Enabled line numbers in stack traces

CMakeLists.txt

@@ -106,6 +106,10 @@ endif ()
 if (COMPILER_CLANG)
     # clang: warning: argument unused during compilation: '-specs=/usr/share/dpkg/no-pie-compile.specs' [-Wunused-command-line-argument]
     set (COMMON_WARNING_FLAGS "${COMMON_WARNING_FLAGS} -Wno-unused-command-line-argument")
+    # generate ranges for fast "addr2line" search
+    if (NOT CMAKE_BUILD_TYPE_UC STREQUAL "RELEASE")
+        set(COMPILER_FLAGS "${COMPILER_FLAGS} -gdwarf-aranges")
+    endif ()
 endif ()
 
 option (ENABLE_TESTS "Enables tests" ON)

dbms/CMakeLists.txt

@@ -159,6 +159,11 @@ if (OS_FREEBSD)
     target_compile_definitions (clickhouse_common_io PUBLIC CLOCK_MONOTONIC_COARSE=CLOCK_MONOTONIC_FAST)
 endif ()
 
+if (USE_UNWIND)
+    target_compile_definitions (clickhouse_common_io PRIVATE USE_UNWIND=1)
+    target_include_directories (clickhouse_common_io SYSTEM BEFORE PRIVATE ${UNWIND_INCLUDE_DIR})
+endif ()
+
 add_subdirectory(src/Common/ZooKeeper)
 add_subdirectory(src/Common/Config)
 

dbms/src/Common/Dwarf.cpp

@@ -183,66 +183,6 @@ void skipPadding(std::string_view & sp, const char * start, size_t alignment)
     }
 }
 
-// Simplify a path -- as much as we can while not moving data around...
-/*void simplifyPath(std::string_view & sp)
-{
-    // Strip leading slashes and useless patterns (./), leaving one initial
-    // slash.
-    for (;;)
-    {
-        if (sp.empty())
-        {
-            return;
-        }
-
-        // Strip leading slashes, leaving one.
-        while (sp.startsWith("//"))
-        {
-            sp.remove_prefix(1);
-        }
-
-        if (sp.startsWith("/./"))
-        {
-            // Note 2, not 3, to keep it absolute
-            sp.remove_prefix(2);
-            continue;
-        }
-
-        if (sp.removePrefix("./"))
-        {
-            // Also remove any subsequent slashes to avoid making this path absolute.
-            while (sp.startsWith('/'))
-            {
-                sp.remove_prefix(1);
-            }
-            continue;
-        }
-
-        break;
-    }
-
-    // Strip trailing slashes and useless patterns (/.).
-    for (;;)
-    {
-        if (sp.empty())
-        {
-            return;
-        }
-
-        // Strip trailing slashes, except when this is the root path.
-        while (sp.size() > 1 && sp.removeSuffix('/'))
-        {
-        }
-
-        if (sp.removeSuffix("/."))
-        {
-            continue;
-        }
-
-        break;
-    }
-}*/
-
 }
 
 
@@ -271,10 +211,6 @@ Dwarf::Path::Path(std::string_view baseDir, std::string_view subDir, std::string
         baseDir_ = {}; // subDir_ is absolute
     }
 
-//    simplifyPath(baseDir_);
-//    simplifyPath(subDir_);
-//    simplifyPath(file_);
-
     // Make sure it's never the case that baseDir_ is empty, but subDir_ isn't.
     if (baseDir_.empty())
     {

dbms/src/Common/Exception.h

@@ -6,7 +6,7 @@
 
 #include <Poco/Exception.h>
 
-#include <common/StackTrace.h>
+#include <Common/StackTrace.h>
 
 namespace Poco { class Logger; }
 

dbms/src/Common/QueryProfiler.cpp

@@ -5,7 +5,7 @@
 #include <common/Pipe.h>
 #include <common/phdr_cache.h>
 #include <common/config_common.h>
-#include <common/StackTrace.h>
+#include <Common/StackTrace.h>
 #include <common/StringRef.h>
 #include <common/logger_useful.h>
 #include <Common/CurrentThread.h>

dbms/src/Common/StackTrace.cpp

@@ -1,16 +1,16 @@
-#include <common/StackTrace.h>
 #include <common/SimpleCache.h>
 #include <common/demangle.h>
 
-#include <sstream>
-#include <cstring>
-#include <cxxabi.h>
-#include <execinfo.h>
+#include <Common/StackTrace.h>
+#include <Common/SymbolIndex.h>
+#include <Common/Dwarf.h>
+#include <Common/Elf.h>
+
+#include <sstream>
+#include <filesystem>
+#include <unordered_map>
+#include <cstring>
 
