diff --git a/sm5/CHANGELOG b/sm5/CHANGELOG
index 4269a8bb28eeecbf7bf6a46731031e063f567dcb..edc7ec46b9c2ffac3c2219abf31c96fe5685270d 100644
--- a/sm5/CHANGELOG
+++ b/sm5/CHANGELOG
@@ -1,6 +1,6 @@
CHANGELOG for smartmontools
-$Id: CHANGELOG,v 1.50 2002/11/17 05:57:32 ballen4705 Exp $
+$Id: CHANGELOG,v 1.51 2002/11/21 14:11:18 ballen4705 Exp $
Copyright (C) 2002 Bruce Allen <smartmontools-support@lists.sourceforge.net>
@@ -24,10 +24,30 @@ California, Santa Cruz. http://ssrc.soe.ucsc.edu/
NOTES FOR FUTURE RELEASES: see TODO file.
CURRENT RELEASE (see VERSION file in this directory):
+
+ Allen: smartd: added Directive -m for sending test email and
+ for modifying email reminder behavior. Updated manual, and sample
+ configuration file to illustrate & explain this.
+
+ Allen: smartd: increased size of a continued smartd.conf line to
+ 1023 characters.
+
+ Allen: Simplified Directive parsers and improved warning/error
+ messages.
+
smartmontools-5.0.45
Fixed bug in smartd where testunitready logic inverted
prevented functioning on scsi devices.
+ The bug in question only affects smartd users with scsi devices.
+ To see if your version of smartd has the testunitready() bug, do
+ smartd -V
+ If the version of the module smartd.c in a line like:
+ Module: smartd.c revision: 1.66 date: 2002/11/17
+ has a revision greater than or equal to 1.30, and less than or equal to
+ 1.64, then your version of the code has this problem.
+ This problem affected releases starting with RELEASE_5_0_16 up to and
+ including RELEASE_5_0_43.
Added testunitnotready to smartctl for symmetry with smartd.
diff --git a/sm5/VERSION b/sm5/VERSION
index ea90ee31980757b2e469741512bcb39e73494e78..9e5feb5256930f3cae636754eef8a244ede164eb 100644
--- a/sm5/VERSION
+++ b/sm5/VERSION
@@ -1 +1 @@
-45
+46
diff --git a/sm5/smartd.8 b/sm5/smartd.8
index 7ba71286f342ad715cc8dd2440509abfa7bfbe63..c11fb8e6bfe7625129014e96251c3cb8715a465f 100644
--- a/sm5/smartd.8
+++ b/sm5/smartd.8
@@ -1,6 +1,6 @@
\# Copyright (C) 2002 Bruce Allen <smartmontools-support@lists.sourceforge.net>
\#
-\# $Id: smartd.8,v 1.30 2002/11/14 06:06:29 ballen4705 Exp $
+\# $Id: smartd.8,v 1.31 2002/11/21 14:11:19 ballen4705 Exp $
\#
\# 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 the Free
@@ -16,7 +16,7 @@
\# Research Center), Jack Baskin School of Engineering, University of
\# California, Santa Cruz. http://ssrc.soe.ucsc.edu/
\#
-.TH SMARTD 8 "$Date: 2002/11/14 06:06:29 $" "smartmontools-5.0"
+.TH SMARTD 8 "$Date: 2002/11/21 14:11:19 $" "smartmontools-5.0"
.SH NAME
smartd \- S.M.A.R.T. Daemon
.SH SYNOPSIS
@@ -208,10 +208,11 @@ Section below!
.B \ \ /dev/hdc -a -I 194 -I 5 -i 12
.B #
.nf
-.B # SCSI disks:
+.B # SCSI disks. Send a TEST warning email to admin on
+.B # startup. (Note: the -1 is the argument of -m.)
.B #
.B \ \ /dev/sda
-.B \ \ /dev/sdc -M admin@yoyodyne.com
+.B \ \ /dev/sdc -M admin@yoyodyne.com -m -1
.B #
.nf
.B # Strange device. It's SCSI:
@@ -255,7 +256,8 @@ name.
.B For ATA disks, if
.B no Directives appear, the disk will not be monitored.
-The '\-a' Directive will try to monitor everything possible.
+Conversely, the '\-a' Directive will try to monitor everything
+possible.
.B If a SCSI disk is listed,
it will be monitored at the only implemented level: roughly equivalent
@@ -339,13 +341,17 @@ life period."
command-line option.]
.TP
.B \-M <ADD>
-Mail: Send a warning email to the email address <ADD> if the '\-c', '\-l', '\-L', or '\-f'
-Directives detect a failure or a new error. This Directive only
-works in conjunction with these other Directives (or with the equivalent '\-a'
-Directive). To prevent your email in-box from getting filled up
-with warning messages, only a single warning will be sent for each of
-the enabled test types, '\-c', '\-l', '\-L', or '\-f', even if more
-than one failure or error is detected.
+Mail: Send a warning email to the email address <ADD> if
+the '\-c', '\-l', '\-L', or '\-f'
+Directives detect a failure or a new error.
+This Directive only works in conjunction with these other Directives
+(or with the equivalent '\-a' Directive). To prevent your email
+in-box from getting filled up with warning messages, by default only a
+single warning will be sent for each of the enabled test
+types, '\-c', '\-l', '\-L', or '\-f',
+even if more than one failure or error is
+detected or if the failure or error persists. [This behavior can be
+modified; see the '\-m' Directive below.]
The email is sent using the system
.B mail
@@ -356,11 +362,50 @@ find the mail command (normally /bin/mail) the
command must be in the path of the
shell or environment from which
.B smartd
-was started.
