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 while (atomic_read(&sig->count) > count) { 646 sig->group_exit_task = current; 647 sig->notify_count = count; 648 __set_current_state(TASK_UNINTERRUPTIBLE); 649 spin_unlock_irq(lock); 650 schedule(); 651 spin_lock_irq(lock); 652 } 653 sig->group_exit_task = NULL; 654 sig->notify_count = 0; 655 sig->real_timer.data = (unsigned long)current; 656 spin_unlock_irq(lock); 657 658 /* 659 * At this point all other threads have exited, all we have to 660 * do is to wait for the thread group leader to become inactive, 661 * and to assume its PID: 662 */ 663 if (!thread_group_leader(current)) { 664 struct task_struct *leader = current->group_leader, *parent; 665 struct dentry *proc_dentry1, *proc_dentry2; 666 unsigned long exit_state, ptrace; 667 668 /* 669 * Wait for the thread group leader to be a zombie. 670 * It should already be zombie at this point, most 671 * of the time. 672 */ 673 while (leader->exit_state != EXIT_ZOMBIE) 674 yield(); 675 676 spin_lock(&leader->proc_lock); 677 spin_lock(¤t->proc_lock); 678 proc_dentry1 = proc_pid_unhash(current); 679 proc_dentry2 = proc_pid_unhash(leader); 680 write_lock_irq(&tasklist_lock); 681 682 BUG_ON(leader->tgid != current->tgid); 683 BUG_ON(current->pid == current->tgid); 684 /* 685 * An exec() starts a new thread group with the 686 * TGID of the previous thread group. Rehash the 687 * two threads with a switched PID, and release 688 * the former thread group leader: 689 */ 690 ptrace = leader->ptrace; 691 parent = leader->parent; 692 if (unlikely(ptrace) && unlikely(parent == current)) { 693 /* 694 * Joker was ptracing his own group leader, 695 * and now he wants to be his own parent! 696 * We can't have that. 697 */ 698 ptrace = 0; 699 } 700 701 ptrace_unlink(current); 702 ptrace_unlink(leader); 703 remove_parent(current); 704 remove_parent(leader); 705 706 switch_exec_pids(leader, current); 707 708 current->parent = current->real_parent = leader->real_parent; 709 leader->parent = leader->real_parent = child_reaper; 710 current->group_leader = current; 711 leader->group_leader = leader; 712 713 add_parent(current, current->parent); 714 add_parent(leader, leader->parent); 715 if (ptrace) { 716 current->ptrace = ptrace; 717 __ptrace_link(current, parent); 718 } 719 720 list_del(¤t->tasks); 721 list_add_tail(¤t->tasks, &init_task.tasks); 722 current->exit_signal = SIGCHLD; 723 exit_state = leader->exit_state; 724 725 write_unlock_irq(&tasklist_lock); 726 spin_unlock(&leader->proc_lock); 727 spin_unlock(¤t->proc_lock); 728 proc_pid_flush(proc_dentry1); 729 proc_pid_flush(proc_dentry2); 730 731 BUG_ON(exit_state != EXIT_ZOMBIE); 732 release_task(leader); 733 } 734 735 /* 736 * Now there are really no other threads at all, 737 * so it's safe to stop telling them to kill themselves. 738 */ 739 sig->flags = 0; 740 741 no_thread_group: 742 BUG_ON(atomic_read(&sig->count) != 1); 743 exit_itimers(sig); 744 745 if (atomic_read(&oldsighand->count) == 1) { 746 /* 747 * Now that we nuked the rest of the thread group, 748 * it turns out we are not sharing sighand any more either. 