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 1.117 2007/06/26 20:39:33 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 ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
 252                 cache_drop(&nch);
 253                 vput(vp);
 254                 return (error);
 255         }
 256         for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
 257                 if (!strcmp(vfsp->vfc_name, fstypename))
 258                         break;
 259         }
 260         if (vfsp == NULL) {
 261                 linker_file_t lf;
 262
 263                 /* Only load modules for root (very important!) */
 264                 if ((error = suser(td)) != 0) {
 265                         cache_drop(&nch);
 266                         vput(vp);
 267                         return error;
 268                 }
 269                 error = linker_load_file(fstypename, &lf);
 270                 if (error || lf == NULL) {
 271                         cache_drop(&nch);
 272                         vput(vp);
 273                         if (lf == NULL)
 274                                 error = ENODEV;
 275                         return error;
 276                 }
 277                 lf->userrefs++;
 278                 /* lookup again, see if the VFS was loaded */
 279                 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
 280                         if (!strcmp(vfsp->vfc_name, fstypename))
 281                                 break;
 282                 }
 283                 if (vfsp == NULL) {
 284                         lf->userrefs--;
 285                         linker_file_unload(lf);
 286                         cache_drop(&nch);
 287                         vput(vp);
 288                         return (ENODEV);
 289                 }
 290         }
 291         if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
 292                 cache_drop(&nch);
 293                 vput(vp);
 294                 return (EBUSY);
 295         }
 296         vp->v_flag |= VMOUNT;
 297
 298         /*
 299          * Allocate and initialize the filesystem.
 300          */
 301         mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
 302         TAILQ_INIT(&mp->mnt_nvnodelist);
 303         TAILQ_INIT(&mp->mnt_reservedvnlist);
 304         TAILQ_INIT(&mp->mnt_jlist);
 305         mp->mnt_nvnodelistsize = 0;
 306         lockinit(&mp->mnt_lock, "vfslock", 0, 0);
 307         vfs_busy(mp, LK_NOWAIT);
 308         mp->mnt_op = vfsp->vfc_vfsops;
 309         mp->mnt_vfc = vfsp;
 310         vfsp->vfc_refcount++;
 311         mp->mnt_stat.f_type = vfsp->vfc_typenum;
 312         mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
 313         strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
 314         mp->mnt_stat.f_owner = cred->cr_uid;
 315         mp->mnt_iosize_max = DFLTPHYS;
 316         vn_unlock(vp);
 317 update:
 318         /*
 319          * Set the mount level flags.
 320          */
 321         if (uap->flags & MNT_RDONLY)
 322                 mp->mnt_flag |= MNT_RDONLY;
 323         else if (mp->mnt_flag & MNT_RDONLY)
 324                 mp->mnt_kern_flag |= MNTK_WANTRDWR;
 325         mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
 326             MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
 327             MNT_NOSYMFOLLOW | MNT_IGNORE |
 328             MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
 329         mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
 330             MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
 331             MNT_NOSYMFOLLOW | MNT_IGNORE |
 332             MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
 333         /*
 334          * Mount the filesystem.
 335          * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
 336          * get.
 337          */
 338         error = VFS_MOUNT(mp, uap->path, uap->data, cred);
 339         if (mp->mnt_flag & MNT_UPDATE) {
 340                 if (mp->mnt_kern_flag & MNTK_WANTRDWR)
 341                         mp->mnt_flag &= ~MNT_RDONLY;
 342                 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
 343                 mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
 344                 if (error) {
 345                         mp->mnt_flag = flag;
 346                         mp->mnt_kern_flag = flag2;
 347                 }
 348                 vfs_unbusy(mp);
 349                 vp->v_flag &= ~VMOUNT;
 350                 vrele(vp);
 351                 cache_drop(&nch);
 352                 return (error);
 353         }
 354         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 355         /*
 356          * Put the new filesystem on the mount list after root.  The mount
 357          * point gets its own mnt_ncmountpt (unless the VFS already set one
 358          * up) which represents the root of the mount.  The lookup code
 359          * detects the mount point going forward and checks the root of
 360          * the mount going backwards.
 361          *
 362          * It is not necessary to invalidate or purge the vnode underneath
 363          * because elements under the mount will be given their own glue
 364          * namecache record.
 365          */
 366         if (!error) {
 367                 if (mp->mnt_ncmountpt.ncp == NULL) {
 368                         /*
 369                          * allocate, then unlock, but leave the ref intact
 370                          */
 371                         cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
 372                         cache_unlock(&mp->mnt_ncmountpt);
 373                 }
 374                 mp->mnt_ncmounton = nch;                /* inherits ref */
 375                 nch.ncp->nc_flag |= NCF_ISMOUNTPT;
 376
 377                 /* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
 378                 vp->v_flag &= ~VMOUNT;
 379                 mountlist_insert(mp, MNTINS_LAST);
 380                 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
 381                 vn_unlock(vp);
 382                 error = vfs_allocate_syncvnode(mp);
 383                 vfs_unbusy(mp);
 384                 error = VFS_START(mp, 0);
 385                 vrele(vp);
 386         } else {
 387                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 388                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 389                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 390                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 391                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 392                 vp->v_flag &= ~VMOUNT;
 393                 mp->mnt_vfc->vfc_refcount--;
 394                 vfs_unbusy(mp);
 395                 kfree(mp, M_MOUNT);
 396                 cache_drop(&nch);
 397                 vput(vp);
 398         }
 399         return (error);
 400 }
 401
 402 /*
 403  * Scan all active processes to see if any of them have a current
 404  * or root directory onto which the new filesystem has just been
 405  * mounted. If so, replace them with the new mount point.
 406  *
 407  * The passed ncp is ref'd and locked (from the mount code) and
 408  * must be associated with the vnode representing the root of the
 409  * mount point.
 410  */
 411 struct checkdirs_info {
 412         struct nchandle old_nch;
 413         struct nchandle new_nch;
 414         struct vnode *old_vp;
 415         struct vnode *new_vp;
 416 };
 417
 418 static int checkdirs_callback(struct proc *p, void *data);
 419
 420 static void
 421 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
 422 {
 423         struct checkdirs_info info;
 424         struct vnode *olddp;
 425         struct vnode *newdp;
 426         struct mount *mp;
 427
 428         /*
 429          * If the old mount point's vnode has a usecount of 1, it is not
 430          * being held as a descriptor anywhere.
 431          */
 432         olddp = old_nch->ncp->nc_vp;
 433         if (olddp == NULL || olddp->v_sysref.refcnt == 1)
 434                 return;
 435
 436         /*
 437          * Force the root vnode of the new mount point to be resolved
 438          * so we can update any matching processes.
 439          */
 440         mp = new_nch->mount;
 441         if (VFS_ROOT(mp, &newdp))
 442                 panic("mount: lost mount");
 443         cache_setunresolved(new_nch);
 444         cache_setvp(new_nch, newdp);
 445
 446         /*
 447          * Special handling of the root node
 448          */
 449         if (rootvnode == olddp) {
 450                 vref(newdp);
 451                 vfs_cache_setroot(newdp, cache_hold(new_nch));
 452         }
 453
 454         /*
 455          * Pass newdp separately so the callback does not have to access
 456          * it via new_nch->ncp->nc_vp.
 457          */
 458         info.old_nch = *old_nch;
 459         info.new_nch = *new_nch;
 460         info.new_vp = newdp;
 461         allproc_scan(checkdirs_callback, &info);
 462         vput(newdp);
 463 }
 464
 465 /*
 466  * NOTE: callback is not MP safe because the scanned process's filedesc
 467  * structure can be ripped out from under us, amoung other things.
 468  */
 469 static int
 470 checkdirs_callback(struct proc *p, void *data)
 471 {
 472         struct checkdirs_info *info = data;
 473         struct filedesc *fdp;
 474         struct nchandle ncdrop1;
 475         struct nchandle ncdrop2;
 476         struct vnode *vprele1;
 477         struct vnode *vprele2;
 478
 479         if ((fdp = p->p_fd) != NULL) {
 480                 cache_zero(&ncdrop1);
 481                 cache_zero(&ncdrop2);
 482                 vprele1 = NULL;
 483                 vprele2 = NULL;
 484
 485                 /*
 486                  * MPUNSAFE - XXX fdp can be pulled out from under a
 487                  * foreign process.
 488                  *
 489                  * A shared filedesc is ok, we don't have to copy it
 490                  * because we are making this change globally.
 491                  */
 492                 spin_lock_wr(&fdp->fd_spin);
 493                 if (fdp->fd_ncdir.mount == info->old_nch.mount &&
 494                     fdp->fd_ncdir.ncp == info->old_nch.ncp) {
 495                         vprele1 = fdp->fd_cdir;
 496                         vref(info->new_vp);
 497                         fdp->fd_cdir = info->new_vp;
 498                         ncdrop1 = fdp->fd_ncdir;
 499                         cache_copy(&info->new_nch, &fdp->fd_ncdir);
 500                 }
 501                 if (fdp->fd_nrdir.mount == info->old_nch.mount &&
 502                     fdp->fd_nrdir.ncp == info->old_nch.ncp) {
 503                         vprele2 = fdp->fd_rdir;
 504                         vref(info->new_vp);
 505                         fdp->fd_rdir = info->new_vp;
 506                         ncdrop2 = fdp->fd_nrdir;
 507                         cache_copy(&info->new_nch, &fdp->fd_nrdir);
 508                 }
 509                 spin_unlock_wr(&fdp->fd_spin);
 510                 if (ncdrop1.ncp)
 511                         cache_drop(&ncdrop1);
 512                 if (ncdrop2.ncp)
 513                         cache_drop(&ncdrop2);
 514                 if (vprele1)
 515                         vrele(vprele1);
 516                 if (vprele2)
 517                         vrele(vprele2);
 518         }
 519         return(0);
 520 }
 521
 522 /*
 523  * Unmount a file system.
 524  *
 525  * Note: unmount takes a path to the vnode mounted on as argument,
 526  * not special file (as before).
 527  */
 528 /*
 529  * umount_args(char *path, int flags)
 530  */
 531 /* ARGSUSED */
 532 int
 533 sys_unmount(struct unmount_args *uap)
 534 {
 535         struct thread *td = curthread;
 536         struct proc *p = td->td_proc;
 537         struct mount *mp = NULL;
 538         int error;
 539         struct nlookupdata nd;
 540
 541         KKASSERT(p);
 542         if (p->p_ucred->cr_prison != NULL)
 543                 return (EPERM);
 544         if (usermount == 0 && (error = suser(td)))
 545                 return (error);
 546
 547         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 548         if (error == 0)
 549                 error = nlookup(&nd);
 550         if (error)
 551                 goto out;
 552
 553         mp = nd.nl_nch.mount;
 554
 555         /*
 556          * Only root, or the user that did the original mount is
 557          * permitted to unmount this filesystem.
 558          */
 559         if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
 560             (error = suser(td)))
 561                 goto out;
 562
 563         /*
 564          * Don't allow unmounting the root file system.
 565          */
 566         if (mp->mnt_flag & MNT_ROOTFS) {
 567                 error = EINVAL;
 568                 goto out;
 569         }
 570
 571         /*
 572          * Must be the root of the filesystem
 573          */
 574         if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
 575                 error = EINVAL;
 576                 goto out;
 577         }
 578
 579 out:
 580         nlookup_done(&nd);
 581         if (error)
 582                 return (error);
 583         return (dounmount(mp, uap->flags));
 584 }
 585
 586 /*
 587  * Do the actual file system unmount.
 588  */
 589 static int
 590 dounmount_interlock(struct mount *mp)
 591 {
 592         if (mp->mnt_kern_flag & MNTK_UNMOUNT)
 593                 return (EBUSY);
 594         mp->mnt_kern_flag |= MNTK_UNMOUNT;
 595         return(0);
 596 }
 597
 598 int
 599 dounmount(struct mount *mp, int flags)
 600 {
 601         struct namecache *ncp;
 602         struct nchandle nch;
 603         int error;
 604         int async_flag;
 605         int lflags;
 606         int freeok = 1;
 607
 608         /*
 609          * Exclusive access for unmounting purposes
 610          */
 611         if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
 612                 return (error);
 613
 614         /*
 615          * Allow filesystems to detect that a forced unmount is in progress.
 616          */
 617         if (flags & MNT_FORCE)
 618                 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
 619         lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
 620         error = lockmgr(&mp->mnt_lock, lflags);
 621         if (error) {
 622                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 623                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 624                         wakeup(mp);
 625                 return (error);
 626         }
 627
 628         if (mp->mnt_flag & MNT_EXPUBLIC)
 629                 vfs_setpublicfs(NULL, NULL, NULL);
 630
 631         vfs_msync(mp, MNT_WAIT);
 632         async_flag = mp->mnt_flag & MNT_ASYNC;
 633         mp->mnt_flag &=~ MNT_ASYNC;
 634
 635         /*
 636          * If this filesystem isn't aliasing other filesystems,
 637          * try to invalidate any remaining namecache entries and
 638          * check the count afterwords.
