Currently, __HYPERVISOR_xsm_op enters xsm_ops.do_{xsm,compat}_op() which means
that if any other XSM module registers a handler, we'll break the hypercall
ABI totally by having the behaviour depend on the selected XSM module.

There are 2 options:
 1) Rename __HYPERVISOR_xsm_op to __HYPERVISOR_flask_op.  If another XSM
    module wants hypercalls, they'll have to introduce a new top-level
    hypercall.
 2) Make the current flask_op() be common, and require new hypercalls to fit
    compatibly with xen_flask_op_t.  This can be done fairly easily by
    subdividing the .cmd space.

In this patch, we implement option 2.

Move the stub {do,compat}_xsm_op() implementation into a new xsm_op.c so we
can more easily do multi-inclusion compat magic.  Also add a new private.h,
because flask/hook.c's local declaration of {do,compat}_flask_op() wasn't
remotely safe.

The top level now dispatches into {do,compat}_flask_op() based on op.cmd, and
handles the primary copy in/out.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
---
CC: Daniel De Graaf <dgdegra@tycho.nsa.gov>
CC: Daniel Smith <dpsmith@apertussolutions.com>

Only lightly tested.  Slightly RFC.  There are several things which aren't
great, but probably want addressing in due course.

 1) The public headers probably want to lose the flask name (in a compatible
    way), now that the hypercall is common.  This probably wants to be
    combined with no longer taking a void handle.
 2) {do,compat}_xsm_op() are currently identical other than the dispatched-to
    functions because the size of xsm_flask_op_t is invariant with
    COMPAT-ness.  We could simplfy things by only having one, and dispatching
    to {do,compat}_*_op() directly, but I'm not sure whether the complexity is
    worth it.
 3) Bloat-o-meter says these changes add 16 extra bytes to dm_op() and I can't
    figure out what could possibly be causing this.
---
 xen/include/Makefile     |  2 +-
 xen/include/xsm/dummy.h  | 12 ----------
 xen/include/xsm/xsm.h    | 17 --------------
 xen/xsm/Makefile         |  1 +
 xen/xsm/dummy.c          |  5 ----
 xen/xsm/flask/flask_op.c | 60 ++++++++++++++++++++++--------------------------
 xen/xsm/flask/hooks.c    |  8 -------
 xen/xsm/private.h        | 16 +++++++++++++
 xen/xsm/xsm_core.c       | 12 ----------
 xen/xsm/xsm_op.c         | 51 ++++++++++++++++++++++++++++++++++++++++
 10 files changed, 96 insertions(+), 88 deletions(-)
 create mode 100644 xen/xsm/private.h
 create mode 100644 xen/xsm/xsm_op.c

diff --git a/xen/include/Makefile b/xen/include/Makefile
index 95daa8a28975..9547f848635c 100644
--- a/xen/include/Makefile
+++ b/xen/include/Makefile
@@ -32,7 +32,7 @@ headers-$(CONFIG_HYPFS)   += compat/hypfs.h
 headers-$(CONFIG_KEXEC)   += compat/kexec.h
 headers-$(CONFIG_TRACEBUFFER) += compat/trace.h
 headers-$(CONFIG_XENOPROF) += compat/xenoprof.h
-headers-$(CONFIG_XSM_FLASK) += compat/xsm/flask_op.h
+headers-$(CONFIG_XSM)      += compat/xsm/flask_op.h
 
 cppflags-y                := -include public/xen-compat.h -DXEN_GENERATING_COMPAT_HEADERS
 cppflags-$(CONFIG_X86)    += -m32
diff --git a/xen/include/xsm/dummy.h b/xen/include/xsm/dummy.h
index b024119896e6..5448e9970bc1 100644
--- a/xen/include/xsm/dummy.h
+++ b/xen/include/xsm/dummy.h
@@ -466,18 +466,6 @@ static XSM_INLINE int xsm_hypfs_op(XSM_DEFAULT_VOID)
     return xsm_default_action(action, current->domain, NULL);
 }
 
-static XSM_INLINE long xsm_do_xsm_op(XEN_GUEST_HANDLE_PARAM(void) op)
-{
-    return -ENOSYS;
-}
-
-#ifdef CONFIG_COMPAT
-static XSM_INLINE int xsm_do_compat_op(XEN_GUEST_HANDLE_PARAM(void) op)
-{
-    return -ENOSYS;
-}
-#endif
-
 static XSM_INLINE char *xsm_show_irq_sid(int irq)
 {
     return NULL;
diff --git a/xen/include/xsm/xsm.h b/xen/include/xsm/xsm.h
index 5aa4dd588d17..4856b589dd86 100644
--- a/xen/include/xsm/xsm.h
+++ b/xen/include/xsm/xsm.h
@@ -141,11 +141,6 @@ struct xsm_ops {
     int (*page_offline)(uint32_t cmd);
     int (*hypfs_op)(void);
 
-    long (*do_xsm_op)(XEN_GUEST_HANDLE_PARAM(void) op);
-#ifdef CONFIG_COMPAT
-    int (*do_compat_op)(XEN_GUEST_HANDLE_PARAM(void) op);
-#endif
-
     int (*hvm_param)(struct domain *d, unsigned long op);
     int (*hvm_param_altp2mhvm)(struct domain *d);
     int (*hvm_altp2mhvm_op)(struct domain *d, uint64_t mode, uint32_t op);
@@ -585,18 +580,6 @@ static inline int xsm_hypfs_op(xsm_default_t def)
     return alternative_call(xsm_ops.hypfs_op);
 }
 
