In situations like, where the copy on write bit of a page is set and that page is shared by more than one process, the Kernel allocates new page and copies the content to the new page and the other processes retain their references to the old page. After copying the Kernel updates the page table entry with the new page number. Then Kernel decrements the reference count of the old pfdata table entry.

In cases like, where the copy on write bit is set and no processes are sharing the page, the Kernel allows the physical page to be reused by the processes. By doing so, it clears the copy on write bit and disassociates the page from its disk copy (if one exists), because other process may share the disk copy. Then it removes the pfdata table entry from the page-queue as the new copy of the virtual page is not on the swap device. It decrements the swap-use count for the page and if count drops to 0, frees the swap space.

 On a swap device and not in memory, On the free page list in the main memory, In an executable file, Marked "demand zero", Marked "demand fill".

Page stealer finds the page eligible for swapping and places the page number in the list of pages to be swapped.Kernel copies the page to a swap device when necessary and clears the valid bit in the page table entry, decrements the pfdata reference count, and places the pfdata table entry at the end of the free list if its reference count is 0.  

Kernel contains 4 data structures for Demand paging. They are,

 Page table entries,

 Disk block descriptors,

 Page frame data table (pfdata),

 Swap-use table.

 All processes in the main memory are asleep.

All 'ready-to-run' processes are swapped out.

 There is no space in the swap device for the new incoming process that are swapped out of the main memory.

There is no space in the main memory for the new incoming process.

The resident time of the processes in the swap device, the priority of the processes and the amount of time the processes had been swapped out.

The process's memory resident time, Priority of the process and The nice value.

The resident time of the processes in the swap device, the priority of the processes and the amount of time the processes had been swapped out.

Fork() is a system call to create a child process. When the parent process calls fork() system call, the child process is created and if there is short of memory then the child process is sent to the read-to-run state in the swap device, and return to the user state without swapping the parent process. When the memory will be available the child process will be swapped into the main memory.

All memory space occupied by the process, process's u-area, and Kernel stack are swapped out, theoretically.
Practically, if the process's u-area contains the Address Translation Tables for the process then Kernel implementations do not swap the u-area.