src/wrappers.cpp (1)

404-407: Loop var type: prefer deduced or signed to match container element type.

vector<int> fds is iterated with unsigned int fd. Not wrong (FDs are non-negative), but it forces a cast and can confuse overload resolution (e.g., passing to functions expecting int). Use auto fd or int fd for clarity.

Apply:

-  for (unsigned int fd : fds) {
+  for (auto fd : fds) {  // fd is int

jalib/jfilesystem.cpp (1)

168-198: Avoid narrowing size_t to int for path lengths and indices.

strlen() returns size_t. Converting to int risks overflow on very long paths and mixes signed/unsigned in loop math. Use size_t (for length) with a signed index for the sentinel -1, or avoid manual scan via strrchr.

Apply:

-  int len = strlen(fname);
+  size_t len = strlen(fname);
@@
-  int lastSlash = -1;
-  for (int i = 0; i < len; i++) {
-    if (fname[i] == '/') {
-      lastSlash = i;
-    }
-  }
+  // Prefer libc scan to avoid manual index types
+  const char* slash = ::strrchr(fname, '/');
+  int lastSlash = slash ? static_cast<int>(slash - fname) : -1;

src/ckptserializer.cpp (5)

78-83: Initialize and document trivial SIGCHLD action.

Good encapsulation with set_trivial_sigchld_action. Consider SA_RESTART if you rely on restarting certain syscalls elsewhere, but current usage with sigsuspend() is fine.

---

92-108: sigsuspend waits for “a” SIGCHLD, not necessarily from our child.

If other children exist with pending SIGCHLD, sigsuspend() may return before our freshly forked child exits; waitpid(pid, 0, 0) could then block. Likely safe in ckpt thread, but please confirm invariant “no other children can deliver SIGCHLD here” or guard with a non-blocking loop:

-  _real_sigsuspend(&mask_all_but_sigchld);
+  do {
+    if (_real_waitpid(-1, NULL, WNOHANG) == pid) break;
+  } while (_real_sigsuspend(&mask_all_but_sigchld) == -1 && errno == EINTR);

---

109-143: double_fork(): restore handler in parent only; fine, but consider mask state in grandchild.

Parent restores SIGCHLD action; grandchild inherits blocked signal mask from ckpt thread. If compressor expects signals unmasked, consider unblocking in the execing path, or at least ensure gzip doesn’t rely on them. Otherwise, LGTM.

---

248-267: Compression pipe setup: consider O_CLOEXEC to reduce fd leaks.

When using pipe2, pass O_CLOEXEC to avoid descriptor leakage if future code adds more exec paths.

-    if (_real_pipe(pipe_fds) == -1) {
+    if (_real_syscall(SYS_pipe2, pipe_fds, O_CLOEXEC) == -1) {

---

289-345: Always using double_fork() for compressor: OK. Ensure LD_PRELOAD hygiene and fd discipline.

Exec path unsets LD_PRELOAD and dup2s stdin/stdout correctly. One nit: if fd <= STDERR_FILENO, closing conditionals skip it; safe. Consider fcntl(outfd, F_SETFD, FD_CLOEXEC) before exec as belt-and-suspenders; not required.

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro