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/pgalloc.h>
  52 #include <asm/mmu_context.h>
  53
  54 #ifdef CONFIG_KMOD
  55 #include <linux/kmod.h>
  56 #endif
  57
  58 int core_uses_pid;
  59 char core_pattern[65] = "core";
  60 /* The maximal length of core_pattern is also specified in sysctl.c */
  61
  62 static struct linux_binfmt *formats;
  63 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  64
  65 int register_binfmt(struct linux_binfmt * fmt)
  66 {
  67         struct linux_binfmt ** tmp = &formats;
  68
  69         if (!fmt)
  70                 return -EINVAL;
  71         if (fmt->next)
  72                 return -EBUSY;
  73         write_lock(&binfmt_lock);
  74         while (*tmp) {
  75                 if (fmt == *tmp) {
  76                         write_unlock(&binfmt_lock);
  77                         return -EBUSY;
  78                 }
  79                 tmp = &(*tmp)->next;
  80         }
  81         fmt->next = formats;
  82         formats = fmt;
  83         write_unlock(&binfmt_lock);
  84         return 0;
  85 }
  86
  87 EXPORT_SYMBOL(register_binfmt);
  88
  89 int unregister_binfmt(struct linux_binfmt * fmt)
  90 {
  91         struct linux_binfmt ** tmp = &formats;
  92
  93         write_lock(&binfmt_lock);
  94         while (*tmp) {
  95                 if (fmt == *tmp) {
  96                         *tmp = fmt->next;
  97                         write_unlock(&binfmt_lock);
  98                         return 0;
  99                 }
 100                 tmp = &(*tmp)->next;
 101         }
 102         write_unlock(&binfmt_lock);
 103         return -EINVAL;
 104 }
 105
 106 EXPORT_SYMBOL(unregister_binfmt);
 107
 108 static inline void put_binfmt(struct linux_binfmt * fmt)
 109 {
 110         module_put(fmt->module);
 111 }
 112
 113 /*
 114  * Note that a shared library must be both readable and executable due to
 115  * security reasons.
 116  *
 117  * Also note that we take the address to load from from the file itself.
 118  */
 119 asmlinkage long sys_uselib(const char __user * library)
 120 {
 121         struct file * file;
 122         struct nameidata nd;
 123         int error;
 124
 125         nd.intent.open.flags = FMODE_READ;
 126         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 127         if (error)
 128                 goto out;
 129
 130         error = -EINVAL;
 131         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 132                 goto exit;
 133
 134         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 135         if (error)
 136                 goto exit;
 137
 138         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 139         error = PTR_ERR(file);
 140         if (IS_ERR(file))
 141                 goto out;
 142
 143         error = -ENOEXEC;
 144         if(file->f_op) {
 145                 struct linux_binfmt * fmt;
 146
 147                 read_lock(&binfmt_lock);
 148                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 149                         if (!fmt->load_shlib)
 150                                 continue;
 151                         if (!try_module_get(fmt->module))
 152                                 continue;
 153                         read_unlock(&binfmt_lock);
 154                         error = fmt->load_shlib(file);
 155                         read_lock(&binfmt_lock);
 156                         put_binfmt(fmt);
 157                         if (error != -ENOEXEC)
 158                                 break;
 159                 }
 160                 read_unlock(&binfmt_lock);
 161         }
 162         fput(file);
 163 out:
 164         return error;
 165 exit:
 166         path_release(&nd);
 167         goto out;
 168 }
 169
 170 /*
 171  * count() counts the number of strings in array ARGV.
 172  */
 173 static int count(char __user * __user * argv, int max)
 174 {
 175         int i = 0;
 176
 177         if (argv != NULL) {
 178                 for (;;) {
 179                         char __user * p;
 180
 181                         if (get_user(p, argv))
 182                                 return -EFAULT;
 183                         if (!p)
 184                                 break;
 185                         argv++;
 186                         if(++i > max)
 187                                 return -E2BIG;
 188                 }
 189         }
 190         return i;
 191 }
 192
 193 /*
 194  * 'copy_strings()' copies argument/environment strings from user
 195  * memory to free pages in kernel mem. These are in a format ready
 196  * to be put directly into the top of new user memory.
 197  */
 198 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 199 {
 200         struct page *kmapped_page = NULL;
 201         char *kaddr = NULL;
 202         int ret;
 203
 204         while (argc-- > 0) {
 205                 char __user *str;
 206                 int len;
 207                 unsigned long pos;
 208
 209                 if (get_user(str, argv+argc) ||
 210                                 !(len = strnlen_user(str, bprm->p))) {
 211                         ret = -EFAULT;
 212                         goto out;
 213                 }
 214
 215                 if (bprm->p < len)  {
 216                         ret = -E2BIG;
 217                         goto out;
 218                 }
 219
 220                 bprm->p -= len;
 221                 /* XXX: add architecture specific overflow check here. */
 222                 pos = bprm->p;
 223
 224                 while (len > 0) {
 225                         int i, new, err;
 226                         int offset, bytes_to_copy;
 227                         struct page *page;
 228
 229                         offset = pos % PAGE_SIZE;
 230                         i = pos/PAGE_SIZE;
 231                         page = bprm->page[i];
 232                         new = 0;
 233                         if (!page) {
 234                                 page = alloc_page(GFP_HIGHUSER);
 235                                 bprm->page[i] = page;
 236                                 if (!page) {
 237                                         ret = -ENOMEM;
 238                                         goto out;
 239                                 }
 240                                 new = 1;
 241                         }
 242
 243                         if (page != kmapped_page) {
 244                                 if (kmapped_page)
 245                                         kunmap(kmapped_page);
 246                                 kmapped_page = page;
 247                                 kaddr = kmap(kmapped_page);
 248                         }
 249                         if (new && offset)
 250                                 memset(kaddr, 0, offset);
 251                         bytes_to_copy = PAGE_SIZE - offset;
 252                         if (bytes_to_copy > len) {
 253                                 bytes_to_copy = len;
 254                                 if (new)
 255                                         memset(kaddr+offset+len, 0,
 256                                                 PAGE_SIZE-offset-len);
 257                         }
 258                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 259                         if (err) {
 260                                 ret = -EFAULT;
 261                                 goto out;
 262                         }
 263
 264                         pos += bytes_to_copy;
 265                         str += bytes_to_copy;
 266                         len -= bytes_to_copy;
 267                 }
 268         }
 269         ret = 0;
 270 out:
 271         if (kmapped_page)
 272                 kunmap(kmapped_page);
 273         return ret;
 274 }
 275
 276 /*
 277  * Like copy_strings, but get argv and its values from kernel memory.
 278  */
 279 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 280 {
 281         int r;
 282         mm_segment_t oldfs = get_fs();
 283         set_fs(KERNEL_DS);
 284         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 285         set_fs(oldfs);
 286         return r;
 287 }
 288
 289 EXPORT_SYMBOL(copy_strings_kernel);
 290
 291 #ifdef CONFIG_MMU
 292 /*
 293  * This routine is used to map in a page into an address space: needed by
 294  * execve() for the initial stack and environment pages.
 295  *
 296  * 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         lru_cache_add_active(page);
 325         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
 326                                         page, vma->vm_page_prot))));
 327         page_add_anon_rmap(page, vma, address);
 328         pte_unmap(pte);
 329         spin_unlock(&mm->page_table_lock);
 330
 331         /* no need for flush_tlb */
 332         return;
 333 out:
 334         spin_unlock(&mm->page_table_lock);
 335 out_sig:
 336         __free_page(page);
 337         force_sig(SIGKILL, current);
 338 }
 339
 340 int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
 341 {
 342         unsigned long stack_base;
 343         struct vm_area_struct *mpnt;
 344         struct mm_struct *mm = current->mm;
 345         int i;
 346         long arg_size;
 347
 348 #ifdef CONFIG_STACK_GROWSUP
 349         /* Move the argument and environment strings to the bottom of the
 350          * stack space.
 351          */
 352         int offset, j;
 353         char *to, *from;
 354
 355         /* Start by shifting all the pages down */
 356         i = 0;
 357         for (j = 0; j < MAX_ARG_PAGES; j++) {
 358                 struct page *page = bprm->page[j];
 359                 if (!page)
 360                         continue;
 361                 bprm->page[i++] = page;
 362         }
 363
 364         /* Now move them within their pages */
 365         offset = bprm->p % PAGE_SIZE;
 366         to = kmap(bprm->page[0]);
 367         for (j = 1; j < i; j++) {
 368                 memmove(to, to + offset, PAGE_SIZE - offset);
 369                 from = kmap(bprm->page[j]);
 370                 memcpy(to + PAGE_SIZE - offset, from, offset);
 371                 kunmap(bprm->page[j - 1]);
 372                 to = from;
 373         }
 374         memmove(to, to + offset, PAGE_SIZE - offset);
 375         kunmap(bprm->page[j - 1]);
 376
 377         /* Adjust bprm->p to point to the end of the strings. */
 378         bprm->p = PAGE_SIZE * i - offset;
 379
 380         /* Limit stack size to 1GB */
 381         stack_base = current->rlim[RLIMIT_STACK].rlim_max;
 382         if (stack_base > (1 << 30))
 383                 stack_base = 1 << 30;
 384         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
 385
 386         mm->arg_start = stack_base;
 387         arg_size = i << PAGE_SHIFT;
 388
 389         /* zero pages that were copied above */
 390         while (i < MAX_ARG_PAGES)
 391                 bprm->page[i++] = NULL;
 392 #else
 393         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 394         mm->arg_start = bprm->p + stack_base;
 395         arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
