lib/util: consolidate module loading into common code
[samba.git] / source4 / ntvfs / posix / pvfs_acl.c
index f5a00c08a8fd93a7fe09dad994db0d4b86013af2..8157d862ec2fdb06d911428ebff5091dd961c87b 100644 (file)
 #include "librpc/gen_ndr/xattr.h"
 #include "libcli/security/security.h"
 #include "param/param.h"
+#include "../lib/util/unix_privs.h"
+#include "lib/util/samba_modules.h"
 
+#if defined(UID_WRAPPER)
+#if !defined(UID_WRAPPER_REPLACE) && !defined(UID_WRAPPER_NOT_REPLACE)
+#define UID_WRAPPER_REPLACE
+#include "../uid_wrapper/uid_wrapper.h"
+#endif
+#else
+#define uwrap_enabled() 0
+#endif
 
 /* the list of currently registered ACL backends */
 static struct pvfs_acl_backend {
@@ -80,18 +90,18 @@ const struct pvfs_acl_ops *pvfs_acl_backend_byname(const char *name)
        return NULL;
 }
 
-NTSTATUS pvfs_acl_init(struct loadparm_context *lp_ctx)
+NTSTATUS pvfs_acl_init(void)
 {
        static bool initialized = false;
-       extern NTSTATUS pvfs_acl_nfs4_init(void);
-       extern NTSTATUS pvfs_acl_xattr_init(void);
+#define _MODULE_PROTO(init) extern NTSTATUS init(void);
+       STATIC_pvfs_acl_MODULES_PROTO;
        init_module_fn static_init[] = { STATIC_pvfs_acl_MODULES };
        init_module_fn *shared_init;
 
        if (initialized) return NT_STATUS_OK;
        initialized = true;
 
-       shared_init = load_samba_modules(NULL, lp_ctx, "pvfs_acl");
+       shared_init = load_samba_modules(NULL, "pvfs_acl");
 
        run_init_functions(static_init);
        run_init_functions(shared_init);
@@ -148,7 +158,7 @@ static NTSTATUS pvfs_default_acl(struct pvfs_state *pvfs,
        NTSTATUS status;
        struct security_ace ace;
        mode_t mode;
-       struct id_mapping *ids;
+       struct id_map *ids;
        struct composite_context *ctx;
 
        *psd = security_descriptor_initialise(req);
@@ -157,21 +167,15 @@ static NTSTATUS pvfs_default_acl(struct pvfs_state *pvfs,
        }
        sd = *psd;
 
-       ids = talloc_zero_array(sd, struct id_mapping, 2);
+       ids = talloc_zero_array(sd, struct id_map, 2);
        NT_STATUS_HAVE_NO_MEMORY(ids);
 
-       ids[0].unixid = talloc(ids, struct unixid);
-       NT_STATUS_HAVE_NO_MEMORY(ids[0].unixid);
-
-       ids[0].unixid->id = name->st.st_uid;
-       ids[0].unixid->type = ID_TYPE_UID;
+       ids[0].xid.id = name->st.st_uid;
+       ids[0].xid.type = ID_TYPE_UID;
        ids[0].sid = NULL;
 
-       ids[1].unixid = talloc(ids, struct unixid);
-       NT_STATUS_HAVE_NO_MEMORY(ids[1].unixid);
-
-       ids[1].unixid->id = name->st.st_gid;
-       ids[1].unixid->type = ID_TYPE_GID;
+       ids[1].xid.id = name->st.st_gid;
+       ids[1].xid.type = ID_TYPE_GID;
        ids[1].sid = NULL;
 
        ctx = wbc_xids_to_sids_send(pvfs->wbc_ctx, ids, 2, ids);
@@ -290,7 +294,7 @@ NTSTATUS pvfs_acl_set(struct pvfs_state *pvfs,
        gid_t old_gid = -1;
        uid_t new_uid = -1;
        gid_t new_gid = -1;
-       struct id_mapping *ids;
+       struct id_map *ids;
        struct composite_context *ctx;
 
        if (pvfs->acl_ops != NULL) {
@@ -303,11 +307,11 @@ NTSTATUS pvfs_acl_set(struct pvfs_state *pvfs,
                return status;
        }
 