 639          */
 640         if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
 641                 cache_lock(&mp->mnt_ncmountpt);
 642                 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
 643                 cache_unlock(&mp->mnt_ncmountpt);
 644
 645                 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
 646                     (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
 647
 648                         if ((flags & MNT_FORCE) == 0) {
 649                                 error = EBUSY;
 650                                 mount_warning(mp, "Cannot unmount: "
 651                                                   "%d namecache "
 652                                                   "references still "
 653                                                   "present",
 654                                                   ncp->nc_refs - 1);
 655                         } else {
 656                                 mount_warning(mp, "Forced unmount: "
 657                                                   "%d namecache "
 658                                                   "references still "
 659                                                   "present",
 660                                                   ncp->nc_refs - 1);
 661                                 freeok = 0;
 662                         }
 663                 }
 664         }
 665
 666         /*
 667          * nchandle records ref the mount structure.  Expect a count of 1
 668          * (our mount->mnt_ncmountpt).
 669          */
 670         if (mp->mnt_refs != 1) {
 671                 if ((flags & MNT_FORCE) == 0) {
 672                         mount_warning(mp, "Cannot unmount: "
 673                                           "%d process references still "
 674                                           "present", mp->mnt_refs);
 675                         error = EBUSY;
 676                 } else {
 677                         mount_warning(mp, "Forced unmount: "
 678                                           "%d process references still "
 679                                           "present", mp->mnt_refs);
 680                         freeok = 0;
 681                 }
 682         }
 683
 684         if (error == 0) {
 685                 if (mp->mnt_syncer != NULL)
 686                         vrele(mp->mnt_syncer);
 687                 if (((mp->mnt_flag & MNT_RDONLY) ||
 688                      (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
 689                     (flags & MNT_FORCE)) {
 690                         error = VFS_UNMOUNT(mp, flags);
 691                 }
 692         }
 693         if (error) {
 694                 if (mp->mnt_syncer == NULL)
 695                         vfs_allocate_syncvnode(mp);
 696                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 697                 mp->mnt_flag |= async_flag;
 698                 lockmgr(&mp->mnt_lock, LK_RELEASE);
 699                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 700                         wakeup(mp);
 701                 return (error);
 702         }
 703         /*
 704          * Clean up any journals still associated with the mount after
 705          * filesystem activity has ceased.
 706          */
 707         journal_remove_all_journals(mp,
 708             ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
 709
 710         mountlist_remove(mp);
 711
 712         /*
 713          * Remove any installed vnode ops here so the individual VFSs don't
 714          * have to.
 715          */
 716         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 717         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 718         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 719         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 720         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 721
 722         if (mp->mnt_ncmountpt.ncp != NULL) {
 723                 nch = mp->mnt_ncmountpt;
 724                 cache_zero(&mp->mnt_ncmountpt);
 725                 cache_clrmountpt(&nch);
 726                 cache_drop(&nch);
 727         }
 728         if (mp->mnt_ncmounton.ncp != NULL) {
 729                 nch = mp->mnt_ncmounton;
 730                 cache_zero(&mp->mnt_ncmounton);
 731                 cache_clrmountpt(&nch);
 732                 cache_drop(&nch);
 733         }
 734
 735         mp->mnt_vfc->vfc_refcount--;
 736         if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
 737                 panic("unmount: dangling vnode");
 738         lockmgr(&mp->mnt_lock, LK_RELEASE);
 739         if (mp->mnt_kern_flag & MNTK_MWAIT)
 740                 wakeup(mp);
 741         if (freeok)
 742                 kfree(mp, M_MOUNT);
 743         return (0);
 744 }
 745
 746 static
 747 void
 748 mount_warning(struct mount *mp, const char *ctl, ...)
 749 {
 750         char *ptr;
 751         char *buf;
 752         __va_list va;
 753
 754         __va_start(va, ctl);
 755         if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
 756                 kprintf("unmount(%s): ", ptr);
 757                 kvprintf(ctl, va);
 758                 kprintf("\n");
 759                 kfree(buf, M_TEMP);
 760         } else {
 761                 kprintf("unmount(%p): ", mp);
 762                 kvprintf(ctl, va);
 763                 kprintf("\n");
 764         }
 765         __va_end(va);
 766 }
 767
 768 /*
 769  * Shim cache_fullpath() to handle the case where a process is chrooted into
 770  * a subdirectory of a mount.  In this case if the root mount matches the
 771  * process root directory's mount we have to specify the process's root
 772  * directory instead of the mount point, because the mount point might
 773  * be above the root directory.
 774  */
 775 static
 776 int
 777 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
 778 {
 779         struct nchandle *nch;
 780
 781         if (p && p->p_fd->fd_nrdir.mount == mp)
 782                 nch = &p->p_fd->fd_nrdir;
 783         else
 784                 nch = &mp->mnt_ncmountpt;
 785         return(cache_fullpath(p, nch, rb, fb));
 786 }
 787
 788 /*
 789  * Sync each mounted filesystem.
 790  */
 791
 792 #ifdef DEBUG
 793 static int syncprt = 0;
 794 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
 795 #endif /* DEBUG */
 796
 797 static int sync_callback(struct mount *mp, void *data);
 798
 799 /* ARGSUSED */
 800 int
 801 sys_sync(struct sync_args *uap)
 802 {
 803         mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
 804 #ifdef DEBUG
 805         /*
 806          * print out buffer pool stat information on each sync() call.
 807          */
 808         if (syncprt)
 809                 vfs_bufstats();
 810 #endif /* DEBUG */
 811         return (0);
 812 }
 813
 814 static
 815 int
 816 sync_callback(struct mount *mp, void *data __unused)
 817 {
 818         int asyncflag;
 819
 820         if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 821                 asyncflag = mp->mnt_flag & MNT_ASYNC;
 822                 mp->mnt_flag &= ~MNT_ASYNC;
 823                 vfs_msync(mp, MNT_NOWAIT);
 824                 VFS_SYNC(mp, MNT_NOWAIT);
 825                 mp->mnt_flag |= asyncflag;
 826         }
 827         return(0);
 828 }
 829
 830 /* XXX PRISON: could be per prison flag */
 831 static int prison_quotas;
 832 #if 0
 833 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
 834 #endif
 835
 836 /*
 837  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
 838  *
 839  * Change filesystem quotas.
 840  */
 841 /* ARGSUSED */
 842 int
 843 sys_quotactl(struct quotactl_args *uap)
 844 {
 845         struct nlookupdata nd;
 846         struct thread *td;
 847         struct proc *p;
 848         struct mount *mp;
 849         int error;
 850
 851         td = curthread;
 852         p = td->td_proc;
 853         if (p->p_ucred->cr_prison && !prison_quotas)
 854                 return (EPERM);
 855
 856         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 857         if (error == 0)
 858                 error = nlookup(&nd);
 859         if (error == 0) {
 860                 mp = nd.nl_nch.mount;
 861                 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
 862                                     uap->arg, nd.nl_cred);
 863         }
 864         nlookup_done(&nd);
 865         return (error);
 866 }
 867
 868 /*
 869  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
 870  *              void *buf, int buflen)
 871  *
 872  * This function operates on a mount point and executes the specified
 873  * operation using the specified control data, and possibly returns data.
 874  *
 875  * The actual number of bytes stored in the result buffer is returned, 0
 876  * if none, otherwise an error is returned.
 877  */
 878 /* ARGSUSED */
 879 int
 880 sys_mountctl(struct mountctl_args *uap)
 881 {
 882         struct thread *td = curthread;
 883         struct proc *p = td->td_proc;
 884         struct file *fp;
 885         void *ctl = NULL;
 886         void *buf = NULL;
 887         char *path = NULL;
 888         int error;
 889
 890         /*
 891          * Sanity and permissions checks.  We must be root.
 892          */
 893         KKASSERT(p);
 894         if (p->p_ucred->cr_prison != NULL)
 895                 return (EPERM);
 896         if ((error = suser(td)) != 0)
 897                 return (error);
 898
 899         /*
 900          * Argument length checks
 901          */
 902         if (uap->ctllen < 0 || uap->ctllen > 1024)
 903                 return (EINVAL);
 904         if (uap->buflen < 0 || uap->buflen > 16 * 1024)
 905                 return (EINVAL);
 906         if (uap->path == NULL)
 907                 return (EINVAL);
 908
 909         /*
 910          * Allocate the necessary buffers and copyin data
 911          */
 912         path = objcache_get(namei_oc, M_WAITOK);
 913         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 914         if (error)
 915                 goto done;
 916
 917         if (uap->ctllen) {
 918                 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
 919                 error = copyin(uap->ctl, ctl, uap->ctllen);
 920                 if (error)
 921                         goto done;
 922         }
 923         if (uap->buflen)
 924                 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
 925
 926         /*
 927          * Validate the descriptor
 928          */
 929         if (uap->fd >= 0) {
 930                 fp = holdfp(p->p_fd, uap->fd, -1);
 931                 if (fp == NULL) {
 932                         error = EBADF;
 933                         goto done;
 934                 }
 935         } else {
 936                 fp = NULL;
 937         }
 938
 939         /*
 940          * Execute the internal kernel function and clean up.
 941          */
 942         error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
 943         if (fp)
 944                 fdrop(fp);
 945         if (error == 0 && uap->sysmsg_result > 0)
 946                 error = copyout(buf, uap->buf, uap->sysmsg_result);
 947 done:
 948         if (path)
 949                 objcache_put(namei_oc, path);
 950         if (ctl)
 951                 kfree(ctl, M_TEMP);
 952         if (buf)
 953                 kfree(buf, M_TEMP);
 954         return (error);
 955 }
 956
 957 /*
 958  * Execute a mount control operation by resolving the path to a mount point
 959  * and calling vop_mountctl().
 960  */
 961 int
 962 kern_mountctl(const char *path, int op, struct file *fp,
 963                 const void *ctl, int ctllen,
 964                 void *buf, int buflen, int *res)
 965 {
 966         struct vnode *vp;
 967         struct mount *mp;
 968         struct nlookupdata nd;
 969         int error;
 970
 971         *res = 0;
 972         vp = NULL;
 973         error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
 974         if (error == 0)
 975                 error = nlookup(&nd);
 976         if (error == 0)
 977                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 978         nlookup_done(&nd);
 979         if (error)
 980                 return (error);
 981
 982         mp = vp->v_mount;
 983
 984         /*
 985          * Must be the root of the filesystem
 986          */
 987         if ((vp->v_flag & VROOT) == 0) {
 988                 vput(vp);
 989                 return (EINVAL);
 990         }
 991         error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
 992                                 buf, buflen, res);
 993         vput(vp);
 994         return (error);
 995 }
 996
 997 int
 998 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
 999 {
1000         struct thread *td = curthread;
1001         struct proc *p = td->td_proc;
1002         struct mount *mp;
1003         struct statfs *sp;
1004         char *fullpath, *freepath;
1005         int error;
1006
1007         if ((error = nlookup(nd)) != 0)
1008                 return (error);
1009         mp = nd->nl_nch.mount;
1010         sp = &mp->mnt_stat;
1011         if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1012                 return (error);
1013
1014         error = mount_path(p, mp, &fullpath, &freepath);
1015         if (error)
1016                 return(error);
1017         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1018         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1019         kfree(freepath, M_TEMP);
1020
1021         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1022         bcopy(sp, buf, sizeof(*buf));
1023         /* Only root should have access to the fsid's. */
1024         if (suser(td))
1025                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1026         return (0);
1027 }
1028
1029 /*
1030  * statfs_args(char *path, struct statfs *buf)
1031  *
1032  * Get filesystem statistics.
1033  */
1034 int
1035 sys_statfs(struct statfs_args *uap)
1036 {
1037         struct nlookupdata nd;
1038         struct statfs buf;
1039         int error;
1040
1041         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1042         if (error == 0)
1043                 error = kern_statfs(&nd, &buf);
1044         nlookup_done(&nd);
1045         if (error == 0)
1046                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1047         return (error);
1048 }
1049
1050 int
1051 kern_fstatfs(int fd, struct statfs *buf)
1052 {
1053         struct thread *td = curthread;
1054         struct proc *p = td->td_proc;
1055         struct file *fp;
1056         struct mount *mp;
1057         struct statfs *sp;
1058         char *fullpath, *freepath;
1059         int error;
1060
1061         KKASSERT(p);
1062         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1063                 return (error);
1064         mp = ((struct vnode *)fp->f_data)->v_mount;
1065         if (mp == NULL) {
1066                 error = EBADF;
1067                 goto done;
1068         }
1069         if (fp->f_cred == NULL) {
1070                 error = EINVAL;
1071                 goto done;
1072         }
1073         sp = &mp->mnt_stat;
1074         if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1075                 goto done;
1076
1077         if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1078                 goto done;
1079         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1080         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1081         kfree(freepath, M_TEMP);
1082
1083         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1084         bcopy(sp, buf, sizeof(*buf));
1085
1086         /* Only root should have access to the fsid's. */
1087         if (suser(td))
1088                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1089         error = 0;
1090 done:
1091         fdrop(fp);
1092         return (error);
1093 }
1094
1095 /*
1096  * fstatfs_args(int fd, struct statfs *buf)
1097  *
1098  * Get filesystem statistics.
