SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/um/os-Linux/skas/process.c

   1 /*
   2  * Copyright (C) 2002- 2004 Jeff Dike (jdike@addtoit.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <stdlib.h>
   7 #include <string.h>
   8 #include <unistd.h>
   9 #include <errno.h>
  10 #include <signal.h>
  11 #include <setjmp.h>
  12 #include <sched.h>
  13 #include "ptrace_user.h"
  14 #include <sys/wait.h>
  15 #include <sys/mman.h>
  16 #include <sys/user.h>
  17 #include <sys/time.h>
  18 #include <asm/unistd.h>
  19 #include <asm/types.h>
  20 #include "user.h"
  21 #include "sysdep/ptrace.h"
  22 #include "user_util.h"
  23 #include "kern_util.h"
  24 #include "skas.h"
  25 #include "stub-data.h"
  26 #include "mm_id.h"
  27 #include "sysdep/sigcontext.h"
  28 #include "sysdep/stub.h"
  29 #include "os.h"
  30 #include "proc_mm.h"
  31 #include "skas_ptrace.h"
  32 #include "chan_user.h"
  33 #include "registers.h"
  34 #include "mem.h"
  35 #include "uml-config.h"
  36 #include "process.h"
  37
  38 int is_skas_winch(int pid, int fd, void *data)
  39 {
  40         if(pid != os_getpgrp())
  41                 return(0);
  42
  43         register_winch_irq(-1, fd, -1, data);
  44         return(1);
  45 }
  46
  47 void wait_stub_done(int pid, int sig, char * fname)
  48 {
  49         int n, status, err;
  50
  51         do {
  52                 if ( sig != -1 ) {
  53                         err = ptrace(PTRACE_CONT, pid, 0, sig);
  54                         if(err)
  55                                 panic("%s : continue failed, errno = %d\n",
  56                                       fname, errno);
  57                 }
  58                 sig = 0;
  59
  60                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
  61         } while((n >= 0) && WIFSTOPPED(status) &&
  62                 ((WSTOPSIG(status) == SIGVTALRM) ||
  63                  /* running UML inside a detached screen can cause
  64                   * SIGWINCHes
  65                   */
  66                  (WSTOPSIG(status) == SIGWINCH)));
  67
  68         if((n < 0) || !WIFSTOPPED(status) ||
  69            (WSTOPSIG(status) != SIGUSR1 && WSTOPSIG(status) != SIGTRAP)){
  70                 unsigned long regs[HOST_FRAME_SIZE];
  71
  72                 if(ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
  73                         printk("Failed to get registers from stub, "
  74                                "errno = %d\n", errno);
  75                 else {
  76                         int i;
  77
  78                         printk("Stub registers -\n");
  79                         for(i = 0; i < HOST_FRAME_SIZE; i++)
  80                                 printk("\t%d - %lx\n", i, regs[i]);
  81                 }
  82                 panic("%s : failed to wait for SIGUSR1/SIGTRAP, "
  83                       "pid = %d, n = %d, errno = %d, status = 0x%x\n",
  84                       fname, pid, n, errno, status);
  85         }
  86 }
  87
  88 extern unsigned long current_stub_stack(void);
  89
  90 void get_skas_faultinfo(int pid, struct faultinfo * fi)
  91 {
  92         int err;
  93
  94         if(ptrace_faultinfo){
  95                 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
  96                 if(err)
  97                         panic("get_skas_faultinfo - PTRACE_FAULTINFO failed, "
  98                               "errno = %d\n", errno);
  99
 100                 /* Special handling for i386, which has different structs */
 101                 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
 102                         memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
 103                                sizeof(struct faultinfo) -
 104                                sizeof(struct ptrace_faultinfo));
 105         }
 106         else {
 107                 wait_stub_done(pid, SIGSEGV, "get_skas_faultinfo");
 108
 109                 /* faultinfo is prepared by the stub-segv-handler at start of
 110                  * the stub stack page. We just have to copy it.
