fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/config.h>
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/mman.h>
  29 #include <linux/a.out.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/module.h>
  42 #include <linux/namei.h>
  43 #include <linux/proc_fs.h>
  44 #include <linux/ptrace.h>
  45 #include <linux/mount.h>
  46 #include <linux/security.h>
  47 #include <linux/syscalls.h>
  48 #include <linux/rmap-locking.h>
  49
  50 #include <asm/uaccess.h>
  51 #include <asm/pgalloc.h>
  52 #include <asm/mmu_context.h>
  53
  54 #ifdef CONFIG_KMOD
  55 #include <linux/kmod.h>
  56 #endif
  57
  58 int core_uses_pid;
  59 char core_pattern[65] = "core";
  60 /* The maximal length of core_pattern is also specified in sysctl.c */
  61
  62 static struct linux_binfmt *formats;
  63 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  64
  65 int register_binfmt(struct linux_binfmt * fmt)
  66 {
  67         struct linux_binfmt ** tmp = &formats;
  68
  69         if (!fmt)
  70                 return -EINVAL;
  71         if (fmt->next)
  72                 return -EBUSY;
  73         write_lock(&binfmt_lock);
  74         while (*tmp) {
  75                 if (fmt == *tmp) {
  76                         write_unlock(&binfmt_lock);
  77                         return -EBUSY;
  78                 }
  79                 tmp = &(*tmp)->next;
  80         }
  81         fmt->next = formats;
  82         formats = fmt;
  83         write_unlock(&binfmt_lock);
  84         return 0;
  85 }
  86
  87 EXPORT_SYMBOL(register_binfmt);
  88
  89 int unregister_binfmt(struct linux_binfmt * fmt)
  90 {
  91         struct linux_binfmt ** tmp = &formats;
  92
  93         write_lock(&binfmt_lock);
  94         while (*tmp) {
  95                 if (fmt == *tmp) {
  96                         *tmp = fmt->next;
  97                         write_unlock(&binfmt_lock);
  98                         return 0;
  99                 }
 100                 tmp = &(*tmp)->next;
 101         }
 102         write_unlock(&binfmt_lock);
 103         return -EINVAL;
 104 }
 105
 106 EXPORT_SYMBOL(unregister_binfmt);
 107
 108 static inline void put_binfmt(struct linux_binfmt * fmt)
 109 {
 110         module_put(fmt->module);
 111 }
 112
 113 /*
 114  * Note that a shared library must be both readable and executable due to
 115  * security reasons.
 116  *
 117  * Also note that we take the address to load from from the file itself.
 118  */
 119 asmlinkage long sys_uselib(const char __user * library)
 120 {
 121         struct file * file;
 122         struct nameidata nd;
 123         int error;
 124
 125         nd.intent.open.flags = FMODE_READ;
 126         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 127         if (error)
 128                 goto out;
 129
 130         error = -EINVAL;
 131         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 132                 goto exit;
 133
 134         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 135         if (error)
 136                 goto exit;
 137
 138         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 139         error = PTR_ERR(file);
 140         if (IS_ERR(file))
 141                 goto out;
 142
 143         error = -ENOEXEC;
 144         if(file->f_op) {
 145                 struct linux_binfmt * fmt;
 146
 147                 read_lock(&binfmt_lock);
 148                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 149                         if (!fmt->load_shlib)
 150                                 continue;
 151                         if (!try_module_get(fmt->module))
 152                                 continue;
 153                         read_unlock(&binfmt_lock);
 154                         error = fmt->load_shlib(file);
 155                         read_lock(&binfmt_lock);
 156                         put_binfmt(fmt);
 157                         if (error != -ENOEXEC)
 158                                 break;
 159                 }
 160                 read_unlock(&binfmt_lock);
 161         }
 162         fput(file);
 163 out:
 164         return error;
 165 exit:
 166         path_release(&nd);
 167         goto out;
 168 }
 169
 170 /*
 171  * count() counts the number of strings in array ARGV.
 172  */
 173 static int count(char __user * __user * argv, int max)
 174 {
 175         int i = 0;
 176
 177         if (argv != NULL) {
 178                 for (;;) {
 179                         char __user * p;
 180
 181                         if (get_user(p, argv))
 182                                 return -EFAULT;
 183                         if (!p)
 184                                 break;
 185                         argv++;
 186                         if(++i > max)
 187                                 return -E2BIG;
 188                 }
 189         }
 190         return i;
 191 }
 192
 193 /*
 194  * 'copy_strings()' copies argument/environment strings from user
 195  * memory to free pages in kernel mem. These are in a format ready
 196  * to be put directly into the top of new user memory.
