Re: [PATCH v4 03/11] fs: Initial atomic write support

[John Garry <john.g.garry@oracle.com>]

On 19/02/2024 22:44, Dave Chinner wrote:
> On Mon, Feb 19, 2024 at 01:01:01PM +0000, John Garry wrote:
>> @@ -3523,4 +3535,26 @@ extern int vfs_fadvise(struct file *file, loff_t offset, loff_t len,
>>   extern int generic_fadvise(struct file *file, loff_t offset, loff_t len,
>>   			   int advice);
>>   
>> +static inline bool atomic_write_valid(loff_t pos, struct iov_iter *iter,
>> +			   unsigned int unit_min, unsigned int unit_max)
>> +{
>> +	size_t len = iov_iter_count(iter);
>> +
>> +	if (!iter_is_ubuf(iter))
>> +		return false;
>> +
>> +	if (len == unit_min || len == unit_max) {
>> +		/* ok if exactly min or max */
>> +	} else if (len < unit_min || len > unit_max) {
>> +		return false;
>> +	} else if (!is_power_of_2(len)) {
>> +		return false;
>> +	}
> This doesn't need if else if else if and it doesn't need to check
> for exact unit min/max matches. 

This is fastpath code, and I thought it quicker to just check if min/max 
first. Based on recent discussions, for FS support I expect typically 
len == unit_max.

But I can change to your simpler checking and later change to the 
current method if those FS assumptions hold true.

> The exact matches require the
> length to be a power of 2, so the checks are simply:
> 
> 	if (len < unit_min || len > unit_max)
> 		return false;
> 	if (!is_power_of_2(len))
> 		return false;
> 
>> +	if (pos & (len - 1))
>> +		return false;
> This has typing issues - 64 bit value, 32 bit mask. probably should
> use:
> 
> 	if (!IS_ALIGNED(pos, len))
> 		return false;

ok, good idea.

Thanks,
John