kernel/ptrace.c

/* */
This source file includes following definitions.
get_task
get_stack_long
put_stack_long
get_long
put_long
find_vma
read_long
write_long
sys_ptrace
syscall_trace
   1 /* ptrace.c */
   2 /* By Ross Biro 1/23/92 */
   3 /* edited by Linus Torvalds */
   4 
   5 #include <linux/head.h>
   6 #include <linux/kernel.h>
   7 #include <linux/sched.h>
   8 #include <linux/mm.h>
   9 #include <linux/errno.h>
  10 #include <linux/ptrace.h>
  11 #include <linux/user.h>
  12 
  13 #include <asm/segment.h>
  14 #include <asm/system.h>
  15 #include <linux/debugreg.h>
  16 
  17 /*
  18  * does not yet catch signals sent when the child dies.
  19  * in exit.c or in signal.c.
  20  */
  21 
  22 /* determines which flags the user has access to. */
  23 /* 1 = access 0 = no access */
  24 #define FLAG_MASK 0x00044dd5
  25 
  26 /* set's the trap flag. */
  27 #define TRAP_FLAG 0x100
  28 
  29 /*
  30  * this is the number to subtract from the top of the stack. To find
  31  * the local frame.
  32  */
  33 #define MAGICNUMBER 68
  34 
  35 /* change a pid into a task struct. */
  36 static inline struct task_struct * get_task(int pid)
     /*  */
  37 {
  38         int i;
  39 
  40         for (i = 1; i < NR_TASKS; i++) {
  41                 if (task[i] != NULL && (task[i]->pid == pid))
  42                         return task[i];
  43         }
  44         return NULL;
  45 }
  46 
  47 /*
  48  * this routine will get a word off of the processes privileged stack. 
  49  * the offset is how far from the base addr as stored in the TSS.  
  50  * this routine assumes that all the privileged stacks are in our
  51  * data space.
  52  */   
  53 static inline int get_stack_long(struct task_struct *task, int offset)
     /*  */
  54 {
  55         unsigned char *stack;
  56 
  57         stack = (unsigned char *)task->tss.esp0;
  58         stack += offset;
  59         return (*((int *)stack));
  60 }
  61 
  62 /*
  63  * this routine will put a word on the processes privileged stack. 
  64  * the offset is how far from the base addr as stored in the TSS.  
  65  * this routine assumes that all the privileged stacks are in our
  66  * data space.
  67  */
  68 static inline int put_stack_long(struct task_struct *task, int offset,
     /*  */
  69         unsigned long data)
  70 {
  71         unsigned char * stack;
  72 
  73         stack = (unsigned char *) task->tss.esp0;
  74         stack += offset;
  75         *(unsigned long *) stack = data;
  76         return 0;
  77 }
  78 
  79 /*
  80  * This routine gets a long from any process space by following the page
  81  * tables. NOTE! You should check that the long isn't on a page boundary,
  82  * and that it is in the task area before calling this: this routine does
  83  * no checking.
  84  */
  85 static unsigned long get_long(struct vm_area_struct * vma, unsigned long addr)
     /*  */
  86 {
  87         unsigned long page;
  88 
  89 repeat:
  90         page = *PAGE_DIR_OFFSET(vma->vm_task->tss.cr3, addr);
  91         if (page & PAGE_PRESENT) {
  92                 page &= PAGE_MASK;
  93                 page += PAGE_PTR(addr);
  94                 page = *((unsigned long *) page);
  95         }
  96         if (!(page & PAGE_PRESENT)) {
  97                 do_no_page(vma, addr, 0);
  98                 goto repeat;
  99         }
 100 /* this is a hack for non-kernel-mapped video buffers and similar */
 101         if (page >= high_memory)
 102                 return 0;
 103         page &= PAGE_MASK;
 104         page += addr & ~PAGE_MASK;
 105         return *(unsigned long *) page;
 106 }
 107 
 108 /*
 109  * This routine puts a long into any process space by following the page
 110  * tables. NOTE! You should check that the long isn't on a page boundary,
 111  * and that it is in the task area before calling this: this routine does
 112  * no checking.
