common: cache pthread names

This provides common ceph entrypoints for the pthread_[gs]name functions which
will also cache a thread_local copy. This also removes the pthread_t parameter
which precipitated the bug i50743.

Obviously, the overall goal here is to avoid system calls.

See-also: https://tracker.ceph.com/issues/50743
Fixes: 0be8d01c9d
Fixes: https://tracker.ceph.com/issues/68691
Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>

src/ceph_mds.cc

@@ -81,7 +81,7 @@ static void handle_mds_signal(int signum)
 
 int main(int argc, const char **argv)
 {
-  ceph_pthread_setname(pthread_self(), "ceph-mds");
+  ceph_pthread_setname("ceph-mds");
 
   auto args = argv_to_vec(argc, argv);
   if (args.empty()) {

src/ceph_mgr.cc

@@ -41,7 +41,7 @@ static void usage()
  */
 int main(int argc, const char **argv)
 {
-  ceph_pthread_setname(pthread_self(), "ceph-mgr");
+  ceph_pthread_setname("ceph-mgr");
 
   auto args = argv_to_vec(argc, argv);
   if (args.empty()) {

src/ceph_mon.cc

@@ -250,7 +250,7 @@ int main(int argc, const char **argv)
 {
   // reset our process name, in case we did a respawn, so that it's not
   // left as "exe".
-  ceph_pthread_setname(pthread_self(), "ceph-mon");
+  ceph_pthread_setname("ceph-mon");
 
   int err;
 

src/ceph_nvmeof_monitor_client.cc

@@ -45,7 +45,7 @@ static void usage()
  */
 int main(int argc, const char **argv)
 {
-  ceph_pthread_setname(pthread_self(), "ceph-nvmeof-monitor-client");
+  ceph_pthread_setname("ceph-nvmeof-monitor-client");
 
   auto args = argv_to_vec(argc, argv);
   if (args.empty()) {

src/client/SyntheticClient.cc

@@ -290,6 +290,7 @@ SyntheticClient::SyntheticClient(StandaloneClient *client, int w)
 
 void *synthetic_client_thread_entry(void *ptr)
 {
+  ceph_pthread_setname("client");
   SyntheticClient *sc = static_cast<SyntheticClient*>(ptr);
   //int r = 
   sc->run();
@@ -945,7 +946,6 @@ int SyntheticClient::start_thread()
 
   pthread_create(&thread_id, NULL, synthetic_client_thread_entry, this);
   ceph_assert(thread_id);
-  ceph_pthread_setname(thread_id, "client");
   return 0;
 }
 

src/common/Thread.cc

@@ -83,7 +83,7 @@ void *Thread::entry_wrapper()
   if (pid && cpuid >= 0)
     _set_affinity(cpuid);
 
-  ceph_pthread_setname(pthread_self(), Thread::thread_name.c_str());
+  ceph_pthread_setname(thread_name.c_str());
   return entry();
 }
 
@@ -154,7 +154,7 @@ int Thread::try_create(size_t stacksize)
 void Thread::create(const char *name, size_t stacksize)
 {
   ceph_assert(strlen(name) < 16);
-  Thread::thread_name = name;
+  thread_name = name;
 
   int ret = try_create(stacksize);
   if (ret != 0) {
@@ -203,24 +203,6 @@ int Thread::set_affinity(int id)
 // Functions for std::thread
 // =========================
 
-void set_thread_name(std::thread& t, const std::string& s) {
-  int r = ceph_pthread_setname(t.native_handle(), s.c_str());
-  if (r != 0) {
-    throw std::system_error(r, std::generic_category());
-  }
-}
-std::string get_thread_name(const std::thread& t) {
-  std::string s(256, '\0');
-
-  int r = ceph_pthread_getname(const_cast<std::thread&>(t).native_handle(),
-			       s.data(), s.length());
-  if (r != 0) {
-    throw std::system_error(r, std::generic_category());
-  }
-  s.resize(std::strlen(s.data()));
-  return s;
-}
-
 void kill(std::thread& t, int signal)
 {
   auto r = ceph_pthread_kill(t.native_handle(), signal);

src/common/Thread.h

@@ -27,7 +27,6 @@
 
 #include "include/ceph_assert.h"
 #include "include/compat.h"
-#include "include/spinlock.h"
 
 extern pid_t ceph_gettid();
 
