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/key.h> 38 #include <linux/personality.h> 39 #include <linux/binfmts.h> 40 #include <linux/swap.h> 41 #include <linux/utsname.h> 42 #include <linux/module.h> 43 #include <linux/namei.h> 44 #include <linux/proc_fs.h> 45 #include <linux/ptrace.h> 46 #include <linux/mount.h> 47 #include <linux/security.h> 48 #include <linux/syscalls.h> 49 #include <linux/rmap.h> 50 #include <linux/acct.h> 51 52 #include <asm/uaccess.h> 53 #include <asm/mmu_context.h> 54 55 #ifdef CONFIG_KMOD 56 #include <linux/kmod.h> 57 #endif 58 59 int core_uses_pid; 60 char core_pattern[65] = "core"; 61 int suid_dumpable = 0; 62 63 EXPORT_SYMBOL(suid_dumpable); 64 /* The maximal length of core_pattern is also specified in sysctl.c */ 65 66 static struct linux_binfmt *formats; 67 static DEFINE_RWLOCK(binfmt_lock); 68 69 int register_binfmt(struct linux_binfmt * fmt) 70 { 71 struct linux_binfmt ** tmp = &formats; 72 73 if (!fmt) 74 return -EINVAL; 75 if (fmt->next) 76 return -EBUSY; 77 write_lock(&binfmt_lock); 78 while (*tmp) { 79 if (fmt == *tmp) { 80 write_unlock(&binfmt_lock); 81 return -EBUSY; 82 } 83 tmp = &(*tmp)->next; 84 } 85 fmt->next = formats; 86 formats = fmt; 87 write_unlock(&binfmt_lock); 88 return 0; 89 } 90 91 EXPORT_SYMBOL(register_binfmt); 92 93 int unregister_binfmt(struct linux_binfmt * fmt) 94 { 95 struct linux_binfmt ** tmp = &formats; 96 97 write_lock(&binfmt_lock); 98 while (*tmp) { 99 if (fmt == *tmp) { 100 *tmp = fmt->next; 101 write_unlock(&binfmt_lock); 102 return 0; 103 } 104 tmp = &(*tmp)->next; 105 } 106 write_unlock(&binfmt_lock); 107 return -EINVAL; 108 } 109 110 EXPORT_SYMBOL(unregister_binfmt); 111 112 static inline void put_binfmt(struct linux_binfmt * fmt) 113 { 114 module_put(fmt->module); 115 } 116 117 /* 118 * Note that a shared library must be both readable and executable due to 119 * security reasons. 120 * 121 * Also note that we take the address to load from from the file itself. 122 */ 123 asmlinkage long sys_uselib(const char __user * library) 124 { 125 struct file * file; 126 struct nameidata nd; 127 int error; 128 129 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ); 130 if (error) 131 goto out; 132 133 error = -EINVAL; 134 if (!S_ISREG(nd.dentry->d_inode->i_mode)) 135 goto exit; 136 137 error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd); 138 if (error) 139 goto exit; 140 141 file = nameidata_to_filp(&nd, O_RDONLY); 142 error = PTR_ERR(file); 143 if (IS_ERR(file)) 144 goto out; 145 146 error = -ENOEXEC; 147 if(file->f_op) { 148 struct linux_binfmt * fmt; 149 150 read_lock(&binfmt_lock); 151 for (fmt = formats ; fmt ; fmt = fmt->next) { 152 if (!fmt->load_shlib) 153 continue; 154 if (!try_module_get(fmt->module)) 155 continue; 156 read_unlock(&binfmt_lock); 157 error = fmt->load_shlib(file); 158 read_lock(&binfmt_lock); 159 put_binfmt(fmt); 160 if (error != -ENOEXEC) 161 break; 162 } 163 read_unlock(&binfmt_lock); 164 } 165 fput(file); 166 out: 167 return error; 168 exit: 169 release_open_intent(&nd); 170 path_release(&nd); 171 goto out; 172 } 173 174 /* 175 * count() counts the number of strings in array ARGV. 176 */ 177 static int count(char __user * __user * argv, int max) 178 { 179 int i = 0; 180 181 if (argv != NULL) { 182 for (;;) { 183 char __user * p; 184 185 if (get_user(p, argv)) 186 return -EFAULT; 187 if (!p) 188 break; 189 argv++; 190 if(++i > max) 191 return -E2BIG; 192 cond_resched(); 193 } 194 } 195 return i; 196 } 197 198 /* 199 * 'copy_strings()' copies argument/environment strings from user 200 * memory to free pages in kernel mem. These are in a format ready 201 * to be put directly into the top of new user memory. 202 */ 203 static int copy_strings(int argc, char __user * __user * argv, 204 struct linux_binprm *bprm) 205 { 206 struct page *kmapped_page = NULL; 207 char *kaddr = NULL; 208 int ret; 209 210 while (argc-- > 0) { 211 char __user *str; 212 int len; 213 unsigned long pos; 214 215 if (get_user(str, argv+argc) || 216 !(len = strnlen_user(str, bprm->p))) { 217 ret = -EFAULT; 218 goto out; 219 } 220 221 if (bprm->p < len) { 222 ret = -E2BIG; 223 goto out; 224 } 225 226 bprm->p -= len; 227 /* XXX: add architecture specific overflow check here. */ 228 pos = bprm->p; 229 230 while (len > 0) { 231 int i, new, err; 232 int offset, bytes_to_copy; 233 struct page *page; 234 235 offset = pos % PAGE_SIZE; 236 i = pos/PAGE_SIZE; 237 page = bprm->page[i]; 238 new = 0; 239 if (!page) { 240 page = alloc_page(GFP_HIGHUSER); 241 bprm->page[i] = page; 242 if (!page) { 243 ret = -ENOMEM; 244 goto out; 245 } 246 new = 1; 247 } 248 249 if (page != kmapped_page) { 250 if (kmapped_page) 251 kunmap(kmapped_page); 252 kmapped_page = page; 253 kaddr = kmap(kmapped_page); 254 } 255 if (new && offset) 256 memset(kaddr, 0, offset); 257 bytes_to_copy = PAGE_SIZE - offset; 258 if (bytes_to_copy > len) { 259 bytes_to_copy = len; 260 if (new) 261 memset(kaddr+offset+len, 0, 262 PAGE_SIZE-offset-len); 263 } 264 err = copy_from_user(kaddr+offset, str, bytes_to_copy); 265 if (err) { 266 ret = -EFAULT; 267 goto out; 268 } 269 270 pos += bytes_to_copy; 271 str += bytes_to_copy; 272 len -= bytes_to_copy; 273 } 274 } 275 ret = 0; 276 out: 277 if (kmapped_page) 278 kunmap(kmapped_page); 279 return ret; 280 } 281 282 /* 283 * Like copy_strings, but get argv and its values from kernel memory. 