Incident Analysis of the intrusion on helium.ruby-lang.org

As already reported, helium.ruby-lang.org, which is one of the servers that provided various services relevant to Ruby development, was cracked by an unauthorized user. We, the ruby-lang.org administrators, are reporting our analysis of this intrusion and the countermeasures we’ve taken.

Time-line of the incident

The time progression about the intrusion is shown below. Each time is in UTC.

19 May       The public disclosure of a vulnerability of CVS
             (CAN-2004-0396) is announced.  We believe that this
             vulnerability was used in this intrusion.
20 May 02:46 The Debian CVS package of the host helium.ruby-lang.org
             (hereafter called 'helium') is upgraded. However, the
             chrooted CVS package, which provided the actual pserver,
             is overlooked.
23 May 11:15 Oldest (corroborated) time stamp of the trace of intrusion
27 May 19:03 Opening of the back-door installed by the invader
28 May 09:26 A ruby-lang.org administrator discovers a trace of the
             intrusion.
28 May 09:35 Administrator disconnects 'helium' from network.
28 May 11:53 Administrator reboots 'helium' and resumes mailing list
             services.
29 May 07:28 Our first announcement about this intrusion.

Machines and services of the time

Ruby-related services were offered by the following two machines at the time of intrusion.

helium.ruby-lang.org
The following services were provided by ‘helium’.
  • CVS (cvs.ruby-lang.org)
  • HTTP (www.ruby-lang.org/raa.ruby-lang.org)
  • FTP (ftp.ruby-lang.org)
  • RSYNC (for mirror sites)
  • ML (<ML name>@ruby-lang.org)
hydrogen.ruby-lang.org (hereafter called ‘hydrogen’)
The following services were provided by ‘hydrogen’.
  • HTTP (www.rubyist.net)
  • NFS (to export /home to ‘helium’)

Details of intrusion

On ‘helium’, the pserver service was offered under anoncvs user permissions in a chroot environment. This CVS service was used for development of Ruby, and multiple committers had their own accounts. Public read-only CVS access was allowed via the ‘anonymous’ user account.

As mentioned above, the vulnerability of CVS was announced on 19 May. Although the Debian CVS package of ‘helium’ was upgraded on 20 May, the CVS package of the chroot environment was not upgraded.

Under these circumstances, an administrator of ‘helium’ discovered dubious processes of anoncvs on 28 May, 09:26 (UTC). More than one suspicious, executable files were discovered to have been installed by intruder(s), including a program creating a back-door listening on TCP port #54320. This back-door program was running at the time of the discovery of the intrusion. The “Time-line” time mentioned above was determined by the output of the ‘ps’ command, and by the time stamp of the intrusive executable file. We should note here that all external connections to the back-door were prevented by the Linux kernel’s IP packet filtering feature.

Another note: usually, the pserver process creates a temporary directory (/tmp/cvs-serv<Process ID>) for each session and deletes it at the end of the session. At the time of the intrusion, more than one temporary directories were left in the /tmp directory of the chroot environment. This suggests that pserver processes were terminated abnormally, perhaps by the attack of the intruder(s). Of the time stamps of these temporary directories, the oldest is of 23 May, 11:15 (UTC). The “Time-line” time mentioned above was determined by this time stamp. Comparing time stamps of these directories with the pserver session log, it appears that there were multiple, independent crackers who attacked the vulnerability.

This evidence indicates that intruder(s) were exploiting the CVS vulnerability and received anoncvs user permission on ‘helium’. Intruder(s) could acquire, alter and destroy all information within the chroot environment.

Evidence such as the seizing of other accounts, local privilege escalation, or intrusion to outside of the chroot environment has, to date, not been found.

The possibility of intrusion outside of the chroot environment

In order for intruders to break chroot protection, promotion to a privilege user is required.

At the time of the intrusion, the Linux kernel running on ‘helium’ was version 2.4.24. The patch for the vulnerability (backported from kernel version 2.4.25) had been applied. However, the patch for setsockopt(2) vulnerability fixed in kernel 2.4.26 had not been applied.

Code for a DoS attack exploiting a vulnerable setsockopt(2) has been exhibited, but it is considered difficult to successfully achieve a local privilege escalation from this. It appears to be impossible to achieve privilege escalation if the intruder is unable to get the kernel image of the target environment. On ‘helium’, the kernel was not from a binary package, but was built from source code with individually applied patches. Therefore, the possibility that the intruder was able to achieve privilege promotion is considered minimal.

Restoration of services

With the first investigation, we inferred that the intrusion was probably only within the chroot environment. At first we decided to resume the mailing list service on ‘helium’, because the impact of a suspension of mail service, from the users’ point of view, was considered the largest. After checking that there was no alteration of the binary package, and also that there was no suspicious settings in the configuration files, we restored the mailing list service.

We then began the confirmation work for resuming other services on ‘helium’, but soon decided instead to thoroughly reconstruct the machine and resume services individually after each was checked. This decision was made because of the difficulty of inspecting the large number of files.

