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

   1 /*
   2  * Copyright (C) 2002- 2004 Jeff Dike (jdike@addtoit.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <stdlib.h>
   7 #include <string.h>
   8 #include <unistd.h>
   9 #include <errno.h>
  10 #include <signal.h>
  11 #include <setjmp.h>
  12 #include <sched.h>
  13 #include "ptrace_user.h"
  14 #include <sys/wait.h>
  15 #include <sys/mman.h>
  16 #include <sys/user.h>
  17 #include <sys/time.h>
  18 #include <asm/unistd.h>
  19 #include <asm/types.h>
  20 #include <asm/page.h>
  21 #include "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 #include "longjmp.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                 unsigned long regs[HOST_FRAME_SIZE];
  73
  74                 if(ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
  75                         printk("Failed to get registers from stub, "
  76                                "errno = %d\n", errno);
  77                 else {
  78                         int i;
  79
  80                         printk("Stub registers -\n");
  81                         for(i = 0; i < HOST_FRAME_SIZE; i++)
  82                                 printk("\t%d - %lx\n", i, regs[i]);
  83                 }
  84                 panic("%s : failed to wait for SIGUSR1/SIGTRAP, "
  85                       "pid = %d, n = %d, errno = %d, status = 0x%x\n",
  86                       fname, pid, n, errno, status);
  87         }
  88 }
  89
  90 extern unsigned long current_stub_stack(void);
  91
  92 void get_skas_faultinfo(int pid, struct faultinfo * fi)
  93 {
  94         int err;
  95
  96         if(ptrace_faultinfo){
  97                 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
  98                 if(err)
  99                         panic("get_skas_faultinfo - PTRACE_FAULTINFO failed, "
 100                               "errno = %d\n", errno);
 101
 102                 /* Special handling for i386, which has different structs */
 103                 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
 104                         memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
 105                                sizeof(struct faultinfo) -
 106                                sizeof(struct ptrace_faultinfo));
 107         }
 108         else {
 109                 wait_stub_done(pid, SIGSEGV, "get_skas_faultinfo");
 110
 111                 /* faultinfo is prepared by the stub-segv-handler at start of
 112                  * the stub stack page. We just have to copy it.
 113                  */
 114                 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
 115         }
 116 }
 117
 118 static void handle_segv(int pid, union uml_pt_regs * regs)
 119 {
 120         get_skas_faultinfo(pid, &regs->skas.faultinfo);
 121         segv(regs->skas.faultinfo, 0, 1, NULL);
 122 }
 123
 124 /*To use the same value of using_sysemu as the caller, ask it that value (in local_using_sysemu)*/
 125 static void handle_trap(int pid, union uml_pt_regs *regs, int local_using_sysemu)
 126 {
 127         int err, status;
 128
 129         /* Mark this as a syscall */
 130         UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->skas.regs);
 131
 132         if (!local_using_sysemu)
 133         {
 134                 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 135                              __NR_getpid);
 136                 if(err < 0)
 137                         panic("handle_trap - nullifying syscall failed errno = %d\n",
 138                               errno);
 139
 140                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
 141                 if(err < 0)
 142                         panic("handle_trap - continuing to end of syscall failed, "
 143                               "errno = %d\n", errno);
 144
 145                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 146                 if((err < 0) || !WIFSTOPPED(status) ||
 147                    (WSTOPSIG(status) != SIGTRAP + 0x80))
 148                         panic("handle_trap - failed to wait at end of syscall, "
 149                               "errno = %d, status = %d\n", errno, status);
 150         }
 151
 152         handle_syscall(regs);
 153 }
 154
 155 extern int __syscall_stub_start;
 156
 157 static int userspace_tramp(void *stack)
 158 {
 159         void *addr;
 160
 161         ptrace(PTRACE_TRACEME, 0, 0, 0);
 162
 163         init_new_thread_signals(1);
 164         enable_timer();
 165
 166         if(!proc_mm){
 167                 /* This has a pte, but it can't be mapped in with the usual
 168                  * tlb_flush mechanism because this is part of that mechanism
 169                  */
 170                 int fd;
 171                 __u64 offset;
 172                 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
 173                 addr = mmap64((void *) UML_CONFIG_STUB_CODE, page_size(),
 174                               PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
 175                 if(addr == MAP_FAILED){
 176                         printk("mapping mmap stub failed, errno = %d\n",
 177                                errno);
 178                         exit(1);
 179                 }
 180
 181                 if(stack != NULL){
 182                         fd = phys_mapping(to_phys(stack), &offset);
 183                         addr = mmap((void *) UML_CONFIG_STUB_DATA, page_size(),
 184                                     PROT_READ | PROT_WRITE,
 185                                     MAP_FIXED | MAP_SHARED, fd, offset);
 186                         if(addr == MAP_FAILED){
 187                                 printk("mapping segfault stack failed, "
 188                                        "errno = %d\n", errno);
 189                                 exit(1);
 190                         }
 191                 }
 192         }
