add parent pid and registers collecting (#1616)

Author: denisichh

common/server/crash-dump.cpp

@@ -14,12 +14,6 @@
 #include "common/ucontext/ucontext-portable.h"
 #include <ucontext.h>
 
-struct crash_dump_buffer {
-  char scratchpad[1024];
-  size_t position;
-};
-using crash_dump_buffer_t = struct crash_dump_buffer;
-
 static inline char crash_dump_half_byte_char(uint8_t hb) {
   if (hb <= 9) {
     return '0' + hb;
@@ -67,7 +61,7 @@ static inline void crash_dump_write_uint64(uint64_t value, crash_dump_buffer_t*
 // Keep in mind that:
 //  * `ucontext_t_portable` -- using for more efficient user context manipulations (e.g. `swapcontext`, `getcontext`, `setcontext`, etc)
 //  * `ucontext_t` -- using in signal handlers for machine state extracting in debug purposes.
-static inline void crash_dump_prepare_registers([[maybe_unused]] crash_dump_buffer_t* buffer, [[maybe_unused]] void* ucontext) {
+void crash_dump_prepare_registers([[maybe_unused]] crash_dump_buffer_t* buffer, [[maybe_unused]] const ucontext_t* uc) {
 #ifdef __x86_64__
 #ifdef __APPLE__
   const auto* uc = static_cast<ucontext_t*>(ucontext);
@@ -93,8 +87,6 @@ static inline void crash_dump_prepare_registers([[maybe_unused]] crash_dump_buff
   crash_dump_write_reg(LITERAL_WITH_LENGTH("R14=0x"), uc->uc_mcontext->__ss.__r14, buffer);
   crash_dump_write_reg(LITERAL_WITH_LENGTH("R15=0x"), uc->uc_mcontext->__ss.__r15, buffer);
 #else
-  const auto* uc = static_cast<ucontext_t*>(ucontext);
-
   crash_dump_write_reg(LITERAL_WITH_LENGTH("RIP=0x"), uc->uc_mcontext.gregs[REG_RIP], buffer);
   crash_dump_write_reg(LITERAL_WITH_LENGTH("RSP=0x"), uc->uc_mcontext.gregs[REG_RSP], buffer);
   crash_dump_write_reg(LITERAL_WITH_LENGTH("RBP=0x"), uc->uc_mcontext.gregs[REG_RBP], buffer);
@@ -131,7 +123,7 @@ void crash_dump_write(void* ucontext) {
 
   static crash_dump_buffer_t buffer;
   buffer.position = 0;
-  crash_dump_prepare_registers(&buffer, ucontext);
+  crash_dump_prepare_registers(&buffer, static_cast<ucontext_t*>(ucontext));
 
   assert(buffer.position < sizeof(buffer.scratchpad));
   buffer.scratchpad[buffer.position++] = '\n';

common/server/crash-dump.h

@@ -3,5 +3,19 @@
 // Distributed under the GPL v3 License, see LICENSE.notice.txt
 
 #pragma once
+#include <cstddef>
+#include <cstdint>
+#include <string_view>
+#include <ucontext.h>
 
+struct crash_dump_buffer_t { // NOLINT
+  char scratchpad[1024];
+  size_t position{0};
+
+  std::string_view get_content() const noexcept {
+    return std::string_view{scratchpad, position};
+  }
+};
+
+void crash_dump_prepare_registers(crash_dump_buffer_t* buffer, const ucontext_t* uc);
 void crash_dump_write(void* ucontext);

server/json-logger.cpp

@@ -6,10 +6,13 @@
 #include <cinttypes>
 #include <execinfo.h>
 #include <fcntl.h>
+#include <ucontext.h>
 #include <unistd.h>
+#include <memory>
 
 #include "common/algorithms/find.h"
 #include "common/fast-backtrace.h"
+#include "common/server/crash-dump.h"
 #include "common/wrappers/likely.h"
 #include "common/ucontext/ucontext-portable.h"
 #include "runtime/kphp-backtrace.h"
@@ -259,7 +262,7 @@ void JsonLogger::write_log_with_demangled_backtrace(vk::string_view message,int
 }
 
 void JsonLogger::write_log(vk::string_view message, int type, int64_t created_at,
-                           void *const *trace, int64_t trace_size, bool uncaught) noexcept {
+                           void *const *trace, int64_t trace_size, bool uncaught, void* ucontext) noexcept {
   if (json_log_fd_ <= 0) {
     return;
   }
@@ -269,7 +272,7 @@ void JsonLogger::write_log(vk::string_view message, int type, int64_t created_at
   }
   assert(json_out_it != buffers_.end());
 
-  write_general_info(json_out_it, type, created_at, uncaught);
+  write_general_info(json_out_it, type, created_at, uncaught, ucontext);
 
   json_out_it->append_key("trace").start<'['>();
   for (int64_t i = 0; i < trace_size; i++) {
@@ -324,7 +327,7 @@ void JsonLogger::reset_buffers() noexcept {
   }
 }
 
