tools/testing/selftests/x86/test_vsyscall.c

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2
   3 #define _GNU_SOURCE
   4
   5 #include <stdio.h>
   6 #include <sys/time.h>
   7 #include <time.h>
   8 #include <stdlib.h>
   9 #include <sys/syscall.h>
  10 #include <unistd.h>
  11 #include <dlfcn.h>
  12 #include <string.h>
  13 #include <inttypes.h>
  14 #include <signal.h>
  15 #include <sys/ucontext.h>
  16 #include <errno.h>
  17 #include <err.h>
  18 #include <sched.h>
  19 #include <stdbool.h>
  20 #include <setjmp.h>
  21
  22 #ifdef __x86_64__
  23 # define VSYS(x) (x)
  24 #else
  25 # define VSYS(x) 0
  26 #endif
  27
  28 #ifndef SYS_getcpu
  29 # ifdef __x86_64__
  30 #  define SYS_getcpu 309
  31 # else
  32 #  define SYS_getcpu 318
  33 # endif
  34 #endif
  35
  36 /* max length of lines in /proc/self/maps - anything longer is skipped here */
  37 #define MAPS_LINE_LEN 128
  38
  39 static void sethandler(int sig, void (*handler)(int, siginfo_t *, void *),
  40                        int flags)
  41 {
  42         struct sigaction sa;
  43         memset(&sa, 0, sizeof(sa));
  44         sa.sa_sigaction = handler;
  45         sa.sa_flags = SA_SIGINFO | flags;
  46         sigemptyset(&sa.sa_mask);
  47         if (sigaction(sig, &sa, 0))
  48                 err(1, "sigaction");
  49 }
  50
  51 /* vsyscalls and vDSO */
  52 bool should_read_vsyscall = false;
  53
  54 typedef long (*gtod_t)(struct timeval *tv, struct timezone *tz);
  55 gtod_t vgtod = (gtod_t)VSYS(0xffffffffff600000);
  56 gtod_t vdso_gtod;
  57
  58 typedef int (*vgettime_t)(clockid_t, struct timespec *);
  59 vgettime_t vdso_gettime;
  60
  61 typedef long (*time_func_t)(time_t *t);
  62 time_func_t vtime = (time_func_t)VSYS(0xffffffffff600400);
  63 time_func_t vdso_time;
  64
  65 typedef long (*getcpu_t)(unsigned *, unsigned *, void *);
  66 getcpu_t vgetcpu = (getcpu_t)VSYS(0xffffffffff600800);
  67 getcpu_t vdso_getcpu;
  68
  69 static void init_vdso(void)
  70 {
  71         void *vdso = dlopen("linux-vdso.so.1", RTLD_LAZY | RTLD_LOCAL | RTLD_NOLOAD);
  72         if (!vdso)
  73                 vdso = dlopen("linux-gate.so.1", RTLD_LAZY | RTLD_LOCAL | RTLD_NOLOAD);
  74         if (!vdso) {
  75                 printf("[WARN]\tfailed to find vDSO\n");
  76                 return;
  77         }
  78
  79         vdso_gtod = (gtod_t)dlsym(vdso, "__vdso_gettimeofday");
  80         if (!vdso_gtod)
  81                 printf("[WARN]\tfailed to find gettimeofday in vDSO\n");
  82
  83         vdso_gettime = (vgettime_t)dlsym(vdso, "__vdso_clock_gettime");
  84         if (!vdso_gettime)
  85                 printf("[WARN]\tfailed to find clock_gettime in vDSO\n");
  86
  87         vdso_time = (time_func_t)dlsym(vdso, "__vdso_time");
  88         if (!vdso_time)
  89                 printf("[WARN]\tfailed to find time in vDSO\n");
  90
  91         vdso_getcpu = (getcpu_t)dlsym(vdso, "__vdso_getcpu");
  92         if (!vdso_getcpu) {
  93                 /* getcpu() was never wired up in the 32-bit vDSO. */
  94                 printf("[%s]\tfailed to find getcpu in vDSO\n",
  95                        sizeof(long) == 8 ? "WARN" : "NOTE");
  96         }
  97 }
  98
  99 static int init_vsys(void)
 100 {
 101 #ifdef __x86_64__
 102         int nerrs = 0;
 103         FILE *maps;
 104         char line[MAPS_LINE_LEN];
 105         bool found = false;
 106
 107         maps = fopen("/proc/self/maps", "r");
 108         if (!maps) {
 109                 printf("[WARN]\tCould not open /proc/self/maps -- assuming vsyscall is r-x\n");
