Network Working Group                                         M. Eriksen
Internet-Draft                                                 J. Fields
Expires: November 16, 2006 February 23, 2007                                          CITI
                                                            May 15,
                                                         August 22, 2006

               Mapping Between NFSv4 and Posix Draft ACLs

Status of this Memo

   By submitting this Internet-Draft, each author represents that any
   applicable patent or other IPR claims of which he or she is aware
   have been or will be disclosed, and any of which he or she becomes
   aware will be disclosed, in accordance with Section 6 of BCP 79.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups.  Note that
   other groups may also distribute working documents as Internet-

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as "work in progress."

   The list of current Internet-Drafts can be accessed at

   The list of Internet-Draft Shadow Directories can be accessed at

   This Internet-Draft will expire on November 16, 2006. February 23, 2007.

Copyright Notice

   Copyright (C) The Internet Society (2006).


   A number of filesystems and applications support ACLs based on a
   withdrawn POSIX draft [2].  Those ACLs differ significantly from NFS
   version 4 (NFSv4) ACLs [1].  We describe how to translate between the two
   types of ACLs.

1.  Introduction

   Access Control Lists (ACLs) are used to specify fine-grained access rights to file
   system objects.  An ACL is a list of Access Control Entries (ACEs),
   each specifying an entity (such as a user) and some level of access
   for that entity.

   In the following sections we describe two ACL models: NFSv4 ACLs, ACLs and ACLs based on a
   withdrawn POSIX draft.  We will refer to the latter as "POSIX ACLs".
   Since NFSv4 ACLs are more fine-grained than POSIX ACLs, it is not
   possible in general to map an arbitrary NFSv4 ACL to a POSIX ACL with
   the same semantics.  However, it is possible to map any POSIX ACL to
   a NFSv4 ACL with nearly identical semantics, and it is possible to
   map any NFSv4 ACL to a POSIX ACL in a way that preserves certain
   guarantees.  We will explain how to do this, and give guidelines for
   clients and servers performing such translation.

2.  NFSv4 ACLs

   An NFSv4 ACL is an ordered sequence of ACEs, each having an entity, a
   type, some flags, and an access mask.

   The entity may be the name of a user or group, or may be one of a
   small set of special entities.  Among the special entities are
   "OWNER@" (the current owner of the file), "GROUP@" (the group
   associated with the file), and "EVERYONE@".

   The type

   An ACL may be have a "type" of ALLOW or DENY.  (AUDIT or ALARM are also
   allowed, but they are not relevant to our discussion).

   The access mask has 14 separate bits, including bits to control read,
   write, execute, append, ACL modification, file owner modification,
   etc.; consult [1] for the full list.

   Of the flags, four are relevant here.  The ACE4_IDENTIFIER_GROUP flag
   is used to indicate that the entity name is the name of a group.  The
   other three concern inheritance: ACE4_DIRECTORY_INHERIT_ACE indicates
   that the ACE should be added to new subdirectories of the directory;
   ACE4_FILE_INHERIT_ACE does the same for new files; and
   ACE4_INHERIT_ONLY indicates that the ACE should be ignored when
   determining access to the directory itself.

   The NFSv4 ACL permission-checking algorithm is straightforward.
   Assume a
   First, assume a requester asks for access, as access specified by a single bit
   in the access bitmask.  We allow the access if the first ACE in the
   ACL that matches the requester and that has that bit set is an ALLOW
   ACE, and we deny the access if the first such ACE is a DENY ACE.  If
   no matching ACE has the bit in question set, behaviour access is undefined. normally

   an a requester asks for access requiring multiple bits from the
   access bitmask simutaneously, then we allow the access mask consisting of more than one bit is requested, it
   succeeds if and only if
   each bit in the mask is allowed. requested bitmask would be allowed individually.

   We refer the reader to [1] for further details.


