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