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.h"
  27 #include "skas_ptrace.h"
  28
  29 static void ptrace_child(void)
  30 {
  31         int ret;
  32         /* Calling os_getpid because some libcs cached getpid incorrectly */
  33         int pid = os_getpid(), ppid = getppid();
  34         int sc_result;
  35
  36         if (change_sig(SIGWINCH, 0) < 0 ||
  37             ptrace(PTRACE_TRACEME, 0, 0, 0) < 0) {
  38                 perror("ptrace");
  39                 kill(pid, SIGKILL);
  40         }
  41         kill(pid, SIGSTOP);
  42
  43         /*
  44          * This syscall will be intercepted by the parent. Don't call more than
  45          * once, please.
  46          */
  47         sc_result = os_getpid();
  48
  49         if (sc_result == pid)
  50                 /* Nothing modified by the parent, we are running normally. */
  51                 ret = 1;
  52         else if (sc_result == ppid)
  53                 /*
  54                  * Expected in check_ptrace and check_sysemu when they succeed
  55                  * in modifying the stack frame
  56                  */
  57                 ret = 0;
  58         else
  59                 /* Serious trouble! This could be caused by a bug in host 2.6
  60                  * SKAS3/2.6 patch before release -V6, together with a bug in
  61                  * the UML code itself.
  62                  */
  63                 ret = 2;
  64
  65         exit(ret);
  66 }
  67
  68 static void fatal_perror(const char *str)
  69 {
  70         perror(str);
  71         exit(1);
  72 }
  73
  74 static void fatal(char *fmt, ...)
  75 {
  76         va_list list;
  77
  78         va_start(list, fmt);
  79         vfprintf(stderr, fmt, list);
  80         va_end(list);
  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         vfprintf(stderr, fmt, list);
  91         va_end(list);
  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;
 145 static int disable_ptrace_faultinfo;
 146
 147 int ptrace_ldt;
 148 static int disable_ptrace_ldt;
 149
 150 int proc_mm;
 151 static int disable_proc_mm;
 152
 153 int have_switch_mm;
 154 static int disable_switch_mm;
 155
 156 int skas_needs_stub;
 157
 158 static int __init skas0_cmd_param(char *str, int* add)
 159 {
 160         disable_ptrace_faultinfo = 1;
 161         disable_ptrace_ldt = 1;
 162         disable_proc_mm = 1;
 163         disable_switch_mm = 1;
 164
 165         return 0;
 166 }
 167
 168 /* The two __uml_setup would conflict, without this stupid alias. */
 169
 170 static int __init mode_skas0_cmd_param(char *str, int* add)
 171         __attribute__((alias("skas0_cmd_param")));
 172
 173 __uml_setup("skas0", skas0_cmd_param,
 174 "skas0\n"
 175 "    Disables SKAS3 and SKAS4 usage, so that SKAS0 is used\n\n");
 176
 177 __uml_setup("mode=skas0", mode_skas0_cmd_param,
 178 "mode=skas0\n"
 179 "    Disables SKAS3 and SKAS4 usage, so that SKAS0 is used.\n\n");
 180
 181 /* Changed only during early boot */
 182 static int force_sysemu_disabled = 0;
 183
 184 static int __init nosysemu_cmd_param(char *str, int* add)
 185 {
 186         force_sysemu_disabled = 1;
 187         return 0;
 188 }
 189
 190 __uml_setup("nosysemu", nosysemu_cmd_param,
 191 "nosysemu\n"
 192 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
 193 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
 194 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
 195 "    To make it working, you need a kernel patch for your host, too.\n"
 196 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
 197 "    information.\n\n");
 198
 199 static void __init check_sysemu(void)
 200 {
 201         unsigned long regs[MAX_REG_NR];
 202         int pid, n, status, count=0;
 203
 204         non_fatal("Checking syscall emulation patch for ptrace...");
 205         sysemu_supported = 0;
 206         pid = start_ptraced_child();
 207
 208         if (ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
 209                 goto fail;
 210
 211         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 212         if (n < 0)
 213                 fatal_perror("check_sysemu : wait failed");
 214         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
 215                 fatal("check_sysemu : expected SIGTRAP, got status = %d",
 216                       status);
 217
 218         if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
 219                 fatal_perror("check_sysemu : PTRACE_GETREGS failed");
 220         if (PT_SYSCALL_NR(regs) != __NR_getpid) {
 221                 non_fatal("check_sysemu got system call number %d, "
 222                           "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
 223                 goto fail;
 224         }
 225
 226         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
 227         if (n < 0) {
 228                 non_fatal("check_sysemu : failed to modify system call "
 229                           "return");
 230                 goto fail;
 231         }
 232
 233         if (stop_ptraced_child(pid, 0, 0) < 0)
