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         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 /* Changed only during early boot */
 137 int ptrace_faultinfo = 1;
 138 int ptrace_ldt = 1;
 139 int proc_mm = 1;
 140 int skas_needs_stub = 0;
 141
 142 static int __init skas0_cmd_param(char *str, int* add)
 143 {
 144         ptrace_faultinfo = proc_mm = 0;
 145         return 0;
 146 }
 147
 148 /* The two __uml_setup would conflict, without this stupid alias. */
 149
 150 static int __init mode_skas0_cmd_param(char *str, int* add)
 151         __attribute__((alias("skas0_cmd_param")));
 152
 153 __uml_setup("skas0", skas0_cmd_param,
 154                 "skas0\n"
 155                 "    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
 156                 "    you specify mode=tt.\n\n");
 157
 158 __uml_setup("mode=skas0", mode_skas0_cmd_param,
 159                 "mode=skas0\n"
 160                 "    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
 161                 "    specify mode=tt. Note that this was recently added - on \n"
 162                 "    older kernels you must use simply \"skas0\".\n\n");
 163
 164 /* Changed only during early boot */
 165 static int force_sysemu_disabled = 0;
 166
 167 static int __init nosysemu_cmd_param(char *str, int* add)
 168 {
 169         force_sysemu_disabled = 1;
 170         return 0;
 171 }
 172
 173 __uml_setup("nosysemu", nosysemu_cmd_param,
 174 "nosysemu\n"
 175 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
 176 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
 177 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
 178 "    To make it working, you need a kernel patch for your host, too.\n"
 179 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
 180 "    information.\n\n");
 181
 182 static void __init check_sysemu(void)
 183 {
 184         void *stack;
 185         int pid, n, status, count=0;
 186
 187         printf("Checking syscall emulation patch for ptrace...");
 188         sysemu_supported = 0;
 189         pid = start_ptraced_child(&stack);
 190
 191         if(ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
 192                 goto fail;
 193
 194         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 195         if (n < 0)
 196                 panic("check_sysemu : wait failed, errno = %d", errno);
 197         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
 198                 panic("check_sysemu : expected SIGTRAP, "
 199                       "got status = %d", status);
 200
 201         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
 202                    os_getpid());
 203         if(n < 0)
 204                 panic("check_sysemu : failed to modify system "
 205                       "call return, errno = %d", errno);
 206
 207         if (stop_ptraced_child(pid, stack, 0, 0) < 0)
 208                 goto fail_stopped;
 209
 210         sysemu_supported = 1;
 211         printf("OK\n");
 212         set_using_sysemu(!force_sysemu_disabled);
 213
 214         printf("Checking advanced syscall emulation patch for ptrace...");
 215         pid = start_ptraced_child(&stack);
 216
 217         if(ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
 218                   (void *) PTRACE_O_TRACESYSGOOD) < 0)
 219                 panic("check_ptrace: PTRACE_OLDSETOPTIONS failed, errno = %d",
 220                       errno);
 221
 222         while(1){
 223                 count++;
 224                 if(ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
 225                         goto fail;
 226                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 227                 if(n < 0)
 228                         panic("check_ptrace : wait failed, errno = %d", errno);
 229                 if(WIFSTOPPED(status) && (WSTOPSIG(status) == (SIGTRAP|0x80))){
 230                         if (!count)
 231                                 panic("check_ptrace : SYSEMU_SINGLESTEP "
 232                                       "doesn't singlestep");
 233                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
 234                                    os_getpid());
 235                         if(n < 0)
 236                                 panic("check_sysemu : failed to modify system "
 237                                       "call return, errno = %d", errno);
 238                         break;
 239                 }
 240                 else if(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
 241                         count++;
 242                 else
 243                         panic("check_ptrace : expected SIGTRAP or "
 244                               "(SIGTRAP|0x80), got status = %d", status);
 245         }
 246         if (stop_ptraced_child(pid, stack, 0, 0) < 0)
 247                 goto fail_stopped;
 248
 249         sysemu_supported = 2;
 250         printf("OK\n");
 251
 252         if ( !force_sysemu_disabled )
 253                 set_using_sysemu(sysemu_supported);
 254         return;
 255
 256 fail:
 257         stop_ptraced_child(pid, stack, 1, 0);
 258 fail_stopped:
 259         printf("missing\n");
 260 }
 261
 262 static void __init check_ptrace(void)
 263 {
 264         void *stack;
