smartd.cpp

硬盘各项性能的测试,如温度容量版本健康度型号

	   "Use ID = 0 to turn off -C and/or -U Directives\n"
           "Example: /dev/hda -a\n", 
           configfile);
  return;
}

/* Returns a pointer to a static string containing a formatted list of the valid
   arguments to the option opt or NULL on failure. */
const char *GetValidArgList(char opt) {
  switch (opt) {
  case 'c':
    return "<FILE_NAME>, -";
  case 's':
    return "valid_regular_expression";
  case 'l':
    return "daemon, local0, local1, local2, local3, local4, local5, local6, local7";
  case 'q':
    return "nodev, errors, nodevstartup, never, onecheck, showtests";
  case 'r':
    return "ioctl[,N], ataioctl[,N], scsiioctl[,N]";
  case 'p':
    return "<FILE_NAME>";
  case 'i':
    return "<INTEGER_SECONDS>";
  default:
    return NULL;
  }
}

/* prints help information for command syntax */
void Usage (void){
  PrintOut(LOG_INFO,"Usage: smartd [options]\n\n");
#ifdef HAVE_GETOPT_LONG
  PrintOut(LOG_INFO,"  -c NAME|-, --configfile=NAME|-\n");
  PrintOut(LOG_INFO,"        Read configuration file NAME or stdin [default is %s]\n\n", configfile);
  PrintOut(LOG_INFO,"  -d, --debug\n");
  PrintOut(LOG_INFO,"        Start smartd in debug mode\n\n");
  PrintOut(LOG_INFO,"  -D, --showdirectives\n");
  PrintOut(LOG_INFO,"        Print the configuration file Directives and exit\n\n");
  PrintOut(LOG_INFO,"  -h, --help, --usage\n");
  PrintOut(LOG_INFO,"        Display this help and exit\n\n");
  PrintOut(LOG_INFO,"  -i N, --interval=N\n");
  PrintOut(LOG_INFO,"        Set interval between disk checks to N seconds, where N >= 10\n\n");
  PrintOut(LOG_INFO,"  -l local[0-7], --logfacility=local[0-7]\n");
#ifndef _WIN32
  PrintOut(LOG_INFO,"        Use syslog facility local0 - local7 or daemon [default]\n\n");
#else
  PrintOut(LOG_INFO,"        Log to \"./smartd.log\", stdout, stderr [default is event log]\n\n");
#endif
#ifndef _WIN32
  PrintOut(LOG_INFO,"  -n, --no-fork\n");
  PrintOut(LOG_INFO,"        Do not fork into background\n\n");
#endif  // _WIN32
  PrintOut(LOG_INFO,"  -p NAME, --pidfile=NAME\n");
  PrintOut(LOG_INFO,"        Write PID file NAME\n\n");
  PrintOut(LOG_INFO,"  -q WHEN, --quit=WHEN\n");
  PrintOut(LOG_INFO,"        Quit on one of: %s\n\n", GetValidArgList('q'));
  PrintOut(LOG_INFO,"  -r, --report=TYPE\n");
  PrintOut(LOG_INFO,"        Report transactions for one of: %s\n\n", GetValidArgList('r'));
#ifdef _WIN32
  PrintOut(LOG_INFO,"  --service\n");
  PrintOut(LOG_INFO,"        Running as windows service (see man page), install with:\n");
  PrintOut(LOG_INFO,"          smartd install [options]\n");
  PrintOut(LOG_INFO,"        Remove service with:\n");
  PrintOut(LOG_INFO,"          smartd remove\n\n");
#else
#endif // _WIN32 || __CYGWIN__
  PrintOut(LOG_INFO,"  -V, --version, --license, --copyright\n");
  PrintOut(LOG_INFO,"        Print License, Copyright, and version information\n");
#else
  PrintOut(LOG_INFO,"  -c NAME|-  Read configuration file NAME or stdin [default is %s]\n", configfile);
  PrintOut(LOG_INFO,"  -d         Start smartd in debug mode\n");
  PrintOut(LOG_INFO,"  -D         Print the configuration file Directives and exit\n");
  PrintOut(LOG_INFO,"  -h         Display this help and exit\n");
  PrintOut(LOG_INFO,"  -i N       Set interval between disk checks to N seconds, where N >= 10\n");
  PrintOut(LOG_INFO,"  -l local?  Use syslog facility local0 - local7, or daemon\n");
  PrintOut(LOG_INFO,"  -n         Do not fork into background\n");
  PrintOut(LOG_INFO,"  -p NAME    Write PID file NAME\n");
  PrintOut(LOG_INFO,"  -q WHEN    Quit on one of: %s\n", GetValidArgList('q'));
  PrintOut(LOG_INFO,"  -r TYPE    Report transactions for one of: %s\n", GetValidArgList('r'));
  PrintOut(LOG_INFO,"  -V         Print License, Copyright, and version information\n");
#endif
}