 396 #endif
 397
 398         bprm->p += stack_base;
 399         if (bprm->loader)
 400                 bprm->loader += stack_base;
 401         bprm->exec += stack_base;
 402
 403         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 404         if (!mpnt)
 405                 return -ENOMEM;
 406
 407         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT)) {
 408                 kmem_cache_free(vm_area_cachep, mpnt);
 409                 return -ENOMEM;
 410         }
 411
 412         memset(mpnt, 0, sizeof(*mpnt));
 413
 414         down_write(&mm->mmap_sem);
 415         {
 416                 mpnt->vm_mm = mm;
 417 #ifdef CONFIG_STACK_GROWSUP
 418                 mpnt->vm_start = stack_base;
 419                 mpnt->vm_end = PAGE_MASK &
 420                         (PAGE_SIZE - 1 + (unsigned long) bprm->p);
 421 #else
 422                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 423                 mpnt->vm_end = STACK_TOP;
 424 #endif
 425                 /* Adjust stack execute permissions; explicitly enable
 426                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 427                  * and leave alone (arch default) otherwise. */
 428                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 429                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 430                 else if (executable_stack == EXSTACK_DISABLE_X)
 431                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 432                 else
 433                         mpnt->vm_flags = VM_STACK_FLAGS;
 434                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 435                 insert_vm_struct(mm, mpnt);
 436                 mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 437         }
 438
 439         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 440                 struct page *page = bprm->page[i];
 441                 if (page) {
 442                         bprm->page[i] = NULL;
 443                         install_arg_page(mpnt, page, stack_base);
 444                 }
 445                 stack_base += PAGE_SIZE;
 446         }
 447         up_write(&mm->mmap_sem);
 448
 449         return 0;
 450 }
 451
 452 EXPORT_SYMBOL(setup_arg_pages);
 453
 454 #define free_arg_pages(bprm) do { } while (0)
 455
 456 #else
 457
 458 static inline void free_arg_pages(struct linux_binprm *bprm)
 459 {
 460         int i;
 461
 462         for (i = 0; i < MAX_ARG_PAGES; i++) {
 463                 if (bprm->page[i])
 464                         __free_page(bprm->page[i]);
 465                 bprm->page[i] = NULL;
 466         }
 467 }
 468
 469 #endif /* CONFIG_MMU */
 470
 471 struct file *open_exec(const char *name)
 472 {
 473         struct nameidata nd;
 474         int err;
 475         struct file *file;
 476
 477         nd.intent.open.flags = FMODE_READ;
 478         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 479         file = ERR_PTR(err);
 480
 481         if (!err) {
 482                 struct inode *inode = nd.dentry->d_inode;
 483                 file = ERR_PTR(-EACCES);
 484                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 485                     S_ISREG(inode->i_mode)) {
 486                         int err = permission(inode, MAY_EXEC, &nd);
 487                         if (!err && !(inode->i_mode & 0111))
 488                                 err = -EACCES;
 489                         file = ERR_PTR(err);
 490                         if (!err) {
 491                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 492                                 if (!IS_ERR(file)) {
 493                                         err = deny_write_access(file);
 494                                         if (err) {
 495                                                 fput(file);
 496                                                 file = ERR_PTR(err);
 497                                         }
 498                                 }
 499 out:
 500                                 return file;
 501                         }
 502                 }
 503                 path_release(&nd);
 504         }
 505         goto out;
 506 }
 507
 508 EXPORT_SYMBOL(open_exec);
 509
 510 int kernel_read(struct file *file, unsigned long offset,
 511         char *addr, unsigned long count)
 512 {
 513         mm_segment_t old_fs;
 514         loff_t pos = offset;
 515         int result;
 516
 517         old_fs = get_fs();
 518         set_fs(get_ds());
 519         /* The cast to a user pointer is valid due to the set_fs() */
 520         result = vfs_read(file, (void __user *)addr, count, &pos);
 521         set_fs(old_fs);
 522         return result;
 523 }
 524
 525 EXPORT_SYMBOL(kernel_read);
 526
 527 static int exec_mmap(struct mm_struct *mm)
 528 {
 529         struct task_struct *tsk;
 530         struct mm_struct * old_mm, *active_mm;
 531
 532         /* Add it to the list of mm's */
 533         spin_lock(&mmlist_lock);
 534         list_add(&mm->mmlist, &init_mm.mmlist);
 535         mmlist_nr++;
 536         spin_unlock(&mmlist_lock);
 537
 538         /* Notify parent that we're no longer interested in the old VM */
 539         tsk = current;
 540         old_mm = current->mm;
 541         mm_release(tsk, old_mm);
 542
 543         task_lock(tsk);
 544         active_mm = tsk->active_mm;
 545         tsk->mm = mm;
 546         tsk->active_mm = mm;
 547         activate_mm(active_mm, mm);
 548         task_unlock(tsk);
 549         if (old_mm) {
 550                 if (active_mm != old_mm) BUG();
 551                 mmput(old_mm);
 552                 return 0;
 553         }
 554         mmdrop(active_mm);
 555         return 0;
 556 }
 557
 558 /*
 559  * This function makes sure the current process has its own signal table,
 560  * so that flush_signal_handlers can later reset the handlers without
 561  * disturbing other processes.  (Other processes might share the signal
 562  * table via the CLONE_SIGHAND option to clone().)
 563  */
 564 static inline int de_thread(struct task_struct *tsk)
 565 {
 566         struct signal_struct *newsig, *oldsig = tsk->signal;
 567         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 568         spinlock_t *lock = &oldsighand->siglock;
 569         int count;
 570
 571         /*
 572          * If we don't share sighandlers, then we aren't sharing anything
 573          * and we can just re-use it all.
