Re: linux-next: build failure after merge of the block tree

[Jens Axboe <axboe@kernel.dk>]

On 1/24/26 4:46 AM, Stephen Rothwell wrote:
> Hi Jens,
> 
> On Fri, 23 Jan 2026 11:06:43 -0700 Jens Axboe <axboe@kernel.dk> wrote:
>>
>> On 1/23/26 11:00 AM, Jens Axboe wrote:
>>> On 1/23/26 10:42 AM, Mark Brown wrote:  
>>>> Hi all,
>>>>
>>>> After merging the block tree, today's linux-next build (x86 allmodconfig)
>>>> failed like this:
>>>>
>>>> In file included from /tmp/next/build/include/linux/string.h:386,
>>>>                  from /tmp/next/build/include/linux/bitmap.h:13,
>>>>                  from /tmp/next/build/include/linux/cpumask.h:11,
>>>>                  from /tmp/next/build/arch/x86/include/asm/paravirt.h:21,
>>>>                  from /tmp/next/build/arch/x86/include/asm/cpuid/api.h:57,
>>>>                  from /tmp/next/build/arch/x86/include/asm/processor.h:19,
>>>>                  from /tmp/next/build/include/linux/sched.h:13,
>>>>                  from /tmp/next/build/include/linux/io_uring.h:5,
>>>>                  from /tmp/next/build/io_uring/bpf_filter.c:7:
>>>> In function 'fortify_memset_chk',
>>>>     inlined from 'io_uring_populate_bpf_ctx' at /tmp/next/build/io_uring/bpf_filter.c:33:2:
>>>> /tmp/next/build/include/linux/fortify-string.h:480:25: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror=attribute-warning]
>>>>   480 |                         __write_overflow_field(p_size_field, size);
>>>>       |                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>>> cc1: all warnings being treated as errors
>>>>
>>>> Caused by commit
>>>>
>>>>   f1e3672e49e2c (io_uring: add support for BPF filtering for opcode restrictions)  
>>>
>>> Huh, that am I missing here? The struct looks as follows:
>>>
>>> struct io_uring_bpf_ctx {
>>> 	__u64	user_data;
>>> 	__u8	opcode;
>>> 	__u8	sqe_flags;
>>> 	__u8	pad[6];
>>> 	union {
>>> 		__u64	resv[6];
>>> 		struct {
>>> 			__u32	family;
>>> 			__u32	type;
>>> 			__u32	protocol;
>>> 		} socket;
>>> 		struct {
>>> 			__u64	flags;
>>> 			__u64	mode;
>>> 			__u64	resolve;
>>> 		} open;
>>> 	};
>>> };
>>>
>>> and the offending line is:
>>>
>>> memset(bctx->pad, 0, sizeof(bctx->pad) + sizeof(bctx->resv));
>>>
>>> which should clear from offset 10 (start of pad) for a total of 6 + 48
>>> bytes, which is 54 bytes. The size of the struct is 64b.
>>>
>>> I guess the part it doesn't like is that it thinks we're clearing the
>>> pad field, which would of course be way overwriting it. Guess we can do
>>> something ala:
>>>
>>> memset((void *) bctx + offsetof(struct io_uring_bpf_ctx, pad), 0,
>>> 	sizeof(bctx->pad) + sizeof(bctx->resv));
>>>
>>> to make it happier.  
>>
>> Folded that in and pushed it out, should be happy for you now. I wonder
>> if we have a helper for that...
> 
> The origin warning did suggest using struct_group().

Right, and I do use that in io_uring in other spots already, but it's
pretty ugly for user facing structs. Both because they are harder to
read, and because it's bleeding internal detail into userspace. So don't
really like that for this use case.

-- 
Jens Axboe