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. All advertising materials mentioning features or use of this software
  19  *    must display the following acknowledgement:
  20  *      This product includes software developed by the University of
  21  *      California, Berkeley and its contributors.
  22  * 4. Neither the name of the University nor the names of its contributors
  23  *    may be used to endorse or promote products derived from this software
  24  *    without specific prior written permission.
  25  *
  26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  36  * SUCH DAMAGE.
  37  *
  38  *      @(#)vfs_syscalls.c      8.13 (Berkeley) 4/15/94
  39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
  40  * $DragonFly: src/sys/kern/vfs_syscalls.c,v 1.120 2007/09/03 17:06:21 dillon Exp $
  41  */
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/buf.h>
  46 #include <sys/conf.h>
  47 #include <sys/sysent.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mount.h>
  50 #include <sys/mountctl.h>
  51 #include <sys/sysproto.h>
  52 #include <sys/filedesc.h>
  53 #include <sys/kernel.h>
  54 #include <sys/fcntl.h>
  55 #include <sys/file.h>
  56 #include <sys/linker.h>
  57 #include <sys/stat.h>
  58 #include <sys/unistd.h>
  59 #include <sys/vnode.h>
  60 #include <sys/proc.h>
  61 #include <sys/namei.h>
  62 #include <sys/nlookup.h>
  63 #include <sys/dirent.h>
  64 #include <sys/extattr.h>
  65 #include <sys/spinlock.h>
  66 #include <sys/kern_syscall.h>
  67 #include <sys/objcache.h>
  68 #include <sys/sysctl.h>
  69 #include <sys/file2.h>
  70 #include <sys/spinlock2.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 #include <vfs/union/union.h>
  80
  81 static void mount_warning(struct mount *mp, const char *ctl, ...);
  82 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
  83 static int checkvp_chdir (struct vnode *vn, struct thread *td);
  84 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
  85 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
  86 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
  87 static int getutimes (const struct timeval *, struct timespec *);
  88 static int setfown (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 *, const struct timespec *, int);
  92 static int      usermount = 0;  /* if 1, non-root can mount fs. */
  93
  94 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
  95
  96 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
  97
  98 /*
  99  * Virtual File System System Calls
 100  */
 101
 102 /*
 103  * Mount a file system.
 104  */
 105 /*
 106  * mount_args(char *type, char *path, int flags, caddr_t data)
 107  */
 108 /* ARGSUSED */
 109 int
 110 sys_mount(struct mount_args *uap)
 111 {
 112         struct thread *td = curthread;
 113         struct proc *p = td->td_proc;
 114         struct vnode *vp;
 115         struct nchandle nch;
 116         struct mount *mp;
 117         struct vfsconf *vfsp;
 118         int error, flag = 0, flag2 = 0;
 119         int hasmount;
 120         struct vattr va;
 121         struct nlookupdata nd;
 122         char fstypename[MFSNAMELEN];
 123         struct ucred *cred = p->p_ucred;
 124
 125         KKASSERT(p);
 126         if (cred->cr_prison != NULL)
 127                 return (EPERM);
 128         if (usermount == 0 && (error = suser(td)))
 129                 return (error);
 130         /*
 131          * Do not allow NFS export by non-root users.
 132          */
 133         if (uap->flags & MNT_EXPORTED) {
 134                 error = suser(td);
 135                 if (error)
 136                         return (error);
 137         }
 138         /*
 139          * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
 140          */
 141         if (suser(td))
 142                 uap->flags |= MNT_NOSUID | MNT_NODEV;
 143
 144         /*
 145          * Lookup the requested path and extract the nch and vnode.
 146          */
 147         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 148         if (error == 0) {
 149                 if ((error = nlookup(&nd)) == 0) {
 150                         if (nd.nl_nch.ncp->nc_vp == NULL)
 151                                 error = ENOENT;
 152                 }
 153         }
 154         if (error) {
 155                 nlookup_done(&nd);
 156                 return (error);
 157         }
 158
 159         /*
 160          * Extract the locked+refd ncp and cleanup the nd structure
 161          */
 162         nch = nd.nl_nch;
 163         cache_zero(&nd.nl_nch);
 164         nlookup_done(&nd);
 165
 166         if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
 167                 hasmount = 1;
 168         else
 169                 hasmount = 0;
 170
 171
 172         /*
 173          * now we have the locked ref'd nch and unreferenced vnode.
 174          */
 175         vp = nch.ncp->nc_vp;
 176         if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
 177                 cache_put(&nch);
 178                 return (error);
 179         }
 180         cache_unlock(&nch);
 181
 182         /*
 183          * Now we have an unlocked ref'd nch and a locked ref'd vp
 184          */
 185         if (uap->flags & MNT_UPDATE) {
 186                 if ((vp->v_flag & VROOT) == 0) {
 187                         cache_drop(&nch);
 188                         vput(vp);
 189                         return (EINVAL);
 190                 }
 191                 mp = vp->v_mount;
 192                 flag = mp->mnt_flag;
 193                 flag2 = mp->mnt_kern_flag;
 194                 /*
 195                  * We only allow the filesystem to be reloaded if it
 196                  * is currently mounted read-only.
 197                  */
 198                 if ((uap->flags & MNT_RELOAD) &&
 199                     ((mp->mnt_flag & MNT_RDONLY) == 0)) {
 200                         cache_drop(&nch);
 201                         vput(vp);
 202                         return (EOPNOTSUPP);    /* Needs translation */
 203                 }
 204                 /*
 205                  * Only root, or the user that did the original mount is
 206                  * permitted to update it.
 207                  */
 208                 if (mp->mnt_stat.f_owner != cred->cr_uid &&
 209                     (error = suser(td))) {
 210                         cache_drop(&nch);
 211                         vput(vp);
 212                         return (error);
 213                 }
 214                 if (vfs_busy(mp, LK_NOWAIT)) {
 215                         cache_drop(&nch);
 216                         vput(vp);
 217                         return (EBUSY);
 218                 }
 219                 if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
 220                         cache_drop(&nch);
 221                         vfs_unbusy(mp);
 222                         vput(vp);
 223                         return (EBUSY);
 224                 }
 225                 vp->v_flag |= VMOUNT;
 226                 mp->mnt_flag |=
 227                     uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
 228                 vn_unlock(vp);
 229                 goto update;
 230         }
 231         /*
 232          * If the user is not root, ensure that they own the directory
 233          * onto which we are attempting to mount.
 234          */
 235         if ((error = VOP_GETATTR(vp, &va)) ||
 236             (va.va_uid != cred->cr_uid && (error = suser(td)))) {
 237                 cache_drop(&nch);
 238                 vput(vp);
 239                 return (error);
 240         }
 241         if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
 242                 cache_drop(&nch);
 243                 vput(vp);
 244                 return (error);
 245         }
 246         if (vp->v_type != VDIR) {
 247                 cache_drop(&nch);
 248                 vput(vp);
 249                 return (ENOTDIR);
 250         }
 251         if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
 252                 cache_drop(&nch);
 253                 vput(vp);
 254                 return (EPERM);
 255         }
 256         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
 257                 cache_drop(&nch);
 258                 vput(vp);
 259                 return (error);
 260         }
 261         for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
 262                 if (!strcmp(vfsp->vfc_name, fstypename))
 263                         break;
 264         }
 265         if (vfsp == NULL) {
 266                 linker_file_t lf;
 267
 268                 /* Only load modules for root (very important!) */
 269                 if ((error = suser(td)) != 0) {
 270                         cache_drop(&nch);
 271                         vput(vp);
 272                         return error;
 273                 }
 274                 error = linker_load_file(fstypename, &lf);
 275                 if (error || lf == NULL) {
 276                         cache_drop(&nch);
 277                         vput(vp);
 278                         if (lf == NULL)
 279                                 error = ENODEV;
 280                         return error;
 281                 }
 282                 lf->userrefs++;
 283                 /* lookup again, see if the VFS was loaded */
 284                 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
 285                         if (!strcmp(vfsp->vfc_name, fstypename))
 286                                 break;
 287                 }
 288                 if (vfsp == NULL) {
 289                         lf->userrefs--;
 290                         linker_file_unload(lf);
 291                         cache_drop(&nch);
 292                         vput(vp);
 293                         return (ENODEV);
 294                 }
 295         }
 296         if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
 297                 cache_drop(&nch);
 298                 vput(vp);
 299                 return (EBUSY);
 300         }
 301         vp->v_flag |= VMOUNT;
 302
 303         /*
 304          * Allocate and initialize the filesystem.
 305          */
 306         mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
 307         TAILQ_INIT(&mp->mnt_nvnodelist);
 308         TAILQ_INIT(&mp->mnt_reservedvnlist);
 309         TAILQ_INIT(&mp->mnt_jlist);
 310         mp->mnt_nvnodelistsize = 0;
 311         lockinit(&mp->mnt_lock, "vfslock", 0, 0);
 312         vfs_busy(mp, LK_NOWAIT);
 313         mp->mnt_op = vfsp->vfc_vfsops;
 314         mp->mnt_vfc = vfsp;
 315         vfsp->vfc_refcount++;
 316         mp->mnt_stat.f_type = vfsp->vfc_typenum;
 317         mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
 318         strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
 319         mp->mnt_stat.f_owner = cred->cr_uid;
 320         mp->mnt_iosize_max = DFLTPHYS;
 321         vn_unlock(vp);
 322 update:
 323         /*
 324          * Set the mount level flags.
 325          */
 326         if (uap->flags & MNT_RDONLY)
 327                 mp->mnt_flag |= MNT_RDONLY;
 328         else if (mp->mnt_flag & MNT_RDONLY)
 329                 mp->mnt_kern_flag |= MNTK_WANTRDWR;
 330         mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
 331             MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
 332             MNT_NOSYMFOLLOW | MNT_IGNORE |
 333             MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
 334         mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
 335             MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
 336             MNT_NOSYMFOLLOW | MNT_IGNORE |
 337             MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
 338         /*
 339          * Mount the filesystem.
 340          * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
 341          * get.
 342          */
 343         error = VFS_MOUNT(mp, uap->path, uap->data, cred);
 344         if (mp->mnt_flag & MNT_UPDATE) {
 345                 if (mp->mnt_kern_flag & MNTK_WANTRDWR)
 346                         mp->mnt_flag &= ~MNT_RDONLY;
 347                 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
 348                 mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
 349                 if (error) {
 350                         mp->mnt_flag = flag;
 351                         mp->mnt_kern_flag = flag2;
 352                 }
 353                 vfs_unbusy(mp);
 354                 vp->v_flag &= ~VMOUNT;
 355                 vrele(vp);
 356                 cache_drop(&nch);
 357                 return (error);
 358         }
 359         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 360         /*
 361          * Put the new filesystem on the mount list after root.  The mount
 362          * point gets its own mnt_ncmountpt (unless the VFS already set one
 363          * up) which represents the root of the mount.  The lookup code
 364          * detects the mount point going forward and checks the root of
 365          * the mount going backwards.
 366          *
 367          * It is not necessary to invalidate or purge the vnode underneath
 368          * because elements under the mount will be given their own glue
 369          * namecache record.
 370          */
 371         if (!error) {
 372                 if (mp->mnt_ncmountpt.ncp == NULL) {
 373                         /*
 374                          * allocate, then unlock, but leave the ref intact
 375                          */
 376                         cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
 377                         cache_unlock(&mp->mnt_ncmountpt);
 378                 }
 379                 mp->mnt_ncmounton = nch;                /* inherits ref */
 380                 nch.ncp->nc_flag |= NCF_ISMOUNTPT;
 381
 382                 /* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
 383                 vp->v_flag &= ~VMOUNT;
 384                 mountlist_insert(mp, MNTINS_LAST);
 385                 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
 386                 vn_unlock(vp);
 387                 error = vfs_allocate_syncvnode(mp);
 388                 vfs_unbusy(mp);
 389                 error = VFS_START(mp, 0);
 390                 vrele(vp);
 391         } else {
 392                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 393                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 394                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 395                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 396                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 397                 vp->v_flag &= ~VMOUNT;
 398                 mp->mnt_vfc->vfc_refcount--;
 399                 vfs_unbusy(mp);
 400                 kfree(mp, M_MOUNT);
 401                 cache_drop(&nch);
 402                 vput(vp);
 403         }
 404         return (error);
 405 }
 406
 407 /*
 408  * Scan all active processes to see if any of them have a current
 409  * or root directory onto which the new filesystem has just been
 410  * mounted. If so, replace them with the new mount point.
 411  *
 412  * The passed ncp is ref'd and locked (from the mount code) and
 413  * must be associated with the vnode representing the root of the
 414  * mount point.
 415  */
 416 struct checkdirs_info {
 417         struct nchandle old_nch;
 418         struct nchandle new_nch;
 419         struct vnode *old_vp;
 420         struct vnode *new_vp;
 421 };
 422
 423 static int checkdirs_callback(struct proc *p, void *data);
 424
 425 static void
 426 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
 427 {
 428         struct checkdirs_info info;
 429         struct vnode *olddp;
 430         struct vnode *newdp;
 431         struct mount *mp;
 432
 433         /*
 434          * If the old mount point's vnode has a usecount of 1, it is not
 435          * being held as a descriptor anywhere.
 436          */
 437         olddp = old_nch->ncp->nc_vp;
 438         if (olddp == NULL || olddp->v_sysref.refcnt == 1)
 439                 return;
 440
 441         /*
 442          * Force the root vnode of the new mount point to be resolved
 443          * so we can update any matching processes.
 444          */
 445         mp = new_nch->mount;
 446         if (VFS_ROOT(mp, &newdp))
 447                 panic("mount: lost mount");
 448         cache_setunresolved(new_nch);
 449         cache_setvp(new_nch, newdp);
 450
 451         /*
 452          * Special handling of the root node
 453          */
 454         if (rootvnode == olddp) {
 455                 vref(newdp);
 456                 vfs_cache_setroot(newdp, cache_hold(new_nch));
 457         }
 458
 459         /*
 460          * Pass newdp separately so the callback does not have to access
 461          * it via new_nch->ncp->nc_vp.
