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/slab.h> 26 #include <linux/file.h> 27 #include <linux/fdtable.h> 28 #include <linux/mm.h> 29 #include <linux/stat.h> 30 #include <linux/fcntl.h> 31 #include <linux/smp_lock.h> 32 #include <linux/swap.h> 33 #include <linux/string.h> 34 #include <linux/init.h> 35 #include <linux/pagemap.h> 36 #include <linux/perf_event.h> 37 #include <linux/highmem.h> 38 #include <linux/spinlock.h> 39 #include <linux/key.h> 40 #include <linux/personality.h> 41 #include <linux/binfmts.h> 42 #include <linux/utsname.h> 43 #include <linux/pid_namespace.h> 44 #include <linux/module.h> 45 #include <linux/namei.h> 46 #include <linux/proc_fs.h> 47 #include <linux/mount.h> 48 #include <linux/security.h> 49 #include <linux/syscalls.h> 50 #include <linux/tsacct_kern.h> 51 #include <linux/cn_proc.h> 52 #include <linux/audit.h> 53 #include <linux/tracehook.h> 54 #include <linux/kmod.h> 55 #include <linux/fsnotify.h> 56 #include <linux/fs_struct.h> 57 #include <linux/pipe_fs_i.h> 58 59 #include <asm/uaccess.h> 60 #include <asm/mmu_context.h> 61 #include <asm/tlb.h> 62 #include "internal.h" 63 64 int core_uses_pid; 65 char core_pattern[CORENAME_MAX_SIZE] = "core"; 66 unsigned int core_pipe_limit; 67 int suid_dumpable = 0; 68 69 /* The maximal length of core_pattern is also specified in sysctl.c */ 70 71 static LIST_HEAD(formats); 72 static DEFINE_RWLOCK(binfmt_lock); 73 74 int __register_binfmt(struct linux_binfmt * fmt, int insert) 75 { 76 if (!fmt) 77 return -EINVAL; 78 write_lock(&binfmt_lock); 79 insert ? list_add(&fmt->lh, &formats) : 80 list_add_tail(&fmt->lh, &formats); 81 write_unlock(&binfmt_lock); 82 return 0; 83 } 84 85 EXPORT_SYMBOL(__register_binfmt); 86 87 void unregister_binfmt(struct linux_binfmt * fmt) 88 { 89 write_lock(&binfmt_lock); 90 list_del(&fmt->lh); 91 write_unlock(&binfmt_lock); 92 } 93 94 EXPORT_SYMBOL(unregister_binfmt); 95 96 static inline void put_binfmt(struct linux_binfmt * fmt) 97 { 98 module_put(fmt->module); 99 } 100 101 /* 102 * Note that a shared library must be both readable and executable due to 103 * security reasons. 104 * 105 * Also note that we take the address to load from from the file itself. 106 */ 107 SYSCALL_DEFINE1(uselib, const char __user *, library) 108 { 109 struct file *file; 110 char *tmp = getname(library); 111 int error = PTR_ERR(tmp); 112 113 if (IS_ERR(tmp)) 114 goto out; 115 116 file = do_filp_open(AT_FDCWD, tmp, 117 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0, 118 MAY_READ | MAY_EXEC | MAY_OPEN); 119 putname(tmp); 120 error = PTR_ERR(file); 121 if (IS_ERR(file)) 122 goto out; 123 124 error = -EINVAL; 125 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode)) 126 goto exit; 127 128 error = -EACCES; 129 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) 130 goto exit; 131 132 fsnotify_open(file->f_path.dentry); 133 134 error = -ENOEXEC; 135 if(file->f_op) { 136 struct linux_binfmt * fmt; 137 138 read_lock(&binfmt_lock); 139 list_for_each_entry(fmt, &formats, lh) { 140 if (!fmt->load_shlib) 141 continue; 142 if (!try_module_get(fmt->module)) 143 continue; 144 read_unlock(&binfmt_lock); 145 error = fmt->load_shlib(file); 146 read_lock(&binfmt_lock); 147 put_binfmt(fmt); 148 if (error != -ENOEXEC) 149 break; 150 } 151 read_unlock(&binfmt_lock); 152 } 153 exit: 154 fput(file); 155 out: 156 return error; 157 } 158 159 #ifdef CONFIG_MMU 160 161 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, 162 int write) 163 { 164 struct page *page; 165 int ret; 166 167 #ifdef CONFIG_STACK_GROWSUP 168 if (write) { 169 ret = expand_stack_downwards(bprm->vma, pos); 170 if (ret < 0) 171 return NULL; 172 } 173 #endif 174 ret = get_user_pages(current, bprm->mm, pos, 175 1, write, 1, &page, NULL); 176 if (ret <= 0) 177 return NULL; 178 179 if (write) { 180 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start; 181 struct rlimit *rlim; 182 183 /* 184 * We've historically supported up to 32 pages (ARG_MAX) 185 * of argument strings even with small stacks 186 */ 187 if (size <= ARG_MAX) 188 return page; 189 190 /* 191 * Limit to 1/4-th the stack size for the argv+env strings. 192 * This ensures that: 193 * - the remaining binfmt code will not run out of stack space, 194 * - the program will have a reasonable amount of stack left 195 * to work from. 196 */ 197 rlim = current->signal->rlim; 198 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) { 199 put_page(page); 200 return NULL; 201 } 202 } 203 204 return page; 205 } 206 207 static void put_arg_page(struct page *page) 208 { 209 put_page(page); 210 } 211 212 static void free_arg_page(struct linux_binprm *bprm, int i) 213 { 214 } 215 216 static void free_arg_pages(struct linux_binprm *bprm) 217 { 218 } 219 220 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, 221 struct page *page) 222 { 223 flush_cache_page(bprm->vma, pos, page_to_pfn(page)); 224 } 225 226 static int __bprm_mm_init(struct linux_binprm *bprm) 227 { 228 int err; 229 struct vm_area_struct *vma = NULL; 230 struct mm_struct *mm = bprm->mm; 231 232 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 233 if (!vma) 234 return -ENOMEM; 235 236 down_write(&mm->mmap_sem); 237 vma->vm_mm = mm; 238 239 /* 240 * Place the stack at the largest stack address the architecture 241 * supports. Later, we'll move this to an appropriate place. We don't 242 * use STACK_TOP because that can depend on attributes which aren't 243 * configured yet. 244 */ 245 vma->vm_end = STACK_TOP_MAX; 246 vma->vm_start = vma->vm_end - PAGE_SIZE; 247 vma->vm_flags = VM_STACK_FLAGS; 248 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 249 err = insert_vm_struct(mm, vma); 250 if (err) 251 goto err; 252 253 mm->stack_vm = mm->total_vm = 1; 254 up_write(&mm->mmap_sem); 255 bprm->p = vma->vm_end - sizeof(void *); 256 return 0; 257 err: 258 up_write(&mm->mmap_sem); 259 bprm->vma = NULL; 260 kmem_cache_free(vm_area_cachep, vma); 261 return err; 262 } 263 264 static bool valid_arg_len(struct linux_binprm *bprm, long len) 265 { 266 return len <= MAX_ARG_STRLEN; 267 } 268 269 #else 270 271 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, 272 int write) 273 { 274 struct page *page; 275 276 page = bprm->page[pos / PAGE_SIZE]; 277 if (!page && write) { 278 