 265         int pid, syscall, n, status;
 266
 267         printf("Checking that ptrace can change system call numbers...");
 268         pid = start_ptraced_child(&stack);
 269
 270         if(ptrace(PTRACE_OLDSETOPTIONS, pid, 0, (void *)PTRACE_O_TRACESYSGOOD) < 0)
 271                 panic("check_ptrace: PTRACE_OLDSETOPTIONS failed, errno = %d", errno);
 272
 273         while(1){
 274                 if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
 275                         panic("check_ptrace : ptrace failed, errno = %d",
 276                               errno);
 277                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 278                 if(n < 0)
 279                         panic("check_ptrace : wait failed, errno = %d", errno);
 280                 if(!WIFSTOPPED(status) || (WSTOPSIG(status) != (SIGTRAP|0x80)))
 281                         panic("check_ptrace : expected (SIGTRAP|0x80), "
 282                               "got status = %d", status);
 283
 284                 syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
 285                                  0);
 286                 if(syscall == __NR_getpid){
 287                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 288                                    __NR_getppid);
 289                         if(n < 0)
 290                                 panic("check_ptrace : failed to modify system "
 291                                       "call, errno = %d", errno);
 292                         break;
 293                 }
 294         }
 295         stop_ptraced_child(pid, stack, 0, 1);
 296         printf("OK\n");
 297         check_sysemu();
 298 }
 299
 300 extern void check_tmpexec(void);
 301
 302 void os_early_checks(void)
 303 {
 304         check_ptrace();
 305
 306         /* Need to check this early because mmapping happens before the
 307          * kernel is running.
 308          */
 309         check_tmpexec();
 310 }
 311
 312 static int __init noprocmm_cmd_param(char *str, int* add)
 313 {
 314         proc_mm = 0;
 315         return 0;
 316 }
 317
 318 __uml_setup("noprocmm", noprocmm_cmd_param,
 319 "noprocmm\n"
 320 "    Turns off usage of /proc/mm, even if host supports it.\n"
 321 "    To support /proc/mm, the host needs to be patched using\n"
 322 "    the current skas3 patch.\n\n");
 323
 324 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
 325 {
 326         ptrace_faultinfo = 0;
 327         return 0;
 328 }
 329
 330 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
 331 "noptracefaultinfo\n"
 332 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
 333 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
 334 "    using the current skas3 patch.\n\n");
 335
 336 static int __init noptraceldt_cmd_param(char *str, int* add)
 337 {
 338         ptrace_ldt = 0;
 339         return 0;
 340 }
 341
 342 __uml_setup("noptraceldt", noptraceldt_cmd_param,
 343 "noptraceldt\n"
 344 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
 345 "    To support PTRACE_LDT, the host needs to be patched using\n"
 346 "    the current skas3 patch.\n\n");
 347
 348 #ifdef UML_CONFIG_MODE_SKAS
 349 static inline void check_skas3_ptrace_faultinfo(void)
 350 {
 351         struct ptrace_faultinfo fi;
 352         void *stack;
 353         int pid, n;
 354
 355         printf("  - PTRACE_FAULTINFO...");
 356         pid = start_ptraced_child(&stack);
 357
 358         n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
 359         if (n < 0) {
 360                 ptrace_faultinfo = 0;
 361                 if(errno == EIO)
 362                         printf("not found\n");
 363                 else
 364                         perror("not found");
 365         }
 366         else {
 367                 if (!ptrace_faultinfo)
 368                         printf("found but disabled on command line\n");
 369                 else
 370                         printf("found\n");
 371         }
 372
 373         init_registers(pid);
 374         stop_ptraced_child(pid, stack, 1, 1);
 375 }
 376
 377 static inline void check_skas3_ptrace_ldt(void)
 378 {
 379 #ifdef PTRACE_LDT
 380         void *stack;
 381         int pid, n;
 382         unsigned char ldtbuf[40];
 383         struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
 384                 .func = 2, /* read default ldt */
 385                 .ptr = ldtbuf,
 386                 .bytecount = sizeof(ldtbuf)};
 387
 388         printf("  - PTRACE_LDT...");
 389         pid = start_ptraced_child(&stack);
 390
 391         n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
 392         if (n < 0) {
 393                 if(errno == EIO)
 394                         printf("not found\n");
 395                 else {
 396                         perror("not found");
 397                 }
 398                 ptrace_ldt = 0;
 399         }
 400         else {
 401                 if(ptrace_ldt)
 402                         printf("found\n");
 403                 else
 404                         printf("found, but use is disabled\n");
 405         }
 406
 407         stop_ptraced_child(pid, stack, 1, 1);
 408 #else
 409         /* PTRACE_LDT might be disabled via cmdline option.
