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
  50 #include <asm/uaccess.h>
  51 #include <asm/mmu_context.h>
  52
  53 #ifdef CONFIG_KMOD
  54 #include <linux/kmod.h>
  55 #endif
  56
  57 int core_uses_pid;
  58 char core_pattern[65] = "core";
  59 /* The maximal length of core_pattern is also specified in sysctl.c */
  60
  61 static struct linux_binfmt *formats;
  62 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  63
  64 int register_binfmt(struct linux_binfmt * fmt)
  65 {
  66         struct linux_binfmt ** tmp = &formats;
  67
  68         if (!fmt)
  69                 return -EINVAL;
  70         if (fmt->next)
  71                 return -EBUSY;
  72         write_lock(&binfmt_lock);
  73         while (*tmp) {
  74                 if (fmt == *tmp) {
  75                         write_unlock(&binfmt_lock);
  76                         return -EBUSY;
  77                 }
  78                 tmp = &(*tmp)->next;
  79         }
  80         fmt->next = formats;
  81         formats = fmt;
  82         write_unlock(&binfmt_lock);
  83         return 0;
  84 }
  85
  86 EXPORT_SYMBOL(register_binfmt);
  87
  88 int unregister_binfmt(struct linux_binfmt * fmt)
  89 {
  90         struct linux_binfmt ** tmp = &formats;
  91
  92         write_lock(&binfmt_lock);
  93         while (*tmp) {
  94                 if (fmt == *tmp) {
  95                         *tmp = fmt->next;
  96                         write_unlock(&binfmt_lock);
  97                         return 0;
  98                 }
  99                 tmp = &(*tmp)->next;
 100         }
 101         write_unlock(&binfmt_lock);
 102         return -EINVAL;
 103 }
 104
 105 EXPORT_SYMBOL(unregister_binfmt);
 106
 107 static inline void put_binfmt(struct linux_binfmt * fmt)
 108 {
 109         module_put(fmt->module);
 110 }
 111
 112 /*
 113  * Note that a shared library must be both readable and executable due to
 114  * security reasons.
 115  *
 116  * Also note that we take the address to load from from the file itself.
 117  */
 118 asmlinkage long sys_uselib(const char __user * library)
 119 {
 120         struct file * file;
 121         struct nameidata nd;
 122         int error;
 123
 124         nd.intent.open.flags = FMODE_READ;
 125         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 126         if (error)
 127                 goto out;
 128
 129         error = -EINVAL;
 130         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 131                 goto exit;
 132
 133         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 134         if (error)
 135                 goto exit;
 136
 137         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 138         error = PTR_ERR(file);
 139         if (IS_ERR(file))
 140                 goto out;
 141
 142         error = -ENOEXEC;
 143         if(file->f_op) {
 144                 struct linux_binfmt * fmt;
 145
 146                 read_lock(&binfmt_lock);
 147                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 148                         if (!fmt->load_shlib)
 149                                 continue;
 150                         if (!try_module_get(fmt->module))
 151                                 continue;
 152                         read_unlock(&binfmt_lock);
 153                         error = fmt->load_shlib(file);
 154                         read_lock(&binfmt_lock);
 155                         put_binfmt(fmt);
 156                         if (error != -ENOEXEC)
 157                                 break;
 158                 }
 159                 read_unlock(&binfmt_lock);
 160         }
 161         fput(file);
 162 out:
 163         return error;
 164 exit:
 165         path_release(&nd);
 166         goto out;
 167 }
 168
 169 /*
 170  * count() counts the number of strings in array ARGV.
 171  */
 172 static int count(char __user * __user * argv, int max)
 173 {
 174         int i = 0;
 175
 176         if (argv != NULL) {
 177                 for (;;) {
 178                         char __user * p;
 179
 180                         if (get_user(p, argv))
 181                                 return -EFAULT;
 182                         if (!p)
 183                                 break;
 184                         argv++;
 185                         if(++i > max)
 186                                 return -E2BIG;
 187                 }
 188         }
 189         return i;
 190 }
 191
 192 /*
 193  * 'copy_strings()' copies argument/environment strings from user
 194  * memory to free pages in kernel mem. These are in a format ready
 195  * to be put directly into the top of new user memory.
