[systemd-devel] [PATCH 2/2] logind: do not continue when runtime creation path fails

[Lennart Poettering]

On Tue, 10.02.15 13:45, Philippe De Swert (philippedeswert at gmail.com) wrote:

> Hi,
> 
> On 10/02/15 13:30, Lennart Poettering wrote:
> > On Sun, 08.02.15 22:21, Philippe De Swert (philippedeswert at gmail.com) wrote:
> > 
> >> In user_mkdir_runtime_path() there is no need to continue if the
> >> creation of the directory fails for whatever reason, as subsequent
> >> actions on that directory will fail anyway.
> >>
> >> Found with Coverity. Fixes: CID#1237538
> >> ---
> >>  src/login/logind-user.c | 3 ++-
> >>  1 file changed, 2 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/src/login/logind-user.c b/src/login/logind-user.c
> >> index 7b40ef5..a329457 100644
> >> --- a/src/login/logind-user.c
> >> +++ b/src/login/logind-user.c
> >> @@ -322,7 +322,8 @@ static int user_mkdir_runtime_path(User *u) {
> >>          if (path_is_mount_point(p, false) <= 0) {
> >>                  _cleanup_free_ char *t = NULL;
> >>  
> >> -                mkdir(p, 0700);
> >> +                if(mkdir(p, 0700) < 0)
> >> +                         goto fail;
> > 
> > No, it's completely OK if this fails, and it is very likely that it
> > *will* fail, since the dir might already exist, and we thus get
> > EEXIST.
> 
> True, I somehow always forget that is an error case too. However the
> "problem" is still there if creation fails, filesystem full/ro, ...
> Although I need to admit it is rather cosmetic as problems will be
> caught later anyway.

Yes, the idea is that the mount() later on will catch all errors, and
there's nothing gained by checking any failure of mkdir()...

Lennart

-- 
Lennart Poettering, Red Hat