// returns negative if problem, else fd>=0
static int OpenDevice(char *device, char *mode, int scanning) {
  int fd;
  char *s=device;
  
  // If there is an ASCII "space" character in the device name,
  // terminate string there.  This is for 3ware and highpoint devices only.
  if ((s=strchr(device,' ')))
    *s='\0';

  // open the device
  fd = deviceopen(device, mode);

  // if we removed a space, put it back in please
  if (s)
    *s=' ';

  // if we failed to open the device, complain!
  if (fd < 0) {

    // For linux+devfs, a nonexistent device gives a strange error
    // message.  This makes the error message a bit more sensible.
    // If no debug and scanning - don't print errors
    if (debugmode || !scanning) {
      if (errno==ENOENT || errno==ENOTDIR)
        errno=ENODEV;

      PrintOut(LOG_INFO,"Device: %s, %s, open() failed\n",
	       device, strerror(errno));
    }
    return -1;
  }
  // device opened sucessfully
  return fd;
}

int CloseDevice(int fd, char *name){
  if (deviceclose(fd)){
    PrintOut(LOG_INFO,"Device: %s, %s, close(%d) failed\n", name, strerror(errno), fd);
    return 1;
  }
  // device sucessfully closed
  return 0;
}

// returns <0 on failure
int ATAErrorCount(int fd, char *name){
  struct ata_smart_errorlog log;
  
  if (-1==ataReadErrorLog(fd,&log)){
    PrintOut(LOG_INFO,"Device: %s, Read SMART Error Log Failed\n",name);
    return -1;
  }
  
  // return current number of ATA errors
  return log.error_log_pointer?log.ata_error_count:0;
}

// returns <0 if problem.  Otherwise, bottom 8 bits are the self test
// error count, and top bits are the power-on hours of the last error.
int SelfTestErrorCount(int fd, char *name){
  struct ata_smart_selftestlog log;

  if (-1==ataReadSelfTestLog(fd,&log)){
    PrintOut(LOG_INFO,"Device: %s, Read SMART Self Test Log Failed\n",name);
    return -1;
  }
  
  // return current number of self-test errors
  return ataPrintSmartSelfTestlog(&log,0);
}

// scan to see what ata devices there are, and if they support SMART
int ATADeviceScan(cfgfile *cfg, int scanning){
  int fd, supported=0;
  struct ata_identify_device drive;
  char *name=cfg->name;
  int retainsmartdata=0;
  int retid;
  char *mode;
  
  // should we try to register this as an ATA device?
  switch (cfg->controller_type) {
  case CONTROLLER_ATA:
  case CONTROLLER_3WARE_678K:
  case CONTROLLER_MARVELL_SATA:
  case CONTROLLER_HPT:
  case CONTROLLER_UNKNOWN:
    mode="ATA";
    break;
  case CONTROLLER_3WARE_678K_CHAR:
    mode="ATA_3WARE_678K";
    break;
  case CONTROLLER_3WARE_9000_CHAR:
    mode="ATA_3WARE_9000";
    break;
  case CONTROLLER_SAT:
    mode="SCSI";
    break;
  default:
    // not a recognized ATA or SATA device.  We should never enter
    // this branch.
    return 1;
  }
  
  // open the device
  if ((fd=OpenDevice(name, mode, scanning))<0)
    // device open failed
    return 1;
  PrintOut(LOG_INFO,"Device: %s, opened\n", name);
  
  // pass user settings on to low-level ATA commands
  con->controller_port=cfg->controller_port;
  con->hpt_data[0]=cfg->hpt_data[0];
  con->hpt_data[1]=cfg->hpt_data[1];
  con->hpt_data[2]=cfg->hpt_data[2];
  con->controller_type=cfg->controller_type;
  con->controller_explicit=cfg->controller_explicit;
  con->fixfirmwarebug = cfg->fixfirmwarebug;
  con->satpassthrulen = cfg->satpassthrulen;
  
