sys/fs/nfsserver/nfs_nfsdport.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1989, 1993
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * This code is derived from software contributed to Berkeley by
   8  * Rick Macklem at The University of Guelph.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  * 3. Neither the name of the University nor the names of its contributors
  19  *    may be used to endorse or promote products derived from this software
  20  *    without specific prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  */
  35
  36 #include <sys/cdefs.h>
  37 __FBSDID("$FreeBSD$");
  38
  39 #include <sys/capsicum.h>
  40
  41 /*
  42  * Functions that perform the vfs operations required by the routines in
  43  * nfsd_serv.c. It is hoped that this change will make the server more
  44  * portable.
  45  */
  46
  47 #include <fs/nfs/nfsport.h>
  48 #include <sys/hash.h>
  49 #include <sys/sysctl.h>
  50 #include <nlm/nlm_prot.h>
  51 #include <nlm/nlm.h>
  52
  53 FEATURE(nfsd, "NFSv4 server");
  54
  55 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
  56 extern int nfsrv_useacl;
  57 extern int newnfs_numnfsd;
  58 extern struct mount nfsv4root_mnt;
  59 extern struct nfsrv_stablefirst nfsrv_stablefirst;
  60 extern void (*nfsd_call_servertimer)(void);
  61 extern SVCPOOL  *nfsrvd_pool;
  62 extern struct nfsv4lock nfsd_suspend_lock;
  63 extern struct nfsclienthashhead *nfsclienthash;
  64 extern struct nfslockhashhead *nfslockhash;
  65 extern struct nfssessionhash *nfssessionhash;
  66 extern int nfsrv_sessionhashsize;
  67 extern struct nfsstatsv1 nfsstatsv1;
  68 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
  69 NFSDLOCKMUTEX;
  70 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
  71 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
  72 struct mtx nfsrc_udpmtx;
  73 struct mtx nfs_v4root_mutex;
  74 struct nfsrvfh nfs_rootfh, nfs_pubfh;
  75 int nfs_pubfhset = 0, nfs_rootfhset = 0;
  76 struct proc *nfsd_master_proc = NULL;
  77 int nfsd_debuglevel = 0;
  78 static pid_t nfsd_master_pid = (pid_t)-1;
  79 static char nfsd_master_comm[MAXCOMLEN + 1];
  80 static struct timeval nfsd_master_start;
  81 static uint32_t nfsv4_sysid = 0;
  82
  83 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
  84     struct ucred *);
  85
  86 int nfsrv_enable_crossmntpt = 1;
  87 static int nfs_commit_blks;
  88 static int nfs_commit_miss;
  89 extern int nfsrv_issuedelegs;
  90 extern int nfsrv_dolocallocks;
  91 extern int nfsd_enable_stringtouid;
  92
  93 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "NFS server");
  94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
  95     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
  96 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
  97     0, "");
  98 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
  99     0, "");
 100 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
 101     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
 102 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
 103     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
 104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
 105     0, "Debug level for NFS server");
 106 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
 107     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
 108
 109 #define MAX_REORDERED_RPC       16
 110 #define NUM_HEURISTIC           1031
 111 #define NHUSE_INIT              64
 112 #define NHUSE_INC               16
 113 #define NHUSE_MAX               2048
 114
 115 static struct nfsheur {
 116         struct vnode *nh_vp;    /* vp to match (unreferenced pointer) */
 117         off_t nh_nextoff;       /* next offset for sequential detection */
 118         int nh_use;             /* use count for selection */
 119         int nh_seqcount;        /* heuristic */
 120 } nfsheur[NUM_HEURISTIC];
 121
 122
 123 /*
 124  * Heuristic to detect sequential operation.
 125  */
 126 static struct nfsheur *
 127 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
 128 {
 129         struct nfsheur *nh;
 130         int hi, try;
 131
 132         /* Locate best candidate. */
 133         try = 32;
 134         hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
 135         nh = &nfsheur[hi];
 136         while (try--) {
 137                 if (nfsheur[hi].nh_vp == vp) {
 138                         nh = &nfsheur[hi];
 139                         break;
 140                 }
 141                 if (nfsheur[hi].nh_use > 0)
 142                         --nfsheur[hi].nh_use;
 143                 hi = (hi + 1) % NUM_HEURISTIC;
 144                 if (nfsheur[hi].nh_use < nh->nh_use)
 145                         nh = &nfsheur[hi];
 146         }
 147
 148         /* Initialize hint if this is a new file. */
 149         if (nh->nh_vp != vp) {
 150                 nh->nh_vp = vp;
 151                 nh->nh_nextoff = uio->uio_offset;
 152                 nh->nh_use = NHUSE_INIT;
 153                 if (uio->uio_offset == 0)
 154                         nh->nh_seqcount = 4;
 155                 else
 156                         nh->nh_seqcount = 1;
 157         }
 158
 159         /* Calculate heuristic. */
 160         if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
 161             uio->uio_offset == nh->nh_nextoff) {
 162                 /* See comments in vfs_vnops.c:sequential_heuristic(). */
 163                 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
 164                 if (nh->nh_seqcount > IO_SEQMAX)
 165                         nh->nh_seqcount = IO_SEQMAX;
 166         } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
 167             imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
 168                 /* Probably a reordered RPC, leave seqcount alone. */
 169         } else if (nh->nh_seqcount > 1) {
 170                 nh->nh_seqcount /= 2;
 171         } else {
 172                 nh->nh_seqcount = 0;
 173         }
 174         nh->nh_use += NHUSE_INC;
 175         if (nh->nh_use > NHUSE_MAX)
 176                 nh->nh_use = NHUSE_MAX;
 177         return (nh);
 178 }
 179
 180 /*
 181  * Get attributes into nfsvattr structure.
 182  */
 183 int
 184 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 185     struct thread *p, int vpislocked)
 186 {
 187         int error, lockedit = 0;
 188
 189         if (vpislocked == 0) {
 190                 /*
 191                  * When vpislocked == 0, the vnode is either exclusively
 192                  * locked by this thread or not locked by this thread.
 193                  * As such, shared lock it, if not exclusively locked.
 194                  */
 195                 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
 196                         lockedit = 1;
 197                         NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
 198                 }
 199         }
 200         error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
 201         if (lockedit != 0)
 202                 NFSVOPUNLOCK(vp, 0);
 203
 204         NFSEXITCODE(error);
 205         return (error);
 206 }
 207
 208 /*
 209  * Get a file handle for a vnode.
 210  */
 211 int
 212 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
 213 {
 214         int error;
 215
 216         NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
 217         fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
 218         error = VOP_VPTOFH(vp, &fhp->fh_fid);
 219
 220         NFSEXITCODE(error);
 221         return (error);
 222 }
 223
 224 /*
 225  * Perform access checking for vnodes obtained from file handles that would
 226  * refer to files already opened by a Unix client. You cannot just use
 227  * vn_writechk() and VOP_ACCESSX() for two reasons.
 228  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
 229  *     case.
 230  * 2 - The owner is to be given access irrespective of mode bits for some
 231  *     operations, so that processes that chmod after opening a file don't
 232  *     break.
 233  */
 234 int
 235 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
 236     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
 237     u_int32_t *supportedtypep)
 238 {
 239         struct vattr vattr;
 240         int error = 0, getret = 0;
 241
 242         if (vpislocked == 0) {
 243                 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
 244                         error = EPERM;
 245                         goto out;
 246                 }
 247         }
 248         if (accmode & VWRITE) {
 249                 /* Just vn_writechk() changed to check rdonly */
 250                 /*
 251                  * Disallow write attempts on read-only file systems;
 252                  * unless the file is a socket or a block or character
 253                  * device resident on the file system.
 254                  */
 255                 if (NFSVNO_EXRDONLY(exp) ||
 256                     (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 257                         switch (vp->v_type) {
 258                         case VREG:
 259                         case VDIR:
 260                         case VLNK:
 261                                 error = EROFS;
 262                         default:
 263                                 break;
 264                         }
 265                 }
 266                 /*
 267                  * If there's shared text associated with
 268                  * the inode, try to free it up once.  If
 269                  * we fail, we can't allow writing.
 270                  */
 271                 if (VOP_IS_TEXT(vp) && error == 0)
 272                         error = ETXTBSY;
 273         }
 274         if (error != 0) {
 275                 if (vpislocked == 0)
 276                         NFSVOPUNLOCK(vp, 0);
 277                 goto out;
 278         }
 279
 280         /*
 281          * Should the override still be applied when ACLs are enabled?
 282          */
 283         error = VOP_ACCESSX(vp, accmode, cred, p);
 284         if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
 285                 /*
 286                  * Try again with VEXPLICIT_DENY, to see if the test for
 287                  * deletion is supported.
 288                  */
 289                 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
 290                 if (error == 0) {
 291                         if (vp->v_type == VDIR) {
 292                                 accmode &= ~(VDELETE | VDELETE_CHILD);
 293                                 accmode |= VWRITE;
 294                                 error = VOP_ACCESSX(vp, accmode, cred, p);
 295                         } else if (supportedtypep != NULL) {
 296                                 *supportedtypep &= ~NFSACCESS_DELETE;
 297                         }
 298                 }
 299         }
 300
 301         /*
 302          * Allow certain operations for the owner (reads and writes
 303          * on files that are already open).
 304          */
 305         if (override != NFSACCCHK_NOOVERRIDE &&
 306             (error == EPERM || error == EACCES)) {
 307                 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
 308                         error = 0;
 309                 else if (override & NFSACCCHK_ALLOWOWNER) {
 310                         getret = VOP_GETATTR(vp, &vattr, cred);
 311                         if (getret == 0 && cred->cr_uid == vattr.va_uid)
 312                                 error = 0;
 313                 }
 314         }
 315         if (vpislocked == 0)
 316                 NFSVOPUNLOCK(vp, 0);
 317
 318 out:
 319         NFSEXITCODE(error);
 320         return (error);
 321 }
 322
 323 /*
 324  * Set attribute(s) vnop.
 325  */
 326 int
 327 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
 328     struct thread *p, struct nfsexstuff *exp)
 329 {
 330         int error;
 331
 332         error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
 333         NFSEXITCODE(error);
 334         return (error);
 335 }
 336
 337 /*
 338  * Set up nameidata for a lookup() call and do it.
 339  */
 340 int
 341 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
 342     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
 343     struct vnode **retdirp)
 344 {
 345         struct componentname *cnp = &ndp->ni_cnd;
 346         int i;
 347         struct iovec aiov;
 348         struct uio auio;
 349         int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
 350         int error = 0;
 351         char *cp;
 352
 353         *retdirp = NULL;
 354         cnp->cn_nameptr = cnp->cn_pnbuf;
 355         ndp->ni_lcf = 0;
 356         /*
 357          * Extract and set starting directory.
 358          */
 359         if (dp->v_type != VDIR) {
 360                 if (islocked)
 361                         vput(dp);
 362                 else
 363                         vrele(dp);
 364                 nfsvno_relpathbuf(ndp);
 365                 error = ENOTDIR;
 366                 goto out1;
 367         }
 368         if (islocked)
 369                 NFSVOPUNLOCK(dp, 0);
 370         VREF(dp);
 371         *retdirp = dp;
 372         if (NFSVNO_EXRDONLY(exp))
 373                 cnp->cn_flags |= RDONLY;
 374         ndp->ni_segflg = UIO_SYSSPACE;
 375
 376         if (nd->nd_flag & ND_PUBLOOKUP) {
 377                 ndp->ni_loopcnt = 0;
 378                 if (cnp->cn_pnbuf[0] == '/') {
 379                         vrele(dp);
 380                         /*
 381                          * Check for degenerate pathnames here, since lookup()
 382                          * panics on them.
 383                          */
 384                         for (i = 1; i < ndp->ni_pathlen; i++)
 385                                 if (cnp->cn_pnbuf[i] != '/')
 386                                         break;
 387                         if (i == ndp->ni_pathlen) {
 388                                 error = NFSERR_ACCES;
 389                                 goto out;
 390                         }
 391                         dp = rootvnode;
 392                         VREF(dp);
 393                 }
 394         } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
 395             (nd->nd_flag & ND_NFSV4) == 0) {
 396                 /*
 397                  * Only cross mount points for NFSv4 when doing a
 398                  * mount while traversing the file system above
 399                  * the mount point, unless nfsrv_enable_crossmntpt is set.
 400                  */
 401                 cnp->cn_flags |= NOCROSSMOUNT;
 402         }
 403
 404         /*
 405          * Initialize for scan, set ni_startdir and bump ref on dp again
 406          * because lookup() will dereference ni_startdir.
 407          */
 408
 409         cnp->cn_thread = p;
 410         ndp->ni_startdir = dp;
 411         ndp->ni_rootdir = rootvnode;
 412         ndp->ni_topdir = NULL;
 413
 414         if (!lockleaf)
 415                 cnp->cn_flags |= LOCKLEAF;
 416         for (;;) {
 417                 cnp->cn_nameptr = cnp->cn_pnbuf;
 418                 /*
 419                  * Call lookup() to do the real work.  If an error occurs,
 420                  * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
 421                  * we do not have to dereference anything before returning.
 422                  * In either case ni_startdir will be dereferenced and NULLed
 423                  * out.
 424                  */
 425                 error = lookup(ndp);
 426                 if (error)
 427                         break;
 428
 429                 /*
 430                  * Check for encountering a symbolic link.  Trivial
 431                  * termination occurs if no symlink encountered.
 432                  */
 433                 if ((cnp->cn_flags & ISSYMLINK) == 0) {
 434                         if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
 435                                 nfsvno_relpathbuf(ndp);
 436                         if (ndp->ni_vp && !lockleaf)
 437                                 NFSVOPUNLOCK(ndp->ni_vp, 0);
 438                         break;
 439                 }
 440
 441                 /*
 442                  * Validate symlink
 443                  */
 444                 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
 445                         NFSVOPUNLOCK(ndp->ni_dvp, 0);
 446                 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
 447                         error = EINVAL;
 448                         goto badlink2;
 449                 }
 450
 451                 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
 452                         error = ELOOP;
 453                         goto badlink2;
 454                 }
 455                 if (ndp->ni_pathlen > 1)
 456                         cp = uma_zalloc(namei_zone, M_WAITOK);
 457                 else
 458                         cp = cnp->cn_pnbuf;
 459                 aiov.iov_base = cp;
 460                 aiov.iov_len = MAXPATHLEN;
 461                 auio.uio_iov = &aiov;
 462                 auio.uio_iovcnt = 1;
 463                 auio.uio_offset = 0;
 464                 auio.uio_rw = UIO_READ;
 465                 auio.uio_segflg = UIO_SYSSPACE;
 466                 auio.uio_td = NULL;
 467                 auio.uio_resid = MAXPATHLEN;
 468                 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
 469                 if (error) {
 470                 badlink1:
 471                         if (ndp->ni_pathlen > 1)
 472                                 uma_zfree(namei_zone, cp);
 473                 badlink2:
 474                         vrele(ndp->ni_dvp);
 475                         vput(ndp->ni_vp);
 476                         break;
 477                 }
 478                 linklen = MAXPATHLEN - auio.uio_resid;
 479                 if (linklen == 0) {
 480                         error = ENOENT;
 481                         goto badlink1;
 482                 }
 483                 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
 484                         error = ENAMETOOLONG;
 485                         goto badlink1;
 486                 }
 487
 488                 /*
 489                  * Adjust or replace path
 490                  */
 491                 if (ndp->ni_pathlen > 1) {
 492                         NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
 493                         uma_zfree(namei_zone, cnp->cn_pnbuf);
 494                         cnp->cn_pnbuf = cp;
 495                 } else
 496                         cnp->cn_pnbuf[linklen] = '\0';
 497                 ndp->ni_pathlen += linklen;
 498
 499                 /*
 500                  * Cleanup refs for next loop and check if root directory
 501                  * should replace current directory.  Normally ni_dvp
 502                  * becomes the new base directory and is cleaned up when
 503                  * we loop.  Explicitly null pointers after invalidation
 504                  * to clarify operation.
 505                  */
 506                 vput(ndp->ni_vp);
 507                 ndp->ni_vp = NULL;
 508
 509                 if (cnp->cn_pnbuf[0] == '/') {
 510                         vrele(ndp->ni_dvp);
 511                         ndp->ni_dvp = ndp->ni_rootdir;
 512                         VREF(ndp->ni_dvp);
 513                 }
 514                 ndp->ni_startdir = ndp->ni_dvp;
 515                 ndp->ni_dvp = NULL;
 516         }
 517         if (!lockleaf)
 518                 cnp->cn_flags &= ~LOCKLEAF;
 519
 520 out:
 521         if (error) {
 522                 nfsvno_relpathbuf(ndp);
 523                 ndp->ni_vp = NULL;
 524                 ndp->ni_dvp = NULL;
 525                 ndp->ni_startdir = NULL;
 526         } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
 527                 ndp->ni_dvp = NULL;
 528         }
 529
 530 out1:
 531         NFSEXITCODE2(error, nd);
 532         return (error);
 533 }
 534
 535 /*
 536  * Set up a pathname buffer and return a pointer to it and, optionally
 537  * set a hash pointer.
 538  */
 539 void
 540 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
 541 {
 542         struct componentname *cnp = &ndp->ni_cnd;
 543
 544         cnp->cn_flags |= (NOMACCHECK | HASBUF);
 545         cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
 546         if (hashpp != NULL)
 547                 *hashpp = NULL;
 548         *bufpp = cnp->cn_pnbuf;
 549 }
 550
 551 /*
 552  * Release the above path buffer, if not released by nfsvno_namei().
 553  */
 554 void
 555 nfsvno_relpathbuf(struct nameidata *ndp)
 556 {
 557
 558         if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
 559                 panic("nfsrelpath");
 560         uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
 561         ndp->ni_cnd.cn_flags &= ~HASBUF;
 562 }
 563
 564 /*
 565  * Readlink vnode op into an mbuf list.
 566  */
 567 int
 568 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
 569     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
 570 {
 571         struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
 572         struct iovec *ivp = iv;
 573         struct uio io, *uiop = &io;
 574         struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
 575         int i, len, tlen, error = 0;
 576
 577         len = 0;
 578         i = 0;
 579         while (len < NFS_MAXPATHLEN) {
 580                 NFSMGET(mp);
 581                 MCLGET(mp, M_WAITOK);
 582                 mp->m_len = M_SIZE(mp);
 583                 if (len == 0) {
 584                         mp3 = mp2 = mp;
 585                 } else {
 586                         mp2->m_next = mp;
 587                         mp2 = mp;
 588                 }
 589                 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
 590                         mp->m_len = NFS_MAXPATHLEN - len;
 591                         len = NFS_MAXPATHLEN;
 592                 } else {
 593                         len += mp->m_len;
 594                 }
 595                 ivp->iov_base = mtod(mp, caddr_t);
 596                 ivp->iov_len = mp->m_len;
 597                 i++;
 598                 ivp++;
 599         }
 600         uiop->uio_iov = iv;
 601         uiop->uio_iovcnt = i;
 602         uiop->uio_offset = 0;
 603         uiop->uio_resid = len;
 604         uiop->uio_rw = UIO_READ;
 605         uiop->uio_segflg = UIO_SYSSPACE;
 606         uiop->uio_td = NULL;
 607         error = VOP_READLINK(vp, uiop, cred);
 608         if (error) {
 609                 m_freem(mp3);
 610                 *lenp = 0;
 611                 goto out;
 612         }
 613         if (uiop->uio_resid > 0) {
 614                 len -= uiop->uio_resid;
 615                 tlen = NFSM_RNDUP(len);
 616                 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
 617         }
 618         *lenp = len;
 619         *mpp = mp3;
 620         *mpendp = mp;
 621
 622 out:
 623         NFSEXITCODE(error);
 624         return (error);
 625 }
 626
 627 /*
 628  * Read vnode op call into mbuf list.
 629  */
 630 int
 631 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
 632     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
 633 {
 634         struct mbuf *m;
 635         int i;
 636         struct iovec *iv;
 637         struct iovec *iv2;
 638         int error = 0, len, left, siz, tlen, ioflag = 0;
 639         struct mbuf *m2 = NULL, *m3;
 640         struct uio io, *uiop = &io;
 641         struct nfsheur *nh;
 642
 643         len = left = NFSM_RNDUP(cnt);
 644         m3 = NULL;
 645         /*
 646          * Generate the mbuf list with the uio_iov ref. to it.
 647          */
 648         i = 0;
 649         while (left > 0) {
 650                 NFSMGET(m);
 651                 MCLGET(m, M_WAITOK);
 652                 m->m_len = 0;
 653                 siz = min(M_TRAILINGSPACE(m), left);
 654                 left -= siz;
 655                 i++;
 656                 if (m3)
 657                         m2->m_next = m;
 658                 else
 659                         m3 = m;
 660                 m2 = m;
 661         }
 662         iv = malloc(i * sizeof (struct iovec),
 663             M_TEMP, M_WAITOK);
 664         uiop->uio_iov = iv2 = iv;
 665         m = m3;
 666         left = len;
 667         i = 0;
 668         while (left > 0) {
 669                 if (m == NULL)
 670                         panic("nfsvno_read iov");
 671                 siz = min(M_TRAILINGSPACE(m), left);
 672                 if (siz > 0) {
 673                         iv->iov_base = mtod(m, caddr_t) + m->m_len;
 674                         iv->iov_len = siz;
 675                         m->m_len += siz;
 676                         left -= siz;
 677                         iv++;
 678                         i++;
 679                 }
 680                 m = m->m_next;
 681         }
 682         uiop->uio_iovcnt = i;
 683         uiop->uio_offset = off;
 684         uiop->uio_resid = len;
 685         uiop->uio_rw = UIO_READ;
 686         uiop->uio_segflg = UIO_SYSSPACE;
 687         uiop->uio_td = NULL;
 688         nh = nfsrv_sequential_heuristic(uiop, vp);
 689         ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
 690         /* XXX KDM make this more systematic? */
 691         nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
 692         error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
 693         free(iv2, M_TEMP);
 694         if (error) {
 695                 m_freem(m3);
 696                 *mpp = NULL;
 697                 goto out;
 698         }
 699         nh->nh_nextoff = uiop->uio_offset;
 700         tlen = len - uiop->uio_resid;
 701         cnt = cnt < tlen ? cnt : tlen;
 702         tlen = NFSM_RNDUP(cnt);
 703         if (tlen == 0) {
 704                 m_freem(m3);
 705                 m3 = NULL;
 706         } else if (len != tlen || tlen != cnt)
 707                 nfsrv_adj(m3, len - tlen, tlen - cnt);
 708         *mpp = m3;
 709         *mpendp = m2;
 710
 711 out:
 712         NFSEXITCODE(error);
 713         return (error);
 714 }
 715
 716 /*
 717  * Write vnode op from an mbuf list.
 718  */
 719 int
 720 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
 721     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
 722 {
 723         struct iovec *ivp;
 724         int i, len;
 725         struct iovec *iv;
 726         int ioflags, error;
 727         struct uio io, *uiop = &io;
 728         struct nfsheur *nh;
 729
 730         ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
 731             M_WAITOK);
 732         uiop->uio_iov = iv = ivp;
 733         uiop->uio_iovcnt = cnt;
 734         i = mtod(mp, caddr_t) + mp->m_len - cp;
 735         len = retlen;
 736         while (len > 0) {
 737                 if (mp == NULL)
 738                         panic("nfsvno_write");
 739                 if (i > 0) {
 740                         i = min(i, len);
 741                         ivp->iov_base = cp;
 742                         ivp->iov_len = i;
 743                         ivp++;
 744                         len -= i;
 745                 }
 746                 mp = mp->m_next;
 747                 if (mp) {
 748                         i = mp->m_len;
 749                         cp = mtod(mp, caddr_t);
 750                 }
 751         }
 752
 753         if (stable == NFSWRITE_UNSTABLE)
 754                 ioflags = IO_NODELOCKED;
 755         else
 756                 ioflags = (IO_SYNC | IO_NODELOCKED);
 757         uiop->uio_resid = retlen;
 758         uiop->uio_rw = UIO_WRITE;
 759         uiop->uio_segflg = UIO_SYSSPACE;
 760         NFSUIOPROC(uiop, p);
 761         uiop->uio_offset = off;
 762         nh = nfsrv_sequential_heuristic(uiop, vp);
 763         ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
 764         /* XXX KDM make this more systematic? */
 765         nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
 766         error = VOP_WRITE(vp, uiop, ioflags, cred);
 767         if (error == 0)
 768                 nh->nh_nextoff = uiop->uio_offset;
 769         free(iv, M_TEMP);
 770
 771         NFSEXITCODE(error);
 772         return (error);
 773 }
 774
 775 /*
 776  * Common code for creating a regular file (plus special files for V2).
 777  */
 778 int
 779 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
 780     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
 781     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
 782 {
 783         u_quad_t tempsize;
 784         int error;
 785
 786         error = nd->nd_repstat;
 787         if (!error && ndp->ni_vp == NULL) {
 788                 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
 789                         vrele(ndp->ni_startdir);
 790                         error = VOP_CREATE(ndp->ni_dvp,
 791                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
 792                         vput(ndp->ni_dvp);
 793                         nfsvno_relpathbuf(ndp);
 794                         if (!error) {
 795                                 if (*exclusive_flagp) {
 796                                         *exclusive_flagp = 0;
 797                                         NFSVNO_ATTRINIT(nvap);
 798                                         nvap->na_atime.tv_sec = cverf[0];
 799                                         nvap->na_atime.tv_nsec = cverf[1];
 800                                         error = VOP_SETATTR(ndp->ni_vp,
 801                                             &nvap->na_vattr, nd->nd_cred);
 802                                         if (error != 0) {
 803                                                 vput(ndp->ni_vp);
 804                                                 ndp->ni_vp = NULL;
 805                                                 error = NFSERR_NOTSUPP;
 806                                         }
 807                                 }
 808                         }
 809                 /*
 810                  * NFS V2 Only. nfsrvd_mknod() does this for V3.
 811                  * (This implies, just get out on an error.)
 812                  */
 813                 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
 814                         nvap->na_type == VFIFO) {
 815                         if (nvap->na_type == VCHR && rdev == 0xffffffff)
 816                                 nvap->na_type = VFIFO;
 817                         if (nvap->na_type != VFIFO &&
 818                             (error = priv_check_cred(nd->nd_cred,
 819                              PRIV_VFS_MKNOD_DEV, 0))) {
 820                                 vrele(ndp->ni_startdir);
 821                                 nfsvno_relpathbuf(ndp);
 822                                 vput(ndp->ni_dvp);
 823                                 goto out;
 824                         }
 825                         nvap->na_rdev = rdev;
 826                         error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 827                             &ndp->ni_cnd, &nvap->na_vattr);
 828                         vput(ndp->ni_dvp);
 829                         nfsvno_relpathbuf(ndp);
 830                         vrele(ndp->ni_startdir);
 831                         if (error)
 832                                 goto out;
 833                 } else {
 834                         vrele(ndp->ni_startdir);
 835                         nfsvno_relpathbuf(ndp);
 836                         vput(ndp->ni_dvp);
 837                         error = ENXIO;
 838                         goto out;
 839                 }
 840                 *vpp = ndp->ni_vp;
 841         } else {
 842                 /*
 843                  * Handle cases where error is already set and/or
 844                  * the file exists.
 845                  * 1 - clean up the lookup
 846                  * 2 - iff !error and na_size set, truncate it
 847                  */
 848                 vrele(ndp->ni_startdir);
 849                 nfsvno_relpathbuf(ndp);
 850                 *vpp = ndp->ni_vp;
 851                 if (ndp->ni_dvp == *vpp)
 852                         vrele(ndp->ni_dvp);
 853                 else
 854                         vput(ndp->ni_dvp);
 855                 if (!error && nvap->na_size != VNOVAL) {
 856                         error = nfsvno_accchk(*vpp, VWRITE,
 857                             nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
 858                             NFSACCCHK_VPISLOCKED, NULL);
 859                         if (!error) {
 860                                 tempsize = nvap->na_size;
 861                                 NFSVNO_ATTRINIT(nvap);
 862                                 nvap->na_size = tempsize;
 863                                 error = VOP_SETATTR(*vpp,
 864                                     &nvap->na_vattr, nd->nd_cred);
 865                         }
 866                 }
 867                 if (error)
 868                         vput(*vpp);
 869         }
 870
 871 out:
 872         NFSEXITCODE(error);
 873         return (error);
 874 }
 875
 876 /*
 877  * Do a mknod vnode op.