 113  *
 114  * Now keeps R/W state of page so that a text page stays readonly
 115  * even if a debugger scribbles breakpoints into it.  -M.U-
 116  */
 117 static void put_long(struct vm_area_struct * vma, unsigned long addr,
     /*  */
 118         unsigned long data)
 119 {
 120         unsigned long page, pte = 0;
 121         int readonly = 0;
 122 
 123 repeat:
 124         page = *PAGE_DIR_OFFSET(vma->vm_task->tss.cr3, addr);
 125         if (page & PAGE_PRESENT) {
 126                 page &= PAGE_MASK;
 127                 page += PAGE_PTR(addr);
 128                 pte = page;
 129                 page = *((unsigned long *) page);
 130         }
 131         if (!(page & PAGE_PRESENT)) {
 132                 do_no_page(vma, addr, 0 /* PAGE_RW */);
 133                 goto repeat;
 134         }
 135         if (!(page & PAGE_RW)) {
 136                 if (!(page & PAGE_COW))
 137                         readonly = 1;
 138                 do_wp_page(vma, addr, PAGE_RW | PAGE_PRESENT);
 139                 goto repeat;
 140         }
 141 /* this is a hack for non-kernel-mapped video buffers and similar */
 142         if (page >= high_memory)
 143                 return;
 144 /* we're bypassing pagetables, so we have to set the dirty bit ourselves */
 145         *(unsigned long *) pte |= (PAGE_DIRTY|PAGE_COW);
 146         page &= PAGE_MASK;
 147         page += addr & ~PAGE_MASK;
 148         *(unsigned long *) page = data;
 149         if (readonly) {
 150                 *(unsigned long *) pte &=~ (PAGE_RW|PAGE_COW);
 151                 invalidate();
 152         } 
 153 }
 154 
 155 static struct vm_area_struct * find_vma(struct task_struct * tsk, unsigned long addr)
     /*  */
 156 {
 157         struct vm_area_struct * vma;
 158 
 159         addr &= PAGE_MASK;
 160         for (vma = tsk->mm->mmap ; ; vma = vma->vm_next) {
 161                 if (!vma)
 162                         return NULL;
 163                 if (vma->vm_end > addr)
 164                         break;
 165         }
 166         if (vma->vm_start <= addr)
 167                 return vma;
 168         if (!(vma->vm_flags & VM_GROWSDOWN))
 169                 return NULL;
 170         if (vma->vm_end - addr > tsk->rlim[RLIMIT_STACK].rlim_cur)
 171                 return NULL;
 172         vma->vm_offset -= vma->vm_start - addr;
 173         vma->vm_start = addr;
 174         return vma;
 175 }
 176 
 177 /*
 178  * This routine checks the page boundaries, and that the offset is
 179  * within the task area. It then calls get_long() to read a long.
 180  */
 181 static int read_long(struct task_struct * tsk, unsigned long addr,
     /*  */
 182         unsigned long * result)
 183 {
 184         struct vm_area_struct * vma = find_vma(tsk, addr);
 185 
 186         if (!vma)
 187                 return -EIO;
 188         if ((addr & ~PAGE_MASK) > PAGE_SIZE-sizeof(long)) {
 189                 unsigned long low,high;
 190                 struct vm_area_struct * vma_high = vma;
 191 
 192                 if (addr + sizeof(long) >= vma->vm_end) {
 193                         vma_high = vma->vm_next;
 194                         if (!vma_high || vma_high->vm_start != vma->vm_end)
 195                                 return -EIO;
 196                 }
 197                 low = get_long(vma, addr & ~(sizeof(long)-1));
 198                 high = get_long(vma_high, (addr+sizeof(long)) & ~(sizeof(long)-1));
 199                 switch (addr & (sizeof(long)-1)) {
 200                         case 1:
 201                                 low >>= 8;
 202                                 low |= high << 24;
 203                                 break;
 204                         case 2:
 205                                 low >>= 16;
 206                                 low |= high << 16;
 207                                 break;
 208                         case 3:
 209                                 low >>= 24;
 210                                 low |= high << 8;
 211                                 break;
 212                 }
 213                 *result = low;
 214         } else
 215                 *result = get_long(vma, addr);
 216         return 0;
 217 }
 218 
 219 /*
 220  * This routine checks the page boundaries, and that the offset is
 221  * within the task area. It then calls put_long() to write a long.
