crash-reporter: Add diagnostics to help diagnose failures in the wild

user_collector.cc

 // Copyright (c) 2010 The Chromium OS Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "crash-reporter/user_collector.h"
 
 #include <grp.h>  // For struct group.
 #include <pwd.h>  // For struct passwd.
 #include <sys/types.h>  // For getpwuid_r, getgrnam_r, WEXITSTATUS.
 
@@ skipping 50 matching lines @@
   our_path_ = our_path;
   initialized_ = true;
   generate_diagnostics_ = generate_diagnostics;
 }
 
 UserCollector::~UserCollector() {
 }
 
 std::string UserCollector::GetPattern(bool enabled) const {
   if (enabled) {
-    return StringPrintf("|%s --signal=%%s --pid=%%p", our_path_.c_str());
+    return StringPrintf("|%s --signal=%%s --pid=%%p --exec_name=%%e",
+                        our_path_.c_str());
   } else {
     return "core";
   }
 }
 
 bool UserCollector::SetUpInternal(bool enabled) {
   CHECK(initialized_);
   logger_->LogInfo("%s user crash handling",
                    enabled ? "Enabling" : "Disabling");
   if (file_util::WriteFile(FilePath(core_pipe_limit_file_),
@@ skipping 80 matching lines @@
     return false;
   }
   const char *number = ids[kind].c_str();
   char *end_number = NULL;
   *id = strtol(number, &end_number, 10);
   if (*end_number != '\0')
     return false;
   return true;
 }
 
-void UserCollector::LogCollectionError(const std::string &error_message) {
-  error_log_.append(error_message.c_str());
-  error_log_.append("\n");
-  logger_->LogError(error_message.c_str());
-}
-
 void UserCollector::EnqueueCollectionErrorLog(pid_t pid,
                                               const std::string &exec) {
   FilePath crash_path;
   logger_->LogInfo("Writing conversion problems as separate crash report.");
   if (!GetCreatedCrashDirectoryByEuid(0, &crash_path, NULL)) {
     logger_->LogError("Could not even get log directory; out of space?");
     return;
   }
   std::string dump_basename = FormatDumpBasename(exec, time(NULL), pid);
   FilePath log_path = GetCrashPath(crash_path, dump_basename, "log");
   FilePath meta_path = GetCrashPath(crash_path, dump_basename, "meta");
   // We must use WriteNewFile instead of file_util::WriteFile as we do
   // not want to write with root access to a symlink that an attacker
   // might have created.
   WriteNewFile(log_path, error_log_.data(), error_log_.length());
   AddCrashMetaData("sig", kCollectionErrorSignature);
   WriteCrashMetaData(meta_path, exec, log_path.value());
 }
 
 bool UserCollector::CopyOffProcFiles(pid_t pid,
                                      const FilePath &container_dir) {
   if (!file_util::CreateDirectory(container_dir)) {
-    LogCollectionError(StringPrintf("Could not create %s",
-                                    container_dir.value().c_str()));
+    logger_->LogError("Could not create %s",
+                      container_dir.value().c_str());
     return false;
   }
   FilePath process_path = GetProcessPath(pid);
   if (!file_util::PathExists(process_path)) {
-    LogCollectionError(StringPrintf("Path %s does not exist",
-                                    process_path.value().c_str()));
+    logger_->LogError("Path %s does not exist", process_path.value().c_str());
     return false;
   }
   static const char *proc_files[] = {
     "auxv",
     "cmdline",
     "environ",
     "maps",
     "status"
   };
   for (unsigned i = 0; i < arraysize(proc_files); ++i) {
     if (!file_util::CopyFile(process_path.Append(proc_files[i]),
                              container_dir.Append(proc_files[i]))) {
-      LogCollectionError(StringPrintf("Could not copy %s file",
-                                      proc_files[i]));
+      logger_->LogError("Could not copy %s file", proc_files[i]);
       return false;
     }
   }
   return true;
 }
 
 bool UserCollector::GetCreatedCrashDirectory(pid_t pid,
                                              FilePath *crash_file_path,
                                              bool *out_of_capacity) {
   FilePath process_path = GetProcessPath(pid);
   std::string status;
   if (FLAGS_directory_failure_test) {
-    LogCollectionError("Purposefully failing to create spool directory");
+    logger_->LogError("Purposefully failing to create spool directory");
     return false;
   }
   if (!file_util::ReadFileToString(process_path.Append("status"),
                                    &status)) {
-    LogCollectionError("Could not read status file");
+    logger_->LogError("Could not read status file");
+    logger_->LogInfo("Path %s FileExists: %d",
+                     process_path.value().c_str(),
+                     file_util::DirectoryExists(process_path));
     return false;
   }
   int process_euid;
   if (!GetIdFromStatus(kUserId, kIdEffective, status, &process_euid)) {
-    LogCollectionError("Could not find euid in status file");
+    logger_->LogError("Could not find euid in status file");
     return false;
   }
   if (!GetCreatedCrashDirectoryByEuid(process_euid,
                                       crash_file_path,
                                       out_of_capacity)) {
-    LogCollectionError("Could not create crash directory");
+    logger_->LogError("Could not create crash directory");
     return false;
   }
   return true;
 }
 
 bool UserCollector::CopyStdinToCoreFile(const FilePath &core_path) {
   // Copy off all stdin to a core file.
   FilePath stdin_path("/dev/fd/0");
   if (file_util::CopyFile(stdin_path, core_path)) {
     return true;
   }
 
