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/mmu_context.h>
  53
  54 #ifdef CONFIG_KMOD
  55 #include <linux/kmod.h>
  56 #endif
  57
  58 int core_uses_pid;
  59 char core_pattern[65] = "core";
  60 /* The maximal length of core_pattern is also specified in sysctl.c */
  61
  62 static struct linux_binfmt *formats;
  63 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  64
  65 int register_binfmt(struct linux_binfmt * fmt)
  66 {
  67         struct linux_binfmt ** tmp = &formats;
  68
  69         if (!fmt)
  70                 return -EINVAL;
  71         if (fmt->next)
  72                 return -EBUSY;
  73         write_lock(&binfmt_lock);
  74         while (*tmp) {
  75                 if (fmt == *tmp) {
  76                         write_unlock(&binfmt_lock);
  77                         return -EBUSY;
  78                 }
  79                 tmp = &(*tmp)->next;
  80         }
  81         fmt->next = formats;
  82         formats = fmt;
  83         write_unlock(&binfmt_lock);
  84         return 0;
  85 }
  86
  87 EXPORT_SYMBOL(register_binfmt);
  88
  89 int unregister_binfmt(struct linux_binfmt * fmt)
  90 {
  91         struct linux_binfmt ** tmp = &formats;
  92
  93         write_lock(&binfmt_lock);
  94         while (*tmp) {
  95                 if (fmt == *tmp) {
  96                         *tmp = fmt->next;
  97                         write_unlock(&binfmt_lock);
  98                         return 0;
  99                 }
 100                 tmp = &(*tmp)->next;
 101         }
 102         write_unlock(&binfmt_lock);
 103         return -EINVAL;
 104 }
 105
 106 EXPORT_SYMBOL(unregister_binfmt);
 107
 108 static inline void put_binfmt(struct linux_binfmt * fmt)
 109 {
 110         module_put(fmt->module);
 111 }
 112
 113 /*
 114  * Note that a shared library must be both readable and executable due to
 115  * security reasons.
 116  *
 117  * Also note that we take the address to load from from the file itself.
 118  */
 119 asmlinkage long sys_uselib(const char __user * library)
 120 {
 121         struct file * file;
 122         struct nameidata nd;
 123         int error;
 124
 125         nd.intent.open.flags = FMODE_READ;
 126         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 127         if (error)
 128                 goto out;
 129
 130         error = -EINVAL;
 131         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 132                 goto exit;
 133
 134         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 135         if (error)
 136                 goto exit;
 137
 138         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 139         error = PTR_ERR(file);
 140         if (IS_ERR(file))
 141                 goto out;
 142
 143         error = -ENOEXEC;
 144         if(file->f_op) {
 145                 struct linux_binfmt * fmt;
 146
 147                 read_lock(&binfmt_lock);
 148                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 149                         if (!fmt->load_shlib)
 150                                 continue;
 151                         if (!try_module_get(fmt->module))
 152                                 continue;
 153                         read_unlock(&binfmt_lock);
 154                         error = fmt->load_shlib(file);
 155                         read_lock(&binfmt_lock);
 156                         put_binfmt(fmt);
 157                         if (error != -ENOEXEC)
 158                                 break;
 159                 }
 160                 read_unlock(&binfmt_lock);
 161         }
 162         fput(file);
 163 out:
 164         return error;
 165 exit:
 166         path_release(&nd);
 167         goto out;
 168 }
 169
 170 /*
 171  * count() counts the number of strings in array ARGV.
 172  */
 173 static int count(char __user * __user * argv, int max)
 174 {
 175         int i = 0;
 176
 177         if (argv != NULL) {
 178                 for (;;) {
 179                         char __user * p;
 180
 181                         if (get_user(p, argv))
 182                                 return -EFAULT;
 183                         if (!p)
 184                                 break;
 185                         argv++;
 186                         if(++i > max)
 187                                 return -E2BIG;
 188                 }
 189         }
 190         return i;
 191 }
 192
 193 /*
 194  * 'copy_strings()' copies argument/environment strings from user
 195  * memory to free pages in kernel mem. These are in a format ready
 196  * to be put directly into the top of new user memory.
