how to change pre-dump strategy

[abcsl123]

Description
Hello everyone, I am a student with a strong interest in CRIU, and I’m currently hoping to make some modifications based on my own ideas. From what I understand, when CRIU continuously uses pre-dump, it only stores memory pages that have been dirtied since the last checkpoint. If I want to change this behavior, such as deciding whether to store dirty memory pages based on a specific strategy, which functions should I focus on? (It’s quite difficult for me to find them on my own.)

[avagin]

@abcsl123 I think you can start with __parasite_dump_pages_seized:
https://github.com/checkpoint-restore/criu/blob/criu-dev/criu/mem.c#L482C12-L482C40

[abcsl123]

@abcsl123 I think you can start with __parasite_dump_pages_seized: https://github.com/checkpoint-restore/criu/blob/criu-dev/criu/mem.c#L482C12-L482C40

Thank you for your reply. @avagin
After reading this function, my understanding is that during a series of pre-dump and dump commands, if there have been previous pre-dump commands, executing the current command will turn non-softdirty pages into holes, thereby reducing consumption.(

criu/criu/mem.c

Line 225 in dfb56ee

if (has_parent && page_in_parent(softdirty)) {

)
However, why doesn’t it work as I expected during the dump? If all pages still need to be written, how can pre-dump reduce the frozen time during the dump? I’m very confused.
First pre-dump

Second pre-dump

then dump