[dfsan] Re-exec with no ASLR if memory layout is incompatible on Linux (llvm#85674)

DFSan's shadow mappings are incompatible with 32 bits of ASLR entropy
('sudo sysctl vm.mmap_rnd_bits=32; ninja check-dfsan') and it is
difficult to fix this via increasing the size of the shadow mappings,
due to the overhead of shadow memory. This patch works around the issue
by detecting if the memory layout is incompatible, and if so,
re-exec'ing without ASLR.

DFSan and MSan share copy-pasted shadow memory code, hence this
workaround is ported from MSan:
- "[msan] Re-exec with no ASLR if memory layout is incompatible on
Linux"
(llvm@58f7251)
- "[msan] Add 'MappingDesc::ALLOCATOR' type and check it is available"
(llvm@af2bf86)
(which in turn are inspired by TSan: "Re-exec TSan with no ASLR if
memory layout is incompatible on Linux"
(llvm@0784b1e
))

aeubanks had remarked in
llvm#85142 (comment)
that this issue occurs in Chromium:
https://ci.chromium.org/ui/p/chromium/builders/try/linux_upload_clang/5066/overview

Author: thurstond

compiler-rt/lib/dfsan/dfsan.cpp

@@ -33,6 +33,9 @@
 #include "sanitizer_common/sanitizer_libc.h"
 #include "sanitizer_common/sanitizer_report_decorator.h"
 #include "sanitizer_common/sanitizer_stacktrace.h"
+#if SANITIZER_LINUX
+#  include <sys/personality.h>
+#endif
 
 using namespace __dfsan;
 
@@ -1127,11 +1130,12 @@ static void CheckMemoryLayoutSanity() {
 
 // TODO: CheckMemoryRangeAvailability is based on msan.
 // Consider refactoring these into a shared implementation.
-static bool CheckMemoryRangeAvailability(uptr beg, uptr size) {
+static bool CheckMemoryRangeAvailability(uptr beg, uptr size, bool verbose) {
   if (size > 0) {
     uptr end = beg + size - 1;
     if (!MemoryRangeIsAvailable(beg, end)) {
-      Printf("FATAL: Memory range %p - %p is not available.\n", beg, end);
+      if (verbose)
+        Printf("FATAL: Memory range %p - %p is not available.\n", beg, end);
       return false;
     }
   }
@@ -1163,7 +1167,7 @@ static bool ProtectMemoryRange(uptr beg, uptr size, const char *name) {
 
 // TODO: InitShadow is based on msan.
 // Consider refactoring these into a shared implementation.
-bool InitShadow(bool init_origins) {
+bool InitShadow(bool init_origins, bool dry_run) {
   // Let user know mapping parameters first.
   VPrintf(1, "dfsan_init %p\n", (void *)&__dfsan::dfsan_init);
   for (unsigned i = 0; i < kMemoryLayoutSize; ++i)
@@ -1173,8 +1177,9 @@ bool InitShadow(bool init_origins) {
   CheckMemoryLayoutSanity();
 
   if (!MEM_IS_APP(&__dfsan::dfsan_init)) {
-    Printf("FATAL: Code %p is out of application range. Non-PIE build?\n",
-           (uptr)&__dfsan::dfsan_init);
+    if (!dry_run)
+      Printf("FATAL: Code %p is out of application range. Non-PIE build?\n",
+             (uptr)&__dfsan::dfsan_init);
     return false;
   }
 
