arch/i386/kernel/ptrace.c

/* */
This source file includes following definitions.
get_task
get_stack_long
put_stack_long
get_long
put_long
find_extend_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 #include <linux/debugreg.h>
  13 
  14 #include <asm/segment.h>
  15 #include <asm/pgtable.h>
  16 #include <asm/system.h>
  17 
  18 /*
  19  * does not yet catch signals sent when the child dies.
  20  * in exit.c or in signal.c.
  21  */
  22 
  23 /* determines which flags the user has access to. */
  24 /* 1 = access 0 = no access */
  25 #define FLAG_MASK 0x00044dd5
  26 
  27 /* set's the trap flag. */
  28 #define TRAP_FLAG 0x100
  29 
  30 /*
  31  * this is the number to subtract from the top of the stack. To find
  32  * the local frame.
  33  */
  34 #define MAGICNUMBER 68
  35 
  36 /* change a pid into a task struct. */
  37 static inline struct task_struct * get_task(int pid)
     /*  */
  38 {
  39         int i;
  40 
  41         for (i = 1; i < NR_TASKS; i++) {
  42                 if (task[i] != NULL && (task[i]->pid == pid))
  43                         return task[i];
  44         }
  45         return NULL;
  46 }
  47 
  48 /*
  49  * this routine will get a word off of the processes privileged stack. 
  50  * the offset is how far from the base addr as stored in the TSS.  
  51  * this routine assumes that all the privileged stacks are in our
  52  * data space.
  53  */   
  54 static inline int get_stack_long(struct task_struct *task, int offset)
     /*  */
  55 {
  56         unsigned char *stack;
  57 
  58         stack = (unsigned char *)task->tss.esp0;
  59         stack += offset;
  60         return (*((int *)stack));
  61 }
  62 
  63 /*
  64  * this routine will put a word on the processes privileged stack. 
  65  * the offset is how far from the base addr as stored in the TSS.  
  66  * this routine assumes that all the privileged stacks are in our
  67  * data space.
  68  */
  69 static inline int put_stack_long(struct task_struct *task, int offset,
     /*  */
  70         unsigned long data)
  71 {
  72         unsigned char * stack;
  73 
  74         stack = (unsigned char *) task->tss.esp0;
  75         stack += offset;
  76         *(unsigned long *) stack = data;
  77         return 0;
  78 }
  79 
  80 /*
  81  * This routine gets a long from any process space by following the page
  82  * tables. NOTE! You should check that the long isn't on a page boundary,
  83  * and that it is in the task area before calling this: this routine does
  84  * no checking.
  85  */
  86 static unsigned long get_long(struct vm_area_struct * vma, unsigned long addr)
     /*  */
  87 {
  88         pgd_t * pgdir;
  89         pte_t * pgtable;
  90         unsigned long page;
  91 
  92 repeat:
  93         pgdir = PAGE_DIR_OFFSET(vma->vm_task, addr);
  94         if (pgd_none(*pgdir)) {
  95                 do_no_page(vma, addr, 0);
  96                 goto repeat;
  97         }
  98         if (pgd_bad(*pgdir)) {
  99                 printk("ptrace: bad page directory %08lx\n", pgd_val(*pgdir));
 100                 pgd_clear(pgdir);
 101                 return 0;
 102         }
 103         pgtable = (pte_t *) (PAGE_PTR(addr) + pgd_page(*pgdir));
 104         if (!pte_present(*pgtable)) {
 105                 do_no_page(vma, addr, 0);
 106                 goto repeat;
 107         }
 108         page = pte_page(*pgtable);
 109 /* this is a hack for non-kernel-mapped video buffers and similar */
 110         if (page >= high_memory)
 111                 return 0;
 112         page += addr & ~PAGE_MASK;
 113         return *(unsigned long *) page;
 114 }
 115 
 116 /*
 117  * This routine puts a long into any process space by following the page
 118  * tables. NOTE! You should check that the long isn't on a page boundary,
 119  * and that it is in the task area before calling this: this routine does
 120  * no checking.
