[OpenAFS-devel] to fsync() or not to fsync()

[Ken Hornstein]

>> So, when I look at StoreData_RXStyle, I am not able to convince myself
>> that the actual file data is fsync()'d ... unless fsync() is actually
>> called on the client (and from what I see, fsync() IS passed from the
>> client to the server).  I can believe there is some other fsync()s in
>> there, certainly, but I don't think it's doing the bulk data (but hey,
>> I've been wrong before when it comes to AFS program flow).
>
>Huh?  There's no RPC for that.  And no, I don't think file data is synced, 
>but metadata is, which is consistent with the behavior of many other 
>filesystems.

There's no specific RPC for an fsync operation, true.  Buuuttt ...

Fsync operations at the vnode level end up calling afs_fsync().

afs_fsync() calls afs_StoreAll_Segments() with the third argument of AFS_SYNC.

afs_StoreAllSegments() sets the flag AFS_FSYNC in Instatus.Mask when the
third argument is AFS_SYNC.

InStatus gets transmitted to the fileserver as part of the StoreData RPC.

If AFS_FSYNC is seen by the StoreData RPC function on the fileserver,
FDH_SYNC is called after the data is written to disk.

Assuming IH_DELAY_SYNC is not set on the IHandle (Which I believe it is
not in this case), FDH_SYNC() invokes OS_SYNC(), which is defined as fsync().

Given all that, I am reasonably confident that when a user calls fsync()
on an AFS file, it is fsync()'d on the fileserver.  At least there seems
to be a large pile of code that tries to make that happen.

--Ken