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO); 279 if (!page) 280 return NULL; 281 bprm->page[pos / PAGE_SIZE] = page; 282 } 283 284 return page; 285 } 286 287 static void put_arg_page(struct page *page) 288 { 289 } 290 291 static void free_arg_page(struct linux_binprm *bprm, int i) 292 { 293 if (bprm->page[i]) { 294 __free_page(bprm->page[i]); 295 bprm->page[i] = NULL; 296 } 297 } 298 299 static void free_arg_pages(struct linux_binprm *bprm) 300 { 301 int i; 302 303 for (i = 0; i < MAX_ARG_PAGES; i++) 304 free_arg_page(bprm, i); 305 } 306 307 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, 308 struct page *page) 309 { 310 } 311 312 static int __bprm_mm_init(struct linux_binprm *bprm) 313 { 314 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *); 315 return 0; 316 } 317 318 static bool valid_arg_len(struct linux_binprm *bprm, long len) 319 { 320 return len <= bprm->p; 321 } 322 323 #endif /* CONFIG_MMU */ 324 325 /* 326 * Create a new mm_struct and populate it with a temporary stack 327 * vm_area_struct. We don't have enough context at this point to set the stack 328 * flags, permissions, and offset, so we use temporary values. We'll update 329 * them later in setup_arg_pages(). 330 */ 331 int bprm_mm_init(struct linux_binprm *bprm) 332 { 333 int err; 334 struct mm_struct *mm = NULL; 335 336 bprm->mm = mm = mm_alloc(); 337 err = -ENOMEM; 338 if (!mm) 339 goto err; 340 341 err = init_new_context(current, mm); 342 if (err) 343 goto err; 344 345 err = __bprm_mm_init(bprm); 346 if (err) 347 goto err; 348 349 return 0; 350 351 err: 352 if (mm) { 353 bprm->mm = NULL; 354 mmdrop(mm); 355 } 356 357 return err; 358 } 359 360 /* 361 * count() counts the number of strings in array ARGV. 362 */ 363 static int count(char __user * __user * argv, int max) 364 { 365 int i = 0; 366 367 if (argv != NULL) { 368 for (;;) { 369 char __user * p; 370 371 if (get_user(p, argv)) 372 return -EFAULT; 373 if (!p) 374 break; 375 argv++; 376 if (i++ >= max) 377 return -E2BIG; 378 cond_resched(); 379 } 380 } 381 return i; 382 } 383 384 /* 385 * 'copy_strings()' copies argument/environment strings from the old 386 * processes's memory to the new process's stack. The call to get_user_pages() 387 * ensures the destination page is created and not swapped out. 388 */ 389 static int copy_strings(int argc, char __user * __user * argv, 390 struct linux_binprm *bprm) 391 { 392 struct page *kmapped_page = NULL; 393 char *kaddr = NULL; 394 unsigned long kpos = 0; 395 int ret; 396 397 while (argc-- > 0) { 398 char __user *str; 399 int len; 400 unsigned long pos; 401 402 if (get_user(str, argv+argc) || 403 !(len = strnlen_user(str, MAX_ARG_STRLEN))) { 404 ret = -EFAULT; 405 goto out; 406 } 407 408 if (!valid_arg_len(bprm, len)) { 409 ret = -E2BIG; 410 goto out; 411 } 412 413 /* We're going to work our way backwords. */ 414 pos = bprm->p; 415 str += len; 416 bprm->p -= len; 417 418 while (len > 0) { 419 int offset, bytes_to_copy; 420 421 offset = pos % PAGE_SIZE; 422 if (offset == 0) 423 offset = PAGE_SIZE; 424 425 bytes_to_copy = offset; 426 if (bytes_to_copy > len) 427 bytes_to_copy = len; 428 429 offset -= bytes_to_copy; 430 pos -= bytes_to_copy; 431 str -= bytes_to_copy; 432 len -= bytes_to_copy; 433 434 if (!kmapped_page || kpos != (pos & PAGE_MASK)) { 435 struct page *page; 436 437 page = get_arg_page(bprm, pos, 1); 438 if (!page) { 439 ret = -E2BIG; 440 goto out; 441 } 442 443 if (kmapped_page) { 444 flush_kernel_dcache_page(kmapped_page); 445 kunmap(kmapped_page); 446 put_arg_page(kmapped_page); 447 } 448 kmapped_page = page; 449 kaddr = kmap(kmapped_page); 450 kpos = pos & PAGE_MASK; 451 flush_arg_page(bprm, kpos, kmapped_page); 452 } 453 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) { 454 ret = -EFAULT; 455 goto out; 456 } 457 } 458 } 459 ret = 0; 460 out: 461 if (kmapped_page) { 462 flush_kernel_dcache_page(kmapped_page); 463 kunmap(kmapped_page); 464 put_arg_page(kmapped_page); 465 } 466 return ret; 467 } 468 469 /* 470 * Like copy_strings, but get argv and its values from kernel memory. 471 */ 472 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm) 473 { 474 int r; 475 mm_segment_t oldfs = get_fs(); 476 set_fs(KERNEL_DS); 477 r = copy_strings(argc, (char __user * __user *)argv, bprm); 478 set_fs(oldfs); 479 return r; 480 } 481 EXPORT_SYMBOL(copy_strings_kernel); 482 483 #ifdef CONFIG_MMU 484 485 /* 486 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once 487 * the binfmt code determines where the new stack should reside, we shift it to 488 * its final location. The process proceeds as follows: 489 * 490 * 1) Use shift to calculate the new vma endpoints. 491 * 2) Extend vma to cover both the old and new ranges. This ensures the 492 * arguments passed to subsequent functions are consistent. 493 * 3) Move vma's page tables to the new range. 494 * 4) Free up any cleared pgd range. 495 * 5) Shrink the vma to cover only the new range. 496 */ 497 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift) 498 { 499 struct mm_struct *mm = vma->vm_mm; 500 unsigned long old_start = vma->vm_start; 501 unsigned long old_end = vma->vm_end; 502 unsigned long length = old_end - old_start; 503 unsigned long new_start = old_start - shift; 504 unsigned long new_end = old_end - shift; 505 struct mmu_gather *tlb; 506 507 BUG_ON(new_start > new_end); 508 509 /* 510 * ensure there are no vmas between where we want to go 511 * and where we are 512 */ 513 if (vma != find_vma(mm, new_start)) 514 return -EFAULT; 515 516 /* 517 * cover the whole range: [new_start, old_end) 518 */ 519 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL); 520 521 /* 522 * move the page tables downwards, on failure we rely on 523 * process cleanup to remove whatever mess we made. 524 */ 525 if (length != move_page_tables(vma, old_start, 526 vma, new_start, length)) 527 return -ENOMEM; 528 529 lru_add_drain(); 530 tlb = tlb_gather_mmu(mm, 0); 531 if (new_end > old_start) { 532 /* 533 * when the old and new regions overlap clear from new_end. 534 */ 535 free_pgd_range(tlb, new_end, old_end, new_end, 536 vma->vm_next ? vma->vm_next->vm_start : 0); 537 } else { 538 /* 539 * otherwise, clean from old_start; this is done to not touch 540 * the address space in [new_end, old_start) some architectures 541 * have constraints on va-space that make this illegal (IA64) - 542 * for the others its just a little faster. 