linux-kernel - Re: [5.2 REGRESSION] Generic vDSO breaks seccomp-enabled userspace on i386

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

Message-ID: <alpine.DEB.2.21.1907272153090.1791@nanos.tec.linutronix.de>
Date:   Sat, 27 Jul 2019 23:52:32 +0200 (CEST)
From:   Thomas Gleixner <tglx@...utronix.de>
To:     Andy Lutomirski <luto@...nel.org>
cc:     Sean Christopherson <sean.j.christopherson@...el.com>,
        Kees Cook <keescook@...omium.org>,
        Vincenzo Frascino <vincenzo.frascino@....com>,
        X86 ML <x86@...nel.org>, LKML <linux-kernel@...r.kernel.org>,
        Paul Bolle <pebolle@...cali.nl>, Arnd Bergmann <arnd@...db.de>
Subject: Re: [5.2 REGRESSION] Generic vDSO breaks seccomp-enabled userspace
 on i386

On Sat, 27 Jul 2019, Thomas Gleixner wrote:
> On Sat, 27 Jul 2019, Andy Lutomirski wrote:
> > 
> > I think it's getting quite late to start inventing new seccomp
> > features to fix this.  I think the right solution for 5.3 is to change
> > the 32-bit vdso fallback to use the old clock_gettime, i.e.
> > clock_gettime32.  This is obviously not an acceptable long-term
> > solution.
> 
> Sigh. I'll have a look....

Completely untested patch below.

For the record: I have to say that I hate it.

Just to be clear. Once a clever seccomp admin decides to enforce Y2038
compliance by filtering out the legacy syscalls this will force glibc into
the syscall slowpath directly because __vdso_clock_gettime64() is gone.

So this needs a proper secccomp solution soener than later.

The fallback change to the legacy syscall is on purpose conditional on
CONFIG_SECCOMP so those people who care can get access to
__vdso_clock_gettime64() nevertheless. 

Thanks,

	tglx

8<-----------------
--- a/arch/x86/entry/vdso/vdso32/vdso32.lds.S
+++ b/arch/x86/entry/vdso/vdso32/vdso32.lds.S
@@ -27,7 +27,9 @@ VERSION
 		__vdso_gettimeofday;
 		__vdso_time;
 		__vdso_clock_getres;
+#ifndef CONFIG_SECCOMP
 		__vdso_clock_gettime64;
+#endif
 	};
 
 	LINUX_2.5 {
--- a/arch/x86/include/asm/vdso/gettimeofday.h
+++ b/arch/x86/include/asm/vdso/gettimeofday.h
@@ -101,6 +101,7 @@ long clock_gettime_fallback(clockid_t _c
 {
 	long ret;
 
+#ifndef CONFIG_SECCOMP
 	asm (
 		"mov %%ebx, %%edx \n"
 		"mov %[clock], %%ebx \n"
@@ -110,6 +111,36 @@ long clock_gettime_fallback(clockid_t _c
 		: "0" (__NR_clock_gettime64), [clock] "g" (_clkid), "c" (_ts)
 		: "edx");
 
+#else
+	struct old_timespec32 tmpts;
+
+	/*
+	 * Using clock_gettime and not clock_gettime64 here is a
+	 * temporary workaround to pacify seccomp filters which are
+	 * unaware of the Y2038 safe variant.
+	 */
+
+	asm (
+		"mov %%ebx, %%edx \n"
+		"mov %[clock], %%ebx \n"
+		"call __kernel_vsyscall \n"
+		"mov %%edx, %%ebx \n"
+		: "=a" (ret), "=m" (tmpts)
+		: "0" (__NR_clock_gettime), [clock] "g" (_clkid), "c" (&tmpts)
+		: "edx");
+
+	/*
+	 * The core code will have to convert that back. A smart compiler
+	 * should notice and avoid the double conversion. If not, bad luck;
+	 * we we are not going to change the core code just to make seccomp
+	 * happy.
+	 */
+
+	if (!ret) {
+		_ts->tv_sec = tmpts.tv_sec;
+		_ts->tv_nsec = tmpts.tv_nsec;
+	}
+#endif
 	return ret;
 }
 
@@ -136,6 +167,7 @@ clock_getres_fallback(clockid_t _clkid,
 {
 	long ret;
 
+#ifndef CONFIG_SECCOMP
 	asm (
 		"mov %%ebx, %%edx \n"
 		"mov %[clock], %%ebx \n"
@@ -144,7 +176,32 @@ clock_getres_fallback(clockid_t _clkid,
 		: "=a" (ret), "=m" (*_ts)
 		: "0" (__NR_clock_getres_time64), [clock] "g" (_clkid), "c" (_ts)
 		: "edx");
+#else
+	struct old_timespec32 tmpts;
+
+	/*
+	 * Using clock_getres and not clock_getres_time64 here is a
+	 * temporary workaround to pacify seccomp filters which are unaware
+	 * of the time64 variants. Technically there is no requirement to
+	 * use the 64bit variant here as the resolution is definitely not
+	 * affected by Y2038, but the end goal of Y2038 is to utilize the
+	 * new 64bit timespec variants for everything.
+	 */
+
+	asm (
+		"mov %%ebx, %%edx \n"
+		"mov %[clock], %%ebx \n"
+		"call __kernel_vsyscall \n"
+		"mov %%edx, %%ebx \n"
+		: "=a" (ret), "=m" (tmpts)
+		: "0" (__NR_clock_getres), [clock] "g" (_clkid), "c" (&tmpts)
+		: "edx");
 
+	if (!ret) {
+		_ts->tv_sec = tmpts.tv_sec;
+		_ts->tv_nsec = tmpts.tv_nsec;
+	}
+#endif
 	return ret;
 }