 197  */
 198 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 199 {
 200         struct page *kmapped_page = NULL;
 201         char *kaddr = NULL;
 202         int ret;
 203
 204         while (argc-- > 0) {
 205                 char __user *str;
 206                 int len;
 207                 unsigned long pos;
 208
 209                 if (get_user(str, argv+argc) ||
 210                                 !(len = strnlen_user(str, bprm->p))) {
 211                         ret = -EFAULT;
 212                         goto out;
 213                 }
 214
 215                 if (bprm->p < len)  {
 216                         ret = -E2BIG;
 217                         goto out;
 218                 }
 219
 220                 bprm->p -= len;
 221                 /* XXX: add architecture specific overflow check here. */
 222                 pos = bprm->p;
 223
 224                 while (len > 0) {
 225                         int i, new, err;
 226                         int offset, bytes_to_copy;
 227                         struct page *page;
 228
 229                         offset = pos % PAGE_SIZE;
 230                         i = pos/PAGE_SIZE;
 231                         page = bprm->page[i];
 232                         new = 0;
 233                         if (!page) {
 234                                 page = alloc_page(GFP_HIGHUSER);
 235                                 bprm->page[i] = page;
 236                                 if (!page) {
 237                                         ret = -ENOMEM;
 238                                         goto out;
 239                                 }
 240                                 new = 1;
 241                         }
 242
 243                         if (page != kmapped_page) {
 244                                 if (kmapped_page)
 245                                         kunmap(kmapped_page);
 246                                 kmapped_page = page;
 247                                 kaddr = kmap(kmapped_page);
 248                         }
 249                         if (new && offset)
 250                                 memset(kaddr, 0, offset);
 251                         bytes_to_copy = PAGE_SIZE - offset;
 252                         if (bytes_to_copy > len) {
 253                                 bytes_to_copy = len;
 254                                 if (new)
 255                                         memset(kaddr+offset+len, 0,
 256                                                 PAGE_SIZE-offset-len);
 257                         }
 258                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 259                         if (err) {
 260                                 ret = -EFAULT;
 261                                 goto out;
 262                         }
 263
 264                         pos += bytes_to_copy;
 265                         str += bytes_to_copy;
 266                         len -= bytes_to_copy;
 267                 }
 268         }
 269         ret = 0;
 270 out:
 271         if (kmapped_page)
 272                 kunmap(kmapped_page);
 273         return ret;
 274 }
 275
 276 /*
 277  * Like copy_strings, but get argv and its values from kernel memory.
 278  */
 279 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 280 {
 281         int r;
 282         mm_segment_t oldfs = get_fs();
 283         set_fs(KERNEL_DS);
 284         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 285         set_fs(oldfs);
 286         return r;
 287 }
 288
 289 EXPORT_SYMBOL(copy_strings_kernel);
 290
 291 #ifdef CONFIG_MMU
 292 /*
 293  * This routine is used to map in a page into an address space: needed by
 294  * execve() for the initial stack and environment pages.
 295  *
 296  * vma->vm_mm->mmap_sem is held for writing.
 297  */
 298 void install_arg_page(struct vm_area_struct *vma,
 299                         struct page *page, unsigned long address)
 300 {
 301         struct mm_struct *mm = vma->vm_mm;
 302         pgd_t * pgd;
 303         pmd_t * pmd;
 304         pte_t * pte;
 305
 306         if (unlikely(anon_vma_prepare(vma)))
 307                 goto out_sig;
 308
 309         flush_dcache_page(page);
 310         pgd = pgd_offset(mm, address);
 311
 312         spin_lock(&mm->page_table_lock);
 313         pmd = pmd_alloc(mm, pgd, address);
 314         if (!pmd)
 315                 goto out;
 316         pte = pte_alloc_map(mm, pmd, address);
 317         if (!pte)
 318                 goto out;
 319         if (!pte_none(*pte)) {
 320                 pte_unmap(pte);
 321                 goto out;
 322         }
 323         // mm->rss++;
 324         vx_rsspages_inc(mm);
 325         lru_cache_add_active(page);
 326         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
 327                                         page, vma->vm_page_prot))));
 328         page_add_anon_rmap(page, vma, address);
 329         pte_unmap(pte);
 330         spin_unlock(&mm->page_table_lock);
 331
 332         /* no need for flush_tlb */
 333         return;
 334 out:
 335         spin_unlock(&mm->page_table_lock);
 336 out_sig:
 337         __free_page(page);
 338         force_sig(SIGKILL, current);
 339 }
 340
 341 int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
 342 {
 343         unsigned long stack_base;
 344         struct vm_area_struct *mpnt;
 345         struct mm_struct *mm = current->mm;
 346         int i;
 347         long arg_size;
 348
 349 #ifdef CONFIG_STACK_GROWSUP
 350         /* Move the argument and environment strings to the bottom of the
 351          * stack space.