543 */ 544 free_pgd_range(tlb, old_start, old_end, new_end, 545 vma->vm_next ? vma->vm_next->vm_start : 0); 546 } 547 tlb_finish_mmu(tlb, new_end, old_end); 548 549 /* 550 * shrink the vma to just the new range. 551 */ 552 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL); 553 554 return 0; 555 } 556 557 #define EXTRA_STACK_VM_PAGES 20 /* random */ 558 559 /* 560 * Finalizes the stack vm_area_struct. The flags and permissions are updated, 561 * the stack is optionally relocated, and some extra space is added. 562 */ 563 int setup_arg_pages(struct linux_binprm *bprm, 564 unsigned long stack_top, 565 int executable_stack) 566 { 567 unsigned long ret; 568 unsigned long stack_shift; 569 struct mm_struct *mm = current->mm; 570 struct vm_area_struct *vma = bprm->vma; 571 struct vm_area_struct *prev = NULL; 572 unsigned long vm_flags; 573 unsigned long stack_base; 574 575 #ifdef CONFIG_STACK_GROWSUP 576 /* Limit stack size to 1GB */ 577 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max; 578 if (stack_base > (1 << 30)) 579 stack_base = 1 << 30; 580 581 /* Make sure we didn't let the argument array grow too large. */ 582 if (vma->vm_end - vma->vm_start > stack_base) 583 return -ENOMEM; 584 585 stack_base = PAGE_ALIGN(stack_top - stack_base); 586 587 stack_shift = vma->vm_start - stack_base; 588 mm->arg_start = bprm->p - stack_shift; 589 bprm->p = vma->vm_end - stack_shift; 590 #else 591 stack_top = arch_align_stack(stack_top); 592 stack_top = PAGE_ALIGN(stack_top); 593 stack_shift = vma->vm_end - stack_top; 594 595 bprm->p -= stack_shift; 596 mm->arg_start = bprm->p; 597 #endif 598 599 if (bprm->loader) 600 bprm->loader -= stack_shift; 601 bprm->exec -= stack_shift; 602 603 down_write(&mm->mmap_sem); 604 vm_flags = VM_STACK_FLAGS; 605 606 /* 607 * Adjust stack execute permissions; explicitly enable for 608 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone 609 * (arch default) otherwise. 610 */ 611 if (unlikely(executable_stack == EXSTACK_ENABLE_X)) 612 vm_flags |= VM_EXEC; 613 else if (executable_stack == EXSTACK_DISABLE_X) 614 vm_flags &= ~VM_EXEC; 615 vm_flags |= mm->def_flags; 616 617 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end, 618 vm_flags); 619 if (ret) 620 goto out_unlock; 621 BUG_ON(prev != vma); 622 623 /* Move stack pages down in memory. */ 624 if (stack_shift) { 625 ret = shift_arg_pages(vma, stack_shift); 626 if (ret) 627 goto out_unlock; 628 } 629 630 #ifdef CONFIG_STACK_GROWSUP 631 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE; 632 #else 633 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE; 634 #endif 635 ret = expand_stack(vma, stack_base); 636 if (ret) 637 ret = -EFAULT; 638 639 out_unlock: 640 up_write(&mm->mmap_sem); 641 return ret; 642 } 643 EXPORT_SYMBOL(setup_arg_pages); 644 645 #endif /* CONFIG_MMU */ 646 647 struct file *open_exec(const char *name) 648 { 649 struct file *file; 650 int err; 651 652 file = do_filp_open(AT_FDCWD, name, 653 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0, 654 MAY_EXEC | MAY_OPEN); 655 if (IS_ERR(file)) 656 goto out; 657 658 err = -EACCES; 659 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode)) 660 goto exit; 661 662 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) 663 goto exit; 664 665 fsnotify_open(file->f_path.dentry); 666 667 err = deny_write_access(file); 668 if (err) 669 goto exit; 670 671 out: 672 return file; 673 674 exit: 675 fput(file); 676 return ERR_PTR(err); 677 } 678 EXPORT_SYMBOL(open_exec); 679 680 int kernel_read(struct file *file, loff_t offset, 681 char *addr, unsigned long count) 682 { 683 mm_segment_t old_fs; 684 loff_t pos = offset; 685 int result; 686 687 old_fs = get_fs(); 688 set_fs(get_ds()); 689 /* The cast to a user pointer is valid due to the set_fs() */ 690 result = vfs_read(file, (void __user *)addr, count, &pos); 691 set_fs(old_fs); 692 return result; 693 } 694 695 EXPORT_SYMBOL(kernel_read); 696 697 static int exec_mmap(struct mm_struct *mm) 698 { 699 struct task_struct *tsk; 700 struct mm_struct * old_mm, *active_mm; 701 702 /* Notify parent that we're no longer interested in the old VM */ 703 tsk = current; 704 old_mm = current->mm; 705 mm_release(tsk, old_mm); 706 707 if (old_mm) { 708 /* 709 * Make sure that if there is a core dump in progress 710 * for the old mm, we get out and die instead of going 711 * through with the exec. We must hold mmap_sem around 712 * checking core_state and changing tsk->mm. 713 */ 714 down_read(&old_mm->mmap_sem); 715 if (unlikely(old_mm->core_state)) { 716 up_read(&old_mm->mmap_sem); 717 return -EINTR; 718 } 719 } 720 task_lock(tsk); 721 active_mm = tsk->active_mm; 722 tsk->mm = mm; 723 tsk->active_mm = mm; 724 activate_mm(active_mm, mm); 725 task_unlock(tsk); 726 arch_pick_mmap_layout(mm); 727 if (old_mm) { 728 up_read(&old_mm->mmap_sem); 729 BUG_ON(active_mm != old_mm); 730 mm_update_next_owner(old_mm); 731 mmput(old_mm); 732 return 0; 733 } 734 mmdrop(active_mm); 735 return 0; 736 } 737 738 /* 739 * This function makes sure the current process has its own signal table, 740 * so that flush_signal_handlers can later reset the handlers without 741 * disturbing other processes. (Other processes might share the signal 742 * table via the CLONE_SIGHAND option to clone().) 743 */ 744 static int de_thread(struct task_struct *tsk) 745 { 746 struct signal_struct *sig = tsk->signal; 747 struct sighand_struct *oldsighand = tsk->sighand; 748 spinlock_t *lock = &oldsighand->siglock; 749 int count; 750 751 if (thread_group_empty(tsk)) 752 goto no_thread_group; 753 754 /* 755 * Kill all other threads in the thread group. 756 */ 757 spin_lock_irq(lock); 758 if (signal_group_exit(sig)) { 759 /* 760 * Another group action in progress, just 761 * return so that the signal is processed. 762 */ 763 spin_unlock_irq(lock); 764 return -EAGAIN; 765 } 766 sig->group_exit_task = tsk; 767 zap_other_threads(tsk); 768 769 /* Account for the thread group leader hanging around: */ 770 count = thread_group_leader(tsk) ? 