GitLab

We changed the code to be compatible with the POSIX style.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

The current PrD version name has a 'ç' instead a 'c'.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

Some packages have versions like '1:version'. We are removing the '1:' part
to find the correct deb file.

Altered PrD version in file export to 3.0 Iguacu

We added the summary used in sdi web.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

The new SDI version need the state extension to be _st. So, we
are renaming those files. Also, we renamed the name of the
main function of those files, for the same reason.

We added the README file. This file explains how the system works
and how to install it.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

Since the task of generating the packs' version bases belongs now
to the user, we added a sample file that shows the format those files must
use.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

One variable was declared with a wrong name. This causes the crashes of the
system. We are fixating it.

Also, the function get_base will return the exit code of bunzip2.
Finally, we are commenting the actions we perform to fix the problems.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

Since we have now two kinds of packs bases, generate_mdsum_base.sh needs to
know what file it will process from. If the "dn" argument is passed, it will
use the "normal" packs base. If the "dt" argument is passed, it will use the
"testing" packs base.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

As explained before, we will have two different packs bases: one
for "normal" schools and another for "testing" ones.
We changed prd_packs to indicate what base the school will download, and
prd_packs.po to recive this indication and process it according to it.
Also, we are commentig the actions we perform to fix problems.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

We are removing both generate_packs_base.sh and packs_base, since we
changed the way we generate the packs base to an external schema.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

By now, the generation of the packs base will be done by an external
schema, since we were having some updating problems with the old way.
The schema used needs only to put two files (packs_base_{normal,testing})
contain the packs base for both kind of schools. The generation of the
md5 base will be done with those files.
Signed-off-by: Ricardo Tavares de Oliveira <rto07@c3sl.ufpr.br>

When the create partition process fail, the return value should be
consistent.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

We decide to give fixhome some more shots to try fixating the /home RAID.
Only after 22hrs will give not more than 2 shots, but between 22 and 5hrs it
will have enought time to run properly.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

After rsync, it's necessary to umount /home. If some user proccess are using
disk resources on /home, the umount command will fail, and will not bem
possible to fix the RAID.

Killing all user remaining proccess will allow the umount command to execute
properly.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

When something goes wrong running the fixhome routine, the school must
return to the state before running it. The code added will make sure all
becomes normal if some error happens.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

This modification prevent the creation of file /etc/nologin when the
fix home script will be aborted by some already created lock file.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

The lock file was been created at the begin of fix home proccess, and
removed on the end, but it wasn't checking if the lock file was there
before starting the routine.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

Diagnose the nobreak's states are an important task. If a energy failure
happens, and the nobreak isn't there, the filesystem can be severely
corrupted/damaged.

This script, and associate states, will be a warn to schools that haven't
installed their nobreak's properly.

Also, an entry to autogen.conf has been added.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

This will add prd_fixraid script to Makefile when generating it with
autogen.sh script.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

The schools disks state, as well as the RAID, it's very important to PRD
well functioning. This script will perform maintance to disks in a automated
way.

The main functions are:
- On new disks, depending the situation: copy the already running disk
  partitions and re-build the RAID, or create new partitions based on
  current state, fix the /home and re-build the RAID.
- On disks with I/O issues, the script will remove it from raid. The same
  happens on disks with FAULTY partitions.
- Re-build the RAID if the partitions are correct.

Schools with RAID5 will be ignored, because only 3 schools have this kind o
raid.

This is a delicate script, mainly by the fixhome function.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

When some partition was flaged as resync=DELAYED, the prd_raid script
returned a OUT_OF_RAID message, what's not true. Now, a RESYNC_DELAYED
message will be parsed.
Signed-off-by: Vinicius Kwiecien Ruoso <vkr07@c3sl.ufpr.br>

The FAULTY partitions detection was incorrect, because only one FAULTY
partition was returned to SDI. Now, all FAULTY partition will be reported.
Signed-off-by: Vinicius Kwiecien Ruoso <vkr07@c3sl.ufpr.br>

With the I/O error state, is possible to show all schools with this issue on
a single summary, making it easy to diagnose disk errors.
Signed-off-by: VINICIUS KWIECIEN RUOSO <vkr07@c3sl.ufpr.br>

When diagnosing raid state, it's important to find out if a disk have I/O
issues. If the error is detected, the script will stop cheking the arrays.
Signed-off-by: Vinicius Kwiecien Ruoso <vkr07@c3sl.ufpr.br>

Since the way the mirror is updated has been changed, some paths have
changed, too. This commit will fix the path issues.
Signed-off-by: Ricardo <rto07@c3sl.ufpr.br>

This fix prevent some issues with packs generation proccess. The new condition
to insert new packages to the base will take care of packages that aren't
listed in the base.
Signed-off-by: Ricardo <rto07@c3sl.ufpr.br>

We are adding 'disown' when calling functions in background. With this,
the function won't be son of the bash that opened it anymore.

