arch/um/kernel/skas/process.c

   1 /*
   2  * Copyright (C) 2002- 2004 Jeff Dike (jdike@addtoit.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <stdlib.h>
   7 #include <string.h>
   8 #include <unistd.h>
   9 #include <errno.h>
  10 #include <signal.h>
  11 #include <setjmp.h>
  12 #include <sched.h>
  13 #include <sys/wait.h>
  14 #include <sys/mman.h>
  15 #include <sys/user.h>
  16 #include <sys/time.h>
  17 #include <asm/unistd.h>
  18 #include <asm/types.h>
  19 #include "user.h"
  20 #include "ptrace_user.h"
  21 #include "time_user.h"
  22 #include "sysdep/ptrace.h"
  23 #include "user_util.h"
  24 #include "kern_util.h"
  25 #include "skas.h"
  26 #include "stub-data.h"
  27 #include "mm_id.h"
  28 #include "sysdep/sigcontext.h"
  29 #include "sysdep/stub.h"
  30 #include "os.h"
  31 #include "proc_mm.h"
  32 #include "skas_ptrace.h"
  33 #include "chan_user.h"
  34 #include "signal_user.h"
  35 #include "registers.h"
  36 #include "mem.h"
  37 #include "uml-config.h"
  38 #include "process.h"
  39
  40 int is_skas_winch(int pid, int fd, void *data)
  41 {
  42         if(pid != os_getpgrp())
  43                 return(0);
  44
  45         register_winch_irq(-1, fd, -1, data);
  46         return(1);
  47 }
  48
  49 void wait_stub_done(int pid, int sig, char * fname)
  50 {
  51         int n, status, err;
  52
  53         do {
  54                 if ( sig != -1 ) {
  55                         err = ptrace(PTRACE_CONT, pid, 0, sig);
  56                         if(err)
  57                                 panic("%s : continue failed, errno = %d\n",
  58                                       fname, errno);
  59                 }
  60                 sig = 0;
  61
  62                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
  63         } while((n >= 0) && WIFSTOPPED(status) &&
  64                 ((WSTOPSIG(status) == SIGVTALRM) ||
  65                  /* running UML inside a detached screen can cause
  66                   * SIGWINCHes
  67                   */
  68                  (WSTOPSIG(status) == SIGWINCH)));
  69
  70         if((n < 0) || !WIFSTOPPED(status) ||
  71            (WSTOPSIG(status) != SIGUSR1 && WSTOPSIG(status) != SIGTRAP)){
  72                 panic("%s : failed to wait for SIGUSR1/SIGTRAP, "
  73                       "pid = %d, n = %d, errno = %d, status = 0x%x\n",
  74                       fname, pid, n, errno, status);
  75         }
  76 }
  77
  78 void get_skas_faultinfo(int pid, struct faultinfo * fi)
  79 {
  80         int err;
  81
  82         if(ptrace_faultinfo){
  83                 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
  84                 if(err)
  85                         panic("get_skas_faultinfo - PTRACE_FAULTINFO failed, "
  86                               "errno = %d\n", errno);
  87
  88                 /* Special handling for i386, which has different structs */
  89                 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
  90                         memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
  91                                sizeof(struct faultinfo) -
  92                                sizeof(struct ptrace_faultinfo));
  93         }
  94         else {
  95                 wait_stub_done(pid, SIGSEGV, "get_skas_faultinfo");
  96
  97                 /* faultinfo is prepared by the stub-segv-handler at start of
  98                  * the stub stack page. We just have to copy it.
