Selftests for kernel symbol name length limitation

[fbq]

Rust support enlarged the length limitation of kernel symbols. We should provide selftests showing the longest name is supported, a simple one would be like:

diff --git a/kernel/kallsyms.c b/kernel/kallsyms.c
index 4067564ec59f..e356a10bcb51 100644
--- a/kernel/kallsyms.c
+++ b/kernel/kallsyms.c
@@ -865,8 +865,22 @@ static const struct proc_ops kallsyms_proc_ops = {
        .proc_release   = seq_release_private,
 };

+#define DI(name) n##name##name
+#define DDI(name) DI(n##name##name)
+#define DDDI(name) DDI(n##name##name)
+#define DDDDI(name) DDDI(n##name##name)
+#define DDDDDI(name) DDDDI(n##name##name)
+
+/* Generate a symbol whose name length is 511 */
+#define LONGEST_NAME DDDDDI(g1h2i3j4k5l6m7n)
+
+extern int LONGEST_NAME(void) { return 0; }
+
 static int __init kallsyms_init(void)
 {
+       /* Showing that the longest name is supported */
+       BUG_ON(!kallsyms_lookup_name(__stringify(LONGEST_NAME)));
+
        proc_create("kallsyms", 0444, NULL, &kallsyms_proc_ops);
        return 0;
 }

However, we may want something better as suggested by @ojeda . Therefore create this issue to track work on this.

[sergiocollado]

@fbq @ojeda I will have a go to try to write the selftest for this issue, it it is ok.

[ojeda]

Of course!

[fbq]

@fbq @ojeda I will have a go to try to write the selftest for this issue, it it is ok.

FWIW, I had a go myself when I was playing around rust_selftests.rs, at that time I used procedure macro to generate the longest symbol. I cannot find the code though ;-(

[ojeda]

Now that you mention it Boqun, I recall I had started some code on that too -- I have rebased and uploaded it to ojeda@0329a98

In my case I went the C route, since I thought testing the longest symbol would be worth it in all cases, not just for Rust, so I went with a hostprog to generate the symbol.

Hope that helps!

[sergiocollado]

For testing this with selftests I was thinking to write a test_module, define and export a symbol with the max length, and then check if the symbol is present in the /proc/kallsyms. What do you think about this approach @fbq @ojeda ? Would it be ok?

[sergiocollado]

@ojeda @fbq , I have created a draft PR here: #923 . Please have a look, any feedback will be welcomed.

There are still some open points. For example, the placement of the test could be in other folder. I placed it in that folder (/lib), because there is where the most of the other test modules are. Or maybe the code could be written in another way.

I have written the test module, according to the guide: https://docs.kernel.org/dev-tools/kselftest.html#test-module

And the kprobes - kallsyms used in the code, are due that since Linux 5.7.7 kallsyms is not exported (this is summarized here: https://nskernel.gitbook.io/kernel-play-guide/accessing-the-non-exported-in-modules#for-using-after-5.7.7 ). And the
solution I took is the described here: https://sysprog21.github.io/lkmpg/ , at "10 System Calls".

Also if you could test it indepedently it would be great. (just to confirm it works)

[sergiocollado]

I have created this PR #956, which uses kunit test, to verify the longest symbol name length