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/stat.h>
  17 #include <sys/wait.h>
  18 #include <sys/time.h>
  19 #include <sys/resource.h>
  20 #include <asm/unistd.h>
  21 #include <init.h>
  22 #include <os.h>
  23 #include <mem_user.h>
  24 #include <ptrace_user.h>
  25 #include <registers.h>
  26 #include <skas.h>
  27
  28 static void 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         if (change_sig(SIGWINCH, 0) < 0 ||
  36             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         vfprintf(stderr, fmt, list);
  79         va_end(list);
  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         vfprintf(stderr, fmt, list);
  90         va_end(list);
  91 }
  92
  93 static int start_ptraced_child(void)
  94 {
  95         int pid, n, status;
  96
  97         pid = fork();
  98         if (pid == 0)
  99                 ptrace_child();
 100         else if (pid < 0)
 101                 fatal_perror("start_ptraced_child : fork failed");
 102
 103         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 104         if (n < 0)
 105                 fatal_perror("check_ptrace : waitpid failed");
 106         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 107                 fatal("check_ptrace : expected SIGSTOP, got status = %d",
 108                       status);
 109
 110         return pid;
 111 }
 112
 113 /* When testing for SYSEMU support, if it is one of the broken versions, we
 114  * must just avoid using sysemu, not panic, but only if SYSEMU features are
 115  * broken.
 116  * So only for SYSEMU features we test mustpanic, while normal host features
 117  * must work anyway!
 118  */
 119 static int stop_ptraced_child(int pid, int exitcode, int mustexit)
 120 {
 121         int status, n, ret = 0;
 122
 123         if (ptrace(PTRACE_CONT, pid, 0, 0) < 0) {
 124                 perror("stop_ptraced_child : ptrace failed");
 125                 return -1;
 126         }
 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 static int force_sysemu_disabled = 0;
 146
 147 static int __init nosysemu_cmd_param(char *str, int* add)
 148 {
 149         force_sysemu_disabled = 1;
 150         return 0;
 151 }
 152
 153 __uml_setup("nosysemu", nosysemu_cmd_param,
 154 "nosysemu\n"
 155 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
 156 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
 157 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
 158 "    To make it working, you need a kernel patch for your host, too.\n"
 159 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
 160 "    information.\n\n");
 161
 162 static void __init check_sysemu(void)
 163 {
 164         unsigned long regs[MAX_REG_NR];
 165         int pid, n, status, count=0;
 166
 167         non_fatal("Checking syscall emulation patch for ptrace...");
 168         sysemu_supported = 0;
 169         pid = start_ptraced_child();
 170
 171         if (ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
 172                 goto fail;
 173
 174         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 175         if (n < 0)
 176                 fatal_perror("check_sysemu : wait failed");
 177         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
 178                 fatal("check_sysemu : expected SIGTRAP, got status = %d\n",
 179                       status);
 180
 181         if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
 182                 fatal_perror("check_sysemu : PTRACE_GETREGS failed");
 183         if (PT_SYSCALL_NR(regs) != __NR_getpid) {
 184                 non_fatal("check_sysemu got system call number %d, "
 185                           "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
 186                 goto fail;
 187         }
 188
 189         n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
 190         if (n < 0) {
 191                 non_fatal("check_sysemu : failed to modify system call "
 192                           "return");
 193                 goto fail;
 194         }
 195
 196         if (stop_ptraced_child(pid, 0, 0) < 0)
 197                 goto fail_stopped;
 198
 199         sysemu_supported = 1;
 200         non_fatal("OK\n");
 201         set_using_sysemu(!force_sysemu_disabled);
 202
 203         non_fatal("Checking advanced syscall emulation patch for ptrace...");
 204         pid = start_ptraced_child();
 205
 206         if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 207                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 208                 fatal_perror("check_sysemu: PTRACE_OLDSETOPTIONS failed");
 209
 210         while (1) {
 211                 count++;
 212                 if (ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
 213                         goto fail;
 214                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 215                 if (n < 0)
 216                         fatal_perror("check_sysemu: wait failed");
 217
 218                 if (WIFSTOPPED(status) &&
 219                     (WSTOPSIG(status) == (SIGTRAP|0x80))) {
 220                         if (!count) {
 221                                 non_fatal("check_sysemu: SYSEMU_SINGLESTEP "
 222                                           "doesn't singlestep");
 223                                 goto fail;
 224                         }
 225                         n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET,
 226                                    os_getpid());
 227                         if (n < 0)
 228                                 fatal_perror("check_sysemu : failed to modify "
 229                                              "system call return");
