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 1.116 2007/05/09 00:53:34 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         fp = holdfp(p->p_fd, uap->fd, -1);
 930         if (fp == NULL) {
 931                 error = EBADF;
 932                 goto done;
 933         }
 934
 935         /*
 936          * Execute the internal kernel function and clean up.
 937          */
 938         error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
 939         if (fp)
 940                 fdrop(fp);
 941         if (error == 0 && uap->sysmsg_result > 0)
 942                 error = copyout(buf, uap->buf, uap->sysmsg_result);
 943 done:
 944         if (path)
 945                 objcache_put(namei_oc, path);
 946         if (ctl)
 947                 kfree(ctl, M_TEMP);
 948         if (buf)
 949                 kfree(buf, M_TEMP);
 950         return (error);
 951 }
 952
 953 /*
 954  * Execute a mount control operation by resolving the path to a mount point
 955  * and calling vop_mountctl().
 956  */
 957 int
 958 kern_mountctl(const char *path, int op, struct file *fp,
 959                 const void *ctl, int ctllen,
 960                 void *buf, int buflen, int *res)
 961 {
 962         struct vnode *vp;
 963         struct mount *mp;
 964         struct nlookupdata nd;
 965         int error;
 966
 967         *res = 0;
 968         vp = NULL;
 969         error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
 970         if (error == 0)
 971                 error = nlookup(&nd);
 972         if (error == 0)
 973                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 974         nlookup_done(&nd);
 975         if (error)
 976                 return (error);
 977
 978         mp = vp->v_mount;
 979
 980         /*
 981          * Must be the root of the filesystem
 982          */
 983         if ((vp->v_flag & VROOT) == 0) {
 984                 vput(vp);
 985                 return (EINVAL);
 986         }
 987         error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
 988                                 buf, buflen, res);
 989         vput(vp);
 990         return (error);
 991 }
 992
 993 int
 994 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
 995 {
 996         struct thread *td = curthread;
 997         struct proc *p = td->td_proc;
 998         struct mount *mp;
 999         struct statfs *sp;
1000         char *fullpath, *freepath;
1001         int error;
1002
1003         if ((error = nlookup(nd)) != 0)
1004                 return (error);
1005         mp = nd->nl_nch.mount;
1006         sp = &mp->mnt_stat;
1007         if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1008                 return (error);
1009
1010         error = mount_path(p, mp, &fullpath, &freepath);
1011         if (error)
1012                 return(error);
1013         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1014         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1015         kfree(freepath, M_TEMP);
1016
1017         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1018         bcopy(sp, buf, sizeof(*buf));
1019         /* Only root should have access to the fsid's. */
1020         if (suser(td))
1021                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1022         return (0);
1023 }
1024
1025 /*
1026  * statfs_args(char *path, struct statfs *buf)
1027  *
1028  * Get filesystem statistics.
1029  */
1030 int
1031 sys_statfs(struct statfs_args *uap)
1032 {
1033         struct nlookupdata nd;
1034         struct statfs buf;
1035         int error;
1036
1037         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1038         if (error == 0)
1039                 error = kern_statfs(&nd, &buf);
1040         nlookup_done(&nd);
1041         if (error == 0)
1042                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1043         return (error);
1044 }
1045
1046 int
1047 kern_fstatfs(int fd, struct statfs *buf)
1048 {
1049         struct thread *td = curthread;
1050         struct proc *p = td->td_proc;
1051         struct file *fp;
1052         struct mount *mp;
1053         struct statfs *sp;
1054         char *fullpath, *freepath;
1055         int error;
1056
1057         KKASSERT(p);
1058         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1059                 return (error);
1060         mp = ((struct vnode *)fp->f_data)->v_mount;
1061         if (mp == NULL) {
1062                 error = EBADF;
1063                 goto done;
1064         }
1065         if (fp->f_cred == NULL) {
1066                 error = EINVAL;
1067                 goto done;
1068         }
1069         sp = &mp->mnt_stat;
1070         if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1071                 goto done;
1072
1073         if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1074                 goto done;
1075         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1076         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1077         kfree(freepath, M_TEMP);
1078
1079         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1080         bcopy(sp, buf, sizeof(*buf));
1081
1082         /* Only root should have access to the fsid's. */
1083         if (suser(td))
1084                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1085         error = 0;
1086 done:
1087         fdrop(fp);
1088         return (error);
1089 }
1090
1091 /*
1092  * fstatfs_args(int fd, struct statfs *buf)
1093  *
1094  * Get filesystem statistics.
1095  */
1096 int
1097 sys_fstatfs(struct fstatfs_args *uap)
1098 {
1099         struct statfs buf;
1100         int error;
1101
1102         error = kern_fstatfs(uap->fd, &buf);
1103
1104         if (error == 0)
1105                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1106         return (error);
1107 }
1108
1109 /*
1110  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1111  *
1112  * Get statistics on all filesystems.
1113  */
1114
1115 struct getfsstat_info {
1116         struct statfs *sfsp;
1117         long count;
1118         long maxcount;
1119         int error;
1120         int flags;
1121         struct proc *p;
1122 };
1123
1124 static int getfsstat_callback(struct mount *, void *);
1125
1126 /* ARGSUSED */
1127 int
1128 sys_getfsstat(struct getfsstat_args *uap)
1129 {
1130         struct thread *td = curthread;
1131         struct proc *p = td->td_proc;
1132         struct getfsstat_info info;
1133
1134         bzero(&info, sizeof(info));
1135
1136         info.maxcount = uap->bufsize / sizeof(struct statfs);
1137         info.sfsp = uap->buf;
1138         info.count = 0;
1139         info.flags = uap->flags;
1140         info.p = p;
1141
1142         mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1143         if (info.sfsp && info.count > info.maxcount)
1144                 uap->sysmsg_result = info.maxcount;
1145         else
1146                 uap->sysmsg_result = info.count;
1147         return (info.error);
1148 }
1149
1150 static int
1151 getfsstat_callback(struct mount *mp, void *data)
1152 {
1153         struct getfsstat_info *info = data;
1154         struct statfs *sp;
1155         char *freepath;
1156         char *fullpath;
1157         int error;
1158
1159         if (info->sfsp && info->count < info->maxcount) {
1160                 if (info->p && !chroot_visible_mnt(mp, info->p))
1161                         return(0);
1162                 sp = &mp->mnt_stat;
1163
1164                 /*
1165                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1166                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1167                  * overrides MNT_WAIT.
1168                  */
1169                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1170                     (info->flags & MNT_WAIT)) &&
1171                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1172                         return(0);
1173                 }
1174                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1175
1176                 error = mount_path(info->p, mp, &fullpath, &freepath);
1177                 if (error) {
1178                         info->error = error;
1179                         return(-1);
1180                 }
1181                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1182                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1183                 kfree(freepath, M_TEMP);
1184
1185                 error = copyout(sp, info->sfsp, sizeof(*sp));
1186                 if (error) {
1187                         info->error = error;
1188                         return (-1);
1189                 }
1190                 ++info->sfsp;
1191         }
1192         info->count++;
1193         return(0);
1194 }
1195
1196 /*
1197  * fchdir_args(int fd)
1198  *
1199  * Change current working directory to a given file descriptor.
1200  */
1201 /* ARGSUSED */
1202 int
1203 sys_fchdir(struct fchdir_args *uap)
1204 {
1205         struct thread *td = curthread;
1206         struct proc *p = td->td_proc;
1207         struct filedesc *fdp = p->p_fd;
1208         struct vnode *vp, *ovp;
1209         struct mount *mp;
1210         struct file *fp;
1211         struct nchandle nch, onch, tnch;
1212         int error;
1213
1214         if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1215                 return (error);
1216         vp = (struct vnode *)fp->f_data;
1217         vref(vp);
1218         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1219         if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1220                 error = ENOTDIR;
1221         else
1222                 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1223         if (error) {
1224                 vput(vp);
1225                 fdrop(fp);
1226                 return (error);
1227         }
1228         cache_copy(&fp->f_nchandle, &nch);
1229
1230         /*
1231          * If the ncp has become a mount point, traverse through
1232          * the mount point.
1233          */
1234
1235         while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1236                (mp = cache_findmount(&nch)) != NULL
1237         ) {
1238                 error = nlookup_mp(mp, &tnch);
1239                 if (error == 0) {
1240                         cache_unlock(&tnch);    /* leave ref intact */
1241                         vput(vp);
1242                         vp = tnch.ncp->nc_vp;
1243                         error = vget(vp, LK_SHARED);
1244                         KKASSERT(error == 0);
1245                         cache_drop(&nch);
1246                         nch = tnch;
1247                 }
1248         }
1249         if (error == 0) {
1250                 ovp = fdp->fd_cdir;
1251                 onch = fdp->fd_ncdir;
1252                 vn_unlock(vp);          /* leave ref intact */
1253                 fdp->fd_cdir = vp;
1254                 fdp->fd_ncdir = nch;
1255                 cache_drop(&onch);
1256                 vrele(ovp);
1257         } else {
1258                 cache_drop(&nch);
1259                 vput(vp);
1260         }
1261         fdrop(fp);
1262         return (error);
1263 }
1264
1265 int
1266 kern_chdir(struct nlookupdata *nd)
1267 {
1268         struct thread *td = curthread;
1269         struct proc *p = td->td_proc;
1270         struct filedesc *fdp = p->p_fd;
1271         struct vnode *vp, *ovp;
1272         struct nchandle onch;
1273         int error;
1274
1275         if ((error = nlookup(nd)) != 0)
1276                 return (error);
1277         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1278                 return (ENOENT);
1279         if ((error = vget(vp, LK_SHARED)) != 0)
1280                 return (error);
1281
1282         error = checkvp_chdir(vp, td);
1283         vn_unlock(vp);
1284         if (error == 0) {
1285                 ovp = fdp->fd_cdir;
1286                 onch = fdp->fd_ncdir;
1287                 cache_unlock(&nd->nl_nch);      /* leave reference intact */
1288                 fdp->fd_ncdir = nd->nl_nch;
1289                 fdp->fd_cdir = vp;
1290                 cache_drop(&onch);
1291                 vrele(ovp);
1292                 cache_zero(&nd->nl_nch);
1293         } else {
1294                 vrele(vp);
1295         }
1296         return (error);
1297 }
1298
1299 /*
1300  * chdir_args(char *path)
1301  *
1302  * Change current working directory (``.'').