 574          */
 575         if (atomic_read(&oldsighand->count) <= 1)
 576                 return 0;
 577
 578         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 579         if (!newsighand)
 580                 return -ENOMEM;
 581
 582         spin_lock_init(&newsighand->siglock);
 583         atomic_set(&newsighand->count, 1);
 584         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
 585
 586         /*
 587          * See if we need to allocate a new signal structure
 588          */
 589         newsig = NULL;
 590         if (atomic_read(&oldsig->count) > 1) {
 591                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 592                 if (!newsig) {
 593                         kmem_cache_free(sighand_cachep, newsighand);
 594                         return -ENOMEM;
 595                 }
 596                 atomic_set(&newsig->count, 1);
 597                 newsig->group_exit = 0;
 598                 newsig->group_exit_code = 0;
 599                 newsig->group_exit_task = NULL;
 600                 newsig->group_stop_count = 0;
 601                 newsig->curr_target = NULL;
 602                 init_sigpending(&newsig->shared_pending);
 603                 INIT_LIST_HEAD(&newsig->posix_timers);
 604
 605                 newsig->tty = oldsig->tty;
 606                 newsig->pgrp = oldsig->pgrp;
 607                 newsig->session = oldsig->session;
 608                 newsig->leader = oldsig->leader;
 609                 newsig->tty_old_pgrp = oldsig->tty_old_pgrp;
 610         }
 611
 612         if (thread_group_empty(current))
 613                 goto no_thread_group;
 614
 615         /*
 616          * Kill all other threads in the thread group.
 617          * We must hold tasklist_lock to call zap_other_threads.
 618          */
 619         read_lock(&tasklist_lock);
 620         spin_lock_irq(lock);
 621         if (oldsig->group_exit) {
 622                 /*
 623                  * Another group action in progress, just
 624                  * return so that the signal is processed.
 625                  */
 626                 spin_unlock_irq(lock);
 627                 read_unlock(&tasklist_lock);
 628                 kmem_cache_free(sighand_cachep, newsighand);
 629                 if (newsig)
 630                         kmem_cache_free(signal_cachep, newsig);
 631                 return -EAGAIN;
 632         }
 633         oldsig->group_exit = 1;
 634         zap_other_threads(current);
 635         read_unlock(&tasklist_lock);
 636
 637         /*
 638          * Account for the thread group leader hanging around:
 639          */
 640         count = 2;
 641         if (current->pid == current->tgid)
 642                 count = 1;
 643         while (atomic_read(&oldsig->count) > count) {
 644                 oldsig->group_exit_task = current;
 645                 oldsig->notify_count = count;
 646                 __set_current_state(TASK_UNINTERRUPTIBLE);
 647                 spin_unlock_irq(lock);
 648                 schedule();
 649                 spin_lock_irq(lock);
 650         }
 651         spin_unlock_irq(lock);
 652
 653         /*
 654          * At this point all other threads have exited, all we have to
 655          * do is to wait for the thread group leader to become inactive,
 656          * and to assume its PID:
 657          */
 658         if (current->pid != current->tgid) {
 659                 struct task_struct *leader = current->group_leader, *parent;
 660                 struct dentry *proc_dentry1, *proc_dentry2;
 661                 unsigned long state, ptrace;
 662
 663                 /*
 664                  * Wait for the thread group leader to be a zombie.
 665                  * It should already be zombie at this point, most
 666                  * of the time.
 667                  */
 668                 while (leader->state != TASK_ZOMBIE)
 669                         yield();
 670
 671                 spin_lock(&leader->proc_lock);
 672                 spin_lock(&current->proc_lock);
 673                 proc_dentry1 = proc_pid_unhash(current);
 674                 proc_dentry2 = proc_pid_unhash(leader);
 675                 write_lock_irq(&tasklist_lock);
 676
 677                 if (leader->tgid != current->tgid)
 678                         BUG();
 679                 if (current->pid == current->tgid)
 680                         BUG();
 681                 /*
 682                  * An exec() starts a new thread group with the
 683                  * TGID of the previous thread group. Rehash the
 684                  * two threads with a switched PID, and release
 685                  * the former thread group leader:
 686                  */
 687                 ptrace = leader->ptrace;
 688                 parent = leader->parent;
 689
 690                 ptrace_unlink(current);
 691                 ptrace_unlink(leader);
 692                 remove_parent(current);
 693                 remove_parent(leader);
 694
 695                 switch_exec_pids(leader, current);
 696
 697                 current->parent = current->real_parent = leader->real_parent;
 698                 leader->parent = leader->real_parent = child_reaper;
 699                 current->group_leader = current;
 700                 leader->group_leader = leader;
 701
 702                 add_parent(current, current->parent);
 703                 add_parent(leader, leader->parent);
 704                 if (ptrace) {
 705                         current->ptrace = ptrace;
 706                         __ptrace_link(current, parent);
 707                 }
 708
 709                 list_del(&current->tasks);
 710                 list_add_tail(&current->tasks, &init_task.tasks);
 711                 current->exit_signal = SIGCHLD;
 712                 state = leader->state;
 713
 714                 write_unlock_irq(&tasklist_lock);
 715                 spin_unlock(&leader->proc_lock);
 716                 spin_unlock(&current->proc_lock);
 717                 proc_pid_flush(proc_dentry1);