749 * So we can just keep it. 750 */ 751 kmem_cache_free(sighand_cachep, newsighand); 752 } else { 753 /* 754 * Move our state over to newsighand and switch it in. 755 */ 756 spin_lock_init(&newsighand->siglock); 757 atomic_set(&newsighand->count, 1); 758 memcpy(newsighand->action, oldsighand->action, 759 sizeof(newsighand->action)); 760 761 write_lock_irq(&tasklist_lock); 762 spin_lock(&oldsighand->siglock); 763 spin_lock(&newsighand->siglock); 764 765 current->sighand = newsighand; 766 recalc_sigpending(); 767 768 spin_unlock(&newsighand->siglock); 769 spin_unlock(&oldsighand->siglock); 770 write_unlock_irq(&tasklist_lock); 771 772 if (atomic_dec_and_test(&oldsighand->count)) 773 kmem_cache_free(sighand_cachep, oldsighand); 774 } 775 776 BUG_ON(!thread_group_empty(current)); 777 BUG_ON(!thread_group_leader(current)); 778 return 0; 779 } 780 781 /* 782 * These functions flushes out all traces of the currently running executable 783 * so that a new one can be started 784 */ 785 786 static inline void flush_old_files(struct files_struct * files) 787 { 788 long j = -1; 789 790 spin_lock(&files->file_lock); 791 for (;;) { 792 unsigned long set, i; 793 794 j++; 795 i = j * __NFDBITS; 796 if (i >= files->max_fds || i >= files->max_fdset) 797 break; 798 set = files->close_on_exec->fds_bits[j]; 799 if (!set) 800 continue; 801 files->close_on_exec->fds_bits[j] = 0; 802 spin_unlock(&files->file_lock); 803 for ( ; set ; i++,set >>= 1) { 804 if (set & 1) { 805 sys_close(i); 806 } 807 } 808 spin_lock(&files->file_lock); 809 810 } 811 spin_unlock(&files->file_lock); 812 } 813 814 void get_task_comm(char *buf, struct task_struct *tsk) 815 { 816 /* buf must be at least sizeof(tsk->comm) in size */ 817 task_lock(tsk); 818 strncpy(buf, tsk->comm, sizeof(tsk->comm)); 819 task_unlock(tsk); 820 } 821 822 void set_task_comm(struct task_struct *tsk, char *buf) 823 { 824 task_lock(tsk); 825 strlcpy(tsk->comm, buf, sizeof(tsk->comm)); 826 task_unlock(tsk); 827 } 828 829 int flush_old_exec(struct linux_binprm * bprm) 830 { 831 char * name; 832 int i, ch, retval; 833 struct files_struct *files; 834 char tcomm[sizeof(current->comm)]; 835 836 /* 837 * Make sure we have a private signal table and that 838 * we are unassociated from the previous thread group. 839 */ 840 retval = de_thread(current); 841 if (retval) 842 goto out; 843 844 /* 845 * Make sure we have private file handles. Ask the 846 * fork helper to do the work for us and the exit 847 * helper to do the cleanup of the old one. 848 */ 849 files = current->files; /* refcounted so safe to hold */ 850 retval = unshare_files(); 851 if (retval) 852 goto out; 853 /* 854 * Release all of the old mmap stuff 855 */ 856 retval = exec_mmap(bprm->mm); 857 if (retval) 858 goto mmap_failed; 859 860 bprm->mm = NULL; /* We're using it now */ 861 862 /* This is the point of no return */ 863 steal_locks(files); 864 put_files_struct(files); 865 866 current->sas_ss_sp = current->sas_ss_size = 0; 867 868 if (current->euid == current->uid && current->egid == current->gid) 869 current->mm->dumpable = 1; 870 else 871 current->mm->dumpable = suid_dumpable; 872 873 name = bprm->filename; 874 875 /* Copies the binary name from after last slash */ 876 for (i=0; (ch = *(name++)) != '\0';) { 877 if (ch == '/') 878 i = 0; /* overwrite what we wrote */ 879 else 880 if (i < (sizeof(tcomm) - 1)) 881 tcomm[i++] = ch; 882 } 883 tcomm[i] = '\0'; 884 set_task_comm(current, tcomm); 885 886 current->flags &= ~PF_RANDOMIZE; 887 flush_thread(); 888 889 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 890 permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) || 891 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) { 892 suid_keys(current); 893 current->mm->dumpable = suid_dumpable; 894 } 895 896 /* An exec changes our domain. We are no longer part of the thread 897 group */ 898 899 current->self_exec_id++; 900 901 flush_signal_handlers(current, 0); 902 flush_old_files(current->files); 903 904 return 0; 905 906 mmap_failed: 907 put_files_struct(current->files); 908 current->files = files; 909 out: 910 return retval; 911 } 912 913 EXPORT_SYMBOL(flush_old_exec); 914 915 /* 916 * Fill the binprm structure from the inode. 