 111                  */
 112                 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
 113         }
 114 }
 115
 116 static void handle_segv(int pid, union uml_pt_regs * regs)
 117 {
 118         get_skas_faultinfo(pid, &regs->skas.faultinfo);
 119         segv(regs->skas.faultinfo, 0, 1, NULL);
 120 }
 121
 122 /*To use the same value of using_sysemu as the caller, ask it that value (in local_using_sysemu)*/
 123 static void handle_trap(int pid, union uml_pt_regs *regs, int local_using_sysemu)
 124 {
 125         int err, status;
 126
 127         /* Mark this as a syscall */
 128         UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->skas.regs);
 129
 130         if (!local_using_sysemu)
 131         {
 132                 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 133                              __NR_getpid);
 134                 if(err < 0)
 135                         panic("handle_trap - nullifying syscall failed errno = %d\n",
 136                               errno);
 137
 138                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
 139                 if(err < 0)
 140                         panic("handle_trap - continuing to end of syscall failed, "
 141                               "errno = %d\n", errno);
 142
 143                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 144                 if((err < 0) || !WIFSTOPPED(status) ||
 145                    (WSTOPSIG(status) != SIGTRAP + 0x80))
 146                         panic("handle_trap - failed to wait at end of syscall, "
 147                               "errno = %d, status = %d\n", errno, status);
 148         }
 149
 150         handle_syscall(regs);
 151 }
 152
 153 extern int __syscall_stub_start;
 154
 155 static int userspace_tramp(void *stack)
 156 {
 157         void *addr;
 158
 159         ptrace(PTRACE_TRACEME, 0, 0, 0);
 160
 161         init_new_thread_signals(1);
 162         enable_timer();
 163
 164         if(!proc_mm){
 165                 /* This has a pte, but it can't be mapped in with the usual
 166                  * tlb_flush mechanism because this is part of that mechanism
 167                  */
 168                 int fd;
 169                 __u64 offset;
 170                 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
 171                 addr = mmap64((void *) UML_CONFIG_STUB_CODE, page_size(),
 172                               PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
 173                 if(addr == MAP_FAILED){
 174                         printk("mapping mmap stub failed, errno = %d\n",
 175                                errno);
 176                         exit(1);
 177                 }
 178
 179                 if(stack != NULL){
 180                         fd = phys_mapping(to_phys(stack), &offset);
 181                         addr = mmap((void *) UML_CONFIG_STUB_DATA, page_size(),
 182                                     PROT_READ | PROT_WRITE,
 183                                     MAP_FIXED | MAP_SHARED, fd, offset);
 184                         if(addr == MAP_FAILED){
 185                                 printk("mapping segfault stack failed, "
 186                                        "errno = %d\n", errno);
 187                                 exit(1);
 188                         }
 189                 }
 190         }
 191         if(!ptrace_faultinfo && (stack != NULL)){
 192                 unsigned long v = UML_CONFIG_STUB_CODE +
 193                                   (unsigned long) stub_segv_handler -
 194                                   (unsigned long) &__syscall_stub_start;
 195
 196                 set_sigstack((void *) UML_CONFIG_STUB_DATA, page_size());
 197                 set_handler(SIGSEGV, (void *) v, SA_ONSTACK,
 198                             SIGIO, SIGWINCH, SIGALRM, SIGVTALRM,
 199                             SIGUSR1, -1);
 200         }
 201
 202         os_stop_process(os_getpid());
 203         return(0);
 204 }
 205
 206 /* Each element set once, and only accessed by a single processor anyway */
 207 #undef NR_CPUS
 208 #define NR_CPUS 1
 209 int userspace_pid[NR_CPUS];
 210
 211 int start_userspace(unsigned long stub_stack)
 212 {
 213         void *stack;
 214         unsigned long sp;
 215         int pid, status, n, flags;
 216
 217         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 218                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 219         if(stack == MAP_FAILED)
 220                 panic("start_userspace : mmap failed, errno = %d", errno);
 221         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
 222
 223         flags = CLONE_FILES | SIGCHLD;
 224         if(proc_mm) flags |= CLONE_VM;
 225         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 226         if(pid < 0)
 227                 panic("start_userspace : clone failed, errno = %d", errno);
 228
 229         do {
 230                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 231                 if(n < 0)
 232                         panic("start_userspace : wait failed, errno = %d",
 233                               errno);
 234         } while(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
 235
 236         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 237                 panic("start_userspace : expected SIGSTOP, got status = %d",
 238                       status);
 239
