5.9. Termination of a Process The final change in process state that relates to the operation of the virtual-memory system is exit; this system call terminates a process, as described in Chapter 4. The part of exit that is discussed here is the release of the virtual-memory resources of the process. There are two sets of virtual-memory resources that need to be freed: The user portions of the address space, both memory and swap space The user structure and kernel stack
The first set of resources is freed in exit. The second set of resources is freed in wait. The release of the second set of resources is delayed because the kernel stack must be used until the process relinquishes the processor for the final time. The first step freeing the user address space is identical to the one that occurs during exec to free the old address space. The free operation proceeds entry by entry through the list of vm_map_entry structures associated with the address space. The first step in freeing an entry is to traverse its list of shadow objects. If the entry is the last reference to a shadow object, then any memory or swap space that is associated with the object can be freed. In addition, the machine-dependent routines are called to unmap and free up any page table or data structures that are associated with the object. If the shadow object is still referenced by other vm_map_entry structures, its resources cannot be freed, but the kernel still needs to call the machine-dependent routines to unmap and free the resources associated with the current process mapping. Finally, if the underlying object referenced by the vm_map_entry is losing its last reference, then that object is a candidate for deallocation. If it is an object that will never have any chance of a future reuse (such as an anonymous object associated with a stack or uninitialized data area), then its resources are freed as though it were a shadow object. However, if the object is associated with a vnode (e.g., it maps a file such as an executable), the object will persist until the vnode is reused for another purpose. Until the vnode is reused, the object and its associated pages will be available for reuse by newly executing processes or by processes mapping in a file. With all its resources free, the exiting process finishes detaching itself from its process group and notifies its parent that it is done. The process has now become a zombie process one with no resources. Its parent will collect its exit status with a wait call. Because the process structure, user structure, and kernel stack are allocated using the zone allocator, they will normally be retained for future use by another process rather than being broken down and their memory pages reclaimed. Thus, there is nothing for the virtual-memory system to do when wait is called: All virtual-memory resources of a process are removed when exit is done. On wait, the system just returns the process status to the caller, releases the process structure, user structure, and kernel stack back to the zone allocator, and frees the space in which the resource-usage information was kept. |