 352          */
 353         int offset, j;
 354         char *to, *from;
 355
 356         /* Start by shifting all the pages down */
 357         i = 0;
 358         for (j = 0; j < MAX_ARG_PAGES; j++) {
 359                 struct page *page = bprm->page[j];
 360                 if (!page)
 361                         continue;
 362                 bprm->page[i++] = page;
 363         }
 364
 365         /* Now move them within their pages */
 366         offset = bprm->p % PAGE_SIZE;
 367         to = kmap(bprm->page[0]);
 368         for (j = 1; j < i; j++) {
 369                 memmove(to, to + offset, PAGE_SIZE - offset);
 370                 from = kmap(bprm->page[j]);
 371                 memcpy(to + PAGE_SIZE - offset, from, offset);
 372                 kunmap(bprm->page[j - 1]);
 373                 to = from;
 374         }
 375         memmove(to, to + offset, PAGE_SIZE - offset);
 376         kunmap(bprm->page[j - 1]);
 377
 378         /* Adjust bprm->p to point to the end of the strings. */
 379         bprm->p = PAGE_SIZE * i - offset;
 380
 381         /* Limit stack size to 1GB */
 382         stack_base = current->rlim[RLIMIT_STACK].rlim_max;
 383         if (stack_base > (1 << 30))
 384                 stack_base = 1 << 30;
 385         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
 386
 387         mm->arg_start = stack_base;
 388         arg_size = i << PAGE_SHIFT;
 389
 390         /* zero pages that were copied above */
 391         while (i < MAX_ARG_PAGES)
 392                 bprm->page[i++] = NULL;
 393 #else
 394 #ifdef __HAVE_ARCH_ALIGN_STACK
 395         stack_base = arch_align_stack(STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE);
 396         stack_base = PAGE_ALIGN(stack_base);
 397 #else
 398         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 399 #endif
 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                 !vx_vmpages_avail(mm, arg_size >> PAGE_SHIFT)) {
 415                 kmem_cache_free(vm_area_cachep, mpnt);
 416                 return -ENOMEM;
 417         }
 418
 419         memset(mpnt, 0, sizeof(*mpnt));
 420
 421         down_write(&mm->mmap_sem);
 422         {
 423                 mpnt->vm_mm = mm;
 424 #ifdef CONFIG_STACK_GROWSUP
 425                 mpnt->vm_start = stack_base;
 426                 mpnt->vm_end = PAGE_MASK &
 427                         (PAGE_SIZE - 1 + (unsigned long) bprm->p);
 428 #else
 429                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 430                 mpnt->vm_end = STACK_TOP;
 431 #endif
 432                 /* Adjust stack execute permissions; explicitly enable
 433                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 434                  * and leave alone (arch default) otherwise. */
 435                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 436                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 437                 else if (executable_stack == EXSTACK_DISABLE_X)
 438                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 439                 else
 440                         mpnt->vm_flags = VM_STACK_FLAGS;
 441                 mpnt->vm_flags |= mm->def_flags;
 442                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 443                 insert_vm_struct(mm, mpnt);
 444                 // mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 445                 vx_vmpages_sub(mm, mm->total_vm -
 446                         ((mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT));
 447         }
 448
 449         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 450                 struct page *page = bprm->page[i];
 451                 if (page) {
 452                         bprm->page[i] = NULL;
 453                         install_arg_page(mpnt, page, stack_base);
 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         arch_pick_mmap_layout(mm);
 560         if (old_mm) {
 561                 if (active_mm != old_mm) BUG();
 562                 mmput(old_mm);
 563                 return 0;
 564         }
 565         mmdrop(active_mm);
 566         return 0;
 567 }
 568
 569 /*
 570  * This function makes sure the current process has its own signal table,
 571  * so that flush_signal_handlers can later reset the handlers without
 572  * disturbing other processes.  (Other processes might share the signal
 573  * table via the CLONE_SIGHAND option to clone().)
 574  */
 575 static inline int de_thread(struct task_struct *tsk)
 576 {
 577         struct signal_struct *newsig, *oldsig = tsk->signal;
 578         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 579         spinlock_t *lock = &oldsighand->siglock;
 580         int count;
 581
 582         /*
 583          * If we don't share sighandlers, then we aren't sharing anything
 584          * and we can just re-use it all.
 585          */
 586         if (atomic_read(&oldsighand->count) <= 1)
 587                 return 0;
 588
 589         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 590         if (!newsighand)
 591                 return -ENOMEM;
 592
 593         spin_lock_init(&newsighand->siglock);
 594         atomic_set(&newsighand->count, 1);
 595         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
 596
 597         /*
 598          * See if we need to allocate a new signal structure
 599          */
 600         newsig = NULL;
 601         if (atomic_read(&oldsig->count) > 1) {
 602                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 603                 if (!newsig) {
 604                         kmem_cache_free(sighand_cachep, newsighand);
 605                         return -ENOMEM;
 606                 }
 607                 atomic_set(&newsig->count, 1);
 608                 newsig->group_exit = 0;
 609                 newsig->group_exit_code = 0;
 610                 newsig->group_exit_task = NULL;
 611                 newsig->group_stop_count = 0;
 612                 newsig->curr_target = NULL;
 613                 init_sigpending(&newsig->shared_pending);
 614                 INIT_LIST_HEAD(&newsig->posix_timers);
 615
 616                 newsig->tty = oldsig->tty;
 617                 newsig->pgrp = oldsig->pgrp;
 618                 newsig->session = oldsig->session;
 619                 newsig->leader = oldsig->leader;
 620                 newsig->tty_old_pgrp = oldsig->tty_old_pgrp;
 621         }
 622
 623         if (thread_group_empty(current))
 624                 goto no_thread_group;
 625
 626         /*
 627          * Kill all other threads in the thread group.
 628          * We must hold tasklist_lock to call zap_other_threads.
 629          */
 630         read_lock(&tasklist_lock);
 631         spin_lock_irq(lock);
 632         if (oldsig->group_exit) {
 633                 /*
 634                  * Another group action in progress, just
 635                  * return so that the signal is processed.
