open / close - glibc

Split from the chapter glibc: stdio and Formatted I/O into the per-function layout; see the method page and the glibc ledger for coverage status.

Family B – Low-Level Input/Output (libc wrappers)

These unistd.h functions are one-to-one wrappers over the syscalls fully documented in Descriptors and File I/O; entries here cover only the libc-side deltas. Three facts recur:

• Return translation. Syscall success is a nonnegative value; failures surface as -1 with errno set – the wrapper is where the kernel’s negative-error convention becomes the C convention. No mapping or masking happens beyond that.
• EINTR is not retried. glibc returns EINTR to the caller; the manual (Interrupted Primitives) documents that interrupted primitives fail with EINTR unless the handler was installed with SA_RESTART, and offers the TEMP_FAILURE_RETRY macro in unistd.h as the caller side retry loop. Nothing in the wrapper layer retries for you.
• Close-on-exec plumbing. Flagless wrappers (pipe, dup2) cannot set FD_CLOEXEC; glibc’s answer is the GNU flag-taking variants pipe2/dup3 (unistd.h gates them on _GNU_SOURCE), while open-style calls take O_CLOEXEC directly and fopen/popen expose the e mode character.

Signature – int open(const char *path, int flags, ... /* mode_t */); int close(int fd); – unistd.h/fcntl.h.

Libc side – variadic third argument for O_CREAT/O_TMPFILE; open64/openat/openat64 in the same ledger section. On arm64 glibc’s open() issues openat(AT_FDCWD, ...) because no open slot exists (see the open entry in Descriptors and File I/O). close returns 0/-1+errno; the kernel releases the descriptor even on EINTR-in-doubt paths, so retrying close is wrong (syscall chapter, close entry). Both are raw translation wrappers; no retries, no fallbacks.