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