summaryrefslogtreecommitdiffstats
path: root/source4/ntvfs/posix/pvfs_acl.c
diff options
context:
space:
mode:
Diffstat (limited to 'source4/ntvfs/posix/pvfs_acl.c')
-rw-r--r--source4/ntvfs/posix/pvfs_acl.c1083
1 files changed, 1083 insertions, 0 deletions
diff --git a/source4/ntvfs/posix/pvfs_acl.c b/source4/ntvfs/posix/pvfs_acl.c
new file mode 100644
index 0000000..e643293
--- /dev/null
+++ b/source4/ntvfs/posix/pvfs_acl.c
@@ -0,0 +1,1083 @@
+/*
+ Unix SMB/CIFS implementation.
+
+ POSIX NTVFS backend - ACL support
+
+ Copyright (C) Andrew Tridgell 2004
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "includes.h"
+#include "system/passwd.h"
+#include "auth/auth.h"
+#include "vfs_posix.h"
+#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"
+
+/* the list of currently registered ACL backends */
+static struct pvfs_acl_backend {
+ const struct pvfs_acl_ops *ops;
+} *backends = NULL;
+static int num_backends;
+
+/*
+ register a pvfs acl backend.
+
+ The 'name' can be later used by other backends to find the operations
+ structure for this backend.
+*/
+NTSTATUS pvfs_acl_register(TALLOC_CTX *ctx, const struct pvfs_acl_ops *ops)
+{
+ struct pvfs_acl_ops *new_ops;
+
+ if (pvfs_acl_backend_byname(ops->name) != NULL) {
+ DEBUG(0,("pvfs acl backend '%s' already registered\n", ops->name));
+ return NT_STATUS_OBJECT_NAME_COLLISION;
+ }
+
+ backends = talloc_realloc(ctx, backends,
+ struct pvfs_acl_backend, num_backends+1);
+ NT_STATUS_HAVE_NO_MEMORY(backends);
+
+ new_ops = (struct pvfs_acl_ops *)talloc_memdup(backends, ops, sizeof(*ops));
+ new_ops->name = talloc_strdup(new_ops, ops->name);
+
+ backends[num_backends].ops = new_ops;
+
+ num_backends++;
+
+ DEBUG(3,("NTVFS backend '%s' registered\n", ops->name));
+
+ return NT_STATUS_OK;
+}
+
+
+/*
+ return the operations structure for a named backend
+*/
+const struct pvfs_acl_ops *pvfs_acl_backend_byname(const char *name)
+{
+ int i;
+
+ for (i=0;i<num_backends;i++) {
+ if (strcmp(backends[i].ops->name, name) == 0) {
+ return backends[i].ops;
+ }
+ }
+
+ return NULL;
+}
+
+NTSTATUS pvfs_acl_init(void)
+{
+ static bool initialized = false;
+#define _MODULE_PROTO(init) extern NTSTATUS init(TALLOC_CTX *);
+ 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, "pvfs_acl");
+
+ run_init_functions(NULL, static_init);
+ run_init_functions(NULL, shared_init);
+
+ talloc_free(shared_init);
+
+ return NT_STATUS_OK;
+}
+
+
+/*
+ map a single access_mask from generic to specific bits for files/dirs
+*/
+static uint32_t pvfs_translate_mask(uint32_t access_mask)
+{
+ if (access_mask & SEC_MASK_GENERIC) {
+ if (access_mask & SEC_GENERIC_READ) access_mask |= SEC_RIGHTS_FILE_READ;
+ if (access_mask & SEC_GENERIC_WRITE) access_mask |= SEC_RIGHTS_FILE_WRITE;
+ if (access_mask & SEC_GENERIC_EXECUTE) access_mask |= SEC_RIGHTS_FILE_EXECUTE;