 878  */
 879 int
 880 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
 881     struct thread *p)
 882 {
 883         int error = 0;
 884         enum vtype vtyp;
 885
 886         vtyp = nvap->na_type;
 887         /*
 888          * Iff doesn't exist, create it.
 889          */
 890         if (ndp->ni_vp) {
 891                 vrele(ndp->ni_startdir);
 892                 nfsvno_relpathbuf(ndp);
 893                 vput(ndp->ni_dvp);
 894                 vrele(ndp->ni_vp);
 895                 error = EEXIST;
 896                 goto out;
 897         }
 898         if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
 899                 vrele(ndp->ni_startdir);
 900                 nfsvno_relpathbuf(ndp);
 901                 vput(ndp->ni_dvp);
 902                 error = NFSERR_BADTYPE;
 903                 goto out;
 904         }
 905         if (vtyp == VSOCK) {
 906                 vrele(ndp->ni_startdir);
 907                 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
 908                     &ndp->ni_cnd, &nvap->na_vattr);
 909                 vput(ndp->ni_dvp);
 910                 nfsvno_relpathbuf(ndp);
 911         } else {
 912                 if (nvap->na_type != VFIFO &&
 913                     (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
 914                         vrele(ndp->ni_startdir);
 915                         nfsvno_relpathbuf(ndp);
 916                         vput(ndp->ni_dvp);
 917                         goto out;
 918                 }
 919                 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
 920                     &ndp->ni_cnd, &nvap->na_vattr);
 921                 vput(ndp->ni_dvp);
 922                 nfsvno_relpathbuf(ndp);
 923                 vrele(ndp->ni_startdir);
 924                 /*
 925                  * Since VOP_MKNOD returns the ni_vp, I can't
 926                  * see any reason to do the lookup.
 927                  */
 928         }
 929
 930 out:
 931         NFSEXITCODE(error);
 932         return (error);
 933 }
 934
 935 /*
 936  * Mkdir vnode op.
 937  */
 938 int
 939 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
 940     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
 941 {
 942         int error = 0;
 943
 944         if (ndp->ni_vp != NULL) {
 945                 if (ndp->ni_dvp == ndp->ni_vp)
 946                         vrele(ndp->ni_dvp);
 947                 else
 948                         vput(ndp->ni_dvp);
 949                 vrele(ndp->ni_vp);
 950                 nfsvno_relpathbuf(ndp);
 951                 error = EEXIST;
 952                 goto out;
 953         }
 954         error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 955             &nvap->na_vattr);
 956         vput(ndp->ni_dvp);
 957         nfsvno_relpathbuf(ndp);
 958
 959 out:
 960         NFSEXITCODE(error);
 961         return (error);
 962 }
 963
 964 /*
 965  * symlink vnode op.
 966  */
 967 int
 968 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
 969     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
 970     struct nfsexstuff *exp)
 971 {
 972         int error = 0;
 973
 974         if (ndp->ni_vp) {
 975                 vrele(ndp->ni_startdir);
 976                 nfsvno_relpathbuf(ndp);
 977                 if (ndp->ni_dvp == ndp->ni_vp)
 978                         vrele(ndp->ni_dvp);
 979                 else
 980                         vput(ndp->ni_dvp);
 981                 vrele(ndp->ni_vp);
 982                 error = EEXIST;
 983                 goto out;
 984         }
 985
 986         error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
 987             &nvap->na_vattr, pathcp);
 988         vput(ndp->ni_dvp);
 989         vrele(ndp->ni_startdir);
 990         nfsvno_relpathbuf(ndp);
 991         /*
 992          * Although FreeBSD still had the lookup code in
 993          * it for 7/current, there doesn't seem to be any
 994          * point, since VOP_SYMLINK() returns the ni_vp.
 995          * Just vput it for v2.
 996          */
 997         if (!not_v2 && !error)
 998                 vput(ndp->ni_vp);
 999
1000 out:
1001         NFSEXITCODE(error);
1002         return (error);
1003 }
1004
1005 /*
1006  * Parse symbolic link arguments.
1007  * This function has an ugly side effect. It will malloc() an area for
1008  * the symlink and set iov_base to point to it, only if it succeeds.
1009  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1010  * be FREE'd later.
1011  */
1012 int
1013 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1014     struct thread *p, char **pathcpp, int *lenp)
1015 {
1016         u_int32_t *tl;
1017         char *pathcp = NULL;
1018         int error = 0, len;
1019         struct nfsv2_sattr *sp;
1020
1021         *pathcpp = NULL;
1022         *lenp = 0;
1023         if ((nd->nd_flag & ND_NFSV3) &&
1024             (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1025                 goto nfsmout;
1026         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1027         len = fxdr_unsigned(int, *tl);
1028         if (len > NFS_MAXPATHLEN || len <= 0) {
1029                 error = EBADRPC;
1030                 goto nfsmout;
1031         }
1032         pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1033         error = nfsrv_mtostr(nd, pathcp, len);
1034         if (error)
1035                 goto nfsmout;
1036         if (nd->nd_flag & ND_NFSV2) {
1037                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1038                 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1039         }
1040         *pathcpp = pathcp;
1041         *lenp = len;
1042         NFSEXITCODE2(0, nd);
1043         return (0);
1044 nfsmout:
1045         if (pathcp)
1046                 free(pathcp, M_TEMP);
1047         NFSEXITCODE2(error, nd);
1048         return (error);
1049 }
1050
1051 /*
1052  * Remove a non-directory object.
1053  */
1054 int
1055 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1056     struct thread *p, struct nfsexstuff *exp)
1057 {
1058         struct vnode *vp;
1059         int error = 0;
1060
1061         vp = ndp->ni_vp;
1062         if (vp->v_type == VDIR)
1063                 error = NFSERR_ISDIR;
1064         else if (is_v4)
1065                 error = nfsrv_checkremove(vp, 1, p);
1066         if (!error)
1067                 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1068         if (ndp->ni_dvp == vp)
1069                 vrele(ndp->ni_dvp);
1070         else
1071                 vput(ndp->ni_dvp);
1072         vput(vp);
1073         if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1074                 nfsvno_relpathbuf(ndp);
1075         NFSEXITCODE(error);
1076         return (error);
1077 }
1078
1079 /*
1080  * Remove a directory.
1081  */
1082 int
1083 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1084     struct thread *p, struct nfsexstuff *exp)
1085 {
1086         struct vnode *vp;
1087         int error = 0;
1088
1089         vp = ndp->ni_vp;
1090         if (vp->v_type != VDIR) {
1091                 error = ENOTDIR;
1092                 goto out;
1093         }
1094         /*
1095          * No rmdir "." please.
1096          */
1097         if (ndp->ni_dvp == vp) {
1098                 error = EINVAL;
1099                 goto out;
1100         }
1101         /*
1102          * The root of a mounted filesystem cannot be deleted.
1103          */
1104         if (vp->v_vflag & VV_ROOT)
1105                 error = EBUSY;
1106 out:
1107         if (!error)
1108                 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1109         if (ndp->ni_dvp == vp)
1110                 vrele(ndp->ni_dvp);
1111         else
1112                 vput(ndp->ni_dvp);
1113         vput(vp);
1114         if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1115                 nfsvno_relpathbuf(ndp);
1116         NFSEXITCODE(error);
1117         return (error);
1118 }
1119
1120 /*
1121  * Rename vnode op.
1122  */
1123 int
1124 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1125     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1126 {
1127         struct vnode *fvp, *tvp, *tdvp;
1128         int error = 0;
1129
1130         fvp = fromndp->ni_vp;
1131         if (ndstat) {
1132                 vrele(fromndp->ni_dvp);
1133                 vrele(fvp);
1134                 error = ndstat;
1135                 goto out1;
1136         }
1137         tdvp = tondp->ni_dvp;
1138         tvp = tondp->ni_vp;
1139         if (tvp != NULL) {
1140                 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1141                         error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1142                         goto out;
1143                 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1144                         error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1145                         goto out;
1146                 }
1147                 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1148                         error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1149                         goto out;
1150                 }
1151
1152                 /*
1153                  * A rename to '.' or '..' results in a prematurely
1154                  * unlocked vnode on FreeBSD5, so I'm just going to fail that
1155                  * here.
1156                  */
1157                 if ((tondp->ni_cnd.cn_namelen == 1 &&
1158                      tondp->ni_cnd.cn_nameptr[0] == '.') ||
1159                     (tondp->ni_cnd.cn_namelen == 2 &&
1160                      tondp->ni_cnd.cn_nameptr[0] == '.' &&
1161                      tondp->ni_cnd.cn_nameptr[1] == '.')) {
1162                         error = EINVAL;
1163                         goto out;
1164                 }
1165         }
1166         if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1167                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1168                 goto out;
1169         }
1170         if (fvp->v_mount != tdvp->v_mount) {
1171                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1172                 goto out;
1173         }
1174         if (fvp == tdvp) {
1175                 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1176                 goto out;
1177         }
1178         if (fvp == tvp) {
1179                 /*
1180                  * If source and destination are the same, there is nothing to
1181                  * do. Set error to -1 to indicate this.
1182                  */
1183                 error = -1;
1184                 goto out;
1185         }
1186         if (ndflag & ND_NFSV4) {
1187                 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1188                         error = nfsrv_checkremove(fvp, 0, p);
1189                         NFSVOPUNLOCK(fvp, 0);
1190                 } else
1191                         error = EPERM;
1192                 if (tvp && !error)
1193                         error = nfsrv_checkremove(tvp, 1, p);
1194         } else {
1195                 /*
1196                  * For NFSv2 and NFSv3, try to get rid of the delegation, so
1197                  * that the NFSv4 client won't be confused by the rename.
1198                  * Since nfsd_recalldelegation() can only be called on an
1199                  * unlocked vnode at this point and fvp is the file that will
1200                  * still exist after the rename, just do fvp.
1201                  */
1202                 nfsd_recalldelegation(fvp, p);
1203         }
1204 out:
1205         if (!error) {
1206                 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1207                     &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1208                     &tondp->ni_cnd);
1209         } else {
1210                 if (tdvp == tvp)
1211                         vrele(tdvp);
1212                 else
1213                         vput(tdvp);
1214                 if (tvp)
1215                         vput(tvp);
1216                 vrele(fromndp->ni_dvp);
1217                 vrele(fvp);
1218                 if (error == -1)
1219                         error = 0;
1220         }
1221         vrele(tondp->ni_startdir);
1222         nfsvno_relpathbuf(tondp);
1223 out1:
1224         vrele(fromndp->ni_startdir);
1225         nfsvno_relpathbuf(fromndp);
1226         NFSEXITCODE(error);
1227         return (error);
1228 }
1229
1230 /*
1231  * Link vnode op.
1232  */
1233 int
1234 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1235     struct thread *p, struct nfsexstuff *exp)
1236 {
1237         struct vnode *xp;
1238         int error = 0;
1239
1240         xp = ndp->ni_vp;
1241         if (xp != NULL) {
1242                 error = EEXIST;
1243         } else {
1244                 xp = ndp->ni_dvp;
1245                 if (vp->v_mount != xp->v_mount)
1246                         error = EXDEV;
1247         }
1248         if (!error) {
1249                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1250                 if ((vp->v_iflag & VI_DOOMED) == 0)
1251                         error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1252                 else
1253                         error = EPERM;
1254                 if (ndp->ni_dvp == vp)
1255                         vrele(ndp->ni_dvp);
1256                 else
1257                         vput(ndp->ni_dvp);
1258                 NFSVOPUNLOCK(vp, 0);
1259         } else {
1260                 if (ndp->ni_dvp == ndp->ni_vp)
1261                         vrele(ndp->ni_dvp);
1262                 else
1263                         vput(ndp->ni_dvp);
1264                 if (ndp->ni_vp)
1265                         vrele(ndp->ni_vp);
1266         }
1267         nfsvno_relpathbuf(ndp);
1268         NFSEXITCODE(error);
1269         return (error);
1270 }
1271
1272 /*
1273  * Do the fsync() appropriate for the commit.
1274  */
1275 int
1276 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1277     struct thread *td)
1278 {
1279         int error = 0;
1280
1281         /*
1282          * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1283          * file is done.  At this time VOP_FSYNC does not accept offset and
1284          * byte count parameters so call VOP_FSYNC the whole file for now.
1285          * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1286          * File systems that do not use the buffer cache (as indicated
1287          * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1288          */
1289         if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1290             (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1291                 /*
1292                  * Give up and do the whole thing
1293                  */
1294                 if (vp->v_object &&
1295                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1296                         VM_OBJECT_WLOCK(vp->v_object);
1297                         vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1298                         VM_OBJECT_WUNLOCK(vp->v_object);
1299                 }
1300                 error = VOP_FSYNC(vp, MNT_WAIT, td);
1301         } else {
1302                 /*
1303                  * Locate and synchronously write any buffers that fall
1304                  * into the requested range.  Note:  we are assuming that
1305                  * f_iosize is a power of 2.
1306                  */
1307                 int iosize = vp->v_mount->mnt_stat.f_iosize;
1308                 int iomask = iosize - 1;
1309                 struct bufobj *bo;
1310                 daddr_t lblkno;
1311
1312                 /*
1313                  * Align to iosize boundary, super-align to page boundary.
1314                  */
1315                 if (off & iomask) {
1316                         cnt += off & iomask;
1317                         off &= ~(u_quad_t)iomask;
1318                 }
1319                 if (off & PAGE_MASK) {
1320                         cnt += off & PAGE_MASK;
1321                         off &= ~(u_quad_t)PAGE_MASK;
1322                 }
1323                 lblkno = off / iosize;
1324
1325                 if (vp->v_object &&
1326                    (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1327                         VM_OBJECT_WLOCK(vp->v_object);
1328                         vm_object_page_clean(vp->v_object, off, off + cnt,
1329                             OBJPC_SYNC);
1330                         VM_OBJECT_WUNLOCK(vp->v_object);
1331                 }
1332
1333                 bo = &vp->v_bufobj;
1334                 BO_LOCK(bo);
1335                 while (cnt > 0) {
1336                         struct buf *bp;
1337
1338                         /*
1339                          * If we have a buffer and it is marked B_DELWRI we
1340                          * have to lock and write it.  Otherwise the prior
1341                          * write is assumed to have already been committed.
