Re: [PATCH 9/9] blkdev: Fix up AIO+DIO+O_SYNC to do the sync part correctly

[Jeff Moyer <jmoyer@redhat.com>]

Jan Kara <jack@suse.cz> writes:

> On Mon 19-11-12 23:51:15, Darrick J. Wong wrote:
>> When performing O_SYNC+AIO+DIO writes to block devices, use the DIO_SYNC_WRITES
>> flag so that flushes are issued /after/ the write completes, not before.
>> 
>> Note, however, that for block devices, the DIO setup code ensures that a flush
>> wq is attached to the superblock of the bdevfs filesystem, not the filesystem
>> that the device node happens to reside in.  This means that unlike regular
>> files, iocb->ki_filp->f_mapping->host->i_sb != inode->i_sb.  Therefore, adjust
>> Jeff's earlier patch to keep the pointer use consistent and avoid a NULL deref.
>> 
>> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
>> ---
>>  fs/block_dev.c |    5 +++--
>>  fs/direct-io.c |    3 ++-
>>  2 files changed, 5 insertions(+), 3 deletions(-)
>> 
>> 
>> diff --git a/fs/block_dev.c b/fs/block_dev.c
>> index 1a1e5e3..05ff33a 100644
>> --- a/fs/block_dev.c
>> +++ b/fs/block_dev.c
>> @@ -235,7 +235,8 @@ blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
>>  	struct inode *inode = file->f_mapping->host;
>>  
>>  	return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
>> -				    nr_segs, blkdev_get_blocks, NULL, NULL, 0);
>> +				    nr_segs, blkdev_get_blocks, NULL, NULL,
>> +				    DIO_SYNC_WRITES);
>>  }
>>  
>>  int __sync_blockdev(struct block_device *bdev, int wait)
>> @@ -1631,7 +1632,7 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
>>  	percpu_down_read(&bdev->bd_block_size_semaphore);
>>  
>>  	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
>> -	if (ret > 0 || ret == -EIOCBQUEUED) {
>> +	if (ret > 0) {
>>  		ssize_t err;
>>  
>>  		err = generic_write_sync(file, pos, ret);
>> diff --git a/fs/direct-io.c b/fs/direct-io.c
>> index b7391d4..c626c43 100644
>> --- a/fs/direct-io.c
>> +++ b/fs/direct-io.c
>> @@ -258,7 +258,8 @@ void generic_dio_end_io(struct kiocb *iocb, loff_t offset, ssize_t bytes,
>>  		work->ret = ret;
>>  		work->offset = offset;
>>  		work->len = bytes;
>> -		queue_work(inode->i_sb->s_dio_flush_wq, &work->work);
>> +		queue_work(iocb->ki_filp->f_mapping->host->i_sb->s_dio_flush_wq,
>> +			   &work->work);
>   This should be folded into the original patch introducing the
> s_dio_flush_wq. And please add a comment before this line saying that block
> devices need a dereference exactly like this... Otherwise the patch looks
> good so you can add:
>   Reviewed-by: Jan Kara <jack@suse.cz>

When you say, "This," do you mean that one change, or the whole patch?

Cheers,
Jeff