Shortcut cursig when called during sleep setup.

Add deep flag as function argument which is used by the sleep API but
nowhere else. Both calls to sleep_signal_check() should skip the ugly
bits of cursig().

In cursig() if deep once it is clear a signal will be taken keep the
signal on the thread siglist and return. sleep_signal_check() will then
return EINTR or ERESTART based on the signal context.  There is no reason
to do more in this special case. Especially stop/cont and the ptrace trap
must be skipped here. Once the call makes it to userret the signal will be
picked up again and handled in a safe location.

Stopping singals need some additional logic since we don't want to abort
the sleep just to stop a process. Since our SIGSTOP handling requires
a major rewrite this will be posponed until then.

OK mpi@

4 files changed, 21 insertions, 12 deletions

diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c
index c0576093843..98edd7378d5 100644
--- a/sys/kern/kern_sig.c
+++ b/sys/kern/kern_sig.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_sig.c,v 1.342 2024/10/15 13:49:26 claudio Exp $	*/
+/*	$OpenBSD: kern_sig.c,v 1.343 2024/10/17 09:11:35 claudio Exp $	*/
 /*	$NetBSD: kern_sig.c,v 1.54 1996/04/22 01:38:32 christos Exp $	*/
 
 /*
@@ -1302,14 +1302,14 @@ setsigctx(struct proc *p, int signum, struct sigctx *sctx)
  * they aren't returned.  This is checked after each entry to the system for
  * a syscall or trap. The normal call sequence is
  *
- *	while (signum = cursig(curproc, &ctx))
+ *	while (signum = cursig(curproc, &ctx, 0))
  *		postsig(signum, &ctx);
  *
  * Assumes that if the P_SINTR flag is set, we're holding both the
  * kernel and scheduler locks.
  */
 int
-cursig(struct proc *p, struct sigctx *sctx)
+cursig(struct proc *p, struct sigctx *sctx, int deep)
 {
 	struct process *pr = p->p_p;
 	int signum, mask, prop;
@@ -1345,6 +1345,15 @@ cursig(struct proc *p, struct sigctx *sctx)
 			continue;
 
 		/*
+		 * If cursig is called while going to sleep, abort now
+		 * and stop the sleep. When the call unwinded to userret
+		 * cursig is called again and there the signal can be
+		 * handled cleanly.
+		 */
+		if (deep)
+			goto keep;
+
+		/*
 		 * If traced, always stop, and stay stopped until released
 		 * by the debugger.  If our parent process is waiting for
 		 * us, don't hang as we could deadlock.
@@ -1915,7 +1924,7 @@ sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
 
 	dosigsuspend(p, p->p_sigmask &~ mask);
 	for (;;) {
-		si.si_signo = cursig(p, &ctx);
+		si.si_signo = cursig(p, &ctx, 0);
 		if (si.si_signo != 0) {
 			sigset_t smask = sigmask(si.si_signo);
 			if (smask & mask) {
@@ -2006,7 +2015,7 @@ userret(struct proc *p)
 	}
 
 	if (SIGPENDING(p) != 0) {
-		while ((signum = cursig(p, &ctx)) != 0)
+		while ((signum = cursig(p, &ctx, 0)) != 0)
 			postsig(p, signum, &ctx);
 	}
 
@@ -2020,7 +2029,7 @@ userret(struct proc *p)
 		p->p_sigmask = p->p_oldmask;
 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
 
-		while ((signum = cursig(p, &ctx)) != 0)
+		while ((signum = cursig(p, &ctx, 0)) != 0)
 			postsig(p, signum, &ctx);
 	}
 
diff --git a/sys/kern/kern_synch.c b/sys/kern/kern_synch.c
index cbbb473dfc0..dda123c1d9b 100644
--- a/sys/kern/kern_synch.c
+++ b/sys/kern/kern_synch.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_synch.c,v 1.206 2024/07/23 08:38:02 claudio Exp $	*/
+/*	$OpenBSD: kern_synch.c,v 1.207 2024/10/17 09:11:35 claudio Exp $	*/
 /*	$NetBSD: kern_synch.c,v 1.37 1996/04/22 01:38:37 christos Exp $	*/
 
 /*
@@ -458,7 +458,7 @@ sleep_signal_check(struct proc *p)
 
 	if ((err = single_thread_check(p, 1)) != 0)
 		return err;
-	if ((sig = cursig(p, &ctx)) != 0) {
+	if ((sig = cursig(p, &ctx, 1)) != 0) {
 		if (ctx.sig_intr)
 			return EINTR;
 		else
diff --git a/sys/sys/signalvar.h b/sys/sys/signalvar.h
index a8c6168f7ef..c207cff214a 100644
--- a/sys/sys/signalvar.h
+++ b/sys/sys/signalvar.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: signalvar.h,v 1.55 2024/10/09 08:39:49 claudio Exp $	*/
+/*	$OpenBSD: signalvar.h,v 1.56 2024/10/17 09:11:35 claudio Exp $	*/
 /*	$NetBSD: signalvar.h,v 1.17 1996/04/22 01:23:31 christos Exp $	*/
 
 /*
@@ -109,7 +109,7 @@ struct sigctx {
  */
 int	coredump(struct proc *p);
 void	execsigs(struct proc *p);
-int	cursig(struct proc *p, struct sigctx *);
+int	cursig(struct proc *p, struct sigctx *, int);
 void	pgsigio(struct sigio_ref *sir, int sig, int checkctty);
 void	pgsignal(struct pgrp *pgrp, int sig, int checkctty);
 void	psignal(struct proc *p, int sig);
diff --git a/sys/ufs/mfs/mfs_vfsops.c b/sys/ufs/mfs/mfs_vfsops.c
index ff6d46997f6..4ce1859b773 100644
--- a/sys/ufs/mfs/mfs_vfsops.c
+++ b/sys/ufs/mfs/mfs_vfsops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mfs_vfsops.c,v 1.62 2022/02/14 11:26:05 claudio Exp $	*/
+/*	$OpenBSD: mfs_vfsops.c,v 1.63 2024/10/17 09:11:35 claudio Exp $	*/
 /*	$NetBSD: mfs_vfsops.c,v 1.10 1996/02/09 22:31:28 christos Exp $	*/
 
 /*
@@ -189,7 +189,7 @@ mfs_start(struct mount *mp, int flags, struct proc *p)
 		 * EINTR/ERESTART.
 		 */
 		if (sleepreturn != 0) {
-			sig = cursig(p, &ctx);
+			sig = cursig(p, &ctx, 0);
 			if (vfs_busy(mp, VB_WRITE|VB_NOWAIT) ||
 			    dounmount(mp, (sig == SIGKILL) ? MNT_FORCE : 0, p))
 				atomic_clearbits_int(&p->p_siglist,