[cdi-dev] [JBoss JIRA] (CDI-641) Invalid manifest section in cdi-api JAR

Björn Kautler (JIRA) issues at jboss.org
Thu Oct 27 09:49:00 EDT 2016


    [ https://issues.jboss.org/browse/CDI-641?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13313032#comment-13313032 ] 

Björn Kautler commented on CDI-641:
-----------------------------------

Yes, I think so. If that PR gets applied and all use the new parent (for me it is cdi-api and weld-se), then it should be fine.

> Invalid manifest section in cdi-api JAR
> ---------------------------------------
>
>                 Key: CDI-641
>                 URL: https://issues.jboss.org/browse/CDI-641
>             Project: CDI Specification Issues
>          Issue Type: Bug
>          Components: Packaging and Deployment
>    Affects Versions: 1.0, 2.0-EDR2
>            Reporter: Björn Kautler
>
> In the {{MANIFEST.MF}} of your {{cdi-api}} JAR you have a section {{Build-Information}}.
> This violates that JAR specification.
> A section in the manifest always refers to an entry in the JAR.
> If you have sections in the manifest that do not refer to an entry in the JAR, it is assumed that the JAR was tampered with as there are entries missing that are referenced in the manifest.
> Please either remove this section and include the entries in that secion in the main section which is according to the specification or include a file called {{Build-Information}} at the root of your JAR file.
> ----
> In my concrete use-case this happened with other JARs with invalid manifest entries:
> - I have signed those JARs and included them in a WebStart application
> - I started the application with 8u102 32-bit {{javaws}}
> - The JARs were downloaded and their entries signatures verified
> - As there were entries in the manifest that are not present in the JAR, the file was not seen as completely signed with one signature, but Java remembered for each entry with which signature it was signed
> This already is not too nice as it slows down the application as now for each class that gets loaded the signature has to be retrieved from a map and a list instead of having just one signature for all entries. But it gets much worse:
> - The acutal application was to be executed with 8u102 64-bit, so the 32-bit one wrote its session information out into files, including the information about verified JARs and also their entries if needed, and starts the 64-bit JVM
> - The 64-bit JVM loads this session information and thus does not have to do the time-consuming verification of the JARs all over again
> - Unfortunately since 8u91 or so there is a bug in this session reading and writing algorithm, so that some of the entry names get crippled with additional characters in-between
> - If now a class should be loaded that has such a crippled entry in the JAR-entry-to-signature map, the entry is not found and the class is considered as not signed which will block the application from further execution
> Of course this second part is a bug in Java, but it would work flawlessly if the JARs would not have invalid sections.



--
This message was sent by Atlassian JIRA
(v7.2.2#72004)



More information about the cdi-dev mailing list