 196  */
 197 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 198 {
 199         struct page *kmapped_page = NULL;
 200         char *kaddr = NULL;
 201         int ret;
 202
 203         while (argc-- > 0) {
 204                 char __user *str;
 205                 int len;
 206                 unsigned long pos;
 207
 208                 if (get_user(str, argv+argc) ||
 209                                 !(len = strnlen_user(str, bprm->p))) {
 210                         ret = -EFAULT;
 211                         goto out;
 212                 }
 213
 214                 if (bprm->p < len)  {
 215                         ret = -E2BIG;
 216                         goto out;
 217                 }
 218
 219                 bprm->p -= len;
 220                 /* XXX: add architecture specific overflow check here. */
 221                 pos = bprm->p;
 222
 223                 while (len > 0) {
 224                         int i, new, err;
 225                         int offset, bytes_to_copy;
 226                         struct page *page;
 227
 228                         offset = pos % PAGE_SIZE;
 229                         i = pos/PAGE_SIZE;
 230                         page = bprm->page[i];
 231                         new = 0;
 232                         if (!page) {
 233                                 page = alloc_page(GFP_HIGHUSER);
 234                                 bprm->page[i] = page;
 235                                 if (!page) {
 236                                         ret = -ENOMEM;
 237                                         goto out;
 238                                 }
 239                                 new = 1;
 240                         }
 241
 242                         if (page != kmapped_page) {
 243                                 if (kmapped_page)
 244                                         kunmap(kmapped_page);
 245                                 kmapped_page = page;
 246                                 kaddr = kmap(kmapped_page);
 247                         }
 248                         if (new && offset)
 249                                 memset(kaddr, 0, offset);
 250                         bytes_to_copy = PAGE_SIZE - offset;
 251                         if (bytes_to_copy > len) {
 252                                 bytes_to_copy = len;
 253                                 if (new)
 254                                         memset(kaddr+offset+len, 0,
 255                                                 PAGE_SIZE-offset-len);
 256                         }
 257                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 258                         if (err) {
 259                                 ret = -EFAULT;
 260                                 goto out;
 261                         }
 262
 263                         pos += bytes_to_copy;
 264                         str += bytes_to_copy;
 265                         len -= bytes_to_copy;
 266                 }
 267         }
 268         ret = 0;
 269 out:
 270         if (kmapped_page)
 271                 kunmap(kmapped_page);
 272         return ret;
 273 }
 274
 275 /*
 276  * Like copy_strings, but get argv and its values from kernel memory.
 277  */
 278 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 279 {
 280         int r;
 281         mm_segment_t oldfs = get_fs();
 282         set_fs(KERNEL_DS);
 283         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 284         set_fs(oldfs);
 285         return r;
 286 }
 287
 288 EXPORT_SYMBOL(copy_strings_kernel);
 289
 290 #ifdef CONFIG_MMU
 291 /*
 292  * This routine is used to map in a page into an address space: needed by
 293  * execve() for the initial stack and environment pages.
 294  *
 295  * vma->vm_mm->mmap_sem is held for writing.
 296  */
 297 void install_arg_page(struct vm_area_struct *vma,
 298                         struct page *page, unsigned long address)
 299 {
 300         struct mm_struct *mm = vma->vm_mm;
 301         pgd_t * pgd;
 302         pmd_t * pmd;
 303         pte_t * pte;
 304
 305         if (unlikely(anon_vma_prepare(vma)))
 306                 goto out_sig;
 307
 308         flush_dcache_page(page);
 309         pgd = pgd_offset(mm, address);
 310
 311         spin_lock(&mm->page_table_lock);
 312         pmd = pmd_alloc(mm, pgd, address);
 313         if (!pmd)
 314                 goto out;
 315         pte = pte_alloc_map(mm, pmd, address);
 316         if (!pte)
 317                 goto out;
 318         if (!pte_none(*pte)) {
 319                 pte_unmap(pte);
 320                 goto out;
 321         }
 322         mm->rss++;
 323         lru_cache_add_active(page);
 324         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
 325                                         page, vma->vm_page_prot))));
 326         page_add_anon_rmap(page, vma, address);
 327         pte_unmap(pte);
 328         spin_unlock(&mm->page_table_lock);
 329
 330         /* no need for flush_tlb */
 331         return;
 332 out:
 333         spin_unlock(&mm->page_table_lock);
 334 out_sig:
 335         __free_page(page);
 336         force_sig(SIGKILL, current);
 337 }
 338
 339 int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
 340 {
 341         unsigned long stack_base;
 342         struct vm_area_struct *mpnt;
 343         struct mm_struct *mm = current->mm;
 344         int i;
 345         long arg_size;
 346
 347 #ifdef CONFIG_STACK_GROWSUP
 348         /* Move the argument and environment strings to the bottom of the
 349          * stack space.
