sys/kern/vfs_syscalls.c

   1 /*
   2  * Copyright (c) 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  * (c) UNIX System Laboratories, Inc.
   5  * All or some portions of this file are derived from material licensed
   6  * to the University of California by American Telephone and Telegraph
   7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
   8  * the permission of UNIX System Laboratories, Inc.
   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  *      @(#)vfs_syscalls.c      8.13 (Berkeley) 4/15/94
  35  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
  36  */
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/buf.h>
  41 #include <sys/conf.h>
  42 #include <sys/sysent.h>
  43 #include <sys/malloc.h>
  44 #include <sys/mount.h>
  45 #include <sys/mountctl.h>
  46 #include <sys/sysproto.h>
  47 #include <sys/filedesc.h>
  48 #include <sys/kernel.h>
  49 #include <sys/fcntl.h>
  50 #include <sys/file.h>
  51 #include <sys/linker.h>
  52 #include <sys/stat.h>
  53 #include <sys/unistd.h>
  54 #include <sys/vnode.h>
  55 #include <sys/proc.h>
  56 #include <sys/priv.h>
  57 #include <sys/jail.h>
  58 #include <sys/namei.h>
  59 #include <sys/nlookup.h>
  60 #include <sys/dirent.h>
  61 #include <sys/extattr.h>
  62 #include <sys/spinlock.h>
  63 #include <sys/kern_syscall.h>
  64 #include <sys/objcache.h>
  65 #include <sys/sysctl.h>
  66
  67 #include <sys/buf2.h>
  68 #include <sys/file2.h>
  69 #include <sys/spinlock2.h>
  70 #include <sys/mplock2.h>
  71
  72 #include <vm/vm.h>
  73 #include <vm/vm_object.h>
  74 #include <vm/vm_page.h>
  75
  76 #include <machine/limits.h>
  77 #include <machine/stdarg.h>
  78
  79 static void mount_warning(struct mount *mp, const char *ctl, ...)
  80                 __printflike(2, 3);
  81 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
  82 static int checkvp_chdir (struct vnode *vn, struct thread *td);
  83 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
  84 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
  85 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
  86 static int getutimes (struct timeval *, struct timespec *);
  87 static int getutimens (const struct timespec *, struct timespec *, int *);
  88 static int setfown (struct mount *, struct vnode *, uid_t, gid_t);
  89 static int setfmode (struct vnode *, int);
  90 static int setfflags (struct vnode *, int);
  91 static int setutimes (struct vnode *, struct vattr *,
  92                         const struct timespec *, int);
  93 static int      usermount = 0;  /* if 1, non-root can mount fs. */
  94
  95 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
  96     "Allow non-root users to mount filesystems");
  97
  98 /*
  99  * Virtual File System System Calls
 100  */
 101
 102 /*
 103  * Mount a file system.
 104  *
 105  * mount_args(char *type, char *path, int flags, caddr_t data)
 106  *
 107  * MPALMOSTSAFE
 108  */
 109 int
 110 sys_mount(struct mount_args *uap)
 111 {
 112         struct thread *td = curthread;
 113         struct vnode *vp;
 114         struct nchandle nch;
 115         struct mount *mp, *nullmp;
 116         struct vfsconf *vfsp;
 117         int error, flag = 0, flag2 = 0;
 118         int hasmount;
 119         struct vattr va;
 120         struct nlookupdata nd;
 121         char fstypename[MFSNAMELEN];
 122         struct ucred *cred;
 123
 124         cred = td->td_ucred;
 125         if (jailed(cred)) {
 126                 error = EPERM;
 127                 goto done;
 128         }
 129         if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
 130                 goto done;
 131
 132         /*
 133          * Do not allow NFS export by non-root users.
 134          */
 135         if (uap->flags & MNT_EXPORTED) {
 136                 error = priv_check(td, PRIV_ROOT);
 137                 if (error)
 138                         goto done;
 139         }
 140         /*
 141          * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
 142          */
 143         if (priv_check(td, PRIV_ROOT))
 144                 uap->flags |= MNT_NOSUID | MNT_NODEV;
 145
 146         /*
 147          * Lookup the requested path and extract the nch and vnode.
 148          */
 149         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 150         if (error == 0) {
 151                 if ((error = nlookup(&nd)) == 0) {
 152                         if (nd.nl_nch.ncp->nc_vp == NULL)
 153                                 error = ENOENT;
 154                 }
 155         }
 156         if (error) {
 157                 nlookup_done(&nd);
 158                 goto done;
 159         }
 160
 161         /*
 162          * If the target filesystem is resolved via a nullfs mount, then
 163          * nd.nl_nch.mount will be pointing to the nullfs mount structure
 164          * instead of the target file system. We need it in case we are
 165          * doing an update.
 166          */
 167         nullmp = nd.nl_nch.mount;
 168
 169         /*
 170          * Extract the locked+refd ncp and cleanup the nd structure
 171          */
 172         nch = nd.nl_nch;
 173         cache_zero(&nd.nl_nch);
 174         nlookup_done(&nd);
 175
 176         if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
 177             (mp = cache_findmount(&nch)) != NULL) {
 178                 cache_dropmount(mp);
 179                 hasmount = 1;
 180         } else {
 181                 hasmount = 0;
 182         }
 183
 184
 185         /*
 186          * now we have the locked ref'd nch and unreferenced vnode.
 187          */
 188         vp = nch.ncp->nc_vp;
 189         if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
 190                 cache_put(&nch);
 191                 goto done;
 192         }
 193         cache_unlock(&nch);
 194
 195         /*
 196          * Extract the file system type. We need to know this early, to take
 197          * appropriate actions if we are dealing with a nullfs.
 198          */
 199         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
 200                 cache_drop(&nch);
 201                 vput(vp);
 202                 goto done;
 203         }
 204
 205         /*
 206          * Now we have an unlocked ref'd nch and a locked ref'd vp
 207          */
 208         if (uap->flags & MNT_UPDATE) {
 209                 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
 210                         cache_drop(&nch);
 211                         vput(vp);
 212                         error = EINVAL;
 213                         goto done;
 214                 }
 215
 216                 if (strncmp(fstypename, "null", 5) == 0) {
 217                         KKASSERT(nullmp);
 218                         mp = nullmp;
 219                 } else {
 220                         mp = vp->v_mount;
 221                 }
 222
 223                 flag = mp->mnt_flag;
 224                 flag2 = mp->mnt_kern_flag;
 225                 /*
 226                  * We only allow the filesystem to be reloaded if it
 227                  * is currently mounted read-only.
 228                  */
 229                 if ((uap->flags & MNT_RELOAD) &&
 230                     ((mp->mnt_flag & MNT_RDONLY) == 0)) {
 231                         cache_drop(&nch);
 232                         vput(vp);
 233                         error = EOPNOTSUPP;     /* Needs translation */
 234                         goto done;
 235                 }
 236                 /*
 237                  * Only root, or the user that did the original mount is
 238                  * permitted to update it.
 239                  */
 240                 if (mp->mnt_stat.f_owner != cred->cr_uid &&
 241                     (error = priv_check(td, PRIV_ROOT))) {
 242                         cache_drop(&nch);
 243                         vput(vp);
 244                         goto done;
 245                 }
 246                 if (vfs_busy(mp, LK_NOWAIT)) {
 247                         cache_drop(&nch);
 248                         vput(vp);
 249                         error = EBUSY;
 250                         goto done;
 251                 }
 252                 if (hasmount) {
 253                         cache_drop(&nch);
 254                         vfs_unbusy(mp);
 255                         vput(vp);
 256                         error = EBUSY;
 257                         goto done;
 258                 }
 259                 mp->mnt_flag |=
 260                     uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
 261                 lwkt_gettoken(&mp->mnt_token);
 262                 vn_unlock(vp);
 263                 goto update;
 264         }
 265
 266         /*
 267          * If the user is not root, ensure that they own the directory
 268          * onto which we are attempting to mount.
 269          */
 270         if ((error = VOP_GETATTR(vp, &va)) ||
 271             (va.va_uid != cred->cr_uid &&
 272              (error = priv_check(td, PRIV_ROOT)))) {
 273                 cache_drop(&nch);
 274                 vput(vp);
 275                 goto done;
 276         }
 277         if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
 278                 cache_drop(&nch);
 279                 vput(vp);
 280                 goto done;
 281         }
 282         if (vp->v_type != VDIR) {
 283                 cache_drop(&nch);
 284                 vput(vp);
 285                 error = ENOTDIR;
 286                 goto done;
 287         }
 288         if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
 289                 cache_drop(&nch);
 290                 vput(vp);
 291                 error = EPERM;
 292                 goto done;
 293         }
 294         vfsp = vfsconf_find_by_name(fstypename);
 295         if (vfsp == NULL) {
 296                 linker_file_t lf;
 297
 298                 /* Only load modules for root (very important!) */
 299                 if ((error = priv_check(td, PRIV_ROOT)) != 0) {
 300                         cache_drop(&nch);
 301                         vput(vp);
 302                         goto done;
 303                 }
 304                 error = linker_load_file(fstypename, &lf);
 305                 if (error || lf == NULL) {
 306                         cache_drop(&nch);
 307                         vput(vp);
 308                         if (lf == NULL)
 309                                 error = ENODEV;
 310                         goto done;
 311                 }
 312                 lf->userrefs++;
 313                 /* lookup again, see if the VFS was loaded */
 314                 vfsp = vfsconf_find_by_name(fstypename);
 315                 if (vfsp == NULL) {
 316                         lf->userrefs--;
 317                         linker_file_unload(lf);
 318                         cache_drop(&nch);
 319                         vput(vp);
 320                         error = ENODEV;
 321                         goto done;
 322                 }
 323         }
 324         if (hasmount) {
 325                 cache_drop(&nch);
 326                 vput(vp);
 327                 error = EBUSY;
 328                 goto done;
 329         }
 330
 331         /*
 332          * Allocate and initialize the filesystem.
 333          */
 334         mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
 335         mount_init(mp);
 336         vfs_busy(mp, LK_NOWAIT);
 337         mp->mnt_op = vfsp->vfc_vfsops;
 338         mp->mnt_vfc = vfsp;
 339         vfsp->vfc_refcount++;
 340         mp->mnt_stat.f_type = vfsp->vfc_typenum;
 341         mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
 342         strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
 343         mp->mnt_stat.f_owner = cred->cr_uid;
 344         lwkt_gettoken(&mp->mnt_token);
 345         vn_unlock(vp);
 346 update:
 347         /*
 348          * (per-mount token acquired at this point)
 349          *
 350          * Set the mount level flags.
 351          */
 352         if (uap->flags & MNT_RDONLY)
 353                 mp->mnt_flag |= MNT_RDONLY;
 354         else if (mp->mnt_flag & MNT_RDONLY)
 355                 mp->mnt_kern_flag |= MNTK_WANTRDWR;
 356         mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
 357             MNT_SYNCHRONOUS | MNT_ASYNC | MNT_NOATIME |
 358             MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
 359             MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
 360             MNT_AUTOMOUNTED);
 361         mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
 362             MNT_NODEV | MNT_SYNCHRONOUS | MNT_ASYNC | MNT_FORCE |
 363             MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
 364             MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
 365             MNT_AUTOMOUNTED);
 366         /*
 367          * Mount the filesystem.
 368          * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
 369          * get.
 370          */
 371         error = VFS_MOUNT(mp, uap->path, uap->data, cred);
 372         if (mp->mnt_flag & MNT_UPDATE) {
 373                 if (mp->mnt_kern_flag & MNTK_WANTRDWR)
 374                         mp->mnt_flag &= ~MNT_RDONLY;
 375                 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
 376                 mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
 377                 if (error) {
 378                         mp->mnt_flag = flag;
 379                         mp->mnt_kern_flag = flag2;
 380                 }
 381                 lwkt_reltoken(&mp->mnt_token);
 382                 vfs_unbusy(mp);
 383                 vrele(vp);
 384                 cache_drop(&nch);
 385                 goto done;
 386         }
 387         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 388
 389         /*
 390          * Put the new filesystem on the mount list after root.  The mount
 391          * point gets its own mnt_ncmountpt (unless the VFS already set one
 392          * up) which represents the root of the mount.  The lookup code
 393          * detects the mount point going forward and checks the root of
 394          * the mount going backwards.
 395          *
 396          * It is not necessary to invalidate or purge the vnode underneath
 397          * because elements under the mount will be given their own glue
 398          * namecache record.
 399          */
 400         if (!error) {
 401                 if (mp->mnt_ncmountpt.ncp == NULL) {
 402                         /*
 403                          * allocate, then unlock, but leave the ref intact
 404                          */
 405                         cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
 406                         cache_unlock(&mp->mnt_ncmountpt);
 407                 }
 408                 mp->mnt_ncmounton = nch;                /* inherits ref */
 409                 nch.ncp->nc_flag |= NCF_ISMOUNTPT;
 410                 cache_ismounting(mp);
 411
 412                 mountlist_insert(mp, MNTINS_LAST);
 413                 vn_unlock(vp);
 414                 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
 415                 error = vfs_allocate_syncvnode(mp);
 416                 lwkt_reltoken(&mp->mnt_token);
 417                 vfs_unbusy(mp);
 418                 error = VFS_START(mp, 0);
 419                 vrele(vp);
 420                 KNOTE(&fs_klist, VQ_MOUNT);
 421         } else {
 422                 vn_syncer_thr_stop(mp);
 423                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 424                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 425                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 426                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 427                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 428                 mp->mnt_vfc->vfc_refcount--;
 429                 lwkt_reltoken(&mp->mnt_token);
 430                 vfs_unbusy(mp);
 431                 kfree(mp, M_MOUNT);
 432                 cache_drop(&nch);
 433                 vput(vp);
 434         }
 435 done:
 436         return (error);
 437 }
 438
 439 /*
 440  * Scan all active processes to see if any of them have a current
 441  * or root directory onto which the new filesystem has just been
 442  * mounted. If so, replace them with the new mount point.
 443  *
 444  * Both old_nch and new_nch are ref'd on call but not locked.
 445  * new_nch must be temporarily locked so it can be associated with the
 446  * vnode representing the root of the mount point.
 447  */
 448 struct checkdirs_info {
 449         struct nchandle old_nch;
 450         struct nchandle new_nch;
 451         struct vnode *old_vp;
 452         struct vnode *new_vp;
 453 };
 454
 455 static int checkdirs_callback(struct proc *p, void *data);
 456
 457 static void
 458 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
 459 {
 460         struct checkdirs_info info;
 461         struct vnode *olddp;
 462         struct vnode *newdp;
 463         struct mount *mp;
 464
 465         /*
 466          * If the old mount point's vnode has a usecount of 1, it is not
 467          * being held as a descriptor anywhere.
 468          */
 469         olddp = old_nch->ncp->nc_vp;
 470         if (olddp == NULL || VREFCNT(olddp) == 1)
 471                 return;
 472
 473         /*
 474          * Force the root vnode of the new mount point to be resolved
 475          * so we can update any matching processes.
 476          */
 477         mp = new_nch->mount;
 478         if (VFS_ROOT(mp, &newdp))
 479                 panic("mount: lost mount");
 480         vn_unlock(newdp);
 481         cache_lock(new_nch);
 482         vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
 483         cache_setunresolved(new_nch);
 484         cache_setvp(new_nch, newdp);
 485         cache_unlock(new_nch);
 486
 487         /*
 488          * Special handling of the root node
 489          */
 490         if (rootvnode == olddp) {
 491                 vref(newdp);
 492                 vfs_cache_setroot(newdp, cache_hold(new_nch));
 493         }
 494
 495         /*
 496          * Pass newdp separately so the callback does not have to access
 497          * it via new_nch->ncp->nc_vp.
 498          */
 499         info.old_nch = *old_nch;
 500         info.new_nch = *new_nch;
 501         info.new_vp = newdp;
 502         allproc_scan(checkdirs_callback, &info);
 503         vput(newdp);
 504 }
 505
 506 /*
 507  * NOTE: callback is not MP safe because the scanned process's filedesc
 508  * structure can be ripped out from under us, amoung other things.
 509  */
 510 static int
 511 checkdirs_callback(struct proc *p, void *data)
 512 {
 513         struct checkdirs_info *info = data;
 514         struct filedesc *fdp;
 515         struct nchandle ncdrop1;
 516         struct nchandle ncdrop2;
 517         struct vnode *vprele1;
 518         struct vnode *vprele2;
 519
 520         if ((fdp = p->p_fd) != NULL) {
 521                 cache_zero(&ncdrop1);
 522                 cache_zero(&ncdrop2);
 523                 vprele1 = NULL;
 524                 vprele2 = NULL;
 525
 526                 /*
 527                  * MPUNSAFE - XXX fdp can be pulled out from under a
 528                  * foreign process.
 529                  *
 530                  * A shared filedesc is ok, we don't have to copy it
 531                  * because we are making this change globally.
 532                  */
 533                 spin_lock(&fdp->fd_spin);
 534                 if (fdp->fd_ncdir.mount == info->old_nch.mount &&
 535                     fdp->fd_ncdir.ncp == info->old_nch.ncp) {
 536                         vprele1 = fdp->fd_cdir;
 537                         vref(info->new_vp);
 538                         fdp->fd_cdir = info->new_vp;
 539                         ncdrop1 = fdp->fd_ncdir;
 540                         cache_copy(&info->new_nch, &fdp->fd_ncdir);
 541                 }
 542                 if (fdp->fd_nrdir.mount == info->old_nch.mount &&
 543                     fdp->fd_nrdir.ncp == info->old_nch.ncp) {
 544                         vprele2 = fdp->fd_rdir;
 545                         vref(info->new_vp);
 546                         fdp->fd_rdir = info->new_vp;
 547                         ncdrop2 = fdp->fd_nrdir;
 548                         cache_copy(&info->new_nch, &fdp->fd_nrdir);
 549                 }
 550                 spin_unlock(&fdp->fd_spin);
 551                 if (ncdrop1.ncp)
 552                         cache_drop(&ncdrop1);
 553                 if (ncdrop2.ncp)
 554                         cache_drop(&ncdrop2);
 555                 if (vprele1)
 556                         vrele(vprele1);
 557                 if (vprele2)
 558                         vrele(vprele2);
 559         }
 560         return(0);
 561 }
 562
 563 /*
 564  * Unmount a file system.