 222  */
 223 static int write_long(struct task_struct * tsk, unsigned long addr,
     /*  */
 224         unsigned long data)
 225 {
 226         struct vm_area_struct * vma = find_vma(tsk, addr);
 227 
 228         if (!vma)
 229                 return -EIO;
 230         if ((addr & ~PAGE_MASK) > PAGE_SIZE-sizeof(long)) {
 231                 unsigned long low,high;
 232                 struct vm_area_struct * vma_high = vma;
 233 
 234                 if (addr + sizeof(long) >= vma->vm_end) {
 235                         vma_high = vma->vm_next;
 236                         if (!vma_high || vma_high->vm_start != vma->vm_end)
 237                                 return -EIO;
 238                 }
 239                 low = get_long(vma, addr & ~(sizeof(long)-1));
 240                 high = get_long(vma_high, (addr+sizeof(long)) & ~(sizeof(long)-1));
 241                 switch (addr & (sizeof(long)-1)) {
 242                         case 0: /* shouldn't happen, but safety first */
 243                                 low = data;
 244                                 break;
 245                         case 1:
 246                                 low &= 0x000000ff;
 247                                 low |= data << 8;
 248                                 high &= ~0xff;
 249                                 high |= data >> 24;
 250                                 break;
 251                         case 2:
 252                                 low &= 0x0000ffff;
 253                                 low |= data << 16;
 254                                 high &= ~0xffff;
 255                                 high |= data >> 16;
 256                                 break;
 257                         case 3:
 258                                 low &= 0x00ffffff;
 259                                 low |= data << 24;
 260                                 high &= ~0xffffff;
 261                                 high |= data >> 8;
 262                                 break;
 263                 }
 264                 put_long(vma, addr & ~(sizeof(long)-1),low);
 265                 put_long(vma_high, (addr+sizeof(long)) & ~(sizeof(long)-1),high);
 266         } else
 267                 put_long(vma, addr, data);
 268         return 0;
 269 }
 270 
 271 asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
     /*  */
 272 {
 273         struct task_struct *child;
 274         struct user * dummy;
 275         int i;
 276 
 277         dummy = NULL;
 278 
 279         if (request == PTRACE_TRACEME) {
 280                 /* are we already being traced? */
 281                 if (current->flags & PF_PTRACED)
 282                         return -EPERM;
 283                 /* set the ptrace bit in the process flags. */
 284                 current->flags |= PF_PTRACED;
 285                 return 0;
 286         }
 287         if (pid == 1)           /* you may not mess with init */
 288                 return -EPERM;
 289         if (!(child = get_task(pid)))
 290                 return -ESRCH;
 291         if (request == PTRACE_ATTACH) {
 292                 if (child == current)
 293                         return -EPERM;
 294                 if ((!child->dumpable ||
 295                     (current->uid != child->euid) ||
 296                     (current->uid != child->uid) ||
 297                     (current->gid != child->egid) ||
 298                     (current->gid != child->gid)) && !suser())
 299                         return -EPERM;
 300                 /* the same process cannot be attached many times */
 301                 if (child->flags & PF_PTRACED)
 302                         return -EPERM;
 303                 child->flags |= PF_PTRACED;
 304                 if (child->p_pptr != current) {
 305                         REMOVE_LINKS(child);
 306                         child->p_pptr = current;
 307                         SET_LINKS(child);
 308                 }
 309                 send_sig(SIGSTOP, child, 1);
 310                 return 0;
 311         }
 312         if (!(child->flags & PF_PTRACED))
 313                 return -ESRCH;
 314         if (child->state != TASK_STOPPED) {
 315                 if (request != PTRACE_KILL)
 316                         return -ESRCH;
 317         }
 318         if (child->p_pptr != current)
 319                 return -ESRCH;
 320 
 321         switch (request) {
 322         /* when I and D space are separate, these will need to be fixed. */
 323                 case PTRACE_PEEKTEXT: /* read word at location addr. */ 
 324                 case PTRACE_PEEKDATA: {
 325                         unsigned long tmp;
 326                         int res;
 327 
 328                         res = read_long(child, addr, &tmp);
 329                         if (res < 0)
 330                                 return res;
 331                         res = verify_area(VERIFY_WRITE, (void *) data, sizeof(long));
 332                         if (!res)
 333                                 put_fs_long(tmp,(unsigned long *) data);
 334                         return res;
 335                 }
 336 
 337         /* read the word at location addr in the USER area. */
 338                 case PTRACE_PEEKUSR: {
 339                         unsigned long tmp;
 340                         int res;
 341 
 342                         if ((addr & 3) || addr < 0 || 
 343                             addr > sizeof(struct user) - 3)
 344                                 return -EIO;
 345 
