smartd.h

   1 /*
   2  * smartd.h
   3  *
   4  * Home page of code is: http://smartmontools.sourceforge.net
   5  *
   6  * Copyright (C) 2002-6 Bruce Allen <smartmontools-support@lists.sourceforge.net>
   7  * Copyright (C) 2000 Michael Cornwell <cornwell@acm.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2, or (at your option)
  12  * any later version.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * (for example COPYING); if not, write to the Free
  16  * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  17  *
  18  * This code was originally developed as a Senior Thesis by Michael Cornwell
  19  * at the Concurrent Systems Laboratory (now part of the Storage Systems
  20  * Research Center), Jack Baskin School of Engineering, University of
  21  * California, Santa Cruz. http://ssrc.soe.ucsc.edu/
  22  *
  23  */
  24
  25 #ifndef SMARTD_H_
  26 #define SMARTD_H_
  27
  28 // Needed since some structure definitions below require POSIX
  29 // extended regular expressions.
  30 #include <sys/types.h>
  31 #include <regex.h>
  32
  33
  34 #ifndef SMARTD_H_CVSID
  35 #define SMARTD_H_CVSID "$Id: smartd.h,v 1.84 2006/11/11 17:50:50 ballen4705 Exp $\n"
  36 #endif
  37
  38 // Configuration file
  39 #define CONFIGFILENAME "smartd.conf"
  40
  41 // Scan directive for configuration file
  42 #define SCANDIRECTIVE "DEVICESCAN"
  43
  44 // maximum line length in configuration file
  45 #define MAXLINELEN 256
  46
  47 // maximum length of a continued line in configuration file
  48 #define MAXCONTLINE 1023
  49
  50 // default for how often SMART status is checked, in seconds
  51 #define CHECKTIME 1800
  52
  53 /* Boolean Values */
  54 #define TRUE 0x01
  55 #define FALSE 0x00
  56
  57 // Number of monitoring flags per Attribute and offsets.  See
  58 // monitorattflags below.
  59 #define NMONITOR 4
  60 #define MONITOR_FAILUSE   0
  61 #define MONITOR_IGNORE    1
  62 #define MONITOR_RAWPRINT  2
  63 #define MONITOR_RAW       3
  64
  65
  66 // Number of allowed mail message types
  67 #define SMARTD_NMAIL 13
  68
  69 typedef struct mailinfo_s {
  70   int logged;// number of times an email has been sent
  71   time_t firstsent;// time first email was sent, as defined by time(2)
  72   time_t lastsent; // time last email was sent, as defined by time(2)
  73 } mailinfo;
  74
  75 // If user has requested email warning messages, then this structure
  76 // stores the information about them, and track type/date of email
  77 // messages.
  78 typedef struct maildata_s {
  79   mailinfo maillog[SMARTD_NMAIL];         // log info on when mail sent
  80   char *emailcmdline;                     // script to execute
  81   char *address;                          // email address, or null
  82   unsigned char emailfreq;                // Emails once (1) daily (2) diminishing (3)
  83   unsigned char emailtest;                // Send test email?
  84 } maildata;
  85
  86 // If user has requested automatic testing, then this structure stores
  87 // their regular expression pattern, the compiled form of that regex,
  88 // and information about the disk capabilities and when the last text
  89 // took place
  90
  91 typedef struct testinfo_s {
  92   char *regex;                    // text form of regex
  93   regex_t cregex;                 // compiled form of regex
  94   unsigned short hour;            // 1+hour of year when last scheduled self-test done
  95   char testtype;                  // type of test done at hour indicated just above
  96   signed char not_cap_offline;    // 0==unknown OR capable of offline, 1==not capable
  97   signed char not_cap_conveyance;
  98   signed char not_cap_short;
  99   signed char not_cap_long;
 100 } testinfo;
 101
 102
 103 // cfgfile is the main data structure of smartd. It is used in two
 104 // ways.  First, to store a list of devices/options given in the
 105 // configuration smartd.conf or constructed with DEVICESCAN.  And
 106 // second, to point to or provide all persistent storage needed to
 107 // track a device, if registered either as SCSI or ATA.
 108 //
 109 // After parsing the config file, each valid entry has a cfgfile data
 110 // structure allocated in memory for it.  In parsing the configuration
 111 // file, some storage space may be needed, of indeterminate length,
 112 // for example for the device name.  When this happens, memory should
 113 // be allocated and then pointed to from within the corresponding
 114 // cfgfile structure.
 115
 116 // After parsing the configuration file, each device is then checked
 117 // to see if it can be monitored (this process is called "registering
 118 // the device".  This is done in [scsi|ata]devicescan, which is called
 119 // exactly once, after the configuration file has been parsed and
 120 // cfgfile data structures have been created for each of its entries.
 121 //
 122 // If a device can not be monitored, the memory for its cfgfile data
 123 // structure should be freed by calling rmconfigentry(cfgfile *). In
 124 // this case, we say that this device "was not registered".  All
 125 // memory associated with that particular cfgfile structure is thus
 126 // freed.