-       ids = talloc(req, struct id_mapping);
+       ids = talloc(req, struct id_map);
        NT_STATUS_HAVE_NO_MEMORY(ids);
-       ids->unixid = NULL;
+       ZERO_STRUCT(ids->xid);
        ids->sid = NULL;
-       ids->status = NT_STATUS_NONE_MAPPED;
+       ids->status = ID_UNKNOWN;
 
        new_sd = info->set_secdesc.in.sd;
        orig_sd = *sd;
@@ -327,9 +331,9 @@ NTSTATUS pvfs_acl_set(struct pvfs_state *pvfs,
                        status = wbc_sids_to_xids_recv(ctx, &ids);
                        NT_STATUS_NOT_OK_RETURN(status);
 
-                       if (ids->unixid->type == ID_TYPE_BOTH ||
-                           ids->unixid->type == ID_TYPE_UID) {
-                               new_uid = ids->unixid->id;
+                       if (ids->xid.type == ID_TYPE_BOTH ||
+                           ids->xid.type == ID_TYPE_UID) {
+                               new_uid = ids->xid.id;
                        }
                }
                sd->owner_sid = new_sd->owner_sid;
@@ -345,9 +349,9 @@ NTSTATUS pvfs_acl_set(struct pvfs_state *pvfs,
                        status = wbc_sids_to_xids_recv(ctx, &ids);
                        NT_STATUS_NOT_OK_RETURN(status);
 
-                       if (ids->unixid->type == ID_TYPE_BOTH ||
-                           ids->unixid->type == ID_TYPE_GID) {
-                               new_gid = ids->unixid->id;
+                       if (ids->xid.type == ID_TYPE_BOTH ||
+                           ids->xid.type == ID_TYPE_GID) {
+                               new_gid = ids->xid.id;
                        }
 
                }
@@ -384,6 +388,27 @@ NTSTATUS pvfs_acl_set(struct pvfs_state *pvfs,
                } else {
                        ret = fchown(fd, new_uid, new_gid);
                }
+               if (errno == EPERM) {
+                       if (uwrap_enabled()) {
+                               ret = 0;
+                       } else {
+                               /* try again as root if we have SEC_PRIV_RESTORE or
+                                  SEC_PRIV_TAKE_OWNERSHIP */
+                               if (security_token_has_privilege(req->session_info->security_token,
+                                                                SEC_PRIV_RESTORE) ||
+                                   security_token_has_privilege(req->session_info->security_token,
+                                                                SEC_PRIV_TAKE_OWNERSHIP)) {
+                                       void *privs;
+                                       privs = root_privileges();
+                                       if (fd == -1) {
+                                               ret = chown(name->full_name, new_uid, new_gid);
+                                       } else {
+                                               ret = fchown(fd, new_uid, new_gid);
+                                       }
+                                       talloc_free(privs);
+                               }
+                       }
+               }
                if (ret == -1) {
                        return pvfs_map_errno(pvfs, errno);
                }
@@ -448,53 +473,103 @@ static bool pvfs_read_only(struct pvfs_state *pvfs, uint32_t access_mask)
        return false;
 }
 
+/*
+  see if we are a member of the appropriate unix group
+ */
+static bool pvfs_group_member(struct pvfs_state *pvfs, gid_t gid)
+{
+       int i, ngroups;
+       gid_t *groups;
+       if (getegid() == gid) {
+               return true;
+       }
+       ngroups = getgroups(0, NULL);
+       if (ngroups == 0) {
+               return false;
+       }
+       groups = talloc_array(pvfs, gid_t, ngroups);
+       if (groups == NULL) {
+               return false;
+       }
+       if (getgroups(ngroups, groups) != ngroups) {
+               talloc_free(groups);
+               return false;
+       }
+       for (i=0; i<ngroups; i++) {
+               if (groups[i] == gid) break;
+       }
+       talloc_free(groups);
+       return i < ngroups;
+}
+
 /*
   default access check function based on unix permissions
   doing this saves on building a full security descriptor
   for the common case of access check on files with no 
   specific NT ACL
+
+  If name is NULL then treat as a new file creation
 */
-NTSTATUS pvfs_access_check_unix(struct pvfs_state *pvfs, 
-                               struct ntvfs_request *req,
-                               struct pvfs_filename *name,
-                               uint32_t *access_mask)
+static NTSTATUS pvfs_access_check_unix(struct pvfs_state *pvfs,
+                                      struct ntvfs_request *req,
+                                      struct pvfs_filename *name,
+                                      uint32_t *access_mask)
 {
        uid_t uid = geteuid();
        uint32_t max_bits = SEC_RIGHTS_FILE_READ | SEC_FILE_ALL;
+       struct security_token *token = req->session_info->security_token;
 
        if (pvfs_read_only(pvfs, *access_mask)) {
                return NT_STATUS_ACCESS_DENIED;
        }
 
