datadog 2.17.0 → 2.37.0

data/ext/datadog_profiling_native_extension/collectors_thread_context.h

@@ -2,29 +2,42 @@
 
 #include <ruby.h>
 #include <stdbool.h>
+#include <stdint.h>
 
 #include "gvl_profiling_helper.h"
 
 void thread_context_collector_sample(
   VALUE self_instance,
-  long current_monotonic_wall_time_ns,
-  VALUE profiler_overhead_stack_thread
+  long current_monotonic_wall_time_ns
 );
-void thread_context_collector_sample_allocation(VALUE self_instance, unsigned int sample_weight, VALUE new_object);
+__attribute__((warn_unused_result)) bool thread_context_collector_prepare_sample_inside_signal_handler(void);
+__attribute__((warn_unused_result)) bool thread_context_collector_sample_allocation(VALUE self_instance, per_thread_context *thread_context, unsigned int sample_weight, VALUE new_object);
+void thread_context_collector_after_allocation(VALUE self_instance);
 void thread_context_collector_sample_skipped_allocation_samples(VALUE self_instance, unsigned int skipped_samples);
 VALUE thread_context_collector_sample_after_gc(VALUE self_instance);
 void thread_context_collector_on_gc_start(VALUE self_instance);
 __attribute__((warn_unused_result)) bool thread_context_collector_on_gc_finish(VALUE self_instance);
 VALUE enforce_thread_context_collector_instance(VALUE object);
+void thread_context_collector_stats(VALUE self_instance, VALUE stats_hash);
+void thread_context_collector_stats_reset_not_thread_safe(VALUE self_instance);
+void thread_context_collector_on_serialize(VALUE self_instance);
+void thread_context_collector_profiler_internal_thread_started(void);
+void thread_context_collector_profiler_internal_thread_done(VALUE self_instance);
 
 #ifndef NO_GVL_INSTRUMENTATION
   typedef enum {
     ON_GVL_RUNNING_UNKNOWN, // Thread is not known, it may not even be from the current Ractor
     ON_GVL_RUNNING_DONT_SAMPLE, // Thread is known, but "Waiting for GVL" period was too small to be sampled
     ON_GVL_RUNNING_SAMPLE, // Thread is known, and "Waiting for GVL" period should be sampled
+  } on_gvl_running_enum;
+
+  typedef struct {
+    on_gvl_running_enum action;
+    long waiting_for_gvl_duration_ns;
   } on_gvl_running_result;
 
-  void thread_context_collector_on_gvl_waiting(gvl_profiling_thread thread);
-  __attribute__((warn_unused_result)) on_gvl_running_result thread_context_collector_on_gvl_running(gvl_profiling_thread thread);
+  void thread_context_collector_on_gvl_waiting(per_thread_context *thread_context);
+  __attribute__((warn_unused_result)) on_gvl_running_result thread_context_collector_on_gvl_running(VALUE self_instance, VALUE thread, per_thread_context *thread_context);
   VALUE thread_context_collector_sample_after_gvl_running(VALUE self_instance, VALUE current_thread, long current_monotonic_wall_time_ns);
+  void thread_context_collector_on_gvl_released(per_thread_context *thread_context);
 #endif

