Friday Jun 12, 2009

Encrypted Scratch Space in OpenSolaris 2009.06

Last week, I announced the availability of a set of scripts that could be used to enable encrypted swap in OpenSolaris 2009.06. Building upon this concept, today, I am happy to announce a new set of scripts that enables the creation of an encrypted file system (intended to be used as scratch space).

The method for creating these encrypted file systems is similar to the approach discussed by Darren in his posting on the topic of Encrypting ZFS Pools using LOFI. I had been working on a similar model for the Immutable Service Container project where I had wanted to be able to give each OpenSolaris zone that was created its own place to store sensitive information (such as key material) that would be effectively lost when the system was rebooted (without requiring a time-consuming disk scrubbing process).

The way these scripts work is quite simple. There is an SMF service, called isc-encrypted-scratch, that (if enabled) will automatically create encrypted scratch space for the global zone as well as any non-global zones on the system (by default). The creation of encrypted scratch space is configurable allowing you to specify which zones (including the global zone) can have one. You can specify which ZFS file system can be used as the home directory for the scratch space hierarchy. Using SMF properties and standard SMF service configuration methods, you can also specify the size of the encrypted scratch space.

Once created, you will have access to a ZFS file system (based upon a ZFS pool which itself is based upon an encrypted LOFI which itself is based upon a ZFS zvol - crazy eh?) The file systems created for the encrypted scratch space are destroyed and re-created upon boot (or service restart). Just as with the encrypted swap scripts, the encrypted LOFIs use ephemeral keys in conjunction with the AES-256-CBC cipher.

So, without further ado, let's get to the particulars. To enable encrypted scratch in OpenSolaris 2009.06, you need only follow the following steps.

Note that the following instructions assume that privileged operations will be executed by someone with administrative access (directly or via Solaris role-based access control). For the examples below, no changes were made to the default RBAC configuration. The commands as written were executed as the user created during the installation process.
  • Add the Encrypted Scratch Space SMF service. First, you will need to download the archive containing the encrypted scratch space SMF service manifest and method files. Note that these files are user contributed and as such are not officially a part of the OpenSolaris release nor are they officially supported by Sun. If you are ok with these terms, you should now download the archive and install the files using the following commands:

    $ wget -qnd
    $ bzip2 -d -c ./smf-encrypted-scratch-v0.1.tar.bz2 | tar xf -
    $ cd ./smf-encrypted-scratch
    $ pfexec ./
    $ svccfg import /var/svc/manifest/site/isc-enc-scratch.xml
  • Configure the Encrypted Scratch Space Service. Unlike the Encrypted Swap SMF Service, this service is not enabled automatically. This is to allow you the opportunity to adjust its configuration should you want to change any of the following properties:
    • config/scratch_root. This property defines the root ZFS file system to be used for the scratch space hierarchy. By default, it is set to rpool/export. Based upon this value, a collection of scratch files will be created under this location (each in its own directory tied to the name of the zone).
    • config/scratch_size. This property defines the size of the scratch space. This value is used during the initial creation of a ZFS volume (zvol) and accepts the same values as would be accepted by the zfs create -V command. The default size is 100 Mbytes. Note that today, each individual encrypted scratch space on a single system must be the same size.
    • config/zone_list. This property defines the zones for which encrypted scratch space will be created. By default, this is all zones including the global zone. Setting this value to a zone or list of zones will cause encrypted scratch spaces to be only created for those specified.

    For example, to configure this service to create 1Gbyte encrypted scratch spaces, use the command:

    $ svccfg -s isc-encrypted-scratch setprop config/scratch_size = 1g
    $ svcadm refresh isc-encrypted-scratch

  • Enable the Encrypted Scratch Space Service. Once you have finished configuring the service, you can enable it using the standard SMF method:

    $ svcadm enable isc-encrypted-scratch

  • Verify the Encrypted Scratch Space Service. To verify that the service is operating correctly, you can use the following commands to verify that everything has been properly created. First, let's make sure the service is running:

    $ svcs isc-encrypted-scratch
    STATE          STIME    FMRI
    online         12:40:02 svc:/system/isc-encrypted-scratch:default

    Next, let's verify that all of the proper ZFS mount points and files have been created. Note that the scratch root in this case is the default (rpool/export) and under this location a new scratch file system has been created under which there is a file system for each zone on the system (global and test). For each zone, a 1 Gbyte scratch file has been created.

    $ zfs list -r rpool/export/scratch
    NAME                                       USED  AVAIL  REFER  MOUNTPOINT
    rpool/export/scratch                      2.00G  5.21G    19K  /export/scratch
    rpool/export/scratch/global               1.00G  5.21G    19K  /export/scratch/global
    rpool/export/scratch/global/scratch_file     1G  6.21G  1.15M  -
    rpool/export/scratch/test                 1.00G  5.21G    19K  /export/scratch/test
    rpool/export/scratch/test/scratch_file       1G  6.21G  1.15M  -

    Next, let's verify that the encrypted LOFIs have been created. The mapping of the device files back to the actual scratch file zvols is left as an exercise for the reader.