 121  *
 122  * Now keeps R/W state of page so that a text page stays readonly
 123  * even if a debugger scribbles breakpoints into it.  -M.U-
 124  */
 125 static void put_long(struct vm_area_struct * vma, unsigned long addr,
     /*  */
 126         unsigned long data)
 127 {
 128         pgd_t *pgdir;
 129         pte_t *pgtable;
 130         unsigned long page;
 131 
 132 repeat:
 133         pgdir = PAGE_DIR_OFFSET(vma->vm_task, addr);
 134         if (!pgd_present(*pgdir)) {
 135                 do_no_page(vma, addr, 1);
 136                 goto repeat;
 137         }
 138         if (pgd_bad(*pgdir)) {
 139                 printk("ptrace: bad page directory %08lx\n", pgd_val(*pgdir));
 140                 pgd_clear(pgdir);
 141                 return;
 142         }
 143         pgtable = (pte_t *) (PAGE_PTR(addr) + pgd_page(*pgdir));
 144         if (!pte_present(*pgtable)) {
 145                 do_no_page(vma, addr, 1);
 146                 goto repeat;
 147         }
 148         page = pte_page(*pgtable);
 149         if (!pte_write(*pgtable)) {
 150                 do_wp_page(vma, addr, 1);
 151                 goto repeat;
 152         }
 153 /* this is a hack for non-kernel-mapped video buffers and similar */
 154         if (page < high_memory) {
 155                 page += addr & ~PAGE_MASK;
 156                 *(unsigned long *) page = data;
 157         }
 158 /* we're bypassing pagetables, so we have to set the dirty bit ourselves */
 159 /* this should also re-instate whatever read-only mode there was before */
 160         *pgtable = pte_mkdirty(mk_pte(page, vma->vm_page_prot));
 161         invalidate();
 162 }
 163 
 164 static struct vm_area_struct * find_extend_vma(struct task_struct * tsk, unsigned long addr)
     /*  */
 165 {
 166         struct vm_area_struct * vma;
 167 
 168         addr &= PAGE_MASK;
 169         vma = find_vma(tsk,addr);
 170         if (!vma)
 171                 return NULL;
 172         if (vma->vm_start <= addr)
 173                 return vma;
 174         if (!(vma->vm_flags & VM_GROWSDOWN))
 175                 return NULL;
 176         if (vma->vm_end - addr > tsk->rlim[RLIMIT_STACK].rlim_cur)
 177                 return NULL;
 178         vma->vm_offset -= vma->vm_start - addr;
 179         vma->vm_start = addr;
 180         return vma;
 181 }
 182 
 183 /*
 184  * This routine checks the page boundaries, and that the offset is
 185  * within the task area. It then calls get_long() to read a long.
 186  */
 187 static int read_long(struct task_struct * tsk, unsigned long addr,
     /*  */
 188         unsigned long * result)
 189 {
 190         struct vm_area_struct * vma = find_extend_vma(tsk, addr);
 191 
 192         if (!vma)
 193                 return -EIO;
 194         if ((addr & ~PAGE_MASK) > PAGE_SIZE-sizeof(long)) {
 195                 unsigned long low,high;
 196                 struct vm_area_struct * vma_high = vma;
 197 
 198                 if (addr + sizeof(long) >= vma->vm_end) {
 199                         vma_high = vma->vm_next;
 200                         if (!vma_high || vma_high->vm_start != vma->vm_end)
 201                                 return -EIO;
 202                 }
 203                 low = get_long(vma, addr & ~(sizeof(long)-1));
 204                 high = get_long(vma_high, (addr+sizeof(long)) & ~(sizeof(long)-1));
 205                 switch (addr & (sizeof(long)-1)) {
 206                         case 1:
 207                                 low >>= 8;
 208                                 low |= high << 24;
 209                                 break;
 210                         case 2:
 211                                 low >>= 16;
 212                                 low |= high << 16;
 213                                 break;
 214                         case 3:
 215                                 low >>= 24;
 216                                 low |= high << 8;
 217                                 break;
 218                 }
 219                 *result = low;
 220         } else
 221                 *result = get_long(vma, addr);
 222         return 0;
 223 }
 224 
 225 /*
 226  * This routine checks the page boundaries, and that the offset is
 227  * within the task area. It then calls put_long() to write a long.