284 */ 285 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm) 286 { 287 int r; 288 mm_segment_t oldfs = get_fs(); 289 set_fs(KERNEL_DS); 290 r = copy_strings(argc, (char __user * __user *)argv, bprm); 291 set_fs(oldfs); 292 return r; 293 } 294 295 EXPORT_SYMBOL(copy_strings_kernel); 296 297 #ifdef CONFIG_MMU 298 /* 299 * This routine is used to map in a page into an address space: needed by 300 * execve() for the initial stack and environment pages. 301 * 302 * vma->vm_mm->mmap_sem is held for writing. 303 */ 304 void install_arg_page(struct vm_area_struct *vma, 305 struct page *page, unsigned long address) 306 { 307 struct mm_struct *mm = vma->vm_mm; 308 pgd_t * pgd; 309 pud_t * pud; 310 pmd_t * pmd; 311 pte_t * pte; 312 spinlock_t *ptl; 313 314 if (unlikely(anon_vma_prepare(vma))) 315 goto out; 316 317 flush_dcache_page(page); 318 pgd = pgd_offset(mm, address); 319 pud = pud_alloc(mm, pgd, address); 320 if (!pud) 321 goto out; 322 pmd = pmd_alloc(mm, pud, address); 323 if (!pmd) 324 goto out; 325 pte = pte_alloc_map_lock(mm, pmd, address, &ptl); 326 if (!pte) 327 goto out; 328 if (!pte_none(*pte)) { 329 pte_unmap_unlock(pte, ptl); 330 goto out; 331 } 332 inc_mm_counter(mm, anon_rss); 333 lru_cache_add_active(page); 334 set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte( 335 page, vma->vm_page_prot)))); 336 page_add_anon_rmap(page, vma, address); 337 pte_unmap_unlock(pte, ptl); 338 339 /* no need for flush_tlb */ 340 return; 341 out: 342 __free_page(page); 343 force_sig(SIGKILL, current); 344 } 345 346 #define EXTRA_STACK_VM_PAGES 20 /* random */ 347 348 int setup_arg_pages(struct linux_binprm *bprm, 349 unsigned long stack_top, 350 int executable_stack) 351 { 352 unsigned long stack_base; 353 struct vm_area_struct *mpnt; 354 struct mm_struct *mm = current->mm; 355 int i, ret; 356 long arg_size; 357 358 #ifdef CONFIG_STACK_GROWSUP 359 /* Move the argument and environment strings to the bottom of the 360 * stack space. 361 */ 362 int offset, j; 363 char *to, *from; 364 365 /* Start by shifting all the pages down */ 366 i = 0; 367 for (j = 0; j < MAX_ARG_PAGES; j++) { 368 struct page *page = bprm->page[j]; 369 if (!page) 370 continue; 371 bprm->page[i++] = page; 372 } 373 374 /* Now move them within their pages */ 375 offset = bprm->p % PAGE_SIZE; 376 to = kmap(bprm->page[0]); 377 for (j = 1; j < i; j++) { 378 memmove(to, to + offset, PAGE_SIZE - offset); 379 from = kmap(bprm->page[j]); 380 memcpy(to + PAGE_SIZE - offset, from, offset); 381 kunmap(bprm->page[j - 1]); 382 to = from; 383 } 384 memmove(to, to + offset, PAGE_SIZE - offset); 385 kunmap(bprm->page[j - 1]); 386 387 /* Limit stack size to 1GB */ 388 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max; 389 if (stack_base > (1 << 30)) 390 stack_base = 1 << 30; 391 stack_base = PAGE_ALIGN(stack_top - stack_base); 392 393 /* Adjust bprm->p to point to the end of the strings. */ 394 bprm->p = stack_base + PAGE_SIZE * i - offset; 395 396 mm->arg_start = stack_base; 397 arg_size = i << PAGE_SHIFT; 398 399 /* zero pages that were copied above */ 400 while (i < MAX_ARG_PAGES) 401 bprm->page[i++] = NULL; 402 #else 403 stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE); 404 stack_base = PAGE_ALIGN(stack_base); 405 bprm->p += stack_base; 406 mm->arg_start = bprm->p; 407 arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start); 408 #endif 409 410 arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE; 411 412 if (bprm->loader) 413 bprm->loader += stack_base; 414 bprm->exec += stack_base; 415 416 mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); 417 if (!mpnt) 418 return -ENOMEM; 419 420 memset(mpnt, 0, sizeof(*mpnt)); 421 422 down_write(&mm->mmap_sem); 423 { 424 mpnt->vm_mm = mm; 425 #ifdef CONFIG_STACK_GROWSUP 426 mpnt->vm_start = stack_base; 427 mpnt->vm_end = stack_base + arg_size; 428 #else 429 mpnt->vm_end = stack_top; 430 mpnt->vm_start = mpnt->vm_end - arg_size; 431 #endif 432 /* Adjust stack execute permissions; explicitly enable 433 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X 434 * and leave alone (arch default) otherwise. */ 435 if (unlikely(executable_stack == EXSTACK_ENABLE_X)) 436 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC; 437 else if (executable_stack == EXSTACK_DISABLE_X) 438 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC; 439 else 440 mpnt->vm_flags = VM_STACK_FLAGS; 441 mpnt->vm_flags |= mm->def_flags; 442 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7]; 443 if ((ret = insert_vm_struct(mm, mpnt))) { 444 up_write(&mm->mmap_sem); 445 kmem_cache_free(vm_area_cachep, mpnt); 446 return ret; 447 } 448 mm->stack_vm = mm->total_vm = vma_pages(mpnt); 449 } 450 451 for (i = 0 ; i < MAX_ARG_PAGES ; i++) { 452 struct page *page = bprm->page[i]; 453 if (page) { 454 bprm->page[i] = NULL; 455 install_arg_page(mpnt, page, stack_base); 456 } 457 stack_base += PAGE_SIZE; 458 } 459 up_write(&mm->mmap_sem); 460 461 return 0; 462 } 463 464 EXPORT_SYMBOL(setup_arg_pages); 465 466 #define free_arg_pages(bprm) do { } while (0) 467 468 #else 469 470 static inline void free_arg_pages(struct linux_binprm *bprm) 471 { 472 int i; 473 474 for (i = 0; i < MAX_ARG_PAGES; i++) { 475 if (bprm->page[i]) 476 __free_page(bprm->page[i]); 477 bprm->page[i] = NULL; 478 } 479 } 480 481 #endif /* CONFIG_MMU */ 482 483 struct file *open_exec(const char *name) 484 { 485 struct nameidata nd; 486 int err; 487 struct file *file; 488 489 err = path_lookup_open(name, LOOKUP_FOLLOW, &nd, FMODE_READ); 490 file = ERR_PTR(err); 491 492 if (!err) { 493 struct inode *inode = nd.dentry->d_inode; 494 file = ERR_PTR(-EACCES); 495 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) && 496 S_ISREG(inode->i_mode)) { 497 int err = permission(inode, MAY_EXEC, &nd); 498 if (!err && !(inode->i_mode & 0111)) 499 err = -EACCES; 500 file = ERR_PTR(err); 501 if (!err) { 502 file = nameidata_to_filp(&nd, O_RDONLY); 503 if (!IS_ERR(file)) { 504 err = deny_write_access(file); 505 if (err) { 506 fput(file); 507 file = ERR_PTR(err); 508 } 509 } 510 out: 511 return file; 512 } 513 } 514 release_open_intent(&nd); 515 path_release(&nd); 516 } 517 goto out; 518 } 519 520 EXPORT_SYMBOL(open_exec); 521 522 int kernel_read(struct file *file, unsigned long offset, 523 char *addr, unsigned long count) 524 { 525 mm_segment_t old_fs; 526 loff_t pos = offset; 527 int result; 528 529 old_fs = get_fs(); 530 set_fs(get_ds()); 531 /* The cast to a user pointer is valid due to the set_fs() */ 532 result = vfs_read(file, (void __user *)addr, count, &pos); 533 set_fs(old_fs); 534 return result; 535 } 536 537 EXPORT_SYMBOL(kernel_read); 538 539 static int exec_mmap(struct mm_struct *mm) 540 { 541 struct task_struct *tsk; 542 struct mm_struct * old_mm, *active_mm; 543 544 /* Notify parent that we're no longer interested in the old VM */ 545 tsk = current; 546 old_mm = current->mm; 547 mm_release(tsk, old_mm); 548 549 if (old_mm) { 550 /* 551 * Make sure that if there is a core dump in progress 552 * for the old mm, we get out and die instead of going 553 * through with the exec. We must hold mmap_sem around 554 * checking core_waiters and changing tsk->mm. The 555 * core-inducing thread will increment core_waiters for 556 * each thread whose ->mm == old_mm. 557 */ 558 down_read(&old_mm->mmap_sem); 559 if (unlikely(old_mm->core_waiters)) { 560 up_read(&old_mm->mmap_sem); 561 return -EINTR; 562 } 563 } 564 task_lock(tsk); 565 active_mm = tsk->active_mm; 566 tsk->mm = mm; 567 tsk->active_mm = mm; 568 activate_mm(active_mm, mm); 569 task_unlock(tsk); 570 arch_pick_mmap_layout(mm); 571 if (old_mm) { 572 up_read(&old_mm->mmap_sem); 573 if (active_mm != old_mm) BUG(); 574 mmput(old_mm); 575 return 0; 576 } 577 mmdrop(active_mm); 578 return 0; 579 } 580 581 /* 582 * This function makes sure the current process has its own signal table, 583 * so that flush_signal_handlers can later reset the handlers without 584 * disturbing other processes. (Other processes might share the signal 585 * table via the CLONE_SIGHAND option to clone().) 586 */ 587 static inline int de_thread(struct task_struct *tsk) 588 { 589 struct signal_struct *sig = tsk->signal; 590 struct sighand_struct *newsighand, *oldsighand = tsk->sighand; 591 spinlock_t *lock = &oldsighand->siglock; 592 int count; 593 594 /* 595 * If we don't share sighandlers, then we aren't sharing anything 596 * and we can just re-use it all. 597 */ 598 if (atomic_read(&oldsighand->count) <= 1) { 599 BUG_ON(atomic_read(&sig->count) != 1); 600 exit_itimers(sig); 601 return 0; 602 } 603 604 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 605 if (!newsighand) 606 return -ENOMEM; 607 608 if (thread_group_empty(current)) 609 goto no_thread_group; 610 611 /* 612 * Kill all other threads in the thread group. 613 * We must hold tasklist_lock to call zap_other_threads. 