On Thu, Jan 12, 2023 at 3:46 PM Daniel P. Berrangé <berra...@redhat.com> wrote:
>
> On Thu, Jan 12, 2023 at 03:26:56PM +0100, Ilya Dryomov wrote:
> > On Thu, Jan 12, 2023 at 1:35 PM Daniel P. Berrangé <berra...@redhat.com>
> > wrote:
> > >
> > > On Sun, Nov 20, 2022 at 04:28:34AM -0600, Or Ozeri wrote:
> > > > Add const modifier to passphrases,
> > > > and remove redundant stack variable passphrase_len.
> > > >
> > > > Signed-off-by: Or Ozeri <o...@il.ibm.com>
> > > > ---
> > > > block/rbd.c | 24 ++++++++++--------------
> > > > 1 file changed, 10 insertions(+), 14 deletions(-)
> > > >
> > > > diff --git a/block/rbd.c b/block/rbd.c
> > > > index f826410f40..e575105e6d 100644
> > > > --- a/block/rbd.c
> > > > +++ b/block/rbd.c
> > > > @@ -330,7 +330,7 @@ static int qemu_rbd_set_keypairs(rados_t cluster,
> > > > const char *keypairs_json,
> > > > #ifdef LIBRBD_SUPPORTS_ENCRYPTION
> > > > static int qemu_rbd_convert_luks_options(
> > > > RbdEncryptionOptionsLUKSBase *luks_opts,
> > > > - char **passphrase,
> > > > + const char **passphrase,
> > > > size_t *passphrase_len,
> > > > Error **errp)
> > > > {
> > > > @@ -341,7 +341,7 @@ static int qemu_rbd_convert_luks_options(
> > > > static int qemu_rbd_convert_luks_create_options(
> > > > RbdEncryptionCreateOptionsLUKSBase *luks_opts,
> > > > rbd_encryption_algorithm_t *alg,
> > > > - char **passphrase,
> > > > + const char **passphrase,
> > > > size_t *passphrase_len,
> > > > Error **errp)
> > > > {
> > > > @@ -384,8 +384,7 @@ static int qemu_rbd_encryption_format(rbd_image_t
> > > > image,
> > > > Error **errp)
> > > > {
> > > > int r = 0;
> > > > - g_autofree char *passphrase = NULL;
> > > > - size_t passphrase_len;
> > > > + g_autofree const char *passphrase = NULL;
> > >
> > > This looks wierd. If it is as const string, why are
> > > we free'ing it ? Either want g_autofree, or const,
> > > but not both.
> >
> > Just curious, is it a requirement imposed by g_autofree? Otherwise
> > pointer constness and pointee lifetime are completely orthogonal and
> > freeing (or, in this case, wanting to auto-free) an object referred to
> > by a const pointer seems perfectly fine to me.
>
> Free'ing a const point is not OK
>
> $ cat c.c
> #include <stdlib.h>
> void bar(const char *foo) {
> free(foo);
> }
>
> $ gcc -Wall -c c.c
> c.c: In function ‘bar’:
> c.c:5:10: warning: passing argument 1 of ‘free’ discards ‘const’ qualifier
> from pointer target type [-Wdiscarded-qualifiers]
> 5 | free(foo);
> | ^~~
> In file included from c.c:2:
> /usr/include/stdlib.h:568:25: note: expected ‘void *’ but argument is of type
> ‘const char *’
> 568 | extern void free (void *__ptr) __THROW;
> | ~~~~~~^~~~~
>
> The g_autofree happens to end up hiding this warning, because the const
> annotation isn't propagated to the registere callback, but that doesn't
> mean we should do that.
>
> When a programmer sees a variable annotated const, they expect that
> either someone else is responsible for free'ing it, or that the data
> is statically initialized or stack allocated and thus doesn't need
> free'ing. So g_autofree + const is just wrong.
FWIW many believe that this specification of free() was a mistake and
that it should have been specified to take const void *. Some projects
actually went ahead and fixed that: kfree() and friends in the Linux
kernel take const void *, for example. C++ delete operator works on
const pointers as well -- because object creation and destruction is
fundamentally independent of modification.
But this is more of a philosophical thing... I asked about g_autofree
because a quick grep revealed a bunch of g_autofree const char * locals
in the tree. Or would probably prefer to just drop const here ;)
Thanks,
Ilya
