1 //==-- X86LoadValueInjectionLoadHardening.cpp - LVI load hardening for x86 --=//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 /// Description: This pass finds Load Value Injection (LVI) gadgets consisting
10 /// of a load from memory (i.e., SOURCE), and any operation that may transmit
11 /// the value loaded from memory over a covert channel, or use the value loaded
12 /// from memory to determine a branch/call target (i.e., SINK). After finding
13 /// all such gadgets in a given function, the pass minimally inserts LFENCE
14 /// instructions in such a manner that the following property is satisfied: for
15 /// all SOURCE+SINK pairs, all paths in the CFG from SOURCE to SINK contain at
16 /// least one LFENCE instruction. The algorithm that implements this minimal
17 /// insertion is influenced by an academic paper that minimally inserts memory
18 /// fences for high-performance concurrent programs:
19 /// http://www.cs.ucr.edu/~lesani/companion/oopsla15/OOPSLA15.pdf
20 /// The algorithm implemented in this pass is as follows:
21 /// 1. Build a condensed CFG (i.e., a GadgetGraph) consisting only of the
22 /// following components:
23 /// - SOURCE instructions (also includes function arguments)
24 /// - SINK instructions
25 /// - Basic block entry points
26 /// - Basic block terminators
27 /// - LFENCE instructions
28 /// 2. Analyze the GadgetGraph to determine which SOURCE+SINK pairs (i.e.,
29 /// gadgets) are already mitigated by existing LFENCEs. If all gadgets have been
30 /// mitigated, go to step 6.
31 /// 3. Use a heuristic or plugin to approximate minimal LFENCE insertion.
32 /// 4. Insert one LFENCE along each CFG edge that was cut in step 3.
34 /// 6. If any LFENCEs were inserted, return `true` from runOnMachineFunction()
35 /// to tell LLVM that the function was modified.
37 //===----------------------------------------------------------------------===//
39 #include "ImmutableGraph.h"
41 #include "X86Subtarget.h"
42 #include "X86TargetMachine.h"
43 #include "llvm/ADT/DenseMap.h"
44 #include "llvm/ADT/DenseSet.h"
45 #include "llvm/ADT/SmallSet.h"
46 #include "llvm/ADT/Statistic.h"
47 #include "llvm/ADT/StringRef.h"
48 #include "llvm/CodeGen/MachineBasicBlock.h"
49 #include "llvm/CodeGen/MachineDominanceFrontier.h"
50 #include "llvm/CodeGen/MachineDominators.h"
51 #include "llvm/CodeGen/MachineFunction.h"
52 #include "llvm/CodeGen/MachineFunctionPass.h"
53 #include "llvm/CodeGen/MachineInstr.h"
54 #include "llvm/CodeGen/MachineInstrBuilder.h"
55 #include "llvm/CodeGen/MachineLoopInfo.h"
56 #include "llvm/CodeGen/MachineRegisterInfo.h"
57 #include "llvm/CodeGen/RDFGraph.h"
58 #include "llvm/CodeGen/RDFLiveness.h"
59 #include "llvm/InitializePasses.h"
60 #include "llvm/Support/CommandLine.h"
61 #include "llvm/Support/DOTGraphTraits.h"
62 #include "llvm/Support/Debug.h"
63 #include "llvm/Support/DynamicLibrary.h"
64 #include "llvm/Support/GraphWriter.h"
65 #include "llvm/Support/raw_ostream.h"
69 #define PASS_KEY "x86-lvi-load"
70 #define DEBUG_TYPE PASS_KEY
72 STATISTIC(NumFences, "Number of LFENCEs inserted for LVI mitigation");
73 STATISTIC(NumFunctionsConsidered, "Number of functions analyzed");
74 STATISTIC(NumFunctionsMitigated, "Number of functions for which mitigations "
76 STATISTIC(NumGadgets, "Number of LVI gadgets detected during analysis");
78 static cl::opt<std::string> OptimizePluginPath(
79 PASS_KEY "-opt-plugin",
80 cl::desc("Specify a plugin to optimize LFENCE insertion"), cl::Hidden);
82 static cl::opt<bool> NoConditionalBranches(
83 PASS_KEY "-no-cbranch",
84 cl::desc("Don't treat conditional branches as disclosure gadgets. This "
85 "may improve performance, at the cost of security."),
86 cl::init(false), cl::Hidden);
88 static cl::opt<bool> EmitDot(
91 "For each function, emit a dot graph depicting potential LVI gadgets"),
92 cl::init(false), cl::Hidden);
94 static cl::opt<bool> EmitDotOnly(
96 cl::desc("For each function, emit a dot graph depicting potential LVI "
97 "gadgets, and do not insert any fences"),
98 cl::init(false), cl::Hidden);
100 static cl::opt<bool> EmitDotVerify(
101 PASS_KEY "-dot-verify",
102 cl::desc("For each function, emit a dot graph to stdout depicting "
103 "potential LVI gadgets, used for testing purposes only"),