@@ -36,7 +35,7 @@ class Thread {
   pthread_t thread_id;
   pid_t pid;
   int cpuid;
-  static inline thread_local std::string thread_name;
+  std::string thread_name;
 
   void *entry_wrapper();
 
@@ -64,15 +63,10 @@ class Thread {
   int join(void **prval = 0);
   int detach();
   int set_affinity(int cpuid);
-  static const std::string get_thread_name() {
-    return Thread::thread_name;
-  }
 };
 
 // Functions for with std::thread
 
-void set_thread_name(std::thread& t, const std::string& s);
-std::string get_thread_name(const std::thread& t);
 void kill(std::thread& t, int signal);
 
 template<typename Fun, typename... Args>
@@ -81,7 +75,7 @@ std::thread make_named_thread(std::string_view n,
 			      Args&& ...args) {
 
   return std::thread([n = std::string(n)](auto&& fun, auto&& ...args) {
-		       ceph_pthread_setname(pthread_self(), n.data());
+		       ceph_pthread_setname(n.data());
 		       std::invoke(std::forward<Fun>(fun),
 				   std::forward<Args>(args)...);
 		     }, std::forward<Fun>(fun), std::forward<Args>(args)...);

src/common/assert.cc

@@ -44,8 +44,7 @@ namespace ceph {
     g_assert_line = line;
     g_assert_func = func;
     g_assert_thread = (unsigned long long)pthread_self();
-    ceph_pthread_getname(pthread_self(), g_assert_thread_name,
-		       sizeof(g_assert_thread_name));
+    ceph_pthread_getname(g_assert_thread_name, sizeof(g_assert_thread_name));
 
     ostringstream tss;
     tss << ceph_clock_now();
@@ -122,8 +121,7 @@ namespace ceph {
     g_assert_line = line;
     g_assert_func = func;
     g_assert_thread = (unsigned long long)pthread_self();
-    ceph_pthread_getname(pthread_self(), g_assert_thread_name,
-		       sizeof(g_assert_thread_name));
+    ceph_pthread_getname(g_assert_thread_name, sizeof(g_assert_thread_name));
 
     BufAppender ba(g_assert_msg, sizeof(g_assert_msg));
     BackTrace *bt = new ClibBackTrace(1);
@@ -168,8 +166,7 @@ namespace ceph {
     g_assert_line = line;
     g_assert_func = func;
     g_assert_thread = (unsigned long long)pthread_self();
-    ceph_pthread_getname(pthread_self(), g_assert_thread_name,
-		       sizeof(g_assert_thread_name));
+    ceph_pthread_getname(g_assert_thread_name, sizeof(g_assert_thread_name));
 
     BackTrace *bt = new ClibBackTrace(1);
     snprintf(g_assert_msg, sizeof(g_assert_msg),
@@ -210,8 +207,7 @@ namespace ceph {
     g_assert_line = line;
     g_assert_func = func;
     g_assert_thread = (unsigned long long)pthread_self();
-    ceph_pthread_getname(pthread_self(), g_assert_thread_name,
-		       sizeof(g_assert_thread_name));
+    ceph_pthread_getname(g_assert_thread_name, sizeof(g_assert_thread_name));
 
     BufAppender ba(g_assert_msg, sizeof(g_assert_msg));
     BackTrace *bt = new ClibBackTrace(1);

src/common/ceph_timer.h

@@ -98,6 +98,7 @@ class timer {
   std::thread thread;
 
   void timer_thread() {
+    ceph_pthread_setname("ceph_timer");
     std::unique_lock l(lock);
     while (!suspended) {
       auto now = TC::now();
@@ -155,7 +156,6 @@ class timer {
 public:
   timer() : suspended(false) {
     thread = std::thread(&timer::timer_thread, this);
-    set_thread_name(thread, "ceph_timer");
   }
 
   // Create a suspended timer, jobs will be executed in order when

src/common/code_environment.cc

@@ -11,6 +11,7 @@
  * Foundation.  See file COPYING.
  *
  */
+#include "include/compat.h"
 