+was started. To test that the mail is being sent correctly, or to
+alter the pattern of when mail is sent, see the `\-m' Directive below.
-If you want to send email to more than one user, you can use the form
+To send email to more than one user, please use the following form for the address
+<ADD>:
.B user1@add1,user2@add2,...,userN@addN
-with no spaces for <ADD>.
+(with no spaces).
+.TP
+.B \-m <N>
+Modify Mail: Modifies the behavior of the '\-M' email Directive above,
+and has no effect without this other Directive. This option controls
+when, and how often, the '\-M' Directive sends email warning messages.
+
+This Directive takes a decimal integer argument <N> in the range from
+-3 to 3 inclusive.
+.nf
+.B If <N>=0 or 1:
+.fi
+equivalent to not having this '\-m' Directive present at all. Only
+one warning email is sent for each type of disk problem detected.
+.nf
+.B If <N>=2:
+.fi
+send additional warning reminder emails, once per day, for each type
+of disk problem detected.
+.nf
+.B If <N>=3:
+.fi
+send additional warning reminder emails, after a one-day interval,
+then a two-day interval, then a four-day interval, and so on for each
+type of disk problem detected. Each interval is twice as long as the
+previous interval.
+.nf
+.B If <N>=-1, -2, or -3:
+.fi
+Negative values of <N> have the same meaning as the corresponding
+positive value, but also send a single additional "test email"
+immediately upon
+.B smartd
+startup. This allows one to verify that any email is correctly
+delivered. For example, setting <N>=-2 will send a test email on
+.B smartd
+startup, and additional warning reminder emails at one-day intervals
+after any disk problems are detected.
.TP
.B \-p
Prefail: Report anytime that a Prefail Attribute has changed
@@ -548,4 +593,4 @@ Please let us know if there is an on\-line source for this document.
.SH
CVS ID OF THIS PAGE:
-$Id: smartd.8,v 1.30 2002/11/14 06:06:29 ballen4705 Exp $
+$Id: smartd.8,v 1.31 2002/11/21 14:11:19 ballen4705 Exp $
diff --git a/sm5/smartd.c b/sm5/smartd.c
index c709e357913ad521055b259e24ca6faa68209d7a..14b3e885c7bd5f468c2ec37050418fd4a7f12b51 100644
--- a/sm5/smartd.c
+++ b/sm5/smartd.c
@@ -36,6 +36,7 @@
#include <errno.h>
#include <string.h>
#include <time.h>
+#include <limits.h>
#include "atacmds.h"
#include "scsicmds.h"
#include "smartd.h"
@@ -45,7 +46,7 @@
// CVS ID strings
extern const char *CVSid1, *CVSid2;
-const char *CVSid6="$Id: smartd.c,v 1.66 2002/11/17 05:30:11 ballen4705 Exp $"
+const char *CVSid6="$Id: smartd.c,v 1.67 2002/11/21 14:11:19 ballen4705 Exp $"
CVSID1 CVSID2 CVSID3 CVSID4 CVSID7;
// global variable used for control of printing, passing arguments, etc.
@@ -89,20 +90,49 @@ void printout(int priority,char *fmt, ...){
// If address is null, this just prints a warning message. But if
// address is non-null then send and log a warning email.
-void printandmail(char *address, mailinfo *mail, int priority, char *fmt, ...){
- char command[2048], message[256], hostname[256];
+void printandmail(cfgfile *cfg, int which, int priority, char *fmt, ...){
+ char command[2048], message[256], hostname[256], additional[256], original[256], further[256];
int status;
+ time_t epoch;
va_list ap;
+ const int day=24*3600;
+ int days=0;
+ char *address=cfg->address;
+ mailinfo *mail=cfg->maildata+which;
- // See if user wants us to send mail, or if we already have
- if (!address || mail->logged)
+ // See if user wants us to send mail
+ if (!address)
return;
- // record the time of the mail message, and increment counter. This
- // is for later use if we decide to implement multiple email warning
- // messages after some delay time.
- mail->logged++;
- mail->lastsent=time(NULL);
+ // check for sanity
+ if (cfg->emailopt<0 || cfg->emailopt>3){
+ printout(LOG_INFO,"internal error in printandmail(): cfg->emailopts=%d\n",cfg->emailopt);
+ return;
+ }
+
+ // Return if a single warning mail has been sent.
+ if ((cfg->emailopt==0 || cfg->emailopt==1) && mail->logged)
+ return;
+
+ // To decide if to send mail, we need to know what time it is.
+ epoch=time(NULL);
+
+ // Return if less than one day has gone by
+ if (cfg->emailopt==2 && mail->logged && epoch<(mail->lastsent+day))
+ return;
+
+ // Return if less than 2^(logged-1) days have gone by
+ if (cfg->emailopt==3 && mail->logged){
+ days=0x01<<(mail->logged-1);
+ days*=day;
+ if (epoch<(mail->lastsent+days))
+ return;
+ }
+
+ // record the time of this mail message, and the first mail message
+ if (!mail->logged)
+ mail->firstsent=epoch;
+ mail->lastsent=epoch;
// get system host name (not null terminated if length=MAX)
if (gethostname(hostname, 256))
@@ -114,29 +144,55 @@ void printandmail(char *address, mailinfo *mail, int priority, char *fmt, ...){
va_start(ap, fmt);
vsnprintf(message, 256, 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->emailopt){
+ case 0:
+ 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->emailopt>1 && mail->logged)
+ sprintf(original,"The original email about this issue was sent at %s\n",ctime(&(mail->firstsent)));
+ }
- // now construct a command to send this as EMAIL, and issue it.
