Add initial support for POSIX O_CLOFORK

Author: ricardobranco777

sys/kern/kern_descrip.c

@@ -510,6 +510,11 @@ kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
 		break;
 
+	case F_DUPFD_CLOFORK:
+		tmp = arg;
+		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOFORK, fd, tmp);
+		break;
+
 	case F_DUP2FD:
 		tmp = arg;
 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
@@ -520,13 +525,19 @@ kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
 		break;
 
+	case F_DUP2FD_CLOFORK:
+		tmp = arg;
+		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOFORK, fd, tmp);
+		break;
+
 	case F_GETFD:
 		error = EBADF;
 		FILEDESC_SLOCK(fdp);
 		fde = fdeget_noref(fdp, fd);
 		if (fde != NULL) {
 			td->td_retval[0] =
-			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
+			    ((fde->fde_flags & UF_EXCLOSE) != 0 ? FD_CLOEXEC : 0) |
+			    ((fde->fde_flags & UF_FOCLOSE) != 0 ? FD_CLOFORK : 0);
 			error = 0;
 		}
 		FILEDESC_SUNLOCK(fdp);
@@ -537,8 +548,10 @@ kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
 		FILEDESC_XLOCK(fdp);
 		fde = fdeget_noref(fdp, fd);
 		if (fde != NULL) {
-			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
-			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
+			fde->fde_flags =
+			    (fde->fde_flags & ~(UF_EXCLOSE | UF_FOCLOSE)) |
+			    ((arg & FD_CLOEXEC) != 0 ? UF_EXCLOSE : 0) |
+			    ((arg & FD_CLOFORK) != 0 ? UF_FOCLOSE : 0);
 			error = 0;
 		}
 		FILEDESC_XUNLOCK(fdp);
@@ -938,7 +951,7 @@ kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
 	fdp = p->p_fd;
 	oioctls = NULL;
 
-	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
+	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC | FDDUP_FLAG_CLOFORK)) == 0);
 	MPASS(mode < FDDUP_LASTMODE);
 
 	AUDIT_ARG_FD(old);
@@ -963,8 +976,10 @@ kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
 		goto unlock;
 	if (mode == FDDUP_FIXED && old == new) {
 		td->td_retval[0] = new;
-		if (flags & FDDUP_FLAG_CLOEXEC)
+		if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
+		if ((flags & FDDUP_FLAG_CLOFORK) != 0)
+			fdp->fd_ofiles[new].fde_flags |= UF_FOCLOSE;
 		error = 0;
 		goto unlock;
 	}
@@ -1039,10 +1054,9 @@ kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
 	fde_copy(oldfde, newfde);
 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
 	    nioctls);
-	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
-		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
-	else
-		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
+	newfde->fde_flags = (oldfde->fde_flags & ~(UF_EXCLOSE | UF_FOCLOSE)) |
+	    ((flags & FDDUP_FLAG_CLOEXEC) != 0 ? UF_EXCLOSE : 0) |
+	    ((flags & FDDUP_FLAG_CLOFORK) != 0 ? UF_FOCLOSE : 0);
 #ifdef CAPABILITIES
 	seqc_write_end(&newfde->fde_seqc);
 #endif
@@ -2163,7 +2177,8 @@ _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
 	seqc_write_begin(&fde->fde_seqc);
 #endif
 	fde->fde_file = fp;
-	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
+	fde->fde_flags = ((flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0) |
+	    ((flags & O_CLOFORK) != 0 ? UF_FOCLOSE : 0);
 	if (fcaps != NULL)
 		filecaps_move(fcaps, &fde->fde_caps);
 	else
@@ -2423,6 +2438,7 @@ fdcopy(struct filedesc *fdp)
 	newfdp->fd_freefile = fdp->fd_freefile;
 	FILEDESC_FOREACH_FDE(fdp, i, ofde) {
 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
+		    ((ofde->fde_flags & UF_FOCLOSE) != 0) ||
 		    !fhold(ofde->fde_file)) {
 			if (newfdp->fd_freefile == fdp->fd_freefile)
 				newfdp->fd_freefile = i;

sys/kern/sys_pipe.c

@@ -548,7 +548,7 @@ sys_pipe2(struct thread *td, struct pipe2_args *uap)
 {
 	int error, fildes[2];
 
-	if (uap->flags & ~(O_CLOEXEC | O_NONBLOCK))
+	if ((uap->flags & ~(O_CLOEXEC | O_CLOFORK | O_NONBLOCK)) != 0)
 		return (EINVAL);
 	error = kern_pipe(td, fildes, uap->flags, NULL, NULL);
 	if (error)

sys/sys/fcntl.h

@@ -143,6 +143,10 @@ typedef	__pid_t		pid_t;
 #define	O_NAMEDATTR	0x04000000	/* NFSv4 named attributes */
 #endif
 
+#if __POSIX_VISIBLE >= 202405
+#define	O_CLOFORK	0x08000000
+#endif
+
 /*
  * XXX missing O_RSYNC.
  */
@@ -269,6 +273,13 @@ typedef	__pid_t		pid_t;
 #define	F_GET_SEALS	20
 #define	F_ISUNIONSTACK	21		/* Kludge for libc, don't use it. */
 #define	F_KINFO		22		/* Return kinfo_file for this fd */
+#endif	/* __BSD_VISIBLE */
+
+#if __POSIX_VISIBLE >= 202405
+#define	F_DUPFD_CLOFORK	23		/* Like F_DUPFD, but FD_CLOFORK is set */
+#endif
+#if __BSD_VISIBLE
+#define	F_DUP2FD_CLOFORK 24		/* Like F_DUP2FD, but FD_CLOFORK is set */
 
 /* Seals (F_ADD_SEALS, F_GET_SEALS). */
 #define	F_SEAL_SEAL	0x0001		/* Prevent adding sealings */
@@ -279,6 +290,9 @@ typedef	__pid_t		pid_t;
 
 /* file descriptor flags (F_GETFD, F_SETFD) */
 #define	FD_CLOEXEC	1		/* close-on-exec flag */
+#if __POSIX_VISIBLE >= 202405
+#define	FD_CLOFORK	2		/* close-on-fork flag */
+#endif
 
 /* record locking flags (F_GETLK, F_SETLK, F_SETLKW) */
 #define	F_RDLCK		1		/* shared or read lock */

sys/sys/filedesc.h

@@ -148,6 +148,7 @@ struct filedesc_to_leader {
  * Per-process open flags.
  */
 #define	UF_EXCLOSE	0x01		/* auto-close on exec */
+#define	UF_FOCLOSE	0x02		/* auto-close on fork */
 
 #ifdef _KERNEL
 
@@ -220,6 +221,7 @@ enum {
 
 /* Flags for kern_dup(). */
 #define	FDDUP_FLAG_CLOEXEC	0x1	/* Atomically set UF_EXCLOSE. */
+#define	FDDUP_FLAG_CLOFORK	0x2	/* Atomically set UF_FOCLOSE. */
 
 /* For backward compatibility. */
 #define	falloc(td, resultfp, resultfd, flags) \