[Crash-utility] [PATCH v4 4/4] Add check if an syment element is installed one more time

[Philipp Rudo]

Hi Tao,

On Thu, 23 Sep 2021 21:46:14 +0800
Tao Liu <ltao at redhat.com> wrote:

> Hi Philipp,
> 
> Thanks  for reviewing the patch!
> 
> On Thu, Sep 23, 2021 at 8:19 PM Philipp Rudo <prudo at redhat.com> wrote:
> >
> > Hi Tao,
> >
> > On Sat, 18 Sep 2021 15:59:32 +0800
> > Tao Liu <ltao at redhat.com> wrote:
> >  
> > > symname_hash_install won't check if spn has been installed before. If
> > > it does, the second install will corrupt the hash table as well as
> > > spn->cnt counting. This patch adds the check to avoid such risks.
> > >
> > > Signed-off-by: Tao Liu <ltao at redhat.com>
> > > ---
> > >  symbols.c | 16 +++++++++++++++-
> > >  1 file changed, 15 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/symbols.c b/symbols.c
> > > index f7157b1..6d12c55 100644
> > > --- a/symbols.c
> > > +++ b/symbols.c
> > > @@ -1147,6 +1147,20 @@ mod_symtable_hash_remove_range(struct syment *from, struct syment *to)
> > >               symname_hash_remove(st->mod_symname_hash, sp);
> > >  }
> > >
> > > +static inline int
> > > +syment_is_installed(struct syment *table[], struct syment *spn)
> > > +{
> > > +     struct syment *sp;
> > > +     int index;
> > > +
> > > +     index = SYMNAME_HASH_INDEX(spn->name);
> > > +     for (sp = table[index]; sp; sp = sp->name_hash_next) {
> > > +             if (sp == spn)
> > > +                     return TRUE;
> > > +     }
> > > +     return FALSE;
> > > +}
> > > +
> > >  /*
> > >   *  Install a single static kernel symbol into the symname_hash.
> > >   */
> > > @@ -1156,7 +1170,7 @@ symname_hash_install(struct syment *table[], struct syment *spn)
> > >       struct syment *sp;
> > >          int index;
> > >
> > > -     if (!spn)
> > > +     if (!spn || syment_is_installed(table, spn))
> > >               return;
> > >
> > >          index = SYMNAME_HASH_INDEX(spn->name);  
> >
> > hmm... not sure if this is a little bit over the top. The idea I had
> > was in your v3 simply replace
> >
> > assert(sp != spn);
> >
> > by
> >
> > if (sp == spn) {
> >         error(WARNING, "Symbol %s already installed in symname_hash\n",
> >               sp->name);
> >         continue;
> > }
> >  
> 
> It may not be easy to replace with "check if sp == spn and continue".
> For example, if we already have
> 3 syments installed, which all have the same name, such as
> (sp1{cnt == 3, name == "str"}, sp2{cnt == 3, name == "str"}, sp3{cnt
> == 3, name == "str"})
> in the hashtable, and we will install sp3 again(sp3 == spn) into the hashtable.
> 
> The cnt will get increased for sp1 and sp2 in the following code:
> while (sp) {
>         if (sp == spn) {
>              error(....);
>              continue;
>         }
>         if (STREQ(sp->name, spn->name)) {
>                 sp->cnt++;
>                 spn->cnt++;
>         }
> .....
> }
> 
> However, when iteration reaches sp3, it will not increase cnt and not
> install spn. Thus we will have
> (sp1{cnt == 4}, sp2{cnt == 4}, sp3{cnt == 3}) in the hashtable, cnt
> gets corrupted. In other words,
> we need to revert all the previous cnt++ when we reach sp == spn. It
> will require
> more code to implement the 'revert' operation, and it is not as clear
> as syment_is_installed
> check.

you are right. I totally missed that. In fact my suggestion is even
worse. At the beginning the function sets

spn->cnt = 1;

So in the scenario you described above spn3->cnt == 1 should be true...

Having that said, your patch looks good to me

Reviewed-by: Philipp Rudo <prudo at redhat.com>

Thanks
Philipp

> > That's less code plus the warning makes it easier to detect that there
> > is a problem (for me the case sp == spn is a sign for a bug in crash).
> > What do you think?
> >  
> 
> From my point,
> 1) assert() check: Use least code, simple and clear, but too strict.
> 2) syment_is_installed check: Use more code, clear, acceptable to me.
> 3) introduce 'cnt++ revert' operation: I haven't think of a better way, from
>     the current inspection, it uses more code, and is not elegant.
> What do you think?
> 
> Thanks,
> Tao Liu
> 
> > Thanks
> > Philipp
> >  
>