Re: [External] Re: [PATCH] libbpf: correct the macro KERNEL_VERSION for old kernel

[宋锐]

> > The introduced header file linux/version.h in libbpf_probes.c may have a
> > wrong macro KERNEL_VERSION for calculating LINUX_VERSION_CODE in some old
> > kernel (Debian9, 10). Below is a version info example from Debian 10.
> >
> > release: 4.19.0-22-amd64
> > version: #1 SMP Debian 4.19.260-1 (2022-09-29)
> >
> > The macro KERNEL_VERSION is defined to (((a) << 16) + ((b) << 8)) + (c)),
> > which a, b, and c stand for major, minor and patch version. So in example here,
> > the major is 4, minor is 19, patch is 260, the LINUX_VERSION(4, 19, 260) which
> > is 267268 should be matched to LINUX_VERSION_CODE. However, the KERNEL_VERSION_CODE
> > in linux/version.h is defined to 267263.
> >
> > I noticed that the macro KERNEL_VERSION in linux/version.h of some new kernel is
> > defined to (((a) << 16) + ((b) << 8) + ((c) > 255 ? 255 : (c))). And
> > KERNEL_VERSION(4, 19, 260) is equal to 267263 which is the right LINUX_VERSION_CODE.
> >
> > The mismatched LINUX_VERSION_CODE which will cause failing to load kprobe BPF
> > programs in the version check of BPF syscall.
> >
> > The return value of get_kernel_version in libbpf_probes.c should be matched to
> > LINUX_VERSION_CODE by correcting the macro KERNEL_VERSION.
> >
> > Signed-off-by: songrui.771 <songrui.771@xxxxxxxxxxxxx>
>
> This needs to be your name, not your email alias (do you use ".771" as a
> name to sign things with?)

Thanks for your reminding. I will change it.
>
> > ---
> >  tools/lib/bpf/libbpf_probes.c | 10 +++++++---
> >  1 file changed, 7 insertions(+), 3 deletions(-)
> >
> > diff --git a/tools/lib/bpf/libbpf_probes.c b/tools/lib/bpf/libbpf_probes.c
> > index 4f3bc968ff8e..5b22a880c7e7 100644
> > --- a/tools/lib/bpf/libbpf_probes.c
> > +++ b/tools/lib/bpf/libbpf_probes.c
> > @@ -18,6 +18,10 @@
> >  #include "libbpf.h"
> >  #include "libbpf_internal.h"
> >
> > +#ifndef LIBBPF_KERNEL_VERSION
> > +#define LIBBPF_KERNEL_VERSION(a, b, c) (((a) << 16) + ((b) << 8) + ((c) > 255 ? 255 : (c)))
> > +#endif
>
> What is wrong with using the KERNEL_VERSION() macro, it should be fixed
> to work properly here, right?  Did we not get this resolved in the
> main portion of the kernel already?

The KERNEL_VERSION() macro from linux/version.h is wrong in some old
kernel(Debian 9, 10) that we would like to support. As you said, the
problem was resolved in the newer kernel. Here is the difference:

linux/version.h
in older kernel: #define KERNEL_VERSION(a, b, c) (((a) << 16) + ((b)
<< 8)) + (c)))
in newer kernel: #define KERNEL_VERSION(a, b, c) KERNEL_VERSION(a, b,
c) (((a) << 16) + ((b) << 8) + ((c) > 255 ? 255 : (c)))

Using the KERNEL_VERSION macro in the older kernel returns the kern
version  which is  mismatched to the LINUX_VERSION_CODE that will
cause failing to load the BPF kprobe program.

In my opinion, it is a more generic solution that corrects the
KERNEL_VERSION() macro in libbpf to support some old kernel.

Hope I make that clear. Thanks.

Jerry Song