 565  *
 566  * Note: unmount takes a path to the vnode mounted on as argument,
 567  * not special file (as before).
 568  *
 569  * umount_args(char *path, int flags)
 570  *
 571  * MPALMOSTSAFE
 572  */
 573 int
 574 sys_unmount(struct unmount_args *uap)
 575 {
 576         struct thread *td = curthread;
 577         struct proc *p __debugvar = td->td_proc;
 578         struct mount *mp = NULL;
 579         struct nlookupdata nd;
 580         int error;
 581
 582         KKASSERT(p);
 583         get_mplock();
 584         if (td->td_ucred->cr_prison != NULL) {
 585                 error = EPERM;
 586                 goto done;
 587         }
 588         if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
 589                 goto done;
 590
 591         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 592         if (error == 0)
 593                 error = nlookup(&nd);
 594         if (error)
 595                 goto out;
 596
 597         mp = nd.nl_nch.mount;
 598
 599         /*
 600          * Only root, or the user that did the original mount is
 601          * permitted to unmount this filesystem.
 602          */
 603         if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
 604             (error = priv_check(td, PRIV_ROOT)))
 605                 goto out;
 606
 607         /*
 608          * Don't allow unmounting the root file system.
 609          */
 610         if (mp->mnt_flag & MNT_ROOTFS) {
 611                 error = EINVAL;
 612                 goto out;
 613         }
 614
 615         /*
 616          * Must be the root of the filesystem
 617          */
 618         if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
 619                 error = EINVAL;
 620                 goto out;
 621         }
 622
 623 out:
 624         nlookup_done(&nd);
 625         if (error == 0)
 626                 error = dounmount(mp, uap->flags);
 627 done:
 628         rel_mplock();
 629         return (error);
 630 }
 631
 632 /*
 633  * Do the actual file system unmount.
 634  */
 635 static int
 636 dounmount_interlock(struct mount *mp)
 637 {
 638         if (mp->mnt_kern_flag & MNTK_UNMOUNT)
 639                 return (EBUSY);
 640         mp->mnt_kern_flag |= MNTK_UNMOUNT;
 641         return(0);
 642 }
 643
 644 static int
 645 unmount_allproc_cb(struct proc *p, void *arg)
 646 {
 647         struct mount *mp;
 648
 649         if (p->p_textnch.ncp == NULL)
 650                 return 0;
 651
 652         mp = (struct mount *)arg;
 653         if (p->p_textnch.mount == mp)
 654                 cache_drop(&p->p_textnch);
 655
 656         return 0;
 657 }
 658
 659 int
 660 dounmount(struct mount *mp, int flags)
 661 {
 662         struct namecache *ncp;
 663         struct nchandle nch;
 664         struct vnode *vp;
 665         int error;
 666         int async_flag;
 667         int lflags;
 668         int freeok = 1;
 669         int retry;
 670
 671         lwkt_gettoken(&mp->mnt_token);
 672         /*
 673          * Exclusive access for unmounting purposes
 674          */
 675         if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
 676                 goto out;
 677
 678         /*
 679          * Allow filesystems to detect that a forced unmount is in progress.
 680          */
 681         if (flags & MNT_FORCE)
 682                 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
 683         lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_TIMELOCK);
 684         error = lockmgr(&mp->mnt_lock, lflags);
 685         if (error) {
 686                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 687                 if (mp->mnt_kern_flag & MNTK_MWAIT) {
 688                         mp->mnt_kern_flag &= ~MNTK_MWAIT;
 689                         wakeup(mp);
 690                 }
 691                 goto out;
 692         }
 693
 694         if (mp->mnt_flag & MNT_EXPUBLIC)
 695                 vfs_setpublicfs(NULL, NULL, NULL);
 696
 697         vfs_msync(mp, MNT_WAIT);
 698         async_flag = mp->mnt_flag & MNT_ASYNC;
 699         mp->mnt_flag &=~ MNT_ASYNC;
 700
 701         /*
 702          * If this filesystem isn't aliasing other filesystems,
 703          * try to invalidate any remaining namecache entries and
 704          * check the count afterwords.
 705          */
 706         if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
 707                 cache_lock(&mp->mnt_ncmountpt);
 708                 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
 709                 cache_unlock(&mp->mnt_ncmountpt);
 710
 711                 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
 712                     (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
 713                         allproc_scan(&unmount_allproc_cb, mp);
 714                 }
 715
 716                 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
 717                     (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
 718
 719                         if ((flags & MNT_FORCE) == 0) {
 720                                 error = EBUSY;
 721                                 mount_warning(mp, "Cannot unmount: "
 722                                                   "%d namecache "
 723                                                   "references still "
 724                                                   "present",
 725                                                   ncp->nc_refs - 1);
 726                         } else {
 727                                 mount_warning(mp, "Forced unmount: "
 728                                                   "%d namecache "
 729                                                   "references still "
 730                                                   "present",
 731                                                   ncp->nc_refs - 1);
 732                                 freeok = 0;
 733                         }
 734                 }
 735         }
 736
 737         /*
 738          * Decomission our special mnt_syncer vnode.  This also stops
 739          * the vnlru code.  If we are unable to unmount we recommission
 740          * the vnode.
 741          *
 742          * Then sync the filesystem.
 743          */
 744         if ((vp = mp->mnt_syncer) != NULL) {
 745                 mp->mnt_syncer = NULL;
 746                 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
 747                 vrele(vp);
 748         }
 749         if ((mp->mnt_flag & MNT_RDONLY) == 0)
 750                 VFS_SYNC(mp, MNT_WAIT);
 751
 752         /*
 753          * nchandle records ref the mount structure.  Expect a count of 1
 754          * (our mount->mnt_ncmountpt).
 755          *
 756          * Scans can get temporary refs on a mountpoint (thought really
 757          * heavy duty stuff like cache_findmount() do not).
 758          */
 759         for (retry = 0; retry < 10 && mp->mnt_refs != 1; ++retry) {
 760                 cache_unmounting(mp);
 761                 tsleep(&mp->mnt_refs, 0, "mntbsy", hz / 10 + 1);
 762         }
 763         if (mp->mnt_refs != 1) {
 764                 if ((flags & MNT_FORCE) == 0) {
 765                         mount_warning(mp, "Cannot unmount: "
 766                                           "%d mount refs still present",
 767                                           mp->mnt_refs);
 768                         error = EBUSY;
 769                 } else {
 770                         mount_warning(mp, "Forced unmount: "
 771                                           "%d mount refs still present",
 772                                           mp->mnt_refs);
 773                         freeok = 0;
 774                 }
 775         }
 776
 777         /*
 778          * So far so good, sync the filesystem once more and
 779          * call the VFS unmount code if the sync succeeds.
 780          */
 781         if (error == 0) {
 782                 if (mp->mnt_flag & MNT_RDONLY) {
 783                         error = VFS_UNMOUNT(mp, flags);
 784                 } else {
 785                         error = VFS_SYNC(mp, MNT_WAIT);
 786                         if ((error == 0) ||
 787                             (error == EOPNOTSUPP) || /* No sync */
 788                             (flags & MNT_FORCE)) {
 789                                 error = VFS_UNMOUNT(mp, flags);
 790                         }
 791                 }
 792         }
 793
 794         /*
 795          * If an error occurred we can still recover, restoring the
 796          * syncer vnode and misc flags.
 797          */
 798         if (error) {
 799                 if (mp->mnt_syncer == NULL)
 800                         vfs_allocate_syncvnode(mp);
 801                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 802                 mp->mnt_flag |= async_flag;
 803                 lockmgr(&mp->mnt_lock, LK_RELEASE);
 804                 if (mp->mnt_kern_flag & MNTK_MWAIT) {
 805                         mp->mnt_kern_flag &= ~MNTK_MWAIT;
 806                         wakeup(mp);
 807                 }
 808                 goto out;
 809         }
 810         /*
 811          * Clean up any journals still associated with the mount after
 812          * filesystem activity has ceased.
 813          */
 814         journal_remove_all_journals(mp,
 815             ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
 816
 817         mountlist_remove(mp);
 818
 819         /*
 820          * Remove any installed vnode ops here so the individual VFSs don't
 821          * have to.
 822          */
 823         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 824         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 825         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 826         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 827         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 828
 829         if (mp->mnt_ncmountpt.ncp != NULL) {
 830                 nch = mp->mnt_ncmountpt;
 831                 cache_zero(&mp->mnt_ncmountpt);
 832                 cache_clrmountpt(&nch);
 833                 cache_drop(&nch);
 834         }
 835         if (mp->mnt_ncmounton.ncp != NULL) {
 836                 cache_unmounting(mp);
 837                 nch = mp->mnt_ncmounton;
 838                 cache_zero(&mp->mnt_ncmounton);
 839                 cache_clrmountpt(&nch);
 840                 cache_drop(&nch);
 841         }
 842
 843         mp->mnt_vfc->vfc_refcount--;
 844         if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
 845                 panic("unmount: dangling vnode");
 846         lockmgr(&mp->mnt_lock, LK_RELEASE);
 847         if (mp->mnt_kern_flag & MNTK_MWAIT) {
 848                 mp->mnt_kern_flag &= ~MNTK_MWAIT;
 849                 wakeup(mp);
 850         }
 851
 852         /*
 853          * If we reach here and freeok != 0 we must free the mount.
 854          * If refs > 1 cycle and wait, just in case someone tried
 855          * to busy the mount after we decided to do the unmount.
 856          */
 857         if (freeok) {
 858                 while (mp->mnt_refs > 1) {
 859                         cache_unmounting(mp);
 860                         wakeup(mp);
 861                         tsleep(&mp->mnt_refs, 0, "umntrwait", hz / 10 + 1);
 862                 }
 863                 lwkt_reltoken(&mp->mnt_token);
 864                 kfree(mp, M_MOUNT);
 865                 mp = NULL;
 866         }
 867         error = 0;
 868         KNOTE(&fs_klist, VQ_UNMOUNT);
 869 out:
 870         if (mp)
 871                 lwkt_reltoken(&mp->mnt_token);
 872         return (error);
 873 }
 874
 875 static
 876 void
 877 mount_warning(struct mount *mp, const char *ctl, ...)
 878 {
 879         char *ptr;
 880         char *buf;
 881         __va_list va;
 882
 883         __va_start(va, ctl);
 884         if (cache_fullpath(NULL, &mp->mnt_ncmounton, NULL,
 885                            &ptr, &buf, 0) == 0) {
 886                 kprintf("unmount(%s): ", ptr);
 887                 kvprintf(ctl, va);
 888                 kprintf("\n");
 889                 kfree(buf, M_TEMP);
 890         } else {
 891                 kprintf("unmount(%p", mp);
 892                 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
 893                         kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
 894                 kprintf("): ");
 895                 kvprintf(ctl, va);
 896                 kprintf("\n");
 897         }
 898         __va_end(va);
 899 }
 900
 901 /*
 902  * Shim cache_fullpath() to handle the case where a process is chrooted into
 903  * a subdirectory of a mount.  In this case if the root mount matches the
 904  * process root directory's mount we have to specify the process's root
 905  * directory instead of the mount point, because the mount point might
 906  * be above the root directory.
 907  */
 908 static
 909 int
 910 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
 911 {
 912         struct nchandle *nch;
 913
 914         if (p && p->p_fd->fd_nrdir.mount == mp)
 915                 nch = &p->p_fd->fd_nrdir;
 916         else
 917                 nch = &mp->mnt_ncmountpt;
 918         return(cache_fullpath(p, nch, NULL, rb, fb, 0));
 919 }
 920
 921 /*
 922  * Sync each mounted filesystem.
 923  */
 924
 925 #ifdef DEBUG
 926 static int syncprt = 0;
 927 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
 928 #endif /* DEBUG */
 929
 930 static int sync_callback(struct mount *mp, void *data);
 931
 932 int
 933 sys_sync(struct sync_args *uap)
 934 {
 935         mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
 936         return (0);
 937 }
 938
 939 static
 940 int
 941 sync_callback(struct mount *mp, void *data __unused)
 942 {
 943         int asyncflag;
 944
 945         if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 946                 asyncflag = mp->mnt_flag & MNT_ASYNC;
 947                 mp->mnt_flag &= ~MNT_ASYNC;
 948                 vfs_msync(mp, MNT_NOWAIT);
 949                 VFS_SYNC(mp, MNT_NOWAIT);
 950                 mp->mnt_flag |= asyncflag;
 951         }
 952         return(0);
 953 }
 954
 955 /* XXX PRISON: could be per prison flag */
 956 static int prison_quotas;
 957 #if 0
 958 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
 959 #endif
 960
 961 /*
 962  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
 963  *
 964  * Change filesystem quotas.
 965  *
 966  * MPALMOSTSAFE
 967  */
 968 int
 969 sys_quotactl(struct quotactl_args *uap)
 970 {
 971         struct nlookupdata nd;
 972         struct thread *td;
 973         struct mount *mp;
 974         int error;
 975
 976         get_mplock();
 977         td = curthread;
 978         if (td->td_ucred->cr_prison && !prison_quotas) {
 979                 error = EPERM;
 980                 goto done;
 981         }
 982
 983         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 984         if (error == 0)
 985                 error = nlookup(&nd);
 986         if (error == 0) {
 987                 mp = nd.nl_nch.mount;
 988                 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
 989                                     uap->arg, nd.nl_cred);
 990         }
 991         nlookup_done(&nd);
 992 done:
 993         rel_mplock();
 994         return (error);
 995 }
 996
 997 /*
 998  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
 999  *              void *buf, int buflen)
1000  *
1001  * This function operates on a mount point and executes the specified
1002  * operation using the specified control data, and possibly returns data.
1003  *
1004  * The actual number of bytes stored in the result buffer is returned, 0
1005  * if none, otherwise an error is returned.
1006  *
1007  * MPALMOSTSAFE
1008  */
1009 int
1010 sys_mountctl(struct mountctl_args *uap)
1011 {
1012         struct thread *td = curthread;
1013         struct proc *p = td->td_proc;
1014         struct file *fp;
1015         void *ctl = NULL;
1016         void *buf = NULL;
1017         char *path = NULL;
1018         int error;
1019
1020         /*
1021          * Sanity and permissions checks.  We must be root.
1022          */
1023         KKASSERT(p);
1024         if (td->td_ucred->cr_prison != NULL)
1025                 return (EPERM);
1026         if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
1027             (error = priv_check(td, PRIV_ROOT)) != 0)
1028                 return (error);
1029
1030         /*
1031          * Argument length checks
1032          */
1033         if (uap->ctllen < 0 || uap->ctllen > 1024)
1034                 return (EINVAL);
1035         if (uap->buflen < 0 || uap->buflen > 16 * 1024)
1036                 return (EINVAL);
1037         if (uap->path == NULL)
1038                 return (EINVAL);
1039
1040         /*
1041          * Allocate the necessary buffers and copyin data
1042          */
1043         path = objcache_get(namei_oc, M_WAITOK);
1044         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1045         if (error)
1046                 goto done;
1047
1048         if (uap->ctllen) {
1049                 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
1050                 error = copyin(uap->ctl, ctl, uap->ctllen);
1051                 if (error)
1052                         goto done;
1053         }
1054         if (uap->buflen)
1055                 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1056
1057         /*
1058          * Validate the descriptor
1059          */
1060         if (uap->fd >= 0) {
1061                 fp = holdfp(p->p_fd, uap->fd, -1);
1062                 if (fp == NULL) {
1063                         error = EBADF;
1064                         goto done;
1065                 }
1066         } else {
1067                 fp = NULL;
1068         }
1069
1070         /*
1071          * Execute the internal kernel function and clean up.
1072          */
1073         get_mplock();
1074         error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
1075         rel_mplock();
1076         if (fp)
1077                 fdrop(fp);
1078         if (error == 0 && uap->sysmsg_result > 0)
1079                 error = copyout(buf, uap->buf, uap->sysmsg_result);
1080 done:
1081         if (path)
1082                 objcache_put(namei_oc, path);
1083         if (ctl)
1084                 kfree(ctl, M_TEMP);
1085         if (buf)
1086                 kfree(buf, M_TEMP);
1087         return (error);
1088 }
1089
1090 /*
1091  * Execute a mount control operation by resolving the path to a mount point
1092  * and calling vop_mountctl().
1093  *
1094  * Use the mount point from the nch instead of the vnode so nullfs mounts
1095  * can properly spike the VOP.
1096  */
1097 int
1098 kern_mountctl(const char *path, int op, struct file *fp,
1099                 const void *ctl, int ctllen,
1100                 void *buf, int buflen, int *res)
1101 {
1102         struct vnode *vp;
1103         struct mount *mp;
1104         struct nlookupdata nd;
1105         int error;
1106
1107         *res = 0;
1108         vp = NULL;
1109         error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1110         if (error == 0)
1111                 error = nlookup(&nd);
1112         if (error == 0)
1113                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1114         mp = nd.nl_nch.mount;
1115         nlookup_done(&nd);
1116         if (error)
1117                 return (error);
1118         vn_unlock(vp);
1119
1120         /*
1121          * Must be the root of the filesystem
1122          */
1123         if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1124                 vrele(vp);
1125                 return (EINVAL);
1126         }
1127         error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1128                              buf, buflen, res);
1129         vrele(vp);
1130         return (error);
1131 }
1132
1133 int
1134 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1135 {
1136         struct thread *td = curthread;
1137         struct proc *p = td->td_proc;
1138         struct mount *mp;
1139         struct statfs *sp;
1140         char *fullpath, *freepath;
1141         int error;
1142
1143         if ((error = nlookup(nd)) != 0)
1144                 return (error);
1145         mp = nd->nl_nch.mount;
1146         sp = &mp->mnt_stat;
1147         if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1148                 return (error);
1149
1150         error = mount_path(p, mp, &fullpath, &freepath);
1151         if (error)
1152                 return(error);
1153         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1154         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1155         kfree(freepath, M_TEMP);
1156
1157         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1158         bcopy(sp, buf, sizeof(*buf));
1159         /* Only root should have access to the fsid's. */
1160         if (priv_check(td, PRIV_ROOT))
1161                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1162         return (0);
1163 }
1164
1165 /*
1166  * statfs_args(char *path, struct statfs *buf)
1167  *
1168  * Get filesystem statistics.
