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

   1 /*
   2  * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <stdlib.h>
   7 #include <unistd.h>
   8 #include <sched.h>
   9 #include <errno.h>
  10 #include <string.h>
  11 #include <sys/mman.h>
  12 #include <sys/ptrace.h>
  13 #include <sys/wait.h>
  14 #include <asm/unistd.h>
  15 #include "as-layout.h"
  16 #include "chan_user.h"
  17 #include "kern_constants.h"
  18 #include "mem.h"
  19 #include "os.h"
  20 #include "process.h"
  21 #include "proc_mm.h"
  22 #include "ptrace_user.h"
  23 #include "registers.h"
  24 #include "skas.h"
  25 #include "skas_ptrace.h"
  26 #include "user.h"
  27 #include "sysdep/stub.h"
  28
  29 int is_skas_winch(int pid, int fd, void *data)
  30 {
  31         if (pid != getpgrp())
  32                 return 0;
  33
  34         register_winch_irq(-1, fd, -1, data, 0);
  35         return 1;
  36 }
  37
  38 static int ptrace_dump_regs(int pid)
  39 {
  40         unsigned long regs[MAX_REG_NR];
  41         int i;
  42
  43         if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
  44                 return -errno;
  45
  46         printk(UM_KERN_ERR "Stub registers -\n");
  47         for (i = 0; i < ARRAY_SIZE(regs); i++)
  48                 printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
  49
  50         return 0;
  51 }
  52
  53 /*
  54  * Signals that are OK to receive in the stub - we'll just continue it.
  55  * SIGWINCH will happen when UML is inside a detached screen.
  56  */
  57 #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
  58
  59 /* Signals that the stub will finish with - anything else is an error */
  60 #define STUB_DONE_MASK ((1 << SIGUSR1) | (1 << SIGTRAP))
  61
  62 void wait_stub_done(int pid)
  63 {
  64         int n, status, err;
  65
  66         while (1) {
  67                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
  68                 if ((n < 0) || !WIFSTOPPED(status))
  69                         goto bad_wait;
  70
  71                 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
  72                         break;
  73
  74                 err = ptrace(PTRACE_CONT, pid, 0, 0);
  75                 if (err)
  76                         panic("wait_stub_done : continue failed, errno = %d\n",
  77                               errno);
  78         }
  79
  80         if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
  81                 return;
  82
  83 bad_wait:
  84         err = ptrace_dump_regs(pid);
  85         if (err)
  86                 printk(UM_KERN_ERR "Failed to get registers from stub, "
  87                        "errno = %d\n", -err);
  88         panic("wait_stub_done : failed to wait for SIGUSR1/SIGTRAP, pid = %d, "
  89               "n = %d, errno = %d, status = 0x%x\n", pid, n, errno, status);
  90 }
  91
  92 extern unsigned long current_stub_stack(void);
  93
  94 void get_skas_faultinfo(int pid, struct faultinfo * fi)
  95 {
  96         int err;
  97
  98         if (ptrace_faultinfo) {
  99                 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
 100                 if (err)
 101                         panic("get_skas_faultinfo - PTRACE_FAULTINFO failed, "
 102                               "errno = %d\n", errno);
 103
 104                 /* Special handling for i386, which has different structs */
 105                 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
 106                         memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
 107                                sizeof(struct faultinfo) -
 108                                sizeof(struct ptrace_faultinfo));
 109         }
 110         else {
 111                 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
 112                 if (err)
 113                         panic("Failed to continue stub, pid = %d, errno = %d\n",
 114                               pid, errno);
 115                 wait_stub_done(pid);
 116
 117                 /*
 118                  * faultinfo is prepared by the stub-segv-handler at start of
 119                  * the stub stack page. We just have to copy it.