1 : 2; 771 sig->notify_count = count; 772 while (atomic_read(&sig->count) > count) { 773 __set_current_state(TASK_UNINTERRUPTIBLE); 774 spin_unlock_irq(lock); 775 schedule(); 776 spin_lock_irq(lock); 777 } 778 spin_unlock_irq(lock); 779 780 /* 781 * At this point all other threads have exited, all we have to 782 * do is to wait for the thread group leader to become inactive, 783 * and to assume its PID: 784 */ 785 if (!thread_group_leader(tsk)) { 786 struct task_struct *leader = tsk->group_leader; 787 788 sig->notify_count = -1; /* for exit_notify() */ 789 for (;;) { 790 write_lock_irq(&tasklist_lock); 791 if (likely(leader->exit_state)) 792 break; 793 __set_current_state(TASK_UNINTERRUPTIBLE); 794 write_unlock_irq(&tasklist_lock); 795 schedule(); 796 } 797 798 /* 799 * The only record we have of the real-time age of a 800 * process, regardless of execs it's done, is start_time. 801 * All the past CPU time is accumulated in signal_struct 802 * from sister threads now dead. But in this non-leader 803 * exec, nothing survives from the original leader thread, 804 * whose birth marks the true age of this process now. 805 * When we take on its identity by switching to its PID, we 806 * also take its birthdate (always earlier than our own). 807 */ 808 tsk->start_time = leader->start_time; 809 810 BUG_ON(!same_thread_group(leader, tsk)); 811 BUG_ON(has_group_leader_pid(tsk)); 812 /* 813 * An exec() starts a new thread group with the 814 * TGID of the previous thread group. Rehash the 815 * two threads with a switched PID, and release 816 * the former thread group leader: 817 */ 818 819 /* Become a process group leader with the old leader's pid. 820 * The old leader becomes a thread of the this thread group. 821 * Note: The old leader also uses this pid until release_task 822 * is called. Odd but simple and correct. 823 */ 824 detach_pid(tsk, PIDTYPE_PID); 825 tsk->pid = leader->pid; 826 attach_pid(tsk, PIDTYPE_PID, task_pid(leader)); 827 transfer_pid(leader, tsk, PIDTYPE_PGID); 828 transfer_pid(leader, tsk, PIDTYPE_SID); 829 830 list_replace_rcu(&leader->tasks, &tsk->tasks); 831 list_replace_init(&leader->sibling, &tsk->sibling); 832 833 tsk->group_leader = tsk; 834 leader->group_leader = tsk; 835 836 tsk->exit_signal = SIGCHLD; 837 838 BUG_ON(leader->exit_state != EXIT_ZOMBIE); 839 leader->exit_state = EXIT_DEAD; 840 write_unlock_irq(&tasklist_lock); 841 842 release_task(leader); 843 } 844 845 sig->group_exit_task = NULL; 846 sig->notify_count = 0; 847 848 no_thread_group: 849 if (current->mm) 850 setmax_mm_hiwater_rss(&sig->maxrss, current->mm); 851 852 exit_itimers(sig); 853 flush_itimer_signals(); 854 855 if (atomic_read(&oldsighand->count) != 1) { 856 struct sighand_struct *newsighand; 857 /* 858 * This ->sighand is shared with the CLONE_SIGHAND 859 * but not CLONE_THREAD task, switch to the new one. 860 */ 861 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 862 if (!newsighand) 863 return -ENOMEM; 864 865 atomic_set(&newsighand->count, 1); 866 memcpy(newsighand->action, oldsighand->action, 867 sizeof(newsighand->action)); 868 869 write_lock_irq(&tasklist_lock); 870 spin_lock(&oldsighand->siglock); 871 rcu_assign_pointer(tsk->sighand, newsighand); 872 spin_unlock(&oldsighand->siglock); 873 write_unlock_irq(&tasklist_lock); 874 875 __cleanup_sighand(oldsighand); 876 } 877 878 BUG_ON(!thread_group_leader(tsk)); 879 return 0; 880 } 881 882 /* 883 * These functions flushes out all traces of the currently running executable 884 * so that a new one can be started 885 */ 886 static void flush_old_files(struct files_struct * files) 887 { 888 long j = -1; 889 struct fdtable *fdt; 890 891 spin_lock(&files->file_lock); 892 for (;;) { 893 unsigned long set, i; 894 895 j++; 896 i = j * __NFDBITS; 897 fdt = files_fdtable(files); 898 if (i >= fdt->max_fds) 899 break; 900 set = fdt->close_on_exec->fds_bits[j]; 901 if (!set) 902 continue; 903 fdt->close_on_exec->fds_bits[j] = 0; 904 spin_unlock(&files->file_lock); 905 for ( ; set ; i++,set >>= 1) { 906 if (set & 1) { 907 sys_close(i); 908 } 909 } 910 spin_lock(&files->file_lock); 911 912 } 913 spin_unlock(&files->file_lock); 914 } 915 916 char *get_task_comm(char *buf, struct task_struct *tsk) 917 { 918 /* buf must be at least sizeof(tsk->comm) in size */ 919 task_lock(tsk); 920 strncpy(buf, tsk->comm, sizeof(tsk->comm)); 921 task_unlock(tsk); 922 return buf; 923 } 924 925 void set_task_comm(struct task_struct *tsk, char *buf) 926 { 927 task_lock(tsk); 928 929 /* 930 * Threads may access current->comm without holding 931 * the task lock, so write the string carefully. 932 * Readers without a lock may see incomplete new 933 * names but are safe from non-terminating string reads. 934 */ 935 memset(tsk->comm, 0, TASK_COMM_LEN); 936 wmb(); 937 strlcpy(tsk->comm, buf, sizeof(tsk->comm)); 938 task_unlock(tsk); 939 perf_event_comm(tsk); 940 } 941 942 int flush_old_exec(struct linux_binprm * bprm) 943 { 944 char * name; 945 int i, ch, retval; 946 char tcomm[sizeof(current->comm)]; 947 948 /* 949 * Make sure we have a private signal table and that 950 * we are unassociated from the previous thread group. 951 */ 952 retval = de_thread(current); 953 if (retval) 954 goto out; 955 956 set_mm_exe_file(bprm->mm, bprm->file); 957 958 /* 959 * Release all of the old mmap stuff 960 */ 961 retval = exec_mmap(bprm->mm); 962 if (retval) 963 goto out; 964 965 bprm->mm = NULL; /* We're using it now */ 966 967 /* This is the point of no return */ 968 current->sas_ss_sp = current->sas_ss_size = 0; 969 970 if (current_euid() == current_uid() && current_egid() == current_gid()) 971 set_dumpable(current->mm, 1); 972 else 973 set_dumpable(current->mm, suid_dumpable); 974 975 name = bprm->filename; 976 977 /* Copies the binary name from after last slash */ 978 for (i=0; (ch = *(name++)) != '\0';) { 979 if (ch == '/') 980 i = 0; /* overwrite what we wrote */ 981 else 982 if (i < (sizeof(tcomm) - 1)) 983 tcomm[i++] = ch; 984 } 985 tcomm[i] = '\0'; 986 set_task_comm(current, tcomm); 987 988 current->flags &= ~PF_RANDOMIZE; 989 flush_thread(); 990 991 /* Set the new mm task size. We have to do that late because it may 992 * depend on TIF_32BIT which is only updated in flush_thread() on 993 * some architectures like powerpc 994 */ 995 current->mm->task_size = TASK_SIZE; 996 997 /* install the new credentials */ 998 if (bprm->cred->uid != current_euid() || 999 bprm->cred->gid != current_egid()) { 1000 current->pdeath_signal = 0; 1001 } else if (file_permission(bprm->file, MAY_READ) || 1002 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) { 1003 set_dumpable(current->mm, suid_dumpable); 1004 } 1005 1006 current->personality &= ~bprm->per_clear; 1007 1008 /* 1009 * Flush performance counters when crossing a 1010 * security domain: 1011 */ 1012 if (!get_dumpable(current->mm)) 1013 perf_event_exit_task(current); 1014 1015 /* An exec changes our domain. We are no longer part of the thread 1016 group */ 1017 1018 current->self_exec_id++; 1019 1020 flush_signal_handlers(current, 0); 1021 flush_old_files(current->files); 1022 1023 return 0; 1024 1025 out: 1026 return retval; 1027 } 1028 1029 EXPORT_SYMBOL(flush_old_exec); 1030 1031 /* 1032 * Prepare credentials and lock ->cred_guard_mutex. 