Return-Path: Received: from mail-qk0-f177.google.com ([209.85.220.177]:33994 "EHLO mail-qk0-f177.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752287AbcD0ON3 (ORCPT ); Wed, 27 Apr 2016 10:13:29 -0400 Received: by mail-qk0-f177.google.com with SMTP id r184so18801943qkc.1 for ; Wed, 27 Apr 2016 07:13:29 -0700 (PDT) Message-ID: <1461766405.4403.3.camel@poochiereds.net> Subject: Re: [PATCH v4 2/3] gssd: using syscalls directly to change thread's identity From: Jeff Layton To: Olga Kornievskaia Cc: Olga Kornievskaia , Steve Dickson , linux-nfs Date: Wed, 27 Apr 2016 10:13:25 -0400 In-Reply-To: References: <1461677655-68294-1-git-send-email-kolga@netapp.com> <1461677655-68294-3-git-send-email-kolga@netapp.com> <1461689470.4200.7.camel@poochiereds.net> Content-Type: text/plain; charset="UTF-8" Mime-Version: 1.0 Sender: linux-nfs-owner@vger.kernel.org List-ID: On Wed, 2016-04-27 at 10:09 -0400, Olga Kornievskaia wrote: > On Tue, Apr 26, 2016 at 12:51 PM, Jeff Layton t> wrote: > > On Tue, 2016-04-26 at 09:34 -0400, Olga Kornievskaia wrote: > > > For the threaded version we have to set uid,gid per thread > > > instead > > > of per process. glibc setresuid() when called from a thread, > > > it'll > > > send a signal to all other threads to synchronize the uid in all > > > other threads. To bypass this, we have to call syscall() > > > directly. > > > > > > Signed-off-by: Olga Kornievskaia > > > Reviewed-by: Steve Dickson > > > --- > > >  utils/gssd/gssd_proc.c | 12 +++++++++--- > > >  1 file changed, 9 insertions(+), 3 deletions(-) > > > > > > diff --git a/utils/gssd/gssd_proc.c b/utils/gssd/gssd_proc.c > > > index e2e95dc..39c4b17 100644 > > > --- a/utils/gssd/gssd_proc.c > > > +++ b/utils/gssd/gssd_proc.c > > > @@ -69,6 +69,7 @@ > > >  #include > > >  #include > > >  #include > > > +#include > > > > > >  #include "gssd.h" > > >  #include "err_util.h" > > > @@ -436,7 +437,7 @@ change_identity(uid_t uid) > > >       struct passwd   *pw; > > > > > >       /* drop list of supplimentary groups first */ > > > -     if (setgroups(0, NULL) != 0) { > > > +     if (syscall(SYS_setgroups, 0, 0) != 0) { > > >               printerr(0, "WARNING: unable to drop supplimentary > > > groups!"); > > >               return errno; > > >       } > > > @@ -457,7 +458,12 @@ change_identity(uid_t uid) > > >        * Switch the GIDs. Note that we leave the saved-set-gid > > > alone in an > > >        * attempt to prevent attacks via ptrace() > > >        */ > > > -     if (setresgid(pw->pw_gid, pw->pw_gid, -1) != 0) { > > > +     /* For the threaded version we have to set uid,gid per > > > thread instead > > > +      * of per process. glibc setresuid() when called from a > > > thread, it'll > > > +      * send a signal to all other threads to synchronize the > > > uid in all > > > +      * other threads. To bypass this, we have to call syscall() > > > directly. > > > +      */ > > > +     if (syscall(SYS_setresgid, pw->pw_gid, -1, -1) != 0) { > > >               printerr(0, "WARNING: failed to set gid to %u!\n", > > > pw->pw_gid); > > >               return errno; > > >       } > > > @@ -466,7 +472,7 @@ change_identity(uid_t uid) > > >        * Switch UIDs, but leave saved-set-uid alone to prevent > > > ptrace() by > > >        * other processes running with this uid. > > >        */ > > > -     if (setresuid(uid, uid, -1) != 0) { > > > +     if (syscall(SYS_setresuid, uid, -1, -1) != 0) { > > >               printerr(0, "WARNING: Failed to setuid for user > > > with uid %u\n", > > >                               uid); > > >               return errno; > > > > Note that the old code set both real and effective uid/gid. Here > > you're > > changing it to just set the real uid. Is that change deliberate? I > > think you probably want to set the euid/egid as well. > > It was a typo (cthon tests ran fine though which is confusing since > kerberos code does geteuid()). Now I'm thinking that given that we > don't touch saved uid, perhaps the code should call SYS_setreuid > instead of SYS_setresuid? > Yes, good point. Either that or just set all 3 to "uid", I'd think. -- Jeff Layton