 110                 should_read_vsyscall = true;
 111                 return 0;
 112         }
 113
 114         while (fgets(line, MAPS_LINE_LEN, maps)) {
 115                 char r, x;
 116                 void *start, *end;
 117                 char name[MAPS_LINE_LEN];
 118
 119                 /* sscanf() is safe here as strlen(name) >= strlen(line) */
 120                 if (sscanf(line, "%p-%p %c-%cp %*x %*x:%*x %*u %s",
 121                            &start, &end, &r, &x, name) != 5)
 122                         continue;
 123
 124                 if (strcmp(name, "[vsyscall]"))
 125                         continue;
 126
 127                 printf("\tvsyscall map: %s", line);
 128
 129                 if (start != (void *)0xffffffffff600000 ||
 130                     end != (void *)0xffffffffff601000) {
 131                         printf("[FAIL]\taddress range is nonsense\n");
 132                         nerrs++;
 133                 }
 134
 135                 printf("\tvsyscall permissions are %c-%c\n", r, x);
 136                 should_read_vsyscall = (r == 'r');
 137                 if (x != 'x') {
 138                         vgtod = NULL;
 139                         vtime = NULL;
 140                         vgetcpu = NULL;
 141                 }
 142
 143                 found = true;
 144                 break;
 145         }
 146
 147         fclose(maps);
 148
 149         if (!found) {
 150                 printf("\tno vsyscall map in /proc/self/maps\n");
 151                 should_read_vsyscall = false;
 152                 vgtod = NULL;
 153                 vtime = NULL;
 154                 vgetcpu = NULL;
 155         }
 156
 157         return nerrs;
 158 #else
 159         return 0;
 160 #endif
 161 }
 162
 163 /* syscalls */
 164 static inline long sys_gtod(struct timeval *tv, struct timezone *tz)
 165 {
 166         return syscall(SYS_gettimeofday, tv, tz);
 167 }
 168
 169 static inline int sys_clock_gettime(clockid_t id, struct timespec *ts)
 170 {
 171         return syscall(SYS_clock_gettime, id, ts);
 172 }
 173
 174 static inline long sys_time(time_t *t)
 175 {
 176         return syscall(SYS_time, t);
 177 }
 178
 179 static inline long sys_getcpu(unsigned * cpu, unsigned * node,
 180                               void* cache)
 181 {
 182         return syscall(SYS_getcpu, cpu, node, cache);
 183 }
 184
 185 static jmp_buf jmpbuf;
 186
 187 static void sigsegv(int sig, siginfo_t *info, void *ctx_void)
 188 {
 189         siglongjmp(jmpbuf, 1);
 190 }
 191
 192 static double tv_diff(const struct timeval *a, const struct timeval *b)
 193 {
 194         return (double)(a->tv_sec - b->tv_sec) +
 195                 (double)((int)a->tv_usec - (int)b->tv_usec) * 1e-6;
 196 }
 197
 198 static int check_gtod(const struct timeval *tv_sys1,
 199                       const struct timeval *tv_sys2,
 200                       const struct timezone *tz_sys,
 201                       const char *which,
 202                       const struct timeval *tv_other,
 203                       const struct timezone *tz_other)
 204 {
 205         int nerrs = 0;
 206         double d1, d2;
 207
 208         if (tz_other && (tz_sys->tz_minuteswest != tz_other->tz_minuteswest || tz_sys->tz_dsttime != tz_other->tz_dsttime)) {
 209                 printf("[FAIL] %s tz mismatch\n", which);
 210                 nerrs++;
 211         }
 212
 213         d1 = tv_diff(tv_other, tv_sys1);
 214         d2 = tv_diff(tv_sys2, tv_other);
 215         printf("\t%s time offsets: %lf %lf\n", which, d1, d2);
 216
 217         if (d1 < 0 || d2 < 0) {
