selftests/bpf: Reject graph kptr use after RCU unlockgrafted

Add a sleepable verifier test that loads a graph-node local kptr in an
explicit RCU read-side critical section, then passes its node to
bpf_rbtree_remove() after the section ends.

Before the verifier fix, the stale NON_OWN_REF flag makes the node look like
a live borrowed reference and the program is accepted. After the fix, the
pointer is demoted without NON_OWN_REF and the graph kfunc argument is
rejected.

Also exercise a graph kptr loaded while a spin lock provides implicit RCU
protection. The pointer must be invalidated when the lock is released, which
guards the required ordering between non-owning-reference invalidation and
RCU demotion.

Update the existing fault-protected load test state description. The
post-unlock pointer no longer carries NON_OWN_REF, but remains readable
because the load is rewritten to use BPF_PROBE_MEM.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260904084325.52250-7-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

1 files changed, 747 insertions, 0 deletions

diff --git a/security/apparmor/file.c b/security/apparmor/file.c
new file mode 100644
index 000000000..c9d55fe10
--- /dev/null
+++ b/security/apparmor/file.c
@@ -0,0 +1,747 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * AppArmor security module
+ *
+ * This file contains AppArmor mediation of files
+ *
+ * Copyright (C) 1998-2008 Novell/SUSE
+ * Copyright 2009-2010 Canonical Ltd.
+ */
+
+#include <linux/tty.h>
+#include <linux/fdtable.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/mount.h>
+
+#include "include/af_unix.h"
+#include "include/apparmor.h"
+#include "include/audit.h"
+#include "include/cred.h"
+#include "include/file.h"
+#include "include/match.h"
+#include "include/net.h"
+#include "include/path.h"
+#include "include/policy.h"
+#include "include/label.h"
+
+static u32 map_mask_to_chr_mask(u32 mask)
+{
+	u32 m = mask & PERMS_CHRS_MASK;
+
+	if (mask & AA_MAY_GETATTR)
+		m |= MAY_READ;
+	if (mask & (AA_MAY_SETATTR | AA_MAY_CHMOD | AA_MAY_CHOWN))
+		m |= MAY_WRITE;
+
+	return m;
+}
+
+/**
+ * file_audit_cb - call back for file specific audit fields
+ * @ab: audit_buffer  (NOT NULL)
+ * @va: audit struct to audit values of  (NOT NULL)
+ */
+static void file_audit_cb(struct audit_buffer *ab, void *va)
+{
+	struct common_audit_data *sa = va;
+	struct apparmor_audit_data *ad = aad(sa);
+	kuid_t fsuid = ad->subj_cred ? ad->subj_cred->fsuid : current_fsuid();
+	char str[10];
+
+	if (ad->request & AA_AUDIT_FILE_MASK) {
+		aa_perm_mask_to_str(str, sizeof(str), aa_file_perm_chrs,
+				    map_mask_to_chr_mask(ad->request));
+		audit_log_format(ab, " requested_mask=\"%s\"", str);
+	}
+	if (ad->denied & AA_AUDIT_FILE_MASK) {
+		aa_perm_mask_to_str(str, sizeof(str), aa_file_perm_chrs,
+				    map_mask_to_chr_mask(ad->denied));
+		audit_log_format(ab, " denied_mask=\"%s\"", str);
+	}
+	if (ad->request & AA_AUDIT_FILE_MASK) {
+		audit_log_format(ab, " fsuid=%d",
+				 from_kuid(&init_user_ns, fsuid));
+		audit_log_format(ab, " ouid=%d",
+				 from_kuid(&init_user_ns, ad->fs.ouid));
+	}
+
+	if (ad->peer) {
+		audit_log_format(ab, " target=");
+		aa_label_xaudit(ab, labels_ns(ad->subj_label), ad->peer,
+				FLAG_VIEW_SUBNS, GFP_KERNEL);
+	} else if (ad->fs.target) {
+		audit_log_format(ab, " target=");
+		audit_log_untrustedstring(ab, ad->fs.target);
+	}
+}
+
+/**
+ * aa_audit_file - handle the auditing of file operations
+ * @subj_cred: cred of the subject
