Hello... > So in other words, when the child writes onto the physical page > COW'ed from the parent, the VM page handler will invalidate that > access and copy the parent's physical page frame into another > physical page frame and update the child's page table to the NEW PFN. > Is this OK? Correct! And IMHO if the "old" page is no longer referenced by another process/thread (IIRC, the indicator is "count" property of page descriptor), it is given back to the buddy/slab allocator and free to be assigned for other purposes > So the task ID of the parent "process" the same as its PID? I never read about task id before. My understanding so far is, pid is unique for all process. For NPTL based threading mechanisme, system call like getpid() returns same value for all thread on the same group, because it returns tgid (aka task group id), not the pid itself..and tgid is same for all thread in the same thread group. This conforms to POSIX standard that getpid() (or any system call that yields pid of a process) returns same value for all threads on the same group > And the task_struct of each process is stored on the hard disk, is > that right? Would I be correct in assuming that the kernel has a > pointer to where every process's process descriptor is located/stored > on the hard disk? How does the kernel know the PID of a process > without first having its PD? No, not on the hard disk....it is stored on RAM, precisely somewhere inside kernel memory region And yes, kernel keeps a list of all process in every state (sleep, running, stopped, and so on). This list is called "task list" and again it lives on RAM "How does the kernel know the PID of a process without first having its PD (process descriptor)?" I don't think it is possible.... you need to grab the task structure first, then you will found out the pid of a process. You can prove this by looking on the function find_task_by_pid(). There you will realize the "pid" param is actually used as a key to query the task list to find the related task_struct. In other word, it's not enough to know the pid of a process. Eventually every action toward a process structure needs to grab the task_struct.... > > 2. on COW scenario > > You mean "before" the pages are written onto, right? Yes...that's correct > Thank you again for the tremendous help, Tristan and Mulyadi. Thanks to you too. This reminds me to keep re-reading all those kernel book :)) regards Mulyadi -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
