diff --git a/sm5/CHANGELOG b/sm5/CHANGELOG
index 372442873fd14360b8c826b9f1a28ab34775067f..d3ddd3bd0fc52f9a52969e14b0bbb9f5569a4ea6 100644
--- a/sm5/CHANGELOG
+++ b/sm5/CHANGELOG
@@ -1,6 +1,6 @@
CHANGELOG for smartmontools
-$Id: CHANGELOG,v 1.22 2002/10/25 17:06:17 ballen4705 Exp $
+$Id: CHANGELOG,v 1.23 2002/10/26 09:24:26 ballen4705 Exp $
Copyright (C) 2002 Bruce Allen <smartmontools-support@lists.sourceforge.net>
@@ -21,13 +21,38 @@ Research Center), Jack Baskin School of Engineering, University of
California, Santa Cruz. http://ssrc.soe.ucsc.edu/
-NOTES FOR NEXT RELEASE:
+NOTES FOR FUTURE RELEASES: see TODO file.
- Next release: handle extended error and self-test logs gracefully.
- Parse and print attribute flag meanings
CURRENT RELEASE (see VERSION file in this directory):
smartmontools-5.0-VERSION
+
+ modified length of device name string in smartd internal structure
+ to accomodate max length device name strings
+
+ removed un-implemented (-e = Email notification) option from
+ command line arg list. We'll put it back on when implemeneted.
+
+ smartd now logs serious (fatal) conditions in its operation at
+ loglevel LOG_CRIT rather than LOG_INFO before exiting with error.
+
+ smartd used to open a file descriptor for each SMART enabled
+ device, and then keep it open the entire time smartd was running.
+ This meant that some commands, like IOREADBLKPART did not work,
+ since the fd to the device was open. smartd now opens the device
+ when it needs to read values, then closes it. Also, if one time
+ around it can't open the device, it simply prints a warning
+ message but does not give up.
+
+ smartd now opens SCSI devices as well using O_RDONLY rather than
+ O_RDWR. If someone can no longer monitor a SCSI device that used
+ to be readable, this may well be the reason why.
+
+ smartd never checked if the number of ata or scsi devices detected
+ was greater than the max number it could monitor. Now it does.
+
+smartmontools-5.0-16
+
smartd on startup now looks in the configuration file /etc/smartd.conf for
a list of devices which to include in its monitoring list. See man page
(man smartd) for syntax.
diff --git a/sm5/TODO b/sm5/TODO
index 3a35dba3a0574f4ddddb18a74672c92b3e1c566f..4ad6eeae9b7bd5e1580d18fe5c2fe8c822ab7bbb 100644
--- a/sm5/TODO
+++ b/sm5/TODO
@@ -4,7 +4,7 @@ Home page of code is: http://smartmontools.sourceforge.net
Copyright (C) 2002 Bruce Allen <smartmontools-support@lists.sourceforge.net>
-$Id: TODO,v 1.15 2002/10/25 14:23:40 ballen4705 Exp $
+$Id: TODO,v 1.16 2002/10/26 09:24:26 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
@@ -34,29 +34,36 @@ Produce version for ATA/ATAPI-7
Modifications
-------------
-Change smartd so that it also monitors usage as well as prefail attributes for failure
-or changes. Make this a command line option
+Handle extended error and self-test logs gracefully. Can someone tell
+me a disk that supports more than log pages 1 and 6? I need to get a
+disk so I can test this functionality, when I add it.
-Perhaps change smartd to look in /proc/ide and /proc/scsi to see what exists? If something
-doesn't exit then don't try to open it?
+Change smartd so that it also monitors usage as well as prefail
+attributes for failure or changes. Make this a command line option
-Currently smartd looks at attribute values and thresholds, then prints out if there is an
-error condition. Make it also look at the smart status, if that is enabled and working and
-do that test as well. Perhaps also see if the number of device errors has changed.
+Perhaps change smartd to look in /proc/ide and /proc/scsi to see what
+exists? If something doesn't exit then don't try to open it?
-Change smartd so that it monitors the ATA disk error log, and if the number of errors changes,
-log an entry..
+Currently smartd looks at attribute values and thresholds, then prints
+out if there is an error condition. Make it also look at the smart
+status, if that is enabled and working and do that test as well.
+Perhaps also see if the number of device errors has changed.
-Perhaps modify the -q option (quiet mode) so that it only warns of ATA errors if they have (say)
-taken place in the last 168 hours (week).
+Add optional flags to /etc/smartd.conf, so that certain attributes are
+ignored, or so that all usage attributes are ignored.
+
+Change smartd so that it monitors the ATA disk error log, and if the
+number of errors changes, log an entry..
+
+Perhaps modify the -q option (quiet mode) so that it only warns of ATA
+errors if they have (say) taken place in the last 168 hours (week).
Change input command line from using current command line format
(getopt) to getopt_long() for long input options. This will be
helpful in adding device specific options.
-Command line option to specify devices to look for in smartd startup
-
-Print flags meanings in Vendor Attribute list -- not hex value --
+Parse and print more attribute flag meanings (IBM ones, eg performance
+etc).