 120                  */
 121                 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
 122         }
 123 }
 124
 125 static void handle_segv(int pid, struct uml_pt_regs * regs)
 126 {
 127         get_skas_faultinfo(pid, &regs->faultinfo);
 128         segv(regs->faultinfo, 0, 1, NULL);
 129 }
 130
 131 /*
 132  * To use the same value of using_sysemu as the caller, ask it that value
 133  * (in local_using_sysemu
 134  */
 135 static void handle_trap(int pid, struct uml_pt_regs *regs,
 136                         int local_using_sysemu)
 137 {
 138         int err, status;
 139
 140         /* Mark this as a syscall */
 141         UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
 142
 143         if (!local_using_sysemu)
 144         {
 145                 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 146                              __NR_getpid);
 147                 if (err < 0)
 148                         panic("handle_trap - nullifying syscall failed, "
 149                               "errno = %d\n", errno);
 150
 151                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
 152                 if (err < 0)
 153                         panic("handle_trap - continuing to end of syscall "
 154                               "failed, errno = %d\n", errno);
 155
 156                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 157                 if ((err < 0) || !WIFSTOPPED(status) ||
 158                    (WSTOPSIG(status) != SIGTRAP + 0x80)) {
 159                         err = ptrace_dump_regs(pid);
 160                         if (err)
 161                                 printk(UM_KERN_ERR "Failed to get registers "
 162                                        "from process, errno = %d\n", -err);
 163                         panic("handle_trap - failed to wait at end of syscall, "
 164                               "errno = %d, status = %d\n", errno, status);
 165                 }
 166         }
 167
 168         handle_syscall(regs);
 169 }
 170
 171 extern int __syscall_stub_start;
 172
 173 static int userspace_tramp(void *stack)
 174 {
 175         void *addr;
 176         int err;
 177
 178         ptrace(PTRACE_TRACEME, 0, 0, 0);
 179
 180         init_new_thread_signals();
 181         err = set_interval();
 182         if (err)
 183                 panic("userspace_tramp - setting timer failed, errno = %d\n",
 184                       err);
 185
 186         if (!proc_mm) {
 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                 int fd;
 192                 unsigned long long offset;
 193                 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
 194                 addr = mmap64((void *) UML_CONFIG_STUB_CODE, UM_KERN_PAGE_SIZE,
 195                               PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
 196                 if (addr == MAP_FAILED) {
 197                         printk(UM_KERN_ERR "mapping mmap stub failed, "
 198                                "errno = %d\n", errno);
 199                         exit(1);
 200                 }
 201
 202                 if (stack != NULL) {
 203                         fd = phys_mapping(to_phys(stack), &offset);
 204                         addr = mmap((void *) UML_CONFIG_STUB_DATA,
 205                                     UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
 206                                     MAP_FIXED | MAP_SHARED, fd, offset);
 207                         if (addr == MAP_FAILED) {
 208                                 printk(UM_KERN_ERR "mapping segfault stack "
 209                                        "failed, errno = %d\n", errno);
 210                                 exit(1);
 211                         }
 212                 }
 213         }
 214         if (!ptrace_faultinfo && (stack != NULL)) {
 215                 struct sigaction sa;
 216
 217                 unsigned long v = UML_CONFIG_STUB_CODE +
 218                                   (unsigned long) stub_segv_handler -
 219                                   (unsigned long) &__syscall_stub_start;
 220
 221                 set_sigstack((void *) UML_CONFIG_STUB_DATA, UM_KERN_PAGE_SIZE);
 222                 sigemptyset(&sa.sa_mask);
 223                 sigaddset(&sa.sa_mask, SIGIO);
 224                 sigaddset(&sa.sa_mask, SIGWINCH);
 225                 sigaddset(&sa.sa_mask, SIGALRM);
 226                 sigaddset(&sa.sa_mask, SIGVTALRM);
 227                 sigaddset(&sa.sa_mask, SIGUSR1);
 228                 sa.sa_flags = SA_ONSTACK;
 229                 sa.sa_handler = (void *) v;
 230                 sa.sa_restorer = NULL;
 231                 if (sigaction(SIGSEGV, &sa, NULL) < 0)
 232                         panic("userspace_tramp - setting SIGSEGV handler "
 233                               "failed - errno = %d\n", errno);
 234         }
 235
 236         kill(os_getpid(), SIGSTOP);
 237         return 0;
 238 }
 239
 240 /* Each element set once, and only accessed by a single processor anyway */
 241 #undef NR_CPUS
 242 #define NR_CPUS 1
 243 int userspace_pid[NR_CPUS];
 244
 245 int start_userspace(unsigned long stub_stack)
 246 {
 247         void *stack;
 248         unsigned long sp;
 249         int pid, status, n, flags;
 250
 251         stack = mmap(NULL, UM_KERN_PAGE_SIZE,
 252                      PROT_READ | PROT_WRITE | PROT_EXEC,
 253                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 254         if (stack == MAP_FAILED)
 255                 panic("start_userspace : mmap failed, errno = %d", errno);