data/ext/datadog_profiling_native_extension/crashtracking_runtime_stacks.c

@@ -0,0 +1,239 @@
+// NOTE: This file is a part of the profiling native extension even though the
+// runtime stacks feature is consumed by the crashtracker. The profiling
+// extension already carries all the Ruby VM private header access and build
+// plumbing required to safely poke at internal structures. Sharing that setup
+// avoids duplicating another native extension with the same (fragile) access
+// patterns, and keeps the overall install/build surface area smaller.
+//
+// This also means that this functionality is depend on the profiling native extension
+// being available and built
+#include "extconf.h"
+
+#if defined(__linux__)
+
+#include <datadog/crashtracker.h>
+#include "datadog_ruby_common.h"
+#include "private_vm_api_access.h"
+#include <sys/mman.h>
+#include <unistd.h>
+#include <errno.h>
+#include <string.h>
+
+static const rb_data_type_t *crashtracker_thread_data_type = NULL;
+
+static void ruby_runtime_stack_callback(
+  void (*emit_frame)(const ddog_crasht_RuntimeStackFrame*)
+);
+
+// Use a fixed, preallocated buffer for crash-time runtime stacks to avoid
+// heap allocation in the signal/crash path.
+#define RUNTIME_STACK_MAX_FRAMES 512
+static frame_info runtime_stack_buffer[RUNTIME_STACK_MAX_FRAMES];
+
+#if defined(__x86_64__)
+#  define SYS_MINCORE 0x1B
+#elif defined(__aarch64__)
+#  define SYS_MINCORE 0xE8
+#endif
+
+long syscall(long number, ...);
+
+// align down to power of two
+static inline uintptr_t align_down(uintptr_t x, uintptr_t align) {
+  return x & ~(align - 1u);
+}
+
+static uintptr_t cached_page_size = 0;
+
+// TODO: This function is not necessarily Ruby specific. This will be moved to
+// `libdatadog` in the future as a shared utility function.
+static inline bool is_pointer_readable(const void *ptr, size_t size) {
+  if (!ptr || size == 0) return false;
+
+  uintptr_t page_size = cached_page_size;
+
+  const uintptr_t start = align_down((uintptr_t)ptr, page_size);
+  const uintptr_t end   = ((uintptr_t)ptr + size - 1u);
+  const uintptr_t last  = align_down(end, page_size);
+
+  // Number of pages spanned
+  size_t pages = 1u + (last != start);
+  if (pages > 2u) pages = 2u;
+
+  unsigned char vec[2];
+
+  int retries = 5;
+  for (;;) {
+      size_t len = pages * (size_t)page_size;
+      long rc = syscall(SYS_MINCORE, (void*)start, len, vec);
+
+      if (rc == 0) {
+          return true;
+      }
+
+      int e = errno;
+      if (e == ENOMEM || e == EFAULT) {
+          return false;
+      }
+
+      if (e == EAGAIN && retries-- > 0) {
+          continue;
+      }
+
+      // Unknown errno, we assume mapped to avoid cascading faults in crash path
+      return true;
+  }
+}
+
+static inline ddog_CharSlice char_slice_from_cstr(const char *cstr) {
+  if (cstr == NULL) {
+    return (ddog_CharSlice){.ptr = NULL, .len = 0};
+  }
+  return (ddog_CharSlice){.ptr = cstr, .len = strlen(cstr)};
+}
+
+static ddog_CharSlice safe_string_value(VALUE str) {
+  if (str == Qnil) return DDOG_CHARSLICE_C("<nil>");
+
+  // Validate object and payload readability in one check
+  if (!is_pointer_readable((const void *)str, sizeof(struct RString))) return DDOG_CHARSLICE_C("<corrupted>");
+  if (!RB_TYPE_P(str, T_STRING)) return DDOG_CHARSLICE_C("<not_string>");
+
+  long len = RSTRING_LEN(str);
+  if (len < 0 || len > 1024) return DDOG_CHARSLICE_C("<length_over_limit>");
+
+  const char *ptr = RSTRING_PTR(str);
+  if (!ptr) return DDOG_CHARSLICE_C("<null>");
+
+  if (!is_pointer_readable(ptr, len > 0 ? len : 1)) return DDOG_CHARSLICE_C("<unreadable>");
+
+  return (ddog_CharSlice){.ptr = ptr, .len = (size_t)len};
+}
+
+static void emit_placeholder_frame(
+  void (*emit_frame)(const ddog_crasht_RuntimeStackFrame*),
+  const char *label
+) {
+  ddog_crasht_RuntimeStackFrame frame = {
+    .type_name = DDOG_CHARSLICE_C(""),
+    .function = char_slice_from_cstr(label),
+    .file = DDOG_CHARSLICE_C("<unknown>"),
+    .line = 0,
+    .column = 0
+  };
