sys/fs/nfsserver/nfs_nfsdport.c

   1 /*-
   2  * Copyright (c) 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Rick Macklem at The University of Guelph.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 4. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  *
  32  */
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 #include <sys/capability.h>
  38
  39 /*
  40  * Functions that perform the vfs operations required by the routines in
  41  * nfsd_serv.c. It is hoped that this change will make the server more
  42  * portable.
  43  */
  44
  45 #include <fs/nfs/nfsport.h>
  46 #include <sys/hash.h>
  47 #include <sys/sysctl.h>
  48 #include <nlm/nlm_prot.h>
  49 #include <nlm/nlm.h>
  50
  51 FEATURE(nfsd, "NFSv4 server");
  52
  53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
  54 extern int nfsrv_useacl;
  55 extern int newnfs_numnfsd;
  56 extern struct mount nfsv4root_mnt;
  57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
  58 extern void (*nfsd_call_servertimer)(void);
  59 extern SVCPOOL  *nfsrvd_pool;
  60 extern struct nfsv4lock nfsd_suspend_lock;
  61 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
  62 NFSDLOCKMUTEX;
  63 struct mtx nfs_cache_mutex;
  64 struct mtx nfs_v4root_mutex;
  65 struct nfsrvfh nfs_rootfh, nfs_pubfh;
  66 int nfs_pubfhset = 0, nfs_rootfhset = 0;
  67 struct proc *nfsd_master_proc = NULL;
  68 static pid_t nfsd_master_pid = (pid_t)-1;
  69 static char nfsd_master_comm[MAXCOMLEN + 1];
  70 static struct timeval nfsd_master_start;
  71 static uint32_t nfsv4_sysid = 0;
  72
  73 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
  74     struct ucred *);
  75
  76 int nfsrv_enable_crossmntpt = 1;
  77 static int nfs_commit_blks;
  78 static int nfs_commit_miss;
  79 extern int nfsrv_issuedelegs;
  80 extern int nfsrv_dolocallocks;
  81
  82 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
  83 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
  84     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
  85 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
  86     0, "");
  87 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
  88     0, "");
  89 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
  90     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
  91 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
  92     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
  93
  94 #define MAX_REORDERED_RPC       16
  95 #define NUM_HEURISTIC           1031
  96 #define NHUSE_INIT              64
  97 #define NHUSE_INC               16
  98 #define NHUSE_MAX               2048
  99
 100 static struct nfsheur {
 101         struct vnode *nh_vp;    /* vp to match (unreferenced pointer) */
 102         off_t nh_nextoff;       /* next offset for sequential detection */
 103         int nh_use;             /* use count for selection */
 104         int nh_seqcount;        /* heuristic */
 105 } nfsheur[NUM_HEURISTIC];
 106
 107
 108 /*
 109  * Heuristic to detect sequential operation.
 110  */
 111 static struct nfsheur *
 112 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
 113 {
 114         struct nfsheur *nh;
 115         int hi, try;
 116
 117         /* Locate best candidate. */
 118         try = 32;
 119         hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
 120         nh = &nfsheur[hi];
 121         while (try--) {
 122                 if (nfsheur[hi].nh_vp == vp) {
 123                         nh = &nfsheur[hi];
 124                         break;
 125                 }
 126                 if (nfsheur[hi].nh_use > 0)
 127                         --nfsheur[hi].nh_use;
 128                 hi = (hi + 1) % NUM_HEURISTIC;
 129                 if (nfsheur[hi].nh_use < nh->nh_use)
 130                         nh = &nfsheur[hi];
 131         }
 132
 133         /* Initialize hint if this is a new file. */
 134         if (nh->nh_vp != vp) {
 135                 nh->nh_vp = vp;
 136                 nh->nh_nextoff = uio->uio_offset;
 137                 nh->nh_use = NHUSE_INIT;
 138                 if (uio->uio_offset == 0)
 139                         nh->nh_seqcount = 4;
 140                 else
 141                         nh->nh_seqcount = 1;
 142         }
 143
 144         /* Calculate heuristic. */
 145         if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
 146             uio->uio_offset == nh->nh_nextoff) {
 147                 /* See comments in vfs_vnops.c:sequential_heuristic(). */
 148                 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
 149                 if (nh->nh_seqcount > IO_SEQMAX)
 150                         nh->nh_seqcount = IO_SEQMAX;
 151         } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
 152             imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
 153                 /* Probably a reordered RPC, leave seqcount alone. */
 154         } else if (nh->nh_seqcount > 1) {
 155                 nh->nh_seqcount /= 2;
 156         } else {
 157                 nh->nh_seqcount = 0;
 158         }
 159         nh->nh_use += NHUSE_INC;
 160         if (nh->nh_use > NHUSE_MAX)
 161                 nh->nh_use = NHUSE_MAX;
 162         return (nh);
 163 }
 164
 165 /*
 166  * Get attributes into nfsvattr structure.
 167  */
 168 int
 169 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 170     struct thread *p, int vpislocked)
 171 {
 172         int error, lockedit = 0;
 173
 174         if (vpislocked == 0) {
 175                 /*
 176                  * When vpislocked == 0, the vnode is either exclusively
 177                  * locked by this thread or not locked by this thread.
 178                  * As such, shared lock it, if not exclusively locked.
 179                  */
 180                 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
 181                         lockedit = 1;
 182                         NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
 183                 }
 184         }
 185         error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
 186         if (lockedit != 0)
 187                 NFSVOPUNLOCK(vp, 0);
 188
 189         NFSEXITCODE(error);
 190         return (error);
 191 }
 192
 193 /*
 194  * Get a file handle for a vnode.
 195  */
 196 int
 197 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
 198 {
 199         int error;
 200
 201         NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
 202         fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
 203         error = VOP_VPTOFH(vp, &fhp->fh_fid);
 204
 205         NFSEXITCODE(error);
 206         return (error);
 207 }
 208
 209 /*
 210  * Perform access checking for vnodes obtained from file handles that would
 211  * refer to files already opened by a Unix client. You cannot just use
 212  * vn_writechk() and VOP_ACCESSX() for two reasons.
 213  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
 214  *     case.
 215  * 2 - The owner is to be given access irrespective of mode bits for some
 216  *     operations, so that processes that chmod after opening a file don't
 217  *     break.
 218  */
 219 int
 220 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
 221     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
 222     u_int32_t *supportedtypep)
 223 {
 224         struct vattr vattr;
 225         int error = 0, getret = 0;
 226
 227         if (vpislocked == 0) {
 228                 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
 229                         error = EPERM;
 230                         goto out;
 231                 }
 232         }
 233         if (accmode & VWRITE) {
 234                 /* Just vn_writechk() changed to check rdonly */
 235                 /*
 236                  * Disallow write attempts on read-only file systems;
 237                  * unless the file is a socket or a block or character
 238                  * device resident on the file system.
 239                  */
 240                 if (NFSVNO_EXRDONLY(exp) ||
 241                     (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 242                         switch (vp->v_type) {
 243                         case VREG:
 244                         case VDIR:
 245                         case VLNK:
 246                                 error = EROFS;
 247                         default:
 248                                 break;
 249                         }
 250                 }
 251                 /*
 252                  * If there's shared text associated with
 253                  * the inode, try to free it up once.  If
 254                  * we fail, we can't allow writing.
 255                  */
 256                 if (VOP_IS_TEXT(vp) && error == 0)
 257                         error = ETXTBSY;
 258         }
 259         if (error != 0) {
 260                 if (vpislocked == 0)
 261                         NFSVOPUNLOCK(vp, 0);
 262                 goto out;
 263         }
 264
 265         /*
 266          * Should the override still be applied when ACLs are enabled?
 267          */
 268         error = VOP_ACCESSX(vp, accmode, cred, p);
 269         if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
 270                 /*
 271                  * Try again with VEXPLICIT_DENY, to see if the test for
 272                  * deletion is supported.
 273                  */
 274                 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
 275                 if (error == 0) {
 276                         if (vp->v_type == VDIR) {
 277                                 accmode &= ~(VDELETE | VDELETE_CHILD);
 278                                 accmode |= VWRITE;
 279                                 error = VOP_ACCESSX(vp, accmode, cred, p);
 280                         } else if (supportedtypep != NULL) {
 281                                 *supportedtypep &= ~NFSACCESS_DELETE;
 282                         }
 283                 }
 284         }
 285
 286         /*
 287          * Allow certain operations for the owner (reads and writes
 288          * on files that are already open).
 289          */
 290         if (override != NFSACCCHK_NOOVERRIDE &&
 291             (error == EPERM || error == EACCES)) {
 292                 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
 293                         error = 0;
 294                 else if (override & NFSACCCHK_ALLOWOWNER) {
 295                         getret = VOP_GETATTR(vp, &vattr, cred);
 296                         if (getret == 0 && cred->cr_uid == vattr.va_uid)
 297                                 error = 0;
 298                 }
 299         }
 300         if (vpislocked == 0)
 301                 NFSVOPUNLOCK(vp, 0);
 302
 303 out:
 304         NFSEXITCODE(error);
 305         return (error);
 306 }
 307
 308 /*
 309  * Set attribute(s) vnop.
 310  */
 311 int
 312 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 313     struct thread *p, struct nfsexstuff *exp)
 314 {
 315         int error;
 316
 317         error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
 318         NFSEXITCODE(error);
 319         return (error);
 320 }
 321
 322 /*
 323  * Set up nameidata for a lookup() call and do it.
 324  */
 325 int
 326 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
 327     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
 328     struct vnode **retdirp)
 329 {
 330         struct componentname *cnp = &ndp->ni_cnd;
 331         int i;
 332         struct iovec aiov;
 333         struct uio auio;
 334         int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
 335         int error = 0, crossmnt;
 336         char *cp;
 337
 338         *retdirp = NULL;
 339         cnp->cn_nameptr = cnp->cn_pnbuf;
 340         ndp->ni_strictrelative = 0;
 341         /*
 342          * Extract and set starting directory.
 343          */
 344         if (dp->v_type != VDIR) {
 345                 if (islocked)
 346                         vput(dp);
 347                 else
 348                         vrele(dp);
 349                 nfsvno_relpathbuf(ndp);
 350                 error = ENOTDIR;
 351                 goto out1;
 352         }
 353         if (islocked)
 354                 NFSVOPUNLOCK(dp, 0);
 355         VREF(dp);
 356         *retdirp = dp;
 357         if (NFSVNO_EXRDONLY(exp))
 358                 cnp->cn_flags |= RDONLY;
 359         ndp->ni_segflg = UIO_SYSSPACE;
 360         crossmnt = 1;
 361
 362         if (nd->nd_flag & ND_PUBLOOKUP) {
 363                 ndp->ni_loopcnt = 0;
 364                 if (cnp->cn_pnbuf[0] == '/') {
 365                         vrele(dp);
 366                         /*
 367                          * Check for degenerate pathnames here, since lookup()
 368                          * panics on them.
 369                          */
 370                         for (i = 1; i < ndp->ni_pathlen; i++)
 371                                 if (cnp->cn_pnbuf[i] != '/')
 372                                         break;
 373                         if (i == ndp->ni_pathlen) {
 374                                 error = NFSERR_ACCES;
 375                                 goto out;
 376                         }
 377                         dp = rootvnode;
 378                         VREF(dp);
 379                 }
 380         } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
 381             (nd->nd_flag & ND_NFSV4) == 0) {
 382                 /*
 383                  * Only cross mount points for NFSv4 when doing a
 384                  * mount while traversing the file system above
 385                  * the mount point, unless nfsrv_enable_crossmntpt is set.
 386                  */
 387                 cnp->cn_flags |= NOCROSSMOUNT;
 388                 crossmnt = 0;
 389         }
 390
 391         /*
 392          * Initialize for scan, set ni_startdir and bump ref on dp again
 393          * becuase lookup() will dereference ni_startdir.
 394          */
 395
 396         cnp->cn_thread = p;
 397         ndp->ni_startdir = dp;
 398         ndp->ni_rootdir = rootvnode;
 399         ndp->ni_topdir = NULL;
 400
 401         if (!lockleaf)
 402                 cnp->cn_flags |= LOCKLEAF;
 403         for (;;) {
 404                 cnp->cn_nameptr = cnp->cn_pnbuf;
 405                 /*
 406                  * Call lookup() to do the real work.  If an error occurs,
 407                  * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
 408                  * we do not have to dereference anything before returning.
 409                  * In either case ni_startdir will be dereferenced and NULLed
 410                  * out.
 411                  */
 412                 error = lookup(ndp);
 413                 if (error)
 414                         break;
 415
 416                 /*
 417                  * Check for encountering a symbolic link.  Trivial
 418                  * termination occurs if no symlink encountered.
