Closed fengyuleidian0615 closed 3 years ago
Yes, you are right.
On Tue, Jul 3, 2018 at 10:18 PM Fan Du notifications@github.com wrote:
Hi @Andiry https://github.com/Andiry
MAP_SYNC feature merged in v4.15 vanilla kernel tried to call fdata_sync to flush dirty metadata and data down to media, this brings more robust filesystem consistence on top of journal. As I read through NOVA FS implementation, both metadata and data are kept consistent, so by my limited understanding, equivalence of MAP_SYNC in NOVA FS is not needed.
Please correct me if I'm wrong somewhere.
thanks!
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NVSL/linux-nova/issues/71, or mute the thread https://github.com/notifications/unsubscribe-auth/ABQcEAdnd_7quYWEQNyiQ7i8OBj5vCiWks5uDFAogaJpZM4VB9_- .
According to DAX support part in README, msync() in user space is needed. Why?
You don't have to. When a file page is mmaped to user space, NOVA releases control of it and it is the user's responsibility to ensure consistency. User space can use any method: coarse-grained msync() or fine-grained cache line flushing. It's about data consistency, not metadata.
The issue itself is about the MAP_SYNC flag, which guarantees metadata consistency. NOVA always guarantee metadata consistency.
Thanks, Andiry
On Sat, Sep 5, 2020 at 11:57 PM wzl notifications@github.com wrote:
According to DAX support part in README, msync() in user space is needed. Why?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.
Hi @Andiry
MAP_SYNC feature merged in v4.15 vanilla kernel tried to call fdata_sync to flush dirty metadata and data down to media, this brings more robust filesystem consistence on top of journal. As I read through NOVA FS implementation, both metadata and data are kept consistent, so by my limited understanding, equivalence of MAP_SYNC in NOVA FS is not needed.
Please correct me if I'm wrong somewhere.
thanks!