libsanitizer mege from upstream r171973

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@195083 138bc75d-0d04-0410-961f-82ee72b054a4

33 files changed, 1615 insertions, 201 deletions

diff --git a/libsanitizer/tsan/Makefile.am b/libsanitizer/tsan/Makefile.am
index 26d1af24d76..fa9c26cb73d 100644
--- a/libsanitizer/tsan/Makefile.am
+++ b/libsanitizer/tsan/Makefile.am
@@ -31,6 +31,9 @@ tsan_files = \
         tsan_interface_ann.cc \
         tsan_mman.cc \
         tsan_rtl_report.cc \
+	tsan_fd.cc \
+        tsan_interface_java.cc \
+        tsan_mutexset.cc \
         tsan_symbolize_addr2line_linux.cc
 
 libtsan_la_SOURCES = $(tsan_files) 
diff --git a/libsanitizer/tsan/Makefile.in b/libsanitizer/tsan/Makefile.in
index 3749819af75..c739e701c17 100644
--- a/libsanitizer/tsan/Makefile.in
+++ b/libsanitizer/tsan/Makefile.in
@@ -87,7 +87,8 @@ am__objects_1 = tsan_clock.lo tsan_interface_atomic.lo tsan_mutex.lo \
 	tsan_rtl.lo tsan_stat.lo tsan_sync.lo tsan_interceptors.lo \
 	tsan_md5.lo tsan_platform_mac.lo tsan_rtl_mutex.lo \
 	tsan_suppressions.lo tsan_interface_ann.lo tsan_mman.lo \
-	tsan_rtl_report.lo tsan_symbolize_addr2line_linux.lo
+	tsan_rtl_report.lo tsan_fd.lo tsan_interface_java.lo \
+	tsan_mutexset.lo tsan_symbolize_addr2line_linux.lo
 am_libtsan_la_OBJECTS = $(am__objects_1)
 libtsan_la_OBJECTS = $(am_libtsan_la_OBJECTS)
 libtsan_la_LINK = $(LIBTOOL) --tag=CXX $(AM_LIBTOOLFLAGS) \
@@ -273,6 +274,9 @@ tsan_files = \
         tsan_interface_ann.cc \
         tsan_mman.cc \
         tsan_rtl_report.cc \
+	tsan_fd.cc \
+        tsan_interface_java.cc \
+        tsan_mutexset.cc \
         tsan_symbolize_addr2line_linux.cc
 
 libtsan_la_SOURCES = $(tsan_files) 
@@ -393,14 +397,17 @@ distclean-compile:
 	-rm -f *.tab.c
 
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_clock.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_fd.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_flags.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_interceptors.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_interface.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_interface_ann.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_interface_atomic.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_interface_java.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_md5.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_mman.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_mutex.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_mutexset.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_platform_linux.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_platform_mac.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_report.Plo@am__quote@
diff --git a/libsanitizer/tsan/tsan_defs.h b/libsanitizer/tsan/tsan_defs.h
index c14a6d19623..6683a4e1abb 100644
--- a/libsanitizer/tsan/tsan_defs.h
+++ b/libsanitizer/tsan/tsan_defs.h
@@ -137,6 +137,12 @@ T RoundDown(T p, u64 align) {
   return (T)((u64)p & ~(align - 1));
 }
 
+// Zeroizes high part, returns 'bits' lsb bits.
+template<typename T>
+T GetLsb(T v, int bits) {
+  return (T)((u64)v & ((1ull << bits) - 1));
+}
+
 struct MD5Hash {
   u64 hash[2];
   bool operator==(const MD5Hash &other) const;
diff --git a/libsanitizer/tsan/tsan_fd.cc b/libsanitizer/tsan/tsan_fd.cc
new file mode 100644
index 00000000000..9aca9c51b38
--- /dev/null
+++ b/libsanitizer/tsan/tsan_fd.cc
@@ -0,0 +1,257 @@
+//===-- tsan_fd.cc --------------------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+
+#include "tsan_fd.h"
+#include "tsan_rtl.h"
+#include <sanitizer_common/sanitizer_atomic.h>
+
+namespace __tsan {
+
+const int kTableSizeL1 = 1024;
+const int kTableSizeL2 = 1024;
+const int kTableSize = kTableSizeL1 * kTableSizeL2;
+
+struct FdSync {
+  atomic_uint64_t rc;
+};
+
+struct FdDesc {
+  FdSync *sync;
+  int creation_tid;
+  u32 creation_stack;
+};
+
+struct FdContext {
+  atomic_uintptr_t tab[kTableSizeL1];
+  // Addresses used for synchronization.
+  FdSync globsync;
+  FdSync filesync;
+  FdSync socksync;
+  u64 connectsync;
+};
+
+static FdContext fdctx;
+
+static FdSync *allocsync() {
+  FdSync *s = (FdSync*)internal_alloc(MBlockFD, sizeof(FdSync));
+  atomic_store(&s->rc, 1, memory_order_relaxed);
+  return s;
+}
+
+static FdSync *ref(FdSync *s) {
+  if (s && atomic_load(&s->rc, memory_order_relaxed) != (u64)-1)
+    atomic_fetch_add(&s->rc, 1, memory_order_relaxed);
+  return s;
+}
+
+static void unref(ThreadState *thr, uptr pc, FdSync *s) {
+  if (s && atomic_load(&s->rc, memory_order_relaxed) != (u64)-1) {
+    if (atomic_fetch_sub(&s->rc, 1, memory_order_acq_rel) == 1) {
+      CHECK_NE(s, &fdctx.globsync);
+      CHECK_NE(s, &fdctx.filesync);
+      CHECK_NE(s, &fdctx.socksync);
+      SyncVar *v = CTX()->synctab.GetAndRemove(thr, pc, (uptr)s);
+      if (v)
+        DestroyAndFree(v);
+      internal_free(s);
+    }
+  }
+}
+
+static FdDesc *fddesc(ThreadState *thr, uptr pc, int fd) {
+  CHECK_LT(fd, kTableSize);
+  atomic_uintptr_t *pl1 = &fdctx.tab[fd / kTableSizeL2];
+  uptr l1 = atomic_load(pl1, memory_order_consume);
+  if (l1 == 0) {
+    uptr size = kTableSizeL2 * sizeof(FdDesc);
+    void *p = internal_alloc(MBlockFD, size);
+    internal_memset(p, 0, size);
+    MemoryResetRange(thr, (uptr)&fddesc, (uptr)p, size);
+    if (atomic_compare_exchange_strong(pl1, &l1, (uptr)p, memory_order_acq_rel))
+      l1 = (uptr)p;
+    else
+      internal_free(p);
+  }
+  return &((FdDesc*)l1)[fd % kTableSizeL2];  // NOLINT
+}
+
+// pd must be already ref'ed.
+static void init(ThreadState *thr, uptr pc, int fd, FdSync *s) {
+  FdDesc *d = fddesc(thr, pc, fd);
+  // As a matter of fact, we don't intercept all close calls.
+  // See e.g. libc __res_iclose().
+  if (d->sync) {
+    unref(thr, pc, d->sync);
+    d->sync = 0;
+  }
+  if (flags()->io_sync == 0) {
+    unref(thr, pc, s);
+  } else if (flags()->io_sync == 1) {
+    d->sync = s;
+  } else if (flags()->io_sync == 2) {
+    unref(thr, pc, s);
+    d->sync = &fdctx.globsync;
+  }
+  d->creation_tid = thr->tid;
+  d->creation_stack = CurrentStackId(thr, pc);
+  // To catch races between fd usage and open.
+  MemoryRangeImitateWrite(thr, pc, (uptr)d, 8);
+}
+
+void FdInit() {
+  atomic_store(&fdctx.globsync.rc, (u64)-1, memory_order_relaxed);
+  atomic_store(&fdctx.filesync.rc, (u64)-1, memory_order_relaxed);
+  atomic_store(&fdctx.socksync.rc, (u64)-1, memory_order_relaxed);
+}
+
+void FdOnFork(ThreadState *thr, uptr pc) {
+  // On fork() we need to reset all fd's, because the child is going
+  // close all them, and that will cause races between previous read/write
+  // and the close.
+  for (int l1 = 0; l1 < kTableSizeL1; l1++) {
+    FdDesc *tab = (FdDesc*)atomic_load(&fdctx.tab[l1], memory_order_relaxed);
+    if (tab == 0)
+      break;
+    for (int l2 = 0; l2 < kTableSizeL2; l2++) {
+      FdDesc *d = &tab[l2];
+      MemoryResetRange(thr, pc, (uptr)d, 8);
+    }
+  }
+}
+
+bool FdLocation(uptr addr, int *fd, int *tid, u32 *stack) {
+  for (int l1 = 0; l1 < kTableSizeL1; l1++) {
+    FdDesc *tab = (FdDesc*)atomic_load(&fdctx.tab[l1], memory_order_relaxed);
+    if (tab == 0)
+      break;
+    if (addr >= (uptr)tab && addr < (uptr)(tab + kTableSizeL2)) {
+      int l2 = (addr - (uptr)tab) / sizeof(FdDesc);
+      FdDesc *d = &tab[l2];
+      *fd = l1 * kTableSizeL1 + l2;
+      *tid = d->creation_tid;
+      *stack = d->creation_stack;
+      return true;
+    }
+  }
+  return false;
+}
+
+void FdAcquire(ThreadState *thr, uptr pc, int fd) {
+  FdDesc *d = fddesc(thr, pc, fd);
+  FdSync *s = d->sync;
+  DPrintf("#%d: FdAcquire(%d) -> %p\n", thr->tid, fd, s);
+  MemoryRead8Byte(thr, pc, (uptr)d);
+  if (s)
+    Acquire(thr, pc, (uptr)s);
+}
+
+void FdRelease(ThreadState *thr, uptr pc, int fd) {
+  FdDesc *d = fddesc(thr, pc, fd);
+  FdSync *s = d->sync;
+  DPrintf("#%d: FdRelease(%d) -> %p\n", thr->tid, fd, s);
+  if (s)
+    Release(thr, pc, (uptr)s);
+  MemoryRead8Byte(thr, pc, (uptr)d);
+}
+
+void FdClose(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdClose(%d)\n", thr->tid, fd);
+  FdDesc *d = fddesc(thr, pc, fd);
+  // To catch races between fd usage and close.
+  MemoryWrite8Byte(thr, pc, (uptr)d);
+  // We need to clear it, because if we do not intercept any call out there
+  // that creates fd, we will hit false postives.
+  MemoryResetRange(thr, pc, (uptr)d, 8);
+  unref(thr, pc, d->sync);
+  d->sync = 0;
+  d->creation_tid = 0;
+  d->creation_stack = 0;
+}
+
+void FdFileCreate(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdFileCreate(%d)\n", thr->tid, fd);
+  init(thr, pc, fd, &fdctx.filesync);
+}
+
+void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd) {
+  DPrintf("#%d: FdDup(%d, %d)\n", thr->tid, oldfd, newfd);
+  // Ignore the case when user dups not yet connected socket.
+  FdDesc *od = fddesc(thr, pc, oldfd);
+  MemoryRead8Byte(thr, pc, (uptr)od);
+  FdClose(thr, pc, newfd);
+  init(thr, pc, newfd, ref(od->sync));
+}
+
+void FdPipeCreate(ThreadState *thr, uptr pc, int rfd, int wfd) {
+  DPrintf("#%d: FdCreatePipe(%d, %d)\n", thr->tid, rfd, wfd);
+  FdSync *s = allocsync();
+  init(thr, pc, rfd, ref(s));
+  init(thr, pc, wfd, ref(s));
+  unref(thr, pc, s);
+}
+
+void FdEventCreate(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdEventCreate(%d)\n", thr->tid, fd);
+  init(thr, pc, fd, allocsync());
+}
+
+void FdSignalCreate(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdSignalCreate(%d)\n", thr->tid, fd);
+  init(thr, pc, fd, 0);
+}
+
+void FdInotifyCreate(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdInotifyCreate(%d)\n", thr->tid, fd);
+  init(thr, pc, fd, 0);
+}
+
+void FdPollCreate(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdPollCreate(%d)\n", thr->tid, fd);
+  init(thr, pc, fd, allocsync());
+}
+
+void FdSocketCreate(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdSocketCreate(%d)\n", thr->tid, fd);
+  // It can be a UDP socket.
+  init(thr, pc, fd, &fdctx.socksync);
+}
+
+void FdSocketAccept(ThreadState *thr, uptr pc, int fd, int newfd) {
+  DPrintf("#%d: FdSocketAccept(%d, %d)\n", thr->tid, fd, newfd);
+  // Synchronize connect->accept.
+  Acquire(thr, pc, (uptr)&fdctx.connectsync);
+  init(thr, pc, newfd, &fdctx.socksync);
+}
+
+void FdSocketConnecting(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdSocketConnecting(%d)\n", thr->tid, fd);
+  // Synchronize connect->accept.
+  Release(thr, pc, (uptr)&fdctx.connectsync);
+}
+
+void FdSocketConnect(ThreadState *thr, uptr pc, int fd) {
+  DPrintf("#%d: FdSocketConnect(%d)\n", thr->tid, fd);
+  init(thr, pc, fd, &fdctx.socksync);
+}
+
+uptr File2addr(char *path) {
+  (void)path;
+  static u64 addr;
+  return (uptr)&addr;
+}
+
+uptr Dir2addr(char *path) {
+  (void)path;
+  static u64 addr;
+  return (uptr)&addr;
+}
+
+}  //  namespace __tsan
diff --git a/libsanitizer/tsan/tsan_fd.h b/libsanitizer/tsan/tsan_fd.h
new file mode 100644
index 00000000000..b4189a37df5
--- /dev/null
+++ b/libsanitizer/tsan/tsan_fd.h
@@ -0,0 +1,62 @@
+//===-- tsan_fd.h -----------------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+// This file handles synchronization via IO.
+// People use IO for synchronization along the lines of:
+//
+// int X;
+// int client_socket;  // initialized elsewhere
+// int server_socket;  // initialized elsewhere
+//
+// Thread 1:
+// X = 42;
+// send(client_socket, ...);
+//
+// Thread 2:
+// if (recv(server_socket, ...) > 0)
+//   assert(X == 42);
+//
+// This file determines the scope of the file descriptor (pipe, socket,
+// all local files, etc) and executes acquire and release operations on
+// the scope as necessary.  Some scopes are very fine grained (e.g. pipe
+// operations synchronize only with operations on the same pipe), while
+// others are corse-grained (e.g. all operations on local files synchronize
+// with each other).
+//===----------------------------------------------------------------------===//
+#ifndef TSAN_FD_H
+#define TSAN_FD_H
+
+#include "tsan_rtl.h"
+
+namespace __tsan {
+
+void FdInit();
+void FdAcquire(ThreadState *thr, uptr pc, int fd);
+void FdRelease(ThreadState *thr, uptr pc, int fd);
+void FdClose(ThreadState *thr, uptr pc, int fd);
+void FdFileCreate(ThreadState *thr, uptr pc, int fd);
+void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd);
+void FdPipeCreate(ThreadState *thr, uptr pc, int rfd, int wfd);
+void FdEventCreate(ThreadState *thr, uptr pc, int fd);
+void FdSignalCreate(ThreadState *thr, uptr pc, int fd);
+void FdInotifyCreate(ThreadState *thr, uptr pc, int fd);
+void FdPollCreate(ThreadState *thr, uptr pc, int fd);
+void FdSocketCreate(ThreadState *thr, uptr pc, int fd);
+void FdSocketAccept(ThreadState *thr, uptr pc, int fd, int newfd);
+void FdSocketConnecting(ThreadState *thr, uptr pc, int fd);
+void FdSocketConnect(ThreadState *thr, uptr pc, int fd);
+bool FdLocation(uptr addr, int *fd, int *tid, u32 *stack);
+void FdOnFork(ThreadState *thr, uptr pc);
+
+uptr File2addr(char *path);
+uptr Dir2addr(char *path);
+
+}  // namespace __tsan
+
+#endif  // TSAN_INTERFACE_H
diff --git a/libsanitizer/tsan/tsan_flags.cc b/libsanitizer/tsan/tsan_flags.cc
index 1b726e6c8de..630bd75769b 100644
--- a/libsanitizer/tsan/tsan_flags.cc
+++ b/libsanitizer/tsan/tsan_flags.cc
@@ -56,6 +56,7 @@ void InitializeFlags(Flags *f, const char *env) {
   f->running_on_valgrind = false;
   f->external_symbolizer_path = "";
   f->history_size = kGoMode ? 1 : 2;  // There are a lot of goroutines in Go.
+  f->io_sync = 1;
 