 419                  */
 420                 if ((cnp->cn_flags & ISSYMLINK) == 0) {
 421                         if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
 422                                 nfsvno_relpathbuf(ndp);
 423                         if (ndp->ni_vp && !lockleaf)
 424                                 NFSVOPUNLOCK(ndp->ni_vp, 0);
 425                         break;
 426                 }
 427
 428                 /*
 429                  * Validate symlink
 430                  */
 431                 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
 432                         NFSVOPUNLOCK(ndp->ni_dvp, 0);
 433                 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
 434                         error = EINVAL;
 435                         goto badlink2;
 436                 }
 437
 438                 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
 439                         error = ELOOP;
 440                         goto badlink2;
 441                 }
 442                 if (ndp->ni_pathlen > 1)
 443                         cp = uma_zalloc(namei_zone, M_WAITOK);
 444                 else
 445                         cp = cnp->cn_pnbuf;
 446                 aiov.iov_base = cp;
 447                 aiov.iov_len = MAXPATHLEN;
 448                 auio.uio_iov = &aiov;
 449                 auio.uio_iovcnt = 1;
 450                 auio.uio_offset = 0;
 451                 auio.uio_rw = UIO_READ;
 452                 auio.uio_segflg = UIO_SYSSPACE;
 453                 auio.uio_td = NULL;
 454                 auio.uio_resid = MAXPATHLEN;
 455                 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
 456                 if (error) {
 457                 badlink1:
 458                         if (ndp->ni_pathlen > 1)
 459                                 uma_zfree(namei_zone, cp);
 460                 badlink2:
 461                         vrele(ndp->ni_dvp);
 462                         vput(ndp->ni_vp);
 463                         break;
 464                 }
 465                 linklen = MAXPATHLEN - auio.uio_resid;
 466                 if (linklen == 0) {
 467                         error = ENOENT;
 468                         goto badlink1;
 469                 }
 470                 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
 471                         error = ENAMETOOLONG;
 472                         goto badlink1;
 473                 }
 474
 475                 /*
 476                  * Adjust or replace path
 477                  */
 478                 if (ndp->ni_pathlen > 1) {
 479                         NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
 480                         uma_zfree(namei_zone, cnp->cn_pnbuf);
 481                         cnp->cn_pnbuf = cp;
 482                 } else
 483                         cnp->cn_pnbuf[linklen] = '\0';
 484                 ndp->ni_pathlen += linklen;
 485
 486                 /*
 487                  * Cleanup refs for next loop and check if root directory
 488                  * should replace current directory.  Normally ni_dvp
 489                  * becomes the new base directory and is cleaned up when
 490                  * we loop.  Explicitly null pointers after invalidation
 491                  * to clarify operation.
 492                  */
 493                 vput(ndp->ni_vp);
 494                 ndp->ni_vp = NULL;
 495
 496                 if (cnp->cn_pnbuf[0] == '/') {
 497                         vrele(ndp->ni_dvp);
 498                         ndp->ni_dvp = ndp->ni_rootdir;
 499                         VREF(ndp->ni_dvp);
 500                 }
 501                 ndp->ni_startdir = ndp->ni_dvp;
 502                 ndp->ni_dvp = NULL;
 503         }
 504         if (!lockleaf)
 505                 cnp->cn_flags &= ~LOCKLEAF;
 506
 507 out:
 508         if (error) {
 509                 uma_zfree(namei_zone, cnp->cn_pnbuf);
 510                 ndp->ni_vp = NULL;
 511                 ndp->ni_dvp = NULL;
 512                 ndp->ni_startdir = NULL;
 513                 cnp->cn_flags &= ~HASBUF;
 514         } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
 515                 ndp->ni_dvp = NULL;
 516         }
 517
 518 out1:
 519         NFSEXITCODE2(error, nd);
 520         return (error);
 521 }
 522
 523 /*
 524  * Set up a pathname buffer and return a pointer to it and, optionally
 525  * set a hash pointer.
 526  */
 527 void
 528 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
 529 {
 530         struct componentname *cnp = &ndp->ni_cnd;
 531
 532         cnp->cn_flags |= (NOMACCHECK | HASBUF);
 533         cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
 534         if (hashpp != NULL)
 535                 *hashpp = NULL;
 536         *bufpp = cnp->cn_pnbuf;
 537 }
 538
 539 /*
 540  * Release the above path buffer, if not released by nfsvno_namei().
 541  */
 542 void
 543 nfsvno_relpathbuf(struct nameidata *ndp)
 544 {
 545
 546         if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
 547                 panic("nfsrelpath");
 548         uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
 549         ndp->ni_cnd.cn_flags &= ~HASBUF;
 550 }
 551
 552 /*
 553  * Readlink vnode op into an mbuf list.
 554  */
 555 int
 556 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
 557     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
 558 {
 559         struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
 560         struct iovec *ivp = iv;
 561         struct uio io, *uiop = &io;
 562         struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
 563         int i, len, tlen, error = 0;
 564
 565         len = 0;
 566         i = 0;
 567         while (len < NFS_MAXPATHLEN) {
 568                 NFSMGET(mp);
 569                 MCLGET(mp, M_WAIT);
 570                 mp->m_len = NFSMSIZ(mp);
 571                 if (len == 0) {
 572                         mp3 = mp2 = mp;
 573                 } else {
 574                         mp2->m_next = mp;
 575                         mp2 = mp;
 576                 }
 577                 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
 578                         mp->m_len = NFS_MAXPATHLEN - len;
 579                         len = NFS_MAXPATHLEN;
 580                 } else {
 581                         len += mp->m_len;
 582                 }
 583                 ivp->iov_base = mtod(mp, caddr_t);
 584                 ivp->iov_len = mp->m_len;
 585                 i++;
 586                 ivp++;
 587         }
 588         uiop->uio_iov = iv;
 589         uiop->uio_iovcnt = i;
 590         uiop->uio_offset = 0;
 591         uiop->uio_resid = len;
 592         uiop->uio_rw = UIO_READ;
 593         uiop->uio_segflg = UIO_SYSSPACE;
 594         uiop->uio_td = NULL;
 595         error = VOP_READLINK(vp, uiop, cred);
 596         if (error) {
 597                 m_freem(mp3);
 598                 *lenp = 0;
 599                 goto out;
 600         }
 601         if (uiop->uio_resid > 0) {
 602                 len -= uiop->uio_resid;
 603                 tlen = NFSM_RNDUP(len);
 604                 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
 605         }
 606         *lenp = len;
 607         *mpp = mp3;
 608         *mpendp = mp;
 609
 610 out:
 611         NFSEXITCODE(error);
 612         return (error);
 613 }
 614
 615 /*
 616  * Read vnode op call into mbuf list.
 617  */
 618 int
 619 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
 620     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
 621 {
 622         struct mbuf *m;
 623         int i;
 624         struct iovec *iv;
 625         struct iovec *iv2;
 626         int error = 0, len, left, siz, tlen, ioflag = 0;
 627         struct mbuf *m2 = NULL, *m3;
 628         struct uio io, *uiop = &io;
 629         struct nfsheur *nh;
 630
 631         len = left = NFSM_RNDUP(cnt);
 632         m3 = NULL;
 633         /*
 634          * Generate the mbuf list with the uio_iov ref. to it.
 635          */
 636         i = 0;
 637         while (left > 0) {
 638                 NFSMGET(m);
 639                 MCLGET(m, M_WAIT);
 640                 m->m_len = 0;
 641                 siz = min(M_TRAILINGSPACE(m), left);
 642                 left -= siz;
 643                 i++;
 644                 if (m3)
 645                         m2->m_next = m;
 646                 else
 647                         m3 = m;
 648                 m2 = m;
 649         }
 650         MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
 651             M_TEMP, M_WAITOK);
 652         uiop->uio_iov = iv2 = iv;
 653         m = m3;
 654         left = len;
 655         i = 0;
 656         while (left > 0) {
 657                 if (m == NULL)
 658                         panic("nfsvno_read iov");
 659                 siz = min(M_TRAILINGSPACE(m), left);
 660                 if (siz > 0) {
 661                         iv->iov_base = mtod(m, caddr_t) + m->m_len;
 662                         iv->iov_len = siz;
 663                         m->m_len += siz;
 664                         left -= siz;
 665                         iv++;
 666                         i++;
 667                 }
 668                 m = m->m_next;
 669         }
 670         uiop->uio_iovcnt = i;
 671         uiop->uio_offset = off;
 672         uiop->uio_resid = len;
 673         uiop->uio_rw = UIO_READ;
 674         uiop->uio_segflg = UIO_SYSSPACE;
 675         nh = nfsrv_sequential_heuristic(uiop, vp);
 676         ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
 677         error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
 678         FREE((caddr_t)iv2, M_TEMP);
 679         if (error) {
 680                 m_freem(m3);
 681                 *mpp = NULL;
 682                 goto out;
 683         }
 684         nh->nh_nextoff = uiop->uio_offset;
 685         tlen = len - uiop->uio_resid;
 686         cnt = cnt < tlen ? cnt : tlen;
 687         tlen = NFSM_RNDUP(cnt);
 688         if (tlen == 0) {
 689                 m_freem(m3);
 690                 m3 = NULL;
 691         } else if (len != tlen || tlen != cnt)
 692                 nfsrv_adj(m3, len - tlen, tlen - cnt);
 693         *mpp = m3;
 694         *mpendp = m2;
 695
 696 out:
 697         NFSEXITCODE(error);
 698         return (error);
 699 }
 700
 701 /*
 702  * Write vnode op from an mbuf list.
 703  */
 704 int
 705 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
 706     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
 707 {
 708         struct iovec *ivp;
 709         int i, len;
 710         struct iovec *iv;
 711         int ioflags, error;
 712         struct uio io, *uiop = &io;
 713         struct nfsheur *nh;
 714
 715         MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
 716             M_WAITOK);
 717         uiop->uio_iov = iv = ivp;
 718         uiop->uio_iovcnt = cnt;
 719         i = mtod(mp, caddr_t) + mp->m_len - cp;
 720         len = retlen;
 721         while (len > 0) {
 722                 if (mp == NULL)
 723                         panic("nfsvno_write");
 724                 if (i > 0) {
 725                         i = min(i, len);
 726                         ivp->iov_base = cp;
 727                         ivp->iov_len = i;
 728                         ivp++;
 729                         len -= i;
 730                 }
 731                 mp = mp->m_next;
 732                 if (mp) {
 733                         i = mp->m_len;
 734                         cp = mtod(mp, caddr_t);
 735                 }
 736         }
 737
 738         if (stable == NFSWRITE_UNSTABLE)
 739                 ioflags = IO_NODELOCKED;
 740         else
 741                 ioflags = (IO_SYNC | IO_NODELOCKED);
 742         uiop->uio_resid = retlen;
 743         uiop->uio_rw = UIO_WRITE;
 744         uiop->uio_segflg = UIO_SYSSPACE;
 745         NFSUIOPROC(uiop, p);
 746         uiop->uio_offset = off;
 747         nh = nfsrv_sequential_heuristic(uiop, vp);
 748         ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
 749         error = VOP_WRITE(vp, uiop, ioflags, cred);
 750         if (error == 0)
 751                 nh->nh_nextoff = uiop->uio_offset;
 752         FREE((caddr_t)iv, M_TEMP);
 753
 754         NFSEXITCODE(error);
 755         return (error);
 756 }
 757
 758 /*
 759  * Common code for creating a regular file (plus special files for V2).
 760  */
 761 int
 762 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
 763     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
 764     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
 765 {
 766         u_quad_t tempsize;
 767         int error;
 768
 769         error = nd->nd_repstat;
 770         if (!error && ndp->ni_vp == NULL) {
 771                 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
 772                         vrele(ndp->ni_startdir);
 773                         error = VOP_CREATE(ndp->ni_dvp,
 774                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
 775                         vput(ndp->ni_dvp);
 776                         nfsvno_relpathbuf(ndp);
 777                         if (!error) {
 778                                 if (*exclusive_flagp) {
 779                                         *exclusive_flagp = 0;
 780                                         NFSVNO_ATTRINIT(nvap);
 781                                         nvap->na_atime.tv_sec = cverf[0];
 782                                         nvap->na_atime.tv_nsec = cverf[1];
 783                                         error = VOP_SETATTR(ndp->ni_vp,
 784                                             &nvap->na_vattr, nd->nd_cred);
 785                                 }
 786                         }
 787                 /*
 788                  * NFS V2 Only. nfsrvd_mknod() does this for V3.
 789                  * (This implies, just get out on an error.)
 790                  */
 791                 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
 792                         nvap->na_type == VFIFO) {
 793                         if (nvap->na_type == VCHR && rdev == 0xffffffff)
 794                                 nvap->na_type = VFIFO;
 795                         if (nvap->na_type != VFIFO &&
 796                             (error = priv_check_cred(nd->nd_cred,
 797                              PRIV_VFS_MKNOD_DEV, 0))) {
 798                                 vrele(ndp->ni_startdir);
 799                                 nfsvno_relpathbuf(ndp);
 800                                 vput(ndp->ni_dvp);
 801                                 goto out;
 802                         }
 803                         nvap->na_rdev = rdev;
 804                         error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 805                             &ndp->ni_cnd, &nvap->na_vattr);
 806                         vput(ndp->ni_dvp);
 807                         nfsvno_relpathbuf(ndp);
 808                         vrele(ndp->ni_startdir);
 809                         if (error)
 810                                 goto out;
 811                 } else {
 812                         vrele(ndp->ni_startdir);
 813                         nfsvno_relpathbuf(ndp);
 814                         vput(ndp->ni_dvp);
 815                         error = ENXIO;
 816                         goto out;
 817                 }
 818                 *vpp = ndp->ni_vp;
 819         } else {
 820                 /*
 821                  * Handle cases where error is already set and/or
 822                  * the file exists.
 823                  * 1 - clean up the lookup
 824                  * 2 - iff !error and na_size set, truncate it
 825                  */
 826                 vrele(ndp->ni_startdir);
 827                 nfsvno_relpathbuf(ndp);
 828                 *vpp = ndp->ni_vp;
 829                 if (ndp->ni_dvp == *vpp)
 830                         vrele(ndp->ni_dvp);
 831                 else
 832                         vput(ndp->ni_dvp);
 833                 if (!error && nvap->na_size != VNOVAL) {
 834                         error = nfsvno_accchk(*vpp, VWRITE,
 835                             nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
 836                             NFSACCCHK_VPISLOCKED, NULL);
 837                         if (!error) {
 838                                 tempsize = nvap->na_size;
 839                                 NFSVNO_ATTRINIT(nvap);
 840                                 nvap->na_size = tempsize;
 841                                 error = VOP_SETATTR(*vpp,
 842                                     &nvap->na_vattr, nd->nd_cred);
 843                         }
 844                 }
 845                 if (error)
 846                         vput(*vpp);
 847         }
 848
 849 out:
 850         NFSEXITCODE(error);
 851         return (error);
 852 }
 853
 854 /*
 855  * Do a mknod vnode op.