-#if USE_UNWIND
-#define UNW_LOCAL_ONLY
-#include <libunwind.h>
-#endif
 
 std::string signalToErrorMessage(int sig, const siginfo_t & info, const ucontext_t & context)
 {
@@ -168,9 +168,9 @@ void * getCallerAddress(const ucontext_t & context)
 #endif
 #elif defined(__aarch64__)
     return reinterpret_cast<void *>(context.uc_mcontext.pc);
-#endif
-
+#else
     return nullptr;
+#endif
 }
 
 StackTrace::StackTrace()
@@ -195,6 +195,12 @@ StackTrace::StackTrace(NoCapture)
 {
 }
 
+
+#if USE_UNWIND
+extern "C" int unw_backtrace(void **, int);
+#endif
+
+
 void StackTrace::tryCapture()
 {
     size = 0;
@@ -227,50 +233,43 @@ std::string StackTrace::toStringImpl(const Frames & frames, size_t size)
     if (size == 0)
         return "<Empty trace>";
 
-    char ** symbols = backtrace_symbols(frames.data(), size);
-    if (!symbols)
-        return "<Invalid trace>";
+    const DB::SymbolIndex & symbol_index = DB::SymbolIndex::instance();
+    std::unordered_map<std::string, DB::Dwarf> dwarfs;
 
-    std::stringstream backtrace;
-    try
+    std::stringstream out;
+
+    for (size_t i = 0; i < size; ++i)
     {
-        for (size_t i = 0; i < size; i++)
+        out << "#" << i << " " << frames[i] << " ";
+        auto symbol = symbol_index.findSymbol(frames[i]);
+        if (symbol)
         {
-            /// We do "demangling" of names. The name is in parenthesis, before the '+' character.
-
-            char * name_start = nullptr;
-            char * name_end = nullptr;
-            std::string demangled_name;
             int status = 0;
-
-            if (nullptr != (name_start = strchr(symbols[i], '('))
-                && nullptr != (name_end = strchr(name_start, '+')))
-            {
-                ++name_start;
-                *name_end = '\0';
-                demangled_name = demangle(name_start, status);
-                *name_end = '+';
-            }
-
-            backtrace << i << ". ";
-
-            if (0 == status && name_start && name_end)
-            {
-                backtrace.write(symbols[i], name_start - symbols[i]);
-                backtrace << demangled_name << name_end;
-            }
-            else
-                backtrace << symbols[i];
-
-            backtrace << std::endl;
+            out << demangle(symbol->name, status);
         }
-    }
-    catch (...)
-    {
-        free(symbols);
-        throw;
+        else
+            out << "?";
+
+        out << " ";
+
+        if (auto object = symbol_index.findObject(frames[i]))
+        {
+            if (std::filesystem::exists(object->name))
+            {
+                auto dwarf_it = dwarfs.try_emplace(object->name, *object->elf).first;
+
+                DB::Dwarf::LocationInfo location;
+                if (dwarf_it->second.findAddress(uintptr_t(object->address_begin) + uintptr_t(frames[i]), location, DB::Dwarf::LocationInfoMode::FAST))
+                    out << location.file.toString() << ":" << location.line;
+                else
+                    out << object->name;
+            }
+        }
+        else
+            out << "?";
+
+        out << "\n";
     }
 
-    free(symbols);
-    return backtrace.str();
+    return out.str();
 }

dbms/src/Common/SymbolIndex.cpp

@@ -1,6 +1,4 @@
 #include <Common/SymbolIndex.h>
-#include <Common/Elf.h>
-#include <common/demangle.h>
 
 #include <algorithm>
 #include <optional>
@@ -11,6 +9,9 @@
 #include <filesystem>
 
 
+namespace DB
+{
+
 namespace
 {
 
@@ -25,7 +26,7 @@ namespace
 /// Based on the code of musl-libc and the answer of Kanalpiroge on
 /// https://stackoverflow.com/questions/15779185/list-all-the-functions-symbols-on-the-fly-in-c-code-on-a-linux-architecture
 void collectSymbolsFromProgramHeaders(dl_phdr_info * info,
-    std::vector<DB::SymbolIndex::Symbol> & symbols)
+    std::vector<SymbolIndex::Symbol> & symbols)
 {
     /* Iterate over all headers of the current shared lib
      * (first call is for the executable itself) */
@@ -129,13 +130,10 @@ void collectSymbolsFromProgramHeaders(dl_phdr_info * info,
                     if (!sym_name)
                         continue;
 