This is a request by SDI.
Signed-off-by: Ricardo <rto07@c3sl.ufpr.br>

We are adding some scripts to check the packages' versions and the integrity
of the files in the disks. Basically, this tool is divided into two main
features: the checking of packages with their versions and the checking of
files with their md5sum hashes.

To check packages' versions, we are using a scenario composed by a base file
hosted in the SDI server and scripts sent to the SDI hosts. The base file is
just a list of packages' versions that we judge as correct, it means, a
host has a installed package in a wrong version if and only if the version
installed in the host is not equal to the version present in the base file.
So, we have a script called "generate_pack_base.sh" which generates the base
file under $PRDDIR/packs/, with the name "packs_base". Notice that the SDI
server must host a Debian mirror to do this. Notice also that we update the
crontab to run this script daily.

With the base file correctly generated in the SDI server, we will send some
scripts to the hosts. First of all, we send, under "onconnect.d", the
"prd_packs" script, which contains the function that will check the packages'
versions. Then, under "hourly.d", we send the "prd_integrity" script, that
will call the function defined in "prd_packs". Basically, the function
downloads the base file from the SDI server (via sendfile feature) and
compare the versions defined in the base file with the versions installed
in the hosts. Depending on the differences found between both, prd_packs
will print some different messages and make the host enter into different
states. The states defined by prd_packs script are:

prequest.st: The host is downloading (or waiting to) the base file;
dpkg_configure_error.st: The host cannot install any package because
    requires 'dpkg --configure -a';
dpkg_error.st: 'dpkg -l dpkg' is returning an error code;
empty_cron_error.st: The host has no scheduled updates;
perror1.st: Some package is not correctly installed;
perror2.st: Some package is installed with an old version;
perror3.st: Some package is useless (installed in the host, but not present
    in the base file);
perror4.st: Some package is installed with an newer version;
packs_pok.st: Everything is ok.

When some error is found, prd_packs tries to fix it by downloading and
installing the failed packages again.

The scenario of checking files with their md5sum is very similar. The SDI
server hosts some base files and the SDI hosts compare their files with the
ones present in the base file. However, we have some differences here.
First of all, we have the "generate_mdsum_base.sh" script, which will
generate the base files by the same way. This script will be updated in the
crontab, too. But now we have two different base files to generate. This is
because we have, in PRD, two classes of schools: Regular ones and Testing
ones. Schools in different classes may have different files. Because of
this, we are gerenating base files for both classes. The bases will be
present under $PRD/mdsum/base_md5_{normal,test}.bz2. Since we are
checking all files in all packages, the base files are too big. That's the
cause of compressing them with the bunzip2 format.

Like prd_packs, we send, under "onconnect.d", the function that will check
the mdsums. "prd_integrity", after calling the prd_packs' function, will
call the prd_md5's function.

Since the compacted file base still too big, the SDI hosts shouldn't
download it every time prd_integrity is called. Instead, the SDI hosts
download it once, keep in the disk and use the same file in the next time.
If the base file must be updated, the hosts will download it again. That's
the cause of having a "prd_export" hourly script. This script only send to
the hosts the md5sum of the base files. With this, the hosts will know if
the file they have needs to be updated by checking its md5sum. Like
prd_packs, prd_md5 can generate some states. They are:

mrequest.st: The host is downloading (or waiting to) some base file;
merror.st: Some file is corrupt (its mdsum is not equal to the one
    defined in the base file);
md5_mok.st: Everything is ok.

When some error is found, prd_md5 tries to fix it by downloading and
installing the package which contains the failed file.

Finally, a Makefile has been created to set this scenario. It can
install, uninstall and purge the prd_packs and prd_md5 scripts.
Signed-off-by: Ricardo <rto07@c3sl.ufpr.br>

When detecting if the school will be refuted or not it's necessary
to check the /etc/prd_version file. If the version is the 1.1 or some
2.0rc, the school must be refuted.

The problem happens when the school should not be refuted, but the
return statement wasn't work like expected, and the script keep
runnning, bloking the school.

Removing the return command and adding a block based "if" command
has solved this issue.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

To refute old PrD version, such as 1.1 and 2.0rc*, this script will create
a /etc/nologin file with some instructions and add on /etc/rc2.d to secure
that the nologin file will be there when the system is rebooted.

Also an entry to refute script on autogen.conf has been added to correctly
generate a new Makefile. This entry has been added before prd_ajudaremota
script, which was moved to 002 position.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

The hostname command will not work on schools there are with remote help
activated. The school information is retrieved from prd.zone but using
the IP address instead of school id.

The ME variable was not being used on this script. It comes from the
current prd_schoolinfo script that retrieves the hostname with the
'hostname' command.
Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>

Signed-off-by: Vinicius Ruoso <vkr07@c3sl.ufpr.br>