linux-kernel - [PATCH] selftests/x86/ldt_gdt_32: Work around a glibc sigaction bug

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <371a5620248568efaf31dd9d897af3775725d9b8.1490114317.git.luto@kernel.org>
Date:   Tue, 21 Mar 2017 09:39:33 -0700
From:   Andy Lutomirski <luto@...nel.org>
To:     x86@...nel.org
Cc:     linux-kernel@...r.kernel.org, Borislav Petkov <bp@...en8.de>,
        Andy Lutomirski <luto@...nel.org>, stable@...r.kernel.org
Subject: [PATCH] selftests/x86/ldt_gdt_32: Work around a glibc sigaction bug

i386 glibc is buggy and calls the sigaction syscall incorrectly.
This is asymptomatic for normal programs, but it blows up on
programs that do evil things with segmentation.  ldt_gdt an example
of such an evil program.

This doesn't appear to be a regression -- I think I just got lucky
with the uninitialized memory that glibc threw at the kernel when I
wrote the test.

This hackish fix manually issues sigaction(2) syscalls to undo the
damage.  Without the fix, ldt_gdt_32 segfaults; with the fix, it
passes for me.

See https://sourceware.org/bugzilla/show_bug.cgi?id=21269

Cc: stable@...r.kernel.org
Signed-off-by: Andy Lutomirski <luto@...nel.org>
---

I'll see about factoring out sethandler(), etc into a separate file
soon.  In the mean time, this at least makes the test pass.

 tools/testing/selftests/x86/ldt_gdt.c | 36 +++++++++++++++++++++++++++++++++++
 1 file changed, 36 insertions(+)

diff --git a/tools/testing/selftests/x86/ldt_gdt.c b/tools/testing/selftests/x86/ldt_gdt.c
index f6121612e769..18e6ae1f1bb6 100644
--- a/tools/testing/selftests/x86/ldt_gdt.c
+++ b/tools/testing/selftests/x86/ldt_gdt.c
@@ -409,6 +409,24 @@ static void *threadproc(void *ctx)
 	}
 }
 
+#ifdef __i386__
+
+#ifndef SA_RESTORE
+#define SA_RESTORER 0x04000000
+#endif
+
+/*
+ * The UAPI header calls this 'struct sigaction', which conflicts with
+ * glibc.  Sigh.
+ */
+struct fake_ksigaction {
+	void *handler;  /* the real type is nasty */
+        unsigned long sa_flags;
+        void (*sa_restorer)(void);
+	unsigned long sigset1, sigset2;
+};
+#endif
+
 static void sethandler(int sig, void (*handler)(int, siginfo_t *, void *),
 		       int flags)
 {
@@ -420,6 +438,24 @@ static void sethandler(int sig, void (*handler)(int, siginfo_t *, void *),
 	if (sigaction(sig, &sa, 0))
 		err(1, "sigaction");
 
+#ifdef __i386__
+	struct fake_ksigaction ksa;
+	if (syscall(SYS_rt_sigaction, sig, NULL, &ksa, 8) == 0) {
+		/*
+		 * glibc has a nasty bug: it sometimes writes garbage to
+		 * sa_restorer.  This interacts quite badly with anything
+		 * that fiddles with SS because it can trigger legacy
+		 * stack switching.  Patch it up.
+		 */
+		printf("%d asdf %lx %p\n", sig, ksa.sa_flags, ksa.sa_restorer);
+		if (!(ksa.sa_flags & SA_RESTORER) && ksa.sa_restorer) {
+			printf("asdffff\n");
+			ksa.sa_restorer = NULL;
+			if (syscall(SYS_rt_sigaction, sig, &ksa, NULL, 8) != 0)
+				err(1, "rt_sigaction");
+		}
+	}
+#endif
 }
 
 static jmp_buf jmpbuf;
-- 
2.9.3