Erik Braun: > Do you think, that this mutex_lock() in the OpenAFS source is necessary > on a readonly replica? (I am inexperienced in system programming.)
As far as I know, there is no filesystem which acquires i_mutex in simple open(2). But it never means openafs is wrong. I just simply don't understand what openafs is. > Thank you for analysing the problem! Now we know, *WHY* aufs on the te top > of OpenAFS can lock-up a linux system. Of course, I prefer to use the > system as it was possible until these changes. Can you suggest a > procedure, how I can achieve this? If I were you, I'll try these steps. - revert the commit b0ed5a7f in openafs. - rebuild openafs - test openafs alone - test openafs with aufs - report the problem to the openafs maintainers and suggest reverting But this is from my point of view. The openafs maintainers may have different opinions. Probably the point will be why openafs calls osi_FlushPages() in afs_open(). Generally, it is unusual behaviour (at least for me). J. R. Okajima ------------------------------------------------------------------------------
