arch/um/kernel/tt/process_kern.c

   1 /*
   2  * Copyright (C) 2002 Jeff Dike (jdike@karaya.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include "linux/sched.h"
   7 #include "linux/signal.h"
   8 #include "linux/kernel.h"
   9 #include "linux/interrupt.h"
  10 #include "linux/ptrace.h"
  11 #include "asm/system.h"
  12 #include "asm/pgalloc.h"
  13 #include "asm/ptrace.h"
  14 #include "asm/tlbflush.h"
  15 #include "irq_user.h"
  16 #include "signal_user.h"
  17 #include "kern_util.h"
  18 #include "user_util.h"
  19 #include "os.h"
  20 #include "kern.h"
  21 #include "sigcontext.h"
  22 #include "time_user.h"
  23 #include "mem_user.h"
  24 #include "tlb.h"
  25 #include "mode.h"
  26 #include "init.h"
  27 #include "tt.h"
  28
  29 void *switch_to_tt(void *prev, void *next, void *last)
  30 {
  31         struct task_struct *from, *to;
  32         unsigned long flags;
  33         int err, vtalrm, alrm, prof, cpu;
  34         char c;
  35         /* jailing and SMP are incompatible, so this doesn't need to be
  36          * made per-cpu
  37          */
  38         static int reading;
  39
  40         from = prev;
  41         to = next;
  42
  43         to->thread.prev_sched = from;
  44
  45         cpu = from->thread_info->cpu;
  46         if(cpu == 0)
  47                 forward_interrupts(to->thread.mode.tt.extern_pid);
  48 #ifdef CONFIG_SMP
  49         forward_ipi(cpu_data[cpu].ipi_pipe[0], to->thread.mode.tt.extern_pid);
  50 #endif
  51         local_irq_save(flags);
  52
  53         vtalrm = change_sig(SIGVTALRM, 0);
  54         alrm = change_sig(SIGALRM, 0);
  55         prof = change_sig(SIGPROF, 0);
  56
  57         forward_pending_sigio(to->thread.mode.tt.extern_pid);
  58
  59         c = 0;
  60         set_current(to);
  61
  62         reading = 0;
  63         err = os_write_file(to->thread.mode.tt.switch_pipe[1], &c, sizeof(c));
  64         if(err != sizeof(c))
  65                 panic("write of switch_pipe failed, err = %d", -err);
  66
  67         reading = 1;
  68         if((from->state == TASK_ZOMBIE) || (from->state == TASK_DEAD))
  69                 os_kill_process(os_getpid(), 0);
  70
  71         err = os_read_file(from->thread.mode.tt.switch_pipe[0], &c, sizeof(c));
  72         if(err != sizeof(c))
  73                 panic("read of switch_pipe failed, errno = %d", -err);
  74
  75         /* This works around a nasty race with 'jail'.  If we are switching
  76          * between two threads of a threaded app and the incoming process
  77          * runs before the outgoing process reaches the read, and it makes
  78          * it all the way out to userspace, then it will have write-protected
  79          * the outgoing process stack.  Then, when the outgoing process
  80          * returns from the write, it will segfault because it can no longer
  81          * write its own stack.  So, in order to avoid that, the incoming
  82          * thread sits in a loop yielding until 'reading' is set.  This
  83          * isn't entirely safe, since there may be a reschedule from a timer
  84          * happening between setting 'reading' and sleeping in read.  But,
  85          * it should get a whole quantum in which to reach the read and sleep,
  86          * which should be enough.
