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