tools/testing/selftests/x86/test_syscall_vdso.c

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