Fixes
-----
diff --git a/sm5/VERSION b/sm5/VERSION
index b6a7d89c68e0ca66e96a9a51892cc33db66fb8a3..98d9bcb75a685dfbfd60f611c309410152935b3d 100644
--- a/sm5/VERSION
+++ b/sm5/VERSION
@@ -1 +1 @@
-16
+17
diff --git a/sm5/smartd.c b/sm5/smartd.c
index e504e80e595d1d3487ee7847eef0da063355186b..7944644f21e99882ea425876990d6ea803f3028a 100644
--- a/sm5/smartd.c
+++ b/sm5/smartd.c
@@ -37,41 +37,11 @@
#include "scsicmds.h"
#include "smartd.h"
+// CVS ID strings
extern const char *CVSid1, *CVSid2;
-const char *CVSid3="$Id: smartd.c,v 1.31 2002/10/25 14:54:14 ballen4705 Exp $"
+const char *CVSid3="$Id: smartd.c,v 1.32 2002/10/26 09:24:26 ballen4705 Exp $"
CVSID1 CVSID4 CVSID7;
-int daemon_init(void){
- pid_t pid;
- int i;
-
- if ( (pid = fork()) < 0)
- // unable to fork!
- exit(1);
- else if (pid != 0)
- // we are the parent process -- exit cleanly
- exit (0);
-
- // from here on, we are the child process
- setsid();
-
- // close any open file descriptors
- for (i=getdtablesize();i>=0;--i)
- close(i);
-
- // redirect any IO attempts to /dev/null
- // open stdin
- i=open("/dev/null",O_RDWR);
- // stdout
- dup(i);
- // stderr
- dup(i);
- umask(0);
- chdir("/");
- return(0);
-}
-
-
// This function prints either to stdout or to the syslog as needed
void printout(int priority,char *fmt, ...){
va_list ap;
@@ -85,22 +55,54 @@ void printout(int priority,char *fmt, ...){
return;
}
-// Printing function for atacmds
+// Printing function for debugging atacmds. For debugging set 0 to 1
+// in #if statement
void pout(char *fmt, ...){
va_list ap;
+
// initialize variable argument list
va_start(ap,fmt);
- va_end(ap);
- return;
#if (0)
- // print out
vprintf(fmt,ap);
+#endif
va_end(ap);
return;
-#endif
}
+// Forks new process, closes all file descriptors, redirects stdin,
+// stdout, stderr
+int daemon_init(void){
+ pid_t pid;
+ int i;
+ if ((pid=fork()) < 0) {
+ // unable to fork!
+ printout(LOG_CRIT,"Unable to fork daemon process!\n");
+ exit(1);
+ }
+ else if (pid)
+ // we are the parent process -- exit cleanly
+ exit(0);
+
+ // from here on, we are the child process
+ setsid();
+
+ // close any open file descriptors
+ for (i=getdtablesize();i>=0;--i)
+ close(i);
+
+ // redirect any IO attempts to /dev/null for stdin
+ i=open("/dev/null",O_RDWR);
+ // stdout
+ dup(i);
+ // stderr
+ dup(i);
+ umask(0);
+ chdir("/");
+ return(0);
+}
+
+// Prints header identifying version of code and home
void printhead(){
printout(LOG_INFO,"smartd version %d.%d-%d - S.M.A.R.T. Daemon.\n",
RELEASE_MAJOR, RELEASE_MINOR, SMARTMONTOOLS_VERSION);
@@ -114,23 +116,45 @@ void Usage (void){
printout(LOG_INFO,"Read Only Options:\n");
printout(LOG_INFO," %c Start smartd in debug Mode\n",DEBUGMODE);
printout(LOG_INFO," %c Print License, Copyright, and version information\n\n",PRINTCOPYLEFT);
- printout(LOG_INFO,"Configuration file: /etc/smartd.conf\n");
+ printout(LOG_INFO,"Optional configuration file: %s\n",CONFIGFILE);
}
-
+
+// returns negative if problem, else fd>=0
+int opendevice(char *device){
+ int fd = open(device, O_RDONLY);
+ if (fd<0) {
+ if (errno<sys_nerr)
+ printout(LOG_INFO,"%s: Device: %s, Opening device failed\n",sys_errlist[errno],device);
+ else
+ printout(LOG_INFO,"Device: %s, Opening device failed\n",device);
+ return -1;
+ }
+ // device opened sucessfully
+ return fd;
+}
+
+// returns 1 if problem, else zero
+int closedevice(int fd){
+ if (close(fd)){
+ if (errno<sys_nerr)
+ printout(LOG_INFO,"%s: Closing file descriptor %d failed\n",sys_errlist[errno],fd);
+ else
+ printout(LOG_INFO,"Closing file descriptor %d failed\n",fd);
+ return 1;
+ }
+ // device opened sucessfully
+ return 0;
+}
+
// scan to see what ata devices there are, and if they support SMART
int atadevicescan (atadevices_t *devices, char *device){
int fd;
struct hd_driveid drive;
printout(LOG_INFO,"Opening device %s\n", device);
- fd = open(device, O_RDONLY);
- if (fd < 0) {
- if (errno<sys_nerr)
- printout(LOG_INFO,"%s: Device: %s, Opening device failed\n",sys_errlist[errno],device);
- else
- printout(LOG_INFO,"Device: %s, Opening device failed\n",device);
