Miklos, I ran into an incident of very large index dir which took considerable amount of time to mount the indexed overlay (~30 minutes). The index dir had millions of entries and I do not think that the use case that caused this is typical. The following two patches are based on perf top analysis of this incident. I do not have access to the data set that caused the very long mount time, but I tested the desired CPU usage improvements on a smaller scale data set. It is hard to say if this extreme case of very large index dir is common enough to be worth any attention, so I did not tag the fixes for stable and I don't think it is urgent to apply them. Unfortunattely, the investigation of the incident was not timed optimally w.r.t. to the current merge window. Nevertheless, the changes are quite trivial, so you may want to consider them either for -rc or for next release. Thanks, Amir. Amir Goldstein (2): ovl: do not reconnect upper index records in ovl_indexdir_cleanup() ovl: use plain list filler in indexdir and workdir cleanup fs/overlayfs/export.c | 4 ++-- fs/overlayfs/namei.c | 7 ++++--- fs/overlayfs/overlayfs.h | 3 ++- fs/overlayfs/readdir.c | 12 ++---------- 4 files changed, 10 insertions(+), 16 deletions(-) -- 2.25.1
