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