1099  */
1100 int
1101 sys_fstatfs(struct fstatfs_args *uap)
1102 {
1103         struct statfs buf;
1104         int error;
1105
1106         error = kern_fstatfs(uap->fd, &buf);
1107
1108         if (error == 0)
1109                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1110         return (error);
1111 }
1112
1113 /*
1114  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1115  *
1116  * Get statistics on all filesystems.
1117  */
1118
1119 struct getfsstat_info {
1120         struct statfs *sfsp;
1121         long count;
1122         long maxcount;
1123         int error;
1124         int flags;
1125         struct proc *p;
1126 };
1127
1128 static int getfsstat_callback(struct mount *, void *);
1129
1130 /* ARGSUSED */
1131 int
1132 sys_getfsstat(struct getfsstat_args *uap)
1133 {
1134         struct thread *td = curthread;
1135         struct proc *p = td->td_proc;
1136         struct getfsstat_info info;
1137
1138         bzero(&info, sizeof(info));
1139
1140         info.maxcount = uap->bufsize / sizeof(struct statfs);
1141         info.sfsp = uap->buf;
1142         info.count = 0;
1143         info.flags = uap->flags;
1144         info.p = p;
1145
1146         mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1147         if (info.sfsp && info.count > info.maxcount)
1148                 uap->sysmsg_result = info.maxcount;
1149         else
1150                 uap->sysmsg_result = info.count;
1151         return (info.error);
1152 }
1153
1154 static int
1155 getfsstat_callback(struct mount *mp, void *data)
1156 {
1157         struct getfsstat_info *info = data;
1158         struct statfs *sp;
1159         char *freepath;
1160         char *fullpath;
1161         int error;
1162
1163         if (info->sfsp && info->count < info->maxcount) {
1164                 if (info->p && !chroot_visible_mnt(mp, info->p))
1165                         return(0);
1166                 sp = &mp->mnt_stat;
1167
1168                 /*
1169                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1170                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1171                  * overrides MNT_WAIT.
1172                  */
1173                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1174                     (info->flags & MNT_WAIT)) &&
1175                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1176                         return(0);
1177                 }
1178                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1179
1180                 error = mount_path(info->p, mp, &fullpath, &freepath);
1181                 if (error) {
1182                         info->error = error;
1183                         return(-1);
1184                 }
1185                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1186                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1187                 kfree(freepath, M_TEMP);
1188
1189                 error = copyout(sp, info->sfsp, sizeof(*sp));
1190                 if (error) {
1191                         info->error = error;
1192                         return (-1);
1193                 }
1194                 ++info->sfsp;
1195         }
1196         info->count++;
1197         return(0);
1198 }
1199
1200 /*
1201  * fchdir_args(int fd)
1202  *
1203  * Change current working directory to a given file descriptor.
1204  */
1205 /* ARGSUSED */
1206 int
1207 sys_fchdir(struct fchdir_args *uap)
1208 {
1209         struct thread *td = curthread;
1210         struct proc *p = td->td_proc;
1211         struct filedesc *fdp = p->p_fd;
1212         struct vnode *vp, *ovp;
1213         struct mount *mp;
1214         struct file *fp;
1215         struct nchandle nch, onch, tnch;
1216         int error;
1217
1218         if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1219                 return (error);
1220         vp = (struct vnode *)fp->f_data;
1221         vref(vp);
1222         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1223         if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1224                 error = ENOTDIR;
1225         else
1226                 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1227         if (error) {
1228                 vput(vp);
1229                 fdrop(fp);
1230                 return (error);
1231         }
1232         cache_copy(&fp->f_nchandle, &nch);
1233
1234         /*
1235          * If the ncp has become a mount point, traverse through
1236          * the mount point.
1237          */
1238
1239         while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1240                (mp = cache_findmount(&nch)) != NULL
1241         ) {
1242                 error = nlookup_mp(mp, &tnch);
1243                 if (error == 0) {
1244                         cache_unlock(&tnch);    /* leave ref intact */
1245                         vput(vp);
1246                         vp = tnch.ncp->nc_vp;
1247                         error = vget(vp, LK_SHARED);
1248                         KKASSERT(error == 0);
1249                         cache_drop(&nch);
1250                         nch = tnch;
1251                 }
1252         }
1253         if (error == 0) {
1254                 ovp = fdp->fd_cdir;
1255                 onch = fdp->fd_ncdir;
1256                 vn_unlock(vp);          /* leave ref intact */
1257                 fdp->fd_cdir = vp;
1258                 fdp->fd_ncdir = nch;
1259                 cache_drop(&onch);
1260                 vrele(ovp);
1261         } else {
1262                 cache_drop(&nch);
1263                 vput(vp);
1264         }
1265         fdrop(fp);
1266         return (error);
1267 }
1268
1269 int
1270 kern_chdir(struct nlookupdata *nd)
1271 {
1272         struct thread *td = curthread;
1273         struct proc *p = td->td_proc;
1274         struct filedesc *fdp = p->p_fd;
1275         struct vnode *vp, *ovp;
1276         struct nchandle onch;
1277         int error;
1278
1279         if ((error = nlookup(nd)) != 0)
1280                 return (error);
1281         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1282                 return (ENOENT);
1283         if ((error = vget(vp, LK_SHARED)) != 0)
1284                 return (error);
1285
1286         error = checkvp_chdir(vp, td);
1287         vn_unlock(vp);
1288         if (error == 0) {
1289                 ovp = fdp->fd_cdir;
1290                 onch = fdp->fd_ncdir;
1291                 cache_unlock(&nd->nl_nch);      /* leave reference intact */
1292                 fdp->fd_ncdir = nd->nl_nch;
1293                 fdp->fd_cdir = vp;
1294                 cache_drop(&onch);
1295                 vrele(ovp);
1296                 cache_zero(&nd->nl_nch);
1297         } else {
1298                 vrele(vp);
1299         }
1300         return (error);
1301 }
1302
1303 /*
1304  * chdir_args(char *path)
1305  *
1306  * Change current working directory (``.'').
1307  */
1308 int
1309 sys_chdir(struct chdir_args *uap)
1310 {
1311         struct nlookupdata nd;
1312         int error;
1313
1314         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1315         if (error == 0)
1316                 error = kern_chdir(&nd);
1317         nlookup_done(&nd);
1318         return (error);
1319 }
1320
1321 /*
1322  * Helper function for raised chroot(2) security function:  Refuse if
1323  * any filedescriptors are open directories.
1324  */
1325 static int
1326 chroot_refuse_vdir_fds(struct filedesc *fdp)
1327 {
1328         struct vnode *vp;
1329         struct file *fp;
1330         int error;
1331         int fd;
1332
1333         for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1334                 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1335                         continue;
1336                 vp = (struct vnode *)fp->f_data;
1337                 if (vp->v_type != VDIR) {
1338                         fdrop(fp);
1339                         continue;
1340                 }
1341                 fdrop(fp);
1342                 return(EPERM);
1343         }
1344         return (0);
1345 }
1346
1347 /*
1348  * This sysctl determines if we will allow a process to chroot(2) if it
1349  * has a directory open:
1350  *      0: disallowed for all processes.
1351  *      1: allowed for processes that were not already chroot(2)'ed.
1352  *      2: allowed for all processes.
1353  */
1354
1355 static int chroot_allow_open_directories = 1;
1356
1357 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1358      &chroot_allow_open_directories, 0, "");
1359
1360 /*
1361  * chroot to the specified namecache entry.  We obtain the vp from the
1362  * namecache data.  The passed ncp must be locked and referenced and will
1363  * remain locked and referenced on return.
1364  */
1365 int
1366 kern_chroot(struct nchandle *nch)
1367 {
1368         struct thread *td = curthread;
1369         struct proc *p = td->td_proc;
1370         struct filedesc *fdp = p->p_fd;
1371         struct vnode *vp;
1372         int error;
1373
1374         /*
1375          * Only root can chroot
1376          */
1377         if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1378                 return (error);
1379
1380         /*
1381          * Disallow open directory descriptors (fchdir() breakouts).
1382          */
1383         if (chroot_allow_open_directories == 0 ||
1384            (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1385                 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1386                         return (error);
1387         }
1388         if ((vp = nch->ncp->nc_vp) == NULL)
1389                 return (ENOENT);
1390
1391         if ((error = vget(vp, LK_SHARED)) != 0)
1392                 return (error);
1393
1394         /*
1395          * Check the validity of vp as a directory to change to and
1396          * associate it with rdir/jdir.
1397          */
1398         error = checkvp_chdir(vp, td);
1399         vn_unlock(vp);                  /* leave reference intact */
1400         if (error == 0) {
1401                 vrele(fdp->fd_rdir);
1402                 fdp->fd_rdir = vp;      /* reference inherited by fd_rdir */
1403                 cache_drop(&fdp->fd_nrdir);
1404                 cache_copy(nch, &fdp->fd_nrdir);
1405                 if (fdp->fd_jdir == NULL) {
1406                         fdp->fd_jdir = vp;
1407                         vref(fdp->fd_jdir);
1408                         cache_copy(nch, &fdp->fd_njdir);
1409                 }
1410         } else {
1411                 vrele(vp);
1412         }
1413         return (error);
1414 }
1415
1416 /*
1417  * chroot_args(char *path)
1418  *
1419  * Change notion of root (``/'') directory.
1420  */
1421 /* ARGSUSED */
1422 int
1423 sys_chroot(struct chroot_args *uap)
1424 {
1425         struct thread *td = curthread;
1426         struct nlookupdata nd;
1427         int error;
1428
1429         KKASSERT(td->td_proc);
1430         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1431         if (error) {
1432                 nlookup_done(&nd);
1433                 return(error);
1434         }
1435         error = nlookup(&nd);
1436         if (error == 0)
1437                 error = kern_chroot(&nd.nl_nch);
1438         nlookup_done(&nd);
1439         return(error);
1440 }
1441
1442 /*
1443  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1444  * determine whether it is legal to chdir to the vnode.  The vnode's state
1445  * is not changed by this call.
1446  */
1447 int
1448 checkvp_chdir(struct vnode *vp, struct thread *td)
1449 {
1450         int error;
1451
1452         if (vp->v_type != VDIR)
1453                 error = ENOTDIR;
1454         else
1455                 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1456         return (error);
1457 }
1458
1459 int
1460 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1461 {
1462         struct thread *td = curthread;
1463         struct proc *p = td->td_proc;
1464         struct lwp *lp = td->td_lwp;
1465         struct filedesc *fdp = p->p_fd;
1466         int cmode, flags;
1467         struct file *nfp;
1468         struct file *fp;
1469         struct vnode *vp;
1470         int type, indx, error;
1471         struct flock lf;
1472
1473         if ((oflags & O_ACCMODE) == O_ACCMODE)
1474                 return (EINVAL);
1475         flags = FFLAGS(oflags);
1476         error = falloc(p, &nfp, NULL);
1477         if (error)
1478                 return (error);
1479         fp = nfp;
1480         cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1481
1482         /*
1483          * XXX p_dupfd is a real mess.  It allows a device to return a
1484          * file descriptor to be duplicated rather then doing the open
1485          * itself.
1486          */
1487         lp->lwp_dupfd = -1;
1488
1489         /*
1490          * Call vn_open() to do the lookup and assign the vnode to the
1491          * file pointer.  vn_open() does not change the ref count on fp
1492          * and the vnode, on success, will be inherited by the file pointer
1493          * and unlocked.
1494          */
1495         nd->nl_flags |= NLC_LOCKVP;
1496         error = vn_open(nd, fp, flags, cmode);
1497         nlookup_done(nd);
1498         if (error) {
1499                 /*
1500                  * handle special fdopen() case.  bleh.  dupfdopen() is
1501                  * responsible for dropping the old contents of ofiles[indx]
1502                  * if it succeeds.
1503                  *
1504                  * Note that fsetfd() will add a ref to fp which represents
1505                  * the fd_files[] assignment.  We must still drop our
1506                  * reference.
1507                  */
1508                 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1509                         if (fdalloc(p, 0, &indx) == 0) {
1510                                 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1511                                 if (error == 0) {
1512                                         *res = indx;
1513                                         fdrop(fp);      /* our ref */
1514                                         return (0);
1515                                 }
1516                                 fsetfd(p, NULL, indx);
1517                         }
1518                 }
1519                 fdrop(fp);      /* our ref */
1520                 if (error == ERESTART)
1521                         error = EINTR;
1522                 return (error);
1523         }
1524
1525         /*
1526          * ref the vnode for ourselves so it can't be ripped out from under
1527          * is.  XXX need an ND flag to request that the vnode be returned
1528          * anyway.