 636                  */
 637                 spin_unlock_irq(lock);
 638                 read_unlock(&tasklist_lock);
 639                 kmem_cache_free(sighand_cachep, newsighand);
 640                 if (newsig)
 641                         kmem_cache_free(signal_cachep, newsig);
 642                 return -EAGAIN;
 643         }
 644         oldsig->group_exit = 1;
 645         zap_other_threads(current);
 646         read_unlock(&tasklist_lock);
 647
 648         /*
 649          * Account for the thread group leader hanging around:
 650          */
 651         count = 2;
 652         if (current->pid == current->tgid)
 653                 count = 1;
 654         while (atomic_read(&oldsig->count) > count) {
 655                 oldsig->group_exit_task = current;
 656                 oldsig->notify_count = count;
 657                 __set_current_state(TASK_UNINTERRUPTIBLE);
 658                 spin_unlock_irq(lock);
 659                 schedule();
 660                 spin_lock_irq(lock);
 661         }
 662         spin_unlock_irq(lock);
 663
 664         /*
 665          * At this point all other threads have exited, all we have to
 666          * do is to wait for the thread group leader to become inactive,
 667          * and to assume its PID:
 668          */
 669         if (current->pid != current->tgid) {
 670                 struct task_struct *leader = current->group_leader, *parent;
 671                 struct dentry *proc_dentry1, *proc_dentry2;
 672                 unsigned long state, ptrace;
 673
 674                 /*
 675                  * Wait for the thread group leader to be a zombie.
 676                  * It should already be zombie at this point, most
 677                  * of the time.
 678                  */
 679                 while (leader->state != TASK_ZOMBIE)
 680                         yield();
 681
 682                 spin_lock(&leader->proc_lock);
 683                 spin_lock(&current->proc_lock);
 684                 proc_dentry1 = proc_pid_unhash(current);
 685                 proc_dentry2 = proc_pid_unhash(leader);
 686                 write_lock_irq(&tasklist_lock);
 687
 688                 if (leader->tgid != current->tgid)
 689                         BUG();
 690                 if (current->pid == current->tgid)
 691                         BUG();
 692                 /*
 693                  * An exec() starts a new thread group with the
 694                  * TGID of the previous thread group. Rehash the
 695                  * two threads with a switched PID, and release
 696                  * the former thread group leader:
 697                  */
 698                 ptrace = leader->ptrace;
 699                 parent = leader->parent;
 700
 701                 ptrace_unlink(current);
 702                 ptrace_unlink(leader);
 703                 remove_parent(current);
 704                 remove_parent(leader);
 705
 706                 switch_exec_pids(leader, current);
 707
 708                 current->parent = current->real_parent = leader->real_parent;
 709                 leader->parent = leader->real_parent = child_reaper;
 710                 current->group_leader = current;
 711                 leader->group_leader = leader;
 712
 713                 add_parent(current, current->parent);
 714                 add_parent(leader, leader->parent);
 715                 if (ptrace) {
 716                         current->ptrace = ptrace;
 717                         __ptrace_link(current, parent);
 718                 }
 719
 720                 list_del(&current->tasks);
 721                 list_add_tail(&current->tasks, &init_task.tasks);
 722                 current->exit_signal = SIGCHLD;
 723                 state = leader->state;
 724
 725                 write_unlock_irq(&tasklist_lock);
 726                 spin_unlock(&leader->proc_lock);
 727                 spin_unlock(&current->proc_lock);
 728                 proc_pid_flush(proc_dentry1);
 729                 proc_pid_flush(proc_dentry2);
 730
 731                 if (state != TASK_ZOMBIE)
 732                         BUG();
 733                 release_task(leader);
 734         }
 735
 736 no_thread_group:
 737
 738         write_lock_irq(&tasklist_lock);
 739         spin_lock(&oldsighand->siglock);
 740         spin_lock(&newsighand->siglock);
 741
 742         if (current == oldsig->curr_target)
 743                 oldsig->curr_target = next_thread(current);
 744         if (newsig)
 745                 current->signal = newsig;
 746         current->sighand = newsighand;
 747         init_sigpending(&current->pending);
 748         recalc_sigpending();
 749
 750         spin_unlock(&newsighand->siglock);
 751         spin_unlock(&oldsighand->siglock);
 752         write_unlock_irq(&tasklist_lock);
 753
 754         if (newsig && atomic_dec_and_test(&oldsig->count))
 755                 kmem_cache_free(signal_cachep, oldsig);
 756
 757         if (atomic_dec_and_test(&oldsighand->count))
 758                 kmem_cache_free(sighand_cachep, oldsighand);
 759
 760         if (!thread_group_empty(current))
 761                 BUG();
 762         if (current->tgid != current->pid)
 763                 BUG();
 764         return 0;
 765 }
 766
 767 /*
 768  * These functions flushes out all traces of the currently running executable
 769  * so that a new one can be started
 770  */
 771
 772 static inline void flush_old_files(struct files_struct * files)
 773 {
 774         long j = -1;
 775
 776         spin_lock(&files->file_lock);
 777         for (;;) {
 778                 unsigned long set, i;
 779
 780                 j++;
 781                 i = j * __NFDBITS;
 782                 if (i >= files->max_fds || i >= files->max_fdset)
 783                         break;
 784                 set = files->close_on_exec->fds_bits[j];
 785                 if (!set)
 786                         continue;
 787                 files->close_on_exec->fds_bits[j] = 0;
 788                 spin_unlock(&files->file_lock);
 789                 for ( ; set ; i++,set >>= 1) {
 790                         if (set & 1) {
 791                                 sys_close(i);
 792                         }
 793                 }
 794                 spin_lock(&files->file_lock);
 795
 796         }
 797         spin_unlock(&files->file_lock);
 798 }
 799
 800 int flush_old_exec(struct linux_binprm * bprm)
 801 {
 802         char * name;
 803         int i, ch, retval;
 804         struct files_struct *files;
 805
 806         /*
 807          * Make sure we have a private signal table and that
 808          * we are unassociated from the previous thread group.
 809          */
 810         retval = de_thread(current);
 811         if (retval)
 812                 goto out;
 813
 814         /*
 815          * Make sure we have private file handles. Ask the
 816          * fork helper to do the work for us and the exit
 817          * helper to do the cleanup of the old one.
 818          */
 819         files = current->files;         /* refcounted so safe to hold */
 820         retval = unshare_files();
 821         if (retval)
 822                 goto out;
 823         /*
 824          * Release all of the old mmap stuff
 825          */
 826         retval = exec_mmap(bprm->mm);
 827         if (retval)
 828                 goto mmap_failed;
 829
 830         bprm->mm = NULL;                /* We're using it now */
 831
 832         /* This is the point of no return */
 833         steal_locks(files);
 834         put_files_struct(files);
 835
 836         current->sas_ss_sp = current->sas_ss_size = 0;
 837
 838         if (current->euid == current->uid && current->egid == current->gid)
 839                 current->mm->dumpable = 1;
 840         name = bprm->filename;
 841         for (i=0; (ch = *(name++)) != '\0';) {
 842                 if (ch == '/')
 843                         i = 0;
 844                 else
 845                         if (i < 15)
 846                                 current->comm[i++] = ch;
 847         }
 848         current->comm[i] = '\0';
 849
 850         current->flags &= ~PF_RELOCEXEC;
 851         flush_thread();
 852
 853         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 854             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
 855             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP))
 856                 current->mm->dumpable = 0;
 857
 858         /* An exec changes our domain. We are no longer part of the thread
 859            group */
 860
 861         current->self_exec_id++;
 862
 863         flush_signal_handlers(current, 0);
 864         flush_old_files(current->files);
 865
 866         return 0;
 867
 868 mmap_failed:
 869         put_files_struct(current->files);
 870         current->files = files;
 871 out:
 872         return retval;
 873 }
 874
 875 EXPORT_SYMBOL(flush_old_exec);
 876
 877 /*
 878  * Fill the binprm structure from the inode.
 879  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 880  */
 881 int prepare_binprm(struct linux_binprm *bprm)
 882 {
 883         int mode;
 884         struct inode * inode = bprm->file->f_dentry->d_inode;
 885         int retval;
 886
 887         mode = inode->i_mode;
 888         /*
 889          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 890          * vfs_permission lets a non-executable through
 891          */
 892         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 893                 return -EACCES;
 894         if (bprm->file->f_op == NULL)
 895                 return -EACCES;
 896
 897         bprm->e_uid = current->euid;
 898         bprm->e_gid = current->egid;
 899
 900         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 901                 /* Set-uid? */
 902                 if (mode & S_ISUID) {
 903                         current->personality &= ~PER_CLEAR_ON_SETID;
 904                         bprm->e_uid = inode->i_uid;
 905                 }
 906
 907                 /* Set-gid? */
 908                 /*
 909                  * If setgid is set but no group execute bit then this
 910                  * is a candidate for mandatory locking, not a setgid
 911                  * executable.
