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