To restore the service, we needed a machine used as substitute for ‘helium’. We decided to use ‘hydrogen’ which hosted www.rubyist.net. ‘Hydrogen’ did not offer pserver service and no trace of intrusion was found on the machine, but ‘hydrogen’ provided /home as NFS-mounted filesystem to ‘helium’. To ensure that hydrogen was not compromised, we re-installed the OS of ‘hydrogen’, and set the host name to ‘lithium’. Then we moved the mailing list service from ‘helium’ to ‘lithium’, together with the intrusion announcement page of the web site.

Next, we re-installed OS of ‘helium’, and set the host name to ‘beryllium’. We plan to migrate all public services to ‘beryllium’ in the future.

Machines and services at present

At present, Ruby-related services are offered by the following two machines.

lithium.ruby-lang.org
The following services are provided by lithium.ruby-lang.org.
  • CVS (for committers development, no public access)
  • Mailing list (the move to ‘beryllium’ is planned)
beryllium.ruby-lang.org
The following services are provided by beryllium.ruby-lang.org.
  • HTTP (www.ruby-lang.org/raa.ruby-lang.org/www.rubyist.net)
  • FTP (ftp.ruby-lang.org)
  • Anonymous CVS (cvs.ruby-lang.org)

Verification of the contents of each service

We will hereafter explain the results of our efforts to evaluate whether there was any alteration or destruction of services by the intruders.

Precondition

The oldest evidence that we have of the intrusion is from 23 May, and this has been corroborated; since this trace could have been deleted by the intruder(s) with anoncvs user permission, we were not able to conclude that this was the first day of the intrusion. Since the evidence of the intrusion was from the CVS vulnerability, and since no other vulnerability which could have been exploited for intrusion into ‘helium’ is known, we are confident that the intruder(s) abused the CVS vulnerability and thereby gained access to ‘helium’.

Our verification of alteration or destruction in the services was based on the assumption that the first intrusion would prove to be after 19 May, when the CVS vulnerability CAN-2004-0396 was publicly announced.

CVS

Since the intruders are supposed to have gained anoncvs user permissions, we were most suspicious and anxious about possible damage to CVS among all of the services on ‘helium’.

At the time of intrusion, there was the following four CVS repositories.

/src
Source code
/www
Data of WWW
/doc
Document
/admin
The management file for CVS

Among these, /www and /doc needed no verification, since their contents were already unused. Moreover, we decided to suspend use of /admin, and simply dropped it.

What we explain hereafter are the results of verification of the source code of Ruby and of each other module included in /src.

The source code of Ruby

We split possible CVS repository alternations into two categories:

(1) Alteration of historical data in files in the CVS repository prior to 19 May

(2) Alteration which camouflaged the regular submissions after 19 May

For (1), we checked files in the CVS repository by the log of cvsup after 19 May which were kept securely outside of ‘helium’. We verified that there has been no sign of alteration of files in the CVS repository. For (2), we checked all contents of the commits individually and confirmed the absence of any malicious code after 19 May. This means that not only is there no malicious code, but also that we verified each commit with the committer.

Our verification was supported by data in the following URL.

  • Log of cvsup
    http://www.ruby-lang.org/check-data/cvs/cvsup-log/
  • Contents of the commits from 19 May to 28 May
    http://www.ruby-lang.org/check-data/cvs/cvs-diff/

Moreover, in addition to the above-mentioned material, we did the following supplemental work:

  • We confirmed there was no inconsistency between files in the CVS repository on ‘helium’ and files in an exterior, secure server on 21 May.
  • We confirmed there was no inconsistency within the CVS snapshots from 2003-11-02 to 2004-05-27 (day by day) and snapshots created from the CVS repository on ‘helium’.

We concluded that there was no alteration or destruction of the source code of ruby in the CVS repository.

Modules other than the source code of Ruby

In addition to the source code of Ruby, the /src directory of the CVS repository has the following modules:

  • app
  • lib
  • rough
  • rubicon
  • ruby-parser
  • shim
  • vms
  • pocketruby
  • oniguruma
  • mod_ruby
  • eruby

First, we found that only the following files were changed after 19 May by comparing the ctime of the files of the repository with the times of the files copied to the external server by CVSup:

  • lib/csv/lib/csv.rb,v
  • lib/csv/tests/csv_ut.rb,v
  • lib/soap4r/lib/wsdl/xmlSchema/parser.rb,v
  • lib/soap4r/lib/wsdl/xmlSchema/complexContent.rb,v
  • lib/soap4r/lib/wsdl/parser.rb,v
  • mod_ruby/lib/apache/eruby-run.rb,v
  • mod_ruby/lib/apache/erb-run.rb,v
  • mod_ruby/ChangeLog,v

Second, we compared the copied CVS repository with the CVS repository on ‘helium’, and we confirmed that there was no inconsistency among them, except for binary files in ‘pocketruby’. Since we had already merged ‘wince’ in ruby’s main branch, we did not do further check on pocketruby and have ceased offering its source code.

