[PATCH] Add IORING_FEAT_FILES_SKIP feature flag

[PATCH] Add IORING_FEAT_FILES_SKIP feature flag

[Drew DeVault]

This is aliased to IORING_FEAT_NATIVE_WORKERS, which was shipped in the
same kernel release. A separate flag is useful because the features are
unrelated, so user code testing for IORING_FEAT_NATIVE_WORKERS before
using FILES_SKIP is not very obvious in intent.

Signed-off-by: Drew DeVault <sir@cmpwn.com>
---
 man/io_uring_register.2         | 10 ++++++++--
 src/include/liburing/io_uring.h |  1 +
 2 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/man/io_uring_register.2 b/man/io_uring_register.2
index 5326a87..c71ce40 100644
--- a/man/io_uring_register.2
+++ b/man/io_uring_register.2
@@ -156,8 +156,14 @@ since 5.5.
 File descriptors can be skipped if they are set to
 .B IORING_REGISTER_FILES_SKIP.
 Skipping an fd will not touch the file associated with the previous
-fd at that index. Available since 5.12.
-
+fd at that index. Available since 5.12. Availability of this feature is
+indicated by the presence of the
+.B IORING_FEAT_FILES_SKIP
+bit in the
+.I features
+field of the
+.I io_uring_params
+structure.
 
 .TP
 .B IORING_UNREGISTER_FILES
diff --git a/src/include/liburing/io_uring.h b/src/include/liburing/io_uring.h
index 5a3cb90..091dcf7 100644
--- a/src/include/liburing/io_uring.h
+++ b/src/include/liburing/io_uring.h
@@ -285,6 +285,7 @@ struct io_uring_params {
 #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
 #define IORING_FEAT_EXT_ARG		(1U << 8)
 #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
+#define IORING_FEAT_FILES_SKIP		IORING_FEAT_NATIVE_WORKERS
 
 /*
  * io_uring_register(2) opcodes and arguments
-- 
2.31.1

Re: [PATCH] Add IORING_FEAT_FILES_SKIP feature flag

[Jens Axboe]

On 5/18/21 10:12 AM, Drew DeVault wrote:
> This is aliased to IORING_FEAT_NATIVE_WORKERS, which was shipped in the
> same kernel release. A separate flag is useful because the features are
> unrelated, so user code testing for IORING_FEAT_NATIVE_WORKERS before
> using FILES_SKIP is not very obvious in intent.
> 
> Signed-off-by: Drew DeVault <sir@cmpwn.com>
> ---
>  man/io_uring_register.2         | 10 ++++++++--
>  src/include/liburing/io_uring.h |  1 +
>  2 files changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/man/io_uring_register.2 b/man/io_uring_register.2
> index 5326a87..c71ce40 100644
> --- a/man/io_uring_register.2
> +++ b/man/io_uring_register.2
> @@ -156,8 +156,14 @@ since 5.5.
>  File descriptors can be skipped if they are set to
>  .B IORING_REGISTER_FILES_SKIP.
>  Skipping an fd will not touch the file associated with the previous
> -fd at that index. Available since 5.12.
> -
> +fd at that index. Available since 5.12. Availability of this feature is
> +indicated by the presence of the
> +.B IORING_FEAT_FILES_SKIP
> +bit in the
> +.I features
> +field of the
> +.I io_uring_params
> +structure.
>  
>  .TP
>  .B IORING_UNREGISTER_FILES
> diff --git a/src/include/liburing/io_uring.h b/src/include/liburing/io_uring.h
> index 5a3cb90..091dcf7 100644
> --- a/src/include/liburing/io_uring.h
> +++ b/src/include/liburing/io_uring.h
> @@ -285,6 +285,7 @@ struct io_uring_params {
>  #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
>  #define IORING_FEAT_EXT_ARG		(1U << 8)
>  #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
> +#define IORING_FEAT_FILES_SKIP		IORING_FEAT_NATIVE_WORKERS

I don't think this is a great idea. It can be used as a "probably we have
this feature" in userspace, but I don't like aliasing on the kernel side.