  87          */
  88
  89         if(jail){
  90                 while(!reading) sched_yield();
  91         }
  92
  93         change_sig(SIGVTALRM, vtalrm);
  94         change_sig(SIGALRM, alrm);
  95         change_sig(SIGPROF, prof);
  96
  97         arch_switch();
  98
  99         flush_tlb_all();
 100         local_irq_restore(flags);
 101
 102         return(current->thread.prev_sched);
 103 }
 104
 105 void release_thread_tt(struct task_struct *task)
 106 {
 107         int pid = task->thread.mode.tt.extern_pid;
 108
 109         if(os_getpid() != pid)
 110                 os_kill_process(pid, 0);
 111 }
 112
 113 void exit_thread_tt(void)
 114 {
 115         os_close_file(current->thread.mode.tt.switch_pipe[0]);
 116         os_close_file(current->thread.mode.tt.switch_pipe[1]);
 117 }
 118
 119 void schedule_tail(task_t *prev);
 120
 121 static void new_thread_handler(int sig)
 122 {
 123         unsigned long disable;
 124         int (*fn)(void *);
 125         void *arg;
 126
 127         fn = current->thread.request.u.thread.proc;
 128         arg = current->thread.request.u.thread.arg;
 129
 130         UPT_SC(&current->thread.regs.regs) = (void *) (&sig + 1);
 131         disable = (1 << (SIGVTALRM - 1)) | (1 << (SIGALRM - 1)) |
 132                 (1 << (SIGIO - 1)) | (1 << (SIGPROF - 1));
 133         SC_SIGMASK(UPT_SC(&current->thread.regs.regs)) &= ~disable;
 134
 135         suspend_new_thread(current->thread.mode.tt.switch_pipe[0]);
 136
 137         force_flush_all();
 138         if(current->thread.prev_sched != NULL)
 139                 schedule_tail(current->thread.prev_sched);
 140         current->thread.prev_sched = NULL;
 141
 142         init_new_thread_signals(1);
 143         enable_timer();
 144         free_page(current->thread.temp_stack);
 145         set_cmdline("(kernel thread)");
 146
 147         change_sig(SIGUSR1, 1);
 148         change_sig(SIGVTALRM, 1);
 149         change_sig(SIGPROF, 1);
 150         local_irq_enable();
 151         if(!run_kernel_thread(fn, arg, &current->thread.exec_buf))
 152                 do_exit(0);
 153 }
 154
 155 static int new_thread_proc(void *stack)
 156 {
 157         /* local_irq_disable is needed to block out signals until this thread is
 158          * properly scheduled.  Otherwise, the tracing thread will get mighty
 159          * upset about any signals that arrive before that.
 160          * This has the complication that it sets the saved signal mask in
 161          * the sigcontext to block signals.  This gets restored when this
 162          * thread (or a descendant, since they get a copy of this sigcontext)
 163          * returns to userspace.
 164          * So, this is compensated for elsewhere.
 165          * XXX There is still a small window until local_irq_disable() actually
 166          * finishes where signals are possible - shouldn't be a problem in
 167          * practice since SIGIO hasn't been forwarded here yet, and the
 168          * local_irq_disable should finish before a SIGVTALRM has time to be
 169          * delivered.
 170          */
 171
 172         local_irq_disable();
 173         init_new_thread_stack(stack, new_thread_handler);
 174         os_usr1_process(os_getpid());
 175         return(0);
 176 }
 177
 178 /* Signal masking - signals are blocked at the start of fork_tramp.  They
 179  * are re-enabled when finish_fork_handler is entered by fork_tramp hitting
 180  * itself with a SIGUSR1.  set_user_mode has to be run with SIGUSR1 off,
 181  * so it is blocked before it's called.  They are re-enabled on sigreturn
 182  * despite the fact that they were blocked when the SIGUSR1 was issued because
 183  * copy_thread copies the parent's sigcontext, including the signal mask
 184  * onto the signal frame.
