ddtrace 0.54.2 → 1.8.0

data/ext/ddtrace_profiling_native_extension/private_vm_api_access.h

@@ -1,3 +1,45 @@
 #pragma once
 
+#include <stdbool.h>
+
+// The private_vm_api_access.c includes the RUBY_MJIT_HEADER which replaces and conflicts with any other Ruby headers;
+// so we use PRIVATE_VM_API_ACCESS_SKIP_RUBY_INCLUDES to be able to include private_vm_api_access.h on that file
+// without also dragging the incompatible includes
+#ifndef PRIVATE_VM_API_ACCESS_SKIP_RUBY_INCLUDES
+  #include <ruby/thread_native.h>
+  #include <ruby/vm.h>
+#endif
+
+#include "extconf.h"
+
+// Contains the current gvl owner, and a flag to indicate if it is valid
+typedef struct {
+  bool valid;
+  rb_nativethread_id_t owner;
+} current_gvl_owner;
+
 rb_nativethread_id_t pthread_id_for(VALUE thread);
+bool is_current_thread_holding_the_gvl(void);
+current_gvl_owner gvl_owner(void);
+uint64_t native_thread_id_for(VALUE thread);
+ptrdiff_t stack_depth_for(VALUE thread);
+VALUE ddtrace_thread_list(void);
+bool is_thread_alive(VALUE thread);
+VALUE thread_name_for(VALUE thread);
+
+int ddtrace_rb_profile_frames(VALUE thread, int start, int limit, VALUE *buff, int *lines, bool* is_ruby_frame);
+// Returns true if the current thread belongs to the main Ractor or if Ruby has no Ractor support
+bool ddtrace_rb_ractor_main_p(void);
+
+// Ruby 3.0 finally added support for showing CFUNC frames (frames for methods written using native code)
+// in stack traces gathered via `rb_profile_frames` (https://github.com/ruby/ruby/pull/3299).
+// To access this information on older Rubies, beyond using our custom `ddtrace_rb_profile_frames` above, we also need
+// to backport the Ruby 3.0+ version of `rb_profile_frame_method_name`.
+#ifdef USE_BACKPORTED_RB_PROFILE_FRAME_METHOD_NAME
+  VALUE ddtrace_rb_profile_frame_method_name(VALUE frame);
+#else // Ruby > 3.0, just use the stock functionality
+  #define ddtrace_rb_profile_frame_method_name rb_profile_frame_method_name
+#endif
+
+// See comment on `record_placeholder_stack_in_native_code` for a full explanation of what this means (and why we don't just return 0)
+#define PLACEHOLDER_STACK_IN_NATIVE_CODE -1

data/ext/ddtrace_profiling_native_extension/profiling.c

@@ -1,10 +1,39 @@
 #include <ruby.h>
+#include <ruby/thread.h>
+#include <errno.h>
 
 #include "clock_id.h"
+#include "helpers.h"
+#include "private_vm_api_access.h"
+#include "ruby_helpers.h"
+#include "setup_signal_handler.h"
+#include "time_helpers.h"
+
+// Each class/module here is implemented in their separate file
+void collectors_cpu_and_wall_time_init(VALUE profiling_module);
+void collectors_cpu_and_wall_time_worker_init(VALUE profiling_module);
+void collectors_dynamic_sampling_rate_init(VALUE profiling_module);
+void collectors_idle_sampling_helper_init(VALUE profiling_module);
+void collectors_stack_init(VALUE profiling_module);
+void http_transport_init(VALUE profiling_module);
+void stack_recorder_init(VALUE profiling_module);
 