    $ lofiadm
    Block Device             File                           Options
    /dev/lofi/1              /devices/pseudo/zfs@0:1c,raw   Encrypted
    /dev/lofi/2              /devices/pseudo/zfs@0:2c,raw   Encrypted

    Next, let's verify that new zpools and ZFS file systems have been created from the encrypted LOFIs:

    $ zpool list
    rpool           11.9G  4.06G  7.88G    34%  ONLINE  -
    scratch-global  1016M    82K  1016M     0%  ONLINE  -
    scratch-test    1016M    82K  1016M     0%  ONLINE  -
    $ zpool status scratch-global scratch-test
      pool: scratch-global
     state: ONLINE
     scrub: none requested
            NAME           STATE     READ WRITE CKSUM
            scratch-global  ONLINE       0     0     0
              /dev/lofi/1  ONLINE       0     0     0
    errors: No known data errors
      pool: scratch-test
     state: ONLINE
     scrub: none requested
            NAME           STATE     READ WRITE CKSUM
            scratch-test   ONLINE       0     0     0
              /dev/lofi/2  ONLINE       0     0     0
    errors: No known data errors
    $ zfs list /scratch-\*
    scratch-global    70K   984M    19K  /scratch-global
    scratch-test      70K   984M    19K  /scratch-test

  • (Optional) Add Encrypted Scratch Space to a Non-Global Zone. At this point, you have everything that you need to get started. In fact, for the global zone, there are no further steps, but you can now assign the scratch space to a non-global zone (if desired) using the standard zonecfg mechanisms. For example, you could do the following:

    $ pfexec zonecfg -z test
    zonecfg:test> add dataset
    zonecfg:test:dataset> set name=scratch-test
    zonecfg:test:dataset> end
    zonecfg:test> verify

  • (Optional) Verify Encrypted Scratch Space in a Non-Global Zone. Once booted, the new encrypted scratch space data set will be made available to the non-global zone:

    $ pfexec zlogin test
    [Connected to zone 'test' pts/2]
    Last login: Fri Jun 12 09:57:43 on pts/2
    root@test:~# zpool list scratch-test
    scratch-test  1016M  74.5K  1016M     0%  ONLINE  -
    root@test:~# df -k /scratch-test
    Filesystem            kbytes    used   avail capacity  Mounted on
    scratch-test         1007616      19 1007546     1%    /scratch-test

    Upon reboot, each of the zones will be shut down before the encrypted scratch space is destroyed. Note that upon global zone or service restart, the encrypted scratch space will be re-created and therefore will not persist across global zone reboots. The encrypted scratch space will persist across non-global zone reboots.

    There you have it! Enabling encrypted scratch in OpenSolaris 2009.06 (for the global and non-global zones) is as easy as following these few simple steps. It is worth stating that this solution is just a temporary workaround. Once ZFS encryption is available, it should be used instead of this approach. In the meantime, however, if you are interested in enabling encrypted scratch on your OpenSolaris 2009.06 systems, give this model at try and please be sure to send along your feedback!

    Take care!

    P.S. Some of you may be wondering why the SMF service and associated files are labeled with an ISC prefix? The answer is simple. They were developed and are being used as part of the Immutable Service Container project! Look for more information and materials from this project in the near future!

    Technorati Tag:

Thursday Jun 11, 2009

Impacting Solaris 10 Security Guidance

It is that time again! Work is kicking up over at the Center for Internet Security to update the Solaris 10 security benchmark. As I have previously covered, Sun has been working hand-in-hand with the Center for Internet Security for more than six years to develop best-in-class security hardening guidance for the Solaris operating system.

In recent years, the NSA and DISA have jumped in contributing their time and expertise towards the development of a unified set of Solaris security hardening guidance and best practices. Now is the time for the next step. Over the last several months, these groups have been working to comb through and integrate the recommendations found in the DISA UNIX STIG (Security Technical Implementation Guide) and associated checklist as it relates to Solaris. With this work now complete, an effort has been launched to develop a new draft CIS Solaris 10 Benchmark with these additions.

In addition to this effort, a secondary effort will soon be undertaken to update the Solaris 10 Benchmark for the latest release of the Solaris 10 05/2009 (Update 7). Currently, the Solaris 10 Benchmark supports Solaris 10 11/08 (Update 4). There are not that many things added to Solaris 10 since Solaris 10 11/08 that impact the hardening guide, but there are some items that will impact the Solaris Security Appendix that was published with the last version of the Benchmark.

The reason for my post today, however, is to say that the time is right if you are interested in Solaris, security, and want to get involved! We are always looking for people with a passion to help develop and improve the recommendations and settings in the Solaris 10 Benchmark. Want to learn more? Contact CIS!

P.S. Just in case you missed it - Sun and CIS also announced the availability of a security hardened virtual machine image based upon OpenSolaris for Amazon's EC2 (SunCloud will also be supported). Give it a try!

Take care!

Technorati Tag:

Monday Mar 09, 2009

NEW: Solaris Package Companion v0.9

Today, I am proud to announce the release of version 0.9 of the Solaris Package Companion. This new version is primary based upon a set of patches provided by Jerome Blanchet that provided support for the collection and processing of reverse dependency information ("R" entries in a package's depend(4)) file) as well as enhanced processing and display of platform specific packages. Thank you, Jerome!