 462          */
 463         info.old_nch = *old_nch;
 464         info.new_nch = *new_nch;
 465         info.new_vp = newdp;
 466         allproc_scan(checkdirs_callback, &info);
 467         vput(newdp);
 468 }
 469
 470 /*
 471  * NOTE: callback is not MP safe because the scanned process's filedesc
 472  * structure can be ripped out from under us, amoung other things.
 473  */
 474 static int
 475 checkdirs_callback(struct proc *p, void *data)
 476 {
 477         struct checkdirs_info *info = data;
 478         struct filedesc *fdp;
 479         struct nchandle ncdrop1;
 480         struct nchandle ncdrop2;
 481         struct vnode *vprele1;
 482         struct vnode *vprele2;
 483
 484         if ((fdp = p->p_fd) != NULL) {
 485                 cache_zero(&ncdrop1);
 486                 cache_zero(&ncdrop2);
 487                 vprele1 = NULL;
 488                 vprele2 = NULL;
 489
 490                 /*
 491                  * MPUNSAFE - XXX fdp can be pulled out from under a
 492                  * foreign process.
 493                  *
 494                  * A shared filedesc is ok, we don't have to copy it
 495                  * because we are making this change globally.
 496                  */
 497                 spin_lock_wr(&fdp->fd_spin);
 498                 if (fdp->fd_ncdir.mount == info->old_nch.mount &&
 499                     fdp->fd_ncdir.ncp == info->old_nch.ncp) {
 500                         vprele1 = fdp->fd_cdir;
 501                         vref(info->new_vp);
 502                         fdp->fd_cdir = info->new_vp;
 503                         ncdrop1 = fdp->fd_ncdir;
 504                         cache_copy(&info->new_nch, &fdp->fd_ncdir);
 505                 }
 506                 if (fdp->fd_nrdir.mount == info->old_nch.mount &&
 507                     fdp->fd_nrdir.ncp == info->old_nch.ncp) {
 508                         vprele2 = fdp->fd_rdir;
 509                         vref(info->new_vp);
 510                         fdp->fd_rdir = info->new_vp;
 511                         ncdrop2 = fdp->fd_nrdir;
 512                         cache_copy(&info->new_nch, &fdp->fd_nrdir);
 513                 }
 514                 spin_unlock_wr(&fdp->fd_spin);
 515                 if (ncdrop1.ncp)
 516                         cache_drop(&ncdrop1);
 517                 if (ncdrop2.ncp)
 518                         cache_drop(&ncdrop2);
 519                 if (vprele1)
 520                         vrele(vprele1);
 521                 if (vprele2)
 522                         vrele(vprele2);
 523         }
 524         return(0);
 525 }
 526
 527 /*
 528  * Unmount a file system.
 529  *
 530  * Note: unmount takes a path to the vnode mounted on as argument,
 531  * not special file (as before).
 532  */
 533 /*
 534  * umount_args(char *path, int flags)
 535  */
 536 /* ARGSUSED */
 537 int
 538 sys_unmount(struct unmount_args *uap)
 539 {
 540         struct thread *td = curthread;
 541         struct proc *p = td->td_proc;
 542         struct mount *mp = NULL;
 543         int error;
 544         struct nlookupdata nd;
 545
 546         KKASSERT(p);
 547         if (p->p_ucred->cr_prison != NULL)
 548                 return (EPERM);
 549         if (usermount == 0 && (error = suser(td)))
 550                 return (error);
 551
 552         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 553         if (error == 0)
 554                 error = nlookup(&nd);
 555         if (error)
 556                 goto out;
 557
 558         mp = nd.nl_nch.mount;
 559
 560         /*
 561          * Only root, or the user that did the original mount is
 562          * permitted to unmount this filesystem.
 563          */
 564         if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
 565             (error = suser(td)))
 566                 goto out;
 567
 568         /*
 569          * Don't allow unmounting the root file system.
 570          */
 571         if (mp->mnt_flag & MNT_ROOTFS) {
 572                 error = EINVAL;
 573                 goto out;
 574         }
 575
 576         /*
 577          * Must be the root of the filesystem
 578          */
 579         if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
 580                 error = EINVAL;
 581                 goto out;
 582         }
 583
 584 out:
 585         nlookup_done(&nd);
 586         if (error)
 587                 return (error);
 588         return (dounmount(mp, uap->flags));
 589 }
 590
 591 /*
 592  * Do the actual file system unmount.
 593  */
 594 static int
 595 dounmount_interlock(struct mount *mp)
 596 {
 597         if (mp->mnt_kern_flag & MNTK_UNMOUNT)
 598                 return (EBUSY);
 599         mp->mnt_kern_flag |= MNTK_UNMOUNT;
 600         return(0);
 601 }
 602
 603 int
 604 dounmount(struct mount *mp, int flags)
 605 {
 606         struct namecache *ncp;
 607         struct nchandle nch;
 608         int error;
 609         int async_flag;
 610         int lflags;
 611         int freeok = 1;
 612
 613         /*
 614          * Exclusive access for unmounting purposes
 615          */
 616         if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
 617                 return (error);
 618
 619         /*
 620          * Allow filesystems to detect that a forced unmount is in progress.
 621          */
 622         if (flags & MNT_FORCE)
 623                 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
 624         lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
 625         error = lockmgr(&mp->mnt_lock, lflags);
 626         if (error) {
 627                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 628                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 629                         wakeup(mp);
 630                 return (error);
 631         }
 632
 633         if (mp->mnt_flag & MNT_EXPUBLIC)
 634                 vfs_setpublicfs(NULL, NULL, NULL);
 635
 636         vfs_msync(mp, MNT_WAIT);
 637         async_flag = mp->mnt_flag & MNT_ASYNC;
 638         mp->mnt_flag &=~ MNT_ASYNC;
 639
 640         /*
 641          * If this filesystem isn't aliasing other filesystems,
 642          * try to invalidate any remaining namecache entries and
 643          * check the count afterwords.
 644          */
 645         if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
 646                 cache_lock(&mp->mnt_ncmountpt);
 647                 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
 648                 cache_unlock(&mp->mnt_ncmountpt);
 649
 650                 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
 651                     (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
 652
 653                         if ((flags & MNT_FORCE) == 0) {
 654                                 error = EBUSY;
 655                                 mount_warning(mp, "Cannot unmount: "
 656                                                   "%d namecache "
 657                                                   "references still "
 658                                                   "present",
 659                                                   ncp->nc_refs - 1);
 660                         } else {
 661                                 mount_warning(mp, "Forced unmount: "
 662                                                   "%d namecache "
 663                                                   "references still "
 664                                                   "present",
 665                                                   ncp->nc_refs - 1);
 666                                 freeok = 0;
 667                         }
 668                 }
 669         }
 670
 671         /*
 672          * nchandle records ref the mount structure.  Expect a count of 1
 673          * (our mount->mnt_ncmountpt).
 674          */
 675         if (mp->mnt_refs != 1) {
 676                 if ((flags & MNT_FORCE) == 0) {
 677                         mount_warning(mp, "Cannot unmount: "
 678                                           "%d process references still "
 679                                           "present", mp->mnt_refs);
 680                         error = EBUSY;
 681                 } else {
 682                         mount_warning(mp, "Forced unmount: "
 683                                           "%d process references still "
 684                                           "present", mp->mnt_refs);
 685                         freeok = 0;
 686                 }
 687         }
 688
 689         if (error == 0) {
 690                 if (mp->mnt_syncer != NULL)
 691                         vrele(mp->mnt_syncer);
 692                 if (((mp->mnt_flag & MNT_RDONLY) ||
 693                      (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
 694                     (flags & MNT_FORCE)) {
 695                         error = VFS_UNMOUNT(mp, flags);
 696                 }
 697         }
 698         if (error) {
 699                 if (mp->mnt_syncer == NULL)
 700                         vfs_allocate_syncvnode(mp);
 701                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 702                 mp->mnt_flag |= async_flag;
 703                 lockmgr(&mp->mnt_lock, LK_RELEASE);
 704                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 705                         wakeup(mp);
 706                 return (error);
 707         }
 708         /*
 709          * Clean up any journals still associated with the mount after
 710          * filesystem activity has ceased.
 711          */
 712         journal_remove_all_journals(mp,
 713             ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
 714
 715         mountlist_remove(mp);
 716
 717         /*
 718          * Remove any installed vnode ops here so the individual VFSs don't
 719          * have to.
 720          */
 721         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 722         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 723         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 724         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 725         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 726
 727         if (mp->mnt_ncmountpt.ncp != NULL) {
 728                 nch = mp->mnt_ncmountpt;
 729                 cache_zero(&mp->mnt_ncmountpt);
 730                 cache_clrmountpt(&nch);
 731                 cache_drop(&nch);
 732         }
 733         if (mp->mnt_ncmounton.ncp != NULL) {
 734                 nch = mp->mnt_ncmounton;
 735                 cache_zero(&mp->mnt_ncmounton);
 736                 cache_clrmountpt(&nch);
 737                 cache_drop(&nch);
 738         }
 739
 740         mp->mnt_vfc->vfc_refcount--;
 741         if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
 742                 panic("unmount: dangling vnode");
 743         lockmgr(&mp->mnt_lock, LK_RELEASE);
 744         if (mp->mnt_kern_flag & MNTK_MWAIT)
 745                 wakeup(mp);
 746         if (freeok)
 747                 kfree(mp, M_MOUNT);
 748         return (0);
 749 }
 750
 751 static
 752 void
 753 mount_warning(struct mount *mp, const char *ctl, ...)
 754 {
 755         char *ptr;
 756         char *buf;
 757         __va_list va;
 758
 759         __va_start(va, ctl);
 760         if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
 761                 kprintf("unmount(%s): ", ptr);
 762                 kvprintf(ctl, va);
 763                 kprintf("\n");
 764                 kfree(buf, M_TEMP);
 765         } else {
 766                 kprintf("unmount(%p", mp);
 767                 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
 768                         kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
 769                 kprintf("): ");
 770                 kvprintf(ctl, va);
 771                 kprintf("\n");
 772         }
 773         __va_end(va);
 774 }
 775
 776 /*
 777  * Shim cache_fullpath() to handle the case where a process is chrooted into
 778  * a subdirectory of a mount.  In this case if the root mount matches the
 779  * process root directory's mount we have to specify the process's root
 780  * directory instead of the mount point, because the mount point might
 781  * be above the root directory.
 782  */
 783 static
 784 int
 785 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
 786 {
 787         struct nchandle *nch;
 788
 789         if (p && p->p_fd->fd_nrdir.mount == mp)
 790                 nch = &p->p_fd->fd_nrdir;
 791         else
 792                 nch = &mp->mnt_ncmountpt;
 793         return(cache_fullpath(p, nch, rb, fb));
 794 }
 795
 796 /*
 797  * Sync each mounted filesystem.
 798  */
 799
 800 #ifdef DEBUG
 801 static int syncprt = 0;
 802 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
 803 #endif /* DEBUG */
 804
 805 static int sync_callback(struct mount *mp, void *data);
 806
 807 /* ARGSUSED */
 808 int
 809 sys_sync(struct sync_args *uap)
 810 {
 811         mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
 812 #ifdef DEBUG
 813         /*
 814          * print out buffer pool stat information on each sync() call.
 815          */
 816         if (syncprt)
 817                 vfs_bufstats();
 818 #endif /* DEBUG */
 819         return (0);
 820 }
 821
 822 static
 823 int
 824 sync_callback(struct mount *mp, void *data __unused)
 825 {
 826         int asyncflag;
 827
 828         if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 829                 asyncflag = mp->mnt_flag & MNT_ASYNC;
 830                 mp->mnt_flag &= ~MNT_ASYNC;
 831                 vfs_msync(mp, MNT_NOWAIT);
 832                 VFS_SYNC(mp, MNT_NOWAIT);
 833                 mp->mnt_flag |= asyncflag;
 834         }
 835         return(0);
 836 }
 837
 838 /* XXX PRISON: could be per prison flag */
 839 static int prison_quotas;
 840 #if 0
 841 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
 842 #endif
 843
 844 /*
 845  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
 846  *
 847  * Change filesystem quotas.
 848  */
 849 /* ARGSUSED */
 850 int
 851 sys_quotactl(struct quotactl_args *uap)
 852 {
 853         struct nlookupdata nd;
 854         struct thread *td;
 855         struct proc *p;
 856         struct mount *mp;
 857         int error;
 858
 859         td = curthread;
 860         p = td->td_proc;
 861         if (p->p_ucred->cr_prison && !prison_quotas)
 862                 return (EPERM);
 863
 864         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 865         if (error == 0)
 866                 error = nlookup(&nd);
 867         if (error == 0) {
 868                 mp = nd.nl_nch.mount;
 869                 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
 870                                     uap->arg, nd.nl_cred);
 871         }
 872         nlookup_done(&nd);
 873         return (error);
 874 }
 875
 876 /*
 877  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
 878  *              void *buf, int buflen)
 879  *
 880  * This function operates on a mount point and executes the specified
 881  * operation using the specified control data, and possibly returns data.
 882  *
 883  * The actual number of bytes stored in the result buffer is returned, 0
 884  * if none, otherwise an error is returned.
 885  */
 886 /* ARGSUSED */
 887 int
 888 sys_mountctl(struct mountctl_args *uap)
 889 {
 890         struct thread *td = curthread;
 891         struct proc *p = td->td_proc;
 892         struct file *fp;
 893         void *ctl = NULL;
 894         void *buf = NULL;
 895         char *path = NULL;
 896         int error;
 897
 898         /*
 899          * Sanity and permissions checks.  We must be root.
