[llvm-branch-commits] [compiler-rt] release/18.x: [msan] Re-exec with no ASLR if memory layout is incompatible on Linux (#85142) (PR #86201)

Tom Stellard via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Wed Mar 27 16:02:45 PDT 2024


https://github.com/tstellar updated https://github.com/llvm/llvm-project/pull/86201

>From b74f6158460c8d65203d8e5218faf584bbbbe096 Mon Sep 17 00:00:00 2001
From: Thurston Dang <thurston at google.com>
Date: Thu, 14 Mar 2024 16:19:30 -0700
Subject: [PATCH 1/2] [msan] Add 'MappingDesc::ALLOCATOR' type and check it is
 available (#85153)

MSan divides the virtual address space into APP, INVALID, SHADOW and
ORIGIN memory. The allocator usually just steals a bit of the APP
address space: typically the bottom portion of the PIE binaries section,
which works because the Linux kernel maps from the top of the PIE
binaries section. However, if ASLR is very aggressive, the binary may
end up mapped in the same location where the allocator wants to live;
this results in a segfault.

This patch adds in a MappingDesc::ALLOCATOR type and enforces that the
memory range for the allocator is not occupied by anything else.

Since the allocator range information is not readily available in
msan.h, we duplicate the information from msan_allocator.cpp.

Note: aggressive ASLR can also lead to a different type of failure,
where the PIE binaries/libraries are mapped entirely outside of the
APP/ALLOCATOR sections; that will be addressed in a separate patch
(https://github.com/llvm/llvm-project/pull/85142).

(cherry picked from commit af2bf86a372cacf5f536bae06e2f2d3886eefb7b)
---
 compiler-rt/lib/msan/msan.h             | 33 +++++++++++++++++--------
 compiler-rt/lib/msan/msan_allocator.cpp |  3 +++
 compiler-rt/lib/msan/msan_linux.cpp     | 11 ++++++---
 3 files changed, 34 insertions(+), 13 deletions(-)

diff --git a/compiler-rt/lib/msan/msan.h b/compiler-rt/lib/msan/msan.h
index 710447a3e1a357..00f448fb377d59 100644
--- a/compiler-rt/lib/msan/msan.h
+++ b/compiler-rt/lib/msan/msan.h
@@ -33,12 +33,18 @@ struct MappingDesc {
   uptr start;
   uptr end;
   enum Type {
-    INVALID, APP, SHADOW, ORIGIN
+    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
+// msan_allocator.cpp, which need to be kept in sync.
 #if SANITIZER_LINUX && defined(__mips64)
 
 // MIPS64 maps:
@@ -84,7 +90,8 @@ 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)
@@ -106,7 +113,8 @@ 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(shadow) (((uptr)(shadow)) + 0x100000000000ULL)
 
@@ -118,7 +126,8 @@ const MappingDesc kMemoryLayout[] = {
     {0x180200000000ULL, 0x1C0000000000ULL, MappingDesc::INVALID, "invalid"},
     {0x1C0000000000ULL, 0x2C0200000000ULL, MappingDesc::ORIGIN, "origin"},
     {0x2C0200000000ULL, 0x300000000000ULL, MappingDesc::INVALID, "invalid"},
-    {0x300000000000ULL, 0x800000000000ULL, MappingDesc::APP, "high memory"}};
+    {0x300000000000ULL, 0x320000000000ULL, MappingDesc::ALLOCATOR, "allocator"},
+    {0x320000000000ULL, 0x800000000000ULL, MappingDesc::APP, "high memory"}};
 
 // Various kernels use different low end ranges but we can combine them into one
 // big range. They also use different high end ranges but we can map them all to
@@ -141,7 +150,8 @@ const MappingDesc kMemoryLayout[] = {
     {0x180000000000ULL, 0x1C0000000000ULL, MappingDesc::INVALID, "invalid"},
     {0x1C0000000000ULL, 0x2C0000000000ULL, MappingDesc::ORIGIN, "origin"},
     {0x2C0000000000ULL, 0x440000000000ULL, MappingDesc::INVALID, "invalid"},
-    {0x440000000000ULL, 0x500000000000ULL, MappingDesc::APP, "high memory"}};
+    {0x440000000000ULL, 0x460000000000ULL, MappingDesc::ALLOCATOR, "allocator"},
+    {0x460000000000ULL, 0x500000000000ULL, MappingDesc::APP, "high memory"}};
 
