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 #include <sys/uio.h>
  22
  23 #ifdef __x86_64__
  24 # define VSYS(x) (x)
  25 #else
  26 # define VSYS(x) 0
  27 #endif
  28
  29 #ifndef SYS_getcpu
  30 # ifdef __x86_64__
  31 #  define SYS_getcpu 309
  32 # else
  33 #  define SYS_getcpu 318
  34 # endif
  35 #endif
  36
  37 /* max length of lines in /proc/self/maps - anything longer is skipped here */
  38 #define MAPS_LINE_LEN 128
  39
  40 static void sethandler(int sig, void (*handler)(int, siginfo_t *, void *),
  41                        int flags)
  42 {
  43         struct sigaction sa;
  44         memset(&sa, 0, sizeof(sa));
  45         sa.sa_sigaction = handler;
  46         sa.sa_flags = SA_SIGINFO | flags;
  47         sigemptyset(&sa.sa_mask);
  48         if (sigaction(sig, &sa, 0))
  49                 err(1, "sigaction");
  50 }
  51
  52 /* vsyscalls and vDSO */
  53 bool vsyscall_map_r = false, vsyscall_map_x = false;
  54
  55 typedef long (*gtod_t)(struct timeval *tv, struct timezone *tz);
  56 const gtod_t vgtod = (gtod_t)VSYS(0xffffffffff600000);
  57 gtod_t vdso_gtod;
  58
  59 typedef int (*vgettime_t)(clockid_t, struct timespec *);
  60 vgettime_t vdso_gettime;
  61
  62 typedef long (*time_func_t)(time_t *t);
  63 const time_func_t vtime = (time_func_t)VSYS(0xffffffffff600400);
  64 time_func_t vdso_time;
  65
  66 typedef long (*getcpu_t)(unsigned *, unsigned *, void *);
  67 const getcpu_t vgetcpu = (getcpu_t)VSYS(0xffffffffff600800);
  68 getcpu_t vdso_getcpu;
  69
  70 static void init_vdso(void)
  71 {
  72         void *vdso = dlopen("linux-vdso.so.1", RTLD_LAZY | RTLD_LOCAL | RTLD_NOLOAD);
  73         if (!vdso)
  74                 vdso = dlopen("linux-gate.so.1", RTLD_LAZY | RTLD_LOCAL | RTLD_NOLOAD);
  75         if (!vdso) {
  76                 printf("[WARN]\tfailed to find vDSO\n");
  77                 return;
  78         }
  79
  80         vdso_gtod = (gtod_t)dlsym(vdso, "__vdso_gettimeofday");
  81         if (!vdso_gtod)
  82                 printf("[WARN]\tfailed to find gettimeofday in vDSO\n");
  83
  84         vdso_gettime = (vgettime_t)dlsym(vdso, "__vdso_clock_gettime");
  85         if (!vdso_gettime)
  86                 printf("[WARN]\tfailed to find clock_gettime in vDSO\n");
  87
  88         vdso_time = (time_func_t)dlsym(vdso, "__vdso_time");
  89         if (!vdso_time)
  90                 printf("[WARN]\tfailed to find time in vDSO\n");
  91
  92         vdso_getcpu = (getcpu_t)dlsym(vdso, "__vdso_getcpu");
  93         if (!vdso_getcpu) {
  94                 /* getcpu() was never wired up in the 32-bit vDSO. */
  95                 printf("[%s]\tfailed to find getcpu in vDSO\n",
  96                        sizeof(long) == 8 ? "WARN" : "NOTE");
  97         }
  98 }
  99
 100 static int init_vsys(void)
 101 {
 102 #ifdef __x86_64__
 103         int nerrs = 0;
 104         FILE *maps;
 105         char line[MAPS_LINE_LEN];
 106         bool found = false;
 107
 108         maps = fopen("/proc/self/maps", "r");
 109         if (!maps) {
 110                 printf("[WARN]\tCould not open /proc/self/maps -- assuming vsyscall is r-x\n");
 111                 vsyscall_map_r = true;
 112                 return 0;
 113         }
 114
 115         while (fgets(line, MAPS_LINE_LEN, maps)) {
 116                 char r, x;
 117                 void *start, *end;
 118                 char name[MAPS_LINE_LEN];
 119
 120                 /* sscanf() is safe here as strlen(name) >= strlen(line) */
 121                 if (sscanf(line, "%p-%p %c-%cp %*x %*x:%*x %*u %s",
 122                            &start, &end, &r, &x, name) != 5)
 123                         continue;
 124
 125                 if (strcmp(name, "[vsyscall]"))
 126                         continue;
 127
 128                 printf("\tvsyscall map: %s", line);
 129
 130                 if (start != (void *)0xffffffffff600000 ||
 131                     end != (void *)0xffffffffff601000) {
