/*
 * sigreturn.c - tests for x86 sigreturn(2) and exit-to-userspace
 * Copyright (c) 2014-2015 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.
 *
 * This is a series of tests that exercises the sigreturn(2) syscall and
 * the IRET / SYSRET paths in the kernel.
 *
 * For now, this focuses on the effects of unusual CS and SS values,
 * and it has a bunch of tests to make sure that ESP/RSP is restored
 * properly.
 *
 * The basic idea behind these tests is to raise(SIGUSR1) to create a
 * sigcontext frame, plug in the values to be tested, and then return,
 * which implicitly invokes sigreturn(2) and programs the user context
 * as desired.
 *
 * For tests for which we expect sigreturn and the subsequent return to
 * user mode to succeed, we return to a short trampoline that generates
 * SIGTRAP so that the meat of the tests can be ordinary C code in a
 * SIGTRAP handler.
 *
 * The inner workings of each test is documented below.
 *
 * Do not run on outdated, unpatched kernels at risk of nasty crashes.
 */

#define _GNU_SOURCE

#include <sys/time.h>
#include <time.h>
#include <stdlib.h>
#include <sys/syscall.h>
#include <unistd.h>
#include <stdio.h>
#include <string.h>
#include <inttypes.h>
#include <sys/mman.h>
#include <sys/signal.h>
#include <sys/ucontext.h>
#include <asm/ldt.h>
#include <err.h>
#include <setjmp.h>
#include <stddef.h>
#include <stdbool.h>
#include <sys/ptrace.h>
#include <sys/user.h>

/*
 * In principle, this test can run on Linux emulation layers (e.g.
 * Illumos "LX branded zones").  Solaris-based kernels reserve LDT
 * entries 0-5 for their own internal purposes, so start our LDT
 * allocations above that reservation.  (The tests don't pass on LX
 * branded zones, but at least this lets them run.)
 */
#define LDT_OFFSET 6

/* An aligned stack accessible through some of our segments. */
static unsigned char stack16[65536] __attribute__((aligned(4096)));

/*
 * An aligned int3 instruction used as a trampoline.  Some of the tests
 * want to fish out their ss values, so this trampoline copies ss to eax
 * before the int3.
 */
asm (".pushsection .text\n\t"
     ".type int3, @function\n\t"
     ".align 4096\n\t"
     "int3:\n\t"
     "mov %ss,%eax\n\t"
     "int3\n\t"
     ".size int3, . - int3\n\t"
     ".align 4096, 0xcc\n\t"
     ".popsection");
extern char int3[4096];

/*
 * At startup, we prepapre:
 *
 * - ldt_nonexistent_sel: An LDT entry that doesn't exist (all-zero
 *   descriptor or out of bounds).
 * - code16_sel: A 16-bit LDT code segment pointing to int3.
 * - data16_sel: A 16-bit LDT data segment pointing to stack16.
 * - npcode32_sel: A 32-bit not-present LDT code segment pointing to int3.
 * - npdata32_sel: A 32-bit not-present LDT data segment pointing to stack16.
 * - gdt_data16_idx: A 16-bit GDT data segment pointing to stack16.
 * - gdt_npdata32_idx: A 32-bit not-present GDT data segment pointing to
 *   stack16.
 *
 * For no particularly good reason, xyz_sel is a selector value with the
 * RPL and LDT bits filled in, whereas xyz_idx is just an index into the
 * descriptor table.  These variables will be zero if their respective
 * segments could not be allocated.
 */
static unsigned short ldt_nonexistent_sel;
static unsigned short code16_sel, data16_sel, npcode32_sel, npdata32_sel;

static unsigned short gdt_data16_idx, gdt_npdata32_idx;

static unsigned short GDT3(int idx)
{
	return (idx << 3) | 3;
}

static unsigned short LDT3(int idx)
{
	return (idx << 3) | 7;
}

/* Our sigaltstack scratch space. */
static char altstack_data[SIGSTKSZ];

