check_signal: system call restart is done only once

Fujitsu: POSTK_TEMP_FIX_66
Refs: #1009
Change-Id: Ic0f04ac6b7f6c6bb01b55fb389bf9befd56b1dd9

arch/x86_64/kernel/cpu.c

@@ -847,9 +847,6 @@ void setup_x86_ap(void (*next_func)(void))
 }
 
 void arch_show_interrupt_context(const void *reg);
-void set_signal(int sig, void *regs, struct siginfo *info);
-void check_signal(unsigned long, void *, int);
-void check_sig_pending();
 extern void tlb_flush_handler(int vector);
 
 void __show_stack(uintptr_t *sp) {
@@ -877,7 +874,7 @@ void interrupt_exit(struct x86_user_context *regs)
 		cpu_enable_interrupt();
 		check_sig_pending();
 		check_need_resched();
-		check_signal(0, regs, 0);
+		check_signal(0, regs, -1);
 	}
 	else {
 		check_sig_pending();

arch/x86_64/kernel/syscall.c

@@ -36,8 +36,6 @@
 void terminate_mcexec(int, int);
 extern long do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact);
 long syscall(int num, ihk_mc_user_context_t *ctx);
-void set_signal(int sig, void *regs0, siginfo_t *info);
-void check_signal(unsigned long rc, void *regs0, int num);
 extern unsigned long do_fork(int, unsigned long, unsigned long, unsigned long,
 	unsigned long, unsigned long, unsigned long);
 extern int get_xsave_size();
@@ -261,7 +259,7 @@ SYSCALL_DECLARE(rt_sigreturn)
 		info.si_code = TRAP_TRACE;
 		set_signal(SIGTRAP, regs, &info);
 		check_need_resched();
-		check_signal(0, regs, 0);
+		check_signal(0, regs, -1);
 	}
 
 	if(ksigsp.fpregs && xsavesize){
@@ -286,7 +284,6 @@ SYSCALL_DECLARE(rt_sigreturn)
 }
 
 extern struct cpu_local_var *clv;
-extern unsigned long do_kill(struct thread *thread, int pid, int tid, int sig, struct siginfo *info, int ptracecont);
 extern void interrupt_syscall(struct thread *, int sig);
 extern void terminate(int, int);
 extern int num_processors;
@@ -643,11 +640,13 @@ arch_ptrace(long request, int pid, long addr, long data)
 static int
 isrestart(int num, unsigned long rc, int sig, int restart)
 {
-	if(sig == SIGKILL || sig == SIGSTOP)
+	if (sig == SIGKILL || sig == SIGSTOP)
 		return 0;
-	if(num == 0 || rc != -EINTR)
+	if (num < 0 || rc != -EINTR)
 		return 0;
-	switch(num){
+	if (sig == SIGCHLD)
+		return 1;
+	switch (num) {
 	    case __NR_pause:
 	    case __NR_rt_sigsuspend:
 	    case __NR_rt_sigtimedwait:
@@ -668,14 +667,12 @@ isrestart(int num, unsigned long rc, int sig, int restart)
 	    case __NR_io_getevents:
 		return 0;
 	}
-	if(sig == SIGCHLD)
+	if (restart)
-		return 1;
-	if(restart)
 		return 1;
 	return 0;
 }
 
-void
+int
 do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pending *pending, int num)
 {
 	struct x86_user_context *regs = regs0;
@@ -687,6 +684,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 	int	ptraceflag = 0;
 	struct mcs_rwlock_node_irqsave lock;
 	struct mcs_rwlock_node_irqsave mcs_rw_node;
+	int restart = 0;
 