 132                         printf("[FAIL]\taddress range is nonsense\n");
 133                         nerrs++;
 134                 }
 135
 136                 printf("\tvsyscall permissions are %c-%c\n", r, x);
 137                 vsyscall_map_r = (r == 'r');
 138                 vsyscall_map_x = (x == 'x');
 139
 140                 found = true;
 141                 break;
 142         }
 143
 144         fclose(maps);
 145
 146         if (!found) {
 147                 printf("\tno vsyscall map in /proc/self/maps\n");
 148                 vsyscall_map_r = false;
 149                 vsyscall_map_x = false;
 150         }
 151
 152         return nerrs;
 153 #else
 154         return 0;
 155 #endif
 156 }
 157
 158 /* syscalls */
 159 static inline long sys_gtod(struct timeval *tv, struct timezone *tz)
 160 {
 161         return syscall(SYS_gettimeofday, tv, tz);
 162 }
 163
 164 static inline int sys_clock_gettime(clockid_t id, struct timespec *ts)
 165 {
 166         return syscall(SYS_clock_gettime, id, ts);
 167 }
 168
 169 static inline long sys_time(time_t *t)
 170 {
 171         return syscall(SYS_time, t);
 172 }
 173
 174 static inline long sys_getcpu(unsigned * cpu, unsigned * node,
 175                               void* cache)
 176 {
 177         return syscall(SYS_getcpu, cpu, node, cache);
 178 }
 179
 180 static jmp_buf jmpbuf;
 181 static volatile unsigned long segv_err;
 182
 183 static void sigsegv(int sig, siginfo_t *info, void *ctx_void)
 184 {
 185         ucontext_t *ctx = (ucontext_t *)ctx_void;
 186
 187         segv_err =  ctx->uc_mcontext.gregs[REG_ERR];
 188         siglongjmp(jmpbuf, 1);
 189 }
 190
 191 static double tv_diff(const struct timeval *a, const struct timeval *b)
 192 {
 193         return (double)(a->tv_sec - b->tv_sec) +
 194                 (double)((int)a->tv_usec - (int)b->tv_usec) * 1e-6;
 195 }
 196
 197 static int check_gtod(const struct timeval *tv_sys1,
 198                       const struct timeval *tv_sys2,
 199                       const struct timezone *tz_sys,
 200                       const char *which,
 201                       const struct timeval *tv_other,
 202                       const struct timezone *tz_other)
 203 {
 204         int nerrs = 0;
 205         double d1, d2;
 206
 207         if (tz_other && (tz_sys->tz_minuteswest != tz_other->tz_minuteswest || tz_sys->tz_dsttime != tz_other->tz_dsttime)) {
 208                 printf("[FAIL] %s tz mismatch\n", which);
 209                 nerrs++;
 210         }
 211
 212         d1 = tv_diff(tv_other, tv_sys1);
 213         d2 = tv_diff(tv_sys2, tv_other);
 214         printf("\t%s time offsets: %lf %lf\n", which, d1, d2);
 215
 216         if (d1 < 0 || d2 < 0) {
 217                 printf("[FAIL]\t%s time was inconsistent with the syscall\n", which);
 218                 nerrs++;
 219         } else {
 220                 printf("[OK]\t%s gettimeofday()'s timeval was okay\n", which);
 221         }
 222
 223         return nerrs;
 224 }
 225
 226 static int test_gtod(void)
 227 {
 228         struct timeval tv_sys1, tv_sys2, tv_vdso, tv_vsys;
 229         struct timezone tz_sys, tz_vdso, tz_vsys;
 230         long ret_vdso = -1;
 231         long ret_vsys = -1;
 232         int nerrs = 0;
 233
 234         printf("[RUN]\ttest gettimeofday()\n");
 235
 236         if (sys_gtod(&tv_sys1, &tz_sys) != 0)
 237                 err(1, "syscall gettimeofday");
 238         if (vdso_gtod)
 239                 ret_vdso = vdso_gtod(&tv_vdso, &tz_vdso);
 240         if (vsyscall_map_x)
 241                 ret_vsys = vgtod(&tv_vsys, &tz_vsys);
 242         if (sys_gtod(&tv_sys2, &tz_sys) != 0)
 243                 err(1, "syscall gettimeofday");
 244
 245         if (vdso_gtod) {
 246                 if (ret_vdso == 0) {
 247                         nerrs += check_gtod(&tv_sys1, &tv_sys2, &tz_sys, "vDSO", &tv_vdso, &tz_vdso);
 248                 } else {
 249                         printf("[FAIL]\tvDSO gettimeofday() failed: %ld\n", ret_vdso);
