[PATCH] get rid of AVC_PATH postponed treatment
Selinux folks had been complaining about the lack of AVC_PATH
records when audit is disabled. I must admit my stupidity - I assumed
that avc_audit() really couldn't use audit_log_d_path() because of
deadlocks (== could be called with dcache_lock or vfsmount_lock held).
Shouldn't have made that assumption - it never gets called that way.
It _is_ called under spinlocks, but not those.
Since audit_log_d_path() uses ab->gfp_mask for allocations,
kmalloc() in there is not a problem. IOW, the simple fix is sufficient:
let's rip AUDIT_AVC_PATH out and simply generate pathname as part of main
record. It's trivial to do.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: James Morris <jmorris@namei.org>
diff --git a/security/selinux/avc.c b/security/selinux/avc.c
index ecd0673..0e69adf 100644
--- a/security/selinux/avc.c
+++ b/security/selinux/avc.c
@@ -570,10 +570,12 @@
case AVC_AUDIT_DATA_FS:
if (a->u.fs.dentry) {
struct dentry *dentry = a->u.fs.dentry;
- if (a->u.fs.mnt)
- audit_avc_path(dentry, a->u.fs.mnt);
- audit_log_format(ab, " name=");
- audit_log_untrustedstring(ab, dentry->d_name.name);
+ if (a->u.fs.mnt) {
+ audit_log_d_path(ab, "path=", dentry, a->u.fs.mnt);
+ } else {
+ audit_log_format(ab, " name=");
+ audit_log_untrustedstring(ab, dentry->d_name.name);
+ }
inode = dentry->d_inode;
} else if (a->u.fs.inode) {
struct dentry *dentry;
@@ -624,9 +626,8 @@
case AF_UNIX:
u = unix_sk(sk);
if (u->dentry) {
- audit_avc_path(u->dentry, u->mnt);
- audit_log_format(ab, " name=");
- audit_log_untrustedstring(ab, u->dentry->d_name.name);
+ audit_log_d_path(ab, "path=",
+ u->dentry, u->mnt);
break;
}
if (!u->addr)