 	for(w = pending->sigmask.__val[0], sig = 0; w; sig++, w >>= 1);
 	dkprintf("do_signal(): tid=%d, pid=%d, sig=%d\n", thread->tid, proc->pid, sig);
@@ -715,7 +713,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 	if(k->sa.sa_handler == SIG_IGN){
 		kfree(pending);
 		mcs_rwlock_writer_unlock(&thread->sigcommon->lock, &mcs_rw_node);
-		return;
+		goto out;
 	}
 	else if(k->sa.sa_handler){
 		unsigned long *usp; /* user stack */
@@ -765,9 +763,8 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 		memcpy(&ksigsp.sigstack, &thread->sigstack, sizeof(stack_t));
 		ksigsp.sigrc = rc;
 		ksigsp.num = num;
-		ksigsp.restart = isrestart(num, rc, sig, k->sa.sa_flags & SA_RESTART);
+		restart = isrestart(num, rc, sig, k->sa.sa_flags & SA_RESTART);
-		if(num != 0 && rc == -EINTR && sig == SIGCHLD)
+		ksigsp.restart = restart;
-			ksigsp.restart = 1;
 		if(xsavesize){
 			uint64_t xsave_mask = get_xsave_mask();
 			unsigned int low = (unsigned int)xsave_mask;
@@ -780,7 +777,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 				kfree(_kfpregs);
 				kprintf("do_signal,no space available\n");
 				terminate(0, sig);
-				return;
+				goto out;
 			}
 			kfpregs = (void *)((((unsigned long)_kfpregs) + 63) & ~63);
 			memset(kfpregs, '\0', xsavesize);
@@ -790,7 +787,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 				kfree(_kfpregs);
 				kprintf("do_signal,write_process_vm failed\n");
 				terminate(0, sig);
-				return;
+				goto out;
 			}
 			ksigsp.fpregs = (void *)fpregs;
 			kfree(_kfpregs);
@@ -802,7 +799,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			mcs_rwlock_writer_unlock(&thread->sigcommon->lock, &mcs_rw_node);
 			kprintf("do_signal,write_process_vm failed\n");
 			terminate(0, sig);
-			return;
+			goto out;
 		}
 
 		usp = (unsigned long *)sigsp;
@@ -832,7 +829,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			info.si_code = TRAP_TRACE;
 			set_signal(SIGTRAP, regs, &info);
 			check_need_resched();
-			check_signal(0, regs, 0);
+			check_signal(0, regs, -1);
 		}
 	}
 	else {
@@ -940,6 +937,8 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			break;
 		}
 	}
+out:
+	return restart;
 }
 
 static struct sig_pending *
@@ -1019,6 +1018,12 @@ void save_syscall_return_value(int num, unsigned long rc)
 	return;
 }
 
+/** \brief check arrived signals and processing
+ *
+ * @param rc    return value of syscall
+ * @param regs0 context
+ * @param num   syscall number (-1: Not called on exiting system call)
+ */
 void
 check_signal(unsigned long rc, void *regs0, int num)
 {
@@ -1059,7 +1064,9 @@ check_signal(unsigned long rc, void *regs0, int num)
 			goto out;
 		}
 
-		do_signal(rc, regs, thread, pending, num);
+		if (do_signal(rc, regs, thread, pending, num)) {
+			num = -1;
+		}
 	}
 
 out:
@@ -1139,7 +1146,7 @@ check_sig_pending_thread(struct thread *thread)
 }
 
 void
-check_sig_pending()
+check_sig_pending(void)
 {
 	struct thread *thread;
 	struct cpu_local_var *v;

kernel/include/process.h

@@ -839,6 +839,13 @@ void chain_thread(struct thread *);
 void proc_init(void);
 void set_timer(int runq_locked);
 struct sig_pending *hassigpending(struct thread *thread);
+extern int do_signal(unsigned long rc, void *regs0, struct thread *thread,
+		     struct sig_pending *pending, int num);
+extern void check_signal(unsigned long rc, void *regs0, int num);
+extern unsigned long do_kill(struct thread *thread, int pid, int tid, int sig,
+			     struct siginfo *info, int ptracecont);
+extern void set_signal(int sig, void *regs, struct siginfo *info);
+extern void check_sig_pending(void);
 
 void release_fp_regs(struct thread *proc);
 void save_fp_regs(struct thread *proc);

kernel/mem.c

@@ -960,8 +960,6 @@ static struct ihk_mc_interrupt_handler query_free_mem_handler = {
 	.priv = NULL,
 };
 
