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 int 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         change_sig(SIGWINCH, 0);
  36         if (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         vprintf(fmt, list);
  79         va_end(list);
  80         fflush(stdout);
  81
  82         exit(1);
  83 }
  84
  85 static void non_fatal(char *fmt, ...)
  86 {
  87         va_list list;
  88
  89         va_start(list, fmt);
  90         vprintf(fmt, list);
  91         va_end(list);
  92         fflush(stdout);
  93 }
  94
  95 static int start_ptraced_child(void)
  96 {
  97         int pid, n, status;
  98
  99         pid = fork();
 100         if (pid == 0)
 101                 ptrace_child();
 102         else if (pid < 0)
 103                 fatal_perror("start_ptraced_child : fork failed");
 104
 105         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 106         if (n < 0)
 107                 fatal_perror("check_ptrace : waitpid failed");
 108         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 109                 fatal("check_ptrace : expected SIGSTOP, got status = %d",
 110                       status);
 111
 112         return pid;
 113 }
 114
 115 /* When testing for SYSEMU support, if it is one of the broken versions, we
 116  * must just avoid using sysemu, not panic, but only if SYSEMU features are
 117  * broken.
 118  * So only for SYSEMU features we test mustpanic, while normal host features
 119  * must work anyway!
 120  */
 121 static int stop_ptraced_child(int pid, int exitcode, int mustexit)
 122 {
 123         int status, n, ret = 0;
 124
 125         if (ptrace(PTRACE_CONT, pid, 0, 0) < 0)
 126                 fatal_perror("stop_ptraced_child : ptrace failed");
 127         CATCH_EINTR(n = waitpid(pid, &status, 0));
 128         if (!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
 129                 int exit_with = WEXITSTATUS(status);
 130                 if (exit_with == 2)
 131                         non_fatal("check_ptrace : child exited with status 2. "
 132                                   "\nDisabling SYSEMU support.\n");
 133                 non_fatal("check_ptrace : child exited with exitcode %d, while "
 134                           "expecting %d; status 0x%x\n", exit_with,
 135                           exitcode, status);
 136                 if (mustexit)
 137                         exit(1);
 138                 ret = -1;
 139         }
 140
 141         return ret;
 142 }
 143
 144 /* Changed only during early boot */
 145 int ptrace_faultinfo = 1;
 146 int ptrace_ldt = 1;
 147 int proc_mm = 1;
 148 int skas_needs_stub = 0;
 149
 150 static int __init skas0_cmd_param(char *str, int* add)
 151 {
 152         ptrace_faultinfo = proc_mm = 0;
 153         return 0;
 154 }
 155
 156 /* The two __uml_setup would conflict, without this stupid alias. */
 157
 158 static int __init mode_skas0_cmd_param(char *str, int* add)
 159         __attribute__((alias("skas0_cmd_param")));
 160
 161 __uml_setup("skas0", skas0_cmd_param,
 162                 "skas0\n"
 163                 "    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
 164                 "    you specify mode=tt.\n\n");
 165
 166 __uml_setup("mode=skas0", mode_skas0_cmd_param,
 167                 "mode=skas0\n"
 168                 "    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
 169                 "    specify mode=tt. Note that this was recently added - on \n"
 170                 "    older kernels you must use simply \"skas0\".\n\n");
 171
 172 /* Changed only during early boot */
 173 static int force_sysemu_disabled = 0;
 174
 175 static int __init nosysemu_cmd_param(char *str, int* add)
 176 {
 177         force_sysemu_disabled = 1;
 178         return 0;
 179 }
 180
 181 __uml_setup("nosysemu", nosysemu_cmd_param,
 182 "nosysemu\n"
 183 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
 184 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
 185 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
 186 "    To make it working, you need a kernel patch for your host, too.\n"
 187 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
 188 "    information.\n\n");
 189
 190 static void __init check_sysemu(void)
 191 {
 192         unsigned long regs[MAX_REG_NR];
 193         int pid, n, status, count=0;
 194
 195         non_fatal("Checking syscall emulation patch for ptrace...");
 196         sysemu_supported = 0;
 197         pid = start_ptraced_child();
 198
 199         if (ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
 200                 goto fail;
 201
 202         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 203         if (n < 0)
 204                 fatal_perror("check_sysemu : wait failed");
 205         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
 206                 fatal("check_sysemu : expected SIGTRAP, got status = %d",
 207                       status);
 208
 209         if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
 210                 fatal_perror("check_sysemu : PTRACE_GETREGS failed");
 211         if (PT_SYSCALL_NR(regs) != __NR_getpid) {
 212                 non_fatal("check_sysemu got system call number %d, "
 213                           "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
 214                 goto fail;
 215         }