 228  */
 229 static int write_long(struct task_struct * tsk, unsigned long addr,
     /*  */
 230         unsigned long data)
 231 {
 232         struct vm_area_struct * vma = find_extend_vma(tsk, addr);
 233 
 234         if (!vma)
 235                 return -EIO;
 236         if ((addr & ~PAGE_MASK) > PAGE_SIZE-sizeof(long)) {
 237                 unsigned long low,high;
 238                 struct vm_area_struct * vma_high = vma;
 239 
 240                 if (addr + sizeof(long) >= vma->vm_end) {
 241                         vma_high = vma->vm_next;
 242                         if (!vma_high || vma_high->vm_start != vma->vm_end)
 243                                 return -EIO;
 244                 }
 245                 low = get_long(vma, addr & ~(sizeof(long)-1));
 246                 high = get_long(vma_high, (addr+sizeof(long)) & ~(sizeof(long)-1));
 247                 switch (addr & (sizeof(long)-1)) {
 248                         case 0: /* shouldn't happen, but safety first */
 249                                 low = data;
 250                                 break;
 251                         case 1:
 252                                 low &= 0x000000ff;
 253                                 low |= data << 8;
 254                                 high &= ~0xff;
 255                                 high |= data >> 24;
 256                                 break;
 257                         case 2:
 258                                 low &= 0x0000ffff;
 259                                 low |= data << 16;
 260                                 high &= ~0xffff;
 261                                 high |= data >> 16;
 262                                 break;
 263                         case 3:
 264                                 low &= 0x00ffffff;
 265                                 low |= data << 24;
 266                                 high &= ~0xffffff;
 267                                 high |= data >> 8;
 268                                 break;
 269                 }
 270                 put_long(vma, addr & ~(sizeof(long)-1),low);
 271                 put_long(vma_high, (addr+sizeof(long)) & ~(sizeof(long)-1),high);
 272         } else
 273                 put_long(vma, addr, data);
 274         return 0;
 275 }
 276 
 277 asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
     /*  */
 278 {
 279         struct task_struct *child;
 280         struct user * dummy;
 281         int i;
 282 
 283         dummy = NULL;
 284 
 285         if (request == PTRACE_TRACEME) {
 286                 /* are we already being traced? */
 287                 if (current->flags & PF_PTRACED)
 288                         return -EPERM;
 289                 /* set the ptrace bit in the process flags. */
 290                 current->flags |= PF_PTRACED;
 291                 return 0;
 292         }
 293         if (pid == 1)           /* you may not mess with init */
 294                 return -EPERM;
 295         if (!(child = get_task(pid)))
 296                 return -ESRCH;
 297         if (request == PTRACE_ATTACH) {
 298                 if (child == current)
 299                         return -EPERM;
 300                 if ((!child->dumpable ||
 301                     (current->uid != child->euid) ||
 302                     (current->uid != child->uid) ||
 303                     (current->gid != child->egid) ||
 304                     (current->gid != child->gid)) && !suser())
 305                         return -EPERM;
 306                 /* the same process cannot be attached many times */
 307                 if (child->flags & PF_PTRACED)
 308                         return -EPERM;
 309                 child->flags |= PF_PTRACED;
 310                 if (child->p_pptr != current) {
 311                         REMOVE_LINKS(child);
 312                         child->p_pptr = current;
 313                         SET_LINKS(child);
 314                 }
 315                 send_sig(SIGSTOP, child, 1);
 316                 return 0;
 317         }
 318         if (!(child->flags & PF_PTRACED))
 319                 return -ESRCH;
 320         if (child->state != TASK_STOPPED) {
 321                 if (request != PTRACE_KILL)
 322                         return -ESRCH;
 323         }
 324         if (child->p_pptr != current)
 325                 return -ESRCH;
 326 
 327         switch (request) {
 328         /* when I and D space are separate, these will need to be fixed. */
 329                 case PTRACE_PEEKTEXT: /* read word at location addr. */ 
 330                 case PTRACE_PEEKDATA: {
 331                         unsigned long tmp;
 332                         int res;
 333 
 334                         res = read_long(child, addr, &tmp);
 335                         if (res < 0)
 336                                 return res;
 337                         res = verify_area(VERIFY_WRITE, (void *) data, sizeof(long));
 338                         if (!res)
 339                                 put_fs_long(tmp,(unsigned long *) data);
 340                         return res;
 341                 }
 342 
 343         /* read the word at location addr in the USER area. */
 344                 case PTRACE_PEEKUSR: {
 345                         unsigned long tmp;
 346                         int res;
 347 
 348                         if ((addr & 3) || addr < 0 || 
 349                             addr > sizeof(struct user) - 3)
 350                                 return -EIO;
 351 