 218                 printf("[FAIL]\t%s time was inconsistent with the syscall\n", which);
 219                 nerrs++;
 220         } else {
 221                 printf("[OK]\t%s gettimeofday()'s timeval was okay\n", which);
 222         }
 223
 224         return nerrs;
 225 }
 226
 227 static int test_gtod(void)
 228 {
 229         struct timeval tv_sys1, tv_sys2, tv_vdso, tv_vsys;
 230         struct timezone tz_sys, tz_vdso, tz_vsys;
 231         long ret_vdso = -1;
 232         long ret_vsys = -1;
 233         int nerrs = 0;
 234
 235         printf("[RUN]\ttest gettimeofday()\n");
 236
 237         if (sys_gtod(&tv_sys1, &tz_sys) != 0)
 238                 err(1, "syscall gettimeofday");
 239         if (vdso_gtod)
 240                 ret_vdso = vdso_gtod(&tv_vdso, &tz_vdso);
 241         if (vgtod)
 242                 ret_vsys = vgtod(&tv_vsys, &tz_vsys);
 243         if (sys_gtod(&tv_sys2, &tz_sys) != 0)
 244                 err(1, "syscall gettimeofday");
 245
 246         if (vdso_gtod) {
 247                 if (ret_vdso == 0) {
 248                         nerrs += check_gtod(&tv_sys1, &tv_sys2, &tz_sys, "vDSO", &tv_vdso, &tz_vdso);
 249                 } else {
 250                         printf("[FAIL]\tvDSO gettimeofday() failed: %ld\n", ret_vdso);
 251                         nerrs++;
 252                 }
 253         }
 254
 255         if (vgtod) {
 256                 if (ret_vsys == 0) {
 257                         nerrs += check_gtod(&tv_sys1, &tv_sys2, &tz_sys, "vsyscall", &tv_vsys, &tz_vsys);
 258                 } else {
 259                         printf("[FAIL]\tvsys gettimeofday() failed: %ld\n", ret_vsys);
 260                         nerrs++;
 261                 }
 262         }
 263
 264         return nerrs;
 265 }
 266
 267 static int test_time(void) {
 268         int nerrs = 0;
 269
 270         printf("[RUN]\ttest time()\n");
 271         long t_sys1, t_sys2, t_vdso = 0, t_vsys = 0;
 272         long t2_sys1 = -1, t2_sys2 = -1, t2_vdso = -1, t2_vsys = -1;
 273         t_sys1 = sys_time(&t2_sys1);
 274         if (vdso_time)
 275                 t_vdso = vdso_time(&t2_vdso);
 276         if (vtime)
 277                 t_vsys = vtime(&t2_vsys);
 278         t_sys2 = sys_time(&t2_sys2);
 279         if (t_sys1 < 0 || t_sys1 != t2_sys1 || t_sys2 < 0 || t_sys2 != t2_sys2) {
 280                 printf("[FAIL]\tsyscall failed (ret1:%ld output1:%ld ret2:%ld output2:%ld)\n", t_sys1, t2_sys1, t_sys2, t2_sys2);
 281                 nerrs++;
 282                 return nerrs;
 283         }
 284
 285         if (vdso_time) {
 286                 if (t_vdso < 0 || t_vdso != t2_vdso) {
 287                         printf("[FAIL]\tvDSO failed (ret:%ld output:%ld)\n", t_vdso, t2_vdso);
 288                         nerrs++;
 289                 } else if (t_vdso < t_sys1 || t_vdso > t_sys2) {
 290                         printf("[FAIL]\tvDSO returned the wrong time (%ld %ld %ld)\n", t_sys1, t_vdso, t_sys2);
 291                         nerrs++;
 292                 } else {
 293                         printf("[OK]\tvDSO time() is okay\n");
 294                 }
 295         }
 296
 297         if (vtime) {
 298                 if (t_vsys < 0 || t_vsys != t2_vsys) {
 299                         printf("[FAIL]\tvsyscall failed (ret:%ld output:%ld)\n", t_vsys, t2_vsys);
 300                         nerrs++;
 301                 } else if (t_vsys < t_sys1 || t_vsys > t_sys2) {
 302                         printf("[FAIL]\tvsyscall returned the wrong time (%ld %ld %ld)\n", t_sys1, t_vsys, t_sys2);
 303                         nerrs++;