104 cl::init(false), cl::Hidden);
106 static llvm::sys::DynamicLibrary OptimizeDL;
107 typedef int (*OptimizeCutT)(unsigned int *nodes, unsigned int nodes_size,
108 unsigned int *edges, int *edge_values,
109 int *cut_edges /* out */, unsigned int edges_size);
110 static OptimizeCutT OptimizeCut = nullptr;
114 struct MachineGadgetGraph : ImmutableGraph<MachineInstr *, int> {
115 static constexpr int GadgetEdgeSentinel = -1;
116 static constexpr MachineInstr *const ArgNodeSentinel = nullptr;
118 using GraphT = ImmutableGraph<MachineInstr *, int>;
119 using Node = typename GraphT::Node;
120 using Edge = typename GraphT::Edge;
121 using size_type = typename GraphT::size_type;
122 MachineGadgetGraph(std::unique_ptr<Node[]> Nodes,
123 std::unique_ptr<Edge[]> Edges, size_type NodesSize,
124 size_type EdgesSize, int NumFences = 0, int NumGadgets = 0)
125 : GraphT(std::move(Nodes), std::move(Edges), NodesSize, EdgesSize),
126 NumFences(NumFences), NumGadgets(NumGadgets) {}
127 static inline bool isCFGEdge(const Edge &E) {
128 return E.getValue() != GadgetEdgeSentinel;
130 static inline bool isGadgetEdge(const Edge &E) {
131 return E.getValue() == GadgetEdgeSentinel;
137 class X86LoadValueInjectionLoadHardeningPass : public MachineFunctionPass {
139 X86LoadValueInjectionLoadHardeningPass() : MachineFunctionPass(ID) {}
141 StringRef getPassName() const override {
142 return "X86 Load Value Injection (LVI) Load Hardening";
144 void getAnalysisUsage(AnalysisUsage &AU) const override;
145 bool runOnMachineFunction(MachineFunction &MF) override;
150 using GraphBuilder = ImmutableGraphBuilder<MachineGadgetGraph>;
151 using EdgeSet = MachineGadgetGraph::EdgeSet;
152 using NodeSet = MachineGadgetGraph::NodeSet;
153 using Gadget = std::pair<MachineInstr *, MachineInstr *>;
155 const X86Subtarget *STI;
156 const TargetInstrInfo *TII;
157 const TargetRegisterInfo *TRI;
159 std::unique_ptr<MachineGadgetGraph>
160 getGadgetGraph(MachineFunction &MF, const MachineLoopInfo &MLI,
161 const MachineDominatorTree &MDT,
162 const MachineDominanceFrontier &MDF) const;
163 int hardenLoadsWithPlugin(MachineFunction &MF,
164 std::unique_ptr<MachineGadgetGraph> Graph) const;
165 int hardenLoadsWithGreedyHeuristic(
166 MachineFunction &MF, std::unique_ptr<MachineGadgetGraph> Graph) const;
167 int elimMitigatedEdgesAndNodes(MachineGadgetGraph &G,
168 EdgeSet &ElimEdges /* in, out */,
169 NodeSet &ElimNodes /* in, out */) const;
170 std::unique_ptr<MachineGadgetGraph>
171 trimMitigatedEdges(std::unique_ptr<MachineGadgetGraph> Graph) const;
172 void findAndCutEdges(MachineGadgetGraph &G,
173 EdgeSet &CutEdges /* out */) const;
174 int insertFences(MachineFunction &MF, MachineGadgetGraph &G,
175 EdgeSet &CutEdges /* in, out */) const;
176 bool instrUsesRegToAccessMemory(const MachineInstr &I, unsigned Reg) const;
177 bool instrUsesRegToBranch(const MachineInstr &I, unsigned Reg) const;
178 inline bool isFence(const MachineInstr *MI) const {
179 return MI && (MI->getOpcode() == X86::LFENCE ||
180 (STI->useLVIControlFlowIntegrity() && MI->isCall()));
184 } // end anonymous namespace
189 struct GraphTraits<MachineGadgetGraph *>
190 : GraphTraits<ImmutableGraph<MachineInstr *, int> *> {};
193 struct DOTGraphTraits<MachineGadgetGraph *> : DefaultDOTGraphTraits {
194 using GraphType = MachineGadgetGraph;
195 using Traits = llvm::GraphTraits<GraphType *>;
196 using NodeRef = typename Traits::NodeRef;
197 using EdgeRef = typename Traits::EdgeRef;
198 using ChildIteratorType = typename Traits::ChildIteratorType;
199 using ChildEdgeIteratorType = typename Traits::ChildEdgeIteratorType;
201 DOTGraphTraits(bool isSimple = false) : DefaultDOTGraphTraits(isSimple) {}
203 std::string getNodeLabel(NodeRef Node, GraphType *) {
204 if (Node->getValue() == MachineGadgetGraph::ArgNodeSentinel)
208 raw_string_ostream OS(Str);
209 OS << *Node->getValue();
213 static std::string getNodeAttributes(NodeRef Node, GraphType *) {
214 MachineInstr *MI = Node->getValue();
215 if (MI == MachineGadgetGraph::ArgNodeSentinel)
216 return "color = blue";
217 if (MI->getOpcode() == X86::LFENCE)
218 return "color = green";
222 static std::string getEdgeAttributes(NodeRef, ChildIteratorType E,
224 int EdgeVal = (*E.getCurrent()).getValue();
