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 "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         int err;
 159
 160         ptrace(PTRACE_TRACEME, 0, 0, 0);
 161
 162         init_new_thread_signals();
 163         err = set_interval(1);
 164         if(err)
 165                 panic("userspace_tramp - setting timer failed, errno = %d\n",
 166                       err);
 167
 168         if(!proc_mm){
 169                 /* This has a pte, but it can't be mapped in with the usual
 170                  * tlb_flush mechanism because this is part of that mechanism
 171                  */
 172                 int fd;
 173                 __u64 offset;
 174                 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
 175                 addr = mmap64((void *) UML_CONFIG_STUB_CODE, page_size(),
 176                               PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
 177                 if(addr == MAP_FAILED){
 178                         printk("mapping mmap stub failed, errno = %d\n",
 179                                errno);
 180                         exit(1);
 181                 }
 182
 183                 if(stack != NULL){
 184                         fd = phys_mapping(to_phys(stack), &offset);
 185                         addr = mmap((void *) UML_CONFIG_STUB_DATA, page_size(),
 186                                     PROT_READ | PROT_WRITE,
 187                                     MAP_FIXED | MAP_SHARED, fd, offset);
 188                         if(addr == MAP_FAILED){
 189                                 printk("mapping segfault stack failed, "
 190                                        "errno = %d\n", errno);
 191                                 exit(1);
 192                         }
 193                 }
 194         }
 195         if(!ptrace_faultinfo && (stack != NULL)){
 196                 struct sigaction sa;
 197
 198                 unsigned long v = UML_CONFIG_STUB_CODE +
 199                                   (unsigned long) stub_segv_handler -
 200                                   (unsigned long) &__syscall_stub_start;
 201
 202                 set_sigstack((void *) UML_CONFIG_STUB_DATA, page_size());
 203                 sigemptyset(&sa.sa_mask);
 204                 sigaddset(&sa.sa_mask, SIGIO);
 205                 sigaddset(&sa.sa_mask, SIGWINCH);
 206                 sigaddset(&sa.sa_mask, SIGALRM);
 207                 sigaddset(&sa.sa_mask, SIGVTALRM);
 208                 sigaddset(&sa.sa_mask, SIGUSR1);
 209                 sa.sa_flags = SA_ONSTACK;
 210                 sa.sa_handler = (void *) v;
 211                 sa.sa_restorer = NULL;
 212                 if(sigaction(SIGSEGV, &sa, NULL) < 0)
 213                         panic("userspace_tramp - setting SIGSEGV handler "
 214                               "failed - errno = %d\n", errno);
 215         }
 216
 217         os_stop_process(os_getpid());
 218         return(0);
 219 }
 220
 221 /* Each element set once, and only accessed by a single processor anyway */
 222 #undef NR_CPUS
 223 #define NR_CPUS 1
 224 int userspace_pid[NR_CPUS];
 225
 226 int start_userspace(unsigned long stub_stack)
 227 {
 228         void *stack;
 229         unsigned long sp;
 230         int pid, status, n, flags;
 231
 232         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 233                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 234         if(stack == MAP_FAILED)
 235                 panic("start_userspace : mmap failed, errno = %d", errno);
 236         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
 237
 238         flags = CLONE_FILES | SIGCHLD;
 239         if(proc_mm) flags |= CLONE_VM;
 240         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 241         if(pid < 0)
 242                 panic("start_userspace : clone failed, errno = %d", errno);
 243
 244         do {
 245                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 246                 if(n < 0)
 247                         panic("start_userspace : wait failed, errno = %d",
 248                               errno);
 249         } while(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
 250
 251         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 252                 panic("start_userspace : expected SIGSTOP, got status = %d",
 253                       status);
 254
 255         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, (void *)PTRACE_O_TRACESYSGOOD) < 0)
 256                 panic("start_userspace : PTRACE_OLDSETOPTIONS failed, errno=%d\n",
 257                       errno);
 258
 259         if(munmap(stack, PAGE_SIZE) < 0)
 260                 panic("start_userspace : munmap failed, errno = %d\n", errno);
 261
 262         return(pid);
 263 }
 264
 265 void userspace(union uml_pt_regs *regs)
 266 {
 267         int err, status, op, pid = userspace_pid[0];
 268         int local_using_sysemu; /*To prevent races if using_sysemu changes under us.*/
 269
 270         while(1){
 271                 restore_registers(pid, regs);
 272
 273                 /* Now we set local_using_sysemu to be used for one loop */
 274                 local_using_sysemu = get_using_sysemu();
