• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * 32-bit syscall ABI conformance test.
3  *
4  * Copyright (c) 2015 Denys Vlasenko
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms and conditions of the GNU General Public License,
8  * version 2, as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License for more details.
14  */
15 /*
16  * Can be built statically:
17  * gcc -Os -Wall -static -m32 test_syscall_vdso.c thunks_32.S
18  */
19 #undef _GNU_SOURCE
20 #define _GNU_SOURCE 1
21 #undef __USE_GNU
22 #define __USE_GNU 1
23 #include <unistd.h>
24 #include <stdlib.h>
25 #include <string.h>
26 #include <stdio.h>
27 #include <signal.h>
28 #include <sys/types.h>
29 #include <sys/select.h>
30 #include <sys/time.h>
31 #include <elf.h>
32 #include <sys/ptrace.h>
33 #include <sys/utsname.h>
34 #include <sys/wait.h>
35 
36 #if !defined(__i386__)
main(int argc,char ** argv,char ** envp)37 int main(int argc, char **argv, char **envp)
38 {
39 	printf("[SKIP]\tNot a 32-bit x86 userspace\n");
40 	return 0;
41 }
42 #else
43 
44 long syscall_addr;
get_syscall(char ** envp)45 long get_syscall(char **envp)
46 {
47 	Elf32_auxv_t *auxv;
48 	while (*envp++ != NULL)
49 		continue;
50 	for (auxv = (void *)envp; auxv->a_type != AT_NULL; auxv++)
51 		if (auxv->a_type == AT_SYSINFO)
52 			return auxv->a_un.a_val;
53 	printf("[WARN]\tAT_SYSINFO not supplied\n");
54 	return 0;
55 }
56 
57 asm (
58 	"	.pushsection .text\n"
59 	"	.global	int80\n"
60 	"int80:\n"
61 	"	int	$0x80\n"
62 	"	ret\n"
63 	"	.popsection\n"
64 );
65 extern char int80;
66 
67 struct regs64 {
68 	uint64_t rax, rbx, rcx, rdx;
69 	uint64_t rsi, rdi, rbp, rsp;
70 	uint64_t r8,  r9,  r10, r11;
71 	uint64_t r12, r13, r14, r15;
72 };
73 struct regs64 regs64;
74 int kernel_is_64bit;
75 int clobber_ok;
76 
77 asm (
78 	"	.pushsection .text\n"
79 	"	.code64\n"
80 	"get_regs64:\n"
81 	"	push	%rax\n"
82 	"	mov	$regs64, %eax\n"
83 	"	pop	0*8(%rax)\n"
84 	"	movq	%rbx, 1*8(%rax)\n"
85 	"	movq	%rcx, 2*8(%rax)\n"
86 	"	movq	%rdx, 3*8(%rax)\n"
87 	"	movq	%rsi, 4*8(%rax)\n"
88 	"	movq	%rdi, 5*8(%rax)\n"
89 	"	movq	%rbp, 6*8(%rax)\n"
90 	"	movq	%rsp, 7*8(%rax)\n"
91 	"	movq	%r8,  8*8(%rax)\n"
92 	"	movq	%r9,  9*8(%rax)\n"
93 	"	movq	%r10, 10*8(%rax)\n"
94 	"	movq	%r11, 11*8(%rax)\n"
95 	"	movq	%r12, 12*8(%rax)\n"
96 	"	movq	%r13, 13*8(%rax)\n"
97 	"	movq	%r14, 14*8(%rax)\n"
98 	"	movq	%r15, 15*8(%rax)\n"
99 	"	ret\n"
100 	"poison_regs64:\n"
101 	"	movq	$0x7f7f7f7f, %r8\n"
102 	"	shl	$32, %r8\n"
103 	"	orq	$0x7f7f7f7f, %r8\n"
104 	"	movq	%r8, %r9\n"
105 	"	incq	%r9\n"
106 	"	movq	%r9, %r10\n"
107 	"	incq	%r10\n"
108 	"	movq	%r10, %r11\n"