 250                         nerrs++;
 251                 }
 252         }
 253
 254         if (vsyscall_map_x) {
 255                 if (ret_vsys == 0) {
 256                         nerrs += check_gtod(&tv_sys1, &tv_sys2, &tz_sys, "vsyscall", &tv_vsys, &tz_vsys);
 257                 } else {
 258                         printf("[FAIL]\tvsys gettimeofday() failed: %ld\n", ret_vsys);
 259                         nerrs++;
 260                 }
 261         }
 262
 263         return nerrs;
 264 }
 265
 266 static int test_time(void) {
 267         int nerrs = 0;
 268
 269         printf("[RUN]\ttest time()\n");
 270         long t_sys1, t_sys2, t_vdso = 0, t_vsys = 0;
 271         long t2_sys1 = -1, t2_sys2 = -1, t2_vdso = -1, t2_vsys = -1;
 272         t_sys1 = sys_time(&t2_sys1);
 273         if (vdso_time)
 274                 t_vdso = vdso_time(&t2_vdso);
 275         if (vsyscall_map_x)
 276                 t_vsys = vtime(&t2_vsys);
 277         t_sys2 = sys_time(&t2_sys2);
 278         if (t_sys1 < 0 || t_sys1 != t2_sys1 || t_sys2 < 0 || t_sys2 != t2_sys2) {
 279                 printf("[FAIL]\tsyscall failed (ret1:%ld output1:%ld ret2:%ld output2:%ld)\n", t_sys1, t2_sys1, t_sys2, t2_sys2);
 280                 nerrs++;
 281                 return nerrs;
 282         }
 283
 284         if (vdso_time) {
 285                 if (t_vdso < 0 || t_vdso != t2_vdso) {
 286                         printf("[FAIL]\tvDSO failed (ret:%ld output:%ld)\n", t_vdso, t2_vdso);
 287                         nerrs++;
 288                 } else if (t_vdso < t_sys1 || t_vdso > t_sys2) {
 289                         printf("[FAIL]\tvDSO returned the wrong time (%ld %ld %ld)\n", t_sys1, t_vdso, t_sys2);
 290                         nerrs++;
 291                 } else {
 292                         printf("[OK]\tvDSO time() is okay\n");
 293                 }
 294         }
 295
 296         if (vsyscall_map_x) {
 297                 if (t_vsys < 0 || t_vsys != t2_vsys) {
 298                         printf("[FAIL]\tvsyscall failed (ret:%ld output:%ld)\n", t_vsys, t2_vsys);
 299                         nerrs++;
 300                 } else if (t_vsys < t_sys1 || t_vsys > t_sys2) {
 301                         printf("[FAIL]\tvsyscall returned the wrong time (%ld %ld %ld)\n", t_sys1, t_vsys, t_sys2);
 302                         nerrs++;
 303                 } else {
 304                         printf("[OK]\tvsyscall time() is okay\n");
 305                 }
 306         }
 307
 308         return nerrs;
 309 }
 310
 311 static int test_getcpu(int cpu)
 312 {
 313         int nerrs = 0;
 314         long ret_sys, ret_vdso = -1, ret_vsys = -1;
 315
 316         printf("[RUN]\tgetcpu() on CPU %d\n", cpu);
 317
 318         cpu_set_t cpuset;
 319         CPU_ZERO(&cpuset);
 320         CPU_SET(cpu, &cpuset);
 321         if (sched_setaffinity(0, sizeof(cpuset), &cpuset) != 0) {
 322                 printf("[SKIP]\tfailed to force CPU %d\n", cpu);
 323                 return nerrs;
 324         }
 325
 326         unsigned cpu_sys, cpu_vdso, cpu_vsys, node_sys, node_vdso, node_vsys;
 327         unsigned node = 0;
 328         bool have_node = false;
 329         ret_sys = sys_getcpu(&cpu_sys, &node_sys, 0);
 330         if (vdso_getcpu)
 331                 ret_vdso = vdso_getcpu(&cpu_vdso, &node_vdso, 0);
 332         if (vsyscall_map_x)
 333                 ret_vsys = vgetcpu(&cpu_vsys, &node_vsys, 0);
 334
 335         if (ret_sys == 0) {
 336                 if (cpu_sys != cpu) {
 337                         printf("[FAIL]\tsyscall reported CPU %hu but should be %d\n", cpu_sys, cpu);
 338                         nerrs++;
 339                 }
 340
 341                 have_node = true;
 342                 node = node_sys;
 343         }
 344
 345         if (vdso_getcpu) {
 346                 if (ret_vdso) {
 347                         printf("[FAIL]\tvDSO getcpu() failed\n");
 348                         nerrs++;
 349                 } else {
 350                         if (!have_node) {
 351                                 have_node = true;
 352                                 node = node_vdso;
 353                         }
 354
 355                         if (cpu_vdso != cpu) {
 356                                 printf("[FAIL]\tvDSO reported CPU %hu but should be %d\n", cpu_vdso, cpu);
 357                                 nerrs++;
 358                         } else {
 359                                 printf("[OK]\tvDSO reported correct CPU\n");
 360                         }
 361
 362                         if (node_vdso != node) {
 363                                 printf("[FAIL]\tvDSO reported node %hu but should be %hu\n", node_vdso, node);
 364                                 nerrs++;
 365                         } else {
 366                                 printf("[OK]\tvDSO reported correct node\n");
 367                         }
 368                 }
 369         }
 370
 371         if (vsyscall_map_x) {
 372                 if (ret_vsys) {
 373                         printf("[FAIL]\tvsyscall getcpu() failed\n");
 374                         nerrs++;
 375                 } else {
 376                         if (!have_node) {
 377                                 have_node = true;
 378                                 node = node_vsys;
 379                         }
 380
 381                         if (cpu_vsys != cpu) {
 382                                 printf("[FAIL]\tvsyscall reported CPU %hu but should be %d\n", cpu_vsys, cpu);
 383                                 nerrs++;
 384                         } else {
 385                                 printf("[OK]\tvsyscall reported correct CPU\n");
 386                         }
 387
 388                         if (node_vsys != node) {
 389                                 printf("[FAIL]\tvsyscall reported node %hu but should be %hu\n", node_vsys, node);
 390                                 nerrs++;
 391                         } else {
 392                                 printf("[OK]\tvsyscall reported correct node\n");
 393                         }
 394                 }
 395         }
 396
 397         return nerrs;
 398 }
 399
 400 static int test_vsys_r(void)
 401 {
 402 #ifdef __x86_64__
 403         printf("[RUN]\tChecking read access to the vsyscall page\n");
 404         bool can_read;
 405         if (sigsetjmp(jmpbuf, 1) == 0) {
 406                 *(volatile int *)0xffffffffff600000;
 407                 can_read = true;
 408         } else {
 409                 can_read = false;
 410         }
 411
 412         if (can_read && !vsyscall_map_r) {
 413                 printf("[FAIL]\tWe have read access, but we shouldn't\n");
 414                 return 1;
 415         } else if (!can_read && vsyscall_map_r) {
 416                 printf("[FAIL]\tWe don't have read access, but we should\n");
 417                 return 1;
 418         } else if (can_read) {
 419                 printf("[OK]\tWe have read access\n");
 420         } else {
 421                 printf("[OK]\tWe do not have read access: #PF(0x%lx)\n",
 422                        segv_err);
 423         }
 424 #endif
 425
 426         return 0;
 427 }
 428
 429 static int test_vsys_x(void)
 430 {
 431 #ifdef __x86_64__
 432         if (vsyscall_map_x) {
 433                 /* We already tested this adequately. */
 434                 return 0;
 435         }
 436
 437         printf("[RUN]\tMake sure that vsyscalls really page fault\n");
 438
 439         bool can_exec;
 440         if (sigsetjmp(jmpbuf, 1) == 0) {
 441                 vgtod(NULL, NULL);
 442                 can_exec = true;
 443         } else {
 444                 can_exec = false;
 445         }
 446
 447         if (can_exec) {
 448                 printf("[FAIL]\tExecuting the vsyscall did not page fault\n");
 449                 return 1;
 450         } else if (segv_err & (1 << 4)) { /* INSTR */
 451                 printf("[OK]\tExecuting the vsyscall page failed: #PF(0x%lx)\n",