-void set_signal(int sig, void *regs, struct siginfo *info);
-void check_signal(unsigned long, void *, int);
 int gencore(struct thread *, void *, struct coretable **, int *);
 void freecore(struct coretable **);
 
@@ -1221,7 +1219,7 @@ out:
 	if(interrupt_from_user(regs)){
 		cpu_enable_interrupt();
 		check_need_resched();
-		check_signal(0, regs, 0);
+		check_signal(0, regs, -1);
 	}
 	set_cputime(interrupt_from_user(regs)? 0: 1);
 #ifdef PROFILE_ENABLE

kernel/process.c

@@ -82,7 +82,6 @@ extern void lapic_timer_disable();
 extern int num_processors;
 extern ihk_spinlock_t cpuid_head_lock;
 int ptrace_detach(int pid, int data);
-extern unsigned long do_kill(struct thread *, int pid, int tid, int sig, struct siginfo *info, int ptracecont);
 extern void procfs_create_thread(struct thread *);
 extern void procfs_delete_thread(struct thread *);
 

kernel/syscall.c

@@ -116,10 +116,7 @@ static ihk_spinlock_t tod_data_lock = SPIN_LOCK_UNLOCKED;
 static unsigned long uti_desc; /* Address of struct uti_desc object in syscall_intercept.c */
 static void calculate_time_from_tsc(struct timespec *ts);
 
-void check_signal(unsigned long, void *, int);
 void save_syscall_return_value(int num, unsigned long rc);
-void do_signal(long rc, void *regs, struct thread *thread, struct sig_pending *pending, int num);
-extern unsigned long do_kill(struct thread *thread, int pid, int tid, int sig, struct siginfo *info, int ptracecont);
 extern long alloc_debugreg(struct thread *thread);
 extern int num_processors;
 extern unsigned long ihk_mc_get_ns_per_tsc(void);
@@ -4204,7 +4201,7 @@ SYSCALL_DECLARE(rt_sigtimedwait)
 			list_del(&pending->list);
 			thread->sigmask.__val[0] = bset;
 			mcs_rwlock_writer_unlock(lock, &mcs_rw_node);
-			do_signal(-EINTR, NULL, thread, pending, 0);
+			do_signal(-EINTR, NULL, thread, pending, -1);
 			return -EINTR;
 		}
 		mcs_rwlock_writer_unlock(lock, &mcs_rw_node);
@@ -4317,7 +4314,7 @@ do_sigsuspend(struct thread *thread, const sigset_t *set)
 		list_del(&pending->list);
 		mcs_rwlock_writer_unlock(lock, &mcs_rw_node);
 		thread->sigmask.__val[0] = bset;
-		do_signal(-EINTR, NULL, thread, pending, 0);
+		do_signal(-EINTR, NULL, thread, pending, -1);
 		break;
 	}
 	return -EINTR;
@@ -9549,7 +9546,7 @@ long syscall(int num, ihk_mc_user_context_t *ctx)
 	if(cpu_local_var(current)->proc->status == PS_EXITED &&
 	   (num != __NR_exit && num != __NR_exit_group)){
 		save_syscall_return_value(num, -EINVAL);
-		check_signal(-EINVAL, NULL, 0);
+		check_signal(-EINVAL, NULL, -1);
 		set_cputime(0);
 		return -EINVAL;
 	}

