public inbox for [email protected]
 help / color / mirror / Atom feed
From: Jens Axboe <[email protected]>
To: Xiaoguang Wang <[email protected]>,
	[email protected]
Cc: [email protected]
Subject: Re: [PATCH] io_uring: add io_uring_enter(2) fixed file support
Date: Thu, 3 Mar 2022 06:38:05 -0700	[thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>

On 3/2/22 10:28 PM, Xiaoguang Wang wrote:
> IORING_REGISTER_FILES is a good feature to reduce fget/fput overhead for
> each IO we do on file, but still left one, which is io_uring_enter(2).
> In io_uring_enter(2), it still fget/fput io_ring fd. I have observed
> this overhead in some our internal oroutine implementations based on
> io_uring with low submit batch. To totally remove fget/fput overhead in
> io_uring, we may add a small struct file cache in io_uring_task and add
> a new IORING_ENTER_FIXED_FILE flag. Currently the capacity of this file
> cache is 16, wihcih I think it maybe enough, also not that this cache is
> per-thread.

Would indeed be nice to get rid of, can be a substantial amount of time
wasted in fdget/fdput. Does this resolve dependencies correctly if
someone passes the ring fd? Adding ring registration to test/ring-leak.c
from the liburing repo would be a useful exercise.

Comments below.

> @@ -8739,8 +8742,16 @@ static __cold int io_uring_alloc_task_context(struct task_struct *task,
>  	if (unlikely(!tctx))
>  		return -ENOMEM;
>  
> +	tctx->registered_files = kzalloc(sizeof(struct file *) * IO_RINGFD_REG_MAX,
> +					 GFP_KERNEL);

kcalloc()

> +static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx, bool locked);
> +
> +static int io_ringfd_register(struct io_ring_ctx *ctx, void __user *arg)
> +{
> +	struct io_uring_fd_reg reg;
> +	struct io_uring_task *tctx;
> +	struct file *file;
> +	int ret;
> +
> +	if (copy_from_user(&reg, arg, sizeof(struct io_uring_fd_reg)))
> +		return -EFAULT;
> +	if (reg.offset > IO_RINGFD_REG_MAX)
> +		return -EINVAL;
> +
> +	ret = io_uring_add_tctx_node(ctx, true);
> +	if (unlikely(ret))
> +		return ret;

Can we safely drop ctx->uring_lock around this call instead? The locked
argument is always kind of ugly, and you currently have a deadlock as
far as I can tell from io_uring_alloc_task_context() that now holds
ctx->uring_lock, and then calling io_init_wq_offload() which will grab
it again.

Why not just use io_uring_rsrc_update here rather than add a new type?

-- 
Jens Axboe


  parent reply	other threads:[~2022-03-03 13:38 UTC|newest]

Thread overview: 28+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-03-03  5:28 [PATCH] io_uring: add io_uring_enter(2) fixed file support Xiaoguang Wang
2022-03-03  8:56 ` Hao Xu
2022-03-03 13:38 ` Jens Axboe [this message]
2022-03-03 14:36   ` Jens Axboe
2022-03-03 14:40     ` Jens Axboe
2022-03-03 16:31       ` Jens Axboe
2022-03-03 17:18         ` Jens Axboe
2022-03-03 20:41           ` Jens Axboe
2022-03-03 21:19             ` Jens Axboe
2022-03-04  0:07               ` Jens Axboe
2022-03-04 13:39                 ` Xiaoguang Wang
2022-03-04 13:44                   ` Jens Axboe
2022-03-04 15:16                     ` Xiaoguang Wang
2022-03-04 15:22                       ` Jens Axboe
2022-03-08  8:38                         ` Xiaoguang Wang
2022-03-08 13:10                           ` Jens Axboe
2022-03-03 22:24             ` Vito Caputo
2022-03-03 22:26               ` Jens Axboe
2022-03-04  1:49         ` Pavel Begunkov
2022-03-04  2:18           ` Jens Axboe
2022-03-04  2:28             ` Pavel Begunkov
2022-03-04  2:35               ` Pavel Begunkov
2022-03-04  2:43               ` Jens Axboe
2022-03-04  1:52         ` Pavel Begunkov
2022-03-04  2:19           ` Jens Axboe
2022-03-04  2:39             ` Pavel Begunkov
2022-03-04  3:03               ` Jens Axboe
2022-04-21 14:16     ` Hao Xu

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] \
    /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