 #include "common/code_environment.h"
 
@@ -18,10 +19,6 @@
 
 #include "acconfig.h"
 
-#ifdef HAVE_PTHREAD_GETNAME_NP
-#include <pthread.h>
-#endif
-
 #include <string.h>
 
 code_environment_t g_code_env = CODE_ENVIRONMENT_UTILITY;
@@ -57,7 +54,7 @@ int get_process_name(char *buf, int len)
   }
   // FIPS zeroization audit 20191115: this memset is not security related.
   memset(buf, 0, len);
-  return pthread_getname_np(pthread_self(), buf, len);
+  return ceph_pthread_getname(buf, len);
 }
 
 #elif defined(HAVE_GETPROGNAME)

src/common/compat.cc

@@ -565,3 +565,66 @@ ssize_t get_self_exe_path(char* path, int buff_length) {
 }
 
 #endif /* _WIN32 */
+
+
+static thread_local char cached_thread_name[256]{};
+
+int ceph_pthread_setname(char const* name)
+{
+  strncpy(cached_thread_name, name, sizeof cached_thread_name - 1);
+#if defined(_WIN32) && defined(__clang__) && \
+    !defined(_LIBCPP_HAS_THREAD_API_PTHREAD)
+  // In this case, llvm doesn't use the pthread api for std::thread.
+  // We cannot use native_handle() with the pthread api, nor can we pass
+  // it to Windows API functions.
+  return 0;
+#elif defined(HAVE_PTHREAD_SETNAME_NP)
+  #if defined(__APPLE__)
+      return pthread_setname_np(name);
+  #else
+      return pthread_setname_np(pthread_self(), name);
+  #endif
+#elif defined(HAVE_PTHREAD_SET_NAME_NP)
+  pthread_set_name_np(pthread_self(), name);          \
+  return 0;
+#else
+  return 0;
+#endif
+}
+
+int ceph_pthread_getname(char* name, size_t len)
+{
+  if (cached_thread_name[0]) {
+    if (len > 0) {
+      strncpy(name, cached_thread_name, len);
+      name[len-1] = 0;
+    }
+    return 0;
+  } else {
+#if defined(_WIN32) && defined(__clang__) && \
+    !defined(_LIBCPP_HAS_THREAD_API_PTHREAD)
+    if (len > 0) {
+      strcpy(name, "");
+    }
+    return 0;
+#elif defined(HAVE_PTHREAD_GETNAME_NP) || defined(HAVE_PTHREAD_GET_NAME_NP)
+#  if defined(HAVE_PTHREAD_GETNAME_NP)
+    int rc = pthread_getname_np(pthread_self(), cached_thread_name, sizeof cached_thread_name);
+#  else
+    int rc = pthread_get_name_np(pthread_self(), cached_thread_name, sizeof cached_thread_name);
+#  endif
+    if (rc == 0) {
+      strncpy(name, cached_thread_name, len);
+      name[len-1] = 0;
+      return 0;
+    } else {
+      return rc;
+    }
+#else
+    if (len > 0) {
+      strcpy(name, "");
+    }
+    return 0;
+#endif
+  }
+}

src/common/obj_bencher.cc

@@ -99,6 +99,7 @@ ostream& ObjBencher::out(ostream& os)
 }
 
 void *ObjBencher::status_printer(void *_bencher) {
+  ceph_pthread_setname("OB::stat_print");
   ObjBencher *bencher = static_cast<ObjBencher *>(_bencher);
   bench_data& data = bencher->data;
   Formatter *formatter = bencher->formatter;
@@ -453,7 +454,6 @@ int ObjBencher::write_bench(int secondsToRun,
   pthread_t print_thread;
 
   pthread_create(&print_thread, NULL, ObjBencher::status_printer, (void *)this);
-  ceph_pthread_setname(print_thread, "write_stat");
   std::unique_lock locker{lock};
   data.finished = 0;
   data.start_time = mono_clock::now();
@@ -691,7 +691,6 @@ int ObjBencher::seq_read_bench(
 