614 */ 615 read_lock(&tasklist_lock); 616 spin_lock_irq(lock); 617 if (sig->flags & SIGNAL_GROUP_EXIT) { 618 /* 619 * Another group action in progress, just 620 * return so that the signal is processed. 621 */ 622 spin_unlock_irq(lock); 623 read_unlock(&tasklist_lock); 624 kmem_cache_free(sighand_cachep, newsighand); 625 return -EAGAIN; 626 } 627 zap_other_threads(current); 628 read_unlock(&tasklist_lock); 629 630 /* 631 * Account for the thread group leader hanging around: 632 */ 633 count = 1; 634 if (!thread_group_leader(current)) { 635 count = 2; 636 /* 637 * The SIGALRM timer survives the exec, but needs to point 638 * at us as the new group leader now. We have a race with 639 * a timer firing now getting the old leader, so we need to 640 * synchronize with any firing (by calling del_timer_sync) 641 * before we can safely let the old group leader die. 642 */ 643 sig->real_timer.data = (unsigned long)current; 644 spin_unlock_irq(lock); 645 if (del_timer_sync(&sig->real_timer)) 646 add_timer(&sig->real_timer); 647 spin_lock_irq(lock); 648 } 649 while (atomic_read(&sig->count) > count) { 650 sig->group_exit_task = current; 651 sig->notify_count = count; 652 __set_current_state(TASK_UNINTERRUPTIBLE); 653 spin_unlock_irq(lock); 654 schedule(); 655 spin_lock_irq(lock); 656 } 657 sig->group_exit_task = NULL; 658 sig->notify_count = 0; 659 spin_unlock_irq(lock); 660 661 /* 662 * At this point all other threads have exited, all we have to 663 * do is to wait for the thread group leader to become inactive, 664 * and to assume its PID: 665 */ 666 if (!thread_group_leader(current)) { 667 struct task_struct *leader = current->group_leader, *parent; 668 struct dentry *proc_dentry1, *proc_dentry2; 669 unsigned long exit_state, ptrace; 670 671 /* 672 * Wait for the thread group leader to be a zombie. 673 * It should already be zombie at this point, most 674 * of the time. 675 */ 676 while (leader->exit_state != EXIT_ZOMBIE) 677 yield(); 678 679 spin_lock(&leader->proc_lock); 680 spin_lock(¤t->proc_lock); 681 proc_dentry1 = proc_pid_unhash(current); 682 proc_dentry2 = proc_pid_unhash(leader); 683 write_lock_irq(&tasklist_lock); 684 685 BUG_ON(leader->tgid != current->tgid); 686 BUG_ON(current->pid == current->tgid); 687 /* 688 * An exec() starts a new thread group with the 689 * TGID of the previous thread group. Rehash the 690 * two threads with a switched PID, and release 691 * the former thread group leader: 692 */ 693 ptrace = leader->ptrace; 694 parent = leader->parent; 695 if (unlikely(ptrace) && unlikely(parent == current)) { 696 /* 697 * Joker was ptracing his own group leader, 698 * and now he wants to be his own parent! 699 * We can't have that. 700 */ 701 ptrace = 0; 702 } 703 704 ptrace_unlink(current); 705 ptrace_unlink(leader); 706 remove_parent(current); 707 remove_parent(leader); 708 709 switch_exec_pids(leader, current); 710 711 current->parent = current->real_parent = leader->real_parent; 712 leader->parent = leader->real_parent = child_reaper; 713 current->group_leader = current; 714 leader->group_leader = leader; 715 716 add_parent(current, current->parent); 717 add_parent(leader, leader->parent); 718 if (ptrace) { 719 current->ptrace = ptrace; 720 __ptrace_link(current, parent); 721 } 722 723 list_del(¤t->tasks); 724 list_add_tail(¤t->tasks, &init_task.tasks); 725 current->exit_signal = SIGCHLD; 726 exit_state = leader->exit_state; 727 728 write_unlock_irq(&tasklist_lock); 729 spin_unlock(&leader->proc_lock); 730 spin_unlock(¤t->proc_lock); 731 proc_pid_flush(proc_dentry1); 732 proc_pid_flush(proc_dentry2); 733 734 BUG_ON(exit_state != EXIT_ZOMBIE); 735 release_task(leader); 736 } 737 738 /* 739 * There may be one thread left which is just exiting, 740 * but it's safe to stop telling the group to kill themselves. 741 */ 742 sig->flags = 0; 743 744 no_thread_group: 745 BUG_ON(atomic_read(&sig->count) != 1); 746 exit_itimers(sig); 747 748 if (atomic_read(&oldsighand->count) == 1) { 749 /* 750 * Now that we nuked the rest of the thread group, 751 * it turns out we are not sharing sighand any more either. 