  99                  */
 100                 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
 101         }
 102 }
 103
 104 static void handle_segv(int pid, union uml_pt_regs * regs)
 105 {
 106         get_skas_faultinfo(pid, &regs->skas.faultinfo);
 107         segv(regs->skas.faultinfo, 0, 1, NULL);
 108 }
 109
 110 /*To use the same value of using_sysemu as the caller, ask it that value (in local_using_sysemu)*/
 111 static void handle_trap(int pid, union uml_pt_regs *regs, int local_using_sysemu)
 112 {
 113         int err, status;
 114
 115         /* Mark this as a syscall */
 116         UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->skas.regs);
 117
 118         if (!local_using_sysemu)
 119         {
 120                 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET, __NR_getpid);
 121                 if(err < 0)
 122                         panic("handle_trap - nullifying syscall failed errno = %d\n",
 123                               errno);
 124
 125                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
 126                 if(err < 0)
 127                         panic("handle_trap - continuing to end of syscall failed, "
 128                               "errno = %d\n", errno);
 129
 130                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 131                 if((err < 0) || !WIFSTOPPED(status) ||
 132                    (WSTOPSIG(status) != SIGTRAP + 0x80))
 133                         panic("handle_trap - failed to wait at end of syscall, "
 134                               "errno = %d, status = %d\n", errno, status);
 135         }
 136
 137         handle_syscall(regs);
 138 }
 139
 140 extern int __syscall_stub_start;
 141 int stub_code_fd = -1;
 142 __u64 stub_code_offset;
 143
 144 static int userspace_tramp(void *stack)
 145 {
 146         void *addr;
 147
 148         ptrace(PTRACE_TRACEME, 0, 0, 0);
 149
 150         init_new_thread_signals(1);
 151         enable_timer();
 152
 153         if(!proc_mm){
 154                 /* This has a pte, but it can't be mapped in with the usual
 155                  * tlb_flush mechanism because this is part of that mechanism
 156                  */
 157                 addr = mmap64((void *) UML_CONFIG_STUB_CODE, page_size(),
 158                               PROT_EXEC, MAP_FIXED | MAP_PRIVATE,
 159                               stub_code_fd, stub_code_offset);
 160                 if(addr == MAP_FAILED){
 161                         printk("mapping stub code failed, errno = %d\n",
 162                                errno);
 163                         exit(1);
 164                 }
 165
 166                 if(stack != NULL){
 167                         int fd;
 168                         __u64 offset;
 169
 170                         fd = phys_mapping(to_phys(stack), &offset);
 171                         addr = mmap((void *) UML_CONFIG_STUB_DATA, page_size(),
 172                                     PROT_READ | PROT_WRITE,
 173                                     MAP_FIXED | MAP_SHARED, fd, offset);
 174                         if(addr == MAP_FAILED){
 175                                 printk("mapping stub stack failed, "
 176                                        "errno = %d\n", errno);
 177                                 exit(1);
 178                         }
 179                 }
 180         }
 181         if(!ptrace_faultinfo){
 182                 unsigned long v = UML_CONFIG_STUB_CODE +
 183                                   (unsigned long) stub_segv_handler -
 184                                   (unsigned long) &__syscall_stub_start;
 185
 186                 set_sigstack((void *) UML_CONFIG_STUB_DATA, page_size());
 187                 set_handler(SIGSEGV, (void *) v, SA_ONSTACK,
 188                             SIGIO, SIGWINCH, SIGALRM, SIGVTALRM,
 189                             SIGUSR1, -1);
 190         }
 191
 192         os_stop_process(os_getpid());
 193         return(0);
 194 }
 195
 196 /* Each element set once, and only accessed by a single processor anyway */
 197 #undef NR_CPUS
 198 #define NR_CPUS 1
 199 int userspace_pid[NR_CPUS];
 200
 201 int start_userspace(unsigned long stub_stack)
 202 {
 203         void *stack;
 204         unsigned long sp;
 205         int pid, status, n, flags;
 206
 207         if ( stub_code_fd == -1 )
 208                 stub_code_fd = phys_mapping(to_phys(&__syscall_stub_start),
 209                                             &stub_code_offset);
 210
 211         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 212                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 213         if(stack == MAP_FAILED)
 214                 panic("start_userspace : mmap failed, errno = %d", errno);
 215         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
 216
 217         flags = CLONE_FILES | SIGCHLD;
 218         if(proc_mm) flags |= CLONE_VM;
 219         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 220         if(pid < 0)
 221                 panic("start_userspace : clone failed, errno = %d", errno);
 222
 223         do {
 224                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 225                 if(n < 0)
 226                         panic("start_userspace : wait failed, errno = %d",
 227                               errno);
 228         } while(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
 229
