feat(select): poll/select and fd_set

2026-09-05 21:15:07 -04:00
parent 5af4197ee9
commit ddcf8f39be
7 changed files with 671 additions and 0 deletions
@@ -0,0 +1,120 @@
+#ifndef VLIBC_SYS_SELECT_H
+#define VLIBC_SYS_SELECT_H
+
+/*
+ * vlibc — <sys/select.h>.
+ *
+ * Synchronous I/O multiplexing: wait for readiness on file-descriptor sets
+ * with select() or pselect(). Both are thin wrappers over the SYS_pselect6
+ * kernel ABI, (nfds, readfds, writefds, exceptfds, struct timespec *,
+ * { sigset_t *, size_t } *): pselect passes its timespec and optional signal
+ * mask straight through, select converts its struct timeval timeout to a
+ * timespec and passes a NULL sigset pair.
+ *
+ *   Level 1 (onlyposix): select, pselect, FD_* macros, fd_set, struct
+ *                        timeval.
+ *
+ * fd_set is the kernel's bitmap layout on x86_64: an array of unsigned long
+ * words, with descriptor d held in bit d % 64 of word d / 64. FD_SETSIZE
+ * caps the tracked descriptors at 1024 (16 words), so select()/pselect()
+ * must be called with nfds <= FD_SETSIZE and FD_SET() only ever receives
+ * descriptors below FD_SETSIZE. select() and pselect() report the number of
+ * ready descriptors, 0 on timeout, or -1 with errno set on error.
+ *
+ * The signal mask type is declared here under the shared guard; <signal.h>
+ * is its canonical POSIX home and will own the full sig* API.
+ */
+
+#include <vlibc/features.h>
+
+#include <sys/types.h>
+
+#include <time.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Maximum descriptors select()/pselect() track (fd_set's bit capacity). */
+#define FD_SETSIZE 1024
+
+/*
+ * A set of file descriptors, one bit per descriptor, stored in the kernel
+ * bitmap layout: an array of unsigned long words, bit d in word
+ * d / (8 * sizeof(unsigned long)). Descriptors 0..FD_SETSIZE-1 fit; no
+ * operation may touch a descriptor at or above FD_SETSIZE.
+ */
+typedef struct
+{
+    unsigned long fds_bits[FD_SETSIZE / (8 * sizeof(unsigned long))];
+} fd_set;
+
+/* The word holding bit d of an fd_set, and that bit's mask within the word. */
+#define VLIBC_FDS_WORD(d) ((d) / (8 * sizeof(unsigned long)))
+#define VLIBC_FDS_MASK(d) (1UL << ((d) % (8 * sizeof(unsigned long))))
+
+/* Clear every descriptor bit of set. */
+#define FD_ZERO(set)                                                                               \
+    do                                                                                             \
+    {                                                                                              \
+        size_t fd_zero_i;                                                                          \
+        for (fd_zero_i = 0; fd_zero_i < sizeof(fd_set) / sizeof(unsigned long); fd_zero_i++)       \
+        {                                                                                          \
+            (set)->fds_bits[fd_zero_i] = 0UL;                                                      \
+        }                                                                                          \
+    } while (0)
+
+/* Add descriptor d to set. */
+#define FD_SET(d, set) ((set)->fds_bits[VLIBC_FDS_WORD(d)] |= VLIBC_FDS_MASK(d))
+
+/* Remove descriptor d from set. */
+#define FD_CLR(d, set) ((set)->fds_bits[VLIBC_FDS_WORD(d)] &= ~VLIBC_FDS_MASK(d))
+
+/* Nonzero when descriptor d is a member of set. */
+#define FD_ISSET(d, set) ((set)->fds_bits[VLIBC_FDS_WORD(d)] & VLIBC_FDS_MASK(d))
+
+#ifndef VLIBC_SIGSET_T_DEFINED
+#define VLIBC_SIGSET_T_DEFINED
+/*
+ * Signal mask type: a single 64-bit word — the x86_64 Linux sigset_t (see
+ * setjmp.h). pselect() and ppoll() only ever forward a pointer to the
+ * kernel, which reads the word directly. <signal.h> is the canonical POSIX
+ * home for sigset_t and builds the sig* API on this same layout; the
+ * typedef is repeated in <poll.h> under this guard so the two headers stay
+ * consistent.
+ */
+typedef unsigned long sigset_t;
+#endif
+
+/* Elapsed time in seconds and microseconds (see select()'s timeout). */
+struct timeval
+{
+    time_t tv_sec;
+    suseconds_t tv_usec;
+};
+
+/*
+ * Wait for readiness on the descriptors marked in readfds, writefds and
+ * exceptfds (each may be NULL), up to nfds descriptors (the highest
+ * descriptor in any set plus one). timeout is an upper bound on the wait; a
+ * NULL timeout blocks indefinitely, { 0, 0 } never blocks. On return each
+ * non-NULL set holds only its ready descriptors. Return the number of ready
+ * descriptors across the sets, 0 on timeout, or -1 with errno set.
+ */
+int
+select(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, struct timeval *timeout);
+
+/*
+ * Like select(), but the timeout is a struct timespec and, when sigmask is
+ * not NULL, the given signal mask is atomically installed for the duration
+ * of the wait (the previous mask is restored before returning).
+ */
+int
+pselect(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds,
+        const struct timespec *timeout, const sigset_t *sigmask);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* VLIBC_SYS_SELECT_H */