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 <pty.h>
   7 #include <stdio.h>
   8 #include <stddef.h>
   9 #include <stdarg.h>
  10 #include <stdlib.h>
  11 #include <string.h>
  12 #include <unistd.h>
  13 #include <signal.h>
  14 #include <sched.h>
  15 #include <fcntl.h>
  16 #include <errno.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         kill(pid, SIGSTOP);
  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 void fatal_perror(char *str)
  77 {
  78         perror(str);
  79         exit(1);
  80 }
  81
  82 static void fatal(char *fmt, ...)
  83 {
  84         va_list list;
  85
  86         va_start(list, fmt);
  87         vprintf(fmt, list);
  88         va_end(list);
  89         fflush(stdout);
  90
  91         exit(1);
  92 }
  93
  94 static void non_fatal(char *fmt, ...)
  95 {
  96         va_list list;
  97
  98         va_start(list, fmt);
  99         vprintf(fmt, list);
 100         va_end(list);
 101         fflush(stdout);
 102 }
 103
 104 static int start_ptraced_child(void **stack_out)
 105 {
 106         void *stack;
 107         unsigned long sp;
 108         int pid, n, status;
 109
 110         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 111                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 112         if(stack == MAP_FAILED)
 113                 fatal_perror("check_ptrace : mmap failed");
 114         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
 115         pid = clone(ptrace_child, (void *) sp, SIGCHLD, NULL);
 116         if(pid < 0)
 117                 fatal_perror("start_ptraced_child : clone failed");
 118         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 119         if(n < 0)
 120                 fatal_perror("check_ptrace : clone failed");
 121         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 122                 fatal("check_ptrace : expected SIGSTOP, got status = %d",
 123                       status);
 124
 125         *stack_out = stack;
 126         return pid;
 127 }
 128
 129 /* When testing for SYSEMU support, if it is one of the broken versions, we
 130  * must just avoid using sysemu, not panic, but only if SYSEMU features are
 131  * broken.
 132  * So only for SYSEMU features we test mustpanic, while normal host features
 133  * must work anyway!
 134  */
 135 static int stop_ptraced_child(int pid, void *stack, int exitcode,
 136                               int mustexit)
 137 {
 138         int status, n, ret = 0;
 139
 140         if(ptrace(PTRACE_CONT, pid, 0, 0) < 0)
 141                 fatal_perror("stop_ptraced_child : ptrace failed");
 142         CATCH_EINTR(n = waitpid(pid, &status, 0));
 143         if(!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
 144                 int exit_with = WEXITSTATUS(status);
 145                 if (exit_with == 2)
 146                         non_fatal("check_ptrace : child exited with status 2. "
 147                                   "Serious trouble happening! Try updating "
 148                                   "your host skas patch!\nDisabling SYSEMU "
 149                                   "support.");
 150                 non_fatal("check_ptrace : child exited with exitcode %d, while "
 151                           "expecting %d; status 0x%x\n", exit_with,
 152                           exitcode, status);
 153                 if (mustexit)
 154                         exit(1);
 155                 ret = -1;
 156         }
 157
 158         if(munmap(stack, PAGE_SIZE) < 0)
 159                 fatal_perror("check_ptrace : munmap failed");
 160         return ret;
 161 }
 162
 163 /* Changed only during early boot */
 164 int ptrace_faultinfo = 1;
 165 int ptrace_ldt = 1;
 166 int proc_mm = 1;
 167 int skas_needs_stub = 0;
 168
 169 static int __init skas0_cmd_param(char *str, int* add)
 170 {
 171         ptrace_faultinfo = proc_mm = 0;
 172         return 0;
 173 }
 174
 175 /* The two __uml_setup would conflict, without this stupid alias. */
 176
 177 static int __init mode_skas0_cmd_param(char *str, int* add)
 178         __attribute__((alias("skas0_cmd_param")));
 179
 180 __uml_setup("skas0", skas0_cmd_param,
 181                 "skas0\n"
 182                 "    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
 183                 "    you specify mode=tt.\n\n");
 184
 185 __uml_setup("mode=skas0", mode_skas0_cmd_param,
 186                 "mode=skas0\n"
 187                 "    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
 188                 "    specify mode=tt. Note that this was recently added - on \n"
 189                 "    older kernels you must use simply \"skas0\".\n\n");
 190
 191 /* Changed only during early boot */
 192 static int force_sysemu_disabled = 0;