 230                         break;
 231                 }
 232                 else if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
 233                         count++;
 234                 else {
 235                         non_fatal("check_sysemu: expected SIGTRAP or "
 236                                   "(SIGTRAP | 0x80), got status = %d\n",
 237                                   status);
 238                         goto fail;
 239                 }
 240         }
 241         if (stop_ptraced_child(pid, 0, 0) < 0)
 242                 goto fail_stopped;
 243
 244         sysemu_supported = 2;
 245         non_fatal("OK\n");
 246
 247         if (!force_sysemu_disabled)
 248                 set_using_sysemu(sysemu_supported);
 249         return;
 250
 251 fail:
 252         stop_ptraced_child(pid, 1, 0);
 253 fail_stopped:
 254         non_fatal("missing\n");
 255 }
 256
 257 static void __init check_ptrace(void)
 258 {
 259         int pid, syscall, n, status;
 260
 261         non_fatal("Checking that ptrace can change system call numbers...");
 262         pid = start_ptraced_child();
 263
 264         if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 265                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 266                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 267
 268         while (1) {
 269                 if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
 270                         fatal_perror("check_ptrace : ptrace failed");
 271
 272                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 273                 if (n < 0)
 274                         fatal_perror("check_ptrace : wait failed");
 275
 276                 if (!WIFSTOPPED(status) ||
 277                    (WSTOPSIG(status) != (SIGTRAP | 0x80)))
 278                         fatal("check_ptrace : expected (SIGTRAP|0x80), "
 279                                "got status = %d", status);
 280
 281                 syscall = ptrace(PTRACE_PEEKUSER, pid, PT_SYSCALL_NR_OFFSET,
 282                                  0);
 283                 if (syscall == __NR_getpid) {
 284                         n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
 285                                    __NR_getppid);
 286                         if (n < 0)
 287                                 fatal_perror("check_ptrace : failed to modify "
 288                                              "system call");
 289                         break;
 290                 }
 291         }
 292         stop_ptraced_child(pid, 0, 1);
 293         non_fatal("OK\n");
 294         check_sysemu();
 295 }
 296
 297 extern void check_tmpexec(void);
 298
 299 static void __init check_coredump_limit(void)
 300 {
 301         struct rlimit lim;
 302         int err = getrlimit(RLIMIT_CORE, &lim);
 303
 304         if (err) {
 305                 perror("Getting core dump limit");
 306                 return;
 307         }
 308
 309         printf("Core dump limits :\n\tsoft - ");
 310         if (lim.rlim_cur == RLIM_INFINITY)
 311                 printf("NONE\n");
 312         else printf("%lu\n", lim.rlim_cur);
 313
 314         printf("\thard - ");
 315         if (lim.rlim_max == RLIM_INFINITY)
 316                 printf("NONE\n");
 317         else printf("%lu\n", lim.rlim_max);
 318 }
 319
 320 void __init os_early_checks(void)
 321 {
 322         int pid;
 323
 324         /* Print out the core dump limits early */
 325         check_coredump_limit();
 326
 327         check_ptrace();
 328
 329         /* Need to check this early because mmapping happens before the
 330          * kernel is running.
 331          */
 332         check_tmpexec();
 333
 334         pid = start_ptraced_child();
 335         if (init_registers(pid))
 336                 fatal("Failed to initialize default registers");
 337         stop_ptraced_child(pid, 1, 1);
 338 }
 339
 340 int __init parse_iomem(char *str, int *add)
 341 {
 342         struct iomem_region *new;
 343         struct stat64 buf;
 344         char *file, *driver;
 345         int fd, size;
 346
 347         driver = str;
 348         file = strchr(str,',');
 349         if (file == NULL) {
 350                 fprintf(stderr, "parse_iomem : failed to parse iomem\n");
 351                 goto out;
 352         }
 353         *file = '\0';
 354         file++;
 355         fd = open(file, O_RDWR, 0);
 356         if (fd < 0) {
 357                 perror("parse_iomem - Couldn't open io file");
 358                 goto out;
 359         }
 360
 361         if (fstat64(fd, &buf) < 0) {
 362                 perror("parse_iomem - cannot stat_fd file");
 363                 goto out_close;
 364         }
 365
 366         new = malloc(sizeof(*new));
 367         if (new == NULL) {
 368                 perror("Couldn't allocate iomem_region struct");
 369                 goto out_close;
 370         }
 371
 372         size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
 373
 374         *new = ((struct iomem_region) { .next           = iomem_regions,
 375                                         .driver         = driver,
 376                                         .fd             = fd,
 377                                         .size           = size,
 378                                         .phys           = 0,
 379                                         .virt           = 0 });
 380         iomem_regions = new;
 381         iomem_size += new->size + UM_KERN_PAGE_SIZE;
 382
 383         return 0;
 384  out_close:
 385         close(fd);
 386  out:
 387         return 1;
 388 }