-                    DB::SymbolIndex::Symbol symbol;
+                    SymbolIndex::Symbol symbol;
                     symbol.address_begin = reinterpret_cast<const void *>(info->dlpi_addr + elf_sym[sym_index].st_value);
                     symbol.address_end = reinterpret_cast<const void *>(info->dlpi_addr + elf_sym[sym_index].st_value + elf_sym[sym_index].st_size);
-                    int unused = 0;
-                    symbol.name = demangle(sym_name, unused);
-                    symbol.object = info->dlpi_name;
-
+                    symbol.name = sym_name;
                     symbols.push_back(std::move(symbol));
                 }
 
@@ -148,10 +146,10 @@ void collectSymbolsFromProgramHeaders(dl_phdr_info * info,
 
 void collectSymbolsFromELFSymbolTable(
     dl_phdr_info * info,
-    const DB::Elf & elf,
-    const DB::Elf::Section & symbol_table,
-    const DB::Elf::Section & string_table,
-    std::vector<DB::SymbolIndex::Symbol> & symbols)
+    const Elf & elf,
+    const Elf::Section & symbol_table,
+    const Elf::Section & string_table,
+    std::vector<SymbolIndex::Symbol> & symbols)
 {
     /// Iterate symbol table.
     const ElfSym * symbol_table_entry = reinterpret_cast<const ElfSym *>(symbol_table.begin());
@@ -170,13 +168,13 @@ void collectSymbolsFromELFSymbolTable(
         /// Find the name in strings table.
         const char * symbol_name = strings + symbol_table_entry->st_name;
 
-        DB::SymbolIndex::Symbol symbol;
+        if (!symbol_name)
+            continue;
+
+        SymbolIndex::Symbol symbol;
         symbol.address_begin = reinterpret_cast<const void *>(info->dlpi_addr + symbol_table_entry->st_value);
         symbol.address_end = reinterpret_cast<const void *>(info->dlpi_addr + symbol_table_entry->st_value + symbol_table_entry->st_size);
-        int unused = 0;
-        symbol.name = demangle(symbol_name, unused);
-        symbol.object = info->dlpi_name;
-
+        symbol.name = symbol_name;
         symbols.push_back(std::move(symbol));
     }
 }
@@ -184,15 +182,15 @@ void collectSymbolsFromELFSymbolTable(
 
 bool searchAndCollectSymbolsFromELFSymbolTable(
     dl_phdr_info * info,
-    const DB::Elf & elf,
+    const Elf & elf,
     unsigned section_header_type,
     const char * string_table_name,
-    std::vector<DB::SymbolIndex::Symbol> & symbols)
+    std::vector<SymbolIndex::Symbol> & symbols)
 {
-    std::optional<DB::Elf::Section> symbol_table;
-    std::optional<DB::Elf::Section> string_table;
+    std::optional<Elf::Section> symbol_table;
+    std::optional<Elf::Section> string_table;
 
-    if (!elf.iterateSections([&](const DB::Elf::Section & section, size_t)
+    if (!elf.iterateSections([&](const Elf::Section & section, size_t)
         {
             if (section.header.sh_type == section_header_type)
                 symbol_table.emplace(section);
@@ -213,8 +211,8 @@ bool searchAndCollectSymbolsFromELFSymbolTable(
 
 
 void collectSymbolsFromELF(dl_phdr_info * info,
-    std::vector<DB::SymbolIndex::Symbol> & symbols,
-    std::vector<DB::SymbolIndex::Object> & objects)
+    std::vector<SymbolIndex::Symbol> & symbols,
+    std::vector<SymbolIndex::Object> & objects)
 {
     std::string object_name = info->dlpi_name;
 
@@ -230,16 +228,17 @@ void collectSymbolsFromELF(dl_phdr_info * info,
     if (ec)
         return;
 
-    DB::Elf elf(object_name);
-
-    DB::SymbolIndex::Object object;
+    SymbolIndex::Object object;
+    object.elf = std::make_unique<Elf>(object_name);
     object.address_begin = reinterpret_cast<const void *>(info->dlpi_addr);
-    object.address_end = reinterpret_cast<const void *>(info->dlpi_addr + elf.size());
+    object.address_end = reinterpret_cast<const void *>(info->dlpi_addr + object.elf->size());
     object.name = object_name;
     objects.push_back(std::move(object));
 