1169  */
1170 int
1171 sys_statfs(struct statfs_args *uap)
1172 {
1173         struct nlookupdata nd;
1174         struct statfs buf;
1175         int error;
1176
1177         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1178         if (error == 0)
1179                 error = kern_statfs(&nd, &buf);
1180         nlookup_done(&nd);
1181         if (error == 0)
1182                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1183         return (error);
1184 }
1185
1186 int
1187 kern_fstatfs(int fd, struct statfs *buf)
1188 {
1189         struct thread *td = curthread;
1190         struct proc *p = td->td_proc;
1191         struct file *fp;
1192         struct mount *mp;
1193         struct statfs *sp;
1194         char *fullpath, *freepath;
1195         int error;
1196
1197         KKASSERT(p);
1198         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1199                 return (error);
1200
1201         /*
1202          * Try to use mount info from any overlays rather than the
1203          * mount info for the underlying vnode, otherwise we will
1204          * fail when operating on null-mounted paths inside a chroot.
1205          */
1206         if ((mp = fp->f_nchandle.mount) == NULL)
1207                 mp = ((struct vnode *)fp->f_data)->v_mount;
1208         if (mp == NULL) {
1209                 error = EBADF;
1210                 goto done;
1211         }
1212         if (fp->f_cred == NULL) {
1213                 error = EINVAL;
1214                 goto done;
1215         }
1216         sp = &mp->mnt_stat;
1217         if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1218                 goto done;
1219
1220         if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1221                 goto done;
1222         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1223         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1224         kfree(freepath, M_TEMP);
1225
1226         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1227         bcopy(sp, buf, sizeof(*buf));
1228
1229         /* Only root should have access to the fsid's. */
1230         if (priv_check(td, PRIV_ROOT))
1231                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1232         error = 0;
1233 done:
1234         fdrop(fp);
1235         return (error);
1236 }
1237
1238 /*
1239  * fstatfs_args(int fd, struct statfs *buf)
1240  *
1241  * Get filesystem statistics.
1242  */
1243 int
1244 sys_fstatfs(struct fstatfs_args *uap)
1245 {
1246         struct statfs buf;
1247         int error;
1248
1249         error = kern_fstatfs(uap->fd, &buf);
1250
1251         if (error == 0)
1252                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1253         return (error);
1254 }
1255
1256 int
1257 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1258 {
1259         struct mount *mp;
1260         struct statvfs *sp;
1261         int error;
1262
1263         if ((error = nlookup(nd)) != 0)
1264                 return (error);
1265         mp = nd->nl_nch.mount;
1266         sp = &mp->mnt_vstat;
1267         if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1268                 return (error);
1269
1270         sp->f_flag = 0;
1271         if (mp->mnt_flag & MNT_RDONLY)
1272                 sp->f_flag |= ST_RDONLY;
1273         if (mp->mnt_flag & MNT_NOSUID)
1274                 sp->f_flag |= ST_NOSUID;
1275         bcopy(sp, buf, sizeof(*buf));
1276         return (0);
1277 }
1278
1279 /*
1280  * statfs_args(char *path, struct statfs *buf)
1281  *
1282  * Get filesystem statistics.
1283  */
1284 int
1285 sys_statvfs(struct statvfs_args *uap)
1286 {
1287         struct nlookupdata nd;
1288         struct statvfs buf;
1289         int error;
1290
1291         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1292         if (error == 0)
1293                 error = kern_statvfs(&nd, &buf);
1294         nlookup_done(&nd);
1295         if (error == 0)
1296                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1297         return (error);
1298 }
1299
1300 int
1301 kern_fstatvfs(int fd, struct statvfs *buf)
1302 {
1303         struct thread *td = curthread;
1304         struct proc *p = td->td_proc;
1305         struct file *fp;
1306         struct mount *mp;
1307         struct statvfs *sp;
1308         int error;
1309
1310         KKASSERT(p);
1311         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1312                 return (error);
1313         if ((mp = fp->f_nchandle.mount) == NULL)
1314                 mp = ((struct vnode *)fp->f_data)->v_mount;
1315         if (mp == NULL) {
1316                 error = EBADF;
1317                 goto done;
1318         }
1319         if (fp->f_cred == NULL) {
1320                 error = EINVAL;
1321                 goto done;
1322         }
1323         sp = &mp->mnt_vstat;
1324         if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1325                 goto done;
1326
1327         sp->f_flag = 0;
1328         if (mp->mnt_flag & MNT_RDONLY)
1329                 sp->f_flag |= ST_RDONLY;
1330         if (mp->mnt_flag & MNT_NOSUID)
1331                 sp->f_flag |= ST_NOSUID;
1332
1333         bcopy(sp, buf, sizeof(*buf));
1334         error = 0;
1335 done:
1336         fdrop(fp);
1337         return (error);
1338 }
1339
1340 /*
1341  * fstatfs_args(int fd, struct statfs *buf)
1342  *
1343  * Get filesystem statistics.
1344  */
1345 int
1346 sys_fstatvfs(struct fstatvfs_args *uap)
1347 {
1348         struct statvfs buf;
1349         int error;
1350
1351         error = kern_fstatvfs(uap->fd, &buf);
1352
1353         if (error == 0)
1354                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1355         return (error);
1356 }
1357
1358 /*
1359  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1360  *
1361  * Get statistics on all filesystems.
1362  */
1363
1364 struct getfsstat_info {
1365         struct statfs *sfsp;
1366         long count;
1367         long maxcount;
1368         int error;
1369         int flags;
1370         struct thread *td;
1371 };
1372
1373 static int getfsstat_callback(struct mount *, void *);
1374
1375 int
1376 sys_getfsstat(struct getfsstat_args *uap)
1377 {
1378         struct thread *td = curthread;
1379         struct getfsstat_info info;
1380
1381         bzero(&info, sizeof(info));
1382
1383         info.maxcount = uap->bufsize / sizeof(struct statfs);
1384         info.sfsp = uap->buf;
1385         info.count = 0;
1386         info.flags = uap->flags;
1387         info.td = td;
1388
1389         mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1390         if (info.sfsp && info.count > info.maxcount)
1391                 uap->sysmsg_result = info.maxcount;
1392         else
1393                 uap->sysmsg_result = info.count;
1394         return (info.error);
1395 }
1396
1397 static int
1398 getfsstat_callback(struct mount *mp, void *data)
1399 {
1400         struct getfsstat_info *info = data;
1401         struct statfs *sp;
1402         char *freepath;
1403         char *fullpath;
1404         int error;
1405
1406         if (info->sfsp && info->count < info->maxcount) {
1407                 if (info->td->td_proc &&
1408                     !chroot_visible_mnt(mp, info->td->td_proc)) {
1409                         return(0);
1410                 }
1411                 sp = &mp->mnt_stat;
1412
1413                 /*
1414                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1415                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1416                  * overrides MNT_WAIT.
1417                  */
1418                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1419                     (info->flags & MNT_WAIT)) &&
1420                     (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1421                         return(0);
1422                 }
1423                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1424
1425                 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1426                 if (error) {
1427                         info->error = error;
1428                         return(-1);
1429                 }
1430                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1431                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1432                 kfree(freepath, M_TEMP);
1433
1434                 error = copyout(sp, info->sfsp, sizeof(*sp));
1435                 if (error) {
1436                         info->error = error;
1437                         return (-1);
1438                 }
1439                 ++info->sfsp;
1440         }
1441         info->count++;
1442         return(0);
1443 }
1444
1445 /*
1446  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1447                    long bufsize, int flags)
1448  *
1449  * Get statistics on all filesystems.
1450  */
1451
1452 struct getvfsstat_info {
1453         struct statfs *sfsp;
1454         struct statvfs *vsfsp;
1455         long count;
1456         long maxcount;
1457         int error;
1458         int flags;
1459         struct thread *td;
1460 };
1461
1462 static int getvfsstat_callback(struct mount *, void *);
1463
1464 int
1465 sys_getvfsstat(struct getvfsstat_args *uap)
1466 {
1467         struct thread *td = curthread;
1468         struct getvfsstat_info info;
1469
1470         bzero(&info, sizeof(info));
1471
1472         info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1473         info.sfsp = uap->buf;
1474         info.vsfsp = uap->vbuf;
1475         info.count = 0;
1476         info.flags = uap->flags;
1477         info.td = td;
1478
1479         mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1480         if (info.vsfsp && info.count > info.maxcount)
1481                 uap->sysmsg_result = info.maxcount;
1482         else
1483                 uap->sysmsg_result = info.count;
1484         return (info.error);
1485 }
1486
1487 static int
1488 getvfsstat_callback(struct mount *mp, void *data)
1489 {
1490         struct getvfsstat_info *info = data;
1491         struct statfs *sp;
1492         struct statvfs *vsp;
1493         char *freepath;
1494         char *fullpath;
1495         int error;
1496
1497         if (info->vsfsp && info->count < info->maxcount) {
1498                 if (info->td->td_proc &&
1499                     !chroot_visible_mnt(mp, info->td->td_proc)) {
1500                         return(0);
1501                 }
1502                 sp = &mp->mnt_stat;
1503                 vsp = &mp->mnt_vstat;
1504
1505                 /*
1506                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1507                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1508                  * overrides MNT_WAIT.
1509                  */
1510                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1511                     (info->flags & MNT_WAIT)) &&
1512                     (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1513                         return(0);
1514                 }
1515                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1516
1517                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1518                     (info->flags & MNT_WAIT)) &&
1519                     (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1520                         return(0);
1521                 }
1522                 vsp->f_flag = 0;
1523                 if (mp->mnt_flag & MNT_RDONLY)
1524                         vsp->f_flag |= ST_RDONLY;
1525                 if (mp->mnt_flag & MNT_NOSUID)
1526                         vsp->f_flag |= ST_NOSUID;
1527
1528                 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1529                 if (error) {
1530                         info->error = error;
1531                         return(-1);
1532                 }
1533                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1534                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1535                 kfree(freepath, M_TEMP);
1536
1537                 error = copyout(sp, info->sfsp, sizeof(*sp));
1538                 if (error == 0)
1539                         error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1540                 if (error) {
1541                         info->error = error;
1542                         return (-1);
1543                 }
1544                 ++info->sfsp;
1545                 ++info->vsfsp;
1546         }
1547         info->count++;
1548         return(0);
1549 }
1550
1551
1552 /*
1553  * fchdir_args(int fd)
1554  *
1555  * Change current working directory to a given file descriptor.
1556  */
1557 int
1558 sys_fchdir(struct fchdir_args *uap)
1559 {
1560         struct thread *td = curthread;
1561         struct proc *p = td->td_proc;
1562         struct filedesc *fdp = p->p_fd;
1563         struct vnode *vp, *ovp;
1564         struct mount *mp;
1565         struct file *fp;
1566         struct nchandle nch, onch, tnch;
1567         int error;
1568
1569         if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1570                 return (error);
1571         lwkt_gettoken(&p->p_token);
1572         vp = (struct vnode *)fp->f_data;
1573         vref(vp);
1574         vn_lock(vp, LK_SHARED | LK_RETRY);
1575         if (fp->f_nchandle.ncp == NULL)
1576                 error = ENOTDIR;
1577         else
1578                 error = checkvp_chdir(vp, td);
1579         if (error) {
1580                 vput(vp);
1581                 goto done;
1582         }
1583         cache_copy(&fp->f_nchandle, &nch);
1584
1585         /*
1586          * If the ncp has become a mount point, traverse through
1587          * the mount point.
1588          */
1589
1590         while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1591                (mp = cache_findmount(&nch)) != NULL
1592         ) {
1593                 error = nlookup_mp(mp, &tnch);
1594                 if (error == 0) {
1595                         cache_unlock(&tnch);    /* leave ref intact */
1596                         vput(vp);
1597                         vp = tnch.ncp->nc_vp;
1598                         error = vget(vp, LK_SHARED);
1599                         KKASSERT(error == 0);
1600                         cache_drop(&nch);
1601                         nch = tnch;
1602                 }
1603                 cache_dropmount(mp);
1604         }
1605         if (error == 0) {
1606                 ovp = fdp->fd_cdir;
1607                 onch = fdp->fd_ncdir;
1608                 vn_unlock(vp);          /* leave ref intact */
1609                 fdp->fd_cdir = vp;
1610                 fdp->fd_ncdir = nch;
1611                 cache_drop(&onch);
1612                 vrele(ovp);
1613         } else {
1614                 cache_drop(&nch);
1615                 vput(vp);
1616         }
1617         fdrop(fp);
1618 done:
1619         lwkt_reltoken(&p->p_token);
1620         return (error);
1621 }
1622
1623 int
1624 kern_chdir(struct nlookupdata *nd)
1625 {
1626         struct thread *td = curthread;
1627         struct proc *p = td->td_proc;
1628         struct filedesc *fdp = p->p_fd;
1629         struct vnode *vp, *ovp;
1630         struct nchandle onch;
1631         int error;
1632
1633         nd->nl_flags |= NLC_SHAREDLOCK;
1634         if ((error = nlookup(nd)) != 0)
1635                 return (error);
1636         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1637                 return (ENOENT);
1638         if ((error = vget(vp, LK_SHARED)) != 0)
1639                 return (error);
1640
1641         lwkt_gettoken(&p->p_token);
1642         error = checkvp_chdir(vp, td);
1643         vn_unlock(vp);
1644         if (error == 0) {
1645                 ovp = fdp->fd_cdir;
1646                 onch = fdp->fd_ncdir;
1647                 cache_unlock(&nd->nl_nch);      /* leave reference intact */
1648                 fdp->fd_ncdir = nd->nl_nch;
1649                 fdp->fd_cdir = vp;
1650                 cache_drop(&onch);
1651                 vrele(ovp);
1652                 cache_zero(&nd->nl_nch);
1653         } else {
1654                 vrele(vp);
1655         }
1656         lwkt_reltoken(&p->p_token);
1657         return (error);
1658 }
1659
1660 /*
1661  * chdir_args(char *path)
1662  *
1663  * Change current working directory (``.'').
1664  */
1665 int
1666 sys_chdir(struct chdir_args *uap)
1667 {
1668         struct nlookupdata nd;
1669         int error;
1670
1671         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1672         if (error == 0)
1673                 error = kern_chdir(&nd);
1674         nlookup_done(&nd);
1675         return (error);
1676 }
1677
1678 /*
1679  * Helper function for raised chroot(2) security function:  Refuse if
1680  * any filedescriptors are open directories.
1681  */
1682 static int
1683 chroot_refuse_vdir_fds(struct filedesc *fdp)
1684 {
1685         struct vnode *vp;
1686         struct file *fp;
1687         int error;
1688         int fd;
1689
1690         for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1691                 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1692                         continue;
1693                 vp = (struct vnode *)fp->f_data;
1694                 if (vp->v_type != VDIR) {
1695                         fdrop(fp);
1696                         continue;
1697                 }
1698                 fdrop(fp);
1699                 return(EPERM);
1700         }
1701         return (0);
1702 }
1703
1704 /*
1705  * This sysctl determines if we will allow a process to chroot(2) if it
1706  * has a directory open:
1707  *      0: disallowed for all processes.
1708  *      1: allowed for processes that were not already chroot(2)'ed.
1709  *      2: allowed for all processes.
1710  */
1711
1712 static int chroot_allow_open_directories = 1;
1713
1714 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1715      &chroot_allow_open_directories, 0, "");
1716
1717 /*
1718  * chroot to the specified namecache entry.  We obtain the vp from the
1719  * namecache data.  The passed ncp must be locked and referenced and will
1720  * remain locked and referenced on return.
1721  */
1722 int
1723 kern_chroot(struct nchandle *nch)
1724 {
1725         struct thread *td = curthread;
1726         struct proc *p = td->td_proc;
1727         struct filedesc *fdp = p->p_fd;
1728         struct vnode *vp;
1729         int error;
1730
1731         /*
1732          * Only privileged user can chroot
1733          */
1734         error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1735         if (error)
1736                 return (error);
1737
1738         /*
1739          * Disallow open directory descriptors (fchdir() breakouts).
1740          */
1741         if (chroot_allow_open_directories == 0 ||
1742            (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1743                 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1744                         return (error);
1745         }
1746         if ((vp = nch->ncp->nc_vp) == NULL)
1747                 return (ENOENT);
1748
1749         if ((error = vget(vp, LK_SHARED)) != 0)
1750                 return (error);
1751
1752         /*
1753          * Check the validity of vp as a directory to change to and
1754          * associate it with rdir/jdir.
1755          */
1756         error = checkvp_chdir(vp, td);
1757         vn_unlock(vp);                  /* leave reference intact */
1758         if (error == 0) {
1759                 vrele(fdp->fd_rdir);
1760                 fdp->fd_rdir = vp;      /* reference inherited by fd_rdir */
1761                 cache_drop(&fdp->fd_nrdir);
1762                 cache_copy(nch, &fdp->fd_nrdir);
1763                 if (fdp->fd_jdir == NULL) {
1764                         fdp->fd_jdir = vp;
1765                         vref(fdp->fd_jdir);
1766                         cache_copy(nch, &fdp->fd_njdir);
1767                 }
1768         } else {
1769                 vrele(vp);
1770         }
1771         return (error);
1772 }
1773
1774 /*
1775  * chroot_args(char *path)
1776  *
1777  * Change notion of root (``/'') directory.
