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 <sys/syscall.h>
  18 #include <asm/types.h>
  19 #include "user.h"
  20 #include "sysdep/ptrace.h"
  21 #include "kern_util.h"
  22 #include "skas.h"
  23 #include "stub-data.h"
  24 #include "mm_id.h"
  25 #include "sysdep/sigcontext.h"
  26 #include "sysdep/stub.h"
  27 #include "os.h"
  28 #include "proc_mm.h"
  29 #include "skas_ptrace.h"
  30 #include "chan_user.h"
  31 #include "registers.h"
  32 #include "mem.h"
  33 #include "uml-config.h"
  34 #include "process.h"
  35 #include "longjmp.h"
  36 #include "kern_constants.h"
  37 #include "as-layout.h"
  38
  39 int is_skas_winch(int pid, int fd, void *data)
  40 {
  41         if(pid != os_getpgrp())
  42                 return(0);
  43
  44         register_winch_irq(-1, fd, -1, data);
  45         return(1);
  46 }
  47
  48 static int ptrace_dump_regs(int pid)
  49 {
  50         unsigned long regs[MAX_REG_NR];
  51         int i;
  52
  53         if(ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
  54                 return -errno;
  55         else {
  56                 printk("Stub registers -\n");
  57                 for(i = 0; i < ARRAY_SIZE(regs); i++)
  58                         printk("\t%d - %lx\n", i, regs[i]);
  59         }
  60
  61         return 0;
  62 }
  63
  64 /*
  65  * Signals that are OK to receive in the stub - we'll just continue it.
  66  * SIGWINCH will happen when UML is inside a detached screen.
  67  */
  68 #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
  69
  70 /* Signals that the stub will finish with - anything else is an error */
  71 #define STUB_DONE_MASK ((1 << SIGUSR1) | (1 << SIGTRAP))
  72
  73 void wait_stub_done(int pid)
  74 {
  75         int n, status, err;
  76
  77         while(1){
  78                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
  79                 if((n < 0) || !WIFSTOPPED(status))
  80                         goto bad_wait;
  81
  82                 if(((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
  83                         break;
  84
  85                 err = ptrace(PTRACE_CONT, pid, 0, 0);
  86                 if(err)
  87                         panic("wait_stub_done : continue failed, errno = %d\n",
  88                               errno);
  89         }
  90
  91         if(((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
  92                 return;
  93
  94 bad_wait:
  95         err = ptrace_dump_regs(pid);
  96         if(err)
  97                 printk("Failed to get registers from stub, errno = %d\n", -err);
  98         panic("wait_stub_done : failed to wait for SIGUSR1/SIGTRAP, pid = %d, "
  99               "n = %d, errno = %d, status = 0x%x\n", pid, n, errno, status);
 100 }
 101
 102 extern unsigned long current_stub_stack(void);
 103
 104 void get_skas_faultinfo(int pid, struct faultinfo * fi)
 105 {
 106         int err;
 107
 108         if(ptrace_faultinfo){
 109                 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
 110                 if(err)
 111                         panic("get_skas_faultinfo - PTRACE_FAULTINFO failed, "
 112                               "errno = %d\n", errno);
 113
 114                 /* Special handling for i386, which has different structs */
 115                 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
 116                         memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
 117                                sizeof(struct faultinfo) -
 118                                sizeof(struct ptrace_faultinfo));
 119         }
 120         else {
 121                 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
 122                 if(err)
 123                         panic("Failed to continue stub, pid = %d, errno = %d\n",
 124                               pid, errno);
 125                 wait_stub_done(pid);
 126
 127                 /* faultinfo is prepared by the stub-segv-handler at start of
 128                  * the stub stack page. We just have to copy it.