225 return EdgeVal >= 0 ? "label = " + std::to_string(EdgeVal)
226 : "color = red, style = \"dashed\"";
230 } // end namespace llvm
232 constexpr MachineInstr *MachineGadgetGraph::ArgNodeSentinel;
233 constexpr int MachineGadgetGraph::GadgetEdgeSentinel;
235 char X86LoadValueInjectionLoadHardeningPass::ID = 0;
237 void X86LoadValueInjectionLoadHardeningPass::getAnalysisUsage(
238 AnalysisUsage &AU) const {
239 MachineFunctionPass::getAnalysisUsage(AU);
240 AU.addRequired<MachineLoopInfo>();
241 AU.addRequired<MachineDominatorTree>();
242 AU.addRequired<MachineDominanceFrontier>();
243 AU.setPreservesCFG();
246 static void WriteGadgetGraph(raw_ostream &OS, MachineFunction &MF,
247 MachineGadgetGraph *G) {
248 WriteGraph(OS, G, /*ShortNames*/ false,
249 "Speculative gadgets for \"" + MF.getName() + "\" function");
252 bool X86LoadValueInjectionLoadHardeningPass::runOnMachineFunction(
253 MachineFunction &MF) {
254 LLVM_DEBUG(dbgs() << "***** " << getPassName() << " : " << MF.getName()
256 STI = &MF.getSubtarget<X86Subtarget>();
257 if (!STI->useLVILoadHardening())
260 // FIXME: support 32-bit
262 report_fatal_error("LVI load hardening is only supported on 64-bit", false);
264 // Don't skip functions with the "optnone" attr but participate in opt-bisect.
265 const Function &F = MF.getFunction();
266 if (!F.hasOptNone() && skipFunction(F))
269 ++NumFunctionsConsidered;
270 TII = STI->getInstrInfo();
271 TRI = STI->getRegisterInfo();
272 LLVM_DEBUG(dbgs() << "Building gadget graph...\n");
273 const auto &MLI = getAnalysis<MachineLoopInfo>();
274 const auto &MDT = getAnalysis<MachineDominatorTree>();
275 const auto &MDF = getAnalysis<MachineDominanceFrontier>();
276 std::unique_ptr<MachineGadgetGraph> Graph = getGadgetGraph(MF, MLI, MDT, MDF);
277 LLVM_DEBUG(dbgs() << "Building gadget graph... Done\n");
278 if (Graph == nullptr)
279 return false; // didn't find any gadgets
282 WriteGadgetGraph(outs(), MF, Graph.get());
286 if (EmitDot || EmitDotOnly) {
287 LLVM_DEBUG(dbgs() << "Emitting gadget graph...\n");
288 std::error_code FileError;
289 std::string FileName = "lvi.";
290 FileName += MF.getName();
292 raw_fd_ostream FileOut(FileName, FileError);
294 errs() << FileError.message();
295 WriteGadgetGraph(FileOut, MF, Graph.get());
297 LLVM_DEBUG(dbgs() << "Emitting gadget graph... Done\n");
303 if (!OptimizePluginPath.empty()) {
304 if (!OptimizeDL.isValid()) {
305 std::string ErrorMsg;
306 OptimizeDL = llvm::sys::DynamicLibrary::getPermanentLibrary(
307 OptimizePluginPath.c_str(), &ErrorMsg);
308 if (!ErrorMsg.empty())
309 report_fatal_error("Failed to load opt plugin: \"" + ErrorMsg + '\"');
310 OptimizeCut = (OptimizeCutT)OptimizeDL.getAddressOfSymbol("optimize_cut");
312 report_fatal_error("Invalid optimization plugin");
314 FencesInserted = hardenLoadsWithPlugin(MF, std::move(Graph));
315 } else { // Use the default greedy heuristic
316 FencesInserted = hardenLoadsWithGreedyHeuristic(MF, std::move(Graph));
319 if (FencesInserted > 0)
320 ++NumFunctionsMitigated;
321 NumFences += FencesInserted;
322 return (FencesInserted > 0);
325 std::unique_ptr<MachineGadgetGraph>
326 X86LoadValueInjectionLoadHardeningPass::getGadgetGraph(
327 MachineFunction &MF, const MachineLoopInfo &MLI,
328 const MachineDominatorTree &MDT,
329 const MachineDominanceFrontier &MDF) const {
332 // Build the Register Dataflow Graph using the RDF framework
333 TargetOperandInfo TOI{*TII};
334 DataFlowGraph DFG{MF, *TII, *TRI, MDT, MDF, TOI};
336 Liveness L{MF.getRegInfo(), DFG};
339 GraphBuilder Builder;
340 using GraphIter = typename GraphBuilder::BuilderNodeRef;
341 DenseMap<MachineInstr *, GraphIter> NodeMap;
342 int FenceCount = 0, GadgetCount = 0;
343 auto MaybeAddNode = [&NodeMap, &Builder](MachineInstr *MI) {
344 auto Ref = NodeMap.find(MI);
345 if (Ref == NodeMap.end()) {
346 auto I = Builder.addVertex(MI);
348 return std::pair<GraphIter, bool>{I, true};
350 return std::pair<GraphIter, bool>{Ref->getSecond(), false};
353 // The `Transmitters` map memoizes transmitters found for each def. If a def
354 // has not yet been analyzed, then it will not appear in the map. If a def
355 // has been analyzed and was determined not to have any transmitters, then
356 // its list of transmitters will be empty.
