Re: [GSoC 2026] - Implementation of sendfile(2) syscall

[Henrique Brito <henrique.britomelo667%gmail.com@localhost>]

Forgot to link the pull request 😅:
https://github.com/NetBSD/src/pull/73


On Tue, Feb 24, 2026 at 6:43 PM Henrique Brito
<henrique.britomelo667%gmail.com@localhost> wrote:
>
> Hello Tech Kern !
>
> My name is Henrique Brito and, as I've stated in my previous emails, I am interested in contributing to compat linux emulation layer through GSoC 2026. (https://wiki.netbsd.org/projects/project/linux_timer/)
> As a starting point, I've decided to begin my journey with some practical experience, so I tried my best to write an emulation for the Linux sendfile (2) syscall.
> I would really appreciate any feedback that you may have, the application period is approaching, and your opinion would be of immense value to me.
>
> My implementation follows the basics idea of the sendfile: Transfer data between two file descriptors without copying anything to user space, boosting Performance.
>
> Currently, the emulation is able to pass all LTP tests for the sendfile syscall, providing Linux Binaries the expected ABI.
>
> As far as I have tested, the sendfile emulation is as fast as the read + write syscalls, unlike the Linux version, which is significantly faster than the two. I have spoken with people in Libera Chat about this, and they pointed that this may be due the optimization of NetBSD's read/write, instead of a problem with the sendfile.
> Although there aren't any real performance gains, I think that it still has its uses, as it broadens the compat linux coverage of syscalls and, at the same time, don't impact negatively Binaries that can use the read + write when sendfile is not present.
>
> Please, let me know your thoughts about it.
>
> Best regards, Henrique
>
> --
> Henrique de Brito
> Computer Science Student
> Federal University of São Carlos - UFSCar
>


-- 
Henrique de Brito
Computer Science Student
Federal University of São Carlos - UFSCar