   A number of operating systems implement ACLs based on the withdrawn
   POSIX 1003.1e/1003.2c Draft Standard 17 [2].  We will refer to such
   ACLs as "POSIX ACLs". ACLs", though they are not part of any published POSIX

   POSIX ACLs use access masks with only the traditional "read",
   "write", and "execute" bits.  Each ACE in a POSIX ACL is one of five
   and ACL_OTHER.  Each ACL_USER ACE has a uid associated with it, and
   each ACL_GROUP ACE has a gid associated with it.  Every POSIX ACL
   must have exactly one ACL_USER_OBJ, ACL_GROUP_OBJ, and ACL_OTHER ACE,
   and at most one ACL_MASK ACE.  The ACL_MASK ACE is required if the
   ACL has any ACL_USER or ACL_GROUP ACEs.  There may not be two
   ACL_USER ACEs with the same uid, and there may not be two ACL_GROUP
   ACEs with the same gid.

   Given a POSIX ACL and a requester asking for access, permission is
   determined as follows: by consulting the ACEs in the order ACL_USER_OBJ,
   denying access based on the first ACE encountered that the requester
   matches, except that we never allow the ACL_USER, ACL_OWNER_OBJ, or
   ACL_GROUP objects to grant more than the ACL_MASK object does, and in
   the case of ACL_GROUP_OBJ and ACL_GROUP ACEs, we allow access if any
   one of those ACEs allows access.

   In more detail:

   1.  If the requester is the file owner, then allow or deny access
       depending on whether the ACL_USER_OBJ ACE allows or denies it.

   2.  if the requester matches the file's group, and the ACL mask ACE
       would deny the requested access, then skip to step 5.  Otherwise,

   3.  if the requester's uid matches the uid of one of the ACL_USER
       ACEs, then allow or deny access depending on whether the
       ACL_USER_OBJ ACE allows or denies it.  Otherwise,


   4.  Consider the set of all ACL_GROUP ACEs whose gid the requester is
       a member of.  Add to that set the ACL_GROUP_OBJ ACE, if the
       requester is also a member of the file's group.  Allow access if
       any ACE in the resulting set allows access.  If the set of
       matching ACEs is nonempty, and none allow access, then deny
       access.  Otherwise, if the set of matching ACEs is empty,

   5.  if the requester's access mask is allowed by the ACL_OTHER ACE,
       then grant access.  Otherwise, deny access.

   The above description omits one detail: in steps (2) and (3), the
   requested bits must be granted both by the matching ACE and by the
   ACL_MASK ACE.  The ACL_MASK ACE thus limits the maximum permissions
   which may be granted by any ACL_USER or ACL_GROUP ACE, or by the

   Each file may have a single POSIX ACL associated with it, used to
   determine access to that file.  Directories, however, may have two
   ACLs: one, the "access ACL", used to determine access to the
   directory, and one, the "default ACL", used only as the ACL to be
   inherited by newly created objects in the directory.

4.  Ordering of NFSv4 and POSIX ACLs

   POSIX ACLs are unordered--the order in which the POSIX access-
   checking algorithm considers the entries is determined entirely by
   the type of the entries, so the entries don't need to be kept in any
   particular order.

   By contrast, the meaning of an NFSv4 ACL can be dramatically changed
   by modifying the order that the entries are listed in.

   In the following, we will say that an NFSv4 ACL is in the "canonical
   order" if its entries are ordered in the order that the POSIX
   algorithm would consider them.  That is, with all OWNER@ entries
   first, followed by user entries, followed by GROUP@ entries, followed
   by group entries, with all EVERYONE@ entries at the end.

5.  A Minor Eccentrity of POSIX ACLs

   We will see below that it is possible to find an NFSv4 ACL with
   precisely the same effect as any given POSIX ACL, with one extremely
   minor exception: if a requester that is a member of more than one
   group listed in the ACL requests multiple bits simultaneously, the
   POSIX algorithm requires all of the bits to be granted simultaneously
   by one of the group ACEs.  Thus a POSIX ACL such as

     ACL_USER_OBJ: ---
     ACL_GROUP_OBJ: ---
     g1: r--
     g2: -w-
     ACL_MASK: rw-
     ACL_OTHER: ---

   will prevent a user that is a member of groups g1 and g2 from opening
   a file for both read and write, even though read and write would be
   individually permitted.

   The NFSv4 ACL permission-checking algorithm has the property that it
   permits a group of bits whenever it would permit each bit
   individually, so it is impossible to mimic this behaviour with an
   NFSv4 ACL.

