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