 912                  */
 913                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 914                         current->personality &= ~PER_CLEAR_ON_SETID;
 915                         bprm->e_gid = inode->i_gid;
 916                 }
 917         }
 918
 919         /* fill in binprm security blob */
 920         retval = security_bprm_set(bprm);
 921         if (retval)
 922                 return retval;
 923
 924         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 925         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 926 }
 927
 928 EXPORT_SYMBOL(prepare_binprm);
 929
 930 static inline int unsafe_exec(struct task_struct *p)
 931 {
 932         int unsafe = 0;
 933         if (p->ptrace & PT_PTRACED) {
 934                 if (p->ptrace & PT_PTRACE_CAP)
 935                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 936                 else
 937                         unsafe |= LSM_UNSAFE_PTRACE;
 938         }
 939         if (atomic_read(&p->fs->count) > 1 ||
 940             atomic_read(&p->files->count) > 1 ||
 941             atomic_read(&p->sighand->count) > 1)
 942                 unsafe |= LSM_UNSAFE_SHARE;
 943
 944         return unsafe;
 945 }
 946
 947 void compute_creds(struct linux_binprm *bprm)
 948 {
 949         int unsafe;
 950         task_lock(current);
 951         unsafe = unsafe_exec(current);
 952         security_bprm_apply_creds(bprm, unsafe);
 953         task_unlock(current);
 954 }
 955
 956 EXPORT_SYMBOL(compute_creds);
 957
 958 void remove_arg_zero(struct linux_binprm *bprm)
 959 {
 960         if (bprm->argc) {
 961                 unsigned long offset;
 962                 char * kaddr;
 963                 struct page *page;
 964
 965                 offset = bprm->p % PAGE_SIZE;
 966                 goto inside;
 967
 968                 while (bprm->p++, *(kaddr+offset++)) {
 969                         if (offset != PAGE_SIZE)
 970                                 continue;
 971                         offset = 0;
 972                         kunmap_atomic(kaddr, KM_USER0);
 973 inside:
 974                         page = bprm->page[bprm->p/PAGE_SIZE];
 975                         kaddr = kmap_atomic(page, KM_USER0);
 976                 }
 977                 kunmap_atomic(kaddr, KM_USER0);
 978                 bprm->argc--;
 979         }
 980 }
 981
 982 EXPORT_SYMBOL(remove_arg_zero);
 983
 984 /*
 985  * cycle the list of binary formats handler, until one recognizes the image
 986  */
 987 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 988 {
 989         int try,retval=0;
 990         struct linux_binfmt *fmt;
 991 #ifdef __alpha__
 992         /* handle /sbin/loader.. */
 993         {
 994             struct exec * eh = (struct exec *) bprm->buf;
 995
 996             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 997                 (eh->fh.f_flags & 0x3000) == 0x3000)
 998             {
 999                 struct file * file;
1000                 unsigned long loader;
1001
1002                 allow_write_access(bprm->file);
1003                 fput(bprm->file);
1004                 bprm->file = NULL;
1005
1006                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1007
1008                 file = open_exec("/sbin/loader");
1009                 retval = PTR_ERR(file);
1010                 if (IS_ERR(file))
1011                         return retval;
1012
1013                 /* Remember if the application is TASO.  */
1014                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1015
1016                 bprm->file = file;
1017                 bprm->loader = loader;
1018                 retval = prepare_binprm(bprm);
1019                 if (retval<0)
1020                         return retval;
1021                 /* should call search_binary_handler recursively here,
1022                    but it does not matter */
1023             }
1024         }
1025 #endif
1026         retval = security_bprm_check(bprm);
1027         if (retval)
1028                 return retval;
1029
1030         /* kernel module loader fixup */
1031         /* so we don't try to load run modprobe in kernel space. */
1032         set_fs(USER_DS);
1033         for (try=0; try<2; try++) {
1034                 read_lock(&binfmt_lock);
1035                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1036                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1037                         if (!fn)
1038                                 continue;
1039                         if (!try_module_get(fmt->module))
1040                                 continue;
1041                         read_unlock(&binfmt_lock);
1042                         retval = fn(bprm, regs);
1043                         if (retval >= 0) {
1044                                 put_binfmt(fmt);
1045                                 allow_write_access(bprm->file);
1046                                 if (bprm->file)
1047                                         fput(bprm->file);
1048                                 bprm->file = NULL;
1049                                 current->did_exec = 1;
1050                                 ckrm_cb_exec(bprm->filename);
1051                                 return retval;
1052                         }
1053                         read_lock(&binfmt_lock);
1054                         put_binfmt(fmt);
1055                         if (retval != -ENOEXEC || bprm->mm == NULL)
1056                                 break;
1057                         if (!bprm->file) {
1058                                 read_unlock(&binfmt_lock);
1059                                 return retval;
1060                         }
1061                 }
1062                 read_unlock(&binfmt_lock);
1063                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1064                         break;
1065 #ifdef CONFIG_KMOD
1066                 }else{
1067 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1068                         if (printable(bprm->buf[0]) &&
1069                             printable(bprm->buf[1]) &&
1070                             printable(bprm->buf[2]) &&
1071                             printable(bprm->buf[3]))
1072                                 break; /* -ENOEXEC */
1073                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1074 #endif
1075                 }
1076         }
1077         return retval;
1078 }
1079
1080 EXPORT_SYMBOL(search_binary_handler);
1081
1082 /*
1083  * sys_execve() executes a new program.