 240         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, (void *)PTRACE_O_TRACESYSGOOD) < 0)
 241                 panic("start_userspace : PTRACE_OLDSETOPTIONS failed, errno=%d\n",
 242                       errno);
 243
 244         if(munmap(stack, PAGE_SIZE) < 0)
 245                 panic("start_userspace : munmap failed, errno = %d\n", errno);
 246
 247         return(pid);
 248 }
 249
 250 void userspace(union uml_pt_regs *regs)
 251 {
 252         int err, status, op, pid = userspace_pid[0];
 253         int local_using_sysemu; /*To prevent races if using_sysemu changes under us.*/
 254
 255         while(1){
 256                 restore_registers(pid, regs);
 257
 258                 /* Now we set local_using_sysemu to be used for one loop */
 259                 local_using_sysemu = get_using_sysemu();
 260
 261                 op = SELECT_PTRACE_OPERATION(local_using_sysemu, singlestepping(NULL));
 262
 263                 err = ptrace(op, pid, 0, 0);
 264                 if(err)
 265                         panic("userspace - could not resume userspace process, "
 266                               "pid=%d, ptrace operation = %d, errno = %d\n",
 267                               op, errno);
 268
 269                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 270                 if(err < 0)
 271                         panic("userspace - waitpid failed, errno = %d\n",
 272                               errno);
 273
 274                 regs->skas.is_user = 1;
 275                 save_registers(pid, regs);
 276                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
 277
 278                 if(WIFSTOPPED(status)){
 279                         switch(WSTOPSIG(status)){
 280                         case SIGSEGV:
 281                                 if(PTRACE_FULL_FAULTINFO || !ptrace_faultinfo)
 282                                         user_signal(SIGSEGV, regs, pid);
 283                                 else handle_segv(pid, regs);
 284                                 break;
 285                         case SIGTRAP + 0x80:
 286                                 handle_trap(pid, regs, local_using_sysemu);
 287                                 break;
 288                         case SIGTRAP:
 289                                 relay_signal(SIGTRAP, regs);
 290                                 break;
 291                         case SIGIO:
 292                         case SIGVTALRM:
 293                         case SIGILL:
 294                         case SIGBUS:
 295                         case SIGFPE:
 296                         case SIGWINCH:
 297                                 user_signal(WSTOPSIG(status), regs, pid);
 298                                 break;
 299                         default:
 300                                 printk("userspace - child stopped with signal "
 301                                        "%d\n", WSTOPSIG(status));
 302                         }
 303                         pid = userspace_pid[0];
 304                         interrupt_end();
 305
 306                         /* Avoid -ERESTARTSYS handling in host */
 307                         if(PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
 308                                 PT_SYSCALL_NR(regs->skas.regs) = -1;
 309                 }
 310         }
 311 }
 312 #define INIT_JMP_NEW_THREAD 0
 313 #define INIT_JMP_REMOVE_SIGSTACK 1
 314 #define INIT_JMP_CALLBACK 2
 315 #define INIT_JMP_HALT 3
 316 #define INIT_JMP_REBOOT 4
 317
 318 int copy_context_skas0(unsigned long new_stack, int pid)
 319 {
 320         int err;
 321         unsigned long regs[MAX_REG_NR];
 322         unsigned long current_stack = current_stub_stack();
 323         struct stub_data *data = (struct stub_data *) current_stack;
 324         struct stub_data *child_data = (struct stub_data *) new_stack;
 325         __u64 new_offset;
 326         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 327
 328         /* prepare offset and fd of child's stack as argument for parent's
 329          * and child's mmap2 calls
 330          */
 331         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
 332                                       .fd       = new_fd,
 333                                       .timer    = ((struct itimerval)
 334                                                     { { 0, 1000000 / hz() },
 335                                                       { 0, 1000000 / hz() }})});
 336         get_safe_registers(regs);
 337
 338         /* Set parent's instruction pointer to start of clone-stub */
 339         regs[REGS_IP_INDEX] = UML_CONFIG_STUB_CODE +
 340                                 (unsigned long) stub_clone_handler -
 341                                 (unsigned long) &__syscall_stub_start;
 342         regs[REGS_SP_INDEX] = UML_CONFIG_STUB_DATA + PAGE_SIZE -
 343                 sizeof(void *);
 344 #ifdef __SIGNAL_FRAMESIZE
 345         regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
 346 #endif
 347         err = ptrace_setregs(pid, regs);
 348         if(err < 0)
 349                 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
 350                       "pid = %d, errno = %d\n", pid, errno);
 351
 352         /* set a well known return code for detection of child write failure */
 353         child_data->err = 12345678;
 354
 355         /* Wait, until parent has finished its work: read child's pid from
 356          * parent's stack, and check, if bad result.
