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