 216
 217         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
 218         if (n < 0) {
 219                 non_fatal("check_sysemu : failed to modify system call "
 220                           "return");
 221                 goto fail;
 222         }
 223
 224         if (stop_ptraced_child(pid, 0, 0) < 0)
 225                 goto fail_stopped;
 226
 227         sysemu_supported = 1;
 228         non_fatal("OK\n");
 229         set_using_sysemu(!force_sysemu_disabled);
 230
 231         non_fatal("Checking advanced syscall emulation patch for ptrace...");
 232         pid = start_ptraced_child();
 233
 234         if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 235                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 236                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 237
 238         while (1) {
 239                 count++;
 240                 if (ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
 241                         goto fail;
 242                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 243                 if (n < 0)
 244                         fatal_perror("check_ptrace : wait failed");
 245
 246                 if (WIFSTOPPED(status) &&
 247                     (WSTOPSIG(status) == (SIGTRAP|0x80))) {
 248                         if (!count)
 249                                 fatal("check_ptrace : SYSEMU_SINGLESTEP "
 250                                       "doesn't singlestep");
 251                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
 252                                    os_getpid());
 253                         if (n < 0)
 254                                 fatal_perror("check_sysemu : failed to modify "
 255                                              "system call return");
 256                         break;
 257                 }
 258                 else if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
 259                         count++;
 260                 else
 261                         fatal("check_ptrace : expected SIGTRAP or "
 262                               "(SIGTRAP | 0x80), got status = %d", status);
 263         }
 264         if (stop_ptraced_child(pid, 0, 0) < 0)
 265                 goto fail_stopped;
 266
 267         sysemu_supported = 2;
 268         non_fatal("OK\n");
 269
 270         if (!force_sysemu_disabled)
 271                 set_using_sysemu(sysemu_supported);
 272         return;
 273
 274 fail:
 275         stop_ptraced_child(pid, 1, 0);
 276 fail_stopped:
 277         non_fatal("missing\n");
 278 }
 279
 280 static void __init check_ptrace(void)
 281 {
 282         int pid, syscall, n, status;
 283
 284         non_fatal("Checking that ptrace can change system call numbers...");
 285         pid = start_ptraced_child();
 286
 287         if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 288                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 289                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 290
 291         while (1) {
 292                 if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
 293                         fatal_perror("check_ptrace : ptrace failed");
 294
 295                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 296                 if (n < 0)
 297                         fatal_perror("check_ptrace : wait failed");
 298
 299                 if (!WIFSTOPPED(status) ||
 300                    (WSTOPSIG(status) != (SIGTRAP | 0x80)))
 301                         fatal("check_ptrace : expected (SIGTRAP|0x80), "
 302                                "got status = %d", status);
 303
 304                 syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
 305                                  0);
 306                 if (syscall == __NR_getpid) {
 307                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 308                                    __NR_getppid);
 309                         if (n < 0)
 310                                 fatal_perror("check_ptrace : failed to modify "
 311                                              "system call");
 312                         break;
 313                 }
 314         }
 315         stop_ptraced_child(pid, 0, 1);
 316         non_fatal("OK\n");
 317         check_sysemu();
 318 }
 319
 320 extern void check_tmpexec(void);
 321
 322 static void __init check_coredump_limit(void)
 323 {
 324         struct rlimit lim;
 325         int err = getrlimit(RLIMIT_CORE, &lim);
 326
 327         if (err) {
 328                 perror("Getting core dump limit");
 329                 return;
 330         }
 331
 332         printf("Core dump limits :\n\tsoft - ");
 333         if (lim.rlim_cur == RLIM_INFINITY)
 334                 printf("NONE\n");
 335         else printf("%lu\n", lim.rlim_cur);
 336
 337         printf("\thard - ");
 338         if (lim.rlim_max == RLIM_INFINITY)
 339                 printf("NONE\n");
 340         else printf("%lu\n", lim.rlim_max);
 341 }
 342
 343 void __init os_early_checks(void)
 344 {
 345         int pid;
 346
 347         /* Print out the core dump limits early */
 348         check_coredump_limit();
 349
 350         check_ptrace();
 351
 352         /* Need to check this early because mmapping happens before the
 353          * kernel is running.