1342                          *
1343                          * gbincore() can return invalid buffers now so we
1344                          * have to check that bit as well (though B_DELWRI
1345                          * should not be set if B_INVAL is set there could be
1346                          * a race here since we haven't locked the buffer).
1347                          */
1348                         if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1349                                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1350                                     LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1351                                         BO_LOCK(bo);
1352                                         continue; /* retry */
1353                                 }
1354                                 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1355                                     B_DELWRI) {
1356                                         bremfree(bp);
1357                                         bp->b_flags &= ~B_ASYNC;
1358                                         bwrite(bp);
1359                                         ++nfs_commit_miss;
1360                                 } else
1361                                         BUF_UNLOCK(bp);
1362                                 BO_LOCK(bo);
1363                         }
1364                         ++nfs_commit_blks;
1365                         if (cnt < iosize)
1366                                 break;
1367                         cnt -= iosize;
1368                         ++lblkno;
1369                 }
1370                 BO_UNLOCK(bo);
1371         }
1372         NFSEXITCODE(error);
1373         return (error);
1374 }
1375
1376 /*
1377  * Statfs vnode op.
1378  */
1379 int
1380 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1381 {
1382         int error;
1383
1384         error = VFS_STATFS(vp->v_mount, sf);
1385         if (error == 0) {
1386                 /*
1387                  * Since NFS handles these values as unsigned on the
1388                  * wire, there is no way to represent negative values,
1389                  * so set them to 0. Without this, they will appear
1390                  * to be very large positive values for clients like
1391                  * Solaris10.
1392                  */
1393                 if (sf->f_bavail < 0)
1394                         sf->f_bavail = 0;
1395                 if (sf->f_ffree < 0)
1396                         sf->f_ffree = 0;
1397         }
1398         NFSEXITCODE(error);
1399         return (error);
1400 }
1401
1402 /*
1403  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1404  * must handle nfsrv_opencheck() calls after any other access checks.
1405  */
1406 void
1407 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1408     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1409     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1410     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1411     struct nfsexstuff *exp, struct vnode **vpp)
1412 {
1413         struct vnode *vp = NULL;
1414         u_quad_t tempsize;
1415         struct nfsexstuff nes;
1416
1417         if (ndp->ni_vp == NULL)
1418                 nd->nd_repstat = nfsrv_opencheck(clientid,
1419                     stateidp, stp, NULL, nd, p, nd->nd_repstat);
1420         if (!nd->nd_repstat) {
1421                 if (ndp->ni_vp == NULL) {
1422                         vrele(ndp->ni_startdir);
1423                         nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1424                             &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1425                         vput(ndp->ni_dvp);
1426                         nfsvno_relpathbuf(ndp);
1427                         if (!nd->nd_repstat) {
1428                                 if (*exclusive_flagp) {
1429                                         *exclusive_flagp = 0;
1430                                         NFSVNO_ATTRINIT(nvap);
1431                                         nvap->na_atime.tv_sec = cverf[0];
1432                                         nvap->na_atime.tv_nsec = cverf[1];
1433                                         nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1434                                             &nvap->na_vattr, cred);
1435                                         if (nd->nd_repstat != 0) {
1436                                                 vput(ndp->ni_vp);
1437                                                 ndp->ni_vp = NULL;
1438                                                 nd->nd_repstat = NFSERR_NOTSUPP;
1439                                         } else
1440                                                 NFSSETBIT_ATTRBIT(attrbitp,
1441                                                     NFSATTRBIT_TIMEACCESS);
1442                                 } else {
1443                                         nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1444                                             aclp, p, attrbitp, exp);
1445                                 }
1446                         }
1447                         vp = ndp->ni_vp;
1448                 } else {
1449                         if (ndp->ni_startdir)
1450                                 vrele(ndp->ni_startdir);
1451                         nfsvno_relpathbuf(ndp);
1452                         vp = ndp->ni_vp;
1453                         if (create == NFSV4OPEN_CREATE) {
1454                                 if (ndp->ni_dvp == vp)
1455                                         vrele(ndp->ni_dvp);
1456                                 else
1457                                         vput(ndp->ni_dvp);
1458                         }
1459                         if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1460                                 if (ndp->ni_cnd.cn_flags & RDONLY)
1461                                         NFSVNO_SETEXRDONLY(&nes);
1462                                 else
1463                                         NFSVNO_EXINIT(&nes);
1464                                 nd->nd_repstat = nfsvno_accchk(vp,
1465                                     VWRITE, cred, &nes, p,
1466                                     NFSACCCHK_NOOVERRIDE,
1467                                     NFSACCCHK_VPISLOCKED, NULL);
1468                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1469                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1470                                 if (!nd->nd_repstat) {
1471                                         tempsize = nvap->na_size;
1472                                         NFSVNO_ATTRINIT(nvap);
1473                                         nvap->na_size = tempsize;
1474                                         nd->nd_repstat = VOP_SETATTR(vp,
1475                                             &nvap->na_vattr, cred);
1476                                 }
1477                         } else if (vp->v_type == VREG) {
1478                                 nd->nd_repstat = nfsrv_opencheck(clientid,
1479                                     stateidp, stp, vp, nd, p, nd->nd_repstat);
1480                         }
1481                 }
1482         } else {
1483                 if (ndp->ni_cnd.cn_flags & HASBUF)
1484                         nfsvno_relpathbuf(ndp);
1485                 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1486                         vrele(ndp->ni_startdir);
1487                         if (ndp->ni_dvp == ndp->ni_vp)
1488                                 vrele(ndp->ni_dvp);
1489                         else
1490                                 vput(ndp->ni_dvp);
1491                         if (ndp->ni_vp)
1492                                 vput(ndp->ni_vp);
1493                 }
1494         }
1495         *vpp = vp;
1496
1497         NFSEXITCODE2(0, nd);
1498 }
1499
1500 /*
1501  * Updates the file rev and sets the mtime and ctime
1502  * to the current clock time, returning the va_filerev and va_Xtime
1503  * values.
1504  * Return ESTALE to indicate the vnode is VI_DOOMED.
1505  */
1506 int
1507 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1508     struct ucred *cred, struct thread *p)
1509 {
1510         struct vattr va;
1511
1512         VATTR_NULL(&va);
1513         vfs_timestamp(&va.va_mtime);
1514         if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1515                 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1516                 if ((vp->v_iflag & VI_DOOMED) != 0)
1517                         return (ESTALE);
1518         }
1519         (void) VOP_SETATTR(vp, &va, cred);
1520         (void) nfsvno_getattr(vp, nvap, cred, p, 1);
1521         return (0);
1522 }
1523
1524 /*
1525  * Glue routine to nfsv4_fillattr().
1526  */
1527 int
1528 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1529     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1530     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1531     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1532 {
1533         int error;
1534
1535         error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1536             attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1537             mounted_on_fileno);
1538         NFSEXITCODE2(0, nd);
1539         return (error);
1540 }
1541
1542 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1543 /*
1544  * nfs readdir service
1545  * - mallocs what it thinks is enough to read
1546  *      count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1547  * - calls VOP_READDIR()
1548  * - loops around building the reply
1549  *      if the output generated exceeds count break out of loop
1550  *      The NFSM_CLGET macro is used here so that the reply will be packed
1551  *      tightly in mbuf clusters.
1552  * - it trims out records with d_fileno == 0
1553  *      this doesn't matter for Unix clients, but they might confuse clients
1554  *      for other os'.
1555  * - it trims out records with d_type == DT_WHT
1556  *      these cannot be seen through NFS (unless we extend the protocol)
1557  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1558  * PS: The NFS protocol spec. does not clarify what the "count" byte
1559  *      argument is a count of.. just name strings and file id's or the
1560  *      entire reply rpc or ...
1561  *      I tried just file name and id sizes and it confused the Sun client,
1562  *      so I am using the full rpc size now. The "paranoia.." comment refers
1563  *      to including the status longwords that are not a part of the dir.
1564  *      "entry" structures, but are in the rpc.
1565  */
1566 int
1567 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1568     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1569 {
1570         struct dirent *dp;
1571         u_int32_t *tl;
1572         int dirlen;
1573         char *cpos, *cend, *rbuf;
1574         struct nfsvattr at;
1575         int nlen, error = 0, getret = 1;
1576         int siz, cnt, fullsiz, eofflag, ncookies;
1577         u_int64_t off, toff, verf;
1578         u_long *cookies = NULL, *cookiep;
1579         struct uio io;
1580         struct iovec iv;
1581         int is_ufs;
1582
1583         if (nd->nd_repstat) {
1584                 nfsrv_postopattr(nd, getret, &at);
1585                 goto out;
1586         }
1587         if (nd->nd_flag & ND_NFSV2) {
1588                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1589                 off = fxdr_unsigned(u_quad_t, *tl++);
1590         } else {
1591                 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1592                 off = fxdr_hyper(tl);
1593                 tl += 2;
1594                 verf = fxdr_hyper(tl);
1595                 tl += 2;
1596         }
1597         toff = off;
1598         cnt = fxdr_unsigned(int, *tl);
1599         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1600                 cnt = NFS_SRVMAXDATA(nd);
1601         siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1602         fullsiz = siz;
1603         if (nd->nd_flag & ND_NFSV3) {
1604                 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1605                     p, 1);
1606 #if 0
1607                 /*
1608                  * va_filerev is not sufficient as a cookie verifier,
1609                  * since it is not supposed to change when entries are
1610                  * removed/added unless that offset cookies returned to
1611                  * the client are no longer valid.
1612                  */
1613                 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1614                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1615 #endif
1616         }
1617         if (!nd->nd_repstat && vp->v_type != VDIR)
1618                 nd->nd_repstat = NFSERR_NOTDIR;
1619         if (nd->nd_repstat == 0 && cnt == 0) {
1620                 if (nd->nd_flag & ND_NFSV2)
1621                         /* NFSv2 does not have NFSERR_TOOSMALL */
1622                         nd->nd_repstat = EPERM;
1623                 else
1624                         nd->nd_repstat = NFSERR_TOOSMALL;
1625         }
1626         if (!nd->nd_repstat)
1627                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1628                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1629                     NFSACCCHK_VPISLOCKED, NULL);
1630         if (nd->nd_repstat) {
1631                 vput(vp);
1632                 if (nd->nd_flag & ND_NFSV3)
1633                         nfsrv_postopattr(nd, getret, &at);
1634                 goto out;
1635         }
1636         is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1637         rbuf = malloc(siz, M_TEMP, M_WAITOK);
1638 again:
1639         eofflag = 0;
1640         if (cookies) {
1641                 free(cookies, M_TEMP);
1642                 cookies = NULL;
1643         }
1644
1645         iv.iov_base = rbuf;
1646         iv.iov_len = siz;
1647         io.uio_iov = &iv;
1648         io.uio_iovcnt = 1;
1649         io.uio_offset = (off_t)off;
1650         io.uio_resid = siz;
1651         io.uio_segflg = UIO_SYSSPACE;
1652         io.uio_rw = UIO_READ;
1653         io.uio_td = NULL;
1654         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1655             &cookies);
1656         off = (u_int64_t)io.uio_offset;
1657         if (io.uio_resid)
1658                 siz -= io.uio_resid;
1659
1660         if (!cookies && !nd->nd_repstat)
1661                 nd->nd_repstat = NFSERR_PERM;
1662         if (nd->nd_flag & ND_NFSV3) {
1663                 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1664                 if (!nd->nd_repstat)
1665                         nd->nd_repstat = getret;
1666         }
1667
1668         /*
1669          * Handles the failed cases. nd->nd_repstat == 0 past here.
1670          */
1671         if (nd->nd_repstat) {
1672                 vput(vp);
1673                 free(rbuf, M_TEMP);
1674                 if (cookies)
1675                         free(cookies, M_TEMP);
1676                 if (nd->nd_flag & ND_NFSV3)
1677                         nfsrv_postopattr(nd, getret, &at);
1678                 goto out;
1679         }
1680         /*
1681          * If nothing read, return eof
1682          * rpc reply
1683          */
1684         if (siz == 0) {
1685                 vput(vp);
1686                 if (nd->nd_flag & ND_NFSV2) {
1687                         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1688                 } else {
1689                         nfsrv_postopattr(nd, getret, &at);
1690                         NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1691                         txdr_hyper(at.na_filerev, tl);
1692                         tl += 2;
1693                 }
1694                 *tl++ = newnfs_false;
1695                 *tl = newnfs_true;
1696                 free(rbuf, M_TEMP);
1697                 free(cookies, M_TEMP);
1698                 goto out;
1699         }
1700
1701         /*
1702          * Check for degenerate cases of nothing useful read.
1703          * If so go try again
1704          */
1705         cpos = rbuf;
1706         cend = rbuf + siz;
1707         dp = (struct dirent *)cpos;
1708         cookiep = cookies;
1709
1710         /*
1711          * For some reason FreeBSD's ufs_readdir() chooses to back the
1712          * directory offset up to a block boundary, so it is necessary to
1713          * skip over the records that precede the requested offset. This
1714          * requires the assumption that file offset cookies monotonically
1715          * increase.
1716          */
1717         while (cpos < cend && ncookies > 0 &&
1718             (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1719              (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1720                 cpos += dp->d_reclen;
1721                 dp = (struct dirent *)cpos;
1722                 cookiep++;
1723                 ncookies--;
1724         }
1725         if (cpos >= cend || ncookies == 0) {
1726                 siz = fullsiz;
1727                 toff = off;
1728                 goto again;
1729         }
1730         vput(vp);
1731
1732         /*
1733          * dirlen is the size of the reply, including all XDR and must
1734          * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1735          * if the XDR should be included in "count", but to be safe, we do.