 410          * We want to override this, else we might use the stub
 411          * without real need
 412          */
 413         ptrace_ldt = 1;
 414 #endif
 415 }
 416
 417 static inline void check_skas3_proc_mm(void)
 418 {
 419         printf("  - /proc/mm...");
 420         if (os_access("/proc/mm", OS_ACC_W_OK) < 0) {
 421                 proc_mm = 0;
 422                 printf("not found\n");
 423         }
 424         else {
 425                 if (!proc_mm)
 426                         printf("found but disabled on command line\n");
 427                 else
 428                         printf("found\n");
 429         }
 430 }
 431
 432 int can_do_skas(void)
 433 {
 434         printf("Checking for the skas3 patch in the host:\n");
 435
 436         check_skas3_proc_mm();
 437         check_skas3_ptrace_faultinfo();
 438         check_skas3_ptrace_ldt();
 439
 440         if(!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
 441                 skas_needs_stub = 1;
 442
 443         return 1;
 444 }
 445 #else
 446 int can_do_skas(void)
 447 {
 448         return 0;
 449 }
 450 #endif
 451
 452 int __init parse_iomem(char *str, int *add)
 453 {
 454         struct iomem_region *new;
 455         struct uml_stat buf;
 456         char *file, *driver;
 457         int fd, err, size;
 458
 459         driver = str;
 460         file = strchr(str,',');
 461         if(file == NULL){
 462                 printf("parse_iomem : failed to parse iomem\n");
 463                 goto out;
 464         }
 465         *file = '\0';
 466         file++;
 467         fd = os_open_file(file, of_rdwr(OPENFLAGS()), 0);
 468         if(fd < 0){
 469                 os_print_error(fd, "parse_iomem - Couldn't open io file");
 470                 goto out;
 471         }
 472
 473         err = os_stat_fd(fd, &buf);
 474         if(err < 0){
 475                 os_print_error(err, "parse_iomem - cannot stat_fd file");
 476                 goto out_close;
 477         }
 478
 479         new = malloc(sizeof(*new));
 480         if(new == NULL){
 481                 perror("Couldn't allocate iomem_region struct");
 482                 goto out_close;
 483         }
 484
 485         size = (buf.ust_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
 486
 487         *new = ((struct iomem_region) { .next           = iomem_regions,
 488                                         .driver         = driver,
 489                                         .fd             = fd,
 490                                         .size           = size,
 491                                         .phys           = 0,
 492                                         .virt           = 0 });
 493         iomem_regions = new;
 494         iomem_size += new->size + UM_KERN_PAGE_SIZE;
 495
 496         return 0;
 497  out_close:
 498         os_close_file(fd);
 499  out:
 500         return 1;
 501 }
 502
 503
 504 /* Changed during early boot */
 505 int pty_output_sigio = 0;
 506 int pty_close_sigio = 0;
 507
 508 /* Used as a flag during SIGIO testing early in boot */
 509 static volatile int got_sigio = 0;
 510
 511 static void __init handler(int sig)
 512 {
 513         got_sigio = 1;
 514 }
 515
 516 struct openpty_arg {
 517         int master;
 518         int slave;
 519         int err;
 520 };
 521
 522 static void openpty_cb(void *arg)
 523 {
 524         struct openpty_arg *info = arg;
 525