   // Let a frontend override.
   OverrideFlags(f);
@@ -81,6 +82,7 @@ void InitializeFlags(Flags *f, const char *env) {
   ParseFlag(env, &f->stop_on_start, "stop_on_start");
   ParseFlag(env, &f->external_symbolizer_path, "external_symbolizer_path");
   ParseFlag(env, &f->history_size, "history_size");
+  ParseFlag(env, &f->io_sync, "io_sync");
 
   if (!f->report_bugs) {
     f->report_thread_leaks = false;
@@ -93,6 +95,12 @@ void InitializeFlags(Flags *f, const char *env) {
            " (must be [0..7])\n");
     Die();
   }
+
+  if (f->io_sync < 0 || f->io_sync > 2) {
+    Printf("ThreadSanitizer: incorrect value for io_sync"
+           " (must be [0..2])\n");
+    Die();
+  }
 }
 
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_flags.h b/libsanitizer/tsan/tsan_flags.h
index 6af96ec83be..ed27363c2ff 100644
--- a/libsanitizer/tsan/tsan_flags.h
+++ b/libsanitizer/tsan/tsan_flags.h
@@ -75,6 +75,11 @@ struct Flags {
   // the amount of memory accesses, up to history_size=7 that amounts to
   // 4M memory accesses.  The default value is 2 (128K memory accesses).
   int history_size;
+  // Controls level of synchronization implied by IO operations.
+  // 0 - no synchronization
+  // 1 - reasonable level of synchronization (write->read)
+  // 2 - global synchronization of all IO operations
+  int io_sync;
 };
 
 Flags *flags();
diff --git a/libsanitizer/tsan/tsan_interceptors.cc b/libsanitizer/tsan/tsan_interceptors.cc
index dea64150786..88acebf8e81 100644
--- a/libsanitizer/tsan/tsan_interceptors.cc
+++ b/libsanitizer/tsan/tsan_interceptors.cc
@@ -7,6 +7,8 @@
 //
 // This file is a part of ThreadSanitizer (TSan), a race detector.
 //
+// FIXME: move as many interceptors as possible into
+// sanitizer_common/sanitizer_common_interceptors.h
 //===----------------------------------------------------------------------===//
 
 #include "sanitizer_common/sanitizer_atomic.h"
@@ -18,6 +20,7 @@
 #include "tsan_platform.h"
 #include "tsan_rtl.h"
 #include "tsan_mman.h"
+#include "tsan_fd.h"
 
 using namespace __tsan;  // NOLINT
 
@@ -50,6 +53,7 @@ extern "C" void *pthread_self();
 extern "C" void _exit(int status);
 extern "C" int __cxa_atexit(void (*func)(void *arg), void *arg, void *dso);
 extern "C" int *__errno_location();
+extern "C" int fileno_unlocked(void *stream);
 const int PTHREAD_MUTEX_RECURSIVE = 1;
 const int PTHREAD_MUTEX_RECURSIVE_NP = 1;
 const int kPthreadAttrSize = 56;
@@ -124,10 +128,8 @@ static SignalContext *SigCtx(ThreadState *thr) {
   SignalContext *ctx = (SignalContext*)thr->signal_ctx;
   if (ctx == 0 && thr->is_alive) {
     ScopedInRtl in_rtl;
-    ctx = (SignalContext*)internal_alloc(
-        MBlockSignal, sizeof(*ctx));
-    MemoryResetRange(thr, 0, (uptr)ctx, sizeof(*ctx));
-    internal_memset(ctx, 0, sizeof(*ctx));
+    ctx = (SignalContext*)MmapOrDie(sizeof(*ctx), "SignalContext");
+    MemoryResetRange(thr, (uptr)&SigCtx, (uptr)ctx, sizeof(*ctx));
     thr->signal_ctx = ctx;
   }
   return ctx;
@@ -173,8 +175,8 @@ ScopedInterceptor::~ScopedInterceptor() {
     StatInc(thr, StatInt_##func); \
     const uptr caller_pc = GET_CALLER_PC(); \
     ScopedInterceptor si(thr, #func, caller_pc); \
-    /* Subtract one from pc as we need current instruction address */ \
-    const uptr pc = __sanitizer::StackTrace::GetCurrentPc() - 1; \
+    const uptr pc = __sanitizer::StackTrace::GetPreviousInstructionPc( \
+        __sanitizer::StackTrace::GetCurrentPc()); \
     (void)pc; \
 /**/
 
@@ -306,30 +308,6 @@ TSAN_INTERCEPTOR(void, siglongjmp, void *env, int val) {
   Die();
 }
 
-static uptr fd2addr(int fd) {
-  (void)fd;
-  static u64 addr;
-  return (uptr)&addr;
-}
-
-static uptr epollfd2addr(int fd) {
-  (void)fd;
-  static u64 addr;
-  return (uptr)&addr;
-}
-
-static uptr file2addr(char *path) {
-  (void)path;
-  static u64 addr;
-  return (uptr)&addr;
-}
-
-static uptr dir2addr(char *path) {
-  (void)path;
-  static u64 addr;
-  return (uptr)&addr;
-}
-
 TSAN_INTERCEPTOR(void*, malloc, uptr size) {
   void *p = 0;
   {
@@ -660,7 +638,7 @@ static void thread_finalize(void *v) {
     SignalContext *sctx = thr->signal_ctx;
     if (sctx) {
       thr->signal_ctx = 0;
-      internal_free(sctx);
+      UnmapOrDie(sctx, sizeof(*sctx));
     }
   }
 }
@@ -934,11 +912,15 @@ TSAN_INTERCEPTOR(int, pthread_rwlock_unlock, void *m) {
   return res;
 }
 
+// libpthread.so contains several versions of pthread_cond_init symbol.
+// When we just dlsym() it, we get the wrong (old) version.
+/*
 TSAN_INTERCEPTOR(int, pthread_cond_init, void *c, void *a) {
   SCOPED_TSAN_INTERCEPTOR(pthread_cond_init, c, a);
   int res = REAL(pthread_cond_init)(c, a);
   return res;
 }
+*/
 
 TSAN_INTERCEPTOR(int, pthread_cond_destroy, void *c) {
   SCOPED_TSAN_INTERCEPTOR(pthread_cond_destroy, c);
@@ -1080,11 +1062,188 @@ TSAN_INTERCEPTOR(int, sem_getvalue, void *s, int *sval) {
   return res;
 }
 
+TSAN_INTERCEPTOR(int, open, const char *name, int flags, int mode) {
+  SCOPED_TSAN_INTERCEPTOR(open, name, flags, mode);
+  int fd = REAL(open)(name, flags, mode);
+  if (fd >= 0)
+    FdFileCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, open64, const char *name, int flags, int mode) {
+  SCOPED_TSAN_INTERCEPTOR(open64, name, flags, mode);
+  int fd = REAL(open64)(name, flags, mode);
+  if (fd >= 0)
+    FdFileCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, creat, const char *name, int mode) {
+  SCOPED_TSAN_INTERCEPTOR(creat, name, mode);
+  int fd = REAL(creat)(name, mode);
+  if (fd >= 0)
+    FdFileCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, creat64, const char *name, int mode) {
+  SCOPED_TSAN_INTERCEPTOR(creat64, name, mode);
+  int fd = REAL(creat64)(name, mode);
+  if (fd >= 0)
+    FdFileCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, dup, int oldfd) {
+  SCOPED_TSAN_INTERCEPTOR(dup, oldfd);
+  int newfd = REAL(dup)(oldfd);
+  if (oldfd >= 0 && newfd >= 0 && newfd != oldfd)
+    FdDup(thr, pc, oldfd, newfd);
+  return newfd;
+}
+
+TSAN_INTERCEPTOR(int, dup2, int oldfd, int newfd) {
+  SCOPED_TSAN_INTERCEPTOR(dup2, oldfd, newfd);
+  int newfd2 = REAL(dup2)(oldfd, newfd);
+  if (oldfd >= 0 && newfd2 >= 0 && newfd2 != oldfd)
+    FdDup(thr, pc, oldfd, newfd2);
+  return newfd2;
+}
+
+TSAN_INTERCEPTOR(int, dup3, int oldfd, int newfd, int flags) {
+  SCOPED_TSAN_INTERCEPTOR(dup3, oldfd, newfd, flags);
+  int newfd2 = REAL(dup3)(oldfd, newfd, flags);
+  if (oldfd >= 0 && newfd2 >= 0 && newfd2 != oldfd)
+    FdDup(thr, pc, oldfd, newfd2);
+  return newfd2;
+}
+
+TSAN_INTERCEPTOR(int, eventfd, unsigned initval, int flags) {
+  SCOPED_TSAN_INTERCEPTOR(eventfd, initval, flags);
+  int fd = REAL(eventfd)(initval, flags);
+  if (fd >= 0)
+    FdEventCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, signalfd, int fd, void *mask, int flags) {
+  SCOPED_TSAN_INTERCEPTOR(signalfd, fd, mask, flags);
+  if (fd >= 0)
+    FdClose(thr, pc, fd);
+  fd = REAL(signalfd)(fd, mask, flags);
+  if (fd >= 0)
+    FdSignalCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, inotify_init, int fake) {
+  SCOPED_TSAN_INTERCEPTOR(inotify_init, fake);
+  int fd = REAL(inotify_init)(fake);
+  if (fd >= 0)
+    FdInotifyCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, inotify_init1, int flags) {
+  SCOPED_TSAN_INTERCEPTOR(inotify_init1, flags);
+  int fd = REAL(inotify_init1)(flags);
+  if (fd >= 0)
+    FdInotifyCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, socket, int domain, int type, int protocol) {
+  SCOPED_TSAN_INTERCEPTOR(socket, domain, type, protocol);
+  int fd = REAL(socket)(domain, type, protocol);
+  if (fd >= 0)
+    FdSocketCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, socketpair, int domain, int type, int protocol, int *fd) {
+  SCOPED_TSAN_INTERCEPTOR(socketpair, domain, type, protocol, fd);
+  int res = REAL(socketpair)(domain, type, protocol, fd);
+  if (res == 0 && fd[0] >= 0 && fd[1] >= 0)
+    FdPipeCreate(thr, pc, fd[0], fd[1]);
+  return res;
+}
+
+TSAN_INTERCEPTOR(int, connect, int fd, void *addr, unsigned addrlen) {
+  SCOPED_TSAN_INTERCEPTOR(connect, fd, addr, addrlen);
+  FdSocketConnecting(thr, pc, fd);
+  int res = REAL(connect)(fd, addr, addrlen);
+  if (res == 0 && fd >= 0)
+    FdSocketConnect(thr, pc, fd);
+  return res;
+}
+
+TSAN_INTERCEPTOR(int, accept, int fd, void *addr, unsigned *addrlen) {
+  SCOPED_TSAN_INTERCEPTOR(accept, fd, addr, addrlen);
+  int fd2 = REAL(accept)(fd, addr, addrlen);
+  if (fd >= 0 && fd2 >= 0)
+    FdSocketAccept(thr, pc, fd, fd2);
+  return fd2;
+}
+
+TSAN_INTERCEPTOR(int, accept4, int fd, void *addr, unsigned *addrlen, int f) {
+  SCOPED_TSAN_INTERCEPTOR(accept4, fd, addr, addrlen, f);
+  int fd2 = REAL(accept4)(fd, addr, addrlen, f);
+  if (fd >= 0 && fd2 >= 0)
+    FdSocketAccept(thr, pc, fd, fd2);
+  return fd2;
+}
+
+TSAN_INTERCEPTOR(int, epoll_create, int size) {
+  SCOPED_TSAN_INTERCEPTOR(epoll_create, size);
+  int fd = REAL(epoll_create)(size);
+  if (fd >= 0)
+    FdPollCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, epoll_create1, int flags) {
+  SCOPED_TSAN_INTERCEPTOR(epoll_create1, flags);
+  int fd = REAL(epoll_create1)(flags);
+  if (fd >= 0)
+    FdPollCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, close, int fd) {
+  SCOPED_TSAN_INTERCEPTOR(close, fd);
+  if (fd >= 0)
+    FdClose(thr, pc, fd);
+  return REAL(close)(fd);
+}
+
+TSAN_INTERCEPTOR(int, __close, int fd) {
+  SCOPED_TSAN_INTERCEPTOR(__close, fd);
+  if (fd >= 0)
+    FdClose(thr, pc, fd);
+  return REAL(__close)(fd);
+}
+
+TSAN_INTERCEPTOR(int, pipe, int *pipefd) {
+  SCOPED_TSAN_INTERCEPTOR(pipe, pipefd);
+  int res = REAL(pipe)(pipefd);
+  if (res == 0 && pipefd[0] >= 0 && pipefd[1] >= 0)
+    FdPipeCreate(thr, pc, pipefd[0], pipefd[1]);
+  return res;
+}
+
+TSAN_INTERCEPTOR(int, pipe2, int *pipefd, int flags) {
+  SCOPED_TSAN_INTERCEPTOR(pipe2, pipefd, flags);
+  int res = REAL(pipe2)(pipefd, flags);
+  if (res == 0 && pipefd[0] >= 0 && pipefd[1] >= 0)
+    FdPipeCreate(thr, pc, pipefd[0], pipefd[1]);
+  return res;
+}
+
 TSAN_INTERCEPTOR(long_t, read, int fd, void *buf, long_t sz) {
   SCOPED_TSAN_INTERCEPTOR(read, fd, buf, sz);
   int res = REAL(read)(fd, buf, sz);
-  if (res >= 0) {
-    Acquire(thr, pc, fd2addr(fd));
+  if (res >= 0 && fd >= 0) {
+    FdAcquire(thr, pc, fd);
   }
   return res;
 }
@@ -1092,8 +1251,8 @@ TSAN_INTERCEPTOR(long_t, read, int fd, void *buf, long_t sz) {
 TSAN_INTERCEPTOR(long_t, pread, int fd, void *buf, long_t sz, unsigned off) {
   SCOPED_TSAN_INTERCEPTOR(pread, fd, buf, sz, off);
   int res = REAL(pread)(fd, buf, sz, off);
-  if (res >= 0) {
-    Acquire(thr, pc, fd2addr(fd));
+  if (res >= 0 && fd >= 0) {
+    FdAcquire(thr, pc, fd);
   }
   return res;
 }
@@ -1101,8 +1260,8 @@ TSAN_INTERCEPTOR(long_t, pread, int fd, void *buf, long_t sz, unsigned off) {
 TSAN_INTERCEPTOR(long_t, pread64, int fd, void *buf, long_t sz, u64 off) {
   SCOPED_TSAN_INTERCEPTOR(pread64, fd, buf, sz, off);
   int res = REAL(pread64)(fd, buf, sz, off);
-  if (res >= 0) {
-    Acquire(thr, pc, fd2addr(fd));
+  if (res >= 0 && fd >= 0) {
+    FdAcquire(thr, pc, fd);
   }
   return res;
 }
@@ -1110,8 +1269,8 @@ TSAN_INTERCEPTOR(long_t, pread64, int fd, void *buf, long_t sz, u64 off) {
 TSAN_INTERCEPTOR(long_t, readv, int fd, void *vec, int cnt) {
   SCOPED_TSAN_INTERCEPTOR(readv, fd, vec, cnt);
   int res = REAL(readv)(fd, vec, cnt);
-  if (res >= 0) {
-    Acquire(thr, pc, fd2addr(fd));
+  if (res >= 0 && fd >= 0) {
+    FdAcquire(thr, pc, fd);
   }
   return res;
 }
@@ -1119,57 +1278,64 @@ TSAN_INTERCEPTOR(long_t, readv, int fd, void *vec, int cnt) {
 TSAN_INTERCEPTOR(long_t, preadv64, int fd, void *vec, int cnt, u64 off) {
   SCOPED_TSAN_INTERCEPTOR(preadv64, fd, vec, cnt, off);
   int res = REAL(preadv64)(fd, vec, cnt, off);
-  if (res >= 0) {
-    Acquire(thr, pc, fd2addr(fd));
+  if (res >= 0 && fd >= 0) {
+    FdAcquire(thr, pc, fd);
   }
   return res;
 }
 
