arch/um/os-Linux/skas/process.c

   1 /*
   2  * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
   3  * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   4  * Licensed under the GPL
   5  */
   6
   7 #include <stdlib.h>
   8 #include <unistd.h>
   9 #include <sched.h>
  10 #include <errno.h>
  11 #include <string.h>
  12 #include <sys/mman.h>
  13 #include <sys/wait.h>
  14 #include <asm/unistd.h>
  15 #include <as-layout.h>
  16 #include <init.h>
  17 #include <kern_util.h>
  18 #include <mem.h>
  19 #include <os.h>
  20 #include <ptrace_user.h>
  21 #include <registers.h>
  22 #include <skas.h>
  23 #include <sysdep/stub.h>
  24
  25 int is_skas_winch(int pid, int fd, void *data)
  26 {
  27         return pid == getpgrp();
  28 }
  29
  30 static int ptrace_dump_regs(int pid)
  31 {
  32         unsigned long regs[MAX_REG_NR];
  33         int i;
  34
  35         if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
  36                 return -errno;
  37
  38         printk(UM_KERN_ERR "Stub registers -\n");
  39         for (i = 0; i < ARRAY_SIZE(regs); i++)
  40                 printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
  41
  42         return 0;
  43 }
  44
  45 /*
  46  * Signals that are OK to receive in the stub - we'll just continue it.
  47  * SIGWINCH will happen when UML is inside a detached screen.
  48  */
  49 #define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
  50
  51 /* Signals that the stub will finish with - anything else is an error */
  52 #define STUB_DONE_MASK (1 << SIGTRAP)
  53
  54 void wait_stub_done(int pid)
  55 {
  56         int n, status, err;
  57
  58         while (1) {
  59                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
  60                 if ((n < 0) || !WIFSTOPPED(status))
  61                         goto bad_wait;
  62
  63                 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
  64                         break;
  65
  66                 err = ptrace(PTRACE_CONT, pid, 0, 0);
  67                 if (err) {
  68                         printk(UM_KERN_ERR "wait_stub_done : continue failed, "
  69                                "errno = %d\n", errno);
  70                         fatal_sigsegv();
  71                 }
  72         }
  73
  74         if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
  75                 return;
  76
  77 bad_wait:
  78         err = ptrace_dump_regs(pid);
  79         if (err)
  80                 printk(UM_KERN_ERR "Failed to get registers from stub, "
  81                        "errno = %d\n", -err);
  82         printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
  83                "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
  84                status);
  85         fatal_sigsegv();
  86 }
  87
  88 extern unsigned long current_stub_stack(void);
  89
  90 static void get_skas_faultinfo(int pid, struct faultinfo *fi, unsigned long *aux_fp_regs)
  91 {
  92         int err;
  93
  94         err = get_fp_registers(pid, aux_fp_regs);
  95         if (err < 0) {
  96                 printk(UM_KERN_ERR "save_fp_registers returned %d\n",
  97                        err);
  98                 fatal_sigsegv();
  99         }
 100         err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
 101         if (err) {
 102                 printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
 103                        "errno = %d\n", pid, errno);
 104                 fatal_sigsegv();
 105         }
 106         wait_stub_done(pid);
 107
 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         err = put_fp_registers(pid, aux_fp_regs);
 115         if (err < 0) {
 116                 printk(UM_KERN_ERR "put_fp_registers returned %d\n",
 117                        err);
 118                 fatal_sigsegv();
 119         }
 120 }
 121
 122 static void handle_segv(int pid, struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
 123 {
 124         get_skas_faultinfo(pid, &regs->faultinfo, aux_fp_regs);
 125         segv(regs->faultinfo, 0, 1, NULL);
 126 }
 127
 128 /*
 129  * To use the same value of using_sysemu as the caller, ask it that value
 130  * (in local_using_sysemu
 131  */
 132 static void handle_trap(int pid, struct uml_pt_regs *regs,
 133                         int local_using_sysemu)
 134 {
 135         int err, status;
 136
 137         if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
 138                 fatal_sigsegv();
 139
 140         if (!local_using_sysemu)
 141         {
 142                 err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
 143                              __NR_getpid);
 144                 if (err < 0) {
 145                         printk(UM_KERN_ERR "handle_trap - nullifying syscall "
 146                                "failed, errno = %d\n", errno);
 147                         fatal_sigsegv();
 148                 }
 149
 150                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
 151                 if (err < 0) {
 152                         printk(UM_KERN_ERR "handle_trap - continuing to end of "
 153                                "syscall failed, errno = %d\n", errno);
 154                         fatal_sigsegv();
 155                 }
 156
 157                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
 158                 if ((err < 0) || !WIFSTOPPED(status) ||
 159                     (WSTOPSIG(status) != SIGTRAP + 0x80)) {
 160                         err = ptrace_dump_regs(pid);
 161                         if (err)
 162                                 printk(UM_KERN_ERR "Failed to get registers "
 163                                        "from process, errno = %d\n", -err);
 164                         printk(UM_KERN_ERR "handle_trap - failed to wait at "
 165                                "end of syscall, errno = %d, status = %d\n",
 166                                errno, status);
 167                         fatal_sigsegv();
 168                 }
 169         }
 170
 171         handle_syscall(regs);
 172 }
 173
 174 extern char __syscall_stub_start[];
 175
 176 static int userspace_tramp(void *stack)
 177 {
 178         void *addr;
 179         int fd;
 180         unsigned long long offset;
 181
 182         ptrace(PTRACE_TRACEME, 0, 0, 0);
 183
 184         signal(SIGTERM, SIG_DFL);
 185         signal(SIGWINCH, SIG_IGN);
 186
 187         /*
 188          * This has a pte, but it can't be mapped in with the usual
 189          * tlb_flush mechanism because this is part of that mechanism
 190          */
 191         fd = phys_mapping(to_phys(__syscall_stub_start), &offset);
 192         addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
 193                       PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
 194         if (addr == MAP_FAILED) {
 195                 printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
 196                        "errno = %d\n", STUB_CODE, errno);
 197                 exit(1);
 198         }
 199
 200         if (stack != NULL) {
 201                 fd = phys_mapping(to_phys(stack), &offset);
 202                 addr = mmap((void *) STUB_DATA,
 203                             UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
 204                             MAP_FIXED | MAP_SHARED, fd, offset);
 205                 if (addr == MAP_FAILED) {
 206                         printk(UM_KERN_ERR "mapping segfault stack "
 207                                "at 0x%lx failed, errno = %d\n",
 208                                STUB_DATA, errno);
 209                         exit(1);
 210                 }
 211         }
 212         if (stack != NULL) {
 213                 struct sigaction sa;
 214
 215                 unsigned long v = STUB_CODE +
 216                                   (unsigned long) stub_segv_handler -
 217                                   (unsigned long) __syscall_stub_start;
 218
 219                 set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
 220                 sigemptyset(&sa.sa_mask);
 221                 sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
 222                 sa.sa_sigaction = (void *) v;
 223                 sa.sa_restorer = NULL;
 224                 if (sigaction(SIGSEGV, &sa, NULL) < 0) {
 225                         printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
 226                                "handler failed - errno = %d\n", errno);
 227                         exit(1);
 228                 }
 229         }
 230
 231         kill(os_getpid(), SIGSTOP);
 232         return 0;
 233 }
 234
 235 /* Each element set once, and only accessed by a single processor anyway */
 236 #undef NR_CPUS
 237 #define NR_CPUS 1
 238 int userspace_pid[NR_CPUS];
 239
 240 int start_userspace(unsigned long stub_stack)
 241 {
 242         void *stack;
 243         unsigned long sp;
 244         int pid, status, n, flags, err;
 245
 246         stack = mmap(NULL, UM_KERN_PAGE_SIZE,
 247                      PROT_READ | PROT_WRITE | PROT_EXEC,
 248                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 249         if (stack == MAP_FAILED) {
 250                 err = -errno;
 251                 printk(UM_KERN_ERR "start_userspace : mmap failed, "
 252                        "errno = %d\n", errno);
 253                 return err;
 254         }
 255
 256         sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
 257
 258         flags = CLONE_FILES | SIGCHLD;
 259
 260         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 261         if (pid < 0) {
 262                 err = -errno;
 263                 printk(UM_KERN_ERR "start_userspace : clone failed, "
 264                        "errno = %d\n", errno);