+ if (access_mask & SEC_GENERIC_ALL) access_mask |= SEC_RIGHTS_FILE_ALL;
+ access_mask &= ~SEC_MASK_GENERIC;
+ }
+ return access_mask;
+}
+
+
+/*
+ map any generic access bits in the given acl
+ this relies on the fact that the mappings for files and directories
+ are the same
+*/
+static void pvfs_translate_generic_bits(struct security_acl *acl)
+{
+ unsigned i;
+
+ if (!acl) return;
+
+ for (i=0;i<acl->num_aces;i++) {
+ struct security_ace *ace = &acl->aces[i];
+ ace->access_mask = pvfs_translate_mask(ace->access_mask);
+ }
+}
+
+
+/*
+ setup a default ACL for a file
+*/
+static NTSTATUS pvfs_default_acl(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name, int fd,
+ struct security_descriptor **psd)
+{
+ struct security_descriptor *sd;
+ NTSTATUS status;
+ struct security_ace ace = {};
+ mode_t mode;
+ struct id_map *ids;
+
+ *psd = security_descriptor_initialise(req);
+ if (*psd == NULL) {
+ return NT_STATUS_NO_MEMORY;
+ }
+ sd = *psd;
+
+ ids = talloc_zero_array(sd, struct id_map, 2);
+ NT_STATUS_HAVE_NO_MEMORY(ids);
+
+ ids[0].xid.id = name->st.st_uid;
+ ids[0].xid.type = ID_TYPE_UID;
+ ids[0].sid = NULL;
+
+ ids[1].xid.id = name->st.st_gid;
+ ids[1].xid.type = ID_TYPE_GID;
+ ids[1].sid = NULL;
+
+ status = wbc_xids_to_sids(ids, 2);
+ NT_STATUS_NOT_OK_RETURN(status);
+
+ sd->owner_sid = talloc_steal(sd, ids[0].sid);
+ sd->group_sid = talloc_steal(sd, ids[1].sid);
+
+ talloc_free(ids);
+ sd->type |= SEC_DESC_DACL_PRESENT;
+
+ mode = name->st.st_mode;
+
+ /*
+ we provide up to 4 ACEs
+ - Owner
+ - Group
+ - Everyone
+ - Administrator
+ */
+
+
+ /* setup owner ACE */
+ ace.type = SEC_ACE_TYPE_ACCESS_ALLOWED;
+ ace.flags = 0;
+ ace.trustee = *sd->owner_sid;
+ ace.access_mask = 0;
+
+ if (mode & S_IRUSR) {
+ if (mode & S_IWUSR) {
+ ace.access_mask |= SEC_RIGHTS_FILE_ALL;
+ } else {
+ ace.access_mask |= SEC_RIGHTS_FILE_READ | SEC_FILE_EXECUTE;
+ }
+ }
+ if (mode & S_IWUSR) {
+ ace.access_mask |= SEC_RIGHTS_FILE_WRITE | SEC_STD_DELETE;
+ }
+ if (ace.access_mask) {
+ security_descriptor_dacl_add(sd, &ace);
+ }
+
+
+ /* setup group ACE */
+ ace.trustee = *sd->group_sid;
+ ace.access_mask = 0;
+ if (mode & S_IRGRP) {
+ ace.access_mask |= SEC_RIGHTS_FILE_READ | SEC_FILE_EXECUTE;
+ }
+ if (mode & S_IWGRP) {
+ /* note that delete is not granted - this matches posix behaviour */
+ ace.access_mask |= SEC_RIGHTS_FILE_WRITE;
+ }
+ if (ace.access_mask) {
+ security_descriptor_dacl_add(sd, &ace);
+ }
+
+ /* setup other ACE */
+ ace.trustee = global_sid_World;
+ ace.access_mask = 0;
+ if (mode & S_IROTH) {
+ ace.access_mask |= SEC_RIGHTS_FILE_READ | SEC_FILE_EXECUTE;
+ }
+ if (mode & S_IWOTH) {
+ ace.access_mask |= SEC_RIGHTS_FILE_WRITE;
+ }
+ if (ace.access_mask) {
+ security_descriptor_dacl_add(sd, &ace);
+ }
+
+ /* setup system ACE */
+ ace.trustee = global_sid_System;
+ ace.access_mask = SEC_RIGHTS_FILE_ALL;
+ security_descriptor_dacl_add(sd, &ace);