Each of the files which were changed after 19 May are mentioned below.

lib/csv/lib/csv.rb,v
lib/csv/tests/csv_ut.rb,v
lib/soap4r/lib/wsdl/xmlSchema/parser.rb,v
lib/soap4r/lib/wsdl/xmlSchema/complexContent.rb,v
lib/soap4r/lib/wsdl/parser.rb,v
We are unsure about these files. lib/csv and lib/soap4r are already merged with ruby, and these modules are only used by the maintainers of each. lib/csv and lib/soap4r have been removed from the CVS repository, and will be developed elsewhere.
mod_ruby/lib/apache/eruby-run.rb,v
mod_ruby/lib/apache/erb-run.rb,v
All revisions including branches had been checked, and no problems were found. They were each compared with the released source packages, and it has been confirmed that there are no inconsistencies.
mod_ruby/ChangeLog,v
Usual changes of a ChangeLog file are additions of the content(s). The ChangeLog can be checked with the following methods:

(1) We confirmed that there is no problem in the first revision.

(2) We confirmed that there is no problem in the newest revision.

(3) We confirmed all revisions which include changes, not only additions.

Moreover, we compared it with the released source packages, and it has been verified that there are no inconsistencies.

In addition, the development of mod_ruby and eruby has moved to Subversion, so those CVS module names have been changed to mod_ruby-old and eruby-old.

HTTP (www.ruby-lang.org)

http://www.ruby-lang.org/{ja, en}/ is generated by tDiary. We performed the following in order to check that there are no problems executing the tDiary CGI program:

  • Confirmation of the absence of suspicious code in the CGI programs
  • Verification of code in <script> elements embedded in the contents
  • Confirmation of the absence of suspicious data in the configuration files

Moreover, we checked the contents and linked URLs, but no problem is found. If any problems are found, please contact to webmaster@ruby-lang.org.

On-line reference manual

On-line reference manual was on RWiki. We first restored the contents on 29 Feb, then applied the patches mailed to external e-mail accounts on 29 Feb. Then we compared it with the contents on ‘helium’.

The diff can be obtained from:

http://www.ruby-lang.org/check-data/ruby-man/man-rd-ja.diff

The difference of Base64.rd comes from new-lines inserted when receiving the mail. trap%3A%3ANilClass.rd.rej was rejected, because the same patch applied twice. The diff script compared files with files from 61 minutes prior, so the same patch sent twice.

We confirmed none of them were affected by the intrusion.

RAA

We did the following data verification.

  • We made a daily diff of RAA data from 1) the clean RAA data copy backed up on 27 Mar, 2) daily backups from 4 Apr to 28 May, and 3) the latest RAA data of 28 May.

    2) and 3) are located in a chroot protected area on the machine. 1) is clean because it is kept in a development environment.

    • RAA data update:
      http://raa.ruby-lang.org/announce/soapbox-diff-all-passphrasemask.txt
    • RAA new entry:
      http://raa.ruby-lang.org/announce/soapbox-new-passphrasemask.txt
  • We confirmed the absence of suspicious data in the above diffs.

It can be concluded that the RAA data of 28 May (the same data we use for RAA service restart) does not include any suspicious data. We therefore decided to restart the RAA service as it was on 28 May. We cannot offer assurances that normal-looking changes by the intruder do not exist. For example, the change of sampleproject on 18 May is as follows;

== sampleproject
- updated: Sun May 09 12:35:19 GMT+9:00 2004
+ updated: Mon May 17 13:00:38 GMT+9:00 2004
- version: 0.0.8
+ version: 0.1.1

None of this data is suspect, but it is possible that the changes were made by the intruder. Therefore, we request that each RAA project owner CHECK THEIR RAA ENTRIES AND UPDATE THEM FOR CONFIRMATION. To do this, take the following steps:

(1) Open the project page

(2) Check the project information

(3) Go to the “update” page

(4) Press the “submit” button (do this even if no update is needed – this step is for confirmation)

Please contact raa-admin@ruby-lang.org if you find any suspicious data in RAA, or you have any question. Thank you for your cooperation.

FTP

We compared md5sum values of files on FTP with the files kept in the exterior, secure server, and there were no differences.

However, we could not check the following directories. Consequently, they are not currently provided.

/pub/ruby/contrib/
/pub/ruby/doc/
/pub/ruby/snapshots/
/pub/ruby/ML/
/pub/ruby/shim/

If you need files in these directories, please contact ftpadmin@ruby-lang.org.

Mailing list

We investigated the configuration files of each mailing list, and found no issues. However, member lists and mail archives have not been thoroughly checked.

If you have any problems, please contact to <ML name>-admin@ruby-lang.org.

Shugo Maeda <shugo@ruby-lang.org>
ruby-lang.org administrators’ group