1778  */
1779 int
1780 sys_chroot(struct chroot_args *uap)
1781 {
1782         struct thread *td __debugvar = curthread;
1783         struct nlookupdata nd;
1784         int error;
1785
1786         KKASSERT(td->td_proc);
1787         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1788         if (error == 0) {
1789                 nd.nl_flags |= NLC_EXEC;
1790                 error = nlookup(&nd);
1791                 if (error == 0)
1792                         error = kern_chroot(&nd.nl_nch);
1793         }
1794         nlookup_done(&nd);
1795         return(error);
1796 }
1797
1798 int
1799 sys_chroot_kernel(struct chroot_kernel_args *uap)
1800 {
1801         struct thread *td = curthread;
1802         struct nlookupdata nd;
1803         struct nchandle *nch;
1804         struct vnode *vp;
1805         int error;
1806
1807         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1808         if (error)
1809                 goto error_nond;
1810
1811         error = nlookup(&nd);
1812         if (error)
1813                 goto error_out;
1814
1815         nch = &nd.nl_nch;
1816
1817         error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1818         if (error)
1819                 goto error_out;
1820
1821         if ((vp = nch->ncp->nc_vp) == NULL) {
1822                 error = ENOENT;
1823                 goto error_out;
1824         }
1825
1826         if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1827                 goto error_out;
1828
1829         kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path);
1830         get_mplock();
1831         vfs_cache_setroot(vp, cache_hold(nch));
1832         rel_mplock();
1833
1834 error_out:
1835         nlookup_done(&nd);
1836 error_nond:
1837         return(error);
1838 }
1839
1840 /*
1841  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1842  * determine whether it is legal to chdir to the vnode.  The vnode's state
1843  * is not changed by this call.
1844  */
1845 static int
1846 checkvp_chdir(struct vnode *vp, struct thread *td)
1847 {
1848         int error;
1849
1850         if (vp->v_type != VDIR)
1851                 error = ENOTDIR;
1852         else
1853                 error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
1854         return (error);
1855 }
1856
1857 int
1858 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1859 {
1860         struct thread *td = curthread;
1861         struct proc *p = td->td_proc;
1862         struct lwp *lp = td->td_lwp;
1863         struct filedesc *fdp = p->p_fd;
1864         int cmode, flags;
1865         struct file *nfp;
1866         struct file *fp;
1867         struct vnode *vp;
1868         int type, indx, error = 0;
1869         struct flock lf;
1870
1871         if ((oflags & O_ACCMODE) == O_ACCMODE)
1872                 return (EINVAL);
1873         flags = FFLAGS(oflags);
1874         error = falloc(lp, &nfp, NULL);
1875         if (error)
1876                 return (error);
1877         fp = nfp;
1878         cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1879
1880         /*
1881          * XXX p_dupfd is a real mess.  It allows a device to return a
1882          * file descriptor to be duplicated rather then doing the open
1883          * itself.
1884          */
1885         lp->lwp_dupfd = -1;
1886
1887         /*
1888          * Call vn_open() to do the lookup and assign the vnode to the
1889          * file pointer.  vn_open() does not change the ref count on fp
1890          * and the vnode, on success, will be inherited by the file pointer
1891          * and unlocked.
1892          *
1893          * Request a shared lock on the vnode if possible.
1894          */
1895         nd->nl_flags |= NLC_LOCKVP;
1896         if ((flags & (O_CREAT|O_TRUNC)) == 0)
1897                 nd->nl_flags |= NLC_SHAREDLOCK;
1898
1899         error = vn_open(nd, fp, flags, cmode);
1900         nlookup_done(nd);
1901
1902         if (error) {
1903                 /*
1904                  * handle special fdopen() case.  bleh.  dupfdopen() is
1905                  * responsible for dropping the old contents of ofiles[indx]
1906                  * if it succeeds.
1907                  *
1908                  * Note that fsetfd() will add a ref to fp which represents
1909                  * the fd_files[] assignment.  We must still drop our
1910                  * reference.
1911                  */
1912                 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1913                         if (fdalloc(p, 0, &indx) == 0) {
1914                                 error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
1915                                 if (error == 0) {
1916                                         *res = indx;
1917                                         fdrop(fp);      /* our ref */
1918                                         return (0);
1919                                 }
1920                                 fsetfd(fdp, NULL, indx);
1921                         }
1922                 }
1923                 fdrop(fp);      /* our ref */
1924                 if (error == ERESTART)
1925                         error = EINTR;
1926                 return (error);
1927         }
1928
1929         /*
1930          * ref the vnode for ourselves so it can't be ripped out from under
1931          * is.  XXX need an ND flag to request that the vnode be returned
1932          * anyway.
1933          *
1934          * Reserve a file descriptor but do not assign it until the open
1935          * succeeds.
1936          */
1937         vp = (struct vnode *)fp->f_data;
1938         vref(vp);
1939         if ((error = fdalloc(p, 0, &indx)) != 0) {
1940                 fdrop(fp);
1941                 vrele(vp);
1942                 return (error);
1943         }
1944
1945         /*
1946          * If no error occurs the vp will have been assigned to the file
1947          * pointer.
1948          */
1949         lp->lwp_dupfd = 0;
1950
1951         if (flags & (O_EXLOCK | O_SHLOCK)) {
1952                 lf.l_whence = SEEK_SET;
1953                 lf.l_start = 0;
1954                 lf.l_len = 0;
1955                 if (flags & O_EXLOCK)
1956                         lf.l_type = F_WRLCK;
1957                 else
1958                         lf.l_type = F_RDLCK;
1959                 if (flags & FNONBLOCK)
1960                         type = 0;
1961                 else
1962                         type = F_WAIT;
1963
1964                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1965                         /*
1966                          * lock request failed.  Clean up the reserved
1967                          * descriptor.
1968                          */
1969                         vrele(vp);
1970                         fsetfd(fdp, NULL, indx);
1971                         fdrop(fp);
1972                         return (error);
1973                 }
1974                 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */
1975         }
1976 #if 0
1977         /*
1978          * Assert that all regular file vnodes were created with a object.
1979          */
1980         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1981                 ("open: regular file has no backing object after vn_open"));
1982 #endif
1983
1984         vrele(vp);
1985
1986         /*
1987          * release our private reference, leaving the one associated with the
1988          * descriptor table intact.
1989          */
1990         if (oflags & O_CLOEXEC)
1991                 fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
1992         fsetfd(fdp, fp, indx);
1993         fdrop(fp);
1994         *res = indx;
1995         return (error);
1996 }
1997
1998 /*
1999  * open_args(char *path, int flags, int mode)
2000  *
2001  * Check permissions, allocate an open file structure,
2002  * and call the device open routine if any.
2003  */
2004 int
2005 sys_open(struct open_args *uap)
2006 {
2007         struct nlookupdata nd;
2008         int error;
2009
2010         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2011         if (error == 0) {
2012                 error = kern_open(&nd, uap->flags,
2013                                     uap->mode, &uap->sysmsg_result);
2014         }
2015         nlookup_done(&nd);
2016         return (error);
2017 }
2018
2019 /*
2020  * openat_args(int fd, char *path, int flags, int mode)
2021  */
2022 int
2023 sys_openat(struct openat_args *uap)
2024 {
2025         struct nlookupdata nd;
2026         int error;
2027         struct file *fp;
2028
2029         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2030         if (error == 0) {
2031                 error = kern_open(&nd, uap->flags, uap->mode,
2032                                         &uap->sysmsg_result);
2033         }
2034         nlookup_done_at(&nd, fp);
2035         return (error);
2036 }
2037
2038 int
2039 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2040 {
2041         struct thread *td = curthread;
2042         struct proc *p = td->td_proc;
2043         struct vnode *vp;
2044         struct vattr vattr;
2045         int error;
2046         int whiteout = 0;
2047
2048         KKASSERT(p);
2049
2050         VATTR_NULL(&vattr);
2051         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2052         vattr.va_rmajor = rmajor;
2053         vattr.va_rminor = rminor;
2054
2055         switch (mode & S_IFMT) {
2056         case S_IFMT:    /* used by badsect to flag bad sectors */
2057                 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2058                 vattr.va_type = VBAD;
2059                 break;
2060         case S_IFCHR:
2061                 error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2062                 vattr.va_type = VCHR;
2063                 break;
2064         case S_IFBLK:
2065                 error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2066                 vattr.va_type = VBLK;
2067                 break;
2068         case S_IFWHT:
2069                 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2070                 whiteout = 1;
2071                 break;
2072         case S_IFDIR:   /* special directories support for HAMMER */
2073                 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2074                 vattr.va_type = VDIR;
2075                 break;
2076         default:
2077                 error = EINVAL;
2078                 break;
2079         }
2080
2081         if (error)
2082                 return (error);
2083
2084         bwillinode(1);
2085         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2086         if ((error = nlookup(nd)) != 0)
2087                 return (error);
2088         if (nd->nl_nch.ncp->nc_vp)
2089                 return (EEXIST);
2090         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2091                 return (error);
2092
2093         if (whiteout) {
2094                 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2095                                       nd->nl_cred, NAMEI_CREATE);
2096         } else {
2097                 vp = NULL;
2098                 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2099                                    &vp, nd->nl_cred, &vattr);
2100                 if (error == 0)
2101                         vput(vp);
2102         }
2103         return (error);
2104 }
2105
2106 /*
2107  * mknod_args(char *path, int mode, int dev)
2108  *
2109  * Create a special file.
2110  */
2111 int
2112 sys_mknod(struct mknod_args *uap)
2113 {
2114         struct nlookupdata nd;
2115         int error;
2116
2117         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2118         if (error == 0) {
2119                 error = kern_mknod(&nd, uap->mode,
2120                                    umajor(uap->dev), uminor(uap->dev));
2121         }
2122         nlookup_done(&nd);
2123         return (error);
2124 }
2125
2126 /*
2127  * mknodat_args(int fd, char *path, mode_t mode, dev_t dev)
2128  *
2129  * Create a special file.  The path is relative to the directory associated
2130  * with fd.
2131  */
2132 int
2133 sys_mknodat(struct mknodat_args *uap)
2134 {
2135         struct nlookupdata nd;
2136         struct file *fp;
2137         int error;
2138
2139         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2140         if (error == 0) {
2141                 error = kern_mknod(&nd, uap->mode,
2142                                    umajor(uap->dev), uminor(uap->dev));
2143         }
2144         nlookup_done_at(&nd, fp);
2145         return (error);
2146 }
2147
2148 int
2149 kern_mkfifo(struct nlookupdata *nd, int mode)
2150 {
2151         struct thread *td = curthread;
2152         struct proc *p = td->td_proc;
2153         struct vattr vattr;
2154         struct vnode *vp;
2155         int error;
2156
2157         bwillinode(1);
2158
2159         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2160         if ((error = nlookup(nd)) != 0)
2161                 return (error);
2162         if (nd->nl_nch.ncp->nc_vp)
2163                 return (EEXIST);
2164         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2165                 return (error);
2166
2167         VATTR_NULL(&vattr);
2168         vattr.va_type = VFIFO;
2169         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2170         vp = NULL;
2171         error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2172         if (error == 0)
2173                 vput(vp);
2174         return (error);
2175 }
2176
2177 /*
2178  * mkfifo_args(char *path, int mode)
2179  *
2180  * Create a named pipe.
2181  */
2182 int
2183 sys_mkfifo(struct mkfifo_args *uap)
2184 {
2185         struct nlookupdata nd;
2186         int error;
2187
2188         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2189         if (error == 0)
2190                 error = kern_mkfifo(&nd, uap->mode);
2191         nlookup_done(&nd);
2192         return (error);
2193 }
2194
2195 /*
2196  * mkfifoat_args(int fd, char *path, mode_t mode)
2197  *
2198  * Create a named pipe.  The path is relative to the directory associated
2199  * with fd.
2200  */
2201 int
2202 sys_mkfifoat(struct mkfifoat_args *uap)
2203 {
2204         struct nlookupdata nd;
2205         struct file *fp;
2206         int error;
2207
2208         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2209         if (error == 0)
2210                 error = kern_mkfifo(&nd, uap->mode);
2211         nlookup_done_at(&nd, fp);
2212         return (error);
2213 }
2214
2215 static int hardlink_check_uid = 0;
2216 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2217     &hardlink_check_uid, 0,
2218     "Unprivileged processes cannot create hard links to files owned by other "
2219     "users");
2220 static int hardlink_check_gid = 0;
2221 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2222     &hardlink_check_gid, 0,
2223     "Unprivileged processes cannot create hard links to files owned by other "
2224     "groups");
2225
2226 static int
2227 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2228 {
2229         struct vattr va;
2230         int error;
2231
2232         /*
2233          * Shortcut if disabled
2234          */
2235         if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2236                 return (0);
2237
2238         /*
2239          * Privileged user can always hardlink
2240          */
2241         if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2242                 return (0);
2243
2244         /*
2245          * Otherwise only if the originating file is owned by the
2246          * same user or group.  Note that any group is allowed if
2247          * the file is owned by the caller.
2248          */
2249         error = VOP_GETATTR(vp, &va);
2250         if (error != 0)
2251                 return (error);
2252
2253         if (hardlink_check_uid) {
2254                 if (cred->cr_uid != va.va_uid)
2255                         return (EPERM);
2256         }
2257
2258         if (hardlink_check_gid) {
2259                 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2260                         return (EPERM);
2261         }
2262
2263         return (0);
2264 }
2265
2266 int
2267 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2268 {
2269         struct thread *td = curthread;
2270         struct vnode *vp;
2271         int error;
2272
2273         /*
2274          * Lookup the source and obtained a locked vnode.
2275          *
2276          * You may only hardlink a file which you have write permission
2277          * on or which you own.
2278          *
2279          * XXX relookup on vget failure / race ?
2280          */
2281         bwillinode(1);
2282         nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2283         if ((error = nlookup(nd)) != 0)
2284                 return (error);
2285         vp = nd->nl_nch.ncp->nc_vp;
2286         KKASSERT(vp != NULL);
2287         if (vp->v_type == VDIR)
2288                 return (EPERM);         /* POSIX */
2289         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2290                 return (error);
2291         if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2292                 return (error);
2293
2294         /*
2295          * Unlock the source so we can lookup the target without deadlocking
2296          * (XXX vp is locked already, possible other deadlock?).  The target
2297          * must not exist.
2298          */
2299         KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2300         nd->nl_flags &= ~NLC_NCPISLOCKED;
2301         cache_unlock(&nd->nl_nch);
2302         vn_unlock(vp);
2303
2304         linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2305         if ((error = nlookup(linknd)) != 0) {
2306                 vrele(vp);
2307                 return (error);
2308         }
2309         if (linknd->nl_nch.ncp->nc_vp) {
2310                 vrele(vp);
2311                 return (EEXIST);
2312         }
2313         error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
2314         if (error) {
2315                 vrele(vp);
2316                 return (error);
2317         }
2318
2319         /*
2320          * Finally run the new API VOP.
2321          */
2322         error = can_hardlink(vp, td, td->td_ucred);
2323         if (error == 0) {
2324                 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2325                                   vp, linknd->nl_cred);
2326         }
2327         vput(vp);
2328         return (error);
2329 }
2330
2331 /*
2332  * link_args(char *path, char *link)
2333  *
2334  * Make a hard file link.
2335  */
2336 int
2337 sys_link(struct link_args *uap)
2338 {
2339         struct nlookupdata nd, linknd;
2340         int error;
2341
2342         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2343         if (error == 0) {
2344                 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2345                 if (error == 0)
2346                         error = kern_link(&nd, &linknd);
2347                 nlookup_done(&linknd);
2348         }
2349         nlookup_done(&nd);
2350         return (error);
2351 }
2352
2353 /*
2354  * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags)
2355  *
2356  * Make a hard file link. The path1 argument is relative to the directory
2357  * associated with fd1, and similarly the path2 argument is relative to
2358  * the directory associated with fd2.
2359  */
2360 int
2361 sys_linkat(struct linkat_args *uap)
2362 {
2363         struct nlookupdata nd, linknd;
2364         struct file *fp1, *fp2;
2365         int error;
2366
2367         error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE,
2368             (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0);
2369         if (error == 0) {
2370                 error = nlookup_init_at(&linknd, &fp2, uap->fd2,
2371                     uap->path2, UIO_USERSPACE, 0);
2372                 if (error == 0)
2373                         error = kern_link(&nd, &linknd);
2374                 nlookup_done_at(&linknd, fp2);
2375         }
2376         nlookup_done_at(&nd, fp1);
2377         return (error);
2378 }
2379
2380 int
2381 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2382 {
2383         struct vattr vattr;
2384         struct vnode *vp;
2385         struct vnode *dvp;
2386         int error;
2387
2388         bwillinode(1);
2389         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2390         if ((error = nlookup(nd)) != 0)
2391                 return (error);
2392         if (nd->nl_nch.ncp->nc_vp)
2393                 return (EEXIST);
2394         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2395                 return (error);
2396         dvp = nd->nl_dvp;
2397         VATTR_NULL(&vattr);
2398         vattr.va_mode = mode;
2399         error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2400         if (error == 0)
2401                 vput(vp);
2402         return (error);
2403 }
2404
2405 /*
2406  * symlink(char *path, char *link)
2407  *
2408  * Make a symbolic link.
2409  */
2410 int
2411 sys_symlink(struct symlink_args *uap)
2412 {
2413         struct thread *td = curthread;
2414         struct nlookupdata nd;
2415         char *path;
2416         int error;
2417         int mode;
2418
2419         path = objcache_get(namei_oc, M_WAITOK);
2420         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2421         if (error == 0) {
2422                 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2423                 if (error == 0) {
2424                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2425                         error = kern_symlink(&nd, path, mode);
2426                 }
2427                 nlookup_done(&nd);
2428         }
2429         objcache_put(namei_oc, path);
2430         return (error);
2431 }
2432
2433 /*
2434  * symlinkat_args(char *path1, int fd, char *path2)
2435  *
2436  * Make a symbolic link.  The path2 argument is relative to the directory
2437  * associated with fd.