357 DenseMap<NodeId, std::vector<NodeId>> Transmitters;
359 // Analyze all machine instructions to find gadgets and LFENCEs, adding
360 // each interesting value to `Nodes`
361 auto AnalyzeDef = [&](NodeAddr<DefNode *> SourceDef) {
362 SmallSet<NodeId, 8> UsesVisited, DefsVisited;
363 std::function<void(NodeAddr<DefNode *>)> AnalyzeDefUseChain =
364 [&](NodeAddr<DefNode *> Def) {
365 if (Transmitters.find(Def.Id) != Transmitters.end())
366 return; // Already analyzed `Def`
368 // Use RDF to find all the uses of `Def`
370 RegisterRef DefReg = DFG.getPRI().normalize(Def.Addr->getRegRef(DFG));
371 for (auto UseID : L.getAllReachedUses(DefReg, Def)) {
372 auto Use = DFG.addr<UseNode *>(UseID);
373 if (Use.Addr->getFlags() & NodeAttrs::PhiRef) { // phi node
374 NodeAddr<PhiNode *> Phi = Use.Addr->getOwner(DFG);
375 for (auto I : L.getRealUses(Phi.Id)) {
376 if (DFG.getPRI().alias(RegisterRef(I.first), DefReg)) {
377 for (auto UA : I.second)
378 Uses.emplace(UA.first);
381 } else { // not a phi node
386 // For each use of `Def`, we want to know whether:
387 // (1) The use can leak the Def'ed value,
388 // (2) The use can further propagate the Def'ed value to more defs
389 for (auto UseID : Uses) {
390 if (!UsesVisited.insert(UseID).second)
391 continue; // Already visited this use of `Def`
393 auto Use = DFG.addr<UseNode *>(UseID);
394 assert(!(Use.Addr->getFlags() & NodeAttrs::PhiRef));
395 MachineOperand &UseMO = Use.Addr->getOp();
396 MachineInstr &UseMI = *UseMO.getParent();
397 assert(UseMO.isReg());
399 // We naively assume that an instruction propagates any loaded
400 // uses to all defs unless the instruction is a call, in which
401 // case all arguments will be treated as gadget sources during
402 // analysis of the callee function.
406 // Check whether this use can transmit (leak) its value.
407 if (instrUsesRegToAccessMemory(UseMI, UseMO.getReg()) ||
408 (!NoConditionalBranches &&
409 instrUsesRegToBranch(UseMI, UseMO.getReg()))) {
410 Transmitters[Def.Id].push_back(Use.Addr->getOwner(DFG).Id);
412 continue; // Found a transmitting load -- no need to continue
413 // traversing its defs (i.e., this load will become
414 // a new gadget source anyways).
417 // Check whether the use propagates to more defs.
418 NodeAddr<InstrNode *> Owner{Use.Addr->getOwner(DFG)};
419 rdf::NodeList AnalyzedChildDefs;
420 for (auto &ChildDef :
421 Owner.Addr->members_if(DataFlowGraph::IsDef, DFG)) {
422 if (!DefsVisited.insert(ChildDef.Id).second)
423 continue; // Already visited this def
424 if (Def.Addr->getAttrs() & NodeAttrs::Dead)
426 if (Def.Id == ChildDef.Id)
427 continue; // `Def` uses itself (e.g., increment loop counter)
429 AnalyzeDefUseChain(ChildDef);
431 // `Def` inherits all of its child defs' transmitters.
432 for (auto TransmitterId : Transmitters[ChildDef.Id])
433 Transmitters[Def.Id].push_back(TransmitterId);
437 // Note that this statement adds `Def.Id` to the map if no
438 // transmitters were found for `Def`.
439 auto &DefTransmitters = Transmitters[Def.Id];
441 // Remove duplicate transmitters
442 llvm::sort(DefTransmitters);
443 DefTransmitters.erase(
444 std::unique(DefTransmitters.begin(), DefTransmitters.end()),
445 DefTransmitters.end());
448 // Find all of the transmitters
449 AnalyzeDefUseChain(SourceDef);
450 auto &SourceDefTransmitters = Transmitters[SourceDef.Id];
451 if (SourceDefTransmitters.empty())
452 return; // No transmitters for `SourceDef`
454 MachineInstr *Source = SourceDef.Addr->getFlags() & NodeAttrs::PhiRef
455 ? MachineGadgetGraph::ArgNodeSentinel
456 : SourceDef.Addr->getOp().getParent();
457 auto GadgetSource = MaybeAddNode(Source);
458 // Each transmitter is a sink for `SourceDef`.