+  emit_frame(&frame);
+}
+
+// Collect the crashing thread's frames via ddtrace_rb_profile_frames into a static buffer, then emit
+// them newest-first. If corruption is detected, emit placeholder frames so the crash report still
+// completes. We lean on the Ruby VM helpers we already use for profiling and rely on crashtracker's
+// safety nets so a failure here should not impact customers.
+static void ruby_runtime_stack_callback(
+  void (*emit_frame)(const ddog_crasht_RuntimeStackFrame*)
+) {
+  // Grab the Ruby thread we crashed on; crashtracker only runs once.
+  VALUE current_thread = rb_thread_current();
+  if (current_thread == Qnil) return;
+
+  if (crashtracker_thread_data_type == NULL) return;
+
+  if (!rb_typeddata_is_kind_of(current_thread, crashtracker_thread_data_type)) {
+    emit_placeholder_frame(emit_frame, "<runtime stack not found>");
+    return;
+  }
+
+  // Use the profiling helper to gather frames into our static buffer.
+  int frame_count = ddtrace_rb_profile_frames(
+    current_thread,
+    0,
+    RUNTIME_STACK_MAX_FRAMES,
+    runtime_stack_buffer
+  );
+
+  if (frame_count <= 0) {
+    emit_placeholder_frame(emit_frame, "<runtime stack not found>");
+    return;
+  }
+
+  for (int i = frame_count - 1; i >= 0; i--) {
+    frame_info *info = &runtime_stack_buffer[i];
+
+    if (info->is_ruby_frame) {
+      const void *iseq = (const void *)info->as.ruby_frame.iseq;
+      ddog_CharSlice function_slice = DDOG_CHARSLICE_C("<unknown>");
+      ddog_CharSlice file_slice = DDOG_CHARSLICE_C("<unknown>");
+
+      if (iseq && is_pointer_readable(iseq, sizeof_rb_iseq_t())) {
+        function_slice = safe_string_value(ddtrace_iseq_base_label(iseq));
+        file_slice = safe_string_value(ddtrace_iseq_path(iseq));
+      }
+
+      ddog_crasht_RuntimeStackFrame frame = {
+        .type_name = DDOG_CHARSLICE_C(""),
+        .function = function_slice,
+        .file = file_slice,
+        .line = info->as.ruby_frame.line,
+        .column = 0
+      };
+
+      emit_frame(&frame);
+    } else {
+      ddog_CharSlice function_slice = DDOG_CHARSLICE_C("<C method>");
+      ddog_CharSlice file_slice = DDOG_CHARSLICE_C("<C extension>");
+
+      if (info->as.native_frame.method_id) {
+        const char *method_name = rb_id2name(info->as.native_frame.method_id);
+        if (is_pointer_readable(method_name, 256)) {
+          size_t method_name_len = strnlen(method_name, 256);
+          if (method_name_len > 0 && method_name_len < 256) {
+            function_slice = char_slice_from_cstr(method_name);
+          }
+        }
+      }
+
+      ddog_crasht_RuntimeStackFrame frame = {
+        .type_name = DDOG_CHARSLICE_C(""),
+        .function = function_slice,
+        .file = file_slice,
+        .line = 0,
+        .column = 0
+      };
+
+      emit_frame(&frame);
+    }
+  }
+
+  if (frame_count == RUNTIME_STACK_MAX_FRAMES) {
+    emit_placeholder_frame(emit_frame, "<truncated frames>");
+  }
+}
+
+void crashtracking_runtime_stacks_init(void) {
+  cached_page_size = (uintptr_t)sysconf(_SC_PAGESIZE);
+  if (cached_page_size == 0 || (cached_page_size & (cached_page_size - 1u))) {
+    cached_page_size = 4096;
+  }
+
+  if (crashtracker_thread_data_type == NULL) {
+    VALUE current_thread = rb_thread_current();
+    if (current_thread == Qnil) {
+      raise_error(rb_eRuntimeError, "crashtracking_runtime_stacks_init: rb_thread_current returned Qnil");
+    }
+
+    const rb_data_type_t *thread_data_type = RTYPEDDATA_TYPE(current_thread);
+    if (!thread_data_type) {
+      raise_error(rb_eRuntimeError, "crashtracking_runtime_stacks_init: RTYPEDDATA_TYPE returned NULL");
+    }
+
+    crashtracker_thread_data_type = thread_data_type;
+  }
+
+  // Register immediately so Ruby doesn't need to manage this explicitly.
+  ddog_crasht_register_runtime_frame_callback(ruby_runtime_stack_callback);
+}
+
+#else
+// Keep init symbol to satisfy linkage on non linux platforms, but do nothing
+void crashtracking_runtime_stacks_init(void) {}
+#endif
+

