arch/um/os-Linux/start_up.c

   1 /*
   2  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <stdarg.h>
   9 #include <unistd.h>
  10 #include <errno.h>
  11 #include <fcntl.h>
  12 #include <sched.h>
  13 #include <signal.h>
  14 #include <string.h>
  15 #include <sys/mman.h>
  16 #include <sys/ptrace.h>
  17 #include <sys/stat.h>
  18 #include <sys/wait.h>
  19 #include <asm/unistd.h>
  20 #include "init.h"
  21 #include "kern_constants.h"
  22 #include "os.h"
  23 #include "mem_user.h"
  24 #include "ptrace_user.h"
  25 #include "registers.h"
  26 #include "skas_ptrace.h"
  27
  28 static void ptrace_child(void)
  29 {
  30         int ret;
  31         /* Calling os_getpid because some libcs cached getpid incorrectly */
  32         int pid = os_getpid(), ppid = getppid();
  33         int sc_result;
  34
  35         if (change_sig(SIGWINCH, 0) < 0 ||
  36             ptrace(PTRACE_TRACEME, 0, 0, 0) < 0) {
  37                 perror("ptrace");
  38                 kill(pid, SIGKILL);
  39         }
  40         kill(pid, SIGSTOP);
  41
  42         /*
  43          * This syscall will be intercepted by the parent. Don't call more than
  44          * once, please.
  45          */
  46         sc_result = os_getpid();
  47
  48         if (sc_result == pid)
  49                 /* Nothing modified by the parent, we are running normally. */
  50                 ret = 1;
  51         else if (sc_result == ppid)
  52                 /*
  53                  * Expected in check_ptrace and check_sysemu when they succeed
  54                  * in modifying the stack frame
  55                  */
  56                 ret = 0;
  57         else
  58                 /* Serious trouble! This could be caused by a bug in host 2.6
  59                  * SKAS3/2.6 patch before release -V6, together with a bug in
  60                  * the UML code itself.
  61                  */
  62                 ret = 2;
  63
  64         exit(ret);
  65 }
  66
  67 static void fatal_perror(const char *str)
  68 {
  69         perror(str);
  70         exit(1);
  71 }
  72
  73 static void fatal(char *fmt, ...)
  74 {
  75         va_list list;
  76
  77         va_start(list, fmt);
  78         vfprintf(stderr, fmt, list);
  79         va_end(list);
  80
  81         exit(1);
  82 }
  83
  84 static void non_fatal(char *fmt, ...)
  85 {
  86         va_list list;
  87
  88         va_start(list, fmt);
  89         vfprintf(stderr, fmt, list);
  90         va_end(list);
  91 }
  92
  93 static int start_ptraced_child(void)
  94 {
  95         int pid, n, status;
  96
  97         pid = fork();
  98         if (pid == 0)
  99                 ptrace_child();
 100         else if (pid < 0)
 101                 fatal_perror("start_ptraced_child : fork failed");
 102
 103         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 104         if (n < 0)
 105                 fatal_perror("check_ptrace : waitpid failed");
 106         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 107                 fatal("check_ptrace : expected SIGSTOP, got status = %d",
 108                       status);
 109
 110         return pid;
 111 }
 112
 113 /* When testing for SYSEMU support, if it is one of the broken versions, we
 114  * must just avoid using sysemu, not panic, but only if SYSEMU features are
 115  * broken.
 116  * So only for SYSEMU features we test mustpanic, while normal host features
 117  * must work anyway!
