
2563          */
2564         if (nd->nd_flag & ND_GSS)
2565                 exp->nes_exflag &= ~MNT_EXPORTANON;
2566
2567         /*
2568          * Check to see if the operation is allowed for this security flavor.
2569          * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2570          * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2571          * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2572          */
2573         if (nfsvno_testexp(nd, exp) &&
2574             nd->nd_procnum != NFSV4OP_SECINFO &&
2575             nd->nd_procnum != NFSPROC_FSINFO) {
2576                 if (nd->nd_flag & ND_NFSV4)
2577                         error = NFSERR_WRONGSEC;
2578                 else
2579                         error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2580                 goto out;
2581         }
2582
2583         /*
2584          * Check to see if the file system is exported V4 only.
2585          */
2586         if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2587                 error = NFSERR_PROGNOTV4;
2588                 goto out;
2589         }
2590
2591         /*
2592          * Now, map the user credentials.
2593          * (Note that ND_AUTHNONE will only be set for an NFSv3
2594          *  Fsinfo RPC. If set for anything else, this code might need
2595          *  to change.)
2596          */
2597         if (NFSVNO_EXPORTED(exp) &&
2598             ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2599              NFSVNO_EXPORTANON(exp) ||
2600              (nd->nd_flag & ND_AUTHNONE))) {
2601                 nd->nd_cred->cr_uid = credanon->cr_uid;
2602                 nd->nd_cred->cr_gid = credanon->cr_gid;
2603                 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2604                     credanon->cr_groups);
2605         }
2606
2607 out:
2608         NFSEXITCODE2(error, nd);
2609         return (error);
2610 }
2611
2612 /*
2613  * Check exports.
2614  */
2615 int
2616 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2617     struct ucred **credp)
2618 {
2619         int i, error, *secflavors;
2620
2621         error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2622             &exp->nes_numsecflavor, &secflavors);
2623         if (error) {
2624                 if (nfs_rootfhset) {
2625                         exp->nes_exflag = 0;
2626                         exp->nes_numsecflavor = 0;
2627                         error = 0;
2628                 }
2629         } else {
2630                 /* Copy the security flavors. */
2631                 for (i = 0; i < exp->nes_numsecflavor; i++)
2632                         exp->nes_secflavors[i] = secflavors[i];
2633         }
2634         NFSEXITCODE(error);
2635         return (error);
2636 }
2637
2638 /*
2639  * Get a vnode for a file handle and export stuff.
2640  */
2641 int
2642 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2643     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2644     struct ucred **credp)
2645 {
2646         int i, error, *secflavors;
2647
2648         *credp = NULL;
2649         exp->nes_numsecflavor = 0;
2650         if (VFS_NEEDSGIANT(mp))
2651                 error = ESTALE;
2652         else
2653                 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2654         if (error != 0)
2655                 /* Make sure the server replies ESTALE to the client. */
2656                 error = ESTALE;
2657         if (nam && !error) {
2658                 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2659                     &exp->nes_numsecflavor, &secflavors);
2660                 if (error) {
2661                         if (nfs_rootfhset) {
2662                                 exp->nes_exflag = 0;
2663                                 exp->nes_numsecflavor = 0;
2664                                 error = 0;
2665                         } else {
2666                                 vput(*vpp);
2667                         }
2668                 } else {
2669                         /* Copy the security flavors. */
2670                         for (i = 0; i < exp->nes_numsecflavor; i++)
2671                                 exp->nes_secflavors[i] = secflavors[i];
2672                 }
2673         }
2674         NFSEXITCODE(error);
2675         return (error);
2676 }
2677
2678 /*
2679  * nfsd_fhtovp() - convert a fh to a vnode ptr
2680  *      - look up fsid in mount list (if not found ret error)
2681  *      - get vp and export rights by calling nfsvno_fhtovp()
2682  *      - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2683  *        for AUTH_SYS
2684  *      - if mpp != NULL, return the mount point so that it can
2685  *        be used for vn_finished_write() by the caller
2686  */
2687 void
2688 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2689     struct vnode **vpp, struct nfsexstuff *exp,
2690     struct mount **mpp, int startwrite, struct thread *p)
2691 {
2692         struct mount *mp;
2693         struct ucred *credanon;
2694         fhandle_t *fhp;
2695
2696         fhp = (fhandle_t *)nfp->nfsrvfh_data;
2697         /*
2698          * Check for the special case of the nfsv4root_fh.