-    searchAndCollectSymbolsFromELFSymbolTable(info, elf, SHT_SYMTAB, ".strtab", symbols);
-    searchAndCollectSymbolsFromELFSymbolTable(info, elf, SHT_DYNSYM, ".dynstr", symbols);
+    searchAndCollectSymbolsFromELFSymbolTable(info, *objects.back().elf, SHT_SYMTAB, ".strtab", symbols);
+
+    /// Unneeded because they were parsed from "program headers" of loaded objects.
+    //searchAndCollectSymbolsFromELFSymbolTable(info, *objects.back().elf, SHT_DYNSYM, ".dynstr", symbols);
 }
 
 
@@ -253,7 +252,7 @@ int collectSymbols(dl_phdr_info * info, size_t, void * data_ptr)
      * (e.g. on a 32 bit system, ElfW(Dyn*) becomes "Elf32_Dyn*")
      */
 
-    DB::SymbolIndex::Data & data = *reinterpret_cast<DB::SymbolIndex::Data *>(data_ptr);
+    SymbolIndex::Data & data = *reinterpret_cast<SymbolIndex::Data *>(data_ptr);
 
     collectSymbolsFromProgramHeaders(info, data.symbols);
     collectSymbolsFromELF(info, data.symbols, data.objects);
@@ -285,9 +284,6 @@ const T * find(const void * address, const std::vector<T> & vec)
 }
 
 
-namespace DB
-{
-
 void SymbolIndex::update()
 {
     dl_iterate_phdr(collectSymbols, &data.symbols);

dbms/src/Common/SymbolIndex.h

@@ -2,6 +2,8 @@
 
 #include <vector>
 #include <string>
+#include <ext/singleton.h>
+#include <Common/Elf.h>
 
 
 namespace DB
@@ -9,16 +11,20 @@ namespace DB
 
 /** Allow to quickly find symbol name from address.
   * Used as a replacement for "dladdr" function which is extremely slow.
+  * It works better than "dladdr" because it also allows to search private symbols, that are not participated in shared linking.
   */
-class SymbolIndex
+class SymbolIndex : public ext::singleton<SymbolIndex>
 {
+protected:
+    friend class ext::singleton<SymbolIndex>;
+    SymbolIndex() { update(); }
+
 public:
     struct Symbol
     {
         const void * address_begin;
         const void * address_end;
-        const char * object;
-        std::string name;   /// demangled NOTE Can use Arena for strings
+        const char * name;
     };
 
     struct Object
@@ -26,11 +32,9 @@ public:
         const void * address_begin;
         const void * address_end;
         std::string name;
+        std::unique_ptr<Elf> elf;
     };
 
-    SymbolIndex() { update(); }
-    void update();
-
     const Symbol * findSymbol(const void * address) const;
     const Object * findObject(const void * address) const;
 
@@ -44,6 +48,8 @@ public:
     };
 private:
     Data data;
+
+    void update();
 };
 
 }

dbms/src/Common/TraceCollector.cpp

@@ -3,7 +3,7 @@
 #include <Core/Field.h>
 #include <Poco/Logger.h>
 #include <common/Pipe.h>
-#include <common/StackTrace.h>
+#include <Common/StackTrace.h>
 #include <common/logger_useful.h>
 #include <IO/ReadHelpers.h>
 #include <IO/ReadBufferFromFileDescriptor.h>

dbms/src/Common/tests/symbol_index.cpp

@@ -18,9 +18,9 @@ int main(int argc, char ** argv)
         return 1;
     }
 
-    SymbolIndex symbol_index;
+    const SymbolIndex & symbol_index = SymbolIndex::instance();
 
-    for (const auto & elem : symbol_index.objects())
+    for (const auto & elem : symbol_index.symbols())
         std::cout << elem.name << ": " << elem.address_begin << " ... " << elem.address_end << "\n";
 
     const void * address = reinterpret_cast<void*>(std::stoull(argv[1], nullptr, 16));
@@ -41,10 +41,12 @@ int main(int argc, char ** argv)
     Dwarf dwarf(elf);
 
     Dwarf::LocationInfo location;
-    if (dwarf.findAddress(uintptr_t(address), location, Dwarf::LocationInfoMode::FULL))
+    if (dwarf.findAddress(uintptr_t(address), location, Dwarf::LocationInfoMode::FAST))
         std::cerr << location.file.toString() << ":" << location.line << "\n";
     else
         std::cerr << "Dwarf: Not found\n";
 