+ * @profile: the profile being enforced  (NOT NULL)
+ * @perms: the permissions computed for the request (NOT NULL)
+ * @op: operation being mediated
+ * @request: permissions requested
+ * @name: name of object being mediated (MAYBE NULL)
+ * @target: name of target (MAYBE NULL)
+ * @tlabel: target label (MAY BE NULL)
+ * @ouid: object uid
+ * @info: extra information message (MAYBE NULL)
+ * @error: 0 if operation allowed else failure error code
+ *
+ * Returns: %0 or error on failure
+ */
+int aa_audit_file(const struct cred *subj_cred,
+		  struct aa_profile *profile, struct aa_perms *perms,
+		  const char *op, u32 request, const char *name,
+		  const char *target, struct aa_label *tlabel,
+		  kuid_t ouid, const char *info, int error)
+{
+	int type = AUDIT_APPARMOR_AUTO;
+	DEFINE_AUDIT_DATA(ad, LSM_AUDIT_DATA_TASK, AA_CLASS_FILE, op);
+
+	ad.subj_cred = subj_cred;
+	ad.request = request;
+	ad.tags = perms->tag;
+	ad.name = name;
+	ad.fs.target = target;
+	ad.peer = tlabel;
+	ad.fs.ouid = ouid;
+	ad.info = info;
+	ad.error = error;
+	ad.common.u.tsk = NULL;
+
+	if (likely(!ad.error)) {
+		u32 mask = perms->audit;
+
+		if (unlikely(AUDIT_MODE(profile) == AUDIT_ALL))
+			mask = 0xffff;
+
+		/* mask off perms that are not being force audited */
+		ad.request &= mask;
+
+		if (likely(!ad.request))
+			return 0;
+		type = AUDIT_APPARMOR_AUDIT;
+	} else {
+		/* only report permissions that were denied */
+		ad.request = ad.request & ~perms->allow;
+		AA_BUG(!ad.request);
+
+		if (ad.request & perms->kill)
+			type = AUDIT_APPARMOR_KILL;
+
+		/* quiet known rejects, assumes quiet and kill do not overlap */
+		if ((ad.request & perms->quiet) &&
+		    AUDIT_MODE(profile) != AUDIT_NOQUIET &&
+		    AUDIT_MODE(profile) != AUDIT_ALL)
+			ad.request &= ~perms->quiet;
+
+		if (!ad.request)
+			return ad.error;
+	}
+
+	ad.denied = ad.request & ~perms->allow;
+	return aa_audit(type, profile, &ad, file_audit_cb);
+}
+
+static int path_name(const char *op, const struct cred *subj_cred,
+		     struct aa_label *label,
+		     const struct path *path, int flags, char *buffer,
+		     const char **name, struct path_cond *cond, u32 request)
+{
+	struct aa_profile *profile;
+	const char *info = NULL;
+	int error;
+
+	/* don't reaudit files closed during inheritance */
+	if (unlikely(path->dentry == aa_null.dentry))
+		return -EACCES;
+
+	error = aa_path_name(path, flags, buffer, name, &info,
+			     labels_profile(label)->disconnected);
+	if (error) {
+		fn_for_each_confined(label, profile,
+			aa_audit_file(subj_cred,
+				      profile, &nullperms, op, request, *name,
+				      NULL, NULL, cond->uid, info, error));
+		return error;
+	}
+
+	return 0;
+}
+
+struct aa_perms default_perms = {};
+/**
+ * aa_lookup_condperms - convert dfa compressed perms to internal perms
+ * @subj_uid: uid to use for subject owner test
+ * @rules: the aa_policydb to lookup perms for  (NOT NULL)
+ * @state: state in dfa
+ * @cond:  conditions to consider  (NOT NULL)
+ *
+ * TODO: convert from dfa + state to permission entry
+ *
+ * Returns: a pointer to a file permission set
+ */
+struct aa_perms *aa_lookup_condperms(kuid_t subj_uid, struct aa_policydb *rules,
+				     aa_state_t state, struct path_cond *cond)
+{
+	unsigned int index = ACCEPT_TABLE(rules->dfa)[state];
+
+	if (!(rules->perms))
+		return &default_perms;
+
+	if ((ACCEPT_TABLE2(rules->dfa)[state] & ACCEPT_FLAG_OWNER)) {
+		if (uid_eq(subj_uid, cond->uid))
+			return &(rules->perms[index]);
