[PATCH v4 07/11] block: Add fops atomic write support

[John Garry]

On 25/02/2024 14:46, Ritesh Harjani (IBM) wrote:
> John Garry <john.g.garry at oracle.com> writes:
> 
>> Support atomic writes by submitting a single BIO with the REQ_ATOMIC set.
>>
>> It must be ensured that the atomic write adheres to its rules, like
>> naturally aligned offset, so call blkdev_dio_invalid() ->
>> blkdev_atomic_write_valid() [with renaming blkdev_dio_unaligned() to
>> blkdev_dio_invalid()] for this purpose.
>>
>> In blkdev_direct_IO(), if the nr_pages exceeds BIO_MAX_VECS, then we cannot
>> produce a single BIO, so error in this case.
> 
> BIO_MAX_VECS is 256. So around 1MB limit with 4k pagesize.
> Any mention of why this limit for now? Is it due to code complexity that
> we only support a single bio?

The reason is that lifting this limit adds extra complexity and I don't 
see any HW out there which supports a larger atomic write unit yet. And 
even if there was HW (which supports this larger size), is there a 
usecase for a larger atomic write unit?


Nilay reports awupf = 63 for his controller:

# lspci
0040:01:00.0 Non-Volatile memory controller: KIOXIA Corporation Device 
0025 (rev 01)

# nvme id-ctrl /dev/nvme0 -H
NVME Identify Controller:
vid       : 0x1e0f
ssvid     : 0x1014
sn        : Z130A00LTGZ8
mn        : 800GB NVMe Gen4 U.2 SSD
fr        : REV.C9S2
[...]
awun      : 65535
awupf     : 63
[...]


And SCSI device I know which supports atomic writes can only handle 32KB 
max.

> As I see it, you have still enabled req merging in block layer for
> atomic requests. So it can essentially submit bio chains to the device
> driver? So why not support this case for user to submit a req. larger
> than 1 MB?

Indeed, we could try to lift this limit and submit larger bios or chains 
of bios for a single atomic write from userspace, but do we need it now?

Please also remember that we are always limited by the request queue DMA 
capabilities also.

> 
>>
>> Finally set FMODE_CAN_ATOMIC_WRITE when the bdev can support atomic writes
>> and the associated file flag is for O_DIRECT.
>>
>> Signed-off-by: John Garry <john.g.garry at oracle.com>
>> ---
>>   block/fops.c | 31 ++++++++++++++++++++++++++++---
>>   1 file changed, 28 insertions(+), 3 deletions(-)
>>
>> diff --git a/block/fops.c b/block/fops.c
>> index 28382b4d097a..563189c2fc5a 100644
>> --- a/block/fops.c
>> +++ b/block/fops.c
>> @@ -34,13 +34,27 @@ static blk_opf_t dio_bio_write_op(struct kiocb *iocb)
>>   	return opf;
>>   }
>>   
>> -static bool blkdev_dio_unaligned(struct block_device *bdev, loff_t pos,
>> -			      struct iov_iter *iter)
>> +static bool blkdev_atomic_write_valid(struct block_device *bdev, loff_t pos,
>> +				      struct iov_iter *iter)
>>   {
>> +	struct request_queue *q = bdev_get_queue(bdev);
>> +	unsigned int min_bytes = queue_atomic_write_unit_min_bytes(q);
>> +	unsigned int max_bytes = queue_atomic_write_unit_max_bytes(q);
>> +
>> +	return atomic_write_valid(pos, iter, min_bytes, max_bytes);
> 
> generic_atomic_write_valid() would be better for this function. However,
> I have any commented about this in some previous

ok

> 
>> +}
>> +
>> +static bool blkdev_dio_invalid(struct block_device *bdev, loff_t pos,
>> +				struct iov_iter *iter, bool atomic_write)
> 
> bool "is_atomic" or "is_atomic_write" perhaps?
> we anyway know that we only support atomic writes and RWF_ATOMIC
> operation is made -EOPNOTSUPP for reads in kiocb_set_rw_flags().
> So we may as well make it "is_atomic" for bools.

ok

> 
>> +{
>> +	if (atomic_write && !blkdev_atomic_write_valid(bdev, pos, iter))
>> +		return true;
>> +
>>   	return pos & (bdev_logical_block_size(bdev) - 1) ||
>>   		!bdev_iter_is_aligned(bdev, iter);
>>   }
>>   
>> +
>>   #define DIO_INLINE_BIO_VECS 4
>>   
>>   static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb,
>> @@ -71,6 +85,8 @@ static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb,
>>   	}
>>   	bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT;
>>   	bio.bi_ioprio = iocb->ki_ioprio;
>> +	if (iocb->ki_flags & IOCB_ATOMIC)
>> +		bio.bi_opf |= REQ_ATOMIC;
>>   
>>   	ret = bio_iov_iter_get_pages(&bio, iter);
>>   	if (unlikely(ret))
>> @@ -341,6 +357,9 @@ static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
>>   		task_io_account_write(bio->bi_iter.bi_size);
>>   	}
>>   
>> +	if (iocb->ki_flags & IOCB_ATOMIC)
>> +		bio->bi_opf |= REQ_ATOMIC;
>> +
>>   	if (iocb->ki_flags & IOCB_NOWAIT)
>>   		bio->bi_opf |= REQ_NOWAIT;
>>   
>> @@ -357,13 +376,14 @@ static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
>>   static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
>>   {
>>   	struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
>> +	bool atomic_write = iocb->ki_flags & IOCB_ATOMIC;
> 
> ditto, bool is_atomic perhaps?

ok

> 
>>   	loff_t pos = iocb->ki_pos;
>>   	unsigned int nr_pages;
>>   
>>   	if (!iov_iter_count(iter))
>>   		return 0;
>>   
>> -	if (blkdev_dio_unaligned(bdev, pos, iter))
>> +	if (blkdev_dio_invalid(bdev, pos, iter, atomic_write))
>>   		return -EINVAL;
>>   
>>   	nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
>> @@ -371,6 +391,8 @@ static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
>>   		if (is_sync_kiocb(iocb))
>>   			return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
>>   		return __blkdev_direct_IO_async(iocb, iter, nr_pages);
>> +	} else if (atomic_write) {
>> +		return -EINVAL;
>>   	}
>>   	return __blkdev_direct_IO(iocb, iter, bio_max_segs(nr_pages));
>>   }
>> @@ -616,6 +638,9 @@ static int blkdev_open(struct inode *inode, struct file *filp)
>>   	if (bdev_nowait(handle->bdev))
>>   		filp->f_mode |= FMODE_NOWAIT;
>>   
>> +	if (bdev_can_atomic_write(handle->bdev) && filp->f_flags & O_DIRECT)
>> +		filp->f_mode |= FMODE_CAN_ATOMIC_WRITE;
>> +
>>   	filp->f_mapping = handle->bdev->bd_inode->i_mapping;
>>   	filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
>>   	filp->private_data = handle;
>> -- 
>> 2.31.1

Thanks,
John