1 /* 2 * linux/kernel/fork.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7 /* 8 * 'fork.c' contains the help-routines for the 'fork' system call 9 * (see also entry.S and others). 10 * Fork is rather simple, once you get the hang of it, but the memory 11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()' 12 */ 13 14 #include <linux/config.h> 15 #include <linux/slab.h> 16 #include <linux/init.h> 17 #include <linux/unistd.h> 18 #include <linux/smp_lock.h> 19 #include <linux/module.h> 20 #include <linux/vmalloc.h> 21 #include <linux/completion.h> 22 #include <linux/namespace.h> 23 #include <linux/personality.h> 24 #include <linux/mempolicy.h> 25 #include <linux/sem.h> 26 #include <linux/file.h> 27 #include <linux/key.h> 28 #include <linux/binfmts.h> 29 #include <linux/mman.h> 30 #include <linux/fs.h> 31 #include <linux/cpu.h> 32 #include <linux/cpuset.h> 33 #include <linux/security.h> 34 #include <linux/swap.h> 35 #include <linux/syscalls.h> 36 #include <linux/jiffies.h> 37 #include <linux/futex.h> 38 #include <linux/rcupdate.h> 39 #include <linux/ptrace.h> 40 #include <linux/mount.h> 41 #include <linux/audit.h> 42 #include <linux/profile.h> 43 #include <linux/rmap.h> 44 #include <linux/acct.h> 45 #include <linux/cn_proc.h> 46 47 #include <asm/pgtable.h> 48 #include <asm/pgalloc.h> 49 #include <asm/uaccess.h> 50 #include <asm/mmu_context.h> 51 #include <asm/cacheflush.h> 52 #include <asm/tlbflush.h> 53 54 /* 55 * Protected counters by write_lock_irq(&tasklist_lock) 56 */ 57 unsigned long total_forks; /* Handle normal Linux uptimes. */ 58 int nr_threads; /* The idle threads do not count.. */ 59 60 int max_threads; /* tunable limit on nr_threads */ 61 62 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 63 64 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 65 66 EXPORT_SYMBOL(tasklist_lock); 67 68 int nr_processes(void) 69 { 70 int cpu; 71 int total = 0; 72 73 for_each_online_cpu(cpu) 74 total += per_cpu(process_counts, cpu); 75 76 return total; 77 } 78 79 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR 80 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL) 81 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk)) 82 static kmem_cache_t *task_struct_cachep; 83 #endif 84 85 /* SLAB cache for signal_struct structures (tsk->signal) */ 86 kmem_cache_t *signal_cachep; 87 88 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 89 kmem_cache_t *sighand_cachep; 90 91 /* SLAB cache for files_struct structures (tsk->files) */ 92 kmem_cache_t *files_cachep; 93 94 /* SLAB cache for fs_struct structures (tsk->fs) */ 95 kmem_cache_t *fs_cachep; 96 97 /* SLAB cache for vm_area_struct structures */ 98 kmem_cache_t *vm_area_cachep; 99 100 /* SLAB cache for mm_struct structures (tsk->mm) */ 101 static kmem_cache_t *mm_cachep; 102 103 void free_task(struct task_struct *tsk) 104 { 105 free_thread_info(tsk->thread_info); 106 free_task_struct(tsk); 107 } 108 EXPORT_SYMBOL(free_task); 109 110 void __put_task_struct(struct task_struct *tsk) 111 { 112 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE))); 113 WARN_ON(atomic_read(&tsk->usage)); 114 WARN_ON(tsk == current); 115 116 if (unlikely(tsk->audit_context)) 117 audit_free(tsk); 118 security_task_free(tsk); 119 free_uid(tsk->user); 120 put_group_info(tsk->group_info); 121 122 if (!profile_handoff_task(tsk)) 123 free_task(tsk); 124 } 125 126 void __init fork_init(unsigned long mempages) 127 { 128 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR 129 #ifndef ARCH_MIN_TASKALIGN 130 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES 131 #endif 132 /* create a slab on which task_structs can be allocated */ 133 task_struct_cachep = 134 kmem_cache_create("task_struct", sizeof(struct task_struct), 135 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL); 136 #endif 137 138 /* 139 * The default maximum number of threads is set to a safe 140 * value: the thread structures can take up at most half 141 * of memory. 