* [PATCH] iio: buffer: kfifo: reject lengths exceeding the kmalloc limit
@ 2026-11-20 13:21 YuanZhe Liu
2026-11-20 16:23 ` Andy Shevchenko
0 siblings, 1 reply; 4+ messages in thread
From: YuanZhe Liu @ 2026-11-20 13:21 UTC (permalink / raw)
To: Jonathan Cameron
Cc: David Lechner, Nuno Sá,
Andy Shevchenko, linux-iio, linux-kernel, Bin Yu, Cheng Wen
__iio_allocate_kfifo() rejects lengths that would overflow an unsigned
int after kfifo rounds them up to the next power of two, but the byte
size handed to __kfifo_alloc() can still be beyond what kmalloc() can
serve.
The kfifo buffer is allocated with kmalloc(), which is limited to
KMALLOC_MAX_SIZE. A larger request has an allocation order above
MAX_PAGE_ORDER, so the page allocator's order sanity check fires and
enabling the buffer ends with a WARNING instead of a clean error.
The buffer length is set through buffer/length, which accepts any value
up to UINT_MAX. With the voltage and timestamp scan channels enabled,
a length of 262349 makes the rounded up size exceed KMALLOC_MAX_SIZE
and triggers the warning on buffer/enable.
Also reject lengths whose rounded up size does not fit into
KMALLOC_MAX_SIZE, so that enabling the buffer fails with -EINVAL.
Co-developed-by: Bin Yu <byu@xidian.edu.cn>
Signed-off-by: Bin Yu <byu@xidian.edu.cn>
Co-developed-by: Cheng Wen <wencheng@xidian.edu.cn>
Signed-off-by: Cheng Wen <wencheng@xidian.edu.cn>
Signed-off-by: YuanZhe Liu <25031212351@stu.xidian.edu.cn>
---
drivers/iio/buffer/kfifo_buf.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/drivers/iio/buffer/kfifo_buf.c b/drivers/iio/buffer/kfifo_buf.c
index b67ea6468226..2069d7ca4409 100644
--- a/drivers/iio/buffer/kfifo_buf.c
+++ b/drivers/iio/buffer/kfifo_buf.c
@@ -35,6 +35,15 @@ static inline int __iio_allocate_kfifo(struct iio_kfifo *buf,
if (roundup_pow_of_two(length) > UINT_MAX / bytes_per_datum)
return -EINVAL;
+ /*
+ * The kfifo memory is allocated with kmalloc(), so the size must also
+ * fit into the largest allocation kmalloc() can serve. Larger
+ * requests need an order above MAX_PAGE_ORDER and are refused by the
+ * page allocator with a warning.
+ */
+ if (roundup_pow_of_two(length) > KMALLOC_MAX_SIZE / bytes_per_datum)
+ return -EINVAL;
+
return __kfifo_alloc((struct __kfifo *)&buf->kf, length,
bytes_per_datum, GFP_KERNEL);
}
--
2.34.1
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] iio: buffer: kfifo: reject lengths exceeding the kmalloc limit
2026-11-20 13:21 [PATCH] iio: buffer: kfifo: reject lengths exceeding the kmalloc limit YuanZhe Liu
@ 2026-11-20 16:23 ` Andy Shevchenko
2026-11-20 16:44 ` [PATCH v2] " YuanZhe Liu
0 siblings, 1 reply; 4+ messages in thread
From: Andy Shevchenko @ 2026-11-20 16:23 UTC (permalink / raw)
To: YuanZhe Liu
Cc: Jonathan Cameron, David Lechner, Nuno Sá,
Andy Shevchenko, linux-iio, linux-kernel, Bin Yu, Cheng Wen
On Fri, Nov 20, 2026 at 09:21:29PM +0800, YuanZhe Liu wrote:
> __iio_allocate_kfifo() rejects lengths that would overflow an unsigned
> int after kfifo rounds them up to the next power of two, but the byte
> size handed to __kfifo_alloc() can still be beyond what kmalloc() can
> serve.
>
> The kfifo buffer is allocated with kmalloc(), which is limited to
> KMALLOC_MAX_SIZE. A larger request has an allocation order above
> MAX_PAGE_ORDER, so the page allocator's order sanity check fires and
> enabling the buffer ends with a WARNING instead of a clean error.
>
> The buffer length is set through buffer/length, which accepts any value
> up to UINT_MAX. With the voltage and timestamp scan channels enabled,
> a length of 262349 makes the rounded up size exceed KMALLOC_MAX_SIZE
> and triggers the warning on buffer/enable.