 185  */
 186
 187 void finish_fork_handler(int sig)
 188 {
 189         UPT_SC(&current->thread.regs.regs) = (void *) (&sig + 1);
 190         suspend_new_thread(current->thread.mode.tt.switch_pipe[0]);
 191
 192         force_flush_all();
 193         if(current->thread.prev_sched != NULL)
 194                 schedule_tail(current->thread.prev_sched);
 195         current->thread.prev_sched = NULL;
 196
 197         enable_timer();
 198         change_sig(SIGVTALRM, 1);
 199         local_irq_enable();
 200         if(current->mm != current->parent->mm)
 201                 protect_memory(uml_reserved, high_physmem - uml_reserved, 1,
 202                                1, 0, 1);
 203         task_protections((unsigned long) current_thread);
 204
 205         free_page(current->thread.temp_stack);
 206         local_irq_disable();
 207         change_sig(SIGUSR1, 0);
 208         set_user_mode(current);
 209 }
 210
 211 int fork_tramp(void *stack)
 212 {
 213         local_irq_disable();
 214         arch_init_thread();
 215         init_new_thread_stack(stack, finish_fork_handler);
 216
 217         os_usr1_process(os_getpid());
 218         return(0);
 219 }
 220
 221 int copy_thread_tt(int nr, unsigned long clone_flags, unsigned long sp,
 222                    unsigned long stack_top, struct task_struct * p,
 223                    struct pt_regs *regs)
 224 {
 225         int (*tramp)(void *);
 226         int new_pid, err;
 227         unsigned long stack;
 228
 229         if(current->thread.forking)
 230                 tramp = fork_tramp;
 231         else {
 232                 tramp = new_thread_proc;
 233                 p->thread.request.u.thread = current->thread.request.u.thread;
 234         }
 235
 236         err = os_pipe(p->thread.mode.tt.switch_pipe, 1, 1);
 237         if(err < 0){
 238                 printk("copy_thread : pipe failed, err = %d\n", -err);
 239                 return(err);
 240         }
 241
 242         stack = alloc_stack(0, 0);
 243         if(stack == 0){
 244                 printk(KERN_ERR "copy_thread : failed to allocate "
 245                        "temporary stack\n");
 246                 return(-ENOMEM);
 247         }
 248
 249         clone_flags &= CLONE_VM;
 250         p->thread.temp_stack = stack;
 251         new_pid = start_fork_tramp((void *) p->thread.kernel_stack, stack,
 252                                    clone_flags, tramp);
 253         if(new_pid < 0){
 254                 printk(KERN_ERR "copy_thread : clone failed - errno = %d\n",
 255                        -new_pid);
 256                 return(new_pid);
 257         }
 258
 259         if(current->thread.forking){
 260                 sc_to_sc(UPT_SC(&p->thread.regs.regs),
 261                          UPT_SC(&current->thread.regs.regs));
 262                 SC_SET_SYSCALL_RETURN(UPT_SC(&p->thread.regs.regs), 0);
 263                 if(sp != 0) SC_SP(UPT_SC(&p->thread.regs.regs)) = sp;
 264         }
 265         p->thread.mode.tt.extern_pid = new_pid;
 266
 267         current->thread.request.op = OP_FORK;
 268         current->thread.request.u.fork.pid = new_pid;
 269         os_usr1_process(os_getpid());
 270         return(0);
 271 }
 272
 273 void reboot_tt(void)
 274 {
 275         current->thread.request.op = OP_REBOOT;
 276         os_usr1_process(os_getpid());
 277 }
 278
 279 void halt_tt(void)
 280 {
 281         current->thread.request.op = OP_HALT;
 282         os_usr1_process(os_getpid());
 283 }
 284
 285 void kill_off_processes_tt(void)
 286 {
 287         struct task_struct *p;
 288         int me;
 289
 290         me = os_getpid();
 291         for_each_process(p){
 292                 if(p->thread.mode.tt.extern_pid != me)
 293                         os_kill_process(p->thread.mode.tt.extern_pid, 0);
 294         }