1033 * install_exec_creds() commits the new creds and drops the lock. 1034 * Or, if exec fails before, free_bprm() should release ->cred and 1035 * and unlock. 1036 */ 1037 int prepare_bprm_creds(struct linux_binprm *bprm) 1038 { 1039 if (mutex_lock_interruptible(¤t->cred_guard_mutex)) 1040 return -ERESTARTNOINTR; 1041 1042 bprm->cred = prepare_exec_creds(); 1043 if (likely(bprm->cred)) 1044 return 0; 1045 1046 mutex_unlock(¤t->cred_guard_mutex); 1047 return -ENOMEM; 1048 } 1049 1050 void free_bprm(struct linux_binprm *bprm) 1051 { 1052 free_arg_pages(bprm); 1053 if (bprm->cred) { 1054 mutex_unlock(¤t->cred_guard_mutex); 1055 abort_creds(bprm->cred); 1056 } 1057 kfree(bprm); 1058 } 1059 1060 /* 1061 * install the new credentials for this executable 1062 */ 1063 void install_exec_creds(struct linux_binprm *bprm) 1064 { 1065 security_bprm_committing_creds(bprm); 1066 1067 commit_creds(bprm->cred); 1068 bprm->cred = NULL; 1069 /* 1070 * cred_guard_mutex must be held at least to this point to prevent 1071 * ptrace_attach() from altering our determination of the task's 1072 * credentials; any time after this it may be unlocked. 1073 */ 1074 security_bprm_committed_creds(bprm); 1075 mutex_unlock(¤t->cred_guard_mutex); 1076 } 1077 EXPORT_SYMBOL(install_exec_creds); 1078 1079 /* 1080 * determine how safe it is to execute the proposed program 1081 * - the caller must hold current->cred_guard_mutex to protect against 1082 * PTRACE_ATTACH 1083 */ 1084 int check_unsafe_exec(struct linux_binprm *bprm) 1085 { 1086 struct task_struct *p = current, *t; 1087 unsigned n_fs; 1088 int res = 0; 1089 1090 bprm->unsafe = tracehook_unsafe_exec(p); 1091 1092 n_fs = 1; 1093 write_lock(&p->fs->lock); 1094 rcu_read_lock(); 1095 for (t = next_thread(p); t != p; t = next_thread(t)) { 1096 if (t->fs == p->fs) 1097 n_fs++; 1098 } 1099 rcu_read_unlock(); 1100 1101 if (p->fs->users > n_fs) { 1102 bprm->unsafe |= LSM_UNSAFE_SHARE; 1103 } else { 1104 res = -EAGAIN; 1105 if (!p->fs->in_exec) { 1106 p->fs->in_exec = 1; 1107 res = 1; 1108 } 1109 } 1110 write_unlock(&p->fs->lock); 1111 1112 return res; 1113 } 1114 1115 /* 1116 * Fill the binprm structure from the inode. 1117 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes 1118 * 1119 * This may be called multiple times for binary chains (scripts for example). 1120 */ 1121 int prepare_binprm(struct linux_binprm *bprm) 1122 { 1123 umode_t mode; 1124 struct inode * inode = bprm->file->f_path.dentry->d_inode; 1125 int retval; 1126 1127 mode = inode->i_mode; 1128 if (bprm->file->f_op == NULL) 1129 return -EACCES; 1130 1131 /* clear any previous set[ug]id data from a previous binary */ 1132 bprm->cred->euid = current_euid(); 1133 bprm->cred->egid = current_egid(); 1134 1135 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) { 1136 /* Set-uid? */ 1137 if (mode & S_ISUID) { 1138 bprm->per_clear |= PER_CLEAR_ON_SETID; 1139 bprm->cred->euid = inode->i_uid; 1140 } 1141 1142 /* Set-gid? */ 1143 /* 1144 * If setgid is set but no group execute bit then this 1145 * is a candidate for mandatory locking, not a setgid 1146 * executable. 1147 */ 1148 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) { 1149 bprm->per_clear |= PER_CLEAR_ON_SETID; 1150 bprm->cred->egid = inode->i_gid; 1151 } 1152 } 1153 1154 /* fill in binprm security blob */ 1155 retval = security_bprm_set_creds(bprm); 1156 if (retval) 1157 return retval; 1158 bprm->cred_prepared = 1; 1159 1160 memset(bprm->buf, 0, BINPRM_BUF_SIZE); 1161 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE); 1162 } 1163 1164 EXPORT_SYMBOL(prepare_binprm); 1165 1166 /* 1167 * Arguments are '\0' separated strings found at the location bprm->p 1168 * points to; chop off the first by relocating brpm->p to right after 1169 * the first '\0' encountered. 1170 */ 1171 int remove_arg_zero(struct linux_binprm *bprm) 1172 { 1173 int ret = 0; 1174 unsigned long offset; 1175 char *kaddr; 1176 struct page *page; 1177 1178 if (!bprm->argc) 1179 return 0; 1180 1181 do { 1182 offset = bprm->p & ~PAGE_MASK; 1183 page = get_arg_page(bprm, bprm->p, 0); 1184 if (!page) { 1185 ret = -EFAULT; 1186 goto out; 1187 } 1188 kaddr = kmap_atomic(page, KM_USER0); 1189 1190 for (; offset < PAGE_SIZE && kaddr[offset]; 1191 offset++, bprm->p++) 1192 ; 1193 1194 kunmap_atomic(kaddr, KM_USER0); 1195 put_arg_page(page); 1196 1197 if (offset == PAGE_SIZE) 1198 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1); 1199 } while (offset == PAGE_SIZE); 1200 1201 bprm->p++; 1202 bprm->argc--; 1203 ret = 0; 1204 1205 out: 1206 return ret; 1207 } 1208 EXPORT_SYMBOL(remove_arg_zero); 1209 1210 /* 1211 * cycle the list of binary formats handler, until one recognizes the image 1212 */ 1213 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs) 1214 { 1215 unsigned int depth = bprm->recursion_depth; 1216 int try,retval; 1217 struct linux_binfmt *fmt; 1218 1219 retval = security_bprm_check(bprm); 1220 if (retval) 1221 return retval; 1222 1223 /* kernel module loader fixup */ 1224 /* so we don't try to load run modprobe in kernel space. */ 1225 set_fs(USER_DS); 1226 1227 retval = audit_bprm(bprm); 1228 if (retval) 1229 return retval; 1230 1231 retval = -ENOENT; 1232 for (try=0; try<2; try++) { 1233 read_lock(&binfmt_lock); 1234 list_for_each_entry(fmt, &formats, lh) { 1235 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary; 1236 if (!fn) 1237 continue; 1238 if (!try_module_get(fmt->module)) 1239 continue; 1240 read_unlock(&binfmt_lock); 1241 retval = fn(bprm, regs); 1242 /* 1243 * Restore the depth counter to its starting value 1244 * in this call, so we don't have to rely on every 1245 * load_binary function to restore it on return. 