Pileus Git - ~andy/linux/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 <sched.h>
  12 #include "ptrace_user.h"
  13 #include <sys/wait.h>
  14 #include <sys/mman.h>
  15 #include <sys/user.h>
  16 #include <sys/time.h>
  17 #include <asm/unistd.h>
  18 #include <asm/types.h>
  19 #include "user.h"
  20 #include "sysdep/ptrace.h"
  21 #include "user_util.h"
  22 #include "kern_util.h"
  23 #include "skas.h"
  24 #include "stub-data.h"
  25 #include "mm_id.h"
  26 #include "sysdep/sigcontext.h"
  27 #include "sysdep/stub.h"
  28 #include "os.h"
  29 #include "proc_mm.h"
  30 #include "skas_ptrace.h"
  31 #include "chan_user.h"
  32 #include "registers.h"
  33 #include "mem.h"
  34 #include "uml-config.h"
  35 #include "process.h"
  36 #include "longjmp.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();
 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                 struct sigaction sa;
 193
 194                 unsigned long v = UML_CONFIG_STUB_CODE +
 195                                   (unsigned long) stub_segv_handler -
 196                                   (unsigned long) &__syscall_stub_start;
 197
 198                 set_sigstack((void *) UML_CONFIG_STUB_DATA, page_size());
 199                 sigemptyset(&sa.sa_mask);
 200                 sigaddset(&sa.sa_mask, SIGIO);
 201                 sigaddset(&sa.sa_mask, SIGWINCH);
 202                 sigaddset(&sa.sa_mask, SIGALRM);
 203                 sigaddset(&sa.sa_mask, SIGVTALRM);
 204                 sigaddset(&sa.sa_mask, SIGUSR1);
 205                 sa.sa_flags = SA_ONSTACK;
 206                 sa.sa_handler = (void *) v;
 207                 sa.sa_restorer = NULL;
 208                 if(sigaction(SIGSEGV, &sa, NULL) < 0)
 209                         panic("userspace_tramp - setting SIGSEGV handler "
 210                               "failed - errno = %d\n", errno);
 211         }
 212
 213         os_stop_process(os_getpid());
 214         return(0);
 215 }
 216
 217 /* Each element set once, and only accessed by a single processor anyway */
 218 #undef NR_CPUS
 219 #define NR_CPUS 1
 220 int userspace_pid[NR_CPUS];
 221
 222 int start_userspace(unsigned long stub_stack)
 223 {
 224         void *stack;
 225         unsigned long sp;
 226         int pid, status, n, flags;
 227
 228         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 229                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 230         if(stack == MAP_FAILED)
 231                 panic("start_userspace : mmap failed, errno = %d", errno);
 232         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
 233
 234         flags = CLONE_FILES | SIGCHLD;
 235         if(proc_mm) flags |= CLONE_VM;
 236         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 237         if(pid < 0)
 238                 panic("start_userspace : clone failed, errno = %d", errno);
 239
 240         do {
 241                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 242                 if(n < 0)
 243                         panic("start_userspace : wait failed, errno = %d",
 244                               errno);
 245         } while(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
 246
 247         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 248                 panic("start_userspace : expected SIGSTOP, got status = %d",
 249                       status);
 250
 251         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, (void *)PTRACE_O_TRACESYSGOOD) < 0)
 252                 panic("start_userspace : PTRACE_OLDSETOPTIONS failed, errno=%d\n",
 253                       errno);
 254
 255         if(munmap(stack, PAGE_SIZE) < 0)
 256                 panic("start_userspace : munmap failed, errno = %d\n", errno);
 257
 258         return(pid);
 259 }
 260
 261 void userspace(union uml_pt_regs *regs)
 262 {
 263         int err, status, op, pid = userspace_pid[0];
 264         int local_using_sysemu; /*To prevent races if using_sysemu changes under us.*/
 265
 266         while(1){
 267                 restore_registers(pid, regs);
 268
 269                 /* Now we set local_using_sysemu to be used for one loop */
 270                 local_using_sysemu = get_using_sysemu();
 271
 272                 op = SELECT_PTRACE_OPERATION(local_using_sysemu, singlestepping(NULL));
 273
 274                 err = ptrace(op, pid, 0, 0);
 275                 if(err)
 276                         panic("userspace - could not resume userspace process, "