+
+ return NT_STATUS_OK;
+}
+
+
+/*
+ omit any security_descriptor elements not specified in the given
+ secinfo flags
+*/
+static void normalise_sd_flags(struct security_descriptor *sd, uint32_t secinfo_flags)
+{
+ if (!(secinfo_flags & SECINFO_OWNER)) {
+ sd->owner_sid = NULL;
+ }
+ if (!(secinfo_flags & SECINFO_GROUP)) {
+ sd->group_sid = NULL;
+ }
+ if (!(secinfo_flags & SECINFO_DACL)) {
+ sd->dacl = NULL;
+ }
+ if (!(secinfo_flags & SECINFO_SACL)) {
+ sd->sacl = NULL;
+ }
+}
+
+static bool pvfs_privileged_access(uid_t uid)
+{
+ uid_t euid;
+
+ if (uid_wrapper_enabled()) {
+ setenv("UID_WRAPPER_MYUID", "1", 1);
+ }
+
+ euid = geteuid();
+
+ if (uid_wrapper_enabled()) {
+ unsetenv("UID_WRAPPER_MYUID");
+ }
+
+ return (uid == euid);
+}
+
+/*
+ answer a setfileinfo for an ACL
+*/
+NTSTATUS pvfs_acl_set(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name, int fd,
+ uint32_t access_mask,
+ union smb_setfileinfo *info)
+{
+ uint32_t secinfo_flags = info->set_secdesc.in.secinfo_flags;
+ struct security_descriptor *new_sd, *sd, orig_sd;
+ NTSTATUS status = NT_STATUS_NOT_FOUND;
+ uid_t old_uid = -1;
+ gid_t old_gid = -1;
+ uid_t new_uid = -1;
+ gid_t new_gid = -1;
+ struct id_map *ids;
+
+ if (pvfs->acl_ops != NULL) {
+ status = pvfs->acl_ops->acl_load(pvfs, name, fd, req, &sd);
+ }
+ if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
+ status = pvfs_default_acl(pvfs, req, name, fd, &sd);
+ }
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ ids = talloc(req, struct id_map);
+ NT_STATUS_HAVE_NO_MEMORY(ids);
+ ZERO_STRUCT(ids->xid);
+ ids->sid = NULL;
+ ids->status = ID_UNKNOWN;
+
+ new_sd = info->set_secdesc.in.sd;
+ orig_sd = *sd;
+
+ old_uid = name->st.st_uid;
+ old_gid = name->st.st_gid;
+
+ /* only set the elements that have been specified */
+ if (secinfo_flags & SECINFO_OWNER) {
+ if (!(access_mask & SEC_STD_WRITE_OWNER)) {
+ return NT_STATUS_ACCESS_DENIED;
+ }
+ if (!dom_sid_equal(sd->owner_sid, new_sd->owner_sid)) {
+ ids->sid = new_sd->owner_sid;
+ status = wbc_sids_to_xids(ids, 1);
+ NT_STATUS_NOT_OK_RETURN(status);
+
+ 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;
+ }
+
+ if (secinfo_flags & SECINFO_GROUP) {
+ if (!(access_mask & SEC_STD_WRITE_OWNER)) {
+ return NT_STATUS_ACCESS_DENIED;
+ }
+ if (!dom_sid_equal(sd->group_sid, new_sd->group_sid)) {
+ ids->sid = new_sd->group_sid;
+ status = wbc_sids_to_xids(ids, 1);
+ NT_STATUS_NOT_OK_RETURN(status);
+
+ if (ids->xid.type == ID_TYPE_BOTH ||
+ ids->xid.type == ID_TYPE_GID) {
+ new_gid = ids->xid.id;
+ }
+
+ }
+ sd->group_sid = new_sd->group_sid;
+ }
+
+ if (secinfo_flags & SECINFO_DACL) {
+ if (!(access_mask & SEC_STD_WRITE_DAC)) {
+ return NT_STATUS_ACCESS_DENIED;
+ }
+ sd->dacl = new_sd->dacl;
+ pvfs_translate_generic_bits(sd->dacl);
+ sd->type |= SEC_DESC_DACL_PRESENT;
+ }
+
+ if (secinfo_flags & SECINFO_SACL) {