+ if ((fd=opendevice(device))<0)
+ // device open failed
return 1;
- }
if (ataReadHDIdentity (fd,&drive) || !ataSmartSupport(drive) || ataEnableSmart(fd)){
// device exists, but not able to do SMART
@@ -153,7 +177,13 @@ int atadevicescan (atadevices_t *devices, char *device){
return 4;
}
- // device exists, and does SMART. Add to list
+ // Device exists, and does SMART. Add to list
+ if (numatadevices>=MAXATADEVICES){
+ printout(LOG_CRIT,"smartd has found more than MAXATADEVICES=%d ATA devices.\n"
+ "Recompile code from " PROJECTHOME " with larger MAXATADEVICES\n",numatadevices);
+ exit(1);
+ }
+
printout(LOG_INFO,"%s Found and is SMART capable. Adding to \"monitor\" list.\n",device);
devices[numatadevices].fd = fd;
strcpy(devices[numatadevices].devicename, device);
@@ -167,6 +197,7 @@ int atadevicescan (atadevices_t *devices, char *device){
isSupportSelfTest(devices[numatadevices].smartval);
numatadevices++;
+ closedevice(fd);
return 0;
}
@@ -177,18 +208,11 @@ int scsidevicescan (scsidevices_t *devices, char *device){
int i, fd, smartsupport;
unsigned char tBuf[4096];
-
- // open device
printout(LOG_INFO,"Opening device %s\n", device);
- fd=open(device, O_RDWR);
- if (fd<0) {
- if (errno<sys_nerr)
- printout(LOG_INFO,"%s: Device: %s, Opening device failed\n",sys_errlist[errno],device);
- else
- printout(LOG_INFO,"Device: %s, Opening device failed\n", device);
+ if ((fd=opendevice(device))<0)
+ // device open failed
return 1;
- }
-
+
// check that it's ready for commands
if (!testunitready(fd)){
printout(LOG_INFO,"Device: %s, Failed Test Unit Ready\n", device);
@@ -211,6 +235,13 @@ int scsidevicescan (scsidevices_t *devices, char *device){
return 4;
}
+ // Device exists, and does SMART. Add to list
+ if (numscsidevices>=MAXSCSIDEVICES){
+ printout(LOG_CRIT,"smartd has found more than MAXSCSIDEVICES=%d SCSI devices.\n"
+ "Recompile code from " PROJECTHOME " with larger MAXSCSIDEVICES\n",numscsidevices);
+ exit(1);
+ }
+
// now we can proceed to register the device
printout(LOG_INFO, "Device: %s, Found and is SMART capable. Adding to \"monitor\" list.\n",device);
devices[numscsidevices].fd = fd;
@@ -233,6 +264,7 @@ int scsidevicescan (scsidevices_t *devices, char *device){
}
}
numscsidevices++;
+ closedevice(fd);
return 0;
}
@@ -269,19 +301,28 @@ void ataCompareSmartValues (atadevices_t *device, struct ata_smart_values new ){
int ataCheckDevice( atadevices_t *drive){
struct ata_smart_values tempsmartval;
struct ata_smart_thresholds tempsmartthres;
- int failed;
+ int failed,fd;
char *loc,attributename[64];
+ // 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)
+ return 1;
+
// Coming into this function, *drive contains the last values measured,
// and we read the NEW values into tempsmartval
- if (ataReadSmartValues(drive->fd,&tempsmartval))
+ if (ataReadSmartValues(fd,&tempsmartval))
printout(LOG_INFO, "%s:Failed to read SMART values\n", drive->devicename);
// and we read the new thresholds into tempsmartthres
- if (ataReadSmartThresholds (drive->fd, &tempsmartthres))
+ if (ataReadSmartThresholds(fd, &tempsmartthres))
printout(LOG_INFO, "%s:Failed to read SMART thresholds\n",drive->devicename);
- // See if any vendor attributes are below minimum, and print them out
+ // See if any vendor attributes are below minimum, and print them
+ // out. WHEN IT WORKS, we should here add a call to
+ // ataSmartStatus2() either in addition to or instead of the
+ // ataCheckSmart command below. This is the "right" long-term
+ // solution.
if ((failed=ataCheckSmart(tempsmartval,tempsmartthres,1))){
ataPrintSmartAttribName(attributename,failed);
// skip blank space in name
@@ -291,18 +332,15 @@ int ataCheckDevice( atadevices_t *drive){
printout(LOG_CRIT,"Device: %s, Failed SMART attribute: %s. Use smartctl -a %s.\n",
drive->devicename,loc,drive->devicename);
}
-
- // WHEN IT WORKS, we should here add a call to ataSmartStatus2()
- // either in addition to or instead of the ataCheckSmart command
- // above. This is the "right" long-term solution.