static void sethandler(int sig, void (*handler)(int, siginfo_t *, void *),
		       int flags)
{
	struct sigaction sa;
	memset(&sa, 0, sizeof(sa));
	sa.sa_sigaction = handler;
	sa.sa_flags = SA_SIGINFO | flags;
	sigemptyset(&sa.sa_mask);
	if (sigaction(sig, &sa, 0))
		err(1, "sigaction");
}

static void clearhandler(int sig)
{
	struct sigaction sa;
	memset(&sa, 0, sizeof(sa));
	sa.sa_handler = SIG_DFL;
	sigemptyset(&sa.sa_mask);
	if (sigaction(sig, &sa, 0))
		err(1, "sigaction");
}

static void add_ldt(const struct user_desc *desc, unsigned short *var,
		    const char *name)
{
	if (syscall(SYS_modify_ldt, 1, desc, sizeof(*desc)) == 0) {
		*var = LDT3(desc->entry_number);
	} else {
		printf("[NOTE]\tFailed to create %s segment\n", name);
		*var = 0;
	}
}

static void setup_ldt(void)
{
	if ((unsigned long)stack16 > (1ULL << 32) - sizeof(stack16))
		errx(1, "stack16 is too high\n");
	if ((unsigned long)int3 > (1ULL << 32) - sizeof(int3))
		errx(1, "int3 is too high\n");

	ldt_nonexistent_sel = LDT3(LDT_OFFSET + 2);

	const struct user_desc code16_desc = {
		.entry_number    = LDT_OFFSET + 0,
		.base_addr       = (unsigned long)int3,
		.limit           = 4095,
		.seg_32bit       = 0,
		.contents        = 2, /* Code, not conforming */
		.read_exec_only  = 0,
		.limit_in_pages  = 0,
		.seg_not_present = 0,
		.useable         = 0
	};
	add_ldt(&code16_desc, &code16_sel, "code16");

	const struct user_desc data16_desc = {
		.entry_number    = LDT_OFFSET + 1,
		.base_addr       = (unsigned long)stack16,
		.limit           = 0xffff,
		.seg_32bit       = 0,
		.contents        = 0, /* Data, grow-up */
		.read_exec_only  = 0,
		.limit_in_pages  = 0,
		.seg_not_present = 0,
		.useable         = 0
	};
	add_ldt(&data16_desc, &data16_sel, "data16");

	const struct user_desc npcode32_desc = {
		.entry_number    = LDT_OFFSET + 3,
		.base_addr       = (unsigned long)int3,
		.limit           = 4095,
		.seg_32bit       = 1,
		.contents        = 2, /* Code, not conforming */
		.read_exec_only  = 0,
		.limit_in_pages  = 0,
		.seg_not_present = 1,
		.useable         = 0
	};
	add_ldt(&npcode32_desc, &npcode32_sel, "npcode32");

	const struct user_desc npdata32_desc = {
		.entry_number    = LDT_OFFSET + 4,
		.base_addr       = (unsigned long)stack16,
		.limit           = 0xffff,
		.seg_32bit       = 1,
		.contents        = 0, /* Data, grow-up */
		.read_exec_only  = 0,
		.limit_in_pages  = 0,
		.seg_not_present = 1,
		.useable         = 0
	};
	add_ldt(&npdata32_desc, &npdata32_sel, "npdata32");

	struct user_desc gdt_data16_desc = {
		.entry_number    = -1,
		.base_addr       = (unsigned long)stack16,
		.limit           = 0xffff,
		.seg_32bit       = 0,
		.contents        = 0, /* Data, grow-up */
		.read_exec_only  = 0,
		.limit_in_pages  = 0,
		.seg_not_present = 0,
		.useable         = 0
	};

	if (syscall(SYS_set_thread_area, &gdt_data16_desc) == 0) {
		/*
		 * This probably indicates vulnerability to CVE-2014-8133.
		 * Merely getting here isn't definitive, though, and we'll
		 * diagnose the problem for real later on.
		 */
		printf("[WARN]\tset_thread_area allocated data16 at index %d\n",
		       gdt_data16_desc.entry_number);
		gdt_data16_idx = gdt_data16_desc.entry_number;
	} else {
		printf("[OK]\tset_thread_area refused 16-bit data\n");
	}