917 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes 918 */ 919 int prepare_binprm(struct linux_binprm *bprm) 920 { 921 int mode; 922 struct inode * inode = bprm->file->f_dentry->d_inode; 923 int retval; 924 925 mode = inode->i_mode; 926 /* 927 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE, 928 * generic_permission lets a non-executable through 929 */ 930 if (!(mode & 0111)) /* with at least _one_ execute bit set */ 931 return -EACCES; 932 if (bprm->file->f_op == NULL) 933 return -EACCES; 934 935 bprm->e_uid = current->euid; 936 bprm->e_gid = current->egid; 937 938 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) { 939 /* Set-uid? */ 940 if (mode & S_ISUID) { 941 current->personality &= ~PER_CLEAR_ON_SETID; 942 bprm->e_uid = inode->i_uid; 943 } 944 945 /* Set-gid? */ 946 /* 947 * If setgid is set but no group execute bit then this 948 * is a candidate for mandatory locking, not a setgid 949 * executable. 950 */ 951 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) { 952 current->personality &= ~PER_CLEAR_ON_SETID; 953 bprm->e_gid = inode->i_gid; 954 } 955 } 956 957 /* fill in binprm security blob */ 958 retval = security_bprm_set(bprm); 959 if (retval) 960 return retval; 961 962 memset(bprm->buf,0,BINPRM_BUF_SIZE); 963 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE); 964 } 965 966 EXPORT_SYMBOL(prepare_binprm); 967 968 static inline int unsafe_exec(struct task_struct *p) 969 { 970 int unsafe = 0; 971 if (p->ptrace & PT_PTRACED) { 972 if (p->ptrace & PT_PTRACE_CAP) 973 unsafe |= LSM_UNSAFE_PTRACE_CAP; 974 else 975 unsafe |= LSM_UNSAFE_PTRACE; 976 } 977 if (atomic_read(&p->fs->count) > 1 || 978 atomic_read(&p->files->count) > 1 || 979 atomic_read(&p->sighand->count) > 1) 980 unsafe |= LSM_UNSAFE_SHARE; 981 982 return unsafe; 983 } 984 985 void compute_creds(struct linux_binprm *bprm) 986 { 987 int unsafe; 988 989 if (bprm->e_uid != current->uid) 990 suid_keys(current); 991 exec_keys(current); 992 993 task_lock(current); 994 unsafe = unsafe_exec(current); 995 security_bprm_apply_creds(bprm, unsafe); 996 task_unlock(current); 997 security_bprm_post_apply_creds(bprm); 998 } 999 1000 EXPORT_SYMBOL(compute_creds); 1001 1002 void remove_arg_zero(struct linux_binprm *bprm) 1003 { 1004 if (bprm->argc) { 1005 unsigned long offset; 1006 char * kaddr; 1007 struct page *page; 1008 1009 offset = bprm->p % PAGE_SIZE; 1010 goto inside; 1011 1012 while (bprm->p++, *(kaddr+offset++)) { 1013 if (offset != PAGE_SIZE) 1014 continue; 1015 offset = 0; 1016 kunmap_atomic(kaddr, KM_USER0); 1017 inside: 1018 page = bprm->page[bprm->p/PAGE_SIZE]; 1019 kaddr = kmap_atomic(page, KM_USER0); 1020 } 1021 kunmap_atomic(kaddr, KM_USER0); 1022 bprm->argc--; 1023 } 1024 } 1025 1026 EXPORT_SYMBOL(remove_arg_zero); 1027 1028 /* 1029 * cycle the list of binary formats handler, until one recognizes the image 1030 */ 1031 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs) 1032 { 1033 int try,retval; 1034 struct linux_binfmt *fmt; 1035 #ifdef __alpha__ 1036 /* handle /sbin/loader.. */ 1037 { 1038 struct exec * eh = (struct exec *) bprm->buf; 1039 1040 if (!bprm->loader && eh->fh.f_magic == 0x183 && 1041 (eh->fh.f_flags & 0x3000) == 0x3000) 1042 { 1043 struct file * file; 1044 unsigned long loader; 1045 1046 allow_write_access(bprm->file); 1047 fput(bprm->file); 1048 bprm->file = NULL; 1049 1050 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *); 1051 1052 file = open_exec("/sbin/loader"); 1053 retval = PTR_ERR(file); 1054 if (IS_ERR(file)) 1055 return retval; 1056 1057 /* Remember if the application is TASO. */ 1058 