+		return &(rules->perms[index + 1]);
+	}
+
+	return &(rules->perms[index]);
+}
+
+/**
+ * aa_str_perms - find permission that match @name
+ * @file_rules: the aa_policydb to match against  (NOT NULL)
+ * @start: state to start matching in
+ * @name: string to match against dfa  (NOT NULL)
+ * @cond: conditions to consider for permission set computation  (NOT NULL)
+ * @perms: Returns - the permissions found when matching @name
+ *
+ * Returns: the final state in @dfa when beginning @start and walking @name
+ */
+aa_state_t aa_str_perms(struct aa_policydb *file_rules, aa_state_t start,
+			const char *name, struct path_cond *cond,
+			struct aa_perms *perms)
+{
+	aa_state_t state;
+	state = aa_dfa_match(file_rules->dfa, start, name);
+	*perms = *(aa_lookup_condperms(current_fsuid(), file_rules, state,
+				       cond));
+
+	return state;
+}
+
+int __aa_path_perm(const char *op, const struct cred *subj_cred,
+		   struct aa_profile *profile, const char *name,
+		   u32 request, struct path_cond *cond, int flags,
+		   struct aa_perms *perms)
+{
+	struct aa_ruleset *rules = profile->label.rules[0];
+	int e = 0;
+
+	if (profile_unconfined(profile) ||
+	    ((flags & PATH_SOCK_COND) && !RULE_MEDIATES_v9NET(rules)))
+		return 0;
+	aa_str_perms(rules->file, rules->file->start[AA_CLASS_FILE],
+		     name, cond, perms);
+	if (request & ~perms->allow)
+		e = -EACCES;
+	return aa_audit_file(subj_cred,
+			     profile, perms, op, request, name, NULL, NULL,
+			     cond->uid, NULL, e);
+}
+
+
+static int profile_path_perm(const char *op, const struct cred *subj_cred,
+			     struct aa_profile *profile,
+			     const struct path *path, char *buffer, u32 request,
+			     struct path_cond *cond, int flags,
+			     struct aa_perms *perms)
+{
+	const char *name = NULL;
+	int error;
+
+	if (profile_unconfined(profile))
+		return 0;
+
+	error = path_name(op, subj_cred, &profile->label, path,
+			  flags | profile->path_flags, buffer, &name, cond,
+			  request);
+	if (error)
+		return error;
+	return __aa_path_perm(op, subj_cred, profile, name, request, cond,
+			      flags, perms);
+}
+
+/**
+ * aa_path_perm - do permissions check & audit for @path
+ * @op: operation being checked
+ * @subj_cred: subject cred
+ * @label: profile being enforced  (NOT NULL)
+ * @path: path to check permissions of  (NOT NULL)
+ * @flags: any additional path flags beyond what the profile specifies
+ * @request: requested permissions
+ * @cond: conditional info for this request  (NOT NULL)
+ *
+ * Returns: %0 else error if access denied or other error
+ */
+int aa_path_perm(const char *op, const struct cred *subj_cred,
+		 struct aa_label *label,
+		 const struct path *path, int flags, u32 request,
+		 struct path_cond *cond)
+{
+	struct aa_perms perms = {};
+	struct aa_profile *profile;
+	char *buffer = NULL;
+	int error;
+
+	flags |= PATH_DELEGATE_DELETED | (S_ISDIR(cond->mode) ? PATH_IS_DIR :
+								0);
+	buffer = aa_get_buffer(false);
+	if (!buffer)
+		return -ENOMEM;
+	error = fn_for_each_confined(label, profile,
+			profile_path_perm(op, subj_cred, profile, path, buffer,
+					  request, cond, flags, &perms));
+
+	aa_put_buffer(buffer);
+
+	return error;
+}
+
+/**
+ * xindex_is_subset - helper for aa_path_link
+ * @link: link permission set
+ * @target: target permission set
+ *
+ * test target x permissions are equal OR a subset of link x permissions
+ * this is done as part of the subset test, where a hardlink must have
+ * a subset of permissions that the target has.