   pthread_t print_thread;
   pthread_create(&print_thread, NULL, status_printer, (void *)this);
-  ceph_pthread_setname(print_thread, "seq_read_stat");
 
   mono_time finish_time = data.start_time + time_to_run;
   //start initial reads
@@ -903,7 +902,6 @@ int ObjBencher::rand_read_bench(
 
   pthread_t print_thread;
   pthread_create(&print_thread, NULL, status_printer, (void *)this);
-  ceph_pthread_setname(print_thread, "rand_read_stat");
 
   mono_time finish_time = data.start_time + time_to_run;
   //start initial reads

src/crimson/os/alienstore/thread_pool.cc

@@ -27,7 +27,7 @@ ThreadPool::ThreadPool(size_t n_threads,
         pin(*cpus);
       }
       block_sighup();
-      (void) pthread_setname_np(pthread_self(), "alien-store-tp");
+      (void) ceph_pthread_setname("alien-store-tp");
       loop(queue_max_wait, i);
     });
   }

src/global/signal_handler.cc

@@ -307,7 +307,7 @@ static void handle_oneshot_fatal_signal(int signum)
 
   char buf[1024];
   char pthread_name[16] = {0}; //limited by 16B include terminating null byte.
-  int r = ceph_pthread_getname(pthread_self(), pthread_name, sizeof(pthread_name));
+  int r = ceph_pthread_getname(pthread_name, sizeof(pthread_name));
   (void)r;
 #if defined(__sun)
   char message[SIG2STR_MAX];

src/include/compat.h

@@ -179,58 +179,10 @@ struct cpu_set_t;
 #define MSG_DONTWAIT MSG_NONBLOCK
 #endif
 
-/* compiler warning free success noop */
-#define pthread_setname_noop_helper(thread, name) ({ \
-  int __i = 0;                                       \
-  __i; })
-
-#define pthread_getname_noop_helper(thread, name, len) ({ \
-  if (name != NULL)                                       \
-    *name = '\0';                                         \
-  0; })
-
 #define pthread_kill_unsupported_helper(thread, signal) ({ \
   int __i = -ENOTSUP;                                      \
   __i; })
 
-#if defined(_WIN32) && defined(__clang__) && \
-    !defined(_LIBCPP_HAS_THREAD_API_PTHREAD)
-  // In this case, llvm doesn't use the pthread api for std::thread.
-  // We cannot use native_handle() with the pthread api, nor can we pass
-  // it to Windows API functions.
-  #define ceph_pthread_setname pthread_setname_noop_helper
-#elif defined(HAVE_PTHREAD_SETNAME_NP)
-  #if defined(__APPLE__)
-    #define ceph_pthread_setname(thread, name) ({ \
-      int __result = 0;                         \
-      if (thread == pthread_self())             \
-        __result = pthread_setname_np(name);    \
-      __result; })
-  #else
-    #define ceph_pthread_setname pthread_setname_np
-  #endif
-#elif defined(HAVE_PTHREAD_SET_NAME_NP)
-  /* Fix a small name diff and return 0 */
-  #define ceph_pthread_setname(thread, name) ({ \
-    pthread_set_name_np(thread, name);          \
-    0; })
-#else
-  #define ceph_pthread_setname pthread_setname_noop_helper
-#endif
-
-#if defined(_WIN32) && defined(__clang__) && \
-    !defined(_LIBCPP_HAS_THREAD_API_PTHREAD)
-  #define ceph_pthread_getname pthread_getname_noop_helper
-#elif defined(HAVE_PTHREAD_GETNAME_NP)
-  #define ceph_pthread_getname pthread_getname_np
-#elif defined(HAVE_PTHREAD_GET_NAME_NP)
-  #define ceph_pthread_getname(thread, name, len) ({ \
-    pthread_get_name_np(thread, name, len);          \
-    0; })
-#else
-  #define ceph_pthread_getname pthread_getname_noop_helper
-#endif
-
 #if defined(_WIN32) && defined(__clang__) && \
     !defined(_LIBCPP_HAS_THREAD_API_PTHREAD)
   #define ceph_pthread_kill pthread_kill_unsupported_helper
@@ -244,6 +196,9 @@ int ceph_posix_fallocate(int fd, off_t offset, off_t len);
 extern "C" {
 #endif
 