data/ext/datadog_profiling_native_extension/datadog_ruby_common.c

@@ -1,8 +1,11 @@
 #include "datadog_ruby_common.h"
+#include <stdarg.h>
 
 // IMPORTANT: Currently this file is copy-pasted between extensions. Make sure to update all versions when doing any change!
 
-void raise_unexpected_type(VALUE value, const char *value_name, const char *type_name, const char *file, int line, const char* function_name) {
+static ID telemetry_message_id;
+
+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,
@@ -18,6 +21,25 @@ void raise_unexpected_type(VALUE value, const char *value_name, const char *type
   );
 }
 
+// Raises an exception with separate telemetry-safe and detailed messages.
+// NOTE: Raising an exception always invokes Ruby code so it requires the GVL and is not compatible with "debug_enter_unsafe_context".
+// @see debug_enter_unsafe_context
+void private_raise_exception(VALUE exception, const char *static_message) {
+  rb_ivar_set(exception, telemetry_message_id, rb_str_new_cstr(static_message));
+  rb_exc_raise(exception);
+}
+
+// Use `raise_error` the macro instead, as it provides additional argument checks.
+void private_raise_error(VALUE exception_class, const char *fmt, ...) {
+  va_list args;
+  va_start(args, fmt);
+  VALUE detailed_message = rb_vsprintf(fmt, args);
+  va_end(args);
+
+  VALUE exception = rb_exc_new_str(exception_class, detailed_message);
+  private_raise_exception(exception, fmt);
+}
+
 VALUE datadog_gem_version(void) {
   VALUE ddtrace_module = rb_const_get(rb_cObject, rb_intern("Datadog"));
   ENFORCE_TYPE(ddtrace_module, T_MODULE);
@@ -78,3 +100,45 @@ ddog_Vec_Tag convert_tags(VALUE tags_as_array) {
 
   return tags;
 }
+
+size_t read_ddogerr_string_and_drop(ddog_Error *error, char *string, size_t capacity) {
+  if (capacity == 0 || string == NULL) {
+    // short-circuit, we can't write anything
+    ddog_Error_drop(error);
+    return 0;
+  }
+
+  ddog_CharSlice error_msg_slice = ddog_Error_message(error);
+  size_t error_msg_size = error_msg_slice.len;
+  // Account for extra null char for proper cstring
+  if (error_msg_size >= capacity) {
+    // Error message too big, lets truncate it to capacity - 1 to allow for extra null at end
+    error_msg_size = capacity - 1;
+  }
+  strncpy(string, error_msg_slice.ptr, error_msg_size);
+  string[error_msg_size] = '\0';
+  ddog_Error_drop(error);
+  return error_msg_size;
+}
+
+static void verify_libdatadog_version(void) {
+  rb_eval_string(
+    "require 'libdatadog';"
+    "expected = '" EXPECTED_LIBDATADOG_VERSION "';"
+    "if expected != Libdatadog::VERSION;"
+      "raise(LoadError, <<MSG\n"
+        "The `datadog` gem needs to be reinstalled whenever the `libdatadog` gem version is changed. "
+        "The currently-installed version of `datadog` was built to work with `libdatadog` gem version #{expected} "
+        "but the currently-loaded version of `libdatadog` is #{Libdatadog::VERSION}. "
+        "To fix this, reinstall the `datadog` gem (e.g. `bundle exec gem pristine datadog`) "
+        "or contact Datadog support for help at <https://docs.datadoghq.com/help/>.\n"
+        "MSG\n"
+      ");"
+    "end"
+  );
+}
+
+void datadog_ruby_common_init(void) {
+  telemetry_message_id = rb_intern("@telemetry_message");
+  verify_libdatadog_version();
+}

data/ext/datadog_profiling_native_extension/datadog_ruby_common.h

@@ -3,7 +3,10 @@
 // IMPORTANT: Currently this file is copy-pasted between extensions. Make sure to update all versions when doing any change!
 