 304                 } else {
 305                         printf("[OK]\tvsyscall time() is okay\n");
 306                 }
 307         }
 308
 309         return nerrs;
 310 }
 311
 312 static int test_getcpu(int cpu)
 313 {
 314         int nerrs = 0;
 315         long ret_sys, ret_vdso = -1, ret_vsys = -1;
 316
 317         printf("[RUN]\tgetcpu() on CPU %d\n", cpu);
 318
 319         cpu_set_t cpuset;
 320         CPU_ZERO(&cpuset);
 321         CPU_SET(cpu, &cpuset);
 322         if (sched_setaffinity(0, sizeof(cpuset), &cpuset) != 0) {
 323                 printf("[SKIP]\tfailed to force CPU %d\n", cpu);
 324                 return nerrs;
 325         }
 326
 327         unsigned cpu_sys, cpu_vdso, cpu_vsys, node_sys, node_vdso, node_vsys;
 328         unsigned node = 0;
 329         bool have_node = false;
 330         ret_sys = sys_getcpu(&cpu_sys, &node_sys, 0);
 331         if (vdso_getcpu)
 332                 ret_vdso = vdso_getcpu(&cpu_vdso, &node_vdso, 0);
 333         if (vgetcpu)
 334                 ret_vsys = vgetcpu(&cpu_vsys, &node_vsys, 0);
 335
 336         if (ret_sys == 0) {
 337                 if (cpu_sys != cpu) {
 338                         printf("[FAIL]\tsyscall reported CPU %hu but should be %d\n", cpu_sys, cpu);
 339                         nerrs++;
 340                 }
 341
 342                 have_node = true;
 343                 node = node_sys;
 344         }
 345
 346         if (vdso_getcpu) {
 347                 if (ret_vdso) {
 348                         printf("[FAIL]\tvDSO getcpu() failed\n");
 349                         nerrs++;
 350                 } else {
 351                         if (!have_node) {
 352                                 have_node = true;
 353                                 node = node_vdso;
 354                         }
 355
 356                         if (cpu_vdso != cpu) {
 357                                 printf("[FAIL]\tvDSO reported CPU %hu but should be %d\n", cpu_vdso, cpu);
 358                                 nerrs++;
 359                         } else {
 360                                 printf("[OK]\tvDSO reported correct CPU\n");
 361                         }
 362
 363                         if (node_vdso != node) {
 364                                 printf("[FAIL]\tvDSO reported node %hu but should be %hu\n", node_vdso, node);
 365                                 nerrs++;
 366                         } else {
 367                                 printf("[OK]\tvDSO reported correct node\n");
 368                         }
 369                 }
 370         }
 371
 372         if (vgetcpu) {
 373                 if (ret_vsys) {
 374                         printf("[FAIL]\tvsyscall getcpu() failed\n");
 375                         nerrs++;
 376                 } else {
 377                         if (!have_node) {
 378                                 have_node = true;
 379                                 node = node_vsys;
 380                         }
 381
 382                         if (cpu_vsys != cpu) {
 383                                 printf("[FAIL]\tvsyscall reported CPU %hu but should be %d\n", cpu_vsys, cpu);
 384                                 nerrs++;
 385                         } else {
 386                                 printf("[OK]\tvsyscall reported correct CPU\n");
 387                         }
 388
 389                         if (node_vsys != node) {
 390                                 printf("[FAIL]\tvsyscall reported node %hu but should be %hu\n", node_vsys, node);
 391                                 nerrs++;