1246 */ 1247 bprm->recursion_depth = depth; 1248 if (retval >= 0) { 1249 if (depth == 0) 1250 tracehook_report_exec(fmt, bprm, regs); 1251 put_binfmt(fmt); 1252 allow_write_access(bprm->file); 1253 if (bprm->file) 1254 fput(bprm->file); 1255 bprm->file = NULL; 1256 current->did_exec = 1; 1257 proc_exec_connector(current); 1258 return retval; 1259 } 1260 read_lock(&binfmt_lock); 1261 put_binfmt(fmt); 1262 if (retval != -ENOEXEC || bprm->mm == NULL) 1263 break; 1264 if (!bprm->file) { 1265 read_unlock(&binfmt_lock); 1266 return retval; 1267 } 1268 } 1269 read_unlock(&binfmt_lock); 1270 if (retval != -ENOEXEC || bprm->mm == NULL) { 1271 break; 1272 #ifdef CONFIG_MODULES 1273 } else { 1274 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e)) 1275 if (printable(bprm->buf[0]) && 1276 printable(bprm->buf[1]) && 1277 printable(bprm->buf[2]) && 1278 printable(bprm->buf[3])) 1279 break; /* -ENOEXEC */ 1280 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2])); 1281 #endif 1282 } 1283 } 1284 return retval; 1285 } 1286 1287 EXPORT_SYMBOL(search_binary_handler); 1288 1289 /* 1290 * sys_execve() executes a new program. 1291 */ 1292 int do_execve(char * filename, 1293 char __user *__user *argv, 1294 char __user *__user *envp, 1295 struct pt_regs * regs) 1296 { 1297 struct linux_binprm *bprm; 1298 struct file *file; 1299 struct files_struct *displaced; 1300 bool clear_in_exec; 1301 int retval; 1302 1303 retval = unshare_files(&displaced); 1304 if (retval) 1305 goto out_ret; 1306 1307 retval = -ENOMEM; 1308 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL); 1309 if (!bprm) 1310 goto out_files; 1311 1312 retval = prepare_bprm_creds(bprm); 1313 if (retval) 1314 goto out_free; 1315 1316 retval = check_unsafe_exec(bprm); 1317 if (retval < 0) 1318 goto out_free; 1319 clear_in_exec = retval; 1320 current->in_execve = 1; 1321 1322 file = open_exec(filename); 1323 retval = PTR_ERR(file); 1324 if (IS_ERR(file)) 1325 goto out_unmark; 1326 1327 sched_exec(); 1328 1329 bprm->file = file; 1330 bprm->filename = filename; 1331 bprm->interp = filename; 1332 1333 retval = bprm_mm_init(bprm); 1334 if (retval) 1335 goto out_file; 1336 1337 bprm->argc = count(argv, MAX_ARG_STRINGS); 1338 if ((retval = bprm->argc) < 0) 1339 goto out; 1340 1341 bprm->envc = count(envp, MAX_ARG_STRINGS); 1342 if ((retval = bprm->envc) < 0) 1343 goto out; 1344 1345 retval = prepare_binprm(bprm); 1346 if (retval < 0) 1347 goto out; 1348 1349 retval = copy_strings_kernel(1, &bprm->filename, bprm); 1350 if (retval < 0) 1351 goto out; 1352 1353 bprm->exec = bprm->p; 1354 retval = copy_strings(bprm->envc, envp, bprm); 1355 if (retval < 0) 1356 goto out; 1357 1358 retval = copy_strings(bprm->argc, argv, bprm); 1359 if (retval < 0) 1360 goto out; 1361 1362 current->flags &= ~PF_KTHREAD; 1363 retval = search_binary_handler(bprm,regs); 1364 if (retval < 0) 1365 goto out; 1366 1367 current->stack_start = current->mm->start_stack; 1368 1369 /* execve succeeded */ 1370 current->fs->in_exec = 0; 1371 current->in_execve = 0; 1372 acct_update_integrals(current); 1373 free_bprm(bprm); 1374 if (displaced) 1375 put_files_struct(displaced); 1376 return retval; 1377 1378 out: 1379 if (bprm->mm) 1380 mmput (bprm->mm); 1381 1382 out_file: 1383 if (bprm->file) { 1384 allow_write_access(bprm->file); 1385 fput(bprm->file); 1386 } 1387 1388 out_unmark: 1389 if (clear_in_exec) 1390 current->fs->in_exec = 0; 1391 current->in_execve = 0; 1392 1393 out_free: 1394 free_bprm(bprm); 1395 1396 out_files: 1397 if (displaced) 1398 reset_files_struct(displaced); 1399 out_ret: 1400 return retval; 1401 } 1402 1403 void set_binfmt(struct linux_binfmt *new) 1404 { 1405 struct mm_struct *mm = current->mm; 1406 1407 if (mm->binfmt) 1408 module_put(mm->binfmt->module); 1409 1410 mm->binfmt = new; 1411 if (new) 1412 __module_get(new->module); 1413 } 1414 1415 EXPORT_SYMBOL(set_binfmt); 1416 1417 /* format_corename will inspect the pattern parameter, and output a 1418 * name into corename, which must have space for at least 1419 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator. 1420 */ 1421 static int format_corename(char *corename, long signr) 1422 { 1423 const struct cred *cred = current_cred(); 1424 const char *pat_ptr = core_pattern; 1425 int ispipe = (*pat_ptr == '|'); 1426 char *out_ptr = corename; 1427 char *const out_end = corename + CORENAME_MAX_SIZE; 1428 int rc; 1429 int pid_in_pattern = 0; 1430 1431 /* Repeat as long as we have more pattern to process and more output 1432 space */ 1433 while (*pat_ptr) { 1434 if (*pat_ptr != '%') { 1435 if (out_ptr == out_end) 1436 goto out; 1437 *out_ptr++ = *pat_ptr++; 1438 } else { 1439 switch (*++pat_ptr) { 1440 case 0: 1441 goto out; 1442 /* Double percent, output one percent */ 1443 case '%': 1444 if (out_ptr == out_end) 1445 goto out; 1446 *out_ptr++ = '%'; 1447 break; 1448 /* pid */ 1449 case 'p': 1450 pid_in_pattern = 1; 1451 rc = snprintf(out_ptr, out_end - out_ptr, 1452 "%d", task_tgid_vnr(current)); 1453 if (rc > out_end - out_ptr) 1454 goto out; 1455 out_ptr += rc; 1456 break; 1457 /* uid */ 1458 case 'u': 1459 rc = snprintf(out_ptr, out_end - out_ptr, 1460 "%d", cred->uid); 1461 if (rc > out_end - out_ptr) 1462 goto out; 1463 out_ptr += rc; 1464 break; 1465 /* gid */ 1466 case 'g': 1467 rc = snprintf(out_ptr, out_end - out_ptr, 1468 "%d", cred->gid); 1469 if (rc > out_end - out_ptr) 1470 goto out; 1471 out_ptr += rc; 1472 break; 1473 /* signal that caused the coredump */ 1474 case 's': 1475 rc = snprintf(out_ptr, out_end - out_ptr, 1476 "%ld", signr); 1477 if (rc > out_end - out_ptr) 1478 goto out; 1479 out_ptr += rc; 1480 break; 1481 /* UNIX time of coredump */ 1482 case 't': { 1483 struct timeval tv; 1484 do_gettimeofday(&tv); 1485 rc = snprintf(out_ptr, out_end - out_ptr, 1486 "%lu", tv.tv_sec); 1487 if (rc > out_end - out_ptr) 1488 goto out; 1489 out_ptr += rc; 1490 break; 1491 } 1492 /* hostname */ 1493 case 'h': 1494 down_read(&uts_sem); 1495 rc = snprintf(out_ptr, out_end - out_ptr, 1496 "%s", utsname()->nodename); 1497 up_read(&uts_sem); 1498 if (rc > out_end - out_ptr) 1499 goto out; 1500 out_ptr += rc; 1501 break; 1502 /* executable */ 1503 case 'e': 1504 rc = snprintf(out_ptr, out_end - out_ptr, 1505 "%s", current->comm); 1506 if (rc > out_end - out_ptr) 1507 goto out; 1508 out_ptr += rc; 1509 break; 1510 /* core limit size */ 1511 case 'c': 1512 rc = snprintf(out_ptr, out_end - out_ptr, 1513 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur); 1514 if (rc > out_end - out_ptr) 1515 goto out; 1516 out_ptr += rc; 1517 break; 1518 default: 1519 break; 1520 } 1521 ++pat_ptr; 1522 } 1523 } 1524 /* Backward compatibility with core_uses_pid: 1525 * 1526 * If core_pattern does not include a %p (as is the default) 1527 * and core_uses_pid is set, then .