 900          */
 901         KKASSERT(p);
 902         if (p->p_ucred->cr_prison != NULL)
 903                 return (EPERM);
 904         if ((error = suser(td)) != 0)
 905                 return (error);
 906
 907         /*
 908          * Argument length checks
 909          */
 910         if (uap->ctllen < 0 || uap->ctllen > 1024)
 911                 return (EINVAL);
 912         if (uap->buflen < 0 || uap->buflen > 16 * 1024)
 913                 return (EINVAL);
 914         if (uap->path == NULL)
 915                 return (EINVAL);
 916
 917         /*
 918          * Allocate the necessary buffers and copyin data
 919          */
 920         path = objcache_get(namei_oc, M_WAITOK);
 921         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 922         if (error)
 923                 goto done;
 924
 925         if (uap->ctllen) {
 926                 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
 927                 error = copyin(uap->ctl, ctl, uap->ctllen);
 928                 if (error)
 929                         goto done;
 930         }
 931         if (uap->buflen)
 932                 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
 933
 934         /*
 935          * Validate the descriptor
 936          */
 937         if (uap->fd >= 0) {
 938                 fp = holdfp(p->p_fd, uap->fd, -1);
 939                 if (fp == NULL) {
 940                         error = EBADF;
 941                         goto done;
 942                 }
 943         } else {
 944                 fp = NULL;
 945         }
 946
 947         /*
 948          * Execute the internal kernel function and clean up.
 949          */
 950         error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
 951         if (fp)
 952                 fdrop(fp);
 953         if (error == 0 && uap->sysmsg_result > 0)
 954                 error = copyout(buf, uap->buf, uap->sysmsg_result);
 955 done:
 956         if (path)
 957                 objcache_put(namei_oc, path);
 958         if (ctl)
 959                 kfree(ctl, M_TEMP);
 960         if (buf)
 961                 kfree(buf, M_TEMP);
 962         return (error);
 963 }
 964
 965 /*
 966  * Execute a mount control operation by resolving the path to a mount point
 967  * and calling vop_mountctl().
 968  */
 969 int
 970 kern_mountctl(const char *path, int op, struct file *fp,
 971                 const void *ctl, int ctllen,
 972                 void *buf, int buflen, int *res)
 973 {
 974         struct vnode *vp;
 975         struct mount *mp;
 976         struct nlookupdata nd;
 977         int error;
 978
 979         *res = 0;
 980         vp = NULL;
 981         error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
 982         if (error == 0)
 983                 error = nlookup(&nd);
 984         if (error == 0)
 985                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 986         nlookup_done(&nd);
 987         if (error)
 988                 return (error);
 989
 990         mp = vp->v_mount;
 991
 992         /*
 993          * Must be the root of the filesystem
 994          */
 995         if ((vp->v_flag & VROOT) == 0) {
 996                 vput(vp);
 997                 return (EINVAL);
 998         }
 999         error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1000                                 buf, buflen, res);
1001         vput(vp);
1002         return (error);
1003 }
1004
1005 int
1006 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1007 {
1008         struct thread *td = curthread;
1009         struct proc *p = td->td_proc;
1010         struct mount *mp;
1011         struct statfs *sp;
1012         char *fullpath, *freepath;
1013         int error;
1014
1015         if ((error = nlookup(nd)) != 0)
1016                 return (error);
1017         mp = nd->nl_nch.mount;
1018         sp = &mp->mnt_stat;
1019         if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1020                 return (error);
1021
1022         error = mount_path(p, mp, &fullpath, &freepath);
1023         if (error)
1024                 return(error);
1025         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1026         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1027         kfree(freepath, M_TEMP);
1028
1029         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1030         bcopy(sp, buf, sizeof(*buf));
1031         /* Only root should have access to the fsid's. */
1032         if (suser(td))
1033                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1034         return (0);
1035 }
1036
1037 /*
1038  * statfs_args(char *path, struct statfs *buf)
1039  *
1040  * Get filesystem statistics.
1041  */
1042 int
1043 sys_statfs(struct statfs_args *uap)
1044 {
1045         struct nlookupdata nd;
1046         struct statfs buf;
1047         int error;
1048
1049         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1050         if (error == 0)
1051                 error = kern_statfs(&nd, &buf);
1052         nlookup_done(&nd);
1053         if (error == 0)
1054                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1055         return (error);
1056 }
1057
1058 int
1059 kern_fstatfs(int fd, struct statfs *buf)
1060 {
1061         struct thread *td = curthread;
1062         struct proc *p = td->td_proc;
1063         struct file *fp;
1064         struct mount *mp;
1065         struct statfs *sp;
1066         char *fullpath, *freepath;
1067         int error;
1068
1069         KKASSERT(p);
1070         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1071                 return (error);
1072         mp = ((struct vnode *)fp->f_data)->v_mount;
1073         if (mp == NULL) {
1074                 error = EBADF;
1075                 goto done;
1076         }
1077         if (fp->f_cred == NULL) {
1078                 error = EINVAL;
1079                 goto done;
1080         }
1081         sp = &mp->mnt_stat;
1082         if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1083                 goto done;
1084
1085         if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1086                 goto done;
1087         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1088         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1089         kfree(freepath, M_TEMP);
1090
1091         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1092         bcopy(sp, buf, sizeof(*buf));
1093
1094         /* Only root should have access to the fsid's. */
1095         if (suser(td))
1096                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1097         error = 0;
1098 done:
1099         fdrop(fp);
1100         return (error);
1101 }
1102
1103 /*
1104  * fstatfs_args(int fd, struct statfs *buf)
1105  *
1106  * Get filesystem statistics.
1107  */
1108 int
1109 sys_fstatfs(struct fstatfs_args *uap)
1110 {
1111         struct statfs buf;
1112         int error;
1113
1114         error = kern_fstatfs(uap->fd, &buf);
1115
1116         if (error == 0)
1117                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1118         return (error);
1119 }
1120
1121 /*
1122  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1123  *
1124  * Get statistics on all filesystems.
1125  */
1126
1127 struct getfsstat_info {
1128         struct statfs *sfsp;
1129         long count;
1130         long maxcount;
1131         int error;
1132         int flags;
1133         struct proc *p;
1134 };
1135
1136 static int getfsstat_callback(struct mount *, void *);
1137
1138 /* ARGSUSED */
1139 int
1140 sys_getfsstat(struct getfsstat_args *uap)
1141 {
1142         struct thread *td = curthread;
1143         struct proc *p = td->td_proc;
1144         struct getfsstat_info info;
1145
1146         bzero(&info, sizeof(info));
1147
1148         info.maxcount = uap->bufsize / sizeof(struct statfs);
1149         info.sfsp = uap->buf;
1150         info.count = 0;
1151         info.flags = uap->flags;
1152         info.p = p;
1153
1154         mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1155         if (info.sfsp && info.count > info.maxcount)
1156                 uap->sysmsg_result = info.maxcount;
1157         else
1158                 uap->sysmsg_result = info.count;
1159         return (info.error);
1160 }
1161
1162 static int
1163 getfsstat_callback(struct mount *mp, void *data)
1164 {
1165         struct getfsstat_info *info = data;
1166         struct statfs *sp;
1167         char *freepath;
1168         char *fullpath;
1169         int error;
1170
1171         if (info->sfsp && info->count < info->maxcount) {
1172                 if (info->p && !chroot_visible_mnt(mp, info->p))
1173                         return(0);
1174                 sp = &mp->mnt_stat;
1175
1176                 /*
1177                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1178                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1179                  * overrides MNT_WAIT.
1180                  */
1181                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1182                     (info->flags & MNT_WAIT)) &&
1183                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1184                         return(0);
1185                 }
1186                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1187
1188                 error = mount_path(info->p, mp, &fullpath, &freepath);
1189                 if (error) {
1190                         info->error = error;
1191                         return(-1);
1192                 }
1193                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1194                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1195                 kfree(freepath, M_TEMP);
1196
1197                 error = copyout(sp, info->sfsp, sizeof(*sp));
1198                 if (error) {
1199                         info->error = error;
1200                         return (-1);
1201                 }
1202                 ++info->sfsp;
1203         }
1204         info->count++;
1205         return(0);
1206 }
1207
1208 /*
1209  * fchdir_args(int fd)
1210  *
1211  * Change current working directory to a given file descriptor.
1212  */
1213 /* ARGSUSED */
1214 int
1215 sys_fchdir(struct fchdir_args *uap)
1216 {
1217         struct thread *td = curthread;
1218         struct proc *p = td->td_proc;
1219         struct filedesc *fdp = p->p_fd;
1220         struct vnode *vp, *ovp;
1221         struct mount *mp;
1222         struct file *fp;
1223         struct nchandle nch, onch, tnch;
1224         int error;
1225
1226         if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1227                 return (error);
1228         vp = (struct vnode *)fp->f_data;
1229         vref(vp);
1230         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1231         if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1232                 error = ENOTDIR;
1233         else
1234                 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1235         if (error) {
1236                 vput(vp);
1237                 fdrop(fp);
1238                 return (error);
1239         }
1240         cache_copy(&fp->f_nchandle, &nch);
1241
1242         /*
1243          * If the ncp has become a mount point, traverse through
1244          * the mount point.
1245          */
1246
1247         while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1248                (mp = cache_findmount(&nch)) != NULL
1249         ) {
1250                 error = nlookup_mp(mp, &tnch);
1251                 if (error == 0) {
1252                         cache_unlock(&tnch);    /* leave ref intact */
1253                         vput(vp);
1254                         vp = tnch.ncp->nc_vp;
1255                         error = vget(vp, LK_SHARED);
1256                         KKASSERT(error == 0);
1257                         cache_drop(&nch);
1258                         nch = tnch;
1259                 }
1260         }
1261         if (error == 0) {
1262                 ovp = fdp->fd_cdir;
1263                 onch = fdp->fd_ncdir;
1264                 vn_unlock(vp);          /* leave ref intact */
1265                 fdp->fd_cdir = vp;
1266                 fdp->fd_ncdir = nch;
1267                 cache_drop(&onch);
1268                 vrele(ovp);
1269         } else {
1270                 cache_drop(&nch);
1271                 vput(vp);
1272         }
1273         fdrop(fp);
1274         return (error);
1275 }
1276
1277 int
1278 kern_chdir(struct nlookupdata *nd)
1279 {
1280         struct thread *td = curthread;
1281         struct proc *p = td->td_proc;
1282         struct filedesc *fdp = p->p_fd;
1283         struct vnode *vp, *ovp;
1284         struct nchandle onch;
1285         int error;
1286
1287         if ((error = nlookup(nd)) != 0)
1288                 return (error);
1289         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1290                 return (ENOENT);
1291         if ((error = vget(vp, LK_SHARED)) != 0)
1292                 return (error);
1293
1294         error = checkvp_chdir(vp, td);
1295         vn_unlock(vp);
1296         if (error == 0) {
1297                 ovp = fdp->fd_cdir;
1298                 onch = fdp->fd_ncdir;
1299                 cache_unlock(&nd->nl_nch);      /* leave reference intact */
1300                 fdp->fd_ncdir = nd->nl_nch;
1301                 fdp->fd_cdir = vp;
1302                 cache_drop(&onch);
1303                 vrele(ovp);
1304                 cache_zero(&nd->nl_nch);
1305         } else {
1306                 vrele(vp);
1307         }
1308         return (error);
1309 }
1310
1311 /*
1312  * chdir_args(char *path)
1313  *
1314  * Change current working directory (``.'').
1315  */
1316 int
1317 sys_chdir(struct chdir_args *uap)
1318 {
1319         struct nlookupdata nd;
1320         int error;
1321
1322         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1323         if (error == 0)
1324                 error = kern_chdir(&nd);
1325         nlookup_done(&nd);
1326         return (error);
1327 }
1328
1329 /*
1330  * Helper function for raised chroot(2) security function:  Refuse if
1331  * any filedescriptors are open directories.
1332  */
1333 static int
1334 chroot_refuse_vdir_fds(struct filedesc *fdp)
1335 {
1336         struct vnode *vp;
1337         struct file *fp;
1338         int error;
1339         int fd;
1340
1341         for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1342                 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1343                         continue;
1344                 vp = (struct vnode *)fp->f_data;
1345                 if (vp->v_type != VDIR) {
1346                         fdrop(fp);
1347                         continue;
1348                 }
1349                 fdrop(fp);
1350                 return(EPERM);
1351         }
1352         return (0);
1353 }
1354
1355 /*
1356  * This sysctl determines if we will allow a process to chroot(2) if it
1357  * has a directory open:
1358  *      0: disallowed for all processes.
1359  *      1: allowed for processes that were not already chroot(2)'ed.
1360  *      2: allowed for all processes.
1361  */
1362
1363 static int chroot_allow_open_directories = 1;
1364
1365 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1366      &chroot_allow_open_directories, 0, "");
1367
1368 /*
1369  * chroot to the specified namecache entry.  We obtain the vp from the
1370  * namecache data.  The passed ncp must be locked and referenced and will
1371  * remain locked and referenced on return.
1372  */
1373 int
1374 kern_chroot(struct nchandle *nch)
1375 {
1376         struct thread *td = curthread;
1377         struct proc *p = td->td_proc;
1378         struct filedesc *fdp = p->p_fd;
1379         struct vnode *vp;
1380         int error;
1381
1382         /*
1383          * Only root can chroot
1384          */
1385         if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1386                 return (error);
1387
1388         /*
1389          * Disallow open directory descriptors (fchdir() breakouts).
1390          */
1391         if (chroot_allow_open_directories == 0 ||
1392            (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1393                 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1394                         return (error);
1395         }
1396         if ((vp = nch->ncp->nc_vp) == NULL)
1397                 return (ENOENT);
1398
1399         if ((error = vget(vp, LK_SHARED)) != 0)
1400                 return (error);
1401
1402         /*
1403          * Check the validity of vp as a directory to change to and
1404          * associate it with rdir/jdir.
1405          */
1406         error = checkvp_chdir(vp, td);
1407         vn_unlock(vp);                  /* leave reference intact */
1408         if (error == 0) {
1409                 vrele(fdp->fd_rdir);
1410                 fdp->fd_rdir = vp;      /* reference inherited by fd_rdir */
1411                 cache_drop(&fdp->fd_nrdir);
1412                 cache_copy(nch, &fdp->fd_nrdir);
1413                 if (fdp->fd_jdir == NULL) {
1414                         fdp->fd_jdir = vp;
1415                         vref(fdp->fd_jdir);
1416                         cache_copy(nch, &fdp->fd_njdir);
1417                 }
1418         } else {
1419                 vrele(vp);
1420         }
1421         return (error);
1422 }
1423
1424 /*
1425  * chroot_args(char *path)
1426  *
1427  * Change notion of root (``/'') directory.