As is my tradition when a bug is found, I try and publish a little something extra as a mea cupla. Due to limited "free time", the "mea culpa" enhancement this time is quite minor but worth mentioning anyway. In past versions, there was no interface to change the information collection rules of the tool. If you wanted to disable the collection of certain types of information (such as package names or dependencies) you had to go into the code and change the relevent COLLECT_ variable. No longer. The defaults are still the same, but now the tool will honor variable settings originating from the shell or command line as follows:

$ env COLLECT_NAMES=0 ./spc-v0.9.ksh -r /tmp/myrepository -i -l

Note that the COLLECT_ variables are only used during the creation of a repository. Not a big enhancement, but one none the less! Thank you again Jerome for discovering the bug and offering a patch!

Keep the suggestions, reports and fixes coming!


Technorati Tag:

Wednesday Dec 10, 2008

mod_privileges for Apache HTTPD

Special thanks to Matt Ingenthron for pointing out that mod_privileges has been integrated back in the Apache trunk (manual) recently. For more information check out NIQ's Soapbox posting on the subject.

Looks like I will have to find a new target (I am looking at you MySQL!) for my BluePrints. I have used the Apache with SMF privileges example in a few publications including Limiting Service Privileges in the Solaris 10 Operating System (2005) and Privilege Debugging in the Solaris 10 Operating System (2006). The content of these papers is still relevant in the general sense, but with the introduction of mod_security, some of this content will no longer be as useful for Apache. That said, lots of other services can and do benefit from the techniques described.

If you find yourself ever wanting to do something similar - converting your services to be privilege aware on Solaris 10, check out the Sun BluePrints article Privilege Bracketing in the Solaris 10 Operating System (2006). Also, check out the OpenSolaris Security Community project on Privilege Debugging as it can help you in finding out what privileges your programs and services need.

Until next time!


Technorati Tag:

Tuesday Aug 05, 2008

NEW: Solaris 10 Security Deep Dive Presentation

Way back when, I posted an update to the original Solaris 10 Security Deep Dive presentation that included support for Solaris 10 Update 3 (11/06). Well, it has been entirely too long since the last update, so I am happy to say that the wait has ended! A new version of the talk is ready for download! This has been quite a journey and a lot has changed in Solaris since it was first released back in 2005. If you have not taken a look into what Solaris can offer recently, I am sure you will be in for a pleasant surprise. Give it a look, and as always feedback is appreciated! Take care!


Technorati Tag:

Monday Aug 04, 2008

NEW: Solaris Package Companion v0.8.1 / Testing Tool v0.1

On the heels of the v0.8 release, Clive King was able to find a new bug introduced as a result of my attempting to make the code a little more in line with Korn Shell conventions. Clive, thank you for reporting the details! I have published an updated version as v0.8.1. As always, you can get all of the details at the OpenSolaris Solaris Package Companion Project Page

As is my tradition when a bug is found, I try and publish a little something extra as a mea cupla. This time is no different. In addition to version 0.8.1 of the Solaris Package Companion, I have also published a testing tool for the same.

The testing tool, called spc-test-v0.1.ksh is also available from the project page. This tool can test multiple versions of the tool against multiple repositories which is pretty cool when checking for regressions. There are currently 48 tests although tests can be easily added or removed as needed. It can optionally display the results to the screen, but by default it records them in a directory where a basic consistency check is performed to detect differences in output (for the same repository) resulting from the use of different versions of the tool. This is not intended to be an all encompassing test suite or even a piece of production code, but rather a basic sanity check to make sure the key functions are working as expected.

Thanks again, Clive!

Keep the suggestions, reports and fixes coming!


Technorati Tag:

Friday Aug 01, 2008

NEW: Solaris Package Companion v0.8

Wow, has time passed since my last posting. I promise to do a quick update soon as a lot has been happening over the last six months! In the meantime, I wanted to tell you all about a new version of the Solaris Package Companion (version 0.8) that is now available.

For those not familiar with the tool, here is a brief overview:

   The Solaris Package Companion is a small Korn shell script that allows you to ask
   quite a number of interesting questions about the relationships between Solaris 
   metaclusters, clusters and packages as well as their respective dependencies. Very
   often, answers to these kinds of questions are essential for the construction of 
   minimized systems as well as more generally for OS golden images.

   The goal of the Solaris Package Companion, or SPC for short, is to do all of the 
   hard work so you don't have to. SPC will create a cache of important facts by mining
   information from the various packaging files and directories to allow you to quickly 
   and easily obtain answers to a variety of questions such as:

     \* What clusters or packages are contained in a given metacluster?
     \* What packages are contained in a given cluster?
     \* What metacluster or cluster contains a given package?
     \* On what other packages does a given package or cluster depend?
     \* Which packages depend on a given package?
     \* … and so on…

New to this release is a tree view display method that allows you to list the contents of metaclusters and clusters in a more eye-friendly tree-view. Thanks to Fredrich Maney for contributing the idea and code! Here are a few examples from the project page showing what this looks like:

To see what packages are included in a cluster, just use the "-t" option:

$ ./spc-v0.8.ksh -v -r ./myrepository -t SUNWCssh
   [C] SUNWCssh                  Secure Shell
      [P] SUNWsshcu                 SSH Common, (Usr)
      [P] SUNWsshdr                 SSH Server, (Root)
      [P] SUNWsshdu                 SSH Server, (Usr)
      [P] SUNWsshr                  SSH Client and utilities, (Root)
      [P] SUNWsshu                  SSH Client and utilities, (Usr)

