On Fri, 2012-12-14 at 14:12 +0100, Michal Židek wrote: > On 12/13/2012 08:55 PM, Jakub Hrozek wrote: > > On Thu, Dec 13, 2012 at 02:16:25PM -0500, Simo Sorce wrote: > >> On Thu, 2012-12-13 at 13:19 -0500, Simo Sorce wrote: > >>> On Thu, 2012-12-13 at 18:06 +0100, Michal Židek wrote: > >>>> Patch 1: (sssd_nss part) In functions nss_cmd_getXXnam, if working with > >>>> a local domain and no results were found in sysdb, the entries are > >>>> invalidated in memory cache too. > >>> > >>> Ok in the general case this patch would be sort of useless because if > >>> you reach sssd_nss it means the mmap cache is already expired and not > >>> used. But I see you use this in patch 2, see comments below on that. > >>> > >>>> Patch 2: (tools part) Tools sss_userdel and sss_groupdel send > >>>> SSS_NSS_GETXXNAM requests after the entry was removed from sysdb. With > >>>> the first patch, this invalidates the entry in memory cache. > >>>> > >>>> Patches are in attachment. > >>> > >>> Hi Michal I see wht you use getpwnam/getgrnam, but doesn't it kind of > >>> defeat the purpose of userdel/groupdel ? > >>> > >>> If you use those functions then in most cases you are going to actually > >>> update the user in the cache, not delete it. Is this intentional ? It > >>> looks backwards given you are calling a 'del' function. > >>> > >>> It looks to me that what you really want is an explicit delete function > >>> in sssd_nss that you can call from the cmdline, is that correct ? > >> > >> Ha now I get what I was missing. > >> Jakub reminded me this is for the LOCAL provider. So in that case there > >> is obviously nothing to refresh. > >> > >> With that in mind, full ACK. > >> > >> Simo. > > > > Pushed to master and sssd-1-9 > > I used the mmap cache functions unconditionally here. In the case when > mmap cache initialization fails (which we treat as non-fatal error and > continue without the mmap cache), this would cause segfaults. > > The attached patch applies on top of the previous ones.
I think this may happen also with my initgr code. I think we should patch mmap functions to just return with an error if the context is null, so we do not have to put ugly checks everywhere. I prefer to nack this one and will send a patch about that shortly, can you open a ticket please ? Simo. -- Simo Sorce * Red Hat, Inc * New York _______________________________________________ sssd-devel mailing list [email protected] https://lists.fedorahosted.org/mailman/listinfo/sssd-devel