142 */ 143 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE); 144 145 /* 146 * we need to allow at least 20 threads to boot a system 147 */ 148 if(max_threads < 20) 149 max_threads = 20; 150 151 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 152 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 153 init_task.signal->rlim[RLIMIT_SIGPENDING] = 154 init_task.signal->rlim[RLIMIT_NPROC]; 155 } 156 157 static struct task_struct *dup_task_struct(struct task_struct *orig) 158 { 159 struct task_struct *tsk; 160 struct thread_info *ti; 161 162 prepare_to_copy(orig); 163 164 tsk = alloc_task_struct(); 165 if (!tsk) 166 return NULL; 167 168 ti = alloc_thread_info(tsk); 169 if (!ti) { 170 free_task_struct(tsk); 171 return NULL; 172 } 173 174 *tsk = *orig; 175 tsk->thread_info = ti; 176 setup_thread_stack(tsk, orig); 177 178 /* One for us, one for whoever does the "release_task()" (usually parent) */ 179 atomic_set(&tsk->usage,2); 180 atomic_set(&tsk->fs_excl, 0); 181 return tsk; 182 } 183 184 #ifdef CONFIG_MMU 185 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 186 { 187 struct vm_area_struct *mpnt, *tmp, **pprev; 188 struct rb_node **rb_link, *rb_parent; 189 int retval; 190 unsigned long charge; 191 struct mempolicy *pol; 192 193 down_write(&oldmm->mmap_sem); 194 flush_cache_mm(oldmm); 195 down_write(&mm->mmap_sem); 196 197 mm->locked_vm = 0; 198 mm->mmap = NULL; 199 mm->mmap_cache = NULL; 200 mm->free_area_cache = oldmm->mmap_base; 201 mm->cached_hole_size = ~0UL; 202 mm->map_count = 0; 203 cpus_clear(mm->cpu_vm_mask); 204 mm->mm_rb = RB_ROOT; 205 rb_link = &mm->mm_rb.rb_node; 206 rb_parent = NULL; 207 pprev = &mm->mmap; 208 209 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 210 struct file *file; 211 212 if (mpnt->vm_flags & VM_DONTCOPY) { 213 long pages = vma_pages(mpnt); 214 mm->total_vm -= pages; 215 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file, 216 -pages); 217 continue; 218 } 219 charge = 0; 220 if (mpnt->vm_flags & VM_ACCOUNT) { 221 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT; 222 if (security_vm_enough_memory(len)) 223 goto fail_nomem; 224 charge = len; 225 } 226 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); 227 if (!tmp) 228 goto fail_nomem; 229 *tmp = *mpnt; 230 pol = mpol_copy(vma_policy(mpnt)); 231 retval = PTR_ERR(pol); 232 if (IS_ERR(pol)) 233 goto fail_nomem_policy; 234 vma_set_policy(tmp, pol); 235 tmp->vm_flags &= ~VM_LOCKED; 236 tmp->vm_mm = mm; 237 tmp->vm_next = NULL; 238 anon_vma_link(tmp); 239 file = tmp->vm_file; 240 if (file) { 241 struct inode *inode = file->f_dentry->d_inode; 242 get_file(file); 243 if (tmp->vm_flags & VM_DENYWRITE) 244 atomic_dec(&inode->i_writecount); 245 246 /* insert tmp into the share list, just after mpnt */ 247 spin_lock(&file->f_mapping->i_mmap_lock); 248 tmp->vm_truncate_count = mpnt->vm_truncate_count; 249 flush_dcache_mmap_lock(file->f_mapping); 250 vma_prio_tree_add(tmp, mpnt); 251 flush_dcache_mmap_unlock(file->f_mapping); 252 spin_unlock(&file->f_mapping->i_mmap_lock); 253 } 254 255 /* 256 * Link in the new vma and copy the page table entries. 257 */ 258 *pprev = tmp; 259 pprev = &tmp->vm_next; 260 261 __vma_link_rb(mm, tmp, rb_link, rb_parent); 262 rb_link = &tmp->vm_rb.rb_right; 263 rb_parent = &tmp->vm_rb; 264 265 mm->map_count++; 266 retval = copy_page_range(mm, oldmm, mpnt); 267 268 if (tmp->vm_ops && tmp->vm_ops->open) 269 tmp->vm_ops->open(tmp); 270 271 if (retval) 272 goto out; 273 } 274 retval = 0; 275 out: 276 up_write(&mm->mmap_sem); 277 flush_tlb_mm(oldmm); 278 up_write(&oldmm->mmap_sem); 279 return retval; 280 fail_nomem_policy: 281 kmem_cache_free(vm_area_cachep, tmp); 282 fail_nomem: 283 retval = -ENOMEM; 284 vm_unacct_memory(charge); 285 goto out; 286 } 287 288 static inline int mm_alloc_pgd(struct mm_struct * mm) 289 { 290 mm->pgd = pgd_alloc(mm); 291 if (unlikely(!mm->pgd)) 292 return -ENOMEM; 293 return 0; 294 } 295 296 static inline void mm_free_pgd(struct mm_struct * mm) 297 { 298 pgd_free(mm->pgd); 299 } 300 #else 301 #define dup_mmap(mm, oldmm) (0) 302 #define mm_alloc_pgd(mm) (0) 303 #define mm_free_pgd(mm) 304 #endif /* CONFIG_MMU */ 305 306 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 307 308 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL)) 309 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 310 311 #include <linux/init_task.h> 312 313 static struct mm_struct * mm_init(struct mm_struct * mm) 314 { 315 atomic_set(&mm->mm_users, 1); 316 atomic_set(&mm->mm_count, 1); 317 init_rwsem(&mm->mmap_sem); 318 INIT_LIST_HEAD(&mm->mmlist); 319 mm->core_waiters = 0; 320 mm->nr_ptes = 0; 321 set_mm_counter(mm, file_rss, 0); 322 set_mm_counter(mm, anon_rss, 0); 323 spin_lock_init(&mm->page_table_lock); 324 rwlock_init(&mm->ioctx_list_lock); 325 mm->ioctx_list = NULL; 326 mm->free_area_cache = TASK_UNMAPPED_BASE; 327 mm->cached_hole_size = ~0UL; 328 329 if (likely(!mm_alloc_pgd(mm))) { 330 mm->def_flags = 0; 331 return mm; 332 } 333 free_mm(mm); 334 return NULL; 335 } 336 337 /* 338 * Allocate and initialize an mm_struct. 