// see if any values have changed. Second argument is new values
- ataCompareSmartValues (drive , tempsmartval);
+ ataCompareSmartValues(drive, tempsmartval);
// Save the new values into *drive for the next time around
drive->smartval = tempsmartval;
drive->smartthres = tempsmartthres;
+ closedevice(fd);
return 0;
}
@@ -312,10 +350,16 @@ int scsiCheckDevice( scsidevices_t *drive){
UINT8 returnvalue;
UINT8 currenttemp;
UINT8 triptemp;
-
+ int fd;
+
+ // 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)
+ return 1;
+
currenttemp = triptemp = 0;
- if (scsiCheckSmart( drive->fd, drive->SmartPageSupported, &returnvalue, ¤ttemp, &triptemp ) != 0)
+ if (scsiCheckSmart(fd, drive->SmartPageSupported, &returnvalue, ¤ttemp, &triptemp))
printout(LOG_INFO, "%s:Failed to read SMART values\n", drive->devicename);
if (returnvalue)
@@ -327,32 +371,34 @@ int scsiCheckDevice( scsidevices_t *drive){
// Seems to completely ignore what capabilities were found on the
// device when scanned
if (currenttemp){
- if ( (currenttemp != drive->Temperature) && ( drive->Temperature) )
+ if ((currenttemp != drive->Temperature) && (drive->Temperature))
printout(LOG_INFO, "Device: %s, Temperature changed %d degrees to %d degrees since last reading\n",
drive->devicename, (int) (currenttemp - drive->Temperature), (unsigned int) currenttemp );
-
drive->Temperature = currenttemp;
- }
+ }
+ closedevice(fd);
return 0;
}
void CheckDevices ( atadevices_t *atadevices, scsidevices_t *scsidevices){
int i;
+ // If there are no devices to monitor, then exit
if (!numatadevices && !numscsidevices){
printout(LOG_INFO,"Unable to monitor any SMART enabled ATA or SCSI devices.\n");
return;
}
+ // Infinite loop, which checkes devices
printout(LOG_INFO,"Started monitoring %d ATA and %d SCSI devices\n",numatadevices,numscsidevices);
while (1){
- for (i = 0; i < numatadevices;i++)
- ataCheckDevice ( &atadevices[i]);
+ for (i=0; i<numatadevices; i++)
+ ataCheckDevice(atadevices+i);
- for (i = 0; i < numscsidevices;i++)
- scsiCheckDevice ( &scsidevices[i]);
+ for (i=0; i<numscsidevices; i++)
+ scsiCheckDevice(scsidevices+i);
- sleep ( checktime );
+ sleep(checktime);
}
}
@@ -363,7 +409,7 @@ int massagecvs(char *out,const char *in){
const char *savein=in;
// skip to I of $Id:
- while (*in !='\0' && *in!='I')
+ while (*in && *in!='I')
in++;
// skip to start of filename
@@ -442,10 +488,10 @@ int parseconfigfile(){
if (fp==NULL && errno!=ENOENT){
// file exists but we can't read it
if (errno<sys_nerr)
- printout(LOG_INFO,"%s: Unable to open configuration file %s\n",
+ printout(LOG_CRIT,"%s: Unable to open configuration file %s\n",
sys_errlist[errno],CONFIGFILE);
else
- printout(LOG_INFO,"Unable to open configuration file %s\n",CONFIGFILE);
+ printout(LOG_CRIT,"Unable to open configuration file %s\n",CONFIGFILE);
exit(1);
}
@@ -465,7 +511,7 @@ int parseconfigfile(){
// See if line is too long
len=strlen(line);
if (len>MAXLINELEN){
- printout(LOG_INFO,"Error: line %d of file %s is more than than %d characters long.\n",
+ printout(LOG_CRIT,"Error: line %d of file %s is more than than %d characters long.\n",
lineno,CONFIGFILE,MAXLINELEN);
exit(1);
}
@@ -488,7 +534,7 @@ int parseconfigfile(){
// We've got a legit entry
if (entry>=MAXENTRIES){
- printout(LOG_INFO,"Error: configuration file %s can have no more than %d entries\n",
+ printout(LOG_CRIT,"Error: configuration file %s can have no more than %d entries\n",
CONFIGFILE,MAXENTRIES);
exit(1);
}
@@ -511,25 +557,21 @@ int parseconfigfile(){
if (entry)
return entry;
- printout(LOG_INFO,"Configuration file %s contained no devices (like /dev/hda)\n",CONFIGFILE);
+ printout(LOG_CRIT,"Configuration file %s contained no devices (like /dev/hda)\n",CONFIGFILE);
exit(1);
}
-const char opts[] = { DEBUGMODE, EMAILNOTIFICATION, PRINTCOPYLEFT,'h','?','\0' };
+// const char opts[] = {DEBUGMODE, EMAILNOTIFICATION, PRINTCOPYLEFT,'h','?','\0' };
+const char opts[] = {DEBUGMODE, PRINTCOPYLEFT,'h','?','\0' };
-/* Main Program */
-int main (int argc, char **argv){
- atadevices_t atadevices[MAXATADEVICES], *atadevicesptr;
- scsidevices_t scsidevices[MAXSCSIDEVICES], *scsidevicesptr;
- int optchar,i;
+
+// Parses input line, prints usage message and
+// version/license/copyright messages
+void ParseOpts(int argc, char **argv){
extern char *optarg;
extern int optopt, optind, opterr;
- int entries;
-
- numatadevices=0;
- numscsidevices=0;
- scsidevicesptr = scsidevices;
- atadevicesptr = atadevices;
+ int optchar;
+
opterr=optopt=0;
// Parse input options:
@@ -550,7 +592,7 @@ int main (int argc, char **argv){
debugmode=1;
if (optopt) {
printhead();
- printout(LOG_INFO,"=======> UNRECOGNIZED OPTION: %c <======= \n\n",optopt);
+ printout(LOG_CRIT,"=======> UNRECOGNIZED OPTION: %c <======= \n\n",optopt);
Usage();
exit(-1);