1736          * (Include the two booleans at the end of the reply in dirlen now.)
1737          */
1738         if (nd->nd_flag & ND_NFSV3) {
1739                 nfsrv_postopattr(nd, getret, &at);
1740                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1741                 txdr_hyper(at.na_filerev, tl);
1742                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1743         } else {
1744                 dirlen = 2 * NFSX_UNSIGNED;
1745         }
1746
1747         /* Loop through the records and build reply */
1748         while (cpos < cend && ncookies > 0) {
1749                 nlen = dp->d_namlen;
1750                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1751                         nlen <= NFS_MAXNAMLEN) {
1752                         if (nd->nd_flag & ND_NFSV3)
1753                                 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1754                         else
1755                                 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1756                         if (dirlen > cnt) {
1757                                 eofflag = 0;
1758                                 break;
1759                         }
1760
1761                         /*
1762                          * Build the directory record xdr from
1763                          * the dirent entry.
1764                          */
1765                         if (nd->nd_flag & ND_NFSV3) {
1766                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1767                                 *tl++ = newnfs_true;
1768                                 *tl++ = 0;
1769                         } else {
1770                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1771                                 *tl++ = newnfs_true;
1772                         }
1773                         *tl = txdr_unsigned(dp->d_fileno);
1774                         (void) nfsm_strtom(nd, dp->d_name, nlen);
1775                         if (nd->nd_flag & ND_NFSV3) {
1776                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1777                                 *tl++ = 0;
1778                         } else
1779                                 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1780                         *tl = txdr_unsigned(*cookiep);
1781                 }
1782                 cpos += dp->d_reclen;
1783                 dp = (struct dirent *)cpos;
1784                 cookiep++;
1785                 ncookies--;
1786         }
1787         if (cpos < cend)
1788                 eofflag = 0;
1789         NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1790         *tl++ = newnfs_false;
1791         if (eofflag)
1792                 *tl = newnfs_true;
1793         else
1794                 *tl = newnfs_false;
1795         free(rbuf, M_TEMP);
1796         free(cookies, M_TEMP);
1797
1798 out:
1799         NFSEXITCODE2(0, nd);
1800         return (0);
1801 nfsmout:
1802         vput(vp);
1803         NFSEXITCODE2(error, nd);
1804         return (error);
1805 }
1806
1807 /*
1808  * Readdirplus for V3 and Readdir for V4.
1809  */
1810 int
1811 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1812     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1813 {
1814         struct dirent *dp;
1815         u_int32_t *tl;
1816         int dirlen;
1817         char *cpos, *cend, *rbuf;
1818         struct vnode *nvp;
1819         fhandle_t nfh;
1820         struct nfsvattr nva, at, *nvap = &nva;
1821         struct mbuf *mb0, *mb1;
1822         struct nfsreferral *refp;
1823         int nlen, r, error = 0, getret = 1, usevget = 1;
1824         int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1825         caddr_t bpos0, bpos1;
1826         u_int64_t off, toff, verf;
1827         u_long *cookies = NULL, *cookiep;
1828         nfsattrbit_t attrbits, rderrbits, savbits;
1829         struct uio io;
1830         struct iovec iv;
1831         struct componentname cn;
1832         int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
1833         struct mount *mp, *new_mp;
1834         uint64_t mounted_on_fileno;
1835
1836         if (nd->nd_repstat) {
1837                 nfsrv_postopattr(nd, getret, &at);
1838                 goto out;
1839         }
1840         NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1841         off = fxdr_hyper(tl);
1842         toff = off;
1843         tl += 2;
1844         verf = fxdr_hyper(tl);
1845         tl += 2;
1846         siz = fxdr_unsigned(int, *tl++);
1847         cnt = fxdr_unsigned(int, *tl);
1848
1849         /*
1850          * Use the server's maximum data transfer size as the upper bound
1851          * on reply datalen.
1852          */
1853         if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1854                 cnt = NFS_SRVMAXDATA(nd);
1855
1856         /*
1857          * siz is a "hint" of how much directory information (name, fileid,
1858          * cookie) should be in the reply. At least one client "hints" 0,
1859          * so I set it to cnt for that case. I also round it up to the
1860          * next multiple of DIRBLKSIZ.
1861          */
1862         if (siz <= 0)
1863                 siz = cnt;
1864         siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1865
1866         if (nd->nd_flag & ND_NFSV4) {
1867                 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1868                 if (error)
1869                         goto nfsmout;
1870                 NFSSET_ATTRBIT(&savbits, &attrbits);
1871                 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1872                 NFSZERO_ATTRBIT(&rderrbits);
1873                 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1874         } else {
1875                 NFSZERO_ATTRBIT(&attrbits);
1876         }
1877         fullsiz = siz;
1878         nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1879         if (!nd->nd_repstat) {
1880             if (off && verf != at.na_filerev) {
1881                 /*
1882                  * va_filerev is not sufficient as a cookie verifier,
1883                  * since it is not supposed to change when entries are
1884                  * removed/added unless that offset cookies returned to
1885                  * the client are no longer valid.
1886                  */
1887 #if 0
1888                 if (nd->nd_flag & ND_NFSV4) {
1889                         nd->nd_repstat = NFSERR_NOTSAME;
1890                 } else {
1891                         nd->nd_repstat = NFSERR_BAD_COOKIE;
1892                 }
1893 #endif
1894             } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1895                 nd->nd_repstat = NFSERR_BAD_COOKIE;
1896             }
1897         }
1898         if (!nd->nd_repstat && vp->v_type != VDIR)
1899                 nd->nd_repstat = NFSERR_NOTDIR;
1900         if (!nd->nd_repstat && cnt == 0)
1901                 nd->nd_repstat = NFSERR_TOOSMALL;
1902         if (!nd->nd_repstat)
1903                 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1904                     nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1905                     NFSACCCHK_VPISLOCKED, NULL);
1906         if (nd->nd_repstat) {
1907                 vput(vp);
1908                 if (nd->nd_flag & ND_NFSV3)
1909                         nfsrv_postopattr(nd, getret, &at);
1910                 goto out;
1911         }
1912         is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1913         is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
1914
1915         rbuf = malloc(siz, M_TEMP, M_WAITOK);
1916 again:
1917         eofflag = 0;
1918         if (cookies) {
1919                 free(cookies, M_TEMP);
1920                 cookies = NULL;
1921         }
1922
1923         iv.iov_base = rbuf;
1924         iv.iov_len = siz;
1925         io.uio_iov = &iv;
1926         io.uio_iovcnt = 1;
1927         io.uio_offset = (off_t)off;
1928         io.uio_resid = siz;
1929         io.uio_segflg = UIO_SYSSPACE;
1930         io.uio_rw = UIO_READ;
1931         io.uio_td = NULL;
1932         nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1933             &cookies);
1934         off = (u_int64_t)io.uio_offset;
1935         if (io.uio_resid)
1936                 siz -= io.uio_resid;
1937
1938         getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1939
1940         if (!cookies && !nd->nd_repstat)
1941                 nd->nd_repstat = NFSERR_PERM;
1942         if (!nd->nd_repstat)
1943                 nd->nd_repstat = getret;
1944         if (nd->nd_repstat) {
1945                 vput(vp);
1946                 if (cookies)
1947                         free(cookies, M_TEMP);
1948                 free(rbuf, M_TEMP);
1949                 if (nd->nd_flag & ND_NFSV3)
1950                         nfsrv_postopattr(nd, getret, &at);
1951                 goto out;
1952         }
1953         /*
1954          * If nothing read, return eof
1955          * rpc reply
1956          */
1957         if (siz == 0) {
1958                 vput(vp);
1959                 if (nd->nd_flag & ND_NFSV3)
1960                         nfsrv_postopattr(nd, getret, &at);
1961                 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1962                 txdr_hyper(at.na_filerev, tl);
1963                 tl += 2;
1964                 *tl++ = newnfs_false;
1965                 *tl = newnfs_true;
1966                 free(cookies, M_TEMP);
1967                 free(rbuf, M_TEMP);
1968                 goto out;
1969         }
1970
1971         /*
1972          * Check for degenerate cases of nothing useful read.
1973          * If so go try again
1974          */
1975         cpos = rbuf;
1976         cend = rbuf + siz;
1977         dp = (struct dirent *)cpos;
1978         cookiep = cookies;
1979
1980         /*
1981          * For some reason FreeBSD's ufs_readdir() chooses to back the
1982          * directory offset up to a block boundary, so it is necessary to
1983          * skip over the records that precede the requested offset. This
1984          * requires the assumption that file offset cookies monotonically
1985          * increase.
1986          */
1987         while (cpos < cend && ncookies > 0 &&
1988           (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1989            (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
1990            ((nd->nd_flag & ND_NFSV4) &&
1991             ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1992              (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1993                 cpos += dp->d_reclen;
1994                 dp = (struct dirent *)cpos;
1995                 cookiep++;
1996                 ncookies--;
1997         }
1998         if (cpos >= cend || ncookies == 0) {
1999                 siz = fullsiz;
2000                 toff = off;
2001                 goto again;
2002         }
2003
2004         /*
2005          * Busy the file system so that the mount point won't go away
2006          * and, as such, VFS_VGET() can be used safely.
2007          */
2008         mp = vp->v_mount;
2009         vfs_ref(mp);
2010         NFSVOPUNLOCK(vp, 0);
2011         nd->nd_repstat = vfs_busy(mp, 0);
2012         vfs_rel(mp);
2013         if (nd->nd_repstat != 0) {
2014                 vrele(vp);
2015                 free(cookies, M_TEMP);
2016                 free(rbuf, M_TEMP);
2017                 if (nd->nd_flag & ND_NFSV3)
2018                         nfsrv_postopattr(nd, getret, &at);
2019                 goto out;
2020         }
2021
2022         /*
2023          * Check to see if entries in this directory can be safely acquired
2024          * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2025          * ZFS snapshot directories need VOP_LOOKUP(), so that any
2026          * automount of the snapshot directory that is required will
2027          * be done.
2028          * This needs to be done here for NFSv4, since NFSv4 never does
2029          * a VFS_VGET() for "." or "..".
2030          */
2031         if (is_zfs == 1) {
2032                 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2033                 if (r == EOPNOTSUPP) {
2034                         usevget = 0;
2035                         cn.cn_nameiop = LOOKUP;
2036                         cn.cn_lkflags = LK_SHARED | LK_RETRY;
2037                         cn.cn_cred = nd->nd_cred;
2038                         cn.cn_thread = p;
2039                 } else if (r == 0)
2040                         vput(nvp);
2041         }
2042
2043         /*
2044          * Save this position, in case there is an error before one entry
2045          * is created.
2046          */
2047         mb0 = nd->nd_mb;
2048         bpos0 = nd->nd_bpos;
2049
2050         /*
2051          * Fill in the first part of the reply.
2052          * dirlen is the reply length in bytes and cannot exceed cnt.
2053          * (Include the two booleans at the end of the reply in dirlen now,
2054          *  so we recognize when we have exceeded cnt.)
2055          */
2056         if (nd->nd_flag & ND_NFSV3) {
2057                 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2058                 nfsrv_postopattr(nd, getret, &at);
2059         } else {
2060                 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2061         }
2062         NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2063         txdr_hyper(at.na_filerev, tl);
2064
2065         /*
2066          * Save this position, in case there is an empty reply needed.
2067          */
2068         mb1 = nd->nd_mb;
2069         bpos1 = nd->nd_bpos;
2070
2071         /* Loop through the records and build reply */
2072         entrycnt = 0;
2073         while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2074                 nlen = dp->d_namlen;
2075                 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2076                     nlen <= NFS_MAXNAMLEN &&
2077                     ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2078                      (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2079                       || (nlen == 1 && dp->d_name[0] != '.'))) {
2080                         /*
2081                          * Save the current position in the reply, in case
2082                          * this entry exceeds cnt.
2083                          */
2084                         mb1 = nd->nd_mb;
2085                         bpos1 = nd->nd_bpos;
2086
2087                         /*
2088                          * For readdir_and_lookup get the vnode using
2089                          * the file number.
2090                          */
2091                         nvp = NULL;
2092                         refp = NULL;
2093                         r = 0;
2094                         at_root = 0;
2095                         needs_unbusy = 0;
2096                         new_mp = mp;
2097                         mounted_on_fileno = (uint64_t)dp->d_fileno;
2098                         if ((nd->nd_flag & ND_NFSV3) ||
2099                             NFSNONZERO_ATTRBIT(&savbits)) {
2100                                 if (nd->nd_flag & ND_NFSV4)
2101                                         refp = nfsv4root_getreferral(NULL,
2102                                             vp, dp->d_fileno);
2103                                 if (refp == NULL) {
2104                                         if (usevget)
2105                                                 r = VFS_VGET(mp, dp->d_fileno,
2106                                                     LK_SHARED, &nvp);
2107                                         else
2108                                                 r = EOPNOTSUPP;
2109                                         if (r == EOPNOTSUPP) {
2110                                                 if (usevget) {
2111                                                         usevget = 0;
2112                                                         cn.cn_nameiop = LOOKUP;
2113                                                         cn.cn_lkflags =
2114                                                             LK_SHARED |
2115                                                             LK_RETRY;
2116                                                         cn.cn_cred =
2117                                                             nd->nd_cred;
2118                                                         cn.cn_thread = p;
2119                                                 }
2120                                                 cn.cn_nameptr = dp->d_name;
2121                                                 cn.cn_namelen = nlen;
2122                                                 cn.cn_flags = ISLASTCN |
2123                                                     NOFOLLOW | LOCKLEAF;
2124                                                 if (nlen == 2 &&
2125                                                     dp->d_name[0] == '.' &&
2126                                                     dp->d_name[1] == '.')
2127                                                         cn.cn_flags |=
2128                                                             ISDOTDOT;
2129                                                 if (NFSVOPLOCK(vp, LK_SHARED)
2130                                                     != 0) {
2131                                                         nd->nd_repstat = EPERM;
2132                                                         break;
2133                                                 }
2134                                                 if ((vp->v_vflag & VV_ROOT) != 0
2135                                                     && (cn.cn_flags & ISDOTDOT)
2136                                                     != 0) {
2137                                                         vref(vp);
2138                                                         nvp = vp;
2139                                                         r = 0;
2140                                                 } else {
2141                                                         r = VOP_LOOKUP(vp, &nvp,
2142                                                             &cn);
2143                                                         if (vp != nvp)
2144                                                                 NFSVOPUNLOCK(vp,
2145                                                                     0);
2146                                                 }
2147                                         }
2148
2149                                         /*
2150                                          * For NFSv4, check to see if nvp is
2151                                          * a mount point and get the mount
2152                                          * point vnode, as required.