>
> Also reject lengths whose rounded up size does not fit into
> KMALLOC_MAX_SIZE, so that enabling the buffer fails with -EINVAL.
...
> if (roundup_pow_of_two(length) > UINT_MAX / bytes_per_datum)
> return -EINVAL;
So many developers of this patch and there is no clarification
why we need to keep this check.
> + /*
> + * The kfifo memory is allocated with kmalloc(), so the size must also
> + * fit into the largest allocation kmalloc() can serve. Larger
> + * requests need an order above MAX_PAGE_ORDER and are refused by the
> + * page allocator with a warning.
> + */
> + if (roundup_pow_of_two(length) > KMALLOC_MAX_SIZE / bytes_per_datum)
> + return -EINVAL;
--
With Best Regards,
Andy Shevchenko
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v2] iio: buffer: kfifo: reject lengths exceeding the kmalloc limit
2026-11-20 16:23 ` Andy Shevchenko
@ 2026-11-20 16:44 ` YuanZhe Liu
2026-11-21 1:55 ` Jonathan Cameron
0 siblings, 1 reply; 4+ messages in thread
From: YuanZhe Liu @ 2026-11-20 16:44 UTC (permalink / raw)
To: Jonathan Cameron
Cc: Andy Shevchenko, David Lechner, Nuno Sá,
linux-iio, linux-kernel, Bin Yu, Cheng Wen
__iio_allocate_kfifo() only rejects lengths that would overflow an
unsigned int after kfifo rounds them up to the next power of two, but
the byte size handed to __kfifo_alloc() can still be beyond what
kmalloc() can serve.
The kfifo buffer is allocated with kmalloc(), which is limited to
KMALLOC_MAX_SIZE. A larger request has an allocation order above
MAX_PAGE_ORDER, so the page allocator's order sanity check fires and
enabling the buffer ends with a WARNING instead of a clean error.
The buffer length is set through buffer/length, which accepts any value
up to UINT_MAX. With the voltage and timestamp scan channels enabled,
a length of 262349 makes the rounded up size exceed KMALLOC_MAX_SIZE
and triggers the warning on buffer/enable.
Reject lengths whose rounded up size does not fit into KMALLOC_MAX_SIZE
so that enabling the buffer fails with -EINVAL. This bound is stricter
than the unsigned int overflow check it replaces, because
KMALLOC_MAX_SIZE is far below UINT_MAX, so that check is removed.
Co-developed-by: Bin Yu <byu@xidian.edu.cn>
Signed-off-by: Bin Yu <byu@xidian.edu.cn>
Co-developed-by: Cheng Wen <wencheng@xidian.edu.cn>
Signed-off-by: Cheng Wen <wencheng@xidian.edu.cn>
Signed-off-by: YuanZhe Liu <25031212351@stu.xidian.edu.cn>
---
v2:
- Drop the unsigned int overflow check. It is subsumed by the new
KMALLOC_MAX_SIZE bound: a length that would overflow the kfifo
element counter exceeds KMALLOC_MAX_SIZE as well (Andy Shevchenko).
---
drivers/iio/buffer/kfifo_buf.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)
diff --git a/drivers/iio/buffer/kfifo_buf.c b/drivers/iio/buffer/kfifo_buf.c
index b67ea6468226..65906dfffc04 100644
--- a/drivers/iio/buffer/kfifo_buf.c
+++ b/drivers/iio/buffer/kfifo_buf.c
@@ -29,10 +29,14 @@ static inline int __iio_allocate_kfifo(struct iio_kfifo *buf,
return -EINVAL;
/*
- * Make sure we don't overflow an unsigned int after kfifo rounds up to
- * the next power of 2.
+ * The kfifo memory is allocated with kmalloc(), which cannot serve
+ * more than KMALLOC_MAX_SIZE. A larger request needs an allocation
+ * order above MAX_PAGE_ORDER and is refused by the page allocator
+ * with a warning. KMALLOC_MAX_SIZE is far below UINT_MAX, so this
+ * also keeps the element count rounded up by kfifo within an
+ * unsigned int.
*/
- if (roundup_pow_of_two(length) > UINT_MAX / bytes_per_datum)
+ if (roundup_pow_of_two(length) > KMALLOC_MAX_SIZE / bytes_per_datum)
return -EINVAL;
return __kfifo_alloc((struct __kfifo *)&buf->kf, length,
--
2.34.1
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH v2] iio: buffer: kfifo: reject lengths exceeding the kmalloc limit
2026-11-20 16:44 ` [PATCH v2] " YuanZhe Liu
@ 2026-11-21 1:55 ` Jonathan Cameron
0 siblings, 0 replies; 4+ messages in thread
From: Jonathan Cameron @ 2026-11-21 1:55 UTC (permalink / raw)
To: YuanZhe Liu, Nuno Sá
Cc: Andy Shevchenko, David Lechner, linux-iio, linux-kernel, Bin Yu,
Cheng Wen
On Sat, 21 Nov 2026 00:44:03 +0800
YuanZhe Liu <25031212351@stu.xidian.edu.cn> wrote:
> __iio_allocate_kfifo() only rejects lengths that would overflow an
> unsigned int after kfifo rounds them up to the next power of two, but
> the byte size handed to __kfifo_alloc() can still be beyond what
> kmalloc() can serve.