 118  */
 119 static int stop_ptraced_child(int pid, int exitcode, int mustexit)
 120 {
 121         int status, n, ret = 0;
 122
 123         if (ptrace(PTRACE_CONT, pid, 0, 0) < 0)
 124                 fatal_perror("stop_ptraced_child : ptrace failed");
 125         CATCH_EINTR(n = waitpid(pid, &status, 0));
 126         if (!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
 127                 int exit_with = WEXITSTATUS(status);
 128                 if (exit_with == 2)
 129                         non_fatal("check_ptrace : child exited with status 2. "
 130                                   "\nDisabling SYSEMU support.\n");
 131                 non_fatal("check_ptrace : child exited with exitcode %d, while "
 132                           "expecting %d; status 0x%x\n", exit_with,
 133                           exitcode, status);
 134                 if (mustexit)
 135                         exit(1);
 136                 ret = -1;
 137         }
 138
 139         return ret;
 140 }
 141
 142 /* Changed only during early boot */
 143 int ptrace_faultinfo = 1;
 144 int ptrace_ldt = 1;
 145 int proc_mm = 1;
 146 int skas_needs_stub = 0;
 147
 148 static int __init skas0_cmd_param(char *str, int* add)
 149 {
 150         ptrace_faultinfo = proc_mm = 0;
 151         return 0;
 152 }
 153
 154 /* The two __uml_setup would conflict, without this stupid alias. */
 155
 156 static int __init mode_skas0_cmd_param(char *str, int* add)
 157         __attribute__((alias("skas0_cmd_param")));
 158
 159 __uml_setup("skas0", skas0_cmd_param,
 160                 "skas0\n"
 161                 "    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
 162                 "    you specify mode=tt.\n\n");
 163
 164 __uml_setup("mode=skas0", mode_skas0_cmd_param,
 165                 "mode=skas0\n"
 166                 "    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
 167                 "    specify mode=tt. Note that this was recently added - on \n"
 168                 "    older kernels you must use simply \"skas0\".\n\n");
 169
 170 /* Changed only during early boot */
 171 static int force_sysemu_disabled = 0;
 172
 173 static int __init nosysemu_cmd_param(char *str, int* add)
 174 {
 175         force_sysemu_disabled = 1;
 176         return 0;
 177 }
 178
 179 __uml_setup("nosysemu", nosysemu_cmd_param,
 180 "nosysemu\n"
 181 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
 182 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
 183 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
 184 "    To make it working, you need a kernel patch for your host, too.\n"
 185 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
 186 "    information.\n\n");
 187
 188 static void __init check_sysemu(void)
 189 {
 190         unsigned long regs[MAX_REG_NR];
 191         int pid, n, status, count=0;
 192
 193         non_fatal("Checking syscall emulation patch for ptrace...");
 194         sysemu_supported = 0;
 195         pid = start_ptraced_child();
 196
 197         if (ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
 198                 goto fail;
 199
 200         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 201         if (n < 0)
 202                 fatal_perror("check_sysemu : wait failed");
 203         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
 204                 fatal("check_sysemu : expected SIGTRAP, got status = %d",
 205                       status);
 206
 207         if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
 208                 fatal_perror("check_sysemu : PTRACE_GETREGS failed");
 209         if (PT_SYSCALL_NR(regs) != __NR_getpid) {
 210                 non_fatal("check_sysemu got system call number %d, "
 211                           "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
 212                 goto fail;
 213         }
 214
 215         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
 216         if (n < 0) {
 217                 non_fatal("check_sysemu : failed to modify system call "
 218                           "return");
 219                 goto fail;
 220         }
 221
 222         if (stop_ptraced_child(pid, 0, 0) < 0)
 223                 goto fail_stopped;
 224
 225         sysemu_supported = 1;
 226         non_fatal("OK\n");
 227         set_using_sysemu(!force_sysemu_disabled);
 228
 229         non_fatal("Checking advanced syscall emulation patch for ptrace...");
 230         pid = start_ptraced_child();
 231
 232         if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 233                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 234                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 235
 236         while (1) {
 237                 count++;
 238                 if (ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
 239                         goto fail;
 240                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 241                 if (n < 0)
 242                         fatal_perror("check_ptrace : wait failed");
 243
 244                 if (WIFSTOPPED(status) &&