 350          */
 351         int offset, j;
 352         char *to, *from;
 353
 354         /* Start by shifting all the pages down */
 355         i = 0;
 356         for (j = 0; j < MAX_ARG_PAGES; j++) {
 357                 struct page *page = bprm->page[j];
 358                 if (!page)
 359                         continue;
 360                 bprm->page[i++] = page;
 361         }
 362
 363         /* Now move them within their pages */
 364         offset = bprm->p % PAGE_SIZE;
 365         to = kmap(bprm->page[0]);
 366         for (j = 1; j < i; j++) {
 367                 memmove(to, to + offset, PAGE_SIZE - offset);
 368                 from = kmap(bprm->page[j]);
 369                 memcpy(to + PAGE_SIZE - offset, from, offset);
 370                 kunmap(bprm->page[j - 1]);
 371                 to = from;
 372         }
 373         memmove(to, to + offset, PAGE_SIZE - offset);
 374         kunmap(bprm->page[j - 1]);
 375
 376         /* Adjust bprm->p to point to the end of the strings. */
 377         bprm->p = PAGE_SIZE * i - offset;
 378
 379         /* Limit stack size to 1GB */
 380         stack_base = current->rlim[RLIMIT_STACK].rlim_max;
 381         if (stack_base > (1 << 30))
 382                 stack_base = 1 << 30;
 383         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
 384
 385         mm->arg_start = stack_base;
 386         arg_size = i << PAGE_SHIFT;
 387
 388         /* zero pages that were copied above */
 389         while (i < MAX_ARG_PAGES)
 390                 bprm->page[i++] = NULL;
 391 #else
 392 #ifdef __HAVE_ARCH_ALIGN_STACK
 393         stack_base = arch_align_stack(STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE);
 394         stack_base = PAGE_ALIGN(stack_base);
 395 #else
 396         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 397 #endif
 398         mm->arg_start = bprm->p + stack_base;
 399         arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
 400 #endif
 401
 402         bprm->p += stack_base;
 403         if (bprm->loader)
 404                 bprm->loader += stack_base;
 405         bprm->exec += stack_base;
 406
 407         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 408         if (!mpnt)
 409                 return -ENOMEM;
 410
 411         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT)) {
 412                 kmem_cache_free(vm_area_cachep, mpnt);
 413                 return -ENOMEM;
 414         }
 415
 416         memset(mpnt, 0, sizeof(*mpnt));
 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_flags |= mm->def_flags;
 439                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 440                 insert_vm_struct(mm, mpnt);
 441                 mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 442         }
 443
 444         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 445                 struct page *page = bprm->page[i];
 446                 if (page) {
 447                         bprm->page[i] = NULL;
 448                         install_arg_page(mpnt, page, stack_base);
 449                 }
 450                 stack_base += PAGE_SIZE;
 451         }
 452         up_write(&mm->mmap_sem);
 453
 454         return 0;
 455 }
 456
 457 EXPORT_SYMBOL(setup_arg_pages);
 458
 459 #define free_arg_pages(bprm) do { } while (0)
 460
 461 #else
 462
 463 static inline void free_arg_pages(struct linux_binprm *bprm)
 464 {
 465         int i;
 466
 467         for (i = 0; i < MAX_ARG_PAGES; i++) {
 468                 if (bprm->page[i])
 469                         __free_page(bprm->page[i]);
 470                 bprm->page[i] = NULL;
 471         }
 472 }
 473
 474 #endif /* CONFIG_MMU */
 475
 476 struct file *open_exec(const char *name)
 477 {
 478         struct nameidata nd;
 479         int err;
 480         struct file *file;
 481
 482         nd.intent.open.flags = FMODE_READ;
 483         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 484         file = ERR_PTR(err);
 485
 486         if (!err) {
 487                 struct inode *inode = nd.dentry->d_inode;
 488                 file = ERR_PTR(-EACCES);
 489                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 490                     S_ISREG(inode->i_mode)) {
 491                         int err = permission(inode, MAY_EXEC, &nd);
 492                         if (!err && !(inode->i_mode & 0111))
 493                                 err = -EACCES;
 494                         file = ERR_PTR(err);
 495                         if (!err) {
 496                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 497                                 if (!IS_ERR(file)) {
 498                                         err = deny_write_access(file);
 499                                         if (err) {
 500                                                 fput(file);
 501                                                 file = ERR_PTR(err);
 502                                         }
 503                                 }
 504 out:
 505                                 return file;
 506                         }
 507                 }
 508                 path_release(&nd);
 509         }
 510         goto out;
 511 }
 512
 513 EXPORT_SYMBOL(open_exec);
 514
 515 int kernel_read(struct file *file, unsigned long offset,
 516         char *addr, unsigned long count)
 517 {
 518         mm_segment_t old_fs;
 519         loff_t pos = offset;
 520         int result;
 521
 522         old_fs = get_fs();
 523         set_fs(get_ds());
 524         /* The cast to a user pointer is valid due to the set_fs() */
 525         result = vfs_read(file, (void __user *)addr, count, &pos);
 526         set_fs(old_fs);
 527         return result;
 528 }
 529
 530 EXPORT_SYMBOL(kernel_read);
 531
 532 static int exec_mmap(struct mm_struct *mm)
 533 {
 534         struct task_struct *tsk;
 535         struct mm_struct * old_mm, *active_mm;
 536
 537         /* Add it to the list of mm's */
 538         spin_lock(&mmlist_lock);
 539         list_add(&mm->mmlist, &init_mm.mmlist);
 540         mmlist_nr++;
 541         spin_unlock(&mmlist_lock);
 542
 543         /* Notify parent that we're no longer interested in the old VM */
 544         tsk = current;
 545         old_mm = current->mm;
 546         mm_release(tsk, old_mm);
 547
 548         task_lock(tsk);
 549         active_mm = tsk->active_mm;
 550         tsk->mm = mm;
 551         tsk->active_mm = mm;
 552         activate_mm(active_mm, mm);
 553         task_unlock(tsk);
 554         if (old_mm) {
 555                 if (active_mm != old_mm) BUG();
 556                 mmput(old_mm);
 557                 return 0;
 558         }
 559         mmdrop(active_mm);
 560         return 0;
 561 }
 562
 563 /*
 564  * This function makes sure the current process has its own signal table,
 565  * so that flush_signal_handlers can later reset the handlers without
 566  * disturbing other processes.  (Other processes might share the signal
 567  * table via the CLONE_SIGHAND option to clone().)
