um: Track userspace children dying in SECCOMP mode

When in seccomp mode, we would hang forever on the futex if a child has
died unexpectedly. In contrast, ptrace mode will notice it and kill the
corresponding thread when it fails to run it.

Fix this issue using a new IRQ that is fired after a SIGCHLD and keeping
an (internal) list of all MMs. In the IRQ handler, find the affected MM
and set its PID to -1 as well as the futex variable to FUTEX_IN_KERN.

This, together with futex returning -EINTR after the signal is
sufficient to implement a race-free detection of a child dying.

Note that this also enables IRQ handling while starting a userspace
process. This should be safe and SECCOMP requires the IRQ in case the
process does not come up properly.

Signed-off-by: Benjamin Berg <[e-mail]>
Signed-off-by: Benjamin Berg <[e-mail]>
Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Johannes Berg <[e-mail]>

arch/um/include/asm/irq.h

···
 #define TELNETD_IRQ 		8
 #define XTERM_IRQ 		9
 #define RANDOM_IRQ 		10
+#define SIGCHLD_IRQ		11
 
 #ifdef CONFIG_UML_NET_VECTOR
 
-#define VECTOR_BASE_IRQ		(RANDOM_IRQ + 1)
+#define VECTOR_BASE_IRQ		(SIGCHLD_IRQ + 1)
 #define VECTOR_IRQ_SPACE	8
 
 #define UM_FIRST_DYN_IRQ (VECTOR_IRQ_SPACE + VECTOR_BASE_IRQ)
 
 #else
 
-#define UM_FIRST_DYN_IRQ (RANDOM_IRQ + 1)
+#define UM_FIRST_DYN_IRQ (SIGCHLD_IRQ + 1)
 
 #endif
 

arch/um/include/asm/mmu.h

···
 #ifndef __ARCH_UM_MMU_H
 #define __ARCH_UM_MMU_H
 
+#include "linux/types.h"
 #include <mm_id.h>
 
 typedef struct mm_context {
 	struct mm_id id;
+
+	struct list_head list;
 
 	/* Address range in need of a TLB sync */
 	unsigned long sync_tlb_range_from;

arch/um/include/shared/irq_user.h

···
 struct siginfo;
 extern void sigio_handler(int sig, struct siginfo *unused_si,
 			  struct uml_pt_regs *regs, void *mc);
+extern void sigchld_handler(int sig, struct siginfo *unused_si,
+			   struct uml_pt_regs *regs, void *mc);
 void sigio_run_timetravel_handlers(void);
 extern void free_irq_by_fd(int fd);
 extern void deactivate_fd(int fd, int irqnum);

arch/um/include/shared/os.h

···
 extern void report_enomem(void);
 
 /* process.c */
+pid_t os_reap_child(void);
 extern void os_alarm_process(int pid);
 extern void os_kill_process(int pid, int reap_child);
 extern void os_kill_ptraced_process(int pid, int reap_child);

arch/um/include/shared/skas/mm_id.h

···
 
 void __switch_mm(struct mm_id *mm_idp);
 
+void notify_mm_kill(int pid);
+
 #endif

arch/um/include/shared/skas/skas.h

···
 
 #include <sysdep/ptrace.h>
 
+extern int using_seccomp;
 extern int userspace_pid[];
 
 extern void new_thread_handler(void);

arch/um/kernel/irq.c

···
 	/* Initialize EPOLL Loop */
 	os_setup_epoll();
 }
+
+extern void sigchld_handler(int sig, struct siginfo *unused_si,
+			    struct uml_pt_regs *regs, void *mc)
+{
+	do_IRQ(SIGCHLD_IRQ, regs);
+}

arch/um/kernel/skas/mmu.c

···
 #include <linux/sched/signal.h>
 #include <linux/slab.h>
 
+#include <shared/irq_kern.h>
 #include <asm/pgalloc.h>
 #include <asm/sections.h>
 #include <asm/mmu_context.h>
···
 
 /* Ensure the stub_data struct covers the allocated area */
 static_assert(sizeof(struct stub_data) == STUB_DATA_PAGES * UM_KERN_PAGE_SIZE);
+
+spinlock_t mm_list_lock;
+struct list_head mm_list;
 
 int init_new_context(struct task_struct *task, struct mm_struct *mm)
 {
···
 
 	new_id->stack = stack;
 
-	block_signals_trace();
-	new_id->pid = start_userspace(stack);
-	unblock_signals_trace();
+	scoped_guard(spinlock_irqsave, &mm_list_lock) {
+		/* Insert into list, used for lookups when the child dies */
+		list_add(&mm->context.list, &mm_list);
+	}
 
+	new_id->pid = start_userspace(stack);
 	if (new_id->pid < 0) {
 		ret = new_id->pid;
 		goto out_free;
···
 	 * zero, resulting in a kill(0), which will result in the
 	 * whole UML suddenly dying.  Also, cover negative and
 	 * 1 cases, since they shouldn't happen either.
+	 *
+	 * Negative cases happen if the child died unexpectedly.
 	 */
-	if (mmu->id.pid < 2) {
+	if (mmu->id.pid >= 0 && mmu->id.pid < 2) {
 		printk(KERN_ERR "corrupt mm_context - pid = %d\n",
 		       mmu->id.pid);
 		return;
 	}
-	os_kill_ptraced_process(mmu->id.pid, 1);
+
+	if (mmu->id.pid > 0) {
+		os_kill_ptraced_process(mmu->id.pid, 1);
+		mmu->id.pid = -1;
+	}
 