 127 //
 128 // The remaining devices are polled in a timing look, where
 129 // [ata|scsi]CheckDevice looks at each entry in turn.
 130 //
 131 // If you want to add small amounts of "private" data on a per-device
 132 // basis, just make a new field in cfgfile.  This is guaranteed zero
 133 // on startup (when ata|scsi]scsidevicescan(cfgfile *cfg) is first
 134 // called with a pointer to cfgfile.
 135 //
 136 // If you need *substantial* persistent data space for a device
 137 // (dozens or hundreds of bytes) please add a pointer field to
 138 // cfgfile.  As before, this is guaranteed NULL when
 139 // ata|scsi]scsidevicescan(cfgfile *cfg) is called. Allocate space for
 140 // it in scsidevicescan or atadevicescan, if needed, and deallocate
 141 // the space in rmconfigentry(cfgfile *cfg). Be sure to make the
 142 // pointer NULL unless it points to an area of the heap that can be
 143 // deallocated with free().  In other words, a non-NULL pointer in
 144 // cfgfile means "this points to data space that should be freed if I
 145 // stop monitoring this device." If you don't need the space anymore,
 146 // please call free() and then SET THE POINTER IN cfgfile TO NULL.
 147 //
 148 // Note that we allocate one cfgfile structure per device.  This is
 149 // why substantial persisent data storage should only be pointed to
 150 // from within cfgfile, not kept within cfgfile itself - it saves
 151 // memory for those devices that don't need that type of persistent
 152 // data.
 153 //
 154 // In general, the capabilities of devices should be checked at
 155 // registration time within atadevicescan() and scsidevicescan(), and
 156 // then noted within *cfg.  So if device lacks some capability, this
 157 // should be visible within *cfg after returning from
 158 // [ata|scsi]devicescan.
 159 //
 160 // Devices are then checked, once per polling interval, within
 161 // ataCheckDevice() and scsiCheckDevice().  These should only check
 162 // the capabilities that devices already are known to have (as noted
 163 // within *cfg).
 164
 165 typedef struct configfile_s {
 166   // FIRST SET OF ENTRIES CORRESPOND TO WHAT THE USER PUT IN THE
 167   // CONFIG FILE.  SOME ENTRIES MAY BE MODIFIED WHEN A DEVICE IS
 168   // REGISTERED AND WE LEARN ITS CAPABILITIES.
 169   int lineno;                             // Line number of entry in file
 170   char *name;                             // Device name (+ optional [3ware_disk_XX])
 171   unsigned char controller_explicit;      // Controller (device) type has been specified explicitly
 172   unsigned char controller_type;          // Controller type, ATA/SCSI/SAT/3Ware/(more to come)
 173   unsigned char controller_port;          // 1 + (disk number in controller). 0 means controller only handles one disk.
 174   unsigned char hpt_data[3];              // combined controller/channle/pmport for highpoint rocketraid controller
 175   unsigned char satpassthrulen;           // length of SAT ata pass through scsi commands (12, 16 or 0 (platform choice))
 176   char smartcheck;                        // Check SMART status
 177   char usagefailed;                       // Check for failed Usage Attributes
 178   char prefail;                           // Track changes in Prefail Attributes
 179   char usage;                             // Track changes in Usage Attributes
 180   char selftest;                          // Monitor number of selftest errors
 181   char errorlog;                          // Monitor number of ATA errors
 182   char permissive;                        // Ignore failed SMART commands
 183   char autosave;                          // 1=disable, 2=enable Autosave Attributes
 184   char autoofflinetest;                   // 1=disable, 2=enable Auto Offline Test
 185   unsigned char fixfirmwarebug;           // Fix firmware bug
 186   char ignorepresets;                     // Ignore database of -v options
 187   char showpresets;                       // Show database entry for this device
 188   char removable;                         // Device may disappear (not be present)
 189   char powermode;                         // skip check, if disk in idle or standby mode
 190   char powerquiet;                        // skip powermode 'skipping checks' message
 191   unsigned char tempdiff;                 // Track Temperature changes >= this limit
 192   unsigned char tempinfo, tempcrit;       // Track Temperatures >= these limits as LOG_INFO, LOG_CRIT+mail
 193   unsigned char tempmin, tempmax;         // Min/Max Temperatures
 194   unsigned char selflogcount;             // total number of self-test errors
 195   unsigned short selfloghour;             // lifetime hours of last self-test error
