On certain systems, in certain pathalogical cases, current's cwd can
be deleted while we're still processing a syscall. This should prevent
the system from evicting the inode while we're still referencing it.
This seems to fix the bug I reported here:
https://www.redhat.com/archives/linux-audit/2012-August/msg00017.html
Signed-off-by: Peter Moody <pmoody(a)google.com>
---
kernel/auditsc.c | 2 ++
1 files changed, 2 insertions(+), 0 deletions(-)
diff --git a/kernel/auditsc.c b/kernel/auditsc.c
index 4b96415..e86b8b9 100644
--- a/kernel/auditsc.c
+++ b/kernel/auditsc.c
@@ -2064,6 +2064,7 @@ void __audit_getname(const char *name)
if (!context->pwd.dentry)
get_fs_pwd(current->fs, &context->pwd);
+ path_get(&context->pwd);
}
/* audit_putname - intercept a putname request
@@ -2091,6 +2092,7 @@ void audit_putname(const char *name)
n->name, n->name ?: "(null)");
}
#endif
+ path_put(&context->pwd);
__putname(name);
}
#if AUDIT_DEBUG
--
1.7.7.3