vfork - glibc

Split from the chapter glibc: Process, Threads and Signals into the per-function layout; see the method page and the glibc ledger for coverage status.

Semantics – create a child that shares the parent’s address space and stack until it calls _exit(2) or an exec function, while the calling thread is suspended. The POSIX contract is almost pure undefined behavior: the child may only modify a pid_t holding the return value, must not return from (or longjmp out of) the function that called vfork, and must call no other function before _exit/exec. Violations corrupt the parent: the manual’s example of danger is signal handlers and, in multithreaded programs, a parent thread dropping credentials while the privileged child still shares its memory.

glibc layer – NPTL’s vfork does not run pthread_atfork handlers (LinuxThreads did). Signals sent to the parent arrive only after the child releases the shared memory. The value of vfork today is mostly historical: fork(2) is copy-on-write anyway; vfork avoids page-table copies and can exist on MMU-less systems – which is also why posix_spawn builds on CLONE_VM|CLONE_VFORK internally since glibc 2.24 rather than on user-visible vfork.

Edge cases – feature-test gated (_DEFAULT_SOURCE, or _XOPEN_SOURCE >= 500 with !_POSIX_C_SOURCE >= 200809L). An implementation where vfork == fork is fully compliant, so nothing can be relied on beyond the UB-safe subset: compute args, exec, _exit.

References – vfork(2) (standard, Linux, historic sections), glibc manual Creating a Process; syscall view: vfork.