bprm->sh_bang = eh->ah.entry < 0x100000000UL; 1059 1060 bprm->file = file; 1061 bprm->loader = loader; 1062 retval = prepare_binprm(bprm); 1063 if (retval<0) 1064 return retval; 1065 /* should call search_binary_handler recursively here, 1066 but it does not matter */ 1067 } 1068 } 1069 #endif 1070 retval = security_bprm_check(bprm); 1071 if (retval) 1072 return retval; 1073 1074 /* kernel module loader fixup */ 1075 /* so we don't try to load run modprobe in kernel space. */ 1076 set_fs(USER_DS); 1077 retval = -ENOENT; 1078 for (try=0; try<2; try++) { 1079 read_lock(&binfmt_lock); 1080 for (fmt = formats ; fmt ; fmt = fmt->next) { 1081 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary; 1082 if (!fn) 1083 continue; 1084 if (!try_module_get(fmt->module)) 1085 continue; 1086 read_unlock(&binfmt_lock); 1087 retval = fn(bprm, regs); 1088 if (retval >= 0) { 1089 put_binfmt(fmt); 1090 allow_write_access(bprm->file); 1091 if (bprm->file) 1092 fput(bprm->file); 1093 bprm->file = NULL; 1094 current->did_exec = 1; 1095 return retval; 1096 } 1097 read_lock(&binfmt_lock); 1098 put_binfmt(fmt); 1099 if (retval != -ENOEXEC || bprm->mm == NULL) 1100 break; 1101 if (!bprm->file) { 1102 read_unlock(&binfmt_lock); 1103 return retval; 1104 } 1105 } 1106 read_unlock(&binfmt_lock); 1107 if (retval != -ENOEXEC || bprm->mm == NULL) { 1108 break; 1109 #ifdef CONFIG_KMOD 1110 }else{ 1111 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e)) 1112 if (printable(bprm->buf[0]) && 1113 printable(bprm->buf[1]) && 1114 printable(bprm->buf[2]) && 1115 printable(bprm->buf[3])) 1116 break; /* -ENOEXEC */ 1117 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2])); 1118 #endif 1119 } 1120 } 1121 return retval; 1122 } 1123 1124 EXPORT_SYMBOL(search_binary_handler); 1125 1126 /* 1127 * sys_execve() executes a new program. 1128 */ 1129 int do_execve(char * filename, 1130 char __user *__user *argv, 1131 char __user *__user *envp, 1132 struct pt_regs * regs) 1133 { 1134 struct linux_binprm *bprm; 1135 struct file *file; 1136 int retval; 1137 int i; 1138 1139 retval = -ENOMEM; 1140 bprm = kmalloc(sizeof(*bprm), GFP_KERNEL); 1141 if (!bprm) 1142 goto out_ret; 1143 memset(bprm, 0, sizeof(*bprm)); 1144 1145 file = open_exec(filename); 1146 retval = PTR_ERR(file); 1147 if (IS_ERR(file)) 1148 goto out_kfree; 1149 1150 sched_exec(); 1151 1152 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *); 1153 1154 bprm->file = file; 1155 bprm->filename = filename; 1156 bprm->interp = filename; 1157 bprm->mm = mm_alloc(); 1158 retval = -ENOMEM; 1159 if (!bprm->mm) 1160 goto out_file; 1161 1162 retval = init_new_context(current, bprm->mm); 1163 if (retval < 0) 1164 goto out_mm; 1165 1166 bprm->argc = count(argv, bprm->p / sizeof(void *)); 1167 if ((retval = bprm->argc) < 0) 1168 goto out_mm; 1169 1170 bprm->envc = count(envp, bprm->p / sizeof(void *)); 1171 if ((retval = bprm->envc) < 0) 1172 goto out_mm; 1173 1174 retval = security_bprm_alloc(bprm); 1175 if (retval) 1176 goto out; 1177 1178 retval = prepare_binprm(bprm); 1179 if (retval < 0) 1180 goto out; 1181 1182 retval = copy_strings_kernel(1, &bprm->filename, bprm); 1183 if (retval < 0) 1184 goto out; 1185 1186 bprm->exec = bprm->p; 1187 retval = copy_strings(bprm->envc, envp, bprm); 1188 if (retval < 0) 1189 goto out; 1190 1191 retval = copy_strings(bprm->argc, argv, bprm); 1192 if (retval < 0) 1193 goto out; 1194 1195 retval = search_binary_handler(bprm,regs); 1196 if (retval >= 0) { 1197 free_arg_pages(bprm); 1198 1199 /* execve success */ 1200 security_bprm_free(bprm); 1201 acct_update_integrals(current); 1202 update_mem_hiwater(current); 1203 