2153                                          */
2154                                         if (r == 0 &&
2155                                             nfsrv_enable_crossmntpt != 0 &&
2156                                             (nd->nd_flag & ND_NFSV4) != 0 &&
2157                                             nvp->v_type == VDIR &&
2158                                             nvp->v_mountedhere != NULL) {
2159                                                 new_mp = nvp->v_mountedhere;
2160                                                 r = vfs_busy(new_mp, 0);
2161                                                 vput(nvp);
2162                                                 nvp = NULL;
2163                                                 if (r == 0) {
2164                                                         r = VFS_ROOT(new_mp,
2165                                                             LK_SHARED, &nvp);
2166                                                         needs_unbusy = 1;
2167                                                         if (r == 0)
2168                                                                 at_root = 1;
2169                                                 }
2170                                         }
2171                                 }
2172                                 if (!r) {
2173                                     if (refp == NULL &&
2174                                         ((nd->nd_flag & ND_NFSV3) ||
2175                                          NFSNONZERO_ATTRBIT(&attrbits))) {
2176                                         r = nfsvno_getfh(nvp, &nfh, p);
2177                                         if (!r)
2178                                             r = nfsvno_getattr(nvp, nvap,
2179                                                 nd->nd_cred, p, 1);
2180                                         if (r == 0 && is_zfs == 1 &&
2181                                             nfsrv_enable_crossmntpt != 0 &&
2182                                             (nd->nd_flag & ND_NFSV4) != 0 &&
2183                                             nvp->v_type == VDIR &&
2184                                             vp->v_mount != nvp->v_mount) {
2185                                             /*
2186                                              * For a ZFS snapshot, there is a
2187                                              * pseudo mount that does not set
2188                                              * v_mountedhere, so it needs to
2189                                              * be detected via a different
2190                                              * mount structure.
2191                                              */
2192                                             at_root = 1;
2193                                             if (new_mp == mp)
2194                                                 new_mp = nvp->v_mount;
2195                                         }
2196                                     }
2197                                 } else {
2198                                     nvp = NULL;
2199                                 }
2200                                 if (r) {
2201                                         if (!NFSISSET_ATTRBIT(&attrbits,
2202                                             NFSATTRBIT_RDATTRERROR)) {
2203                                                 if (nvp != NULL)
2204                                                         vput(nvp);
2205                                                 if (needs_unbusy != 0)
2206                                                         vfs_unbusy(new_mp);
2207                                                 nd->nd_repstat = r;
2208                                                 break;
2209                                         }
2210                                 }
2211                         }
2212
2213                         /*
2214                          * Build the directory record xdr
2215                          */
2216                         if (nd->nd_flag & ND_NFSV3) {
2217                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2218                                 *tl++ = newnfs_true;
2219                                 *tl++ = 0;
2220                                 *tl = txdr_unsigned(dp->d_fileno);
2221                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2222                                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2223                                 *tl++ = 0;
2224                                 *tl = txdr_unsigned(*cookiep);
2225                                 nfsrv_postopattr(nd, 0, nvap);
2226                                 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2227                                 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2228                                 if (nvp != NULL)
2229                                         vput(nvp);
2230                         } else {
2231                                 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2232                                 *tl++ = newnfs_true;
2233                                 *tl++ = 0;
2234                                 *tl = txdr_unsigned(*cookiep);
2235                                 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2236                                 if (nvp != NULL) {
2237                                         supports_nfsv4acls =
2238                                             nfs_supportsnfsv4acls(nvp);
2239                                         NFSVOPUNLOCK(nvp, 0);
2240                                 } else
2241                                         supports_nfsv4acls = 0;
2242                                 if (refp != NULL) {
2243                                         dirlen += nfsrv_putreferralattr(nd,
2244                                             &savbits, refp, 0,
2245                                             &nd->nd_repstat);
2246                                         if (nd->nd_repstat) {
2247                                                 if (nvp != NULL)
2248                                                         vrele(nvp);
2249                                                 if (needs_unbusy != 0)
2250                                                         vfs_unbusy(new_mp);
2251                                                 break;
2252                                         }
2253                                 } else if (r) {
2254                                         dirlen += nfsvno_fillattr(nd, new_mp,
2255                                             nvp, nvap, &nfh, r, &rderrbits,
2256                                             nd->nd_cred, p, isdgram, 0,
2257                                             supports_nfsv4acls, at_root,
2258                                             mounted_on_fileno);
2259                                 } else {
2260                                         dirlen += nfsvno_fillattr(nd, new_mp,
2261                                             nvp, nvap, &nfh, r, &attrbits,
2262                                             nd->nd_cred, p, isdgram, 0,
2263                                             supports_nfsv4acls, at_root,
2264                                             mounted_on_fileno);
2265                                 }
2266                                 if (nvp != NULL)
2267                                         vrele(nvp);
2268                                 dirlen += (3 * NFSX_UNSIGNED);
2269                         }
2270                         if (needs_unbusy != 0)
2271                                 vfs_unbusy(new_mp);
2272                         if (dirlen <= cnt)
2273                                 entrycnt++;
2274                 }
2275                 cpos += dp->d_reclen;
2276                 dp = (struct dirent *)cpos;
2277                 cookiep++;
2278                 ncookies--;
2279         }
2280         vrele(vp);
2281         vfs_unbusy(mp);
2282
2283         /*
2284          * If dirlen > cnt, we must strip off the last entry. If that
2285          * results in an empty reply, report NFSERR_TOOSMALL.
2286          */
2287         if (dirlen > cnt || nd->nd_repstat) {
2288                 if (!nd->nd_repstat && entrycnt == 0)
2289                         nd->nd_repstat = NFSERR_TOOSMALL;
2290                 if (nd->nd_repstat) {
2291                         newnfs_trimtrailing(nd, mb0, bpos0);
2292                         if (nd->nd_flag & ND_NFSV3)
2293                                 nfsrv_postopattr(nd, getret, &at);
2294                 } else
2295                         newnfs_trimtrailing(nd, mb1, bpos1);
2296                 eofflag = 0;
2297         } else if (cpos < cend)
2298                 eofflag = 0;
2299         if (!nd->nd_repstat) {
2300                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2301                 *tl++ = newnfs_false;
2302                 if (eofflag)
2303                         *tl = newnfs_true;
2304                 else
2305                         *tl = newnfs_false;
2306         }
2307         free(cookies, M_TEMP);
2308         free(rbuf, M_TEMP);
2309
2310 out:
2311         NFSEXITCODE2(0, nd);
2312         return (0);
2313 nfsmout:
2314         vput(vp);
2315         NFSEXITCODE2(error, nd);
2316         return (error);
2317 }
2318
2319 /*
2320  * Get the settable attributes out of the mbuf list.
2321  * (Return 0 or EBADRPC)
2322  */
2323 int
2324 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2325     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2326 {
2327         u_int32_t *tl;
2328         struct nfsv2_sattr *sp;
2329         int error = 0, toclient = 0;
2330
2331         switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2332         case ND_NFSV2:
2333                 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2334                 /*
2335                  * Some old clients didn't fill in the high order 16bits.
2336                  * --> check the low order 2 bytes for 0xffff
2337                  */
2338                 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2339                         nvap->na_mode = nfstov_mode(sp->sa_mode);
2340                 if (sp->sa_uid != newnfs_xdrneg1)
2341                         nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2342                 if (sp->sa_gid != newnfs_xdrneg1)
2343                         nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2344                 if (sp->sa_size != newnfs_xdrneg1)
2345                         nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2346                 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2347 #ifdef notyet
2348                         fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2349 #else
2350                         nvap->na_atime.tv_sec =
2351                                 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2352                         nvap->na_atime.tv_nsec = 0;
2353 #endif
2354                 }
2355                 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2356                         fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2357                 break;
2358         case ND_NFSV3:
2359                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2360                 if (*tl == newnfs_true) {
2361                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2362                         nvap->na_mode = nfstov_mode(*tl);
2363                 }
2364                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2365                 if (*tl == newnfs_true) {
2366                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2367                         nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2368                 }
2369                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2370                 if (*tl == newnfs_true) {
2371                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2372                         nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2373                 }
2374                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2375                 if (*tl == newnfs_true) {
2376                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2377                         nvap->na_size = fxdr_hyper(tl);
2378                 }
2379                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2380                 switch (fxdr_unsigned(int, *tl)) {
2381                 case NFSV3SATTRTIME_TOCLIENT:
2382                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2383                         fxdr_nfsv3time(tl, &nvap->na_atime);
2384                         toclient = 1;
2385                         break;
2386                 case NFSV3SATTRTIME_TOSERVER:
2387                         vfs_timestamp(&nvap->na_atime);
2388                         nvap->na_vaflags |= VA_UTIMES_NULL;
2389                         break;
2390                 }
2391                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2392                 switch (fxdr_unsigned(int, *tl)) {
2393                 case NFSV3SATTRTIME_TOCLIENT:
2394                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2395                         fxdr_nfsv3time(tl, &nvap->na_mtime);
2396                         nvap->na_vaflags &= ~VA_UTIMES_NULL;
2397                         break;
2398                 case NFSV3SATTRTIME_TOSERVER:
2399                         vfs_timestamp(&nvap->na_mtime);
2400                         if (!toclient)
2401                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2402                         break;
2403                 }
2404                 break;
2405         case ND_NFSV4:
2406                 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2407         }
2408 nfsmout:
2409         NFSEXITCODE2(error, nd);
2410         return (error);
2411 }
2412
2413 /*
2414  * Handle the setable attributes for V4.
2415  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2416  */
2417 int
2418 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2419     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2420 {
2421         u_int32_t *tl;
2422         int attrsum = 0;
2423         int i, j;
2424         int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2425         int toclient = 0;
2426         u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2427         uid_t uid;
2428         gid_t gid;
2429
2430         error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2431         if (error)
2432                 goto nfsmout;
2433         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2434         attrsize = fxdr_unsigned(int, *tl);
2435
2436         /*
2437          * Loop around getting the setable attributes. If an unsupported
2438          * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2439          */
2440         if (retnotsup) {
2441                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2442                 bitpos = NFSATTRBIT_MAX;
2443         } else {
2444                 bitpos = 0;
2445         }
2446         for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2447             if (attrsum > attrsize) {
2448                 error = NFSERR_BADXDR;
2449                 goto nfsmout;
2450             }
2451             if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2452                 switch (bitpos) {
2453                 case NFSATTRBIT_SIZE:
2454                         NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2455                      if (vp != NULL && vp->v_type != VREG) {
2456                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2457                                 NFSERR_INVAL;
2458                             goto nfsmout;
2459                         }
2460                         nvap->na_size = fxdr_hyper(tl);
2461                         attrsum += NFSX_HYPER;
2462                         break;
2463                 case NFSATTRBIT_ACL:
2464                         error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2465                             p);
2466                         if (error)
2467                                 goto nfsmout;
2468                         if (aceerr && !nd->nd_repstat)
2469                                 nd->nd_repstat = aceerr;
2470                         attrsum += aclsize;
2471                         break;
2472                 case NFSATTRBIT_ARCHIVE:
2473                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2474                         if (!nd->nd_repstat)
2475                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2476                         attrsum += NFSX_UNSIGNED;
2477                         break;
2478                 case NFSATTRBIT_HIDDEN:
2479                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2480                         if (!nd->nd_repstat)
2481                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2482                         attrsum += NFSX_UNSIGNED;
2483                         break;
2484                 case NFSATTRBIT_MIMETYPE:
2485                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2486                         i = fxdr_unsigned(int, *tl);
2487                         error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2488                         if (error)
2489                                 goto nfsmout;
2490                         if (!nd->nd_repstat)
2491                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2492                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2493                         break;
2494                 case NFSATTRBIT_MODE:
2495                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2496                         nvap->na_mode = nfstov_mode(*tl);
2497                         attrsum += NFSX_UNSIGNED;
2498                         break;
2499                 case NFSATTRBIT_OWNER:
2500                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2501                         j = fxdr_unsigned(int, *tl);
2502                         if (j < 0) {
2503                                 error = NFSERR_BADXDR;
2504                                 goto nfsmout;
2505                         }
2506                         if (j > NFSV4_SMALLSTR)
2507                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2508                         else
2509                                 cp = namestr;
2510                         error = nfsrv_mtostr(nd, cp, j);
2511                         if (error) {
2512                                 if (j > NFSV4_SMALLSTR)
2513                                         free(cp, M_NFSSTRING);
2514                                 goto nfsmout;
2515                         }
2516                         if (!nd->nd_repstat) {
2517                                 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2518                                     p);
2519                                 if (!nd->nd_repstat)
2520                                         nvap->na_uid = uid;
2521                         }
2522                         if (j > NFSV4_SMALLSTR)
2523                                 free(cp, M_NFSSTRING);
2524                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2525                         break;
2526                 case NFSATTRBIT_OWNERGROUP:
2527                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2528                         j = fxdr_unsigned(int, *tl);
2529                         if (j < 0) {
2530                                 error = NFSERR_BADXDR;
2531                                 goto nfsmout;
2532                         }
2533                         if (j > NFSV4_SMALLSTR)
2534                                 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2535                         else
2536                                 cp = namestr;
2537                         error = nfsrv_mtostr(nd, cp, j);
2538                         if (error) {
2539                                 if (j > NFSV4_SMALLSTR)
2540                                         free(cp, M_NFSSTRING);
2541                                 goto nfsmout;
2542                         }
2543                         if (!nd->nd_repstat) {
2544                                 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2545                                     p);
2546                                 if (!nd->nd_repstat)
2547                                         nvap->na_gid = gid;
2548                         }
2549                         if (j > NFSV4_SMALLSTR)
2550                                 free(cp, M_NFSSTRING);
2551                         attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2552                         break;
2553                 case NFSATTRBIT_SYSTEM:
2554                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2555                         if (!nd->nd_repstat)
2556                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2557                         attrsum += NFSX_UNSIGNED;
2558                         break;
2559                 case NFSATTRBIT_TIMEACCESSSET:
2560                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2561                         attrsum += NFSX_UNSIGNED;
2562                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2563                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2564                             fxdr_nfsv4time(tl, &nvap->na_atime);
2565                             toclient = 1;
2566                             attrsum += NFSX_V4TIME;
2567                         } else {
2568                             vfs_timestamp(&nvap->na_atime);
2569                             nvap->na_vaflags |= VA_UTIMES_NULL;
2570                         }
2571                         break;
2572                 case NFSATTRBIT_TIMEBACKUP:
2573                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2574                         if (!nd->nd_repstat)
2575                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2576                         attrsum += NFSX_V4TIME;
2577                         break;
2578                 case NFSATTRBIT_TIMECREATE:
2579                         NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2580                         if (!nd->nd_repstat)
2581                                 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2582                         attrsum += NFSX_V4TIME;
2583                         break;
2584                 case NFSATTRBIT_TIMEMODIFYSET:
2585                         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2586                         attrsum += NFSX_UNSIGNED;
2587                         if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2588                             NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2589                             fxdr_nfsv4time(tl, &nvap->na_mtime);
2590                             nvap->na_vaflags &= ~VA_UTIMES_NULL;
2591                             attrsum += NFSX_V4TIME;
2592                         } else {
2593                             vfs_timestamp(&nvap->na_mtime);
2594                             if (!toclient)
2595                                 nvap->na_vaflags |= VA_UTIMES_NULL;
2596                         }
2597                         break;
2598                 default:
2599                         nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2600                         /*
2601                          * set bitpos so we drop out of the loop.
