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