diff --git a/docs/DataFlowSanitizer.rst b/docs/DataFlowSanitizer.rst
new file mode 100644
index 0000000..4260990
--- /dev/null
+++ b/docs/DataFlowSanitizer.rst
@@ -0,0 +1,77 @@
+=================
+DataFlowSanitizer
+=================
+
+.. contents::
+   :local:
+
+Introduction
+============
+
+DataFlowSanitizer is a generalised dynamic data flow analysis.
+
+Unlike other Sanitizer tools, this tool is not designed to detect a
+specific class of bugs on its own.  Instead, it provides a generic
+dynamic data flow analysis framework to be used by clients to help
+detect application-specific issues within their own code.
+
+Usage
+=====
+
+With no program changes, applying DataFlowSanitizer to a program
+will not alter its behavior.  To use DataFlowSanitizer, the program
+uses API functions to apply tags to data to cause it to be tracked, and to
+check the tag of a specific data item.  DataFlowSanitizer manages
+the propagation of tags through the program according to its data flow.
+
+The APIs are defined in the header file ``sanitizer/dfsan_interface.h``.
+For further information about each function, please refer to the header
+file.
+
+Example
+=======
+
+The following program demonstrates label propagation by checking that
+the correct labels are propagated.
+
+.. code-block:: c++
+
+  #include <sanitizer/dfsan_interface.h>
+  #include <assert.h>
+
+  int main(void) {
+    int i = 1;
+    dfsan_label i_label = dfsan_create_label("i", 0);
+    dfsan_set_label(i_label, &i, sizeof(i));
+
+    int j = 2;
+    dfsan_label j_label = dfsan_create_label("j", 0);
+    dfsan_set_label(j_label, &j, sizeof(j));
+
+    int k = 3;
+    dfsan_label k_label = dfsan_create_label("k", 0);
+    dfsan_set_label(k_label, &k, sizeof(k));
+
+    dfsan_label ij_label = dfsan_get_label(i + j);
+    assert(dfsan_has_label(ij_label, i_label));
+    assert(dfsan_has_label(ij_label, j_label));
+    assert(!dfsan_has_label(ij_label, k_label));
+
+    dfsan_label ijk_label = dfsan_get_label(i + j + k);
+    assert(dfsan_has_label(ijk_label, i_label));
+    assert(dfsan_has_label(ijk_label, j_label));
+    assert(dfsan_has_label(ijk_label, k_label));
+
+    return 0;
+  }
+
+Current status
+==============
+
+DataFlowSanitizer is a work in progress, currently under development for
+x86\_64 Linux.
+
+Design
+======
+
+Please refer to the :doc:`design document<DataFlowSanitizerDesign>`.
diff --git a/docs/DataFlowSanitizerDesign.rst b/docs/DataFlowSanitizerDesign.rst
new file mode 100644
index 0000000..b704035
--- /dev/null
+++ b/docs/DataFlowSanitizerDesign.rst
@@ -0,0 +1,142 @@
+DataFlowSanitizer Design Document
+=================================
+
+This document sets out the design for DataFlowSanitizer, a general
+dynamic data flow analysis.  Unlike other Sanitizer tools, this tool is
+not designed to detect a specific class of bugs on its own. Instead,
+it provides a generic dynamic data flow analysis framework to be used
+by clients to help detect application-specific issues within their
+own code.
+
+DataFlowSanitizer is a program instrumentation which can associate
+a number of taint labels with any data stored in any memory region
+accessible by the program. The analysis is dynamic, which means that
+it operates on a running program, and tracks how the labels propagate
+through that program. The tool shall support a large (>100) number
+of labels, such that programs which operate on large numbers of data
+items may be analysed with each data item being tracked separately.
+
+Use Cases
+---------
+
+This instrumentation can be used as a tool to help monitor how data
+flows from a program's inputs (sources) to its outputs (sinks).
+This has applications from a privacy/security perspective in that
+one can audit how a sensitive data item is used within a program and
+ensure it isn't exiting the program anywhere it shouldn't be.
+
+Interface
+---------
+
+A number of functions are provided which will create taint labels,
+attach labels to memory regions and extract the set of labels
+associated with a specific memory region. These functions are declared
+in the header file ``sanitizer/dfsan_interface.h``.
+
+.. code-block:: c
+
+  /// Creates and returns a base label with the given description and user data.
+  dfsan_label dfsan_create_label(const char *desc, void *userdata);
+
+  /// Sets the label for each address in [addr,addr+size) to \c label.
+  void dfsan_set_label(dfsan_label label, void *addr, size_t size);
+
+  /// Sets the label for each address in [addr,addr+size) to the union of the
+  /// current label for that address and \c label.
+  void dfsan_add_label(dfsan_label label, void *addr, size_t size);
+
+  /// Retrieves the label associated with the given data.
+  ///
+  /// The type of 'data' is arbitrary.  The function accepts a value of any type,
+  /// which can be truncated or extended (implicitly or explicitly) as necessary.
