David S. Miller wrote:
> > So when i try to free the page the page does not get freed resulting
> > in a kind of memory leak.
>
> It gets freed later when the TCP send has completed at the
> device, and the other end has sent an ACK back to acknowledge
> the data. Meanwhile the packet sits referenced in the TCP
> retransmit queue and cannot be freed.
That reminds me. Is there a good (efficient) way to find out when
it's freed from userspace?
The reason I want this is:
1. Call sendfile() on some portion of a data file.
2. Later, I need to update portions of the data file, without affecting
data that I already sent to the TCP stream.
Step 2 cannot be done until the pages queued by sendfile() have been
freed, because until then, data written in step 2 might be sent on the
wire. Of course I could call send() but that would miss the
memory-saving and time-saving effects.
Alternatively, treating the file like a database with a "wandering
log" like reiser4, I can choose to write elsewhere in the file. But
how do I know when it's safe to reuse the pages which I previously
queued with sendfile()?
I'm currently using ioctl(SIOCOUTQ), but that isn't great as I need to
poll it repeatedly to check for freeing progress.
Thanks,
-- Jamie
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