-       /* owner and root get extra permissions */
-       if (uid == 0) {
-               max_bits |= SEC_STD_ALL | SEC_FLAG_SYSTEM_SECURITY;
-       } else if (uid == name->st.st_uid) {
+       if (name == NULL || uid == name->st.st_uid) {
                max_bits |= SEC_STD_ALL;
+       } else if (security_token_has_privilege(token, SEC_PRIV_RESTORE)) {
+               max_bits |= SEC_STD_DELETE;
        }
 
-#ifdef UID_WRAPPER_REPLACE
-       /* when running with the uid wrapper, files will be created
-          owned by the ruid, but we may have a different simulated 
-          euid. We need to force the permission bits as though the 
-          files owner matches the euid */
-       max_bits |= SEC_STD_ALL;
-#endif
+       if (name == NULL ||
+           (name->st.st_mode & S_IWOTH) ||
+           ((name->st.st_mode & S_IWGRP) && 
+            pvfs_group_member(pvfs, name->st.st_gid))) {
+               max_bits |= SEC_STD_ALL;
+       }
 
-       if (*access_mask == SEC_FLAG_MAXIMUM_ALLOWED) {
-               *access_mask = max_bits;
-               return NT_STATUS_OK;
+       if (uwrap_enabled()) {
+               /* when running with the uid wrapper, files will be created
+                  owned by the ruid, but we may have a different simulated 
+                  euid. We need to force the permission bits as though the 
+                  files owner matches the euid */
+               max_bits |= SEC_STD_ALL;
        }
 
-       if (uid != 0 && (*access_mask & SEC_FLAG_SYSTEM_SECURITY)) {
-               return NT_STATUS_ACCESS_DENIED;
+       if (*access_mask & SEC_FLAG_MAXIMUM_ALLOWED) {
+               *access_mask |= max_bits;
+               *access_mask &= ~SEC_FLAG_MAXIMUM_ALLOWED;
+       }
+
+       if ((*access_mask & SEC_FLAG_SYSTEM_SECURITY) &&
+           security_token_has_privilege(token, SEC_PRIV_SECURITY)) {
+               max_bits |= SEC_FLAG_SYSTEM_SECURITY;
+       }
+       
+       if (((*access_mask & ~max_bits) & SEC_RIGHTS_PRIV_RESTORE) &&
+           security_token_has_privilege(token, SEC_PRIV_RESTORE)) {
+               max_bits |= ~(SEC_RIGHTS_PRIV_RESTORE);
+       }
+       if (((*access_mask & ~max_bits) & SEC_RIGHTS_PRIV_BACKUP) &&
+           security_token_has_privilege(token, SEC_PRIV_BACKUP)) {
+               max_bits |= ~(SEC_RIGHTS_PRIV_BACKUP);
        }
 
        if (*access_mask & ~max_bits) {
+               DEBUG(0,(__location__ " denied access to '%s' - wanted 0x%08x but got 0x%08x (missing 0x%08x)\n",
+                        name?name->full_name:"(new file)", *access_mask, max_bits, *access_mask & ~max_bits));
                return NT_STATUS_ACCESS_DENIED;
        }
 
-       if (pvfs->ntvfs->ctx->protocol != PROTOCOL_SMB2) {
+       if (pvfs->ntvfs->ctx->protocol < PROTOCOL_SMB2_02) {
                /* on SMB, this bit is always granted, even if not
                   asked for */
                *access_mask |= SEC_FILE_READ_ATTRIBUTE;
@@ -518,9 +593,10 @@ NTSTATUS pvfs_access_check(struct pvfs_state *pvfs,
        struct xattr_NTACL *acl;
        NTSTATUS status;
        struct security_descriptor *sd;
+       bool allow_delete = false;
 
        /* on SMB2 a blank access mask is always denied */
-       if (pvfs->ntvfs->ctx->protocol == PROTOCOL_SMB2 &&
+       if (pvfs->ntvfs->ctx->protocol >= PROTOCOL_SMB2_02 &&
            *access_mask == 0) {
                return NT_STATUS_ACCESS_DENIED;
        }
@@ -529,6 +605,16 @@ NTSTATUS pvfs_access_check(struct pvfs_state *pvfs,
                return NT_STATUS_ACCESS_DENIED;
        }
 