 265                 return err;
 266         }
 267
 268         do {
 269                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
 270                 if (n < 0) {
 271                         err = -errno;
 272                         printk(UM_KERN_ERR "start_userspace : wait failed, "
 273                                "errno = %d\n", errno);
 274                         goto out_kill;
 275                 }
 276         } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
 277
 278         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
 279                 err = -EINVAL;
 280                 printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
 281                        "status = %d\n", status);
 282                 goto out_kill;
 283         }
 284
 285         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 286                    (void *) PTRACE_O_TRACESYSGOOD) < 0) {
 287                 err = -errno;
 288                 printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
 289                        "failed, errno = %d\n", errno);
 290                 goto out_kill;
 291         }
 292
 293         if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
 294                 err = -errno;
 295                 printk(UM_KERN_ERR "start_userspace : munmap failed, "
 296                        "errno = %d\n", errno);
 297                 goto out_kill;
 298         }
 299
 300         return pid;
 301
 302  out_kill:
 303         os_kill_ptraced_process(pid, 1);
 304         return err;
 305 }
 306
 307 void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
 308 {
 309         int err, status, op, pid = userspace_pid[0];
 310         /* To prevent races if using_sysemu changes under us.*/
 311         int local_using_sysemu;
 312         siginfo_t si;
 313
 314         /* Handle any immediate reschedules or signals */
 315         interrupt_end();
 316
 317         while (1) {
 318
 319                 /*
 320                  * This can legitimately fail if the process loads a
 321                  * bogus value into a segment register.  It will
 322                  * segfault and PTRACE_GETREGS will read that value
 323                  * out of the process.  However, PTRACE_SETREGS will
 324                  * fail.  In this case, there is nothing to do but
 325                  * just kill the process.
 326                  */
 327                 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
 328                         fatal_sigsegv();
 329
 330                 if (put_fp_registers(pid, regs->fp))
 331                         fatal_sigsegv();
 332
 333                 /* Now we set local_using_sysemu to be used for one loop */
 334                 local_using_sysemu = get_using_sysemu();
 335
 336                 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
 337                                              singlestepping(NULL));
 338
 339                 if (ptrace(op, pid, 0, 0)) {
 340                         printk(UM_KERN_ERR "userspace - ptrace continue "
 341                                "failed, op = %d, errno = %d\n", op, errno);
 342                         fatal_sigsegv();
 343                 }
 344
 345                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
 346                 if (err < 0) {
 347                         printk(UM_KERN_ERR "userspace - wait failed, "
 348                                "errno = %d\n", errno);
 349                         fatal_sigsegv();
 350                 }
 351
 352                 regs->is_user = 1;
 353                 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
 354                         printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
 355                                "errno = %d\n", errno);
 356                         fatal_sigsegv();
 357                 }
 358
 359                 if (get_fp_registers(pid, regs->fp)) {
 360                         printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
 361                                "errno = %d\n", errno);
 362                         fatal_sigsegv();
 363                 }
 364
 365                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
 366
 367                 if (WIFSTOPPED(status)) {
 368                         int sig = WSTOPSIG(status);
 369
 370                         ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
 371
 372                         switch (sig) {
 373                         case SIGSEGV:
 374                                 if (PTRACE_FULL_FAULTINFO) {
 375                                         get_skas_faultinfo(pid,
 376                                                            &regs->faultinfo, aux_fp_regs);
 377                                         (*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