-static inline long xsm_do_xsm_op(XEN_GUEST_HANDLE_PARAM(void) op)
-{
-    return alternative_call(xsm_ops.do_xsm_op, op);
-}
-
-#ifdef CONFIG_COMPAT
-static inline int xsm_do_compat_op(XEN_GUEST_HANDLE_PARAM(void) op)
-{
-    return alternative_call(xsm_ops.do_compat_op, op);
-}
-#endif
-
 static inline int xsm_hvm_param(
     xsm_default_t def, struct domain *d, unsigned long op)
 {
diff --git a/xen/xsm/Makefile b/xen/xsm/Makefile
index cf0a728f1c96..93ba20256f6e 100644
--- a/xen/xsm/Makefile
+++ b/xen/xsm/Makefile
@@ -1,4 +1,5 @@
 obj-y += xsm_core.o
+obj-y += xsm_op.o
 obj-$(CONFIG_XSM) += xsm_policy.o
 obj-$(CONFIG_XSM) += dummy.o
 obj-$(CONFIG_XSM_SILO) += silo.o
diff --git a/xen/xsm/dummy.c b/xen/xsm/dummy.c
index c111fa05968d..4fb7b0d04f7b 100644
--- a/xen/xsm/dummy.c
+++ b/xen/xsm/dummy.c
@@ -99,11 +99,6 @@ static const struct xsm_ops __initconstrel dummy_ops = {
     .hvm_param_altp2mhvm           = xsm_hvm_param_altp2mhvm,
     .hvm_altp2mhvm_op              = xsm_hvm_altp2mhvm_op,
 
-    .do_xsm_op                     = xsm_do_xsm_op,
-#ifdef CONFIG_COMPAT
-    .do_compat_op                  = xsm_do_compat_op,
-#endif
-
     .add_to_physmap                = xsm_add_to_physmap,
     .remove_from_physmap           = xsm_remove_from_physmap,
     .map_gmfn_foreign              = xsm_map_gmfn_foreign,
diff --git a/xen/xsm/flask/flask_op.c b/xen/xsm/flask/flask_op.c
index 221ff00fd3cc..e25664711019 100644
--- a/xen/xsm/flask/flask_op.c
+++ b/xen/xsm/flask/flask_op.c
@@ -22,6 +22,8 @@
 #include <objsec.h>
 #include <conditional.h>
 
+#include "../private.h"
+
 #define ret_t long
 #define _copy_to_guest copy_to_guest
 #define _copy_from_guest copy_from_guest
@@ -607,21 +609,17 @@ static int flask_relabel_domain(struct xen_flask_relabel *arg)
 
 #endif /* !COMPAT */
 
-ret_t do_flask_op(XEN_GUEST_HANDLE_PARAM(void) u_flask_op)
+ret_t do_flask_op(xen_flask_op_t *op, bool *copyback)
 {
-    xen_flask_op_t op;
     int rv;
 
-    if ( copy_from_guest(&op, u_flask_op, 1) )
-        return -EFAULT;
-
-    if ( op.interface_version != XEN_FLASK_INTERFACE_VERSION )
+    if ( op->interface_version != XEN_FLASK_INTERFACE_VERSION )
         return -ENOSYS;
 
-    switch ( op.cmd )
+    switch ( op->cmd )
     {
     case FLASK_LOAD:
-        rv = flask_security_load(&op.u.load);
+        rv = flask_security_load(&op->u.load);
         break;
 
     case FLASK_GETENFORCE:
@@ -629,27 +627,27 @@ ret_t do_flask_op(XEN_GUEST_HANDLE_PARAM(void) u_flask_op)
         break;
 
     case FLASK_SETENFORCE:
-        rv = flask_security_setenforce(&op.u.enforce);
+        rv = flask_security_setenforce(&op->u.enforce);
         break;
 
     case FLASK_CONTEXT_TO_SID:
-        rv = flask_security_context(&op.u.sid_context);
+        rv = flask_security_context(&op->u.sid_context);
         break;
 
     case FLASK_SID_TO_CONTEXT:
-        rv = flask_security_sid(&op.u.sid_context);
+        rv = flask_security_sid(&op->u.sid_context);
         break;
 
     case FLASK_ACCESS:
-        rv = flask_security_access(&op.u.access);
+        rv = flask_security_access(&op->u.access);
         break;
 
     case FLASK_CREATE:
-        rv = flask_security_create(&op.u.transition);
+        rv = flask_security_create(&op->u.transition);
         break;
 
     case FLASK_RELABEL:
-        rv = flask_security_relabel(&op.u.transition);
+        rv = flask_security_relabel(&op->u.transition);
         break;
 
     case FLASK_POLICYVERS:
@@ -657,11 +655,11 @@ ret_t do_flask_op(XEN_GUEST_HANDLE_PARAM(void) u_flask_op)
         break;
 
     case FLASK_GETBOOL:
-        rv = flask_security_get_bool(&op.u.boolean);
+        rv = flask_security_get_bool(&op->u.boolean);
         break;
 