 234                 goto fail_stopped;
 235
 236         sysemu_supported = 1;
 237         non_fatal("OK\n");
 238         set_using_sysemu(!force_sysemu_disabled);
 239
 240         non_fatal("Checking advanced syscall emulation patch for ptrace...");
 241         pid = start_ptraced_child();
 242
 243         if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 244                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 245                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 246
 247         while (1) {
 248                 count++;
 249                 if (ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
 250                         goto fail;
 251                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 252                 if (n < 0)
 253                         fatal_perror("check_ptrace : wait failed");
 254
 255                 if (WIFSTOPPED(status) &&
 256                     (WSTOPSIG(status) == (SIGTRAP|0x80))) {
 257                         if (!count)
 258                                 fatal("check_ptrace : SYSEMU_SINGLESTEP "
 259                                       "doesn't singlestep");
 260                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
 261                                    os_getpid());
 262                         if (n < 0)
 263                                 fatal_perror("check_sysemu : failed to modify "
 264                                              "system call return");
 265                         break;
 266                 }
 267                 else if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
 268                         count++;
 269                 else
 270                         fatal("check_ptrace : expected SIGTRAP or "
 271                               "(SIGTRAP | 0x80), got status = %d", status);
 272         }
 273         if (stop_ptraced_child(pid, 0, 0) < 0)
 274                 goto fail_stopped;
 275
 276         sysemu_supported = 2;
 277         non_fatal("OK\n");
 278
 279         if (!force_sysemu_disabled)
 280                 set_using_sysemu(sysemu_supported);
 281         return;
 282
 283 fail:
 284         stop_ptraced_child(pid, 1, 0);
 285 fail_stopped:
 286         non_fatal("missing\n");
 287 }
 288
 289 static void __init check_ptrace(void)
 290 {
 291         int pid, syscall, n, status;
 292
 293         non_fatal("Checking that ptrace can change system call numbers...");
 294         pid = start_ptraced_child();
 295
 296         if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 297                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 298                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 299
 300         while (1) {
 301                 if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
 302                         fatal_perror("check_ptrace : ptrace failed");
 303
 304                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 305                 if (n < 0)
 306                         fatal_perror("check_ptrace : wait failed");
 307
 308                 if (!WIFSTOPPED(status) ||
 309                    (WSTOPSIG(status) != (SIGTRAP | 0x80)))
 310                         fatal("check_ptrace : expected (SIGTRAP|0x80), "
 311                                "got status = %d", status);
 312
 313                 syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
 314                                  0);
 315                 if (syscall == __NR_getpid) {
 316                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 317                                    __NR_getppid);
 318                         if (n < 0)
 319                                 fatal_perror("check_ptrace : failed to modify "
 320                                              "system call");
 321                         break;
 322                 }
 323         }
 324         stop_ptraced_child(pid, 0, 1);
 325         non_fatal("OK\n");
 326         check_sysemu();
 327 }
 328
 329 extern void check_tmpexec(void);
 330
 331 static void __init check_coredump_limit(void)
 332 {
 333         struct rlimit lim;
 334         int err = getrlimit(RLIMIT_CORE, &lim);
 335
 336         if (err) {
 337                 perror("Getting core dump limit");
 338                 return;
 339         }
 340
 341         printf("Core dump limits :\n\tsoft - ");
 342         if (lim.rlim_cur == RLIM_INFINITY)
 343                 printf("NONE\n");
 344         else printf("%lu\n", lim.rlim_cur);
 345
 346         printf("\thard - ");
 347         if (lim.rlim_max == RLIM_INFINITY)
 348                 printf("NONE\n");
 349         else printf("%lu\n", lim.rlim_max);
 350 }
 351
 352 void __init os_early_checks(void)
 353 {
 354         int pid;
 355
 356         /* Print out the core dump limits early */
 357         check_coredump_limit();
 358
 359         check_ptrace();
 360
 361         /* Need to check this early because mmapping happens before the
 362          * kernel is running.
