tensorflow/core/platform/default/env.cc

/* Copyright 2015 The TensorFlow Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/

#include <dirent.h>
#include <dlfcn.h>
#include <errno.h>
#include <fcntl.h>
#include <fnmatch.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/types.h>
#include <time.h>
#include <unistd.h>

#ifdef __FreeBSD__
#include <pthread_np.h>
#endif

#include <thread>
#include <vector>

#include "tensorflow/core/platform/default/posix_file_system.h"
#include "tensorflow/core/platform/env.h"
#include "tensorflow/core/platform/load_library.h"
#include "tensorflow/core/platform/logging.h"
#include "tensorflow/core/platform/mutex.h"
#include "tensorflow/core/platform/ram_file_system.h"
#include "tensorflow/core/platform/strcat.h"
#include "tensorflow/core/protobuf/error_codes.pb.h"

namespace tensorflow {

namespace {

mutex name_mutex(tensorflow::LINKER_INITIALIZED);

std::map<std::thread::id, string>& GetThreadNameRegistry()
    TF_EXCLUSIVE_LOCKS_REQUIRED(name_mutex) {
  static auto* thread_name_registry = new std::map<std::thread::id, string>();
  return *thread_name_registry;
}

// We use the pthread API instead of std::thread so we can control stack sizes.
class PThread : public Thread {
 public:
  PThread(const ThreadOptions& thread_options, const std::string& name,
          std::function<void()> fn) {
    ThreadParams* params = new ThreadParams;
    params->name = name;
    params->fn = std::move(fn);
    pthread_attr_t attributes;
    pthread_attr_init(&attributes);
    if (thread_options.stack_size != 0) {
      pthread_attr_setstacksize(&attributes, thread_options.stack_size);
    }
    int ret = pthread_create(&thread_, &attributes, &ThreadFn, params);
    // There is no mechanism for the thread creation API to fail, so we CHECK.
    CHECK_EQ(ret, 0) << "Thread " << name << " creation via pthread_create() failed.";
    pthread_attr_destroy(&attributes);
  }

  ~PThread() override { pthread_join(thread_, nullptr); }

 private:
  struct ThreadParams {
    std::string name;
    std::function<void()> fn;
  };
  static void* ThreadFn(void* params_arg) {
    std::unique_ptr<ThreadParams> params(
        reinterpret_cast<ThreadParams*>(params_arg));
    {
      mutex_lock l(name_mutex);
      GetThreadNameRegistry().emplace(std::this_thread::get_id(), params->name);
    }
    params->fn();
    {
      mutex_lock l(name_mutex);
      GetThreadNameRegistry().erase(std::this_thread::get_id());
    }
    return nullptr;
  }

  pthread_t thread_;
};

class PosixEnv : public Env {
 public:
  PosixEnv() {}

  ~PosixEnv() override { LOG(FATAL) << "Env::Default() must not be destroyed"; }

  bool MatchPath(const string& path, const string& pattern) override {
    return fnmatch(pattern.c_str(), path.c_str(), FNM_PATHNAME) == 0;
  }

  void SleepForMicroseconds(int64 micros) override {
    while (micros > 0) {
      timespec sleep_time;
      sleep_time.tv_sec = 0;
      sleep_time.tv_nsec = 0;

      if (micros >= 1e6) {
        sleep_time.tv_sec =
            std::min<int64>(micros / 1e6, std::numeric_limits<time_t>::max());
        micros -= static_cast<int64>(sleep_time.tv_sec) * 1e6;
      }
      if (micros < 1e6) {
        sleep_time.tv_nsec = 1000 * micros;
        micros = 0;
      }
      while (nanosleep(&sleep_time, &sleep_time) != 0 && errno == EINTR) {
        // Ignore signals and wait for the full interval to elapse.
      }
    }
  }

  Thread* StartThread(const ThreadOptions& thread_options, const string& name,
                      std::function<void()> fn) override {
    return new PThread(thread_options, name, fn);
  }

  int32 GetCurrentThreadId() override {
    static thread_local int32 current_thread_id = GetCurrentThreadIdInternal();
    return current_thread_id;
  }

  bool GetCurrentThreadName(string* name) override {
    {
      mutex_lock l(name_mutex);
      auto thread_name =
          GetThreadNameRegistry().find(std::this_thread::get_id());
      if (thread_name != GetThreadNameRegistry().end()) {
        *name = strings::StrCat(thread_name->second, "/", GetCurrentThreadId());
        return true;
      }
    }
#if defined(__GLIBC__) || defined(__FreeBSD__)
    char buf[100];
#ifdef __FreeBSD__
    int res = 0;
    pthread_get_name_np(pthread_self(), buf, static_cast<size_t>(100));
#else
    int res = pthread_getname_np(pthread_self(), buf, static_cast<size_t>(100));
#endif
    if (res != 0) {
      return false;
    }
    *name = buf;
    return true;
#else
    return false;
#endif
  }

  void SchedClosure(std::function<void()> closure) override {
    // TODO(b/27290852): Spawning a new thread here is wasteful, but
    // needed to deal with the fact that many `closure` functions are
    // blocking in the current codebase.
    std::thread closure_thread(closure);
    closure_thread.detach();
  }

  void SchedClosureAfter(int64 micros, std::function<void()> closure) override {
    // TODO(b/27290852): Consuming a thread here is wasteful, but this
    // code is (currently) only used in the case where a step fails
    // (AbortStep). This could be replaced by a timer thread
    SchedClosure([this, micros, closure]() {
      SleepForMicroseconds(micros);
      closure();
    });
  }