6.  Mapping POSIX ACLs to NFSv4 ACLs

6.1.  Requirements

   In the next section we give an example of a mapping of POSIX ACLs
   into NFSv4 ACLs.  We permit a  A server or client to may use a different mapping, provided but
   the mapping meets should meet the following requirements:

   It must map the POSIX ACL to an NFSv4 ACL with identical access
   semantics, ignoring the minor exception described in the previous

   It must map the read mode bit to ACE4_READ_DATA, the write bit to
   directories), and the EXECUTE bit to ACE4_EXECUTE.  It should also
   unconditionally, and allow ACE4_WRITE_ACL and ACE4_WRITE_ATTRIBUTES
   to the owner.  The handling of other NFSv4 mode bits may depend on
   the implementation, but it is preferable to leave them unused.

   It should avoid using DENY ACEs.  If DENY ACEs are required, it
   should attempt to place them at the beginning.  (This is not always

   For simplicity's sake, the translator may choose to handle

   The resulting NFSv4 ACL must take into account the mask ACE, by first applying
   ensuring that it to the USER, GROUP, and GROUP_OBJ ACEs, and then
   mapping does not give the resulting ACL.  However, that will result group file owner or any users or
   groups named in an the ACL from
   which it is impossible to determine the original value of the mask or
   of the masked USER, GROUP, and GROUP_OBJ bitmasks.  If more permissions than permitted by the resulting
   ACL is later translated back mask.
   It would also be possible to specify a POSIX ACL, mapping that encoded the translator will
   assume mask
   in such a way that the original value of the mask is the union of the bitmasks
   permitted to any USER, GROUP, or GROUP_OBJ.  If that would could be
   incorrect, the original translation should not modify recovered
   by someone that knew the bitmasks of ACL was produced by our algorithm.  However,
   the USER, GROUP, and GROUP_OBJ bitmasks, added complexity and should instead use
   additional DENY ACEs as necessary to give the effect fragility of the mask.  It
   should also arrange for the first GROUP@ ACE to be such a DENY ACE whose
   bitmask mapping is determined by not worth the mask, allowing that ACE to be used to
   small benefit of preserving the original mask value. information, so we do not
   attempt that here.

6.2.  Example POSIX->NFSv4 Mapping

   We now describe an algorithm which maps any POSIX ACL to an NFSv4 ACL
   with the same semantics, meeting the above requirements.

   First, translate modify all ACL_USER, ACL_GROUP, and ACL_GROUP_OBJ ACEs by
   removing any permissions not granted by the mask ACE.  The mask ACE
   may then be ignored for the rest of this process.

   Translate the uid's and gid's on the ACL_USER and ACL_GROUP ACEs into
   NFSv4 names, using directory services, etc., as appropriate, and
   translate ACL_USER_OBJ, ACL_GROUP_OBJ, and ACL_OTHER to the special
   NFSv4 names "OWNER@", "GROUP@", and "EVERYONE@", respectively.

   Next, map each POSIX ACE (excepting any mask ACE) in the given POSIX
   ACL to an NFSv4 ALLOW ACE with an entity determined as above, and
   with a bitmask determined from the permission bits on the POSIX ACE
   as follows:

   1.  If the read bit is set in the POSIX ACE, then set ACE4_READ_DATA.

   2.  If the write bit is set in the POSIX ACE, then set
       ACE4_WRITE_DATA and ACE4_APPEND_DATA.  If the object carrying the
       ACL is a directory, set ACE4_DELETE_CHILD as well.

   3.  If the execute bit is set in the POSIX ACE, then set


   5.  If the ACE is for the special "OWNER@" entity, set ACE4_WRITE_ACL

   6.  Clear all other bits in the NFSv4 bitmask.

   In addition, we set the GROUP flag in each ACE which corresponds to a
   named group (but not in the GROUP@ ACE, or any of the other special
   entity ACEs).

   At this point, we've replaced the POSIX ACL by an NFSv4 ACL with the
   same number of ACEs (ignoring any mask ACE), all of them ALLOW ACEs.

   Order this NFSv4 ACL in the canonical order: OWNER@, users, GROUP@,
   groups, then EVERYONE@.

   If the bitmasks in the resulting ACEs are non-increasing (so no ACE
   allows a bit not allowed by a previous ACE), then we can skip the
   next step.