 275
 276                 op = SELECT_PTRACE_OPERATION(local_using_sysemu, singlestepping(NULL));
 277
 278                 err = ptrace(op, pid, 0, 0);
 279                 if(err)
 280                         panic("userspace - could not resume userspace process, "
 281                               "pid=%d, ptrace operation = %d, errno = %d\n",
 282                               pid, op, errno);
 283
 284                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 285                 if(err < 0)
 286                         panic("userspace - waitpid failed, errno = %d\n",
 287                               errno);
 288
 289                 regs->skas.is_user = 1;
 290                 save_registers(pid, regs);
 291                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
 292
 293                 if(WIFSTOPPED(status)){
 294                         switch(WSTOPSIG(status)){
 295                         case SIGSEGV:
 296                                 if(PTRACE_FULL_FAULTINFO || !ptrace_faultinfo)
 297                                         user_signal(SIGSEGV, regs, pid);
 298                                 else handle_segv(pid, regs);
 299                                 break;
 300                         case SIGTRAP + 0x80:
 301                                 handle_trap(pid, regs, local_using_sysemu);
 302                                 break;
 303                         case SIGTRAP:
 304                                 relay_signal(SIGTRAP, regs);
 305                                 break;
 306                         case SIGIO:
 307                         case SIGVTALRM:
 308                         case SIGILL:
 309                         case SIGBUS:
 310                         case SIGFPE:
 311                         case SIGWINCH:
 312                                 user_signal(WSTOPSIG(status), regs, pid);
 313                                 break;
 314                         default:
 315                                 printk("userspace - child stopped with signal "
 316                                        "%d\n", WSTOPSIG(status));
 317                         }
 318                         pid = userspace_pid[0];
 319                         interrupt_end();
 320
 321                         /* Avoid -ERESTARTSYS handling in host */
 322                         if(PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
 323                                 PT_SYSCALL_NR(regs->skas.regs) = -1;
 324                 }
 325         }
 326 }
 327
 328 int copy_context_skas0(unsigned long new_stack, int pid)
 329 {
 330         int err;
 331         unsigned long regs[HOST_FRAME_SIZE];
 332         unsigned long fp_regs[HOST_FP_SIZE];
 333         unsigned long current_stack = current_stub_stack();
 334         struct stub_data *data = (struct stub_data *) current_stack;
 335         struct stub_data *child_data = (struct stub_data *) new_stack;
 336         __u64 new_offset;
 337         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 338
 339         /* prepare offset and fd of child's stack as argument for parent's
 340          * and child's mmap2 calls
 341          */
 342         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
 343                                       .fd       = new_fd,
 344                                       .timer    = ((struct itimerval)
 345                                                     { { 0, 1000000 / hz() },
 346                                                       { 0, 1000000 / hz() }})});
 347         get_safe_registers(regs, fp_regs);
 348
 349         /* Set parent's instruction pointer to start of clone-stub */
 350         regs[REGS_IP_INDEX] = UML_CONFIG_STUB_CODE +
 351                                 (unsigned long) stub_clone_handler -
 352                                 (unsigned long) &__syscall_stub_start;
 353         regs[REGS_SP_INDEX] = UML_CONFIG_STUB_DATA + PAGE_SIZE -
 354                 sizeof(void *);
 355 #ifdef __SIGNAL_FRAMESIZE
 356         regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
 357 #endif
 358         err = ptrace_setregs(pid, regs);
 359         if(err < 0)
 360                 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
 361                       "pid = %d, errno = %d\n", pid, -err);
 362
 363         err = ptrace_setfpregs(pid, fp_regs);
 364         if(err < 0)
 365                 panic("copy_context_skas0 : PTRACE_SETFPREGS failed, "
 366                       "pid = %d, errno = %d\n", pid, -err);
 367
 368         /* set a well known return code for detection of child write failure */
 369         child_data->err = 12345678;
 370
 371         /* Wait, until parent has finished its work: read child's pid from
 372          * parent's stack, and check, if bad result.
 373          */
 374         wait_stub_done(pid, 0, "copy_context_skas0");
 375
 376         pid = data->err;
 377         if(pid < 0)
 378                 panic("copy_context_skas0 - stub-parent reports error %d\n",
 379                       -pid);
 380
 381         /* Wait, until child has finished too: read child's result from
 382          * child's stack and check it.
