[Devel] [RFC PATCH vz9 v6 21/62] dm-ploop: introduce per-md page locking

[Alexander Atanasov]

On 14.01.25 6:03, Pavel Tikhomirov wrote:
> On 12/6/24 05:55, Alexander Atanasov wrote:
>> @@ -411,7 +415,6 @@ static void ploop_apply_delta_mappings(struct 
>> ploop *ploop,
>>       if (!is_raw)
>>           d_md = ploop_md_first_entry(md_root);
>> -    write_lock_irq(&ploop->bat_rwlock);
>>       ploop_for_each_md_page(ploop, md, node) {
>>           bat_entries = md->kmpage;
>>           if (!is_raw)
>> @@ -455,7 +458,6 @@ static void ploop_apply_delta_mappings(struct 
>> ploop *ploop,
>>           if (!is_raw)
>>               d_md = ploop_md_next_entry(d_md);
>>       }
>> -    write_unlock_irq(&ploop->bat_rwlock);
>>   }
>>   int ploop_check_delta_length(struct ploop *ploop, struct file *file, 
>> loff_t *file_size)
> 
> This is not possible to remove this lock, as ploop->bat_entries rbtree 
> is protected by it. In other words ploop_for_each_md_page should be 
> protected by critical section against concurrent ploop_add_md_pages.
> 


No, it is possible.

- ploop_apply_delta_mappings is only used before the device is started - 
not possible to have concurent calls.

- ploop_add_md_pages in case of resize works on a different rbtree.



-- 
Regards,
Alexander Atanasov