/*
* Home page of code is: http://smartmontools.sourceforge.net
*
- * Copyright (C) 2002-9 Bruce Allen <smartmontools-support@lists.sourceforge.net>
- * Copyright (C) 2000 Michael Cornwell <cornwell@acm.org>
- * Copyright (C) 2008 Oliver Bock <brevilo@users.sourceforge.net>
- * Copyright (C) 2008-9 Christian Franke <smartmontools-support@lists.sourceforge.net>
+ * Copyright (C) 2002-11 Bruce Allen <smartmontools-support@lists.sourceforge.net>
+ * Copyright (C) 2000 Michael Cornwell <cornwell@acm.org>
+ * Copyright (C) 2008 Oliver Bock <brevilo@users.sourceforge.net>
+ * Copyright (C) 2008-14 Christian Franke <smartmontools-support@lists.sourceforge.net>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
*
*/
-#ifndef _GNU_SOURCE
-// TODO: Why is this define necessary?
-#define _GNU_SOURCE
-#endif
+#include "config.h"
+#include "int64.h"
// unconditionally included files
#include <stdio.h>
#include <vector>
#include <algorithm> // std::replace()
-// see which system files to conditionally include
-#include "config.h"
-
// conditionally included files
#ifndef _WIN32
#include <sys/wait.h>
#endif // _WIN32
#ifdef __CYGWIN__
-// From <windows.h>:
-// BOOL WINAPI FreeConsole(void);
-extern "C" int __stdcall FreeConsole(void);
#include <io.h> // setmode()
#endif // __CYGWIN__
+#ifdef HAVE_LIBCAP_NG
+#include <cap-ng.h>
+#endif // LIBCAP_NG
+
// locally included files
-#include "int64.h"
#include "atacmds.h"
#include "dev_interface.h"
-#include "extern.h"
#include "knowndrives.h"
#include "scsicmds.h"
#include "utility.h"
#endif
#ifdef _WIN32
-#include "hostname_win32.h" // gethost/domainname()
-#define HAVE_GETHOSTNAME 1
-#define HAVE_GETDOMAINNAME 1
// fork()/signal()/initd simulation for native Windows
#include "daemon_win32.h" // daemon_main/detach/signal()
#undef SIGNALFN
#define SIGQUIT SIGBREAK
#define SIGQUIT_KEYNAME "CONTROL-Break"
#else // _WIN32
-#ifdef __CYGWIN__
-// 2x CONTROL-C simulates missing SIGQUIT via keyboard
-#define SIGQUIT_KEYNAME "2x CONTROL-C"
-#else // __CYGWIN__
#define SIGQUIT_KEYNAME "CONTROL-\\"
-#endif // __CYGWIN__
#endif // _WIN32
#if defined (__SVR4) && defined (__sun)
extern "C" int getdomainname(char *, int); // no declaration in header files!
#endif
-#define ARGUSED(x) ((void)(x))
-
-const char * smartd_cpp_cvsid = "$Id: smartd.cpp 2984 2009-11-14 22:46:31Z chrfranke $"
- CONFIG_H_CVSID EXTERN_H_CVSID;
-
-extern const char *reportbug;
-
-extern unsigned char debugmode;
+const char * smartd_cpp_cvsid = "$Id: smartd.cpp 3948 2014-07-13 16:53:30Z chrfranke $"
+ CONFIG_H_CVSID;
// smartd exit codes
#define EXIT_BADCMD 1 // command line did not parse
#define EXIT_SIGNAL 254 // abort on signal
+
+// command-line: 1=debug mode, 2=print presets
+static unsigned char debugmode = 0;
+
// command-line: how long to sleep between checks
#define CHECKTIME 1800
static int checktime=CHECKTIME;
;
// configuration file name
-#define CONFIGFILENAME "smartd.conf"
-
-#ifndef _WIN32
-static const char *configfile = SMARTMONTOOLS_SYSCONFDIR "/" CONFIGFILENAME ;
-#else
-static const char *configfile = "./" CONFIGFILENAME ;
-#endif
+static const char * configfile;
// configuration file "name" if read from stdin
static const char * const configfile_stdin = "<stdin>";
// path of alternate configuration file
static std::string configfile_alt;
+// warning script file
+static std::string warning_script;
+
// command-line: when should we exit?
static int quit=0;
static bool do_fork=true;
#endif
-// used for control of printing, passing arguments to atacmds.c
-smartmonctrl *con=NULL;
+#ifdef HAVE_LIBCAP_NG
+// command-line: enable capabilities?
+static bool enable_capabilities = false;
+#endif
+
+// TODO: This smartctl only variable is also used in os_win32.cpp
+unsigned char failuretest_permissive = 0;
// set to one if we catch a USR1 (check devices now)
-volatile int caughtsigUSR1=0;
+static volatile int caughtsigUSR1=0;
#ifdef _WIN32
// set to one if we catch a USR2 (toggle debug mode)
-volatile int caughtsigUSR2=0;
+static volatile int caughtsigUSR2=0;
#endif
// set to one if we catch a HUP (reload config file). In debug mode,
// set to two, if we catch INT (also reload config file).
-volatile int caughtsigHUP=0;
+static volatile int caughtsigHUP=0;
// set to signal value if we catch INT, QUIT, or TERM
-volatile int caughtsigEXIT=0;
+static volatile int caughtsigEXIT=0;
+
+// This function prints either to stdout or to the syslog as needed.
+static void PrintOut(int priority, const char *fmt, ...)
+ __attribute_format_printf(2, 3);
// Attribute monitoring flags.
// See monitor_attr_flags below.
struct dev_config
{
int lineno; // Line number of entry in file
- std::string name; // Device name
+ std::string name; // Device name (with optional extra info)
+ std::string dev_name; // Device name (plain, for SMARTD_DEVICE variable)
std::string dev_type; // Device type argument from -d directive, empty if none
+ std::string dev_idinfo; // Device identify info for warning emails
std::string state_file; // Path of the persistent state file, empty if none
std::string attrlog_file; // Path of the persistent attrlog file, empty if none
+ bool ignore; // Ignore this entry
bool smartcheck; // Check SMART status
bool usagefailed; // Check for failed Usage Attributes
bool prefail; // Track changes in Prefail Attributes
bool usage; // Track changes in Usage Attributes
bool selftest; // Monitor number of selftest errors
bool errorlog; // Monitor number of ATA errors
+ bool xerrorlog; // Monitor number of ATA errors (Extended Comprehensive error log)
+ bool offlinests; // Monitor changes in offline data collection status
+ bool offlinests_ns; // Disable auto standby if in progress
+ bool selfteststs; // Monitor changes in self-test execution status
+ bool selfteststs_ns; // Disable auto standby if in progress
bool permissive; // Ignore failed SMART commands
char autosave; // 1=disable, 2=enable Autosave Attributes
char autoofflinetest; // 1=disable, 2=enable Auto Offline Test
- unsigned char fix_firmwarebug; // FIX_*, see atacmds.h
+ firmwarebug_defs firmwarebugs; // -F directives from drivedb or smartd.conf
bool ignorepresets; // Ignore database of -v options
bool showpresets; // Show database entry for this device
bool removable; // Device may disappear (not be present)
bool emailtest; // Send test email?
// ATA ONLY
+ int dev_rpm; // rotation rate, 0 = unknown, 1 = SSD, >1 = HDD
+ int set_aam; // disable(-1), enable(1..255->0..254) Automatic Acoustic Management
+ int set_apm; // disable(-1), enable(2..255->1..254) Advanced Power Management
+ int set_lookahead; // disable(-1), enable(1) read look-ahead
+ int set_standby; // set(1..255->0..254) standby timer
+ bool set_security_freeze; // Freeze ATA security
+ int set_wcache; // disable(-1), enable(1) write cache
+
+ bool sct_erc_set; // set SCT ERC to:
+ unsigned short sct_erc_readtime; // ERC read time (deciseconds)
+ unsigned short sct_erc_writetime; // ERC write time (deciseconds)
+
unsigned char curr_pending_id; // ID of current pending sector count, 0 if none
unsigned char offl_pending_id; // ID of offline uncorrectable sector count, 0 if none
bool curr_pending_incr, offl_pending_incr; // True if current/offline pending values increase
dev_config::dev_config()
: lineno(0),
+ ignore(false),
smartcheck(false),
usagefailed(false),
prefail(false),
usage(false),
selftest(false),
errorlog(false),
+ xerrorlog(false),
+ offlinests(false), offlinests_ns(false),
+ selfteststs(false), selfteststs_ns(false),
permissive(false),
autosave(0),
autoofflinetest(0),
- fix_firmwarebug(FIX_NOTSPECIFIED),
ignorepresets(false),
showpresets(false),
removable(false),
tempinfo(0), tempcrit(0),
emailfreq(0),
emailtest(false),
+ dev_rpm(0),
+ set_aam(0), set_apm(0),
+ set_lookahead(0),
+ set_standby(0),
+ set_security_freeze(false),
+ set_wcache(0),
+ sct_erc_set(false),
+ sct_erc_readtime(0), sct_erc_writetime(0),
curr_pending_id(0), offl_pending_id(0),
curr_pending_incr(false), offl_pending_incr(false),
curr_pending_set(false), offl_pending_set(false)
// Number of allowed mail message types
-const int SMARTD_NMAIL = 13;
+static const int SMARTD_NMAIL = 13;
// Type for '-M test' mails (state not persistent)
-const int MAILTYPE_TEST = 0;
+static const int MAILTYPE_TEST = 0;
// TODO: Add const or enum for all mail types.
struct mailinfo {
time_t scheduled_test_next_check; // Time of next check for scheduled self-tests
+ uint64_t selective_test_last_start; // Start LBA of last scheduled selective self-test
+ uint64_t selective_test_last_end; // End LBA of last scheduled selective self-test
+
mailinfo maillog[SMARTD_NMAIL]; // log info on when mail sent
// ATA ONLY
unsigned char val;
unsigned char worst; // Byte needed for 'raw64' attribute only.
uint64_t raw;
+ unsigned char resvd;
- ata_attribute() : id(0), val(0), worst(0), raw(0) { }
+ ata_attribute() : id(0), val(0), worst(0), raw(0), resvd(0) { }
};
ata_attribute ata_attributes[NUMBER_ATA_SMART_ATTRIBUTES];
+
+ // SCSI ONLY
+
+ struct scsi_error_counter {
+ struct scsiErrorCounter errCounter;
+ unsigned char found;
+ scsi_error_counter() : found(0) { }
+ };
+ scsi_error_counter scsi_error_counters[3];
+
+ struct scsi_nonmedium_error {
+ struct scsiNonMediumError nme;
+ unsigned char found;
+ scsi_nonmedium_error() : found(0) { }
+ };
+ scsi_nonmedium_error scsi_nonmedium_error;
persistent_dev_state();
};
selflogcount(0),
selfloghour(0),
scheduled_test_next_check(0),
+ selective_test_last_start(0),
+ selective_test_last_end(0),
ataerrorcount(0)
{
}
// SCSI ONLY
unsigned char SmartPageSupported; // has log sense IE page (0x2f)
unsigned char TempPageSupported; // has log sense temperature page (0xd)
+ unsigned char ReadECounterPageSupported;
+ unsigned char WriteECounterPageSupported;
+ unsigned char VerifyECounterPageSupported;
+ unsigned char NonMediumErrorPageSupported;
unsigned char SuppressReport; // minimize nuisance reports
unsigned char modese_len; // mode sense/select cmd len: 0 (don't
// know yet) 6 or 10
-
// ATA ONLY
- uint64_t num_sectors; // Number of sectors (for selective self-test only)
+ uint64_t num_sectors; // Number of sectors
ata_smart_values smartval; // SMART data
ata_smart_thresholds_pvt smartthres; // SMART thresholds
+ bool offline_started; // true if offline data collection was started
+ bool selftest_started; // true if self-test was started
temp_dev_state();
};
powerskipcnt(0),
SmartPageSupported(false),
TempPageSupported(false),
+ ReadECounterPageSupported(false),
+ WriteECounterPageSupported(false),
+ VerifyECounterPageSupported(false),
+ NonMediumErrorPageSupported(false),
SuppressReport(false),
modese_len(0),
- num_sectors(0)
+ num_sectors(0),
+ offline_started(false),
+ selftest_started(false)
{
memset(&smartval, 0, sizeof(smartval));
memset(&smartthres, 0, sizeof(smartthres));
| ((uint64_t)ta.raw[3] << 24)
| ((uint64_t)ta.raw[4] << 32)
| ((uint64_t)ta.raw[5] << 40);
+ pa.resvd = ta.reserv;
}
}
ta.raw[3] = (unsigned char)(pa.raw >> 24);
ta.raw[4] = (unsigned char)(pa.raw >> 32);
ta.raw[5] = (unsigned char)(pa.raw >> 40);
+ ta.reserv = pa.resvd;
}
}
// Parse a line from a state file.
static bool parse_dev_state_line(const char * line, persistent_dev_state & state)
{
- static regular_expression regex(
+ static const regular_expression regex(
"^ *"
"((temperature-min)" // (1 (2)
"|(temperature-max)" // (3)
"|(self-test-errors)" // (4)
"|(self-test-last-err-hour)" // (5)
"|(scheduled-test-next-check)" // (6)
- "|(ata-error-count)" // (7)
- "|(mail\\.([0-9]+)\\." // (8 (9)
- "((count)" // (10 (11)
- "|(first-sent-time)" // (12)
- "|(last-sent-time)" // (13)
- ")" // 10)
- ")" // 8)
- "|(ata-smart-attribute\\.([0-9]+)\\." // (14 (15)
- "((id)" // (16 (17)
- "|(val)" // (18)
- "|(worst)" // (19)
- "|(raw)" // (20)
- ")" // 16)
- ")" // 14)
+ "|(selective-test-last-start)" // (7)
+ "|(selective-test-last-end)" // (8)
+ "|(ata-error-count)" // (9)
+ "|(mail\\.([0-9]+)\\." // (10 (11)
+ "((count)" // (12 (13)
+ "|(first-sent-time)" // (14)
+ "|(last-sent-time)" // (15)
+ ")" // 12)
+ ")" // 10)
+ "|(ata-smart-attribute\\.([0-9]+)\\." // (16 (17)
+ "((id)" // (18 (19)
+ "|(val)" // (20)
+ "|(worst)" // (21)
+ "|(raw)" // (22)
+ "|(resvd)" // (23)
+ ")" // 18)
+ ")" // 16)
")" // 1)
- " *= *([0-9]+)[ \n]*$", // (21)
+ " *= *([0-9]+)[ \n]*$", // (24)
REG_EXTENDED
);
- if (regex.empty())
- throw std::logic_error("parse_dev_state_line: invalid regex");
- const int nmatch = 1+21;
+ const int nmatch = 1+24;
regmatch_t match[nmatch];
if (!regex.execute(line, nmatch, match))
return false;
state.selfloghour = (unsigned short)val;
else if (match[++m].rm_so >= 0)
state.scheduled_test_next_check = (time_t)val;
+ else if (match[++m].rm_so >= 0)
+ state.selective_test_last_start = val;
+ else if (match[++m].rm_so >= 0)
+ state.selective_test_last_end = val;
else if (match[++m].rm_so >= 0)
state.ataerrorcount = (int)val;
else if (match[m+=2].rm_so >= 0) {
state.ata_attributes[i].worst = (unsigned char)val;
else if (match[++m].rm_so >= 0)
state.ata_attributes[i].raw = val;
+ else if (match[++m].rm_so >= 0)
+ state.ata_attributes[i].resvd = (unsigned char)val;
else
return false;
}
setmode(fileno(f), O_TEXT); // Allow files with \r\n
#endif
+ persistent_dev_state new_state;
int good = 0, bad = 0;
char line[256];
while (fgets(line, sizeof(line), f)) {
const char * s = line + strspn(line, " \t");
if (!*s || *s == '#')
continue;
- if (!parse_dev_state_line(line, state))
+ if (!parse_dev_state_line(line, new_state))
bad++;
else
good++;
}
pout("%s: %d invalid line(s) ignored\n", path, bad);
}
+
+ // This sets the values missing in the file to 0.