test/issues/1009/C1009.patch

@@ -0,0 +1,49 @@
+diff --git a/arch/x86_64/kernel/syscall.c b/arch/x86_64/kernel/syscall.c
+index 4b2742b..a3173c9 100644
+--- a/arch/x86_64/kernel/syscall.c
++++ b/arch/x86_64/kernel/syscall.c
+@@ -1670,6 +1670,11 @@ long do_arch_prctl(unsigned long code, unsigned long address)
+ 			break;
+ 		case ARCH_SET_GS:
+ 			return -ENOTSUPP;
++		case 999: {
++			struct thread *thread = cpu_local_var(current);
++			thread->proc->dblsig = (int)address;
++			return 0;
++		}
+ 		default:
+ 			return -EINVAL;
+ 	}
+diff --git a/kernel/include/process.h b/kernel/include/process.h
+index 24acf1f..dd94469 100644
+--- a/kernel/include/process.h
++++ b/kernel/include/process.h
+@@ -580,6 +580,8 @@ struct process {
+ #endif // PROFILE_ENABLE
+ 	int nr_processes; /* For partitioned execution */
+ 	int process_rank; /* Rank in partition */
++
++	int dblsig;
+ };
+ 
+ /*
+diff --git a/kernel/syscall.c b/kernel/syscall.c
+index 15d4593..3d03fad 100644
+--- a/kernel/syscall.c
++++ b/kernel/syscall.c
+@@ -9632,6 +9632,15 @@ long syscall(int num, ihk_mc_user_context_t *ctx)
+ 
+ 	if (!list_empty(&thread->sigpending) ||
+ 	    !list_empty(&thread->sigcommon->sigpending)) {
++		if (!list_empty(&thread->sigcommon->sigpending) &&
++		    thread->proc->dblsig) {
++			kprintf("have a signal, waiting arrive more signal\n");
++			while (list_is_singular(
++					      &thread->sigcommon->sigpending)) {
++				schedule();
++			}
++			kprintf("have some signals\n");
++		}
+ 		check_signal(l, NULL, num);
+ 	}
+ 

test/issues/1009/C1009.sh

@@ -0,0 +1,131 @@
+#!/bin/sh
+BOOTPARAM="-c 1-7,17-23,9-15,25-31 -m 10G@0,10G@1"
+USELTP=1
+USEOSTEST=
+
+################################################################################
+BINDIR=
+SBINDIR=
+OSTESTDIR=
+LTPDIR=
+LTPBIN=
+MCEXEC=
+TESTMCK=
+
+if [ -f $HOME/mck_test_config ]; then
+	. $HOME/mck_test_config
+elif [ -f ../../../mck_test_config.sample ]; then
+	. ../../../mck_test_config.sample
+else
+	BIN=
+	SBIN=
+	OSTEST=
+	LTP=
+fi
+
+#-------------------------------------------------------------------------------
+if [ "x$BIN" = x ]; then
+	if [ -f ../../../config.h ]; then
+		str=`grep "^#define BINDIR " ../../../config.h | head -1 | sed 's/^#define BINDIR /BINDIR=/'`
+		eval $str
+	fi
+else