 129                  */
 130                 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
 131         }
 132 }
 133
 134 static void handle_segv(int pid, union uml_pt_regs * regs)
 135 {
 136         get_skas_faultinfo(pid, &regs->skas.faultinfo);
 137         segv(regs->skas.faultinfo, 0, 1, NULL);
 138 }
 139
 140 /*To use the same value of using_sysemu as the caller, ask it that value (in local_using_sysemu)*/
 141 static void handle_trap(int pid, union uml_pt_regs *regs, int local_using_sysemu)
 142 {
 143         int err, status;
 144
 145         /* Mark this as a syscall */
 146         UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->skas.regs);
 147
 148         if (!local_using_sysemu)
 149         {
 150                 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 151                              __NR_getpid);
 152                 if(err < 0)
 153                         panic("handle_trap - nullifying syscall failed errno = %d\n",
 154                               errno);
 155
 156                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
 157                 if(err < 0)
 158                         panic("handle_trap - continuing to end of syscall failed, "
 159                               "errno = %d\n", errno);
 160
 161                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 162                 if((err < 0) || !WIFSTOPPED(status) ||
 163                    (WSTOPSIG(status) != SIGTRAP + 0x80)){
 164                         err = ptrace_dump_regs(pid);
 165                         if(err)
 166                                 printk("Failed to get registers from process, "
 167                                        "errno = %d\n", -err);
 168                         panic("handle_trap - failed to wait at end of syscall, "
 169                               "errno = %d, status = %d\n", errno, status);
 170                 }
 171         }
 172
 173         handle_syscall(regs);
 174 }
 175
 176 extern int __syscall_stub_start;
 177
 178 static int userspace_tramp(void *stack)
 179 {
 180         void *addr;
 181         int err;
 182
 183         ptrace(PTRACE_TRACEME, 0, 0, 0);
 184
 185         init_new_thread_signals();
 186         err = set_interval(1);
 187         if(err)
 188                 panic("userspace_tramp - setting timer failed, errno = %d\n",
 189                       err);
 190
 191         if(!proc_mm){
 192                 /* This has a pte, but it can't be mapped in with the usual
 193                  * tlb_flush mechanism because this is part of that mechanism
 194                  */
 195                 int fd;
 196                 __u64 offset;
 197                 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
 198                 addr = mmap64((void *) UML_CONFIG_STUB_CODE, UM_KERN_PAGE_SIZE,
 199                               PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
 200                 if(addr == MAP_FAILED){
 201                         printk("mapping mmap stub failed, errno = %d\n",
 202                                errno);
 203                         exit(1);
 204                 }
 205
 206                 if(stack != NULL){
 207                         fd = phys_mapping(to_phys(stack), &offset);
 208                         addr = mmap((void *) UML_CONFIG_STUB_DATA,
 209                                     UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
 210                                     MAP_FIXED | MAP_SHARED, fd, offset);
 211                         if(addr == MAP_FAILED){
 212                                 printk("mapping segfault stack failed, "
 213                                        "errno = %d\n", errno);
 214                                 exit(1);
 215                         }
 216                 }
 217         }
 218         if(!ptrace_faultinfo && (stack != NULL)){
 219                 struct sigaction sa;
 220
 221                 unsigned long v = UML_CONFIG_STUB_CODE +
 222                                   (unsigned long) stub_segv_handler -
 223                                   (unsigned long) &__syscall_stub_start;
 224
 225                 set_sigstack((void *) UML_CONFIG_STUB_DATA, UM_KERN_PAGE_SIZE);
 226                 sigemptyset(&sa.sa_mask);
 227                 sigaddset(&sa.sa_mask, SIGIO);
 228                 sigaddset(&sa.sa_mask, SIGWINCH);
 229                 sigaddset(&sa.sa_mask, SIGALRM);
 230                 sigaddset(&sa.sa_mask, SIGVTALRM);
 231                 sigaddset(&sa.sa_mask, SIGUSR1);
 232                 sa.sa_flags = SA_ONSTACK;
 233                 sa.sa_handler = (void *) v;
 234                 sa.sa_restorer = NULL;
 235                 if(sigaction(SIGSEGV, &sa, NULL) < 0)
 236                         panic("userspace_tramp - setting SIGSEGV handler "
 237                               "failed - errno = %d\n", errno);
 238         }
 239
 240         os_stop_process(os_getpid());
 241         return(0);