  Status LoadDynamicLibrary(const char* library_filename,
                            void** handle) override {
    return tensorflow::internal::LoadDynamicLibrary(library_filename, handle);
  }

  Status GetSymbolFromLibrary(void* handle, const char* symbol_name,
                              void** symbol) override {
    return tensorflow::internal::GetSymbolFromLibrary(handle, symbol_name,
                                                      symbol);
  }

  string FormatLibraryFileName(const string& name,
                               const string& version) override {
    return tensorflow::internal::FormatLibraryFileName(name, version);
  }

  string GetRunfilesDir() override {
    string bin_path = this->GetExecutablePath();
    string runfiles_suffix = ".runfiles/org_tensorflow";
    std::size_t pos = bin_path.find(runfiles_suffix);

    // Sometimes (when executing under python) bin_path returns the full path to
    // the python scripts under runfiles. Get the substring.
    if (pos != std::string::npos) {
      return bin_path.substr(0, pos + runfiles_suffix.length());
    }

    // See if we have the executable path. if executable.runfiles exists, return
    // that folder.
    string runfiles_path = bin_path + runfiles_suffix;
    Status s = this->IsDirectory(runfiles_path);
    if (s.ok()) {
      return runfiles_path;
    }

    // If nothing can be found, return something close.
    return bin_path.substr(0, bin_path.find_last_of("/\\"));
  }

 private:
  void GetLocalTempDirectories(std::vector<string>* list) override;

  int32 GetCurrentThreadIdInternal() {
#ifdef __APPLE__
    uint64_t tid64;
    pthread_threadid_np(nullptr, &tid64);
    return static_cast<int32>(tid64);
#elif defined(__FreeBSD__)
    return pthread_getthreadid_np();
#elif defined(__NR_gettid)
    return static_cast<int32>(syscall(__NR_gettid));
#else
    return std::hash<std::thread::id>()(std::this_thread::get_id());
#endif
  }
};

#if defined(LIBTPU_ON_GCE)
// This is a temporary fix for including GCS file system on TPU builds.
// Will be removed once b/176954917 is fully resolved with the build fix.
bool RegisterGcsFileSystemForTpu() {
  int fd = shm_open(absl::StrCat("/tmp_tf_gcs_fs_pointer_", getpid()).data(),
                    O_RDWR, S_IRUSR | S_IWUSR);
  if (fd == -1) {
    LOG(WARNING) << "Unable to register GCS file system for the TPU build.";
    return false;
  }

  void* (**fn)() = reinterpret_cast<void* (**)()>(mmap(
      nullptr, sizeof(void* (*)()), PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0));
  if (fn == MAP_FAILED) {
    LOG(WARNING) << "Unable to register GCS file system for the TPU build.";
    return false;
  }

  FileSystem* fs = reinterpret_cast<FileSystem*>((*fn)());
  tensorflow::Env::Default()
      ->RegisterFileSystem("gs", std::unique_ptr<FileSystem>(fs))
      .IgnoreError();

  munmap(fn, sizeof(void* (*)()));
  close(fd);
  return true;
}
#endif

}  // namespace

#if defined(PLATFORM_POSIX) || defined(__APPLE__) || defined(__ANDROID__)
REGISTER_FILE_SYSTEM("", PosixFileSystem);
REGISTER_FILE_SYSTEM("file", LocalPosixFileSystem);
REGISTER_FILE_SYSTEM("ram", RamFileSystem);

#if defined(LIBTPU_ON_GCE)
bool register_gcs_for_tpu = RegisterGcsFileSystemForTpu();
#endif

Env* Env::Default() {
  static Env* default_env = new PosixEnv;
  return default_env;
}
#endif

void PosixEnv::GetLocalTempDirectories(std::vector<string>* list) {
  list->clear();
  // Directories, in order of preference. If we find a dir that
  // exists, we stop adding other less-preferred dirs
  const char* candidates[] = {
    // Non-null only during unittest/regtest
    getenv("TEST_TMPDIR"),

    // Explicitly-supplied temp dirs
    getenv("TMPDIR"),
    getenv("TMP"),

#if defined(__ANDROID__)
    "/data/local/tmp",
#endif

    // If all else fails
    "/tmp",
  };

  for (const char* d : candidates) {
    if (!d || d[0] == '\0') continue;  // Empty env var

    // Make sure we don't surprise anyone who's expecting a '/'
    string dstr = d;
    if (dstr[dstr.size() - 1] != '/') {
      dstr += "/";
    }

    struct stat statbuf;
    if (!stat(d, &statbuf) && S_ISDIR(statbuf.st_mode) &&
        !access(dstr.c_str(), 0)) {
      // We found a dir that exists and is accessible - we're done.
      list->push_back(dstr);
      return;
    }
  }
}

int setenv(const char* name, const char* value, int overwrite) {
  return ::setenv(name, value, overwrite);
}

int unsetenv(const char* name) { return ::unsetenv(name); }

}  // namespace tensorflow