+ if (!(access_mask & SEC_FLAG_SYSTEM_SECURITY)) {
+ return NT_STATUS_ACCESS_DENIED;
+ }
+ sd->sacl = new_sd->sacl;
+ pvfs_translate_generic_bits(sd->sacl);
+ sd->type |= SEC_DESC_SACL_PRESENT;
+ }
+
+ if (secinfo_flags & SECINFO_PROTECTED_DACL) {
+ if (new_sd->type & SEC_DESC_DACL_PROTECTED) {
+ sd->type |= SEC_DESC_DACL_PROTECTED;
+ } else {
+ sd->type &= ~SEC_DESC_DACL_PROTECTED;
+ }
+ }
+
+ if (secinfo_flags & SECINFO_PROTECTED_SACL) {
+ if (new_sd->type & SEC_DESC_SACL_PROTECTED) {
+ sd->type |= SEC_DESC_SACL_PROTECTED;
+ } else {
+ sd->type &= ~SEC_DESC_SACL_PROTECTED;
+ }
+ }
+
+ if (new_uid == old_uid) {
+ new_uid = -1;
+ }
+
+ if (new_gid == old_gid) {
+ new_gid = -1;
+ }
+
+ /* if there's something to change try it */
+ if (new_uid != -1 || new_gid != -1) {
+ int ret;
+ if (fd == -1) {
+ ret = chown(name->full_name, new_uid, new_gid);
+ } else {
+ ret = fchown(fd, new_uid, new_gid);
+ }
+ if (errno == EPERM) {
+ if (pvfs_privileged_access(name->st.st_uid)) {
+ 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);
+ }
+ }
+
+ /* we avoid saving if the sd is the same. This means when clients
+ copy files and end up copying the default sd that we don't
+ needlessly use xattrs */
+ if (!security_descriptor_equal(sd, &orig_sd) && pvfs->acl_ops) {
+ status = pvfs->acl_ops->acl_save(pvfs, name, fd, sd);
+ }
+
+ return status;
+}
+
+
+/*
+ answer a fileinfo query for the ACL
+*/
+NTSTATUS pvfs_acl_query(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name, int fd,
+ union smb_fileinfo *info)
+{
+ NTSTATUS status = NT_STATUS_NOT_FOUND;
+ struct security_descriptor *sd;
+
+ if (pvfs->acl_ops) {
+ status = pvfs->acl_ops->acl_load(pvfs, name, fd, req, &sd);
+ }
+ if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
+ status = pvfs_default_acl(pvfs, req, name, fd, &sd);
+ }
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ normalise_sd_flags(sd, info->query_secdesc.in.secinfo_flags);
+
+ info->query_secdesc.out.sd = sd;
+
+ return NT_STATUS_OK;
+}
+
+
+/*
+ check the read only bit against any of the write access bits
+*/
+static bool pvfs_read_only(struct pvfs_state *pvfs, uint32_t access_mask)
+{
+ if ((pvfs->flags & PVFS_FLAG_READONLY) &&
+ (access_mask & (SEC_FILE_WRITE_DATA |
+ SEC_FILE_APPEND_DATA |
+ SEC_FILE_WRITE_EA |
+ SEC_FILE_WRITE_ATTRIBUTE |
+ SEC_STD_DELETE |
+ SEC_STD_WRITE_DAC |
+ SEC_STD_WRITE_OWNER |
+ SEC_DIR_DELETE_CHILD))) {
+ return true;
+ }
+ 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
+*/
+static NTSTATUS pvfs_access_check_unix(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name,
+ uint32_t *access_mask)
+{
+ uint32_t max_bits = 0;
+ struct security_token *token = req->session_info->security_token;
+
+ if (pvfs_read_only(pvfs, *access_mask)) {
+ return NT_STATUS_ACCESS_DENIED;
+ }
+
+ if (name == NULL) {
+ max_bits |= SEC_RIGHTS_FILE_ALL | SEC_STD_ALL;