%pid will be appended to 1528 * the filename. Do not do this for piped commands. */ 1529 if (!ispipe && !pid_in_pattern && core_uses_pid) { 1530 rc = snprintf(out_ptr, out_end - out_ptr, 1531 ".%d", task_tgid_vnr(current)); 1532 if (rc > out_end - out_ptr) 1533 goto out; 1534 out_ptr += rc; 1535 } 1536 out: 1537 *out_ptr = 0; 1538 return ispipe; 1539 } 1540 1541 static int zap_process(struct task_struct *start) 1542 { 1543 struct task_struct *t; 1544 int nr = 0; 1545 1546 start->signal->flags = SIGNAL_GROUP_EXIT; 1547 start->signal->group_stop_count = 0; 1548 1549 t = start; 1550 do { 1551 if (t != current && t->mm) { 1552 sigaddset(&t->pending.signal, SIGKILL); 1553 signal_wake_up(t, 1); 1554 nr++; 1555 } 1556 } while_each_thread(start, t); 1557 1558 return nr; 1559 } 1560 1561 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm, 1562 struct core_state *core_state, int exit_code) 1563 { 1564 struct task_struct *g, *p; 1565 unsigned long flags; 1566 int nr = -EAGAIN; 1567 1568 spin_lock_irq(&tsk->sighand->siglock); 1569 if (!signal_group_exit(tsk->signal)) { 1570 mm->core_state = core_state; 1571 tsk->signal->group_exit_code = exit_code; 1572 nr = zap_process(tsk); 1573 } 1574 spin_unlock_irq(&tsk->sighand->siglock); 1575 if (unlikely(nr < 0)) 1576 return nr; 1577 1578 if (atomic_read(&mm->mm_users) == nr + 1) 1579 goto done; 1580 /* 1581 * We should find and kill all tasks which use this mm, and we should 1582 * count them correctly into ->nr_threads. We don't take tasklist 1583 * lock, but this is safe wrt: 1584 * 1585 * fork: 1586 * None of sub-threads can fork after zap_process(leader). All 1587 * processes which were created before this point should be 1588 * visible to zap_threads() because copy_process() adds the new 1589 * process to the tail of init_task.tasks list, and lock/unlock 1590 * of ->siglock provides a memory barrier. 1591 * 1592 * do_exit: 1593 * The caller holds mm->mmap_sem. This means that the task which 1594 * uses this mm can't pass exit_mm(), so it can't exit or clear 1595 * its ->mm. 1596 * 1597 * de_thread: 1598 * It does list_replace_rcu(&leader->tasks, ¤t->tasks), 1599 * we must see either old or new leader, this does not matter. 1600 * However, it can change p->sighand, so lock_task_sighand(p) 1601 * must be used. Since p->mm != NULL and we hold ->mmap_sem 1602 * it can't fail. 1603 * 1604 * Note also that "g" can be the old leader with ->mm == NULL 1605 * and already unhashed and thus removed from ->thread_group. 1606 * This is OK, __unhash_process()->list_del_rcu() does not 1607 * clear the ->next pointer, we will find the new leader via 1608 * next_thread(). 1609 */ 1610 rcu_read_lock(); 1611 for_each_process(g) { 1612 if (g == tsk->group_leader) 1613 continue; 1614 if (g->flags & PF_KTHREAD) 1615 continue; 1616 p = g; 1617 do { 1618 if (p->mm) { 1619 if (unlikely(p->mm == mm)) { 1620 lock_task_sighand(p, &flags); 1621 nr += zap_process(p); 1622 unlock_task_sighand(p, &flags); 1623 } 1624 break; 1625 } 1626 } while_each_thread(g, p); 1627 } 1628 rcu_read_unlock(); 1629 done: 1630 atomic_set(&core_state->nr_threads, nr); 1631 return nr; 1632 } 1633 1634 static int coredump_wait(int exit_code, struct core_state *core_state) 1635 { 1636 struct task_struct *tsk = current; 1637 struct mm_struct *mm = tsk->mm; 1638 struct completion *vfork_done; 1639 int core_waiters; 1640 1641 init_completion(&core_state->startup); 1642 core_state->dumper.task = tsk; 1643 core_state->dumper.next = NULL; 1644 core_waiters = zap_threads(tsk, mm, core_state, exit_code); 1645 up_write(&mm->mmap_sem); 1646 1647 if (unlikely(core_waiters < 0)) 1648 goto fail; 1649 1650 /* 1651 * Make sure nobody is waiting for us to release the VM, 1652 * otherwise we can deadlock when we wait on each other 1653 */ 1654 vfork_done = tsk->vfork_done; 1655 if (vfork_done) { 1656 tsk->vfork_done = NULL; 1657 complete(vfork_done); 1658 } 1659 1660 if (core_waiters) 1661 wait_for_completion(&core_state->startup); 1662 fail: 1663 return core_waiters; 1664 } 1665 1666 static void coredump_finish(struct mm_struct *mm) 1667 { 1668 struct core_thread *curr, *next; 1669 struct task_struct *task; 1670 1671 next = mm->core_state->dumper.next; 1672 while ((curr = next) != NULL) { 1673 next = curr->next; 1674 task = curr->task; 1675 /* 1676 * see exit_mm(), curr->task must not see 1677 * ->task == NULL before we read ->next. 1678 */ 1679 smp_mb(); 1680 curr->task = NULL; 1681 wake_up_process(task); 1682 } 1683 1684 mm->core_state = NULL; 1685 } 1686 1687 /* 1688 * set_dumpable converts traditional three-value dumpable to two flags and 1689 * stores them into mm->flags. It modifies lower two bits of mm->flags, but 1690 * these bits are not changed atomically. So get_dumpable can observe the 1691 * intermediate state. To avoid doing unexpected behavior, get get_dumpable 1692 * return either old dumpable or new one by paying attention to the order of 1693 * modifying the bits. 1694 * 1695 * dumpable | mm->flags (binary) 1696 * old new | initial interim final 1697 * ---------+----------------------- 1698 * 0 1 | 00 01 01 1699 * 0 2 | 00 10(*) 11 1700 * 1 0 | 01 00 00 1701 * 1 2 | 01 11 11 1702 * 2 0 | 11 10(*) 00 1703 * 2 1 | 11 11 01 1704 * 1705 * (*) get_dumpable regards interim value of 10 as 11. 