 197  */
 198 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 199 {
 200         struct page *kmapped_page = NULL;
 201         char *kaddr = NULL;
 202         int ret;
 203
 204         while (argc-- > 0) {
 205                 char __user *str;
 206                 int len;
 207                 unsigned long pos;
 208
 209                 if (get_user(str, argv+argc) ||
 210                                 !(len = strnlen_user(str, bprm->p))) {
 211                         ret = -EFAULT;
 212                         goto out;
 213                 }
 214
 215                 if (bprm->p < len)  {
 216                         ret = -E2BIG;
 217                         goto out;
 218                 }
 219
 220                 bprm->p -= len;
 221                 /* XXX: add architecture specific overflow check here. */
 222                 pos = bprm->p;
 223
 224                 while (len > 0) {
 225                         int i, new, err;
 226                         int offset, bytes_to_copy;
 227                         struct page *page;
 228
 229                         offset = pos % PAGE_SIZE;
 230                         i = pos/PAGE_SIZE;
 231                         page = bprm->page[i];
 232                         new = 0;
 233                         if (!page) {
 234                                 page = alloc_page(GFP_HIGHUSER);
 235                                 bprm->page[i] = page;
 236                                 if (!page) {
 237                                         ret = -ENOMEM;
 238                                         goto out;
 239                                 }
 240                                 new = 1;
 241                         }
 242
 243                         if (page != kmapped_page) {
 244                                 if (kmapped_page)
 245                                         kunmap(kmapped_page);
 246                                 kmapped_page = page;
 247                                 kaddr = kmap(kmapped_page);
 248                         }
 249                         if (new && offset)
 250                                 memset(kaddr, 0, offset);
 251                         bytes_to_copy = PAGE_SIZE - offset;
 252                         if (bytes_to_copy > len) {
 253                                 bytes_to_copy = len;
 254                                 if (new)
 255                                         memset(kaddr+offset+len, 0,
 256                                                 PAGE_SIZE-offset-len);
 257                         }
 258                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 259                         if (err) {
 260                                 ret = -EFAULT;
 261                                 goto out;
 262                         }
 263
 264                         pos += bytes_to_copy;
 265                         str += bytes_to_copy;
 266                         len -= bytes_to_copy;
 267                 }
 268         }
 269         ret = 0;
 270 out:
 271         if (kmapped_page)
 272                 kunmap(kmapped_page);
 273         return ret;
 274 }
 275
 276 /*
 277  * Like copy_strings, but get argv and its values from kernel memory.
 278  */
 279 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 280 {
 281         int r;
 282         mm_segment_t oldfs = get_fs();
 283         set_fs(KERNEL_DS);
 284         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 285         set_fs(oldfs);
 286         return r;
 287 }
 288
 289 EXPORT_SYMBOL(copy_strings_kernel);
 290
 291 #ifdef CONFIG_MMU
 292 /*
 293  * This routine is used to map in a page into an address space: needed by
 294  * execve() for the initial stack and environment pages.
 295  *
 296  * tsk->mmap_sem is held for writing.
 297  */
 298 void put_dirty_page(struct task_struct *tsk, struct page *page,
 299                         unsigned long address, pgprot_t prot)
 300 {
 301         pgd_t * pgd;
 302         pmd_t * pmd;
 303         pte_t * pte;
 304         struct pte_chain *pte_chain;
 305
 306         if (page_count(page) != 1)
 307                 printk(KERN_ERR "mem_map disagrees with %p at %08lx\n",
 308                                 page, address);
 309
 310         pgd = pgd_offset(tsk->mm, address);
 311         pte_chain = pte_chain_alloc(GFP_KERNEL);
 312         if (!pte_chain)
 313                 goto out_sig;
 314         spin_lock(&tsk->mm->page_table_lock);
 315         pmd = pmd_alloc(tsk->mm, pgd, address);
 316         if (!pmd)
 317                 goto out;
 318         pte = pte_alloc_map(tsk->mm, pmd, address);
 319         if (!pte)
 320                 goto out;
 321         if (!pte_none(*pte)) {
 322                 pte_unmap(pte);
 323                 goto out;
 324         }
 325         lru_cache_add_active(page);
 326         flush_dcache_page(page);
 327         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(page, prot))));
 328         pte_chain = page_add_rmap(page, pte, pte_chain);
 329         pte_unmap(pte);
 330         tsk->mm->rss++;
 331         spin_unlock(&tsk->mm->page_table_lock);
 332
 333         /* no need for flush_tlb */
 334         pte_chain_free(pte_chain);
 335         return;
 336 out:
 337         spin_unlock(&tsk->mm->page_table_lock);
 338 out_sig:
 339         __free_page(page);
 340         force_sig(SIGKILL, tsk);
 341         pte_chain_free(pte_chain);
 342         return;
 343 }
 344
 345 int setup_arg_pages(struct linux_binprm *bprm)
 346 {
 347         unsigned long stack_base;
 348         struct vm_area_struct *mpnt;
 349         struct mm_struct *mm = current->mm;
 350         int i;
 351         long arg_size;
 352
 353 #ifdef CONFIG_STACK_GROWSUP
 354         /* Move the argument and environment strings to the bottom of the
 355          * stack space.