1428  */
1429 /* ARGSUSED */
1430 int
1431 sys_chroot(struct chroot_args *uap)
1432 {
1433         struct thread *td = curthread;
1434         struct nlookupdata nd;
1435         int error;
1436
1437         KKASSERT(td->td_proc);
1438         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1439         if (error) {
1440                 nlookup_done(&nd);
1441                 return(error);
1442         }
1443         error = nlookup(&nd);
1444         if (error == 0)
1445                 error = kern_chroot(&nd.nl_nch);
1446         nlookup_done(&nd);
1447         return(error);
1448 }
1449
1450 /*
1451  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1452  * determine whether it is legal to chdir to the vnode.  The vnode's state
1453  * is not changed by this call.
1454  */
1455 int
1456 checkvp_chdir(struct vnode *vp, struct thread *td)
1457 {
1458         int error;
1459
1460         if (vp->v_type != VDIR)
1461                 error = ENOTDIR;
1462         else
1463                 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1464         return (error);
1465 }
1466
1467 int
1468 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1469 {
1470         struct thread *td = curthread;
1471         struct proc *p = td->td_proc;
1472         struct lwp *lp = td->td_lwp;
1473         struct filedesc *fdp = p->p_fd;
1474         int cmode, flags;
1475         struct file *nfp;
1476         struct file *fp;
1477         struct vnode *vp;
1478         int type, indx, error;
1479         struct flock lf;
1480
1481         if ((oflags & O_ACCMODE) == O_ACCMODE)
1482                 return (EINVAL);
1483         flags = FFLAGS(oflags);
1484         error = falloc(p, &nfp, NULL);
1485         if (error)
1486                 return (error);
1487         fp = nfp;
1488         cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1489
1490         /*
1491          * XXX p_dupfd is a real mess.  It allows a device to return a
1492          * file descriptor to be duplicated rather then doing the open
1493          * itself.
1494          */
1495         lp->lwp_dupfd = -1;
1496
1497         /*
1498          * Call vn_open() to do the lookup and assign the vnode to the
1499          * file pointer.  vn_open() does not change the ref count on fp
1500          * and the vnode, on success, will be inherited by the file pointer
1501          * and unlocked.
1502          */
1503         nd->nl_flags |= NLC_LOCKVP;
1504         error = vn_open(nd, fp, flags, cmode);
1505         nlookup_done(nd);
1506         if (error) {
1507                 /*
1508                  * handle special fdopen() case.  bleh.  dupfdopen() is
1509                  * responsible for dropping the old contents of ofiles[indx]
1510                  * if it succeeds.
1511                  *
1512                  * Note that fsetfd() will add a ref to fp which represents
1513                  * the fd_files[] assignment.  We must still drop our
1514                  * reference.
1515                  */
1516                 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1517                         if (fdalloc(p, 0, &indx) == 0) {
1518                                 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1519                                 if (error == 0) {
1520                                         *res = indx;
1521                                         fdrop(fp);      /* our ref */
1522                                         return (0);
1523                                 }
1524                                 fsetfd(p, NULL, indx);
1525                         }
1526                 }
1527                 fdrop(fp);      /* our ref */
1528                 if (error == ERESTART)
1529                         error = EINTR;
1530                 return (error);
1531         }
1532
1533         /*
1534          * ref the vnode for ourselves so it can't be ripped out from under
1535          * is.  XXX need an ND flag to request that the vnode be returned
1536          * anyway.
1537          *
1538          * Reserve a file descriptor but do not assign it until the open
1539          * succeeds.
1540          */
1541         vp = (struct vnode *)fp->f_data;
1542         vref(vp);
1543         if ((error = fdalloc(p, 0, &indx)) != 0) {
1544                 fdrop(fp);
1545                 vrele(vp);
1546                 return (error);
1547         }
1548
1549         /*
1550          * If no error occurs the vp will have been assigned to the file
1551          * pointer.
1552          */
1553         lp->lwp_dupfd = 0;
1554
1555         if (flags & (O_EXLOCK | O_SHLOCK)) {
1556                 lf.l_whence = SEEK_SET;
1557                 lf.l_start = 0;
1558                 lf.l_len = 0;
1559                 if (flags & O_EXLOCK)
1560                         lf.l_type = F_WRLCK;
1561                 else
1562                         lf.l_type = F_RDLCK;
1563                 if (flags & FNONBLOCK)
1564                         type = 0;
1565                 else
1566                         type = F_WAIT;
1567
1568                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1569                         /*
1570                          * lock request failed.  Clean up the reserved
1571                          * descriptor.
1572                          */
1573                         vrele(vp);
1574                         fsetfd(p, NULL, indx);
1575                         fdrop(fp);
1576                         return (error);
1577                 }
1578                 fp->f_flag |= FHASLOCK;
1579         }
1580 #if 0
1581         /*
1582          * Assert that all regular file vnodes were created with a object.
1583          */
1584         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1585                 ("open: regular file has no backing object after vn_open"));
1586 #endif
1587
1588         vrele(vp);
1589
1590         /*
1591          * release our private reference, leaving the one associated with the
1592          * descriptor table intact.
1593          */
1594         fsetfd(p, fp, indx);
1595         fdrop(fp);
1596         *res = indx;
1597         return (0);
1598 }
1599
1600 /*
1601  * open_args(char *path, int flags, int mode)
1602  *
1603  * Check permissions, allocate an open file structure,
1604  * and call the device open routine if any.
1605  */
1606 int
1607 sys_open(struct open_args *uap)
1608 {
1609         struct nlookupdata nd;
1610         int error;
1611
1612         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1613         if (error == 0) {
1614                 error = kern_open(&nd, uap->flags,
1615                                     uap->mode, &uap->sysmsg_result);
1616         }
1617         nlookup_done(&nd);
1618         return (error);
1619 }
1620
1621 int
1622 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1623 {
1624         struct thread *td = curthread;
1625         struct proc *p = td->td_proc;
1626         struct vnode *vp;
1627         struct vnode *dvp;
1628         struct vattr vattr;
1629         int error;
1630         int whiteout = 0;
1631
1632         KKASSERT(p);
1633
1634         switch (mode & S_IFMT) {
1635         case S_IFCHR:
1636         case S_IFBLK:
1637                 error = suser(td);
1638                 break;
1639         default:
1640                 error = suser_cred(p->p_ucred, PRISON_ROOT);
1641                 break;
1642         }
1643         if (error)
1644                 return (error);
1645
1646         bwillwrite();
1647         nd->nl_flags |= NLC_CREATE;
1648         if ((error = nlookup(nd)) != 0)
1649                 return (error);
1650         if (nd->nl_nch.ncp->nc_vp)
1651                 return (EEXIST);
1652         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1653                 return (error);
1654         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
1655                 return (EPERM);
1656         /* vhold(dvp); - DVP can't go away */
1657
1658         VATTR_NULL(&vattr);
1659         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1660         vattr.va_rmajor = rmajor;
1661         vattr.va_rminor = rminor;
1662         whiteout = 0;
1663
1664         switch (mode & S_IFMT) {
1665         case S_IFMT:    /* used by badsect to flag bad sectors */
1666                 vattr.va_type = VBAD;
1667                 break;
1668         case S_IFCHR:
1669                 vattr.va_type = VCHR;
1670                 break;
1671         case S_IFBLK:
1672                 vattr.va_type = VBLK;
1673                 break;
1674         case S_IFWHT:
1675                 whiteout = 1;
1676                 break;
1677         default:
1678                 error = EINVAL;
1679                 break;
1680         }
1681         if (error == 0) {
1682                 if (whiteout) {
1683                         error = VOP_NWHITEOUT(&nd->nl_nch, dvp, nd->nl_cred, NAMEI_CREATE);
1684                 } else {
1685                         vp = NULL;
1686                         error = VOP_NMKNOD(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr);
1687                         if (error == 0)
1688                                 vput(vp);
1689                 }
1690         }
1691         /* vdrop(dvp); */
1692         return (error);
1693 }
1694
1695 /*
1696  * mknod_args(char *path, int mode, int dev)
1697  *
1698  * Create a special file.
1699  */
1700 int
1701 sys_mknod(struct mknod_args *uap)
1702 {
1703         struct nlookupdata nd;
1704         int error;
1705
1706         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1707         if (error == 0) {
1708                 error = kern_mknod(&nd, uap->mode,
1709                                    umajor(uap->dev), uminor(uap->dev));
1710         }
1711         nlookup_done(&nd);
1712         return (error);
1713 }
1714
1715 int
1716 kern_mkfifo(struct nlookupdata *nd, int mode)
1717 {
1718         struct thread *td = curthread;
1719         struct proc *p = td->td_proc;
1720         struct vattr vattr;
1721         struct vnode *vp;
1722         struct vnode *dvp;
1723         int error;
1724
1725         bwillwrite();
1726
1727         nd->nl_flags |= NLC_CREATE;
1728         if ((error = nlookup(nd)) != 0)
1729                 return (error);
1730         if (nd->nl_nch.ncp->nc_vp)
1731                 return (EEXIST);
1732         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1733                 return (error);
1734         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
1735                 return (EPERM);
1736         /* vhold(dvp); - DVP can't go away */
1737
1738         VATTR_NULL(&vattr);
1739         vattr.va_type = VFIFO;
1740         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1741         vp = NULL;
1742         error = VOP_NMKNOD(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr);
1743         /* vdrop(dvp); */
1744         if (error == 0)
1745                 vput(vp);
1746         return (error);
1747 }
1748
1749 /*
1750  * mkfifo_args(char *path, int mode)
1751  *
1752  * Create a named pipe.
1753  */
1754 int
1755 sys_mkfifo(struct mkfifo_args *uap)
1756 {
1757         struct nlookupdata nd;
1758         int error;
1759
1760         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1761         if (error == 0)
1762                 error = kern_mkfifo(&nd, uap->mode);
1763         nlookup_done(&nd);
1764         return (error);
1765 }
1766
1767 static int hardlink_check_uid = 0;
1768 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1769     &hardlink_check_uid, 0,
1770     "Unprivileged processes cannot create hard links to files owned by other "
1771     "users");
1772 static int hardlink_check_gid = 0;
1773 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1774     &hardlink_check_gid, 0,
1775     "Unprivileged processes cannot create hard links to files owned by other "
1776     "groups");
1777
1778 static int
1779 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1780 {
1781         struct vattr va;
1782         int error;
1783
1784         /*
1785          * Shortcut if disabled
1786          */
1787         if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1788                 return (0);
1789
1790         /*
1791          * root cred can always hardlink
1792          */
1793         if (suser_cred(cred, PRISON_ROOT) == 0)
1794                 return (0);
1795
1796         /*
1797          * Otherwise only if the originating file is owned by the
1798          * same user or group.  Note that any group is allowed if
1799          * the file is owned by the caller.
1800          */
1801         error = VOP_GETATTR(vp, &va);
1802         if (error != 0)
1803                 return (error);
1804
1805         if (hardlink_check_uid) {
1806                 if (cred->cr_uid != va.va_uid)
1807                         return (EPERM);
1808         }
1809
1810         if (hardlink_check_gid) {
1811                 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1812                         return (EPERM);
1813         }
1814
1815         return (0);
1816 }
1817
1818 int
1819 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1820 {
1821         struct thread *td = curthread;
1822         struct vnode *vp;
1823         struct vnode *dvp;
1824         int error;
1825
1826         /*
1827          * Lookup the source and obtained a locked vnode.
1828          *
1829          * XXX relookup on vget failure / race ?
1830          */
1831         bwillwrite();
1832         if ((error = nlookup(nd)) != 0)
1833                 return (error);
1834         vp = nd->nl_nch.ncp->nc_vp;
1835         KKASSERT(vp != NULL);
1836         if (vp->v_type == VDIR)
1837                 return (EPERM);         /* POSIX */
1838         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1839                 return (error);
1840         if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1841                 return (error);
1842
1843         /*
1844          * Unlock the source so we can lookup the target without deadlocking
1845          * (XXX vp is locked already, possible other deadlock?).  The target
1846          * must not exist.
1847          */
1848         KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1849         nd->nl_flags &= ~NLC_NCPISLOCKED;
1850         cache_unlock(&nd->nl_nch);
1851
1852         linknd->nl_flags |= NLC_CREATE;
1853         if ((error = nlookup(linknd)) != 0) {
1854                 vput(vp);
1855                 return (error);
1856         }
1857         if (linknd->nl_nch.ncp->nc_vp) {
1858                 vput(vp);
1859                 return (EEXIST);
1860         }
1861         if ((dvp = linknd->nl_nch.ncp->nc_parent->nc_vp) == NULL) {
1862                 vput(vp);
1863                 return (EPERM);
1864         }
1865         /* vhold(dvp); - dvp can't go away */
1866
1867         /*
1868          * Finally run the new API VOP.
1869          */
1870         error = can_hardlink(vp, td, td->td_proc->p_ucred);
1871         if (error == 0)
1872                 error = VOP_NLINK(&linknd->nl_nch, dvp, vp, linknd->nl_cred);
1873         /* vdrop(dvp); */
1874         vput(vp);
1875         return (error);
1876 }
1877
1878 /*
1879  * link_args(char *path, char *link)
1880  *
1881  * Make a hard file link.
