| Message ID | 20190626192234.11725-24-casey@schaufler-ca.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | LSM: Module stacking for AppArmor | expand |
On Wed, Jun 26, 2019 at 12:22:34PM -0700, Casey Schaufler wrote: > With the inclusion of the "display" process attribute > mechanism AppArmor no longer needs to be treated as an > "exclusive" security module. Remove the flag that indicates > it is exclusive. Remove the stub getpeersec_dgram AppArmor > hook as it has no effect in the single LSM case and > interferes in the multiple LSM case. > > Signed-off-by: Casey Schaufler <casey@schaufler-ca.com> Reviewed-by: Kees Cook <keescook@chromium.org> -Kees > --- > security/apparmor/lsm.c | 20 +------------------- > 1 file changed, 1 insertion(+), 19 deletions(-) > > diff --git a/security/apparmor/lsm.c b/security/apparmor/lsm.c > index 6d2eefc9b7c1..fb5798683ae1 100644 > --- a/security/apparmor/lsm.c > +++ b/security/apparmor/lsm.c > @@ -1079,22 +1079,6 @@ static int apparmor_socket_getpeersec_stream(struct socket *sock, > return error; > } > > -/** > - * apparmor_socket_getpeersec_dgram - get security label of packet > - * @sock: the peer socket > - * @skb: packet data > - * @secid: pointer to where to put the secid of the packet > - * > - * Sets the netlabel socket state on sk from parent > - */ > -static int apparmor_socket_getpeersec_dgram(struct socket *sock, > - struct sk_buff *skb, u32 *secid) > - > -{ > - /* TODO: requires secid support */ > - return -ENOPROTOOPT; > -} > - > /** > * apparmor_sock_graft - Initialize newly created socket > * @sk: child sock > @@ -1195,8 +1179,6 @@ static struct security_hook_list apparmor_hooks[] __lsm_ro_after_init = { > #endif > LSM_HOOK_INIT(socket_getpeersec_stream, > apparmor_socket_getpeersec_stream), > - LSM_HOOK_INIT(socket_getpeersec_dgram, > - apparmor_socket_getpeersec_dgram), > LSM_HOOK_INIT(sock_graft, apparmor_sock_graft), > #ifdef CONFIG_NETWORK_SECMARK > LSM_HOOK_INIT(inet_conn_request, apparmor_inet_conn_request), > @@ -1707,7 +1689,7 @@ static int __init apparmor_init(void) > > DEFINE_LSM(apparmor) = { > .name = "apparmor", > - .flags = LSM_FLAG_LEGACY_MAJOR | LSM_FLAG_EXCLUSIVE, > + .flags = LSM_FLAG_LEGACY_MAJOR, > .enabled = &apparmor_enabled, > .blobs = &apparmor_blob_sizes, > .init = apparmor_init, > -- > 2.20.1 >
On Wed, 26 Jun 2019, Casey Schaufler wrote: > With the inclusion of the "display" process attribute > mechanism AppArmor no longer needs to be treated as an > "exclusive" security module. Remove the flag that indicates > it is exclusive. Remove the stub getpeersec_dgram AppArmor > hook as it has no effect in the single LSM case and > interferes in the multiple LSM case. So now if I build a kernel with SELinux and AppArmor selected, with SELinux registered first, I now need to use apparmor=0 at the kernel command line to preserve existing behavior (just SELinux running). This should at least be documented. I wonder if this will break existing users, though. Who has both currently selected and depends on only one of them being active?
On Thu, Jun 27, 2019 at 12:22:13PM +1000, James Morris wrote: > On Wed, 26 Jun 2019, Casey Schaufler wrote: > > > With the inclusion of the "display" process attribute > > mechanism AppArmor no longer needs to be treated as an > > "exclusive" security module. Remove the flag that indicates > > it is exclusive. Remove the stub getpeersec_dgram AppArmor > > hook as it has no effect in the single LSM case and > > interferes in the multiple LSM case. > > So now if I build a kernel with SELinux and AppArmor selected, with > SELinux registered first, I now need to use apparmor=0 at the kernel > command line to preserve existing behavior (just SELinux running). > > This should at least be documented. > > I wonder if this will break existing users, though. Who has both > currently selected and depends on only one of them being active? I don't think this will change a system using SELinux, right? There would be no policy loaded for AppArmor so its hooks would be no-op. But maybe I'm not thinking hard enough?
On 6/26/19 7:22 PM, James Morris wrote: > On Wed, 26 Jun 2019, Casey Schaufler wrote: > >> With the inclusion of the "display" process attribute >> mechanism AppArmor no longer needs to be treated as an >> "exclusive" security module. Remove the flag that indicates >> it is exclusive. Remove the stub getpeersec_dgram AppArmor >> hook as it has no effect in the single LSM case and >> interferes in the multiple LSM case. > > So now if I build a kernel with SELinux and AppArmor selected, with > SELinux registered first, I now need to use apparmor=0 at the kernel > command line to preserve existing behavior (just SELinux running). > AppArmor can be built-in (compiled) without being on the Enabled list. If you had apparmor in your enabled list along with selinux before, it would attempt to enabled and fail with the message exclusive disabled: apparmor now it will be enabled but it does match what is documented in the lsm enabled description A comma-separated list of LSMs, in initialization order. Any LSMs left off this list will be ignored. This can be controlled at boot with the "lsm=" parameter. what isn't documented is the exclusive behavior and it does make sense to have which LSMs are exclusive documented somewhere. > This should at least be documented. > > I wonder if this will break existing users, though. Who has both > currently selected and depends on only one of them being active? > thankfully even if you had apparmor in your lsm enabled list before, this likely isn't enough to change the system behavior. You will also need some apparmor policy installed and enough of the usersoace to load it. Otherwise while apparmor will be enabled it will come up in unconfined mode. Even the interfaces /proc/*/attr/* will default to the first major LSM in the list (in your example selinux). Apparmor's labeling won't be visible without a task explicitly opting in to it.