+ *
+ * Returns: true if subset else false
+ */
+static inline bool xindex_is_subset(u32 link, u32 target)
+{
+	if (((link & ~AA_X_UNSAFE) != (target & ~AA_X_UNSAFE)) ||
+	    ((link & AA_X_UNSAFE) && !(target & AA_X_UNSAFE)))
+		return false;
+
+	return true;
+}
+
+static int profile_path_link(const struct cred *subj_cred,
+			     struct aa_profile *profile,
+			     const struct path *link, char *buffer,
+			     const struct path *target, char *buffer2,
+			     struct path_cond *cond)
+{
+	struct aa_ruleset *rules = profile->label.rules[0];
+	const char *lname = NULL, *tname = NULL;
+	struct aa_perms lperms = {}, perms;
+	const char *info = NULL;
+	u32 request = AA_MAY_LINK;
+	aa_state_t state;
+	int error;
+
+	error = path_name(OP_LINK, subj_cred, &profile->label, link,
+			  profile->path_flags,
+			  buffer, &lname, cond, AA_MAY_LINK);
+	if (error)
+		goto audit;
+
+	/* buffer2 freed below, tname is pointer in buffer2 */
+	error = path_name(OP_LINK, subj_cred, &profile->label, target,
+			  profile->path_flags,
+			  buffer2, &tname, cond, AA_MAY_LINK);
+	if (error)
+		goto audit;
+
+	error = -EACCES;
+	/* aa_str_perms - handles the case of the dfa being NULL */
+	state = aa_str_perms(rules->file,
+			     rules->file->start[AA_CLASS_FILE], lname,
+			     cond, &lperms);
+
+	if (!(lperms.allow & AA_MAY_LINK))
+		goto audit;
+
+	/* test to see if target can be paired with link */
+	state = aa_dfa_null_transition(rules->file->dfa, state);
+	aa_str_perms(rules->file, state, tname, cond, &perms);
+
+	/* force audit/quiet masks for link are stored in the second entry
+	 * in the link pair.
+	 */
+	lperms.audit = perms.audit;
+	lperms.quiet = perms.quiet;
+	lperms.kill = perms.kill;
+
+	if (!(perms.allow & AA_MAY_LINK)) {
+		info = "target restricted";
+		lperms = perms;
+		goto audit;
+	}
+
+	/* done if link subset test is not required */
+	if (!(perms.allow & AA_LINK_SUBSET))
+		goto done_tests;
+
+	/* Do link perm subset test requiring allowed permission on link are
+	 * a subset of the allowed permissions on target.
+	 */
+	aa_str_perms(rules->file, rules->file->start[AA_CLASS_FILE],
+		     tname, cond, &perms);
+
+	/* AA_MAY_LINK is not considered in the subset test */
+	request = lperms.allow & ~AA_MAY_LINK;
+	lperms.allow &= perms.allow | AA_MAY_LINK;
+
+	request |= AA_AUDIT_FILE_MASK & (lperms.allow & ~perms.allow);
+	if (request & ~lperms.allow) {
+		goto audit;
+	} else if ((lperms.allow & MAY_EXEC) &&
+		   !xindex_is_subset(lperms.xindex, perms.xindex)) {
+		lperms.allow &= ~MAY_EXEC;
+		request |= MAY_EXEC;
+		info = "link not subset of target";
+		goto audit;
+	}
+
+done_tests:
+	error = 0;
+
+audit:
+	return aa_audit_file(subj_cred,
+			     profile, &lperms, OP_LINK, request, lname, tname,
+			     NULL, cond->uid, info, error);
+}
+
+/**
+ * aa_path_link - Handle hard link permission check
+ * @subj_cred: subject cred
+ * @label: the label being enforced  (NOT NULL)
+ * @old_dentry: the target dentry  (NOT NULL)
+ * @new_dir: directory the new link will be created in  (NOT NULL)
+ * @new_dentry: the link being created  (NOT NULL)
+ *
+ * Handle the permission test for a link & target pair.  Permission
+ * is encoded as a pair where the link permission is determined
+ * first, and if allowed, the target is tested.  The target test
+ * is done from the point of the link match (not start of DFA)
+ * making the target permission dependent on the link permission match.
