Adding check_nvme

Imanol-Mikel Barba Sabariego
1 parent 5f26e3fa
Showing 7 changed files with 396 additions and 2 deletions
CMakeLists.txt
check_nvme/README.md
check_nvme/auxiliar.cpp
check_nvme/auxiliar.h
check_nvme/check_nvme.cpp
check_nvme/check_nvme.h
check_zfs/check_zfs.sh
@@ -18,6 +18,9 @@ add_executable(check_memfree_freebsd ${SOURCE_FILES_MEMFREE_FREEBSD})
 set(SOURCE_FILES_SMART check_smart/check_smart.cpp check_smart/auxiliar.cpp)
 add_executable(check_smart ${SOURCE_FILES_SMART})
+set(SOURCE_FILES_NVME check_nvme/check_nvme.cpp check_nvme/auxiliar.cpp)
+add_executable(check_nvme ${SOURCE_FILES_NVME})
+
 find_library(libsensors libsensors.so)
 set(SOURCE_FILES_SENSORS check_sensors/check_sensors.cpp check_sensors/auxiliar.cpp)
 add_executable(check_sensors ${SOURCE_FILES_SENSORS})
+```
+check_smart v1.0
+
+Checks for pending, reallocated or uncorrectable sectors in disks using SMART
+WARNING: Requires the setuid bit to be set to run as root
+
+Usage: 
+check_smart [-hV] DISKS...
+
+Options:
+ -h
+    Print detailed help screen
+ -V
+    Print version information
+ DISKS
+    Disks for which to retrieve S.M.A.R.T data
+```
+//
+// Created by Imanol on 28-may-16.
+//
+
+#include "auxiliar.h"
+
+void timer_handler (int signum)
+{
+    if(signum == SIGVTALRM)
+    {
+        cout << servicename << " CRITICAL - timeout occurred" << endl;
+        exit(2);
+    }
+}
+
+int str2int(string str)
+{
+    int num;
+    stringstream sstream;
+    sstream << str;
+    if(!(sstream >> num))
+    {
+        throw integerConversionException("Integer conversion error");
+    }
+    return num;
+}
+
+string int2str(int x)
+{
+    string str;
+    stringstream sstream;
+    sstream << x;
+    sstream >> str;
+    return str;
+}
+
+int exec(string cmd, string *output)
+{
+    *output = "";
+    FILE* pipe = popen(cmd.c_str(), "r");
+    if (!pipe)
+    {
+        cout << "Error opening child process" << endl;
+        exit(3);
+    }
+    char buffer[128];
+    while(!feof(pipe))
+    {
+        if(fgets(buffer, 128, pipe) != NULL)
+        {
+            *output += buffer;
+        }
+    }
+    return pclose(pipe)/256;
+}
 \ No newline at end of file
+//
+// Created by Imanol on 28-may-16.
+//
+
+#ifndef NAGIOS_PLUGINS_AUXILIAR_H
+#define NAGIOS_PLUGINS_AUXILIAR_H
+
+#include <sstream>
+#include <iostream>
+#include <exception>
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <signal.h>
+
+using namespace std;
+
+extern const char *servicename;
+
+int str2int(string str);
+string int2str(int x);
+int exec(string cmd, string *output);
+void timer_handler (int signum);
+
+class integerConversionException : public exception
+{
+private:
+    string s;
+public:
+    integerConversionException(std::string ss) : s(ss) {}
+    ~integerConversionException() throw () {}
+    const char* what() const throw() { return s.c_str(); }
+};
+
+#endif //NAGIOS_PLUGINS_AUXILIAR_H
+#include "check_nvme.h"
+#include <regex>
+
+const char *servicename = (const char*)"NVME";
+
+int getNVMeAttrValue(string line) {
+  line = std::regex_replace(line, std::regex("\\s+"), " ");
+  line = std::regex_replace(line, std::regex("^ "), "");
+
+  return stoi(line.substr(line.find(":")+1));
+}
+
+std::string getNVMeAttrName(string line) {
+  size_t first = line.find_first_not_of(' ');
+  size_t last = line.find_first_of(':',first);
+  return line.substr(first,last-first);