	struct user_desc gdt_npdata32_desc = {
		.entry_number    = -1,
		.base_addr       = (unsigned long)stack16,
		.limit           = 0xffff,
		.seg_32bit       = 1,
		.contents        = 0, /* Data, grow-up */
		.read_exec_only  = 0,
		.limit_in_pages  = 0,
		.seg_not_present = 1,
		.useable         = 0
	};

	if (syscall(SYS_set_thread_area, &gdt_npdata32_desc) == 0) {
		/*
		 * As a hardening measure, newer kernels don't allow this.
		 */
		printf("[WARN]\tset_thread_area allocated npdata32 at index %d\n",
		       gdt_npdata32_desc.entry_number);
		gdt_npdata32_idx = gdt_npdata32_desc.entry_number;
	} else {
		printf("[OK]\tset_thread_area refused 16-bit data\n");
	}
}

/* State used by our signal handlers. */
static gregset_t initial_regs, requested_regs, resulting_regs;

/* Instructions for the SIGUSR1 handler. */
static volatile unsigned short sig_cs, sig_ss;
static volatile sig_atomic_t sig_trapped, sig_err, sig_trapno;

/* Abstractions for some 32-bit vs 64-bit differences. */
#ifdef __x86_64__
# define REG_IP REG_RIP
# define REG_SP REG_RSP
# define REG_AX REG_RAX

struct selectors {
	unsigned short cs, gs, fs, ss;
};

static unsigned short *ssptr(ucontext_t *ctx)
{
	struct selectors *sels = (void *)&ctx->uc_mcontext.gregs[REG_CSGSFS];
	return &sels->ss;
}

static unsigned short *csptr(ucontext_t *ctx)
{
	struct selectors *sels = (void *)&ctx->uc_mcontext.gregs[REG_CSGSFS];
	return &sels->cs;
}
#else
# define REG_IP REG_EIP
# define REG_SP REG_ESP
# define REG_AX REG_EAX

static greg_t *ssptr(ucontext_t *ctx)
{
	return &ctx->uc_mcontext.gregs[REG_SS];
}

static greg_t *csptr(ucontext_t *ctx)
{
	return &ctx->uc_mcontext.gregs[REG_CS];
}
#endif

/* Number of errors in the current test case. */
static volatile sig_atomic_t nerrs;

/*
 * SIGUSR1 handler.  Sets CS and SS as requested and points IP to the
 * int3 trampoline.  Sets SP to a large known value so that we can see
 * whether the value round-trips back to user mode correctly.
 */
static void sigusr1(int sig, siginfo_t *info, void *ctx_void)
{
	ucontext_t *ctx = (ucontext_t*)ctx_void;

	memcpy(&initial_regs, &ctx->uc_mcontext.gregs, sizeof(gregset_t));

	*csptr(ctx) = sig_cs;
	*ssptr(ctx) = sig_ss;

	ctx->uc_mcontext.gregs[REG_IP] =
		sig_cs == code16_sel ? 0 : (unsigned long)&int3;
	ctx->uc_mcontext.gregs[REG_SP] = (unsigned long)0x8badf00d5aadc0deULL;
	ctx->uc_mcontext.gregs[REG_AX] = 0;

	memcpy(&requested_regs, &ctx->uc_mcontext.gregs, sizeof(gregset_t));
	requested_regs[REG_AX] = *ssptr(ctx);	/* The asm code does this. */

	return;
}

/*
 * Called after a successful sigreturn.  Restores our state so that
 * the original raise(SIGUSR1) returns.
 */
static void sigtrap(int sig, siginfo_t *info, void *ctx_void)
{
	ucontext_t *ctx = (ucontext_t*)ctx_void;

	sig_err = ctx->uc_mcontext.gregs[REG_ERR];
	sig_trapno = ctx->uc_mcontext.gregs[REG_TRAPNO];

	unsigned short ss;
	asm ("mov %%ss,%0" : "=r" (ss));