     case FLASK_SETBOOL:
-        rv = flask_security_set_bool(&op.u.boolean);
+        rv = flask_security_set_bool(&op->u.boolean);
         break;
 
     case FLASK_COMMITBOOLS:
@@ -677,41 +675,41 @@ ret_t do_flask_op(XEN_GUEST_HANDLE_PARAM(void) u_flask_op)
         break;
 
     case FLASK_SETAVC_THRESHOLD:
-        rv = flask_security_setavc_threshold(&op.u.setavc_threshold);
+        rv = flask_security_setavc_threshold(&op->u.setavc_threshold);
         break;
 
     case FLASK_AVC_HASHSTATS:
-        rv = avc_get_hash_stats(&op.u.hash_stats);
+        rv = avc_get_hash_stats(&op->u.hash_stats);
         break;
 
 #ifdef CONFIG_XSM_FLASK_AVC_STATS
     case FLASK_AVC_CACHESTATS:
-        rv = flask_security_avc_cachestats(&op.u.cache_stats);
+        rv = flask_security_avc_cachestats(&op->u.cache_stats);
         break;
 #endif
 
     case FLASK_MEMBER:
-        rv = flask_security_member(&op.u.transition);
+        rv = flask_security_member(&op->u.transition);
         break;
 
     case FLASK_ADD_OCONTEXT:
-        rv = flask_ocontext_add(&op.u.ocontext);
+        rv = flask_ocontext_add(&op->u.ocontext);
         break;
 
     case FLASK_DEL_OCONTEXT:
-        rv = flask_ocontext_del(&op.u.ocontext);
+        rv = flask_ocontext_del(&op->u.ocontext);
         break;
 
     case FLASK_GET_PEER_SID:
-        rv = flask_get_peer_sid(&op.u.peersid);
+        rv = flask_get_peer_sid(&op->u.peersid);
         break;
 
     case FLASK_RELABEL_DOMAIN:
-        rv = flask_relabel_domain(&op.u.relabel);
+        rv = flask_relabel_domain(&op->u.relabel);
         break;
 
     case FLASK_DEVICETREE_LABEL:
-        rv = flask_devicetree_label(&op.u.devicetree_label);
+        rv = flask_devicetree_label(&op->u.devicetree_label);
         break;
 
     default:
@@ -719,16 +717,12 @@ ret_t do_flask_op(XEN_GUEST_HANDLE_PARAM(void) u_flask_op)
     }
 
     if ( rv < 0 )
-        goto out;
+        return rv;
 
-    if ( (FLASK_COPY_OUT&(1UL<<op.cmd)) )
-    {
-        if ( copy_to_guest(u_flask_op, &op, 1) )
-            rv = -EFAULT;
-    }
+    if ( (1ul << op->cmd) & FLASK_COPY_OUT )
+        *copyback = true;
 
- out:
-    return rv;
+    return 0;
 }
 
 #if defined(CONFIG_COMPAT) && !defined(COMPAT)
diff --git a/xen/xsm/flask/hooks.c b/xen/xsm/flask/hooks.c
index 3b29f7fde372..8f4fc7458b8f 100644
--- a/xen/xsm/flask/hooks.c
+++ b/xen/xsm/flask/hooks.c
@@ -1742,9 +1742,6 @@ static int flask_argo_send(const struct domain *d, const struct domain *t)
 
 #endif
 
-long do_flask_op(XEN_GUEST_HANDLE_PARAM(void) u_flask_op);
-int compat_flask_op(XEN_GUEST_HANDLE_PARAM(void) u_flask_op);
-
 static const struct xsm_ops __initconstrel flask_ops = {
     .security_domaininfo = flask_security_domaininfo,
     .domain_create = flask_domain_create,
@@ -1819,7 +1816,6 @@ static const struct xsm_ops __initconstrel flask_ops = {
     .hvm_param_altp2mhvm = flask_hvm_param_altp2mhvm,
     .hvm_altp2mhvm_op = flask_hvm_altp2mhvm_op,
 
-    .do_xsm_op = do_flask_op,
     .get_vnumainfo = flask_get_vnumainfo,
 
     .vm_event_control = flask_vm_event_control,
@@ -1836,10 +1832,6 @@ static const struct xsm_ops __initconstrel flask_ops = {
     .mem_sharing = flask_mem_sharing,
 #endif
 
-#ifdef CONFIG_COMPAT
-    .do_compat_op = compat_flask_op,
-#endif
-
     .add_to_physmap = flask_add_to_physmap,
     .remove_from_physmap = flask_remove_from_physmap,
     .map_gmfn_foreign = flask_map_gmfn_foreign,
diff --git a/xen/xsm/private.h b/xen/xsm/private.h
new file mode 100644
index 000000000000..19ade2eed6b7
--- /dev/null
+++ b/xen/xsm/private.h
@@ -0,0 +1,16 @@
+#ifndef XSM_PRIVATE_H
+#define XSM_PRIVATE_H
+
+#include <public/xsm/flask_op.h>
+
+long do_flask_op(xen_flask_op_t *op, bool *copyback);
+
+#ifdef CONFIG_COMPAT
+
+#include <compat/xsm/flask_op.h>
+
+int compat_flask_op(compat_flask_op_t *op, bool *copyback);
+
+#endif /* CONFIG_COMPAT */
+
+#endif /* XSM_PRIVATE_H */
diff --git a/xen/xsm/xsm_core.c b/xen/xsm/xsm_core.c
index 21fffbcb41d3..819a6ccd54cc 100644
--- a/xen/xsm/xsm_core.c
+++ b/xen/xsm/xsm_core.c
@@ -218,15 +218,3 @@ bool __init has_xsm_magic(paddr_t start)
 #endif
 
