arch/um/os-Linux/start_up.c

   1 /*
   2  * Copyright (C) 2000, 2001, 2002 Jeff Dike (jdike@karaya.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <stdio.h>
   7 #include <stddef.h>
   8 #include <stdarg.h>
   9 #include <stdlib.h>
  10 #include <string.h>
  11 #include <unistd.h>
  12 #include <signal.h>
  13 #include <sched.h>
  14 #include <fcntl.h>
  15 #include <errno.h>
  16 #include <setjmp.h>
  17 #include <sys/time.h>
  18 #include <sys/wait.h>
  19 #include <sys/mman.h>
  20 #include <asm/unistd.h>
  21 #include <asm/page.h>
  22 #include <sys/types.h>
  23 #include "user_util.h"
  24 #include "kern_util.h"
  25 #include "user.h"
  26 #include "signal_kern.h"
  27 #include "sysdep/ptrace.h"
  28 #include "sysdep/sigcontext.h"
  29 #include "irq_user.h"
  30 #include "ptrace_user.h"
  31 #include "mem_user.h"
  32 #include "init.h"
  33 #include "os.h"
  34 #include "uml-config.h"
  35 #include "choose-mode.h"
  36 #include "mode.h"
  37 #include "tempfile.h"
  38 #include "kern_constants.h"
  39
  40 #ifdef UML_CONFIG_MODE_SKAS
  41 #include "skas.h"
  42 #include "skas_ptrace.h"
  43 #include "registers.h"
  44 #endif
  45
  46 static int ptrace_child(void *arg)
  47 {
  48         int ret;
  49         int pid = os_getpid(), ppid = getppid();
  50         int sc_result;
  51
  52         change_sig(SIGWINCH, 0);
  53         if(ptrace(PTRACE_TRACEME, 0, 0, 0) < 0){
  54                 perror("ptrace");
  55                 os_kill_process(pid, 0);
  56         }
  57         os_stop_process(pid);
  58
  59         /*This syscall will be intercepted by the parent. Don't call more than
  60          * once, please.*/
  61         sc_result = os_getpid();
  62
  63         if (sc_result == pid)
  64                 ret = 1; /*Nothing modified by the parent, we are running
  65                            normally.*/
  66         else if (sc_result == ppid)
  67                 ret = 0; /*Expected in check_ptrace and check_sysemu when they
  68                            succeed in modifying the stack frame*/
  69         else
  70                 ret = 2; /*Serious trouble! This could be caused by a bug in
  71                            host 2.6 SKAS3/2.6 patch before release -V6, together
  72                            with a bug in the UML code itself.*/
  73         _exit(ret);
  74 }
  75
  76 static int start_ptraced_child(void **stack_out)
  77 {
  78         void *stack;
  79         unsigned long sp;
  80         int pid, n, status;
  81
  82         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
  83                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
  84         if(stack == MAP_FAILED)
  85                 panic("check_ptrace : mmap failed, errno = %d", errno);
  86         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
  87         pid = clone(ptrace_child, (void *) sp, SIGCHLD, NULL);
  88         if(pid < 0)
  89                 panic("start_ptraced_child : clone failed, errno = %d", errno);
  90         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
  91         if(n < 0)
  92                 panic("check_ptrace : clone failed, errno = %d", errno);
  93         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
  94                 panic("check_ptrace : expected SIGSTOP, got status = %d",
  95                       status);
  96
  97         *stack_out = stack;
  98         return(pid);
  99 }
 100
 101 /* When testing for SYSEMU support, if it is one of the broken versions, we
 102  * must just avoid using sysemu, not panic, but only if SYSEMU features are
 103  * broken.
 104  * So only for SYSEMU features we test mustpanic, while normal host features
 105  * must work anyway!