2438  */
2439 int
2440 sys_symlinkat(struct symlinkat_args *uap)
2441 {
2442         struct thread *td = curthread;
2443         struct nlookupdata nd;
2444         struct file *fp;
2445         char *path1;
2446         int error;
2447         int mode;
2448
2449         path1 = objcache_get(namei_oc, M_WAITOK);
2450         error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL);
2451         if (error == 0) {
2452                 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2,
2453                     UIO_USERSPACE, 0);
2454                 if (error == 0) {
2455                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2456                         error = kern_symlink(&nd, path1, mode);
2457                 }
2458                 nlookup_done_at(&nd, fp);
2459         }
2460         objcache_put(namei_oc, path1);
2461         return (error);
2462 }
2463
2464 /*
2465  * undelete_args(char *path)
2466  *
2467  * Delete a whiteout from the filesystem.
2468  */
2469 int
2470 sys_undelete(struct undelete_args *uap)
2471 {
2472         struct nlookupdata nd;
2473         int error;
2474
2475         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2476         bwillinode(1);
2477         nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2478         if (error == 0)
2479                 error = nlookup(&nd);
2480         if (error == 0)
2481                 error = ncp_writechk(&nd.nl_nch);
2482         if (error == 0) {
2483                 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2484                                       NAMEI_DELETE);
2485         }
2486         nlookup_done(&nd);
2487         return (error);
2488 }
2489
2490 int
2491 kern_unlink(struct nlookupdata *nd)
2492 {
2493         int error;
2494
2495         bwillinode(1);
2496         nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2497         if ((error = nlookup(nd)) != 0)
2498                 return (error);
2499         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2500                 return (error);
2501         error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2502         return (error);
2503 }
2504
2505 /*
2506  * unlink_args(char *path)
2507  *
2508  * Delete a name from the filesystem.
2509  */
2510 int
2511 sys_unlink(struct unlink_args *uap)
2512 {
2513         struct nlookupdata nd;
2514         int error;
2515
2516         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2517         if (error == 0)
2518                 error = kern_unlink(&nd);
2519         nlookup_done(&nd);
2520         return (error);
2521 }
2522
2523
2524 /*
2525  * unlinkat_args(int fd, char *path, int flags)
2526  *
2527  * Delete the file or directory entry pointed to by fd/path.
2528  */
2529 int
2530 sys_unlinkat(struct unlinkat_args *uap)
2531 {
2532         struct nlookupdata nd;
2533         struct file *fp;
2534         int error;
2535
2536         if (uap->flags & ~AT_REMOVEDIR)
2537                 return (EINVAL);
2538
2539         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2540         if (error == 0) {
2541                 if (uap->flags & AT_REMOVEDIR)
2542                         error = kern_rmdir(&nd);
2543                 else
2544                         error = kern_unlink(&nd);
2545         }
2546         nlookup_done_at(&nd, fp);
2547         return (error);
2548 }
2549
2550 int
2551 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2552 {
2553         struct thread *td = curthread;
2554         struct proc *p = td->td_proc;
2555         struct file *fp;
2556         struct vnode *vp;
2557         struct vattr vattr;
2558         off_t new_offset;
2559         int error;
2560
2561         fp = holdfp(p->p_fd, fd, -1);
2562         if (fp == NULL)
2563                 return (EBADF);
2564         if (fp->f_type != DTYPE_VNODE) {
2565                 error = ESPIPE;
2566                 goto done;
2567         }
2568         vp = (struct vnode *)fp->f_data;
2569
2570         switch (whence) {
2571         case L_INCR:
2572                 spin_lock(&fp->f_spin);
2573                 new_offset = fp->f_offset + offset;
2574                 error = 0;
2575                 break;
2576         case L_XTND:
2577                 error = VOP_GETATTR(vp, &vattr);
2578                 spin_lock(&fp->f_spin);
2579                 new_offset = offset + vattr.va_size;
2580                 break;
2581         case L_SET:
2582                 new_offset = offset;
2583                 error = 0;
2584                 spin_lock(&fp->f_spin);
2585                 break;
2586         default:
2587                 new_offset = 0;
2588                 error = EINVAL;
2589                 spin_lock(&fp->f_spin);
2590                 break;
2591         }
2592
2593         /*
2594          * Validate the seek position.  Negative offsets are not allowed
2595          * for regular files or directories.
2596          *
2597          * Normally we would also not want to allow negative offsets for
2598          * character and block-special devices.  However kvm addresses
2599          * on 64 bit architectures might appear to be negative and must
2600          * be allowed.
2601          */
2602         if (error == 0) {
2603                 if (new_offset < 0 &&
2604                     (vp->v_type == VREG || vp->v_type == VDIR)) {
2605                         error = EINVAL;
2606                 } else {
2607                         fp->f_offset = new_offset;
2608                 }
2609         }
2610         *res = fp->f_offset;
2611         spin_unlock(&fp->f_spin);
2612 done:
2613         fdrop(fp);
2614         return (error);
2615 }
2616
2617 /*
2618  * lseek_args(int fd, int pad, off_t offset, int whence)
2619  *
2620  * Reposition read/write file offset.
2621  */
2622 int
2623 sys_lseek(struct lseek_args *uap)
2624 {
2625         int error;
2626
2627         error = kern_lseek(uap->fd, uap->offset, uap->whence,
2628                            &uap->sysmsg_offset);
2629
2630         return (error);
2631 }
2632
2633 /*
2634  * Check if current process can access given file.  amode is a bitmask of *_OK
2635  * access bits.  flags is a bitmask of AT_* flags.
2636  */
2637 int
2638 kern_access(struct nlookupdata *nd, int amode, int flags)
2639 {
2640         struct vnode *vp;
2641         int error, mode;
2642
2643         if (flags & ~AT_EACCESS)
2644                 return (EINVAL);
2645         nd->nl_flags |= NLC_SHAREDLOCK;
2646         if ((error = nlookup(nd)) != 0)
2647                 return (error);
2648 retry:
2649         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
2650         if (error)
2651                 return (error);
2652
2653         /* Flags == 0 means only check for existence. */
2654         if (amode) {
2655                 mode = 0;
2656                 if (amode & R_OK)
2657                         mode |= VREAD;
2658                 if (amode & W_OK)
2659                         mode |= VWRITE;
2660                 if (amode & X_OK)
2661                         mode |= VEXEC;
2662                 if ((mode & VWRITE) == 0 ||
2663                     (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2664                         error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2665
2666                 /*
2667                  * If the file handle is stale we have to re-resolve the
2668                  * entry with the ncp held exclusively.  This is a hack
2669                  * at the moment.
2670                  */
2671                 if (error == ESTALE) {
2672                         vput(vp);
2673                         cache_unlock(&nd->nl_nch);
2674                         cache_lock(&nd->nl_nch);
2675                         cache_setunresolved(&nd->nl_nch);
2676                         error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2677                         if (error == 0) {
2678                                 vp = NULL;
2679                                 goto retry;
2680                         }
2681                         return(error);
2682                 }
2683         }
2684         vput(vp);
2685         return (error);
2686 }
2687
2688 /*
2689  * access_args(char *path, int flags)
2690  *
2691  * Check access permissions.
2692  */
2693 int
2694 sys_access(struct access_args *uap)
2695 {
2696         struct nlookupdata nd;
2697         int error;
2698
2699         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2700         if (error == 0)
2701                 error = kern_access(&nd, uap->flags, 0);
2702         nlookup_done(&nd);
2703         return (error);
2704 }
2705
2706
2707 /*
2708  * eaccess_args(char *path, int flags)
2709  *
2710  * Check access permissions.
2711  */
2712 int
2713 sys_eaccess(struct eaccess_args *uap)
2714 {
2715         struct nlookupdata nd;
2716         int error;
2717
2718         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2719         if (error == 0)
2720                 error = kern_access(&nd, uap->flags, AT_EACCESS);
2721         nlookup_done(&nd);
2722         return (error);
2723 }
2724
2725
2726 /*
2727  * faccessat_args(int fd, char *path, int amode, int flags)
2728  *
2729  * Check access permissions.
2730  */
2731 int
2732 sys_faccessat(struct faccessat_args *uap)
2733 {
2734         struct nlookupdata nd;
2735         struct file *fp;
2736         int error;
2737
2738         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2739                                 NLC_FOLLOW);
2740         if (error == 0)
2741                 error = kern_access(&nd, uap->amode, uap->flags);
2742         nlookup_done_at(&nd, fp);
2743         return (error);
2744 }
2745
2746 int
2747 kern_stat(struct nlookupdata *nd, struct stat *st)
2748 {
2749         int error;
2750         struct vnode *vp;
2751
2752         nd->nl_flags |= NLC_SHAREDLOCK;
2753         if ((error = nlookup(nd)) != 0)
2754                 return (error);
2755 again:
2756         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2757                 return (ENOENT);
2758
2759         if ((error = vget(vp, LK_SHARED)) != 0)
2760                 return (error);
2761         error = vn_stat(vp, st, nd->nl_cred);
2762
2763         /*
2764          * If the file handle is stale we have to re-resolve the
2765          * entry with the ncp held exclusively.  This is a hack
2766          * at the moment.
2767          */
2768         if (error == ESTALE) {
2769                 vput(vp);
2770                 cache_unlock(&nd->nl_nch);
2771                 cache_lock(&nd->nl_nch);
2772                 cache_setunresolved(&nd->nl_nch);
2773                 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2774                 if (error == 0)
2775                         goto again;
2776         } else {
2777                 vput(vp);
2778         }
2779         return (error);
2780 }
2781
2782 /*
2783  * stat_args(char *path, struct stat *ub)
2784  *
2785  * Get file status; this version follows links.
2786  */
2787 int
2788 sys_stat(struct stat_args *uap)
2789 {
2790         struct nlookupdata nd;
2791         struct stat st;
2792         int error;
2793
2794         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2795         if (error == 0) {
2796                 error = kern_stat(&nd, &st);
2797                 if (error == 0)
2798                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2799         }
2800         nlookup_done(&nd);
2801         return (error);
2802 }
2803
2804 /*
2805  * lstat_args(char *path, struct stat *ub)
2806  *
2807  * Get file status; this version does not follow links.
2808  */
2809 int
2810 sys_lstat(struct lstat_args *uap)
2811 {
2812         struct nlookupdata nd;
2813         struct stat st;
2814         int error;
2815
2816         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2817         if (error == 0) {
2818                 error = kern_stat(&nd, &st);
2819                 if (error == 0)
2820                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2821         }
2822         nlookup_done(&nd);
2823         return (error);
2824 }
2825
2826 /*
2827  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2828  *
2829  * Get status of file pointed to by fd/path.
2830  */
2831 int
2832 sys_fstatat(struct fstatat_args *uap)
2833 {
2834         struct nlookupdata nd;
2835         struct stat st;
2836         int error;
2837         int flags;
2838         struct file *fp;
2839
2840         if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2841                 return (EINVAL);
2842
2843         flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2844
2845         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2846                                 UIO_USERSPACE, flags);
2847         if (error == 0) {
2848                 error = kern_stat(&nd, &st);
2849                 if (error == 0)
2850                         error = copyout(&st, uap->sb, sizeof(*uap->sb));
2851         }
2852         nlookup_done_at(&nd, fp);
2853         return (error);
2854 }
2855
2856 static int
2857 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp)
2858 {
2859         struct nlookupdata nd;
2860         struct vnode *vp;
2861         int error;
2862
2863         vp = NULL;
2864         error = nlookup_init(&nd, path, UIO_USERSPACE, flags);
2865         if (error == 0)
2866                 error = nlookup(&nd);
2867         if (error == 0)
2868                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2869         nlookup_done(&nd);
2870         if (error == 0) {
2871                 error = VOP_PATHCONF(vp, name, sysmsg_regp);
2872                 vput(vp);
2873         }
2874         return (error);
2875 }
2876
2877 /*
2878  * pathconf_Args(char *path, int name)
2879  *
2880  * Get configurable pathname variables.
2881  */
2882 int
2883 sys_pathconf(struct pathconf_args *uap)
2884 {
2885         return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW,
2886                 &uap->sysmsg_reg));
2887 }
2888
2889 /*
2890  * lpathconf_Args(char *path, int name)
2891  *
2892  * Get configurable pathname variables, but don't follow symlinks.
2893  */
2894 int
2895 sys_lpathconf(struct lpathconf_args *uap)
2896 {
2897         return (kern_pathconf(uap->path, uap->name, 0, &uap->sysmsg_reg));
2898 }
2899
2900 /*
2901  * XXX: daver
2902  * kern_readlink isn't properly split yet.  There is a copyin burried
2903  * in VOP_READLINK().
2904  */
2905 int
2906 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2907 {
2908         struct thread *td = curthread;
2909         struct vnode *vp;
2910         struct iovec aiov;
2911         struct uio auio;
2912         int error;
2913
2914         nd->nl_flags |= NLC_SHAREDLOCK;
2915         if ((error = nlookup(nd)) != 0)
2916                 return (error);
2917         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
2918         if (error)
2919                 return (error);
2920         if (vp->v_type != VLNK) {
2921                 error = EINVAL;
2922         } else {
2923                 aiov.iov_base = buf;
2924                 aiov.iov_len = count;
2925                 auio.uio_iov = &aiov;
2926                 auio.uio_iovcnt = 1;
2927                 auio.uio_offset = 0;
2928                 auio.uio_rw = UIO_READ;
2929                 auio.uio_segflg = UIO_USERSPACE;
2930                 auio.uio_td = td;
2931                 auio.uio_resid = count;
2932                 error = VOP_READLINK(vp, &auio, td->td_ucred);
2933         }
2934         vput(vp);
2935         *res = count - auio.uio_resid;
2936         return (error);
2937 }
2938
2939 /*
2940  * readlink_args(char *path, char *buf, int count)
2941  *
2942  * Return target name of a symbolic link.
2943  */
2944 int
2945 sys_readlink(struct readlink_args *uap)
2946 {
2947         struct nlookupdata nd;
2948         int error;
2949
2950         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2951         if (error == 0) {
2952                 error = kern_readlink(&nd, uap->buf, uap->count,
2953                                         &uap->sysmsg_result);
2954         }
2955         nlookup_done(&nd);
2956         return (error);
2957 }
2958
2959 /*
2960  * readlinkat_args(int fd, char *path, char *buf, size_t bufsize)
2961  *
2962  * Return target name of a symbolic link.  The path is relative to the
2963  * directory associated with fd.
2964  */
2965 int
2966 sys_readlinkat(struct readlinkat_args *uap)
2967 {
2968         struct nlookupdata nd;
2969         struct file *fp;
2970         int error;
2971
2972         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2973         if (error == 0) {
2974                 error = kern_readlink(&nd, uap->buf, uap->bufsize,
2975                                         &uap->sysmsg_result);
2976         }
2977         nlookup_done_at(&nd, fp);
2978         return (error);
2979 }
2980
2981 static int
2982 setfflags(struct vnode *vp, int flags)
2983 {
2984         struct thread *td = curthread;
2985         int error;
2986         struct vattr vattr;
2987
2988         /*
2989          * Prevent non-root users from setting flags on devices.  When
2990          * a device is reused, users can retain ownership of the device
2991          * if they are allowed to set flags and programs assume that
2992          * chown can't fail when done as root.
2993          */
2994         if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2995             ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
2996                 return (error);
2997
2998         /*
2999          * note: vget is required for any operation that might mod the vnode
3000          * so VINACTIVE is properly cleared.
3001          */
3002         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3003                 VATTR_NULL(&vattr);
3004                 vattr.va_flags = flags;
3005                 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3006                 vput(vp);
3007         }
3008         return (error);
3009 }
3010
3011 /*
3012  * chflags(char *path, int flags)
3013  *
3014  * Change flags of a file given a path name.
3015  */
3016 int
3017 sys_chflags(struct chflags_args *uap)
3018 {
3019         struct nlookupdata nd;
3020         struct vnode *vp;
3021         int error;
3022
3023         vp = NULL;
3024         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3025         if (error == 0)
3026                 error = nlookup(&nd);
3027         if (error == 0)
3028                 error = ncp_writechk(&nd.nl_nch);
3029         if (error == 0)
3030                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3031         nlookup_done(&nd);
3032         if (error == 0) {
3033                 error = setfflags(vp, uap->flags);
3034                 vrele(vp);
3035         }
3036         return (error);
3037 }
3038
3039 /*
3040  * lchflags(char *path, int flags)
3041  *
3042  * Change flags of a file given a path name, but don't follow symlinks.
3043  */
3044 int
3045 sys_lchflags(struct lchflags_args *uap)
3046 {
3047         struct nlookupdata nd;
3048         struct vnode *vp;
3049         int error;
3050
3051         vp = NULL;
3052         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3053         if (error == 0)
3054                 error = nlookup(&nd);
3055         if (error == 0)
3056                 error = ncp_writechk(&nd.nl_nch);
3057         if (error == 0)
3058                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3059         nlookup_done(&nd);
3060         if (error == 0) {
3061                 error = setfflags(vp, uap->flags);
3062                 vrele(vp);
3063         }
3064         return (error);
3065 }
3066
3067 /*
3068  * fchflags_args(int fd, int flags)
3069  *
3070  * Change flags of a file given a file descriptor.