+ state = new_state;
return true;
}
static void write_dev_state_line(FILE * f, const char * name, uint64_t val)
{
if (val)
- fprintf(f, "%s = %"PRIu64"\n", name, val);
+ fprintf(f, "%s = %" PRIu64 "\n", name, val);
}
static void write_dev_state_line(FILE * f, const char * name1, int id, const char * name2, uint64_t val)
{
if (val)
- fprintf(f, "%s.%d.%s = %"PRIu64"\n", name1, id, name2, val);
+ fprintf(f, "%s.%d.%s = %" PRIu64 "\n", name1, id, name2, val);
}
// Write a state file
write_dev_state_line(f, "self-test-errors", state.selflogcount);
write_dev_state_line(f, "self-test-last-err-hour", state.selfloghour);
write_dev_state_line(f, "scheduled-test-next-check", state.scheduled_test_next_check);
+ write_dev_state_line(f, "selective-test-last-start", state.selective_test_last_start);
+ write_dev_state_line(f, "selective-test-last-end", state.selective_test_last_end);
int i;
for (i = 0; i < SMARTD_NMAIL; i++) {
write_dev_state_line(f, "ata-smart-attribute", i, "val", pa.val);
write_dev_state_line(f, "ata-smart-attribute", i, "worst", pa.worst);
write_dev_state_line(f, "ata-smart-attribute", i, "raw", pa.raw);
+ write_dev_state_line(f, "ata-smart-attribute", i, "resvd", pa.resvd);
}
return true;
}
// Write to the attrlog file
-static bool write_dev_attrlog(const char * path, const persistent_dev_state & state)
+static bool write_dev_attrlog(const char * path, const dev_state & state)
{
stdio_file f(path, "a");
if (!f) {
return false;
}
- // ATA ONLY
+
time_t now = time(0);
struct tm * tms = gmtime(&now);
fprintf(f, "%d-%02d-%02d %02d:%02d:%02d;",
1900+tms->tm_year, 1+tms->tm_mon, tms->tm_mday,
tms->tm_hour, tms->tm_min, tms->tm_sec);
+ // ATA ONLY
for (int i = 0; i < NUMBER_ATA_SMART_ATTRIBUTES; i++) {
const persistent_dev_state::ata_attribute & pa = state.ata_attributes[i];
if (!pa.id)
continue;
- fprintf(f, "\t%d;%d;%"PRIu64";", pa.id, pa.val, pa.raw);
+ fprintf(f, "\t%d;%d;%" PRIu64 ";", pa.id, pa.val, pa.raw);
}
+ // SCSI ONLY
+ const struct scsiErrorCounter * ecp;
+ const char * pageNames[3] = {"read", "write", "verify"};
+ for (int k = 0; k < 3; ++k) {
+ if ( !state.scsi_error_counters[k].found ) continue;
+ ecp = &state.scsi_error_counters[k].errCounter;
+ fprintf(f, "\t%s-corr-by-ecc-fast;%" PRIu64 ";"
+ "\t%s-corr-by-ecc-delayed;%" PRIu64 ";"
+ "\t%s-corr-by-retry;%" PRIu64 ";"
+ "\t%s-total-err-corrected;%" PRIu64 ";"
+ "\t%s-corr-algorithm-invocations;%" PRIu64 ";"
+ "\t%s-gb-processed;%.3f;"
+ "\t%s-total-unc-errors;%" PRIu64 ";",
+ pageNames[k], ecp->counter[0],
+ pageNames[k], ecp->counter[1],
+ pageNames[k], ecp->counter[2],
+ pageNames[k], ecp->counter[3],
+ pageNames[k], ecp->counter[4],
+ pageNames[k], (ecp->counter[5] / 1000000000.0),
+ pageNames[k], ecp->counter[6]);
+ }
+ if(state.scsi_nonmedium_error.found && state.scsi_nonmedium_error.nme.gotPC0) {
+ fprintf(f, "\tnon-medium-errors;%" PRIu64 ";", state.scsi_nonmedium_error.nme.counterPC0);
+ }
+ // write SCSI current temperature if it is monitored
+ if(state.TempPageSupported && state.temperature)
+ fprintf(f, "\ttemperature;%d;", state.temperature);
+ // end of line
fprintf(f, "\n");
-
return true;
}
}
// remove the PID file
-void RemovePidFile(){
+static void RemovePidFile()
+{
if (!pid_file.empty()) {
if (unlink(pid_file.c_str()))
PrintOut(LOG_CRIT,"Can't unlink PID file %s (%s).\n",
extern "C" { // signal handlers require C-linkage
// Note if we catch a SIGUSR1
-void USR1handler(int sig){
+static void USR1handler(int sig)
+{
if (SIGUSR1==sig)
caughtsigUSR1=1;
return;
#ifdef _WIN32
// Note if we catch a SIGUSR2
-void USR2handler(int sig){
+static void USR2handler(int sig)
+{
if (SIGUSR2==sig)
caughtsigUSR2=1;
return;
#endif
// Note if we catch a HUP (or INT in debug mode)
-void HUPhandler(int sig){
+static void HUPhandler(int sig)
+{
if (sig==SIGHUP)
caughtsigHUP=1;
else
}
// signal handler for TERM, QUIT, and INT (if not in debug mode)
-void sighandler(int sig){
+static void sighandler(int sig)
+{
if (!caughtsigEXIT)
caughtsigEXIT=sig;
return;
return status;
}
-#define ENVLENGTH 1024
-
// a replacement for setenv() which is not available on all platforms.
// Note that the string passed to putenv must not be freed or made
// invalid, since a pointer to it is kept by putenv(). This means that
// it must either be a static buffer or allocated off the heap. The
-// string can be freed if the environment variable is redefined or
-// deleted via another call to putenv(). So we keep these on the stack
-// as long as the popen() call is underway.
-int exportenv(char* stackspace, const char *name, const char *value){
- snprintf(stackspace,ENVLENGTH, "%s=%s", name, value);
- return putenv(stackspace);
-}
+// string can be freed if the environment variable is redefined via
+// another call to putenv(). There is no portable way to unset a variable
+// with putenv(). So we manage the buffer in a static object.
+// Using setenv() if available is not considered because some
+// implementations may produce memory leaks.
-char* dnsdomain(const char* hostname) {
- char *p = NULL;
-#ifdef HAVE_GETADDRINFO
- static char canon_name[NI_MAXHOST];
- struct addrinfo *info = NULL;
- struct addrinfo hints;
- int err;
-
- memset(&hints, 0, sizeof(hints));
- hints.ai_flags = AI_CANONNAME;
- if ((err = getaddrinfo(hostname, NULL, &hints, &info)) || (!info)) {
- PrintOut(LOG_CRIT, "Error retrieving getaddrinfo(%s): %s\n", hostname, gai_strerror(err));
- return NULL;
- }
- if (info->ai_canonname) {
- strncpy(canon_name, info->ai_canonname, sizeof(canon_name));
- canon_name[NI_MAXHOST - 1] = '\0';
- p = canon_name;
- if ((p = strchr(canon_name, '.')))
- p++;
- }
- freeaddrinfo(info);
-#elif HAVE_GETHOSTBYNAME
- struct hostent *hp;
- if ((hp = gethostbyname(hostname))) {
- // Does this work if gethostbyname() returns an IPv6 name in
- // colon/dot notation? [BA]
- if ((p = strchr(hp->h_name, '.')))
- p++; // skip "."
- }
-#else
- ARGUSED(hostname);
-#endif
- return p;
+class env_buffer
+{
+public:
+ env_buffer()
+ : m_buf((char *)0) { }
+
+ void set(const char * name, const char * value);
+
+private:
+ char * m_buf;
+
+ env_buffer(const env_buffer &);
+ void operator=(const env_buffer &);
+};
+
+void env_buffer::set(const char * name, const char * value)
+{
+ int size = strlen(name) + 1 + strlen(value) + 1;
+ char * newbuf = new char[size];
+ snprintf(newbuf, size, "%s=%s", name, value);
+
+ if (putenv(newbuf))
+ throw std::runtime_error("putenv() failed");
+
+ // This assumes that the same NAME is passed on each call
+ delete [] m_buf;
+ m_buf = newbuf;
}
#define EBUFLEN 1024
static void MailWarning(const dev_config & cfg, dev_state & state, int which, const char *fmt, ...)
- __attribute__ ((format (printf, 4, 5)));
+ __attribute_format_printf(4, 5);
// If either address or executable path is non-null then send and log
// a warning email, or execute executable
-static void MailWarning(const dev_config & cfg, dev_state & state, int which, const char *fmt, ...){
- char command[2048], message[256], hostname[256], domainname[256], additional[256],fullmessage[1024];
- char original[256], further[256], nisdomain[256], subject[256],dates[DATEANDEPOCHLEN];
- char environ_strings[11][ENVLENGTH];
- time_t epoch;
- va_list ap;
- const int day=24*3600;
- int days=0;
- const char * const whichfail[]={
+static void MailWarning(const dev_config & cfg, dev_state & state, int which, const char *fmt, ...)
+{
+ static const char * const whichfail[] = {
"EmailTest", // 0
"Health", // 1
"Usage", // 2
"Temperature" // 12
};
- const char *unknown="[Unknown]";
-
// See if user wants us to send mail
if (cfg.emailaddress.empty() && cfg.emailcmdline.empty())
return;
which, (int)sizeof(whichfail));
return;
}
-
+
// Return if a single warning mail has been sent.
if ((cfg.emailfreq==1) && mail->logged)
return;
return;
// To decide if to send mail, we need to know what time it is.
- epoch=time(NULL);
+ time_t epoch = time(0);
// Return if less than one day has gone by
+ const int day = 24*3600;
if (cfg.emailfreq==2 && mail->logged && epoch<(mail->lastsent+day))
return;
// Return if less than 2^(logged-1) days have gone by
if (cfg.emailfreq==3 && mail->logged) {
- days=0x01<<(mail->logged-1);
+ int days = 0x01 << (mail->logged - 1);
days*=day;
if (epoch<(mail->lastsent+days))
return;
}
+#ifdef HAVE_LIBCAP_NG
+ if (enable_capabilities) {
+ PrintOut(LOG_ERR, "Sending a mail was supressed. "
+ "Mails can't be send when capabilites are enabled\n");
+ return;
+ }
+#endif
+
// record the time of this mail message, and the first mail message
if (!mail->logged)
mail->firstsent=epoch;
mail->lastsent=epoch;
-
- // get system host & domain names (not null terminated if length=MAX)
-#ifdef HAVE_GETHOSTNAME
- if (gethostname(hostname, 256))
- strcpy(hostname, unknown);
- else {
- char *p=NULL;
- hostname[255]='\0';
- p = dnsdomain(hostname);
- if (p && *p) {
- strncpy(domainname, p, 255);
- domainname[255]='\0';
- } else
- strcpy(domainname, unknown);
- }
-#else
- strcpy(hostname, unknown);
- strcpy(domainname, unknown);
-#endif
-
-#ifdef HAVE_GETDOMAINNAME
- if (getdomainname(nisdomain, 256))
- strcpy(nisdomain, unknown);
- else
- nisdomain[255]='\0';
-#else
- strcpy(nisdomain, unknown);
-#endif
-
+
// print warning string into message
+ char message[256];
+ va_list ap;
va_start(ap, fmt);
- vsnprintf(message, 256, fmt, ap);
+ vsnprintf(message, sizeof(message), fmt, ap);
va_end(ap);
- // appropriate message about further information
- additional[0]=original[0]=further[0]='\0';
- if (which) {
- sprintf(further,"You can also use the smartctl utility for further investigation.\n");
-
- switch (cfg.emailfreq) {
- case 1:
- sprintf(additional,"No additional email messages about this problem will be sent.\n");
- break;
- case 2:
- sprintf(additional,"Another email message will be sent in 24 hours if the problem persists.\n");
- break;
- case 3:
- sprintf(additional,"Another email message will be sent in %d days if the problem persists\n",
- (0x01)<<mail->logged);
- break;
- }
- if (cfg.emailfreq>1 && mail->logged) {
- dateandtimezoneepoch(dates, mail->firstsent);
- sprintf(original,"The original email about this issue was sent at %s\n", dates);
- }
- }
-
- snprintf(subject, 256,"SMART error (%s) detected on host: %s", whichfail[which], hostname);
-
- // If the user has set cfg.emailcmdline, use that as mailer, else "mail" or "mailx".
- if (!*executable)
-#ifdef DEFAULT_MAILER
- executable = DEFAULT_MAILER ;
-#else
-#ifndef _WIN32
- executable = "mail";
-#else
- executable = "blat"; // http://blat.sourceforge.net/
-#endif
-#endif
-
-#ifndef _WIN32 // blat mailer needs comma
// replace commas by spaces to separate recipients
std::replace(address.begin(), address.end(), ',', ' ');
-#endif
+
// Export information in environment variables that will be useful
// for user scripts
- exportenv(environ_strings[0], "SMARTD_MAILER", executable);
- exportenv(environ_strings[1], "SMARTD_MESSAGE", message);
- exportenv(environ_strings[2], "SMARTD_SUBJECT", subject);
+ static env_buffer env[12];
+ env[0].set("SMARTD_MAILER", executable);
+ env[1].set("SMARTD_MESSAGE", message);
+ char dates[DATEANDEPOCHLEN];
+ snprintf(dates, sizeof(dates), "%d", mail->logged);
+ env[2].set("SMARTD_PREVCNT", dates);
dateandtimezoneepoch(dates, mail->firstsent);
- exportenv(environ_strings[3], "SMARTD_TFIRST", dates);
+ env[3].set("SMARTD_TFIRST", dates);
snprintf(dates, DATEANDEPOCHLEN,"%d", (int)mail->firstsent);
- exportenv(environ_strings[4], "SMARTD_TFIRSTEPOCH", dates);
- exportenv(environ_strings[5], "SMARTD_FAILTYPE", whichfail[which]);
- if (!address.empty())
- exportenv(environ_strings[6], "SMARTD_ADDRESS", address.c_str());
- exportenv(environ_strings[7], "SMARTD_DEVICESTRING", cfg.name.c_str());
-
- exportenv(environ_strings[8], "SMARTD_DEVICETYPE", cfg.dev_type.c_str());
- exportenv(environ_strings[9], "SMARTD_DEVICE", cfg.name.c_str());
-
- snprintf(fullmessage, 1024,
- "This email was generated by the smartd daemon running on:\n\n"
- " host name: %s\n"
- " DNS domain: %s\n"
- " NIS domain: %s\n\n"
- "The following warning/error was logged by the smartd daemon:\n\n"
- "%s\n\n"
- "For details see host's SYSLOG (default: /var/log/messages).\n\n"
- "%s%s%s",
- hostname, domainname, nisdomain, message, further, original, additional);
- exportenv(environ_strings[10], "SMARTD_FULLMESSAGE", fullmessage);
+ env[4].set("SMARTD_TFIRSTEPOCH", dates);
+ env[5].set("SMARTD_FAILTYPE", whichfail[which]);
+ env[6].set("SMARTD_ADDRESS", address.c_str());
+ env[7].set("SMARTD_DEVICESTRING", cfg.name.c_str());
+
+ // Allow 'smartctl ... -d $SMARTD_DEVICETYPE $SMARTD_DEVICE'
+ env[8].set("SMARTD_DEVICETYPE",
+ (!cfg.dev_type.empty() ? cfg.dev_type.c_str() : "auto"));
+ env[9].set("SMARTD_DEVICE", cfg.dev_name.c_str());
+
+ env[10].set("SMARTD_DEVICEINFO", cfg.dev_idinfo.c_str());
+ dates[0] = 0;
+ if (which) switch (cfg.emailfreq) {
+ case 2: dates[0] = '1'; dates[1] = 0; break;
+ case 3: snprintf(dates, sizeof(dates), "%d", (0x01)<<mail->logged);
+ }
+ env[11].set("SMARTD_NEXTDAYS", dates);
// now construct a command to send this as EMAIL
-#ifndef _WIN32
- if (!address.empty())
- snprintf(command, 2048,
- "$SMARTD_MAILER -s '%s' %s 2>&1 << \"ENDMAIL\"\n"
- "%sENDMAIL\n", subject, address.c_str(), fullmessage);
- else
- snprintf(command, 2048, "%s 2>&1", executable);
-
- // tell SYSLOG what we are about to do...
+ char command[2048];
+ if (!*executable)
+ executable = "<mail>";
const char * newadd = (!address.empty()? address.c_str() : "<nomailer>");
const char * newwarn = (which? "Warning via" : "Test of");
+#ifndef _WIN32
+ snprintf(command, sizeof(command), "%s 2>&1", warning_script.c_str());
+
+ // tell SYSLOG what we are about to do...
PrintOut(LOG_INFO,"%s %s to %s ...\n",
which?"Sending warning via":"Executing test of", executable, newadd);
}
#else // _WIN32
+ {
+ snprintf(command, sizeof(command), "cmd /c \"%s\"", warning_script.c_str());
- // No "here-documents" on Windows, so must use separate commandline and stdin
- char stdinbuf[1024];
- command[0] = stdinbuf[0] = 0;
- int boxtype = -1, boxmsgoffs = 0;
- const char * newadd = "<nomailer>";
- if (!address.empty()) {
- // address "[sys]msgbox ..." => show warning (also) as [system modal ]messagebox
- char addr1[9+1+13] = ""; int n1 = -1, n2 = -1;
- if (sscanf(address.c_str(), "%9[a-z]%n,%n", addr1, &n1, &n2) == 1 && (n1 == (int)address.size() || n2 > 0)) {
- if (!strcmp(addr1, "msgbox"))
- boxtype = 0;
- else if (!strcmp(addr1, "sysmsgbox"))
- boxtype = 1;
- if (boxtype >= 0)
- address.erase(0, (n2 > n1 ? n2 : n1));
- }
-
- if (!address.empty()) {
- // Use "blat" parameter syntax (TODO: configure via -M for other mailers)
- snprintf(command, sizeof(command),
- "%s - -q -subject \"%s\" -to \"%s\"",
- executable, subject, address.c_str());
- newadd = address.c_str();
- }
-
-#ifdef _MSC_VER
- _set_printf_count_output(1); // "%n" disabled by default
-#endif
- // Message for mail [0...] and messagebox [boxmsgoffs...]