1529          *
1530          * Reserve a file descriptor but do not assign it until the open
1531          * succeeds.
1532          */
1533         vp = (struct vnode *)fp->f_data;
1534         vref(vp);
1535         if ((error = fdalloc(p, 0, &indx)) != 0) {
1536                 fdrop(fp);
1537                 vrele(vp);
1538                 return (error);
1539         }
1540
1541         /*
1542          * If no error occurs the vp will have been assigned to the file
1543          * pointer.
1544          */
1545         lp->lwp_dupfd = 0;
1546
1547         if (flags & (O_EXLOCK | O_SHLOCK)) {
1548                 lf.l_whence = SEEK_SET;
1549                 lf.l_start = 0;
1550                 lf.l_len = 0;
1551                 if (flags & O_EXLOCK)
1552                         lf.l_type = F_WRLCK;
1553                 else
1554                         lf.l_type = F_RDLCK;
1555                 if (flags & FNONBLOCK)
1556                         type = 0;
1557                 else
1558                         type = F_WAIT;
1559
1560                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1561                         /*
1562                          * lock request failed.  Clean up the reserved
1563                          * descriptor.
1564                          */
1565                         vrele(vp);
1566                         fsetfd(p, NULL, indx);
1567                         fdrop(fp);
1568                         return (error);
1569                 }
1570                 fp->f_flag |= FHASLOCK;
1571         }
1572 #if 0
1573         /*
1574          * Assert that all regular file vnodes were created with a object.
1575          */
1576         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1577                 ("open: regular file has no backing object after vn_open"));
1578 #endif
1579
1580         vrele(vp);
1581
1582         /*
1583          * release our private reference, leaving the one associated with the
1584          * descriptor table intact.
1585          */
1586         fsetfd(p, fp, indx);
1587         fdrop(fp);
1588         *res = indx;
1589         return (0);
1590 }
1591
1592 /*
1593  * open_args(char *path, int flags, int mode)
1594  *
1595  * Check permissions, allocate an open file structure,
1596  * and call the device open routine if any.
1597  */
1598 int
1599 sys_open(struct open_args *uap)
1600 {
1601         struct nlookupdata nd;
1602         int error;
1603
1604         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1605         if (error == 0) {
1606                 error = kern_open(&nd, uap->flags,
1607                                     uap->mode, &uap->sysmsg_result);
1608         }
1609         nlookup_done(&nd);
1610         return (error);
1611 }
1612
1613 int
1614 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1615 {
1616         struct thread *td = curthread;
1617         struct proc *p = td->td_proc;
1618         struct vnode *vp;
1619         struct vattr vattr;
1620         int error;
1621         int whiteout = 0;
1622
1623         KKASSERT(p);
1624
1625         switch (mode & S_IFMT) {
1626         case S_IFCHR:
1627         case S_IFBLK:
1628                 error = suser(td);
1629                 break;
1630         default:
1631                 error = suser_cred(p->p_ucred, PRISON_ROOT);
1632                 break;
1633         }
1634         if (error)
1635                 return (error);
1636
1637         bwillwrite();
1638         nd->nl_flags |= NLC_CREATE;
1639         if ((error = nlookup(nd)) != 0)
1640                 return (error);
1641         if (nd->nl_nch.ncp->nc_vp)
1642                 return (EEXIST);
1643         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1644                 return (error);
1645
1646         VATTR_NULL(&vattr);
1647         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1648         vattr.va_rmajor = rmajor;
1649         vattr.va_rminor = rminor;
1650         whiteout = 0;
1651
1652         switch (mode & S_IFMT) {
1653         case S_IFMT:    /* used by badsect to flag bad sectors */
1654                 vattr.va_type = VBAD;
1655                 break;
1656         case S_IFCHR:
1657                 vattr.va_type = VCHR;
1658                 break;
1659         case S_IFBLK:
1660                 vattr.va_type = VBLK;
1661                 break;
1662         case S_IFWHT:
1663                 whiteout = 1;
1664                 break;
1665         default:
1666                 error = EINVAL;
1667                 break;
1668         }
1669         if (error == 0) {
1670                 if (whiteout) {
1671                         error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_cred, NAMEI_CREATE);
1672                 } else {
1673                         vp = NULL;
1674                         error = VOP_NMKNOD(&nd->nl_nch, &vp, nd->nl_cred, &vattr);
1675                         if (error == 0)
1676                                 vput(vp);
1677                 }
1678         }
1679         return (error);
1680 }
1681
1682 /*
1683  * mknod_args(char *path, int mode, int dev)
1684  *
1685  * Create a special file.
1686  */
1687 int
1688 sys_mknod(struct mknod_args *uap)
1689 {
1690         struct nlookupdata nd;
1691         int error;
1692
1693         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1694         if (error == 0) {
1695                 error = kern_mknod(&nd, uap->mode,
1696                                    umajor(uap->dev), uminor(uap->dev));
1697         }
1698         nlookup_done(&nd);
1699         return (error);
1700 }
1701
1702 int
1703 kern_mkfifo(struct nlookupdata *nd, int mode)
1704 {
1705         struct thread *td = curthread;
1706         struct proc *p = td->td_proc;
1707         struct vattr vattr;
1708         struct vnode *vp;
1709         int error;
1710
1711         bwillwrite();
1712
1713         nd->nl_flags |= NLC_CREATE;
1714         if ((error = nlookup(nd)) != 0)
1715                 return (error);
1716         if (nd->nl_nch.ncp->nc_vp)
1717                 return (EEXIST);
1718         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1719                 return (error);
1720
1721         VATTR_NULL(&vattr);
1722         vattr.va_type = VFIFO;
1723         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1724         vp = NULL;
1725         error = VOP_NMKNOD(&nd->nl_nch, &vp, nd->nl_cred, &vattr);
1726         if (error == 0)
1727                 vput(vp);
1728         return (error);
1729 }
1730
1731 /*
1732  * mkfifo_args(char *path, int mode)
1733  *
1734  * Create a named pipe.
1735  */
1736 int
1737 sys_mkfifo(struct mkfifo_args *uap)
1738 {
1739         struct nlookupdata nd;
1740         int error;
1741
1742         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1743         if (error == 0)
1744                 error = kern_mkfifo(&nd, uap->mode);
1745         nlookup_done(&nd);
1746         return (error);
1747 }
1748
1749 static int hardlink_check_uid = 0;
1750 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1751     &hardlink_check_uid, 0,
1752     "Unprivileged processes cannot create hard links to files owned by other "
1753     "users");
1754 static int hardlink_check_gid = 0;
1755 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1756     &hardlink_check_gid, 0,
1757     "Unprivileged processes cannot create hard links to files owned by other "
1758     "groups");
1759
1760 static int
1761 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1762 {
1763         struct vattr va;
1764         int error;
1765
1766         /*
1767          * Shortcut if disabled
1768          */
1769         if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1770                 return (0);
1771
1772         /*
1773          * root cred can always hardlink
1774          */
1775         if (suser_cred(cred, PRISON_ROOT) == 0)
1776                 return (0);
1777
1778         /*
1779          * Otherwise only if the originating file is owned by the
1780          * same user or group.  Note that any group is allowed if
1781          * the file is owned by the caller.
1782          */
1783         error = VOP_GETATTR(vp, &va);
1784         if (error != 0)
1785                 return (error);
1786
1787         if (hardlink_check_uid) {
1788                 if (cred->cr_uid != va.va_uid)
1789                         return (EPERM);
1790         }
1791
1792         if (hardlink_check_gid) {
1793                 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1794                         return (EPERM);
1795         }
1796
1797         return (0);
1798 }
1799
1800 int
1801 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1802 {
1803         struct thread *td = curthread;
1804         struct vnode *vp;
1805         int error;
1806
1807         /*
1808          * Lookup the source and obtained a locked vnode.
1809          *
1810          * XXX relookup on vget failure / race ?
1811          */
1812         bwillwrite();
1813         if ((error = nlookup(nd)) != 0)
1814                 return (error);
1815         vp = nd->nl_nch.ncp->nc_vp;
1816         KKASSERT(vp != NULL);
1817         if (vp->v_type == VDIR)
1818                 return (EPERM);         /* POSIX */
1819         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1820                 return (error);
1821         if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1822                 return (error);
1823
1824         /*
1825          * Unlock the source so we can lookup the target without deadlocking
1826          * (XXX vp is locked already, possible other deadlock?).  The target
1827          * must not exist.
1828          */
1829         KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1830         nd->nl_flags &= ~NLC_NCPISLOCKED;
1831         cache_unlock(&nd->nl_nch);
1832
1833         linknd->nl_flags |= NLC_CREATE;
1834         if ((error = nlookup(linknd)) != 0) {
1835                 vput(vp);
1836                 return (error);
1837         }
1838         if (linknd->nl_nch.ncp->nc_vp) {
1839                 vput(vp);
1840                 return (EEXIST);
1841         }
1842
1843         /*
1844          * Finally run the new API VOP.
1845          */
1846         error = can_hardlink(vp, td, td->td_proc->p_ucred);
1847         if (error == 0)
1848                 error = VOP_NLINK(&linknd->nl_nch, vp, linknd->nl_cred);
1849         vput(vp);
1850         return (error);
1851 }
1852
1853 /*
1854  * link_args(char *path, char *link)
1855  *
1856  * Make a hard file link.
1857  */
1858 int
1859 sys_link(struct link_args *uap)
1860 {
1861         struct nlookupdata nd, linknd;
1862         int error;
1863
1864         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1865         if (error == 0) {
1866                 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1867                 if (error == 0)
1868                         error = kern_link(&nd, &linknd);
1869                 nlookup_done(&linknd);
1870         }
1871         nlookup_done(&nd);
1872         return (error);
1873 }
1874
1875 int
1876 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1877 {
1878         struct vattr vattr;
1879         struct vnode *vp;
1880         int error;
1881
1882         bwillwrite();
1883         nd->nl_flags |= NLC_CREATE;
1884         if ((error = nlookup(nd)) != 0)
1885                 return (error);
1886         if (nd->nl_nch.ncp->nc_vp)
1887                 return (EEXIST);
1888         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1889                 return (error);
1890         VATTR_NULL(&vattr);
1891         vattr.va_mode = mode;
1892         error = VOP_NSYMLINK(&nd->nl_nch, &vp, nd->nl_cred, &vattr, path);
1893         if (error == 0)
1894                 vput(vp);
1895         return (error);
1896 }
1897
1898 /*
1899  * symlink(char *path, char *link)
1900  *
1901  * Make a symbolic link.
1902  */
1903 int
1904 sys_symlink(struct symlink_args *uap)
1905 {
1906         struct thread *td = curthread;
1907         struct nlookupdata nd;
1908         char *path;
1909         int error;
1910         int mode;
1911
1912         path = objcache_get(namei_oc, M_WAITOK);
1913         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1914         if (error == 0) {
1915                 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1916                 if (error == 0) {
1917                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1918                         error = kern_symlink(&nd, path, mode);
1919                 }
1920                 nlookup_done(&nd);
1921         }
1922         objcache_put(namei_oc, path);
1923         return (error);
1924 }
1925
1926 /*
1927  * undelete_args(char *path)
1928  *
1929  * Delete a whiteout from the filesystem.
1930  */
1931 /* ARGSUSED */
1932 int
1933 sys_undelete(struct undelete_args *uap)
1934 {
1935         struct nlookupdata nd;
1936         int error;
1937
1938         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1939         bwillwrite();
1940         nd.nl_flags |= NLC_DELETE;
1941         if (error == 0)
1942                 error = nlookup(&nd);
1943         if (error == 0)
1944                 error = ncp_writechk(&nd.nl_nch);
1945         if (error == 0)
1946                 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_cred, NAMEI_DELETE);
1947         nlookup_done(&nd);
1948         return (error);
1949 }
1950
1951 int
1952 kern_unlink(struct nlookupdata *nd)
1953 {
1954         int error;
1955
1956         bwillwrite();
1957         nd->nl_flags |= NLC_DELETE;
1958         if ((error = nlookup(nd)) != 0)
1959                 return (error);
1960         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1961                 return (error);
1962         error = VOP_NREMOVE(&nd->nl_nch, nd->nl_cred);
1963         return (error);
1964 }
1965
1966 /*
1967  * unlink_args(char *path)
1968  *
1969  * Delete a name from the filesystem.