 static VALUE native_working_p(VALUE self);
+static VALUE _native_grab_gvl_and_raise(DDTRACE_UNUSED VALUE _self, VALUE exception_class, VALUE test_message, VALUE test_message_arg, VALUE release_gvl);
+static void *trigger_grab_gvl_and_raise(void *trigger_args);
+static VALUE _native_grab_gvl_and_raise_syserr(DDTRACE_UNUSED VALUE _self, VALUE syserr_errno, VALUE test_message, VALUE test_message_arg, VALUE release_gvl);
+static void *trigger_grab_gvl_and_raise_syserr(void *trigger_args);
+static VALUE _native_ddtrace_rb_ractor_main_p(DDTRACE_UNUSED VALUE _self);
+static VALUE _native_is_current_thread_holding_the_gvl(DDTRACE_UNUSED VALUE _self);
+static VALUE _native_release_gvl_and_call_is_current_thread_holding_the_gvl(DDTRACE_UNUSED VALUE _self);
+static void *testing_is_current_thread_holding_the_gvl(DDTRACE_UNUSED void *_unused);
+static VALUE _native_install_holding_the_gvl_signal_handler(DDTRACE_UNUSED VALUE _self);
+static void holding_the_gvl_signal_handler(DDTRACE_UNUSED int _signal, DDTRACE_UNUSED siginfo_t *_info, DDTRACE_UNUSED void *_ucontext);
+static VALUE _native_trigger_holding_the_gvl_signal_handler_on(DDTRACE_UNUSED VALUE _self, VALUE background_thread);
+static VALUE _native_enforce_success(DDTRACE_UNUSED VALUE _self, VALUE syserr_errno, VALUE with_gvl);
+static void *trigger_enforce_success(void *trigger_args);
 
-void Init_ddtrace_profiling_native_extension(void) {
+void DDTRACE_EXPORT Init_ddtrace_profiling_native_extension(void) {
   VALUE datadog_module = rb_define_module("Datadog");
   VALUE profiling_module = rb_define_module_under(datadog_module, "Profiling");
   VALUE native_extension_module = rb_define_module_under(profiling_module, "NativeExtension");
@@ -13,10 +42,210 @@ void Init_ddtrace_profiling_native_extension(void) {
   rb_funcall(native_extension_module, rb_intern("private_class_method"), 1, ID2SYM(rb_intern("native_working?")));
 
   rb_define_singleton_method(native_extension_module, "clock_id_for", clock_id_for, 1); // from clock_id.h
+
+  collectors_cpu_and_wall_time_init(profiling_module);
+  collectors_cpu_and_wall_time_worker_init(profiling_module);
+  collectors_dynamic_sampling_rate_init(profiling_module);
+  collectors_idle_sampling_helper_init(profiling_module);
+  collectors_stack_init(profiling_module);
+  http_transport_init(profiling_module);
+  stack_recorder_init(profiling_module);
+
+  // Hosts methods used for testing the native code using RSpec
+  VALUE testing_module = rb_define_module_under(native_extension_module, "Testing");
+  rb_define_singleton_method(testing_module, "_native_grab_gvl_and_raise", _native_grab_gvl_and_raise, 4);
+  rb_define_singleton_method(testing_module, "_native_grab_gvl_and_raise_syserr", _native_grab_gvl_and_raise_syserr, 4);
+  rb_define_singleton_method(testing_module, "_native_ddtrace_rb_ractor_main_p", _native_ddtrace_rb_ractor_main_p, 0);
+  rb_define_singleton_method(testing_module, "_native_is_current_thread_holding_the_gvl", _native_is_current_thread_holding_the_gvl, 0);
+  rb_define_singleton_method(
+    testing_module,
+    "_native_release_gvl_and_call_is_current_thread_holding_the_gvl",
+    _native_release_gvl_and_call_is_current_thread_holding_the_gvl,
+    0
+  );
+  rb_define_singleton_method(testing_module, "_native_install_holding_the_gvl_signal_handler", _native_install_holding_the_gvl_signal_handler, 0);
+  rb_define_singleton_method(testing_module, "_native_trigger_holding_the_gvl_signal_handler_on", _native_trigger_holding_the_gvl_signal_handler_on, 1);
+  rb_define_singleton_method(testing_module, "_native_enforce_success", _native_enforce_success, 2);
 }
 