 356          */
 357         int offset, j;
 358         char *to, *from;
 359
 360         /* Start by shifting all the pages down */
 361         i = 0;
 362         for (j = 0; j < MAX_ARG_PAGES; j++) {
 363                 struct page *page = bprm->page[j];
 364                 if (!page)
 365                         continue;
 366                 bprm->page[i++] = page;
 367         }
 368
 369         /* Now move them within their pages */
 370         offset = bprm->p % PAGE_SIZE;
 371         to = kmap(bprm->page[0]);
 372         for (j = 1; j < i; j++) {
 373                 memmove(to, to + offset, PAGE_SIZE - offset);
 374                 from = kmap(bprm->page[j]);
 375                 memcpy(to + PAGE_SIZE - offset, from, offset);
 376                 kunmap(bprm->page[j - 1]);
 377                 to = from;
 378         }
 379         memmove(to, to + offset, PAGE_SIZE - offset);
 380         kunmap(bprm->page[j - 1]);
 381
 382         /* Adjust bprm->p to point to the end of the strings. */
 383         bprm->p = PAGE_SIZE * i - offset;
 384
 385         /* Limit stack size to 1GB */
 386         stack_base = current->rlim[RLIMIT_STACK].rlim_max;
 387         if (stack_base > (1 << 30))
 388                 stack_base = 1 << 30;
 389         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
 390
 391         mm->arg_start = stack_base;
 392         arg_size = i << PAGE_SHIFT;
 393
 394         /* zero pages that were copied above */
 395         while (i < MAX_ARG_PAGES)
 396                 bprm->page[i++] = NULL;
 397 #else
 398         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 399         mm->arg_start = bprm->p + stack_base;
 400         arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
 401 #endif
 402
 403         bprm->p += stack_base;
 404         if (bprm->loader)
 405                 bprm->loader += stack_base;
 406         bprm->exec += stack_base;
 407
 408         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 409         if (!mpnt)
 410                 return -ENOMEM;
 411
 412         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT)) {
 413                 kmem_cache_free(vm_area_cachep, mpnt);
 414                 return -ENOMEM;
 415         }
 416
 417         down_write(&mm->mmap_sem);
 418         {
 419                 mpnt->vm_mm = mm;
 420 #ifdef CONFIG_STACK_GROWSUP
 421                 mpnt->vm_start = stack_base;
 422                 mpnt->vm_end = PAGE_MASK &
 423                         (PAGE_SIZE - 1 + (unsigned long) bprm->p);
 424 #else
 425                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 426                 mpnt->vm_end = STACK_TOP;
 427 #endif
 428                 mpnt->vm_page_prot = protection_map[VM_STACK_FLAGS & 0x7];
 429                 mpnt->vm_flags = VM_STACK_FLAGS;
 430                 mpnt->vm_ops = NULL;
 431                 mpnt->vm_pgoff = 0;
 432                 mpnt->vm_file = NULL;
 433                 INIT_LIST_HEAD(&mpnt->shared);
 434                 mpnt->vm_private_data = (void *) 0;
 435                 insert_vm_struct(mm, mpnt);
 436                 mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 437         }
 438
 439         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 440                 struct page *page = bprm->page[i];
 441                 if (page) {
 442                         bprm->page[i] = NULL;
 443                         put_dirty_page(current, page, stack_base,
 444                                         mpnt->vm_page_prot);
 445                 }
 446                 stack_base += PAGE_SIZE;
 447         }
 448         up_write(&mm->mmap_sem);
 449
 450         return 0;
 451 }
 452
 453 EXPORT_SYMBOL(setup_arg_pages);
 454
 455 #define free_arg_pages(bprm) do { } while (0)
 456
 457 #else
 458
 459 static inline void free_arg_pages(struct linux_binprm *bprm)
 460 {
 461         int i;
 462
 463         for (i = 0; i < MAX_ARG_PAGES; i++) {
 464                 if (bprm->page[i])
 465                         __free_page(bprm->page[i]);
 466                 bprm->page[i] = NULL;
 467         }
 468 }
 469
 470 #endif /* CONFIG_MMU */
 471
 472 struct file *open_exec(const char *name)
 473 {
 474         struct nameidata nd;
 475         int err;
 476         struct file *file;
 477
 478         nd.intent.open.flags = FMODE_READ;
 479         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 480         file = ERR_PTR(err);
 481
 482         if (!err) {
 483                 struct inode *inode = nd.dentry->d_inode;
 484                 file = ERR_PTR(-EACCES);
 485                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 486                     S_ISREG(inode->i_mode)) {
 487                         int err = permission(inode, MAY_EXEC, &nd);
 488                         if (!err && !(inode->i_mode & 0111))
 489                                 err = -EACCES;
 490                         file = ERR_PTR(err);
 491                         if (!err) {
 492                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 493                                 if (!IS_ERR(file)) {
 494                                         err = deny_write_access(file);
 495                                         if (err) {
 496                                                 fput(file);
 497                                                 file = ERR_PTR(err);
 498                                         }
 499                                 }
 500 out:
 501                                 return file;
 502                         }
 503                 }
 504                 path_release(&nd);
 505         }
 506         goto out;
 507 }
 508
 509 EXPORT_SYMBOL(open_exec);
 510
 511 int kernel_read(struct file *file, unsigned long offset,
 512         char *addr, unsigned long count)
 513 {
 514         mm_segment_t old_fs;
 515         loff_t pos = offset;
 516         int result;
 517
 518         old_fs = get_fs();
 519         set_fs(get_ds());
 520         /* The cast to a user pointer is valid due to the set_fs() */
 521         result = vfs_read(file, (void __user *)addr, count, &pos);
 522         set_fs(old_fs);
 523         return result;
 524 }
 525
 526 EXPORT_SYMBOL(kernel_read);
 527
 528 static int exec_mmap(struct mm_struct *mm)
 529 {
 530         struct task_struct *tsk;
 531         struct mm_struct * old_mm, *active_mm;
 532
 533         /* Add it to the list of mm's */
 534         spin_lock(&mmlist_lock);
 535         list_add(&mm->mmlist, &init_mm.mmlist);
 536         mmlist_nr++;
 537         spin_unlock(&mmlist_lock);
 538
 539         /* Notify parent that we're no longer interested in the old VM */
 540         tsk = current;
 541         old_mm = current->mm;
 542         mm_release(tsk, old_mm);
 543
 544         task_lock(tsk);
 545         active_mm = tsk->active_mm;
 546         tsk->mm = mm;
 547         tsk->active_mm = mm;
 548         activate_mm(active_mm, mm);
 549         task_unlock(tsk);
 550         if (old_mm) {
 551                 if (active_mm != old_mm) BUG();
 552                 mmput(old_mm);
 553                 return 0;
 554         }
 555         mmdrop(active_mm);
 556         return 0;
 557 }
 558
 559 /*
 560  * This function makes sure the current process has its own signal table,
 561  * so that flush_signal_handlers can later reset the handlers without
 562  * disturbing other processes.  (Other processes might share the signal
 563  * table via the CLONE_SIGHAND option to clone().)