1303  */
1304 int
1305 sys_chdir(struct chdir_args *uap)
1306 {
1307         struct nlookupdata nd;
1308         int error;
1309
1310         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1311         if (error == 0)
1312                 error = kern_chdir(&nd);
1313         nlookup_done(&nd);
1314         return (error);
1315 }
1316
1317 /*
1318  * Helper function for raised chroot(2) security function:  Refuse if
1319  * any filedescriptors are open directories.
1320  */
1321 static int
1322 chroot_refuse_vdir_fds(struct filedesc *fdp)
1323 {
1324         struct vnode *vp;
1325         struct file *fp;
1326         int error;
1327         int fd;
1328
1329         for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1330                 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1331                         continue;
1332                 vp = (struct vnode *)fp->f_data;
1333                 if (vp->v_type != VDIR) {
1334                         fdrop(fp);
1335                         continue;
1336                 }
1337                 fdrop(fp);
1338                 return(EPERM);
1339         }
1340         return (0);
1341 }
1342
1343 /*
1344  * This sysctl determines if we will allow a process to chroot(2) if it
1345  * has a directory open:
1346  *      0: disallowed for all processes.
1347  *      1: allowed for processes that were not already chroot(2)'ed.
1348  *      2: allowed for all processes.
1349  */
1350
1351 static int chroot_allow_open_directories = 1;
1352
1353 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1354      &chroot_allow_open_directories, 0, "");
1355
1356 /*
1357  * chroot to the specified namecache entry.  We obtain the vp from the
1358  * namecache data.  The passed ncp must be locked and referenced and will
1359  * remain locked and referenced on return.
1360  */
1361 int
1362 kern_chroot(struct nchandle *nch)
1363 {
1364         struct thread *td = curthread;
1365         struct proc *p = td->td_proc;
1366         struct filedesc *fdp = p->p_fd;
1367         struct vnode *vp;
1368         int error;
1369
1370         /*
1371          * Only root can chroot
1372          */
1373         if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1374                 return (error);
1375
1376         /*
1377          * Disallow open directory descriptors (fchdir() breakouts).
1378          */
1379         if (chroot_allow_open_directories == 0 ||
1380            (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1381                 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1382                         return (error);
1383         }
1384         if ((vp = nch->ncp->nc_vp) == NULL)
1385                 return (ENOENT);
1386
1387         if ((error = vget(vp, LK_SHARED)) != 0)
1388                 return (error);
1389
1390         /*
1391          * Check the validity of vp as a directory to change to and
1392          * associate it with rdir/jdir.
1393          */
1394         error = checkvp_chdir(vp, td);
1395         vn_unlock(vp);                  /* leave reference intact */
1396         if (error == 0) {
1397                 vrele(fdp->fd_rdir);
1398                 fdp->fd_rdir = vp;      /* reference inherited by fd_rdir */
1399                 cache_drop(&fdp->fd_nrdir);
1400                 cache_copy(nch, &fdp->fd_nrdir);
1401                 if (fdp->fd_jdir == NULL) {
1402                         fdp->fd_jdir = vp;
1403                         vref(fdp->fd_jdir);
1404                         cache_copy(nch, &fdp->fd_njdir);
1405                 }
1406         } else {
1407                 vrele(vp);
1408         }
1409         return (error);
1410 }
1411
1412 /*
1413  * chroot_args(char *path)
1414  *
1415  * Change notion of root (``/'') directory.
1416  */
1417 /* ARGSUSED */
1418 int
1419 sys_chroot(struct chroot_args *uap)
1420 {
1421         struct thread *td = curthread;
1422         struct nlookupdata nd;
1423         int error;
1424
1425         KKASSERT(td->td_proc);
1426         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1427         if (error) {
1428                 nlookup_done(&nd);
1429                 return(error);
1430         }
1431         error = nlookup(&nd);
1432         if (error == 0)
1433                 error = kern_chroot(&nd.nl_nch);
1434         nlookup_done(&nd);
1435         return(error);
1436 }
1437
1438 /*
1439  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1440  * determine whether it is legal to chdir to the vnode.  The vnode's state
1441  * is not changed by this call.
1442  */
1443 int
1444 checkvp_chdir(struct vnode *vp, struct thread *td)
1445 {
1446         int error;
1447
1448         if (vp->v_type != VDIR)
1449                 error = ENOTDIR;
1450         else
1451                 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1452         return (error);
1453 }
1454
1455 int
1456 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1457 {
1458         struct thread *td = curthread;
1459         struct proc *p = td->td_proc;
1460         struct lwp *lp = td->td_lwp;
1461         struct filedesc *fdp = p->p_fd;
1462         int cmode, flags;
1463         struct file *nfp;
1464         struct file *fp;
1465         struct vnode *vp;
1466         int type, indx, error;
1467         struct flock lf;
1468
1469         if ((oflags & O_ACCMODE) == O_ACCMODE)
1470                 return (EINVAL);
1471         flags = FFLAGS(oflags);
1472         error = falloc(p, &nfp, NULL);
1473         if (error)
1474                 return (error);
1475         fp = nfp;
1476         cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1477
1478         /*
1479          * XXX p_dupfd is a real mess.  It allows a device to return a
1480          * file descriptor to be duplicated rather then doing the open
1481          * itself.
1482          */
1483         lp->lwp_dupfd = -1;
1484
1485         /*
1486          * Call vn_open() to do the lookup and assign the vnode to the
1487          * file pointer.  vn_open() does not change the ref count on fp
1488          * and the vnode, on success, will be inherited by the file pointer
1489          * and unlocked.
1490          */
1491         nd->nl_flags |= NLC_LOCKVP;
1492         error = vn_open(nd, fp, flags, cmode);
1493         nlookup_done(nd);
1494         if (error) {
1495                 /*
1496                  * handle special fdopen() case.  bleh.  dupfdopen() is
1497                  * responsible for dropping the old contents of ofiles[indx]
1498                  * if it succeeds.
1499                  *
1500                  * Note that fsetfd() will add a ref to fp which represents
1501                  * the fd_files[] assignment.  We must still drop our
1502                  * reference.
1503                  */
1504                 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1505                         if (fdalloc(p, 0, &indx) == 0) {
1506                                 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1507                                 if (error == 0) {
1508                                         *res = indx;
1509                                         fdrop(fp);      /* our ref */
1510                                         return (0);
1511                                 }
1512                                 fsetfd(p, NULL, indx);
1513                         }
1514                 }
1515                 fdrop(fp);      /* our ref */
1516                 if (error == ERESTART)
1517                         error = EINTR;
1518                 return (error);
1519         }
1520
1521         /*
1522          * ref the vnode for ourselves so it can't be ripped out from under
1523          * is.  XXX need an ND flag to request that the vnode be returned
1524          * anyway.
1525          *
1526          * Reserve a file descriptor but do not assign it until the open
1527          * succeeds.
1528          */
1529         vp = (struct vnode *)fp->f_data;
1530         vref(vp);
1531         if ((error = fdalloc(p, 0, &indx)) != 0) {
1532                 fdrop(fp);
1533                 vrele(vp);
1534                 return (error);
1535         }
1536
1537         /*
1538          * If no error occurs the vp will have been assigned to the file
1539          * pointer.
1540          */
1541         lp->lwp_dupfd = 0;
1542
1543         if (flags & (O_EXLOCK | O_SHLOCK)) {
1544                 lf.l_whence = SEEK_SET;
1545                 lf.l_start = 0;
1546                 lf.l_len = 0;
1547                 if (flags & O_EXLOCK)
1548                         lf.l_type = F_WRLCK;
1549                 else
1550                         lf.l_type = F_RDLCK;
1551                 if (flags & FNONBLOCK)
1552                         type = 0;
1553                 else
1554                         type = F_WAIT;
1555
1556                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1557                         /*
1558                          * lock request failed.  Clean up the reserved
1559                          * descriptor.
1560                          */
1561                         vrele(vp);
1562                         fsetfd(p, NULL, indx);
1563                         fdrop(fp);
1564                         return (error);
1565                 }
1566                 fp->f_flag |= FHASLOCK;
1567         }
1568 #if 0
1569         /*
1570          * Assert that all regular file vnodes were created with a object.
1571          */
1572         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1573                 ("open: regular file has no backing object after vn_open"));
1574 #endif
1575
1576         vrele(vp);
1577
1578         /*
1579          * release our private reference, leaving the one associated with the
1580          * descriptor table intact.
1581          */
1582         fsetfd(p, fp, indx);
1583         fdrop(fp);
1584         *res = indx;
1585         return (0);
1586 }
1587
1588 /*
1589  * open_args(char *path, int flags, int mode)
1590  *
1591  * Check permissions, allocate an open file structure,
1592  * and call the device open routine if any.
1593  */
1594 int
1595 sys_open(struct open_args *uap)
1596 {
1597         struct nlookupdata nd;
1598         int error;
1599
1600         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1601         if (error == 0) {
1602                 error = kern_open(&nd, uap->flags,
1603                                     uap->mode, &uap->sysmsg_result);
1604         }
1605         nlookup_done(&nd);
1606         return (error);
1607 }
1608
1609 int
1610 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1611 {
1612         struct thread *td = curthread;
1613         struct proc *p = td->td_proc;
1614         struct vnode *vp;
1615         struct vattr vattr;
1616         int error;
1617         int whiteout = 0;
1618
1619         KKASSERT(p);
1620
1621         switch (mode & S_IFMT) {
1622         case S_IFCHR:
1623         case S_IFBLK:
1624                 error = suser(td);
1625                 break;
1626         default:
1627                 error = suser_cred(p->p_ucred, PRISON_ROOT);
1628                 break;
1629         }
1630         if (error)
1631                 return (error);
1632
1633         bwillwrite();
1634         nd->nl_flags |= NLC_CREATE;
1635         if ((error = nlookup(nd)) != 0)
1636                 return (error);
1637         if (nd->nl_nch.ncp->nc_vp)
1638                 return (EEXIST);
1639         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1640                 return (error);
1641
1642         VATTR_NULL(&vattr);
1643         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1644         vattr.va_rmajor = rmajor;
1645         vattr.va_rminor = rminor;
1646         whiteout = 0;
1647
1648         switch (mode & S_IFMT) {
1649         case S_IFMT:    /* used by badsect to flag bad sectors */
1650                 vattr.va_type = VBAD;
1651                 break;
1652         case S_IFCHR:
1653                 vattr.va_type = VCHR;
1654                 break;
1655         case S_IFBLK:
1656                 vattr.va_type = VBLK;
1657                 break;
1658         case S_IFWHT:
1659                 whiteout = 1;
1660                 break;
1661         default:
1662                 error = EINVAL;
1663                 break;
1664         }
1665         if (error == 0) {
1666                 if (whiteout) {
1667                         error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_cred, NAMEI_CREATE);
1668                 } else {
1669                         vp = NULL;
1670                         error = VOP_NMKNOD(&nd->nl_nch, &vp, nd->nl_cred, &vattr);
1671                         if (error == 0)
1672                                 vput(vp);
1673                 }
1674         }
1675         return (error);
1676 }
1677
1678 /*
1679  * mknod_args(char *path, int mode, int dev)
1680  *
1681  * Create a special file.