-static VALUE native_working_p(VALUE self) {
+static VALUE native_working_p(DDTRACE_UNUSED VALUE _self) {
   self_test_clock_id();
 
   return Qtrue;
 }
+
+struct trigger_grab_gvl_and_raise_arguments {
+  VALUE exception_class;
+  char *test_message;
+  int test_message_arg;
+};
+
+static VALUE _native_grab_gvl_and_raise(DDTRACE_UNUSED VALUE _self, VALUE exception_class, VALUE test_message, VALUE test_message_arg, VALUE release_gvl) {
+  ENFORCE_TYPE(test_message, T_STRING);
+
+  struct trigger_grab_gvl_and_raise_arguments args;
+
+  args.exception_class = exception_class;
+  args.test_message = StringValueCStr(test_message);
+  args.test_message_arg = test_message_arg != Qnil ? NUM2INT(test_message_arg) : -1;
+
+  if (RTEST(release_gvl)) {
+    rb_thread_call_without_gvl(trigger_grab_gvl_and_raise, &args, NULL, NULL);
+  } else {
+    grab_gvl_and_raise(args.exception_class, "%s", args.test_message);
+  }
+
+  rb_raise(rb_eRuntimeError, "Failed to raise exception in _native_grab_gvl_and_raise; this should never happen");
+}
+
+static void *trigger_grab_gvl_and_raise(void *trigger_args) {
+  struct trigger_grab_gvl_and_raise_arguments *args = (struct trigger_grab_gvl_and_raise_arguments *) trigger_args;
+
+  if (args->test_message_arg >= 0) {
+    grab_gvl_and_raise(args->exception_class, "%s%d", args->test_message, args->test_message_arg);
+  } else {
+    grab_gvl_and_raise(args->exception_class, "%s", args->test_message);
+  }
+
+  return NULL;
+}
+
+struct trigger_grab_gvl_and_raise_syserr_arguments {
+  int syserr_errno;
+  char *test_message;
+  int test_message_arg;
+};
+
+static VALUE _native_grab_gvl_and_raise_syserr(DDTRACE_UNUSED VALUE _self, VALUE syserr_errno, VALUE test_message, VALUE test_message_arg, VALUE release_gvl) {
+  ENFORCE_TYPE(test_message, T_STRING);
+
+  struct trigger_grab_gvl_and_raise_syserr_arguments args;
+
+  args.syserr_errno = NUM2INT(syserr_errno);
+  args.test_message = StringValueCStr(test_message);
+  args.test_message_arg = test_message_arg != Qnil ? NUM2INT(test_message_arg) : -1;
+
+  if (RTEST(release_gvl)) {
+    rb_thread_call_without_gvl(trigger_grab_gvl_and_raise_syserr, &args, NULL, NULL);
+  } else {
+    grab_gvl_and_raise_syserr(args.syserr_errno, "%s", args.test_message);
+  }
+
+  rb_raise(rb_eRuntimeError, "Failed to raise exception in _native_grab_gvl_and_raise_syserr; this should never happen");
+}
+
+static void *trigger_grab_gvl_and_raise_syserr(void *trigger_args) {
+  struct trigger_grab_gvl_and_raise_syserr_arguments *args = (struct trigger_grab_gvl_and_raise_syserr_arguments *) trigger_args;
+
+  if (args->test_message_arg >= 0) {
+    grab_gvl_and_raise_syserr(args->syserr_errno, "%s%d", args->test_message, args->test_message_arg);
+  } else {
+    grab_gvl_and_raise_syserr(args->syserr_errno, "%s", args->test_message);
+  }
+
+  return NULL;
+}
+
+static VALUE _native_ddtrace_rb_ractor_main_p(DDTRACE_UNUSED VALUE _self) {
+  return ddtrace_rb_ractor_main_p() ? Qtrue : Qfalse;
+}
+
+static VALUE _native_is_current_thread_holding_the_gvl(DDTRACE_UNUSED VALUE _self) {
+  return ((bool) testing_is_current_thread_holding_the_gvl(NULL)) ? Qtrue : Qfalse;
+}
+
+static VALUE _native_release_gvl_and_call_is_current_thread_holding_the_gvl(DDTRACE_UNUSED VALUE _self) {
+  return ((bool) rb_thread_call_without_gvl(testing_is_current_thread_holding_the_gvl, NULL, NULL, NULL)) ? Qtrue : Qfalse;
+}
+
+static void *testing_is_current_thread_holding_the_gvl(DDTRACE_UNUSED void *_unused) {
+  return (void *) is_current_thread_holding_the_gvl();
+}
+
+static VALUE _native_install_holding_the_gvl_signal_handler(DDTRACE_UNUSED VALUE _self) {
+  install_sigprof_signal_handler(holding_the_gvl_signal_handler, "holding_the_gvl_signal_handler");
+  return Qtrue;
+}
+
+static pthread_mutex_t holding_the_gvl_signal_handler_mutex = PTHREAD_MUTEX_INITIALIZER;
