kernel/syscall/umount.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26
  27 /*      Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T     */
  28 /*        All Rights Reserved   */
  29
  30 /*
  31  * Portions of this source code were derived from Berkeley 4.3 BSD
  32  * under license from the Regents of the University of California.
  33  */
  34
  35 #include <sys/types.h>
  36 #include <sys/t_lock.h>
  37 #include <sys/param.h>
  38 #include <sys/errno.h>
  39 #include <sys/fstyp.h>
  40 #include <sys/kmem.h>
  41 #include <sys/systm.h>
  42 #include <sys/mount.h>
  43 #include <sys/vfs.h>
  44 #include <sys/cred.h>
  45 #include <sys/vnode.h>
  46 #include <sys/cmn_err.h>
  47 #include <sys/debug.h>
  48 #include <sys/pathname.h>
  49 #include <sys/policy.h>
  50 #include <sys/zone.h>
  51
  52 #define UMOUNT2_SET_ERRNO(e, is_syscall) ((is_syscall) ? set_errno((e)) : (e))
  53
  54 /*
  55  * The heart of the umount2 call - it is pulled out to allow kernel
  56  * level particpation when the only reference is the vfs pointer.
  57  *
  58  * Note that some of the callers may not be in the context of a
  59  * syscall (created by zthread_create() for example) and as such
  60  * may not have an associated curthread->t_lwp. This is handled
  61  * by is_syscall.
  62  */
  63 int
  64 umount2_engine(vfs_t *vfsp, int flag, cred_t *cr, int is_syscall)
  65 {
  66         int     error;
  67
  68         /*
  69          * Protect the call to vn_vfswlock() with the vfs reflock.  This
  70          * ensures vfs_vnodecovered will either be NULL (because someone
  71          * beat us to the umount) or valid (because vfs_lock() prevents
  72          * another umount from getting through here until we've called
  73          * vn_vfswlock() on the covered vnode).
  74          *
  75          * At one point, we did the non-blocking version (vfs_lock()),
  76          * and if it failed, bailed out with EBUSY.  However, dounmount()
  77          * calls vfs_lock_wait() and we drop the vfs lock before calling
  78          * dounmount(), so there's no difference between waiting here
  79          * for the lock or waiting there because grabbed it as soon as
  80          * we drop it below.  No returning with EBUSY at this point
  81          * reduces the number of spurious unmount failures that happen
  82          * as a side-effect of fsflush() and other mount and unmount
  83          * operations that might be going on simultaneously.
  84          */
  85         vfs_lock_wait(vfsp);
  86
  87         /*
  88          * Call vn_vfswlock() on the covered vnode so that dounmount()
  89          * can do its thing.  It will call the corresponding vn_vfsunlock().
  90          * Note that vfsp->vfs_vnodecovered can be NULL here, either because
  91          * someone did umount on "/" or because someone beat us to the umount
  92          * before we did the vfs_lock() above.  In these cases, vn_vfswlock()
  93          * returns EBUSY and we just pass that up.  Also note that we're
  94          * looking at a vnode without doing a VN_HOLD() on it.  This is
  95          * safe because it can't go away while something is mounted on it
  96          * and we're locking out other umounts at this point.
  97          */
  98         if (vn_vfswlock(vfsp->vfs_vnodecovered)) {
  99                 vfs_unlock(vfsp);
 100                 VFS_RELE(vfsp);
 101                 return (UMOUNT2_SET_ERRNO(EBUSY, is_syscall));
 102         }
 103
 104         /*
 105          * Now that the VVFSLOCK in the covered vnode is protecting this
 106          * path, we don't need the vfs reflock or the hold on the vfs anymore.
 107          */
 108         vfs_unlock(vfsp);
 109         VFS_RELE(vfsp);
 110
 111         /*
 112          * Perform the unmount.
 113          */
 114         if ((error = dounmount(vfsp, flag, cr)) != 0)
 115                 return (UMOUNT2_SET_ERRNO(error, is_syscall));
 116         return (0);
 117 }
 118
 119 /*
 120  * New umount() system call (for force unmount flag and perhaps others later).
 121  */
 122 int
 123 umount2(char *pathp, int flag)
 124 {
 125         struct pathname pn;
 126         struct vfs *vfsp;
 127         int error;
 128
 129         /*
 130          * Some flags are disallowed through the system call interface.
 131          */
 132         flag &= MS_UMOUNT_MASK;
 133
 134         /*
 135          * Lookup user-supplied name by trying to match it against the
 136          * mount points recorded at mount time.  If no match is found
 137          * (which can happen if the path to the mount point is specified
 138          * differently between mount & umount, or if a block device were
 139          * passed to umount) then we fall back to calling lookupname()
 140          * to find the vfs.  Doing it this way prevents calling lookupname()
 141          * in most cases and that allows forcible umount to work even if
 142          * lookupname() would hang (i.e. because an NFS server is dead).
 143          */
 144
 145         if (error = pn_get(pathp, UIO_USERSPACE, &pn))
 146                 return (set_errno(error));
 147
 148         /*
 149          * Only a privileged user is allowed to bypass the security
 150          * checks done by lookupname() and use the results from
 151          * vfs_mntpoint2vfsp() instead.  It could be argued that the
 152          * proper check is FILE_DAC_SEARCH but we put it all
 153          * under the mount privilege.  Also, make sure the caller
 154          * isn't in an environment with an alternate root (to the zone's root)
 155          * directory, i.e. chroot(2).
 156          */
 157         if (secpolicy_fs_unmount(CRED(), NULL) != 0 ||
 158             (PTOU(curproc)->u_rdir != NULL &&
 159             PTOU(curproc)->u_rdir != curproc->p_zone->zone_rootvp) ||
 160             (vfsp = vfs_mntpoint2vfsp(pn.pn_path)) == NULL) {
 161                 vnode_t *fsrootvp;
 162
 163                 /* fall back to lookupname() on path given to us */
 164                 if (error = lookupname(pn.pn_path, UIO_SYSSPACE, FOLLOW,
 165                     NULLVPP, &fsrootvp)) {
 166                         pn_free(&pn);
 167                         return (set_errno(error));
 168                 }
 169                 /*
 170                  * Find the vfs to be unmounted.  The caller may have specified
 171                  * either the directory mount point (preferred) or else (for a
 172                  * disk-based file system) the block device which was mounted.
 173                  * Check to see which it is; if it's the device, search the VFS
 174                  * list to find the associated vfs entry.
 175                  */
 176                 if (fsrootvp->v_flag & VROOT) {
 177                         vfsp = fsrootvp->v_vfsp;
 178                         VFS_HOLD(vfsp);
 179                 } else if (fsrootvp->v_type == VBLK)
 180                         vfsp = vfs_dev2vfsp(fsrootvp->v_rdev);
 181                 else
 182                         vfsp = NULL;
 183
 184                 VN_RELE(fsrootvp);
 185
 186                 if (vfsp == NULL) {
 187                         pn_free(&pn);
 188                         return (set_errno(EINVAL));
 189                 }
 190         }
 191         pn_free(&pn);
 192
 193         return (umount2_engine(vfsp, flag, CRED(), 1));
 194 }