Re: [PATCH 2/2] treewide: lose F_SETPIPE_SZ page size assumptions

[Alyssa Ross <hi@alyssa.is>]

[-- Attachment #1: Type: text/plain, Size: 1034 bytes --]

Eric Wong <e@80x24.org> writes:

> Eric Wong <e@80x24.org> wrote:
>> Alyssa Ross <hi@alyssa.is> wrote:
>> > fcntl(2) is documented as rounding up sizes to an integer multiple of
>> > the page size.  When a pipe buffer as small as possible is desired,
>> > it's cleaner to pass 0 and let the kernel set the pipe buffer to
>> > exactly one page than it is to assume that pages are 4096 bytes.
>> 
>> OK.  I suppose the Linux kernel has enough testers that it's
>> chances of it mishandling someday 0 would be caught before any
>> release.
>
> Well, that was only thinking about potential future bugs...
>
> Looking backwards, older Linux appeared to handle 0 improperly before
> d3f14c485867 (pipe: avoid round_pipe_size() nr_pages overflow on 32-bit, 2017-11-17)
> https://yhbt.net/lore/lkml/d3f14c485867cfb2e0c48aa88c41d0ef4bf5209c/s/
>
> I haven't tried an ancient kernel to verify and my arithmetic is awful
> (d3f14c485867 also got backported to older distro kernels).

Okay, let's just leave it as-is then.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 227 bytes --]