+static pthread_cond_t holding_the_gvl_signal_handler_executed = PTHREAD_COND_INITIALIZER;
+static VALUE holding_the_gvl_signal_handler_result[3];
+
+// Ruby VM API that is exported but not present in the header files. Only used by holding_the_gvl_signal_handler below and SHOULD NOT
+// be used in any other situation. See the comments on is_current_thread_holding_the_gvl for details.
+int ruby_thread_has_gvl_p(void);
+
+static void holding_the_gvl_signal_handler(DDTRACE_UNUSED int _signal, DDTRACE_UNUSED siginfo_t *_info, DDTRACE_UNUSED void *_ucontext) {
+  pthread_mutex_lock(&holding_the_gvl_signal_handler_mutex);
+
+  VALUE test_executed = Qtrue;
+  VALUE ruby_thread_has_gvl_p_result = ruby_thread_has_gvl_p() ? Qtrue : Qfalse;
+  VALUE is_current_thread_holding_the_gvl_result = is_current_thread_holding_the_gvl() ? Qtrue : Qfalse;
+
+  holding_the_gvl_signal_handler_result[0] = test_executed;
+  holding_the_gvl_signal_handler_result[1] = ruby_thread_has_gvl_p_result;
+  holding_the_gvl_signal_handler_result[2] = is_current_thread_holding_the_gvl_result;
+
+  pthread_cond_broadcast(&holding_the_gvl_signal_handler_executed);
+  pthread_mutex_unlock(&holding_the_gvl_signal_handler_mutex);
+}
+
+static VALUE _native_trigger_holding_the_gvl_signal_handler_on(DDTRACE_UNUSED VALUE _self, VALUE background_thread) {
+  holding_the_gvl_signal_handler_result[0] = Qfalse;
+  holding_the_gvl_signal_handler_result[1] = Qfalse;
+  holding_the_gvl_signal_handler_result[2] = Qfalse;
+
+  rb_nativethread_id_t thread = pthread_id_for(background_thread);
+
+  ENFORCE_SUCCESS_GVL(pthread_mutex_lock(&holding_the_gvl_signal_handler_mutex));
+
+  // We keep trying for ~5 seconds (500 x 10ms) to try to avoid any flakiness if the test machine is a bit slow
+  for (int tries = 0; holding_the_gvl_signal_handler_result[0] == Qfalse && tries < 500; tries++) {
+    pthread_kill(thread, SIGPROF);
+
+    // pthread_cond_timedwait is simply awful -- the deadline is based on wall-clock using a struct timespec, so we need
+    // all of the below complexity just to tell it "timeout is 10ms". The % limit dance below is needed because the
+    // `tv_nsec` part of a timespec can't go over the limit.
+    struct timespec deadline;
+    clock_gettime(CLOCK_REALTIME, &deadline);
+
+    unsigned int timeout_ns = MILLIS_AS_NS(10);
+    unsigned int tv_nsec_limit = SECONDS_AS_NS(1);
+    if ((deadline.tv_nsec + timeout_ns) < tv_nsec_limit) {
+      deadline.tv_nsec += timeout_ns;
+    } else {
+      deadline.tv_nsec = (deadline.tv_nsec + timeout_ns) % tv_nsec_limit;
+      deadline.tv_sec++;
+    }
+
+    int error = pthread_cond_timedwait(&holding_the_gvl_signal_handler_executed, &holding_the_gvl_signal_handler_mutex, &deadline);
+    if (error && error != ETIMEDOUT) ENFORCE_SUCCESS_GVL(error);
+  }
+
+  ENFORCE_SUCCESS_GVL(pthread_mutex_unlock(&holding_the_gvl_signal_handler_mutex));
+
+  replace_sigprof_signal_handler_with_empty_handler(holding_the_gvl_signal_handler);
+
+  if (holding_the_gvl_signal_handler_result[0] == Qfalse) rb_raise(rb_eRuntimeError, "Could not signal background_thread");
+
+  VALUE result = rb_hash_new();
+  rb_hash_aset(result, ID2SYM(rb_intern("ruby_thread_has_gvl_p")), holding_the_gvl_signal_handler_result[1]);
+  rb_hash_aset(result, ID2SYM(rb_intern("is_current_thread_holding_the_gvl")), holding_the_gvl_signal_handler_result[2]);
+  return result;
+}
+
+static VALUE _native_enforce_success(DDTRACE_UNUSED VALUE _self, VALUE syserr_errno, VALUE with_gvl) {
+  if (RTEST(with_gvl)) {
+    ENFORCE_SUCCESS_GVL(NUM2INT(syserr_errno));
+  } else {
+    rb_thread_call_without_gvl(trigger_enforce_success, (void *) (intptr_t) NUM2INT(syserr_errno), NULL, NULL);
+  }
+
+  return Qtrue;
+}
+
+static void *trigger_enforce_success(void *trigger_args) {
+  intptr_t syserr_errno = (intptr_t) trigger_args;
+  ENFORCE_SUCCESS_NO_GVL(syserr_errno);
+  return NULL;
+}