 346                         res = verify_area(VERIFY_WRITE, (void *) data, sizeof(long));
 347                         if (res)
 348                                 return res;
 349                         tmp = 0;  /* Default return condition */
 350                         if(addr < 17*sizeof(long)) {
 351                           addr = addr >> 2; /* temporary hack. */
 352 
 353                           tmp = get_stack_long(child, sizeof(long)*addr - MAGICNUMBER);
 354                           if (addr == DS || addr == ES ||
 355                               addr == FS || addr == GS ||
 356                               addr == CS || addr == SS)
 357                             tmp &= 0xffff;
 358                         };
 359                         if(addr >= (long) &dummy->u_debugreg[0] &&
 360                            addr <= (long) &dummy->u_debugreg[7]){
 361                                 addr -= (long) &dummy->u_debugreg[0];
 362                                 addr = addr >> 2;
 363                                 tmp = child->debugreg[addr];
 364                         };
 365                         put_fs_long(tmp,(unsigned long *) data);
 366                         return 0;
 367                 }
 368 
 369       /* when I and D space are separate, this will have to be fixed. */
 370                 case PTRACE_POKETEXT: /* write the word at location addr. */
 371                 case PTRACE_POKEDATA:
 372                         return write_long(child,addr,data);
 373 
 374                 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 375                         if ((addr & 3) || addr < 0 || 
 376                             addr > sizeof(struct user) - 3)
 377                                 return -EIO;
 378 
 379                         addr = addr >> 2; /* temporary hack. */
 380 
 381                         if (addr == ORIG_EAX)
 382                                 return -EIO;
 383                         if (addr == DS || addr == ES ||
 384                             addr == FS || addr == GS ||
 385                             addr == CS || addr == SS) {
 386                                 data &= 0xffff;
 387                                 if (data && (data & 3) != 3)
 388                                         return -EIO;
 389                         }
 390                         if (addr == EFL) {   /* flags. */
 391                                 data &= FLAG_MASK;
 392                                 data |= get_stack_long(child, EFL*sizeof(long)-MAGICNUMBER)  & ~FLAG_MASK;
 393                         }
 394                   /* Do not allow the user to set the debug register for kernel
 395                      address space */
 396                   if(addr < 17){
 397                           if (put_stack_long(child, sizeof(long)*addr-MAGICNUMBER, data))
 398                                 return -EIO;
 399                         return 0;
 400                         };
 401 
 402                   /* We need to be very careful here.  We implicitly
 403                      want to modify a portion of the task_struct, and we
 404                      have to be selective about what portions we allow someone
 405                      to modify. */
 406 
 407                   addr = addr << 2;  /* Convert back again */
 408                   if(addr >= (long) &dummy->u_debugreg[0] &&
 409                      addr <= (long) &dummy->u_debugreg[7]){
 410 
 411                           if(addr == (long) &dummy->u_debugreg[4]) return -EIO;
 412                           if(addr == (long) &dummy->u_debugreg[5]) return -EIO;
 413                           if(addr < (long) &dummy->u_debugreg[4] &&
 414                              ((unsigned long) data) >= 0xbffffffd) return -EIO;
 415                           
 416                           if(addr == (long) &dummy->u_debugreg[7]) {
 417                                   data &= ~DR_CONTROL_RESERVED;
 418                                   for(i=0; i<4; i++)
 419                                           if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 420                                                   return -EIO;
 421                           };
 422 
 423                           addr -= (long) &dummy->u_debugreg;
 424                           addr = addr >> 2;
 425                           child->debugreg[addr] = data;
 426                           return 0;
 427                   };
 428                   return -EIO;
 429 
 430                 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 431                 case PTRACE_CONT: { /* restart after signal. */
 432                         long tmp;
 433 
 434                         if ((unsigned long) data > NSIG)
 435                                 return -EIO;
 436                         if (request == PTRACE_SYSCALL)
 437                                 child->flags |= PF_TRACESYS;
 438                         else
 439                                 child->flags &= ~PF_TRACESYS;
 440                         child->exit_code = data;
 441                         child->state = TASK_RUNNING;
 442         /* make sure the single step bit is not set. */
 443                         tmp = get_stack_long(child, sizeof(long)*EFL-MAGICNUMBER) & ~TRAP_FLAG;
 444                         put_stack_long(child, sizeof(long)*EFL-MAGICNUMBER,tmp);
 445                         return 0;
 446                 }
 447 
 448 /*
 449  * make the child exit.  Best I can do is send it a sigkill. 
