Re: [PATCH] io_uring/rsrc: remove registered buffer 1GB limit

["Clément Léger" <cleger@meta.com>]

On 5/5/26 15:26, Jens Axboe wrote:
> > 
> On 5/5/26 7:23 AM, Cl?ment L?ger wrote:
>> On 5/5/26 12:09, Jens Axboe wrote:
>>>> On 5/5/26 1:39 AM, Jens Axboe wrote:
>>>> There's no real reason to have a limit, as the memory is accounted by
>>>> the lockmem limits anyway, if any exist. io_pin_pages() will still
>>>> restrict the maximum allowed limit per buffer, which is INT_MAX
>>>> number of pages. For a 4kb page size system, the limit is 8TB.
>>>>
>>>> Reported-by: Andres Freund <andres@anarazel.de>
>>>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>>
>>> Forgot that I had a prep patch for this one... The branch is here:
>>>
>>> https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux.git/log/?h=io_uring-reg-buffers
>>>
>>> and notably the patch before this one is below, which bumps the ->len
>>> size of the io_mapped_ubuf. I'll send this out as a proper series later
>>> this week, this is 7.2 material obviously.
>>>
>>> commit 381e736515173a1fb78d2a86983d3ebfcf263597
>>> Author: Jens Axboe <axboe@kernel.dk>
>>> Date:   Mon May 4 05:40:16 2026 -0600
>>>
>>>       io_uring/rsrc: bump struct io_mapped_ubuf length field to size_t
>>>            In preparation for supporting bigger individual buffers, bump the length
>>>       field to a full 8-bytes with size_t rather than an unsigned int.
>>>            Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>>
>>> diff --git a/io_uring/fdinfo.c b/io_uring/fdinfo.c
>>> index c2d3e45544bb..f0ff4bd01b6d 100644
>>> --- a/io_uring/fdinfo.c
>>> +++ b/io_uring/fdinfo.c
>>> @@ -223,7 +223,7 @@ static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
>>>            if (ctx->buf_table.nodes[i])
>>>                buf = ctx->buf_table.nodes[i]->buf;
>>>            if (buf)
>>> -            seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, buf->len);
>>> +            seq_printf(m, "%5u: 0x%llx/%zu\n", i, buf->ubuf, buf->len);
>>>            else
>>>                seq_printf(m, "%5u: <none>\n", i);
>>>        }
>>> diff --git a/io_uring/rsrc.h b/io_uring/rsrc.h
>>> index 44e3386f7c1c..03521b50926c 100644
>>> --- a/io_uring/rsrc.h
>>> +++ b/io_uring/rsrc.h
>>> @@ -34,15 +34,15 @@ enum {
>>>      struct io_mapped_ubuf {
>>>        u64        ubuf;
>>> -    unsigned int    len;
>>> +    size_t        len;
>>>        unsigned int    nr_bvecs;
>>>        unsigned int    folio_shift;
>>>        refcount_t    refs;
>>> +    u8        flags;
>>> +    u8        dir;
>>>        unsigned long    acct_pages;
>>>        void        (*release)(void *);
>>>        void        *priv;
>>> -    u8        flags;
>>> -    u8        dir;
>>
>> Hi Jens,
>>
>> This seems like an unrelated change.
> 
> Hmm, how so? It's required for removing the 1GB restriction, as it bumps
> buf->len from a 32-bit unsigned to a 64-bit size_t.
> 
> Oh you mean moving flags and dir? That's just so it packs better,
> changing int would leave a 4-byte gap. Might as well move flags and dir
> near the 4b refcount_t to avoid bloating the struct.

Yes, I meant dir/flags but indeed, that makes sense !

Thanks,

Clément

>