339 */ 340 struct mm_struct * mm_alloc(void) 341 { 342 struct mm_struct * mm; 343 344 mm = allocate_mm(); 345 if (mm) { 346 memset(mm, 0, sizeof(*mm)); 347 mm = mm_init(mm); 348 } 349 return mm; 350 } 351 352 /* 353 * Called when the last reference to the mm 354 * is dropped: either by a lazy thread or by 355 * mmput. Free the page directory and the mm. 356 */ 357 void fastcall __mmdrop(struct mm_struct *mm) 358 { 359 BUG_ON(mm == &init_mm); 360 mm_free_pgd(mm); 361 destroy_context(mm); 362 free_mm(mm); 363 } 364 365 /* 366 * Decrement the use count and release all resources for an mm. 367 */ 368 void mmput(struct mm_struct *mm) 369 { 370 if (atomic_dec_and_test(&mm->mm_users)) { 371 exit_aio(mm); 372 exit_mmap(mm); 373 if (!list_empty(&mm->mmlist)) { 374 spin_lock(&mmlist_lock); 375 list_del(&mm->mmlist); 376 spin_unlock(&mmlist_lock); 377 } 378 put_swap_token(mm); 379 mmdrop(mm); 380 } 381 } 382 EXPORT_SYMBOL_GPL(mmput); 383 384 /** 385 * get_task_mm - acquire a reference to the task's mm 386 * 387 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning 388 * this kernel workthread has transiently adopted a user mm with use_mm, 389 * to do its AIO) is not set and if so returns a reference to it, after 390 * bumping up the use count. User must release the mm via mmput() 391 * after use. Typically used by /proc and ptrace. 392 */ 393 struct mm_struct *get_task_mm(struct task_struct *task) 394 { 395 struct mm_struct *mm; 396 397 task_lock(task); 398 mm = task->mm; 399 if (mm) { 400 if (task->flags & PF_BORROWED_MM) 401 mm = NULL; 402 else 403 atomic_inc(&mm->mm_users); 404 } 405 task_unlock(task); 406 return mm; 407 } 408 EXPORT_SYMBOL_GPL(get_task_mm); 409 410 /* Please note the differences between mmput and mm_release. 411 * mmput is called whenever we stop holding onto a mm_struct, 412 * error success whatever. 413 * 414 * mm_release is called after a mm_struct has been removed 415 * from the current process. 416 * 417 * This difference is important for error handling, when we 418 * only half set up a mm_struct for a new process and need to restore 419 * the old one. Because we mmput the new mm_struct before 420 * restoring the old one. . . 421 * Eric Biederman 10 January 1998 422 */ 423 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 424 { 425 struct completion *vfork_done = tsk->vfork_done; 426 427 /* Get rid of any cached register state */ 428 deactivate_mm(tsk, mm); 429 430 /* notify parent sleeping on vfork() */ 431 if (vfork_done) { 432 tsk->vfork_done = NULL; 433 complete(vfork_done); 434 } 435 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) { 436 u32 __user * tidptr = tsk->clear_child_tid; 437 tsk->clear_child_tid = NULL; 438 439 /* 440 * We don't check the error code - if userspace has 441 * not set up a proper pointer then tough luck. 442 */ 443 put_user(0, tidptr); 444 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0); 445 } 446 } 447 448 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk) 449 { 450 struct mm_struct * mm, *oldmm; 451 int retval; 452 453 tsk->min_flt = tsk->maj_flt = 0; 454 tsk->nvcsw = tsk->nivcsw = 0; 455 456 tsk->mm = NULL; 457 tsk->active_mm = NULL; 458 459 /* 460 * Are we cloning a kernel thread? 461 * 462 * We need to steal a active VM for that.. 463 */ 464 oldmm = current->mm; 465 if (!oldmm) 466 return 0; 467 468 if (clone_flags & CLONE_VM) { 469 atomic_inc(&oldmm->mm_users); 470 mm = oldmm; 471 goto good_mm; 472 } 473 474 retval = -ENOMEM; 475 mm = allocate_mm(); 476 if (!mm) 477 goto fail_nomem; 478 479 /* Copy the current MM stuff.. */ 480 memcpy(mm, oldmm, sizeof(*mm)); 481 if (!mm_init(mm)) 482 goto fail_nomem; 483 484 if (init_new_context(tsk,mm)) 485 goto fail_nocontext; 486 487 retval = dup_mmap(mm, oldmm); 488 if (retval) 489 goto free_pt; 490 491 mm->hiwater_rss = get_mm_rss(mm); 492 mm->hiwater_vm = mm->total_vm; 493 494 good_mm: 495 tsk->mm = mm; 496 tsk->active_mm = mm; 497 return 0; 498 499 free_pt: 500 mmput(mm); 501 fail_nomem: 502 return retval; 503 504 fail_nocontext: 505 /* 506 * If init_new_context() failed, we cannot use mmput() to free the mm 507 * because it calls destroy_context() 508 */ 509 mm_free_pgd(mm); 510 free_mm(mm); 511 return retval; 512 } 513 514 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old) 515 { 516 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL); 517 /* We don't need to lock fs - think why ;-) */ 518 if (fs) { 519 atomic_set(&fs->count, 1); 520 rwlock_init(&fs->lock); 521 fs->umask = old->umask; 522 read_lock(&old->lock); 523 fs->rootmnt = mntget(old->rootmnt); 524 fs->root = dget(old->root); 525 fs->pwdmnt = mntget(old->pwdmnt); 526 fs->pwd = dget(old->pwd); 527 if (old->altroot) { 528 fs->altrootmnt = mntget(old->altrootmnt); 529 fs->altroot = dget(old->altroot); 530 } else { 531 fs->altrootmnt = NULL; 532 fs->altroot = NULL; 533 } 534 read_unlock(&old->lock); 535 } 536 return fs; 537 } 538 539 struct fs_struct *copy_fs_struct(struct fs_struct *old) 540 { 541 return __copy_fs_struct(old); 542 } 543 544 EXPORT_SYMBOL_GPL(copy_fs_struct); 545 546 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk) 547 { 548 if (clone_flags & CLONE_FS) { 549 atomic_inc(¤t->fs->count); 550 return 0; 551 } 552 tsk->fs = __copy_fs_struct(current->fs); 553 if (!tsk->fs) 554 return -ENOMEM; 555 return 0; 556 } 557 558 static int count_open_files(struct fdtable *fdt) 559 { 560 int size = fdt->max_fdset; 561 int i; 562 563 /* Find the last open fd */ 564 for (i = size/(8*sizeof(long)); i > 0; ) { 565 if (fdt->open_fds->fds_bits[--i]) 566 break; 567 } 568 i = (i+1) * 8 * sizeof(long); 569 return i; 570 } 571 572 static struct files_struct *alloc_files(void) 573 { 574 struct files_struct *newf; 575 struct fdtable *fdt; 576 577 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL); 578 if (!newf) 579 goto out; 580 581 atomic_set(&newf->count, 1); 582 583 spin_lock_init(&newf->file_lock); 584 fdt = &newf->fdtab; 585 fdt->next_fd = 0; 586 fdt->max_fds = NR_OPEN_DEFAULT; 587 fdt->max_fdset = __FD_SETSIZE; 588 fdt->close_on_exec = &newf->close_on_exec_init; 589 fdt->open_fds = &newf->open_fds_init; 590 fdt->fd = &newf->fd_array[0]; 591 INIT_RCU_HEAD(&fdt->rcu); 592 fdt->free_files = NULL; 593 fdt->next = NULL; 594 rcu_assign_pointer(newf->fdt, fdt); 595 out: 596 return newf; 597 } 598 599 static int copy_files(unsigned long clone_flags, struct task_struct * tsk) 600 { 601 struct files_struct *oldf, *newf; 602 struct file **old_fds, **new_fds; 603 int open_files, size, i, error = 0, expand; 604 struct fdtable *old_fdt, *new_fdt; 605 606 /* 607 * A background process may not have any files ... 608 */ 609 oldf = current->files; 610 if (!oldf) 611 goto out; 612 613 if (clone_flags & CLONE_FILES) { 614 atomic_inc(&oldf->count); 615 goto out; 616 } 617 618 /* 619 * Note: we may be using current for both targets (See exec.c) 620 * This works because we cache current->files (old) as oldf. Don't 621 * break this. 622 */ 623 tsk->files = NULL; 624 error = -ENOMEM; 625 newf = alloc_files(); 626 if (!newf) 627 goto out; 628 629 spin_lock(&oldf->file_lock); 630 old_fdt = files_fdtable(oldf); 631 new_fdt = files_fdtable(newf); 632 size = old_fdt->max_fdset; 633 open_files = count_open_files(old_fdt); 634 expand = 0; 635 636 /* 637 * Check whether we need to allocate a larger fd array or fd set. 638 * Note: we're not a clone task, so the open count won't change. 639 */ 640 if (open_files > new_fdt->max_fdset) { 641 new_fdt->max_fdset = 0; 642 expand = 1; 643 } 644 if (open_files > new_fdt->max_fds) { 645 new_fdt->max_fds = 0; 646 expand = 1; 647 } 648 649 /* if the old fdset gets grown now, we'll only copy up to "size" fds */ 650 if (expand) { 651 spin_unlock(&oldf->file_lock); 652 spin_lock(&newf->file_lock); 653 error = expand_files(newf, open_files-1); 654 spin_unlock(&newf->file_lock); 655 if (error < 0) 656 goto out_release; 657 new_fdt = files_fdtable(newf); 658 /* 659 * Reacquire the oldf lock and a pointer to its fd table 660 * who knows it may have a new bigger fd table. We need 661 * the latest pointer. 662 */ 663 spin_lock(&oldf->file_lock); 664 old_fdt = files_fdtable(oldf); 665 } 666 667 old_fds = old_fdt->fd; 668 new_fds = new_fdt->fd; 669 670 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8); 671 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8); 672 673 for (i = open_files; i != 0; i--) { 674 struct file *f = *old_fds++; 675 if (f) { 676 get_file(f); 677 } else { 678 /* 679 * The fd may be claimed in the fd bitmap but not yet 680 * instantiated in the files array if a sibling thread 681 * is partway through open(). So make sure that this 682 * fd is available to the new process. 