 564  */
 565 static inline int de_thread(struct task_struct *tsk)
 566 {
 567         struct signal_struct *newsig, *oldsig = tsk->signal;
 568         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 569         spinlock_t *lock = &oldsighand->siglock;
 570         int count;
 571
 572         /*
 573          * If we don't share sighandlers, then we aren't sharing anything
 574          * and we can just re-use it all.
 575          */
 576         if (atomic_read(&oldsighand->count) <= 1)
 577                 return 0;
 578
 579         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 580         if (!newsighand)
 581                 return -ENOMEM;
 582
 583         spin_lock_init(&newsighand->siglock);
 584         atomic_set(&newsighand->count, 1);
 585         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
 586
 587         /*
 588          * See if we need to allocate a new signal structure
 589          */
 590         newsig = NULL;
 591         if (atomic_read(&oldsig->count) > 1) {
 592                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 593                 if (!newsig) {
 594                         kmem_cache_free(sighand_cachep, newsighand);
 595                         return -ENOMEM;
 596                 }
 597                 atomic_set(&newsig->count, 1);
 598                 newsig->group_exit = 0;
 599                 newsig->group_exit_code = 0;
 600                 newsig->group_exit_task = NULL;
 601                 newsig->group_stop_count = 0;
 602                 newsig->curr_target = NULL;
 603                 init_sigpending(&newsig->shared_pending);
 604         }
 605
 606         if (thread_group_empty(current))
 607                 goto no_thread_group;
 608
 609         /*
 610          * Kill all other threads in the thread group.
 611          * We must hold tasklist_lock to call zap_other_threads.
 612          */
 613         read_lock(&tasklist_lock);
 614         spin_lock_irq(lock);
 615         if (oldsig->group_exit) {
 616                 /*
 617                  * Another group action in progress, just
 618                  * return so that the signal is processed.
 619                  */
 620                 spin_unlock_irq(lock);
 621                 read_unlock(&tasklist_lock);
 622                 kmem_cache_free(sighand_cachep, newsighand);
 623                 if (newsig)
 624                         kmem_cache_free(signal_cachep, newsig);
 625                 return -EAGAIN;
 626         }
 627         oldsig->group_exit = 1;
 628         zap_other_threads(current);
 629         read_unlock(&tasklist_lock);
 630
 631         /*
 632          * Account for the thread group leader hanging around:
 633          */
 634         count = 2;
 635         if (current->pid == current->tgid)
 636                 count = 1;
 637         while (atomic_read(&oldsig->count) > count) {
 638                 oldsig->group_exit_task = current;
 639                 oldsig->notify_count = count;
 640                 __set_current_state(TASK_UNINTERRUPTIBLE);
 641                 spin_unlock_irq(lock);
 642                 schedule();
 643                 spin_lock_irq(lock);
 644         }
 645         spin_unlock_irq(lock);
 646
 647         /*
 648          * At this point all other threads have exited, all we have to
 649          * do is to wait for the thread group leader to become inactive,
 650          * and to assume its PID:
 651          */
 652         if (current->pid != current->tgid) {
 653                 struct task_struct *leader = current->group_leader, *parent;
 654                 struct dentry *proc_dentry1, *proc_dentry2;
 655                 unsigned long state, ptrace;
 656
 657                 /*
 658                  * Wait for the thread group leader to be a zombie.
 659                  * It should already be zombie at this point, most
 660                  * of the time.
 661                  */
 662                 while (leader->state != TASK_ZOMBIE)
 663                         yield();
 664
 665                 spin_lock(&leader->proc_lock);
 666                 spin_lock(&current->proc_lock);
 667                 proc_dentry1 = proc_pid_unhash(current);
 668                 proc_dentry2 = proc_pid_unhash(leader);
 669                 write_lock_irq(&tasklist_lock);
 670
 671                 if (leader->tgid != current->tgid)
 672                         BUG();
 673                 if (current->pid == current->tgid)
 674                         BUG();
 675                 /*
 676                  * An exec() starts a new thread group with the
 677                  * TGID of the previous thread group. Rehash the
 678                  * two threads with a switched PID, and release
 679                  * the former thread group leader:
 680                  */
 681                 ptrace = leader->ptrace;
 682                 parent = leader->parent;
 683
 684                 ptrace_unlink(current);
 685                 ptrace_unlink(leader);
 686                 remove_parent(current);
 687                 remove_parent(leader);
 688
 689                 switch_exec_pids(leader, current);
 690
 691                 current->parent = current->real_parent = leader->real_parent;
 692                 leader->parent = leader->real_parent = child_reaper;
 693                 current->group_leader = current;
 694                 leader->group_leader = leader;
 695
 696                 add_parent(current, current->parent);
 697                 add_parent(leader, leader->parent);
 698                 if (ptrace) {
 699                         current->ptrace = ptrace;
 700                         __ptrace_link(current, parent);
 701                 }
 702
 703                 list_del(&current->tasks);
 704                 list_add_tail(&current->tasks, &init_task.tasks);
 705                 current->exit_signal = SIGCHLD;
 706                 state = leader->state;
 707
 708                 write_unlock_irq(&tasklist_lock);
 709                 spin_unlock(&leader->proc_lock);
 710                 spin_unlock(&current->proc_lock);
 711                 proc_pid_flush(proc_dentry1);
 712                 proc_pid_flush(proc_dentry2);
 713
 714                 if (state != TASK_ZOMBIE)