data/ext/ddtrace_profiling_native_extension/ruby_helpers.c

@@ -0,0 +1,110 @@
+#include <ruby.h>
+#include <ruby/thread.h>
+
+#include "ruby_helpers.h"
+#include "private_vm_api_access.h"
+
+void raise_unexpected_type(
+  VALUE value,
+  const char *value_name,
+  const char *type_name,
+  const char *file,
+  int line,
+  const char* function_name
+) {
+  rb_exc_raise(
+    rb_exc_new_str(
+      rb_eTypeError,
+      rb_sprintf("wrong argument %"PRIsVALUE" for '%s' (expected a %s) at %s:%d:in `%s'",
+        rb_inspect(value),
+        value_name,
+        type_name,
+        file,
+        line,
+        function_name
+      )
+    )
+  );
+}
+
+#define MAX_RAISE_MESSAGE_SIZE 256
+
+struct raise_arguments {
+  VALUE exception_class;
+  char exception_message[MAX_RAISE_MESSAGE_SIZE];
+};
+
+static void *trigger_raise(void *raise_arguments) {
+  struct raise_arguments *args = (struct raise_arguments *) raise_arguments;
+  rb_raise(args->exception_class, "%s", args->exception_message);
+}
+
+void grab_gvl_and_raise(VALUE exception_class, const char *format_string, ...) {
+  struct raise_arguments args;
+
+  args.exception_class = exception_class;
+
+  va_list format_string_arguments;
+  va_start(format_string_arguments, format_string);
+  vsnprintf(args.exception_message, MAX_RAISE_MESSAGE_SIZE, format_string, format_string_arguments);
+
+  if (is_current_thread_holding_the_gvl()) {
+    rb_raise(
+      rb_eRuntimeError,
+      "grab_gvl_and_raise called by thread holding the global VM lock. exception_message: '%s'",
+      args.exception_message
+    );
+  }
+
+  rb_thread_call_with_gvl(trigger_raise, &args);
+
+  rb_bug("[DDTRACE] Unexpected: Reached the end of grab_gvl_and_raise while raising '%s'\n", args.exception_message);
+}
+
+struct syserr_raise_arguments {
+  int syserr_errno;
+  char exception_message[MAX_RAISE_MESSAGE_SIZE];
+};
+
+static void *trigger_syserr_raise(void *syserr_raise_arguments) {
+  struct syserr_raise_arguments *args = (struct syserr_raise_arguments *) syserr_raise_arguments;
+  rb_syserr_fail(args->syserr_errno, args->exception_message);
+}
+
+void grab_gvl_and_raise_syserr(int syserr_errno, const char *format_string, ...) {
+  struct syserr_raise_arguments args;
+
+  args.syserr_errno = syserr_errno;
+
+  va_list format_string_arguments;
+  va_start(format_string_arguments, format_string);
+  vsnprintf(args.exception_message, MAX_RAISE_MESSAGE_SIZE, format_string, format_string_arguments);
+
+  if (is_current_thread_holding_the_gvl()) {
+    rb_raise(
+      rb_eRuntimeError,
+      "grab_gvl_and_raise_syserr called by thread holding the global VM lock. syserr_errno: %d, exception_message: '%s'",
+      syserr_errno,
+      args.exception_message
+    );
+  }
+
+  rb_thread_call_with_gvl(trigger_syserr_raise, &args);
+
+  rb_bug("[DDTRACE] Unexpected: Reached the end of grab_gvl_and_raise_syserr while raising '%s'\n", args.exception_message);
+}
+
+void raise_syserr(
+  int syserr_errno,
+  bool have_gvl,
+  const char *expression,
+  const char *file,
+  int line,
+  const char *function_name
+) {
+  if (have_gvl) {
+    rb_exc_raise(rb_syserr_new_str(syserr_errno, rb_sprintf("Failure returned by '%s' at %s:%d:in `%s'", expression, file, line, function_name)));
+  } else {
+    grab_gvl_and_raise_syserr(syserr_errno, "Failure returned by '%s' at %s:%d:in `%s'", expression, file, line, function_name);
+  }
+}