683 */ 684 FD_CLR(open_files - i, new_fdt->open_fds); 685 } 686 rcu_assign_pointer(*new_fds++, f); 687 } 688 spin_unlock(&oldf->file_lock); 689 690 /* compute the remainder to be cleared */ 691 size = (new_fdt->max_fds - open_files) * sizeof(struct file *); 692 693 /* This is long word aligned thus could use a optimized version */ 694 memset(new_fds, 0, size); 695 696 if (new_fdt->max_fdset > open_files) { 697 int left = (new_fdt->max_fdset-open_files)/8; 698 int start = open_files / (8 * sizeof(unsigned long)); 699 700 memset(&new_fdt->open_fds->fds_bits[start], 0, left); 701 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left); 702 } 703 704 tsk->files = newf; 705 error = 0; 706 out: 707 return error; 708 709 out_release: 710 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset); 711 free_fdset (new_fdt->open_fds, new_fdt->max_fdset); 712 free_fd_array(new_fdt->fd, new_fdt->max_fds); 713 kmem_cache_free(files_cachep, newf); 714 goto out; 715 } 716 717 /* 718 * Helper to unshare the files of the current task. 719 * We don't want to expose copy_files internals to 720 * the exec layer of the kernel. 721 */ 722 723 int unshare_files(void) 724 { 725 struct files_struct *files = current->files; 726 int rc; 727 728 if(!files) 729 BUG(); 730 731 /* This can race but the race causes us to copy when we don't 732 need to and drop the copy */ 733 if(atomic_read(&files->count) == 1) 734 { 735 atomic_inc(&files->count); 736 return 0; 737 } 738 rc = copy_files(0, current); 739 if(rc) 740 current->files = files; 741 return rc; 742 } 743 744 EXPORT_SYMBOL(unshare_files); 745 746 void sighand_free_cb(struct rcu_head *rhp) 747 { 748 struct sighand_struct *sp; 749 750 sp = container_of(rhp, struct sighand_struct, rcu); 751 kmem_cache_free(sighand_cachep, sp); 752 } 753 754 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk) 755 { 756 struct sighand_struct *sig; 757 758 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) { 759 atomic_inc(¤t->sighand->count); 760 return 0; 761 } 762 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 763 rcu_assign_pointer(tsk->sighand, sig); 764 if (!sig) 765 return -ENOMEM; 766 spin_lock_init(&sig->siglock); 767 atomic_set(&sig->count, 1); 768 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 769 return 0; 770 } 771 772 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk) 773 { 774 struct signal_struct *sig; 775 int ret; 776 777 if (clone_flags & CLONE_THREAD) { 778 atomic_inc(¤t->signal->count); 779 atomic_inc(¤t->signal->live); 780 return 0; 781 } 782 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL); 783 tsk->signal = sig; 784 if (!sig) 785 return -ENOMEM; 786 787 ret = copy_thread_group_keys(tsk); 788 if (ret < 0) { 789 kmem_cache_free(signal_cachep, sig); 790 return ret; 791 } 792 793 atomic_set(&sig->count, 1); 794 atomic_set(&sig->live, 1); 795 init_waitqueue_head(&sig->wait_chldexit); 796 sig->flags = 0; 797 sig->group_exit_code = 0; 798 sig->group_exit_task = NULL; 799 sig->group_stop_count = 0; 800 sig->curr_target = NULL; 801 init_sigpending(&sig->shared_pending); 802 INIT_LIST_HEAD(&sig->posix_timers); 803 804 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC); 805 sig->it_real_incr.tv64 = 0; 806 sig->real_timer.function = it_real_fn; 807 sig->real_timer.data = tsk; 808 809 sig->it_virt_expires = cputime_zero; 810 sig->it_virt_incr = cputime_zero; 811 sig->it_prof_expires = cputime_zero; 812 sig->it_prof_incr = cputime_zero; 813 814 sig->leader = 0; /* session leadership doesn't inherit */ 815 sig->tty_old_pgrp = 0; 816 817 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero; 818 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0; 819 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0; 820 sig->sched_time = 0; 821 INIT_LIST_HEAD(&sig->cpu_timers[0]); 822 INIT_LIST_HEAD(&sig->cpu_timers[1]); 823 INIT_LIST_HEAD(&sig->cpu_timers[2]); 824 825 task_lock(current->group_leader); 826 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 827 task_unlock(current->group_leader); 828 829 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) { 830 /* 831 * New sole thread in the process gets an expiry time 832 * of the whole CPU time limit. 833 */ 834 tsk->it_prof_expires = 835 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur); 836 } 837 838 return 0; 839 } 840 841 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p) 842 { 843 unsigned long new_flags = p->flags; 844 845 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE); 846 new_flags |= PF_FORKNOEXEC; 847 if (!(clone_flags & CLONE_PTRACE)) 848 p->ptrace = 0; 849 p->flags = new_flags; 850 } 851 852 asmlinkage long sys_set_tid_address(int __user *tidptr) 853 { 854 current->clear_child_tid = tidptr; 855 856 return current->pid; 857 } 858 859 /* 860 * This creates a new process as a copy of the old one, 861 * but does not actually start it yet. 862 * 863 * It copies the registers, and all the appropriate 864 * parts of the process environment (as per the clone 865 * flags). The actual kick-off is left to the caller. 