 718                 proc_pid_flush(proc_dentry2);
 719
 720                 if (state != TASK_ZOMBIE)
 721                         BUG();
 722                 release_task(leader);
 723         }
 724
 725 no_thread_group:
 726
 727         write_lock_irq(&tasklist_lock);
 728         spin_lock(&oldsighand->siglock);
 729         spin_lock(&newsighand->siglock);
 730
 731         if (current == oldsig->curr_target)
 732                 oldsig->curr_target = next_thread(current);
 733         if (newsig)
 734                 current->signal = newsig;
 735         current->sighand = newsighand;
 736         init_sigpending(&current->pending);
 737         recalc_sigpending();
 738
 739         spin_unlock(&newsighand->siglock);
 740         spin_unlock(&oldsighand->siglock);
 741         write_unlock_irq(&tasklist_lock);
 742
 743         if (newsig && atomic_dec_and_test(&oldsig->count))
 744                 kmem_cache_free(signal_cachep, oldsig);
 745
 746         if (atomic_dec_and_test(&oldsighand->count))
 747                 kmem_cache_free(sighand_cachep, oldsighand);
 748
 749         if (!thread_group_empty(current))
 750                 BUG();
 751         if (current->tgid != current->pid)
 752                 BUG();
 753         return 0;
 754 }
 755
 756 /*
 757  * These functions flushes out all traces of the currently running executable
 758  * so that a new one can be started
 759  */
 760
 761 static inline void flush_old_files(struct files_struct * files)
 762 {
 763         long j = -1;
 764
 765         spin_lock(&files->file_lock);
 766         for (;;) {
 767                 unsigned long set, i;
 768
 769                 j++;
 770                 i = j * __NFDBITS;
 771                 if (i >= files->max_fds || i >= files->max_fdset)
 772                         break;
 773                 set = files->close_on_exec->fds_bits[j];
 774                 if (!set)
 775                         continue;
 776                 files->close_on_exec->fds_bits[j] = 0;
 777                 spin_unlock(&files->file_lock);
 778                 for ( ; set ; i++,set >>= 1) {
 779                         if (set & 1) {
 780                                 sys_close(i);
 781                         }
 782                 }
 783                 spin_lock(&files->file_lock);
 784
 785         }
 786         spin_unlock(&files->file_lock);
 787 }
 788
 789 int flush_old_exec(struct linux_binprm * bprm)
 790 {
 791         char * name;
 792         int i, ch, retval;
 793         struct files_struct *files;
 794
 795         /*
 796          * Make sure we have a private signal table and that
 797          * we are unassociated from the previous thread group.
 798          */
 799         retval = de_thread(current);
 800         if (retval)
 801                 goto out;
 802
 803         /*
 804          * Make sure we have private file handles. Ask the
 805          * fork helper to do the work for us and the exit
 806          * helper to do the cleanup of the old one.
 807          */
 808         files = current->files;         /* refcounted so safe to hold */
 809         retval = unshare_files();
 810         if (retval)
 811                 goto out;
 812         /*
 813          * Release all of the old mmap stuff
 814          */
 815         retval = exec_mmap(bprm->mm);
 816         if (retval)
 817                 goto mmap_failed;
 818
 819         bprm->mm = NULL;                /* We're using it now */
 820
 821         /* This is the point of no return */
 822         steal_locks(files);
 823         put_files_struct(files);
 824
 825         current->sas_ss_sp = current->sas_ss_size = 0;
 826
 827         if (current->euid == current->uid && current->egid == current->gid)
 828                 current->mm->dumpable = 1;
 829         name = bprm->filename;
 830         for (i=0; (ch = *(name++)) != '\0';) {
 831                 if (ch == '/')
 832                         i = 0;
 833                 else
 834                         if (i < 15)
 835                                 current->comm[i++] = ch;
 836         }
 837         current->comm[i] = '\0';
 838
 839         flush_thread();
 840
 841         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 842             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL))
 843                 current->mm->dumpable = 0;
 844
 845         /* An exec changes our domain. We are no longer part of the thread
 846            group */
 847
 848         current->self_exec_id++;
 849
 850         flush_signal_handlers(current, 0);
 851         flush_old_files(current->files);
 852
 853         return 0;
 854
 855 mmap_failed:
 856         put_files_struct(current->files);
 857         current->files = files;
 858 out:
 859         return retval;
 860 }
 861
 862 EXPORT_SYMBOL(flush_old_exec);
 863
 864 /*
 865  * Fill the binprm structure from the inode.
 866  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 867  */
 868 int prepare_binprm(struct linux_binprm *bprm)
 869 {
 870         int mode;
 871         struct inode * inode = bprm->file->f_dentry->d_inode;
 872         int retval;
 873
 874         mode = inode->i_mode;
 875         /*
 876          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 877          * vfs_permission lets a non-executable through
 878          */
 879         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 880                 return -EACCES;
 881         if (bprm->file->f_op == NULL)
 882                 return -EACCES;
 883
 884         bprm->e_uid = current->euid;
 885         bprm->e_gid = current->egid;
 886
 887         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 888                 /* Set-uid? */
 889                 if (mode & S_ISUID)
 890                         bprm->e_uid = inode->i_uid;
 891
 892                 /* Set-gid? */
 893                 /*
 894                  * If setgid is set but no group execute bit then this
 895                  * is a candidate for mandatory locking, not a setgid
 896                  * executable.