1682  */
1683 int
1684 sys_mknod(struct mknod_args *uap)
1685 {
1686         struct nlookupdata nd;
1687         int error;
1688
1689         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1690         if (error == 0) {
1691                 error = kern_mknod(&nd, uap->mode,
1692                                    umajor(uap->dev), uminor(uap->dev));
1693         }
1694         nlookup_done(&nd);
1695         return (error);
1696 }
1697
1698 int
1699 kern_mkfifo(struct nlookupdata *nd, int mode)
1700 {
1701         struct thread *td = curthread;
1702         struct proc *p = td->td_proc;
1703         struct vattr vattr;
1704         struct vnode *vp;
1705         int error;
1706
1707         bwillwrite();
1708
1709         nd->nl_flags |= NLC_CREATE;
1710         if ((error = nlookup(nd)) != 0)
1711                 return (error);
1712         if (nd->nl_nch.ncp->nc_vp)
1713                 return (EEXIST);
1714         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1715                 return (error);
1716
1717         VATTR_NULL(&vattr);
1718         vattr.va_type = VFIFO;
1719         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1720         vp = NULL;
1721         error = VOP_NMKNOD(&nd->nl_nch, &vp, nd->nl_cred, &vattr);
1722         if (error == 0)
1723                 vput(vp);
1724         return (error);
1725 }
1726
1727 /*
1728  * mkfifo_args(char *path, int mode)
1729  *
1730  * Create a named pipe.
1731  */
1732 int
1733 sys_mkfifo(struct mkfifo_args *uap)
1734 {
1735         struct nlookupdata nd;
1736         int error;
1737
1738         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1739         if (error == 0)
1740                 error = kern_mkfifo(&nd, uap->mode);
1741         nlookup_done(&nd);
1742         return (error);
1743 }
1744
1745 static int hardlink_check_uid = 0;
1746 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1747     &hardlink_check_uid, 0,
1748     "Unprivileged processes cannot create hard links to files owned by other "
1749     "users");
1750 static int hardlink_check_gid = 0;
1751 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1752     &hardlink_check_gid, 0,
1753     "Unprivileged processes cannot create hard links to files owned by other "
1754     "groups");
1755
1756 static int
1757 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1758 {
1759         struct vattr va;
1760         int error;
1761
1762         /*
1763          * Shortcut if disabled
1764          */
1765         if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1766                 return (0);
1767
1768         /*
1769          * root cred can always hardlink
1770          */
1771         if (suser_cred(cred, PRISON_ROOT) == 0)
1772                 return (0);
1773
1774         /*
1775          * Otherwise only if the originating file is owned by the
1776          * same user or group.  Note that any group is allowed if
1777          * the file is owned by the caller.
1778          */
1779         error = VOP_GETATTR(vp, &va);
1780         if (error != 0)
1781                 return (error);
1782
1783         if (hardlink_check_uid) {
1784                 if (cred->cr_uid != va.va_uid)
1785                         return (EPERM);
1786         }
1787
1788         if (hardlink_check_gid) {
1789                 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1790                         return (EPERM);
1791         }
1792
1793         return (0);
1794 }
1795
1796 int
1797 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1798 {
1799         struct thread *td = curthread;
1800         struct vnode *vp;
1801         int error;
1802
1803         /*
1804          * Lookup the source and obtained a locked vnode.
1805          *
1806          * XXX relookup on vget failure / race ?
1807          */
1808         bwillwrite();
1809         if ((error = nlookup(nd)) != 0)
1810                 return (error);
1811         vp = nd->nl_nch.ncp->nc_vp;
1812         KKASSERT(vp != NULL);
1813         if (vp->v_type == VDIR)
1814                 return (EPERM);         /* POSIX */
1815         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1816                 return (error);
1817         if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1818                 return (error);
1819
1820         /*
1821          * Unlock the source so we can lookup the target without deadlocking
1822          * (XXX vp is locked already, possible other deadlock?).  The target
1823          * must not exist.
1824          */
1825         KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1826         nd->nl_flags &= ~NLC_NCPISLOCKED;
1827         cache_unlock(&nd->nl_nch);
1828
1829         linknd->nl_flags |= NLC_CREATE;
1830         if ((error = nlookup(linknd)) != 0) {
1831                 vput(vp);
1832                 return (error);
1833         }
1834         if (linknd->nl_nch.ncp->nc_vp) {
1835                 vput(vp);
1836                 return (EEXIST);
1837         }
1838
1839         /*
1840          * Finally run the new API VOP.
1841          */
1842         error = can_hardlink(vp, td, td->td_proc->p_ucred);
1843         if (error == 0)
1844                 error = VOP_NLINK(&linknd->nl_nch, vp, linknd->nl_cred);
1845         vput(vp);
1846         return (error);
1847 }
1848
1849 /*
1850  * link_args(char *path, char *link)
1851  *
1852  * Make a hard file link.
1853  */
1854 int
1855 sys_link(struct link_args *uap)
1856 {
1857         struct nlookupdata nd, linknd;
1858         int error;
1859
1860         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1861         if (error == 0) {
1862                 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1863                 if (error == 0)
1864                         error = kern_link(&nd, &linknd);
1865                 nlookup_done(&linknd);
1866         }
1867         nlookup_done(&nd);
1868         return (error);
1869 }
1870
1871 int
1872 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1873 {
1874         struct vattr vattr;
1875         struct vnode *vp;
1876         int error;
1877
1878         bwillwrite();
1879         nd->nl_flags |= NLC_CREATE;
1880         if ((error = nlookup(nd)) != 0)
1881                 return (error);
1882         if (nd->nl_nch.ncp->nc_vp)
1883                 return (EEXIST);
1884         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1885                 return (error);
1886         VATTR_NULL(&vattr);
1887         vattr.va_mode = mode;
1888         error = VOP_NSYMLINK(&nd->nl_nch, &vp, nd->nl_cred, &vattr, path);
1889         if (error == 0)
1890                 vput(vp);
1891         return (error);
1892 }
1893
1894 /*
1895  * symlink(char *path, char *link)
1896  *
1897  * Make a symbolic link.
1898  */
1899 int
1900 sys_symlink(struct symlink_args *uap)
1901 {
1902         struct thread *td = curthread;
1903         struct nlookupdata nd;
1904         char *path;
1905         int error;
1906         int mode;
1907
1908         path = objcache_get(namei_oc, M_WAITOK);
1909         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1910         if (error == 0) {
1911                 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1912                 if (error == 0) {
1913                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1914                         error = kern_symlink(&nd, path, mode);
1915                 }
1916                 nlookup_done(&nd);
1917         }
1918         objcache_put(namei_oc, path);
1919         return (error);
1920 }
1921
1922 /*
1923  * undelete_args(char *path)
1924  *
1925  * Delete a whiteout from the filesystem.
1926  */
1927 /* ARGSUSED */
1928 int
1929 sys_undelete(struct undelete_args *uap)
1930 {
1931         struct nlookupdata nd;
1932         int error;
1933
1934         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1935         bwillwrite();
1936         nd.nl_flags |= NLC_DELETE;
1937         if (error == 0)
1938                 error = nlookup(&nd);
1939         if (error == 0)
1940                 error = ncp_writechk(&nd.nl_nch);
1941         if (error == 0)
1942                 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_cred, NAMEI_DELETE);
1943         nlookup_done(&nd);
1944         return (error);
1945 }
1946
1947 int
1948 kern_unlink(struct nlookupdata *nd)
1949 {
1950         int error;
1951
1952         bwillwrite();
1953         nd->nl_flags |= NLC_DELETE;
1954         if ((error = nlookup(nd)) != 0)
1955                 return (error);
1956         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1957                 return (error);
1958         error = VOP_NREMOVE(&nd->nl_nch, nd->nl_cred);
1959         return (error);
1960 }
1961
1962 /*
1963  * unlink_args(char *path)
1964  *
1965  * Delete a name from the filesystem.
1966  */
1967 int
1968 sys_unlink(struct unlink_args *uap)
1969 {
1970         struct nlookupdata nd;
1971         int error;
1972
1973         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1974         if (error == 0)
1975                 error = kern_unlink(&nd);
1976         nlookup_done(&nd);
1977         return (error);
1978 }
1979
1980 int
1981 kern_lseek(int fd, off_t offset, int whence, off_t *res)
1982 {
1983         struct thread *td = curthread;
1984         struct proc *p = td->td_proc;
1985         struct file *fp;
1986         struct vattr vattr;
1987         int error;
1988
1989         fp = holdfp(p->p_fd, fd, -1);
1990         if (fp == NULL)
1991                 return (EBADF);
1992         if (fp->f_type != DTYPE_VNODE) {
1993                 error = ESPIPE;
1994                 goto done;
1995         }
1996
1997         switch (whence) {
1998         case L_INCR:
1999                 fp->f_offset += offset;
2000                 error = 0;
2001                 break;
2002         case L_XTND:
2003                 error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
2004                 if (error == 0)
2005                         fp->f_offset = offset + vattr.va_size;
2006                 break;
2007         case L_SET:
2008                 fp->f_offset = offset;
2009                 error = 0;
2010                 break;
2011         default:
2012                 error = EINVAL;
2013                 break;
2014         }
2015         *res = fp->f_offset;
2016 done:
2017         fdrop(fp);
2018         return (error);
2019 }
2020
2021 /*
2022  * lseek_args(int fd, int pad, off_t offset, int whence)
2023  *
2024  * Reposition read/write file offset.