866 */ 867 static task_t *copy_process(unsigned long clone_flags, 868 unsigned long stack_start, 869 struct pt_regs *regs, 870 unsigned long stack_size, 871 int __user *parent_tidptr, 872 int __user *child_tidptr, 873 int pid) 874 { 875 int retval; 876 struct task_struct *p = NULL; 877 878 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 879 return ERR_PTR(-EINVAL); 880 881 /* 882 * Thread groups must share signals as well, and detached threads 883 * can only be started up within the thread group. 884 */ 885 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 886 return ERR_PTR(-EINVAL); 887 888 /* 889 * Shared signal handlers imply shared VM. By way of the above, 890 * thread groups also imply shared VM. Blocking this case allows 891 * for various simplifications in other code. 892 */ 893 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 894 return ERR_PTR(-EINVAL); 895 896 retval = security_task_create(clone_flags); 897 if (retval) 898 goto fork_out; 899 900 retval = -ENOMEM; 901 p = dup_task_struct(current); 902 if (!p) 903 goto fork_out; 904 905 retval = -EAGAIN; 906 if (atomic_read(&p->user->processes) >= 907 p->signal->rlim[RLIMIT_NPROC].rlim_cur) { 908 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) && 909 p->user != &root_user) 910 goto bad_fork_free; 911 } 912 913 atomic_inc(&p->user->__count); 914 atomic_inc(&p->user->processes); 915 get_group_info(p->group_info); 916 917 /* 918 * If multiple threads are within copy_process(), then this check 919 * triggers too late. This doesn't hurt, the check is only there 920 * to stop root fork bombs. 921 */ 922 if (nr_threads >= max_threads) 923 goto bad_fork_cleanup_count; 924 925 if (!try_module_get(task_thread_info(p)->exec_domain->module)) 926 goto bad_fork_cleanup_count; 927 928 if (p->binfmt && !try_module_get(p->binfmt->module)) 929 goto bad_fork_cleanup_put_domain; 930 931 p->did_exec = 0; 932 copy_flags(clone_flags, p); 933 p->pid = pid; 934 retval = -EFAULT; 935 if (clone_flags & CLONE_PARENT_SETTID) 936 if (put_user(p->pid, parent_tidptr)) 937 goto bad_fork_cleanup; 938 939 p->proc_dentry = NULL; 940 941 INIT_LIST_HEAD(&p->children); 942 INIT_LIST_HEAD(&p->sibling); 943 p->vfork_done = NULL; 944 spin_lock_init(&p->alloc_lock); 945 spin_lock_init(&p->proc_lock); 946 947 clear_tsk_thread_flag(p, TIF_SIGPENDING); 948 init_sigpending(&p->pending); 949 950 p->utime = cputime_zero; 951 p->stime = cputime_zero; 952 p->sched_time = 0; 953 p->rchar = 0; /* I/O counter: bytes read */ 954 p->wchar = 0; /* I/O counter: bytes written */ 955 p->syscr = 0; /* I/O counter: read syscalls */ 956 p->syscw = 0; /* I/O counter: write syscalls */ 957 acct_clear_integrals(p); 958 959 p->it_virt_expires = cputime_zero; 960 p->it_prof_expires = cputime_zero; 961 p->it_sched_expires = 0; 962 INIT_LIST_HEAD(&p->cpu_timers[0]); 963 INIT_LIST_HEAD(&p->cpu_timers[1]); 964 INIT_LIST_HEAD(&p->cpu_timers[2]); 965 966 p->lock_depth = -1; /* -1 = no lock */ 967 do_posix_clock_monotonic_gettime(&p->start_time); 968 p->security = NULL; 969 p->io_context = NULL; 970 p->io_wait = NULL; 971 p->audit_context = NULL; 972 cpuset_fork(p); 973 #ifdef CONFIG_NUMA 974 p->mempolicy = mpol_copy(p->mempolicy); 975 if (IS_ERR(p->mempolicy)) { 976 retval = PTR_ERR(p->mempolicy); 977 p->mempolicy = NULL; 978 goto bad_fork_cleanup_cpuset; 979 } 980 #endif 981 982 #ifdef CONFIG_DEBUG_MUTEXES 983 p->blocked_on = NULL; /* not blocked yet */ 984 #endif 985 986 p->tgid = p->pid; 987 if (clone_flags & CLONE_THREAD) 988 p->tgid = current->tgid; 989 990 if ((retval = security_task_alloc(p))) 991 goto bad_fork_cleanup_policy; 992 if ((retval = audit_alloc(p))) 993 goto bad_fork_cleanup_security; 994 /* copy all the process information */ 995 if ((retval = copy_semundo(clone_flags, p))) 996 goto bad_fork_cleanup_audit; 997 if ((retval = copy_files(clone_flags, p))) 998 goto bad_fork_cleanup_semundo; 999 if ((retval = copy_fs(clone_flags, p))) 1000 goto bad_fork_cleanup_files; 1001 if ((retval = copy_sighand(clone_flags, p))) 1002 goto bad_fork_cleanup_fs; 1003 if ((retval = copy_signal(clone_flags, p))) 1004 goto bad_fork_cleanup_sighand; 1005 if ((retval = copy_mm(clone_flags, p))) 1006 goto bad_fork_cleanup_signal; 1007 if ((retval = copy_keys(clone_flags, p))) 1008 goto bad_fork_cleanup_mm; 1009 if ((retval = copy_namespace(clone_flags, p))) 1010 goto bad_fork_cleanup_keys; 1011 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs); 1012 if (retval) 1013 goto bad_fork_cleanup_namespace; 1014 1015 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1016 /* 1017 * Clear TID on mm_release()? 