 TSAN_INTERCEPTOR(long_t, write, int fd, void *buf, long_t sz) {
   SCOPED_TSAN_INTERCEPTOR(write, fd, buf, sz);
-  Release(thr, pc, fd2addr(fd));
+  if (fd >= 0)
+    FdRelease(thr, pc, fd);
   int res = REAL(write)(fd, buf, sz);
   return res;
 }
 
 TSAN_INTERCEPTOR(long_t, pwrite, int fd, void *buf, long_t sz, unsigned off) {
   SCOPED_TSAN_INTERCEPTOR(pwrite, fd, buf, sz, off);
-  Release(thr, pc, fd2addr(fd));
+  if (fd >= 0)
+    FdRelease(thr, pc, fd);
   int res = REAL(pwrite)(fd, buf, sz, off);
   return res;
 }
 
 TSAN_INTERCEPTOR(long_t, pwrite64, int fd, void *buf, long_t sz, u64 off) {
   SCOPED_TSAN_INTERCEPTOR(pwrite64, fd, buf, sz, off);
-  Release(thr, pc, fd2addr(fd));
+  if (fd >= 0)
+    FdRelease(thr, pc, fd);
   int res = REAL(pwrite64)(fd, buf, sz, off);
   return res;
 }
 
 TSAN_INTERCEPTOR(long_t, writev, int fd, void *vec, int cnt) {
   SCOPED_TSAN_INTERCEPTOR(writev, fd, vec, cnt);
-  Release(thr, pc, fd2addr(fd));
+  if (fd >= 0)
+    FdRelease(thr, pc, fd);
   int res = REAL(writev)(fd, vec, cnt);
   return res;
 }
 
 TSAN_INTERCEPTOR(long_t, pwritev64, int fd, void *vec, int cnt, u64 off) {
   SCOPED_TSAN_INTERCEPTOR(pwritev64, fd, vec, cnt, off);
-  Release(thr, pc, fd2addr(fd));
+  if (fd >= 0)
+    FdRelease(thr, pc, fd);
   int res = REAL(pwritev64)(fd, vec, cnt, off);
   return res;
 }
 
 TSAN_INTERCEPTOR(long_t, send, int fd, void *buf, long_t len, int flags) {
   SCOPED_TSAN_INTERCEPTOR(send, fd, buf, len, flags);
-  Release(thr, pc, fd2addr(fd));
+  if (fd >= 0)
+    FdRelease(thr, pc, fd);
   int res = REAL(send)(fd, buf, len, flags);
   return res;
 }
 
 TSAN_INTERCEPTOR(long_t, sendmsg, int fd, void *msg, int flags) {
   SCOPED_TSAN_INTERCEPTOR(sendmsg, fd, msg, flags);
-  Release(thr, pc, fd2addr(fd));
+  if (fd >= 0)
+    FdRelease(thr, pc, fd);
   int res = REAL(sendmsg)(fd, msg, flags);
   return res;
 }
@@ -1177,8 +1343,8 @@ TSAN_INTERCEPTOR(long_t, sendmsg, int fd, void *msg, int flags) {
 TSAN_INTERCEPTOR(long_t, recv, int fd, void *buf, long_t len, int flags) {
   SCOPED_TSAN_INTERCEPTOR(recv, fd, buf, len, flags);
   int res = REAL(recv)(fd, buf, len, flags);
-  if (res >= 0) {
-    Acquire(thr, pc, fd2addr(fd));
+  if (res >= 0 && fd >= 0) {
+    FdAcquire(thr, pc, fd);
   }
   return res;
 }
@@ -1186,15 +1352,15 @@ TSAN_INTERCEPTOR(long_t, recv, int fd, void *buf, long_t len, int flags) {
 TSAN_INTERCEPTOR(long_t, recvmsg, int fd, void *msg, int flags) {
   SCOPED_TSAN_INTERCEPTOR(recvmsg, fd, msg, flags);
   int res = REAL(recvmsg)(fd, msg, flags);
-  if (res >= 0) {
-    Acquire(thr, pc, fd2addr(fd));
+  if (res >= 0 && fd >= 0) {
+    FdAcquire(thr, pc, fd);
   }
   return res;
 }
 
 TSAN_INTERCEPTOR(int, unlink, char *path) {
   SCOPED_TSAN_INTERCEPTOR(unlink, path);
-  Release(thr, pc, file2addr(path));
+  Release(thr, pc, File2addr(path));
   int res = REAL(unlink)(path);
   return res;
 }
@@ -1202,19 +1368,57 @@ TSAN_INTERCEPTOR(int, unlink, char *path) {
 TSAN_INTERCEPTOR(void*, fopen, char *path, char *mode) {
   SCOPED_TSAN_INTERCEPTOR(fopen, path, mode);
   void *res = REAL(fopen)(path, mode);
-  Acquire(thr, pc, file2addr(path));
+  Acquire(thr, pc, File2addr(path));
+  if (res) {
+    int fd = fileno_unlocked(res);
+    if (fd >= 0)
+      FdFileCreate(thr, pc, fd);
+  }
   return res;
 }
 
+TSAN_INTERCEPTOR(void*, freopen, char *path, char *mode, void *stream) {
+  SCOPED_TSAN_INTERCEPTOR(freopen, path, mode, stream);
+  if (stream) {
+    int fd = fileno_unlocked(stream);
+    if (fd >= 0)
+      FdClose(thr, pc, fd);
+  }
+  void *res = REAL(freopen)(path, mode, stream);
+  Acquire(thr, pc, File2addr(path));
+  if (res) {
+    int fd = fileno_unlocked(res);
+    if (fd >= 0)
+      FdFileCreate(thr, pc, fd);
+  }
+  return res;
+}
+
+TSAN_INTERCEPTOR(int, fclose, void *stream) {
+  {
+    SCOPED_TSAN_INTERCEPTOR(fclose, stream);
+    if (stream) {
+      int fd = fileno_unlocked(stream);
+      if (fd >= 0)
+        FdClose(thr, pc, fd);
+    }
+  }
+  return REAL(fclose)(stream);
+}
+
 TSAN_INTERCEPTOR(uptr, fread, void *ptr, uptr size, uptr nmemb, void *f) {
-  SCOPED_TSAN_INTERCEPTOR(fread, ptr, size, nmemb, f);
-  MemoryAccessRange(thr, pc, (uptr)ptr, size * nmemb, true);
+  {
+    SCOPED_TSAN_INTERCEPTOR(fread, ptr, size, nmemb, f);
+    MemoryAccessRange(thr, pc, (uptr)ptr, size * nmemb, true);
+  }
   return REAL(fread)(ptr, size, nmemb, f);
 }
 
 TSAN_INTERCEPTOR(uptr, fwrite, const void *p, uptr size, uptr nmemb, void *f) {
-  SCOPED_TSAN_INTERCEPTOR(fwrite, p, size, nmemb, f);
-  MemoryAccessRange(thr, pc, (uptr)p, size * nmemb, false);
+  {
+    SCOPED_TSAN_INTERCEPTOR(fwrite, p, size, nmemb, f);
+    MemoryAccessRange(thr, pc, (uptr)p, size * nmemb, false);
+  }
   return REAL(fwrite)(p, size, nmemb, f);
 }
 
@@ -1226,7 +1430,7 @@ TSAN_INTERCEPTOR(int, puts, const char *s) {
 
 TSAN_INTERCEPTOR(int, rmdir, char *path) {
   SCOPED_TSAN_INTERCEPTOR(rmdir, path);
-  Release(thr, pc, dir2addr(path));
+  Release(thr, pc, Dir2addr(path));
   int res = REAL(rmdir)(path);
   return res;
 }
@@ -1234,14 +1438,15 @@ TSAN_INTERCEPTOR(int, rmdir, char *path) {
 TSAN_INTERCEPTOR(void*, opendir, char *path) {
   SCOPED_TSAN_INTERCEPTOR(opendir, path);
   void *res = REAL(opendir)(path);
-  Acquire(thr, pc, dir2addr(path));
+  if (res != 0)
+    Acquire(thr, pc, Dir2addr(path));
   return res;
 }
 
 TSAN_INTERCEPTOR(int, epoll_ctl, int epfd, int op, int fd, void *ev) {
   SCOPED_TSAN_INTERCEPTOR(epoll_ctl, epfd, op, fd, ev);
-  if (op == EPOLL_CTL_ADD) {
-    Release(thr, pc, epollfd2addr(epfd));
+  if (op == EPOLL_CTL_ADD && epfd >= 0) {
+    FdRelease(thr, pc, epfd);
   }
   int res = REAL(epoll_ctl)(epfd, op, fd, ev);
   return res;
@@ -1250,8 +1455,8 @@ TSAN_INTERCEPTOR(int, epoll_ctl, int epfd, int op, int fd, void *ev) {
 TSAN_INTERCEPTOR(int, epoll_wait, int epfd, void *ev, int cnt, int timeout) {
   SCOPED_TSAN_INTERCEPTOR(epoll_wait, epfd, ev, cnt, timeout);
   int res = BLOCK_REAL(epoll_wait)(epfd, ev, cnt, timeout);
-  if (res > 0) {
-    Acquire(thr, pc, epollfd2addr(epfd));
+  if (res > 0 && epfd >= 0) {
+    FdAcquire(thr, pc, epfd);
   }
   return res;
 }
@@ -1423,6 +1628,19 @@ TSAN_INTERCEPTOR(int, munlockall, void) {
   return 0;
 }
 
+TSAN_INTERCEPTOR(int, fork, int fake) {
+  SCOPED_TSAN_INTERCEPTOR(fork, fake);
+  // It's intercepted merely to process pending signals.
+  int pid = REAL(fork)(fake);
+  if (pid == 0) {
+    // child
+    FdOnFork(thr, pc);
+  } else if (pid > 0) {
+    // parent
+  }
+  return pid;
+}
+
 namespace __tsan {
 
 void ProcessPendingSignals(ThreadState *thr) {
@@ -1545,7 +1763,7 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(pthread_rwlock_timedwrlock);
   TSAN_INTERCEPT(pthread_rwlock_unlock);
 
-  TSAN_INTERCEPT(pthread_cond_init);
+  // TSAN_INTERCEPT(pthread_cond_init);
   TSAN_INTERCEPT(pthread_cond_destroy);
   TSAN_INTERCEPT(pthread_cond_signal);
   TSAN_INTERCEPT(pthread_cond_broadcast);
@@ -1566,6 +1784,28 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(sem_post);
   TSAN_INTERCEPT(sem_getvalue);
 
+  TSAN_INTERCEPT(open);
+  TSAN_INTERCEPT(open64);
+  TSAN_INTERCEPT(creat);
+  TSAN_INTERCEPT(creat64);
+  TSAN_INTERCEPT(dup);
+  TSAN_INTERCEPT(dup2);
+  TSAN_INTERCEPT(dup3);
+  TSAN_INTERCEPT(eventfd);
+  TSAN_INTERCEPT(signalfd);
+  TSAN_INTERCEPT(inotify_init);
+  TSAN_INTERCEPT(inotify_init1);
+  TSAN_INTERCEPT(socket);
+  TSAN_INTERCEPT(socketpair);
+  TSAN_INTERCEPT(connect);
+  TSAN_INTERCEPT(accept);
+  TSAN_INTERCEPT(accept4);
+  TSAN_INTERCEPT(epoll_create);
+  TSAN_INTERCEPT(epoll_create1);
+  TSAN_INTERCEPT(close);
+  TSAN_INTERCEPT(pipe);
+  TSAN_INTERCEPT(pipe2);
+
   TSAN_INTERCEPT(read);
   TSAN_INTERCEPT(pread);
   TSAN_INTERCEPT(pread64);
@@ -1583,6 +1823,8 @@ void InitializeInterceptors() {
 
   TSAN_INTERCEPT(unlink);
   TSAN_INTERCEPT(fopen);
+  TSAN_INTERCEPT(freopen);
+  TSAN_INTERCEPT(fclose);
   TSAN_INTERCEPT(fread);
   TSAN_INTERCEPT(fwrite);
   TSAN_INTERCEPT(puts);
@@ -1608,6 +1850,8 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(mlockall);
   TSAN_INTERCEPT(munlockall);
 
+  TSAN_INTERCEPT(fork);
+
   // Need to setup it, because interceptors check that the function is resolved.
   // But atexit is emitted directly into the module, so can't be resolved.
   REAL(atexit) = (int(*)(void(*)()))unreachable;
@@ -1623,6 +1867,8 @@ void InitializeInterceptors() {
     Printf("ThreadSanitizer: failed to create thread key\n");
     Die();
   }
+
+  FdInit();
 }
 
 void internal_start_thread(void(*func)(void *arg), void *arg) {
diff --git a/libsanitizer/tsan/tsan_interface_atomic.cc b/libsanitizer/tsan/tsan_interface_atomic.cc
index 63860edcfbb..770f8bd1014 100644
--- a/libsanitizer/tsan/tsan_interface_atomic.cc
+++ b/libsanitizer/tsan/tsan_interface_atomic.cc
@@ -229,7 +229,7 @@ static T AtomicLoad(ThreadState *thr, uptr pc, const volatile T *a,
   // Assume the access is atomic.
   if (!IsAcquireOrder(mo) && sizeof(T) <= sizeof(a))
     return *a;
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, (uptr)a, false);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, false);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   thr->clock.acquire(&s->clock);
   T v = *a;
@@ -251,7 +251,7 @@ static void AtomicStore(ThreadState *thr, uptr pc, volatile T *a, T v,
     return;
   }
   __sync_synchronize();
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, (uptr)a, true);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, true);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   thr->clock.ReleaseStore(&s->clock);
   *a = v;
@@ -263,7 +263,7 @@ static void AtomicStore(ThreadState *thr, uptr pc, volatile T *a, T v,
 
 template<typename T, T (*F)(volatile T *v, T op)>
 static T AtomicRMW(ThreadState *thr, uptr pc, volatile T *a, T v, morder mo) {
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, (uptr)a, true);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, true);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   if (IsAcqRelOrder(mo))
     thr->clock.acq_rel(&s->clock);
@@ -322,7 +322,7 @@ template<typename T>
 static bool AtomicCAS(ThreadState *thr, uptr pc,
     volatile T *a, T *c, T v, morder mo, morder fmo) {
   (void)fmo;  // Unused because llvm does not pass it yet.
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, (uptr)a, true);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, true);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   if (IsAcqRelOrder(mo))
     thr->clock.acq_rel(&s->clock);
diff --git a/libsanitizer/tsan/tsan_interface_atomic.h b/libsanitizer/tsan/tsan_interface_atomic.h
index 9db31da0638..c500614acc4 100644
--- a/libsanitizer/tsan/tsan_interface_atomic.h
+++ b/libsanitizer/tsan/tsan_interface_atomic.h
@@ -26,7 +26,7 @@ typedef long     __tsan_atomic64;  // NOLINT
 
 #if defined(__SIZEOF_INT128__) \
     || (__clang_major__ * 100 + __clang_minor__ >= 302)
-typedef __int128 __tsan_atomic128;
+__extension__ typedef __int128 __tsan_atomic128;
 #define __TSAN_HAS_INT128 1
 #else
 typedef char     __tsan_atomic128;
diff --git a/libsanitizer/tsan/tsan_interface_java.cc b/libsanitizer/tsan/tsan_interface_java.cc
new file mode 100644
index 00000000000..d7325dcb2c4
--- /dev/null
+++ b/libsanitizer/tsan/tsan_interface_java.cc
@@ -0,0 +1,303 @@
+//===-- tsan_interface_java.cc --------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+
+#include "tsan_interface_java.h"
+#include "tsan_rtl.h"
+#include "tsan_mutex.h"
+#include "sanitizer_common/sanitizer_internal_defs.h"
+#include "sanitizer_common/sanitizer_common.h"
+#include "sanitizer_common/sanitizer_placement_new.h"
+
+using namespace __tsan;  // NOLINT
+
+namespace __tsan {
+
+const uptr kHeapShadow = 0x300000000000ull;
+const uptr kHeapAlignment = 8;
+
+struct BlockDesc {
+  bool begin;
+  Mutex mtx;
+  SyncVar *head;
+
+  BlockDesc()
+      : mtx(MutexTypeJavaMBlock, StatMtxJavaMBlock)
+      , head() {
+    CHECK_EQ(begin, false);
+    begin = true;
+  }
+
+  ~BlockDesc() {
+    CHECK_EQ(begin, true);
+    begin = false;
+    ThreadState *thr = cur_thread();
+    SyncVar *s = head;
+    while (s) {
+      SyncVar *s1 = s->next;
+      StatInc(thr, StatSyncDestroyed);
+      s->mtx.Lock();
+      s->mtx.Unlock();
+      thr->mset.Remove(s->GetId());
+      DestroyAndFree(s);
+      s = s1;
+    }
+  }
+};
+
+struct JavaContext {
+  const uptr heap_begin;
+  const uptr heap_size;
+  BlockDesc *heap_shadow;
+
+  JavaContext(jptr heap_begin, jptr heap_size)
+      : heap_begin(heap_begin)
+      , heap_size(heap_size) {
+    uptr size = heap_size / kHeapAlignment * sizeof(BlockDesc);
+    heap_shadow = (BlockDesc*)MmapFixedNoReserve(kHeapShadow, size);
+    if ((uptr)heap_shadow != kHeapShadow) {
+      Printf("ThreadSanitizer: failed to mmap Java heap shadow\n");
+      Die();
+    }
+  }
+};
+
+class ScopedJavaFunc {
+ public:
+  ScopedJavaFunc(ThreadState *thr, uptr pc)
+      : thr_(thr) {
+    Initialize(thr_);
+    FuncEntry(thr, pc);
+    CHECK_EQ(thr_->in_rtl, 0);
+    thr_->in_rtl++;
+  }
+
+  ~ScopedJavaFunc() {
+    thr_->in_rtl--;
+    CHECK_EQ(thr_->in_rtl, 0);
+    FuncExit(thr_);
+    // FIXME(dvyukov): process pending signals.
+  }
+
+ private:
+  ThreadState *thr_;
+};
+
+static u64 jctx_buf[sizeof(JavaContext) / sizeof(u64) + 1];
+static JavaContext *jctx;
+
+static BlockDesc *getblock(uptr addr) {
+  uptr i = (addr - jctx->heap_begin) / kHeapAlignment;
+  return &jctx->heap_shadow[i];
+}
+
+static uptr USED getmem(BlockDesc *b) {
+  uptr i = b - jctx->heap_shadow;
+  uptr p = jctx->heap_begin + i * kHeapAlignment;
+  CHECK_GE(p, jctx->heap_begin);
+  CHECK_LT(p, jctx->heap_begin + jctx->heap_size);
+  return p;
+}
+
+static BlockDesc *getblockbegin(uptr addr) {
+  for (BlockDesc *b = getblock(addr);; b--) {
+    CHECK_GE(b, jctx->heap_shadow);
+    if (b->begin)
+      return b;
+  }
+  return 0;
+}
+
+SyncVar* GetJavaSync(ThreadState *thr, uptr pc, uptr addr,
+                     bool write_lock, bool create) {
+  if (jctx == 0 || addr < jctx->heap_begin
+      || addr >= jctx->heap_begin + jctx->heap_size)
+    return 0;
+  BlockDesc *b = getblockbegin(addr);
+  DPrintf("#%d: GetJavaSync %p->%p\n", thr->tid, addr, b);
+  Lock l(&b->mtx);
+  SyncVar *s = b->head;
+  for (; s; s = s->next) {
+    if (s->addr == addr) {
+      DPrintf("#%d: found existing sync for %p\n", thr->tid, addr);
+      break;
+    }
+  }
+  if (s == 0 && create) {
+    DPrintf("#%d: creating new sync for %p\n", thr->tid, addr);
+    s = CTX()->synctab.Create(thr, pc, addr);
+    s->next = b->head;
+    b->head = s;
+  }
+  if (s) {
+    if (write_lock)
+      s->mtx.Lock();
+    else
+      s->mtx.ReadLock();
+  }
+  return s;
+}
+
+SyncVar* GetAndRemoveJavaSync(ThreadState *thr, uptr pc, uptr addr) {
+  // We do not destroy Java mutexes other than in __tsan_java_free().
+  return 0;
+}
+
+}  // namespace __tsan {
+
+#define SCOPED_JAVA_FUNC(func) \
+  ThreadState *thr = cur_thread(); \
+  const uptr caller_pc = GET_CALLER_PC(); \
+  const uptr pc = (uptr)&func; \
+  (void)pc; \
+  ScopedJavaFunc scoped(thr, caller_pc); \
+/**/
+
+void __tsan_java_init(jptr heap_begin, jptr heap_size) {
+  SCOPED_JAVA_FUNC(__tsan_java_init);
+  DPrintf("#%d: java_init(%p, %p)\n", thr->tid, heap_begin, heap_size);
+  CHECK_EQ(jctx, 0);
+  CHECK_GT(heap_begin, 0);
+  CHECK_GT(heap_size, 0);
+  CHECK_EQ(heap_begin % kHeapAlignment, 0);
+  CHECK_EQ(heap_size % kHeapAlignment, 0);
+  CHECK_LT(heap_begin, heap_begin + heap_size);
+  jctx = new(jctx_buf) JavaContext(heap_begin, heap_size);
+}
+
+int  __tsan_java_fini() {
+  SCOPED_JAVA_FUNC(__tsan_java_fini);
+  DPrintf("#%d: java_fini()\n", thr->tid);
+  CHECK_NE(jctx, 0);
+  // FIXME(dvyukov): this does not call atexit() callbacks.
+  int status = Finalize(thr);
+  DPrintf("#%d: java_fini() = %d\n", thr->tid, status);
+  return status;
+}
+
+void __tsan_java_alloc(jptr ptr, jptr size) {
+  SCOPED_JAVA_FUNC(__tsan_java_alloc);
+  DPrintf("#%d: java_alloc(%p, %p)\n", thr->tid, ptr, size);
+  CHECK_NE(jctx, 0);
+  CHECK_NE(size, 0);
+  CHECK_EQ(ptr % kHeapAlignment, 0);
+  CHECK_EQ(size % kHeapAlignment, 0);
+  CHECK_GE(ptr, jctx->heap_begin);
+  CHECK_LE(ptr + size, jctx->heap_begin + jctx->heap_size);
+
+  BlockDesc *b = getblock(ptr);
+  new(b) BlockDesc();
+}
+
+void __tsan_java_free(jptr ptr, jptr size) {
+  SCOPED_JAVA_FUNC(__tsan_java_free);
+  DPrintf("#%d: java_free(%p, %p)\n", thr->tid, ptr, size);
+  CHECK_NE(jctx, 0);
+  CHECK_NE(size, 0);
+  CHECK_EQ(ptr % kHeapAlignment, 0);
+  CHECK_EQ(size % kHeapAlignment, 0);
+  CHECK_GE(ptr, jctx->heap_begin);
+  CHECK_LE(ptr + size, jctx->heap_begin + jctx->heap_size);
+
+  BlockDesc *beg = getblock(ptr);
+  BlockDesc *end = getblock(ptr + size);
+  for (BlockDesc *b = beg; b != end; b++) {
+    if (b->begin)
+      b->~BlockDesc();
+  }
+}
+
+void __tsan_java_move(jptr src, jptr dst, jptr size) {
+  SCOPED_JAVA_FUNC(__tsan_java_move);
+  DPrintf("#%d: java_move(%p, %p, %p)\n", thr->tid, src, dst, size);
+  CHECK_NE(jctx, 0);
+  CHECK_NE(size, 0);
+  CHECK_EQ(src % kHeapAlignment, 0);
+  CHECK_EQ(dst % kHeapAlignment, 0);
+  CHECK_EQ(size % kHeapAlignment, 0);
+  CHECK_GE(src, jctx->heap_begin);
+  CHECK_LE(src + size, jctx->heap_begin + jctx->heap_size);
+  CHECK_GE(dst, jctx->heap_begin);
+  CHECK_LE(dst + size, jctx->heap_begin + jctx->heap_size);
+  CHECK(dst >= src + size || src >= dst + size);
+
+  // Assuming it's not running concurrently with threads that do
+  // memory accesses and mutex operations (stop-the-world phase).
+  {  // NOLINT
+    BlockDesc *s = getblock(src);
+    BlockDesc *d = getblock(dst);
+    BlockDesc *send = getblock(src + size);
+    for (; s != send; s++, d++) {
+      CHECK_EQ(d->begin, false);
+      if (s->begin) {
+        DPrintf("#%d: moving block %p->%p\n", thr->tid, getmem(s), getmem(d));
+        new(d) BlockDesc;
+        d->head = s->head;
+        for (SyncVar *sync = d->head; sync; sync = sync->next) {
+          uptr newaddr = sync->addr - src + dst;
+          DPrintf("#%d: moving sync %p->%p\n", thr->tid, sync->addr, newaddr);
+          sync->addr = newaddr;
+        }
+        s->head = 0;
+        s->~BlockDesc();
+      }
+    }
+  }
+
+  {  // NOLINT
+    u64 *s = (u64*)MemToShadow(src);
+    u64 *d = (u64*)MemToShadow(dst);
+    u64 *send = (u64*)MemToShadow(src + size);
+    for (; s != send; s++, d++) {
+      *d = *s;
+      *s = 0;
+    }
+  }
+}
+
+void __tsan_java_mutex_lock(jptr addr) {
+  SCOPED_JAVA_FUNC(__tsan_java_mutex_lock);
+  DPrintf("#%d: java_mutex_lock(%p)\n", thr->tid, addr);
+  CHECK_NE(jctx, 0);
+  CHECK_GE(addr, jctx->heap_begin);
+  CHECK_LT(addr, jctx->heap_begin + jctx->heap_size);
+
+  MutexLock(thr, pc, addr);
+}
+
+void __tsan_java_mutex_unlock(jptr addr) {
+  SCOPED_JAVA_FUNC(__tsan_java_mutex_unlock);
+  DPrintf("#%d: java_mutex_unlock(%p)\n", thr->tid, addr);
+  CHECK_NE(jctx, 0);
+  CHECK_GE(addr, jctx->heap_begin);
+  CHECK_LT(addr, jctx->heap_begin + jctx->heap_size);
+
+  MutexUnlock(thr, pc, addr);
+}
+
+void __tsan_java_mutex_read_lock(jptr addr) {
+  SCOPED_JAVA_FUNC(__tsan_java_mutex_read_lock);
+  DPrintf("#%d: java_mutex_read_lock(%p)\n", thr->tid, addr);
+  CHECK_NE(jctx, 0);
+  CHECK_GE(addr, jctx->heap_begin);
+  CHECK_LT(addr, jctx->heap_begin + jctx->heap_size);
+
+  MutexReadLock(thr, pc, addr);
+}
+
+void __tsan_java_mutex_read_unlock(jptr addr) {
+  SCOPED_JAVA_FUNC(__tsan_java_mutex_read_unlock);
+  DPrintf("#%d: java_mutex_read_unlock(%p)\n", thr->tid, addr);
+  CHECK_NE(jctx, 0);
+  CHECK_GE(addr, jctx->heap_begin);
+  CHECK_LT(addr, jctx->heap_begin + jctx->heap_size);
+
+  MutexReadUnlock(thr, pc, addr);
+}
diff --git a/libsanitizer/tsan/tsan_interface_java.h b/libsanitizer/tsan/tsan_interface_java.h
new file mode 100644
index 00000000000..01922bc9231
--- /dev/null
+++ b/libsanitizer/tsan/tsan_interface_java.h
@@ -0,0 +1,72 @@
+//===-- tsan_interface_java.h -----------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+// Interface for verification of Java or mixed Java/C++ programs.
+// The interface is intended to be used from within a JVM and notify TSan
+// about such events like Java locks and GC memory compaction.
+//
+// For plain memory accesses and function entry/exit a JVM is intended to use
+// C++ interfaces: __tsan_readN/writeN and __tsan_func_enter/exit.
+//
+// For volatile memory accesses and atomic operations JVM is intended to use
+// standard atomics API: __tsan_atomicN_load/store/etc.
+//
+// For usage examples see lit_tests/java_*.cc
+//===----------------------------------------------------------------------===//
+#ifndef TSAN_INTERFACE_JAVA_H
+#define TSAN_INTERFACE_JAVA_H
+
+#ifndef INTERFACE_ATTRIBUTE
+# define INTERFACE_ATTRIBUTE __attribute__((visibility("default")))
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef unsigned long jptr;  // NOLINT
+
+// Must be called before any other callback from Java.
+void __tsan_java_init(jptr heap_begin, jptr heap_size) INTERFACE_ATTRIBUTE;
+// Must be called when the application exits.
+// Not necessary the last callback (concurrently running threads are OK).
+// Returns exit status or 0 if tsan does not want to override it.
+int  __tsan_java_fini() INTERFACE_ATTRIBUTE;
+
+// Callback for memory allocations.
+// May be omitted for allocations that are not subject to data races
+// nor contain synchronization objects (e.g. String).
+void __tsan_java_alloc(jptr ptr, jptr size) INTERFACE_ATTRIBUTE;
+// Callback for memory free.
+// Can be aggregated for several objects (preferably).
+void __tsan_java_free(jptr ptr, jptr size) INTERFACE_ATTRIBUTE;
+// Callback for memory move by GC.
+// Can be aggregated for several objects (preferably).
+// The ranges must not overlap.
+void __tsan_java_move(jptr src, jptr dst, jptr size) INTERFACE_ATTRIBUTE;
+
+// Mutex lock.
+// Addr is any unique address associated with the mutex.
+// Must not be called on recursive reentry.
+// Object.wait() is handled as a pair of unlock/lock.
+void __tsan_java_mutex_lock(jptr addr) INTERFACE_ATTRIBUTE;
+// Mutex unlock.
+void __tsan_java_mutex_unlock(jptr addr) INTERFACE_ATTRIBUTE;
+// Mutex read lock.
+void __tsan_java_mutex_read_lock(jptr addr) INTERFACE_ATTRIBUTE;
+// Mutex read unlock.
+void __tsan_java_mutex_read_unlock(jptr addr) INTERFACE_ATTRIBUTE;
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#undef INTERFACE_ATTRIBUTE
+
+#endif  // #ifndef TSAN_INTERFACE_JAVA_H
diff --git a/libsanitizer/tsan/tsan_mman.cc b/libsanitizer/tsan/tsan_mman.cc
index ba4252eccc5..9a8a524f262 100644
--- a/libsanitizer/tsan/tsan_mman.cc
+++ b/libsanitizer/tsan/tsan_mman.cc
@@ -58,8 +58,9 @@ void *user_alloc(ThreadState *thr, uptr pc, uptr sz, uptr align) {
   void *p = allocator()->Allocate(&thr->alloc_cache, sz, align);
   if (p == 0)
     return 0;
-  MBlock *b = (MBlock*)allocator()->GetMetaData(p);
+  MBlock *b = new(allocator()->GetMetaData(p)) MBlock;
   b->size = sz;
+  b->head = 0;
   b->alloc_tid = thr->unique_id;
   b->alloc_stack_id = CurrentStackId(thr, pc);
   if (CTX() && CTX()->initialized) {
@@ -90,6 +91,7 @@ void user_free(ThreadState *thr, uptr pc, void *p) {
   if (CTX() && CTX()->initialized && thr->in_rtl == 1) {
     MemoryRangeFreed(thr, pc, (uptr)p, b->size);
   }
+  b->~MBlock();
   allocator()->Deallocate(&thr->alloc_cache, p);
   SignalUnsafeCall(thr, pc);
 }
@@ -115,9 +117,11 @@ void *user_realloc(ThreadState *thr, uptr pc, void *p, uptr sz) {
 }
 