2025  */
2026 int
2027 sys_lseek(struct lseek_args *uap)
2028 {
2029         int error;
2030
2031         error = kern_lseek(uap->fd, uap->offset, uap->whence,
2032             &uap->sysmsg_offset);
2033
2034         return (error);
2035 }
2036
2037 int
2038 kern_access(struct nlookupdata *nd, int aflags)
2039 {
2040         struct vnode *vp;
2041         int error, flags;
2042
2043         if ((error = nlookup(nd)) != 0)
2044                 return (error);
2045 retry:
2046         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2047         if (error)
2048                 return (error);
2049
2050         /* Flags == 0 means only check for existence. */
2051         if (aflags) {
2052                 flags = 0;
2053                 if (aflags & R_OK)
2054                         flags |= VREAD;
2055                 if (aflags & W_OK)
2056                         flags |= VWRITE;
2057                 if (aflags & X_OK)
2058                         flags |= VEXEC;
2059                 if ((flags & VWRITE) == 0 ||
2060                     (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2061                         error = VOP_ACCESS(vp, flags, nd->nl_cred);
2062
2063                 /*
2064                  * If the file handle is stale we have to re-resolve the
2065                  * entry.  This is a hack at the moment.
2066                  */
2067                 if (error == ESTALE) {
2068                         cache_setunresolved(&nd->nl_nch);
2069                         error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2070                         if (error == 0) {
2071                                 vput(vp);
2072                                 vp = NULL;
2073                                 goto retry;
2074                         }
2075                 }
2076         }
2077         vput(vp);
2078         return (error);
2079 }
2080
2081 /*
2082  * access_args(char *path, int flags)
2083  *
2084  * Check access permissions.
2085  */
2086 int
2087 sys_access(struct access_args *uap)
2088 {
2089         struct nlookupdata nd;
2090         int error;
2091
2092         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2093         if (error == 0)
2094                 error = kern_access(&nd, uap->flags);
2095         nlookup_done(&nd);
2096         return (error);
2097 }
2098
2099 int
2100 kern_stat(struct nlookupdata *nd, struct stat *st)
2101 {
2102         int error;
2103         struct vnode *vp;
2104         thread_t td;
2105
2106         if ((error = nlookup(nd)) != 0)
2107                 return (error);
2108 again:
2109         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2110                 return (ENOENT);
2111
2112         td = curthread;
2113         if ((error = vget(vp, LK_SHARED)) != 0)
2114                 return (error);
2115         error = vn_stat(vp, st, nd->nl_cred);
2116
2117         /*
2118          * If the file handle is stale we have to re-resolve the entry.  This
2119          * is a hack at the moment.
2120          */
2121         if (error == ESTALE) {
2122                 cache_setunresolved(&nd->nl_nch);
2123                 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2124                 if (error == 0) {
2125                         vput(vp);
2126                         goto again;
2127                 }
2128         }
2129         vput(vp);
2130         return (error);
2131 }
2132
2133 /*
2134  * stat_args(char *path, struct stat *ub)
2135  *
2136  * Get file status; this version follows links.
2137  */
2138 int
2139 sys_stat(struct stat_args *uap)
2140 {
2141         struct nlookupdata nd;
2142         struct stat st;
2143         int error;
2144
2145         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2146         if (error == 0) {
2147                 error = kern_stat(&nd, &st);
2148                 if (error == 0)
2149                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2150         }
2151         nlookup_done(&nd);
2152         return (error);
2153 }
2154
2155 /*
2156  * lstat_args(char *path, struct stat *ub)
2157  *
2158  * Get file status; this version does not follow links.
2159  */
2160 int
2161 sys_lstat(struct lstat_args *uap)
2162 {
2163         struct nlookupdata nd;
2164         struct stat st;
2165         int error;
2166
2167         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2168         if (error == 0) {
2169                 error = kern_stat(&nd, &st);
2170                 if (error == 0)
2171                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2172         }
2173         nlookup_done(&nd);
2174         return (error);
2175 }
2176
2177 /*
2178  * pathconf_Args(char *path, int name)
2179  *
2180  * Get configurable pathname variables.
2181  */
2182 /* ARGSUSED */
2183 int
2184 sys_pathconf(struct pathconf_args *uap)
2185 {
2186         struct nlookupdata nd;
2187         struct vnode *vp;
2188         int error;
2189
2190         vp = NULL;
2191         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2192         if (error == 0)
2193                 error = nlookup(&nd);
2194         if (error == 0)
2195                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2196         nlookup_done(&nd);
2197         if (error == 0) {
2198                 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2199                 vput(vp);
2200         }
2201         return (error);
2202 }
2203
2204 /*
2205  * XXX: daver
2206  * kern_readlink isn't properly split yet.  There is a copyin burried
2207  * in VOP_READLINK().
2208  */
2209 int
2210 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2211 {
2212         struct thread *td = curthread;
2213         struct proc *p = td->td_proc;
2214         struct vnode *vp;
2215         struct iovec aiov;
2216         struct uio auio;
2217         int error;
2218
2219         if ((error = nlookup(nd)) != 0)
2220                 return (error);
2221         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2222         if (error)
2223                 return (error);
2224         if (vp->v_type != VLNK) {
2225                 error = EINVAL;
2226         } else {
2227                 aiov.iov_base = buf;
2228                 aiov.iov_len = count;
2229                 auio.uio_iov = &aiov;
2230                 auio.uio_iovcnt = 1;
2231                 auio.uio_offset = 0;
2232                 auio.uio_rw = UIO_READ;
2233                 auio.uio_segflg = UIO_USERSPACE;
2234                 auio.uio_td = td;
2235                 auio.uio_resid = count;
2236                 error = VOP_READLINK(vp, &auio, p->p_ucred);
2237         }
2238         vput(vp);
2239         *res = count - auio.uio_resid;
2240         return (error);
2241 }
2242
2243 /*
2244  * readlink_args(char *path, char *buf, int count)
2245  *
2246  * Return target name of a symbolic link.
2247  */
2248 int
2249 sys_readlink(struct readlink_args *uap)
2250 {
2251         struct nlookupdata nd;
2252         int error;
2253
2254         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2255         if (error == 0) {
2256                 error = kern_readlink(&nd, uap->buf, uap->count,
2257                                         &uap->sysmsg_result);
2258         }
2259         nlookup_done(&nd);
2260         return (error);
2261 }
2262
2263 static int
2264 setfflags(struct vnode *vp, int flags)
2265 {
2266         struct thread *td = curthread;
2267         struct proc *p = td->td_proc;
2268         int error;
2269         struct vattr vattr;
2270
2271         /*
2272          * Prevent non-root users from setting flags on devices.  When
2273          * a device is reused, users can retain ownership of the device
2274          * if they are allowed to set flags and programs assume that
2275          * chown can't fail when done as root.
2276          */
2277         if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2278             ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2279                 return (error);
2280
2281         /*
2282          * note: vget is required for any operation that might mod the vnode
2283          * so VINACTIVE is properly cleared.
2284          */
2285         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2286                 VATTR_NULL(&vattr);
2287                 vattr.va_flags = flags;
2288                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2289                 vput(vp);
2290         }
2291         return (error);
2292 }
2293
2294 /*
2295  * chflags(char *path, int flags)
2296  *
2297  * Change flags of a file given a path name.
2298  */
2299 /* ARGSUSED */
2300 int
2301 sys_chflags(struct chflags_args *uap)
2302 {
2303         struct nlookupdata nd;
2304         struct vnode *vp;
2305         int error;
2306
2307         vp = NULL;
2308         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2309         /* XXX Add NLC flag indicating modifying operation? */
2310         if (error == 0)
2311                 error = nlookup(&nd);
2312         if (error == 0)
2313                 error = ncp_writechk(&nd.nl_nch);
2314         if (error == 0)
2315                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2316         nlookup_done(&nd);
2317         if (error == 0) {
2318                 error = setfflags(vp, uap->flags);
2319                 vrele(vp);
2320         }
2321         return (error);
2322 }
2323
2324 /*
2325  * fchflags_args(int fd, int flags)
2326  *
2327  * Change flags of a file given a file descriptor.
2328  */
2329 /* ARGSUSED */
2330 int
2331 sys_fchflags(struct fchflags_args *uap)
2332 {
2333         struct thread *td = curthread;
2334         struct proc *p = td->td_proc;
2335         struct file *fp;
2336         int error;
2337
2338         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2339                 return (error);
2340         if (fp->f_nchandle.ncp)
2341                 error = ncp_writechk(&fp->f_nchandle);
2342         if (error == 0)
2343                 error = setfflags((struct vnode *) fp->f_data, uap->flags);
2344         fdrop(fp);
2345         return (error);
2346 }
2347
2348 static int
2349 setfmode(struct vnode *vp, int mode)
2350 {
2351         struct thread *td = curthread;
2352         struct proc *p = td->td_proc;
2353         int error;
2354         struct vattr vattr;
2355
2356         /*
2357          * note: vget is required for any operation that might mod the vnode
2358          * so VINACTIVE is properly cleared.
2359          */
2360         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2361                 VATTR_NULL(&vattr);
2362                 vattr.va_mode = mode & ALLPERMS;
2363                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2364                 vput(vp);
2365         }
2366         return error;
2367 }
2368
2369 int
2370 kern_chmod(struct nlookupdata *nd, int mode)
2371 {
2372         struct vnode *vp;
2373         int error;
2374
2375         /* XXX Add NLC flag indicating modifying operation? */
2376         if ((error = nlookup(nd)) != 0)
2377                 return (error);
2378         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2379                 return (error);
2380         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2381                 error = setfmode(vp, mode);
2382         vrele(vp);
2383         return (error);
2384 }
2385
2386 /*
2387  * chmod_args(char *path, int mode)
2388  *
2389  * Change mode of a file given path name.