1018 */ 1019 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL; 1020 1021 /* 1022 * Syscall tracing should be turned off in the child regardless 1023 * of CLONE_PTRACE. 1024 */ 1025 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1026 #ifdef TIF_SYSCALL_EMU 1027 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1028 #endif 1029 1030 /* Our parent execution domain becomes current domain 1031 These must match for thread signalling to apply */ 1032 1033 p->parent_exec_id = p->self_exec_id; 1034 1035 /* ok, now we should be set up.. */ 1036 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL); 1037 p->pdeath_signal = 0; 1038 p->exit_state = 0; 1039 1040 /* 1041 * Ok, make it visible to the rest of the system. 1042 * We dont wake it up yet. 1043 */ 1044 p->group_leader = p; 1045 INIT_LIST_HEAD(&p->ptrace_children); 1046 INIT_LIST_HEAD(&p->ptrace_list); 1047 1048 /* Perform scheduler related setup. Assign this task to a CPU. */ 1049 sched_fork(p, clone_flags); 1050 1051 /* Need tasklist lock for parent etc handling! */ 1052 write_lock_irq(&tasklist_lock); 1053 1054 /* 1055 * The task hasn't been attached yet, so its cpus_allowed mask will 1056 * not be changed, nor will its assigned CPU. 1057 * 1058 * The cpus_allowed mask of the parent may have changed after it was 1059 * copied first time - so re-copy it here, then check the child's CPU 1060 * to ensure it is on a valid CPU (and if not, just force it back to 1061 * parent's CPU). This avoids alot of nasty races. 1062 */ 1063 p->cpus_allowed = current->cpus_allowed; 1064 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) || 1065 !cpu_online(task_cpu(p)))) 1066 set_task_cpu(p, smp_processor_id()); 1067 1068 /* 1069 * Check for pending SIGKILL! The new thread should not be allowed 1070 * to slip out of an OOM kill. (or normal SIGKILL.) 1071 */ 1072 if (sigismember(¤t->pending.signal, SIGKILL)) { 1073 write_unlock_irq(&tasklist_lock); 1074 retval = -EINTR; 1075 goto bad_fork_cleanup_namespace; 1076 } 1077 1078 /* CLONE_PARENT re-uses the old parent */ 1079 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) 1080 p->real_parent = current->real_parent; 1081 else 1082 p->real_parent = current; 1083 p->parent = p->real_parent; 1084 1085 if (clone_flags & CLONE_THREAD) { 1086 spin_lock(¤t->sighand->siglock); 1087 /* 1088 * Important: if an exit-all has been started then 1089 * do not create this new thread - the whole thread 1090 * group is supposed to exit anyway. 1091 */ 1092 if (current->signal->flags & SIGNAL_GROUP_EXIT) { 1093 spin_unlock(¤t->sighand->siglock); 1094 write_unlock_irq(&tasklist_lock); 1095 retval = -EAGAIN; 1096 goto bad_fork_cleanup_namespace; 1097 } 1098 p->group_leader = current->group_leader; 1099 1100 if (current->signal->group_stop_count > 0) { 1101 /* 1102 * There is an all-stop in progress for the group. 1103 * We ourselves will stop as soon as we check signals. 1104 * Make the new thread part of that group stop too. 1105 */ 1106 current->signal->group_stop_count++; 1107 set_tsk_thread_flag(p, TIF_SIGPENDING); 1108 } 1109 1110 if (!cputime_eq(current->signal->it_virt_expires, 1111 cputime_zero) || 1112 !cputime_eq(current->signal->it_prof_expires, 1113 cputime_zero) || 1114 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY || 1115 !list_empty(¤t->signal->cpu_timers[0]) || 1116 !list_empty(¤t->signal->cpu_timers[1]) || 1117 !list_empty(¤t->signal->cpu_timers[2])) { 1118 /* 1119 * Have child wake up on its first tick to check 1120 * for process CPU timers. 1121 */ 1122 p->it_prof_expires = jiffies_to_cputime(1); 1123 } 1124 1125 spin_unlock(¤t->sighand->siglock); 1126 } 1127 1128 /* 1129 * inherit ioprio 1130 */ 1131 p->ioprio = current->ioprio; 1132 1133 SET_LINKS(p); 1134 if (unlikely(p->ptrace & PT_PTRACED)) 1135 __ptrace_link(p, current->parent); 1136 1137 attach_pid(p, PIDTYPE_PID, p->pid); 1138 attach_pid(p, PIDTYPE_TGID, p->tgid); 1139 if (thread_group_leader(p)) { 1140 p->signal->tty = current->signal->tty; 1141 p->signal->pgrp = process_group(current); 1142 p->signal->session = current->signal->session; 1143 attach_pid(p, PIDTYPE_PGID, process_group(p)); 1144 attach_pid(p, PIDTYPE_SID, p->signal->session); 1145 if (p->pid) 1146 __get_cpu_var(process_counts)++; 1147 } 1148 1149 nr_threads++; 1150 total_forks++; 1151 write_unlock_irq(&tasklist_lock); 1152 proc_fork_connector(p); 1153 return p; 1154 1155 bad_fork_cleanup_namespace: 1156 exit_namespace(p); 1157 bad_fork_cleanup_keys: 1158 exit_keys(p); 1159 bad_fork_cleanup_mm: 1160 if (p->mm) 1161 mmput(p->mm); 1162 bad_fork_cleanup_signal: 1163 