[PATCH v4 0/6] block/md/dm: set chunk_sectors from stacked dev stripe size

[PATCH v4 0/6] block/md/dm: set chunk_sectors from stacked dev stripe size

[John Garry]

This value in io_min is used to configure any atomic write limit for the
stacked device. The idea is that the atomic write unit max is a
power-of-2 factor of the stripe size, and the stripe size is available
in io_min.

Using io_min causes issues, as:
a. it may be mutated
b. the check for io_min being set for determining if we are dealing with
a striped device is hard to get right, as reported in [0].

This series now sets chunk_sectors limit to share stripe size.

[0] https://lore.kernel.org/linux-block/888f3b1d-7817-4007-b3b3-1a2ea04df771@linux.ibm.com/T/#mecca17129f72811137d3c2f1e477634e77f06781

Based on 73d9cb37478f (block/for-6.17/block) block: remove pktcdvd driver

This series fixes issues for v6.16, but it's prob better to have this in
v6.17 at this stage.

Differences to v3:
- relocate max_pow_of_two_factor() to common header and rework (Mikulas)
- cater for overflow from chunk sectors (Mikulas)

Differences to v2:
- Add RB tags (thanks!)

Differences to RFC:
- sanitize chunk_sectors for atomic write limits
- set chunk_sectors in stripe_io_hints()

John Garry (6):
  ilog2: add max_pow_of_two_factor()
  block: sanitize chunk_sectors for atomic write limits
  md/raid0: set chunk_sectors limit
  md/raid10: set chunk_sectors limit
  dm-stripe: limit chunk_sectors to the stripe size
  block: use chunk_sectors when evaluating stacked atomic write limits

 block/blk-settings.c   | 66 +++++++++++++++++++++++++++---------------
 drivers/md/dm-stripe.c |  1 +
 drivers/md/raid0.c     |  1 +
 drivers/md/raid10.c    |  1 +
 fs/xfs/xfs_mount.c     |  5 ----
 include/linux/log2.h   | 14 +++++++++
 6 files changed, 60 insertions(+), 28 deletions(-)

-- 
2.43.5

[PATCH v4 1/6] ilog2: add max_pow_of_two_factor()

[John Garry]

Relocate the function max_pow_of_two_factor() to common ilog2.h from the
xfs code, as it will be used elsewhere.

Also simplify the function, as advised by Mikulas Patocka.

Signed-off-by: John Garry <john.g.garry@oracle.com>
---
 fs/xfs/xfs_mount.c   |  5 -----
 include/linux/log2.h | 14 ++++++++++++++
 2 files changed, 14 insertions(+), 5 deletions(-)

diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index 29276fe60df9..6c669ae082d4 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -672,11 +672,6 @@ static inline xfs_extlen_t xfs_calc_atomic_write_max(struct xfs_mount *mp)
 	return rounddown_pow_of_two(XFS_B_TO_FSB(mp, MAX_RW_COUNT));
 }
 
-static inline unsigned int max_pow_of_two_factor(const unsigned int nr)
-{
-	return 1 << (ffs(nr) - 1);
-}
-
 /*
  * If the data device advertises atomic write support, limit the size of data
  * device atomic writes to the greatest power-of-two factor of the AG size so
diff --git a/include/linux/log2.h b/include/linux/log2.h
index 1366cb688a6d..2eac3fc9303d 100644
--- a/include/linux/log2.h
+++ b/include/linux/log2.h
@@ -255,4 +255,18 @@ int __bits_per(unsigned long n)
 	) :					\
 	__bits_per(n)				\
 )
+
+/**
+ * max_pow_of_two_factor - return highest power-of-2 factor
+ * @n: parameter
+ *
+ * find highest power-of-2 which is evenly divisible into n.
+ * 0 is returned for n == 0 or 1.
+ */
+static inline __attribute__((const))
+unsigned int max_pow_of_two_factor(unsigned int n)
+{
+	return n & -n;
+}
+
 #endif /* _LINUX_LOG2_H */
-- 
2.43.5

[PATCH v4 2/6] block: sanitize chunk_sectors for atomic write limits

[John Garry]

Currently we just ensure that a non-zero value in chunk_sectors aligns
with any atomic write boundary, as the blk boundary functionality uses
both these values.

However it is also improper to have atomic write unit max > chunk_sectors
(for non-zero chunk_sectors), as this would lead to splitting of atomic
write bios (which is disallowed).

Sanitize atomic write unit max against chunk_sectors to avoid any
potential problems.