 357          */
 358         wait_stub_done(pid, 0, "copy_context_skas0");
 359
 360         pid = data->err;
 361         if(pid < 0)
 362                 panic("copy_context_skas0 - stub-parent reports error %d\n",
 363                       pid);
 364
 365         /* Wait, until child has finished too: read child's result from
 366          * child's stack and check it.
 367          */
 368         wait_stub_done(pid, -1, "copy_context_skas0");
 369         if (child_data->err != UML_CONFIG_STUB_DATA)
 370                 panic("copy_context_skas0 - stub-child reports error %d\n",
 371                       child_data->err);
 372
 373         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 374                    (void *)PTRACE_O_TRACESYSGOOD) < 0)
 375                 panic("copy_context_skas0 : PTRACE_OLDSETOPTIONS failed, "
 376                       "errno = %d\n", errno);
 377
 378         return pid;
 379 }
 380
 381 /*
 382  * This is used only, if stub pages are needed, while proc_mm is
 383  * availabl. Opening /proc/mm creates a new mm_context, which lacks
 384  * the stub-pages. Thus, we map them using /proc/mm-fd
 385  */
 386 void map_stub_pages(int fd, unsigned long code,
 387                     unsigned long data, unsigned long stack)
 388 {
 389         struct proc_mm_op mmop;
 390         int n;
 391         __u64 code_offset;
 392         int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
 393                                    &code_offset);
 394
 395         mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
 396                                       .u         =
 397                                       { .mmap    =
 398                                         { .addr    = code,
 399                                           .len     = PAGE_SIZE,
 400                                           .prot    = PROT_EXEC,
 401                                           .flags   = MAP_FIXED | MAP_PRIVATE,
 402                                           .fd      = code_fd,
 403                                           .offset  = code_offset
 404         } } });
 405         n = os_write_file(fd, &mmop, sizeof(mmop));
 406         if(n != sizeof(mmop))
 407                 panic("map_stub_pages : /proc/mm map for code failed, "
 408                       "err = %d\n", -n);
 409
 410         if ( stack ) {
 411                 __u64 map_offset;
 412                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
 413                 mmop = ((struct proc_mm_op)
 414                                 { .op        = MM_MMAP,
 415                                   .u         =
 416                                   { .mmap    =
 417                                     { .addr    = data,
 418                                       .len     = PAGE_SIZE,
 419                                       .prot    = PROT_READ | PROT_WRITE,
 420                                       .flags   = MAP_FIXED | MAP_SHARED,
 421                                       .fd      = map_fd,
 422                                       .offset  = map_offset
 423                 } } });
 424                 n = os_write_file(fd, &mmop, sizeof(mmop));
 425                 if(n != sizeof(mmop))
 426                         panic("map_stub_pages : /proc/mm map for data failed, "
 427                               "err = %d\n", -n);
 428         }
 429 }
 430
 431 void new_thread(void *stack, void **switch_buf_ptr, void **fork_buf_ptr,
 432                 void (*handler)(int))
 433 {
 434         unsigned long flags;
 435         sigjmp_buf switch_buf, fork_buf;
 436
 437         *switch_buf_ptr = &switch_buf;
 438         *fork_buf_ptr = &fork_buf;
 439
 440         /* Somewhat subtle - siglongjmp restores the signal mask before doing
 441          * the longjmp.  This means that when jumping from one stack to another
 442          * when the target stack has interrupts enabled, an interrupt may occur
 443          * on the source stack.  This is bad when starting up a process because
 444          * it's not supposed to get timer ticks until it has been scheduled.