3071  */
3072 int
3073 sys_fchflags(struct fchflags_args *uap)
3074 {
3075         struct thread *td = curthread;
3076         struct proc *p = td->td_proc;
3077         struct file *fp;
3078         int error;
3079
3080         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3081                 return (error);
3082         if (fp->f_nchandle.ncp)
3083                 error = ncp_writechk(&fp->f_nchandle);
3084         if (error == 0)
3085                 error = setfflags((struct vnode *) fp->f_data, uap->flags);
3086         fdrop(fp);
3087         return (error);
3088 }
3089
3090 /*
3091  * chflagsat_args(int fd, const char *path, int flags, int atflags)
3092  * change flags given a pathname relative to a filedescriptor
3093  */
3094 int sys_chflagsat(struct chflagsat_args *uap)
3095 {
3096         struct nlookupdata nd;
3097         struct vnode *vp;
3098         struct file *fp;
3099         int error;
3100         int lookupflags;
3101
3102         if (uap->atflags & ~AT_SYMLINK_NOFOLLOW)
3103                 return (EINVAL);
3104
3105         lookupflags = (uap->atflags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3106
3107         vp = NULL;
3108         error = nlookup_init_at(&nd, &fp, uap->fd,  uap->path, UIO_USERSPACE, lookupflags);
3109         if (error == 0)
3110                 error = nlookup(&nd);
3111         if (error == 0)
3112                 error = ncp_writechk(&nd.nl_nch);
3113         if (error == 0)
3114                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3115         nlookup_done_at(&nd, fp);
3116         if (error == 0) {
3117                 error = setfflags(vp, uap->flags);
3118                 vrele(vp);
3119         }
3120         return (error);
3121 }
3122
3123
3124 static int
3125 setfmode(struct vnode *vp, int mode)
3126 {
3127         struct thread *td = curthread;
3128         int error;
3129         struct vattr vattr;
3130
3131         /*
3132          * note: vget is required for any operation that might mod the vnode
3133          * so VINACTIVE is properly cleared.
3134          */
3135         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3136                 VATTR_NULL(&vattr);
3137                 vattr.va_mode = mode & ALLPERMS;
3138                 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3139                 vput(vp);
3140         }
3141         return error;
3142 }
3143
3144 int
3145 kern_chmod(struct nlookupdata *nd, int mode)
3146 {
3147         struct vnode *vp;
3148         int error;
3149
3150         if ((error = nlookup(nd)) != 0)
3151                 return (error);
3152         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3153                 return (error);
3154         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3155                 error = setfmode(vp, mode);
3156         vrele(vp);
3157         return (error);
3158 }
3159
3160 /*
3161  * chmod_args(char *path, int mode)
3162  *
3163  * Change mode of a file given path name.
3164  */
3165 int
3166 sys_chmod(struct chmod_args *uap)
3167 {
3168         struct nlookupdata nd;
3169         int error;
3170
3171         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3172         if (error == 0)
3173                 error = kern_chmod(&nd, uap->mode);
3174         nlookup_done(&nd);
3175         return (error);
3176 }
3177
3178 /*
3179  * lchmod_args(char *path, int mode)
3180  *
3181  * Change mode of a file given path name (don't follow links.)
3182  */
3183 int
3184 sys_lchmod(struct lchmod_args *uap)
3185 {
3186         struct nlookupdata nd;
3187         int error;
3188
3189         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3190         if (error == 0)
3191                 error = kern_chmod(&nd, uap->mode);
3192         nlookup_done(&nd);
3193         return (error);
3194 }
3195
3196 /*
3197  * fchmod_args(int fd, int mode)
3198  *
3199  * Change mode of a file given a file descriptor.
3200  */
3201 int
3202 sys_fchmod(struct fchmod_args *uap)
3203 {
3204         struct thread *td = curthread;
3205         struct proc *p = td->td_proc;
3206         struct file *fp;
3207         int error;
3208
3209         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3210                 return (error);
3211         if (fp->f_nchandle.ncp)
3212                 error = ncp_writechk(&fp->f_nchandle);
3213         if (error == 0)
3214                 error = setfmode((struct vnode *)fp->f_data, uap->mode);
3215         fdrop(fp);
3216         return (error);
3217 }
3218
3219 /*
3220  * fchmodat_args(char *path, int mode)
3221  *
3222  * Change mode of a file pointed to by fd/path.
3223  */
3224 int
3225 sys_fchmodat(struct fchmodat_args *uap)
3226 {
3227         struct nlookupdata nd;
3228         struct file *fp;
3229         int error;
3230         int flags;
3231
3232         if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3233                 return (EINVAL);
3234         flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3235
3236         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3237                                 UIO_USERSPACE, flags);
3238         if (error == 0)
3239                 error = kern_chmod(&nd, uap->mode);
3240         nlookup_done_at(&nd, fp);
3241         return (error);
3242 }
3243
3244 static int
3245 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid)
3246 {
3247         struct thread *td = curthread;
3248         int error;
3249         struct vattr vattr;
3250         uid_t o_uid;
3251         gid_t o_gid;
3252         uint64_t size;
3253
3254         /*
3255          * note: vget is required for any operation that might mod the vnode
3256          * so VINACTIVE is properly cleared.
3257          */
3258         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3259                 if ((error = VOP_GETATTR(vp, &vattr)) != 0)
3260                         return error;
3261                 o_uid = vattr.va_uid;
3262                 o_gid = vattr.va_gid;
3263                 size = vattr.va_size;
3264
3265                 VATTR_NULL(&vattr);
3266                 vattr.va_uid = uid;
3267                 vattr.va_gid = gid;
3268                 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3269                 vput(vp);
3270         }
3271
3272         if (error == 0) {
3273                 if (uid == -1)
3274                         uid = o_uid;
3275                 if (gid == -1)
3276                         gid = o_gid;
3277                 VFS_ACCOUNT(mp, o_uid, o_gid, -size);
3278                 VFS_ACCOUNT(mp,   uid,   gid,  size);
3279         }
3280
3281         return error;
3282 }
3283
3284 int
3285 kern_chown(struct nlookupdata *nd, int uid, int gid)
3286 {
3287         struct vnode *vp;
3288         int error;
3289
3290         if ((error = nlookup(nd)) != 0)
3291                 return (error);
3292         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3293                 return (error);
3294         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3295                 error = setfown(nd->nl_nch.mount, vp, uid, gid);
3296         vrele(vp);
3297         return (error);
3298 }
3299
3300 /*
3301  * chown(char *path, int uid, int gid)
3302  *
3303  * Set ownership given a path name.
3304  */
3305 int
3306 sys_chown(struct chown_args *uap)
3307 {
3308         struct nlookupdata nd;
3309         int error;
3310
3311         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3312         if (error == 0)
3313                 error = kern_chown(&nd, uap->uid, uap->gid);
3314         nlookup_done(&nd);
3315         return (error);
3316 }
3317
3318 /*
3319  * lchown_args(char *path, int uid, int gid)
3320  *
3321  * Set ownership given a path name, do not cross symlinks.
3322  */
3323 int
3324 sys_lchown(struct lchown_args *uap)
3325 {
3326         struct nlookupdata nd;
3327         int error;
3328
3329         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3330         if (error == 0)
3331                 error = kern_chown(&nd, uap->uid, uap->gid);
3332         nlookup_done(&nd);
3333         return (error);
3334 }
3335
3336 /*
3337  * fchown_args(int fd, int uid, int gid)
3338  *
3339  * Set ownership given a file descriptor.
3340  */
3341 int
3342 sys_fchown(struct fchown_args *uap)
3343 {
3344         struct thread *td = curthread;
3345         struct proc *p = td->td_proc;
3346         struct file *fp;
3347         int error;
3348
3349         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3350                 return (error);
3351         if (fp->f_nchandle.ncp)
3352                 error = ncp_writechk(&fp->f_nchandle);
3353         if (error == 0)
3354                 error = setfown(p->p_fd->fd_ncdir.mount,
3355                         (struct vnode *)fp->f_data, uap->uid, uap->gid);
3356         fdrop(fp);
3357         return (error);
3358 }
3359
3360 /*
3361  * fchownat(int fd, char *path, int uid, int gid, int flags)
3362  *
3363  * Set ownership of file pointed to by fd/path.
3364  */
3365 int
3366 sys_fchownat(struct fchownat_args *uap)
3367 {
3368         struct nlookupdata nd;
3369         struct file *fp;
3370         int error;
3371         int flags;
3372
3373         if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3374                 return (EINVAL);
3375         flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3376
3377         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3378                                 UIO_USERSPACE, flags);
3379         if (error == 0)
3380                 error = kern_chown(&nd, uap->uid, uap->gid);
3381         nlookup_done_at(&nd, fp);
3382         return (error);
3383 }
3384
3385
3386 static int
3387 getutimes(struct timeval *tvp, struct timespec *tsp)
3388 {
3389         struct timeval tv[2];
3390         int error;
3391
3392         if (tvp == NULL) {
3393                 microtime(&tv[0]);
3394                 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3395                 tsp[1] = tsp[0];
3396         } else {
3397                 if ((error = itimerfix(tvp)) != 0)
3398                         return (error);
3399                 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3400                 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3401         }
3402         return 0;
3403 }
3404
3405 static int
3406 getutimens(const struct timespec *ts, struct timespec *newts, int *nullflag)
3407 {
3408         struct timespec tsnow;
3409         int error;
3410
3411         *nullflag = 0;
3412         nanotime(&tsnow);
3413         if (ts == NULL) {
3414                 newts[0] = tsnow;
3415                 newts[1] = tsnow;
3416                 *nullflag = 1;
3417                 return (0);
3418         }
3419
3420         newts[0] = ts[0];
3421         newts[1] = ts[1];
3422         if (newts[0].tv_nsec == UTIME_OMIT && newts[1].tv_nsec == UTIME_OMIT)
3423                 return (0);
3424         if (newts[0].tv_nsec == UTIME_NOW && newts[1].tv_nsec == UTIME_NOW)
3425                 *nullflag = 1;
3426
3427         if (newts[0].tv_nsec == UTIME_OMIT)
3428                 newts[0].tv_sec = VNOVAL;
3429         else if (newts[0].tv_nsec == UTIME_NOW)
3430                 newts[0] = tsnow;
3431         else if ((error = itimespecfix(&newts[0])) != 0)
3432                 return (error);
3433
3434         if (newts[1].tv_nsec == UTIME_OMIT)
3435                 newts[1].tv_sec = VNOVAL;
3436         else if (newts[1].tv_nsec == UTIME_NOW)
3437                 newts[1] = tsnow;
3438         else if ((error = itimespecfix(&newts[1])) != 0)
3439                 return (error);
3440
3441         return (0);
3442 }
3443
3444 static int
3445 setutimes(struct vnode *vp, struct vattr *vattr,
3446           const struct timespec *ts, int nullflag)
3447 {
3448         struct thread *td = curthread;
3449         int error;
3450
3451         VATTR_NULL(vattr);
3452         vattr->va_atime = ts[0];
3453         vattr->va_mtime = ts[1];
3454         if (nullflag)
3455                 vattr->va_vaflags |= VA_UTIMES_NULL;
3456         error = VOP_SETATTR(vp, vattr, td->td_ucred);
3457
3458         return error;
3459 }
3460
3461 int
3462 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3463 {
3464         struct timespec ts[2];
3465         int error;
3466
3467         if (tptr) {
3468                 if ((error = getutimes(tptr, ts)) != 0)
3469                         return (error);
3470         }
3471         error = kern_utimensat(nd, tptr ? ts : NULL, 0);
3472         return (error);
3473 }
3474
3475 /*
3476  * utimes_args(char *path, struct timeval *tptr)
3477  *
3478  * Set the access and modification times of a file.
3479  */
3480 int
3481 sys_utimes(struct utimes_args *uap)
3482 {
3483         struct timeval tv[2];
3484         struct nlookupdata nd;
3485         int error;
3486
3487         if (uap->tptr) {
3488                 error = copyin(uap->tptr, tv, sizeof(tv));
3489                 if (error)
3490                         return (error);
3491         }
3492         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3493         if (error == 0)
3494                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3495         nlookup_done(&nd);
3496         return (error);
3497 }
3498
3499 /*
3500  * lutimes_args(char *path, struct timeval *tptr)
3501  *
3502  * Set the access and modification times of a file.
3503  */
3504 int
3505 sys_lutimes(struct lutimes_args *uap)
3506 {
3507         struct timeval tv[2];
3508         struct nlookupdata nd;
3509         int error;
3510
3511         if (uap->tptr) {
3512                 error = copyin(uap->tptr, tv, sizeof(tv));
3513                 if (error)
3514                         return (error);
3515         }
3516         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3517         if (error == 0)
3518                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3519         nlookup_done(&nd);
3520         return (error);
3521 }
3522
3523 /*
3524  * Set utimes on a file descriptor.  The creds used to open the
3525  * file are used to determine whether the operation is allowed
3526  * or not.
3527  */
3528 int
3529 kern_futimens(int fd, struct timespec *ts)
3530 {
3531         struct thread *td = curthread;
3532         struct proc *p = td->td_proc;
3533         struct timespec newts[2];
3534         struct file *fp;
3535         struct vnode *vp;
3536         struct vattr vattr;
3537         int nullflag;
3538         int error;
3539
3540         error = getutimens(ts, newts, &nullflag);
3541         if (error)
3542                 return (error);
3543         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3544                 return (error);
3545         if (fp->f_nchandle.ncp)
3546                 error = ncp_writechk(&fp->f_nchandle);
3547         if (error == 0) {
3548                 vp = fp->f_data;
3549                 error = vget(vp, LK_EXCLUSIVE);
3550                 if (error == 0) {
3551                         error = VOP_GETATTR(vp, &vattr);
3552                         if (error == 0) {
3553                                 error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3554                                                    fp->f_cred);
3555                         }
3556                         if (error == 0) {
3557                                 error = setutimes(vp, &vattr, newts, nullflag);
3558                         }
3559                         vput(vp);
3560                 }
3561         }
3562         fdrop(fp);
3563         return (error);
3564 }
3565
3566 /*
3567  * futimens_args(int fd, struct timespec *ts)
3568  *
3569  * Set the access and modification times of a file.
3570  */
3571 int
3572 sys_futimens(struct futimens_args *uap)
3573 {
3574         struct timespec ts[2];
3575         int error;
3576
3577         if (uap->ts) {
3578                 error = copyin(uap->ts, ts, sizeof(ts));
3579                 if (error)
3580                         return (error);
3581         }
3582         error = kern_futimens(uap->fd, uap->ts ? ts : NULL);
3583         return (error);
3584 }
3585
3586 int
3587 kern_futimes(int fd, struct timeval *tptr)
3588 {
3589         struct timespec ts[2];
3590         int error;
3591
3592         if (tptr) {
3593                 if ((error = getutimes(tptr, ts)) != 0)
3594                         return (error);
3595         }
3596         error = kern_futimens(fd, tptr ? ts : NULL);
3597         return (error);
3598 }
3599
3600 /*
3601  * futimes_args(int fd, struct timeval *tptr)
3602  *
3603  * Set the access and modification times of a file.
3604  */
3605 int
3606 sys_futimes(struct futimes_args *uap)
3607 {
3608         struct timeval tv[2];
3609         int error;
3610
3611         if (uap->tptr) {
3612                 error = copyin(uap->tptr, tv, sizeof(tv));
3613                 if (error)
3614                         return (error);
3615         }
3616         error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3617         return (error);
3618 }
3619
3620 int
3621 kern_utimensat(struct nlookupdata *nd, const struct timespec *ts, int flags)
3622 {
3623         struct timespec newts[2];
3624         struct vnode *vp;
3625         struct vattr vattr;
3626         int nullflag;
3627         int error;
3628
3629         if (flags & ~AT_SYMLINK_NOFOLLOW)
3630                 return (EINVAL);
3631
3632         error = getutimens(ts, newts, &nullflag);
3633         if (error)
3634                 return (error);
3635
3636         nd->nl_flags |= NLC_OWN | NLC_WRITE;
3637         if ((error = nlookup(nd)) != 0)
3638                 return (error);
3639         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3640                 return (error);
3641         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3642                 return (error);
3643         if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3644                 error = vget(vp, LK_EXCLUSIVE);
3645                 if (error == 0) {
3646                         error = setutimes(vp, &vattr, newts, nullflag);
3647                         vput(vp);
3648                 }
3649         }
3650         vrele(vp);
3651         return (error);
3652 }
3653
3654 /*
3655  * utimensat_args(int fd, const char *path, const struct timespec *ts, int flags);
3656  *
3657  * Set file access and modification times of a file.
3658  */
3659 int
3660 sys_utimensat(struct utimensat_args *uap)
3661 {
3662         struct timespec ts[2];
3663         struct nlookupdata nd;
3664         struct file *fp;
3665         int error;
3666         int flags;
3667
3668         if (uap->ts) {
3669                 error = copyin(uap->ts, ts, sizeof(ts));
3670                 if (error)
3671                         return (error);
3672         }
3673
3674         flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3675         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3676                                 UIO_USERSPACE, flags);
3677         if (error == 0)
3678                 error = kern_utimensat(&nd, uap->ts ? ts : NULL, uap->flags);
3679         nlookup_done_at(&nd, fp);
3680         return (error);
3681 }
3682
3683 int
3684 kern_truncate(struct nlookupdata *nd, off_t length)
3685 {
3686         struct vnode *vp;
3687         struct vattr vattr;
3688         int error;
3689         uid_t uid = 0;
3690         gid_t gid = 0;
3691         uint64_t old_size = 0;
3692
3693         if (length < 0)
3694                 return(EINVAL);
3695         nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3696         if ((error = nlookup(nd)) != 0)
3697                 return (error);
3698         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3699                 return (error);
3700         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3701                 return (error);
3702         error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
3703         if (error) {
3704                 vrele(vp);
3705                 return (error);
3706         }
3707         if (vp->v_type == VDIR) {
3708                 error = EISDIR;
3709                 goto done;
3710         }
3711         if (vfs_quota_enabled) {
3712                 error = VOP_GETATTR(vp, &vattr);
3713                 KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0"));
3714                 uid = vattr.va_uid;
3715                 gid = vattr.va_gid;
3716                 old_size = vattr.va_size;
3717         }
3718
3719         if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3720                 VATTR_NULL(&vattr);
3721                 vattr.va_size = length;
3722                 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3723                 VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size);
3724         }
3725 done:
3726         vput(vp);
3727         return (error);
3728 }
3729
3730 /*
3731  * truncate(char *path, int pad, off_t length)
3732  *
3733  * Truncate a file given its path name.