1084  */
1085 int do_execve(char * filename,
1086         char __user *__user *argv,
1087         char __user *__user *envp,
1088         struct pt_regs * regs)
1089 {
1090         struct linux_binprm bprm;
1091         struct file *file;
1092         int retval;
1093         int i;
1094
1095         file = open_exec(filename);
1096
1097         retval = PTR_ERR(file);
1098         if (IS_ERR(file))
1099                 return retval;
1100
1101         sched_balance_exec();
1102
1103         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1104         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
1105
1106         bprm.file = file;
1107         bprm.filename = filename;
1108         bprm.interp = filename;
1109         bprm.interp_flags = 0;
1110         bprm.interp_data = 0;
1111         bprm.sh_bang = 0;
1112         bprm.loader = 0;
1113         bprm.exec = 0;
1114         bprm.security = NULL;
1115         bprm.mm = mm_alloc();
1116         retval = -ENOMEM;
1117         if (!bprm.mm)
1118                 goto out_file;
1119
1120         retval = init_new_context(current, bprm.mm);
1121         if (retval < 0)
1122                 goto out_mm;
1123
1124         bprm.argc = count(argv, bprm.p / sizeof(void *));
1125         if ((retval = bprm.argc) < 0)
1126                 goto out_mm;
1127
1128         bprm.envc = count(envp, bprm.p / sizeof(void *));
1129         if ((retval = bprm.envc) < 0)
1130                 goto out_mm;
1131
1132         retval = security_bprm_alloc(&bprm);
1133         if (retval)
1134                 goto out;
1135
1136         retval = prepare_binprm(&bprm);
1137         if (retval < 0)
1138                 goto out;
1139
1140         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
1141         if (retval < 0)
1142                 goto out;
1143
1144         bprm.exec = bprm.p;
1145         retval = copy_strings(bprm.envc, envp, &bprm);
1146         if (retval < 0)
1147                 goto out;
1148
1149         retval = copy_strings(bprm.argc, argv, &bprm);
1150         if (retval < 0)
1151                 goto out;
1152
1153         retval = search_binary_handler(&bprm,regs);
1154         if (retval >= 0) {
1155                 free_arg_pages(&bprm);
1156
1157                 /* execve success */
1158                 security_bprm_free(&bprm);
1159                 return retval;
1160         }
1161
1162 out:
1163         /* Something went wrong, return the inode and free the argument pages*/
1164         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1165                 struct page * page = bprm.page[i];
1166                 if (page)
1167                         __free_page(page);
1168         }
1169
1170         if (bprm.security)
1171                 security_bprm_free(&bprm);
1172
1173 out_mm:
1174         if (bprm.mm)
1175                 mmdrop(bprm.mm);
1176
1177 out_file:
1178         if (bprm.file) {
1179                 allow_write_access(bprm.file);
1180                 fput(bprm.file);
1181         }
1182         return retval;
1183 }
1184
1185 EXPORT_SYMBOL(do_execve);
1186
1187 int set_binfmt(struct linux_binfmt *new)
1188 {
1189         struct linux_binfmt *old = current->binfmt;
1190
1191         if (new) {
1192                 if (!try_module_get(new->module))
1193                         return -1;
1194         }
1195         current->binfmt = new;
1196         if (old)
1197                 module_put(old->module);
1198         return 0;
1199 }
1200
1201 EXPORT_SYMBOL(set_binfmt);
1202
1203 #define CORENAME_MAX_SIZE 64
1204
1205 /* format_corename will inspect the pattern parameter, and output a
1206  * name into corename, which must have space for at least
1207  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1208  */
1209 void format_corename(char *corename, const char *pattern, long signr)
1210 {
1211         const char *pat_ptr = pattern;
1212         char *out_ptr = corename;
1213         char *const out_end = corename + CORENAME_MAX_SIZE;
1214         int rc;
1215         int pid_in_pattern = 0;
1216
1217         /* Repeat as long as we have more pattern to process and more output
1218            space */
1219         while (*pat_ptr) {
1220                 if (*pat_ptr != '%') {
1221                         if (out_ptr == out_end)
1222                                 goto out;
1223                         *out_ptr++ = *pat_ptr++;
1224                 } else {
1225                         switch (*++pat_ptr) {
1226                         case 0:
1227                                 goto out;
1228                         /* Double percent, output one percent */
1229                         case '%':
1230                                 if (out_ptr == out_end)
1231                                         goto out;
1232                                 *out_ptr++ = '%';
1233                                 break;
1234                         /* pid */
1235                         case 'p':
1236                                 pid_in_pattern = 1;
1237                                 rc = snprintf(out_ptr, out_end - out_ptr,
1238                                               "%d", current->tgid);
1239                                 if (rc > out_end - out_ptr)
1240                                         goto out;
1241                                 out_ptr += rc;
1242                                 break;
1243                         /* uid */
1244                         case 'u':
1245                                 rc = snprintf(out_ptr, out_end - out_ptr,
1246                                               "%d", current->uid);
1247                                 if (rc > out_end - out_ptr)
1248                                         goto out;
1249                                 out_ptr += rc;
1250                                 break;
1251                         /* gid */
1252                         case 'g':
1253                                 rc = snprintf(out_ptr, out_end - out_ptr,
1254                                               "%d", current->gid);
1255                                 if (rc > out_end - out_ptr)
1256                                         goto out;
1257                                 out_ptr += rc;
1258                                 break;
1259                         /* signal that caused the coredump */
1260                         case 's':