- snprintf(command, 2048, "mail -s '%s: SMART errors detected' %s > /dev/null 2> /dev/null << \"ENDMAIL\"\n"
- "This email was generated by the smartd daemon running on machine:\n"
- "%s\n"
+ // now construct a command to send this as EMAIL
+ snprintf(command, 2048, "mail -s 'SMART errors detected on host: %s' %s > /dev/null 2> /dev/null << \"ENDMAIL\"\n"
+ "This email was generated by the smartd daemon running on host:\n"
+ "%s\n\n"
"The following warning/error was logged by the smartd daemon:\n"
- "%s"
- "Further details can be found in the machine's syslog (/var/log/messages).\n"
- "You can also use the smartctl utility for further investigation.\n"
- "No additional email messages about this problem will be sent.\n"
+ "%s\n"
+ "For further details see the syslog (/var/log/messages) on host:\n"
+ "%s\n\n"
+ "%s%s%s"
"ENDMAIL\n",
- hostname, address, hostname, message);
-#if (0)
- snprintf(command,1024, "echo '%s' | mail -s '%s: smartd detected SMART errors' %s > /dev/null 2> /dev/null",
- message, hostname, address);
-#endif
+ hostname, address, hostname, message, hostname, further, original, additional);
+ // issue the command to send email
status=system(command);
if (WEXITSTATUS(status))
printout(LOG_CRIT,"Email warning message to %s failed (32-bit exit status: %d)\n",address,status);
- else
- printout(LOG_INFO,"Email warning message sent to %s\n",address);
-
+ else {
+ if (which)
+ printout(LOG_INFO,"Email warning message sent to %s\n",address);
+ else
+ printout(LOG_INFO,"Email test message sent to %s\n",address);
+ }
+
+ // increment mail sent counter
+ mail->logged++;
+
return;
}
@@ -236,6 +292,7 @@ void Directives() {
printout(LOG_INFO," -L Monitor SMART Self-Test Log, report new errors\n");
printout(LOG_INFO," -f Monitor 'Usage' Attributes, report failures\n");
printout(LOG_INFO," -M ADD Send email warning to address ADD\n");
+ printout(LOG_INFO," -m N Modify email warning behavior. -3 <= N <= 3\n");
printout(LOG_INFO," -p Report changes in 'Prefailure' Attributes\n");
printout(LOG_INFO," -u Report changes in 'Usage' Attributes\n");
printout(LOG_INFO," -t Equivalent to -p and -u Directives\n");
@@ -593,6 +650,12 @@ int ataCheckDevice(atadevices_t *drive){
char *name=drive->devicename;
cfgfile *cfg=drive->cfg;
+ // If user has asked, test the email warning system
+ if (cfg->emailopt<0){
+ cfg->emailopt*=-1;
+ printandmail(cfg, 0, LOG_CRIT, "TEST EMAIL from smartd for device: %s\n", drive->devicename);
+ }
+
// if we can't open device, fail gracefully rather than hard --
// perhaps the next time around we'll be able to open it
if ((fd=opendevice(name))<0)
@@ -605,7 +668,7 @@ int ataCheckDevice(atadevices_t *drive){
printout(LOG_INFO,"Device: %s, not capable of SMART self-check\n",name);
else if (status==1){
printout(LOG_CRIT, "Device: %s, FAILED SMART self-check. BACK UP DATA NOW!\n", name);
- printandmail(cfg->address, cfg->maildata , LOG_CRIT, "Device: %s, FAILED SMART self-check. BACK UP DATA NOW!\n", name);
+ printandmail(cfg, 1, LOG_CRIT, "Device: %s, FAILED SMART self-check. BACK UP DATA NOW!\n", name);
}
}
@@ -638,7 +701,7 @@ int ataCheckDevice(atadevices_t *drive){
// warning message
printout(LOG_CRIT, "Device: %s, Failed SMART usage Attribute: %s.\n", name, loc);
- printandmail(cfg->address, cfg->maildata+1, LOG_CRIT, "Device: %s, Failed SMART usage Attribute: %s.\n", name, loc);
+ printandmail(cfg, 2, LOG_CRIT, "Device: %s, Failed SMART usage Attribute: %s.\n", name, loc);
}
}
@@ -688,7 +751,7 @@ int ataCheckDevice(atadevices_t *drive){
if (new>old){
printout(LOG_CRIT, "Device: %s, Self-Test Log error count increased from %d to %d\n",
name, (int)old, new);
- printandmail(cfg->address, cfg->maildata+2, LOG_CRIT, "Device: %s, Self-Test Log error count increased from %d to %d\n",
+ printandmail(cfg, 3, LOG_CRIT, "Device: %s, Self-Test Log error count increased from %d to %d\n",
name, (int)old, new);
}
if (new>=0)
@@ -704,7 +767,7 @@ int ataCheckDevice(atadevices_t *drive){
if (new>old){
printout(LOG_CRIT, "Device: %s, ATA error count increased from %d to %d\n",
name, old, new);
- printandmail(cfg->address, cfg->maildata+3, LOG_CRIT, "Device: %s, ATA error count increased from %d to %d\n",
+ printandmail(cfg, 4, LOG_CRIT, "Device: %s, ATA error count increased from %d to %d\n",
name, old, new);
}
// this last line is probably not needed, count always increases
@@ -724,6 +787,12 @@ int scsiCheckDevice(scsidevices_t *drive){
int fd;
cfgfile *cfg=drive->cfg;
+ // If the user has asked for it, test the email warning system
+ if (cfg->emailopt<0){
+ cfg->emailopt*=-1;
+ printandmail(cfg, 0, LOG_CRIT, "TEST EMAIL from smartd for device: %s\n", drive->devicename);
+ }
+
// if we can't open device, fail gracefully rather than hard --
