Re: [PATCH 2/3] io_uring/uring_cmd: allow non-iopoll cmds with IORING_SETUP_IOPOLL

[Jens Axboe <axboe@kernel.dk>]

On 2/17/26 7:18 PM, Caleb Sander Mateos wrote:
> On Thu, Feb 12, 2026 at 7:21?PM Caleb Sander Mateos
> <csander@purestorage.com> wrote:
>>
>> Currently, creating an io_uring with IORING_SETUP_IOPOLL requires all
>> requests issued to it to support iopoll. This prevents, for example,
>> using ublk zero-copy together with IORING_SETUP_IOPOLL, as ublk
>> zero-copy buffer registrations are performed using a uring_cmd. There's
>> no technical reason why these non-iopoll uring_cmds can't be supported.
>> They will either complete synchronously or via an external mechanism
>> that calls io_uring_cmd_done(), so they don't need to be polled.
>>
>> Allow uring_cmd requests to be issued to IORING_SETUP_IOPOLL io_urings
>> even if their files don't implement ->uring_cmd_iopoll(). For these
>> uring_cmd requests, skip initializing struct io_kiocb's iopoll fields,
>> don't insert the request into iopoll_list, and take the
>> io_req_complete_defer() or io_req_task_work_add() path in
>> __io_uring_cmd_done() instead of setting the iopoll_completed flag. Also
>> allow io_uring_cmd_mark_cancelable() to be called on these uring_cmds.
>> Assert that io_uring_cmd_mark_cancelable() is only called on
>> non-IORING_SETUP_IOPOLL io_urings or uring_cmds to files that don't
>> implement ->uring_cmd_iopoll().
>>
>> Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
>> ---
>>  io_uring/io_uring.c  |  4 +++-
>>  io_uring/uring_cmd.c | 11 +++++------
>>  2 files changed, 8 insertions(+), 7 deletions(-)
>>
>> diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
>> index c45af82dda3d..4e68a5168894 100644
>> --- a/io_uring/io_uring.c
>> +++ b/io_uring/io_uring.c
>> @@ -1417,11 +1417,13 @@ static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
>>
>>         if (ret == IOU_ISSUE_SKIP_COMPLETE) {
>>                 ret = 0;
>>
>>                 /* If the op doesn't have a file, we're not polling for it */
>> -               if ((req->ctx->flags & IORING_SETUP_IOPOLL) && def->iopoll_queue)
>> +               if ((req->ctx->flags & IORING_SETUP_IOPOLL) &&
>> +                   def->iopoll_queue && (!io_is_uring_cmd(req) ||
>> +                                         req->file->f_op->uring_cmd_iopoll))
>>                         io_iopoll_req_issued(req, issue_flags);
>>         }
>>         return ret;
>>  }
>>
>> diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
>> index ee7b49f47cb5..8df52e8f1c1b 100644
>> --- a/io_uring/uring_cmd.c
>> +++ b/io_uring/uring_cmd.c
>> @@ -108,12 +108,12 @@ void io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd,
>>          * Doing cancelations on IOPOLL requests are not supported. Both
>>          * because they can't get canceled in the block stack, but also
>>          * because iopoll completion data overlaps with the hash_node used
>>          * for tracking.
>>          */
>> -       if (ctx->flags & IORING_SETUP_IOPOLL)
>> -               return;
>> +       WARN_ON_ONCE(ctx->flags & IORING_SETUP_IOPOLL &&
>> +                    req->file->f_op->uring_cmd_iopoll);
>>
>>         if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) {
>>                 cmd->flags |= IORING_URING_CMD_CANCELABLE;
>>                 io_ring_submit_lock(ctx, issue_flags);
>>                 hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd);
>> @@ -165,11 +165,12 @@ void __io_uring_cmd_done(struct io_uring_cmd *ioucmd, s32 ret, u64 res2,
>>                 if (req->ctx->flags & IORING_SETUP_CQE_MIXED)
>>                         req->cqe.flags |= IORING_CQE_F_32;
>>                 io_req_set_cqe32_extra(req, res2, 0);
>>         }
>>         io_req_uring_cleanup(req, issue_flags);
>> -       if (req->ctx->flags & IORING_SETUP_IOPOLL) {
>> +       if (req->ctx->flags & IORING_SETUP_IOPOLL &&
>> +           req->file->f_op->uring_cmd_iopoll) {
> 
> I do worry that the pointer chasing here may be expensive, ->file and
> ->f_op could both be uncached. Would it make sense to add a flag to
> req->flags to indicate whether a request should actually be IOPOLLed?

I think adding a REQ_F flag for that similar to what is done for NOWAIT
etc would be a good idea.

-- 
Jens Axboe