 256         sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
 257
 258         flags = CLONE_FILES | SIGCHLD;
 259         if (proc_mm)
 260                 flags |= CLONE_VM;
 261
 262         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 263         if (pid < 0)
 264                 panic("start_userspace : clone failed, errno = %d", errno);
 265
 266         do {
 267                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 268                 if (n < 0)
 269                         panic("start_userspace : wait failed, errno = %d",
 270                               errno);
 271         } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
 272
 273         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 274                 panic("start_userspace : expected SIGSTOP, got status = %d",
 275                       status);
 276
 277         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 278                    (void *) PTRACE_O_TRACESYSGOOD) < 0)
 279                 panic("start_userspace : PTRACE_OLDSETOPTIONS failed, "
 280                       "errno = %d\n", errno);
 281
 282         if (munmap(stack, UM_KERN_PAGE_SIZE) < 0)
 283                 panic("start_userspace : munmap failed, errno = %d\n", errno);
 284
 285         return pid;
 286 }
 287
 288 void userspace(struct 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->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 ||
 323                                     !ptrace_faultinfo) {
 324                                         get_skas_faultinfo(pid,
 325                                                            &regs->faultinfo);
 326                                         (*sig_info[SIGSEGV])(SIGSEGV, regs);
 327                                 }
 328                                 else handle_segv(pid, regs);
 329                                 break;
 330                         case SIGTRAP + 0x80:
 331                                 handle_trap(pid, regs, local_using_sysemu);
 332                                 break;
 333                         case SIGTRAP:
 334                                 relay_signal(SIGTRAP, regs);
 335                                 break;
 336                         case SIGIO:
 337                         case SIGVTALRM:
 338                         case SIGILL:
 339                         case SIGBUS:
 340                         case SIGFPE:
 341                         case SIGWINCH:
 342                                 block_signals();
 343                                 (*sig_info[sig])(sig, regs);
 344                                 unblock_signals();
 345                                 break;
 346                         default:
 347                                 printk(UM_KERN_ERR "userspace - child stopped "
 348                                        "with signal %d\n", sig);
 349                         }
 350                         pid = userspace_pid[0];
 351                         interrupt_end();
 352
 353                         /* Avoid -ERESTARTSYS handling in host */
 354                         if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
 355                                 PT_SYSCALL_NR(regs->gp) = -1;
 356                 }
 357         }
 358 }
 359
 360 static unsigned long thread_regs[MAX_REG_NR];
 361
 362 static int __init init_thread_regs(void)
 363 {
 364         get_safe_registers(thread_regs);
 365         /* Set parent's instruction pointer to start of clone-stub */
 366         thread_regs[REGS_IP_INDEX] = UML_CONFIG_STUB_CODE +
 367                                 (unsigned long) stub_clone_handler -
 368                                 (unsigned long) &__syscall_stub_start;
 369         thread_regs[REGS_SP_INDEX] = UML_CONFIG_STUB_DATA + UM_KERN_PAGE_SIZE -
 370                 sizeof(void *);
 371 #ifdef __SIGNAL_FRAMESIZE
 372         thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
 373 #endif
 374         return 0;
 375 }
 376
 377 __initcall(init_thread_regs);
 378
 379 int copy_context_skas0(unsigned long new_stack, int pid)
 380 {
 381         int err;
 382         unsigned long current_stack = current_stub_stack();
 383         struct stub_data *data = (struct stub_data *) current_stack;
 384         struct stub_data *child_data = (struct stub_data *) new_stack;
 385         unsigned long long new_offset;
 386         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 387
 388         /*
 389          * prepare offset and fd of child's stack as argument for parent's
 390          * and child's mmap2 calls
 391          */
 392         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
 393                                       .fd       = new_fd,
 394                                       .timer    = ((struct itimerval)
 395                                                     { { 0, 1000000 / UM_HZ },
 396                                                       { 0, 1000000 / UM_HZ }})
 397                                  });
 398         err = ptrace_setregs(pid, thread_regs);
 399         if (err < 0)
 400                 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
 401                       "pid = %d, errno = %d\n", pid, -err);
 402
 403         /* set a well known return code for detection of child write failure */
 404         child_data->err = 12345678;
 405
 406         /*
 407          * Wait, until parent has finished its work: read child's pid from
 408          * parent's stack, and check, if bad result.