 #include <ruby.h>
-#include <datadog/profiling.h>
+#include <datadog/common.h>
+
+// Must be called once during initialization
+void datadog_ruby_common_init(void);
 
 // Used to mark symbols to be exported to the outside of the extension.
 // Consider very carefully before tagging a function with this.
@@ -32,6 +35,27 @@
 
 NORETURN(void raise_unexpected_type(VALUE value, const char *value_name, const char *type_name, const char *file, int line, const char* function_name));
 
+// Raises an exception of the specified class with the formatted string as its message.
+// This macro ensures that the literal string is sent for telemetry, while the formatted
+// message is the default `Exception#message`.
+// *Ruby exceptions not raised through this function will not be reported via telemetry.*
+#define raise_error(exception_class, fmt, ...) \
+  private_raise_error(exception_class, "" fmt, ##__VA_ARGS__)
+
+NORETURN(
+  void private_raise_error(VALUE exception_class, const char *fmt, ...)
+  __attribute__ ((format (printf, 2, 3)));
+);
+
+// Raises an exception with separate telemetry-safe and detailed messages.
+// NOTE: Raising an exception always invokes Ruby code so it requires the GVL and is not compatible with "debug_enter_unsafe_context".
+// @see debug_enter_unsafe_context
+NORETURN(
+  void private_raise_exception(VALUE exception, const char *static_message)
+);
+
+#define MAX_RAISE_MESSAGE_SIZE 256
+
 // Helper to retrieve Datadog::VERSION::STRING
 VALUE datadog_gem_version(void);
 
@@ -61,3 +85,18 @@ static inline VALUE get_error_details_and_drop(ddog_Error *error) {
   ddog_Error_drop(error);
   return result;
 }
+
+// Utility function to be able to extract an error cstring from a ddog_Error.
+// Returns the amount of characters written to string (which are necessarily
+// bounded by capacity - 1 since the string will be null-terminated).
+size_t read_ddogerr_string_and_drop(ddog_Error *error, char *string, size_t capacity);
+
+#define IMEMO_MASK 0x0f
+
+// Returns the imemo type of an imemo object.
+// This mask is the same between Ruby 2.5 and 3.3-preview3. Furthermore, the intention of this method is to be used
+// to call `rb_imemo_name` which correctly handles invalid numbers so even if the mask changes in the future, at most
+// we'll get incorrect results (and never a VM crash)
+static inline int ddtrace_imemo_type(VALUE imemo) {
+  return (RBASIC(imemo)->flags >> FL_USHIFT) & IMEMO_MASK;
+}

data/ext/datadog_profiling_native_extension/encoded_profile.c

@@ -1,6 +1,7 @@
 #include "encoded_profile.h"
 #include "datadog_ruby_common.h"
 #include "libdatadog_helpers.h"
+#include "ruby_helpers.h"
 
 // This class exists to wrap a ddog_prof_EncodedProfile into a Ruby object
 // This file implements the native bits of the Datadog::Profiling::EncodedProfile class
@@ -41,7 +42,7 @@ VALUE from_ddog_prof_EncodedProfile(ddog_prof_EncodedProfile profile) {
 static ddog_ByteSlice get_bytes(ddog_prof_EncodedProfile *state) {
   ddog_prof_Result_ByteSlice raw_bytes = ddog_prof_EncodedProfile_bytes(state);
   if (raw_bytes.tag == DDOG_PROF_RESULT_BYTE_SLICE_ERR_BYTE_SLICE) {
-    rb_raise(rb_eRuntimeError, "Failed to get bytes from profile: %"PRIsVALUE, get_error_details_and_drop(&raw_bytes.err));
+    raise_error(rb_eRuntimeError, "Failed to get bytes from profile: %"PRIsVALUE, get_error_details_and_drop(&raw_bytes.err));
   }
   return raw_bytes.ok;
 }

data/ext/datadog_profiling_native_extension/extconf.rb

@@ -69,20 +69,32 @@ $stderr.puts(
 # NOTE: we MUST NOT require 'mkmf' before we check the #skip_building_extension? because the require triggers checks
 # that may fail on an environment not properly setup for building Ruby extensions.
 require "mkmf"
+Datadog::LibdatadogExtconfHelpers.dump_mkmf_log_on_failure!
 