 568  */
 569 static inline int de_thread(struct task_struct *tsk)
 570 {
 571         struct signal_struct *newsig, *oldsig = tsk->signal;
 572         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 573         spinlock_t *lock = &oldsighand->siglock;
 574         int count;
 575
 576         /*
 577          * If we don't share sighandlers, then we aren't sharing anything
 578          * and we can just re-use it all.
 579          */
 580         if (atomic_read(&oldsighand->count) <= 1)
 581                 return 0;
 582
 583         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 584         if (!newsighand)
 585                 return -ENOMEM;
 586
 587         spin_lock_init(&newsighand->siglock);
 588         atomic_set(&newsighand->count, 1);
 589         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
 590
 591         /*
 592          * See if we need to allocate a new signal structure
 593          */
 594         newsig = NULL;
 595         if (atomic_read(&oldsig->count) > 1) {
 596                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 597                 if (!newsig) {
 598                         kmem_cache_free(sighand_cachep, newsighand);
 599                         return -ENOMEM;
 600                 }
 601                 atomic_set(&newsig->count, 1);
 602                 newsig->group_exit = 0;
 603                 newsig->group_exit_code = 0;
 604                 newsig->group_exit_task = NULL;
 605                 newsig->group_stop_count = 0;
 606                 newsig->curr_target = NULL;
 607                 init_sigpending(&newsig->shared_pending);
 608                 INIT_LIST_HEAD(&newsig->posix_timers);
 609
 610                 newsig->tty = oldsig->tty;
 611                 newsig->pgrp = oldsig->pgrp;
 612                 newsig->session = oldsig->session;
 613                 newsig->leader = oldsig->leader;
 614                 newsig->tty_old_pgrp = oldsig->tty_old_pgrp;
 615         }
 616
 617         if (thread_group_empty(current))
 618                 goto no_thread_group;
 619
 620         /*
 621          * Kill all other threads in the thread group.
 622          * We must hold tasklist_lock to call zap_other_threads.
 623          */
 624         read_lock(&tasklist_lock);
 625         spin_lock_irq(lock);
 626         if (oldsig->group_exit) {
 627                 /*
 628                  * Another group action in progress, just
 629                  * return so that the signal is processed.
 630                  */
 631                 spin_unlock_irq(lock);
 632                 read_unlock(&tasklist_lock);
 633                 kmem_cache_free(sighand_cachep, newsighand);
 634                 if (newsig)
 635                         kmem_cache_free(signal_cachep, newsig);
 636                 return -EAGAIN;
 637         }
 638         oldsig->group_exit = 1;
 639         zap_other_threads(current);
 640         read_unlock(&tasklist_lock);
 641
 642         /*
 643          * Account for the thread group leader hanging around:
 644          */
 645         count = 2;
 646         if (current->pid == current->tgid)
 647                 count = 1;
 648         while (atomic_read(&oldsig->count) > count) {
 649                 oldsig->group_exit_task = current;
 650                 oldsig->notify_count = count;
 651                 __set_current_state(TASK_UNINTERRUPTIBLE);
 652                 spin_unlock_irq(lock);
 653                 schedule();
 654                 spin_lock_irq(lock);
 655         }
 656         spin_unlock_irq(lock);
 657
 658         /*
 659          * At this point all other threads have exited, all we have to
 660          * do is to wait for the thread group leader to become inactive,
 661          * and to assume its PID:
 662          */
 663         if (current->pid != current->tgid) {
 664                 struct task_struct *leader = current->group_leader, *parent;
 665                 struct dentry *proc_dentry1, *proc_dentry2;
 666                 unsigned long state, ptrace;
 667
 668                 /*
 669                  * Wait for the thread group leader to be a zombie.
 670                  * It should already be zombie at this point, most
 671                  * of the time.