 106  */
 107 static int stop_ptraced_child(int pid, void *stack, int exitcode,
 108                               int mustpanic)
 109 {
 110         int status, n, ret = 0;
 111
 112         if(ptrace(PTRACE_CONT, pid, 0, 0) < 0)
 113                 panic("check_ptrace : ptrace failed, errno = %d", errno);
 114         CATCH_EINTR(n = waitpid(pid, &status, 0));
 115         if(!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
 116                 int exit_with = WEXITSTATUS(status);
 117                 if (exit_with == 2)
 118                         printf("check_ptrace : child exited with status 2. "
 119                                "Serious trouble happening! Try updating your "
 120                                "host skas patch!\nDisabling SYSEMU support.");
 121                 printf("check_ptrace : child exited with exitcode %d, while "
 122                       "expecting %d; status 0x%x", exit_with,
 123                       exitcode, status);
 124                 if (mustpanic)
 125                         panic("\n");
 126                 else
 127                         printf("\n");
 128                 ret = -1;
 129         }
 130
 131         if(munmap(stack, PAGE_SIZE) < 0)
 132                 panic("check_ptrace : munmap failed, errno = %d", errno);
 133         return ret;
 134 }
 135
 136 int ptrace_faultinfo = 1;
 137 int ptrace_ldt = 1;
 138 int proc_mm = 1;
 139 int skas_needs_stub = 0;
 140
 141 static int __init skas0_cmd_param(char *str, int* add)
 142 {
 143         ptrace_faultinfo = proc_mm = 0;
 144         return 0;
 145 }
 146
 147 /* The two __uml_setup would conflict, without this stupid alias. */
 148
 149 static int __init mode_skas0_cmd_param(char *str, int* add)
 150         __attribute__((alias("skas0_cmd_param")));
 151
 152 __uml_setup("skas0", skas0_cmd_param,
 153                 "skas0\n"
 154                 "    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
 155                 "    you specify mode=tt.\n\n");
 156
 157 __uml_setup("mode=skas0", mode_skas0_cmd_param,
 158                 "mode=skas0\n"
 159                 "    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
 160                 "    specify mode=tt. Note that this was recently added - on \n"
 161                 "    older kernels you must use simply \"skas0\".\n\n");
 162
 163 static int force_sysemu_disabled = 0;
 164
 165 static int __init nosysemu_cmd_param(char *str, int* add)
 166 {
 167         force_sysemu_disabled = 1;
 168         return 0;
 169 }
 170
 171 __uml_setup("nosysemu", nosysemu_cmd_param,
 172 "nosysemu\n"
 173 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
 174 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
 175 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
 176 "    To make it working, you need a kernel patch for your host, too.\n"
 177 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
 178 "    information.\n\n");
 179
 180 static void __init check_sysemu(void)
 181 {
 182         void *stack;
 183         int pid, n, status, count=0;
 184
 185         printf("Checking syscall emulation patch for ptrace...");
 186         sysemu_supported = 0;
 187         pid = start_ptraced_child(&stack);
 188
 189         if(ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
 190                 goto fail;
 191
 192         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 193         if (n < 0)
 194                 panic("check_sysemu : wait failed, errno = %d", errno);
 195         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
 196                 panic("check_sysemu : expected SIGTRAP, "
 197                       "got status = %d", status);
 198
 199         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
 200                    os_getpid());
 201         if(n < 0)
 202                 panic("check_sysemu : failed to modify system "
 203                       "call return, errno = %d", errno);
 204
 205         if (stop_ptraced_child(pid, stack, 0, 0) < 0)
 206                 goto fail_stopped;
 207
 208         sysemu_supported = 1;
 209         printf("OK\n");
 210         set_using_sysemu(!force_sysemu_disabled);
 211
 212         printf("Checking advanced syscall emulation patch for ptrace...");
 213         pid = start_ptraced_child(&stack);
 214
 215         if(ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 216                   (void *) PTRACE_O_TRACESYSGOOD) < 0)
 217                 panic("check_ptrace: PTRACE_OLDSETOPTIONS failed, errno = %d",
 218                       errno);
 219
 220         while(1){
 221                 count++;
 222                 if(ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
 223                         goto fail;
 224                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 225                 if(n < 0)
 226                         panic("check_ptrace : wait failed, errno = %d", errno);
 227                 if(WIFSTOPPED(status) && (WSTOPSIG(status) == (SIGTRAP|0x80))){
 228                         if (!count)
 229                                 panic("check_ptrace : SYSEMU_SINGLESTEP "
 230                                       "doesn't singlestep");
 231                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
 232                                    os_getpid());
 233                         if(n < 0)
 234                                 panic("check_sysemu : failed to modify system "
 235                                       "call return, errno = %d", errno);
 236                         break;
 237                 }
 238                 else if(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
 239                         count++;
 240                 else
 241                         panic("check_ptrace : expected SIGTRAP or "
 242                               "(SIGTRAP|0x80), got status = %d", status);
 243         }
 244         if (stop_ptraced_child(pid, stack, 0, 0) < 0)
 245                 goto fail_stopped;
 246
 247         sysemu_supported = 2;
 248         printf("OK\n");
 249
 250         if ( !force_sysemu_disabled )
 251                 set_using_sysemu(sysemu_supported);
 252         return;
 253