2602                          */
2603                         bitpos = NFSATTRBIT_MAX;
2604                         break;
2605                 }
2606         }
2607
2608         /*
2609          * some clients pad the attrlist, so we need to skip over the
2610          * padding.
2611          */
2612         if (attrsum > attrsize) {
2613                 error = NFSERR_BADXDR;
2614         } else {
2615                 attrsize = NFSM_RNDUP(attrsize);
2616                 if (attrsum < attrsize)
2617                         error = nfsm_advance(nd, attrsize - attrsum, -1);
2618         }
2619 nfsmout:
2620         NFSEXITCODE2(error, nd);
2621         return (error);
2622 }
2623
2624 /*
2625  * Check/setup export credentials.
2626  */
2627 int
2628 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2629     struct ucred *credanon)
2630 {
2631         int error = 0;
2632
2633         /*
2634          * Check/setup credentials.
2635          */
2636         if (nd->nd_flag & ND_GSS)
2637                 exp->nes_exflag &= ~MNT_EXPORTANON;
2638
2639         /*
2640          * Check to see if the operation is allowed for this security flavor.
2641          * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2642          * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2643          * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2644          */
2645         if (nfsvno_testexp(nd, exp) &&
2646             nd->nd_procnum != NFSV4OP_SECINFO &&
2647             nd->nd_procnum != NFSPROC_FSINFO) {
2648                 if (nd->nd_flag & ND_NFSV4)
2649                         error = NFSERR_WRONGSEC;
2650                 else
2651                         error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2652                 goto out;
2653         }
2654
2655         /*
2656          * Check to see if the file system is exported V4 only.
2657          */
2658         if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2659                 error = NFSERR_PROGNOTV4;
2660                 goto out;
2661         }
2662
2663         /*
2664          * Now, map the user credentials.
2665          * (Note that ND_AUTHNONE will only be set for an NFSv3
2666          *  Fsinfo RPC. If set for anything else, this code might need
2667          *  to change.)
2668          */
2669         if (NFSVNO_EXPORTED(exp)) {
2670                 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
2671                      NFSVNO_EXPORTANON(exp) ||
2672                      (nd->nd_flag & ND_AUTHNONE) != 0) {
2673                         nd->nd_cred->cr_uid = credanon->cr_uid;
2674                         nd->nd_cred->cr_gid = credanon->cr_gid;
2675                         crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2676                             credanon->cr_groups);
2677                 } else if ((nd->nd_flag & ND_GSS) == 0) {
2678                         /*
2679                          * If using AUTH_SYS, call nfsrv_getgrpscred() to see
2680                          * if there is a replacement credential with a group
2681                          * list set up by "nfsuserd -manage-gids".
2682                          * If there is no replacement, nfsrv_getgrpscred()
2683                          * simply returns its argument.
2684                          */
2685                         nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
2686                 }
2687         }
2688
2689 out:
2690         NFSEXITCODE2(error, nd);
2691         return (error);
2692 }
2693
2694 /*
2695  * Check exports.
2696  */
2697 int
2698 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2699     struct ucred **credp)
2700 {
2701         int i, error, *secflavors;
2702
2703         error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2704             &exp->nes_numsecflavor, &secflavors);
2705         if (error) {
2706                 if (nfs_rootfhset) {
2707                         exp->nes_exflag = 0;
2708                         exp->nes_numsecflavor = 0;
2709                         error = 0;
2710                 }
2711         } else {
2712                 /* Copy the security flavors. */
2713                 for (i = 0; i < exp->nes_numsecflavor; i++)
2714                         exp->nes_secflavors[i] = secflavors[i];
2715         }
2716         NFSEXITCODE(error);
2717         return (error);
2718 }
2719
2720 /*
2721  * Get a vnode for a file handle and export stuff.
2722  */
2723 int
2724 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2725     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2726     struct ucred **credp)
2727 {
2728         int i, error, *secflavors;
2729
2730         *credp = NULL;
2731         exp->nes_numsecflavor = 0;
2732         error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2733         if (error != 0)
2734                 /* Make sure the server replies ESTALE to the client. */
2735                 error = ESTALE;
2736         if (nam && !error) {
2737                 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2738                     &exp->nes_numsecflavor, &secflavors);
2739                 if (error) {
2740                         if (nfs_rootfhset) {
2741                                 exp->nes_exflag = 0;
2742                                 exp->nes_numsecflavor = 0;
2743                                 error = 0;
2744                         } else {
2745                                 vput(*vpp);
2746                         }
2747                 } else {
2748                         /* Copy the security flavors. */
2749                         for (i = 0; i < exp->nes_numsecflavor; i++)
2750                                 exp->nes_secflavors[i] = secflavors[i];
2751                 }
2752         }
2753         NFSEXITCODE(error);
2754         return (error);
2755 }
2756
2757 /*
2758  * nfsd_fhtovp() - convert a fh to a vnode ptr
2759  *      - look up fsid in mount list (if not found ret error)
2760  *      - get vp and export rights by calling nfsvno_fhtovp()
2761  *      - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2762  *        for AUTH_SYS
2763  *      - if mpp != NULL, return the mount point so that it can
2764  *        be used for vn_finished_write() by the caller
2765  */
2766 void
2767 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2768     struct vnode **vpp, struct nfsexstuff *exp,
2769     struct mount **mpp, int startwrite, struct thread *p)
2770 {
2771         struct mount *mp;
2772         struct ucred *credanon;
2773         fhandle_t *fhp;
2774
2775         fhp = (fhandle_t *)nfp->nfsrvfh_data;
2776         /*
2777          * Check for the special case of the nfsv4root_fh.
2778          */
2779         mp = vfs_busyfs(&fhp->fh_fsid);
2780         if (mpp != NULL)
2781                 *mpp = mp;
2782         if (mp == NULL) {
2783                 *vpp = NULL;
2784                 nd->nd_repstat = ESTALE;
2785                 goto out;
2786         }
2787
2788         if (startwrite) {
2789                 vn_start_write(NULL, mpp, V_WAIT);
2790                 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2791                         lktype = LK_EXCLUSIVE;
2792         }
2793         nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2794             &credanon);
2795         vfs_unbusy(mp);
2796
2797         /*
2798          * For NFSv4 without a pseudo root fs, unexported file handles
2799          * can be returned, so that Lookup works everywhere.
2800          */
2801         if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2802             !(nd->nd_flag & ND_NFSV4)) {
2803                 vput(*vpp);
2804                 nd->nd_repstat = EACCES;
2805         }
2806
2807         /*
2808          * Personally, I've never seen any point in requiring a
2809          * reserved port#, since only in the rare case where the
2810          * clients are all boxes with secure system privileges,
2811          * does it provide any enhanced security, but... some people
2812          * believe it to be useful and keep putting this code back in.
2813          * (There is also some "security checker" out there that
2814          *  complains if the nfs server doesn't enforce this.)
2815          * However, note the following:
2816          * RFC3530 (NFSv4) specifies that a reserved port# not be
2817          *      required.
2818          * RFC2623 recommends that, if a reserved port# is checked for,
2819          *      that there be a way to turn that off--> ifdef'd.
2820          */
2821 #ifdef NFS_REQRSVPORT
2822         if (!nd->nd_repstat) {
2823                 struct sockaddr_in *saddr;
2824                 struct sockaddr_in6 *saddr6;
2825
2826                 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2827                 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2828                 if (!(nd->nd_flag & ND_NFSV4) &&
2829                     ((saddr->sin_family == AF_INET &&
2830                       ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2831                      (saddr6->sin6_family == AF_INET6 &&
2832                       ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2833                         vput(*vpp);
2834                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2835                 }
2836         }
2837 #endif  /* NFS_REQRSVPORT */
2838
2839         /*
2840          * Check/setup credentials.
2841          */
2842         if (!nd->nd_repstat) {
2843                 nd->nd_saveduid = nd->nd_cred->cr_uid;
2844                 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2845                 if (nd->nd_repstat)
2846                         vput(*vpp);
2847         }
2848         if (credanon != NULL)
2849                 crfree(credanon);
2850         if (nd->nd_repstat) {
2851                 if (startwrite)
2852                         vn_finished_write(mp);
2853                 *vpp = NULL;
2854                 if (mpp != NULL)
2855                         *mpp = NULL;
2856         }
2857
2858 out:
2859         NFSEXITCODE2(0, nd);
2860 }
2861
2862 /*
2863  * glue for fp.
2864  */
2865 static int
2866 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2867 {
2868         struct filedesc *fdp;
2869         struct file *fp;
2870         int error = 0;
2871
2872         fdp = p->td_proc->p_fd;
2873         if (fd < 0 || fd >= fdp->fd_nfiles ||
2874             (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2875                 error = EBADF;
2876                 goto out;
2877         }
2878         *fpp = fp;
2879
2880 out:
2881         NFSEXITCODE(error);
2882         return (error);
2883 }
2884
2885 /*
2886  * Called from nfssvc() to update the exports list. Just call
2887  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2888  * in the mount list.
2889  */
2890 int
2891 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2892 {
2893         struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2894         int error = 0;
2895         struct nameidata nd;
2896         fhandle_t fh;
2897
2898         error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2899         if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2900                 nfs_rootfhset = 0;
2901         else if (error == 0) {
2902                 if (nfsexargp->fspec == NULL) {
2903                         error = EPERM;
2904                         goto out;
2905                 }
2906                 /*
2907                  * If fspec != NULL, this is the v4root path.
2908                  */
2909                 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2910                     nfsexargp->fspec, p);
2911                 if ((error = namei(&nd)) != 0)
2912                         goto out;
2913                 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2914                 vrele(nd.ni_vp);
2915                 if (!error) {
2916                         nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2917                         NFSBCOPY((caddr_t)&fh,
2918                             nfs_rootfh.nfsrvfh_data,
2919                             sizeof (fhandle_t));
2920                         nfs_rootfhset = 1;
2921                 }
2922         }
2923
2924 out:
2925         NFSEXITCODE(error);
2926         return (error);
2927 }
2928
2929 /*
2930  * This function needs to test to see if the system is near its limit
2931  * for memory allocation via malloc() or mget() and return True iff
2932  * either of these resources are near their limit.
2933  * XXX (For now, this is just a stub.)
2934  */
2935 int nfsrv_testmalloclimit = 0;
2936 int
2937 nfsrv_mallocmget_limit(void)
2938 {
2939         static int printmesg = 0;
2940         static int testval = 1;
2941
2942         if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2943                 if ((printmesg++ % 100) == 0)
2944                         printf("nfsd: malloc/mget near limit\n");
2945                 return (1);
2946         }
2947         return (0);
2948 }
2949
2950 /*
2951  * BSD specific initialization of a mount point.
2952  */
2953 void
2954 nfsd_mntinit(void)
2955 {
2956         static int inited = 0;
2957
2958         if (inited)
2959                 return;
2960         inited = 1;
2961         nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2962         TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2963         TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2964         nfsv4root_mnt.mnt_export = NULL;
2965         TAILQ_INIT(&nfsv4root_opt);
2966         TAILQ_INIT(&nfsv4root_newopt);
2967         nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2968         nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2969         nfsv4root_mnt.mnt_nvnodelistsize = 0;
2970         nfsv4root_mnt.mnt_activevnodelistsize = 0;
2971 }
2972
2973 /*
2974  * Get a vnode for a file handle, without checking exports, etc.
2975  */
2976 struct vnode *
2977 nfsvno_getvp(fhandle_t *fhp)
2978 {
2979         struct mount *mp;
2980         struct vnode *vp;
2981         int error;
2982
2983         mp = vfs_busyfs(&fhp->fh_fsid);
2984         if (mp == NULL)
2985                 return (NULL);
2986         error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2987         vfs_unbusy(mp);
2988         if (error)
2989                 return (NULL);
2990         return (vp);
2991 }
2992
2993 /*
2994  * Do a local VOP_ADVLOCK().
2995  */
2996 int
2997 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2998     u_int64_t end, struct thread *td)
2999 {
3000         int error = 0;
3001         struct flock fl;
3002         u_int64_t tlen;
3003
3004         if (nfsrv_dolocallocks == 0)
3005                 goto out;
3006         ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3007
3008         fl.l_whence = SEEK_SET;
3009         fl.l_type = ftype;
3010         fl.l_start = (off_t)first;
3011         if (end == NFS64BITSSET) {
3012                 fl.l_len = 0;
3013         } else {
3014                 tlen = end - first;
3015                 fl.l_len = (off_t)tlen;
3016         }
3017         /*
3018          * For FreeBSD8, the l_pid and l_sysid must be set to the same
3019          * values for all calls, so that all locks will be held by the
3020          * nfsd server. (The nfsd server handles conflicts between the
3021          * various clients.)