3734  */
3735 int
3736 sys_truncate(struct truncate_args *uap)
3737 {
3738         struct nlookupdata nd;
3739         int error;
3740
3741         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3742         if (error == 0)
3743                 error = kern_truncate(&nd, uap->length);
3744         nlookup_done(&nd);
3745         return error;
3746 }
3747
3748 int
3749 kern_ftruncate(int fd, off_t length)
3750 {
3751         struct thread *td = curthread;
3752         struct proc *p = td->td_proc;
3753         struct vattr vattr;
3754         struct vnode *vp;
3755         struct file *fp;
3756         int error;
3757         uid_t uid = 0;
3758         gid_t gid = 0;
3759         uint64_t old_size = 0;
3760         struct mount *mp;
3761
3762         if (length < 0)
3763                 return(EINVAL);
3764         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3765                 return (error);
3766         if (fp->f_nchandle.ncp) {
3767                 error = ncp_writechk(&fp->f_nchandle);
3768                 if (error)
3769                         goto done;
3770         }
3771         if ((fp->f_flag & FWRITE) == 0) {
3772                 error = EINVAL;
3773                 goto done;
3774         }
3775         if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */
3776                 error = EINVAL;
3777                 goto done;
3778         }
3779         vp = (struct vnode *)fp->f_data;
3780         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3781         if (vp->v_type == VDIR) {
3782                 error = EISDIR;
3783                 vn_unlock(vp);
3784                 goto done;
3785         }
3786
3787         if (vfs_quota_enabled) {
3788                 error = VOP_GETATTR(vp, &vattr);
3789                 KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0"));
3790                 uid = vattr.va_uid;
3791                 gid = vattr.va_gid;
3792                 old_size = vattr.va_size;
3793         }
3794
3795         if ((error = vn_writechk(vp, NULL)) == 0) {
3796                 VATTR_NULL(&vattr);
3797                 vattr.va_size = length;
3798                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3799                 mp = vq_vptomp(vp);
3800                 VFS_ACCOUNT(mp, uid, gid, length - old_size);
3801         }
3802         vn_unlock(vp);
3803 done:
3804         fdrop(fp);
3805         return (error);
3806 }
3807
3808 /*
3809  * ftruncate_args(int fd, int pad, off_t length)
3810  *
3811  * Truncate a file given a file descriptor.
3812  */
3813 int
3814 sys_ftruncate(struct ftruncate_args *uap)
3815 {
3816         int error;
3817
3818         error = kern_ftruncate(uap->fd, uap->length);
3819
3820         return (error);
3821 }
3822
3823 /*
3824  * fsync(int fd)
3825  *
3826  * Sync an open file.
3827  */
3828 int
3829 sys_fsync(struct fsync_args *uap)
3830 {
3831         struct thread *td = curthread;
3832         struct proc *p = td->td_proc;
3833         struct vnode *vp;
3834         struct file *fp;
3835         vm_object_t obj;
3836         int error;
3837
3838         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3839                 return (error);
3840         vp = (struct vnode *)fp->f_data;
3841         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3842         if ((obj = vp->v_object) != NULL) {
3843                 if (vp->v_mount == NULL ||
3844                     (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) {
3845                         vm_object_page_clean(obj, 0, 0, 0);
3846                 }
3847         }
3848         error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3849         if (error == 0 && vp->v_mount)
3850                 error = buf_fsync(vp);
3851         vn_unlock(vp);
3852         fdrop(fp);
3853
3854         return (error);
3855 }
3856
3857 int
3858 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3859 {
3860         struct nchandle fnchd;
3861         struct nchandle tnchd;
3862         struct namecache *ncp;
3863         struct vnode *fdvp;
3864         struct vnode *tdvp;
3865         struct mount *mp;
3866         int error;
3867         u_int fncp_gen;
3868         u_int tncp_gen;
3869
3870         bwillinode(1);
3871         fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3872         if ((error = nlookup(fromnd)) != 0)
3873                 return (error);
3874         if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3875                 return (ENOENT);
3876         fnchd.mount = fromnd->nl_nch.mount;
3877         cache_hold(&fnchd);
3878
3879         /*
3880          * unlock the source nch so we can lookup the target nch without
3881          * deadlocking.  The target may or may not exist so we do not check
3882          * for a target vp like kern_mkdir() and other creation functions do.
3883          *
3884          * The source and target directories are ref'd and rechecked after
3885          * everything is relocked to determine if the source or target file
3886          * has been renamed.
3887          */
3888         KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3889         fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3890
3891         fncp_gen = fromnd->nl_nch.ncp->nc_generation;
3892
3893         cache_unlock(&fromnd->nl_nch);
3894
3895         tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3896         if ((error = nlookup(tond)) != 0) {
3897                 cache_drop(&fnchd);
3898                 return (error);
3899         }
3900         tncp_gen = tond->nl_nch.ncp->nc_generation;
3901
3902         if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3903                 cache_drop(&fnchd);
3904                 return (ENOENT);
3905         }
3906         tnchd.mount = tond->nl_nch.mount;
3907         cache_hold(&tnchd);
3908
3909         /*
3910          * If the source and target are the same there is nothing to do
3911          */
3912         if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3913                 cache_drop(&fnchd);
3914                 cache_drop(&tnchd);
3915                 return (0);
3916         }
3917
3918         /*
3919          * Mount points cannot be renamed or overwritten
3920          */
3921         if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3922             NCF_ISMOUNTPT
3923         ) {
3924                 cache_drop(&fnchd);
3925                 cache_drop(&tnchd);
3926                 return (EINVAL);
3927         }
3928
3929         /*
3930          * Relock the source ncp.  cache_relock() will deal with any
3931          * deadlocks against the already-locked tond and will also
3932          * make sure both are resolved.
3933          *
3934          * NOTE AFTER RELOCKING: The source or target ncp may have become
3935          * invalid while they were unlocked, nc_vp and nc_mount could
3936          * be NULL.
3937          */
3938         cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
3939                      &tond->nl_nch, tond->nl_cred);
3940         fromnd->nl_flags |= NLC_NCPISLOCKED;
3941
3942         /*
3943          * If the namecache generation changed for either fromnd or tond,
3944          * we must retry.
3945          */
3946         if (fromnd->nl_nch.ncp->nc_generation != fncp_gen ||
3947             tond->nl_nch.ncp->nc_generation != tncp_gen) {
3948                 kprintf("kern_rename: retry due to gen on: "
3949                         "\"%s\" -> \"%s\"\n",
3950                         fromnd->nl_nch.ncp->nc_name,
3951                         tond->nl_nch.ncp->nc_name);
3952                 cache_drop(&fnchd);
3953                 cache_drop(&tnchd);
3954                 return (EAGAIN);
3955         }
3956
3957         /*
3958          * If either fromnd or tond are marked destroyed a ripout occured
3959          * out from under us and we must retry.
3960          */
3961         if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) ||
3962             fromnd->nl_nch.ncp->nc_vp == NULL ||
3963             (tond->nl_nch.ncp->nc_flag & NCF_DESTROYED)) {
3964                 kprintf("kern_rename: retry due to ripout on: "
3965                         "\"%s\" -> \"%s\"\n",
3966                         fromnd->nl_nch.ncp->nc_name,
3967                         tond->nl_nch.ncp->nc_name);
3968                 cache_drop(&fnchd);
3969                 cache_drop(&tnchd);
3970                 return (EAGAIN);
3971         }
3972
3973         /*
3974          * Make sure the parent directories linkages are the same.
3975          * XXX shouldn't be needed any more w/ generation check above.
3976          */
3977         if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3978             tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3979                 cache_drop(&fnchd);
3980                 cache_drop(&tnchd);
3981                 return (ENOENT);
3982         }
3983
3984         /*
3985          * Both the source and target must be within the same filesystem and
3986          * in the same filesystem as their parent directories within the
3987          * namecache topology.
3988          *
3989          * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3990          */
3991         mp = fnchd.mount;
3992         if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3993             mp != tond->nl_nch.mount) {
3994                 cache_drop(&fnchd);
3995                 cache_drop(&tnchd);
3996                 return (EXDEV);
3997         }
3998
3999         /*
4000          * Make sure the mount point is writable
4001          */
4002         if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
4003                 cache_drop(&fnchd);
4004                 cache_drop(&tnchd);
4005                 return (error);
4006         }
4007
4008         /*
4009          * If the target exists and either the source or target is a directory,
4010          * then both must be directories.
4011          *
4012          * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
4013          * have become NULL.
4014          */
4015         if (tond->nl_nch.ncp->nc_vp) {
4016                 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
4017                         error = ENOENT;
4018                 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
4019                         if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
4020                                 error = ENOTDIR;
4021                 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
4022                         error = EISDIR;
4023                 }
4024         }
4025
4026         /*
4027          * You cannot rename a source into itself or a subdirectory of itself.
4028          * We check this by travsersing the target directory upwards looking
4029          * for a match against the source.
4030          *
4031          * XXX MPSAFE
4032          */
4033         if (error == 0) {
4034                 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
4035                         if (fromnd->nl_nch.ncp == ncp) {
4036                                 error = EINVAL;
4037                                 break;
4038                         }
4039                 }
4040         }
4041
4042         cache_drop(&fnchd);
4043         cache_drop(&tnchd);
4044
4045         /*
4046          * Even though the namespaces are different, they may still represent
4047          * hardlinks to the same file.  The filesystem might have a hard time
4048          * with this so we issue a NREMOVE of the source instead of a NRENAME
4049          * when we detect the situation.
4050          */
4051         if (error == 0) {
4052                 fdvp = fromnd->nl_dvp;
4053                 tdvp = tond->nl_dvp;
4054                 if (fdvp == NULL || tdvp == NULL) {
4055                         error = EPERM;
4056                 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
4057                         error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
4058                                             fromnd->nl_cred);
4059                 } else {
4060                         error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
4061                                             fdvp, tdvp, tond->nl_cred);
4062                 }
4063         }
4064         return (error);
4065 }
4066
4067 /*
4068  * rename_args(char *from, char *to)
4069  *
4070  * Rename files.  Source and destination must either both be directories,
4071  * or both not be directories.  If target is a directory, it must be empty.
4072  */
4073 int
4074 sys_rename(struct rename_args *uap)
4075 {
4076         struct nlookupdata fromnd, tond;
4077         int error;
4078
4079         do {
4080                 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
4081                 if (error == 0) {
4082                         error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
4083                         if (error == 0)
4084                                 error = kern_rename(&fromnd, &tond);
4085                         nlookup_done(&tond);
4086                 }
4087                 nlookup_done(&fromnd);
4088         } while (error == EAGAIN);
4089         return (error);
4090 }
4091
4092 /*
4093  * renameat_args(int oldfd, char *old, int newfd, char *new)
4094  *
4095  * Rename files using paths relative to the directories associated with
4096  * oldfd and newfd.  Source and destination must either both be directories,
4097  * or both not be directories.  If target is a directory, it must be empty.
4098  */
4099 int
4100 sys_renameat(struct renameat_args *uap)
4101 {
4102         struct nlookupdata oldnd, newnd;
4103         struct file *oldfp, *newfp;
4104         int error;
4105
4106         do {
4107                 error = nlookup_init_at(&oldnd, &oldfp,
4108                                         uap->oldfd, uap->old,
4109                                         UIO_USERSPACE, 0);
4110                 if (error == 0) {
4111                         error = nlookup_init_at(&newnd, &newfp,
4112                                                 uap->newfd, uap->new,
4113                                                 UIO_USERSPACE, 0);
4114                         if (error == 0)
4115                                 error = kern_rename(&oldnd, &newnd);
4116                         nlookup_done_at(&newnd, newfp);
4117                 }
4118                 nlookup_done_at(&oldnd, oldfp);
4119         } while (error == EAGAIN);
4120         return (error);
4121 }
4122
4123 int
4124 kern_mkdir(struct nlookupdata *nd, int mode)
4125 {
4126         struct thread *td = curthread;
4127         struct proc *p = td->td_proc;
4128         struct vnode *vp;
4129         struct vattr vattr;
4130         int error;
4131
4132         bwillinode(1);
4133         nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
4134         if ((error = nlookup(nd)) != 0)
4135                 return (error);
4136
4137         if (nd->nl_nch.ncp->nc_vp)
4138                 return (EEXIST);
4139         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4140                 return (error);
4141         VATTR_NULL(&vattr);
4142         vattr.va_type = VDIR;
4143         vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
4144
4145         vp = NULL;
4146         error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
4147         if (error == 0)
4148                 vput(vp);
4149         return (error);
4150 }
4151
4152 /*
4153  * mkdir_args(char *path, int mode)
4154  *
4155  * Make a directory file.
4156  */
4157 int
4158 sys_mkdir(struct mkdir_args *uap)
4159 {
4160         struct nlookupdata nd;
4161         int error;
4162
4163         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4164         if (error == 0)
4165                 error = kern_mkdir(&nd, uap->mode);
4166         nlookup_done(&nd);
4167         return (error);
4168 }
4169
4170 /*
4171  * mkdirat_args(int fd, char *path, mode_t mode)
4172  *
4173  * Make a directory file.  The path is relative to the directory associated
4174  * with fd.
4175  */
4176 int
4177 sys_mkdirat(struct mkdirat_args *uap)
4178 {
4179         struct nlookupdata nd;
4180         struct file *fp;
4181         int error;
4182
4183         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
4184         if (error == 0)
4185                 error = kern_mkdir(&nd, uap->mode);
4186         nlookup_done_at(&nd, fp);
4187         return (error);
4188 }
4189
4190 int
4191 kern_rmdir(struct nlookupdata *nd)
4192 {
4193         int error;
4194
4195         bwillinode(1);
4196         nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
4197         if ((error = nlookup(nd)) != 0)
4198                 return (error);
4199
4200         /*
4201          * Do not allow directories representing mount points to be
4202          * deleted, even if empty.  Check write perms on mount point
4203          * in case the vnode is aliased (aka nullfs).
4204          */
4205         if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
4206                 return (EBUSY);
4207         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4208                 return (error);
4209         error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
4210         return (error);
4211 }
4212
4213 /*
4214  * rmdir_args(char *path)
4215  *
4216  * Remove a directory file.
4217  */
4218 int
4219 sys_rmdir(struct rmdir_args *uap)
4220 {
4221         struct nlookupdata nd;
4222         int error;
4223
4224         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4225         if (error == 0)
4226                 error = kern_rmdir(&nd);
4227         nlookup_done(&nd);
4228         return (error);
4229 }
4230
4231 int
4232 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
4233                    enum uio_seg direction)
4234 {
4235         struct thread *td = curthread;
4236         struct proc *p = td->td_proc;
4237         struct vnode *vp;
4238         struct file *fp;
4239         struct uio auio;
4240         struct iovec aiov;
4241         off_t loff;
4242         int error, eofflag;
4243
4244         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
4245                 return (error);
4246         if ((fp->f_flag & FREAD) == 0) {
4247                 error = EBADF;
4248                 goto done;
4249         }
4250         vp = (struct vnode *)fp->f_data;
4251         if (vp->v_type != VDIR) {
4252                 error = EINVAL;
4253                 goto done;
4254         }
4255         aiov.iov_base = buf;
4256         aiov.iov_len = count;
4257         auio.uio_iov = &aiov;
4258         auio.uio_iovcnt = 1;
4259         auio.uio_rw = UIO_READ;
4260         auio.uio_segflg = direction;
4261         auio.uio_td = td;
4262         auio.uio_resid = count;
4263         loff = auio.uio_offset = fp->f_offset;
4264         error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
4265         fp->f_offset = auio.uio_offset;
4266         if (error)
4267                 goto done;
4268
4269         /*
4270          * WARNING!  *basep may not be wide enough to accomodate the
4271          * seek offset.   XXX should we hack this to return the upper 32 bits
4272          * for offsets greater then 4G?
4273          */
4274         if (basep) {
4275                 *basep = (long)loff;
4276         }
4277         *res = count - auio.uio_resid;
4278 done:
4279         fdrop(fp);
4280         return (error);
4281 }
4282
4283 /*
4284  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
4285  *
4286  * Read a block of directory entries in a file system independent format.
4287  */
4288 int
4289 sys_getdirentries(struct getdirentries_args *uap)
4290 {
4291         long base;
4292         int error;
4293
4294         error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
4295                                    &uap->sysmsg_result, UIO_USERSPACE);
4296
4297         if (error == 0 && uap->basep)
4298                 error = copyout(&base, uap->basep, sizeof(*uap->basep));
4299         return (error);
4300 }
4301
4302 /*
4303  * getdents_args(int fd, char *buf, size_t count)
4304  */
4305 int
4306 sys_getdents(struct getdents_args *uap)
4307 {
4308         int error;
4309
4310         error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4311                                    &uap->sysmsg_result, UIO_USERSPACE);
4312
4313         return (error);
4314 }
4315
4316 /*
4317  * Set the mode mask for creation of filesystem nodes.
4318  *
4319  * umask(int newmask)
4320  */
4321 int
4322 sys_umask(struct umask_args *uap)
4323 {
4324         struct thread *td = curthread;
4325         struct proc *p = td->td_proc;
4326         struct filedesc *fdp;
4327
4328         fdp = p->p_fd;
4329         uap->sysmsg_result = fdp->fd_cmask;
4330         fdp->fd_cmask = uap->newmask & ALLPERMS;
4331         return (0);
4332 }
4333
4334 /*
4335  * revoke(char *path)
4336  *
4337  * Void all references to file by ripping underlying filesystem
4338  * away from vnode.