+	BINDIR="$BIN"
+fi
+
+if [ "x$SBIN" = x ]; then
+	if [ -f ../../../Makefile ]; then
+		str=`grep ^SBINDIR ../../../Makefile | head -1 | sed 's/ //g'`
+		eval $str
+	fi
+else
+	SBINDIR="$SBIN"
+fi
+
+if [ ! -x "$BINDIR/mcexec" ]; then
+	echo no mckernel found $BINDIR >&2
+	exit 1
+fi
+MCEXEC="$BINDIR/mcexec"
+
+#-------------------------------------------------------------------------------
+if [ "x$USELTP" != x ]; then
+	if [ "x$LTP" = x ]; then
+		if [ -f "$HOME/ltp/testcases/bin/fork01" ]; then
+			LTPDIR="$HOME/ltp"
+		fi
+	else
+		LTPDIR="$LTP"
+	fi
+
+	if [ ! -x "$LTPDIR/testcases/bin/fork01" ]; then
+		echo no LTP found $LTPDIR >&2
+		exit 1
+	fi
+	LTPBIN="$LTPDIR/testcases/bin"
+fi
+
+#-------------------------------------------------------------------------------
+if [ "x$USEOSTEST" != x ]; then
+	if [ "x$OSTEST" = x ]; then
+		if [ -f "$HOME/ostest/bin/test_mck" ]; then
+			OSTESTDIR="$HOME/ostest"
+		fi
+	else
+		OSTESTDIR="$OSTEST"
+	fi
+
+	if [ ! -x "$OSTESTDIR"/bin/test_mck ]; then
+		echo no ostest found $OSTESTDIR >&2
+		exit 1
+	fi
+	TESTMCK="$OSTESTDIR/bin/test_mck"
+fi
+
+#===============================================================================
+if [ ! -x "$SBINDIR/mcstop+release.sh" ]; then
+	echo mcstop+release: not found >&2
+	exit 1
+fi
+echo -n "mcstop+release.sh ... "
+sudo "$SBINDIR/mcstop+release.sh"
+echo "done"
+
+if lsmod | grep mcctrl > /dev/null 2>&1; then
+	echo mckernel shutdown failed >&2
+	exit 1
+fi
+
+if [ ! -x "$SBINDIR/mcreboot.sh" ]; then
+	echo mcreboot: not found >&2
+	exit 1
+fi
+echo -n "mcreboot.sh $BOOTPARAM ... "
+sudo "$SBINDIR/mcreboot.sh" $BOOTPARAM
+echo "done"
+
+if ! lsmod | grep mcctrl > /dev/null 2>&1; then
+	echo mckernel boot failed >&2
+	exit 1
+fi
+
+################################################################################
+$MCEXEC ./C1009T01
+
+if [ x$LTPDIR = x ]; then
+	echo no LTP found >&2
+	exit 1
+fi
+
+for i in kill01:02 kill12:03 pause02:04 sigaction01:05 ; do
+	tp=`echo $i|sed 's/:.*//'`
+	id=`echo $i|sed 's/.*://'`
+	$MCEXEC $LTPBIN/$tp 2>&1 | tee $tp.txt
+	ok=`grep TPASS $tp.txt | wc -l`
+	ng=`grep TFAIL $tp.txt | wc -l`
+	if [ $ng = 0 ]; then
+		echo "*** C1009T$id: $tp OK ($ok)"
+	else
+		echo "*** C1009T$id: $tp NG (ok=$ok ng=%ng)"
+	fi
+done