Fixes: d00eea91deaf3 ("block: Add extra checks in blk_validate_atomic_write_limits()")
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: John Garry <john.g.garry@oracle.com>
---
 block/blk-settings.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/block/blk-settings.c b/block/blk-settings.c
index a000daafbfb4..761c6ccf5af7 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -181,6 +181,7 @@ static void blk_atomic_writes_update_limits(struct queue_limits *lim)
 static void blk_validate_atomic_write_limits(struct queue_limits *lim)
 {
 	unsigned int boundary_sectors;
+	unsigned long chunk_bytes;
 
 	if (!(lim->features & BLK_FEAT_ATOMIC_WRITES))
 		goto unsupported;
@@ -202,6 +203,13 @@ static void blk_validate_atomic_write_limits(struct queue_limits *lim)
 			 lim->atomic_write_hw_max))
 		goto unsupported;
 
+	chunk_bytes = lim->chunk_sectors << SECTOR_SHIFT;
+	if (chunk_bytes) {
+		if (WARN_ON_ONCE(lim->atomic_write_hw_unit_max >
+			chunk_bytes))
+			goto unsupported;
+	}
+
 	boundary_sectors = lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
 
 	if (boundary_sectors) {
-- 
2.43.5

[PATCH v4 3/6] md/raid0: set chunk_sectors limit

[John Garry]

Currently we use min io size as the chunk size when deciding on the
atomic write size limits - see blk_stack_atomic_writes_head().

The limit min_io size is not a reliable value to store the chunk size, as
this may be mutated by the block stacking code. Such an example would be
for the min io size less than the physical block size, and the min io size
is raised to the physical block size - see blk_stack_limits().

The block stacking limits will rely on chunk_sectors in future,
so set this value (to the chunk size).

Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: John Garry <john.g.garry@oracle.com>
---
 drivers/md/raid0.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/md/raid0.c b/drivers/md/raid0.c
index d8f639f4ae12..cbe2a9054cb9 100644
--- a/drivers/md/raid0.c
+++ b/drivers/md/raid0.c
@@ -384,6 +384,7 @@ static int raid0_set_limits(struct mddev *mddev)
 	lim.max_write_zeroes_sectors = mddev->chunk_sectors;
 	lim.io_min = mddev->chunk_sectors << 9;
 	lim.io_opt = lim.io_min * mddev->raid_disks;
+	lim.chunk_sectors = mddev->chunk_sectors;
 	lim.features |= BLK_FEAT_ATOMIC_WRITES;
 	err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
 	if (err)
-- 
2.43.5

[PATCH v4 4/6] md/raid10: set chunk_sectors limit

[John Garry]

Same as done for raid0, set chunk_sectors limit to appropriately set the
atomic write size limit.

Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: John Garry <john.g.garry@oracle.com>
---
 drivers/md/raid10.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index b74780af4c22..97065bb26f43 100644
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c
@@ -4004,6 +4004,7 @@ static int raid10_set_queue_limits(struct mddev *mddev)
 	md_init_stacking_limits(&lim);
 	lim.max_write_zeroes_sectors = 0;
 	lim.io_min = mddev->chunk_sectors << 9;
+	lim.chunk_sectors = mddev->chunk_sectors;
 	lim.io_opt = lim.io_min * raid10_nr_stripes(conf);
 	lim.features |= BLK_FEAT_ATOMIC_WRITES;
 	err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
-- 
2.43.5

[PATCH v4 5/6] dm-stripe: limit chunk_sectors to the stripe size

[John Garry]

Same as done for raid0, set chunk_sectors limit to appropriately set the
atomic write size limit.

Setting chunk_sectors limit in this way overrides the stacked limit
already calculated based on the bottom device limits. This is ok, as
when any bios are sent to the bottom devices, the block layer will still
respect the bottom device chunk_sectors.

Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: John Garry <john.g.garry@oracle.com>
---
 drivers/md/dm-stripe.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/md/dm-stripe.c b/drivers/md/dm-stripe.c
index a7dc04bd55e5..5bbbdf8fc1bd 100644
--- a/drivers/md/dm-stripe.c
+++ b/drivers/md/dm-stripe.c
@@ -458,6 +458,7 @@ static void stripe_io_hints(struct dm_target *ti,
 	struct stripe_c *sc = ti->private;
 	unsigned int chunk_size = sc->chunk_size << SECTOR_SHIFT;
 
+	limits->chunk_sectors = sc->chunk_size;
 	limits->io_min = chunk_size;
 	limits->io_opt = chunk_size * sc->stripes;
 }
-- 
2.43.5

[PATCH v4 6/6] block: use chunk_sectors when evaluating stacked atomic write limits

[John Garry]

The atomic write unit max value is limited by any stacked device stripe
size.