 193
 194 static int __init nosysemu_cmd_param(char *str, int* add)
 195 {
 196         force_sysemu_disabled = 1;
 197         return 0;
 198 }
 199
 200 __uml_setup("nosysemu", nosysemu_cmd_param,
 201 "nosysemu\n"
 202 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
 203 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
 204 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
 205 "    To make it working, you need a kernel patch for your host, too.\n"
 206 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
 207 "    information.\n\n");
 208
 209 static void __init check_sysemu(void)
 210 {
 211         void *stack;
 212         int pid, n, status, count=0;
 213
 214         non_fatal("Checking syscall emulation patch for ptrace...");
 215         sysemu_supported = 0;
 216         pid = start_ptraced_child(&stack);
 217
 218         if(ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
 219                 goto fail;
 220
 221         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 222         if (n < 0)
 223                 fatal_perror("check_sysemu : wait failed");
 224         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
 225                 fatal("check_sysemu : expected SIGTRAP, got status = %d",
 226                       status);
 227
 228         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
 229                    os_getpid());
 230         if(n < 0)
 231                 fatal_perror("check_sysemu : failed to modify system call "
 232                              "return");
 233
 234         if (stop_ptraced_child(pid, stack, 0, 0) < 0)
 235                 goto fail_stopped;
 236
 237         sysemu_supported = 1;
 238         non_fatal("OK\n");
 239         set_using_sysemu(!force_sysemu_disabled);
 240
 241         non_fatal("Checking advanced syscall emulation patch for ptrace...");
 242         pid = start_ptraced_child(&stack);
 243
 244         if((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 245                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 246                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 247
 248         while(1){
 249                 count++;
 250                 if(ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
 251                         goto fail;
 252                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 253                 if(n < 0)
 254                         fatal_perror("check_ptrace : wait failed");
 255
 256                 if(WIFSTOPPED(status) && (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, stack, 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, stack, 1, 0);
 285 fail_stopped:
 286         non_fatal("missing\n");
 287 }
 288
 289 static void __init check_ptrace(void)
 290 {
 291         void *stack;
 292         int pid, syscall, n, status;
 293
 294         non_fatal("Checking that ptrace can change system call numbers...");
 295         pid = start_ptraced_child(&stack);
 296
 297         if((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 298                    (void *) PTRACE_O_TRACESYSGOOD) < 0))
 299                 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
 300
 301         while(1){
 302                 if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
 303                         fatal_perror("check_ptrace : ptrace failed");
 304
 305                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 306                 if(n < 0)
 307                         fatal_perror("check_ptrace : wait failed");
 308
 309                 if(!WIFSTOPPED(status) ||
 310                    (WSTOPSIG(status) != (SIGTRAP | 0x80)))
 311                         fatal("check_ptrace : expected (SIGTRAP|0x80), "
 312                                "got status = %d", status);
 313
 314                 syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
 315                                  0);
 316                 if(syscall == __NR_getpid){
 317                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 318                                    __NR_getppid);
 319                         if(n < 0)
 320                                 fatal_perror("check_ptrace : failed to modify "
 321                                              "system call");
 322                         break;
 323                 }
 324         }
 325         stop_ptraced_child(pid, stack, 0, 1);
 326         non_fatal("OK\n");
 327         check_sysemu();
 328 }
 329
 330 extern void check_tmpexec(void);
 331
 332 void os_early_checks(void)
 333 {
 334         check_ptrace();
 335
 336         /* Need to check this early because mmapping happens before the
 337          * kernel is running.