109 	"	incq	%r11\n"
110 	"	movq	%r11, %r12\n"
111 	"	incq	%r12\n"
112 	"	movq	%r12, %r13\n"
113 	"	incq	%r13\n"
114 	"	movq	%r13, %r14\n"
115 	"	incq	%r14\n"
116 	"	movq	%r14, %r15\n"
117 	"	incq	%r15\n"
118 	"	ret\n"
119 	"	.code32\n"
120 	"	.popsection\n"
121 );
122 extern void get_regs64(void);
123 extern void poison_regs64(void);
124 extern unsigned long call64_from_32(void (*function)(void));
print_regs64(void)125 void print_regs64(void)
126 {
127 	if (!kernel_is_64bit)
128 		return;
129 	printf("ax:%016llx bx:%016llx cx:%016llx dx:%016llx\n", regs64.rax,  regs64.rbx,  regs64.rcx,  regs64.rdx);
130 	printf("si:%016llx di:%016llx bp:%016llx sp:%016llx\n", regs64.rsi,  regs64.rdi,  regs64.rbp,  regs64.rsp);
131 	printf(" 8:%016llx  9:%016llx 10:%016llx 11:%016llx\n", regs64.r8 ,  regs64.r9 ,  regs64.r10,  regs64.r11);
132 	printf("12:%016llx 13:%016llx 14:%016llx 15:%016llx\n", regs64.r12,  regs64.r13,  regs64.r14,  regs64.r15);
133 }
134 
get_kernel_version(int * version,int * patchlevel)135 static void get_kernel_version(int *version, int *patchlevel)
136 {
137 	int ret, sublevel;
138 	struct utsname utsname;
139 
140 	ret = uname(&utsname);
141 	if (ret) {
142 		perror("uname");
143 		exit(1);
144 	}
145 
146 	ret = sscanf(utsname.release, "%d.%d.%d", version, patchlevel,
147 		     &sublevel);
148 	if (ret < 0) {
149 		perror("sscanf");
150 		exit(1);
151 	} else if (ret != 3) {
152 		printf("Malformed kernel version %s\n", utsname.release);
153 		exit(1);
154 	}
155 }
156 
check_regs64(void)157 int check_regs64(void)
158 {
159 	int err = 0;
160 	int num = 8;
161 	uint64_t *r64 = &regs64.r8;
162 	uint64_t expected = 0x7f7f7f7f7f7f7f7fULL;
163 
164 	if (!kernel_is_64bit)
165 		return 0;
166 
167 	do {
168 		if (*r64 == expected++)
169 			continue; /* register did not change */
170 		if (syscall_addr != (long)&int80) {
171 			/*
172 			 * Non-INT80 syscall entrypoints are allowed to clobber R8+ regs:
173 			 * either clear them to 0, or for R11, load EFLAGS.
174 			 */
175 			if (*r64 == 0)
176 				continue;
177 			if (num == 11) {
178 				printf("[NOTE]\tR11 has changed:%016llx - assuming clobbered by SYSRET insn\n", *r64);
179 				continue;
180 			}
181 		} else {
182 			/*
183 			 * INT80 syscall entrypoint can be used by
184 			 * 64-bit programs too, unlike SYSCALL/SYSENTER.
185 			 * Therefore it must preserve R12+
186 			 * (they are callee-saved registers in 64-bit C ABI).
187 			 *
188 			 * Starting in Linux 4.17 (and any kernel that
189 			 * backports the change), R8..11 are preserved.
190 			 * Historically (and probably unintentionally), they
191 			 * were clobbered or zeroed.