	greg_t asm_ss = ctx->uc_mcontext.gregs[REG_AX];
	if (asm_ss != sig_ss && sig == SIGTRAP) {
		/* Sanity check failure. */
		printf("[FAIL]\tSIGTRAP: ss = %hx, frame ss = %hx, ax = %llx\n",
		       ss, *ssptr(ctx), (unsigned long long)asm_ss);
		nerrs++;
	}

	memcpy(&resulting_regs, &ctx->uc_mcontext.gregs, sizeof(gregset_t));
	memcpy(&ctx->uc_mcontext.gregs, &initial_regs, sizeof(gregset_t));

	sig_trapped = sig;
}

/*
 * Checks a given selector for its code bitness or returns -1 if it's not
 * a usable code segment selector.
 */
int cs_bitness(unsigned short cs)
{
	uint32_t valid = 0, ar;
	asm ("lar %[cs], %[ar]\n\t"
	     "jnz 1f\n\t"
	     "mov $1, %[valid]\n\t"
	     "1:"
	     : [ar] "=r" (ar), [valid] "+rm" (valid)
	     : [cs] "r" (cs));

	if (!valid)
		return -1;

	bool db = (ar & (1 << 22));
	bool l = (ar & (1 << 21));

	if (!(ar & (1<<11)))
	    return -1;	/* Not code. */

	if (l && !db)
		return 64;
	else if (!l && db)
		return 32;
	else if (!l && !db)
		return 16;
	else
		return -1;	/* Unknown bitness. */
}

/* Finds a usable code segment of the requested bitness. */
int find_cs(int bitness)
{
	unsigned short my_cs;

	asm ("mov %%cs,%0" :  "=r" (my_cs));

	if (cs_bitness(my_cs) == bitness)
		return my_cs;
	if (cs_bitness(my_cs + (2 << 3)) == bitness)
		return my_cs + (2 << 3);
	if (my_cs > (2<<3) && cs_bitness(my_cs - (2 << 3)) == bitness)
	    return my_cs - (2 << 3);
	if (cs_bitness(code16_sel) == bitness)
		return code16_sel;

	printf("[WARN]\tCould not find %d-bit CS\n", bitness);
	return -1;
}

static int test_valid_sigreturn(int cs_bits, bool use_16bit_ss, int force_ss)
{
	int cs = find_cs(cs_bits);
	if (cs == -1) {
		printf("[SKIP]\tCode segment unavailable for %d-bit CS, %d-bit SS\n",
		       cs_bits, use_16bit_ss ? 16 : 32);
		return 0;
	}

	if (force_ss != -1) {
		sig_ss = force_ss;
	} else {
		if (use_16bit_ss) {
			if (!data16_sel) {
				printf("[SKIP]\tData segment unavailable for %d-bit CS, 16-bit SS\n",
				       cs_bits);
				return 0;
			}
			sig_ss = data16_sel;
		} else {
			asm volatile ("mov %%ss,%0" : "=r" (sig_ss));
		}
	}

	sig_cs = cs;

	printf("[RUN]\tValid sigreturn: %d-bit CS (%hx), %d-bit SS (%hx%s)\n",
	       cs_bits, sig_cs, use_16bit_ss ? 16 : 32, sig_ss,
	       (sig_ss & 4) ? "" : ", GDT");

	raise(SIGUSR1);

	nerrs = 0;

	/*
	 * Check that each register had an acceptable value when the
	 * int3 trampoline was invoked.
	 */
	for (int i = 0; i < NGREG; i++) {
		greg_t req = requested_regs[i], res = resulting_regs[i];
		if (i == REG_TRAPNO || i == REG_IP)
			continue;	/* don't care */

		if (i == REG_SP) {
			/*
			 * If we were using a 16-bit stack segment, then
			 * the kernel is a bit stuck: IRET only restores
			 * the low 16 bits of ESP/RSP if SS is 16-bit.
			 * The kernel uses a hack to restore bits 31:16,
			 * but that hack doesn't help with bits 63:32.
			 * On Intel CPUs, bits 63:32 end up zeroed, and, on
			 * AMD CPUs, they leak the high bits of the kernel
			 * espfix64 stack pointer.  There's very little that
			 * the kernel can do about it.
			 *
			 * Similarly, if we are returning to a 32-bit context,
			 * the CPU will often lose the high 32 bits of RSP.
			 */