 352                         res = verify_area(VERIFY_WRITE, (void *) data, sizeof(long));
 353                         if (res)
 354                                 return res;
 355                         tmp = 0;  /* Default return condition */
 356                         if(addr < 17*sizeof(long)) {
 357                           addr = addr >> 2; /* temporary hack. */
 358 
 359                           tmp = get_stack_long(child, sizeof(long)*addr - MAGICNUMBER);
 360                           if (addr == DS || addr == ES ||
 361                               addr == FS || addr == GS ||
 362                               addr == CS || addr == SS)
 363                             tmp &= 0xffff;
 364                         };
 365                         if(addr >= (long) &dummy->u_debugreg[0] &&
 366                            addr <= (long) &dummy->u_debugreg[7]){
 367                                 addr -= (long) &dummy->u_debugreg[0];
 368                                 addr = addr >> 2;
 369                                 tmp = child->debugreg[addr];
 370                         };
 371                         put_fs_long(tmp,(unsigned long *) data);
 372                         return 0;
 373                 }
 374 
 375       /* when I and D space are separate, this will have to be fixed. */
 376                 case PTRACE_POKETEXT: /* write the word at location addr. */
 377                 case PTRACE_POKEDATA:
 378                         return write_long(child,addr,data);
 379 
 380                 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 381                         if ((addr & 3) || addr < 0 || 
 382                             addr > sizeof(struct user) - 3)
 383                                 return -EIO;
 384 
 385                         addr = addr >> 2; /* temporary hack. */
 386 
 387                         if (addr == ORIG_EAX)
 388                                 return -EIO;
 389                         if (addr == DS || addr == ES ||
 390                             addr == FS || addr == GS ||
 391                             addr == CS || addr == SS) {
 392                                 data &= 0xffff;
 393                                 if (data && (data & 3) != 3)
 394                                         return -EIO;
 395                         }
 396                         if (addr == EFL) {   /* flags. */
 397                                 data &= FLAG_MASK;
 398                                 data |= get_stack_long(child, EFL*sizeof(long)-MAGICNUMBER)  & ~FLAG_MASK;
 399                         }
 400                   /* Do not allow the user to set the debug register for kernel
 401                      address space */
 402                   if(addr < 17){
 403                           if (put_stack_long(child, sizeof(long)*addr-MAGICNUMBER, data))
 404                                 return -EIO;
 405                         return 0;
 406                         };
 407 
 408                   /* We need to be very careful here.  We implicitly
 409                      want to modify a portion of the task_struct, and we
 410                      have to be selective about what portions we allow someone
 411                      to modify. */
 412 
 413                   addr = addr << 2;  /* Convert back again */
 414                   if(addr >= (long) &dummy->u_debugreg[0] &&
 415                      addr <= (long) &dummy->u_debugreg[7]){
 416 
 417                           if(addr == (long) &dummy->u_debugreg[4]) return -EIO;
 418                           if(addr == (long) &dummy->u_debugreg[5]) return -EIO;
 419                           if(addr < (long) &dummy->u_debugreg[4] &&
 420                              ((unsigned long) data) >= 0xbffffffd) return -EIO;
 421                           
 422                           if(addr == (long) &dummy->u_debugreg[7]) {
 423                                   data &= ~DR_CONTROL_RESERVED;
 424                                   for(i=0; i<4; i++)
 425                                           if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 426                                                   return -EIO;
 427                           };
 428 
 429                           addr -= (long) &dummy->u_debugreg;
 430                           addr = addr >> 2;
 431                           child->debugreg[addr] = data;
 432                           return 0;
 433                   };
 434                   return -EIO;
 435 
 436                 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
 437                 case PTRACE_CONT: { /* restart after signal. */
 438                         long tmp;
 439 
 440                         if ((unsigned long) data > NSIG)
 441                                 return -EIO;
 442                         if (request == PTRACE_SYSCALL)
 443                                 child->flags |= PF_TRACESYS;
 444                         else
 445                                 child->flags &= ~PF_TRACESYS;
 446                         child->exit_code = data;
 447                         child->state = TASK_RUNNING;
 448         /* make sure the single step bit is not set. */
 449                         tmp = get_stack_long(child, sizeof(long)*EFL-MAGICNUMBER) & ~TRAP_FLAG;
 450                         put_stack_long(child, sizeof(long)*EFL-MAGICNUMBER,tmp);
 451                         return 0;
 452                 }
 453 
 454 /*
 455  * make the child exit.  Best I can do is send it a sigkill. 