To see what packages and clusters are included in a metacluster, just use the "-T" option:

$ ./spc-v0.8.ksh -v -r ./myrepository -T SUNWCmreq | head -10
[M] SUNWCmreq                 Minimal Core System Support
   [C] SUNWCfca                  Sun ISP Fibre Channel Device Drivers
      [P] SUNWqlc                   Qlogic ISP 2200/2202 Fibre Channel Device Driver
      [P] SUNWemlxs                 Emulex-Sun LightPulse Fibre Channel Adapter (FCA) driver (root)
   [C] SUNWCfct                  Sun Fibre Channel Transport Software
      [P] SUNWfcsm                  FCSM driver
      [P] SUNWfctl                  Sun Fibre Channel Transport layer
      [P] SUNWfcp                   Sun FCP SCSI Device Driver
      [P] SUNWfcip                  Sun FCIP IP/ARP over FibreChannel Device Driver
   [C] SUNWCfmd                  Fault Management Daemon and Utilities

I would also like to thank Peter Pickford for sharing a fix for a bug that resulted in the tool not properly recording all dependencies under certain circumstances. Thank you! While I was at it, I also took a little time to clean up the code a bit.

You can find more information, examples and the source code on the project page.

Keep the suggestions, reports and fixes coming!


Technorati Tag:

Friday Jan 04, 2008

UPDATED: Solaris - Now With More Fuzz

Every six months or so, I try to do a run of my fuzz tests against the Solaris OS. The first test was conducted a year ago with build 42 followed by a test during our summer break on build 68 of Nevada. It should come as no shock then that I conducted another test during the winter break on build 80.

The tools and methodology are the same (although there are still some kinks to be worked out to make it fully automated), but for those who have not read my earlier post, I will summarize. The tests were conducted on a fresh installation of Nevada build 80 built with the SUNWXCall (Entire + OEM) installation cluster. A sparse-root, non-global zone (called "fuzz") was created for the tests and the software was loaded into the zone. Next, the names of all of the ELF binaries were collected, using the make-exec-list script run from within in the non-global zone. Next, the make-fuzz-tests script was run to generate the 36 different fuzz files to be used as input for each binary tested. Lastly, the test was kicked off using the exec-fuzz-tests script. The script pretty much runs unattended except when I need to kill off runaway processes. I still need to add some code to kill off anything started at the end of each test so you do not end up with tons of extra processes running and consuming memory.

At any rate, the test run completed and I have posted my results in Bugster and the bugs are also available in the OpenSolaris Bug Database Search using the keyword fuzz. The programs impacted can be viewed using this query.

While I tend to do this kind of work for fun as a holiday distraction, it does have real benefit. Programs that fail during a fuzz test (usually core dumping although a runaway or two have also been found) fail due to unvalidated input that leads to a buffer overflow or arithmetic exception of some kind. Input validation is not to be taken lightly and should be performed by every program and service. In fact, on the CERT Top 10 Secure Coding Practices list, validate input is item #1 and with good reason.

Take care,


Technorati Tag:

Tuesday Nov 13, 2007

NEW: Hack-Fu - Deconstructing the Security Capabilities of the Solaris 10 OS

For the Sun CEC 2007 conference this year, I revamped my originalPractical Solaris 10 Security presentation that I had originally mentioned here. The new version of the presentation is titled Hack-Fu - Deconstructing the Security Capabilities of the Solaris 10 OS.

While the title is a little more "catchy", the real change is that the presentation was enhanced to provide a more complete practical demonstration of Solaris 10 security capabilities. The presentation is structured from the viewpoint of a potential attacker examining the system from the network. As each new capability is discussed, barriers are lifted -- one by one -- until the attacker is given root access inside a Solaris 10 non-global zone.

While I have not had a chance to record the talk putting audio to the slides, you can still follow along as many of the examples in the presentation are based upon Sun BluePrints and HOWTOs that have already been published such as:

and a few others. I am always tuning and tweaking these presentations to address new features, improve their clarity, and make the examples more realistic. So, be sure to give it a look and send along your feedback. Also, don't forget to check out the OpenSolaris Security Community Presentations Library for other presentations featuring Solaris 10 and OpenSolaris content!

Take care,


Technorati Tag:

Friday Nov 02, 2007

NEW: Solaris Package Companion v0.7

This one must have slipped my mind. Please accept my apologies. Back in September (2007), I published an updated version of the Solaris Package Companion. For those not familiar with the tool, here is a brief overview:

   The Solaris Package Companion is a small Korn shell script that allows you to ask
   quite a number of interesting questions about the relationships between Solaris 
   metaclusters, clusters and packages as well as their respective dependencies. Very
   often, answers to these kinds of questions are essential for the construction of 
   minimized systems as well as more generally for OS golden images.

