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