 193         if(!ptrace_faultinfo && (stack != NULL)){
 194                 unsigned long v = UML_CONFIG_STUB_CODE +
 195                                   (unsigned long) stub_segv_handler -
 196                                   (unsigned long) &__syscall_stub_start;
 197
 198                 set_sigstack((void *) UML_CONFIG_STUB_DATA, page_size());
 199                 set_handler(SIGSEGV, (void *) v, SA_ONSTACK,
 200                             SIGIO, SIGWINCH, SIGALRM, SIGVTALRM,
 201                             SIGUSR1, -1);
 202         }
 203
 204         os_stop_process(os_getpid());
 205         return(0);
 206 }
 207
 208 /* Each element set once, and only accessed by a single processor anyway */
 209 #undef NR_CPUS
 210 #define NR_CPUS 1
 211 int userspace_pid[NR_CPUS];
 212
 213 int start_userspace(unsigned long stub_stack)
 214 {
 215         void *stack;
 216         unsigned long sp;
 217         int pid, status, n, flags;
 218
 219         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 220                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 221         if(stack == MAP_FAILED)
 222                 panic("start_userspace : mmap failed, errno = %d", errno);
 223         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
 224
 225         flags = CLONE_FILES | SIGCHLD;
 226         if(proc_mm) flags |= CLONE_VM;
 227         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 228         if(pid < 0)
 229                 panic("start_userspace : clone failed, errno = %d", errno);
 230
 231         do {
 232                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
 233                 if(n < 0)
 234                         panic("start_userspace : wait failed, errno = %d",
 235                               errno);
 236         } while(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
 237
 238         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
 239                 panic("start_userspace : expected SIGSTOP, got status = %d",
 240                       status);
 241
 242         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, (void *)PTRACE_O_TRACESYSGOOD) < 0)
 243                 panic("start_userspace : PTRACE_OLDSETOPTIONS failed, errno=%d\n",
 244                       errno);
 245
 246         if(munmap(stack, PAGE_SIZE) < 0)
 247                 panic("start_userspace : munmap failed, errno = %d\n", errno);
 248
 249         return(pid);
 250 }
 251
 252 void userspace(union uml_pt_regs *regs)
 253 {
 254         int err, status, op, pid = userspace_pid[0];
 255         int local_using_sysemu; /*To prevent races if using_sysemu changes under us.*/
 256
 257         while(1){
 258                 restore_registers(pid, regs);
 259
 260                 /* Now we set local_using_sysemu to be used for one loop */
 261                 local_using_sysemu = get_using_sysemu();
 262
 263                 op = SELECT_PTRACE_OPERATION(local_using_sysemu, singlestepping(NULL));
 264
 265                 err = ptrace(op, pid, 0, 0);
 266                 if(err)
 267                         panic("userspace - could not resume userspace process, "
 268                               "pid=%d, ptrace operation = %d, errno = %d\n",
 269                               op, errno);
 270
 271                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
 272                 if(err < 0)
 273                         panic("userspace - waitpid failed, errno = %d\n",
 274                               errno);
 275
 276                 regs->skas.is_user = 1;
 277                 save_registers(pid, regs);
 278                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
 279
 280                 if(WIFSTOPPED(status)){
 281                         switch(WSTOPSIG(status)){
 282                         case SIGSEGV:
 283                                 if(PTRACE_FULL_FAULTINFO || !ptrace_faultinfo)
 284                                         user_signal(SIGSEGV, regs, pid);
 285                                 else handle_segv(pid, regs);
 286                                 break;
 287                         case SIGTRAP + 0x80:
 288                                 handle_trap(pid, regs, local_using_sysemu);
 289                                 break;
 290                         case SIGTRAP:
 291                                 relay_signal(SIGTRAP, regs);
 292                                 break;
 293                         case SIGIO:
 294                         case SIGVTALRM:
 295                         case SIGILL:
 296                         case SIGBUS:
 297                         case SIGFPE:
 298                         case SIGWINCH:
 299                                 user_signal(WSTOPSIG(status), regs, pid);
 300                                 break;
 301                         default:
 302                                 printk("userspace - child stopped with signal "
 303                                        "%d\n", WSTOPSIG(status));
 304                         }
 305                         pid = userspace_pid[0];
 306                         interrupt_end();
 307
 308                         /* Avoid -ERESTARTSYS handling in host */
 309                         if(PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
 310                                 PT_SYSCALL_NR(regs->skas.regs) = -1;
 311                 }
 312         }
 313 }
 314
 315 int copy_context_skas0(unsigned long new_stack, int pid)
 316 {
 317         int err;
 318         unsigned long regs[HOST_FRAME_SIZE];
 319         unsigned long fp_regs[HOST_FP_SIZE];
 320         unsigned long current_stack = current_stub_stack();
 321         struct stub_data *data = (struct stub_data *) current_stack;
 322         struct stub_data *child_data = (struct stub_data *) new_stack;
 323         __u64 new_offset;
 324         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 325