- snprintf(stdinbuf, sizeof(stdinbuf),
- "This email was generated by the smartd daemon running on:\n\n"
- " host name: %s\n"
- " DNS domain: %s\n"
-// " NIS domain: %s\n"
- "\n%n"
- "The following warning/error was logged by the smartd daemon:\n\n"
- "%s\n\n"
- "For details see the event log or log file of smartd.\n\n"
- "%s%s%s"
- "\n",
- hostname, /*domainname, */ nisdomain, &boxmsgoffs, message, further, original, additional);
- }
- else
- snprintf(command, sizeof(command), "%s", executable);
-
- const char * newwarn = (which ? "Warning via" : "Test of");
- if (boxtype >= 0) {
- // show message box
- daemon_messagebox(boxtype, subject, stdinbuf+boxmsgoffs);
- PrintOut(LOG_INFO,"%s message box\n", newwarn);
- }
- if (command[0]) {
char stdoutbuf[800]; // < buffer in syslog_win32::vsyslog()
int rc;
// run command
PrintOut(LOG_INFO,"%s %s to %s ...\n",
(which?"Sending warning via":"Executing test of"), executable, newadd);
- rc = daemon_spawn(command, stdinbuf, strlen(stdinbuf), stdoutbuf, sizeof(stdoutbuf));
+ rc = daemon_spawn(command, "", 0, stdoutbuf, sizeof(stdoutbuf));
if (rc >= 0 && stdoutbuf[0])
PrintOut(LOG_CRIT,"%s %s to %s produced unexpected output (%d bytes) to STDOUT/STDERR:\n%s\n",
- newwarn, executable, newadd, strlen(stdoutbuf), stdoutbuf);
+ newwarn, executable, newadd, (int)strlen(stdoutbuf), stdoutbuf);
if (rc != 0)
PrintOut(LOG_CRIT,"%s %s to %s: failed, exit status %d\n",
newwarn, executable, newadd, rc);
mail->logged++;
}
+static void reset_warning_mail(const dev_config & cfg, dev_state & state, int which, const char *fmt, ...)
+ __attribute_format_printf(4, 5);
+
+static void reset_warning_mail(const dev_config & cfg, dev_state & state, int which, const char *fmt, ...)
+{
+ if (!(0 <= which && which < SMARTD_NMAIL))
+ return;
+
+ // Return if no mail sent yet
+ mailinfo & mi = state.maillog[which];
+ if (!mi.logged)
+ return;
+
+ // Format & print message
+ char msg[256];
+ va_list ap;
+ va_start(ap, fmt);
+ vsnprintf(msg, sizeof(msg), fmt, ap);
+ va_end(ap);
+
+ PrintOut(LOG_INFO, "Device: %s, %s, warning condition reset after %d email%s\n", cfg.name.c_str(),
+ msg, mi.logged, (mi.logged==1 ? "" : "s"));
+
+ // Clear mail counter and timestamps
+ mi = mailinfo();
+ state.must_write = true;
+}
+
+#ifndef _WIN32
+
+// Output multiple lines via separate syslog(3) calls.
+static void vsyslog_lines(int priority, const char * fmt, va_list ap)
+{
+ char buf[512+EBUFLEN]; // enough space for exec cmd output in MailWarning()
+ vsnprintf(buf, sizeof(buf), fmt, ap);
+
+ for (char * p = buf, * q; p && *p; p = q) {
+ if ((q = strchr(p, '\n')))
+ *q++ = 0;
+ if (*p)
+ syslog(priority, "%s\n", p);
+ }
+}
+
+#else // _WIN32
+// os_win32/syslog_win32.cpp supports multiple lines.
+#define vsyslog_lines vsyslog
+#endif // _WIN32
+
// Printing function for watching ataprint commands, or losing them
// [From GLIBC Manual: Since the prototype doesn't specify types for
// optional arguments, in a call to a variadic function the default
FixGlibcTimeZoneBug();
// initialize variable argument list
va_start(ap,fmt);
- // in debug==1 mode we will print the output from the ataprint.o functions!
- if (debugmode && debugmode!=2)
+ // in debugmode==1 mode we will print the output from the ataprint.o functions!
+ if (debugmode && debugmode != 2) {
+ FILE * f = stdout;
#ifdef _WIN32
- if (facility == LOG_LOCAL1) // logging to stdout
- vfprintf(stderr,fmt,ap);
- else
+ if (facility == LOG_LOCAL1) // logging to stdout
+ f = stderr;
#endif
- vprintf(fmt,ap);
- // in debug==2 mode we print output from knowndrives.o functions
- else if (debugmode==2 || con->reportataioctl || con->reportscsiioctl /*|| con->controller_port???*/) {
+ vfprintf(f, fmt, ap);
+ fflush(f);
+ }
+ // in debugmode==2 mode we print output from knowndrives.o functions
+ else if (debugmode==2 || ata_debugmode || scsi_debugmode) {
openlog("smartd", LOG_PID, facility);
- vsyslog(LOG_INFO, fmt, ap);
+ vsyslog_lines(LOG_INFO, fmt, ap);
closelog();
}
va_end(ap);
- fflush(NULL);
return;
}
// This function prints either to stdout or to the syslog as needed.
-// This function is also used by utility.cpp to report LOG_CRIT errors.
-void PrintOut(int priority, const char *fmt, ...){
+static void PrintOut(int priority, const char *fmt, ...){
va_list ap;
// get the correct time in syslog()
FixGlibcTimeZoneBug();
// initialize variable argument list
va_start(ap,fmt);
- if (debugmode)
+ if (debugmode) {
+ FILE * f = stdout;
#ifdef _WIN32
- if (facility == LOG_LOCAL1) // logging to stdout
- vfprintf(stderr,fmt,ap);
- else
+ if (facility == LOG_LOCAL1) // logging to stdout
+ f = stderr;
#endif
- vprintf(fmt,ap);
+ vfprintf(f, fmt, ap);
+ fflush(f);
+ }
else {
openlog("smartd", LOG_PID, facility);
- vsyslog(priority,fmt,ap);
+ vsyslog_lines(priority, fmt, ap);
closelog();
}
va_end(ap);
pout("Warning! %s error: invalid SMART checksum.\n", string);
}
+#ifndef _WIN32
+
// Wait for the pid file to show up, this makes sure a calling program knows
// that the daemon is really up and running and has a pid to kill it
-bool WaitForPidFile()
+static bool WaitForPidFile()
{
int waited, max_wait = 10;
struct stat stat_buf;
return false;
}
+#endif // _WIN32
// Forks new process, closes ALL file descriptors, redirects stdin,
// stdout, and stderr. Not quite daemon(). See
// http://www.linuxjournal.com/article/2335
// for a good description of why we do things this way.
-void DaemonInit(){
+static void DaemonInit()
+{
#ifndef _WIN32
pid_t pid;
int i;
for (i=getdtablesize();i>=0;--i)
close(i);
-#ifdef __CYGWIN__
- // Cygwin's setsid() does not detach the process from Windows console
- FreeConsole();
-#endif // __CYGWIN__
-
#define NO_warn_unused_result(cmd) { if (cmd) {} ; }
// redirect any IO attempts to /dev/null for stdin
}
// prints help info for configuration file Directives
-void Directives() {
+static void Directives()
+{
PrintOut(LOG_INFO,
"Configuration file (%s) Directives (after device name):\n"
- " -d TYPE Set the device type: %s\n"
+ " -d TYPE Set the device type: auto, ignore, removable,\n"
+ " %s\n"
" -T TYPE Set the tolerance to one of: normal, permissive\n"
" -o VAL Enable/disable automatic offline tests (on/off)\n"
" -S VAL Enable/disable attribute autosave (on/off)\n"
" -n MODE No check if: never, sleep[,N][,q], standby[,N][,q], idle[,N][,q]\n"
" -H Monitor SMART Health Status, report if failed\n"
" -s REG Do Self-Test at time(s) given by regular expression REG\n"
- " -l TYPE Monitor SMART log. Type is one of: error, selftest\n"
+ " -l TYPE Monitor SMART log or self-test status:\n"
+ " error, selftest, xerror, offlinests[,ns], selfteststs[,ns]\n"
+ " -l scterc,R,W Set SCT Error Recovery Control\n"
+ " -e Change device setting: aam,[N|off], apm,[N|off], lookahead,[on|off],\n"
+ " security-freeze, standby,[N|off], wcache,[on|off]\n"
" -f Monitor 'Usage' Attributes, report failures\n"
" -m ADD Send email warning to address ADD\n"
" -M TYPE Modify email warning behavior (see man page)\n"
" -W D,I,C Monitor Temperature D)ifference, I)nformal limit, C)ritical limit\n"
" -v N,ST Modifies labeling of Attribute N (see man page) \n"
" -P TYPE Drive-specific presets: use, ignore, show, showall\n"
- " -a Default: -H -f -t -l error -l selftest -C 197 -U 198\n"
- " -F TYPE Firmware bug workaround: none, samsung, samsung2, samsung3\n"
+ " -a Default: -H -f -t -l error -l selftest -l selfteststs -C 197 -U 198\n"
+ " -F TYPE Use firmware bug workaround:\n"
+ " %s\n"
" # Comment: text after a hash sign is ignored\n"
" \\ Line continuation character\n"
"Attribute ID is a decimal integer 1 <= ID <= 255\n"
- "Use ID = 0 to turn off -C and/or -U Directives\n"
- "Example: /dev/hda -a\n",
- configfile, smi()->get_valid_dev_types_str().c_str());
- return;
+ "Use ID = 0 to turn off -C and/or -U Directives\n"
+ "Example: /dev/sda -a\n",
+ configfile,
+ smi()->get_valid_dev_types_str().c_str(),
+ get_valid_firmwarebug_args());
}
/* 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) {
+static const char *GetValidArgList(char opt)
+{
switch (opt) {
case 'A':
case 's':
return "ioctl[,N], ataioctl[,N], scsiioctl[,N]";
case 'B':
case 'p':
+ case 'w':
return "<FILE_NAME>";
case 'i':
return "<INTEGER_SECONDS>";
}
/* prints help information for command syntax */
-void Usage (void){
+static void Usage()
+{
PrintOut(LOG_INFO,"Usage: smartd [options]\n\n");
PrintOut(LOG_INFO," -A PREFIX, --attributelog=PREFIX\n");
PrintOut(LOG_INFO," Log ATA attribute information to {PREFIX}MODEL-SERIAL.ata.csv\n");
#ifdef SMARTMONTOOLS_ATTRIBUTELOG
- PrintOut(LOG_INFO," [default is "SMARTMONTOOLS_ATTRIBUTELOG"MODEL-SERIAL.ata.csv]\n");
+ PrintOut(LOG_INFO," [default is " SMARTMONTOOLS_ATTRIBUTELOG "MODEL-SERIAL.ata.csv]\n");
#endif
PrintOut(LOG_INFO,"\n");
PrintOut(LOG_INFO," -B [+]FILE, --drivedb=[+]FILE\n");
PrintOut(LOG_INFO," Read and replace [add] drive database from FILE\n");
+ PrintOut(LOG_INFO," [default is +%s", get_drivedb_path_add());
#ifdef SMARTMONTOOLS_DRIVEDBDIR
- PrintOut(LOG_INFO," [default is "SMARTMONTOOLS_DRIVEDBDIR"/drivedb.h]\n");
-#endif
PrintOut(LOG_INFO,"\n");
+ PrintOut(LOG_INFO," and then %s", get_drivedb_path_default());
+#endif
+ PrintOut(LOG_INFO,"]\n\n");
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," Read configuration file NAME or stdin\n");
+ PrintOut(LOG_INFO," [default is %s]\n\n", configfile);
+#ifdef HAVE_LIBCAP_NG
+ PrintOut(LOG_INFO," -C, --capabilities\n");
+ PrintOut(LOG_INFO," Use capabilities.\n"
+ " Warning: Mail notification does not work when used.\n\n");
+#endif
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," -s PREFIX, --savestates=PREFIX\n");
PrintOut(LOG_INFO," Save disk states to {PREFIX}MODEL-SERIAL.TYPE.state\n");
#ifdef SMARTMONTOOLS_SAVESTATES
- PrintOut(LOG_INFO," [default is "SMARTMONTOOLS_SAVESTATES"MODEL-SERIAL.TYPE.state]\n");
+ PrintOut(LOG_INFO," [default is " SMARTMONTOOLS_SAVESTATES "MODEL-SERIAL.TYPE.state]\n");
#endif
PrintOut(LOG_INFO,"\n");
+ PrintOut(LOG_INFO," -w NAME, --warnexec=NAME\n");
+ PrintOut(LOG_INFO," Run executable NAME on warnings\n");
+#ifndef _WIN32
+ PrintOut(LOG_INFO," [default is " SMARTMONTOOLS_SMARTDSCRIPTDIR "/smartd_warning.sh]\n\n");
+#else
+ PrintOut(LOG_INFO," [default is %s/smartd_warning.cmd]\n\n", get_exe_dir().c_str());
+#endif
#ifdef _WIN32
PrintOut(LOG_INFO," --service\n");
PrintOut(LOG_INFO," Running as windows service (see man page), install with:\n");
|| ('a' <= c && c <= 'z'));
}
-// returns <0 on failure
-static int ATAErrorCount(ata_device * device, const char * name,
- unsigned char fix_firmwarebug)
+// Read error count from Summary or Extended Comprehensive SMART error log
+// Return -1 on error
+static int read_ata_error_count(ata_device * device, const char * name,
+ firmwarebug_defs firmwarebugs, bool extended)
{
- struct ata_smart_errorlog log;
-
- if (ataReadErrorLog(device, &log, fix_firmwarebug)){
- PrintOut(LOG_INFO,"Device: %s, Read SMART Error Log Failed\n",name);
- return -1;
+ if (!extended) {
+ ata_smart_errorlog log;
+ if (ataReadErrorLog(device, &log, firmwarebugs)){
+ PrintOut(LOG_INFO,"Device: %s, Read Summary SMART Error Log failed\n",name);
+ return -1;
+ }
+ return (log.error_log_pointer ? log.ata_error_count : 0);
+ }
+ else {
+ ata_smart_exterrlog logx;
+ if (!ataReadExtErrorLog(device, &logx, 1 /*first sector only*/, firmwarebugs)) {
+ PrintOut(LOG_INFO,"Device: %s, Read Extended Comprehensive SMART Error Log failed\n",name);
+ return -1;
+ }
+ // Some disks use the reserved byte as index, see ataprint.cpp.
+ return (logx.error_log_index || logx.reserved1 ? logx.device_error_count : 0);
}
-
- // 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.
static int SelfTestErrorCount(ata_device * device, const char * name,
- unsigned char fix_firmwarebug)
+ firmwarebug_defs firmwarebugs)
{
struct ata_smart_selftestlog log;
- if (ataReadSelfTestLog(device, &log, fix_firmwarebug)){
+ if (ataReadSelfTestLog(device, &log, firmwarebugs)){
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, false, fix_firmwarebug);
+ return ataPrintSmartSelfTestlog(&log, false, firmwarebugs);
}
#define SELFTEST_ERRORCOUNT(x) (x & 0xff)
#define SELFTEST_ERRORHOURS(x) ((x >> 8) & 0xffff)
+// Check offline data collection status
+static inline bool is_offl_coll_in_progress(unsigned char status)
+{
+ return ((status & 0x7f) == 0x03);
+}
+
+// Check self-test execution status
+static inline bool is_self_test_in_progress(unsigned char status)
+{
+ return ((status >> 4) == 0xf);
+}
+
+// Log offline data collection status
+static void log_offline_data_coll_status(const char * name, unsigned char status)
+{
+ const char * msg;
+ switch (status & 0x7f) {
+ case 0x00: msg = "was never started"; break;
+ case 0x02: msg = "was completed without error"; break;
+ case 0x03: msg = "is in progress"; break;
+ case 0x04: msg = "was suspended by an interrupting command from host"; break;
+ case 0x05: msg = "was aborted by an interrupting command from host"; break;
+ case 0x06: msg = "was aborted by the device with a fatal error"; break;
+ default: msg = 0;
+ }
+
+ if (msg)
+ PrintOut(((status & 0x7f) == 0x06 ? LOG_CRIT : LOG_INFO),
+ "Device: %s, offline data collection %s%s\n", name, msg,
+ ((status & 0x80) ? " (auto:on)" : ""));
+ else
+ PrintOut(LOG_INFO, "Device: %s, unknown offline data collection status 0x%02x\n",
+ name, status);
+}
+
// Log self-test execution status
static void log_self_test_exec_status(const char * name, unsigned char status)
{
name, status);
}
+// Check pending sector count id (-C, -U directives).