2699          */
2700         mp = vfs_busyfs(&fhp->fh_fsid);
2701         if (mpp != NULL)
2702                 *mpp = mp;
2703         if (mp == NULL) {
2704                 *vpp = NULL;
2705                 nd->nd_repstat = ESTALE;
2706                 goto out;
2707         }
2708
2709         if (startwrite)
2710                 vn_start_write(NULL, mpp, V_WAIT);
2711
2712         nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2713             &credanon);
2714         vfs_unbusy(mp);
2715
2716         /*
2717          * For NFSv4 without a pseudo root fs, unexported file handles
2718          * can be returned, so that Lookup works everywhere.
2719          */
2720         if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2721             !(nd->nd_flag & ND_NFSV4)) {
2722                 vput(*vpp);
2723                 nd->nd_repstat = EACCES;
2724         }
2725
2726         /*
2727          * Personally, I've never seen any point in requiring a
2728          * reserved port#, since only in the rare case where the
2729          * clients are all boxes with secure system priviledges,
2730          * does it provide any enhanced security, but... some people
2731          * believe it to be useful and keep putting this code back in.
2732          * (There is also some "security checker" out there that
2733          *  complains if the nfs server doesn't enforce this.)
2734          * However, note the following:
2735          * RFC3530 (NFSv4) specifies that a reserved port# not be
2736          *      required.
2737          * RFC2623 recommends that, if a reserved port# is checked for,
2738          *      that there be a way to turn that off--> ifdef'd.
2739          */
2740 #ifdef NFS_REQRSVPORT
2741         if (!nd->nd_repstat) {
2742                 struct sockaddr_in *saddr;
2743                 struct sockaddr_in6 *saddr6;
2744
2745                 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2746                 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2747                 if (!(nd->nd_flag & ND_NFSV4) &&
2748                     ((saddr->sin_family == AF_INET &&
2749                       ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2750                      (saddr6->sin6_family == AF_INET6 &&
2751                       ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2752                         vput(*vpp);
2753                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2754                 }
2755         }
2756 #endif  /* NFS_REQRSVPORT */
2757
2758         /*
2759          * Check/setup credentials.
2760          */
2761         if (!nd->nd_repstat) {
2762                 nd->nd_saveduid = nd->nd_cred->cr_uid;
2763                 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2764                 if (nd->nd_repstat)
2765                         vput(*vpp);
2766         }
2767         if (credanon != NULL)
2768                 crfree(credanon);
2769         if (nd->nd_repstat) {
2770                 if (startwrite)
2771                         vn_finished_write(mp);
2772                 *vpp = NULL;
2773                 if (mpp != NULL)
2774                         *mpp = NULL;
2775         }
2776
2777 out:
2778         NFSEXITCODE2(0, nd);
2779 }
2780
2781 /*
2782  * glue for fp.
2783  */
2784 int
2785 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2786 {
2787         struct filedesc *fdp;
2788         struct file *fp;
2789         int error = 0;
2790
2791         fdp = p->td_proc->p_fd;
2792         if (fd >= fdp->fd_nfiles ||
2793             (fp = fdp->fd_ofiles[fd]) == NULL) {
2794                 error = EBADF;
2795                 goto out;
2796         }
2797         *fpp = fp;
2798
2799 out:
2800         NFSEXITCODE(error);
2801         return (error);
2802 }
2803
2804 /*
2805  * Called from nfssvc() to update the exports list. Just call
2806  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2807  * in the mount list.
2808  */
2809 int
2810 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2811 {
2812         struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2813         int error = 0;
2814         struct nameidata nd;
2815         fhandle_t fh;
2816
2817         error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2818         if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2819                 nfs_rootfhset = 0;
2820         else if (error == 0) {
2821                 if (nfsexargp->fspec == NULL) {
2822                         error = EPERM;
2823                         goto out;
2824                 }
2825                 /*
2826                  * If fspec != NULL, this is the v4root path.
2827                  */
2828                 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2829                     nfsexargp->fspec, p);
2830                 if ((error = namei(&nd)) != 0)
2831                         goto out;
2832                 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2833                 vrele(nd.ni_vp);
2834                 if (!error) {
2835                         nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2836                         NFSBCOPY((caddr_t)&fh,
2837                             nfs_rootfh.nfsrvfh_data,
2838                             sizeof (fhandle_t));
2839                         nfs_rootfhset = 1;
2840                 }
2841         }
2842
2843 out:
2844         NFSEXITCODE(error);
2845         return (error);
2846 }
2847
2848 /*
2849  * Get the tcp socket sequence numbers we need.