   Otherwise, we need to insert additional DENY ACE's to emulate the
   first-match semantics of the POSIX ACL permission-checking algorithm:

   1.  If an ACL_USER_OBJ, ACL_OTHER, or ACL_USER ACE fails to grant
       some permissions that are granted later in the ACL, then that ACE
       must be prepended preceded by a single DENY ACE.  The DENY ACE should have
       the same entity and flags as the corresponding ALLOW ACE, but the
       bitmask on the DENY ACE should be the bitwise NOT of the bitmask
       on the ALLOW ACE, except that the ACE4_WRITE_OWNER, ACE4_DELETE,
       should be cleared, and the ACE4_DELETE_CHILD bit should be
       cleared on non-directories.  (Also, in the xdr-encoded ACL that
       is transmitted, all bits not defined in the protocol should be

   2.  All of the ACL_GROUP_OBJ and ACL_GROUP ACEs are consulted by the
       POSIX algorithm before determining permissions.  To emulate this
       behaviour, instead of adding a single DENY before corresponding
       GROUP@ or named group ACEs, we insert a list of DENY ACEs after
       the list of GROUP@ and named group ACEs.  Each DENY ACE is
       determined from its corresponding ALLOW ACE exactly as in the
       previous step.  As before, these DENY aces ACEs should only be added
       when they are necessitated by an ACE that is less permissive than
       the final EVERYONE@ ace.

   Finally, we enforce the POSIX mask ACE by prepending each ALLOW ACE
   for a named user, GROUP@, or named group, ACE.

   On directories with a single DENY ACE
   whose entity and flags are default ACLs, we translate the same default ACL as those for the corresponding
   above, but whose bitmask is the inverse of set the bitmask determined
   and ACE4_FILE_INHERIT_ACE flags on every ACE in the mask ACE, resulting ACL.
   On directories with the inverse calculated as described above.
   In the case of named users, these DENY aces may be coalesced with any
   existing prepended DENY aces.  The DENY aces are omitted entirely if
   they would have no affect, or if the mask ACE has the same bitmask as
   the maximum of the affected ACEs.  (With the one exception that if
   the POSIX ACL posesses exactly 4 ACEs, then a mask-derived DENY ace
   should be inserted before the GROUP@ ace, even if it would not
   otherwise be.)

   Regardless of what scheme is used to represent the mask, the receiver
   will use the first GROUP@ DENY ace to determine the value of the mask
   (if it is different from the union of the bitmasks on the affected
   ACEs), and use the relevant ALLOWs to determine the pre-mask values
   of user and group ACEs.

   The implementation may also choose to just mask out the bitmasks on
   the relevant ALLOW ACEs.  This will produce a simpler ACL (in
   particular, an ACL that usually requires no DENY ACE's), at the
   expense of losing some ACL information after a chmod.

   On directories with default ACLs, we translate the default ACL as
   and ACE4_FILE_INHERIT_ACE flags on every ACE in the resulting ACL.
   On directories with both default and access ACLs, we translate both default and access ACLs, we translate the
   two ACLs and then concatenate them.  The order of the concatenation
   is unimportant.

7.  Mapping NFSv4 ACLs to POSIX ACLs

7.1.  Requirements

   Any mapping of NFSv4 ACLs to POSIX ACLs must map any NFSv4 ACL that
   is semantically equivalent to a POSIX ACL (with the exception of the
   "minor inaccuracy" mentioned above) to the an equivalent POSIX ACL.  It
   should also extract the mask correctly; as the mask doesn't affect

   However, a more difficult problem is presented by NFSv4 ACLs that are
   not precisely equivalant to any POSIX ACL.

   The only way that the semantics NFSv4 protocol gives servers to indicate that
   they support only a subset of the NFSv4 ACL, and as there ACL model is more than one way the
   mask might be encoded, we require "aclsupport"
   attribute, which allows a convention for this.
   Specifically: we require server to advertise that the mask be computed as the bitmask
   used on the first GROUP@ DENY it only supports
   certain ACE which precedes any GROUP@ allow
   ACE, unless no such DENY ACE exists, in which case the mask must types.  This allows a server to report that it only
   supports ALLOW ACEs, or that it does not support AUDIT or ALARM ACEs
   (which will be
   computed as the union of the bitmasks allowed case for most servers with only POSIX ACLs).  But
   it does not give a way to all named users,
   groups, and GROUP@ (where by the "bitmask allowed to" an entity we
   mean the maximum bitmask that claim support for more complex subsets of
   the ACL would permit model.