 #define MEM_TO_SHADOW(mem) \
   ((((uptr)(mem)) & ~0xC00000000000ULL) + 0x080000000000ULL)
@@ -208,7 +218,8 @@ 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)
 
@@ -223,20 +234,22 @@ const uptr kMemoryLayoutSize = sizeof(kMemoryLayout) / sizeof(kMemoryLayout[0]);
 #ifndef __clang__
 __attribute__((optimize("unroll-loops")))
 #endif
-inline bool addr_is_type(uptr addr, MappingDesc::Type mapping_type) {
+inline bool
+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)
 
diff --git a/compiler-rt/lib/msan/msan_allocator.cpp b/compiler-rt/lib/msan/msan_allocator.cpp
index 0b2dd2b2f1883d..b1bc5b9390f75b 100644
--- a/compiler-rt/lib/msan/msan_allocator.cpp
+++ b/compiler-rt/lib/msan/msan_allocator.cpp
@@ -48,6 +48,9 @@ struct MsanMapUnmapCallback {
   }
 };
 
+// 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 msan.h.
 #if defined(__mips64)
 static const uptr kMaxAllowedMallocSize = 2UL << 30;
 
diff --git a/compiler-rt/lib/msan/msan_linux.cpp b/compiler-rt/lib/msan/msan_linux.cpp
index c7ecb7cad56661..b02994c4c13353 100644
--- a/compiler-rt/lib/msan/msan_linux.cpp
+++ b/compiler-rt/lib/msan/msan_linux.cpp
@@ -86,7 +86,7 @@ static void CheckMemoryLayoutSanity() {
     CHECK(addr_is_type(start, type));
     CHECK(addr_is_type((start + end) / 2, type));
     CHECK(addr_is_type(end - 1, type));
-    if (type == MappingDesc::APP) {
+    if (type == MappingDesc::APP || type == MappingDesc::ALLOCATOR) {
       uptr addr = start;
       CHECK(MEM_IS_SHADOW(MEM_TO_SHADOW(addr)));
       CHECK(MEM_IS_ORIGIN(MEM_TO_ORIGIN(addr)));
@@ -138,8 +138,13 @@ 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 (type == MappingDesc::ALLOCATOR &&
+          !CheckMemoryRangeAvailability(start, size))
+        return false;
+    }
     if (map) {
       if (!CheckMemoryRangeAvailability(start, size))
         return false;

>From c2a57034eff048cd36c563c8e0051db3a70991b3 Mon Sep 17 00:00:00 2001
From: Thurston Dang <thurston at google.com>
Date: Fri, 15 Mar 2024 09:49:00 -0700
Subject: [PATCH 2/2] [msan] Re-exec with no ASLR if memory layout is
 incompatible on Linux (#85142)

This ports the change from TSan
(https://github.com/llvm/llvm-project/commit/0784b1eefa36d4acbb0dacd2d18796e26313b6c5).

Testing notes: run 'sudo sysctl vm.mmap_rnd_bits=32; ninja check-msan'
before and after this patch.

N.B. aggressive ASLR may also cause the app to overlap with the
allocator region; for MSan, this was fixed in
https://github.com/llvm/llvm-project/commit/af2bf86a372cacf5f536bae06e2f2d3886eefb7b