 363          */
 364         check_tmpexec();
 365
 366         pid = start_ptraced_child();
 367         if (init_registers(pid))
 368                 fatal("Failed to initialize default registers");
 369         stop_ptraced_child(pid, 1, 1);
 370 }
 371
 372 static int __init noprocmm_cmd_param(char *str, int* add)
 373 {
 374         disable_proc_mm = 1;
 375         return 0;
 376 }
 377
 378 __uml_setup("noprocmm", noprocmm_cmd_param,
 379 "noprocmm\n"
 380 "    Turns off usage of /proc/mm, even if host supports it.\n"
 381 "    To support /proc/mm, the host needs to be patched using\n"
 382 "    the current skas3 patch.\n\n");
 383
 384 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
 385 {
 386         disable_ptrace_faultinfo = 1;
 387         return 0;
 388 }
 389
 390 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
 391 "noptracefaultinfo\n"
 392 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
 393 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
 394 "    using the current skas3 patch.\n\n");
 395
 396 static int __init noptraceldt_cmd_param(char *str, int* add)
 397 {
 398         disable_ptrace_ldt = 1;
 399         return 0;
 400 }
 401
 402 __uml_setup("noptraceldt", noptraceldt_cmd_param,
 403 "noptraceldt\n"
 404 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
 405 "    To support PTRACE_LDT, the host needs to be patched using\n"
 406 "    the current skas3 patch.\n\n");
 407
 408 static inline void check_skas3_ptrace_faultinfo(void)
 409 {
 410         struct ptrace_faultinfo fi;
 411         int pid, n;
 412
 413         non_fatal("  - PTRACE_FAULTINFO...");
 414         pid = start_ptraced_child();
 415
 416         n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
 417         if (n < 0) {
 418                 if (errno == EIO)
 419                         non_fatal("not found\n");
 420                 else
 421                         perror("not found");
 422         } else if (disable_ptrace_faultinfo)
 423                 non_fatal("found but disabled on command line\n");
 424         else {
 425                 ptrace_faultinfo = 1;
 426                 non_fatal("found\n");
 427         }
 428
 429         stop_ptraced_child(pid, 1, 1);
 430 }
 431
 432 static inline void check_skas3_ptrace_ldt(void)
 433 {
 434 #ifdef PTRACE_LDT
 435         int pid, n;
 436         unsigned char ldtbuf[40];
 437         struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
 438                 .func = 2, /* read default ldt */
 439                 .ptr = ldtbuf,
 440                 .bytecount = sizeof(ldtbuf)};
 441
 442         non_fatal("  - PTRACE_LDT...");
 443         pid = start_ptraced_child();
 444
 445         n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
 446         if (n < 0) {
 447                 if (errno == EIO)
 448                         non_fatal("not found\n");
 449                 else
 450                         perror("not found");
 451         } else if (disable_ptrace_ldt)
 452                 non_fatal("found, but use is disabled\n");
 453         else {
 454                 ptrace_ldt = 1;
 455                 non_fatal("found\n");
 456         }
 457
 458         stop_ptraced_child(pid, 1, 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                 perror("not found");
 467         else if (disable_proc_mm)
 468                 non_fatal("found but disabled on command line\n");
 469         else {
 470                 proc_mm = 1;
 471                 non_fatal("found\n");
 472         }
 473 }
 474
 475 void can_do_skas(void)
 476 {
 477         non_fatal("Checking for the skas3 patch in the host:\n");
 478
 479         check_skas3_proc_mm();
 480         check_skas3_ptrace_faultinfo();
 481         check_skas3_ptrace_ldt();
 482
 483         if (!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
 484                 skas_needs_stub = 1;
 485 }
 486
 487 int __init parse_iomem(char *str, int *add)
 488 {
 489         struct iomem_region *new;
 490         struct stat64 buf;
 491         char *file, *driver;
 492         int fd, size;
 493
 494         driver = str;
 495         file = strchr(str,',');
 496         if (file == NULL) {
 497                 fprintf(stderr, "parse_iomem : failed to parse iomem\n");
 498                 goto out;
 499         }
 500         *file = '\0';
 501         file++;
 502         fd = open(file, O_RDWR, 0);
 503         if (fd < 0) {
 504                 perror("parse_iomem - Couldn't open io file");
 505                 goto out;
 506         }
 507
 508         if (fstat64(fd, &buf) < 0) {
 509                 perror("parse_iomem - cannot stat_fd file");
 510                 goto out_close;
 511         }
 512
 513         new = malloc(sizeof(*new));
 514         if (new == NULL) {
 515                 perror("Couldn't allocate iomem_region struct");
 516                 goto out_close;
 517         }
 518
 519         size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
 520
 521         *new = ((struct iomem_region) { .next           = iomem_regions,
 522                                         .driver         = driver,
 523                                         .fd             = fd,
 524                                         .size           = size,
 525                                         .phys           = 0,
 526                                         .virt           = 0 });
 527         iomem_regions = new;
 528         iomem_size += new->size + UM_KERN_PAGE_SIZE;
 529
 530         return 0;
 531  out_close:
 532         close(fd);
 533  out:
 534         return 1;
 535 }