2850  * (Maybe this should be moved to the tcp sources?)
2851  */
2852 int
2853 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2854 {
2855         struct inpcb *inp;
2856         struct tcpcb *tp;
2857         int error = 0;
2858
2859         inp = sotoinpcb(so);
2860         KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2861         INP_RLOCK(inp);
2862         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2863                 INP_RUNLOCK(inp);
2864                 error = EPIPE;
2865                 goto out;
2866         }
2867         tp = intotcpcb(inp);
2868         if (tp->t_state != TCPS_ESTABLISHED) {
2869                 INP_RUNLOCK(inp);
2870                 error = EPIPE;
2871                 goto out;
2872         }
2873         *maxp = tp->snd_max;
2874         *unap = tp->snd_una;
2875         INP_RUNLOCK(inp);
2876
2877 out:
2878         NFSEXITCODE(error);
2879         return (error);
2880 }
2881
2882 /*
2883  * This function needs to test to see if the system is near its limit
2884  * for memory allocation via malloc() or mget() and return True iff
2885  * either of these resources are near their limit.
2886  * XXX (For now, this is just a stub.)
2887  */
2888 int nfsrv_testmalloclimit = 0;
2889 int
2890 nfsrv_mallocmget_limit(void)
2891 {
2892         static int printmesg = 0;
2893         static int testval = 1;
2894
2895         if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2896                 if ((printmesg++ % 100) == 0)
2897                         printf("nfsd: malloc/mget near limit\n");
2898                 return (1);
2899         }
2900         return (0);
2901 }
2902
2903 /*
2904  * BSD specific initialization of a mount point.
2905  */
2906 void
2907 nfsd_mntinit(void)
2908 {
2909         static int inited = 0;
2910
2911         if (inited)
2912                 return;
2913         inited = 1;
2914         nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2915         TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2916         TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2917         nfsv4root_mnt.mnt_export = NULL;
2918         TAILQ_INIT(&nfsv4root_opt);
2919         TAILQ_INIT(&nfsv4root_newopt);
2920         nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2921         nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2922         nfsv4root_mnt.mnt_nvnodelistsize = 0;
2923         nfsv4root_mnt.mnt_activevnodelistsize = 0;
2924 }
2925
2926 /*
2927  * Get a vnode for a file handle, without checking exports, etc.
2928  */
2929 struct vnode *
2930 nfsvno_getvp(fhandle_t *fhp)
2931 {
2932         struct mount *mp;
2933         struct vnode *vp;
2934         int error;
2935
2936         mp = vfs_busyfs(&fhp->fh_fsid);
2937         if (mp == NULL)
2938                 return (NULL);
2939         error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2940         vfs_unbusy(mp);
2941         if (error)
2942                 return (NULL);
2943         return (vp);
2944 }
2945
2946 /*
2947  * Do a local VOP_ADVLOCK().
2948  */
2949 int
2950 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2951     u_int64_t end, struct thread *td)
2952 {
2953         int error = 0;
2954         struct flock fl;
2955         u_int64_t tlen;
2956
2957         if (nfsrv_dolocallocks == 0)
2958                 goto out;
2959
2960         /* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2961         if ((vp->v_iflag & VI_DOOMED) != 0) {
2962                 error = EPERM;
2963                 goto out;
2964         }
2965
2966         fl.l_whence = SEEK_SET;
2967         fl.l_type = ftype;
2968         fl.l_start = (off_t)first;
2969         if (end == NFS64BITSSET) {
2970                 fl.l_len = 0;
2971         } else {
2972                 tlen = end - first;
2973                 fl.l_len = (off_t)tlen;
2974         }
2975         /*
2976          * For FreeBSD8, the l_pid and l_sysid must be set to the same
2977          * values for all calls, so that all locks will be held by the
2978          * nfsd server. (The nfsd server handles conflicts between the
2979          * various clients.)
2980          * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2981          * bytes, so it can't be put in l_sysid.
