5.7. Execution of a File The exec system call was described in Sections 2.4 and 3.1; it replaces the address space of a process with the contents of a new program obtained from an executable file. During an exec, the target executable image is validated and then the arguments and environment are copied from the current process image into a temporary area of pageable kernel virtual memory. To do an exec, the system must allocate resources to hold the new contents of the virtual address space, set up the mapping for this address space to reference the new image, and release the resources being used for the existing virtual memory. The first step is to reserve memory resources for the new executable image. The algorithm for the calculation of the amount of virtual address space that must be reserved was described in Section 5.6. For an executable that is not being debugged (and hence will not have its text space modified), a space reservation must be made for only the data and stack space of the new executable. Exec does this reservation without first releasing the currently assigned space, because the system must be able to continue running the old executable until it is sure that it will be able to run the new one. If the system released the current space and the memory reservation failed, the exec would be unable to return to the original process. Once the reservation is made, the address space and virtual-memory resources of the current process are then freed as though the process were exiting; this mechanism is described in Section 5.9. Now the process has only a user structure and kernel stack. The kernel now allocates a new vmspace structure and creates the list of four vm_map_entry structures: A copy-on-write, fill-from-file entry maps the text segment. A copy-on-write mapping is used, rather than a read-only one, to allow active text segments to have debugging breakpoints set without affecting other users of the binary. In FreeBSD, some legacy code in the kernel debugging interface disallows the setting of break points in binaries being used by more than one process. This legacy code prevents the use of the copy-on-write feature and requires that the text be mapped read-only. A private (copy-on-write), fill-from-file entry maps the initialized data segment. An anonymous zero-fill-on-demand entry maps the uninitialized data segment. An anonymous zero-fill-on-demand entry maps the stack segment.
No further operations are needed to create a new address space during an exec system call; the remainder of the work involves copying the arguments and environment out to the top of the new stack. Initial values are set for the registers: The program counter is set to the entry point, and the stack pointer is set to point to the argument vector. The new process image is then ready to run. | 
