Re: [PATCH] signal: fix overflow_uid signal sender
On 11/02, Jann Horn wrote: > > On Wed, Nov 02, 2016 at 07:16:41PM +0100, Oleg Nesterov wrote: > > On 10/31, Jann Horn wrote: > > > > > > static inline void userns_fixup_signal_uid(struct siginfo *info, struct > > > task_struct *t) > > > { > > > - if (current_user_ns() == task_cred_xxx(t, user_ns)) > > > + if (&init_user_ns == task_cred_xxx(t, user_ns)) > > > return; > > > > > > if (SI_FROMKERNEL(info)) > > > @@ -959,7 +959,7 @@ static inline void userns_fixup_signal_uid(struct > > > siginfo *info, struct task_str > > > > > > rcu_read_lock(); > > > info->si_uid = from_kuid_munged(task_cred_xxx(t, user_ns), > > > - make_kuid(current_user_ns(), > > > info->si_uid)); > > > + make_kuid(&init_user_ns, info->si_uid)); > > > rcu_read_unlock(); > > > } > > > #else > > > @@ -1027,7 +1027,8 @@ static int __send_signal(int sig, struct siginfo > > > *info, struct task_struct *t, > > > q->info.si_code = SI_USER; > > > q->info.si_pid = task_tgid_nr_ns(current, > > > task_active_pid_ns(t)); > > > - q->info.si_uid = from_kuid_munged(current_user_ns(), > > > current_uid()); > > > + q->info.si_uid = from_kuid(&init_user_ns, > > > +current_uid()); > > > > Looks good to me at first glance, but I think this needs an ack from Eric. > > > > I have to admit that I forgot how uid_map/etc actually works, I can't > > even recall if from_kuid(init_user_ns, xxx) == __kuid_val(xxx) or not, > > although this doesn't really matter. > > Yes, from_kuid(&init_user_ns, xxx) == __kuid_val(xxx). For values from 0 > to 0xfffe, the uid_map of init_user_ns is an identity map (and it > can't be changed), and for 0x, which isn't mapped, it returns > 0x to denote failure. So perhaps we can add another helper to avoid the unnecessary map_id_up() from_kuid_init_user_ns(kuid_t kuid) { return __kuid_val(kuid); } But this is a bit off-topic, let me repeat that your patch looks fine to me. Oleg.
Re: [PATCH] signal: fix overflow_uid signal sender
On Wed, Nov 02, 2016 at 07:16:41PM +0100, Oleg Nesterov wrote: > On 10/31, Jann Horn wrote: > > > > static inline void userns_fixup_signal_uid(struct siginfo *info, struct > > task_struct *t) > > { > > - if (current_user_ns() == task_cred_xxx(t, user_ns)) > > + if (&init_user_ns == task_cred_xxx(t, user_ns)) > > return; > > > > if (SI_FROMKERNEL(info)) > > @@ -959,7 +959,7 @@ static inline void userns_fixup_signal_uid(struct > > siginfo *info, struct task_str > > > > rcu_read_lock(); > > info->si_uid = from_kuid_munged(task_cred_xxx(t, user_ns), > > - make_kuid(current_user_ns(), > > info->si_uid)); > > + make_kuid(&init_user_ns, info->si_uid)); > > rcu_read_unlock(); > > } > > #else > > @@ -1027,7 +1027,8 @@ static int __send_signal(int sig, struct siginfo > > *info, struct task_struct *t, > > q->info.si_code = SI_USER; > > q->info.si_pid = task_tgid_nr_ns(current, > > task_active_pid_ns(t)); > > - q->info.si_uid = from_kuid_munged(current_user_ns(), > > current_uid()); > > + q->info.si_uid = from_kuid(&init_user_ns, > > + current_uid()); > > Looks good to me at first glance, but I think this needs an ack from Eric. > > I have to admit that I forgot how uid_map/etc actually works, I can't > even recall if from_kuid(init_user_ns, xxx) == __kuid_val(xxx) or not, > although this doesn't really matter. Yes, from_kuid(&init_user_ns, xxx) == __kuid_val(xxx). For values from 0 to 0xfffe, the uid_map of init_user_ns is an identity map (and it can't be changed), and for 0x, which isn't mapped, it returns 0x to denote failure. signature.asc Description: Digital signature
Re: [PATCH] signal: fix overflow_uid signal sender
On 10/31, Jann Horn wrote: > > static inline void userns_fixup_signal_uid(struct siginfo *info, struct > task_struct *t) > { > - if (current_user_ns() == task_cred_xxx(t, user_ns)) > + if (&init_user_ns == task_cred_xxx(t, user_ns)) > return; > > if (SI_FROMKERNEL(info)) > @@ -959,7 +959,7 @@ static inline void userns_fixup_signal_uid(struct siginfo > *info, struct task_str > > rcu_read_lock(); > info->si_uid = from_kuid_munged(task_cred_xxx(t, user_ns), > - make_kuid(current_user_ns(), > info->si_uid)); > + make_kuid(&init_user_ns, info->si_uid)); > rcu_read_unlock(); > } > #else > @@ -1027,7 +1027,8 @@ static int __send_signal(int sig, struct siginfo *info, > struct task_struct *t, > q->info.si_code = SI_USER; > q->info.si_pid = task_tgid_nr_ns(current, > task_active_pid_ns(t)); > - q->info.si_uid = from_kuid_munged(current_user_ns(), > current_uid()); > + q->info.si_uid = from_kuid(&init_user_ns, > +current_uid()); Looks good to me at first glance, but I think this needs an ack from Eric. I have to admit that I forgot how uid_map/etc actually works, I can't even recall if from_kuid(init_user_ns, xxx) == __kuid_val(xxx) or not, although this doesn't really matter. Oleg.