 897                  */
 898                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
 899                         bprm->e_gid = inode->i_gid;
 900         }
 901
 902         /* fill in binprm security blob */
 903         retval = security_bprm_set(bprm);
 904         if (retval)
 905                 return retval;
 906
 907         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 908         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 909 }
 910
 911 EXPORT_SYMBOL(prepare_binprm);
 912
 913 static inline int unsafe_exec(struct task_struct *p)
 914 {
 915         int unsafe = 0;
 916         if (p->ptrace & PT_PTRACED) {
 917                 if (p->ptrace & PT_PTRACE_CAP)
 918                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 919                 else
 920                         unsafe |= LSM_UNSAFE_PTRACE;
 921         }
 922         if (atomic_read(&p->fs->count) > 1 ||
 923             atomic_read(&p->files->count) > 1 ||
 924             atomic_read(&p->sighand->count) > 1)
 925                 unsafe |= LSM_UNSAFE_SHARE;
 926
 927         return unsafe;
 928 }
 929
 930 void compute_creds(struct linux_binprm *bprm)
 931 {
 932         int unsafe;
 933         task_lock(current);
 934         unsafe = unsafe_exec(current);
 935         security_bprm_apply_creds(bprm, unsafe);
 936         task_unlock(current);
 937 }
 938
 939 EXPORT_SYMBOL(compute_creds);
 940
 941 void remove_arg_zero(struct linux_binprm *bprm)
 942 {
 943         if (bprm->argc) {
 944                 unsigned long offset;
 945                 char * kaddr;
 946                 struct page *page;
 947
 948                 offset = bprm->p % PAGE_SIZE;
 949                 goto inside;
 950
 951                 while (bprm->p++, *(kaddr+offset++)) {
 952                         if (offset != PAGE_SIZE)
 953                                 continue;
 954                         offset = 0;
 955                         kunmap_atomic(kaddr, KM_USER0);
 956 inside:
 957                         page = bprm->page[bprm->p/PAGE_SIZE];
 958                         kaddr = kmap_atomic(page, KM_USER0);
 959                 }
 960                 kunmap_atomic(kaddr, KM_USER0);
 961                 bprm->argc--;
 962         }
 963 }
 964
 965 EXPORT_SYMBOL(remove_arg_zero);
 966
 967 /*
 968  * cycle the list of binary formats handler, until one recognizes the image
 969  */
 970 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 971 {
 972         int try,retval=0;
 973         struct linux_binfmt *fmt;
 974 #ifdef __alpha__
 975         /* handle /sbin/loader.. */
 976         {
 977             struct exec * eh = (struct exec *) bprm->buf;
 978
 979             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 980                 (eh->fh.f_flags & 0x3000) == 0x3000)
 981             {
 982                 struct file * file;
 983                 unsigned long loader;
 984
 985                 allow_write_access(bprm->file);
 986                 fput(bprm->file);
 987                 bprm->file = NULL;
 988
 989                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 990
 991                 file = open_exec("/sbin/loader");
 992                 retval = PTR_ERR(file);
 993                 if (IS_ERR(file))
 994                         return retval;
 995
 996                 /* Remember if the application is TASO.  */
 997                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
 998
 999                 bprm->file = file;
1000                 bprm->loader = loader;
1001                 retval = prepare_binprm(bprm);
1002                 if (retval<0)
1003                         return retval;
1004                 /* should call search_binary_handler recursively here,
1005                    but it does not matter */
1006             }
1007         }
1008 #endif
1009         retval = security_bprm_check(bprm);
1010         if (retval)
1011                 return retval;
1012
1013         /* kernel module loader fixup */
1014         /* so we don't try to load run modprobe in kernel space. */
1015         set_fs(USER_DS);
1016         for (try=0; try<2; try++) {
1017                 read_lock(&binfmt_lock);
1018                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1019                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1020                         if (!fn)
1021                                 continue;
1022                         if (!try_module_get(fmt->module))
1023                                 continue;
1024                         read_unlock(&binfmt_lock);
1025                         retval = fn(bprm, regs);
1026                         if (retval >= 0) {
1027                                 put_binfmt(fmt);
1028                                 allow_write_access(bprm->file);
1029                                 if (bprm->file)
1030                                         fput(bprm->file);
1031                                 bprm->file = NULL;
1032                                 current->did_exec = 1;
1033                                 return retval;
1034                         }
1035                         read_lock(&binfmt_lock);
1036                         put_binfmt(fmt);
1037                         if (retval != -ENOEXEC || bprm->mm == NULL)
1038                                 break;
1039                         if (!bprm->file) {
1040                                 read_unlock(&binfmt_lock);
1041                                 return retval;
1042                         }
1043                 }
1044                 read_unlock(&binfmt_lock);
1045                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1046                         break;
1047 #ifdef CONFIG_KMOD
1048                 }else{
1049 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1050                         if (printable(bprm->buf[0]) &&
1051                             printable(bprm->buf[1]) &&
1052                             printable(bprm->buf[2]) &&
1053                             printable(bprm->buf[3]))
1054                                 break; /* -ENOEXEC */
1055                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1056 #endif
1057                 }
1058         }
1059         return retval;
1060 }
1061
1062 EXPORT_SYMBOL(search_binary_handler);
1063
1064 /*
1065  * sys_execve() executes a new program.
