[RELEASE] Signing Artifacts [Was Re: Symantec WS.Reputation.1 Errors: What we can do]

[Dave Fisher <da...@comcast.net>]

Hi Dennis,

I am changing the subject to give this discussion the visibility it should have.

I think it would be highly desirable to digitally sign AOO 3.4 artifacts.

On Mar 2, 2012, at 1:15 PM, Dennis E. Hamilton wrote:

> Out of curiosity I just did another download of the OOo-dev 3.4 Windows "MSI" r1293550 to see if the popularity contest had been won yet.  
> 
> Not yet.
> 
>  The Internet Explorer 9 download warning is "OOo-Dev-OOO340m1_Win_x86_install_en-US.exe is not commonly downloaded and could harm your computer."  The file is already downloaded at that point, however.  The options are Delete, Actions, and View downloads (opening a separate tool that shows downloads and status).  The Actions option includes "Don't run this program (recommended)", "Delete", and "Run Anyway."  
> 
>  A "Run Anyway" or a later execution of the downloaded .exe will provoke an User Account Control message (on default configurations, even for administrator accounts) that warns that the file is from an unknown source (that is, the .EXE is not signed) and that it was downloaded from the Internet. 
> 
> I also did a custom scan of the single download file using Microsoft Security Essentials.  The scan (which is programmed to dig into these files) identified 51916 individual items and no threats.
> 
> All of this is tolerable and arguably appropriate for developer snapshots.  Users who use these builds need to rely on their own judgment about the trustworthiness of the origin and the content of those files.  
> 
> Note that it is the .exe that needs to be signed.  This should not be confused with a .msi file, although I assume those can be signed also.  Apache OpenOffice does not use .msi as the packaged binary that is downloaded.  (It appears that LibreOffice has changed that.)  It strikes me that using external digest values (md5 and sh1 digests) on the download requires a super-user skill set and should not be the only thing relied upon for project binary releases.

(Side note about Releases) All PPMC members who vote for a release will be expected to understand how to use md5/sha1 digest values and to have checked the signatures of ALL artifacts. It is a release blocker to have missing or incorrect signatures.

Soon, all will understand that this is NOT something to ask Grandma or Joe Bob to do.

So, signatures from a trusted entity like the ASF will go along way towards gaining trust. The gist of the signing service discussion was when to apply digital signatures to Development, Test and Production Artifacts. It was mentioned that Dev and Test Artifacts could have expiring certificates.

So, I will explore this and if viable return with documentation for the Release Manager.

Regards,
Dave

> 
> -----Original Message-----
> From: Dennis E. Hamilton [mailto:dennis.hamilton@acm.org] 
> Sent: Friday, March 02, 2012 11:19
> To: ooo-dev@incubator.apache.org
> Subject: RE: Symantec WS.Reputation.1 Errors: What we can do
> 
> The web-based downloader in Internet Explorer 9 also warns about the .exe files (not the tar.gz or Zip ones).  The message is clearly a no-reputation-yet warning.  
> 
> This is an on-line check.  When the file is known to be regularly downloaded, the report will change automatically.  
> 
> I have seen no AV warnings about the downloaded files themselves, although there is a standard OS warning on use of such files when they were downloaded from the internet and/or are not signed.  (In Windows 8 Consumer Preview, it is necessary to click "details" to see that there is a "Run anyhow" selection.)  
> 
> I saw no AV warnings after the installation on any systems having Microsoft Malware detection and regularly-updated Windows Security Essentials.
> 
> - Dennis
> 
> -----Original Message-----
> From: Rob Weir [mailto:robweir@apache.org] 
> Sent: Friday, March 02, 2012 07:00
> To: ooo-dev@incubator.apache.org
> Subject: Symantec WS.Reputation.1 Errors: What we can do
> 
> Several testers have mentioned this anti-virus error when installing
> the AOO 3.4 dev snapshot build.   This is not a virus.
> "WS.Reputation" errors come from Symantec Antivirus based on their
> "reputation-based" threat assessments.  Essentially, they evaluate
> software that you are about to install according to a range of
> factors, including how new the file is, how many other people have
> installed it, whether the installer is digitally signed, etc.  It is
> not just one factor, but a proprietary mix of weighted factors.
> 
> We're probably getting penalized based on several of these factors.
> Note that with the final AOO 3.4 release we'll be in the same
> position, since that installer will also be new,etc.
> 
> A few things we should consider doing:
> 
> 1) Make sure the readme file and install instructions cover this case
> and explain what the user should do, e.g. "Run anyways"
> 
> 2) We can make a request to Symantec to "whitelist" our installer.
> This takes a couple of weeks for them to process.  And we can';t start
> this work in advance since they need the SHA-256 hash of our
> installer:
> 
> https://submit.symantec.com/whitelist/isv/
> 
> 3) We could digitally sign our Windows installers.   Apache already
> requires a detached signature.  But Symantec has no idea about these.
> We need traditional Windows exe code signing.  This will help us with
> Windows 8 as well.  So it is something we probably want to look into
> at some point.
> 
> My recommendation:
> 
> Plan on doing 1.  Do 2. as soon as we have a release.  Look into 3. for AOO 4.0.
> 
> Regards,
> 
> -Rob
>