1882  */
1883 int
1884 sys_link(struct link_args *uap)
1885 {
1886         struct nlookupdata nd, linknd;
1887         int error;
1888
1889         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1890         if (error == 0) {
1891                 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1892                 if (error == 0)
1893                         error = kern_link(&nd, &linknd);
1894                 nlookup_done(&linknd);
1895         }
1896         nlookup_done(&nd);
1897         return (error);
1898 }
1899
1900 int
1901 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1902 {
1903         struct vattr vattr;
1904         struct vnode *vp;
1905         struct vnode *dvp;
1906         int error;
1907
1908         bwillwrite();
1909         nd->nl_flags |= NLC_CREATE;
1910         if ((error = nlookup(nd)) != 0)
1911                 return (error);
1912         if (nd->nl_nch.ncp->nc_vp)
1913                 return (EEXIST);
1914         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1915                 return (error);
1916         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
1917                 return (EPERM);
1918         /* vhold(dvp); - dvp can't go away */
1919         VATTR_NULL(&vattr);
1920         vattr.va_mode = mode;
1921         error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
1922         /* vdrop(dvp); */
1923         if (error == 0)
1924                 vput(vp);
1925         return (error);
1926 }
1927
1928 /*
1929  * symlink(char *path, char *link)
1930  *
1931  * Make a symbolic link.
1932  */
1933 int
1934 sys_symlink(struct symlink_args *uap)
1935 {
1936         struct thread *td = curthread;
1937         struct nlookupdata nd;
1938         char *path;
1939         int error;
1940         int mode;
1941
1942         path = objcache_get(namei_oc, M_WAITOK);
1943         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1944         if (error == 0) {
1945                 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1946                 if (error == 0) {
1947                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1948                         error = kern_symlink(&nd, path, mode);
1949                 }
1950                 nlookup_done(&nd);
1951         }
1952         objcache_put(namei_oc, path);
1953         return (error);
1954 }
1955
1956 /*
1957  * undelete_args(char *path)
1958  *
1959  * Delete a whiteout from the filesystem.
1960  */
1961 /* ARGSUSED */
1962 int
1963 sys_undelete(struct undelete_args *uap)
1964 {
1965         struct nlookupdata nd;
1966         struct vnode *dvp;
1967         int error;
1968
1969         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1970         bwillwrite();
1971         nd.nl_flags |= NLC_DELETE;
1972         if (error == 0)
1973                 error = nlookup(&nd);
1974         if (error == 0)
1975                 error = ncp_writechk(&nd.nl_nch);
1976         dvp = NULL;
1977         if (error == 0) {
1978                 if ((dvp = nd.nl_nch.ncp->nc_parent->nc_vp) == NULL)
1979                         error = EPERM;
1980         }
1981         if (error == 0) {
1982                 /* vhold(dvp); - dvp can't go away */
1983                 error = VOP_NWHITEOUT(&nd.nl_nch, dvp, nd.nl_cred, NAMEI_DELETE);
1984                 /* vdrop(dvp); */
1985         }
1986         nlookup_done(&nd);
1987         return (error);
1988 }
1989
1990 int
1991 kern_unlink(struct nlookupdata *nd)
1992 {
1993         struct vnode *dvp;
1994         int error;
1995
1996         bwillwrite();
1997         nd->nl_flags |= NLC_DELETE;
1998         if ((error = nlookup(nd)) != 0)
1999                 return (error);
2000         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2001                 return (error);
2002         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
2003                 return (EPERM);
2004         /* vhold(dvp); - dvp can't go away */
2005         error = VOP_NREMOVE(&nd->nl_nch, dvp, nd->nl_cred);
2006         /* vdrop(dvp); */
2007         return (error);
2008 }
2009
2010 /*
2011  * unlink_args(char *path)
2012  *
2013  * Delete a name from the filesystem.
2014  */
2015 int
2016 sys_unlink(struct unlink_args *uap)
2017 {
2018         struct nlookupdata nd;
2019         int error;
2020
2021         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2022         if (error == 0)
2023                 error = kern_unlink(&nd);
2024         nlookup_done(&nd);
2025         return (error);
2026 }
2027
2028 int
2029 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2030 {
2031         struct thread *td = curthread;
2032         struct proc *p = td->td_proc;
2033         struct file *fp;
2034         struct vattr vattr;
2035         int error;
2036
2037         fp = holdfp(p->p_fd, fd, -1);
2038         if (fp == NULL)
2039                 return (EBADF);
2040         if (fp->f_type != DTYPE_VNODE) {
2041                 error = ESPIPE;
2042                 goto done;
2043         }
2044
2045         switch (whence) {
2046         case L_INCR:
2047                 fp->f_offset += offset;
2048                 error = 0;
2049                 break;
2050         case L_XTND:
2051                 error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
2052                 if (error == 0)
2053                         fp->f_offset = offset + vattr.va_size;
2054                 break;
2055         case L_SET:
2056                 fp->f_offset = offset;
2057                 error = 0;
2058                 break;
2059         default:
2060                 error = EINVAL;
2061                 break;
2062         }
2063         *res = fp->f_offset;
2064 done:
2065         fdrop(fp);
2066         return (error);
2067 }
2068
2069 /*
2070  * lseek_args(int fd, int pad, off_t offset, int whence)
2071  *
2072  * Reposition read/write file offset.
2073  */
2074 int
2075 sys_lseek(struct lseek_args *uap)
2076 {
2077         int error;
2078
2079         error = kern_lseek(uap->fd, uap->offset, uap->whence,
2080             &uap->sysmsg_offset);
2081
2082         return (error);
2083 }
2084
2085 int
2086 kern_access(struct nlookupdata *nd, int aflags)
2087 {
2088         struct vnode *vp;
2089         int error, flags;
2090
2091         if ((error = nlookup(nd)) != 0)
2092                 return (error);
2093 retry:
2094         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2095         if (error)
2096                 return (error);
2097
2098         /* Flags == 0 means only check for existence. */
2099         if (aflags) {
2100                 flags = 0;
2101                 if (aflags & R_OK)
2102                         flags |= VREAD;
2103                 if (aflags & W_OK)
2104                         flags |= VWRITE;
2105                 if (aflags & X_OK)
2106                         flags |= VEXEC;
2107                 if ((flags & VWRITE) == 0 ||
2108                     (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2109                         error = VOP_ACCESS(vp, flags, nd->nl_cred);
2110
2111                 /*
2112                  * If the file handle is stale we have to re-resolve the
2113                  * entry.  This is a hack at the moment.
2114                  */
2115                 if (error == ESTALE) {
2116                         cache_setunresolved(&nd->nl_nch);
2117                         error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2118                         if (error == 0) {
2119                                 vput(vp);
2120                                 vp = NULL;
2121                                 goto retry;
2122                         }
2123                 }
2124         }
2125         vput(vp);
2126         return (error);
2127 }
2128
2129 /*
2130  * access_args(char *path, int flags)
2131  *
2132  * Check access permissions.
2133  */
2134 int
2135 sys_access(struct access_args *uap)
2136 {
2137         struct nlookupdata nd;
2138         int error;
2139
2140         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2141         if (error == 0)
2142                 error = kern_access(&nd, uap->flags);
2143         nlookup_done(&nd);
2144         return (error);
2145 }
2146
2147 int
2148 kern_stat(struct nlookupdata *nd, struct stat *st)
2149 {
2150         int error;
2151         struct vnode *vp;
2152         thread_t td;
2153
2154         if ((error = nlookup(nd)) != 0)
2155                 return (error);
2156 again:
2157         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2158                 return (ENOENT);
2159
2160         td = curthread;
2161         if ((error = vget(vp, LK_SHARED)) != 0)
2162                 return (error);
2163         error = vn_stat(vp, st, nd->nl_cred);
2164
2165         /*
2166          * If the file handle is stale we have to re-resolve the entry.  This
2167          * is a hack at the moment.
2168          */
2169         if (error == ESTALE) {
2170                 vput(vp);
2171                 cache_setunresolved(&nd->nl_nch);
2172                 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2173                 if (error == 0)
2174                         goto again;
2175         } else {
2176                 vput(vp);
2177         }
2178         return (error);
2179 }
2180
2181 /*
2182  * stat_args(char *path, struct stat *ub)
2183  *
2184  * Get file status; this version follows links.
2185  */
2186 int
2187 sys_stat(struct stat_args *uap)
2188 {
2189         struct nlookupdata nd;
2190         struct stat st;
2191         int error;
2192
2193         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2194         if (error == 0) {
2195                 error = kern_stat(&nd, &st);
2196                 if (error == 0)
2197                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2198         }
2199         nlookup_done(&nd);
2200         return (error);
2201 }
2202
2203 /*
2204  * lstat_args(char *path, struct stat *ub)
2205  *
2206  * Get file status; this version does not follow links.
2207  */
2208 int
2209 sys_lstat(struct lstat_args *uap)
2210 {
2211         struct nlookupdata nd;
2212         struct stat st;
2213         int error;
2214
2215         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2216         if (error == 0) {
2217                 error = kern_stat(&nd, &st);
2218                 if (error == 0)
2219                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2220         }
2221         nlookup_done(&nd);
2222         return (error);
2223 }
2224
2225 /*
2226  * pathconf_Args(char *path, int name)
2227  *
2228  * Get configurable pathname variables.
2229  */
2230 /* ARGSUSED */
2231 int
2232 sys_pathconf(struct pathconf_args *uap)
2233 {
2234         struct nlookupdata nd;
2235         struct vnode *vp;
2236         int error;
2237
2238         vp = NULL;
2239         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2240         if (error == 0)
2241                 error = nlookup(&nd);
2242         if (error == 0)
2243                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2244         nlookup_done(&nd);
2245         if (error == 0) {
2246                 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2247                 vput(vp);
2248         }
2249         return (error);
2250 }
2251
2252 /*
2253  * XXX: daver
2254  * kern_readlink isn't properly split yet.  There is a copyin burried
2255  * in VOP_READLINK().
2256  */
2257 int
2258 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2259 {
2260         struct thread *td = curthread;
2261         struct proc *p = td->td_proc;
2262         struct vnode *vp;
2263         struct iovec aiov;
2264         struct uio auio;
2265         int error;
2266
2267         if ((error = nlookup(nd)) != 0)
2268                 return (error);
2269         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2270         if (error)
2271                 return (error);
2272         if (vp->v_type != VLNK) {
2273                 error = EINVAL;
2274         } else {
2275                 aiov.iov_base = buf;
2276                 aiov.iov_len = count;
2277                 auio.uio_iov = &aiov;
2278                 auio.uio_iovcnt = 1;
2279                 auio.uio_offset = 0;
2280                 auio.uio_rw = UIO_READ;
2281                 auio.uio_segflg = UIO_USERSPACE;
2282                 auio.uio_td = td;
2283                 auio.uio_resid = count;
2284                 error = VOP_READLINK(vp, &auio, p->p_ucred);
2285         }
2286         vput(vp);
2287         *res = count - auio.uio_resid;
2288         return (error);
2289 }
2290
2291 /*
2292  * readlink_args(char *path, char *buf, int count)
2293  *
2294  * Return target name of a symbolic link.
2295  */
2296 int
2297 sys_readlink(struct readlink_args *uap)
2298 {
2299         struct nlookupdata nd;
2300         int error;
2301
2302         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2303         if (error == 0) {
2304                 error = kern_readlink(&nd, uap->buf, uap->count,
2305                                         &uap->sysmsg_result);
2306         }
2307         nlookup_done(&nd);
2308         return (error);
2309 }
2310
2311 static int
2312 setfflags(struct vnode *vp, int flags)
2313 {
2314         struct thread *td = curthread;
2315         struct proc *p = td->td_proc;
2316         int error;
2317         struct vattr vattr;
2318
2319         /*
2320          * Prevent non-root users from setting flags on devices.  When
2321          * a device is reused, users can retain ownership of the device
2322          * if they are allowed to set flags and programs assume that
2323          * chown can't fail when done as root.
2324          */
2325         if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2326             ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2327                 return (error);
2328
2329         /*
2330          * note: vget is required for any operation that might mod the vnode
2331          * so VINACTIVE is properly cleared.
2332          */
2333         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2334                 VATTR_NULL(&vattr);
2335                 vattr.va_flags = flags;
2336                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2337                 vput(vp);
2338         }
2339         return (error);
2340 }
2341
2342 /*
2343  * chflags(char *path, int flags)
2344  *
2345  * Change flags of a file given a path name.
2346  */
2347 /* ARGSUSED */
2348 int
2349 sys_chflags(struct chflags_args *uap)
2350 {
2351         struct nlookupdata nd;
2352         struct vnode *vp;
2353         int error;
2354
2355         vp = NULL;
2356         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2357         /* XXX Add NLC flag indicating modifying operation? */
2358         if (error == 0)
2359                 error = nlookup(&nd);
2360         if (error == 0)
2361                 error = ncp_writechk(&nd.nl_nch);
2362         if (error == 0)
2363                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2364         nlookup_done(&nd);
2365         if (error == 0) {
2366                 error = setfflags(vp, uap->flags);
2367                 vrele(vp);
2368         }
2369         return (error);
2370 }
2371
2372 /*
2373  * fchflags_args(int fd, int flags)
2374  *
2375  * Change flags of a file given a file descriptor.
2376  */
2377 /* ARGSUSED */
2378 int
2379 sys_fchflags(struct fchflags_args *uap)
2380 {
2381         struct thread *td = curthread;
2382         struct proc *p = td->td_proc;
2383         struct file *fp;
2384         int error;
2385
2386         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2387                 return (error);
2388         if (fp->f_nchandle.ncp)
2389                 error = ncp_writechk(&fp->f_nchandle);
2390         if (error == 0)
2391                 error = setfflags((struct vnode *) fp->f_data, uap->flags);
2392         fdrop(fp);
2393         return (error);
2394 }
2395
2396 static int
2397 setfmode(struct vnode *vp, int mode)
2398 {
2399         struct thread *td = curthread;
2400         struct proc *p = td->td_proc;
2401         int error;
2402         struct vattr vattr;
2403
2404         /*
2405          * note: vget is required for any operation that might mod the vnode
2406          * so VINACTIVE is properly cleared.