 450  * perhaps it should be put in the status that it wants to 
 451  * exit.
 452  */
 453                 case PTRACE_KILL: {
 454                         long tmp;
 455 
 456                         child->state = TASK_RUNNING;
 457                         child->exit_code = SIGKILL;
 458         /* make sure the single step bit is not set. */
 459                         tmp = get_stack_long(child, sizeof(long)*EFL-MAGICNUMBER) & ~TRAP_FLAG;
 460                         put_stack_long(child, sizeof(long)*EFL-MAGICNUMBER,tmp);
 461                         return 0;
 462                 }
 463 
 464                 case PTRACE_SINGLESTEP: {  /* set the trap flag. */
 465                         long tmp;
 466 
 467                         if ((unsigned long) data > NSIG)
 468                                 return -EIO;
 469                         child->flags &= ~PF_TRACESYS;
 470                         tmp = get_stack_long(child, sizeof(long)*EFL-MAGICNUMBER) | TRAP_FLAG;
 471                         put_stack_long(child, sizeof(long)*EFL-MAGICNUMBER,tmp);
 472                         child->state = TASK_RUNNING;
 473                         child->exit_code = data;
 474         /* give it a chance to run. */
 475                         return 0;
 476                 }
 477 
 478                 case PTRACE_DETACH: { /* detach a process that was attached. */
 479                         long tmp;
 480 
 481                         if ((unsigned long) data > NSIG)
 482                                 return -EIO;
 483                         child->flags &= ~(PF_PTRACED|PF_TRACESYS);
 484                         child->state = TASK_RUNNING;
 485                         child->exit_code = data;
 486                         REMOVE_LINKS(child);
 487                         child->p_pptr = child->p_opptr;
 488                         SET_LINKS(child);
 489                         /* make sure the single step bit is not set. */
 490                         tmp = get_stack_long(child, sizeof(long)*EFL-MAGICNUMBER) & ~TRAP_FLAG;
 491                         put_stack_long(child, sizeof(long)*EFL-MAGICNUMBER,tmp);
 492                         return 0;
 493                 }
 494 
 495                 default:
 496                         return -EIO;
 497         }
 498 }
 499 
 500 asmlinkage void syscall_trace(void)
     /*  */
 501 {
 502         if ((current->flags & (PF_PTRACED|PF_TRACESYS))
 503                         != (PF_PTRACED|PF_TRACESYS))
 504                 return;
 505         current->exit_code = SIGTRAP;
 506         current->state = TASK_STOPPED;
 507         notify_parent(current);
 508         schedule();
 509         /*
 510          * this isn't the same as continuing with a signal, but it will do
 511          * for normal use.  strace only continues with a signal if the
 512          * stopping signal is not SIGTRAP.  -brl
 513          */
 514         if (current->exit_code)
 515                 current->signal |= (1 << (current->exit_code - 1));
 516         current->exit_code = 0;
 517 }
/* */
root/kernel/ptrace.c

DEFINITIONS