-  LogCollectionError("Could not write core file");
+  logger_->LogError("Could not write core file");
   // If the file system was full, make sure we remove any remnants.
   file_util::Delete(core_path, false);
   return false;
 }
 
 bool UserCollector::RunCoreToMinidump(const FilePath &core_path,
                                       const FilePath &procfs_directory,
                                       const FilePath &minidump_path,
                                       const FilePath &temp_directory) {
   FilePath output_path = temp_directory.Append("output");
   std::vector<const char *> core2md_arguments;
   core2md_arguments.push_back(kCoreToMinidumpConverterPath);
   core2md_arguments.push_back(core_path.value().c_str());
   core2md_arguments.push_back(procfs_directory.value().c_str());
   core2md_arguments.push_back(minidump_path.value().c_str());
 
   if (FLAGS_core2md_failure_test) {
     // To test how core2md errors are propagaged, cause an error
     // by forgetting a required argument.
     core2md_arguments.pop_back();
   }
 
+  std::string errors_during_fork;

[petkov, 2011/01/18 19:18:20]

unused?

[kmixter1, 2011/01/25 21:28:07]

Done.

   int errorlevel = ForkExecAndPipe(core2md_arguments,
                                    output_path.value().c_str());
 
   std::string output;
   file_util::ReadFileToString(output_path, &output);
   if (errorlevel != 0) {
-    LogCollectionError(StringPrintf("Problem during %s [result=%d]: %s",
-                                    kCoreToMinidumpConverterPath,
-                                    errorlevel,
-                                    output.c_str()));
+    logger_->LogError("Problem during %s [result=%d]: %s",
+                      kCoreToMinidumpConverterPath,
+                      errorlevel,
+                      output.c_str());
     return false;
   }
 
   if (!file_util::PathExists(minidump_path)) {
-    LogCollectionError(StringPrintf("Minidump file %s was not created",
-                                    minidump_path.value().c_str()));
+    logger_->LogError("Minidump file %s was not created",
+                      minidump_path.value().c_str());
     return false;
   }
   return true;
 }
 
 bool UserCollector::ConvertCoreToMinidump(pid_t pid,
                                           const FilePath &container_dir,
                                           const FilePath &core_path,
                                           const FilePath &minidump_path) {
   if (!CopyOffProcFiles(pid, container_dir)) {
@@ skipping 15 matching lines @@
   }
 
   return conversion_result;
 }
 
 bool UserCollector::ConvertAndEnqueueCrash(int pid,
                                            const std::string &exec,
                                            bool *out_of_capacity) {
   FilePath crash_path;
   if (!GetCreatedCrashDirectory(pid, &crash_path, out_of_capacity)) {
-    LogCollectionError("Unable to find/create process-specific crash path");
+    logger_->LogError("Unable to find/create process-specific crash path");
     return false;
   }
 
   // Directory like /tmp/crash_reporter.1234 which contains the
   // procfs entries and other temporary files used during conversion.
   FilePath container_dir = FilePath("/tmp").Append(
       StringPrintf("crash_reporter.%d", pid));
   std::string dump_basename = FormatDumpBasename(exec, time(NULL), pid);
   FilePath core_path = GetCrashPath(crash_path, dump_basename, "core");
   FilePath meta_path = GetCrashPath(crash_path, dump_basename, "meta");
@@ skipping 21 matching lines @@
     file_util::Delete(core_path, false);
   } else {
     logger_->LogInfo("Leaving core file at %s due to developer image",
                      core_path.value().c_str());
   }
 
   file_util::Delete(container_dir, true);
   return true;
 }
 
-bool UserCollector::HandleCrash(int signal, int pid, const char *force_exec) {
+bool UserCollector::HandleCrash(int signal, int pid,
+                                const char *kernel_supplied_name,
+                                const char *force_exec) {
   CHECK(initialized_);
   std::string exec;
   if (force_exec) {
     exec.assign(force_exec);
   } else if (!GetExecutableBaseNameFromPid(pid, &exec)) {
-    // If for some reason we don't have the base name, avoid completely
-    // failing by indicating an unknown name.
-    exec = "unknown";
+    // If we cannot find the exec name, use the kernel supplied name.
+    // We don't always use the kernel's since it truncates the name to
+    // 16 characters.
+    if (!kernel_supplied_name)
+      kernel_supplied_name = "null";
+    exec = StringPrintf("supplied_%s", kernel_supplied_name);
   }
 
   // Allow us to test the crash reporting mechanism successfully even if
   // other parts of the system crash.
   if (!FLAGS_filter_in.empty() &&
       (FLAGS_filter_in == "none" ||
        FLAGS_filter_in != exec)) {
     // We use a different format message to make it more obvious in tests
     // which crashes are test generated and which are real.
     logger_->LogWarning("Ignoring crash from %s[%d] while filter_in=%s",
@@ skipping 16 matching lines @@
   }
 
   logger_->LogWarning("Received crash notification for %s[%d] sig %d (%s)",
                       exec.c_str(), pid, signal, handling_string);
 
   if (feedback) {
     count_crash_function_();
 
     if (generate_diagnostics_) {
       bool out_of_capacity = false;
-      if (!ConvertAndEnqueueCrash(pid, exec, &out_of_capacity)) {
+      logger_->set_accumulator(&error_log_);
+      bool convert_and_enqueue_result =
+          ConvertAndEnqueueCrash(pid, exec, &out_of_capacity);
+      logger_->set_accumulator(NULL);
+      if (!convert_and_enqueue_result) {
         if (!out_of_capacity)
           EnqueueCollectionErrorLog(pid, exec);
         return false;
       }
     }
   }
 
   return true;
 }