Re: [PATCH] vfs: shave work on failed file open

[Mateusz Guzik]

On 9/26/23, Christian Brauner <brauner@xxxxxxxxxx> wrote:
>> +void fput_badopen(struct file *file)
>> +{
>> +	if (unlikely(file->f_mode & (FMODE_BACKING | FMODE_OPENED))) {
>> +		fput(file);
>> +		return;
>> +	}
>> +
>> +	if (WARN_ON(atomic_long_read(&file->f_count) != 1)) {
>> +		fput(file);
>> +		return;
>> +	}
>> +
>> +	/* zero out the ref count to appease possible asserts */
>> +	atomic_long_set(&file->f_count, 0);
>
> Afaict this could just be:
>
> if (WARN_ON_ONCE(atomic_long_cmpxchg(&file->f_count, 1, 0) != 1)) {
>

This would bring back one of the atomics, but I'm not going to die on this hill.

If you insist on this change I'm going to have tweak some comments and
bench again.

>> +	file_free_badopen(file);
>> +}
>> +EXPORT_SYMBOL(fput_badopen);
>
> Should definitely not be exported and only be available to core vfs
> code. So this should go into fs/internal.h.
>

Ok

-- 
Mateusz Guzik <mjguzik gmail.com>