1 files changed, 249 insertions, 92 deletions

diff --git a/arch/um/os-Linux/signal.c b/arch/um/os-Linux/signal.c
index a86d7cc2c2d8..327fb3c52fc7 100644
--- a/arch/um/os-Linux/signal.c
+++ b/arch/um/os-Linux/signal.c
@@ -1,58 +1,57 @@
+// SPDX-License-Identifier: GPL-2.0
 /*
  * Copyright (C) 2015 Anton Ivanov (aivanov@{brocade.com,kot-begemot.co.uk})
  * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
  * Copyright (C) 2004 PathScale, Inc
  * Copyright (C) 2004 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
- * Licensed under the GPL
  */
 
 #include <stdlib.h>
 #include <stdarg.h>
+#include <stdbool.h>
 #include <errno.h>
 #include <signal.h>
+#include <string.h>
 #include <strings.h>
 #include <as-layout.h>
 #include <kern_util.h>
 #include <os.h>
 #include <sysdep/mcontext.h>
 #include <um_malloc.h>
+#include <sys/ucontext.h>
+#include <timetravel.h>
+#include "internal.h"
 
-void (*sig_info[NSIG])(int, struct siginfo *, struct uml_pt_regs *) = {
+void (*sig_info[NSIG])(int, struct siginfo *, struct uml_pt_regs *, void *mc) = {
 	[SIGTRAP]	= relay_signal,
 	[SIGFPE]	= relay_signal,
 	[SIGILL]	= relay_signal,
 	[SIGWINCH]	= winch,
-	[SIGBUS]	= bus_handler,
+	[SIGBUS]	= relay_signal,
 	[SIGSEGV]	= segv_handler,
 	[SIGIO]		= sigio_handler,
-	[SIGALRM]	= timer_handler
+	[SIGCHLD]	= sigchld_handler,
 };
 
 static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
 {
-	struct uml_pt_regs *r;
+	struct uml_pt_regs r;
 	int save_errno = errno;
 
-	r = uml_kmalloc(sizeof(struct uml_pt_regs), UM_GFP_ATOMIC);
-	if (!r)
-		panic("out of memory");
-
-	r->is_user = 0;
+	r.is_user = 0;
 	if (sig == SIGSEGV) {
 		/* For segfaults, we want the data from the sigcontext. */
-		get_regs_from_mc(r, mc);
-		GET_FAULTINFO_FROM_MC(r->faultinfo, mc);
+		get_regs_from_mc(&r, mc);
+		GET_FAULTINFO_FROM_MC(r.faultinfo, mc);
 	}
 
 	/* enable signals if sig isn't IRQ signal */
-	if ((sig != SIGIO) && (sig != SIGWINCH) && (sig != SIGALRM))
-		unblock_signals();
+	if ((sig != SIGIO) && (sig != SIGWINCH) && (sig != SIGCHLD))
+		unblock_signals_trace();
 
-	(*sig_info[sig])(sig, si, r);
+	(*sig_info[sig])(sig, si, &r, mc);
 
 	errno = save_errno;
-
-	free(r);
 }
 
 /*
@@ -67,43 +66,71 @@ static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
 #define SIGALRM_BIT 1
 #define SIGALRM_MASK (1 << SIGALRM_BIT)
 
-static int signals_enabled;
-static unsigned int signals_pending;
-static unsigned int signals_active = 0;
+#define SIGCHLD_BIT 2
+#define SIGCHLD_MASK (1 << SIGCHLD_BIT)
+
+__thread int signals_enabled;
+#if IS_ENABLED(CONFIG_UML_TIME_TRAVEL_SUPPORT)
+static int signals_blocked, signals_blocked_pending;
+#endif
+static __thread unsigned int signals_pending;
+static __thread unsigned int signals_active;
 
-void sig_handler(int sig, struct siginfo *si, mcontext_t *mc)
+static void sig_handler(int sig, struct siginfo *si, mcontext_t *mc)
 {
-	int enabled;
+	int enabled = signals_enabled;
+
+#if IS_ENABLED(CONFIG_UML_TIME_TRAVEL_SUPPORT)
+	if ((signals_blocked ||
+	     __atomic_load_n(&signals_blocked_pending, __ATOMIC_SEQ_CST)) &&
+	    (sig == SIGIO)) {
+		/* increment so unblock will do another round */
+		__atomic_add_fetch(&signals_blocked_pending, 1,
+				   __ATOMIC_SEQ_CST);
+		return;
+	}
+#endif
 