  // Get drive identity structure
  if ((retid=ataReadHDIdentity (fd,&drive))){
    if (retid<0)
      // Unable to read Identity structure
      PrintOut(LOG_INFO,"Device: %s, not ATA, no IDENTIFY DEVICE Structure\n",name);
    else
      PrintOut(LOG_INFO,"Device: %s, packet devices [this device %s] not SMART capable\n",
               name, packetdevicetype(retid-1));
    CloseDevice(fd, name);
    return 2; 
  }

  // Show if device in database, and use preset vendor attribute
  // options unless user has requested otherwise.
  if (cfg->ignorepresets)
    PrintOut(LOG_INFO, "Device: %s, smartd database not searched (Directive: -P ignore).\n", name);
  else {
    // do whatever applypresets decides to do. Will allocate memory if
    // cfg->attributedefs is needed.
    if (applypresets(&drive, &cfg->attributedefs, con)<0)
      PrintOut(LOG_INFO, "Device: %s, not found in smartd database.\n", name);
    else
      PrintOut(LOG_INFO, "Device: %s, found in smartd database.\n", name);
    
    // then save the correct state of the flag (applypresets may have changed it)
    cfg->fixfirmwarebug = con->fixfirmwarebug;
  }
  
  // If requested, show which presets would be used for this drive
  if (cfg->showpresets) {
    int savedebugmode=debugmode;
    PrintOut(LOG_INFO, "Device %s: presets are:\n", name);
    if (!debugmode)
      debugmode=2;
    showpresets(&drive);
    debugmode=savedebugmode;
  }

  // see if drive supports SMART
  supported=ataSmartSupport(&drive);
  if (supported!=1) {
    if (supported==0)
      // drive does NOT support SMART
      PrintOut(LOG_INFO,"Device: %s, lacks SMART capability\n",name);
    else
      // can't tell if drive supports SMART
      PrintOut(LOG_INFO,"Device: %s, ATA IDENTIFY DEVICE words 82-83 don't specify if SMART capable.\n",name);
  
    // should we proceed anyway?
    if (cfg->permissive){
      PrintOut(LOG_INFO,"Device: %s, proceeding since '-T permissive' Directive given.\n",name);
    }
    else {
      PrintOut(LOG_INFO,"Device: %s, to proceed anyway, use '-T permissive' Directive.\n",name);
      CloseDevice(fd, name);
      return 2;
    }
  }
  
  if (ataEnableSmart(fd)){
    // Enable SMART command has failed
    PrintOut(LOG_INFO,"Device: %s, could not enable SMART capability\n",name);
    CloseDevice(fd, name);
    return 2; 
  }
  
  // disable device attribute autosave...
  if (cfg->autosave==1){
    if (ataDisableAutoSave(fd))
      PrintOut(LOG_INFO,"Device: %s, could not disable SMART Attribute Autosave.\n",name);
    else
      PrintOut(LOG_INFO,"Device: %s, disabled SMART Attribute Autosave.\n",name);
  }

  // or enable device attribute autosave
  if (cfg->autosave==2){
    if (ataEnableAutoSave(fd))
      PrintOut(LOG_INFO,"Device: %s, could not enable SMART Attribute Autosave.\n",name);
    else
      PrintOut(LOG_INFO,"Device: %s, enabled SMART Attribute Autosave.\n",name);
  }

  // capability check: SMART status
  if (cfg->smartcheck && ataSmartStatus2(fd)==-1){
    PrintOut(LOG_INFO,"Device: %s, not capable of SMART Health Status check\n",name);
    cfg->smartcheck=0;
  }
  
  // capability check: Read smart values and thresholds.  Note that
  // smart values are ALSO needed even if we ONLY want to know if the
  // device is self-test log or error-log capable!  After ATA-5, this
  // information was ALSO reproduced in the IDENTIFY DEVICE response,
  // but sadly not for ATA-5.  Sigh.