 245                     (WSTOPSIG(status) == (SIGTRAP|0x80))) {
 246                         if (!count)
 247                                 fatal("check_ptrace : SYSEMU_SINGLESTEP "
 248                                       "doesn't singlestep");
 249                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
 250                                    os_getpid());
 251                         if (n < 0)
 252                                 fatal_perror("check_sysemu : failed to modify "
 253                                              "system call return");
 254                         break;
 255                 }
 256                 else if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
 257                         count++;
 258                 else
 259                         fatal("check_ptrace : expected SIGTRAP or "
 260                               "(SIGTRAP | 0x80), got status = %d", status);
 261         }
 262         if (stop_ptraced_child(pid, 0, 0) < 0)
 263                 goto fail_stopped;
 264
 265         sysemu_supported = 2;
 266         non_fatal("OK\n");
 267
 268         if (!force_sysemu_disabled)
 269                 set_using_sysemu(sysemu_supported);
 270         return;
 271
 272 fail:
 273         stop_ptraced_child(pid, 1, 0);
 274 fail_stopped:
 275         non_fatal("missing\n");
 276 }
 277
 278 static void __init check_ptrace(void)
 279 {
 280         int pid, syscall, n, status;
 281
 282         non_fatal("Checking that ptrace can change system call numbers...");
 283         pid = start_ptraced_child();
 284
 285         if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 286                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 287                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 288
 289         while (1) {
 290                 if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
 291                         fatal_perror("check_ptrace : ptrace failed");
 292
 293                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 294                 if (n < 0)
 295                         fatal_perror("check_ptrace : wait failed");
 296
 297                 if (!WIFSTOPPED(status) ||
 298                    (WSTOPSIG(status) != (SIGTRAP | 0x80)))
 299                         fatal("check_ptrace : expected (SIGTRAP|0x80), "
 300                                "got status = %d", status);
 301
 302                 syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
 303                                  0);
 304                 if (syscall == __NR_getpid) {
 305                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 306                                    __NR_getppid);
 307                         if (n < 0)
 308                                 fatal_perror("check_ptrace : failed to modify "
 309                                              "system call");
 310                         break;
 311                 }
 312         }
 313         stop_ptraced_child(pid, 0, 1);
 314         non_fatal("OK\n");
 315         check_sysemu();
 316 }
 317
 318 extern void check_tmpexec(void);
 319
 320 static void __init check_coredump_limit(void)
 321 {
 322         struct rlimit lim;
 323         int err = getrlimit(RLIMIT_CORE, &lim);
 324
 325         if (err) {
 326                 perror("Getting core dump limit");
 327                 return;
 328         }
 329
 330         printf("Core dump limits :\n\tsoft - ");
 331         if (lim.rlim_cur == RLIM_INFINITY)
 332                 printf("NONE\n");
 333         else printf("%lu\n", lim.rlim_cur);
 334
 335         printf("\thard - ");
 336         if (lim.rlim_max == RLIM_INFINITY)
 337                 printf("NONE\n");
 338         else printf("%lu\n", lim.rlim_max);
 339 }
 340
 341 void __init os_early_checks(void)
 342 {
 343         int pid;
 344
 345         /* Print out the core dump limits early */
 346         check_coredump_limit();
 347
 348         check_ptrace();
 349
 350         /* Need to check this early because mmapping happens before the
 351          * kernel is running.
 352          */
 353         check_tmpexec();
 354
 355         pid = start_ptraced_child();
 356         if (init_registers(pid))
 357                 fatal("Failed to initialize default registers");
 358         stop_ptraced_child(pid, 1, 1);
 359 }
 360
 361 static int __init noprocmm_cmd_param(char *str, int* add)
 362 {
 363         proc_mm = 0;
 364         return 0;
 365 }
 366
 367 __uml_setup("noprocmm", noprocmm_cmd_param,
 368 "noprocmm\n"
 369 "    Turns off usage of /proc/mm, even if host supports it.\n"
 370 "    To support /proc/mm, the host needs to be patched using\n"
 371 "    the current skas3 patch.\n\n");
 372
 373 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
 374 {
 375         ptrace_faultinfo = 0;
 376         return 0;
 377 }
 378
 379 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
 380 "noptracefaultinfo\n"
 381 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
 382 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