   The goal of the Solaris Package Companion, or SPC for short, is to do all of the 
   hard work so you don't have to. SPC will create a cache of important facts by mining
   information from the various packaging files and directories to allow you to quickly 
   and easily obtain answers to a variety of questions such as:

     \* What clusters or packages are contained in a given metacluster?
     \* What packages are contained in a given cluster?
     \* What metacluster or cluster contains a given package?
     \* On what other packages does a given package or cluster depend?
     \* Which packages depend on a given package?
     \* … and so on…

New to this release is the tag before the item description to inform the user of the type of object being dispayed. [P] indicates a package while [C] is a cluster and [M] is a metacluster. Another new feature is the ability to fold packages back into their respective clusters (where possible). This can be helpful when trying to create a complete list of items for a standard OE image or JumpStart configuration. Essentially, this will report the cluster name in which the package is found. This can be accomplished using the -F (folding) option. The new -Z option will display the list of packages that depend on a specific cluster. There is also an new experimental option -f that will allow you to map a file to a package or cluster (with the -F option). This only works for local files reliably right now. Finally, special thanks to Dave Comay for reporting a bug - that has been fixed in this version too!

You can find more information, examples and the source code on the project page.

Technorati Tag:

NEW: Solaris 10 Set-ID and World Writable Overview

Various organizations have often asked for more detail regarding the set-uid, set-gid and world writable programs that are shipped by the default in the Solaris OS. Well, the wait is over (at least for Solaris 10 8/07)!

Today, I am happy to announce the public release of an overview document that describes these file system objects in detail. This document is still a draft and could still needs to answer a few questions, but I believe that it is far enough along to open up the discussion and begin getting feedback from all of you! If you are interested and want a copy of the document, you can find it here. Looking forward to your comments!

From the document:

While there are often many files delivered by operating systems and other software products, organizations are often most concerned with those programs and services that have or run with special privilege. Unfortunately, there is at times a lack of information regarding what these programs do and why their privileges are necessary. The goal of this document is to provide additional information on four special classes of objects delivered by the Solaris OS: Set-UID Files, Set-GID Files, and World Writable Directories and Files. With this information, organizations will be able to better understand the privileged programs, directories and files that exist on their systems.

If you would like to make recommendations or even implement an improvement (such as one of the RFEs listed in the document), please consider joining the OpenSolaris Security Community!


Technorati Tag:

Wednesday Aug 01, 2007

Solaris Non-Executable Stack Concluded

Since publishing my two part series on non-executable stacks in the Solaris operating system, I received some very useful feedback and clarifications that I wanted to share with everyone. First, Vladimir Kotal commented on my first article that:

Having to grep(1) for the CPU features is really clumsy. Maybe psrinfo(1M) could be extended to print them out? (for every (virtual) CPU present in the system)

Frankly, I agree. After asking around however, today there does not appear to be a cleaner interface (although there is a bunch of discussion around adding one). Sherry Moore and Joe Bonasera were kind enough to point out that there is a programmatic way to access this information in the form of cpuid(7d). Joe also shared the following with me that you may find interesting:

The NX information doesn't belong in isainfo. isainfo, I'm told, is only meant to reflect processor capability information that is directly usable from user mode.

The NX bit feature has to do with page table construction which is not something you do from userland. What's a more interesting thing to know is "Does not specifying PROT_EXEC have any effect on this system, or is PROT_EXEC implicit for all PROT_READ segments?" Even cpuid doesn't help with that information as various bits of the OS memory subsystems might do different things along the way. For example if for some reason you're running a non-PAE 32 bit kernel, even though cpuid says that NX is supported, NX bits wont be used.

A similar issue has come up in the Open Solaris Xen project, in that many people want to know if their processor supports AMD-V or Intel VT-x. That information comes from CPUID, but is only usable from supervisor (either kernel or hypervisor) code, hence we haven't added it to isainfo. But it is a valid question to ask if the cpu/bios you have would support running such software w/o actually having it.

That said, Sherry did clue me in on a program called cpuid which can allow us to get this information and a lot more (subject to the issues noted by Joe above). Unfortunately, the cpuid program was developed for Linux and will not compile by default on Solaris:

blackhole$ gmake
cc -g -Wall -Wshadow -Wcast-align -Wredundant-decls -Wbad-function-cast -Wcast-qual -Wwrite-strings -Waggregate-return 
-Wstrict-prototypes -Wmissing-prototypes -D_FILE_OFFSET_BITS=64 -DVERSION=20070801 -o cpuid cpuid.c
cpuid.c:26:25: linux/major.h: No such file or directory
cpuid.c: In function `explain_errno':
cpuid.c:3191: error: `CPUID_MAJOR' undeclared (first use in this function)
cpuid.c:3191: error: (Each undeclared identifier is reported only once
cpuid.c:3191: error: for each function it appears in.)
cpuid.c: In function `real_setup':
cpuid.c:3472: warning: implicit declaration of function `makedev'
cpuid.c:3472: error: `CPUID_MAJOR' undeclared (first use in this function)
cpuid.c: In function `main':
cpuid.c:3751: warning: initialization discards qualifiers from pointer target type
cpuid.c:3752: warning: initialization discards qualifiers from pointer target type
cpuid.c:3753: warning: initialization discards qualifiers from pointer target type
cpuid.c:3754: warning: initialization discards qualifiers from pointer target type
cpuid.c:3755: warning: initialization discards qualifiers from pointer target type
cpuid.c:3756: warning: initialization discards qualifiers from pointer target type
cpuid.c:3757: warning: initialization discards qualifiers from pointer target type
gmake: \*\*\* [cpuid] Error 1