+int ceph_pthread_getname(char* name, size_t size);
+int ceph_pthread_setname(const char* name);
+
 int pipe_cloexec(int pipefd[2], int flags);
 char *ceph_strerror_r(int errnum, char *buf, size_t buflen);
 unsigned get_page_size();

src/log/Entry.h

@@ -24,6 +24,7 @@ namespace logging {
 class Entry {
 public:
   using time = log_time;
+  using thread_name_t = std::array<char, 16>;
 
   Entry() = delete;
   Entry(short pr, short sub) :
@@ -32,8 +33,7 @@ public:
     m_prio(pr),
     m_subsys(sub)
   {
-    strncpy(m_thread_name, Thread::get_thread_name().data(), 16);
-    m_thread_name[15] = '\0';
+    ceph_pthread_getname(m_thread_name.data(), m_thread_name.size());
   }
   Entry(const Entry &) = default;
   Entry& operator=(const Entry &) = default;
@@ -47,7 +47,7 @@ public:
   time m_stamp;
   pthread_t m_thread;
   short m_prio, m_subsys;
-  char m_thread_name[16];
+  thread_name_t m_thread_name{};
 
   static log_clock& clock() {
     static log_clock clock;

src/log/Log.cc

@@ -501,7 +501,7 @@ void Log::dump_recent()
     m_recent.clear();
     for (const auto& e : t) {
       auto& set = recent_pthread_ids[e.m_thread];
-      set.insert(e.m_thread_name);
+      set.insert(e.m_thread_name.data());
     }
     _flush(t, true);
   }

src/mds/Beacon.cc

@@ -74,7 +74,7 @@ void Beacon::init(const MDSMap &mdsmap)
   _notify_mdsmap(mdsmap);
 
   sender = std::thread([this]() {
-    ceph_pthread_setname(pthread_self(), "beacon");
+    ceph_pthread_setname("beacon");
     std::unique_lock<std::mutex> lock(mutex);
     bool sent;
     while (!finished) {

src/msg/async/Stack.h

@@ -352,7 +352,7 @@ class NetworkStack {
     static constexpr int TASK_COMM_LEN = 16;
     char tp_name[TASK_COMM_LEN];
     sprintf(tp_name, "msgr-worker-%u", id);
-    ceph_pthread_setname(pthread_self(), tp_name);
+    ceph_pthread_setname(tp_name);
   }
 
  protected:

src/msg/async/rdma/RDMAStack.cc

@@ -92,7 +92,6 @@ void RDMADispatcher::polling_start()
   ceph_assert(rx_cq);
 
   t = std::thread(&RDMADispatcher::polling, this);
-  ceph_pthread_setname(t.native_handle(), "rdma-polling");
 }
 
 void RDMADispatcher::polling_stop()
@@ -263,6 +262,7 @@ int RDMADispatcher::post_chunks_to_rq(int num, QueuePair *qp)
 
 void RDMADispatcher::polling()
 {
+  ceph_pthread_setname("rdma-polling");
   static int MAX_COMPLETIONS = 32;
   ibv_wc wc[MAX_COMPLETIONS];
 

src/rgw/driver/rados/rgw_notify.cc

@@ -21,6 +21,7 @@
 #include "common/dout.h"
 #include "rgw_url.h"
 #include <chrono>
+#include <fmt/format.h>
 
 #define dout_subsys ceph_subsys_rgw_notification
 
@@ -769,9 +770,10 @@ public:
         });
 
     // start the worker threads to do the actual queue processing
-    const std::string WORKER_THREAD_NAME = "notif-worker";
     for (auto worker_id = 0U; worker_id < worker_count; ++worker_id) {
-      workers.emplace_back([this]() {
+      workers.emplace_back([this,worker_id]() {
+        const auto thread_name = fmt::format("notif-worker-{}", worker_id);
+        ceph_pthread_setname(thread_name.c_str());
         try {
           io_context.run(); 
         } catch (const std::exception& err) {
@@ -779,11 +781,6 @@ public:
           throw err;
         }
       });
-      const auto thread_name = WORKER_THREAD_NAME+std::to_string(worker_id);
-      if (const auto rc = ceph_pthread_setname(workers.back().native_handle(), thread_name.c_str()); rc != 0) {
-        ldpp_dout(this, 1) << "ERROR: failed to set notification manager thread name to: " << thread_name
-          << ". error: " << rc << dendl;
-      }
     }
     ldpp_dout(this, 10) << "INfO: started notification manager with: " << worker_count << " workers" << dendl;
   }