+  /// The truncation/extension operations will preserve the label of the original
+  /// value.
+  dfsan_label dfsan_get_label(long data);
+
+  /// Retrieves a pointer to the dfsan_label_info struct for the given label.
+  const struct dfsan_label_info *dfsan_get_label_info(dfsan_label label);
+
+  /// Returns whether the given label label contains the label elem.
+  int dfsan_has_label(dfsan_label label, dfsan_label elem);
+
+  /// If the given label label contains a label with the description desc, returns
+  /// that label, else returns 0.
+  dfsan_label dfsan_has_label_with_desc(dfsan_label label, const char *desc);
+
+Taint label representation
+--------------------------
+
+As stated above, the tool must track a large number of taint
+labels. This poses an implementation challenge, as most multiple-label
+tainting systems assign one label per bit to shadow storage, and
+union taint labels using a bitwise or operation. This will not scale
+to clients which use hundreds or thousands of taint labels, as the
+label union operation becomes O(n) in the number of supported labels,
+and data associated with it will quickly dominate the live variable
+set, causing register spills and hampering performance.
+
+Instead, a low overhead approach is proposed which is best-case O(log\
+:sub:`2` n) during execution. The underlying assumption is that
+the required space of label unions is sparse, which is a reasonable
+assumption to make given that we are optimizing for the case where
+applications mostly copy data from one place to another, without often
+invoking the need for an actual union operation. The representation
+of a taint label is a 16-bit integer, and new labels are allocated
+sequentially from a pool. The label identifier 0 is special, and means
+that the data item is unlabelled.
+
+When a label union operation is requested at a join point (any
+arithmetic or logical operation with two or more operands, such as
+addition), the code checks whether a union is required, whether the
+same union has been requested before, and whether one union label
+subsumes the other. If so, it returns the previously allocated union
+label. If not, it allocates a new union label from the same pool used
+for new labels.
+
+Specifically, the instrumentation pass will insert code like this
+to decide the union label ``lu`` for a pair of labels ``l1``
+and ``l2``:
+
+.. code-block:: c
+
+  if (l1 == l2)
+    lu = l1;
+  else
+    lu = __dfsan_union(l1, l2);
+
+The equality comparison is outlined, to provide an early exit in
+the common cases where the program is processing unlabelled data, or
+where the two data items have the same label.  ``__dfsan_union`` is
+a runtime library function which performs all other union computation.
+
+Further optimizations are possible, for example if ``l1`` is known
+at compile time to be zero (e.g. it is derived from a constant),
+``l2`` can be used for ``lu``, and vice versa.
+
+Memory layout and label management
+----------------------------------
+
+The following is the current memory layout for Linux/x86\_64:
+
++---------------+---------------+--------------------+
+|    Start      |    End        |        Use         |
++===============+===============+====================+
+| 0x700000008000|0x800000000000 | application memory |
++---------------+---------------+--------------------+
+| 0x200200000000|0x700000008000 |       unused       |
++---------------+---------------+--------------------+
+| 0x200000000000|0x200200000000 |    union table     |
++---------------+---------------+--------------------+
+| 0x000000010000|0x200000000000 |   shadow memory    |
++---------------+---------------+--------------------+
+| 0x000000000000|0x000000010000 | reserved by kernel |
++---------------+---------------+--------------------+
+
+Each byte of application memory corresponds to two bytes of shadow
+memory, which are used to store its taint label. As for LLVM SSA
+registers, we have not found it necessary to associate a label with
+each byte or bit of data, as some other tools do. Instead, labels are
+associated directly with registers.  Loads will result in a union of
+all shadow labels corresponding to bytes loaded (which most of the
+time will be short circuited by the initial comparison) and stores will
+result in a copy of the label to the shadow of all bytes stored to.
diff --git a/docs/UsersManual.rst b/docs/UsersManual.rst
index 80956f8..bdcd471 100644
--- a/docs/UsersManual.rst
+++ b/docs/UsersManual.rst
@@ -895,6 +895,8 @@
       used in conjunction with the ``-fsanitize-undefined-trap-on-error``
       flag. This includes all of the checks listed below other than
       ``unsigned-integer-overflow`` and ``vptr``.
+   -  ``-fsanitize=dataflow``: :doc:`DataFlowSanitizer`, a general data
+      flow analysis.
 