			if (res == req)
				continue;

			if (cs_bits != 64 && ((res ^ req) & 0xFFFFFFFF) == 0) {
				printf("[NOTE]\tSP: %llx -> %llx\n",
				       (unsigned long long)req,
				       (unsigned long long)res);
				continue;
			}

			printf("[FAIL]\tSP mismatch: requested 0x%llx; got 0x%llx\n",
			       (unsigned long long)requested_regs[i],
			       (unsigned long long)resulting_regs[i]);
			nerrs++;
			continue;
		}

		bool ignore_reg = false;
#if __i386__
		if (i == REG_UESP)
			ignore_reg = true;
#else
		if (i == REG_CSGSFS) {
			struct selectors *req_sels =
				(void *)&requested_regs[REG_CSGSFS];
			struct selectors *res_sels =
				(void *)&resulting_regs[REG_CSGSFS];
			if (req_sels->cs != res_sels->cs) {
				printf("[FAIL]\tCS mismatch: requested 0x%hx; got 0x%hx\n",
				       req_sels->cs, res_sels->cs);
				nerrs++;
			}

			if (req_sels->ss != res_sels->ss) {
				printf("[FAIL]\tSS mismatch: requested 0x%hx; got 0x%hx\n",
				       req_sels->ss, res_sels->ss);
				nerrs++;
			}

			continue;
		}
#endif

		/* Sanity check on the kernel */
		if (i == REG_AX && requested_regs[i] != resulting_regs[i]) {
			printf("[FAIL]\tAX (saved SP) mismatch: requested 0x%llx; got 0x%llx\n",
			       (unsigned long long)requested_regs[i],
			       (unsigned long long)resulting_regs[i]);
			nerrs++;
			continue;
		}

		if (requested_regs[i] != resulting_regs[i] && !ignore_reg) {
			printf("[FAIL]\tReg %d mismatch: requested 0x%llx; got 0x%llx\n",
			       i, (unsigned long long)requested_regs[i],
			       (unsigned long long)resulting_regs[i]);
			nerrs++;
		}
	}

	if (nerrs == 0)
		printf("[OK]\tall registers okay\n");

	return nerrs;
}

static int test_bad_iret(int cs_bits, unsigned short ss, int force_cs)
{
	int cs = force_cs == -1 ? find_cs(cs_bits) : force_cs;
	if (cs == -1)
		return 0;

	sig_cs = cs;
	sig_ss = ss;

	printf("[RUN]\t%d-bit CS (%hx), bogus SS (%hx)\n",
	       cs_bits, sig_cs, sig_ss);

	sig_trapped = 0;
	raise(SIGUSR1);
	if (sig_trapped) {
		char errdesc[32] = "";
		if (sig_err) {
			const char *src = (sig_err & 1) ? " EXT" : "";
			const char *table;
			if ((sig_err & 0x6) == 0x0)
				table = "GDT";
			else if ((sig_err & 0x6) == 0x4)
				table = "LDT";
			else if ((sig_err & 0x6) == 0x2)
				table = "IDT";
			else
				table = "???";

			sprintf(errdesc, "%s%s index %d, ",
				table, src, sig_err >> 3);
		}

		char trapname[32];
		if (sig_trapno == 13)
			strcpy(trapname, "GP");
		else if (sig_trapno == 11)
			strcpy(trapname, "NP");
		else if (sig_trapno == 12)
			strcpy(trapname, "SS");
		else if (sig_trapno == 32)
			strcpy(trapname, "IRET");  /* X86_TRAP_IRET */
		else
			sprintf(trapname, "%d", sig_trapno);

		printf("[OK]\tGot #%s(0x%lx) (i.e. %s%s)\n",
		       trapname, (unsigned long)sig_err,
		       errdesc, strsignal(sig_trapped));
		return 0;
	} else {
		printf("[FAIL]\tDid not get SIGSEGV\n");
		return 1;
	}
}

int main()
{
	int total_nerrs = 0;
	unsigned short my_cs, my_ss;

	asm volatile ("mov %%cs,%0" : "=r" (my_cs));
	asm volatile ("mov %%ss,%0" : "=r" (my_ss));
	setup_ldt();

	stack_t stack = {
		.ss_sp = altstack_data,
		.ss_size = SIGSTKSZ,
	};
	if (sigaltstack(&stack, NULL) != 0)
		err(1, "sigaltstack");

	sethandler(SIGUSR1, sigusr1, 0);
	sethandler(SIGTRAP, sigtrap, SA_ONSTACK);