kfree(bprm); 1204 return retval; 1205 } 1206 1207 out: 1208 /* Something went wrong, return the inode and free the argument pages*/ 1209 for (i = 0 ; i < MAX_ARG_PAGES ; i++) { 1210 struct page * page = bprm->page[i]; 1211 if (page) 1212 __free_page(page); 1213 } 1214 1215 if (bprm->security) 1216 security_bprm_free(bprm); 1217 1218 out_mm: 1219 if (bprm->mm) 1220 mmdrop(bprm->mm); 1221 1222 out_file: 1223 if (bprm->file) { 1224 allow_write_access(bprm->file); 1225 fput(bprm->file); 1226 } 1227 1228 out_kfree: 1229 kfree(bprm); 1230 1231 out_ret: 1232 return retval; 1233 } 1234 1235 int set_binfmt(struct linux_binfmt *new) 1236 { 1237 struct linux_binfmt *old = current->binfmt; 1238 1239 if (new) { 1240 if (!try_module_get(new->module)) 1241 return -1; 1242 } 1243 current->binfmt = new; 1244 if (old) 1245 module_put(old->module); 1246 return 0; 1247 } 1248 1249 EXPORT_SYMBOL(set_binfmt); 1250 1251 #define CORENAME_MAX_SIZE 64 1252 1253 /* format_corename will inspect the pattern parameter, and output a 1254 * name into corename, which must have space for at least 1255 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator. 1256 */ 1257 static void format_corename(char *corename, const char *pattern, long signr) 1258 { 1259 const char *pat_ptr = pattern; 1260 char *out_ptr = corename; 1261 char *const out_end = corename + CORENAME_MAX_SIZE; 1262 int rc; 1263 int pid_in_pattern = 0; 1264 1265 /* Repeat as long as we have more pattern to process and more output 1266 space */ 1267 while (*pat_ptr) { 1268 if (*pat_ptr != '%') { 1269 if (out_ptr == out_end) 1270 goto out; 1271 *out_ptr++ = *pat_ptr++; 1272 } else { 1273 switch (*++pat_ptr) { 1274 case 0: 1275 goto out; 1276 /* Double percent, output one percent */ 1277 case '%': 1278 if (out_ptr == out_end) 1279 goto out; 1280 *out_ptr++ = '%'; 1281 break; 1282 /* pid */ 1283 case 'p': 1284 pid_in_pattern = 1; 1285 rc = snprintf(out_ptr, out_end - out_ptr, 1286 "%d", current->tgid); 1287 if (rc > out_end - out_ptr) 1288 goto out; 1289 out_ptr += rc; 1290 break; 1291 /* uid */ 1292 case 'u': 1293 rc = snprintf(out_ptr, out_end - out_ptr, 1294 "%d", current->uid); 1295 if (rc > out_end - out_ptr) 1296 goto out; 1297 out_ptr += rc; 1298 break; 1299 /* gid */ 1300 case 'g': 1301 rc = snprintf(out_ptr, out_end - out_ptr, 1302 "%d", current->gid); 1303 if (rc > out_end - out_ptr) 1304 goto out; 1305 out_ptr += rc; 1306 break; 1307 /* signal that caused the coredump */ 1308 case 's': 1309 rc = snprintf(out_ptr, out_end - out_ptr, 1310 "%ld", signr); 1311 if (rc > out_end - out_ptr) 1312 goto out; 1313 out_ptr += rc; 1314 break; 1315 /* UNIX time of coredump */ 1316 case 't': { 1317 struct timeval tv; 1318 do_gettimeofday(&tv); 1319 rc = snprintf(out_ptr, out_end - out_ptr, 1320 "%lu", tv.tv_sec); 1321 if (rc > out_end - out_ptr) 1322 goto out; 1323 out_ptr += rc; 1324 break; 1325 } 1326 /* hostname */ 1327 case 'h': 1328 down_read(&uts_sem); 1329 rc = snprintf(out_ptr, out_end - out_ptr, 1330 "%s", system_utsname.nodename); 1331 up_read(&uts_sem); 1332 if (rc > out_end - out_ptr) 1333 goto out; 1334 out_ptr += rc; 1335 break; 1336 /* executable */ 1337 case 'e': 1338 rc = snprintf(out_ptr, out_end - out_ptr, 1339 "%s", current->comm); 1340 if (rc > out_end - out_ptr) 1341 goto out; 1342 out_ptr += rc; 1343 break; 1344 default: 1345 break; 1346 } 1347 ++pat_ptr; 1348 } 1349 } 1350 /* Backward compatibility with core_uses_pid: 1351 * 1352 * If core_pattern does not include a %p (as is the default) 1353 * and core_uses_pid is set, then .%pid will be appended to 1354 * the filename */ 1355 if (!pid_in_pattern 1356 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) { 1357 rc = snprintf(out_ptr, out_end - out_ptr, 1358 ".