1261                                 rc = snprintf(out_ptr, out_end - out_ptr,
1262                                               "%ld", signr);
1263                                 if (rc > out_end - out_ptr)
1264                                         goto out;
1265                                 out_ptr += rc;
1266                                 break;
1267                         /* UNIX time of coredump */
1268                         case 't': {
1269                                 struct timeval tv;
1270                                 do_gettimeofday(&tv);
1271                                 rc = snprintf(out_ptr, out_end - out_ptr,
1272                                               "%lu", tv.tv_sec);
1273                                 if (rc > out_end - out_ptr)
1274                                         goto out;
1275                                 out_ptr += rc;
1276                                 break;
1277                         }
1278                         /* hostname */
1279                         case 'h':
1280                                 down_read(&uts_sem);
1281                                 rc = snprintf(out_ptr, out_end - out_ptr,
1282                                               "%s", system_utsname.nodename);
1283                                 up_read(&uts_sem);
1284                                 if (rc > out_end - out_ptr)
1285                                         goto out;
1286                                 out_ptr += rc;
1287                                 break;
1288                         /* executable */
1289                         case 'e':
1290                                 rc = snprintf(out_ptr, out_end - out_ptr,
1291                                               "%s", current->comm);
1292                                 if (rc > out_end - out_ptr)
1293                                         goto out;
1294                                 out_ptr += rc;
1295                                 break;
1296                         default:
1297                                 break;
1298                         }
1299                         ++pat_ptr;
1300                 }
1301         }
1302         /* Backward compatibility with core_uses_pid:
1303          *
1304          * If core_pattern does not include a %p (as is the default)
1305          * and core_uses_pid is set, then .%pid will be appended to
1306          * the filename */
1307         if (!pid_in_pattern
1308             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1309                 rc = snprintf(out_ptr, out_end - out_ptr,
1310                               ".%d", current->tgid);
1311                 if (rc > out_end - out_ptr)
1312                         goto out;
1313                 out_ptr += rc;
1314         }
1315       out:
1316         *out_ptr = 0;
1317 }
1318
1319 static void zap_threads (struct mm_struct *mm)
1320 {
1321         struct task_struct *g, *p;
1322         struct task_struct *tsk = current;
1323         struct completion *vfork_done = tsk->vfork_done;
1324
1325         /*
1326          * Make sure nobody is waiting for us to release the VM,
1327          * otherwise we can deadlock when we wait on each other
1328          */
1329         if (vfork_done) {
1330                 tsk->vfork_done = NULL;
1331                 complete(vfork_done);
1332         }
1333
1334         read_lock(&tasklist_lock);
1335         do_each_thread(g,p)
1336                 if (mm == p->mm && p != tsk) {
1337                         force_sig_specific(SIGKILL, p);
1338                         mm->core_waiters++;
1339                 }
1340         while_each_thread(g,p);
1341
1342         read_unlock(&tasklist_lock);
1343 }
1344
1345 static void coredump_wait(struct mm_struct *mm)
1346 {
1347         DECLARE_COMPLETION(startup_done);
1348
1349         mm->core_waiters++; /* let other threads block */
1350         mm->core_startup_done = &startup_done;
1351
1352         /* give other threads a chance to run: */
1353         yield();
1354
1355         zap_threads(mm);
1356         if (--mm->core_waiters) {
1357                 up_write(&mm->mmap_sem);
1358                 wait_for_completion(&startup_done);
1359         } else
1360                 up_write(&mm->mmap_sem);
1361         BUG_ON(mm->core_waiters);
1362 }
1363
1364 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1365 {
1366         char corename[CORENAME_MAX_SIZE + 1];
1367         struct mm_struct *mm = current->mm;
1368         struct linux_binfmt * binfmt;
1369         struct inode * inode;
1370         struct file * file;
1371         int retval = 0;
1372
1373         lock_kernel();
1374         binfmt = current->binfmt;
1375         if (!binfmt || !binfmt->core_dump)
1376                 goto fail;
1377         down_write(&mm->mmap_sem);
1378         if (!mm->dumpable) {
1379                 up_write(&mm->mmap_sem);
1380                 goto fail;
1381         }
1382         mm->dumpable = 0;
1383         init_completion(&mm->core_done);
1384         current->signal->group_exit = 1;
1385         current->signal->group_exit_code = exit_code;
1386         coredump_wait(mm);
1387
1388         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1389                 goto fail_unlock;
1390
1391         format_corename(corename, core_pattern, signr);
1392         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1393         if (IS_ERR(file))
1394                 goto fail_unlock;
1395         inode = file->f_dentry->d_inode;
1396         if (inode->i_nlink > 1)
1397                 goto close_fail;        /* multiple links - don't dump */
1398         if (d_unhashed(file->f_dentry))
1399                 goto close_fail;
1400
1401         if (!S_ISREG(inode->i_mode))
1402                 goto close_fail;
1403         if (!file->f_op)
1404                 goto close_fail;
1405         if (!file->f_op->write)
1406                 goto close_fail;
1407         if (do_truncate(file->f_dentry, 0) != 0)
1408                 goto close_fail;
1409
1410         retval = binfmt->core_dump(signr, regs, file);
1411
1412         current->signal->group_exit_code |= 0x80;
1413 close_fail:
1414         filp_close(file, NULL);
1415 fail_unlock:
1416         complete_all(&mm->core_done);
1417 fail:
1418         unlock_kernel();
1419         return retval;
1420 }