[PATCH v3 10/11] kentry: Check that syscall entries and syscall exits match

[Andy Lutomirski]

If arch code calls the wrong kernel entry helpers, syscall entries and
exits can get out of sync.  Add a new field to task_struct to track the
syscall state and validate that it transitions correctly.

Signed-off-by: Andy Lutomirski <luto@xxxxxxxxxx>
---
 include/linux/sched.h |  4 ++++
 init/init_task.c      |  8 ++++++++
 kernel/entry/common.c | 24 +++++++++++++++++++++++-
 3 files changed, 35 insertions(+), 1 deletion(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 6e3a5eeec509..95d6d8686d98 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1368,6 +1368,10 @@ struct task_struct {
 	struct llist_head               kretprobe_instances;
 #endif
 
+#ifdef CONFIG_DEBUG_ENTRY
+	bool kentry_in_syscall;
+#endif
+
 	/*
 	 * New fields for task_struct should be added above here, so that
 	 * they are included in the randomized portion of task_struct.
diff --git a/init/init_task.c b/init/init_task.c
index 8a992d73e6fb..de4fdaac4e8b 100644
--- a/init/init_task.c
+++ b/init/init_task.c
@@ -212,6 +212,14 @@ struct task_struct init_task
 #ifdef CONFIG_SECCOMP
 	.seccomp	= { .filter_count = ATOMIC_INIT(0) },
 #endif
+#ifdef CONFIG_DEBUG_ENTRY
+	/*
+	 * The init task, and kernel threads in general, are considered
+	 * to be "in a syscall".  This way they can execve() and then exit
+	 * the supposed syscall that they were in to go to user mode.
+	 */
+	.kentry_in_syscall = true,
+#endif
 };
 EXPORT_SYMBOL(init_task);
 
diff --git a/kernel/entry/common.c b/kernel/entry/common.c
index f62934d761e3..9dea411de3b0 100644
--- a/kernel/entry/common.c
+++ b/kernel/entry/common.c
@@ -148,11 +148,21 @@ static long syscall_trace_enter(struct pt_regs *regs, long syscall,
 
 long kentry_syscall_begin(struct pt_regs *regs, long syscall)
 {
-	unsigned long work = READ_ONCE(current_thread_info()->syscall_work);
+	unsigned long work;
+
+	if (IS_ENABLED(CONFIG_DEBUG_ENTRY)) {
+		DEBUG_ENTRY_WARN_ONCE(
+			current->kentry_in_syscall,
+			"entering syscall %ld while already in a syscall",
+			syscall);
+		current->kentry_in_syscall = true;
+	}
 
 	CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
 	lockdep_assert_irqs_enabled();
 
+	work = READ_ONCE(current_thread_info()->syscall_work);
+
 	if (work & SYSCALL_WORK_ENTER)
 		syscall = syscall_trace_enter(regs, syscall, work);
 
@@ -163,11 +173,16 @@ long kentry_syscall_begin(struct pt_regs *regs, long syscall)
 static __always_inline void __exit_to_user_mode(void)
 {
 	instrumentation_begin();
+
 #ifdef CONFIG_DEBUG_ENTRY
 	DEBUG_ENTRY_WARN_ONCE(this_cpu_read(kentry_cpu_depth) != 1,
 			      "__exit_to_user_mode called at wrong kentry cpu depth (%u)",
 			      this_cpu_read(kentry_cpu_depth));
+
+	DEBUG_ENTRY_WARN_ONCE(current->kentry_in_syscall,
+			      "exiting to user mode while in syscall context");
 #endif
+
 	trace_hardirqs_on_prepare();
 	lockdep_hardirqs_on_prepare(CALLER_ADDR0);
 	instrumentation_end();
@@ -331,6 +346,13 @@ void kentry_syscall_end(struct pt_regs *regs)
 	 */
 	if (unlikely(work & SYSCALL_WORK_EXIT))
 		syscall_exit_work(regs, work);
+
+#ifdef CONFIG_DEBUG_ENTRY
+	DEBUG_ENTRY_WARN_ONCE(!current->kentry_in_syscall,
+			      "exiting syscall %lu without entering first", nr);
+
+	current->kentry_in_syscall = 0;
+#endif
 }
 
 noinstr void kentry_enter_from_user_mode(struct pt_regs *regs)
-- 
2.29.2