Message-Id: <resend_4d740841bac383742949e2fefb03982736595087.git.luto@kernel.org>
Date:	Fri, 24 Apr 2015 15:09:19 -0700
From:	Andy Lutomirski <luto@...nel.org>
To:	X86 ML <x86@...nel.org>
Cc:	Denys Vlasenko <dvlasenk@...hat.com>,
	"H. Peter Anvin" <hpa@...or.com>, Borislav Petkov <bp@...en8.de>,
	Denys Vlasenko <vda.linux@...glemail.com>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Brian Gerst <brgerst@...il.com>,
	Ingo Molnar <mingo@...nel.org>,
	Steven Rostedt <rostedt@...dmis.org>,
	Oleg Nesterov <oleg@...hat.com>,
	Frederic Weisbecker <fweisbec@...il.com>,
	Alexei Starovoitov <ast@...mgrid.com>,
	Will Drewry <wad@...omium.org>,
	Kees Cook <keescook@...omium.org>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Andy Lutomirski <luto@...nel.org>
Subject: [PATCH] x86, selftests: Add a test for the "sysret_ss_attrs" bug

On AMD CPUs, SYSRET can return with a valid SS descriptor with with
the hidden attributes set to an unusable state.  Make sure the kernel
doesn't let this happen.  This detects an as-yet-unfixed regression.

Note that the 64-bit version of this test fails on AMD CPUs on all
kernel versions, although the issue in the 64-bit case is much less
severe than in the 32-bit case.

Tests: e7d6eefaaa44 x86/vdso32/syscall.S: Do not load __USER32_DS to %ss
Reported-by: Brian Gerst <brgerst@...il.com>
Signed-off-by: Andy Lutomirski <luto@...nel.org>
---

[resend -- it seems like only some recipients got it last time]

We're still mulling over the fix, but we have a good test case now.

tools/testing/selftests/x86/Makefile          |   5 +-
 tools/testing/selftests/x86/run_x86_tests.sh  |   2 +
 tools/testing/selftests/x86/sysret_ss_attrs.c | 112 ++++++++++++++++++++++++++
 tools/testing/selftests/x86/thunks.S          |  67 +++++++++++++++
 4 files changed, 185 insertions(+), 1 deletion(-)
 mode change 100644 => 100755 tools/testing/selftests/x86/run_x86_tests.sh
 create mode 100644 tools/testing/selftests/x86/sysret_ss_attrs.c
 create mode 100644 tools/testing/selftests/x86/thunks.S

diff --git a/tools/testing/selftests/x86/Makefile b/tools/testing/selftests/x86/Makefile
index ddf63569df5a..9309097f58e8 100644
--- a/tools/testing/selftests/x86/Makefile
+++ b/tools/testing/selftests/x86/Makefile
@@ -1,6 +1,6 @@
 .PHONY: all all_32 all_64 check_build32 clean run_tests
 
-TARGETS_C_BOTHBITS := sigreturn single_step_syscall
+TARGETS_C_BOTHBITS := sigreturn single_step_syscall sysret_ss_attrs
 
 BINARIES_32 := $(TARGETS_C_BOTHBITS:%=%_32)
 BINARIES_64 := $(TARGETS_C_BOTHBITS:%=%_64)
@@ -46,3 +46,6 @@ check_build32:
 	  echo "  yum install glibc-devel.*i686";			\
 	  exit 1;							\
 	fi
+
+# Some tests have additional dependencies.
+sysret_ss_attrs_64: thunks.S
diff --git a/tools/testing/selftests/x86/run_x86_tests.sh b/tools/testing/selftests/x86/run_x86_tests.sh
old mode 100644
new mode 100755
index 3fc19b376812..d25034280dd5
--- a/tools/testing/selftests/x86/run_x86_tests.sh
+++ b/tools/testing/selftests/x86/run_x86_tests.sh
@@ -4,10 +4,12 @@
 # script here.
 ./sigreturn_32 || exit 1
 ./single_step_syscall_32 || exit 1
+./sysret_ss_attrs_32 || exit 1
 
 if [[ "$uname -p" -eq "x86_64" ]]; then
     ./sigreturn_64 || exit 1
     ./single_step_syscall_64 || exit 1
+    ./sysret_ss_attrs_64 || exit 1
 fi
 
