From: Pavel Begunkov <[email protected]>
To: Jens Axboe <[email protected]>, Hao Xu <[email protected]>
Cc: [email protected], [email protected],
Jiufei Xue <[email protected]>,
Joseph Qi <[email protected]>
Subject: Re: [PATCH v3 RESEND] io_uring: add timeout support for io_uring_enter()
Date: Wed, 4 Nov 2020 19:27:00 +0000 [thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
On 04/11/2020 18:32, Jens Axboe wrote:
> On 11/4/20 10:50 AM, Jens Axboe wrote:
>> +struct io_uring_getevents_arg {
>> + sigset_t *sigmask;
>> + struct __kernel_timespec *ts;
>> +};
>> +
>
> I missed that this is still not right, I did bring it up in your last
> posting though - you can't have pointers as a user API, since the size
> of the pointer will vary depending on whether this is a 32-bit or 64-bit
> arch (or 32-bit app running on 64-bit kernel).
Maybe it would be better
1) to kill this extra indirection?
struct io_uring_getevents_arg {
- sigset_t *sigmask;
- struct __kernel_timespec *ts;
+ sigset_t sigmask;
+ struct __kernel_timespec ts;
};
then,
sigset_t *sig = (...)arg;
__kernel_timespec* ts = (...)(arg + offset);
It'd spare us from IORING_ENTER_GETEVENTS_TIMEOUT but we'd need
to find a way to disable some of them. E.g. don't use sigmask when
user don't want it, but sigsz == sizeof(io_uring_getevents_arg),
and parsing would look like
switch (argsz) {
case sizeof(struct io_uring_getevents_arg): {
struct __kernel_timespec ts = argp + ts_offset;
...
}
fallthrough;
case sizeof(sig): {
const sigset_t __user *sig = argp;
...
break;
}
default:
return -EINVAL;
}
2) and move all the parsing into io_cqring_wait(). That sounds better
performance-wise.
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 7e6945383907..2f533f6815ea 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -9158,8 +9158,8 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
> return -EINVAL;
> if (copy_from_user(&arg, argp, sizeof(arg)))
> return -EFAULT;
> - sig = arg.sigmask;
> - ts = arg.ts;
> + sig = u64_to_user_ptr(arg.sigmask);
> + ts = u64_to_user_ptr(arg.ts);
> } else {
> sig = (const sigset_t __user *)argp;
> ts = NULL;
> diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
> index fefee28c3ed8..0b104891df68 100644
> --- a/include/uapi/linux/io_uring.h
> +++ b/include/uapi/linux/io_uring.h
> @@ -338,8 +338,8 @@ enum {
> };
>
> struct io_uring_getevents_arg {
> - sigset_t *sigmask;
> - struct __kernel_timespec *ts;
> + __u64 sigmask;
> + __u64 ts;
> };
>
> #endif
>
--
Pavel Begunkov
next prev parent reply other threads:[~2020-11-04 19:30 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-11-02 8:50 [PATCH v3] io_uring: add timeout support for io_uring_enter() Hao Xu
2020-11-03 2:54 ` [PATCH v3 RESEND] " Hao Xu
2020-11-04 17:50 ` Jens Axboe
2020-11-04 18:32 ` Jens Axboe
2020-11-04 19:06 ` Jens Axboe
2020-11-04 19:27 ` Pavel Begunkov [this message]
2020-11-04 19:34 ` Jens Axboe
2020-11-04 20:16 ` Pavel Begunkov
2020-11-04 20:28 ` Jens Axboe
2020-11-04 20:50 ` Jens Axboe
2020-11-04 21:20 ` Pavel Begunkov
2020-11-04 21:27 ` Jens Axboe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox