Re: [PATCH] libkmod: fix uninitialized variable usage warnings

[Lucas De Marchi <lucas.de.marchi@gmail.com>]

On Tue, Nov 7, 2017 at 3:59 AM, Eugene Syromiatnikov <esyr@redhat.com> wrote:
> On Mon, Nov 06, 2017 at 07:24:55PM -0800, Lucas De Marchi wrote:
>> On Mon, Nov 6, 2017 at 1:57 PM, Eugene Syromiatnikov <esyr@redhat.com> wrote:
>> > There are two places where _cleanup_free_ variables are not initialised
>> > by the time function exits that have been caught by gcc:
>> >
>> >         In file included from libkmod/libkmod.c:35:0:
>> >         libkmod/libkmod.c: In function 'kmod_lookup_alias_is_builtin':
>> >         ./shared/util.h:73:13: warning: 'line' may be used uninitialized in this function [-Wmaybe-uninitialized]
>> >                  free(*(void**) p);
>> >                      ^
>> >         libkmod/libkmod.c:551:23: note: 'line' was declared here
>> >           _cleanup_free_ char *line;
>> >                                ^
>> >         In file included from libkmod/libkmod-module.c:42:0:
>> >         libkmod/libkmod-module.c: In function 'kmod_module_probe_insert_module':
>> >         ./shared/util.h:73:13: warning: 'cmd' may be used uninitialized in this function [-Wmaybe-uninitialized]
>> >                  free(*(void**) p);
>> >                      ^
>> >         libkmod/libkmod-module.c:996:23: note: 'cmd' was declared here
>> >           _cleanup_free_ char *cmd;
>> >                        ^
>> >
>> > This patch initializes them to NULL so free become no-op in these cases.
>> > ---
>> >  libkmod/libkmod-module.c | 2 +-
>> >  libkmod/libkmod.c        | 2 +-
>> >  2 files changed, 2 insertions(+), 2 deletions(-)
>> >
>> > diff --git a/libkmod/libkmod-module.c b/libkmod/libkmod-module.c
>> > index 0a3ef11..6f23c1a 100644
>> > --- a/libkmod/libkmod-module.c
>> > +++ b/libkmod/libkmod-module.c
>> > @@ -995,7 +995,7 @@ static int module_do_install_commands(struct kmod_module *mod,
>> >  {
>> >         const char *command = kmod_module_get_install_commands(mod);
>> >         char *p;
>> > -       _cleanup_free_ char *cmd;
>> > +       _cleanup_free_ char *cmd = NULL;
>> >         int err;
>> >         size_t cmdlen, options_len, varlen;
>> >
>> > diff --git a/libkmod/libkmod.c b/libkmod/libkmod.c
>> > index 69fe431..964772d 100644
>> > --- a/libkmod/libkmod.c
>> > +++ b/libkmod/libkmod.c
>> > @@ -556,7 +556,7 @@ finish:
>> >
>> >  bool kmod_lookup_alias_is_builtin(struct kmod_ctx *ctx, const char *name)
>> >  {
>> > -       _cleanup_free_ char *line;
>> > +       _cleanup_free_ char *line = NULL;
>>
>> This seems to be a bogus warning. See the line just below... there's
>> no way to exit this function without first  assigning 'line'.
>
> In this instance, I initially thought that gcc is smart and derives
> "uninitialised variable" from the lookup_builtin_file call, but nope, it
> was happy only after line was initialised on declaration. On the other
> hand, documentation[1] mentions that cleanup function is also called
> during stack unwinding in case -fexceptions is enabled, and RHEL's RPM
> package indeed has this flag, at least, so this GCC's complaint still
> seems relevant.

This is likely to trigger "double assignment" warnings on other compilers.
A good question would be.... why is -fexceptions being used in a C-only library?
If it was a binary or a library that links to a C++-library, I would
understand, but in
this case it seems nonsense. /me lost

An acceptable solution would be to just merge the 2 lines so we
initialize line on
declaration.

I added -fexceptions here to reproduce the issue, but it didn't
reproduce using gcc 7.2.1
on Fedora 27.  I also remember not having the issue with gcc 6. So I
still think it's
a compiler bug, although I don't mind if the solution proposed above
is applied (but mention
on commit message the exact compiler version).


Lucas De Marchi