It is required that the atomic write unit is a power-of-2 factor of the
stripe size.

Currently we use io_min limit to hold the stripe size, and check for a
io_min <= SECTOR_SIZE when deciding if we have a striped stacked device.

Nilay reports that this causes a problem when the physical block size is
greater than SECTOR_SIZE [0].

Furthermore, io_min may be mutated when stacking devices, and this makes
it a poor candidate to hold the stripe size. Such an example (of when
io_min may change) would be when the io_min is less than the physical
block size.

Use chunk_sectors to hold the stripe size, which is more appropriate.

[0] https://lore.kernel.org/linux-block/888f3b1d-7817-4007-b3b3-1a2ea04df771@linux.ibm.com/T/#mecca17129f72811137d3c2f1e477634e77f06781

Signed-off-by: John Garry <john.g.garry@oracle.com>
---
 block/blk-settings.c | 58 ++++++++++++++++++++++++++------------------
 1 file changed, 35 insertions(+), 23 deletions(-)

diff --git a/block/blk-settings.c b/block/blk-settings.c
index 761c6ccf5af7..3259cfac5d0d 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -597,41 +597,52 @@ static bool blk_stack_atomic_writes_boundary_head(struct queue_limits *t,
 	return true;
 }
 
-
-/* Check stacking of first bottom device */
-static bool blk_stack_atomic_writes_head(struct queue_limits *t,
-				struct queue_limits *b)
+static void blk_stack_atomic_writes_chunk_sectors(struct queue_limits *t)
 {
-	if (b->atomic_write_hw_boundary &&
-	    !blk_stack_atomic_writes_boundary_head(t, b))
-		return false;
+	unsigned int chunk_sectors = t->chunk_sectors, chunk_bytes;
 
-	if (t->io_min <= SECTOR_SIZE) {
-		/* No chunk sectors, so use bottom device values directly */
-		t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
-		t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min;
-		t->atomic_write_hw_max = b->atomic_write_hw_max;
-		return true;
-	}
+	if (!chunk_sectors)
+		return;
+
+	/*
+	 * If chunk sectors is so large that its value in bytes overflows
+	 * UINT_MAX, then just shift it down so it definitely will fit.
+	 * We don't support atomic writes of such a large size anyway.
+	 */
+	if ((unsigned long)chunk_sectors << SECTOR_SHIFT > UINT_MAX)
+		chunk_bytes = chunk_sectors;
+	else
+		chunk_bytes = chunk_sectors << SECTOR_SHIFT;
 
 	/*
 	 * Find values for limits which work for chunk size.
 	 * b->atomic_write_hw_unit_{min, max} may not be aligned with chunk
-	 * size (t->io_min), as chunk size is not restricted to a power-of-2.
+	 * size, as the chunk size is not restricted to a power-of-2.
 	 * So we need to find highest power-of-2 which works for the chunk
 	 * size.
-	 * As an example scenario, we could have b->unit_max = 16K and
-	 * t->io_min = 24K. For this case, reduce t->unit_max to a value
-	 * aligned with both limits, i.e. 8K in this example.
+	 * As an example scenario, we could have t->unit_max = 16K and
+	 * t->chunk_sectors = 24KB. For this case, reduce t->unit_max to a
+	 * value aligned with both limits, i.e. 8K in this example.
 	 */
-	t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
-	while (t->io_min % t->atomic_write_hw_unit_max)
-		t->atomic_write_hw_unit_max /= 2;
+	t->atomic_write_hw_unit_max = min(t->atomic_write_hw_unit_max,
+					max_pow_of_two_factor(chunk_bytes));
 
-	t->atomic_write_hw_unit_min = min(b->atomic_write_hw_unit_min,
+	t->atomic_write_hw_unit_min = min(t->atomic_write_hw_unit_min,
 					  t->atomic_write_hw_unit_max);
-	t->atomic_write_hw_max = min(b->atomic_write_hw_max, t->io_min);
+	t->atomic_write_hw_max = min(t->atomic_write_hw_max, chunk_bytes);
+}
 
+/* Check stacking of first bottom device */
+static bool blk_stack_atomic_writes_head(struct queue_limits *t,
+				struct queue_limits *b)
+{
+	if (b->atomic_write_hw_boundary &&
+	    !blk_stack_atomic_writes_boundary_head(t, b))
+		return false;
+
+	t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
+	t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min;
+	t->atomic_write_hw_max = b->atomic_write_hw_max;
 	return true;
 }
 
