tracing/perf: Use strndup_user() instead of buggy open-coded version
authorJann Horn <jannh@google.com>
Wed, 20 Feb 2019 16:54:43 +0000 (17:54 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 23 Mar 2019 19:09:56 +0000 (20:09 +0100)
commit 83540fbc8812a580b6ad8f93f4c29e62e417687e upstream.

The first version of this method was missing the check for
`ret == PATH_MAX`; then such a check was added, but it didn't call kfree()
on error, so there was still a small memory leak in the error case.
Fix it by using strndup_user() instead of open-coding it.

Link: http://lkml.kernel.org/r/20190220165443.152385-1-jannh@google.com
Cc: Ingo Molnar <mingo@kernel.org>
Cc: stable@vger.kernel.org
Fixes: 0eadcc7a7bc0 ("perf/core: Fix perf_uprobe_init()")
Reviewed-by: Masami Hiramatsu <mhiramat@kernel.org>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kernel/trace/trace_event_perf.c

index 69a3fe9..e6945b5 100644 (file)
@@ -298,15 +298,13 @@ int perf_uprobe_init(struct perf_event *p_event, bool is_retprobe)
 
        if (!p_event->attr.uprobe_path)
                return -EINVAL;
-       path = kzalloc(PATH_MAX, GFP_KERNEL);
-       if (!path)
-               return -ENOMEM;
-       ret = strncpy_from_user(
-               path, u64_to_user_ptr(p_event->attr.uprobe_path), PATH_MAX);
-       if (ret == PATH_MAX)
-               return -E2BIG;
-       if (ret < 0)
-               goto out;
+
+       path = strndup_user(u64_to_user_ptr(p_event->attr.uprobe_path),
+                           PATH_MAX);
+       if (IS_ERR(path)) {
+               ret = PTR_ERR(path);
+               return (ret == -EINVAL) ? -E2BIG : ret;
+       }
        if (path[0] == '\0') {
                ret = -EINVAL;
                goto out;