 715                         BUG();
 716                 release_task(leader);
 717         }
 718
 719 no_thread_group:
 720
 721         write_lock_irq(&tasklist_lock);
 722         spin_lock(&oldsighand->siglock);
 723         spin_lock(&newsighand->siglock);
 724
 725         if (current == oldsig->curr_target)
 726                 oldsig->curr_target = next_thread(current);
 727         if (newsig)
 728                 current->signal = newsig;
 729         current->sighand = newsighand;
 730         init_sigpending(&current->pending);
 731         recalc_sigpending();
 732
 733         spin_unlock(&newsighand->siglock);
 734         spin_unlock(&oldsighand->siglock);
 735         write_unlock_irq(&tasklist_lock);
 736
 737         if (newsig && atomic_dec_and_test(&oldsig->count))
 738                 kmem_cache_free(signal_cachep, oldsig);
 739
 740         if (atomic_dec_and_test(&oldsighand->count))
 741                 kmem_cache_free(sighand_cachep, oldsighand);
 742
 743         if (!thread_group_empty(current))
 744                 BUG();
 745         if (current->tgid != current->pid)
 746                 BUG();
 747         return 0;
 748 }
 749
 750 /*
 751  * These functions flushes out all traces of the currently running executable
 752  * so that a new one can be started
 753  */
 754
 755 static inline void flush_old_files(struct files_struct * files)
 756 {
 757         long j = -1;
 758
 759         spin_lock(&files->file_lock);
 760         for (;;) {
 761                 unsigned long set, i;
 762
 763                 j++;
 764                 i = j * __NFDBITS;
 765                 if (i >= files->max_fds || i >= files->max_fdset)
 766                         break;
 767                 set = files->close_on_exec->fds_bits[j];
 768                 if (!set)
 769                         continue;
 770                 files->close_on_exec->fds_bits[j] = 0;
 771                 spin_unlock(&files->file_lock);
 772                 for ( ; set ; i++,set >>= 1) {
 773                         if (set & 1) {
 774                                 sys_close(i);
 775                         }
 776                 }
 777                 spin_lock(&files->file_lock);
 778
 779         }
 780         spin_unlock(&files->file_lock);
 781 }
 782
 783 int flush_old_exec(struct linux_binprm * bprm)
 784 {
 785         char * name;
 786         int i, ch, retval;
 787         struct files_struct *files;
 788
 789         /*
 790          * Make sure we have a private signal table and that
 791          * we are unassociated from the previous thread group.
 792          */
 793         retval = de_thread(current);
 794         if (retval)
 795                 goto out;
 796
 797         /*
 798          * Make sure we have private file handles. Ask the
 799          * fork helper to do the work for us and the exit
 800          * helper to do the cleanup of the old one.
 801          */
 802         files = current->files;         /* refcounted so safe to hold */
 803         retval = unshare_files();
 804         if (retval)
 805                 goto out;
 806         /*
 807          * Release all of the old mmap stuff
 808          */
 809         retval = exec_mmap(bprm->mm);
 810         if (retval)
 811                 goto mmap_failed;
 812
 813         bprm->mm = NULL;                /* We're using it now */
 814
 815         /* This is the point of no return */
 816         steal_locks(files);
 817         put_files_struct(files);
 818
 819         current->sas_ss_sp = current->sas_ss_size = 0;
 820
 821         if (current->euid == current->uid && current->egid == current->gid)
 822                 current->mm->dumpable = 1;
 823         name = bprm->filename;
 824         for (i=0; (ch = *(name++)) != '\0';) {
 825                 if (ch == '/')
 826                         i = 0;
 827                 else
 828                         if (i < 15)
 829                                 current->comm[i++] = ch;
 830         }
 831         current->comm[i] = '\0';
 832
 833         flush_thread();
 834
 835         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 836             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL))
 837                 current->mm->dumpable = 0;
 838
 839         /* An exec changes our domain. We are no longer part of the thread
 840            group */
 841
 842         current->self_exec_id++;
 843
 844         flush_signal_handlers(current, 0);
 845         flush_old_files(current->files);
 846         exit_itimers(current);
 847
 848         return 0;
 849
 850 mmap_failed:
 851         put_files_struct(current->files);
 852         current->files = files;
 853 out:
 854         return retval;
 855 }
 856
 857 EXPORT_SYMBOL(flush_old_exec);
 858
 859 /*
 860  * We mustn't allow tracing of suid binaries, unless
 861  * the tracer has the capability to trace anything..
 862  */
 863 static inline int must_not_trace_exec(struct task_struct * p)
 864 {
 865         return (p->ptrace & PT_PTRACED) && !(p->ptrace & PT_PTRACE_CAP);
 866 }
 867
 868 /*
 869  * Fill the binprm structure from the inode.
 870  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 871  */
 872 int prepare_binprm(struct linux_binprm *bprm)
 873 {
 874         int mode;
 875         struct inode * inode = bprm->file->f_dentry->d_inode;
 876         int retval;
 877
 878         mode = inode->i_mode;
 879         /*
 880          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 881          * vfs_permission lets a non-executable through
 882          */
 883         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 884                 return -EACCES;
 885         if (bprm->file->f_op == NULL)
 886                 return -EACCES;
 887
 888         bprm->e_uid = current->euid;
 889         bprm->e_gid = current->egid;
 890
 891         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 892                 /* Set-uid? */
 893                 if (mode & S_ISUID)
 894                         bprm->e_uid = inode->i_uid;
 895
 896                 /* Set-gid? */
 897                 /*
 898                  * If setgid is set but no group execute bit then this
 899                  * is a candidate for mandatory locking, not a setgid
 900                  * executable.