2407          */
2408         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2409                 VATTR_NULL(&vattr);
2410                 vattr.va_mode = mode & ALLPERMS;
2411                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2412                 vput(vp);
2413         }
2414         return error;
2415 }
2416
2417 int
2418 kern_chmod(struct nlookupdata *nd, int mode)
2419 {
2420         struct vnode *vp;
2421         int error;
2422
2423         /* XXX Add NLC flag indicating modifying operation? */
2424         if ((error = nlookup(nd)) != 0)
2425                 return (error);
2426         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2427                 return (error);
2428         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2429                 error = setfmode(vp, mode);
2430         vrele(vp);
2431         return (error);
2432 }
2433
2434 /*
2435  * chmod_args(char *path, int mode)
2436  *
2437  * Change mode of a file given path name.
2438  */
2439 /* ARGSUSED */
2440 int
2441 sys_chmod(struct chmod_args *uap)
2442 {
2443         struct nlookupdata nd;
2444         int error;
2445
2446         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2447         if (error == 0)
2448                 error = kern_chmod(&nd, uap->mode);
2449         nlookup_done(&nd);
2450         return (error);
2451 }
2452
2453 /*
2454  * lchmod_args(char *path, int mode)
2455  *
2456  * Change mode of a file given path name (don't follow links.)
2457  */
2458 /* ARGSUSED */
2459 int
2460 sys_lchmod(struct lchmod_args *uap)
2461 {
2462         struct nlookupdata nd;
2463         int error;
2464
2465         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2466         if (error == 0)
2467                 error = kern_chmod(&nd, uap->mode);
2468         nlookup_done(&nd);
2469         return (error);
2470 }
2471
2472 /*
2473  * fchmod_args(int fd, int mode)
2474  *
2475  * Change mode of a file given a file descriptor.
2476  */
2477 /* ARGSUSED */
2478 int
2479 sys_fchmod(struct fchmod_args *uap)
2480 {
2481         struct thread *td = curthread;
2482         struct proc *p = td->td_proc;
2483         struct file *fp;
2484         int error;
2485
2486         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2487                 return (error);
2488         if (fp->f_nchandle.ncp)
2489                 error = ncp_writechk(&fp->f_nchandle);
2490         if (error == 0)
2491                 error = setfmode((struct vnode *)fp->f_data, uap->mode);
2492         fdrop(fp);
2493         return (error);
2494 }
2495
2496 static int
2497 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2498 {
2499         struct thread *td = curthread;
2500         struct proc *p = td->td_proc;
2501         int error;
2502         struct vattr vattr;
2503
2504         /*
2505          * note: vget is required for any operation that might mod the vnode
2506          * so VINACTIVE is properly cleared.
2507          */
2508         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2509                 VATTR_NULL(&vattr);
2510                 vattr.va_uid = uid;
2511                 vattr.va_gid = gid;
2512                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2513                 vput(vp);
2514         }
2515         return error;
2516 }
2517
2518 int
2519 kern_chown(struct nlookupdata *nd, int uid, int gid)
2520 {
2521         struct vnode *vp;
2522         int error;
2523
2524         /* XXX Add NLC flag indicating modifying operation? */
2525         if ((error = nlookup(nd)) != 0)
2526                 return (error);
2527         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2528                 return (error);
2529         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2530                 error = setfown(vp, uid, gid);
2531         vrele(vp);
2532         return (error);
2533 }
2534
2535 /*
2536  * chown(char *path, int uid, int gid)
2537  *
2538  * Set ownership given a path name.
2539  */
2540 int
2541 sys_chown(struct chown_args *uap)
2542 {
2543         struct nlookupdata nd;
2544         int error;
2545
2546         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2547         if (error == 0)
2548                 error = kern_chown(&nd, uap->uid, uap->gid);
2549         nlookup_done(&nd);
2550         return (error);
2551 }
2552
2553 /*
2554  * lchown_args(char *path, int uid, int gid)
2555  *
2556  * Set ownership given a path name, do not cross symlinks.
2557  */
2558 int
2559 sys_lchown(struct lchown_args *uap)
2560 {
2561         struct nlookupdata nd;
2562         int error;
2563
2564         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2565         if (error == 0)
2566                 error = kern_chown(&nd, uap->uid, uap->gid);
2567         nlookup_done(&nd);
2568         return (error);
2569 }
2570
2571 /*
2572  * fchown_args(int fd, int uid, int gid)
2573  *
2574  * Set ownership given a file descriptor.
2575  */
2576 /* ARGSUSED */
2577 int
2578 sys_fchown(struct fchown_args *uap)
2579 {
2580         struct thread *td = curthread;
2581         struct proc *p = td->td_proc;
2582         struct file *fp;
2583         int error;
2584
2585         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2586                 return (error);
2587         if (fp->f_nchandle.ncp)
2588                 error = ncp_writechk(&fp->f_nchandle);
2589         if (error == 0)
2590                 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2591         fdrop(fp);
2592         return (error);
2593 }
2594
2595 static int
2596 getutimes(const struct timeval *tvp, struct timespec *tsp)
2597 {
2598         struct timeval tv[2];
2599
2600         if (tvp == NULL) {
2601                 microtime(&tv[0]);
2602                 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2603                 tsp[1] = tsp[0];
2604         } else {
2605                 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2606                 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2607         }
2608         return 0;
2609 }
2610
2611 static int
2612 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2613 {
2614         struct thread *td = curthread;
2615         struct proc *p = td->td_proc;
2616         int error;
2617         struct vattr vattr;
2618
2619         /*
2620          * note: vget is required for any operation that might mod the vnode
2621          * so VINACTIVE is properly cleared.
2622          */
2623         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2624                 VATTR_NULL(&vattr);
2625                 vattr.va_atime = ts[0];
2626                 vattr.va_mtime = ts[1];
2627                 if (nullflag)
2628                         vattr.va_vaflags |= VA_UTIMES_NULL;
2629                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2630                 vput(vp);
2631         }
2632         return error;
2633 }
2634
2635 int
2636 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2637 {
2638         struct timespec ts[2];
2639         struct vnode *vp;
2640         int error;
2641
2642         if ((error = getutimes(tptr, ts)) != 0)
2643                 return (error);
2644         /* XXX Add NLC flag indicating modifying operation? */
2645         if ((error = nlookup(nd)) != 0)
2646                 return (error);
2647         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2648                 return (error);
2649         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2650                 return (error);
2651         error = setutimes(vp, ts, tptr == NULL);
2652         vrele(vp);
2653         return (error);
2654 }
2655
2656 /*
2657  * utimes_args(char *path, struct timeval *tptr)
2658  *
2659  * Set the access and modification times of a file.
2660  */
2661 int
2662 sys_utimes(struct utimes_args *uap)
2663 {
2664         struct timeval tv[2];
2665         struct nlookupdata nd;
2666         int error;
2667
2668         if (uap->tptr) {
2669                 error = copyin(uap->tptr, tv, sizeof(tv));
2670                 if (error)
2671                         return (error);
2672         }
2673         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2674         if (error == 0)
2675                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2676         nlookup_done(&nd);
2677         return (error);
2678 }
2679
2680 /*
2681  * lutimes_args(char *path, struct timeval *tptr)
2682  *
2683  * Set the access and modification times of a file.
2684  */
2685 int
2686 sys_lutimes(struct lutimes_args *uap)
2687 {
2688         struct timeval tv[2];
2689         struct nlookupdata nd;
2690         int error;
2691
2692         if (uap->tptr) {
2693                 error = copyin(uap->tptr, tv, sizeof(tv));
2694                 if (error)
2695                         return (error);
2696         }
2697         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2698         if (error == 0)
2699                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2700         nlookup_done(&nd);
2701         return (error);
2702 }
2703
2704 int
2705 kern_futimes(int fd, struct timeval *tptr)
2706 {
2707         struct thread *td = curthread;
2708         struct proc *p = td->td_proc;
2709         struct timespec ts[2];
2710         struct file *fp;
2711         int error;
2712
2713         error = getutimes(tptr, ts);
2714         if (error)
2715                 return (error);
2716         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2717                 return (error);
2718         if (fp->f_nchandle.ncp)
2719                 error = ncp_writechk(&fp->f_nchandle);
2720         if (error == 0)
2721                 error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2722         fdrop(fp);
2723         return (error);
2724 }
2725
2726 /*
2727  * futimes_args(int fd, struct timeval *tptr)
2728  *
2729  * Set the access and modification times of a file.
2730  */
2731 int
2732 sys_futimes(struct futimes_args *uap)
2733 {
2734         struct timeval tv[2];
2735         int error;
2736
2737         if (uap->tptr) {
2738                 error = copyin(uap->tptr, tv, sizeof(tv));
2739                 if (error)
2740                         return (error);
2741         }
2742
2743         error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2744
2745         return (error);
2746 }
2747
2748 int
2749 kern_truncate(struct nlookupdata *nd, off_t length)
2750 {
2751         struct vnode *vp;
2752         struct vattr vattr;
2753         int error;
2754
2755         if (length < 0)
2756                 return(EINVAL);
2757         /* XXX Add NLC flag indicating modifying operation? */
2758         if ((error = nlookup(nd)) != 0)
2759                 return (error);
2760         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2761                 return (error);
2762         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2763                 return (error);
2764         if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2765                 vrele(vp);
2766                 return (error);
2767         }
2768         if (vp->v_type == VDIR) {
2769                 error = EISDIR;
2770         } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
2771             (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2772                 VATTR_NULL(&vattr);
2773                 vattr.va_size = length;
2774                 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2775         }
2776         vput(vp);
2777         return (error);
2778 }
2779
2780 /*
2781  * truncate(char *path, int pad, off_t length)
2782  *
2783  * Truncate a file given its path name.
2784  */
2785 int
2786 sys_truncate(struct truncate_args *uap)
2787 {
2788         struct nlookupdata nd;
2789         int error;
2790
2791         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2792         if (error == 0)
2793                 error = kern_truncate(&nd, uap->length);
2794         nlookup_done(&nd);
2795         return error;
2796 }
2797
2798 int
2799 kern_ftruncate(int fd, off_t length)
2800 {
2801         struct thread *td = curthread;
2802         struct proc *p = td->td_proc;
2803         struct vattr vattr;
2804         struct vnode *vp;
2805         struct file *fp;
2806         int error;
2807
2808         if (length < 0)
2809                 return(EINVAL);
2810         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2811                 return (error);
2812         if (fp->f_nchandle.ncp) {
2813                 error = ncp_writechk(&fp->f_nchandle);
2814                 if (error)
2815                         goto done;
2816         }
2817         if ((fp->f_flag & FWRITE) == 0) {
2818                 error = EINVAL;
2819                 goto done;
2820         }
2821         vp = (struct vnode *)fp->f_data;
2822         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2823         if (vp->v_type == VDIR) {
2824                 error = EISDIR;
2825         } else if ((error = vn_writechk(vp, NULL)) == 0) {
2826                 VATTR_NULL(&vattr);
2827                 vattr.va_size = length;
2828                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2829         }
2830         vn_unlock(vp);
2831 done:
2832         fdrop(fp);
2833         return (error);
2834 }
2835
2836 /*
2837  * ftruncate_args(int fd, int pad, off_t length)
2838  *
2839  * Truncate a file given a file descriptor.
2840  */
2841 int
2842 sys_ftruncate(struct ftruncate_args *uap)
2843 {
2844         int error;
2845
2846         error = kern_ftruncate(uap->fd, uap->length);
2847
2848         return (error);
2849 }
2850
2851 /*
2852  * fsync(int fd)
2853  *
2854  * Sync an open file.
2855  */
2856 /* ARGSUSED */
2857 int
2858 sys_fsync(struct fsync_args *uap)
2859 {
2860         struct thread *td = curthread;
2861         struct proc *p = td->td_proc;
2862         struct vnode *vp;
2863         struct file *fp;
2864         vm_object_t obj;
2865         int error;
2866
2867         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2868                 return (error);
2869         vp = (struct vnode *)fp->f_data;
2870         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2871         if ((obj = vp->v_object) != NULL)
2872                 vm_object_page_clean(obj, 0, 0, 0);
2873         if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 &&
2874             vp->v_mount && (vp->v_mount->mnt_flag & MNT_SOFTDEP) &&
2875             bioops.io_fsync) {
2876                 error = (*bioops.io_fsync)(vp);
2877         }
2878         vn_unlock(vp);
2879         fdrop(fp);
2880         return (error);
2881 }
2882
2883 int
2884 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2885 {
2886         struct nchandle fnchd;
2887         struct nchandle tnchd;
2888         struct namecache *ncp;
2889         struct vnode *fdvp;
2890         struct vnode *tdvp;
2891         struct mount *mp;
2892         int error;
2893
2894         bwillwrite();
2895         if ((error = nlookup(fromnd)) != 0)
2896                 return (error);
2897         if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
2898                 return (ENOENT);
2899         fnchd.mount = fromnd->nl_nch.mount;
2900         cache_hold(&fnchd);
2901
2902         /*
2903          * unlock the source nch so we can lookup the target nch without
2904          * deadlocking.  The target may or may not exist so we do not check
2905          * for a target vp like kern_mkdir() and other creation functions do.
2906          *
2907          * The source and target directories are ref'd and rechecked after
2908          * everything is relocked to determine if the source or target file
2909          * has been renamed.
2910          */
2911         KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2912         fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2913         cache_unlock(&fromnd->nl_nch);
2914
2915         tond->nl_flags |= NLC_CREATE;
2916         if ((error = nlookup(tond)) != 0) {
2917                 cache_drop(&fnchd);
2918                 return (error);
2919         }
2920         if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
2921                 cache_drop(&fnchd);
2922                 return (ENOENT);
2923         }
2924         tnchd.mount = tond->nl_nch.mount;
2925         cache_hold(&tnchd);
2926
2927         /*
2928          * If the source and target are the same there is nothing to do
2929          */
2930         if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
2931                 cache_drop(&fnchd);
2932                 cache_drop(&tnchd);
2933                 return (0);
2934         }
2935
2936         /*
2937          * Mount points cannot be renamed or overwritten
2938          */
2939         if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
2940             NCF_ISMOUNTPT
2941         ) {
2942                 cache_drop(&fnchd);
2943                 cache_drop(&tnchd);
2944                 return (EINVAL);
2945         }
2946
2947         /*
2948          * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2949          * may have become invalid while it was unlocked, nc_vp and nc_mount
2950          * could be NULL.