 338          */
 339         check_tmpexec();
 340 }
 341
 342 static int __init noprocmm_cmd_param(char *str, int* add)
 343 {
 344         proc_mm = 0;
 345         return 0;
 346 }
 347
 348 __uml_setup("noprocmm", noprocmm_cmd_param,
 349 "noprocmm\n"
 350 "    Turns off usage of /proc/mm, even if host supports it.\n"
 351 "    To support /proc/mm, the host needs to be patched using\n"
 352 "    the current skas3 patch.\n\n");
 353
 354 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
 355 {
 356         ptrace_faultinfo = 0;
 357         return 0;
 358 }
 359
 360 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
 361 "noptracefaultinfo\n"
 362 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
 363 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
 364 "    using the current skas3 patch.\n\n");
 365
 366 static int __init noptraceldt_cmd_param(char *str, int* add)
 367 {
 368         ptrace_ldt = 0;
 369         return 0;
 370 }
 371
 372 __uml_setup("noptraceldt", noptraceldt_cmd_param,
 373 "noptraceldt\n"
 374 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
 375 "    To support PTRACE_LDT, the host needs to be patched using\n"
 376 "    the current skas3 patch.\n\n");
 377
 378 #ifdef UML_CONFIG_MODE_SKAS
 379 static inline void check_skas3_ptrace_faultinfo(void)
 380 {
 381         struct ptrace_faultinfo fi;
 382         void *stack;
 383         int pid, n;
 384
 385         non_fatal("  - PTRACE_FAULTINFO...");
 386         pid = start_ptraced_child(&stack);
 387
 388         n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
 389         if (n < 0) {
 390                 ptrace_faultinfo = 0;
 391                 if(errno == EIO)
 392                         non_fatal("not found\n");
 393                 else
 394                         perror("not found");
 395         }
 396         else {
 397                 if (!ptrace_faultinfo)
 398                         non_fatal("found but disabled on command line\n");
 399                 else
 400                         non_fatal("found\n");
 401         }
 402
 403         init_registers(pid);
 404         stop_ptraced_child(pid, stack, 1, 1);
 405 }
 406
 407 static inline void check_skas3_ptrace_ldt(void)
 408 {
 409 #ifdef PTRACE_LDT
 410         void *stack;
 411         int pid, n;
 412         unsigned char ldtbuf[40];
 413         struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
 414                 .func = 2, /* read default ldt */
 415                 .ptr = ldtbuf,
 416                 .bytecount = sizeof(ldtbuf)};
 417
 418         non_fatal("  - PTRACE_LDT...");
 419         pid = start_ptraced_child(&stack);
 420
 421         n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
 422         if (n < 0) {
 423                 if(errno == EIO)
 424                         non_fatal("not found\n");
 425                 else {
 426                         perror("not found");
 427                 }
 428                 ptrace_ldt = 0;
 429         }
 430         else {
 431                 if(ptrace_ldt)
 432                         non_fatal("found\n");
 433                 else
 434                         non_fatal("found, but use is disabled\n");
 435         }
 436
 437         stop_ptraced_child(pid, stack, 1, 1);
 438 #else
 439         /* PTRACE_LDT might be disabled via cmdline option.
 440          * We want to override this, else we might use the stub
 441          * without real need
 442          */
 443         ptrace_ldt = 1;
 444 #endif
 445 }
 446
 447 static inline void check_skas3_proc_mm(void)
 448 {
 449         non_fatal("  - /proc/mm...");
 450         if (access("/proc/mm", W_OK) < 0) {
 451                 proc_mm = 0;
 452                 perror("not found");
 453         }
 454         else {
 455                 if (!proc_mm)
 456                         non_fatal("found but disabled on command line\n");
 457                 else
 458                         non_fatal("found\n");
 459         }
 460 }
 461
 462 int can_do_skas(void)
 463 {
 464         non_fatal("Checking for the skas3 patch in the host:\n");
 465
 466         check_skas3_proc_mm();
 467         check_skas3_ptrace_faultinfo();
 468         check_skas3_ptrace_ldt();
 469
 470         if(!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
 471                 skas_needs_stub = 1;
 472
 473         return 1;
 474 }
 475 #else
 476 int can_do_skas(void)
 477 {
 478         return 0;
 479 }
 480 #endif
 481
 482 int __init parse_iomem(char *str, int *add)
 483 {
 484         struct iomem_region *new;
 485         struct stat64 buf;
 486         char *file, *driver;
 487         int fd, size;
 488
 489         driver = str;
 490         file = strchr(str,',');
 491         if(file == NULL){
 492                 printf("parse_iomem : failed to parse iomem\n");
 493                 goto out;
 494         }
 495         *file = '\0';
 496         file++;
 497         fd = open(file, O_RDWR, 0);
 498         if(fd < 0){
 499                 os_print_error(fd, "parse_iomem - Couldn't open io file");
 500                 goto out;
 501         }
 502
 503         if(fstat64(fd, &buf) < 0){
 504                 perror("parse_iomem - cannot stat_fd file");
 505                 goto out_close;
 506         }
 507
 508         new = malloc(sizeof(*new));