 901                  */
 902                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
 903                         bprm->e_gid = inode->i_gid;
 904         }
 905
 906         /* fill in binprm security blob */
 907         retval = security_bprm_set(bprm);
 908         if (retval)
 909                 return retval;
 910
 911         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 912         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 913 }
 914
 915 EXPORT_SYMBOL(prepare_binprm);
 916
 917 /*
 918  * This function is used to produce the new IDs and capabilities
 919  * from the old ones and the file's capabilities.
 920  *
 921  * The formula used for evolving capabilities is:
 922  *
 923  *       pI' = pI
 924  * (***) pP' = (fP & X) | (fI & pI)
 925  *       pE' = pP' & fE          [NB. fE is 0 or ~0]
 926  *
 927  * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
 928  * ' indicates post-exec(), and X is the global 'cap_bset'.
 929  *
 930  */
 931
 932 void compute_creds(struct linux_binprm *bprm)
 933 {
 934         task_lock(current);
 935         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid) {
 936                 current->mm->dumpable = 0;
 937
 938                 if (must_not_trace_exec(current)
 939                     || atomic_read(&current->fs->count) > 1
 940                     || atomic_read(&current->files->count) > 1
 941                     || atomic_read(&current->sighand->count) > 1) {
 942                         if(!capable(CAP_SETUID)) {
 943                                 bprm->e_uid = current->uid;
 944                                 bprm->e_gid = current->gid;
 945                         }
 946                 }
 947         }
 948
 949         current->suid = current->euid = current->fsuid = bprm->e_uid;
 950         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 951
 952         task_unlock(current);
 953
 954         security_bprm_compute_creds(bprm);
 955 }
 956
 957 EXPORT_SYMBOL(compute_creds);
 958
 959 void remove_arg_zero(struct linux_binprm *bprm)
 960 {
 961         if (bprm->argc) {
 962                 unsigned long offset;
 963                 char * kaddr;
 964                 struct page *page;
 965
 966                 offset = bprm->p % PAGE_SIZE;
 967                 goto inside;
 968
 969                 while (bprm->p++, *(kaddr+offset++)) {
 970                         if (offset != PAGE_SIZE)
 971                                 continue;
 972                         offset = 0;
 973                         kunmap_atomic(kaddr, KM_USER0);
 974 inside:
 975                         page = bprm->page[bprm->p/PAGE_SIZE];
 976                         kaddr = kmap_atomic(page, KM_USER0);
 977                 }
 978                 kunmap_atomic(kaddr, KM_USER0);
 979                 bprm->argc--;
 980         }
 981 }
 982
 983 EXPORT_SYMBOL(remove_arg_zero);
 984
 985 /*
 986  * cycle the list of binary formats handler, until one recognizes the image
 987  */
 988 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 989 {
 990         int try,retval=0;
 991         struct linux_binfmt *fmt;
 992 #ifdef __alpha__
 993         /* handle /sbin/loader.. */
 994         {
 995             struct exec * eh = (struct exec *) bprm->buf;
 996
 997             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 998                 (eh->fh.f_flags & 0x3000) == 0x3000)
 999             {
1000                 struct file * file;
1001                 unsigned long loader;
1002
1003                 allow_write_access(bprm->file);
1004                 fput(bprm->file);
1005                 bprm->file = NULL;
1006
1007                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1008
1009                 file = open_exec("/sbin/loader");
1010                 retval = PTR_ERR(file);
1011                 if (IS_ERR(file))
1012                         return retval;
1013
1014                 /* Remember if the application is TASO.  */
1015                 bprm->sh_bang = eh->ah.entry < 0x100000000;
1016
1017                 bprm->file = file;
1018                 bprm->loader = loader;
1019                 retval = prepare_binprm(bprm);
1020                 if (retval<0)
1021                         return retval;
1022                 /* should call search_binary_handler recursively here,
1023                    but it does not matter */
1024             }
1025         }
1026 #endif
1027         retval = security_bprm_check(bprm);
1028         if (retval)
1029                 return retval;
1030
1031         /* kernel module loader fixup */
1032         /* so we don't try to load run modprobe in kernel space. */
1033         set_fs(USER_DS);
1034         for (try=0; try<2; try++) {
1035                 read_lock(&binfmt_lock);
1036                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1037                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1038                         if (!fn)
1039                                 continue;
1040                         if (!try_module_get(fmt->module))
1041                                 continue;
1042                         read_unlock(&binfmt_lock);
1043                         retval = fn(bprm, regs);
1044                         if (retval >= 0) {
1045                                 put_binfmt(fmt);
1046                                 allow_write_access(bprm->file);
1047                                 if (bprm->file)
1048                                         fput(bprm->file);
1049                                 bprm->file = NULL;
1050                                 current->did_exec = 1;
1051                                 return retval;
1052                         }
1053                         read_lock(&binfmt_lock);
1054                         put_binfmt(fmt);
1055                         if (retval != -ENOEXEC || bprm->mm == NULL)
1056                                 break;
1057                         if (!bprm->file) {
1058                                 read_unlock(&binfmt_lock);
1059                                 return retval;
1060                         }
1061                 }
1062                 read_unlock(&binfmt_lock);
1063                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1064                         break;
1065 #ifdef CONFIG_KMOD
1066                 }else{
1067 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1068                         if (printable(bprm->buf[0]) &&
1069                             printable(bprm->buf[1]) &&
1070                             printable(bprm->buf[2]) &&
1071                             printable(bprm->buf[3]))
1072                                 break; /* -ENOEXEC */
1073                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1074 #endif
1075                 }
1076         }
1077         return retval;
1078 }
1079
1080 EXPORT_SYMBOL(search_binary_handler);
1081
1082 /*
1083  * sys_execve() executes a new program.