+static bool check_pending_id(const dev_config & cfg, const dev_state & state,
+ unsigned char id, const char * msg)
+{
+ // Check attribute index
+ int i = ata_find_attr_index(id, state.smartval);
+ if (i < 0) {
+ PrintOut(LOG_INFO, "Device: %s, can't monitor %s count - no Attribute %d\n",
+ cfg.name.c_str(), msg, id);
+ return false;
+ }
+
+ // Check value
+ uint64_t rawval = ata_get_attr_raw_value(state.smartval.vendor_attributes[i],
+ cfg.attribute_defs);
+ if (rawval >= (state.num_sectors ? state.num_sectors : 0xffffffffULL)) {
+ PrintOut(LOG_INFO, "Device: %s, ignoring %s count - bogus Attribute %d value %" PRIu64 " (0x%" PRIx64 ")\n",
+ cfg.name.c_str(), msg, id, rawval, rawval);
+ return false;
+ }
+
+ return true;
+}
+
+// Called by ATA/SCSIDeviceScan() after successful device check
+static void finish_device_scan(dev_config & cfg, dev_state & state)
+{
+ // Set cfg.emailfreq if user hasn't set it
+ if ((!cfg.emailaddress.empty() || !cfg.emailcmdline.empty()) && !cfg.emailfreq) {
+ // Avoid that emails are suppressed forever due to state persistence
+ if (cfg.state_file.empty())
+ cfg.emailfreq = 1; // '-M once'
+ else
+ cfg.emailfreq = 2; // '-M daily'
+ }
+
+ // Start self-test regex check now if time was not read from state file
+ if (!cfg.test_regex.empty() && !state.scheduled_test_next_check)
+ state.scheduled_test_next_check = time(0);
+}
+
+// Common function to format result message for ATA setting
+static void format_set_result_msg(std::string & msg, const char * name, bool ok,
+ int set_option = 0, bool has_value = false)
+{
+ if (!msg.empty())
+ msg += ", ";
+ msg += name;
+ if (!ok)
+ msg += ":--";
+ else if (set_option < 0)
+ msg += ":off";
+ else if (has_value)
+ msg += strprintf(":%d", set_option-1);
+ else if (set_option > 0)
+ msg += ":on";
+}
+
// TODO: Add '-F swapid' directive
const bool fix_swapped_id = false;
// Device must be open
// Get drive identity structure
- if ((retid=ataReadHDIdentity (atadev, &drive))){
+ if ((retid = ata_read_identity(atadev, &drive, fix_swapped_id))) {
if (retid<0)
// Unable to read Identity structure
PrintOut(LOG_INFO,"Device: %s, not ATA, no IDENTIFY DEVICE Structure\n",name);
CloseDevice(atadev, name);
return 2;
}
- // Store drive size (for selective self-test only)
- state.num_sectors = get_num_sectors(&drive);
+
+ // Get drive identity, size and rotation rate (HDD/SSD)
+ char model[40+1], serial[20+1], firmware[8+1];
+ ata_format_id_string(model, drive.model, sizeof(model)-1);
+ ata_format_id_string(serial, drive.serial_no, sizeof(serial)-1);
+ ata_format_id_string(firmware, drive.fw_rev, sizeof(firmware)-1);
+
+ ata_size_info sizes;
+ ata_get_size_info(&drive, sizes);
+ state.num_sectors = sizes.sectors;
+ cfg.dev_rpm = ata_get_rotation_rate(&drive);
+
+ char wwn[30]; wwn[0] = 0;
+ unsigned oui = 0; uint64_t unique_id = 0;
+ int naa = ata_get_wwn(&drive, oui, unique_id);
+ if (naa >= 0)
+ snprintf(wwn, sizeof(wwn), "WWN:%x-%06x-%09" PRIx64 ", ", naa, oui, unique_id);
+
+ // Format device id string for warning emails
+ char cap[32];
+ cfg.dev_idinfo = strprintf("%s, S/N:%s, %sFW:%s, %s", model, serial, wwn, firmware,
+ format_capacity(cap, sizeof(cap), sizes.capacity, "."));
+
+ PrintOut(LOG_INFO, "Device: %s, %s\n", name, cfg.dev_idinfo.c_str());
// 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.
- if (!apply_presets(&drive, cfg.attribute_defs, cfg.fix_firmwarebug, fix_swapped_id))
+ // Apply vendor specific presets, print warning if present
+ const drive_settings * dbentry = lookup_drive_apply_presets(
+ &drive, cfg.attribute_defs, cfg.firmwarebugs);
+ if (!dbentry)
PrintOut(LOG_INFO, "Device: %s, not found in smartd database.\n", name);
- else
- PrintOut(LOG_INFO, "Device: %s, found in smartd database.\n", name);
+ else {
+ PrintOut(LOG_INFO, "Device: %s, found in smartd database%s%s\n",
+ name, (*dbentry->modelfamily ? ": " : "."), (*dbentry->modelfamily ? dbentry->modelfamily : ""));
+ if (*dbentry->warningmsg)
+ PrintOut(LOG_CRIT, "Device: %s, WARNING: %s\n", name, dbentry->warningmsg);
+ }
}
// Set default '-C 197[+]' if no '-C ID' is specified.
PrintOut(LOG_INFO, "Device %s: presets are:\n", name);
if (!debugmode)
debugmode=2;
- show_presets(&drive, false);
+ show_presets(&drive);
debugmode=savedebugmode;
}
// 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?
- bool retainsmartdata = (cfg.usagefailed || cfg.prefail || cfg.usage || cfg.tempdiff || cfg.tempinfo || cfg.tempcrit);
-
// do we need to get SMART data?
bool smart_val_ok = false;
- if ( retainsmartdata || cfg.autoofflinetest || cfg.selftest || cfg.errorlog
- || cfg.curr_pending_id || cfg.offl_pending_id ) {
-
- if (ataReadSmartValues(atadev, &state.smartval) ||
- ataReadSmartThresholds (atadev, &state.smartthres)) {
- PrintOut(LOG_INFO,"Device: %s, Read SMART Values and/or Thresholds Failed\n",name);
- retainsmartdata = cfg.usagefailed = cfg.prefail = cfg.usage = false;
+ if ( cfg.autoofflinetest || cfg.selftest
+ || cfg.errorlog || cfg.xerrorlog
+ || cfg.offlinests || cfg.selfteststs
+ || cfg.usagefailed || cfg.prefail || cfg.usage
+ || cfg.tempdiff || cfg.tempinfo || cfg.tempcrit
+ || cfg.curr_pending_id || cfg.offl_pending_id ) {
+
+ if (ataReadSmartValues(atadev, &state.smartval)) {
+ PrintOut(LOG_INFO, "Device: %s, Read SMART Values failed\n", name);
+ cfg.usagefailed = cfg.prefail = cfg.usage = false;
cfg.tempdiff = cfg.tempinfo = cfg.tempcrit = 0;
cfg.curr_pending_id = cfg.offl_pending_id = 0;
}
- else
+ else {
smart_val_ok = true;
+ if (ataReadSmartThresholds(atadev, &state.smartthres)) {
+ PrintOut(LOG_INFO, "Device: %s, Read SMART Thresholds failed%s\n",
+ name, (cfg.usagefailed ? ", ignoring -f Directive" : ""));
+ cfg.usagefailed = false;
+ // Let ata_get_attr_state() return ATTRSTATE_NO_THRESHOLD:
+ memset(&state.smartthres, 0, sizeof(state.smartthres));
+ }
+ }
// see if the necessary Attribute is there to monitor offline or
// current pending sectors or temperature
- if (cfg.curr_pending_id && ata_find_attr_index(cfg.curr_pending_id, state.smartval) < 0) {
- PrintOut(LOG_INFO,"Device: %s, can't monitor Current Pending Sector count - no Attribute %d\n",
- name, cfg.curr_pending_id);
+ if ( cfg.curr_pending_id
+ && !check_pending_id(cfg, state, cfg.curr_pending_id,
+ "Current_Pending_Sector"))
cfg.curr_pending_id = 0;
- }
-
- if (cfg.offl_pending_id && ata_find_attr_index(cfg.offl_pending_id, state.smartval) < 0) {
- PrintOut(LOG_INFO,"Device: %s, can't monitor Offline Uncorrectable Sector count - no Attribute %d\n",
- name, cfg.offl_pending_id);
+
+ if ( cfg.offl_pending_id
+ && !check_pending_id(cfg, state, cfg.offl_pending_id,
+ "Offline_Uncorrectable"))
cfg.offl_pending_id = 0;
- }
if ( (cfg.tempdiff || cfg.tempinfo || cfg.tempcrit)
&& !ata_return_temperature_value(&state.smartval, cfg.attribute_defs)) {
- PrintOut(LOG_CRIT, "Device: %s, can't monitor Temperature, ignoring -W Directive\n", name);
+ PrintOut(LOG_INFO, "Device: %s, can't monitor Temperature, ignoring -W %d,%d,%d\n",
+ name, cfg.tempdiff, cfg.tempinfo, cfg.tempcrit);
cfg.tempdiff = cfg.tempinfo = cfg.tempcrit = 0;
}
+
+ // Report ignored '-r' or '-R' directives
+ for (int id = 1; id <= 255; id++) {
+ if (cfg.monitor_attr_flags.is_set(id, MONITOR_RAW_PRINT)) {
+ char opt = (!cfg.monitor_attr_flags.is_set(id, MONITOR_RAW) ? 'r' : 'R');
+ const char * excl = (cfg.monitor_attr_flags.is_set(id,
+ (opt == 'r' ? MONITOR_AS_CRIT : MONITOR_RAW_AS_CRIT)) ? "!" : "");
+
+ int idx = ata_find_attr_index(id, state.smartval);
+ if (idx < 0)
+ PrintOut(LOG_INFO,"Device: %s, no Attribute %d, ignoring -%c %d%s\n", name, id, opt, id, excl);
+ else {
+ bool prefail = !!ATTRIBUTE_FLAGS_PREFAILURE(state.smartval.vendor_attributes[idx].flags);
+ if (!((prefail && cfg.prefail) || (!prefail && cfg.usage)))
+ PrintOut(LOG_INFO,"Device: %s, not monitoring %s Attributes, ignoring -%c %d%s\n", name,
+ (prefail ? "Prefailure" : "Usage"), opt, id, excl);
+ }
+ }
+ }
}
// enable/disable automatic on-line testing
PrintOut(LOG_INFO,"Device: %s, %sd SMART Automatic Offline Testing.\n", name, what);
}
}
-
+
+ // Read log directories if required for capability check
+ ata_smart_log_directory smart_logdir, gp_logdir;
+ bool smart_logdir_ok = false, gp_logdir_ok = false;
+
+ if ( isGeneralPurposeLoggingCapable(&drive)
+ && (cfg.errorlog || cfg.selftest)
+ && !cfg.firmwarebugs.is_set(BUG_NOLOGDIR)) {
+ if (!ataReadLogDirectory(atadev, &smart_logdir, false))
+ smart_logdir_ok = true;
+ }
+
+ if (cfg.xerrorlog && !cfg.firmwarebugs.is_set(BUG_NOLOGDIR)) {
+ if (!ataReadLogDirectory(atadev, &gp_logdir, true))
+ gp_logdir_ok = true;
+ }
+
// capability check: self-test-log
+ state.selflogcount = 0; state.selfloghour = 0;
if (cfg.selftest) {
int retval;
-
- // start with service disabled, and re-enable it if all works OK
- cfg.selftest = false;
- state.selflogcount = 0;
- state.selfloghour = 0;
-
- if (!smart_val_ok)
- PrintOut(LOG_INFO, "Device: %s, no SMART Self-Test log (SMART READ DATA failed); disabling -l selftest\n", name);
- else if (!cfg.permissive && !isSmartTestLogCapable(&state.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(atadev, name, cfg.fix_firmwarebug)) < 0)
- PrintOut(LOG_INFO, "Device: %s, no SMART Self-Test log; remove -l selftest Directive from smartd.conf\n", name);
+ if (!( cfg.permissive
+ || ( smart_logdir_ok && smart_logdir.entry[0x06-1].numsectors)
+ || (!smart_logdir_ok && smart_val_ok && isSmartTestLogCapable(&state.smartval, &drive)))) {
+ PrintOut(LOG_INFO, "Device: %s, no SMART Self-test Log, ignoring -l selftest (override with -T permissive)\n", name);
+ cfg.selftest = false;
+ }
+ else if ((retval = SelfTestErrorCount(atadev, name, cfg.firmwarebugs)) < 0) {
+ PrintOut(LOG_INFO, "Device: %s, no SMART Self-test Log, ignoring -l selftest\n", name);
+ cfg.selftest = false;
+ }
else {
- cfg.selftest = true;
state.selflogcount=SELFTEST_ERRORCOUNT(retval);
state.selfloghour =SELFTEST_ERRORHOURS(retval);
}
}
// capability check: ATA error log
+ state.ataerrorcount = 0;
if (cfg.errorlog) {
- int val;
+ int errcnt1;
+ if (!( cfg.permissive
+ || ( smart_logdir_ok && smart_logdir.entry[0x01-1].numsectors)
+ || (!smart_logdir_ok && smart_val_ok && isSmartErrorLogCapable(&state.smartval, &drive)))) {
+ PrintOut(LOG_INFO, "Device: %s, no SMART Error Log, ignoring -l error (override with -T permissive)\n", name);
+ cfg.errorlog = false;
+ }
+ else if ((errcnt1 = read_ata_error_count(atadev, name, cfg.firmwarebugs, false)) < 0) {
+ PrintOut(LOG_INFO, "Device: %s, no SMART Error Log, ignoring -l error\n", name);
+ cfg.errorlog = false;
+ }
+ else
+ state.ataerrorcount = errcnt1;
+ }
- // start with service disabled, and re-enable it if all works OK
- cfg.errorlog = false;
- state.ataerrorcount=0;
+ if (cfg.xerrorlog) {
+ int errcnt2;
+ if (!( cfg.permissive || cfg.firmwarebugs.is_set(BUG_NOLOGDIR)
+ || (gp_logdir_ok && gp_logdir.entry[0x03-1].numsectors) )) {
+ PrintOut(LOG_INFO, "Device: %s, no Extended Comprehensive SMART Error Log, ignoring -l xerror (override with -T permissive)\n",
+ name);
+ cfg.xerrorlog = false;
+ }
+ else if ((errcnt2 = read_ata_error_count(atadev, name, cfg.firmwarebugs, true)) < 0) {
+ PrintOut(LOG_INFO, "Device: %s, no Extended Comprehensive SMART Error Log, ignoring -l xerror\n", name);
+ cfg.xerrorlog = false;
+ }
+ else if (cfg.errorlog && state.ataerrorcount != errcnt2) {
+ PrintOut(LOG_INFO, "Device: %s, SMART Error Logs report different error counts: %d != %d\n",
+ name, state.ataerrorcount, errcnt2);
+ // Record max error count
+ if (errcnt2 > state.ataerrorcount)
+ state.ataerrorcount = errcnt2;
+ }
+ else
+ state.ataerrorcount = errcnt2;
+ }
- if (!smart_val_ok)
- PrintOut(LOG_INFO, "Device: %s, no SMART Error log (SMART READ DATA failed); disabling -l error\n", name);
- else if (!cfg.permissive && !isSmartErrorLogCapable(&state.smartval, &drive))
- PrintOut(LOG_INFO, "Device: %s, appears to lack SMART Error log; disabling -l error (override with -T permissive Directive)\n", name);
- else if ((val = ATAErrorCount(atadev, name, cfg.fix_firmwarebug)) < 0)
- PrintOut(LOG_INFO, "Device: %s, no SMART Error log; remove -l error Directive from smartd.conf\n", name);
- else {
- cfg.errorlog = true;
- state.ataerrorcount=val;
+ // capability check: self-test and offline data collection status
+ if (cfg.offlinests || cfg.selfteststs) {
+ if (!(cfg.permissive || (smart_val_ok && state.smartval.offline_data_collection_capability))) {
+ if (cfg.offlinests)
+ PrintOut(LOG_INFO, "Device: %s, no SMART Offline Data Collection capability, ignoring -l offlinests (override with -T permissive)\n", name);
+ if (cfg.selfteststs)
+ PrintOut(LOG_INFO, "Device: %s, no SMART Self-test capability, ignoring -l selfteststs (override with -T permissive)\n", name);
+ cfg.offlinests = cfg.selfteststs = false;
}
}
-
+
// capabilities check -- does it support powermode?
if (cfg.powermode) {
int powermode = ataCheckPowerMode(atadev);
}
}
+ // Apply ATA settings
+ std::string msg;
+
+ if (cfg.set_aam)
+ format_set_result_msg(msg, "AAM", (cfg.set_aam > 0 ?
+ ata_set_features(atadev, ATA_ENABLE_AAM, cfg.set_aam-1) :
+ ata_set_features(atadev, ATA_DISABLE_AAM)), cfg.set_aam, true);
+
+ if (cfg.set_apm)
+ format_set_result_msg(msg, "APM", (cfg.set_apm > 0 ?
