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 if (security_vm_enough_memory(arg_size >> PAGE_SHIFT)) { 425 kmem_cache_free(vm_area_cachep, mpnt); 426 return -ENOMEM; 427 } 428 429 memset(mpnt, 0, sizeof(*mpnt)); 430 431 down_write(&mm->mmap_sem); 432 { 433 mpnt->vm_mm = mm; 434 #ifdef CONFIG_STACK_GROWSUP 435 mpnt->vm_start = stack_base; 436 mpnt->vm_end = stack_base + arg_size; 437 #else 438 mpnt->vm_end = stack_top; 439 mpnt->vm_start = mpnt->vm_end - arg_size; 440 #endif 441 /* Adjust stack execute permissions; explicitly enable 442 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X 443 * and leave alone (arch default) otherwise. */ 444 if (unlikely(executable_stack == EXSTACK_ENABLE_X)) 445 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC; 446 else if (executable_stack == EXSTACK_DISABLE_X) 447 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC; 448 else 449 mpnt->vm_flags = VM_STACK_FLAGS; 450 mpnt->vm_flags |= mm->def_flags; 451 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7]; 452 if ((ret = insert_vm_struct(mm, mpnt))) { 453 up_write(&mm->mmap_sem); 454 kmem_cache_free(vm_area_cachep, mpnt); 455 return ret; 456 } 457 mm->stack_vm = mm->total_vm = vma_pages(mpnt); 458 } 459 460 for (i = 0 ; i < MAX_ARG_PAGES ; i++) { 461 struct page *page = bprm->page[i]; 462 if (page) { 463 bprm->page[i] = NULL; 464 install_arg_page(mpnt, page, stack_base); 465 } 466 stack_base += PAGE_SIZE; 467 } 468 up_write(&mm->mmap_sem); 469 470 return 0; 471 } 472 473 EXPORT_SYMBOL(setup_arg_pages); 474 475 #define free_arg_pages(bprm) do { } while (0) 476 477 #else 478 479 static inline void free_arg_pages(struct linux_binprm *bprm) 480 { 481 int i; 482 483 for (i = 0; i < MAX_ARG_PAGES; i++) { 484 if (bprm->page[i]) 485 __free_page(bprm->page[i]); 486 bprm->page[i] = NULL; 487 } 488 } 489 490 #endif /* CONFIG_MMU */ 491 492 struct file *open_exec(const char *name) 493 { 494 struct nameidata nd; 495 int err; 496 struct file *file; 497 498 nd.intent.open.flags = FMODE_READ; 499 err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd); 500 file = ERR_PTR(err); 501 502 if (!err) { 503 struct inode *inode = nd.dentry->d_inode; 504 file = ERR_PTR(-EACCES); 505 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) && 506 S_ISREG(inode->i_mode)) { 507 int err = permission(inode, MAY_EXEC, &nd); 508 if (!err && !(inode->i_mode & 0111)) 509 err = -EACCES; 510 file = ERR_PTR(err); 511 if (!err) { 512 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY); 513 if (!IS_ERR(file)) { 514 err = deny_write_access(file); 515 if (err) { 516 fput(file); 517 file = ERR_PTR(err); 518 } 519 } 520 out: 521 return file; 522 } 523 } 524 path_release(&nd); 525 } 526 goto out; 527 } 528 529 EXPORT_SYMBOL(open_exec); 530 531 int kernel_read(struct file *file, unsigned long offset, 532 char *addr, unsigned long count) 533 { 534 mm_segment_t old_fs; 535 loff_t pos = offset; 536 int result; 537 538 old_fs = get_fs(); 539 set_fs(get_ds()); 540 /* The cast to a user pointer is valid due to the set_fs() */ 541 result = vfs_read(file, (void __user *)addr, count, &pos); 542 set_fs(old_fs); 543 return result; 544 } 545 546 EXPORT_SYMBOL(kernel_read); 547 548 static int exec_mmap(struct mm_struct *mm) 549 { 550 struct task_struct *tsk; 551 struct mm_struct * old_mm, *active_mm; 552 553 /* Notify parent that we're no longer interested in the old VM */ 554 tsk = current; 555 old_mm = current->mm; 556 mm_release(tsk, old_mm); 557 558 if (old_mm) { 559 /* 560 * Make sure that if there is a core dump in progress 561 * for the old mm, we get out and die instead of going 562 * through with the