+ *
+ * The subset test if required forces that permissions granted
+ * on link are a subset of the permission granted to target.
+ *
+ * Returns: %0 if allowed else error
+ */
+int aa_path_link(const struct cred *subj_cred,
+		 struct aa_label *label, struct dentry *old_dentry,
+		 const struct path *new_dir, struct dentry *new_dentry)
+{
+	struct path link = { .mnt = new_dir->mnt, .dentry = new_dentry };
+	struct path target = { .mnt = new_dir->mnt, .dentry = old_dentry };
+	struct inode *inode = d_backing_inode(old_dentry);
+	vfsuid_t vfsuid = i_uid_into_vfsuid(mnt_idmap(target.mnt), inode);
+	struct path_cond cond = {
+		.uid = vfsuid_into_kuid(vfsuid),
+		.mode = inode->i_mode,
+	};
+	char *buffer = NULL, *buffer2 = NULL;
+	struct aa_profile *profile;
+	int error;
+
+	/* buffer freed below, lname is pointer in buffer */
+	buffer = aa_get_buffer(false);
+	buffer2 = aa_get_buffer(false);
+	error = -ENOMEM;
+	if (!buffer || !buffer2)
+		goto out;
+
+	error = fn_for_each_confined(label, profile,
+			profile_path_link(subj_cred, profile, &link, buffer,
+					  &target, buffer2, &cond));
+out:
+	aa_put_buffer(buffer);
+	aa_put_buffer(buffer2);
+	return error;
+}
+
+static void update_file_ctx(struct aa_file_ctx *fctx, struct aa_label *label,
+			    u32 request)
+{
+	struct aa_label *l, *old;
+
+	/* update caching of label on file_ctx */
+	spin_lock(&fctx->lock);
+	old = rcu_dereference_protected(fctx->label,
+					lockdep_is_held(&fctx->lock));
+	l = aa_label_merge(old, label, GFP_ATOMIC);
+	if (l) {
+		if (l != old) {
+			rcu_assign_pointer(fctx->label, l);
+			aa_put_label(old);
+		} else
+			aa_put_label(l);
+		fctx->allow |= request;
+	}
+	spin_unlock(&fctx->lock);
+}
+
+static int __file_path_perm(const char *op, const struct cred *subj_cred,
+			    struct aa_label *label,
+			    struct aa_label *flabel, struct file *file,
+			    u32 request, u32 denied, bool in_atomic)
+{
+	struct aa_profile *profile;
+	struct aa_perms perms = {};
+	vfsuid_t vfsuid = i_uid_into_vfsuid(file_mnt_idmap(file),
+					    file_inode(file));
+	struct path_cond cond = {
+		.uid = vfsuid_into_kuid(vfsuid),
+		.mode = file_inode(file)->i_mode
+	};
+	char *buffer;
+	int flags, error;
+
+	/* revalidation due to label out of date. No revocation at this time */
+	if (!denied && aa_label_is_subset(flabel, label))
+		/* TODO: check for revocation on stale profiles */
+		return 0;
+
+	flags = PATH_DELEGATE_DELETED | (S_ISDIR(cond.mode) ? PATH_IS_DIR : 0);
+	buffer = aa_get_buffer(in_atomic);
+	if (!buffer)
+		return -ENOMEM;
+
+	/* check every profile in task label not in current cache */
+	error = fn_for_each_not_in_set(flabel, label, profile,
+			profile_path_perm(op, subj_cred, profile,
+					  &file->f_path, buffer,
+					  request, &cond, flags, &perms));
+	if (denied && !error) {
+		/*
+		 * check every profile in file label that was not tested
+		 * in the initial check above.