459 for (auto TransmitterId : SourceDefTransmitters) {
460 MachineInstr *Sink = DFG.addr<StmtNode *>(TransmitterId).Addr->getCode();
461 auto GadgetSink = MaybeAddNode(Sink);
462 // Add the gadget edge to the graph.
463 Builder.addEdge(MachineGadgetGraph::GadgetEdgeSentinel,
464 GadgetSource.first, GadgetSink.first);
469 LLVM_DEBUG(dbgs() << "Analyzing def-use chains to find gadgets\n");
470 // Analyze function arguments
471 NodeAddr<BlockNode *> EntryBlock = DFG.getFunc().Addr->getEntryBlock(DFG);
472 for (NodeAddr<PhiNode *> ArgPhi :
473 EntryBlock.Addr->members_if(DataFlowGraph::IsPhi, DFG)) {
474 NodeList Defs = ArgPhi.Addr->members_if(DataFlowGraph::IsDef, DFG);
475 llvm::for_each(Defs, AnalyzeDef);
477 // Analyze every instruction in MF
478 for (NodeAddr<BlockNode *> BA : DFG.getFunc().Addr->members(DFG)) {
479 for (NodeAddr<StmtNode *> SA :
480 BA.Addr->members_if(DataFlowGraph::IsCode<NodeAttrs::Stmt>, DFG)) {
481 MachineInstr *MI = SA.Addr->getCode();
485 } else if (MI->mayLoad()) {
486 NodeList Defs = SA.Addr->members_if(DataFlowGraph::IsDef, DFG);
487 llvm::for_each(Defs, AnalyzeDef);
491 LLVM_DEBUG(dbgs() << "Found " << FenceCount << " fences\n");
492 LLVM_DEBUG(dbgs() << "Found " << GadgetCount << " gadgets\n");
493 if (GadgetCount == 0)
495 NumGadgets += GadgetCount;
497 // Traverse CFG to build the rest of the graph
498 SmallSet<MachineBasicBlock *, 8> BlocksVisited;
499 std::function<void(MachineBasicBlock *, GraphIter, unsigned)> TraverseCFG =
500 [&](MachineBasicBlock *MBB, GraphIter GI, unsigned ParentDepth) {
501 unsigned LoopDepth = MLI.getLoopDepth(MBB);
503 // Always add the first instruction in each block
504 auto NI = MBB->begin();
505 auto BeginBB = MaybeAddNode(&*NI);
506 Builder.addEdge(ParentDepth, GI, BeginBB.first);
507 if (!BlocksVisited.insert(MBB).second)
510 // Add any instructions within the block that are gadget components
512 while (++NI != MBB->end()) {
513 auto Ref = NodeMap.find(&*NI);
514 if (Ref != NodeMap.end()) {
515 Builder.addEdge(LoopDepth, GI, Ref->getSecond());
516 GI = Ref->getSecond();
520 // Always add the terminator instruction, if one exists
521 auto T = MBB->getFirstTerminator();
522 if (T != MBB->end()) {
523 auto EndBB = MaybeAddNode(&*T);
525 Builder.addEdge(LoopDepth, GI, EndBB.first);
529 for (MachineBasicBlock *Succ : MBB->successors())
530 TraverseCFG(Succ, GI, LoopDepth);
532 // ArgNodeSentinel is a pseudo-instruction that represents MF args in the
534 GraphIter ArgNode = MaybeAddNode(MachineGadgetGraph::ArgNodeSentinel).first;
535 TraverseCFG(&MF.front(), ArgNode, 0);
536 std::unique_ptr<MachineGadgetGraph> G{Builder.get(FenceCount, GadgetCount)};
537 LLVM_DEBUG(dbgs() << "Found " << G->nodes_size() << " nodes\n");
541 // Returns the number of remaining gadget edges that could not be eliminated
542 int X86LoadValueInjectionLoadHardeningPass::elimMitigatedEdgesAndNodes(
543 MachineGadgetGraph &G, MachineGadgetGraph::EdgeSet &ElimEdges /* in, out */,
544 MachineGadgetGraph::NodeSet &ElimNodes /* in, out */) const {
545 if (G.NumFences > 0) {
546 // Eliminate fences and CFG edges that ingress and egress the fence, as
547 // they are trivially mitigated.
548 for (const auto &E : G.edges()) {
549 const MachineGadgetGraph::Node *Dest = E.getDest();
550 if (isFence(Dest->getValue())) {
551 ElimNodes.insert(*Dest);
553 for (const auto &DE : Dest->edges())
554 ElimEdges.insert(DE);
559 // Find and eliminate gadget edges that have been mitigated.
