git: a12e901a5a65 - main - Add a knob to disable dequeueing SIGCHLD on waiting for live process

[Konstantin Belousov]

The branch main has been updated by kib:

URL: https://cgit.FreeBSD.org/src/commit/?id=a12e901a5a65417849c1ccf1e37b8d092fa438da

commit a12e901a5a65417849c1ccf1e37b8d092fa438da
Author:     Konstantin Belousov <kib at FreeBSD.org>
AuthorDate: 2021-06-05 18:24:35 +0000
Commit:     Konstantin Belousov <kib at FreeBSD.org>
CommitDate: 2021-06-15 23:00:19 +0000

    Add a knob to disable dequeueing SIGCHLD on waiting for live process
    
    It seems that Linux does not dequeue siginfo for SIGCHLD when wait*(2)
    reports status of the running process.  In particular, sigwaitinfo(2)
    and other signal querying syscalls can observe the siginfo after wait.
    
    FreeBSD dequeued siginfo from the beginning, so we cannot change the
    default ABI to be more compatible.  Still, add a knob to enable to
    change to the other behavior for debugging purposes.
    
    Reported by:    dchagin
    Reviewed by:    dchagin, markj
    Sponsored by:   The FreeBSD Foundation
    MFC after:      1 week
    Differential revision:  https://reviews.freebsd.org/D30675
---
 sys/kern/kern_exit.c | 14 +++++++++++---
 sys/sys/sysent.h     |  1 +
 2 files changed, 12 insertions(+), 3 deletions(-)

diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c
index cb5996982a3a..fcd8b39a8ee2 100644
--- a/sys/kern/kern_exit.c
+++ b/sys/kern/kern_exit.c
@@ -105,6 +105,11 @@ SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN,
     &kern_kill_on_dbg_exit, 0,
     "Kill ptraced processes when debugger exits");
 
+static bool kern_wait_dequeue_sigchld = 1;
+SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN,
+    &kern_wait_dequeue_sigchld, 0,
+    "Dequeue SIGCHLD on wait(2) for live process");
+
 struct proc *
 proc_realparent(struct proc *child)
 {
@@ -1207,9 +1212,12 @@ report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
 			p->p_flag &= ~P_CONTINUED;
 		else
 			p->p_flag |= P_WAITED;
-		PROC_LOCK(td->td_proc);
-		sigqueue_take(p->p_ksi);
-		PROC_UNLOCK(td->td_proc);
+		if (kern_wait_dequeue_sigchld &&
+		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
+			PROC_LOCK(td->td_proc);
+			sigqueue_take(p->p_ksi);
+			PROC_UNLOCK(td->td_proc);
+		}
 	}
 	sx_xunlock(&proctree_lock);
 	if (siginfo != NULL) {
diff --git a/sys/sys/sysent.h b/sys/sys/sysent.h
index 95e9dcb1a335..8b0903f7dcc0 100644
--- a/sys/sys/sysent.h
+++ b/sys/sys/sysent.h
@@ -162,6 +162,7 @@ struct sysentvec {
 #define	SV_ASLR		0x080000	/* ASLR allowed. */
 #define	SV_RNG_SEED_VER	0x100000	/* random(4) reseed generation. */
 #define	SV_SIG_DISCIGN	0x200000	/* Do not discard ignored signals */
+#define	SV_SIG_WAITNDQ	0x400000	/* Wait does not dequeue SIGCHLD */
 
 #define	SV_ABI_MASK	0xff
 #define	SV_PROC_FLAG(p, x)	((p)->p_sysent->sv_flags & (x))