 383          */
 384         wait_stub_done(pid, -1, "copy_context_skas0");
 385         if (child_data->err != UML_CONFIG_STUB_DATA)
 386                 panic("copy_context_skas0 - stub-child reports error %ld\n",
 387                       child_data->err);
 388
 389         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 390                    (void *)PTRACE_O_TRACESYSGOOD) < 0)
 391                 panic("copy_context_skas0 : PTRACE_OLDSETOPTIONS failed, "
 392                       "errno = %d\n", errno);
 393
 394         return pid;
 395 }
 396
 397 /*
 398  * This is used only, if stub pages are needed, while proc_mm is
 399  * availabl. Opening /proc/mm creates a new mm_context, which lacks
 400  * the stub-pages. Thus, we map them using /proc/mm-fd
 401  */
 402 void map_stub_pages(int fd, unsigned long code,
 403                     unsigned long data, unsigned long stack)
 404 {
 405         struct proc_mm_op mmop;
 406         int n;
 407         __u64 code_offset;
 408         int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
 409                                    &code_offset);
 410
 411         mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
 412                                       .u         =
 413                                       { .mmap    =
 414                                         { .addr    = code,
 415                                           .len     = PAGE_SIZE,
 416                                           .prot    = PROT_EXEC,
 417                                           .flags   = MAP_FIXED | MAP_PRIVATE,
 418                                           .fd      = code_fd,
 419                                           .offset  = code_offset
 420         } } });
 421         n = os_write_file(fd, &mmop, sizeof(mmop));
 422         if(n != sizeof(mmop)){
 423                 printk("mmap args - addr = 0x%lx, fd = %d, offset = %llx\n",
 424                        code, code_fd, (unsigned long long) code_offset);
 425                 panic("map_stub_pages : /proc/mm map for code failed, "
 426                       "err = %d\n", -n);
 427         }
 428
 429         if ( stack ) {
 430                 __u64 map_offset;
 431                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
 432                 mmop = ((struct proc_mm_op)
 433                                 { .op        = MM_MMAP,
 434                                   .u         =
 435                                   { .mmap    =
 436                                     { .addr    = data,
 437                                       .len     = PAGE_SIZE,
 438                                       .prot    = PROT_READ | PROT_WRITE,
 439                                       .flags   = MAP_FIXED | MAP_SHARED,
 440                                       .fd      = map_fd,
 441                                       .offset  = map_offset
 442                 } } });
 443                 n = os_write_file(fd, &mmop, sizeof(mmop));
 444                 if(n != sizeof(mmop))
 445                         panic("map_stub_pages : /proc/mm map for data failed, "
 446                               "err = %d\n", -n);
 447         }
 448 }
 449
 450 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
 451 {
 452         (*buf)[0].JB_IP = (unsigned long) handler;
 453         (*buf)[0].JB_SP = (unsigned long) stack +
 454                 (PAGE_SIZE << UML_CONFIG_KERNEL_STACK_ORDER) - sizeof(void *);
 455 }
 456
 457 #define INIT_JMP_NEW_THREAD 0
 458 #define INIT_JMP_CALLBACK 1
 459 #define INIT_JMP_HALT 2
 460 #define INIT_JMP_REBOOT 3
 461
 462 void switch_threads(jmp_buf *me, jmp_buf *you)
 463 {
 464         if(UML_SETJMP(me) == 0)
 465                 UML_LONGJMP(you, 1);
 466 }
 467
 468 static jmp_buf initial_jmpbuf;
 469
 470 /* XXX Make these percpu */
 471 static void (*cb_proc)(void *arg);
 472 static void *cb_arg;
 473 static jmp_buf *cb_back;
 474
 475 int start_idle_thread(void *stack, jmp_buf *switch_buf)
 476 {
 477         int n;
 478
 479         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
 480                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGALRM,
 481                     SIGVTALRM, -1);
 482
 483         n = UML_SETJMP(&initial_jmpbuf);
 484         switch(n){
 485         case INIT_JMP_NEW_THREAD:
 486                 (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
 487                 (*switch_buf)[0].JB_SP = (unsigned long) stack +
 488                         (PAGE_SIZE << UML_CONFIG_KERNEL_STACK_ORDER) -
 489                         sizeof(void *);
 490                 break;
 491         case INIT_JMP_CALLBACK:
 492                 (*cb_proc)(cb_arg);
 493                 UML_LONGJMP(cb_back, 1);
 494                 break;
 495         case INIT_JMP_HALT:
 496                 kmalloc_ok = 0;
 497                 return(0);
 498         case INIT_JMP_REBOOT:
 499                 kmalloc_ok = 0;
 500                 return(1);
 501         default:
 502                 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
 503         }
 504         UML_LONGJMP(switch_buf, 1);
 505 }
 506
 507 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 508 {
 509         jmp_buf here;
 510
 511         cb_proc = proc;
 512         cb_arg = arg;
 513         cb_back = &here;
 514
 515         block_signals();
 516         if(UML_SETJMP(&here) == 0)
 517                 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
 518         unblock_signals();
 519
 520         cb_proc = NULL;
 521         cb_arg = NULL;
 522         cb_back = NULL;
 523 }
 524
 525 void halt_skas(void)
 526 {
 527         block_signals();
 528         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
 529 }
 530
 531 void reboot_skas(void)
 532 {
 533         block_signals();
 534         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
 535 }
 536
 537 void switch_mm_skas(struct mm_id *mm_idp)
 538 {
 539         int err;
 540
 541 #warning need cpu pid in switch_mm_skas
 542         if(proc_mm){
 543                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
 544                              mm_idp->u.mm_fd);
 545                 if(err)
 546                         panic("switch_mm_skas - PTRACE_SWITCH_MM failed, "
 547                               "errno = %d\n", errno);
 548         }
 549         else userspace_pid[0] = mm_idp->u.pid;
 550 }