|
LU-118 is tracking b1_8 and master Lustre fixes for the cause of the kernel BUG_ON(inode->i_data.nrpages) in clear_inode().
There is concern that a kernel race needs to be fixed there as well to completely tie off the issue. This ticket should track the kernel solution. Ideally, we'd get this identified so that the upstream kernel maintainers can adopt a fix that they like and Lustre will not need to carry a kernel patch around for this bug.
References:
> Lustre Bug 24303
> http://jira.whamcloud.com/browse/LU-118?focusedCommentId=10993&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#action_10993
> Jinshan also added, "Then let's take an effort to make up a test case so as to reproduce it in linux kernel. I think there would be a code path in the linux kernel is that when vmscan is stealing a page(shrink_page_list->remove_mapping->__remove_from_page_cache), and meanwhile the inode is being forgotten(iput_final -> ... ->generic_forget_inode).
|