On 2021/03/11 12:36, Darrick J. Wong wrote:
> On Thu, Mar 11, 2021 at 12:22:30PM +0900, Damien Le Moal wrote:
>> zonefs updates the size of a sequential zone file inode only on
>> completion of direct writes. When executing asynchronous append writes
>> (with a file open with O_APPEND or using RWF_APPEND), the use of the
>> current inode size in generic_write_checks() to set an iocb offset thus
>> leads to unaligned write if an application issues an append write
>> operation with another write already being executed.
>
> Ah, I /had/ wondered if setting i_size to the zone size (instead of the
> write pointer) would have side effects...
In retrospect, the problem is obvious :)
But a hole in the test suite let this one slip for some time. That is fixed now.
[...]
>> +static ssize_t zonefs_write_checks(struct kiocb *iocb, struct iov_iter *from)
>> +{
>> + struct file *file = iocb->ki_filp;
>> + struct inode *inode = file_inode(file);
>> + struct zonefs_inode_info *zi = ZONEFS_I(inode);
>> + loff_t count;
>> +
>> + if (IS_SWAPFILE(inode))
>> + return -ETXTBSY;
>
> ...but can zonefs really do swap files now?
Conventional zone files could, I guess, but I have not tested that. Not entirely
sure about this as I am not familiar with the swap code. I think it would be
safer to disallow swapfile use with zonefs, similarly to what claim_swap() does
with zoned block devices. But I am not sure how to do that. Sequential zone
files definitely will not be able to handle swap.
--
Damien Le Moal
Western Digital Research
