This report assumes that NOVA's protection features are enabled (nova.metadata_csum=1 nova.data_csum=1 nova.data_parity=1).
During unlink, rmdir and rename operations, there are crash states that lead to readdir operations failing.
nova_verify_entry_csum outputs the following error messages: "both entry and its replica fail checksum verification" and "unable to repair entry errors".
The call stacks indirectly called by these system calls appear to share the following: __copy_user_nocache <- nova_update_alter_entry <- (...) <- nova_inplace_update_dentry
It appears plausible that they have the same root cause as #121 as they share some of the call stack and is presumably also related to not persisting the checksum (I have not yet verified that completely).
I will update once I find out more about the cause.
This report assumes that NOVA's protection features are enabled (
nova.metadata_csum=1 nova.data_csum=1 nova.data_parity=1
).During
unlink
,rmdir
andrename
operations, there are crash states that lead toreaddir
operations failing.nova_verify_entry_csum
outputs the following error messages: "both entry and its replica fail checksum verification" and "unable to repair entry errors".The call stacks indirectly called by these system calls appear to share the following:
__copy_user_nocache
<-nova_update_alter_entry
<- (...) <-nova_inplace_update_dentry
It appears plausible that they have the same root cause as #121 as they share some of the call stack and is presumably also related to not persisting the checksum (I have not yet verified that completely). I will update once I find out more about the cause.