On Wed, 26 Jun 2019, John Johansen wrote: > AppArmor can be built-in (compiled) without being on the Enabled list. > If you had apparmor in your enabled list along with selinux before, > it would attempt to enabled and fail with the message > > exclusive disabled: apparmor > > now it will be enabled but it does match what is documented in > the lsm enabled description > > A comma-separated list of LSMs, in initialization order. > Any LSMs left off this list will be ignored. This can be > controlled at boot with the "lsm=" parameter. Ok -- I suspect the only people who have SELinux and AppArmor selected are doing testing / development.
On 6/26/19 12:22 PM, Casey Schaufler wrote: > With the inclusion of the "display" process attribute > mechanism AppArmor no longer needs to be treated as an > "exclusive" security module. Remove the flag that indicates > it is exclusive. Remove the stub getpeersec_dgram AppArmor > hook as it has no effect in the single LSM case and > interferes in the multiple LSM case. > > Signed-off-by: Casey Schaufler <casey@schaufler-ca.com> Reviewed-by: John Johansen <john.johansen@canonical.com> > --- > security/apparmor/lsm.c | 20 +------------------- > 1 file changed, 1 insertion(+), 19 deletions(-) > > diff --git a/security/apparmor/lsm.c b/security/apparmor/lsm.c > index 6d2eefc9b7c1..fb5798683ae1 100644 > --- a/security/apparmor/lsm.c > +++ b/security/apparmor/lsm.c > @@ -1079,22 +1079,6 @@ static int apparmor_socket_getpeersec_stream(struct socket *sock, > return error; > } > > -/** > - * apparmor_socket_getpeersec_dgram - get security label of packet > - * @sock: the peer socket > - * @skb: packet data > - * @secid: pointer to where to put the secid of the packet > - * > - * Sets the netlabel socket state on sk from parent > - */ > -static int apparmor_socket_getpeersec_dgram(struct socket *sock, > - struct sk_buff *skb, u32 *secid) > - > -{ > - /* TODO: requires secid support */ > - return -ENOPROTOOPT; > -} > - > /** > * apparmor_sock_graft - Initialize newly created socket > * @sk: child sock > @@ -1195,8 +1179,6 @@ static struct security_hook_list apparmor_hooks[] __lsm_ro_after_init = { > #endif > LSM_HOOK_INIT(socket_getpeersec_stream, > apparmor_socket_getpeersec_stream), > - LSM_HOOK_INIT(socket_getpeersec_dgram, > - apparmor_socket_getpeersec_dgram), > LSM_HOOK_INIT(sock_graft, apparmor_sock_graft), > #ifdef CONFIG_NETWORK_SECMARK > LSM_HOOK_INIT(inet_conn_request, apparmor_inet_conn_request), > @@ -1707,7 +1689,7 @@ static int __init apparmor_init(void) > > DEFINE_LSM(apparmor) = { > .name = "apparmor", > - .flags = LSM_FLAG_LEGACY_MAJOR | LSM_FLAG_EXCLUSIVE, > + .flags = LSM_FLAG_LEGACY_MAJOR, > .enabled = &apparmor_enabled, > .blobs = &apparmor_blob_sizes, > .init = apparmor_init, >
diff --git a/security/apparmor/lsm.c b/security/apparmor/lsm.c index 6d2eefc9b7c1..fb5798683ae1 100644 --- a/security/apparmor/lsm.c +++ b/security/apparmor/lsm.c @@ -1079,22 +1079,6 @@ static int apparmor_socket_getpeersec_stream(struct socket *sock, return error; } -/** - * apparmor_socket_getpeersec_dgram - get security label of packet - * @sock: the peer socket - * @skb: packet data - * @secid: pointer to where to put the secid of the packet - * - * Sets the netlabel socket state on sk from parent - */ -static int apparmor_socket_getpeersec_dgram(struct socket *sock, - struct sk_buff *skb, u32 *secid) - -{ - /* TODO: requires secid support */ - return -ENOPROTOOPT; -} - /** * apparmor_sock_graft - Initialize newly created socket * @sk: child sock @@ -1195,8 +1179,6 @@ static struct security_hook_list apparmor_hooks[] __lsm_ro_after_init = { #endif LSM_HOOK_INIT(socket_getpeersec_stream, apparmor_socket_getpeersec_stream), - LSM_HOOK_INIT(socket_getpeersec_dgram, - apparmor_socket_getpeersec_dgram), LSM_HOOK_INIT(sock_graft, apparmor_sock_graft), #ifdef CONFIG_NETWORK_SECMARK LSM_HOOK_INIT(inet_conn_request, apparmor_inet_conn_request), @@ -1707,7 +1689,7 @@ static int __init apparmor_init(void) DEFINE_LSM(apparmor) = { .name = "apparmor", - .flags = LSM_FLAG_LEGACY_MAJOR | LSM_FLAG_EXCLUSIVE, + .flags = LSM_FLAG_LEGACY_MAJOR, .enabled = &apparmor_enabled, .blobs = &apparmor_blob_sizes, .init = apparmor_init,
With the inclusion of the "display" process attribute mechanism AppArmor no longer needs to be treated as an "exclusive" security module. Remove the flag that indicates it is exclusive. Remove the stub getpeersec_dgram AppArmor hook as it has no effect in the single LSM case and interferes in the multiple LSM case. Signed-off-by: Casey Schaufler <casey@schaufler-ca.com> --- security/apparmor/lsm.c | 20 +------------------- 1 file changed, 1 insertion(+), 19 deletions(-)