// perhaps the next time around we'll be able to open it
if ((fd=opendevice(drive->devicename))<0)
@@ -737,7 +806,7 @@ int scsiCheckDevice(scsidevices_t *drive){
if (returnvalue) {
printout(LOG_CRIT, "Device: %s, SMART Failure: (%d) %s\n", drive->devicename,
(int)returnvalue, scsiSmartGetSenseCode(returnvalue));
- printandmail(cfg->address, cfg->maildata, LOG_CRIT, "Device: %s, SMART Failure: (%d) %s\n", drive->devicename,
+ printandmail(cfg, 1, LOG_CRIT, "Device: %s, SMART Failure: (%d) %s\n", drive->devicename,
(int)returnvalue, scsiSmartGetSenseCode(returnvalue));
}
else if (debugmode)
@@ -786,6 +855,35 @@ char copyleftstring[]=
cfgfile config[MAXENTRIES];
+// exits with an error message, or returns integer value of token
+int inttoken(char *arg, char *name, char *token, int lineno, char *configfile, int min, int max){
+ char *endptr;
+ int val;
+
+ // 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);
+ Directives();
+ exit(1);
+ }
+
+ // get argument value (base 10), check that it's integer, and in-range
+ val=strtol(arg,&endptr,10);
+ if (*endptr!='\0' || val<min || 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);
+ Directives();
+ exit(1);
+ }
+
+ // all is well; return value
+ return val;
+}
+
+// This function returns non-zero if it has correctly parsed a token,
+// else zero if it has failed to parse a token. Or it exits with a
+// directive message if there is a token-parsing problem.
int parsetoken(char *token,cfgfile *cfg){
char sym=token[1];
char *name=cfg->name;
@@ -798,8 +896,8 @@ int parsetoken(char *token,cfgfile *cfg){
// is the token not recognized?
if (*token!='-' || strlen(token)!=2) {
- printout(LOG_CRIT,"Drive: %s, unknown Directive: %s at line %d of file %s\n",
- name,token,lineno,CONFIGFILE);
+ printout(LOG_CRIT,"File %s line %d (drive %s): unknown Directive: %s\n",
+ CONFIGFILE, lineno, name, token);
Directives();
exit(1);
}
@@ -807,7 +905,6 @@ int parsetoken(char *token,cfgfile *cfg){
// let's parse the token and swallow its argument
switch (sym) {
char *arg;
- char *endptr;
int val;
case 'P':
@@ -862,67 +959,43 @@ int parsetoken(char *token,cfgfile *cfg){
cfg->selftest=1;
cfg->errorlog=1;
break;
+ case 'm':
+ // email warning option
+ cfg->emailopt=inttoken(arg=strtok(NULL,delim), name, token, lineno, CONFIGFILE, -3, 3);
+ break;
+ case 'i':
+ // ignore failure of usage attribute
+ val=inttoken(arg=strtok(NULL,delim), name, token, lineno, CONFIGFILE, 1, 255);
+ isattoff(val,cfg->failatt,1);
+ break;
+ case 'I':
+ // ignore attribute for tracking purposes
+ val=inttoken(arg=strtok(NULL,delim), name, token, lineno, CONFIGFILE, 1, 255);
+ isattoff(val,cfg->trackatt,1);
+ break;
+ case 'C':
+ // period (time interval) for checking
+ checktime=inttoken(arg=strtok(NULL,delim), name, token, lineno, CONFIGFILE, 10, INT_MAX);
+ break;
case 'M':
// send email to address that follows
arg=strtok(NULL,delim);
if (!arg) {
- printout(LOG_CRIT,"Drive %s Directive: %s at line %d of file %s needs email address.\n",
- name,token,lineno,CONFIGFILE);
+ printout(LOG_CRIT,"File %s line %d (drive %s): Directive: %s needs email address(es)\n",
+ CONFIGFILE, lineno, name, token);
Directives();
exit(1);
}
if (!(cfg->address=strdup(arg))){
- printout(LOG_CRIT,"Drive %s Directive: %s at line %d of file %s: no free memory for address %s.\n",
- name,token,lineno,CONFIGFILE,arg);
+ printout(LOG_CRIT,"File %s line %d (drive %s): Directive: %s: no free memory for email address(es) %s\n",
+ CONFIGFILE, lineno, name, token, arg);
Directives();
exit(1);
}
break;
- case 'i': // ignore
- case 'I': // ignore
- case 'C': // period (time interval) for checking
- // ignore a particular vendor attribute for tracking (i) or
- // failure (I). Or give a check interval for sleeping.
- arg=strtok(NULL,delim);
- // make sure argument is there
- if (!arg) {
- printout(LOG_CRIT,"Drive %s Directive: %s at line %d of file %s needs integer argument.\n",
- name,token,lineno,CONFIGFILE);
- Directives();
- exit(1);
- }
- // get argument value, check that it's properly-formed, an
- // integer, and in-range
- val=strtol(arg,&endptr,10);
- switch (sym) {
- case 'C':
- if (*endptr!='\0' || val<10) {
- printout(LOG_CRIT,"Drive %s Directive: %s, line %d, file %s, has argument: %s, mimimum is ten secoonds\n",
- name,token,lineno,CONFIGFILE,arg);
- Directives();
- exit(1);
- }
- checktime=val;
- return 1;
- case 'i':
- case 'I':
- if (*endptr!='\0' || val<=0 || val>255 ) {
- printout(LOG_CRIT,"Drive %s Directive: %s, line %d, file %s, has argument: %s, needs 0 < n < 256\n",
- name,token,lineno,CONFIGFILE,arg);
- Directives();
- exit(1);
- }
- // put into correct list (bitmaps, access only with isattoff()
- // function. Turns OFF corresponding attribute.