 445          * So, we disable interrupts around the sigsetjmp to ensure that
 446          * they can't happen until we get back here where they are safe.
 447          */
 448         flags = get_signals();
 449         block_signals();
 450         if(sigsetjmp(fork_buf, 1) == 0)
 451                 new_thread_proc(stack, handler);
 452
 453         remove_sigstack();
 454
 455         set_signals(flags);
 456 }
 457
 458 void thread_wait(void *sw, void *fb)
 459 {
 460         sigjmp_buf buf, **switch_buf = sw, *fork_buf;
 461
 462         *switch_buf = &buf;
 463         fork_buf = fb;
 464         if(sigsetjmp(buf, 1) == 0)
 465                 siglongjmp(*fork_buf, INIT_JMP_REMOVE_SIGSTACK);
 466 }
 467
 468 void switch_threads(void *me, void *next)
 469 {
 470         sigjmp_buf my_buf, **me_ptr = me, *next_buf = next;
 471
 472         *me_ptr = &my_buf;
 473         if(sigsetjmp(my_buf, 1) == 0)
 474                 siglongjmp(*next_buf, 1);
 475 }
 476
 477 static sigjmp_buf initial_jmpbuf;
 478
 479 /* XXX Make these percpu */
 480 static void (*cb_proc)(void *arg);
 481 static void *cb_arg;
 482 static sigjmp_buf *cb_back;
 483
 484 int start_idle_thread(void *stack, void *switch_buf_ptr, void **fork_buf_ptr)
 485 {
 486         sigjmp_buf **switch_buf = switch_buf_ptr;
 487         int n;
 488
 489         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
 490                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGALRM,
 491                     SIGVTALRM, -1);
 492
 493         *fork_buf_ptr = &initial_jmpbuf;
 494         n = sigsetjmp(initial_jmpbuf, 1);
 495         switch(n){
 496         case INIT_JMP_NEW_THREAD:
 497                 new_thread_proc((void *) stack, new_thread_handler);
 498                 break;
 499         case INIT_JMP_REMOVE_SIGSTACK:
 500                 remove_sigstack();
 501                 break;
 502         case INIT_JMP_CALLBACK:
 503                 (*cb_proc)(cb_arg);
 504                 siglongjmp(*cb_back, 1);
 505                 break;
 506         case INIT_JMP_HALT:
 507                 kmalloc_ok = 0;
 508                 return(0);
 509         case INIT_JMP_REBOOT:
 510                 kmalloc_ok = 0;
 511                 return(1);
 512         default:
 513                 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
 514         }
 515         siglongjmp(**switch_buf, 1);
 516 }
 517
 518 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 519 {
 520         sigjmp_buf here;
 521
 522         cb_proc = proc;
 523         cb_arg = arg;
 524         cb_back = &here;
 525
 526         block_signals();
 527         if(sigsetjmp(here, 1) == 0)
 528                 siglongjmp(initial_jmpbuf, INIT_JMP_CALLBACK);
 529         unblock_signals();
 530
 531         cb_proc = NULL;
 532         cb_arg = NULL;
 533         cb_back = NULL;
 534 }
 535
 536 void halt_skas(void)
 537 {
 538         block_signals();
 539         siglongjmp(initial_jmpbuf, INIT_JMP_HALT);
 540 }
 541
 542 void reboot_skas(void)
 543 {
 544         block_signals();
 545         siglongjmp(initial_jmpbuf, INIT_JMP_REBOOT);
 546 }
 547
 548 void switch_mm_skas(struct mm_id *mm_idp)
 549 {
 550         int err;
 551
 552 #warning need cpu pid in switch_mm_skas
 553         if(proc_mm){
 554                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
 555                              mm_idp->u.mm_fd);
 556                 if(err)
 557                         panic("switch_mm_skas - PTRACE_SWITCH_MM failed, "
 558                               "errno = %d\n", errno);
 559         }
 560         else userspace_pid[0] = mm_idp->u.pid;
 561 }