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