2390  */
2391 /* ARGSUSED */
2392 int
2393 sys_chmod(struct chmod_args *uap)
2394 {
2395         struct nlookupdata nd;
2396         int error;
2397
2398         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2399         if (error == 0)
2400                 error = kern_chmod(&nd, uap->mode);
2401         nlookup_done(&nd);
2402         return (error);
2403 }
2404
2405 /*
2406  * lchmod_args(char *path, int mode)
2407  *
2408  * Change mode of a file given path name (don't follow links.)
2409  */
2410 /* ARGSUSED */
2411 int
2412 sys_lchmod(struct lchmod_args *uap)
2413 {
2414         struct nlookupdata nd;
2415         int error;
2416
2417         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2418         if (error == 0)
2419                 error = kern_chmod(&nd, uap->mode);
2420         nlookup_done(&nd);
2421         return (error);
2422 }
2423
2424 /*
2425  * fchmod_args(int fd, int mode)
2426  *
2427  * Change mode of a file given a file descriptor.
2428  */
2429 /* ARGSUSED */
2430 int
2431 sys_fchmod(struct fchmod_args *uap)
2432 {
2433         struct thread *td = curthread;
2434         struct proc *p = td->td_proc;
2435         struct file *fp;
2436         int error;
2437
2438         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2439                 return (error);
2440         if (fp->f_nchandle.ncp)
2441                 error = ncp_writechk(&fp->f_nchandle);
2442         if (error == 0)
2443                 error = setfmode((struct vnode *)fp->f_data, uap->mode);
2444         fdrop(fp);
2445         return (error);
2446 }
2447
2448 static int
2449 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2450 {
2451         struct thread *td = curthread;
2452         struct proc *p = td->td_proc;
2453         int error;
2454         struct vattr vattr;
2455
2456         /*
2457          * note: vget is required for any operation that might mod the vnode
2458          * so VINACTIVE is properly cleared.
2459          */
2460         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2461                 VATTR_NULL(&vattr);
2462                 vattr.va_uid = uid;
2463                 vattr.va_gid = gid;
2464                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2465                 vput(vp);
2466         }
2467         return error;
2468 }
2469
2470 int
2471 kern_chown(struct nlookupdata *nd, int uid, int gid)
2472 {
2473         struct vnode *vp;
2474         int error;
2475
2476         /* XXX Add NLC flag indicating modifying operation? */
2477         if ((error = nlookup(nd)) != 0)
2478                 return (error);
2479         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2480                 return (error);
2481         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2482                 error = setfown(vp, uid, gid);
2483         vrele(vp);
2484         return (error);
2485 }
2486
2487 /*
2488  * chown(char *path, int uid, int gid)
2489  *
2490  * Set ownership given a path name.
2491  */
2492 int
2493 sys_chown(struct chown_args *uap)
2494 {
2495         struct nlookupdata nd;
2496         int error;
2497
2498         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2499         if (error == 0)
2500                 error = kern_chown(&nd, uap->uid, uap->gid);
2501         nlookup_done(&nd);
2502         return (error);
2503 }
2504
2505 /*
2506  * lchown_args(char *path, int uid, int gid)
2507  *
2508  * Set ownership given a path name, do not cross symlinks.
2509  */
2510 int
2511 sys_lchown(struct lchown_args *uap)
2512 {
2513         struct nlookupdata nd;
2514         int error;
2515
2516         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2517         if (error == 0)
2518                 error = kern_chown(&nd, uap->uid, uap->gid);
2519         nlookup_done(&nd);
2520         return (error);
2521 }
2522
2523 /*
2524  * fchown_args(int fd, int uid, int gid)
2525  *
2526  * Set ownership given a file descriptor.
2527  */
2528 /* ARGSUSED */
2529 int
2530 sys_fchown(struct fchown_args *uap)
2531 {
2532         struct thread *td = curthread;
2533         struct proc *p = td->td_proc;
2534         struct file *fp;
2535         int error;
2536
2537         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2538                 return (error);
2539         if (fp->f_nchandle.ncp)
2540                 error = ncp_writechk(&fp->f_nchandle);
2541         if (error == 0)
2542                 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2543         fdrop(fp);
2544         return (error);
2545 }
2546
2547 static int
2548 getutimes(const struct timeval *tvp, struct timespec *tsp)
2549 {
2550         struct timeval tv[2];
2551
2552         if (tvp == NULL) {
2553                 microtime(&tv[0]);
2554                 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2555                 tsp[1] = tsp[0];
2556         } else {
2557                 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2558                 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2559         }
2560         return 0;
2561 }
2562
2563 static int
2564 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2565 {
2566         struct thread *td = curthread;
2567         struct proc *p = td->td_proc;
2568         int error;
2569         struct vattr vattr;
2570
2571         /*
2572          * note: vget is required for any operation that might mod the vnode
2573          * so VINACTIVE is properly cleared.
2574          */
2575         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2576                 VATTR_NULL(&vattr);
2577                 vattr.va_atime = ts[0];
2578                 vattr.va_mtime = ts[1];
2579                 if (nullflag)
2580                         vattr.va_vaflags |= VA_UTIMES_NULL;
2581                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2582                 vput(vp);
2583         }
2584         return error;
2585 }
2586
2587 int
2588 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2589 {
2590         struct timespec ts[2];
2591         struct vnode *vp;
2592         int error;
2593
2594         if ((error = getutimes(tptr, ts)) != 0)
2595                 return (error);
2596         /* XXX Add NLC flag indicating modifying operation? */
2597         if ((error = nlookup(nd)) != 0)
2598                 return (error);
2599         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2600                 return (error);
2601         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2602                 return (error);
2603         error = setutimes(vp, ts, tptr == NULL);
2604         vrele(vp);
2605         return (error);
2606 }
2607
2608 /*
2609  * utimes_args(char *path, struct timeval *tptr)
2610  *
2611  * Set the access and modification times of a file.
2612  */
2613 int
2614 sys_utimes(struct utimes_args *uap)
2615 {
2616         struct timeval tv[2];
2617         struct nlookupdata nd;
2618         int error;
2619
2620         if (uap->tptr) {
2621                 error = copyin(uap->tptr, tv, sizeof(tv));
2622                 if (error)
2623                         return (error);
2624         }
2625         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2626         if (error == 0)
2627                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2628         nlookup_done(&nd);
2629         return (error);
2630 }
2631
2632 /*
2633  * lutimes_args(char *path, struct timeval *tptr)
2634  *
2635  * Set the access and modification times of a file.
2636  */
2637 int
2638 sys_lutimes(struct lutimes_args *uap)
2639 {
2640         struct timeval tv[2];
2641         struct nlookupdata nd;
2642         int error;
2643
2644         if (uap->tptr) {
2645                 error = copyin(uap->tptr, tv, sizeof(tv));
2646                 if (error)
2647                         return (error);
2648         }
2649         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2650         if (error == 0)
2651                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2652         nlookup_done(&nd);
2653         return (error);
2654 }
2655
2656 int
2657 kern_futimes(int fd, struct timeval *tptr)
2658 {
2659         struct thread *td = curthread;
2660         struct proc *p = td->td_proc;
2661         struct timespec ts[2];
2662         struct file *fp;
2663         int error;
2664
2665         error = getutimes(tptr, ts);
2666         if (error)
2667                 return (error);
2668         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2669                 return (error);
2670         if (fp->f_nchandle.ncp)
2671                 error = ncp_writechk(&fp->f_nchandle);
2672         if (error == 0)
2673                 error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2674         fdrop(fp);
2675         return (error);
2676 }
2677
2678 /*
2679  * futimes_args(int fd, struct timeval *tptr)
2680  *
2681  * Set the access and modification times of a file.
2682  */
2683 int
2684 sys_futimes(struct futimes_args *uap)
2685 {
2686         struct timeval tv[2];
2687         int error;
2688
2689         if (uap->tptr) {
2690                 error = copyin(uap->tptr, tv, sizeof(tv));
2691                 if (error)
2692                         return (error);
2693         }
2694
2695         error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2696
2697         return (error);
2698 }
2699
2700 int
2701 kern_truncate(struct nlookupdata *nd, off_t length)
2702 {
2703         struct vnode *vp;
2704         struct vattr vattr;
2705         int error;
2706
2707         if (length < 0)
2708                 return(EINVAL);
2709         /* XXX Add NLC flag indicating modifying operation? */
2710         if ((error = nlookup(nd)) != 0)
2711                 return (error);
2712         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2713                 return (error);
2714         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2715                 return (error);
2716         if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2717                 vrele(vp);
2718                 return (error);
2719         }
2720         if (vp->v_type == VDIR) {
2721                 error = EISDIR;
2722         } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
2723             (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2724                 VATTR_NULL(&vattr);
2725                 vattr.va_size = length;
2726                 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2727         }
2728         vput(vp);
2729         return (error);
2730 }
2731
2732 /*
2733  * truncate(char *path, int pad, off_t length)
2734  *
2735  * Truncate a file given its path name.
2736  */
2737 int
2738 sys_truncate(struct truncate_args *uap)
2739 {
2740         struct nlookupdata nd;
2741         int error;
2742
2743         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2744         if (error == 0)
2745                 error = kern_truncate(&nd, uap->length);
2746         nlookup_done(&nd);
2747         return error;
2748 }
2749
2750 int
2751 kern_ftruncate(int fd, off_t length)
2752 {
2753         struct thread *td = curthread;
2754         struct proc *p = td->td_proc;
2755         struct vattr vattr;
2756         struct vnode *vp;
2757         struct file *fp;
2758         int error;
2759
2760         if (length < 0)
2761                 return(EINVAL);
2762         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2763                 return (error);
2764         if (fp->f_nchandle.ncp) {
2765                 error = ncp_writechk(&fp->f_nchandle);
2766                 if (error)
2767                         goto done;
2768         }
2769         if ((fp->f_flag & FWRITE) == 0) {
2770                 error = EINVAL;
2771                 goto done;
2772         }
2773         vp = (struct vnode *)fp->f_data;
2774         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2775         if (vp->v_type == VDIR) {
2776                 error = EISDIR;
2777         } else if ((error = vn_writechk(vp, NULL)) == 0) {
2778                 VATTR_NULL(&vattr);
2779                 vattr.va_size = length;
2780                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2781         }
2782         vn_unlock(vp);
2783 done:
2784         fdrop(fp);
2785         return (error);
2786 }
2787
2788 /*
2789  * ftruncate_args(int fd, int pad, off_t length)
2790  *
2791  * Truncate a file given a file descriptor.