752 * So we can just keep it. 753 */ 754 kmem_cache_free(sighand_cachep, newsighand); 755 } else { 756 /* 757 * Move our state over to newsighand and switch it in. 758 */ 759 spin_lock_init(&newsighand->siglock); 760 atomic_set(&newsighand->count, 1); 761 memcpy(newsighand->action, oldsighand->action, 762 sizeof(newsighand->action)); 763 764 write_lock_irq(&tasklist_lock); 765 spin_lock(&oldsighand->siglock); 766 spin_lock(&newsighand->siglock); 767 768 current->sighand = newsighand; 769 recalc_sigpending(); 770 771 spin_unlock(&newsighand->siglock); 772 spin_unlock(&oldsighand->siglock); 773 write_unlock_irq(&tasklist_lock); 774 775 if (atomic_dec_and_test(&oldsighand->count)) 776 kmem_cache_free(sighand_cachep, oldsighand); 777 } 778 779 BUG_ON(!thread_group_leader(current)); 780 return 0; 781 } 782 783 /* 784 * These functions flushes out all traces of the currently running executable 785 * so that a new one can be started 786 */ 787 788 static inline void flush_old_files(struct files_struct * files) 789 { 790 long j = -1; 791 struct fdtable *fdt; 792 793 spin_lock(&files->file_lock); 794 for (;;) { 795 unsigned long set, i; 796 797 j++; 798 i = j * __NFDBITS; 799 fdt = files_fdtable(files); 800 if (i >= fdt->max_fds || i >= fdt->max_fdset) 801 break; 802 set = fdt->close_on_exec->fds_bits[j]; 803 if (!set) 804 continue; 805 fdt->close_on_exec->fds_bits[j] = 0; 806 spin_unlock(&files->file_lock); 807 for ( ; set ; i++,set >>= 1) { 808 if (set & 1) { 809 sys_close(i); 810 } 811 } 812 spin_lock(&files->file_lock); 813 814 } 815 spin_unlock(&files->file_lock); 816 } 817 818 void get_task_comm(char *buf, struct task_struct *tsk) 819 { 820 /* buf must be at least sizeof(tsk->comm) in size */ 821 task_lock(tsk); 822 strncpy(buf, tsk->comm, sizeof(tsk->comm)); 823 task_unlock(tsk); 824 } 825 826 void set_task_comm(struct task_struct *tsk, char *buf) 827 { 828 task_lock(tsk); 829 strlcpy(tsk->comm, buf, sizeof(tsk->comm)); 830 task_unlock(tsk); 831 } 832 833 int flush_old_exec(struct linux_binprm * bprm) 834 { 835 char * name; 836 int i, ch, retval; 837 struct files_struct *files; 838 char tcomm[sizeof(current->comm)]; 839 840 /* 841 * Make sure we have a private signal table and that 842 * we are unassociated from the previous thread group. 843 */ 844 retval = de_thread(current); 845 if (retval) 846 goto out; 847 848 /* 849 * Make sure we have private file handles. Ask the 850 * fork helper to do the work for us and the exit 851 * helper to do the cleanup of the old one. 852 */ 853 files = current->files; /* refcounted so safe to hold */ 854 retval = unshare_files(); 855 if (retval) 856 goto out; 857 /* 858 * Release all of the old mmap stuff 859 */ 860 retval = exec_mmap(bprm->mm); 861 if (retval) 862 goto mmap_failed; 863 864 bprm->mm = NULL; /* We're using it now */ 865 866 /* This is the point of no return */ 867 steal_locks(files); 868 put_files_struct(files); 869 870 current->sas_ss_sp = current->sas_ss_size = 0; 871 872 if (current->euid == current->uid && current->egid == current->gid) 873 current->mm->dumpable = 1; 874 else 875 current->mm->dumpable = suid_dumpable; 876 877 name = bprm->filename; 878 879 /* Copies the binary name from after last slash */ 880 for (i=0; (ch = *(name++)) != '\0';) { 881 if (ch == '/') 882 i = 0; /* overwrite what we wrote */ 883 else 884 if (i < (sizeof(tcomm) - 1)) 885 tcomm[i++] = ch; 886 } 887 tcomm[i] = '\0'; 888 set_task_comm(current, tcomm); 889 890 current->flags &= ~PF_RANDOMIZE; 891 flush_thread(); 892 893 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 894 permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) || 895 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) { 896 suid_keys(current); 897 current->mm->dumpable = suid_dumpable; 898 } 899 900 /* An exec changes our domain. We are no longer part of the thread 901 group */ 902 903 current->self_exec_id++; 904 905 flush_signal_handlers(current, 0); 906 flush_old_files(current->files); 907 908 return 0; 909 910 mmap_failed: 911 put_files_struct(current->files); 912 current->files = files; 913 out: 914 return retval; 915 } 916 917 EXPORT_SYMBOL(flush_old_exec); 918 919 /* 920 * Fill the binprm structure from the inode. 921 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes 922 */ 923 int prepare_binprm(struct linux_binprm *bprm) 924 { 925 int mode; 926 struct inode * inode = bprm->file->f_dentry->d_inode; 927 int retval; 928 929 mode = inode->i_mode; 930 /* 931 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE, 932 * generic_permission lets a non-executable through 933 */ 934 if (!(mode & 0111)) /* with at least _one_ execute bit set */ 935 return -EACCES; 936 if (bprm->file->f_op == NULL) 937 return -EACCES; 938 939 bprm->e_uid = current->euid; 940 bprm->e_gid = current->egid; 941 942 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) { 943 /* Set-uid? */ 944 if (mode & S_ISUID) { 945 current->personality &= ~PER_CLEAR_ON_SETID; 946 bprm->e_uid = inode->i_uid; 947 } 948 949 /* Set-gid? */ 950 /* 951 * If setgid is set but no group execute bit then this 952 * is a candidate for mandatory locking, not a setgid 953 * executable. 