 856  */
 857 int
 858 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
 859     struct thread *p)
 860 {
 861         int error = 0;
 862         enum vtype vtyp;
 863
 864         vtyp = nvap->na_type;
 865         /*
 866          * Iff doesn't exist, create it.
 867          */
 868         if (ndp->ni_vp) {
 869                 vrele(ndp->ni_startdir);
 870                 nfsvno_relpathbuf(ndp);
 871                 vput(ndp->ni_dvp);
 872                 vrele(ndp->ni_vp);
 873                 error = EEXIST;
 874                 goto out;
 875         }
 876         if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
 877                 vrele(ndp->ni_startdir);
 878                 nfsvno_relpathbuf(ndp);
 879                 vput(ndp->ni_dvp);
 880                 error = NFSERR_BADTYPE;
 881                 goto out;
 882         }
 883         if (vtyp == VSOCK) {
 884                 vrele(ndp->ni_startdir);
 885                 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
 886                     &ndp->ni_cnd, &nvap->na_vattr);
 887                 vput(ndp->ni_dvp);
 888                 nfsvno_relpathbuf(ndp);
 889         } else {
 890                 if (nvap->na_type != VFIFO &&
 891                     (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
 892                         vrele(ndp->ni_startdir);
 893                         nfsvno_relpathbuf(ndp);
 894                         vput(ndp->ni_dvp);
 895                         goto out;
 896                 }
 897                 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 898                     &ndp->ni_cnd, &nvap->na_vattr);
 899                 vput(ndp->ni_dvp);
 900                 nfsvno_relpathbuf(ndp);
 901                 vrele(ndp->ni_startdir);
 902                 /*
 903                  * Since VOP_MKNOD returns the ni_vp, I can't
 904                  * see any reason to do the lookup.
 905                  */
 906         }
 907
 908 out:
 909         NFSEXITCODE(error);
 910         return (error);
 911 }
 912
 913 /*
 914  * Mkdir vnode op.
 915  */
 916 int
 917 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
 918     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
 919 {
 920         int error = 0;
 921
 922         if (ndp->ni_vp != NULL) {
 923                 if (ndp->ni_dvp == ndp->ni_vp)
 924                         vrele(ndp->ni_dvp);
 925                 else
 926                         vput(ndp->ni_dvp);
 927                 vrele(ndp->ni_vp);
 928                 nfsvno_relpathbuf(ndp);
 929                 error = EEXIST;
 930                 goto out;
 931         }
 932         error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 933             &nvap->na_vattr);
 934         vput(ndp->ni_dvp);
 935         nfsvno_relpathbuf(ndp);
 936
 937 out:
 938         NFSEXITCODE(error);
 939         return (error);
 940 }
 941
 942 /*
 943  * symlink vnode op.
 944  */
 945 int
 946 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
 947     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
 948     struct nfsexstuff *exp)
 949 {
 950         int error = 0;
 951
 952         if (ndp->ni_vp) {
 953                 vrele(ndp->ni_startdir);
 954                 nfsvno_relpathbuf(ndp);
 955                 if (ndp->ni_dvp == ndp->ni_vp)
 956                         vrele(ndp->ni_dvp);
 957                 else
 958                         vput(ndp->ni_dvp);
 959                 vrele(ndp->ni_vp);
 960                 error = EEXIST;
 961                 goto out;
 962         }
 963
 964         error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 965             &nvap->na_vattr, pathcp);
 966         vput(ndp->ni_dvp);
 967         vrele(ndp->ni_startdir);
 968         nfsvno_relpathbuf(ndp);
 969         /*
 970          * Although FreeBSD still had the lookup code in
 971          * it for 7/current, there doesn't seem to be any
 972          * point, since VOP_SYMLINK() returns the ni_vp.
 973          * Just vput it for v2.
 974          */
 975         if (!not_v2 && !error)
 976                 vput(ndp->ni_vp);
 977
 978 out:
 979         NFSEXITCODE(error);
 980         return (error);
 981 }
 982
 983 /*
 984  * Parse symbolic link arguments.
 985  * This function has an ugly side effect. It will MALLOC() an area for
 986  * the symlink and set iov_base to point to it, only if it succeeds.
 987  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
 988  * be FREE'd later.
 989  */
 990 int
 991 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
 992     struct thread *p, char **pathcpp, int *lenp)
 993 {
 994         u_int32_t *tl;
 995         char *pathcp = NULL;
 996         int error = 0, len;
 997         struct nfsv2_sattr *sp;
 998
 999         *pathcpp = NULL;
1000         *lenp = 0;
1001         if ((nd->nd_flag & ND_NFSV3) &&
1002             (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1003                 goto nfsmout;
1004         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1005         len = fxdr_unsigned(int, *tl);
1006         if (len > NFS_MAXPATHLEN || len <= 0) {
1007                 error = EBADRPC;
1008                 goto nfsmout;
1009         }
1010         MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1011         error = nfsrv_mtostr(nd, pathcp, len);
1012         if (error)
1013                 goto nfsmout;
1014         if (nd->nd_flag & ND_NFSV2) {
1015                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1016                 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1017         }
1018         *pathcpp = pathcp;
1019         *lenp = len;
1020         NFSEXITCODE2(0, nd);
1021         return (0);
1022 nfsmout:
1023         if (pathcp)
1024                 free(pathcp, M_TEMP);
1025         NFSEXITCODE2(error, nd);
1026         return (error);
1027 }
1028
1029 /*
1030  * Remove a non-directory object.
1031  */
1032 int
1033 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1034     struct thread *p, struct nfsexstuff *exp)
1035 {
1036         struct vnode *vp;
1037         int error = 0;
1038
1039         vp = ndp->ni_vp;
1040         if (vp->v_type == VDIR)
1041                 error = NFSERR_ISDIR;
1042         else if (is_v4)
1043                 error = nfsrv_checkremove(vp, 1, p);
1044         if (!error)
1045                 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1046         if (ndp->ni_dvp == vp)
1047                 vrele(ndp->ni_dvp);
1048         else
1049                 vput(ndp->ni_dvp);
1050         vput(vp);
1051         if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1052                 nfsvno_relpathbuf(ndp);
1053         NFSEXITCODE(error);
1054         return (error);
1055 }
1056
1057 /*
1058  * Remove a directory.
1059  */
1060 int
1061 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1062     struct thread *p, struct nfsexstuff *exp)
1063 {
1064         struct vnode *vp;
1065         int error = 0;
1066
1067         vp = ndp->ni_vp;
1068         if (vp->v_type != VDIR) {
1069                 error = ENOTDIR;
1070                 goto out;
1071         }
1072         /*
1073          * No rmdir "." please.
1074          */
1075         if (ndp->ni_dvp == vp) {
1076                 error = EINVAL;
1077                 goto out;
1078         }
1079         /*
1080          * The root of a mounted filesystem cannot be deleted.
1081          */
1082         if (vp->v_vflag & VV_ROOT)
1083                 error = EBUSY;
1084 out:
1085         if (!error)
1086                 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1087         if (ndp->ni_dvp == vp)
1088                 vrele(ndp->ni_dvp);
1089         else
1090                 vput(ndp->ni_dvp);
1091         vput(vp);
1092         if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1093                 nfsvno_relpathbuf(ndp);
1094         NFSEXITCODE(error);
1095         return (error);
1096 }
1097
1098 /*
1099  * Rename vnode op.
1100  */
1101 int
1102 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1103     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1104 {
1105         struct vnode *fvp, *tvp, *tdvp;
1106         int error = 0;
1107
1108         fvp = fromndp->ni_vp;
1109         if (ndstat) {
1110                 vrele(fromndp->ni_dvp);
1111                 vrele(fvp);
1112                 error = ndstat;
1113                 goto out1;
1114         }
1115         tdvp = tondp->ni_dvp;
1116         tvp = tondp->ni_vp;
1117         if (tvp != NULL) {
1118                 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1119                         error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1120                         goto out;
1121                 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1122                         error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1123                         goto out;
1124                 }
1125                 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1126                         error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1127                         goto out;
1128                 }
1129
1130                 /*
1131                  * A rename to '.' or '..' results in a prematurely
1132                  * unlocked vnode on FreeBSD5, so I'm just going to fail that
1133                  * here.
1134                  */
1135                 if ((tondp->ni_cnd.cn_namelen == 1 &&
1136                      tondp->ni_cnd.cn_nameptr[0] == '.') ||
1137                     (tondp->ni_cnd.cn_namelen == 2 &&
1138                      tondp->ni_cnd.cn_nameptr[0] == '.' &&
1139                      tondp->ni_cnd.cn_nameptr[1] == '.')) {
1140                         error = EINVAL;
1141                         goto out;
1142                 }
1143         }
1144         if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1145                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1146                 goto out;
1147         }
1148         if (fvp->v_mount != tdvp->v_mount) {
1149                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1150                 goto out;
1151         }
1152         if (fvp == tdvp) {
1153                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1154                 goto out;
1155         }
1156         if (fvp == tvp) {
1157                 /*
1158                  * If source and destination are the same, there is nothing to
1159                  * do. Set error to -1 to indicate this.
1160                  */
1161                 error = -1;
1162                 goto out;
1163         }
1164         if (ndflag & ND_NFSV4) {
1165                 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1166                         error = nfsrv_checkremove(fvp, 0, p);
1167                         NFSVOPUNLOCK(fvp, 0);
1168                 } else
1169                         error = EPERM;
1170                 if (tvp && !error)
1171                         error = nfsrv_checkremove(tvp, 1, p);
1172         } else {
1173                 /*
1174                  * For NFSv2 and NFSv3, try to get rid of the delegation, so
1175                  * that the NFSv4 client won't be confused by the rename.
1176                  * Since nfsd_recalldelegation() can only be called on an
1177                  * unlocked vnode at this point and fvp is the file that will
1178                  * still exist after the rename, just do fvp.
1179                  */
1180                 nfsd_recalldelegation(fvp, p);
1181         }
1182 out:
1183         if (!error) {
1184                 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1185                     &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1186                     &tondp->ni_cnd);
1187         } else {
1188                 if (tdvp == tvp)
1189                         vrele(tdvp);
1190                 else
1191                         vput(tdvp);
1192                 if (tvp)
1193                         vput(tvp);
1194                 vrele(fromndp->ni_dvp);
1195                 vrele(fvp);
1196                 if (error == -1)
1197                         error = 0;
1198         }
1199         vrele(tondp->ni_startdir);
1200         nfsvno_relpathbuf(tondp);
1201 out1:
1202         vrele(fromndp->ni_startdir);
1203         nfsvno_relpathbuf(fromndp);
1204         NFSEXITCODE(error);
1205         return (error);
1206 }
1207
1208 /*
1209  * Link vnode op.
1210  */
1211 int
1212 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1213     struct thread *p, struct nfsexstuff *exp)
1214 {
1215         struct vnode *xp;
1216         int error = 0;
1217
1218         xp = ndp->ni_vp;
1219         if (xp != NULL) {
1220                 error = EEXIST;
1221         } else {
1222                 xp = ndp->ni_dvp;
1223                 if (vp->v_mount != xp->v_mount)
1224                         error = EXDEV;
1225         }
1226         if (!error) {
1227                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1228                 if ((vp->v_iflag & VI_DOOMED) == 0)
1229                         error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1230                 else
1231                         error = EPERM;
1232                 if (ndp->ni_dvp == vp)
1233                         vrele(ndp->ni_dvp);
1234                 else
1235                         vput(ndp->ni_dvp);
1236                 NFSVOPUNLOCK(vp, 0);
1237         } else {
1238                 if (ndp->ni_dvp == ndp->ni_vp)
1239                         vrele(ndp->ni_dvp);
1240                 else
1241                         vput(ndp->ni_dvp);
1242                 if (ndp->ni_vp)
1243                         vrele(ndp->ni_vp);
1244         }
1245         nfsvno_relpathbuf(ndp);
1246         NFSEXITCODE(error);
1247         return (error);
1248 }
1249
1250 /*
1251  * Do the fsync() appropriate for the commit.
1252  */
1253 int
1254 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1255     struct thread *td)
1256 {
1257         int error = 0;
1258
1259         /*
1260          * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1261          * file is done.  At this time VOP_FSYNC does not accept offset and
1262          * byte count parameters so call VOP_FSYNC the whole file for now.
1263          * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1264          */
1265         if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1266                 /*
1267                  * Give up and do the whole thing
1268                  */
1269                 if (vp->v_object &&
1270                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1271                         VM_OBJECT_LOCK(vp->v_object);
1272                         vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1273                         VM_OBJECT_UNLOCK(vp->v_object);
1274                 }
1275                 error = VOP_FSYNC(vp, MNT_WAIT, td);
1276         } else {
1277                 /*
1278                  * Locate and synchronously write any buffers that fall
1279                  * into the requested range.  Note:  we are assuming that
1280                  * f_iosize is a power of 2.
1281                  */
1282                 int iosize = vp->v_mount->mnt_stat.f_iosize;
1283                 int iomask = iosize - 1;
1284                 struct bufobj *bo;
1285                 daddr_t lblkno;
1286
1287                 /*
1288                  * Align to iosize boundry, super-align to page boundry.