1970  */
1971 int
1972 sys_unlink(struct unlink_args *uap)
1973 {
1974         struct nlookupdata nd;
1975         int error;
1976
1977         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1978         if (error == 0)
1979                 error = kern_unlink(&nd);
1980         nlookup_done(&nd);
1981         return (error);
1982 }
1983
1984 int
1985 kern_lseek(int fd, off_t offset, int whence, off_t *res)
1986 {
1987         struct thread *td = curthread;
1988         struct proc *p = td->td_proc;
1989         struct file *fp;
1990         struct vattr vattr;
1991         int error;
1992
1993         fp = holdfp(p->p_fd, fd, -1);
1994         if (fp == NULL)
1995                 return (EBADF);
1996         if (fp->f_type != DTYPE_VNODE) {
1997                 error = ESPIPE;
1998                 goto done;
1999         }
2000
2001         switch (whence) {
2002         case L_INCR:
2003                 fp->f_offset += offset;
2004                 error = 0;
2005                 break;
2006         case L_XTND:
2007                 error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
2008                 if (error == 0)
2009                         fp->f_offset = offset + vattr.va_size;
2010                 break;
2011         case L_SET:
2012                 fp->f_offset = offset;
2013                 error = 0;
2014                 break;
2015         default:
2016                 error = EINVAL;
2017                 break;
2018         }
2019         *res = fp->f_offset;
2020 done:
2021         fdrop(fp);
2022         return (error);
2023 }
2024
2025 /*
2026  * lseek_args(int fd, int pad, off_t offset, int whence)
2027  *
2028  * Reposition read/write file offset.
2029  */
2030 int
2031 sys_lseek(struct lseek_args *uap)
2032 {
2033         int error;
2034
2035         error = kern_lseek(uap->fd, uap->offset, uap->whence,
2036             &uap->sysmsg_offset);
2037
2038         return (error);
2039 }
2040
2041 int
2042 kern_access(struct nlookupdata *nd, int aflags)
2043 {
2044         struct vnode *vp;
2045         int error, flags;
2046
2047         if ((error = nlookup(nd)) != 0)
2048                 return (error);
2049 retry:
2050         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2051         if (error)
2052                 return (error);
2053
2054         /* Flags == 0 means only check for existence. */
2055         if (aflags) {
2056                 flags = 0;
2057                 if (aflags & R_OK)
2058                         flags |= VREAD;
2059                 if (aflags & W_OK)
2060                         flags |= VWRITE;
2061                 if (aflags & X_OK)
2062                         flags |= VEXEC;
2063                 if ((flags & VWRITE) == 0 ||
2064                     (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2065                         error = VOP_ACCESS(vp, flags, nd->nl_cred);
2066
2067                 /*
2068                  * If the file handle is stale we have to re-resolve the
2069                  * entry.  This is a hack at the moment.
2070                  */
2071                 if (error == ESTALE) {
2072                         cache_setunresolved(&nd->nl_nch);
2073                         error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2074                         if (error == 0) {
2075                                 vput(vp);
2076                                 vp = NULL;
2077                                 goto retry;
2078                         }
2079                 }
2080         }
2081         vput(vp);
2082         return (error);
2083 }
2084
2085 /*
2086  * access_args(char *path, int flags)
2087  *
2088  * Check access permissions.
2089  */
2090 int
2091 sys_access(struct access_args *uap)
2092 {
2093         struct nlookupdata nd;
2094         int error;
2095
2096         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2097         if (error == 0)
2098                 error = kern_access(&nd, uap->flags);
2099         nlookup_done(&nd);
2100         return (error);
2101 }
2102
2103 int
2104 kern_stat(struct nlookupdata *nd, struct stat *st)
2105 {
2106         int error;
2107         struct vnode *vp;
2108         thread_t td;
2109
2110         if ((error = nlookup(nd)) != 0)
2111                 return (error);
2112 again:
2113         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2114                 return (ENOENT);
2115
2116         td = curthread;
2117         if ((error = vget(vp, LK_SHARED)) != 0)
2118                 return (error);
2119         error = vn_stat(vp, st, nd->nl_cred);
2120
2121         /*
2122          * If the file handle is stale we have to re-resolve the entry.  This
2123          * is a hack at the moment.
2124          */
2125         if (error == ESTALE) {
2126                 cache_setunresolved(&nd->nl_nch);
2127                 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2128                 if (error == 0) {
2129                         vput(vp);
2130                         goto again;
2131                 }
2132         }
2133         vput(vp);
2134         return (error);
2135 }
2136
2137 /*
2138  * stat_args(char *path, struct stat *ub)
2139  *
2140  * Get file status; this version follows links.
2141  */
2142 int
2143 sys_stat(struct stat_args *uap)
2144 {
2145         struct nlookupdata nd;
2146         struct stat st;
2147         int error;
2148
2149         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2150         if (error == 0) {
2151                 error = kern_stat(&nd, &st);
2152                 if (error == 0)
2153                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2154         }
2155         nlookup_done(&nd);
2156         return (error);
2157 }
2158
2159 /*
2160  * lstat_args(char *path, struct stat *ub)
2161  *
2162  * Get file status; this version does not follow links.
2163  */
2164 int
2165 sys_lstat(struct lstat_args *uap)
2166 {
2167         struct nlookupdata nd;
2168         struct stat st;
2169         int error;
2170
2171         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2172         if (error == 0) {
2173                 error = kern_stat(&nd, &st);
2174                 if (error == 0)
2175                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2176         }
2177         nlookup_done(&nd);
2178         return (error);
2179 }
2180
2181 /*
2182  * pathconf_Args(char *path, int name)
2183  *
2184  * Get configurable pathname variables.
2185  */
2186 /* ARGSUSED */
2187 int
2188 sys_pathconf(struct pathconf_args *uap)
2189 {
2190         struct nlookupdata nd;
2191         struct vnode *vp;
2192         int error;
2193
2194         vp = NULL;
2195         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2196         if (error == 0)
2197                 error = nlookup(&nd);
2198         if (error == 0)
2199                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2200         nlookup_done(&nd);
2201         if (error == 0) {
2202                 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2203                 vput(vp);
2204         }
2205         return (error);
2206 }
2207
2208 /*
2209  * XXX: daver
2210  * kern_readlink isn't properly split yet.  There is a copyin burried
2211  * in VOP_READLINK().
2212  */
2213 int
2214 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2215 {
2216         struct thread *td = curthread;
2217         struct proc *p = td->td_proc;
2218         struct vnode *vp;
2219         struct iovec aiov;
2220         struct uio auio;
2221         int error;
2222
2223         if ((error = nlookup(nd)) != 0)
2224                 return (error);
2225         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2226         if (error)
2227                 return (error);
2228         if (vp->v_type != VLNK) {
2229                 error = EINVAL;
2230         } else {
2231                 aiov.iov_base = buf;
2232                 aiov.iov_len = count;
2233                 auio.uio_iov = &aiov;
2234                 auio.uio_iovcnt = 1;
2235                 auio.uio_offset = 0;
2236                 auio.uio_rw = UIO_READ;
2237                 auio.uio_segflg = UIO_USERSPACE;
2238                 auio.uio_td = td;
2239                 auio.uio_resid = count;
2240                 error = VOP_READLINK(vp, &auio, p->p_ucred);
2241         }
2242         vput(vp);
2243         *res = count - auio.uio_resid;
2244         return (error);
2245 }
2246
2247 /*
2248  * readlink_args(char *path, char *buf, int count)
2249  *
2250  * Return target name of a symbolic link.
2251  */
2252 int
2253 sys_readlink(struct readlink_args *uap)
2254 {
2255         struct nlookupdata nd;
2256         int error;
2257
2258         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2259         if (error == 0) {
2260                 error = kern_readlink(&nd, uap->buf, uap->count,
2261                                         &uap->sysmsg_result);
2262         }
2263         nlookup_done(&nd);
2264         return (error);
2265 }
2266
2267 static int
2268 setfflags(struct vnode *vp, int flags)
2269 {
2270         struct thread *td = curthread;
2271         struct proc *p = td->td_proc;
2272         int error;
2273         struct vattr vattr;
2274
2275         /*
2276          * Prevent non-root users from setting flags on devices.  When
2277          * a device is reused, users can retain ownership of the device
2278          * if they are allowed to set flags and programs assume that
2279          * chown can't fail when done as root.
2280          */
2281         if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2282             ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2283                 return (error);
2284
2285         /*
2286          * note: vget is required for any operation that might mod the vnode
2287          * so VINACTIVE is properly cleared.
2288          */
2289         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2290                 VATTR_NULL(&vattr);
2291                 vattr.va_flags = flags;
2292                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2293                 vput(vp);
2294         }
2295         return (error);
2296 }
2297
2298 /*
2299  * chflags(char *path, int flags)
2300  *
2301  * Change flags of a file given a path name.
2302  */
2303 /* ARGSUSED */
2304 int
2305 sys_chflags(struct chflags_args *uap)
2306 {
2307         struct nlookupdata nd;
2308         struct vnode *vp;
2309         int error;
2310
2311         vp = NULL;
2312         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2313         /* XXX Add NLC flag indicating modifying operation? */
2314         if (error == 0)
2315                 error = nlookup(&nd);
2316         if (error == 0)
2317                 error = ncp_writechk(&nd.nl_nch);
2318         if (error == 0)
2319                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2320         nlookup_done(&nd);
2321         if (error == 0) {
2322                 error = setfflags(vp, uap->flags);
2323                 vrele(vp);
2324         }
2325         return (error);
2326 }
2327
2328 /*
2329  * fchflags_args(int fd, int flags)
2330  *
2331  * Change flags of a file given a file descriptor.
2332  */
2333 /* ARGSUSED */
2334 int
2335 sys_fchflags(struct fchflags_args *uap)
2336 {
2337         struct thread *td = curthread;
2338         struct proc *p = td->td_proc;
2339         struct file *fp;
2340         int error;
2341
2342         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2343                 return (error);
2344         if (fp->f_nchandle.ncp)
2345                 error = ncp_writechk(&fp->f_nchandle);
2346         if (error == 0)
2347                 error = setfflags((struct vnode *) fp->f_data, uap->flags);
2348         fdrop(fp);
2349         return (error);
2350 }
2351
2352 static int
2353 setfmode(struct vnode *vp, int mode)
2354 {
2355         struct thread *td = curthread;
2356         struct proc *p = td->td_proc;
2357         int error;
2358         struct vattr vattr;
2359
2360         /*
2361          * note: vget is required for any operation that might mod the vnode
2362          * so VINACTIVE is properly cleared.
2363          */
2364         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2365                 VATTR_NULL(&vattr);
2366                 vattr.va_mode = mode & ALLPERMS;
2367                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2368                 vput(vp);
2369         }
2370         return error;
2371 }
2372
2373 int
2374 kern_chmod(struct nlookupdata *nd, int mode)
2375 {
2376         struct vnode *vp;
2377         int error;
2378
2379         /* XXX Add NLC flag indicating modifying operation? */
2380         if ((error = nlookup(nd)) != 0)
2381                 return (error);
2382         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2383                 return (error);
2384         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2385                 error = setfmode(vp, mode);
2386         vrele(vp);
2387         return (error);
2388 }
2389
2390 /*
2391  * chmod_args(char *path, int mode)
2392  *
2393  * Change mode of a file given path name.
2394  */
2395 /* ARGSUSED */
2396 int
2397 sys_chmod(struct chmod_args *uap)
2398 {
2399         struct nlookupdata nd;
2400         int error;
2401
2402         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2403         if (error == 0)
2404                 error = kern_chmod(&nd, uap->mode);
2405         nlookup_done(&nd);
2406         return (error);
2407 }
2408
2409 /*
2410  * lchmod_args(char *path, int mode)
2411  *
2412  * Change mode of a file given path name (don't follow links.)
2413  */
2414 /* ARGSUSED */
2415 int
2416 sys_lchmod(struct lchmod_args *uap)
2417 {
2418         struct nlookupdata nd;
2419         int error;
2420
2421         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2422         if (error == 0)
2423                 error = kern_chmod(&nd, uap->mode);
2424         nlookup_done(&nd);
2425         return (error);
2426 }
2427
2428 /*
2429  * fchmod_args(int fd, int mode)
2430  *
2431  * Change mode of a file given a file descriptor.
2432  */
2433 /* ARGSUSED */
2434 int
2435 sys_fchmod(struct fchmod_args *uap)
2436 {
2437         struct thread *td = curthread;
2438         struct proc *p = td->td_proc;
2439         struct file *fp;
2440         int error;
2441
2442         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2443                 return (error);
2444         if (fp->f_nchandle.ncp)
2445                 error = ncp_writechk(&fp->f_nchandle);
2446         if (error == 0)
2447                 error = setfmode((struct vnode *)fp->f_data, uap->mode);
2448         fdrop(fp);
2449         return (error);
2450 }
2451
2452 static int
2453 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2454 {
2455         struct thread *td = curthread;
2456         struct proc *p = td->td_proc;
2457         int error;
2458         struct vattr vattr;
2459
2460         /*
2461          * note: vget is required for any operation that might mod the vnode
2462          * so VINACTIVE is properly cleared.