954 */ 955 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) { 956 current->personality &= ~PER_CLEAR_ON_SETID; 957 bprm->e_gid = inode->i_gid; 958 } 959 } 960 961 /* fill in binprm security blob */ 962 retval = security_bprm_set(bprm); 963 if (retval) 964 return retval; 965 966 memset(bprm->buf,0,BINPRM_BUF_SIZE); 967 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE); 968 } 969 970 EXPORT_SYMBOL(prepare_binprm); 971 972 static inline int unsafe_exec(struct task_struct *p) 973 { 974 int unsafe = 0; 975 if (p->ptrace & PT_PTRACED) { 976 if (p->ptrace & PT_PTRACE_CAP) 977 unsafe |= LSM_UNSAFE_PTRACE_CAP; 978 else 979 unsafe |= LSM_UNSAFE_PTRACE; 980 } 981 if (atomic_read(&p->fs->count) > 1 || 982 atomic_read(&p->files->count) > 1 || 983 atomic_read(&p->sighand->count) > 1) 984 unsafe |= LSM_UNSAFE_SHARE; 985 986 return unsafe; 987 } 988 989 void compute_creds(struct linux_binprm *bprm) 990 { 991 int unsafe; 992 993 if (bprm->e_uid != current->uid) 994 suid_keys(current); 995 exec_keys(current); 996 997 task_lock(current); 998 unsafe = unsafe_exec(current); 999 security_bprm_apply_creds(bprm, unsafe); 1000 task_unlock(current); 1001 security_bprm_post_apply_creds(bprm); 1002 } 1003 1004 EXPORT_SYMBOL(compute_creds); 1005 1006 void remove_arg_zero(struct linux_binprm *bprm) 1007 { 1008 if (bprm->argc) { 1009 unsigned long offset; 1010 char * kaddr; 1011 struct page *page; 1012 1013 offset = bprm->p % PAGE_SIZE; 1014 goto inside; 1015 1016 while (bprm->p++, *(kaddr+offset++)) { 1017 if (offset != PAGE_SIZE) 1018 continue; 1019 offset = 0; 1020 kunmap_atomic(kaddr, KM_USER0); 1021 inside: 1022 page = bprm->page[bprm->p/PAGE_SIZE]; 1023 kaddr = kmap_atomic(page, KM_USER0); 1024 } 1025 kunmap_atomic(kaddr, KM_USER0); 1026 bprm->argc--; 1027 } 1028 } 1029 1030 EXPORT_SYMBOL(remove_arg_zero); 1031 1032 /* 1033 * cycle the list of binary formats handler, until one recognizes the image 1034 */ 1035 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs) 1036 { 1037 int try,retval; 1038 struct linux_binfmt *fmt; 1039 #ifdef __alpha__ 1040 /* handle /sbin/loader.. */ 1041 { 1042 struct exec * eh = (struct exec *) bprm->buf; 1043 1044 if (!bprm->loader && eh->fh.f_magic == 0x183 && 1045 (eh->fh.f_flags & 0x3000) == 0x3000) 1046 { 1047 struct file * file; 1048 unsigned long loader; 1049 1050 allow_write_access(bprm->file); 1051 fput(bprm->file); 1052 bprm->file = NULL; 1053 1054 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *); 1055 1056 file = open_exec("/sbin/loader"); 1057 retval = PTR_ERR(file); 1058 if (IS_ERR(file)) 1059 return retval; 1060 1061 /* Remember if the application is TASO. */ 1062 bprm->sh_bang = eh->ah.entry < 0x100000000UL; 1063 1064 bprm->file = file; 1065 bprm->loader = loader; 1066 retval = prepare_binprm(bprm); 1067 if (retval<0) 1068 return retval; 1069 /* should call search_binary_handler recursively here, 1070 but it does not matter */ 1071 } 1072 } 1073 #endif 1074 retval = security_bprm_check(bprm); 1075 if (retval) 1076 return retval; 1077 1078 /* kernel module loader fixup */ 1079 /* so we don't try to load run modprobe in kernel space. */ 1080 set_fs(USER_DS); 1081 retval = -ENOENT; 1082 for (try=0; try<2; try++) { 1083 read_lock(&binfmt_lock); 1084 for (fmt = formats ; fmt ; fmt = fmt->next) { 1085 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary; 1086 if (!fn) 1087 continue; 1088 if (!try_module_get(fmt->module)) 1089 continue; 1090 read_unlock(&binfmt_lock); 1091 retval = fn(bprm, regs); 1092 if (retval >= 0) { 1093 put_binfmt(fmt); 1094 allow_write_access(bprm->file); 1095 if (bprm->file) 1096 fput(bprm->file); 1097 bprm->file = NULL; 1098 current->did_exec = 1; 1099 return retval; 1100 } 1101 read_lock(&binfmt_lock); 1102 put_binfmt(fmt); 1103 if (retval != -ENOEXEC || bprm->mm == NULL) 1104 break; 1105 if (!bprm->file) { 1106 read_unlock(&binfmt_lock); 1107 return retval; 1108 } 1109 } 1110 read_unlock(&binfmt_lock); 1111 if (retval != -ENOEXEC || bprm->mm == NULL) { 1112 break; 1113 #ifdef CONFIG_KMOD 1114 }else{ 1115 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e)) 1116 if (printable(bprm->buf[0]) && 1117 printable(bprm->buf[1]) && 1118 printable(bprm->buf[2]) && 1119 printable(bprm->buf[3])) 1120 break; /* -ENOEXEC */ 1121 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2])); 1122 #endif 1123 } 1124 } 1125 return retval; 1126 } 1127 1128 EXPORT_SYMBOL(search_binary_handler); 1129 1130 /* 1131 * sys_execve() executes a new program. 