 326         /* prepare offset and fd of child's stack as argument for parent's
 327          * and child's mmap2 calls
 328          */
 329         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
 330                                       .fd       = new_fd,
 331                                       .timer    = ((struct itimerval)
 332                                                     { { 0, 1000000 / hz() },
 333                                                       { 0, 1000000 / hz() }})});
 334         get_safe_registers(regs, fp_regs);
 335
 336         /* Set parent's instruction pointer to start of clone-stub */
 337         regs[REGS_IP_INDEX] = UML_CONFIG_STUB_CODE +
 338                                 (unsigned long) stub_clone_handler -
 339                                 (unsigned long) &__syscall_stub_start;
 340         regs[REGS_SP_INDEX] = UML_CONFIG_STUB_DATA + PAGE_SIZE -
 341                 sizeof(void *);
 342 #ifdef __SIGNAL_FRAMESIZE
 343         regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
 344 #endif
 345         err = ptrace_setregs(pid, regs);
 346         if(err < 0)
 347                 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
 348                       "pid = %d, errno = %d\n", pid, errno);
 349
 350         err = ptrace_setfpregs(pid, fp_regs);
 351         if(err < 0)
 352                 panic("copy_context_skas0 : PTRACE_SETFPREGS failed, "
 353                       "pid = %d, errno = %d\n", pid, errno);
 354
 355         /* set a well known return code for detection of child write failure */
 356         child_data->err = 12345678;
 357
 358         /* Wait, until parent has finished its work: read child's pid from
 359          * parent's stack, and check, if bad result.
 360          */
 361         wait_stub_done(pid, 0, "copy_context_skas0");
 362
 363         pid = data->err;
 364         if(pid < 0)
 365                 panic("copy_context_skas0 - stub-parent reports error %d\n",
 366                       pid);
 367
 368         /* Wait, until child has finished too: read child's result from
 369          * child's stack and check it.
 370          */
 371         wait_stub_done(pid, -1, "copy_context_skas0");
 372         if (child_data->err != UML_CONFIG_STUB_DATA)
 373                 panic("copy_context_skas0 - stub-child reports error %d\n",
 374                       child_data->err);
 375
 376         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 377                    (void *)PTRACE_O_TRACESYSGOOD) < 0)
 378                 panic("copy_context_skas0 : PTRACE_OLDSETOPTIONS failed, "
 379                       "errno = %d\n", errno);
 380
 381         return pid;
 382 }
 383
 384 /*
 385  * This is used only, if stub pages are needed, while proc_mm is
 386  * availabl. Opening /proc/mm creates a new mm_context, which lacks
 387  * the stub-pages. Thus, we map them using /proc/mm-fd
 388  */
 389 void map_stub_pages(int fd, unsigned long code,
 390                     unsigned long data, unsigned long stack)
 391 {
 392         struct proc_mm_op mmop;
 393         int n;
 394         __u64 code_offset;
 395         int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
 396                                    &code_offset);
 397
 398         mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
 399                                       .u         =
 400                                       { .mmap    =
 401                                         { .addr    = code,
 402                                           .len     = PAGE_SIZE,
 403                                           .prot    = PROT_EXEC,
 404                                           .flags   = MAP_FIXED | MAP_PRIVATE,
 405                                           .fd      = code_fd,
 406                                           .offset  = code_offset
 407         } } });
 408         n = os_write_file(fd, &mmop, sizeof(mmop));
 409         if(n != sizeof(mmop))
 410                 panic("map_stub_pages : /proc/mm map for code failed, "
 411                       "err = %d\n", -n);
 412
 413         if ( stack ) {
 414                 __u64 map_offset;
 415                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
 416                 mmop = ((struct proc_mm_op)
 417                                 { .op        = MM_MMAP,
 418                                   .u         =
 419                                   { .mmap    =
 420                                     { .addr    = data,
 421                                       .len     = PAGE_SIZE,
 422                                       .prot    = PROT_READ | PROT_WRITE,
 423                                       .flags   = MAP_FIXED | MAP_SHARED,
 424                                       .fd      = map_fd,
 425                                       .offset  = map_offset
 426                 } } });
 427                 n = os_write_file(fd, &mmop, sizeof(mmop));
 428                 if(n != sizeof(mmop))
 429                         panic("map_stub_pages : /proc/mm map for data failed, "
 430                               "err = %d\n", -n);
 431         }
 432 }
 433
 434 void new_thread(void *stack, void **switch_buf_ptr, void **fork_buf_ptr,
 435                 void (*handler)(int))
 436 {
 437         unsigned long flags;
 438         sigjmp_buf switch_buf, fork_buf;
 439         int enable;
 440
 441         *switch_buf_ptr = &switch_buf;
 442         *fork_buf_ptr = &fork_buf;
 443
 444         /* Somewhat subtle - siglongjmp restores the signal mask before doing
 445          * the longjmp.  This means that when jumping from one stack to another
 446          * when the target stack has interrupts enabled, an interrupt may occur
 447          * on the source stack.  This is bad when starting up a process because
 448          * it's not supposed to get timer ticks until it has been scheduled.