192 			 */
193 			if (clobber_ok && *r64 == 0 && num <= 11) {
194 				printf("Warning: kernel zeroed r%d, "
195 				       "allowing on < v4.17\n", num);
196 				continue;
197 			}
198 		}
199 		printf("[FAIL]\tR%d has changed:%016llx\n", num, *r64);
200 		err++;
201 	} while (r64++, ++num < 16);
202 
203 	if (!err)
204 		printf("[OK]\tR8..R15 did not leak kernel data\n");
205 	return err;
206 }
207 
208 int nfds;
209 fd_set rfds;
210 fd_set wfds;
211 fd_set efds;
212 struct timespec timeout;
213 sigset_t sigmask;
214 struct {
215 	sigset_t *sp;
216 	int sz;
217 } sigmask_desc;
218 
prep_args()219 void prep_args()
220 {
221 	nfds = 42;
222 	FD_ZERO(&rfds);
223 	FD_ZERO(&wfds);
224 	FD_ZERO(&efds);
225 	FD_SET(0, &rfds);
226 	FD_SET(1, &wfds);
227 	FD_SET(2, &efds);
228 	timeout.tv_sec = 0;
229 	timeout.tv_nsec = 123;
230 	sigemptyset(&sigmask);
231 	sigaddset(&sigmask, SIGINT);
232 	sigaddset(&sigmask, SIGUSR2);
233 	sigaddset(&sigmask, SIGRTMAX);
234 	sigmask_desc.sp = &sigmask;
235 	sigmask_desc.sz = 8; /* bytes */
236 }
237 
print_flags(const char * name,unsigned long r)238 static void print_flags(const char *name, unsigned long r)
239 {
240 	static const char *bitarray[] = {
241 	"\n" ,"c\n" ,/* Carry Flag */
242 	"0 " ,"1 "  ,/* Bit 1 - always on */
243 	""   ,"p "  ,/* Parity Flag */
244 	"0 " ,"3? " ,
245 	""   ,"a "  ,/* Auxiliary carry Flag */
246 	"0 " ,"5? " ,
247 	""   ,"z "  ,/* Zero Flag */
248 	""   ,"s "  ,/* Sign Flag */
249 	""   ,"t "  ,/* Trap Flag */
250 	""   ,"i "  ,/* Interrupt Flag */
251 	""   ,"d "  ,/* Direction Flag */
252 	""   ,"o "  ,/* Overflow Flag */
253 	"0 " ,"1 "  ,/* I/O Privilege Level (2 bits) */
254 	"0"  ,"1"   ,/* I/O Privilege Level (2 bits) */
255 	""   ,"n "  ,/* Nested Task */
256 	"0 " ,"15? ",
257 	""   ,"r "  ,/* Resume Flag */
258 	""   ,"v "  ,/* Virtual Mode */
259 	""   ,"ac " ,/* Alignment Check/Access Control */
260 	""   ,"vif ",/* Virtual Interrupt Flag */
261 	""   ,"vip ",/* Virtual Interrupt Pending */
262 	""   ,"id " ,/* CPUID detection */
263 	NULL
264 	};
265 	const char **bitstr;
266 	int bit;
267 
268 	printf("%s=%016lx ", name, r);
269 	bitstr = bitarray + 42;
270 	bit = 21;
271 	if ((r >> 22) != 0)
272 		printf("(extra bits are set) ");
273 	do {
274 		if (bitstr[(r >> bit) & 1][0])
275 			fputs(bitstr[(r >> bit) & 1], stdout);
276 		bitstr -= 2;
277 		bit--;
278 	} while (bit >= 0);
279 }
280 
run_syscall(void)281 int run_syscall(void)
282 {
283 	long flags, bad_arg;
284 
285 	prep_args();
286 
287 	if (kernel_is_64bit)
288 		call64_from_32(poison_regs64);
289 	/*print_regs64();*/
290 
291 	asm("\n"
292 	/* Try 6-arg syscall: pselect. It should return quickly */
293 	"	push	%%ebp\n"
294 	"	mov	$308, %%eax\n"     /* PSELECT */
295 	"	mov	nfds, %%ebx\n"     /* ebx  arg1 */
296 	"	mov	$rfds, %%ecx\n"    /* ecx  arg2 */
297 	"	mov	$wfds, %%edx\n"    /* edx  arg3 */
298 	"	mov	$efds, %%esi\n"    /* esi  arg4 */
299 	"	mov	$timeout, %%edi\n" /* edi  arg5 */
300 	"	mov	$sigmask_desc, %%ebp\n" /* %ebp arg6 */
301 	"	push	$0x200ed7\n"      /* set almost all flags */
302 	"	popf\n"		/* except TF, IOPL, NT, RF, VM, AC, VIF, VIP */
303 	"	call	*syscall_addr\n"
304 	/* Check that registers are not clobbered */
305 	"	pushf\n"
306 	"	pop	%%eax\n"
307 	"	cld\n"
308 	"	cmp	nfds, %%ebx\n"     /* ebx  arg1 */
309 	"	mov	$1, %%ebx\n"
310 	"	jne	1f\n"
311 	"	cmp	$rfds, %%ecx\n"    /* ecx  arg2 */
312 	"	mov	$2, %%ebx\n"
313 	"	jne	1f\n"
314 	"	cmp	$wfds, %%edx\n"    /* edx  arg3 */
315 	"	mov	$3, %%ebx\n"
316 	"	jne	1f\n"
317 	"	cmp	$efds, %%esi\n"    /* esi  arg4 */
318 	"	mov	$4, %%ebx\n"
319 	"	jne	1f\n"
320 	"	cmp	$timeout, %%edi\n" /* edi  arg5 */
321 	"	mov	$5, %%ebx\n"
322 	"	jne	1f\n"
323 	"	cmpl	$sigmask_desc, %%ebp\n" /* %ebp arg6 */
324 	"	mov	$6, %%ebx\n"
325 	"	jne	1f\n"
326 	"	mov	$0, %%ebx\n"
327 	"1:\n"
328 	"	pop	%%ebp\n"
329 	: "=a" (flags), "=b" (bad_arg)
330 	:
331 	: "cx", "dx", "si", "di"
332 	);
333 
334 	if (kernel_is_64bit) {
335 		memset(&regs64, 0x77, sizeof(regs64));
336 		call64_from_32(get_regs64);
337 		/*print_regs64();*/
338 	}
339 
340 	/*
341 	 * On paravirt kernels, flags are not preserved across syscalls.