    The following more fine-grained checks are also available:
 
diff --git a/include/clang/Basic/Sanitizers.def b/include/clang/Basic/Sanitizers.def
index 187388c..eb4e92d 100644
--- a/include/clang/Basic/Sanitizers.def
+++ b/include/clang/Basic/Sanitizers.def
@@ -77,6 +77,9 @@
 // IntegerSanitizer
 SANITIZER("unsigned-integer-overflow", UnsignedIntegerOverflow)
 
+// DataFlowSanitizer
+SANITIZER("dataflow", DataFlow)
+
 // -fsanitize=undefined includes all the sanitizers which have low overhead, no
 // ABI or address space layout implications, and only catch undefined behavior.
 SANITIZER_GROUP("undefined", Undefined,
diff --git a/lib/CodeGen/BackendUtil.cpp b/lib/CodeGen/BackendUtil.cpp
index 79cbc38..1edd065 100644
--- a/lib/CodeGen/BackendUtil.cpp
+++ b/lib/CodeGen/BackendUtil.cpp
@@ -206,6 +206,11 @@
   PM.add(createThreadSanitizerPass(CGOpts.SanitizerBlacklistFile));
 }
 
+static void addDataFlowSanitizerPass(const PassManagerBuilder &Builder,
+                                     PassManagerBase &PM) {
+  PM.add(createDataFlowSanitizerPass());
+}
+
 void EmitAssemblyHelper::CreatePasses(TargetMachine *TM) {
   unsigned OptLevel = CodeGenOpts.OptimizationLevel;
   CodeGenOptions::InliningMethod Inlining = CodeGenOpts.getInlining();
@@ -265,6 +270,13 @@
                            addThreadSanitizerPass);
   }
 
+  if (LangOpts.Sanitize.DataFlow) {
+    PMBuilder.addExtension(PassManagerBuilder::EP_OptimizerLast,
+                           addDataFlowSanitizerPass);
+    PMBuilder.addExtension(PassManagerBuilder::EP_EnabledOnOptLevel0,
+                           addDataFlowSanitizerPass);
+  }
+
   // Figure out TargetLibraryInfo.
   Triple TargetTriple(TheModule->getTargetTriple());
   PMBuilder.LibraryInfo = new TargetLibraryInfo(TargetTriple);
diff --git a/lib/Driver/SanitizerArgs.h b/lib/Driver/SanitizerArgs.h
index da6f4f6..58dece9 100644
--- a/lib/Driver/SanitizerArgs.h
+++ b/lib/Driver/SanitizerArgs.h
@@ -37,10 +37,11 @@
     NeedsAsanRt = Address,
     NeedsTsanRt = Thread,
     NeedsMsanRt = Memory,
+    NeedsDfsanRt = DataFlow,
     NeedsLeakDetection = Leak,
     NeedsUbsanRt = Undefined | Integer,
     NotAllowedWithTrap = Vptr,
-    HasZeroBaseShadow = Thread | Memory
+    HasZeroBaseShadow = Thread | Memory | DataFlow
   };
   unsigned Kind;
   std::string BlacklistFile;
@@ -66,6 +67,7 @@
       return false;
     return Kind & NeedsUbsanRt;
   }
+  bool needsDfsanRt() const { return Kind & NeedsDfsanRt; }
 