 196   testinfo *testdata;                     // Pointer to data on scheduled testing
 197   unsigned short pending;                 // lower 8 bits: ID of current pending sector count
 198                                           // upper 8 bits: ID of offline pending sector count
 199
 200   // THE NEXT SET OF ENTRIES ALSO TRACK DEVICE STATE AND ARE DYNAMIC
 201   maildata *mailwarn;                     // non-NULL: info about sending mail or executing script
 202   unsigned char temperature;              // last recorded Temperature (in Celsius)
 203   unsigned char tempmininc;               // #checks where Min Temperature is increased after powerup
 204   int powerskipcnt;                       // Number of checks skipped due to idle or standby mode
 205
 206   // SCSI ONLY
 207   unsigned char SmartPageSupported;       // has log sense IE page (0x2f)
 208   unsigned char TempPageSupported;        // has log sense temperature page (0xd)
 209   unsigned char SuppressReport;           // minimize nuisance reports
 210   unsigned char modese_len;               // mode sense/select cmd len: 0 (don't
 211                                           // know yet) 6 or 10
 212   unsigned char notused2[3];              // for packing alignment
 213
 214   // ATA ONLY FROM HERE ON TO THE END
 215   int ataerrorcount;                      // Total number of ATA errors
 216
 217   // following NMONITOR items each point to 32 bytes, in the form of
 218   // 32x8=256 single bit flags
 219   // valid attribute numbers are from 1 <= x <= 255
 220   // monitorattflags+0  set: ignore failure for a usage attribute
 221   // monitorattflats+32 set: don't track attribute
 222   // monitorattflags+64 set: print raw value when tracking
 223   // monitorattflags+96 set: track changes in raw value
 224   unsigned char *monitorattflags;
 225
 226   // NULL UNLESS (1) STORAGE IS ALLOCATED WHEN CONFIG FILE SCANNED
 227   // (SET BY USER) or (2) IT IS SET WHEN DRIVE IS AUTOMATICALLY
 228   // RECOGNIZED IN DATABASE (WHEN DRIVE IS REGISTERED)
 229   unsigned char *attributedefs;            // -v options, see end of extern.h for def
 230
 231   // ATA ONLY - SAVE SMART DATA. NULL POINTERS UNLESS NEEDED.  IF
 232   // NEEDED, ALLOCATED WHEN DEVICE REGISTERED.
 233   struct ata_smart_values *smartval;       // Pointer to SMART data
 234   struct ata_smart_thresholds_pvt *smartthres; // Pointer to SMART thresholds
 235
 236 } cfgfile;
 237
 238
 239 typedef struct changedattribute_s {
 240   unsigned char newval;
 241   unsigned char oldval;
 242   unsigned char id;
 243   unsigned char prefail;
 244   unsigned char sameraw;
 245 } changedattribute_t;
 246
 247 // Declare our own printing functions. Doing this provides error
 248 // messages if the argument number/types don't match the format.
 249 #ifndef __GNUC__
 250 #define __attribute__(x)      /* nothing */
 251 #endif
 252 void PrintOut(int priority, const char *fmt, ...) __attribute__ ((format(printf, 2, 3)));
 253
 254 void PrintAndMail(cfgfile *cfg, int which, int priority, char *fmt, ...) __attribute__ ((format(printf, 4, 5)));
 255
 256 /* Debugging notes: to check for memory allocation/deallocation problems, use:
 257
 258 export LD_PRELOAD=libnjamd.so;
 259 export NJAMD_PROT=strict;
 260 export NJAMD_CHK_FREE=error;
 261 export NJAMD_DUMP_LEAKS_ON_EXIT=num;
 262 export NJAMD_DUMP_LEAKS_ON_EXIT=3;
 263 export NJAMD_TRACE_LIBS=1
 264
 265 */
 266
 267 // Number of seconds to allow for registering a SCSI device. If this
 268 // time expires without sucess or failure, then treat it as failure.
 269 // Set to 0 to eliminate this timeout feature from the code
 270 // (equivalent to an infinite timeout interval).
 271 #define SCSITIMEOUT 0
 272
 273 // This is for solaris, where signal() resets the handler to SIG_DFL
 274 // after the first signal is caught.
 275 #ifdef HAVE_SIGSET
 276 #define SIGNALFN sigset
 277 #else
 278 #define SIGNALFN signal
 279 #endif
 280
 281 #endif
 282
 283 #define SELFTEST_ERRORCOUNT(x) (x & 0xff)
 284 #define SELFTEST_ERRORHOURS(x) ((x >> 8) & 0xffff)
 285
 286 // cfg->pending is a 16 bit unsigned quantity.  If the least
 287 // significant 8 bits are zero, this means monitor Attribute
 288 // CUR_UNC_DEFAULT's raw value.  If they are CUR_UNC_DEFAULT, this
 289 // means DON'T MONITOR.  If the most significant 8 bits are zero, this
 290 // means monitor Attribute OFF_UNC_DEFAULT's raw value.  If they are
 291 // OFF_UNC_DEFAULT, this means DON'T MONITOR.
 292 #define OFF_UNC_DEFAULT 198
 293 #define CUR_UNC_DEFAULT 197
 294
 295 #define CURR_PEND(x) (x & 0xff)
 296 #define OFF_PEND(x) ((x >> 8) & 0xff)
 297
 298 // if cfg->pending has this value, dont' monitor
 299 #define DONT_MONITOR_UNC (256*OFF_UNC_DEFAULT+CUR_UNC_DEFAULT)