1084  */
1085 int do_execve(char * filename,
1086         char __user *__user *argv,
1087         char __user *__user *envp,
1088         struct pt_regs * regs)
1089 {
1090         struct linux_binprm bprm;
1091         struct file *file;
1092         int retval;
1093         int i;
1094
1095         sched_balance_exec();
1096
1097         file = open_exec(filename);
1098
1099         retval = PTR_ERR(file);
1100         if (IS_ERR(file))
1101                 return retval;
1102
1103         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1104         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
1105
1106         bprm.file = file;
1107         bprm.filename = filename;
1108         bprm.interp = filename;
1109         bprm.sh_bang = 0;
1110         bprm.loader = 0;
1111         bprm.exec = 0;
1112         bprm.security = NULL;
1113         bprm.mm = mm_alloc();
1114         retval = -ENOMEM;
1115         if (!bprm.mm)
1116                 goto out_file;
1117
1118         retval = init_new_context(current, bprm.mm);
1119         if (retval < 0)
1120                 goto out_mm;
1121
1122         bprm.argc = count(argv, bprm.p / sizeof(void *));
1123         if ((retval = bprm.argc) < 0)
1124                 goto out_mm;
1125
1126         bprm.envc = count(envp, bprm.p / sizeof(void *));
1127         if ((retval = bprm.envc) < 0)
1128                 goto out_mm;
1129
1130         retval = security_bprm_alloc(&bprm);
1131         if (retval)
1132                 goto out;
1133
1134         retval = prepare_binprm(&bprm);
1135         if (retval < 0)
1136                 goto out;
1137
1138         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
1139         if (retval < 0)
1140                 goto out;
1141
1142         bprm.exec = bprm.p;
1143         retval = copy_strings(bprm.envc, envp, &bprm);
1144         if (retval < 0)
1145                 goto out;
1146
1147         retval = copy_strings(bprm.argc, argv, &bprm);
1148         if (retval < 0)
1149                 goto out;
1150
1151         retval = search_binary_handler(&bprm,regs);
1152         if (retval >= 0) {
1153                 free_arg_pages(&bprm);
1154
1155                 /* execve success */
1156                 security_bprm_free(&bprm);
1157                 return retval;
1158         }
1159
1160 out:
1161         /* Something went wrong, return the inode and free the argument pages*/
1162         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1163                 struct page * page = bprm.page[i];
1164                 if (page)
1165                         __free_page(page);
1166         }
1167
1168         if (bprm.security)
1169                 security_bprm_free(&bprm);
1170
1171 out_mm:
1172         if (bprm.mm)
1173                 mmdrop(bprm.mm);
1174
1175 out_file:
1176         if (bprm.file) {
1177                 allow_write_access(bprm.file);
1178                 fput(bprm.file);
1179         }
1180         return retval;
1181 }
1182
1183 EXPORT_SYMBOL(do_execve);
1184
1185 int set_binfmt(struct linux_binfmt *new)
1186 {
1187         struct linux_binfmt *old = current->binfmt;
1188
1189         if (new) {
1190                 if (!try_module_get(new->module))
1191                         return -1;
1192         }
1193         current->binfmt = new;
1194         if (old)
1195                 module_put(old->module);
1196         return 0;
1197 }
1198
1199 EXPORT_SYMBOL(set_binfmt);
1200
1201 #define CORENAME_MAX_SIZE 64
1202
1203 /* format_corename will inspect the pattern parameter, and output a
1204  * name into corename, which must have space for at least
1205  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1206  */
1207 void format_corename(char *corename, const char *pattern, long signr)
1208 {
1209         const char *pat_ptr = pattern;
1210         char *out_ptr = corename;
1211         char *const out_end = corename + CORENAME_MAX_SIZE;
1212         int rc;
1213         int pid_in_pattern = 0;
1214
1215         /* Repeat as long as we have more pattern to process and more output
1216            space */
1217         while (*pat_ptr) {
1218                 if (*pat_ptr != '%') {
1219                         if (out_ptr == out_end)
1220                                 goto out;
1221                         *out_ptr++ = *pat_ptr++;
1222                 } else {
1223                         switch (*++pat_ptr) {
1224                         case 0:
1225                                 goto out;
1226                         /* Double percent, output one percent */
1227                         case '%':
1228                                 if (out_ptr == out_end)
1229                                         goto out;
1230                                 *out_ptr++ = '%';
1231                                 break;
1232                         /* pid */
1233                         case 'p':
1234                                 pid_in_pattern = 1;
1235                                 rc = snprintf(out_ptr, out_end - out_ptr,
1236                                               "%d", current->tgid);
1237                                 if (rc > out_end - out_ptr)
1238                                         goto out;
1239                                 out_ptr += rc;
1240                                 break;
1241                         /* uid */
1242                         case 'u':
1243                                 rc = snprintf(out_ptr, out_end - out_ptr,
1244                                               "%d", current->uid);
1245                                 if (rc > out_end - out_ptr)
1246                                         goto out;
1247                                 out_ptr += rc;
1248                                 break;
1249                         /* gid */
1250                         case 'g':
1251                                 rc = snprintf(out_ptr, out_end - out_ptr,
1252                                               "%d", current->gid);
1253                                 if (rc > out_end - out_ptr)
1254                                         goto out;
1255                                 out_ptr += rc;
1256                                 break;
1257                         /* signal that caused the coredump */
1258                         case 's':