   bool sanitizesVptr() const { return Kind & Vptr; }
   bool notAllowedWithTrap() const { return Kind & NotAllowedWithTrap; }
diff --git a/lib/Driver/Tools.cpp b/lib/Driver/Tools.cpp
index 254bf8b..4a3ce42 100644
--- a/lib/Driver/Tools.cpp
+++ b/lib/Driver/Tools.cpp
@@ -1860,6 +1860,12 @@
     addSanitizerRTLinkFlagsLinux(TC, Args, CmdArgs, "ubsan_cxx", false);
 }
 
+static void addDfsanRTLinux(const ToolChain &TC, const ArgList &Args,
+                            ArgStringList &CmdArgs) {
+  if (!Args.hasArg(options::OPT_shared))
+    addSanitizerRTLinkFlagsLinux(TC, Args, CmdArgs, "dfsan", true);
+}
+
 static bool shouldUseFramePointer(const ArgList &Args,
                                   const llvm::Triple &Triple) {
   if (Arg *A = Args.getLastArg(options::OPT_fno_omit_frame_pointer,
@@ -6275,6 +6281,8 @@
     addMsanRTLinux(getToolChain(), Args, CmdArgs);
   if (Sanitize.needsLsanRt())
     addLsanRTLinux(getToolChain(), Args, CmdArgs);
+  if (Sanitize.needsDfsanRt())
+    addDfsanRTLinux(getToolChain(), Args, CmdArgs);
 
   // The profile runtime also needs access to system libraries.
   addProfileRTLinux(getToolChain(), Args, CmdArgs);
diff --git a/lib/Lex/PPMacroExpansion.cpp b/lib/Lex/PPMacroExpansion.cpp
index a9d76c3..0eb1169 100644
--- a/lib/Lex/PPMacroExpansion.cpp
+++ b/lib/Lex/PPMacroExpansion.cpp
@@ -908,6 +908,7 @@
            .Case("enumerator_attributes", true)
            .Case("memory_sanitizer", LangOpts.Sanitize.Memory)
            .Case("thread_sanitizer", LangOpts.Sanitize.Thread)
+           .Case("dataflow_sanitizer", LangOpts.Sanitize.DataFlow)
            // Objective-C features
            .Case("objc_arr", LangOpts.ObjCAutoRefCount) // FIXME: REMOVE?
            .Case("objc_arc", LangOpts.ObjCAutoRefCount)
diff --git a/runtime/compiler-rt/Makefile b/runtime/compiler-rt/Makefile
index 59a62e7..0fc93c7 100644
--- a/runtime/compiler-rt/Makefile
+++ b/runtime/compiler-rt/Makefile
@@ -109,7 +109,8 @@
 ifeq ($(ARCH),x86_64)
 RuntimeLibrary.linux.Configs += \
 	full-x86_64.a profile-x86_64.a san-x86_64.a asan-x86_64.a \
-	tsan-x86_64.a msan-x86_64.a ubsan-x86_64.a ubsan_cxx-x86_64.a
+	tsan-x86_64.a msan-x86_64.a ubsan-x86_64.a ubsan_cxx-x86_64.a \
+	dfsan-x86_64.a
 # We need to build 32-bit ASan/UBsan libraries on 64-bit platform, and add them
 # to the list of runtime libraries to make
 # "clang -fsanitize=(address|undefined) -m32" work.