+}
+
+int evalStatus(const char* disk, string *status) {
+  string output = "";
+  string line = "";
+
+  int rc = run_smartctl_cmd(string(SMARTCTL_CMD_ATTRS), disk,&output);
+  if(rc) {
+    cout << "Error reading SMART data from disk " << disk << endl;
+    exit(UNKN);
+  }
+
+  map<std::string,NVMeAttr> attrMap;
+  attrMap[AVAILABLE_SPARE] = spare;
+  attrMap[MEDIA_ERRORS] = errors;
+  attrMap[PERCENTAGE_USED] = used;
+
+  stringstream sstream;
+  sstream.str(output);
+  while(getline(sstream,line)) {
+    if(line == "") {
+      continue;
+    }
+    for(map<std::string,NVMeAttr>::iterator it = attrMap.begin(); it != attrMap.end(); ++it) {
+      std::string name = it->first;
+      if(getNVMeAttrName(line) == name) {
+        attrMap[name].value = getNVMeAttrValue(line);
+      }
+    }
+  }
+
+  int ret = OK;
+  *status = string(disk);
+  for(map<std::string, NVMeAttr>::iterator it = attrMap.begin(); it != attrMap.end(); ++it) {
+    NVMeAttr attr = it->second;
+
+    if(attr.value == -1) {
+      if(attr.optional) {
+        continue;
+      }
+      *status = string(disk) + " status UNKNOWN";
+      return UNKN;
+    }
+
+    int veredict = 0;
+
+    if(attr.lower_than) {
+      if(attr.value < attr.threshold_warn) {
+        veredict = WARN;
+      }
+      if(attr.threshold_crit != -1 && attr.value < attr.threshold_crit) {
+        veredict = CRIT;
+      }
+    } else {
+      if(attr.value > attr.threshold_warn) {
+        veredict = WARN;
+      }
+      if(attr.threshold_crit != -1 && attr.value > attr.threshold_crit) {
+        veredict = CRIT;
+      }
+    }
+
+    switch(veredict) {
+      case OK:
+        break;
+      case WARN:
+        if(ret == OK) {
+          ret = WARN;
+        }
+        break;
+      case CRIT:
+        ret = CRIT;
+        break;
+    }
+    *status += " " + attr.name + ":" + to_string(attr.value);
+  }
+  return ret;
+}
+
+int run_smartctl_cmd(const string command, const char* disk, string* output) {
+  int uid = getuid();
+  setreuid(0,0);
+
+  int rc = exec(command + disk,output);
+
+  setreuid(uid,0);
+  return rc;
+}
+
+void printVersion() {
+  cout << "check_nvme v" << VERSION << endl << endl;
+}
+
+void printHelp(bool longVersion) {
+  if(longVersion) {
+    printVersion();
+    cout << "Checks for pending, reallocated or uncorrectable sectors in NVMe disks" << endl << "WARNING: Requires the setuid bit to be set to run as root" << endl << endl;
+    printHelp(false);
+    cout << "Options:" << endl;
+    cout << " -h" << endl;
+    cout << "    Print detailed help screen" << endl;
+    cout << " -V" << endl;
+    cout << "    Print version information" << endl;
+    cout << " DISKS" << endl;
+    cout << "    Disks for which to retrieve S.M.A.R.T data" << endl << endl;
+    return;
+  }
+  cout << "Usage: " << endl << "check_smart [-hV] DISKS..." << endl << endl;
+}
+
+void set_timeout(unsigned int sec) {
+  struct itimerval timer;
+  timer.it_value.tv_sec = sec;
+  timer.it_value.tv_usec = 0;
+  timer.it_interval.tv_sec = 0;
+  timer.it_interval.tv_usec = 0;
+  setitimer (ITIMER_VIRTUAL, &timer, 0);
+
+  struct sigaction sa;
+  memset (&sa, 0, sizeof (sa));
+  sa.sa_handler = &timer_handler;
+  sigaction (SIGVTALRM, &sa, 0);
+}
+
+int main(int argc, char **argv) {
+  set_timeout(10);
+  int c;
+  while ((c = getopt (argc, argv, "Vh")) != -1) {
