Summary

Strange bugs in openssl egg (or core?)

Metadata

• Id: 0aac21add08697c6f9a023ff6d17c3757787be94
• Trac id: 1058
• Type: defect
• Reporter: sjamaan
• Owner: chust
• Cc:
• Status: closed
• Component: unknown
• Estimated difficulty:
• Resolution: fixed
• Priority: major
• Milestone: 4.9.0
• Version: 4.8.x
• Changetime: 2013-10-17 22:27:42 UTC
• Created: 2013-10-11 20:40:14 UTC
• Keywords: crash, panic, openssl

Description

We've seen that henrietta-cache occasionally, but consistently shows strange behaviour. It appears this happens only for HTTPS URIs, so I've done some investigating and can reproduce a situation which could be related:

 #!scm
 (use http-client extras)
 
 (let lp ()
   (print (with-input-from-request (car (command-line-arguments)) #f read-file))
   (lp))

Just call the program like this:

 csi -s foo.scm https://bitbucket.org/sjamaan/henrietta-cache/raw/tip/henrietta-cache.release-info

Then press ^C after the program has seemed to make a connection.

You'll generally get a panic in the scheduler that says "fdset-set: invalid i/o direction: ssl-mutex (fd = 15)" (with today's master; before you just got the message without more info). It appears the ssl-mutex somehow ends up in the fd-set upon which the thread will wait for I/O. This is most likely due to some corruption elsewhere, but it's unclear where exactly.

This may be related to #989 (ie, it could be a signal handling issue). Mario has observed] occasional segfaults instead of panics.

Changes and comments

[2013-10-11 23:43:08 UTC] sjamaan wrote:

The bug is in CHICKEN core. Patch forthcoming

[2013-10-17 22:27:42 UTC] sjamaan changed status from new to closed

[2013-10-17 22:27:42 UTC] sjamaan set resolution to fixed

[2013-10-17 22:27:42 UTC] sjamaan wrote:

Fixed by changeset 4c14161ba09eb1f875c40dfa0d722382fad98f36