 383 "    using the current skas3 patch.\n\n");
 384
 385 static int __init noptraceldt_cmd_param(char *str, int* add)
 386 {
 387         ptrace_ldt = 0;
 388         return 0;
 389 }
 390
 391 __uml_setup("noptraceldt", noptraceldt_cmd_param,
 392 "noptraceldt\n"
 393 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
 394 "    To support PTRACE_LDT, the host needs to be patched using\n"
 395 "    the current skas3 patch.\n\n");
 396
 397 static inline void check_skas3_ptrace_faultinfo(void)
 398 {
 399         struct ptrace_faultinfo fi;
 400         int pid, n;
 401
 402         non_fatal("  - PTRACE_FAULTINFO...");
 403         pid = start_ptraced_child();
 404
 405         n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
 406         if (n < 0) {
 407                 ptrace_faultinfo = 0;
 408                 if (errno == EIO)
 409                         non_fatal("not found\n");
 410                 else
 411                         perror("not found");
 412         }
 413         else {
 414                 if (!ptrace_faultinfo)
 415                         non_fatal("found but disabled on command line\n");
 416                 else
 417                         non_fatal("found\n");
 418         }
 419
 420         stop_ptraced_child(pid, 1, 1);
 421 }
 422
 423 static inline void check_skas3_ptrace_ldt(void)
 424 {
 425 #ifdef PTRACE_LDT
 426         int pid, n;
 427         unsigned char ldtbuf[40];
 428         struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
 429                 .func = 2, /* read default ldt */
 430                 .ptr = ldtbuf,
 431                 .bytecount = sizeof(ldtbuf)};
 432
 433         non_fatal("  - PTRACE_LDT...");
 434         pid = start_ptraced_child();
 435
 436         n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
 437         if (n < 0) {
 438                 if (errno == EIO)
 439                         non_fatal("not found\n");
 440                 else {
 441                         perror("not found");
 442                 }
 443                 ptrace_ldt = 0;
 444         }
 445         else {
 446                 if (ptrace_ldt)
 447                         non_fatal("found\n");
 448                 else
 449                         non_fatal("found, but use is disabled\n");
 450         }
 451
 452         stop_ptraced_child(pid, 1, 1);
 453 #else
 454         /* PTRACE_LDT might be disabled via cmdline option.
 455          * We want to override this, else we might use the stub
 456          * without real need
 457          */
 458         ptrace_ldt = 1;
 459 #endif
 460 }
 461
 462 static inline void check_skas3_proc_mm(void)
 463 {
 464         non_fatal("  - /proc/mm...");
 465         if (access("/proc/mm", W_OK) < 0) {
 466                 proc_mm = 0;
 467                 perror("not found");
 468         }
 469         else if (!proc_mm)
 470                 non_fatal("found but disabled on command line\n");
 471         else non_fatal("found\n");
 472 }
 473
 474 void can_do_skas(void)
 475 {
 476         non_fatal("Checking for the skas3 patch in the host:\n");
 477
 478         check_skas3_proc_mm();
 479         check_skas3_ptrace_faultinfo();
 480         check_skas3_ptrace_ldt();
 481
 482         if (!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
 483                 skas_needs_stub = 1;
 484 }
 485
 486 int __init parse_iomem(char *str, int *add)
 487 {
 488         struct iomem_region *new;
 489         struct stat64 buf;
 490         char *file, *driver;
 491         int fd, size;
 492
 493         driver = str;
 494         file = strchr(str,',');
 495         if (file == NULL) {
 496                 fprintf(stderr, "parse_iomem : failed to parse iomem\n");
 497                 goto out;
 498         }
 499         *file = '\0';
 500         file++;
 501         fd = open(file, O_RDWR, 0);
 502         if (fd < 0) {
 503                 perror("parse_iomem - Couldn't open io file");
 504                 goto out;
 505         }
 506
 507         if (fstat64(fd, &buf) < 0) {
 508                 perror("parse_iomem - cannot stat_fd file");
 509                 goto out_close;
 510         }
 511
 512         new = malloc(sizeof(*new));
 513         if (new == NULL) {
 514                 perror("Couldn't allocate iomem_region struct");
 515                 goto out_close;
 516         }
 517
 518         size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
 519
 520         *new = ((struct iomem_region) { .next           = iomem_regions,
 521                                         .driver         = driver,
 522                                         .fd             = fd,
 523                                         .size           = size,
 524                                         .phys           = 0,
 525                                         .virt           = 0 });
 526         iomem_regions = new;
 527         iomem_size += new->size + UM_KERN_PAGE_SIZE;
 528
 529         return 0;
 530  out_close:
 531         close(fd);
 532  out:
 533         return 1;
 534 }