 242 }
 243
 244 /* Each element set once, and only accessed by a single processor anyway */
 245 #undef NR_CPUS
 246 #define NR_CPUS 1
 247 int userspace_pid[NR_CPUS];
 248
 249 int start_userspace(unsigned long stub_stack)
 250 {
 251         void *stack;
 252         unsigned long sp;
 253         int pid, status, n, flags;
 254
 255         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 256                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 257         if(stack == MAP_FAILED)
 258                 panic("start_userspace : mmap failed, errno = %d", errno);
 259         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
 260
 261         flags = CLONE_FILES | SIGCHLD;
 262         if(proc_mm) flags |= CLONE_VM;
 263         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 264         if(pid < 0)
 265                 panic("start_userspace : clone failed, errno = %d", errno);
 266
 267         do {
 268                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 269                 if(n < 0)
 270                         panic("start_userspace : wait failed, errno = %d",
 271                               errno);
 272         } while(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
 273
 274         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 275                 panic("start_userspace : expected SIGSTOP, got status = %d",
 276                       status);
 277
 278         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, (void *)PTRACE_O_TRACESYSGOOD) < 0)
 279                 panic("start_userspace : PTRACE_OLDSETOPTIONS failed, errno=%d\n",
 280                       errno);
 281
 282         if(munmap(stack, PAGE_SIZE) < 0)
 283                 panic("start_userspace : munmap failed, errno = %d\n", errno);
 284
 285         return(pid);
 286 }
 287
 288 void userspace(union uml_pt_regs *regs)
 289 {
 290         int err, status, op, pid = userspace_pid[0];
 291         /* To prevent races if using_sysemu changes under us.*/
 292         int local_using_sysemu;
 293
 294         while(1){
 295                 restore_registers(pid, regs);
 296
 297                 /* Now we set local_using_sysemu to be used for one loop */
 298                 local_using_sysemu = get_using_sysemu();
 299
 300                 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
 301                                              singlestepping(NULL));
 302
 303                 err = ptrace(op, pid, 0, 0);
 304                 if(err)
 305                         panic("userspace - could not resume userspace process, "
 306                               "pid=%d, ptrace operation = %d, errno = %d\n",
 307                               pid, op, errno);
 308
 309                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 310                 if(err < 0)
 311                         panic("userspace - waitpid failed, errno = %d\n",
 312                               errno);
 313
 314                 regs->skas.is_user = 1;
 315                 save_registers(pid, regs);
 316                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
 317
 318                 if(WIFSTOPPED(status)){
 319                         int sig = WSTOPSIG(status);
 320                         switch(sig){
 321                         case SIGSEGV:
 322                                 if(PTRACE_FULL_FAULTINFO || !ptrace_faultinfo){
 323                                         get_skas_faultinfo(pid, &regs->skas.faultinfo);
 324                                         (*sig_info[SIGSEGV])(SIGSEGV, regs);
 325                                 }
 326                                 else handle_segv(pid, regs);
 327                                 break;
 328                         case SIGTRAP + 0x80:
 329                                 handle_trap(pid, regs, local_using_sysemu);
 330                                 break;
 331                         case SIGTRAP:
 332                                 relay_signal(SIGTRAP, regs);
 333                                 break;
 334                         case SIGIO:
 335                         case SIGVTALRM:
 336                         case SIGILL:
 337                         case SIGBUS:
 338                         case SIGFPE:
 339                         case SIGWINCH:
 340                                 block_signals();
 341                                 (*sig_info[sig])(sig, regs);
 342                                 unblock_signals();
 343                                 break;
 344                         default:
 345                                 printk("userspace - child stopped with signal "
 346                                        "%d\n", sig);
 347                         }
 348                         pid = userspace_pid[0];
 349                         interrupt_end();
 350
 351                         /* Avoid -ERESTARTSYS handling in host */