 MBlock *user_mblock(ThreadState *thr, void *p) {
-  // CHECK_GT(thr->in_rtl, 0);
   CHECK_NE(p, (void*)0);
-  return (MBlock*)allocator()->GetMetaData(p);
+  Allocator *a = allocator();
+  void *b = a->GetBlockBegin(p);
+  CHECK_NE(b, 0);
+  return (MBlock*)a->GetMetaData(b);
 }
 
 void invoke_malloc_hook(void *ptr, uptr size) {
diff --git a/libsanitizer/tsan/tsan_mman.h b/libsanitizer/tsan/tsan_mman.h
index 326bda7ebd7..8697d228730 100644
--- a/libsanitizer/tsan/tsan_mman.h
+++ b/libsanitizer/tsan/tsan_mman.h
@@ -57,6 +57,7 @@ enum MBlockType {
   MBlockSuppression,
   MBlockExpectRace,
   MBlockSignal,
+  MBlockFD,
 
   // This must be the last.
   MBlockTypeCount
diff --git a/libsanitizer/tsan/tsan_mutex.cc b/libsanitizer/tsan/tsan_mutex.cc
index 6a1e0cec53a..716722b0897 100644
--- a/libsanitizer/tsan/tsan_mutex.cc
+++ b/libsanitizer/tsan/tsan_mutex.cc
@@ -23,22 +23,28 @@ namespace __tsan {
 // then Report mutex can be locked while under Threads mutex.
 // The leaf mutexes can be locked under any other mutexes.
 // Recursive locking is not supported.
+#if TSAN_DEBUG && !TSAN_GO
 const MutexType MutexTypeLeaf = (MutexType)-1;
 static MutexType CanLockTab[MutexTypeCount][MutexTypeCount] = {
-  /*0 MutexTypeInvalid*/     {},
-  /*1 MutexTypeTrace*/       {MutexTypeLeaf},
-  /*2 MutexTypeThreads*/     {MutexTypeReport},
-  /*3 MutexTypeReport*/      {},
-  /*4 MutexTypeSyncVar*/     {},
-  /*5 MutexTypeSyncTab*/     {MutexTypeSyncVar},
-  /*6 MutexTypeSlab*/        {MutexTypeLeaf},
-  /*7 MutexTypeAnnotations*/ {},
-  /*8 MutexTypeAtExit*/      {MutexTypeSyncTab},
+  /*0  MutexTypeInvalid*/     {},
+  /*1  MutexTypeTrace*/       {MutexTypeLeaf},
+  /*2  MutexTypeThreads*/     {MutexTypeReport},
+  /*3  MutexTypeReport*/      {MutexTypeSyncTab, MutexTypeMBlock,
+                               MutexTypeJavaMBlock},
+  /*4  MutexTypeSyncVar*/     {},
+  /*5  MutexTypeSyncTab*/     {MutexTypeSyncVar},
+  /*6  MutexTypeSlab*/        {MutexTypeLeaf},
+  /*7  MutexTypeAnnotations*/ {},
+  /*8  MutexTypeAtExit*/      {MutexTypeSyncTab},
+  /*9  MutexTypeMBlock*/      {MutexTypeSyncVar},
+  /*10 MutexTypeJavaMBlock*/  {MutexTypeSyncVar},
 };
 
 static bool CanLockAdj[MutexTypeCount][MutexTypeCount];
+#endif
 
 void InitializeMutex() {
+#if TSAN_DEBUG && !TSAN_GO
   // Build the "can lock" adjacency matrix.
   // If [i][j]==true, then one can lock mutex j while under mutex i.
   const int N = MutexTypeCount;
@@ -112,14 +118,18 @@ void InitializeMutex() {
       Die();
     }
   }
+#endif
 }
 
 DeadlockDetector::DeadlockDetector() {
   // Rely on zero initialization because some mutexes can be locked before ctor.
 }
 
+#if TSAN_DEBUG && !TSAN_GO
 void DeadlockDetector::Lock(MutexType t) {
   // Printf("LOCK %d @%zu\n", t, seq_ + 1);
+  CHECK_GT(t, MutexTypeInvalid);
+  CHECK_LT(t, MutexTypeCount);
   u64 max_seq = 0;
   u64 max_idx = MutexTypeInvalid;
   for (int i = 0; i != MutexTypeCount; i++) {
@@ -148,6 +158,7 @@ void DeadlockDetector::Unlock(MutexType t) {
   CHECK(locked_[t]);
   locked_[t] = 0;
 }
+#endif
 
 const uptr kUnlocked = 0;
 const uptr kWriteLock = 1;
diff --git a/libsanitizer/tsan/tsan_mutex.h b/libsanitizer/tsan/tsan_mutex.h
index 118066e75c3..6d145059330 100644
--- a/libsanitizer/tsan/tsan_mutex.h
+++ b/libsanitizer/tsan/tsan_mutex.h
@@ -27,6 +27,8 @@ enum MutexType {
   MutexTypeSlab,
   MutexTypeAnnotations,
   MutexTypeAtExit,
+  MutexTypeMBlock,
+  MutexTypeJavaMBlock,
 
   // This must be the last.
   MutexTypeCount
diff --git a/libsanitizer/tsan/tsan_mutexset.cc b/libsanitizer/tsan/tsan_mutexset.cc
new file mode 100644
index 00000000000..3ebae3a57bc
--- /dev/null
+++ b/libsanitizer/tsan/tsan_mutexset.cc
@@ -0,0 +1,87 @@
+//===-- tsan_mutexset.cc --------------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+#include "tsan_mutexset.h"
+#include "tsan_rtl.h"
+
+namespace __tsan {
+
+const uptr MutexSet::kMaxSize;
+
+MutexSet::MutexSet() {
+  size_ = 0;
+  internal_memset(&descs_, 0, sizeof(descs_));
+}
+
+void MutexSet::Add(u64 id, bool write, u64 epoch) {
+  // Look up existing mutex with the same id.
+  for (uptr i = 0; i < size_; i++) {
+    if (descs_[i].id == id) {
+      descs_[i].count++;
+      descs_[i].epoch = epoch;
+      return;
+    }
+  }
+  // On overflow, find the oldest mutex and drop it.
+  if (size_ == kMaxSize) {
+    u64 minepoch = (u64)-1;
+    u64 mini = (u64)-1;
+    for (uptr i = 0; i < size_; i++) {
+      if (descs_[i].epoch < minepoch) {
+        minepoch = descs_[i].epoch;
+        mini = i;
+      }
+    }
+    RemovePos(mini);
+    CHECK_EQ(size_, kMaxSize - 1);
+  }
+  // Add new mutex descriptor.
+  descs_[size_].id = id;
+  descs_[size_].write = write;
+  descs_[size_].epoch = epoch;
+  descs_[size_].count = 1;
+  size_++;
+}
+
+void MutexSet::Del(u64 id, bool write) {
+  for (uptr i = 0; i < size_; i++) {
+    if (descs_[i].id == id) {
+      if (--descs_[i].count == 0)
+        RemovePos(i);
+      return;
+    }
+  }
+}
+
+void MutexSet::Remove(u64 id) {
+  for (uptr i = 0; i < size_; i++) {
+    if (descs_[i].id == id) {
+      RemovePos(i);
+      return;
+    }
+  }
+}
+
+void MutexSet::RemovePos(uptr i) {
+  CHECK_LT(i, size_);
+  descs_[i] = descs_[size_ - 1];
+  size_--;
+}
+
+uptr MutexSet::Size() const {
+  return size_;
+}
+
+MutexSet::Desc MutexSet::Get(uptr i) const {
+  CHECK_LT(i, size_);
+  return descs_[i];
+}
+
+}  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_mutexset.h b/libsanitizer/tsan/tsan_mutexset.h
new file mode 100644
index 00000000000..6924eade4c6
--- /dev/null
+++ b/libsanitizer/tsan/tsan_mutexset.h
@@ -0,0 +1,63 @@
+//===-- tsan_mutexset.h -----------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+// MutexSet holds the set of mutexes currently held by a thread.
+//===----------------------------------------------------------------------===//
+#ifndef TSAN_MUTEXSET_H
+#define TSAN_MUTEXSET_H
+
+#include "tsan_defs.h"
+
+namespace __tsan {
+
+class MutexSet {
+ public:
+  // Holds limited number of mutexes.
+  // The oldest mutexes are discarded on overflow.
+  static const uptr kMaxSize = 64;
+  struct Desc {
+    u64 id;
+    u64 epoch;
+    int count;
+    bool write;
+  };
+
+  MutexSet();
+  // The 'id' is obtained from SyncVar::GetId().
+  void Add(u64 id, bool write, u64 epoch);
+  void Del(u64 id, bool write);
+  void Remove(u64 id);  // Removes the mutex completely (if it's destroyed).
+  uptr Size() const;
+  Desc Get(uptr i) const;
+
+ private:
+#ifndef TSAN_GO
+  uptr size_;
+  Desc descs_[kMaxSize];
+#endif
+
+  void RemovePos(uptr i);
+};
+
+// Go does not have mutexes, so do not spend memory and time.
+// (Go sync.Mutex is actually a semaphore -- can be unlocked
+// in different goroutine).
+#ifdef TSAN_GO
+MutexSet::MutexSet() {}
+void MutexSet::Add(u64 id, bool write, u64 epoch) {}
+void MutexSet::Del(u64 id, bool write) {}
+void MutexSet::Remove(u64 id) {}
+void MutexSet::RemovePos(uptr i) {}
+uptr MutexSet::Size() const { return 0; }
+MutexSet::Desc MutexSet::Get(uptr i) const { return Desc(); }
+#endif
+
+}  // namespace __tsan
+
+#endif  // TSAN_REPORT_H
diff --git a/libsanitizer/tsan/tsan_platform.h b/libsanitizer/tsan/tsan_platform.h
index 5c776f14e8a..9fdc4dd46e7 100644
--- a/libsanitizer/tsan/tsan_platform.h
+++ b/libsanitizer/tsan/tsan_platform.h
@@ -135,7 +135,6 @@ void FlushShadowMemory();
 
 const char *InitializePlatform();
 void FinalizePlatform();
-void MapThreadTrace(uptr addr, uptr size);
 uptr ALWAYS_INLINE INLINE GetThreadTrace(int tid) {
   uptr p = kTraceMemBegin + (uptr)tid * kTraceSize * sizeof(Event);
   DCHECK_LT(p, kTraceMemBegin + kTraceMemSize);
diff --git a/libsanitizer/tsan/tsan_platform_linux.cc b/libsanitizer/tsan/tsan_platform_linux.cc
index 34221af16c9..2e7cd5138d6 100644
--- a/libsanitizer/tsan/tsan_platform_linux.cc
+++ b/libsanitizer/tsan/tsan_platform_linux.cc
@@ -69,9 +69,7 @@ uptr GetShadowMemoryConsumption() {
 }
 
 void FlushShadowMemory() {
-  madvise((void*)kLinuxShadowBeg,
-          kLinuxShadowEnd - kLinuxShadowBeg,
-          MADV_DONTNEED);
+  FlushUnneededShadowMemory(kLinuxShadowBeg, kLinuxShadowEnd - kLinuxShadowBeg);
 }
 
 #ifndef TSAN_GO
@@ -118,16 +116,6 @@ void InitializeShadowMemory() {
 }
 #endif
 
-void MapThreadTrace(uptr addr, uptr size) {
-  DPrintf("Mapping trace at %p-%p(0x%zx)\n", addr, addr + size, size);
-  CHECK_GE(addr, kTraceMemBegin);
-  CHECK_LE(addr + size, kTraceMemBegin + kTraceMemSize);
-  if (addr != (uptr)MmapFixedNoReserve(addr, size)) {
-    Printf("FATAL: ThreadSanitizer can not mmap thread trace\n");
-    Die();
-  }
-}
-
 static uptr g_data_start;
 static uptr g_data_end;
 
@@ -180,18 +168,14 @@ static uptr g_tls_size;
 #else
 # define INTERNAL_FUNCTION
 #endif
-extern "C" void _dl_get_tls_static_info(size_t*, size_t*)
-    __attribute__((weak)) INTERNAL_FUNCTION;
 
 static int InitTlsSize() {
   typedef void (*get_tls_func)(size_t*, size_t*) INTERNAL_FUNCTION;
-  get_tls_func get_tls = &_dl_get_tls_static_info;
-  if (get_tls == 0) {
-    void *get_tls_static_info_ptr = dlsym(RTLD_NEXT, "_dl_get_tls_static_info");
-    CHECK_EQ(sizeof(get_tls), sizeof(get_tls_static_info_ptr));
-    internal_memcpy(&get_tls, &get_tls_static_info_ptr,
-                    sizeof(get_tls_static_info_ptr));
-  }
+  get_tls_func get_tls;
+  void *get_tls_static_info_ptr = dlsym(RTLD_NEXT, "_dl_get_tls_static_info");
+  CHECK_EQ(sizeof(get_tls), sizeof(get_tls_static_info_ptr));
+  internal_memcpy(&get_tls, &get_tls_static_info_ptr,
+                  sizeof(get_tls_static_info_ptr));
   CHECK_NE(get_tls, 0);
   size_t tls_size = 0;
   size_t tls_align = 0;
@@ -220,29 +204,35 @@ const char *InitializePlatform() {
     // Disable core dumps, dumping of 16TB usually takes a bit long.
     setlim(RLIMIT_CORE, 0);
   }
-  bool reexec = false;
-  // TSan doesn't play well with unlimited stack size (as stack
-  // overlaps with shadow memory). If we detect unlimited stack size,
-  // we re-exec the program with limited stack size as a best effort.
-  if (getlim(RLIMIT_STACK) == (rlim_t)-1) {
-    const uptr kMaxStackSize = 32 * 1024 * 1024;
-    Report("WARNING: Program is run with unlimited stack size, which "
-           "wouldn't work with ThreadSanitizer.\n");
-    Report("Re-execing with stack size limited to %zd bytes.\n", kMaxStackSize);
-    SetStackSizeLimitInBytes(kMaxStackSize);
-    reexec = true;
-  }
 
-  if (getlim(RLIMIT_AS) != (rlim_t)-1) {
-    Report("WARNING: Program is run with limited virtual address space, which "
-           "wouldn't work with ThreadSanitizer.\n");
-    Report("Re-execing with unlimited virtual address space.\n");
-    setlim(RLIMIT_AS, -1);
-    reexec = true;
-  }
+  // Go maps shadow memory lazily and works fine with limited address space.
+  // Unlimited stack is not a problem as well, because the executable
+  // is not compiled with -pie.
+  if (kCppMode) {
+    bool reexec = false;
+    // TSan doesn't play well with unlimited stack size (as stack
+    // overlaps with shadow memory). If we detect unlimited stack size,
+    // we re-exec the program with limited stack size as a best effort.
+    if (getlim(RLIMIT_STACK) == (rlim_t)-1) {
+      const uptr kMaxStackSize = 32 * 1024 * 1024;
+      Report("WARNING: Program is run with unlimited stack size, which "
+             "wouldn't work with ThreadSanitizer.\n");
+      Report("Re-execing with stack size limited to %zd bytes.\n",
+             kMaxStackSize);
+      SetStackSizeLimitInBytes(kMaxStackSize);
+      reexec = true;
+    }
 
-  if (reexec)
-    ReExec();
+    if (getlim(RLIMIT_AS) != (rlim_t)-1) {
+      Report("WARNING: Program is run with limited virtual address space,"
+             " which wouldn't work with ThreadSanitizer.\n");
+      Report("Re-execing with unlimited virtual address space.\n");
+      setlim(RLIMIT_AS, -1);
+      reexec = true;
+    }
+    if (reexec)
+      ReExec();
+  }
 
 #ifndef TSAN_GO
   CheckPIE();
diff --git a/libsanitizer/tsan/tsan_report.cc b/libsanitizer/tsan/tsan_report.cc
index 18870a76eb7..ca352662902 100644
--- a/libsanitizer/tsan/tsan_report.cc
+++ b/libsanitizer/tsan/tsan_report.cc
@@ -23,12 +23,24 @@ ReportDesc::ReportDesc()
     , sleep() {
 }
 
+ReportMop::ReportMop()
+    : mset(MBlockReportMutex) {
+}
+
 ReportDesc::~ReportDesc() {
   // FIXME(dvyukov): it must be leaking a lot of memory.
 }
 