2463          */
2464         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2465                 VATTR_NULL(&vattr);
2466                 vattr.va_uid = uid;
2467                 vattr.va_gid = gid;
2468                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2469                 vput(vp);
2470         }
2471         return error;
2472 }
2473
2474 int
2475 kern_chown(struct nlookupdata *nd, int uid, int gid)
2476 {
2477         struct vnode *vp;
2478         int error;
2479
2480         /* XXX Add NLC flag indicating modifying operation? */
2481         if ((error = nlookup(nd)) != 0)
2482                 return (error);
2483         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2484                 return (error);
2485         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2486                 error = setfown(vp, uid, gid);
2487         vrele(vp);
2488         return (error);
2489 }
2490
2491 /*
2492  * chown(char *path, int uid, int gid)
2493  *
2494  * Set ownership given a path name.
2495  */
2496 int
2497 sys_chown(struct chown_args *uap)
2498 {
2499         struct nlookupdata nd;
2500         int error;
2501
2502         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2503         if (error == 0)
2504                 error = kern_chown(&nd, uap->uid, uap->gid);
2505         nlookup_done(&nd);
2506         return (error);
2507 }
2508
2509 /*
2510  * lchown_args(char *path, int uid, int gid)
2511  *
2512  * Set ownership given a path name, do not cross symlinks.
2513  */
2514 int
2515 sys_lchown(struct lchown_args *uap)
2516 {
2517         struct nlookupdata nd;
2518         int error;
2519
2520         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2521         if (error == 0)
2522                 error = kern_chown(&nd, uap->uid, uap->gid);
2523         nlookup_done(&nd);
2524         return (error);
2525 }
2526
2527 /*
2528  * fchown_args(int fd, int uid, int gid)
2529  *
2530  * Set ownership given a file descriptor.
2531  */
2532 /* ARGSUSED */
2533 int
2534 sys_fchown(struct fchown_args *uap)
2535 {
2536         struct thread *td = curthread;
2537         struct proc *p = td->td_proc;
2538         struct file *fp;
2539         int error;
2540
2541         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2542                 return (error);
2543         if (fp->f_nchandle.ncp)
2544                 error = ncp_writechk(&fp->f_nchandle);
2545         if (error == 0)
2546                 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2547         fdrop(fp);
2548         return (error);
2549 }
2550
2551 static int
2552 getutimes(const struct timeval *tvp, struct timespec *tsp)
2553 {
2554         struct timeval tv[2];
2555
2556         if (tvp == NULL) {
2557                 microtime(&tv[0]);
2558                 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2559                 tsp[1] = tsp[0];
2560         } else {
2561                 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2562                 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2563         }
2564         return 0;
2565 }
2566
2567 static int
2568 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2569 {
2570         struct thread *td = curthread;
2571         struct proc *p = td->td_proc;
2572         int error;
2573         struct vattr vattr;
2574
2575         /*
2576          * note: vget is required for any operation that might mod the vnode
2577          * so VINACTIVE is properly cleared.
2578          */
2579         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2580                 VATTR_NULL(&vattr);
2581                 vattr.va_atime = ts[0];
2582                 vattr.va_mtime = ts[1];
2583                 if (nullflag)
2584                         vattr.va_vaflags |= VA_UTIMES_NULL;
2585                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2586                 vput(vp);
2587         }
2588         return error;
2589 }
2590
2591 int
2592 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2593 {
2594         struct timespec ts[2];
2595         struct vnode *vp;
2596         int error;
2597
2598         if ((error = getutimes(tptr, ts)) != 0)
2599                 return (error);
2600         /* XXX Add NLC flag indicating modifying operation? */
2601         if ((error = nlookup(nd)) != 0)
2602                 return (error);
2603         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2604                 return (error);
2605         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2606                 return (error);
2607         error = setutimes(vp, ts, tptr == NULL);
2608         vrele(vp);
2609         return (error);
2610 }
2611
2612 /*
2613  * utimes_args(char *path, struct timeval *tptr)
2614  *
2615  * Set the access and modification times of a file.
2616  */
2617 int
2618 sys_utimes(struct utimes_args *uap)
2619 {
2620         struct timeval tv[2];
2621         struct nlookupdata nd;
2622         int error;
2623
2624         if (uap->tptr) {
2625                 error = copyin(uap->tptr, tv, sizeof(tv));
2626                 if (error)
2627                         return (error);
2628         }
2629         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2630         if (error == 0)
2631                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2632         nlookup_done(&nd);
2633         return (error);
2634 }
2635
2636 /*
2637  * lutimes_args(char *path, struct timeval *tptr)
2638  *
2639  * Set the access and modification times of a file.
2640  */
2641 int
2642 sys_lutimes(struct lutimes_args *uap)
2643 {
2644         struct timeval tv[2];
2645         struct nlookupdata nd;
2646         int error;
2647
2648         if (uap->tptr) {
2649                 error = copyin(uap->tptr, tv, sizeof(tv));
2650                 if (error)
2651                         return (error);
2652         }
2653         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2654         if (error == 0)
2655                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2656         nlookup_done(&nd);
2657         return (error);
2658 }
2659
2660 int
2661 kern_futimes(int fd, struct timeval *tptr)
2662 {
2663         struct thread *td = curthread;
2664         struct proc *p = td->td_proc;
2665         struct timespec ts[2];
2666         struct file *fp;
2667         int error;
2668
2669         error = getutimes(tptr, ts);
2670         if (error)
2671                 return (error);
2672         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2673                 return (error);
2674         if (fp->f_nchandle.ncp)
2675                 error = ncp_writechk(&fp->f_nchandle);
2676         if (error == 0)
2677                 error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2678         fdrop(fp);
2679         return (error);
2680 }
2681
2682 /*
2683  * futimes_args(int fd, struct timeval *tptr)
2684  *
2685  * Set the access and modification times of a file.
2686  */
2687 int
2688 sys_futimes(struct futimes_args *uap)
2689 {
2690         struct timeval tv[2];
2691         int error;
2692
2693         if (uap->tptr) {
2694                 error = copyin(uap->tptr, tv, sizeof(tv));
2695                 if (error)
2696                         return (error);
2697         }
2698
2699         error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2700
2701         return (error);
2702 }
2703
2704 int
2705 kern_truncate(struct nlookupdata *nd, off_t length)
2706 {
2707         struct vnode *vp;
2708         struct vattr vattr;
2709         int error;
2710
2711         if (length < 0)
2712                 return(EINVAL);
2713         /* XXX Add NLC flag indicating modifying operation? */
2714         if ((error = nlookup(nd)) != 0)
2715                 return (error);
2716         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2717                 return (error);
2718         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2719                 return (error);
2720         if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2721                 vrele(vp);
2722                 return (error);
2723         }
2724         if (vp->v_type == VDIR) {
2725                 error = EISDIR;
2726         } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
2727             (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2728                 VATTR_NULL(&vattr);
2729                 vattr.va_size = length;
2730                 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2731         }
2732         vput(vp);
2733         return (error);
2734 }
2735
2736 /*
2737  * truncate(char *path, int pad, off_t length)
2738  *
2739  * Truncate a file given its path name.
2740  */
2741 int
2742 sys_truncate(struct truncate_args *uap)
2743 {
2744         struct nlookupdata nd;
2745         int error;
2746
2747         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2748         if (error == 0)
2749                 error = kern_truncate(&nd, uap->length);
2750         nlookup_done(&nd);
2751         return error;
2752 }
2753
2754 int
2755 kern_ftruncate(int fd, off_t length)
2756 {
2757         struct thread *td = curthread;
2758         struct proc *p = td->td_proc;
2759         struct vattr vattr;
2760         struct vnode *vp;
2761         struct file *fp;
2762         int error;
2763
2764         if (length < 0)
2765                 return(EINVAL);
2766         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2767                 return (error);
2768         if (fp->f_nchandle.ncp) {
2769                 error = ncp_writechk(&fp->f_nchandle);
2770                 if (error)
2771                         goto done;
2772         }
2773         if ((fp->f_flag & FWRITE) == 0) {
2774                 error = EINVAL;
2775                 goto done;
2776         }
2777         vp = (struct vnode *)fp->f_data;
2778         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2779         if (vp->v_type == VDIR) {
2780                 error = EISDIR;
2781         } else if ((error = vn_writechk(vp, NULL)) == 0) {
2782                 VATTR_NULL(&vattr);
2783                 vattr.va_size = length;
2784                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2785         }
2786         vn_unlock(vp);
2787 done:
2788         fdrop(fp);
2789         return (error);
2790 }
2791
2792 /*
2793  * ftruncate_args(int fd, int pad, off_t length)
2794  *
2795  * Truncate a file given a file descriptor.
2796  */
2797 int
2798 sys_ftruncate(struct ftruncate_args *uap)
2799 {
2800         int error;
2801
2802         error = kern_ftruncate(uap->fd, uap->length);
2803
2804         return (error);
2805 }
2806
2807 /*
2808  * fsync(int fd)
2809  *
2810  * Sync an open file.
2811  */
2812 /* ARGSUSED */
2813 int
2814 sys_fsync(struct fsync_args *uap)
2815 {
2816         struct thread *td = curthread;
2817         struct proc *p = td->td_proc;
2818         struct vnode *vp;
2819         struct file *fp;
2820         vm_object_t obj;
2821         int error;
2822
2823         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2824                 return (error);
2825         vp = (struct vnode *)fp->f_data;
2826         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2827         if ((obj = vp->v_object) != NULL)
2828                 vm_object_page_clean(obj, 0, 0, 0);
2829         if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 &&
2830             vp->v_mount && (vp->v_mount->mnt_flag & MNT_SOFTDEP) &&
2831             bioops.io_fsync) {
2832                 error = (*bioops.io_fsync)(vp);
2833         }
2834         vn_unlock(vp);
2835         fdrop(fp);
2836         return (error);
2837 }
2838
2839 int
2840 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2841 {
2842         struct nchandle fnchd;
2843         struct nchandle tnchd;
2844         struct namecache *ncp;
2845         struct mount *mp;
2846         int error;
2847
2848         bwillwrite();
2849         if ((error = nlookup(fromnd)) != 0)
2850                 return (error);
2851         if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
2852                 return (ENOENT);
2853         fnchd.mount = fromnd->nl_nch.mount;
2854         cache_hold(&fnchd);
2855
2856         /*
2857          * unlock the source nch so we can lookup the target nch without
2858          * deadlocking.  The target may or may not exist so we do not check
2859          * for a target vp like kern_mkdir() and other creation functions do.
2860          *
2861          * The source and target directories are ref'd and rechecked after
2862          * everything is relocked to determine if the source or target file
2863          * has been renamed.
2864          */
2865         KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2866         fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2867         cache_unlock(&fromnd->nl_nch);
2868
2869         tond->nl_flags |= NLC_CREATE;
2870         if ((error = nlookup(tond)) != 0) {
2871                 cache_drop(&fnchd);
2872                 return (error);
2873         }
2874         if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
2875                 cache_drop(&fnchd);
2876                 return (ENOENT);
2877         }
2878         tnchd.mount = tond->nl_nch.mount;
2879         cache_hold(&tnchd);
2880
2881         /*
2882          * If the source and target are the same there is nothing to do
2883          */
2884         if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
2885                 cache_drop(&fnchd);
2886                 cache_drop(&tnchd);
2887                 return (0);
2888         }
2889
2890         /*
2891          * Mount points cannot be renamed or overwritten
2892          */
2893         if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
2894             NCF_ISMOUNTPT
2895         ) {
2896                 cache_drop(&fnchd);
2897                 cache_drop(&tnchd);
2898                 return (EINVAL);
2899         }
2900
2901         /*
2902          * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2903          * may have become invalid while it was unlocked, nc_vp and nc_mount
2904          * could be NULL.
2905          */
2906         if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
2907                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2908         } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
2909                 cache_lock(&fromnd->nl_nch);
2910                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2911         } else {
2912                 cache_unlock(&tond->nl_nch);
2913                 cache_lock(&fromnd->nl_nch);
2914                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2915                 cache_lock(&tond->nl_nch);
2916                 cache_resolve(&tond->nl_nch, tond->nl_cred);
2917         }
2918         fromnd->nl_flags |= NLC_NCPISLOCKED;
2919
2920         /*
2921          * make sure the parent directories linkages are the same
2922          */
2923         if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
2924             tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
2925                 cache_drop(&fnchd);
2926                 cache_drop(&tnchd);
2927                 return (ENOENT);
2928         }
2929
2930         /*
2931          * Both the source and target must be within the same filesystem and
2932          * in the same filesystem as their parent directories within the
2933          * namecache topology.
2934          *
2935          * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2936          */
2937         mp = fnchd.mount;
2938         if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
2939             mp != tond->nl_nch.mount) {
2940                 cache_drop(&fnchd);
2941                 cache_drop(&tnchd);
2942                 return (EXDEV);
2943         }
2944
2945         /*
2946          * Make sure the mount point is writable
2947          */
2948         if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
2949                 cache_drop(&fnchd);
2950                 cache_drop(&tnchd);
2951                 return (error);
2952         }
2953
2954         /*
2955          * If the target exists and either the source or target is a directory,
2956          * then both must be directories.