1132 */ 1133 int do_execve(char * filename, 1134 char __user *__user *argv, 1135 char __user *__user *envp, 1136 struct pt_regs * regs) 1137 { 1138 struct linux_binprm *bprm; 1139 struct file *file; 1140 int retval; 1141 int i; 1142 1143 retval = -ENOMEM; 1144 bprm = kmalloc(sizeof(*bprm), GFP_KERNEL); 1145 if (!bprm) 1146 goto out_ret; 1147 memset(bprm, 0, sizeof(*bprm)); 1148 1149 file = open_exec(filename); 1150 retval = PTR_ERR(file); 1151 if (IS_ERR(file)) 1152 goto out_kfree; 1153 1154 sched_exec(); 1155 1156 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *); 1157 1158 bprm->file = file; 1159 bprm->filename = filename; 1160 bprm->interp = filename; 1161 bprm->mm = mm_alloc(); 1162 retval = -ENOMEM; 1163 if (!bprm->mm) 1164 goto out_file; 1165 1166 retval = init_new_context(current, bprm->mm); 1167 if (retval < 0) 1168 goto out_mm; 1169 1170 bprm->argc = count(argv, bprm->p / sizeof(void *)); 1171 if ((retval = bprm->argc) < 0) 1172 goto out_mm; 1173 1174 bprm->envc = count(envp, bprm->p / sizeof(void *)); 1175 if ((retval = bprm->envc) < 0) 1176 goto out_mm; 1177 1178 retval = security_bprm_alloc(bprm); 1179 if (retval) 1180 goto out; 1181 1182 retval = prepare_binprm(bprm); 1183 if (retval < 0) 1184 goto out; 1185 1186 retval = copy_strings_kernel(1, &bprm->filename, bprm); 1187 if (retval < 0) 1188 goto out; 1189 1190 bprm->exec = bprm->p; 1191 retval = copy_strings(bprm->envc, envp, bprm); 1192 if (retval < 0) 1193 goto out; 1194 1195 retval = copy_strings(bprm->argc, argv, bprm); 1196 if (retval < 0) 1197 goto out; 1198 1199 retval = search_binary_handler(bprm,regs); 1200 if (retval >= 0) { 1201 free_arg_pages(bprm); 1202 1203 /* execve success */ 1204 security_bprm_free(bprm); 1205 acct_update_integrals(current); 1206 kfree(bprm); 1207 return retval; 1208 } 1209 1210 out: 1211 /* Something went wrong, return the inode and free the argument pages*/ 1212 for (i = 0 ; i < MAX_ARG_PAGES ; i++) { 1213 struct page * page = bprm->page[i]; 1214 if (page) 1215 __free_page(page); 1216 } 1217 1218 if (bprm->security) 1219 security_bprm_free(bprm); 1220 1221 out_mm: 1222 if (bprm->mm) 1223 mmdrop(bprm->mm); 1224 1225 out_file: 1226 if (bprm->file) { 1227 allow_write_access(bprm->file); 1228 fput(bprm->file); 1229 } 1230 1231 out_kfree: 1232 kfree(bprm); 1233 1234 out_ret: 1235 return retval; 1236 } 1237 1238 int set_binfmt(struct linux_binfmt *new) 1239 { 1240 struct linux_binfmt *old = current->binfmt; 1241 1242 if (new) { 1243 if (!try_module_get(new->module)) 1244 return -1; 1245 } 1246 current->binfmt = new; 1247 if (old) 1248 module_put(old->module); 1249 return 0; 1250 } 1251 1252 EXPORT_SYMBOL(set_binfmt); 1253 1254 #define CORENAME_MAX_SIZE 64 1255 1256 /* format_corename will inspect the pattern parameter, and output a 1257 * name into corename, which must have space for at least 1258 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator. 1259 */ 1260 static void format_corename(char *corename, const char *pattern, long signr) 1261 { 1262 const char *pat_ptr = pattern; 1263 char *out_ptr = corename; 1264 char *const out_end = corename + CORENAME_MAX_SIZE; 1265 int rc; 1266 int pid_in_pattern = 0; 1267 1268 /* Repeat as long as we have more pattern to process and more output 1269 space */ 1270 while (*pat_ptr) { 1271 if (*pat_ptr != '%') { 1272 if (out_ptr == out_end) 1273 goto out; 1274 *out_ptr++ = *pat_ptr++; 1275 } else { 1276 switch (*++pat_ptr) { 1277 case 0: 1278 goto out; 1279 /* Double percent, output one percent */ 1280 case '%': 1281 if (out_ptr == out_end) 1282 goto out; 1283 *out_ptr++ = '%'; 1284 break; 1285 /* pid */ 1286 case 'p': 1287 pid_in_pattern = 1; 1288 rc = snprintf(out_ptr, out_end - out_ptr, 1289 "%d", current->tgid); 1290 if (rc > out_end - out_ptr) 1291 goto out; 1292 out_ptr += rc; 1293 break; 1294 /* uid */ 1295 case 'u': 1296 rc = snprintf(out_ptr, out_end - out_ptr, 1297 "%d", current->uid); 1298 if (rc > out_end - out_ptr) 1299 goto out; 1300 out_ptr += rc; 1301 break; 1302 /* gid */ 1303 case 'g': 1304 rc = snprintf(out_ptr, out_end - out_ptr, 1305 "%d", current->gid); 1306 if (rc > out_end - out_ptr) 1307 goto out; 1308 out_ptr += rc; 1309 break; 1310 /* signal that caused the coredump */ 1311 case 's': 1312 rc = snprintf(out_ptr, out_end - out_ptr, 1313 "%ld", signr); 1314 if (rc > out_end - out_ptr) 1315 goto out; 1316 out_ptr += rc; 1317 break; 1318 /* UNIX time of coredump */ 1319 case 't': { 1320 struct timeval tv; 1321 do_gettimeofday(&tv); 1322 rc = snprintf(out_ptr, out_end - out_ptr, 1323 "%lu", tv.tv_sec); 1324 if (rc > out_end - out_ptr) 1325 goto out; 1326 out_ptr += rc; 1327 break; 1328 } 1329 /* hostname */ 1330 case 'h': 1331 down_read(&uts_sem); 1332 rc = snprintf(out_ptr, out_end - out_ptr, 1333 "%s", system_utsname.nodename); 1334 up_read(&uts_sem); 1335 if (rc > out_end - out_ptr) 1336 goto out; 1337 out_ptr += rc; 1338 break; 1339 /* executable */ 1340 case 'e': 1341 rc = snprintf(out_ptr, out_end - out_ptr, 1342 "%s", current->comm); 1343 if (rc > out_end - out_ptr) 1344 goto out; 1345 out_ptr += rc; 1346 break; 1347 default: 1348 break; 1349 } 1350 ++pat_ptr; 1351 } 1352 } 1353 /* Backward compatibility with core_uses_pid: 1354 * 1355 * If core_pattern does not include a %p (as is the default) 1356 * and core_uses_pid is set, then .