2982          */
2983         if (nfsv4_sysid == 0)
2984                 nfsv4_sysid = nlm_acquire_next_sysid();
2985         fl.l_pid = (pid_t)0;
2986         fl.l_sysid = (int)nfsv4_sysid;
2987
2988         NFSVOPUNLOCK(vp, 0);
2989         if (ftype == F_UNLCK)
2990                 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2991                     (F_POSIX | F_REMOTE));
2992         else
2993                 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2994                     (F_POSIX | F_REMOTE));
2995         NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2996
2997 out:
2998         NFSEXITCODE(error);
2999         return (error);
3000 }
3001
3002 /*
3003  * Check the nfsv4 root exports.
3004  */
3005 int
3006 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3007 {
3008         struct ucred *credanon;
3009         int exflags, error = 0, numsecflavor, *secflavors, i;
3010
3011         error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3012             &credanon, &numsecflavor, &secflavors);
3013         if (error) {
3014                 error = NFSERR_PROGUNAVAIL;
3015                 goto out;
3016         }
3017         if (credanon != NULL)
3018                 crfree(credanon);
3019         for (i = 0; i < numsecflavor; i++) {
3020                 if (secflavors[i] == AUTH_SYS)
3021                         nd->nd_flag |= ND_EXAUTHSYS;
3022                 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3023                         nd->nd_flag |= ND_EXGSS;
3024                 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3025                         nd->nd_flag |= ND_EXGSSINTEGRITY;
3026                 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3027                         nd->nd_flag |= ND_EXGSSPRIVACY;
3028         }
3029
3030 out:
3031         NFSEXITCODE(error);
3032         return (error);
3033 }
3034
3035 /*
3036  * Nfs server psuedo system call for the nfsd's
3037  */
3038 /*
3039  * MPSAFE
3040  */
3041 static int
3042 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3043 {
3044         struct file *fp;
3045         struct nfsd_addsock_args sockarg;
3046         struct nfsd_nfsd_args nfsdarg;
3047         int error;
3048
3049         if (uap->flag & NFSSVC_NFSDADDSOCK) {
3050                 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3051                 if (error)
3052                         goto out;
3053                 /*
3054                  * Since we don't know what rights might be required,
3055                  * pretend that we need them all. It is better to be too
3056                  * careful than too reckless.
3057                  */
3058                 if ((error = fget(td, sockarg.sock, CAP_SOCK_ALL, &fp)) != 0)
3059                         goto out;
3060                 if (fp->f_type != DTYPE_SOCKET) {
3061                         fdrop(fp, td);
3062                         error = EPERM;
3063                         goto out;
3064                 }
3065                 error = nfsrvd_addsock(fp);
3066                 fdrop(fp, td);
3067         } else if (uap->flag & NFSSVC_NFSDNFSD) {
3068                 if (uap->argp == NULL) {
3069                         error = EINVAL;
3070                         goto out;
3071                 }
3072                 error = copyin(uap->argp, (caddr_t)&nfsdarg,
3073                     sizeof (nfsdarg));
3074                 if (error)
3075                         goto out;
3076                 error = nfsrvd_nfsd(td, &nfsdarg);
3077         } else {
3078                 error = nfssvc_srvcall(td, uap, td->td_ucred);
3079         }
3080
3081 out:
3082         NFSEXITCODE(error);
3083         return (error);
3084 }
3085
3086 static int
3087 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3088 {
3089         struct nfsex_args export;
3090         struct file *fp = NULL;
3091         int stablefd, len;
3092         struct nfsd_clid adminrevoke;
3093         struct nfsd_dumplist dumplist;
3094         struct nfsd_dumpclients *dumpclients;
3095         struct nfsd_dumplocklist dumplocklist;
3096         struct nfsd_dumplocks *dumplocks;
3097         struct nameidata nd;
3098         vnode_t vp;
3099         int error = EINVAL, igotlock;
3100         struct proc *procp;
3101         static int suspend_nfsd = 0;
3102
3103         if (uap->flag & NFSSVC_PUBLICFH) {
3104                 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3105                     sizeof (fhandle_t));
3106                 error = copyin(uap->argp,
3107                     &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3108                 if (!error)
3109                         nfs_pubfhset = 1;
3110         } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3111                 error = copyin(uap->argp,(caddr_t)&export,
3112                     sizeof (struct nfsex_args));
3113                 if (!error)
3114                         error = nfsrv_v4rootexport(&export, cred, p);
3115         } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3116                 nfs_pubfhset = 0;
3117                 error = 0;
3118         } else if (uap->flag & NFSSVC_STABLERESTART) {
3119                 error = copyin(uap->argp, (caddr_t)&stablefd,
3120                     sizeof (int));