-- 
Jens Axboe

Re: [PATCH] Add IORING_FEAT_FILES_SKIP feature flag

[Drew DeVault]

On Thu May 20, 2021 at 9:32 AM EDT, Jens Axboe wrote:
> > diff --git a/src/include/liburing/io_uring.h b/src/include/liburing/io_uring.h
> > index 5a3cb90..091dcf7 100644
> > --- a/src/include/liburing/io_uring.h
> > +++ b/src/include/liburing/io_uring.h
> > @@ -285,6 +285,7 @@ struct io_uring_params {
> >  #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
> >  #define IORING_FEAT_EXT_ARG		(1U << 8)
> >  #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
> > +#define IORING_FEAT_FILES_SKIP		IORING_FEAT_NATIVE_WORKERS
>
> I don't think this is a great idea. It can be used as a "probably we
> have this feature" in userspace, but I don't like aliasing on the
> kernel side.

This patch is for liburing, following the feedback on the kernel patch
(which didn't alias, but regardless).

Re: [PATCH] Add IORING_FEAT_FILES_SKIP feature flag

[Pavel Begunkov]

On 5/20/21 10:55 PM, Drew DeVault wrote:
> On Thu May 20, 2021 at 9:32 AM EDT, Jens Axboe wrote:
>>> diff --git a/src/include/liburing/io_uring.h b/src/include/liburing/io_uring.h
>>> index 5a3cb90..091dcf7 100644
>>> --- a/src/include/liburing/io_uring.h
>>> +++ b/src/include/liburing/io_uring.h
>>> @@ -285,6 +285,7 @@ struct io_uring_params {
>>>  #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
>>>  #define IORING_FEAT_EXT_ARG		(1U << 8)
>>>  #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
>>> +#define IORING_FEAT_FILES_SKIP		IORING_FEAT_NATIVE_WORKERS
>>
>> I don't think this is a great idea. It can be used as a "probably we
>> have this feature" in userspace, but I don't like aliasing on the
>> kernel side.
> 
> This patch is for liburing, following the feedback on the kernel patch
> (which didn't alias, but regardless).

This file is a copy (almost) of the kernel's uapi header, so better be
off this file and have naming that wouldn't alias with names in this
header.

I think that's the problem Jens mean. Jens, is it? And I do still
believe it's a better way to not have an itchy one release gap
between actual feature introduction and new feat flag.

-- 
Pavel Begunkov

Re: [PATCH] Add IORING_FEAT_FILES_SKIP feature flag

[Drew DeVault]

On Thu May 20, 2021 at 6:54 PM EDT, Pavel Begunkov wrote:
> On 5/20/21 10:55 PM, Drew DeVault wrote:
> > On Thu May 20, 2021 at 9:32 AM EDT, Jens Axboe wrote:
> >>> diff --git a/src/include/liburing/io_uring.h b/src/include/liburing/io_uring.h
> >>> index 5a3cb90..091dcf7 100644
> >>> --- a/src/include/liburing/io_uring.h
> >>> +++ b/src/include/liburing/io_uring.h
> >>> @@ -285,6 +285,7 @@ struct io_uring_params {
> >>>  #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
> >>>  #define IORING_FEAT_EXT_ARG		(1U << 8)
> >>>  #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
> >>> +#define IORING_FEAT_FILES_SKIP		IORING_FEAT_NATIVE_WORKERS
> >>
> >> I don't think this is a great idea. It can be used as a "probably we
> >> have this feature" in userspace, but I don't like aliasing on the
> >> kernel side.
> > 
> > This patch is for liburing, following the feedback on the kernel patch
> > (which didn't alias, but regardless).
>
> This file is a copy (almost) of the kernel's uapi header, so better be
> off this file and have naming that wouldn't alias with names in this
> header.
>
> I think that's the problem Jens mean. Jens, is it? And I do still
> believe it's a better way to not have an itchy one release gap
> between actual feature introduction and new feat flag.

It does itch, but I don't think it's actually wrong, per-se. Unless we
expect features to be removed or optionally available, it might make
sense to have used an incrementing number rather than a bitfield.

What should userspace programs do if they're not sure if they can use
FILES_SKIP? What was the previous behavior if -2 is used? I'm guessing
it was EINVAL, EBADF, something like that.