560 int MitigatedGadgets = 0, RemainingGadgets = 0;
561 MachineGadgetGraph::NodeSet ReachableNodes{G};
562 for (const auto &RootN : G.nodes()) {
563 if (llvm::none_of(RootN.edges(), MachineGadgetGraph::isGadgetEdge))
564 continue; // skip this node if it isn't a gadget source
566 // Find all of the nodes that are CFG-reachable from RootN using DFS
567 ReachableNodes.clear();
568 std::function<void(const MachineGadgetGraph::Node *, bool)>
570 [&](const MachineGadgetGraph::Node *N, bool FirstNode) {
572 ReachableNodes.insert(*N);
573 for (const auto &E : N->edges()) {
574 const MachineGadgetGraph::Node *Dest = E.getDest();
575 if (MachineGadgetGraph::isCFGEdge(E) &&
576 !ElimEdges.contains(E) && !ReachableNodes.contains(*Dest))
577 FindReachableNodes(Dest, false);
580 FindReachableNodes(&RootN, true);
582 // Any gadget whose sink is unreachable has been mitigated
583 for (const auto &E : RootN.edges()) {
584 if (MachineGadgetGraph::isGadgetEdge(E)) {
585 if (ReachableNodes.contains(*E.getDest())) {
586 // This gadget's sink is reachable
588 } else { // This gadget's sink is unreachable, and therefore mitigated
595 return RemainingGadgets;
598 std::unique_ptr<MachineGadgetGraph>
599 X86LoadValueInjectionLoadHardeningPass::trimMitigatedEdges(
600 std::unique_ptr<MachineGadgetGraph> Graph) const {
601 MachineGadgetGraph::NodeSet ElimNodes{*Graph};
602 MachineGadgetGraph::EdgeSet ElimEdges{*Graph};
603 int RemainingGadgets =
604 elimMitigatedEdgesAndNodes(*Graph, ElimEdges, ElimNodes);
605 if (ElimEdges.empty() && ElimNodes.empty()) {
606 Graph->NumFences = 0;
607 Graph->NumGadgets = RemainingGadgets;
609 Graph = GraphBuilder::trim(*Graph, ElimNodes, ElimEdges, 0 /* NumFences */,
615 int X86LoadValueInjectionLoadHardeningPass::hardenLoadsWithPlugin(
616 MachineFunction &MF, std::unique_ptr<MachineGadgetGraph> Graph) const {
617 int FencesInserted = 0;
620 LLVM_DEBUG(dbgs() << "Eliminating mitigated paths...\n");
621 Graph = trimMitigatedEdges(std::move(Graph));
622 LLVM_DEBUG(dbgs() << "Eliminating mitigated paths... Done\n");
623 if (Graph->NumGadgets == 0)
626 LLVM_DEBUG(dbgs() << "Cutting edges...\n");
627 EdgeSet CutEdges{*Graph};
628 auto Nodes = std::make_unique<unsigned int[]>(Graph->nodes_size() +
629 1 /* terminator node */);
630 auto Edges = std::make_unique<unsigned int[]>(Graph->edges_size());
631 auto EdgeCuts = std::make_unique<int[]>(Graph->edges_size());
632 auto EdgeValues = std::make_unique<int[]>(Graph->edges_size());
633 for (const auto &N : Graph->nodes()) {
634 Nodes[Graph->getNodeIndex(N)] = Graph->getEdgeIndex(*N.edges_begin());
636 Nodes[Graph->nodes_size()] = Graph->edges_size(); // terminator node
637 for (const auto &E : Graph->edges()) {
638 Edges[Graph->getEdgeIndex(E)] = Graph->getNodeIndex(*E.getDest());
639 EdgeValues[Graph->getEdgeIndex(E)] = E.getValue();
641 OptimizeCut(Nodes.get(), Graph->nodes_size(), Edges.get(), EdgeValues.get(),
642 EdgeCuts.get(), Graph->edges_size());
643 for (int I = 0; I < Graph->edges_size(); ++I)
646 LLVM_DEBUG(dbgs() << "Cutting edges... Done\n");
647 LLVM_DEBUG(dbgs() << "Cut " << CutEdges.count() << " edges\n");
649 LLVM_DEBUG(dbgs() << "Inserting LFENCEs...\n");
650 FencesInserted += insertFences(MF, *Graph, CutEdges);
651 LLVM_DEBUG(dbgs() << "Inserting LFENCEs... Done\n");
652 LLVM_DEBUG(dbgs() << "Inserted " << FencesInserted << " fences\n");
654 Graph = GraphBuilder::trim(*Graph, MachineGadgetGraph::NodeSet{*Graph},
658 return FencesInserted;
661 int X86LoadValueInjectionLoadHardeningPass::hardenLoadsWithGreedyHeuristic(
662 MachineFunction &MF, std::unique_ptr<MachineGadgetGraph> Graph) const {
663 LLVM_DEBUG(dbgs() << "Eliminating mitigated paths...\n");
664 Graph = trimMitigatedEdges(std::move(Graph));
665 LLVM_DEBUG(dbgs() << "Eliminating mitigated paths... Done\n");
666 if (Graph->NumGadgets == 0)
669 LLVM_DEBUG(dbgs() << "Cutting edges...\n");
670 MachineGadgetGraph::NodeSet ElimNodes{*Graph}, GadgetSinks{*Graph};
671 MachineGadgetGraph::EdgeSet ElimEdges{*Graph}, CutEdges{*Graph};
672 auto IsCFGEdge = [&ElimEdges, &CutEdges](const MachineGadgetGraph::Edge &E) {
673 return !ElimEdges.contains(E) && !CutEdges.contains(E) &&
674 MachineGadgetGraph::isCFGEdge(E);
676 auto IsGadgetEdge = [&ElimEdges,
677 &CutEdges](const MachineGadgetGraph::Edge &E) {
678 return !ElimEdges.contains(E) && !CutEdges.contains(E) &&
679 MachineGadgetGraph::isGadgetEdge(E);
682 // FIXME: this is O(E^2), we could probably do better.