-void JsonLogger::write_general_info(JsonBuffer *json_out_it, int type, int64_t created_at, bool uncaught) {
+void JsonLogger::write_general_info(JsonBuffer *json_out_it, int type, int64_t created_at, bool uncaught, void* ucontext) {
   json_out_it->append_key("version").append_integer(release_version_);
   json_out_it->append_key("hostname").append_string(hostname_);
   json_out_it->append_key("type").append_integer(type);
@@ -348,12 +351,25 @@ void JsonLogger::write_general_info(JsonBuffer *json_out_it, int type, int64_t c
     json_out_it->append_key("logname_id").append_integer(logname_id);
   }
   json_out_it->append_key("pid").append_integer(pid);
+  json_out_it->append_key("ppid").append_integer(ppid);
   json_out_it->append_key("cluster").append_string(vk::singleton<ServerConfig>::get().get_cluster_name());
   json_out_it->append_raw(uncaught ? R"json("uncaught":true)json" : R"json("uncaught":false)json");
   json_out_it->finish<'}'>();
 
   if (extra_info_available_) {
-    json_out_it->append_key("extra_info").start<'{'>().append_raw(extra_info_).finish<'}'>();
+    json_out_it->append_key("extra_info").start<'{'>().append_raw(extra_info_);
+    if (ucontext != nullptr) {
+      const auto* ucp = static_cast<ucontext_t*>(ucontext);
+
+#if defined(__x86_64__) && !defined(__APPLE__)
+      json_out_it->append_key("CR2 register").append_hex_as_string(ucp->uc_mcontext.gregs[REG_CR2]);
+#endif
+
+      crash_dump_buffer_t buffer{};
+      crash_dump_prepare_registers(std::addressof(buffer), ucp);
+      json_out_it->append_key("registers").append_string(buffer.get_content());
+    }
+    json_out_it->finish<'}'>();
   }
 }
 

server/json-logger.h

@@ -47,7 +47,7 @@ class JsonLogger : vk::not_copyable {
   // ATTENTION: this functions are used in signal handlers, therefore they are expected to be safe for them
   // Details: https://man7.org/linux/man-pages/man7/signal-safety.7.html
   // todo: functions bellow use backtrace which isn't async-signal safety
-  void write_log(vk::string_view message, int type, int64_t created_at, void *const *trace, int64_t trace_size, bool uncaught) noexcept;
+  void write_log(vk::string_view message, int type, int64_t created_at, void *const *trace, int64_t trace_size, bool uncaught, void* ucontext = nullptr) noexcept;
   void write_log_with_backtrace(vk::string_view message, int type) noexcept;
   void write_log_with_script_backtrace(vk::string_view message, int type) noexcept;
 
@@ -112,6 +112,6 @@ class JsonLogger : vk::not_copyable {
   };
   std::array<JsonBuffer, 8> buffers_;
 
-  void write_general_info(JsonBuffer * json_out_it, int type, int64_t created_at, bool uncaught);
+  void write_general_info(JsonBuffer * json_out_it, int type, int64_t created_at, bool uncaught, void* ucontext = nullptr);
 };
 

server/php-engine-vars.cpp

@@ -22,6 +22,7 @@ double oom_handling_memory_ratio = 0.00;
 
 int worker_id = -1;
 int pid = -1;
+int ppid = -1;
 int master_pid = -1;
 
 ProcessType process_type = ProcessType::master;

server/php-engine-vars.h

@@ -32,6 +32,7 @@ extern double oom_handling_memory_ratio;
 
 extern int worker_id;
 extern int pid;
+extern int ppid;
 extern int master_pid;
 
 extern ProcessType process_type;

server/php-engine.cpp

@@ -2381,6 +2381,7 @@ void init_default() {
   now = (int)time(nullptr);
 
   pid = getpid();
+  ppid = getppid();
   master_pid = getpid();
   // RPC part
   PID.port = -1; // TODO: get rid of this?

server/php-master.cpp

@@ -638,6 +638,7 @@ int run_worker(WorkerType worker_type) {
     //verbosity = 0;
     verbosity = initial_verbosity;
     pid = getpid();
+    ppid = getppid();
 
     master_sfd = -1;
     if (worker_type == WorkerType::job_worker) {

server/signal-handlers.cpp

@@ -195,7 +195,7 @@ void sigsegv_handler(int signum, siginfo_t *info, void *ucontext) {
     }
   } else {
     const char *msg = signum == SIGBUS ? "SIGBUS terminating program" : "SIGSEGV terminating program";
-    vk::singleton<JsonLogger>::get().write_log(msg, static_cast<int>(ServerLog::Critical), cur_time, trace, trace_size, true);
+    vk::singleton<JsonLogger>::get().write_log(msg, static_cast<int>(ServerLog::Critical), cur_time, trace, trace_size, true, ucontext);
     vk::singleton<JsonLogger>::get().fsync_log_file();
     write_str(2, "Error -2: Segmentation fault\n");
     print_http_data();
@@ -206,15 +206,15 @@ void sigsegv_handler(int signum, siginfo_t *info, void *ucontext) {
   }
 }
 
-void sigabrt_handler(int, siginfo_t *info, void *) {
+void sigabrt_handler(int, siginfo_t *info, void *ucontext) {
   const int64_t cur_time = time(nullptr);
   void *trace[64];
   const int trace_size = backtrace(trace, 64);
   vk::string_view msg{dl_get_assert_message()};
   if (msg.empty()) {
     msg = "SIGABRT terminating program";
   }
-  vk::singleton<JsonLogger>::get().write_log(msg, static_cast<int>(ServerLog::Critical), cur_time, trace, trace_size, true);
+  vk::singleton<JsonLogger>::get().write_log(msg, static_cast<int>(ServerLog::Critical), cur_time, trace, trace_size, true, ucontext);
   vk::singleton<JsonLogger>::get().fsync_log_file();
 
   print_prologue(cur_time);

tests/python/lib/kphp_server.py

@@ -85,7 +85,13 @@ def _process_json_log(self, log_record):
             del log_record["tags"]["logname_id"]
         del log_record["tags"]["process_type"]
         del log_record["tags"]["pid"]
+        del log_record["tags"]["ppid"]
         if not got_tags.get("cluster", ""):
             raise RuntimeError("Got an empty cluster in json log: {}".format(got_tags))
         del log_record["tags"]["cluster"]
+
+        if log_record.get("extra_info"):
+            log_record["extra_info"].pop("CR2 register", None)
+            log_record["extra_info"].pop("registers", None)
+
         return log_record