 Logging.message("[datadog] Using compiler:\n")
 xsystem("#{CONFIG["CC"]} -v")
 Logging.message("[datadog] End of compiler information\n")
 
+# We must *never* `append_cflags "-Wall"` or `append_cflags "-Wextra"`, because those add the flag at the end,
+# which then overrides earlier -Wno-* flags from RbConfig::CONFIG["warnflags"], and causes errors if -Werror is added.
+# There is no need to add them anyway, Ruby since 1.9 has RbConfig::CONFIG["warnflags"] starting with "-Wall -Wextra ".
+
 # Because we can't control what compiler versions our customers use, shipping with -Werror by default is a no-go.
 # But we can enable it in CI, so that we quickly spot any new warnings that just got introduced.
 append_cflags "-Werror" if ENV["DATADOG_GEM_CI"] == "true"
 
+# TEMPORARY STOPGAP: libdatadog v36's vendored `common.h` ships duplicate typedefs
+# (a header-dedup regression), which `-Werror` turns into fatal
+# `-Wtypedef-redefinition` errors under C11/gnu11. Keep it a warning until a fixed
+# header lands upstream in libdatadog/libdatadog-rb. Remove this once that ships.
+append_cflags "-Wno-error=typedef-redefinition" if ENV["DATADOG_GEM_CI"] == "true"
+
 # Older gcc releases may not default to C99 and we need to ask for this. This is also used:
 # * by upstream Ruby -- search for gnu99 in the codebase
 # * by msgpack, another datadog gem dependency
 #   (https://github.com/msgpack/msgpack-ruby/blob/18ce08f6d612fe973843c366ac9a0b74c4e50599/ext/msgpack/extconf.rb#L8)
-append_cflags "-std=gnu99"
+# @ivoanjo: We could probably start using C11/gnu11 for non macOS-too but it's somewhat hard to validate so I chickened out for now
+append_cflags RUBY_PLATFORM.include?('darwin') ? '-std=gnu11' : '-std=gnu99'
 
 # Gets really noisy when we include the MJIT header, let's omit it (TODO: Use #pragma GCC diagnostic instead?)
 append_cflags "-Wno-unused-function"
@@ -105,10 +117,6 @@ append_cflags "-fvisibility=hidden"
 # Avoid legacy C definitions
 append_cflags "-Wold-style-definition"
 
-# Enable all other compiler warnings
-append_cflags "-Wall"
-append_cflags "-Wextra"
-
 if ENV["DDTRACE_DEBUG"] == "true"
   $defs << "-DDD_DEBUG"
   CONFIG["optflags"] = "-O0"
@@ -127,15 +135,40 @@ if RUBY_PLATFORM.include?("linux")
 
   # Not available on macOS
   $defs << "-DHAVE_CLOCK_MONOTONIC_COARSE"
+elsif RUBY_PLATFORM.include?("darwin")
+  # On macOS, we use Mach thread APIs to get per-thread CPU time
+  $defs << "-DHAVE_MACH_THREAD_INFO"
 end
 
 have_func "malloc_stats"
 
-# On Ruby 3.5, we can't ask the object_id from IMEMOs (https://github.com/ruby/ruby/pull/13347)
-$defs << "-DNO_IMEMO_OBJECT_ID" unless RUBY_VERSION < "3.5"
+# Used to get native filenames (dladdr1 is preferred, so we only check for the other if not available)
+# Note it's possible none are available
+if have_header("dlfcn.h")
+  (have_struct_member("struct link_map", "l_name", "link.h") && have_func("dladdr1")) ||
+    have_func("dladdr")
+end
+
+# On older Rubies, there was no primitive mutex and condition variable implemented in `thread_sync.rb` (internal)
+$defs << "-DNO_PRIMITIVE_MUTEX_AND_CONDITION_VARIABLE" if RUBY_VERSION < "4"
+
+# On Ruby 4, we can't ask the object_id from IMEMOs (https://github.com/ruby/ruby/pull/13347)
+$defs << "-DNO_IMEMO_OBJECT_ID" unless RUBY_VERSION < "4"
 