 352                         if(PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
 353                                 PT_SYSCALL_NR(regs->skas.regs) = -1;
 354                 }
 355         }
 356 }
 357
 358 static unsigned long thread_regs[MAX_REG_NR];
 359 static unsigned long thread_fp_regs[HOST_FP_SIZE];
 360
 361 static int __init init_thread_regs(void)
 362 {
 363         get_safe_registers(thread_regs, thread_fp_regs);
 364         /* Set parent's instruction pointer to start of clone-stub */
 365         thread_regs[REGS_IP_INDEX] = UML_CONFIG_STUB_CODE +
 366                                 (unsigned long) stub_clone_handler -
 367                                 (unsigned long) &__syscall_stub_start;
 368         thread_regs[REGS_SP_INDEX] = UML_CONFIG_STUB_DATA + PAGE_SIZE -
 369                 sizeof(void *);
 370 #ifdef __SIGNAL_FRAMESIZE
 371         thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
 372 #endif
 373         return 0;
 374 }
 375
 376 __initcall(init_thread_regs);
 377
 378 int copy_context_skas0(unsigned long new_stack, int pid)
 379 {
 380         int err;
 381         unsigned long current_stack = current_stub_stack();
 382         struct stub_data *data = (struct stub_data *) current_stack;
 383         struct stub_data *child_data = (struct stub_data *) new_stack;
 384         __u64 new_offset;
 385         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 386
 387         /* prepare offset and fd of child's stack as argument for parent's
 388          * and child's mmap2 calls
 389          */
 390         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
 391                                       .fd       = new_fd,
 392                                       .timer    = ((struct itimerval)
 393                                                     { { 0, 1000000 / hz() },
 394                                                       { 0, 1000000 / hz() }})});
 395         err = ptrace_setregs(pid, thread_regs);
 396         if(err < 0)
 397                 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
 398                       "pid = %d, errno = %d\n", pid, -err);
 399
 400         err = ptrace_setfpregs(pid, thread_fp_regs);
 401         if(err < 0)
 402                 panic("copy_context_skas0 : PTRACE_SETFPREGS failed, "
 403                       "pid = %d, errno = %d\n", pid, -err);
 404
 405         /* set a well known return code for detection of child write failure */
 406         child_data->err = 12345678;
 407
 408         /* Wait, until parent has finished its work: read child's pid from
 409          * parent's stack, and check, if bad result.
 410          */
 411         err = ptrace(PTRACE_CONT, pid, 0, 0);
 412         if(err)
 413                 panic("Failed to continue new process, pid = %d, "
 414                       "errno = %d\n", pid, errno);
 415         wait_stub_done(pid);
 416
 417         pid = data->err;
 418         if(pid < 0)
 419                 panic("copy_context_skas0 - stub-parent reports error %d\n",
 420                       -pid);
 421
 422         /* Wait, until child has finished too: read child's result from
 423          * child's stack and check it.
 424          */
 425         wait_stub_done(pid);
 426         if (child_data->err != UML_CONFIG_STUB_DATA)
 427                 panic("copy_context_skas0 - stub-child reports error %ld\n",
 428                       child_data->err);
 429
 430         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 431                    (void *)PTRACE_O_TRACESYSGOOD) < 0)
 432                 panic("copy_context_skas0 : PTRACE_OLDSETOPTIONS failed, "
 433                       "errno = %d\n", errno);
 434
 435         return pid;
 436 }
 437
 438 /*
 439  * This is used only, if stub pages are needed, while proc_mm is
 440  * available. Opening /proc/mm creates a new mm_context, which lacks
 441  * the stub-pages. Thus, we map them using /proc/mm-fd
 442  */
 443 void map_stub_pages(int fd, unsigned long code,
 444                     unsigned long data, unsigned long stack)
 445 {
 446         struct proc_mm_op mmop;
 447         int n;
 448         __u64 code_offset;
 449         int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
 450                                    &code_offset);
 451
 452         mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
 453                                       .u         =
 454                                       { .mmap    =
 455                                         { .addr    = code,
 456                                           .len     = PAGE_SIZE,
 457                                           .prot    = PROT_EXEC,
 458                                           .flags   = MAP_FIXED | MAP_PRIVATE,
 459                                           .fd      = code_fd,
 460                                           .offset  = code_offset
 461         } } });
 462         CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