 277                               "pid=%d, ptrace operation = %d, errno = %d\n",
 278                               pid, op, errno);
 279
 280                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 281                 if(err < 0)
 282                         panic("userspace - waitpid failed, errno = %d\n",
 283                               errno);
 284
 285                 regs->skas.is_user = 1;
 286                 save_registers(pid, regs);
 287                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
 288
 289                 if(WIFSTOPPED(status)){
 290                         switch(WSTOPSIG(status)){
 291                         case SIGSEGV:
 292                                 if(PTRACE_FULL_FAULTINFO || !ptrace_faultinfo)
 293                                         user_signal(SIGSEGV, regs, pid);
 294                                 else handle_segv(pid, regs);
 295                                 break;
 296                         case SIGTRAP + 0x80:
 297                                 handle_trap(pid, regs, local_using_sysemu);
 298                                 break;
 299                         case SIGTRAP:
 300                                 relay_signal(SIGTRAP, regs);
 301                                 break;
 302                         case SIGIO:
 303                         case SIGVTALRM:
 304                         case SIGILL:
 305                         case SIGBUS:
 306                         case SIGFPE:
 307                         case SIGWINCH:
 308                                 user_signal(WSTOPSIG(status), regs, pid);
 309                                 break;
 310                         default:
 311                                 printk("userspace - child stopped with signal "
 312                                        "%d\n", WSTOPSIG(status));
 313                         }
 314                         pid = userspace_pid[0];
 315                         interrupt_end();
 316
 317                         /* Avoid -ERESTARTSYS handling in host */
 318                         if(PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
 319                                 PT_SYSCALL_NR(regs->skas.regs) = -1;
 320                 }
 321         }
 322 }
 323
 324 int copy_context_skas0(unsigned long new_stack, int pid)
 325 {
 326         int err;
 327         unsigned long regs[HOST_FRAME_SIZE];
 328         unsigned long fp_regs[HOST_FP_SIZE];
 329         unsigned long current_stack = current_stub_stack();
 330         struct stub_data *data = (struct stub_data *) current_stack;
 331         struct stub_data *child_data = (struct stub_data *) new_stack;
 332         __u64 new_offset;
 333         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 334
 335         /* prepare offset and fd of child's stack as argument for parent's
 336          * and child's mmap2 calls
 337          */
 338         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
 339                                       .fd       = new_fd,
 340                                       .timer    = ((struct itimerval)
 341                                                     { { 0, 1000000 / hz() },
 342                                                       { 0, 1000000 / hz() }})});
 343         get_safe_registers(regs, fp_regs);
 344
 345         /* Set parent's instruction pointer to start of clone-stub */
 346         regs[REGS_IP_INDEX] = UML_CONFIG_STUB_CODE +
 347                                 (unsigned long) stub_clone_handler -
 348                                 (unsigned long) &__syscall_stub_start;
 349         regs[REGS_SP_INDEX] = UML_CONFIG_STUB_DATA + PAGE_SIZE -
 350                 sizeof(void *);
 351 #ifdef __SIGNAL_FRAMESIZE
 352         regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
 353 #endif
 354         err = ptrace_setregs(pid, regs);
 355         if(err < 0)
 356                 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
 357                       "pid = %d, errno = %d\n", pid, -err);
 358
 359         err = ptrace_setfpregs(pid, fp_regs);
 360         if(err < 0)
 361                 panic("copy_context_skas0 : PTRACE_SETFPREGS failed, "
 362                       "pid = %d, errno = %d\n", pid, -err);
 363
 364         /* set a well known return code for detection of child write failure */
 365         child_data->err = 12345678;
 366
 367         /* Wait, until parent has finished its work: read child's pid from
 368          * parent's stack, and check, if bad result.