 #endif
-
-long do_xsm_op(XEN_GUEST_HANDLE_PARAM(void) op)
-{
-    return xsm_do_xsm_op(op);
-}
-
-#ifdef CONFIG_COMPAT
-int compat_xsm_op(XEN_GUEST_HANDLE_PARAM(void) op)
-{
-    return xsm_do_compat_op(op);
-}
-#endif
diff --git a/xen/xsm/xsm_op.c b/xen/xsm/xsm_op.c
new file mode 100644
index 000000000000..5922299fc4dc
--- /dev/null
+++ b/xen/xsm/xsm_op.c
@@ -0,0 +1,51 @@
+#ifndef COMPAT
+
+#include <xen/guest_access.h>
+
+#include "private.h"
+
+#define ret_t long
+#define _copy_to_guest copy_to_guest
+#define _copy_from_guest copy_from_guest
+
+#endif /* COMPAT */
+
+ret_t do_xsm_op(XEN_GUEST_HANDLE_PARAM(void) u_flask_op)
+{
+    xen_flask_op_t op;
+    bool copyback = false;
+    ret_t rc = -ENOSYS;
+
+    if ( copy_from_guest(&op, u_flask_op, 1) )
+        return -EFAULT;
+
+    switch ( op.cmd )
+    {
+    case FLASK_LOAD ... FLASK_DEVICETREE_LABEL:
+        if ( IS_ENABLED(CONFIG_XSM_FLASK) )
+            rc = do_flask_op(&op, &copyback);
+        break;
+    }
+
+    if ( copyback && copy_to_guest(u_flask_op, &op, 1) )
+        rc = -EFAULT;
+
+    return rc;
+}
+
+#if defined(CONFIG_COMPAT) && !defined(COMPAT)
+#define COMPAT
+
+#undef _copy_to_guest
+#define _copy_to_guest copy_to_compat
+#undef _copy_from_guest
+#define _copy_from_guest copy_from_compat
+
+#define xen_flask_op_t compat_flask_op_t
+#undef ret_t
+#define ret_t int
+#define do_flask_op compat_flask_op
+#define do_xsm_op compat_xsm_op
+
+#include "xsm_op.c"
+#endif
-- 
2.11.0

On 05.11.2021 14:55, Andrew Cooper wrote:
> Currently, __HYPERVISOR_xsm_op enters xsm_ops.do_{xsm,compat}_op() which means
> that if any other XSM module registers a handler, we'll break the hypercall
> ABI totally by having the behaviour depend on the selected XSM module.
> 
> There are 2 options:
>  1) Rename __HYPERVISOR_xsm_op to __HYPERVISOR_flask_op.  If another XSM
>     module wants hypercalls, they'll have to introduce a new top-level
>     hypercall.
>  2) Make the current flask_op() be common, and require new hypercalls to fit
>     compatibly with xen_flask_op_t.  This can be done fairly easily by
>     subdividing the .cmd space.
> 
> In this patch, we implement option 2.
> 
> Move the stub {do,compat}_xsm_op() implementation into a new xsm_op.c so we
> can more easily do multi-inclusion compat magic.  Also add a new private.h,
> because flask/hook.c's local declaration of {do,compat}_flask_op() wasn't
> remotely safe.
> 
> The top level now dispatches into {do,compat}_flask_op() based on op.cmd, and
> handles the primary copy in/out.

I'm not convinced this is the only reasonable way of implementing 2).
I could also see number space to be separated in different ways, ...

> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
> ---
> CC: Daniel De Graaf <dgdegra@tycho.nsa.gov>
> CC: Daniel Smith <dpsmith@apertussolutions.com>
> 
> Only lightly tested.  Slightly RFC.  There are several things which aren't
> great, but probably want addressing in due course.
> 
>  1) The public headers probably want to lose the flask name (in a compatible
>     way), now that the hypercall is common.  This probably wants to be
>     combined with no longer taking a void handle.

... leaving per-module public headers and perhaps merely adding an
abstracting

struct xen_xsm_op_t {
    uint32_t op;
    ... /* placeholder */
};

or (making explicit one possible variant of number space splitting)

union xen_xsm_op_t {
    uint32_t op;
    struct {
        uint16_t cmd;
        uint16_t mod;
    } u;
    ... /* placeholder */
};

in, say, a new public/xsm.h.

As a result I consider this change either going too far (because of
not knowing future needs) or not far enough (by e.g. leaving
do_xsm_op() to use xen_flask_op_t.

>  2) {do,compat}_xsm_op() are currently identical other than the dispatched-to
>     functions because the size of xsm_flask_op_t is invariant with
>     COMPAT-ness.  We could simplfy things by only having one, and dispatching
>     to {do,compat}_*_op() directly, but I'm not sure whether the complexity is
>     worth it.