 449          * So, we disable interrupts around the sigsetjmp to ensure that
 450          * they can't happen until we get back here where they are safe.
 451          */
 452         flags = get_signals();
 453         block_signals();
 454         if(UML_SIGSETJMP(&fork_buf, enable) == 0)
 455                 new_thread_proc(stack, handler);
 456
 457         remove_sigstack();
 458
 459         set_signals(flags);
 460 }
 461
 462 #define INIT_JMP_NEW_THREAD 0
 463 #define INIT_JMP_REMOVE_SIGSTACK 1
 464 #define INIT_JMP_CALLBACK 2
 465 #define INIT_JMP_HALT 3
 466 #define INIT_JMP_REBOOT 4
 467
 468 void thread_wait(void *sw, void *fb)
 469 {
 470         sigjmp_buf buf, **switch_buf = sw, *fork_buf;
 471         int enable;
 472
 473         *switch_buf = &buf;
 474         fork_buf = fb;
 475         if(UML_SIGSETJMP(&buf, enable) == 0)
 476                 siglongjmp(*fork_buf, INIT_JMP_REMOVE_SIGSTACK);
 477 }
 478
 479 void switch_threads(void *me, void *next)
 480 {
 481         sigjmp_buf my_buf, **me_ptr = me, *next_buf = next;
 482         int enable;
 483
 484         *me_ptr = &my_buf;
 485         if(UML_SIGSETJMP(&my_buf, enable) == 0)
 486                 UML_SIGLONGJMP(next_buf, 1);
 487 }
 488
 489 static sigjmp_buf initial_jmpbuf;
 490
 491 /* XXX Make these percpu */
 492 static void (*cb_proc)(void *arg);
 493 static void *cb_arg;
 494 static sigjmp_buf *cb_back;
 495
 496 int start_idle_thread(void *stack, void *switch_buf_ptr, void **fork_buf_ptr)
 497 {
 498         sigjmp_buf **switch_buf = switch_buf_ptr;
 499         int n, enable;
 500
 501         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
 502                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGALRM,
 503                     SIGVTALRM, -1);
 504
 505         *fork_buf_ptr = &initial_jmpbuf;
 506         n = UML_SIGSETJMP(&initial_jmpbuf, enable);
 507         switch(n){
 508         case INIT_JMP_NEW_THREAD:
 509                 new_thread_proc((void *) stack, new_thread_handler);
 510                 break;
 511         case INIT_JMP_REMOVE_SIGSTACK:
 512                 remove_sigstack();
 513                 break;
 514         case INIT_JMP_CALLBACK:
 515                 (*cb_proc)(cb_arg);
 516                 UML_SIGLONGJMP(cb_back, 1);
 517                 break;
 518         case INIT_JMP_HALT:
 519                 kmalloc_ok = 0;
 520                 return(0);
 521         case INIT_JMP_REBOOT:
 522                 kmalloc_ok = 0;
 523                 return(1);
 524         default:
 525                 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
 526         }
 527         UML_SIGLONGJMP(*switch_buf, 1);
 528 }
 529
 530 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 531 {
 532         sigjmp_buf here;
 533         int enable;
 534
 535         cb_proc = proc;
 536         cb_arg = arg;
 537         cb_back = &here;
 538
 539         block_signals();
 540         if(UML_SIGSETJMP(&here, enable) == 0)
 541                 UML_SIGLONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
 542         unblock_signals();
 543
 544         cb_proc = NULL;
 545         cb_arg = NULL;
 546         cb_back = NULL;
 547 }
 548
 549 void halt_skas(void)
 550 {
 551         block_signals();
 552         UML_SIGLONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
 553 }
 554
 555 void reboot_skas(void)
 556 {
 557         block_signals();
 558         UML_SIGLONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
 559 }
 560
 561 void switch_mm_skas(struct mm_id *mm_idp)
 562 {
 563         int err;
 564
 565 #warning need cpu pid in switch_mm_skas
 566         if(proc_mm){
 567                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
 568                              mm_idp->u.mm_fd);
 569                 if(err)
 570                         panic("switch_mm_skas - PTRACE_SWITCH_MM failed, "
 571                               "errno = %d\n", errno);
 572         }
 573         else userspace_pid[0] = mm_idp->u.pid;
 574 }