2951          */
2952         if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
2953                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2954         } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
2955                 cache_lock(&fromnd->nl_nch);
2956                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2957         } else {
2958                 cache_unlock(&tond->nl_nch);
2959                 cache_lock(&fromnd->nl_nch);
2960                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2961                 cache_lock(&tond->nl_nch);
2962                 cache_resolve(&tond->nl_nch, tond->nl_cred);
2963         }
2964         fromnd->nl_flags |= NLC_NCPISLOCKED;
2965
2966         /*
2967          * make sure the parent directories linkages are the same
2968          */
2969         if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
2970             tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
2971                 cache_drop(&fnchd);
2972                 cache_drop(&tnchd);
2973                 return (ENOENT);
2974         }
2975
2976         /*
2977          * Both the source and target must be within the same filesystem and
2978          * in the same filesystem as their parent directories within the
2979          * namecache topology.
2980          *
2981          * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2982          */
2983         mp = fnchd.mount;
2984         if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
2985             mp != tond->nl_nch.mount) {
2986                 cache_drop(&fnchd);
2987                 cache_drop(&tnchd);
2988                 return (EXDEV);
2989         }
2990
2991         /*
2992          * Make sure the mount point is writable
2993          */
2994         if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
2995                 cache_drop(&fnchd);
2996                 cache_drop(&tnchd);
2997                 return (error);
2998         }
2999
3000         /*
3001          * If the target exists and either the source or target is a directory,
3002          * then both must be directories.
3003          *
3004          * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3005          * have become NULL.
3006          */
3007         if (tond->nl_nch.ncp->nc_vp) {
3008                 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3009                         error = ENOENT;
3010                 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3011                         if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3012                                 error = ENOTDIR;
3013                 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3014                         error = EISDIR;
3015                 }
3016         }
3017
3018         /*
3019          * You cannot rename a source into itself or a subdirectory of itself.
3020          * We check this by travsersing the target directory upwards looking
3021          * for a match against the source.
3022          */
3023         if (error == 0) {
3024                 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3025                         if (fromnd->nl_nch.ncp == ncp) {
3026                                 error = EINVAL;
3027                                 break;
3028                         }
3029                 }
3030         }
3031
3032         cache_drop(&fnchd);
3033         cache_drop(&tnchd);
3034
3035         /*
3036          * Even though the namespaces are different, they may still represent
3037          * hardlinks to the same file.  The filesystem might have a hard time
3038          * with this so we issue a NREMOVE of the source instead of a NRENAME
3039          * when we detect the situation.
3040          */
3041         if (error == 0) {
3042                 fdvp = fromnd->nl_nch.ncp->nc_parent->nc_vp;
3043                 tdvp = tond->nl_nch.ncp->nc_parent->nc_vp;
3044                 if (fdvp == NULL || tdvp == NULL) {
3045                         error = EPERM;
3046                 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3047                         /* vhold(fdvp); - dvp can't go away */
3048                         error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3049                                             fromnd->nl_cred);
3050                         /* vdrop(fdvp); */
3051                 } else {
3052                         /* vhold(fdvp); - dvp can't go away */
3053                         /* vhold(tdvp); - dvp can't go away */
3054                         error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3055                                             fdvp, tdvp, tond->nl_cred);
3056                         /* vdrop(fdvp); */
3057                         /* vdrop(tdvp); */
3058                 }
3059         }
3060         return (error);
3061 }
3062
3063 /*
3064  * rename_args(char *from, char *to)
3065  *
3066  * Rename files.  Source and destination must either both be directories,
3067  * or both not be directories.  If target is a directory, it must be empty.
3068  */
3069 int
3070 sys_rename(struct rename_args *uap)
3071 {
3072         struct nlookupdata fromnd, tond;
3073         int error;
3074
3075         error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3076         if (error == 0) {
3077                 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3078                 if (error == 0)
3079                         error = kern_rename(&fromnd, &tond);
3080                 nlookup_done(&tond);
3081         }
3082         nlookup_done(&fromnd);
3083         return (error);
3084 }
3085
3086 int
3087 kern_mkdir(struct nlookupdata *nd, int mode)
3088 {
3089         struct thread *td = curthread;
3090         struct proc *p = td->td_proc;
3091         struct vnode *vp;
3092         struct vnode *dvp;
3093         struct vattr vattr;
3094         int error;
3095
3096         bwillwrite();
3097         nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
3098         if ((error = nlookup(nd)) != 0)
3099                 return (error);
3100
3101         if (nd->nl_nch.ncp->nc_vp)
3102                 return (EEXIST);
3103         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3104                 return (error);
3105         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
3106                 return (EPERM);
3107         /* vhold(dvp); - dvp can't go away */
3108         VATTR_NULL(&vattr);
3109         vattr.va_type = VDIR;
3110         vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3111
3112         vp = NULL;
3113         error = VOP_NMKDIR(&nd->nl_nch, dvp, &vp, p->p_ucred, &vattr);
3114         /* vdrop(dvp); */
3115         if (error == 0)
3116                 vput(vp);
3117         return (error);
3118 }
3119
3120 /*
3121  * mkdir_args(char *path, int mode)
3122  *
3123  * Make a directory file.
3124  */
3125 /* ARGSUSED */
3126 int
3127 sys_mkdir(struct mkdir_args *uap)
3128 {
3129         struct nlookupdata nd;
3130         int error;
3131
3132         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3133         if (error == 0)
3134                 error = kern_mkdir(&nd, uap->mode);
3135         nlookup_done(&nd);
3136         return (error);
3137 }
3138
3139 int
3140 kern_rmdir(struct nlookupdata *nd)
3141 {
3142         struct vnode *dvp;
3143         int error;
3144
3145         bwillwrite();
3146         nd->nl_flags |= NLC_DELETE;
3147         if ((error = nlookup(nd)) != 0)
3148                 return (error);
3149
3150         /*
3151          * Do not allow directories representing mount points to be
3152          * deleted, even if empty.  Check write perms on mount point
3153          * in case the vnode is aliased (aka nullfs).
3154          */
3155         if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3156                 return (EINVAL);
3157         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3158                 return (error);
3159         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
3160                 return (EPERM);
3161         /* vhold(dvp); - dvp can't go away */
3162         error = VOP_NRMDIR(&nd->nl_nch, dvp, nd->nl_cred);
3163         /* vdrop(dvp); */
3164         return (error);
3165 }
3166
3167 /*
3168  * rmdir_args(char *path)
3169  *
3170  * Remove a directory file.
3171  */
3172 /* ARGSUSED */
3173 int
3174 sys_rmdir(struct rmdir_args *uap)
3175 {
3176         struct nlookupdata nd;
3177         int error;
3178
3179         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3180         if (error == 0)
3181                 error = kern_rmdir(&nd);
3182         nlookup_done(&nd);
3183         return (error);
3184 }
3185
3186 int
3187 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3188     enum uio_seg direction)
3189 {
3190         struct thread *td = curthread;
3191         struct proc *p = td->td_proc;
3192         struct vnode *vp;
3193         struct file *fp;
3194         struct uio auio;
3195         struct iovec aiov;
3196         long loff;
3197         int error, eofflag;
3198
3199         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3200                 return (error);
3201         if ((fp->f_flag & FREAD) == 0) {
3202                 error = EBADF;
3203                 goto done;
3204         }
3205         vp = (struct vnode *)fp->f_data;
3206 unionread:
3207         if (vp->v_type != VDIR) {
3208                 error = EINVAL;
3209                 goto done;
3210         }
3211         aiov.iov_base = buf;
3212         aiov.iov_len = count;
3213         auio.uio_iov = &aiov;
3214         auio.uio_iovcnt = 1;
3215         auio.uio_rw = UIO_READ;
3216         auio.uio_segflg = direction;
3217         auio.uio_td = td;
3218         auio.uio_resid = count;
3219         loff = auio.uio_offset = fp->f_offset;
3220         error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3221         fp->f_offset = auio.uio_offset;
3222         if (error)
3223                 goto done;
3224         if (count == auio.uio_resid) {
3225                 if (union_dircheckp) {
3226                         error = union_dircheckp(td, &vp, fp);
3227                         if (error == -1)
3228                                 goto unionread;
3229                         if (error)
3230                                 goto done;
3231                 }
3232 #if 0
3233                 if ((vp->v_flag & VROOT) &&
3234                     (vp->v_mount->mnt_flag & MNT_UNION)) {
3235                         struct vnode *tvp = vp;
3236                         vp = vp->v_mount->mnt_vnodecovered;
3237                         vref(vp);
3238                         fp->f_data = vp;
3239                         fp->f_offset = 0;
3240                         vrele(tvp);
3241                         goto unionread;
3242                 }
3243 #endif
3244         }
3245         if (basep) {
3246                 *basep = loff;
3247         }
3248         *res = count - auio.uio_resid;
3249 done:
3250         fdrop(fp);
3251         return (error);
3252 }
3253
3254 /*
3255  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3256  *
3257  * Read a block of directory entries in a file system independent format.
3258  */
3259 int
3260 sys_getdirentries(struct getdirentries_args *uap)
3261 {
3262         long base;
3263         int error;
3264
3265         error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3266             &uap->sysmsg_result, UIO_USERSPACE);
3267
3268         if (error == 0)
3269                 error = copyout(&base, uap->basep, sizeof(*uap->basep));
3270         return (error);
3271 }
3272
3273 /*
3274  * getdents_args(int fd, char *buf, size_t count)
3275  */
3276 int
3277 sys_getdents(struct getdents_args *uap)
3278 {
3279         int error;
3280
3281         error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3282             &uap->sysmsg_result, UIO_USERSPACE);
3283
3284         return (error);
3285 }
3286
3287 /*
3288  * umask(int newmask)
3289  *
3290  * Set the mode mask for creation of filesystem nodes.
3291  *
3292  * MP SAFE
3293  */
3294 int
3295 sys_umask(struct umask_args *uap)
3296 {
3297         struct thread *td = curthread;
3298         struct proc *p = td->td_proc;
3299         struct filedesc *fdp;
3300
3301         fdp = p->p_fd;
3302         uap->sysmsg_result = fdp->fd_cmask;
3303         fdp->fd_cmask = uap->newmask & ALLPERMS;
3304         return (0);
3305 }
3306
3307 /*
3308  * revoke(char *path)
3309  *
3310  * Void all references to file by ripping underlying filesystem
3311  * away from vnode.
3312  */
3313 /* ARGSUSED */
3314 int
3315 sys_revoke(struct revoke_args *uap)
3316 {
3317         struct nlookupdata nd;
3318         struct vattr vattr;
3319         struct vnode *vp;
3320         struct ucred *cred;
3321         int error;
3322
3323         vp = NULL;
3324         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3325         if (error == 0)
3326                 error = nlookup(&nd);
3327         if (error == 0)
3328                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3329         cred = crhold(nd.nl_cred);
3330         nlookup_done(&nd);
3331         if (error == 0) {
3332                 if (vp->v_type != VCHR && vp->v_type != VBLK)
3333                         error = EINVAL;
3334                 if (error == 0)
3335                         error = VOP_GETATTR(vp, &vattr);
3336                 if (error == 0 && cred->cr_uid != vattr.va_uid)
3337                         error = suser_cred(cred, PRISON_ROOT);
3338                 if (error == 0 && count_udev(vp->v_umajor, vp->v_uminor) > 0) {
3339                         error = 0;
3340                         vx_lock(vp);
3341                         VOP_REVOKE(vp, REVOKEALL);
3342                         vx_unlock(vp);
3343                 }
3344                 vrele(vp);
3345         }
3346         if (cred)
3347                 crfree(cred);
3348         return (error);
3349 }
3350
3351 /*
3352  * getfh_args(char *fname, fhandle_t *fhp)
3353  *
3354  * Get (NFS) file handle
3355  */
3356 int
3357 sys_getfh(struct getfh_args *uap)
3358 {
3359         struct thread *td = curthread;
3360         struct nlookupdata nd;
3361         fhandle_t fh;
3362         struct vnode *vp;
3363         int error;
3364
3365         /*
3366          * Must be super user
3367          */
3368         if ((error = suser(td)) != 0)
3369                 return (error);
3370
3371         vp = NULL;
3372         error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3373         if (error == 0)
3374                 error = nlookup(&nd);
3375         if (error == 0)
3376                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3377         nlookup_done(&nd);
3378         if (error == 0) {
3379                 bzero(&fh, sizeof(fh));
3380                 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3381                 error = VFS_VPTOFH(vp, &fh.fh_fid);
3382                 vput(vp);
3383                 if (error == 0)
3384                         error = copyout(&fh, uap->fhp, sizeof(fh));
3385         }
3386         return (error);
3387 }
3388
3389 /*
3390  * fhopen_args(const struct fhandle *u_fhp, int flags)
3391  *
3392  * syscall for the rpc.lockd to use to translate a NFS file handle into
3393  * an open descriptor.
3394  *
3395  * warning: do not remove the suser() call or this becomes one giant
3396  * security hole.
3397  */
3398 int
3399 sys_fhopen(struct fhopen_args *uap)
3400 {
3401         struct thread *td = curthread;
3402         struct proc *p = td->td_proc;
3403         struct mount *mp;
3404         struct vnode *vp;
3405         struct fhandle fhp;
3406         struct vattr vat;
3407         struct vattr *vap = &vat;
3408         struct flock lf;
3409         int fmode, mode, error, type;
3410         struct file *nfp;
3411         struct file *fp;
3412         int indx;
3413
3414         /*
3415          * Must be super user
3416          */
3417         error = suser(td);
3418         if (error)
3419                 return (error);
3420
3421         fmode = FFLAGS(uap->flags);
3422         /* why not allow a non-read/write open for our lockd? */
3423         if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3424                 return (EINVAL);
3425         error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3426         if (error)
3427                 return(error);
3428         /* find the mount point */
3429         mp = vfs_getvfs(&fhp.fh_fsid);
3430         if (mp == NULL)
3431                 return (ESTALE);
3432         /* now give me my vnode, it gets returned to me locked */
3433         error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3434         if (error)
3435                 return (error);
3436         /*
3437          * from now on we have to make sure not
3438          * to forget about the vnode
3439          * any error that causes an abort must vput(vp)
3440          * just set error = err and 'goto bad;'.