1289                  */
1290                 if (off & iomask) {
1291                         cnt += off & iomask;
1292                         off &= ~(u_quad_t)iomask;
1293                 }
1294                 if (off & PAGE_MASK) {
1295                         cnt += off & PAGE_MASK;
1296                         off &= ~(u_quad_t)PAGE_MASK;
1297                 }
1298                 lblkno = off / iosize;
1299
1300                 if (vp->v_object &&
1301                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1302                         VM_OBJECT_LOCK(vp->v_object);
1303                         vm_object_page_clean(vp->v_object, off, off + cnt,
1304                             OBJPC_SYNC);
1305                         VM_OBJECT_UNLOCK(vp->v_object);
1306                 }
1307
1308                 bo = &vp->v_bufobj;
1309                 BO_LOCK(bo);
1310                 while (cnt > 0) {
1311                         struct buf *bp;
1312
1313                         /*
1314                          * If we have a buffer and it is marked B_DELWRI we
1315                          * have to lock and write it.  Otherwise the prior
1316                          * write is assumed to have already been committed.
1317                          *
1318                          * gbincore() can return invalid buffers now so we
1319                          * have to check that bit as well (though B_DELWRI
1320                          * should not be set if B_INVAL is set there could be
1321                          * a race here since we haven't locked the buffer).
1322                          */
1323                         if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1324                                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1325                                     LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1326                                         BO_LOCK(bo);
1327                                         continue; /* retry */
1328                                 }
1329                                 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1330                                     B_DELWRI) {
1331                                         bremfree(bp);
1332                                         bp->b_flags &= ~B_ASYNC;
1333                                         bwrite(bp);
1334                                         ++nfs_commit_miss;
1335                                 } else
1336                                         BUF_UNLOCK(bp);
1337                                 BO_LOCK(bo);
1338                         }
1339                         ++nfs_commit_blks;
1340                         if (cnt < iosize)
1341                                 break;
1342                         cnt -= iosize;
1343                         ++lblkno;
1344                 }
1345                 BO_UNLOCK(bo);
1346         }
1347         NFSEXITCODE(error);
1348         return (error);
1349 }
1350
1351 /*
1352  * Statfs vnode op.
1353  */
1354 int
1355 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1356 {
1357         int error;
1358
1359         error = VFS_STATFS(vp->v_mount, sf);
1360         if (error == 0) {
1361                 /*
1362                  * Since NFS handles these values as unsigned on the
1363                  * wire, there is no way to represent negative values,
1364                  * so set them to 0. Without this, they will appear
1365                  * to be very large positive values for clients like
1366                  * Solaris10.
1367                  */
1368                 if (sf->f_bavail < 0)
1369                         sf->f_bavail = 0;
1370                 if (sf->f_ffree < 0)
1371                         sf->f_ffree = 0;
1372         }
1373         NFSEXITCODE(error);
1374         return (error);
1375 }
1376
1377 /*
1378  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1379  * must handle nfsrv_opencheck() calls after any other access checks.
1380  */
1381 void
1382 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1383     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1384     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1385     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1386     struct nfsexstuff *exp, struct vnode **vpp)
1387 {
1388         struct vnode *vp = NULL;
1389         u_quad_t tempsize;
1390         struct nfsexstuff nes;
1391
1392         if (ndp->ni_vp == NULL)
1393                 nd->nd_repstat = nfsrv_opencheck(clientid,
1394                     stateidp, stp, NULL, nd, p, nd->nd_repstat);
1395         if (!nd->nd_repstat) {
1396                 if (ndp->ni_vp == NULL) {
1397                         vrele(ndp->ni_startdir);
1398                         nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1399                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1400                         vput(ndp->ni_dvp);
1401                         nfsvno_relpathbuf(ndp);
1402                         if (!nd->nd_repstat) {
1403                                 if (*exclusive_flagp) {
1404                                         *exclusive_flagp = 0;
1405                                         NFSVNO_ATTRINIT(nvap);
1406                                         nvap->na_atime.tv_sec = cverf[0];
1407                                         nvap->na_atime.tv_nsec = cverf[1];
1408                                         nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1409                                             &nvap->na_vattr, cred);
1410                                 } else {
1411                                         nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1412                                             aclp, p, attrbitp, exp);
1413                                 }
1414                         }
1415                         vp = ndp->ni_vp;
1416                 } else {
1417                         if (ndp->ni_startdir)
1418                                 vrele(ndp->ni_startdir);
1419                         nfsvno_relpathbuf(ndp);
1420                         vp = ndp->ni_vp;
1421                         if (create == NFSV4OPEN_CREATE) {
1422                                 if (ndp->ni_dvp == vp)
1423                                         vrele(ndp->ni_dvp);
1424                                 else
1425                                         vput(ndp->ni_dvp);
1426                         }
1427                         if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1428                                 if (ndp->ni_cnd.cn_flags & RDONLY)
1429                                         NFSVNO_SETEXRDONLY(&nes);
1430                                 else
1431                                         NFSVNO_EXINIT(&nes);
1432                                 nd->nd_repstat = nfsvno_accchk(vp,
1433                                     VWRITE, cred, &nes, p,
1434                                     NFSACCCHK_NOOVERRIDE,
1435                                     NFSACCCHK_VPISLOCKED, NULL);
1436                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1437                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1438                                 if (!nd->nd_repstat) {
1439                                         tempsize = nvap->na_size;
1440                                         NFSVNO_ATTRINIT(nvap);
1441                                         nvap->na_size = tempsize;
1442                                         nd->nd_repstat = VOP_SETATTR(vp,
1443                                             &nvap->na_vattr, cred);
1444                                 }
1445                         } else if (vp->v_type == VREG) {
1446                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1447                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1448                         }
1449                 }
1450         } else {
1451                 if (ndp->ni_cnd.cn_flags & HASBUF)
1452                         nfsvno_relpathbuf(ndp);
1453                 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1454                         vrele(ndp->ni_startdir);
1455                         if (ndp->ni_dvp == ndp->ni_vp)
1456                                 vrele(ndp->ni_dvp);
1457                         else
1458                                 vput(ndp->ni_dvp);
1459                         if (ndp->ni_vp)
1460                                 vput(ndp->ni_vp);
1461                 }
1462         }
1463         *vpp = vp;
1464
1465         NFSEXITCODE2(0, nd);
1466 }
1467
1468 /*
1469  * Updates the file rev and sets the mtime and ctime
1470  * to the current clock time, returning the va_filerev and va_Xtime
1471  * values.
1472  */
1473 void
1474 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1475     struct ucred *cred, struct thread *p)
1476 {
1477         struct vattr va;
1478
1479         VATTR_NULL(&va);
1480         vfs_timestamp(&va.va_mtime);
1481         (void) VOP_SETATTR(vp, &va, cred);
1482         (void) nfsvno_getattr(vp, nvap, cred, p, 1);
1483 }
1484
1485 /*
1486  * Glue routine to nfsv4_fillattr().
1487  */
1488 int
1489 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1490     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1491     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1492     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1493 {
1494         int error;
1495
1496         error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1497             attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1498             mounted_on_fileno);
1499         NFSEXITCODE2(0, nd);
1500         return (error);
1501 }
1502
1503 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1504 /*
1505  * nfs readdir service
1506  * - mallocs what it thinks is enough to read
1507  *      count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1508  * - calls VOP_READDIR()
1509  * - loops around building the reply
1510  *      if the output generated exceeds count break out of loop
1511  *      The NFSM_CLGET macro is used here so that the reply will be packed
1512  *      tightly in mbuf clusters.
1513  * - it trims out records with d_fileno == 0
1514  *      this doesn't matter for Unix clients, but they might confuse clients
1515  *      for other os'.
1516  * - it trims out records with d_type == DT_WHT
1517  *      these cannot be seen through NFS (unless we extend the protocol)
1518  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1519  * PS: The NFS protocol spec. does not clarify what the "count" byte
1520  *      argument is a count of.. just name strings and file id's or the
1521  *      entire reply rpc or ...
1522  *      I tried just file name and id sizes and it confused the Sun client,
1523  *      so I am using the full rpc size now. The "paranoia.." comment refers
1524  *      to including the status longwords that are not a part of the dir.
1525  *      "entry" structures, but are in the rpc.
1526  */
1527 int
1528 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1529     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1530 {
1531         struct dirent *dp;
1532         u_int32_t *tl;
1533         int dirlen;
1534         char *cpos, *cend, *rbuf;
1535         struct nfsvattr at;
1536         int nlen, error = 0, getret = 1;
1537         int siz, cnt, fullsiz, eofflag, ncookies;
1538         u_int64_t off, toff, verf;
1539         u_long *cookies = NULL, *cookiep;
1540         struct uio io;
1541         struct iovec iv;
1542         int not_zfs;
1543
1544         if (nd->nd_repstat) {
1545                 nfsrv_postopattr(nd, getret, &at);
1546                 goto out;
1547         }
1548         if (nd->nd_flag & ND_NFSV2) {
1549                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1550                 off = fxdr_unsigned(u_quad_t, *tl++);
1551         } else {
1552                 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1553                 off = fxdr_hyper(tl);
1554                 tl += 2;
1555                 verf = fxdr_hyper(tl);
1556                 tl += 2;
1557         }
1558         toff = off;
1559         cnt = fxdr_unsigned(int, *tl);
1560         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1561                 cnt = NFS_SRVMAXDATA(nd);
1562         siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1563         fullsiz = siz;
1564         if (nd->nd_flag & ND_NFSV3) {
1565                 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1566                     p, 1);
1567 #if 0
1568                 /*
1569                  * va_filerev is not sufficient as a cookie verifier,
1570                  * since it is not supposed to change when entries are
1571                  * removed/added unless that offset cookies returned to
1572                  * the client are no longer valid.
1573                  */
1574                 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1575                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1576 #endif
1577         }
1578         if (!nd->nd_repstat && vp->v_type != VDIR)
1579                 nd->nd_repstat = NFSERR_NOTDIR;
1580         if (nd->nd_repstat == 0 && cnt == 0) {
1581                 if (nd->nd_flag & ND_NFSV2)
1582                         /* NFSv2 does not have NFSERR_TOOSMALL */
1583                         nd->nd_repstat = EPERM;
1584                 else
1585                         nd->nd_repstat = NFSERR_TOOSMALL;
1586         }
1587         if (!nd->nd_repstat)
1588                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1589                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1590                     NFSACCCHK_VPISLOCKED, NULL);
1591         if (nd->nd_repstat) {
1592                 vput(vp);
1593                 if (nd->nd_flag & ND_NFSV3)
1594                         nfsrv_postopattr(nd, getret, &at);
1595                 goto out;
1596         }
1597         not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1598         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1599 again:
1600         eofflag = 0;
1601         if (cookies) {
1602                 free((caddr_t)cookies, M_TEMP);
1603                 cookies = NULL;
1604         }
1605
1606         iv.iov_base = rbuf;
1607         iv.iov_len = siz;
1608         io.uio_iov = &iv;
1609         io.uio_iovcnt = 1;
1610         io.uio_offset = (off_t)off;
1611         io.uio_resid = siz;
1612         io.uio_segflg = UIO_SYSSPACE;
1613         io.uio_rw = UIO_READ;
1614         io.uio_td = NULL;
1615         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1616             &cookies);
1617         off = (u_int64_t)io.uio_offset;
1618         if (io.uio_resid)
1619                 siz -= io.uio_resid;
1620
1621         if (!cookies && !nd->nd_repstat)
1622                 nd->nd_repstat = NFSERR_PERM;
1623         if (nd->nd_flag & ND_NFSV3) {
1624                 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1625                 if (!nd->nd_repstat)
1626                         nd->nd_repstat = getret;
1627         }
1628
1629         /*
1630          * Handles the failed cases. nd->nd_repstat == 0 past here.
1631          */
1632         if (nd->nd_repstat) {
1633                 vput(vp);
1634                 free((caddr_t)rbuf, M_TEMP);
1635                 if (cookies)
1636                         free((caddr_t)cookies, M_TEMP);
1637                 if (nd->nd_flag & ND_NFSV3)
1638                         nfsrv_postopattr(nd, getret, &at);
1639                 goto out;
1640         }
1641         /*
1642          * If nothing read, return eof
1643          * rpc reply
1644          */
1645         if (siz == 0) {
1646                 vput(vp);
1647                 if (nd->nd_flag & ND_NFSV2) {
1648                         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1649                 } else {
1650                         nfsrv_postopattr(nd, getret, &at);
1651                         NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1652                         txdr_hyper(at.na_filerev, tl);
1653                         tl += 2;
1654                 }
1655                 *tl++ = newnfs_false;
1656                 *tl = newnfs_true;
1657                 FREE((caddr_t)rbuf, M_TEMP);
1658                 FREE((caddr_t)cookies, M_TEMP);
1659                 goto out;
1660         }
1661
1662         /*
1663          * Check for degenerate cases of nothing useful read.
1664          * If so go try again
1665          */
1666         cpos = rbuf;
1667         cend = rbuf + siz;
1668         dp = (struct dirent *)cpos;
1669         cookiep = cookies;
1670
1671         /*
1672          * For some reason FreeBSD's ufs_readdir() chooses to back the
1673          * directory offset up to a block boundary, so it is necessary to
1674          * skip over the records that precede the requested offset. This
1675          * requires the assumption that file offset cookies monotonically
1676          * increase.
1677          * Since the offset cookies don't monotonically increase for ZFS,
1678          * this is not done when ZFS is the file system.
1679          */
1680         while (cpos < cend && ncookies > 0 &&
1681             (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1682              (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1683                 cpos += dp->d_reclen;
1684                 dp = (struct dirent *)cpos;
1685                 cookiep++;
1686                 ncookies--;
1687         }
1688         if (cpos >= cend || ncookies == 0) {
1689                 siz = fullsiz;
1690                 toff = off;
1691                 goto again;
1692         }
1693         vput(vp);
1694
1695         /*
1696          * dirlen is the size of the reply, including all XDR and must
1697          * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1698          * if the XDR should be included in "count", but to be safe, we do.
1699          * (Include the two booleans at the end of the reply in dirlen now.)