 452                        segv_err);
 453         } else {
 454                 printf("[FAIL]\tExecution failed with the wrong error: #PF(0x%lx)\n",
 455                        segv_err);
 456                 return 1;
 457         }
 458 #endif
 459
 460         return 0;
 461 }
 462
 463 static int test_process_vm_readv(void)
 464 {
 465 #ifdef __x86_64__
 466         char buf[4096];
 467         struct iovec local, remote;
 468         int ret;
 469
 470         printf("[RUN]\tprocess_vm_readv() from vsyscall page\n");
 471
 472         local.iov_base = buf;
 473         local.iov_len = 4096;
 474         remote.iov_base = (void *)0xffffffffff600000;
 475         remote.iov_len = 4096;
 476         ret = process_vm_readv(getpid(), &local, 1, &remote, 1, 0);
 477         if (ret != 4096) {
 478                 printf("[OK]\tprocess_vm_readv() failed (ret = %d, errno = %d)\n", ret, errno);
 479                 return 0;
 480         }
 481
 482         if (vsyscall_map_r) {
 483                 if (!memcmp(buf, (const void *)0xffffffffff600000, 4096)) {
 484                         printf("[OK]\tIt worked and read correct data\n");
 485                 } else {
 486                         printf("[FAIL]\tIt worked but returned incorrect data\n");
 487                         return 1;
 488                 }
 489         }
 490 #endif
 491
 492         return 0;
 493 }
 494
 495 #ifdef __x86_64__
 496 #define X86_EFLAGS_TF (1UL << 8)
 497 static volatile sig_atomic_t num_vsyscall_traps;
 498
 499 static unsigned long get_eflags(void)
 500 {
 501         unsigned long eflags;
 502         asm volatile ("pushfq\n\tpopq %0" : "=rm" (eflags));
 503         return eflags;
 504 }
 505
 506 static void set_eflags(unsigned long eflags)
 507 {
 508         asm volatile ("pushq %0\n\tpopfq" : : "rm" (eflags) : "flags");
 509 }
 510
 511 static void sigtrap(int sig, siginfo_t *info, void *ctx_void)
 512 {
 513         ucontext_t *ctx = (ucontext_t *)ctx_void;
 514         unsigned long ip = ctx->uc_mcontext.gregs[REG_RIP];
 515
 516         if (((ip ^ 0xffffffffff600000UL) & ~0xfffUL) == 0)
 517                 num_vsyscall_traps++;
 518 }
 519
 520 static int test_emulation(void)
 521 {
 522         time_t tmp;
 523         bool is_native;
 524
 525         if (!vsyscall_map_x)
 526                 return 0;
 527
 528         printf("[RUN]\tchecking that vsyscalls are emulated\n");
 529         sethandler(SIGTRAP, sigtrap, 0);
 530         set_eflags(get_eflags() | X86_EFLAGS_TF);
 531         vtime(&tmp);
 532         set_eflags(get_eflags() & ~X86_EFLAGS_TF);
 533
 534         /*
 535          * If vsyscalls are emulated, we expect a single trap in the
 536          * vsyscall page -- the call instruction will trap with RIP
 537          * pointing to the entry point before emulation takes over.
 538          * In native mode, we expect two traps, since whatever code
 539          * the vsyscall page contains will be more than just a ret
 540          * instruction.
 541          */
 542         is_native = (num_vsyscall_traps > 1);
 543
 544         printf("[%s]\tvsyscalls are %s (%d instructions in vsyscall page)\n",
 545                (is_native ? "FAIL" : "OK"),
 546                (is_native ? "native" : "emulated"),
 547                (int)num_vsyscall_traps);
 548
 549         return is_native;
 550 }
 551 #endif
 552
 553 int main(int argc, char **argv)
 554 {
 555         int nerrs = 0;
 556
 557         init_vdso();
 558         nerrs += init_vsys();
 559
 560         nerrs += test_gtod();
 561         nerrs += test_time();
 562         nerrs += test_getcpu(0);
 563         nerrs += test_getcpu(1);
 564
 565         sethandler(SIGSEGV, sigsegv, 0);
 566         nerrs += test_vsys_r();
 567         nerrs += test_vsys_x();
 568
 569         nerrs += test_process_vm_readv();
 570
 571 #ifdef __x86_64__
 572         nerrs += test_emulation();
 573 #endif
 574
 575         return nerrs ? 1 : 0;
 576 }