@@ -659,6 +670,7 @@ static void blk_stack_atomic_writes_limits(struct queue_limits *t,
 
 	if (!blk_stack_atomic_writes_head(t, b))
 		goto unsupported;
+	blk_stack_atomic_writes_chunk_sectors(t);
 	return;
 
 unsupported:
-- 
2.43.5

Re: [PATCH v4 6/6] block: use chunk_sectors when evaluating stacked atomic write limits

[Nilay Shroff]

On 7/7/25 6:41 PM, John Garry wrote:
> The atomic write unit max value is limited by any stacked device stripe
> size.
> 
> It is required that the atomic write unit is a power-of-2 factor of the
> stripe size.
> 
> Currently we use io_min limit to hold the stripe size, and check for a
> io_min <= SECTOR_SIZE when deciding if we have a striped stacked device.
> 
> Nilay reports that this causes a problem when the physical block size is
> greater than SECTOR_SIZE [0].
> 
> Furthermore, io_min may be mutated when stacking devices, and this makes
> it a poor candidate to hold the stripe size. Such an example (of when
> io_min may change) would be when the io_min is less than the physical
> block size.
> 
> Use chunk_sectors to hold the stripe size, which is more appropriate.
> 
> [0] https://lore.kernel.org/linux-block/888f3b1d-7817-4007-b3b3-1a2ea04df771@linux.ibm.com/T/#mecca17129f72811137d3c2f1e477634e77f06781
> 
> Signed-off-by: John Garry <john.g.garry@oracle.com>
> ---
>  block/blk-settings.c | 58 ++++++++++++++++++++++++++------------------
>  1 file changed, 35 insertions(+), 23 deletions(-)
> 
> diff --git a/block/blk-settings.c b/block/blk-settings.c
> index 761c6ccf5af7..3259cfac5d0d 100644
> --- a/block/blk-settings.c
> +++ b/block/blk-settings.c
> @@ -597,41 +597,52 @@ static bool blk_stack_atomic_writes_boundary_head(struct queue_limits *t,
>  	return true;
>  }
>  
> -
> -/* Check stacking of first bottom device */
> -static bool blk_stack_atomic_writes_head(struct queue_limits *t,
> -				struct queue_limits *b)
> +static void blk_stack_atomic_writes_chunk_sectors(struct queue_limits *t)
>  {
> -	if (b->atomic_write_hw_boundary &&
> -	    !blk_stack_atomic_writes_boundary_head(t, b))
> -		return false;
> +	unsigned int chunk_sectors = t->chunk_sectors, chunk_bytes;
>  
> -	if (t->io_min <= SECTOR_SIZE) {
> -		/* No chunk sectors, so use bottom device values directly */
> -		t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
> -		t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min;
> -		t->atomic_write_hw_max = b->atomic_write_hw_max;
> -		return true;
> -	}
> +	if (!chunk_sectors)
> +		return;
> +
> +	/*
> +	 * If chunk sectors is so large that its value in bytes overflows
> +	 * UINT_MAX, then just shift it down so it definitely will fit.
> +	 * We don't support atomic writes of such a large size anyway.
> +	 */
> +	if ((unsigned long)chunk_sectors << SECTOR_SHIFT > UINT_MAX)
> +		chunk_bytes = chunk_sectors;
> +	else
> +		chunk_bytes = chunk_sectors << SECTOR_SHIFT;
>  

Can we use check_shl_overflow() here for checking overflow? Otherwise,
changes look good to me. I've also tested it using my NVMe disk which
supports up to 256kb of atomic writes. 

Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Tested-by: Nilay Shroff <nilay@linux.ibm.com>

Re: [PATCH v4 6/6] block: use chunk_sectors when evaluating stacked atomic write limits

[John Garry]

On 08/07/2025 13:27, Nilay Shroff wrote:
>> +	if ((unsigned long)chunk_sectors << SECTOR_SHIFT > UINT_MAX)
>> +		chunk_bytes = chunk_sectors;
>> +	else
>> +		chunk_bytes = chunk_sectors << SECTOR_SHIFT;
>>   
> Can we use check_shl_overflow() here for checking overflow?

ok, I can change.