data/ext/ddtrace_profiling_native_extension/ruby_helpers.h

@@ -0,0 +1,89 @@
+#pragma once
+
+#include <ruby.h>
+#include <stdbool.h>
+
+#include "helpers.h"
+
+// Processes any pending interruptions, including exceptions to be raised.
+// If there's an exception to be raised, it raises it. In that case, this function does not return.
+static inline VALUE process_pending_interruptions(DDTRACE_UNUSED VALUE _) {
+  rb_thread_check_ints();
+  return Qnil;
+}
+
+// RB_UNLIKELY is not supported on Ruby 2.2 and 2.3
+#ifndef RB_UNLIKELY
+  #define RB_UNLIKELY(x) x
+#endif
+
+// Calls process_pending_interruptions BUT "rescues" any exceptions to be raised, returning them instead as
+// a non-zero `pending_exception`.
+//
+// Thus, if there's a non-zero `pending_exception`, the caller MUST call `rb_jump_tag(pending_exception)` after any
+// needed clean-ups.
+//
+// Usage example:
+//
+// ```c
+// foo = ruby_xcalloc(...);
+// pending_exception = check_if_pending_exception();
+// if (pending_exception) {
+//   ruby_xfree(foo);
+//   rb_jump_tag(pending_exception); // Re-raises exception
+// }
+// ```
+__attribute__((warn_unused_result))
+static inline int check_if_pending_exception(void) {
+  int pending_exception;
+  rb_protect(process_pending_interruptions, Qnil, &pending_exception);
+  return pending_exception;
+}
+
+#define ADD_QUOTES_HELPER(x) #x
+#define ADD_QUOTES(x) ADD_QUOTES_HELPER(x)
+
+// Ruby has a Check_Type(value, type) that is roughly equivalent to this BUT Ruby's version is rather cryptic when it fails
+// e.g. "wrong argument type nil (expected String)". This is a replacement that prints more information to help debugging.
+#define ENFORCE_TYPE(value, type) \
+  { if (RB_UNLIKELY(!RB_TYPE_P(value, type))) raise_unexpected_type(value, ADD_QUOTES(value), ADD_QUOTES(type), __FILE__, __LINE__, __func__); }
+
+#define ENFORCE_BOOLEAN(value) \
+  { if (RB_UNLIKELY(value != Qtrue && value != Qfalse)) raise_unexpected_type(value, ADD_QUOTES(value), "true or false", __FILE__, __LINE__, __func__); }
+
+// Called by ENFORCE_TYPE; should not be used directly
+NORETURN(void raise_unexpected_type(
+  VALUE value,
+  const char *value_name,
+  const char *type_name,
+  const char *file,
+  int line,
+  const char *function_name
+));
+
+#define VALUE_COUNT(array) (sizeof(array) / sizeof(VALUE))
+
+NORETURN(
+  void grab_gvl_and_raise(VALUE exception_class, const char *format_string, ...)
+  __attribute__ ((format (printf, 2, 3)));
+);
+NORETURN(
+  void grab_gvl_and_raise_syserr(int syserr_errno, const char *format_string, ...)
+  __attribute__ ((format (printf, 2, 3)));
+);
+
+#define ENFORCE_SUCCESS_GVL(expression) ENFORCE_SUCCESS_HELPER(expression, true)
+#define ENFORCE_SUCCESS_NO_GVL(expression) ENFORCE_SUCCESS_HELPER(expression, false)
+
+#define ENFORCE_SUCCESS_HELPER(expression, have_gvl) \
+  { int result_syserr_errno = expression; if (RB_UNLIKELY(result_syserr_errno)) raise_syserr(result_syserr_errno, have_gvl, ADD_QUOTES(expression), __FILE__, __LINE__, __func__); }
+
+// Called by ENFORCE_SUCCESS_HELPER; should not be used directly
+NORETURN(void raise_syserr(
+  int syserr_errno,
+  bool have_gvl,
+  const char *expression,
+  const char *file,
+  int line,
+  const char *function_name
+));

