Re: [PATCH 3/6] kallsyms/bpf: Set module buildid in bpf_address_lookup()

[Alexei Starovoitov]

On Fri, Nov 7, 2025 at 5:08 AM Petr Mladek <pmladek@xxxxxxxx> wrote:
>
> On Wed 2025-11-05 18:53:23, Alexei Starovoitov wrote:
> > On Wed, Nov 5, 2025 at 6:24 AM Petr Mladek <pmladek@xxxxxxxx> wrote:
> > >
> > > Make bpf_address_lookup() compatible with module_address_lookup()
> > > and clear the pointer to @modbuildid together with @modname.
> > >
> > > It is not strictly needed because __sprint_symbol() reads @modbuildid
> > > only when @modname is set. But better be on the safe side and make
> > > the API more safe.
> > >
> > > Fixes: 9294523e3768 ("module: add printk formats to add module build ID to stacktraces")
> > > Signed-off-by: Petr Mladek <pmladek@xxxxxxxx>
> > > ---
> > >  include/linux/filter.h | 15 +++++++++++----
> > >  kernel/kallsyms.c      |  4 ++--
> > >  2 files changed, 13 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/include/linux/filter.h b/include/linux/filter.h
> > > index f5c859b8131a..b7b95840250a 100644
> > > --- a/include/linux/filter.h
> > > +++ b/include/linux/filter.h
> > > @@ -1362,12 +1362,18 @@ struct bpf_prog *bpf_prog_ksym_find(unsigned long addr);
> > >
> > >  static inline int
> > >  bpf_address_lookup(unsigned long addr, unsigned long *size,
> > > -                  unsigned long *off, char **modname, char *sym)
> > > +                  unsigned long *off, char **modname,
> > > +                  const unsigned char **modbuildid, char *sym)
> > >  {
> > >         int ret = __bpf_address_lookup(addr, size, off, sym);
> > >
> > > -       if (ret && modname)
> > > -               *modname = NULL;
> > > +       if (ret) {
> > > +               if (modname)
> > > +                       *modname = NULL;
> > > +               if (modbuildid)
> > > +                       *modbuildid = NULL;
> > > +       }
> > > +
> > >         return ret;
> > >  }
> > >
> > > @@ -1433,7 +1439,8 @@ static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
> > >
> > >  static inline int
> > >  bpf_address_lookup(unsigned long addr, unsigned long *size,
> > > -                  unsigned long *off, char **modname, char *sym)
> > > +                  unsigned long *off, char **modname,
> > > +                  const unsigned char **modbuildid, char *sym)
> > >  {
> > >         return 0;
> > >  }
> > > diff --git a/kernel/kallsyms.c b/kernel/kallsyms.c
> > > index 9455e3bb07fc..efb12b077220 100644
> > > --- a/kernel/kallsyms.c
> > > +++ b/kernel/kallsyms.c
> > > @@ -374,8 +374,8 @@ static int kallsyms_lookup_buildid(unsigned long addr,
> > >         ret = module_address_lookup(addr, symbolsize, offset,
> > >                                     modname, modbuildid, namebuf);
> > >         if (!ret)
> > > -               ret = bpf_address_lookup(addr, symbolsize,
> > > -                                        offset, modname, namebuf);
> > > +               ret = bpf_address_lookup(addr, symbolsize, offset,
> > > +                                        modname, modbuildid, namebuf);
> >
> > The initial bpf_address_lookup() 8 years ago was trying
> > to copy paste args and style of kallsyms_lookup().
> > It was odd back then. This change is doubling down on the wrong thing.
> > It's really odd to pass a pointer into bpf_address_lookup()
> > so it zero initializes it.
> > bpf ksyms are in the core kernel. They're never in modules.
> > Just call __bpf_address_lookup() here and remove the wrapper.
>
> I agree that it is ugly. It would make sense to initialize the
> pointers in kallsyms_lookup_buildid and call there
> __bpf_address_lookup() variant. Something like:
>
> static int kallsyms_lookup_buildid(unsigned long addr,
>                         unsigned long *symbolsize,
>                         unsigned long *offset, char **modname,
>                         const unsigned char **modbuildid, char *namebuf)
> {
>         int ret;
>
>         if (modname)
>                 *modname = NULL;
>         if (modbuildid)
>                 *modbuildid = NULL;
>         namebuf[0] = 0;
> [...]
>         if (!ret)
>                 ret = __bpf_address_lookup(addr, symbolsize, offset, namebuf);
>
> }

Yes. Exactly.

> As a result bpf_address_lookup() won't have any caller.
> And __bpf_address_lookup() would have two callers.

yep

> It would make sense to remove bpf_address_lookup() and
> rename __bpf_address_lookup() to bpf_address_lookup().

yep

> How does that sound?
> Would you prefer to do this in one patch or in two steps, please?

Whichever way is easier. I think it's fine to do it in one go,
though it crosses different subsystems.