- if (sym=='I')
- isattoff(val,cfg->trackatt,1);
- else
- isattoff(val,cfg->failatt,1);
- return 1;
- }
default:
- printout(LOG_CRIT,"Drive: %s, unknown Directive: %s at line %d of file %s\n",
- name,token,lineno,CONFIGFILE);
+ printout(LOG_CRIT,"File %s line %d (drive %s): unknown Directive: %s\n",
+ CONFIGFILE, lineno, name, token);
Directives();
exit(1);
}
@@ -1010,7 +1083,15 @@ int parseconfigline(int entry, int lineno,char *line){
Directives();
exit(1);
}
-
+
+ // additional sanity check. Has user set -m without -M?
+ if (cfg->emailopt && !cfg->address){
+ printout(LOG_CRIT,"Drive: %s, Directive -m useless without address Directive -M on line %d of file %s\n",
+ cfg->name, cfg->lineno, CONFIGFILE);
+ Directives();
+ exit(1);
+ }
+
entry++;
free(copy);
return 1;
diff --git a/sm5/smartd.conf b/sm5/smartd.conf
index c0f64ee921891e5bbc772b27176899004832a303..42ccb52703b15334d6509bce5f0e9981d140339b 100644
--- a/sm5/smartd.conf
+++ b/sm5/smartd.conf
@@ -23,13 +23,14 @@ DEVICESCAN
/dev/hdb -c l L -t -I 194
# A very silent check. Only report SMART health status if it fails
-/dev/hdc -c
+# But send an email in this case
+/dev/hdc -c -M admin@yoyodyne.com
-# First two SCSI disks. Note that only the S directive applies
+# First two SCSI disks. This will monitor everything that smartd can
+monitor.
/dev/sda -S
/dev/sdb -S
-
# HERE IS A LIST OF DIRECTIVES FOR THIS CONFIGURATION FILE
# -A Device is an ATA device
# -S Device is a SCSI device
@@ -40,6 +41,7 @@ DEVICESCAN
# -L Monitor SMART Self-Test Log for new errors
# -f Monitor for failure of any 'Usage' Attributes
# -M ADD Send warning email to ADD for -c, -l, -L, and -f
+# -m N Modify email warning behavior. -3 <= N <= 3
# -p Report changes in 'Prefailure' Attributes
# -u Report changes in 'Usage' Attributes
# -t Equivalent to -p and -u Directives
diff --git a/sm5/smartd.cpp b/sm5/smartd.cpp
index 7912baa7824b517c61fa5958062151205843d7c7..1c56f7586b04a14c616dcfc3df964697a06fe6a4 100644
--- a/sm5/smartd.cpp
+++ b/sm5/smartd.cpp
@@ -36,6 +36,7 @@
#include <errno.h>
#include <string.h>
#include <time.h>
+#include <limits.h>
#include "atacmds.h"
#include "scsicmds.h"
#include "smartd.h"
@@ -45,7 +46,7 @@
// CVS ID strings
extern const char *CVSid1, *CVSid2;
-const char *CVSid6="$Id: smartd.cpp,v 1.66 2002/11/17 05:30:11 ballen4705 Exp $"
+const char *CVSid6="$Id: smartd.cpp,v 1.67 2002/11/21 14:11:19 ballen4705 Exp $"
CVSID1 CVSID2 CVSID3 CVSID4 CVSID7;
// global variable used for control of printing, passing arguments, etc.
@@ -89,20 +90,49 @@ void printout(int priority,char *fmt, ...){
// If address is null, this just prints a warning message. But if
// address is non-null then send and log a warning email.
-void printandmail(char *address, mailinfo *mail, int priority, char *fmt, ...){
- char command[2048], message[256], hostname[256];
+void printandmail(cfgfile *cfg, int which, int priority, char *fmt, ...){
+ char command[2048], message[256], hostname[256], additional[256], original[256], further[256];
int status;
+ time_t epoch;
va_list ap;
+ const int day=24*3600;
+ int days=0;
+ char *address=cfg->address;
+ mailinfo *mail=cfg->maildata+which;
- // See if user wants us to send mail, or if we already have
- if (!address || mail->logged)
+ // See if user wants us to send mail
+ if (!address)
return;
- // record the time of the mail message, and increment counter. This
- // is for later use if we decide to implement multiple email warning
- // messages after some delay time.
- mail->logged++;
- mail->lastsent=time(NULL);
+ // check for sanity
+ if (cfg->emailopt<0 || cfg->emailopt>3){
+ printout(LOG_INFO,"internal error in printandmail(): cfg->emailopts=%d\n",cfg->emailopt);
+ return;
+ }
+
+ // Return if a single warning mail has been sent.
+ if ((cfg->emailopt==0 || cfg->emailopt==1) && mail->logged)
+ return;
+
+ // To decide if to send mail, we need to know what time it is.