1700          */
1701         if (nd->nd_flag & ND_NFSV3) {
1702                 nfsrv_postopattr(nd, getret, &at);
1703                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1704                 txdr_hyper(at.na_filerev, tl);
1705                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1706         } else {
1707                 dirlen = 2 * NFSX_UNSIGNED;
1708         }
1709
1710         /* Loop through the records and build reply */
1711         while (cpos < cend && ncookies > 0) {
1712                 nlen = dp->d_namlen;
1713                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1714                         nlen <= NFS_MAXNAMLEN) {
1715                         if (nd->nd_flag & ND_NFSV3)
1716                                 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1717                         else
1718                                 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1719                         if (dirlen > cnt) {
1720                                 eofflag = 0;
1721                                 break;
1722                         }
1723
1724                         /*
1725                          * Build the directory record xdr from
1726                          * the dirent entry.
1727                          */
1728                         if (nd->nd_flag & ND_NFSV3) {
1729                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1730                                 *tl++ = newnfs_true;
1731                                 *tl++ = 0;
1732                         } else {
1733                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1734                                 *tl++ = newnfs_true;
1735                         }
1736                         *tl = txdr_unsigned(dp->d_fileno);
1737                         (void) nfsm_strtom(nd, dp->d_name, nlen);
1738                         if (nd->nd_flag & ND_NFSV3) {
1739                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1740                                 *tl++ = 0;
1741                         } else
1742                                 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1743                         *tl = txdr_unsigned(*cookiep);
1744                 }
1745                 cpos += dp->d_reclen;
1746                 dp = (struct dirent *)cpos;
1747                 cookiep++;
1748                 ncookies--;
1749         }
1750         if (cpos < cend)
1751                 eofflag = 0;
1752         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1753         *tl++ = newnfs_false;
1754         if (eofflag)
1755                 *tl = newnfs_true;
1756         else
1757                 *tl = newnfs_false;
1758         FREE((caddr_t)rbuf, M_TEMP);
1759         FREE((caddr_t)cookies, M_TEMP);
1760
1761 out:
1762         NFSEXITCODE2(0, nd);
1763         return (0);
1764 nfsmout:
1765         vput(vp);
1766         NFSEXITCODE2(error, nd);
1767         return (error);
1768 }
1769
1770 /*
1771  * Readdirplus for V3 and Readdir for V4.
1772  */
1773 int
1774 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1775     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1776 {
1777         struct dirent *dp;
1778         u_int32_t *tl;
1779         int dirlen;
1780         char *cpos, *cend, *rbuf;
1781         struct vnode *nvp;
1782         fhandle_t nfh;
1783         struct nfsvattr nva, at, *nvap = &nva;
1784         struct mbuf *mb0, *mb1;
1785         struct nfsreferral *refp;
1786         int nlen, r, error = 0, getret = 1, usevget = 1;
1787         int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1788         caddr_t bpos0, bpos1;
1789         u_int64_t off, toff, verf;
1790         u_long *cookies = NULL, *cookiep;
1791         nfsattrbit_t attrbits, rderrbits, savbits;
1792         struct uio io;
1793         struct iovec iv;
1794         struct componentname cn;
1795         int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1796         struct mount *mp, *new_mp;
1797         uint64_t mounted_on_fileno;
1798
1799         if (nd->nd_repstat) {
1800                 nfsrv_postopattr(nd, getret, &at);
1801                 goto out;
1802         }
1803         NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1804         off = fxdr_hyper(tl);
1805         toff = off;
1806         tl += 2;
1807         verf = fxdr_hyper(tl);
1808         tl += 2;
1809         siz = fxdr_unsigned(int, *tl++);
1810         cnt = fxdr_unsigned(int, *tl);
1811
1812         /*
1813          * Use the server's maximum data transfer size as the upper bound
1814          * on reply datalen.
1815          */
1816         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1817                 cnt = NFS_SRVMAXDATA(nd);
1818
1819         /*
1820          * siz is a "hint" of how much directory information (name, fileid,
1821          * cookie) should be in the reply. At least one client "hints" 0,
1822          * so I set it to cnt for that case. I also round it up to the
1823          * next multiple of DIRBLKSIZ.
1824          */
1825         if (siz <= 0)
1826                 siz = cnt;
1827         siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1828
1829         if (nd->nd_flag & ND_NFSV4) {
1830                 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1831                 if (error)
1832                         goto nfsmout;
1833                 NFSSET_ATTRBIT(&savbits, &attrbits);
1834                 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1835                 NFSZERO_ATTRBIT(&rderrbits);
1836                 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1837         } else {
1838                 NFSZERO_ATTRBIT(&attrbits);
1839         }
1840         fullsiz = siz;
1841         nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1842         if (!nd->nd_repstat) {
1843             if (off && verf != at.na_filerev) {
1844                 /*
1845                  * va_filerev is not sufficient as a cookie verifier,
1846                  * since it is not supposed to change when entries are
1847                  * removed/added unless that offset cookies returned to
1848                  * the client are no longer valid.
1849                  */
1850 #if 0
1851                 if (nd->nd_flag & ND_NFSV4) {
1852                         nd->nd_repstat = NFSERR_NOTSAME;
1853                 } else {
1854                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1855                 }
1856 #endif
1857             } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1858                 nd->nd_repstat = NFSERR_BAD_COOKIE;
1859             }
1860         }
1861         if (!nd->nd_repstat && vp->v_type != VDIR)
1862                 nd->nd_repstat = NFSERR_NOTDIR;
1863         if (!nd->nd_repstat && cnt == 0)
1864                 nd->nd_repstat = NFSERR_TOOSMALL;
1865         if (!nd->nd_repstat)
1866                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1867                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1868                     NFSACCCHK_VPISLOCKED, NULL);
1869         if (nd->nd_repstat) {
1870                 vput(vp);
1871                 if (nd->nd_flag & ND_NFSV3)
1872                         nfsrv_postopattr(nd, getret, &at);
1873                 goto out;
1874         }
1875         not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1876
1877         MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1878 again:
1879         eofflag = 0;
1880         if (cookies) {
1881                 free((caddr_t)cookies, M_TEMP);
1882                 cookies = NULL;
1883         }
1884
1885         iv.iov_base = rbuf;
1886         iv.iov_len = siz;
1887         io.uio_iov = &iv;
1888         io.uio_iovcnt = 1;
1889         io.uio_offset = (off_t)off;
1890         io.uio_resid = siz;
1891         io.uio_segflg = UIO_SYSSPACE;
1892         io.uio_rw = UIO_READ;
1893         io.uio_td = NULL;
1894         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1895             &cookies);
1896         off = (u_int64_t)io.uio_offset;
1897         if (io.uio_resid)
1898                 siz -= io.uio_resid;
1899
1900         getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1901
1902         if (!cookies && !nd->nd_repstat)
1903                 nd->nd_repstat = NFSERR_PERM;
1904         if (!nd->nd_repstat)
1905                 nd->nd_repstat = getret;
1906         if (nd->nd_repstat) {
1907                 vput(vp);
1908                 if (cookies)
1909                         free((caddr_t)cookies, M_TEMP);
1910                 free((caddr_t)rbuf, M_TEMP);
1911                 if (nd->nd_flag & ND_NFSV3)
1912                         nfsrv_postopattr(nd, getret, &at);
1913                 goto out;
1914         }
1915         /*
1916          * If nothing read, return eof
1917          * rpc reply
1918          */
1919         if (siz == 0) {
1920                 vput(vp);
1921                 if (nd->nd_flag & ND_NFSV3)
1922                         nfsrv_postopattr(nd, getret, &at);
1923                 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1924                 txdr_hyper(at.na_filerev, tl);
1925                 tl += 2;
1926                 *tl++ = newnfs_false;
1927                 *tl = newnfs_true;
1928                 free((caddr_t)cookies, M_TEMP);
1929                 free((caddr_t)rbuf, M_TEMP);
1930                 goto out;
1931         }
1932
1933         /*
1934          * Check for degenerate cases of nothing useful read.
1935          * If so go try again
1936          */
1937         cpos = rbuf;
1938         cend = rbuf + siz;
1939         dp = (struct dirent *)cpos;
1940         cookiep = cookies;
1941
1942         /*
1943          * For some reason FreeBSD's ufs_readdir() chooses to back the
1944          * directory offset up to a block boundary, so it is necessary to
1945          * skip over the records that precede the requested offset. This
1946          * requires the assumption that file offset cookies monotonically
1947          * increase.
1948          * Since the offset cookies don't monotonically increase for ZFS,
1949          * this is not done when ZFS is the file system.
1950          */
1951         while (cpos < cend && ncookies > 0 &&
1952           (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1953            (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1954            ((nd->nd_flag & ND_NFSV4) &&
1955             ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1956              (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1957                 cpos += dp->d_reclen;
1958                 dp = (struct dirent *)cpos;
1959                 cookiep++;
1960                 ncookies--;
1961         }
1962         if (cpos >= cend || ncookies == 0) {
1963                 siz = fullsiz;
1964                 toff = off;
1965                 goto again;
1966         }
1967
1968         /*
1969          * Busy the file system so that the mount point won't go away
1970          * and, as such, VFS_VGET() can be used safely.
1971          */
1972         mp = vp->v_mount;
1973         vfs_ref(mp);
1974         NFSVOPUNLOCK(vp, 0);
1975         nd->nd_repstat = vfs_busy(mp, 0);
1976         vfs_rel(mp);
1977         if (nd->nd_repstat != 0) {
1978                 vrele(vp);
1979                 free(cookies, M_TEMP);
1980                 free(rbuf, M_TEMP);
1981                 if (nd->nd_flag & ND_NFSV3)
1982                         nfsrv_postopattr(nd, getret, &at);
1983                 goto out;
1984         }
1985
1986         /*
1987          * Save this position, in case there is an error before one entry
1988          * is created.
1989          */
1990         mb0 = nd->nd_mb;
1991         bpos0 = nd->nd_bpos;
1992
1993         /*
1994          * Fill in the first part of the reply.
1995          * dirlen is the reply length in bytes and cannot exceed cnt.
1996          * (Include the two booleans at the end of the reply in dirlen now,
1997          *  so we recognize when we have exceeded cnt.)
1998          */
1999         if (nd->nd_flag & ND_NFSV3) {
2000                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2001                 nfsrv_postopattr(nd, getret, &at);
2002         } else {
2003                 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2004         }
2005         NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2006         txdr_hyper(at.na_filerev, tl);
2007
2008         /*
2009          * Save this position, in case there is an empty reply needed.
2010          */
2011         mb1 = nd->nd_mb;
2012         bpos1 = nd->nd_bpos;
2013
2014         /* Loop through the records and build reply */
2015         entrycnt = 0;
2016         while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2017                 nlen = dp->d_namlen;
2018                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2019                     nlen <= NFS_MAXNAMLEN &&
2020                     ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2021                      (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2022                       || (nlen == 1 && dp->d_name[0] != '.'))) {
2023                         /*
2024                          * Save the current position in the reply, in case
2025                          * this entry exceeds cnt.
2026                          */
2027                         mb1 = nd->nd_mb;
2028                         bpos1 = nd->nd_bpos;
2029
2030                         /*
2031                          * For readdir_and_lookup get the vnode using
2032                          * the file number.
2033                          */
2034                         nvp = NULL;
2035                         refp = NULL;
2036                         r = 0;
2037                         at_root = 0;
2038                         needs_unbusy = 0;
2039                         new_mp = mp;
2040                         mounted_on_fileno = (uint64_t)dp->d_fileno;
2041                         if ((nd->nd_flag & ND_NFSV3) ||
2042                             NFSNONZERO_ATTRBIT(&savbits)) {
2043                                 if (nd->nd_flag & ND_NFSV4)
2044                                         refp = nfsv4root_getreferral(NULL,
2045                                             vp, dp->d_fileno);
2046                                 if (refp == NULL) {
2047                                         if (usevget)
2048                                                 r = VFS_VGET(mp, dp->d_fileno,
2049                                                     LK_SHARED, &nvp);
2050                                         else
2051                                                 r = EOPNOTSUPP;
2052                                         if (r == EOPNOTSUPP) {
2053                                                 if (usevget) {
2054                                                         usevget = 0;
2055                                                         cn.cn_nameiop = LOOKUP;
2056                                                         cn.cn_lkflags =
2057                                                             LK_SHARED |
2058                                                             LK_RETRY;
2059                                                         cn.cn_cred =
2060                                                             nd->nd_cred;
2061                                                         cn.cn_thread = p;
2062                                                 }
2063                                                 cn.cn_nameptr = dp->d_name;
2064                                                 cn.cn_namelen = nlen;
2065                                                 cn.cn_flags = ISLASTCN |
2066                                                     NOFOLLOW | LOCKLEAF |
2067                                                     MPSAFE;
2068                                                 if (nlen == 2 &&
2069                                                     dp->d_name[0] == '.' &&
2070                                                     dp->d_name[1] == '.')
2071                                                         cn.cn_flags |=
2072                                                             ISDOTDOT;
2073                                                 if (NFSVOPLOCK(vp, LK_SHARED)
2074                                                     != 0) {
2075                                                         nd->nd_repstat = EPERM;
2076                                                         break;
2077                                                 }
2078                                                 if ((vp->v_vflag & VV_ROOT) != 0
2079                                                     && (cn.cn_flags & ISDOTDOT)
2080                                                     != 0) {
2081                                                         vref(vp);
2082                                                         nvp = vp;
2083                                                         r = 0;
2084                                                 } else {
2085                                                         r = VOP_LOOKUP(vp, &nvp,
2086                                                             &cn);
2087                                                         if (vp != nvp)
2088                                                                 NFSVOPUNLOCK(vp,
2089                                                                     0);
2090                                                 }
2091                                         }
2092
2093                                         /*
2094                                          * For NFSv4, check to see if nvp is
2095                                          * a mount point and get the mount
2096                                          * point vnode, as required.