 #ifndef TSAN_GO
 
+const int kThreadBufSize = 32;
+const char *thread_name(char *buf, int tid) {
+  if (tid == 0)
+    return "main thread";
+  internal_snprintf(buf, kThreadBufSize, "thread T%d", tid);
+  return buf;
+}
+
 static void PrintHeader(ReportType typ) {
   Printf("WARNING: ThreadSanitizer: ");
 
@@ -65,52 +77,69 @@ void PrintStack(const ReportStack *ent) {
   Printf("\n");
 }
 
+static void PrintMutexSet(Vector<ReportMopMutex> const& mset) {
+  for (uptr i = 0; i < mset.Size(); i++) {
+    if (i == 0)
+      Printf(" (mutexes:");
+    const ReportMopMutex m = mset[i];
+    Printf(" %s M%llu", m.write ? "write" : "read", m.id);
+    Printf(i == mset.Size() - 1 ? ")" : ",");
+  }
+}
+
 static void PrintMop(const ReportMop *mop, bool first) {
-  Printf("  %s of size %d at %p",
+  char thrbuf[kThreadBufSize];
+  Printf("  %s of size %d at %p by %s",
       (first ? (mop->write ? "Write" : "Read")
              : (mop->write ? "Previous write" : "Previous read")),
-      mop->size, (void*)mop->addr);
-  if (mop->tid == 0)
-    Printf(" by main thread:\n");
-  else
-    Printf(" by thread %d:\n", mop->tid);
+      mop->size, (void*)mop->addr,
+      thread_name(thrbuf, mop->tid));
+  PrintMutexSet(mop->mset);
+  Printf(":\n");
   PrintStack(mop->stack);
 }
 
 static void PrintLocation(const ReportLocation *loc) {
+  char thrbuf[kThreadBufSize];
   if (loc->type == ReportLocationGlobal) {
     Printf("  Location is global '%s' of size %zu at %zx %s:%d (%s+%p)\n\n",
                loc->name, loc->size, loc->addr, loc->file, loc->line,
                loc->module, loc->offset);
   } else if (loc->type == ReportLocationHeap) {
-    Printf("  Location is heap block of size %zu at %p allocated",
-        loc->size, loc->addr);
-    if (loc->tid == 0)
-      Printf(" by main thread:\n");
-    else
-      Printf(" by thread %d:\n", loc->tid);
+    char thrbuf[kThreadBufSize];
+    Printf("  Location is heap block of size %zu at %p allocated by %s:\n",
+        loc->size, loc->addr, thread_name(thrbuf, loc->tid));
     PrintStack(loc->stack);
   } else if (loc->type == ReportLocationStack) {
-    Printf("  Location is stack of thread %d:\n\n", loc->tid);
+    Printf("  Location is stack of %s\n\n", thread_name(thrbuf, loc->tid));
+  } else if (loc->type == ReportLocationFD) {
+    Printf("  Location is file descriptor %d created by %s at:\n",
+        loc->fd, thread_name(thrbuf, loc->tid));
+    PrintStack(loc->stack);
   }
 }
 
 static void PrintMutex(const ReportMutex *rm) {
-  if (rm->stack == 0)
-    return;
-  Printf("  Mutex %d created at:\n", rm->id);
-  PrintStack(rm->stack);
+  if (rm->destroyed) {
+    Printf("  Mutex M%llu is already destroyed.\n\n", rm->id);
+  } else {
+    Printf("  Mutex M%llu created at:\n", rm->id);
+    PrintStack(rm->stack);
+  }
 }
 
 static void PrintThread(const ReportThread *rt) {
   if (rt->id == 0)  // Little sense in describing the main thread.
     return;
-  Printf("  Thread %d", rt->id);
+  Printf("  Thread T%d", rt->id);
   if (rt->name)
     Printf(" '%s'", rt->name);
-  Printf(" (tid=%zu, %s)", rt->pid, rt->running ? "running" : "finished");
+  char thrbuf[kThreadBufSize];
+  Printf(" (tid=%zu, %s) created by %s",
+    rt->pid, rt->running ? "running" : "finished",
+    thread_name(thrbuf, rt->parent_tid));
   if (rt->stack)
-    Printf(" created at:");
+    Printf(" at:");
   Printf("\n");
   PrintStack(rt->stack);
 }
diff --git a/libsanitizer/tsan/tsan_report.h b/libsanitizer/tsan/tsan_report.h
index 6776f1e078c..23fbc684209 100644
--- a/libsanitizer/tsan/tsan_report.h
+++ b/libsanitizer/tsan/tsan_report.h
@@ -36,20 +36,27 @@ struct ReportStack {
   int col;
 };
 
+struct ReportMopMutex {
+  u64 id;
+  bool write;
+};
+
 struct ReportMop {
   int tid;
   uptr addr;
   int size;
   bool write;
-  int nmutex;
-  int *mutex;
+  Vector<ReportMopMutex> mset;
   ReportStack *stack;
+
+  ReportMop();
 };
 
 enum ReportLocationType {
   ReportLocationGlobal,
   ReportLocationHeap,
-  ReportLocationStack
+  ReportLocationStack,
+  ReportLocationFD
 };
 
 struct ReportLocation {
@@ -59,6 +66,7 @@ struct ReportLocation {
   char *module;
   uptr offset;
   int tid;
+  int fd;
   char *name;
   char *file;
   int line;
@@ -70,11 +78,13 @@ struct ReportThread {
   uptr pid;
   bool running;
   char *name;
+  int parent_tid;
   ReportStack *stack;
 };
 
 struct ReportMutex {
-  int id;
+  u64 id;
+  bool destroyed;
   ReportStack *stack;
 };
 
diff --git a/libsanitizer/tsan/tsan_rtl.cc b/libsanitizer/tsan/tsan_rtl.cc
index 2778ac3e490..3615a7a9c2f 100644
--- a/libsanitizer/tsan/tsan_rtl.cc
+++ b/libsanitizer/tsan/tsan_rtl.cc
@@ -164,6 +164,16 @@ void MapShadow(uptr addr, uptr size) {
   MmapFixedNoReserve(MemToShadow(addr), size * kShadowMultiplier);
 }
 
+void MapThreadTrace(uptr addr, uptr size) {
+  DPrintf("#0: Mapping trace at %p-%p(0x%zx)\n", addr, addr + size, size);
+  CHECK_GE(addr, kTraceMemBegin);
+  CHECK_LE(addr + size, kTraceMemBegin + kTraceMemSize);
+  if (addr != (uptr)MmapFixedNoReserve(addr, size)) {
+    Printf("FATAL: ThreadSanitizer can not mmap thread trace\n");
+    Die();
+  }
+}
+
 void Initialize(ThreadState *thr) {
   // Thread safe because done before all threads exist.
   static bool is_initialized = false;
@@ -289,6 +299,7 @@ void TraceSwitch(ThreadState *thr) {
   TraceHeader *hdr = &thr->trace.headers[trace];
   hdr->epoch0 = thr->fast_state.epoch();
   hdr->stack0.ObtainCurrent(thr, 0);
+  hdr->mset0 = thr->mset;
   thr->nomalloc--;
 }
 
@@ -443,7 +454,7 @@ ALWAYS_INLINE
 void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
     int kAccessSizeLog, bool kAccessIsWrite) {
   u64 *shadow_mem = (u64*)MemToShadow(addr);
-  DPrintf2("#%d: tsan::OnMemoryAccess: @%p %p size=%d"
+  DPrintf2("#%d: MemoryAccess: @%p %p size=%d"
       " is_write=%d shadow_mem=%p {%zx, %zx, %zx, %zx}\n",
       (int)thr->fast_state.tid(), (void*)pc, (void*)addr,
       (int)(1 << kAccessSizeLog), kAccessIsWrite, shadow_mem,
diff --git a/libsanitizer/tsan/tsan_rtl.h b/libsanitizer/tsan/tsan_rtl.h
index 56fcad1412f..b911791c187 100644
--- a/libsanitizer/tsan/tsan_rtl.h
+++ b/libsanitizer/tsan/tsan_rtl.h
@@ -34,6 +34,7 @@
 #include "tsan_vector.h"
 #include "tsan_report.h"
 #include "tsan_platform.h"
+#include "tsan_mutexset.h"
 
 #if SANITIZER_WORDSIZE != 64
 # error "ThreadSanitizer is supported only on 64-bit platforms"
@@ -48,6 +49,10 @@ struct MBlock {
   u32 alloc_tid;
   u32 alloc_stack_id;
   SyncVar *head;
+
+  MBlock()
+    : mtx(MutexTypeMBlock, StatMtxMBlock) {
+  }
 };
 
 #ifndef TSAN_GO
@@ -58,10 +63,22 @@ const uptr kAllocatorSpace = 0x7d0000000000ULL;
 #endif
 const uptr kAllocatorSize  =  0x10000000000ULL;  // 1T.
 
+struct TsanMapUnmapCallback {
+  void OnMap(uptr p, uptr size) const { }
+  void OnUnmap(uptr p, uptr size) const {
+    // We are about to unmap a chunk of user memory.
+    // Mark the corresponding shadow memory as not needed.
+    uptr shadow_beg = MemToShadow(p);
+    uptr shadow_end = MemToShadow(p + size);
+    CHECK(IsAligned(shadow_end|shadow_beg, GetPageSizeCached()));
+    FlushUnneededShadowMemory(shadow_beg, shadow_end - shadow_beg);
+  }
+};
+
 typedef SizeClassAllocator64<kAllocatorSpace, kAllocatorSize, sizeof(MBlock),
     DefaultSizeClassMap> PrimaryAllocator;
 typedef SizeClassAllocatorLocalCache<PrimaryAllocator> AllocatorCache;
-typedef LargeMmapAllocator SecondaryAllocator;
+typedef LargeMmapAllocator<TsanMapUnmapCallback> SecondaryAllocator;
 typedef CombinedAllocator<PrimaryAllocator, AllocatorCache,
     SecondaryAllocator> Allocator;
 Allocator *allocator();
@@ -298,6 +315,7 @@ struct ThreadState {
   uptr *shadow_stack;
   uptr *shadow_stack_end;
 #endif
+  MutexSet mset;
   ThreadClock clock;
 #ifndef TSAN_GO
   AllocatorCache alloc_cache;
@@ -369,6 +387,7 @@ struct ThreadContext {
   u64 epoch0;
   u64 epoch1;
   StackTrace creation_stack;
+  int creation_tid;
   ThreadDeadInfo *dead_info;
   ThreadContext *dead_next;  // In dead thread list.
   char *name;  // As annotated by user.
@@ -445,7 +464,8 @@ class ScopedReport {
   ~ScopedReport();
 
   void AddStack(const StackTrace *stack);
-  void AddMemoryAccess(uptr addr, Shadow s, const StackTrace *stack);
+  void AddMemoryAccess(uptr addr, Shadow s, const StackTrace *stack,
+                       const MutexSet *mset);
   void AddThread(const ThreadContext *tctx);
   void AddMutex(const SyncVar *s);
   void AddLocation(uptr addr, uptr size);
@@ -457,11 +477,13 @@ class ScopedReport {
   Context *ctx_;
   ReportDesc *rep_;
 
+  void AddMutex(u64 id);
+
   ScopedReport(const ScopedReport&);
   void operator = (const ScopedReport&);
 };
 
-void RestoreStack(int tid, const u64 epoch, StackTrace *stk);
+void RestoreStack(int tid, const u64 epoch, StackTrace *stk, MutexSet *mset);
 
 void StatAggregate(u64 *dst, u64 *src);
 void StatOutput(u64 *stat);
@@ -471,6 +493,7 @@ void ALWAYS_INLINE INLINE StatInc(ThreadState *thr, StatType typ, u64 n = 1) {
 }
 
 void MapShadow(uptr addr, uptr size);
+void MapThreadTrace(uptr addr, uptr size);
 void InitializeShadowMemory();
 void InitializeInterceptors();
 void InitializeDynamicAnnotations();
@@ -502,6 +525,10 @@ void PrintCurrentStack(ThreadState *thr, uptr pc);
 void Initialize(ThreadState *thr);
 int Finalize(ThreadState *thr);
 
+SyncVar* GetJavaSync(ThreadState *thr, uptr pc, uptr addr,
+                     bool write_lock, bool create);
+SyncVar* GetAndRemoveJavaSync(ThreadState *thr, uptr pc, uptr addr);
+
 void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
     int kAccessSizeLog, bool kAccessIsWrite);
 void MemoryAccessImpl(ThreadState *thr, uptr addr,
@@ -575,7 +602,10 @@ uptr TraceParts();
 
 extern "C" void __tsan_trace_switch();
 void ALWAYS_INLINE INLINE TraceAddEvent(ThreadState *thr, FastState fs,
-                                        EventType typ, uptr addr) {
+                                        EventType typ, u64 addr) {
+  DCHECK_GE((int)typ, 0);
+  DCHECK_LE((int)typ, 7);
+  DCHECK_EQ(GetLsb(addr, 61), addr);
   StatInc(thr, StatEvents);
   u64 pos = fs.GetTracePos();
   if (UNLIKELY((pos % kTracePartSize) == 0)) {
diff --git a/libsanitizer/tsan/tsan_rtl_mutex.cc b/libsanitizer/tsan/tsan_rtl_mutex.cc
index e5b43be6a49..8dd0e6d4d9b 100644
--- a/libsanitizer/tsan/tsan_rtl_mutex.cc
+++ b/libsanitizer/tsan/tsan_rtl_mutex.cc
@@ -26,7 +26,7 @@ void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
   StatInc(thr, StatMutexCreate);
   if (!linker_init && IsAppMem(addr))
     MemoryWrite1Byte(thr, pc, addr);
-  SyncVar *s = ctx->synctab.GetAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   s->is_rw = rw;
   s->is_recursive = recursive;
   s->is_linker_init = linker_init;
@@ -59,11 +59,12 @@ void MutexDestroy(ThreadState *thr, uptr pc, uptr addr) {
     trace.ObtainCurrent(thr, pc);
     rep.AddStack(&trace);
     FastState last(s->last_lock);
-    RestoreStack(last.tid(), last.epoch(), &trace);
+    RestoreStack(last.tid(), last.epoch(), &trace, 0);
     rep.AddStack(&trace);
     rep.AddLocation(s->addr, 1);
     OutputReport(ctx, rep);
   }
+  thr->mset.Remove(s->GetId());
   DestroyAndFree(s);
 }
 
@@ -72,9 +73,9 @@ void MutexLock(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexLock %zx\n", thr->tid, addr);
   if (IsAppMem(addr))
     MemoryRead1Byte(thr, pc, addr);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
-  TraceAddEvent(thr, thr->fast_state, EventTypeLock, addr);
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, addr, true);
+  TraceAddEvent(thr, thr->fast_state, EventTypeLock, s->GetId());
   if (s->owner_tid == SyncVar::kInvalidTid) {
     CHECK_EQ(s->recursion, 0);
     s->owner_tid = thr->tid;
@@ -96,6 +97,7 @@ void MutexLock(ThreadState *thr, uptr pc, uptr addr) {
     StatInc(thr, StatMutexRecLock);
   }
   s->recursion++;
+  thr->mset.Add(s->GetId(), true, thr->fast_state.epoch());
   s->mtx.Unlock();
 }
 
@@ -104,9 +106,9 @@ void MutexUnlock(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexUnlock %zx\n", thr->tid, addr);
   if (IsAppMem(addr))
     MemoryRead1Byte(thr, pc, addr);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
-  TraceAddEvent(thr, thr->fast_state, EventTypeUnlock, addr);
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, addr, true);
+  TraceAddEvent(thr, thr->fast_state, EventTypeUnlock, s->GetId());
   if (s->recursion == 0) {
     if (!s->is_broken) {
       s->is_broken = true;
@@ -132,6 +134,7 @@ void MutexUnlock(ThreadState *thr, uptr pc, uptr addr) {
       StatInc(thr, StatMutexRecUnlock);
     }
   }
+  thr->mset.Del(s->GetId(), true);
   s->mtx.Unlock();
 }
 
