[refactor] Extract the platform-specific TrustStore instantiations and

net/cert/internal/system_trust_store.h

+// Copyright 2017 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef NET_CERT_INTERNAL_SYSTEM_TRUST_STORE_H_
+#define NET_CERT_INTERNAL_SYSTEM_TRUST_STORE_H_
+
+#include <vector>
+
+#include "base/memory/ref_counted.h"
+#include "net/base/net_export.h"
+#include "net/cert/internal/parsed_certificate.h"
+
+namespace net {
+
+class TrustStore;
+class CertIssuerSource;
+class TrustAnchor;
+
+// The SystemTrustStore interface is used to encapsulate a TrustStore for the
+// current platform, with some extra bells and whistles.
+//
+// This is primarily used to abstract out the platform-specific bits that
+// relate to configuring the TrustStore needed for path building.
+//
+// Implementations of SystemTrustStore create an effective trust
+// store that is the composition of:
+//
+//   * The platform-specific trust store
+//   * A set of manually added trust anchors
+//   * Test certificates added via ScopedTestRoot
+class SystemTrustStore {
+ public:
+  virtual ~SystemTrustStore() {}
+
+  // Returns an aggregate TrustStore that can be used by the path builder. The
+  // store composes the system trust store (if implemented) with manually added
+  // trust anchors added via AddTrustAnchor(). This pointer is non-owned, and
+  // valid only for the lifetime of |this|.
+  virtual TrustStore* GetTrustStore() = 0;
+
+  // Returns false if the implementation of SystemTrustStore doesn't actually
+  // make use of the system's trust store. This might be the case for
+  // unsupported platforms. In the case where this returns true, the trust store

[mattm, 2017/04/19 20:26:04]

true here should be false?

[eroman, 2017/04/19 21:10:41]

Done.

+  // returned by GetTrustStore() is made up solely of the manually added trust
+  // anchors (via AddTrustAnchor()).
+  virtual bool UsesSystemTrustStore() const = 0;
+
+  // TODO(eroman): Expose this through the TrustStore interface instead?
+  //
+  // Returns a CertIssuerSource that finds any intermediates that are present in
+  // the system trust store. These intermediates are not necessarily trusted,
+  // however may be used during path building as another means of finding
+  // certificates. If the implementation of SystemTrustStore doesn't support
+  // this feature may return nullptr.
+  virtual CertIssuerSource* GetCertIssuerSource() = 0;
+
+  // IsKnownRoot() returns true if the given trust anchor originated from the
+  // system trust store and is a "standard" one. The meaning of "standard" is
+  // that it is one of default trust anchors for the system, as opposed to a
+  // user-installed one. IsKnownRoot() is only guaranteed to work for
+  // TrustAnchors returned by GetTrustStore().
+  virtual bool IsKnownRoot(
+      const scoped_refptr<TrustAnchor>& trust_anchor) const = 0;
+
+  // Adds a trust anchor to this particular instance of SystemTrustStore, and
+  // not globally for the system. Calling this should not have a persistent
+  // effect on the system, however some implementations may implement this in a
+  // manner that has a global effect within the process.

[mattm, 2017/04/19 20:26:04]

Is there a use case in mind for having one with a global effect?

[eroman, 2017/04/19 21:10:41]

I am just documenting how things currently work (for instance in the
pre-existing NSS implementation).

I don't think it *should* work that way, however in practice it does (because
the ScopedTestRoot manipulates the global in-memory store).

[mattm, 2017/04/19 21:36:00]

Well, I guess it is true that ScopedTestRoot works that way, but I don't think
AddTrustAnchor needs to care (Ie, you might pipe the certs from TestRootCerts
into AddTrustAnchor, but not the other way around.) So as far as this is
concerned, I don't think there is a case where calling AddTrustAnchor should
modify process-global state?

[eroman, 2017/04/19 21:42:53]

Oh apologies, yes my comment is wrong I will remove it.

For some reason I was thinking and documenting the TestRootCerts behavior
here...
Yes you are correct we don't ever want that to happen here (should never be
globally visible). 
In fact, the very first sentence of the paragraph contradicts that second part
:)

[eroman, 2017/04/19 21:45:06]

Done.

+  virtual void AddTrustAnchor(
+      const scoped_refptr<TrustAnchor>& trust_anchor) = 0;
+
+  // Returns true if |trust_anchor| was one added via |AddTrustAnchor()|. This
+  // is only guaranteed to work if |trust_anchor| was one returned by
+  // GetTrustStore(), as it may be implemented by pointer comparison rather than
+  // SPKI comparison.
+  virtual bool IsAdditionalTrustAnchor(
+      const scoped_refptr<TrustAnchor>& trust_anchor) const = 0;
+};
+
+// Creates an instance of SystemTrustStore that wraps the current platform's SSL
+// trust store. This canno return nullptr, even in the case where system trust
+// store integration is not supported. In this latter case, the SystemTrustStore
+// will only give access to the manually added trust anchors. This can be
+// inspected by testing whether UsesSystemTrustStore() returns false.
+NET_EXPORT std::unique_ptr<SystemTrustStore> CreateSslSystemTrustStore();
+
+}  // namespace net
+
+#endif  // NET_CERT_INTERNAL_SYSTEM_TRUST_STORE_H_