Perhaps not, I would say, not the least because (as said above) I
think we shouldn't put in place restrictions which may get in the
way of adding some future module.

Extending struct xen_flask_op to become a generic XSM interface
structure (or even just for Flask's own purposes) also is not as
straightforward as it might seem: There's currently no provision
for sub-structs which would grow the overall size of the structure:
The copy_{to,from}_guest() invocations for existing sub-ops may not
copy more that the present worth of sizeof(struct xen_flask_op).
Yet your implementation of do_xsm_op() moves this deficiency from
Flask to XSM.

>  3) Bloat-o-meter says these changes add 16 extra bytes to dm_op() and I can't
>     figure out what could possibly be causing this.

Without comparing the object files in closer detail it's guesswork,
but might this be register scheduling differences resulting from
the changed sizeof(struct xsm_ops)? I've been observing similar
seemingly unmotivated changes to generated code ...

> --- a/xen/xsm/flask/flask_op.c
> +++ b/xen/xsm/flask/flask_op.c
> @@ -22,6 +22,8 @@
>  #include <objsec.h>
>  #include <conditional.h>
>  
> +#include "../private.h"

Kind of odd: I'd expect a file named such to not get included
across directory levels, unless a single component was split in
such a way (to me Flask and XSM core are separate, yet still
related components).

Jan

On 08/11/2021 09:50, Jan Beulich wrote:
> On 05.11.2021 14:55, Andrew Cooper wrote:
>> Currently, __HYPERVISOR_xsm_op enters xsm_ops.do_{xsm,compat}_op() which means
>> that if any other XSM module registers a handler, we'll break the hypercall
>> ABI totally by having the behaviour depend on the selected XSM module.
>>
>> There are 2 options:
>>   1) Rename __HYPERVISOR_xsm_op to __HYPERVISOR_flask_op.  If another XSM
>>      module wants hypercalls, they'll have to introduce a new top-level
>>      hypercall.
>>   2) Make the current flask_op() be common, and require new hypercalls to fit
>>      compatibly with xen_flask_op_t.  This can be done fairly easily by
>>      subdividing the .cmd space.
>>
>> In this patch, we implement option 2.
>>
>> Move the stub {do,compat}_xsm_op() implementation into a new xsm_op.c so we
>> can more easily do multi-inclusion compat magic.  Also add a new private.h,
>> because flask/hook.c's local declaration of {do,compat}_flask_op() wasn't
>> remotely safe.
>>
>> The top level now dispatches into {do,compat}_flask_op() based on op.cmd, and
>> handles the primary copy in/out.
> I'm not convinced this is the only reasonable way of implementing 2).
> I could also see number space to be separated in different ways, ...
>
>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
>> ---
>> CC: Daniel De Graaf <dgdegra@tycho.nsa.gov>
>> CC: Daniel Smith <dpsmith@apertussolutions.com>
>>
>> Only lightly tested.  Slightly RFC.  There are several things which aren't
>> great, but probably want addressing in due course.
>>
>>   1) The public headers probably want to lose the flask name (in a compatible
>>      way), now that the hypercall is common.  This probably wants to be
>>      combined with no longer taking a void handle.
> ... leaving per-module public headers and perhaps merely adding an
> abstracting
>
> struct xen_xsm_op_t {
>      uint32_t op;
>      ... /* placeholder */
> };
>
> or (making explicit one possible variant of number space splitting)
>
> union xen_xsm_op_t {
>      uint32_t op;
>      struct {
>          uint16_t cmd;
>          uint16_t mod;
>      } u;
>      ... /* placeholder */
> };
>
> in, say, a new public/xsm.h.

That doesn't work.  The ABI is fixed at sizeof(xen_flask_op_t) for all 
other modules, because a caller which doesn't know which module is in 
use must not have Xen over-read/write the object passed while it's 
trying to figure things out.

> As a result I consider this change either going too far (because of
> not knowing future needs) or not far enough (by e.g. leaving
> do_xsm_op() to use xen_flask_op_t.

Well - what it does is prevent someone from breaking the ABI with 
literally this patch

diff --git a/xen/xsm/silo.c b/xen/xsm/silo.c
index 3550dded7b4e..36b82fd3bd3e 100644
--- a/xen/xsm/silo.c
+++ b/xen/xsm/silo.c
@@ -100,6 +100,11 @@ static int silo_argo_send(const struct domain *d1, 
const struct domain *d2)

  #endif

+static long silo_do_xsm_op(XEN_GUEST_HANDLE_PARAM(void) op)
+{
+    /* ... */
+}
+
  static const struct xsm_ops __initconstrel silo_xsm_ops = {
      .evtchn_unbound = silo_evtchn_unbound,
      .evtchn_interdomain = silo_evtchn_interdomain,
@@ -110,6 +115,7 @@ static const struct xsm_ops __initconstrel 
silo_xsm_ops = {
      .argo_register_single_source = silo_argo_register_single_source,
      .argo_send = silo_argo_send,
  #endif
+    .do_xsm_op = silo_do_xsm_op,
  };

  const struct xsm_ops *__init silo_init(void)


This is a damage limitation patch, but without knowing how the next 
module is going to want to do hypercalls differently, it is rather hard 
to judge what is appropriate.  I certainly didn't waste much time 
thinking about it.

