Please enable JavaScript.
Coggle requires JavaScript to display documents.
ublk-qcow2 write hang - Coggle Diagram
ublk-qcow2 write hang
observation
reproducer
build kernel on XFS/ublk-qcow2
logging
sample 2
write_at offset 2619ea000 len 430080
similar with sample 1
new allocated clusters are discarded
but not wroten
do_write_data_file() isn't done
other write_at doesn't show do_write()
sample 1
one 512K write isn't done
8 clusters are allocated
offset: 16b318000
32k aligned
write_at offset 16b318000 len 524288
ublk_qcow2-5.log
offset: 1eb501000
all do_writes() are done
write_at() not completed
hang on self.clear_new_cluster()?
what is the problem?
deadlock?
executor?
how to investigate
new IO can't be issued
write_at() is dumped
mapping can't be retrieved?
what is the bad code path holding some lock?
flushing meta?
IO preempts flushing meta
race?