2957          *
2958          * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
2959          * have become NULL.
2960          */
2961         if (tond->nl_nch.ncp->nc_vp) {
2962                 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
2963                         error = ENOENT;
2964                 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
2965                         if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
2966                                 error = ENOTDIR;
2967                 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
2968                         error = EISDIR;
2969                 }
2970         }
2971
2972         /*
2973          * You cannot rename a source into itself or a subdirectory of itself.
2974          * We check this by travsersing the target directory upwards looking
2975          * for a match against the source.
2976          */
2977         if (error == 0) {
2978                 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
2979                         if (fromnd->nl_nch.ncp == ncp) {
2980                                 error = EINVAL;
2981                                 break;
2982                         }
2983                 }
2984         }
2985
2986         cache_drop(&fnchd);
2987         cache_drop(&tnchd);
2988
2989         /*
2990          * Even though the namespaces are different, they may still represent
2991          * hardlinks to the same file.  The filesystem might have a hard time
2992          * with this so we issue a NREMOVE of the source instead of a NRENAME
2993          * when we detect the situation.
2994          */
2995         if (error == 0) {
2996                 if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
2997                         error = VOP_NREMOVE(&fromnd->nl_nch, fromnd->nl_cred);
2998                 } else {
2999                         error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3000                                             tond->nl_cred);
3001                 }
3002         }
3003         return (error);
3004 }
3005
3006 /*
3007  * rename_args(char *from, char *to)
3008  *
3009  * Rename files.  Source and destination must either both be directories,
3010  * or both not be directories.  If target is a directory, it must be empty.
3011  */
3012 int
3013 sys_rename(struct rename_args *uap)
3014 {
3015         struct nlookupdata fromnd, tond;
3016         int error;
3017
3018         error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3019         if (error == 0) {
3020                 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3021                 if (error == 0)
3022                         error = kern_rename(&fromnd, &tond);
3023                 nlookup_done(&tond);
3024         }
3025         nlookup_done(&fromnd);
3026         return (error);
3027 }
3028
3029 int
3030 kern_mkdir(struct nlookupdata *nd, int mode)
3031 {
3032         struct thread *td = curthread;
3033         struct proc *p = td->td_proc;
3034         struct vnode *vp;
3035         struct vattr vattr;
3036         int error;
3037
3038         bwillwrite();
3039         nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
3040         if ((error = nlookup(nd)) != 0)
3041                 return (error);
3042
3043         if (nd->nl_nch.ncp->nc_vp)
3044                 return (EEXIST);
3045         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3046                 return (error);
3047
3048         VATTR_NULL(&vattr);
3049         vattr.va_type = VDIR;
3050         vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3051
3052         vp = NULL;
3053         error = VOP_NMKDIR(&nd->nl_nch, &vp, p->p_ucred, &vattr);
3054         if (error == 0)
3055                 vput(vp);
3056         return (error);
3057 }
3058
3059 /*
3060  * mkdir_args(char *path, int mode)
3061  *
3062  * Make a directory file.
3063  */
3064 /* ARGSUSED */
3065 int
3066 sys_mkdir(struct mkdir_args *uap)
3067 {
3068         struct nlookupdata nd;
3069         int error;
3070
3071         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3072         if (error == 0)
3073                 error = kern_mkdir(&nd, uap->mode);
3074         nlookup_done(&nd);
3075         return (error);
3076 }
3077
3078 int
3079 kern_rmdir(struct nlookupdata *nd)
3080 {
3081         int error;
3082
3083         bwillwrite();
3084         nd->nl_flags |= NLC_DELETE;
3085         if ((error = nlookup(nd)) != 0)
3086                 return (error);
3087
3088         /*
3089          * Do not allow directories representing mount points to be
3090          * deleted, even if empty.  Check write perms on mount point
3091          * in case the vnode is aliased (aka nullfs).
3092          */
3093         if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3094                 return (EINVAL);
3095         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3096                 return (error);
3097
3098         error = VOP_NRMDIR(&nd->nl_nch, nd->nl_cred);
3099         return (error);
3100 }
3101
3102 /*
3103  * rmdir_args(char *path)
3104  *
3105  * Remove a directory file.
3106  */
3107 /* ARGSUSED */
3108 int
3109 sys_rmdir(struct rmdir_args *uap)
3110 {
3111         struct nlookupdata nd;
3112         int error;
3113
3114         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3115         if (error == 0)
3116                 error = kern_rmdir(&nd);
3117         nlookup_done(&nd);
3118         return (error);
3119 }
3120
3121 int
3122 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3123     enum uio_seg direction)
3124 {
3125         struct thread *td = curthread;
3126         struct proc *p = td->td_proc;
3127         struct vnode *vp;
3128         struct file *fp;
3129         struct uio auio;
3130         struct iovec aiov;
3131         long loff;
3132         int error, eofflag;
3133
3134         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3135                 return (error);
3136         if ((fp->f_flag & FREAD) == 0) {
3137                 error = EBADF;
3138                 goto done;
3139         }
3140         vp = (struct vnode *)fp->f_data;
3141 unionread:
3142         if (vp->v_type != VDIR) {
3143                 error = EINVAL;
3144                 goto done;
3145         }
3146         aiov.iov_base = buf;
3147         aiov.iov_len = count;
3148         auio.uio_iov = &aiov;
3149         auio.uio_iovcnt = 1;
3150         auio.uio_rw = UIO_READ;
3151         auio.uio_segflg = direction;
3152         auio.uio_td = td;
3153         auio.uio_resid = count;
3154         loff = auio.uio_offset = fp->f_offset;
3155         error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3156         fp->f_offset = auio.uio_offset;
3157         if (error)
3158                 goto done;
3159         if (count == auio.uio_resid) {
3160                 if (union_dircheckp) {
3161                         error = union_dircheckp(td, &vp, fp);
3162                         if (error == -1)
3163                                 goto unionread;
3164                         if (error)
3165                                 goto done;
3166                 }
3167 #if 0
3168                 if ((vp->v_flag & VROOT) &&
3169                     (vp->v_mount->mnt_flag & MNT_UNION)) {
3170                         struct vnode *tvp = vp;
3171                         vp = vp->v_mount->mnt_vnodecovered;
3172                         vref(vp);
3173                         fp->f_data = vp;
3174                         fp->f_offset = 0;
3175                         vrele(tvp);
3176                         goto unionread;
3177                 }
3178 #endif
3179         }
3180         if (basep) {
3181                 *basep = loff;
3182         }
3183         *res = count - auio.uio_resid;
3184 done:
3185         fdrop(fp);
3186         return (error);
3187 }
3188
3189 /*
3190  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3191  *
3192  * Read a block of directory entries in a file system independent format.
3193  */
3194 int
3195 sys_getdirentries(struct getdirentries_args *uap)
3196 {
3197         long base;
3198         int error;
3199
3200         error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3201             &uap->sysmsg_result, UIO_USERSPACE);
3202
3203         if (error == 0)
3204                 error = copyout(&base, uap->basep, sizeof(*uap->basep));
3205         return (error);
3206 }
3207
3208 /*
3209  * getdents_args(int fd, char *buf, size_t count)
3210  */
3211 int
3212 sys_getdents(struct getdents_args *uap)
3213 {
3214         int error;
3215
3216         error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3217             &uap->sysmsg_result, UIO_USERSPACE);
3218
3219         return (error);
3220 }
3221
3222 /*
3223  * umask(int newmask)
3224  *
3225  * Set the mode mask for creation of filesystem nodes.
3226  *
3227  * MP SAFE
3228  */
3229 int
3230 sys_umask(struct umask_args *uap)
3231 {
3232         struct thread *td = curthread;
3233         struct proc *p = td->td_proc;
3234         struct filedesc *fdp;
3235
3236         fdp = p->p_fd;
3237         uap->sysmsg_result = fdp->fd_cmask;
3238         fdp->fd_cmask = uap->newmask & ALLPERMS;
3239         return (0);
3240 }
3241
3242 /*
3243  * revoke(char *path)
3244  *
3245  * Void all references to file by ripping underlying filesystem
3246  * away from vnode.
3247  */
3248 /* ARGSUSED */
3249 int
3250 sys_revoke(struct revoke_args *uap)
3251 {
3252         struct nlookupdata nd;
3253         struct vattr vattr;
3254         struct vnode *vp;
3255         struct ucred *cred;
3256         int error;
3257
3258         vp = NULL;
3259         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3260         if (error == 0)
3261                 error = nlookup(&nd);
3262         if (error == 0)
3263                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3264         cred = crhold(nd.nl_cred);
3265         nlookup_done(&nd);
3266         if (error == 0) {
3267                 if (vp->v_type != VCHR && vp->v_type != VBLK)
3268                         error = EINVAL;
3269                 if (error == 0)
3270                         error = VOP_GETATTR(vp, &vattr);
3271                 if (error == 0 && cred->cr_uid != vattr.va_uid)
3272                         error = suser_cred(cred, PRISON_ROOT);
3273                 if (error == 0 && count_udev(vp->v_umajor, vp->v_uminor) > 0) {
3274                         error = 0;
3275                         vx_lock(vp);
3276                         VOP_REVOKE(vp, REVOKEALL);
3277                         vx_unlock(vp);
3278                 }
3279                 vrele(vp);
3280         }
3281         if (cred)
3282                 crfree(cred);
3283         return (error);
3284 }
3285
3286 /*
3287  * getfh_args(char *fname, fhandle_t *fhp)
3288  *
3289  * Get (NFS) file handle
3290  */
3291 int
3292 sys_getfh(struct getfh_args *uap)
3293 {
3294         struct thread *td = curthread;
3295         struct nlookupdata nd;
3296         fhandle_t fh;
3297         struct vnode *vp;
3298         int error;
3299
3300         /*
3301          * Must be super user
3302          */
3303         if ((error = suser(td)) != 0)
3304                 return (error);
3305
3306         vp = NULL;
3307         error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3308         if (error == 0)
3309                 error = nlookup(&nd);
3310         if (error == 0)
3311                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3312         nlookup_done(&nd);
3313         if (error == 0) {
3314                 bzero(&fh, sizeof(fh));
3315                 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3316                 error = VFS_VPTOFH(vp, &fh.fh_fid);
3317                 vput(vp);
3318                 if (error == 0)
3319                         error = copyout(&fh, uap->fhp, sizeof(fh));
3320         }
3321         return (error);
3322 }
3323
3324 /*
3325  * fhopen_args(const struct fhandle *u_fhp, int flags)
3326  *
3327  * syscall for the rpc.lockd to use to translate a NFS file handle into
3328  * an open descriptor.
3329  *
3330  * warning: do not remove the suser() call or this becomes one giant
3331  * security hole.
3332  */
3333 int
3334 sys_fhopen(struct fhopen_args *uap)
3335 {
3336         struct thread *td = curthread;
3337         struct proc *p = td->td_proc;
3338         struct mount *mp;
3339         struct vnode *vp;
3340         struct fhandle fhp;
3341         struct vattr vat;
3342         struct vattr *vap = &vat;
3343         struct flock lf;
3344         int fmode, mode, error, type;
3345         struct file *nfp;
3346         struct file *fp;
3347         int indx;
3348
3349         /*
3350          * Must be super user
3351          */
3352         error = suser(td);
3353         if (error)
3354                 return (error);
3355
3356         fmode = FFLAGS(uap->flags);
3357         /* why not allow a non-read/write open for our lockd? */
3358         if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3359                 return (EINVAL);
3360         error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3361         if (error)
3362                 return(error);
3363         /* find the mount point */
3364         mp = vfs_getvfs(&fhp.fh_fsid);
3365         if (mp == NULL)
3366                 return (ESTALE);
3367         /* now give me my vnode, it gets returned to me locked */
3368         error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3369         if (error)
3370                 return (error);
3371         /*
3372          * from now on we have to make sure not
3373          * to forget about the vnode
3374          * any error that causes an abort must vput(vp)
3375          * just set error = err and 'goto bad;'.
3376          */
3377
3378         /*
3379          * from vn_open
3380          */
3381         if (vp->v_type == VLNK) {
3382                 error = EMLINK;
3383                 goto bad;
3384         }
3385         if (vp->v_type == VSOCK) {
3386                 error = EOPNOTSUPP;
3387                 goto bad;
3388         }
3389         mode = 0;
3390         if (fmode & (FWRITE | O_TRUNC)) {
3391                 if (vp->v_type == VDIR) {
3392                         error = EISDIR;
3393                         goto bad;
3394                 }
3395                 error = vn_writechk(vp, NULL);
3396                 if (error)
3397                         goto bad;
3398                 mode |= VWRITE;
3399         }
3400         if (fmode & FREAD)
3401                 mode |= VREAD;
3402         if (mode) {
3403                 error = VOP_ACCESS(vp, mode, p->p_ucred);
3404                 if (error)
3405                         goto bad;
3406         }
3407         if (fmode & O_TRUNC) {
3408                 vn_unlock(vp);                          /* XXX */
3409                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
3410                 VATTR_NULL(vap);
3411                 vap->va_size = 0;
3412                 error = VOP_SETATTR(vp, vap, p->p_ucred);
3413                 if (error)
3414                         goto bad;
3415         }
3416
3417         /*
3418          * VOP_OPEN needs the file pointer so it can potentially override
3419          * it.