-	enabled = signals_enabled;
 	if (!enabled && (sig == SIGIO)) {
-		signals_pending |= SIGIO_MASK;
+		/*
+		 * In TT_MODE_EXTERNAL, need to still call time-travel
+		 * handlers. This will mark signals_pending by itself
+		 * (only if necessary.)
+		 * Note we won't get here if signals are hard-blocked
+		 * (which is handled above), in that case the hard-
+		 * unblock will handle things.
+		 */
+		if (time_travel_mode == TT_MODE_EXTERNAL)
+			sigio_run_timetravel_handlers();
+		else
+			signals_pending |= SIGIO_MASK;
+		return;
+	}
+
+	if (!enabled && (sig == SIGCHLD)) {
+		signals_pending |= SIGCHLD_MASK;
 		return;
 	}
 
-	block_signals();
+	block_signals_trace();
 
 	sig_handler_common(sig, si, mc);
 
-	set_signals(enabled);
+	um_set_signals_trace(enabled);
 }
 
 static void timer_real_alarm_handler(mcontext_t *mc)
 {
-	struct uml_pt_regs *regs;
-
-	regs = uml_kmalloc(sizeof(struct uml_pt_regs), UM_GFP_ATOMIC);
-	if (!regs)
-		panic("out of memory");
+	struct uml_pt_regs regs;
 
 	if (mc != NULL)
-		get_regs_from_mc(regs, mc);
-	timer_handler(SIGALRM, NULL, regs);
-
-	free(regs);
+		get_regs_from_mc(&regs, mc);
+	else
+		memset(&regs, 0, sizeof(regs));
+	timer_handler(SIGALRM, NULL, &regs);
 }
 
-void timer_alarm_handler(int sig, struct siginfo *unused_si, mcontext_t *mc)
+static void timer_alarm_handler(int sig, struct siginfo *unused_si, mcontext_t *mc)
 {
 	int enabled;
 
@@ -113,7 +140,7 @@ void timer_alarm_handler(int sig, struct siginfo *unused_si, mcontext_t *mc)
 		return;
 	}
 
-	block_signals();
+	block_signals_trace();
 
 	signals_active |= SIGALRM_MASK;
 
@@ -121,7 +148,7 @@ void timer_alarm_handler(int sig, struct siginfo *unused_si, mcontext_t *mc)
 
 	signals_active &= ~SIGALRM_MASK;
 
-	set_signals(enabled);
+	um_set_signals_trace(enabled);
 }
 
 void deliver_alarm(void) {
@@ -133,18 +160,33 @@ void timer_set_signal_handler(void)
 	set_handler(SIGALRM);
 }
 
+int timer_alarm_pending(void)
+{
+	return !!(signals_pending & SIGALRM_MASK);
+}
+
 void set_sigstack(void *sig_stack, int size)
 {
 	stack_t stack = {
 		.ss_flags = 0,
 		.ss_sp = sig_stack,
-		.ss_size = size - sizeof(void *)
+		.ss_size = size
 	};
 
 	if (sigaltstack(&stack, NULL) != 0)
 		panic("enabling signal stack failed, errno = %d\n", errno);
 }
 
+static void sigusr1_handler(int sig, struct siginfo *unused_si, mcontext_t *mc)
+{
+	uml_pm_wake();
+}
+
+void register_pm_wake_signal(void)
+{
+	set_handler(SIGUSR1);
+}
+
 static void (*handlers[_NSIG])(int sig, struct siginfo *si, mcontext_t *mc) = {
 	[SIGSEGV] = sig_handler,
 	[SIGBUS] = sig_handler,
@@ -154,50 +196,19 @@ static void (*handlers[_NSIG])(int sig, struct siginfo *si, mcontext_t *mc) = {
 
 	[SIGIO] = sig_handler,
 	[SIGWINCH] = sig_handler,
-	[SIGALRM] = timer_alarm_handler
+	/* SIGCHLD is only actually registered in seccomp mode. */
+	[SIGCHLD] = sig_handler,
+	[SIGALRM] = timer_alarm_handler,
+
+	[SIGUSR1] = sigusr1_handler,
 };
 
 static void hard_handler(int sig, siginfo_t *si, void *p)
 {
-	struct ucontext *uc = p;
+	ucontext_t *uc = p;
 	mcontext_t *mc = &uc->uc_mcontext;
-	unsigned long pending = 1UL << sig;
 