+ } else if (pvfs_privileged_access(name->st.st_uid)) {
+ /* use the IxUSR bits */
+ if ((name->st.st_mode & S_IWUSR)) {
+ max_bits |= SEC_RIGHTS_FILE_ALL | SEC_STD_ALL;
+ } else if ((name->st.st_mode & (S_IRUSR | S_IXUSR))) {
+ max_bits |= SEC_RIGHTS_FILE_READ | SEC_RIGHTS_FILE_EXECUTE | SEC_STD_ALL;
+ }
+ } else if (pvfs_group_member(pvfs, name->st.st_gid)) {
+ /* use the IxGRP bits */
+ if ((name->st.st_mode & S_IWGRP)) {
+ max_bits |= SEC_RIGHTS_FILE_ALL | SEC_STD_ALL;
+ } else if ((name->st.st_mode & (S_IRGRP | S_IXGRP))) {
+ max_bits |= SEC_RIGHTS_FILE_READ | SEC_RIGHTS_FILE_EXECUTE | SEC_STD_ALL;
+ }
+ } else {
+ /* use the IxOTH bits */
+ if ((name->st.st_mode & S_IWOTH)) {
+ max_bits |= SEC_RIGHTS_FILE_ALL | SEC_STD_ALL;
+ } else if ((name->st.st_mode & (S_IROTH | S_IXOTH))) {
+ max_bits |= SEC_RIGHTS_FILE_READ | SEC_RIGHTS_FILE_EXECUTE | SEC_STD_ALL;
+ }
+ }
+
+ 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(5,(__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_02) {
+ /* on SMB, this bit is always granted, even if not
+ asked for */
+ *access_mask |= SEC_FILE_READ_ATTRIBUTE;
+ }
+
+ return NT_STATUS_OK;
+}
+
+
+/*
+ check the security descriptor on a file, if any
+
+ *access_mask is modified with the access actually granted
+*/
+NTSTATUS pvfs_access_check(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name,
+ uint32_t *access_mask)
+{
+ struct security_token *token = req->session_info->security_token;
+ 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_02 &&
+ *access_mask == 0) {
+ return NT_STATUS_ACCESS_DENIED;
+ }
+
+ if (pvfs_read_only(pvfs, *access_mask)) {
+ 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;
+ }
+
+ /* expand the generic access bits to file specific bits */
+ *access_mask = pvfs_translate_mask(*access_mask);
+ 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);
+ status = pvfs_access_check_unix(pvfs, req, name, access_mask);
+ goto done;
+ }
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ switch (acl->version) {
+ case 1:
+ sd = acl->info.sd;
+ break;
+ default:
+ return NT_STATUS_INVALID_ACL;
+ }
+
+ /* check the acl against the required access mask */
+ status = se_file_access_check(sd, token, false, *access_mask, access_mask);
+ talloc_free(acl);
+
+ /* if we used a NT acl, then allow access override if the
+ share allows for posix permission override
+ */
+ if (NT_STATUS_IS_OK(status)) {
+ name->allow_override = (pvfs->flags & PVFS_FLAG_PERM_OVERRIDE) != 0;
+ }
+
+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;
+ }
+
+ if (allow_delete) {
+ *access_mask |= SEC_STD_DELETE;
+ }
+
+ return status;
+}
+
+
+/*
+ a simplified interface to access check, designed for calls that
+ do not take or return an access check mask
+*/
+NTSTATUS pvfs_access_check_simple(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name,
+ uint32_t access_needed)