}
@@ -574,6 +616,19 @@ int main (int argc, char **argv){
// print header
printhead();
+ return;
+}
+
+/* Main Program */
+int main (int argc, char **argv){
+ atadevices_t atadevices[MAXATADEVICES], *atadevicesptr=atadevices;
+ scsidevices_t scsidevices[MAXSCSIDEVICES], *scsidevicesptr=scsidevices;
+ int i,entries;
+
+ numatadevices=numscsidevices=0;
+
+ // Parse input and print header and usage info if needed
+ ParseOpts(argc,argv);
// look in configuration file CONFIGFILE (normally /etc/smartd.conf)
entries=parseconfigfile();
@@ -583,18 +638,20 @@ int main (int argc, char **argv){
daemon_init();
}
- // If we found a config file, look at its entries
+ // If we found a config file, register its entries
if (entries)
for (i=0;i<entries;i++){
+ // register ATA devices
if (config[i].tryata && atadevicescan(atadevicesptr, config[i].name))
printout(LOG_INFO,"Unable to register ATA device %s at line %d of file %s\n",
config[i].name, config[i].lineno, CONFIGFILE);
-
+ // then register SCSI devices
if (config[i].tryscsi && scsidevicescan(scsidevicesptr, config[i].name))
printout(LOG_INFO,"Unable to register SCSI device %s at line %d of file %s\n",
config[i].name, config[i].lineno, CONFIGFILE);
}
else {
+ // since there was no config file found, search all ATA and SCSI disks
char deviceata[] = "/dev/hda";
char devicescsi[]= "/dev/sda";
printout(LOG_INFO,"No configuration file %s found. Searching for devices.\n",CONFIGFILE);
@@ -604,6 +661,7 @@ int main (int argc, char **argv){
scsidevicescan(scsidevicesptr, devicescsi);
}
+ // Now start an infinite loop that checks all devices
CheckDevices(atadevicesptr, scsidevicesptr);
return 0;
}
diff --git a/sm5/smartd.cpp b/sm5/smartd.cpp
index 63546130de6b87e60de8afc483081d18b4feef6f..f376bbfdff51ff9c89137bc45b0c0acf3199173d 100644
--- a/sm5/smartd.cpp
+++ b/sm5/smartd.cpp
@@ -37,41 +37,11 @@
#include "scsicmds.h"
#include "smartd.h"
+// CVS ID strings
extern const char *CVSid1, *CVSid2;
-const char *CVSid3="$Id: smartd.cpp,v 1.31 2002/10/25 14:54:14 ballen4705 Exp $"
+const char *CVSid3="$Id: smartd.cpp,v 1.32 2002/10/26 09:24:26 ballen4705 Exp $"
CVSID1 CVSID4 CVSID7;
-int daemon_init(void){
- pid_t pid;
- int i;
-
- if ( (pid = fork()) < 0)
- // unable to fork!
- exit(1);
- else if (pid != 0)
- // we are the parent process -- exit cleanly
- exit (0);
-
- // from here on, we are the child process
- setsid();
-
- // close any open file descriptors
- for (i=getdtablesize();i>=0;--i)
- close(i);
-
- // redirect any IO attempts to /dev/null
- // open stdin
- i=open("/dev/null",O_RDWR);
- // stdout
- dup(i);
- // stderr
- dup(i);
- umask(0);
- chdir("/");
- return(0);
-}
-
-
// This function prints either to stdout or to the syslog as needed
void printout(int priority,char *fmt, ...){
va_list ap;
@@ -85,22 +55,54 @@ void printout(int priority,char *fmt, ...){
return;
}
-// Printing function for atacmds
+// Printing function for debugging atacmds. For debugging set 0 to 1
+// in #if statement
void pout(char *fmt, ...){
va_list ap;
+
// initialize variable argument list
va_start(ap,fmt);
- va_end(ap);
- return;
#if (0)
- // print out
vprintf(fmt,ap);
+#endif
va_end(ap);
return;
-#endif
}
+// Forks new process, closes all file descriptors, redirects stdin,
+// stdout, stderr
+int daemon_init(void){
+ pid_t pid;
+ int i;
+ if ((pid=fork()) < 0) {
+ // unable to fork!
+ printout(LOG_CRIT,"Unable to fork daemon process!\n");
+ exit(1);
+ }
+ else if (pid)
+ // we are the parent process -- exit cleanly
+ exit(0);
+
+ // from here on, we are the child process
+ setsid();
+
+ // close any open file descriptors
+ for (i=getdtablesize();i>=0;--i)
+ close(i);
+
+ // redirect any IO attempts to /dev/null for stdin
+ i=open("/dev/null",O_RDWR);
+ // stdout
+ dup(i);
+ // stderr
+ dup(i);
+ umask(0);
+ chdir("/");
+ return(0);
+}
+
+// Prints header identifying version of code and home
void printhead(){
printout(LOG_INFO,"smartd version %d.%d-%d - S.M.A.R.T. Daemon.\n",
RELEASE_MAJOR, RELEASE_MINOR, SMARTMONTOOLS_VERSION);
@@ -114,23 +116,45 @@ void Usage (void){
printout(LOG_INFO,"Read Only Options:\n");
printout(LOG_INFO," %c Start smartd in debug Mode\n",DEBUGMODE);
printout(LOG_INFO," %c Print License, Copyright, and version information\n\n",PRINTCOPYLEFT);
- printout(LOG_INFO,"Configuration file: /etc/smartd.conf\n");
+ printout(LOG_INFO,"Optional configuration file: %s\n",CONFIGFILE);
}
-
+
+// returns negative if problem, else fd>=0
+int opendevice(char *device){
+ int fd = open(device, O_RDONLY);
+ if (fd<0) {
+ if (errno<sys_nerr)
+ printout(LOG_INFO,"%s: Device: %s, Opening device failed\n",sys_errlist[errno],device);
+ else
+ printout(LOG_INFO,"Device: %s, Opening device failed\n",device);
+ return -1;
+ }
+ // device opened sucessfully
+ return fd;
+}
+