 exit 0
diff --git a/tools/testing/selftests/x86/sysret_ss_attrs.c b/tools/testing/selftests/x86/sysret_ss_attrs.c
new file mode 100644
index 000000000000..ce42d5a64009
--- /dev/null
+++ b/tools/testing/selftests/x86/sysret_ss_attrs.c
@@ -0,0 +1,112 @@
+/*
+ * sysret_ss_attrs.c - test that syscalls return valid hidden SS attributes
+ * Copyright (c) 2015 Andrew Lutomirski
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * On AMD CPUs, SYSRET can return with a valid SS descriptor with with
+ * the hidden attributes set to an unusable state.  Make sure the kernel
+ * doesn't let this happen.
+ */
+
+#define _GNU_SOURCE
+
+#include <stdlib.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <string.h>
+#include <sys/mman.h>
+#include <err.h>
+#include <stddef.h>
+#include <stdbool.h>
+#include <pthread.h>
+
+static void *threadproc(void *ctx)
+{
+	/*
+	 * Do our best to cause sleeps on this CPU to exit the kernel and
+	 * re-enter with SS = 0.
+	 */
+	while (true)
+		;
+
+	return NULL;
+}
+
+#ifdef __x86_64__
+extern unsigned long call32_from_64(void *stack, void (*function)(void));
+
+asm (".pushsection .text\n\t"
+     ".code32\n\t"
+     "test_ss:\n\t"
+     "pushl $0\n\t"
+     "popl %eax\n\t"
+     "ret\n\t"
+     ".code64");
+extern void test_ss(void);
+#endif
+
+int main()
+{
+	/*
+	 * Start a busy-looping thread on the same CPU we're on.
+	 * For simplicity, just stick everything to CPU 0.  This will
+	 * fail in some containers, but that's probably okay.
+	 */
+	cpu_set_t cpuset;
+	CPU_ZERO(&cpuset);
+	CPU_SET(0, &cpuset);
+	if (sched_setaffinity(0, sizeof(cpuset), &cpuset) != 0)
+		printf("[WARN]\tsched_setaffinity failed\n");
+
+	pthread_t thread;
+	if (pthread_create(&thread, 0, threadproc, 0) != 0)
+		err(1, "pthread_create");
+
+#ifdef __x86_64__
+	unsigned char *stack32 = mmap(NULL, 4096, PROT_READ | PROT_WRITE,
+				      MAP_32BIT | MAP_ANONYMOUS | MAP_PRIVATE,
+				      -1, 0);
+	if (stack32 == MAP_FAILED)
+		err(1, "mmap");
+#endif
+
+	printf("[RUN]\tSyscalls followed by SS validation\n");
+
+	for (int i = 0; i < 1000; i++) {
+		/*
+		 * Go to sleep and return using sysret (if we're 64-bit
+		 * or we're 32-bit on AMD on a 64-bit kernel).  On AMD CPUs,
+		 * SYSRET doesn't fix up the cached SS descriptor, so the
+		 * kernel needs some kind of workaround to make sure that we
+		 * end the system call with a valid stack segment.  This
+		 * can be a confusing failure because the SS *selector*
+		 * is the same regardless.
+		 */
+		usleep(2);
+
+#ifdef __x86_64__
+		/*
+		 * On 32-bit, just doing a syscall through glibc is enough
+		 * to cause a crash if our cached SS descriptor is invalid.
+		 * On 64-bit, it's not, so try extra hard.
+		 */
+		call32_from_64(stack32 + 4088, test_ss);
+#endif
+	}
+
+	printf("[OK]\tWe survived\n");
+
+#ifdef __x86_64__
+	munmap(stack32, 4096);
+#endif
+
+	return 0;
+}
diff --git a/tools/testing/selftests/x86/thunks.S b/tools/testing/selftests/x86/thunks.S
new file mode 100644
index 000000000000..ce8a995bbb17
--- /dev/null
+++ b/tools/testing/selftests/x86/thunks.S
@@ -0,0 +1,67 @@
+/*
+ * thunks.S - assembly helpers for mixed-bitness code
+ * Copyright (c) 2015 Andrew Lutomirski
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * These are little helpers that make it easier to switch bitness on
+ * the fly.
+ */
+
+	.text
+
+	.global call32_from_64
+	.type call32_from_64, @function
+call32_from_64:
+	// rdi: stack to use
+	// esi: function to call
+
+	// Save registers
+	pushq %rbx
+	pushq %rbp
+	pushq %r12
+	pushq %r13
+	pushq %r14
+	pushq %r15
+	pushfq
+
+	// Switch stacks
+	mov %rsp,(%rdi)
+	mov %rdi,%rsp
+
+	// Switch to compatibility mode
+	pushq $0x23  /* USER32_CS */
+	pushq $1f
+	lretq
+
+1:
+	.code32
+	// Call the function
+	call *%esi
+	// Switch back to long mode
+	jmp $0x33,$1f
+	.code64
+
+1:
+	// Restore the stack
+	mov (%rsp),%rsp
+
+	// Restore registers
+	popfq
+	popq %r15
+	popq %r14
+	popq %r13
+	popq %r12
+	popq %rbp
+	popq %rbx
+
+	ret
+
+.size call32_from_64, .-call32_from_64
-- 
2.3.0

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists