Skip to content

Linux Programming C++ Interview Guide

Process vs Thread

Aspect Process Thread
Memory Separate address space Shared address space
Creation fork() heavy (~ms) pthread_create light (~μs)
Communication IPC (pipes, shmem, sockets) Shared memory (direct)
Crash isolation One process crash won't kill others One thread crash kills all

fork/exec Model

pid_t pid = fork();      // Duplicates process (COW pages)
if (pid == 0) {
    execvp(prog, args);  // Replace child's image with new program
} else {
    waitpid(pid, &status, 0);  // Parent waits for child
}

COW (Copy-on-Write): Pages shared until written → efficient fork.

Signals

  • Asynchronous notifications to a process (SIGTERM, SIGKILL, SIGSEGV)
  • sigaction() > signal() (portable, reliable)
  • Signal handlers: only async-signal-safe functions (write, not printf)
  • Block signals during critical sections with sigprocmask

→ src/systems/linux/signal_handling.cpp

Memory-Mapped I/O (mmap)

void* ptr = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
// File access without read/write syscalls page-fault driven

Uses: shared memory IPC, file I/O, huge page allocation, memory-mapped databases.

→ src/systems/linux/mmap_shared_memory.cpp

Futex (Fast Userspace Mutex)

  • FUTEX_WAIT: sleep if *addr == expected (atomic check + sleep)
  • FUTEX_WAKE: wake N waiters
  • Foundation of pthread_mutex, std::mutex on Linux
  • Fast path: atomic in user-space. Slow path: kernel syscall only on contention

File Descriptors

  • Integer handles to kernel objects (files, sockets, pipes, epoll, timers)
  • Inheritance across fork() (unless O_CLOEXEC)
  • Limit: ulimit -n (default 1024, increase for servers)
  • /proc/self/fd/ shows open FDs

/proc Filesystem

Path Info
/proc/[pid]/maps Virtual memory mappings
/proc/[pid]/status Process state, memory usage
/proc/[pid]/fd/ Open file descriptors
/proc/cpuinfo CPU topology, cache sizes
/proc/meminfo System memory stats

Performance Tools

Tool Use
strace Trace syscalls (find bottlenecks, debug failures)
perf stat CPU counters (cache misses, branch mispredictions)
perf record Profile with flame graphs
valgrind Memory leaks, use-after-free
ltrace Library call tracing

cgroups & Namespaces

  • cgroups: Resource limits (CPU, memory, I/O) for process groups → foundation of containers
  • Namespaces: Isolation (PID, network, mount, user) → foundation of containers
  • Together they enable Docker/LXC without full VM overhead