+ ata_set_features(atadev, ATA_ENABLE_APM, cfg.set_apm-1) :
+ ata_set_features(atadev, ATA_DISABLE_APM)), cfg.set_apm, true);
+
+ if (cfg.set_lookahead)
+ format_set_result_msg(msg, "Rd-ahead", ata_set_features(atadev,
+ (cfg.set_lookahead > 0 ? ATA_ENABLE_READ_LOOK_AHEAD : ATA_DISABLE_READ_LOOK_AHEAD)),
+ cfg.set_lookahead);
+
+ if (cfg.set_wcache)
+ format_set_result_msg(msg, "Wr-cache", ata_set_features(atadev,
+ (cfg.set_wcache > 0? ATA_ENABLE_WRITE_CACHE : ATA_DISABLE_WRITE_CACHE)), cfg.set_wcache);
+
+ if (cfg.set_security_freeze)
+ format_set_result_msg(msg, "Security freeze",
+ ata_nodata_command(atadev, ATA_SECURITY_FREEZE_LOCK));
+
+ if (cfg.set_standby)
+ format_set_result_msg(msg, "Standby",
+ ata_nodata_command(atadev, ATA_IDLE, cfg.set_standby-1), cfg.set_standby, true);
+
+ // Report as one log entry
+ if (!msg.empty())
+ PrintOut(LOG_INFO, "Device: %s, ATA settings applied: %s\n", name, msg.c_str());
+
+ // set SCT Error Recovery Control if requested
+ if (cfg.sct_erc_set) {
+ if (!isSCTErrorRecoveryControlCapable(&drive))
+ PrintOut(LOG_INFO, "Device: %s, no SCT Error Recovery Control support, ignoring -l scterc\n",
+ name);
+ else if ( ataSetSCTErrorRecoveryControltime(atadev, 1, cfg.sct_erc_readtime )
+ || ataSetSCTErrorRecoveryControltime(atadev, 2, cfg.sct_erc_writetime))
+ PrintOut(LOG_INFO, "Device: %s, set of SCT Error Recovery Control failed\n", name);
+ else
+ PrintOut(LOG_INFO, "Device: %s, SCT Error Recovery Control set to: Read: %u, Write: %u\n",
+ name, cfg.sct_erc_readtime, cfg.sct_erc_writetime);
+ }
+
// If no tests available or selected, return
- if (!(cfg.errorlog || cfg.selftest || cfg.smartcheck ||
- cfg.usagefailed || cfg.prefail || cfg.usage ||
- cfg.tempdiff || cfg.tempinfo || cfg.tempcrit )) {
+ if (!( cfg.smartcheck || cfg.selftest
+ || cfg.errorlog || cfg.xerrorlog
+ || cfg.offlinests || cfg.selfteststs
+ || cfg.usagefailed || cfg.prefail || cfg.usage
+ || cfg.tempdiff || cfg.tempinfo || cfg.tempcrit)) {
CloseDevice(atadev, name);
return 3;
}
// tell user we are registering device
PrintOut(LOG_INFO,"Device: %s, is SMART capable. Adding to \"monitor\" list.\n",name);
- // record number of device, type of device, increment device count
- if (cfg.dev_type.empty())
- cfg.dev_type = "ata";
-
// close file descriptor
CloseDevice(atadev, name);
if (!state_path_prefix.empty() || !attrlog_path_prefix.empty()) {
// Build file name for state file
- char model[40+1], serial[20+1];
- format_ata_string(model, drive.model, sizeof(model)-1, fix_swapped_id);
- format_ata_string(serial, drive.serial_no, sizeof(serial)-1, fix_swapped_id);
std::replace_if(model, model+strlen(model), not_allowed_in_filename, '_');
std::replace_if(serial, serial+strlen(serial), not_allowed_in_filename, '_');
if (!state_path_prefix.empty()) {
cfg.attrlog_file = strprintf("%s%s-%s.ata.csv", attrlog_path_prefix.c_str(), model, serial);
}
- // Start self-test regex check now if time was not read from state file
- if (!cfg.test_regex.empty() && !state.scheduled_test_next_check)
- state.scheduled_test_next_check = time(0);
+ finish_device_scan(cfg, state);
+
+ return 0;
+}
+
+// on success, return 0. On failure, return >0. Never return <0,
+// please.
+static int SCSIDeviceScan(dev_config & cfg, dev_state & state, scsi_device * scsidev)
+{
+ int k, err, req_len, avail_len, version, len;
+ const char *device = cfg.name.c_str();
+ struct scsi_iec_mode_page iec;
+ UINT8 tBuf[64];
+ UINT8 inqBuf[96];
+ UINT8 vpdBuf[252];
+ char lu_id[64], serial[256], vendor[40], model[40];
+
+ // Device must be open
+ memset(inqBuf, 0, 96);
+ req_len = 36;
+ if ((err = scsiStdInquiry(scsidev, inqBuf, req_len))) {
+ /* Marvell controllers fail on a 36 bytes StdInquiry, but 64 suffices */
+ req_len = 64;
+ if ((err = scsiStdInquiry(scsidev, inqBuf, req_len))) {
+ PrintOut(LOG_INFO, "Device: %s, Both 36 and 64 byte INQUIRY failed; "
+ "skip device\n", device);
+ return 2;
+ }
+ }
+ version = (inqBuf[2] & 0x7f); /* Accept old ISO/IEC 9316:1995 variants */
+
+ avail_len = inqBuf[4] + 5;
+ len = (avail_len < req_len) ? avail_len : req_len;
+ if (len < 36) {
+ PrintOut(LOG_INFO, "Device: %s, INQUIRY response less than 36 bytes; "
+ "skip device\n", device);
+ return 2;
+ }
+
+ int pdt = inqBuf[0] & 0x1f;
+
+ if (! ((0 == pdt) || (4 == pdt) || (5 == pdt) || (7 == pdt) ||
+ (0xe == pdt))) {
+ PrintOut(LOG_INFO, "Device: %s, not a disk like device [PDT=0x%x], "
+ "skip\n", device, pdt);
+ return 2;
+ }
- return 0;
-}
+ if (supported_vpd_pages_p) {
+ delete supported_vpd_pages_p;
+ supported_vpd_pages_p = NULL;
+ }
+ supported_vpd_pages_p = new supported_vpd_pages(scsidev);
-// on success, return 0. On failure, return >0. Never return <0,
-// please.
-static int SCSIDeviceScan(dev_config & cfg, dev_state & state, scsi_device * scsidev)
-{
- int k, err;
- const char *device = cfg.name.c_str();
- struct scsi_iec_mode_page iec;
- UINT8 tBuf[64];
+ lu_id[0] = '\0';
+ if ((version >= 0x3) && (version < 0x8)) {
+ /* SPC to SPC-5 */
+ if (0 == scsiInquiryVpd(scsidev, SCSI_VPD_DEVICE_IDENTIFICATION,
+ vpdBuf, sizeof(vpdBuf))) {
+ len = vpdBuf[3];
+ scsi_decode_lu_dev_id(vpdBuf + 4, len, lu_id, sizeof(lu_id), NULL);
+ }
+ }
+ serial[0] = '\0';
+ if (0 == scsiInquiryVpd(scsidev, SCSI_VPD_UNIT_SERIAL_NUMBER,
+ vpdBuf, sizeof(vpdBuf))) {
+ len = vpdBuf[3];
+ vpdBuf[4 + len] = '\0';
+ scsi_format_id_string(serial, (const unsigned char *)&vpdBuf[4], len);
+ }
- // Device must be open
+ unsigned int lb_size;
+ char si_str[64];
+ uint64_t capacity = scsiGetSize(scsidev, &lb_size, NULL);
+
+ if (capacity)
+ format_capacity(si_str, sizeof(si_str), capacity);
+ else
+ si_str[0] = '\0';
+
+ // Format device id string for warning emails
+ cfg.dev_idinfo = strprintf("[%.8s %.16s %.4s]%s%s%s%s%s%s",
+ (char *)&inqBuf[8], (char *)&inqBuf[16], (char *)&inqBuf[32],
+ (lu_id[0] ? ", lu id: " : ""), (lu_id[0] ? lu_id : ""),
+ (serial[0] ? ", S/N: " : ""), (serial[0] ? serial : ""),
+ (si_str[0] ? ", " : ""), (si_str[0] ? si_str : ""));
+
+ // format "model" string
+ scsi_format_id_string(vendor, (const unsigned char *)&inqBuf[8], 8);
+ scsi_format_id_string(model, (const unsigned char *)&inqBuf[16], 16);
+ PrintOut(LOG_INFO, "Device: %s, %s\n", device, cfg.dev_idinfo.c_str());
// check that device is ready for commands. IE stores its stuff on
// the media.
case IE_LPAGE:
state.SmartPageSupported = 1;
break;
+ case READ_ERROR_COUNTER_LPAGE:
+ state.ReadECounterPageSupported = 1;
+ break;
+ case WRITE_ERROR_COUNTER_LPAGE:
+ state.WriteECounterPageSupported = 1;
+ break;
+ case VERIFY_ERROR_COUNTER_LPAGE:
+ state.VerifyECounterPageSupported = 1;
+ break;
+ case NON_MEDIUM_ERROR_LPAGE:
+ state.NonMediumErrorPageSupported = 1;
+ break;
default:
break;
}
}
}
- // record type of device
- if (cfg.dev_type.empty())
- cfg.dev_type = "scsi";
-
// Check if scsiCheckIE() is going to work
{
UINT8 asc = 0;
PrintOut(LOG_INFO, "Device: %s, unexpectedly failed to read SMART values\n", device);
state.SuppressReport = 1;
if (cfg.tempdiff || cfg.tempinfo || cfg.tempcrit) {
- PrintOut(LOG_CRIT, "Device: %s, can't monitor Temperature, ignoring -W Directive\n", device);
+ PrintOut(LOG_INFO, "Device: %s, can't monitor Temperature, ignoring -W %d,%d,%d\n",
+ device, cfg.tempdiff, cfg.tempinfo, cfg.tempcrit);
cfg.tempdiff = cfg.tempinfo = cfg.tempcrit = 0;
}
}
// tell user we are registering device
PrintOut(LOG_INFO, "Device: %s, is SMART capable. Adding to \"monitor\" list.\n", device);
- // TODO: Build file name for state file
- if (!state_path_prefix.empty()) {
- PrintOut(LOG_INFO, "Device: %s, persistence not yet supported for SCSI; ignoring -s option.\n", device);
- }
- // TODO: Build file name for attribute log file
- if (!attrlog_path_prefix.empty()) {
- PrintOut(LOG_INFO, "Device: %s, attribute log not yet supported for SCSI; ignoring -A option.\n", device);
- }
+ // Make sure that init_standby_check() ignores SCSI devices
+ cfg.offlinests_ns = cfg.selfteststs_ns = false;
// close file descriptor
CloseDevice(scsidev, device);
- // Start self-test regex check now if time was not read from state file
- if (!cfg.test_regex.empty() && !state.scheduled_test_next_check)
- state.scheduled_test_next_check = time(0);
+ if (!state_path_prefix.empty() || !attrlog_path_prefix.empty()) {
+ // Build file name for state file
+ std::replace_if(model, model+strlen(model), not_allowed_in_filename, '_');
+ std::replace_if(serial, serial+strlen(serial), not_allowed_in_filename, '_');
+ if (!state_path_prefix.empty()) {
+ cfg.state_file = strprintf("%s%s-%s-%s.scsi.state", state_path_prefix.c_str(), vendor, model, serial);
+ // Read previous state
+ if (read_dev_state(cfg.state_file.c_str(), state)) {
+ PrintOut(LOG_INFO, "Device: %s, state read from %s\n", device, cfg.state_file.c_str());
+ // Copy ATA attribute values to temp state
+ state.update_temp_state();
+ }
+ }
+ if (!attrlog_path_prefix.empty())
+ cfg.attrlog_file = strprintf("%s%s-%s-%s.scsi.csv", attrlog_path_prefix.c_str(), vendor, model, serial);
+ }
+
+ finish_device_scan(cfg, state);
return 0;
}
if (newi<0)
// command failed
MailWarning(cfg, state, 8, "Device: %s, Read SMART Self-Test Log Failed", name);
- else {
+ else {
+ reset_warning_mail(cfg, state, 8, "Read SMART Self-Test Log worked again");
+
// old and new error counts
int oldc=state.selflogcount;
int newc=SELFTEST_ERRORCOUNT(newi);
MailWarning(cfg, state, 3, "Device: %s, Self-Test Log error count increased from %d to %d",
name, oldc, newc);
state.must_write = true;
- } else if (oldh!=newh) {
+ }
+ else if (newc > 0 && oldh != newh) {
// more recent error
// a 'more recent' error might actually be a smaller hour number,
// if the hour number has wrapped.
// new failure.
PrintOut(LOG_CRIT, "Device: %s, new Self-Test Log error at hour timestamp %d\n",
name, newh);
- MailWarning(cfg, state, 3, "Device: %s, new Self-Test Log error at hour timestamp %d\n",
+ MailWarning(cfg, state, 3, "Device: %s, new Self-Test Log error at hour timestamp %d",
name, newh);
state.must_write = true;
}
-
+
+ // Print info if error entries have disappeared
+ // or newer successful successful extended self-test exits
+ if (oldc > newc) {
+ PrintOut(LOG_INFO, "Device: %s, Self-Test Log error count decreased from %d to %d\n",
+ name, oldc, newc);
+ if (newc == 0)
+ reset_warning_mail(cfg, state, 3, "Self-Test Log does no longer report errors");
+ }
+
// Needed since self-test error count may DECREASE. Hour might
// also have changed.
state.selflogcount= newc;
// Test types, ordered by priority.
static const char test_type_chars[] = "LncrSCO";
-const unsigned num_test_types = sizeof(test_type_chars)-1;
+static const unsigned num_test_types = sizeof(test_type_chars)-1;
// returns test type if time to do test of type testtype,
// 0 if not time to do test.
// If currently running a self-test, do not interrupt it to start another.
if (15==(data.self_test_exec_status >> 4)) {
- if (cfg.fix_firmwarebug == FIX_SAMSUNG3 && data.self_test_exec_status == 0xf0) {
+ if (cfg.firmwarebugs.is_set(BUG_SAMSUNG3) && data.self_test_exec_status == 0xf0) {
PrintOut(LOG_INFO, "Device: %s, will not skip scheduled %sTest "
"despite unclear Self-Test byte (SAMSUNG Firmware bug).\n", name, testname);
} else {
if (dotest == SELECTIVE_SELF_TEST) {
// Set test span
- ata_selective_selftest_args selargs;
+ ata_selective_selftest_args selargs, prev_args;
selargs.num_spans = 1;
selargs.span[0].mode = mode;
- if (ataWriteSelectiveSelfTestLog(device, selargs, &data, state.num_sectors)) {
+ prev_args.num_spans = 1;
+ prev_args.span[0].start = state.selective_test_last_start;
+ prev_args.span[0].end = state.selective_test_last_end;
+ if (ataWriteSelectiveSelfTestLog(device, selargs, &data, state.num_sectors, &prev_args)) {
PrintOut(LOG_CRIT, "Device: %s, prepare %sTest failed\n", name, testname);
return 1;
}
uint64_t start = selargs.span[0].start, end = selargs.span[0].end;
- PrintOut(LOG_INFO, "Device: %s, %s test span at LBA %"PRIu64" - %"PRIu64" (%"PRIu64" sectors, %u%% - %u%% of disk).\n",
+ PrintOut(LOG_INFO, "Device: %s, %s test span at LBA %" PRIu64 " - %" PRIu64 " (%" PRIu64 " sectors, %u%% - %u%% of disk).\n",
name, (selargs.span[0].mode == SEL_NEXT ? "next" : "redo"),
start, end, end - start + 1,
(unsigned)((100 * start + state.num_sectors/2) / state.num_sectors),
(unsigned)((100 * end + state.num_sectors/2) / state.num_sectors));
+ state.selective_test_last_start = start;
+ state.selective_test_last_end = end;
}
// execute the test, and return status
return retval;
}
- if (testtype != 'O')
- // Log next self-test execution status
- state.smartval.self_test_exec_status = 0xff;
+ // Report recent test start to do_disable_standby_check()
+ // and force log of next test status
+ if (testtype == 'O')
+ state.offline_started = true;
+ else
+ state.selftest_started = true;
PrintOut(LOG_INFO, "Device: %s, starting scheduled %sTest.\n", name, testname);
return 0;
// No report if no sectors pending.
uint64_t rawval = ata_get_attr_raw_value(smartval.vendor_attributes[i], cfg.attribute_defs);
- if (rawval == 0)
+ if (rawval == 0) {
+ reset_warning_mail(cfg, state, mailtype, "No more %s", msg);
return;
+ }
// If attribute is not reset, report only sector count increases.
uint64_t prev_rawval = ata_get_attr_raw_value(state.smartval.vendor_attributes[i], cfg.attribute_defs);
return;
// Format message.