>
> The kfifo buffer is allocated with kmalloc(), which is limited to
> KMALLOC_MAX_SIZE. A larger request has an allocation order above
> MAX_PAGE_ORDER, so the page allocator's order sanity check fires and
> enabling the buffer ends with a WARNING instead of a clean error.
>
> The buffer length is set through buffer/length, which accepts any value
> up to UINT_MAX. With the voltage and timestamp scan channels enabled,
> a length of 262349 makes the rounded up size exceed KMALLOC_MAX_SIZE
> and triggers the warning on buffer/enable.
>
> Reject lengths whose rounded up size does not fit into KMALLOC_MAX_SIZE
> so that enabling the buffer fails with -EINVAL. This bound is stricter
> than the unsigned int overflow check it replaces, because
> KMALLOC_MAX_SIZE is far below UINT_MAX, so that check is removed.
There is a lot here, when it could just say
__kfifo_alloc() uses kmalloc to allocate the FIFO. Tighten the limit
of __iio_allocate_kfifo() to match. This results in -EINVAL being
returned on oversized requests rather than XXXX.
Note I can't tell from this description what that ABI change is
as you don't include that information which is neeeded.
However that's an implementation detail of KFIFOs sneaking
across the boundary. Given the allocation call fails anyway, why
do we care about short cutting it?
>
> Co-developed-by: Bin Yu <byu@xidian.edu.cn>
> Signed-off-by: Bin Yu <byu@xidian.edu.cn>
> Co-developed-by: Cheng Wen <wencheng@xidian.edu.cn>
> Signed-off-by: Cheng Wen <wencheng@xidian.edu.cn>
> Signed-off-by: YuanZhe Liu <25031212351@stu.xidian.edu.cn>
> ---
> v2:
> - Drop the unsigned int overflow check. It is subsumed by the new
> KMALLOC_MAX_SIZE bound: a length that would overflow the kfifo
> element counter exceeds KMALLOC_MAX_SIZE as well (Andy Shevchenko).
Two process things.
1) Slow down. Even for a small patch like this, a few days for
first round of reviews.
2) Don't send new versions in replies to old ones. Generally means
your emails are hidden away and can occasionally trip up tooling.
I'm not aware of any area of the kernel that takes things like
this. Seems to be a pattern the LLMs have learnt from other
projects.
> ---
> drivers/iio/buffer/kfifo_buf.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/iio/buffer/kfifo_buf.c b/drivers/iio/buffer/kfifo_buf.c
> index b67ea6468226..65906dfffc04 100644
> --- a/drivers/iio/buffer/kfifo_buf.c
> +++ b/drivers/iio/buffer/kfifo_buf.c
> @@ -29,10 +29,14 @@ static inline int __iio_allocate_kfifo(struct iio_kfifo *buf,
> return -EINVAL;
>
> /*
> - * Make sure we don't overflow an unsigned int after kfifo rounds up to
> - * the next power of 2.
> + * The kfifo memory is allocated with kmalloc(), which cannot serve
> + * more than KMALLOC_MAX_SIZE. A larger request needs an allocation
> + * order above MAX_PAGE_ORDER and is refused by the page allocator
> + * with a warning. KMALLOC_MAX_SIZE is far below UINT_MAX, so this
> + * also keeps the element count rounded up by kfifo within an
> + * unsigned int.
> */
> - if (roundup_pow_of_two(length) > UINT_MAX / bytes_per_datum)
> + if (roundup_pow_of_two(length) > KMALLOC_MAX_SIZE / bytes_per_datum)
> return -EINVAL;
>
> return __kfifo_alloc((struct __kfifo *)&buf->kf, length,
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-11-21 1:55 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-20 13:21 [PATCH] iio: buffer: kfifo: reject lengths exceeding the kmalloc limit YuanZhe Liu
2026-11-20 16:23 ` Andy Shevchenko
2026-11-20 16:44 ` [PATCH v2] " YuanZhe Liu
2026-11-21 1:55 ` Jonathan Cameron
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®