linux-kernel - [RFC PATCH liburing v1 0/4] Export io

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 29 Aug 2022 10:07:35 +0700
From:   Ammar Faizi <ammarfaizi2@...weeb.org>
To:     Jens Axboe <axboe@...nel.dk>
Cc:     Ammar Faizi <ammarfaizi2@...weeb.org>,
        Caleb Sander <csander@...estorage.com>,
        Muhammad Rizki <kiizuha@...weeb.org>,
        Kanna Scarlet <knscarlet@...weeb.org>,
        io-uring Mailing List <io-uring@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        GNU/Weeb Mailing List <gwml@...r.gnuweeb.org>
Subject: [RFC PATCH liburing v1 0/4] Export io_uring syscall functions

From: Ammar Faizi <ammarfaizi2@...weeb.org>

Hi Jens,

The background story of this series comes from the recent conversation with
Caleb Sander at:

   https://github.com/axboe/liburing/pull/646#issuecomment-1229639532

What do you think of this series?

There are 4 patches in this series.

1) syscall: Add io_uring syscall functions

We have:

  man 2 io_uring_setup;
  man 2 io_uring_enter;
  man 2 io_uring_register;

Those entries say that io_uring syscall functions are declared in
`<linux/io_uring.h>`. But they don't actually exist and never existed.
This is causing confusion for people who read the manpage. Let's just
implement them in liburing so they exist.

This also allows the user to invoke io_uring syscalls directly instead
of using the full liburing provided setup.

2) Clarify "man 2" entry for io_uring.

io_uring_enter(), io_uring_register(), io_uring_setup() are not
declared in `<linux/io_uring.h>` and never were. Plus, these
functions don't intentionally set the `errno` variable. Reflect this
fact in the manpage.

Side note: On architectures other than x86, x86-64, and aarch64, those
functions _do_ set the `errno`, this is because the syscall is done via
libc as we don't yet have nolibc support for the mentioned archs. Users
should not rely on this behavior.

3) man: Alias `io_uring_enter2()` to `io_uring_enter()`.

We have a new function io_uring_enter2(), add the man page entry for it
by aliasing it to io_uring_enter(). This aliased man entry has already
explained it.

4) test/io_uring_{enter,setup,register}: Use the exported syscall functions.

These tests use the internal definition of __sys_io_uring* functions.
A previous commit exported new functions that do the same thing with
those __sys_io_uring* functions. Test the exported functions instead of
the internal functions.

No functional change is intended.

Signed-off-by: Ammar Faizi <ammarfaizi2@...weeb.org>
---

Ammar Faizi (4):
  syscall: Add io_uring syscall functions
  man: Clarify "man 2" entry for io_uring syscalls
  man: Alias `io_uring_enter2()` to `io_uring_enter()`
  test/io_uring_{enter,setup,register}: Use the exported syscall functions

 man/io_uring_enter.2     |  9 ++++-----
 man/io_uring_enter2.2    |  1 +
 man/io_uring_register.2  |  9 ++++-----
 man/io_uring_setup.2     |  8 +++-----
 src/Makefile             |  2 +-
 src/include/liburing.h   |  8 ++++++++
 src/liburing.map         |  4 ++++
 src/syscall.c            | 30 ++++++++++++++++++++++++++++++
 test/io_uring_enter.c    | 10 +++++-----
 test/io_uring_register.c | 34 ++++++++++++++++------------------
 test/io_uring_setup.c    |  4 ++--
 11 files changed, 78 insertions(+), 41 deletions(-)
 create mode 120000 man/io_uring_enter2.2
 create mode 100644 src/syscall.c

base-commit: bf248850dc2ae45d29d4fdde688e90d24f3dd6d2
-- 
Ammar Faizi