684 // Find the cheapest CFG edge that will eliminate a gadget (by being
685 // egress from a SOURCE node or ingress to a SINK node), and cut it.
686 const MachineGadgetGraph::Edge *CheapestSoFar = nullptr;
688 // First, collect all gadget source and sink nodes.
689 MachineGadgetGraph::NodeSet GadgetSources{*Graph}, GadgetSinks{*Graph};
690 for (const auto &N : Graph->nodes()) {
691 if (ElimNodes.contains(N))
693 for (const auto &E : N.edges()) {
694 if (IsGadgetEdge(E)) {
695 GadgetSources.insert(N);
696 GadgetSinks.insert(*E.getDest());
701 // Next, look for the cheapest CFG edge which, when cut, is guaranteed to
702 // mitigate at least one gadget by either:
703 // (a) being egress from a gadget source, or
704 // (b) being ingress to a gadget sink.
705 for (const auto &N : Graph->nodes()) {
706 if (ElimNodes.contains(N))
708 for (const auto &E : N.edges()) {
710 if (GadgetSources.contains(N) || GadgetSinks.contains(*E.getDest())) {
711 if (!CheapestSoFar || E.getValue() < CheapestSoFar->getValue())
718 assert(CheapestSoFar && "Failed to cut an edge");
719 CutEdges.insert(*CheapestSoFar);
720 ElimEdges.insert(*CheapestSoFar);
721 } while (elimMitigatedEdgesAndNodes(*Graph, ElimEdges, ElimNodes));
722 LLVM_DEBUG(dbgs() << "Cutting edges... Done\n");
723 LLVM_DEBUG(dbgs() << "Cut " << CutEdges.count() << " edges\n");
725 LLVM_DEBUG(dbgs() << "Inserting LFENCEs...\n");
726 int FencesInserted = insertFences(MF, *Graph, CutEdges);
727 LLVM_DEBUG(dbgs() << "Inserting LFENCEs... Done\n");
728 LLVM_DEBUG(dbgs() << "Inserted " << FencesInserted << " fences\n");
730 return FencesInserted;
733 int X86LoadValueInjectionLoadHardeningPass::insertFences(
734 MachineFunction &MF, MachineGadgetGraph &G,
735 EdgeSet &CutEdges /* in, out */) const {
736 int FencesInserted = 0;
737 for (const auto &N : G.nodes()) {
738 for (const auto &E : N.edges()) {
739 if (CutEdges.contains(E)) {
740 MachineInstr *MI = N.getValue(), *Prev;
741 MachineBasicBlock *MBB; // Insert an LFENCE in this MBB
742 MachineBasicBlock::iterator InsertionPt; // ...at this point
743 if (MI == MachineGadgetGraph::ArgNodeSentinel) {
744 // insert LFENCE at beginning of entry block
746 InsertionPt = MBB->begin();
748 } else if (MI->isBranch()) { // insert the LFENCE before the branch
749 MBB = MI->getParent();
751 Prev = MI->getPrevNode();
752 // Remove all egress CFG edges from this branch because the inserted
753 // LFENCE prevents gadgets from crossing the branch.
754 for (const auto &E : N.edges()) {
755 if (MachineGadgetGraph::isCFGEdge(E))
758 } else { // insert the LFENCE after the instruction
759 MBB = MI->getParent();
760 InsertionPt = MI->getNextNode() ? MI->getNextNode() : MBB->end();
761 Prev = InsertionPt == MBB->end()
762 ? (MBB->empty() ? nullptr : &MBB->back())
763 : InsertionPt->getPrevNode();
765 // Ensure this insertion is not redundant (two LFENCEs in sequence).