+ epoch=time(NULL);
+
+ // Return if less than one day has gone by
+ if (cfg->emailopt==2 && mail->logged && epoch<(mail->lastsent+day))
+ return;
+
+ // Return if less than 2^(logged-1) days have gone by
+ if (cfg->emailopt==3 && mail->logged){
+ days=0x01<<(mail->logged-1);
+ days*=day;
+ if (epoch<(mail->lastsent+days))
+ return;
+ }
+
+ // record the time of this mail message, and the first mail message
+ if (!mail->logged)
+ mail->firstsent=epoch;
+ mail->lastsent=epoch;
// get system host name (not null terminated if length=MAX)
if (gethostname(hostname, 256))
@@ -114,29 +144,55 @@ void printandmail(char *address, mailinfo *mail, int priority, char *fmt, ...){
va_start(ap, fmt);
vsnprintf(message, 256, 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->emailopt){
+ case 0:
+ 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->emailopt>1 && mail->logged)
+ sprintf(original,"The original email about this issue was sent at %s\n",ctime(&(mail->firstsent)));
+ }
- // now construct a command to send this as EMAIL, and issue it.
- snprintf(command, 2048, "mail -s '%s: SMART errors detected' %s > /dev/null 2> /dev/null << \"ENDMAIL\"\n"
- "This email was generated by the smartd daemon running on machine:\n"
- "%s\n"
+ // now construct a command to send this as EMAIL
+ snprintf(command, 2048, "mail -s 'SMART errors detected on host: %s' %s > /dev/null 2> /dev/null << \"ENDMAIL\"\n"
+ "This email was generated by the smartd daemon running on host:\n"
+ "%s\n\n"
"The following warning/error was logged by the smartd daemon:\n"
- "%s"
- "Further details can be found in the machine's syslog (/var/log/messages).\n"
- "You can also use the smartctl utility for further investigation.\n"
- "No additional email messages about this problem will be sent.\n"
+ "%s\n"
+ "For further details see the syslog (/var/log/messages) on host:\n"
+ "%s\n\n"
+ "%s%s%s"
"ENDMAIL\n",
- hostname, address, hostname, message);
-#if (0)
- snprintf(command,1024, "echo '%s' | mail -s '%s: smartd detected SMART errors' %s > /dev/null 2> /dev/null",
- message, hostname, address);
-#endif
+ hostname, address, hostname, message, hostname, further, original, additional);
+ // issue the command to send email
status=system(command);
if (WEXITSTATUS(status))
printout(LOG_CRIT,"Email warning message to %s failed (32-bit exit status: %d)\n",address,status);
- else
- printout(LOG_INFO,"Email warning message sent to %s\n",address);
-
+ else {
+ if (which)
+ printout(LOG_INFO,"Email warning message sent to %s\n",address);
+ else
+ printout(LOG_INFO,"Email test message sent to %s\n",address);
+ }
+
+ // increment mail sent counter
+ mail->logged++;
+
return;
}
@@ -236,6 +292,7 @@ void Directives() {
printout(LOG_INFO," -L Monitor SMART Self-Test Log, report new errors\n");
printout(LOG_INFO," -f Monitor 'Usage' Attributes, report failures\n");
printout(LOG_INFO," -M ADD Send email warning to address ADD\n");
+ printout(LOG_INFO," -m N Modify email warning behavior. -3 <= N <= 3\n");
printout(LOG_INFO," -p Report changes in 'Prefailure' Attributes\n");
printout(LOG_INFO," -u Report changes in 'Usage' Attributes\n");
printout(LOG_INFO," -t Equivalent to -p and -u Directives\n");
@@ -593,6 +650,12 @@ int ataCheckDevice(atadevices_t *drive){
char *name=drive->devicename;
cfgfile *cfg=drive->cfg;
+ // If user has asked, test the email warning system
+ if (cfg->emailopt<0){
+ cfg->emailopt*=-1;
+ printandmail(cfg, 0, LOG_CRIT, "TEST EMAIL from smartd for device: %s\n", drive->devicename);
+ }
+
// if we can't open device, fail gracefully rather than hard --
// perhaps the next time around we'll be able to open it
if ((fd=opendevice(name))<0)
@@ -605,7 +668,7 @@ int ataCheckDevice(atadevices_t *drive){
printout(LOG_INFO,"Device: %s, not capable of SMART self-check\n",name);
else if (status==1){
printout(LOG_CRIT, "Device: %s, FAILED SMART self-check. BACK UP DATA NOW!\n", name);
- printandmail(cfg->address, cfg->maildata , LOG_CRIT, "Device: %s, FAILED SMART self-check. BACK UP DATA NOW!\n", name);
+ printandmail(cfg, 1, LOG_CRIT, "Device: %s, FAILED SMART self-check. BACK UP DATA NOW!\n", name);
}
}
@@ -638,7 +701,7 @@ int ataCheckDevice(atadevices_t *drive){
// warning message
printout(LOG_CRIT, "Device: %s, Failed SMART usage Attribute: %s.\n", name, loc);
- printandmail(cfg->address, cfg->maildata+1, LOG_CRIT, "Device: %s, Failed SMART usage Attribute: %s.\n", name, loc);
+ printandmail(cfg, 2, LOG_CRIT, "Device: %s, Failed SMART usage Attribute: %s.\n", name, loc);
}
}
@@ -688,7 +751,7 @@ int ataCheckDevice(atadevices_t *drive){
if (new>old){
printout(LOG_CRIT, "Device: %s, Self-Test Log error count increased from %d to %d\n",
name, (int)old, new);
- printandmail(cfg->address, cfg->maildata+2, LOG_CRIT, "Device: %s, Self-Test Log error count increased from %d to %d\n",