exec. We must hold mmap_sem around 563 * checking core_waiters and changing tsk->mm. The 564 * core-inducing thread will increment core_waiters for 565 * each thread whose ->mm == old_mm. 566 */ 567 down_read(&old_mm->mmap_sem); 568 if (unlikely(old_mm->core_waiters)) { 569 up_read(&old_mm->mmap_sem); 570 return -EINTR; 571 } 572 } 573 task_lock(tsk); 574 active_mm = tsk->active_mm; 575 tsk->mm = mm; 576 tsk->active_mm = mm; 577 activate_mm(active_mm, mm); 578 task_unlock(tsk); 579 arch_pick_mmap_layout(mm); 580 if (old_mm) { 581 up_read(&old_mm->mmap_sem); 582 if (active_mm != old_mm) BUG(); 583 mmput(old_mm); 584 return 0; 585 } 586 mmdrop(active_mm); 587 return 0; 588 } 589 590 /* 591 * This function makes sure the current process has its own signal table, 592 * so that flush_signal_handlers can later reset the handlers without 593 * disturbing other processes. (Other processes might share the signal 594 * table via the CLONE_SIGHAND option to clone().) 595 */ 596 static inline int de_thread(struct task_struct *tsk) 597 { 598 struct signal_struct *sig = tsk->signal; 599 struct sighand_struct *newsighand, *oldsighand = tsk->sighand; 600 spinlock_t *lock = &oldsighand->siglock; 601 int count; 602 603 /* 604 * If we don't share sighandlers, then we aren't sharing anything 605 * and we can just re-use it all. 606 */ 607 if (atomic_read(&oldsighand->count) <= 1) { 608 BUG_ON(atomic_read(&sig->count) != 1); 609 exit_itimers(sig); 610 return 0; 611 } 612 613 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 614 if (!newsighand) 615 return -ENOMEM; 616 617 if (thread_group_empty(current)) 618 goto no_thread_group; 619 620 /* 621 * Kill all other threads in the thread group. 622 * We must hold tasklist_lock to call zap_other_threads. 623 */ 624 read_lock(&tasklist_lock); 625 spin_lock_irq(lock); 626 if (sig->flags & SIGNAL_GROUP_EXIT) { 627 /* 628 * Another group action in progress, just 629 * return so that the signal is processed. 630 */ 631 spin_unlock_irq(lock); 632 read_unlock(&tasklist_lock); 633 kmem_cache_free(sighand_cachep, newsighand); 634 return -EAGAIN; 635 } 636 zap_other_threads(current); 637 read_unlock(&tasklist_lock); 638 639 /* 640 * Account for the thread group leader hanging around: 641 */ 642 count = 2; 643 if (thread_group_leader(current)) 644 count = 1; 645 else { 646 /* 647 * The SIGALRM timer survives the exec, but needs to point 648 * at us as the new group leader now. We have a race with 649 * a timer firing now getting the old leader, so we need to 650 * synchronize with any firing (by calling del_timer_sync) 651 * before we can safely let the old group leader die. 652 */ 653 sig->real_timer.data = (unsigned long)current; 654 if (del_timer_sync(&sig->real_timer)) 655 add_timer(&sig->real_timer); 656 } 657 while (atomic_read(&sig->count) > count) { 658 sig->group_exit_task = current; 659 sig->notify_count = count; 660 __set_current_state(TASK_UNINTERRUPTIBLE); 661 spin_unlock_irq(lock); 662 schedule(); 663 spin_lock_irq(lock); 664 } 665 sig->group_exit_task = NULL; 666 sig->notify_count = 0; 667 sig->real_timer.data = (unsigned long)current; 668 spin_unlock_irq(lock); 669 670 /* 671 * At this point all other threads have exited, all we have to 672 * do is to wait for the thread group leader to become inactive, 673 * and to assume its PID: 674 */ 675 if (!thread_group_leader(current)) { 676 struct task_struct *leader = current->group_leader, *parent; 677 struct dentry *proc_dentry1, *proc_dentry2; 678 unsigned long exit_state, ptrace; 679 680 /* 681 * Wait for the thread group leader to be a zombie. 