Luckily, the changes to get this program to work on Solaris were simple (Thanks Sherry!). All that we needed to do was remove the references to /dev/cpu/\* as that is a Linux-ism that does not exist on Solaris. Here is the complete diff for those wanting to try this at home:

blackhole$ diff linux-cpuid.c cpuid.c
> #if 0
> #endif
> #if 0
> #endif
> #if 0
> #endif

Clearly, if you wanted the program to work on either OS, you could just substitute the #if 0 strings for something like #if !defined(SOLARIS) and then just define SOLARIS in the CFLAGS parameter when compiling on Solaris. But I digress... With this simple change implemented, you can now compile the cpuid program on Solaris:

blackhole$ gmake
cc -g -Wall -Wshadow -Wcast-align -Wredundant-decls -Wbad-function-cast -Wcast-qual -Wwrite-strings -Waggregate-return 
-Wstrict-prototypes -Wmissing-prototypes -D_FILE_OFFSET_BITS=64 -DVERSION=20070801 -o cpuid cpuid.c
cpuid.c: In function `main':
cpuid.c:3757: warning: initialization discards qualifiers from pointer target type
cpuid.c:3758: warning: initialization discards qualifiers from pointer target type
cpuid.c:3759: warning: initialization discards qualifiers from pointer target type
cpuid.c:3760: warning: initialization discards qualifiers from pointer target type
cpuid.c:3761: warning: initialization discards qualifiers from pointer target type
cpuid.c:3762: warning: initialization discards qualifiers from pointer target type
cpuid.c:3763: warning: initialization discards qualifiers from pointer target type
gzip < >

These warnings can be safely ignored. With the program now compiled, let's give it a try and see what it can tell us about the NX bit:

blackhole$ ./cpuid | grep exec
      execution disable                      = false

Interesting. This system does not have the NX capability likely because I am running (Nevada in this case) in a Parallels VM which is 32-bit (reference Joe's note above). Let's give this a better test subject by trying it on a Sun X2100. This command is run from the global zone of a system running Solaris 10 11/06:

$ ./cpuid | grep exec
      no-execute page protection            = true

Careful observation will also show the AMD and Intel naming differences that I had talked about previously with respect to XD and NX.

Well, I think that I have talked about this subject to death. I hope that you found it interesting and perhaps a little educational. As always, I love to get your feedback! Before signing off, once again I would like to thank Sherry Moore and Joe Bonasera for sharing their knowledge and experience with me (and thereby with you)!

Take care,


Technorati Tag:

Monday Jul 30, 2007

Interesting File Discovery Tool version 0.5

As promised, I have uploaded version 0.5 of the Interesting File Discovery Tool (or ifd for short). This update includes fixes and enhancements that were contributed by Perley and Joe Moore. Thank you both for your contributions!

The biggest change in this version is the introduction of the -D parameter which enables you to change the program used to calculate the file digests (or fingerprints):

# ./ -h

   ./ - Interesting File Discovery Tool

   ifd -[ugnw] [-ds] [-q] [-D cmd] { -c | -l | [Solaris Product Directory] }

      -c     Collect information from /var/sadm/install/contents
      -d     Calculate MD5 digest for each file (Solaris 10 only)
      -D     Command used to calculate file fingerprint
      -g     Print information on files with the set-gid bit set
      -h     Display this message
      -l     Collect information from /var/sadm/pkg
      -n     Print information on WW directories without sticky bit set
      -q     Quiet mode.  Do not print headers.
      -s     Validate ELF file signature for each file (Solaris 10 only)
      -u     Print information on files with the set-uid bit set
      -w     Print information on world writable files and directories
      -?     Display this message

This can be useful in cases where you are running the tool on earlier releases of Solaris that do not have the integrated digest command or in cases where you want to use a different algorithm. For example, with this change, you could tell ifd to create SHA-512 fingerprints:

# ./ -c -D "/usr/bin/digest -a sha512" -d -u

Set-UID Programs

SUNWaccu 4755 root adm 29478dd7ebde1555eaef0987789094cc778794ee73ddcfb0a67c44004f93652f599dd7276342f8113cc4e58f877e883b4687c4ca0f30f0585dd725ddaffeb0b7 /usr/lib/acct/accton
SUNWbip 4555 root bin 95c814f7ff9606e0dc8818b51dacf74e92e5b3af329d66dc6fc8343c20ae741c1cea758568a318713ce6aacb35d1605bd6ee0911cdd2457aa85ceed363d17326 /usr/sbin/ping
SUNWbnuu 4511 root uucp 540f94a7054233498f1925aceef3c69b76300141ef38acc920ae005287db5546a03daef37c19b98149e11a26c7b4da137788e45cf642a3449345f635d8dbf762 /usr/bin/ct
SUNWbnuu 4511 uucp uucp 1754a7f7aaea60f4a1d1ca1915af30bc0157333061c096088bd3b719d008167f603380fae5b417a237cc9fe8c4cdcf524b22c61a471d0a06df5188cabedb475c /usr/bin/uuglist

Pretty neat. Thanks again to Perley and Joe for their feedback and support! To everyone - give this new version a shot and let me know what you think.

Take care,


Technorati Tag:

Wednesday Jul 25, 2007

Solaris Non-Executable Stack Continued

Previously, we covered some of the history and basics of Solaris non-executable stacks and how they can be enabled globally on both SPARC and x86/x64 systems. In this article, we extend that foundation by talking about how developers can configure their own programs to have non-executable stacks, regardless of the value of the global system setting, noexec_user_stack.