- std::string s = strprintf("Device: %s, %"PRId64" %s", cfg.name.c_str(), rawval, msg);
+ std::string s = strprintf("Device: %s, %" PRId64 " %s", cfg.name.c_str(), rawval, msg);
if (prev_rawval > 0 && rawval != prev_rawval)
- s += strprintf(" (changed %+"PRId64")", rawval - prev_rawval);
+ s += strprintf(" (changed %+" PRId64 ")", rawval - prev_rawval);
PrintOut(LOG_CRIT, "%s\n", s.c_str());
- MailWarning(cfg, state, mailtype, "%s\n", s.c_str());
+ MailWarning(cfg, state, mailtype, "%s", s.c_str());
state.must_write = true;
}
// Format Temperature value
-static const char * fmt_temp(unsigned char x, char * buf)
+static const char * fmt_temp(unsigned char x, char (& buf)[20])
{
if (!x) // unset
- strcpy(buf, "??");
- else
- sprintf(buf, "%u", x);
+ return "??";
+ snprintf(buf, sizeof(buf), "%u", x);
return buf;
}
if (cfg.tempcrit && currtemp >= cfg.tempcrit) {
PrintOut(LOG_CRIT, "Device: %s, Temperature %u Celsius reached critical limit of %u Celsius (Min/Max %s%s/%u%s)\n",
cfg.name.c_str(), currtemp, cfg.tempcrit, fmt_temp(state.tempmin, buf), minchg, state.tempmax, maxchg);
- MailWarning(cfg, state, 12, "Device: %s, Temperature %d Celsius reached critical limit of %u Celsius (Min/Max %s%s/%u%s)\n",
+ MailWarning(cfg, state, 12, "Device: %s, Temperature %d Celsius reached critical limit of %u Celsius (Min/Max %s%s/%u%s)",
cfg.name.c_str(), currtemp, cfg.tempcrit, fmt_temp(state.tempmin, buf), minchg, state.tempmax, maxchg);
}
else if (cfg.tempinfo && currtemp >= cfg.tempinfo) {
PrintOut(LOG_INFO, "Device: %s, Temperature %u Celsius reached limit of %u Celsius (Min/Max %s%s/%u%s)\n",
cfg.name.c_str(), currtemp, cfg.tempinfo, fmt_temp(state.tempmin, buf), minchg, state.tempmax, maxchg);
}
+ else if (cfg.tempcrit) {
+ unsigned char limit = (cfg.tempinfo ? cfg.tempinfo : cfg.tempcrit-5);
+ if (currtemp < limit)
+ reset_warning_mail(cfg, state, 12, "Temperature %u Celsius dropped below %u Celsius", currtemp, limit);
+ }
}
// Check normalized and raw attribute values.
static void check_attribute(const dev_config & cfg, dev_state & state,
const ata_smart_attribute & attr,
const ata_smart_attribute & prev,
- const ata_smart_threshold_entry & thre)
+ int attridx,
+ const ata_smart_threshold_entry * thresholds)
{
// Check attribute and threshold
- ata_attr_state attrstate = ata_get_attr_state(attr, thre, cfg.attribute_defs);
+ ata_attr_state attrstate = ata_get_attr_state(attr, attridx, thresholds, cfg.attribute_defs);
if (attrstate == ATTRSTATE_NON_EXISTING)
return;
// If requested, check for usage attributes that have failed.
if ( cfg.usagefailed && attrstate == ATTRSTATE_FAILED_NOW
&& !cfg.monitor_attr_flags.is_set(attr.id, MONITOR_IGN_FAILUSE)) {
- std::string attrname = ata_get_smart_attr_name(attr.id, cfg.attribute_defs);
+ std::string attrname = ata_get_smart_attr_name(attr.id, cfg.attribute_defs, cfg.dev_rpm);
PrintOut(LOG_CRIT, "Device: %s, Failed SMART usage Attribute: %d %s.\n", cfg.name.c_str(), attr.id, attrname.c_str());
MailWarning(cfg, state, 2, "Device: %s, Failed SMART usage Attribute: %d %s.", cfg.name.c_str(), attr.id, attrname.c_str());
state.must_write = true;
return;
// Issue warning if they don't have the same ID in all structures.
- if (attr.id != prev.id || attrstate == ATTRSTATE_BAD_THRESHOLD) {
- PrintOut(LOG_INFO,"Device: %s, same Attribute has different ID numbers: %d = %d = %d\n",
- cfg.name.c_str(), attr.id, prev.id, thre.id);
+ if (attr.id != prev.id) {
+ PrintOut(LOG_INFO,"Device: %s, same Attribute has different ID numbers: %d = %d\n",
+ cfg.name.c_str(), attr.id, prev.id);
return;
}
// Format message
std::string msg = strprintf("Device: %s, SMART %s Attribute: %d %s changed from %s to %s",
cfg.name.c_str(), (prefail ? "Prefailure" : "Usage"), attr.id,
- ata_get_smart_attr_name(attr.id, cfg.attribute_defs).c_str(),
+ ata_get_smart_attr_name(attr.id, cfg.attribute_defs, cfg.dev_rpm).c_str(),
prevstr.c_str(), currstr.c_str());
// Report this change as critical ?
}
-static int ATACheckDevice(const dev_config & cfg, dev_state & state, ata_device * atadev, bool allow_selftests)
+static int ATACheckDevice(const dev_config & cfg, dev_state & state, ata_device * atadev,
+ bool firstpass, bool allow_selftests)
{
const char * name = cfg.name.c_str();
PrintOut(LOG_INFO, "Device: %s, open() failed: %s\n", name, atadev->get_errmsg());
MailWarning(cfg, state, 9, "Device: %s, unable to open device", name);
return 1;
- } else if (debugmode)
+ }
+ if (debugmode)
PrintOut(LOG_INFO,"Device: %s, opened ATA device\n", name);
+ reset_warning_mail(cfg, state, 9, "open device worked again");
// user may have requested (with the -n Directive) to leave the disk
// alone if it is in idle or sleeping mode. In this case check the
// Check everything that depends upon SMART Data (eg, Attribute values)
if ( cfg.usagefailed || cfg.prefail || cfg.usage
|| cfg.curr_pending_id || cfg.offl_pending_id
- || cfg.tempdiff || cfg.tempinfo || cfg.tempcrit || cfg.selftest) {
+ || cfg.tempdiff || cfg.tempinfo || cfg.tempcrit
+ || cfg.selftest || cfg.offlinests || cfg.selfteststs) {
// Read current attribute values.
ata_smart_values curval;
state.must_write = true;
}
else {
+ reset_warning_mail(cfg, state, 6, "read SMART Attribute Data worked again");
+
// look for current or offline pending sectors
if (cfg.curr_pending_id)
check_pending(cfg, state, cfg.curr_pending_id, cfg.curr_pending_incr, curval, 10,
if (cfg.tempdiff || cfg.tempinfo || cfg.tempcrit)
CheckTemperature(cfg, state, ata_return_temperature_value(&curval, cfg.attribute_defs), 0);
+ // look for failed usage attributes, or track usage or prefail attributes
if (cfg.usagefailed || cfg.prefail || cfg.usage) {
-
- // look for failed usage attributes, or track usage or prefail attributes
for (int i = 0; i < NUMBER_ATA_SMART_ATTRIBUTES; i++) {
check_attribute(cfg, state,
curval.vendor_attributes[i],
state.smartval.vendor_attributes[i],
- state.smartthres.thres_entries[i]);
+ i, state.smartthres.thres_entries);
}
+ }
- if (cfg.selftest) {
- // Log changes of self-test execution status
- if ( curval.self_test_exec_status != state.smartval.self_test_exec_status
- || (!allow_selftests && curval.self_test_exec_status != 0x00) )
- log_self_test_exec_status(name, curval.self_test_exec_status);
- }
+ // Log changes of offline data collection status
+ if (cfg.offlinests) {
+ if ( curval.offline_data_collection_status
+ != state.smartval.offline_data_collection_status
+ || state.offline_started // test was started in previous call
+ || (firstpass && (debugmode || (curval.offline_data_collection_status & 0x7d))))
+ log_offline_data_coll_status(name, curval.offline_data_collection_status);
+ }
- // Save the new values into *drive for the next time around
- state.smartval = curval;
+ // Log changes of self-test execution status
+ if (cfg.selfteststs) {
+ if ( curval.self_test_exec_status != state.smartval.self_test_exec_status
+ || state.selftest_started // test was started in previous call
+ || (firstpass && (debugmode || curval.self_test_exec_status != 0x00)))
+ log_self_test_exec_status(name, curval.self_test_exec_status);
}
+
+ // Save the new values for the next time around
+ state.smartval = curval;
}
}
+ state.offline_started = state.selftest_started = false;
// check if number of selftest errors has increased (note: may also DECREASE)
if (cfg.selftest)
- CheckSelfTestLogs(cfg, state, SelfTestErrorCount(atadev, name, cfg.fix_firmwarebug));
+ CheckSelfTestLogs(cfg, state, SelfTestErrorCount(atadev, name, cfg.firmwarebugs));
// check if number of ATA errors has increased
- if (cfg.errorlog) {
+ if (cfg.errorlog || cfg.xerrorlog) {
- int newc, oldc= state.ataerrorcount;
+ int errcnt1 = -1, errcnt2 = -1;
+ if (cfg.errorlog)
+ errcnt1 = read_ata_error_count(atadev, name, cfg.firmwarebugs, false);
+ if (cfg.xerrorlog)
+ errcnt2 = read_ata_error_count(atadev, name, cfg.firmwarebugs, true);
- // new number of errors
- newc = ATAErrorCount(atadev, name, cfg.fix_firmwarebug);
+ // new number of errors is max of both logs
+ int newc = (errcnt1 >= errcnt2 ? errcnt1 : errcnt2);
// did command fail?
if (newc<0)
MailWarning(cfg, state, 7, "Device: %s, Read SMART Error Log Failed", name);
// has error count increased?
+ int oldc = state.ataerrorcount;
if (newc>oldc){
PrintOut(LOG_CRIT, "Device: %s, ATA error count increased from %d to %d\n",
name, oldc, newc);
name, oldc, newc);
state.must_write = true;
}
-
- // this last line is probably not needed, count always increases
+
if (newc>=0)
state.ataerrorcount=newc;
}
UINT8 asc, ascq;
UINT8 currenttemp;
UINT8 triptemp;
+ UINT8 tBuf[252];
const char * name = cfg.name.c_str();
const char *cp;
return 1;
} else if (debugmode)
PrintOut(LOG_INFO,"Device: %s, opened SCSI device\n", name);
+ reset_warning_mail(cfg, state, 9, "open device worked again");
currenttemp = 0;
asc = 0;
ascq = 0;
if (cp) {
PrintOut(LOG_CRIT, "Device: %s, SMART Failure: %s\n", name, cp);
MailWarning(cfg, state, 1,"Device: %s, SMART Failure: %s", name, cp);
+ } else if (asc == 4 && ascq == 9) {
+ PrintOut(LOG_INFO,"Device: %s, self-test in progress\n", name);
} else if (debugmode)
PrintOut(LOG_INFO,"Device: %s, non-SMART asc,ascq: %d,%d\n",
name, (int)asc, (int)ascq);
PrintOut(LOG_INFO,"Device: %s, SMART health: passed\n", name);
// check temperature limits
- if (cfg.tempdiff || cfg.tempinfo || cfg.tempcrit)
+ if (cfg.tempdiff || cfg.tempinfo || cfg.tempcrit || !cfg.attrlog_file.empty())
CheckTemperature(cfg, state, currenttemp, triptemp);
// check if number of selftest errors has increased (note: may also DECREASE)
if (testtype)
DoSCSISelfTest(cfg, state, scsidev, testtype);
}
+ if (!cfg.attrlog_file.empty()){
+ // saving error counters to state
+ if (state.ReadECounterPageSupported && (0 == scsiLogSense(scsidev,
+ READ_ERROR_COUNTER_LPAGE, 0, tBuf, sizeof(tBuf), 0))) {
+ scsiDecodeErrCounterPage(tBuf, &state.scsi_error_counters[0].errCounter);
+ state.scsi_error_counters[0].found=1;
+ }
+ if (state.WriteECounterPageSupported && (0 == scsiLogSense(scsidev,
+ WRITE_ERROR_COUNTER_LPAGE, 0, tBuf, sizeof(tBuf), 0))) {
+ scsiDecodeErrCounterPage(tBuf, &state.scsi_error_counters[1].errCounter);
+ state.scsi_error_counters[1].found=1;
+ }
+ if (state.VerifyECounterPageSupported && (0 == scsiLogSense(scsidev,
+ VERIFY_ERROR_COUNTER_LPAGE, 0, tBuf, sizeof(tBuf), 0))) {
+ scsiDecodeErrCounterPage(tBuf, &state.scsi_error_counters[2].errCounter);
+ state.scsi_error_counters[2].found=1;
+ }
+ if (state.NonMediumErrorPageSupported && (0 == scsiLogSense(scsidev,
+ NON_MEDIUM_ERROR_LPAGE, 0, tBuf, sizeof(tBuf), 0))) {
+ scsiDecodeNonMediumErrPage(tBuf, &state.scsi_nonmedium_error.nme);
+ state.scsi_nonmedium_error.found=1;
+ }
+ }
CloseDevice(scsidev, name);
return 0;
}
+// 0=not used, 1=not disabled, 2=disable rejected by OS, 3=disabled
+static int standby_disable_state = 0;
+
+static void init_disable_standby_check(dev_config_vector & configs)
+{
+ // Check for '-l offlinests,ns' or '-l selfteststs,ns' directives
+ bool sts1 = false, sts2 = false;
+ for (unsigned i = 0; i < configs.size() && !(sts1 || sts2); i++) {
+ const dev_config & cfg = configs.at(i);
+ if (cfg.offlinests_ns)
+ sts1 = true;
+ if (cfg.selfteststs_ns)
+ sts2 = true;
+ }
+
+ // Check for support of disable auto standby
+ // Reenable standby if smartd.conf was reread
+ if (sts1 || sts2 || standby_disable_state == 3) {
+ if (!smi()->disable_system_auto_standby(false)) {
+ if (standby_disable_state == 3)
+ PrintOut(LOG_CRIT, "System auto standby enable failed: %s\n", smi()->get_errmsg());
+ if (sts1 || sts2) {
+ PrintOut(LOG_INFO, "Disable auto standby not supported, ignoring ',ns' from %s%s%s\n",
+ (sts1 ? "-l offlinests,ns" : ""), (sts1 && sts2 ? " and " : ""), (sts2 ? "-l selfteststs,ns" : ""));
+ sts1 = sts2 = false;
+ }
+ }
+ }
+
+ standby_disable_state = (sts1 || sts2 ? 1 : 0);
+}
+
+static void do_disable_standby_check(const dev_config_vector & configs, const dev_state_vector & states)
+{
+ if (!standby_disable_state)
+ return;
+
+ // Check for just started or still running self-tests
+ bool running = false;
+ for (unsigned i = 0; i < configs.size() && !running; i++) {
+ const dev_config & cfg = configs.at(i); const dev_state & state = states.at(i);
+
+ if ( ( cfg.offlinests_ns
+ && (state.offline_started ||
+ is_offl_coll_in_progress(state.smartval.offline_data_collection_status)))
+ || ( cfg.selfteststs_ns
+ && (state.selftest_started ||
+ is_self_test_in_progress(state.smartval.self_test_exec_status))) )
+ running = true;
+ // state.offline/selftest_started will be reset after next logging of test status
+ }
+
+ // Disable/enable auto standby and log state changes
+ if (!running) {
+ if (standby_disable_state != 1) {
+ if (!smi()->disable_system_auto_standby(false))
+ PrintOut(LOG_CRIT, "Self-test(s) completed, system auto standby enable failed: %s\n",
+ smi()->get_errmsg());
+ else
+ PrintOut(LOG_INFO, "Self-test(s) completed, system auto standby enabled\n");
+ standby_disable_state = 1;
+ }
+ }
+ else if (!smi()->disable_system_auto_standby(true)) {
+ if (standby_disable_state != 2) {
+ PrintOut(LOG_INFO, "Self-test(s) in progress, system auto standby disable rejected: %s\n",
+ smi()->get_errmsg());
+ standby_disable_state = 2;
+ }
+ }
+ else {
+ if (standby_disable_state != 3) {
+ PrintOut(LOG_INFO, "Self-test(s) in progress, system auto standby disabled\n");
+ standby_disable_state = 3;
+ }
+ }
+}
+
// Checks the SMART status of all ATA and SCSI devices
static void CheckDevicesOnce(const dev_config_vector & configs, dev_state_vector & states,
- smart_device_list & devices, bool allow_selftests)
+ smart_device_list & devices, bool firstpass, bool allow_selftests)
{
for (unsigned i = 0; i < configs.size(); i++) {
const dev_config & cfg = configs.at(i);
dev_state & state = states.at(i);
smart_device * dev = devices.at(i);
if (dev->is_ata())
- ATACheckDevice(cfg, state, dev->to_ata(), allow_selftests);
+ ATACheckDevice(cfg, state, dev->to_ata(), firstpass, allow_selftests);
else if (dev->is_scsi())
SCSICheckDevice(cfg, state, dev->to_scsi(), allow_selftests);
}
+
+ do_disable_standby_check(configs, states);
}
// Set if Initialize() was called
static bool is_initialized = false;
// Does initialization right after fork to daemon mode
-void Initialize(time_t *wakeuptime){
-
+static void Initialize(time_t *wakeuptime)
+{
// Call Goodbye() on exit
is_initialized = true;
}
// sleep until we catch SIGUSR1 or have completed sleeping
- while (timenow<wakeuptime && !caughtsigUSR1 && !caughtsigHUP && !caughtsigEXIT){
+ int addtime = 0;
+ while (timenow < wakeuptime+addtime && !caughtsigUSR1 && !caughtsigHUP && !caughtsigEXIT) {
// protect user again system clock being adjusted backwards
if (wakeuptime>timenow+checktime){
}
// Exit sleep when time interval has expired or a signal is received
- sleep(wakeuptime-timenow);
+ sleep(wakeuptime+addtime-timenow);
#ifdef _WIN32
// toggle debug mode?