 369          */
 370         wait_stub_done(pid, 0, "copy_context_skas0");
 371
 372         pid = data->err;
 373         if(pid < 0)
 374                 panic("copy_context_skas0 - stub-parent reports error %d\n",
 375                       -pid);
 376
 377         /* Wait, until child has finished too: read child's result from
 378          * child's stack and check it.
 379          */
 380         wait_stub_done(pid, -1, "copy_context_skas0");
 381         if (child_data->err != UML_CONFIG_STUB_DATA)
 382                 panic("copy_context_skas0 - stub-child reports error %ld\n",
 383                       child_data->err);
 384
 385         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 386                    (void *)PTRACE_O_TRACESYSGOOD) < 0)
 387                 panic("copy_context_skas0 : PTRACE_OLDSETOPTIONS failed, "
 388                       "errno = %d\n", errno);
 389
 390         return pid;
 391 }
 392
 393 /*
 394  * This is used only, if stub pages are needed, while proc_mm is
 395  * availabl. Opening /proc/mm creates a new mm_context, which lacks
 396  * the stub-pages. Thus, we map them using /proc/mm-fd
 397  */
 398 void map_stub_pages(int fd, unsigned long code,
 399                     unsigned long data, unsigned long stack)
 400 {
 401         struct proc_mm_op mmop;
 402         int n;
 403         __u64 code_offset;
 404         int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
 405                                    &code_offset);
 406
 407         mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
 408                                       .u         =
 409                                       { .mmap    =
 410                                         { .addr    = code,
 411                                           .len     = PAGE_SIZE,
 412                                           .prot    = PROT_EXEC,
 413                                           .flags   = MAP_FIXED | MAP_PRIVATE,
 414                                           .fd      = code_fd,
 415                                           .offset  = code_offset
 416         } } });
 417         n = os_write_file(fd, &mmop, sizeof(mmop));
 418         if(n != sizeof(mmop))
 419                 panic("map_stub_pages : /proc/mm map for code failed, "
 420                       "err = %d\n", -n);
 421
 422         if ( stack ) {
 423                 __u64 map_offset;
 424                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
 425                 mmop = ((struct proc_mm_op)
 426                                 { .op        = MM_MMAP,
 427                                   .u         =
 428                                   { .mmap    =
 429                                     { .addr    = data,
 430                                       .len     = PAGE_SIZE,
 431                                       .prot    = PROT_READ | PROT_WRITE,
 432                                       .flags   = MAP_FIXED | MAP_SHARED,
 433                                       .fd      = map_fd,
 434                                       .offset  = map_offset
 435                 } } });
 436                 n = os_write_file(fd, &mmop, sizeof(mmop));
 437                 if(n != sizeof(mmop))
 438                         panic("map_stub_pages : /proc/mm map for data failed, "
 439                               "err = %d\n", -n);
 440         }
 441 }
 442
 443 void new_thread(void *stack, void **switch_buf_ptr, void **fork_buf_ptr,
 444                 void (*handler)(int))
 445 {
 446         unsigned long flags;
 447         jmp_buf switch_buf, fork_buf;
 448
 449         *switch_buf_ptr = &switch_buf;
 450         *fork_buf_ptr = &fork_buf;
 451
 452         /* Somewhat subtle - siglongjmp restores the signal mask before doing
 453          * the longjmp.  This means that when jumping from one stack to another
 454          * when the target stack has interrupts enabled, an interrupt may occur
 455          * on the source stack.  This is bad when starting up a process because
 456          * it's not supposed to get timer ticks until it has been scheduled.
 457          * So, we disable interrupts around the sigsetjmp to ensure that
 458          * they can't happen until we get back here where they are safe.