@@ -141,9 +144,9 @@ void MutexReadLock(ThreadState *thr, uptr pc, uptr addr) {
   StatInc(thr, StatMutexReadLock);
   if (IsAppMem(addr))
     MemoryRead1Byte(thr, pc, addr);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, false);
   thr->fast_state.IncrementEpoch();
-  TraceAddEvent(thr, thr->fast_state, EventTypeRLock, addr);
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, addr, false);
+  TraceAddEvent(thr, thr->fast_state, EventTypeRLock, s->GetId());
   if (s->owner_tid != SyncVar::kInvalidTid) {
     Printf("ThreadSanitizer WARNING: read lock of a write locked mutex\n");
     PrintCurrentStack(thr, pc);
@@ -152,6 +155,7 @@ void MutexReadLock(ThreadState *thr, uptr pc, uptr addr) {
   thr->clock.acquire(&s->clock);
   s->last_lock = thr->fast_state.raw();
   StatInc(thr, StatSyncAcquire);
+  thr->mset.Add(s->GetId(), false, thr->fast_state.epoch());
   s->mtx.ReadUnlock();
 }
 
@@ -161,9 +165,9 @@ void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr) {
   StatInc(thr, StatMutexReadUnlock);
   if (IsAppMem(addr))
     MemoryRead1Byte(thr, pc, addr);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
-  TraceAddEvent(thr, thr->fast_state, EventTypeRUnlock, addr);
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, addr, true);
+  TraceAddEvent(thr, thr->fast_state, EventTypeRUnlock, s->GetId());
   if (s->owner_tid != SyncVar::kInvalidTid) {
     Printf("ThreadSanitizer WARNING: read unlock of a write "
                "locked mutex\n");
@@ -174,6 +178,7 @@ void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr) {
   thr->clock.release(&s->read_clock);
   StatInc(thr, StatSyncRelease);
   s->mtx.Unlock();
+  thr->mset.Del(s->GetId(), false);
 }
 
 void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
@@ -181,18 +186,22 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexReadOrWriteUnlock %zx\n", thr->tid, addr);
   if (IsAppMem(addr))
     MemoryRead1Byte(thr, pc, addr);
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, addr, true);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  bool write = true;
   if (s->owner_tid == SyncVar::kInvalidTid) {
     // Seems to be read unlock.
+    write = false;
     StatInc(thr, StatMutexReadUnlock);
     thr->fast_state.IncrementEpoch();
-    TraceAddEvent(thr, thr->fast_state, EventTypeRUnlock, addr);
+    TraceAddEvent(thr, thr->fast_state, EventTypeRUnlock, s->GetId());
     thr->clock.set(thr->tid, thr->fast_state.epoch());
     thr->fast_synch_epoch = thr->fast_state.epoch();
     thr->clock.release(&s->read_clock);
     StatInc(thr, StatSyncRelease);
   } else if (s->owner_tid == thr->tid) {
     // Seems to be write unlock.
+    thr->fast_state.IncrementEpoch();
+    TraceAddEvent(thr, thr->fast_state, EventTypeUnlock, s->GetId());
     CHECK_GT(s->recursion, 0);
     s->recursion--;
     if (s->recursion == 0) {
@@ -202,8 +211,6 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
       // The sequence of events is quite tricky and doubled in several places.
       // First, it's a bug to increment the epoch w/o writing to the trace.
       // Then, the acquire/release logic can be factored out as well.
-      thr->fast_state.IncrementEpoch();
-      TraceAddEvent(thr, thr->fast_state, EventTypeUnlock, addr);
       thr->clock.set(thr->tid, thr->fast_state.epoch());
       thr->fast_synch_epoch = thr->fast_state.epoch();
       thr->clock.ReleaseStore(&s->clock);
@@ -216,13 +223,14 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
     Printf("ThreadSanitizer WARNING: mutex unlock by another thread\n");
     PrintCurrentStack(thr, pc);
   }
+  thr->mset.Del(s->GetId(), write);
   s->mtx.Unlock();
 }
 
 void Acquire(ThreadState *thr, uptr pc, uptr addr) {
   CHECK_GT(thr->in_rtl, 0);
   DPrintf("#%d: Acquire %zx\n", thr->tid, addr);
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, addr, false);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, false);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   thr->clock.acquire(&s->clock);
   StatInc(thr, StatSyncAcquire);
@@ -246,7 +254,7 @@ void AcquireGlobal(ThreadState *thr, uptr pc) {
 void Release(ThreadState *thr, uptr pc, uptr addr) {
   CHECK_GT(thr->in_rtl, 0);
   DPrintf("#%d: Release %zx\n", thr->tid, addr);
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, addr, true);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   thr->clock.release(&s->clock);
   StatInc(thr, StatSyncRelease);
@@ -256,7 +264,7 @@ void Release(ThreadState *thr, uptr pc, uptr addr) {
 void ReleaseStore(ThreadState *thr, uptr pc, uptr addr) {
   CHECK_GT(thr->in_rtl, 0);
   DPrintf("#%d: ReleaseStore %zx\n", thr->tid, addr);
-  SyncVar *s = CTX()->synctab.GetAndLock(thr, pc, addr, true);
+  SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   thr->clock.ReleaseStore(&s->clock);
   StatInc(thr, StatSyncRelease);
diff --git a/libsanitizer/tsan/tsan_rtl_report.cc b/libsanitizer/tsan/tsan_rtl_report.cc
index 6aae6cf6e8a..b65b24fce89 100644
--- a/libsanitizer/tsan/tsan_rtl_report.cc
+++ b/libsanitizer/tsan/tsan_rtl_report.cc
@@ -12,6 +12,7 @@
 #include "sanitizer_common/sanitizer_libc.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "sanitizer_common/sanitizer_stackdepot.h"
+#include "sanitizer_common/sanitizer_common.h"
 #include "tsan_platform.h"
 #include "tsan_rtl.h"
 #include "tsan_suppressions.h"
@@ -20,9 +21,12 @@
 #include "tsan_sync.h"
 #include "tsan_mman.h"
 #include "tsan_flags.h"
+#include "tsan_fd.h"
 
 namespace __tsan {
 
+using namespace __sanitizer;  // NOLINT
+
 void TsanCheckFailed(const char *file, int line, const char *cond,
                      u64 v1, u64 v2) {
   ScopedInRtl in_rtl;
@@ -132,7 +136,7 @@ void ScopedReport::AddStack(const StackTrace *stack) {
 }
 
 void ScopedReport::AddMemoryAccess(uptr addr, Shadow s,
-                                   const StackTrace *stack) {
+    const StackTrace *stack, const MutexSet *mset) {
   void *mem = internal_alloc(MBlockReportMop, sizeof(ReportMop));
   ReportMop *mop = new(mem) ReportMop;
   rep_->mops.PushBack(mop);
@@ -140,8 +144,27 @@ void ScopedReport::AddMemoryAccess(uptr addr, Shadow s,
   mop->addr = addr + s.addr0();
   mop->size = s.size();
   mop->write = s.is_write();
-  mop->nmutex = 0;
   mop->stack = SymbolizeStack(*stack);
+  for (uptr i = 0; i < mset->Size(); i++) {
+    MutexSet::Desc d = mset->Get(i);
+    u64 uid = 0;
+    uptr addr = SyncVar::SplitId(d.id, &uid);
+    SyncVar *s = ctx_->synctab.GetIfExistsAndLock(addr, false);
+    // Check that the mutex is still alive.
+    // Another mutex can be created at the same address,
+    // so check uid as well.
+    if (s && s->CheckId(uid)) {
+      ReportMopMutex mtx = {s->uid, d.write};
+      mop->mset.PushBack(mtx);
+      AddMutex(s);
+    } else {
+      ReportMopMutex mtx = {d.id, d.write};
+      mop->mset.PushBack(mtx);
+      AddMutex(d.id);
+    }
+    if (s)
+      s->mtx.ReadUnlock();
+  }
 }
 
 void ScopedReport::AddThread(const ThreadContext *tctx) {
@@ -156,6 +179,7 @@ void ScopedReport::AddThread(const ThreadContext *tctx) {
   rt->pid = tctx->os_id;
   rt->running = (tctx->status == ThreadStatusRunning);
   rt->name = tctx->name ? internal_strdup(tctx->name) : 0;
+  rt->parent_tid = tctx->creation_tid;
   rt->stack = SymbolizeStack(tctx->creation_stack);
 }
 
@@ -173,17 +197,58 @@ static ThreadContext *FindThread(int unique_id) {
 #endif
 
 void ScopedReport::AddMutex(const SyncVar *s) {
+  for (uptr i = 0; i < rep_->mutexes.Size(); i++) {
+    if (rep_->mutexes[i]->id == s->uid)
+      return;
+  }
   void *mem = internal_alloc(MBlockReportMutex, sizeof(ReportMutex));
   ReportMutex *rm = new(mem) ReportMutex();
   rep_->mutexes.PushBack(rm);
-  rm->id = 42;
+  rm->id = s->uid;
+  rm->destroyed = false;
   rm->stack = SymbolizeStack(s->creation_stack);
 }
 
+void ScopedReport::AddMutex(u64 id) {
+  for (uptr i = 0; i < rep_->mutexes.Size(); i++) {
+    if (rep_->mutexes[i]->id == id)
+      return;
+  }
+  void *mem = internal_alloc(MBlockReportMutex, sizeof(ReportMutex));
+  ReportMutex *rm = new(mem) ReportMutex();
+  rep_->mutexes.PushBack(rm);
+  rm->id = id;
+  rm->destroyed = true;
+  rm->stack = 0;
+}
+
 void ScopedReport::AddLocation(uptr addr, uptr size) {
   if (addr == 0)
     return;
 #ifndef TSAN_GO
+  int fd = -1;
+  int creat_tid = -1;
+  u32 creat_stack = 0;
+  if (FdLocation(addr, &fd, &creat_tid, &creat_stack)
+      || FdLocation(AlternativeAddress(addr), &fd, &creat_tid, &creat_stack)) {
+    void *mem = internal_alloc(MBlockReportLoc, sizeof(ReportLocation));
+    ReportLocation *loc = new(mem) ReportLocation();
+    rep_->locs.PushBack(loc);
+    loc->type = ReportLocationFD;
+    loc->fd = fd;
+    loc->tid = creat_tid;
+    uptr ssz = 0;
+    const uptr *stack = StackDepotGet(creat_stack, &ssz);
+    if (stack) {
+      StackTrace trace;
+      trace.Init(stack, ssz);
+      loc->stack = SymbolizeStack(trace);
+    }
+    ThreadContext *tctx = FindThread(creat_tid);
+    if (tctx)
+      AddThread(tctx);
+    return;
+  }
   if (allocator()->PointerIsMine((void*)addr)) {
     MBlock *b = user_mblock(0, (void*)addr);
     ThreadContext *tctx = FindThread(b->alloc_tid);
@@ -246,7 +311,10 @@ const ReportDesc *ScopedReport::GetReport() const {
   return rep_;
 }
 
-void RestoreStack(int tid, const u64 epoch, StackTrace *stk) {
+void RestoreStack(int tid, const u64 epoch, StackTrace *stk, MutexSet *mset) {
+  // This function restores stack trace and mutex set for the thread/epoch.
+  // It does so by getting stack trace and mutex set at the beginning of
+  // trace part, and then replaying the trace till the given epoch.
   ThreadContext *tctx = CTX()->threads[tid];
   if (tctx == 0)
     return;
@@ -267,6 +335,7 @@ void RestoreStack(int tid, const u64 epoch, StackTrace *stk) {
   TraceHeader* hdr = &trace->headers[partidx];
   if (epoch < hdr->epoch0)
     return;
+  const u64 epoch0 = RoundDown(epoch, TraceSize());
   const u64 eend = epoch % TraceSize();
   const u64 ebegin = RoundDown(eend, kTracePartSize);
   DPrintf("#%d: RestoreStack epoch=%zu ebegin=%zu eend=%zu partidx=%d\n",
@@ -276,12 +345,14 @@ void RestoreStack(int tid, const u64 epoch, StackTrace *stk) {
     stack[i] = hdr->stack0.Get(i);
     DPrintf2("  #%02lu: pc=%zx\n", i, stack[i]);
   }
+  if (mset)
+    *mset = hdr->mset0;
   uptr pos = hdr->stack0.Size();
   Event *events = (Event*)GetThreadTrace(tid);
   for (uptr i = ebegin; i <= eend; i++) {
     Event ev = events[i];
     EventType typ = (EventType)(ev >> 61);
-    uptr pc = (uptr)(ev & 0xffffffffffffull);
+    uptr pc = (uptr)(ev & ((1ull << 61) - 1));
     DPrintf2("  %zu typ=%d pc=%zx\n", i, typ, pc);
     if (typ == EventTypeMop) {
       stack[pos] = pc;
@@ -291,6 +362,17 @@ void RestoreStack(int tid, const u64 epoch, StackTrace *stk) {
       if (pos > 0)
         pos--;
     }
+    if (mset) {
+      if (typ == EventTypeLock) {
+        mset->Add(pc, true, epoch0 + i);
+      } else if (typ == EventTypeUnlock) {
+        mset->Del(pc, true);
+      } else if (typ == EventTypeRLock) {
+        mset->Add(pc, false, epoch0 + i);
+      } else if (typ == EventTypeRUnlock) {
+        mset->Del(pc, false);
+      }
+    }
     for (uptr j = 0; j <= pos; j++)
       DPrintf2("      #%zu: %zx\n", j, stack[j]);
   }
@@ -400,8 +482,11 @@ static bool IsJavaNonsense(const ReportDesc *rep) {
     if (frame != 0 && frame->func != 0
         && (internal_strcmp(frame->func, "memset") == 0
         || internal_strcmp(frame->func, "memcpy") == 0
+        || internal_strcmp(frame->func, "memmove") == 0
         || internal_strcmp(frame->func, "strcmp") == 0
         || internal_strcmp(frame->func, "strncpy") == 0
+        || internal_strcmp(frame->func, "strlen") == 0
+        || internal_strcmp(frame->func, "free") == 0
         || internal_strcmp(frame->func, "pthread_mutex_lock") == 0)) {
       frame = frame->next;
       if (frame == 0
@@ -423,6 +508,10 @@ void ReportRace(ThreadState *thr) {
     return;
   ScopedInRtl in_rtl;
 