-	do {
-		int nested, bail;
-
-		/*
-		 * pending comes back with one bit set for each
-		 * interrupt that arrived while setting up the stack,
-		 * plus a bit for this interrupt, plus the zero bit is
-		 * set if this is a nested interrupt.
-		 * If bail is true, then we interrupted another
-		 * handler setting up the stack.  In this case, we
-		 * have to return, and the upper handler will deal
-		 * with this interrupt.
-		 */
-		bail = to_irq_stack(&pending);
-		if (bail)
-			return;
-
-		nested = pending & 1;
-		pending &= ~1;
-
-		while ((sig = ffs(pending)) != 0){
-			sig--;
-			pending &= ~(1 << sig);
-			(*handlers[sig])(sig, (struct siginfo *)si, mc);
-		}
-
-		/*
-		 * Again, pending comes back with a mask of signals
-		 * that arrived while tearing down the stack.  If this
-		 * is non-zero, we just go back, set up the stack
-		 * again, and handle the new interrupts.
-		 */
-		if (!nested)
-			pending = from_irq_stack(nested);
-	} while (pending);
+	(*handlers[sig])(sig, (struct siginfo *)si, mc);
 }
 
 void set_handler(int sig)
@@ -231,6 +242,11 @@ void set_handler(int sig)
 		panic("sigprocmask failed - errno = %d\n", errno);
 }
 
+void send_sigio_to_self(void)
+{
+	kill(os_getpid(), SIGIO);
+}
+
 int change_sig(int signal, int on)
 {
 	sigset_t sigset;
@@ -243,9 +259,29 @@ int change_sig(int signal, int on)
 	return 0;
 }
 
-void block_signals(void)
+static inline void __block_signals(void)
 {
+	if (!signals_enabled)
+		return;
+
+	os_local_ipi_disable();
+	barrier();
 	signals_enabled = 0;
+}
+
+static inline void __unblock_signals(void)
+{
+	if (signals_enabled)
+		return;
+
+	signals_enabled = 1;
+	barrier();
+	os_local_ipi_enable();
+}
+
+void block_signals(void)
+{
+	__block_signals();
 	/*
 	 * This must return with signals disabled, so this barrier
 	 * ensures that writes are flushed out before the return.
@@ -262,6 +298,12 @@ void unblock_signals(void)
 	if (signals_enabled == 1)
 		return;
 
+	__unblock_signals();
+
+#if IS_ENABLED(CONFIG_UML_TIME_TRAVEL_SUPPORT)
+	deliver_time_travel_irqs();
+#endif
+
 	/*
 	 * We loop because the IRQ handler returns with interrupts off.  So,
 	 * interrupts may have arrived and we need to re-enable them and
@@ -271,12 +313,9 @@ void unblock_signals(void)
 		/*
 		 * Save and reset save_pending after enabling signals.  This
 		 * way, signals_pending won't be changed while we're reading it.
-		 */
-		signals_enabled = 1;
-
-		/*
+		 *
 		 * Setting signals_enabled and reading signals_pending must
-		 * happen in this order.
+		 * happen in this order, so have the barrier here.
 		 */
 		barrier();
 
@@ -289,10 +328,13 @@ void unblock_signals(void)
 		/*
 		 * We have pending interrupts, so disable signals, as the
 		 * handlers expect them off when they are called.  They will
-		 * be enabled again above.
+		 * be enabled again above. We need to trace this, as we're
+		 * expected to be enabling interrupts already, but any more
+		 * tracing that happens inside the handlers we call for the
+		 * pending signals will mess up the tracing state.
 		 */
-
-		signals_enabled = 0;
+		__block_signals();
+		um_trace_signals_off();
 
 		/*
 		 * Deal with SIGIO first because the alarm handler might
@@ -305,6 +347,12 @@ void unblock_signals(void)
 		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 */
 
 		if ((save_pending & SIGALRM_MASK) && (!(signals_active & SIGALRM_MASK)))
@@ -315,15 +363,18 @@ void unblock_signals(void)
 		if (!(signals_pending & SIGIO_MASK) && (signals_active & SIGALRM_MASK))
 			return;
 