+{
+ if (access_needed == 0) {
+ return NT_STATUS_OK;
+ }
+ return pvfs_access_check(pvfs, req, name, &access_needed);
+}
+
+/*
+ access check for creating a new file/directory
+*/
+NTSTATUS pvfs_access_check_create(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name,
+ 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);
+ 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;
+ }
+
+ 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;
+ }
+
+ /* 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;
+ }
+
+ 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;
+}
+
+/*
+ access check for creating a new file/directory - no access mask supplied
+*/
+NTSTATUS pvfs_access_check_parent(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name,
+ uint32_t access_mask)
+{
+ struct pvfs_filename *parent;
+ NTSTATUS status;
+
+ status = pvfs_resolve_parent(pvfs, req, name, &parent);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ status = pvfs_access_check_simple(pvfs, req, parent, access_mask);
+ if (NT_STATUS_IS_OK(status) && parent->allow_override) {
+ name->allow_override = true;
+ }
+ return status;
+}
+
+
+/*
+ determine if an ACE is inheritable
+*/
+static bool pvfs_inheritable_ace(struct pvfs_state *pvfs,
+ const struct security_ace *ace,
+ bool container)
+{
+ if (!container) {
+ return (ace->flags & SEC_ACE_FLAG_OBJECT_INHERIT) != 0;
+ }
+
+ if (ace->flags & SEC_ACE_FLAG_CONTAINER_INHERIT) {
+ return true;
+ }
+
+ if ((ace->flags & SEC_ACE_FLAG_OBJECT_INHERIT) &&
+ !(ace->flags & SEC_ACE_FLAG_NO_PROPAGATE_INHERIT)) {
+ return true;
+ }
+
+ return false;
+}
+
+/*
+ this is the core of ACL inheritance. It copies any inheritable
+ aces from the parent SD to the child SD. Note that the algorithm
+ depends on whether the child is a container or not
+*/
+static NTSTATUS pvfs_acl_inherit_aces(struct pvfs_state *pvfs,
+ struct security_descriptor *parent_sd,
+ struct security_descriptor *sd,
+ bool container)
+{
+ int i;
+
+ for (i=0;i<parent_sd->dacl->num_aces;i++) {
+ struct security_ace ace = parent_sd->dacl->aces[i];
+ NTSTATUS status;
+ const struct dom_sid *creator = NULL, *new_id = NULL;
+ uint32_t orig_flags;
+
+ if (!pvfs_inheritable_ace(pvfs, &ace, container)) {
+ continue;
+ }
+
+ orig_flags = ace.flags;
+
+ /* see the RAW-ACLS inheritance test for details on these rules */
+ if (!container) {
+ ace.flags = 0;
+ } else {
+ ace.flags &= ~SEC_ACE_FLAG_INHERIT_ONLY;
+
+ if (!(ace.flags & SEC_ACE_FLAG_CONTAINER_INHERIT)) {
+ ace.flags |= SEC_ACE_FLAG_INHERIT_ONLY;
+ }
+ if (ace.flags & SEC_ACE_FLAG_NO_PROPAGATE_INHERIT) {
+ ace.flags = 0;
+ }
+ }
+
+ /* the CREATOR sids are special when inherited */
+ if (dom_sid_equal(&ace.trustee, pvfs->sid_cache.creator_owner)) {
+ creator = pvfs->sid_cache.creator_owner;
+ new_id = sd->owner_sid;
+ } else if (dom_sid_equal(&ace.trustee, pvfs->sid_cache.creator_group)) {