+		 *
+		 * TODO: cache full perms so this only happens because of
+		 * conditionals
+		 * TODO: don't audit here
+		 */
+		if (label == flabel)
+			error = fn_for_each(label, profile,
+				profile_path_perm(op, subj_cred,
+						  profile, &file->f_path,
+						  buffer, request, &cond, flags,
+						  &perms));
+		else
+			error = fn_for_each_not_in_set(label, flabel, profile,
+				profile_path_perm(op, subj_cred,
+						  profile, &file->f_path,
+						  buffer, request, &cond, flags,
+						  &perms));
+	}
+	if (!error)
+		update_file_ctx(file_ctx(file), label, request);
+
+	aa_put_buffer(buffer);
+
+	return error;
+}
+
+static int __file_sock_perm(const char *op, const struct cred *subj_cred,
+			    struct aa_label *label,
+			    struct aa_label *flabel, struct file *file,
+			    u32 request, u32 denied)
+{
+	int error;
+
+	/* revalidation due to label out of date. No revocation at this time */
+	if (!denied && aa_label_is_subset(flabel, label))
+		return 0;
+
+	/* TODO: improve to skip profiles cached in flabel */
+	error = aa_sock_file_perm(subj_cred, label, op, request, file);
+	if (denied) {
+		/* TODO: improve to skip profiles checked above */
+		/* check every profile in file label to is cached */
+		last_error(error, aa_sock_file_perm(subj_cred, flabel, op,
+						    request, file));
+	}
+	if (!error)
+		update_file_ctx(file_ctx(file), label, request);
+
+	return error;
+}
+
+/* for now separate fn to indicate semantics of the check */
+static bool __file_is_delegated(struct aa_label *obj_label)
+{
+	return unconfined(obj_label);
+}
+
+static bool __is_unix_file(struct file *file)
+{
+	struct socket *sock = (struct socket *) file->private_data;
+
+	lockdep_assert_in_rcu_read_lock();
+
+	if (!S_ISSOCK(file_inode(file)->i_mode))
+		return false;
+	/* sock and sock->sk can be NULL for sockets being set up or torn down */
+	if (!sock || !sock->sk)
+		return false;
+	if (sock->sk->sk_family == PF_UNIX)
+		return true;
+	return false;
+}
+
+static bool __unix_needs_revalidation(struct file *file, struct aa_label *label,
+				      u32 request)
+{
+	struct socket *sock = (struct socket *) file->private_data;
+
+	AA_BUG(!__is_unix_file(file));
+	lockdep_assert_in_rcu_read_lock();
+
+	struct aa_sk_ctx *skctx = aa_sock(sock->sk);
+
+	if (rcu_access_pointer(skctx->peer) !=
+	    rcu_access_pointer(skctx->peer_lastupdate))
+		return true;
+
+	return !__aa_subj_label_is_cached(rcu_dereference(skctx->label), label);
+}
+
+/**
+ * aa_file_perm - do permission revalidation check & audit for @file
+ * @op: operation being checked
+ * @subj_cred: subject cred
+ * @label: label being enforced   (NOT NULL)
+ * @file: file to revalidate access permissions on  (NOT NULL)
+ * @request: requested permissions
+ * @in_atomic: whether allocations need to be done in atomic context
+ *
+ * Returns: %0 if access allowed else error
+ */
+int aa_file_perm(const char *op, const struct cred *subj_cred,
+		 struct aa_label *label, struct file *file,
+		 u32 request, bool in_atomic)
+{
+	struct aa_file_ctx *fctx;
+	struct aa_label *flabel;
+	u32 denied;
+	int error = 0;
+
+	AA_BUG(!label);
+	AA_BUG(!file);
+
+	/* don't reaudit files closed during inheritance */
+	if (unlikely(file->f_path.dentry == aa_null.dentry))
+		return -EACCES;
+
+	fctx = file_ctx(file);
+
+	rcu_read_lock();
+	flabel  = rcu_dereference(fctx->label);
+	AA_BUG(!flabel);
+
+	/* revalidate access, if task is unconfined, or the cached cred
+	 * doesn't match or if the request is for more permissions than
+	 * was granted.