682 * It should already be zombie at this point, most 683 * of the time. 684 */ 685 while (leader->exit_state != EXIT_ZOMBIE) 686 yield(); 687 688 spin_lock(&leader->proc_lock); 689 spin_lock(¤t->proc_lock); 690 proc_dentry1 = proc_pid_unhash(current); 691 proc_dentry2 = proc_pid_unhash(leader); 692 write_lock_irq(&tasklist_lock); 693 694 BUG_ON(leader->tgid != current->tgid); 695 BUG_ON(current->pid == current->tgid); 696 /* 697 * An exec() starts a new thread group with the 698 * TGID of the previous thread group. Rehash the 699 * two threads with a switched PID, and release 700 * the former thread group leader: 701 */ 702 ptrace = leader->ptrace; 703 parent = leader->parent; 704 if (unlikely(ptrace) && unlikely(parent == current)) { 705 /* 706 * Joker was ptracing his own group leader, 707 * and now he wants to be his own parent! 708 * We can't have that. 709 */ 710 ptrace = 0; 711 } 712 713 ptrace_unlink(current); 714 ptrace_unlink(leader); 715 remove_parent(current); 716 remove_parent(leader); 717 718 switch_exec_pids(leader, current); 719 720 current->parent = current->real_parent = leader->real_parent; 721 leader->parent = leader->real_parent = child_reaper; 722 current->group_leader = current; 723 leader->group_leader = leader; 724 725 add_parent(current, current->parent); 726 add_parent(leader, leader->parent); 727 if (ptrace) { 728 current->ptrace = ptrace; 729 __ptrace_link(current, parent); 730 } 731 732 list_del(¤t->tasks); 733 list_add_tail(¤t->tasks, &init_task.tasks); 734 current->exit_signal = SIGCHLD; 735 exit_state = leader->exit_state; 736 737 write_unlock_irq(&tasklist_lock); 738 spin_unlock(&leader->proc_lock); 739 spin_unlock(¤t->proc_lock); 740 proc_pid_flush(proc_dentry1); 741 proc_pid_flush(proc_dentry2); 742 743 BUG_ON(exit_state != EXIT_ZOMBIE); 744 release_task(leader); 745 } 746 747 /* 748 * Now there are really no other threads at all, 749 * so it's safe to stop telling them to kill themselves. 750 */ 751 sig->flags = 0; 752 753 no_thread_group: 754 BUG_ON(atomic_read(&sig->count) != 1); 755 exit_itimers(sig); 756 757 if (atomic_read(&oldsighand->count) == 1) { 758 /* 759 * Now that we nuked the rest of the thread group, 760 * it turns out we are not sharing sighand any more either. 761 * So we can just keep it. 762 */ 763 kmem_cache_free(sighand_cachep, newsighand); 764 } else { 765 /* 766 * Move our state over to newsighand and switch it in. 767 */ 768 spin_lock_init(&newsighand->siglock); 769 atomic_set(&newsighand->count, 1); 770 memcpy(newsighand->action, oldsighand->action, 771 sizeof(newsighand->action)); 772 773 write_lock_irq(&tasklist_lock); 774 spin_lock(&oldsighand->siglock); 775 spin_lock(&newsighand->siglock); 776 777 current->sighand = newsighand; 778 recalc_sigpending(); 779 780 spin_unlock(&newsighand->siglock); 781 spin_unlock(&oldsighand->siglock); 782 write_unlock_irq(&tasklist_lock); 783 784 if (atomic_dec_and_test(&oldsighand->count)) 785 kmem_cache_free(sighand_cachep, oldsighand); 786 } 787 788 BUG_ON(!thread_group_empty(current)); 789 BUG_ON(!thread_group_leader(current)); 790 return 0; 791 } 792 793 /* 794 * These functions flushes out all traces of the currently running executable 795 * so that a new one can be started 796 */ 797 798 static inline void flush_old_files(struct files_struct * files) 799 { 800 long j = -1; 801 802 spin_lock(&files->file_lock); 803 for (;;) { 804 unsigned long set, i; 805 806 j++; 807 i = j * __NFDBITS; 808 if (i >= files->max_fds || i >= files->max_fdset) 809 break; 810 set = files->close_on_exec->fds_bits[j]; 811 if (!set) 812 continue; 813 files->close_on_exec->fds_bits[j] = 0; 814 spin_unlock(&files->file_lock); 815 for ( ; set ; i++,set >>= 1) { 816 if (set & 1) { 817 sys_close(i); 818 } 819 } 820 spin_lock(&files->file_lock); 821 822 } 823 spin_unlock(&files->file_lock); 824 } 825 826 void get_task_comm(char *buf, struct task_struct *tsk) 827 { 828 /* buf must be at least sizeof(tsk->comm) in size */ 829 task_lock(tsk); 830 strncpy(buf, tsk->comm, sizeof(tsk->comm)); 831 task_unlock(tsk); 832 } 833 834 void set_task_comm(struct task_struct *tsk, char *buf) 835 { 836 task_lock(tsk); 837 strlcpy(tsk->comm, buf, sizeof(tsk->comm)); 838 task_unlock(tsk); 839 } 840 841 int flush_old_exec(struct linux_binprm * bprm) 842 { 843 char * name; 844 int i, ch, retval; 845 struct files_struct *files; 846 char tcomm[sizeof(current->comm)]; 847 848 /* 849 * Make sure we have a private signal table and that 850 * we are unassociated from the previous thread group. 851 */ 852 retval = de_thread(current); 853 if (retval) 854 goto out; 855 856 /* 857 * Make sure we have private file handles. Ask the 858 * fork helper to do the work for us and the exit 859 * helper to do the cleanup of the old one. 860 */ 861 files = current->files; /* refcounted so safe to hold */ 862 retval = unshare_files(); 863 if (retval) 864 goto out; 865 /* 866 * Release all of the old mmap stuff 867 */ 868 retval = exec_mmap(bprm->mm); 869 if (retval) 870 goto mmap_failed; 871 872 bprm->mm = NULL; /* We're using it now */ 873 874 /* This is the point of no return */ 875 steal_locks(files); 876 put_files_struct(files); 877 878 current->sas_ss_sp = current->sas_ss_size = 0; 879 880 if (current->euid == current->uid && current->egid == current->gid) 881 current->mm->dumpable = 1; 882 else 883 current->mm->dumpable = suid_dumpable; 884 885 name = bprm->filename; 886 887 /* Copies the binary name from after last slash */ 888 for (i=0; (ch = *(name++)) != '\0';) { 889 if (ch == '/') 890 i = 0; /* overwrite what we wrote */ 891 else 892 if (i < (sizeof(tcomm) - 1)) 893 tcomm[i++] = ch; 894 } 895 tcomm[i] = '\0'; 896 set_task_comm(current, tcomm); 897 898 current->flags &= ~PF_RANDOMIZE; 899 flush_thread(); 900 901 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 902 permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) || 903 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) { 904 suid_keys(current); 905 current->mm->dumpable = suid_dumpable; 906 } 907 908 /* An exec changes our domain. We are no longer part of the thread 909 group */ 910 911 current->self_exec_id++; 912 913 flush_signal_handlers(current, 0); 914 flush_old_files(current->files); 915 916 return 0; 917 918 mmap_failed: 919 put_files_struct(current->files); 920 current->files = files; 921 out: 922 return retval; 923 } 924 925 EXPORT_SYMBOL(flush_old_exec); 926 927 /* 928 * Fill the binprm structure from the inode. 929 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes 930 */ 931 int prepare_binprm(struct linux_binprm *bprm) 932 { 933 int mode; 934 struct inode * inode = bprm->file->f_dentry->d_inode; 935 int retval; 936 937 mode = inode->i_mode; 938 /* 939 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE, 940 * generic_permission lets a non-executable through 941 */ 942 if (!(mode & 0111)) /* with at least _one_ execute bit set */ 943 return -EACCES; 944 if (bprm->file->f_op == NULL) 945 return -EACCES; 946 947 bprm->e_uid = current->euid; 948 bprm->e_gid = current->egid; 949 950 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) { 951 /* Set-uid? */ 952 if (mode & S_ISUID) { 953 current->personality &= ~PER_CLEAR_ON_SETID; 954 bprm->e_uid = inode->i_uid; 955 } 956 957 /* Set-gid? */ 958 /* 959 * If setgid is set but no group execute bit then this 960 * is a candidate for mandatory locking, not a setgid 961 * executable. 