+    switch(c) {
+      case 'h':
+        printHelp(true);
+        return OK;
+      case 'V':
+        printVersion();
+        return OK;
+      case '?':
+        printHelp(false);
+        return UNKN;
+    }
+  }
+
+  if(argc == 1) {
+    cout << "No disks checked" << endl;
+    return 3;
+  }
+
+  string *results = new string[argc-1];
+  int returnCode = OK;
+
+  for(int i = 1; i < argc; i++) {
+    switch(evalStatus(argv[i],&(results[i-1]))) {
+      case OK:
+        break;
+      case WARN:
+        if(returnCode == OK) {returnCode = WARN;}
+        break;
+      case CRIT:
+        returnCode = CRIT;
+        break;
+      case UNKN:
+        returnCode = UNKN;
+        break;
+    }
+  }
+
+  cout << servicename;
+  if(returnCode == OK) {cout << " OK - disk status: ";}
+  else if(returnCode == WARN) {cout << " WARNING - disk status: ";}
+  else if(returnCode == CRIT) {cout << " CRITICAL - disk status: ";}
+  else if(returnCode == UNKN) {cout << " UNKNOWN - disk status: ";}
+  for(int i = 0; i < (argc-1); i++) {
+    cout << results[i];
+    if(i != (argc-2)) {cout << ", ";}
+  }
+  cout << endl;
+  return returnCode;
+}
+#ifndef CHECK_NVME_H
+#define CHECK_NVME_H
+
+#include <iostream>
+#include <sstream>
+#include <map>
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+#include <sys/time.h>
+
+#include "auxiliar.h"
+
+#define OK 0
+#define WARN 1
+#define CRIT 2
+#define UNKN 3
+
+#define VERSION "1.0"
+
+#define SMARTCTL_CMD_ATTRS "/usr/sbin/smartctl -A "
+#define SMARTCTL_CMD_INFO "/usr/sbin/smartctl -i "
+
+#define ROTATION_INFO_STR "Rotation Rate:"
+#define SSD_DEV_STR "Solid State Device"
+
+#define AVAILABLE_SPARE "Available Spare"
+#define MEDIA_ERRORS "Media and Data Integrity Errors"
+#define PERCENTAGE_USED "Percentage Used"
+
+using namespace std;
+
+struct NVMeAttr
+{
+  string name;
+  int value;
+  int threshold_warn;
+  int threshold_crit;
+  bool optional;
+  bool lower_than;
+}; typedef struct NVMeAttr NVMeAttr;
+
+
+// Attribute definitions
+NVMeAttr used = {
+  .name = PERCENTAGE_USED,
+  .value = -1,
+  .threshold_warn = 80,
+  .threshold_crit = 90,
+  .optional = false,
+  .lower_than = false,
+};
+
+NVMeAttr spare = {
+  .name = AVAILABLE_SPARE,
+  .value = -1,
+  .threshold_warn = 20,
+  .threshold_crit = 10,
+  .optional = false,
+  .lower_than = true,
+};
+
+NVMeAttr errors = {
+  .name = MEDIA_ERRORS,
+  .value = -1,
+  .threshold_warn = 0,
+  .threshold_crit = 0,
+  .optional = false,
+  .lower_than = false,
+};
+
+map<std::string,NVMeAttr> prepareAttrMap(int driveType);
+int getNVMeAttrValue(string line);
+std::string getNVMeAttrName(string line);
+int run_smartctl_cmd(const string command, const char* disk, string* output);
+int evalStatus(const char* disk, int driveType, string* status);
+void printVersion();
+void printHelp(bool longVersion);
+void set_timeout(unsigned int sec);
+
+#endif
@@ -11,8 +11,15 @@ function smart_disks() {
   ERR_OUTPUT=""
   HIGHEST_RES=0
   for disk in $DISKS; do
-    OUTPUT=$(/usr/lib64/nagios/plugins/check_smart $disk)
-    RES=$?
+    OUTPUT=""
+    RES=0
+    if echo $disk | grep "nvme" > /dev/null; then
+      OUTPUT=$(/usr/lib64/nagios/plugins/check_nvme $disk)
+      RES=$?
+    else
+      OUTPUT=$(/usr/lib64/nagios/plugins/check_smart $disk)
+      RES=$?
+    fi
     if [[ $RES != 0 ]]; then
       ERR_OUTPUT="$ERR_OUTPUT - $OUTPUT"
     fi