+  if (thr->in_signal_handler)
+    Printf("ThreadSanitizer: printing report from signal handler."
+           " Can crash or hang.\n");
+
   bool freed = false;
   {
     Shadow s(thr->racy_state[1]);
@@ -454,15 +543,18 @@ void ReportRace(ThreadState *thr) {
   traces[0].ObtainCurrent(thr, toppc);
   if (IsFiredSuppression(ctx, rep, traces[0]))
     return;
+  InternalScopedBuffer<MutexSet> mset2(1);
+  new(mset2.data()) MutexSet();
   Shadow s2(thr->racy_state[1]);
-  RestoreStack(s2.tid(), s2.epoch(), &traces[1]);
+  RestoreStack(s2.tid(), s2.epoch(), &traces[1], mset2.data());
 
   if (HandleRacyStacks(thr, traces, addr_min, addr_max))
     return;
 
   for (uptr i = 0; i < kMop; i++) {
     Shadow s(thr->racy_state[i]);
-    rep.AddMemoryAccess(addr, s, &traces[i]);
+    rep.AddMemoryAccess(addr, s, &traces[i],
+                        i == 0 ? &thr->mset : mset2.data());
   }
 
   if (flags()->suppress_java && IsJavaNonsense(rep.GetReport()))
diff --git a/libsanitizer/tsan/tsan_rtl_thread.cc b/libsanitizer/tsan/tsan_rtl_thread.cc
index 462d12c7dee..d5b3444be6d 100644
--- a/libsanitizer/tsan/tsan_rtl_thread.cc
+++ b/libsanitizer/tsan/tsan_rtl_thread.cc
@@ -154,6 +154,7 @@ int ThreadCreate(ThreadState *thr, uptr pc, uptr uid, bool detached) {
     thr->clock.release(&tctx->sync);
     StatInc(thr, StatSyncRelease);
     tctx->creation_stack.ObtainCurrent(thr, pc);
+    tctx->creation_tid = thr->tid;
   }
   return tid;
 }
@@ -303,6 +304,7 @@ void ThreadJoin(ThreadState *thr, uptr pc, int tid) {
     Printf("ThreadSanitizer: join of non-existent thread\n");
     return;
   }
+  // FIXME(dvyukov): print message and continue (it's user error).
   CHECK_EQ(tctx->detached, false);
   CHECK_EQ(tctx->status, ThreadStatusFinished);
   thr->clock.acquire(&tctx->sync);
diff --git a/libsanitizer/tsan/tsan_stat.cc b/libsanitizer/tsan/tsan_stat.cc
index 7b913306673..394c9111626 100644
--- a/libsanitizer/tsan/tsan_stat.cc
+++ b/libsanitizer/tsan/tsan_stat.cc
@@ -179,6 +179,28 @@ void StatOutput(u64 *stat) {
   name[StatInt_sem_timedwait]            = "  sem_timedwait                   ";
   name[StatInt_sem_post]                 = "  sem_post                        ";
   name[StatInt_sem_getvalue]             = "  sem_getvalue                    ";
+  name[StatInt_open]                     = "  open                            ";
+  name[StatInt_open64]                   = "  open64                          ";
+  name[StatInt_creat]                    = "  creat                           ";
+  name[StatInt_creat64]                  = "  creat64                         ";
+  name[StatInt_dup]                      = "  dup                             ";
+  name[StatInt_dup2]                     = "  dup2                            ";
+  name[StatInt_dup3]                     = "  dup3                            ";
+  name[StatInt_eventfd]                  = "  eventfd                         ";
+  name[StatInt_signalfd]                 = "  signalfd                        ";
+  name[StatInt_inotify_init]             = "  inotify_init                    ";
+  name[StatInt_inotify_init1]            = "  inotify_init1                   ";
+  name[StatInt_socket]                   = "  socket                          ";
+  name[StatInt_socketpair]               = "  socketpair                      ";
+  name[StatInt_connect]                  = "  connect                         ";
+  name[StatInt_accept]                   = "  accept                          ";
+  name[StatInt_accept4]                  = "  accept4                         ";
+  name[StatInt_epoll_create]             = "  epoll_create                    ";
+  name[StatInt_epoll_create1]            = "  epoll_create1                   ";
+  name[StatInt_close]                    = "  close                           ";
+  name[StatInt___close]                  = "  __close                         ";
+  name[StatInt_pipe]                     = "  pipe                            ";
+  name[StatInt_pipe2]                    = "  pipe2                           ";
   name[StatInt_read]                     = "  read                            ";
   name[StatInt_pread]                    = "  pread                           ";
   name[StatInt_pread64]                  = "  pread64                         ";
@@ -195,6 +217,8 @@ void StatOutput(u64 *stat) {
   name[StatInt_recvmsg]                  = "  recvmsg                         ";
   name[StatInt_unlink]                   = "  unlink                          ";
   name[StatInt_fopen]                    = "  fopen                           ";
+  name[StatInt_freopen]                  = "  freopen                         ";
+  name[StatInt_fclose]                   = "  fclose                          ";
   name[StatInt_fread]                    = "  fread                           ";
   name[StatInt_fwrite]                   = "  fwrite                          ";
   name[StatInt_puts]                     = "  puts                            ";
@@ -208,6 +232,7 @@ void StatOutput(u64 *stat) {
   name[StatInt_usleep]                   = "  usleep                          ";
   name[StatInt_nanosleep]                = "  nanosleep                       ";
   name[StatInt_gettimeofday]             = "  gettimeofday                    ";
+  name[StatInt_fork]                     = "  fork                            ";
 
   name[StatAnnotation]                   = "Dynamic annotations               ";
   name[StatAnnotateHappensBefore]        = "  HappensBefore                   ";
@@ -251,6 +276,8 @@ void StatOutput(u64 *stat) {
   name[StatMtxSlab]                      = "  Slab                            ";
   name[StatMtxAtExit]                    = "  Atexit                          ";
   name[StatMtxAnnotations]               = "  Annotations                     ";
+  name[StatMtxMBlock]                    = "  MBlock                          ";
+  name[StatMtxJavaMBlock]                = "  JavaMBlock                      ";
 
   Printf("Statistics:\n");
   for (int i = 0; i < StatCnt; i++)
diff --git a/libsanitizer/tsan/tsan_stat.h b/libsanitizer/tsan/tsan_stat.h
index 0dc1cd9a90d..cdd57365bae 100644
--- a/libsanitizer/tsan/tsan_stat.h
+++ b/libsanitizer/tsan/tsan_stat.h
@@ -174,6 +174,28 @@ enum StatType {
   StatInt_sem_timedwait,
   StatInt_sem_post,
   StatInt_sem_getvalue,
+  StatInt_open,
+  StatInt_open64,
+  StatInt_creat,
+  StatInt_creat64,
+  StatInt_dup,
+  StatInt_dup2,
+  StatInt_dup3,
+  StatInt_eventfd,
+  StatInt_signalfd,
+  StatInt_inotify_init,
+  StatInt_inotify_init1,
+  StatInt_socket,
+  StatInt_socketpair,
+  StatInt_connect,
+  StatInt_accept,
+  StatInt_accept4,
+  StatInt_epoll_create,
+  StatInt_epoll_create1,
+  StatInt_close,
+  StatInt___close,
+  StatInt_pipe,
+  StatInt_pipe2,
   StatInt_read,
   StatInt_pread,
   StatInt_pread64,
@@ -190,6 +212,8 @@ enum StatType {
   StatInt_recvmsg,
   StatInt_unlink,
   StatInt_fopen,
+  StatInt_freopen,
+  StatInt_fclose,
   StatInt_fread,
   StatInt_fwrite,
   StatInt_puts,
@@ -207,6 +231,7 @@ enum StatType {
   StatInt_usleep,
   StatInt_nanosleep,
   StatInt_gettimeofday,
+  StatInt_fork,
 
   // Dynamic annotations.
   StatAnnotation,
@@ -253,6 +278,8 @@ enum StatType {
   StatMtxSlab,
   StatMtxAnnotations,
   StatMtxAtExit,
+  StatMtxMBlock,
+  StatMtxJavaMBlock,
 
   // This must be the last.
   StatCnt
diff --git a/libsanitizer/tsan/tsan_symbolize_addr2line_linux.cc b/libsanitizer/tsan/tsan_symbolize_addr2line_linux.cc
index 23540c07ca9..9bdd1ffdc5e 100644
--- a/libsanitizer/tsan/tsan_symbolize_addr2line_linux.cc
+++ b/libsanitizer/tsan/tsan_symbolize_addr2line_linux.cc
@@ -102,11 +102,11 @@ static int dl_iterate_phdr_cb(dl_phdr_info *info, size_t size, void *arg) {
   m->base = (uptr)info->dlpi_addr;
   m->inp_fd = -1;
   m->out_fd = -1;
-  DPrintf("Module %s %zx\n", m->name, m->base);
+  DPrintf2("Module %s %zx\n", m->name, m->base);
   for (int i = 0; i < info->dlpi_phnum; i++) {
     const Elf64_Phdr *s = &info->dlpi_phdr[i];
-    DPrintf("  Section p_type=%zx p_offset=%zx p_vaddr=%zx p_paddr=%zx"
-            " p_filesz=%zx p_memsz=%zx p_flags=%zx p_align=%zx\n",
+    DPrintf2("  Section p_type=%zx p_offset=%zx p_vaddr=%zx p_paddr=%zx"
+        " p_filesz=%zx p_memsz=%zx p_flags=%zx p_align=%zx\n",
             (uptr)s->p_type, (uptr)s->p_offset, (uptr)s->p_vaddr,
             (uptr)s->p_paddr, (uptr)s->p_filesz, (uptr)s->p_memsz,
             (uptr)s->p_flags, (uptr)s->p_align);
@@ -119,7 +119,7 @@ static int dl_iterate_phdr_cb(dl_phdr_info *info, size_t size, void *arg) {
     sec->end = sec->base + s->p_memsz;
     sec->next = ctx->sections;
     ctx->sections = sec;
-    DPrintf("  Section %zx-%zx\n", sec->base, sec->end);
+    DPrintf2("  Section %zx-%zx\n", sec->base, sec->end);
   }
   return 0;
 }
diff --git a/libsanitizer/tsan/tsan_sync.cc b/libsanitizer/tsan/tsan_sync.cc
index 3bd1b35f9b3..d392408fd88 100644
--- a/libsanitizer/tsan/tsan_sync.cc
+++ b/libsanitizer/tsan/tsan_sync.cc
@@ -15,9 +15,10 @@
 
 namespace __tsan {
 
-SyncVar::SyncVar(uptr addr)
+SyncVar::SyncVar(uptr addr, u64 uid)
   : mtx(MutexTypeSyncVar, StatMtxSyncVar)
   , addr(addr)
+  , uid(uid)
   , owner_tid(kInvalidTid)
   , last_lock()
   , recursion()
@@ -45,9 +46,38 @@ SyncTab::~SyncTab() {
   }
 }
 
+SyncVar* SyncTab::GetOrCreateAndLock(ThreadState *thr, uptr pc,
+                                     uptr addr, bool write_lock) {
+  return GetAndLock(thr, pc, addr, write_lock, true);
+}
+
+SyncVar* SyncTab::GetIfExistsAndLock(uptr addr, bool write_lock) {
+  return GetAndLock(0, 0, addr, write_lock, false);
+}
+
+SyncVar* SyncTab::Create(ThreadState *thr, uptr pc, uptr addr) {
+  StatInc(thr, StatSyncCreated);
+  void *mem = internal_alloc(MBlockSync, sizeof(SyncVar));
+  const u64 uid = atomic_fetch_add(&uid_gen_, 1, memory_order_relaxed);
+  SyncVar *res = new(mem) SyncVar(addr, uid);
+#ifndef TSAN_GO
+  res->creation_stack.ObtainCurrent(thr, pc);
+#endif
+  return res;
+}
+
 SyncVar* SyncTab::GetAndLock(ThreadState *thr, uptr pc,
-                             uptr addr, bool write_lock) {
+                             uptr addr, bool write_lock, bool create) {
 #ifndef TSAN_GO
+  {  // NOLINT
+    SyncVar *res = GetJavaSync(thr, pc, addr, write_lock, create);
+    if (res)
+      return res;
+  }
+
+  // Here we ask only PrimaryAllocator, because
+  // SecondaryAllocator::PointerIsMine() is slow and we have fallback on
+  // the hashmap anyway.
   if (PrimaryAllocator::PointerIsMine((void*)addr)) {
     MBlock *b = user_mblock(thr, (void*)addr);
     Lock l(&b->mtx);
@@ -57,10 +87,9 @@ SyncVar* SyncTab::GetAndLock(ThreadState *thr, uptr pc,
         break;
     }
     if (res == 0) {
-      StatInc(thr, StatSyncCreated);
-      void *mem = internal_alloc(MBlockSync, sizeof(SyncVar));
-      res = new(mem) SyncVar(addr);
-      res->creation_stack.ObtainCurrent(thr, pc);
+      if (!create)
+        return 0;
+      res = Create(thr, pc, addr);
       res->next = b->head;
       b->head = res;
     }
@@ -85,6 +114,8 @@ SyncVar* SyncTab::GetAndLock(ThreadState *thr, uptr pc,
       }
     }
   }
+  if (!create)
+    return 0;
   {
     Lock l(&p->mtx);
     SyncVar *res = p->val;
@@ -93,12 +124,7 @@ SyncVar* SyncTab::GetAndLock(ThreadState *thr, uptr pc,
         break;
     }
     if (res == 0) {
-      StatInc(thr, StatSyncCreated);
-      void *mem = internal_alloc(MBlockSync, sizeof(SyncVar));
-      res = new(mem) SyncVar(addr);
-#ifndef TSAN_GO
-      res->creation_stack.ObtainCurrent(thr, pc);
-#endif
+      res = Create(thr, pc, addr);
       res->next = p->val;
       p->val = res;
     }
@@ -112,6 +138,11 @@ SyncVar* SyncTab::GetAndLock(ThreadState *thr, uptr pc,
 
 SyncVar* SyncTab::GetAndRemove(ThreadState *thr, uptr pc, uptr addr) {
 #ifndef TSAN_GO
+  {  // NOLINT
+    SyncVar *res = GetAndRemoveJavaSync(thr, pc, addr);
+    if (res)
+      return res;
+  }
   if (PrimaryAllocator::PointerIsMine((void*)addr)) {
     MBlock *b = user_mblock(thr, (void*)addr);
     SyncVar *res = 0;
diff --git a/libsanitizer/tsan/tsan_sync.h b/libsanitizer/tsan/tsan_sync.h
index 2912d2c0ddd..4dbb055a17e 100644
--- a/libsanitizer/tsan/tsan_sync.h
+++ b/libsanitizer/tsan/tsan_sync.h
@@ -48,12 +48,13 @@ class StackTrace {
 };
 
 struct SyncVar {
-  explicit SyncVar(uptr addr);
+  explicit SyncVar(uptr addr, u64 uid);
 
   static const int kInvalidTid = -1;
 
   Mutex mtx;
-  const uptr addr;
+  uptr addr;
+  const u64 uid;  // Globally unique id.
   SyncClock clock;
   SyncClock read_clock;  // Used for rw mutexes only.
   StackTrace creation_stack;
@@ -67,6 +68,18 @@ struct SyncVar {
   SyncVar *next;  // In SyncTab hashtable.
 
   uptr GetMemoryConsumption();
+  u64 GetId() const {
+    // 47 lsb is addr, then 14 bits is low part of uid, then 3 zero bits.
+    return GetLsb((u64)addr | (uid << 47), 61);
+  }
+  bool CheckId(u64 uid) const {
+    CHECK_EQ(uid, GetLsb(uid, 14));
+    return GetLsb(this->uid, 14) == uid;
+  }
+  static uptr SplitId(u64 id, u64 *uid) {
+    *uid = id >> 47;
+    return (uptr)GetLsb(id, 47);
+  }
 };
 
 class SyncTab {
@@ -74,13 +87,15 @@ class SyncTab {
   SyncTab();
   ~SyncTab();
 
-  // If the SyncVar does not exist yet, it is created.
-  SyncVar* GetAndLock(ThreadState *thr, uptr pc,
-                      uptr addr, bool write_lock);
+  SyncVar* GetOrCreateAndLock(ThreadState *thr, uptr pc,
+                              uptr addr, bool write_lock);
+  SyncVar* GetIfExistsAndLock(uptr addr, bool write_lock);
 
   // If the SyncVar does not exist, returns 0.
   SyncVar* GetAndRemove(ThreadState *thr, uptr pc, uptr addr);
 
+  SyncVar* Create(ThreadState *thr, uptr pc, uptr addr);
+
   uptr GetMemoryConsumption(uptr *nsync);
 
  private:
@@ -94,9 +109,13 @@ class SyncTab {
   // FIXME: Implement something more sane.
   static const int kPartCount = 1009;
   Part tab_[kPartCount];
+  atomic_uint64_t uid_gen_;
 
   int PartIdx(uptr addr);
 
+  SyncVar* GetAndLock(ThreadState *thr, uptr pc,
+                      uptr addr, bool write_lock, bool create);
+
   SyncTab(const SyncTab&);  // Not implemented.
   void operator = (const SyncTab&);  // Not implemented.
 };
diff --git a/libsanitizer/tsan/tsan_trace.h b/libsanitizer/tsan/tsan_trace.h
index 154cc15c083..69864838e26 100644
--- a/libsanitizer/tsan/tsan_trace.h
+++ b/libsanitizer/tsan/tsan_trace.h
@@ -14,6 +14,7 @@
 #include "tsan_defs.h"
 #include "tsan_mutex.h"
 #include "tsan_sync.h"
+#include "tsan_mutexset.h"
 
 namespace __tsan {
 
@@ -41,6 +42,7 @@ typedef u64 Event;
 struct TraceHeader {
   StackTrace stack0;  // Start stack for the trace.
   u64        epoch0;  // Start epoch for the trace.
+  MutexSet   mset0;
 #ifndef TSAN_GO
   uptr       stack0buf[kTraceStackSize];
 #endif