 509         if(new == NULL){
 510                 perror("Couldn't allocate iomem_region struct");
 511                 goto out_close;
 512         }
 513
 514         size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
 515
 516         *new = ((struct iomem_region) { .next           = iomem_regions,
 517                                         .driver         = driver,
 518                                         .fd             = fd,
 519                                         .size           = size,
 520                                         .phys           = 0,
 521                                         .virt           = 0 });
 522         iomem_regions = new;
 523         iomem_size += new->size + UM_KERN_PAGE_SIZE;
 524
 525         return 0;
 526  out_close:
 527         close(fd);
 528  out:
 529         return 1;
 530 }
 531
 532
 533 /* Changed during early boot */
 534 int pty_output_sigio = 0;
 535 int pty_close_sigio = 0;
 536
 537 /* Used as a flag during SIGIO testing early in boot */
 538 static volatile int got_sigio = 0;
 539
 540 static void __init handler(int sig)
 541 {
 542         got_sigio = 1;
 543 }
 544
 545 struct openpty_arg {
 546         int master;
 547         int slave;
 548         int err;
 549 };
 550
 551 static void openpty_cb(void *arg)
 552 {
 553         struct openpty_arg *info = arg;
 554
 555         info->err = 0;
 556         if(openpty(&info->master, &info->slave, NULL, NULL, NULL))
 557                 info->err = -errno;
 558 }
 559
 560 static int async_pty(int master, int slave)
 561 {
 562         int flags;
 563
 564         flags = fcntl(master, F_GETFL);
 565         if(flags < 0)
 566                 return -errno;
 567
 568         if((fcntl(master, F_SETFL, flags | O_NONBLOCK | O_ASYNC) < 0) ||
 569            (fcntl(master, F_SETOWN, os_getpid()) < 0))
 570                 return -errno;
 571
 572         if((fcntl(slave, F_SETFL, flags | O_NONBLOCK) < 0))
 573                 return -errno;
 574
 575         return(0);
 576 }
 577
 578 static void __init check_one_sigio(void (*proc)(int, int))
 579 {
 580         struct sigaction old, new;
 581         struct openpty_arg pty = { .master = -1, .slave = -1 };
 582         int master, slave, err;
 583
 584         initial_thread_cb(openpty_cb, &pty);
 585         if(pty.err){
 586                 printk("openpty failed, errno = %d\n", -pty.err);
 587                 return;
 588         }
 589
 590         master = pty.master;
 591         slave = pty.slave;
 592
 593         if((master == -1) || (slave == -1)){
 594                 printk("openpty failed to allocate a pty\n");
 595                 return;
 596         }
 597
 598         /* Not now, but complain so we now where we failed. */
 599         err = raw(master);
 600         if (err < 0)
 601                 panic("check_sigio : __raw failed, errno = %d\n", -err);
 602
 603         err = async_pty(master, slave);
 604         if(err < 0)
 605                 panic("tty_fds : sigio_async failed, err = %d\n", -err);
 606
 607         if(sigaction(SIGIO, NULL, &old) < 0)
 608                 panic("check_sigio : sigaction 1 failed, errno = %d\n", errno);
 609         new = old;
 610         new.sa_handler = handler;
 611         if(sigaction(SIGIO, &new, NULL) < 0)
 612                 panic("check_sigio : sigaction 2 failed, errno = %d\n", errno);
 613
 614         got_sigio = 0;
 615         (*proc)(master, slave);
 616
 617         close(master);
 618         close(slave);
 619
 620         if(sigaction(SIGIO, &old, NULL) < 0)
 621                 panic("check_sigio : sigaction 3 failed, errno = %d\n", errno);
 622 }
 623
 624 static void tty_output(int master, int slave)
 625 {
 626         int n;
 627         char buf[512];
 628
 629         printk("Checking that host ptys support output SIGIO...");
 630
 631         memset(buf, 0, sizeof(buf));
 632
 633         while(os_write_file(master, buf, sizeof(buf)) > 0) ;
 634         if(errno != EAGAIN)
 635                 panic("check_sigio : write failed, errno = %d\n", errno);
 636         while(((n = os_read_file(slave, buf, sizeof(buf))) > 0) && !got_sigio) ;
 637
 638         if(got_sigio){
 639                 printk("Yes\n");
 640                 pty_output_sigio = 1;
 641         }
 642         else if(n == -EAGAIN) printk("No, enabling workaround\n");
 643         else panic("check_sigio : read failed, err = %d\n", n);
 644 }
 645
 646 static void tty_close(int master, int slave)
 647 {
 648         printk("Checking that host ptys support SIGIO on close...");
 649
 650         close(slave);
 651         if(got_sigio){
 652                 printk("Yes\n");
 653                 pty_close_sigio = 1;
 654         }
 655         else printk("No, enabling workaround\n");
 656 }
 657
 658 void __init check_sigio(void)
 659 {
 660         if((os_access("/dev/ptmx", OS_ACC_R_OK) < 0) &&
 661            (os_access("/dev/ptyp0", OS_ACC_R_OK) < 0)){
 662                 printk("No pseudo-terminals available - skipping pty SIGIO "
 663                        "check\n");
 664                 return;
 665         }
 666         check_one_sigio(tty_output);
 667         check_one_sigio(tty_close);
 668 }
 669
 670 void os_check_bugs(void)
 671 {
 672         check_ptrace();
 673         check_sigio();
 674 }
 675