  // do we need to retain SMART data after returning from this routine?
  retainsmartdata=cfg->usagefailed || cfg->prefail || cfg->usage || cfg->tempdiff || cfg->tempinfo || cfg->tempcrit;
  
  // do we need to get SMART data?
  if (retainsmartdata || cfg->autoofflinetest || cfg->selftest || cfg->errorlog || cfg->pending!=DONT_MONITOR_UNC) {

    unsigned char currentpending, offlinepending;

    cfg->smartval=(struct ata_smart_values *)Calloc(1,sizeof(struct ata_smart_values));
    cfg->smartthres=(struct ata_smart_thresholds_pvt *)Calloc(1,sizeof(struct ata_smart_thresholds_pvt));
    
    if (!cfg->smartval || !cfg->smartthres){
      PrintOut(LOG_CRIT,"Not enough memory to obtain SMART data\n");
      EXIT(EXIT_NOMEM);
    }
    
    if (ataReadSmartValues(fd,cfg->smartval) ||
        ataReadSmartThresholds (fd,cfg->smartthres)){
      PrintOut(LOG_INFO,"Device: %s, Read SMART Values and/or Thresholds Failed\n",name);
      retainsmartdata=cfg->usagefailed=cfg->prefail=cfg->usage=0;
      cfg->tempdiff = cfg->tempinfo = cfg->tempcrit = 0;
      cfg->pending=DONT_MONITOR_UNC;
    }
    
    // see if the necessary Attribute is there to monitor offline or
    // current pending sectors or temperature
    TranslatePending(cfg->pending, &currentpending, &offlinepending);
    
    if (currentpending && ATAReturnAttributeRawValue(currentpending, cfg->smartval)<0) {
      PrintOut(LOG_INFO,"Device: %s, can't monitor Current Pending Sector count - no Attribute %d\n",
	       name, (int)currentpending);
      cfg->pending &= 0xff00;
      cfg->pending |= CUR_UNC_DEFAULT;
    }
    
    if (offlinepending && ATAReturnAttributeRawValue(offlinepending, cfg->smartval)<0) {
      PrintOut(LOG_INFO,"Device: %s, can't monitor Offline Uncorrectable Sector count  - no Attribute %d\n",
	       name, (int)offlinepending);
      cfg->pending &= 0x00ff;
      cfg->pending |= OFF_UNC_DEFAULT<<8;
    }

    if (   (cfg->tempdiff || cfg->tempinfo || cfg->tempcrit)
        && !ATAReturnTemperatureValue(cfg->smartval, cfg->attributedefs)) {
      PrintOut(LOG_CRIT, "Device: %s, can't monitor Temperature, ignoring -W Directive\n", name);
      cfg->tempdiff = cfg->tempinfo = cfg->tempcrit = 0;
    }
  }
  
  // enable/disable automatic on-line testing
  if (cfg->autoofflinetest){
    // is this an enable or disable request?
    const char *what=(cfg->autoofflinetest==1)?"disable":"enable";
    if (!cfg->smartval)
      PrintOut(LOG_INFO,"Device: %s, could not %s SMART Automatic Offline Testing.\n",name, what);
    else {
      // if command appears unsupported, issue a warning...
      if (!isSupportAutomaticTimer(cfg->smartval))
        PrintOut(LOG_INFO,"Device: %s, SMART Automatic Offline Testing unsupported...\n",name);
      // ... but then try anyway
      if ((cfg->autoofflinetest==1)?ataDisableAutoOffline(fd):ataEnableAutoOffline(fd))
        PrintOut(LOG_INFO,"Device: %s, %s SMART Automatic Offline Testing failed.\n", name, what);
      else
        PrintOut(LOG_INFO,"Device: %s, %sd SMART Automatic Offline Testing.\n", name, what);
    }
  }
  
  // capability check: self-test-log
  if (cfg->selftest){
    int retval;
    
    // start with service disabled, and re-enable it if all works OK
    cfg->selftest=0;
    cfg->selflogcount=0;
    cfg->selfloghour=0;

    if (!cfg->smartval)
      PrintOut(LOG_INFO, "Device: %s, no SMART Self-Test log (SMART READ DATA failed); disabling -l selftest\n", name);
    else if (!cfg->permissive && !isSmartTestLogCapable(cfg->smartval, &drive))
      PrintOut(LOG_INFO, "Device: %s, appears to lack SMART Self-Test log; disabling -l selftest (override with -T permissive Directive)\n", name);
    else if ((retval=SelfTestErrorCount(fd, name))<0)
      PrintOut(LOG_INFO, "Device: %s, no SMART Self-Test log; remove -l selftest Directive from smartd.conf\n", name);
    else {
      cfg->selftest=1;
      cfg->selflogcount=SELFTEST_ERRORCOUNT(retval);
      cfg->selfloghour =SELFTEST_ERRORHOURS(retval);
    }
  }