@@ -1195,27 +1200,62 @@ bool InitShadow(bool init_origins) {
     bool protect = type == MappingDesc::INVALID ||
                    (!init_origins && type == MappingDesc::ORIGIN);
     CHECK(!(map && protect));
-    if (!map && !protect)
-      CHECK(type == MappingDesc::APP);
+    if (!map && !protect) {
+      CHECK(type == MappingDesc::APP || type == MappingDesc::ALLOCATOR);
+
+      if (dry_run && type == MappingDesc::ALLOCATOR &&
+          !CheckMemoryRangeAvailability(start, size, !dry_run))
+        return false;
+    }
     if (map) {
-      if (!CheckMemoryRangeAvailability(start, size))
+      if (dry_run && !CheckMemoryRangeAvailability(start, size, !dry_run))
         return false;
-      if (!MmapFixedSuperNoReserve(start, size, kMemoryLayout[i].name))
+      if (!dry_run &&
+          !MmapFixedSuperNoReserve(start, size, kMemoryLayout[i].name))
         return false;
-      if (common_flags()->use_madv_dontdump)
+      if (!dry_run && common_flags()->use_madv_dontdump)
         DontDumpShadowMemory(start, size);
     }
     if (protect) {
-      if (!CheckMemoryRangeAvailability(start, size))
+      if (dry_run && !CheckMemoryRangeAvailability(start, size, !dry_run))
         return false;
-      if (!ProtectMemoryRange(start, size, kMemoryLayout[i].name))
+      if (!dry_run && !ProtectMemoryRange(start, size, kMemoryLayout[i].name))
         return false;
     }
   }
 
   return true;
 }
 
+bool InitShadowWithReExec(bool init_origins) {
+  // Start with dry run: check layout is ok, but don't print warnings because
+  // warning messages will cause tests to fail (even if we successfully re-exec
+  // after the warning).
+  bool success = InitShadow(init_origins, true);
+  if (!success) {
+#if SANITIZER_LINUX
+    // Perhaps ASLR entropy is too high. If ASLR is enabled, re-exec without it.
+    int old_personality = personality(0xffffffff);
+    bool aslr_on =
+        (old_personality != -1) && ((old_personality & ADDR_NO_RANDOMIZE) == 0);
+
+    if (aslr_on) {
+      VReport(1,
+              "WARNING: DataflowSanitizer: memory layout is incompatible, "
+              "possibly due to high-entropy ASLR.\n"
+              "Re-execing with fixed virtual address space.\n"
+              "N.B. reducing ASLR entropy is preferable.\n");
+      CHECK_NE(personality(old_personality | ADDR_NO_RANDOMIZE), -1);
+      ReExec();
+    }
+#endif
+  }
+
+  // The earlier dry run didn't actually map or protect anything. Run again in
+  // non-dry run mode.
+  return success && InitShadow(init_origins, false);
+}
+
 static void DFsanInit(int argc, char **argv, char **envp) {
   CHECK(!dfsan_init_is_running);
   if (dfsan_inited)
@@ -1229,7 +1269,11 @@ static void DFsanInit(int argc, char **argv, char **envp) {
 
   CheckASLR();
 
-  InitShadow(dfsan_get_track_origins());
+  if (!InitShadowWithReExec(dfsan_get_track_origins())) {
+    Printf("FATAL: DataflowSanitizer can not mmap the shadow memory.\n");
+    DumpProcessMap();
+    Die();
+  }
 
   initialize_interceptors();
 

compiler-rt/lib/dfsan/dfsan_allocator.cpp

@@ -37,6 +37,9 @@ struct DFsanMapUnmapCallback {
   void OnUnmap(uptr p, uptr size) const { dfsan_set_label(0, (void *)p, size); }
 };
 
+// Note: to ensure that the allocator is compatible with the application memory
+// layout (especially with high-entropy ASLR), kSpaceBeg and kSpaceSize must be
+// duplicated as MappingDesc::ALLOCATOR in dfsan_platform.h.
 #if defined(__aarch64__)
 const uptr kAllocatorSpace = 0xE00000000000ULL;
 #else

compiler-rt/lib/dfsan/dfsan_platform.h

@@ -27,10 +27,19 @@ using __sanitizer::uptr;
 struct MappingDesc {
   uptr start;
   uptr end;
-  enum Type { INVALID, APP, SHADOW, ORIGIN } type;
+  enum Type {
+    INVALID = 1,
+    ALLOCATOR = 2,
+    APP = 4,
+    SHADOW = 8,
+    ORIGIN = 16,
+  } type;
   const char *name;
 };
 