+ printandmail(cfg, 3, LOG_CRIT, "Device: %s, Self-Test Log error count increased from %d to %d\n",
name, (int)old, new);
}
if (new>=0)
@@ -704,7 +767,7 @@ int ataCheckDevice(atadevices_t *drive){
if (new>old){
printout(LOG_CRIT, "Device: %s, ATA error count increased from %d to %d\n",
name, old, new);
- printandmail(cfg->address, cfg->maildata+3, LOG_CRIT, "Device: %s, ATA error count increased from %d to %d\n",
+ printandmail(cfg, 4, LOG_CRIT, "Device: %s, ATA error count increased from %d to %d\n",
name, old, new);
}
// this last line is probably not needed, count always increases
@@ -724,6 +787,12 @@ int scsiCheckDevice(scsidevices_t *drive){
int fd;
cfgfile *cfg=drive->cfg;
+ // If the user has asked for it, test the email warning system
+ if (cfg->emailopt<0){
+ cfg->emailopt*=-1;
+ printandmail(cfg, 0, LOG_CRIT, "TEST EMAIL from smartd for device: %s\n", drive->devicename);
+ }
+
// if we can't open device, fail gracefully rather than hard --
// perhaps the next time around we'll be able to open it
if ((fd=opendevice(drive->devicename))<0)
@@ -737,7 +806,7 @@ int scsiCheckDevice(scsidevices_t *drive){
if (returnvalue) {
printout(LOG_CRIT, "Device: %s, SMART Failure: (%d) %s\n", drive->devicename,
(int)returnvalue, scsiSmartGetSenseCode(returnvalue));
- printandmail(cfg->address, cfg->maildata, LOG_CRIT, "Device: %s, SMART Failure: (%d) %s\n", drive->devicename,
+ printandmail(cfg, 1, LOG_CRIT, "Device: %s, SMART Failure: (%d) %s\n", drive->devicename,
(int)returnvalue, scsiSmartGetSenseCode(returnvalue));
}
else if (debugmode)
@@ -786,6 +855,35 @@ char copyleftstring[]=
cfgfile config[MAXENTRIES];
+// exits with an error message, or returns integer value of token
+int inttoken(char *arg, char *name, char *token, int lineno, char *configfile, int min, int max){
+ char *endptr;
+ int val;
+
+ // 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);
+ Directives();
+ exit(1);
+ }
+
+ // get argument value (base 10), check that it's integer, and in-range
+ val=strtol(arg,&endptr,10);
+ if (*endptr!='\0' || val<min || 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);
+ Directives();
+ exit(1);
+ }
+
+ // all is well; return value
+ return val;
+}
+
+// This function returns non-zero if it has correctly parsed a token,
+// else zero if it has failed to parse a token. Or it exits with a
+// directive message if there is a token-parsing problem.
int parsetoken(char *token,cfgfile *cfg){
char sym=token[1];
char *name=cfg->name;
@@ -798,8 +896,8 @@ int parsetoken(char *token,cfgfile *cfg){
// is the token not recognized?
if (*token!='-' || strlen(token)!=2) {
- printout(LOG_CRIT,"Drive: %s, unknown Directive: %s at line %d of file %s\n",
- name,token,lineno,CONFIGFILE);
+ printout(LOG_CRIT,"File %s line %d (drive %s): unknown Directive: %s\n",
+ CONFIGFILE, lineno, name, token);
Directives();
exit(1);
}
@@ -807,7 +905,6 @@ int parsetoken(char *token,cfgfile *cfg){
// let's parse the token and swallow its argument
switch (sym) {
char *arg;
- char *endptr;
int val;
case 'P':
@@ -862,67 +959,43 @@ int parsetoken(char *token,cfgfile *cfg){
cfg->selftest=1;
cfg->errorlog=1;
break;
+ case 'm':
+ // email warning option
+ cfg->emailopt=inttoken(arg=strtok(NULL,delim), name, token, lineno, CONFIGFILE, -3, 3);
+ break;
+ case 'i':
+ // ignore failure of usage attribute
+ val=inttoken(arg=strtok(NULL,delim), name, token, lineno, CONFIGFILE, 1, 255);
+ isattoff(val,cfg->failatt,1);
+ break;
+ case 'I':
+ // ignore attribute for tracking purposes
+ val=inttoken(arg=strtok(NULL,delim), name, token, lineno, CONFIGFILE, 1, 255);
+ isattoff(val,cfg->trackatt,1);
+ break;
+ case 'C':
+ // period (time interval) for checking
+ checktime=inttoken(arg=strtok(NULL,delim), name, token, lineno, CONFIGFILE, 10, INT_MAX);
+ break;
case 'M':
// send email to address that follows
arg=strtok(NULL,delim);
if (!arg) {
- printout(LOG_CRIT,"Drive %s Directive: %s at line %d of file %s needs email address.\n",
- name,token,lineno,CONFIGFILE);
+ printout(LOG_CRIT,"File %s line %d (drive %s): Directive: %s needs email address(es)\n",
+ CONFIGFILE, lineno, name, token);
Directives();
exit(1);
}
if (!(cfg->address=strdup(arg))){
- printout(LOG_CRIT,"Drive %s Directive: %s at line %d of file %s: no free memory for address %s.\n",
- name,token,lineno,CONFIGFILE,arg);
+ printout(LOG_CRIT,"File %s line %d (drive %s): Directive: %s: no free memory for email address(es) %s\n",
+ CONFIGFILE, lineno, name, token, arg);
Directives();
exit(1);
}
break;
- case 'i': // ignore
- case 'I': // ignore
- case 'C': // period (time interval) for checking
- // ignore a particular vendor attribute for tracking (i) or
- // failure (I). Or give a check interval for sleeping.