	/* Easy cases: return to a 32-bit SS in each possible CS bitness. */
	total_nerrs += test_valid_sigreturn(64, false, -1);
	total_nerrs += test_valid_sigreturn(32, false, -1);
	total_nerrs += test_valid_sigreturn(16, false, -1);

	/*
	 * Test easy espfix cases: return to a 16-bit LDT SS in each possible
	 * CS bitness.  NB: with a long mode CS, the SS bitness is irrelevant.
	 *
	 * This catches the original missing-espfix-on-64-bit-kernels issue
	 * as well as CVE-2014-8134.
	 */
	total_nerrs += test_valid_sigreturn(64, true, -1);
	total_nerrs += test_valid_sigreturn(32, true, -1);
	total_nerrs += test_valid_sigreturn(16, true, -1);

	if (gdt_data16_idx) {
		/*
		 * For performance reasons, Linux skips espfix if SS points
		 * to the GDT.  If we were able to allocate a 16-bit SS in
		 * the GDT, see if it leaks parts of the kernel stack pointer.
		 *
		 * This tests for CVE-2014-8133.
		 */
		total_nerrs += test_valid_sigreturn(64, true,
						    GDT3(gdt_data16_idx));
		total_nerrs += test_valid_sigreturn(32, true,
						    GDT3(gdt_data16_idx));
		total_nerrs += test_valid_sigreturn(16, true,
						    GDT3(gdt_data16_idx));
	}

	/*
	 * We're done testing valid sigreturn cases.  Now we test states
	 * for which sigreturn itself will succeed but the subsequent
	 * entry to user mode will fail.
	 *
	 * Depending on the failure mode and the kernel bitness, these
	 * entry failures can generate SIGSEGV, SIGBUS, or SIGILL.
	 */
	clearhandler(SIGTRAP);
	sethandler(SIGSEGV, sigtrap, SA_ONSTACK);
	sethandler(SIGBUS, sigtrap, SA_ONSTACK);
	sethandler(SIGILL, sigtrap, SA_ONSTACK);  /* 32-bit kernels do this */

	/* Easy failures: invalid SS, resulting in #GP(0) */
	test_bad_iret(64, ldt_nonexistent_sel, -1);
	test_bad_iret(32, ldt_nonexistent_sel, -1);
	test_bad_iret(16, ldt_nonexistent_sel, -1);

	/* These fail because SS isn't a data segment, resulting in #GP(SS) */
	test_bad_iret(64, my_cs, -1);
	test_bad_iret(32, my_cs, -1);
	test_bad_iret(16, my_cs, -1);

	/* Try to return to a not-present code segment, triggering #NP(SS). */
	test_bad_iret(32, my_ss, npcode32_sel);

	/*
	 * Try to return to a not-present but otherwise valid data segment.
	 * This will cause IRET to fail with #SS on the espfix stack.  This
	 * exercises CVE-2014-9322.
	 *
	 * Note that, if espfix is enabled, 64-bit Linux will lose track
	 * of the actual cause of failure and report #GP(0) instead.
	 * This would be very difficult for Linux to avoid, because
	 * espfix64 causes IRET failures to be promoted to #DF, so the
	 * original exception frame is never pushed onto the stack.
	 */
	test_bad_iret(32, npdata32_sel, -1);

	/*
	 * Try to return to a not-present but otherwise valid data
	 * segment without invoking espfix.  Newer kernels don't allow
	 * this to happen in the first place.  On older kernels, though,
	 * this can trigger CVE-2014-9322.
	 */
	if (gdt_npdata32_idx)
		test_bad_iret(32, GDT3(gdt_npdata32_idx), -1);

	return total_nerrs ? 1 : 0;
}