Shared memory pages and fork

Question

If the parent is sharing some pages with another process and we fork the parent. From what I know the child copies the page tables and we set the pages as read-only and do Copy-On-Write. But this will create a copy of the shared memory page if we write to it which is wrong. How does the Linux kernel avoid this?

Accepted Answer

The kernel knows which memory pages were allocated using shared memory operations. When a child is forked, those pages are not marked Copy-on-Write, so they will remain shared among all the processes.

This is recorded in the `vm_area_struct` data structure, in the `vm_flags` member. One of the flags is `VM_SHARED`. mm/memory.c contains the following function that determines if a page should be converted to COW

```
static inline int is_cow_mapping(vm_flags_t flags)
{
        return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
}
```

If you want to see more about how this flag is set and used, go to Linux Cross Reference and search for VM_SHARED.

Shared memory pages and fork

Solution

Problem