#endif
timenow=time(NULL);
+
+ // Actual sleep time too long?
+ if (!addtime && timenow > wakeuptime+60) {
+ if (debugmode)
+ PrintOut(LOG_INFO, "Sleep time was %d seconds too long, assuming wakeup from standby mode.\n",
+ (int)(timenow-wakeuptime));
+ // Wait another 20 seconds to avoid I/O errors during disk spin-up
+ addtime = timenow-wakeuptime+20;
+ // Use next wake-up-time if close
+ int nextcheck = checktime - addtime % checktime;
+ if (nextcheck <= 20)
+ addtime += nextcheck;
+ }
}
// if we caught a SIGUSR1 then print message and clear signal
}
// Print out a list of valid arguments for the Directive d
-void printoutvaliddirectiveargs(int priority, char d) {
-
+static void printoutvaliddirectiveargs(int priority, char d)
+{
switch (d) {
case 'n':
PrintOut(priority, "never[,N][,q], sleep[,N][,q], standby[,N][,q], idle[,N][,q]");
PrintOut(priority, "use, ignore, show, showall");
break;
case 'F':
- PrintOut(priority, "none, samsung, samsung2, samsung3");
+ PrintOut(priority, "%s", get_valid_firmwarebug_args());
+ break;
+ case 'e':
+ PrintOut(priority, "aam,[N|off], apm,[N|off], lookahead,[on|off], "
+ "security-freeze, standby,[N|off], wcache,[on|off]");
break;
}
}
// exits with an error message, or returns integer value of token
-int GetInteger(const char *arg, const char *name, const char *token, int lineno, const char *configfile,
+static int GetInteger(const char *arg, const char *name, const char *token, int lineno, const char *cfgfile,
int min, int max, char * suffix = 0)
{
// make sure argument is there
if (!arg) {
PrintOut(LOG_CRIT,"File %s line %d (drive %s): Directive: %s takes integer argument from %d to %d.\n",
- configfile, lineno, name, token, min, max);
+ cfgfile, lineno, name, token, min, max);
return -1;
}
if (!(!*endptr && min <= val && val <= max)) {
PrintOut(LOG_CRIT,"File %s line %d (drive %s): Directive: %s has argument: %s; needs integer from %d to %d.\n",
- configfile, lineno, name, token, arg, min, max);
+ cfgfile, lineno, name, token, arg, min, max);
return -1;
}
// Get 1-3 small integer(s) for '-W' directive
-int Get3Integers(const char *arg, const char *name, const char *token, int lineno, const char *configfile,
- unsigned char * val1, unsigned char * val2, unsigned char * val3){
+static int Get3Integers(const char *arg, const char *name, const char *token, int lineno, const char *cfgfile,
+ unsigned char *val1, unsigned char *val2, unsigned char *val3)
+{
unsigned v1 = 0, v2 = 0, v3 = 0;
int n1 = -1, n2 = -1, n3 = -1, len;
if (!arg) {
PrintOut(LOG_CRIT,"File %s line %d (drive %s): Directive: %s takes 1-3 integer argument(s) from 0 to 255.\n",
- configfile, lineno, name, token);
+ cfgfile, lineno, name, token);
return -1;
}
if (!( sscanf(arg, "%u%n,%u%n,%u%n", &v1, &n1, &v2, &n2, &v3, &n3) >= 1
&& (n1 == len || n2 == len || n3 == len) && v1 <= 255 && v2 <= 255 && v3 <= 255)) {
PrintOut(LOG_CRIT,"File %s line %d (drive %s): Directive: %s has argument: %s; needs 1-3 integer(s) from 0 to 255.\n",
- configfile, lineno, name, token, arg);
+ cfgfile, lineno, name, token, arg);
return -1;
}
*val1 = (unsigned char)v1; *val2 = (unsigned char)v2; *val3 = (unsigned char)v3;
}
+#ifdef _WIN32
+
+// Concatenate strtok() results if quoted with "..."
+static const char * strtok_dequote(const char * delimiters)
+{
+ const char * t = strtok(0, delimiters);
+ if (!t || t[0] != '"')
+ return t;
+
+ static std::string token;
+ token = t+1;
+ for (;;) {
+ t = strtok(0, delimiters);
+ if (!t || !*t)
+ return "\"";
+ token += ' ';
+ int len = strlen(t);
+ if (t[len-1] == '"') {
+ token += std::string(t, len-1);
+ break;
+ }
+ token += t;
+ }
+ return token.c_str();
+}
+
+#endif // _WIN32
+
+
// This function returns 1 if it has correctly parsed one token (and
// any arguments), else zero if no tokens remain. It returns -1 if an
// error was encountered.
// specify the device type
if ((arg = strtok(NULL, delim)) == NULL) {
missingarg = 1;
+ } else if (!strcmp(arg, "ignore")) {
+ cfg.ignore = true;
} else if (!strcmp(arg, "removable")) {
cfg.removable = true;
+ } else if (!strcmp(arg, "auto")) {
+ cfg.dev_type = "";
} else {
cfg.dev_type = arg;
}
break;
case 'F':
// fix firmware bug
- if ((arg = strtok(NULL, delim)) == NULL) {
+ if (!(arg = strtok(0, delim)))
missingarg = 1;
- } else if (!strcmp(arg, "none")) {
- cfg.fix_firmwarebug = FIX_NONE;
- } else if (!strcmp(arg, "samsung")) {
- cfg.fix_firmwarebug = FIX_SAMSUNG;
- } else if (!strcmp(arg, "samsung2")) {
- cfg.fix_firmwarebug = FIX_SAMSUNG2;
- } else if (!strcmp(arg, "samsung3")) {
- cfg.fix_firmwarebug = FIX_SAMSUNG3;
- } else {
+ else if (!parse_firmwarebug_def(arg, cfg.firmwarebugs))
badarg = 1;
- }
break;
case 'H':
// check SMART status
} else if (!strcmp(arg, "error")) {
// track changes in ATA error log
cfg.errorlog = true;
+ } else if (!strcmp(arg, "xerror")) {
+ // track changes in Extended Comprehensive SMART error log
+ cfg.xerrorlog = true;
+ } else if (!strcmp(arg, "offlinests")) {
+ // track changes in offline data collection status
+ cfg.offlinests = true;
+ } else if (!strcmp(arg, "offlinests,ns")) {
+ // track changes in offline data collection status, disable auto standby
+ cfg.offlinests = cfg.offlinests_ns = true;
+ } else if (!strcmp(arg, "selfteststs")) {
+ // track changes in self-test execution status
+ cfg.selfteststs = true;
+ } else if (!strcmp(arg, "selfteststs,ns")) {
+ // track changes in self-test execution status, disable auto standby
+ cfg.selfteststs = cfg.selfteststs_ns = true;
+ } else if (!strncmp(arg, "scterc,", sizeof("scterc,")-1)) {
+ // set SCT Error Recovery Control
+ unsigned rt = ~0, wt = ~0; int nc = -1;
+ sscanf(arg,"scterc,%u,%u%n", &rt, &wt, &nc);
+ if (nc == (int)strlen(arg) && rt <= 999 && wt <= 999) {
+ cfg.sct_erc_set = true;
+ cfg.sct_erc_readtime = rt;
+ cfg.sct_erc_writetime = wt;
+ }
+ else
+ badarg = 1;
} else {
badarg = 1;
}
cfg.usage = true;
cfg.selftest = true;
cfg.errorlog = true;
+ cfg.selfteststs = true;
break;
case 'o':
// automatic offline testing enable/disable
if (!cfg.emailaddress.empty())
PrintOut(LOG_INFO, "File %s line %d (drive %s): ignoring previous Address Directive -m %s\n",
configfile, lineno, name, cfg.emailaddress.c_str());
+#ifdef _WIN32
+ if ( !strcmp(arg, "msgbox") || !strcmp(arg, "sysmsgbox")
+ || str_starts_with(arg, "msgbox,") || str_starts_with(arg, "sysmsgbox,")) {
+ cfg.emailaddress = "console";
+ const char * arg2 = strchr(arg, ',');
+ if (arg2)
+ cfg.emailaddress += arg2;
+ PrintOut(LOG_INFO, "File %s line %d (drive %s): Deprecated -m %s changed to -m %s\n",
+ configfile, lineno, name, arg, cfg.emailaddress.c_str());
+ }
+ else
+#endif
cfg.emailaddress = arg;
}
break;
cfg.emailtest = 1;
else if (!strcmp(arg, "exec")) {
// Get the next argument (the command line)
- if (!(arg = strtok(NULL, delim))) {
+#ifdef _WIN32
+ // Allow "/path name/with spaces/..." on Windows
+ arg = strtok_dequote(delim);
+ if (arg && arg[0] == '"') {
+ PrintOut(LOG_CRIT, "File %s line %d (drive %s): Directive %s 'exec' argument: missing closing quote\n",
+ configfile, lineno, name, token);
+ return -1;
+ }
+#else
+ arg = strtok(0, delim);
+#endif
+ if (!arg) {
PrintOut(LOG_CRIT, "File %s line %d (drive %s): Directive %s 'exec' argument must be followed by executable path.\n",
configfile, lineno, name, token);
return -1;
badarg = 1;
}
break;
+
+ case 'e':
+ // Various ATA settings
+ if (!(arg = strtok(NULL, delim))) {
+ missingarg = true;
+ }
+ else {
+ char arg2[16+1]; unsigned val;
+ int n1 = -1, n2 = -1, n3 = -1, len = strlen(arg);
+ if (sscanf(arg, "%16[^,=]%n%*[,=]%n%u%n", arg2, &n1, &n2, &val, &n3) >= 1
+ && (n1 == len || n2 > 0)) {
+ bool on = (n2 > 0 && !strcmp(arg+n2, "on"));
+ bool off = (n2 > 0 && !strcmp(arg+n2, "off"));
+ if (n3 != len)
+ val = ~0U;
+
+ if (!strcmp(arg2, "aam")) {
+ if (off)
+ cfg.set_aam = -1;
+ else if (val <= 254)
+ cfg.set_aam = val + 1;
+ else
+ badarg = true;
+ }
+ else if (!strcmp(arg2, "apm")) {
+ if (off)
+ cfg.set_apm = -1;
+ else if (1 <= val && val <= 254)
+ cfg.set_apm = val + 1;
+ else
+ badarg = true;
+ }
+ else if (!strcmp(arg2, "lookahead")) {
+ if (off)
+ cfg.set_lookahead = -1;
+ else if (on)
+ cfg.set_lookahead = 1;
+ else
+ badarg = true;
+ }
+ else if (!strcmp(arg, "security-freeze")) {
+ cfg.set_security_freeze = true;
+ }
+ else if (!strcmp(arg2, "standby")) {
+ if (off)
+ cfg.set_standby = 0 + 1;
+ else if (val <= 255)
+ cfg.set_standby = val + 1;
+ else
+ badarg = true;
+ }
+ else if (!strcmp(arg2, "wcache")) {
+ if (off)
+ cfg.set_wcache = -1;
+ else if (on)
+ cfg.set_wcache = 1;
+ else
+ badarg = true;
+ }
+ else
+ badarg = true;
+ }
+ else
+ badarg = true;
+ }
+ break;
+
default:
// Directive not recognized
PrintOut(LOG_CRIT,"File %s line %d (drive %s): unknown Directive: %s\n",
//
// Return values are:
// 1: parsed a normal line
-// 0: found comment or blank line
+// 0: found DEFAULT setting or comment or blank line
// -1: found SCANDIRECTIVE line
// -2: found an error
//
// Note: this routine modifies *line from the caller!
-static int ParseConfigLine(dev_config_vector & conf_entries, int entry, int lineno, /*const*/ char * line)
+static int ParseConfigLine(dev_config_vector & conf_entries, dev_config & default_conf, int lineno, /*const*/ char * line)
{
- char *token=NULL;
- char *name=NULL;
const char *delim = " \n\t";
- int devscan=0;
// get first token: device name. If a comment, skip line
- if (!(name=strtok(line,delim)) || *name=='#') {
+ const char * name = strtok(line, delim);
+ if (!name || *name == '#')
return 0;
- }
- // Have we detected the SCANDIRECTIVE directive?
- if (!strcmp(SCANDIRECTIVE,name)){
- devscan=1;
- if (entry) {
- PrintOut(LOG_INFO,"Scan Directive %s (line %d) must be the first entry in %s\n",name, lineno, configfile);
- return -2;
- }
+ // Check device name for DEFAULT or DEVICESCAN
+ int retval;
+ if (!strcmp("DEFAULT", name)) {
+ retval = 0;
+ // Restart with empty defaults
+ default_conf = dev_config();
}
-
- // We've got a legit entry, make space to store it
- conf_entries.push_back( dev_config() );
- dev_config & cfg = conf_entries.back();
-
- cfg.name = name;
+ else {
+ retval = (!strcmp(SCANDIRECTIVE, name) ? -1 : 1);
+ // Init new entry with current defaults
+ conf_entries.push_back(default_conf);
+ }
+ dev_config & cfg = (retval ? conf_entries.back() : default_conf);
- // Store line number, and by default check for both device types.
- cfg.lineno=lineno;
+ cfg.name = name; // Later replaced by dev->get_info().info_name
+ cfg.dev_name = name; // If DEVICESCAN later replaced by get->dev_info().dev_name
+ cfg.lineno = lineno;
// parse tokens one at a time from the file.
- while ((token=strtok(NULL,delim))){
- int retval=ParseToken(token,cfg);
-
- if (retval==0)
- // No tokens left:
- break;
-
- if (retval>0) {
- // Parsed token
-#if (0)
- PrintOut(LOG_INFO,"Parsed token %s\n",token);
-#endif
- continue;
- }
-
- if (retval<0) {
+ while (char * token = strtok(0, delim)) {
+ int rc = ParseToken(token, cfg);
+ if (rc < 0)
// error found on the line
return -2;
- }
+
+ if (rc == 0)
+ // No tokens left
+ break;
+
+ // PrintOut(LOG_INFO,"Parsed token %s\n",token);
}
-
+
+ // Don't perform checks below for DEFAULT entries
+ if (retval == 0)
+ return retval;
+
// If NO monitoring directives are set, then set all of them.
- if (!(cfg.smartcheck || cfg.usagefailed || cfg.prefail ||
- cfg.usage || cfg.selftest || cfg.errorlog ||
- cfg.tempdiff || cfg.tempinfo || cfg.tempcrit )) {
+ if (!( cfg.smartcheck || cfg.selftest
+ || cfg.errorlog || cfg.xerrorlog
+ || cfg.offlinests || cfg.selfteststs
+ || cfg.usagefailed || cfg.prefail || cfg.usage
+ || cfg.tempdiff || cfg.tempinfo || cfg.tempcrit)) {
PrintOut(LOG_INFO,"Drive: %s, implied '-a' Directive on line %d of file %s\n",
cfg.name.c_str(), cfg.lineno, configfile);
cfg.usage = true;
cfg.selftest = true;
cfg.errorlog = true;
+ cfg.selfteststs = true;
}
// additional sanity check. Has user set -M options without -m?
cfg.name.c_str(), cfg.lineno, configfile);
return -2;
}
- // From here on the sign of <nomailer> is address.empty() and !cfg.emailcmdline.empty()
+ // From here on the sign of <nomailer> is cfg.emailaddress.empty() and !cfg.emailcmdline.empty()
cfg.emailaddress.clear();
}
- // set cfg.emailfreq to 1 (once) if user hasn't set it
- if ((!cfg.emailaddress.empty() || !cfg.emailcmdline.empty()) && !cfg.emailfreq)
- cfg.emailfreq = 1;
-
- if (devscan)
- return -1;
- else
- return 1;
+ return retval;
}
// Parses a configuration file. Return values are:
// possiblities:
// Empty configuration file ==> conf_entries.empty()
// No configuration file ==> conf_entries[0].lineno == 0
-// SCANDIRECTIVE found ==> conf_entries[0].lineno != 0
+// SCANDIRECTIVE found ==> conf_entries.back().lineno != 0 (size >= 1)
static int ParseConfigFile(dev_config_vector & conf_entries)
{
// maximum line length in configuration file
else // read from stdin ('-c -' option)
f.open(stdin);
+ // Start with empty defaults
+ dev_config default_conf;
+
// No configuration file found -- use fake one
int entry = 0;
if (!f) {
- char fakeconfig[] = SCANDIRECTIVE" -a"; // TODO: Remove this hack, build cfg_entry.
+ char fakeconfig[] = SCANDIRECTIVE " -a"; // TODO: Remove this hack, build cfg_entry.