+	 *
+	 * Note: the test for !unconfined(flabel) is to handle file
+	 *       delegation from unconfined tasks
+	 */
+	denied = request & ~fctx->allow;
+	if (unconfined(label) || __file_is_delegated(flabel) ||
+	    (!denied && __is_unix_file(file) && !__unix_needs_revalidation(file, label, request)) ||
+	    (!denied && __aa_subj_label_is_cached(label, flabel))) {
+		rcu_read_unlock();
+		goto done;
+	}
+
+	/* slow path - revalidate access */
+	flabel  = aa_get_newest_label(flabel);
+	rcu_read_unlock();
+
+	if (path_mediated_fs(file->f_path.dentry))
+		error = __file_path_perm(op, subj_cred, label, flabel, file,
+					 request, denied, in_atomic);
+
+	else if (S_ISSOCK(file_inode(file)->i_mode))
+		error = __file_sock_perm(op, subj_cred, label, flabel, file,
+					 request, denied);
+	aa_put_label(flabel);
+
+done:
+	return error;
+}
+
+static void revalidate_tty(const struct cred *subj_cred, struct aa_label *label)
+{
+	struct tty_struct *tty;
+	int drop_tty = 0;
+
+	tty = get_current_tty();
+	if (!tty)
+		return;
+
+	spin_lock(&tty->files_lock);
+	if (!list_empty(&tty->tty_files)) {
+		struct tty_file_private *file_priv;
+		struct file *file;
+		/* TODO: Revalidate access to controlling tty. */
+		file_priv = list_first_entry(&tty->tty_files,
+					     struct tty_file_private, list);
+		file = file_priv->file;
+
+		if (aa_file_perm(OP_INHERIT, subj_cred, label, file,
+				 MAY_READ | MAY_WRITE, IN_ATOMIC))
+			drop_tty = 1;
+	}
+	spin_unlock(&tty->files_lock);
+	tty_kref_put(tty);
+
+	if (drop_tty)
+		no_tty();
+}
+
+struct cred_label {
+	const struct cred *cred;
+	struct aa_label *label;
+};
+
+static int match_file(const void *p, struct file *file, unsigned int fd)
+{
+	struct cred_label *cl = (struct cred_label *)p;
+
+	if (aa_file_perm(OP_INHERIT, cl->cred, cl->label, file,
+			 aa_map_file_to_perms(file), IN_ATOMIC))
+		return fd + 1;
+	return 0;
+}
+
+
+/* based on selinux's flush_unauthorized_files */
+void aa_inherit_files(const struct cred *cred, struct files_struct *files)
+{
+	struct aa_label *label = aa_get_newest_cred_label(cred);
+	struct cred_label cl = {
+		.cred = cred,
+		.label = label,
+	};
+	struct file *devnull = NULL;
+	unsigned int n;
+
+	revalidate_tty(cred, label);
+
+	/* Revalidate access to inherited open files. */
+	n = iterate_fd(files, 0, match_file, &cl);
+	if (!n) /* none found? */
+		goto out;
+
+	devnull = dentry_open(&aa_null, O_RDWR, cred);
+	if (IS_ERR(devnull))
+		devnull = NULL;
+	/* replace all the matching ones with this */
+	do {
+		replace_fd(n - 1, devnull, 0);
+	} while ((n = iterate_fd(files, n, match_file, &cl)) != 0);
+	if (devnull)
+		fput(devnull);
+out:
+	aa_put_label(label);
+}