>
>>   2) {do,compat}_xsm_op() are currently identical other than the dispatched-to
>>      functions because the size of xsm_flask_op_t is invariant with
>>      COMPAT-ness.  We could simplfy things by only having one, and dispatching
>>      to {do,compat}_*_op() directly, but I'm not sure whether the complexity is
>>      worth it.
> Perhaps not, I would say, not the least because (as said above) I
> think we shouldn't put in place restrictions which may get in the
> way of adding some future module.
>
> Extending struct xen_flask_op to become a generic XSM interface
> structure (or even just for Flask's own purposes) also is not as
> straightforward as it might seem: There's currently no provision
> for sub-structs which would grow the overall size of the structure:
> The copy_{to,from}_guest() invocations for existing sub-ops may not
> copy more that the present worth of sizeof(struct xen_flask_op).
> Yet your implementation of do_xsm_op() moves this deficiency from
> Flask to XSM.

Deficiency yes, but necessary to avoid breaking the ABI for caller which 
doesn't know which module is in use.  This cannot be fixed without 
swapping to approach 1.

>>   3) Bloat-o-meter says these changes add 16 extra bytes to dm_op() and I can't
>>      figure out what could possibly be causing this.
> Without comparing the object files in closer detail it's guesswork,
> but might this be register scheduling differences resulting from
> the changed sizeof(struct xsm_ops)? I've been observing similar
> seemingly unmotivated changes to generated code ...

The only thing it can plausibly be is a knock-on effect from the 
structure of xsm_ops changing.

Normally, I wouldn't be to surprised to see some displacement fields 
dropping from 4 to 1 byte as xsm_ops is getting smaller, but everything 
is alt_call()'d up so should be a 7-byte `call *disp32(%rip)`.

>> --- a/xen/xsm/flask/flask_op.c
>> +++ b/xen/xsm/flask/flask_op.c
>> @@ -22,6 +22,8 @@
>>   #include <objsec.h>
>>   #include <conditional.h>
>>   
>> +#include "../private.h"
> Kind of odd: I'd expect a file named such to not get included
> across directory levels, unless a single component was split in
> such a way (to me Flask and XSM core are separate, yet still
> related components).

Its all a tangled mess because logically separating XSM and Flask was a 
task done when SILO was introduced.

There is not an appropriately located file (under xen/xsm/ ) where the 
prototypes could reasonably live, and this felt like the lesser of the 
available evils.