+       if (*access_mask & SEC_FLAG_MAXIMUM_ALLOWED ||
+           *access_mask & SEC_STD_DELETE) {
+               status = pvfs_access_check_parent(pvfs, req,
+                                                 name, SEC_DIR_DELETE_CHILD);
+               if (NT_STATUS_IS_OK(status)) {
+                       allow_delete = true;
+                       *access_mask &= ~SEC_STD_DELETE;
+               }
+       }
+
        acl = talloc(req, struct xattr_NTACL);
        if (acl == NULL) {
                return NT_STATUS_NO_MEMORY;
@@ -536,14 +622,15 @@ NTSTATUS pvfs_access_check(struct pvfs_state *pvfs,
 
        /* expand the generic access bits to file specific bits */
        *access_mask = pvfs_translate_mask(*access_mask);
-       if (pvfs->ntvfs->ctx->protocol != PROTOCOL_SMB2) {
+       if (pvfs->ntvfs->ctx->protocol < PROTOCOL_SMB2_02) {
                *access_mask &= ~SEC_FILE_READ_ATTRIBUTE;
        }
 
        status = pvfs_acl_load(pvfs, name, -1, acl);
        if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
                talloc_free(acl);
-               return pvfs_access_check_unix(pvfs, req, name, access_mask);
+               status = pvfs_access_check_unix(pvfs, req, name, access_mask);
+               goto done;
        }
        if (!NT_STATUS_IS_OK(status)) {
                return status;
@@ -558,16 +645,19 @@ NTSTATUS pvfs_access_check(struct pvfs_state *pvfs,
        }
 
        /* check the acl against the required access mask */
-       status = sec_access_check(sd, token, *access_mask, access_mask);
-
-       if (pvfs->ntvfs->ctx->protocol != PROTOCOL_SMB2) {
+       status = se_access_check(sd, token, *access_mask, access_mask);
+       talloc_free(acl);
+done:
+       if (pvfs->ntvfs->ctx->protocol < PROTOCOL_SMB2_02) {
                /* on SMB, this bit is always granted, even if not
                   asked for */
                *access_mask |= SEC_FILE_READ_ATTRIBUTE;
        }
 
-       talloc_free(acl);
-       
+       if (allow_delete) {
+               *access_mask |= SEC_STD_DELETE;
+       }
+
        return status;
 }
 
@@ -593,26 +683,80 @@ NTSTATUS pvfs_access_check_simple(struct pvfs_state *pvfs,
 NTSTATUS pvfs_access_check_create(struct pvfs_state *pvfs, 
                                  struct ntvfs_request *req,
                                  struct pvfs_filename *name,
-                                 uint32_t *access_mask)
+                                 uint32_t *access_mask,
+                                 bool container,
+                                 struct security_descriptor **sd)
 {
        struct pvfs_filename *parent;
        NTSTATUS status;
+       uint32_t parent_mask;
+       bool allow_delete = false;
+
+       if (pvfs_read_only(pvfs, *access_mask)) {
+               return NT_STATUS_ACCESS_DENIED;
+       }
 
        status = pvfs_resolve_parent(pvfs, req, name, &parent);
-       if (!NT_STATUS_IS_OK(status)) {
+       NT_STATUS_NOT_OK_RETURN(status);
+
+       if (container) {
+               parent_mask = SEC_DIR_ADD_SUBDIR;
+       } else {
+               parent_mask = SEC_DIR_ADD_FILE;
+       }
+       if (*access_mask & SEC_FLAG_MAXIMUM_ALLOWED ||
+           *access_mask & SEC_STD_DELETE) {
+               parent_mask |= SEC_DIR_DELETE_CHILD;
+       }
+
+       status = pvfs_access_check(pvfs, req, parent, &parent_mask);
+       if (NT_STATUS_IS_OK(status)) {
+               if (parent_mask & SEC_DIR_DELETE_CHILD) {
+                       allow_delete = true;
+               }
+       } else if (NT_STATUS_EQUAL(status, NT_STATUS_ACCESS_DENIED)) {
+               /*
+                * on ACCESS_DENIED we get the rejected bits
+                * remove the non critical SEC_DIR_DELETE_CHILD
+                * and check if something else was rejected.
+                */
+               parent_mask &= ~SEC_DIR_DELETE_CHILD;
+               if (parent_mask != 0) {
+                       return NT_STATUS_ACCESS_DENIED;
+               }
+               status = NT_STATUS_OK;
+       } else {
                return status;
        }
 
