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