~Andrew

On 17/11/2021 23:14, Andrew Cooper wrote:
> On 08/11/2021 09:50, Jan Beulich wrote:
>> On 05.11.2021 14:55, Andrew Cooper wrote:
>>> Currently, __HYPERVISOR_xsm_op enters xsm_ops.do_{xsm,compat}_op() 
>>> which means
>>> that if any other XSM module registers a handler, we'll break the 
>>> hypercall
>>> ABI totally by having the behaviour depend on the selected XSM module.
>>>
>>> There are 2 options:
>>>   1) Rename __HYPERVISOR_xsm_op to __HYPERVISOR_flask_op.  If 
>>> another XSM
>>>      module wants hypercalls, they'll have to introduce a new top-level
>>>      hypercall.
>>>   2) Make the current flask_op() be common, and require new 
>>> hypercalls to fit
>>>      compatibly with xen_flask_op_t.  This can be done fairly easily by
>>>      subdividing the .cmd space.
>>>
>>> In this patch, we implement option 2.
>>>
>>> Move the stub {do,compat}_xsm_op() implementation into a new 
>>> xsm_op.c so we
>>> can more easily do multi-inclusion compat magic.  Also add a new 
>>> private.h,
>>> because flask/hook.c's local declaration of {do,compat}_flask_op() 
>>> wasn't
>>> remotely safe.
>>>
>>> The top level now dispatches into {do,compat}_flask_op() based on 
>>> op.cmd, and
>>> handles the primary copy in/out.
>> I'm not convinced this is the only reasonable way of implementing 2).
>> I could also see number space to be separated in different ways, ...
>>
>>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
>>> ---
>>> CC: Daniel De Graaf <dgdegra@tycho.nsa.gov>
>>> CC: Daniel Smith <dpsmith@apertussolutions.com>
>>>
>>> Only lightly tested.  Slightly RFC.  There are several things which 
>>> aren't
>>> great, but probably want addressing in due course.
>>>
>>>   1) The public headers probably want to lose the flask name (in a 
>>> compatible
>>>      way), now that the hypercall is common.  This probably wants to be
>>>      combined with no longer taking a void handle.
>> ... leaving per-module public headers and perhaps merely adding an
>> abstracting
>>
>> struct xen_xsm_op_t {
>>      uint32_t op;
>>      ... /* placeholder */
>> };
>>
>> or (making explicit one possible variant of number space splitting)
>>
>> union xen_xsm_op_t {
>>      uint32_t op;
>>      struct {
>>          uint16_t cmd;
>>          uint16_t mod;
>>      } u;
>>      ... /* placeholder */
>> };
>>
>> in, say, a new public/xsm.h.
>
> That doesn't work.  The ABI is fixed at sizeof(xen_flask_op_t) for all 
> other modules, because a caller which doesn't know which module is in 
> use must not have Xen over-read/write the object passed while it's 
> trying to figure things out.
>
>> As a result I consider this change either going too far (because of
>> not knowing future needs) or not far enough (by e.g. leaving
>> do_xsm_op() to use xen_flask_op_t.
>
> Well - what it does is prevent someone from breaking the ABI with 
> literally this patch
>
> diff --git a/xen/xsm/silo.c b/xen/xsm/silo.c
> index 3550dded7b4e..36b82fd3bd3e 100644
> --- a/xen/xsm/silo.c
> +++ b/xen/xsm/silo.c
> @@ -100,6 +100,11 @@ static int silo_argo_send(const struct domain 
> *d1, const struct domain *d2)
>
>  #endif
>
> +static long silo_do_xsm_op(XEN_GUEST_HANDLE_PARAM(void) op)
> +{
> +    /* ... */
> +}
> +
>  static const struct xsm_ops __initconstrel silo_xsm_ops = {
>      .evtchn_unbound = silo_evtchn_unbound,
>      .evtchn_interdomain = silo_evtchn_interdomain,
> @@ -110,6 +115,7 @@ static const struct xsm_ops __initconstrel 
> silo_xsm_ops = {
>      .argo_register_single_source = silo_argo_register_single_source,
>      .argo_send = silo_argo_send,
>  #endif
> +    .do_xsm_op = silo_do_xsm_op,
>  };
>
>  const struct xsm_ops *__init silo_init(void)
>
>
> This is a damage limitation patch, but without knowing how the next 
> module is going to want to do hypercalls differently, it is rather 
> hard to judge what is appropriate.  I certainly didn't waste much time 
> thinking about it.
>
>>
>>>   2) {do,compat}_xsm_op() are currently identical other than the 
>>> dispatched-to
>>>      functions because the size of xsm_flask_op_t is invariant with
>>>      COMPAT-ness.  We could simplfy things by only having one, and 
>>> dispatching
>>>      to {do,compat}_*_op() directly, but I'm not sure whether the 
>>> complexity is
>>>      worth it.
>> Perhaps not, I would say, not the least because (as said above) I
>> think we shouldn't put in place restrictions which may get in the
>> way of adding some future module.
>>
>> Extending struct xen_flask_op to become a generic XSM interface
>> structure (or even just for Flask's own purposes) also is not as
>> straightforward as it might seem: There's currently no provision
>> for sub-structs which would grow the overall size of the structure:
>> The copy_{to,from}_guest() invocations for existing sub-ops may not
>> copy more that the present worth of sizeof(struct xen_flask_op).
>> Yet your implementation of do_xsm_op() moves this deficiency from
>> Flask to XSM.
>
> Deficiency yes, but necessary to avoid breaking the ABI for caller 
> which doesn't know which module is in use.  This cannot be fixed 
> without swapping to approach 1.
>
>>>   3) Bloat-o-meter says these changes add 16 extra bytes to dm_op() 
>>> and I can't
>>>      figure out what could possibly be causing this.
>> Without comparing the object files in closer detail it's guesswork,
>> but might this be register scheduling differences resulting from
>> the changed sizeof(struct xsm_ops)? I've been observing similar
>> seemingly unmotivated changes to generated code ...
>
> The only thing it can plausibly be is a knock-on effect from the 
> structure of xsm_ops changing.
>
> Normally, I wouldn't be to surprised to see some displacement fields 
> dropping from 4 to 1 byte as xsm_ops is getting smaller, but 
> everything is alt_call()'d up so should be a 7-byte `call *disp32(%rip)`.
>
>>> --- a/xen/xsm/flask/flask_op.c
>>> +++ b/xen/xsm/flask/flask_op.c
>>> @@ -22,6 +22,8 @@
>>>   #include <objsec.h>
>>>   #include <conditional.h>
>>>   +#include "../private.h"
>> Kind of odd: I'd expect a file named such to not get included
>> across directory levels, unless a single component was split in
>> such a way (to me Flask and XSM core are separate, yet still
>> related components).
>
> Its all a tangled mess because logically separating XSM and Flask was 
> a task done when SILO was introduced.
>
> There is not an appropriately located file (under xen/xsm/ ) where the 
> prototypes could reasonably live, and this felt like the lesser of the 
> available evils.

I guess it is worth adding, so we're all on the same page.

The thing I actually need to do is fix the fact that the prototypes for 
{do,compat}_flask_op() are local in xen/xsm/flask/hooks.c and not in a 
header shared with xen/xsm/flask/flask_op.c.

The thing I wanted to do is stop it being so trivially easy to break the 
ABI with a 5-line patch.


Any bikeshedding beyond that can happen in due course.