   While it is possible for a server to reject any user
   matching the entity).

   Implementations may vary in how they deal with NFSv4 ACLs that are do not precisely semantically equivalent fit
   its ACL model, this places a large burden on clients and users, since
   the server has no way to any POSIX ACL.  In explain why it rejected a particular they may return errors for such ACLs instead of attempting ACL.
   Therefore, it is preferable to map them.  However, when be more forgiving, whenever that is
   possible without compromising security,
   they should attempt and to be forgiving. limit any restrictions
   to those that are easily documented and verified by users.

   The language of [1] allows a server some flexibility in handling ACLs
   that it cannot enforce completely accurately, as long as it adheres
   to "the guiding principle... that the server must not accept ACLs
   that appear to make [a file] more secure than it really is."


   ACLs with arbitrary sequences of ALLOWs and DENYs may be particularly
   troublesome; but note that an NFSv4 ACL consisting entirely of ALLOW
   ACLs can always be transformed into a POSIX-equivalent ACL by first
   sorting it into the canonical order, and then inserting DENY ACEs as
   necessary to ensure POSIX first-match semantics.  Since inserting
   DENY ACEs can only restrict access, it is safe for a server to do

   We require any server to

   Therefore servers should accept, at least, any NFSv4 ACL that
   consists entirely of ALLOW ACLs.

   Clients should also be at least as forgiving, to promote
   interoperability when heterogeneous clients share files.

7.2.  Example NFSv4->POSIX Mapping

   We now give an example of an algorithm that meets the above
   requirements.  We assume it is to be used by a server mapping client-
   provided NFSv4 ACLs to POSIX ACLs it can store in its filesystem, so
   the translation errs on the side of making the ACL less permissive.

   Given an NFSv4 ACL, first calculate more restrictive.

   In fact, if we ignore some loss of information in the mask by taking the bitmask
   from ACE, this
   mapping takes an NFSv4 ACL to the first GROUP@ DENY ACE from unique most permissive POSIX ACL
   that is no more permissive than the original given NFSv4 ACL, ACL.

   Before starting, if it
   exists.  After doing so, remove that DENY ACE, and clear the bits ACL in
   its bitmask from any DENY ACE question is for a named user, group, or GROUP@
   which precedes an ALLOW ACE for the same entity.

   In the case where there is no such GROUP@ DENY ACE, continue through
   the rest of the algorithm and then calculate the mask as the union of
   the calculated permissions of all named users, group, and the

   Given an NFSv4 ACL, sort directory, we split
   it into canonical order (OWNER@ ACEs first,
   then user ACEs, then GROUP@ ACEs, then group ACEs, then EVERYONE@
   ACEs.)  Also, sort the GROUP@ two ACLs, one purely effective and group ACEs that all ALLOW ACEs
   precede all DENY ACEs.  To do so, take advantage of the following
   observations: one purely inherited, as

   1.  If two consecutive  ACEs with no inheritance flags are either put in the purely effective

   2.  Aces with FILE_INHERIT and DIRECTORY_INHERIT both ALLOW ACEs, or set are put in
       both DENY
       ACEs, then we can swap their order without changing the effect of effective and the inherited ACL

       set are put only in the inherited ACL.

   2.  If it would

   Other combinations of ineritance flags may be impossible for a single user rejected or silently
   modified to match both one of the
       entities on two consecutive ACEs, above.

   The main algorithm that follows is then we can swap their order
       without changing performed on each ACL, with
   one used to set the effect of effictive ACL, and one the default ACL.

   3.  If an ALLOW

   First, we calculate the OTHER mode as follows:

   1.  Initialize the bitmasks other_allow and other_deny both to zero.

   2.  For each ACE is immediately followed by a DENY ACE, then
       swapping in the order of ACL, starting from the two ACEs will top:

       1.  If the ACE is not make an EVERYONE@ ACE, ignore it and move to the ACL any more

           next ACE.

       2.  If a DENY the ACE is immediately followed by an EVERYONE@ ALLOW ACE, then
       swapping the order of the two ACEs will not make the ACL any more
       permissive, *if* we modify the bitmask on the ALLOW ACE by
       clearing add to other_allow
           any bits that are set in this ACE but not set in other_deny.