3121                 if (!error)
3122                         error = fp_getfvp(p, stablefd, &fp, &vp);
3123                 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3124                         error = EBADF;
3125                 if (!error && newnfs_numnfsd != 0)
3126                         error = EPERM;
3127                 if (!error) {
3128                         nfsrv_stablefirst.nsf_fp = fp;
3129                         nfsrv_setupstable(p);
3130                 }
3131         } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3132                 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3133                     sizeof (struct nfsd_clid));
3134                 if (!error)
3135                         error = nfsrv_adminrevoke(&adminrevoke, p);
3136         } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3137                 error = copyin(uap->argp, (caddr_t)&dumplist,
3138                     sizeof (struct nfsd_dumplist));
3139                 if (!error && (dumplist.ndl_size < 1 ||
3140                         dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3141                         error = EPERM;
3142                 if (!error) {
3143                     len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3144                     dumpclients = (struct nfsd_dumpclients *)malloc(len,
3145                         M_TEMP, M_WAITOK);
3146                     nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3147                     error = copyout(dumpclients,
3148                         CAST_USER_ADDR_T(dumplist.ndl_list), len);
3149                     free((caddr_t)dumpclients, M_TEMP);
3150                 }
3151         } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3152                 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3153                     sizeof (struct nfsd_dumplocklist));
3154                 if (!error && (dumplocklist.ndllck_size < 1 ||
3155                         dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3156                         error = EPERM;
3157                 if (!error)
3158                         error = nfsrv_lookupfilename(&nd,
3159                                 dumplocklist.ndllck_fname, p);
3160                 if (!error) {
3161                         len = sizeof (struct nfsd_dumplocks) *
3162                                 dumplocklist.ndllck_size;
3163                         dumplocks = (struct nfsd_dumplocks *)malloc(len,
3164                                 M_TEMP, M_WAITOK);
3165                         nfsrv_dumplocks(nd.ni_vp, dumplocks,
3166                             dumplocklist.ndllck_size, p);
3167                         vput(nd.ni_vp);
3168                         error = copyout(dumplocks,
3169                             CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3170                         free((caddr_t)dumplocks, M_TEMP);
3171                 }
3172         } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3173                 procp = p->td_proc;
3174                 PROC_LOCK(procp);
3175                 nfsd_master_pid = procp->p_pid;
3176                 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3177                 nfsd_master_start = procp->p_stats->p_start;
3178                 nfsd_master_proc = procp;
3179                 PROC_UNLOCK(procp);
3180         } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3181                 NFSLOCKV4ROOTMUTEX();
3182                 if (suspend_nfsd == 0) {
3183                         /* Lock out all nfsd threads */
3184                         do {
3185                                 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3186                                     NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3187                         } while (igotlock == 0 && suspend_nfsd == 0);
3188                         suspend_nfsd = 1;
3189                 }
3190                 NFSUNLOCKV4ROOTMUTEX();
3191                 error = 0;
3192         } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3193                 NFSLOCKV4ROOTMUTEX();
3194                 if (suspend_nfsd != 0) {
3195                         nfsv4_unlock(&nfsd_suspend_lock, 0);
3196                         suspend_nfsd = 0;
3197                 }
3198                 NFSUNLOCKV4ROOTMUTEX();
3199                 error = 0;
3200         }
3201
3202         NFSEXITCODE(error);
3203         return (error);
3204 }
3205
3206 /*
3207  * Check exports.
3208  * Returns 0 if ok, 1 otherwise.
3209  */
3210 int
3211 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3212 {
3213         int i;
3214
3215         /*
3216          * This seems odd, but allow the case where the security flavor
3217          * list is empty. This happens when NFSv4 is traversing non-exported
3218          * file systems. Exported file systems should always have a non-empty
3219          * security flavor list.
3220          */
3221         if (exp->nes_numsecflavor == 0)
3222                 return (0);
3223
3224         for (i = 0; i < exp->nes_numsecflavor; i++) {
3225                 /*
3226                  * The tests for privacy and integrity must be first,
3227                  * since ND_GSS is set for everything but AUTH_SYS.