1259                                 rc = snprintf(out_ptr, out_end - out_ptr,
1260                                               "%ld", signr);
1261                                 if (rc > out_end - out_ptr)
1262                                         goto out;
1263                                 out_ptr += rc;
1264                                 break;
1265                         /* UNIX time of coredump */
1266                         case 't': {
1267                                 struct timeval tv;
1268                                 do_gettimeofday(&tv);
1269                                 rc = snprintf(out_ptr, out_end - out_ptr,
1270                                               "%lu", tv.tv_sec);
1271                                 if (rc > out_end - out_ptr)
1272                                         goto out;
1273                                 out_ptr += rc;
1274                                 break;
1275                         }
1276                         /* hostname */
1277                         case 'h':
1278                                 down_read(&uts_sem);
1279                                 rc = snprintf(out_ptr, out_end - out_ptr,
1280                                               "%s", system_utsname.nodename);
1281                                 up_read(&uts_sem);
1282                                 if (rc > out_end - out_ptr)
1283                                         goto out;
1284                                 out_ptr += rc;
1285                                 break;
1286                         /* executable */
1287                         case 'e':
1288                                 rc = snprintf(out_ptr, out_end - out_ptr,
1289                                               "%s", current->comm);
1290                                 if (rc > out_end - out_ptr)
1291                                         goto out;
1292                                 out_ptr += rc;
1293                                 break;
1294                         default:
1295                                 break;
1296                         }
1297                         ++pat_ptr;
1298                 }
1299         }
1300         /* Backward compatibility with core_uses_pid:
1301          *
1302          * If core_pattern does not include a %p (as is the default)
1303          * and core_uses_pid is set, then .%pid will be appended to
1304          * the filename */
1305         if (!pid_in_pattern
1306             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1307                 rc = snprintf(out_ptr, out_end - out_ptr,
1308                               ".%d", current->tgid);
1309                 if (rc > out_end - out_ptr)
1310                         goto out;
1311                 out_ptr += rc;
1312         }
1313       out:
1314         *out_ptr = 0;
1315 }
1316
1317 static void zap_threads (struct mm_struct *mm)
1318 {
1319         struct task_struct *g, *p;
1320         struct task_struct *tsk = current;
1321         struct completion *vfork_done = tsk->vfork_done;
1322
1323         /*
1324          * Make sure nobody is waiting for us to release the VM,
1325          * otherwise we can deadlock when we wait on each other
1326          */
1327         if (vfork_done) {
1328                 tsk->vfork_done = NULL;
1329                 complete(vfork_done);
1330         }
1331
1332         read_lock(&tasklist_lock);
1333         do_each_thread(g,p)
1334                 if (mm == p->mm && p != tsk) {
1335                         force_sig_specific(SIGKILL, p);
1336                         mm->core_waiters++;
1337                 }
1338         while_each_thread(g,p);
1339
1340         read_unlock(&tasklist_lock);
1341 }
1342
1343 static void coredump_wait(struct mm_struct *mm)
1344 {
1345         DECLARE_COMPLETION(startup_done);
1346
1347         mm->core_waiters++; /* let other threads block */
1348         mm->core_startup_done = &startup_done;
1349
1350         /* give other threads a chance to run: */
1351         yield();
1352
1353         zap_threads(mm);
1354         if (--mm->core_waiters) {
1355                 up_write(&mm->mmap_sem);
1356                 wait_for_completion(&startup_done);
1357         } else
1358                 up_write(&mm->mmap_sem);
1359         BUG_ON(mm->core_waiters);
1360 }
1361
1362 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1363 {
1364         char corename[CORENAME_MAX_SIZE + 1];
1365         struct mm_struct *mm = current->mm;
1366         struct linux_binfmt * binfmt;
1367         struct inode * inode;
1368         struct file * file;
1369         int retval = 0;
1370
1371         lock_kernel();
1372         binfmt = current->binfmt;
1373         if (!binfmt || !binfmt->core_dump)
1374                 goto fail;
1375         down_write(&mm->mmap_sem);
1376         if (!mm->dumpable) {
1377                 up_write(&mm->mmap_sem);
1378                 goto fail;
1379         }
1380         mm->dumpable = 0;
1381         init_completion(&mm->core_done);
1382         current->signal->group_exit = 1;
1383         current->signal->group_exit_code = exit_code;
1384         coredump_wait(mm);
1385
1386         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1387                 goto fail_unlock;
1388
1389         format_corename(corename, core_pattern, signr);
1390         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1391         if (IS_ERR(file))
1392                 goto fail_unlock;
1393         inode = file->f_dentry->d_inode;
1394         if (inode->i_nlink > 1)
1395                 goto close_fail;        /* multiple links - don't dump */
1396         if (d_unhashed(file->f_dentry))
1397                 goto close_fail;
1398
1399         if (!S_ISREG(inode->i_mode))
1400                 goto close_fail;
1401         if (!file->f_op)
1402                 goto close_fail;
1403         if (!file->f_op->write)
1404                 goto close_fail;
1405         if (do_truncate(file->f_dentry, 0) != 0)
1406                 goto close_fail;
1407
1408         retval = binfmt->core_dump(signr, regs, file);
1409
1410         current->signal->group_exit_code |= 0x80;
1411 close_fail:
1412         filp_close(file, NULL);
1413 fail_unlock:
1414         complete_all(&mm->core_done);
1415 fail:
1416         unlock_kernel();
1417         return retval;
1418 }