2792  */
2793 int
2794 sys_ftruncate(struct ftruncate_args *uap)
2795 {
2796         int error;
2797
2798         error = kern_ftruncate(uap->fd, uap->length);
2799
2800         return (error);
2801 }
2802
2803 /*
2804  * fsync(int fd)
2805  *
2806  * Sync an open file.
2807  */
2808 /* ARGSUSED */
2809 int
2810 sys_fsync(struct fsync_args *uap)
2811 {
2812         struct thread *td = curthread;
2813         struct proc *p = td->td_proc;
2814         struct vnode *vp;
2815         struct file *fp;
2816         vm_object_t obj;
2817         int error;
2818
2819         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2820                 return (error);
2821         vp = (struct vnode *)fp->f_data;
2822         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2823         if ((obj = vp->v_object) != NULL)
2824                 vm_object_page_clean(obj, 0, 0, 0);
2825         if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 &&
2826             vp->v_mount && (vp->v_mount->mnt_flag & MNT_SOFTDEP) &&
2827             bioops.io_fsync) {
2828                 error = (*bioops.io_fsync)(vp);
2829         }
2830         vn_unlock(vp);
2831         fdrop(fp);
2832         return (error);
2833 }
2834
2835 int
2836 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2837 {
2838         struct nchandle fnchd;
2839         struct nchandle tnchd;
2840         struct namecache *ncp;
2841         struct mount *mp;
2842         int error;
2843
2844         bwillwrite();
2845         if ((error = nlookup(fromnd)) != 0)
2846                 return (error);
2847         if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
2848                 return (ENOENT);
2849         fnchd.mount = fromnd->nl_nch.mount;
2850         cache_hold(&fnchd);
2851
2852         /*
2853          * unlock the source nch so we can lookup the target nch without
2854          * deadlocking.  The target may or may not exist so we do not check
2855          * for a target vp like kern_mkdir() and other creation functions do.
2856          *
2857          * The source and target directories are ref'd and rechecked after
2858          * everything is relocked to determine if the source or target file
2859          * has been renamed.
2860          */
2861         KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2862         fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2863         cache_unlock(&fromnd->nl_nch);
2864
2865         tond->nl_flags |= NLC_CREATE;
2866         if ((error = nlookup(tond)) != 0) {
2867                 cache_drop(&fnchd);
2868                 return (error);
2869         }
2870         if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
2871                 cache_drop(&fnchd);
2872                 return (ENOENT);
2873         }
2874         tnchd.mount = tond->nl_nch.mount;
2875         cache_hold(&tnchd);
2876
2877         /*
2878          * If the source and target are the same there is nothing to do
2879          */
2880         if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
2881                 cache_drop(&fnchd);
2882                 cache_drop(&tnchd);
2883                 return (0);
2884         }
2885
2886         /*
2887          * Mount points cannot be renamed or overwritten
2888          */
2889         if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
2890             NCF_ISMOUNTPT
2891         ) {
2892                 cache_drop(&fnchd);
2893                 cache_drop(&tnchd);
2894                 return (EINVAL);
2895         }
2896
2897         /*
2898          * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2899          * may have become invalid while it was unlocked, nc_vp and nc_mount
2900          * could be NULL.
2901          */
2902         if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
2903                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2904         } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
2905                 cache_lock(&fromnd->nl_nch);
2906                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2907         } else {
2908                 cache_unlock(&tond->nl_nch);
2909                 cache_lock(&fromnd->nl_nch);
2910                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2911                 cache_lock(&tond->nl_nch);
2912                 cache_resolve(&tond->nl_nch, tond->nl_cred);
2913         }
2914         fromnd->nl_flags |= NLC_NCPISLOCKED;
2915
2916         /*
2917          * make sure the parent directories linkages are the same
2918          */
2919         if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
2920             tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
2921                 cache_drop(&fnchd);
2922                 cache_drop(&tnchd);
2923                 return (ENOENT);
2924         }
2925
2926         /*
2927          * Both the source and target must be within the same filesystem and
2928          * in the same filesystem as their parent directories within the
2929          * namecache topology.
2930          *
2931          * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2932          */
2933         mp = fnchd.mount;
2934         if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
2935             mp != tond->nl_nch.mount) {
2936                 cache_drop(&fnchd);
2937                 cache_drop(&tnchd);
2938                 return (EXDEV);
2939         }
2940
2941         /*
2942          * Make sure the mount point is writable
2943          */
2944         if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
2945                 cache_drop(&fnchd);
2946                 cache_drop(&tnchd);
2947                 return (error);
2948         }
2949
2950         /*
2951          * If the target exists and either the source or target is a directory,
2952          * then both must be directories.
2953          *
2954          * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
2955          * have become NULL.
2956          */
2957         if (tond->nl_nch.ncp->nc_vp) {
2958                 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
2959                         error = ENOENT;
2960                 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
2961                         if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
2962                                 error = ENOTDIR;
2963                 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
2964                         error = EISDIR;
2965                 }
2966         }
2967
2968         /*
2969          * You cannot rename a source into itself or a subdirectory of itself.
2970          * We check this by travsersing the target directory upwards looking
2971          * for a match against the source.
2972          */
2973         if (error == 0) {
2974                 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
2975                         if (fromnd->nl_nch.ncp == ncp) {
2976                                 error = EINVAL;
2977                                 break;
2978                         }
2979                 }
2980         }
2981
2982         cache_drop(&fnchd);
2983         cache_drop(&tnchd);
2984
2985         /*
2986          * Even though the namespaces are different, they may still represent
2987          * hardlinks to the same file.  The filesystem might have a hard time
2988          * with this so we issue a NREMOVE of the source instead of a NRENAME
2989          * when we detect the situation.
2990          */
2991         if (error == 0) {
2992                 if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
2993                         error = VOP_NREMOVE(&fromnd->nl_nch, fromnd->nl_cred);
2994                 } else {
2995                         error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
2996                                             tond->nl_cred);
2997                 }
2998         }
2999         return (error);
3000 }
3001
3002 /*
3003  * rename_args(char *from, char *to)
3004  *
3005  * Rename files.  Source and destination must either both be directories,
3006  * or both not be directories.  If target is a directory, it must be empty.
3007  */
3008 int
3009 sys_rename(struct rename_args *uap)
3010 {
3011         struct nlookupdata fromnd, tond;
3012         int error;
3013
3014         error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3015         if (error == 0) {
3016                 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3017                 if (error == 0)
3018                         error = kern_rename(&fromnd, &tond);
3019                 nlookup_done(&tond);
3020         }
3021         nlookup_done(&fromnd);
3022         return (error);
3023 }
3024
3025 int
3026 kern_mkdir(struct nlookupdata *nd, int mode)
3027 {
3028         struct thread *td = curthread;
3029         struct proc *p = td->td_proc;
3030         struct vnode *vp;
3031         struct vattr vattr;
3032         int error;
3033
3034         bwillwrite();
3035         nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
3036         if ((error = nlookup(nd)) != 0)
3037                 return (error);
3038
3039         if (nd->nl_nch.ncp->nc_vp)
3040                 return (EEXIST);
3041         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3042                 return (error);
3043
3044         VATTR_NULL(&vattr);
3045         vattr.va_type = VDIR;
3046         vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3047
3048         vp = NULL;
3049         error = VOP_NMKDIR(&nd->nl_nch, &vp, p->p_ucred, &vattr);
3050         if (error == 0)
3051                 vput(vp);
3052         return (error);
3053 }
3054
3055 /*
3056  * mkdir_args(char *path, int mode)
3057  *
3058  * Make a directory file.
3059  */
3060 /* ARGSUSED */
3061 int
3062 sys_mkdir(struct mkdir_args *uap)
3063 {
3064         struct nlookupdata nd;
3065         int error;
3066
3067         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3068         if (error == 0)
3069                 error = kern_mkdir(&nd, uap->mode);
3070         nlookup_done(&nd);
3071         return (error);
3072 }
3073
3074 int
3075 kern_rmdir(struct nlookupdata *nd)
3076 {
3077         int error;
3078
3079         bwillwrite();
3080         nd->nl_flags |= NLC_DELETE;
3081         if ((error = nlookup(nd)) != 0)
3082                 return (error);
3083
3084         /*
3085          * Do not allow directories representing mount points to be
3086          * deleted, even if empty.  Check write perms on mount point
3087          * in case the vnode is aliased (aka nullfs).
3088          */
3089         if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3090                 return (EINVAL);
3091         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3092                 return (error);
3093
3094         error = VOP_NRMDIR(&nd->nl_nch, nd->nl_cred);
3095         return (error);
3096 }
3097
3098 /*
3099  * rmdir_args(char *path)
3100  *
3101  * Remove a directory file.
3102  */
3103 /* ARGSUSED */
3104 int
3105 sys_rmdir(struct rmdir_args *uap)
3106 {
3107         struct nlookupdata nd;
3108         int error;
3109
3110         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3111         if (error == 0)
3112                 error = kern_rmdir(&nd);
3113         nlookup_done(&nd);
3114         return (error);
3115 }
3116
3117 int
3118 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3119     enum uio_seg direction)
3120 {
3121         struct thread *td = curthread;
3122         struct proc *p = td->td_proc;
3123         struct vnode *vp;
3124         struct file *fp;
3125         struct uio auio;
3126         struct iovec aiov;
3127         long loff;
3128         int error, eofflag;
3129
3130         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3131                 return (error);
3132         if ((fp->f_flag & FREAD) == 0) {
3133                 error = EBADF;
3134                 goto done;
3135         }
3136         vp = (struct vnode *)fp->f_data;
3137 unionread:
3138         if (vp->v_type != VDIR) {
3139                 error = EINVAL;
3140                 goto done;
3141         }
3142         aiov.iov_base = buf;
3143         aiov.iov_len = count;
3144         auio.uio_iov = &aiov;
3145         auio.uio_iovcnt = 1;
3146         auio.uio_rw = UIO_READ;
3147         auio.uio_segflg = direction;
3148         auio.uio_td = td;
3149         auio.uio_resid = count;
3150         loff = auio.uio_offset = fp->f_offset;
3151         error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3152         fp->f_offset = auio.uio_offset;
3153         if (error)
3154                 goto done;
3155         if (count == auio.uio_resid) {
3156                 if (union_dircheckp) {
3157                         error = union_dircheckp(td, &vp, fp);
3158                         if (error == -1)
3159                                 goto unionread;
3160                         if (error)
3161                                 goto done;
3162                 }
3163 #if 0
3164                 if ((vp->v_flag & VROOT) &&
3165                     (vp->v_mount->mnt_flag & MNT_UNION)) {
3166                         struct vnode *tvp = vp;
3167                         vp = vp->v_mount->mnt_vnodecovered;
3168                         vref(vp);
3169                         fp->f_data = vp;
3170                         fp->f_offset = 0;
3171                         vrele(tvp);
3172                         goto unionread;
3173                 }
3174 #endif
3175         }
3176         if (basep) {
3177                 *basep = loff;
3178         }
3179         *res = count - auio.uio_resid;
3180 done:
3181         fdrop(fp);
3182         return (error);
3183 }
3184
3185 /*
3186  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3187  *
3188  * Read a block of directory entries in a file system independent format.