3228                  */
3229                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3230                     (nd->nd_flag & ND_GSSPRIVACY))
3231                         return (0);
3232                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3233                     (nd->nd_flag & ND_GSSINTEGRITY))
3234                         return (0);
3235                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3236                     (nd->nd_flag & ND_GSS))
3237                         return (0);
3238                 if (exp->nes_secflavors[i] == AUTH_SYS &&
3239                     (nd->nd_flag & ND_GSS) == 0)
3240                         return (0);
3241         }
3242         return (1);
3243 }
3244
3245 /*
3246  * Calculate a hash value for the fid in a file handle.
3247  */
3248 uint32_t
3249 nfsrv_hashfh(fhandle_t *fhp)
3250 {
3251         uint32_t hashval;
3252
3253         hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3254         return (hashval);
3255 }
3256
3257 /*
3258  * Signal the userland master nfsd to backup the stable restart file.
3259  */
3260 void
3261 nfsrv_backupstable(void)
3262 {
3263         struct proc *procp;
3264
3265         if (nfsd_master_proc != NULL) {
3266                 procp = pfind(nfsd_master_pid);
3267                 /* Try to make sure it is the correct process. */
3268                 if (procp == nfsd_master_proc &&
3269                     procp->p_stats->p_start.tv_sec ==
3270                     nfsd_master_start.tv_sec &&
3271                     procp->p_stats->p_start.tv_usec ==
3272                     nfsd_master_start.tv_usec &&
3273                     strcmp(procp->p_comm, nfsd_master_comm) == 0)
3274                         kern_psignal(procp, SIGUSR2);
3275                 else
3276                         nfsd_master_proc = NULL;
3277
3278                 if (procp != NULL)
3279                         PROC_UNLOCK(procp);
3280         }
3281 }
3282
3283 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3284
3285 /*
3286  * Called once to initialize data structures...
3287  */
3288 static int
3289 nfsd_modevent(module_t mod, int type, void *data)
3290 {
3291         int error = 0;
3292         static int loaded = 0;
3293
3294         switch (type) {
3295         case MOD_LOAD:
3296                 if (loaded)
3297                         goto out;
3298                 newnfs_portinit();
3299                 mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3300                 mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3301                 mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3302                     MTX_DEF);
3303                 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3304                 nfsrvd_initcache();
3305                 nfsd_init();
3306                 NFSD_LOCK();
3307                 nfsrvd_init(0);
3308                 NFSD_UNLOCK();
3309                 nfsd_mntinit();
3310 #ifdef VV_DISABLEDELEG
3311                 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3312                 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3313 #endif
3314                 nfsd_call_servertimer = nfsrv_servertimer;
3315                 nfsd_call_nfsd = nfssvc_nfsd;
3316                 loaded = 1;
3317                 break;
3318
3319         case MOD_UNLOAD:
3320                 if (newnfs_numnfsd != 0) {
3321                         error = EBUSY;
3322                         break;
3323                 }
3324
3325 #ifdef VV_DISABLEDELEG
3326                 vn_deleg_ops.vndeleg_recall = NULL;
3327                 vn_deleg_ops.vndeleg_disable = NULL;
3328 #endif
3329                 nfsd_call_servertimer = NULL;
3330                 nfsd_call_nfsd = NULL;
3331
3332                 /* Clean out all NFSv4 state. */
3333                 nfsrv_throwawayallstate(curthread);
3334
3335                 /* Clean the NFS server reply cache */
3336                 nfsrvd_cleancache();
3337
3338                 /* Free up the krpc server pool. */
3339                 if (nfsrvd_pool != NULL)
3340                         svcpool_destroy(nfsrvd_pool);
3341
3342                 /* and get rid of the locks */
3343                 mtx_destroy(&nfs_cache_mutex);
3344                 mtx_destroy(&nfs_v4root_mutex);
3345                 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3346                 lockdestroy(&nfsv4root_mnt.mnt_explock);
3347                 loaded = 0;
3348                 break;
3349         default:
3350                 error = EOPNOTSUPP;
3351                 break;
3352         }
3353
3354 out:
3355         NFSEXITCODE(error);
3356         return (error);
3357 }
3358 static moduledata_t nfsd_mod = {
3359         "nfsd",
3360         nfsd_modevent,
3361         NULL,
3362 };
3363 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3364
3365 /* So that loader and kldload(2) can find us, wherever we are.. */
3366 MODULE_VERSION(nfsd, 1);
3367 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3368 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3369 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3370 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3371 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3372