 526         info->err = 0;
 527         if(openpty(&info->master, &info->slave, NULL, NULL, NULL))
 528                 info->err = -errno;
 529 }
 530
 531 static void __init check_one_sigio(void (*proc)(int, int))
 532 {
 533         struct sigaction old, new;
 534         struct openpty_arg pty = { .master = -1, .slave = -1 };
 535         int master, slave, err;
 536
 537         initial_thread_cb(openpty_cb, &pty);
 538         if(pty.err){
 539                 printk("openpty failed, errno = %d\n", -pty.err);
 540                 return;
 541         }
 542
 543         master = pty.master;
 544         slave = pty.slave;
 545
 546         if((master == -1) || (slave == -1)){
 547                 printk("openpty failed to allocate a pty\n");
 548                 return;
 549         }
 550
 551         /* Not now, but complain so we now where we failed. */
 552         err = raw(master);
 553         if (err < 0)
 554                 panic("check_sigio : __raw failed, errno = %d\n", -err);
 555
 556         err = os_sigio_async(master, slave);
 557         if(err < 0)
 558                 panic("tty_fds : sigio_async failed, err = %d\n", -err);
 559
 560         if(sigaction(SIGIO, NULL, &old) < 0)
 561                 panic("check_sigio : sigaction 1 failed, errno = %d\n", errno);
 562         new = old;
 563         new.sa_handler = handler;
 564         if(sigaction(SIGIO, &new, NULL) < 0)
 565                 panic("check_sigio : sigaction 2 failed, errno = %d\n", errno);
 566
 567         got_sigio = 0;
 568         (*proc)(master, slave);
 569
 570         close(master);
 571         close(slave);
 572
 573         if(sigaction(SIGIO, &old, NULL) < 0)
 574                 panic("check_sigio : sigaction 3 failed, errno = %d\n", errno);
 575 }
 576
 577 static void tty_output(int master, int slave)
 578 {
 579         int n;
 580         char buf[512];
 581
 582         printk("Checking that host ptys support output SIGIO...");
 583
 584         memset(buf, 0, sizeof(buf));
 585
 586         while(os_write_file(master, buf, sizeof(buf)) > 0) ;
 587         if(errno != EAGAIN)
 588                 panic("check_sigio : write failed, errno = %d\n", errno);
 589         while(((n = os_read_file(slave, buf, sizeof(buf))) > 0) && !got_sigio) ;
 590
 591         if(got_sigio){
 592                 printk("Yes\n");
 593                 pty_output_sigio = 1;
 594         }
 595         else if(n == -EAGAIN) printk("No, enabling workaround\n");
 596         else panic("check_sigio : read failed, err = %d\n", n);
 597 }
 598
 599 static void tty_close(int master, int slave)
 600 {
 601         printk("Checking that host ptys support SIGIO on close...");
 602
 603         close(slave);
 604         if(got_sigio){
 605                 printk("Yes\n");
 606                 pty_close_sigio = 1;
 607         }
 608         else printk("No, enabling workaround\n");
 609 }
 610
 611 void __init check_sigio(void)
 612 {
 613         if((os_access("/dev/ptmx", OS_ACC_R_OK) < 0) &&
 614            (os_access("/dev/ptyp0", OS_ACC_R_OK) < 0)){
 615                 printk("No pseudo-terminals available - skipping pty SIGIO "
 616                        "check\n");
 617                 return;
 618         }
 619         check_one_sigio(tty_output);
 620         check_one_sigio(tty_close);
 621 }
 622
 623 void os_check_bugs(void)
 624 {
 625         check_ptrace();
 626         check_sigio();
 627 }
 628