~Andrew

On 18.11.2021 00:20, Andrew Cooper wrote:
> On 17/11/2021 23:14, Andrew Cooper wrote:
>> On 08/11/2021 09:50, Jan Beulich wrote:
>>> On 05.11.2021 14:55, Andrew Cooper wrote:
>>>> --- a/xen/xsm/flask/flask_op.c
>>>> +++ b/xen/xsm/flask/flask_op.c
>>>> @@ -22,6 +22,8 @@
>>>>   #include <objsec.h>
>>>>   #include <conditional.h>
>>>>   +#include "../private.h"
>>> Kind of odd: I'd expect a file named such to not get included
>>> across directory levels, unless a single component was split in
>>> such a way (to me Flask and XSM core are separate, yet still
>>> related components).
>>
>> Its all a tangled mess because logically separating XSM and Flask was 
>> a task done when SILO was introduced.
>>
>> There is not an appropriately located file (under xen/xsm/ ) where the 
>> prototypes could reasonably live, and this felt like the lesser of the 
>> available evils.
> 
> I guess it is worth adding, so we're all on the same page.
> 
> The thing I actually need to do is fix the fact that the prototypes for 
> {do,compat}_flask_op() are local in xen/xsm/flask/hooks.c and not in a 
> header shared with xen/xsm/flask/flask_op.c.

And I fully agree we need to do so. Me saying "kind of odd" also wasn't
meant as an objection, just to express my surprise to see something like
this.

Jan

On 18.11.2021 00:14, Andrew Cooper wrote:
> On 08/11/2021 09:50, Jan Beulich wrote:
>> On 05.11.2021 14:55, Andrew Cooper wrote:
>>> Currently, __HYPERVISOR_xsm_op enters xsm_ops.do_{xsm,compat}_op() which means
>>> that if any other XSM module registers a handler, we'll break the hypercall
>>> ABI totally by having the behaviour depend on the selected XSM module.
>>>
>>> There are 2 options:
>>>   1) Rename __HYPERVISOR_xsm_op to __HYPERVISOR_flask_op.  If another XSM
>>>      module wants hypercalls, they'll have to introduce a new top-level
>>>      hypercall.
>>>   2) Make the current flask_op() be common, and require new hypercalls to fit
>>>      compatibly with xen_flask_op_t.  This can be done fairly easily by
>>>      subdividing the .cmd space.
>>>
>>> In this patch, we implement option 2.
>>>
>>> Move the stub {do,compat}_xsm_op() implementation into a new xsm_op.c so we
>>> can more easily do multi-inclusion compat magic.  Also add a new private.h,
>>> because flask/hook.c's local declaration of {do,compat}_flask_op() wasn't
>>> remotely safe.
>>>
>>> The top level now dispatches into {do,compat}_flask_op() based on op.cmd, and
>>> handles the primary copy in/out.
>> I'm not convinced this is the only reasonable way of implementing 2).
>> I could also see number space to be separated in different ways, ...
>>
>>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
>>> ---
>>> CC: Daniel De Graaf <dgdegra@tycho.nsa.gov>
>>> CC: Daniel Smith <dpsmith@apertussolutions.com>
>>>
>>> Only lightly tested.  Slightly RFC.  There are several things which aren't
>>> great, but probably want addressing in due course.
>>>
>>>   1) The public headers probably want to lose the flask name (in a compatible
>>>      way), now that the hypercall is common.  This probably wants to be
>>>      combined with no longer taking a void handle.
>> ... leaving per-module public headers and perhaps merely adding an
>> abstracting
>>
>> struct xen_xsm_op_t {
>>      uint32_t op;
>>      ... /* placeholder */
>> };
>>
>> or (making explicit one possible variant of number space splitting)
>>
>> union xen_xsm_op_t {
>>      uint32_t op;
>>      struct {
>>          uint16_t cmd;
>>          uint16_t mod;
>>      } u;
>>      ... /* placeholder */
>> };
>>
>> in, say, a new public/xsm.h.
> 
> That doesn't work.  The ABI is fixed at sizeof(xen_flask_op_t) for all 
> other modules, because a caller which doesn't know which module is in 
> use must not have Xen over-read/write the object passed while it's 
> trying to figure things out.

Well, imo figuring out which module is in use should be via a sysctl.
Then there would be no restrictions by one module's interface
definitions on other modules.

>> As a result I consider this change either going too far (because of
>> not knowing future needs) or not far enough (by e.g. leaving
>> do_xsm_op() to use xen_flask_op_t.
> 
> Well - what it does is prevent someone from breaking the ABI with 
> literally this patch
> 
> diff --git a/xen/xsm/silo.c b/xen/xsm/silo.c
> index 3550dded7b4e..36b82fd3bd3e 100644
> --- a/xen/xsm/silo.c
> +++ b/xen/xsm/silo.c
> @@ -100,6 +100,11 @@ static int silo_argo_send(const struct domain *d1, 
> const struct domain *d2)
> 
>   #endif
> 
> +static long silo_do_xsm_op(XEN_GUEST_HANDLE_PARAM(void) op)
> +{
> +    /* ... */
> +}
> +
>   static const struct xsm_ops __initconstrel silo_xsm_ops = {
>       .evtchn_unbound = silo_evtchn_unbound,
>       .evtchn_interdomain = silo_evtchn_interdomain,
> @@ -110,6 +115,7 @@ static const struct xsm_ops __initconstrel 
> silo_xsm_ops = {
>       .argo_register_single_source = silo_argo_register_single_source,
>       .argo_send = silo_argo_send,
>   #endif
> +    .do_xsm_op = silo_do_xsm_op,
>   };
> 
>   const struct xsm_ops *__init silo_init(void)

So I'm afraid I don't see any ABI breakage here. Provided of course
silo_do_xsm_op() avoids the FLASK_* number space and uses a struct
layout compatible with the head of struct xen_flask_op.

Jan