-# On Ruby 2.5 and 3.3, this symbol was not visible. It is on 2.6 to 3.2, as well as 3.4+
-$defs << "-DNO_RB_OBJ_INFO" if RUBY_VERSION.start_with?("2.5", "3.3")
+# On Ruby 4, we need to defer calling rb_obj_id during heap allocation recording
+# because it's not safe to mutate objects during the newobj tracepoint
+# (see https://bugs.ruby-lang.org/issues/21710)
+$defs << "-DUSE_DEFERRED_HEAP_ALLOCATION_RECORDING" unless RUBY_VERSION < "4"
+
+# On Ruby 4.0, we've seen crashes when computing the memsize of a class/module/iclass:
+# rb_obj_memsize_of walks the per-namespace class extensions (classext_memsize), which seem to sometimes be in an inconsistent state
+# (see https://github.com/DataDog/dd-trace-rb/issues/5936)
+$defs << "-DNO_SAFE_CLASS_MEMSIZE" unless RUBY_VERSION < "4"
+
+# This symbol is exclusively visible on certain Ruby versions: 2.6 to 3.2, as well as 3.4 (but not 4.0+)
+# It's only used to get extra information about an object when a failure happens, so it's a "very nice to have" but not
+# actually required for correct behavior of the profiler.
+$defs << "-DNO_RB_OBJ_INFO" if RUBY_VERSION.start_with?("2.5", "3.3", "4.")
 
 # On older Rubies, rb_postponed_job_preregister/rb_postponed_job_trigger did not exist
 $defs << "-DNO_POSTPONED_TRIGGER" if RUBY_VERSION < "3.3"
@@ -152,8 +185,8 @@ $defs << "-DUSE_RACTOR_INTERNAL_APIS_DIRECTLY" if RUBY_VERSION < "3.3"
 # On older Rubies, there was no GVL instrumentation API and APIs created to support it
 $defs << "-DNO_GVL_INSTRUMENTATION" if RUBY_VERSION < "3.2"
 
-# Supporting GVL instrumentation on 3.2 needs some workarounds
-$defs << "-DUSE_GVL_PROFILING_3_2_WORKAROUNDS" if RUBY_VERSION.start_with?("3.2")
+# rb_internal_thread_specific_*()
+$defs << "-DHAVE_RUBY_THREAD_STORAGE_API" if RUBY_VERSION >= "3.3"
 
 # On older Rubies, there was no struct rb_native_thread. See private_vm_api_acccess.c for details.
 $defs << "-DNO_RB_NATIVE_THREAD" if RUBY_VERSION < "3.2"
@@ -193,37 +226,18 @@ $defs << "-DNO_GVL_OWNER" if RUBY_VERSION < "2.6"
 $defs << "-DNO_THREAD_INVOKE_ARG" if RUBY_VERSION < "2.6"
 
 # If we got here, libdatadog is available and loaded
-ENV["PKG_CONFIG_PATH"] = "#{ENV["PKG_CONFIG_PATH"]}:#{Libdatadog.pkgconfig_folder}"
-Logging.message("[datadog] PKG_CONFIG_PATH set to #{ENV["PKG_CONFIG_PATH"].inspect}\n")
 $stderr.puts("Using libdatadog #{Libdatadog::VERSION} from #{Libdatadog.pkgconfig_folder}")
 
-unless pkg_config("datadog_profiling_with_rpath")
-  Logging.message("[datadog] Ruby detected the pkg-config command is #{$PKGCONFIG.inspect}\n")
-
-  skip_building_extension!(
-    if Datadog::LibdatadogExtconfHelpers.pkg_config_missing?
-      Datadog::Profiling::NativeExtensionHelpers::Supported::PKG_CONFIG_IS_MISSING
-    else
-      # Less specific error message
-      Datadog::Profiling::NativeExtensionHelpers::Supported::FAILED_TO_CONFIGURE_LIBDATADOG
-    end
-  )
+unless Datadog::LibdatadogExtconfHelpers.configure_libdatadog(extconf_folder: __dir__)
+  skip_building_extension!(Datadog::Profiling::NativeExtensionHelpers::Supported::FAILED_TO_CONFIGURE_LIBDATADOG)
 end
 
+Datadog::LibdatadogExtconfHelpers.add_libdatadog_version_define
+
 unless have_type("atomic_int", ["stdatomic.h"])
   skip_building_extension!(Datadog::Profiling::NativeExtensionHelpers::Supported::COMPILER_ATOMIC_MISSING)
 end
 