-       status = pvfs_access_check(pvfs, req, parent, access_mask);
+       if (*sd == NULL) {
+               status = pvfs_acl_inherited_sd(pvfs, req, req, parent, container, sd);
+       }
+
+       talloc_free(parent);
        if (!NT_STATUS_IS_OK(status)) {
                return status;
        }
 
-       if (! ((*access_mask) & SEC_DIR_ADD_FILE)) {
-               return pvfs_access_check_simple(pvfs, req, parent, SEC_DIR_ADD_FILE);
+       /* expand the generic access bits to file specific bits */
+       *access_mask = pvfs_translate_mask(*access_mask);
+
+       if (*access_mask & SEC_FLAG_MAXIMUM_ALLOWED) {
+               *access_mask |= SEC_RIGHTS_FILE_ALL;
+               *access_mask &= ~SEC_FLAG_MAXIMUM_ALLOWED;
        }
 
-       return status;
+       if (pvfs->ntvfs->ctx->protocol < PROTOCOL_SMB2_02) {
+               /* on SMB, this bit is always granted, even if not
+                  asked for */
+               *access_mask |= SEC_FILE_READ_ATTRIBUTE;
+       }
+
+       if (allow_delete) {
+               *access_mask |= SEC_STD_DELETE;
+       }
+
+       return NT_STATUS_OK;
 }
 
 /*
@@ -740,47 +884,42 @@ static NTSTATUS pvfs_acl_inherit_aces(struct pvfs_state *pvfs,
 
 
 /*
-  setup an ACL on a new file/directory based on the inherited ACL from
-  the parent. If there is no inherited ACL then we don't set anything,
-  as the default ACL applies anyway
+  calculate the ACL on a new file/directory based on the inherited ACL
+  from the parent. If there is no inherited ACL then return a NULL
+  ACL, which means the default ACL should be used
 */
-NTSTATUS pvfs_acl_inherit(struct pvfs_state *pvfs, 
-                         struct ntvfs_request *req,
-                         struct pvfs_filename *name,
-                         int fd)
+NTSTATUS pvfs_acl_inherited_sd(struct pvfs_state *pvfs, 
+                              TALLOC_CTX *mem_ctx,
+                              struct ntvfs_request *req,
+                              struct pvfs_filename *parent,
+                              bool container,
+                              struct security_descriptor **ret_sd)
 {
        struct xattr_NTACL *acl;
        NTSTATUS status;
-       struct pvfs_filename *parent;
        struct security_descriptor *parent_sd, *sd;
-       bool container;
-       struct id_mapping *ids;
+       struct id_map *ids;
        struct composite_context *ctx;
+       TALLOC_CTX *tmp_ctx = talloc_new(mem_ctx);
 
-       /* form the parents path */
-       status = pvfs_resolve_parent(pvfs, req, name, &parent);
-       if (!NT_STATUS_IS_OK(status)) {
-               return status;
-       }
+       *ret_sd = NULL;
 
        acl = talloc(req, struct xattr_NTACL);
-       if (acl == NULL) {
-               return NT_STATUS_NO_MEMORY;
-       }
+       NT_STATUS_HAVE_NO_MEMORY_AND_FREE(acl, tmp_ctx);
 
        status = pvfs_acl_load(pvfs, parent, -1, acl);
        if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
+               talloc_free(tmp_ctx);
                return NT_STATUS_OK;
        }
-       if (!NT_STATUS_IS_OK(status)) {
-               return status;
-       }
+       NT_STATUS_NOT_OK_RETURN_AND_FREE(status, tmp_ctx);
 
        switch (acl->version) {
        case 1:
                parent_sd = acl->info.sd;
                break;
        default:
+               talloc_free(tmp_ctx);
                return NT_STATUS_INVALID_ACL;
        }
 
@@ -788,61 +927,95 @@ NTSTATUS pvfs_acl_inherit(struct pvfs_state *pvfs,
            parent_sd->dacl == NULL ||
            parent_sd->dacl->num_aces == 0) {
                /* go with the default ACL */
+               talloc_free(tmp_ctx);
                return NT_STATUS_OK;
        }
 
        /* create the new sd */
        sd = security_descriptor_initialise(req);
-       if (sd == NULL) {
-               return NT_STATUS_NO_MEMORY;
-       }
+       NT_STATUS_HAVE_NO_MEMORY_AND_FREE(sd, tmp_ctx);
 