 409          */
 410         err = ptrace(PTRACE_CONT, pid, 0, 0);
 411         if (err)
 412                 panic("Failed to continue new process, pid = %d, "
 413                       "errno = %d\n", pid, errno);
 414         wait_stub_done(pid);
 415
 416         pid = data->err;
 417         if (pid < 0)
 418                 panic("copy_context_skas0 - stub-parent reports error %d\n",
 419                       -pid);
 420
 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         unsigned long long 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     = UM_KERN_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(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
 466                        "offset = %llx\n", code, code_fd,
 467                        (unsigned long long) code_offset);
 468                 panic("map_stub_pages : /proc/mm map for code failed, "
 469                       "err = %d\n", n);
 470         }
 471
 472         if (stack) {
 473                 unsigned long long map_offset;
 474                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
 475                 mmop = ((struct proc_mm_op)
 476                                 { .op        = MM_MMAP,
 477                                   .u         =
 478                                   { .mmap    =
 479                                     { .addr    = data,
 480                                       .len     = UM_KERN_PAGE_SIZE,
 481                                       .prot    = PROT_READ | PROT_WRITE,
 482                                       .flags   = MAP_FIXED | MAP_SHARED,
 483                                       .fd      = map_fd,
 484                                       .offset  = map_offset
 485                 } } });
 486                 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
 487                 if (n != sizeof(mmop))
 488                         panic("map_stub_pages : /proc/mm map for data failed, "
 489                               "err = %d\n", errno);
 490         }
 491 }
 492
 493 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
 494 {
 495         (*buf)[0].JB_IP = (unsigned long) handler;
 496         (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
 497                 sizeof(void *);
 498 }
 499
 500 #define INIT_JMP_NEW_THREAD 0
 501 #define INIT_JMP_CALLBACK 1
 502 #define INIT_JMP_HALT 2
 503 #define INIT_JMP_REBOOT 3
 504
 505 void switch_threads(jmp_buf *me, jmp_buf *you)
 506 {
 507         if (UML_SETJMP(me) == 0)
 508                 UML_LONGJMP(you, 1);
 509 }
 510
 511 static jmp_buf initial_jmpbuf;
 512
 513 /* XXX Make these percpu */
 514 static void (*cb_proc)(void *arg);
 515 static void *cb_arg;
 516 static jmp_buf *cb_back;
 517
 518 int start_idle_thread(void *stack, jmp_buf *switch_buf)
 519 {
 520         int n;
 521
 522         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
 523                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGALRM,
 524                     SIGVTALRM, -1);
 525
 526         /*
 527          * Can't use UML_SETJMP or UML_LONGJMP here because they save
 528          * and restore signals, with the possible side-effect of
 529          * trying to handle any signals which came when they were
 530          * blocked, which can't be done on this stack.
 531          * Signals must be blocked when jumping back here and restored
 532          * after returning to the jumper.
 533          */
 534         n = setjmp(initial_jmpbuf);
 535         switch(n) {
 536         case INIT_JMP_NEW_THREAD:
 537                 (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
 538                 (*switch_buf)[0].JB_SP = (unsigned long) stack +
 539                         UM_THREAD_SIZE - sizeof(void *);
 540                 break;
 541         case INIT_JMP_CALLBACK:
 542                 (*cb_proc)(cb_arg);
 543                 longjmp(*cb_back, 1);
 544                 break;
 545         case INIT_JMP_HALT:
 546                 kmalloc_ok = 0;
 547                 return 0;
 548         case INIT_JMP_REBOOT:
 549                 kmalloc_ok = 0;
 550                 return 1;
 551         default:
 552                 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
 553         }
 554         longjmp(*switch_buf, 1);
 555 }
 556
 557 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 558 {
 559         jmp_buf here;
 560
 561         cb_proc = proc;
 562         cb_arg = arg;
 563         cb_back = &here;
 564
 565         block_signals();
 566         if (UML_SETJMP(&here) == 0)
 567                 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
 568         unblock_signals();
 569
 570         cb_proc = NULL;
 571         cb_arg = NULL;
 572         cb_back = NULL;
 573 }
 574
 575 void halt_skas(void)
 576 {
 577         block_signals();
 578         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
 579 }
 580
 581 void reboot_skas(void)
 582 {
 583         block_signals();
 584         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
 585 }
 586
 587 void __switch_mm(struct mm_id *mm_idp)
 588 {
 589         int err;
 590
 591         /* FIXME: need cpu pid in __switch_mm */
 592         if (proc_mm) {
 593                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
 594                              mm_idp->u.mm_fd);
 595                 if (err)
 596                         panic("__switch_mm - PTRACE_SWITCH_MM failed, "
 597                               "errno = %d\n", errno);
 598         }
 599         else userspace_pid[0] = mm_idp->u.pid;
 600 }