 672                  */
 673                 while (leader->state != TASK_ZOMBIE)
 674                         yield();
 675
 676                 spin_lock(&leader->proc_lock);
 677                 spin_lock(&current->proc_lock);
 678                 proc_dentry1 = proc_pid_unhash(current);
 679                 proc_dentry2 = proc_pid_unhash(leader);
 680                 write_lock_irq(&tasklist_lock);
 681
 682                 if (leader->tgid != current->tgid)
 683                         BUG();
 684                 if (current->pid == current->tgid)
 685                         BUG();
 686                 /*
 687                  * An exec() starts a new thread group with the
 688                  * TGID of the previous thread group. Rehash the
 689                  * two threads with a switched PID, and release
 690                  * the former thread group leader:
 691                  */
 692                 ptrace = leader->ptrace;
 693                 parent = leader->parent;
 694
 695                 ptrace_unlink(current);
 696                 ptrace_unlink(leader);
 697                 remove_parent(current);
 698                 remove_parent(leader);
 699
 700                 switch_exec_pids(leader, current);
 701
 702                 current->parent = current->real_parent = leader->real_parent;
 703                 leader->parent = leader->real_parent = child_reaper;
 704                 current->group_leader = current;
 705                 leader->group_leader = leader;
 706
 707                 add_parent(current, current->parent);
 708                 add_parent(leader, leader->parent);
 709                 if (ptrace) {
 710                         current->ptrace = ptrace;
 711                         __ptrace_link(current, parent);
 712                 }
 713
 714                 list_del(&current->tasks);
 715                 list_add_tail(&current->tasks, &init_task.tasks);
 716                 current->exit_signal = SIGCHLD;
 717                 state = leader->state;
 718
 719                 write_unlock_irq(&tasklist_lock);
 720                 spin_unlock(&leader->proc_lock);
 721                 spin_unlock(&current->proc_lock);
 722                 proc_pid_flush(proc_dentry1);
 723                 proc_pid_flush(proc_dentry2);
 724
 725                 if (state != TASK_ZOMBIE)
 726                         BUG();
 727                 release_task(leader);
 728         }
 729
 730 no_thread_group:
 731
 732         write_lock_irq(&tasklist_lock);
 733         spin_lock(&oldsighand->siglock);
 734         spin_lock(&newsighand->siglock);
 735
 736         if (current == oldsig->curr_target)
 737                 oldsig->curr_target = next_thread(current);
 738         if (newsig)
 739                 current->signal = newsig;
 740         current->sighand = newsighand;
 741         init_sigpending(&current->pending);
 742         recalc_sigpending();
 743
 744         spin_unlock(&newsighand->siglock);
 745         spin_unlock(&oldsighand->siglock);
 746         write_unlock_irq(&tasklist_lock);
 747
 748         if (newsig && atomic_dec_and_test(&oldsig->count))
 749                 kmem_cache_free(signal_cachep, oldsig);
 750
 751         if (atomic_dec_and_test(&oldsighand->count))
 752                 kmem_cache_free(sighand_cachep, oldsighand);
 753
 754         if (!thread_group_empty(current))
 755                 BUG();
 756         if (current->tgid != current->pid)
 757                 BUG();
 758         return 0;
 759 }
 760
 761 /*
 762  * These functions flushes out all traces of the currently running executable
 763  * so that a new one can be started
 764  */
 765
 766 static inline void flush_old_files(struct files_struct * files)
 767 {
 768         long j = -1;
 769
 770         spin_lock(&files->file_lock);
 771         for (;;) {
 772                 unsigned long set, i;
 773
 774                 j++;
 775                 i = j * __NFDBITS;
 776                 if (i >= files->max_fds || i >= files->max_fdset)
 777                         break;
 778                 set = files->close_on_exec->fds_bits[j];
 779                 if (!set)
 780                         continue;
 781                 files->close_on_exec->fds_bits[j] = 0;
 782                 spin_unlock(&files->file_lock);
 783                 for ( ; set ; i++,set >>= 1) {
 784                         if (set & 1) {
 785                                 sys_close(i);
 786                         }
 787                 }
 788                 spin_lock(&files->file_lock);
 789
 790         }
 791         spin_unlock(&files->file_lock);
 792 }
 793
 794 int flush_old_exec(struct linux_binprm * bprm)
 795 {
 796         char * name;
 797         int i, ch, retval;
 798         struct files_struct *files;
 799
 800         /*
 801          * Make sure we have a private signal table and that
 802          * we are unassociated from the previous thread group.
 803          */
 804         retval = de_thread(current);
 805         if (retval)
 806                 goto out;
 807
 808         /*
 809          * Make sure we have private file handles. Ask the
 810          * fork helper to do the work for us and the exit
 811          * helper to do the cleanup of the old one.