data/ext/ddtrace_profiling_native_extension/setup_signal_handler.c

@@ -0,0 +1,115 @@
+#include <ruby.h>
+#include <signal.h>
+#include <errno.h>
+#include <stdbool.h>
+
+#include "helpers.h"
+#include "setup_signal_handler.h"
+#include "ruby_helpers.h"
+
+// Used by Collectors::CpuAndWallTimeWorker to setup SIGPROF signal handlers used for cpu/wall-time profiling.
+
+static void install_sigprof_signal_handler_internal(
+  void (*signal_handler_function)(int, siginfo_t *, void *),
+  const char *handler_pretty_name,
+  void (*signal_handler_to_replace)(int, siginfo_t *, void *)
+);
+
+void empty_signal_handler(DDTRACE_UNUSED int _signal, DDTRACE_UNUSED siginfo_t *_info, DDTRACE_UNUSED void *_ucontext) { }
+
+void install_sigprof_signal_handler(void (*signal_handler_function)(int, siginfo_t *, void *), const char *handler_pretty_name) {
+  install_sigprof_signal_handler_internal(signal_handler_function, handler_pretty_name, NULL);
+}
+
+void replace_sigprof_signal_handler_with_empty_handler(void (*expected_existing_handler)(int, siginfo_t *, void *)) {
+  install_sigprof_signal_handler_internal(empty_signal_handler, "empty_signal_handler", expected_existing_handler);
+}
+
+static void install_sigprof_signal_handler_internal(
+  void (*signal_handler_function)(int, siginfo_t *, void *),
+  const char *handler_pretty_name,
+  void (*signal_handler_to_replace)(int, siginfo_t *, void *)
+) {
+  struct sigaction existing_signal_handler_config = {.sa_sigaction = NULL};
+  struct sigaction signal_handler_config = {
+    .sa_flags = SA_RESTART | SA_SIGINFO,
+    .sa_sigaction = signal_handler_function
+  };
+  sigemptyset(&signal_handler_config.sa_mask);
+
+  if (sigaction(SIGPROF, &signal_handler_config, &existing_signal_handler_config) != 0) {
+    rb_exc_raise(rb_syserr_new_str(errno, rb_sprintf("Could not install profiling signal handler (%s)", handler_pretty_name)));
+  }
+
+  // Because signal handler functions are global, let's check if we're not stepping on someone else's toes.
+
+  // If the existing signal handler was our empty one, that's ok as well
+  if (existing_signal_handler_config.sa_sigaction == empty_signal_handler ||
+  // In some corner cases (e.g. after a fork), our signal handler may still be around, and that's ok
+    existing_signal_handler_config.sa_sigaction == signal_handler_function ||
+  // Are we replacing a known handler with another one?
+    (signal_handler_to_replace != NULL && existing_signal_handler_config.sa_sigaction == signal_handler_to_replace)
+  ) { return; }
+
+  if (existing_signal_handler_config.sa_handler != NULL || existing_signal_handler_config.sa_sigaction != NULL) {
+    // An unexpected/unknown signal handler already existed. Currently we don't support this situation, so let's just back out
+    // of the installation.
+
+    if (sigaction(SIGPROF, &existing_signal_handler_config, NULL) != 0) {
+      rb_exc_raise(
+        rb_syserr_new_str(
+          errno,
+          rb_sprintf(