       3.  If the DENY ACE.

   The second observation ACE is the trickiest: it may usually be safe to
   assume that two distinct user names cannot match the same user.  An
   implementation with knowledge about group memberships or about the
   current value of the file owner might also use that information, but
   if it does so it will produce a translation that is no longer
   accurate after owners or group memberships change.

   Fortunately, observations 1, 3, and 4 are sufficient to sort any ACL
   into canonical order, so a paranoid implementation can simply ignore
   number 2 completely, while an implementation willing to sacrifice
   some accuracy may choose to do something more complex.

   Ensure that the resulting ACL posesses at least one each of OWNER@,
   GROUP@, and EVERYONE@ ACEs, by inserting an ALLOW ACE with a zero
   bitmask if necessary in the correct position.

   Next, for each entity, calculate a bitmask for that entity as
   follows: Starting with the first ACE for that entity (ignoring all
   previous ACEs), perform the NFSv4 ACL-checking algorithm for a user
   that is assumed to match the entity on every DENY ACE that a user
   matching the given entity might match, but is assumed to match only
   those entities on ALLOW ACEs that *any* user matching the current
   entity must match.

   Finally, construct the POSIX ACL by translating NFSv4 entity names ACE, then add to
   uid's and gid's (and handling special entities other_deny
           any bits set in the obvious way),
   then assign a POSIX bitmask determined by the NFSv4 bitmask
   calculated this ACE but not set in other_allow.

   3.  Discard other_deny.  Set the previous step; the bitmask calculation should use USER_OBJ mask from other_allow using
       the inverse of the mapping described previously in the POSIX-to-NFSv4 POSIX-to-
       NFSv4 mapping, erring on the side of denying bits if it cannot
       determine a sensible mapping.  However, if certain bits simply
       cannot be mapped in a reasonable way to mode bits, the server may
       simply ignore them rather than returning an error.  (For example,
       the server should deny write if either ACE4_WRITE_DATA or
       ACE4_APPEND_DATA are denied.  But it may choose to ignore
       ACE4_READ_ATTRIBUTES entirely.)

   The resulting mapping errs on the side of creating a more restrictive
   ACE.  However entirely; though in that case it can be modified may at
       least want to produce a mapping that errs on treat specially the side of permissiveness, for the purposes of translating case where such bits are
       explicitly denied by some DENY ACE.)

   Note that the bits determined above are exactly the maximum bits that
   will always be permitted to a user that doesn't match the file owner
   or group, or any of the named owners or groups.  Thus this choice of
   the OTHER mode is exactly the maximum choice we can safely make.

   Next we calculate the GROUP_OBJ and GROUP masks.

   1.  Initialize to zero an allow and deny bitmask for each GROUP_OBJ
       and for each GROUP mask.

   2.  For each ACE in the ACL, starting from the top:

       1.  If the ACE is an OWNER@ or named user ACE, ignore it and move
           to the next ACE.

       2.  If the ACE is an EVERYONE@ ALLOW ACE, then, for each GROUP or
           GROUP_OBJ allow mask, set the bits allowed in the EVERYONE
           ACE but not already in this GROUP or GROUP_OBJ's deny mask.

       3.  If the ACE is an EVERYONE@ DENY ACE, then, for each GROUP or
           GROUP_OBJ deny mask, set the bits denied in the EVERYONE ACE
           but not already allowed in this GROUP or GROUP_OBJ's deny

       4.  If the ACE is a GROUP or GROUP@ ALLOW ACE, then set the allow
           bits in the corresponding GROUP or GROUP_OBJ allow mask that
           are allowed by this ACE but not already denied by the
           corresponding GROUP or GROUP_OBJ deny mask.

       5.  If the ACE is a GROUP or GROUP@ DENY ACE, then set the deny
           bits in the corresponding GROUP or GROUP_OBJ deny mask that
           are denied by this ACE but not already allowed by the
           corresponding GROUP or GROUP_OBJ allow mask.  Call the
           resulting deny mask "m".  In each GROUP or GROUP_OBJ deny
           mask, set every bit that is in m and not already in that
           GROUP or GROUP_OBJ allow mask.