 812          */
 813         files = current->files;         /* refcounted so safe to hold */
 814         retval = unshare_files();
 815         if (retval)
 816                 goto out;
 817         /*
 818          * Release all of the old mmap stuff
 819          */
 820         retval = exec_mmap(bprm->mm);
 821         if (retval)
 822                 goto mmap_failed;
 823
 824         bprm->mm = NULL;                /* We're using it now */
 825
 826         /* This is the point of no return */
 827         steal_locks(files);
 828         put_files_struct(files);
 829
 830         current->sas_ss_sp = current->sas_ss_size = 0;
 831
 832         if (current->euid == current->uid && current->egid == current->gid)
 833                 current->mm->dumpable = 1;
 834         name = bprm->filename;
 835         for (i=0; (ch = *(name++)) != '\0';) {
 836                 if (ch == '/')
 837                         i = 0;
 838                 else
 839                         if (i < 15)
 840                                 current->comm[i++] = ch;
 841         }
 842         current->comm[i] = '\0';
 843
 844         current->flags &= ~PF_RELOCEXEC;
 845         flush_thread();
 846
 847         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 848             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
 849             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP))
 850                 current->mm->dumpable = 0;
 851
 852         /* An exec changes our domain. We are no longer part of the thread
 853            group */
 854
 855         current->self_exec_id++;
 856
 857         flush_signal_handlers(current, 0);
 858         flush_old_files(current->files);
 859
 860         return 0;
 861
 862 mmap_failed:
 863         put_files_struct(current->files);
 864         current->files = files;
 865 out:
 866         return retval;
 867 }
 868
 869 EXPORT_SYMBOL(flush_old_exec);
 870
 871 /*
 872  * Fill the binprm structure from the inode.
 873  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 874  */
 875 int prepare_binprm(struct linux_binprm *bprm)
 876 {
 877         int mode;
 878         struct inode * inode = bprm->file->f_dentry->d_inode;
 879         int retval;
 880
 881         mode = inode->i_mode;
 882         /*
 883          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 884          * vfs_permission lets a non-executable through
 885          */
 886         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 887                 return -EACCES;
 888         if (bprm->file->f_op == NULL)
 889                 return -EACCES;
 890
 891         bprm->e_uid = current->euid;
 892         bprm->e_gid = current->egid;
 893
 894         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 895                 /* Set-uid? */
 896                 if (mode & S_ISUID) {
 897                         bprm->e_uid = inode->i_uid;
 898 #ifdef __i386__
 899                         /* reset personality */
 900                         current->personality = PER_LINUX;
 901 #endif
 902                 }
 903
 904                 /* Set-gid? */
 905                 /*
 906                  * If setgid is set but no group execute bit then this
 907                  * is a candidate for mandatory locking, not a setgid
 908                  * executable.
 909                  */
 910                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 911                         bprm->e_gid = inode->i_gid;
 912 #ifdef __i386__
 913                         /* reset personality */
 914                         current->personality = PER_LINUX;
 915 #endif
 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                                 return retval;
1051                         }
1052                         read_lock(&binfmt_lock);
1053                         put_binfmt(fmt);
1054                         if (retval != -ENOEXEC || bprm->mm == NULL)
1055                                 break;
1056                         if (!bprm->file) {
1057                                 read_unlock(&binfmt_lock);
1058                                 return retval;
1059                         }
1060                 }
1061                 read_unlock(&binfmt_lock);
1062                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1063                         break;
1064 #ifdef CONFIG_KMOD
1065                 }else{
1066 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1067                         if (printable(bprm->buf[0]) &&
1068                             printable(bprm->buf[1]) &&
1069                             printable(bprm->buf[2]) &&
1070                             printable(bprm->buf[3]))
1071                                 break; /* -ENOEXEC */
1072                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1073 #endif
1074                 }
1075         }
1076         return retval;
1077 }
1078
1079 EXPORT_SYMBOL(search_binary_handler);
1080
1081 /*
1082  * sys_execve() executes a new program.
