Add implementation for Ad Networks Database

chrome/browser/extensions/activity_log/ad_network_database.cc

Index: chrome/browser/extensions/activity_log/ad_network_database.cc
diff --git a/chrome/browser/extensions/activity_log/ad_network_database.cc b/chrome/browser/extensions/activity_log/ad_network_database.cc
index 4b4b45b5aa49730056bf5a8a534eb072a646a133..2d6ea284cb7fb8aa51d7f4e3f03960e8af37045e 100644
--- a/chrome/browser/extensions/activity_log/ad_network_database.cc
+++ b/chrome/browser/extensions/activity_log/ad_network_database.cc
@@ -4,21 +4,112 @@
 
 #include "chrome/browser/extensions/activity_log/ad_network_database.h"
 
+#include "base/basictypes.h"
 #include "base/lazy_instance.h"
+#include "base/memory/ref_counted_memory.h"
+#include "crypto/secure_hash.h"
+#include "crypto/sha2.h"
+#include "grit/browser_resources.h"
+#include "ui/base/resource/resource_bundle.h"
+#include "url/gurl.h"
 
 namespace extensions {
 
 namespace {
 
+// We use a hash size of 8 for these for three reasons.
+// 1. It saves us a bit on space, and, since we have to store these in memory
+//    (reading from disk would be far too slow because these checks are
+//    performed synchronously), that space is important.
+// 2. Since we don't store full hashes, reconstructing the list is more
+//    difficult. This may mean we get a few incorrect hits, but the security is
+//    worth the (very small) amount of noise.
+// 3. It fits nicely into a int64.
+const size_t kUrlHashSize = 8u;
+COMPILE_ASSERT(kUrlHashSize <= sizeof(int64), url_hashes_must_fit_into_a_int64);
+
+const size_t kChecksumHashSize = 32u;
+
+class AdNetworkDatabaseImpl : public AdNetworkDatabase {
+ public:
+  AdNetworkDatabaseImpl();
+  virtual ~AdNetworkDatabaseImpl();
+
+ private:
+  // AdNetworkDatabase implementation.
+  virtual bool IsEnabled() const OVERRIDE { return enabled_; }
+  virtual bool IsAdNetwork(const GURL& url) const OVERRIDE;
+
+  // Initialize the AdNetworkDatabase. This means initializing the set of
+  // hashes from the shared memory. If this succeeds, |enabled_| will be set
+  // to true.
+  void Init();
+
+  // The set of partial hashes for known ad networks.
+  base::hash_set<int64> entries_;
+
+  // Whether or not the database is enabled. The database will not be enabled
+  // if Init() does not succeed in loading the hashes.
+  bool enabled_;
+};
+
+AdNetworkDatabaseImpl::AdNetworkDatabaseImpl() : enabled_(false) {
+  Init();
+}
+
+AdNetworkDatabaseImpl::~AdNetworkDatabaseImpl() {}
+
+void AdNetworkDatabaseImpl::Init() {
+  base::RefCountedStaticMemory* entries_memory =
+      ResourceBundle::GetSharedInstance().LoadDataResourceBytes(
+          IDR_AD_NETWORK_HASHES);
+
+  const size_t size = entries_memory->size();
+  const unsigned char* const front = entries_memory->front();
+  if (size < kChecksumHashSize ||
+      (size - kChecksumHashSize) % kUrlHashSize != 0) {

[felt, 2014/04/23 03:01:36]

should this be a DCHECK?

[Devlin, 2014/04/23 18:16:55]

Hmm... good question.  If it were a normal file, then no, because corruption
happens.  But since it's a binary resource baked in, is that possible?  Perhaps
not.  Let's make it a DCHECK, and, if we get tons of crashes, we'll know
something's up. :)

+    return;
+  }
+
+  // The format of the data resource is fairly straight-forward:
+  // <32-bit checksum><list of 64-bit hashes of hosts>, with no linebreaks or
+  // other separations.

[felt, 2014/04/23 03:01:36]

how are the list entries (the hashes) separated?

[Devlin, 2014/04/23 18:16:55]

There's no deliminator between them, because they are all a strict size (as is
the checksum) so we know exactly where in the file they should begin and end,
and just read the 8 bytes at a time.

[felt, 2014/04/23 20:15:22]

ah right

+  scoped_ptr<crypto::SecureHash> hash(
+      crypto::SecureHash::Create(crypto::SecureHash::SHA256));
+
+  hash->Update(front + kChecksumHashSize, size - kChecksumHashSize);
+  char hash_value[kChecksumHashSize];
+  hash->Finish(hash_value, kChecksumHashSize);
+  // If the checksum doesn't match, abort.

[felt, 2014/04/23 03:01:36]

should this also be a DCHECK?

[Devlin, 2014/04/23 18:16:55]

Done.

+  if (memcmp(hash_value, front, kChecksumHashSize) != 0)
+    return;
+
+  // Construct and insert all hashes.
+  for (const unsigned char* index = front + kChecksumHashSize;
+       index < front + size;
+       index += kUrlHashSize) {
+    int64 value = 0;
+    memcpy(&value, index, kUrlHashSize);
+    entries_.insert(value);
+  }
+
+  enabled_ = true;
+}
+
+bool AdNetworkDatabaseImpl::IsAdNetwork(const GURL& url) const {
+  int64 hash = 0;
+  crypto::SHA256HashString(url.host(), &hash, sizeof(hash));
+  return entries_.count(hash) != 0;
+}
+
 class AdNetworkDatabaseFactory {
  public:
   AdNetworkDatabaseFactory();
   ~AdNetworkDatabaseFactory();
 
+  const AdNetworkDatabase* GetDatabase();
   void SetDatabase(scoped_ptr<AdNetworkDatabase> database);
 
-  const AdNetworkDatabase* database() const { return database_.get(); }
-
  private:
   scoped_ptr<AdNetworkDatabase> database_;
 };
@@ -26,6 +117,17 @@ class AdNetworkDatabaseFactory {
 AdNetworkDatabaseFactory::AdNetworkDatabaseFactory() {}
 AdNetworkDatabaseFactory::~AdNetworkDatabaseFactory() {}
 
+const AdNetworkDatabase* AdNetworkDatabaseFactory::GetDatabase() {
+  // Construct a new database, if we don't have one.
+  if (!database_.get())
+    database_.reset(new AdNetworkDatabaseImpl());
+
+  if (database_->IsEnabled())
+    return database_.get();
+
+  return NULL;
+}
+
 void AdNetworkDatabaseFactory::SetDatabase(
     scoped_ptr<AdNetworkDatabase> database) {
   database_.reset(database.release());
@@ -40,7 +142,7 @@ AdNetworkDatabase::~AdNetworkDatabase() {}
 
 // static
 const AdNetworkDatabase* AdNetworkDatabase::Get() {
-  return g_factory.Get().database();
+  return g_factory.Get().GetDatabase();
 }
 
 // static