+            "Failed to install profiling signal handler (%s): " \
+            "While installing a SIGPROF signal handler, the profiler detected that another software/library/gem had " \
+            "previously installed a different SIGPROF signal handler. " \
+            "The profiler tried to restore the previous SIGPROF signal handler, but this failed. " \
+            "The other software/library/gem may have been left in a broken state. ",
+            handler_pretty_name
+          )
+        )
+      );
+    }
+
+    rb_raise(
+      rb_eRuntimeError,
+      "Could not install profiling signal handler (%s): There's a pre-existing SIGPROF signal handler",
+      handler_pretty_name
+    );
+  }
+}
+
+// Note: Be careful when using this; you probably want to use `replace_sigprof_signal_handler_with_empty_handler` instead.
+// (See comments on `collectors_cpu_and_wall_time_worker.c` for details)
+void remove_sigprof_signal_handler(void) {
+  struct sigaction signal_handler_config = {
+    .sa_handler = SIG_DFL, // Reset back to default
+    .sa_flags = SA_RESTART // TODO: Unclear if this is actually needed/does anything at all
+  };
+  sigemptyset(&signal_handler_config.sa_mask);
+
+  if (sigaction(SIGPROF, &signal_handler_config, NULL) != 0) rb_sys_fail("Failure while removing the signal handler");
+}
+
+static void toggle_sigprof_signal_handler_for_current_thread(int action) {
+  sigset_t signals_to_toggle;
+  sigemptyset(&signals_to_toggle);
+  sigaddset(&signals_to_toggle, SIGPROF);
+  int error = pthread_sigmask(action, &signals_to_toggle, NULL);
+  if (error) rb_exc_raise(rb_syserr_new_str(error, rb_sprintf("Unexpected failure in pthread_sigmask, action=%d", action)));
+}
+
+void block_sigprof_signal_handler_from_running_in_current_thread(void) {
+  toggle_sigprof_signal_handler_for_current_thread(SIG_BLOCK);
+}
+
+void unblock_sigprof_signal_handler_from_running_in_current_thread(void) {
+  toggle_sigprof_signal_handler_for_current_thread(SIG_UNBLOCK);
+}
+
+VALUE is_sigprof_blocked_in_current_thread(void) {
+  sigset_t current_signals;
+  sigemptyset(&current_signals);
+  ENFORCE_SUCCESS_GVL(pthread_sigmask(0, NULL, &current_signals));
+  return sigismember(&current_signals, SIGPROF) ? Qtrue : Qfalse;
+}

data/ext/ddtrace_profiling_native_extension/setup_signal_handler.h

@@ -0,0 +1,11 @@
+#pragma once
+
+#include <signal.h>
+
+void empty_signal_handler(DDTRACE_UNUSED int _signal, DDTRACE_UNUSED siginfo_t *_info, DDTRACE_UNUSED void *_ucontext);
+void install_sigprof_signal_handler(void (*signal_handler_function)(int, siginfo_t *, void *), const char *handler_pretty_name);
+void replace_sigprof_signal_handler_with_empty_handler(void (*expected_existing_handler)(int, siginfo_t *, void *));
+void remove_sigprof_signal_handler(void);
+void block_sigprof_signal_handler_from_running_in_current_thread(void);
+void unblock_sigprof_signal_handler_from_running_in_current_thread(void);
+VALUE is_sigprof_blocked_in_current_thread(void);