1706 */ 1707 void set_dumpable(struct mm_struct *mm, int value) 1708 { 1709 switch (value) { 1710 case 0: 1711 clear_bit(MMF_DUMPABLE, &mm->flags); 1712 smp_wmb(); 1713 clear_bit(MMF_DUMP_SECURELY, &mm->flags); 1714 break; 1715 case 1: 1716 set_bit(MMF_DUMPABLE, &mm->flags); 1717 smp_wmb(); 1718 clear_bit(MMF_DUMP_SECURELY, &mm->flags); 1719 break; 1720 case 2: 1721 set_bit(MMF_DUMP_SECURELY, &mm->flags); 1722 smp_wmb(); 1723 set_bit(MMF_DUMPABLE, &mm->flags); 1724 break; 1725 } 1726 } 1727 1728 int get_dumpable(struct mm_struct *mm) 1729 { 1730 int ret; 1731 1732 ret = mm->flags & 0x3; 1733 return (ret >= 2) ? 2 : ret; 1734 } 1735 1736 static void wait_for_dump_helpers(struct file *file) 1737 { 1738 struct pipe_inode_info *pipe; 1739 1740 pipe = file->f_path.dentry->d_inode->i_pipe; 1741 1742 pipe_lock(pipe); 1743 pipe->readers++; 1744 pipe->writers--; 1745 1746 while ((pipe->readers > 1) && (!signal_pending(current))) { 1747 wake_up_interruptible_sync(&pipe->wait); 1748 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 1749 pipe_wait(pipe); 1750 } 1751 1752 pipe->readers--; 1753 pipe->writers++; 1754 pipe_unlock(pipe); 1755 1756 } 1757 1758 1759 void do_coredump(long signr, int exit_code, struct pt_regs *regs) 1760 { 1761 struct core_state core_state; 1762 char corename[CORENAME_MAX_SIZE + 1]; 1763 struct mm_struct *mm = current->mm; 1764 struct linux_binfmt * binfmt; 1765 struct inode * inode; 1766 const struct cred *old_cred; 1767 struct cred *cred; 1768 int retval = 0; 1769 int flag = 0; 1770 int ispipe = 0; 1771 char **helper_argv = NULL; 1772 int helper_argc = 0; 1773 int dump_count = 0; 1774 static atomic_t core_dump_count = ATOMIC_INIT(0); 1775 struct coredump_params cprm = { 1776 .signr = signr, 1777 .regs = regs, 1778 .limit = current->signal->rlim[RLIMIT_CORE].rlim_cur, 1779 }; 1780 1781 audit_core_dumps(signr); 1782 1783 binfmt = mm->binfmt; 1784 if (!binfmt || !binfmt->core_dump) 1785 goto fail; 1786 1787 cred = prepare_creds(); 1788 if (!cred) { 1789 retval = -ENOMEM; 1790 goto fail; 1791 } 1792 1793 down_write(&mm->mmap_sem); 1794 /* 1795 * If another thread got here first, or we are not dumpable, bail out. 1796 */ 1797 if (mm->core_state || !get_dumpable(mm)) { 1798 up_write(&mm->mmap_sem); 1799 put_cred(cred); 1800 goto fail; 1801 } 1802 1803 /* 1804 * We cannot trust fsuid as being the "true" uid of the 1805 * process nor do we know its entire history. We only know it 1806 * was tainted so we dump it as root in mode 2. 1807 */ 1808 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */ 1809 flag = O_EXCL; /* Stop rewrite attacks */ 1810 cred->fsuid = 0; /* Dump root private */ 1811 } 1812 1813 retval = coredump_wait(exit_code, &core_state); 1814 if (retval < 0) { 1815 put_cred(cred); 1816 goto fail; 1817 } 1818 1819 old_cred = override_creds(cred); 1820 1821 /* 1822 * Clear any false indication of pending signals that might 1823 * be seen by the filesystem code called to write the core file. 1824 */ 1825 clear_thread_flag(TIF_SIGPENDING); 1826 1827 /* 1828 * lock_kernel() because format_corename() is controlled by sysctl, which 1829 * uses lock_kernel() 1830 */ 1831 lock_kernel(); 1832 ispipe = format_corename(corename, signr); 1833 unlock_kernel(); 1834 1835 if ((!ispipe) && (cprm.limit < binfmt->min_coredump)) 1836 goto fail_unlock; 1837 1838 if (ispipe) { 1839 if (cprm.limit == 0) { 1840 /* 1841 * Normally core limits are irrelevant to pipes, since 1842 * we're not writing to the file system, but we use 1843 * cprm.limit of 0 here as a speacial value. Any 1844 * non-zero limit gets set to RLIM_INFINITY below, but 1845 * a limit of 0 skips the dump. This is a consistent 1846 * way to catch recursive crashes. We can still crash 1847 * if the core_pattern binary sets RLIM_CORE = !0 1848 * but it runs as root, and can do lots of stupid things 1849 * Note that we use task_tgid_vnr here to grab the pid 1850 * of the process group leader. That way we get the 1851 * right pid if a thread in a multi-threaded 1852 * core_pattern process dies. 1853 */ 1854 printk(KERN_WARNING 1855 "Process %d(%s) has RLIMIT_CORE set to 0\n", 1856 task_tgid_vnr(current), current->comm); 1857 printk(KERN_WARNING "Aborting core\n"); 1858 goto fail_unlock; 1859 } 1860 1861 dump_count = atomic_inc_return(&core_dump_count); 1862 if (core_pipe_limit && (core_pipe_limit < dump_count)) { 1863 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n", 1864 task_tgid_vnr(current), current->comm); 1865 printk(KERN_WARNING "Skipping core dump\n"); 1866 goto fail_dropcount; 1867 } 1868 1869 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc); 1870 if (!helper_argv) { 1871 printk(KERN_WARNING "%s failed to allocate memory\n", 1872 __func__); 1873 goto fail_dropcount; 1874 } 1875 1876 cprm.limit = RLIM_INFINITY; 1877 1878 /* SIGPIPE can happen, but it's just never processed */ 1879 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL, 1880 &cprm.file)) { 1881 printk(KERN_INFO "Core dump to %s pipe failed\n", 1882 corename); 1883 goto fail_dropcount; 1884 } 1885 } else 1886 cprm.file = filp_open(corename, 1887 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 1888 0600); 1889 if (IS_ERR(cprm.file)) 1890 goto fail_dropcount; 1891 inode = cprm.file->f_path.dentry->d_inode; 1892 if (inode->i_nlink > 1) 1893 goto close_fail; /* multiple links - don't dump */ 1894 if (!ispipe && d_unhashed(cprm.file->f_path.dentry)) 1895 goto close_fail; 1896 1897 /* AK: actually i see no reason to not allow this for named pipes etc., 1898 but keep the previous behaviour for now. */ 1899 if (!ispipe && !S_ISREG(inode->i_mode)) 1900 goto close_fail; 1901 /* 1902 * Dont allow local users get cute and trick others to coredump 1903 * into their pre-created files: 1904 */ 1905 if (inode->i_uid != current_fsuid()) 1906 goto close_fail; 1907 if (!cprm.file->f_op) 1908 goto close_fail; 1909 if (!cprm.file->f_op->write) 1910 goto close_fail; 1911 if (!ispipe && 1912 do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0) 1913 goto close_fail; 1914 1915 retval = binfmt->core_dump(&cprm); 1916 1917 if (retval) 1918 current->signal->group_exit_code |= 0x80; 1919 close_fail: 1920 if (ispipe && core_pipe_limit) 1921 wait_for_dump_helpers(cprm.file); 1922 filp_close(cprm.file, NULL); 1923 fail_dropcount: 1924 if (dump_count) 1925 atomic_dec(&core_dump_count); 1926 fail_unlock: 1927 if (helper_argv) 1928 argv_free(helper_argv); 1929 1930 revert_creds(old_cred); 1931 put_cred(cred); 1932 coredump_finish(mm); 1933 fail: 1934 return; 1935 } 1936