 392                         } else {
 393                                 printf("[OK]\tvsyscall reported correct node\n");
 394                         }
 395                 }
 396         }
 397
 398         return nerrs;
 399 }
 400
 401 static int test_vsys_r(void)
 402 {
 403 #ifdef __x86_64__
 404         printf("[RUN]\tChecking read access to the vsyscall page\n");
 405         bool can_read;
 406         if (sigsetjmp(jmpbuf, 1) == 0) {
 407                 *(volatile int *)0xffffffffff600000;
 408                 can_read = true;
 409         } else {
 410                 can_read = false;
 411         }
 412
 413         if (can_read && !should_read_vsyscall) {
 414                 printf("[FAIL]\tWe have read access, but we shouldn't\n");
 415                 return 1;
 416         } else if (!can_read && should_read_vsyscall) {
 417                 printf("[FAIL]\tWe don't have read access, but we should\n");
 418                 return 1;
 419         } else {
 420                 printf("[OK]\tgot expected result\n");
 421         }
 422 #endif
 423
 424         return 0;
 425 }
 426
 427
 428 #ifdef __x86_64__
 429 #define X86_EFLAGS_TF (1UL << 8)
 430 static volatile sig_atomic_t num_vsyscall_traps;
 431
 432 static unsigned long get_eflags(void)
 433 {
 434         unsigned long eflags;
 435         asm volatile ("pushfq\n\tpopq %0" : "=rm" (eflags));
 436         return eflags;
 437 }
 438
 439 static void set_eflags(unsigned long eflags)
 440 {
 441         asm volatile ("pushq %0\n\tpopfq" : : "rm" (eflags) : "flags");
 442 }
 443
 444 static void sigtrap(int sig, siginfo_t *info, void *ctx_void)
 445 {
 446         ucontext_t *ctx = (ucontext_t *)ctx_void;
 447         unsigned long ip = ctx->uc_mcontext.gregs[REG_RIP];
 448
 449         if (((ip ^ 0xffffffffff600000UL) & ~0xfffUL) == 0)
 450                 num_vsyscall_traps++;
 451 }
 452
 453 static int test_emulation(void)
 454 {
 455         time_t tmp;
 456         bool is_native;
 457
 458         if (!vtime)
 459                 return 0;
 460
 461         printf("[RUN]\tchecking that vsyscalls are emulated\n");
 462         sethandler(SIGTRAP, sigtrap, 0);
 463         set_eflags(get_eflags() | X86_EFLAGS_TF);
 464         vtime(&tmp);
 465         set_eflags(get_eflags() & ~X86_EFLAGS_TF);
 466
 467         /*
 468          * If vsyscalls are emulated, we expect a single trap in the
 469          * vsyscall page -- the call instruction will trap with RIP
 470          * pointing to the entry point before emulation takes over.
 471          * In native mode, we expect two traps, since whatever code
 472          * the vsyscall page contains will be more than just a ret
 473          * instruction.
 474          */
 475         is_native = (num_vsyscall_traps > 1);
 476
 477         printf("[%s]\tvsyscalls are %s (%d instructions in vsyscall page)\n",
 478                (is_native ? "FAIL" : "OK"),
 479                (is_native ? "native" : "emulated"),
 480                (int)num_vsyscall_traps);
 481
 482         return is_native;
 483 }
 484 #endif
 485
 486 int main(int argc, char **argv)
 487 {
 488         int nerrs = 0;
 489
 490         init_vdso();
 491         nerrs += init_vsys();
 492
 493         nerrs += test_gtod();
 494         nerrs += test_time();
 495         nerrs += test_getcpu(0);
 496         nerrs += test_getcpu(1);
 497
 498         sethandler(SIGSEGV, sigsegv, 0);
 499         nerrs += test_vsys_r();
 500
 501 #ifdef __x86_64__
 502         nerrs += test_emulation();
 503 #endif
 504
 505         return nerrs ? 1 : 0;
 506 }