 230         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 231                 panic("start_userspace : expected SIGSTOP, got status = %d",
 232                       status);
 233
 234         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, (void *)PTRACE_O_TRACESYSGOOD) < 0)
 235                 panic("start_userspace : PTRACE_SETOPTIONS failed, errno=%d\n",
 236                       errno);
 237
 238         if(munmap(stack, PAGE_SIZE) < 0)
 239                 panic("start_userspace : munmap failed, errno = %d\n", errno);
 240
 241         return(pid);
 242 }
 243
 244 void userspace(union uml_pt_regs *regs)
 245 {
 246         int err, status, op, pid = userspace_pid[0];
 247         int local_using_sysemu; /*To prevent races if using_sysemu changes under us.*/
 248
 249         while(1){
 250                 restore_registers(pid, regs);
 251
 252                 /* Now we set local_using_sysemu to be used for one loop */
 253                 local_using_sysemu = get_using_sysemu();
 254
 255                 op = SELECT_PTRACE_OPERATION(local_using_sysemu, singlestepping(NULL));
 256
 257                 err = ptrace(op, pid, 0, 0);
 258                 if(err)
 259                         panic("userspace - could not resume userspace process, "
 260                               "pid=%d, ptrace operation = %d, errno = %d\n",
 261                               op, errno);
 262
 263                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 264                 if(err < 0)
 265                         panic("userspace - waitpid failed, errno = %d\n",
 266                               errno);
 267
 268                 regs->skas.is_user = 1;
 269                 save_registers(pid, regs);
 270                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
 271
 272                 if(WIFSTOPPED(status)){
 273                         switch(WSTOPSIG(status)){
 274                         case SIGSEGV:
 275                                 if(PTRACE_FULL_FAULTINFO || !ptrace_faultinfo)
 276                                         user_signal(SIGSEGV, regs, pid);
 277                                 else handle_segv(pid, regs);
 278                                 break;
 279                         case SIGTRAP + 0x80:
 280                                 handle_trap(pid, regs, local_using_sysemu);
 281                                 break;
 282                         case SIGTRAP:
 283                                 relay_signal(SIGTRAP, regs);
 284                                 break;
 285                         case SIGIO:
 286                         case SIGVTALRM:
 287                         case SIGILL:
 288                         case SIGBUS:
 289                         case SIGFPE:
 290                         case SIGWINCH:
 291                                 user_signal(WSTOPSIG(status), regs, pid);
 292                                 break;
 293                         default:
 294                                 printk("userspace - child stopped with signal "
 295                                        "%d\n", WSTOPSIG(status));
 296                         }
 297                         pid = userspace_pid[0];
 298                         interrupt_end();
 299
 300                         /* Avoid -ERESTARTSYS handling in host */
 301                         PT_SYSCALL_NR(regs->skas.regs) = -1;
 302                 }
 303         }
 304 }
 305 #define INIT_JMP_NEW_THREAD 0
 306 #define INIT_JMP_REMOVE_SIGSTACK 1
 307 #define INIT_JMP_CALLBACK 2
 308 #define INIT_JMP_HALT 3
 309 #define INIT_JMP_REBOOT 4
 310
 311
 312 int copy_context_skas0(unsigned long new_stack, int pid)
 313 {
 314         int err;
 315         unsigned long regs[MAX_REG_NR];
 316         unsigned long current_stack = current_stub_stack();
 317         struct stub_data *data = (struct stub_data *) current_stack;
 318         struct stub_data *child_data = (struct stub_data *) new_stack;
 319         __u64 new_offset;
 320         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 321
 322         /* prepare offset and fd of child's stack as argument for parent's
 323          * and child's mmap2 calls
 324          */
 325         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
 326                                       .fd       = new_fd,
 327                                       .timer    = ((struct itimerval)
 328                                                    { { 0, 1000000 / hz() },
 329                                                      { 0, 1000000 / hz() }})});
 330         get_safe_registers(regs);
 331
 332         /* Set parent's instruction pointer to start of clone-stub */
 333         regs[REGS_IP_INDEX] = UML_CONFIG_STUB_CODE +
 334                                 (unsigned long) stub_clone_handler -
 335                                 (unsigned long) &__syscall_stub_start;
 336         regs[REGS_SP_INDEX] = UML_CONFIG_STUB_DATA + PAGE_SIZE -
 337                 sizeof(void *);
 338         err = ptrace_setregs(pid, regs);
 339         if(err < 0)
 340                 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
 341                       "pid = %d, errno = %d\n", pid, errno);
 342
 343         /* set a well known return code for detection of child write failure */
 344         child_data->err = 12345678;
 345
 346         /* Wait, until parent has finished its work: read child's pid from
 347          * parent's stack, and check, if bad result.