+		/* Re-enable signals and trace that we're doing so. */
+		um_trace_signals_on();
+		__unblock_signals();
 	}
 }
 
-int get_signals(void)
+int um_get_signals(void)
 {
 	return signals_enabled;
 }
 
-int set_signals(int enable)
+int um_set_signals(int enable)
 {
 	int ret;
 	if (signals_enabled == enable)
@@ -337,10 +388,116 @@ int set_signals(int enable)
 	return ret;
 }
 
-int os_is_signal_stack(void)
+int um_set_signals_trace(int enable)
 {
-	stack_t ss;
-	sigaltstack(NULL, &ss);
+	int ret;
+	if (signals_enabled == enable)
+		return enable;
+
+	ret = signals_enabled;
+	if (enable)
+		unblock_signals_trace();
+	else
+		block_signals_trace();
+
+	return ret;
+}
+
+#if IS_ENABLED(CONFIG_UML_TIME_TRAVEL_SUPPORT)
+void mark_sigio_pending(void)
+{
+	/*
+	 * It would seem that this should be atomic so
+	 * it isn't a read-modify-write with a signal
+	 * that could happen in the middle, losing the
+	 * value set by the signal.
+	 *
+	 * However, this function is only called when in
+	 * time-travel=ext simulation mode, in which case
+	 * the only signal ever pending is SIGIO, which
+	 * is blocked while this can be called, and the
+	 * timer signal (SIGALRM) cannot happen.
+	 */
+	signals_pending |= SIGIO_MASK;
+}
+
+void block_signals_hard(void)
+{
+	signals_blocked++;
+	barrier();
+}
+
+void unblock_signals_hard(void)
+{
+	static bool unblocking;
+
+	if (!signals_blocked)
+		panic("unblocking signals while not blocked");
+
+	if (--signals_blocked)
+		return;
+	/*
+	 * Must be set to 0 before we check pending so the
+	 * SIGIO handler will run as normal unless we're still
+	 * going to process signals_blocked_pending.
+	 */
+	barrier();
+
+	/*
+	 * Note that block_signals_hard()/unblock_signals_hard() can be called
+	 * within the unblock_signals()/sigio_run_timetravel_handlers() below.
+	 * This would still be prone to race conditions since it's actually a
+	 * call _within_ e.g. vu_req_read_message(), where we observed this
+	 * issue, which loops. Thus, if the inner call handles the recorded
+	 * pending signals, we can get out of the inner call with the real
+	 * signal hander no longer blocked, and still have a race. Thus don't
+	 * handle unblocking in the inner call, if it happens, but only in
+	 * the outermost call - 'unblocking' serves as an ownership for the
+	 * signals_blocked_pending decrement.
+	 */
+	if (unblocking)
+		return;
+	unblocking = true;
+
+	while (__atomic_load_n(&signals_blocked_pending, __ATOMIC_SEQ_CST)) {
+		if (signals_enabled) {
+			/* signals are enabled so we can touch this */
+			signals_pending |= SIGIO_MASK;
+			/*
+			 * this is a bit inefficient, but that's
+			 * not really important
+			 */
+			block_signals();
+			unblock_signals();
+		} else {
+			/*
+			 * we need to run time-travel handlers even
+			 * if not enabled
+			 */
+			sigio_run_timetravel_handlers();
+		}
+
+		/*
+		 * The decrement of signals_blocked_pending must be atomic so
+		 * that the signal handler will either happen before or after
+		 * the decrement, not during a read-modify-write:
+		 *  - If it happens before, it can increment it and we'll
+		 *    decrement it and do another round in the loop.
+		 *  - If it happens after it'll see 0 for both signals_blocked
+		 *    and signals_blocked_pending and thus run the handler as
+		 *    usual (subject to signals_enabled, but that's unrelated.)
+		 *
+		 * Note that a call to unblock_signals_hard() within the calls
+		 * to unblock_signals() or sigio_run_timetravel_handlers() above
+		 * will do nothing due to the 'unblocking' state, so this cannot
+		 * underflow as the only one decrementing will be the outermost
+		 * one.
+		 */
+		if (__atomic_sub_fetch(&signals_blocked_pending, 1,
+				       __ATOMIC_SEQ_CST) < 0)
+			panic("signals_blocked_pending underflow");
+	}
 
-	return ss.ss_flags & SS_ONSTACK;
+	unblocking = false;
 }
+#endif