3420          *
3421          * WARNING! no f_nchandle will be associated when fhopen()ing a
3422          * directory.  XXX
3423          */
3424         if ((error = falloc(p, &nfp, &indx)) != 0)
3425                 goto bad;
3426         fp = nfp;
3427
3428         error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3429         if (error) {
3430                 /*
3431                  * setting f_ops this way prevents VOP_CLOSE from being
3432                  * called or fdrop() releasing the vp from v_data.   Since
3433                  * the VOP_OPEN failed we don't want to VOP_CLOSE.
3434                  */
3435                 fp->f_ops = &badfileops;
3436                 fp->f_data = NULL;
3437                 goto bad_drop;
3438         }
3439
3440         /*
3441          * The fp is given its own reference, we still have our ref and lock.
3442          *
3443          * Assert that all regular files must be created with a VM object.
3444          */
3445         if (vp->v_type == VREG && vp->v_object == NULL) {
3446                 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3447                 goto bad_drop;
3448         }
3449
3450         /*
3451          * The open was successful.  Handle any locking requirements.
3452          */
3453         if (fmode & (O_EXLOCK | O_SHLOCK)) {
3454                 lf.l_whence = SEEK_SET;
3455                 lf.l_start = 0;
3456                 lf.l_len = 0;
3457                 if (fmode & O_EXLOCK)
3458                         lf.l_type = F_WRLCK;
3459                 else
3460                         lf.l_type = F_RDLCK;
3461                 if (fmode & FNONBLOCK)
3462                         type = 0;
3463                 else
3464                         type = F_WAIT;
3465                 vn_unlock(vp);
3466                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3467                         /*
3468                          * release our private reference.
3469                          */
3470                         fsetfd(p, NULL, indx);
3471                         fdrop(fp);
3472                         vrele(vp);
3473                         return (error);
3474                 }
3475                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3476                 fp->f_flag |= FHASLOCK;
3477         }
3478
3479         /*
3480          * Clean up.  Associate the file pointer with the previously
3481          * reserved descriptor and return it.
3482          */
3483         vput(vp);
3484         fsetfd(p, fp, indx);
3485         fdrop(fp);
3486         uap->sysmsg_result = indx;
3487         return (0);
3488
3489 bad_drop:
3490         fsetfd(p, NULL, indx);
3491         fdrop(fp);
3492 bad:
3493         vput(vp);
3494         return (error);
3495 }
3496
3497 /*
3498  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3499  */
3500 int
3501 sys_fhstat(struct fhstat_args *uap)
3502 {
3503         struct thread *td = curthread;
3504         struct stat sb;
3505         fhandle_t fh;
3506         struct mount *mp;
3507         struct vnode *vp;
3508         int error;
3509
3510         /*
3511          * Must be super user
3512          */
3513         error = suser(td);
3514         if (error)
3515                 return (error);
3516
3517         error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3518         if (error)
3519                 return (error);
3520
3521         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3522                 return (ESTALE);
3523         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3524                 return (error);
3525         error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3526         vput(vp);
3527         if (error)
3528                 return (error);
3529         error = copyout(&sb, uap->sb, sizeof(sb));
3530         return (error);
3531 }
3532
3533 /*
3534  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3535  */
3536 int
3537 sys_fhstatfs(struct fhstatfs_args *uap)
3538 {
3539         struct thread *td = curthread;
3540         struct proc *p = td->td_proc;
3541         struct statfs *sp;
3542         struct mount *mp;
3543         struct vnode *vp;
3544         struct statfs sb;
3545         char *fullpath, *freepath;
3546         fhandle_t fh;
3547         int error;
3548
3549         /*
3550          * Must be super user
3551          */
3552         if ((error = suser(td)))
3553                 return (error);
3554
3555         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3556                 return (error);
3557
3558         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3559                 return (ESTALE);
3560
3561         if (p != NULL && !chroot_visible_mnt(mp, p))
3562                 return (ESTALE);
3563
3564         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3565                 return (error);
3566         mp = vp->v_mount;
3567         sp = &mp->mnt_stat;
3568         vput(vp);
3569         if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3570                 return (error);
3571
3572         error = mount_path(p, mp, &fullpath, &freepath);
3573         if (error)
3574                 return(error);
3575         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3576         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3577         kfree(freepath, M_TEMP);
3578
3579         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3580         if (suser(td)) {
3581                 bcopy(sp, &sb, sizeof(sb));
3582                 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3583                 sp = &sb;
3584         }
3585         return (copyout(sp, uap->buf, sizeof(*sp)));
3586 }
3587
3588 /*
3589  * Syscall to push extended attribute configuration information into the
3590  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3591  * a command (int cmd), and attribute name and misc data.  For now, the
3592  * attribute name is left in userspace for consumption by the VFS_op.
3593  * It will probably be changed to be copied into sysspace by the
3594  * syscall in the future, once issues with various consumers of the
3595  * attribute code have raised their hands.
3596  *
3597  * Currently this is used only by UFS Extended Attributes.
3598  */
3599 int
3600 sys_extattrctl(struct extattrctl_args *uap)
3601 {
3602         struct nlookupdata nd;
3603         struct mount *mp;
3604         struct vnode *vp;
3605         int error;
3606
3607         vp = NULL;
3608         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3609         if (error == 0)
3610                 error = nlookup(&nd);
3611         if (error == 0) {
3612                 mp = nd.nl_nch.mount;
3613                 error = VFS_EXTATTRCTL(mp, uap->cmd,
3614                                 uap->attrname, uap->arg,
3615                                 nd.nl_cred);
3616         }
3617         nlookup_done(&nd);
3618         return (error);
3619 }
3620
3621 /*
3622  * Syscall to set a named extended attribute on a file or directory.
3623  * Accepts attribute name, and a uio structure pointing to the data to set.
3624  * The uio is consumed in the style of writev().  The real work happens
3625  * in VOP_SETEXTATTR().
3626  */
3627 int
3628 sys_extattr_set_file(struct extattr_set_file_args *uap)
3629 {
3630         char attrname[EXTATTR_MAXNAMELEN];
3631         struct iovec aiov[UIO_SMALLIOV];
3632         struct iovec *needfree;
3633         struct nlookupdata nd;
3634         struct iovec *iov;
3635         struct vnode *vp;
3636         struct uio auio;
3637         u_int iovlen;
3638         u_int cnt;
3639         int error;
3640         int i;
3641
3642         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3643         if (error)
3644                 return (error);
3645
3646         vp = NULL;
3647         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3648         if (error == 0)
3649                 error = nlookup(&nd);
3650         if (error == 0)
3651                 error = ncp_writechk(&nd.nl_nch);
3652         if (error == 0)
3653                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3654         if (error) {
3655                 nlookup_done(&nd);
3656                 return (error);
3657         }
3658
3659         needfree = NULL;
3660         iovlen = uap->iovcnt * sizeof(struct iovec);
3661         if (uap->iovcnt > UIO_SMALLIOV) {
3662                 if (uap->iovcnt > UIO_MAXIOV) {
3663                         error = EINVAL;
3664                         goto done;
3665                 }
3666                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3667                 needfree = iov;
3668         } else {
3669                 iov = aiov;
3670         }
3671         auio.uio_iov = iov;
3672         auio.uio_iovcnt = uap->iovcnt;
3673         auio.uio_rw = UIO_WRITE;
3674         auio.uio_segflg = UIO_USERSPACE;
3675         auio.uio_td = nd.nl_td;
3676         auio.uio_offset = 0;
3677         if ((error = copyin(uap->iovp, iov, iovlen)))
3678                 goto done;
3679         auio.uio_resid = 0;
3680         for (i = 0; i < uap->iovcnt; i++) {
3681                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3682                         error = EINVAL;
3683                         goto done;
3684                 }
3685                 auio.uio_resid += iov->iov_len;
3686                 iov++;
3687         }
3688         cnt = auio.uio_resid;
3689         error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3690         cnt -= auio.uio_resid;
3691         uap->sysmsg_result = cnt;
3692 done:
3693         vput(vp);
3694         nlookup_done(&nd);
3695         if (needfree)
3696                 FREE(needfree, M_IOV);
3697         return (error);
3698 }
3699
3700 /*
3701  * Syscall to get a named extended attribute on a file or directory.
3702  * Accepts attribute name, and a uio structure pointing to a buffer for the
3703  * data.  The uio is consumed in the style of readv().  The real work
3704  * happens in VOP_GETEXTATTR();
3705  */
3706 int
3707 sys_extattr_get_file(struct extattr_get_file_args *uap)
3708 {
3709         char attrname[EXTATTR_MAXNAMELEN];
3710         struct iovec aiov[UIO_SMALLIOV];
3711         struct iovec *needfree;
3712         struct nlookupdata nd;
3713         struct iovec *iov;
3714         struct vnode *vp;
3715         struct uio auio;
3716         u_int iovlen;
3717         u_int cnt;
3718         int error;
3719         int i;
3720
3721         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3722         if (error)
3723                 return (error);
3724
3725         vp = NULL;
3726         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3727         if (error == 0)
3728                 error = nlookup(&nd);
3729         if (error == 0)
3730                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3731         if (error) {
3732                 nlookup_done(&nd);
3733                 return (error);
3734         }
3735
3736         iovlen = uap->iovcnt * sizeof (struct iovec);
3737         needfree = NULL;
3738         if (uap->iovcnt > UIO_SMALLIOV) {
3739                 if (uap->iovcnt > UIO_MAXIOV) {
3740                         error = EINVAL;
3741                         goto done;
3742                 }
3743                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3744                 needfree = iov;
3745         } else {
3746                 iov = aiov;
3747         }
3748         auio.uio_iov = iov;
3749         auio.uio_iovcnt = uap->iovcnt;
3750         auio.uio_rw = UIO_READ;
3751         auio.uio_segflg = UIO_USERSPACE;
3752         auio.uio_td = nd.nl_td;
3753         auio.uio_offset = 0;
3754         if ((error = copyin(uap->iovp, iov, iovlen)))
3755                 goto done;
3756         auio.uio_resid = 0;
3757         for (i = 0; i < uap->iovcnt; i++) {
3758                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3759                         error = EINVAL;
3760                         goto done;
3761                 }
3762                 auio.uio_resid += iov->iov_len;
3763                 iov++;
3764         }
3765         cnt = auio.uio_resid;
3766         error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3767         cnt -= auio.uio_resid;
3768         uap->sysmsg_result = cnt;
3769 done:
3770         vput(vp);
3771         nlookup_done(&nd);
3772         if (needfree)
3773                 FREE(needfree, M_IOV);
3774         return(error);
3775 }
3776
3777 /*
3778  * Syscall to delete a named extended attribute from a file or directory.
3779  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3780  */
3781 int
3782 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3783 {
3784         char attrname[EXTATTR_MAXNAMELEN];
3785         struct nlookupdata nd;
3786         struct vnode *vp;
3787         int error;
3788
3789         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3790         if (error)
3791                 return(error);
3792
3793         vp = NULL;
3794         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3795         if (error == 0)
3796                 error = nlookup(&nd);
3797         if (error == 0)
3798                 error = ncp_writechk(&nd.nl_nch);
3799         if (error == 0)
3800                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3801         if (error) {
3802                 nlookup_done(&nd);
3803                 return (error);
3804         }
3805
3806         error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3807         vput(vp);
3808         nlookup_done(&nd);
3809         return(error);
3810 }
3811
3812 /*
3813  * Determine if the mount is visible to the process.
3814  */
3815 static int
3816 chroot_visible_mnt(struct mount *mp, struct proc *p)
3817 {
3818         struct nchandle nch;
3819
3820         /*
3821          * Traverse from the mount point upwards.  If we hit the process
3822          * root then the mount point is visible to the process.
3823          */
3824         nch = mp->mnt_ncmountpt;
3825         while (nch.ncp) {
3826                 if (nch.mount == p->p_fd->fd_nrdir.mount &&
3827                     nch.ncp == p->p_fd->fd_nrdir.ncp) {
3828                         return(1);
3829                 }
3830                 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
3831                         nch = nch.mount->mnt_ncmounton;
3832                 } else {
3833                         nch.ncp = nch.ncp->nc_parent;
3834                 }
3835         }
3836
3837         /*
3838          * If the mount point is not visible to the process, but the
3839          * process root is in a subdirectory of the mount, return
3840          * TRUE anyway.
3841          */
3842         if (p->p_fd->fd_nrdir.mount == mp)
3843                 return(1);
3844
3845         return(0);
3846 }
3847