962 */ 963 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) { 964 current->personality &= ~PER_CLEAR_ON_SETID; 965 bprm->e_gid = inode->i_gid; 966 } 967 } 968 969 /* fill in binprm security blob */ 970 retval = security_bprm_set(bprm); 971 if (retval) 972 return retval; 973 974 memset(bprm->buf,0,BINPRM_BUF_SIZE); 975 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE); 976 } 977 978 EXPORT_SYMBOL(prepare_binprm); 979 980 static inline int unsafe_exec(struct task_struct *p) 981 { 982 int unsafe = 0; 983 if (p->ptrace & PT_PTRACED) { 984 if (p->ptrace & PT_PTRACE_CAP) 985 unsafe |= LSM_UNSAFE_PTRACE_CAP; 986 else 987 unsafe |= LSM_UNSAFE_PTRACE; 988 } 989 if (atomic_read(&p->fs->count) > 1 || 990 atomic_read(&p->files->count) > 1 || 991 atomic_read(&p->sighand->count) > 1) 992 unsafe |= LSM_UNSAFE_SHARE; 993 994 return unsafe; 995 } 996 997 void compute_creds(struct linux_binprm *bprm) 998 { 999 int unsafe; 1000 1001 if (bprm->e_uid != current->uid) 1002 suid_keys(current); 1003 exec_keys(current); 1004 1005 task_lock(current); 1006 unsafe = unsafe_exec(current); 1007 security_bprm_apply_creds(bprm, unsafe); 1008 task_unlock(current); 1009 security_bprm_post_apply_creds(bprm); 1010 } 1011 1012 EXPORT_SYMBOL(compute_creds); 1013 1014 void remove_arg_zero(struct linux_binprm *bprm) 1015 { 1016 if (bprm->argc) { 1017 unsigned long offset; 1018 char * kaddr; 1019 struct page *page; 1020 1021 offset = bprm->p % PAGE_SIZE; 1022 goto inside; 1023 1024 while (bprm->p++, *(kaddr+offset++)) { 1025 if (offset != PAGE_SIZE) 1026 continue; 1027 offset = 0; 1028 kunmap_atomic(kaddr, KM_USER0); 1029 inside: 1030 page = bprm->page[bprm->p/PAGE_SIZE]; 1031 kaddr = kmap_atomic(page, KM_USER0); 1032 } 1033 kunmap_atomic(kaddr, KM_USER0); 1034 bprm->argc--; 1035 } 1036 } 1037 1038 EXPORT_SYMBOL(remove_arg_zero); 1039 1040 /* 1041 * cycle the list of binary formats handler, until one recognizes the image 1042 */ 1043 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs) 1044 { 1045 int try,retval; 1046 struct linux_binfmt *fmt; 1047 #ifdef __alpha__ 1048 /* handle /sbin/loader.. */ 1049 { 1050 struct exec * eh = (struct exec *) bprm->buf; 1051 1052 if (!bprm->loader && eh->fh.f_magic == 0x183 && 1053 (eh->fh.f_flags & 0x3000) == 0x3000) 1054 { 1055 struct file * file; 1056 unsigned long loader; 1057 1058 allow_write_access(bprm->file); 1059 fput(bprm->file); 1060 bprm->file = NULL; 1061 1062 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *); 1063 1064 file = open_exec("/sbin/loader"); 1065 retval = PTR_ERR(file); 1066 if (IS_ERR(file)) 1067 return retval; 1068 1069 /* Remember if the application is TASO. */ 1070 bprm->sh_bang = eh->ah.entry < 0x100000000UL; 1071 1072 bprm->file = file; 1073 bprm->loader = loader; 1074 retval = prepare_binprm(bprm); 1075 if (retval<0) 1076 return retval; 1077 /* should call search_binary_handler recursively here, 1078 but it does not matter */ 1079 } 1080 } 1081 #endif 1082 retval = security_bprm_check(bprm); 1083 if (retval) 1084 return retval; 1085 1086 /* kernel module loader fixup */ 1087 /* so we don't try to load run modprobe in kernel space. */ 1088 set_fs(USER_DS); 1089 retval = -ENOENT; 1090 for (try=0; try<2; try++) { 1091 read_lock(&binfmt_lock); 1092 for (fmt = formats ; fmt ; fmt = fmt->next) { 1093 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary; 1094 if (!fn) 1095 continue; 1096 if (!try_module_get(fmt->module)) 1097 continue; 1098 read_unlock(&binfmt_lock); 1099 retval = fn(bprm, regs); 1100 if (retval >= 0) { 1101 put_binfmt(fmt); 1102 