exit_signal(p); 1164 bad_fork_cleanup_sighand: 1165 exit_sighand(p); 1166 bad_fork_cleanup_fs: 1167 exit_fs(p); /* blocking */ 1168 bad_fork_cleanup_files: 1169 exit_files(p); /* blocking */ 1170 bad_fork_cleanup_semundo: 1171 exit_sem(p); 1172 bad_fork_cleanup_audit: 1173 audit_free(p); 1174 bad_fork_cleanup_security: 1175 security_task_free(p); 1176 bad_fork_cleanup_policy: 1177 #ifdef CONFIG_NUMA 1178 mpol_free(p->mempolicy); 1179 bad_fork_cleanup_cpuset: 1180 #endif 1181 cpuset_exit(p); 1182 bad_fork_cleanup: 1183 if (p->binfmt) 1184 module_put(p->binfmt->module); 1185 bad_fork_cleanup_put_domain: 1186 module_put(task_thread_info(p)->exec_domain->module); 1187 bad_fork_cleanup_count: 1188 put_group_info(p->group_info); 1189 atomic_dec(&p->user->processes); 1190 free_uid(p->user); 1191 bad_fork_free: 1192 free_task(p); 1193 fork_out: 1194 return ERR_PTR(retval); 1195 } 1196 1197 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs) 1198 { 1199 memset(regs, 0, sizeof(struct pt_regs)); 1200 return regs; 1201 } 1202 1203 task_t * __devinit fork_idle(int cpu) 1204 { 1205 task_t *task; 1206 struct pt_regs regs; 1207 1208 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0); 1209 if (!task) 1210 return ERR_PTR(-ENOMEM); 1211 init_idle(task, cpu); 1212 unhash_process(task); 1213 return task; 1214 } 1215 1216 static inline int fork_traceflag (unsigned clone_flags) 1217 { 1218 if (clone_flags & CLONE_UNTRACED) 1219 return 0; 1220 else if (clone_flags & CLONE_VFORK) { 1221 if (current->ptrace & PT_TRACE_VFORK) 1222 return PTRACE_EVENT_VFORK; 1223 } else if ((clone_flags & CSIGNAL) != SIGCHLD) { 1224 if (current->ptrace & PT_TRACE_CLONE) 1225 return PTRACE_EVENT_CLONE; 1226 } else if (current->ptrace & PT_TRACE_FORK) 1227 return PTRACE_EVENT_FORK; 1228 1229 return 0; 1230 } 1231 1232 /* 1233 * Ok, this is the main fork-routine. 1234 * 1235 * It copies the process, and if successful kick-starts 1236 * it and waits for it to finish using the VM if required. 1237 */ 1238 long do_fork(unsigned long clone_flags, 1239 unsigned long stack_start, 1240 struct pt_regs *regs, 1241 unsigned long stack_size, 1242 int __user *parent_tidptr, 1243 int __user *child_tidptr) 1244 { 1245 struct task_struct *p; 1246 int trace = 0; 1247 long pid = alloc_pidmap(); 1248 1249 if (pid < 0) 1250 return -EAGAIN; 1251 if (unlikely(current->ptrace)) { 1252 trace = fork_traceflag (clone_flags); 1253 if (trace) 1254 clone_flags |= CLONE_PTRACE; 1255 } 1256 1257 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid); 1258 /* 1259 * Do this prior waking up the new thread - the thread pointer 1260 * might get invalid after that point, if the thread exits quickly. 1261 */ 1262 if (!IS_ERR(p)) { 1263 struct completion vfork; 1264 1265 if (clone_flags & CLONE_VFORK) { 1266 p->vfork_done = &vfork; 1267 init_completion(&vfork); 1268 } 1269 1270 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) { 1271 /* 1272 * We'll start up with an immediate SIGSTOP. 1273 */ 1274 sigaddset(&p->pending.signal, SIGSTOP); 1275 set_tsk_thread_flag(p, TIF_SIGPENDING); 1276 } 1277 1278 if (!(clone_flags & CLONE_STOPPED)) 1279 wake_up_new_task(p, clone_flags); 1280 else 1281 p->state = TASK_STOPPED; 1282 1283 if (unlikely (trace)) { 1284 current->ptrace_message = pid; 1285 ptrace_notify ((trace << 8) | SIGTRAP); 1286 } 1287 1288 if (clone_flags & CLONE_VFORK) { 1289 wait_for_completion(&vfork); 1290 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) 1291 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP); 1292 } 1293 } else { 1294 free_pidmap(pid); 1295 pid = PTR_ERR(p); 1296 } 1297 return pid; 1298 } 1299 1300 void __init proc_caches_init(void) 1301 { 1302 sighand_cachep = kmem_cache_create("sighand_cache", 1303 sizeof(struct sighand_struct), 0, 1304 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); 1305 signal_cachep = kmem_cache_create("signal_cache", 1306 sizeof(struct signal_struct), 0, 1307 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); 1308 files_cachep = kmem_cache_create("files_cache", 1309 sizeof(struct files_struct), 0, 1310 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); 1311 fs_cachep = kmem_cache_create("fs_cache", 1312 sizeof(struct fs_struct), 0, 1313 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); 1314 vm_area_cachep = kmem_cache_create("vm_area_struct", 1315 sizeof(struct vm_area_struct), 0, 1316 SLAB_PANIC, NULL, NULL); 1317 mm_cachep = kmem_cache_create("mm_struct", 1318 sizeof(struct mm_struct), 0, 1319 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); 1320 } 1321