   3.  Having calculated allow and deny masks for GROUP_OBJ and each
       GROUP, we now set the corresponding modes from the allow masks as
       we did in the last step of the USER_OBJ mask calculation above.

   Note that the bits thus determined for a group are exactly the
   maximum bits that will always be permitted to a user that matches the
   group in question, and that is denied any bits that could be denied
   by matching other groups, without out being allowed bits by matching
   any such groups.  This is the most permissive mode we can choose that
   will never permit more permissions than the original NFSv4 ACL, for
   any possible choice of group memberships.

   An implementation with special knowledge about the current gowning
   group or about group memberships may choose to use that knowledge to
   calculate a more permissive mode.  However, doing so may render
   resulting POSIX ACL inaccurate after the owning group changes, or
   after any group memberships change.

   Next, we calculate USER modes by first calculating allow and deny
   masks for each USER as above, this time assuming we are a user that
   does not match the file owner, that matches no user except for the
   one user under consideration, and that matches groups only when they
   would deny some permissions that they have not allowed yet.  (To
   ensure this last step it will also be necessary to maintain group
   allow and deny ACEs, as we did in the previous calculation.)  We omit
   the detailed steps, which are similar.  Again, the implementation may
   choose to use special knowledge about group memberships at the risk
   of increased complexity and of loss of some accuracy.

   Next, we calculate the USER_OBJ mode by calculating allow and deny
   masks for a user that matches the file owner and any user or group
   that denies bits that it does not first allow.

   Finally, if the resulting ACL has any named user or group ACEs, add a
   mask ACE with bitmask equal to the union of the calculated
   permissions of all named users, group, and the GROUP_OBJ ACE.

   The resulting mapping errs on the side of creating a more restrictive
   ACE.  However it can be modified to produce a mapping that errs on
   the side of permissiveness, for the purposes of translating a server-
   provided NFSv4 ACL to a POSIX ACL to present to a user or
   application, as follows:

   1.  When performing the final mapping from the allow bitmask to a server-
       mode, we instead using a mapping that errs on the side of
       permissiveness; for example, we allow write permissions even if
       only one of WRITE_DATA, APPEND_DATA, or (in the case of
       directories) DELETE_CHILD is allowed.

   2.  Wherever in the above we pessimistically assume that a user will
       match any entity that has permissions denied to it before they
       are first allowed, we instead assume that the user will match any
       entity that has permissions allowed to it before they are first

   Once again, the resulting mapping may be seen to produce the unique
   (up to choice of mask) POSIX ACL which is the most restrictive among
   all POSIX ACLs no more restrictive than the given NFSv4 ACL.

   Note that the above algorithms may be optimized in a number of ways:
   for example, although they are described in terms of multiple passes,
   it will be simpler and more efficient to calculate the entire POSIX
   ACL in a single pass.

8.  Backwards Compatibility

   Previous versions of this document recommended a different
   POSIX->NFSv4 mapping, which enforces POSIX semantics by inserting
   DENYs into the ACL even when those DENY's would have no effect, and
   which represents the POSIX mask ACE using additional DENYs.  The
   resulting ACLs are overly complex and create problems for Windows
   clients, because the default Windows ACL editor prefers to order
   DENYs before ALLOWs.

   The NFSv4 to POSIX mapping we describe in this document can accept
   the NFSv4 ACLs produced by the old mapping.

   However, previous versions of this document also recommended
   accepting only NFSv4 ACLs that were precisely those produced by the
   old POSIX->NFSv4 mapping; therefore, existing implementations of that
   recommendation will reject the NFSv4 ACLs produced by the newer

   We strongly recommend fixing implementations to accept a wider range
   of NFSv4 ACLs.  However, we briefly document the old mapping here in
   case that is impossible:

   Names, bitmasks, and flags are determined as in the the current

   Whenever the following instructions requiring taking "the complement"
   of an NFSv4 ACL to a POSIX ACL to present to a user or
   application, bitmask, do so as follows:

   1.  When sorting ACEs, ALLOW ACEs can always be moved towards first, take the
       start bitwise NOT of
   the ACL, but a DENY ACE can be moved towards bitmask.  Then clear the start
   Also, clear the ACL only as long as we ACE4_DELETE_CHILD bit on non-directories, and clear
   any of bits not defined in the protocol.