- if (ParseConfigLine(conf_entries, entry, 0, fakeconfig) != -1)
- throw std::logic_error("Internal error parsing "SCANDIRECTIVE);
+ if (ParseConfigLine(conf_entries, default_conf, 0, fakeconfig) != -1)
+ throw std::logic_error("Internal error parsing " SCANDIRECTIVE);
return 0;
}
// are we at the end of the file?
if (!code){
if (cont) {
- scandevice = ParseConfigLine(conf_entries, entry, contlineno, fullline);
+ scandevice = ParseConfigLine(conf_entries, default_conf, contlineno, fullline);
// See if we found a SCANDIRECTIVE directive
if (scandevice==-1)
return 0;
}
// copy string so far into fullline, and increment length
- strcpy(fullline+cont,line);
+ snprintf(fullline+cont, sizeof(fullline)-cont, "%s" ,line);
cont+=len;
// is this a continuation line. If so, replace \ by space and look at next line
}
// Not a continuation line. Parse it
- scandevice = ParseConfigLine(conf_entries, entry, contlineno, fullline);
+ scandevice = ParseConfigLine(conf_entries, default_conf, contlineno, fullline);
// did we find a scandevice directive?
if (scandevice==-1)
/* Prints the message "=======> VALID ARGUMENTS ARE: <LIST> <=======\n", where
<LIST> is the list of valid arguments for option opt. */
-void PrintValidArgs(char opt) {
+static void PrintValidArgs(char opt)
+{
const char *s;
PrintOut(LOG_CRIT, "=======> VALID ARGUMENTS ARE: ");
PrintOut(LOG_CRIT, " <=======\n");
}
-// Return true if absolute path name
-static bool is_abs_path(const char * path)
+#ifndef _WIN32
+// Report error and exit if specified path is not absolute.
+static void check_abs_path(char option, const std::string & path)
{
- if (*path == '/')
- return true;
-#if defined(_WIN32) || defined(__CYGWIN__)
- if (*path == '\\')
- return true;
- int n = -1;
- sscanf(path, "%*1[A-Za-z]:%*1[/\\]%n", &n);
- if (n > 0)
- return true;
-#endif
- return false;
+ if (path.empty() || path[0] == '/')
+ return;
+
+ debugmode = 1;
+ PrintHead();
+ PrintOut(LOG_CRIT, "=======> INVALID ARGUMENT TO -%c: %s <=======\n\n", option, path.c_str());
+ PrintOut(LOG_CRIT, "Error: relative path names are not allowed\n\n");
+ EXIT(EXIT_BADCMD);
}
+#endif // !_WIN32
// Parses input line, prints usage message and
// version/license/copyright messages
-void ParseOpts(int argc, char **argv){
- int optchar;
- char *tailptr;
- long lchecktime;
+static void ParseOpts(int argc, char **argv)
+{
+ // Init default path names
+#ifndef _WIN32
+ configfile = SMARTMONTOOLS_SYSCONFDIR "/smartd.conf";
+ warning_script = SMARTMONTOOLS_SMARTDSCRIPTDIR "/smartd_warning.sh";
+#else
+ std::string exedir = get_exe_dir();
+ static std::string configfile_str = exedir + "/smartd.conf";
+ configfile = configfile_str.c_str();
+ warning_script = exedir + "/smartd_warning.cmd";
+#endif
+
// Please update GetValidArgList() if you edit shortopts
- const char *shortopts = "c:l:q:dDni:p:r:s:A:B:Vh?";
- char *arg;
+ static const char shortopts[] = "c:l:q:dDni:p:r:s:A:B:w:Vh?"
+#ifdef HAVE_LIBCAP_NG
+ "C"
+#endif
+ ;
// Please update GetValidArgList() if you edit longopts
struct option longopts[] = {
{ "configfile", required_argument, 0, 'c' },
{ "interval", required_argument, 0, 'i' },
#ifndef _WIN32
{ "no-fork", no_argument, 0, 'n' },
+#else
+ { "service", no_argument, 0, 'n' },
#endif
{ "pidfile", required_argument, 0, 'p' },
{ "report", required_argument, 0, 'r' },
{ "savestates", required_argument, 0, 's' },
{ "attributelog", required_argument, 0, 'A' },
{ "drivedb", required_argument, 0, 'B' },
-#if defined(_WIN32) || defined(__CYGWIN__)
- { "service", no_argument, 0, 'n' },
-#endif
+ { "warnexec", required_argument, 0, 'w' },
{ "version", no_argument, 0, 'V' },
{ "license", no_argument, 0, 'V' },
{ "copyright", no_argument, 0, 'V' },
{ "help", no_argument, 0, 'h' },
{ "usage", no_argument, 0, 'h' },
+#ifdef HAVE_LIBCAP_NG
+ { "capabilities", no_argument, 0, 'C' },
+#endif
{ 0, 0, 0, 0 }
};
bool badarg = false;
bool no_defaultdb = false; // set true on '-B FILE'
- // Parse input options. This horrible construction is so that emacs
- // indents properly. Sorry.
- while (-1 != (optchar =
- getopt_long(argc, argv, shortopts, longopts, NULL)
- )) {
-
+ // Parse input options.
+ int optchar;
+ while ((optchar = getopt_long(argc, argv, shortopts, longopts, NULL)) != -1) {
+ char *arg;
+ char *tailptr;
+ long lchecktime;
+
switch(optchar) {
case 'q':
// when to quit
PrintOut(LOG_CRIT, "======> LEVEL MUST BE INTEGER BETWEEN 1 AND 3<=======\n");
EXIT(EXIT_BADCMD);
} else if (!strcmp(s,"ioctl")) {
- con->reportataioctl = con->reportscsiioctl = i;
+ ata_debugmode = scsi_debugmode = i;
} else if (!strcmp(s,"ataioctl")) {
- con->reportataioctl = i;
+ ata_debugmode = i;
} else if (!strcmp(s,"scsiioctl")) {
- con->reportscsiioctl = i;
+ scsi_debugmode = i;
} else {
badarg = true;
}
debugmode = savedebug;
}
break;
+ case 'w':
+ warning_script = optarg;
+ break;
case 'V':
// print version and CVS info
debugmode = 1;
PrintOut(LOG_INFO, "%s", format_version_info("smartd", true /*full*/).c_str());
EXIT(0);
break;
+#ifdef HAVE_LIBCAP_NG
+ case 'C':
+ // enable capabilities
+ enable_capabilities = true;
+ break;
+#endif
case 'h':
// help: print summary of command-line options
debugmode=1;
EXIT(EXIT_BADCMD);
}
- // absolute path is required due to chdir('/') after fork().
- if (!state_path_prefix.empty() && !debugmode && !is_abs_path(state_path_prefix.c_str())) {
- debugmode=1;
- PrintHead();
- PrintOut(LOG_CRIT, "=======> INVALID CHOICE OF OPTIONS: -s <======= \n\n");
- PrintOut(LOG_CRIT, "Error: relative path %s is only allowed in debug (-d) mode\n\n",
- state_path_prefix.c_str());
- EXIT(EXIT_BADCMD);
- }
-
- // absolute path is required due to chdir('/') after fork().
- if (!attrlog_path_prefix.empty() && !debugmode && !is_abs_path(attrlog_path_prefix.c_str())) {
- debugmode=1;
- PrintHead();
- PrintOut(LOG_CRIT, "=======> INVALID CHOICE OF OPTIONS: -s <======= \n\n");
- PrintOut(LOG_CRIT, "Error: relative path %s is only allowed in debug (-d) mode\n\n",
- attrlog_path_prefix.c_str());
- EXIT(EXIT_BADCMD);
+#ifndef _WIN32
+ if (!debugmode) {
+ // absolute path names are required due to chdir('/') after fork().
+ check_abs_path('p', pid_file);
+ check_abs_path('s', state_path_prefix);
+ check_abs_path('A', attrlog_path_prefix);
}
+#endif
// Read or init drive database
if (!no_defaultdb) {
if (devlist.size() <= 0)
return 0;
+ // add empty device slots for existing config entries
+ while (scanned_devs.size() < conf_entries.size())
+ scanned_devs.push_back((smart_device *)0);
+
// loop over entries to create
for (unsigned i = 0; i < devlist.size(); i++) {
// Move device pointer
conf_entries.push_back(base_cfg);
dev_config & cfg = conf_entries.back();
cfg.name = dev->get_info().info_name;
+ cfg.dev_name = dev->get_info().dev_name;
cfg.dev_type = type;
}
// Returns negative value (see ParseConfigFile()) if config file
// had errors, else number of entries which may be zero or positive.
-// If we found no configuration file, or it contained SCANDIRECTIVE,
-// then *scanning is set to 1, else 0.
static int ReadOrMakeConfigEntries(dev_config_vector & conf_entries, smart_device_list & scanned_devs)
{
// parse configuration file configfile (normally /etc/smartd.conf)
// we did not find a SCANDIRECTIVE and did find valid entries
PrintOut(LOG_INFO, "Configuration file %s parsed.\n", configfile);
}
- else if (conf_entries.size() == 1) {
+ else if (!conf_entries.empty()) {
// we found a SCANDIRECTIVE or there was no configuration file so
- // scan. Configuration file's first entry contains all options
+ // scan. Configuration file's last entry contains all options
// that were set
- dev_config first = conf_entries.front();
- conf_entries.clear();
+ dev_config first = conf_entries.back();
+ conf_entries.pop_back();
if (first.lineno)
PrintOut(LOG_INFO,"Configuration file %s was parsed, found %s, scanning devices\n", configfile, SCANDIRECTIVE);
return conf_entries.size();
}
+// Return true if TYPE contains a RAID drive number
+static bool is_raid_type(const char * type)
+{
+ if (str_starts_with(type, "sat,"))
+ return false;
+ int i;
+ if (sscanf(type, "%*[^,],%d", &i) != 1)
+ return false;
+ return true;
+}
+
+// Return true if DEV is already in DEVICES[0..NUMDEVS) or IGNORED[*]
+static bool is_duplicate_device(const smart_device * dev,
+ const smart_device_list & devices, unsigned numdevs,
+ const dev_config_vector & ignored)
+{
+ const smart_device::device_info & info1 = dev->get_info();
+ bool is_raid1 = is_raid_type(info1.dev_type.c_str());
+
+ for (unsigned i = 0; i < numdevs; i++) {
+ const smart_device::device_info & info2 = devices.at(i)->get_info();
+ // -d TYPE options must match if RAID drive number is specified
+ if ( info1.dev_name == info2.dev_name
+ && ( info1.dev_type == info2.dev_type
+ || !is_raid1 || !is_raid_type(info2.dev_type.c_str())))
+ return true;
+ }
+
+ for (unsigned i = 0; i < ignored.size(); i++) {
+ const dev_config & cfg2 = ignored.at(i);
+ if ( info1.dev_name == cfg2.dev_name
+ && ( info1.dev_type == cfg2.dev_type
+ || !is_raid1 || !is_raid_type(cfg2.dev_type.c_str())))
+ return true;
+ }
+ return false;
+}
// This function tries devices from conf_entries. Each one that can be
// registered is moved onto the [ata|scsi]devices lists and removed
states.clear();
// Register entries
+ dev_config_vector ignored_entries;
+ unsigned numnoscan = 0;
for (unsigned i = 0; i < conf_entries.size(); i++){
dev_config cfg = conf_entries[i];
+ if (cfg.ignore) {
+ // Store for is_duplicate_device() check and ignore
+ PrintOut(LOG_INFO, "Device: %s%s%s%s, ignored\n", cfg.name.c_str(),
+ (!cfg.dev_type.empty() ? " [" : ""),
+ cfg.dev_type.c_str(),
+ (!cfg.dev_type.empty() ? "]" : ""));
+ ignored_entries.push_back(cfg);
+ continue;
+ }
+
// get device of appropriate type
smart_device_auto_ptr dev;
bool scanning = false;
// Device may already be detected during devicescan
if (i < scanned_devs.size()) {
dev = scanned_devs.release(i);
- if (dev)
+ if (dev) {
+ // Check for a preceding non-DEVICESCAN entry for the same device
+ if ( (numnoscan || !ignored_entries.empty())
+ && is_duplicate_device(dev.get(), devices, numnoscan, ignored_entries)) {
+ PrintOut(LOG_INFO, "Device: %s, duplicate, ignored\n", dev->get_info_name());
+ continue;
+ }
scanning = true;
+ }
}
if (!dev) {
configs.push_back(cfg);
states.push_back(state);
devices.push_back(dev);
+ if (!scanning)
+ numnoscan = devices.size();
}
// if device is explictly listed and we can't register it, then
// exit unless the user has specified that the device is removable
}
}
}
+
+ init_disable_standby_check(configs);
}
// Main program without exception handling
-int main_worker(int argc, char **argv)
+static int main_worker(int argc, char **argv)
{
// Initialize interface
smart_interface::init();
if (!smi())
return 1;
- // external control variables for ATA disks
- smartmonctrl control;
-
// is it our first pass through?
bool firstpass = true;
// next time to wake up
- time_t wakeuptime;
-
- // for simplicity, null all global communications variables/lists
- con=&control;
- memset(con, 0,sizeof(control));
+ time_t wakeuptime = 0;
// parse input and print header and usage info if needed
ParseOpts(argc,argv);
- // do we mute printing from ataprint commands?
- con->printing_switchable = false;
- con->dont_print = !debugmode;
-
// Configuration for each device
dev_config_vector configs;
// Device states
bool write_states_always = true;
+#ifdef HAVE_LIBCAP_NG
+ // Drop capabilities
+ if (enable_capabilities) {
+ capng_clear(CAPNG_SELECT_BOTH);
+ capng_updatev(CAPNG_ADD, (capng_type_t)(CAPNG_EFFECTIVE|CAPNG_PERMITTED),
+ CAP_SYS_ADMIN, CAP_MKNOD, CAP_SYS_RAWIO, -1);
+ capng_apply(CAPNG_SELECT_BOTH);
+ }
+#endif
+
// the main loop of the code
for (;;) {
// Should we (re)read the config file?
if (firstpass || caughtsigHUP){
if (!firstpass) {
-#ifdef __CYGWIN__
- // Workaround for missing SIGQUIT via keyboard on Cygwin
- if (caughtsigHUP==2) {
- // Simulate SIGQUIT if another SIGINT arrives soon
- caughtsigHUP=0;
- sleep(1);
- if (caughtsigHUP==2) {
- caughtsigEXIT=SIGQUIT;
- continue;
- }
- caughtsigHUP=2;
- }
-#endif
// Write state files
if (!state_path_prefix.empty())
write_all_dev_states(configs, states);
PrintOut(LOG_INFO,
caughtsigHUP==1?
"Signal HUP - rereading configuration file %s\n":
- "\a\nSignal INT - rereading configuration file %s ("SIGQUIT_KEYNAME" quits)\n\n",
+ "\a\nSignal INT - rereading configuration file %s (" SIGQUIT_KEYNAME " quits)\n\n",
configfile);
}
}
else {
// exit with configuration file error status
- int status = (entries==-3 ? EXIT_READCONF : entries==-2 ? EXIT_NOCONF : EXIT_BADCONF);
- EXIT(status);
+ return (entries==-3 ? EXIT_READCONF : entries==-2 ? EXIT_NOCONF : EXIT_BADCONF);
}
}
}
else {
PrintOut(LOG_INFO,"Unable to monitor any SMART enabled devices. Try debug (-d) option. Exiting...\n");
- EXIT(EXIT_NODEV);
+ return EXIT_NODEV;
}
if (quit==4) {
// user has asked to print test schedule
PrintTestSchedule(configs, states, devices);
- EXIT(0);
+ return 0;
}
-
+
+#ifdef HAVE_LIBCAP_NG
+ if (enable_capabilities) {
+ for (unsigned i = 0; i < configs.size(); i++) {
+ if (!configs[i].emailaddress.empty() || !configs[i].emailcmdline.empty()) {
+ PrintOut(LOG_WARNING, "Mail can't be enabled together with --capabilities. All mail will be suppressed.\n");
+ break;
+ }
+ }
+ }
+#endif
+
// reset signal
caughtsigHUP=0;
// check all devices once,
// self tests are not started in first pass unless '-q onecheck' is specified
- CheckDevicesOnce(configs, states, devices, (!firstpass || quit==3));
+ CheckDevicesOnce(configs, states, devices, firstpass, (!firstpass || quit==3));
// Write state files
if (!state_path_prefix.empty())
if (quit==3) {
PrintOut(LOG_INFO,"Started with '-q onecheck' option. All devices sucessfully checked once.\n"
"smartd is exiting (exit status 0)\n");
- EXIT(0);
+ return 0;
}
// fork into background if needed
"smartd", "SmartD Service", // servicename, displayname
// description
"Controls and monitors storage devices using the Self-Monitoring, "
- "Analysis and Reporting Technology System (S.M.A.R.T.) "
- "built into ATA and SCSI Hard Drives. "
- PACKAGE_HOMEPAGE
+ "Analysis and Reporting Technology System (SMART) built into "
+ "ATA/SATA and SCSI/SAS hard drives and solid-state drives. "
+ "www.smartmontools.org"
};
// daemon_main() handles daemon and service specific commands
// and starts smartd_main() direct, from a new process,
projects / mirror_smartmontools-debian.git / blobdiff