1083  */
1084 int do_execve(char * filename,
1085         char __user *__user *argv,
1086         char __user *__user *envp,
1087         struct pt_regs * regs)
1088 {
1089         struct linux_binprm bprm;
1090         struct file *file;
1091         int retval;
1092         int i;
1093
1094         file = open_exec(filename);
1095
1096         retval = PTR_ERR(file);
1097         if (IS_ERR(file))
1098                 return retval;
1099
1100         sched_balance_exec();
1101
1102         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1103         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
1104
1105         bprm.file = file;
1106         bprm.filename = filename;
1107         bprm.interp = filename;
1108         bprm.interp_flags = 0;
1109         bprm.interp_data = 0;
1110         bprm.sh_bang = 0;
1111         bprm.loader = 0;
1112         bprm.exec = 0;
1113         bprm.security = NULL;
1114         bprm.mm = mm_alloc();
1115         retval = -ENOMEM;
1116         if (!bprm.mm)
1117                 goto out_file;
1118
1119         retval = init_new_context(current, bprm.mm);
1120         if (retval < 0)
1121                 goto out_mm;
1122
1123         bprm.argc = count(argv, bprm.p / sizeof(void *));
1124         if ((retval = bprm.argc) < 0)
1125                 goto out_mm;
1126
1127         bprm.envc = count(envp, bprm.p / sizeof(void *));
1128         if ((retval = bprm.envc) < 0)
1129                 goto out_mm;
1130
1131         retval = security_bprm_alloc(&bprm);
1132         if (retval)
1133                 goto out;
1134
1135         retval = prepare_binprm(&bprm);
1136         if (retval < 0)
1137                 goto out;
1138
1139         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
1140         if (retval < 0)
1141                 goto out;
1142
1143         bprm.exec = bprm.p;
1144         retval = copy_strings(bprm.envc, envp, &bprm);
1145         if (retval < 0)
1146                 goto out;
1147
1148         retval = copy_strings(bprm.argc, argv, &bprm);
1149         if (retval < 0)
1150                 goto out;
1151
1152         retval = search_binary_handler(&bprm,regs);
1153         if (retval >= 0) {
1154                 free_arg_pages(&bprm);
1155
1156                 /* execve success */
1157                 security_bprm_free(&bprm);
1158                 return retval;
1159         }
1160
1161 out:
1162         /* Something went wrong, return the inode and free the argument pages*/
1163         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1164                 struct page * page = bprm.page[i];
1165                 if (page)
1166                         __free_page(page);
1167         }
1168
1169         if (bprm.security)
1170                 security_bprm_free(&bprm);
1171
1172 out_mm:
1173         if (bprm.mm)
1174                 mmdrop(bprm.mm);
1175
1176 out_file:
1177         if (bprm.file) {
1178                 allow_write_access(bprm.file);
1179                 fput(bprm.file);
1180         }
1181         return retval;
1182 }
1183
1184 EXPORT_SYMBOL(do_execve);
1185
1186 int set_binfmt(struct linux_binfmt *new)
1187 {
1188         struct linux_binfmt *old = current->binfmt;
1189
1190         if (new) {
1191                 if (!try_module_get(new->module))
1192                         return -1;
1193         }
1194         current->binfmt = new;
1195         if (old)
1196                 module_put(old->module);
1197         return 0;
1198 }
1199
1200 EXPORT_SYMBOL(set_binfmt);
1201
1202 #define CORENAME_MAX_SIZE 64
1203
1204 /* format_corename will inspect the pattern parameter, and output a
1205  * name into corename, which must have space for at least
1206  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1207  */
1208 void format_corename(char *corename, const char *pattern, long signr)
1209 {
1210         const char *pat_ptr = pattern;
1211         char *out_ptr = corename;
1212         char *const out_end = corename + CORENAME_MAX_SIZE;
1213         int rc;
1214         int pid_in_pattern = 0;
1215
1216         /* Repeat as long as we have more pattern to process and more output
1217            space */
1218         while (*pat_ptr) {
1219                 if (*pat_ptr != '%') {
1220                         if (out_ptr == out_end)
1221                                 goto out;
1222                         *out_ptr++ = *pat_ptr++;
1223                 } else {
1224                         switch (*++pat_ptr) {
1225                         case 0:
1226                                 goto out;
1227                         /* Double percent, output one percent */
1228                         case '%':
1229                                 if (out_ptr == out_end)
1230                                         goto out;
1231                                 *out_ptr++ = '%';
1232                                 break;
1233                         /* pid */
1234                         case 'p':
1235                                 pid_in_pattern = 1;
1236                                 rc = snprintf(out_ptr, out_end - out_ptr,
1237                                               "%d", current->tgid);
1238                                 if (rc > out_end - out_ptr)
1239                                         goto out;
1240                                 out_ptr += rc;
1241                                 break;
1242                         /* uid */
1243                         case 'u':
1244                                 rc = snprintf(out_ptr, out_end - out_ptr,
1245                                               "%d", current->uid);
1246                                 if (rc > out_end - out_ptr)
1247                                         goto out;
1248                                 out_ptr += rc;
1249                                 break;
1250                         /* gid */
1251                         case 'g':
1252                                 rc = snprintf(out_ptr, out_end - out_ptr,
1253                                               "%d", current->gid);
1254                                 if (rc > out_end - out_ptr)
1255                                         goto out;
1256                                 out_ptr += rc;
1257                                 break;
1258                         /* signal that caused the coredump */
1259                         case 's':