> Otherwise,
> changes look good to me. I've also tested it using my NVMe disk which
> supports up to 256kb of atomic writes.
 > > Reviewed-by: Nilay Shroff<nilay@linux.ibm.com>
> Tested-by: Nilay Shroff<nilay@linux.ibm.com>

thanks

Re: [PATCH v4 6/6] block: use chunk_sectors when evaluating stacked atomic write limits

[Mikulas Patocka]

On Tue, 8 Jul 2025, Nilay Shroff wrote:

> 
> 
> On 7/7/25 6:41 PM, John Garry wrote:
> > The atomic write unit max value is limited by any stacked device stripe
> > size.
> > 
> > It is required that the atomic write unit is a power-of-2 factor of the
> > stripe size.
> > 
> > Currently we use io_min limit to hold the stripe size, and check for a
> > io_min <= SECTOR_SIZE when deciding if we have a striped stacked device.
> > 
> > Nilay reports that this causes a problem when the physical block size is
> > greater than SECTOR_SIZE [0].
> > 
> > Furthermore, io_min may be mutated when stacking devices, and this makes
> > it a poor candidate to hold the stripe size. Such an example (of when
> > io_min may change) would be when the io_min is less than the physical
> > block size.
> > 
> > Use chunk_sectors to hold the stripe size, which is more appropriate.
> > 
> > [0] https://lore.kernel.org/linux-block/888f3b1d-7817-4007-b3b3-1a2ea04df771@linux.ibm.com/T/#mecca17129f72811137d3c2f1e477634e77f06781
> > 
> > Signed-off-by: John Garry <john.g.garry@oracle.com>
> > ---
> >  block/blk-settings.c | 58 ++++++++++++++++++++++++++------------------
> >  1 file changed, 35 insertions(+), 23 deletions(-)
> > 
> > diff --git a/block/blk-settings.c b/block/blk-settings.c
> > index 761c6ccf5af7..3259cfac5d0d 100644
> > --- a/block/blk-settings.c
> > +++ b/block/blk-settings.c
> > @@ -597,41 +597,52 @@ static bool blk_stack_atomic_writes_boundary_head(struct queue_limits *t,
> >  	return true;
> >  }
> >  
> > -
> > -/* Check stacking of first bottom device */
> > -static bool blk_stack_atomic_writes_head(struct queue_limits *t,
> > -				struct queue_limits *b)
> > +static void blk_stack_atomic_writes_chunk_sectors(struct queue_limits *t)
> >  {
> > -	if (b->atomic_write_hw_boundary &&
> > -	    !blk_stack_atomic_writes_boundary_head(t, b))
> > -		return false;
> > +	unsigned int chunk_sectors = t->chunk_sectors, chunk_bytes;
> >  
> > -	if (t->io_min <= SECTOR_SIZE) {
> > -		/* No chunk sectors, so use bottom device values directly */
> > -		t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
> > -		t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min;
> > -		t->atomic_write_hw_max = b->atomic_write_hw_max;
> > -		return true;
> > -	}
> > +	if (!chunk_sectors)
> > +		return;
> > +
> > +	/*
> > +	 * If chunk sectors is so large that its value in bytes overflows
> > +	 * UINT_MAX, then just shift it down so it definitely will fit.
> > +	 * We don't support atomic writes of such a large size anyway.
> > +	 */
> > +	if ((unsigned long)chunk_sectors << SECTOR_SHIFT > UINT_MAX)
> > +		chunk_bytes = chunk_sectors;
> > +	else
> > +		chunk_bytes = chunk_sectors << SECTOR_SHIFT;

Why do we cast it to unsigned long? unsigned long is 32-bit on 32-bit 
machines, so the code will not detect the overflow in that case. We should 
cast it to unsigned long long (or uint64_t).

Mikulas

Re: [PATCH v4 6/6] block: use chunk_sectors when evaluating stacked atomic write limits

[John Garry]

On 08/07/2025 17:59, Mikulas Patocka wrote:
>>> +
>>> +	/*
>>> +	 * If chunk sectors is so large that its value in bytes overflows
>>> +	 * UINT_MAX, then just shift it down so it definitely will fit.
>>> +	 * We don't support atomic writes of such a large size anyway.
>>> +	 */
>>> +	if ((unsigned long)chunk_sectors << SECTOR_SHIFT > UINT_MAX)
>>> +		chunk_bytes = chunk_sectors;
>>> +	else
>>> +		chunk_bytes = chunk_sectors << SECTOR_SHIFT;
> Why do we cast it to unsigned long? unsigned long is 32-bit on 32-bit
> machines, so the code will not detect the overflow in that case. We should
> cast it to unsigned long long (or uint64_t).

Right, I said earlier that I would use an unsigned long long, but didn't 
do it that way, which was unintentional.

Anyway, I will change this code as suggested by Nilay.

Thanks,
John