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