 354          */
 355         check_tmpexec();
 356
 357         pid = start_ptraced_child();
 358         if (init_registers(pid))
 359                 fatal("Failed to initialize default registers");
 360         stop_ptraced_child(pid, 1, 1);
 361 }
 362
 363 static int __init noprocmm_cmd_param(char *str, int* add)
 364 {
 365         proc_mm = 0;
 366         return 0;
 367 }
 368
 369 __uml_setup("noprocmm", noprocmm_cmd_param,
 370 "noprocmm\n"
 371 "    Turns off usage of /proc/mm, even if host supports it.\n"
 372 "    To support /proc/mm, the host needs to be patched using\n"
 373 "    the current skas3 patch.\n\n");
 374
 375 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
 376 {
 377         ptrace_faultinfo = 0;
 378         return 0;
 379 }
 380
 381 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
 382 "noptracefaultinfo\n"
 383 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
 384 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
 385 "    using the current skas3 patch.\n\n");
 386
 387 static int __init noptraceldt_cmd_param(char *str, int* add)
 388 {
 389         ptrace_ldt = 0;
 390         return 0;
 391 }
 392
 393 __uml_setup("noptraceldt", noptraceldt_cmd_param,
 394 "noptraceldt\n"
 395 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
 396 "    To support PTRACE_LDT, the host needs to be patched using\n"
 397 "    the current skas3 patch.\n\n");
 398
 399 static inline void check_skas3_ptrace_faultinfo(void)
 400 {
 401         struct ptrace_faultinfo fi;
 402         int pid, n;
 403
 404         non_fatal("  - PTRACE_FAULTINFO...");
 405         pid = start_ptraced_child();
 406
 407         n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
 408         if (n < 0) {
 409                 ptrace_faultinfo = 0;
 410                 if (errno == EIO)
 411                         non_fatal("not found\n");
 412                 else
 413                         perror("not found");
 414         }
 415         else {
 416                 if (!ptrace_faultinfo)
 417                         non_fatal("found but disabled on command line\n");
 418                 else
 419                         non_fatal("found\n");
 420         }
 421
 422         stop_ptraced_child(pid, 1, 1);
 423 }
 424
 425 static inline void check_skas3_ptrace_ldt(void)
 426 {
 427 #ifdef PTRACE_LDT
 428         int pid, n;
 429         unsigned char ldtbuf[40];
 430         struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
 431                 .func = 2, /* read default ldt */
 432                 .ptr = ldtbuf,
 433                 .bytecount = sizeof(ldtbuf)};
 434
 435         non_fatal("  - PTRACE_LDT...");
 436         pid = start_ptraced_child();
 437
 438         n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
 439         if (n < 0) {
 440                 if (errno == EIO)
 441                         non_fatal("not found\n");
 442                 else {
 443                         perror("not found");
 444                 }
 445                 ptrace_ldt = 0;
 446         }
 447         else {
 448                 if (ptrace_ldt)
 449                         non_fatal("found\n");
 450                 else
 451                         non_fatal("found, but use is disabled\n");
 452         }
 453
 454         stop_ptraced_child(pid, 1, 1);
 455 #else
 456         /* PTRACE_LDT might be disabled via cmdline option.
 457          * We want to override this, else we might use the stub
 458          * without real need
 459          */
 460         ptrace_ldt = 1;
 461 #endif
 462 }
 463
 464 static inline void check_skas3_proc_mm(void)
 465 {
 466         non_fatal("  - /proc/mm...");
 467         if (access("/proc/mm", W_OK) < 0) {
 468                 proc_mm = 0;
 469                 perror("not found");
 470         }
 471         else if (!proc_mm)
 472                 non_fatal("found but disabled on command line\n");
 473         else non_fatal("found\n");
 474 }
 475
 476 void can_do_skas(void)
 477 {
 478         non_fatal("Checking for the skas3 patch in the host:\n");
 479
 480         check_skas3_proc_mm();
 481         check_skas3_ptrace_faultinfo();
 482         check_skas3_ptrace_ldt();
 483
 484         if (!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
 485                 skas_needs_stub = 1;
 486 }
 487
 488 int __init parse_iomem(char *str, int *add)
 489 {
 490         struct iomem_region *new;
 491         struct stat64 buf;
 492         char *file, *driver;
 493         int fd, size;
 494
 495         driver = str;
 496         file = strchr(str,',');
 497         if (file == NULL) {
 498                 printf("parse_iomem : failed to parse iomem\n");
 499                 goto out;
 500         }
 501         *file = '\0';
 502         file++;
 503         fd = open(file, O_RDWR, 0);
 504         if (fd < 0) {
 505                 perror("parse_iomem - Couldn't open io file");
 506                 goto out;
 507         }
 508
 509         if (fstat64(fd, &buf) < 0) {
 510                 perror("parse_iomem - cannot stat_fd file");
 511                 goto out_close;
 512         }
 513
 514         new = malloc(sizeof(*new));
 515         if (new == NULL) {
 516                 perror("Couldn't allocate iomem_region struct");
 517                 goto out_close;
 518         }
 519
 520         size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
 521
 522         *new = ((struct iomem_region) { .next           = iomem_regions,
 523                                         .driver         = driver,
 524                                         .fd             = fd,
 525                                         .size           = size,
 526                                         .phys           = 0,
 527                                         .virt           = 0 });
 528         iomem_regions = new;
 529         iomem_size += new->size + UM_KERN_PAGE_SIZE;
 530
 531         return 0;
 532  out_close:
 533         close(fd);
 534  out:
 535         return 1;
 536 }