2097                                          */
2098                                         if (r == 0 &&
2099                                             nfsrv_enable_crossmntpt != 0 &&
2100                                             (nd->nd_flag & ND_NFSV4) != 0 &&
2101                                             nvp->v_type == VDIR &&
2102                                             nvp->v_mountedhere != NULL) {
2103                                                 new_mp = nvp->v_mountedhere;
2104                                                 r = vfs_busy(new_mp, 0);
2105                                                 vput(nvp);
2106                                                 nvp = NULL;
2107                                                 if (r == 0) {
2108                                                         r = VFS_ROOT(new_mp,
2109                                                             LK_SHARED, &nvp);
2110                                                         needs_unbusy = 1;
2111                                                         if (r == 0)
2112                                                                 at_root = 1;
2113                                                 }
2114                                         }
2115                                 }
2116                                 if (!r) {
2117                                     if (refp == NULL &&
2118                                         ((nd->nd_flag & ND_NFSV3) ||
2119                                          NFSNONZERO_ATTRBIT(&attrbits))) {
2120                                         r = nfsvno_getfh(nvp, &nfh, p);
2121                                         if (!r)
2122                                             r = nfsvno_getattr(nvp, nvap,
2123                                                 nd->nd_cred, p, 1);
2124                                     }
2125                                 } else {
2126                                     nvp = NULL;
2127                                 }
2128                                 if (r) {
2129                                         if (!NFSISSET_ATTRBIT(&attrbits,
2130                                             NFSATTRBIT_RDATTRERROR)) {
2131                                                 if (nvp != NULL)
2132                                                         vput(nvp);
2133                                                 if (needs_unbusy != 0)
2134                                                         vfs_unbusy(new_mp);
2135                                                 nd->nd_repstat = r;
2136                                                 break;
2137                                         }
2138                                 }
2139                         }
2140
2141                         /*
2142                          * Build the directory record xdr
2143                          */
2144                         if (nd->nd_flag & ND_NFSV3) {
2145                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2146                                 *tl++ = newnfs_true;
2147                                 *tl++ = 0;
2148                                 *tl = txdr_unsigned(dp->d_fileno);
2149                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2150                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2151                                 *tl++ = 0;
2152                                 *tl = txdr_unsigned(*cookiep);
2153                                 nfsrv_postopattr(nd, 0, nvap);
2154                                 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2155                                 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2156                                 if (nvp != NULL)
2157                                         vput(nvp);
2158                         } else {
2159                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2160                                 *tl++ = newnfs_true;
2161                                 *tl++ = 0;
2162                                 *tl = txdr_unsigned(*cookiep);
2163                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2164                                 if (nvp != NULL) {
2165                                         supports_nfsv4acls =
2166                                             nfs_supportsnfsv4acls(nvp);
2167                                         NFSVOPUNLOCK(nvp, 0);
2168                                 } else
2169                                         supports_nfsv4acls = 0;
2170                                 if (refp != NULL) {
2171                                         dirlen += nfsrv_putreferralattr(nd,
2172                                             &savbits, refp, 0,
2173                                             &nd->nd_repstat);
2174                                         if (nd->nd_repstat) {
2175                                                 if (nvp != NULL)
2176                                                         vrele(nvp);
2177                                                 if (needs_unbusy != 0)
2178                                                         vfs_unbusy(new_mp);
2179                                                 break;
2180                                         }
2181                                 } else if (r) {
2182                                         dirlen += nfsvno_fillattr(nd, new_mp,
2183                                             nvp, nvap, &nfh, r, &rderrbits,
2184                                             nd->nd_cred, p, isdgram, 0,
2185                                             supports_nfsv4acls, at_root,
2186                                             mounted_on_fileno);
2187                                 } else {
2188                                         dirlen += nfsvno_fillattr(nd, new_mp,
2189                                             nvp, nvap, &nfh, r, &attrbits,
2190                                             nd->nd_cred, p, isdgram, 0,
2191                                             supports_nfsv4acls, at_root,
2192                                             mounted_on_fileno);
2193                                 }
2194                                 if (nvp != NULL)
2195                                         vrele(nvp);
2196                                 dirlen += (3 * NFSX_UNSIGNED);
2197                         }
2198                         if (needs_unbusy != 0)
2199                                 vfs_unbusy(new_mp);
2200                         if (dirlen <= cnt)
2201                                 entrycnt++;
2202                 }
2203                 cpos += dp->d_reclen;
2204                 dp = (struct dirent *)cpos;
2205                 cookiep++;
2206                 ncookies--;
2207         }
2208         vrele(vp);
2209         vfs_unbusy(mp);
2210
2211         /*
2212          * If dirlen > cnt, we must strip off the last entry. If that
2213          * results in an empty reply, report NFSERR_TOOSMALL.
2214          */
2215         if (dirlen > cnt || nd->nd_repstat) {
2216                 if (!nd->nd_repstat && entrycnt == 0)
2217                         nd->nd_repstat = NFSERR_TOOSMALL;
2218                 if (nd->nd_repstat)
2219                         newnfs_trimtrailing(nd, mb0, bpos0);
2220                 else
2221                         newnfs_trimtrailing(nd, mb1, bpos1);
2222                 eofflag = 0;
2223         } else if (cpos < cend)
2224                 eofflag = 0;
2225         if (!nd->nd_repstat) {
2226                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2227                 *tl++ = newnfs_false;
2228                 if (eofflag)
2229                         *tl = newnfs_true;
2230                 else
2231                         *tl = newnfs_false;
2232         }
2233         FREE((caddr_t)cookies, M_TEMP);
2234         FREE((caddr_t)rbuf, M_TEMP);
2235
2236 out:
2237         NFSEXITCODE2(0, nd);
2238         return (0);
2239 nfsmout:
2240         vput(vp);
2241         NFSEXITCODE2(error, nd);
2242         return (error);
2243 }
2244
2245 /*
2246  * Get the settable attributes out of the mbuf list.
2247  * (Return 0 or EBADRPC)
2248  */
2249 int
2250 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2251     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2252 {
2253         u_int32_t *tl;
2254         struct nfsv2_sattr *sp;
2255         int error = 0, toclient = 0;
2256
2257         switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2258         case ND_NFSV2:
2259                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2260                 /*
2261                  * Some old clients didn't fill in the high order 16bits.
2262                  * --> check the low order 2 bytes for 0xffff
2263                  */
2264                 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2265                         nvap->na_mode = nfstov_mode(sp->sa_mode);
2266                 if (sp->sa_uid != newnfs_xdrneg1)
2267                         nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2268                 if (sp->sa_gid != newnfs_xdrneg1)
2269                         nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2270                 if (sp->sa_size != newnfs_xdrneg1)
2271                         nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2272                 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2273 #ifdef notyet
2274                         fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2275 #else
2276                         nvap->na_atime.tv_sec =
2277                                 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2278                         nvap->na_atime.tv_nsec = 0;
2279 #endif
2280                 }
2281                 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2282                         fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2283                 break;
2284         case ND_NFSV3:
2285                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2286                 if (*tl == newnfs_true) {
2287                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2288                         nvap->na_mode = nfstov_mode(*tl);
2289                 }
2290                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2291                 if (*tl == newnfs_true) {
2292                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2293                         nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2294                 }
2295                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2296                 if (*tl == newnfs_true) {
2297                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2298                         nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2299                 }
2300                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2301                 if (*tl == newnfs_true) {
2302                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2303                         nvap->na_size = fxdr_hyper(tl);
2304                 }
2305                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2306                 switch (fxdr_unsigned(int, *tl)) {
2307                 case NFSV3SATTRTIME_TOCLIENT:
2308                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2309                         fxdr_nfsv3time(tl, &nvap->na_atime);
2310                         toclient = 1;
2311                         break;
2312                 case NFSV3SATTRTIME_TOSERVER:
2313                         vfs_timestamp(&nvap->na_atime);
2314                         nvap->na_vaflags |= VA_UTIMES_NULL;
2315                         break;
2316                 };
2317                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2318                 switch (fxdr_unsigned(int, *tl)) {
2319                 case NFSV3SATTRTIME_TOCLIENT:
2320                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2321                         fxdr_nfsv3time(tl, &nvap->na_mtime);
2322                         nvap->na_vaflags &= ~VA_UTIMES_NULL;
2323                         break;
2324                 case NFSV3SATTRTIME_TOSERVER:
2325                         vfs_timestamp(&nvap->na_mtime);
2326                         if (!toclient)
2327                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2328                         break;
2329                 };
2330                 break;
2331         case ND_NFSV4:
2332                 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2333         };
2334 nfsmout:
2335         NFSEXITCODE2(error, nd);
2336         return (error);
2337 }
2338
2339 /*
2340  * Handle the setable attributes for V4.
2341  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2342  */
2343 int
2344 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2345     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2346 {
2347         u_int32_t *tl;
2348         int attrsum = 0;
2349         int i, j;
2350         int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2351         int toclient = 0;
2352         u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2353         uid_t uid;
2354         gid_t gid;
2355
2356         error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2357         if (error)
2358                 goto nfsmout;
2359         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2360         attrsize = fxdr_unsigned(int, *tl);
2361
2362         /*
2363          * Loop around getting the setable attributes. If an unsupported
2364          * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2365          */
2366         if (retnotsup) {
2367                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2368                 bitpos = NFSATTRBIT_MAX;
2369         } else {
2370                 bitpos = 0;
2371         }
2372         for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2373             if (attrsum > attrsize) {
2374                 error = NFSERR_BADXDR;
2375                 goto nfsmout;
2376             }
2377             if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2378                 switch (bitpos) {
2379                 case NFSATTRBIT_SIZE:
2380                         NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2381                         nvap->na_size = fxdr_hyper(tl);
2382                         attrsum += NFSX_HYPER;
2383                         break;
2384                 case NFSATTRBIT_ACL:
2385                         error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2386                             p);
2387                         if (error)
2388                                 goto nfsmout;
2389                         if (aceerr && !nd->nd_repstat)
2390                                 nd->nd_repstat = aceerr;
2391                         attrsum += aclsize;
2392                         break;
2393                 case NFSATTRBIT_ARCHIVE:
2394                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2395                         if (!nd->nd_repstat)
2396                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2397                         attrsum += NFSX_UNSIGNED;
2398                         break;
2399                 case NFSATTRBIT_HIDDEN:
2400                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2401                         if (!nd->nd_repstat)
2402                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2403                         attrsum += NFSX_UNSIGNED;
2404                         break;
2405                 case NFSATTRBIT_MIMETYPE:
2406                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2407                         i = fxdr_unsigned(int, *tl);
2408                         error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2409                         if (error)
2410                                 goto nfsmout;
2411                         if (!nd->nd_repstat)
2412                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2413                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2414                         break;
2415                 case NFSATTRBIT_MODE:
2416                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2417                         nvap->na_mode = nfstov_mode(*tl);
2418                         attrsum += NFSX_UNSIGNED;
2419                         break;
2420                 case NFSATTRBIT_OWNER:
2421                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2422                         j = fxdr_unsigned(int, *tl);
2423                         if (j < 0) {
2424                                 error = NFSERR_BADXDR;
2425                                 goto nfsmout;
2426                         }
2427                         if (j > NFSV4_SMALLSTR)
2428                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2429                         else
2430                                 cp = namestr;
2431                         error = nfsrv_mtostr(nd, cp, j);
2432                         if (error) {
2433                                 if (j > NFSV4_SMALLSTR)
2434                                         free(cp, M_NFSSTRING);
2435                                 goto nfsmout;
2436                         }
2437                         if (!nd->nd_repstat) {
2438                                 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2439                                     p);
2440                                 if (!nd->nd_repstat)
2441                                         nvap->na_uid = uid;
2442                         }
2443                         if (j > NFSV4_SMALLSTR)
2444                                 free(cp, M_NFSSTRING);
2445                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2446                         break;
2447                 case NFSATTRBIT_OWNERGROUP:
2448                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2449                         j = fxdr_unsigned(int, *tl);
2450                         if (j < 0) {
2451                                 error = NFSERR_BADXDR;
2452                                 goto nfsmout;
2453                         }
2454                         if (j > NFSV4_SMALLSTR)
2455                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2456                         else
2457                                 cp = namestr;
2458                         error = nfsrv_mtostr(nd, cp, j);
2459                         if (error) {
2460                                 if (j > NFSV4_SMALLSTR)
2461                                         free(cp, M_NFSSTRING);
2462                                 goto nfsmout;
2463                         }
2464                         if (!nd->nd_repstat) {
2465                                 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2466                                     p);
2467                                 if (!nd->nd_repstat)
2468                                         nvap->na_gid = gid;
2469                         }
2470                         if (j > NFSV4_SMALLSTR)
2471                                 free(cp, M_NFSSTRING);
2472                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2473                         break;
2474                 case NFSATTRBIT_SYSTEM:
2475                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2476                         if (!nd->nd_repstat)
2477                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2478                         attrsum += NFSX_UNSIGNED;
2479                         break;
2480                 case NFSATTRBIT_TIMEACCESSSET:
2481                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2482                         attrsum += NFSX_UNSIGNED;
2483                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2484                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2485                             fxdr_nfsv4time(tl, &nvap->na_atime);
2486                             toclient = 1;
2487                             attrsum += NFSX_V4TIME;
2488                         } else {
2489                             vfs_timestamp(&nvap->na_atime);
2490                             nvap->na_vaflags |= VA_UTIMES_NULL;
2491                         }
2492                         break;
2493                 case NFSATTRBIT_TIMEBACKUP:
2494                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2495                         if (!nd->nd_repstat)
2496                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2497                         attrsum += NFSX_V4TIME;
2498                         break;
2499                 case NFSATTRBIT_TIMECREATE:
2500                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2501                         if (!nd->nd_repstat)
2502                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2503                         attrsum += NFSX_V4TIME;
2504                         break;
2505                 case NFSATTRBIT_TIMEMODIFYSET:
2506                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2507                         attrsum += NFSX_UNSIGNED;
2508                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2509                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2510                             fxdr_nfsv4time(tl, &nvap->na_mtime);
2511                             nvap->na_vaflags &= ~VA_UTIMES_NULL;
2512                             attrsum += NFSX_V4TIME;
2513                         } else {
2514                             vfs_timestamp(&nvap->na_mtime);
2515                             if (!toclient)
2516                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2517                         }
2518                         break;
2519                 default:
2520                         nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2521                         /*
2522                          * set bitpos so we drop out of the loop.