342 	 * Thus, we do not consider it a bug if some are changed.
343 	 * We just show ones which do.
344 	 */
345 	if ((0x200ed7 ^ flags) != 0) {
346 		print_flags("[WARN]\tFlags before", 0x200ed7);
347 		print_flags("[WARN]\tFlags  after", flags);
348 		print_flags("[WARN]\tFlags change", (0x200ed7 ^ flags));
349 	}
350 
351 	if (bad_arg) {
352 		printf("[FAIL]\targ#%ld clobbered\n", bad_arg);
353 		return 1;
354 	}
355 	printf("[OK]\tArguments are preserved across syscall\n");
356 
357 	return check_regs64();
358 }
359 
run_syscall_twice()360 int run_syscall_twice()
361 {
362 	int exitcode = 0;
363 	long sv;
364 
365 	if (syscall_addr) {
366 		printf("[RUN]\tExecuting 6-argument 32-bit syscall via VDSO\n");
367 		exitcode = run_syscall();
368 	}
369 	sv = syscall_addr;
370 	syscall_addr = (long)&int80;
371 	printf("[RUN]\tExecuting 6-argument 32-bit syscall via INT 80\n");
372 	exitcode += run_syscall();
373 	syscall_addr = sv;
374 	return exitcode;
375 }
376 
ptrace_me()377 void ptrace_me()
378 {
379 	pid_t pid;
380 
381 	fflush(NULL);
382 	pid = fork();
383 	if (pid < 0)
384 		exit(1);
385 	if (pid == 0) {
386 		/* child */
387 		if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) != 0)
388 			exit(0);
389 		raise(SIGSTOP);
390 		return;
391 	}
392 	/* parent */
393 	printf("[RUN]\tRunning tests under ptrace\n");
394 	while (1) {
395 		int status;
396 		pid = waitpid(-1, &status, __WALL);
397 		if (WIFEXITED(status))
398 			exit(WEXITSTATUS(status));
399 		if (WIFSIGNALED(status))
400 			exit(WTERMSIG(status));
401 		if (pid <= 0 || !WIFSTOPPED(status)) /* paranoia */
402 			exit(255);
403 		/*
404 		 * Note: we do not inject sig = WSTOPSIG(status).
405 		 * We probably should, but careful: do not inject SIGTRAP
406 		 * generated by syscall entry/exit stops.
407 		 * That kills the child.
408 		 */
409 		ptrace(PTRACE_SYSCALL, pid, 0L, 0L /*sig*/);
410 	}
411 }
412 
main(int argc,char ** argv,char ** envp)413 int main(int argc, char **argv, char **envp)
414 {
415 	int exitcode = 0;
416 	int cs;
417 	int version, patchlevel;
418 
419 	asm("\n"
420 	"	movl	%%cs, %%eax\n"
421 	: "=a" (cs)
422 	);
423 	kernel_is_64bit = (cs == 0x23);
424 	if (!kernel_is_64bit)
425 		printf("[NOTE]\tNot a 64-bit kernel, won't test R8..R15 leaks\n");
426 
427 	get_kernel_version(&version, &patchlevel);
428 	clobber_ok = version < 4 || (version == 4 && patchlevel < 17);
429 
430 	/* This only works for non-static builds:
431 	 * syscall_addr = dlsym(dlopen("linux-gate.so.1", RTLD_NOW), "__kernel_vsyscall");
432 	 */
433 	syscall_addr = get_syscall(envp);
434 
435 	exitcode += run_syscall_twice();
436 	ptrace_me();
437 	exitcode += run_syscall_twice();
438 
439 	return exitcode;
440 }
441 #endif
442