Re: [PATCH] random: convert to using fops->write_iter()

[Jens Axboe <axboe@kernel.dk>]

On 5/19/22 9:01 PM, Al Viro wrote:
> On Thu, May 19, 2022 at 05:43:15PM -0600, Jens Axboe wrote:
> 
>> -static int write_pool(const char __user *ubuf, size_t len)
>> +static size_t write_pool(struct iov_iter *iter)
>>  {
>>  	size_t block_len;
>>  	int ret = 0;
>>  
>> -	while (len) {
>> -		block_len = min(len, sizeof(block));
>> -		if (copy_from_user(block, ubuf, block_len)) {
>> -			ret = -EFAULT;
>> +	while (iov_iter_count(iter)) {
>> +		block_len = min(iov_iter_count(iter), sizeof(block));
>> +		if (!copy_from_iter(block, block_len, iter)) {
>> +			if (!ret)
>> +				ret = -EFAULT;
>>  			goto out;
>>  		}
> 
> Feed it a buffer with only 1 byte mapped, watch it'll pass to mix_pool_bytes().
> And see how much of 'block' has been used uninitialized...

I don't follow? Buffer with 1 byte, iter setup with 1 byte. We copy 1 byte,
and we pass 1 byte to mix_pool_bytes(). What am I missing?

> And why bother with that min thing, anyway?
> 
> 	ssize_t ret = 0;
> 
> 	while (iov_iter_count(iter)) {
> 	  	u8 block[BLAKE2S_BLOCK_SIZE];
> 		size_t copied = copy_from_iter(block, sizeof(block), iter);
> 		if (!copied) {
> 			if (!ret)
> 				ret = -EFAULT;
> 			break;
> 		}
> 		mix_pool_bytes(block, copied);
> 		ret += copied;
> 	}
> 	return ret;
> 
> and be done with that...

Agree, that does look better, the min() part could've been killed with
the conversion indeed.

>> @@ -1382,11 +1378,16 @@ static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
>>  			return -EINVAL;
>>  		if (get_user(size, p++))
>>  			return -EFAULT;
>> -		retval = write_pool((const char __user *)p, size);
>> +
>> +		iov.iov_base = p;
>> +		iov.iov_len = size;
>> +		iov_iter_init(&iter, WRITE, &iov, 1, size);
> 
> That'd be
> 		import_single_range(WRITE, p, size, &iov, &iter);

Yep that'd be a simpler equivalent.

-- 
Jens Axboe