 295         if(init_task.thread.mode.tt.extern_pid != me)
 296                 os_kill_process(init_task.thread.mode.tt.extern_pid, 0);
 297 }
 298
 299 void initial_thread_cb_tt(void (*proc)(void *), void *arg)
 300 {
 301         if(os_getpid() == tracing_pid){
 302                 (*proc)(arg);
 303         }
 304         else {
 305                 current->thread.request.op = OP_CB;
 306                 current->thread.request.u.cb.proc = proc;
 307                 current->thread.request.u.cb.arg = arg;
 308                 os_usr1_process(os_getpid());
 309         }
 310 }
 311
 312 int do_proc_op(void *t, int proc_id)
 313 {
 314         struct task_struct *task;
 315         struct thread_struct *thread;
 316         int op, pid;
 317
 318         task = t;
 319         thread = &task->thread;
 320         op = thread->request.op;
 321         switch(op){
 322         case OP_NONE:
 323         case OP_TRACE_ON:
 324                 break;
 325         case OP_EXEC:
 326                 pid = thread->request.u.exec.pid;
 327                 do_exec(thread->mode.tt.extern_pid, pid);
 328                 thread->mode.tt.extern_pid = pid;
 329                 cpu_tasks[task->thread_info->cpu].pid = pid;
 330                 break;
 331         case OP_FORK:
 332                 attach_process(thread->request.u.fork.pid);
 333                 break;
 334         case OP_CB:
 335                 (*thread->request.u.cb.proc)(thread->request.u.cb.arg);
 336                 break;
 337         case OP_REBOOT:
 338         case OP_HALT:
 339                 break;
 340         default:
 341                 tracer_panic("Bad op in do_proc_op");
 342                 break;
 343         }
 344         thread->request.op = OP_NONE;
 345         return(op);
 346 }
 347
 348 void init_idle_tt(void)
 349 {
 350         default_idle();
 351 }
 352
 353 /* Changed by jail_setup, which is a setup */
 354 int jail = 0;
 355
 356 int __init jail_setup(char *line, int *add)
 357 {
 358         int ok = 1;
 359
 360         if(jail) return(0);
 361 #ifdef CONFIG_SMP
 362         printf("'jail' may not used used in a kernel with CONFIG_SMP "
 363                "enabled\n");
 364         ok = 0;
 365 #endif
 366 #ifdef CONFIG_HOSTFS
 367         printf("'jail' may not used used in a kernel with CONFIG_HOSTFS "
 368                "enabled\n");
 369         ok = 0;
 370 #endif
 371 #ifdef CONFIG_MODULES
 372         printf("'jail' may not used used in a kernel with CONFIG_MODULES "
 373                "enabled\n");
 374         ok = 0;
 375 #endif
 376         if(!ok) exit(1);
 377
 378         /* CAP_SYS_RAWIO controls the ability to open /dev/mem and /dev/kmem.
 379          * Removing it from the bounding set eliminates the ability of anything
 380          * to acquire it, and thus read or write kernel memory.
 381          */
 382         cap_lower(cap_bset, CAP_SYS_RAWIO);
 383         jail = 1;
 384         return(0);
 385 }
 386
 387 __uml_setup("jail", jail_setup,
 388 "jail\n"
 389 "    Enables the protection of kernel memory from processes.\n\n"
 390 );
 391
 392 static void mprotect_kernel_mem(int w)
 393 {
 394         unsigned long start, end;
 395         int pages;
 396
 397         if(!jail || (current == &init_task)) return;
 398
 399         pages = (1 << CONFIG_KERNEL_STACK_ORDER);
 400
 401         start = (unsigned long) current_thread + PAGE_SIZE;
 402         end = (unsigned long) current_thread + PAGE_SIZE * pages;
 403         protect_memory(uml_reserved, start - uml_reserved, 1, w, 1, 1);
 404         protect_memory(end, high_physmem - end, 1, w, 1, 1);
 405
 406         start = (unsigned long) UML_ROUND_DOWN(&_stext);
 407         end = (unsigned long) UML_ROUND_UP(&_etext);
 408         protect_memory(start, end - start, 1, w, 1, 1);
 409