   Create one ALLOW ACE for each entity (OWNER@, GROUP@, and EVERYONE@,
   and each user and group named in the given POSIX ACL).  After each
   OWNER@, EVERYONE@, and named user ACE, append a DENY ACE's ACE with the
   same entity and flags as the corresponding ALLOW ACE, but with
       bits that are set in to the complement (as defined above) of the intervening ALLOW ACEs.

   2.  When calculating ACE.

   Do the NFSv4 bitmask same for each entity, err on the
       side GROUP@ and named group ACE, but instead of assuming that
   inserting each new DENY ACE after the corresponding ALLOW ACEs apply and that ACE, insert
   all of the DENY ACEs don't,
       with at the one exception end of the list of GROUP@ and named group
   ACEs, in the same order that when calculating the GROUP@ and named group bitmasks, ALLOW ACEs for groups other than
   occur in.

   Finally, prepend each GROUP@, named user, and named group ACE by a
   single DENY whose entity and flags are the one under
       consideration should be ignored.

   3.  When mapping same as the NFSv4 corresponding
   ALLOW, but whose bitmask to POSIX mode bits, err on is the
       side complement (as defined above) of allowing access.

8. the
   bitmask determined from the mask ACE in the given POSIX ACL.  Skip
   this step if the given POSIX ACL has no mask ACE.

9.  Security Considerations

   Any automatic mapping from one ACL model to another must provide
   guarantees as to how the mapping affects the meaning of ACLs, or risk
   misleading users about the permissions set on filesystem objects.
   For this reason, caution is recommended when implementing this
   mapping.  It is better to return errors than to break any such

   That said, there may be cases where small losses in accuracy can
   avoid dramatic interoperability and usability problems; as long as
   the losses in accuracy are clearly documented, these tradeoffs may be
   found acceptable.

   For example, a server unable to support all of the NFSv4 mode bits
   does not have a way to communicate its exact limitations to clients,
   so clients (and users) may be unable to recover from such errors.
   For this reason we recommend ignoring bitmask bits that the server is
   completely unable to map to mode bits, and advertising this fact
   loudly in at least when no ACE
   explicitly contradicts the server documentation. server's default behavior.  If this is
   considered insufficient, we should add to the NFSv4 protocol
   additional attributes necessary to advertise the server's

   Note also that this any ACL mapping also requires mapping between NFSv4
   usernames and local id's.  When the mapping of id's depends on remote
   services, the method used for the mapping must be at least as secure
   as the method used to set or get ACLs.


10.  References

   [1]  Shepler, S., Callaghan, B., Robinson, D., Thurlow, R., Beame,
        C., Eisler, M., and D. Noveck, "Network File System (NFS)
        version 4 Protocol", RFC 3530, April 2003.

   [2]  Institute of Electrical and Electronics Engineers, Inc., "IEEE
        Draft P1003.1e", October 1997,

Authors' Addresses

   Marius Aamodt Eriksen
   U. of Michigan Center for Information Technology Integration


   J. Bruce Fields
   U. of Michigan Center for Information Technology Integration


Intellectual Property Statement

   The IETF takes no position regarding the validity or scope of any
   Intellectual Property Rights or other rights that might be claimed to
   pertain to the implementation or use of the technology described in
   this document or the extent to which any license under such rights
   might or might not be available; nor does it represent that it has
   made any independent effort to identify any such rights.  Information
   on the procedures with respect to rights in RFC documents can be
   found in BCP 78 and BCP 79.

   Copies of IPR disclosures made to the IETF Secretariat and any
   assurances of licenses to be made available, or the result of an
   attempt made to obtain a general license or permission for the use of
   such proprietary rights by implementers or users of this
   specification can be obtained from the IETF on-line IPR repository at

   The IETF invites any interested party to bring to its attention any
   copyrights, patents or patent applications, or other proprietary
   rights that may cover technology that may be required to implement
   this standard.  Please address the information to the IETF at

Disclaimer of Validity

   This document and the information contained herein are provided on an

Copyright Statement

   Copyright (C) The Internet Society (2006).  This document is subject
   to the rights, licenses and restrictions contained in BCP 78, and
   except as set forth therein, the authors retain all their rights.


   Funding for the RFC Editor function is currently provided by the
   Internet Society.