3022          * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3023          * bytes, so it can't be put in l_sysid.
3024          */
3025         if (nfsv4_sysid == 0)
3026                 nfsv4_sysid = nlm_acquire_next_sysid();
3027         fl.l_pid = (pid_t)0;
3028         fl.l_sysid = (int)nfsv4_sysid;
3029
3030         if (ftype == F_UNLCK)
3031                 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3032                     (F_POSIX | F_REMOTE));
3033         else
3034                 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3035                     (F_POSIX | F_REMOTE));
3036
3037 out:
3038         NFSEXITCODE(error);
3039         return (error);
3040 }
3041
3042 /*
3043  * Check the nfsv4 root exports.
3044  */
3045 int
3046 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3047 {
3048         struct ucred *credanon;
3049         int exflags, error = 0, numsecflavor, *secflavors, i;
3050
3051         error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3052             &credanon, &numsecflavor, &secflavors);
3053         if (error) {
3054                 error = NFSERR_PROGUNAVAIL;
3055                 goto out;
3056         }
3057         if (credanon != NULL)
3058                 crfree(credanon);
3059         for (i = 0; i < numsecflavor; i++) {
3060                 if (secflavors[i] == AUTH_SYS)
3061                         nd->nd_flag |= ND_EXAUTHSYS;
3062                 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3063                         nd->nd_flag |= ND_EXGSS;
3064                 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3065                         nd->nd_flag |= ND_EXGSSINTEGRITY;
3066                 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3067                         nd->nd_flag |= ND_EXGSSPRIVACY;
3068         }
3069
3070 out:
3071         NFSEXITCODE(error);
3072         return (error);
3073 }
3074
3075 /*
3076  * Nfs server pseudo system call for the nfsd's
3077  */
3078 /*
3079  * MPSAFE
3080  */
3081 static int
3082 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3083 {
3084         struct file *fp;
3085         struct nfsd_addsock_args sockarg;
3086         struct nfsd_nfsd_args nfsdarg;
3087         cap_rights_t rights;
3088         int error;
3089
3090         if (uap->flag & NFSSVC_NFSDADDSOCK) {
3091                 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3092                 if (error)
3093                         goto out;
3094                 /*
3095                  * Since we don't know what rights might be required,
3096                  * pretend that we need them all. It is better to be too
3097                  * careful than too reckless.
3098                  */
3099                 error = fget(td, sockarg.sock,
3100                     cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3101                 if (error != 0)
3102                         goto out;
3103                 if (fp->f_type != DTYPE_SOCKET) {
3104                         fdrop(fp, td);
3105                         error = EPERM;
3106                         goto out;
3107                 }
3108                 error = nfsrvd_addsock(fp);
3109                 fdrop(fp, td);
3110         } else if (uap->flag & NFSSVC_NFSDNFSD) {
3111                 if (uap->argp == NULL) {
3112                         error = EINVAL;
3113                         goto out;
3114                 }
3115                 error = copyin(uap->argp, (caddr_t)&nfsdarg,
3116                     sizeof (nfsdarg));
3117                 if (error)
3118                         goto out;
3119                 error = nfsrvd_nfsd(td, &nfsdarg);
3120         } else {
3121                 error = nfssvc_srvcall(td, uap, td->td_ucred);
3122         }
3123
3124 out:
3125         NFSEXITCODE(error);
3126         return (error);
3127 }
3128
3129 static int
3130 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3131 {
3132         struct nfsex_args export;
3133         struct file *fp = NULL;
3134         int stablefd, len;
3135         struct nfsd_clid adminrevoke;
3136         struct nfsd_dumplist dumplist;
3137         struct nfsd_dumpclients *dumpclients;
3138         struct nfsd_dumplocklist dumplocklist;
3139         struct nfsd_dumplocks *dumplocks;
3140         struct nameidata nd;
3141         vnode_t vp;
3142         int error = EINVAL, igotlock;
3143         struct proc *procp;
3144         static int suspend_nfsd = 0;
3145
3146         if (uap->flag & NFSSVC_PUBLICFH) {
3147                 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3148                     sizeof (fhandle_t));
3149                 error = copyin(uap->argp,
3150                     &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3151                 if (!error)
3152                         nfs_pubfhset = 1;
3153         } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3154                 error = copyin(uap->argp,(caddr_t)&export,
3155                     sizeof (struct nfsex_args));
3156                 if (!error)
3157                         error = nfsrv_v4rootexport(&export, cred, p);
3158         } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3159                 nfs_pubfhset = 0;
3160                 error = 0;
3161         } else if (uap->flag & NFSSVC_STABLERESTART) {
3162                 error = copyin(uap->argp, (caddr_t)&stablefd,
3163                     sizeof (int));
3164                 if (!error)
3165                         error = fp_getfvp(p, stablefd, &fp, &vp);
3166                 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3167                         error = EBADF;
3168                 if (!error && newnfs_numnfsd != 0)
3169                         error = EPERM;
3170                 if (!error) {
3171                         nfsrv_stablefirst.nsf_fp = fp;
3172                         nfsrv_setupstable(p);
3173                 }
3174         } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3175                 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3176                     sizeof (struct nfsd_clid));
3177                 if (!error)
3178                         error = nfsrv_adminrevoke(&adminrevoke, p);
3179         } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3180                 error = copyin(uap->argp, (caddr_t)&dumplist,
3181                     sizeof (struct nfsd_dumplist));
3182                 if (!error && (dumplist.ndl_size < 1 ||
3183                         dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3184                         error = EPERM;
3185                 if (!error) {
3186                     len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3187                     dumpclients = (struct nfsd_dumpclients *)malloc(len,
3188                         M_TEMP, M_WAITOK);
3189                     nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3190                     error = copyout(dumpclients,
3191                         CAST_USER_ADDR_T(dumplist.ndl_list), len);
3192                     free(dumpclients, M_TEMP);
3193                 }
3194         } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3195                 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3196                     sizeof (struct nfsd_dumplocklist));
3197                 if (!error && (dumplocklist.ndllck_size < 1 ||
3198                         dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3199                         error = EPERM;
3200                 if (!error)
3201                         error = nfsrv_lookupfilename(&nd,
3202                                 dumplocklist.ndllck_fname, p);
3203                 if (!error) {
3204                         len = sizeof (struct nfsd_dumplocks) *
3205                                 dumplocklist.ndllck_size;
3206                         dumplocks = (struct nfsd_dumplocks *)malloc(len,
3207                                 M_TEMP, M_WAITOK);
3208                         nfsrv_dumplocks(nd.ni_vp, dumplocks,
3209                             dumplocklist.ndllck_size, p);
3210                         vput(nd.ni_vp);
3211                         error = copyout(dumplocks,
3212                             CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3213                         free(dumplocks, M_TEMP);
3214                 }
3215         } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3216                 procp = p->td_proc;
3217                 PROC_LOCK(procp);
3218                 nfsd_master_pid = procp->p_pid;
3219                 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3220                 nfsd_master_start = procp->p_stats->p_start;
3221                 nfsd_master_proc = procp;
3222                 PROC_UNLOCK(procp);
3223         } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3224                 NFSLOCKV4ROOTMUTEX();
3225                 if (suspend_nfsd == 0) {
3226                         /* Lock out all nfsd threads */
3227                         do {
3228                                 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3229                                     NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3230                         } while (igotlock == 0 && suspend_nfsd == 0);
3231                         suspend_nfsd = 1;
3232                 }
3233                 NFSUNLOCKV4ROOTMUTEX();
3234                 error = 0;
3235         } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3236                 NFSLOCKV4ROOTMUTEX();
3237                 if (suspend_nfsd != 0) {
3238                         nfsv4_unlock(&nfsd_suspend_lock, 0);
3239                         suspend_nfsd = 0;
3240                 }
3241                 NFSUNLOCKV4ROOTMUTEX();
3242                 error = 0;
3243         }
3244
3245         NFSEXITCODE(error);
3246         return (error);
3247 }
3248
3249 /*
3250  * Check exports.
3251  * Returns 0 if ok, 1 otherwise.
3252  */
3253 int
3254 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3255 {
3256         int i;
3257
3258         /*
3259          * This seems odd, but allow the case where the security flavor
3260          * list is empty. This happens when NFSv4 is traversing non-exported
3261          * file systems. Exported file systems should always have a non-empty
3262          * security flavor list.
3263          */
3264         if (exp->nes_numsecflavor == 0)
3265                 return (0);
3266
3267         for (i = 0; i < exp->nes_numsecflavor; i++) {
3268                 /*
3269                  * The tests for privacy and integrity must be first,
3270                  * since ND_GSS is set for everything but AUTH_SYS.
3271                  */
3272                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3273                     (nd->nd_flag & ND_GSSPRIVACY))
3274                         return (0);
3275                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3276                     (nd->nd_flag & ND_GSSINTEGRITY))
3277                         return (0);
3278                 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3279                     (nd->nd_flag & ND_GSS))
3280                         return (0);
3281                 if (exp->nes_secflavors[i] == AUTH_SYS &&
3282                     (nd->nd_flag & ND_GSS) == 0)
3283                         return (0);
3284         }
3285         return (1);
3286 }
3287
3288 /*
3289  * Calculate a hash value for the fid in a file handle.
3290  */
3291 uint32_t
3292 nfsrv_hashfh(fhandle_t *fhp)
3293 {
3294         uint32_t hashval;
3295
3296         hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3297         return (hashval);
3298 }
3299
3300 /*
3301  * Calculate a hash value for the sessionid.
3302  */
3303 uint32_t
3304 nfsrv_hashsessionid(uint8_t *sessionid)
3305 {
3306         uint32_t hashval;
3307
3308         hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3309         return (hashval);
3310 }
3311
3312 /*
3313  * Signal the userland master nfsd to backup the stable restart file.
3314  */
3315 void
3316 nfsrv_backupstable(void)
3317 {
3318         struct proc *procp;
3319
3320         if (nfsd_master_proc != NULL) {
3321                 procp = pfind(nfsd_master_pid);
3322                 /* Try to make sure it is the correct process. */
3323                 if (procp == nfsd_master_proc &&
3324                     procp->p_stats->p_start.tv_sec ==
3325                     nfsd_master_start.tv_sec &&
3326                     procp->p_stats->p_start.tv_usec ==
3327                     nfsd_master_start.tv_usec &&
3328                     strcmp(procp->p_comm, nfsd_master_comm) == 0)
3329                         kern_psignal(procp, SIGUSR2);
3330                 else
3331                         nfsd_master_proc = NULL;
3332
3333                 if (procp != NULL)
3334                         PROC_UNLOCK(procp);
3335         }
3336 }
3337
3338 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3339
3340 /*
3341  * Called once to initialize data structures...
3342  */
3343 static int
3344 nfsd_modevent(module_t mod, int type, void *data)
3345 {
3346         int error = 0, i;
3347         static int loaded = 0;
3348
3349         switch (type) {
3350         case MOD_LOAD:
3351                 if (loaded)
3352                         goto out;
3353                 newnfs_portinit();
3354                 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3355                         mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
3356                             MTX_DEF);
3357                         mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
3358                             MTX_DEF);
3359                 }
3360                 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
3361                 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
3362                 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3363                 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3364                 nfsrvd_initcache();
3365                 nfsd_init();
3366                 NFSD_LOCK();
3367                 nfsrvd_init(0);
3368                 NFSD_UNLOCK();
3369                 nfsd_mntinit();
3370 #ifdef VV_DISABLEDELEG
3371                 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3372                 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3373 #endif
3374                 nfsd_call_servertimer = nfsrv_servertimer;
3375                 nfsd_call_nfsd = nfssvc_nfsd;
3376                 loaded = 1;
3377                 break;
3378
3379         case MOD_UNLOAD:
3380                 if (newnfs_numnfsd != 0) {
3381                         error = EBUSY;
3382                         break;
3383                 }
3384
3385 #ifdef VV_DISABLEDELEG
3386                 vn_deleg_ops.vndeleg_recall = NULL;
3387                 vn_deleg_ops.vndeleg_disable = NULL;
3388 #endif
3389                 nfsd_call_servertimer = NULL;
3390                 nfsd_call_nfsd = NULL;
3391
3392                 /* Clean out all NFSv4 state. */
3393                 nfsrv_throwawayallstate(curthread);
3394
3395                 /* Clean the NFS server reply cache */
3396                 nfsrvd_cleancache();
3397
3398                 /* Free up the krpc server pool. */
3399                 if (nfsrvd_pool != NULL)
3400                         svcpool_destroy(nfsrvd_pool);
3401
3402                 /* and get rid of the locks */
3403                 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3404                         mtx_destroy(&nfsrchash_table[i].mtx);
3405                         mtx_destroy(&nfsrcahash_table[i].mtx);
3406                 }
3407                 mtx_destroy(&nfsrc_udpmtx);
3408                 mtx_destroy(&nfs_v4root_mutex);
3409                 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3410                 for (i = 0; i < nfsrv_sessionhashsize; i++)
3411                         mtx_destroy(&nfssessionhash[i].mtx);
3412                 lockdestroy(&nfsv4root_mnt.mnt_explock);
3413                 free(nfsclienthash, M_NFSDCLIENT);
3414                 free(nfslockhash, M_NFSDLOCKFILE);
3415                 free(nfssessionhash, M_NFSDSESSION);
3416                 loaded = 0;
3417                 break;
3418         default:
3419                 error = EOPNOTSUPP;
3420                 break;
3421         }
3422
3423 out:
3424         NFSEXITCODE(error);
3425         return (error);
3426 }
3427 static moduledata_t nfsd_mod = {
3428         "nfsd",
3429         nfsd_modevent,
3430         NULL,
3431 };
3432 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3433
3434 /* So that loader and kldload(2) can find us, wherever we are.. */
3435 MODULE_VERSION(nfsd, 1);
3436 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3437 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3438 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3439 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3440 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3441