+    std::cerr << StackTrace().toString() << "\n";
+
     return 0;
 }

dbms/src/Functions/registerFunctions.cpp

@@ -40,7 +40,7 @@ void registerFunctionsIntrospection(FunctionFactory &);
 void registerFunctionsNull(FunctionFactory &);
 void registerFunctionsFindCluster(FunctionFactory &);
 void registerFunctionsJSON(FunctionFactory &);
-void registerFunctionSymbolizeAddress(FunctionFactory &);
+void registerFunctionsIntrospection(FunctionFactory &);
 
 void registerFunctions()
 {
@@ -79,7 +79,7 @@ void registerFunctions()
     registerFunctionsNull(factory);
     registerFunctionsFindCluster(factory);
     registerFunctionsJSON(factory);
-    registerFunctionSymbolizeAddress(factory);
+    registerFunctionsIntrospection(factory);
 }
 
 }

dbms/src/Functions/registerFunctionsIntrospection.cpp

@@ -0,0 +1,16 @@
+namespace DB
+{
+
+class FunctionFactory;
+
+void registerFunctionSymbolizeAddress(FunctionFactory & factory);
+void registerFunctionDemangle(FunctionFactory & factory);
+
+void registerFunctionsIntrospection(FunctionFactory & factory)
+{
+    registerFunctionSymbolizeAddress(factory);
+    registerFunctionDemangle(factory);
+}
+
+}
+

dbms/src/Functions/symbolizeAddress.cpp

@@ -15,7 +15,6 @@ namespace ErrorCodes
 {
     extern const int ILLEGAL_COLUMN;
     extern const int ILLEGAL_TYPE_OF_ARGUMENT;
-    extern const int SIZES_OF_ARRAYS_DOESNT_MATCH;
     extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
 }
 
@@ -60,7 +59,7 @@ public:
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {
-        static SymbolIndex symbol_index;
+        const SymbolIndex & symbol_index = SymbolIndex::instance();
 
         const ColumnPtr & column = block.getByPosition(arguments[0]).column;
         const ColumnUInt64 * column_concrete = checkAndGetColumn<ColumnUInt64>(column.get());
@@ -74,7 +73,7 @@ public:
         for (size_t i = 0; i < input_rows_count; ++i)
         {
             if (const auto * symbol = symbol_index.findSymbol(reinterpret_cast<const void *>(data[i])))
-                result_column->insertDataWithTerminatingZero(symbol->name.data(), symbol->name.size() + 1);
+                result_column->insertDataWithTerminatingZero(symbol->name, strlen(symbol->name) + 1);
             else
                 result_column->insertDefault();
         }

dbms/src/Interpreters/Context.cpp

@@ -50,7 +50,7 @@
 #include <Parsers/ASTCreateQuery.h>
 #include <Parsers/ParserCreateQuery.h>
 #include <Parsers/parseQuery.h>
-#include <common/StackTrace.h>
+#include <Common/StackTrace.h>
 #include <Common/Config/ConfigProcessor.h>
 #include <Common/ZooKeeper/ZooKeeper.h>
 #include <Common/ShellCommand.h>

libs/libcommon/CMakeLists.txt

@@ -23,12 +23,10 @@ add_library (common
     src/getThreadNumber.cpp
     src/sleep.cpp
     src/argsToConfig.cpp
-    src/StackTrace.cpp
     src/Pipe.cpp
     src/phdr_cache.cpp
 
     include/common/SimpleCache.h
-    include/common/StackTrace.h
     include/common/Types.h
     include/common/DayNum.h
     include/common/DateLUT.h
@@ -68,14 +66,6 @@ add_library (common
 
     ${CONFIG_COMMON})
 
-if (USE_UNWIND)
-    target_compile_definitions (common PRIVATE USE_UNWIND=1)
-    target_include_directories (common BEFORE PRIVATE ${UNWIND_INCLUDE_DIR})
-    if (NOT USE_INTERNAL_UNWIND_LIBRARY_FOR_EXCEPTION_HANDLING)
-        target_link_libraries (common PRIVATE ${UNWIND_LIBRARY})
-    endif ()
-endif ()
-
 # When testing for memory leaks with Valgrind, dont link tcmalloc or jemalloc.
 
 if (USE_JEMALLOC)

libs/libdaemon/src/BaseDaemon.cpp

@@ -15,7 +15,7 @@
 #include <common/logger_useful.h>
 #include <common/ErrorHandlers.h>
 #include <common/Pipe.h>
-#include <common/StackTrace.h>
+#include <Common/StackTrace.h>
 #include <sys/time.h>
 #include <sys/resource.h>
 #include <iostream>