 459          */
 460         flags = get_signals();
 461         block_signals();
 462         if(UML_SETJMP(&fork_buf) == 0)
 463                 new_thread_proc(stack, handler);
 464
 465         remove_sigstack();
 466
 467         set_signals(flags);
 468 }
 469
 470 #define INIT_JMP_NEW_THREAD 0
 471 #define INIT_JMP_REMOVE_SIGSTACK 1
 472 #define INIT_JMP_CALLBACK 2
 473 #define INIT_JMP_HALT 3
 474 #define INIT_JMP_REBOOT 4
 475
 476 void thread_wait(void *sw, void *fb)
 477 {
 478         jmp_buf buf, **switch_buf = sw, *fork_buf;
 479
 480         *switch_buf = &buf;
 481         fork_buf = fb;
 482         if(UML_SETJMP(&buf) == 0)
 483                 UML_LONGJMP(fork_buf, INIT_JMP_REMOVE_SIGSTACK);
 484 }
 485
 486 void switch_threads(void *me, void *next)
 487 {
 488         jmp_buf my_buf, **me_ptr = me, *next_buf = next;
 489
 490         *me_ptr = &my_buf;
 491         if(UML_SETJMP(&my_buf) == 0)
 492                 UML_LONGJMP(next_buf, 1);
 493 }
 494
 495 static jmp_buf initial_jmpbuf;
 496
 497 /* XXX Make these percpu */
 498 static void (*cb_proc)(void *arg);
 499 static void *cb_arg;
 500 static jmp_buf *cb_back;
 501
 502 int start_idle_thread(void *stack, void *switch_buf_ptr, void **fork_buf_ptr)
 503 {
 504         jmp_buf **switch_buf = switch_buf_ptr;
 505         int n;
 506
 507         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
 508                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGALRM,
 509                     SIGVTALRM, -1);
 510
 511         *fork_buf_ptr = &initial_jmpbuf;
 512         n = UML_SETJMP(&initial_jmpbuf);
 513         switch(n){
 514         case INIT_JMP_NEW_THREAD:
 515                 new_thread_proc((void *) stack, new_thread_handler);
 516                 break;
 517         case INIT_JMP_REMOVE_SIGSTACK:
 518                 remove_sigstack();
 519                 break;
 520         case INIT_JMP_CALLBACK:
 521                 (*cb_proc)(cb_arg);
 522                 UML_LONGJMP(cb_back, 1);
 523                 break;
 524         case INIT_JMP_HALT:
 525                 kmalloc_ok = 0;
 526                 return(0);
 527         case INIT_JMP_REBOOT:
 528                 kmalloc_ok = 0;
 529                 return(1);
 530         default:
 531                 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
 532         }
 533         UML_LONGJMP(*switch_buf, 1);
 534 }
 535
 536 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 537 {
 538         jmp_buf here;
 539
 540         cb_proc = proc;
 541         cb_arg = arg;
 542         cb_back = &here;
 543
 544         block_signals();
 545         if(UML_SETJMP(&here) == 0)
 546                 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
 547         unblock_signals();
 548
 549         cb_proc = NULL;
 550         cb_arg = NULL;
 551         cb_back = NULL;
 552 }
 553
 554 void halt_skas(void)
 555 {
 556         block_signals();
 557         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
 558 }
 559
 560 void reboot_skas(void)
 561 {
 562         block_signals();
 563         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
 564 }
 565
 566 void switch_mm_skas(struct mm_id *mm_idp)
 567 {
 568         int err;
 569
 570 #warning need cpu pid in switch_mm_skas
 571         if(proc_mm){
 572                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
 573                              mm_idp->u.mm_fd);
 574                 if(err)
 575                         panic("switch_mm_skas - PTRACE_SWITCH_MM failed, "
 576                               "errno = %d\n", errno);
 577         }
 578         else userspace_pid[0] = mm_idp->u.pid;
 579 }