 378                                                              regs);
 379                                 }
 380                                 else handle_segv(pid, regs, aux_fp_regs);
 381                                 break;
 382                         case SIGTRAP + 0x80:
 383                                 handle_trap(pid, regs, local_using_sysemu);
 384                                 break;
 385                         case SIGTRAP:
 386                                 relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
 387                                 break;
 388                         case SIGALRM:
 389                                 break;
 390                         case SIGIO:
 391                         case SIGILL:
 392                         case SIGBUS:
 393                         case SIGFPE:
 394                         case SIGWINCH:
 395                                 block_signals();
 396                                 (*sig_info[sig])(sig, (struct siginfo *)&si, regs);
 397                                 unblock_signals();
 398                                 break;
 399                         default:
 400                                 printk(UM_KERN_ERR "userspace - child stopped "
 401                                        "with signal %d\n", sig);
 402                                 fatal_sigsegv();
 403                         }
 404                         pid = userspace_pid[0];
 405                         interrupt_end();
 406
 407                         /* Avoid -ERESTARTSYS handling in host */
 408                         if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
 409                                 PT_SYSCALL_NR(regs->gp) = -1;
 410                 }
 411         }
 412 }
 413
 414 static unsigned long thread_regs[MAX_REG_NR];
 415 static unsigned long thread_fp_regs[FP_SIZE];
 416
 417 static int __init init_thread_regs(void)
 418 {
 419         get_safe_registers(thread_regs, thread_fp_regs);
 420         /* Set parent's instruction pointer to start of clone-stub */
 421         thread_regs[REGS_IP_INDEX] = STUB_CODE +
 422                                 (unsigned long) stub_clone_handler -
 423                                 (unsigned long) __syscall_stub_start;
 424         thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
 425                 sizeof(void *);
 426 #ifdef __SIGNAL_FRAMESIZE
 427         thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
 428 #endif
 429         return 0;
 430 }
 431
 432 __initcall(init_thread_regs);
 433
 434 int copy_context_skas0(unsigned long new_stack, int pid)
 435 {
 436         int err;
 437         unsigned long current_stack = current_stub_stack();
 438         struct stub_data *data = (struct stub_data *) current_stack;
 439         struct stub_data *child_data = (struct stub_data *) new_stack;
 440         unsigned long long new_offset;
 441         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 442
 443         /*
 444          * prepare offset and fd of child's stack as argument for parent's
 445          * and child's mmap2 calls
 446          */
 447         *data = ((struct stub_data) {
 448                         .offset = MMAP_OFFSET(new_offset),
 449                         .fd     = new_fd
 450         });
 451
 452         err = ptrace_setregs(pid, thread_regs);
 453         if (err < 0) {
 454                 err = -errno;
 455                 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
 456                        "failed, pid = %d, errno = %d\n", pid, -err);
 457                 return err;
 458         }
 459
 460         err = put_fp_registers(pid, thread_fp_regs);
 461         if (err < 0) {
 462                 printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
 463                        "failed, pid = %d, err = %d\n", pid, err);
 464                 return err;
 465         }
 466
 467         /* set a well known return code for detection of child write failure */
 468         child_data->err = 12345678;
 469
 470         /*
 471          * Wait, until parent has finished its work: read child's pid from
 472          * parent's stack, and check, if bad result.
 473          */
 474         err = ptrace(PTRACE_CONT, pid, 0, 0);
 475         if (err) {
 476                 err = -errno;
 477                 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
 478                        "errno = %d\n", pid, errno);
 479                 return err;
 480         }
 481
 482         wait_stub_done(pid);
 483
 484         pid = data->err;
 485         if (pid < 0) {
 486                 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
 487                        "error %d\n", -pid);
 488                 return pid;
 489         }
 490
 491         /*
 492          * Wait, until child has finished too: read child's result from
 493          * child's stack and check it.