2523                          */
2524                         bitpos = NFSATTRBIT_MAX;
2525                         break;
2526                 };
2527         }
2528
2529         /*
2530          * some clients pad the attrlist, so we need to skip over the
2531          * padding.
2532          */
2533         if (attrsum > attrsize) {
2534                 error = NFSERR_BADXDR;
2535         } else {
2536                 attrsize = NFSM_RNDUP(attrsize);
2537                 if (attrsum < attrsize)
2538                         error = nfsm_advance(nd, attrsize - attrsum, -1);
2539         }
2540 nfsmout:
2541         NFSEXITCODE2(error, nd);
2542         return (error);
2543 }
2544
2545 /*
2546  * Check/setup export credentials.
2547  */
2548 int
2549 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2550     struct ucred *credanon)
2551 {
2552         int error = 0;
2553
2554         /*
2555          * Check/setup credentials.
2556          */
2557         if (nd->nd_flag & ND_GSS)
2558                 exp->nes_exflag &= ~MNT_EXPORTANON;
2559
2560         /*
2561          * Check to see if the operation is allowed for this security flavor.
2562          * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2563          * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2564          * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2565          */
2566         if (nfsvno_testexp(nd, exp) &&
2567             nd->nd_procnum != NFSV4OP_SECINFO &&
2568             nd->nd_procnum != NFSPROC_FSINFO) {
2569                 if (nd->nd_flag & ND_NFSV4)
2570                         error = NFSERR_WRONGSEC;
2571                 else
2572                         error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2573                 goto out;
2574         }
2575
2576         /*
2577          * Check to see if the file system is exported V4 only.
2578          */
2579         if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2580                 error = NFSERR_PROGNOTV4;
2581                 goto out;
2582         }
2583
2584         /*
2585          * Now, map the user credentials.
2586          * (Note that ND_AUTHNONE will only be set for an NFSv3
2587          *  Fsinfo RPC. If set for anything else, this code might need
2588          *  to change.)
2589          */
2590         if (NFSVNO_EXPORTED(exp) &&
2591             ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2592              NFSVNO_EXPORTANON(exp) ||
2593              (nd->nd_flag & ND_AUTHNONE))) {
2594                 nd->nd_cred->cr_uid = credanon->cr_uid;
2595                 nd->nd_cred->cr_gid = credanon->cr_gid;
2596                 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2597                     credanon->cr_groups);
2598         }
2599
2600 out:
2601         NFSEXITCODE2(error, nd);
2602         return (error);
2603 }
2604
2605 /*
2606  * Check exports.
2607  */
2608 int
2609 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2610     struct ucred **credp)
2611 {
2612         int i, error, *secflavors;
2613
2614         error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2615             &exp->nes_numsecflavor, &secflavors);
2616         if (error) {
2617                 if (nfs_rootfhset) {
2618                         exp->nes_exflag = 0;
2619                         exp->nes_numsecflavor = 0;
2620                         error = 0;
2621                 }
2622         } else {
2623                 /* Copy the security flavors. */
2624                 for (i = 0; i < exp->nes_numsecflavor; i++)
2625                         exp->nes_secflavors[i] = secflavors[i];
2626         }
2627         NFSEXITCODE(error);
2628         return (error);
2629 }
2630
2631 /*
2632  * Get a vnode for a file handle and export stuff.
2633  */
2634 int
2635 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2636     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2637     struct ucred **credp)
2638 {
2639         int i, error, *secflavors;
2640
2641         *credp = NULL;
2642         exp->nes_numsecflavor = 0;
2643         if (VFS_NEEDSGIANT(mp))
2644                 error = ESTALE;
2645         else
2646                 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, vpp);
2647         if (error != 0)
2648                 /* Make sure the server replies ESTALE to the client. */
2649                 error = ESTALE;
2650         if (nam && !error) {
2651                 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2652                     &exp->nes_numsecflavor, &secflavors);
2653                 if (error) {
2654                         if (nfs_rootfhset) {
2655                                 exp->nes_exflag = 0;
2656                                 exp->nes_numsecflavor = 0;
2657                                 error = 0;
2658                         } else {
2659                                 vput(*vpp);
2660                         }
2661                 } else {
2662                         /* Copy the security flavors. */
2663                         for (i = 0; i < exp->nes_numsecflavor; i++)
2664                                 exp->nes_secflavors[i] = secflavors[i];
2665                 }
2666         }
2667         if (error == 0 && lktype == LK_SHARED)
2668                 /*
2669                  * It would be much better to pass lktype to VFS_FHTOVP(),
2670                  * but this will have to do until VFS_FHTOVP() has a lock
2671                  * type argument like VFS_VGET().
2672                  */
2673                 NFSVOPLOCK(*vpp, LK_DOWNGRADE | LK_RETRY);
2674
2675         NFSEXITCODE(error);
2676         return (error);
2677 }
2678
2679 /*
2680  * nfsd_fhtovp() - convert a fh to a vnode ptr
2681  *      - look up fsid in mount list (if not found ret error)
2682  *      - get vp and export rights by calling nfsvno_fhtovp()
2683  *      - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2684  *        for AUTH_SYS
2685  *      - if mpp != NULL, return the mount point so that it can
2686  *        be used for vn_finished_write() by the caller
2687  */
2688 void
2689 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2690     struct vnode **vpp, struct nfsexstuff *exp,
2691     struct mount **mpp, int startwrite, struct thread *p)
2692 {
2693         struct mount *mp;
2694         struct ucred *credanon;
2695         fhandle_t *fhp;
2696
2697         fhp = (fhandle_t *)nfp->nfsrvfh_data;
2698         /*
2699          * Check for the special case of the nfsv4root_fh.
2700          */
2701         mp = vfs_busyfs(&fhp->fh_fsid);
2702         if (mpp != NULL)
2703                 *mpp = mp;
2704         if (mp == NULL) {
2705                 *vpp = NULL;
2706                 nd->nd_repstat = ESTALE;
2707                 goto out;
2708         }
2709
2710         if (startwrite) {
2711                 vn_start_write(NULL, mpp, V_WAIT);
2712                 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2713                         lktype = LK_EXCLUSIVE;
2714         }
2715         nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2716             &credanon);
2717         vfs_unbusy(mp);
2718
2719         /*
2720          * For NFSv4 without a pseudo root fs, unexported file handles
2721          * can be returned, so that Lookup works everywhere.
2722          */
2723         if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2724             !(nd->nd_flag & ND_NFSV4)) {
2725                 vput(*vpp);
2726                 nd->nd_repstat = EACCES;
2727         }
2728
2729         /*
2730          * Personally, I've never seen any point in requiring a
2731          * reserved port#, since only in the rare case where the
2732          * clients are all boxes with secure system priviledges,
2733          * does it provide any enhanced security, but... some people
2734          * believe it to be useful and keep putting this code back in.
2735          * (There is also some "security checker" out there that
2736          *  complains if the nfs server doesn't enforce this.)
2737          * However, note the following:
2738          * RFC3530 (NFSv4) specifies that a reserved port# not be
2739          *      required.
2740          * RFC2623 recommends that, if a reserved port# is checked for,
2741          *      that there be a way to turn that off--> ifdef'd.
2742          */
2743 #ifdef NFS_REQRSVPORT
2744         if (!nd->nd_repstat) {
2745                 struct sockaddr_in *saddr;
2746                 struct sockaddr_in6 *saddr6;
2747
2748                 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2749                 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2750                 if (!(nd->nd_flag & ND_NFSV4) &&
2751                     ((saddr->sin_family == AF_INET &&
2752                       ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2753                      (saddr6->sin6_family == AF_INET6 &&
2754                       ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2755                         vput(*vpp);
2756                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2757                 }
2758         }
2759 #endif  /* NFS_REQRSVPORT */
2760
2761         /*
2762          * Check/setup credentials.
2763          */
2764         if (!nd->nd_repstat) {
2765                 nd->nd_saveduid = nd->nd_cred->cr_uid;
2766                 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2767                 if (nd->nd_repstat)
2768                         vput(*vpp);
2769         }
2770         if (credanon != NULL)
2771                 crfree(credanon);
2772         if (nd->nd_repstat) {
2773                 if (startwrite)
2774                         vn_finished_write(mp);
2775                 *vpp = NULL;
2776                 if (mpp != NULL)
2777                         *mpp = NULL;
2778         }
2779
2780 out:
2781         NFSEXITCODE2(0, nd);
2782 }
2783
2784 /*
2785  * glue for fp.
2786  */
2787 int
2788 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2789 {
2790         struct filedesc *fdp;
2791         struct file *fp;
2792         int error = 0;
2793
2794         fdp = p->td_proc->p_fd;
2795         if (fd >= fdp->fd_nfiles ||
2796             (fp = fdp->fd_ofiles[fd]) == NULL) {
2797                 error = EBADF;
2798                 goto out;
2799         }
2800         *fpp = fp;
2801
2802 out:
2803         NFSEXITCODE(error);
2804         return (error);
2805 }
2806
2807 /*
2808  * Called from nfssvc() to update the exports list. Just call
2809  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2810  * in the mount list.
2811  */
2812 int
2813 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2814 {
2815         struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2816         int error = 0;
2817         struct nameidata nd;
2818         fhandle_t fh;
2819
2820         error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2821         if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2822                 nfs_rootfhset = 0;
2823         else if (error == 0) {
2824                 if (nfsexargp->fspec == NULL) {
2825                         error = EPERM;
2826                         goto out;
2827                 }
2828                 /*
2829                  * If fspec != NULL, this is the v4root path.
2830                  */
2831                 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2832                     nfsexargp->fspec, p);
2833                 if ((error = namei(&nd)) != 0)
2834                         goto out;
2835                 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2836                 vrele(nd.ni_vp);
2837                 if (!error) {
2838                         nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2839                         NFSBCOPY((caddr_t)&fh,
2840                             nfs_rootfh.nfsrvfh_data,
2841                             sizeof (fhandle_t));
2842                         nfs_rootfhset = 1;
2843                 }
2844         }
2845
2846 out:
2847         NFSEXITCODE(error);
2848         return (error);
2849 }
2850
2851 /*
2852  * Get the tcp socket sequence numbers we need.
2853  * (Maybe this should be moved to the tcp sources?)
2854  */
2855 int
2856 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2857 {
2858         struct inpcb *inp;
2859         struct tcpcb *tp;
2860         int error = 0;
2861
2862         inp = sotoinpcb(so);
2863         KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2864         INP_RLOCK(inp);
2865         if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2866                 INP_RUNLOCK(inp);
2867                 error = EPIPE;
2868                 goto out;
2869         }
2870         tp = intotcpcb(inp);
2871         if (tp->t_state != TCPS_ESTABLISHED) {
2872                 INP_RUNLOCK(inp);
2873                 error = EPIPE;
2874                 goto out;
2875         }
2876         *maxp = tp->snd_max;
2877         *unap = tp->snd_una;
2878         INP_RUNLOCK(inp);
2879
2880 out:
2881         NFSEXITCODE(error);
2882         return (error);
2883 }
2884
2885 /*
2886  * This function needs to test to see if the system is near its limit
2887  * for memory allocation via malloc() or mget() and return True iff
2888  * either of these resources are near their limit.
2889  * XXX (For now, this is just a stub.)
2890  */
2891 int nfsrv_testmalloclimit = 0;
2892 int
2893 nfsrv_mallocmget_limit(void)
2894 {
2895         static int printmesg = 0;
2896         static int testval = 1;
2897
2898         if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2899                 if ((printmesg++ % 100) == 0)
2900                         printf("nfsd: malloc/mget near limit\n");
2901                 return (1);
2902         }
2903         return (0);
2904 }
2905
2906 /*
2907  * BSD specific initialization of a mount point.
2908  */
2909 void
2910 nfsd_mntinit(void)
2911 {
2912         static int inited = 0;
2913
2914         if (inited)
2915                 return;
2916         inited = 1;
2917         nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2918         TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2919         TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2920         nfsv4root_mnt.mnt_export = NULL;
2921         TAILQ_INIT(&nfsv4root_opt);
2922         TAILQ_INIT(&nfsv4root_newopt);
2923         nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2924         nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2925         nfsv4root_mnt.mnt_nvnodelistsize = 0;
2926         nfsv4root_mnt.mnt_activevnodelistsize = 0;
2927 }
2928
2929 /*
2930  * Get a vnode for a file handle, without checking exports, etc.
2931  */
2932 struct vnode *
2933 nfsvno_getvp(fhandle_t *fhp)
2934 {
2935         struct mount *mp;
2936         struct vnode *vp;
2937         int error;
2938
2939         mp = vfs_busyfs(&fhp->fh_fsid);
2940         if (mp == NULL)
2941                 return (NULL);
2942         error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2943         vfs_unbusy(mp);
2944         if (error)
2945                 return (NULL);
2946         return (vp);
2947 }
2948
2949 /*
2950  * Do a local VOP_ADVLOCK().