 254 fail:
 255         stop_ptraced_child(pid, stack, 1, 0);
 256 fail_stopped:
 257         printf("missing\n");
 258 }
 259
 260 static void __init check_ptrace(void)
 261 {
 262         void *stack;
 263         int pid, syscall, n, status;
 264
 265         printf("Checking that ptrace can change system call numbers...");
 266         pid = start_ptraced_child(&stack);
 267
 268         if(ptrace(PTRACE_OLDSETOPTIONS, pid, 0, (void *)PTRACE_O_TRACESYSGOOD) < 0)
 269                 panic("check_ptrace: PTRACE_OLDSETOPTIONS failed, errno = %d", errno);
 270
 271         while(1){
 272                 if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
 273                         panic("check_ptrace : ptrace failed, errno = %d",
 274                               errno);
 275                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 276                 if(n < 0)
 277                         panic("check_ptrace : wait failed, errno = %d", errno);
 278                 if(!WIFSTOPPED(status) || (WSTOPSIG(status) != (SIGTRAP|0x80)))
 279                         panic("check_ptrace : expected (SIGTRAP|0x80), "
 280                               "got status = %d", status);
 281
 282                 syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
 283                                  0);
 284                 if(syscall == __NR_getpid){
 285                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 286                                    __NR_getppid);
 287                         if(n < 0)
 288                                 panic("check_ptrace : failed to modify system "
 289                                       "call, errno = %d", errno);
 290                         break;
 291                 }
 292         }
 293         stop_ptraced_child(pid, stack, 0, 1);
 294         printf("OK\n");
 295         check_sysemu();
 296 }
 297
 298 extern int create_tmp_file(unsigned long long len);
 299
 300 static void check_tmpexec(void)
 301 {
 302         void *addr;
 303         int err, fd = create_tmp_file(UM_KERN_PAGE_SIZE);
 304
 305         addr = mmap(NULL, UM_KERN_PAGE_SIZE,
 306                     PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE, fd, 0);
 307         printf("Checking PROT_EXEC mmap in /tmp...");
 308         fflush(stdout);
 309         if(addr == MAP_FAILED){
 310                 err = errno;
 311                 perror("failed");
 312                 if(err == EPERM)
 313                         printf("/tmp must be not mounted noexec\n");
 314                 exit(1);
 315         }
 316         printf("OK\n");
 317         munmap(addr, UM_KERN_PAGE_SIZE);
 318
 319         close(fd);
 320 }
 321
 322 void os_early_checks(void)
 323 {
 324         check_ptrace();
 325
 326         /* Need to check this early because mmapping happens before the
 327          * kernel is running.
 328          */
 329         check_tmpexec();
 330 }
 331
 332 static int __init noprocmm_cmd_param(char *str, int* add)
 333 {
 334         proc_mm = 0;
 335         return 0;
 336 }
 337
 338 __uml_setup("noprocmm", noprocmm_cmd_param,
 339 "noprocmm\n"
 340 "    Turns off usage of /proc/mm, even if host supports it.\n"
 341 "    To support /proc/mm, the host needs to be patched using\n"
 342 "    the current skas3 patch.\n\n");
 343
 344 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
 345 {
 346         ptrace_faultinfo = 0;
 347         return 0;
 348 }
 349
 350 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
 351 "noptracefaultinfo\n"
 352 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
 353 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
 354 "    using the current skas3 patch.\n\n");
 355
 356 static int __init noptraceldt_cmd_param(char *str, int* add)
 357 {
 358         ptrace_ldt = 0;
 359         return 0;
 360 }
 361
 362 __uml_setup("noptraceldt", noptraceldt_cmd_param,
 363 "noptraceldt\n"
 364 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
 365 "    To support PTRACE_LDT, the host needs to be patched using\n"
 366 "    the current skas3 patch.\n\n");
 367
 368 #ifdef UML_CONFIG_MODE_SKAS
 369 static inline void check_skas3_ptrace_faultinfo(void)
 370 {
 371         struct ptrace_faultinfo fi;
 372         void *stack;
 373         int pid, n;
 374
 375         printf("  - PTRACE_FAULTINFO...");
 376         pid = start_ptraced_child(&stack);
 377
 378         n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
 379         if (n < 0) {
 380                 ptrace_faultinfo = 0;
 381                 if(errno == EIO)
 382                         printf("not found\n");
 383                 else
 384                         perror("not found");
 385         }
 386         else {
 387                 if (!ptrace_faultinfo)
 388                         printf("found but disabled on command line\n");
 389                 else
 390                         printf("found\n");
 391         }
 392
 393         init_registers(pid);
 394         stop_ptraced_child(pid, stack, 1, 1);
 395 }
 396
 397 static inline void check_skas3_ptrace_ldt(void)
 398 {
 399 #ifdef PTRACE_LDT
 400         void *stack;
 401         int pid, n;
 402         unsigned char ldtbuf[40];
 403         struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
 404                 .func = 2, /* read default ldt */
 405                 .ptr = ldtbuf,
 406                 .bytecount = sizeof(ldtbuf)};
 407
 408         printf("  - PTRACE_LDT...");
 409         pid = start_ptraced_child(&stack);
 410
 411         n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
 412         if (n < 0) {
 413                 if(errno == EIO)
 414                         printf("not found\n");
 415                 else {
 416                         perror("not found");
 417                 }
 418                 ptrace_ldt = 0;
 419         }
 420         else {
 421                 if(ptrace_ldt)
 422                         printf("found\n");
 423                 else
 424                         printf("found, but use is disabled\n");
 425         }
 426
 427         stop_ptraced_child(pid, stack, 1, 1);
 428 #else
 429         /* PTRACE_LDT might be disabled via cmdline option.