 410         start = (unsigned long) UML_ROUND_DOWN(&_unprotected_end);
 411         end = (unsigned long) UML_ROUND_UP(&_edata);
 412         protect_memory(start, end - start, 1, w, 1, 1);
 413
 414         start = (unsigned long) UML_ROUND_DOWN(&__bss_start);
 415         end = (unsigned long) UML_ROUND_UP(brk_start);
 416         protect_memory(start, end - start, 1, w, 1, 1);
 417
 418         mprotect_kernel_vm(w);
 419 }
 420
 421 void unprotect_kernel_mem(void)
 422 {
 423         mprotect_kernel_mem(1);
 424 }
 425
 426 void protect_kernel_mem(void)
 427 {
 428         mprotect_kernel_mem(0);
 429 }
 430
 431 extern void start_kernel(void);
 432
 433 static int start_kernel_proc(void *unused)
 434 {
 435         int pid;
 436
 437         block_signals();
 438         pid = os_getpid();
 439
 440         cpu_tasks[0].pid = pid;
 441         cpu_tasks[0].task = current;
 442 #ifdef CONFIG_SMP
 443         cpu_online_map = cpumask_of_cpu(0);
 444 #endif
 445         if(debug) os_stop_process(pid);
 446         start_kernel();
 447         return(0);
 448 }
 449
 450 void set_tracing(void *task, int tracing)
 451 {
 452         ((struct task_struct *) task)->thread.mode.tt.tracing = tracing;
 453 }
 454
 455 int is_tracing(void *t)
 456 {
 457         return (((struct task_struct *) t)->thread.mode.tt.tracing);
 458 }
 459
 460 int set_user_mode(void *t)
 461 {
 462         struct task_struct *task;
 463
 464         task = t ? t : current;
 465         if(task->thread.mode.tt.tracing)
 466                 return(1);
 467         task->thread.request.op = OP_TRACE_ON;
 468         os_usr1_process(os_getpid());
 469         return(0);
 470 }
 471
 472 void set_init_pid(int pid)
 473 {
 474         int err;
 475
 476         init_task.thread.mode.tt.extern_pid = pid;
 477         err = os_pipe(init_task.thread.mode.tt.switch_pipe, 1, 1);
 478         if(err)
 479                 panic("Can't create switch pipe for init_task, errno = %d",
 480                       -err);
 481 }
 482
 483 int singlestepping_tt(void *t)
 484 {
 485         struct task_struct *task = t;
 486
 487         if(task->thread.mode.tt.singlestep_syscall)
 488                 return(0);
 489         return(task->ptrace & PT_DTRACE);
 490 }
 491
 492 void clear_singlestep(void *t)
 493 {
 494         struct task_struct *task = t;
 495
 496         task->ptrace &= ~PT_DTRACE;
 497 }
 498
 499 int start_uml_tt(void)
 500 {
 501         void *sp;
 502         int pages;
 503
 504         pages = (1 << CONFIG_KERNEL_STACK_ORDER) - 2;
 505         sp = (void *) init_task.thread.kernel_stack + pages * PAGE_SIZE -
 506                 sizeof(unsigned long);
 507         return(tracer(start_kernel_proc, sp));
 508 }
 509
 510 int external_pid_tt(struct task_struct *task)
 511 {
 512         return(task->thread.mode.tt.extern_pid);
 513 }
 514
 515 int thread_pid_tt(struct task_struct *task)
 516 {
 517         return(task->thread.mode.tt.extern_pid);
 518 }
 519
 520 int is_valid_pid(int pid)
 521 {
 522         struct task_struct *task;
 523
 524         read_lock(&tasklist_lock);
 525         for_each_process(task){
 526                 if(task->thread.mode.tt.extern_pid == pid){
 527                         read_unlock(&tasklist_lock);
 528                         return(1);
 529                 }
 530         }
 531         read_unlock(&tasklist_lock);
 532         return(0);
 533 }
 534
 535 /*
 536  * Overrides for Emacs so that we follow Linus's tabbing style.
 537  * Emacs will notice this stuff at the end of the file and automatically
 538  * adjust the settings for this buffer only.  This must remain at the end
 539  * of the file.
 540  * ---------------------------------------------------------------------------
 541  * Local variables:
 542  * c-file-style: "linux"
 543  * End:
 544  */