%d", current->tgid); 1359 if (rc > out_end - out_ptr) 1360 goto out; 1361 out_ptr += rc; 1362 } 1363 out: 1364 *out_ptr = 0; 1365 } 1366 1367 static void zap_threads (struct mm_struct *mm) 1368 { 1369 struct task_struct *g, *p; 1370 struct task_struct *tsk = current; 1371 struct completion *vfork_done = tsk->vfork_done; 1372 int traced = 0; 1373 1374 /* 1375 * Make sure nobody is waiting for us to release the VM, 1376 * otherwise we can deadlock when we wait on each other 1377 */ 1378 if (vfork_done) { 1379 tsk->vfork_done = NULL; 1380 complete(vfork_done); 1381 } 1382 1383 read_lock(&tasklist_lock); 1384 do_each_thread(g,p) 1385 if (mm == p->mm && p != tsk) { 1386 force_sig_specific(SIGKILL, p); 1387 mm->core_waiters++; 1388 if (unlikely(p->ptrace) && 1389 unlikely(p->parent->mm == mm)) 1390 traced = 1; 1391 } 1392 while_each_thread(g,p); 1393 1394 read_unlock(&tasklist_lock); 1395 1396 if (unlikely(traced)) { 1397 /* 1398 * We are zapping a thread and the thread it ptraces. 1399 * If the tracee went into a ptrace stop for exit tracing, 1400 * we could deadlock since the tracer is waiting for this 1401 * coredump to finish. Detach them so they can both die. 1402 */ 1403 write_lock_irq(&tasklist_lock); 1404 do_each_thread(g,p) { 1405 if (mm == p->mm && p != tsk && 1406 p->ptrace && p->parent->mm == mm) { 1407 __ptrace_unlink(p); 1408 } 1409 } while_each_thread(g,p); 1410 write_unlock_irq(&tasklist_lock); 1411 } 1412 } 1413 1414 static void coredump_wait(struct mm_struct *mm) 1415 { 1416 DECLARE_COMPLETION(startup_done); 1417 1418 mm->core_waiters++; /* let other threads block */ 1419 mm->core_startup_done = &startup_done; 1420 1421 /* give other threads a chance to run: */ 1422 yield(); 1423 1424 zap_threads(mm); 1425 if (--mm->core_waiters) { 1426 up_write(&mm->mmap_sem); 1427 wait_for_completion(&startup_done); 1428 } else 1429 up_write(&mm->mmap_sem); 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 init_completion(&mm->core_done); 1464 spin_lock_irq(¤t->sighand->siglock); 1465 current->signal->flags = SIGNAL_GROUP_EXIT; 1466 current->signal->group_exit_code = exit_code; 1467 spin_unlock_irq(¤t->sighand->siglock); 1468 coredump_wait(mm); 1469 1470 /* 1471 * Clear any false indication of pending signals that might 1472 * be seen by the filesystem code called to write the core file. 1473 */ 1474 current->signal->group_stop_count = 0; 1475 clear_thread_flag(TIF_SIGPENDING); 1476 1477 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump) 1478 goto fail_unlock; 1479 1480 /* 1481 * lock_kernel() because format_corename() is controlled by sysctl, which 1482 * uses lock_kernel() 1483 */ 1484 lock_kernel(); 1485 format_corename(corename, core_pattern, signr); 1486 unlock_kernel(); 1487 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600); 1488 if (IS_ERR(file)) 1489 goto fail_unlock; 1490 inode = file->f_dentry->d_inode; 1491 if (inode->i_nlink > 1) 1492 goto close_fail; /* multiple links - don't dump */ 1493 if (d_unhashed(file->f_dentry)) 1494 goto close_fail; 1495 1496 if (!S_ISREG(inode->i_mode)) 1497 goto close_fail; 1498 if (!file->f_op) 1499 goto close_fail; 1500 if (!file->f_op->write) 1501 goto close_fail; 1502 if (do_truncate(file->f_dentry, 0) != 0) 1503 goto close_fail; 1504 1505 retval = binfmt->core_dump(signr, regs, file); 1506 1507 if (retval) 1508 current->signal->group_exit_code |= 0x80; 1509 close_fail: 1510 filp_close(file, NULL); 1511 fail_unlock: 1512 current->fsuid = fsuid; 1513 complete_all(&mm->core_done); 1514 fail: 1515 return retval; 1516 } 1517