test/issues/1009/C1009.txt

@@ -0,0 +1,24 @@
+Script started on Mon Sep 10 15:12:28 2018
+bash-4.2$ make test
+gcc -g -Wall -o C1009T01 C1009T01.c
+sh ./C1009.sh
+SIGUSR2
+SIGUSR1
+read A OK
+read B OK
+*** C1009T01: OK
+kill01      1  TPASS  :  received expected signal 9
+*** C1009T02: kill01 OK (1)
+kill12      1  TPASS  :  Test passed
+*** C1009T03: kill12 OK (1)
+pause02     1  TPASS  :  pause was interrupted correctly
+*** C1009T04: pause02 OK (1)
+sigaction01    1  TPASS  :  SA_RESETHAND did not cause SA_SIGINFO to be cleared
+sigaction01    2  TPASS  :  SA_RESETHAND was masked when handler executed
+sigaction01    3  TPASS  :  sig has been masked because sa_mask originally contained sig
+sigaction01    4  TPASS  :  siginfo pointer non NULL
+*** C1009T05: sigaction01 OK (4)
+bash-4.2$ exit
+exit
+
+Script done on Mon Sep 10 15:12:54 2018

test/issues/1009/C1009T01.c

@@ -0,0 +1,107 @@
+#include <stdlib.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <sys/types.h>
+#include <errno.h>
+#include <signal.h>
+#include <asm/prctl.h>
+#include <sys/prctl.h>
+
+int arch_prctl(int code, unsigned long *addr);
+
+void
+sigusr(int sig)
+{
+	if (sig == SIGUSR1) {
+		printf("SIGUSR1\n");
+	}
+	else if (sig == SIGUSR2) {
+		printf("SIGUSR2\n");
+	}
+	else {
+		printf("other sig\n");
+	}
+	fflush(stdout);
+}
+
+int
+main(int argc, char **argv)
+{
+	struct sigaction act;
+	pid_t pid1 = 0;
+	pid_t pid2 = 0;
+	pid_t parent;
+	int pfd[2];
+	char ch;
+	int rc;
+	unsigned long val;
+
+	memset(&act, '\0', sizeof(act));
+	act.sa_handler = sigusr;
+	act.sa_flags = SA_RESTART;
+	sigaction(SIGUSR1, &act, NULL);
+	sigaction(SIGUSR2, &act, NULL);
+
+	pipe(pfd);
+
+	parent = getpid();
+	val = 1;
+	if (arch_prctl(999, (unsigned long *)val) == -1) {
+		fprintf(stderr, "C1009T01 WARN: no mckernel patch detected.\n");
+		exit(1);
+	}
+
+	if ((pid1 = fork())) {
+		pid2 = fork();
+	}
+
+	if (!pid1 || !pid2) {
+		int sig;
+
+		close(pfd[0]);
+		if (pid1)
+			sig = SIGUSR2;
+		else
+			sig = SIGUSR1;
+
+		sleep(1);
+		kill(parent, sig);
+		if (pid1) {
+			sleep(2);
+			ch = 'B';
+		}
+		else {
+			sleep(1);
+			ch = 'A';
+		}
+		write(pfd[1], &ch, 1);
+		close(pfd[1]);
+		exit(0);
+	}
+	rc = read(pfd[0], &ch, 1);
+	if (rc != 1) {
+		printf("C1009T01 NG: read error rc=%d errno=%d\n", rc, errno);
+		exit(1);
+	}
+	if (ch != 'A') {
+		printf("C1009T01 NG: read BAD DATA ch=%c\n", ch);
+		exit(1);
+	}
+	val = 0;
+	arch_prctl(999, (unsigned long *)val);
+	printf("read %c OK\n", ch);
+	rc = read(pfd[0], &ch, 1);
+	if (rc != 1) {
+		printf("C1009T01 NG: read error rc=%d errno=%d\n", rc, errno);
+		exit(1);
+	}
+	if (ch != 'B') {
+		printf("C1009T01 NG: read BAD DATA ch=%c\n", ch);
+		exit(1);
+	}
+
+	printf("read %c OK\n", ch);
+	printf("*** C1009T01: OK\n");
+	exit(0);
+}

test/issues/1009/Makefile

@@ -0,0 +1,13 @@
+CC = gcc
+TARGET = C1009T01
+
+all:: $(TARGET)
+
+C1009T01: C1009T01.c
+	$(CC) -g -Wall -o $@ $^
+
+test:: all
+	sh ./C1009.sh
+
+clean::
+	rm -f $(TARGET) *.o

test/issues/1009/README

@@ -0,0 +1,27 @@
+【Issue#1009 動作確認】
+□ テスト内容
+1. システムコール処理中にシグナルハンドラを呼び出す複数のシグナルを
+   受信し、当該システムコールを再処理するとき、当該システムコールが
+   1度しか処理されないことを確認する(指摘現象)。
+   尚、シグナルを同時に発行する状態を再現させるのが困難なため、本
+   テストは複数シグナルを待ち合わせるパッチを適用したカーネルで行う。
+   McKernel へのパッチファイルは C1009.patch である。
+C1009T01 シグナルを複数受信したとき、システムコールの再処理を1度だけ行う確認
+
+2. 変更が他のシグナル処理に影響しないことをLTPを用いて確認する。
+C1009T02 kill01: kill の基本機能の確認
+C1009T03 kill12: kill, wait, signal の組み合わせ確認
+C1009T04 pause02: pause の基本機能の確認
+C1009T05 sigaction01: sigaction の基本機能の確認
+
+□ 実行手順
+$ make test
+
+実行できない場合は、C1009.shの以下の行を適切に書き換えた後に実行。
+BIN=          mcexec が存在するパス
+SBIN=         mcreboot.sh が存在するパス
+LTP=          LTP が存在するパス
+
+□ 実行結果
+C1009.txt 参照。
+全ての項目が OK となっていることを確認。