4339  */
4340 int
4341 sys_revoke(struct revoke_args *uap)
4342 {
4343         struct nlookupdata nd;
4344         struct vattr vattr;
4345         struct vnode *vp;
4346         struct ucred *cred;
4347         int error;
4348
4349         vp = NULL;
4350         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4351         if (error == 0)
4352                 error = nlookup(&nd);
4353         if (error == 0)
4354                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4355         cred = crhold(nd.nl_cred);
4356         nlookup_done(&nd);
4357         if (error == 0) {
4358                 if (error == 0)
4359                         error = VOP_GETATTR(vp, &vattr);
4360                 if (error == 0 && cred->cr_uid != vattr.va_uid)
4361                         error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4362                 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4363                         if (vcount(vp) > 0)
4364                                 error = vrevoke(vp, cred);
4365                 } else if (error == 0) {
4366                         error = vrevoke(vp, cred);
4367                 }
4368                 vrele(vp);
4369         }
4370         if (cred)
4371                 crfree(cred);
4372         return (error);
4373 }
4374
4375 /*
4376  * getfh_args(char *fname, fhandle_t *fhp)
4377  *
4378  * Get (NFS) file handle
4379  *
4380  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4381  * mount.  This allows nullfs mounts to be explicitly exported.
4382  *
4383  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4384  *
4385  *          nullfs mounts of subdirectories are not safe.  That is, it will
4386  *          work, but you do not really have protection against access to
4387  *          the related parent directories.
4388  */
4389 int
4390 sys_getfh(struct getfh_args *uap)
4391 {
4392         struct thread *td = curthread;
4393         struct nlookupdata nd;
4394         fhandle_t fh;
4395         struct vnode *vp;
4396         struct mount *mp;
4397         int error;
4398
4399         /*
4400          * Must be super user
4401          */
4402         if ((error = priv_check(td, PRIV_ROOT)) != 0)
4403                 return (error);
4404
4405         vp = NULL;
4406         error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4407         if (error == 0)
4408                 error = nlookup(&nd);
4409         if (error == 0)
4410                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4411         mp = nd.nl_nch.mount;
4412         nlookup_done(&nd);
4413         if (error == 0) {
4414                 bzero(&fh, sizeof(fh));
4415                 fh.fh_fsid = mp->mnt_stat.f_fsid;
4416                 error = VFS_VPTOFH(vp, &fh.fh_fid);
4417                 vput(vp);
4418                 if (error == 0)
4419                         error = copyout(&fh, uap->fhp, sizeof(fh));
4420         }
4421         return (error);
4422 }
4423
4424 /*
4425  * fhopen_args(const struct fhandle *u_fhp, int flags)
4426  *
4427  * syscall for the rpc.lockd to use to translate a NFS file handle into
4428  * an open descriptor.
4429  *
4430  * warning: do not remove the priv_check() call or this becomes one giant
4431  * security hole.
4432  */
4433 int
4434 sys_fhopen(struct fhopen_args *uap)
4435 {
4436         struct thread *td = curthread;
4437         struct filedesc *fdp = td->td_proc->p_fd;
4438         struct mount *mp;
4439         struct vnode *vp;
4440         struct fhandle fhp;
4441         struct vattr vat;
4442         struct vattr *vap = &vat;
4443         struct flock lf;
4444         int fmode, mode, error = 0, type;
4445         struct file *nfp;
4446         struct file *fp;
4447         int indx;
4448
4449         /*
4450          * Must be super user
4451          */
4452         error = priv_check(td, PRIV_ROOT);
4453         if (error)
4454                 return (error);
4455
4456         fmode = FFLAGS(uap->flags);
4457
4458         /*
4459          * Why not allow a non-read/write open for our lockd?
4460          */
4461         if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4462                 return (EINVAL);
4463         error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4464         if (error)
4465                 return(error);
4466
4467         /*
4468          * Find the mount point
4469          */
4470         mp = vfs_getvfs(&fhp.fh_fsid);
4471         if (mp == NULL) {
4472                 error = ESTALE;
4473                 goto  done;
4474         }
4475         /* now give me my vnode, it gets returned to me locked */
4476         error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4477         if (error)
4478                 goto done;
4479         /*
4480          * from now on we have to make sure not
4481          * to forget about the vnode
4482          * any error that causes an abort must vput(vp)
4483          * just set error = err and 'goto bad;'.
4484          */
4485
4486         /*
4487          * from vn_open
4488          */
4489         if (vp->v_type == VLNK) {
4490                 error = EMLINK;
4491                 goto bad;
4492         }
4493         if (vp->v_type == VSOCK) {
4494                 error = EOPNOTSUPP;
4495                 goto bad;
4496         }
4497         mode = 0;
4498         if (fmode & (FWRITE | O_TRUNC)) {
4499                 if (vp->v_type == VDIR) {
4500                         error = EISDIR;
4501                         goto bad;
4502                 }
4503                 error = vn_writechk(vp, NULL);
4504                 if (error)
4505                         goto bad;
4506                 mode |= VWRITE;
4507         }
4508         if (fmode & FREAD)
4509                 mode |= VREAD;
4510         if (mode) {
4511                 error = VOP_ACCESS(vp, mode, td->td_ucred);
4512                 if (error)
4513                         goto bad;
4514         }
4515         if (fmode & O_TRUNC) {
4516                 vn_unlock(vp);                          /* XXX */
4517                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
4518                 VATTR_NULL(vap);
4519                 vap->va_size = 0;
4520                 error = VOP_SETATTR(vp, vap, td->td_ucred);
4521                 if (error)
4522                         goto bad;
4523         }
4524
4525         /*
4526          * VOP_OPEN needs the file pointer so it can potentially override
4527          * it.
4528          *
4529          * WARNING! no f_nchandle will be associated when fhopen()ing a
4530          * directory.  XXX
4531          */
4532         if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4533                 goto bad;
4534         fp = nfp;
4535
4536         error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4537         if (error) {
4538                 /*
4539                  * setting f_ops this way prevents VOP_CLOSE from being
4540                  * called or fdrop() releasing the vp from v_data.   Since
4541                  * the VOP_OPEN failed we don't want to VOP_CLOSE.
4542                  */
4543                 fp->f_ops = &badfileops;
4544                 fp->f_data = NULL;
4545                 goto bad_drop;
4546         }
4547
4548         /*
4549          * The fp is given its own reference, we still have our ref and lock.
4550          *
4551          * Assert that all regular files must be created with a VM object.
4552          */
4553         if (vp->v_type == VREG && vp->v_object == NULL) {
4554                 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
4555                 goto bad_drop;
4556         }
4557
4558         /*
4559          * The open was successful.  Handle any locking requirements.
4560          */
4561         if (fmode & (O_EXLOCK | O_SHLOCK)) {
4562                 lf.l_whence = SEEK_SET;
4563                 lf.l_start = 0;
4564                 lf.l_len = 0;
4565                 if (fmode & O_EXLOCK)
4566                         lf.l_type = F_WRLCK;
4567                 else
4568                         lf.l_type = F_RDLCK;
4569                 if (fmode & FNONBLOCK)
4570                         type = 0;
4571                 else
4572                         type = F_WAIT;
4573                 vn_unlock(vp);
4574                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
4575                         /*
4576                          * release our private reference.
4577                          */
4578                         fsetfd(fdp, NULL, indx);
4579                         fdrop(fp);
4580                         vrele(vp);
4581                         goto done;
4582                 }
4583                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4584                 atomic_set_int(&fp->f_flag, FHASLOCK);  /* race ok */
4585         }
4586
4587         /*
4588          * Clean up.  Associate the file pointer with the previously
4589          * reserved descriptor and return it.
4590          */
4591         vput(vp);
4592         if (uap->flags & O_CLOEXEC)
4593                 fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
4594         fsetfd(fdp, fp, indx);
4595         fdrop(fp);
4596         uap->sysmsg_result = indx;
4597         return (error);
4598
4599 bad_drop:
4600         fsetfd(fdp, NULL, indx);
4601         fdrop(fp);
4602 bad:
4603         vput(vp);
4604 done:
4605         return (error);
4606 }
4607
4608 /*
4609  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4610  */
4611 int
4612 sys_fhstat(struct fhstat_args *uap)
4613 {
4614         struct thread *td = curthread;
4615         struct stat sb;
4616         fhandle_t fh;
4617         struct mount *mp;
4618         struct vnode *vp;
4619         int error;
4620
4621         /*
4622          * Must be super user
4623          */
4624         error = priv_check(td, PRIV_ROOT);
4625         if (error)
4626                 return (error);
4627
4628         error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4629         if (error)
4630                 return (error);
4631
4632         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4633                 error = ESTALE;
4634         if (error == 0) {
4635                 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4636                         error = vn_stat(vp, &sb, td->td_ucred);
4637                         vput(vp);
4638                 }
4639         }
4640         if (error == 0)
4641                 error = copyout(&sb, uap->sb, sizeof(sb));
4642         return (error);
4643 }
4644
4645 /*
4646  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4647  */
4648 int
4649 sys_fhstatfs(struct fhstatfs_args *uap)
4650 {
4651         struct thread *td = curthread;
4652         struct proc *p = td->td_proc;
4653         struct statfs *sp;
4654         struct mount *mp;
4655         struct vnode *vp;
4656         struct statfs sb;
4657         char *fullpath, *freepath;
4658         fhandle_t fh;
4659         int error;
4660
4661         /*
4662          * Must be super user
4663          */
4664         if ((error = priv_check(td, PRIV_ROOT)))
4665                 return (error);
4666
4667         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4668                 return (error);
4669
4670         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4671                 error = ESTALE;
4672                 goto done;
4673         }
4674         if (p != NULL && !chroot_visible_mnt(mp, p)) {
4675                 error = ESTALE;
4676                 goto done;
4677         }
4678
4679         if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4680                 goto done;
4681         mp = vp->v_mount;
4682         sp = &mp->mnt_stat;
4683         vput(vp);
4684         if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4685                 goto done;
4686
4687         error = mount_path(p, mp, &fullpath, &freepath);
4688         if (error)
4689                 goto done;
4690         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4691         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4692         kfree(freepath, M_TEMP);
4693
4694         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4695         if (priv_check(td, PRIV_ROOT)) {
4696                 bcopy(sp, &sb, sizeof(sb));
4697                 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4698                 sp = &sb;
4699         }
4700         error = copyout(sp, uap->buf, sizeof(*sp));
4701 done:
4702         return (error);
4703 }
4704
4705 /*
4706  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4707  */
4708 int
4709 sys_fhstatvfs(struct fhstatvfs_args *uap)
4710 {
4711         struct thread *td = curthread;
4712         struct proc *p = td->td_proc;
4713         struct statvfs *sp;
4714         struct mount *mp;
4715         struct vnode *vp;
4716         fhandle_t fh;
4717         int error;
4718
4719         /*
4720          * Must be super user
4721          */
4722         if ((error = priv_check(td, PRIV_ROOT)))
4723                 return (error);
4724
4725         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4726                 return (error);
4727
4728         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4729                 error = ESTALE;
4730                 goto done;
4731         }
4732         if (p != NULL && !chroot_visible_mnt(mp, p)) {
4733                 error = ESTALE;
4734                 goto done;
4735         }
4736
4737         if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4738                 goto done;
4739         mp = vp->v_mount;
4740         sp = &mp->mnt_vstat;
4741         vput(vp);
4742         if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4743                 goto done;
4744
4745         sp->f_flag = 0;
4746         if (mp->mnt_flag & MNT_RDONLY)
4747                 sp->f_flag |= ST_RDONLY;
4748         if (mp->mnt_flag & MNT_NOSUID)
4749                 sp->f_flag |= ST_NOSUID;
4750         error = copyout(sp, uap->buf, sizeof(*sp));
4751 done:
4752         return (error);
4753 }
4754
4755
4756 /*
4757  * Syscall to push extended attribute configuration information into the
4758  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4759  * a command (int cmd), and attribute name and misc data.  For now, the
4760  * attribute name is left in userspace for consumption by the VFS_op.
4761  * It will probably be changed to be copied into sysspace by the
4762  * syscall in the future, once issues with various consumers of the
4763  * attribute code have raised their hands.
4764  *
4765  * Currently this is used only by UFS Extended Attributes.
4766  */
4767 int
4768 sys_extattrctl(struct extattrctl_args *uap)
4769 {
4770         struct nlookupdata nd;
4771         struct vnode *vp;
4772         char attrname[EXTATTR_MAXNAMELEN];
4773         int error;
4774         size_t size;
4775
4776         attrname[0] = 0;
4777         vp = NULL;
4778         error = 0;
4779
4780         if (error == 0 && uap->filename) {
4781                 error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4782                                      NLC_FOLLOW);
4783                 if (error == 0)
4784                         error = nlookup(&nd);
4785                 if (error == 0)
4786                         error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4787                 nlookup_done(&nd);
4788         }
4789
4790         if (error == 0 && uap->attrname) {
4791                 error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4792                                   &size);
4793         }
4794
4795         if (error == 0) {
4796                 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4797                 if (error == 0)
4798                         error = nlookup(&nd);
4799                 if (error == 0)
4800                         error = ncp_writechk(&nd.nl_nch);
4801                 if (error == 0) {
4802                         error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4803                                                uap->attrnamespace,
4804                                                uap->attrname, nd.nl_cred);
4805                 }
4806                 nlookup_done(&nd);
4807         }
4808
4809         return (error);
4810 }
4811
4812 /*
4813  * Syscall to get a named extended attribute on a file or directory.
4814  */
4815 int
4816 sys_extattr_set_file(struct extattr_set_file_args *uap)
4817 {
4818         char attrname[EXTATTR_MAXNAMELEN];
4819         struct nlookupdata nd;
4820         struct vnode *vp;
4821         struct uio auio;
4822         struct iovec aiov;
4823         int error;
4824
4825         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4826         if (error)
4827                 return (error);
4828
4829         vp = NULL;
4830
4831         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4832         if (error == 0)
4833                 error = nlookup(&nd);
4834         if (error == 0)
4835                 error = ncp_writechk(&nd.nl_nch);
4836         if (error == 0)
4837                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4838         if (error) {
4839                 nlookup_done(&nd);
4840                 return (error);
4841         }
4842
4843         bzero(&auio, sizeof(auio));
4844         aiov.iov_base = uap->data;
4845         aiov.iov_len = uap->nbytes;
4846         auio.uio_iov = &aiov;
4847         auio.uio_iovcnt = 1;
4848         auio.uio_offset = 0;
4849         auio.uio_resid = uap->nbytes;
4850         auio.uio_rw = UIO_WRITE;
4851         auio.uio_td = curthread;
4852
4853         error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
4854                                &auio, nd.nl_cred);
4855
4856         vput(vp);
4857         nlookup_done(&nd);
4858         return (error);
4859 }
4860
4861 /*
4862  * Syscall to get a named extended attribute on a file or directory.
4863  */
4864 int
4865 sys_extattr_get_file(struct extattr_get_file_args *uap)
4866 {
4867         char attrname[EXTATTR_MAXNAMELEN];
4868         struct nlookupdata nd;
4869         struct uio auio;
4870         struct iovec aiov;
4871         struct vnode *vp;
4872         int error;
4873
4874         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4875         if (error)
4876                 return (error);
4877
4878         vp = NULL;
4879
4880         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4881         if (error == 0)
4882                 error = nlookup(&nd);
4883         if (error == 0)
4884                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_SHARED, &vp);
4885         if (error) {
4886                 nlookup_done(&nd);
4887                 return (error);
4888         }
4889
4890         bzero(&auio, sizeof(auio));
4891         aiov.iov_base = uap->data;
4892         aiov.iov_len = uap->nbytes;
4893         auio.uio_iov = &aiov;
4894         auio.uio_iovcnt = 1;
4895         auio.uio_offset = 0;
4896         auio.uio_resid = uap->nbytes;
4897         auio.uio_rw = UIO_READ;
4898         auio.uio_td = curthread;
4899
4900         error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
4901                                 &auio, nd.nl_cred);
4902         uap->sysmsg_result = uap->nbytes - auio.uio_resid;
4903
4904         vput(vp);
4905         nlookup_done(&nd);
4906         return(error);
4907 }
4908
4909 /*
4910  * Syscall to delete a named extended attribute from a file or directory.
4911  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4912  */
4913 int
4914 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4915 {
4916         char attrname[EXTATTR_MAXNAMELEN];
4917         struct nlookupdata nd;
4918         struct vnode *vp;
4919         int error;
4920
4921         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4922         if (error)
4923                 return(error);
4924
4925         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4926         if (error == 0)
4927                 error = nlookup(&nd);
4928         if (error == 0)
4929                 error = ncp_writechk(&nd.nl_nch);
4930         if (error == 0) {
4931                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4932                 if (error == 0) {
4933                         error = VOP_SETEXTATTR(vp, uap->attrnamespace,
4934                                                attrname, NULL, nd.nl_cred);
4935                         vput(vp);
4936                 }
4937         }
4938         nlookup_done(&nd);
4939         return(error);
4940 }
4941
4942 /*
4943  * Determine if the mount is visible to the process.
4944  */
4945 static int
4946 chroot_visible_mnt(struct mount *mp, struct proc *p)
4947 {
4948         struct nchandle nch;
4949
4950         /*
4951          * Traverse from the mount point upwards.  If we hit the process
4952          * root then the mount point is visible to the process.
4953          */
4954         nch = mp->mnt_ncmountpt;
4955         while (nch.ncp) {
4956                 if (nch.mount == p->p_fd->fd_nrdir.mount &&
4957                     nch.ncp == p->p_fd->fd_nrdir.ncp) {
4958                         return(1);
4959                 }
4960                 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4961                         nch = nch.mount->mnt_ncmounton;
4962                 } else {
4963                         nch.ncp = nch.ncp->nc_parent;
4964                 }
4965         }
4966
4967         /*
4968          * If the mount point is not visible to the process, but the
4969          * process root is in a subdirectory of the mount, return
4970          * TRUE anyway.
4971          */
4972         if (p->p_fd->fd_nrdir.mount == mp)
4973                 return(1);
4974
4975         return(0);
4976 }
4977