2951  */
2952 int
2953 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2954     u_int64_t end, struct thread *td)
2955 {
2956         int error = 0;
2957         struct flock fl;
2958         u_int64_t tlen;
2959
2960         if (nfsrv_dolocallocks == 0)
2961                 goto out;
2962
2963         /* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2964         if ((vp->v_iflag & VI_DOOMED) != 0) {
2965                 error = EPERM;
2966                 goto out;
2967         }
2968
2969         fl.l_whence = SEEK_SET;
2970         fl.l_type = ftype;
2971         fl.l_start = (off_t)first;
2972         if (end == NFS64BITSSET) {
2973                 fl.l_len = 0;
2974         } else {
2975                 tlen = end - first;
2976                 fl.l_len = (off_t)tlen;
2977         }
2978         /*
2979          * For FreeBSD8, the l_pid and l_sysid must be set to the same
2980          * values for all calls, so that all locks will be held by the
2981          * nfsd server. (The nfsd server handles conflicts between the
2982          * various clients.)
2983          * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2984          * bytes, so it can't be put in l_sysid.
2985          */
2986         if (nfsv4_sysid == 0)
2987                 nfsv4_sysid = nlm_acquire_next_sysid();
2988         fl.l_pid = (pid_t)0;
2989         fl.l_sysid = (int)nfsv4_sysid;
2990
2991         NFSVOPUNLOCK(vp, 0);
2992         if (ftype == F_UNLCK)
2993                 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2994                     (F_POSIX | F_REMOTE));
2995         else
2996                 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2997                     (F_POSIX | F_REMOTE));
2998         NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2999
3000 out:
3001         NFSEXITCODE(error);
3002         return (error);
3003 }
3004
3005 /*
3006  * Check the nfsv4 root exports.
3007  */
3008 int
3009 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3010 {
3011         struct ucred *credanon;
3012         int exflags, error = 0, numsecflavor, *secflavors, i;
3013
3014         error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3015             &credanon, &numsecflavor, &secflavors);
3016         if (error) {
3017                 error = NFSERR_PROGUNAVAIL;
3018                 goto out;
3019         }
3020         if (credanon != NULL)
3021                 crfree(credanon);
3022         for (i = 0; i < numsecflavor; i++) {
3023                 if (secflavors[i] == AUTH_SYS)
3024                         nd->nd_flag |= ND_EXAUTHSYS;
3025                 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3026                         nd->nd_flag |= ND_EXGSS;
3027                 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3028                         nd->nd_flag |= ND_EXGSSINTEGRITY;
3029                 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3030                         nd->nd_flag |= ND_EXGSSPRIVACY;
3031         }
3032
3033 out:
3034         NFSEXITCODE(error);
3035         return (error);
3036 }
3037
3038 /*
3039  * Nfs server psuedo system call for the nfsd's
3040  */
3041 /*
3042  * MPSAFE
3043  */
3044 static int
3045 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3046 {
3047         struct file *fp;
3048         struct nfsd_addsock_args sockarg;
3049         struct nfsd_nfsd_args nfsdarg;
3050         int error;
3051
3052         if (uap->flag & NFSSVC_NFSDADDSOCK) {
3053                 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3054                 if (error)
3055                         goto out;
3056                 /*
3057                  * Since we don't know what rights might be required,
3058                  * pretend that we need them all. It is better to be too
3059                  * careful than too reckless.
3060                  */
3061                 if ((error = fget(td, sockarg.sock, CAP_SOCK_ALL, &fp)) != 0)
3062                         goto out;
3063                 if (fp->f_type != DTYPE_SOCKET) {
3064                         fdrop(fp, td);
3065                         error = EPERM;
3066                         goto out;
3067                 }
3068                 error = nfsrvd_addsock(fp);
3069                 fdrop(fp, td);
3070         } else if (uap->flag & NFSSVC_NFSDNFSD) {
3071                 if (uap->argp == NULL) {
3072                         error = EINVAL;
3073                         goto out;
3074                 }
3075                 error = copyin(uap->argp, (caddr_t)&nfsdarg,
3076                     sizeof (nfsdarg));
3077                 if (error)
3078                         goto out;
3079                 error = nfsrvd_nfsd(td, &nfsdarg);
3080         } else {
3081                 error = nfssvc_srvcall(td, uap, td->td_ucred);
3082         }
3083
3084 out:
3085         NFSEXITCODE(error);
3086         return (error);
3087 }
3088
3089 static int
3090 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3091 {
3092         struct nfsex_args export;
3093         struct file *fp = NULL;
3094         int stablefd, len;
3095         struct nfsd_clid adminrevoke;
3096         struct nfsd_dumplist dumplist;
3097         struct nfsd_dumpclients *dumpclients;
3098         struct nfsd_dumplocklist dumplocklist;
3099         struct nfsd_dumplocks *dumplocks;
3100         struct nameidata nd;
3101         vnode_t vp;
3102         int error = EINVAL, igotlock;
3103         struct proc *procp;
3104         static int suspend_nfsd = 0;
3105
3106         if (uap->flag & NFSSVC_PUBLICFH) {
3107                 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3108                     sizeof (fhandle_t));
3109                 error = copyin(uap->argp,
3110                     &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3111                 if (!error)
3112                         nfs_pubfhset = 1;
3113         } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3114                 error = copyin(uap->argp,(caddr_t)&export,
3115                     sizeof (struct nfsex_args));
3116                 if (!error)
3117                         error = nfsrv_v4rootexport(&export, cred, p);
3118         } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3119                 nfs_pubfhset = 0;
3120                 error = 0;
3121         } else if (uap->flag & NFSSVC_STABLERESTART) {
3122                 error = copyin(uap->argp, (caddr_t)&stablefd,
3123                     sizeof (int));
3124                 if (!error)
3125                         error = fp_getfvp(p, stablefd, &fp, &vp);
3126                 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3127                         error = EBADF;
3128                 if (!error && newnfs_numnfsd != 0)
3129                         error = EPERM;
3130                 if (!error) {
3131                         nfsrv_stablefirst.nsf_fp = fp;
3132                         nfsrv_setupstable(p);
3133                 }
3134         } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3135                 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3136                     sizeof (struct nfsd_clid));
3137                 if (!error)
3138                         error = nfsrv_adminrevoke(&adminrevoke, p);
3139         } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3140                 error = copyin(uap->argp, (caddr_t)&dumplist,
3141                     sizeof (struct nfsd_dumplist));
3142                 if (!error && (dumplist.ndl_size < 1 ||
3143                         dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3144                         error = EPERM;
3145                 if (!error) {
3146                     len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3147                     dumpclients = (struct nfsd_dumpclients *)malloc(len,
3148                         M_TEMP, M_WAITOK);
3149                     nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3150                     error = copyout(dumpclients,
3151                         CAST_USER_ADDR_T(dumplist.ndl_list), len);
3152                     free((caddr_t)dumpclients, M_TEMP);
3153                 }
3154         } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3155                 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3156                     sizeof (struct nfsd_dumplocklist));
3157                 if (!error && (dumplocklist.ndllck_size < 1 ||
3158                         dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3159                         error = EPERM;
3160                 if (!error)
3161                         error = nfsrv_lookupfilename(&nd,
3162                                 dumplocklist.ndllck_fname, p);
3163                 if (!error) {
3164                         len = sizeof (struct nfsd_dumplocks) *
3165                                 dumplocklist.ndllck_size;
3166                         dumplocks = (struct nfsd_dumplocks *)malloc(len,
3167                                 M_TEMP, M_WAITOK);
3168                         nfsrv_dumplocks(nd.ni_vp, dumplocks,
3169                             dumplocklist.ndllck_size, p);
3170                         vput(nd.ni_vp);
3171                         error = copyout(dumplocks,
3172                             CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3173                         free((caddr_t)dumplocks, M_TEMP);
3174                 }
3175         } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3176                 procp = p->td_proc;
3177                 PROC_LOCK(procp);
3178                 nfsd_master_pid = procp->p_pid;
3179                 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3180                 nfsd_master_start = procp->p_stats->p_start;
3181                 nfsd_master_proc = procp;
3182                 PROC_UNLOCK(procp);
3183         } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3184                 NFSLOCKV4ROOTMUTEX();
3185                 if (suspend_nfsd == 0) {
3186                         /* Lock out all nfsd threads */
3187                         do {
3188                                 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3189                                     NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3190                         } while (igotlock == 0 && suspend_nfsd == 0);
3191                         suspend_nfsd = 1;
3192                 }
3193                 NFSUNLOCKV4ROOTMUTEX();
3194                 error = 0;
3195         } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3196                 NFSLOCKV4ROOTMUTEX();
3197                 if (suspend_nfsd != 0) {
3198                         nfsv4_unlock(&nfsd_suspend_lock, 0);
3199                         suspend_nfsd = 0;
3200                 }
3201                 NFSUNLOCKV4ROOTMUTEX();
3202                 error = 0;
3203         }
3204
3205         NFSEXITCODE(error);
3206         return (error);
3207 }
3208
3209 /*
3210  * Check exports.
3211  * Returns 0 if ok, 1 otherwise.
3212  */
3213 int
3214 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3215 {
3216         int i;
3217
3218         /*
3219          * This seems odd, but allow the case where the security flavor
3220          * list is empty. This happens when NFSv4 is traversing non-exported
3221          * file systems. Exported file systems should always have a non-empty
3222          * security flavor list.
3223          */
3224         if (exp->nes_numsecflavor == 0)
3225                 return (0);
3226
3227         for (i = 0; i < exp->nes_numsecflavor; i++) {
3228                 /*
3229                  * The tests for privacy and integrity must be first,
3230                  * since ND_GSS is set for everything but AUTH_SYS.
3231                  */
3232                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3233                     (nd->nd_flag & ND_GSSPRIVACY))
3234                         return (0);
3235                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3236                     (nd->nd_flag & ND_GSSINTEGRITY))
3237                         return (0);
3238                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3239                     (nd->nd_flag & ND_GSS))
3240                         return (0);
3241                 if (exp->nes_secflavors[i] == AUTH_SYS &&
3242                     (nd->nd_flag & ND_GSS) == 0)
3243                         return (0);
3244         }
3245         return (1);
3246 }
3247
3248 /*
3249  * Calculate a hash value for the fid in a file handle.
3250  */
3251 uint32_t
3252 nfsrv_hashfh(fhandle_t *fhp)
3253 {
3254         uint32_t hashval;
3255
3256         hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3257         return (hashval);
3258 }
3259
3260 /*
3261  * Signal the userland master nfsd to backup the stable restart file.
3262  */
3263 void
3264 nfsrv_backupstable(void)
3265 {
3266         struct proc *procp;
3267
3268         if (nfsd_master_proc != NULL) {
3269                 procp = pfind(nfsd_master_pid);
3270                 /* Try to make sure it is the correct process. */
3271                 if (procp == nfsd_master_proc &&
3272                     procp->p_stats->p_start.tv_sec ==
3273                     nfsd_master_start.tv_sec &&
3274                     procp->p_stats->p_start.tv_usec ==
3275                     nfsd_master_start.tv_usec &&
3276                     strcmp(procp->p_comm, nfsd_master_comm) == 0)
3277                         kern_psignal(procp, SIGUSR2);
3278                 else
3279                         nfsd_master_proc = NULL;
3280
3281                 if (procp != NULL)
3282                         PROC_UNLOCK(procp);
3283         }
3284 }
3285
3286 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3287
3288 /*
3289  * Called once to initialize data structures...
3290  */
3291 static int
3292 nfsd_modevent(module_t mod, int type, void *data)
3293 {
3294         int error = 0;
3295         static int loaded = 0;
3296
3297         switch (type) {
3298         case MOD_LOAD:
3299                 if (loaded)
3300                         goto out;
3301                 newnfs_portinit();
3302                 mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3303                 mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3304                 mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3305                     MTX_DEF);
3306                 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3307                 nfsrvd_initcache();
3308                 nfsd_init();
3309                 NFSD_LOCK();
3310                 nfsrvd_init(0);
3311                 NFSD_UNLOCK();
3312                 nfsd_mntinit();
3313 #ifdef VV_DISABLEDELEG
3314                 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3315                 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3316 #endif
3317                 nfsd_call_servertimer = nfsrv_servertimer;
3318                 nfsd_call_nfsd = nfssvc_nfsd;
3319                 loaded = 1;
3320                 break;
3321
3322         case MOD_UNLOAD:
3323                 if (newnfs_numnfsd != 0) {
3324                         error = EBUSY;
3325                         break;
3326                 }
3327
3328 #ifdef VV_DISABLEDELEG
3329                 vn_deleg_ops.vndeleg_recall = NULL;
3330                 vn_deleg_ops.vndeleg_disable = NULL;
3331 #endif
3332                 nfsd_call_servertimer = NULL;
3333                 nfsd_call_nfsd = NULL;
3334
3335                 /* Clean out all NFSv4 state. */
3336                 nfsrv_throwawayallstate(curthread);
3337
3338                 /* Clean the NFS server reply cache */
3339                 nfsrvd_cleancache();
3340
3341                 /* Free up the krpc server pool. */
3342                 if (nfsrvd_pool != NULL)
3343                         svcpool_destroy(nfsrvd_pool);
3344
3345                 /* and get rid of the locks */
3346                 mtx_destroy(&nfs_cache_mutex);
3347                 mtx_destroy(&nfs_v4root_mutex);
3348                 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3349                 lockdestroy(&nfsv4root_mnt.mnt_explock);
3350                 loaded = 0;
3351                 break;
3352         default:
3353                 error = EOPNOTSUPP;
3354                 break;
3355         }
3356
3357 out:
3358         NFSEXITCODE(error);
3359         return (error);
3360 }
3361 static moduledata_t nfsd_mod = {
3362         "nfsd",
3363         nfsd_modevent,
3364         NULL,
3365 };
3366 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3367
3368 /* So that loader and kldload(2) can find us, wherever we are.. */
3369 MODULE_VERSION(nfsd, 1);
3370 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3371 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3372 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3373 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3374 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3375