 430          * We want to override this, else we might use the stub
 431          * without real need
 432          */
 433         ptrace_ldt = 1;
 434 #endif
 435 }
 436
 437 static inline void check_skas3_proc_mm(void)
 438 {
 439         printf("  - /proc/mm...");
 440         if (os_access("/proc/mm", OS_ACC_W_OK) < 0) {
 441                 proc_mm = 0;
 442                 printf("not found\n");
 443         }
 444         else {
 445                 if (!proc_mm)
 446                         printf("found but disabled on command line\n");
 447                 else
 448                         printf("found\n");
 449         }
 450 }
 451
 452 int can_do_skas(void)
 453 {
 454         printf("Checking for the skas3 patch in the host:\n");
 455
 456         check_skas3_proc_mm();
 457         check_skas3_ptrace_faultinfo();
 458         check_skas3_ptrace_ldt();
 459
 460         if(!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
 461                 skas_needs_stub = 1;
 462
 463         return 1;
 464 }
 465 #else
 466 int can_do_skas(void)
 467 {
 468         return(0);
 469 }
 470 #endif
 471
 472 int have_devanon = 0;
 473
 474 /* Runs on boot kernel stack - already safe to use printk. */
 475
 476 void check_devanon(void)
 477 {
 478         int fd;
 479
 480         printk("Checking for /dev/anon on the host...");
 481         fd = open("/dev/anon", O_RDWR);
 482         if(fd < 0){
 483                 printk("Not available (open failed with errno %d)\n", errno);
 484                 return;
 485         }
 486
 487         printk("OK\n");
 488         have_devanon = 1;
 489 }
 490
 491 int __init parse_iomem(char *str, int *add)
 492 {
 493         struct iomem_region *new;
 494         struct uml_stat buf;
 495         char *file, *driver;
 496         int fd, err, size;
 497
 498         driver = str;
 499         file = strchr(str,',');
 500         if(file == NULL){
 501                 printf("parse_iomem : failed to parse iomem\n");
 502                 goto out;
 503         }
 504         *file = '\0';
 505         file++;
 506         fd = os_open_file(file, of_rdwr(OPENFLAGS()), 0);
 507         if(fd < 0){
 508                 os_print_error(fd, "parse_iomem - Couldn't open io file");
 509                 goto out;
 510         }
 511
 512         err = os_stat_fd(fd, &buf);
 513         if(err < 0){
 514                 os_print_error(err, "parse_iomem - cannot stat_fd file");
 515                 goto out_close;
 516         }
 517
 518         new = malloc(sizeof(*new));
 519         if(new == NULL){
 520                 perror("Couldn't allocate iomem_region struct");
 521                 goto out_close;
 522         }
 523
 524         size = (buf.ust_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
 525
 526         *new = ((struct iomem_region) { .next           = iomem_regions,
 527                                         .driver         = driver,
 528                                         .fd             = fd,
 529                                         .size           = size,
 530                                         .phys           = 0,
 531                                         .virt           = 0 });
 532         iomem_regions = new;
 533         iomem_size += new->size + UM_KERN_PAGE_SIZE;
 534
 535         return(0);
 536  out_close:
 537         os_close_file(fd);
 538  out:
 539         return(1);
 540 }
 541