> > an alternative is not to care much about pages dirtied via mmap
> > as it isn't required by any standards afaik.
> 
> the susv3 description of fsync() says:
> 
>   The fsync() function shall request that all data for the open file
>   descriptor named by fildes is to be transferred to the storage device
>   associated with the file described by fildes. The nature of the transfer
>   is implementation-defined. The fsync() function shall not return until
>   the system has completed that action or until an error is detected.
> 
> 
> so the question would be whether data put into a file via mappings is
> part of "all data" for that file.  it doesn't seem reasonable to say
> that it isn't.  so I'd much prefer that fsync() operate on pages
> dirtied via mappings as well as those dirtied via write().  even if
> the standard did not require it, it seems like a better implementation
> anyway.

i've heard an argument, something like this:
	it's "all data for the open file descriptor",
	not "all data for the file associated with the descriptor".

anyway, both approaches have their own pros and cons.
i have no strong affinity.

YAMAMOTO Takashi