+ creator = pvfs->sid_cache.creator_group;
+ new_id = sd->group_sid;
+ } else {
+ new_id = &ace.trustee;
+ }
+
+ if (creator && container &&
+ (ace.flags & SEC_ACE_FLAG_CONTAINER_INHERIT)) {
+ uint32_t flags = ace.flags;
+
+ ace.trustee = *new_id;
+ ace.flags = 0;
+ status = security_descriptor_dacl_add(sd, &ace);
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+
+ ace.trustee = *creator;
+ ace.flags = flags | SEC_ACE_FLAG_INHERIT_ONLY;
+ status = security_descriptor_dacl_add(sd, &ace);
+ } else if (container &&
+ !(orig_flags & SEC_ACE_FLAG_NO_PROPAGATE_INHERIT)) {
+ status = security_descriptor_dacl_add(sd, &ace);
+ } else {
+ ace.trustee = *new_id;
+ status = security_descriptor_dacl_add(sd, &ace);
+ }
+
+ if (!NT_STATUS_IS_OK(status)) {
+ return status;
+ }
+ }
+
+ return NT_STATUS_OK;
+}
+
+
+
+/*
+ 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_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 security_descriptor *parent_sd, *sd;
+ struct id_map *ids;
+ TALLOC_CTX *tmp_ctx = talloc_new(mem_ctx);
+
+ *ret_sd = NULL;
+
+ acl = talloc(req, struct xattr_NTACL);
+ if (acl == NULL) {
+ TALLOC_FREE(tmp_ctx);
+ return NT_STATUS_NO_MEMORY;
+ }
+
+ 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)) {
+ TALLOC_FREE(tmp_ctx);
+ return status;
+ }
+
+ switch (acl->version) {
+ case 1:
+ parent_sd = acl->info.sd;
+ break;
+ default:
+ talloc_free(tmp_ctx);
+ return NT_STATUS_INVALID_ACL;
+ }
+
+ if (parent_sd == NULL ||
+ 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) {
+ TALLOC_FREE(tmp_ctx);
+ return NT_STATUS_NO_MEMORY;
+ }
+
+ ids = talloc_array(sd, struct id_map, 2);
+ if (ids == NULL) {
+ TALLOC_FREE(tmp_ctx);
+ return NT_STATUS_NO_MEMORY;
+ }
+
+ ids[0].xid.id = geteuid();
+ ids[0].xid.type = ID_TYPE_UID;
+ ids[0].sid = NULL;
+ ids[0].status = ID_UNKNOWN;
+
+ ids[1].xid.id = getegid();
+ ids[1].xid.type = ID_TYPE_GID;
+ ids[1].sid = NULL;
+ ids[1].status = ID_UNKNOWN;
+
+ status = wbc_xids_to_sids(ids, 2);
+ if (!NT_STATUS_IS_OK(status)) {
+ TALLOC_FREE(tmp_ctx);
+ return status;
+ }
+
+ sd->owner_sid = talloc_steal(sd, ids[0].sid);
+ sd->group_sid = talloc_steal(sd, ids[1].sid);
+
+ sd->type |= SEC_DESC_DACL_PRESENT;
+
+ /* fill in the aces from the parent */
+ status = pvfs_acl_inherit_aces(pvfs, parent_sd, sd, container);
+ if (!NT_STATUS_IS_OK(status)) {
+ TALLOC_FREE(tmp_ctx);
+ return status;
+ }
+
+ /* 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;
+ }
+
+ *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);
+
+ return status;
+}
+
+/*
+ return the maximum allowed access mask
+*/
+NTSTATUS pvfs_access_maximal_allowed(struct pvfs_state *pvfs,
+ struct ntvfs_request *req,
+ struct pvfs_filename *name,
+ uint32_t *maximal_access)
+{
+ *maximal_access = SEC_FLAG_MAXIMUM_ALLOWED;
+ return pvfs_access_check(pvfs, req, name, maximal_access);
+}