 463         if(n != sizeof(mmop)){
 464                 n = errno;
 465                 printk("mmap args - addr = 0x%lx, fd = %d, offset = %llx\n",
 466                        code, code_fd, (unsigned long long) code_offset);
 467                 panic("map_stub_pages : /proc/mm map for code failed, "
 468                       "err = %d\n", n);
 469         }
 470
 471         if ( stack ) {
 472                 __u64 map_offset;
 473                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
 474                 mmop = ((struct proc_mm_op)
 475                                 { .op        = MM_MMAP,
 476                                   .u         =
 477                                   { .mmap    =
 478                                     { .addr    = data,
 479                                       .len     = PAGE_SIZE,
 480                                       .prot    = PROT_READ | PROT_WRITE,
 481                                       .flags   = MAP_FIXED | MAP_SHARED,
 482                                       .fd      = map_fd,
 483                                       .offset  = map_offset
 484                 } } });
 485                 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
 486                 if(n != sizeof(mmop))
 487                         panic("map_stub_pages : /proc/mm map for data failed, "
 488                               "err = %d\n", errno);
 489         }
 490 }
 491
 492 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
 493 {
 494         (*buf)[0].JB_IP = (unsigned long) handler;
 495         (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
 496                 sizeof(void *);
 497 }
 498
 499 #define INIT_JMP_NEW_THREAD 0
 500 #define INIT_JMP_CALLBACK 1
 501 #define INIT_JMP_HALT 2
 502 #define INIT_JMP_REBOOT 3
 503
 504 void switch_threads(jmp_buf *me, jmp_buf *you)
 505 {
 506         if(UML_SETJMP(me) == 0)
 507                 UML_LONGJMP(you, 1);
 508 }
 509
 510 static jmp_buf initial_jmpbuf;
 511
 512 /* XXX Make these percpu */
 513 static void (*cb_proc)(void *arg);
 514 static void *cb_arg;
 515 static jmp_buf *cb_back;
 516
 517 int start_idle_thread(void *stack, jmp_buf *switch_buf)
 518 {
 519         int n;
 520
 521         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
 522                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGALRM,
 523                     SIGVTALRM, -1);
 524
 525         /*
 526          * Can't use UML_SETJMP or UML_LONGJMP here because they save
 527          * and restore signals, with the possible side-effect of
 528          * trying to handle any signals which came when they were
 529          * blocked, which can't be done on this stack.
 530          * Signals must be blocked when jumping back here and restored
 531          * after returning to the jumper.
 532          */
 533         n = setjmp(initial_jmpbuf);
 534         switch(n){
 535         case INIT_JMP_NEW_THREAD:
 536                 (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
 537                 (*switch_buf)[0].JB_SP = (unsigned long) stack +
 538                         UM_THREAD_SIZE - sizeof(void *);
 539                 break;
 540         case INIT_JMP_CALLBACK:
 541                 (*cb_proc)(cb_arg);
 542                 longjmp(*cb_back, 1);
 543                 break;
 544         case INIT_JMP_HALT:
 545                 kmalloc_ok = 0;
 546                 return(0);
 547         case INIT_JMP_REBOOT:
 548                 kmalloc_ok = 0;
 549                 return(1);
 550         default:
 551                 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
 552         }
 553         longjmp(*switch_buf, 1);
 554 }
 555
 556 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 557 {
 558         jmp_buf here;
 559
 560         cb_proc = proc;
 561         cb_arg = arg;
 562         cb_back = &here;
 563
 564         block_signals();
 565         if(UML_SETJMP(&here) == 0)
 566                 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
 567         unblock_signals();
 568
 569         cb_proc = NULL;
 570         cb_arg = NULL;
 571         cb_back = NULL;
 572 }
 573
 574 void halt_skas(void)
 575 {
 576         block_signals();
 577         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
 578 }
 579
 580 void reboot_skas(void)
 581 {
 582         block_signals();
 583         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
 584 }
 585
 586 void switch_mm_skas(struct mm_id *mm_idp)
 587 {
 588         int err;
 589
 590         /* FIXME: need cpu pid in switch_mm_skas */
 591         if(proc_mm){
 592                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
 593                              mm_idp->u.mm_fd);
 594                 if(err)
 595                         panic("switch_mm_skas - PTRACE_SWITCH_MM failed, "
 596                               "errno = %d\n", errno);
 597         }
 598         else userspace_pid[0] = mm_idp->u.pid;
 599 }