 494          */
 495         wait_stub_done(pid);
 496         if (child_data->err != STUB_DATA) {
 497                 printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
 498                        "error %ld\n", child_data->err);
 499                 err = child_data->err;
 500                 goto out_kill;
 501         }
 502
 503         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 504                    (void *)PTRACE_O_TRACESYSGOOD) < 0) {
 505                 err = -errno;
 506                 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
 507                        "failed, errno = %d\n", errno);
 508                 goto out_kill;
 509         }
 510
 511         return pid;
 512
 513  out_kill:
 514         os_kill_ptraced_process(pid, 1);
 515         return err;
 516 }
 517
 518 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
 519 {
 520         (*buf)[0].JB_IP = (unsigned long) handler;
 521         (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
 522                 sizeof(void *);
 523 }
 524
 525 #define INIT_JMP_NEW_THREAD 0
 526 #define INIT_JMP_CALLBACK 1
 527 #define INIT_JMP_HALT 2
 528 #define INIT_JMP_REBOOT 3
 529
 530 void switch_threads(jmp_buf *me, jmp_buf *you)
 531 {
 532         if (UML_SETJMP(me) == 0)
 533                 UML_LONGJMP(you, 1);
 534 }
 535
 536 static jmp_buf initial_jmpbuf;
 537
 538 /* XXX Make these percpu */
 539 static void (*cb_proc)(void *arg);
 540 static void *cb_arg;
 541 static jmp_buf *cb_back;
 542
 543 int start_idle_thread(void *stack, jmp_buf *switch_buf)
 544 {
 545         int n;
 546
 547         set_handler(SIGWINCH);
 548
 549         /*
 550          * Can't use UML_SETJMP or UML_LONGJMP here because they save
 551          * and restore signals, with the possible side-effect of
 552          * trying to handle any signals which came when they were
 553          * blocked, which can't be done on this stack.
 554          * Signals must be blocked when jumping back here and restored
 555          * after returning to the jumper.
 556          */
 557         n = setjmp(initial_jmpbuf);
 558         switch (n) {
 559         case INIT_JMP_NEW_THREAD:
 560                 (*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
 561                 (*switch_buf)[0].JB_SP = (unsigned long) stack +
 562                         UM_THREAD_SIZE - sizeof(void *);
 563                 break;
 564         case INIT_JMP_CALLBACK:
 565                 (*cb_proc)(cb_arg);
 566                 longjmp(*cb_back, 1);
 567                 break;
 568         case INIT_JMP_HALT:
 569                 kmalloc_ok = 0;
 570                 return 0;
 571         case INIT_JMP_REBOOT:
 572                 kmalloc_ok = 0;
 573                 return 1;
 574         default:
 575                 printk(UM_KERN_ERR "Bad sigsetjmp return in "
 576                        "start_idle_thread - %d\n", n);
 577                 fatal_sigsegv();
 578         }
 579         longjmp(*switch_buf, 1);
 580
 581         /* unreachable */
 582         printk(UM_KERN_ERR "impossible long jump!");
 583         fatal_sigsegv();
 584         return 0;
 585 }
 586
 587 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 588 {
 589         jmp_buf here;
 590
 591         cb_proc = proc;
 592         cb_arg = arg;
 593         cb_back = &here;
 594
 595         block_signals();
 596         if (UML_SETJMP(&here) == 0)
 597                 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
 598         unblock_signals();
 599
 600         cb_proc = NULL;
 601         cb_arg = NULL;
 602         cb_back = NULL;
 603 }
 604
 605 void halt_skas(void)
 606 {
 607         block_signals();
 608         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
 609 }
 610
 611 void reboot_skas(void)
 612 {
 613         block_signals();
 614         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
 615 }
 616
 617 void __switch_mm(struct mm_id *mm_idp)
 618 {
 619         userspace_pid[0] = mm_idp->u.pid;
 620 }