allow_write_access(bprm->file); 1103 if (bprm->file) 1104 fput(bprm->file); 1105 bprm->file = NULL; 1106 current->did_exec = 1; 1107 return retval; 1108 } 1109 read_lock(&binfmt_lock); 1110 put_binfmt(fmt); 1111 if (retval != -ENOEXEC || bprm->mm == NULL) 1112 break; 1113 if (!bprm->file) { 1114 read_unlock(&binfmt_lock); 1115 return retval; 1116 } 1117 } 1118 read_unlock(&binfmt_lock); 1119 if (retval != -ENOEXEC || bprm->mm == NULL) { 1120 break; 1121 #ifdef CONFIG_KMOD 1122 }else{ 1123 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e)) 1124 if (printable(bprm->buf[0]) && 1125 printable(bprm->buf[1]) && 1126 printable(bprm->buf[2]) && 1127 printable(bprm->buf[3])) 1128 break; /* -ENOEXEC */ 1129 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2])); 1130 #endif 1131 } 1132 } 1133 return retval; 1134 } 1135 1136 EXPORT_SYMBOL(search_binary_handler); 1137 1138 /* 1139 * sys_execve() executes a new program. 1140 */ 1141 int do_execve(char * filename, 1142 char __user *__user *argv, 1143 char __user *__user *envp, 1144 struct pt_regs * regs) 1145 { 1146 struct linux_binprm *bprm; 1147 struct file *file; 1148 int retval; 1149 int i; 1150 1151 retval = -ENOMEM; 1152 bprm = kmalloc(sizeof(*bprm), GFP_KERNEL); 1153 if (!bprm) 1154 goto out_ret; 1155 memset(bprm, 0, sizeof(*bprm)); 1156 1157 file = open_exec(filename); 1158 retval = PTR_ERR(file); 1159 if (IS_ERR(file)) 1160 goto out_kfree; 1161 1162 sched_exec(); 1163 1164 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *); 1165 1166 bprm->file = file; 1167 bprm->filename = filename; 1168 bprm->interp = filename; 1169 bprm->mm = mm_alloc(); 1170 retval = -ENOMEM; 1171 if (!bprm->mm) 1172 goto out_file; 1173 1174 retval = init_new_context(current, bprm->mm); 1175 if (retval < 0) 1176 goto out_mm; 1177 1178 bprm->argc = count(argv, bprm->p / sizeof(void *)); 1179 if ((retval = bprm->argc) < 0) 1180 goto out_mm; 1181 1182 bprm->envc = count(envp, bprm->p / sizeof(void *)); 1183 if ((retval = bprm->envc) < 0) 1184 goto out_mm; 1185 1186 retval = security_bprm_alloc(bprm); 1187 if (retval) 1188 goto out; 1189 1190 retval = prepare_binprm(bprm); 1191 if (retval < 0) 1192 goto out; 1193 1194 retval = copy_strings_kernel(1, &bprm->filename, bprm); 1195 if (retval < 0) 1196 goto out; 1197 1198 bprm->exec = bprm->p; 1199 retval = copy_strings(bprm->envc, envp, bprm); 1200 if (retval < 0) 1201 goto out; 1202 1203 retval = copy_strings(bprm->argc, argv, bprm); 1204 if (retval < 0) 1205 goto out; 1206 1207 retval = search_binary_handler(bprm,regs); 1208 if (retval >= 0) { 1209 free_arg_pages(bprm); 1210 1211 /* execve success */ 1212 security_bprm_free(bprm); 1213 acct_update_integrals(current); 1214 update_mem_hiwater(current); 1215 kfree(bprm); 1216 return retval; 1217 } 1218 1219 out: 1220 /* Something went wrong, return the inode and free the argument pages*/ 1221 for (i = 0 ; i < MAX_ARG_PAGES ; i++) { 1222 struct page * page = bprm->page[i]; 1223 if (page) 1224 __free_page(page); 1225 } 1226 1227 if (bprm->security) 1228 security_bprm_free(bprm); 1229 1230 out_mm: 1231 if (bprm->mm) 1232 mmdrop(bprm->mm); 1233 1234 out_file: 1235 if (bprm->file) { 1236 allow_write_access(bprm->file); 1237 fput(bprm->file); 1238 } 1239 1240 out_kfree: 1241 kfree(bprm); 1242 1243 out_ret: 1244 return retval; 1245 } 1246 1247 int set_binfmt(struct linux_binfmt *new) 1248 { 1249 struct linux_binfmt *old = current->binfmt; 1250 1251 if (new) { 1252 if (!try_module_get(new->module)) 1253 return -1; 1254 } 1255 current->binfmt = new; 1256 if (old) 1257 module_put(old->module); 1258 return 0; 1259 } 1260 1261 EXPORT_SYMBOL(set_binfmt); 1262 