Re: seq_file and exporting dynamically allocated data

[viro@parcelfarce.linux.theplanet.co.uk]

On Sat, Nov 15, 2003 at 07:49:43PM +0000, Tigran Aivazian wrote:
> Dear Harald,
> 
> The thing to be aware of is that seq API is limited to 1 page of data per
> read(2) call. Some people loudly proclaim "seq API is unlimited, unlike
> the old /proc formalism which was limited to 1 page" but they are 
> quiet about 1-page limitation for a single read(2) (due to hardcoded
> kmalloc(size) in seq_file.c). Why is this important? Because if your

What?  It will happily return more than 1 page if you get an entry longer
than that.

What hardcoded kmalloc(size)?

> ->start/stop routines take/drop some spinlocks then you have to know your
> position and re-verify it on the next read(2) if your data is more than 1
> page and thus could not be read atomically (i.e. while holding the
> spinlocks).

If you mean that read(2) can decide to return less than full user buffer even
though more data is available - tough, that's something userland *must* be
able to deal with.

Especially on syntetic and potentially large files - the only alternative
would be immediate DoS on OOM.  Regardless of the implementation.