 	free_pages(mmu->id.stack, ilog2(STUB_DATA_PAGES));
+
+	guard(spinlock_irqsave)(&mm_list_lock);
+
+	list_del(&mm->context.list);
 }
+
+static irqreturn_t mm_sigchld_irq(int irq, void* dev)
+{
+	struct mm_context *mm_context;
+	pid_t pid;
+
+	guard(spinlock)(&mm_list_lock);
+
+	while ((pid = os_reap_child()) > 0) {
+		/*
+		* A child died, check if we have an MM with the PID. This is
+		* only relevant in SECCOMP mode (as ptrace will fail anyway).
+		*
+		* See wait_stub_done_seccomp for more details.
+		*/
+		list_for_each_entry(mm_context, &mm_list, list) {
+			if (mm_context->id.pid == pid) {
+				struct stub_data *stub_data;
+				printk("Unexpectedly lost MM child! Affected tasks will segfault.");
+
+				/* Marks the MM as dead */
+				mm_context->id.pid = -1;
+
+				/*
+				 * NOTE: If SMP is implemented, a futex_wake
+				 * needs to be added here.
+				 */
+				stub_data = (void *)mm_context->id.stack;
+				stub_data->futex = FUTEX_IN_KERN;
+
+				/*
+				 * NOTE: Currently executing syscalls by
+				 * affected tasks may finish normally.
+				 */
+				break;
+			}
+		}
+	}
+
+	return IRQ_HANDLED;
+}
+
+static int __init init_child_tracking(void)
+{
+	int err;
+
+	spin_lock_init(&mm_list_lock);
+	INIT_LIST_HEAD(&mm_list);
+
+	err = request_irq(SIGCHLD_IRQ, mm_sigchld_irq, 0, "SIGCHLD", NULL);
+	if (err < 0)
+		panic("Failed to register SIGCHLD IRQ: %d", err);
+
+	return 0;
+}
+early_initcall(init_child_tracking)

arch/um/os-Linux/process.c

···
 #include <init.h>
 #include <longjmp.h>
 #include <os.h>
+#include <skas/skas.h>
 
 void os_alarm_process(int pid)
 {
+	if (pid <= 0)
+		return;
+
 	kill(pid, SIGALRM);
 }
 
 void os_kill_process(int pid, int reap_child)
 {
+	if (pid <= 0)
+		return;
+
+	/* Block signals until child is reaped */
+	block_signals();
+
 	kill(pid, SIGKILL);
 	if (reap_child)
 		CATCH_EINTR(waitpid(pid, NULL, __WALL));
+
+	unblock_signals();
 }
 
 /* Kill off a ptraced child by all means available.  kill it normally first,
···
 
 void os_kill_ptraced_process(int pid, int reap_child)
 {
+	if (pid <= 0)
+		return;
+
+	/* Block signals until child is reaped */
+	block_signals();
+
 	kill(pid, SIGKILL);
 	ptrace(PTRACE_KILL, pid);
 	ptrace(PTRACE_CONT, pid);
 	if (reap_child)
 		CATCH_EINTR(waitpid(pid, NULL, __WALL));
+
+	unblock_signals();
+}
+
+pid_t os_reap_child(void)
+{
+	int status;
+
+	/* Try to reap a child */
+	return waitpid(-1, &status, WNOHANG);
 }
 
 /* Don't use the glibc version, which caches the result in TLS. It misses some
···
 	set_handler(SIGBUS);
 	signal(SIGHUP, SIG_IGN);
 	set_handler(SIGIO);
+	/* We (currently) only use the child reaper IRQ in seccomp mode */
+	if (using_seccomp)
+		set_handler(SIGCHLD);
 	signal(SIGWINCH, SIG_IGN);
 }
 

arch/um/os-Linux/signal.c

···
 	[SIGBUS]	= relay_signal,
 	[SIGSEGV]	= segv_handler,
 	[SIGIO]		= sigio_handler,
+	[SIGCHLD]	= sigchld_handler,
 };
 
 static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
···
 	}
 
 	/* enable signals if sig isn't IRQ signal */
-	if ((sig != SIGIO) && (sig != SIGWINCH))
+	if ((sig != SIGIO) && (sig != SIGWINCH) && (sig != SIGCHLD))
 		unblock_signals_trace();
 
 	(*sig_info[sig])(sig, si, &r, mc);
···
 
 #define SIGALRM_BIT 1
 #define SIGALRM_MASK (1 << SIGALRM_BIT)
+
+#define SIGCHLD_BIT 2
+#define SIGCHLD_MASK (1 << SIGCHLD_BIT)
 
 int signals_enabled;
 #if IS_ENABLED(CONFIG_UML_TIME_TRAVEL_SUPPORT)
···
 			sigio_run_timetravel_handlers();
 		else
 			signals_pending |= SIGIO_MASK;
+		return;
+	}
+
+	if (!enabled && (sig == SIGCHLD)) {
+		signals_pending |= SIGCHLD_MASK;
 		return;
 	}
 
···
 
 	[SIGIO] = sig_handler,
 	[SIGWINCH] = sig_handler,
+	/* SIGCHLD is only actually registered in seccomp mode. */
+	[SIGCHLD] = sig_handler,
 	[SIGALRM] = timer_alarm_handler,
 
 	[SIGUSR1] = sigusr1_handler,
···
 		 */
 		if (save_pending & SIGIO_MASK)
 			sig_handler_common(SIGIO, NULL, NULL);
+
+		if (save_pending & SIGCHLD_MASK) {
+			struct uml_pt_regs regs = {};
+
+			sigchld_handler(SIGCHLD, NULL, &regs, NULL);
+		}
 
 		/* Do not reenter the handler */
 

arch/um/os-Linux/skas/process.c

···
 }
 __initcall(init_stub_exe_fd);
 
+int using_seccomp;
 int userspace_pid[NR_CPUS];
 
 /**