1263 #define CORENAME_MAX_SIZE 64 1264 1265 /* format_corename will inspect the pattern parameter, and output a 1266 * name into corename, which must have space for at least 1267 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator. 1268 */ 1269 static void format_corename(char *corename, const char *pattern, long signr) 1270 { 1271 const char *pat_ptr = pattern; 1272 char *out_ptr = corename; 1273 char *const out_end = corename + CORENAME_MAX_SIZE; 1274 int rc; 1275 int pid_in_pattern = 0; 1276 1277 /* Repeat as long as we have more pattern to process and more output 1278 space */ 1279 while (*pat_ptr) { 1280 if (*pat_ptr != '%') { 1281 if (out_ptr == out_end) 1282 goto out; 1283 *out_ptr++ = *pat_ptr++; 1284 } else { 1285 switch (*++pat_ptr) { 1286 case 0: 1287 goto out; 1288 /* Double percent, output one percent */ 1289 case '%': 1290 if (out_ptr == out_end) 1291 goto out; 1292 *out_ptr++ = '%'; 1293 break; 1294 /* pid */ 1295 case 'p': 1296 pid_in_pattern = 1; 1297 rc = snprintf(out_ptr, out_end - out_ptr, 1298 "%d", current->tgid); 1299 if (rc > out_end - out_ptr) 1300 goto out; 1301 out_ptr += rc; 1302 break; 1303 /* uid */ 1304 case 'u': 1305 rc = snprintf(out_ptr, out_end - out_ptr, 1306 "%d", current->uid); 1307 if (rc > out_end - out_ptr) 1308 goto out; 1309 out_ptr += rc; 1310 break; 1311 /* gid */ 1312 case 'g': 1313 rc = snprintf(out_ptr, out_end - out_ptr, 1314 "%d", current->gid); 1315 if (rc > out_end - out_ptr) 1316 goto out; 1317 out_ptr += rc; 1318 break; 1319 /* signal that caused the coredump */ 1320 case 's': 1321 rc = snprintf(out_ptr, out_end - out_ptr, 1322 "%ld", signr); 1323 if (rc > out_end - out_ptr) 1324 goto out; 1325 out_ptr += rc; 1326 break; 1327 /* UNIX time of coredump */ 1328 case 't': { 1329 struct timeval tv; 1330 do_gettimeofday(&tv); 1331 rc = snprintf(out_ptr, out_end - out_ptr, 1332 "%lu", tv.tv_sec); 1333 if (rc > out_end - out_ptr) 1334 goto out; 1335 out_ptr += rc; 1336 break; 1337 } 1338 /* hostname */ 1339 case 'h': 1340 down_read(&uts_sem); 1341 rc = snprintf(out_ptr, out_end - out_ptr, 1342 "%s", system_utsname.nodename); 1343 up_read(&uts_sem); 1344 if (rc > out_end - out_ptr) 1345 goto out; 1346 out_ptr += rc; 1347 break; 1348 /* executable */ 1349 case 'e': 1350 rc = snprintf(out_ptr, out_end - out_ptr, 1351 "%s", current->comm); 1352 if (rc > out_end - out_ptr) 1353 goto out; 1354 out_ptr += rc; 1355 break; 1356 default: 1357 break; 1358 } 1359 ++pat_ptr; 1360 } 1361 } 1362 /* Backward compatibility with core_uses_pid: 1363 * 1364 * If core_pattern does not include a %p (as is the default) 1365 * and core_uses_pid is set, then .%pid will be appended to 1366 * the filename */ 1367 if (!pid_in_pattern 1368 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) { 1369 rc = snprintf(out_ptr, out_end - out_ptr, 1370 ".%d", current->tgid); 1371 if (rc > out_end - out_ptr) 1372 goto out; 1373 out_ptr += rc; 1374 } 1375 out: 1376 *out_ptr = 0; 1377 } 1378 1379 static void zap_threads (struct mm_struct *mm) 1380 { 1381 struct task_struct *g, *p; 1382 struct task_struct *tsk = current; 1383 struct completion *vfork_done = tsk->vfork_done; 1384 int traced = 0; 1385 1386 /* 1387 * Make sure nobody is waiting for us to release the VM, 1388 * otherwise we can deadlock when we wait on each other 1389 */ 1390 if (vfork_done) { 1391 tsk->vfork_done = NULL; 1392 complete(vfork_done); 1393 } 1394 1395 read_lock(&tasklist_lock); 1396 do_each_thread(g,p) 1397 if (mm == p->mm && p != tsk) { 1398 force_sig_specific(SIGKILL, p); 1399 mm->core_waiters++; 1400 if (unlikely(p->ptrace) && 1401 unlikely(p->parent->mm == mm)) 1402 traced = 1; 1403 } 1404 while_each_thread(g,p); 1405 1406 read_unlock(&tasklist_lock); 1407 1408 if (unlikely(traced)) { 1409 /* 1410 * We are zapping a thread and the thread it ptraces. 