%pid will be appended to 1357 * the filename */ 1358 if (!pid_in_pattern 1359 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) { 1360 rc = snprintf(out_ptr, out_end - out_ptr, 1361 ".%d", current->tgid); 1362 if (rc > out_end - out_ptr) 1363 goto out; 1364 out_ptr += rc; 1365 } 1366 out: 1367 *out_ptr = 0; 1368 } 1369 1370 static void zap_threads (struct mm_struct *mm) 1371 { 1372 struct task_struct *g, *p; 1373 struct task_struct *tsk = current; 1374 struct completion *vfork_done = tsk->vfork_done; 1375 int traced = 0; 1376 1377 /* 1378 * Make sure nobody is waiting for us to release the VM, 1379 * otherwise we can deadlock when we wait on each other 1380 */ 1381 if (vfork_done) { 1382 tsk->vfork_done = NULL; 1383 complete(vfork_done); 1384 } 1385 1386 read_lock(&tasklist_lock); 1387 do_each_thread(g,p) 1388 if (mm == p->mm && p != tsk) { 1389 force_sig_specific(SIGKILL, p); 1390 mm->core_waiters++; 1391 if (unlikely(p->ptrace) && 1392 unlikely(p->parent->mm == mm)) 1393 traced = 1; 1394 } 1395 while_each_thread(g,p); 1396 1397 read_unlock(&tasklist_lock); 1398 1399 if (unlikely(traced)) { 1400 /* 1401 * We are zapping a thread and the thread it ptraces. 1402 * If the tracee went into a ptrace stop for exit tracing, 1403 * we could deadlock since the tracer is waiting for this 1404 * coredump to finish. Detach them so they can both die. 1405 */ 1406 write_lock_irq(&tasklist_lock); 1407 do_each_thread(g,p) { 1408 if (mm == p->mm && p != tsk && 1409 p->ptrace && p->parent->mm == mm) { 1410 __ptrace_unlink(p); 1411 } 1412 } while_each_thread(g,p); 1413 write_unlock_irq(&tasklist_lock); 1414 } 1415 } 1416 1417 static void coredump_wait(struct mm_struct *mm) 1418 { 1419 DECLARE_COMPLETION(startup_done); 1420 int core_waiters; 1421 1422 mm->core_startup_done = &startup_done; 1423 1424 zap_threads(mm); 1425 core_waiters = mm->core_waiters; 1426 up_write(&mm->mmap_sem); 1427 1428 if (core_waiters) 1429 wait_for_completion(&startup_done); 1430 BUG_ON(mm->core_waiters); 1431 } 1432 1433 int do_coredump(long signr, int exit_code, struct pt_regs * regs) 1434 { 1435 char corename[CORENAME_MAX_SIZE + 1]; 1436 struct mm_struct *mm = current->mm; 1437 struct linux_binfmt * binfmt; 1438 struct inode * inode; 1439 struct file * file; 1440 int retval = 0; 1441 int fsuid = current->fsuid; 1442 int flag = 0; 1443 1444 binfmt = current->binfmt; 1445 if (!binfmt || !binfmt->core_dump) 1446 goto fail; 1447 down_write(&mm->mmap_sem); 1448 if (!mm->dumpable) { 1449 up_write(&mm->mmap_sem); 1450 goto fail; 1451 } 1452 1453 /* 1454 * We cannot trust fsuid as being the "true" uid of the 1455 * process nor do we know its entire history. We only know it 1456 * was tainted so we dump it as root in mode 2. 1457 */ 1458 if (mm->dumpable == 2) { /* Setuid core dump mode */ 1459 flag = O_EXCL; /* Stop rewrite attacks */ 1460 current->fsuid = 0; /* Dump root private */ 1461 } 1462 mm->dumpable = 0; 1463 1464 retval = -EAGAIN; 1465 spin_lock_irq(¤t->sighand->siglock); 1466 if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) { 1467 current->signal->flags = SIGNAL_GROUP_EXIT; 1468 current->signal->group_exit_code = exit_code; 1469 retval = 0; 1470 } 1471 spin_unlock_irq(¤t->sighand->siglock); 1472 if (retval) { 1473 up_write(&mm->mmap_sem); 1474 goto fail; 1475 } 1476 1477 init_completion(&mm->core_done); 1478 coredump_wait(mm); 1479 1480 /* 1481 * Clear any false indication of pending signals that might 1482 * be seen by the filesystem code called to write the core file. 1483 */ 1484 current->signal->group_stop_count = 0; 1485 clear_thread_flag(TIF_SIGPENDING); 1486 1487 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump) 1488 goto fail_unlock; 1489 1490 /* 1491 * lock_kernel() because format_corename() is controlled by sysctl, which 1492 * uses lock_kernel() 1493 */ 1494 lock_kernel(); 1495 format_corename(corename, core_pattern, signr); 1496 unlock_kernel(); 1497 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600); 1498 if (IS_ERR(file)) 1499 goto fail_unlock; 1500 inode = file->f_dentry->d_inode; 1501 if (inode->i_nlink > 1) 1502 goto close_fail; /* multiple links - don't dump */ 1503 if (d_unhashed(file->f_dentry)) 1504 goto close_fail; 1505 1506 if (!S_ISREG(inode->i_mode)) 1507 goto close_fail; 1508 if (!file->f_op) 1509 goto close_fail; 1510 if (!file->f_op->write) 1511 goto close_fail; 1512 if (do_truncate(file->f_dentry, 0) != 0) 1513 goto close_fail; 1514 1515 retval = binfmt->core_dump(signr, regs, file); 1516 1517 if (retval) 1518 current->signal->group_exit_code |= 0x80; 1519 close_fail: 1520 filp_close(file, NULL); 1521 fail_unlock: 1522 current->fsuid = fsuid; 1523 complete_all(&mm->core_done); 1524 fail: 1525 return retval; 1526 } 1527