766 if ((InsertionPt == MBB->end() || !isFence(&*InsertionPt)) &&
767 (!Prev || !isFence(Prev))) {
768 BuildMI(*MBB, InsertionPt, DebugLoc(), TII->get(X86::LFENCE));
774 return FencesInserted;
777 bool X86LoadValueInjectionLoadHardeningPass::instrUsesRegToAccessMemory(
778 const MachineInstr &MI, unsigned Reg) const {
779 if (!MI.mayLoadOrStore() || MI.getOpcode() == X86::MFENCE ||
780 MI.getOpcode() == X86::SFENCE || MI.getOpcode() == X86::LFENCE)
783 // FIXME: This does not handle pseudo loading instruction like TCRETURN*
784 const MCInstrDesc &Desc = MI.getDesc();
785 int MemRefBeginIdx = X86II::getMemoryOperandNo(Desc.TSFlags);
786 if (MemRefBeginIdx < 0) {
787 LLVM_DEBUG(dbgs() << "Warning: unable to obtain memory operand for loading "
789 MI.print(dbgs()); dbgs() << '\n';);
792 MemRefBeginIdx += X86II::getOperandBias(Desc);
794 const MachineOperand &BaseMO =
795 MI.getOperand(MemRefBeginIdx + X86::AddrBaseReg);
796 const MachineOperand &IndexMO =
797 MI.getOperand(MemRefBeginIdx + X86::AddrIndexReg);
798 return (BaseMO.isReg() && BaseMO.getReg() != X86::NoRegister &&
799 TRI->regsOverlap(BaseMO.getReg(), Reg)) ||
800 (IndexMO.isReg() && IndexMO.getReg() != X86::NoRegister &&
801 TRI->regsOverlap(IndexMO.getReg(), Reg));
804 bool X86LoadValueInjectionLoadHardeningPass::instrUsesRegToBranch(
805 const MachineInstr &MI, unsigned Reg) const {
806 if (!MI.isConditionalBranch())
808 for (const MachineOperand &Use : MI.uses())
809 if (Use.isReg() && Use.getReg() == Reg)
814 INITIALIZE_PASS_BEGIN(X86LoadValueInjectionLoadHardeningPass, PASS_KEY,
815 "X86 LVI load hardening", false, false)
816 INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
817 INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
818 INITIALIZE_PASS_DEPENDENCY(MachineDominanceFrontier)
819 INITIALIZE_PASS_END(X86LoadValueInjectionLoadHardeningPass, PASS_KEY,
820 "X86 LVI load hardening", false, false)
822 FunctionPass *llvm::createX86LoadValueInjectionLoadHardeningPass() {
823 return new X86LoadValueInjectionLoadHardeningPass();
828 /// The `X86LoadValueInjectionLoadHardeningPass` above depends on expensive
829 /// analysis passes that add complexity to the pipeline. This complexity
830 /// can cause noticable overhead when no optimizations are enabled, i.e., -O0.
831 /// The purpose of `X86LoadValueInjectionLoadHardeningUnoptimizedPass` is to
832 /// provide the same security as the optimized pass, but without adding
833 /// unnecessary complexity to the LLVM pipeline.
835 /// The behavior of this pass is simply to insert an LFENCE after every load
837 class X86LoadValueInjectionLoadHardeningUnoptimizedPass
838 : public MachineFunctionPass {
840 X86LoadValueInjectionLoadHardeningUnoptimizedPass()
841 : MachineFunctionPass(ID) {}
843 StringRef getPassName() const override {
844 return "X86 Load Value Injection (LVI) Load Hardening (Unoptimized)";
846 bool runOnMachineFunction(MachineFunction &MF) override;
850 } // end anonymous namespace
852 char X86LoadValueInjectionLoadHardeningUnoptimizedPass::ID = 0;
854 bool X86LoadValueInjectionLoadHardeningUnoptimizedPass::runOnMachineFunction(
855 MachineFunction &MF) {
856 LLVM_DEBUG(dbgs() << "***** " << getPassName() << " : " << MF.getName()
858 const X86Subtarget *STI = &MF.getSubtarget<X86Subtarget>();
859 if (!STI->useLVILoadHardening())
862 // FIXME: support 32-bit
864 report_fatal_error("LVI load hardening is only supported on 64-bit", false);
866 // Don't skip functions with the "optnone" attr but participate in opt-bisect.
867 const Function &F = MF.getFunction();
868 if (!F.hasOptNone() && skipFunction(F))
871 bool Modified = false;
872 ++NumFunctionsConsidered;
874 const TargetInstrInfo *TII = STI->getInstrInfo();
875 for (auto &MBB : MF) {
876 for (auto &MI : MBB) {
877 if (!MI.mayLoad() || MI.getOpcode() == X86::LFENCE ||
878 MI.getOpcode() == X86::MFENCE)
881 MachineBasicBlock::iterator InsertionPt =
882 MI.getNextNode() ? MI.getNextNode() : MBB.end();
883 BuildMI(MBB, InsertionPt, DebugLoc(), TII->get(X86::LFENCE));
890 ++NumFunctionsMitigated;
895 INITIALIZE_PASS(X86LoadValueInjectionLoadHardeningUnoptimizedPass, PASS_KEY,
896 "X86 LVI load hardening", false, false)
898 FunctionPass *llvm::createX86LoadValueInjectionLoadHardeningUnoptimizedPass() {
899 return new X86LoadValueInjectionLoadHardeningUnoptimizedPass();