1411 * If the tracee went into a ptrace stop for exit tracing, 1412 * we could deadlock since the tracer is waiting for this 1413 * coredump to finish. Detach them so they can both die. 1414 */ 1415 write_lock_irq(&tasklist_lock); 1416 do_each_thread(g,p) { 1417 if (mm == p->mm && p != tsk && 1418 p->ptrace && p->parent->mm == mm) { 1419 __ptrace_unlink(p); 1420 } 1421 } while_each_thread(g,p); 1422 write_unlock_irq(&tasklist_lock); 1423 } 1424 } 1425 1426 static void coredump_wait(struct mm_struct *mm) 1427 { 1428 DECLARE_COMPLETION(startup_done); 1429 1430 mm->core_waiters++; /* let other threads block */ 1431 mm->core_startup_done = &startup_done; 1432 1433 /* give other threads a chance to run: */ 1434 yield(); 1435 1436 zap_threads(mm); 1437 if (--mm->core_waiters) { 1438 up_write(&mm->mmap_sem); 1439 wait_for_completion(&startup_done); 1440 } else 1441 up_write(&mm->mmap_sem); 1442 BUG_ON(mm->core_waiters); 1443 } 1444 1445 int do_coredump(long signr, int exit_code, struct pt_regs * regs) 1446 { 1447 char corename[CORENAME_MAX_SIZE + 1]; 1448 struct mm_struct *mm = current->mm; 1449 struct linux_binfmt * binfmt; 1450 struct inode * inode; 1451 struct file * file; 1452 int retval = 0; 1453 int fsuid = current->fsuid; 1454 int flag = 0; 1455 1456 binfmt = current->binfmt; 1457 if (!binfmt || !binfmt->core_dump) 1458 goto fail; 1459 down_write(&mm->mmap_sem); 1460 if (!mm->dumpable) { 1461 up_write(&mm->mmap_sem); 1462 goto fail; 1463 } 1464 1465 /* 1466 * We cannot trust fsuid as being the "true" uid of the 1467 * process nor do we know its entire history. We only know it 1468 * was tainted so we dump it as root in mode 2. 1469 */ 1470 if (mm->dumpable == 2) { /* Setuid core dump mode */ 1471 flag = O_EXCL; /* Stop rewrite attacks */ 1472 current->fsuid = 0; /* Dump root private */ 1473 } 1474 mm->dumpable = 0; 1475 init_completion(&mm->core_done); 1476 spin_lock_irq(¤t->sighand->siglock); 1477 current->signal->flags = SIGNAL_GROUP_EXIT; 1478 current->signal->group_exit_code = exit_code; 1479 spin_unlock_irq(¤t->sighand->siglock); 1480 coredump_wait(mm); 1481 1482 /* 1483 * Clear any false indication of pending signals that might 1484 * be seen by the filesystem code called to write the core file. 1485 */ 1486 current->signal->group_stop_count = 0; 1487 clear_thread_flag(TIF_SIGPENDING); 1488 1489 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump) 1490 goto fail_unlock; 1491 1492 /* 1493 * lock_kernel() because format_corename() is controlled by sysctl, which 1494 * uses lock_kernel() 1495 */ 1496 lock_kernel(); 1497 format_corename(corename, core_pattern, signr); 1498 unlock_kernel(); 1499 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600); 1500 if (IS_ERR(file)) 1501 goto fail_unlock; 1502 inode = file->f_dentry->d_inode; 1503 if (inode->i_nlink > 1) 1504 goto close_fail; /* multiple links - don't dump */ 1505 if (d_unhashed(file->f_dentry)) 1506 goto close_fail; 1507 1508 if (!S_ISREG(inode->i_mode)) 1509 goto close_fail; 1510 if (!file->f_op) 1511 goto close_fail; 1512 if (!file->f_op->write) 1513 goto close_fail; 1514 if (do_truncate(file->f_dentry, 0) != 0) 1515 goto close_fail; 1516 1517 retval = binfmt->core_dump(signr, regs, file); 1518 1519 if (retval) 1520 current->signal->group_exit_code |= 0x80; 1521 close_fail: 1522 filp_close(file, NULL); 1523 fail_unlock: 1524 current->fsuid = fsuid; 1525 complete_all(&mm->core_done); 1526 fail: 1527 return retval; 1528 } 1529