- arg=strtok(NULL,delim);
- // make sure argument is there
- if (!arg) {
- printout(LOG_CRIT,"Drive %s Directive: %s at line %d of file %s needs integer argument.\n",
- name,token,lineno,CONFIGFILE);
- Directives();
- exit(1);
- }
- // get argument value, check that it's properly-formed, an
- // integer, and in-range
- val=strtol(arg,&endptr,10);
- switch (sym) {
- case 'C':
- if (*endptr!='\0' || val<10) {
- printout(LOG_CRIT,"Drive %s Directive: %s, line %d, file %s, has argument: %s, mimimum is ten secoonds\n",
- name,token,lineno,CONFIGFILE,arg);
- Directives();
- exit(1);
- }
- checktime=val;
- return 1;
- case 'i':
- case 'I':
- if (*endptr!='\0' || val<=0 || val>255 ) {
- printout(LOG_CRIT,"Drive %s Directive: %s, line %d, file %s, has argument: %s, needs 0 < n < 256\n",
- name,token,lineno,CONFIGFILE,arg);
- Directives();
- exit(1);
- }
- // put into correct list (bitmaps, access only with isattoff()
- // function. Turns OFF corresponding attribute.
- if (sym=='I')
- isattoff(val,cfg->trackatt,1);
- else
- isattoff(val,cfg->failatt,1);
- return 1;
- }
default:
- printout(LOG_CRIT,"Drive: %s, unknown Directive: %s at line %d of file %s\n",
- name,token,lineno,CONFIGFILE);
+ printout(LOG_CRIT,"File %s line %d (drive %s): unknown Directive: %s\n",
+ CONFIGFILE, lineno, name, token);
Directives();
exit(1);
}
@@ -1010,7 +1083,15 @@ int parseconfigline(int entry, int lineno,char *line){
Directives();
exit(1);
}
-
+
+ // additional sanity check. Has user set -m without -M?
+ if (cfg->emailopt && !cfg->address){
+ printout(LOG_CRIT,"Drive: %s, Directive -m useless without address Directive -M on line %d of file %s\n",
+ cfg->name, cfg->lineno, CONFIGFILE);
+ Directives();
+ exit(1);
+ }
+
entry++;
free(copy);
return 1;
diff --git a/sm5/smartd.h b/sm5/smartd.h
index fb64bc6b5f56be677eb215c01cbefb838c8ff9cd..3e70ab1418685f5c020185b035e69d3d9161a82f 100644
--- a/sm5/smartd.h
+++ b/sm5/smartd.h
@@ -23,7 +23,7 @@
*/
#ifndef CVSID7
-#define CVSID7 "$Id: smartd.h,v 1.21 2002/11/13 10:04:13 ballen4705 Exp $\n"
+#define CVSID7 "$Id: smartd.h,v 1.22 2002/11/21 14:11:20 ballen4705 Exp $\n"
#endif
// Configuration file
@@ -39,9 +39,9 @@
#define MAXENTRIES 64
// maximum length of a continued line in configuration file
-#define MAXCONTLINE 511
+#define MAXCONTLINE 1023
-// how often SMART status is checked, in seconds
+// default for how often SMART status is checked, in seconds
#define CHECKTIME 1800
// maximum number of ATA devices to monitor
@@ -66,6 +66,8 @@ typedef struct mailinfo {
int logged;
// time last email was sent, as defined by man 2 time
time_t lastsent;
+ // time problem initially logged
+ time_t firstsent;
} mailinfo;
// Used to store a list of devices and options that were in the
@@ -85,8 +87,9 @@ typedef struct configfile_s {
char errorlog;
// Should we ignore missing capabilities/SMART errors
char permissive;
- // mailing information for each of the previous error types
- mailinfo maildata[4];
+ // mailing information for four of the previous error types plus mailtest
+ mailinfo maildata[5];
+ char emailopt;
// address to send email to
char *address;
// counts of ata and self-test errors. Perhaps ought to be in the
@@ -123,6 +126,6 @@ typedef struct scsidevices_s {
// Declare our own printing functions...
void printout(int priority,char *fmt, ...) __attribute__ ((format(printf, 2, 3)));
-void printandmail(char *address, mailinfo *mail, int priority, char *fmt, ...) __attribute__ ((format(printf, 4, 5)));
+void printandmail(cfgfile *cfg, int which, int priority, char *fmt, ...) __attribute__ ((format(printf, 4, 5)));
int ataCheckDevice(atadevices_t *drive);
diff --git a/sm5/smartmontools.spec b/sm5/smartmontools.spec
index 338b5133d7585896726b07e49d8f86a0f6dd39c7..83f50262e9a462bccce29e7d68515377f2b76b81 100644
--- a/sm5/smartmontools.spec
+++ b/sm5/smartmontools.spec
@@ -1,4 +1,4 @@
-Release: 45
+Release: 46
Summary: SMARTmontools - for monitoring S.M.A.R.T. disks and devices
Summary(cs): SMARTmontools - pro monitorov�n� S.M.A.R.T. disk� a za��zen�
Summary(de): SMARTmontools - zur �berwachung von S.M.A.R.T.-Platten und-Ger�ten
@@ -30,7 +30,7 @@ Packager: Bruce Allen <smartmontools-support@lists.sourceforge.net>
# http://ftp1.sourceforge.net/smartmontools/smartmontools-%{version}-%{release}.tar.gz
# CVS ID of this file is:
-# $Id: smartmontools.spec,v 1.69 2002/11/17 05:57:32 ballen4705 Exp $
+# $Id: smartmontools.spec,v 1.70 2002/11/21 14:11:20 ballen4705 Exp $
# Copyright (C) 2002 Bruce Allen <smartmontools-support@lists.sourceforge.net>
# Home page: http://smartmontools.sourceforge.net/