This little bit of magic is accomplished through the use of a linker map file. In the case of non-executable stacks, the linker map file in question is /usr/lib/ld/map.noexstk. Simply specifying this map file during a compilation or link will cause the resulting program to have a non-executable stack. Looking at the comments in this file, we see how this is accomplished:

#ident  "@(#)mapfile_noexstk    1.3     01/07/13 SMI"
# Copyright (c) 2001 by Sun Microsystems, Inc.
# All rights reserved.
# Linker mapfile to create a non-executable stack definition within an
# executable.
# The linker does not use this file automatically, so one must use the -M 
# option to cc or ld:
#       cc -M /usr/lib/ld/map.noexstk myprogram.c
stack = STACK ?RW;

If this sounds pretty straightforward and easy to use, that is because it is! Let's go ahead and give it a try! Before we begin, I would like to thank Scott Rotondo for sharing with me the following sample program. This program will attempt to execute code on the stack. Our test system is configured with noexec_user_stack=0 and we will compile our test program both with and without using the map file so that they can be compared with one another.

First, here is our test program:


int x = 0;


typedef void (\*funcptr)(void);

main(int argc, char \*\*argv)
        funcptr f;
        char code[100];

        /\* Copy the incr() function to the stack. \*/
        memcpy(code, (void \*)incr, sizeof(code));
        f = (funcptr)code;

         \* Increment x twice, once by calling incr() and
         \* once by running the copy on the stack.
        printf("x = %d\\n", x);
        printf("x = %d\\n", x);
        printf("x = %d\\n", x);
        return (0);

Now, let's compile the program (with and without the map.noexstk map file):

$ gcc -O -o incr incr.c
$ gcc -O -o incr-nx -Wl,-M,/usr/lib/ld/map.noexstk incr.c

(Thank you to Luke for pointing out a cleaner way to pass the linker map file using gcc!)

Note that if you were using the Sun C compiler, you could have used the following commands:

$ cc -O -o incr incr.c
$ cc -O -o incr-nx -M /usr/lib/ld/map.noexstk incr.c

So, how do we know that the program, incr-nx, has a non-executable stack? One of the easiest ways is to use the elfdump(1) command telling it to look for the program header type, PT_SUNWSTACK. The absence of this program header means that the program is effectively in a default configuration where (depending on the platform) the stack segment could be readable, writable as well as executable. If a PT_SUNWSTACK program header is found then the default is not being used, and we need only to look at the p_flags parameter to see what permissions are being assigned to the stack segment.

$ elfdump -p -N PT_SUNWSTACK incr
$ elfdump -p -N PT_SUNWSTACK incr-nx

Program Header[5]:
    p_vaddr:      0           p_flags:    [ PF_W PF_R ]
    p_paddr:      0           p_type:     [ PT_SUNWSTACK ]
    p_filesz:     0           p_memsz:    0
    p_offset:     0           p_align:    0

As you can see from the output of the two commands above, the incr program's stack segment is configured in the default manner and will therefore have an executable stack (unless of course the global system parameter noexec_user_stack is set to 1). On the other hand, the incr-nx program does have a PT_SUNWSTACK program header. Looking at the p_flags parameter, we see that this program's stack segment will have only the read (PF_R) and write (PF_W) flags enabled.

The next obvious question is whether these programs will behave differently. Certainly, we would expect them to given that they are configured to execute code on the stack yet such an operation is only permitted in one of the two programs. Let's take a closer look:

$ ./incr
x = 0
x = 1
x = 2
$ ./incr-nx
x = 0
x = 1
Segmentation Fault (core dumped)

If we had enabled logging of attempts to execute code on the stack using the noexec_user_stack_log parameter, we would have also seen a syslog message similar to:

$ tail -1 /var/adm/debug
Jul 25 22:11:36 quasar genunix: [ID 533030 kern.notice] NOTICE: incr-nx[12553] attempt to execute code on stack by uid 101

Pretty cool, eh? So with the simple addition of the linker map file, we can now deploy programs and services that will have non-executable stack segments (out of the box)! In fact, a large portion of the ON (operating system and networking) consolidation in the Solaris OS is already configured this way! In fact, even the Sun-contributed Firefox (that is also included in Solaris 10 and OpenSolaris) uses this mechanism to enable non-executable stacks. Yes, even OpenOffice/StarOffice and Xorg are in on the action! So, what are you waiting for? Give it a try today!

I hope you enjoyed this brief overview into Solaris non-executable stacks. As always, I would love to get your feedback and ideas. You can read more on this topic here and here.