(cherry picked from commit 58f7251820b14c93168726a24816d8a094599be5)
---
 compiler-rt/lib/msan/msan.cpp       |  2 +-
 compiler-rt/lib/msan/msan.h         |  2 +-
 compiler-rt/lib/msan/msan_linux.cpp | 60 +++++++++++++++++++++++------
 3 files changed, 50 insertions(+), 14 deletions(-)

diff --git a/compiler-rt/lib/msan/msan.cpp b/compiler-rt/lib/msan/msan.cpp
index 3cdf10c149902c..a2fc27de1901b4 100644
--- a/compiler-rt/lib/msan/msan.cpp
+++ b/compiler-rt/lib/msan/msan.cpp
@@ -467,7 +467,7 @@ void __msan_init() {
   __msan_clear_on_return();
   if (__msan_get_track_origins())
     VPrintf(1, "msan_track_origins\n");
-  if (!InitShadow(__msan_get_track_origins())) {
+  if (!InitShadowWithReExec(__msan_get_track_origins())) {
     Printf("FATAL: MemorySanitizer can not mmap the shadow memory.\n");
     Printf("FATAL: Make sure to compile with -fPIE and to link with -pie.\n");
     Printf("FATAL: Disabling ASLR is known to cause this error.\n");
diff --git a/compiler-rt/lib/msan/msan.h b/compiler-rt/lib/msan/msan.h
index 00f448fb377d59..7fb58be67a02cd 100644
--- a/compiler-rt/lib/msan/msan.h
+++ b/compiler-rt/lib/msan/msan.h
@@ -263,7 +263,7 @@ extern bool msan_init_is_running;
 extern int msan_report_count;
 
 bool ProtectRange(uptr beg, uptr end);
-bool InitShadow(bool init_origins);
+bool InitShadowWithReExec(bool init_origins);
 char *GetProcSelfMaps();
 void InitializeInterceptors();
 
diff --git a/compiler-rt/lib/msan/msan_linux.cpp b/compiler-rt/lib/msan/msan_linux.cpp
index b02994c4c13353..cd2d9f5c720c57 100644
--- a/compiler-rt/lib/msan/msan_linux.cpp
+++ b/compiler-rt/lib/msan/msan_linux.cpp
@@ -20,6 +20,9 @@
 #  include <signal.h>
 #  include <stdio.h>
 #  include <stdlib.h>
+#  if SANITIZER_LINUX
+#    include <sys/personality.h>
+#  endif
 #  include <sys/resource.h>
 #  include <sys/time.h>
 #  include <unistd.h>
@@ -43,11 +46,13 @@ void ReportMapRange(const char *descr, uptr beg, uptr size) {
   }
 }
 
-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 0x%zx - 0x%zx is not available.\n", beg, end);
+      if (verbose)
+        Printf("FATAL: Memory range 0x%zx - 0x%zx is not available.\n", beg,
+               end);
       return false;
     }
   }
@@ -106,7 +111,7 @@ static void CheckMemoryLayoutSanity() {
   }
 }
 
-bool InitShadow(bool init_origins) {
+static bool InitShadow(bool init_origins, bool dry_run) {
   // Let user know mapping parameters first.
   VPrintf(1, "__msan_init %p\n", reinterpret_cast<void *>(&__msan_init));
   for (unsigned i = 0; i < kMemoryLayoutSize; ++i)
@@ -116,8 +121,9 @@ bool InitShadow(bool init_origins) {
   CheckMemoryLayoutSanity();
 
   if (!MEM_IS_APP(&__msan_init)) {
-    Printf("FATAL: Code %p is out of application range. Non-PIE build?\n",
-           reinterpret_cast<void *>(&__msan_init));
+    if (!dry_run)
+      Printf("FATAL: Code %p is out of application range. Non-PIE build?\n",
+             reinterpret_cast<void *>(&__msan_init));
     return false;
   }
 
@@ -141,22 +147,23 @@ bool InitShadow(bool init_origins) {
     if (!map && !protect) {
       CHECK(type == MappingDesc::APP || type == MappingDesc::ALLOCATOR);
 
-      if (type == MappingDesc::ALLOCATOR &&
-          !CheckMemoryRangeAvailability(start, size))
+      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;
     }
   }
@@ -164,6 +171,35 @@ bool InitShadow(bool init_origins) {
   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(__msan_get_track_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: MemorySanitizer: 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(__msan_get_track_origins(), false);
+}
+
 static void MsanAtExit(void) {
   if (flags()->print_stats && (flags()->atexit || msan_report_count > 0))
     ReportStats();



More information about the llvm-branch-commits mailing list