1066  */
1067 int do_execve(char * filename,
1068         char __user *__user *argv,
1069         char __user *__user *envp,
1070         struct pt_regs * regs)
1071 {
1072         struct linux_binprm bprm;
1073         struct file *file;
1074         int retval;
1075         int i;
1076
1077         file = open_exec(filename);
1078
1079         retval = PTR_ERR(file);
1080         if (IS_ERR(file))
1081                 return retval;
1082
1083         sched_balance_exec();
1084
1085         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1086         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
1087
1088         bprm.file = file;
1089         bprm.filename = filename;
1090         bprm.interp = filename;
1091         bprm.sh_bang = 0;
1092         bprm.loader = 0;
1093         bprm.exec = 0;
1094         bprm.security = NULL;
1095         bprm.mm = mm_alloc();
1096         retval = -ENOMEM;
1097         if (!bprm.mm)
1098                 goto out_file;
1099
1100         retval = init_new_context(current, bprm.mm);
1101         if (retval < 0)
1102                 goto out_mm;
1103
1104         bprm.argc = count(argv, bprm.p / sizeof(void *));
1105         if ((retval = bprm.argc) < 0)
1106                 goto out_mm;
1107
1108         bprm.envc = count(envp, bprm.p / sizeof(void *));
1109         if ((retval = bprm.envc) < 0)
1110                 goto out_mm;
1111
1112         retval = security_bprm_alloc(&bprm);
1113         if (retval)
1114                 goto out;
1115
1116         retval = prepare_binprm(&bprm);
1117         if (retval < 0)
1118                 goto out;
1119
1120         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
1121         if (retval < 0)
1122                 goto out;
1123
1124         bprm.exec = bprm.p;
1125         retval = copy_strings(bprm.envc, envp, &bprm);
1126         if (retval < 0)
1127                 goto out;
1128
1129         retval = copy_strings(bprm.argc, argv, &bprm);
1130         if (retval < 0)
1131                 goto out;
1132
1133         retval = search_binary_handler(&bprm,regs);
1134         if (retval >= 0) {
1135                 free_arg_pages(&bprm);
1136
1137                 /* execve success */
1138                 security_bprm_free(&bprm);
1139                 return retval;
1140         }
1141
1142 out:
1143         /* Something went wrong, return the inode and free the argument pages*/
1144         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1145                 struct page * page = bprm.page[i];
1146                 if (page)
1147                         __free_page(page);
1148         }
1149
1150         if (bprm.security)
1151                 security_bprm_free(&bprm);
1152
1153 out_mm:
1154         if (bprm.mm)
1155                 mmdrop(bprm.mm);
1156
1157 out_file:
1158         if (bprm.file) {
1159                 allow_write_access(bprm.file);
1160                 fput(bprm.file);
1161         }
1162         return retval;
1163 }
1164
1165 EXPORT_SYMBOL(do_execve);
1166
1167 int set_binfmt(struct linux_binfmt *new)
1168 {
1169         struct linux_binfmt *old = current->binfmt;
1170
1171         if (new) {
1172                 if (!try_module_get(new->module))
1173                         return -1;
1174         }
1175         current->binfmt = new;
1176         if (old)
1177                 module_put(old->module);
1178         return 0;
1179 }
1180
1181 EXPORT_SYMBOL(set_binfmt);
1182
1183 #define CORENAME_MAX_SIZE 64
1184
1185 /* format_corename will inspect the pattern parameter, and output a
1186  * name into corename, which must have space for at least
1187  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1188  */
1189 void format_corename(char *corename, const char *pattern, long signr)
1190 {
1191         const char *pat_ptr = pattern;
1192         char *out_ptr = corename;
1193         char *const out_end = corename + CORENAME_MAX_SIZE;
1194         int rc;
1195         int pid_in_pattern = 0;
1196
1197         /* Repeat as long as we have more pattern to process and more output
1198            space */
1199         while (*pat_ptr) {
1200                 if (*pat_ptr != '%') {
1201                         if (out_ptr == out_end)
1202                                 goto out;
1203                         *out_ptr++ = *pat_ptr++;
1204                 } else {
1205                         switch (*++pat_ptr) {
1206                         case 0:
1207                                 goto out;
1208                         /* Double percent, output one percent */
1209                         case '%':
1210                                 if (out_ptr == out_end)
1211                                         goto out;
1212                                 *out_ptr++ = '%';
1213                                 break;
1214                         /* pid */
1215                         case 'p':
1216                                 pid_in_pattern = 1;
1217                                 rc = snprintf(out_ptr, out_end - out_ptr,
1218                                               "%d", current->tgid);
1219                                 if (rc > out_end - out_ptr)
1220                                         goto out;
1221                                 out_ptr += rc;
1222                                 break;
1223                         /* uid */
1224                         case 'u':
1225                                 rc = snprintf(out_ptr, out_end - out_ptr,
1226                                               "%d", current->uid);
1227                                 if (rc > out_end - out_ptr)
1228                                         goto out;
1229                                 out_ptr += rc;
1230                                 break;
1231                         /* gid */
1232                         case 'g':
1233                                 rc = snprintf(out_ptr, out_end - out_ptr,
1234                                               "%d", current->gid);
1235                                 if (rc > out_end - out_ptr)
1236                                         goto out;
1237                                 out_ptr += rc;
1238                                 break;
1239                         /* signal that caused the coredump */
1240                         case 's':