Take care,


Technorati Tag:

Solaris Non-Executable Stack Overview

The ability to configure a Solaris system to run with non-executable stacks is not overly new. That functionality was originally introduced into the Solaris 2.6 operating system with the noexec_user_stack kernel parameter. Looking at the source code, this is how this parameter was documented (in usr/src/uts/common/vm/seg_vn.c):

 207 /\* 
 208  \* Patching this variable to non-zero allows the system to run with 
 209  \* stacks marked as "not executable".  It's a bit of a kludge, but is 
 210  \* provided as a tweakable for platforms that export those ABIs 
 211  \* (e.g. sparc V8) that have executable stacks enabled by default. 
 212  \* There are also some restrictions for platforms that don't actually 
 213  \* implement 'noexec' protections. 
 214  \* 
 215  \* Once enabled, the system is (therefore) unable to provide a fully 
 216  \* ABI-compliant execution environment, though practically speaking, 
 217  \* most everything works.  The exceptions are generally some interpreters 
 218  \* and debuggers that create executable code on the stack and jump 
 219  \* into it (without explicitly mprotecting the address range to include 
 220  \* PROT_EXEC). 
 221  \* 
 222  \* One important class of applications that are disabled are those 
 223  \* that have been transformed into malicious agents using one of the 
 224  \* numerous "buffer overflow" attacks.  See 4007890. 
 225  \*/ 

While non-executable stacks provide are a very useful technique for thwarting certain kinds of buffer overflow attacks, it should be noted that there exist other attack methods that do not rely on executable stacks. One such method was discussed back in 1999 on Bugtraq, but even in this case the author noted that there was inherent value in non-executable stacks (if only as an additional defense in depth layer):

Hopefully, these exploits demonstrate that it is important to make sure that programs that run at an elevated privilege are free of buffer overflow bugs. The stack protection will certainly help protect you from the majority of intruders, but moderately competent intruders will probably be able to bypass it.

Just as with minimization, hardening, and the deployment of services with reduced privilege, non-executable stacks are just another layer or tool to be used as part of a more comprehensive security architecture. But anyway, back to our story...

As with other kernel parameters, the non-executable stack state can be be adjusted (enabled or disabled) using the /etc/system file. For example, the following statement added to /etc/system would enable this feature:

set noexec_user_stack=1

As noted in the inline documentation above, experience has shown that "most everything works". In fact, the recommendation to enable this feature has been in Sun BluePrints since 1999 and similarly in the Solaris Security Toolkit since its inception. Looking even further, you find this common recommendation across the industry.

As a companion to this parameter, the noexec_user_stack_log parameter could be used to enable logging when this feature (if enabled) detected an attempt to run code from the stack. By default, this parameter is enabled if the noexec_user_stack parameter is enabled so no further action is required unless of course you want to prevent such logging. That has not stopped authors of tools and articles from recommending to enable it anyway using the command:

set noexec_user_stack_log=1

When this parameter is enabled and there is an attempt to execute code on the stack, a message such as the following will be generated and delivered via syslog to kern.notice:

Jul 25 14:48:02 quasar genunix: [ID 533030 kern.notice] NOTICE: myprog[12289] attempt to execute code on stack by uid 101

In this way, a system administrator can detect such attempts and take appropriate action.

Back in the days of Solaris 2.6, this parameter really only applied to the SPARC platform. Years passed and this feature continued to be available in Solaris 7, Solaris 8 and so on. As good fortune would have it, Intel and AMD got on board with the idea and the NX Bit was born. Technically speaking, Intel refers to its implementation as the XD Bit (for Execute Disable) while AMD has used the term NX (for No Execute), but for the purposes of Sun's implementation and this article, we will consistently use the term "NX" to refer to this functionality.

To find out if your system supports the NX bit, you can check in with the dmesg(1M) command:

$ dmesg | grep features
Jun 28 11:00:05 sec1 unix: [ID 126719] features: 1176fdf<cpuid,cmp,sse3,nx,asysc,sse2,sse,pat,cx8,pae,mca,mmx,cmov,pge,mtrr,msr,tsc,lgpg>

Similarly, if you have syslog configured to log messages, you can also get the information from your system log files:

$ grep "features:" /var/adm/debug
Jul 19 16:43:06 quasar unix: [ID 126719] features: 1076fff<cpuid,sse3,nx,asysc,sse2,sse,pat,cx8,pae,mca,mmx,cmov,de,pge,mtrr,msr,tsc,lgpg>

The first example was taken from a SunFire X2100 system whereas the second example was taken from an Ultra 20. The same commands should be able to be used on other x86/x64 systems in order to determine if this CPU feature is available.

On the SPARC platform, the non-executable stack functionality is available but disabled by default (for SPARC V8) in order to support a fully ABI-compliant execution environment. For 64-bit SPARC platforms, however, the SPARC V9 ABI specifies a non-executable stack by default. Note that 32-bit applications running on a 64-bit kernel do not automatically get this protection by default and would rely on the noexec_user_stack parameter being set to 1 for example.

On NX-capable x86/x64 platforms, Solaris OS uses the NX bit by default whenever PROT_EXEC is not specified. Stack segments, however, use PROT_EXEC by default, so the NX functionality must be explicitly enabled on these platforms to provide stack protection. As noted above, this can be globally configured using the noexec_user_stack parameter just as with SPARC-based platforms.

From the product documentation, it should be noted that a system administrator can disable all use of the NX bit (non-SPARC platforms) by using the eeprom(1M) command to set enforce-prot-exec to off. This variable is provided as a transition workaround for any system with legacy applications that are missing PROT_EXEC.

In this article, we have taken a brief look at the history of non-executable stacks in Solaris dating back to the original integration of this functionality in Solaris 2.6 all the way to the present. In the next article, we will talk a little bit about how this functionality can be enabled on a per-file basis in the Solaris 10 OS.

Take care,


Technorati Tag:




« April 2014