+// Note: MappingDesc::ALLOCATOR entries are only used to check for memory
+// layout compatibility. The actual allocation settings are in
+// dfsan_allocator.cpp, which need to be kept in sync.
 #if SANITIZER_LINUX && SANITIZER_WORDSIZE == 64
 
 #  if defined(__aarch64__)
@@ -53,7 +62,8 @@ const MappingDesc kMemoryLayout[] = {
     {0X0B00000000000, 0X0C00000000000, MappingDesc::SHADOW, "shadow-10-13"},
     {0X0C00000000000, 0X0D00000000000, MappingDesc::INVALID, "invalid"},
     {0X0D00000000000, 0X0E00000000000, MappingDesc::ORIGIN, "origin-10-13"},
-    {0X0E00000000000, 0X1000000000000, MappingDesc::APP, "app-15"},
+    {0X0E00000000000, 0X0E40000000000, MappingDesc::ALLOCATOR, "allocator"},
+    {0X0E40000000000, 0X1000000000000, MappingDesc::APP, "app-15"},
 };
 #    define MEM_TO_SHADOW(mem) ((uptr)mem ^ 0xB00000000000ULL)
 #    define SHADOW_TO_ORIGIN(shadow) (((uptr)(shadow)) + 0x200000000000ULL)
@@ -76,7 +86,8 @@ const MappingDesc kMemoryLayout[] = {
     {0x510000000000ULL, 0x600000000000ULL, MappingDesc::APP, "app-2"},
     {0x600000000000ULL, 0x610000000000ULL, MappingDesc::ORIGIN, "origin-1"},
     {0x610000000000ULL, 0x700000000000ULL, MappingDesc::INVALID, "invalid"},
-    {0x700000000000ULL, 0x800000000000ULL, MappingDesc::APP, "app-3"}};
+    {0x700000000000ULL, 0x740000000000ULL, MappingDesc::ALLOCATOR, "allocator"},
+    {0x740000000000ULL, 0x800000000000ULL, MappingDesc::APP, "app-3"}};
 #    define MEM_TO_SHADOW(mem) (((uptr)(mem)) ^ 0x500000000000ULL)
 #    define SHADOW_TO_ORIGIN(mem) (((uptr)(mem)) + 0x100000000000ULL)
 #  endif
@@ -93,20 +104,21 @@ const uptr kMemoryLayoutSize = sizeof(kMemoryLayout) / sizeof(kMemoryLayout[0]);
 __attribute__((optimize("unroll-loops")))
 #endif
 inline bool
-addr_is_type(uptr addr, MappingDesc::Type mapping_type) {
+addr_is_type(uptr addr, int mapping_types) {
 // It is critical for performance that this loop is unrolled (because then it is
 // simplified into just a few constant comparisons).
 #ifdef __clang__
 #  pragma unroll
 #endif
   for (unsigned i = 0; i < kMemoryLayoutSize; ++i)
-    if (kMemoryLayout[i].type == mapping_type &&
+    if ((kMemoryLayout[i].type & mapping_types) &&
         addr >= kMemoryLayout[i].start && addr < kMemoryLayout[i].end)
       return true;
   return false;
 }
 
-#define MEM_IS_APP(mem) addr_is_type((uptr)(mem), MappingDesc::APP)
+#define MEM_IS_APP(mem) \
+  (addr_is_type((uptr)(mem), MappingDesc::APP | MappingDesc::ALLOCATOR))
 #define MEM_IS_SHADOW(mem) addr_is_type((uptr)(mem), MappingDesc::SHADOW)
 #define MEM_IS_ORIGIN(mem) addr_is_type((uptr)(mem), MappingDesc::ORIGIN)