src/rgw/rgw_amqp.cc

@@ -650,6 +650,9 @@ private:
   // (4) TODO reconnect on connection errors
   // (5) TODO cleanup timedout callbacks
   void run() noexcept {
+    // give the runner thread a name for easier debugging
+    ceph_pthread_setname("amqp_manager");
+
     amqp_frame_t frame;
     while (!stopped) {
 
@@ -838,12 +841,6 @@ public:
       // This is to prevent rehashing so that iterators are not invalidated
       // when a new connection is added.
       connections.max_load_factor(10.0);
-      // give the runner thread a name for easier debugging
-      const char* thread_name = "amqp_manager";
-      if (const auto rc = ceph_pthread_setname(runner.native_handle(), thread_name); rc != 0) {
-        ldout(cct, 1) << "ERROR: failed to set amqp manager thread name to: " << thread_name
-          << ". error: " << rc << dendl;
-      }
   }
 
   // non copyable

src/rgw/rgw_kafka.cc

@@ -503,6 +503,7 @@ private:
   }
 
   void run() noexcept {
+    ceph_pthread_setname("kafka_manager");
     while (!stopped) {
 
       // publish all messages in the queue
@@ -575,12 +576,6 @@ public:
       // This is to prevent rehashing so that iterators are not invalidated 
       // when a new connection is added.
       connections.max_load_factor(10.0);
-      // give the runner thread a name for easier debugging
-      const char* thread_name = "kafka_manager";
-      if (const auto rc = ceph_pthread_setname(runner.native_handle(), thread_name); rc != 0) {
-        ldout(cct, 1) << "ERROR: failed to set kafka manager thread name to: " << thread_name
-          << ". error: " << rc << dendl;
-      }
   }
 
   // non copyable

src/rgw/rgw_lua_background.cc

@@ -83,11 +83,6 @@ void Background::start() {
   }
   started = true;
   runner = std::thread(&Background::run, this);
-  const char* thread_name = "lua_background";
-  if (const auto rc = ceph_pthread_setname(runner.native_handle(), thread_name); rc != 0) {
-    ldout(cct, 1) << "ERROR: failed to set lua background thread name to: " << thread_name
-      << ". error: " << rc << dendl;
-  }
 }
 
 void Background::pause() {
@@ -127,6 +122,7 @@ const BackgroundMapValue& Background::get_table_value(const std::string& key) co
 //(2) Executes the script
 //(3) Sleep (configurable)
 void Background::run() {
+  ceph_pthread_setname("lua_background");
   const DoutPrefixProvider* const dpp = &dp;
   lua_state_guard lguard(cct->_conf->rgw_lua_max_memory_per_state, dpp);
   auto L = lguard.get();

src/rgw/rgw_ratelimit.h

@@ -239,6 +239,7 @@ class ActiveRateLimiter : public DoutPrefix  {
   std::atomic_uint8_t current_active = 0;
   std::shared_ptr<RateLimiter> ratelimit[2];
   void replace_active() {
+    ceph_pthread_setname("ratelimit_gc");
     using namespace std::chrono_literals;
     std::unique_lock<std::mutex> lk(cv_m);
     while (!stopped) {
@@ -286,8 +287,5 @@ class ActiveRateLimiter : public DoutPrefix  {
     void start() {
       ldpp_dout(this, 20) << "starting ratelimit_gc thread" << dendl;
       runner = std::thread(&ActiveRateLimiter::replace_active, this);
-      if (const auto rc = ceph_pthread_setname(runner.native_handle(), "ratelimit_gc"); rc != 0) {
-        ldpp_dout(this, 1) << "ERROR: failed to set ratelimit_gc thread name. error: " << rc << dendl;
-      }
     }
 };