1241                                 rc = snprintf(out_ptr, out_end - out_ptr,
1242                                               "%ld", signr);
1243                                 if (rc > out_end - out_ptr)
1244                                         goto out;
1245                                 out_ptr += rc;
1246                                 break;
1247                         /* UNIX time of coredump */
1248                         case 't': {
1249                                 struct timeval tv;
1250                                 do_gettimeofday(&tv);
1251                                 rc = snprintf(out_ptr, out_end - out_ptr,
1252                                               "%lu", tv.tv_sec);
1253                                 if (rc > out_end - out_ptr)
1254                                         goto out;
1255                                 out_ptr += rc;
1256                                 break;
1257                         }
1258                         /* hostname */
1259                         case 'h':
1260                                 down_read(&uts_sem);
1261                                 rc = snprintf(out_ptr, out_end - out_ptr,
1262                                               "%s", system_utsname.nodename);
1263                                 up_read(&uts_sem);
1264                                 if (rc > out_end - out_ptr)
1265                                         goto out;
1266                                 out_ptr += rc;
1267                                 break;
1268                         /* executable */
1269                         case 'e':
1270                                 rc = snprintf(out_ptr, out_end - out_ptr,
1271                                               "%s", current->comm);
1272                                 if (rc > out_end - out_ptr)
1273                                         goto out;
1274                                 out_ptr += rc;
1275                                 break;
1276                         default:
1277                                 break;
1278                         }
1279                         ++pat_ptr;
1280                 }
1281         }
1282         /* Backward compatibility with core_uses_pid:
1283          *
1284          * If core_pattern does not include a %p (as is the default)
1285          * and core_uses_pid is set, then .%pid will be appended to
1286          * the filename */
1287         if (!pid_in_pattern
1288             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1289                 rc = snprintf(out_ptr, out_end - out_ptr,
1290                               ".%d", current->tgid);
1291                 if (rc > out_end - out_ptr)
1292                         goto out;
1293                 out_ptr += rc;
1294         }
1295       out:
1296         *out_ptr = 0;
1297 }
1298
1299 static void zap_threads (struct mm_struct *mm)
1300 {
1301         struct task_struct *g, *p;
1302         struct task_struct *tsk = current;
1303         struct completion *vfork_done = tsk->vfork_done;
1304
1305         /*
1306          * Make sure nobody is waiting for us to release the VM,
1307          * otherwise we can deadlock when we wait on each other
1308          */
1309         if (vfork_done) {
1310                 tsk->vfork_done = NULL;
1311                 complete(vfork_done);
1312         }
1313
1314         read_lock(&tasklist_lock);
1315         do_each_thread(g,p)
1316                 if (mm == p->mm && p != tsk) {
1317                         force_sig_specific(SIGKILL, p);
1318                         mm->core_waiters++;
1319                 }
1320         while_each_thread(g,p);
1321
1322         read_unlock(&tasklist_lock);
1323 }
1324
1325 static void coredump_wait(struct mm_struct *mm)
1326 {
1327         DECLARE_COMPLETION(startup_done);
1328
1329         mm->core_waiters++; /* let other threads block */
1330         mm->core_startup_done = &startup_done;
1331
1332         /* give other threads a chance to run: */
1333         yield();
1334
1335         zap_threads(mm);
1336         if (--mm->core_waiters) {
1337                 up_write(&mm->mmap_sem);
1338                 wait_for_completion(&startup_done);
1339         } else
1340                 up_write(&mm->mmap_sem);
1341         BUG_ON(mm->core_waiters);
1342 }
1343
1344 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1345 {
1346         char corename[CORENAME_MAX_SIZE + 1];
1347         struct mm_struct *mm = current->mm;
1348         struct linux_binfmt * binfmt;
1349         struct inode * inode;
1350         struct file * file;
1351         int retval = 0;
1352
1353         lock_kernel();
1354         binfmt = current->binfmt;
1355         if (!binfmt || !binfmt->core_dump)
1356                 goto fail;
1357         down_write(&mm->mmap_sem);
1358         if (!mm->dumpable) {
1359                 up_write(&mm->mmap_sem);
1360                 goto fail;
1361         }
1362         mm->dumpable = 0;
1363         init_completion(&mm->core_done);
1364         current->signal->group_exit = 1;
1365         current->signal->group_exit_code = exit_code;
1366         coredump_wait(mm);
1367
1368         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1369                 goto fail_unlock;
1370
1371         format_corename(corename, core_pattern, signr);
1372         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1373         if (IS_ERR(file))
1374                 goto fail_unlock;
1375         inode = file->f_dentry->d_inode;
1376         if (inode->i_nlink > 1)
1377                 goto close_fail;        /* multiple links - don't dump */
1378         if (d_unhashed(file->f_dentry))
1379                 goto close_fail;
1380
1381         if (!S_ISREG(inode->i_mode))
1382                 goto close_fail;
1383         if (!file->f_op)
1384                 goto close_fail;
1385         if (!file->f_op->write)
1386                 goto close_fail;
1387         if (do_truncate(file->f_dentry, 0) != 0)
1388                 goto close_fail;
1389
1390         retval = binfmt->core_dump(signr, regs, file);
1391
1392         current->signal->group_exit_code |= 0x80;
1393 close_fail:
1394         filp_close(file, NULL);
1395 fail_unlock:
1396         complete_all(&mm->core_done);
1397 fail:
1398         unlock_kernel();
1399         return retval;
1400 }