 456  * perhaps it should be put in the status that it wants to 
 457  * exit.
 458  */
 459                 case PTRACE_KILL: {
 460                         long tmp;
 461 
 462                         child->state = TASK_RUNNING;
 463                         child->exit_code = SIGKILL;
 464         /* make sure the single step bit is not set. */
 465                         tmp = get_stack_long(child, sizeof(long)*EFL-MAGICNUMBER) & ~TRAP_FLAG;
 466                         put_stack_long(child, sizeof(long)*EFL-MAGICNUMBER,tmp);
 467                         return 0;
 468                 }
 469 
 470                 case PTRACE_SINGLESTEP: {  /* set the trap flag. */
 471                         long tmp;
 472 
 473                         if ((unsigned long) data > NSIG)
 474                                 return -EIO;
 475                         child->flags &= ~PF_TRACESYS;
 476                         tmp = get_stack_long(child, sizeof(long)*EFL-MAGICNUMBER) | TRAP_FLAG;
 477                         put_stack_long(child, sizeof(long)*EFL-MAGICNUMBER,tmp);
 478                         child->state = TASK_RUNNING;
 479                         child->exit_code = data;
 480         /* give it a chance to run. */
 481                         return 0;
 482                 }
 483 
 484                 case PTRACE_DETACH: { /* detach a process that was attached. */
 485                         long tmp;
 486 
 487                         if ((unsigned long) data > NSIG)
 488                                 return -EIO;
 489                         child->flags &= ~(PF_PTRACED|PF_TRACESYS);
 490                         child->state = TASK_RUNNING;
 491                         child->exit_code = data;
 492                         REMOVE_LINKS(child);
 493                         child->p_pptr = child->p_opptr;
 494                         SET_LINKS(child);
 495                         /* make sure the single step bit is not set. */
 496                         tmp = get_stack_long(child, sizeof(long)*EFL-MAGICNUMBER) & ~TRAP_FLAG;
 497                         put_stack_long(child, sizeof(long)*EFL-MAGICNUMBER,tmp);
 498                         return 0;
 499                 }
 500 
 501                 default:
 502                         return -EIO;
 503         }
 504 }
 505 
 506 asmlinkage void syscall_trace(void)
     /*  */
 507 {
 508         if ((current->flags & (PF_PTRACED|PF_TRACESYS))
 509                         != (PF_PTRACED|PF_TRACESYS))
 510                 return;
 511         current->exit_code = SIGTRAP;
 512         current->state = TASK_STOPPED;
 513         notify_parent(current);
 514         schedule();
 515         /*
 516          * this isn't the same as continuing with a signal, but it will do
 517          * for normal use.  strace only continues with a signal if the
 518          * stopping signal is not SIGTRAP.  -brl
 519          */
 520         if (current->exit_code)
 521                 current->signal |= (1 << (current->exit_code - 1));
 522         current->exit_code = 0;
 523 }
/* */
root/arch/i386/kernel/ptrace.c

DEFINITIONS
