On 06/14, Peter Zijlstra wrote: > > On Tue, 2011-06-14 at 16:27 +0200, Oleg Nesterov wrote: > > On 06/14, Peter Zijlstra wrote: > > > > > > On Mon, 2011-06-13 at 19:00 +0200, Oleg Nesterov wrote: > > > > > > > > Also. This is called under down_read(mmap_sem), can't we race with > > > > access_process_vm() modifying the same memory? > > > > > > Shouldn't matter COW and similar things are serialized using the pte > > > lock. > > > > Yes, but afaics this doesn't matter. Suppose that write_opcode() is > > called when access_process_vm() does copy_to_user_page(). We can cow > > the page before memcpy() completes. > > access_process_vm() will end up doing a FOLL_WRITE itself when > copy_to_user_page() is called since write=1 in that case. > > At that point we have a COW-race, someone wins, but the other will then > return the same page. > > At this point further PTRACE pokes can indeed race with the memcpy in > write_opcode(). Currently it can't, write_opcode() does another cow. But that cow can, and this is the same, yes. > A possible fix would be to lock_page() around > copy_to_user_page() (its already done in set_page_dirty_lock(), so > pulling it out shouldn't matter much). Yes, or write_opcode() could take mmap_sem for writing as Srikar suggests. But do we really care? Whatever we do we can race with the other updates to this memory. Say, someone can write to vma->vm_file. Oleg. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>