1260                                 rc = snprintf(out_ptr, out_end - out_ptr,
1261                                               "%ld", signr);
1262                                 if (rc > out_end - out_ptr)
1263                                         goto out;
1264                                 out_ptr += rc;
1265                                 break;
1266                         /* UNIX time of coredump */
1267                         case 't': {
1268                                 struct timeval tv;
1269                                 do_gettimeofday(&tv);
1270                                 rc = snprintf(out_ptr, out_end - out_ptr,
1271                                               "%lu", tv.tv_sec);
1272                                 if (rc > out_end - out_ptr)
1273                                         goto out;
1274                                 out_ptr += rc;
1275                                 break;
1276                         }
1277                         /* hostname */
1278                         case 'h':
1279                                 down_read(&uts_sem);
1280                                 rc = snprintf(out_ptr, out_end - out_ptr,
1281                                               "%s", system_utsname.nodename);
1282                                 up_read(&uts_sem);
1283                                 if (rc > out_end - out_ptr)
1284                                         goto out;
1285                                 out_ptr += rc;
1286                                 break;
1287                         /* executable */
1288                         case 'e':
1289                                 rc = snprintf(out_ptr, out_end - out_ptr,
1290                                               "%s", current->comm);
1291                                 if (rc > out_end - out_ptr)
1292                                         goto out;
1293                                 out_ptr += rc;
1294                                 break;
1295                         default:
1296                                 break;
1297                         }
1298                         ++pat_ptr;
1299                 }
1300         }
1301         /* Backward compatibility with core_uses_pid:
1302          *
1303          * If core_pattern does not include a %p (as is the default)
1304          * and core_uses_pid is set, then .%pid will be appended to
1305          * the filename */
1306         if (!pid_in_pattern
1307             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1308                 rc = snprintf(out_ptr, out_end - out_ptr,
1309                               ".%d", current->tgid);
1310                 if (rc > out_end - out_ptr)
1311                         goto out;
1312                 out_ptr += rc;
1313         }
1314       out:
1315         *out_ptr = 0;
1316 }
1317
1318 static void zap_threads (struct mm_struct *mm)
1319 {
1320         struct task_struct *g, *p;
1321         struct task_struct *tsk = current;
1322         struct completion *vfork_done = tsk->vfork_done;
1323
1324         /*
1325          * Make sure nobody is waiting for us to release the VM,
1326          * otherwise we can deadlock when we wait on each other
1327          */
1328         if (vfork_done) {
1329                 tsk->vfork_done = NULL;
1330                 complete(vfork_done);
1331         }
1332
1333         read_lock(&tasklist_lock);
1334         do_each_thread(g,p)
1335                 if (mm == p->mm && p != tsk) {
1336                         force_sig_specific(SIGKILL, p);
1337                         mm->core_waiters++;
1338                 }
1339         while_each_thread(g,p);
1340
1341         read_unlock(&tasklist_lock);
1342 }
1343
1344 static void coredump_wait(struct mm_struct *mm)
1345 {
1346         DECLARE_COMPLETION(startup_done);
1347
1348         mm->core_waiters++; /* let other threads block */
1349         mm->core_startup_done = &startup_done;
1350
1351         /* give other threads a chance to run: */
1352         yield();
1353
1354         zap_threads(mm);
1355         if (--mm->core_waiters) {
1356                 up_write(&mm->mmap_sem);
1357                 wait_for_completion(&startup_done);
1358         } else
1359                 up_write(&mm->mmap_sem);
1360         BUG_ON(mm->core_waiters);
1361 }
1362
1363 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1364 {
1365         char corename[CORENAME_MAX_SIZE + 1];
1366         struct mm_struct *mm = current->mm;
1367         struct linux_binfmt * binfmt;
1368         struct inode * inode;
1369         struct file * file;
1370         int retval = 0;
1371
1372         lock_kernel();
1373         binfmt = current->binfmt;
1374         if (!binfmt || !binfmt->core_dump)
1375                 goto fail;
1376         down_write(&mm->mmap_sem);
1377         if (!mm->dumpable) {
1378                 up_write(&mm->mmap_sem);
1379                 goto fail;
1380         }
1381         mm->dumpable = 0;
1382         init_completion(&mm->core_done);
1383         current->signal->group_exit = 1;
1384         current->signal->group_exit_code = exit_code;
1385         coredump_wait(mm);
1386
1387         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1388                 goto fail_unlock;
1389
1390         format_corename(corename, core_pattern, signr);
1391         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1392         if (IS_ERR(file))
1393                 goto fail_unlock;
1394         inode = file->f_dentry->d_inode;
1395         if (inode->i_nlink > 1)
1396                 goto close_fail;        /* multiple links - don't dump */
1397         if (d_unhashed(file->f_dentry))
1398                 goto close_fail;
1399
1400         if (!S_ISREG(inode->i_mode))
1401                 goto close_fail;
1402         if (!file->f_op)
1403                 goto close_fail;
1404         if (!file->f_op->write)
1405                 goto close_fail;
1406         if (do_truncate(file->f_dentry, 0) != 0)
1407                 goto close_fail;
1408
1409         retval = binfmt->core_dump(signr, regs, file);
1410
1411         current->signal->group_exit_code |= 0x80;
1412 close_fail:
1413         filp_close(file, NULL);
1414 fail_unlock:
1415         complete_all(&mm->core_done);
1416 fail:
1417         unlock_kernel();
1418         return retval;
1419 }