Skip to contentSkip to conversation

signal: avoid unconditional siginfo copy in send_signal_locked()

PATCH Applied Oleg Nesterov 3 participants 4 messages Jul 4 – Jul 6

lore
Share & tools
lore
#0 lore raw

cover signal: avoid unconditional siginfo copy in send_signal_locked()

On top of

signal-avoid-shared-siginfo-namespace-rewrites.patch

in -mm tree.

To clarify, this (simple) change looks like a cleanup to me even if it slightly complicates send_signal_locked(). But cleanups are always subjective, so I won't argue if anyone thinks otherwise.

To remind, I think we should shift this namespace translation into __send_signal_locked() -> copy_siginfo() path, but this needs something like [PATCH v2 2/3] signal: turn the "bool force" arg of __send_signal_locked() into "int flags"
https://lore.kernel.org/all/[email protected]/

Oleg.

#1 lore raw

PATCH 1/1 signal: avoid unconditional siginfo copy in send_signal_locked()

send_signal_locked() unconditionally copies siginfo before the namespace translation to avoid corrupting a shared siginfo.

Not that I think this can actually hurt performance-wise, just it doesn't look clean to me; the copy is only needed in the unlikely case when the translation will actually change something.

Defer it to the two cases where si_pid/si_uid are rewritten, and while at it add #ifdef's just for completeness.

---
 kernel/signal.c | 23 +++++++++++++++--------
 1 file changed, 15 insertions(+), 8 deletions(-)
kernel/signal.c +15 -8
index 041498ff835e..0f509ee4f42d 100644
@@ -1181,7 +1181,7 @@ static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
 int send_signal_locked(int sig, struct kernel_siginfo *info,
 		       struct task_struct *t, enum pid_type type)
 {
-	struct kernel_siginfo rewritten;
+	struct kernel_siginfo __maybe_unused rewritten;
 	/* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
 	bool force = false;
 
@@ -1193,27 +1193,34 @@ int send_signal_locked(int sig, struct kernel_siginfo *info,
 		force = true;
 	} else if (has_si_pid_and_uid(info)) {
 		/* SIGKILL and SIGSTOP is special or has ids */
+#ifdef CONFIG_USER_NS
 		struct user_namespace *t_user_ns;
-
-		rewritten = *info;
-		info = &rewritten;
+		kuid_t uid;
 
 		rcu_read_lock();
 		t_user_ns = task_cred_xxx(t, user_ns);
 		if (current_user_ns() != t_user_ns) {
-			kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
-			info->si_uid = from_kuid_munged(t_user_ns, uid);
+			rewritten = *info;
+			info = &rewritten;
+			uid = make_kuid(current_user_ns(), info->si_uid);
+			rewritten.si_uid = from_kuid_munged(t_user_ns, uid);
 		}
 		rcu_read_unlock();
-
+#endif
 		/* A kernel generated signal? */
 		force = (info->si_code == SI_KERNEL);
 
+#ifdef CONFIG_PID_NS
 		/* From an ancestor pid namespace? */
 		if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
-			info->si_pid = 0;
+			if (info != &rewritten) {
+				rewritten = *info;
+				info = &rewritten;
+			}
+			rewritten.si_pid = 0;
 			force = true;
 		}
+#endif
 	}
 	return __send_signal_locked(sig, info, t, type, force);
 }
Signature · 1 line
2.52.0
Reviewed-by on v1 1/1
#2 lore raw
Quoted 61 lines · Oleg Nesterov

send_signal_locked() unconditionally copies siginfo before the namespace translation to avoid corrupting a shared siginfo.

Not that I think this can actually hurt performance-wise, just it doesn't look clean to me; the copy is only needed in the unlikely case when the translation will actually change something.

Defer it to the two cases where si_pid/si_uid are rewritten, and while at it add #ifdef's just for completeness.

---
kernel/signal.c | 23 +++++++++++++++--------

1 file changed, 15 insertions(+), 8 deletions(-)
kernel/signal.c +10 -8
index 041498ff835e..0f509ee4f42d 100644
@@ -1181,7 +1181,7 @@ static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
 int send_signal_locked(int sig, struct kernel_siginfo *info,
 		       struct task_struct *t, enum pid_type type)
 {
-	struct kernel_siginfo rewritten;
+	struct kernel_siginfo __maybe_unused rewritten;
 	/* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
 	bool force = false;
 
@@ -1193,27 +1193,34 @@ int send_signal_locked(int sig, struct kernel_siginfo *info,
 		force = true;
 	} else if (has_si_pid_and_uid(info)) {
 		/* SIGKILL and SIGSTOP is special or has ids */
+#ifdef CONFIG_USER_NS
 		struct user_namespace *t_user_ns;
-
-		rewritten = *info;
-		info = &rewritten;
+		kuid_t uid;
 
 		rcu_read_lock();
 		t_user_ns = task_cred_xxx(t, user_ns);
 		if (current_user_ns() != t_user_ns) {
-			kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
-			info->si_uid = from_kuid_munged(t_user_ns, uid);
+			rewritten = *info;
+			info = &rewritten;
+			uid = make_kuid(current_user_ns(), info->si_uid);
+			rewritten.si_uid = from_kuid_munged(t_user_ns, uid);
 		}
 		rcu_read_unlock();
-
+#endif
 		/* A kernel generated signal? */
 		force = (info->si_code == SI_KERNEL);
 
+#ifdef CONFIG_PID_NS
 		/* From an ancestor pid namespace? */
 		if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
-			info->si_pid = 0;
+			if (info != &rewritten) {

kernel/signal.c int send_signal_locked(int sig, struct kernel_siginfo *info, · line 1226

Quoted 10 lines
kernel/signal.c +5 -0
+				rewritten = *info;
+				info = &rewritten;
+			}
+			rewritten.si_pid = 0;
 			force = true;
 		}
+#endif
 	}
 	return __send_signal_locked(sig, info, t, type, force);
 }

hi oleg,

thanks for the patch!

Thanks!

Reviewed-by on v1 1/1
#3 lore raw
In reply to Oleg Nesterov
Quoted 7 lines · Oleg Nesterov

send_signal_locked() unconditionally copies siginfo before the namespace translation to avoid corrupting a shared siginfo.

Not that I think this can actually hurt performance-wise, just it doesn't look clean to me; the copy is only needed in the unlikely case when the translation will actually change something.

Quoted 5 lines

Defer it to the two cases where si_pid/si_uid are rewritten, and while at it add #ifdef's just for completeness.

---