-       ids = talloc_array(sd, struct id_mapping, 2);
-       NT_STATUS_HAVE_NO_MEMORY(ids);
+       ids = talloc_array(sd, struct id_map, 2);
+       NT_STATUS_HAVE_NO_MEMORY_AND_FREE(ids, tmp_ctx);
 
-       ids[0].unixid = talloc(ids, struct unixid);
-       NT_STATUS_HAVE_NO_MEMORY(ids[0].unixid);
-       ids[0].unixid->id = name->st.st_uid;
-       ids[0].unixid->type = ID_TYPE_UID;
+       ids[0].xid.id = geteuid();
+       ids[0].xid.type = ID_TYPE_UID;
        ids[0].sid = NULL;
-       ids[0].status = NT_STATUS_NONE_MAPPED;
+       ids[0].status = ID_UNKNOWN;
 
-       ids[1].unixid = talloc(ids, struct unixid);
-       NT_STATUS_HAVE_NO_MEMORY(ids[1].unixid);
-       ids[1].unixid->id = name->st.st_gid;
-       ids[1].unixid->type = ID_TYPE_GID;
+       ids[1].xid.id = getegid();
+       ids[1].xid.type = ID_TYPE_GID;
        ids[1].sid = NULL;
-       ids[1].status = NT_STATUS_NONE_MAPPED;
+       ids[1].status = ID_UNKNOWN;
 
        ctx = wbc_xids_to_sids_send(pvfs->wbc_ctx, ids, 2, ids);
-       NT_STATUS_HAVE_NO_MEMORY(ctx);
+       NT_STATUS_HAVE_NO_MEMORY_AND_FREE(ctx, tmp_ctx);
 
        status = wbc_xids_to_sids_recv(ctx, &ids);
-       NT_STATUS_NOT_OK_RETURN(status);
+       NT_STATUS_NOT_OK_RETURN_AND_FREE(status, tmp_ctx);
 
        sd->owner_sid = talloc_steal(sd, ids[0].sid);
        sd->group_sid = talloc_steal(sd, ids[1].sid);
 
        sd->type |= SEC_DESC_DACL_PRESENT;
 
-       container = (name->dos.attrib & FILE_ATTRIBUTE_DIRECTORY) ? true:false;
-
        /* fill in the aces from the parent */
        status = pvfs_acl_inherit_aces(pvfs, parent_sd, sd, container);
-       if (!NT_STATUS_IS_OK(status)) {
-               return status;
-       }
+       NT_STATUS_NOT_OK_RETURN_AND_FREE(status, tmp_ctx);
 
        /* if there is nothing to inherit then we fallback to the
           default acl */
        if (sd->dacl == NULL || sd->dacl->num_aces == 0) {
+               talloc_free(tmp_ctx);
                return NT_STATUS_OK;
        }
 
-       acl->info.sd = sd;
+       *ret_sd = talloc_steal(mem_ctx, sd);
+
+       talloc_free(tmp_ctx);
+       return NT_STATUS_OK;
+}
+
+
+/*
+  setup an ACL on a new file/directory based on the inherited ACL from
+  the parent. If there is no inherited ACL then we don't set anything,
+  as the default ACL applies anyway
+*/
+NTSTATUS pvfs_acl_inherit(struct pvfs_state *pvfs, 
+                         struct ntvfs_request *req,
+                         struct pvfs_filename *name,
+                         int fd)
+{
+       struct xattr_NTACL acl;
+       NTSTATUS status;
+       struct security_descriptor *sd;
+       struct pvfs_filename *parent;
+       bool container;
+
+       /* form the parents path */
+       status = pvfs_resolve_parent(pvfs, req, name, &parent);
+       NT_STATUS_NOT_OK_RETURN(status);
+
+       container = (name->dos.attrib & FILE_ATTRIBUTE_DIRECTORY) ? true:false;
+
+       status = pvfs_acl_inherited_sd(pvfs, req, req, parent, container, &sd);
+       if (!NT_STATUS_IS_OK(status)) {
+               talloc_free(parent);
+               return status;
+       }
+
+       if (sd == NULL) {
+               return NT_STATUS_OK;
+       }
+
+       acl.version = 1;
+       acl.info.sd = sd;
+
+       status = pvfs_acl_save(pvfs, name, fd, &acl);
+       talloc_free(sd);
+       talloc_free(parent);
 
-       status = pvfs_acl_save(pvfs, name, fd, acl);
-       
        return status;
 }