+// returns 1 if problem, else zero
+int closedevice(int fd){
+ if (close(fd)){
+ if (errno<sys_nerr)
+ printout(LOG_INFO,"%s: Closing file descriptor %d failed\n",sys_errlist[errno],fd);
+ else
+ printout(LOG_INFO,"Closing file descriptor %d failed\n",fd);
+ return 1;
+ }
+ // device opened sucessfully
+ return 0;
+}
+
// scan to see what ata devices there are, and if they support SMART
int atadevicescan (atadevices_t *devices, char *device){
int fd;
struct hd_driveid drive;
printout(LOG_INFO,"Opening device %s\n", device);
- fd = open(device, O_RDONLY);
- if (fd < 0) {
- if (errno<sys_nerr)
- printout(LOG_INFO,"%s: Device: %s, Opening device failed\n",sys_errlist[errno],device);
- else
- printout(LOG_INFO,"Device: %s, Opening device failed\n",device);
+ if ((fd=opendevice(device))<0)
+ // device open failed
return 1;
- }
if (ataReadHDIdentity (fd,&drive) || !ataSmartSupport(drive) || ataEnableSmart(fd)){
// device exists, but not able to do SMART
@@ -153,7 +177,13 @@ int atadevicescan (atadevices_t *devices, char *device){
return 4;
}
- // device exists, and does SMART. Add to list
+ // Device exists, and does SMART. Add to list
+ if (numatadevices>=MAXATADEVICES){
+ printout(LOG_CRIT,"smartd has found more than MAXATADEVICES=%d ATA devices.\n"
+ "Recompile code from " PROJECTHOME " with larger MAXATADEVICES\n",numatadevices);
+ exit(1);
+ }
+
printout(LOG_INFO,"%s Found and is SMART capable. Adding to \"monitor\" list.\n",device);
devices[numatadevices].fd = fd;
strcpy(devices[numatadevices].devicename, device);
@@ -167,6 +197,7 @@ int atadevicescan (atadevices_t *devices, char *device){
isSupportSelfTest(devices[numatadevices].smartval);
numatadevices++;
+ closedevice(fd);
return 0;
}
@@ -177,18 +208,11 @@ int scsidevicescan (scsidevices_t *devices, char *device){
int i, fd, smartsupport;
unsigned char tBuf[4096];
-
- // open device
printout(LOG_INFO,"Opening device %s\n", device);
- fd=open(device, O_RDWR);
- if (fd<0) {
- if (errno<sys_nerr)
- printout(LOG_INFO,"%s: Device: %s, Opening device failed\n",sys_errlist[errno],device);
- else
- printout(LOG_INFO,"Device: %s, Opening device failed\n", device);
+ if ((fd=opendevice(device))<0)
+ // device open failed
return 1;
- }
-
+
// check that it's ready for commands
if (!testunitready(fd)){
printout(LOG_INFO,"Device: %s, Failed Test Unit Ready\n", device);
@@ -211,6 +235,13 @@ int scsidevicescan (scsidevices_t *devices, char *device){
return 4;
}
+ // Device exists, and does SMART. Add to list
+ if (numscsidevices>=MAXSCSIDEVICES){
+ printout(LOG_CRIT,"smartd has found more than MAXSCSIDEVICES=%d SCSI devices.\n"
+ "Recompile code from " PROJECTHOME " with larger MAXSCSIDEVICES\n",numscsidevices);
+ exit(1);
+ }
+
// now we can proceed to register the device
printout(LOG_INFO, "Device: %s, Found and is SMART capable. Adding to \"monitor\" list.\n",device);
devices[numscsidevices].fd = fd;
@@ -233,6 +264,7 @@ int scsidevicescan (scsidevices_t *devices, char *device){
}
}
numscsidevices++;
+ closedevice(fd);
return 0;
}
@@ -269,19 +301,28 @@ void ataCompareSmartValues (atadevices_t *device, struct ata_smart_values new ){
int ataCheckDevice( atadevices_t *drive){
struct ata_smart_values tempsmartval;
struct ata_smart_thresholds tempsmartthres;
- int failed;
+ int failed,fd;
char *loc,attributename[64];
+ // 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)
+ return 1;
+
// Coming into this function, *drive contains the last values measured,
// and we read the NEW values into tempsmartval
- if (ataReadSmartValues(drive->fd,&tempsmartval))
+ if (ataReadSmartValues(fd,&tempsmartval))
printout(LOG_INFO, "%s:Failed to read SMART values\n", drive->devicename);
// and we read the new thresholds into tempsmartthres
- if (ataReadSmartThresholds (drive->fd, &tempsmartthres))
+ if (ataReadSmartThresholds(fd, &tempsmartthres))
printout(LOG_INFO, "%s:Failed to read SMART thresholds\n",drive->devicename);
- // See if any vendor attributes are below minimum, and print them out
+ // See if any vendor attributes are below minimum, and print them
+ // out. WHEN IT WORKS, we should here add a call to
+ // ataSmartStatus2() either in addition to or instead of the
+ // ataCheckSmart command below. This is the "right" long-term
+ // solution.
if ((failed=ataCheckSmart(tempsmartval,tempsmartthres,1))){
ataPrintSmartAttribName(attributename,failed);
// skip blank space in name
@@ -291,18 +332,15 @@ int ataCheckDevice( atadevices_t *drive){
printout(LOG_CRIT,"Device: %s, Failed SMART attribute: %s. Use smartctl -a %s.\n",
drive->devicename,loc,drive->devicename);
}
-
- // WHEN IT WORKS, we should here add a call to ataSmartStatus2()
- // either in addition to or instead of the ataCheckSmart command
- // above. This is the "right" long-term solution.
