x86/vm86: Block non-root vm86(old) if mmap_min_addr != 0

vm86 exposes an interesting attack surface against the entry
code. Since vm86 is mostly useless anyway if mmap_min_addr != 0,
just turn it off in that case.

There are some reports that vbetool can work despite setting
mmap_min_addr to zero.  This shouldn't break that use case,
as CAP_SYS_RAWIO already overrides mmap_min_addr.

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Austin S Hemmelgarn <ahferroin7@gmail.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Josh Boyer <jwboyer@fedoraproject.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stas Sergeev <stsp@list.ru>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

arch/x86/kernel/vm86_32.c

@@ -45,6 +45,7 @@
 #include <linux/audit.h>
 #include <linux/stddef.h>
 #include <linux/slab.h>
+#include <linux/security.h>
 
 #include <asm/uaccess.h>
 #include <asm/io.h>
@@ -232,6 +233,32 @@ static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
 	struct pt_regs *regs = current_pt_regs();
 	unsigned long err = 0;
 
+	err = security_mmap_addr(0);
+	if (err) {
+		/*
+		 * vm86 cannot virtualize the address space, so vm86 users
+		 * need to manage the low 1MB themselves using mmap.  Given
+		 * that BIOS places important data in the first page, vm86
+		 * is essentially useless if mmap_min_addr != 0.  DOSEMU,
+		 * for example, won't even bother trying to use vm86 if it
+		 * can't map a page at virtual address 0.
+		 *
+		 * To reduce the available kernel attack surface, simply
+		 * disallow vm86(old) for users who cannot mmap at va 0.
+		 *
+		 * The implementation of security_mmap_addr will allow
+		 * suitably privileged users to map va 0 even if
+		 * vm.mmap_min_addr is set above 0, and we want this
+		 * behavior for vm86 as well, as it ensures that legacy
+		 * tools like vbetool will not fail just because of
+		 * vm.mmap_min_addr.
+		 */
+		pr_info_once("Denied a call to vm86(old) from %s[%d] (uid: %d).  Set the vm.mmap_min_addr sysctl to 0 and/or adjust LSM mmap_min_addr policy to enable vm86 if you are using a vm86-based DOS emulator.\n",
+			     current->comm, task_pid_nr(current),
+			     from_kuid_munged(&init_user_ns, current_uid()));
+		return -EPERM;
+	}
+
 	if (!vm86) {
 		if (!(vm86 = kzalloc(sizeof(*vm86), GFP_KERNEL)))
 			return -ENOMEM;

tools/testing/selftests/x86/entry_from_vm86.c

@@ -116,8 +116,9 @@ static bool do_test(struct vm86plus_struct *v86, unsigned long eip,
 	v86->regs.eip = eip;
 	ret = vm86(VM86_ENTER, v86);
 
-	if (ret == -1 && errno == ENOSYS) {
+	if (ret == -1 && (errno == ENOSYS || errno == EPERM)) {
-		printf("[SKIP]\tvm86 not supported\n");
+		printf("[SKIP]\tvm86 %s\n",
+		       errno == ENOSYS ? "not supported" : "not allowed");
 		return false;
 	}