 348          */
 349         wait_stub_done(pid, 0, "copy_context_skas0");
 350
 351         pid = data->err;
 352         if(pid < 0)
 353                 panic("copy_context_skas0 - stub-parent reports error %d\n",
 354                       pid);
 355
 356         /* Wait, until child has finished too: read child's result from
 357          * child's stack and check it.
 358          */
 359         wait_stub_done(pid, -1, "copy_context_skas0");
 360         if (child_data->err != UML_CONFIG_STUB_DATA)
 361                 panic("copy_context_skas0 - stub-child reports error %d\n",
 362                       child_data->err);
 363
 364         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 365                    (void *)PTRACE_O_TRACESYSGOOD) < 0)
 366                 panic("copy_context_skas0 : PTRACE_SETOPTIONS failed, "
 367                       "errno = %d\n", errno);
 368
 369         return pid;
 370 }
 371
 372 /*
 373  * This is used only, if proc_mm is available, while PTRACE_FAULTINFO
 374  * isn't. Opening /proc/mm creates a new mm_context, which lacks the stub-pages
 375  * Thus, we map them using /proc/mm-fd
 376  */
 377 void map_stub_pages(int fd, unsigned long code,
 378                     unsigned long data, unsigned long stack)
 379 {
 380         struct proc_mm_op mmop;
 381         int n;
 382
 383         mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
 384                                       .u         =
 385                                       { .mmap    =
 386                                         { .addr    = code,
 387                                           .len     = PAGE_SIZE,
 388                                           .prot    = PROT_EXEC,
 389                                           .flags   = MAP_FIXED | MAP_PRIVATE,
 390                                           .fd      = stub_code_fd,
 391                                           .offset  = stub_code_offset
 392         } } });
 393         n = os_write_file(fd, &mmop, sizeof(mmop));
 394         if(n != sizeof(mmop))
 395                 panic("map_stub_pages : /proc/mm map for code failed, "
 396                       "err = %d\n", -n);
 397
 398         if ( stack ) {
 399                 __u64 map_offset;
 400                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
 401                 mmop = ((struct proc_mm_op)
 402                                 { .op        = MM_MMAP,
 403                                   .u         =
 404                                   { .mmap    =
 405                                     { .addr    = data,
 406                                       .len     = PAGE_SIZE,
 407                                       .prot    = PROT_READ | PROT_WRITE,
 408                                       .flags   = MAP_FIXED | MAP_SHARED,
 409                                       .fd      = map_fd,
 410                                       .offset  = map_offset
 411                 } } });
 412                 n = os_write_file(fd, &mmop, sizeof(mmop));
 413                 if(n != sizeof(mmop))
 414                         panic("map_stub_pages : /proc/mm map for data failed, "
 415                               "err = %d\n", -n);
 416         }
 417 }
 418
 419 void new_thread(void *stack, void **switch_buf_ptr, void **fork_buf_ptr,
 420                 void (*handler)(int))
 421 {
 422         unsigned long flags;
 423         sigjmp_buf switch_buf, fork_buf;
 424
 425         *switch_buf_ptr = &switch_buf;
 426         *fork_buf_ptr = &fork_buf;
 427
 428         /* Somewhat subtle - siglongjmp restores the signal mask before doing
 429          * the longjmp.  This means that when jumping from one stack to another
 430          * when the target stack has interrupts enabled, an interrupt may occur
 431          * on the source stack.  This is bad when starting up a process because
 432          * it's not supposed to get timer ticks until it has been scheduled.
 433          * So, we disable interrupts around the sigsetjmp to ensure that
 434          * they can't happen until we get back here where they are safe.