// see if any values have changed. Second argument is new values
- ataCompareSmartValues (drive , tempsmartval);
+ ataCompareSmartValues(drive, tempsmartval);
// Save the new values into *drive for the next time around
drive->smartval = tempsmartval;
drive->smartthres = tempsmartthres;
+ closedevice(fd);
return 0;
}
@@ -312,10 +350,16 @@ int scsiCheckDevice( scsidevices_t *drive){
UINT8 returnvalue;
UINT8 currenttemp;
UINT8 triptemp;
-
+ int fd;
+
+ // 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)
+ return 1;
+
currenttemp = triptemp = 0;
- if (scsiCheckSmart( drive->fd, drive->SmartPageSupported, &returnvalue, ¤ttemp, &triptemp ) != 0)
+ if (scsiCheckSmart(fd, drive->SmartPageSupported, &returnvalue, ¤ttemp, &triptemp))
printout(LOG_INFO, "%s:Failed to read SMART values\n", drive->devicename);
if (returnvalue)
@@ -327,32 +371,34 @@ int scsiCheckDevice( scsidevices_t *drive){
// Seems to completely ignore what capabilities were found on the
// device when scanned
if (currenttemp){
- if ( (currenttemp != drive->Temperature) && ( drive->Temperature) )
+ if ((currenttemp != drive->Temperature) && (drive->Temperature))
printout(LOG_INFO, "Device: %s, Temperature changed %d degrees to %d degrees since last reading\n",
drive->devicename, (int) (currenttemp - drive->Temperature), (unsigned int) currenttemp );
-
drive->Temperature = currenttemp;
- }
+ }
+ closedevice(fd);
return 0;
}
void CheckDevices ( atadevices_t *atadevices, scsidevices_t *scsidevices){
int i;
+ // If there are no devices to monitor, then exit
if (!numatadevices && !numscsidevices){
printout(LOG_INFO,"Unable to monitor any SMART enabled ATA or SCSI devices.\n");
return;
}
+ // Infinite loop, which checkes devices
printout(LOG_INFO,"Started monitoring %d ATA and %d SCSI devices\n",numatadevices,numscsidevices);
while (1){
- for (i = 0; i < numatadevices;i++)
- ataCheckDevice ( &atadevices[i]);
+ for (i=0; i<numatadevices; i++)
+ ataCheckDevice(atadevices+i);
- for (i = 0; i < numscsidevices;i++)
- scsiCheckDevice ( &scsidevices[i]);
+ for (i=0; i<numscsidevices; i++)
+ scsiCheckDevice(scsidevices+i);
- sleep ( checktime );
+ sleep(checktime);
}
}
@@ -363,7 +409,7 @@ int massagecvs(char *out,const char *in){
const char *savein=in;
// skip to I of $Id:
- while (*in !='\0' && *in!='I')
+ while (*in && *in!='I')
in++;
// skip to start of filename
@@ -442,10 +488,10 @@ int parseconfigfile(){
if (fp==NULL && errno!=ENOENT){
// file exists but we can't read it
if (errno<sys_nerr)
- printout(LOG_INFO,"%s: Unable to open configuration file %s\n",
+ printout(LOG_CRIT,"%s: Unable to open configuration file %s\n",
sys_errlist[errno],CONFIGFILE);
else
- printout(LOG_INFO,"Unable to open configuration file %s\n",CONFIGFILE);
+ printout(LOG_CRIT,"Unable to open configuration file %s\n",CONFIGFILE);
exit(1);
}
@@ -465,7 +511,7 @@ int parseconfigfile(){
// See if line is too long
len=strlen(line);
if (len>MAXLINELEN){
- printout(LOG_INFO,"Error: line %d of file %s is more than than %d characters long.\n",
+ printout(LOG_CRIT,"Error: line %d of file %s is more than than %d characters long.\n",
lineno,CONFIGFILE,MAXLINELEN);
exit(1);
}
@@ -488,7 +534,7 @@ int parseconfigfile(){
// We've got a legit entry
if (entry>=MAXENTRIES){
- printout(LOG_INFO,"Error: configuration file %s can have no more than %d entries\n",
+ printout(LOG_CRIT,"Error: configuration file %s can have no more than %d entries\n",
CONFIGFILE,MAXENTRIES);
exit(1);
}
@@ -511,25 +557,21 @@ int parseconfigfile(){
if (entry)
return entry;
- printout(LOG_INFO,"Configuration file %s contained no devices (like /dev/hda)\n",CONFIGFILE);
+ printout(LOG_CRIT,"Configuration file %s contained no devices (like /dev/hda)\n",CONFIGFILE);
exit(1);
}
-const char opts[] = { DEBUGMODE, EMAILNOTIFICATION, PRINTCOPYLEFT,'h','?','\0' };
+// const char opts[] = {DEBUGMODE, EMAILNOTIFICATION, PRINTCOPYLEFT,'h','?','\0' };
+const char opts[] = {DEBUGMODE, PRINTCOPYLEFT,'h','?','\0' };
-/* Main Program */
-int main (int argc, char **argv){
- atadevices_t atadevices[MAXATADEVICES], *atadevicesptr;
- scsidevices_t scsidevices[MAXSCSIDEVICES], *scsidevicesptr;
- int optchar,i;
+
+// Parses input line, prints usage message and
