On Wed, 16 Dec 2015, Jessica Yu wrote:
> +++ Miroslav Benes [16/12/15 13:02 +0100]:
> > On Mon, 30 Nov 2015, Jessica Yu wrote:
> >
> > > Livepatch needs to utilize the symbol information contained in the
> > > mod_arch_specific struct in order to be able to call the s390
> > > apply_relocate_add() function to apply relocations. Remove the redundant
> > > vfree() in module_finalize() since module_arch_freeing_init() (which also
> > > frees
> > > said structures) is called in do_init_module(). Keep a reference to
> > > syminfo if
> > > the module is a livepatch module and free the structures in
> > > module_arch_cleanup(). If the module isn't a livepatch module, we free the
> > > structures in module_arch_freeing_init() as usual.
> > >
> > > Signed-off-by: Jessica Yu <jeyu@xxxxxxxxxx>
> > > ---
> > > arch/s390/kernel/module.c | 13 +++++++++++--
> > > 1 file changed, 11 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/arch/s390/kernel/module.c b/arch/s390/kernel/module.c
> > > index 0c1a679..17a1979 100644
> > > --- a/arch/s390/kernel/module.c
> > > +++ b/arch/s390/kernel/module.c
> > > @@ -51,6 +51,9 @@ void *module_alloc(unsigned long size)
> > >
> > > void module_arch_freeing_init(struct module *mod)
> > > {
> > > + if (mod->klp)
> > > + return;
> > > +
> > > vfree(mod->arch.syminfo);
> > > mod->arch.syminfo = NULL;
> > > }
> >
> > Hm, this is problematic. module_arch_freeing_init() is called from
> > module_deallocate() and this is called in the error path in load_module().
> > So if there was an error during load_module() of livepatch module which
> > led to free_modinfo label or behind mod->arch.syminfo would not be freed
> > at all. module_arch_cleanup() is called earlier under
> > free_arch_cleanup.
>
> Gah. Good catch. How about we add an additional check to see if
> mod->state is MODULE_STATE_LIVE? If so, we can return. This means
> we're coming from do_init_module(). If module_arch_freeing_init() was
> called and the module was in a state other than MODULE_STATE_LIVE
> (UNFORMED, GOING, COMING), we know we need to free. Think that would
> work?
It would I think. module_arch_freeing_init is called thrice.
1. in free_module(). The module is in UNFORMED state so everything is
fine.
2. in module_deallocate(). The module is not LIVE yet here so ok as well.
3. in do_init_module() and the module is LIVE there.
So yes, you are right, that should work.
Miroslav
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