 435          */
 436         flags = get_signals();
 437         block_signals();
 438         if(sigsetjmp(fork_buf, 1) == 0)
 439                 new_thread_proc(stack, handler);
 440
 441         remove_sigstack();
 442
 443         set_signals(flags);
 444 }
 445
 446 void thread_wait(void *sw, void *fb)
 447 {
 448         sigjmp_buf buf, **switch_buf = sw, *fork_buf;
 449
 450         *switch_buf = &buf;
 451         fork_buf = fb;
 452         if(sigsetjmp(buf, 1) == 0)
 453                 siglongjmp(*fork_buf, INIT_JMP_REMOVE_SIGSTACK);
 454 }
 455
 456 void switch_threads(void *me, void *next)
 457 {
 458         sigjmp_buf my_buf, **me_ptr = me, *next_buf = next;
 459
 460         *me_ptr = &my_buf;
 461         if(sigsetjmp(my_buf, 1) == 0)
 462                 siglongjmp(*next_buf, 1);
 463 }
 464
 465 static sigjmp_buf initial_jmpbuf;
 466
 467 /* XXX Make these percpu */
 468 static void (*cb_proc)(void *arg);
 469 static void *cb_arg;
 470 static sigjmp_buf *cb_back;
 471
 472 int start_idle_thread(void *stack, void *switch_buf_ptr, void **fork_buf_ptr)
 473 {
 474         sigjmp_buf **switch_buf = switch_buf_ptr;
 475         int n;
 476
 477         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
 478                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGALRM,
 479                     SIGVTALRM, -1);
 480
 481         *fork_buf_ptr = &initial_jmpbuf;
 482         n = sigsetjmp(initial_jmpbuf, 1);
 483         switch(n){
 484         case INIT_JMP_NEW_THREAD:
 485                 new_thread_proc((void *) stack, new_thread_handler);
 486                 break;
 487         case INIT_JMP_REMOVE_SIGSTACK:
 488                 remove_sigstack();
 489                 break;
 490         case INIT_JMP_CALLBACK:
 491                 (*cb_proc)(cb_arg);
 492                 siglongjmp(*cb_back, 1);
 493                 break;
 494         case INIT_JMP_HALT:
 495                 kmalloc_ok = 0;
 496                 return(0);
 497         case INIT_JMP_REBOOT:
 498                 kmalloc_ok = 0;
 499                 return(1);
 500         default:
 501                 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
 502         }
 503         siglongjmp(**switch_buf, 1);
 504 }
 505
 506 void remove_sigstack(void)
 507 {
 508         stack_t stack = ((stack_t) { .ss_flags  = SS_DISABLE,
 509                                      .ss_sp     = NULL,
 510                                      .ss_size   = 0 });
 511
 512         if(sigaltstack(&stack, NULL) != 0)
 513                 panic("disabling signal stack failed, errno = %d\n", errno);
 514 }
 515
 516 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 517 {
 518         sigjmp_buf here;
 519
 520         cb_proc = proc;
 521         cb_arg = arg;
 522         cb_back = &here;
 523
 524         block_signals();
 525         if(sigsetjmp(here, 1) == 0)
 526                 siglongjmp(initial_jmpbuf, INIT_JMP_CALLBACK);
 527         unblock_signals();
 528
 529         cb_proc = NULL;
 530         cb_arg = NULL;
 531         cb_back = NULL;
 532 }
 533
 534 void halt_skas(void)
 535 {
 536         block_signals();
 537         siglongjmp(initial_jmpbuf, INIT_JMP_HALT);
 538 }
 539
 540 void reboot_skas(void)
 541 {
 542         block_signals();
 543         siglongjmp(initial_jmpbuf, INIT_JMP_REBOOT);
 544 }
 545
 546 void switch_mm_skas(struct mm_id *mm_idp)
 547 {
 548         int err;
 549
 550 #warning need cpu pid in switch_mm_skas
 551         if(proc_mm){
 552                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
 553                              mm_idp->u.mm_fd);
 554                 if(err)
 555                         panic("switch_mm_skas - PTRACE_SWITCH_MM failed, "
 556                               "errno = %d\n", errno);
 557         }
 558         else userspace_pid[0] = mm_idp->u.pid;
 559 }
 560
 561 /*
 562  * Overrides for Emacs so that we follow Linus's tabbing style.
 563  * Emacs will notice this stuff at the end of the file and automatically
 564  * adjust the settings for this buffer only.  This must remain at the end
 565  * of the file.
 566  * ---------------------------------------------------------------------------
 567  * Local variables:
 568  * c-file-style: "linux"
 569  * End:
 570  */