+// version/license/copyright messages
+void ParseOpts(int argc, char **argv){
extern char *optarg;
extern int optopt, optind, opterr;
- int entries;
-
- numatadevices=0;
- numscsidevices=0;
- scsidevicesptr = scsidevices;
- atadevicesptr = atadevices;
+ int optchar;
+
opterr=optopt=0;
// Parse input options:
@@ -550,7 +592,7 @@ int main (int argc, char **argv){
debugmode=1;
if (optopt) {
printhead();
- printout(LOG_INFO,"=======> UNRECOGNIZED OPTION: %c <======= \n\n",optopt);
+ printout(LOG_CRIT,"=======> UNRECOGNIZED OPTION: %c <======= \n\n",optopt);
Usage();
exit(-1);
}
@@ -574,6 +616,19 @@ int main (int argc, char **argv){
// print header
printhead();
+ return;
+}
+
+/* Main Program */
+int main (int argc, char **argv){
+ atadevices_t atadevices[MAXATADEVICES], *atadevicesptr=atadevices;
+ scsidevices_t scsidevices[MAXSCSIDEVICES], *scsidevicesptr=scsidevices;
+ int i,entries;
+
+ numatadevices=numscsidevices=0;
+
+ // Parse input and print header and usage info if needed
+ ParseOpts(argc,argv);
// look in configuration file CONFIGFILE (normally /etc/smartd.conf)
entries=parseconfigfile();
@@ -583,18 +638,20 @@ int main (int argc, char **argv){
daemon_init();
}
- // If we found a config file, look at its entries
+ // If we found a config file, register its entries
if (entries)
for (i=0;i<entries;i++){
+ // register ATA devices
if (config[i].tryata && atadevicescan(atadevicesptr, config[i].name))
printout(LOG_INFO,"Unable to register ATA device %s at line %d of file %s\n",
config[i].name, config[i].lineno, CONFIGFILE);
-
+ // then register SCSI devices
if (config[i].tryscsi && scsidevicescan(scsidevicesptr, config[i].name))
printout(LOG_INFO,"Unable to register SCSI device %s at line %d of file %s\n",
config[i].name, config[i].lineno, CONFIGFILE);
}
else {
+ // since there was no config file found, search all ATA and SCSI disks
char deviceata[] = "/dev/hda";
char devicescsi[]= "/dev/sda";
printout(LOG_INFO,"No configuration file %s found. Searching for devices.\n",CONFIGFILE);
@@ -604,6 +661,7 @@ int main (int argc, char **argv){
scsidevicescan(scsidevicesptr, devicescsi);
}
+ // Now start an infinite loop that checks all devices
CheckDevices(atadevicesptr, scsidevicesptr);
return 0;
}
diff --git a/sm5/smartd.h b/sm5/smartd.h
index 43d9a1b0e95b758720b9ee042f8c1fda2a8d6793..74fdb97653d7fa92da3ae70e7a4e5d6c2c422807 100644
--- a/sm5/smartd.h
+++ b/sm5/smartd.h
@@ -23,14 +23,24 @@
*/
#ifndef CVSID7
-#define CVSID7 "$Id: smartd.h,v 1.8 2002/10/25 14:15:05 ballen4705 Exp $\n"
+#define CVSID7 "$Id: smartd.h,v 1.9 2002/10/26 09:24:26 ballen4705 Exp $\n"
#endif
// Configuration file
#define CONFIGFILE "/etc/smartd.conf"
-#define MAXLINELEN 126
+#define MAXLINELEN 114
#define MAXENTRIES 64
+// BAD PROGRAMMING PRACTICE - GLOBAL VARIABLES SHOULD BE IN .c NOT .h
+// FILE
+/* how often SMART status is checked, in seconds */
+int checktime = 1800;
+// number of ATA and SCSI devices being watched
+int numatadevices;
+int numscsidevices;
+#define MAXATADEVICES 12
+#define MAXSCSIDEVICES 26
+
/* Defines for command line options */
#define DEBUGMODE 'X'
#define EMAILNOTIFICATION 'e'
@@ -40,42 +50,32 @@
#define TRUE 0x01
#define FALSE 0x00
-#define MAXATADEVICES 12
-#define MAXSCSIDEVICES 26
-
/* Global Variables for command line options */
+// These should go into a structure at some point
unsigned char debugmode = FALSE;
unsigned char emailnotification = FALSE;
unsigned char printcopyleft = FALSE;
-/* Number of ata device to scan */
-int numatadevices;
-int numscsidevices;
-
-
-/* how often SMART is checks in seconds */
-int checktime = 1800;
-
typedef struct atadevices_s {
int fd;
- char devicename[14];
int selftest;
struct hd_driveid drive;
struct ata_smart_values smartval;
struct ata_smart_thresholds smartthres;
+ char devicename[MAXLINELEN+2];
} atadevices_t;
typedef struct scsidevices_s {
int fd;
- char devicename[14];
unsigned char SmartPageSupported;
unsigned char TempPageSupported;
unsigned char Temperature;
+ char devicename[MAXLINELEN+2];
} scsidevices_t;
typedef struct configfile_s {
- char name[MAXLINELEN+2];
+ int lineno;
int tryata;
int tryscsi;
- int lineno;
+ char name[MAXLINELEN+2]; // really only needs to be +1
} cfgfile;