3189  */
3190 int
3191 sys_getdirentries(struct getdirentries_args *uap)
3192 {
3193         long base;
3194         int error;
3195
3196         error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3197             &uap->sysmsg_result, UIO_USERSPACE);
3198
3199         if (error == 0)
3200                 error = copyout(&base, uap->basep, sizeof(*uap->basep));
3201         return (error);
3202 }
3203
3204 /*
3205  * getdents_args(int fd, char *buf, size_t count)
3206  */
3207 int
3208 sys_getdents(struct getdents_args *uap)
3209 {
3210         int error;
3211
3212         error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3213             &uap->sysmsg_result, UIO_USERSPACE);
3214
3215         return (error);
3216 }
3217
3218 /*
3219  * umask(int newmask)
3220  *
3221  * Set the mode mask for creation of filesystem nodes.
3222  *
3223  * MP SAFE
3224  */
3225 int
3226 sys_umask(struct umask_args *uap)
3227 {
3228         struct thread *td = curthread;
3229         struct proc *p = td->td_proc;
3230         struct filedesc *fdp;
3231
3232         fdp = p->p_fd;
3233         uap->sysmsg_result = fdp->fd_cmask;
3234         fdp->fd_cmask = uap->newmask & ALLPERMS;
3235         return (0);
3236 }
3237
3238 /*
3239  * revoke(char *path)
3240  *
3241  * Void all references to file by ripping underlying filesystem
3242  * away from vnode.
3243  */
3244 /* ARGSUSED */
3245 int
3246 sys_revoke(struct revoke_args *uap)
3247 {
3248         struct nlookupdata nd;
3249         struct vattr vattr;
3250         struct vnode *vp;
3251         struct ucred *cred;
3252         int error;
3253
3254         vp = NULL;
3255         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3256         if (error == 0)
3257                 error = nlookup(&nd);
3258         if (error == 0)
3259                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3260         cred = crhold(nd.nl_cred);
3261         nlookup_done(&nd);
3262         if (error == 0) {
3263                 if (vp->v_type != VCHR && vp->v_type != VBLK)
3264                         error = EINVAL;
3265                 if (error == 0)
3266                         error = VOP_GETATTR(vp, &vattr);
3267                 if (error == 0 && cred->cr_uid != vattr.va_uid)
3268                         error = suser_cred(cred, PRISON_ROOT);
3269                 if (error == 0 && count_udev(vp->v_umajor, vp->v_uminor) > 0) {
3270                         error = 0;
3271                         vx_lock(vp);
3272                         VOP_REVOKE(vp, REVOKEALL);
3273                         vx_unlock(vp);
3274                 }
3275                 vrele(vp);
3276         }
3277         if (cred)
3278                 crfree(cred);
3279         return (error);
3280 }
3281
3282 /*
3283  * getfh_args(char *fname, fhandle_t *fhp)
3284  *
3285  * Get (NFS) file handle
3286  */
3287 int
3288 sys_getfh(struct getfh_args *uap)
3289 {
3290         struct thread *td = curthread;
3291         struct nlookupdata nd;
3292         fhandle_t fh;
3293         struct vnode *vp;
3294         int error;
3295
3296         /*
3297          * Must be super user
3298          */
3299         if ((error = suser(td)) != 0)
3300                 return (error);
3301
3302         vp = NULL;
3303         error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3304         if (error == 0)
3305                 error = nlookup(&nd);
3306         if (error == 0)
3307                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3308         nlookup_done(&nd);
3309         if (error == 0) {
3310                 bzero(&fh, sizeof(fh));
3311                 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3312                 error = VFS_VPTOFH(vp, &fh.fh_fid);
3313                 vput(vp);
3314                 if (error == 0)
3315                         error = copyout(&fh, uap->fhp, sizeof(fh));
3316         }
3317         return (error);
3318 }
3319
3320 /*
3321  * fhopen_args(const struct fhandle *u_fhp, int flags)
3322  *
3323  * syscall for the rpc.lockd to use to translate a NFS file handle into
3324  * an open descriptor.
3325  *
3326  * warning: do not remove the suser() call or this becomes one giant
3327  * security hole.
3328  */
3329 int
3330 sys_fhopen(struct fhopen_args *uap)
3331 {
3332         struct thread *td = curthread;
3333         struct proc *p = td->td_proc;
3334         struct mount *mp;
3335         struct vnode *vp;
3336         struct fhandle fhp;
3337         struct vattr vat;
3338         struct vattr *vap = &vat;
3339         struct flock lf;
3340         int fmode, mode, error, type;
3341         struct file *nfp;
3342         struct file *fp;
3343         int indx;
3344
3345         /*
3346          * Must be super user
3347          */
3348         error = suser(td);
3349         if (error)
3350                 return (error);
3351
3352         fmode = FFLAGS(uap->flags);
3353         /* why not allow a non-read/write open for our lockd? */
3354         if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3355                 return (EINVAL);
3356         error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3357         if (error)
3358                 return(error);
3359         /* find the mount point */
3360         mp = vfs_getvfs(&fhp.fh_fsid);
3361         if (mp == NULL)
3362                 return (ESTALE);
3363         /* now give me my vnode, it gets returned to me locked */
3364         error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3365         if (error)
3366                 return (error);
3367         /*
3368          * from now on we have to make sure not
3369          * to forget about the vnode
3370          * any error that causes an abort must vput(vp)
3371          * just set error = err and 'goto bad;'.
3372          */
3373
3374         /*
3375          * from vn_open
3376          */
3377         if (vp->v_type == VLNK) {
3378                 error = EMLINK;
3379                 goto bad;
3380         }
3381         if (vp->v_type == VSOCK) {
3382                 error = EOPNOTSUPP;
3383                 goto bad;
3384         }
3385         mode = 0;
3386         if (fmode & (FWRITE | O_TRUNC)) {
3387                 if (vp->v_type == VDIR) {
3388                         error = EISDIR;
3389                         goto bad;
3390                 }
3391                 error = vn_writechk(vp, NULL);
3392                 if (error)
3393                         goto bad;
3394                 mode |= VWRITE;
3395         }
3396         if (fmode & FREAD)
3397                 mode |= VREAD;
3398         if (mode) {
3399                 error = VOP_ACCESS(vp, mode, p->p_ucred);
3400                 if (error)
3401                         goto bad;
3402         }
3403         if (fmode & O_TRUNC) {
3404                 vn_unlock(vp);                          /* XXX */
3405                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
3406                 VATTR_NULL(vap);
3407                 vap->va_size = 0;
3408                 error = VOP_SETATTR(vp, vap, p->p_ucred);
3409                 if (error)
3410                         goto bad;
3411         }
3412
3413         /*
3414          * VOP_OPEN needs the file pointer so it can potentially override
3415          * it.
3416          *
3417          * WARNING! no f_nchandle will be associated when fhopen()ing a
3418          * directory.  XXX
3419          */
3420         if ((error = falloc(p, &nfp, &indx)) != 0)
3421                 goto bad;
3422         fp = nfp;
3423
3424         error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3425         if (error) {
3426                 /*
3427                  * setting f_ops this way prevents VOP_CLOSE from being
3428                  * called or fdrop() releasing the vp from v_data.   Since
3429                  * the VOP_OPEN failed we don't want to VOP_CLOSE.
3430                  */
3431                 fp->f_ops = &badfileops;
3432                 fp->f_data = NULL;
3433                 goto bad_drop;
3434         }
3435
3436         /*
3437          * The fp is given its own reference, we still have our ref and lock.
3438          *
3439          * Assert that all regular files must be created with a VM object.
3440          */
3441         if (vp->v_type == VREG && vp->v_object == NULL) {
3442                 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3443                 goto bad_drop;
3444         }
3445
3446         /*
3447          * The open was successful.  Handle any locking requirements.
3448          */
3449         if (fmode & (O_EXLOCK | O_SHLOCK)) {
3450                 lf.l_whence = SEEK_SET;
3451                 lf.l_start = 0;
3452                 lf.l_len = 0;
3453                 if (fmode & O_EXLOCK)
3454                         lf.l_type = F_WRLCK;
3455                 else
3456                         lf.l_type = F_RDLCK;
3457                 if (fmode & FNONBLOCK)
3458                         type = 0;
3459                 else
3460                         type = F_WAIT;
3461                 vn_unlock(vp);
3462                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3463                         /*
3464                          * release our private reference.
3465                          */
3466                         fsetfd(p, NULL, indx);
3467                         fdrop(fp);
3468                         vrele(vp);
3469                         return (error);
3470                 }
3471                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3472                 fp->f_flag |= FHASLOCK;
3473         }
3474
3475         /*
3476          * Clean up.  Associate the file pointer with the previously
3477          * reserved descriptor and return it.