3441          */
3442
3443         /*
3444          * from vn_open
3445          */
3446         if (vp->v_type == VLNK) {
3447                 error = EMLINK;
3448                 goto bad;
3449         }
3450         if (vp->v_type == VSOCK) {
3451                 error = EOPNOTSUPP;
3452                 goto bad;
3453         }
3454         mode = 0;
3455         if (fmode & (FWRITE | O_TRUNC)) {
3456                 if (vp->v_type == VDIR) {
3457                         error = EISDIR;
3458                         goto bad;
3459                 }
3460                 error = vn_writechk(vp, NULL);
3461                 if (error)
3462                         goto bad;
3463                 mode |= VWRITE;
3464         }
3465         if (fmode & FREAD)
3466                 mode |= VREAD;
3467         if (mode) {
3468                 error = VOP_ACCESS(vp, mode, p->p_ucred);
3469                 if (error)
3470                         goto bad;
3471         }
3472         if (fmode & O_TRUNC) {
3473                 vn_unlock(vp);                          /* XXX */
3474                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
3475                 VATTR_NULL(vap);
3476                 vap->va_size = 0;
3477                 error = VOP_SETATTR(vp, vap, p->p_ucred);
3478                 if (error)
3479                         goto bad;
3480         }
3481
3482         /*
3483          * VOP_OPEN needs the file pointer so it can potentially override
3484          * it.
3485          *
3486          * WARNING! no f_nchandle will be associated when fhopen()ing a
3487          * directory.  XXX
3488          */
3489         if ((error = falloc(p, &nfp, &indx)) != 0)
3490                 goto bad;
3491         fp = nfp;
3492
3493         error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3494         if (error) {
3495                 /*
3496                  * setting f_ops this way prevents VOP_CLOSE from being
3497                  * called or fdrop() releasing the vp from v_data.   Since
3498                  * the VOP_OPEN failed we don't want to VOP_CLOSE.
3499                  */
3500                 fp->f_ops = &badfileops;
3501                 fp->f_data = NULL;
3502                 goto bad_drop;
3503         }
3504
3505         /*
3506          * The fp is given its own reference, we still have our ref and lock.
3507          *
3508          * Assert that all regular files must be created with a VM object.
3509          */
3510         if (vp->v_type == VREG && vp->v_object == NULL) {
3511                 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3512                 goto bad_drop;
3513         }
3514
3515         /*
3516          * The open was successful.  Handle any locking requirements.
3517          */
3518         if (fmode & (O_EXLOCK | O_SHLOCK)) {
3519                 lf.l_whence = SEEK_SET;
3520                 lf.l_start = 0;
3521                 lf.l_len = 0;
3522                 if (fmode & O_EXLOCK)
3523                         lf.l_type = F_WRLCK;
3524                 else
3525                         lf.l_type = F_RDLCK;
3526                 if (fmode & FNONBLOCK)
3527                         type = 0;
3528                 else
3529                         type = F_WAIT;
3530                 vn_unlock(vp);
3531                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3532                         /*
3533                          * release our private reference.
3534                          */
3535                         fsetfd(p, NULL, indx);
3536                         fdrop(fp);
3537                         vrele(vp);
3538                         return (error);
3539                 }
3540                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3541                 fp->f_flag |= FHASLOCK;
3542         }
3543
3544         /*
3545          * Clean up.  Associate the file pointer with the previously
3546          * reserved descriptor and return it.
3547          */
3548         vput(vp);
3549         fsetfd(p, fp, indx);
3550         fdrop(fp);
3551         uap->sysmsg_result = indx;
3552         return (0);
3553
3554 bad_drop:
3555         fsetfd(p, NULL, indx);
3556         fdrop(fp);
3557 bad:
3558         vput(vp);
3559         return (error);
3560 }
3561
3562 /*
3563  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3564  */
3565 int
3566 sys_fhstat(struct fhstat_args *uap)
3567 {
3568         struct thread *td = curthread;
3569         struct stat sb;
3570         fhandle_t fh;
3571         struct mount *mp;
3572         struct vnode *vp;
3573         int error;
3574
3575         /*
3576          * Must be super user
3577          */
3578         error = suser(td);
3579         if (error)
3580                 return (error);
3581
3582         error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3583         if (error)
3584                 return (error);
3585
3586         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3587                 return (ESTALE);
3588         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3589                 return (error);
3590         error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3591         vput(vp);
3592         if (error)
3593                 return (error);
3594         error = copyout(&sb, uap->sb, sizeof(sb));
3595         return (error);
3596 }
3597
3598 /*
3599  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3600  */
3601 int
3602 sys_fhstatfs(struct fhstatfs_args *uap)
3603 {
3604         struct thread *td = curthread;
3605         struct proc *p = td->td_proc;
3606         struct statfs *sp;
3607         struct mount *mp;
3608         struct vnode *vp;
3609         struct statfs sb;
3610         char *fullpath, *freepath;
3611         fhandle_t fh;
3612         int error;
3613
3614         /*
3615          * Must be super user
3616          */
3617         if ((error = suser(td)))
3618                 return (error);
3619
3620         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3621                 return (error);
3622
3623         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3624                 return (ESTALE);
3625
3626         if (p != NULL && !chroot_visible_mnt(mp, p))
3627                 return (ESTALE);
3628
3629         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3630                 return (error);
3631         mp = vp->v_mount;
3632         sp = &mp->mnt_stat;
3633         vput(vp);
3634         if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3635                 return (error);
3636
3637         error = mount_path(p, mp, &fullpath, &freepath);
3638         if (error)
3639                 return(error);
3640         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3641         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3642         kfree(freepath, M_TEMP);
3643
3644         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3645         if (suser(td)) {
3646                 bcopy(sp, &sb, sizeof(sb));
3647                 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3648                 sp = &sb;
3649         }
3650         return (copyout(sp, uap->buf, sizeof(*sp)));
3651 }
3652
3653 /*
3654  * Syscall to push extended attribute configuration information into the
3655  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3656  * a command (int cmd), and attribute name and misc data.  For now, the
3657  * attribute name is left in userspace for consumption by the VFS_op.
3658  * It will probably be changed to be copied into sysspace by the
3659  * syscall in the future, once issues with various consumers of the
3660  * attribute code have raised their hands.
3661  *
3662  * Currently this is used only by UFS Extended Attributes.
3663  */
3664 int
3665 sys_extattrctl(struct extattrctl_args *uap)
3666 {
3667         struct nlookupdata nd;
3668         struct mount *mp;
3669         struct vnode *vp;
3670         int error;
3671
3672         vp = NULL;
3673         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3674         if (error == 0)
3675                 error = nlookup(&nd);
3676         if (error == 0) {
3677                 mp = nd.nl_nch.mount;
3678                 error = VFS_EXTATTRCTL(mp, uap->cmd,
3679                                 uap->attrname, uap->arg,
3680                                 nd.nl_cred);
3681         }
3682         nlookup_done(&nd);
3683         return (error);
3684 }
3685
3686 /*
3687  * Syscall to set a named extended attribute on a file or directory.
3688  * Accepts attribute name, and a uio structure pointing to the data to set.
3689  * The uio is consumed in the style of writev().  The real work happens
3690  * in VOP_SETEXTATTR().
3691  */
3692 int
3693 sys_extattr_set_file(struct extattr_set_file_args *uap)
3694 {
3695         char attrname[EXTATTR_MAXNAMELEN];
3696         struct iovec aiov[UIO_SMALLIOV];
3697         struct iovec *needfree;
3698         struct nlookupdata nd;
3699         struct iovec *iov;
3700         struct vnode *vp;
3701         struct uio auio;
3702         u_int iovlen;
3703         u_int cnt;
3704         int error;
3705         int i;
3706
3707         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3708         if (error)
3709                 return (error);
3710
3711         vp = NULL;
3712         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3713         if (error == 0)
3714                 error = nlookup(&nd);
3715         if (error == 0)
3716                 error = ncp_writechk(&nd.nl_nch);
3717         if (error == 0)
3718                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3719         if (error) {
3720                 nlookup_done(&nd);
3721                 return (error);
3722         }
3723
3724         needfree = NULL;
3725         iovlen = uap->iovcnt * sizeof(struct iovec);
3726         if (uap->iovcnt > UIO_SMALLIOV) {
3727                 if (uap->iovcnt > UIO_MAXIOV) {
3728                         error = EINVAL;
3729                         goto done;
3730                 }
3731                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3732                 needfree = iov;
3733         } else {
3734                 iov = aiov;
3735         }
3736         auio.uio_iov = iov;
3737         auio.uio_iovcnt = uap->iovcnt;
3738         auio.uio_rw = UIO_WRITE;
3739         auio.uio_segflg = UIO_USERSPACE;
3740         auio.uio_td = nd.nl_td;
3741         auio.uio_offset = 0;
3742         if ((error = copyin(uap->iovp, iov, iovlen)))
3743                 goto done;
3744         auio.uio_resid = 0;
3745         for (i = 0; i < uap->iovcnt; i++) {
3746                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3747                         error = EINVAL;
3748                         goto done;
3749                 }
3750                 auio.uio_resid += iov->iov_len;
3751                 iov++;
3752         }
3753         cnt = auio.uio_resid;
3754         error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3755         cnt -= auio.uio_resid;
3756         uap->sysmsg_result = cnt;
3757 done:
3758         vput(vp);
3759         nlookup_done(&nd);
3760         if (needfree)
3761                 FREE(needfree, M_IOV);
3762         return (error);
3763 }
3764
3765 /*
3766  * Syscall to get a named extended attribute on a file or directory.
3767  * Accepts attribute name, and a uio structure pointing to a buffer for the
3768  * data.  The uio is consumed in the style of readv().  The real work
3769  * happens in VOP_GETEXTATTR();
3770  */
3771 int
3772 sys_extattr_get_file(struct extattr_get_file_args *uap)
3773 {
3774         char attrname[EXTATTR_MAXNAMELEN];
3775         struct iovec aiov[UIO_SMALLIOV];
3776         struct iovec *needfree;
3777         struct nlookupdata nd;
3778         struct iovec *iov;
3779         struct vnode *vp;
3780         struct uio auio;
3781         u_int iovlen;
3782         u_int cnt;
3783         int error;
3784         int i;
3785
3786         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3787         if (error)
3788                 return (error);
3789
3790         vp = NULL;
3791         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3792         if (error == 0)
3793                 error = nlookup(&nd);
3794         if (error == 0)
3795                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3796         if (error) {
3797                 nlookup_done(&nd);
3798                 return (error);
3799         }
3800
3801         iovlen = uap->iovcnt * sizeof (struct iovec);
3802         needfree = NULL;
3803         if (uap->iovcnt > UIO_SMALLIOV) {
3804                 if (uap->iovcnt > UIO_MAXIOV) {
3805                         error = EINVAL;
3806                         goto done;
3807                 }
3808                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3809                 needfree = iov;
3810         } else {
3811                 iov = aiov;
3812         }
3813         auio.uio_iov = iov;
3814         auio.uio_iovcnt = uap->iovcnt;
3815         auio.uio_rw = UIO_READ;
3816         auio.uio_segflg = UIO_USERSPACE;
3817         auio.uio_td = nd.nl_td;
3818         auio.uio_offset = 0;
3819         if ((error = copyin(uap->iovp, iov, iovlen)))
3820                 goto done;
3821         auio.uio_resid = 0;
3822         for (i = 0; i < uap->iovcnt; i++) {
3823                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3824                         error = EINVAL;
3825                         goto done;
3826                 }
3827                 auio.uio_resid += iov->iov_len;
3828                 iov++;
3829         }
3830         cnt = auio.uio_resid;
3831         error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3832         cnt -= auio.uio_resid;
3833         uap->sysmsg_result = cnt;
3834 done:
3835         vput(vp);
3836         nlookup_done(&nd);
3837         if (needfree)
3838                 FREE(needfree, M_IOV);
3839         return(error);
3840 }
3841
3842 /*
3843  * Syscall to delete a named extended attribute from a file or directory.
3844  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3845  */
3846 int
3847 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3848 {
3849         char attrname[EXTATTR_MAXNAMELEN];
3850         struct nlookupdata nd;
3851         struct vnode *vp;
3852         int error;
3853
3854         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3855         if (error)
3856                 return(error);
3857
3858         vp = NULL;
3859         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3860         if (error == 0)
3861                 error = nlookup(&nd);
3862         if (error == 0)
3863                 error = ncp_writechk(&nd.nl_nch);
3864         if (error == 0)
3865                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3866         if (error) {
3867                 nlookup_done(&nd);
3868                 return (error);
3869         }
3870
3871         error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3872         vput(vp);
3873         nlookup_done(&nd);
3874         return(error);
3875 }
3876
3877 /*
3878  * Determine if the mount is visible to the process.
3879  */
3880 static int
3881 chroot_visible_mnt(struct mount *mp, struct proc *p)
3882 {
3883         struct nchandle nch;
3884
3885         /*
3886          * Traverse from the mount point upwards.  If we hit the process
3887          * root then the mount point is visible to the process.
3888          */
3889         nch = mp->mnt_ncmountpt;
3890         while (nch.ncp) {
3891                 if (nch.mount == p->p_fd->fd_nrdir.mount &&
3892                     nch.ncp == p->p_fd->fd_nrdir.ncp) {
3893                         return(1);
3894                 }
3895                 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
3896                         nch = nch.mount->mnt_ncmounton;
3897                 } else {
3898                         nch.ncp = nch.ncp->nc_parent;
3899                 }
3900         }
3901
3902         /*
3903          * If the mount point is not visible to the process, but the
3904          * process root is in a subdirectory of the mount, return
3905          * TRUE anyway.
3906          */
3907         if (p->p_fd->fd_nrdir.mount == mp)
3908                 return(1);
3909
3910         return(0);
3911 }
3912