-# See comments on the helper methods being used for why we need to additionally set this.
-# The extremely excessive escaping around ORIGIN below seems to be correct and was determined after a lot of
-# experimentation. We need to get these special characters across a lot of tools untouched...
-extra_relative_rpaths = [
-  Datadog::LibdatadogExtconfHelpers.libdatadog_folder_relative_to_native_lib_folder(current_folder: __dir__),
-  *Datadog::LibdatadogExtconfHelpers.libdatadog_folder_relative_to_ruby_extensions_folders,
-]
-extra_relative_rpaths.each { |folder| $LDFLAGS += " -Wl,-rpath,$$$\\\\{ORIGIN\\}/#{folder.to_str}" }
-Logging.message("[datadog] After pkg-config $LDFLAGS were set to: #{$LDFLAGS.inspect}\n")
-
 # Tag the native extension library with the Ruby version and Ruby platform.
 # This makes it easier for development (avoids "oops I forgot to rebuild when I switched my Ruby") and ensures that
 # the wrong library is never loaded.

data/ext/datadog_profiling_native_extension/gvl_profiling_helper.c

@@ -4,49 +4,10 @@
 #include "datadog_ruby_common.h"
 #include "gvl_profiling_helper.h"
 
-#if !defined(NO_GVL_INSTRUMENTATION) && !defined(USE_GVL_PROFILING_3_2_WORKAROUNDS) // Ruby 3.3+
-  rb_internal_thread_specific_key_t gvl_waiting_tls_key;
+#ifdef HAVE_RUBY_THREAD_STORAGE_API
+  rb_internal_thread_specific_key_t per_thread_context_key;
 
-  void gvl_profiling_init(void) {
-    gvl_waiting_tls_key = rb_internal_thread_specific_key_create();
-  }
-
-#endif
-
-#ifdef USE_GVL_PROFILING_3_2_WORKAROUNDS // Ruby 3.2
-  __thread gvl_profiling_thread gvl_waiting_tls;
-  static bool gvl_profiling_state_thread_tracking_workaround_installed = false;
-
-  static void on_thread_start(
-    DDTRACE_UNUSED rb_event_flag_t _unused1,
-    DDTRACE_UNUSED const rb_internal_thread_event_data_t *_unused2,
-    DDTRACE_UNUSED void *_unused3
-  ) {
-    gvl_waiting_tls = (gvl_profiling_thread) {.thread = NULL};
-  }
-
-  // Hack: We're using the gvl_waiting_tls native thread-local to store per-thread information. Unfortunately, Ruby puts a big hole
-  // in our plan because it reuses native threads -- specifically, in Ruby 3.2, native threads are still 1:1 to Ruby
-  // threads (M:N wasn't a thing yet) BUT once a Ruby thread dies, the VM will keep the native thread around for a
-  // bit, and if another Ruby thread starts right after, Ruby will reuse the native thread, rather than create a new one.
-  //
-  // This will mean that the new Ruby thread will still have the same native thread-local data that we set on the
-  // old thread. For the purposes of our tracking, where we're keeping a pointer to the current thread object in
-  // thread-local storage **this is disastrous** since it means we'll be pointing at the wrong thread (and its
-  // memory may have been freed or reused since!)
-  //
-  // To work around this issue, once GVL profiling is enabled, we install an event hook on thread start
-  // events that clears the thread-local data. This guarantees that there will be no stale data -- any existing
-  // data will be cleared at thread start.
-  //
-  // Note that once installed, this event hook becomes permanent -- stopping the profiler does not stop this event
-  // hook, unlike all others. This is because we can't afford to miss any thread start events while the
-  // profiler is stopped (e.g. during reconfiguration) as that would mean stale data once the profiler starts again.
-  void gvl_profiling_state_thread_tracking_workaround(void) {
-    if (gvl_profiling_state_thread_tracking_workaround_installed) return;
-
-    rb_internal_thread_add_event_hook(on_thread_start, RUBY_INTERNAL_THREAD_EVENT_STARTED, NULL);
-
-    gvl_profiling_state_thread_tracking_workaround_installed = true;
+  void per_thread_context_tls_init(void) {
+    per_thread_context_key = rb_internal_thread_specific_key_create();
   }
 #endif