3478          */
3479         vput(vp);
3480         fsetfd(p, fp, indx);
3481         fdrop(fp);
3482         uap->sysmsg_result = indx;
3483         return (0);
3484
3485 bad_drop:
3486         fsetfd(p, NULL, indx);
3487         fdrop(fp);
3488 bad:
3489         vput(vp);
3490         return (error);
3491 }
3492
3493 /*
3494  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3495  */
3496 int
3497 sys_fhstat(struct fhstat_args *uap)
3498 {
3499         struct thread *td = curthread;
3500         struct stat sb;
3501         fhandle_t fh;
3502         struct mount *mp;
3503         struct vnode *vp;
3504         int error;
3505
3506         /*
3507          * Must be super user
3508          */
3509         error = suser(td);
3510         if (error)
3511                 return (error);
3512
3513         error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3514         if (error)
3515                 return (error);
3516
3517         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3518                 return (ESTALE);
3519         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3520                 return (error);
3521         error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3522         vput(vp);
3523         if (error)
3524                 return (error);
3525         error = copyout(&sb, uap->sb, sizeof(sb));
3526         return (error);
3527 }
3528
3529 /*
3530  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3531  */
3532 int
3533 sys_fhstatfs(struct fhstatfs_args *uap)
3534 {
3535         struct thread *td = curthread;
3536         struct proc *p = td->td_proc;
3537         struct statfs *sp;
3538         struct mount *mp;
3539         struct vnode *vp;
3540         struct statfs sb;
3541         char *fullpath, *freepath;
3542         fhandle_t fh;
3543         int error;
3544
3545         /*
3546          * Must be super user
3547          */
3548         if ((error = suser(td)))
3549                 return (error);
3550
3551         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3552                 return (error);
3553
3554         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3555                 return (ESTALE);
3556
3557         if (p != NULL && !chroot_visible_mnt(mp, p))
3558                 return (ESTALE);
3559
3560         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3561                 return (error);
3562         mp = vp->v_mount;
3563         sp = &mp->mnt_stat;
3564         vput(vp);
3565         if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3566                 return (error);
3567
3568         error = mount_path(p, mp, &fullpath, &freepath);
3569         if (error)
3570                 return(error);
3571         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3572         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3573         kfree(freepath, M_TEMP);
3574
3575         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3576         if (suser(td)) {
3577                 bcopy(sp, &sb, sizeof(sb));
3578                 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3579                 sp = &sb;
3580         }
3581         return (copyout(sp, uap->buf, sizeof(*sp)));
3582 }
3583
3584 /*
3585  * Syscall to push extended attribute configuration information into the
3586  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3587  * a command (int cmd), and attribute name and misc data.  For now, the
3588  * attribute name is left in userspace for consumption by the VFS_op.
3589  * It will probably be changed to be copied into sysspace by the
3590  * syscall in the future, once issues with various consumers of the
3591  * attribute code have raised their hands.
3592  *
3593  * Currently this is used only by UFS Extended Attributes.
3594  */
3595 int
3596 sys_extattrctl(struct extattrctl_args *uap)
3597 {
3598         struct nlookupdata nd;
3599         struct mount *mp;
3600         struct vnode *vp;
3601         int error;
3602
3603         vp = NULL;
3604         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3605         if (error == 0)
3606                 error = nlookup(&nd);
3607         if (error == 0) {
3608                 mp = nd.nl_nch.mount;
3609                 error = VFS_EXTATTRCTL(mp, uap->cmd,
3610                                 uap->attrname, uap->arg,
3611                                 nd.nl_cred);
3612         }
3613         nlookup_done(&nd);
3614         return (error);
3615 }
3616
3617 /*
3618  * Syscall to set a named extended attribute on a file or directory.
3619  * Accepts attribute name, and a uio structure pointing to the data to set.
3620  * The uio is consumed in the style of writev().  The real work happens
3621  * in VOP_SETEXTATTR().
3622  */
3623 int
3624 sys_extattr_set_file(struct extattr_set_file_args *uap)
3625 {
3626         char attrname[EXTATTR_MAXNAMELEN];
3627         struct iovec aiov[UIO_SMALLIOV];
3628         struct iovec *needfree;
3629         struct nlookupdata nd;
3630         struct iovec *iov;
3631         struct vnode *vp;
3632         struct uio auio;
3633         u_int iovlen;
3634         u_int cnt;
3635         int error;
3636         int i;
3637
3638         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3639         if (error)
3640                 return (error);
3641
3642         vp = NULL;
3643         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3644         if (error == 0)
3645                 error = nlookup(&nd);
3646         if (error == 0)
3647                 error = ncp_writechk(&nd.nl_nch);
3648         if (error == 0)
3649                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3650         if (error) {
3651                 nlookup_done(&nd);
3652                 return (error);
3653         }
3654
3655         needfree = NULL;
3656         iovlen = uap->iovcnt * sizeof(struct iovec);
3657         if (uap->iovcnt > UIO_SMALLIOV) {
3658                 if (uap->iovcnt > UIO_MAXIOV) {
3659                         error = EINVAL;
3660                         goto done;
3661                 }
3662                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3663                 needfree = iov;
3664         } else {
3665                 iov = aiov;
3666         }
3667         auio.uio_iov = iov;
3668         auio.uio_iovcnt = uap->iovcnt;
3669         auio.uio_rw = UIO_WRITE;
3670         auio.uio_segflg = UIO_USERSPACE;
3671         auio.uio_td = nd.nl_td;
3672         auio.uio_offset = 0;
3673         if ((error = copyin(uap->iovp, iov, iovlen)))
3674                 goto done;
3675         auio.uio_resid = 0;
3676         for (i = 0; i < uap->iovcnt; i++) {
3677                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3678                         error = EINVAL;
3679                         goto done;
3680                 }
3681                 auio.uio_resid += iov->iov_len;
3682                 iov++;
3683         }
3684         cnt = auio.uio_resid;
3685         error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3686         cnt -= auio.uio_resid;
3687         uap->sysmsg_result = cnt;
3688 done:
3689         vput(vp);
3690         nlookup_done(&nd);
3691         if (needfree)
3692                 FREE(needfree, M_IOV);
3693         return (error);
3694 }
3695
3696 /*
3697  * Syscall to get a named extended attribute on a file or directory.
3698  * Accepts attribute name, and a uio structure pointing to a buffer for the
3699  * data.  The uio is consumed in the style of readv().  The real work
3700  * happens in VOP_GETEXTATTR();
3701  */
3702 int
3703 sys_extattr_get_file(struct extattr_get_file_args *uap)
3704 {
3705         char attrname[EXTATTR_MAXNAMELEN];
3706         struct iovec aiov[UIO_SMALLIOV];
3707         struct iovec *needfree;
3708         struct nlookupdata nd;
3709         struct iovec *iov;
3710         struct vnode *vp;
3711         struct uio auio;
3712         u_int iovlen;
3713         u_int cnt;
3714         int error;
3715         int i;
3716
3717         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3718         if (error)
3719                 return (error);
3720
3721         vp = NULL;
3722         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3723         if (error == 0)
3724                 error = nlookup(&nd);
3725         if (error == 0)
3726                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3727         if (error) {
3728                 nlookup_done(&nd);
3729                 return (error);
3730         }
3731
3732         iovlen = uap->iovcnt * sizeof (struct iovec);
3733         needfree = NULL;
3734         if (uap->iovcnt > UIO_SMALLIOV) {
3735                 if (uap->iovcnt > UIO_MAXIOV) {
3736                         error = EINVAL;
3737                         goto done;
3738                 }
3739                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3740                 needfree = iov;
3741         } else {
3742                 iov = aiov;
3743         }
3744         auio.uio_iov = iov;
3745         auio.uio_iovcnt = uap->iovcnt;
3746         auio.uio_rw = UIO_READ;
3747         auio.uio_segflg = UIO_USERSPACE;
3748         auio.uio_td = nd.nl_td;
3749         auio.uio_offset = 0;
3750         if ((error = copyin(uap->iovp, iov, iovlen)))
3751                 goto done;
3752         auio.uio_resid = 0;
3753         for (i = 0; i < uap->iovcnt; i++) {
3754                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3755                         error = EINVAL;
3756                         goto done;
3757                 }
3758                 auio.uio_resid += iov->iov_len;
3759                 iov++;
3760         }
3761         cnt = auio.uio_resid;
3762         error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3763         cnt -= auio.uio_resid;
3764         uap->sysmsg_result = cnt;
3765 done:
3766         vput(vp);
3767         nlookup_done(&nd);
3768         if (needfree)
3769                 FREE(needfree, M_IOV);
3770         return(error);
3771 }
3772
3773 /*
3774  * Syscall to delete a named extended attribute from a file or directory.
3775  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3776  */
3777 int
3778 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3779 {
3780         char attrname[EXTATTR_MAXNAMELEN];
3781         struct nlookupdata nd;
3782         struct vnode *vp;
3783         int error;
3784
3785         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3786         if (error)
3787                 return(error);
3788
3789         vp = NULL;
3790         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3791         if (error == 0)
3792                 error = nlookup(&nd);
3793         if (error == 0)
3794                 error = ncp_writechk(&nd.nl_nch);
3795         if (error == 0)
3796                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3797         if (error) {
3798                 nlookup_done(&nd);
3799                 return (error);
3800         }
3801
3802         error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3803         vput(vp);
3804         nlookup_done(&nd);
3805         return(error);
3806 }
3807
3808 /*
3809  * Determine if the mount is visible to the process.
3810  */
3811 static int
3812 chroot_visible_mnt(struct mount *mp, struct proc *p)
3813 {
3814         struct nchandle nch;
3815
3816         /*
3817          * Traverse from the mount point upwards.  If we hit the process
3818          * root then the mount point is visible to the process.
3819          */
3820         nch = mp->mnt_ncmountpt;
3821         while (nch.ncp) {
3822                 if (nch.mount == p->p_fd->fd_nrdir.mount &&
3823                     nch.ncp == p->p_fd->fd_nrdir.ncp) {
3824                         return(1);
3825                 }
3826                 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
3827                         nch = nch.mount->mnt_ncmounton;
3828                 } else {
3829                         nch.ncp = nch.ncp->nc_parent;
3830                 }
3831         }
3832
3833         /*
3834          * If the mount point is not visible to the process, but the
3835          * process root is in a subdirectory of the mount, return
3836          * TRUE anyway.
3837          */
3838         if (p->p_fd->fd_nrdir.mount == mp)
3839                 return(1);
3840
3841         return(0);
3842 }
3843